Degree

Notes for a Licenciatura
Based on Lectures of J. Sancho
July 12, 2021

Preface
During the seventies, the Licenciatura –the university studies– in Mathematics at the University
of Salamanca had a highly coherent structure devised by late Prof. Juan B. Sancho Guimerá.
In his lectures he taught us how the main ideas of Grothendieck pervade mathematics:
How separable extensions of a field k may be understood as coverings of a one point space
Spec k, the localization process being the change of the base field and the globalization process
encoding Galois theory.
How any morphism X → T may be understood as a family of spaces parameterized by T ,
so that absolute statements about a space X really refer to the projection X → p onto the one
point space, and how absolute statements always have relative versions about families X → T ,
which greatly simplify the theory because of the freedom they provide.
How any morphism T → X may be understood as a point of X (parameterized by T ) and
that obviously such functor of points of X fully determines the whole structure of X, so reducing
definitions of morphisms to the silly case of sets –provided that they are natural definitions–
and constructions to the problem of representing a functor.
How sheaf cohomology is well-suited for Algebraic Geometry, Analysis and Topology.
How Grothendieck’s representability theorem is very simple in the case of categories of
abelian sheaves, and that it directly provides the existence of dualizing sheaves and dualiz-
ing complexes in the case of smooth curves (representing the functor H 1 (C, −)∗ , Riemann-
Roch’s theorem), smooth varieties (Serre’s duality), topological manifolds (Poincaré’s duality)
and proper morphisms (Grothendieck’s duality).
How toposes may unify Algebraic Geometry, Arithmetic and Topology.
And above all, how deeply rooted we have the wish of simple and natural definitions, state-
ments and proofs, and that such yearning may be always overwhelmingly accomplished.
The aim of these student notes based on the lectures of Sancho and his collaborators, updat-
ing them and redacting any topic as best as I know today, is to achieve a unified presentation of a
significant part of the university teaching in Mathematics. And also to develop the courses with
complete proofs, in a concise and coordinated way; mainly focusing on the logical and meaning
interdependencies of the involved topics, devoid of anything that would put the central points
in the shade. Hence the pages will be filled with definitions, statements and proofs, while with
scarce examples, applications and motivations. And the aim of this preface is to bring these
dry pages into life, to invite the reader to place them into the perspective of the much broader
Grothendieck’s oeuvre and life.
Fortunately Grothendieck has given us a very inspired and appealing description of its own
mathematical works in the Promenade à travers une oeuvre – ou l’enfant et la Mère 1 at the very
beginning of Récoltes et Semailles 2 , an unpublished work where (following the path of Descartes,
Pascal, Leibnitz,...) he has contributed to embed Mathematics as a significant part of a much
broader spiritual adventure: man’s self-conscience unfolding. In this Promenade he presents the
relationship of any man with the spiritual goods under a double aspect. On the one hand, the
luminous aspect (the passion for knowledge) represented by the figure of a child. On the other
hand, the obscure aspect of la peur et de ses antidotes vaniteux, et les insidieux blocages de la
1
Promenade through an oeuvre – or the child and the Mother.
2
Reapings and Sowings.
ii
creativité qui en derivent 3 . A deep and valuable lesson of the main body of Récoltes et Semailles
is summarized at the end of the Introduction:
Si quelque chose pourtant est saccagé et mutilé, et desamorcé de sa force originelle,

c’est en ceux qui oublient la force qui repose en eux-mêmes et qui s’imaginent saccager
une chose à leur merci, alors qu’ils se coupent seulement de la vertue créatrice de ce
qui est à leur disposition comme elle est à la disposition de tous, mais nullement à
leur merci ni au pouvoir de personne.4
This Promenade touched me deeply. First by the innocence of the child making mathematics,
of our children untiringly asking why? and what is it? instead of the weary and clumsy what is
it for? of adults. And above all by the silently, quite, attentive, feminine attitude of listening
our whispering interior voice and readiness to accept it, resounding me Mary’s fiat: be it unto
me according to your word. But the exigence of solitude in any creative work shocked me, and
when I wrote Grothendieck in 1987 that I missed it, he said me in a letter:
Vous soulignez, à juste titre, la difficulté psychique de la création solitaire... C’est

là la situation qui a été la mienne plus ou moins pendant toute ma vie, depuis mon
enfance, tant sur le plan de la création mathématique, que dans mon itinéraire spi-
rituel... Et sans cela, rien de grand ne s’accomplit, ni dans l’aventure individuelle,
ni dans l’aventure collective – que ce soit au plan intellectuel, ou au plan spirituel...
Au niveau spirituel, la plus grande ouevre (à mes yeux) qu’un homme ait accomplie,
était la Passion du Christ et sa mort sur la croix... Cette oeuvre était et ne pouvait
être que solitaire. Et même c’était la solitude suprême, car Dieu Lui-même s’est
retiré, pour que l’Oeuvre s’accomplisse sans le secours d’une consolation.5
In fact any great creative labour usually has a harsh period of solitude, and Grothendieck and
his parents had a very tough life, which is a paradigmatic and astonishing incarnation of the
whole XXth century history. A life that we may foresee in the autobiographical chapters III and
VI of La Clef des Songes 6 , another unpublished work where he embeds mathematics into man’s
religious adventure. He says:
Les lois mathématiques peuvent être découvertes par l’homme, mais elles ne sont
créés ni par l’homme ni même par Dieu. Que deux plus deux égale quatre n’est pas
un décret de Dieu, qu’Il aurait été libre de changer en deux plus deux égale trois,
ou cinq. Je sens les lois mathématiques comme faisant partie de la nature même de
Dieu – une partie infime, certes, la plus superficielle en quelque sorte, et la seule qui
soit accesible à la seule raison.7
3
the fear and the antidotes of vanity, and the insidious ways they block creativity (Avant-propos, page A3).
4
If something is despoiled and mutilated, defused of its initial force, it is in those unaware of the force resting
inside them, who think they are despoiling something at their mercy, while they are cutting themselves off from
the creative virtue of that which is at their disposal as this virtue is at the disposal of everyone, but not at their
mercy nor under the power of anyone (Introduction II.10, page xxii).
5
You remark, deservedly, the psychic difficulty of any lonely creation... Such has been my situation all my life,
since my childhood, in the mathematical creation and in my spiritual itinerary... And without that, nothing great
is accomplished, in the individual adventure, nor in the collective adventure – at the intellectual level or at the
spiritual level... At the spiritual level, the biggest oeuvre (in my eyes) accomplished by a man, was the Passion
of the Christ and his death on the cross... This oeuvre was and could only be solitary. It was even the supreme
solitude, since God Itself moved away, so that the Oeuvre is accomplished without the help of any consolation.
6
The Key of Dreams.
7
Mathematical laws may be discovered by man, but they are not created by man, nor even by God. That two
plus two equals four is not a decree of God that He is free to change into two plus two equals three, or five. I
sense the mathematical laws as being part of the very nature of God – a tiny part, certainly, the most superficial
in some sense, and the only part accessible to reason alone (III 31, Les retrouvailles perdues..., page 100).
iii
And without this skin, without that flesh and blood, and without that fire resting inside us,
the following hundreds of tight pages will only be a nude and dead skeleton.
Juan A. Navarro González

iv
Contents
Introduction 1
First Year 13
1 Analysis I 13
1.1 Integer and Rational Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Cardinal and Ordinal Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Real and Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.4 Metric Spaces and Topological Spaces . . . . . . . . . . . . . . . . . . . . . . . . 21
1.4.1 Continuous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.2 Completion of a Metric Space . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.5 Differential Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.6 Integral Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.7 Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.7.1 Elementary Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2 Linear Algebra 41
2.1 Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.1.1 The Quotient Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.2 Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.3 Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3.1 Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.4 The Dual Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.5 Euclidean Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.6 Diagonalization of Endomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.7 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.7.1 Alternate Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3 Algebra I 69
3.1 The Quotient Ring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2 Principal Ideal Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.3 Roots and Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3.1 Quadratic Irrationals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.3.2 Simple Fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.3.3 Operator Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.3.4 Root Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.3.5 Multiple Roots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.4 Rings of Fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.5 The Resultant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
v
vi CONTENTS
Second Year 91
4 Analysis II 91
4.1 Topological Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.3 The Lebesgue measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.4 Integral Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.4.1 Change of Variables Formula . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.5 Convergence in C m (U ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5 Algebra II 117
5.1 Actions of a Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.2 Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.2.1 Injective and Projective Modules . . . . . . . . . . . . . . . . . . . . . . . 122
5.2.2 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.2.3 Tensor Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.3 Categories and Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.3.1 Grothendieck Representability Theorem . . . . . . . . . . . . . . . . . . . 132
5.4 The Spectrum of a Ring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.6 Finite Algebras over a Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.6.1 Trivial and Separable Algebras . . . . . . . . . . . . . . . . . . . . . . . . 143
5.6.2 Galois Theory of Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
5.6.3 The Frobënius Automorphism . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.6.4 Radical Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.6.5 Inseparable Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
6 Projective Geometry 155

6.1 Projective Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
6.1.1 Affine Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.2 Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
6.2.1 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.2.2 Quadrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.3 Modules over a Principal Ideal Domain . . . . . . . . . . . . . . . . . . . . . . . . 170
6.3.1 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
6.3.2 Grothendieck K-Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.4 Classification of Pairs of Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6.5 Semisimple Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
6.5.1 The Brauer Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Third Year 189
7 Commutative Algebra 189

7.1 The Spectrum of a Ring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
7.2 Primary Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
7.3 Completion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
7.4 Dimension Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
7.5 Finite Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
7.6 Valuations and Dedekind Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
CONTENTS vii
7.7 Birrational Finite Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

7.8 Faithfully Flat Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
7.9 Galois Theory of Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
7.9.1 The Fundamental Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
8 Topology 215
8.1 Lattice Semirings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
8.2 Compact Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
8.2.1 Proper Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
8.3 Separation Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
8.4 Noetherian and Finite Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
8.5 Compactifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
8.6 Dimension Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
8.7 Uniform Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
8.8 Galois Theory of Coverings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
8.8.1 The Fundamental Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
8.8.2 Triangulated Compact Surfaces . . . . . . . . . . . . . . . . . . . . . . . . 245
9 Analysis III 249

9.1 Rings of Smooth Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
9.2 Ordinary Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
9.2.1 Uniparametric Groups and Lie Derivative . . . . . . . . . . . . . . . . . . 254
9.2.2 Pfaff Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
9.3 Integration of Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
9.3.1 Harmonic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
9.3.2 De Rham Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
9.4 Functions of a Complex Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
9.4.1 Meromorphic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
9.4.2 Convergence in O(X) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
10 Differential Geometry I 275

10.1 Smooth Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
10.1.1 Tensor Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
10.1.2 Smooth Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
10.2 Linear Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
10.2.1 Torsion and Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
10.3 Riemannian Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
10.3.1 Normal Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
10.3.2 Surfaces of Constant Curvature . . . . . . . . . . . . . . . . . . . . . . . . 292
10.4 Riemannian Embeddings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
10.5 Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
Fourth Year 300
11 Algebraic Geometry I 301

11.1 Sheaves and Presheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
11.2 Sheaf Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
11.3 Schemes and Coherent Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
11.4 Riemann-Roch Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
viii CONTENTS
11.4.1 Calculation of the Canonical Sheaf . . . . . . . . . . . . . . . . . . . . . . 316

11.5 The Projective Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
11.5.1 Projective Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
11.6 Complete Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
12 Algebraic Topology I 327

12.1 Cohomology with Supports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
12.2 Homological Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
12.2.1 The Functors TorA n
n and ExtA . . . . . . . . . . . . . . . . . . . . . . . . . 333
12.2.2 Derived Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
12.3 Inverse Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
12.4 Cup Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
12.4.1 Universal Coefficients and Künneth’s Theorem . . . . . . . . . . . . . . . 342
12.5 Locally Trivial Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
12.5.1 Vector Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
12.6 Local Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
12.6.1 Topological Intersection Theory . . . . . . . . . . . . . . . . . . . . . . . . 351
12.7 Duality Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
12.7.1 Degree Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
12.8 Characteristic Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
12.9 Spectral Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
13 Analysis IV 369
13.1 Dirichlet Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
13.1.1 Uniformization Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
13.2 Fréchet Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
13.2.1 Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
13.2.2 The Transpose Linear Mapping . . . . . . . . . . . . . . . . . . . . . . . . 381
13.3 Compact Riemann Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
13.3.1 Riemann-Roch Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
14 Differential Geometry II 395

14.1 Valued Differential Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
14.1.1 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
14.2 Calculus of Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
14.2.1 Problems in Dimension 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
14.3 Natural Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
14.4 Chern Classes and Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
Fifth Year 411
15 Algebraic Geometry II 411

15.1 Injective Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
15.2 Local Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
15.2.1 Regular Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
15.2.2 Depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
15.2.3 Local Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
15.3 Quasi-Coherent Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
15.4 K-Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
CONTENTS ix
15.4.1 Graded K-Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427

15.4.2 Cohomology theories and Chern Classes . . . . . . . . . . . . . . . . . . . 431
15.4.3 Grothendieck’s Riemann-Roch Theorem . . . . . . . . . . . . . . . . . . . 435
15.5 Duality Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
15.5.1 Calculation of the Dualizing Complex . . . . . . . . . . . . . . . . . . . . 442
15.5.2 Local Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
15.5.3 Biduality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
15.6 Formal Functions Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
15.7 Grothendieck Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
15.7.1 The Faithfully Flat Topology . . . . . . . . . . . . . . . . . . . . . . . . . 455
15.7.2 Sheafification of a Presheaf . . . . . . . . . . . . . . . . . . . . . . . . . . 459
Exercises 465
1. Analysis I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
2. Linear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
3. Algebra I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
4. Analysis II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
5. Algebra II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
6. Projective Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
7. Commutative Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
8. Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 508
9. Analysis III . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513
10. Differential Geometry I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 518
11. Algebraic Geometry I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523
12. Algebraic Topology I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 530
13. Analysis IV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
14. Differential Geometry II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
15. Algebraic Geometry II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
Index of Terms 547
Bibliography 563
x CONTENTS
Introduction
On the occasion of the Symposium in memory of late Prof. J. Sancho Guimerá held at
Salamanca in 2014, I decided to write some notes based on Sancho’s lectures, with the same
initial aim as Bourbaki at the 40’s: to undertake the university mathematical teaching from the
outset, with precise definitions, clear statements and complete proofs.
But, contrary to Bourbaki’s book, without avoiding sheaves and categories.
Hence, all the theorems included in these notes have concise and complete proofs, but this
is not a textbook for self-study of students in a standard sense:
1. It tries to reflect the terse style of the study notes of a student, always filled in with the
explanations of a lecturer. And always assuming the atmosphere of each course, with some
implicit assumptions and conventions, that surely will be clear to anyone who carefully reads
any section from the beginning. However, the defined concepts are always written in boldface
and they figure in the index of terms.
2. It focuses on the core of each course, the simple concepts and ideas that directly lead to
the main results, leaving aside details, examples and applications that surely are required to
assimilate the theory, but would put the central points in the shadow.
3. All the courses of a year must be studied simultaneously, as students actually do, and no
logical order may be imposed upon them, each one massively using definitions, results and
ideas from the companion chapters, sometimes placed at subsequent pages, due to the linear
character of any book. It is non sense to read a chapter alone, and all the courses of a year
must be studied simultaneously as a whole. So the course Linear Algebra uses some properties
of polynomials, proved in the companion but posterior chapter Algebra I ; the course Analysis
III massively uses from the start smooth manifolds, vector fields and differential forms (and
by the end metrics of constant curvature), studied in the companion but posterior chapter
Differential Geometry I ; and so on.
I do not pretend to develop each course in a self-contained way, without references to the
previous or simultaneous courses. On the contrary, I pretend to enlighten the mutual connec-
tions, to show that it would be unnatural to parcel out mathematics in several unrelated fields
(algebra, analysis, differential geometry, topology,...), that the main ideas are simple and every-
where fruitful, so reflecting the essential unity of Mathematics. Even if this choice force us to
use concepts, results and ideas eventually placed in a posterior chapter, when all the courses of
a year are written in a single book.
I have tried to be brief, since my aim is to exhibit many university courses, underlining
the logical interdependencies and the central concepts and methods. To show that, when each
course is developed with a regard to the resources and needs of the other parts of Mathematics,
the main results have more natural and simple proofs. To show that, again and again, when a
gentle hand provides the correct definitions, the adequate point of view, surprisingly the knots
untie, the difficulties dissolve and the backbone of each theory may be exposed in a handful of
pages.
I have been able to put black on white fifteen annual courses, with the courses of each year
assumed to be studied at the same time:
2 INTRODUCTION
First Year
• Analysis I: Cardinal and ordinal numbers. Real and complex numbers. Metric spaces and
topological spaces. Differential and integral calculus. Power series.
• Linear Algebra: Vector spaces and linear maps. The dual space. Euclidean vector spaces.
Tensors and p-forms.
• Algebra I: The quotient ring. Principal ideal domains. Field extensions and roots. Rings of
fractions. Elimination.
Second Year
• Analysis II: σ-compact spaces. The Lebesgue measure. Differential and integral calculus
with several real variables. Convergence in rings of differentiable functions.
• Algebra II: Actions of a group. Modules. Categories. Spectrum of a ring. Module of

differentials. Finite algebras over a field. Galois theory of fields.
• Projective Geometry: Projective and affine spaces. Classification of metrics, modules over
a principal ideal domain, and pairs of metrics. Semisimple rings and the Brauer group.
Third Year
• Commutative Algebra: Noetherian rings. Primary decomposition. Completion. Dimen-

sion theory. Integral dependence. Dedekind domains. Galois theory of rings.
• Topology: Lattice semirings. Separation properties. Noetherian spaces. Compactifications.

Dimension theory. Uniform spaces. Galois theory of coverings. The fundamental group.
• Analysis III: Rings of smooth functions. Ordinary differential equations. Pfaff systems.
Integration of differential forms. De Rham cohomology. Riemann surfaces. Meromorphic
functions. Riemann mapping theorem.
• Differential Geometry I: Smooth manifolds. Tensor fields and the exterior differential
calculus. Linear connections. Riemannian metrics. Lie groups.
Fourth Year
• Algebraic Geometry I: Sheaf cohomology. Schemes and coherent sheaves. Algebraic curves
and the Riemann-Roch theorem. The projective spectrum.
• Algebraic Topology I: The cohomology ring. Homological algebra. Local cohomology.

Duality theorem. Characteristic classes. Spectral sequences.
• Analysis IV: Uniformization theorem. Fréchet spaces. Compact Riemann surfaces.
• Differential Geometry II: Valued differential calculus. Calculus of variations. Natural

bundles. Chern classes and curvature.
Fifth Year
• Algebraic Geometry II: Local algebra. Quasi-coherent sheaves. K-theory. Duality theory.
Grothendieck topologies and descent theory.
3
and many significant parts of the university teaching lack (functional analysis and partial dif-
ferential equations, homology and homotopy theory, number theory, measure theory and statis-
tics, mathematical physics, etc.), but I think that these courses suffice to show that many
Grothendieck ideas may be introduced at the undergraduate level when the courses are devel-
oped in a coordinate way.
The first three years are mainly devoted to the concepts of structure and morphism, the
central results being “Galois type” theorems stating the equivalence of two different structures
(i.e. of two categories), while the last two years focus on the concepts of sheaf and cohomology,
the central results being duality theorems. Moreover, representable functors and the concepts
of point and space pervade the notes from start to finish, as well as the clarification of the
adjectives natural, intrinsic, canonical and universal that we frequently use in a naive sense.
Here it is a brief description of the sense and content of each annual course:
Analysis I: We begin with the construction of integer, rational, real and complex numbers,
the genetic method of reducing all mathematical concepts to the process of counting (sets and
natural numbers); so initiating the accomplishment of the fantastic pythagorean vision8 : all is
a music of numbers. This genetic method is the backbone of Analysis and culminates in the
theory of functions of several complex variables and the theory of partial differential equations.
We study cardinal numbers, and we put |X| ≤ |Y | when there is an injective map X → Y .
The main result is that any set of cardinal numbers has a first element. So, there is the first
infinite cardinal ℵ0 = |N|, the first cardinal ℵ1 bigger than ℵ0 ,... and so we may parameterize
the infinite cardinals by the ordinal numbers.
We construct the real numbers using Cauchy sequences of rational numbers, and generalize
this method to complete any metric space. We also prove that the compact sets in Rn are just
the bounded closed sets, a central result that readily proves that the image of any continuous
function f : [a, b] → R is a closed interval, f ([a, b]) = [m, M ], a statement encoding the main
results on continuous functions.
After an exposition of the Infinitesimal Calculus, using the Riemann integral, we show that
a bounded function f : [a, b] → R is Riemann-integrable if and only if it is almost everywhere
continuous. We end with an introduction to the power series expansions, and we use them to
study the exponential and trigonometric functions.
Linear Algebra: In this chapter we take the path of the axiomatic method. Geometry bursts
luminously in Greece when the Greek genius realized that properties of such subtle and funda-
mental concepts as point, line, right angle, etc., should not be reduced to properties of previous
concepts, but that we should better put on show their basic mutual relations, stating them as
axioms or postulates of the theory, and then derive the remaining properties. In modern terms,
any question is submerged in an implicit structure, and a breaking point is to put it into light.
Besides, in the 19th century the German world discovered that it is crucial to clarify what struc-
tures are considered to be equal (the isomorphisms) and it placed as the central problem of any
theory the classification of all possible structures up to isomorphisms9 . As a first example of the
8
Aristotle (Metaphysics, Book I, Part 5, 986a): τ α τ ων αριθµων στ ωιχια τ ων oντ ων στ ωιχια παντ ων
υπλαβoν ιναι, και τ oν oλoν oυρανoν αρµoνιαν ιναι και αριθµoν. ([The pythagoreans supposed] the
numbers to be the elements of all things, and the whole heaven to be a musical scale and a number).
9
One of the first to glimpse such classification problems was Goethe, as he wrote in his diary entries (Naples,
May 17, 1787): Die Urpflanze wird das wunderlichste Geschöpf von der Welt, um welches mich die Natur selbst
beneiden soll. Mit diesem Modell und dem Schlüssel dazu kann man alsdann noch Pflanzen ins Unendliche
erfinden, die konsequent sein müssen, das heißt, die, wenn sie auch nicht existieren, doch existieren könnten
und nicht etwa malerische oder dichterische Schatten und Scheine sind, sondern eine innerliche Wahrheit und
Notwendigkeit haben. (The Urpflanze is going to be the most wonderful creation of the world, which Nature herself
shall envy me. With this model and the key to it, one can then go on inventing plants ad infinitum, which must
4 INTRODUCTION
great insight of the German point of view, just compare the genetic definition of the addition
of integer numbers given in p. 13 with the definition steaming from the classification of cyclic
groups (p. 44): it is the unique possible operation defining a structure of infinite cyclic group
(so recovering our infantile vision: what is added does not matter, only how to do the addition).
After a very concise introduction to the structures of group and ring, the course focuses
on the structure of vector space (so that we may exhibit for the first time the structure of the
Euclidean Geometry10 , at least with a fixed point and a fixed length unit) and we prove that
vector spaces are classified by the dimension, whenever it is finite.
A big emphasis is put on universal properties (our first contact with representable functors)
which simplify the theory in many points. For example in the intrinsic definition of the con-
traction of indices, otherwise cumbersome and in need of some checking, and in the proof of the
natural isomorphism (Λp E)∗ = Λp E ∗ . The systematic use of linear maps and exact sequences
from the very beginning of the course also simplifies the theory.
Algebra I: This chapter is devoted to the study of the structure of ring. We understand it
as a generalized arithmetic: numbers are ideals and the divisibility relation is the inclusion
relation, so that prime numbers are maximal ideals, the quotient ring encoding the gaussian
theory of congruences. The main result is Kronecker’s formula for a root of any polynomial
equation p(x) = 0, discovered when he realized that the extraction of a root involves two quite
different steps. The first one, subtle, humble and usually of long gestation, is the highlight of a
structure of field (radical expressions, real or complex numbers, etc.); while the second one is to
express a root inside such structure. But men were looking for a root without questioning the
convenience of the structure where they limited the search, enclosed in an invisible iron circle11
that Kronecker innocently crossed when he discovered that, if p(x) is irreducible, an obvious
be consistent; that is, even if they do not exist, they could exist and not as picturesque or poetic shadows and
illusions, but with inner truth and necessity). In his quiet walks along the Italian botanical gardens, he dreamed
about the plant structure and its possible realizations, so that he could know all the plants that Linnaeus had
catalogued, the unknown plants of Africa and America that brave and courageous explorers were looking for, the
extinct plants, the future ones and those that will remain in God’s mind forever. Even if he could not achieve
this wonderful dream, it is not a mere coincidence that the periodic table of elements and the first classifications
of mathematical structures and elementary particles were so close to Goethe, in time, space and spirit.
10
Stricto senso, the higher dimensional real Euclidean geometries studied in this course were the first “Non
Euclidean” geometries, glimpsed in 1747 by Immanuel Kant (Gedanken von der wahren Schätzung der lebendi-
gen Kräfte §10–11): Diesem zu folge halte ich dafür: daß die Substanzen in der existirenden Welt, wovon wir
ein Theil sind, wesentliche Kräfte von der Art haben, daß sie in Vereinigung miteinander nach dem doppelten
umgekehrten Verhältniß der Weiten ihre Wirkungen von sich ausbreiten; zweitens, daß das Ganze, was daher
entspringt, vermöge dieses Gesetzes die Eigenschaft der dreifachen Dimension habe; drittens, daß dieses Gesetz
willkürlich sei, und da Gott dafür ein anderes, zum Exempel des umgekehrten dreifachen Verhältnisses, hätte
wählen können; daß endlich viertens aus einem andern Gesetze auch eine Ausdehnung von andern Eigenschaften
und Abmessungen geflossen wäre. Eine Wissenschaft von allen diesen möglichen Raumesarten wäre unfehlbar die
höchste Geometrie, die ein endlicher Verstand unternehmen könnte... Wenn es möglich ist, daß es Ausdehnungen
von andern Abmessungen gebe, so ist es auch sehr wahrscheinlich, daß sie Gott wirklich irgendwo angebracht hat.
Denn seine Werke haben alle die Größe und Mannigfaltigkeit, ist, daß es die sie nur fassen können. (Thoughts
on the True Estimation of Living Forces §10–11: Accordingly, I am of the opinion that substances in the existing
world, of which we are a part, have essential forces of such a kind that they propagate their effects in union with
each other according to the inverse square of the distances; secondly, that the whole to which this gives rise has,
by virtue of this law, the property of being three dimensional; thirdly, that this law is arbitrary, and that God
could have chosen another, e.g., the inverse-cube, relation; fourthly, and finally, that an extension with different
properties and dimensions would also have resulted from a different law. A science of all these possible kinds of
space would undoubtedly be the highest geometry that a finite understanding could undertake... If it is possible
that there are extensions of different dimensions, then it is also very probable that God has really produced them
somewhere. For His works have all the greatness and diversity that they can possibly contain).
11
Our questions always have an implicit horizon embracing all the possible answers. To make it clear, and to
extend it if necessary, are the strong moments of man’s self-conscience unfolding.
5
root is just the class [x] in the field of residue classes modulo p(x), so reducing the extraction of
roots to the problem of decomposing polynomials into irreducible factors.
At first glance Kronecker’s astonishing answer may seem a mere tautology or a sterile for-
mality. Certainly it is a tautology presented with all the formal rigor we are able, as any other
theorem; but, after applying it to prove D’Alembert’s and Hamilton-Cayley theorems (p. 76
and p. 59), and to solve the millennial questions on ruler-and-compass constructions (p. 78),
we leave to the reader the consideration that the adjectives mere and sterile deserve.
Analysis II: After a brief study of topological spaces, we study the differential calculus with
functions of several real variables. The main tool is a key lemma in the calculus with infinitesi-
mals: If fPis a function of class C m on an open ball U around the origin 0 ∈ Rn and f (0) = 0,
then f = i fi xi for some functions fi of class C m−1 . This key lemma is proved using Barrow’s
rule and the differentiation rule under the integral sign, and it readily gives Schwarz’s theorem,
∂i ∂j f = ∂j ∂i f , and Taylor’s expansions; and it also gives the inverse mapping theorem, using
the relative
P version along the diagonal: any function f ∈ C m (U × U ), vanishing on the diagonal,
is f = i fi (yi − xi ) for some functions fi ∈ C m−1 (U × U ).
Then we study the Lebesgue measure and the corresponding integration of functions, the
main results being the change of variables formula and Sard’s theorem. We prove both using
the key lemma along the diagonal, that shows that in a neighborhood of a point p, the image of
any cube is very close to the image by the tangent linear map at p.
Finally, we introduce topological vector spaces and we prove that C m (U ) is complete with
the topology of the compact convergence of the functions and the iterated partial derivatives.
Algebra II: This is a crucial course with an explosion of many central ideas and constructions
that will pervade the remaining chapters.
1. A big step is given in the clarification of the concepts of point and function. Any commutative
ring A defines a topological space Spec A whose points correspond to the prime ideals of A,
the elements of A are viewed as functions on Spec A, and the ideals of A as equations of
geometric loci. Integers, gaussian integers, etc., may be understood as functions, and we may
apply them our geometric intuitions and resources, so starting the unification of Arithmetic
and Geometry dreamed by Kronecker.
2. Localization of modules, proving that most properties of modules are local12 , in the sense
that they only have to be checked on the local rings of the points of Spec A.
3. Tensor products. Hence the base change of algebras and modules, so that we may introduce
the concepts of geometric property (stable under changes k → K of the base field) and local
property (valid whenever it holds after some base change k → K).
4. Categories and functors, aiming to give a first rigorous approximation to the fundamental
concepts of structure, natural and canonical.
5. Points of Spec A with coordinates in a field K are viewed as morphisms Spec K → Spec A and,
in general, Grothendieck’s representability theorem gives necessary and sufficient conditions
for a functor to be a functor of points.
6. The module of differentials, giving an algebraic treatment of the differential calculus.

12
A basic fact that we constantly use throughout all the notes, even without further mention. Surprisingly, any
other property is easily reduced to the critical property of being 0. Here, as in any other case, nothingness, as
soon as it is senseful, becomes a driving force.
6 INTRODUCTION
7. Injective and projective modules, a basic tool of the forthcoming Homological Algebra.
The central result is Galois theorem, stated as an antiequivalence of the category13 of k-

algebras trivial on a given Galois extension k → L of Galois group G with the category of finite
G-sets. The proof14 directly follows from the fundamental exact sequence k → A ⇒ A ⊗k A and
the stability of the algebra of invariants AG under base changes, (AG ) ⊗k L = (A ⊗k L)G .
We also introduce the Frobënius automorphism of a polynomial q(x) ∈ Z[x] at a prime p, so
relating the global properties of q(x) to the behavior of its reductions q̄(x) ∈ Fp [x].
Finally, we prove that the maximal separable subalgebra is stable under base changes k → K,
and the elementary theory of inseparable k-algebras readily follows.
Projective Geometry: We introduce the projective space P(E) attached to a vector space E,
but only as a mere set, without elucidating the underlying structure. To figure out the implicit
structure grounding projective geometry has been a major topic in the history of Geometry,
and so it will be in these notes. Nevertheless, in this course at least we know the group of
automorphisms of such elusive structure: an isomorphism of a k-vector space E into a k 0 -vector
space E 0 clearly should be defined as a group isomorphism f : E → E 0 such that f (λe) =
σ(λ)f (e) for some ring isomorphism σ : k → k 0 ; hence the group of automorphisms of P(E) is
the group P Sl(E) of all bijections P(E) → P(E) represented by a semilinear automorphism
E → E.
Klein, in his foundational Erlangen Programme, states that the structure of the projective
space is given by the action of the group P Gl(E) of all projectivities15 on P(E), and that in
general any action of a group G on a set defines a geometry: concepts of the geometry being
just G-invariant concepts, statements being relations between concepts, and theorems being
true relations. Considering different subgroups of P Gl(E), in this course we also study affine
geometry, Euclidean geometry and Non-Euclidean geometries. In this course we show that
the structure of the projective spaces of dimension ≥ 2 is captured by the lattice of linear
subvarieties16 ; but to figure out the structure of the real projective line as a ringed space (so
grounding projective geometry on non-algebraically closed fields) was a first success of the theory
of schemes.
On the other hand, any concept of a geometry gives rise to a classification problem, con-
sidered as the determination of the set of orbits of the induced G-action. In this course we
study the projective and affine classifications of quadrics (reduced to the linear classification of
symmetric metrics), the projective classification of projectivities (reduced to the linear classi-
fication of endomorphims) and the projective classification of pencils of quadrics (reduced to
the classification of metrics on modules over the finite k-algebras k[x]/(p(x)) considered in the
companion course Algebra II ). In this sense, this is a second course in Linear Algebra.
We classify endomorphisms, and finitely generated modules over a principal ideal domain A,
using the crucial fact that B = A/pn A is an injective B-module when p is irreducible, a result
that follows directly from the Ideal Criterion studied in Algebra II.
It is worthy of attention that, in the former course Linear Algebra, the characteristic polyno-
mial of an endomorphism was introduced in a non-intrinsic way, defining it in a base and checking
13
In fact the category of finite direct sums of intermediate fields, so that the theorem not only determines the
intermediate fields but also their k-morphisms.
14
Two alternative proofs are sketched in exercises 105 and 106, p. 488. The first one reduces the proof to the
simple case k = L, once the exactness properties of the involved functors are checked; while the second one uses
Grothendieck’s characterization of the forgetful functor on the category of finite G-sets.
15
so that the group of automorphisms of P(E) is just the normalizer of P Gl(E) in the group of all bijections
P(E) → P(E), which may be shown (ex. 9, p. 492) to be P Sl(E).
16
The so called fundamental theorem of projective geometry (p. 159); but today such fundamental theorem,
illuminating the structure of projective spaces, is the universal property of P(E) given in p. 322.
7
that it does not depend on the base. Now we introduce the K-group to give Grothendieck’s
astonishing definition of the characteristic polynomial: it is the universal additive function on
the finitely generated torsion k[x]-modules.
Finally, we study semisimple rings and the Brauer group, obtaining that the quaternions are
the unique non-commutative finite extension of the real numbers.
Commutative Algebra: First we introduce the “structural sheaf” of the spectrum of a ring,
so that we may present all the topics with its full geometric comprehension and flavor:
1. Primary decomposition of ideals and completion.
2. Krull dimension theory and regular rings.
3. Finite morphisms and birrational finite morphisms, obtaining the desingularization of curves,
and the calculation of intersection numbers, by quadratic transformations.
Finally we show that the theory of unramified finite flat ring morphisms A → B is totally
analogous to the Galois theory of separable finite k-algebras and that, when considering the
induced morphism Spec B → Spec A, it is totally analogous to the theory of coverings of a
topological space, so obtaining the definition of the fundamental group π1 (Spec A, p) of Spec A
at a geometric point p : Spec k̄ → Spec A.
In these notes the Galois theory of fields, of noetherian rings and of coverings of a topological
space, have a unified presentation, with equal statements and proofs.
Topology: When studying continuous functions, the particular value of a function f at a point
x usually does not matter, only wether f (x) = 0 or f (x) 6= 0. So it is natural to identify all the
non-zero real numbers, and so we obtain K = {0, g}, with a closed point 0 and a dense point g
(the generic real number). We have a natural structure of ”field” with unity g = 1 on K,
0+0=0 0+g =g+0=g g+g =g

0·0=0 0·g =g·0=0 g·g =g
and it is clear that any closed set Y in a topological space X is the zero-set of a unique continuous
function fY : X → K, so that the lattice A(X) of all closed sets in a topological space X may
be viewed as a ”ring” of continuous functions, A(X) = C(X, K), the addition corresponding to
the intersection, and the product to the union17 ,
fY + fZ = fY ∩Z , fY · fZ = fY ∪Z .
Many operations introduced in the elementary theory of rings (quotient ring, localization,
tensor product, spectrum, etc.) remain meaningful in the framework of lattice semirings, and
they preserve their geometric meanings. Moreover, most statements and proofs remain valid,
and the purpose of this chapter is to show that this comprehension of the lattice of closed sets
A(X) as a ”ring” of functions provides a handy tool in some topics:
1. The functor Spec clarifies the theory of compactifications.
2. Finite topological spaces are dual to finite lattice semirings, and essentially encode the theory
of polyhedra.
17
In fact A(X) is not a true ring, since the existence of opposite element fails, but it comes with the extra
properties 1 + a = 1 and a2 = a. We name them lattice semirings. In a suggestive way, Connes uses to say that
semirings where 1 + 1 = 1 have characteristic 1, so that K is a semifield of characteristic 1.
8 INTRODUCTION
3. Typically the Krull dimension of A(X) is infinite, but Sancho realized that the minimal Krull
dimension of a base18 B ⊆ A(X) provides a good definition for the dimension of a topological
space X, since it coincides with Grothendieck’s dimension (the maximal length of a chain of
irreducible closed sets) when X is a noetherian space, and with the inductive dimension and
Lebesgue cover dimension when X is a separable metric space.
Finally we study uniform spaces, and the theory of coverings X → S is developed parallel to
the Galois theory of fields and rings, taking advantage of the existence of a universal covering
with Galois group π1 (S, p).
Analysis III: First we study partitions of unity and we show how a smooth manifold X may
be reconstructed from the ring of smooth functions C ∞ (X), the basic facts being the natural
homeomorphism X = SpecR C ∞ (X) and the coincidence of C ∞ (U ) with the ring of fractions of
C ∞ (X) with respect to the multiplicative system of all smooth functions without zeros in the
open subset U .
In the theory of ordinary differential equations, we show that the argument of the contrac-
tive map, classically used to prove the existence and uniqueness of a solution, also proves the
continuous and differentiable dependence on the initial conditions when considering adequate
functional spaces.
Then we study Pfaff systems, the central result being a geometrically obvious proof of a
projection theorem given necessary and sufficient condition for a Pfaff system P on a smooth
manifold X to be projectable by a regular projection π : X → Y , in the sense that P = π ∗ Q for
some Pfaff system Q on Y . As a direct consequence we obtain Frobënius theorem (involutive
distributions are integrable) and Darboux’s classification of germs of 1-forms.
We also study the integration of differential forms on smooth manifolds, obtaining the fun-
damental Stoke’s theorem. We apply it to harmonic functions, De Rham cohomology and the
theory of complex analytic functions.
Finally, we introduce Riemann surfaces as ringed spaces. It is easy to prove that locally any
non constant analytic morphism is u = z n for some exponent n ≥ 1 and, as a direct consequence
of this local classification of morphisms, most elementary properties of Riemann surfaces follow.
We prove the Riemann mapping theorem with a systematic use of the hyperbolic plane geometry.
Differential Geometry I: We introduce smooth manifolds as ringed spaces X locally isomor-

phic to open sets in Euclidean spaces with the sheaf of smooth functions (so eliminating the
cumbersome atlases) and submanifolds as topological subspaces Y ,→ X such that (Y, CY∞ ) is
a smooth manifold, where CY∞ is the sheaf of restrictions of smooth functions on X. Then we
develop the tensor calculus and the exterior differential calculus on smooth manifolds.
We study linear connections, the main result being that any torsion free linear connection of
null curvature is locally Euclidean, and riemannian manifolds, proving that any simply connected
and complete riemannian surface of constant curvature is isometric to the Euclidean plane, the
hyperbolic plane or the sphere. Then we present the theory of riemannian embeddings.
Finally we present a bit of the theory of Lie groups, up to the point of obtaining the classi-
fication of abelian Lie groups.
Algebraic Geometry I: This chapter may be considered the hearth of these notes. We in-
troduce sheaves and we define the cohomology groups H p (X, F) using Godement’s resolution.
Then we prove a cohomological bound theorem
H p (X, F) = 0, p > dim X,

18
a subring whose complements form a base of open sets of X in the usual sense.
9
when X is a noetherian space, and the acyclicity theorem for arbitrary rings,
H p (Spec A, M
f ) = 0, p > 0.
Now, after a massive use of the spectrum of a ring in the courses Algebra II, Topology and
Commutative Algebra, and the comprehension of smooth manifolds and Riemann surfaces as
ringed spaces in Differential Geometry I and Analysis III, we may naturally introduce schemes
as ringed spaces locally isomorphic to (Spec A, Ae ). The major example of scheme in this course
will be the Riemann variety of a finite extension of k(t), the Riemann variety of k(t) itself being
the projective line P1 over the field k.
First we calculate the cohomology groups of the structural sheaf of P1 , projecting it onto a
finite space with two closed points and a dense point, and it readily follows the determination
of the cohomology groups of the line sheaves OP1 (n). Then we prove the fundamental finiteness
theorem for coherent sheaves, dim H p (X, M) < ∞, projecting the curve X onto P1 , since the
direct image preserves cohomology.
Now the weak Riemann-Roch theorem for line sheaves LD is immediate,
dim H 0 (X, LD ) − dim H 1 (X, LD ) = 1 − g + deg D.
Since, H 0 (X, LK−D ) = HomOX (LD , LK ), to prove the strong Riemann-Roch theorem we
only have to see that
dim H 1 (X, LD ) = dim HomOX (LD , LK ),
where K stands for a canonical divisor of the curve X. Since H 2 (X, M) = 0, Grothendieck’s
representability theorem directly gives the existence of a dualizing sheaf:
H 1 (X, M)∗ = HomOX (M, DX )
for some quasi-coherent sheaf DX , and the problem is to determine DX .

It is easy to show that DX is a line sheaf, but it is hard to prove that it is the sheaf of
differentials. We present two proofs of this fundamental result, the first one based on a laborious
local calculation of the conductor of a projection X → P1 , while the more natural second proof
uses the diagonal embedding X → X ×k X and the stability of the cohomology of coherent
sheaves under base changes (so pointing that the theory of curves is naturally entangled with
the theory of higher dimensional varieties).
Finally we introduce the projective spectrum of a graded ring, and determine the cohomology
groups of the line sheaves OPd (n) over the projective space Pd = Proj k[x0 , . . . , xd ].
Algebraic Topology I: This chapter is devoted to the cohomology of sheaves over σ-compact
spaces. After introducing the fundamental exact sequences (Mayer-Vietoris, closed subspace
and local cohomology exact sequences) we determine the cohomology groups H p ([0, 1]n , Z) of
cubes, hence of spheres and many classical spaces.
Then we introduce the inverse image and the cup product, and we prove some fundamental
results: calculation of the cohomology of the fibres, theorem of base change, finiteness theorem,
universal coefficients formula, Künneth theorem, and the classification of line sheaves. We also
use local cohomology groups to define the cohomology class of a closed submanifold and to
develop a topological intersection theory.
Now we get down to the duality theorem. So as to include non locally Euclidean spaces, we
do not consider the dual of the last cohomology group Hcn (X, F), but of a complex determining
all the cohomology groups with compact supports. Again the representability theorem gives the
existence of a dualizing complex DX • such that
•
RHom(RΓc (F), Z) = RHom(F, DX )
10 INTRODUCTION
and in this case it is not hard to determine it when X is a topological manifold of dimension n:
the dualizing complex DX • has a unique non-zero cohomology sheaf19 T , locally constant and
X
placed at degree −n, so that HomZ (Hcn (X, F), Z) = Hom(F, TX ).
Then we determine the cohomology groups of a projective bundle, a result grounding the
theory of characteristic classes for vector bundles.
Finally we introduce some useful spectral sequences.
Analysis IV: After using Perron’s method to solve the Dirichlet problem, we use it and sheaf
cohomology to prove the uniformization theorem: any simply connected Riemann surface is the
complex plane, the complex projective line or the unit disk.
Then we present a brief introduction to Fréchet spaces and the duality theory of locally
convex spaces, up to the point of obtaining Schwartz theorem, the crucial technical tool to prove
(using Cech cohomology) that locally free sheaves on Riemann compact surfaces always have
finite dimensional cohomology groups.
This finiteness theorem readily gives the existence of non constant meromorphic functions,
and the equivalence of the category of compact Riemann surfaces to the category of complete
and non singular algebraic curves over the field C, studied in the companion course Algebraic
Geometry I. Hence, most of the results there obtained for complete non singular curves over
arbitrary fields may be extended to compact Riemann surfaces, in particular the fundamental
Riemann-Roch theorem. However, we show that now the residue theorem provides a very simple
and natural proof of the coincidence of the dualizing sheaf with the sheaf of analytic differentials,
the fundamental result that was so hard to prove for algebraic curves.
Differential Geometry II: Many mathematical and physical concepts may be viewed as dif-
ferential forms with values in a vector bundle (or a locally free sheaf, the point of view adopted
in this chapter) and the differential calculus with such forms requires a linear connection on
the vector bundle. We develop this fruitful differential calculus up to the point of obtaining
Bianchi’s identities and Cartan’s structure equations. We also determine the Chern classes of a
complex vector bundle in terms of the curvature 2-form of any linear connection.
We present the variational calculus, obtaining the fundamental Poincaré-Cartan form and
Noether’s theorem on the infinitesimal symmetries of a variational problem.
Finally, since most bundles used in Geometry and Physics are natural bundles, we include
the Galois theory of natural bundles (of a given finite order). The statements are parallel to
those of the Galois theory of fields, rings and coverings, but unfortunately proofs differ.
Algebraic Geometry II: We begin with a study of Local Algebra, obtaining Serre’s theorem
on regular rings and some characterizations of Cohen-Macaulay rings. Then, after a brief study
of quasi-coherent sheaves (including Deligne’s formula and the base change theorem) we get
down to the central topics of this course: the Riemann-Roch and duality theorems.
1. First we study the K-theory of coherent sheaves up to the point of obtaining Chern classes of
vector bundles with values in the graded K-theory GK • (X). Then we prove the fundamental
result: the K-theory is the universal multiplicative20 cohomology theory.
Now, if a cohomology theory A(X) follows the additive law c1 (L ⊗ L0 ) = c1 (L) + c1 (L0 ), then
we may modify the direct image in A(X) ⊗ Q with an exponential, so that it follows the
multiplicative law and we have a functorial ring morphism K(X) → A(X) ⊗ Q preserving
the new direct image. This is just the Riemann-Roch-Grothendieck theorem.
19
that would be obtained in case of dualizing the last cohomology group, as in Algebraic Geometry I.
in the sense that Chern classes of line sheaves follow the law c1 (L ⊗ L0 ) = c1 (L) + c1 (L0 ) − c1 (L)c1 (L0 ) of the
20
multiplicative group. Remark that (1 − x)(1 − y) = 1 − (x + y − xy).

11
A direct consequence, in the complex case, is the existence of a natural ring morphism
GK • (X) ⊗ Q → H 2• (Xan , Q) preserving inverse and direct images, hence Chern classes.
Since GK 0 (Spec C) ⊗ Q = H 0 (pt, Q) = Q, the algebraic and topological definitions of any
numerical cohomological invariant coincide. For example, if X is a projective smooth variety,
the topological and algebraic self-intersection numbers of the diagonal coincide,
i i p+q dim H p (X, Ωq ).
P P
i (−1) dimQ H (Xan , Q) = p,q (−1) C X
2. Given a projective morphism f : X → S, the representability theorem gives the existence of

a dualizing complex DX/S such that for any quasi-coherent OX -module M,
RHomOS (Rf∗ (M), OS ) = RHomOX (M, DX/S )
and again the crux is to calculate it in the case of a smooth morphism; to prove that it is the
highest exterior power of the sheaf of differentials, DX/S ' ΩdX/S .
Using the Koszul complex, we may see that the dualizing sheaf of a closed regular embedding
Y → X of codimension d is just Λd NY /X , where NY /X stands for the normal bundle. Now
the calculation of the dualizing sheaf of a smooth projective morphism X → S is a simple
question. Just consider the composition
∆ π
X −−−→ X ×S X −−−1−→ X
of the diagonal embedding with the first projection (obviously it is the identity), and the
normal bundle N to the diagonal embedding (which is dual to the sheaf of differentials ΩX/S
by the very definition). We have
OX = DX/X = DX/X×X ⊗ ∆∗ DX×X/X = Λd N ⊗ ∆∗ (π1∗ DX/S ) = Λd N ⊗ DX/S
and we are done: DX/S = (Λd N )∗ = ΩdX/S . Once the question is placed in the relative and
general setting (morphisms and arbitrary dimension instead of the absolute case of curves)
the obvious compatibility properties of the theory dissolve the question.
Then we prove the local duality theorem. Again, dualizing the local cohomology groups at
a point x, the representability theorem directly gives the existence of a local dualizing complex,
and the central point is to show that, in the case of a projective variety, it is just the stalk at x of
the global dualizing complex. In the crucial case of a smooth variety X, we prove it determining
the local cohomology groups Hxp (X, OX ).
Finally, we introduce Grothendieck topologies, proving that quasi-coherent OS -modules and
S-schemes define sheaves on the category of S-schemes with the fpqc topology, and that locally
quasi-coherent sheaves are quasi-coherent (descent of quasi-coherent sheaves).
Requirements: From a logical point of view, only the elementary properties of natural numbers
and the intuitive set theory (including Zorn’s lemma) are required, and further foundational
questions are avoided. In fact, I feel that Zermelo-Fraenkel’s axioms do not ground properly set
theory: they include unions of arbitrary sets and assume that the elements of a set are always
sets, while in mathematics, only unions of subsets of a given set are sensible, and it is non-sense
to consider the elements of the elements of any given set21 .
21
So, even if the spectrum Spec A of a ring A is defined to be the set of all prime ideals of A, the statements
0 ∈ x ∈ Spec A and 1 ∈ / x ∈ Spec A are non-sense: points of Spec A are not prime ideals of A, but naturally
correspond to prime ideals. And the elements of the quotient set are not equivalence classes, but naturally
correspond to equivalence classes.
12 INTRODUCTION
However, along the five years, an increasing maturity level is assumed.
Notations: Exceptionally we use some non-standard notations. The covariant derivative of a

tensor field T in the direction of a vector field D is denoted D∇ T , instead of the usual ∇D T ;
and the Lie derivative DL T , instead of the usual LD T .
Sometimes Tpq denotes a tensor of type (p, q); so R2,2 denotes the Riemann-Christoffel tensor,
R2 denotes the Ricci tensor, a p-form is Ωp , a metric is S2 , and so on.
When a fixed map f : X → Y is unambiguously understood, and Z ⊆ Y , we put
X ∩ Z := f −1 (Z) = {x ∈ X : f (x) ∈ Z},
so that X ∩ Z is a subset of X but not a subset of Z.

The one point set is denoted by ∗.
Acknowledgements: My debt and gratitude to my teachers at the University of Salamanca

in the seventies, mainly D. Juan Bautista Sancho Guimerá and his collaborators D. Antonio
Pérez-Rendón Collantes, D. Pedro Luis Garcı́a Pérez, D. Cristóbal Garcı́a-Loygorri y Urzaiz, D.
Jesús Muñoz Dı́az and D. Jaime Muñoz Masqué.
Also my grateful acknowledgement to many companions who have written notes on these
courses, with many improvements of their own, since I have made good use of them: Daniel
Hernández, José M. Muñoz, Juan Sancho Jr. and his brothers Teresa, Carlos, Pedro and Fer-
nando, my friend Ricardo Faro and my sons José and Alberto.
Chapter 1
Analysis I
1.1 Integer and Rational Numbers

Definitions: A relation ≡ on a set X is an equivalence relation if it is
1. Reflexive: x ≡ x, ∀x ∈ X.
2. Symmetric: x, y ∈ X, x ≡ y ⇒ y ≡ x.
3. Transitive: x, y, z ∈ X, x ≡ y, y ≡ z ⇒ x ≡ z.
The equivalence class of x ∈ X is x̄ = [x] = {y ∈ X : x ≡ y}.
A subset C ⊆ X is an equivalence class when C = [x] for some x ∈ X.
The quotient set X/ ≡ is the set of all equivalence classes (or even better, it is a set whose
elements correspond to the equivalence classes).
The surjective map π : X → X/ ≡ , π(x) = [x], is the canonical projection.
Theorem: In X/ ≡ , only equivalent elements are identified; [x] = [y] ⇔ x ≡ y .
Proof: If [x] = [y], then y ∈ [y] = [x], and x ≡ y.

Conversely, if x ≡ y, since ≡ is reflexive, it is enough to show that [y] ⊆ [x].
Now, if z ∈ [y], then y ≡ z; hence x ≡ z and z ∈ [x].
Corollary: Each element of X is in a unique equivalence class.
Proof: If x ∈ [y], then y ≡ x; hence [y] = [x].
Construction of Z: Let N = {0, 1, 2, 3, . . .} be the set of natural numbers. The set of integer
numbers Z is defined to be the quotient set of N × N by the equivalence relation
(m, n) ≡ (m0 , n0 ) when m + n0 = m0 + n,
and m − n denotes the class of (m, n). Any natural number n defines an integer number n − 0,
and so we may identify N with a subset of Z = {. . . , −2, −1, 0, 1, 2, . . .}.
The sum and the product of a = m − n, b = r − s, are defined to be
a + b = (m + r) − (n + s) , a · b = (mr + ns) − (nr + ms)
and both are well-defined. In fact, if a = m0 − n0 , then m + n0 = m0 + n, and
m + n0 + r + s = m0 + n + r + s
(m + n0 )r + (m0 + n)s = (m0 + n)r + (m + n0 )s
13
14 CHAPTER 1. ANALYSIS I
and we also have a + b = m0 + r − (n0 + s), ab = m0 r + n0 s − (n0 r + m0 s).

So we reduce statements on Z to statements on N. The theory of integer numbers is just a
part of the theory of natural numbers.
If the natural numbers are free from contradiction, so are the integer numbers.
Example: Let us fix a natural number n ≥ 2. The congruence relation, a ≡ b (mod. n) when
b − a ∈ nZ, is an equivalence relation on Z, and [a] = a + nZ = {a + cn : c ∈ Z}.
The quotient set Z/nZ = {[1], [2], . . . , [n] = [0]} has n elements.
Construction of Q: The set of rational numbers Q is defined to be the quotient set of the
direct product Z × (Z − {0}) by the equivalence relation
(a, s) ≡ (b, t) when at = bs,
and as denotes the class of (a, s). Each integer a defines a rational number a1 and we may identify
Z with a subset of Q. The sum and the product of q = as and r = bt are defined to be
at + bs ab
q+r = , q·r =
st st
a0
and both are well-defined. In fact, if q = s0 , then as0 = a0 s, so that
(at + bs)s0 t = a0 tst + bss0 t = (a0 t + bs0 )st,

abs0 t = a0 bst,
0 0 0
and we also have q + r = a t+bs ab
s0 t , qr = s0 t .
n
We say that a rational number is ≥ 0 when it may be represented as a quotient m of two
natural numbers. If q, r ∈ Q, we put q ≤ r when r − q ≥ 0.
So we reduce statements on Q to statements on Z. The theory of rational numbers is just a
part of the theory of integer numbers.
If the natural numbers are free from contradiction, so are the rational numbers.
1.2 Cardinal and Ordinal Numbers

Definitions: Let X, Y be two sets. A map f : X → Y assigns to any element x ∈ X a unique
element f (x) ∈ Y , and the composition with another map g : Y → Z is

g ◦ f : X → Z , (g ◦ f )(x) = g f (x) .
If A ⊆ X, we put f (A) = {f (a) : a ∈ A} ⊆ Y .

If B ⊆ Y , we put B ∩ X := f −1 (B) = {x ∈ X : f (x) ∈ B} ⊆ X.
A map f : X → Y is injective if f (x) = f (x0 ) ⇒ x = x0 , surjective when f (X) = Y , and
bijective if it is injective and surjective: any element y ∈ Y is the image of a unique element
f −1 (y) ∈ X, so that f −1 : Y → X also is a bijection and f −1 ◦ f = IdX , f ◦ f −1 = IdY .
Compositions of injective (surjective, bijective) maps are injective (surjective, bijective).
Definition: Let X, Y be two sets. If there exists a bijection f : X → Y , then we say that X
and Y have the same cardinal and we put |X| = |Y |. If there exists an injection f : X → Y ,
then we say that |X| ≤ |Y |, and we put |X| < |Y | if moreover |X| =
6 |Y |.
Sets of cardinal ≤ ℵ0 := |N| are said to be countable.
Schröder-Bernstein Theorem: If |X| ≤ |Y | and |Y | ≤ |X|, then |X| = |Y |.

1.2. CARDINAL AND ORDINAL NUMBERS 15
Proof: Let f : X → Y and g : Y → X be injective maps, so that h = gf : X0 := X → X2 := h(X)

is a bijection, and we have X2 =T(gf )(X) ⊆ X1 := g(Y ), where |Y | = |X1 |.
Put Xn+2 := h(Xn ), X∞ := n Xn , and Zn := Xn − Xn+1 .
The map h defines bijections Z2n → Z2n+2 , n ≥ 0, and the identity defines bijections
Z2n+1 → Z2n+1 , n ≥ 0. We see that |X| = |X1 | = |Y |, since we have a bijection
X = Z0 ∪ Z1 ∪ Z2 ∪ . . . ∪ X∞ → Z1 ∪ Z2 ∪ Z3 ∪ . . . ∪ X∞ = X1 = g(Y ).
Definitions: Given two cardinals a = |X|, b = |Y |, the sum a + b is the cardinal of the disjoint
union X q Y , the product ab is the cardinal of the cartesian product X × Y , and the power
ab is the cardinal of the set X Y of all maps1 Y → X.
Any subset Y of a set X may be viewed as a map IY : X → {0, 1}, where IY (Y ) = 1 and
IY (X − Y ) = 0; hence the cardinal of the set P(X) of all subsets of X is just 2|X| .
Cantor’s Theorem: |X| < 2|X| .
First proof : Obviously |X| ≤ |P(X)|.

If there exists a bijection f : X → P(X), we consider the subset Y := {x ∈ X : x ∈
/ f (x)}.
Since f is surjective, we have Y = f (y) for some y ∈ X.
If y ∈ Y , then y ∈
/ f (y) = Y . If y ∈
/ Y , then y ∈ f (y) = Y . Absurd.
Hence such bijection does not exist, and |X| = 6 |P(X)| = 2|X| .
Second proof : Let us give an alternative proof, using Cantor’s Diagonalverfahren. Given a map
φ : X → {0, 1}X , φ 7→ φx , we may consider the map ϕ : X → {0, 1} such that ϕ(x) = 1 when
φx (x) = 0, and ϕ(x) = 0 when φx (x) = 1, so that ϕ 6= φx , ∀x ∈ X, and we see that φ is not
6 |{0, 1}X | = 2|X| .
surjective. Hence |X| =
Definitions: A relation ≤ on a set X is an order relation if it is

1. Reflexive: x ≤ x, ∀x ∈ X.
2. Antisymmetric: x, y ∈ X, x ≤ y, y ≤ x ⇒ x = y.
3. Transitive: x, y, z ∈ X, x ≤ y, y ≤ z ⇒ x ≤ z.
and we put x < y when x ≤ y and x 6= y.
Any order on X clearly induces an order on any subset Y ⊆ X.
Let X, X 0 be two ordered sets. A map f : X → X 0 is a morhism of ordered sets when
x1 ≤ x2 ⇒ f (x1 ) ≤ f (x2 ).
It is an isomorphism of ordered sets when it is bijective and f −1 also is a morphism (it is

bijective and x1 ≤ x2 ⇔ f (x1 ) ≤ f (x2 )), and it is an anti-isomorphism when it is bijective
and x1 ≤ x2 ⇔ f (x1 ) ≥ f (x2 ).
Definitions: Let X be an order. We say that x ∈ X is the first element when x ≤ y, ∀y ∈ X

(if it exists, it is unique) and that it is the last element when y ≤ x, ∀y ∈ X. The first (resp.
last) element of a subset Y ⊆ X is the maximum (resp. minimum) of Y .
We say that x ∈ X is maximal when x ≤ y ⇒ x = y, and minimal when y ≤ x ⇒ y = x.
We say that x ∈ X is an upper (resp. lower) bound of a subset Y ⊆ X when y ≤ x (resp.
x ≤ y) for all y ∈ Y . The supremum of Y is the first element of the subset of all upper bounds
1
We assume that any set X admits a unique map ∅ → X (in particular there is a unique map ∅ → ∅, the
identity), while there is no map X → ∅ when X 6= ∅.
of Y (if it exists, it is unique) and the infimum of Y is the last element of the subset of all
lower bounds of Y (if it exists, it is unique). An order is a lattice if has first and last element,
and any pair x1 , x2 ∈ X has supremum and infimum.
An order X is total if any pair is comparable (x ≤ y or y ≤ x for any x, y ∈ X) and it is a
well-order when every non empty subset has a first element (hence it is a total order).
The chains in an order X are the totally ordered subsets of X.
In the founding principles of Mathematics we admit the so named Zorn’s lemma:
Zorn’s Lemma: Let X be a non empty order. If any chain of X has an upper bound in X,
then X has some maximal element.
Axiom of Choice: Any surjective map p : X → Y admits a section s : Y → X (a map such

that p ◦ s = IdY ).
Proof: Let S be the set of pairs (A, s) where A ⊆ Y and s : A → X is a map such that p◦s = IdA .
The set S is not empty (just take a point y ∈ Y , a point x ∈ p−1 (y), and put s(y) = x) and
we order S as follows:
(A, s) ≤ (B, t) when A ⊆ B and s = t|A .
S
Any chain {(Ai , si )}i∈I admits the upper bound ( i Ai , s), where s(a) = si (a) when a ∈ Ai .
Hence, by Zorn’s lemma, S has a maximal element (Z, s).
If Z 6= Y , and y ∈ Y − Z, we may extend s to Z ∪ {y} (just put s(y) = x, where x ∈ p−1 (y))
against the maximal character of (Z, s). Hence Z = Y and s is a section of p.
Corollary: If p : X → Y is a surjective map, then |Y | ≤ |X|.
Proof: Let s be a section of p. Since s is injective, we have |Y | = |s(Y )| ≤ |X|. q.e.d.
1. N × N is a well-order set with the lexicographical order 00 < 01 < . . . < 10 < 11 < . . .
2. Any set of cardinals is an ordered set.
3. The subsets of a set, subgroups of a group, ideals of a ring, vector subspaces of a vector space,
etc., ordered by inclusion, form a lattice.
4. Any ordered set (X, ≤) defines a dual order X ∗ = (X, ≤∗ ), where we put x ≤∗ y ⇔ y ≤ x.
It is clear that X ∗∗ = X, and that a map f : X → Y is an anti-isomorphism if and only if
f : X ∗ → Y is an isomorphism. Moreover, X is a lattice if and only if so is X ∗ .
5. Since f : Z → N, f (n) = 2n when n ≥ 0 and f (−n) = 2n − 1 when n ≥ 1, is bijective, we see

that |Z| = ℵ0 .
6. Since f : N × N → N, f (n, m) = 2n 3m , is injective, we see that ℵ20 = ℵ0 ; hence ℵn0 = ℵ0 ,

n ≥ 1.
7. Since Q is a quotient of Z × (N − 0), we see that |Q| = ℵ0 ; hence |Qn | = ℵ0 , n ≥ 1.
8. The set of infinite decimal expressions 0.c1 c2 c3 . . . has cardinal 10ℵ0 : it is uncountable.
9. Cantor’s theorem shows that “the set X of all sets” is non sense: since P(X) ⊆ X, we would
have 2|X| = |P(X)| ≤ |X|. Analogously there is no “set of all groups”, etc.
1.2. CARDINAL AND ORDINAL NUMBERS 17
Ordinal Numbers
Definition: Let X be a well-order. A set Y ⊆ X is an initial ray of X when y < x for all
y ∈ Y , x ∈ X − Y . When Y is incomplete, Y 6= X, then Y = (← x) := {y ∈ X : y < x}, where
x is the supremum of the ray, so that X is isomorphic to the set of incomplete initial rays with
the inclusion order.
Theorem: Any set X admits a well-order.
Proof: Consider the set W of all pairs (Y, ≤) where Y ⊆ X and ≤ is a well-order on Y .
The set W is not empty (finite subsets admit a well-order) and we order W as follows:
(Y 0 , ≤0 ) ≤ (Y, ≤) when (Y 0 , ≤0 ) is an initial ray of (Y, ≤).
In W any chain {(Yi , ≤i )}i∈I admits an upper bound:

S
In fact Y = i Yi has an obvious order ≤ inducing ≤i on Yi , and Yi is an initial ray of Y :
if y ∈ Y, y ∈
/ Yi , then y is in some Yj containing Yi so that y bounds above Yi . Finally, Y is a
well-ordered set: if ∅ 6= Z ⊆ Y , we have ∅ 6= Yi ∩ Z for some index i, and the first element of
Yi ∩ Z is the first element of Z, because Yi is an initial ray of Y .
According to Zorn’s lemma, W has a maximal element (Y, ≤).
If Y 6= X, and x ∈ X − Y , we may extend the well-order of Y to Y ∪ {x} so that y ≤ x,
∀y ∈ Y , against the maximal character of (Y, ≤). Hence Y = X and ≤ is a well-order on X.
Zermelo’s Theorem: Let X be a well ordered set. If f : X → X is a strictly increasing map

(x < y ⇒ f (x) < f (y)) then x ≤ f (x), ∀x ∈ X.
Proof: Otherwise, let x be the first element such that f (x) < x, so that f (x) ≤ f (f (x)).
Since f is increasing, we have f (f (x)) < f (x). Absurd. q.e.d.
1. Any well-order X has a unique automorphism: the identity.

If τ : X → X is an automorphism and x < τ (x), then τ −1 (x) < x. Impossible.
2. If two well-orders are isomorphic, then the isomorphism is unique. (So, we may identify
isomorphic well-orders and name them ordinal numbers. The ordinal of N is ω.)
If f, g : X → X 0 are isomorphisms, then g −1 f is an automorphism; hence g −1 f = IdX .
3. Any two initial rays R1 ⊂ R2 of a well-order never are isomorphic.

If f : R2 → R1 ⊂ R2 is an isomorphism and x ∈ R2 − R1 , then f (x) < x. Impossible.
Definition: Let α, β be two ordinal numbers. We put β < α when β is isomorphic to an

incomplete initial ray of α. By the above results, ≤ defines an order on any set of ordinal
numbers, and the set (← α) of ordinal numbers β < α is a well-order, since it is just α.
Lemma: Let X, Y be two well-orders. If any incomplete initial ray of X is isomorphic to an

incomplete initial ray of Y , then X ≤ Y .
Proof: Any
incomplete initial ray (← x) of X is isomorphic to a unique incomplete initial ray
← f (x) of Y . If x < x0 , by the above results it is clear that f (x) < f (x0 ), and that f is an
isomorphism of X onto an initial ray of Y (eventually complete).
Lemma: If α and β are two ordinal numbers, then α ≤ β or β ≤ α.

Proof: Otherwise, by the lemma we may consider the first incomplete initial ray (← x) of α
(resp. (←, y) of β) not isomorphic to an incomplete initial ray of β (resp. of α).
Now (← x) and (←, y) satisfy the hypothesis of the above lemma, so that (← x) ≤ (← y)
and (← y) ≤ (← x). Hence (← x) and (← y) are isomorphic. Absurd.
Theorem: If S is a set of ordinal numbers, then (S, ≤) is a well-order.
Proof: If R ⊆ S is not empty, we consider an ordinal α ∈ R.

The initial rays of α give all the ordinals β < α, so that {β ∈ R : β ≤ α} has a first element
β0 and, by the lemma, β0 ≤ α < γ for any other γ ∈ R.
Cardinal Numbers: Let ω be the ordered set of all finite ordinal numbers, so that |ω| = ℵ0 .
Let ω1 be the ordered set of all ordinal numbers of cardinal ≤ ℵ0 , so that ω1 is the first
ordinal number of cardinal > ℵ0 . We put ℵ1 := |ω1 |, so that ℵ0 < ℵ1 and there is no intermediate
cardinal: if |σ| < ℵ1 , then σ < ω1 , so that |σ| ≤ ℵ0 .
Let ω2 be the ordered set of all ordinal numbers of cardinal ≤ ℵ1 , so that ω2 is the first
ordinal number of cardinal > ℵ1 , and we put ℵ2 := |ω2 |. So we obtain an increasing S sequence
ℵ0 < ℵ1 < ℵ2 < . . . of cardinals with no intermediate cardinal. Now we put ℵω := | n ωn |, and
so on:
ℵ0 < ℵ1 < ℵ2 < . . . < ℵω < ℵω+1 < . . .
In general, if α = β + 1, we consider the ordered set ωαSof all ordinal numbers of cardinal
≤ ℵβ , and we put ℵα := |ωα |. Otherwise, we consider ωα = β<α ωβ , and we put ℵα := |ωα |.
Now, given an infinite cardinal |σ|, if we consider the first ordinal α such that |σ| ≤ ℵα , then
|σ| = ℵα . Otherwise σ < ωα , so that |σ| ≤ ℵβ for some β < α. Absurd.
So we see how the ordinal numbers correspond to the infinite cardinal numbers.
1.3 Real and Complex Numbers

Now we consider the ring (commutative with unity) of all sequences (qn ) = (q0 , q1 , q2 , . . .) of
rational numbers, with the termwise addition and product.
Definition: A sequence of rational numbers (qn ) is a Cauchy sequence when for every positive
rational ε ∈ Q+ there is an index nε such that |qn − qm | < ε, ∀m, n ≥ nε ; and it is a null
sequence when for every ε ∈ Q+ there is an index nε such that |qn | < ε, ∀n ≥ nε .
Constant sequences (q) = (q, q, q, . . .) are Cauchy sequences.
Any null sequence is a Cauchy sequence, and any Cauchy sequence is bounded: there exists
c ∈ N such that |qn | ≤ c for any index n (just take c ≥ ε + max{|q0 |, |q1 |, . . . , |qnε |}.).
Any subsequence (qin ), i0 < i1 < i2 < . . ., of a Cauchy (resp. null) sequence (qn ) also is a
Cauchy (resp. null) sequence, and the difference (qn − qin ) is a null sequence.
Theorem: Cauchy sequences of rational numbers form a subring, and the null sequences define
an ideal of this subring.
Proof: First we prove that the Cauchy sequences form a subring and, since constant sequences
are Cauchy sequences, we only have to prove that the sum (qn ) + (qn0 ) = (qn + qn0 ) and product
(qn ) · (qn0 ) = (qn qn0 ) of two Cauchy sequences (qn ), (qn0 ) also are Cauchy sequences.
The sum is clear, since |(qn + qn0 ) − (qm + qm 0 )| ≤ |q − q | + |q 0 − q 0 |.
n m n m
1.3. REAL AND COMPLEX NUMBERS 19
For the product, consider a constant c ∈ N bounding (qn ) and (qn0 ). Given ε ∈ Q+ , there is
an index nε such that |qn − qm |, |qn0 − qm
0 | < ε, ∀m, n ≥ n , so that
ε
|qn qn0 − qm qm
0
| = |qn (qn0 − qm
0 0
) + (qn − qm )qm | ≤ |qn | · |qn0 − qm
0 0
| + |qn − qm | · |qm | < 2cε.
Finally the null sequences clearly form an additive subgroup and, when (qn0 ) is a null sequence,
we consider a constant c bounding (qn ) and an index nε such that |qn0 | < εc , ∀n ≥ nε , so that
|qn qn0 | ≤ c|qn0 | < ε, and we see that (qn qn0 ) also is a null sequence.
Definition: The ring of real numbers R is the quotient of the ring of Cauchy sequences of
rational numbers by the ideal of null sequences.
A real number is ≥ 0 when it may be represented by some Cauchy sequence (qn ) with all
the terms qn ≥ 0. If a, b ∈ R, we put a ≤ b when b − a ≥ 0.
We have a canonical injective ring morphism Q → R, q 7→ [(q)], (preserving inequalities).
A Cauchy sequence (qn ) and any subsequence (qin ) represent the same real number.
Lemma: Given a real number x 6= 0, there exists k ∈ N such that x may be represented by a
Cauchy sequence (qn ) with all the terms qn ≥ k1 , or with all the terms qn ≤ − k1 .
Proof: By definition, any non-null Cauchy sequence admits a subsequence (qn ) with all the terms
|qn | ≥ k1 for some constant k ∈ N.
Now (qn ) admits a subsequence of positive terms or a subsequence of negative terms.
Theorem: The ring R is a field and ≤ is total order compatible with the additive and multi-
plicative structure:
1. If a ≤ b, then a + c ≤ b + c.
2. If 0 ≤ a and 0 ≤ b, then 0 ≤ ab.
Proof: Clearly ≤ is reflexive and transitive. It is antisymmetric: If a ≤ b and b ≤ a, then b − a

and a − b are represented by Cauchy sequences (qn ) and (qn0 ) with all the terms qn , qn0 ≥ 0. Now
(qn + qn0 ) is a null sequence and so is (qn ), since |qn | ≤ |qn + qn0 |, so that b − a = 0.
It is a total order: If a 6= b, by the above lemma b − a > 0 or a − b > 0; hence a < b or b < a.
Finally we prove that R is a field: By the lemma, any non-zero real number is represented
by a Cauchy sequence (qn ) with all the terms |qn | ≥ k1 for some constant k ∈ N. If (qn−1 ) is a
Cauchy sequence, then 1 = [(qn )] · [(qn−1 )], and we conclude. Now,

1 1 qm − qn |qm − qn |
−
qn qm = qn qm ≤ ·

k2
Definition: The absolute value of x ∈ R is defined to be |x| = max{x, −x}.

A sequence of real numbers (xn ) is a Cauchy sequence when for every positive real number
ε ∈ R+ there is an index nε such that |xn − xm | < ε, ∀m, n ≥ nε ; and it converges to a limit
x ∈ R (and we put x = lim xn ) when there is an index nε such that |xn − x| < ε, ∀n ≥ nε .
Any convergent sequence is a Cauchy sequence.
Theorem: For all x ∈ R and ε ∈ Q+ there exists q ∈ Q with |q − x| < ε.
Proof: If (qn ) represents x, then there exists n0 ∈ N such that |qn − qn0 | < 2ε , ∀n ≥ n0 .
Hence |x − qn0 | ≤ 2ε < ε.
Theorem: Any Cauchy sequence (xn ) of real numbers converges to a unique real limit.
Proof: Let us see the existence of the limit:

Pick qn ∈ Q such that |qn − xn | < n1 . Then (qn ) is a Cauchy sequence of rational numbers:
Given ε ∈ Q+ , there exists an index n0 such that n0 > 3ε and |xn − xm | < 3ε , ∀n ≥ n0 , so
that we have |qn − qm | ≤ |qn − xn | + |xn − xm | + |xm − qm | < ε.
In particular, if we put x = [(qn )], we have |qn − x| ≤ ε, ∀n ≥ n0 .
Now, given ε > 0, there exists an index nε such that nε > 2ε and |qn − x| < 2ε , ∀n ≥ nε , so
that we have |xn − x| ≤ |xn − qn | + |qn − x| < n1 + 2ε ≤ ε, ∀n ≥ nε . Hence x = lim xn .
If y ∈ R is a different limit, there is an index n0 such that |xn −x| < |y−x|
2 and |xn −y| < |y−x|
2 ,
|y−x| |y−x|
∀n ≥ n0 , and we get a contradiction: |y − x| ≤ |y − xn | + |xn − x| < 2 + 2 = |y − x|.
Corollary: Any bounded above set of real numbers has a supremum, and any bounded below set
of real numbers has an infimum.
Proof: If X ⊂ R is bounded above, for each natural number n we consider the first fraction
qn = 2an , a ∈ Z, which is an upper bound of X. Now (qn ) is a Cauchy sequence, because
|qn − qm | < 2−n , and x := lim qn is the supremum of X.
If X ⊂ R is bounded below, then −X is bounded above.
If x is the supremum of −X, then −x is the infimum of X.
Corollary: Any bounded above non-decreasing sequence (xn ) converges, hence so does any
bounded below non-increasing sequence.
Proof: Put x = sup{xn }. Given ε > 0, there exists and index nε such that x − ε < xnε ≤ x;
hence x − ε < xn ≤ x for all n ≥ nε , since the sequence is non-decreasing.
Corollary: Any positive real number b has a unique positive square root.
2
Proof: If a is positive and a2 < b, then a < 21 a + ab and 41 a + ab = 14 a2 + ab2 b + 2b < b.

If we put a0 = a and an+1 = 12 an + abn , we obtain (it was a summerian method of extracting

square roots) a bounded above increasing sequence with limit l = 12 l + bl . Hence l2 = b.

The positive square root is unique because x < y ⇒ x2 < y 2 .
Complex Numbers: The set C of complex numbers is the set of pairs x + yi of real numbers,
endowed with the operations (i2 = −1)
(x1 + y1 i) + (x2 + y2 i) = (x1 + x2 ) + (y1 + y2 )i ,

(x1 + y1 i) · (x2 + y2 i) = (x1 x2 − y1 y2 ) + (x1 y2 + x2 y1 )i .
The conjugate of z √= x + yi ispthe complex number z̄ = x − yi, and the modulus of z is

the real number |z| = + z · z̄ = + x2 + y 2 ≥ 0.
z + u = z̄ + ū , zu = z̄ ū , z̄¯ = z , |z| = |z̄| ,

|z| = 0 ⇔ z = 0 , |zu| = |z| · |u| , z + z̄ ≤ 2|z| , |z + u| ≤ |z| + |u| .
These properties follow directly from the definitions, except the last one. Now,
|z + u|2 = (z + u)(z + u) = (z + u)(z̄ + ū) = |z|2 + |u|2 + z ū + z̄u

= |z|2 + |u|2 + z ū + z ū ≤ |z|2 + |u|2 + 2|z ū|
= |z|2 + |u|2 + 2|z| · |ū| = |z|2 + |u|2 + 2|z| · |u| = (|z| + |u|)2 .
1.4. METRIC SPACES AND TOPOLOGICAL SPACES 21
Theorem: |R| = |C| = 2ℵ0 .
Proof: The set S of all sequences of rational numbers has cardinal ℵℵ0 0 = 2ℵ0 .
Since R is a quotient of a subset of S, we see that |R| ≤ 2ℵ0 .
On the other hand, any infinite decimal 0.c0 c1 c2 . . ., where 0 ≤ ci ≤ 9, defines a Cauchy
sequence of rational numbers (0.c0 , 0.c0 c1 , 0.c0 c1 c2 , . . .). When we only consider digits ci ∈ {0, 1},
we obtain an injection 2N ,→ R; hence 2ℵ0 ≤ |R| and we conclude that |R| = 2ℵ0 .
Finally, |C| = |R × R| = |R|2 = 22ℵ0 = 2ℵ0 .
1.4 Metric Spaces and Topological Spaces

Definition: A metric d on a set X is a map d : X × X → R≥0 such that
1. d(x, y) = d(y, x).
2. d(x, z) ≤ d(x, y) + d(x, z). (Triangle inequality).
3. d(x, y) = 0 if and only if x = y.
and a metric space is a set X endowed with a metric d.
Given two metric spaces (X, d) and (X 0 , d0 ), a map f : X → X 0 is a metric morphism
when d(x, y) = d f (x), f (y) , ∀x, y ∈ X; and it is an isometry, or metric isomorphism, if
moreover it is bijective (so that f −1 : X 0 → X also is an isometry).
Compositions of metric morphisms also are metric morphisms.
Examples: The usual metric on R is d(x, y) = |y − x|.

The usual metric on C is d(x, y) = |y − x|.
Any Euclidean vector space E admits the metric d(p, q) = kq − pk (see p. 55).
In particular, the usual scalar product in Rn induces the metric
p
d (x1 , . . . , xn ), (y1 , . . . , yn ) = (y1 − x1 )2 + . . . + (yn − xn )2
and the existence of orthonormal bases shows that any Euclidean vector space of dimension n
is metrically isomorphic to Rn .
A metric on the set of all bounded real-valued functions on a set X is
d(f, h) = sup {|h(x) − f (x)|}.
x∈X
A metric d on a set X obviously induces a metric dY (y1 , y2 ) := d(y1 , y2 ) on any subset

Y ⊆ X, and we say that (Y, dY ) is a subspace of the metric space (X, d).
Definition: In a metric space X, the ball with center at a point x ∈ X and radius r ∈ R+ is
Br (x) = B(x, r) = Bd (x, r) := {y ∈ X : d(x, y) < r}
and U ⊆ X is said to be an open set in X when for any point x ∈ U there exists a ball
Bε (x) ⊆ U (so that the empty set ∅ and X are open sets).
A subset Y ⊆ X is a closed set in X when the complement X − Y is an open set.
Theorem: Arbitrary unions and finite intersections of open sets are open sets.
Proof: Arbitrary unions of open sets are obviously open, and so are finite intersections since
B(x, ε1 ) ∩ . . . ∩ B(x, εn ) = B(x, ε) , ε = min{ε1 , . . . , εn }.
Definition: A topology on a set X is a family of subsets, named open sets, such that
1. The subsets ∅ and X are open sets.
2. Arbitrary unions of open sets are open sets.
3. Finite intersections of open sets are open sets.
and we say that Y ⊆ X is a closed set when X − Y is open, so that arbitrary intersections and
finite unions of closed sets are closed sets, and X and ∅ are closed sets.
A topological space is a set X endowed with a topology, and it is metrizable if the
topology is defined by a (non necessarily unique) metric on X.
o
Definitions: Given a topology on X, the interior of Y ⊆ X is the biggest open set Y ⊆ Y (the
union of all the open sets contained in Y ), and Y is a neighborhood of a subset Z, when Z is
contained in the interior of Y .
The closure of Y ⊆ X is the smallest closed set Ȳ containing Y (the intersection of all the
closed sets containing Y ), so that x ∈ Ȳ just when any neighborhood of x intersects Y .
The boundary of Y is the set ∂Y of points in the closure Ȳ not in the interior of Y .
When Ȳ = X, we say that Y is dense in X.
A sequence of points (xn ) in X converges to a limit x ∈ X (and we put x = lim xn or
xn → x) when for any neighborhood U of x there is an index nU such that xn ∈ U , ∀n ≥ nU
(in a metric space, this condition means that lim d(x, xn ) = 0). We say that a sequence (xn ) in
X has an adherent point x ∈ X when for any neighborhood U of x and any index n we have
xm ∈ U for some index m ≥ n. When all the points xn are in a subset Y , then clearly x ∈ Ȳ .
1. In a metric space X, any ball Br (x) is open and any open set is a (may be infinite) union of
balls, while {y ∈ X : d(x, y) = r} and {y ∈ X : d(x, y) ≤ r} are closed subspaces. Moreover,
a point x ∈ X is in the closure of a subspace Y if and only if x = lim yn for some sequence
yn ∈ Y . In fact, if x ∈ Ȳ , there are points yn ∈ B x, n1 ∩ Y , so that yn → x.
2. Let X be a topological space and Y ⊆ X. The subsets U ∩ Y , where U is an open set in

X, clearly define a topology on Y and, when endowed with this topology, we say that Y is a
subspace of X. When X is a metric space, this topology is defined by the induced metric
dY , because BdY (y, r) = BdX (y, r) ∩ Y .
3. In R = (−∞, ∞), the intervals (a, b) := {x ∈ R : a < x < b}, (−∞, b) and (a, ∞) are open,
while the intervals [a, b] := {x ∈ R : a ≤ x ≤ b}, (−∞, b] and [a, ∞) are closed. When a < b,
in fact (a, b) is the interior of [a, b] and [a, b] is the closure of (a, b).
4. In Rd , any product U1 × . . . × Ud of open sets Ui ⊆ R is an open set, and any product

Y1 × . . . × Yd of closed sets Yi ⊆ R is a closed set.
5. The rational numbers are dense in R (see page 19).
6. If all the subsets of a set X are defined to be open sets, we obtain the discrete topology on
X. If only ∅ and X are defined to be open sets, we obtain the trivial topology on X.
Definition: If a topological space X has a unique open closed set, then X = ∅. If X only has
two open closed sets (namely ∅ and X 6= ∅) we say that X is connected. If X admits an open
closed set U different from ∅ and X (so that X = U ∪ U c is a disjoint union of two non-empty
open sets) we say that X is disconnected. A subset Y of X is connected (or disconnected)
when so it is with the induced topology.
A subset I ⊆ R is an interval (eventually with infinite ends) if [a, b] ⊆ I whenever a, b ∈ I.
Theorem: A subset X ⊆ R is connected if and only if it is a non-empty interval.
Proof: If X is connected, put a = inf X, b = sup X, so that X ⊆ [a, b].

If x ∈ (a, b) is not in X, then X ∩ [x, ∞) = X ∩ (x, ∞) is an open closed set in X.
Hence it is empty (and we have b ≤ x), or it is X (and we have x ≤ a). Absurd.
We conclude that (a, b) ⊆ X ⊆ [a, b], so that X is an interval.
Now let us see that any bounded closed interval [a, b] is connected. If [a, b] = U ∪ U c is a
disjoint union of two non-empty open closed sets, we may assume that a ∈ U . Let c = inf U c ,
so that [a, c) ⊆ U .Then c ∈ U c and c ∈ U , because U c and U are closed. Absurd.
Finally, any non-empty interval is an increasing union of bounded closed intervals, so that
it is easy to check that it also is connected.
S
Definition: An open cover of XSis a family of open sets {Ui }i∈I such that X = i Ui , and X is
compact if any open cover X = i Ui has a finite subcover: T X = Ui1 ∪ . . . ∪ Uin . Equivalently, if
a family of closed sets {Yi }i∈I has empty intersection, ∅ = i Yi , then so does some finite family
∅ = Yi1 ∩ . . . ∩ Yin . A subset Y of a topological space space X is said to be compact S when so
it is with the induced topology: If {Ui }i∈I is a family of open sets in X and Y ⊆ i Ui , then
Y ⊆ Ui1 ∪ . . . ∪ Uin for some finite set of indices.
Proposition: Any sequence (xn ) in a compact space K has an adherent point x ∈ K.

T
Proof: Let Yn be the closure of {xn , xn+1 , . . .}, so that Yn+1 ⊆ Yn . If n Yn = ∅, then so does a
finite family ∅ = Yn1 ∩ . . . ∩ Ynr = Ymax n1 ,...,nr . Absurd.
T T
Hence n Yn 6= ∅, and any point x ∈ n Yn is adherent to (xn ).
Lemma: Any closed subspace Y of a compact space K also is compact.

S S
Proof: If Y ⊆ i Ui is an open cover, then K = (K − Y ) ∪ ( i Ui ) is an open cover of K, so that
it admits a finite subcover K = (K − Y ) ∪ Ui1 ∪ . . . ∪ Uin .
We see that Y ⊆ Ui1 ∪ . . . ∪ Uin . Hence Y is compact.
Lemma: Any compact subspace K of a metric space X is closed and bounded (i.e. contained
in a ball).
Proof: If x ∈ X − K, for each point y ∈ K we may find disjoint balls x ∈ By and y ∈ By0 .
Then K ⊆ y By0 , so that K admits a finite subcover, K ⊆ By0 1 ∪ . . . ∪ By0 n .
S
Hence x ∈ By1 ∩ . . . ∩ Byn and (By1 ∩ . . . ∩ Byn ) ∩ K = ∅, so that X − K is open.

S
Finally, fix a point x ∈ X. The open cover K ⊆ r∈N Br (x) admits a finite subcover
K ⊆ Br1 (x) ∪ . . . ∪ Brn (x) = Br (x), so that K is bounded.
Heine-Borel Theorem: Compact subsets of Rd are just bounded closed subsets.
Proof: If K ⊂ Rd is compact, it is a closed bounded set by the lemma.

Conversely, any closed bounded set K ⊂ Rd is contained in a cube
I0 = [a1 , a1 + l] × . . . × [ad , ad + l]
and, by the above lemma, we only have to show that I0 is compact.

S
Assume that I0 is not compact, and let I0 ⊆ i Ui be an open cover with no finite subcover.
Bisecting the edges of I0 , we obtain 2d cubes, and at least one of such cubes, say I1 , does not
admit a finite subcover. So we obtain a decreasing sequence of cubes
I0 ⊃ I1 ⊃ I2 ⊃ . . . ⊃ In ⊃ . . . ,
the size of In being l/2n . Let us consider a sequence (yn ) such that yn ∈ In .
The coordinates xi (yn ) of these points define d Cauchy sequences of real numbers, since
|xi (yn ) − xi (ym )| ≤ 2−n0 l when m, n ≥ n0 . Hence (yn ) converges to a point y ∈ Rd , and y ∈ In
since all the terms of the sequence, up to a finite number, are in In .
Since y ∈ I0 , we have y ∈ Ui for some index i,
√ and nthere is a ball Br (y) ⊆ Ui .
Since y ∈ In , when n is big enough (namely dl/2 < r) we have In ⊆ Br (y) ⊆ Ui .
Absurd, In does not admit any finite subcover.
S
Alternative Proof in the crucial case I = [a, b]: Let I ⊆ i Ui be an open cover. Since I is
connected, it is enough to show that the set C of all points x ∈ I such that [a, x] admits a finite
subcover is a closed open set in I.
Now, given a point x ∈ I, we have x ∈ Br (x) ⊆ Ui for some index i and some r > 0. It
is clear that Br (x) ⊆ C whenever x ∈ C (hence C is open) and that Br (x) ⊆ I − C whenever
x ∈ I − C (hence C is closed).
1.4.1 Continuous Functions

Definition: A map f : X → Y between topological spaces is continuous at a point x ∈
X when f −1 (V ) is a neighborhood of x for any neighborhood V of f (x). We say that f is
continuous when it is continuous at any point of X: when f −1 (V ) is an open set in X for any
open set V in Y ; i.e. f −1 (Z) is a closed set in X for any closed set Z in Y . We say that a
continuous bijective map f is an homeomorphism if f −1 : Y → X also is continuous.
1. Compositions of continuous maps also are continuous.
2. Any continuous map f preserves limits: if x = lim xn , then f (x) = lim f (xn ).
Conversely, when X and Y are metric spaces, any map f : X → Y preserving limits is
continuous: if Z ⊆ Y is closed and a sequence xn ∈ f −1 (Z) converges, then lim xn ∈ f −1 (Z)
because f (lim xn ) = lim f (xn ) ∈ Z.
3. The functions R × R → R, (x, y) 7→ x + y, and R × R → R, (x, y) 7→ xy, are continuous;

hence, sums and products of real-valued continuous functions are continuous.
The function R − {0} → R − {0}, x 7→ 1/x, is continuous; hence the inverse of a continuous
function (without zeroes) also is continuous.
Theorem: Let f : X → Y be a continuous map. If X is compact, then f (X) is compact. If X
is connected, then f (X) is connected.
Proof: Let us consider an open cover f (X) ⊆ i Vi , so that X = i f −1 (Vi ).

S S
Now, if X is compact, it admits a finite subcover, X = f −1 (Vi1 ) ∪ . . . ∪ f −1 (Vin ).
Hence f (X) ⊆ Vi1 ∪ . . . ∪ Vin , and f (X) is compact.
Let V be an open closed set in f (X), so that f −1 (V ) is an open closed set in X.
If X is connected, then f −1 (V ) = ∅ or f −1 (V ) = X. Hence V = ∅ or V = f (X).
Corollary: Any continuous function f : K → R on a compact topological space K 6= ∅ attains

a maximum and a minimum (there exist p, q ∈ K such that f (p) ≤ f (x) ≤ f (q) for all x ∈ K).
Proof: Since f (K) ⊂ R is bounded, it admits an infimum a and a supremum b.

Since f (K) is closed, we have that a, b ∈ f (K), and we conclude.
Corollary: If a continuous function f : X → R on a connected topological space X attains two

values a < b, then it also attains any value a < c < b (i.e. c = f (x) for some x ∈ X.)
Intermediate Value Theorem: Any continuous function f : [a, b] → R attains any value
between f (a) and f (b).
Corollary: If ρ is a positive real number and n ≥ 1 is a natural number, there exists a unique
√ √
positive real number n ρ such that ( n ρ)n = ρ.
Proof: Consider the continuous function f : [0, b] → R, f (x) = xn , where bn > ρ.

The existence of a positive n-th root follows from the above corollary, since f (0) < ρ < f (b),
and it is unique since x < y ⇒ xn < y n .
Bolzano’s Theorem: Let f : [a, b] → R be a continuous function. If f (a) and f (b) have
opposite sign, then we have f (x) = 0 at some point a < x < b.
1.4.2 Completion of a Metric Space

Definition: A map f : X → Y between metric spaces is uniformly
continuous when for
every ε ∈ R+ there exists δ ∈ R+ such that f Bδ (x) ⊆ Bε f (x) , ∀x ∈ X; i.e. such that
d(x, y) < δ ⇒ d(f (x), f (y)) < ε.
Compositions of uniformly continuous maps also are uniformly continuous.
Proposition: If Y is a subset of a metric space X, then the function

f : X → R, f (x) = d(x, Y ) = inf{d(x, y); y ∈ Y }
is uniformly continuous. Moreover, f (Ȳ ) = 0 and f (x) > 0 whenever x ∈
/ Ȳ .
Proof: Since d(x0 , y) ≤ d(x0 , x) + d(x, y), ∀y ∈ Y , we have |f (x0 ) − f (x)| ≤ d(x0 , x).
Moreover, f (x) = 0 if and only if any ball Bε (x) intersects Y , i.e. x ∈ Ȳ .
Theorem: Any continuous map f from a compact metric space K to a metric space Y is
uniformly continuous.
Proof: Since
f is continuous, given ε > 0, at any point x ∈ K we have a ball B(x, δx ) such that
f B(x, δx ) ⊆ B f (x), ε , and the balls B(x, 12 δx ) define an open cover of K.

Since K is compact, it admits a finite subcover, K = B(x1 , 21 δx1 ) ∪ . . . ∪ B(xn , 21 δxn ).

Put δ := min{ 12 δx1 , . . . , 12 δxn } > 0.
Now, if d(x, y) < δ, there is an index i such that d(xi , x) < 21 δxi , so that
d(xi , y) ≤ d(xi , x) + d(x, y) < 21 δxi + δ ≤ δxi .

Hence d f (xi ), f (y) < ε, and we conclude:

d f (x), f (y) ≤ d f (x), f (xi ) + d f (xi ), f (y) < ε + ε = 2ε.
Definitions: A sequence of points (xn ) of a metric space X is a Cauchy sequence when for
every ε ∈ R+ there is an index nε such that d(xn , xm ) < ε, ∀m, n ≥ nε .
Any convergent sequence is a Cauchy sequence, and a metric space X is complete when
every Cauchy sequence in X converges to a limit x ∈ X (clearly unique).
Uniformly continuous maps preserve Cauchy sequences.
1. Any Euclidean vector space is complete, hence C is complete. In fact, a sequence of points
converges (resp. is a Cauchy sequence) if and only if the coordinates converge (resp. define
Cauchy sequences) in R.
P
2. An infinite series of complex numbers n an is always understood as the sequence of partial
sums sn = a0 + . . . + an . Hence, it converges if for any ε > 0 there exists and index nε such
that |an + . . . + am | < ε whenever m ≥ n ≥ nε .
P
3. If a series n an converges, then lim an = 0.
We have lim an = lim(sn − sn−1 ) = lim sn − lim sn−1 = 0.
P P
4. If |an | ≤ bn when n 0, and n bn converges, then so does n an .
When n 0, we have |an + . . . + am | ≤ bn + . . . + bm = |bn + . . . + bm |.
5. If lim |a|an+1 | P P
n|
< 1, then n |an | (hence n an ) converges.
We may assume that |a|an+1 | n , and n 1

P
n | ≤ r < 1. Hence |a n | ≤ |a 0 |r n |a0 |r = |a0 | 1−r .
Proposition: Any closed subspace Y of a complete metric space X is complete.
Proof: Any Cauchy sequence (yn ) in Y converges to a point x ∈ X, because X is complete, and
x ∈ Y when Y is closed.
Proposition: Any complete subspace Y of a metric space X is a closed subspace.
Proof: If x ∈ Ȳ , then we have points yn ∈ B x, n1 ∩ Y , so that yn → x.

Now (yn ) is a Cauchy sequence because d(yn , ym ) ≤ d(yn , x) + d(x, ym ) < n1 + m

1
.
Hence (yn ) converges to a point y ∈ Y , and y = x because the limit point is unique.
Proposition: Any compact metric space K is complete.
Proof: Let (xn ) be a Cauchy sequence in K.

Given ε > 0, there is an index nε such that d(xn , xm ) < ε, ∀n, m ≥ nε .
Now, let x ∈ K be an adherent point to (xn ). We have d(x, xn ) < ε for some n ≥ nε , so that
d(x, xm ) ≤ d(x, xn ) + d(xn , xm ) < 2ε, ∀m ≥ n. We conclude that xn → x.
Definition: A topological space is Baire when the intersection of any countable family of dense
open sets is dense (the union of any countable family of closed sets with empty interior
S has dense
complement). In particular, if a non-empty Baire space is a countable union X = n Yn of closed
sets, then some Yn has non-empty interior.
Baire Theorem: Any complete metric space X is Baire.
Proof: Let {Un } be a countable

T family of dense open sets in X, and let us see that any non-empty
open set V ⊆ X intersects n Un .
Since U1 is dense, there is a closed ball B̄(x1 , r1 ) ⊆ U1 ∩ V .
Recursively, there are closed balls B̄(xn , rn ) ⊆ B(xn−1 , rn−1 ) ∩ Un , 0 < rn < n1 .
Since xm ∈ B(xn , rn ) when m ≥ n, we have that (xn ) is a Cauchy sequence.
Hence xn → x ∈ X. Since B̄(xn , rn ) is closed, we conclude that x ∈ B̄(xn , rn ) ⊆ Un ∩ V .
Definition: In a metric space X, two Cauchy sequences (xn ), (x0n ) are said to be equivalent if
lim d(xn , x0n ) = 0, and it is an equivalence relation on the set of all Cauchy sequences.
The quotient set will be denoted X

b and it is the completion of the metric space X.
Theorem: The completion X

b is a complete metric space with the metric
d(x̂, ŷ) := lim d(xn , yn ),
where (xn ) and (yn ) are Cauchy sequences representing x̂, ŷ ∈ X

b respectively.
We have a canonical metric morphism i : X → X, i(x) = [(x, x, . . .)], with dense image, and
b
it is an isometry when X is complete.
Proof: First we see that d(xn , yn ) is a Cauchy sequence in R (hence it converges):
d(xn , yn ) − d(xm , ym ) ≤ d(xn , xm ) + d(xm , ym ) + d(ym , yn ) − d(xm , ym ) = d(xn , xm ) + d(ym , yn )
hence |d(xn , yn ) − d(xm , ym )| ≤ d(xn , xm ) + d(ym , yn ), and we conclude.

1. d is well-defined :
If (x0n ) also represents x̂, then d(x0n , yn ) ≤ d(x0n , xn ) + d(xn , yn ); hence lim d(x0n , yn ) ≤
lim d(xn , yn ), since lim d(x0n , xn ) = 0. By symmetry lim d(xn , yn ) ≤ lim d(x0n , yn ) and we
conclude.
2. d is a metric on X:
b
Clearly d(x̂, ŷ) = d(ŷ, x̂) and, if d(x̂, ŷ) = 0, then (xn ) and (yn ) are equivalent, and x̂ = ŷ. If
a Cauchy sequence (zn ) represents a point ẑ ∈ X, b then d(xn , zn ) ≤ d(xn , yn ) + d(yn , zn ), and
taking limits we see that d(x̂, ẑ) ≤ d(x̂, ŷ) + d(ŷ, ẑ).
3. The metric morphism i has dense image:

Given x̂ ∈ X
b and ε > 0, let (xn ) be a Cauchy sequence representing x̂. There is an index n0
such that d(xn , xm ) < 21 ε, ∀n, m ≥ nε . Hence d x̂, i(xn0 ) ≤ 12 ε < ε, and i(xn0 ) ∈ Bε (x̂).

If moreover X is complete, then x̂ = lim xn ∈ X, so that i is bijective.

Finally, Xb is complete (compare with the proof in p. 20, where X = Q and X b = R):
b pick qn ∈ X such that d(qn , xn ) < . 1
Given a Cauchy sequence (xn ) in X, n
Then (qn ) is a Cauchy sequence in X: Given ε > 0, there is an index n0 such that n0 > 3ε
and d(xn , xm ) < 3ε , ∀n, m ≥ n0 , so that d(qn , qm ) ≤ d(qn , xn ) + d(xn , xm ) + d(xm , qm ) < ε.
In particular, if we put x = [(qn )] ∈ X, b we have d(qn , x) ≤ ε, ∀n ≥ n0 .
Now, given ε > 0, there exists an index nε such that nε > 2ε and d(qn , x) < 2ε , ∀n ≥ nε , so
that d(xn , x) ≤ d(xn , qn ) + d(qn , x) < n1 + 2ε ≤ ε, ∀n ≥ nε . Hence x = lim xn .
Theorem: Let X, Y be two metric spaces. Any uniformly continuous map f : X → Y induces
a uniformly continuous map fˆ: X
b → Yb such that the following square is commutative:
f
X /Y
i i

fˆ
X
b /Y
b
Proof: If s = (xn ) is a Cauchy sequence in X, then f (s) = (f (xn )) is a Cauchy sequence in Y ,

because f is uniformly continuous.
If (x0n ) is equivalent to (xn ), for any ε > 0 there exists δ > 0 such that d(f (xn ), f (x0n )) < ε
whenever d(xn , x0n ) < δ. Hence (f (xn )) is equivalent to (f (x0n )), and f induces a natural map
ˆ ˆ

f : X → Y , f [s] = f (s) , such that the square is commutative.
b b
Let us see that fˆ is uniformly continuous:

Given ε > 0, there exists δ > 0 such that d(x, y) < δ ⇒ d f (x), f (y) < ε.
Take x̂ = [(x n )], ŷ = [(yn )] ∈ X such that
d(x̂, ŷ) = lim d(xn, yn ) < δ. When n 0, we have
b
ˆ ˆ
d f (xn ), f (yn ) < ε, so that d f (x̂), f (ŷ) = lim d f (xn ), f (yn ) ≤ ε < 2ε.
Universal Property: If f : X → Y is a uniformly continuous map and Y is complete, there

b → Y (a map such that f = φ ◦ i.)
exists a unique uniformly continuous extension φ : X
Proof: The existence follows from the above theorem, since i : Y → Yb is an isometry.
The uniformly continuous extension is unique since it is determined on the dense set i(X).
Corollary: If X is complete, the completion of any subspace Y ⊆ X is just the closure Ȳ .
Proof: The natural map φ : Yb → Ȳ preserves distances, d(x̂, ŷ) = d(φ(x̂), φ(ŷ)), so that in
particular it is injective, and it is surjective because any point ȳ ∈ Ȳ is the limit of some Cauchy
sequence (yn ) in Y : just take points yn ∈ Y ∩ B(ȳ, 1/n).
1.5 Differential Calculus

In this section U will be a non empty open set in R, and a ∈ U .
Definition: A function f : U → R is differentiable at a point a ∈ U when
f (a + t) − f (a) = u(t)t
for some function u(t) continuous at t = 0 (in particular f (x) is continuous at x = a) and we
say that f 0 (a) := u(0) is the derivative of f (x) at the point x = a. Since u(t) = f (a+t)−f
t
(a)
when t 6= 0, such extension u(t) continuous at t = 0 is unique because we must have
f (a + t) − f (a)
f 0 (a) = u(0) = lim
t→0 t

(for any ε ∈ R+ there is δ ∈ R+ such that f (a+t)−f (a)
− f 0 (a) < ε when |t| < δ).

t
We say that f is differentiable when so it is at any point of U (in particular f is continuous),
and the function f 0 : U → R is the derivative of f .
Chain’s Rule: Let U, V be non-empty open sets in R. If y : U → V is differentiable at a

point a ∈ U and f : V → R is differentiable at b = y(a) ∈ V , then the composition z = f (y) is
differentiable at x = a and z 0 (a) = f 0 (b)y 0 (a).
Hence, if y and f are differentiable, so is z = f (y) and z 0 = f 0 (y)y 0 .
Proof: We have y(a + t) = b + ut, f (b + t) = f (b) + vt, where u, v are continuous at t = 0, and
u(0) = y 0 (a), v(0) = f 0 (b). Then

z(a + t) = f b + ut = z(a) + v(ut)u t ,

where the function v u(t)t u(t) is continuous at t = 0, since so are u(t) and v(t).
Hence z is differentiable at x = a and z 0 (a) = v(0)u(0) = f 0 (b)y 0 (a).
1.5. DIFFERENTIAL CALCULUS 29
Theorem: Constant functions have null derivative, and the derivative of x is 1. Sums and
products of differentiable functions also are differentiable and, if a differentiable function f does
not vanish at any point of U , then f1 also is differentiable. Moreover
0
0 0 0 0 0 0 1 −f 0
(f + g) = f + g , (f g) = f g + f g , = 2 ·
f f
Proof: We have f (a + t) = b + ut, g(a + t) = c + vt where u, v are continuous and b = f (a),

u(0) = f 0 (a), c = g(a), v(0) = g 0 (a). Now,
(f + g)(a + t) = b + c + (u + v)t,
(f g)(a + t) = bc + (bv + uc + uvt)t,
1 1 1 1 −u
− = − = t.
f (a + t) f (a) b + ut b (b + ut)b
Definition: A function f : U → R has a local maximum (resp. local minimum) at a ∈ U if

there is a neighborhood V of a where f (x) ≤ f (a), ∀x ∈ V (resp. f (x) ≥ f (a), ∀x ∈ V ).
Theorem: If a differentiable function f : U → R has a local maximum or minimum at a point

a ∈ U , then f 0 (a) = 0.
Proof: We have f (a + t) = f (a) + ut, where u is continuous.

If f has a local minimum, in a neighborhood of t = 0 we have u(t) ≥ 0 when t > 0 (hence
u(0) ≥ 0) and u(t) ≤ 0 when t < 0 (hence u(0) ≤ 0). We conclude that u(0) = 0.
If f has a local maximum, then −f has a local minimum, so that −f 0 (a) = 0.
Rolle’s Theorem: Let f : [a, b] → R be a continuous function, differentiable in (a, b). If

f (a) = f (b), then f 0 (ξ) = 0 at some point a < ξ < b.
Proof: Since f attains a maximum and a minimum in [a, b] and f (a) = f (b), it attains a maximum
or a minimum at an interior point a < ξ < b, so that f 0 (ξ) = 0.
Cauchy’s Mean Value Theorem: Let f, g : [a, b] → R be continuous functions, differentiable
in (a, b). There exists a < ξ < b such that f (b) − f (a) g 0 (ξ) = g(b) − g(a) f 0 (ξ).

0 f 0 (x) g 0 (x)

1 f (x) g(x)
Proof: Apply Rolle’s theorem to h(x) = 1 f (a) g(a), and use h0 (x) = 1 f (a) g(a) .

1 f (b) g(b) 1 f (b) g(b)
When g(x) = x, we obtain:
Lagrange’s Mean Value Theorem: Let f : [a, b] → R be a continuous function, differentiable

in (a, b). There exists a < ξ < b such that f (b) − f (a) = f 0 (ξ)(b − a).
Corollary: If f 0 (ξ) is positive at any point a < ξ < b, then f is a strictly increasing function:
x < y ⇒ f (x) < f (y). If f 0 (ξ) is negative at any point a < ξ < b, then f is a strictly
decreasing function: x < y ⇒ f (x) > f (y).
Theorem: Let f : (a, b) → R be a differentiable function. If f 0 (x) = 0, ∀x ∈ (a, b), then f is a

constant function.
Proof: Apply Lagrange’s theorem to any closed interval [c, d] ⊂ (a, b).
Definition: A function f is of class C 1 if it is differentiable and f 0 is continuous. Inductively,

we say that f is a function of class C m if it is differentiable and f 0 is of class C m−1 . We say
that f is of class C 0 when it is continuous, and of class C ∞ when it is of class C m , ∀m ∈ N, and
C m (U ) will denote the ring of all functions of class C m on U .
The n-th iterated derivative of a function f will be denoted f (n .
Theorem: Let f : U → R be a function of class C 1 . If f 0 (a) > 0 at a point a ∈ U , then f is a

strictly increasing function on a neighborhood V of a.
Hence, if f 0 (a) < 0, then f is a strictly decreasing function on a neighborhood V of a.
Proof: Since f 0 is continuous, it is positive on a neighborhood of a.
Lemma: Let f be a differentiable function of class C n on a connected open set U ⊆ R. If

f 0 (a) = . . . = f (n−1 (a) = 0, then for any point x ∈ U there is an intermediate point ξ between a
and x such that
f (n (ξ)
f (x) − f (a) = (x − a)n ·
n!
Proof: By induction on n, and it is Lagrange’s theorem when n = 1.
When n > 1, applying Cauchy’s mean value theorem to f (x) and g(x) = (x − a)n , we obtain
f (x) − f (a) f 0 (ξ) 1 f 0 (ξ) − f 0 (a)
n
= n−1
=
(x − a) n(ξ − a) n (ξ − a)n−1
where a < ξ < x (or x < ξ < a). Now f 0 is a function of class C n−1 and the first n − 2 derivatives
at the point x = a vanish. By induction, there is a point a < ξ 0 < ξ < x (or...) such that
1 f 0 (ξ) − f 0 (a) 1 f (n (ξ 0 ) f (n (ξ 0 )
= = ·
n (ξ − a)n−1 n (n − 1)! n!
Corollary: Let f : U → R be a function of class C n , n ≥ 2, such that f 0 (a) = . . . = f (n−1 (a) = 0.

If n is even and f (n (a) > 0, then f has a local minimum at x = a.
If n is even and f (n (a) < 0, then f has a local maximum at x = a.
If n is odd and f (n (a) 6= 0, then f has not an extremum at x = a.
Proof: Since f (n is continuous, in a neighborhood of a we have f (x) − f (a) = n1 f (n (ξ)(x − a)n ,

where the sign of f (n (ξ) coincides with the sign of f (n (a). Now the result is clear.
Definition: The n-th Taylor polynomial at a point x = a of a function f of class C n is
f 00 (a) f (n (a)
(Tan f )(x) := f (a) + f 0 (a)(x − a) + (x − a)2 + . . . + (x − a)n .
2! n!
Remark that P (a) = f (a), P 0 (a) = f 0 (a), P 00 (a) = f 00 (a), . . . , P (n (a) = f (n (a).
Taylor’s Formula: Let f be a function of class C n+1 on a connected open set U ⊆ R. For any
point x ∈ U there is an intermediate point ξ between a and x such that
f (n+1 (ξ)
f (x) = (Tan f )(x) + (x − a)n+1 .
(n + 1)!
Proof 2 : The function h(x) = f (x) − (Tan f )(x) is of class C n+1 and h0 (a) = . . . = h(n (a) = 0.
Moreover h(a) = 0 and h(n+1 = f (n+1 . We conclude by the lemma.
2
The proofs of the lemma and this theorem only use the existence of the last derivative, not its continuity.
1.6. INTEGRAL CALCULUS 31
Primitives: Let f : U → R be an arbitrary function. If F : U → R is a differentiable function

and F 0 = f , we say that F is a primitive of f , and we put dF = f dx (because infinitesimally
the increments of F at x = a are the product of f (a) by the increments of x). When U is
connected, if it exists, the primitive is unique
R up to the addition of a Rconstant c and (in view of
the forthcoming Barrow’s rule) we put f (x)dx = F (x) + c, so that dF = F + c.
If u, v : U → R are differentiable, we have (uv)0 = u0 v + uv 0 , so that d(uv) = vdu + udv, and
we obtain the integration by parts formula:
Z Z
udv = uv − vdu.
Moreover, the chain’s rule gives the integration by substitution formula:

Z Z
f (y)dy = f y(x) y 0 (x)dx,

(if F is a primitive of f : V → R, and y = y(x) is a differentiable function U → V , a primitive of

f (y)y 0 is F (y) : U → R). When y(x) is bijective and
the inverse x = x(y) also is differentiable,
if G(x) is a primitive of f y(x) y 0 (x), then G x(y) : V → R is a primitive of f .
1.6 Integral Calculus

Let f : [a, b] → R be a bounded function (there is c ∈ R+ such that |f (x)| ≤ c, ∀x ∈ [a, b]).
Definitions: A partition of the interval [a, b] is any finite subset P containing both ends. We
usually consider it as an increasing sequence a = x0 < x1 < . . . < xn = b, and we put
Ii = [xi−1 , xi ], µ(Ii ) = xi − xi−1 , mi = inf{f (x); x ∈ Ii }, Mi = sup{f (x); x ∈ Ii },
and the lower and upper Riemann sums of f over [a, b] with partition P are
n
P n
P
sf (P ) = mi µ(Ii ) , Sf (P ) = Mi µ(Ii ).
i=1 i=1
Lemma: Given partitions P, P 0 of [a, b], if we put P 00 = P ∪ P 0 , then

sf (P ) ≤ sf (P 00 ) ≤ Sf (P 00 ) ≤ Sf (P 0 ).
Definition: By the lemma, the supremum of the lower sums is bounded above by the infimum of
the upper sums. We say that a bounded function f is Riemann-integrable or just integrable
when both coincide, and we put
Z b
f dx = sup{sf (P )} = inf {Sf (P )}
a P P
because it is “the sum of the product of f by the infinitesimal differences of x”. Hence, f is
integrable when for any ε > 0 there are partitions P, P 0 such that Sf (P 0 ) − sf (P ) < ε (or, by
the lemma, when there is a partition P such that Sf (P ) − sf (P ) < ε).
1. If f, h : [a, b] → R are integrable, then so is any R-linear combination λf + µh and
Z b Z b Z b
(λf + µh)dx = λ f dx + µ hdx .
a a a
In fact we have sf (P ) + sh (P ) ≤ sf +h (P ) ≤ Sf +h (P ) ≤ Sf (P ) + Sh (P ).
Moreover sλf (P ) = λsf (P ) and Sλf (P ) = λSf (P ) when λ ≥ 0. And finally, we also have
s−f (P ) = −Sf (P ) and S−f (P ) = −sf (P ).
Z b
2. If f : [a, b] → R is integrable and m ≤ f ≤ M , then m(b − a) ≤ f dx ≤ M (b − a).
a
Z b Z b
3. If f, h : [a, b] → R are integrable and f ≤ h, then f dx ≤ hdx.
a a
Z b Z b Z b
If f ≤ h, then 0 ≤ h − f , so that 0 ≤ (h − f )dx = hdx − f dx.
a a a
Z b Z b

4. If f : [a, b] → R is integrable, then so is |f | and
f dx ≤ |f |dx.
a a
If m̄i and M̄i are the infimum and supremum of |f | on an interval [xi−1 , xi ] of a partition P
of [a, b], then M̄i − m̄i ≤ Mi − mi because |y| − |x| ≤ |y − x|.
Hence S|f | (P )−s|f | (P ) ≤ Sf (P )−sf (P ) and |f | is integrable. R f ≤ |f | and −f ≤ |f |;
R Moreover
Rb Rb Rb Rb b b
hence a f dx ≤ a |f |dx and − a f dx ≤ a |f |dx, so that a f dx ≤ a |f |dx.

5. Any continuous function f : [a, b] → R is integrable.

Since f is uniformly continuous, given ε > 0, there is δ > 0 such that |f (y) − f (x)| < ε when
|y − x| < δ, so that we have
P Mi − mi < ε when P is a partition with intervals of length < δ.
Hence Sf (P ) − sf (P ) < i εµ(Ii ) = ε(b − a), and f is integrable.
6. Any increasing function (x ≤ y ⇒ f (x) ≤ f (y)) is integrable; hence so is any decreasing
function (x ≤ y ⇒ f (x) ≥ f (y)).
Since mi = f (xi−1 ) and Mi = f (xi ), when we consider the partition Pn of [a, b] in n intervals
of equal length, we have
n
X b − a b − a
Sf (Pn ) − sf (Pn ) ≤ f (xi ) − f (xi−1 ) = f (b) − f (a) ·
n n
i=1
7. If a < c < b, then f is integrable if and only if so are the restrictions of f to [a, c] and [c, b],
and we have Z b Z c Z b
f dx = f dx + f dx.
a a c
Any partition of [a, b] is contained in a partition P = P 0 ∪ P 00 , where P 0 is a partition of [a, c]
and P 00 of [c, b], and we have sf (P ) = sf (P 0 ) + sf (P 00 ), Sf (P ) = Sf (P 0 ) + Sf (P 00 ).
Rx
Theorem: If a bounded function f : [a, b] → R is integrable, then F (x) = a f dt is continuous.
If moreover, f is continuous at an interior point a < c < b, then F (x) is differentiable at the
point x = c and F 0 (c) = f (c).
Ry Ry
Proof: If |f | ≤ c, then |F (y) − F (x)| = | x f dt| ≤ x |f |dt ≤ c|y − x|.
R c+t
Hence f is continuous and F (c + t) − F (c) = c f dx = u(t)t, where
inf{f (x); |x − c| ≤ t} ≤ u(t) ≤ sup{f (x); |x − c| ≤ t}.
If f is continuous at x = c, then u(t) is continuous at t = 0, and u(0) = f (c).
Hence F (x) is differentiable at t = c and F 0 (c) = u(0) = f (c).
Barrow’s Rule: Let f : [a, b] → R be an integrable function. If F : [a, b] → R is a continuous

function, differentiable in (a, b), and F 0 = f , then
Z b
f dx = F (b) − F (a).
a
Proof: Given a partition P , by the mean value theorem there exist ξi ∈ (xi−1 , xi ) such that
n
P Pn
F (b) − F (a) = F (xi ) − F (xi−1 ) = f (ξi )(xi − xi−1 ),
i=1 i=1
Rb
so that sf (P ) ≤ F (b) − F (a) ≤ Sf (P ). Since f is integrable, F (b) − F (a) = a f dx.
Z b
Corollary: If F : U → R is of class C1 and [a, b] ⊂ U , then dF = F (b) − F (a).
a
1
Change of Variable Formula: Let y(x) be a function of class C on a neighborhood of [a, b].
If f is a continuous function on y [a, b] , then
Z y(b) Z b
f y(x) y 0 (x)dx.

f (y)dy =
y(a) a
Proof: If F 0 = f , by the chain’s rule F y(x) is a primitive of f y(x) y 0 (x), so that

Z b Z y(b)
0
f y(x) y (x)dx = F y(b) − F y(a) = f (y)dy.
a y(a)
Notation: When f (x1 , . . . , xn ) is a function defined on a subset of Rn , the partial derivative

∂f
∂xi (a1 , . . . , an ) denotes the derivative, if it exists, of f (a1 , . . . , xi , . . . , an ) at xi = ai (in particular
f must be defined at (a1 , . . . , xi , . . . , an ) when xi is in a small interval around ai ):
∂f f (a1 , . . . , ai + t, . . . , an ) − f (a1 , . . . , ai , . . . , an )
(a1 , . . . , an ) = lim ·
∂xi t→0 t
Differentiation under the Integral Sign: If f (x, t) is a continuous function on a rectangle
Rb
[a, b] × (c, d), then F (t) = a f (x, t)dx is continuous. If moreover ∂f
∂t (x, t) exists and defines a
continuous function on the rectangle, then F (t) is differentiable and
Z b Z b
∂ ∂f
f (x, t)dx = (x, t)dx.
∂t a a ∂t
Proof: Since f (x, t) is continuous, it is uniformly continuous on a compact set [a, b] × [t − c, t + c]

and, given ε > 0, there exists δ > 0 such that, when |h| < δ, we have |f (x, t + h) − f (x, t)| < ε
for any x ∈ [a, b]. Integrating we conclude:
Z b Z b

|F (t + h) − F (t)| = f (x, t + h)dx − f (x, t) < ε(b − a).
a a
Now assume that ∂f ∂t (x, t) exists and it is continuous.

Fix a point (x, t) ∈ [a, b] × (c, d). By the mean value theorem, when h is a small real number
we have f (x, t + h) − f (x, t) = h ∂f
∂t (x, t + ξ), where |ξ| ≤ |h|. Hence,
f (x, t + h) − f (x, t) ∂f ∂f ∂f
− (x, t) = (x, t + ξ) − (x, t),
h ∂t ∂t ∂t
and, since ∂f
∂t is uniformly continuous on a compact set [a, b] × [t − c, t + c], given ε > 0, there
exists δ > 0 such that, when |h| < δ, we have | ∂f ∂f
∂t (x, t + ξ) − ∂t (x, t)| < ε for any x ∈ [a, b].
Integrating we conclude:
R b
f (x, t + h)dx − R b f (x, t) Z b ∂f
a a
− (x, t)dx < ε(b − a).

h a ∂t

Definition:
S A set A ⊂ R has null measure when P for any ε ∈ R+ there is a countable cover
A ⊆ n In by open intervals with total measure n µ(In ) ≤ ε (in the sense that all the finite
sums are bounded above by ε). In particular, any subset B ⊆ A also has null measure.
S
Lemma: Any countable union A = n An of sets of null measure also has null measure.
Proof: Given ε > 0, each set An admits a countable open cover by open intervals with total
measure ≤ 2−n ε. Now, the family
P of all these intervals define a countable cover of A by open
intervals with total measure ≤ n 2−n ε = ε.
Example: Let us divide [0, 1] in three parts, and pick both extremes, A1 = [0, 31 ]∪[ 23 , 1]. Now we
divide both in three parts, and pick the extremes, A2 = [0, 19 ] ∪ [ 29 , 31 ] ∪ [ 23 , 79 ] ∪ [ 89 , 1]. Repeating
the process we obtain a decreasing sequence A1 ⊃ A2 ⊃ . . . ⊃ An ⊃ . . . of closed sets, and
A = n An is the Cantor set. It is a closed set (of cardinal 2ℵ0 ) with null measure.
T
In fact A ⊂ An and An is a disjoint union of 2n closed intervals with total measure 2n /3n ,
so thatit may be covered with open intervals of total measure 2n+1 /3n .
Definition: Let f : X → R be a bounded function on a topological space X. The oscillation

of f on a subset Z ⊆ X is o(f, Z) = supx∈Z f (x) − inf x∈Z f (x), and the oscillation of f at a
point x ∈ X is the infimum o(f, x) of the oscillations of f on the neighborhoods of x.
By definition, f is continuous at the point x if and only if o(f, x) = 0.
The set {x ∈ X : o(f, x) < n1 } is open; hence An = {x ∈ X : o(f, x) ≥ n1 } is closed.
Lebesgue’s Theorem: A bounded function f : [a, b] → R is integrable if and only if the set
A = {x ∈ [a, b] : o(f, x) 6= 0} of points of discontinuity has null measure.
S
Proof: If f is integrable, since A = n An , we only have to show that An has null measure.
Given ε > 0, take a partition P such that Sf (P ) − sf (P ) < nε , and consider the intervals
Uj = (xj−1 , xj ) of the partition intersecting An , so that o(f, Uj ) ≥ n1 . We have
1 P
o(f, Ii )µ(Ii ) = Sf (P ) − sf (P ) < n1 ε
P P
n µ(Uj ) ≤ o(f, Uj )µ(Ij ) ≤
j j Ii ∈P
P
so that j µ(Uj ) < ε. Now, these intervals Uj cover An up to a finite number of points, and we
conclude that An admits a finite cover by open intervals with total measure < 2ε.
Conversely, if A has null measure, then so does Aε = {x ∈ [a, b] : o(f, x) ≥ ε} ⊆ A and, since
Aε is compact, it admits a finite open cover Aε ⊂ I1 ∪ . . . ∪ In by disjoint intervals with total
measure µ(I1 ) + . . . + µ(In ) < ε.
The end points of these intervals define a partition P 0 , and any point of the compact set
K = (I1 ∪ . . . ∪ In )c has a neighborhood where the oscillation of f is < ε. Hence K admits a
finite cover by open intervals where the oscillation of f is < ε, and we may add points of K so
as to obtain a partition P such that the oscillation of f on the remaining intervals Ii is < ε.
We conclude that f is integrable because, if |f | ≤ c, then we have
n
P P n
P P
Sf (P ) − sf (P ) = o(f, Ij )µ(Ij ) + o(f, Ii )µ(Ii ) ≤ 2cµ(Ij ) + εµ(Ii ) < 2cε + (b − a)ε.
j=1 i j=1 i
Corollary: If f, h : [a, b] → R are integrable functions, then so are f h, max(f, h) and min(f, h).
Hence so are |f | and f+ = max(f, 0) and f− = max(−f, 0).
Proof: The points of discontinuity are contained in {disc. of f } ∪ {disc. of h}.
Corollary: If f : [a, b] → R is integrable and g : f ([a, b]) → R is a bounded continuous function,

then the composition g ◦ f is integrable.
1.7. POWER SERIES 35
1.7 Power Series

Definition: Let Y be a metric space. A sequence of maps fn : X → Y converges uniformly
to a limit map f : X → Y if for any ε ∈ R+ there is an index nε such that d f (x), fn (x) < ε,
∀n ≥ nε , x ∈ X. When Y is complete, we say that (fn ) converges uniformly if for any ε > 0
there is an index nε such that d fm (x), fn (x) < ε, ∀n, m ≥ nε , x ∈ X, so that at any point
x ∈ X we have a Cauchy sequence fn (x) in Y , with a limit f (x), and (fn ) converges uniformly
to f : X → Y because d f (x), fn (x) = lim d fm (x), fn (x) ≤ ε, ∀n ≥ nε .
P
Example: If fn : X → C are complex-valued functions, the infinite series n fn converges
uniformly when |fn (x) + . . . + fm (x)| P< ε, for all x ∈ X, 0 n ≤ m. P In particular, since
|fn + . . . + fm | ≤ |fn | + . . . + |fm |, if n |fn | converges uniformly, so does n fn .
Theorem: Let X be a topological space. If a sequence of maps fn : X → Y converges uniformly

on X to a limit map f : X → Y and all the maps fn are continuous at a point p ∈ X, then f
also is continuous at p.

Proof: Given ε > 0 there is an index n such that d f (x), fn (x) < ε, ∀x ∈ X. Since fn is
continuous at p, in a neighborhood U of p we have d fn (x), fn (p) < ε, ∀x ∈ U ; hence

d f (x), f (p) ≤ d f (x), fn (x) + d fn (x), fn (p) + d fn (p), f (p) < ε + ε + ε.
Theorem: Let fn : [a, b] → R be integrable

Rx functions. If (fn ) converges uniformly
Rx to a function
f , then f is integrable, and Fn (x) = a fn dt converge uniformly to F (x) = a f dt. In particular,
Z b Z b
lim fn dx = (lim fn )f dx.
a a
Proof: Given ε > 0, there is an index n such that |f (x) − fn (x)| < ε, ∀x ∈ [a, b], and there is a
partition P such that Sfn (P ) − sfn (P ) < ε. Hence,
Sf (P ) − sf (P ) ≤ |Sf (P ) − Sfn (P )| + |Sfn (P ) − sfn (P )| + |sfn (P ) − sf (P )| < (b − a)ε + ε + (b − a)ε
and we see that f is integrable. Moreover,

Z x
|F (x) − Fn (x)| ≤ |f (t) − fn (t)|dt ≤ (b − a)ε.
a
Theorem: If some differentiable functions fn : (a, b) → R uniformly converge to a function f ,

and the derivatives fn0 uniformly converge to a function h, then f is differentiable and f 0 = h.
Proof: Fix a point c ∈ (a, b), and we may assume c = 0.

Since fn is differentiable, we have fn (t) = fn (0) + un (t)t, where un (t) is continuous and
un (0) = fn0 (0). By the mean value theorem, we have
fn (t) − fm (t) − (fn (0) − fm (0))

un (t) − um (t) = = fn0 (ξ) − fm
0
(ξ) , a < ξ < b.
t−0
Since (fn0 ) converges uniformly, (un ) converges uniformly to a continuous function u.
Since (fn (0)) converges, fn = fn (0) + un t converge uniformly to f (t) = f (0) + u(t)t.
Now, f (t) is differentiable at the point t = 0, because u(t) is continuous, and we have
f 0 (0) = u(0) = lim un (0) = lim fn0 (0) = h(0); i.e. f 0 (c) = h(c). q.e.d.

In fact, only the convergence of fn (c) at a point c ∈ (a, b) is required: in such case this
argument shows that (fn ) uniformly converges to a function f .
an xn be a power series with complex coefficients. If n an bn con-

P P
Abel’s Theorem: Let n P
verges for some b ∈ C, then n |an z n | (hence n an z n ) converges when |z| < |b|.
P
Proof: If P n an bn converges, lim |an bn | = 0. Let c be an upper bound of n

P
n
P z then
n {|an b |}.
z
Now n |an z | ≤ c n b , and the geometric series converges when b < 1.

n with complex co-

P
Definition: The radius of convergence of a power series n an (x − a)P
efficients is the supremum r of all non-negative real numbers c such that n |an |cn converges
∞), so that n |an (z − a)n | converges when |z − a| < r and, by Abel’s
P
(eventually Pr = 0 or r = n
theorem, n an (z − a) does not converge when |z − a| > r.
complex function s(z) = n an (z −a)n on the open
P
When 0 < r ≤ ∞, we have P a well-defined
ball Br (a), and the series n an xn converges uniformly to the function s on any compact set
K ⊂ Br (a).PIn fact, if M is the maximum of |z − a| on K, then we have |an (z − a)n | ≤ |an |M n
on K, and n |an |M n converges because M < r.
Proposition: The radius of convergence r of a power series n an (x − a)n is

P
r−1 = lim n |an | := inf{k ∈ R+ : n |an | ≤ k for any index n, up to a finite number}.
p p
p p
Proof: If c lim n |an | < θP< 1, then c n |an | < θ for Pany index n, up to a finite number and,
since the geometric series θ n converges, so does |a |cn.
p nP n n
If c lim n |an | > 1, then n |an |cn does not converge, since it has infinite terms > 1.

Corollary: n an (x − a)n and n an n(x − a)n−1 have the same radius of convergence.
P P
√
Proof: Let us see that lim n
n = 1. If 1 < θ, when n 0 we have
1
1+ n (n + 1)θ−(n+1)
1> = ·
θ nθ−n
Hence n nθ−n converges and lim nθ−n = 0.
P
√
When n 0, wePhave nθ−n < 1, henceP n < θn , hence 1 < n n <P
θ.
Now, the series n an n(x − a) n−1 and n an n(x − a) = (x − a) n an n(x − a)n−1 have the
n
same radius of convergence, and

p √ p p
lim n n|an | = lim n n lim n |an | = lim n |an |.
Theorem: Let n an (x − a)nPbe a power series with real coefficients and radius of convergence
P
0 < r ≤ n ∞ on (a − r, a + r), and
0
P∞. Then s(x) n−1
= n an (x − a) is a function of class C
s (x) = n nan (x − a) .
n n
P P
Corollary: Let n an (x − a) and n bn (x − a) be power series with real coefficients and
positive radius of convergence. If both define the same function f (x) on some neighborhood of
the point x = a, then an = bn for any index n.
Proof: We have n!an = f (n (a) = n!bn .
Definition: Let U ⊆ R be an open set. A function f : U → R is analytic when

P each point
a ∈ U has an open neighborhood where f (x) is defined by some power series n an (x − a)n
(unique by the above corollary) with positive radius of convergence.
P P
Mertens’
Pn Theorem:
P Let P n an , n bn be two series P
of complex numbers and put cn =
i=0 ai bn−i . If n |an | and n bn converge, then so does n cn and
∞
∞ ∞
P P P
cn = an bn .
i=0 i=0 i=0
Proof: Put An = ni=0 ai , Bn = ni=0 bi , Cn = ni=0 ci = ni=0 ai Bn−i .

P P P P
P P
If A = n an = lim An , and B = n bn = lim Bn , then we have
n
Pn n
P P n
P
|Cn − AB| ≤ ai Bn−i − B
ai + B
ai − AB ≤ |ai | · |B − Bn−i | + |B| · A −
ai .
i=0 i=0 i=0 i=0
Since the second term converges to 0, we only have to show that so does the first term.
P
Put c = sup{|B − Bn |}, k = n |an |.
Given ε > 0, there is an index m such that ∞
P
i=m |ai | < ε and |B − Bn | < ε, ∀n ≥ m.
Hence, when n ≥ 2m, we have
n
P m−1
P n
P
|ai ||B − Bn−i | = |ai ||B − Bn−i | + |ai ||B − Bn−i | < kε + εc.
i=0 i=0 i=m
Corollary: Any product of real analytic functions also is analytic.
1.7.1 Elementary Functions

Let U ⊆ R be an open set. A function f : U → C, f (t) = x(t) + iy(t) is continuous if and only
if so are x(t) and y(t), and we say that f (t) is differentiable (integrable, analytic etc.) when so
Rb Rb Rb
are x(t) and y(t), and we put f 0 (t) := x0 (t) + iy 0 (t), a f (t)dt := a x(t)dt + i a y(t)dt.
P 1 n
Definition: The series n n! c (where 0! := 1) converges for any positive real number c, since
n!cn+1 c P 1 n
lim (n+1)!cn = lim n+1 = 0. Hence, the power series n n! x has infinite radius of convergence
and it defines a continuous function
∞
X zn z2 z3
exp : C −→ C , exp(z) = =1+z+ + + ...
n! 2! 3!
n=0
Now, by Mertens’ theorem, for any a, b ∈ C we have

∞ n ∞
!
X X am bn−m X (a + b)n
exp(a) · exp(b) = = = exp(a + b).
m! (n − m)! n!
n=0 m=0 n=0
1
Since exp(z) exp(−z) = exp(0) = 1, we have exp(−z) = exp(z) (in particular exp(z) 6= 0).
1 1 1
Now we put e := exp(1) = 1 + 1 + + + . . . + + . . ..
2! 3! n!
n! n!
It is an irrational number, since n!e = n! + . . . + n! + (n+1)! + . . . never is an integer, because
n!
n! + . . . + n! is a natural number and
∞
1 1 X 1 1
0< + ... + + ... < r
= ·
n+1 (n + 1) . . . (n + r) (n + 1) n
r=1
n n m
= exp(n) = en .

For any rational number m we have exp m
n
n
= e m , and for any complex number we put ez := exp(z).

Hence exp m
n
The analytic function et : R → R+ is an increasing function, since so are the functions tn! , and
the derivative is (et )0 = et . Since e > 2, we have that et is arbitrarily large when t → ∞ (hence
et is arbitrarily small when t → −∞) and we see that et : R → R+ is a group isomorphism.
Since ez = ez̄ , we have |eit |2 = eit e−it = 1 when t ∈ R. Hence |eit | = 1, and we put
eit = cos t + i sin t,

∞
eit + e−it X (−1)n 2n
cos t = = t ,
2 (2n)!
n=0
∞
eit − e−it X (−1)n 2n+1
sin t = = t ,
2i (2n + 1)!
n=0
so that cos2 t + sin2 t = |eit |2 = 1.

Moreover, cos(−t) = cos t, sin(−t) = − sin t because e−it = eit , and eit eis = ei(t+s) gives
cos(t + s) = (cos t)(cos s) − (sin t)(sin s),

sin(t + s) = (cos t)(sin s) + (sin t)(cos s).
Since (eit )0 = ieit , we have (cos t)0 = − sin t, (sin t)0 = cos t.
Now we have cos 0 = 1, sin 0 = 0 and
22 24 26 28 22 24 1
cos 2 = 1 − + − + − ... < 1 − + =−
2! 4! 6! 8! 2! 4! 3
because the terms are decreasing with alternate signs.
Let t0 be the minimum positive real number where cos t0 = 0. We put π := 2t0 .
Since sin0 (t) = cos t is positive in [0, t0 ), we have sin t0 > 0. Since sin2 t0 = 1, we have
π
e2i = i , cos(t + π2 ) = − sin t , sin(t + π2 ) = cos t,
eπi = −1 , cos(t + π) = − cos t , sin(t + π) = − sin t,
2πi
e =1 , cos(t + 2π) = cos t , sin(t + 2π) = sin t.
Now, cos t is a decreasing function on [0, π2 ], while sin t is increasing, and cos 0 = sin π2 = 1,
cos π2 = sin 0 = 0. Hence, any real number −1 ≤ x ≤ 1 is x = cos α for a unique 0 ≤ α ≤ π.
Moreover, if a2 + b2 = 1, then we have a + bi = eiθ for some real number θ (well defined up
to the addition of an integer multiple of 2π). Summarizing:
Theorem: The exponential et : R → R+ is a group isomorphism.

The exponential eit : R → {z ∈ C : |z| = 1} is a group epimorphism with kernel 2πiZ.
The exponential ez : C → C − {0} is a group epimorphism with kernel 2πiZ.
Definition: If z is a complex number of modulus ρ 6= 0, then z = ρeiθ = ρ(cos θ + i sin θ)

for some real number θ (well defined up to the addition of an integer multiple of 2π) named
0 0
argument of z. We have arg(zz 0 ) = arg z + arg z 0 because eiθ eiθ = ei(θ+θ ) .
Hence, any non null complex number z = ρeiθ has n complex n-th roots, namely
√ θ+2kπ
n
ρ ei n , k = 0, . . . , n − 1.
Lemma: Let P (x) = c0 xn + c1 xn−1 + . . . + cn ∈ C[x]. If we put c = max{|c1 |, . . . , |cn |}, then
for any complex number z of modulus ρ we have

c
ρn |c0 | − ρ−1 ≤ |P (z)|.
Proof: |P (z)| ≥ |c0 z n | − |c1 z n−1 + . . . + cn | ≥ |c0 |ρn − c(ρn−1 + . . . + 1) =

= |c0 |ρn − c(ρn − 1)/(ρ − 1) ≥ |c0 |ρn − cρn /(ρ − 1) = ρn (|c0 | − c/(ρ − 1)).
D’Alembert’s Theorem: Any non constant polynomial P (x) ∈ C[x] has a complex root.
Proof: The disk Dρ := {z ∈ C : |z| ≤ ρ} is a closed bounded set in C; hence it is compact and
the continuous function f : Dρ → R, f (z) = |P (z)|, attains a minimum at some point a ∈ Dρ .
By the above lemma, we may fix ρ so that |P (z)| > |P (0)| whenever |z| = ρ, so that a must
be an interior point of Dρ .
If |P (a)| > 0, then we consider the first monomial with non null coefficient in P (a + z):
P (a + z) = P (a) + cd z d + cd+1 z d+1 + . . . + cn z n , cd 6= 0.
Now we may choose z so that |z| < 1 and |P (a) + cd z d | = |P (a)| − |cd z d | (just take z of small
modulus and argument d1 (π + arg P (a) − arg cd )). Hence
|P (a + z)| ≤ |P (a)| − |cd z d | + |cd+1 z d+1 + . . . + cn z n | ≤ |P (a)| − |cd | · |z d | + c|z|d+1 ,
where c := (n − d) max{|cd+1 |, . . . , |cn |}. It is clear that we may choose z of so small modulus
that |P (a + z)| < |P (a)| and a + z ∈ Dρ . Absurd.
Hence |P (a)| = 0, and z = a is a complex root of P .
Chapter 2
Linear Algebra
2.1 Groups
Definitions: A permutation of n elements is a bijection σ : {1, . . . , n} → {1, . . . , n}. The
product of permutations is the composition and Sn is the set of permutations of n elements.
Given different numbers 1 ≤ a1 , . . . , ar ≤ n, the cycle (a1 . . . ar ) ∈ Sn is the permutation
transforming ai into ai+1 (and ar into a1 ) and fixing any other number.
The cycles (a1 a2 ) of length 2 are named transpositions.
Two cycles σ = (a1 . . . ar ), τ = (b1 . . . bs ) are disjoint when ai 6= bj for any pair of indices
i, j; so that στ = τ σ. Any permutation is a product σ = α1 . . . αr of disjoint cycles, uniquely
up to the order of the factors. If di is the length of αi , the form of σ is d1 , . . . , dr .
Lemma: Any permutation is a product of transpositions.
Proof: (a1 . . . ar ) = (a1 a2 )(a2 a3 ) . . . (ar−1 ar ).
Proposition: Two permutations σ, τ ∈ Sn are conjugate (τ = γσγ −1 for some γ ∈ Sn ) if and

only if both have equal form.
Proof: Let σ = (a1 . . . ad1 )(b1 . . . bd2 ) . . . (c1 . . . cdr ) be a product of disjoint cycles.
If γ ∈ Sn , and we put i0 = γ(i), then γσγ −1 = (a01 . . . a0d1 )(b01 . . . b0d2 ) . . . (c01 . . . c0dr ).
Q Q
Definition: Let ∆ = i<j (xj − xi ). If σ ∈ Sn , then the factors of σ∆ = i<j (xσ(j) − xσ(i) )
coincide, up to a sign, with the factors of ∆, and the sign of σ is defined to be
(∗) σ∆ = (sgn σ)∆ , sgn σ = ±1.
Proposition: sgn τ σ = (sgn τ )(sgn σ), and sgn (a1 . . . ar ) = (−1)r−1 .
Proof: To show that sgn τ σ = (sgn τ )(sgn σ), just apply τ to the variables in (∗),
τ (σ∆) = (sgn σ)(τ ∆) = (sgn σ)(sgn τ )∆.
A direct calculation shows that sgn (12) = −1, and for any other transposition τ we have
τ = γ(12)γ −1 , so that sgn τ = (sgn γ)(sgn (12))(sgn γ −1 ) = sgn (12) = −1.
Now, (a1 . . . ar ) is a product of r − 1 transpositions, hence the sign is (−1)r−1 .
.
Definition: A map G × G →
− G defines a group structure on the set G if
1. a · (b · c) = (a · b) · c; ∀a, b, c ∈ G.
41
42 CHAPTER 2. LINEAR ALGEBRA
2. There exists1 1 ∈ G such that 1 · a = a · 1 = a, ∀a ∈ G.
3. For any a ∈ G there exists2 a−1 ∈ G such that a · a−1 = a−1 · a = 1.
If a · b = b · a, ∀a, b ∈ G, the group is said to be abelian, the operation is denoted by +, the

neutral element 0, and the inverse −a (and it is named opposite).
Examples: Z is an abelian group with the sum.

Sn is a group (non abelian when n ≥ 3) with the product of permutations.
Groups allow simplification: if ab = ac, then a−1 ab = a−1 ac, and b = c.
Definitions: Let G, G0 be two groups. A map f : G → G0 is a group morphism if
f (a · b) = f (a) · f (b), ∀a, b ∈ G.
If moreover f is bijective, it is a group isomorphism, (automorphism when G0 = G) and in

such a case so is f −1 , and we put G ' G0 .
Isomorphic groups have the same properties in Group Theory.
1. If f is a group morphism, then f (1) = 1 and f (a−1 ) = f (a)−1 .

f (1) = f (1 · 1) = f (1) · f (1), and f (a) · f (a−1 ) = f (aa−1 ) = f (1) = 1.
2. Compositions of group morphisms are group morphisms.

(gf )(ab) = g(f (ab)) = g(f (a)·f (b)) = g(f (a))·g(f (b)) = (gf )(a)·(gf )(b).
3. If f : G → G0 is a group isomorphism, so is the bijection f −1 : G0 → G.

f f −1 (a0 ) · f −1 (b0 ) = f f −1 (a0 ) · f f −1 (b0 ) = a0 · b0 = f f −1 (a0 b0 ) , and f is injective.

Definition: A subset H of a group G is a subgroup if it is a group with the product of G,
1. a, b ∈ H ⇒ a · b ∈ H.
2. 1 ∈ H.
3. a ∈ H ⇒ a−1 ∈ H.
Examples: The subgroups 1 and G are the trivial subgroups of G.

The intersection of any family of subgroups is also a subgroup.
The subgroup (a1 , . . . , an ) generated by a1 , . . . , an ∈ G is the minimal subgroup containing
them (the intersection of all subgroups containing them).
The subgroup generated by an element a ∈ G is (a) = {. . . , a−2 , a−1 , 1, a, a2 , . . .}.
The subgroup of Z generated by a, b ∈ Z is aZ + bZ = {ax + by : x, y ∈ Z}.
Theorem: Any subgroup H of Z is generated by a unique natural number, H = nZ .
Proof: If H = 0, take n = 0.
If H 6= 0, let n be the first positive number in H (it exists since −H = H).
Now nZ ⊆ H, because H is a subgroup, and, if m ∈ H, we divide m by n,
m = cn + r; 0 ≤ r ≤ n − 1,
r = m − cn ∈ H;
1
This neutral element is unique: if e is also neutral, then e = e · 1 = 1.
2
This inverse element is unique: if a · b = 1, then a−1 = a−1 · 1 = a−1 · (a · b) = (a−1 · a) · b = 1 · b = b.
2.1. GROUPS 43
hence r = 0, by the choice of n, and m ∈ nZ. Therefore H = nZ.

The uniqueness is evident.
Example: Given n1 , n2 ∈ N, we have n1 Z + n2 Z = dZ, n1 Z ∩ n2 Z = mZ for some d, m ∈ N, the

greatest common divisor and the least common multiple of n1 and n2 .
Hence n1 and n2 are coprime if and only if n1 Z + n2 Z = Z.
Proposition: Let f : G → Ḡ be a group morphism. If H is a subgroup of G, then f (H) is a

subgroup of Ḡ. If H̄ is a subgroup of Ḡ, then f −1 (H̄) is a subgroup of G. Hence the image
Im f = f (G) is a subgroup of Ḡ, and the kernel Ker f = f −1 (1) is a subgroup of G.
Proof: 1 ∈ f (H) because f (1) = 1 and 1 ∈ H.

If f (h0 ), f (h) ∈ f (H), then f (h0 )f (h) = f (h0 h) ∈ f (H), and f (h)−1 = f (h−1 ) ∈ f (H).
1 ∈ f −1 (H̄), because f (1) = 1 ∈ H̄.
If g 0 , g ∈ f −1 (H̄), then f (g 0 g) = f (g 0 )f (g) ∈ H̄, and f (g −1 ) = f (g)−1 ∈ H̄.
Proposition: A group morphism f is injective if and only if Ker(f ) = 1.
Proof: If f is injective, and f (g) = 1 = f (1), then g = 1.

If Ker f = 1, and f (a) = f (b), then f (a−1 b) = 1; hence a−1 b = 1, and a = b.
2.1.1 The Quotient Group

Definitions: Two elements a, b ∈ G are congruent modulo a subgroup H when a−1 b ∈ H; i.e.,
b ∈ aH, and we put a ≡ b (mod. H). This is an equivalence relation,
1. a ≡ a, because a−1 a = 1 ∈ H for all a ∈ G.
−1
2. If a ≡ b, then a−1 b ∈ H; hence b−1 a = a−1 b ∈ H, and b ≡ a.
3. If a ≡ b and b ≡ c, then a−1 b, b−1 c ∈ H; hence a−1 c = (a−1 b)(b−1 c) ∈ H, and a ≡ c.
G/H denotes the quotient set, and the class of a ∈ G is just aH = {ah : h ∈ H}.
The order of G is the cardinal |G|, and the index of H is the cardinal of G/H.
Lagrange’s Theorem: Let H be a subgroup of a finite group G. The order of H divides the
order of G, and the quotient is the index of H,
|G/H| = |G| / |H| .
Proof: All the equivalence classes aH have equal cardinal, because the surjective maps
H −→ aH, h 7→ ah,
are injective: If ax = ay, then x = a−1 (ax) = a−1 (ay) = y.

Hence |G| is the product of |H| by the number |G/H| of equivalence classes.
Definition: A subgroup H of G is normal, and we put H G, when gHg −1 ⊆ H, ∀g ∈ G.

The kernel of any group morphism f : G → G0 is a normal subgroup:
If h ∈ Ker f , then f (ghg −1 ) = f (g)f (h)f (g)−1 = f (g)f (g)−1 = 1, hence ghg −1 ∈ Ker f .
Theorem: If H is a normal subgroup of G, there exists a unique group structure on the quotient
set G/H such that π : G → G/H is a group morphism. Moreover, Ker π = H .
Proof: The only possible product, [a] · [b] = [ab] is well-defined, and it defines a group structure,
1. If [a0 ] = [a], then a0 = ah, a0 b = ahb = ab(b−1 hb) ∈ abH; hence [a0 b] = [ab].
2. ([a] · [b]) · [c] = [ab] · [c] = [(ab)c] = [a(bc)] = [a] · [bc] = [a] · ([b] · [c]).
3. [a] · [1] = [a · 1] = [a], [1] · [a] = [1 · a] = [a].
4. [a] · [a−1 ] = [a · a−1 ] = [1], [a−1 ] · [a] = [a−1 · a] = [1].
5. Kerπ = {a ∈ G : [a] = [1]} = [1] = H.

Universal Property: Let H be a normal subgroup of G. If f : G → G0 is a group morphism
and f (H) = 1, then there is a unique group morphism φ : G/H → G0 such that φ([a]) = f (a),
f
G / G0 f = φπ
A
π
φ
G/H
Proof: The map φ : G/H → G0 , φ([g]) = f (g), is well-defined,
[g 0 ] = [g], g 0 = gh ∈ gH ⊆ g(Ker f ), f (g 0 ) = f (g)f (h) = f (g) · 1 = f (g),
and it is a group morphism: φ([a] · [b]) = φ([ab]) = f (ab) = f (a) · f (b) = φ([a]) · φ([b]).
Isomorphism Theorem: If f : G → G0 is a group morphism, then we have a group isomor-

∼
phism φ : G/Ker f −
→ Im f , φ([g]) = f (g).
Proof: By the universal property we have an epimorphism φ : G/Ker f → Im f , φ([g]) = f (g),

and it is injective: If 1 = φ([g]) = f (g), then g ∈ Ker f and [g] = 1.
Definition: A group G is cyclic if it is generated by some element,
G = (g) = {. . . , g −n , . . . , g −1 , g 0 = 1, g, g 2 , . . . , g n , . . .}.
For example, Z/nZ is a cyclic group, generated by the class [1].
Classification of Cyclic groups: Any cyclic group G is isomorphic to Z/nZ for some n ∈ N.
Proof: If G = (g), the morphism f : Z → G, f (m) = g m , is surjective.

Since Ker f = nZ for some n ∈ N, we have an isomorphism φ : Z/nZ ' G, φ([m]) = g m .
Definition: The order of an element g ∈ G is the order of the subgroup (g).

1. The order of g is the first natural number r 6= 0 such that g r = 1 (if it exists), and in such a
case g m = 1 if and only if m is a multiple of r.
In fact, φ : Z/rZ → (g), φ([m]) = g m , is an isomorphism.
2. If G is a finite group of order n, then g n = 1 for all g ∈ G.
3. The order of a permutation of form d1 , . . . , dr is the l.c.m.(d1 , . . . , dr ) .
4. The generators of the group Z/nZ are just the classes [m] of numbers coprime to n.
In fact [m] generates Z/nZ if and only if π −1 ([mZ]) = mZ + nZ coincides with Z.
5. The alternate subgroup An is the kernel of the group morphism sgn : Sn → {±1}. It is a
normal subgroup of index 2, because Sn /An ' {±1}; hence |An | = n!/2 .
2.2. RINGS 45
2.2 Rings
+ .
Definition: Two maps A × A −→ A, A × A →
− A define a structure of ring (commutative with
unit) on the set A if
1. (A, +) is a commutative group.
2. a · (b · c) = (a · b) · c, ∀a, b, c ∈ A.
3. a · (b + c) = a · b + a · c, ∀a, b, c ∈ A.
4. a · b = b · a, ∀a, b ∈ A.
5. There exists3 1 ∈ A such that a · 1 = a, ∀a ∈ A.
a·
By (3), the map A −−→ A is a group morphism; hence a · 0 = 0, a · (−b) = −(a · b).
Let A, B be two rings. A group morphism f : A → B is a ring morphism if
f (a · b) = f (a) · f (b),
f (1) = 1.
If moreover f is bijective, it is a ring isomorphism, (automorphism when B = A) and in

such a case so is f −1 , and we put A ' B.
Isomorphic rings have the same properties in Ring Theory.
Compositions of ring morphisms are also ring morphisms.
Definition: An element a ∈ A is a unit or invertible if ab = 1 for some b ∈ A.
A ring k 6= 0 is a field if any non null element is invertible. In this notes, unless otherwise
stated, k always denotes a field, and the elements of k are named scalars.
Definition: An additive subgroup B of a ring A is a subring if it is stable under products and
1 ∈ B, so that B also is a ring with the operations of A.
An additive subgroup a ⊆ A is an ideal if it is closed under products by arbitrary elements
of A (a ∈ A, b ∈ a ⇒ ab ∈ a ).
Examples: Z is a ring, while Q, R and C are fields.

Polynomials in a variable x with coefficients in a ring A form a ring A[x].
The invertible elements of a ring A form a multiplicative abelian group A∗ .
If f : A → B is a ring morphism and a is invertible in A, then f (a−1 ) · f (a) = f (a−1 a) =
f (1) = 1; hence f (a) is invertible in B. Moreover, f : A∗ → B ∗ is a group morphism.
If an ideal a contains an invertible element, then 1 ∈ a and a = A. In particular, the unique
ideals of a field k are 0 and k.
Intersections of ideals of A are also ideals of A.
Any ideal of Z is nZ, where n is a natural number, because so is any subgroup.
The kernel of a ring morphism A → B is an ideal of A, and the image is a subring of B.
Theorem: Let a be an ideal of a ring A. The quotient group A/a has a unique ring structure
such that the canonical projection π : A → A/a is a ring morphism. Moreover, Ker π = a.
Proof: The unique possible product, [a] · [b] = [ab], is well-defined (and it is easy to check that
it defines a ring structure in A/a, the unity being [1]):
[a] = [a0 ], a0 = a + c ∈ a + a, a0 b = ab + cb ∈ ab + a, [a0 b] = [ab].

3
This identity element is unique: if e is another identity, then e = e · 1 = 1.
Example: If p is a prime number, then Fp := Z/pZ is a field. In fact, if 0 < n < p, then
pZ ⊂ nZ + pZ; hence nZ + pZ = Z, and we have nm + pa = 1, so that [n] · [m] = 1.
Universal Property: If a ring morphism f : A → B vanishes on an ideal a of A, then there

is a unique ring morphism φ : A/a → B such that φ([a]) = f (a),
f
A /B f = φπ
B
π
φ
A/a
Proof: By the universal property of the quotient group, there is a unique group morphism
φ : A/a → B such that f = φπ, and φ is a ring morphism,
φ([a] · [b]) =φ([ab]) = f (ab) = f (a)f (b) = φ([a])φ([b]),

φ(1) = φ(π(1)) = f (1) = 1.
Isomorphism Theorem: If f : A → B is a ring morphism, then we have a ring isomorphism

∼
φ : A/Ker f −
→ Im f , φ([a]) = f (a).
Proof: It is a group isomorphism, and a ring morphism by the universal property.
Definitions: An element a ∈ A is a zero divisor if ab = 0 for some b 6= 0; i.e., the group

a·
morphism A −−→ A is not injective. A ring A 6= 0 is integral (or a domain) when ab = 0 ⇒
a = 0 or b = 0.
A proper element (non null nor invertible) of a domain is irreducible if it is not a product
of two proper elements.
Any field is a domain: If ab = 0 and b 6= 0, then 0 = abb−1 = a.
If A is an integral ring, and P, Q ∈ A[x] are non null, then

P
P Q = (an xn + . . . + a0 )(bm xm + . . . + b0 ) = an bm xn+m + . . . + ai bj xd + . . . + a0 b0 6= 0
i+j=d
because an bm 6= 0 when an , bm 6= 0. Moreover deg(P Q) = deg P + deg Q. Hence A[x] also is a

domain and any invertible polynomial is constant: A[x]∗ = A∗ .
Division Theorem: Let Q 6= 0 be a polynomial with coefficients in a field k. For any P ∈ k[x]
there is a unique pair C, R ∈ k[x] (named quotient and remainder) such that
P = CQ + R, deg R < deg Q or R = 0.
Proof: The existence is given by the usual division algorithm.

Now, if P = C1 Q + R1 , where deg R1 < deg Q or R1 = 0, then Q(C1 − C) = R − R1 .
Since deg(R − R1 ) < deg Q, then C1 − C = 0, and R − R1 = 0.
Ruffini’s Rule: A polynomial P (x) ∈ k[x] is a multiple of x − a if and only if P (a) = 0.
Proof: If P (x) = C(x)(x − a) + r, then P (a) = C(a) · 0 + r = r. q.e.d.
1. a ∈ k is a root of P ∈ k[x] if and only if P is a multiple of x − a in the ring k[x].
2. Any irreducible polynomial in k[x] of degree > 1 has no root in k.

2.3. VECTOR SPACES 47
3. A polynomial of degree 2 or 3 is irreducible in k[x] if and only if it has no root in k.

If P = Q1 Q2 , then some factor has degree 1, and it has some root in k.
4. If a1 , . . . , ar ∈ k are different roots of P ∈ k[x], then P is a multiple of (x − a1 ) . . . (x − ar ).

By Ruffini’s rule, P = (x − a1 )Q, and a2 , . . . , ar are roots of Q. Hence (x − a2 ) . . . (x − ar )
divides Q by induction on r, and (x − a1 ) . . . (x − ar ) divides P .
5. Any polynomial of degree 2 with complex coefficients has a complex root.

It is enough to show that any complex number z has a complex square root, and we may
1+z 2 (1+z)2
= 1+z z z̄+z

assume that |z| = 1. Now, if z =
6 −1, we have |1+z| = (1+z)(1+z̄) 1+z̄ = 1+z̄ = z.
6. The number of roots of a polynomial 0 6= P ∈ k[x] in k is bounded by the degree of P .
7. Let a1 , . . . , ar ∈ k be different elements of a field. Given b1 , . . . , br ∈ k, there is a unique

polynomial P (x) ∈ k[x], deg P < r, such that P (ai ) = bi .
In fact we have the Lagrange’s interpolation formula:
r
X Qj (x) (x − a1 ) . . . (x − ar )
P (x) = bj , Qj (x) = ·
Qj (aj ) (x − aj )
j=1
2.3 Vector Spaces

Definition: Let k be a field. A structure of k-vector space on an abelian group E (the
.
elements of E are named vectors) is defined by a map k × E −
→ E such that
1. λ(e1 + e2 ) = λe1 + λe2 , ∀λ ∈ k, e1 , e2 ∈ E.
2. (λ1 + λ2 )e = λ1 e + λ2 e, ∀λ1 , λ2 ∈ k, e ∈ E.
3. (λµ)e = λ(µe), ∀λ, µ ∈ k, e ∈ E.
4. 1 · e = e for all e ∈ E.
and a subgroup V ⊆ E is a vector subspace if λV ⊆ V , ∀λ ∈ k.

λ· ·e
By (1) and (2), the maps E −−→ E and k −−→ E are group morphisms; hence
λ · 0 = 0, λ · (−e) = −(λe), 0 · e = 0, (−λ)e = −(λe).
Moreover, if λe = 0, and λ 6= 0, then 0 = λ−1 (λe) = 1 · e = e.

Definition: A group morphism f : E → F between k-vector spaces is k-linear if
f (λ · e) = λ · f (e); ∀λ ∈ k, e ∈ E.
If moreover it is bijective, it is a linear isomorphism (automorphism when F = E) and in

such a case so is f −1 , and we put E ' F .
Isomorphic vector spaces have the same properties in Linear Algebra.
Compositions of linear maps are also linear maps.
Proposition: Let f : E → F be a linear map. If V ⊆ is a vector subspace, then f (V ) is a vector

subspace of F . If W ⊆ F is a vector subspace, then f −1 (W ) is a vector subspace of E. Hence
Im f = f (E) is a vector subspace of F and Ker f = f −1 (0) is a vector subspace of E.
Proof: f (V ) and f −1 (W ) are subgroups (p. 43) and f (v) ∈ f (V ) ⇒ λf (v) = f (λv) ∈ f (V ).
Moreover, if e ∈ f −1 (W ), then f (λe) = λf (e) ∈ W ; hence λe ∈ f −1 (W ).
Examples: k n is a k-vector space with the product λ(µ1 , . . . , µn ) = (λµ1 , . . . , λµn ).

E and 0 are the trivial vector subspaces of E.
If V and W are vector subspaces of E, so are V ∩ W and the sum
V + W = {v + w : v ∈ V and w ∈ W }.
Let A be a matrix m × n with coefficients in k. The map f : k n → k m , f (X) = AX, is linear,

the kernel being the space of solutions of the homogeneous system AX = 0, and B ∈ Im f just
when the system of linear equations AX = B is consistent.
If e1 , . . . , en ∈ E, the map f : k n → E, f (λ1 , . . . , λn , ) = λ1 e1 + . . . + λn en , is linear, the
image being the minimal vector subspace containing e1 , . . . , en ,
he1 , . . . , en i = ke1 + . . . + ken = {λ1 e1 + . . . + λn en : λ1 , . . . , λn ∈ k}.
Theorem: If f, g : E → F are linear maps, then so are the maps
λf : E −→ F, (λf )(e) = λ · f (e),

f + g : E −→ F, (f + g)(e) = f (e) + g(e),
where λ ∈ k; and the set Homk (E, F ) of all k-linear maps E → F is a k-vector space.
Proof: It follows directly from the definitions.
Theorem: If V is a vector subspace of E, there exists a unique vector space structure on the
quotient group E/V such that π : E → E/V is a linear map. Moreover, Ker π = V .
Proof: The unique possible product, λ · [e ] = [ λe ], is well-defined,
[e] = [e0 ], e0 − e ∈ V, λe0 − λe = λ(e0 − e) ∈ V, [ λe ] = [ λe0 ],
and it is easy to check that it defines a vector space structure on E/V .
Definitions: X = p + V is the linear subvariety of E with direction V passing through the

point p, and linear subvarieties of direction V form the vector space E/V .
Two linear subvarieties p + V , q + W are parallel if V ⊆ W or W ⊆ V .
Universal Property: Let V be a vector subspace of E. Any linear map f : E → F vanishing

on V uniquely factors through a linear map φ : E/V → F such that φ([e]) = f (e),
f
E /F f = φπ
B
π φ
E/F
Proof: By the universal property of the quotient group, there is a unique group morphism
φ : E/V → F such that f = φ ◦ π, and φ is linear,
φ(λ[e]) = φ([λe]) = f (λe) = λf (e) = λφ([e]).
Isomorphism Theorem: If f : E → F is a k-linear map, then we have a k-linear isomorphism

∼
φ : E/Ker f −
→ Im f , φ([e]) = f (e).
Proof: It is a group isomorphism by the isomorphism theorem for group morphisms, and it is
linear by the universal property of E/Ker f .
Theorem: If p : E → Ē is a linear surjective map, we have a lattice isomorphism

Vector sub- Vector subspaces of
∼
−−→ , V̄ 7→ p−1 (V̄ ),
spaces of Ē E containing Ker p
the inverse isomorphism being the map V 7→ p(V ). If V = p−1 (V̄ ), then we have a natural
isomorphism E/V ' Ē/V̄ .
Proof: We have p(p−1 V̄ ) = V̄ because p is surjective and, when Ker p ⊆ V , we also have
p−1 p(V ) = V + Ker p = V,

p π
so that the kernel of the epimorphism E →
− Ē −
→ Ē/V̄ is just V ; hence E/V ' Ē/V̄ .
fn−1 fn
Definition: A sequence . . . → En−1 −−−→ En −−→ En+1 → . . . of linear maps is exact when
Im fn−1 = Ker fn for any index n.
i p
A sequence 0 → E 0 → − E → − E 00 → 0 is exact when i is injective and p is a surjective map
with kernel Im i, so that E 00 ' E/Im i. These are named short exact sequences.
Definitions: A vector space E 6= 0 is simple if 0 and E are the only vector subspaces of E.
A flag in E of length n is a maximal increasing sequence E0 ⊂ E1 ⊂ . . . ⊂ En of vector
subspaces (E0 = 0, En = E and the quotients Ei /Ei−1 are simple). E is finite-dimensional if
it admits some flag, and the dimension of E is the minimal length of all flags.
The unique vector space of dimension 0 is E = 0.
Vector spaces of dimension 1 are just simple spaces, and k is simple.
Theorem: Let E be a vector space of finite dimension.
1. dim V ≤ dim E for any subspace V ⊆ E; and V = E when the equality holds.
2. All flags in E have equal length, namely the dimension of E.

i p
3. If 0 → E 0 −−→ E −−→ E 00 → 0 is exact, then dim E = dim E 0 + dim E 00 .
Proof: (1) Let 0 = E0 ⊂ E1 ⊂ . . . ⊂ En = E be a flag, n = dim E, and put Vi = Ei ∩ V .

The kernel of Vi → Ei /Ei−1 is Ei−1 ∩ Vi = Vi−1 , so that we have an injective linear map
Vi /Vi−1 ,→ Ei /Ei−1 and we see that Vi /Vi−1 is simple or null (i.e., Vi−1 = Vi ).
Eliminating repetitions in 0 ⊆ V1 . . . ⊆ Vn = V we obtain a flag; hence dim V ≤ dim E.
If dim V = n, then all the inclusions Vi−1 ⊂ Vi are strict.
Now, if Ei−1 = Vi−1 ⊂ Vi ⊆ Ei , then Vi = Ei because Ei /Ei−1 is simple.
Since V0 = E0 , we see that V = Vn = En = E.
(2) If 0 ⊂ Ē1 ⊂ . . . ⊂ Ēd = E is a flag, then d ≤ dim E because dim Ēi−1 < dim Ēi .
By definition dim E ≤ d; hence d = dim E.
0 = E 0 and 0 ⊂ . . . ⊂ E 00 = E 00 , then
(3) Given flags 0 ⊂ . . . ⊂ Em d
0
0 ⊂ . . . ⊂ i(Em ) = p−1 (0) ⊂ . . . ⊂ p−1 (Ed00 ) = E
is a flag in E by the former theorem; hence dim E = m + d = dim E 0 + dim E 00 .

Corollary: dim E/V = dim E − dim V .

dim E = dim (Ker f ) + dim (Im f ), for any linear map f : E → F .
dim (E × F ) = dim E + dim F .
dim (V + W ) = dim V + dim W − dim (V ∩ W ).
i π
Proof: 0 → V →
− E−
→ E/V → 0 is exact, because V = Ker π.
i f
0 → Ker f →
− E−
→ Im f → 0 is exact.
1 i 2 π
0→E−
→ E × F −→ F → 0 is exact; i1 (e) = (e, 0), π2 (e, v) = v.
i s
0→V ∩W →
− V ×W →
− V + W → 0 is exact; i(e) = (e, −e), s(v, w) = v + w.
Corollary: Any injective or surjective k-linear map f : E → F between k-vector spaces of equal
dimension (finite, of course) is an isomorphism.
Proof: dim (Ker f ) = dim F − dim (Im f ); hence, Ker f = 0 if and only if Im f = F .
2.3.1 Bases
Any family of vectors e1 , . . . , en ∈ E defines a linear map
f : k n −→ E, f (λ1 , . . . , λn ) = λ1 e1 + . . . + λn en .
Definitions: The vectors e1 , . . . , en form a base of E if f is an isomorphism (any vector e ∈ E

may be uniquely written as a linear combination e = x1 e1 + . . . + xn en ).
In such a case n = dim E, and (x1 , . . . , xn ) ∈ k n are the coordinates of e in such base.
The usual base of k n is e1 = (1, 0, . . . , 0), e2 = (0, 1, . . . , 0), . . . , en = (0, . . . , 0, 1).
The vectors e1 , . . . , en generate or span E if f is surjective, ke1 + . . . + ken = E.
In such a case n ≥ dim E, and they form a base of E when the equality holds.
The vectors e1 , . . . , en are linearly independent if f is injective (the unique null linear
combination, λ1 e1 + . . . + λn en = 0, is the trivial one, λ1 = . . . = λn = 0).
In such a case n ≤ dim E, and they form a base of E when the equality holds.
Proposition: Any generating system e1 , . . . , en of E contains a base of E.
Proof: Put Ei = ke1 + . . . + kei . Eliminating ei when Ei−1 = Ei , we may assume that we have
0 ⊂ E1 ⊂ . . . ⊂ En = E, so that n ≤ dim E. Hence n = dim E, and e1 , . . . , en is a base of E.
Proposition: In a finite dimensional vector space E, any linearly independent family e1 , . . . , en

may be extended to a base of E.
Proof: Put Ei = ke1 + . . . + kei . The inclusions Ei−1 ⊂ Ei are strict because dim Ei = i, and we
may fix en+1 , . . . , en+r ∈ E so that 0 ⊂ E1 ⊂ . . . ⊂ En+r = E, with En+j = ke1 + . . . + ken+j .
Hence n + r ≤ dim E, and the generating system e1 , . . . , en , . . . , en+r is a base of E.
Theorem: Any vector space E 6= 0 of finite dimension n has a base, E ' k n . Hence, two finite
dimensional k-vector spaces are isomorphic if and only if both have equal dimension.
Proof: Any non null vector of E may be extended to a base of E.

Rouché-Frobënius Theorem: A system of linear equations AX = B is consistent if and only

if rk A = rk (A|B).
Proof: Let A1 , . . . , An ∈ k m be the columns of A. The system x1 A1 +. . .+xn An = B is consistent

if and only if B ∈ hA1 , . . . , An i; i.e., hA1 , . . . , An i = hA1 , . . . , An , Bi.
This is equivalent to the coincidence of the dimensions (p. 49).
Now, the rank of a matrix (the maximum number of linearly independent columns) is just
the dimension of the vector subspace of k m spanned by the columns.
Proposition: dim (Homk (E, F )) = (dim E)(dim F ).

In fact, if e1 , . . . , en is a base of E, then we have an isomorphism
∼
Homk (E, F ) −−→ F × . n. . ×F, f 7→ (f (e1 ), . . . , f (en )),
Proof: If f (e1 ) = . . . = f (en ) = 0, then f = 0 because
f (E) = f (ke1 + . . . + ken ) = k · f (e1 ) + . . . + k · f (en ) = 0.
Moreover, the map f (x1 e1 + . . . + xn en ) = x1 v1 + . . . + xn vn is linear, and f (ei ) = vi .

Now dim (Homk (E, F )) = dim F n = n(dim F ).
Definition: Let f : E → F be a linear map. If (e1 , . . . , en ) is a base of E, and (v1 , . . . , vm ) is a

base of F , then we have
f (ej ) = a1j v1 + . . . + amj vm .
and A = (aij ) is the matrix of f in such bases. Since the j-th column of A are the coordinates
of f (ej ); the rank of A is just the dimension of hf (e1 ), . . . , f (en )i = Im f ,
dim (Im f ) = rk A,
dim (Ker f ) = n − rk A.
P
If we write the coordinates of a vector e = j xj ej ∈ E as a column X = (xj ), then
0 aij xj e0i ,
P P P P P
f (e) = j xj f (ej ) = j i xj aij ei = i j
so that the coordinates Y of f (e) in the base of F are just
Y = AX. (2.1)
If h : F → W is linear, (w1 , . . . , wr ) is a base of W , and B = (bki ) is the matrix of h, then

P
h(vi ) = k bki wk
P P P P P
(hf )(ej ) = h i aij vi = i aij h(vi ) = i,j aij bki wk = k i bki aij wk .
P
The matrix of h ◦ f in the bases e1 , . . . , en and w1 , . . . , wr is BA := i bki aij .
The matrix product is associative (so is the composition of maps) and the matrix of any
isomorphism f is invertible, the inverse being the matrix of f −1 .
Definition: Let (e01 , . . . , e0n ) be a new base of E. The base change matrix is the matrix
B = (bij ) ∈ Mn×n (k) of the identity map E → E in the bases (e01 , . . . , e0n ) and (e1 , . . . , en ),
e0j = b1j e1 + . . . + bnj en .

By 2.1, if X and X 0 are the coordinates of e ∈ E in these bases,
X = BX 0 , X 0 = B −1 X.
If we also fix a new base (v10 , . . . , vm

0 ) in F , we have a base change matrix C ∈ M
m×m (k),
and the commutative square
f
e01 , . . . , e0n F −−−→ F v10 , . . . , vm
0
Id ↓ ↑ Id
f
e1 , . . . , en E −−−→ E v1 , . . . , vm
shows that the matrix A0 ∈ Mm×n (k) of f in the new bases is just
A0 = C −1 AB. (2.2)
Definition: The sum of some vector subspaces V1 , . . . , Vr of a vector space is direct, and we
put V1 ⊕ . . . ⊕ Vr , if the always surjective linear map
s : V1 × . . . × Vr −→ V1 + . . . + Vr , s(v1 , . . . , vr ) = v1 + . . . + vr ,
is an isomorphism, so that dim (V1 ⊕ . . . ⊕ Vr ) = dim V1 + . . . + dim Vr .
Examples: If e1 , . . . , en is a base of E, then E = ke1 ⊕ . . . ⊕ ken .

If r = 2, then Ker s = {(v, −v) : v ∈ V1 ∩ V2 }, and the sum is direct ⇔ V1 ∩ V2 = 0.
If E = V ⊕ W , (V + W = E, V ∩ W = 0) we say that W is a supplement of V in E.
Theorem: Any vector subspace V ⊆ E admits a supplement in E.
Proof: Let X be the set of all vector subspaces W ⊆ E such that V ∩ W = 0, ordered by
inclusion. It is not empty because V ∩ 0 = 0. S
Any chainS {Wi } hasS an upper bound: theScrux is to show that W = i Wi is a subgroup,
since λW = i λWi ⊆ i Wi = W , V ∩ W = i (V ∩ Wi ) = 0 and Wi ⊆ W .
Now, if u, w ∈ W , then u, w ∈ Wi by the chain condition; hence u + w, −u ∈ Wi ⊆ W .
By Zorn’s lemma, X has a maximal element W , and V ∩ W = 0 since W ∈ X.
Let us show that E = V + W . If e ∈ E, and e ∈/ W , then W ⊂ W + ke, so that V ∩ (W + ke)
is not zero. There is a vector 0 6= v = w + λe, where v ∈ V , w ∈ W .
Now λ 6= 0, because V ∩ W = 0, and we see that e = λ−1 (v − w) ∈ V + W .
i p
Theorem: Any exact sequence 0 −→ E 0 −−→ E −−→ E 00 −→ 0 of linear maps admits a linear
section (a map s : E 00 → E such that ps = IdE 00 ). Hence i + s : E 0 ⊕ E 00 → E is an isomorphism
and the following diagram commutes, where i1 (e0 ) = (e0 , 0), π2 (e0 , e00 ) = e00 ,
i π
0 −→ E 0 −−1→ E 0 ⊕ E 00 −−2→ E 00 −→ 0
k o|i + s k
i p
0 −→ E 0 −→ E −−→ E 00 −→ 0
Proof: Let W be a supplement of Im i = Ker p in E, so that p : W − → E 00 .

∼
−1 00
The required map is s = p : E → W ⊆ E, and the diagram clearly commutes.
Let us see that i + s : E 0 ⊕ E 00 → E is an isomorphism.
If 0 = i(e0 ) + s(e00 ), then 0 = pi(e0 ) + ps(e00 ) = e00 , i(e0 ) = 0; hence e0 = 0 since i is injective.
If e ∈ E, we put e00 = p(e), so that e − s(e00 ) ∈ Ker p = Im i; hence e − s(e00 ) = i(e0 ), and we
conclude that e = i(e0 ) + s(e00 ).
2.4. THE DUAL SPACE 53
2.4 The Dual Space

Definitions: Linear forms on a k-vector space E are k-linear maps ω : E → k.
The dual space of E is E ∗ = Homk (E, k).
Theorem: If e1 , . . . , en is a base of E, there is a unique base ω1 , . . . , ωn of E ∗ , named dual

base, such that ωi (ej ) = δij , (where δij = 1 when i = j, and 0 otherwise).
Proof: The map ωi (x1 e1 + . . . + xn en ) = xi is a linear form and ωi (ej ) = δij .

The linear forms ω1 , . . . , ωn span E ∗ because, for any linear form ω, we have
ω = ω(e1 )ω1 + . . . + ω(en )ωn (2.3)
since both coincide on the base e1 , . . . , en . Hence ω1 , . . . , ωn form a base of E ∗ .
Note: 2.3 states that the coordinates of ω in the dual base are Z = (ω(e1 ), . . . , ω(en )), which is
the matrix of ω : E → k when we consider in k the base defined by the unity.
If (e01 , . . . , e0n ) is another base of E, and (ω10 , . . . , ωn0 ) is the dual base, the coordinates Z 0 of
ω in the new base are Z 0 = ZB by 2.2, where B is the base change matrix, and Z, Z 0 are rows;
hence Z t = (B −1 )t (Z 0 )t and the base change matrix in E ∗ is just (B −1 )t .
Lemma: If e ∈ E is non null, there exists ω ∈ E ∗ such that ω(e) = 1.
Proof: Let V be a supplement of ke. The linear form ω : E = ke ⊕ V → k, ω(λe + v) = λ, is

well defined (since e 6= 0) and ω(e) = 1. q.e.d.
Any vector e ∈ E defines a linear form ψ(e) : E ∗ −→ k, ψ(e)(ω) = ω(e); so that we have a
natural linear map ψ : E → E ∗∗ .
Reflexivity Theorem: E = E ∗∗ , for any finite dimensional vector space E.
Proof: Since dim E ∗∗ = dim E ∗ = dim E, we only have to show that ψ is injective.
If ψ(e) = 0, then ω(e) = ψ(e)(ω) = 0, ∀ω ∈ E ∗ , and e = 0 by the above lemma.
Definition: The incident of a vector subspace V ⊆ E is the vector subspace of E ∗
V o = {ω ∈ E ∗ : ω(v) = 0, for all v ∈ V }.
Lemma: If E is finite dimensional, then V = (V o )o via the identification E = E ∗∗ .
Proof: Clearly V ⊆ (V o )o = {e ∈ E : ω(e) = 0, for all ω ∈ V o }.

Conversely, if e ∈ (V o )o , and π : E → E/V is the canonical map, we have 0 = ω̄(πe) for all
ω̄ ∈ (E/V )∗ , since ω̄ ◦ π ∈ V o . Hence π(e) = 0, and e ∈ V .
Theorem: If E is finite dimensional, we have a natural lattice anti-isomorphism

Vector sub- Vector sub-
∼
−−→ , V 7→ V o .
spaces of E spaces of E ∗
Proof: It is a bijective map according to the lemma, the inverse being the incidence in E ∗ .
Moreover, if V1o ⊆ V2o , then (V1o )o ⊇ (V2o )o , and V1 ⊇ V2 by the lemma.
Corollary: (V + W )o = V o ∩ W o .
(V ∩ W )o = V o + W o .
dim V o = dim E − dim V .
Proof: These properties are lattice properties.
Example: The equations a1 x1 + . . . + an xn = 0 of the hyperplanes passing through V form the

incident subspace V o . If L = kω1 + . . . + kωm is a vector subspace of E ∗ , then Lo is the space
of solutions of the homogeneous system

ω1 = a11 x1 + . . . . + a1n xn = 0 
.....................
ωm = am1 x1 + . . . + amn xn = 0

Definition: The transpose of a linear map f : F → E is the linear map

f ∗ : E ∗ −→ F ∗ , f ∗ (ω) = ω ◦ f.
It is clear that (g ◦ f )∗ = f ∗ ◦ g ∗ , and (λf1 + µf2 )∗ = λf1∗ + µf2∗ .
If i : V → E is the inclusion, i(v) = v, then i∗ (ω) is the restriction of ω to V .
If (aij ) is the matrix of f in some bases v1 , . . . , vm of F and e1 , . . . , en of E, then the matrix
of f ∗ in the dual bases θ1 , . . . , θm of F ∗ and ω1 , . . . , ωn of E ∗ , is the transpose matrix
(f ∗ ωj )(vi ) = ωj (f (vi )) = aji .
j p
Frobënius Theorem: If 0 −→ E1 −−→ E −−→ E2 −→ 0 is an exact sequence of linear maps,
so is the sequence of transpose maps:
p∗ j∗
0 −→ E2∗ −−→ E ∗ −−→ E1∗ −→ 0.
Proof: It is enough (p. 52) to consider the exact sequence
i π
0 −→ E1 −−1→ E1 ⊕ E2 −−2→ E2 −→ 0
We have that i∗1 is surjective, π2∗ is injective, and Im π2∗ ⊆ Ker i∗1 because
i∗1 π1∗ = (π1 i1 )∗ = Id,
i∗2 π2∗ = (π2 i2 )∗ = Id,
i∗1 π2∗ = (π2 i1 )∗ = 0.
Finally, if ω ∈ Ker i∗1 , then ω = (i1 π1 + i2 π2 )∗ ω = (π1∗ i∗1 + π2∗ i∗2 )ω = π2∗ (i∗2 ω) ∈ Im π2∗ .
Corollary: If f : E → F is a linear map, then Ker f ∗ = (Im f )o , Im f ∗ = (Ker f )o .

i f π
Proof: The exact sequence 0 → Ker f →
− E−
→F −
→ F/Im f → 0 induces an exact sequence
π∗ f∗ i∗
0 −→ (F/Im f )∗ −−−→ F ∗ −−−→ E ∗ −−→ (Ker f )∗ −→ 0.
Hence Im f ∗ = Ker i∗ = (Ker f )o and Ker f ∗ = Im π ∗ .
Finally, Im π ∗ = (Im f )o by the universal property of the quotient space F/Im f .
Corollary: Im f ∗ = (Im f )∗ , (column and row ranks of any matrix coincide).
Proof: By the universal property of the quotient space, (E/Ker f )∗ ' (Ker f )o = Im (f ∗ ), and
by the Isomorphism theorem, (E/Ker f )∗ ' (Im f )∗ .
π∗ i∗
Note: The exact sequence 0 → (E/V )∗ −→ E ∗ − → V ∗ → 0, where V o = Ker i∗ , proves again
that dim V = dim E − dim V ; hence that V = (V o )o .
o
2.5. EUCLIDEAN VECTOR SPACES 55
2.5 Euclidean Vector Spaces

In this section we put K = R or C, and a group morphisms f : E → E 0 between K-vector spaces
is said to be semilinear when f (λe) = λ̄f (e), λ ∈ K.
When K = R semilinear maps are just linear maps and, when K = C, semilinear maps are
just linear maps if E 0 is endowed with the conjugate complex structure λ · e0 := λ̄e0 .
.
Definition: A scalar product on a K-vector space E is a map E × E −→ K which is
1. Left-semilinear and right-linear :
(λe + µe0 ) · v = λ̄(e · v) + µ̄(e0 · v) , e · (λv + µv 0 ) = λ(e · v) + µ(e · v 0 ).
2. v · e = e · v.
3. Positive definite: e · e ≥ 0, and e = 0 if the equality holds.

Two vectors e, v are orthogonal when e · v = 0, and √ the modulus of a vector is the non-
√
negative real number kek = + e · e. We have kλek = λ2 e · e = |λ|·kek, and the Pythagorean
theorem holds: If e · v = 0, then ke + vk2 = kek2 + e · v + v · e + kvk2 = kek2 + kvk2 .
Lemma: |e · v| ≤ kek · kvk, (Cauchy-Schwarz inequality).

ke + vk ≤ kek + kvk, (triangle inequality).
e
Proof: When e 6= 0, put u = kek , so that u · u = 1 and e = αu, kek = |α|.
There is a scalar β such that v = βu + w and u · w = 0; just take β = u · v.
Hence e · v = ᾱβ(u · u) = ᾱβ,and we conclude by the Pythagorean theorem:
p
kek · kvk = |α| |β|2 + kwk2 ≥ |α| · |β| = |ᾱβ| = |e · v|.
Finally, the triangle inequality follows from the Cauchy-Schwarz inequality:
ke + vk2 = (e + v) · (e + v) = e · e + v · v + e · v + e · v ≤
≤ e · e + v · v + 2|e · v| ≤ kek2 + kvk2 + 2kek · kvk = (kek + kvk)2 .
Definition: In the real case, the (measure in radians of the) angle formed by two non-null
e·v
vectors e, v is defined to be (p. 38) the unique real number α ∈ [0, π] such that cos α = kek·kvk ,
so that e · v = 0 means that α = π/2.
Definition: A Euclidean vector space is a finite dimensional real or complex vector space E
endowed with a scalar product4 , so that we have a semilinear map (named polarity)
φ : E −→ E ∗ , (φe)(v) = e · v.
Theorem: The polarity φ : E → E ∗ is a semilinear isomorphism.
Proof: If φ(e) = 0, then e · e = (φe)(e) = 0; hence e = 0, and φ is injective.

Since dim E = dim E ∗ , we conclude that φ is an isomorphism.
Definitions: The orthogonal of a vector subspace V ⊆ E is the vector subspace
V ⊥ = {e ∈ E : e · v = 0 for all v ∈ V } = {e ∈ E : φ(e) ∈ V o }.

4
In fact, in the classical Euclidean Geometry the scalar product is well-defined up to a positive factor, and to
fix it is just to fix the length unit.
Corollary: We have a lattice anti-automorphism

Vector sub- Vector sub-
∼
−−→ , V 7→ V ⊥ .
spaces of E spaces of E
so that (V + W )⊥ = V ⊥ ∩ W ⊥ , (V ∩ W )⊥ = V ⊥ + W ⊥ and dim V ⊥ = dim E − dim V .
Proof: It follows from the properties of the incidence, since φ(V ⊥ ) = V o .
Corollary: (V ⊥ )⊥ = V .
E =V⊥⊕V.
Proof: We have V = (V ⊥ )⊥ because V ⊆ (V ⊥ )⊥ , and
dim (V ⊥ )⊥ = dim E − dim V ⊥ = dim E − (dim E − dim V ) = dim V.
Finally, V ∩ V ⊥ = 0 since v · v = 0 ⇒ v = 0, and V + V ⊥ = E because
dim (V + V ⊥ ) = dim V + dim V ⊥ = dim E.
Examples: In the 3 following examples, we consider a real Euclidean space:

a+b+c
1. In any triangle, the three medians intersect at the barycenter g = 3 , mutually dividing
at the ratio 2:1, as the following identities show:
a+b+c a 2b+c b 2a+c c 2a+b
= + = + = + ·
3 3 3 2 3 3 2 3 3 2
2. Moreover, if h is the intersection point of the heights trough the vertices a and c, then
(b − a) · (h − c) = 0 (2.4)
(c − b) · (h − a) = 0 (2.5)
and adding them we obtain (c − a) · (h − b) = 0, so that the third height also passes through
the orthocenter h. On the other hand, if f is the intersection point of the bisectors of the
sides ab and bc, then
0 = 2(b − a) · f − a+b

2 = (b − a) · 2f + a · a − b · b (2.6)
0 = 2(c − b) · f − b+c

2 = (c − b) · 2f + b · b − c · c (2.7)
and adding them we obtain 0 = (c − a) · 2f + a · a − c · c = 2(c − a) · f − a+c

2 , so that the
bisector of the side ab also passes through the circumcenter f .
a+b+c
3. Now, adding 2.4 with 2.6, and 2.5 with 2.7, we obtain (recall that g = 3 )
(b − a) · (h + 2f − 3g) = 0,
(c − b) · (h + 2f − 3g) = 0.
Since the vectors b − a, c − b are linearly independent (because a, b, c are not collinear) and
the polarity is an isomorphism, we see that h + 2f − 3g = 0. The barycenter divides the
segment determined by the orthocenter and the circumcenter in the ratio 2:1,
h 2f 2
g= + = h + (f − h).
3 3 3
2.6. DIAGONALIZATION OF ENDOMORPHISMS 57
Definition: A base e1 , . . . , en of E is orthonormal when ei · ej = δij , so that
(x1 e1 + . . . + xn en ) · (y1 e1 + . . . + yn en ) = x̄1 y1 + . . . + x̄n yn .
Theorem: Any Euclidean vector space E 6= 0 admits orthonormal bases.

v
Proof: If E = Kv, an orthonormal base is e = kvk , since e · e = 1.
If n = dim E > 1, we take en ∈ E of module 1. Now dim (Ken )⊥ = n − 1, and
E = (Ken )⊥ ⊕ Ken .
By induction (Ken )⊥ admits an orthonormal base e1 , . . . , en−1 .

Now e1 , . . . , en span E and ei · ej = δij , hence they form an orthonormal base of E.
2.6 Diagonalization of Endomorphisms

Definition: Let E be a k-vector space of dimension n. The endomorphisms of E are the
linear maps T : E → E, and Endk (E) = Homk (E, E) is a vector space of dimension n2 .
The product of endomorphisms is the composition, ST = S ◦ T .
In general, given a polynomial p(x) = a0 + a1 x + a2 x2 + . . . + ad xd ∈ k[x], we put
p(T ) = a0 IdE + a1 T + a2 T 2 + . . . + ad T d .
Once we fix a base e1 , . . . , en of E, any endomorphism T has a n × n matrix A = (aij ),

n
P
T (ej ) = aij ei ; j = 1, . . . , n,
i=1
and the matrix A0 of T in another base of E is (p. 52)
A0 = B −1 AB. (2.8)
where B is the base change matrix. Hence, if I is the unit n × n matrix, using properties
of the determinants (that we shall prove in the next section) we may see that the polynomial
cT (x) = |xI − A| is intrinsic: it does not depend on the base
|xI − A0 | = |xB −1 IB − B −1 AB| = |B −1 (xI − A)B|

= |B −1 | · |xI − A| · |B| = |B|−1 |B| · |xI − A| = |xI − A|
Definitions: The characteristic polynomial cT (x) of the endomorphism T is

x − a11 −a12 . . . −a1n

−a21 x − a22 . . . −a2n Pn
n
cT (x) = = x − a ii xn−1 + . . . + (−1)n |A|.
... ... ... . . . i=1
−an1 −an2 . . . x − ann
A scalar α ∈ k is an eigenvalue of T if there exists a non null vector e ∈ E such that

T (e) = αe. We say that e is an eigenvector of T of eigenvalue α, and that Vα := Ker (αId−T ) =
{e ∈ E : T (e) = αe} is an eigenspace of T .
Theorem: The roots in k of the characteristic polynomial are just the eigenvalues of T .
Proof: A scalar α is an eigenvalue of T if and only if
0 6= dim (Ker (αId − T )) = n − rk (αI − A);
or equivalently rk (αI − A) < n; that is to say, cT (α) = |αI − A| = 0.
Corollary: The number of eigenvalues of T always is ≤ dim E.
Proof: The number of roots of cT (x) is bounded by deg cT (x) = n = dim E.
Corollary: When k = C, any endomorphism admits an eigenvector.
Definition: An endomorphism T of an Euclidean vector space E is selfadjoint or hermitian

(or symmetric in the real case) when T (e) · v = e · T (V ), ∀e, v ∈ E.
If A is the matrix of T in an orthonormal base of E, this condition states that A is an
hermitian matrix, Āt = A. Hence, in the real case, that A is symmetric, At = A.
Spectral Theorem: If T : E → E is a selfadjoint endomorphism, then all the roots of the

characteristic polynomial cT (x) are real, any two eigenvectors of T with different eigenvalues
are orthogonal, and E admits an orthonormal base of eigenvectors of T .
Proof: If 0 6= e ∈ E is an eigenvector, T (e) = αe, then e · e 6= 0 and α is real:
ᾱ(e · e) = (αe) · e = T (e) · e = e · T (e) = e · (αe) = α(e · e).
In the complex case, we see that any root of cT (x) = |xI − A| is real; hence also in the real case,
since any real symmetric matrix A is a complex hermitian matrix.
Now, if eα , eβ are eigenvector with eigenvalues α 6= β, then ᾱ = α, and vα · vβ = 0 because
α(vα · vβ ) = ᾱ(vα · vβ ) = (αvα ) · vβ = T (vα ) · vβ = vα · T (vβ ) = vα · (βvβ ) = β(vα · vβ ).
Finally, fix some eigenvector e and, dividing e by kek, we may assume that kek = 1.
The orthogonal subspace V := hei⊥ has dimension n − 1 and T (V ) ⊆ V : If v ∈ V , then
T (v) · e = v · T (e) = v · (αe) = α(v · e) = 0, so that T (v) ∈ V .
By induction on the dimension of E, we may assume that V admits an orthonormal base
e2 , . . . , en of eigenvectors of the selfadjoint endomorphism T |V : V → V , hence of T .
Now e, e2 , . . . , en is an orthonormal base of E, of eigenvectors of T .
Example: In Quantum Mechanics, the state space of a system is a complex vector space E
(tipically of infinite dimension) with a scalar product h | i, the states being the unidimensional
vector subspaces (normally represented by a vector of norm 1), and any observable magnitude
is described by an hermitian endomorphism T : E → E. The possible values of the magnitude
are just the eigenvalues of T (hence real numbers). If the state of the system is represented by
an eigenvector ψ of eigenvalue α, then the measurement of such magnitude is certainly α; but
if the state is represented by a sum ψ = ψ1 + . . . + ψr of eigenvectors with different eigenvalues
α1 , . . . , αr (and such decomposition is unique because the sum Vα1 + . . . + Vαr always is a direct
sum, see p. 79, and it exists at least when E is finite dimensional) then the possible values of
the measurement are α1 , . . . , αr , with probability
kψi k2 hψi |ψi i

P(αi ) = 2
=
kψk hψ|ψi
2.6. DIAGONALIZATION OF ENDOMORPHISMS 59
and, the states ψ1 , . . . , ψr being mutually orthogonal, by the Pythagorean theorem we have that
kψk2 = kψ1 k2 + . . . + kψr k2 , P(α1 ) + . . . + P(αr ) = 1.
In particular, the mean value of the measurements in the state Cψ is just
P P
P h i ψi | i αi ψi i hψ|T (ψ)i
i αi P(αi ) = = ·
hψ|ψi hψ|ψi
Hamilton-Cayley Theorem: The characteristic polinomial c(x) = xn + . . . + c1 x + c0 anni-

hilates the endomorphism, c(T ) = T n + . . . + c1 T + c0 Id = 0.
Proof: It is obvious when n = 1, since T = aId, and c(x) = x − a.

If n > 1, then the matrix of c(T ) is c(A) = An + . . . + c0 I and, to prove that c(A) = 0, we
may consider an extension of k.
By Kronecker’s theorem, we may assume that c(x) has some root α ∈ k.
Let e1 , . . . , en be a base of E, where T (e1 ) = αe1 ,
r
α ... rα ...
A= , A = .
0 Ā 0 Ār
Now c(x) = (x − α)c̄(x), where c̄(x) = |xI − Ā| and, by induction, c̄(Ā) = 0.

0 ... c̄(α) . . . 0 ... c̄(α) . . .
c(A) = (A − αI)c̄(A) = = = 0.
0 B 0 c̄(Ā) 0 B 0 0
Definition: The linear map k[x] → Endk (E), p(x) 7→ p(T ), is not injective, since dim k[x] = ∞,
and the annihilator polynomial of T is the unitary generator φT (x) = xd + . . . of the ideal
{p(x) ∈ k[x] : p(T ) = 0}. In k[x], it is the polynomial of least degree annihilating T , and it
divides any other polynomial in k[x] annihilating T .
Theorem: The roots in k of the annihilator polynomial φT (x) are just the eigenvalues of T .
Proof: If e 6= 0 and T (e) = αe, then 0 = φ(T )e = φ(α)e; hence φ(α) = 0.

Conversely, if φ(x) = (x − α)p(x), then p(T ) 6= 0, and p(T )v 6= 0 for some vector v.

Hence (T − α) p(T )v = φ(T )v = 0, and α is an eigenvalue.
Examples: The symmetry S of R3 with respect to a plane has the eigenvalues x = 1, −1. Since
S 2 = Id, the annihilator divides x2 − 1; hence φS (x) = x2 − 1.
A rotation T of right angle in R3 only has the eigenvalue x = 1. Since T 4 = Id, the annihilator
divides x4 − 1 = (x − 1)(x + 1)(x2 + 1); hence φT (x) = (x − 1)(x2 + 1).
Definition: An endomorphism T : E → E is diagonalizable if E admits a base e1 , . . . , en of

eigenvectors, T (ej ) = αj ej , so that the matrix of T is diagonal,
 
α1 0 0
 
 0 
 
 
 0 
 
0 0 αn
Examples: By 2.8, an endomorphism T of matrix A is diagonalizable if and only if there is an

invertible matrix B such that D = B −1 AB is diagonal. In such a case
 
α1 0 0
 
 0 
Am = (BDB −1 )(BDB −1 ) . . . = BDm B −1 = B 
 −1
B

 0 
 
0 0 m
αn
and so we obtain the general solution Xm = Am X0 of the system of difference equations


xm+1 = a11 xm + . . . + a1n zm

Xm+1 = AXm , ........................

zm+1 = an1 xm + . . . + ann zm

When k = R or C, the general solution of the system of differential equations


0
x1 = a11 x1 + . . . + a1n xn

X 0 = AX , .....................

 0
xn = an1 x1 + . . . + ann xn
is X = B X̄, where X̄ is the general solution of the system X̄ 0 = DX̄. In fact,
X 0 = B X̄ 0 = BDX̄ = BDB −1 X = AX.
Now, the differential equations x̄0i = αi x̄i of the system X̄ 0 = DX̄ are solved in p. 81, so
that x̄i (t) = ci eαi t , ci ∈ k (= R or C), and the solutions of the initial system are
c1 eα1 t
   
x1
 ..   . 
 .  = B  ..  .
xn cn eαn t
Lemma: An endomorphism of eigenvalues α1 , . . . , αr is diagonalizable if and only if
Vα1 + . . . + Vαr = E.
Proof: If T is diagonalizable, then Vα1 + . . . + Vαr is E, since it contains a base of E.

If Vα1 + . . . + Vαr = E, considering bases in Vαi we see that E admits a generating system
of eigenvectors, hence a base.
Lemma: dim (Vα1 + . . . + Vαm ) = dim Vα1 + . . . + dim Vαm , ( αi 6= αj ).
Proof: Ker (T − α1 ) . . . (T − αm ) = Ker (T − α1 ) ⊕ . . . ⊕ Ker (T − αm ), by p. 79.
Corollary: If an endomorphism of a vector space of dimension n has n different eigenvalues,

it is diagonalizable.
Proof: dim (Vα1 + . . . + Vαn ) = dim Vα1 + . . . + dim Vαn ≥ n = dim E.
Diagonalization Criterion: An endomorphism T is diagonalizable if and only if all the roots

of the annihilator polynomial φT (x) are in k and are simple roots.
2.7. TENSORS 61
Proof: If all the roots of φT (x) are in k and are simple, φT (x) = (x − α1 ) . . . (x − αr ), then we
have 0 = (T − α1 ) . . . (T − αr ) and T is diagonalizable (p. 79),
E = Ker (T − α1 ) ⊕ . . . ⊕ Ker (T − αr ) = Vα1 ⊕ . . . ⊕ Vαr .
Conversely, if α1 , . . . , αr are the eigenvalues of T , then (T − α1 ) . . . (T − αr ) vanishes on a

base when T is diagonalizable; hence it is null and φT (x) divides (x − α1 ) . . . (x − αr ), so that
any root of φT (x) is simple and it is in k.
Diagonalization Criterion: An endomorphism T is diagonalizable if and only if all the roots

of the characteristic polynomial cT (x) are in k and the multiplicity mi of any root αi is
mi = dim Vαi .
Proof: If T is diagonalizable, in some base the matrix is diagonal

 
α1 0 0
 
 0 
D=
 

 0 
 
0 0 αn
and any root of cT (x) = |xI − D| = (x − α1 ) . . . (x − αn ) is in k.

The multiplicity mi of any root αi is the number of repetitions in α1 , . . . , αn ; hence

mi = n − rk (D − αi I) = dim Ker (T − αi Id) = dim Vαi .
P
Conversely, if all roots of cT (x) are P i mi = deg cT (x) = dim
P in k, then PE.
Now, if mi = dim Vαi , then dim V
i αi = i dim Vαi = dim E, and i Vαi = E.
2.7 Tensors
Definition: Let E1 , . . . , Er , F be k-vector spaces. A map T : E1 ×. . .×Er → F is k-multilinear
if it is k-linear in any variable,
T (. . . , ei + vi , . . .) = T (. . . , ei , . . .) + T (. . . , vi , . . .),
T (. . . , λei , . . .) = λT (. . . , ei , . . .).
Multilinear maps E1 × . . . × Er → F form a k-vector space:
(T + T̄ )(e1 , . . . , er ) = T (e1 , . . . , er ) + T̄ (e1 , . . . , er ),

(λT )(e1 , . . . , er ) = λ · T (e1 , . . . , er ).
Definitions: Tensors of type (p, q) on a k-vector space E of finite dimension are multilinear
maps T : E× . p. . ×E × E ∗ × . q. . ×E ∗ −→ k, and they form a vector space Tpq E. Tensors of type
(p, 0) are covariant5 of order p, and (0, q)-tensors are contravariant of order q.
In particular, T10 E = E ∗ , and T01 E = E ∗∗ = E, and we agree that T00 E = k.
Any linear map f : F → E naturally induces linear maps
f ∗ : Tp0 E −→ Tp0 F , (f ∗ T )(v1 , . . . , vp ) = T f (v1 ), . . . , f (vp ) ,

f∗ : T0q F −→ T0q E , (f∗ T )(ω1 , . . . , ωq ) = T f ∗ (ω1 ), . . . , f ∗ (ωq ) .

5
They have p covariant indices or variables.
If g : E → V is also linear, then (g ◦ f )∗ = f ∗ ◦ g ∗ and (g ◦ f )∗ = g∗ ◦ f∗ .
Definition: The tensor product of a (p, q)-tensor T with a (p0 , q 0 )-tensor T 0 is the following
(p + p0 , q + q 0 )-tensor
(T ⊗T 0 )(e1 , . . . , ep+p0 , ω1 , . . . , ωq+q0 ) = T (e1 , . . . , ep , ω1 , . . . , ωq )·T 0 (ep+1 , . . . ep+p0 , ωq+1 , . . . ωq+q0 ).
1. (λT + µT̄ ) ⊗ T 0 = λ(T ⊗ T 0 ) + µ(T̄ ⊗ T 0 ).

T ⊗ (λT 0 + µT̄ 0 ) = λ(T ⊗ T 0 ) + µ(T ⊗ T̄ 0 ).
2. (T ⊗ T 0 ) ⊗ T 00 = T ⊗ (T 0 ⊗ T 00 ).
3. f ∗ (T ⊗ T 0 ) = f ∗ (T ) ⊗ f ∗ (T 0 ).
Proof: Put e = (e1 , . . . , ep ), e0 = (ep+1 , . . . , ep+p0 ), ω = (ω1 , . . . , ωq ), ω 0 = (ωq+1 , . . . , ωq+q0 ).
(λT + µT̄ ) ⊗ T 0 (e, e0 , ω, ω 0 ) = (λT + µT̄ )(e, ω)T 0 (e0 , ω 0 )

= λT (e, ω)T 0 (e0 , ω 0 ) + µT̄ (e, ω)T 0 (e0 , ω 0 ) = λ(T ⊗ T 0 ) + µ(T̄ ⊗ T 0 ) (e, e0 , ω, ω 0 )

and analogously T ⊗ (λT 0 + µT̄ 0 ) = λ(T ⊗ T 0 ) + µ(T ⊗ T̄ 0 ).

The second property is obvious, and finally we prove the third:
(f ∗ (T ⊗ T 0 ))(e, e0 ) = (T ⊗ T 0 )(f (e), f (e0 )) = T (f (e)) · T 0 (f (e0 ))

= (f ∗ T )(e) · (f ∗ T 0 )(e0 ) = (f ∗ (T ) ⊗ f ∗ (T 0 ))(e, e0 ).
Note: We also put T q E = T0q E and T p E ∗ = Tp0 E. The

L tensor product defines a structure of (non
•E = q
commutative) k-algebra on the vector space T• p,q Tp E of all finite formal sums L
of tensors
P q ∗ is a morphism of k-algebras, only defined on the subalgebra T • E ∗ = p ∗
T
p,q p , and f pT E .
Theorem: dim Tpq E = (dim E)p+q . In fact, if e1 , . . . , en is a base of E, and ω1 , . . . , ωn is the

dual base, then a base of Tpq E is defined by the tensors
ωi1 ⊗ . . . ⊗ ωip ⊗ ej1 ⊗ . . . ⊗ ejq .
Proof: Let us fix a sequence i1 , . . . , ip , j1 , . . . , jq . Since ωi (ej ) = δij ,
(ωi1 ⊗ . . . ⊗ ωip ⊗ ej1 ⊗ . . . ⊗ ejq )(ei1 , . . . , eip , ωj1 , . . . , ωjq ) = 1,
while the remaining tensors in the considered family vanish on (ei1 , . . . , eip , ωj1 , . . . , ωjq ).
Hence ωi1 ⊗ . . . ⊗ ωip ⊗ ej1 ⊗ . . . ⊗ ejq is not a linear combination of the remaining tensors:
it is a family of linearly independent tensors.
To show that this family spans Tpq E, we prove that any (p, q)-tensor T is
P
T = T (ei1 , . . . , eip , ωj1 , . . . , ωjq ) ωi1 ⊗ . . . ⊗ ωip ⊗ ej1 ⊗ . . . ⊗ ejq .
1≤i1 ,...,ip ≤n
1≤j1 ,...,jq ≤n
In fact, the difference vanishes on any sequence (ei1 , . . . , eip , ωj1 , . . . , ωjq ); hence it is null.
j ...j
The coordinates of a (p, q)-tensor T are just λi11...ipq = T (ei1 , . . . , eip , ωj1 , . . . , ωjq ).
Example: Any endomorphism T defines a (1, 1)-tensor T (e, ω) = ω(T e), and this linear map
Endk (E) → T11 E is injective: if ω(T e) = 0, ∀ω ∈ E ∗ , then T (e) = 0 (p. 53).
Since both vector spaces have dimension (dim E)2 , it is an isomorphism, T11 E = Endk (E).
2.7. TENSORS 63
P
If (aij ) is the matrix of T , i.e., T (ej ) = i aij ei , then the coordinates of the corresponding
tensor T = ij λji ωi ⊗ ej are λji = T (ei , ωj ) = ωj (T ei ) = aji .
P
Universal Property: For any multilinear map T : E ∗ × . p. . ×E ∗ ×E× . q. . ×E → F there exists

a unique linear map f : Tpq E → F such that
f (ω1 ⊗ . . . ⊗ ωp ⊗ e1 ⊗ . . . ⊗ eq ) = T (ω1 , . . . , ωp , e1 , . . . , eq ). (2.9)
Proof: We fix a base of E, and we define f : Tpq E → F in the corresponding base of Tpq E by 2.9.
Now both terms in 2.9 are multilinear maps coinciding on the sequences of vectors of a base;
hence they coincide. Uniqueness is obvious.
Corollary: We have canonical isomorphisms (Tpq E)∗ = Tqp E.
Proof: Put F = k in the universal property.
Index Contraction: There exists a unique linear map C11 : Tpq E → Tp−1
q−1
E such that
C11 (ω1 ⊗ ω2 ⊗ . . . ⊗ ωp ⊗ e1 ⊗ e2 ⊗ . . . ⊗ eq ) = ω1 (e1 ) ω2 ⊗ . . . ⊗ ωp ⊗ e2 ⊗ . . . ⊗ eq .
Proof: Just apply the universal property to the multilinear map

q−1
T : E ∗ × . p. . ×E ∗ × E× . q. . ×E −→ Tp−1 E
T (ω1 , . . . , ωp , e1 , . . . , eq ) = ω1 (e1 ) ω2 ⊗ . . . ⊗ ωp ⊗ e2 ⊗ . . . ⊗ eq .
Note: We have a contraction Cij : Tpq E → Tp−1

q−1
E for any other pair of indices i, j.
The trace of an endomorphism T = ij aji ωi ⊗ ej is trT = C11 T = i aii .
P P
2.7.1 Alternate Tensors

If T ∈ T p E ∗ is a covariant p-tensor and σ ∈ Sp , we put
(σT )(e1 , . . . , ep ) = T (eσ(1) , . . . , eσ(p) ). (2.10)
(the permutation σ is applied to the places, not to the indices).

1. The map σ : T p E ∗ −→ T p E ∗ is linear.
2. For any linear map f : F → E, we have f ∗ (σT ) = σ(f ∗ T ).
3. τ (σT ) = (τ σ)T ; σ, τ ∈ Sp .
4. σ(ω1 ⊗ . . . ⊗ ωp ) = ωσ−1 (1) ⊗ . . . ⊗ ωσ−1 (p) ; ω1 , . . . , ωp ∈ E ∗ .

Proof: The two first properties are obvious. Let us see the third and fourth,

τ (σT ) (e1 , . . . , ep ) = (σT )(eτ (1) , . . . , eτ (p) ) = T (eτ (σ1) , . . . , eτ (σp) )

= T (e(τ σ)1 , . . . , e(τ σ)p ) = (τ σ)T (e1 , . . . , ep ),
(σ(ω1 ⊗ . . . ⊗ ωp ))(e1 , . . . , ep ) = ω1 (eσ(1) ) · . . . · ωp (eσ(p) )
= ωσ−1 (1) (e1 ) · . . . ·ωσ−1 (p) (ep ) = (ωσ−1 (1) ⊗ . . . ⊗ ωσ−1 (p) )(e1 , . . . , ep ).
Definition: A covariant tensor Ωp ∈ T p E ∗ is alternate, or a p-form, when
Ωp (. . . , e, . . . , e, . . .) = 0 , ∀e ∈ E.
The alternate tensors of order p form a vector subspace Λp E ∗ ⊆ T p E ∗ , and we agree that
Λ0 E ∗ = k and Λ1 E ∗ = E ∗ . Analogously we have a subspace Λq E ⊆ T q E of alternate contravari-
ant q-tensors.
If f : F → E is linear, then f ∗ Ωp is alternate, and f ∗ : Λp E ∗ → Λp F ∗ is linear.
Lemma: σΩp = (sgn σ)Ωp , for any p-form Ωp .
Proof: We may assume that σ = (ij) is a transposition,
0 = Ωp (. . . , ei + ej , . . . , ei + ej , . . .)
= Ω(. . . ei . . . ei . . .) + Ω(. . . ei . . . ej . . .) + Ω(. . . ej . . . ei . . .) + Ω(. . . ej . . . ej . . .)
= Ωp (. . . , ei , . . . , ej , . . .) + Ωp (. . . , ej , . . . , ei , . . .).
Definition: The anti-symmetrization of a covariant p-tensor T ∈ T p E ∗ is the tensor

P
h(T ) = (sgn σ)(σT )
σ∈Sp
and we agree that h(T ) = T when p = 0 or 1. By property 4,

P
h(ω1 ⊗ . . . ⊗ ωp ) = (sgn σ)(ωσ(1) ⊗ . . . ⊗ ωσ(p) ). (2.11)
σ∈Sp
Lemma: h : T • E ∗ −→ Λ• E ∗ is a surjective linear map, and the kernel is a two-sided ideal,
h(T ) = 0 ⇒ h(T ⊗ T 0 ) = h(T 0 ⊗ T ) = 0. (2.12)
Proof: If τ transposes two repeated terms in a sequence (. . . , e, . . . , e, . . .), then

P P P
(sgn σ)(σT ) and τ (sgn σ)(σT ) = − sgn (τ σ)(τ σT ),
σ∈Ap σ∈Ap σ∈Ap
coincide on (. . . , e, . . . , e, . . .), and the difference h(T ) vanishes; i.e., h(T ) ∈ Λp E ∗ .

Let us see that h is surjective. Let e1 , . . . , en be a base of E, and ω1 , . . . , ωn the dual base.
Put λi1 ...ip = Ωp (ei1 , . . . , eip ), so that λσ(i1 )...σ(ip ) = (sgn σ)λi1 ...ip , and
P
Ωp = λi1 ...ip ωi1 ⊗ . . . ⊗ ωip
1≤i1 ,...,ip ≤n
P P
= (sgn σ)λi1 ...ip ωσ(i1 ) ⊗ . . . ⊗ ωσ(ip )
i1 ≤...≤ip σ∈Sp
P P
= λi1 ...ip h(ωi1 ⊗ . . . ⊗ ωip ) = h λi1 ...ip ωi1 ⊗ . . . ⊗ ωip . (2.13)
i1 ≤...≤ip i1 ≤...≤ip
Finally, we identify any permutation σ ∈ Sp with a permutation σ ∈ Sp+q fixing the numbers
p + 1, . . . , p + q, so that for any covariant q-tensor T 0 we have
(sgn σ)σ(T ⊗ T 0 ) = h(T ) ⊗ T 0 = 0,

P
σ∈Sp
(sgn τ σ) τ σ(T ⊗ T 0 ) = 0, for all τ ∈ Sp+q ,

P
σ∈Sp
and the sum in h(T ⊗ T 0 ) corresponding to any equivalence class τ Sp is null.

Hence the total sum vanishes. Analogously h(T 0 ⊗ T ) = 0.
2.7. TENSORS 65
Definition: Since Λ• E ∗ is the quotient of T • E ∗ by an ideal, we have a product on Λ• E ∗ , induced

by the tensor product. The exterior product of two alternate tensors Ωp , Ωq is
Ωp ∧ Ωq = h(Tp ⊗ Tq ); where Ωp = h(Tp ), Ωq = h(Tq ).
Since h(Ωp ) = (p!)Ωp , then Ωp = h p!1 Ωp when char k = 0, and

1
(char k = 0) Ω p ∧ Ωq = h(Ωp ⊗ Ωq ).
p!q!
1. (λΩp + µΩ̄p ) ∧ Ωq = λ(Ωp ∧ Ωq ) + µ(Ω̄p ∧ Ωq ).

Ωp ∧ (λΩq + µΩ̄q ) = λ(Ωp ∧ Ωq ) + µ(Ωp ∧ Ω̄q ).
2. (Ωp ∧ Ωq ) ∧ Ωr = Ωp ∧ (Ωq ∧ Ωr ).
3. Ωp ∧ Ωq = (−1)pq Ωq ∧ Ωp .
4. ω ∧ ω = 0, ω ∈ E ∗ .
5. f ∗ (Ωp ∧ Ωq ) = f ∗ (Ωp ) ∧ f ∗ (Ωq ).

P
6. ω1 ∧ . . . ∧ ωp = σ∈Sp (sgn σ)(ωσ(1) ⊗ . . . ⊗ ωσ(p) ).
7. dim Λp E ∗ = np , so that Λp E ∗ = 0 when p > n. In fact, if ω1 , . . . , ωn is a base of E ∗ , then

the p-forms ωi1 ∧ . . . ∧ ωip , i1 < . . . < ip , form a base of Λp E ∗ , and

P
Ωp = Ωp (ei1 , . . . , eip ) ωi1 ∧ . . . ∧ ωip .
i1 ≤...≤ip
8. ω1 , . . . , ωp ∈ E ∗ are linearly independent if and only if ω1 ∧ . . . ∧ ωp 6= 0.
Proof: Properties 1, 2 and 5 follow from the corresponding properties of the tensor product.
To prove 4, when τ = (12), we have,
ω ∧ ω = h(ω ⊗ ω) = ω ⊗ ω − τ (ω ⊗ ω) = ω ⊗ ω − ω ⊗ ω = 0.
(3) First we prove the case p = q = 1. Given two 1-forms ω, ω 0 ,
0 = (ω + ω 0 ) ∧ (ω + ω 0 ) = ω ∧ ω + ω ∧ ω 0 + ω 0 ∧ ω + ω 0 ∧ ω 0 = ω ∧ ω 0 + ω 0 ∧ ω ;
hence ω ∧ ω 0 = −ω 0 ∧ ω. Now the case Ωp = ω1 ∧ . . . ∧ ωp , Ωq = θ1 ∧ . . . ∧ θq follows directly, and

we may easily conclude the general case.
(6) It follows from 2.11.
(7) The p-forms ωi1 ∧ . . . ∧ ωip span Λp E ∗ since, by 2.13,
P
Ωp = Ωp (ei1 , . . . , eip ) ωi1 ∧ . . . ∧ ωip ,
i1 ≤...≤ip
and they are linearly independent since (ωi1 ∧ . . . ∧ ωip )(ej1 , . . . , ejp ) = 0, except if j1 , . . . , jp is
a reordering of i1 < . . . < ip .
(8) Any linearly independent family ω1 , . . . , ωp may be extended to a base of E ∗ .
Hence ω1 ∧ . . . ∧ ωp is in a base of Λp E ∗ , and it is non null.
If ω1 , . . . , ωp are linearly dependent, someone is a linear combination of the remaining, and
property 4 shows that ω1 ∧ . . . ∧ ωp = 0.
Note: The kernel of h : T • E → Λ• E is the ideal I generated by the tensors e ⊗ e.

In fact we have a surjective morphism T • E/I → Λ• E, since h(e⊗e) = 0. Moreover, in T • E/I
we have [e] · [e0 ] = −[e0 ] · [e], because [e] · [e] = 0. If we fix a base, the products [ei1 ] . . . [eip ],
i1 < . . . < ip , span T • E/I. Hence T • E/I → Λ• E is an isomorphism: I = Ker h.
Definition: If Ωp ∈ Λp E ∗ , e ∈ E, the interior contraction is the (p − 1)-form
(ie Ωp )(e2 , . . . , ep ) = Ωp (e, e2 , . . . , ep ).
Theorem: ie (Ωp ∧ Ωq ) = (ie Ωp ) ∧ Ωq + (−1)p Ωp ∧ (ie Ωq ).
Proof: Let us fix a base e = e1 , . . . , en and the dual base ω1 , . . . , ωn .

We may assume that Ωp = ωi1 ∧ . . . ∧ ωip = ωi1 ∧ ωI , and Ωq = ωj1 ∧ . . . ∧ ωjq = ωj1 ∧ ωJ .
When i1 > 1 and j1 > 1, both terms are 0. When i1 = 1 and j1 > 1,
ie (Ωp ∧ Ωq ) = ie (ω1 ∧ ωI ∧ Ωq ) = ωI ∧ Ωq ,
(ie Ωp ) ∧ Ωq + (−1) Ωp ∧ (ie Ωq ) = ωI ∧ Ωq + (−1)p Ωp ∧ 0 = ωI ∧ Ωq ,
p
and analogously when i1 > 1 and j1 = 1. When i1 = 1 and j1 = 1,
ie (Ωp ∧ Ωq ) = ie (ω1 ∧ ωI ∧ ω1 ∧ ωJ ) = ie (0) = 0,

(ie Ωp ) ∧ Ωq + (−1) Ωp ∧ (ie Ωq ) = ωI ∧ Ωq + (−1)p ω1 ∧ ωI ∧ ωJ = ωI ∧ Ωq − ωI ∧ Ωq = 0.
p
Universal Property: For any alternate multilinear map H : E× . p. . ×E → F there exists a

unique linear map f : Λp E → F such that
f (e1 ∧ . . . ∧ ep ) = H(e1 , . . . , ep ). (2.14)
Proof: We fix a base v1 , . . . , vn of E, and we define a linear map f : Λp E → F imposing 2.14 on

the corresponding base {vi1 ∧ . . . ∧ vip } of Λp E. Now both terms in 2.14 are multilinear maps
E× . p. . ×E → F coinciding on the sequences of vectors of a base; hence both coincide.
Uniqueness is clear, since the products e1 ∧ . . . ∧ ep span Λp E.
Corollary: We have canonical isomorphisms (Λp E)∗ = Λp E ∗ ,
(ω1 ∧ . . . ∧ ωp )(e1 ∧ . . . ∧ ep ) = (ω1 ∧ . . . ∧ ωp )(e1 , . . . , ep ).
Proof: Just put F = k in the above universal property.
Determinants and Volume Forms

Definition: The determinant of a matrix A = (aij ) with n rows and columns is
P
|A| = (sgn σ) a1,σ(1) . . . an,σ(n) .
σ∈Sn
Proposition: (ω1 ∧ . . . ∧ ωp )(e1 , . . . , ep ) = |ωi (ej )| = |ωj (ei )|.

P P
Proof: σ (sgn σ)(σ(ω1 ⊗ . . . ⊗ ωp )) = σ (sgn σ)(ωσ(1) ⊗ . . . ⊗ ωσ(p) ). q.e.d.
The coordinates of ω1 ∧ . . . ∧ ωp in a base e1 , . . . , en of E are

ω1 (ei1 ) . . . ω1 (eip )

(ω1 ∧ . . . ∧ ωp )(ei1 , . . . , eip ) = . . . ... . . . .
ωp (ei ) . . . ωp (ei )
1 p
2.7. TENSORS 67
This determinant is the p-minor formed with the columns i1 , . . . , ip of the matrix whose rows
are the coordinates of ω1 , . . . , ωp in the dual base. Hence, p given rows of a matrix are linearly
independent if and only if they contain a non null minor of order p:
The rank of a matrix is the largest order of all non-zero minors (Rank Theorem).
Definition: Since dim Λn E ∗ = 1, n = dim E, any endomorphism Λn E ∗ → Λn E ∗ is a homothety.

Hence any endomorphism T : E → E induces an endomorphism T ∗ : Λn E ∗ → Λn E ∗ , the
product by some scalar det(T ), named determinant of T .
Proposition: The determinant of an endomorphism T coincides with the determinant of the

matrix A = (aij ) in any base, det(T ) = |A|.
Proof: T ∗ (ω1 ∧ . . . ∧ ωn ) = (T ∗ ω1 ) ∧ . . . ∧ (T ∗ ωn ), and T ∗ (ωi ) = j aij ωj .

P
Theorem: det(T S) = (det T )(det S).
Proof: (T S)∗ = S ∗ T ∗ .
Lemma: Let n = dim E and 0 6= ΩE ∈ Λn E ∗ . Some vectors e1 , . . . , en form a base of E if and

only if ΩE (e1 , . . . , en ) 6= 0.
Proof: If e1 , . . . , en form a base of E, the coordinate of ΩE in the corresponding base of Λn E ∗ is

ΩE (e1 , . . . , en ), and it is not 0 since ΩE 6= 0.
If e1 , . . . , en are linearly dependent, someone is a linear combination of the remaining; hence
ΩE (e1 , . . . , en ) = 0, because ΩE is alternate.
Theorem: An endomorphism T is invertible if and only if det(T ) 6= 0.
Proof: Fix a base e1 , . . . , en of E, and a non-zero n-form ΩE .

ΩE T (e1 ), . . . , T (en ) = (T ∗ ΩE )(e1 , . . . , en ) = (det T ) ΩE (e1 , . . . , en ).

(2.15)
Hence T (e1 ), . . . , T (en ) is a base if and only if det(T ) 6= 0.
Definitions: Let E be a real vector space of dimension n. A non null n-form ΩE is a vol-
ume form, and ΩE (e1 , . . . , en ) is the volume (with sign) of the parallelepiped determined by
e1 , . . . , en . Two volume forms ΩE , Ω0E define the same orientation of E when Ω0E = λΩE for
some λ > 0. Equivalence classes are orientations of E. Once we fix an orientation [ΩE ], a base
e1 , . . . , en of E is direct if ΩE (e1 , . . . , en ) > 0. By 2.15,

Volume of Volume of
= (det T )
T e1 , . . . , T en e1 , . . . , e n
If we fix a scalar product in E, the polarity φ : E → E ∗ induces isomorphisms
φ : Λp E ∗ −→ Λp E = (Λp E ∗ )∗ ,
hence a scalar product in Λp E ∗ : Ωp · Ω0p = (φΩp )(Ω0p ).
If ω1 , . . . , ωn is the dual base of an orthonormal base e1 , . . . , en , the p-forms ωi1 ∧ . . . ∧ ωip
define an orthonormal base of Λp E ∗ :
φ(ωi1 ∧ . . . ∧ ωip )(ωj1 ∧ . . . ∧ ωjp ) = (ei1 ∧ . . . ∧ eip )(ωj1 ∧ . . . ∧ ωjp ) = δi1 j1 . . . δip jp .
Theorem: Let E be an Euclidean oriented vector space E. There is a unique volume form ΩE
such that the volume of any direct orthonormal base is 1.
Proof: Since dim Λn E ∗ = 1, the orientation has a unique volume form ΩE of module 1 and, if
e1 , . . . , en is a direct orthonormal base, we have ΩE = ω1 ∧ . . . ∧ ωn ; hence ΩE (e1 , . . . , en ) = 1.
Chapter 3
Algebra I
3.1 The Quotient Ring

Definitions: An ideal m 6= A is maximal if A is the unique ideal strictly containing m.
An ideal p 6= A is prime when ab ∈ p ⇒ a ∈ p or b ∈ p.
The sum of ideals a + b = {a + b : a ∈ a, b ∈ b} is the minimal ideal containing them.
The ideal generated by a1 , . . . , an ∈ A is the ideal (a1 , . . . , an ) = a1 A + . . . + an A.
An ideal a is principal if it is generated by some element, a = aA.
A principal ideal domain is a domain where any ideal is principal.
The product of two ideals a and b is the ideal
ab = {a1 b1 + . . . + an bn ; a1 , . . . , an ∈ a, b1 , . . . , bn ∈ b}
Theorem: An ideal a of a ring A is prime ⇔ A/a is an integral ring.

An ideal a of a ring A is maximal ⇔ A/a is a field.
Proof: If a is prime and [a] · [b] = 0, then ab ∈ a, and a ∈ a or b ∈ a; hence, [a] = 0 or [b] = 0.
Conversely, if A/a is integral and ab ∈ a, then [a] · [b] = [ab] = 0, and [a] = 0 or [b] = 0;
hence, a ∈ a or b ∈ a.
If a is maximal and [a] 6= 0, then the inclusion a ⊂ a + aA is strict; hence A = a + aA, and
1 = x + ab, where x ∈ a, b ∈ A. Therefore [a] · [b] = 1 and [a] is invertible in A/a.
Conversely, if A/a is a field and a ⊂ b, we take b ∈ b not in a. Since [b] 6= 0 in A/a, there
exists [a] ∈ A/a such that [a][b] = 1. Hence 1 ∈ ab + a ⊆ b, and b = A.
Corollary: Any maximal ideal is a prime ideal.
Proof: Any field is a domain.
Euclid’s Lemma: If a prime number p divides a product, it divides some factor.
Proof: Since pZ is a maximal ideal of Z, it is a prime ideal.
Theorem: If p : A → Ā is a surjective ring morphism, we have a lattice isomorphism

Ideals Ideals of A
(I := Ker p) −−∼
→ , ā 7→ p−1 (ā),
of Ā containing I
the inverse map being a 7→ p(a). If ā = p(a), we have a natural ring isomorphism A/a ' Ā/ā;
hence a is a maximal (resp. prime) ideal if and only if so is ā.
69
70 CHAPTER 3. ALGEBRA I
Proof: We have p(p−1 ā) = ā because p is surjective and, when I ⊆ a, we also have
p−1 p(a) = a + Ker p = a,

p π
so that the kernel of the epimorphism A −−→ Ā −−→ Ā/ā is just a; hence A/a ' Ā/ā.
Chinese Remainder Theorem: Let a and b be two ideals of a ring A. If a + b = A, then

a ∩ b = ab, and we have a natural ring isomorphism
∼
φ : A/a ∩ b −−→ (A/a) × (A/b), φ([x]ab ) = ([x]a , [x]b ).
Proof: Let 1 = a + b ∈ a + b. If c ∈ a ∩ b, then c = c(a + b) = ca + cb ∈ ab.

Hence a ∩ b = ab, since the inclusion ab ⊆ a ∩ b always holds.
Moreover, the kernel of the morphism f : A → (A/a) × (A/b), f (x) = ([x]a , [x]b ), is a ∩ b,
and f is surjective because f (bx + ay) = ([x]a , [y]b ),
x = (a + b)x ≡ bx ≡ bx + ay (mod. a)
y = (a + b)y ≡ ay ≡ bx + ay (mod. b)
Corollary: Z/mnZ = (Z/nZ) × (Z/nZ), when m and n are coprime.
Definition: The Euler indicator φ(n) is the number of integers between 1 and n which are
prime to n; i.e. the number of generators of any cyclic group of order n, and the order of the
group of units (Z/nZ)∗ of the ring Z/nZ. In fact, a class [m] is invertible, [1] = [a] · [m] = a[m],
if and only if it generates the group Z/nZ.
Proposition: φ(pr ) = (p − 1)pr−1 , when p is prime.

φ(n · m) = φ(n) · φ(m), when n and m are coprime.
Proof: The integers between 1 and pr which are not prime to pr are just p, 2p, . . . , pr−1 p.
With respect to the second formula, by the chinese remainder theorem
(Z/nmZ)∗ = (Z/nZ × Z/mZ)∗ = (Z/nZ)∗ × (Z/mZ)∗ .
Euler’s Congruence: aφ(n) ≡ 1 (mod. n), when a, n are coprime.
Proof: If [a] ∈ (Z/nZ)∗ , then [1] = [a]φ(n) = [aφ(n) ], (p. 44).
Fermat’s Congruence: np−1 ≡ 1 (mod. p), when a prime p does not divide n.
Corollary: np ≡ n (mod. p), when p is prime.
Wilson’s congruence: (p − 1)! ≡ −1 (mod. p) when p is a prime number.
Proof: xp−1 − 1 = (x − 1)(x − 2) . . . (x − (p − 1)) since 1, . . . , p − 1 are roots of xp−1 − 1 in Fp .

a
Definition: If a ∈ Z is not a multiple of an odd prime p, the Legendre’s symbol p is 1
when ā ∈ F∗p is a square, and −1 otherwise.

p−1
Corollary: If p is odd, then F∗2 ∗
p is a subgroup of Fp of order 2 , and ab
p = a
p
b
p .
3.1. THE QUOTIENT RING 71
Proof: The kernel of the group morphism f : F∗p → F∗p , f (x) = x2 , is {±1} because x2 − 1 =
(x + 1)(x − 1); and −1 6= 1 since p 6= 2.
p−1
By the isomorphism theorem, the image F∗2 p of f has order 2 .
Now, F∗p /F∗2
p ' {±1} since it is a group of order 2, and the Legendre’s symbol is just the
canonical projection π : F∗p → F∗p /F∗2
p ' {±1}; hence it is a group morphism.
p−1
a
Corollary: p ≡a 2 (mod. p).
p−1 p−1
Proof: If a ∈ F∗p , then a 2 = ±1 because (a 2 )2 = ap−1 = 1.
p−1 p−1
If a is a square, a = b2 , then a 2 = bp−1 = 1. Since the number of roots of x 2 − 1 is
p−1
bounded by the degree, we see that a 2 = 1 if and only if a is a square in Fp .
Corollary: −1 is a quadratic residue modulo p 6= 2 if and only if p ≡ 1 (mod. 4).
Gauss Lemma: If P ∈ Z[x] is a product of polynomials with rational coefficients, P = Q1 Q2 ,

multiplying both factors by some constants we have a decomposition P = Q01 Q02 in Z[x].
Proof: We may assume that the coefficients of Q1 and Q2 have a common denominator
1 1
P = (a0 xn + . . . + an ) · (b0 xm + . . . + bm )
a b
abP = (a0 xn + . . . + an )(b0 xm + . . . + bm )
where a, a0 , . . . , an , b, b0 , . . . , bm ∈ Z.
If a prime p divides ab, by the following lemma p divides some factor of the second term.
So eliminating all the prime factors of ab, we get a decomposition in Z[x],
P = (a00 xn + . . . + a0n )(b00 xm + . . . + b0m ).
Lemma: If p is a prime ideal of a ring A, then p[x] is a prime ideal of A[x]
Proof: The ideal p[x] is the kernel of the surjective ring morphism
i i
P P
φ : A[x] −→ (A/p)[x], φ i ai x = i āi x .
Hence A[x]/p[x] ' (A/p)[x] is a domain, and p[x] is a prime ideal.
Corollary: If P ∈ Z[x] is irreducible of degree ≥ 1, then P (x) is irreducible in Q[x].
Note: Given P ∈ Z[x], let us fix different integers a0 , a1 , . . . , ad . If Q ∈ Z[x] divides P , then
Q(ai ) divides P (ai ). Since any integer has a finite number of divisors, interpolating (p. 47)
we obtain all the possible divisors of P (x) in Z[x] of a given degree d. So we may obtain the
irreducible factor decomposition of P in Z[x]; hence in Q[x] by Gauss lemma.
Reduction Criterion: Let Q = c0 xn + . . . + cn ∈ Z[x], and let p be a prime not dividing c0 . If

Q has a factor of degree d in Z[x], then the reduction Q̄ = c̄0 xn + . . . + c̄n ∈ Fp [x] has a factor
of degree d. Hence, if Q̄ is irreducible in Fp [x], then so is Q in Q[x].
Proof: We have deg Q̄ = deg Q, since c̄0 6= 0.

If we have a decomposition Q = AB in Z[x], then Q̄ = ĀB̄ in Fp [x] and
deg Ā + deg B̄ = deg Q̄ = deg Q = deg A + deg B.

Since deg Ā ≤ deg A, deg B̄ ≤ deg B, then deg Ā = deg A, and deg B̄ = deg B.
Finally, if Q̄ is irreducible, then Q has no factor of degree 1, . . . , n − 1 in Z[x].
Hence Q has no factor of degree 1, . . . , n − 1 in Q[x] by Gauss lemma.
Eisenstein’s Criterion: Q(x) = c0 xn + . . . + cn ∈ Z[x] is irreducible in Q[x] if there is a prime

p not dividing c0 such that p divides c1 , . . . , cn and p2 does not divide cn .
Proof: By Gauss lemma, if Q is not irreducible, it is a product of non constant polynomials with
integer coefficients, and reducing modulo p we have
Q = (a0 + a1 x + . . . + ar xr )(b0 + b1 x + . . . + bn−r xn−r ),

c̄0 xn = (ā0 + ā1 x + . . . + ār xr )(b̄0 + b̄1 x + . . . + b̄n−r xn−r ),
c̄0 xn = (ār xr )(b̄n−r xn−r ).
Hence ā0 = b̄0 = 0, and cn = a0 b0 is a multiple of p2 , a contradiction.
Corollary: The polynomial xp−1 + . . . + x + 1 is irreducible in Q[x] when p is prime.

xp −1
Proof: x−1 = xp−1 + . . . + x + 1 is irreducible if and only if so is
(x + 1)p − 1

p−1 p p−2 p p−i−1 p
=x + x + ... + x + ... + .
x 1 i p−1
The number pi = p(p−1)...(p−i+1) p

i! is a multiple of p when 1 ≤ i ≤ p − 1, and p−1 = p is not
2
a multiple of p . Hence it is irreducible by the Eisenstein’s criterion.
3.2 Principal Ideal Domains

Definitions: Let A be a principal ideal domain. If a, b ∈ A, then we have aA + bA = dA
and aA ∩ bA = mA for some d, m ∈ A, well-defined up to invertible factors. Then d is the
greatest common divisor of a and b (a common divisor which is a multiple of any other
common divisor), and m is the least common multiple, (a common multiple dividing any
other common multiple). The equality dA = aA + bA proves
Bézout’s Identity: d = αa + βb, for some α, β ∈ A.
Euclid’s Lemma: Let 0 6= p ∈ A. The following conditions are equivalent:
1. p is irreducible.
2. pA is a maximal ideal.
3. pA is a prime ideal.
Proof: (1 ⇒ 2) If p is irreducible and pA ⊆ bA, then p = bc; hence b or c is invertible in A.

If so is b, then bA = A; and if so is c, then bA = bcA = pA.
(2 ⇒ 3) Any maximal ideal is prime (p. 69).
(3 ⇒ 1) If pA is prime, and ab = p ∈ pA, then a ∈ pA or b ∈ pA. If a ∈ pA, then
a = pc, p = ab = pcb, bc = 1,
3.2. PRINCIPAL IDEAL DOMAINS 73
and b is invertible. Analogously, a is invertible if b ∈ pA. Hence p is irreducible.
Decomposition Theorem: Any proper element of a principal ideal domain A decomposes,

uniquely up to the order and units, as a product of irreducible elements, a = p1 . . . pr , r ≥ 1.
Proof: First we prove the uniqueness. If a = p1 . . . pr , then the number of factors coinciding, up
to units, with a given factor p is the greatest exponent n such that pn divides a.
In fact, if pn |a, by Euclid’s lemma p divides some factor pi ; hence p coincides with pi up to
a unit and pn−1 divides p1 . . . pbi . . . pr . Iterating we see that n factors coincide with p.
Now we prove the existence. If a proper element a ∈ A does not decompose as a product of
irreducible elements, then we have a = bc, where aA ⊂ bA, aA ⊂ cA, and some factor is proper
and does not decompose as a product of irreducible elements.
So we obtain an infinite increasing sequence of ideals of A, against the following
Lemma: Any chain of ideals I1 ⊆ I2 ⊆ . . . of A stabilizes, In = In+1 = . . ..
Proof: If a, b ∈ ∪i Ii , then a, b ∈ Ii for some index i, so that a + b, xa ∈ Ii ⊆ ∪i Ii , ∀x ∈ A, and

we see that ∪i Ii is an ideal of A. Let c be a generator of ∪i Ii .
Now c ∈ In for some index n, and In ⊆ In+j ⊆ ∪i Ii = cA ⊆ In ; hence In = In+j , ∀j > 0.
Definition: A domain A is Euclidean if there is a map δ : A − {0} → N such that

1. If 0 6= a, b ∈ A, then b = ac + r for some c, r ∈ A, where δ(r) < δ(a) or r = 0.
2. δ(a) ≤ δ(ab) for any non null a, b ∈ A.
Theorem: Any ideal a of an Euclidean ring A is principal, a = aA.
Proof: If a = 0, we take a = 0. If a 6= 0, we take a ∈ a with δ(a) minimal, and we have aA ⊆ a

because a ∈ a. Now, if b ∈ a, dividing b by a
b = ac + r, where δ(r) < δ(a) or r = 0,
r = b − ac ∈ a;
we see that r = 0, by the choice of a. Hence b = ac ∈ aA, and a = aA.
Euclid’s Algorithm: If a = cb + r, then g.c.d.(a, b) = g.c.d.(b, r).
Proof: (a, b) ⊆ (b, r) since a = cb + r, and (b, r) ⊆ (a, b) since r = a − cb. q.e.d.
1. The ring Z is Euclidean, with δ(n) = |n|.
2. If k is a field, the ring k[x] is Euclidean, with δ(P ) = deg P . Hence, any non null ideal a is
generated by a unitary polynomial xn + a1 xn−1 + . . . + an , clearly unique.
3. The ring of gaussian integers A = Z[ i ] = Z + Zi is Euclidean, with δ(z) = z · z̄ = |z|2 .
In fact, for any complex number u + vi there is x + yi ∈ A such that |(u + vi) − (x + yi)| < 1
(just take |u − x|, |v − y| ≤ 1/2). Hence, if a, b ∈ A, a 6= 0, there is c ∈ A such that | ab − c| < 1;
so that r = b − ac ∈ A, and |r| = |a ab − c | < |a|.

4. In a Euclidean ring, iterated divisions let us calculate d =g.c.d.(a, b) as the last non null
remainder, and the coefficients of Bézout’s Identity. In fact, if d = αr + βb, then
d = αr + βb = α(a − cb) + βb = αa + (β − cα)b.
5. The Diophantine equation ax + by = c has integer solution if and only if c ∈ aZ + bZ = dZ,

and in such a case, if d = αa + βb, a solution is xo = αc/d, yo = βc/d.
3.3 Roots and Extensions

Definitions: Let k be a field. A k-algebra is a ring A endowed with a ring morphism j : k → A,
so that A is a k-vector space, λa := j(λ)a, and we identify λ with j(λ).
A morphism of k-algebras f : A → B is a ring morphisms such that f (λ) = λ, λ ∈ k, and
it is an isomorphism if it is bijective.
The dimension of A as a k-vector space is the degree [A : k] of A over k, and A is a finite
k-algebra if so is the degree.
Definitions: A k-algebra k → L is an extension of k if L is a field, and it is a finite extension
∼
if so is the degree [L : k]. It is a trivial extension if k −
→ L; i.e. [L : k] = 1.
The extension generated by α1 , . . . , αn ∈ L is
k(α1 , . . . , αn ) = {a/b : a, b ∈ k[α1 , . . . , αn ], b 6= 0} .
A root of P = c0 xn + . . . + cn ∈ k[x] is an element α of some extension L of k such that

P (α) = c0 αn + . . . + cn = 0, and the multiplicity is the greatest exponent m such that (x − α)m
divides P in L[x]. Let us consider the irreducible factor decomposition of P in L[x]:
P = c0 (x − α1 )m1 . . . (x − αr )mr Qn1 1 . . . Qns s
where deg Qi > 1. The roots of P in L are α1 , . . . , αr , with multiplicities m1 , . . . mr .

The number of roots in an extension, counted with multiplicities, is bounded by the degree,
m1 + . . . + mr ≤ deg P (x),
and we say that P has all the roots in L if the equality holds. In such a case
c0 xn + c1 xn−1 + . . . + cn−1 x + cn = c0 (x − α1 ) . . . (x − αn ),
where α1 , . . . , αn are the roots of P in L, repeated with their multiplicities.

Identifying coefficients we obtain Cardano Formulae:
(−1)r ccr0 =
P
αi1 . . . αir , 1 ≤ r ≤ n.
i1 <...<ir
Definition: A polynomial P (x1 , . . . , xn ) with coefficients in a ring A is said to be symmetric

when P (xσ(1) , . . . , xσ(n) ) = P (x1 , . . . , xn ), for any permutation σ ∈ Sn .
The elementary symmetric functions are the polynomials
P
sr (x1 , . . . , xn ) = xi1 . . . xir , 1 ≤ r ≤ n.
i1 <...<ir
For example, s1 (x1 , . . . , xn ) = x1 + . . . + xn , and sn (x1 , . . . , xn ) = x1 . . . xn .
Symmetric Functions Theorem: If P (x1 , . . . , xn ) is a symmetric polynomial, then there

exists a unique polynomial Q(x1 , . . . , xn ) with coefficients in A such that
P (x1 , . . . , xn ) = Q(s1 , . . . , sn ), sr = sr (x1 , . . . , xn ).
Proof: We prove the existence by induction on n and deg P , and we may assume that P is
homogeneous.
Since P (x1 , . . . , xn−1 , 0) is symmetric, there is a polynomial Q̄(x1 , . . . , xn−1 ) such that
P (x1 , . . . , xn−1 , 0) = Q̄(s̄1 , . . . , s̄n−1 ), s̄r = sr (x1 , . . . , xn−1 ).

3.3. ROOTS AND EXTENSIONS 75
Now P (x1 , . . . , xn ) − Q̄(s1 , . . . , sn−1 ) vanishes in the quotient by the ideal (xn ).
Hence it is a multiple of xn and, being symmetric, it is a multiple of sn = x1 . . . xn ,
P (x1 , . . . , xn ) − Q̄(s1 , . . . , sn−1 ) = sn P 0 (x1 , . . . , xn ).
Since P 0 (x1 , . . . , xn ) is symmetric and homogeneous of degree deg P − n, by induction
P 0 (x1 , . . . , xn ) = Q0 (s1 , . . . , sn )
for some polynomial Q0 (x1 , . . . , xn ), and we take Q = Q̄ + xn Q0 ,
P (x1 , . . . , xn ) = Q̄(s1 , . . . , sn−1 ) + sn Q0 (s1 , . . . , sn ) = Q(s1 , . . . , sn ).
Uniqueness follows from the following theorem.
Theorem: Let Q ∈ A[x1 , . . . , xn ]. If Q(s1 , . . . , sn ) = 0, then Q(x1 , . . . , xn ) = 0.
Proof: It is clear when n = 1. If n ≥ 2 and the theorem is false, we take Q(x1 , . . . , xn ) 6= 0 of

minimum degree in xn such that Q(s1 , . . . , sn ) = 0,
Q = Q0 (x1 , . . . , xn−1 ) + Q1 (x1 , . . . , xn−1 )xn + . . . + Qd (x1 , . . . , xn−1 )xdn .
If we put xn = 0 in the identity Q(s1 , . . . , sn ) = 0, we obtain Q0 (s̄1 , . . . , s̄n−1 ) = 0.

By induction Q0 = 0, and
Q(x1 , . . . , xn ) = xn · (Q1 + . . . + Qd xnd−1 ) = xn · R(x1 , . . . , xn ) .
Hence R(x1 , . . . , xn ) 6= 0 and R(s1 , . . . , sn ) = 0, against the choice of Q.
Lemma: k[x]/(P ) is a k-algebra of degree d = deg P , and 1, x̄, . . . , x̄d−1 form a base.
Proof: If V = k ⊕ . . . ⊕ kxd−1 , we have to show that the linear map
π : V → k[x]/(P ), π(Q) = [Q],
is an isomorphism. It is injective because V ∩(P ) = 0, and it is surjective because any polynomial

coincides in k[x]/(P ) with the remainder of the division by P .
Kronecker’s Theorem: If P ∈ k[x] is irreducible, then a root of P is x̄ ∈ k[x]/(P ); and if

α ∈ L is another root, there is an isomorphism of k-algebras k[x]/(P ) ' k(α), x̄ 7→ α.
Proof: The ideal (P ) is maximal by Euclid’s lemma; hence k[x]/(P ) is an extension of k.

Let us see that x̄ is a root of P = i ai xi ,
P
P (x̄) = i ai [ x ]i = i = [P (x)] = 0.
P P
i ai x
If α ∈ L is another root, the image of the morphism of k-algebras k[x] → L, x 7→ α, is k[α],

and the kernel contains the maximal ideal (P ); hence it is (P ).
By the isomorphism theorem, we have an isomorphism
∼
k[x]/(P ) −−→ k[α], [Q(x)] 7→ Q(α).
Therefore k[α] is a field, because so is k[x]/(P ), and k[α] = k(α).

Corollary: Let P ∈ k[x] be an irreducible polynomial of degree d. If α is a root of P , then P

divides any other polynomial with coefficients in k admitting the root α, and
k(α) = k ⊕ kα ⊕ kα2 ⊕ . . . ⊕ kαd−1 .

√
Examples: The polynomial xn − 2 is irreducible in Q[x] by Eisenstein criterion; hence Q( n 2 )
is a finite extension of Q of degree n.
The equality k(α) = k[α] is just the rationalization of algebraic expressions: if Q ∈ k[x] and
1
Q(α) 6= 0, we put x = α in Bézout’s identity, 1 = AP + BQ, and we obtain Q(α) = B(α).
Theorem: Any non constant polynomial P ∈ k[x] has all the roots in a finite extension of k.
Proof: P (x) has a root α in a finite extension K; hence P = (x − α)Q, where Q ∈ K[x].
By induction on the degree, Q has all the roots in finite extension L of K.
Hence P = (x − α)Q has all the roots in L, a finite extension of k by the following result:
Degree Theorem: If k → K → L are finite extensions, then
[L : k] = [L : K] · [K : k] .
Proof: If K = ku1 ⊕ . . . ⊕ kun , and L = Kv1 ⊕ . . . ⊕ Kvm , then

L
L = (ku1 ⊕ . . . ⊕ kun )v1 ⊕ . . . ⊕ (ku1 ⊕ . . . ⊕ kun )vm = i,j kui vj .
D’Alembert’s Theorem: Any non constant polynomial P ∈ C[x] has a complex root.
Proof: If P ∈ R[x] has degree n = 2d m, with m odd, we proceed by induction on d.

It holds when d = 0 by Bolzano’s theorem.
In general, by Kronecker’s theorem, we have P = (x − α1 ) . . . (x − αn ) in some extension L
of C.
Given a ∈ R, the polynomial with roots αi + αj + aαi αj has degree n2 = 2d−1 m(n − 1) and

real coefficients (they are symmetric functions of the roots αi ). By induction this polynomial
has a complex root. Hence there are indices i, j and real numbers a 6= b such that
αi + αj + aαi αj , αi + αj + bαi αj ∈ C .
Therefore αi +αj , αi αj ∈ C, so that αi , αj are roots of a polynomial of degree 2 with complex

coefficients; hence αi , αj ∈ C.
If P ∈ C[x], and P̄ denotes the conjugate polynomial, then P P̄ ∈ R[x], and P P̄ has a
complex root α. It is a root of P or P̄ , in which case ᾱ is a root of P .
3.3.1 Quadratic Irrationals

Definitions: An element α of an extension L of k is algebraic over k if it is a root of some
non null polynomial with coefficients in k, hence of an irreducible factor Pα (the minimal
polynomial of α). An extension k → L is algebraic if any element of L is algebraic over k.
Pα divides any other polynomial with coefficients in k admitting the root α (p. 76) and
[k(α) : k] = deg Pα .
Lemma: α ∈ L is algebraic over k if and only if k(α) is a finite extension of k.

Proof: If k(α) is a finite extension of degree d, then 1, α, α2 , . . . , αd are linearly dependent,

a0 + a1 α + . . . + ad αd = 0, where ai ∈ k, and α is algebraic over k.
Corollary: If α1 , . . . , αn ∈ L are algebraic over k, then k → k(α1 , . . . , αn ) is a finite extension.
Proof: By induction on n, the following extensions are finite
k −→ k(α1 , . . . , αn−1 ) −→ k(α1 , . . . , αn−1 , αn ).
Theorem: The elements of L algebraic over k form an extension of k.
Proof: If α, β are algebraic over k, then k(α, β) is a finite extension of k.

Any element of k(α, β) is algebraic; hence so are α + β, αβ, and α/β when β 6= 0.
Theorem: Let k → K be an algebraic extension, and let L be an extension of K. If α ∈ L is

algebraic over K, then α is algebraic over k.
Proof: If α is a root of c0 xn + . . . + cn ∈ K[x], then k(c0 , . . . , cn , α) is a finite extension of

k(c0 , . . . , cn ), which is a finite extension of k, and α is algebraic over k.
Definition: A field K ⊂ C is an extension of Q by quadratic radicals if K = Q(α1 , . . . , αn ),

where αi2 ∈ Q(α1 , . . . , αi−1 ), 1 ≤ i ≤ n. A complex number is a quadratic irrational if it is in
some extension of Q by quadratic radicals.
√
Lemma: If a ∈ k, then the degree of k( a ) over k is 1 or 2.
√
Proof: Put α = a. Then [k(α) : k] = deg Pα = 1 or 2, since Pα divides x2 − a.
Theorem: The degree of any extension Q → K by quadratic radicals is a power of 2.
Proof: Let K = Q(α1 , . . . , αn ) where αi2 ∈ Q(α1 , . . . , αi−1 ), 1 ≤ i ≤ n.

Now [K : Q] is a power of 2 because we have extensions of degree 1 or 2
Q −→ Q(α1 ) −→ Q(α1 , α2 ) −→ . . . −→ Q(α1 . . . , αn−1 ) −→ Q(α1 . . . , αn ) = K.
Corollary: If α is a quadratic irrational, then the degree of Pα is a power of 2.
Proof: Let K be an extension of Q by quadratic radicals.

If α ∈ K, then Q(α) ⊆ K, so that deg Pα = [Q(α) : Q] divides [K : Q] = 2n .
Definitions: The complex roots of xn − 1 are the complex n-th roots of unity.
2πk n
Since e n i = e2πki = 1 when k ∈ Z, and the degree of a polynomial bounds the number
of roots, the complex n-th roots of unity form a cyclic group µn of order n,
2π
µn = {εn , ε2n , . . . , εnn = 1}, εn = e n i = cos 2π 2π
n + i sin n ·
The primitive n-th roots of unity are the generators of the cyclic group µn . They are εm
n
where m, n are coprime (p. 44), and they are the roots of the n-th cyclotomic polynomial
Y 2πi
m

Φn (x) = x−e n .
1≤m≤n
g.c.d.(m,n)=1
Since the order d of any element of µn divides n, and the elements of µn of order d are just
the primitive d-th roots of unity, we have
xn − 1 =
Q Q
(x − α) = Φd (x) = Φ1 (x) . . . Φn (x).
α∈µn d|n
So we may calculate Φn (x) inductively and, being unitary, it has integer coefficients1 .
For example, Φ1 = x − 1, Φ2 = x + 1, Φ3 = x2 + x + 1, Φ4 = x2 + 1, Φ6 = x2 − x + 1,
Φ8 = x4 + 1, Φ9 = x6 + x3 + 1 and Φp = xp−1 + xp−2 + . . . + x + 1 when p is prime.
Ruler-and-Compass Constructions: Fix a segment in a plane, identify the end points with
0 and 1, and the points of the plane with complex numbers. Since quadratic irrationals are
obtained from 0 and 1 by means of sums, subtractions, products, quotients and quadratic radi-
cals, they are constructible using straightedge and compass. Conversely constructible points are
quadratic irrationals, because intersections of straight lines and circles are determined in terms
of sums, subtractions, products, quotients and quadratic radicals2 , as well as the line passing
through two given points, and the circle with given center and radius.
√
2
1. The roots −b± 2a b −4ac
of a polynomial ax2 + bx + c with rational coefficients are quadratic
irrationals, as well as the roots of ax4 + bx2 + c and ax4 + bx3 + cx2 + bx + a, since the
substitutions y = x2 and y = x−1 + x transform them into quadratic equations.
2πi 2πi
p 2πi 2πi 2πi
2. If e n is a quadratic irrational, then so is e 2n = e n . Moreover, e 3 , e 5 are quadratic
irrationals, since Φ3 = x2 + x + 1 and Φ5 = x4 + x3 + x2 + x + 1. Hence so is
2πi 2πi 2πi 2πi 2πi
e 15 = (e 15 )6 (e 15 )−5 = (e 5 )2 (e 3 )−1 .
The regular polygons of 2n , 2n 3, 2n 5 and 2n 15 sides are ruler-and-compass constructible.

√
3. 3 2 is not a quadratic irrational, since it is a root of the irreducible polynomial x3 − 2.
Doubling the cube is impossible using straight-edge and compass.
2πi 2πi
4. If p is prime, the irreducible polynomial of e p is Φp = xp−1 + . . . + 1 (p. 72); hence e p is
not a quadratic irrational when p − 1 has an odd factor.
The regular polygons of 7, 11, 13, 19,. . . sides are not ruler-and-compass constructible.
5. The reduction of Φ9 = x6 + x3 + 1 modulo 2 is irreducible: it has no root in F2 and no

irreducible polynomial of degree 2 or 3, x2 + x + 1, x3 + x + 1, x3 + x2 + 1, divides Φ̄9 . Hence
2πi
Φ9 is irreducible in Q[x], and e 9 is not a quadratic irrational.
2π
Trisection of the angle of 3 radians is impossible using straight-edge and compass.
3.3.2 Simple Fractions

P B1 Br
Lemma: = + ··· + when Q1 , . . . , Qr ∈ k[x] are pairwise coprime.
Q1 . . . Qr Q1 Qr
Proof: Put Q = Q2 . . . Qr . By Bézout’s identity 1 = AQ1 + BQ, where A, B ∈ k[x].

1
and we shall see (p. 150) that Φn (x) is irreducible in Q[x].
The intersection points of two circles P = x2 + y 2 + ax + by + c = 0, P 0 = x2 + y 2 + a0 x + b0 y + c0 = 0 are
2
just the intersection points of one circle with the straight line P 0 − P = 0.
Hence P = P AQ1 + P BQ, and we conclude by induction on r,

P P BQ P AQ1 B1 PA B1 B2 Br
= + = + = + + ··· + ·
Q1 Q Q1 Q Q1 Q Q1 Q2 . . . Qr Q1 Q2 Qr
Lemma: Let Q ∈ k[x] be of degree d ≥ 1. If B ∈ k[x], then there exist A0 , . . . , An ∈ k[x], of
degree < d, such that B = A0 + A1 Q + A2 Q2 + . . . + An Qn .
Proof: Let C, A0 ∈ k[x] such that B = A0 + QC, deg A0 < d. By induction on the degree, there
are A1 , A2 , . . . ∈ k[x] of degree < d such that B = A0 + Q(A1 + A2 Q + A3 Q2 + . . .).
P
Definition: A rational fraction is simple if it is a monomial axn or Qn , where Q is irreducible
and deg P < deg Q.
P (x)
Theorem: Any rational fraction Q(x) with coefficients in a field k decomposes, uniquely up to
the order, as a sum of simple fractions.
Proof: If Q = Qn1 1 . . . Qnr r is the irreducible factor decomposition, then

P B1 Br
= n1 + . . . + nr
Q Q1 Qr
and there are polynomials Ai0 , Ai1 , . . . ∈ k[x], deg Aij < deg Qi , such that
Bi Ai0 + Ai1 Qi + Ai2 Q2i + . . . nPi −1 Aij
ni = ni = ni −j
+ Polynomial.
Qi Qi j=0 Qi
Now we prove the uniqueness. Given two decompositions

A B
n1 + . . . = + . . . , A 6= 0,
Q1 Qn1 1
where Qn1 1 is the greatest power of Q1 in both decomposotions (eventually B = 0), multiplying
by the greatest common multiple of all denominators we obtain an equality
(A − B)Qn2 2 . . . Qnr r = Q1 C.
Hence Q1 divides A − B and, since deg(A − B) < deg Q1 , we have A = B.
We conclude by induction on the number of terms of a decomposition.
3.3.3 Operator Theory

Theorem: Let T be an endomorphism of a k-vector space E. If P, Q ∈ k[x] are coprime,
Ker (P Q)(T ) = Ker P (T ) ⊕ Ker Q(T ).
Proof: Bézout’s identity 1 = AP + BQ shows that any vector e ∈ E decomposes
e = AP (T )e + BQ(T )e.
If P Q(T )e = 0, then AP (T )e ∈ Ker Q(T ), and BQ(T )e ∈ Ker P (T ).
Hence Ker P Q(T ) = Ker P (T ) + Ker Q(T ).
If e ∈ Ker P (T ) ∩ Ker Q(T ), then e = AP (T )e + BQ(T )e = 0 + 0.
Corollary: If T is an endomorphism of a complex vector space, and P ∈ C[x] is a polynomial

with complex roots α1 , . . . , αr of multiplicities m1 , . . . , mr , then
Ker P (T ) = Ker (T − α1 )m1 ⊕ . . . ⊕ Ker (T − αr )mr .
Difference Equations
Let E be the vector space of all complex sequences (s0 , s1 , . . .). Let ∇ : E → E be the operator
∇sn = sn+1 , and let ∆ = ∇ − 1 be the difference operator, ∆sn = sn+1 − sn .
Commutation Formula: P (∇)(αn sn ) = αn P (α∇)sn ; P ∈ C[x], α ∈ C.
Proof: We may assume that P (x) = xk , and we have
∇k (αn sn ) = αn+k sn+k = αn (α∇)k sn .
Corollary: The sequences αn , nαn , . . . , nk−1 αn form a base of Ker (∇ − α)k .
Proof: Since (∇ − α)k αn sn = αn (α∇ − α)k sn = αn+k ∆k sn , we have to show that Ker ∆k is
defined by all polynomial sequences a0 + a1 n + . . . + ak−1 nk−1 .
We proceed by induction on k, and we use that any sequence s = (sn ) is
n
n n
X n
(∆i s)0 .

sn = (∇ s)0 = (1 + ∆) s 0
=
i
i=0
n n(n−1)...(n−i+1)
If ∆k s = 0, then s is a polynomial sequence of degree < k, because

i = i! is
a polynomial function of n of degree i,
n n
(∆k−1 s)0 .

sn = s0 + 1 (∆s)0 + . . . + k−1
Finally, ∆d+1 nd = 0 because we have ∆d+1 nd = ∆d (∆nd ) = ∆d ((n + 1)d − nd ) = 0, and

(n + 1)d − nd is a polynomial function of degree d − 1.
Particular Solution: The complex roots of P solve the equation P (∇)xn = 0, and solutions
of P (∇)xn = yn are obtained adding a particular solution. Iterating, the search of a solution is
reduced to the equation (∇ − α)xn = yn . By the commutation formula,
yn = (∇ − α)xn = (∇ − α)αn α−n xn = αn+1 ∆(α−n xn ),
n−1
and ∆(v0 + . . . + vn−1 ) = vn . Hence a solution of (∇ − α)xn = yn is xn = αn−1 α−i yi .
P
i=0
In case that Q(∇)yn = 0 for some polynomial Q(x) coprime to P (x), Bézout’s identity
Id = P (∇)A(∇) + B(∇)Q(∇) shows that xn = A(∇)yn is a solution of P (∇)xn = yn .
Example: Fibonacci’s sequence 0, 1, 1, 2, 3, 5, 8, 13,... √satisfies the equation xn+2 = xn+1 + xn ,

and the roots of x2 − x − 1 are φ, −φ−1 , where φ = 1+2 5 is the golden ratio. Hence
xn = c1 φn + c2 (−φ)−n ,
and the constants c1 , c2 may be determined with the initial terms

c1 + c2 = x0 = 0 1 1
c1 = = √ , c2 = −c1 .
c1 φ − c2 φ−1 = x1 = 1 φ + φ−1 5
Differential Equations
Let E be the complex vector space of all infinitely derivable functions f : R → C and let D be
the derivative operator, Df = D x(t) + y(t)i = x (t) + y 0 (t)i.
0

Commutation Formula: P (D)(eαt f ) = eαt P (D + α)f ; P ∈ C[x], α ∈ C.
Proof: We may assume that P (x) = xk , and we proceed by induction on k.

If k = 1, then D(eαt f ) = αeαt f + eαt Df = eαt (D + α)f . In general,
Dk (eαt f ) = D(Dk−1 eαt f ) = D(eαt (D + α)k−1 f )

= αeαt (D + α)k−1 f + eαt D(D + α)k−1 f = eαt (D + α)k f.
Corollary: The functions eαt , teαt , . . . , tk−1 eαt form a base of Ker (D − α)k .
Proof: (D − α)k (eαt f ) = eαt Dk f , and Ker Dk = h1, t, . . . , tk−1 i.
Particular Solution: The complex roots of P solve the equation P (D)f = 0, and the solutions
of P (D)f = g are obtained adding a particular solution. Iterating, the search of a solution is
reduced to the equation (D − α)f = g. By the commutation formula,
(D − α)f = (D − α)eαt e−αt f = eαt D(e−αt f ),
and a particular solution of (D − α)f = g is just f (t) = eαt e−αt g(t)dt .

R
We may also use the decomposition as a sum of simple fractions

1 X aij
f (t) = g= g
(D − α1 )m1
. . . (D − αr )m r (D − αi )j
i,j
ZZ Z
αi t 1
X X
−αi t
= aij e e g= aij e αi t
. . . e−αi t g(t) dt.
j
Dj
i,j i,j
3.3.4 Root Separation

P (x)
Definitions: Let P, Q ∈ R[x] be coprime. We say that f (x) = Q(x) has a pole of order m at
x = a if Q has a root of multiplicity m at this point, and that the excess of f at x = a is
 P
P +1 if Q switches from −∞ to +∞ at x = a.

Ea = −1 si Q P
switches from +∞ to −∞ at x = a.
Q 
0 otherwise.

The excess Eab f between a and b is defined to be the sum of excesses of f at all points of the
interval [a, b], assuming that the end points a, b are not poles of f .
The sign variation V (c1 , . . . , cn ) of a sequence is the number of sign changes, after eliminating
null terms, and the variation between a and b of some polynomials P1 , . . . , Pn is
Vab (P1 , . . . , Pn ) = V (P1 (a), . . . , Pn (a)) − V (P1 (b), . . . , Pn (b)).
1. Eab (Polynomial) = 0.
2. Eab (f1 + f2 ) = Eab f1 + Eab f2 , when f1 and f2 have no common pole.
3. Eab (λf ) = (sgn λ)Eab f , for all non null λ ∈ R.

4. Eab Q1 = 12 (sgn Q(b) − sgn Q(a)).

P
5. Eab Q + Eab Q b
P = Va (P, Q).
P Q
Proof: Only property 5 is not obvious. Since Q and P have no common pole,
2 2
P
Eab Q + Eab Q b P +Q b 1 1 b
P = Ea P Q = Ea P Q = 2 (sgn P (b)Q(b) − sgn P (a)Q(a)) = Va (P, Q).
Excess Calculation: Applying Euclid’s algorithm (changing the remainders sign)

R1
P = Q1 Q − R1 P
Q = Q1 − Q
P
Eab Q = −Eab RQ1
Q
Q = Q2 R1 − R2 R1 = Q2 − R2
R1 Eab RQ1 = −Eab R
R1
2
......... ......... .........

Rn−1
Rn−1 = Qn+1 Rn Rn = Qn+1 Eab RRn−1
n
=0
Eab Q Eab Q bP b R1 b Q b R2 b Rn b Rn−1
P = P + Ea Q + Ea Q + Ea R1 + Ea R1 + . . . + Ea Rn−1 + Ea Rn
= VaB (P, Q) + Vab (Q, R1 ) + . . . + Vab (Rn−1 , Rn ) = Vab (P, Q, R1 , . . . , Rn )

Number of roots of
Sturm’s Theorem: = Vab (P, P 0 , R1 , . . . , Rn ) ; P (a)P (b) 6= 0 .
P between a and b
Proof: Let P = (x − a1 )m1 . . . (x − ar )mr Q, where Q has no real root. Then
P0 m1 mr Q0
= + ... + + ·
P x − a1 x − ar Q
0 0 0
Since Eab QQ = 0, because QQ has no pole, Eab PP coincides with the number of roots of P
between a and b (counted without multiplicity).
Note: This proof, and the excess calculation, require that Q, R1 , . . . , Rn do not vanish at a
nor at b. If someone vanishes, we move a little a and b. First we remark that no consecutive
remainders may vanish at a (or b), since a would be a root of the g.c.d., hence of P . Moreover,
if Ri (a) = 0, since Ri−1 = Qi Ri − Ri+1 , then Ri−1 (a) and Ri+1 (a) have opposite sign.
Modifying the end points, a0 = a + ε, b0 = b + ε, so that no remainder vanishes and excesses
do not change, nor signs of non null remainders, we are in the former case,
0 0
Eab Q b Q b b
P = Ea0 P = Va0 (P, Q, R1 , . . . , Rn ) = Va (P, Q, R1 , . . . , Rn ).
3.3.5 Multiple Roots

Definition: The derivative of a polynomal P (x) = a0 + a1 x + . . . + an xn ∈ k[x] is
P 0 (x) = a1 + 2a2 x + . . . + nan xn−1 ,
where we put iai = ai + . .i . +ai ∈ k. It may be that deg P 0 < n − 1, since 1+ . .i . +1 may be
zero in k. For example, when k = Fp , then the derivative of P = xp + 1 is P 0 = 0.
The following equalities hold, since they hold when P = xi , Q = xj ,
(aP + bQ)0 = aP 0 + bQ0 ; a, b ∈ k

(P · Q) = P · Q + P · Q0 .
0 0
Now, if P (x) = c0 xn + . . . + cn = c0 (x − α1 ) . . . (x − αn ), we shall determine the sums

σr = α1r + . . . + αnr , and we agree that σ0 = n.
n n
P0 X 1 α2

X 1 αi
= = + 2 + 3i + . . .
P x − αi x x x
i=1 i=1
P0 σ0 σ1 σ2
and we obtain Girard Formula: = + 2 + 3 + ...
P x x x
nc0 xn−1 + . . . + cn−1 = (c0 xn + . . . + cn )(σ0 x−1 + σ1 x−2 + σ2 x−3 + . . .)
Equating the coefficients of xn−r−1 , we obtain
P r
P
(n − r)cr = ci σj = ci σr−i , 1 ≤ r ≤ n − 1,
i+j=r i=0
P n
P
0= ci σj = ci σr−i , r ≥ n,
i+j=r i=0
(
0 = c0 σr + c1 σr−1 + . . . + cr−1 σ1 + rcr , r≤n
Newton Formulae:
0 = c0 σr + c1 σr−1 + c2 σr−2 + . . . + cn σr−n , r≥n
Theorem: A root α of P ∈ k[x] is multiple if and only if it is a root of P 0 (x).
Proof: Let P = (x − α)m Q, where Q(α) 6= 0.

If m = 1, then P 0 = Q + (x − α)Q0 ; hence P 0 (α) = Q(α) 6= 0.
If m ≥ 2, then P 0 = m(x − α)m−1 + (x − α)m Q0 ; hence P 0 (α) = 0.
Corollary: The multiple roots of P (x) are just the roots of D(x) = g.c.d. P, P 0 .

Proof: Roots of D are roots of P and P 0 ; hence multiple roots of P .

Conversely, by Bézout’s identity D = AP + BP 0 , if α is a multiple root of P (hence a root
of P 0 ), then α is a root of D.
Corollary: All the roots of an irreducible polynomial P ∈ k[x] are simple, or P 0 (x) = 0.
Proof: D = g.c.d.(P, P 0 ) is 1 or P , since P is irreducible.

If D = 1, any root of P is simple by the former result.
If D = P , since it divides P 0 and deg P 0 < deg P , we have P 0 = 0.
Definition: Let A be a ring. The kernel of the unique ring morphism Z → A is an ideal dZ,
and d is the characteristic of A.
It is 0 when 1+ . n. . +1 6= 0, ∀n ≥ 1, and it is positive when 1+ . n. . +1 = 0 for some n ≥ 1.
Z, Q, R and C have null characteristic, and the characteristic of Z/nZ is n.

√
2
The formula x = −b± 2ab −4ac
for the roots of a quadratic polynomial ax2 + bx + c ∈ k[x] only
holds when the characteristic of the field k is not 2 (so that 2a 6= 0).
Null Characteristic
Theorem: If char k = 0, any root of an irreducible polynomial P ∈ k[x] is simple.
Proof: P 0 (x) 6= 0, because deg P 0 = deg P − 1.
Theorem: If char k = 0, any root of multiplicity m of P ∈ k[x] has multiplicity m − 1 in P 0 .
Proof: If α is a root of multiplicity m, then P = (x − α)m Q, where Q(α) 6= 0. Now
P 0 = (x − α)m−1 (mQ + (x − α)Q0 ),
where mQ(α) + (α − α)Q0 (α) = mQ(α) 6= 0, since m 6= 0 when char k = 0.
Descartes Rule: The number of positive roots r+ (P ) of a polynomial with real coefficients
P (x) = a0 + a1 x + . . . + an xn , counted with multiplicity, is bounded by the number of sign
changes V (P ) in the sequence of coefficients,
r+ (P ) ≤ V (a0 , a1 , . . . , an ),
and both coincide if all the roots of P (x) are real.
Proof: Eliminating the null roots, we may assume that a0 > 0, and we may also assume that
the rule holds for polynomials of degree < n, so that r+ (P 0 ) ≤ V (P 0 ).
(
V (P ) if a0 has the sign of the consecutive non-zero coefficient aj (I)
V (P 0 ) =
V (P ) − 1 if a0 and aj has opposite sign (II)
Let us compare r+ (P 0 ) and r+ (P ). If the positive roots of P are α1 < . . . < αr , with
multiplicities m1 , . . . , mr , then αi is a root of P 0 of multiplicity mi − 1 and, by Rolle’s theorem,
P 0 vanishes between any two consecutive roots, so that r+ (P 0 ) ≥ r+ (P ) − 1.
In case (II) the rule follows.
In case (I), the derivative P 0 has one more root between 0 and α1 , because P (0) = a0 > 0 and
the first non null derivative of P at x = 0 is positive. Hence r+ (P 0 ) ≥ r+ (P ) and we conclude.
The number of negative roots of P is r− (P ) = r+ (P̄ ), where P̄ (x) = P (−x).
We have r+ (P ) ≤ V (P ), r− (P ) ≤ V (P̄ ), and V (P ) + V (P̄ ) ≤ n.
If all the roots of P are real, then r+ (P ) + r− (P ) = n, since a0 6= 0.
Hence r+ (P ) = V (P ), and r− (P ) = V (P̄ ).
Positive Characteristic
Theorem: The characteristic of any domain is 0 or a prime number.
Proof: Let A be a domain of positive characteristic d = nm.

In A we have nm = 0; hence n = 0 or m = 0, and we conclude that n = d or m = d in Z.
Lemma3 : (a + b)p = ap + bp , when the characteristic is a prime number p.
Proof: The number pi = p(p−1)...(p−i+1)

i! is a multiple of p when 0 < i < p, since p does not
divide the denominator. Hence
p
p i p−i
(a + b)p = = ap + bp .
P
i a b
i=0
3
This lemma proves again Fermat’s congruence: np = (1 + . . . + 1)p = 1p + . . . + 1p = n in Fp .
3.4. RINGS OF FRACTIONS 85
Theorem: If Q ∈ Fp [x] is irreducible, then any root of Q(x) is simple.
Proof: If Q = i ai xi has a multiple root, then 0 = Q0 = i iai xi−1 . Hence ai = 0 when i is

P P
not a multiple of p, and Q is not irreducible by Fermat’s congruence,
Q(x) = a0 + ap xp + a2p x2p + . . . = (a0 + ap x + a2p x2 + . . .)p .
Example: Let k = F2 (t). The polynomial x2 − t is irreducible in k[x], since it has no root in k;
but any root α is multiple, because x2 − t = x2 − α2 = (x − α)2 .
3.4 Rings of Fractions

Definition: Let A be a ring. S ⊆ A is a multiplicative system if 1 ∈ S, and s, t ∈ S ⇒ st ∈ S.
We consider in A × S the equivalence relation, clearly reflexive and symmetric,
(a, s) ≡ (b, t) ⇔ there are u, v ∈ S such that au = bv, su = tv.
It is transitive: if (a, s) ≡ (b, t), (b, t) ≡ (c, r), there are u, v, u0 , v 0 ∈ S such that au = bv,
su = tv, bu0 = cv 0 , tu0 = rv 0 . Hence auu0 = bvu0 = cvv 0 , suu0 = tvu0 = rvv 0 , and (a, s) ≡ (c, r).
Definition: The localization AS of A at S is the quotient set with the ring structure
a b at + bs a b ab
+ = , · =
s t st s t st
a a au
where s is the class of (a, s). These operations are well-defined: if we replace s by su ,
au b (at + bs)u at + bs au b abu ab

+ = = , · = =
su t stu st su t stu st
In AS ; we have 0 = 01 , 1 = 11 , and − as = −a a
s . Moreover, s = 0 if and only if ua = 0 for some
a b
u ∈ S; hence s = t if and only if u(at − bs) = 0 for some u ∈ S.
The canonical ring morphism γ : A → AS , γ(a) = a1 , is the localization morphism, and
γ(s) = 1s is invertible in AS for all s ∈ S, the inverse being 1s .
Theorem: If A is a domain, then S = A − {0} is a multiplicative system, AS is a field (the

field of fractions of A) and γ : A → AS is injective.
Proof: S is a multiplicative system because 1 6= 0 and a, b 6= 0 ⇒ ab 6= 0.

If a1 = 0, then sa = 0, where s 6= 0; hence a = 0, so that γ is injective and AS 6= 0.
Now, if as 6= 0, then a 6= 0, and as ∈ AS is the inverse of as .
Universal Property: If f : A → B is a ring morphism and f (s) is invertible in B for all

s ∈ S, then there is a unique ring morphism ψ : AS → B such that ψ( a1 ) = f (a),
f
A /B f = ψγ
C
γ
ψ
AS
Proof: The unique possible morphism, ψ : AS → B, ψ( as ) = f (a)f (s)−1 , is well-defined,

−1
ψ au = f (a)f (u)f (s)−1 f (u)−1 = f (a)f (s)−1 .

su = f (au)f (su)
a
Theorem: If J is an ideal of AS , then I = A ∩ J = {a ∈ A : 1 ∈ J} is an ideal of A, and
J = IAS = { as ; a ∈ I}.
Proof: It is clear that A ∩ J is an ideal of A and that IAS ⊆ J.

Now, if bt ∈ J, then 1b = 1t bt ∈ J; hence b ∈ I and bt ∈ IAS .
Definition: A unique factorization domain is an integral ring A where any proper element
can be written as a product of irreducible elements, uniquely up to the order and units.
Euclid’s lemma holds in such rings: If an irreducible element p divides a product, then it
divides some factor (pA is a prime ideal). In fact, if bc = pa, writing b and c as a product of
irreducibles, some factor coincides, up to a unit, with p; hence b or c is a multiple of p.
In general, if a ∈ A divides a product bc and it has no common irreducible factor with b,
then it divides c. Moreover, Gauss lemma and its proof hold in A[x].
Proposition: Let A be a unique factorization domain. Any root of c0 xn + . . . + cn ∈ A[x],

c0 cn 6= 0, in the field of fractions of A is x = ab , where a ∈ A divides cn and b ∈ A divides c0 .
a
Proof: If x = b is a root, then 0 = c0 an + c1 an−1 b + . . . + cn−1 abn−1 + cn bn , and
cn bn = −a(c0 an−1 + . . . + cn−1 bn−1 ).
If a and b have no common irreducible factor, then a divides cn .
Analogously b divides c0 an , and we conclude that b divides c0 .
Lemma: If A is a domain, then so is A[x], and A[x]∗ = A∗ .
Proof: (an xn + . . .)(bm xm + . . .) = an bm xn+m + . . ., so that deg (P Q) = deg P + deg Q.

Now it is clear that A[x] is integral and that invertible polynomials have degree 0.
Theorem: If A is a unique factorization domain, then so is A[x].
Proof: We prove the existence of irreducible factor decompositions by induction on the degree,
and we put P = dQ, where the coefficients of Q have no common irreducible factor.
If Q is irreducible, P = dQ is a product of irreducibles. Otherwise Q = Q1 Q2 , where Q1 and
Q2 are products of irreducibles by induction, hence so is P = dQ1 Q2 .
To prove the uniqueness, we consider two irreducible factor decompositions,
p1 . . . pr P1 (x) . . . Ps (x) = q1 . . . qm Q1 (x) . . . Qn (x),
where pi , qj ∈ A; deg Pi , deg Qj ≥ 1. Let Σ be the field of fractions of A.
The ring Σ[x] is Euclidean, and all factors Pi , Qj are irreducible in Σ[x] by Gauss lemma.
Hence s = n, and reordering Qi = abii Pi (where ai , bi have no common irreducible factor);
bi Qi = ai Pi , and any irreducible factor of bi (resp. ai ) divides Pi (resp. Qi ), which is irreducible:
ai and bi are invertible, and p1 . . . pr = uq1 . . . qm , where u ∈ A is invertible.
Hence r = m, and reordering, pi = qi up to units.
Corollary: Z[x1 , . . . , xn ] and k[x1 , . . . , xn ] are unique factorization domains.
Example: Let us see the Vandermonde determinant,

1 1 ... 1

x1 x2 . . . xn Y

... = (xj − xi ).
... ... ...

xn−1 xn−1 . . . xn−1 1≤i<j≤n
1 2 n
3.5. THE RESULTANT 87
This determinant is a polynomial V (x1 , . . . , xn ) ∈ Z[x

Q 1 , . . . , xn ], and it vanishes in the quo-
tient by the ideal (xj − xi ); hence it is a multiple of i<j (xj − xi ), this ring being a unique
factorization domain, and both coincide up to a constant factor c because both polynomials
n(n−1) n

have degree 0 + 1 + 2 + . . . + (n − 1) = 2 = 2 .
Let us see that c = 1. The diagonal monomial of V (x1 , . . . , xn ) is xnn−1 . . . x23 x2 while, by
induction on n, the other term is
n−2
(xj − xi ) = (xnn−1 + . . .)(xn−1 . . . x2 + . . .) = xnn−1 . . . x2 + . . .
Q
(xn − x1 ) . . . (xn − xn−1 )
i<j<n
3.5 The Resultant

Definition: Let us express two polynomials P, Q ∈ k[x] in terms of their roots:
P = a0 xn + a1 xn−1 + . . . + an = a0 (x − α1 ) . . . (x − αn )
Q = b0 xm + b1 xn−1 + . . . + bm = b0 (x − β1 ) . . . (x − βm )
P and Q have a common root if and only if the resultant R(P, Q) vanishes,
R(P, Q) = am n Q
0 b0 i,j (αi − βj ) ∈ k.
1. If m = 0, then R(P, Q) = bn0 .
2. R(Q, P ) = (−1)nm R(P, Q).

R(Q, P ) = am n nm am bn nm R(P, Q).
Q Q
0 b0 i,j (βj − αi ) = (−1) 0 0 i,j (αi − βj ) = (−1)
3. R(P, Q) = am
Q
0 i Q(αi ).
R(P, Q) = a0 b0 i j (αi − βj ) = am
m n m
Q Q Q Q Q
0 i b0 j (αi − βj ) = a0 i Q(αi ).
Now we consider the polynomial ring Z[a0 , α1 . . . , αn , b0 , β1 , . . . , βm ].
4. R(P, Q) ∈ Z[a0 , . . . , an , b0 , . . . , bm ].
The resultant is a symmetric polynomial in αi , with coefficients in Z[a0 , b0 , . . . , bm ]; hence
it is a polynomial in the elementary symmetric functions of the roots αi , which are ± aa0i by
Cardano’s formulae:

(∗) R(P, Q) = F a0 , b0 , . . . , bm , aa10 , . . . , aan0 = T (a0 ,...,aanr,b0 ,...,bm )
0
T̄ (a0 ,...,an ,b0 ,...,bm )

(∗∗) R(P, Q) = ±R(Q, P ) = bs0
Now, Z[a0 , . . . , an , b0 , . . . , bm ] is a polynomial ring (p. 75); hence a unique factorization

domain, and comparing (∗) and (∗∗) we see that r = s = 0.
5. R(P, Q) is a homogeneous polynomial of degree m in the variables a0 , . . . , an , and homoge-

neous of degree n in the variables b0 , . . . , bm .
If we multiply the variables a0 , . . . , an with an indeterminate t, then we replace P by tP =
ta0 xn + ta1 xn−1 + . . . + tan = ta0 (x − α1 ) . . . (x − αn ), and
R(tP, Q) = (ta0 )m = tm R(P, Q),

Q
i Q(αi )
so that R(P, Q) is homogeneous of degree m in a0 , . . . , an . Since R(P, Q) = ±R(Q, P ), it is

homogeneous of degree n in b0 , . . . , bm .
Theorem: If P = CQ + R, then R(P, Q) = (−1)nm b0n−r R(Q, R), where r = deg R.
Proof: Since P (βj ) = Q(βj )C(βj ) + R(βj ) = R(βj ), we have
R(Q, P ) = bn0 = bn0 = bn−r

Q Q
j P (βj ) j R(βj ) 0 R(Q, R) .
Example: The discriminant ∆ of a polynomial P = xn + a1 xn−1 + . . . + an is

n Q n Q n
(αj − αi )2 = (−1)( 2 ) (αj − αi ) = (−1)( 2 ) P 0 (αi ) = (−1)( 2 ) R(P, P 0 ).
Q
∆=
i<j i,j i

1
n−1

1 ... 1 1 α1 ... α1 n σ1 . . . σn−1
α2n−1

α1 α 2 ... αn 1 α2 ... σ σ2 ... σn
= 1

∆ = ·
. . . . . . ... . . . . . . ... ... ... ... ...
... . . .
αn−1 αn−1 ... αnn−1 1 αn . . . αnn−1 σn−1 σn . . . σ2n−2
1 2
Proposition: P and Q have no common root if and only if g.c.d.(P, Q) = 1.
Proof: If they are coprime, then 1 = AP + BQ, so that they have no common root.
Otherwise, any root of g.c.d.(P, Q) is a common root.
Definition: The Bézout resultant is Rb (P, Q) = am

0 (det hQ ), where
hQ : k[x]/(P ) −→ k[x]/(P ), hQ (B) = BQ.
Theorem: Rb (P, Q) = R(P, Q).
Proof: If we consider L[x]/(P ) instead of k[x]/(P ), the determinant of hQ is the same, because
so is the matrix of hQ in the base 1, x, . . . , xn−1 .
L[x]/(P ) ' L[x]/(x − α1 ) ⊕ . . . ⊕ L[x]/(x − αn ) ' L ⊕ . . . ⊕ L
where the isomorphism L[x]/(x − αi ) ' L replaces x by αi . Hence the class [Q] corresponds to
(Q(α1 ), . . . , Q(αn )) and, in the usual base of Ln , the matrix of hQ is diagonal,
Rb (P, Q) = am m
0 (det hQ ) = a0 Q(α1 ) . . . Q(αn ) = R(P, Q).
Definition: The condition (P, Q) = k[x] means that the linear map
f : k[x]/(Q) ⊕ k[x]/(P ) −→ k[x]/(P Q), f (A, B) = AP + BQ,
is surjective. Both vector spaces have dimension m + n, and the Euler resultant is the deter-
minant of the matrix of f when we consider the base (xm−1 , 0), . . . , (1, 0), (0, xn−1 ), . . . , (0, 1) in
k[x]/(Q) ⊕ k[x]/(P ), and the base xn+m−1 , . . . , 1 in k[x]/(P Q),

a0 a1 . . . ... an 0 ... ... 0

0 a0 a1 ... ... an 0 ... 0

. . . . . . . . . ... ... ... . . . . . . . . .

e
0 0 ... ... ... ... . . . an−1 an
R (P, Q) =
b 0 b1 ... ... bm 0 ... ... 0
0 b0 b1 ... ... bm 0 ... 0

. . . . . . . . . ... ... ... . . . . . . . . .

0 0 ... ... ... ... . . . bm−1 bm
3.5. THE RESULTANT 89
where m rows have the coefficients of P and n rows those of Q.

Clearly Re (P, Q) = 0 if and only if P and Q have a common root.
When we consider a0 , a1 . . . , an , b0 , b1 , . . . , bm as indeterminates, Re (P, Q) is a homogeneous
polynomial of degree m in a0 , . . . , an , and homogeneous of degree n in b0 , . . . , bm .
Lemma: R(P, Q) is not a multiple of b0 in Z[a0 , . . . , an , b0 , . . . , bm ] when m 6= 0.
Proof: Let Q̄(x) = b1 xm−1 + . . . + bm , so that Q = b0 xm + Q̄.

If R(P, Q) is a multiple of b0 , then we have
0 = R(P, Q) = am mQ
Q
0 i Q(αi ) = a0 i Q̄(αi ) = a0 R(P, Q̄),
in the polynomial ring Z[a0 , α1 , . . . , αn , b1 , . . . , bm ] = Z[a0 , α1 , . . . , αn , b0 , . . . , bm ]/(b0 ).

Hence R(P, Q̄) = 0, and any two polynomials of degrees n, m − 1 would have null resultant.
Theorem: Re (P, Q) = R(P, Q).
Proof: Let us see Re (P, Q) in the polynomial ring Z[a0 , α1 , . . . , αn , b0 , β1Q

, . . . , βm ].
It vanishes in the quotient by (αi − βj ); hence Re is a multiple of i,j (αi − βj ), since the
ring is a unique factorization domain, so that am n e
0 b0 R (P, Q) is a multiple of the resultant,
am n e
0 b0 R (P, Q) = F · R(P, Q).
F is a symmetric function of the roots αi and the roots βj , since so are R and Re .
b
Hence F is a polynomial in the elementary symmetric functions which are ± aa0i and ± b0j by
Cardano’s formulae. Eliminating denominators, we obtain
ar0 bs0 Re (P, Q) = F̄ · R(P, Q)
and the above lemma shows that Re is a multiple of R. Since both polynomials have degree m
in a0 , . . . , an and degree n in b0 , . . . , bm , we have Re = cR for some constant c.
Let us see that c = 1. A monomial in Re is the diagonal am n
0 bm , while
Y
R(P, Q) = am Q(αi ) = am m n
Q
0 0 i (bm + bm−1 αi + . . .) = a0 bm + . . .
i
Elimination
Let us consider a system of algebraic equations with complex coefficients
(
0 = P (x, y) = a0 (y)xn + a1 (y)xn−1 + . . . + an (y)
0 = Q(x, y) = b0 (y)xm + b1 (y)xm−1 + . . . + bm (y)
Let R(y) be the resultant of P and Q, considered as polynomials in x with coefficients in

C(y). Since the resultant is a polynomial in the coefficients, R(y) ∈ C[y].
Theorem: The roots of R(y) are just the common roots of a0 (y) and b0 (y), and the ordinates
of the solutions of the given system of algebraic equations.
Proof: If a0 (β) = b0 (β) = 0, clearly β is a root of the Euler resultant.

If a0 (β) 6= 0 (and analogously if b0 (β) 6= 0), the Bézout resultant shows that R(β) is the
determinant of the endomorphism

hQ(x,β) : C[x]/ P (x, β) −→ C[x]/ P (x, β) .

Hence R(β) = 0 if and only if Q(x, β) is not invertible in C[x]/ P (x, β) , a condition stating
that Q(x, β) and P (x, β) have a common root α ∈ C: the system has a solution x = α, y = β.
Theorem: If P (x, y) and Q(x, y) have no common irreducible factor, then the above system of
algebraic equations only has a finite number of complex solutions.
Proof: According to Gauss lemma, both polynomials have no common irreducible factor in
C(y)[x]; hence the resultant R(y) is not 0, and it has a finite number of complex roots.
The complex solutions of the system have a finite number of possible ordinates.
Analogously, replacing y by x, they have a finite number of possible abscissas.
The system has a finite number of complex solutions.
Bézout’s Theorem: Let P, Q ∈ C[x, y] be polynomials of degrees n and m, with no common

irreducible factor. Then nm bounds the number of complex solutions of

P (x, y) = 0
Q(x, y) = 0
Proof: Fixing suitable axes we may assume that the solutions have different ordinates (so that
the degree of the resultant R(y) bounds the number of solutions) and that
P (x, y) = a0 xn + a1 (y)xn−1 + . . . + an (y),

Q(x, y) = b0 xm + b1 (y)xm−1 + . . . + bm (y).
The coefficients pij (y) of the Euler resultant are polynomials of degree
deg pij ≤ n + i − j, 1 ≤ i ≤ n,
deg pm+i,j ≤ m + i − j, 1 ≤ i ≤ n.
For the terms p1,σ(1) . . . pm+n,σ(m+n) of the Euler resultant we have
deg p1,σ(1) ≤ n + 1 − σ(1)

... ......
deg pm,σ(m) ≤ n + m − σ(m)
deg pm+1,σ(m+1) ≤ m + 1 − σ(m + 1)
... ......
deg pm+n,σ(m+n) ≤ m + n − σ(m + n)
so that the degree of any term is bounded above by

P
(n + 1)+ . . . + (n + m) + (m + 1) + . . . + (m + n) − i σ(i) =

= (n + 1) + . . . + (n + m) + (m + 1) + . . . + (m + n) − 1 + . . . + (m + n)
= ((n + 1) + (n + m)) m n m+n
2 + ((m + 1) + (m + n)) 2 − (1 + (m + n)) 2 = nm.
Chapter 4
Analysis II
4.1 Topological Spaces

Definitions: Recall that a topology T on a set X is any family of subsets (named open sets)
such that ∅ and X are open sets, and arbitrary unions and finite intersections of open sets are
open sets. Given two topologies T , T 0 on X, if T ⊆ T 0 , we say that T is coarser than T 0 or
that T 0 is finer than T . T
Let {Ti } be a family of topologies on X. Then T = i Ti also is a topology on X.
Hence, any family of subsets {Ui } of X generates a topology: the coarsest topology (the
intersection of the topologies) where all the sets Ui are open. The open sets are the arbitrary
unions of finite1 intersections Ui1 ∩ . . . ∩ Uin .
A family {Vi } of neighborhoods of a point x ∈ X is a base of neighborhoods of x when
for any other neighborhood V of x, we have Vi ⊆ V for some index i. A family {Ui } of open
sets is a base of the topology when any other open set is a (possibly infinite) union of open sets
in the family ({Ui : x ∈ Ui } is a base of neighborhoods of x, for any point x ∈ X).
Let X be a set. Given a family of topological spaces {Yi } and maps fi : Yi → X, the final
topology is the finest topology on X such that all the maps fi are continuous: U ⊆ X is open
when fi−1 (Ui ) is an open set in Yi for any index i.
Given maps fi : X → Yi , the initial topology is the coarsest topology on X such that all
the maps fi are continuous: it is the topology generated by the subsets fi−1 (V ), where V ⊆ Yi
is open. In particular, when X is a subset of a topological space Y , the induced topology is just
the initial topology of the inclusion j : X ,→ Y .
Q
Definitions: Given a family of topological spaces {Xi }, the direct Q product i Xi always is
endowed with the initial topology of the natural projections
Q pi : i Xi → Xi , so that a basis of
the topology is defined by the products of open sets i Ui , where Ui = Xi up to a finite number
of indices, and for any topological space Y we have that
Q Q
HomTop (Y, i Xi ) = i HomTop (Y, Xi ) , f 7→ (f ◦ pi ).
The disjoint union qi Xi always

` is endowed with the final topology of the natural inclusions
ji : Xi → i Xi , so that U ⊆ i Xi is open when ji−1 (U ) is open for any index i. For any
`
topological space Y we have that
Q
HomTop (qi Xi , Y ) = i HomTop (Xi , Y ) , f 7→ (ji ◦ f ).
If R ⊆ X × X is an equivalence relation on a topological space X, the quotient set X/R

always is endowed with the final topology of the canonical projection π : X → X/R, so that
1 T S
By definition, i∈I Ui = X and i∈I Ui = ∅ when the set of indices I is empty.
91
92 CHAPTER 4. ANALYSIS II
Ū ⊆ X/R is open when π −1 (Ū ) is open, and for any topological space Y we have an exact
sequence of sets (in the sense that the first map is injective and the image is just the subset
where both right arrows coincide):
π p1 ,p2
HomTop (X/R, Y ) −−→ HomTop (X, Y ) ⇒ HomTop (X, R) .
Examples: The product topology in Rn is the topology induced by the usual metric.
Any point x of a metric space X admits a countable base of neighborhoods B x, n1 . If

moreover X has a countable dense set {xm }, then X admits a countable base B xm , n1 of open
sets. In particular, the topology of Rn (hence of any subspace) admits a countable base.
Proposition: Let X be a topological space such that any point has a countable base of neigh-
borhoods. A point x ∈ X is in the closure of Y ⊂ X if and only if there is a sequence (yn ) in Y
which converges to x.
Proof: If x ∈ Ȳ , take a countable base {Un } of neighborhoods of x. Replacing Un by U1 ∩. . .∩Un ,

we may assume that Un+1 ⊆ Un . Pick yn ∈ Un ∩ Y . It is clear that (yn ) converges to x.
Proposition: Let X be a topological space such that any point has a countable base of neighbor-
hoods. A map f : X → Y is continuous if and only if it preserves limits: lim f (xn ) = f (lim xn )
for any convergent sequence (xn ) in X.
Proof: Continuous maps always preserve limits. Conversely, if f is not continuous at a point
x ∈ X, there is a neighborhood V of f (x) in Y such that f −1 (V ) is not a neighborhood of x.
Take a countable base {Un }, Un+1 ⊆ Un , of neighborhoods of x. For any n ∈ N we have a
point xn ∈ Un such that f (xn ) ∈
/ V , so that xn → x, while f (xn ) does not converge to f (x).
Proposition: If any point of a compact space K has a countable base of neighborhoods, then
any sequence (xn ) in K has a convergent subsequence.
Proof: Take an adherent point x ∈ X and consider a countable base {Um } of neighborhoods of
x. We have xn0 ∈ U0 for some index n1 , and xn1 ∈ U1 for some index n1 > n0 , and so on.
So we obtain a subsequence (xn0 , xn1 , xn2 . . .) which converges to x.
Proposition: A topological space X with a countable base of open sets is compact if and only if
any sequence has a convergent subsequence.
S
Proof: Assume that X admits an open cover X = α Uα with no finite subcover.
Since Uα is a union of open sets in the countable base, we see that we may assume that it is
a countable cover. Pick xn ∈ X − (U1 ∪ . . . ∪ Un ).
If (xn ) admits a convergent subsequence xni → x ∈ X, we have x ∈ Um for some index m.
Now, Um contains all the terms xni up to a finite number. Absurd when ni ≥ m.
Definitions: A topological space X is T0 when for any two points x 6= y there is an open set
containing one of them but not the other; i.e. x̄ 6= ȳ.
X is T1 when for any two points x 6= y there are two open sets, one containing x but not y,
and the other containing y but not x; i.e. when any point is a closed set.
X is T2 , separated or Hausdorff when any two points x 6= y have disjoint neighborhoods.
X is T3 or regular when any closed set Y and any point x ∈/ Y have disjoint neighborhoods:
at any point the closed neighborhoods form a basis of neighborhoods.
X is T4 or normal when any two disjoint closed sets have disjoint neighborhoods.
4.1. TOPOLOGICAL SPACES 93
It is clear that (T4 + T1 ) ⇒ (T3 + T1 ) ⇒ T2 ⇒ T1 ⇒ T0 .
Proposition: In a separated space X, the limit of a sequence (xn ) is unique, if it exists, and
any compact set K ⊆ X is closed.
Proof: Given two limits x 6= y of (xn ), any neighborhood of x or y contains all the points xn
when n 0. Absurd since x and y have disjoint neighborhoods.
Finally, if K ⊆ X is compact, and x ∈ X − K, for each point y ∈ K we may find disjoint
open neighborhoods
S x ∈ Uy and y ∈ Vy .
Then K ⊆ y Vy , so that K admits a finite subcover, K ⊆ Vy1 ∪ . . . ∪ Vyn .
Hence x ∈ Uy1 ∩ . . . ∩ Uyn and (Uy1 ∩ . . . ∩ Uyn ) ∩ K = ∅, so that X − K is open.
Corollary: Any continuous bijection f : K → K 0 between compact separated spaces is a home-

omorphism.
Proof: If Y ⊆ K is closed, then it is compact (p. 23), so that f (Y ) is compact.

Hence f (Y ) is closed, because K 0 is separated, and we conclude.
Theorem: The diagonal ∆ = {(x, x)} is closed in X × X if and only if X is separated.
Proof: ∆ is closed if and only if for any point (x, y) ∈

/ ∆ there are neighborhoods Ux and Uy of
x and y respectively, such that (Ux × Uy ) ∩ ∆ = ∅; i.e. Ux ∩ Uy = ∅.
Q
Corollary: Any direct product i Xi of separated spaces is separated.
Q Q Q Q
Proof: If ∆i is closed in Xi × Xi , the i ∆i is closed in i (Xi × Xi ) = ( i Xi ) × ( i Xi ).
Corollary: Let X be a separated topological space. If two continuous maps f, h : T → X coincide

on a dense subset of T , then f = h.

Proof: Let us consider the continuous map (f, h) : T → X × X, (f, h)(t) = f (t), h(t) .
Then {t ∈ T : f (t) = h(t)} = (f, h)−1 (∆) is closed in T and contains a dense set: it is T .
Proposition: Any compact separated space K is normal.
Proof: Let X, Y be disjoint closed (hence compact) sets and fix a point x ∈ X.
For any point y ∈ Y we have disjoint neighborhoods x ∈ Uy , y ∈ Vy .
If Y ⊆ Vy1 ∪ . . . ∪ Vyn , then Vx = Vy1 ∪ . . . ∪ Vyn and Ux = Uy1 ∩ . . . ∩ Uyn are disjoint
neighborhoods of Y and x respectively. If X ⊆ Ux1 ∪ . . . ∪ Uxm , then Ux1 ∪ . . . ∪ Uxm and
Vx1 ∩ . . . ∩ Vxm are disjoint neighborhoods of X and Y respectively.
Proposition: Given disjoint closed sets Y, Z in a metric space X, there is a continuous function
f : X → [0, 1] such that Y = f −1 (0), Z = f −1 (1). In particular, any metric space is a normal
T1 space.
Proof: The functions g(x) = d(x, Y ) and h(x) = d(x, Z) are continuous (p. 25). Just put
f = g/(g + h).
Definition: Let X be a topological space. A connected component of X is a maximal

connected subspace, and X is locally connected if any point has a base of connected neigh-
borhoods.
Theorem: (1) If a set C ⊆ X is connected, then the closure C̄ also is connected.

S
(2) If some connected sets Ci ⊆ X have non-empty intersection, then i Ci is connected.
(3) Any point of X lies in a unique connected component of X.
(4) Any connected component of X is closed and, if X is locally connected, it is also open.
Proof: (1) Let U be a non-empty open closed set in C̄. Then U ∩ C is non-empty, because C is
dense in C̄; hence U ∩ C = C, and we conclude
S that U = C̄ because U is T closed.
(2) Now let U be an open closed set in i Ci and pick a point x ∈ i Ci . We may assume
that x ∈ U (otherwise replace U by U c ). Now Ci ∩ U is a non-empty
S open closed set in Ci ;
hence Ci ∩ U = Ci for any index i, and we conclude that U = i Ci .
(3) Given x ∈ X, the union of all connected sets Ci ⊆ X containing x is connected by (2),
and it is the unique connected component of X containing x.
(4) If C is a connected component of X, then C ⊆ C̄ and C̄ is connected by (1); hence
C = C̄.
If moreover X is locally connected and U is a connected neighborhood of x ∈ C, then U ∪ C
is connected; hence U ∪ C = C, so that U ⊆ C and we see that C is open.
Definition: A topological space is locally compact if any point has a compact neighborhood,
and it is σ-compact if moreover it is separated and with a countable base of open sets. A
subset of a topological space is relatively compact if it has compact closure.
Proposition: In a locally compact separated space, any point has a base of compact neighbor-
hoods.
Proof: We may assume that X is a compact separated space.

If U is a neighborhood of x ∈ X, then x and U c have disjoint open neighborhoods, x ∈ U1 ,
U ⊆ U2 , so that U2c ⊆ U is a closed (hence compact) neighborhood of x.
c
S
Proposition: Any σ-compact space X admits a countable cover by compact sets, X = n Kn ,
such that Kn is contained in the interior of Kn+1 .
S
Proof: Let X = n Un be a countable cover by relatively compact open sets.
Put K1 = U 1 . Given Kn−1 , it admits a finite open cover, Kn−1 ⊆ Ui1 ∪ . . . ∪ Uir , and we
put Kn = U i1 ∪ . . . U ir ∪ U n .
S
Proposition: If X is a connected and locally connected σ-compact space, then X = n Kn ,
where Kn is a compact set contained in the interior of Kn+1 , and X − Kn has no relatively
compact connected component.
Proof: If K ⊂ X is compact, let K [ be the union of K with the relatively compact connected
components of X − K. Then K [ is closed, because the connected components of X − K [ are
just the non relatively compact components of X − K, and moreover K [ is compact:
Let V be a relatively compact open set containing K, and U a connected component of
X − K. Since X is connected, U ∩ V 6= ∅ (if U ⊆ X − V , then U = Ū because ∂U ⊆ K).
Since U is connected, U ⊆ V or U ∩ (∂V ) 6= ∅ (otherwise U ⊆ V q (X − V̄ )). Only a finite
number U1 , . . . , Un of relatively compact components intersect the compact set ∂V , so that
K [ ⊆ U1 ∪ . . . ∪ Un ∪ V is relatively compact; hence compact, since it is closed.
Finally, put X = n Qn , with Qn compact and Qn ⊆ Qn+1 . Take K0 = Q[0 and, inductively,
S
Kn = L[n , where Ln is a compact set containing Kn−1 ∪ Qn−1 in the interior.
Examples: Any open set in Rn is a locally compact and locally connected space; hence any
topological manifold X of dimension n (any point of X has an open neighborhood homeo-
morphic2 to an open set in Rn ) also is locally compact and locally connected.
Hence any open or closed subset of Rn is a σ-compact space.
Any separated topological manifold with a countable base of open sets is σ-compact.
Any set X, with the discrete topology (any subset is an open set), is separated, locally
connected and locally compact, but not σ-compact when |X| > ℵ0 .
Let X be a locally compact non-compact separated space. If we consider on the disjoint union
X ∗ := X ∪ {∞} the topology such that X is an open subspace and the open neighborhoods of
∞ are the complements of the compact sets K ⊂ X, then X ∗ is a compact separated space, the
one-point compactification of X.
Theorem: Any open subset U of a compact separated space K is Baire.
Proof: Replacing U by Ū , we may assume that U is dense; hence, we only have to show that
K is Baire. Let {Un } be a countable Tfamily of dense open sets in K, and let us see that any
non-empty open set V ⊆ K intersects n Un .
Since U1 ∩ V 6= ∅ and K is regular, there is a non-empty open set V1 with V̄1 ⊆ U1 ∩ V .
Recursively, we may find non-empty open sets Vn withT V̄n ⊆ Vn−1 ∩TUn . Since {V̄n } is a
decreasing sequence of non-empty closed sets, there is x ∈ n V̄n ⊆ V ∩ n Un .
Corollary: Any locally compact separated space X is Baire.
Proof: X is an open subset of the one-point compactification X ∗ .

On any real affine space (A, E) we consider the initial topology of the affine functions A → R
(the finest topology where the affine functions are continuous). If we consider the coordinates
in an affine reference system, the bijection (x1 , . . . , xn ) : A → Rn is in fact a homeomorphism,
since any affine function a1 x1 + . . . + an xn + b is continuous in Rn with the product topology
+ a·
(because the maps R × R − → R and R − → R clearly are continuous).
In this section (A, E), (A0 , E 0 ) will be Euclidean spaces of dimension n and m respectively
and ϕ : U → V is a map, where U and V are open sets in A and A0 respectively.
Definitions: We say that ϕ is differentiable at a point p ∈ U when there is a linear map

ϕ∗,p : E → E 0 (named tangent linear map of ϕ at p) such that
ϕ(p + e) = ϕ(p) + ϕ∗,p (e) + O(e)kek , p + e ∈ U,
for some map O : E → E 0 defined in an open neighborhood of 0, continuous at 0 and such that
O(0) = 0 (hence ϕ is continuous at p).
We say that ϕ is differentiable, when so it is at any point (in particular ϕ is continuous).
In the case of a differentiable function f : U → R, we say that the tangent linear map is the
differential of f at p. It is a 1-form that we denote dp f : E → R.
When ϕ is an affine map, the tangent linear map at any point is just ϕ ~.
Hence, if we fix affine coordinates (x1 , . . . , xn ) in A (hence a basis e1 , . . . , en in E), then
dp x1 , . . . , dp xn is just the dual basis, (dp xi )(ej ) = δij .
2
This definition only is sensible if an open set in Rn can not be homeomorphic to an open set in Rm when
n 6= m, a fact that will be proved in p. 233.
If it exists, the tangent linear map is unique, because for any vector v ∈ E we have
ϕ(p + tv) − ϕ(p)
ϕ∗,p (v) = lim , (t ∈ R). (4.1)
t→0 t
In the case of a function f : U → R, this limit is named directional derivative (∂v f )(p) of
f at the point p in the direction v:
f (p + tv) − f (p) d(f ◦ σ)
(dp f )(v) = (∂v f )(p) = lim = , σ(t) = p + tv .
t dt

t→0 t=0
When we fix affine coordinates (x1 , . . . , xn ) in A, the derivatives in the directions e1 , . . . , en

are said to be the partial derivatives of f at the point p:
∂f f (x1 , . . . , xi + t, . . . , xn ) − f (x1 , . . . , xi , . . . xn )
(∂i f )(p) = (p) = (dp f )(ei ) = lim ,
∂xi t→0 t
∂f ∂f
dp f = (p)dp x1 + . . . + (p)dp xn .
∂x1 ∂xn
In the general case, if we also fix affine coordinates (y1 , . . . , ym ) in A0 , we have

ϕ(x1 , . . . , xn ) = f1 (x1 , . . . , xn ), . . . , fm (x1 , . . . , xn )
for some functions f1 , . . . , fm : U → R, and we say that yi = fi (x1 , . . . , xn ) are the equations of
ϕ in the fixed coordinate systems. The map ϕ is differentiable at a point p if and only if so are
all the components f1 , . . . , fm , because a map O : E → Rm is continuous and vanishes at 0 if
∂f1
and only if so do the components. In such case, 4.1 states that ϕ∗,p (ej ) = ∂x j
(p), . . . , ∂fm
∂xj (p) ,
so that the matrix of the tangent linear map ϕ∗,p in the fixed bases of E and E 0 is just the
∂fi
jacobian matrix (and the determinant | ∂x j
(p)| is named the jacobian of ϕ at p):
 
∂f1 ∂f1
∂x1 (p)
... ∂xn (p)
ϕ∗,p =  ... ... ... 
 
∂fm ∂fm
∂x1 (p) . . . ∂xn (p)
Chain’s Rule: Let us consider maps ϕ : U → V and φ : V → W , where W is an open set of

another Euclidean space A00 . If ϕ is differentiable at a point p ∈ U and φ is differentiable at
q = ϕ(p) ∈ V , then φ ◦ ϕ is differentiable at p and (φ ◦ ϕ)∗,p = φ∗,q ◦ ϕ∗,p .
Proof: We have ϕ(p + e) = q + ϕ∗ (e) + O1 (e)kek and φ(q + v) = φ(q) + φ∗ (v) + O2 (v)kvk, where
O1 and O2 are continuous and vanish at 0. Therefore:

(φϕ)(p + e) = φ q + ϕ∗ (e) + O1 (e)kek

= φ(q) + φ∗ ϕ∗ (e) + O1 (e)kek + O2 ϕ∗ (e) + O1 (e)kek kϕ∗ (e) + O1 (e)kekk
= (φϕ)(p) + (φ∗ ϕ∗ )(e) + O3 (e)kek ,
where O3 (e) = φ∗ O1 (e) + O2 ϕ∗ (e) + O1 (e)kek kϕ∗ (e)+O 1 (e)kekk

kek . Now,
kϕ∗ (e) + O1 (e)kekk ≤ kϕ∗ (e)k + kO1 (e)k · kek ≤ ckek + kO1 (e)k · kek
for some constant c, and we see that O3 is continuous and vanishes at 0. We conclude.
Definition: A differentiable map ϕ : U → V is of class C 1 when ϕ0 : U → Hom(E, E 0 ), p 7→ ϕ∗,p ,

is continuous. Inductively, ϕ is of class C m if it is differentiable and ϕ0 is of class C m−1 . We say
that ϕ is of class C 0 when it is continuous, and of class C ∞ when it is of class C m , ∀m ∈ N.
A map ϕ = (f1 , . . . , fm ) is of class C m if and only if so are the functions f1 , . . . , fm .
Theorem: If ϕ : U → V and φ : V → W are maps of class C m , then so is φ ◦ ϕ.
Proof: We proceed by induction on m, and it is clear when m = 0.

When m ≥ 1, the chain’s rule states that (φϕ)0 is of class C m−1 , since it is just the composition
of the following maps of class C m−1 :
1×ϕ ϕ0 ×φ0 ◦
U −−−−−→ U × V −−−−−→ Hom(E, E 0 ) × Hom(E 0 , E 00 ) −−−→ Hom(E, E 00 ) .
∂f
Theorem: A function f : U → R is of class C 1 if and only if the partial derivatives ∂x i
exist
m
and are continuous. Hence f is of class C if and only if the partial derivatives of order ≤ m
exist and are continuous.
Proof: If f is differentiable, then all the partial derivatives exist, and if f is of class C 1 , then the
∂f ∂f
map ( ∂x 1
, . . . , ∂xn
) : U → Rn is continuous.
Conversely, assume that the partial derivatives exist and are continuous.
We shall give the proof in the case n = 2, since the generalization is obvious.
In a ball with center at p = (a, b) and contained in U , by the mean value theorem

f (a + s, b + t) − f (a, b) = f (a + s, b + t) − f (a, b + t) + f (a, b + t) − f (a, b)
= ∂1 f (a + ξ, b + t) · s + ∂2 f (a, b + ξ 0 ) · t , |ξ| ≤ |s|, |ξ 0 | ≤ |t| ;
f (a + s, b + t) − f (a, b) − ∂1 f (a, b) · s − ∂2 f (a, b) · t =
= [∂1 f (a + ξ, b + t) − ∂1 f (a, b)] s + ∂2 f (a, b + ξ 0 ) − ∂2 f (a, b) t ,

√
and this function, even if divided by s2 + t2 , is continuous and vanishes at s = t = 0, because
both partial derivatives are continuous. Hence f is differentiable.
Finally, U → E ∗ , p 7→ dp f , is continuous because so are the partial derivatives.
Theorem: Any affine function f is differentiable, and df = f~. Sums and products of differen-
tiable functions also are differentiable and, if a differentiable function f does not vanish at any
point of U , then f1 also is differentiable. Moreover
dp (f + g) = dp f + dp g , dp (f g) = g(p)dp f + f (p)dp g , dp f1 = − f 21(p) dp f .

Hence, sums, products and inverses of functions of class C m , are of class C m , and C m (U )
will denote the ring of all functions of class C m on U .
Proof: Taking partial derivatives, we are reduced to the case n = 1 considered in p. 29.
Key Lemma: Let U be a ball and f ∈ C m (U ), m ≥ 1. If f vanishes at p = (a1 , . . . , an ) ∈ U ,

there are functions f1 , . . . , fn ∈ C m−1 (U ) such that
f = f1 (x1 − a1 ) + . . . + fn (xn − an ),
P ∂f
and for any such decomposition f = i fi (xi − ai ) we have fj (p) = ∂xj (p).
Proof: If f (p) = 0, and we fix a point x = (x1 , . . . , xn ) ∈ U , we have

h(t) = f p + t(x − p) , −ε < t < 1 + ε ,
Z 1 n Z 1
X ∂f
h0 (t) dt =

f (x) = h(1) − h(0) = (xi − ai ) p + t(x − p) dt
0 ∂xi
i=1 0
n Z 1 n
X ∂f X
= (xi − ai ) p + t(x − p) dt = (xi − ai )fi (x) ,
i=1 0 ∂xi i=1
m−1 by the differentiation rule under the integral sign (p. 33).
P C
where fi is of class
Now, if f = i fi (xi − ai ), and the functions fi are C m−1 (hence continuous), then
∂f fj (a1 , . . . , aj + t, . . . , an )t − 0
(p) = lim = lim fj (a1 , . . . , aj + t, . . . , an ) = fi (p).
∂xj t→0 t t→0
Note: If f is a C ∞ function on a ball U around the origin of Rn , iterating the lemma,

P P P P P P
f =a+ fi xi = a + ai xi + fij xi xj = a + ai xi + aij xi xj + fijk xi xj xk ,
i i i≤j i i≤j i≤j≤k
where fijk ∈ C ∞ (U ). Deriving we see that (∂i ∂j f )(0) = aij = (∂j ∂i f )(0) when i < j.
Moreover, for any natural number m there is a polynomial Pm of degree ≤ m such that
Oi1 ...in (x)xi11 . . . xinn ,

P
f (x) = Pm (x) +
i1 +...+in =m
where Oi1 ...in ∈ C ∞ (U )

and Oi1 ...in (0) = 0, and deriving we see that f and Pm have the same
partial derivatives of order ≤ m at the origin.
In fact, these results holds for functions of class C m , m < ∞:
∂2f ∂2f
Schwarz’s Theorem: If f is a function of class C 2 , then = ·
∂xi ∂xj ∂xj ∂xi
Proof: We may assume that f isP a function on an open ball around the origin and f (0) = 0.
By the lemma, weP have f = r fr xr for some functions fr of class C 1 .
Hence ∂i f = fi + r (∂i fr )xr , where (∂i fr ) is continuous, and we conclude by the lemma
P
(∂j ∂i f )(0) = (∂j fi )(0) + ∂j r (∂i fr )xr (0) = (∂j fi )(0) + (∂i fj )(0).
Definition: A multi-index of length n is a sequence α = (α1 , . . . , αn ) of n natural numbers, and

we put |α| = α1 + . . . + αn , α! = (α1 !) . . . (αn !), xα = xα1 1 . . . xαnn , and
∂ |α| f ∂ |α| f
∂α f = = ·
∂xα ∂xα1 1 . . . ∂xαnn
The m-th Taylor polynomial at a point x = a of a function f of class C m is the unique

polynomial of degree ≤ m with the same partial derivatives of order ≤ m at such point, namely
X 1 ∂ |α| f
(Tam f )(x) := (a) · (x1 − a1 )α1 . . . (xn − an )αn .
α! ∂xα
|α|≤m
Taylor’s Formula: Let f be a function of class C m on an open ball U . If a ∈ U , then there is

a continuous function O(x), vanishing at x = a, such that
f (x) = (Tam f )(x) + O(x)kx − akm .
Proof: Replacing f by f − Tam f , we may assume that f and its partial derivatives of order ≤ m
vanish at x = a. We may also assume that a is the origin of Rn .
Given a point x ∈ U , we consider the function h(t) = f (tx), where −ε < t < 1 + ε.
By the chain’s rule, we have h(0) = h0 (0) = h00 (0) = . . . = h(m (0) = 0 and
n X m! ∂ |α| f
X ∂mf
h(m (t) = (tx)xi1 . . . xim = (tx)xα .
∂xi1 . . . ∂xim α! ∂xα
i1 ,...,im =1 |α|=m
Applying to h(t) the 1-dimensional Taylor formula at t = 0 and t = 1, we obtain
1 (m X 1 ∂ |α| f
f (x) = h(1) = h (ξ) = (ξx)xα ,
m! α! ∂xα
|α|=m
|α|
where 0 < ξ < 1. Since the partial derivatives ∂∂xαf are continuous and vanish at x = 0, we see
that f (x), even when divided by kxkm , is continuous, and vanishes at x = 0.
Theorem: If a differentiable function f has a local extremum at a point p ∈ U , then dp f = 0.
f (p+te)−f (p)
Proof: For any vector e ∈ E, we have (dp f )(e) = limt→0 t = 0 because f has a local
extremum at the point p of the line p + Re.
Definition: The hessian of a function f : U → R of class C 2 at a point p ∈ U is the bilinear

map H : E × E → R, H(e, v) = (∂e ∂v f )(p). When we fix affine coordinates, the matrix of the
hessian is just (∂i ∂j f )(p) ; hence the hessian is a symmetric bilinear form.
Theorem: Let p be a critical point (i.e. dp f = 0) of a function f : U → R of class C 2 .

If the hessian is positive definite, then p is a local minimum of f .
If the hessian is negative definite, then p is a local maximum of f .
If the hessian has positive index, then p is not a local extremum of f .
Proof: We may assume that p is the origin of Rn . If q is the quadratic form defined by the
hessian of f at p, in some open neighborhood of p we have
f (x) = a + 12 q(x) + O(x)kxk2 .
When q is positive definite, and c > 0 is the minimum of q on the unit sphere, we have
q(x) ≥ ckxk2 . Now, since O(x) is continuous and O(0) = 0, there is δ > 0 such that |O(x)| < 2c
when |x| < δ, so that, in the ball of radius δ, the function f attains a minimum at x = 0:
f (x) − a = q(x) + O(x)kxk2 ≥ ckxk2 − 21 ckxk2 = 12 ckxk2 > 0.
When q is negative definite, the hessian of −f is positive definite; hence −f has a local
minimum at p , so that f has a local maximum.
If there are directions e, v such that q(e) > 0 and q(v) < 0, then, f has a local minimum on
the line p + Re, and a local maximum on p + Rv; hence p is not a local extremum of f .
Definition: A bijective map ϕ : U → V of class C m , m ≥ 1, is a C m -diffeomorphim when the

inverse map ϕ−1 : V → U also is of class C m .
A map ϕ : U → V of class C m , m ≥ 1, is a local diffeomorphim at p ∈ U if it induces a
m
C -diffeomorphism of some open neighborhood of p onto an open neighborhood of ϕ(p).
Key Lemma (along the Diagonal): Let U be an open ball, and f ∈ C m (U ), m ≥ 1. There are
functions F1 , . . . , Fn of class C m−1 on U × U such that
f (y) − f (x) = F1 (x, y)(y1 − x1 ) + . . . + Fn (x, y)(yn − xn ),

∂f
Fi (x, x) = ∂xi (x).

Proof: Given points x, y ∈ U , we consider the function h(t) = f ty + (1 − t)x , −ε ≤ t ≤ 1 + ε:
Z 1 n Z 1
0
X ∂f
f (y) − f (x) = h(1) − h(0) = h (t) dt = (yi − xi ) ty + (1 − t)x dt
0 0 ∂xi
i=1
n Z 1 n
X ∂f X
= (yi − xi ) ty + (1 − t)x dt = (yi − xi )Fi (x, y).
0 ∂xi
i=1 i=1
and Fi ∈ C m−1 (U × U ) by the differentiation rule under the integral sign (p. 33).
Inverse Mapping Theorem: A map ϕ : U → U 0 of class C m is a local diffeomorphism at a

point p if and only if the linear tangent map ϕ∗,p : E → E 0 is an isomorphism.
Proof: If ϕ is a local diffeomorphism at p, then ϕ∗,p and ϕ−1∗,ϕ(p) are mutually inverse by the
chain’s rule. Hence ϕ∗,p is an isomorphism.
Conversely, if ϕ∗,p is an isomorphism, consider the equations x0i = fi (x1 , . . . , xn ) of ϕ.
1. ϕ is locally injective:
∂f
Since ∂x i
(p) 6= 0, and the determinant is a continuous function, we may assume that the
j
jacobian has no zero in U . We may also assume that U is a ball so that, by the lemma, we
have functions Fij ∈ C m−1 (U × U ) such that
n
P
fi (y) − fi (x) = Fij (x, y)(yj − xj ),
j=1
∂fi
Fij (x, x) = ∂xj (x),

f (y) − f (x) =
F · (y −
x), where F is the matrix F ij (x, y) . Since Fij (p) 6= 0, we may
assume that Fij (x, y) has no zero in U × U . Hence, if f (y) − f (x) = 0, then y − x = 0.
2. ϕ(U ) is a neighborhood of p0 = ϕ(p):

We may assume that ϕ is injective and the jacobian has no zero in U . Let us consider a
closed ball B̄ ⊂ U with center at p, and let S be the sphere. Since ϕ is continuous and S is
compact, ϕ(S) is compact and p0 ∈/ ϕ(S) because ϕ is injective. Let us see that ϕ(U ) contains
the open set
W = {x0 ∈ U 0 : kx0 − p0 k < d(x0 , ϕ(S))}.
Given a point x0 ∈ W , we consider the following function f : U → R of class C m :
n
h(x) = kϕ(x) − x0 k2 = (fj (x) − x0j )2 .
P
j=1
It attains a minimum at a point x ∈ B̄, and x is not in S because d(x0 , ϕ(S)) > d(x0 , p0 ).
Hence it is a local minimum and
n
∂h X ∂fj
0= (x) = 2 (fj (x) − x0j ) (x).
∂xi ∂xi
j=1
∂f
(x) 6= 0, we see that fj (x) − x0j = 0; hence x0 = ϕ(x).

i
Since ∂xj
3. Locally ϕ is an open map:

We may assume that ϕ is injective and the jacobian has no zero in U . Then ϕ(U ) is an open
set. Hence ϕ(V ) is open when V ⊆ U is open, and ϕ is an open map. It is an homeomorphism,
and we have the inverse homeomorphism ϕ−1 : U 0 → U .
4.3. THE LEBESGUE MEASURE 101
4. ϕ−1 is of class C 1 :
Let xi = gi (x01 , . . . , x0n ) be the equations of ϕ−1 . Since Fij (x, y) is invertible at any point

of U × U , we also have continuous functions Gij on U 0 × U 0 such that

gi (y 0 ) − gi (x0 ) = j Gij (x0 , y 0 )(yj0 − x0j )
P
∂gi
We see that gi is of class C 1 , since ∂x0j
(x0 ) = Gij (x0 , x0 ).
5. ϕ−1 is of class C m : By the chain’s rule, we have a commutative square

ϕ−1
V /U
(ϕ−1 )0 ϕ0

Isom(E 0 , E) o
inv
Isom(E, E 0 )
inv
Now, Isom(E, E 0 ) −−−→ Isom(E 0 , E) is C ∞ , ϕ0 is C m−1 , and by induction on m, we may
assume that ϕ−1 is C m−1 . Hence (ϕ−1 )0 is C m−1 , and ϕ−1 is C m .
4.3 The Lebesgue measure

Definition: Let X be a set. A family of subsets A ⊆ P(X) is an algebra (resp. σ-algebra) if
it is closed under finite (resp. countable) unions and intersections, complement, and ∅, X ∈ A.
A measurable space (X, A) is a set endowed with a σ-algebra of subsets. If Y is a subset of
X, then Y ∩ A = {Y ∩ A; A ∈ A} defines a structure of measurable space on Y .
A map f : (Y, B) → (X, A) is measurable when f −1 (A) ∈ B for all A ∈ A.
Compositions of measurable maps are also measurable maps.
T
Any intersection i Ai of σ-algebras Ai ⊆ P(X) is a σ-algebra; hence any family of subsets
B ⊆ P(X) generates a σ-algebra: the intersection of all σ-algebras containing B. The Borel
σ-algebra B(X) of a topological space X is the σ-algebra generated by the open subsets and,
when Y is a subspace of X, we have B(Y ) = Y ∩ B(X).
Definition: A measure µ on a measurable space (X, A) is a map µ : A → [0, ∞] such that
1. µ(∅) = 0.
2. µ n∈N An = ∞
` P
n=0 µ(An ).
and a measure space is a measurable space endowed with a measure. When A ∈ A, then µ
also defines a measure on (A, A ∩ A).
S
Proposition: If An is an increasing sequence in A, then µ n An = sup µ(An ).
Proof: Put Bn = Acn−1 ∩ An . Then µ(

S P
An ) = n µ(Bn ) = lim µ(B1 ∪ . . . ∪ Bn ) = sup µ(An ).
Definitions: Given a (bounded) interval I = (a, b), (a, b], [a, b) or [a, b], with a ≤ b, we put
µ(I) = b − a. The empty set (a, a) and a point [a, a] are intervals, and µ(∅) = µ(a) = 0.
In Rd , d ≥ 2, a (bounded) rectangle is direct product I = I1 ×. . .×Id of (bounded) intervals
and we put µ(I) = µ(I1 )·. . .·µ(Id ). We admit open and closed rectangles, as well as degenerated
rectangles I (at least one factor is empty or a point) with µ(I) = 0.
If I, J ⊂ R are intervals, so is I ∩ J, and I c ∩ J is a finite union of disjoint intervals; hence,
if I, J ⊂ Rd are rectangles, so is I ∩ J, and I c ∩ J is a finite union of disjoint rectangles.
Lemma: Let I1 , . . . , In , I be rectangles in Rd . If I ⊆ I1 ∪. . .∪In , then µ(I) ≤ µ(I1 )+. . .+µ(In ).

If I1 , . . . , In are disjoint and I1 ∪ . . . ∪ In ⊆ I, then µ(I1 ) + . . . + µ(In ) ≤ µ(I).
Proof: By induction on d, and it is obvious when d = 0 (assuming that µ(R0 ) = µ(0) = 1).
When d ≥ 1, any rectangle I ⊂ Rd defines a family Ix = {y ∈ Rd−1 : (x, y) ∈ I} of rectangles
in Rd−1 , and Z
µ(I) = µ(Ix )dx.
R

By induction µ (I1 )x + . . . + µ (In )x ≤ µ(Ix ), and the result follows from the properties
of the Riemann integral on R.
Definition: The (d-dimensional) exterior measure of A ⊆ Rd is the infimum of the total

volume of all countable covers of A by rectangles:
∞
P S
me (A) = inf µ(In ) ; A ⊆ n In .
n=1
Given ε > 0, it is clear that

P In is contained
P in an open (resp. closed) rectangle Jn with
n
µ(Jn ) < µ(In ) + ε/2 , so that n µ(Jn ) < ε + n µ(In ). Hence, in the definition of me (A) we
may only consider countable covers by open (resp. closed) rectangles.
When me (A) = 0, we say that A has null measure.
1. me (∅) = 0.
2. me (A) ≤ me (B) whenever A ⊆ B.

S P
3. me n An ≤ n me (An ).
n
S P
Given
S ε >S0, we consider covers A n ⊆ m In,m such that m µ(In,m ) < ε/2 + me (An ). Now
n An ⊆ n,m In,m , and
n
S P P P P
me n An ≤ n m µ(In,m ) < n ε/2 + me (An ) = ε + n me (An ).
4. me (I) = µ(I) for any rectangle I.

S
It is clear that me (I) ≤ µ(I). Now, given a cover I ⊆ n In and ε > 0, pick a closed rectangle
K ⊆ I such that µ(K) > µ(I)−ε and open rectangles Un ⊇ In such that µ(Un ) < µ(IP n
n )+ε/2 .
Since K is compact, it admits a finite cover K ⊂ Un1 ∪ . . . ∪ Unr , so that µ(K) ≤ n µ(Un )
by the lemma. Hence,
P P
µ(I) < ε + µ(K) ≤ ε + n µ(Un ) < 2ε + n µ(In ).
5. The exterior measure is invariant under translations: me (a + A) = me (A).
Examples: The argument of p. 34 shows that any countable union of sets of null measure also
has null measure. Hence, any countable set has null measure, as well as any hyperplane xi = ai ,
and the set of all points with some rational coordinate.
In R, we have me (A) ≤ me (Q ∩ A) + me (Qc ∩ A) = me (Qc ∩ A) ≤ me (A) for any set A ⊆ R;
hence me (Qc ∩ A) = me (A). In particular me (Qc ∩ [a, b]) = b − a, and me (Qc ) = ∞.
The set of algebraic numbers is countable, hence, if T is the set of transcendent real numbers,
we have me (T ∩[a, b]) = b−a, and if we consider in C the set T̄ of transcendent complex numbers,
we have me (T̄ ∩ A) = me (A) for any subset A ⊆ C.
Caratheodory’s Lemma: If L is a subset of Rd , the following conditions are equivalent:

1. µ(I) = me (L ∩ I) + me (Lc ∩ I), for any rectangle I.
2. me (A) = me (L ∩ A) + me (Lc ∩ A), for any set A ⊆ Rd .
Proof: (1 ⇒ 2) We always have me (A) = me (L ∩ A) ∪ (Lc ∩ A) ≤ me (L ∩ A) + me (Lc ∩ A).

S P
Given ε > 0, consider a cover A ⊆ n In such that n µ(In ) < ε + me (A).
We have (L ∩ A) ⊆ n (L ∩ In ), (Lc ∩ A) ⊆ n (Lc ∩ In ); hence
S S
me (L ∩ A) + me (Lc ∩ A) ≤ c
P P P
n me (L ∩ In ) + n me (L ∩ In ) = n me (In ) < ε + me (A).
Definition: A set L ⊆ Rd is measurable when it satisfies the above equivalent conditions.
Note: If L is a bounded set, contained in a rectangle K, then µ(K) − me (K − L) deserves the

name of interior measure of L, so that condition µ(K) = me (L ∩ K) + me (Lc ∩ K) states that
the interior and exterior measures of L coincide.
Theorem: The Lebesgue measurable sets in Rd form a σ-algebra containing the rectangles (hence
all the Borel sets) and any set of null measure, and m = me defines a measure (the Lebesgue
measure) invariant under translations.
Proof: If L is measurable, clearly so is Lc , and ∅ and Rd are obviously measurable.

Any set N of null measure is measurable, since me (N ∩I)+me (N c ∩I) = me (N c ∩I) ≤ me (I).
Any rectangle J is measurable because, for any rectangle I, we have that J c ∩ I is a finite
union of disjoint rectangles, J c ∩ I = I1 q . . . q In , and by the lemma
n n
me (J ∩ I) + me (J c ∩ I) ≤ µ(J ∩ I) +
P P
me (Ii ) = µ(J ∩ I) + µ(Ii ) = µ(I).
i=1 i=1
1. If L1 and L2 are measurable, then so is L1 ∪ L2 (hence so is L1 ∩ L2 = (Lc1 ∪ Lc2 )c ):

me (L1 ∪ L2 ) ∩ A + me (L1 ∪ L2 )c ∩ A =

= me L1 ∩ (L1 ∪ L2 ) ∩ A + me Lc1 ∩ (L1 ∪ L2 ) ∩ A + me (L1 ∪ L2 )c ∩ A

= me (L1 ∩ A) + me (L2 ∩ Lc1 ∩ A) + me (Lc2 ∩ Lc1 ∩ A) = me (L1 ∩ A) + me (Lc1 ∩ A) = me (A).

P
2. Countable disjoint unions of measurable sets are measurable, and m qn Ln = n m(Ln ):
Tk−1 c Tk−1 c c
Tk−1 c Tk c

me n=1 Ln ∩ A = me n=1 Ln ∩ Lk ∩ A + me n=1 Ln ∩ Lk ∩ A = me n=1 Ln ∩
A + me (Lk ∩ A). Hence
2
me (A) = me (Lc1 ∩ A) + me (L1 ∩ A) = me (Lc1 ∩ Lc2 ∩ A) +
P
me (Ln ∩ A) = . . .
n=1
Tk k
c ∩A + me (Ln ∩ A) ≥ me (Lc ∩ A) + kn=1 me (Ln ∩ A),
P P
= me L
n=1 n
n=1
where L = qn Ln . Considering the limit as k → ∞, we see that L is measurable:

∞
me (A) ≥ me (Lc ∩ A) +
P
me (Ln ∩ A) ≥
n=1
≥ me (Lc ∩ A) + me ∩ A = me (Lc ∩ A) + me (L ∩ A) .
S
n Ln
P P P
When A = L, we have m(L) ≥ n m(Ln ∩ L) = n m(Ln ); hence m(L) = n m(Ln ).
3. Countable unions of measurable sets are measurable (hence countable intersections):

In fact L = n Ln is the disjoint union of the measurable sets Lc1 ∩ . . . ∩ Lcn−1 ∩ Ln .
S
4. If L is measurable, so is any translation a + L:

We have (a + L) ∩ I = a + L ∩ (I − a) , (a + L)c ∩ I = a + Lc ∩ (I − a) and the exterior

measure me is invariant under translations.
Proposition: A set L ⊆ Rd is measurable if and only if L = B ∪ N with B a Borel set and

m(N ) = 0; hence if and only if L is contained in a Borel set B such that m(B − L) = 0.
Proof: If m(L) < ∞, there is a Borel set B ⊇ L (in fact a countable intersection of countable
covers of L by rectangles) such that m(L) = m(B); hence m(B − L) = 0.
When m(L) = ∞, we put Rd = qn In as a countable disjoint union S of rectangles. We have
Borel sets Bn ⊇ (L ∩ In ) with m Bn − (L ∩ In ) = 0; hence L ⊆ B = n Bn and m(B − L) = 0.
Finally, if B is a Borel set containing Lc and m(B − Lc ) = 0, then L = B c ∪ (B ∩ L), and
m(B ∩ L) = m(B − Lc ) = 0.
Proposition: If L1 ⊆ Rn1 , L2 ⊆ Rn2 are measurable, so is L1 × L2 ⊆ Rn1 +n2 and
m(L1 × L2 ) = m1 (L1 )m2 (L2 ) , (where 0 · ∞ = 0).
Proof: Let I ⊂ Rn2 be a rectangle. Since µ(B) = m(B × I) is an invariant measure on B(Rn1 ),
we have µ = m2 (I)m1 ; hence m(B × I) = m1 (B)m2 (I) for any Borel set B ⊆ Rn1 .
Now, if m1 (B1 ) < ∞, then µ(B) = m(B1 × B) is an invariant measure on B(Rn2 ).
It follows that µ = m1 (B1 )m2 ; hence m(B1 × B2 ) = m1 (B1 )m2 (B2 ).
If m1 (B1 ) = m2 (B2 ) = ∞, there n2
is a rectangle I ⊂ R such that 0 < m2 (B2 ∩ I) < ∞.
Hence ∞ = m B1 × (B2 ∩ I) ≤ m(B1 × B2 ), and m(B1 × B2 ) = ∞.
Finally, in general Li = Bi ∪ Ni , with Bi a Borel set and mi (Ni ) = 0, so that
L1 × L2 = (B1 × B2 ) ∪ (N1 × L2 ) ∪ (L1 × N2 )
is measurable, because B1 × B2 is a Borel set and m(N1 × Rm ) = m(Rn × N2 ) = 0, since

m(N1 × I) = 0 for any interval I.
Moreover, m(L1 × L2 ) = m(B1 × B2 ) = m1 (B1 )m2 (B2 ) = m1 (L1 )m2 (L2 ).
Lemma: There is a unique measure m on B(Rd ) such that m(I) = µ(I) for any rectangle I.
Proof: Let S
m̄ be a measure suchPthat m̄(I) =P µ(I) for any rectangle I, and let B be a Borel set.
If B ⊆ n In , then m̄(B) ≤ n m̄(In ) = n µ(In ); hence m̄(B) ≤ m(B).
Now, if B is contained in a rectangle I, then m̄(B) ≤ m̄(I) = µ(I) < ∞ and
µ(I) − m̄(B) = m̄(I) − m̄(B) = m̄(I − B) ≤ m(I − B) = µ(I) − m(B) ,
so that m(B) ≤ m̄(B). When B is unbounded, we consider a partition Rd = qn In as a disjoint

union of rectangles In , and we conclude:
P P
m̄(B) = n m̄(In ∩ B) = n m(In ∩ B) = m(B) .
Theorem: If a measure m̃ on B(Rd ) is invariant under translations and m̃ [0, 1)d = c < ∞,

then we have m̃ = cm.
Proof: If c = 0, then m̄(Rd ) = 0, because Rd = qn1 ,...,nd ∈Z (n1 , . . . , nd ) + [0, 1)d ; hence m̄ = 0.

If c > 0, replacing m̃ by 1c m̃, we may assume that c = 1.

Q
For any rectangle I = i [ai , bi ), with bi − ai = ni ∈ N, we have m̃(I) = m(I), because I is
a disjoint union of n1 · · · nd translations of [0, 1)d .
Hence, for any rectangle J = i [ai , bi ), with bi − ai = pqii ∈ Q, we have m̃(J) = m(J),
Q
because the disjoint union of q1 · · · qd translations of J is a rectangle I with integer edges.
Now, for any ε > 0 and any open rectangle K, there are rectangles J1 , J2 with rational edges
such that J1 ⊆ K ⊆ J2 and m(J2 ) − m(J1 ) < ε; hence m̃(K) = m(K).
Finally, if L is a degenerated rectangle, for any ε > 0 there is an open rectangle K ⊃ L such
that m̃(K) = m(K) < ε; hence m̃(L) = m(L) = 0.
We conclude by the previous lemma.
Corollary: The Lebesgue measure is invariant under rigid motions, and any linear subvariety
of Rn of dimension < n has null measure.

Proof: Let ϕ is a rigid motion. The measure µ(B) = m ϕ(B) is invariant under translations,
and m ϕ(D) = m(D) > 0 when D is a ball, because ϕ(D) is a translation of D.
Finally, if X is a linear subvariety of dimension d < n, then ϕ(X) = Rd × 0 for some rigid
motion ϕ, and Rd × 0 has null measure.
Theorem: Given vectors e1 = (a11 , . . . , an1 ), . . . , en = (a1n , . . . , ann ) in Rn , we have

det(aij ) = ±m {x1 e1 + . . . + xn en ; 0 ≤ x1 , . . . , xn < 1} .
Proof: We have to show that the right term Vol(e1 , . . . , en ) is an alternate tensor.
Clearly it is alternate, and Vol(λe1 , . . . , en ) = λVol(e1 , . . . , en ) when λ ∈ Z, because m is
invariant under translations; hence when λ−1 ∈ Z; hence when λ ∈ Q; hence for all λ ∈ R.
It vanishes when e1 , . . . , en are linearly dependent, because they are in a proper vector
subspace, and moreover Vol(e1 , . . . , en ) = Vol(e1 , . . . , en + e1 ), because the non-common parts
of both parallelepipeds differ in a translation:
e1 + {x1 e1 + xn en ; 0 ≤ x1 < xn < 1} = {y1 e1 + yn (en + e1 ); 0 ≤ y1 , yn < 1 ≤ y1 + yn }

{x1 e1 + xn en ; 0 ≤ xn ≤ x1 < 1} = {y1 e1 + yn (en + e1 ); 0 ≤ y1 , yn , y1 + yn < 1}
en
e1
Now, Vol(e1 , . . . , en + λe1 ) = λ1 Vol(λe1 , . . . , en + λe1 ) = λ1 Vol(λe1 , . . . , en ) = Vol(e1 , . . . , en ),

and we see that Vol(e1 , . . . , en + u) = Vol(e1 , . . . , en ) when u ∈ he1 , . . . , en−1 i. We conclude:
Vol(e1 , . . . , en + vn ) = Vol(e1 , . . . , en + λen + u) = Vol(e1 , . . . , en + λen )

= (1 + λ)Vol(e1 , . . . , en ) = Vol(e1 , . . . , en ) + Vol(e1 , . . . , λen )
= Vol(e1 , . . . , en ) + Vol(e1 , . . . , λen + u) = Vol(e1 , . . . , en ) + Vol(e1 , . . . , vn ).
Corollary: If T : Rn → Rn is linear and B ∈ B(Rn ), then m T (B) = |det T |m(B).

Proof: If det T = 0, then m T (B) = 0 because T (B) ⊆ Im T and dim (Im T ) < n.
If det T 6= 0, then the measure µ(B) = m T (B) is invariant under translations, and
µ [0, 1)n = |det T | by the theorem.
4.4 Integral Calculus

Let (X, A) be a measurable space, and recall that a function f : X → [0, ∞] is measurable when
f −1 (B) ∈ A, ∀B ∈ B [0, ∞] . In fact, it is enough that {x ∈ X : f (x) > c}∈ A, ∀c ∈ R+ (and
we may replace the sign > by <, ≥ or ≤) because (c, ∞] generate B [0, ∞] .
If f, h : X → [0, ∞] are measurable, so is f × h : X × X → [0, ∞] × [0, ∞], because any open
set in [0, ∞] × [0, ∞] is a countable union of open sets U1 × U2 ; hence f + h is measurable, the
set {x ∈ X : f (x) = h(x)} is in A, etc.
Definition: A measurable function s : X → [0, ∞) is simple when s = a1 IA1 + . . . + an IAn for

some partition X = qi Ai , with Ai ∈ A, ai < ∞.
Lemma: If some functions fn : X → [0, ∞] are measurable, then so are h(x) = sup fn (x) and
g(x) = inf fn (x), and the function f (x) = lim fn (x) when it exists.
S
Proof: {x ∈ X : sup fn (x) > c} = {x ∈ X : fn (x) > c} ∈ A,
Tn
{x ∈ X : inf fn (x) ≥ c} = n {x ∈ X : fn (x) ≥ c} ∈ A,

lim fn = inf n≥1 supi≥n fi .
Theorem: Any measurable function f : X → [0, ∞] is the limit f = lim sn of an increasing

sequence of simple functions sn .
Proof: Let Qn = 2mn ; 0 ≤ m ≤ 2n n be the set of all rationals in [0, n] with denominator 2n .

Now put sn (x) = max{q ∈ Qn : q ≤ f (x)} ∈ Qn , and sn is a measurable function because

{x ∈ X : sn (x) ≥ q} = {x ∈ X : f (x) ≥ q}, ∀q ∈ Qn .
Definition: Let (X, A, µ) be a measure space. The integral of a simple function s is

Z
P P P
ai IAi dµ = ai µ(Ai ) ; s = ai IAi , X = qi Ai ,
X i i i
(whereP0 · ∞ = 0)P and it does not depend on the decomposition:

If i ai IAi = j bj IBj , with X = qj Bj , then ai = bj whenever Ai ∩ Bj 6= ∅, and
P P P P P P
i ai µ(Ai ) = i j ai µ(Ai ∩ Bj ) = j i bj µ(Ai ∩ Bj ) = j bj µ(Bj ).
The integral of a measurable function f : X → [0, ∞] is defined to be

Z Z
f dµ = sup sdµ ; s is simple and 0 ≤ s ≤ f ∈ [0, ∞].
X X
For any set A ∈ A, the functions f |A and IA f also are measurable, and we have
Z Z
(f |A )dµ = IA f dµ.
A X
1. When f is simple, this definition coincides with the former one:

P P
If s = j bj IBj ≤ f = ai IAi , X = qi Ai = qj Bj , then bj ≤ ai when Ai ∩ Bj 6= ∅, and
P P P P P P
j bj µ(Bj ) = j i bj µ(Ai ∩ Bj ) ≤ i j ai µ(Ai ∩ Bj ) = i ai µ(Ai ).
Z Z
2. If f is measurable, then cf dµ = c f dµ for any c ∈ [0, ∞).
Z Z
3. If f, h are measurable and f ≤ h, then f dµ ≤ hdµ.
Z Z
4. If A, B ∈ A and A ⊆ B, then f dµ ≤ f dµ.
A B
Z
5. If A ∈ A and µ(A) = 0, then f dµ = 0.
A
R R R
We have A sdµ = 0 when s is simple; hence A f dµ = sup A sdµ; s ≤ f = 0.
6. Monotone Convergence Theorem: If fn is an increasing sequence of measurable func-

tions, then Z Z
lim fn dµ = (lim fn )dµ.
The sequence (fn ) RconvergesRto a function f because

R (fnR) is increasing, and f = lim fn is
measurable; hence fn dµ ≤ f dµ, so that lim fn dµ ≤ f dµ.
P
Now, given a simple function s = i ai IAi ≤ f , we fix r < 1 and we consider
Bn = {x ∈ X : rs(x) ≤ fn (x)}.
S
We have Bn ⊆ Bn+1 because (fn ) is increasing, and X = n Bn because fn (x) → f (x) and
rs(x) < f (x) (since r < 1 and s(x) < ∞ by the definition of simple function).
Z Z Z Z
r sdµ ≤ fn dµ ≤ fn dµ ≤ lim fn dµ.
Bn Bn X X
R R
Now, r Bn sdµ →r sdµ when n → ∞, because µ(Bn ∩ Ai ) → µ(Ai ).
X
R R R R
When r → 1, we see that X sdµ ≤ lim X fn dµ; hence X f dµ ≤ lim X fn dµ.
Z Z Z
7. If f and h are measurable, then (f + h)dµ = f dµ + hdµ.
P P P
If f = ai IAi and h = bj IBj , then f + h = i,j (ai + bj )IAi ∩Bj and
P P P P P
(ai + bj )µ(Ai ∩ Bj ) = ai µ(Ai ∩ Bj ) + bj µ(Ai ∩ Bj ) = ai µ(Ai ) + bj µ(Bj ).
i,j i,j i,j i j
In the general case, f = lim sn and h = lim tn , where (sn ) and (tn ) are increasing sequences of
simple functions, so that f + h = lim(sn + tn ) and we conclude by the monotone convergence
theorem.
∞ Z ∞
Z X !
X
8. If fn are measurable functions, then fn dµ = fn dµ.
n=1 n=1
Z Z Z
9. If A, B ∈ A and µ(A ∩ B) = 0, then f dµ = f dµ + f dµ.
A∪B A B
R
In fact IA + IB = IA∪B + IA∩B , and f dm = 0 when µ(A ∩ B) = 0.
A∩B
Z Z
10. If f and h are measurable and f = h almost everywhere, then f dµ = hdµ.
Z
11. If f is measurable and f dµ = 0, then f = 0 almost everywhere.
Z Z
µ({f (x) ≥ ε}) = dµ ≤ ε−1 f dµ = 0.
{f (x)≥ε}
Z
12. If f is measurable, then µf (A) := f dµ is a measure on A.
A
RP P R P
If A = qn An , then µf (A) = n IAn f dµ = n IAn f dµ = n µf (An ).
Definition: If f : X → R is measurable, so are f+R= max{f, 0},

R f− = max{−f, 0}, and we have
f = f+ − f− . We say that f is integrable when f+ dµ and f− dµ are finite, and we put
Z Z Z
f dµ = f+ dµ − f− dµ ∈ R.
X X X
Z Z
1. If f is integrable, so is cf , ∀c ∈ R, and cf dµ = c f dµ.
Z Z Z
2. If f and h are integrable, so is f + h and (f + h)dµ = f dµ + hdµ.
We have (f + h)+ − (f + h)− = f + h = f+ − f− + h+ − h− .

Hence (f + h)+ + f− + h− = (f + h)− + f+ + h+ , and integrating we conclude.
Z Z
3. If f and h are integrable and f ≤ h, then f dµ ≤ hdµ.
4. If f is measurable and |f | ≤ h for some integrable function h, then f is integrable and

Z Z

f dµ ≤ |f |dµ.

R R R
|f | = f+ + f− , and (f+ + f− )dµ is finite if and only if so are f+ dµ and f− dµ.
5. If fn is an increasing (or decreasing) sequence of integrable functions with integrable limit,

Z Z
lim fn dµ = (lim fn )dµ.
If (fn ) is increasing, then 0 ≤ fn − f1 ≤ f − f1 , and we conclude by the monotone convergence

theorem. If (fn ) is decreasing, then (−fn ) is increasing, and we conclude.
Dominated Convergence Theorem: Let fn , f : X → R be measurable functions such that

f = lim fn . If there is an integrable function h such that |fn | ≤ h for any index n, then
Z Z
lim fn dµ = f dµ.
Proof: Since |fn | → |f |, we also have |f | ≤ h; hence f is integrable. Moreover, mn = inf k≥n fk
is an increasing sequence, and Mn = supk≥n fk is a decreasing sequence, both with limit f .
Since mn , Mn are integrable, because |mn |, |Mn | ≤ h, and mn ≤ fn ≤ Mn , we conclude:
Z Z Z
mn dµ ≤ fn dµ ≤ Mn dµ,
Z Z Z Z Z
lim mn dµ = (lim mn )dµ = f dµ = (lim Mn )dµ = lim Mn dµ.
Differentiation under the Integral Sign: Let U ⊆ Rn be an open set, and let f : U × X → R
be a function such that ft (x) = f (t, x) is measurable, ∀t ∈ U , and fx (t) = f (t, x) is of class C m ,
0 ≤ m ≤ ∞, for all x ∈ X. If for any multi-index |α| ≤ mR we have an integrable function gα

such that |∂α f (t, x)| ≤ gα (x), ∀(t, x) ∈ U × X, then F (t) = X f (t, x)dµ is of class C m and
∂ |α| ∂ |α| f
Z Z
f (t1 , . . . , tn , x)dµ = (t1 , . . . , tn , x)dµ , |α| ≤ m.
∂tα X X ∂tα
Proof: Fix a ∈ U . Given any sequence tn → a, we have

Z Z Z
F (tn ) = f (tn , x)dµ → lim f (tn , x) dµ = f (a, x)dµ = F (a)
tn →a
by the dominated convergence theorem. Hence F (t) is continuous. Moreover,
F (a + hi ) − F (a) f (a + hi , x) − f (a, x)
Z
= dµ , hi = (0, . . . , h, . . . , 0),
h h
and the integrand may be extended to a continuous function on a neighborhood of h = 0, where

∂f
the value is ∂ti
(a, x). By the mean value theorem this extension is absolutely bounded by gi .
Hence the first term also may be extended to a continuous function on a neighborhood of
R ∂f ∂f
h = 0, so that ∂F∂ti exists and ∂F
∂ti = ∂F
∂ti dµ. In particular, ∂ti is continuous, because so is ∂ti .
4.4.1 Change of Variables Formula

Now we consider Rn with the σ-algebra of Borel sets and the Lebesgue measure m; hence a
function f is measurable when f −1 (B) is a Borel set for any Borel set B ⊆ R, and we put
Z Z
f (x1 , . . . , xn )dx1 . . . dxn = f dm.
Rn Rn
Theorem: If f : Rn → [0, ∞] is measurable and Γf = {(x, y) ∈ Rn × R : 0 ≤ y < f (x)}, then

Z
f dm = m(Γf ).
Rn
P
Proof: It holds when f = i ai IBi is a simple function, because m Bi × [0, ai ) = ai m(Bi ).
InSgeneral, f = lim sn is a limit of an increasing sequence of simple functions, so that
Γf = s Γsn and we conclude by the monotone convergence theorem:
Z Z Z
m(Γf ) = lim m(Γsn ) = lim sn dm = (lim sn )dm = f dm.
Monotone Class Theorem: The σ-algebra generated by an algebra A is the least family
M ⊇ A closed under countable increasing unions and countable decreasing intersections.
Proof: The family B = {B ∈ M : B ∩SA ∈ M, ∀A ∈ S A} contains A, and it is closed under

countable increasing unions,
T because (
T n n B ) ∩ A = n (Bn ∩ A), and countable decreasing
intersections, because ( n Bn ) ∩ A = n (Bn ∩ A); hence B = M.
Now C = {B ∈ M : B ∩ A ∈ M, ∀A ∈ M} contains A, and it is closed under countable
increasing unions and decreasing intersections; hence C = M is closed under finite intersections.
Analogously D = {B ∈ M : B c ∈ M} contains A, and it is closed under countable increasing
unions and decreasing intersections; hence B = M is closed under complements.
Since M is closed under complements
S S and finite intersections, it is closed under finite unions;
hence under countable unions An = n (A1 ∪ . . . ∪ An ), and M is a σ-algebra.
Lemma: If B ⊆ Rn+m is a Borel set, and we put By = {x ∈ Rn : (x, y) ∈ B}, where y ∈ Rm ,

then the function hB (y) = m(By ) is measurable, and
Z
m(B) = m(By )dy1 . . . dym .
Rm
Proof: Replace Rn+m by I × Rm , where I is a rectangle, and let M be the family of all Borel
sets B ⊆ I × Rm such that hB is measurable.
IfShB is measurable, so is hB c = m(I) − hB , and if we have a countable increasing union
B = n Bn of Borel sets Bn ∈ M, then B ∈ M, because hB = sup hBn ; hence M also is closed
under countable decreasing intersections.
Clearly M contains the finite disjoint unions of (eventually unbounded) rectangles in I ×Rm ,
and such family is an algebra (if J, J 0 are rectangles, then J ∩ J 0 is a rectangle, and J c is a finite
disjoint union of rectangles): By the monotone class theorem, M = B(I × Rm ).
In general, hB = i hIi ∩B is measurable, where Rn = qIi is a countable partition.
P
Now, µ(B) = Rm hB dm is a measure on B(Rn+m ):

R
Z Z
` P P P
µ Bn = hBn dm = hBn dm = µ(Bn ),
n Rm n n Rm n
and it is invariant. Since µ [0, 1)n+m = 1, we conclude that m(B) = Rm hB dm.

R
Cavalieri’s Principle: If m(By ) = m(B̄y ), ∀y ∈ Rm , then m(B) = m(B̄).
Fubini’s Theorem: If f : Rn+m → [0, ∞] is measurable, so is hf (y) = Rn f (x, y)dm and

R
Z Z Z
f (x, y)dx1 . . . dxn dy1 . . . dym = f (x, y)dx1 . . . dxn dy1 . . . dym .
Rn ×Rm Rm Rn
Proof: By the lemma, it holds for any simple function.

In general, we put f = lim sn where (sn ) is an increasing sequence of simple functions.
By the monotone convergence theorem, hf = lim hsn and we conclude:
Z Z Z Z
hf dm = lim hsn dm = lim sn dm = f dm.
Rm Rm Rn+m Rn+m
Corollary: IfR f : Rn+m → R is integrable, and fy (x) = f (x, y) is integrable for any y ∈ Rm ,
then hf (y) = Rn f (x, y)dm is integrable and Fubini’s formula holds for f .
R R
Proof: By the theorem, h+ (y) =R Rn f+ dmRand h− (y)R = Rn f− dm R have finite R integral;
R hence
hf = h+ − h− is integrable, and hf dm = h+ dm − h− dm = f+ dm − f− dm = f dm.

Definition: Let ϕ : U → V , ϕ(x1 , . . . , xn ) = y1 (x1 , . . . , xn ), . . . , yn (x1 , . . . , xn ) , be a map of
class C 1 , where U, V ⊆ Rn are open sets. The jacobian Jϕ (p) at a point p ∈ U is

∂y1 . . . ∂y1
∂x1 ∂xn
Jϕ (p) = det ϕ∗,p = . . . . . . . . . (p).

∂yn ∂yn
∂x . . . ∂x
1 n
Change of Variables Formula: Let ϕ : U → V be a C 1 -diffeomorphism. If f : V → [0, ∞] is

measurable, then so is (f ◦ ϕ)|Jϕ | and
Z Z
∂yi
f (y1 , . . . , yn )dy1 . . . dyn = f y1 (x), . . . , yn (x) det
dx1 . . . dxn .
V U ∂xj
Proof: If f is measurable, so is f ◦ ϕ because ϕ is aR homeomorphism;

R hence (f ◦ ϕ)|Jϕ | because
|Jϕ | is continuous. And we only have to show that V f dm ≤ U (f ◦ ϕ)|Jϕ |dm, since the reverse
inequality follows applying this inequality to ϕ−1 : V → U .
If we prove this inequality when f is the indicator function of a Borel set, the monotone
convergence theorem shows that it holds when f is measurable.
1. Given ε > 0, each point p ∈ U has a neighborhood Up such that

ϕ(x + C) ⊆ ϕ(x) + (1 + ε)ϕ∗,p (C)
for any cube3 x + C ⊆ Up . In particular, m ϕ(x + C) ≤ (1 + ε)n |Jϕ (p)|m(C).

By the key lemma of infinitesimals along the diagonal, in a neighborhood of p we have

ϕ(y) − ϕ(x) = F (x, y)(y − x), where F (x, y) is a continuous family of endomorphisms of Rn
such that F (x, x) = ϕ∗,x ; hence
ϕ(y) − ϕ(x) = ϕ∗,p (y − x) + H(x, y)(y − x),
where H(x, y) = F (x, y) − ϕ∗,p is a continuous family such that H(p, p) = 0.

Since ϕ∗,p is an isomorphism, there exists k > 0 such that C ⊆ kϕ∗,p (C) for any cube C
centered at the origin, and we may fix a neighborhood Up where H(x, y)(C) ⊆ kε C, ∀x, y ∈ Up .
We conclude:
ϕ(x + C) ⊆ ϕ(x) + ϕ∗,p (C) + kε C ⊆ ϕ(x) + ϕ∗,p (C) + εϕ∗,p (C) = ϕ(x) + (1 + ε)ϕ∗,p (C).
R
2. Given α > 1, each point p ∈ U has a neighborhood Up such that m(ϕB) ≤ α B |Jϕ |dm for
any Borel set B ⊆ Up .
√
In a neighborhood Up we have m ϕ(x + C) < α|Jϕ (p)|m(C) for any cube x + C ⊆ Up , and
√
|Jϕ (p)| < αJ, where J is the infimum of |Jϕ | on Up ; so that m ϕ(x + C) < αJm(C).
P so that any Borel set B ⊆ Up admits a countable cover

We may assume that Up is a rectangle,
by cubes xn + Cn ⊆ Up such that n m(Cn ) < αm(B), because any rectangle is a countable
disjoint union of cubes; hence
Z
m(ϕB) ≤ m ϕ(xn + Cn ) ≤ αJ m(Cn ) ≤ α2 Jm(B) ≤ α2
P P
|Jϕ |dm.
n n B
S R
Now, given α > 1, fix a countable open cover U = i Ui such that m(ϕBi ) ≤ α Bi |Jϕ |dm
for any Borel set Bi ⊆ Ui .
Any Borel set B ⊆ U admits a partition B = qi Bi , where Bi is a Borel set in Ui , hence
Z Z
P
m(ϕB) = m(ϕBi ) ≤ α |Jϕ |dm = α |Jϕ |dm.
i Bi B
R
Since it holds for any α > 1, we conclude that m(ϕB) ≤ B |Jϕ |dm.
Corollary: If N ⊂ U has null measure, so does ϕ(N ). If L ⊆ U is measurable, so is ϕ(L).

R
Proof: Take a Borel set N ⊆ B with m(B) = 0. Then m(ϕN ) ≤ m(ϕB) = B |Jϕ |dm = 0.
Now, any measurable set is L = B ∪ N , where B is a Borel set and m(N ) = 0.
Hence ϕ(L) = ϕ(B) ∪ ϕ(N ) is measurable, since ϕ(B) is a Borel set and m(ϕN ) = 0.
Corollary: If f : V → R is integrable, then so is (f ◦ ϕ)|Jϕ | and the change of variables formula

holds for f .
3
That is to say, C is a rectangle with center at the origin and sides of equal length.

Proof: If f+ , f− have finite integrals, so have (f+ ◦ ϕ)|Jϕ | = (f ◦ ϕ)|Jϕ | + and (f− ◦ ϕ)|Jϕ | =

(f ◦ ϕ)|Jϕ | − ; hence (f ◦ ϕ)|Jϕ | is integrable.
Definition: Let U ⊆ Rn be an open set and f : U → Rd a map of class C 1 . We say that x ∈ U

is a critical point of f when rk f∗,x < d, and the critical set ∆f is the set of critical points.
Sard’s Theorem: Let U ⊆ Rn be an open set and f : U → Rd a map of class C m , m < nd . The
set of critical values f (∆f ) has null measure in Rd .
Proof: Let us consider the equations yi = yi (x1 , . . . , xn ) of f , and let ∆kf be the set of points
x ∈ U where all the partial derivatives of the functions yi (x1 , . . . , xn ) up to order k vanish at x.
The case n = 0 is trivial, and we proceed by induction on n.
1. The set f (∆f − ∆1f ) has null measure: Let us show that any point x ∈ ∆f − ∆1f has an open
neighborhood V such that f V ∩ (∆f − ∆1f ) has null measure.

Re-labelling the coordinates if necessary, we may assume that (∂1 y1 )(x) 6= 0. Hence, replacing
U by a smaller neighborhood, and composing f with a diffeomorphism, we may assume that
y1 (x1 , . . . , xn ) = x1 . Now, for any t ∈ R we have a map
ft : Ut = {(x2 , . . . , xn ) ∈ Rn−1 : (t, x2 , . . . , xn ) ∈ U } −→ Rd−1

ft (x2 , . . . , xn ) = f (t, x2 , . . . , xn ),
f (∆f − ∆1f ) = t {t} × ft ∆ft .
S
By induction ft ∆ft has null measure in Rd−1 , so that f (∆f − ∆1f ) has null measure in Rd

by Cavalieri’s principle.
2. The set f (∆kf − ∆k+1

f ) has null measure for all k ≥ 1: Let us show that any point x ∈
∆f − ∆f has an open neighborhood V such that f V ∩ (∆kf − ∆k+1
k k+1
f ) has null measure.
Replacing U by a smaller neighborhood, we may assume that ∆k+1f = ∅, and composing f

k
with a diffeomorphism, we may assume that ∆f is contained in the hyperplane x1 = 0. Now
we have a map
f0 : U0 = {(x2 , . . . , xn ) ∈ Rn−1 : (0, x2 , . . . , xn ) ∈ U } −→ Rd−1

f0 (x2 , . . . , xn ) = f (0, x2 , . . . , xn ),
f (∆kf ) = f0 ∆kf0 ,

and by induction f0 ∆kf0 has null measure in Rd−1 .

n
3. The set f (∆kf ) has null measure when k > m: Let us show that f (K ∩ ∆kf ) has null measure
for any compact set K ⊂ U .
By the key lemma along the diagonal, there is a constant c0 > 0 and a finite cover of K ∩ ∆kf
by open sets V such that kf (x2 ) − f (x1 )k ≤ ckx2 − x1 kk whenever x1 ∈ K ∩ ∆kf , x2 ∈ V .
Hence, we have a constant c > 0 such that if C ⊂ V is a cube with edge r which intersects
K ∩ ∆kf , then the exterior measure is
me f (C) ≤ crmk = c · m(C)mk/n .

4.5. CONVERGENCE IN C M (U ) 113
Cover K ∩ ∆kf with a finite number of such cubes {Cj }, with disjoint interiors. Given a
natural number N , subdivide each cube Cj into N n cubes {Cji } of equal sizes. If a cube Cji
intersects K ∩ ∆kf , then me f (Cji ) ≤= c · m(Cji )mk/n = cN −mk m(Cj ). Hence,

me f (Cj ∩ ∆kf ) ≤ i me f (Cji ∩ ∆kf ) ≤ N n−mk m(Cj ).

P
Since n − mk < 0, when N → ∞, we obtain that me f (Cj ∩ ∆kf ) = 0, and we conclude.

4.5 Convergence in C m (U )
Definition: Let X, Y be topological spaces and let C(X, Y ) be the set of all continuous maps.
The compact-open topology on C(X, Y ) is generated by the subsets
hK, U i = {f ∈ C(X, Y ) : f (K) ⊆ U },
where K ⊆ X is compact and U ⊆ Y is open.

A base of open sets are just the finite intersections hK1 , U1 i ∩ . . . ∩ hKn , Un i.
If U ⊆ V , then hK, U i ⊆ hK, V i. If K ⊆ L, then hL, U i ⊆ hK, U i. Hence, if X is σ-compact
and Y has a countable base of open sets, then C(X, Y ) also has a countable base.
Definition: Put K = R or C. A topological vector space is a K-vector space E with a

linear topology, a topology such that the structural maps E × E → E and K × E → E are
continuous, where E × E and K × E are endowed with the direct product topology, so that also
are continuous the natural maps
Kn × E n −→ E , (λ1 , . . . , λn , e1 , . . . , en ) 7→ λ1 e1 + . . . + λn en .
The topological K-vector spaces, with the continuous K-linear maps, define a category.
Since the translations are continuous, they are homeomorphisms, and we see that any linear
topology is determined by the family {Ui } of neighborhoods of 0, since {p+Ui } is just the family
of neighborhoods of any point p ∈ E.
Definition: A seminorm on a real vector space E is a map q : E → R≥0 such that
1. q(e + v) ≤ q(e) + q(v); ∀e, v ∈ E.
2. q(λe) = |λ| · q(e); ∀λ ∈ K, e ∈ E.
and it is a norm when q(e) = 0 ⇒ e = 0. Sometime we put q(e) = kek.

A seminorm defines a pseudometric4 d(x, y) = q(y − x), and it is a metric when q is a
norm. Hence, a seminorm defines a topology on E. A base of neighborhoods of x ∈ E is defined
by the balls
Br (x) = B(x, r) = Bd (x, r) := {y ∈ E : d(x, y) < r} = {y ∈ E : ky − xk < r} , r ∈ R+ ,
and this topology is a linear topology because
Br (x) + Bs (y) ⊆ Br+s (x + y) , (λ − ε, λ + ε)Br (x) ⊆ B(|λ|+|ε|)r (λx).

4
A map d : X × X → R≥0 such that d(x, x) = 0, d(x, y) = d(y, x), and d(x, z) ≤ d(x, y) + d(x, z). Any family
of pseudometrics {di }Tinduces a topology on X, a base of neighborhoods of x ∈ X being defined by the finite
intersections of balls i Bdi (x, ri ), ri > 0, where Bd (x, r) = {y ∈ X : d(x, y) < r}. This topology is separated
when x 6= y ⇒ di (x, y) 6= 0 for some index i.
In general, any family of seminorms {qi } on E defines a linear topology, and a topological
vector space is locally convex when the topology is defined by a family of seminorms. Then a
base of neighborhoods of x ∈ E is defined by the finite intersections of balls
{y ∈ E : qi1 (y − x) < r1 } ∩ . . . ∩ {y ∈ E : qin (y − x) < rn }.
When the topology of E is separated and may be defined by a countable family of semi-
norms {qnP}, then E is metrizable, the topology being defined by the translation-invariant metric
d(x, y) = n 2−n min{1, qn (y − x)}. Now, a sequence (en ) is a Cauchy sequence for d when
for any neighborhood U of 0, there is an index k such that en − em ∈ U , ∀m, n ≥ k. Hence
Cauchy sequences do not depend on the fixed family {qn }, only on the linear topology of E, and
we say that E is complete if any Cauchy sequence converges.
√
Examples: Any scalar product on a real vector space E defines a norm q(e) = e · e.
If B(X) is the ring of all bounded functions onR a set X, then q(f ) = supx∈X |f (x)| is a norm.
If (X, A) is a measurable space, then q(f ) = X |f |dµ is a seminorm on the vector space of
all integrable functions.
Let q be a seminorm on a vector space F . If f : E → F is a linear map, then (f ∗ q)(e) :=
q f (e) is a seminorm on E, and the corresponding topology is just the initial topology. In
particular, if V ⊂ F is a vector subspace, then the restriction q|V is a seminorm, and the
corresponding topology on V is the induced topology.
If q1 , q2 are two seminorms on a vector space E, then q1 + q2 and sup{q1 , q2 } also are
seminorms, defining the same topology as the pair of seminorms {q1 , q2 }.
Definition: Any compact set K in a topological space X defines a seminorm k kK on the ring
of real (or complex) valued continuous functions C(X),
kf kK = supp∈K |f (p)| , dK (f, h) = kh − f kK ,
and the seminorms k kK define the compact convergence topology on C(X).

S o
When X is σ-compact, X = n Kn , Kn ⊆K n+1 , this linear topology is defined by the
seminorms k kKn , because K ⊆ Kn for some index n.
Proposition: The topology of the compact convergence on C(X) coincides with the compact-open
topology. In particular, C(X) admits a countable base of open sets when X is σ-compact.
Proof: If f ∈ hK, U i and ε is smaller than the distance of f (K) to X−U , then BdK (f, ε) ⊆ hK, U i,
and we see that hK, U i is open in the topology of the compact convergence.
Now, given a ball BdK (f, 2ε), any point p ∈ K has a compact neighborhood Ūp in K such
that f (Ūp ) ⊆ f (p) − ε, f (p) + ε , so that dŪp (h − f ) < 2ε when h ∈ hŪp , f (p) − ε, f (p) + ε i.
Since K is compact, it admits a finite cover K = Ūp1 ∪ . . . ∪ Ūpn , and we see that BdK (f, 2ε)
is open in the compact-open topology because

hŪp1 , f (p1 ) − ε, f (p1 ) + ε i ∩ . . . ∩ hŪpn , f (pn ) − ε, f (pn ) + ε i ⊆ BdK (f, 2ε).
Theorem: If X is a σ-compact space, then C(X) is metrizable and complete.

Proof: Any Cauchy sequence converges (p. 35) to a function continuous on any compact set.
Definition: Let U ⊆ Rd be an open set. Any compact set K ⊂ U defines seminorms k kK,r on
C m (X), 1 ≤ m ≤ ∞, (see p. 98 for notations)
4.5. CONVERGENCE IN C M (U ) 115
1

kf kK,r = sup|α|≤r α! k∂α f kK ; r ≤ m, r < ∞,
where the factorials are introduced so that kf gkK,r ≤ kf kK,r kgkK,r . In fact
X α!
∂α (f g) = (∂β f )(∂α−β g).
β!(α − β)!
β≤α
These seminorms define a linear topology on C m (U ), the topology of compact convergence

of derivatives. We have lim fn = f when (fn ) uniformly converges to f , and (∂α fn ), |α| ≤ m,
uniformly converges to ∂α f , on any compact set K ⊂ U .
S o
If we put U = n Kn , Kn ⊆K n+1 , we see that it is defined by the countable family of
seminorms k kKn ,r , because K ⊆ Kn for some index n.
Moreover, C m (U ) admits a countable base of open sets, because this topology is just the
initial topology of the maps ∂α : C m (U ) → C(U ), |α| ≤ m, and C(U ) has a countable base.
Theorem: C m (U ) is metrizable and complete, 1 ≤ m ≤ ∞.
Proof: Any Cauchy sequence (fn ) uniformly converges to a continuous function g, and the partial
derivatives ∂α f , |α| ≤ m, uniformly converge to certain continuous functions gα , and we have
to prove that gα = ∂α g, so that g ∈ C m (U ).
We may use that the derivative preserves uniform limits (p. 35) or we may argue directly.
If p = (a1 , . . . , ad ) ∈ U , then
Z xi
∂fn
fn (a1 , . . . , xi , . . . , ad ) = fn (a1 , . . . , ai , . . . , ad ) + (a1 , . . . , ti , . . . , ad ) dti ,
ai ∂xi
Z xi
g(a1 , . . . , xi , . . . , ad ) = g(a1 , . . . , ai , . . . , ad ) + gi (a1 , . . . , ti , . . . , ad ) dti ,
ai
∂g
and we see that g1 (p) = ∂x1 (p). Iterating, we conclude.
Chapter 5
Algebra II
5.1 Actions of a Group

.
Definitions: An action1 of a group G on a set X, or a G-set, is a map G × X −
→ X such that
1. (g1 g2 ) · x = g1 · (g2 · x) for all g1 , g2 ∈ G, x ∈ X.
2. 1 · x = x, for all x ∈ X.
It defines on X an equivalence relation, x ≡ x0 when x0 = gx for some g ∈ G, and the orbit of

x ∈ X is the equivalence class Gx = {gx : g ∈ G}. The isotropy subgroup is
Ix ={g ∈ G : gx = x}.
Igx =gIx g −1 . (5.1)
An action is transitive if there is a unique orbit, G = Gx for some x ∈ X.

A point x ∈ X is a fixed point when Ix = G, and X G denotes the set of fixed points.
A map f : X → Y is a morphism of G-sets if f (g · x) = g · f (x) for all g ∈ G, x ∈ X; and
the set of all G-morphisms X → Y is denoted by HomG (X, Y ).
Isomorphisms of G-sets are bijective morphisms of G-sets.
If H ⊆ G is a subgroup, G/H is a G-set with the action g1 · [g2 ] = [g1 g2 ].
Theorem: HomG (G/H, X) = X H , f 7→ f ([1]), for any subgroup H ⊆ G.
Proof: The map f : G/H → X, f (gH) = gx, is well defined if and only if x ∈ X H .
Theorem: G/Ix = Gx, [g] 7→ gx.
Proof: If g1 x = g2 x, then g1−1 g2 x = x, g1−1 g2 ∈ Ix , [g1 ] = [g2 ].
Class Formula: If a finite group G acts on a finite set X, then there are non-fixed points
xi ∈ X and divisors di > 1 of the order of G such that
|X| = |X G | + xi [G : Ixi ] = |X G | + i di .
P P
Proof: X is the disjoint union of all the orbits, |X G | is the number of one point orbits, and the
cardinals di = [G : Ixi ] of the other orbits divide |G| by Lagrange’s theorem.
Definition: A p-group is a group of order some power of a prime number p.

1
Right actions correspond to left actions by the formula x · g = g −1 · x.
117
118 CHAPTER 5. ALGEBRA II
Lemma: If G is a p-group, and X a finite G-set, then |X| ≡ |X G | (mod. p).
Proof: In the class formula, any divisor di is a power pni , ni ≥ 1.
Theorem: The center of any p-group G 6= 1 is non trivial,
Z(G) = {a ∈ G : ag = ga, ∀g ∈ G} =
6 1.
Proof: G acts on G by conjugation, and the set of fixed points is Z(G).

By the lemma, |Z(G)| ≡ |G| ≡ 0 (mod. p), and |Z(G)| = 6 1.
Cauchy’s Theorem: If a prime p divides |G|, then G has some subgroup of order p.
Proof: The group Z/pZ acts on X = {(g1 , . . . , gp ) ∈ Gp : g1 . . . gp = 1} by means of a cyclic

permutation,
g1 . . . gp = 1, g1 . . . gp−1 = gp−1 , gp g1 . . . gp−1 = 1.
Since |X| = |G|p−1 is a multiple of p, there is a fixed point (g, . . . , g) 6= (1, . . . , 1).
Hence, g p = 1, g 6= 1, and the order of (g) is p.
Definition: Let p be a prime number. The Sylow p-subgroups of a finite group G are the
subgroups of order the greatest power pn dividing |G| = pn m.
Lemma: Any subgroup H of order pi , i < n, is contained in a subgroup H 0 of order pi+1 such
that H H 0 (i.e., H is a normal subgroup of H 0 ).
Proof: The p-group H acts on G/H, the set of fixed points being N (H)/H, where the subgroup
N (H) = {g ∈ G : gHg −1 = H} is the normalizer of H in G.
Since |G/H| is a multiple of p, so is |N (H)/H|.
By Cauchy’s theorem, N (H)/H has some subgroup H̄ of order p.
Now H 0 = π −1 (H̄) is a subgroup of N (H) of order pi+1 and H H 0 .
Corollary: Any p-group G admits a sequence 1 = H0 H1 . . . Hn−1 Hn = G, where Hi

is a subgroup of order pi .
Corollary (First Sylow Theorem): There exist Sylow p-subgroups of G.
Second Sylow Theorem: All Sylow p-subgroups of G are conjugate.
Proof: Let P 0 , P be Sylow p-subgroups. P 0 acts on G/P and p does not divide |G/P |.
There is a fixed point ḡ ∈ G/P , P 0 ⊆ gP g −1 , and P 0 = gP g −1 since both have order pn .
Third Sylow Theorem: Put |G| = pn m. The number of Sylow p-subgroups of G divides the
common index m and it is congruent to 1 modulo p.
Proof: G acts transitively, by conjugation, on the set X of all Sylow p-subgroups of G.

N (P ) is the isotropy subgroup of P ; hence |X| = [G : N (P )] divides [G : P ] = m.
Let us show that P is the unique fixed point under the action of P .
If gP 0 g −1 = P 0 , ∀g ∈ P , then P ⊂ N (P 0 ), and P , P 0 are Sylow p-subgroups of N (P ).
Hence P 0 = P by the second Sylow theorem. Now |X| ≡ |X P | = 1 (mod. p).
Definition: A finite group G is solvable if there are subgroups 1 = H0 H1 . . . Hn = G

with abelian quotients Hi /Hi−1 .
5.2. MODULES 119
Theorem: Let G be a finite group. If G is solvable, so is any subgroup H.

If H G, then G is solvable if and only if so are H and G/H.
Proof: If 1 = H0 H1 . . . Hn = G is a resolution, we put Hi0 = Hi ∩ H.

Then Hi−1 0 Hi0 , and Hi0 /Hi−1
0 ,→ Hi /Hi−1 ; hence Hi0 /Hi−1
0 is abelian, and H is solvable.
If H G, and π : G → G/H is the canonical projection, we put H̄i = π(Hi ). Then H̄i−1 H̄i
and H̄i /H̄i−1 is a quotient of Hi /Hi−1 ; hence it is abelian, and G/H is solvable.
Conversely, if we have resolutions of H and G/H,
1 = H0 H1 . . . Hn−1 Hn = H,
1 = H̄0 H̄1 . . . H̄d−1 H̄d = G/H,
then 1 H1 . . . Hn = π −1 (H̄0 ) π −1 (H̄1 ) . . . π −1 (H̄d ) = G is a resolution of G.
5.2 Modules
Definitions: Let A be a ring (commutative with unity). A structure of A-module on an
.
abelian group M is given by a product A × M −
→ M such that
1. a(m1 + m2 ) = am1 + am2 ; ∀m1 , m2 ∈ M .
2. (a + b)m = am + bm; ∀a, b ∈ A, m ∈ M .
3. (ab)m = a(bm); ∀a, b ∈ A, m ∈ M .
4. 1 · m = m; ∀m ∈ M
and a subgroup N ⊆ M is a submodule if a ∈ A, m ∈ N ⇒ am ∈ N .
A group morphism f : M → M 0 is a morphism of A-modules (an isomorphism of A-
modules if moreover it is bijective) when it is A-linear,
f (am) = a · f (m) ; ∀a ∈ A, m ∈ M.
Examples: Modules over a field k are just k-vector spaces, submodules are just vector subspaces
and morphisms of k-modules are just k-linear maps.
Any abelian group admits a unique structure of Z-module, submodules are just subgroups
and morphisms of Z-modules are just group morphisms.
Sums and intersections of submodules are submodules.
If I is an ideal, IM = {a1 m1 + . . . + an mn : ai ∈ I, mi ∈ M } is a submodule of M .
The submodules of A are just the ideals of A.
The set HomA (M, N ) of all A-linear maps M → N is an A-module with the obvious structure,
(f + h)(m) = f (m) + h(m)
Q and (af )(m) = a(f (m)).L
Direct products i∈I Mi and direct sums i∈I Mi (formed by all sequences {mi }i∈I
with a finite number of non-zero terms) of A-modules are A-modules.
Any family {mi }i∈I of elements of M defines a morphism of A-modules
L P
f: I A −→ M, f ((ai )i∈I ) = i ai mi ,
P
and the image i Ami is the submodule generated by {mi }i∈I . If f is an isomorphism, we say
that M is a free module of base {mi }i∈I .
M is said to be a finitely generated A-module if M = Am1 + . . . + Amn .
The proofs given in the case of vector spaces (pp. 47–49) show that
1. If N is a submodule of an A-module M , the quotient group M/N admits a unique structure

of A-module such that the canonical projection π : M → M/N is a morphism of A-modules,
and M/N has the corresponding universal property.
2. If f : M → N is a morphism of A-modules, then Ker f and Im f are submodules, and the

isomorphism theorem M/Ker f ' Im f holds.
3. If p : M → M̄ is an A-linear surjective map, we have a lattice isomorphism (where a sub-

module P̄ ⊆ M̄ corresponds to the kernel P = π −1 (P̄ ) of M → M̄ → M̄ /P̄ , so that
M/P = M̄ /P̄ )

Submodules ∼ Submodules of M
(N := Ker p) −−→
of M̄ containing N
4. Modules of finite length are defined to be modules admitting some flag, and the common
length of all flags is the length l(M ) of the module (but simple A-modules are residue
fields A/m of maximal ideals, so that typically A has not finite length).
5. If 0 → M 0 → M → M 00 → 0 is an exact sequence of A-modules, then l(M ) = l(M 0 ) + l(M 00 ).

Theorem: Any ring A 6= 0 has some maximal ideal.
S
Proof: Let X be the ordered set of all ideals I 6= A. If {Ij } is a chain in X, then I = j Ij is
an ideal 6= A (p. 52) containing any ideal Ij . By Zorn’s lemma, X has a maximal element.
Corollary: Any ideal I 6= A is contained in some maximal ideal.
Proof: Any maximal ideal of A/I defines a maximal ideal of A containing I (p. 69).
Corollary: An element f ∈ A is invertible if and only if it is in no maximal ideal of A.
Proof: If f is not a unit, then f A 6= A, and some maximal ideal contains f A.
Corollary: Let A 6= 0. If there is a surjective morphism An → Am , then n ≥ m. Hence, any

base of An has n elements.
Proof: Let m be a maximal ideal of A, and put k = A/m.

If a morphism An → Am is surjective, so is the k-linear map An /mAn → Am /mAm .
Since An /mAn = An /mn = (A/m)n = k n , we have n ≥ m. q.e.d.
Morphisms of A-modules f : M 0 → M naturally induce morphisms of A-modules
f∗ : HomA (N, M 0 ) −→ HomA (N, M ), f∗ (g) = f ◦ g,

f ∗ : HomA (M, N ) −→ HomA (M 0 , N ), f ∗ (g) = g ◦ f.
i p
Theorem: A sequence of morphisms of A-modules M 0 −−→ M −−→ M 00 −→ 0 is exact if and
only if for any A-module N the following sequence is exact
p∗ i∗
0 −→ HomA (M 00 , N ) −−→ HomA (M, N ) −−→ HomA (M 0 , N )
Proof: Assume that Im i = Ker p and p is surjective.

If f : M 00 → N vanishes on Im p = M 00 , then f = 0; hence p∗ is injective.
Since pi = 0, we have 0 = (pi)∗ = i∗ p∗ ; hence Im p∗ ⊆ Ker i∗ .
5.2. MODULES 121
Finally, if f : M → N vanishes on Im i = Ker p, by the universal property f factors through

p : M → M/Ker p ' M 00 , so that f ∈ Im p∗ .
Conversely, since p∗ is injective when N = M 00 /Im p, the canonical map M 00 → N is zero;
hence Im p = M 00 , and p is surjective.
Im i ⊆ Ker p, because pi = (pi)∗ (IdM 00 ) = i∗ p∗ (IdM 00 ) = 0.
π : M → N = M/Im i. Since i∗ (π) = 0, there is a morphism f : M 00 → N such that
π = p∗ (f ) = f p; hence Ker p ⊆ Ker π = Im i,
p
M0
i /M / M 00 /0
π
f
z
M/Im i
i p
Theorem: A sequence of morphisms of A-modules 0 −→ M 0 −−→ M −−→ M 00 is exact if and
only if for any A-module N the following sequence is exact
i p∗
0 −→ HomA (N, M 0 ) −−∗→ HomA (N, M ) −−→ HomA (N, M 00 )
Proof: The direct statement is easy to check. To prove the converse, just take N = A, since we
have natural isomorphisms HomA (A, M ) = M , f 7→ f (1).
i p
Theorem: Let 0 −→ M 0 −−→ M −−→ M 00 −→ 0 be an exact sequence of A-modules. The
following conditions are equivalent (and we say that the sequence splits),
1. There is an A-linear section s : M 00 → M ; ps = IdM 00 .
2. There is an A-linear retract r : M → M 0 ; ri = IdM 0 .
3. p∗ : HomA (N, M ) −→ HomA (N, M 00 ) is surjective for any A-module N .
4. i∗ : HomA (M, N ) −→ HomA (M 0 , N ) is surjective for any A-module N .
5. There is an isomorphism M 0 ⊕ M 00 ' M such that the following diagram commutes,

i π
0 −→ M 0 −−1→ M 0 ⊕ M 00 −−2→ M 00 −→ 0
k |o k
i p
0 −→ M 0 −→ M −−→ M 00 −→ 0
Proof: (1 ⇒ 3) Because p∗ s∗ = (ps)∗ = Id.

(3 ⇒ 1) Just take N = M 00 , and consider the identity of M 00 .
(2 ⇒ 4) Because i∗ r∗ = (ri)∗ = Id.
(4 ⇒ 2) Just take N = M 0 , and consider the identity of M 0 .
(1 ⇒ 5) The required isomorphism is i + s : M 0 ⊕ M 00 → M (see p. 52).
(2 ⇒ 5) The morphism (r, p) : M → M 0 ⊕ M 00 clearly gives a commutative diagram.
Let us prove that it is an isomorphism:
If r(m) = 0 and p(m) = 0, then m = i(m0 ) and m0 = ri(m0 ) = r(m) = 0; hence m = 0.
Given m0 ∈ M 0 and m00 = p(m) ∈ M 00 , there exists x ∈ m0 such that r(m + i(x)) = m0 .
Finally, (5 ⇒ 1) and (5 ⇒ 2) are obvious, since π2 admits the section i2 (m00 ) = (0, m00 ) and
i1 admits the retract π1 (m0 , m00 ) = m0 .
5.2.1 Injective and Projective Modules

Definitions: An A-module P is projective if HomA (P, −) preserves exact sequences; i.e., if a
morphism of A-modules p : M → M 00 is surjective, so is the morphism
p∗ : HomA (P, M ) −→ HomA (P, M 00 ).
Dually, an A-module Q is injective if HomA (−, Q) preserves exact sequences; i.e., if a

morphism of A-modules i : M 0 → M is injective, then the following morphism is surjective
i∗ : HomA (M 0 , Q) −→ HomA (M, Q).

i p
Theorem: If P is projective, any exact sequence 0 −→ M 0 −−→ M −−→ P −→ 0 splits.
i p
Dually, if Q is injective, any exact sequence 0 −→ Q −−→ M −−→ M 00 −→ 0 splits.
Proof: If P is projective, then p∗ : HomA (P, M ) → HomA (P, P ) is surjective, and there is a
morphism s : P → M such that IdP = p∗ (s) = ps.
If Q is injective, then i∗ : HomA (M, Q) → HomA (Q, Q) is surjective, and there is a morphism
r : M → Q such that IdQ = i∗ (r) = ri.
⊕i Pi is projective if and only if so are the modules Pi .

Theorem: Q
Dually, i Qi is injective if and only if so are the modules Qi .
Proof: It follows from the universal properties of the direct sum and the direct product,
Q
HomA (⊕i Pi , M ) = i HomA (Pi , M )
Q Q
HomA (M, i Qi ) = i HomA (M, Qi )
Theorem: Any free module is projective. Any module is a quotient of a projective module.
Proof: A is projective because HomA (A, M ) = M ; hence ⊕I A is projective.

Moreover, any generating system {mi }i∈I of M defines an epimorphism ⊕I A → M .
Ideal Criterion: If the restriction morphism HomA (A, Q) → HomA (I, Q) is surjective for any
ideal I of A, then Q is an injective A-module.
Proof: If N is a submodule of M , we must show that any morphism f : N → Q is the restriction

of a morphism M → Q.
Take m ∈ M not in N , and consider the ideal I = {a ∈ A : am ∈ N } and the morphism
φ : I → Q, φ(a) = f (am). By hypothesis it may be extended to a morphism φ0 : A → Q.
Since φ vanishes on the kernel of A → Am, it induces a morphism φ0 : Am → Q coinciding
with f on N ∩ Am = Im. Now the exact sequence
j s
0 −→ N ∩ Am −−−→ N ⊕ Am −−−→ N + Am −→ 0
where j(n) = (n, −n), shows that f + φ0 : N ⊕ Am → Q defines a morphism N + Am → Q

coinciding with f on N . Now, applying Zorn’s lemma to the pairs (M 0 , f 0 ), where M 0 is a
submodule of M containing N and f 0 : M 0 → Q extends f , with the order
(M10 , f10 ) ≤ (M20 , f20 ) ⇔ M10 ⊆ M20 , and f20 extends f10 ,
we see that there exists (M 0 , f 0 ) maximal, and M 0 = M since we may not extend f 0 .
5.2. MODULES 123
a·
Definition: An A-module Q is divisible if the morphisms Q −
→ Q, a 6= 0, are surjective.
Corollary: Divisible modules over a principal ideal domain are injective.

a·
Proof: Q −
→ Q is just the morphism HomA (A, Q) → HomA (aA, Q) ' HomA (A, Q), a 6= 0.
Corollary: If p is an irreducible element of a principal ideal domain A, then B = A/pn A is an

injective B-module.
Proof: Given an ideal pr A/pn A of B and a morphism of B-modules φ : pr A/pn A → B, we have

0 = φ(p̄n ) = p̄n−r φ(p̄r ); hence φ(p̄r ) = bp̄r for some b ∈ B.
The required extension φ0 : B → B is just φ0 (x) = bx.
5.2.2 Localization
Definition: The localization MS of an A-module M by a multiplicative system S of A is the
quotient of M × S by the equivalence relation
(m, s) ≡ (n, t) ⇔ there are u, v ∈ S such that mu = nv, su = tv,
and it is an AS -module with the operations

m n tm + sn a m am
+ = , · =
s t st s t st
where m m
s is the class of (m, s). Therefore s = 0 if and only if um = 0 for some u ∈ S.
Any morphism of A-modules f : M → N defines a morphism of AS -modules
f (m)
fS : MS −→ NS , fS m s = s ·
m
We have a canonical localization morphism of A-modules M → MS , m 7→ 1·
Universal Property: Let N be an AS -module. Any morphism of A-modules f : M → N

uniquely factors through a morphism of AS -modules φ : MS → N , φ( m
1 ) = f (m).
HomAS (MS , N ) = HomA (M, N ).
Proof: The unique possible morphism, φ( m −1

s ) = s f (m), is well defined,
−1 −1 −1 −1
φ um

us = (us) f (um) = s u uf (m) = s f (m).
Notation: Mf denotes the localization by S = {1, f, . . . , f n , . . .}.

If p is a prime ideal, Mp denotes the localization by S = A − p.
f g
Theorem: If M 0 −−→ M −−→ M 00 is an exact sequence, so is the sequence
f g
MS0 −−−
S
→ MS00 .
S
→ MS −−−
m g(m)
Proof: If s ∈ Ker gS , then s = 0, and 0 = tg(m)
= g(tm) for some t ∈ S.
f (m0 ) m0
Hence tm = f (m0 ), m tm
and s = ts = ts = fS ts ∈ Im fS . q.e.d.
If N is a submodule of M , then NS may be identified with a submodule of MS , and

1. (N + N 0 )S = NS + NS0 .
2. (N ∩ N 0 )S = NS ∩ NS0 .
3. (M ⊕ M 0 )S = MS ⊕ MS0 .
4. (M/N )S = MS /NS .
5. (Ker f )S = Ker fS .
6. (Im f )S = Im fS .
Proof: The non obvious equalities follow localizing the exact sequences
0 −→ M 0 −→ M 0 ⊕ M −→ M −→ 0
0 −→ N ∩ N 0 −→ N ⊕ N 0 −→ N + N 0 −→ 0
0 −→ N −→M −→ M/N −→ 0
0 −→ Ker f −→ M −→ N
5.2.3 Tensor Product

Definitions: An ordered set I is filtering if for any pair i, j ∈ I there is k ∈ I such that i, j ≤ k.
An inductive system of sets is a family of sets {Xi }i∈I and maps φji : Xi → Xj , i ≤ j,
such that φii = IdXi and, whenever i ≤ j ≤ k,
Xi / Xj φki = φkj φji

Xk

The inductive limit lim
−→
Xi is the quotient set q i Xi / ≡ of the disjoint union, where
xi ≡ xj when xi = xj in Xk , for some k ≥ i, j,
(if xi = xj in Xk , and xj = xl in Xk0 , then xi = xj = xl in Xr for any index r ≥ k, k 0 .)

We have canonical maps φj : Xj → lim −→
Xi , φj (xj ) = [xj ], such that
Xi / Xj φi = φj φji

lim
−→
Xi
Definitions: A projective system of sets is a family of sets {Xi }i∈I and maps φij : Xj → Xi ,
i ≤ j, such that φii = IdXi and, whenever i ≤ j ≤ k,
Xk φik = φij φjk

Xj / Xi
Q
The subset of i Xi formed by all congruent sequences (xi ), φij (xj ) = xi when i ≤ j, is the
projective limit lim
←−
Xi . (If I is the empty set, lim
←−
Xi := {p}.)
5.2. MODULES 125
We have canonical maps φj : lim

←−
Xi → Xj , φj (xi ) = xj , such that
lim
←−
Xi φi = φij φj

Xj / Xi
If the maps φji are group (ring,...) morphisms, then the limits inherit a group (ring,...)
structure and the canonical maps φi are group (ring,...) morphisms.
Examples: The ring morphisms K[x]/(xn ) → K[x]/(xm ), m ≤ n, form a projective system of
rings, and the projective limit is the ring of formal series K[[x]].
Any module M is the inductive limit of its submodules Mi ⊆ M , and the projective limit of
its quotients M → Ni , since lim
−→
Xi = Xk and lim←−
Xi = Xk when I has a final element k.
Universal Property: Let (Xi , φji ) be an inductive system. Given maps fi : Xi → Y such that
fi = fj φji when i ≤ j, there exists a unique map f : lim
−→
Xi → Y such that fi = f φi ,
Hom(lim
−→
Xi , Y ) = lim
←−
Hom(Xi , Y ), f 7→ (f φi ).
Dually, let (Xi , φij ) be a projective system. Given maps fi : Y → Xi such that fi = φij fj
when i ≤ j, there exists a unique map f : Y → lim ←−
Xi such that fi = φi f ,
Hom(Y, lim
←−
Xi ) = lim
←−
Hom(Y, Xi ), f 7→ (φi f ).
Proof: In the first case, the unique possible map is f ([xi ]) = fi (xi ), and in the second case,
f (y) = (fi (y)). Both are well defined since fi = fj φij , and fi = φji fj . q.e.d.
Let M, N, P be three A-modules. The A-bilinear maps M × N → P form an A-module

F (P ) = BilA (M, N ; P ). We shall construct a bilinear map ξ : M × N → M ⊗A N such that any
bilinear map g : M × N → P uniquely factors through a morphism f : M ⊗A N → P ,
HomA (M ⊗A N, P ) = BilA (M, N ; P ), f 7→ f ◦ ξ.
Remark that f is surjective if and only if the image of g spans P ; i.e., g does not factor
through a strict submodule of P . Since M ⊗A N must be the projective limit of its quotients, we
consider pairs Pg , where P is an A-module and g : M × N → P is A-bilinear, and morphisms
of pairs f : Pg0 0 → Pg are defined to be A-linear maps f : P 0 → P such that g = f ◦ g 0 .
Definition: A pair Qξ is minimal if any injective morphism of pairs Pg → Qξ is an isomorphism

(so that Qξ is a quotient of M ⊗A N , and we should have M ⊗A N = lim ←−
Qξ ).
i p
Lemma: If 0 −→ P 0 −−→ P −−→ P 00 is an exact sequence, then so is the sequence
i p∗
0 −→ F (P 0 ) = BilA (M, N ; P 0 ) −−∗→ F (P ) = BilA (M, N ; P ) −−→ F (P 00 ) = BilA (M, N ; P 00 )
and F preserves direct products and projective limits,
BilA (M, N ; P × P 0 ) = BilA (M, N ; P ) × BilA (M, N ; P 0 ),

BilA (M, N ; lim
←−
Pi ) = lim
←−
BilA (M, N ; Pi ).
Proof: Straightforward. q.e.d.

If Qξ is a minimal pair, any two morphisms of pairs f1 , f2 : Qξ → Pg coincide:

i f2 −f1
0 −→ Ker (f2 − f1 ) −→ Q −−−−−→ P
i f ∗ −f ∗
0 −→ F (Ker (f2 − f1 )) −−∗→ F (Q) −−2−−−
1
→ F (P )
(f2∗ − f1∗ )(ξ) = f2 ξ − f1 ξ = g − g = 0
and there is ξ 0 ∈ F (Ker (f2 − f1 )) such that i : Ker (f2 − f1 )ξ0 → Qξ is a morphism of pairs.
Since Qξ is minimal, Ker (f2 − f1 ) = Q, and f1 = f2 .
We order minimal pairs (identifying isomorphic pairs): Q0ξ0 ≥ Qξ if there is a morphism of
pairs f 0 : Q0ξ0 → Qξ . If also Qξ ≥ Q0ξ0 , there is a morphism f : Qξ → Q0ξ0 , and we have morphisms
f 0 f : Qξ → Qξ , f f 0 : Q0ξ0 → Q0ξ0 ; hence f 0 f = IdQ , f f 0 = IdQ0 , and Qξ = Q0ξ0 .
Lemma: Any pair Pg is dominated by a minimal pair: there exists a morphism of pairs Qξ → Pg ,
where Qξ is minimal.
Proof: Just take the submodule Q generated by the image of g : M × N → P , with the bilinear
map ξ : M × N → Q, ξ(m, n) = g(m, n). q.e.d.
Let us show that this is a filtering order because F preserves direct products:
If Qξ and Q0ξ0 are minimal pairs, we have morphisms
3 Qξ
(Q × Q0 )ξ×ξ0
+
Q0ξ0
and any minimal pair dominating (Q × Q0 )ξ×ξ0 dominates Qξ and Q0ξ0 .

Hence the minimal pairs form a projective system {(Qi )ξi }, and the projective limit M ⊗A N
is a pair with ξ = (ξi ) ∈ F (M ⊗A N ) = lim
←−
F (Qi ).
A minimal pair dominating (M ⊗A N )ξ dominates any other minimal pair; hence there is a
final minimal pair, obviously the projective limit (M ⊗A N )ξ , and for every pair Pg we have a
unique morphism of pairs (M ⊗A N )ξ → Pg . We see that ξ : M × N → M ⊗A N is the universal
bilinear map and, if we put m ⊗ n := ξ(m, n),
(am + a0 m0 ) ⊗ n = a(m ⊗ n) + a0 (m0 ⊗ n),

m ⊗ (an + a0 n0 ) = a(m ⊗ n) + a0 (m ⊗ n0 ).
Universal Property: Any bilinear map g : M × N −→ P uniquely factors through a morphism

of A-modules f : M ⊗A N → P , f (m ⊗ n) = g(m, n),
HomA (M ⊗A N, P ) = BilA (M, N ; P ).
If f : M → M 0 , h : N → N 0 are morphisms of A-modules, then M × N → M 0 ⊗A N 0 ,

(m, n) 7→ f (m) ⊗ h(n), is A-bilinear, and it induces a morphism of A-modules
f ⊗ h : M ⊗A N 0 → M 0 ⊗A N 0 , (f ⊗ h)(m ⊗ n) = f (m) ⊗ h(n).
Lemma: BilA (M, N ; P ) = HomA (M, HomA (N, P )).
Proof: The morphism f : M → HomA (N, P ) corresponds to g(m, n) = f (m)(n).

5.2. MODULES 127
i p
Theorem: If M 0 −−→ M −−→ M 00 −→ 0 is an exact sequence, so is exact the sequence
i⊗1 p⊗1
M 0 ⊗A N −−−→ M ⊗A N −−−→ M 00 ⊗A N −→ 0
Proof: If E is the exact sequence M 0 → M → M 00 → 0, for any A-module P the sequence

HomA (E, HomA (N, P )) = HomA (E ⊗A N, P ) is exact; hence E ⊗A N is exact. q.e.d.
1. (M ⊗A N ) ⊗A P = M ⊗A (N ⊗A P ), where (m ⊗ n) ⊗ p = m ⊗ (n ⊗ p).
2. M ⊗A N = N ⊗A M , where m ⊗ n = n ⊗ m.
P P
3. M ⊗A (⊕i Ni ) = ⊕i (M ⊗A Ni ), where m ⊗ ( i ni ) = i m ⊗ ni .
4. M ⊗A (lim
−→
Ni ) = lim
−→
(M ⊗A Ni ).
5. A ⊗A M = M , where a ⊗ m = am.
6. (A/I) ⊗A M = M/IM , where ā ⊗ m = [am].

Proof of the Properties of the Tensor Product of Modules:
(1) Hom((M ⊗ N ) ⊗ P, X) = Hom(M ⊗ N, Hom(P, X)) = Hom(M, Hom(N, Hom(P, X)))
= Hom(M, Hom(N ⊗ P, X)) = Hom(M ⊗ (N ⊗ P ), X).
(2) Hom(M ⊗ N, X) = BilA (M, N ; X) = BilA (N, M ; X) = Hom(N ⊗ M, X).

Q
(3) Hom(M ⊗ (⊕i Ni ), X) = Hom(M, Hom(⊕i Ni , X)) = Hom(M, i Hom(Ni , X))
Q Q
= i Hom(M, Hom(Ni , X)) = i Hom(M ⊗ Ni , X) = Hom(⊕i (M ⊗ Ni ), X).
(4) Hom(M ⊗ (lim

−→
Ni ), X) = Hom(M, Hom(lim
−→
Ni , X)) = Hom(M, lim
←−
Hom(Ni , X))
= lim
←−
Hom(M, Hom(Ni , X)) = lim
←−
Hom(M ⊗ Ni , X) = Hom(lim
−→
(M ⊗ Ni ), X).
(5) Hom(A ⊗ M, X) = Hom(A, Hom(M, X)) = Hom(M, X).
(6) The sequence I ⊗A M −→ A ⊗A M = M −→ (A/I) ⊗A M −→ 0 is exact.
Definition: An A-module P is flat if (−) ⊗A P preserves exact sequences; i.e., if M 0 → M is

an injective morphism of A-modules, so is M 0 ⊗A P → M ⊗A P .
Any free (or projective) module is flat.
If i : E 0 → E is an injective linear map, then i ⊗ 1 : E 0 ⊗k F → E ⊗k F is also injective, since
i admits a retract (p. 52). Any k-vector space is a flat k-module.
Definition: Let A → B be a ring morphism. P The morphisms

P 1 ⊗ b : M ⊗A B → M ⊗A B
define a structure of B-module on M ⊗A B, b m
i i ⊗ b i = i mi ⊗ (bbi ). We denote it by
MB = M ⊗A B, and we say that it is the base change of M .
The morphism of A-modules M → MB , m 7→ m ⊗ 1, is the base change morphism.
Any morphism of A-modules f : M → M 0 induces a morphism of B-modules
fB = f ⊗ 1 : MB −→ MB0 , fB
P P
i mi ⊗ bi = i f (mi ) ⊗ bi .
Universal Property: Let N be a B-module. For any morphism of A-modules f : M → N

there exists a unique morphism of B-modules φ : MB → N such that φ(m ⊗ 1) = f (m),
HomA (M, N ) = HomB (MB , N ).

Proof: By the universal property of the tensor product, there exists a unique morphism of
A-modules φ : M ⊗A B → N , φ(m ⊗ b) = bf (m), and it is a morphism of B-modules.
Corollary: MS = M ⊗A AS .
Proof: HomAS (MS , N ) = HomA (M, N ) = HomAS (M ⊗A AS , N ).
Theorem: (M ⊗A B) ⊗B N = M ⊗A N, (m ⊗ b) ⊗ n = m ⊗ (bn), where N is a B-module.
Proof: In the isomorphism HomA (M ⊗A N, X) = HomA (M, HomA (N, X)) it is easy to check
that HomB (M ⊗A N, X) corresponds to HomA (M, HomB (N, X)); hence
HomB (MB ⊗ N, X) = HomB (MB , HomB (N, X))

= HomA (M, HomB (N, X)) = HomB (M ⊗A N, X).
Corollary: (MB )C = MC , where (m ⊗ b) ⊗ c = m ⊗ (bc).
Corollary: (MB ) ⊗B (MB0 ) = (M ⊗A M 0 )B , where (m ⊗ b1 ) ⊗ (m0 ⊗ b2 ) = (m ⊗ m0 ) ⊗ b1 b2 .
Proof: (M ⊗A B) ⊗B (MB0 ) = M ⊗A (MB0 ) = (M ⊗A M 0 ) ⊗A B.
Definition: We fix a ring k (a field in this course) and we name k-algebra any ring morphism
k → A, and morphisms of k-algebras are ring morphisms A → B such that the following
triangle is commutative
k

A /B
If A, B are k-algebras, the morphism of k-modules
A ⊗k B ⊗k A ⊗k B −→ A ⊗k B, a1 ⊗ b1 ⊗ a2 ⊗ b2 7→ a1 a2 ⊗ b1 b2 ,
induces a k-bilinear map (A ⊗k B) × (A ⊗k B) → A ⊗k B. This product

P P P
i ai ⊗ bi · j aj ⊗ bj = i,j ai aj ⊗ bi bj
defines a ring structure on A ⊗k B, and it is also a k-algebra with the natural morphism
k −→ A ⊗k B, λ 7→ λ ⊗ 1 = 1 ⊗ λ.
Universal Property: If f : A → C, h : B → C are morphisms of k-algebras, there is a unique

morphism of k-algebras φ : A ⊗k B → C such that φ(a ⊗ 1) = f (a), φ(1 ⊗ b) = h(b).
Homk-alg (A ⊗k B, C) = Homk-alg (A, C) × Homk-alg (B, C).
Proof: By the universal property of the tensor product, there exists a unique morphism of
k-modules φ : A ⊗k B → C, φ(a ⊗ b) = f (a)h(b), and it is a morphism of k-algebras.
Corollary: k[x1 , . . . , xn ] ⊗k L = L[x1 , . . . , xn ].

(k[x]/(P )) ⊗k L = L[x]/(P ).
5.3. CATEGORIES AND FUNCTORS 129
5.3 Categories and Functors

Definition: A category C is given by a family of objects, a set HomC (M, N ) for each pair
M, N of objects (the morphisms M → N ), and a map (named composition)
HomC (M, N ) × HomC (N, P ) −→ HomC (M, P ), (f, g) 7→ g ◦ f.
for any three objects M, N, P ; satisfying the following conditions:
1. Composition of morphisms is associative: (f ◦ g) ◦ h = f ◦ (g ◦ h).
2. For any object M there is a identity2 morphism IdM : M → M such that f ◦ IdM = f , and
IdM ◦ g = g, for all morphisms f : M → N , g : N → M .
So we have the category Sets of sets, the category Groups of groups, of rings, of A-modules,
of topological spaces, etc.
Definitions: A morphism f : M → N is an isomorphism (automorphism when M = N ) if

there is a morphism g : N → M such that f g = idN , gf = idM . This inverse morphism g is
obviously unique and we put g = f −1 .
The opposite category of C has the same objects, but HomCop (M, N ) = HomC (N, M ), the
composition f ◦ g of two morphisms in Cop being the composition g ◦ f in C.
Let C,C’ be two categories.
A covariant functor F : C C0 associates an object F (M ) of C0 to any object M of C,
and a morphism F (f ) : F (M ) → F (N ) of C0 to any morphism f : M → N of C, so that
1. F (IdM ) = IdF (M ) for any object M of C.

g f
2. F (f ◦ g) = F (f ) ◦ F (g) for any pair of morphisms M →
− N−
→ P of C.
Analogously, contravariant functors associate a morphism F (f ) : F (N ) → F (M ) to any

morphism f : M → N , so that F (f ◦ g) = F (g) ◦ F (f ). Contravariant functors C C0 are just
op 0
covariant functors C C , so that any functor may be always assumed to be covariant.
Given two (covariant) functors F, G : C C0 , to give a natural transformation or mor-
phism of functors θ : F → G is just to give a morphism θM : F (M ) → G(M ) in C’ for any
object M in C, such that for any morphism f : M → N in C the following square commutes
F (f )
F (M ) / F (N )
θM θN θN ◦ F (f ) = G(f ) ◦ θM
G(f )
G(M ) / G(N )
and it is an isomorphism of functors if θM is an isomorphism in C’ for any object M , so that

−1
θ−1 : G → F , (θ−1 )M := (θM )−1 , also is an isomorphism of functors, and θN ◦F (f )◦θM = G(f ).
0 3
A covariant functor F : C C is an equivalence of categories when there exist a covariant
functor G : C0 C and isomorphisms of functors θ : G ◦ F − ∼
→ IdC and θ0 : F ◦ G −∼
→ IdC0 ().
0
Given an equivalence F : C C , for any two objects M, N of C, we have natural maps
F G θ
HomC (M, N ) −−→ HomC0 (F (M ), F (N )) −−→ HomC (GF (M ), GF (N )) ' HomC (M, N )
2
The identity is unique: if Id0M is another identity, then Id0M = Id0M ◦ IdM = IdM .
3
and we say that a contravariant functor F : C C0 is an antiequivalence when it induces an equivalence
op 0
C C.
−1
and the composition is the identity, θN ◦ G(F (f )) ◦ θM = f , because θ is functorial.
Therefore F is injective (hence so is G) and we see that we have bijections4
F : HomC (M, N ) −→ HomC0 (F (M ), F (N )) , f 7→ F (f ).
Definitions: Let X be an object of a category C. We say that X • (T ) = HomC (T, X) is the set
of points of X parameterized by T , or just T -points of X.
Any morphism t : S → T in C induces a natural map
X • (T ) = HomC (T, X) −→ HomC (S, X) = X • (S), x 7→ x|t = x ◦ t,
so that X • : C Sets is a contravariant functor, named functor of points of X, and we say

that x|t is the specialization of the T -point x at the point t of the parameter space T . A point
of X is generic if any other point of X is a specialization. So, the identity xg : X → X is a
generic point of X, since any point x : T → X is a specialization, x = IdX ◦ x = xg |x .
Any morphism f : X → Y naturally induces a morphism f = f∗ : X • → Y • , f (x) := f ◦ x.
Yoneda’s Lemma: For any contravariant functor F : C Sets we have a canonical bijection
Homnat (X • , F ) = F (X), θ 7→ θ(IdX ).
In particular, HomC (X, Y ) = Homnat (X • , Y • ) for any two objects X, Y of C.
Proof: Any morphism of functors X • → F preserves specializations, hence it is fully determined

by the image of a generic point, and the considered map is injective.
Finally, any element ξ ∈ F (X) defines a morphism θ : X • → F , θ(x) = F (x)(ξ) ∈ F (T ) for
any point x : T → X. Now θ(Id) = F (Id)(ξ) = ξ, so that the map is surjective.
Definitions: The direct product of two objects X, Y is a pair of morphisms
X ×Y
π1 π2
~
X Y
with the universal property (π1 , π2 ) : HomC (T, X × Y ) = HomC (T, X) × HomC (T, Y ); i.e. the
points of the direct product are the direct product of points, (X × Y )• = X • × Y • .
The final object pt is defined to be an object with a unique T -point T → pt for any object
T ; i.e. with the universal property HomC (T, pt) = ∗. `
Dually we define the direct sum X ⊕ Y (or coproduct X Y ) as a pair of morphisms
X> ⊕ _ Y
j1 j2
X Y
with the universal property (j1∗ , j2∗ ) : HomC (X ⊕ Y, T ) = HomC (X, T ) × HomC (Y, T ), and the
initial object ∅ is defined by the universal
Q L T ) = ∗.
property HomC (∅,
In general,
Q the product i Xi and coproduct i Xi of any set of objects {Xi }i∈I
we define L
(so that i Xi = pt and i Xi = ∅ when I is empty) by the universal properties
Q Q L Q
Hom(T, i Xi ) = i Hom(T, Xi ) , Hom( i Xi , T ) = i Hom(Xi , T ).
4
We have isomorphisms F (θM ), θF0 (M ) : F GF (M ) = F (M ), G(θM 0 0 0
0 ), θG(M 0 ) : GF G(M ) = G(M ), and it would
0 0
be sensible to assume that F (θM ) = θF (M ) , G(θM 0 ) = θG(M 0 ) . Given the isomorphism of functors θ, just consider
0 −1
the isomorphisms θM 0 := G (θG(M 0 ) ), since θGF (M ) = GF (θM ) by the functorial character of θ.
Projective limits and inductive limits are defined by the universal properties
Hom(T, lim
←−
Xi ) = lim
←−
Hom(T, Xi ) , Hom(lim
−→
Xi , T ) = lim
←−
Hom(Xi , T ).
Definition: Fix an object S in C. The morphisms X → S, named S-objects, with the sets
HomS (X, Y ) of S-morphisms, i.e. of morphisms f : X → Y fitting in a commutative diagram
f
X /Y

S
form a category C|S , and the direct product in C|S of two S-objects f : X → S, g : Y → S is
said to be the fibred product of the morphisms f and g. It is a S-object X ×S Y → S endowed
with S-morphisms π1 : X ×S Y → X, π2 : X ×S Y → Y such that, for any S-object T , we have
a bijection
(π1 , π2 ) : HomS (T, X ×S Y ) = HomS (T, X) × HomS (T, Y ).
When T = X, and we fix the identity morphism T → X, we obtain the very important
Graph Formula: HomS (X, Y ) = HomX (X, X ×S Y ), f 7→ IdX ×f .
1. F (E) = Λp E, F (X) = X/G, F (X) = X G , F (N ) = HomA (M, N ), F (M ) = MS , F (P ) =

BilA (M, N ; P ), F (N ) = M ⊗A N , F (M ) = MB are covariant functors.
2. F (E) = E ∗ , F (E) = Tp E, F (N ) = HomA (N, M ) are contravariant functors.
3. The Reflexivity Theorem shows that the functor E E ∗ induces an antiequivalence of the
category of finite dimensional k-vector spaces with itself, the inverse functor being F itself.
The Fundamental Theorem of Projective Geometry points that the lattice of linear subva-
rieties defines an equivalence of the category of projective spaces of dimension n ≥ 2 and
projectivizations of semilinear transformations with a certain a certain category of ordered
sets and isomorphisms, but the construction of the inverse functor requires additional work
(see [2]).
Given a group G, the categories of left G-sets and right G-sets are equivalent, a left action
corresponding to the right action x · g = g −1 · x.
4. In the category of sets (and topological spaces), the direct sum is the disjoint union, the
initial object is the empty set and the final object is the one point set.
5. In the category Sets of sets, the fibred product of two maps f : X → S, g : Y → S is the set
X ×S Y = {(x, y) ∈ X ×Y : f (x) = g(y)}, with the projections π1 : X ×S Y → X, π1 (x, y) = x,
and π2 : X ×S Y → Y , π2 (x, y) = y. In particular, X ×pt Y = X × Y , X ×S y = f −1 (y) when
y is a point of S, and X ×S Y = X ∩ Y when X and Y are subsets of S.
6. In an arbitrary category, the universal property of X ×S Y shows that for any object T we
have a natural bijection (π1 , π2 ) : Hom(T, X ×S Y ) = Hom(T, X) ×Hom(T,S) Hom(T, Y ); i.e.
the points of the fibred product are the fibred product of points, (X ×S Y )• = X • ×S • Y • .
7. In the opposite category of rings, the direct sum is A × B (that we also denote A ⊕ B),
k-algebras are just k-objects, and the fibred product is A ⊗k B. Hence, the graph formula
states that Homk-alg (A, B) = HomB -alg (A ⊗k B, B).
5.3.1 Grothendieck Representability Theorem

Definition: A contravariant functor F : C Sets is representable when it is isomorphic to
the functor of points Q• of some object Q. By Yoneda’s lemma, any functorial isomorphism
F ' Q• is defined by a unique element ξ ∈ F (Q), and we say that the pair Qξ represents F :
for any η ∈ F (T ), there is a unique morphism f : T → Q such that F (f )(ξ) = η,
HomC (T, Q) = F (T ), f 7→ F (f )(ξ).
Dually, a covariant functor F : C Sets is representable when for some pair Qξ , where
ξ ∈ F (Q), the following natural maps are bijective,
HomC (Q, M ) −→ F (M ), f 7→ F (f )(ξ).
Theorem: If it exists, the representant of a functor is unique, up to canonical isomorphisms:

If two pairs Qξ and Q0ξ0 represent a covariant (resp. contravariant) functor F , then there is a
unique isomorphism f : Q → Q0 such that F (f )(ξ) = ξ 0 (resp. F (f )(ξ 0 ) = ξ).
Proof: There is a unique morphism f : Q → Q0 such that F (f )(ξ) = ξ 0 , and a unique morphism
f 0 : Q0 → Q such that F (f 0 )(ξ 0 ) = ξ.
Hence F (f 0 f )(ξ) = ξ, F (f f 0 )(ξ 0 ) = ξ 0 , so that f 0 f = IdQ , f f 0 = IdQ0 .
Examples: All universal properties state that certain pair Qξ represents a functor F (in quo-
tients, ξ is the canonical projection, in localizations it is the localization morphism, in base
changes it is the base change morphism, etc.). Hence any universal property determines a well
defined (up to a canonical isomorphism) object, if it exists.
Definitions: In the category of sets, a sequence of maps

i f,g
X −→ Y ⇒ Z
is exact when i is injective with image Im i = {y ∈ Y : f (y) = g(y)}.

In an arbitrary category C, a sequence of morphisms
i f,g
M −→ N ⇒ P
is exact (i is the kernel of f and g) when, for any object X, so is the sequence of maps
i f∗ ,g∗
HomC (X, M ) −−∗→ HomC (X, N ) ⇒ HomC (X, P )
and we say that a covariant functor is left exact when it preserves finite direct products (hence
the final object) and kernels. By definition, any representable covariant functor is left exact and
preserves projective limits, and Grothendieck’s representability theorem states that the converse
is true under a very mild condition. Dually a sequence of morphisms
f,g p
M ⇒ N −−→ P
is exact (p is the cokernel of f and g) when so are sequences of maps

p∗ f ∗,g ∗
HomC (P, X) −−→ HomC (N, X) ⇒ HomC (M, X)
and a covariant functor is right exact when it preserves finite coproducts (hence the initial
object) and cokernels, and it is exact when it is left and right exact. Finally, a contravariant
functor F : C C0 is left or right exact when so is the covariant functor F : Cop C0 .
Examples: In the category of A-modules, the first condition states that the sequence of A-linear
i f −g f −g p
maps 0 → M → − N −−→ P is exact, and the second one states that so is M −−→ N → − P → 0.
In the category of sets, the cokernel is the quotient of N by the coarse equivalence relation
such that f (m) ≡ g(m). In the category of topological spaces, it is the same quotient endowed
with the quotient topology (the finest topology such that p is continuous), and in the category
of G-sets with the obvious action of G.
In an arbitrary category C, the action of a finite group G = {g1 , . . . , gn } on an object X
is defined to be a group morphism G → AutC (X), so that we have morphisms gi : X → X.
Now the objects X G and X/G are defined to be the kernel and cokernel respectively of these
morphisms g1 , . . . , gn : X → X.
Representability Theorem: If a covariant functor F : A-mod Sets is left exact and any
pair is dominated by a minimal pair, then F is an inductive limit of representable subfunctors,
F (M ) = lim
−→
HomA (Qi , M ).
If moreover F preserves projective limits, F (lim
←−
Mi ) = lim
←−
F (Mi ), then it is representable,
F (M ) = HomA (Q, M ).
If a contravariant functor F : A-mod Sets is left exact and any pair is dominated by a
minimal pair, then F is an inductive limit of representable subfunctors. If moreover F transforms
inductive limits into projective limits, F (lim
−→
Mi ) = lim
←−
F (Mi ), then it is representable,
F (M ) = HomA (M, Q).
Proof: In the covariant case, the proof given in p. 126 holds (pairs exist since F (0) = {∗} =
6 ∅!)
replacing the exact sequence 0 → Ker(f2 − f1 ) → Q → P by the exact sequence
i f1 ,f2
Ker(f2 − f1 ) −→ Q ⇒ P,
and it also holds in the contravariant case with the following obvious changes:
A pair Qξ is minimal if any surjective morphism of pairs Qξ → Pg is an isomorphism.
A pair Qξ dominates a pair Pg if there is a morphism of pairs Pg → Qξ .
Minimal pairs are just submodules of the representant; hence, if it exists, it is the inductive
limits of all minimal pairs. In this case, the coincidence of all morphisms f1 , f2 : Pg → Qξ , when
Qξ is minimal, follows from the exact sequence P ⇒ Q → Q/Im(f2 − f1 ) → 0.
Corollary: Any covariant (resp. contravariant) left exact functor A-mod Sets preserving
projective limits (resp. taking inductive limits into projective limits) is representable.
Proof: We have to show that any pair is dominated by a minimal pair.

In the covariant case, given a pair Pξ , consider the set of submodules M ⊆ P such that
ξ ∈ F (M ) with the inclusion order (since F (M ) → F (P ) is injective, we identify F (M ) with a
subset of F (P )). Now a minimal element M defines a minimal pair Mξ dominating
T Pξ , and it
exists by Zorn’sTlemma because any chain {Mi } hasTa lower bound M := i Mi , since we have
ξ ∈ F (M ) = F ( i Mi ) = F (lim
←−
Mi ) = lim
←−
F (Mi ) = i F (Mi ).
S M such that ξ ∈ F (P/M ) ⊆ F (P ), and

In the contravariant case we consider submodules
any chain {Mi } has an upper bound because M/( i Mi ) = lim
−→
(M/Mi ).
Definition: A covariant functor F : A-mod A-mod is A-linear when so are all the maps
F : HomA (M, N ) → HomA (F (M ), F (N )), and F is (linearly) representable if we have an
isomorphism of functors F ' HomA (Q, −) for some A-module Q (we have natural A-linear
isomorphisms F (M ) = HomA (Q, M ), not just natural bijections, so that, when a pair Qξ rep-
resents F , we have that F (Q) = HomA (Q, Q) has a structure of (eventually non commutative)
A-algebra with unity ξ ∈ F (Q)).
We have analogous definitions for contravariant functors F : A-mod A-mod.
Corollary: A covariant linear functor F : A-mod A-mod is representable if and only if it

preserves projective limits and for any exact sequence of A-modules 0 → M 0 → M → M 00 , so is
the sequence 0 → F (M 0 ) → F (M ) → F (M 00 ).
A contravariant linear functor F : A-mod A-mod is representable if and only if it
transforms inductive limits into projective limits and for any exact sequence of A-modules
M 0 → M → M 00 → 0, so is the sequence 0 → F (M 00 ) → F (M ) → F (M 0 ).
Proof: Any covariant Z-linear functor F preserves split exact sequences, hence F (M × N ) =
F (M ) × F (N ) and F (0) = 0, and it is left exact if and only if 0 → F (M 0 ) → F (M ) → F (M 00 )
is exact whenever so is 0 → M 0 → M → M 00 , because F (f − g) = F (f ) − F (g).
Finally, the bijection HomA (Q, M ) = F (M ), f 7→ F (f )(ξ), is A-linear when so is F .
An analogous argument holds in the contravariant case.
Theorem: Any A-module M is a submodule of an injective A-module.
Proof: Any cyclic group is a subgroup of the injective Z-module Q = Q ⊕ Q/Z, so that for any
non zero m ∈ M we have a group morphism ω : M → Q not vanishing on m.
If we put M ∗ = HomZ (M, Q), the natural morphism of A-modules M → M ∗∗ is injective,
and it is enough to embed M ∗∗ into an injective A-module.
Now, F (M ) = M ∗ is a (linearly) representable functor, M ∗ = HomA (M, D), and the repre-
sentant D is injective because F preserves exact sequences (Q is an injective Z-module).
When M = A, we see that A∗ is an injective A-module,Q ∗ and considering an epimorphism
∗ ∗∗
⊕I A → M , we obtain an injective morphism M → I A into an injective A-module.
Set Theoretic Questions: Rigorously, in the Representability theorem we must prove that
minimal pairs (up to isomorphisms of pairs) form a set. In the contravariant case, any minimal
pair Mξ is fully determined (up to isomorphisms) by the subset M • (A) ⊆ F (A), because a
submodule M ,→ M 0 is fully determined by the morphisms A → M 0 factoring through M , and
any two minimal pairs are submodules of a third minimal pair.
In the covariant case, if Q is an A-module such that for any A-module N and any 0 6= n ∈ N
there is a morphism f : N → Q such that f (n) 6= 0, then any minimal pair Mξ is fully determined
(up to isomorphisms) by the subset HomA (M, Q) ⊆ F (Q). Since An ' Q A/a, just consider for
any ideal a ⊂ A an injective A-module Qa containing A/a, and put Q = a Qa .
For a representability theorem in abstract categories see exercises 59-63 in pp. 484-485.
5.4 The Spectrum of a Ring

Definitions: The spectrum of a ring A is the set Spec A of all prime ideals, and any element
f ∈ A is said to be a function on Spec A, where the value f (x) of f at a point x ∈ Spec A,
defined by a prime ideal p, is just the class of f into the residue field κ(x)
f (x) = [f ] ∈ κ(x) := (A/p)p = Ap /pAp .
Although the residue field κ(x) changes with the point, the zero value is absolute.
5.4. THE SPECTRUM OF A RING 135
The ideal of all functions vanishing at x is just p, and zeros of ideals

\
Points of Spec A where Prime ideals of
(I)0 = = (f )0 = = Spec (A/I)
all functions in I vanish A containing I
f ∈I
are closed sets of a topology on Spec A, named Zariski topology,
(A)0 = ∅ , (0)0 = Spec A

P T
j Ij 0 = j (Ij )0
(IJ)0 = (I ∩ J)0 = (I)0 ∪ (J)0
and only the last equality requires a proof: if f1 ∈ I, f2 ∈ J do not vanish at a point x, then
neither does f1 f2 , and f1 f2 ∈ IJ ⊆ I ∩ J.
The basic open sets Uf = Spec A − (f )0 define a base of this topology since closed sets
are intersections of zeros of functions: Given a closed set Y and an exterior point x, there is a
function f ∈ A vanishing on Y such that f (x) 6= 0. Hence, if IY is the ideal of all the functions
f ∈ A vanishing on a given closed set Y , we have Y = (IY )0 .
Recall (p. 120) that Spec 6= ∅ when A 6= 0, so that I = A when (I)0 = ∅. Hence
Chinese Remainder Theorem: If (I)0 ∩(J)0 = ∅, then I∩J = IJ and A/IJ = (A/I)×/A/J).
Corollary: A/(m1 . . . mn ) = (A/m1 ) × . . . × (A/mn ) when m1 , . . . , mn are maximal ideals.
Proof: By induction on n, since (m1 )0 ∩ (m2 . . . mn )0 = (m1 )0 ∩ {(m2 )0 ∪ . . . ∪ (mn )0 } = ∅.
Proposition: x = (p)0 , and we say that x is the generic point of its closure x.
Hence Spec A is T0 and closed points correspond to maximal ideals of A.
Proof: A point x ∈ Spec A is in a closed set (I)0 when I ⊆ p, so that (p)0 ⊆ (I)0 .
Hence (p)0 is the minimal closed set containing x.
Theorem: Spec A is a compact topological space.

T P
Proof:
P Given a family of closed sets with empty intersection, ∅ = j (Ij )0 = j Ij 0 , then
I
j j = A, since any ideal I 6
= A is contained in some prime ideal.
Hence 1 = f1 + . . . + fn , where fi ∈ Iji , and a finite family has empty intersection,
(Ij1 )0 ∩ . . . ∩ (Ijn )0 = (Ij1 + . . . + Ijn )0 = (A)0 = ∅.
Definitions: Let X be a non-empty topological space. It is irreducible if it is not a union of

two smaller closed sets, and an irreducible component of X is a maximal irreducible subspace.
The closure of any irreducible subspace is irreducible; hence the irreducible components are
closed. By Zorn’s lemma, any irreducible subspace of X is in some irreducible component; hence
X is the union of its irreducible components.
When C is an irreducible closed set in Spec A, the ideal IC is prime (if C ⊆ (f h)0 = (f )0 ∪(h)0
then C ⊆ (f )0 or C ⊆ (h)0 ) and C = (IC )0 ; hence C is the closure of a (unique) point:
Proposition: The irreducible components of Spec A correspond to the minimal prime ideals of
A, and any prime ideal contains some minimal prime ideal. In fact, we have a lattice anti-
isomorphism:
Prime ideals ∼ Irreducible closed
−−→ , p 7→ (p)0
of A subspaces of Spec A
1. The spectrum Spec k of a field has a unique point.
2. Spec Z has a closed point for every prime number p, with residue field Fp , and a generic point
with residue field Q.
3. Spec k[x] has a closed point for any unitary irreducible polynomial P , with residue field
k[x]/(P ), and a generic point with residue field k(x).
4. The bijection (I)0 = Spec (A/I) is a homeomorphism since (f )0 ∩ (I)0 = (f¯ )0 . Hence,
Spec (A1 ⊕ A2 ) = (Spec A1 ) q (Spec A2 ).
In fact, A = A1 ⊕ A2 = I1 + I2 , and I1 ∩ I2 = 0, where I1 = 0 ⊕ A2 , and I2 = A1 ⊕ 0, so that
we have Spec A = (I1 )0 ∪ (I2 )0 , (I1 )0 ∩ (I2 )0 = ∅, and (Ii )0 = Spec A/Ii = Spec Ai .
Definition: A ring morphism j : A → B induces a map φ : Spec B → Spec A, where x = φ(y)

when px = py ∩ A := {a ∈ A : j(a) ∈ py }, so that κ(x) ,→ κ(y).
By definition (jf )(y) = f (x) = f (φ(y)); hence φ−1 (f )0 = (f B)0 , and φ is continuous:
φ−1 (I)0 = φ−1

T T −1 T
(f )0 = φ (f )0 = (f B)0 = (IB)0 .
f ∈I f ∈I f ∈I
Theorem: The canonical projection A → A/I induces a homeomorphism Spec A/I = (I)0 .
Proof: We know (p. 69) that it is a continuous bijection, and it is a homeomorphism because
we have (J)0 ∩ (Spec A/I) = π(J) 0 for any ideal J of A.
Theorem: Let S be a multiplicative system of A. The canonical morphism A → AS induces a

homeomorphism of Spec AS onto the subspace of Spec A defined by the prime ideals not inter-
secting S.
Proof: If q is a prime ideal of AS , then p = A ∩ q is disjoint from S and (p. 86) q = pAS , so
that Spec AS → Spec A is injective.
If a prime ideal p of A is disjoint from S, then A ∩ pAS = p:
a/1 = b/s, b ∈ p ⇒ au = bv ∈ p for some u, v ∈ S ⇒ a ∈ p.
Hence pAS 6= AS , and it is a prime ideal: as · bt ∈ pAS ⇒ ab ∈ A ∩ pAS = p ⇒ a or b ∈ p.

This continuous map defines a homeomorphism of Spec AS onto the image because we have
(I)0 ∩ (Spec AS ) = (IAS )0 for any ideal I of A.
Corollary: Prime ideals of Ap correspond to prime ideals of A contained in p; hence Ap has a

unique maximal ideal, pAp .
Corollary: Spec Af = Uf = Spec A − (f )0 .
Definition: The radical of a ring A is the set rad A = {a ∈ A : an = 0, for some n ≥ 1} of all
nilpotent elements, and a ring is reduced if rad A = 0.
Corollary: Nilpotent functions are just functions vanishing at any point of the spectrum. (The
radical of a ring is the intersection of all prime ideals).
Proof: If f n = 0, then (f )0 = (f n )0 = Spec A.

If (f )0 = Spec A, then Spec Af = ∅; hence Af = 0, and f is nilpotent.
5.4. THE SPECTRUM OF A RING 137
Proposition: If a ring morphism A → B is injective, Spec B → Spec A has dense image.
Proof: Let x ∈ Spec A be the point of a minimal prime. Now, Bx 6= 0 since Ax → Bx is injective;
hence x = Spec Ax is in the image, and it is dense. q.e.d.
In fact, the spectrum of a ring A must be considered to be the pair (A, Spec A), and
morphisms X = Spec B → Y = Spec A are defined to be ring morphisms A → B, i.e.
Hom(Y, X) = Homrings (A, B), so that any morphism induces a continuous map, which does
not determine it. Hence, if X = Spec A and Y = Spec B, then a point φ : Y → X of X
parameterized by Y is just a ring morphism A → B, and
X ×Spec k Y = Spec (A ⊗k B) , X q Y = Spec (A ⊕ B).
Now we shall calculate the fibre φ−1 (x) over the point x : Spec κ(x) → X = Spec A defined
by a prime ideal p of A:
Fibre Formula: φ−1 (x) = Spec (Bp /pBp ) = Spec B ⊗A κ(x) = Y ×X x.

Proof: If p = m is a maximal ideal, the fibre (mB)0 is the spectrum of B/mB = B ⊗A κ(x).
In general, since φ−1 (x) ⊆ Spec Bp , it is the fibre of Spec Bp → Spec Ap over the unique
closed point of Spec Ap , defined by the maximal ideal pAp .
Corollary: If x ∈ Spec A is defined by a maximal ideal m, then φ−1 (x) = Spec B/mB.
If x ∈ Spec A is defined by a minimal prime ideal p, then φ−1 (x) = Spec Bp .
1. The fibre of the projection A2,C = Spec C[x, y] → Spec C[x] = A1,C over the point x = a is
Spec C[x, y]/(x − a), the points being defined by the ideal (x − a) and the maximal ideals
(x − a, y − b). The fibre over the generic point is Spec C(x)[y], the points being defined by
the ideal 0 and the ideals (P ), where P (x, y) is irreducible of non zero degree in y. The
complex plane Spec C[x, y] is formed by the closed points x = a, y = b, the generic points of
the irreducible curves P (x, y) = 0, and the generic point of the plane.
2. The fibre of the projection A1,Z = Spec Z[x] → Spec Z over a prime number p is Spec Fp [x],
the points being defined by the ideal (p) and the maximal ideals (p, Q), where the reduction
Q̄ modulo p is irreducible. The fibre over the generic point is Spec Q[x], the points being
defined by the ideal 0 and the ideals (P ), where P is irreducible in Q[x].
3. Let us consider Z[i] = Z[x]/(x2 + 1). The fibre of Spec Z[i] → Spec Z over a prime p is

one point x = 1, if p = 2

2
Spec Fp [x]/(x + 1) = one point, if −1 is not a quadratic residue mod. p

2 points x = ±a, if −1 is a quadratic residue mod. p

Hence (p. 71) the maximal ideals of Z[i] are (2, 1 + i) = (1 + i), the ideals (p) when p ≡ 3
(mod. 4), and the ideals (p, i ± a), with a2 ≡ −1 (mod. p), when p ≡ 1 (mod. 4).
Definition: The localization of an A-module M at the prime ideal of a point x ∈ Spec A is
denoted by Mx . The support of an element m ∈ M is
supp (m) = {x ∈ Spec A : mx 6= 0},

S
(where mx := m/1) and supp (M ) = {x ∈ Spec A : Mx 6= 0} = m∈M supp (m) is the support
of M .
Lemma: (Ann (m))0 = supp (m). Therefore, m = 0 if and only if mx = 0, ∀x ∈ Spec A.
Proof: Condition mx = 0 states that f m = 0 for some function f not vanishing at x; i.e., x is
not in the zeros of the ideal Ann(m) = {f ∈ A : f m = 0}.
Now, if mx = 0 at any point, then (Ann(m))0 = ∅; hence Ann(m) = A, and m = 0.
Corollary: M = 0 if and only if Mx = 0 at any point x ∈ Spec A.

f g
Theorem: A sequence M 0 −−→ M −−→ M 00 of morphisms of A-modules is exact if and only if
fx gx
so is the localization Mx0 −−−→ Mx −−−→ Mx00 at any point x ∈ Spec A.
Proof: If it is exact at any point, then (Im gf )x = Im (gf )x = Im (gx fx ) = 0.

Hence Im gf = 0, and Im f ⊆ Ker g.
Now, localizing Ker g/Im f we see that it is 0,
(Ker g/Im f )x = (Ker g)x /(Im f )x = (Ker gx )/(Im fx ) = 0.
Definition: A ring O is local if it has a unique maximal ideal m. For example, Ax .
Nakayama’s Lemma: Let O be a local ring of maximal ideal m. If M is a finite O-module

and mM = M , then M = 0.
Proof: If M 6= 0, we consider a minimal generating system m1 , . . . , mn .
M = mM = m(Om1 + . . . + Omn ) = mm1 + . . . + mmn ,
and m1 = f1 m1 + f2 m2 + . . . + fn mn for some functions f1 , . . . , fn ∈ m.

Hence 1 − f1 is invertible (it is not in m) and
(1 − f1 )m1 = f2 m2 + . . . + fn mn ;
m1 ∈ Om2 + . . . + Omn , and m2 , . . . , mn generate M . Absurd.
Corollary: M = Om1 + . . . + Omn if and only if M/mM = k m̄1 + . . . + k m̄n ; (k = O/m).
Proof: Put M 0 = M/(Om1 + . . . + Omn ).

The exact sequence On → M → M 0 → 0 induces an exact sequence
k n −→ M ⊗O k −→ M 0 ⊗O k −→ 0
and m̄1 , . . . , m̄n span M ⊗O k = M/m if and only if 0 = M 0 /mM 0 ; i.e., M 0 = 0.

Definition: In this section k will be a ring (not a field, as in the rest of these notes). A k-
derivation of a k-algebra A into an A-module M is a group morphism D : A → M vanishing
on constants (Dλ = 0 for all λ ∈ k) such that
D(ab) = (Da)b + a(Db); a, b ∈ A.
Derivations A → M are k-linear and they form an A-module Derk (A, M ) with the operations
(D + D0 )a = Da + D0 a , (bD)a = b(Da).
Any morphism of A-modules f : M → M 0 induces a morphism of A-modules
f∗ : Derk (A, M ) −→ Derk (A, M 0 ), f∗ D = f ◦ D,
and any morphism of k-algebras j : A → B induces B-linear morphisms (N a B-module)
j ∗ : Derk (B, N ) −→ Derk (A, N ), j ∗ D = D ◦ j.
Example: Any derivation D : k[x1 , . . . , xn ] → M is determined by the derivatives Dxi , so that

∂
, and Derk (k[x1 , . . . , xn ], M ) = M ∂x∂ 1 ⊕ . . . ⊕ M ∂x∂n .
P
D = i (Dxi ) ∂x i
Derivations First Exact Sequence: If A → B is a morphism of k-algebras, for any B-module

N we have an exact sequence of B-modules
0 −→ DerA (B, N ) −→ Derk (B, N ) −→ Derk (A, N )
Derivations Second Exact Sequence: When B = A/I, we have an exact sequence
0 −→ Derk (B, N ) −→ Derk (A, N ) −→ HomA (I, N ) = HomB (I/I 2 , N )
Proof: The restriction of a derivation D : A → N to I is A-linear since f ∈ I annihilates any

B-module: D(af ) = a(Df ) + f (Da) = a(Df ).
Finally, a derivation D : A → N factors through B if and only if it vanishes on I.
∼
Example: If k = A/m is a field, then Derk (A, k) − → Homk (m/m2 , k). In fact, this morphism
is injective by the second exact sequence, and to show that it is surjective, we introduce the
differential dp : A → m/m2 , dp f = [∆f ] = [f − f (p)], at the point p defined by m,
f g = (f (p) + ∆f )(g(p) + ∆g),

∆(f g) = f (p)(∆g) + g(p)(∆f ) + (∆f )(∆g),
dp (f g) = f (p)dp f + g(p)dp f.
Now, if ω : m/m2 → k is k-linear, the derivation Df = ω(dp f ) coincides with ω on m.
Definitions: The existence of a universal derivation follows from the representability theorem;
but we shall give a direct construction. The diagonal ideal is the kernel ∆ of the morphism
µ : A ⊗k A → A, µ(a ⊗ b) = ab, and the module of differentials is ΩA/k = ∆/∆2 (both sides
define the same A-module structure, since it is annihilated by a ⊗ 1 − 1 ⊗ a ∈ ∆).
The differential is the k-derivation d : A → ΩA/k , da = [a ⊗ 1 − 1 ⊗ a],
d(ab) = (b ⊗ 1)[a ⊗ 1 − 1 ⊗ a] + (1 ⊗ a)[b ⊗ 1 − 1 ⊗ b] = b(da) + a(db).
Lemma: ΩA/k is generated by the image of the differential d : A → ΩA/k .

P P P
Proof: If i ai ⊗ bi ∈ ∆, then i ai bi = 0; hence i1 ⊗ ai bi = 0, and
P P P P
i ai ⊗ bi = i ai ⊗ bi − i1 ⊗ ai bi = i bi (ai ⊗ 1 − 1 ⊗ ai ).
Universal Property: For any k-derivation D : A → M there exists a unique morphism of

A-modules f : ΩA/k → M such that f (da) = Da,
Derk (A, M ) = HomA (ΩA/k , M ).

Proof: The A-linear morphism φ : A ⊗k A → M , φ(a ⊗ b) = b(Da), vanishes on ∆2 ,
φ((a ⊗ 1 − 1 ⊗ a)(b ⊗ 1 − 1 ⊗ b)) = D(ab) − a(Db) − b(Da) + ab(D1) = 0,
and it induces an A-linear map f : ∆/∆2 → M , f (da) = φ(a ⊗ 1 − 1 ⊗ a) = Da.

The uniqueness follows from the above lemma.
Corollary: ΩA/k = Adx1 ⊕ . . . ⊕ Adxn , when A = k[x1 , . . . , xn ].
∂ ∂
Proof: Derk (k[x1 , . . . , xn ], M ) = M ∂x1 ⊕ ... ⊕ M ∂xn .

Theorem: ΩAS /k = ΩA/k S
.
Proof: If M is an AS -module, then Derk (AS , M ) = Derk (A, M ), since a derivation D : A → M

only may be defined by the derivation
a sDa − aDs
D̄ : AS −→ M, D̄ =
s s2
a
and it is well defined: if s = bt , then rat = rbs, r ∈ S. Deriving and dividing by rst
Da aDt Db bDs
+ = +
s st t st
sDa − aDs tDb − bDt
=
s2 t2
HomAS (ΩAS /k , M ) = Derk (AS , M ) = Derk (A, M ) = HomA (ΩA/k , M ) = HomAS ((ΩA/k )S , M ).
Corollary: Ω(A1 ⊕A2 )/k = ΩA1 /k ⊕ ΩA2 /k .
Proof: We have A = A1 ⊕ A2 = Af1 ⊕ Af2 , where f1 = (1, 0), f2 = (0, 1).

Hence any A-module M is M = Mf1 ⊕ Mf2 .
Now, (ΩA/k )fi = ΩAf /k = ΩAi /k .
i
Theorem: The module of differentials is stable under base changes, ΩAK /K = ΩA/k ⊗k K.
Proof: If M is an AK -module, any k-derivation D : A → M uniquely factors through a K-

derivation D ⊗ 1 : AK → M , and
HomAK (ΩAK /K , M ) = DerK (AK , M ) = Derk (A, M ) = HomA (ΩA/k , M )

= HomAK (ΩA/k ⊗A AK , M ) = HomAK (ΩA/k ⊗k K, M ).
Differentials First Exact Sequence: If A → B is a morphism of k-algebras, we have an

exact sequence of B-modules
ΩA/k ⊗A B −→ ΩB/k −→ ΩB/A −→ 0
Proof: For any B-module N , we have an exact sequence
0 −→ HomB (ΩB/A , N ) −→ HomB (ΩB/k , N ) −→ HomB (ΩA/k ⊗A B, N )

|| || ||
DerA (B, N ) Derk (B, N ) Derk (A, N )
5.6. FINITE ALGEBRAS OVER A FIELD 141
Differentials Second Exact Sequence: When B = A/I, we have an exact sequence

d⊗1
I/I 2 −−−−→ ΩA/k ⊗A B −→ ΩB/k −→ 0
Proof: For any B-module N we have an exact sequence
0 −→ HomB (ΩB/k , N ) −→ HomB (ΩA/k ⊗A B, N ) −→ HomB (I/I 2 , N )

|| || ||
Derk (B, N ) Derk (A, N ) HomA (I, N )
Corollary: ΩA/k = (k[x]/(P, P 0 ))dx, where A = k[x]/(P ).

∼
Example: If k = A/m is a field, then d ⊗ 1 : m/m2 −− → ΩA/k ⊗A k is an isomorphism, since so is
(p. 139) the transpose Derk (A, k) = HomA (ΩA/k , k) = Homk (ΩA/k ⊗A k, k) → Homk (m/m2 , k).
5.6 Finite Algebras over a Field

Lemma: Any prime ideal of a finite k-algebra A is a maximal ideal.
Proof: If A is integral, the linear map ha : A → A, ha (x) = ax, is injective when a 6= 0.

Hence it is surjective, and 1 = ha (b) = ab for some b ∈ A, so that A is a field.
Lemma: The number of maximal ideals of any finite k-algebra A is ≤ [A : k].
Proof: We have n ≤ [(A/m1 ) × . . . × (A/mn ) : k] = [A/m1 . . . mn : k] ≤ [A : k] (p. 135).
Theorem: The spectrum of a finite k-algebra is a finite discrete space, Spec A = {x1 , . . . , xn },
and A decomposes as a direct sum of local algebras (finite extensions if A is reduced)
A = Ax1 ⊕ . . . ⊕ Axn .
Proof: Spec A is a finite space and any point is closed, hence it is discrete.
The natural morphism A → Ax1 ⊕ . . . ⊕ Axn is an isomorphism at any point y ∈ Spec A,
(Ax1 ⊕ . . . ⊕ Axn )y = (Ax1 )y ⊕ . . . ⊕ (Axn )y = Ay
because (Ax )y = 0 when x 6= y, since prime ideals of (Ax )y correspond to prime ideals of A
contained in mx and my , and there are no such ideals.
If A is reduced, Axi is a reduced local algebra; hence it is a field.
Theorem: If A, B are finite k-algebras, any injective morphism A → B induces a surjective

continuous map Spec B → Spec A.
Proof: If x ∈ Spec A, the morphism Ax → Bx is injective.

Hence Bx 6= 0, and any point of Spec Bx is in the fibre over x = Spec Ax .
Definitions: The points of a k-algebra A with values in an extension L of k, or L-points, are

the morphisms of k-algebras A → L.
If A is a k-algebra, a point x ∈ Spec A is rational when k = κ(x).
A finite k-algebra is rational when so is any point of Spec A.
1. L-valued points of A = k[x]/(P ) are just roots of P (x) in L.

2. When A = L is a finite extension, L-points of L are just automorphisms of L over k.
3. Any k-morphism φ : Spec B → Spec A preserves rational points since κ(y) is an extension of
κ(x) when x = φ(y). Hence subalgebras of a rational algebra are also rational (by the above
theorem) and it is clear that quotients, direct sums and tensor products of rational algebras
are also rational.
4. By the Chinese remainder theorem, k[x]/(P ) = i k[x]/(Pini ), where P = P1n1 . . . Psns is the
L
irreducible factor decomposition. The points of Spec k[x]/(P ) correspond to the irreducible
factors Pi , and the residue field is k[x]/(Pi ), rational points correspond to roots of P in k,
and k[x]/(P ) is rational if and only if all the roots of P are in k.

Points of Spec AL
Points Formula: Homk-alg (A, L) =
of residue field L
Proof: When L = k, any morphism A → k is surjective and it is fully determined by the kernel,
which defines a rational point of Spec A. In fact, two morphisms with equal kernel differ in an
automorphism of k, which is the identity.
The general case now follows from the graph formula,
Homk-alg (A, L) = HomL-alg (A ⊗k L, L).
Theorem: The concept of rational local algebra is geometric (stable under base changes k → K;
if A is a finite local rational k-algebra, then AK is a finite local rational K-algebra).
Proof: Let m be the unique maximal ideal of a finite rational local k-algebra O.
Any element of m is nilpotent, and k = O/m. Now the exact sequence
0 −→ m ⊗k K −→ O ⊗k K −→ k ⊗k K = K −→ 0
shows that mK is a maximal ideal of OK , and OK /mK = K.

Since m ⊗k K is generated by nilpotent elements, OK is a rational local K-algebra.
Kronecker’s Theorem: If A is a finite k-algebra, there is a finite extension k → L such that

AL is a rational L-algebra.
Proof: By induction on [A : k]. When [A : k] = 1, then A = k is rational.

When [A : k] > 1, if A has a rational point, A = A1 ⊕ B, and by induction, B is rational
over a finite extension L, and AL = (A1 )L ⊕ BL is rational.
If A has no rational point, we consider a residue field A → A/m = K. The points formula
shows that AK has some rational point; hence there is a finite extension K → L such that
(AK )L = AL is a rational L-algebra.
Definitions: Given a commutative square of extensions
k / K1

K2 /L
the field of fractions of the image of K1 ⊗k K2 → L is the composite K1 K2 , formed by sums,

products and quotients of elements of K1 and K2 .
The composites of K1 and K2 are just the residue fields of Spec (K1 ⊗k K2 ); hence they exist,
and when k → K1 is a finite extension, any composite is a quotient of K1 ⊗k K2 .
The proof of Kronecker’s theorem shows that any finite k-algebra A is rational over a quotient
L of an iterated tensor product A⊗n . With this additional condition, L is unique up to (non
canonical) isomorphisms and it is the decomposition field of A.
In fact, if A is rational over another quotient L0 of A⊗m , then so are A⊗n and its quotient L,
and there is a morphism L → L0 by the points formula. Hence [L : k] ≤ [L0 : k], and analogously
[L0 : k] ≤ [L : k], so that L → L0 is an isomorphism.
The condition of A = k[x]/(P ) being rational over L states that P has all its roots α1 , . . . , αn
in L, and that of being a quotient of A⊗n states that L is generated by some roots of P , so that
the decomposition field of P (i.e. of k[x]/(P )) over k is just L = k(α1 , . . . , αn ).
Definitions: A field k̄ is algebraically closed if any non constant polynomial with coefficients
in k̄ has a root in k̄ (any finite extension k̄ → K is trivial).
An algebraic extension k → k̄ is an algebraic closure of k if k̄ is algebraically closed.
For example, C is algebraically closed, so that it is an algebraic closure of R, and the field
Q̄ of all algebraic complex numbers is an algebraic closure of Q.
Theorem: Any field has an algebraic closure, unique up to non canonical isomorphisms.
Proof: Let LP be a decomposition field of P ∈ k[x] and let us consider the k-algebra
A = lim
−→
LP1 ⊗k . . . ⊗k LPn
and the residue field k̄ of a maximal ideal m of A, which is an algebraic extension of k since it
is generated by the images of the morphisms LP → A → k̄.
These morphisms also show that any polynomial P ∈ k[x] has all the roots in k̄, hence k̄ is
algebraically closed (p. 77).
If k → k 0 is another algebraic closure, we consider a composite k 0 k̄, and we have k̄ −
∼
→ k 0 k̄,
since k 0 is algebraic over k. Analogously k 0 −
→ k 0 k̄, and k̄ ' k 0 .
∼
5.6.1 Trivial and Separable Algebras

Definition: A finite k-algebra A is trivial when A = k ⊕ . . . ⊕ k = ⊕X k = Hom(X, k); i.e.,
when the number of points of Spec A coincides with the degree [A : k]. In such a case, A = ⊕k,
a 7→ (p1 (a), . . . , pn (a)), where Homk-alg (A, k) = {p1 , . . . , pn }.
If A, B are trivial k-algebras, clearly so are A ⊕ B, A ⊗k B and any quotient A/I; and also
any subalgebra C ,→ A, since trivial algebras are just reduced rational algebras.
Finally, given an extension k → L, by the points formula:
1. The number of L-points of a finite k-algebra A is bounded by [A : k], and the equality holds if
and only if AL is trivial, A⊗k L = ⊕L. In such a case AL = ⊕L, a⊗λ 7→ (p1 (a)λ, . . . , pn (a)λ),
where Homk-alg (A, L) = {p1 , . . . , pn }.
2. The number of automorphisms of a finite extension k → L is bounded by [L : k], and the equal-
ity holds if and only if L⊗k L = ⊕L. In such case L⊗k L = ⊕L, µ⊗λ 7→ (g1 (µ)λ, . . . , gn (µ)λ),
where Autk-alg (L) = {g1 , . . . , gn }.
Theorem: The functors F (A) = Spec A = Homk-alg (A, k) and R(X) = ⊕X k = Hom(X, k)
define an equivalence of categories
op
Trivial k-algebras ! Finite Sets .
Proof: The natural morphism A → Hom(Spec A, k) is injective because A = ⊕k is reduced;

hence it is an isomorphism, since both algebras
have equal degree (the cardinal of Spec A).
The natural map X → Spec Hom(X, k) is injective; hence it is bijective, since the number
of points of the spectrum is bounded by the degree.
Theorem: A finite k-algebra A is separable if the following equivalent conditions hold
1. A is locally5 trivial: AL = ⊕L for some finite extension k → L.
2. A is geometrically reduced: AK is reduced for any extension k → K.
3. A ⊗k A is reduced.
4. ΩA/k = 0.
Proof: (1 ⇒ 2) 0 → AK → AKL = (AL )KL = (⊕L)KL = ⊕KL; hence AK is reduced.

(2 ⇒ 3) A is reduced, A = ⊕i Ki ; hence A ⊗k A = ⊕i (A ⊗k Ki ) is reduced.
(3 ⇒ 4) If A ⊗k A = ⊕i Ki , any ideal I of A ⊗k A is a direct sum of components.
Hence I = I 2 , and ΩA/k = ∆/∆2 = 0.
(4 ⇒ 1) If AL = B ⊕ . . ., then 0 = (ΩA/k )L = ΩAL /L = ΩB/L ⊕ . . .; hence ΩB/L = 0.
By Kronecker’s theorem, there is a finite extension k → L such that AL is a rational L-
algebra.
Now (see p. 141) for any maximal ideal m of AL we have m/m2 = ΩB/L ⊗B L = 0.
Hence m = 0 by Nakayama’s lemma, so that B = L and AL is a trivial L-algebra.
Corollary: Separability is a local and geometric concept. Subalgebras, quotients, direct sums,
and tensor products of separable algebras are also separable. Moreover,

Components of
Homk-alg (A, K) = (A separable finite k-algebra)
A ⊗k K equal to K

Components of
Autk-alg (L) = (L separable finite extension)
L ⊗k L equal to L
Primitive Element Theorem: If the field k is infinite, any finite separable k-algebra is gen-
erated by some element, A = k[a].
Proof: A is locally trivial, 0 → A → AL = L⊕ . n. . ⊕L, where n = [A : k].

Elements (λ1 , . . . , λn ) ∈ A such that λi = λj form
S a proper vector subspace Vij ⊂ A, since
A contains a base of AL = Ln . When k is infinite6 , Vij 6= A and there is a ∈ A such that the
n morphisms k[a] → AL → L are different. The degree of k[a] is ≥ n, and A = k[a].
Definitions: A polynomial P ∈ k[x] is separable when so is the k-algebra k[x]/(P ); i.e., when
any root of P is simple, g.c.d.(P, P 0 ) = 1.
An element a of a finite k-algebra A is separable when so is the k-algebra k[a] ' k[x]/(Pa );
i.e., when so is the annihilator polynomial Pa (x).
Proposition: A = k[a1 , . . . , an ] is separable if and only if so are a1 , . . . , an .

5
If k → L is a finite extension, Spec L → Spec k should be viewed as an open cover of Spec k in a convenient
“Grothendieck topology”.
6
Including Vij into an hyperplane, in the dual projective space P(A∗ ) we have a finite number of points, and
there exists an exterior hyperplane; just project from a point and use induction on dimension.
Proof: If A is separable, so are the subalgebras k[ai ].

If k[a1 ], . . . , k[an ] are separable algebras, so is k[a1 ] ⊗k . . . ⊗k k[an ].
Hence A, since we have an epimorphism k[a1 ] ⊗k . . . ⊗k k[an ] → A.
Definition: A field k is perfect if any finite extension is separable; i.e., any irreducible poly-
nomial in k[x] has simple roots.
Fields of null characteristic and the finite fields Fp = Z/pZ are perfect (pp. 84, 85) and the
example of p. 85 shows that the field F2 (t) is not perfect.
Proposition: A field k of characteristic p > 0 is perfect if and only if the Frobënius morphism
F : k → k, F (a) = ap , is surjective. Therefore, finite fields are perfect.
√ √
Proof: If k is perfect, then k → k( p a ) is separable; hence p a ∈ k.
Conversely, if F is surjective, and a polynomial Q ∈ k[x] has a multiple root, then Q0 = 0
(p. 83) and Q is not irreducible,
Q = a0 + ap xp + a2p x2p + . . . = (bp0 + bpp x + bp2p x2 + . . .) = (b0 + bp x + b2p x2 + . . .)p .
Definition: The trace of a ∈ A is the trace of the endomorphism ha : A → A, ha (x) = ax.

So we get a linear form tr : A → k, tr(a) = tr ha , and a trace metric,
Tr(a, b) = tr(ab).
The trace of a nilpotent endomorphism T is null since (p. 175) the characteristic polynomial
is cT (x) = xn ; hence rad A ⊆ rad Tr.
Theorem: The trace metric is stable under base changes k → K.
Proof: The matrix of ha in a base e1 , . . . , en of A is just the matrix of ha⊗1 : AK → AK in the

base e1 ⊗ 1, . . . , en ⊗ 1 of AK , so that tr(a) = tr(a ⊗ 1).
Hence the trace’s metric in AK is just Tr ⊗ 1 : AK ⊗K AK = (A ⊗k A) ⊗k K → K.

Radical of the Radical of the
Corollary: ⊗k K =
trace metric in A trace metric in AK
Proof: Let us consider the polarity φ : A → A∗ , φ(a)(b) = Tr(a, b) = tr(ab).

φ
0 −→ rad Tr −→ A −−→ Homk (A, k)
φ⊗1
0 −→ (rad Tr)K −→ AK −−−−→ Homk (A, k) ⊗k K = HomAK (AK , K)
Since φ ⊗ 1 is the polarity of the trace metric in AK , we see that (rad Tr)K is just the radical
of the trace metric in AK .
Theorem: A finite k-algebra is separable if and only if the trace metric is non singular.
Proof: The trace metric of a trivial algebra is non singular (the matrix in the usual base is I).
Now, if A is locally trivial, then (rad Tr)L = 0, and rad Tr = 0.
If rad Tr = 0, then rad AK ⊆ (rad Tr)K = 0, and A is geometrically reduced.
Corollary: A finite extension k → K is separable if and only if the trace tr : K → k is non null.
Proof: The radical rad Tr is an ideal of the field K; hence it is null or K.

5.6.2 Galois Theory

Theorem: The sequence k → A ⇒ A ⊗k A is exact for any k-algebra A 6= 0.
Proof: Take a k-linear retract ω : A → k (p. 53). If a ⊗ 1 = 1 ⊗ a, then
a = (1 ⊗ ω)(a ⊗ 1) = (1 ⊗ ω)(1 ⊗ a) = ω(a) ∈ k.
Theorem: If a finite group G acts on a k-algebra A, then the algebra of invariants is stable
under base changes k → B,
(A ⊗k B)G = AG ⊗k B.
Proof: The functor (−) ⊗k B being exact, it preserves invariants under a finite group.
f,τ
That is to say, put G = {τ1 , . . . , τn }; then the exact sequence AG → A ⇒ A⊕ . n. . ⊕A of
k-linear maps, where f (a) = (a, . . . , a) and τ (a) = (τ1 (a), . . . , τn (a)), induces an exact sequence
f ⊗1,τ ⊗1
(AG )B −→ AB ⇒ AB ⊕ . n. . ⊕AB .
Definition: A finite extension k → L is a Galois extension when L ⊗k L is a trivial L-algebra
L ⊗k L = ⊕G L = Hom(G, L) , (a ⊗ b)(τ ) = τ (a)b,
where G = Homk-alg (L, L) = Autk-alg (L) =: Aut(L/k) is the Galois group, and the exact
sequence k → L ⇒ L ⊗k L = Hom(G, L) shows that k = LG .
1. The decomposition field L of a separable k-algebra A is a quotient of A⊗m . Since AL trivial,
so is L ⊗k L, and k → L is a Galois extension: the Galois envelope of A over k. Hence the
decomposition field k(α1 , . . . , αd ) of a separable polynomial P ∈ k[x] is a Galois extension of
k, and the Galois group G is a permutation group of the roots of P .
2. Any Galois extension k → L also is a Galois extension of any intermediate field K, since we
have an epimorphism ⊕L = L ⊗k L → L ⊗K L.
Moreover, any composite of L with L is ' L, so that any two morphisms f1 , f2 : L → E
into another extension always have equal image. Hence, if an intermediate field K is also a
Galois extension of k, then any automorphism τ ∈ G preserves it, τ (K) = K, and we have a
restriction morphism G = Aut(L/k) → Aut(K/k).
3. Since a Galois extension k → L is trivial over L, so is any intermediate field K; hence a

k-algebra is trivial over L if and only if it is (isomorphic to) a direct sum K1 ⊕ . . . ⊕ Kn of
intermediate fields (i.e. subrings of L containing k).
Let k → L be a Galois extension of group G, and let A be a finite k-algebra trivial over L,
A ⊗k L = ⊕F (A) L = Hom(F (A), L) , (a ⊗ λ)(p) = p(a)λ,
where F (A) := Spec AL = Homk-alg (A, L) is a finite G-set (τ p = τ ◦ p for any L-point p : A → L)
fully determining A:
A = A ⊗k LG = (A ⊗k L)G = Hom(F (A), L)G ,
where the action of G on Hom(F (A), L) is just τ · f = τ ◦ f ◦ τ −1 .

Hence we define the associated covering of a finite G-set ∆ to be
R(∆) = Hom(∆, L)G = HomG (∆, L),

so that the natural morphism A → RF (A) = HomG (Homk-alg (A, L), L) is an isomorphism
whenever A is trivial over L.
x
Moreover, any point x ∈ ∆ defines a morphism of k-algebras R(∆) ,→ Hom(∆, L) −
→ L, and
let us see that this natural morphism ∆ → F R(∆) also is an isomorphism:
Galois Theorem: If k → L is a Galois extension of group G, then the functors F and R define
an equivalence of categories
op
Finite k-algebras Finite RF (A) = A
! ,
trivial over L G-sets F R(∆) = ∆
Proof: We determine the k-morphisms R(∆) → L with the points formula:
R(∆) ⊗k L = Hom(∆, L)G ⊗k L = Hom(∆, L ⊗k L)G = Hom(∆, Hom(G, L))G ,
where the action of G on the left factor of L ⊗k L corresponds to the action (τ f )(g) = f (gτ ) on
Hom(G, L). Hence
Hom(∆, Hom(G, L))G = Hom(∆ × G, L)G = Hom((∆ × G)/G, L) = Hom(∆, L),
and we conclude that ∆ = Homk-alg (R(∆), L) = F R(∆).
Corollary: Homk-alg (A, B) = HomG (F (B), F (A)), HomG (∆, ∆0 ) = Homk-alg (R(∆0 ), R(∆)).
Proof: See page 130.
Corollary: If H ⊆ G is a subgroup, then R(G/H) = LH and F (LH ) = G/H.

Intermediate fields Subgroups K 7−→ Aut(L/K)
1. = , , is a lattice anti-isomorphism.
between k and L of G LH ←−p H
0
2. Homk-alg (LH , LH ) = {τ̄ ∈ G/H : H 0 ⊆ τ Hτ −1 }.
0
3. LH ' LH if and only if H 0 and H are conjugate subgroups.
4. Aut(LH /k) = N (H)/H.
5. LH is a Galois extension of k if and only if H is a normal subgroup of G, and in such a case

the Galois group is Aut(LH /k) = G/H.
Proof: We have R(G/H) = HomG (G/H, L) = LH , F (LH ) = F R(G/H) = G/H.

(1) Let i : LH → L be the inclusion. The fibre H of the restriction map G = F (L) → G/H =
F (LH ) over [Id] = i is just HomLH -alg (L, L) = Aut(L/LH ); hence H = Aut(L/LH ).
Moreover, K = LAut(L/K) , since L is a Galois extension of K.
The remaining statements readily follow from the equality
0 0
Homk-alg (LH , LH ) = HomG (G/H 0 , G/H) = (G/H)H = {ḡ ∈ G/H : H 0 ⊆ gHg −1 }.
Lagrange’s Theorem: Let k → L be a Galois extension of group G. Any composite LE is a

Galois extension of E, and the Galois group is Aut(L/L ∩ E).
Proof: LE is a quotient of L ⊗k E; hence LE ⊗E LE = ⊕LE, being a quotient of
L ⊗k E ⊗E LE = L ⊗K LE = L ⊗K L ⊗L LE = (⊕L) ⊗L LE = ⊕LE.
Moreover Aut(LE/E) ,→ Aut(L/k), and LAut(LE/E) = L ∩ (LE)Aut(LE/E) = L ∩ E.
Corollary: If G is the Galois group of a separable polynomial P (x), then the roots of any
irreducible factor of P form an orbit under the action of G.
Proof: Let L be a decomposition field of P (x). If P = P1 . . . Pr , where Pi is irreducible, then

A = k[x]/(P ) = K1 ⊕ . . . ⊕ Kr , where Ki = k[x]/(Pi ).
Hence F (A) = F (K1 ) ⊕ . . . ⊕ F (Kr ), where F (Ki ) = Homk-alg (k[x]/(Pi ), L) is an orbit,
formed by the roots of Pi in L.
Artin’s Theorem: Let H be a group of automorphisms of a finite extension k → L. If k = LH ,

then L is a Galois extension of k and H = Aut(L/k).
Proof: By the points formula, each element of H = {τ1 , . . . , τn } defines a rational component Ai
of L ⊗k L; hence L ⊗k L = (A1 ⊕ . . . ⊕ An ) ⊕ (B1 ⊕ . . .) and
L = LH ⊗k L = (L ⊗k L)H = (A1 ⊕ . . . ⊕ An )H ⊕ (B1 ⊕ . . .)H .
Since L is a domain, B1 ⊕ . . . = 0.
Moreover Ai = L, since any nilpotent a ∈ A1 would define a nilpotent in L,
(τ1 (a), . . . , τn (a)) ∈ (A1 ⊕ . . . ⊕ An )H = L.
5.6.3 The Frobënius Automorphism

Let L be a finite field of characteristic p, so that L is an extension of Fp .
If n = [L : Fp ], we have a Fp -linear isomorphism L ' Fnp ; hence L has q = pn elements.
Theorem: Let q = pn be a power of a prime number. There exists a unique field with q elements,
namely the decomposition field Fq of xq − x over Fp .
Proof: The automorphism F : Fq → Fq , F (a) = aq , fixes any root of xq − x; hence it is the

identity, and any element of Fq is a root of Q = xq − x.
Since Q is separable, because Q0 = −1, it has q different roots, and Fq has q elements.
If L is another field with q elements, then the non null elements of L form a multiplicative
group of order q − 1, so that they are roots of xq−1 − 1.
Hence all the roots of xq − x are in L, and L is the decomposition field of xq − x over Fp .
Theorem: The extension Fp → Fq is a Galois extension of cyclic Galois group, generated by

the Frobënius automorphism F (α) = αp .
Proof: Let H = (F ). Since FH

q = Fp , Artin’s theorem let us conclude.
Corollary: Let Q ∈ Fp [x] be separable, product of irreducible polynomials of degrees n1 , . . . , nr .

The Galois group of Q over Fp is generated by a permutation of form n1 , . . . , nr .
Lemma: Any finite subgroup H of the multiplicative group k ∗ of a field k is cyclic.
Proof: If d is the annihilator of H, then all the elements of H are roots of xd − 1.

Hence d ≥ |H|, and the classification of abelian groups shows that H is cyclic.
Corollary: In Fp [x] there are irreducible polynomials of arbitrary degree.

Proof: Let q = pn . Since F∗q is a cyclic group, Fq = Fp (θ).

The irreducible polynomial of θ over Fp has degree n.
Reduction Theorem: Let G be the Galois group of Q = xn + c1 xn−1 + . . . + cn ∈ Z[x] over Q.

If Ḡ is the Galois group over Fp of the reduction Q̄ ∈ Fp [x], there is a subgroup H ⊆ G and an
epimorphism ϕ : H → Ḡ. If Q̄ is separable, then ϕ is an isomorphism, and τ ∈ H and ϕ(τ ) ∈ Ḡ
have equal form as permutations of the roots.
Proof : Let α1 , . . . , αn be the complex roots of Q(x).

1. A = Z[α1 , . . . , αn ] is a finite Z-module, because αin = −c1 αin−1 − . . . − cn .
2. AG = Z. If ab ∈ AG = A ∩ Q, then Z[ ab ] ⊆ A is a finite Z-module (p. 171); hence its elements
have bounded denominator, so that ab ∈ Z.
3. A/pA is a finite Fp -algebra, hence Spec (A/pA) = {x1 , . . . , xd } is finite. Let mi be the maximal
ideal of A defined by xi , and put Ki = A/mi . Since
(m1 + (m2 ∩ . . . ∩ md ))0 = {x1 } ∩ {x2 , . . . , xd } = ∅,
by the chinese remainder theorem we have epimorphisms
A −→ A/(m1 ∩ . . . ∩ md ) = K1 ⊕ . . . ⊕ Kd −→ K1 = Fp [ᾱ1 , . . . , ᾱn ].
Q Q
Since Q(x) = i (x − αi ), then Q̄(x) = i (x − ᾱi ); hence K1 is the decomposition field of Q̄
and Ḡ = Aut(K1 /Fp ). Now the subgroup H = {τ ∈ G : τ (m1 ) = m1 } acts on K1 = A/m1
and we have a natural morphism ϕ : H −→ Ḡ, ϕ(τ )(ā) = [τ (a)].
4. ϕ is surjective. We put K1 = Fp (θ̄), and we fixQθ ∈ A with value θ̄ at x1 and vanishing at
x2 , . . . , xd . Now all the coefficients of R(x) = τ ∈G (x − τ θ) are in AG = Z, and τ̄ (θ̄) is a
root of R̄(x), ∀τ̄ ∈ Ḡ. Hence τ̄ (θ̄) = [τ θ] for some τ ∈ G.
Since τ (θ) does not vanish at x1 , we conclude that τ ∈ H, and ϕ(τ ) = τ̄ .
5. If Q̄ is separable, then τ ∈ H defines a permutation of the roots αi , and ϕ(τ ) defines the
same permutation of the roots ᾱi . Hence ϕ is injective.
Definition: If Q̄ is separable, the Frobënius automorphism of Q(x) at a prime number p is
the unique element Fp ∈ H such that ϕ(Fp ) is the Frobënius automorphism of K1 ,
Fp (a) ≡ ap (mod. m1 ), a ∈ A.
Fp depends on the maximal ideal m1 ; but it is well defined up to conjugation since it is
σFp σ −1 when we fix the maximal ideal mi = σ(m1 ), and G acts transitively on Spec(A/pA):
If Spec(A/pA) has two orbits, we fix a function f ∈ A only
Q vanishing at the orbit of x1 (it
G
exists since A → K1 ⊕ . . . ⊕ Kd is surjective). Now N (f ) = τ ∈G τ (f ) ∈ A = Z only vanishes
at the orbit of x1 . Absurd, N (f ) ∈ m1 ∩ Z = pZ ⊆ mi . q.e.d.
1. There are polynomials of arbitrary degree n with Galois group Sn over Q.
Let Q2 ∈ F2 [x] irreducible of degree n; Q3 ∈ F3 [x] with an irreducible factor of degree n − 1
and a root in F3 ; and Qp ∈ Fp [x] with an irreducible factor of degree 2 and n−2 different roots
in Fp , p 6= 2, 3. Since Z/6pZ = F2 ⊕ F3 ⊕ Fp , there is Q ∈ Z[x] with reductions Q2 , Q3 , Qp
modulo 2, 3, p. The Galois group of Q is a transitive subgroup G ⊆ Sn with a (n − 1)-cycle
and a transposition: (2, . . . , n), (ij) ∈ G.
Since G is transitive, it contains a transposition (1k). Conjugating (1k) with (2, . . . , n) we
see that (12), (13), . . . , (1n) ∈ G and these transpositions generate Sn = G.
2. If G does not contain n-cycles, then the reduction Q̄ never is irreducible. Any quartic of
group {Id, (12)(34), (13)(24), (14)(23)}, for example x4 + 1, is irreducible; but it has reducible
reduction at any prime p, since it is inseparable, or it has 4 roots in Fp (when Fp = Id) or it
is a product of two quadratic factors (when Fp has form 2,2).
3. If any automorphism τ ∈ G fixes

√ some root√of Q,√then Q̄ has a root in Fp (if Q̄ is separable).
Since any automorphism of Q( 2, i) fixes 2, i 2 or 1 + i, it fixes a root of x − 24 ; hence
8
x8 − 24 has a root in Fp when p 6= 2, and we see that 24 is a 8-th power at any prime.
The polynomial xn − 1 is separable when p = car k does not divide n, and its roots in a
decomposition field L form a subgroup µn of L∗ ; hence a cyclic group
µn = {εn , ε2n , . . . , εnn = 1}.
Theorem: k → k(εn ) is an abelian extension of group G ⊆ (Z/nZ)∗ .
Proof: k(εn ) is a Galois extension because it is the decomposition field of a separable polynomial,
and τ ∈ G induces an automorphism of µn ; hence τ (εn ) = εin , i ∈ (Z/nZ)∗ .
Now, the morphism G → (Z/nZ)∗ , τ 7→ i, is injective since τ is determined by τ (εn ).
Theorem: The Frobënius automorphism of xn − 1 at any prime p not dividing n is
Fp = [p ] ∈ (Z/nZ)∗ .
Proof: If the reduction of xn − 1 is separable, εin 6= εjn , i 6= j; so that εin ≡

/ εjn (mod. m1 ).
Since Fp (εn ) ≡ εpn (mod. m1 ), we see that Fp (εn ) = εpn .
2πi
Corollary: Q → Q e n is an abelian extension of group (Z/nZ)∗ , hence of degree φ(n), and
the cyclotomic polynomial Φn (x) is irreducible.
Proof: G contains any prime not dividing a n, and these primes generate (Z/nZ)∗ .
q−1
Lemma: If q is odd, the discriminant of Q(x) = xq − 1 is ∆ = (−1) 2 qq .
q−1 q−1 q−1 q−1

0 qq qq .
Q Q
Proof: ∆ = (−1) 2
i Q (αi ) = (−1) 2
i αi = (−1) 2

p p−1 q−1 q
Quadratic Reciprocity Law: = (−1) 2 2 , where p, q are odd primes.
q p
√ 2πi
Proof: K = Q( ∆ ) is the unique extension of degree 2 contained in Q e q , and the Frobënius
automorphism Fp of xq − 1 is the identity on K whenever [p] is in the unique subgroup of index
2 of (Z/qZ)∗ , formed by all the quadratic residues.
√ 2πi
Since Z[ ∆] ⊂ Z[e q ], the restriction of Fp to K is the Frobënius automorphism of x2 − ∆
at p, which is the identity when ∆ is a quadratic residue modulo p. Hence (p. 71)
q−1
p ∆ −1 2 q p−1 q−1 q
= = = (−1) 2 2 .
q p p p p
5.6.4 Radical Extensions

√
Lemma: If car k 6= 2, any extension k → L of degree 2 is L = k( a ) for some a ∈ k.
Proof: If β ∈ L is not√in k, it is a root of some polynomial

√ x2 + bx + c ∈ k[x].
2
Hence 2β = −b ± b − 4c, and L = k(β) = k( b − 4c ). 2
Proposition: All the complex roots of a polynomial P ∈ Q[x] are quadratic irrationals if and
only if the Galois group G of P (x) over Q is a 2-group.
Proof: If all the roots of P are quadratic irrationals, then the decomposition field L is contained
in an extension by quadratic radicals, and |G| = [L : Q] = 2d (p. 77).
If G is a 2-group, there exist subgroups 1 ⊂ H1 ⊂ . . . ⊂ Hd = G, |Hi | = 2i , (p. 118).
Now we see that L is an extension by quadratic radicals,
2 2 2 2
Q = LHd −−−→ LHd−1 −−−→ . . . −−−→ LH1 −−−→ L
Corollary: An algebraic number α ∈ C is a quadratic irrational if and only if the Galois group
G over Q of the irreducible polynomial Pα (x) is a 2-group.
Proof: If G is a 2-group, any root of Pα (hence α) is a quadratic irrational.

Conversely, if α is a quadratic irrational, we have to show that so is any complex root β of
Pα , and by definition α ∈ Q(α1 , . . . , αr ), where αi2 ∈ Q(α1 , . . . , αi−1 ).
Let L ⊂ C be the Galois envelope of Q(α1 , . . . , αr ).
Since Pα is irreducible, Q(α) ' Q(β), and τ (α) = β for some automorphism τ : L → L;
hence β ∈ Q(τ α1 , . . . , τ αr ) and (τ αi )2 ∈ Q(τ α1 , . . . , τ αi−1 ). q.e.d.
2πi 2πi
1. Q → Q e n is a Galois extension of degree φ(n); hence e n is a quadratic irrational if and
only if φ(n) is a power of 2.
2πi
2. If p = 2k + 1 is prime, then e p is a quadratic irrational. The regular polygons of 17, 257,
and 65537 sides are ruler-and-compass constructible.
3. Other proof of D’Alembert’s theorem: We have to show that any finite extension C → L
is trivial, and we may assume that L is a Galois extension of R. Let G be the Galois group,
and let H be a Sylow 2-subgroup, so that [LH : R] is odd. Then gr Pα = [R(α) : R] is odd for
all α ∈ LH , and Pα has a real root by Bolzano’s theorem. Hence α ∈ R, LH = R, H = G,
and L is an extension of C by quadratic radicals; L = C.
Independence Theorem: Let A be a k-algebra and L an extension of k. Any family of
morphisms of k-algebras pi : A → L is L-linearly independent.
Proof: When L = k, the morphisms pi : A → k are surjective, so that mi = Ker pi is a maximal

ideal, and we conclude since A/(m1 . . . mr ) = (A/m1 ) ⊕ . . . ⊕ (A/mr ) = k r (p. 135).
In the general case, just consider the rational points pi ⊗ 1 : AL → L.
Definitions: A finite extension k → K is a radical extension if
K = k(α1 , . . . , αr ) , αini ∈ k(α1 , . . . , αi−1 ).
A polynomial P ∈ k[x] is solvable in radicals if the decomposition field L over k admits a

finite extension L → K such that K is a radical extension of k.
Theorem: If k contains all the n-th roots of unity, and char k does not divide n, then
1. Any extension k(α), αn = a ∈ k, is cyclic, of degree a divisor d of n, and αd ∈ k.

√
2. If k → L is a cyclic extension of degree n, then L = k( n a ) for some a ∈ k.
Proof: (1) The decomposition field of the separable polynomial xn − a is k(α), hence it is a
Galois extension. If τ ∈ G, then (τ α)n = τ (αn ) = a, and τ (α) = uα, u ∈ µn .
So we get an injective morphism G ,→ µn .
Since µn is a cyclic group of order n, then G is a cyclic group of order a divisor d of n.
If G = (σ), then σ(α) = vα, where v d = 1. Hence αd ∈ k, since σ(αd ) = (σα)d = (vα)d = αd .
(2) Let G = (σ) be the Galois group.
Since σ n = Id, the annihilator polynomial of σ is xn − 1 by the independence theorem.
Hence εn is an eigenvalue of σ and there exists 0 6= α ∈ L such that σ(α) = εn α.
Now σ(αn ) = (εn α)n = αn , and αn ∈ k. Moreover, L = k(α) since σ i (α) = εin α 6= α, i < n.
Lemma: Let k → L be a Galois extension of group G. If k → E is an abelian extension, then

Aut(EL/E) = H G, and G/H is abelian.
Proof: E ∩ L is an abelian extension of k, hence it defines a subgroup H G such that G/H is

abelian, and Lagrange’s theorem states that H = Aut(EL/E).
Theorem: Let k be a field of characteristic 0. A polynomial P ∈ k[x] is solvable in radicals if

and only if the Galois group G is solvable.
Proof: Let L be the decomposition field of P , and G = Aut(L/k).

If L → k(α1 , . . . , αr ), where αini ∈ k(α1 , . . . , αi−1 ), we put Ei = k(εn , α1 , . . . , αi ), where
n = n1 . . . nr . Now E0 = k(εn ) is an abelian extension of k (p. 150), and Ei is a cyclic extension
of Ei−1 . If Hi is the Galois group of Ei L over Ei ,
LO / E0 L / E1 L / ... / Er L
O O O
G H0 H1 Hr
k / E0 / E1 / ... / Er
the above lemma states that Hr . . . H0 G and the successive quotients are abelian.
Since Hr = 1, because L ⊂ Er , we see that G is solvable.
Conversely, if G is solvable of order n, by Lagrange’s theorem the Galois group of L(εn ) over
k(εn ) is a subgroup H of G; hence it is solvable: we have subgroups 1 = H0 H1 . . . Hr = H,
where Hi /Hi−1 is cyclic, of order a divisor of n.
Hence L(εn ) is a radical extension of k(εn ), and P is solvable in radicals. q.e.d.
1. 1 A3 S3 is a resolution of S3 , and a resolution of S4 is 1 V A4 S4 , where V is the
Klein group {Id, (12)(34), (13)(24), (14)(23)}. But Sn is not solvable when n ≥ 5 because any
3-cycle is the commutator of two 3-cycles,
(ijk) = στ σ −1 τ −1 ; σ = (ijl), τ = (ikm),
so that a resolution of Sn , since Hi−1 contains any commutator of elements of Hi , would lead
to an absurd: H0 = 1 contains any 3-cycle. Equations of degree 2, 3 and 4 are solvable in
radicals; but equations of group S5 (they exist, p. 149) are not solvable in radicals.
2. If the Galois group is solvable of order p1 p2 . . . pr , where pi is prime, the above proof shows
√ √
that the equation is solved with r radicals p1 , . . . , pr and pi -th roots of unity. Since
|S3 | = 2 · 3, cubics are solved with a square root and a cubic root, and since |S4 | = 23 3,
quartics are solved with three square roots and a cubic root.
5.6.5 Inseparable Algebras

Definition: The separable elements of a finite k-algebra A form a subalgebra π0k (A), and the
separability degree of A over k is [A : k]s = [π0k (A) : k].
Theorem: The maximal separable subalgebra is stable under base changes k → K,
π0k (A) ⊗k K = π0K (A ⊗k K).
Proof: The kernel A1 of the differential A → ΩA/k contains π0k (A), hence π0k (A) ⊆ n An , where
T
k
An+1 = (An )1 . When
k
T An+1 = An , the algebra An is separable (p. 144)T and An ⊆ π0 (A).
Hence π0 (A) = n An , and we conclude because the subalgebra n An is stable under base
changes, since so is the module of differentials (p. 140).
Corollary: |Homk-alg (A, K)| ≤ [A : k]s , and the equality holds if and only if AK is rational.
|Aut(L/k)| ≤ [L : k]s , and the equality holds if and only if L ⊗k L is a rational L-algebra (L
is a normal extension of k).
Proof: By the above theorem we may assume that k = K.

Let B = A1 ⊕. . .⊕An be the rational components of A. Since separable algebras are reduced,
π0 (B) ,→ B/rad B = k n , and since k n ⊆ π0k (B), we see that π0k (B) = k n .
k
Hence |Homk-alg (A, k)| ≤ [A : k]s , and the equality holds if and only if A is rational.
Corollary: [L : k]s = [L : K]s [K : k]s .
Proof: The extensions L and K are rational over some extension E, so that
[L : k]s = |Homk-alg (L, E)| , [K : k]s = |Homk-alg (K, E)|.
The map Homk-alg (L, E) → Homk-alg (K, E) is surjective by the points formula (p. 141), and
the fibre over any point is HomK -alg (L, E), hence of cardinal [L : K]s since L ⊗K E is a rational
E-algebra (it is a quotient of L ⊗k E).
Corollary: The functor π0k preserves epimorphisms.
Proof: After a base change (p. 142) we may assume that A is a rational local k-algebra.
Corollary: A finite k-algebra A is said to be purely inseparable if the following equivalent

conditions hold
1. A is geometrically local (AK is local for any extension k → K).
2. AL is a rational local L-algebra for some extension k → L.
3. π0k (A) = k (that is to say, [A : k]s = 1).
Proof: (1 ⇒ 2) When AL is rational (p. 142) it is local and rational.

(2 ⇒ 3) If AL is local and rational, then π0k (A)L = π0L (AL ) = L; hence π0k (A) = k.
(3 ⇒ 1) π0K (AK ) = π0k (A)K = K, and we conclude by the following lemma.
Lemma: A finite k-algebra A is local if and only if π0k (A) is a field.
Proof: π0k (A) is local if so is A (p. 141) and, being reduced, it is a field.
If A = A1 ⊕ A2 ⊕ . . . is not local, then π0k (A) = π0k (A1 ) ⊕ π0k (A2 ) ⊕ . . . is not integral.
Corollary: If G is the group of automorphisms of a normal extension k → L, then k → LG is

a purely inseparable extension (and LG → L is a Galois extension by Artin’s theorem).
Proof: Since [L : k]s = |G| = [L : LG ], it follows that [LG : k]s = 1.
Examples: The decomposition field L of a finite k-algebra A is a quotient of A⊗m and, AL

being rational, so is L ⊗k L; hence L is a normal extension of k.
A polynomial Q is purely inseparable when so is the k-algebra k[x]/(Q); i.e., when all the
n
roots coincide. So, xp − a is purely inseparable when char k = p. In fact, if α is a root, then
n n n n n
αp = a, and xp − a = xp − αp = (x − α)p .
Corollary: Let p = car k. If k → L is a purely inseparable extension, then any element α ∈ L

n
is a root of a polynomial xp − a ∈ k[x].
Proof: Let Qα be the irreducible polynomial of α over k, and q = pn the greatest power such
that Qα (x) = Q(xq ) for some polynomial Q (separable, it is irreducible and Q0 6= 0).
Now a = αq ∈ π0k (L) = k, and α is a root of xq − a (in fact Qα (x) = xq − a, because
deg Qα (x) = deg Q(xq ) ≥ q).
Chapter 6
Projective Geometry
6.1 Projective Spaces

Definitions: The projective space P(E) of a vector space E is the set of all 1-dimensional
6 0)
vector subspaces. We have a projection (defined when e =
π : E −→ P(E), π(e) = hei.
The dimension of P(E) is n = dim E − 1, and we put Pn = P(E).

Linear subvarieties are subsets X = π(V ) = P(V ), where V is a vector subspace.
The dimension of X is dim V − 1, and the unique subvariety of dimension −1 is ∅ = π(0).
The codimension of X is codim X = dim P(E) − dim X = dim E − dim V.
Hyperplanes are linear subvarieties of codimension 1.
S
Since V = e∈V hei, we have a lattice isomorphism

Vector sub- π Linear sub-
−−→ , V 7→ π(V ).
spaces of E varieties of P(E)
so that the maximum of X = π(V ), Y = π(W ) is X + Y = π(V + W ); and (p. 50)
dim (X + Y ) = dim X + dim Y − dim (X ∩ Y ).
For example, two different lines in P2 cut one another in a unique point.
In general, since dim (X + Y ) ≤ dim P(E), we have
codim (X ∩ Y ) ≤ codim X + codim Y.
Theorem: Let X = π(V ). Linear subvarieties of dimension 1 + dim X containing X correspond

to points of P(E/V ) and we have a lattice isomorphism

Linear subvarieties ∼ Linear subvarieties
−−→ , π(W ) 7→ π(W/V ).
of P(E) containing X of P(E/V )
Proof: The isomorphism of p. 49, and the formula dim W/V = dim W − dim V .
Theorem: Hyperplanes of P(E) correspond to points of the dual space P(E ∗ ), and we have a
lattice anti-isomorphism (where linear subvarieties of codimension d correspond to linear subva-
rieties of dimension d − 1)

Linear sub- Linear sub-
∼
−−→ , π(V ) 7→ π(V o ).
varieties of P(E) varieties of P(E ∗ )
155
156 CHAPTER 6. PROJECTIVE GEOMETRY
Proof: The isomorphism of p. 53, and the formula dim V o = dim E − dim V .
Duality Principle: Any ordered set (X, ≤) defines a dual order X ∗ = (X, ≤∗ ), where we put
x ≤∗ y whenever y ≤ x. Since dim P(E ∗ ) = dim P(E), the dual of a n-dimensional projective
lattice (i.e. isomorphic to the lattice of linear subvarieties of a projective space of dimension
n) also is a n-dimensional projective lattice. Hence, any statement on projective lattices has a
dual equivalent statement. The statement “any two different points in a plane may be joined by
a unique line”, is dual to “any two different lines in a plane intersect at a unique point”, and
the dual figure of a triangle in a plane (three non collinear points) is again a triangle (three non
concurrent lines).
Desargues Theorem: If the three straight lines joining corresponding vertices of two triangles
ABC and A0 B 0 C 0 all meet in a point P , then the three intersection points L, M, N of pairs of
corresponding sides lie on a straight line.
Proof: If p represents P , there are representants a, b, c, a0 , b0 , c0 of A, B, C, A0 , B 0 , C 0 such that

p = a + a0 = b + b0 = c + c0 . Therefore
l = a − b = b0 − a0 represents L,
m = a − c = c0 − a0 represents M,
n = b − c = c0 − b0 represents N,
so that l − m + n = 0, and the points L, M, N are collinear.
Lemma: Given n + 2 points (P0 , . . . , Pn ; U ) in Pn , no hyperplane containing n + 1, there is a

base e0 , . . . , en of E, unique up to a common factor, such that
P0 = π(e0 ), . . . , Pn = π(en ), U = π(e0 + . . . + en ).
Proof: To prove the existence, we put Pi = π(vi ).

Now v0 , . . . , vn span E because P0 + . . . + Pn = Pn . If
U = π(λ0 v0 + . . . + λn vn ),
then λi 6= 0, because P0 + . . . + Pbi + . . . + U = Pn , and we put ei = λi vi .

If we have another base e00 , . . . , e0n , and
e00 = λ0 e0 , . . . , e0n = λn en , e00 + . . . + e0n = µ(e0 + . . . + en ),
then λ0 = µ, . . . , λn = µ, and the factors λi coincide.
Definitions: A projective reference system in Pn is a sequence (P0 , . . . , Pn ; U ) of n + 2

points, no hyperplane containing n + 1. If P = π(e), and e = x0 e0 + . . . + xn en in the normalized
base of the lemma, (x0 , . . . , xn ) are the homogeneous coordinates of P .
They are well defined up to a common factor, and may be not all zero.
The cross-ratio of four (different) collinear points is
x0
(P1 , P2 ; P3 , P4 ) = ∈ k,
x1
where (x0 , x1 ) are the coordinates of P4 in the reference system (P1 , P2 ; P3 ).
The cross-ratio defines a bijection P1 −{P1 , P2 , P3 } → k−{0, 1}. The projective parameter
of P = (x0 , x1 ) is θ = xx01 , so defining a bijection θ : P1 → k ∪ {∞}.
6.1. PROJECTIVE SPACES 157
Definition: A map τ : P(E) → P(E 0 ) is a projectivity if τ = π(T ) for some isomorphism

T : E → E 0 , in the sense that the following square commutes
E
T / E0
π π

P(E)
τ / P(E 0 )
so that τ induces a lattice isomorphism between linear subvarieties of P(E) and P(E 0 ), and any
projective statement is invariant under projectivities.
Projectivities P(E) → P(E) form a group P Gl(E), and homographies are projectivities
P1 → P1 . The equations of a projectivity Pn → Pn are X 0 = AX, where A is an invertible
matrix of n + 1 rows and columns. In particular, homographies are
(
x00 = ax0 + bx1

0 aθ + b a b
6= 0.
, θ = ;
x01 = cx0 + dx1 cθ + d c d
Lemma: The linear representant T : E → E 0 of a projectivity τ : P(E) → P(E 0 ) is well defined,

up to a non null factor.
Proof: If π(T ) = π(T̄ ), and we put S = T −1 T̄ , then π(S) : P(E) → P(E) is the identity.
The former lemma states that S = λId, so that T̄ = λT .
Theorem: If (P0 , . . . , Pn ; U ), (P00 , . . . , Pn0 ; U 0 ) are two projective reference systems in Pn , there
τ
is a unique projectivity Pn − → Pn such that Pi0 = τ Pi , U 0 = τ U .
Proof: Any two bases of E differ in a unique linear automorphism T : E → E.
Corollary: The cross-ratio classifies, up to projectivities, quadruples of collinear points.
Proof: If (P1 , P2 ; P3 , P4 ) = (P10 , P20 ; P30 , P40 ), and we take a homography τ transforming P1 , P2 , P3
into P10 , P20 , P30 , we have that P40 = τ P4 since
(P10 , P20 ; P30 , P40 ) = (P1 , P2 ; P3 , P4 ) = (τ P1 , τ P2 ; τ P3 , τ P4 ) = (P10 , P20 ; P30 , τ P4 ).
(θ1 − θ3 )(θ2 − θ4 )
Corollary: (P1 , P2 ; P3 , P4 ) = where θi is the projective parameter of Pi .
(θ1 − θ4 )(θ2 − θ3 )
(θ1 − θ3 )(θ2 − θ)
Proof: The homography transforming P1 , P2 , P3 into ∞, 0, 1 is τ (θ) = ·
(θ1 − θ)(θ2 − θ3 )
Corollary: (P1 , P2 ; P3 , P4 ) = (P2 , P1 ; P4 , P3 ) = (P3 , P4 ; P1 , P2 ) = (P4 , P3 ; P2 , P1 ).
Definition: Four collinear points have 24 orderings. Since the cross-ratio is invariant under the
Klein group V = {Id, (12)(34), (13)(24), (14)(23)}, there are at most 6 different values
1
(P1 , P2 ; P3 , P4 ) = λ (P1 , P3 ; P2 , P4 ) = 1 − λ (P2 , P1 ; P3 , P4 ) =
λ
1 1 λ
(P2 , P3 ; P1 , P4 ) = 1 − (P3 , P1 ; P2 , P4 ) = (P3 , P2 ; P1 , P4 ) =
λ 1−λ λ−1
and, when (P1 , P2 ; P3 , P4 ) = −1, we say that P4 is the harmonic conjugate of P3 with respect
to P1 , P2 . In such a case we also have (P2 , P1 ; P3 , P4 ) = (P1 , P2 ; P4 , P3 ) = −1.
6.1.1 Affine Spaces

Definition: An affine space of dimension n over a field k is a set An (the points) with an
action of a k-vector space V of dimension n (the free vectors) such that, if p, q ∈ An , then
q = p + e for a unique e ∈ V .
~
A map φ : An → A0m is affine when φ(p + e) = φ(p) + φ(e) ~ : V → V 0,
for some linear map φ
~ is an isomorphism).
obviously unique. Affinities are bijective affine maps (φ
Lemma: Given points p ∈ An , p0 ∈ A0m , and a linear map h : V → V 0 , there is a unique affine
~
map φ : An → A0m such that p0 = φ(p), h = φ.
Proof: The unique possible map, φ(p + e) = p0 + h(e), is affine:
φ(p + e + v) = p0 + h(e + v) = p0 + h(e) + h(v) = φ(p + e) + h(v).
Definition: The affine functions f : An → k form a vector space F of dimension n + 1.

The linear map V ,→ E = F ∗ , v(f ) = f~(v), identifies V with {ω ∈ E : ω(1) = 0}, and
the vector extension An ,→ E, x(f ) = f (x), is an affine map identifying An with the linear
subvariety {ω ∈ E : ω(1) = 1} of direction V .
Hence we have a map An ,→ Pn = P(E) identifying An with the complement in Pn of
H = π(V ), the hyperplane at infinity (the hyperplane of improper points).
Any affinity φ : An → An induces an isomorphism φ∗ : F → F , hence an isomorphism
φ : E → E, and φ(V ) = V ; therefore it induces a projectivity τ : Pn → Pn , leaving invariant the
hyperplane at infinity, and it coincides with φ on the affine part.
This normalized representant φ of the projectivity τ induces the identity on E/V .
Theorem1 : The group of affinities of An is canonically isomorphic to the group of projectivities

of Pn leaving invariant the hyperplane at infinity H.
Proof: If a projectivity τ : Pn → Pn leaves invariant H = π(V ), it admits a unique linear

representant T : E → E such that T (An ) = An , and the restriction φ = T |An : An → An is an
affinity, which induces the projectivity τ since both coincide out of H, where there are projective
reference systems.
Definitions: A projective reference system (P0 , . . . , Pn ; U ) is affine when P1 + . . . + Pn = H.

The origin is P0 , the axis are lines the P0 + Pi , and U is the unit point.
In homogeneous coordinates, the equation of the hyperplane at infinity is x0 = 0, and any
proper point has well defined affine coordinates y1 = xx01 , . . . , yn = xxn0 .
Affine subvarieties are affine parts of linear subvarieties and, except ∅, they correspond to
linear subvarieties of Pn not contained in H.
Two affine subvarieties are parallel if the infinity zones are incident.
The simple ratio (A, B, C) of three collinear points is the cross-ratio (A, B; C, P ) with the
improper point P of the line ABC, and C is the middle point of A and B when (A, B, C) = −1.
Affinities fixing any point at infinity are homotheties and translations, according to they
fix or not a proper point.
1
F. Klein, in the Erlangen Program, views Geometry as the action of a group G on a set: concepts are
invariants of the group action, statements are relations between invariants, and theorems are true relations.
Projective Geometry is defined by the action of the group of projectivities on Pn , and Affine Geometry by the
action of the group of affinities: Affine geometry is the projective geometry of a fixed hyperplane.
6.2. METRICS 159
Definitions: A semilinear transformation of a k-vector space E into a k 0 -vector space E 0

is a group isomorphism T : E → E 0 such that T (λe) = σ(λ)T (e) for some ring isomorphism2
σ : k → k 0 ; so that V ⊆ E is a vector subspace if and only if T (V ) is a vector subspace of E 0 .
A bijective map τ : P(E) → P(E 0 ) is a collineation if it induces a lattice isomorphism
between the lattices of linear subvarieties. Since X ⊆ P(E) is a linear subvariety if and only
if it contains the line passing through any two different points of X, a bijective map τ is a
collineation if and only if it preserves lines.
The projectivization π(T ) of a semilinear transformation T : E → E 0 is always a collineation.
Fundamental Theorem of Projective Geometry: Any collineation τ : P(E) → P(E 0 ) is

induced by a semilinear transformation (σ, T ) : (k, E) → (k 0 , E 0 ), when dim P(E) ≥ 2.
Proof: We put Pn = P(E), P0n = P(E 0 ) and we fix a reference (P0 , . . . , Pn ; U ) in Pn , the reference
(P00 , . . . , Pn0 ; U 0 ) in P0n obtained with τ , and the normalized bases e0 , . . . , en , and e00 , . . . , e0n .
We consider the affine structures defined by H = P1 + . . . + Pn and H 0 = P10 + . . . + Pn0 (and
the corresponding affine coordinates), so that τ preserves parallelism.
1. There is a bijection σ : k → k 0 such that τ (y1 , . . . , yn ) = (σ(y1 ), . . . , σ(yn )).

Since τ defines a bijection of the axis Li = P0 +Pi with the axis L0i = P00 +Pi0 , and it preserves
parallelism, we have τ (y1 , . . . , yn ) = (σ1 (y1 ), . . . , σn (yn )) for some bijections σi : k → k 0 .
Moreover, since τ transforms the diagonal of the plane Πij = Li + Lj into the diagonal of the
plane Π0ij = L0i + L0j , we see that σi = σj .
2. σ(0) = 0, since τ (P0 ) = P00 ; and σ(1) = 1, since τ (U ) = U 0 .
3. σ preserves the product: σ(ab) = σ(a)σ(b). In fact, since the line (x, ax) of the plane Π12
passes through the origin, so does (σ(x), σ(ax)); hence σ(ax) = cσ(x) for some constant c.
When x = 1, we see that c = σ(a).
4. σ preserves the sum: σ(a + b) = σ(a) + σ(b). In fact, since the line (x, x + a) is parallel to
the diagonal of Π12 , we have that (σ(x), σ(x + a)) is parallel to the diagonal of Π012 ; hence
σ(x + a) = σ(x) + c for some constant c. When x = 0, we see that c = σ(a).
σ(xi )e0i , is semilinear and it defines a collineation τ1 : Pn → P0n coinciding

P P
5. T ( xi ei ) =
with τ out of H; hence τ̄ = τ1−1 τ is the identity out of H. If P ∈ H, we consider two lines
R1 , R2 intersecting H at P . Since each line has two points out of H, they are invariant under
τ̄ , and we see that τ̄ (P ) = P . Hence, τ = τ1 .
6.2 Metrics
Let E be a vector space of finite dimension over a field k of characteristic 6= 2.
Definitions: A (symmetric) metric is a 2-covariant symmetric tensor S, and we put
e · v := S(e, v) = φ(e)(v),
where φ : E → E ∗ , φ(e) = ie S = S(e, −), is the polarity of S. A metric vector space is a

vector space endowed with a metric. A linear map f : (E, S) → (Ē, S̄) is a metric morphism
when e · v = f (e) · f (v), and an isometry is a metric isomorphism.
2
Obviously unique when E 6= 0. When k = k0 = R, semilinear transformations are linear because the unique
automorphism σ of the field R is the identity. In fact, σ(R+ ) = σ(R∗2 ) = R∗2 = R+ , hence σ preserves the order
and it is continuous. Since σ is the identity in Q, we have that σ = Id.
Two vectors are orthogonal if e · v = 0. Orthogonal vectors to a vector subspace V form a

vector subspace V ⊥ , the orthogonal of V , and it is clear that V ⊆ V ⊥⊥ .
The radical is the kernel of the polarity, rad E = E ⊥ (hence rad V = V ⊥ ∩ V ), and the
rank is dim (E/rad E). The space E is non singular when the polarity is an isomorphism,
rad E = 0, and totally isotropic when the polarity is null, rad E = E.
A non null vector e ∈ E is isotropic if e · e = 0.
A space is elliptic if there are no isotropic vectors (in particular it is non singular).
The orthogonal sum E ⊥ E 0 of two metric vector spaces is the direct sum E ⊕ E 0 , endowed
with the following natural metric, so that rad (E ⊥ E 0 ) = (rad E) ⊥ (rad E 0 ),
(e1 + e01 ) · (e2 + e02 ) = e1 · e2 + e01 · e02 .
A metric S on E is projectable by an epimorphism p : E → Ē if there exists a metric S̄ in

Ē, clearly unique, such that p is a metric morphism.
Theorem: A metric is projectable if and only if Ker p ⊆ rad E.
Proof: If it is projectable and p(e) = 0, then e · v = p(e) · p(v) = 0; hence e ∈ rad E.

Conversely, if Ker p ⊆ rad E, the metric p(e) · p(v) = e · v is well defined: if e0 = e + u,
u ∈ rad E, then e0 · v = e · v + u · v = e · v.
Corollary: Any metric projects onto E/rad E, and the projection is non singular.
Theorem: Any metric vector space E decomposes, uniquely up to isometries, as an orthogonal

sum of a totally isotropic space and a non singular space,
E = (rad E) ⊥ (E/rad E).
Proof: If E = (rad E) ⊕ V , then E = (rad E) ⊥ V , and V is non singular.

Moreover, if E = T ⊥ F , where T is totally isotropic and F is non singular, then
rad E = (rad T ) ⊥ (rad F ) = T
and the metric morphism F → E/rad E = E/T is an isometry.
Lemma: If E is non singular, then dim V ⊥ = dim E − dim V , and V = V ⊥⊥ .

If moreover V is non singular, E = V ⊥ V ⊥ .
∼
Proof: Composing the polarity E −
→ E ∗ with the epimorphism E ∗ → V ∗ we get an exact
⊥ ∗
sequence 0 −→ V −→ E −→ V −→ 0 and we conclude.
Corollary: Let E be non singular. If V is a non singular subspace, there is an isometry E → E,

the symmetry with respect to V , which is the identity on V and −Id on V ⊥ .
Definitions: A plane is hyperbolic if it is non singular and has some isotropic vector. A
hyperbolic space is an orthogonal sum of hyperbolic planes. Two isotropic vectors e, e0 with
e · e0 = 1 form a hyperbolic pair, so that they span a hyperbolic plane he, e0 i.
Lemma: Let E be non singular. Any base e1 , . . . , ei of a totally isotropic subspace T may be
completed so as to obtain i mutually orthogonal hyperbolic pairs; hence T is contained in a
hyperbolic space he1 , e01 i ⊥ . . . ⊥ hei , e0i i.
6.2. METRICS 161
Proof: By induction on i. By the above lemma, there is a vector v orthogonal to e2 , . . . , ei , and

such that e1 · v = 1. If we put e01 = v − v·v
2 e1 , we obtain a hyperbolic pair e1 , e1 .
0
By induction, in he1 , e1 i we have vectors e2 , . . . , ei such that the pairs ej , e0j are hyperbolic
0 ⊥ 0 0
and mutually orthogonal.
Corollary: Any hyperbolic plane has some hyperbolic pair, and all hyperbolic spaces of equal
dimension are isometric.
Witt’s Theorem: Let E be non singular. Any isometry σ : V → V 0 between vector subspaces
may be extended to an isometry of E.
Proof: We put V = rad V ⊥ F , and we complete a base of rad V in F ⊥ so as to obtain mutually

orthogonal hyperbolic pairs, which span a hyperbolic space H ⊆ F ⊥ .
Now V 0 = rad V 0 ⊥ F 0 , with F 0 = σF , and analogously completing a base of rad V 0 in F 0⊥ ,
so as to obtain a hyperbolic space H 0 ⊆ F 0⊥ , we see that we may extend σ to an isometry
H ⊥ F → H 0 ⊥ F 0 ; and we may assume that V is non singular.
When V = V 0 , we may extend σ by the identity on V ⊥ .
If V = hei, then V 0 = he0 i, where e0 = σ(e). Since V + V 0 is a plane not totally isotropic,
and the vectors e0 + e, e0 − e are non null and orthogonal, someone is not isotropic.
If so is e0 − e, the symmetry with respect to he0 − ei⊥ transforms e into e0 , and it extends σ.
If so is e0 + e, the symmetry with respect to he0 + ei extends σ.
When dim V > 1, we decomposes it as an orthogonal sum of two non singular subspaces,
V = F ⊥ G, so that V 0 = F 0 ⊥ G0 . By induction, we may extend σ : F → F 0 to an isometry σ1 of
E; we put G1 = σ1 (G). Since G1 and G0 are contained in F 0⊥ , the isometry σσ1−1 : G1 → G0 may
be extended to an isometry of F 0⊥ which, extended by the identity on F 0 , defines an isometry
σ2 of E such that the isometry σ2 σ1 : E → E extends σ.
Corollary: If E is non singular, all maximal totally isotropic subspaces have equal dimension,
named index of the metric.
Corollary: Let E be non singular. If E ⊥ F ' E ⊥ F 0 , then F ' F 0 .
Proof: Since rad F = rad (E ⊥ F ) ' rad (E ⊥ F 0 ) = rad F 0 , we have that
E ⊥ (F/rad F ) ' E ⊥ (F 0 /rad F 0 ),
and, by Witt’s theorem, F/rad F ' F 0 /rad F 0 , and F ' F 0 .
Theorem: Any metric vector space E decomposes, uniquely up to isometries, as an orthogonal

sum of a totally isotropic space, a hyperbolic space and an elliptic space.
Proof: If E is non singular, any maximal totally isotropic subspace T is contained in a hyperbolic
space H, and we have E = H ⊥ H ⊥ , where H ⊥ is elliptic, since T is maximal.
The uniqueness follows from the above corollary.
6.2.1 Classification
Theorem: The dimension and rank classify metrics when k is algebraically closed.
Proof: The elliptic part has dimension 0 or 1 (according to the rank being even or odd) because
if it would have two linearly independent vectors e, v, we may fix λ ∈ k so that
(λe + v)2 = (e2 )λ2 + 2(e · v)λ + v 2 = 0.
Moreover, all elliptic spaces of dimension 1 are isometric since they have some vector of
√
square e2 = 1: just divide any vector e 6= 0 by e · e.
Definition: The matrix of a metric S in a base e1 , . . . , en of E is A = (aij ), aij = ei · ej .

Remark that the rank of S is just the rank of the matrix A.
Lemma: When k = R, any elliptic space is positive definite or negative definite.
Proof: If e2 > 0 and v 2 < 0, then (λe + v)2 = (e2 )λ2 + 2(e · v)λ + v 2 = 0 for some λ ∈ R, since
the discriminant of the polynomial is 4(e · v)2 − 4(e2 )(v 2 ) > 0.
Theorem: If k = R, the dimension, rank, index and sign of the elliptic part classify metrics.
Proof: The existence of orthonormal bases shows that the dimension classifies elliptic spaces of
positive sign; hence also those of negative sign. q.e.d.
In the real case, to calculate the rank, index and sign of a metric S, we fix a scalar product
e · v in E, and we put e ∗ v = S(e, v). The scalar product defines an isomorphism E ' E ∗ , and
the polarity T : E → E ∗ = E of S is an endomorphism such that e ∗ v = (T e) · v, so that T is a
symmetric, (T e) · v = e · (T v), and, according to the Spectral theorem:
The endomorphism T diagonalizes in some orthonormal base of E. Hence all the roots of the
characteristic polynomial cT (x) are real. If n is the degree of cT (x), r+ and r− are the number
of positive and negative roots (counted with multiplicities), and r0 is the multiplicity of the null
root, then the rank r, index i, and sign s of S are
r = n − r0 , i = min(r+ , r− ), s = sgn (r+ − r− ).
In fact, let α1 , . . . , αn be the roots of cT and let e1 , . . . , en be an orthonormal

p base of eigenvectors,
T (ei ) = αi ei . Then ei ∗ ej = (T ei ) · ej = αi δij and, dividing ei by |αi |, when αi 6= 0, the matrix
of S is diag(0, .r.0., 0, 1, .r.+., 1, −1, .r.−., −1). q.e.d.
If A is the matrix of S in a base e1 , . . . , en , and we consider the scalar product ei · ej = δij ,
then the matrix of the endomorphism T is A, since ei · (T ej ) = ei ∗ ej = aij . The characteristic
polynomial cT (x) = |xI − A| is not an invariant of S; but so are the numbers r+ , r− , r0 .
Definition: Any non-singular metric on a 1-dimensional vector space is obviously classified by

the square e · e of any non null vector, well defined in the group k ∗ /k ∗2 . Now a non-singular
metric S on a vector space E of dimension n defines an isomorphism E → E ∗ ; hence an
isomorphism Λn E → Λn E ∗ = (Λn E)∗ , and a non singular metric on Λn E, classified by an
element disc S ∈ k ∗ /k ∗2 named discriminant of S.
If A is the matrix of S in a base of E, then disc S = |A| ∈ k ∗ /k ∗2 .
In general, the discriminant of a metric is defined to be that of the non singular part.
It is clear that disc (E ⊥ E 0 ) = (disc E)(disc E 0 ).
6.2. METRICS 163
Lemma: Let Fq be a finite field of characteristic 6= 2. If a, b ∈ Fq are non null, then the equation
ax2 + by 2 = 1 has some solution in Fq .
Proof: Since F2q has q+1

2 elements (p. 70), so does the image of the map f : Fq → Fq , f (x) =
−1 2
b (1 − ax ). Hence they intersect, and we conclude.
Theorem: The dimension, rank and discriminant classify metrics when k is a finite field.
Proof: By induction on the rank r.

When r = 1, the discriminant classifies non singular metrics of dimension 1.
If r ≥ 2, by the above lemma e · e = 1 for some e ∈ E.
Hence E = hei ⊥ hei⊥ , and we conclude by induction since disc E = disc hei⊥ .
Definition: An alternate metric is a 2-form Ω ∈ Λ2 E ∗ . The above proofs remain valid for
alternate metrics; but, since any vector is isotropic, non singular spaces are hyperbolic.
Theorem: The rank classifies alternate metrics, and it is always even. Any alternate metric of
rank 2r, in some base ω1 , . . . , ωn of E ∗ , is Ω = ω1 ∧ ω2 + . . . + ω2r−1 ∧ ω2r .
Definition: Let σ : K → K be an involution (σ 2 = Id, σ 6= Id) of a field K. Then the equality

α = α+σα2 + α−σα
2 shows that K is an extension of degree 2 of k := {α ∈ K : σα = α}, and that
K = k ⊕ kj, where σ(j) = −j. We put ᾱ = σα, and a = −j 2 ∈ k. An hermitian metric on a
finite-dimensional K-vector space E is a right-linear, left-semilinear (of automorphism σ) map
H : E × E → K, such that H(v, e) = H(e, v).
The orthogonal of V is V ⊥ = {e ∈ E : H(e, v) = 0, ∀v ∈ V }, and the radical is rad H = E ⊥ .
H defines a symmetric metric S and an alternate metric Ω on the k-vector space E,
H(e, v) = S(e, v) + Ω(e, v)j,
Moreover, we have S(e, v) = Ω(e, jv) and Ω(e, v) = a−1 S(je, v). Hence,
1. The orthogonal V ⊥ of a K-vector subspace V ⊆ E is the same for H and S. In particular,

rad H = rad S, and rk S = 2 rk H.
2. H(e, e) = S(e, e), and the isotropic vectors of H are those of S.
3. Any maximal totally isotropic subspace T of H so is for S.

If T ⊕ ke would be totally isotropic for S, then e ∈ T ⊥ , and H(e, e) = S(e, e) = 0; and T ⊕ Ke
would be totally isotropic for H.
Theorem: Two hermitian metrics H, H 0 are equivalent if and only if the associated symmetric
metrics S, S 0 are equivalent.
Proof: If τ : (E, H) → (E 0 , H 0 ) is a K-linear isometry, then τ : (E, S) → (E 0 , S 0 ) is a k-linear

isometry, S(e, e) = H(e, e) = H 0 (τ e, τ e) = S 0 (τ e, τ e).
Conversely, if τ : (E, S) → (E 0 , S 0 ) is a k-linear isometry, take e ∈ E such that S(e, e) 6= 0
and put e0 = τ (e), so that H 0 (e0 , e0 ) = S 0 (e0 , e0 ) = S(e, e) = H(e, e), and (Ke, H) ' (Ke0 , H 0 );
hence (Ke, S) ' (Ke0 , S 0 ).
(Ke, S) being non singular, we have ((Ke)⊥ , S) ' ((Ke0 )⊥ , S 0 ) by Witt’s theorem.
Now ((Ke)⊥ , H) ' ((Ke0 )⊥ , H 0 ) by induction on the dimension, and (E, H) ' (E 0 , H 0 ).
6.2.2 Quadrics
Definition: A map q : E → k is said to be a quadratic form when, in the coordinates
(x1 , . . . , xn ) of a base e0 , . . . , en of E (hence of any base) it is defined by some homogeneous
polynomial of degree 2:
P
q(x0 e0 + . . . + xn en ) = cij xi xj .
0≤i≤j≤n
The quadratic form of a metric S is the map q : E → k, q(e) = e2 = S(e, e).

It vanishes on every isotropic vector, and it fully determines the metric,
1

e·v = 2 q(e + v) − q(e) − q(v) .
If (aij ) = (ei · ej ) is the matrix of S in a base, the quadratic form is
aii x2i +
P P P
q(x0 , . . . , xn ) = aij xi xj = 2aij xi xj
i,j i i<j
so that any quadratic form is associated to a unique metric (recall that car k 6= 2).Definitions:
Definitions: A quadric in P(E) is a non null metric on E, up to a factor, Q = hSi. That is to
say, quadrics in P(E) are just points of the projective space P(S2 E).
Two quadrics hSi, hS 0 i in P(E) and P(E 0 ) are projectively equivalent if some projectivity
π(T ) : P(E) → P(E 0 ) transforms one into the other: S 0 (T e, T v) = λS(e, v).
Two points are conjugate with respect to a quadric if they are represented by orthogonal
vectors, and points of the quadric are represented by isotropic vectors.
The points in the vertex π(rad S) are the singular points of the quadric. If X = π(V ) is
not contained in the vertex, then Q ∩ X denotes the quadric defined by the restriction of S to V .
The directrix of a quadric Q is the non singular quadric Q ∩ X, where X is a linear subvariety
defined by some supplement of rad E in E.
The polar variety of X = π(V ) is X ⊥ = π(V ⊥ ), and if a point P is not in the vertex, then
⊥
P is the polar hyperplane of P .
The tangent hyperplane at a non singular point P of the quadric is just its polar hyperplane,
and a linear subvariety X is said to be tangent at P when P ∈ X ⊆ P ⊥ .
If Q is non singular, then φ : E → E ∗ is an isomorphism, and we have the dual quadric hS ∗ i
in the dual space P(E ∗ ), where S ∗ (φe, φv) = S(e, v), and points of the dual quadric are tangent
hyperplanes of Q.
When we multiply a metric by a non null factor λ, the rank and index do not change, but
in the real case the sign changes when λ is negative. Therefore,
Theorem: If the field k is algebraically closed, the dimension and rank (n, r) projectively classify
quadrics, and 1 ≤ r ≤ n + 1.
Conics in P2,C Locus Quadrics in P3,C Locus

x20 = 0 double line x20 = 0 double plane
x20 + x21 = 0 pair of lines x20 + x21 = 0 pair of planes
x20 + x21 + x22 = 0 non singular conic x20 + x21 + x22 = 0 cone
x20 + x21 + x22 + x23 = 0 non singular quadric
Theorem: If k = R, the dimension, rank and index (n, r, i) projectively classify quadrics, and
we have 0 ≤ 2i ≤ r ≤ n + 1.
6.2. METRICS 165
Example: The fundamental hypothesis of the Special Theory of Relativity is that the light cone
is a quadratic cone in a 4-dimensional affine space (A4 , V ); that it is defined by some quadratic
form q : V → R of index 1 (well-defined up to a factor3 ): A Minkowski spacetime is a real
affine space A4 whose space of free vectors V is endowed with a 1-dimensional vector space Q
of symmetric metrics of index 1. Let us fix a Lorentz metric g ∈ Q (i.e. a metric of type
(+, −,p−, −), named time metric, the proper time interval between two events p and q = p+v
being g(v, v), whenever g(v, v) > 0), and let us fix a metric h ∈ Q of type (−, +, +, +), named
space metric, so that we have h = −c2 g for some positive constant c, the speed of light.
Isotropic vectors are light vectors.
The other hypothesis p is that all inertial observers are equivalent, so that the equation of
the light cone must be x2 + y 2 + z 2 = ct in any inertial reference frame. Hence an inertial
reference system is defined to be an event p ∈ A4 (the origin) and a base e0 , e1 , e2 , e3 of V
such that the matrix of g is diag(1, −c−2 , −c−2 , −c−2 ); i.e. the matrix of h is diag(−c−2 , 1, 1, 1).
Now any event is q = p + te0 + xe1 + ye2 + ze3 , where t is the observed time and the spatial
vector ~r = xe1 + ye2 + ze3 is the position of q (both relative to the inertial observer), and we
say that (t, x, y, z) are the inertial coordinates of q in the considered inertial reference. The
straight line p+Re0 is the trajectory of the inertial observer (e0 being the velocity, the spacetime
displacement per time unit) and e1 , e2 , e3 is an orthonormal base of the Euclidean structure that
the space metric h = −c2 g defines on the subspace (Re0 )⊥ = he1 , e2 , e3 i.
Inertial trajectories are straight lines where g is positive-definite, velocities of inertial ob-
servers being vectors4 u such that g(u, u) = 1, so that h defines a Euclidean structure on (Ru)⊥ ,
and vectors in (Ru)⊥ are spatial (or simultaneity) vectors for inertial observers of velocity u.
The metrics g and h are well defined up to a positive factor. To replace g by λ2 g modifies
time intervals by a factor λ, so that to fix the time metric g ∈ Q is just to fix the time unit, and
to replace h by λ2 h modifies lengths by a factor λ, so that to fix the space metric h ∈ Q is just
to fix the length unit.
In presence of electromagnetic forces, the special theory of relativity assumes the existence of
an endomorphism F̃ : V → V such that F̃ (u) is the force acting on a unit charge at rest, measured
by an inertial observer of velocity u. Let us consider the 2-covariant tensor F (e, v) = h F̃ (e), v .
Since any force always is a spatial vector, we have F (u, u) = 0, so that F is an alternate tensor,
and an electromagnetic field is defined to be a 2-form F on a Minkowski spacetime. Once
we fix an inertial reference e0 , e1 , e2 , e3 , we have
F = E1 ω0 ∧ ω1 + E2 ω0 ∧ ω2 + E3 ω0 ∧ ω3 − 1c (B3 ω1 ∧ ω2 + B2 ω3 ∧ ω1 + B1 ω2 ∧ ω3 ),
where the spatial vectors E~ = E1 e1 + E2 e2 + E3 e3 and B~ = B1 e1 + B2 e2 + B3 e3 depend on the

inertial observer.
However, the characteristic polynomial of the endomorphism F̃ is x4 − c12 (kEk~ 2 − kBk
~ 2 )x2 −
1 ~ ~ 2 ~ 2 − kBk
~ 2 and hE|~ Bi
~ 2 do not depend on the inertial reference
c4
hE|Bi , so that the scalars kEk
frame.
Affine Elements of Quadrics: A quadric in an affine space (P(E), H) is parabolic when the
infinity H is a tangent hyperplane.
A center of a quadric is a point conjugate to any point at infinity.
A line passing through a center is a diameter, and it is an asymptote if it is tangent.
Two quadrics are affinely equivalent if some affinity transforms one into the other.
3
The special theory of relativity is just the geometry of a sphere in P3,R !
4
In fact, the Special Theory of Relativity always assume that time is oriented; that the instantaneous 4-velocity
U of any punctual body always is in a fixed connected component of the hyperboloid g(u, u) = 1.
Lemma: If (r, i) are the rank and index of a quadric Q, and (r0 , i0 ) are those of the intersection
with the hyperplane at infinity H = π(V ), there are three possible cases,
1. V does not contain rad E. In this case rad V = V ∩ rad E, and r0 = r, i0 = i.
2. rad E = rad V . In this case r0 = r − 1, and i0 = i, i − 1 (according to V contains or not a
maximal totally isotropic subspace of E).
3. rad E is a hyperplane of rad V . In this case r0 = r − 2, i0 = i − 1.
Proof: If the hyperplane V does not contain rad E, then E = V ⊥ hei, e ∈ rad E; so that
rad V = V ∩ rad E, since any vector in V is orthogonal to e, and we have an isometry V /rad V =
E/rad E; hence r0 = r, i0 = i. This is case 1.
If rad V = rad E, then V /rad V is a non singular hyperplane of E/rad E; hence r0 = r − 1,
and i0 = i, i − 1, according to V /rad V contains or not a maximal totally isotropic subspace of
E/rad E. This is case 2.
If rad E ⊂ rad V , projecting to E/rad E we may assume rad E = 0. Then dim V ⊥ = 1.
Since rad V = V ⊥ ∩ V 6= 0, we have dim rad V = 1 (hence r0 = r − 2), and V = V ⊥⊥ is the
orthogonal of an isotropic vector: i0 = i − 1. This is case 3.
Lemma: If the intersections Q ∩ H1 , Q ∩ H2 of a quadric with two hyperplanes are projectively

equivalent, there is a projectivity τ such that τ (Q) = Q, τ (H1 ) = H2 .
Proof: We put Hi = π(Vi ), and we distinguish the three possible cases.

1. We put rad E = rad Vi ⊥ Ti , and Vi = (rad Vi ) ⊥ Fi . We have
E = rad V1 ⊥ T1 ⊥ F1 = rad V2 ⊥ T2 ⊥ F2 .
By hypothesis we have an isometry (V1 , λS) → (V2 , S) and, Ti being totally isotropic, it may
be extended to an isometry T : (E, λS) → (E, S), and T (V1 ) = V2 .
2. We fix a supplement Ē of rad E, and we put Fi = Vi ∩ Ē, Ē = Fi ⊥ hei i. We have that
Vi = rad E ⊥ Fi , and
E = rad E ⊥ F1 ⊥ he1 i = rad E ⊥ F2 ⊥ he2 i
where a1 = e1 · e1 , a2 = e2 · e2 are non null. By hypothesis (F1 , λS) and (F2 , S) are isometric;
hence, if m = dim F1 , in the group k ∗ /k ∗2 we have
a1 disc F1 = disc Ē = a2 disc F2 = a2 λm disc F1
and a1 = λm a2 in k ∗ /k ∗2 . If m is even, we have an isometry he1 i ' he2 i and, by Witt’s

theorem, it may be extended to an isometry T̄ : Ē → Ē transforming he1 i⊥ = F1 into
he2 i⊥ = F2 , and it may be extended by the identity T : rad E ⊥ Ē → rad E ⊥ Ē so as to
obtain an isometry transforming V1 into V2 .
If m is odd, a2 = λa1 in k ∗ /k ∗2 , and (he1 i, λS) ' (he2 i, S). Since (F1 , λS) ' (F2 , S), we
obtain an isometry T : (E, λS) → (E, S), and T (V1 ) = V2 .
3. We put rad Vi = rad E ⊥ hei i, and let Ēi be a supplement of rad E containing hei i. The
orthogonal of hei i in Ēi is just Vi ∩ Ēi . Now, Ē1 and Ē2 are isometric and non singular,
hence by Witt’s theorem the isometry he1 i ' he2 i may be extended to an isometry Ē1 → Ē2 ,
transforming V1 ∩ Ē1 into V2 ∩ Ē2 . We obtain an isometry T : rad E ⊥ Ē1 → rad E ⊥ Ē2 ,
and T (V1 ) = V2 .
6.2. METRICS 167
Theorem: Two quadrics are affinely equivalent if and only if both, and their intersections with
the hyperplane at infinity, are projectively equivalent.
Proof: If Q and Q0 are projectively equivalent, Q0 = τ (Q) for some projectivity τ , then Q ∩ H
and Q0 ∩ τ (H) are projectively equivalent. If moreover Q ∩ H and Q0 ∩ H are projectively
equivalent, so are Q0 ∩ τ (H) and Q0 ∩ H. By the above lemma, σ(Q0 ) = Q0 , and σ(τ H) = H for
some projectivity σ. Hence Q0 = στ Q, and στ (H) = H.
Euclidean Geometry: An Euclidean structure on a real affine space An is given by a scalar

product Ω on the space V of free vectors, well defined up to a positive factor; i.e. a non singular
quadric hΩi of index 0 at infinity H = π(V ), named absolute: classical Euclidean geometry is
the projective geometry of a fixed hyperplane with a fixed imaginary non-singular quadric. An
affine reference system (P0 , . . . , Pn ; U ) is said to be Euclidean if there is a normalized base
e0 , . . . , en such that e1 , . . . , en is an orthonormal base of Ω, so that the absolute is
x21 + . . . + x2n = 0.
An affinity τ : Pn → Pn is a similarity if the absolute is invariant, τ Ω = ρ2 Ω (where τ

denotes the normalized linear representant, inducing the identity on E/V ) for some positive real
number ρ, the similarity ratio of τ . Motions are similarities of ratio 1.
Lemma: Let Q1 , Q2 be the points where a line P1 + P2 intersects a quadric Q = hSi. If φ is a

e1 ·e2
complex number5 such that cos φ = √ 2 2
, where Pi = π(ei ), then
e1 e2
(P1 , P2 ; Q1 , Q2 ) = e2φi .
Proof: We have Qi = π(αi e1 + e2 ), where α1 , α2 are the roots of 0 = e21 t2 + 2(e1 · e2 )t + e22 .
q
e1 ·e2 (e1 ·e2 )2
p
2 2 √ + −1
α2 e1 · e2 + (e1 · e2 )2 − e1 e2 e21 e22 e21 e22
(P1 , P2 ; Q1 , Q2 ) = = p = q
α1 e1 · e2 − (e1 · e2 )2 − e21 e22 e1 ·e2
√
2
− (ee12·ee22) − 1
2
e1 e2 2
1 2
p
2
cos φ + cos φ − 1 cos φ + i sin φ e φi
= p = = −φi = e2φi .
2
cos φ − cos φ − 1 cos φ − i sin φ e
Definition: The point at infinity of a line being represented by the direction of the line, the
1
angle determined by two intersecting lines r1 , r2 may be defined to be φ = | 2i ln (r1 , r2 ; i, j)|,
where i, j are the self-conjugate lines with respect to the absolute in the pencil determined by
r1 and r2 . The lines are perpendicular when (r1 , r2 ; i, j) = −1; i.e., φ = π/2.
Now let Q be a quadric of rank r in an Euclidean space.
Quadrics differing in a motion are considered to be equal.
Since any symmetric endomorphism diagonalizes in an orthonormal base (p. 162) there are
Euclidean references where the intersection Q ∩ H with the hyperplane at infinity is
a1 x21 + . . . + ar0 x2r0 = 0,
and we shall place the origin P0 so that the equation of the quadric becomes simple.
We distinguish the three possible cases (p. 166) and we assume that rad E = 0 (if rad E 6= 0,
the equation is the same, just put E = Ē ⊥ rad E):
5
Well-defined up to a sign and an integer multiple of π, and φ is real when Q1 , Q2 are conjugate complex
points, (e1 · e2 )2 < e21 e22 . Since cos di = cosh d = 12 (ed + e−d ), it may be fixed to be a pure imaginary φ = di when
Q1 , Q2 are real points, (e1 · e2 )2 ≥ e21 e22 , not separating P1 and P2 (i.e., e21 and e22 have equal sign).
1. The vertex has some proper point P0 , and the quadric is a1 y12 + . . . + ar yr2 = 0.
2. The pole of the infinity is a center P0 , and the quadric is a1 y12 + . . . + ar−1 yr−1
2 = 1.
3. The quadric is a paraboloid tangent to H at a point, say P1 (hence a1 = 0). The polar variety
of P2 + . . . + Pn is a line that passing through P1 and intersecting the quadric at a proper
point P0 . The quadric is y1 = a2 y22 + . . . + ar−1 yr−1
2 .
Metric Elements of Conics: Given a non singular conic in an Euclidean plane, an axis is
a line conjugate to the perpendicular lines, and the vertices are the intersection points with
the axes. A focus is any point where conjugate lines are just perpendicular lines (and the
corresponding polar line is said to be a directrix), so that the focuses of a conic hSi are defined
by the singular conics of the focal series hS ∗ + λΩ∗ i, where hS ∗ i is the dual conic and hΩ∗ i is
the absolute, viewed as a pair of imaginary lines in the dual plane.
Non Euclidean Geometries: In Euclidean geometry, the absolute defines a singular imag-
inary quadric hΩ∗ i in the dual space P∗n , with vertex at a point, the hyperplane at infinity.
Perpendicularity is just conjugation with respect to hΩ∗ i, proper points are points defining hy-
perplanes in P∗n not intersecting hΩ∗ i at real points, and proper lines are lines passing through
some proper point. Non Euclidean geometries are obtained by fixing the absolute to be an
arbitrary non singular quadric in Pn (non ruled, so that proper points exist).
Hyperbolic Geometry: The absolute hΩ∗ i is a real quadric of index 1, hence the dual of a
real quadric Q = hΩi in Pn . Proper points of this geometry are interior points (without real
tangents) of Q, and points at infinity are those of Q. Proper lines are lines intersecting Q at
1
two points, and the angle determined by two intersecting lines r1 , r2 is φ = | 2i ln (r1 , r2 ; i, j)|,
where i, j are the self-conjugate lines in the pencil determined by r1 and r2 , which are imaginary
conjugate lines. In this geometry the distance is an absolute concept!

1 |e ·e |
d(P1 , P2 ) = 2 ln (P1 , P2 ; Q1 , Q2 ) = arc cosh √ 2 2
1 2
e1 e2
where Pi = π(ei ) and the line P1 + P2 intersects Q at the points Q1 , Q2 (the cross-ratio being
positive because Q1 , Q2 do not separate P1 , P2 ). Straight lines have infinite length; but this
geometry violates Euclid’s fifth postulate: by an exterior point pass infinite parallel lines.
Hence, the fifth postulate is independent of the other four postulates.
Elliptic Geometry: The absolute hΩ∗ i is an imaginary non singular quadric, dual to an
imaginary quadric Q = hΩi in Pn . All points are proper, hence there are no parallel lines.
1
Angles are defined to be φ = | 2i ln (r1 , r2 ; i, j)|, and we also have an absolute distance (but
now Q1 , Q2 are imaginary conjugate points and straight lines have finite length π)

ln (P1 , P2 ; Q1 , Q2 ) = arccos |e|e11|·|e
·e2 |
1
d(P1 , P2 ) = 2i 2|
.
6.3 Modules over a Principal Ideal Domain

Let Σ be the field of fractions of a principal ideal domain A.
Let M be an A-module, and MΣ = M ⊗A Σ the localization of M by S = A − {0}.
Definition: The rank of M is rk M = dimΣ MΣ . In particular rk Ar = r.
If 0 → M 0 → M → M 00 → 0 is an exact sequence, then rk M = rk M 0 + rk M 00 .
Lemma: Any submodule N of a free A-module L of rank r is free, of rank ≤ r.
Proof: By induction on r. If r = 1, then N ' aA ' A or 0.

If r > 1, then L is a direct sum of two non null free modules, L = L0 ⊕ L00 , and we have a
commutative diagram with exact rows
π
0 −→ L0 −→ L −−→ L00 −→ 0
∪ ∪ ∪
π
0 −→ N 0 = N ∩ L0 −→ N −−→ N 00 = π(N ) −→ 0
6.3. MODULES OVER A PRINCIPAL IDEAL DOMAIN 171
By induction, N 0 and N 00 are free, and the sum of ranks is ≤ r.

Since N 00 is free, the bottom exact sequence splits and N ' N 0 ⊕ N 00 is free, of rank ≤ r.
Corollary: Any submodule of a finitely generated A-module is finitely generated.
Proof: Any finitely generated module is a quotient of a free module Ar .
Definitions: The kernel of the localization morphism M → MΣ = M ⊗A Σ is the torsion

submodule T (M ) = {m ∈ M : am = 0, for some non null a ∈ A}.
M is a torsion module if M = T (M ), and M is torsion free if T (M ) = 0.
Lemma: Any finitely generated torsion free A-module M is free.
Proof: Let m1 , . . . , mn be a generating system of M , where m1 , . . . , mr are linearly independent

and ai mi ∈ Am1 + . . . + Amr , with ai 6= 0. Let b = ar+1 . . . an 6= 0.
Since bM is a submodule of the free module Am1 + . . . + Amr , it is free.
If M is torsion free, M → bM , m 7→ bm, is an isomorphism, and we conclude.
Theorem: Any finitely generated A-module decomposes, uniquely up to isomorphisms, as a

direct sum of a free module and a torsion module
M ' Ar ⊕ T (M ), r = rk M.
Proof: M/T (M ) is torsion free; hence it is free and the following exact sequence splits,
0 −→T (M ) −→ M −→ M/T (M ) −→ 0
M ' (M/T (M )) ⊕ T (M ).
Now, if M ' L ⊕ T , where L is free and T is torsion free, localizing we have that MΣ ' LΣ ,
and L is free of rank r = rk M . Moreover,
T (M ) ' T (L) ⊕ T (T ) = 0 ⊕ T = T.
Definition: The annihilator ideal of an A-module M is Ann M = {b ∈ A : bM = 0}.

When A is a principal ideal domain, the generator a is the annihilator of M , and it is well
defined up to invertible factors. We put Ker b = {m ∈ M : bm = 0}, so that Ker a = M .
Torsion finitely generated A-modules have non null annihilator a, and finite length, since
l(A/aA) < ∞. Moreover, Ann(M1 ⊕ M2 ) = Ann(M1 ) ∩ Ann(M2 ).
Lemma: Ker pq = Ker p ⊕ Ker q, when p and q are coprime.
Proof: (See p. 79). By Bézout’s identity, 1 = λp + µq, and for all m ∈ M ,
m = λpm + µqm.
If m ∈ Ker pq, then λpm ∈ Ker q, and µqm ∈ Ker p; hence Ker pq = Ker p + Ker q.
If m ∈ Ker p ∩ Ker q, then m = λpm + µqm = 0 + 0 = 0.
First Decomposition Theorem: Let pn1 1 · · · pns s be the irreducible factor decomposition of the
annihilator of a finitely generated torsion A-module M . Then M decomposes uniquely as a direct
sum of submodules annihilated by pni i ,
M = Ker pn1 1 ⊕ . . . ⊕ Ker pns s .

Proof: Since M = Ker (pn1 1 · · · pns s ), the existence follows from the above lemma.
Now, if M = M1 ⊕ . . . ⊕ Ms , where pni i Mi = 0, then Mi ⊆ Ker pni i .
If some inclusion is strict, so is M = ⊕i Mi ⊂ ⊕i Ker pni i = M . Absurd.
Definition: A finitely generated A-module is primary if it is annihilated by some power of an

irreducible element p ∈ A. Modules isomorphic to A/pn A are named primary monogenous.
Second Decomposition Theorem: Any primary finitely generated A-module M decomposes,

uniquely up to isomorphisms, as a direct sum of primary monogenous modules
M ' (A/pn1 A) ⊕ . . . ⊕ (A/pns A), n1 ≥ · · · ≥ ns .
Proof: We consider a minimal generating system m1 , . . . , ms of M , and we proceed by induction

on s, since theorem is obvious when M is monogenous.
If the annihilator of M is pn , some generator, say m1 , is not annihilated by pn−1 , and we
have an exact sequence of modules over the ring B = A/pn A,
0 −→ B = Am1 −→ M −→ M̄ −→ 0.
It splits, since B is an injective B-module (p. 123); hence M ' A/pn A ⊕ M̄ , where M̄ is
generated by m̄2 , . . . , m̄s , and we obtain the existence by induction.
Finally, put k = A/pA. We have (pi A/pj A) ⊗A k ' k when 0 ≤ i < j, so that the number
νj of terms A/pj A in a decomposition of M does not depend on the decomposition
dimk (M ⊗ k) = ν1 + . . . + νn ,
dimk ((pM ) ⊗A k) = ν2 + . . . + νn ,
..................
dimk ((pn−1 M ) ⊗A k) = νn .
6.3.1 Classification
Let M be a finitely generated A-module. By the above theorem, there are irreducible elements
n
pi ∈ A and powers pi ij , well defined up to units and named elementary divisors of M , such
that M decomposes as a direct sum
L
n
M ' (A⊕ . r. . ⊕A) ⊕ A/pi ijA , r = rk M.
i,j
Classification Theorem: The rank and elementary divisors classify finitely generated modules
over a principal ideal domain A.
Definitions: If M is a module over an arbitrary ring A, the tensor algebra of M
TA• M = ⊗n
L
nM = A ⊕ M ⊕ (M ⊗A M ) ⊕ . . .
is a (non commutative) algebra with a canonical morphism M → T • M such that any morphism
of A-modules M → B into a (not necessarily commutative) A-algebra B uniquely factors trough
a morphism of A-algebras T • M → B.
HomA (M, B) = HomA-alg (TA• M, B).

The symmetric algebra SA • M of M is the quotient of T • M by the two sided ideal generated
A
by the elements m ⊗ m0 − m0 ⊗ m, while the exterior algebra Λ•A M of M is the quotient of
TA• M by the two sided ideal generated by the elements m ⊗ m,
•
S n M = (TA• M )/(m ⊗ m0 − m0 ⊗ m),
L
SA M=
Ln
Λ•A M = nΛ
n
M = (T • M )/(m ⊗ m).
Both commute with base changes, (S • M ) ⊗A B = S • (MB ) and (Λ• M ) ⊗A B = Λ• (MB ),

since so do tensor products; hence they commute with localizations.
Moreover, Λ0 M = A and Λ1 M = M since the ideal (m ⊗ m) has no element of degree ≤ 1,
and SA• M is a commutative algebra while Λ• M is an anticommutative algebra,
A
an bm = (−1)nm bm an ; an ∈ Λn M, bm ∈ Λm M.
We have a canonical morphism M → S • M such that any A-linear map M → B into a

commutative A-algebra B uniquely factors trough a morphism of A-algebras S • M → B,
•
HomA (M, B) = HomA-alg (SA M, B).
Proposition: Λ• (M ⊕ N ) = (Λ• M ) ⊗A (Λ• N ), where the algebra structure of the right term is
defined by the product (an ⊗ bm )(ar ⊗ bs ) = (−1)mr an ar ⊗ bm bs .
Proof: The natural morphism M ⊕ N → (Λ• M ) ⊗ (Λ• N ) induces a morphism of algebras

T • (M ⊕ N ) → (Λ• M ) ⊗ (Λ• N ), and it factors through Λ• (M ⊕ N ).
On the other hand, the natural morphisms Λ• M → Λ• (M ⊕ N ), Λ• N → Λ• (M ⊕ N ) induce
the inverse morphism (Λ• M ) ⊗ (Λ• N ) → Λ• (M ⊕ N ).
n
Corollary: If L is a free module of rank n, then Λp L is a free module of rank

p . Hence, if M
is monogenous, Λp M = 0 for all p > 1.
Definition: The invariant factors of a finitely generated module M over a principal ideal
domain A are the annihilators φj (well defined up to units) of Λj M , j ≥ 1.
Theorem: The invariant factors classify finitely generated modules over a principal ideal domain
n
A. If r is the rank of M , and pi ij are the elementary divisors (ni1 ≥ ni2 ≥ . . .), then
n n
φ1 = . . . = φr = 0, φr+j = p1 1j p2 2j . . .
M ' A/φ1 A ⊕ . . . ⊕ A/φd A
Proof: Λ• (N1 ⊕ N2 ) = (Λ• N1 ) ⊗A (Λ• N2 ), and Λ• N = A ⊕ N when N is monogenous.

Since A/I ⊗A A/J = A/(I + J), the annihilator φj A of Λj M is the intersection Q nof all sums
n
of j ideals of the family 0, . r. ., 0, (pi ij ), so that φ1 = . . . = φr = 0, and φr+j A = i pi ij A.
The chinese remainder theorem shows that M ' A/φ1 A ⊕ . . . ⊕ A/φd A.
f π
Definition: Let us consider a presentation L0n −−→ Lm −−→ M −→ 0, where L0n and Lm are free
modules of ranks n and m. The image of the morphism
Λf ⊗1
Λm−i L0 ⊗ (Λm−i L)∗ −−−−−→ Λm−i L ⊗ (Λm−i L)∗ −→ A
is the i-th Fitting ideal Fi (M ) = ci A. It is generated by all minors of order m − i of the matrix
of f (we agree that Fi = 0 when m − i > n, and Fi = A when m − i < 1).
ci−1
Proposition: ci = φi+1 . . . φd , so that φi = ci ·
Proof: Localizing at a point of Spec A we may assume that A is a local ring.

Let k = A/pA be the residue field. Now a decomposition
M = A/φ1 A ⊕ . . . ⊕ A/φd A = A⊕ . r. . ⊕A ⊕ A/φr+1 ⊕ . . . ⊕ A/φd A
defines generators m1 , . . . , md of M , and we consider the exact sequence

f ⊗1 π⊗1
L0n ⊗A k −−−−→ Lm ⊗A k −−−−→ M ⊗A k = k d −→ 0.
P lemma, there is a base e1 , . . . , emPof L such that π(ei ) = mi , i ≤ d.

By Nakayama’s
If π(ej ) = i aij mi , j > d, replacing ej by ej − i aij ei we may assume that π(ej ) = 0, so
that φr+1 er+1 , . . . , φd ed , ed+1 , . . . , em is base of Ker π.
Using the same argument with L0 , we obtain bases where any coefficient of the matrix of f
is null, except some diagonal coefficients φr+1 , . . . , φd , 1, . m−d
. . . . ., 1.
Now the theorem is obvious.
Classification of Abelian Groups: The invariant factors classify finitely generated abelian
groups. Any finitely generated abelian group is a direct sum of infinite cyclic groups and cyclic
groups of order some powers of prime numbers.
Corollary: Any finite abelian group of order n has subgroups of any order dividing n.
Proof: In fact, pn−i Z/pn Z is a subgroup of Z/pn Z of order pi .
Corollary: A system of diophantine linear equations AX = B admits an integer solution if and

only if A and (A|B) have equal rank r, and cr (A) = cr (A|B).
Proof: Let us consider the morphism f : Zn → Zm , f (X) = AX.

The rank of M = Zm /Im f is rk (M ) = m − rk (A), and the order of T (M ) is cr (A).
If the system has integer solutions, then M = M/ZB, and ci (A) = ci (A|B) for any index i.
If it has no integer solution, then B 6= 0 in M . If B is not a torsion element of M , then
rk (M/ZB) = rk (M ) − 1, and rk (A|B) = rk (A) + 1.
Otherwise the torsion of M is bigger than the torsion of M/ZB, and cr (A) 6= cr (A|B).
Definition: Let T be an endomorphism of a k-vector space E of finite dimension n. We put
p(x) · e = p(T )(e)
where p(x) ∈ k[x], e ∈ E. This k[x]-module ET is a finite torsion k[x]-module since dimk E < ∞,
and submodules V ⊆ E are just invariant vector subspaces, T (V ) ⊆ V .
Two endomorphisms T, T 0 of E are equivalent if the corresponding modules are isomorphic,
ET ' ET 0 ; i.e., if there is a linear automorphism τ of E such that
T 0 = τ ◦ T ◦ τ −1 .
The invariant factors of T are defined to be those of the k[x]-module ET , and we may
calculate them with a matrix A of T in a base e1 , . . . , en of E. In fact the exact sequence
(x−y)·
0 −→ k[x, y] −−−−−→ k[x, y] −→ k[x] −→ 0
x⊗1−1⊗x
0 −→k[x] ⊗k k[x] −−−−−−−→ k[x] ⊗k k[x] −→ k[x] −→ 0
splits because k[x] is free, and applying (−) ⊗k[x] ET we obtain an exact sequence
x⊗1−1⊗T
0 −→ k[x] ⊗k E −−−−−−−→ k[x] ⊗k E −→ ET −→ 0
where k[x] ⊗k E is a free k[x]-module of base (1 ⊗ e1 , . . . , 1 ⊗ en ), and the matrix of x ⊗ 1 − 1 ⊗ T

is just the characteristic matrix xI − A (and the characteristic polynomial of T may be
intrisically defined to be the determinant of the endomorphism x ⊗ 1 − 1 ⊗ T of the free k[x]-
module k[x] ⊗k E of rank n).
Classification of Endomorphisms: If ci is the greatest common divisor of all minors of order

n − i of xI − A, then the invariant factors of T are φi = ci−1
ci ·
Corollary: The characteristic polynomial c0 = |xI − A| is the product of the invariant factors,
c0 = φ1 . . . φd . Hence, the characteristic polynomial is a multiple of the annihilator polynomial
φ1 , and both have the same irreducible factors.
Lemma: 1, x − α, . . . , (x − α)n−1 form a base of k[x]/((x − α)n ).
Proof: A base of k[y]/(y n ) is (1, y, . . . , y n−1 ); where y = x − α. q.e.d.
When ET ' k[x]/((x − α)n ), the matrix of T in the base {ej = (x − α)j−1 } is
T (ej ) = x · (x − α)j−1 = (x − α)(x − α)j−1 + α(x − α)j−1 = ej+1 + αej ,

 
α 0 0
 
 1 α 
 
 0 (6.1)
 

 

 0
0 0 1 α
If the elementary divisors of T are (x − αi )nij , in some base the matrix of T is a Jordan
matrix (Bij is a matrix nij × nij of the form 6.1, with αi in the diagonal),
 
..
 . 

 Bij 
 (6.2)
..
.
so obtaining reduced equations of endomorphisms if k is algebraically closed.

If k = R, we consider a complex number α = a + bi, b 6= 0, and the irreducible polynomial
(x − α)(x − ᾱ) = x2 − 2ax + (a2 + b2 ).
Lemma: R[x]/(x − α)n (x − ᾱ)n = C[x]/(x − α)n , isomorphism of R[x]-modules.
Proof: The annihilator of C[x]/((x − α)n ) is a multiple of (x − α)n (x − ᾱ)n , since it has real
coefficients. We conclude because dimR C[x]/((x − α)n ) = 2n. q.e.d.
The matrix of T in the base {ej = (x − α)j−1 , e0j = i(x − α)j−1 } is
T (ej ) = x · (x − α)j−1 = (x − α)j + α(x − α)j−1

= (x − α)j + a(x − α)j−1 + bi(x − α)j−1 = ej+1 + aej + be0j
T (e0j ) = x · i(x − α)j−1 = i(x − α)j + iα(x − α)j−1
= i(x − α)j + ai(x − α)j−1 − b(x − α)j−1 = e0j+1 + ae0j − bej
 
A 0 0
 
 I A 
a −b 1 0
 
 0  , A= , I= (6.3)
 
  b a 0 1

 0 

0 0 I A
In general, in some base of E the matrix of T is a Jordan matrix 6.2, where the matrices Bij
have the form 6.1 or 6.3.
Definition: Two projectivities τ, τ 0 : P(E) → P(E) are equivalent if τ 0 = στ σ −1 for some

projectivity σ (if T and T 0 are linear representants of τ and τ 0 , this condition states that the
endomorphisms T 0 and λT are equivalent for some non null constant λ ∈ k).
Lemma: If φi (x) are the invariant factors of T , then those of λT are φi (x/λ).
Proof: If E = k[x]/(φ1 (x)) ⊕ . . . ⊕ k[x]/(φn (x)) and we put y = λx, then
E ' k[y]/ φ1 ( λy ) ⊕ . . . ⊕ k[y]/ φn ( λy )

where the structure of k[y]-module of E is defined by the endomorphism λT .

This isomorphism shows that the invariant factors of λT are just φi ( λy ).
Classification of Projectivities: Two projectivities are equivalent if and only if the invariant
factors of the linear representants have proportional (non null) roots, φ0i (x) = φi (λx).
P Gl(2, C)
Invariant Factors Equation Fixed Points
φ1 = (t − 1)(t − a) θ0 = aθ 2 points
φ1 = (t − 1)2 θ0 =θ+1 1 point
φ1 = φ2 = t − 1 θ0 = θ all
P Gl(3, C)
Invariant Factors Equations Fixed Points
φ1 = (t − 1)(t − a)(t − b) x0 = ax, y0 = by 3 points
φ1 = (t − 1)2 (t − a) x0 = x + 1, y 0 = ay 2 points
φ1 = (t − 1)3 x0 = x + 1, y 0 = x + y 1 point
φ1 = (t − 1)(t − a), φ2 = t − 1 x0 = x, y 0 = ay 1 line, 1 point
φ1 = (t − 1)2 , φ2 = t − 1 x0 = x + 1, y 0 = y 1 line
φ1 = φ2 = φ3 = t − 1 x0 = x, y 0 = y all
P Gl(4, C)
Invariant Factors Equations Fixed Points
φ1 = (t − 1)(t − a)(t − b)(t − c) x0 = ax, y 0 = by, z 0 = cz 4 points
φ1 = (t − 1)2 (t − a)(t − b) x0 = x + 1, y 0 = ay, z 0 = bz 3 points
φ1 = (x − 1)2 (t − a)2 x0 = x + 1, y 0 = ay, z 0 = y + az 2 points
φ1 = (t − 1)3 (t − a) x0 = x + 1, y 0 = x + y, z 0 = az 2 points
φ1 = (t − 1)4 x0 = x + 1, y 0 = x + y, z 0 = y + z 1 point
φ1 = (t − 1)(t − a)(t − b), φ2 = t − 1 x0 = x, y 0 = ay, z 0 = bz 1 line, 2 points
φ1 = (t − 1)2 (t − a), φ2 = t − 1 x0 = x + 1, y 0 = y, z 0 = az 1 line, 1 point
φ1 = (t − 1)2 (t − a), φ2 = t − a x0 = x + 1, y 0 = ay, z 0 = az 1 line, 1 point
φ1 = (t − 1)3 , φ2 = t − 1 x0 = x + 1, y 0 = x + y, z 0 = z 1 line
φ1 = φ2 = (t − 1)(t − a) x0 = x, y 0 = ay, z 0 = az 2 lines
φ1 = (t − 1)(t − a), φ2 = φ3 = t − 1 x0 = x, y 0 = y, z 0 = az 1 plane, 1 point
φ1 = (t − 1)2 , φ2 = φ3 = t − 1 x0 = x + 1, y 0 = y, z 0 = z 1 plane
φ1 = φ2 = (t − 1)2 x0 = x + 1, y 0 = y, z 0 = y + z 1 line
φ1 = φ2 = φ3 = φ4 = t − 1 x0 = x, y 0 = y, z 0 = z all
6.3.2 Grothendieck K-Group

Definitions: An additive function on the category C of finitely generated A-modules, with
values in an abelian group G, assigns to any finitely generated A-module M an element χ(M ) ∈ G
so that χ(M ) = χ(M 0 ) + χ(M 00 ) for any exact sequence
(∗) 0 −→ M 0 −→ M −→ M 00 −→ 0
An additive function ξ : C → K is universal if any other additive function χ : C → G

uniquely factors trough ξ; i.e., there is a unique group morphism f : K → G such that χ = f ◦ ξ.
If ξ1 : C → K1 , ξ2 : C → K2 are universal additive functions, then there is a unique group
isomorphism f : K1 → K2 such that ξ2 = f ◦ ξ1 .
(We also have a K-group of finite length A-modules, of finite projective A-modules, etc.)
The existence of a universal additive function follows from the representability theorem; but
we shall give a direct construction. Let us consider the free abelian group on isomorphism
classes of finitely generated A-modules, and the quotient K(A) by the subgroup generated by
elements M − M 0 − M 00 , one for each exact sequence (∗). If [M ] denotes the class of M in K(A),
then [M ] = [M 0 ] + [M 00 ] and, for any additive function χ there is a unique group morphism
f : K(A) → G such that χ(M ) = f ([M ]). Remark that [M ⊕ N ] = [M ] + [N ], so that any
element of K(A) is a difference [M1 ] − [M2 ], but not uniquely.
Example: If A is a principal ideal domain, the decomposition theorems show that we have
n f
[M ] = r[A] + ij [A/pi ij A] in K(A). Moreover, the exact sequence 0 → A −
P
→ A → A/f A → 0,
where f 6= 0, shows that [A/f A] = [A] − [A] = 0. Hence [M ] = r[A], and rg : K(A) → Z is an
isomorphism.
Definition: If A is a principal ideal domain, the divisors on X = Spec A are the formal linear
combinations of closed points n1 x1 +. . .+ns xs with integer coefficients, and they form an abelian
group Div(X). The divisor of a function f = pn1 1 . . . pns s ∈ A is
D(f ) = n1 x1 + . . . + ns xs , mxi = pi A.
Examples: D : Q∗+ → Div(Spec Z) is a group isomorphism.

xn +...
Div(Spec k[x]) is isomorphic to the multiplicative group of rational functions xm +... and a
rational function f /g corresponds to D(f ) − D(g).
n
pi ij =
Q Q
Theorem: The product of the elementary divisors, c0 = ij i φi , defines the universal
additive function on finite length A-modules,
P
Klf (A) = Div(X), [M ] 7→ D(c0 (M )) = i,j nij xi .
n
A/pi ij A. Since l(A/pni A) = n,
L
Proof: M = M1 ⊕ . . . ⊕ Ms , where Mi ' j
P P
D(c0 (M )) = i ( j nij )xi = l(M1 )x1 + . . . + l(Ms )xs ,
and this is an additive function since so is length. Therefore it defines a group morphism
Klf (A) → Div(X), [A/mx ] 7→ x, and it is an isomorphism because the elements [A/mx ] generate
Klf (A), since [A/pn A] = n[A/pA], as the following exact sequences show
pn
0 −→ A/pA −−−→ A/pn+1 A −→ A/pn A −→ 0
Corollary: The order is the universal additive function on finite abelian groups.
The characteristic polynomial is the universal additive function on endomorphisms.
6.4 Classification of Pairs of Metrics

Let S, S 0 be symmetric metrics, where S is non singular, on a k-vector space E of finite dimension,
char k 6= 2. We put e · v = S(e, v).
S being non singular, we have S 0 (e, v) = (T e) · v for a unique endomorphism T : E → E, and
T defines a structure of k[x]-module on E. Since S and S 0 are symmetric, we have
(T e) · v = e · (T v),
and the polarity φ : E −→∼

E ∗ of S is an isomorphism of k[x]-modules.
If T̃ : F → F is the endomorphism of another pair of metrics S̃, S̃ 0 , where S̃ is non singular,
then a k-linear isomorphism F − ∼
→ E transforms the pair (S, S 0 ) into (S̃, S̃ 0 ) if and only if it is
an isomorphism of k[x]-modules transforming S into S̃.
Let A = k[x]/(φ), where φ(x) is the annihilator polynomial of T .
By the representability theorem, the functor M M ∗ is representable, so that there is an

A-module D such that
M ∗ = HomA (M, D).
When M = A, we see that D = A∗ ; hence E ∗ = HomA (E, A∗ ).
Now, since the annihilator of the k[x]-module A∗ = Homk (A, k) is also φ(x), we have A∗ = Aξ
for some linear form ξ : A → k, and
E ∗ ' HomA (E, A), ω = ω A when ω(e) = ξ ω A (e) .

Therefore, φ induces an isomorphism of A-modules

∼
φ : E −−→ E ∗ ' HomA (E, A)
6.4. CLASSIFICATION OF PAIRS OF METRICS 179
and there is a unique A-bilinear map S A : E × E → A such that e · v = ξ S A (e, v) . It is

non-singular and symmetric because so is S,
ξ S A (v, e) = v · e = e · v.

Moreover, an A-linear isomorphism F → E transforms a non singular metric S̃ into S if and

only if it transforms S̃ A into S A : Classification of pairs of metrics, with fixed annihilator φ(x),
is reduced to the classification of non singular metrics on finitely generated A-modules.
Fix an A-bilinear metric S on a finitely generated A-module E, non singular in the sense
that the polarity φ : E → HomA (E, A) is an isomorphism, and put e · v = S(e, v) ∈ A.
If φ(x) = pn1 1 . . . pns s is the irreducible factor decomposition, then (p. 171)
E = E1 ⊕ . . . ⊕ Es , Ei = ker pni i .
First Decomposition Theorem: This decomposition is orthogonal, Ei · Ej = 0, i 6= j.

n n
Proof: Ei · Ej = (pni i A + pj j A)(Ei · Ej ) = (pni i Ei ) · Ej + Ei · (pj j Ej ) = 0. q.e.d.
Hence, we may assume that A = k[x]/(pn ) is local, and we put m = pA, K = A/m.
Any metric S : E ⊗A E → A defines, by the base change A → K (p. 128), a K-bilinear
metric S̄ : Ē ⊗K Ē → K, where Ē = E/pE. By definition, ē · v̄ = e · v ∈ K.
Remark that, when E is free and S is non singular, then so is S̄.
Lemma: Let (L, S), (L0 , S 0 ) be free A-modules with non singular metrics. Any K-linear isometry
σ̄ : (L̄, S̄) → (L̄0 , S̄ 0 ) may be lifted to an A-linear isometry σ : (L, S) → (L0 , S 0 ).
Proof: By induction on r = rk L and n. If L = Ae, we put q = pn−1 , and we must lift any
isometry σ̄ : L/qL → L/qL0 to an isometry σ : L → L0 . We fix a representant e0 ∈ L0 of σ̄(ē).
Since σ̄ is an isometry, e0 · e0 = (1 + aq)(e · e). If we put b = 1 − 12 aq, then (be0 ) · (be0 ) = e · e
since q 2 = 0, and σ(e) = be0 is the required isometry.
If r > 1, we fix e ∈ L with ē ∈ L̄ non isotropic, and a representant e0 ∈ L0 of the non isotropic
vector σ̄(ē). Then L = (Ae) ⊥ F , L0 = (Ae0 ) ⊥ F 0 .
By induction the isometries σ̄ : K ē → K ē0 , σ̄ : F̄ → F̄ 0 may be lifted to isometries Ae → Ae0 ,
F → F 0 , defining the required isometry σ : L → L0 .
Definition: A submodule H is homogeneous if H ' A/pd A ⊕ . . . ⊕ A/pd A.

In such a case, pd H = 0 and S takes values in {a ∈ A : pd a = 0} = mn−d ' A/md .
Hence we may view S as an (A/md )-bilinear metric on the (A/md )-module H, defining by
base change a metric K-vector space (H̄, S̄).
We also have a K-bilinear metric S̄ on (ker pd ), and the radical contains the image of ker pd−1
and pE ∩ ker pd ; hence it projects to the quotient, and we obtain a metric morphism
ker pd
(H̄, S̄) −→ Ēd = , S̄ .
[ker pd−1 + (pE ∩ ker pd )]
Second Decomposition Theorem: E is an orthogonal sum E = H1 ⊥ . . . ⊥ Hn of homoge-

neous modules Hd of annihilator pd , and uniquely up to isometries.
Proof: We prove the existence by induction on dim E.

Since pn−1 E 6= 0, we have pn−1 e · v 6= 0 for some vectors e, v, so that the metric S̄ on
Ē = E/pE is not zero; hence e · e is invertible in A for some e ∈ E.
Now E = (Ae) ⊥ (Ae)⊥ , and by induction E is an orthogonal sum of monogenous modules.
Grouping terms with equal annihilator, we see that E is an orthogonal sum of homogeneous
modules.
Finally, if E = H1 ⊕. . .⊕Hn , then the metric morphisms (H̄d , S̄) → (Ēd , S̄) are isometries (in
particular Ēd is non singular) becauseHr ∩ ker pd ⊆ pE ∩ ker pd when r > d, and Hr ⊆ ker pd−1
when r < d. Now uniqueness follows from the above lemma.
Classification of Metrics on Modules: Non singular metrics on finitely generated A-modules

(E, S) are classified by the invariant factors of E and the non singular metric (A/mi )-vector
spaces (H̄i,d , S̄), where Hi,d is a homogeneous component of E of annihilator mdi .
Classification of Pairs of Metrics: Pairs of metrics on a k-vector space E (the first non
singular) are classified by the invariant factors of the associated endomorphisms T and the non
singular metric k[x]/(pi )-vector spaces (H̄i,d , S̄ A ).
Corollary: If k is algebraically closed, pairs of metrics, the first non singular, are classified by
the elementary divisors of the associated endomorphism.
Proof: The dimension classifies non singular metrics on k-vector spaces.
Reduced equations: A pair of metrics with a unique elementary divisor (x − α)n is

 
  0 0 1 α
0 0 1  
 0
0
   
 , S0 = 

S= 0

 0   
 
   1 
1 0 0
 
α 0 0
Definitions: Pencils of quadrics λS + µS 0 in P(E) are lines in the space of quadrics P(S2 E).
Every point in P(E) which is not a base point (lying on all the quadrics of the pencil) lies on
exactly one quadric. The fundamental points of a pencil are the points of the vertices of the
singular quadrics. We only may consider pencils with some non singular quadric.
The endomorphisms associated with two pairs of quadrics in the pencil differ in a homog-
raphy. Even if the invariant factors are not invariants of the pencil, so are the degrees and
multiplicities of the roots (and cross-ratios if there are 4 or more roots).
Pencils of Conics in P2,C
1. Annihilator with 3 simple roots. A pencil defined by 2 conics intersecting at 4 points. There
are 3 pairs of lines, and 3 non-collinear fundamental points.
λ(x20 + x22 + x22 ) + (x20 − x21 ) = 0.
2. Annihilator with a simple and a double root. A pencil defined by 2 conics intersecting at
3 points, with a common tangent at one of them. There are 2 fundamental points, one
being a base point, where any conic in the pencil is tangent to the line passing through the
fundamental points.
λ(2x0 x1 + x22 ) + (x20 + x22 ) = 0.
6.4. CLASSIFICATION OF PAIRS OF METRICS 181
3. Annihilator with a triple root. A pencil defined by 2 conics intersecting at 2 points, with a
common tangent at one of them. There is a unique fundamental point.
λ(2x0 x1 + x22 ) + 2x0 x1 = 0.
4. Annihilator with 2 simple roots. A pencil defined by 2 conics intersecting at 2 points, with
common tangent at both. There is a line of fundamental points, and an exterior fundamental
point.
λ(x20 + x21 + x22 ) + x20 = 0.
5. Annihilator with a double root. A pencil defined by 2 conics intersecting at a unique point,
with common tangent. There is a line of fundamental points.
λ(2x0 x1 + x22 ) + x20 = 0.
Pencils of Quadrics in P3,C
1. Annihilator with 4 simple roots. There are 4 cones, with non coplanar vertices. There are
infinite equivalence classes, according to the cross-ratio a of the 4 roots.
λ(x20 + x21 + x22 + x23 ) + (ax20 + x21 − x22 ) = 0 ; a 6= 0, 1, −1.
2. Annihilator with 2 simple roots and a double root. There are 3 cones, with non collinear
vertices. The vertex of a cone lies in the other two cones.
λ(2x0 x1 + x22 + x23 ) + (x20 − x22 + x23 ) = 0.
3. Annihilator with 2 double roots. There are 2 cones, and the line joining the vertices is a
common generatrix.
λ(2x0 x1 + 2x2 x3 ) + (x20 + 2x0 x1 + x22 ) = 0.
4. Annihilator with a simple and a triple root. There are 2 cones, the vertex of a cone lying on
the other cone.
λ(x21 + 2x0 x2 + x23 ) + (2x0 x1 + x23 ) = 0.
5. Annihilator with a root of multiplicity 4. There is a unique cone.
λ(2x0 x3 + 2x1 x2 ) + (2x0 x2 + x21 ) = 0.
6. Annihilator with 3 simple roots, one a simple root of φ2 . There is a pair of planes, and 2
cones with exterior vertices.
λ(x20 + x21 + x22 + x23 ) + (x22 − x23 ) = 0.
7. Annihilator with a double and a simple root, also a simple root of φ2 . There is a pair of
planes, and a cone with incident vertex.
λ(2x0 x1 + x22 + x23 ) + (2x0 x1 + x20 ) = 0.
8. Annihilator with a simple and a double root, also a simple root of φ2 . There is a pair of
planes, and a cone with exterior vertex, tangent to the vertex of the pair of planes.
λ(2x0 x1 + x22 + x23 ) + (x20 + x23 ) = 0.

9. Annihilator with a triple root, also a simple root of φ2 . There is a pair of planes, one being
tangent to any quadric in the pencil.
λ(2x0 x2 + x21 + x23 ) + 2x0 x1 = 0.
10. Annihilator φ1 = φ2 , with 2 simple roots. There are 2 pairs of planes, with skew vertices.
λ(x20 + x21 + x22 + x23 ) + (x20 + x21 ) = 0.
11. Annihilator with 2 simple roots, one a simple root of φ2 and φ3 . There is a a double plane,
and a cone with exterior vertex.
λ(x20 + x21 + x22 + x23 ) + x20 = 0.
12. Annihilator φ1 = φ2 , with a double root. There is a unique pair of planes.
λ(2x0 x1 + 2x2 x3 ) + (x20 + x22 ) = 0.
13. Annihilator with a double root, also a simple root of φ2 = φ3 . There is a double plane.
λ(2x0 x1 + x22 + x23 ) + x20 = 0.
6.5 Semisimple Rings

In this section any ring A has a unity, but may be non-commutative, and Ao = EndA (A) will
denote the opposite ring: the set A with the same addition, and product a · b = ba.
The ring of n × n matrices with coefficients in A is denoted by Mn (A), and the transposition
of matrices defines an isomorphism Mn (Ao ) = Mn (A)o .
Except otherwise stated, all modules, submodules and ideals are on the left.
Definitions: A module S 6= 0 is simple if it has no proper submodules. A finitely generated

module M is semisimple (resp. homogeneous of type S) if it is a sum, hence a finite sum, of
simple submodules (resp. submodules isomorphic to S); hence so is any quotient of M .
Example: Let k be a field and G = {g1 , . . . , gn } a finite group. Then k[G] = kg1 ⊕ . . . ⊕ kgn is a
k-algebra with the product induced by G. If E is a k-vector space, any linear representation
(that is to say, group morphism) G → Autk (E) extends by linearity to a morphism of k-algebras
k[G] → Endk (E), so that E is a k[G]-module, and it is simple when the representation is
irreducible (E 6= 0 and any G-invariant vector subspace is trivial).
If the characteristic of k does not divide the order of G, any finite k[G]-module is semisimple
(Maschke’s theorem). In fact, it is enough to show that any exact sequence of k[G]-modules
π
0 → M0 → M − → M 00 → 0 splits. If a k-linear sectionPs of π is G-invariant, s = gsg −1 , then it is
k[G]-linear. Otherwise the arithmetic mean s̃ = |G| 1 −1 is a G-invariant section:
g gsg
1 −1 1 −1 1
P P P
πs̃ = |G| g∈G gπsg = |G| g∈G g(IdM 00 )g = |G| g∈G IdM 00 = IdM 00 .
Schur’s lemma: Any non-zero morphism f : S → S 0 between simple A-modules is an isomor-

phism. In particular, the ring EndA (S) is a division ring.
Proof: Since Ker f 6= S and Im f 6= 0, we have Ker f = 0 and Im f = S 0 .

6.5. SEMISIMPLE RINGS 183
Lemma: Let N be a submodule of an A-module M . If M = N + S1 + . . . + Sn , where the

submodules Si are simple, then M = N ⊕ Si1 ⊕ . . . ⊕ Sir for some indices i1 , . . . , ir . Hence, any
semisimple module is a direct sum of simple modules.
Proof: If N ∩ S1 = S1 , we drop S1 . If N ∩ S1 = 0, we replace N by N ⊕ S1 .

We conclude by induction on n.
Theorem: Any submodule N of a semisimple module M is a direct summand, M = N ⊕ N 0 .

P P
Proof: If M = i Si , then M = N + i Si , and we conclude by the lemma.
Corollary: If a module is semisimple (resp. homogeneous of type S), then so is any submodule.
In particular, the type of a homogeneous module is unique.
Theorem: Any semisimple module M admits a unique decomposition as a direct sum of homo-
geneous submodules of different types (the isotypic decomposition).
Proof: If M = S1 ⊕ . . . ⊕ Sn , grouping isomorphic summands we obtain the existence.

If S ⊆ M is a simple submodule, the projection S → Si is null or an isomorphism by Schur’s
lemma. Hence S is contained in the component of type S, and the uniqueness follows.
Definitions: A division ring is a ring where any non-zero element is invertible.

The center of a ring A is the subring Z(A) = {a ∈ A : ab = ba, ∀b ∈ A}.
When k is a (commutative) field, a k-algebra is a ring A endowed with a ring morphism
k → Z(A), and we say that it is a central k-algebra if k = Z(A).
A ring A is semisimple if it is a semisimple A-module (so that any finite A-module, being
a quotient of An , is semisimple), and it is a simple ring if it is a homogeneous A-module (so
that all finite A-modules are homogeneous of the same type).
The Fundamental Example: Let us see that the ring A = EndD (E) ' Mn (Do ) of endomor-
phisms of a n-dimensional vector space E over a division ring D is simple and how it determines
E and D. A simple A-module is E, and a basis e1 , . . . , en of E gives an isomorphism
EndD (E) = E⊕ . n. . ⊕E , T 7→ (T e1 , . . . , T en ),
hence A is a simple ring, and E is the unique simple A-module (any other being a quotient of
A). Moreover D = EndA (E) and, in particular, Z(A) = Z(D). In fact,
Mn (D) = Ao = HomA (A, A) = HomA (⊕n E, ⊕n E) = Mn (EndA (E)).
Skolem-Noether Theorem: Let E be a finite-dimensional vector space over a division ring D.

Any automorphism of the ring A = EndD (E) is induced by the conjugation with a semi-linear
automorphism E → E, well-defined up to the product with a homothety λId, λ ∈ D. Hence, the
group of automorphisms of the k-algebra Endk (E) is the group P Gl(E).
Proof: Since E is the unique simple module over the ring EndD (E), for any ring isomorphism
φ : A = EndD (E) → A0 = EndD0 (E 0 ) we have a group isomorphism T : E → E 0 such that
T (ae) = (φa)(T e), ∀a ∈ A, e ∈ E. That is to say, φ(a) = T aT −1 .
This isomorphism T is well-defined up to A-linear isomorphisms λId ∈ EndA (E) = D.
Moreover, T induces a ring isomorphism σ : D = EndA (E) → D0 = EndA0 (E 0 ) such that, for
any λ ∈ D, the following square commutes
E
T / E0
λ σ(λ)

E
T / E0
Hence T is a semi-linear transformation with associated isomorphism σ.

Finally, condition T (λId)T −1 = λId states that T is a linear map.
Theorem: Let A be a semisimple ring. Up to isomorphisms, there is a finite number of simple

modules S1 , . . . , Sr , and A is a direct product of matrix rings over division rings,
A = EndD1 (S1 ) × . . . × EndDr (Sr ) , Di = EndA (Si ).
Proof: Let A = S1n1 ⊕ . . . ⊕ Srnr be the isotypic decomposition.

Any simple module, being a quotient of A, appears in this decomposition.
Moreover, by Schur’s lemma, HomA (Si , Sj ) = 0 when i 6= j. Hence
Ao = HomA (⊕i Sini , ⊕i Sini ) = i HomA (Sini , Sini ) = i Mni (Di ).

Q Q
Corollary: Let A be a simple ring. Up to isomorphisms, there is a unique simple module S,

and A is a matrix ring over a division ring,
A = EndD (S) , D = EndA (S).
Corollary: Any simple finite algebra A over an algebraically closed field k is A ' Mn (k).
Proof: We have k ,→ D = EndA (S) because k ,→ Z(A). Now, if α ∈ D, then k[α] is an integral
finite commutative k-algebra; hence it is a field and k = k[α], so that k = D.
Corollary: Let G be a finite group and k an algebraically closed field whose characteristic does
not divide |G|. The number of different irreducible representations G → Autk (Ei ) coincides with
the number of conjugacy classes in G, and the degrees di = dimk Ei satisfy that
|G| = d21 + . . . + d2r .
Proof: Since k[G] is semisimple and k is algebraically closed, k[G] = Endk (E1 ) × . . . × Endk (Er ).
Now it is clear that |G| = dim k k[G] = i d2i , and that r is just the dimension of
P
Z(k[G]) = Z(Endk (E1 )) × . . . × Z(Endk (Er )) = k× . r. . ×k.
We have i λi gi ∈ Z(k[G]) if and only if i λi (ggi g −1 ) = i λi gi , ∀g ∈ G. Hence a basis of

P P P
Z(k[G]) is defined by the elements g1 + . . . + gs , where {g1 , . . . , gs } is a conjugacy class.
Theorem: A ring A is simple if and only if it admits a minimal ideal (a simple submodule)
and any two-sided ideal of A is trivial.
Proof: Let A be a simple ring. If I 6= A is a two-sided ideal, since I annihilates A/I ' S ⊕. . .⊕S,
we have IS = 0; hence IA = I(S ⊕ . . . ⊕ S) = 0, and I = 0.
Conversely, if A admits a minimal ideal I and any two-sided ideal is trivial, then
P
A = IA = a∈A Ia.
Since we have an epimorphism I → Ia, we see that Ia is a simple module isomorphic to I

whenever non null. Hence A is a homogeneous A-module.
Wedderburn’s Theorem: Let A be a simple ring and S a simple A-module. The category of
A-modules is equivalent to the category of right modules over the division ring D = EndA (S).
Proof: If M is an A-module, then F (M ) = HomA (S, M ) is a right D-module.

If E is a right D-module, then G(E) = E ⊗D S is a left A-module via the action on S.
The natural morphisms G(F (S)) → S, D → F (G(D)) are isomorphisms, and both functors
preserve direct sums and inductive limits.
Since any A-module (resp. right D-module) is an inductive limit of direct sums of modules
isomorphic to S (resp. D), we conclude that F and G define an equivalence of categories.
6.5.1 The Brauer Group

Definition: An Azumaya k-algebra is a finite simple central k-algebra; hence it is isomorphic
to a matrix algebra Mn (D) over a division ring D such that k = Z(D).
Theorem: A finite k-algebra A is an Azumaya algebra if and only if the natural morphism
φA : A ⊗k Ao → Endk (A), φA (a ⊗ b)(x) = axb, is an isomorphism.
Proof: If A is an Azumaya algebra, we consider the image B of φA ,
B ,→ Endk (A) = A ⊕ . . . ⊕ A.
Since two-sided ideals of A are just (A ⊗k Ao )-submodules and A is a simple ring, it is a

simple B-module. Hence B is a homogeneous B-module: it is a simple ring.
It follows that B is the ring of endomorphisms of the simple module,
B = EndD (A) , D = EndB (A) = EndA⊗Ao (A) = Z(A) = k,
so that B = Endk (A), and φA is surjective.

Since dim (A ⊗k Ao ) = dim Endk (A), it is an isomorphism.
Conversely, if φA is an isomorphism, then A is a simple (A ⊗k Ao )-module, because A is
always a simple Endk (A)-module. Hence A is a simple ring, and moreover
Z(A) = EndA⊗k Ao (A) = EndEndk (A) (A) = k.
Corollary: If A and B are two Azumaya k-algebras, then so is A ⊗k B.
Proof: φA⊗k B = φA ⊗ φB .
Corollary: Let k → L be a commutative extension. A finite k-algebra A is an Azumaya k-

algebra if and only if A ⊗k L is an Azumaya L-algebra.
Proof: φA⊗k L = φA ⊗ 1.
Corollary: A finite k-algebra A is an Azumaya k-algebra if and only if A ⊗k k̄ ' Mn (k̄). In

particular, the dimension of any Azumaya k-algebra is a perfect square, dimk A = n2 .
Definition: The Azumaya k-algebras form an abelian group Br(k) under the tensor product
if we identify Mn (D) and Mm (D0 ) whenever D ' D0 . The neutral element is the class of all
trivial Azumaya k-algebras Mn (k), and the inverse of a class [A] is the class [Ao ]. It is the
Brauer group of k, and its elements correspond to the central extensions k → D. For any
commutative extension k → L we have a group morphism Br(k) → Br(L), [A] 7→ [A ⊗k L], the
kernel Br(L/k) being the subgroup defined by the Azumaya k-algebras trivial over L.
Theorem: An Azumaya k-algebra A of degree n2 is trivial, A ' Mn (k), if and only if contains
a trivial commutative subalgebra B of degree n, B ' k⊕ . n. . ⊕k.
Proof: If A ' Mn (k), then the diagonal matrices define a subalgebra B ' k⊕ . n. . ⊕k.
Conversely, assume that A admits a trivial subalgebra B ' k⊕ . n. . ⊕k.
Then we have an isomorphism of A-modules
A = A ⊗B B = A ⊗B (k⊕ . n. . ⊕k) = (A ⊗B k)⊕ . n. . ⊕(A ⊗B k) = I1 ⊕ . n. . ⊕In ,
so that dim Ii ≤ n for some index i. Now, the ring morphism A → Endk (Ii ) is injective because
A is simple, and it is an isomorphism since dim A = n2 ≥ dim Endk (Ii ).
Lemma: Any Azumaya k-algebra A of degree n2 contains a separable commutative subalgebra

of degree n.
Proof: If a ∈ A, then the commutative subalgebra k[a] ' k[x]/(q(x)) is separable of degree n
when so is the annihilator polynomial q(x). Hence, it is enough to show that the characteristic
polynomial of the endomorphism a : A → A is separable for some a ∈ A.
Now, the discriminant of the characteristic polynomial is a polynomial on A, and it is not
the zero polynomial since it does not vanish at some points of A ⊗k k̄ ' Mn (k̄).
When k is infinite, we conclude that it does not vanish at some point a ∈ A.
When k is a finite field, any extension is separable, and it is enough to show that any division
ring D of degree n2 contains a commutative field of degree n.
Pick a commutative field L ⊂ D of degree d < n.
Since ⊕k̄ ' L ⊗k k̄ ⊂ D ⊗k k̄ ' Mn (k̄), we have L ⊗k k̄ = k̄[T ], where the annihilator of the
endomorphism T is a separable polynomial of degree d.
Now T is diagonalizable and commutes with all the diagonal matrices.
The commutator of L ⊗k k̄ in D ⊗k k̄ has degree ≥ n, and the commutator of L in D strictly
contains L: there exists α ∈ D, α ∈ / L, such that the field L[α] is commutative.
Theorem: If A is an Azumaya k-algebra, then there exists a finite Galois extension k → L such
that A ⊗k L is trivial, A ⊗k L ' Mn (L).
Proof: If dim A = n2 , pick a separable commutative algebra B ⊂ A of degree n.

Now (p. 146) B is trivial over a finite Galois extension k → L, and A ⊗k L contains a trivial
commutative L-algebra B ⊗k L of degree n; hence it is trivial, A ⊗k L ' Mn (L).
Definition: Let k → L be a Galois extension of Galois group G. An action of G on a L-vector

space E is a group morphism ρ : G → Autk (E) such that ρ(g) is a semi-linear transformation of
automorphism g. Two actions on E and F are equivalent if there is a L-linear isomorphism
T : E → F preserving the actions, T (ge) = g(T e).
An action of G on the projective space P(E) is a group morphism ρ of G into the group
of all collineations, such that g is the automorphism of the collineation ρ(g). Two actions are
equivalent if some projectivity transforms one into the other.
An action of G on a L-algebra A is a group morphism ρ : G → Autk-alg (A) such that ρ(g)
is a semi-linear transformation of automorphism g. Two actions on A and B are equivalent if
there is an isomorphism of L-algebras A → B preserving the actions.
Let us consider the ring L[G] = ⊕g∈G Lg with the product (λP1 g1 )(λ2 g2 ) =Pλ1 g1 (λ2 ) · (g1 g2 ),
so that L[G]-modules are just L-vector spaces with a G-action, ( i λi gi )e = i λi (gi e).
Lemma: The natural morphism L[G] → Endk (L) is an isomorphism. Hence L[G] is a simple
ring, and the simple L[G]-module is just L with the natural G-action.
Proof: We have that Homk (L, L) = HomL (L ⊗k L, L) = HomL (⊕L, L) and the elements of G
correspond to the projections ⊕L → L. Now the result is obvious.
Corollary: The natural morphism E G ⊗k L → E is an isomorphism for any L[G]-module E.
Proof: It holds when E = L, because k = LG (p. 147), and in general E ' ⊕L.
Theorem: Let k → L be a Galois extension of group G. The algebra of invariants Mn (L)G of

any action of G on the L-algebra Mn (L) is an Azumaya k-algebra of degree n2 trivial over L,
and so we obtain a bijection
 
Azumaya k-algebras of
degree n2 trivial over L = G-actions on Mn (L) = G-actions on Pn−1
up to equivalence up to equivalence
up to isomorphisms
Proof: The natural morphism Mn (L)G ⊗k L → Mn (L) is an isomorphism; hence Mn (L)G is an

Azumaya k-algebra of degree n2 trivial over L, and it determines the action of G on Mn (L) up
to equivalence. Moreover, for any Azumaya k-algebra A of degree n2 trivial over L, we have
that A = A ⊗k (LG ) = (A ⊗k L)G = Mn (L)G .
Finally, the last bijection is a direct consequence of the Skolem-Noether theorem.
Definition: Let k → L be a cyclic Galois extension of degree d and let us fix a generator g of the
Galois group G. Now a G-action on a projective space is just a collineation τ of automorphism g
and order d. If T is a semi-linear representant, condition τ d = Id states that T d = µId, µ ∈ L∗ .
Since µT = T d T = T T d = T µ, we have g(µ) = µ, so that in fact µ = inv(T ) is in k ∗ .
But it depends on the semi-linear representant T ,
inv(λT ) = (λT )d = λ(gλ) . . . (g d−1 λ)T d = N (λ)inv(T ),
where N (λ) = λ(gλ) . . . (g d−1 λ) is the norm of λ ∈ L∗ . Hence the invariant inv(τ ) = [µ] is
well-defined in the group k ∗ /N (L∗ ), but it depends on the fixed generator g of G.
Remark that inv(T ⊕ . . . ⊕ T ) = inv(T ), so that inv(τ ) only depends on the class of the
Azumaya k-algebra, and we obtain a map inv : Br(L/k) → k ∗ /N (L∗ ) which is a group morphism
because inv(T ⊗ T̄ ) = inv(T ) · inv(T̄ ).
Now, the representant T : E → E defines on E a structure of module over the ring
Lµ [G] = L ⊕ Lg ⊕ . . . ⊕ Lg d−1 ; g d = µ, g · λ = g(λ) · g.
Lemma: The ring Lµ [G] is simple (in fact it is an Azumaya k-algebra).

√
Proof: We have an isomorphism Lµ [G] ⊗k k̄ → L[G] ⊗k k̄ = Md (k̄), g ⊗ 1 7→ g ⊗ d µ.
Theorem: If k → L is a cyclic extension, inv : Br(L/k) → k ∗ /N (L∗ ) is an isomorphism.
Proof: The Azumaya k-algebra A defined by a simple Lµ [G]-module S has invariant [µ], and any
other Azumaya k-algebra B defined by a semi-linear transformation of invariant µ corresponds
to a module E ' S⊕ . n. . ⊕S; hence B ' Mn (A) and we conclude.
Frobenius Theorem: There is a unique non-commutative finite extension of R,
Br(R) = Z/2Z.
Proof: Br(R) = Br(C/R) = R∗ /N (C∗ ) = R∗ /R∗+ = {±1}.
Wedderburn’s Theorem: Every finite division ring is commutative, Br(Fq ) = 0.
Proof: Any finite extension Fq → Fqd is cyclic and the Galois group is generated by the Frobënius
automorphism F (x) = xq . Hence the norm N : F∗qd → F∗q is
2 d−1 q d −1
N (x) = xF (x)F 2 (x) . . . F d−1 (x) = xxq xq . . . xq =x q−1
−1d
and the kernel of N has order ≤ qq−1 , so that N is surjective and Br(Fqd /Fq ) = 0.
S
We conclude that Br(Fq ) = d Br(Fqd /Fq ) = 0.
Example: Let G = {Id, j} be the Galois group of C over R, where j stands for the complex
conjugation. Since C−1 [G] is a non-trivial Azumaya R-algebra of degree 4, it is the unique
non-commutative finite extension of R. We denote it H and name it the quaternions,
H = C−1 [G] = C ⊕ Cj = R ⊕ Ri ⊕ Rj ⊕ Rij ; i2 = j 2 = −1, ji = −ij.

Chapter 7
Commutative Algebra
7.1 The Spectrum of a Ring

Let A be a ring and M an A-module.
If U is an open set in X = Spec A, then AU and MU = M ⊗A AU denote the localization of
A and M by the multiplicative system of all functions without zeros in U .
When U = Uf = Spec Af is a basic open set, then the functions without zeros in U are
invertible in Af , so that AU = Af (hence MU = Mf ), and U = Spec AU .
Now let us consider the map
f = ` Mx −→ X, π(mx ) = x.
π: M
x∈X
When x ∈ U , we have a morphism MU → Mx , so that m ∈ MU defines a section m̃ : U → M f

of π. If two sections m̃, ñ coincide at a point, mx = nx , then f n = f m, where f (x) 6= 0, and
both sections coincide on the neighborhood Uf ∩ U . The images of these sections m̃ form a base
of a topology on Mf, the sections m̃ are continuous, and π is a local homeomorphism.
Moreover, any continuous section of π locally coincides with some section m̃, and Mx is
just the set of germs at x of continuous sections. The continuous sections of π on U form an
AU -module HomX (U, M f), and we have a natural morphism of AU -modules
MU −→ HomX (U, M
f).
Theorem: MU = HomX (U, M

f), when U = Uf is a basic open set.
Proof: Since Uf = Spec AUf , we may assume that U = X.

We know that the morphism is injective (p. 138). Let us see that it is surjective.
S → M be a continuous section. Since X is compact, there is a finite basic open
Let s : X f
cover X = Ui , and elements si ∈ MUi such that s|Ui = s̃i .
The sections s̃i and s̃j coincide on the basic open set Ui ∩ Uj and, since the morphism is
injective, si = sj in MUi ∩Uj ; hence we have to prove the exacteness of the sequence
i L i1 ,i2 L
0 −→ M −→ i MUi ⇒ MUi ∩Uj
i,j
(∗) || ||
M ⊗A B M ⊗A B ⊗A B
where B = ⊕i AUi , i(m) = m ⊗ 1, and i1 (m ⊗ b) = m ⊗ b ⊗ 1, i2 (m ⊗ b) = m ⊗ 1 ⊗ b.

It is clear when the natural ring morphism A → B admits a retract r : B → A: if m ⊗ b ⊗ 1 =
m ⊗ 1 ⊗ b, applying 1 ⊗ 1 ⊗ r we obtain m ⊗ b = r(b)m ∈ M .
189
190 CHAPTER 7. COMMUTATIVE ALGEBRA
S
If A → B = ⊕i AUi has no retract, since X = i Ui , it is enough to prove the exactness after
applying the functor M MB := M ⊗A B (p. 138)
MB −→ MB ⊗B BB ⇒ MB ⊗B BB ⊗B BB
i.e. the corresponding sequence for the B-module MB and the morphism B → BB = B ⊗A B,
b 7→ 1 ⊗ b, admitting the obvious retract µ : B ⊗A B → B, µ(b0 ⊗ b) = b0 b.
Direct proof: Let Ui = X − (fi )0 . Given elements si ∈ Mfi , such that si = sj in Mfi fj , we must
prove the existence of m ∈ M such that si = m in Mfi .
Put si = m f n , with a common exponent n.
i
i
mi mj
Since fin = fjn in Mfi fj , there exists r such that fir fjr fjn mi = fir fjr fin mj .
mi fir mi
Since fin = we may assume that fjn mi = fin mj .
fir fin , when n 0
We have X = j Uj ; hence A = j fjn A, and 1 = j φj fjn , where φj ∈ A.
S P P
We put m = j φj mj , and let us prove that m = m

P
f n in Mfi ,
i
i
fin m = φj fin mj = φj fjn mi = φj fjn mi = 1 · mi .

P P P
j j j q.e.d.
1. If the spectrum disconnects, X = U q V , then the ring decomposes, A = AU ⊕ AV .

U and V are basic open sets: the section with value 0 on U and 1 on V is continuous.
2. If M is supported on a finite number of closed points x1 , . . . , xn (i.e. Mx = 0 when x 6= xi ),

then M = Mx1 ⊕ . . . ⊕ Mxn .
3. If M is annihilated by some power of a maximal ideal m, then M = Mm .
Note: From now on, in these notes the spectrum Spec A of a ring A will be considered to
be the pair (Spec A, A), and a morphism Spec B → Spec A is defined to be a ring morphism
A → B, so that it induces a continuous map Spec B → Spec A, but different morphisms may
induce the same continuous map. When A and B are algebras over a field k, then the k-
morphisms Spec B → Spec A are defined to be the morphisms of k-algebras A → B, so that
Homk (Spec B, Spec A) = Homk-alg (A, B).
7.2 Primary Decomposition

Definition: A module is noetherian if any submodule is finitely generated (equivalently, any
strictly increasing sequence of submodules is finite).
A ring is noetherian if any ideal is finitely generated.
Lemma: If 0 → M 0 → M → M 00 → 0 is exact, then M is noetherian ⇔ so are M 0 and M 00 .
Proof: It is enough to show that M is finitely generated when so are M 0 and M 00 , since any
submodule of M has an analogous exact sequence.
Let us fix epimorphisms L0 = An → M 0 , L00 = Am → M 00 .
We extend the last one to a morphism L00 → M , and we conclude since p is surjective,
0 / L0 / L0 ⊕ L00 / L00 /0
p
z
0 / M0 /M / M 00 /0
7.2. PRIMARY DECOMPOSITION 191
Theorem: Any finitely generated module over a noetherian ring is noetherian.
Proof: If A is noetherian, so is An , and any quotient of An , by the lemma.
Theorem: If A is a noetherian ring, so is A[x].
Proof: Let I be an ideal of A[x].

The leading coefficients of polynomials in I form an ideal a1 A + . . . + ar A.
Fix polynomials Pi (x) = ai xn + . . . ∈ I, of P
equal degree n.
If Q(x) = axm + . . . ∈ I, m ≥ n, then a = i bi ai , and deg (Q − i bi xm−n Pi ) < m,
P
I = (P1 , . . . , Pr ) + (I ∩ L),
where L = A ⊕ Ax ⊕ . . . ⊕ Axn−1 is a noetherian A-module.

Therefore I ∩ L is a finitely generated A-module, and the ideal I is finitely generated.
Corollary: If A is a noetherian ring, so is any finitely generated A-algebra.
Proof: The ring A[x1 , . . . , xn ] is noetherian; hence so is any quotient.
Definitions: An ideal q 6= A is irreducible if it is not a proper intersection of two ideals.

The radical rad I = {a ∈ A : an ∈ I, for some n ≥ 1} of an ideal I is (p. 136) the intersection
of all prime ideals containing I, and it is the greatest ideal such that (rad I)0 = (I)0 .
b·
An ideal q 6= A is primary if any homothety A/q −
→ A/q, b ∈ A, is injective or nilpotent,
/ q ⇒ bn ∈ q.
ab ∈ q, a ∈
In such a case p = rad q is a prime ideal, and we say that q is a p-primary ideal.
Lemma: If A is noetherian, any irreducible ideal is primary.
Proof: Let b : A/q → A/q be a homothety. Ker b ⊆ Ker b2 ⊆ . . . ⊆ Ker bn ⊆ . . .

Since A is noetherian, Ker bn = Ker bn+1 for some exponent n,
Ker b ∩ Im bn = 0.
Since q is irreducible, we have that Ker b = 0 or Im bn = 0.

If Ker b = 0, then b is injective. If Im bn = 0, then b is nilpotent.
Theorem: Any ideal I of a noetherian ring A is a finite intersection of primary ideals.
Proof: If I is not irreducible, I = I1 ∩ I2 . If some ideal Ii is not irreducible,... The process ends
by noetherianness, and I is a finite intersection of irreducible (hence primary) ideals. q.e.d.
Now, since intersections of p-primary ideals are p-primary, grouping ideals and eliminating
redundancies, I admits a reduced primary decomposition I = q1 ∩ . . . ∩ qn , where pi 6= pj , and
I 6= q1 ∩ . . . b
qi . . . ∩ qn (we put pi = rad qi ).
Corollary: The number of minimal prime ideals of a noetherian ring is finite.
Proof: Let 0 = q1 ∩ . . . ∩ qn be a primary decomposition.

We have Spec A = (q1 )0 ∪ . . . ∪ (qn )0 = (p1 )0 ∪ . . . ∪ (pn )0 .
Now it is clear that any minimal prime coincides with some ideal pi .
Lemma: Any p-primary ideal q is defined by infinitesimal conditions at the point p,
q = A ∩ qAp .
Proof: If f ∈ A ∩ qAp , then sf ∈ q, where s ∈

/ p = rad q; hence f ∈ q.
Theorem: Let I = q1 ∩ . . . ∩ qn be a reduced primary decomposition of an ideal I of a noetherian

ring A. The primary component qi does not depend on the decomposition when (pi )0 is an
irreducible component of (I)0 .
Proof: When j 6= i, we have qj Api = Api since pj = rad qj intersects A − pi .
Ipi = q1 Api ∩ . . . ∩ qn Api = qi Api ,
so that qi = A ∩ qi Api = A ∩ Ipi only depends on I and pi . q.e.d.
The annihilator of ā ∈ A/I is denoted by (I : a) := {b ∈ A : ba ∈ I} ⊇ I.

If q is p-primary, then (q : a) is p-primary when a ∈
/ q, and (q : a) = A when a ∈ q.
Theorem: Let I = q1 ∩ . . . ∩ qn be a reduced primary decomposition of an ideal I of a noethe-

rian ring A. The associated primes pi = rad qi are just the prime ideals coinciding with the
annihilator of some element of A/I; hence they do not depend on the decomposition.
T
Proof: If p = (I : a) = i (qi : a) is prime, taking radicals we see that p is an intersection of
some ideals pi ; hence p = pi for some index i.
Conversely, if a ∈ q2 ∩ . . . ∩ qn , a ∈
/ q1 , then (I : a) = (q1 : a) is a p1 -primary ideal, and we
may consider the first power pr1 contained in (q1 : a).
Now, if b ∈ pr−1
1 and b ∈/ (q1 : a), then (I : ab) = (q1 : ab) = p1 .
Corollary: The union of the associated primes of the ideal 0 is the set of all zero divisors.
Proof: Let 0 = q1 ∩ . . . ∩ qn . If a is zero divisor, ab = 0 and b ∈

/ qi , then a ∈ pi .
Corollary: Any finitely generated module M over a noetherian ring A admits a chain of sub-
modules 0 = M0 ⊂ M1 . . . ⊂ Mn = M such that Mi /Mi−1 ' A/pi , with pi a prime ideal.
Proof: By the above theorem, any monogenous submodule has some element m of prime anni-
hilator p. We put M1 = Am ' A/p, and we conclude since M is noetherian.
7.3 Completion
Definition: Let I be a proper ideal of a ring A, and let M be an A-module. We put
(
e−n if m ∈ I n M, m ∈ / I n+1 M
kmk = T n
0 if m ∈ I M
The pseudometric d(m, m0 ) = km0 − mk defines the I-adic topology of M , and the sub-
modules I n M form a base of open (and closed) neighborhoods of 0.
Artin-Rees Lemma: Let A be a noetherian ring. If M 0 is a submodule of a finitely generated

module M , there is an exponent h such that
I n−h (M 0 ∩ I h M ) = M 0 ∩ I n M, n ≥ h.
7.3. COMPLETION 193
Proof: Put AD = A ⊕ I ⊕ . . . ⊕ I n ⊕ . . .
MD = M ⊕ IM ⊕ . . . ⊕ I n M ⊕ . . .
MD is an AD -module, with the obvious product I n × I m M → I n+m M .
AD and MD are noetherian: If I = (ξ1 , . . . , ξs ), then AD = A[ξ1 t, . . . , ξs t] ⊂ A[t], and MD is
a finite AD -module. The following submodule of MD ,
N = M 0 ⊕ M10 ⊕ . . . ⊕ Mn0 . . . , where Mn0 = M 0 ∩ I n M,
is finite, generated by e1 , . . . , er , that we may assume homogeneous of degree ≤ h.

Now any homogeneous element of N of degree ≥ h is
r
P r P
P
ai ei = bij aij ei , where deg(aij ei ) = h.
i=1 i=1 j
Since aij ei ∈ Mh0 , we see that AD Mh0 = 0 0

i.e., I n−h Mn−h = Mn0 .
L
n≥h Mn ;
Corollary: The I-adic topology of M induces on M 0 the I-adic topology of M 0 .
Proof: I n M 0 ⊆ Mn0 , and Mn+h

0 = I n Mh0 ⊆ I n M 0 .
nM
T
Corollary: If A is noetherian, any finitely generated A-module M is separated, nI = 0.
Proof: Localizing at a point of Spec A we may assume that A is local.

Since the topology induced on N = n I n M is trivial, we have N = IN .
T
Finally, N = 0 by Nakayama’s lemma.
Definition: Mc denotes the completion of M , formed by classes of Cauchy sequences.

We have a canonical morphism M → M c, and any morphism of A-modules f : M → N
b c→N
naturally induces a morphism of A-modules fb: M b , fb(mn ) = (f (mn )).
c = lim M/I n M.
Proposition: M ←−
Proof: If (mn ) is a Cauchy sequence, there is a subsequence such that kmi − mj k ≤ e−n when
i, j ≥ n, and (m̄n ) ∈ lim
←−
M/I n M since mi − mn ∈ I n M for all i ≥ n.
If (m̄n ) ∈ lim
←−
M/I n M , then kmi − mj k ≤ e−n when i, j ≥ n, and (mn ) is a Cauchy sequence.
These morphisms are well defined and mutually inverse. q.e.d.
1. Any Cauchy sequence (mn ) is equivalent to a series n sn , sn = mn+1 − mn ∈ I n M .

P
2. Let O be the local ring at the origin of An := Spec k[x1 , . . . , xn ], m = (x1 , . . . , xn ), O/md+1 =
{Polynomials of degree ≤ d}. The m-adic completion is O b = k[[x1 , . . . , xn ]], and the mor-
phism O → O assigns to any germ the Taylor expansion at the origin.
b
3. Let (O, m) be the local ring at the origin of a plane curve 0 = y + (terms of degree≥ 2). By
Nakayama’s lemma m = (x); hence mn = (xn ), k[x]/(xn ) = O/mn , and O b = k[[x]].
4. If O is the ring of germs at the origin of C ∞ functions on Rd , then mn is the ideal of all
T
2
germs with null Taylor expansion, and e−1/x ∈ mn . Hence O is not noetherian.
T
b p = lim Z/pn Z is the ring of p-adic numbers.

5. Z ←−
i p
Theorem: Let A be a noetherian ring. If 0 → M 0 → − M → − M 00 → 0 is an exact sequence of
finitely generated A-modules, then so is exact the sequence
î p̂
c0 −→
0 −→ M c −−
M →M c00 −→ 0
Proof: We put Mn0 = M 0 ∩ I n M . We have exact sequences
0 −→ M 0 /Mn0 −→ M I n M −→ M 00 /I n M 00 −→ 0
î p̂
0 −→ lim
←−
M 0 /Mn0 −→ Mc −−
→M c00
and lim
←−
M 0 /Mn0 = M
c0 since {M 0 } defines, by Artin-Rees, the I-adic topology on M 0 .
n
Now p̂ is surjective: I M → I n M 00 is surjective, and any series n s00n , with s00n ∈ I n M 00 , is
n
P
P 00 P P n
n sn = n p(sn ) = p̂ n sn ), where sn ∈ I M .
Theorem: M ⊗A A
b=M
c, when A is noetherian and M is finitely generated.
Proof: If L is free, L = An , then we have L

b=A bn = L ⊗A A.
b
m 0
In general, we fix a presentation A = L → L → M → 0, and we conclude
L0 ⊗A Ab −→ L ⊗A Ab −→ M ⊗A A
b −→ 0
|| || ↓
L
b 0 −→ L
b −→ M
c −→ 0
Corollary: If A is noetherian, A
b is a flat A-algebra.
Proof: Let 0 → M 0 → M be exact. Now, M is an inductive limit of finitely generated submodules

Mi , the submodules Mi0 = M 0 ∩ Mi are finitely generated, and M 0 = lim
−→
Mi0 . Therefore
0 −→ Mi0 ⊗A A
b −→ Mi ⊗A A
b
and 0 → M 0 ⊗A A
b → M ⊗A A
b is exact because inductive limits preserve exact sequences.
Corollary: I n A
b = Ib n , A/I n = A/
b Ib n , I n /I n+1 = Ib n /Ib n+1 , when A is noetherian.
Proof: 0 −→ I −→ A is exact and, since A

b is A-flat,
n
0 −→ I ⊗A A
b −→ A
b b = I ⊗A A
is exact. Hence I A b and I n A
b = I, b = Ib .
0 −→ In −→ A −→ A/I n −→ 0 is exact; hence so is
0 −→ I n ⊗A A b −→ A/I n −→ 0
b −→ A
since A/I n is complete (as any other module annihilated by I n ).

This exact sequence shows that A/I n = A/I
b nA b Ibn ; hence I n /I n+1 = Ib n /Ib n+1 .
b = A/
Corollary: If A is noetherian, Ab is complete and separated for the I-adic

b topology.
If (O, m) is a noetherian local ring, O
b is a complete local ring of maximal ideal m.
b
Proof: lim b Ib n = lim A/I n = A.

A/ b
←− ←−
To prove that O
b is local, of maximal ideal m,
b we have to show that 1 + f is invertible for
b Since O is complete, the series (1 + f )−1 = 1 − f + f 2 − f 3 + . . . converges.
any f ∈ m. b
7.3. COMPLETION 195
Definitions: An I-filtration of M is a chain of submodules M = M0 ⊇ M1 ⊇ . . . such that

IMn ⊆ Mn+1 . The associated graded module
GM = M/M1 ⊕ M1 /M2 ⊕ . . . ⊕ Mn /Mn+1 . . .
is a module over the graded ring GI A = A/I ⊕ I/I 2 ⊕ . . . ⊕ I n /I n+1 . . ., and the completion
M
c = lim M/Mn is an A-module.
←−
b
A morphism f : M → N is compatible with the filtrations if f (Mn ) ⊆ Nn , so that it induces
a morphism of GI A-modules f 0 : GM → GN , and a morphism of A-modules
b fˆ: M
c→N b.
Example: Let m = (x1 , . . . , xn ) ⊂ k[x1 , . . . , xn ], and let A = k[x1 , . . . , xn ]/I be the ring of
functions on a subvariety X = Spec A passing through the origin, m̄ = (x1 , . . . , xn ) ⊂ A.
The exact sequence mr ∩ I → mr → m̄r → 0 induces an exact sequence
mr ∩ I −→ mr /mr+1 −→ m̄r /m̄r+1 −→ 0
and Gm̄ A = k[x1 , . . . , xn ]/Iin , where Iin is the ideal of the tangent cone to X at the origin,
Iin = (fr )f ∈I , f = fr + (terms of higher degree).
Now, in Algebra we have a weak Inverse Function Theorem:
Lemma: If f 0 : GM → GN is surjective (injective), so is fˆ: M

c→N
b.
Proof: If the morphisms Mi /Mi+1 → Ni /Ni+1 are surjective, and n ∈ N

b,
n = f (m0 ) in N/N1 , where m0 ∈ M,

n = f (m0 ) + f (m1 ) in N/N2 , where m1 ∈ M1 , and so on
P
so that m = i mi ∈ M c, and fˆ(m) = P f (mi ) = n.
i
If f 0 is injective, and 0 6= (mi ) ∈ M
c = lim M/Mi , we consider the first index i such that
←−
0 6= mi ∈ M/Mi , so that mi ∈ Mi−1 /Mi ,→ Ni−1 /Ni , and we see that (f (mi )) 6= 0.
Lemma: Let A be a complete and separated ring with the I-adic topology. If GI A is a noetherian
ring, then A is a noetherian ring .
Proof: Let q be an ideal of A, that we filter with q ∩ I n . The ideal Gq of GI A is generated by

a finite number of homogeneous elements, and each generator of degree d defines a morphism
A → q, compatible with the filtrations if we put A0 = . . . = Ad = A, Ad+n = I n .
So we get a morphism Ar = L → q such that GL → Gq is surjective and, by the above
lemma, L b→b q is surjective; hence L → q is surjective,
∼
L −→ L
b
↓ ↓
q ,→ b
q (q is separated since so is A)
Theorem: If A is a noetherian ring, then A

b is a noetherian ring.
Proof: A
b is complete and separated for the I-adic
b topology, and GIbA
b = GI A is noetherian.
7.4 Dimension Theory

Definition: Let A be a ring. The (Krull) dimension of A is the maximal length of all chains
p0 ⊂ p1 ⊂ . . . ⊂ pn of prime ideals (or irreducible closed sets in X = Spec A).
If x, y ∈ X, we put x ≤ y when x ∈ y, i.e. py ⊆ px , and we say that x is a specialization
of y. So, dim A is the maximal length of all specializations x0 > x1 > . . . > xn .
Definitions: Let A = A0 ⊕ A1 ⊕ . . . ⊕ An ⊕ . . . be a graded ring, Am · An ⊆ Am+n .

Let M = ⊕n Mn be a graded A-module, Am · Mn ⊆ Mm+n .
If A = A0 [ξ1 , . . . , ξd ], where A0 is a ring of finite length (hence noetherian) and deg ξi = 1,
and M is a finitely generated A-module, then the A0 -modules Mn are of finite length. The
Hilbert and Samuel functions of M are H(n) = l(Mn ) and S(n) = l(M0 ) + . . . + l(Mn−1 );
so that ∆S(n) = S(n + 1) − S(n) = H(n).
Theorem: H(n) is a polynomial function of degree < d when n 0; hence S(n) is a polynomial
function of degree ≤ d when n 0.
Proof: By induction on d. If A = A0 , then M = A0 m1 + . . . + A0 ms , and Mn = 0 when n is

bigger that the degree of any generator mi . The Hilbert polynomial is H = 0.
If d ≥ 1, we consider the following exact sequences
ξ
0 −→ Kern −→ Mn −−d→ Mn+1 −→ Cokern+1 −→ 0
∆H(n) = H(n + 1) − H(n) = l(Cokern+1 ) − l(Ker n )
Since A is noetherian, Ker = ⊕n Kern and Coker = ⊕n Cokern are finite A-modules, and they
are annihilated by ξd ; hence both are finite A0 [ξ1 , . . . , ξd−1 ]-modules.
By induction, l(Cokern+1 ) and l(Kern ), hence the difference ∆H(n), are polynomial functions
of degree < d − 1 when n 0, and H(n) is a polynomial function of degree < d.
n+d−1 nd

Example: The Samuel function of k[x1 , . . . , xd ] is S(n) = d = d! + ....
n+d−1

In fact, Ker = 0, Coker = k[x1 , . . . , xd−1 ], and ∆S(n) = l(Cokern+1 ) = d−1 by induction
on d. Hence S(n) = n+d−1

d + c. Since S(1) = 1, the constant c is null.
Theorem: A noetherian local ring O has finite length if and only if dim O = 0.
Proof: If l(O) < ∞, then mm = mn+1 for some exponent, and mn = 0 by Nakayama’s lemma.
The unique prime ideal of O is m, and dim O = 0.
If dim O = 0, then the unique prime ideal of O is m, and some power of m = rad O is null,
O = O/mn . Hence the O/m-vector spaces mi /mi+1 have finite dimension, and
l(O) = l(O/mn ) = l(O/m) + l(m/m2 ) + . . . + l(mn−1 /mn ) < ∞.
Definition: Let (O, m) be a noetherian local ring. If q = (f1 , . . . , fd ) is an ideal of radical m,

then l(O/q) < ∞, and Gq O = (O/q)[ξ1 , . . . , ξd ], where ξi = [fi ] ∈ q/q2 has degree 1.
A q-filtration M = M0 ⊇ M1 ⊇ M2 ⊇ . . . is stable if there is an index h such that
n
q Mh = Mn+h (recall Artin-Rees). In such a case, if M is finitely generated, then GM is a
finitely generated Gq O-module, with a Samuel polynomial S(n) of degree ≤ d,
S(n) = l(M/M1 ) + . . . + l(Mn−1 /Mn ) = l(M/Mn ), n 0.
Lemma: The degree and first coefficient of S(n) do not depend on the stable q-filtration.
7.4. DIMENSION THEORY 197
Proof: Let us consider the q-adic filtration, S 0 (n) = l(M/qn M ), n 0.

We have qn M ⊆ Mn in any q-filtration, and Mn+h ⊆ qn M if it is stable.
Hence the degree and first coefficient of S(n) and S 0 (n) coincide,
S 0 (n) ≥ S(n), S(n + h) ≥ S 0 (n) , n 0.
Lemma: The degree of the Samuel polynomial does not depend on the m-primary ideal q.
Proof: Let S(n) = l(M/qn M ), and S 0 (n) = l(M/mn M ) when n 0. Since mk ⊆ q ⊆ m, we

have S 0 (kn) ≥ S(n) ≥ S 0 (n), n 0; so that the degrees of S(n) and S 0 (n) coincide.
Definition: The Samuel polynomial SO of a local ring (O, m) is the Samuel polynomial of
the m-adic filtration, SO (n) = l(O/mn O), n 0.
Since SO (n) is integer when n 0, we have SO (n) = md nd + . . . + m1 n1 + m0 = md!d nd + . . .

for some integers m0 , . . . , md , and md is said to be the multiplicity of O.
Example: Let O be the local ring at the origin of 0 = Pm (x1 , . . . , xd ) + (terms of degree > m).
We have an exact sequence
P m
0 −→ k[x1 , . . . , xd ] −−−→ k[x1 , . . . , xd ] −→ k[x1 , . . . , xd ]/(Pm ) = Gm O −→ 0
n+d−1 n+d−m−1 m
nd−1 + . . ., and the multiplicity is m.

so that SO (n) = d − d = (d−1)!
Theorem: deg SO/f O < deg SO , when f ∈ m is not a zero divisor.
Proof: 0 → f O/(f O ∩ mn ) → O/mn → Ō/m̄n → 0 is exact, where Ō = O/f O.
SŌ (n) = SO (n) − l(f O/(f O ∩ mn )) , n 0.
Now, mn ∩ f O is a stable m-filtration of f O by Artin-Res, and f O ' O since f is not a zero

divisor. Hence, l(f O/(f O ∩mn )) and SO (n) are polynomials of equal degree and first coefficient,
when n 0, and the difference has lower degree.
Definition: Some functions f1 , . . . , fd ∈ O form a system of parameters if they generate an

ideal of radical m; i.e., dim O/(f1 , . . . , fd ) = 0.
Theorem: The dimension of a noetherian local ring O coincides with the minimum number of
parameters, and with the degree of the Samuel polynomial SO (n).
1. dim O ≥ minimum number of parameters.

Let x1 , . . . , xs be the generic points of the irreducible components of Spec O. Since functions
separate points and closed sets, there exists fi ∈ O vanishing at all of them, except xi . Now
f = f1 + . . . + fs does not vanish at any point xi , and therefore dim O/f O < dim O. We
conclude by induction on dim O.
2. Minimum number of parameters ≥ deg SO .

If q = (f1 , . . . , fd ) is m-primary, the degree of the Samuel polynomial of any stable q-filtration
is ≤ d; hence so is the degree of SO by the above lemma.
3. deg SO ≥ dim O.
By induction on the degree of SO . If it is 0, then l(O/mn ) is constant when n 0, and
mn = mn+1 . Hence mn = 0 by Nakayama, and dim O = 0.
If deg SO ≥ 1, and p0 ⊂ . . . ⊂ pd is a chain of prime ideals of O, take f ∈ p1 not in p0 and put
Ō = O/(p0 + f O). By the above theorem, deg SO ≥ deg SO/p0 > deg SŌ , and by induction
deg SŌ ≥ d − 1, since p̄1 ⊂ . . . ⊂ p̄d is a chain of prime ideals of Ō. Hence deg SO ≥ d, and
deg SO ≥ dim O.
Corollary: 1. The dimension of a noetherian local ring O is finite.
2. dim (f )0 ≥ dim O − 1, when f ∈ m. (Krull’s Theorem)
3. dim O = dim O.b
4. dim O ≤ dimK (m/m2 ), where K = O/m.
Proof: (1) The minimal number of parameters (or the degree of SO ) is finite.
(2) If f¯1 , . . . , f¯r is a system of parameters of O/f O, then f1 , . . . , fr , f is a system of param-
eters of O; hence dim O ≤ 1 + dim O/f O.
(3) O and O b have equal Samuel functions, since O/mn = O/ b m b n.
(4) By Nakayama’s lemma, a base of m/m2 defines a generating system of m.
Definitions: Let O be a noetherian local ring, of residue field K = O/m.

The tangent cone is Spec Gm O, and the cotangent space is m/m2 . The ring O is regular
if m is generated by a minimal system of parameters, dim O = dimK (m/m2 ).
Proposition: A noetherian local ring O is regular ⇔ the tangent cone is an affine space,
Gm O = K[x1 , . . . , xd ] (isomorphism of graded rings).
Proof: If Gm O = K[x1 , . . . , xd ], then dimK (m/m2 ) = d = deg SO = dim O.

If O is regular of dimension d, we have an epimorphism
K[x1 , . . . , xd ] −→ Gm O −→ 0,
and the Samuel polynomial of both rings at the origin has degree d.
Since the local ring of K[x1 , . . . , xn ] is integral, and the degree of the Samuel polynomial of
the quotient by a non zero divisor decreases, it is an isomorphism.
Corollary: A noetherian local ring O is regular if and only if O

b is regular.
Proof: Gm O = Gmb O.
b
Corollary: Regular local rings are integral.
Proof: Let a, b ∈ O be non null. Since 0 = n mn , some ā ∈ mi /mi+1 , b̄ ∈ mj /mj+1 are non null
T
and, Gm O being integral, 0 6= ā · b̄ = [ab] ∈ mi+j /mi+j+1 . Hence ab 6= 0.
Theorem: If O is a regular local k-algebra and k = O/m, then O

b = k[[x1 , . . . , xd ]].
Proof: Let m = (f1 , . . . , fd ), d = dim O. The morphism k[x1 , . . . , xd ] → O, xi 7→ fi , induces an

isomorphism after completion, since it induces an isomorphism (p. 195)
∼
k[x1 , . . . , xd ] −−→ Gm O = (O/m)[f¯1 , . . . , f¯d ].
7.5. FINITE MORPHISMS 199
Proposition: Let O be a noetherian local k-algebra. If K = O/m is a finite separable extension

of k, the ring morphism O
b → K admits a section, and O b = K[[x1 , . . . , xd ]] if O is regular.
Proof: The epimorphisms π0k (O/mn ) → π0k (K) are isomorphisms, because π0k (O/mn ) is a field
(p. 153); hence we have a section K = π0k (K) = lim
←−
π0k (O/mn ) → lim
←−
O/mn = O.
b
7.5 Finite Morphisms

Definition: A ring morphism A → B is finite if B is a finitely generated A-module,
B = Ab1 + . . . + Abn .
P
Finite morphisms are stable under base changes A → C, since B ⊗A C = i C(bi ⊗ 1), and
composition: If B = Ab1 + . . . + Abn , and C = Bc1 + . . . + Bcm , then
P P P
C= i Abi 1c + . . . + i Abi cm = i,j Abi cj .
Lemma: If f : M → M is an endomorphism of a finitely generated A-module, then we have a

relation f n + a1 f n−1 + . . . + an Id = 0, where a1 , . . . , an ∈ A.
Proof: Since M is a quotient of some free module An , we may assume that M = An .

If the lemma holds for f , then it holds for f ⊗ 1 for any base change A → B; hence we
may assume that A is the polynomial ring Z[xij ] and f is the endomorphism of An given by the
generic matrix (xij ), because any endomorphism is obtained from f by a base change.
Now, by Hamilton-Cayley’s theorem, the lemma holds after the base change Z[xij ] → Q(xij ),
because Q(xij ) is a field, and we conclude that it also holds in Z[xij ], because Z[xij ] → Q(xij )
is injective.
Theorem: Let A → B be a morphism, and b ∈ B. The following conditions are equivalent:
1. b satisfies an integral dependence relation bn + a1 bn−1 + . . . + an = 0, ai ∈ A.
2. A[b] is a finite subalgebra of B.
3. b is in a finite subalgebra of B.
Proof: (1 ⇒ 2) If bn + a1 bn−1 + . . . + an , then A[b] = A + Ab + . . . + Abn−1 .

b·
(3 ⇒ 1) If C ⊆ B is a finite subalgebra, and b ∈ C, then C −→ C is an endomorphism of
a finitely generated A-module, and by the lemma there exist a1 , . . . , an ∈ A such that (bn +
a1 bn−1 + . . . + an )c = 0, ∀c ∈ C. Now put c = 1.
Corollary: The integral elements of B form a subalgebra (the integral closure of A in B).
Proof: If b1 , b2 ∈ B are integral, the morphisms A → A[b1 ] → A[b1 , b2 ] are finite.

Hence any element of A[b1 , b2 ] (in particular b1 + b2 , b1 b2 and ab1 , ∀a ∈ A) is integral.
Example: The morphism A = k[x] → k[x, y]/(P ) is finite if and only if the curve P (x, y) = 0
has no vertical asymptote (if the projective curve P (x0 , x1 , x2 ) = 0 passes through the point
p = (0, 0, 1), the tangent cone at p is the line at infinity, with certain multiplicity).
In fact, y is integral when a multiple of P (hence P ) is y n + p1 (x)y n−1 + . . . + pn (x).

Homogenizing, and then dehomogenizing with x2 so that p = (0, 0, 1) is the origin, this condition
states that, if the curve passes through p, the tangent cone is xd−n
0 = 0,
0 = xd−n
0 + (terms of degree > d − n), where d = deg P.
Theorem: Any finite morphism π : Spec B −→ Spec A has finite discrete fibres, and π is sur-
jective when A → B is injective.
Proof: In fact π −1 (x) = Spec (B ⊗A κ(x)), and B ⊗A κ(x) is a finite κ(x)-algebra (p. 141).
If A → B is injective, then so is Ax → Bx , and Bx 6= 0.
Now Bx /px Bx 6= 0 by Nakayama, and π −1 (x) = Spec (Bx /px Bx ) is not void.
Corollary: If π : Spec B → Spec A is a finite morphism, then dim B ≤ dim A.
Proof: No chain of specializations y0 > y1 > . . . > yn in Spec B may have a coincidence
π(yi−1 ) = π(yi ), since the fibre of π(yi ) is discrete.
Going Up Theorem: Finite morphisms are closed maps.
Proof: If J is an ideal of B, the morphism A/I → B/J is finite and injective, where I = A ∩ J.
Now the following commutative square shows that π((J)0 ) = (I)0 ,
π
Spec B −−→ Spec A
∪ ∪
(J)0 = Spec B/J −−→ Spec A/I = (I)0
Corollary: If x0 < x = π(y), there exists y 0 < y such that π(y 0 ) = x0 .
Proof: The closure of x = π(y) is contained in π(y) since π is closed.
Corollary: If A → B is a finite injective morphism, then dim A = dim B.
Proof: Let x0 > . . . > xn be a chain of specializations in Spec A, and x0 = π(y0 ).

There are specializations y0 > . . . > yn such that π(yi ) = xi ; hence dim B ≥ dim A.
Definition: An integral ring A is normal if it is integrally closed in the field of fractions ΣA .

Unique factorization domains are normal (p. 86).
Lemma: Let A be a normal ring, let L be a normal extension of the field of fractions ΣA , and
let B ⊂ L be a finite subalgebra. If B is stable under the action of the group G = Aut(L/ΣA ),
then G acts transitively on the fibres of π : Spec B → Spec A.
Proof: Let G = {σ1 , . . . , σn }, and let y, y 0 be two points of a fibre π −1 (x).

If y 0 ∈
/ Gy, since the fibres are discrete, there exists f ∈ B vanishing at y 0 and without zeros
on the orbit Gy (p. 197). The roots of the irreducible polynomial td + c1 td−1 + . . . + cd of f over
Σ are σi (f ) ∈ B, with certain multiplicities. Since A is normal, B ∩ ΣA = A, and
mi
Q
cd = i σi (f ) ∈ B ∩ ΣA = A.
Now cd (x) = cd (y 0 ) = 0, since f (y 0 ) = 0, and cd (x) = cd (y) 6= 0, since (σi f )(y) 6= 0. Absurd.
7.5. FINITE MORPHISMS 201
Going Down Theorem: Let A → B be a finite injective morphism between integral rings. If
A is normal, then π : Spec B → Spec A is open.
Proof: Let L be the normal envelop of ΣB over ΣA , and put G = Aut(L/ΣA ).

If B = A[b1 , . . . , bn ], we put B 0 = A[σi (bj )], σi ∈ G. Since ρ : Spec B 0 → Spec B is surjective,
we have U = ρ(ρ−1 (U )), and we may assume that B is G-invariant.
Finally, π(U ) is open when so is U , because π is a closed map and by the above lemma
S
Spec A − π(U ) = π Spec B − σ(U ) .
σ∈G
Corollary: If x0 > x = π(y), there exists y 0 > y such that π(y 0 ) = x0 .
Proof: We have to show that y is in the closure F of the fibre of x0 , since it is finite.
Since x0 is not in the open set π(Spec B − F ), neither is the specialization x = π(y).
Corollary: dim By = dim Ax , when x = π(y).
Hilbert’s Nullstellensatz
Definition: The affine space of dimension n over a field k is An := Spec k[x1 , . . . , xn ]. If
A = k[ξ1 . . . , ξn ] is a finitely generated k-algebra, its spectrum X = (A, Spec A) is said to be an
affine algebraic variety over k, and recall that Homk (Spec B, Spec A) := Homk-alg (A, B).
Normalization Lemma: There is a finite injective morphism k[x1 , . . . , xd ] → A = k[ξ1 . . . , ξn ],

where d ≤ n, and d < n when ξ1 . . . , ξn are algebraically dependent.
Proof: If ξ1 , . . . , ξn satisfy an algebraic relation, write it in decreasing lexicographical order,
aξ1r1 ξ2r2 · · · ξnrn + . . . = 0.
We put ξi0 = ξi − ξ1di , where d1 d2 . . . dn−1 .

0
A = k[ξ10 , . . . , ξn−1 , ξn ], where ξ10 , . . . , ξn−1
0 , ξn satisfy a relation, with leading term in ξn ,
aξnr1 d1 +...+rn−1 dn−1 +rn ,
so that k[ξ10 , . . . , ξn−1

0 ] ,→ A is a finite morphism. By induction, there exists a finite morphism
k[x1 , . . . , xd ] ,→ k[ξ10 , . . . , ξn−1
0 ], d ≤ n − 1, and we conclude.
Corollary: dim k[x1 , . . . , xn ] = n.
Proof: By induction on n. If 0 ⊂ p1 ⊂ . . . ⊂ pm is a chain of prime ideals in k[x1 , . . . , xn ], let

P ∈ p1 be non null, so that the dimension of A = k[x1 , . . . , xn ]/(P ) is ≥ m − 1.
The proof of the lemma shows the existence of a finite injective morphism k[y1 , . . . , yd ] → A,
d < n. By induction dim A = d; hence m ≤ n and dim k[x1 , . . . , xn ] ≤ n.
We conclude since it is clear that dim k[x1 , . . . , xn ] ≥ n.
Corollary: Any affine algebraic variety X of dimension d admits a finite projection X → Ad .
Proof: If k[x1 , . . . , xd ] ,→ A is finite, then d = dim k[x1 , . . . , xd ] = dim A.
Theorem of Zeros: Finitely generated k-algebras of dimension 0 are finite k-algebras. Hence,
any residue field A/m of a finitely generated k-algebra A is a finite extension of k.
Proof: If dim A = 0, there exists a finite morphism k → A.
Corollary: Morphisms φ : Y → X of affine algebraic varieties preserve closed points.
Proof: If φ(y) = x, we have κ(x) ,→ κ(y). Now, if y is closed, κ(y) is a finite extension of k;
hence so is κ(x), and A/px ⊆ κ(x) is a field.
Corollary: The radical of A is the intersection of all maximal ideals.
Proof: If f ∈ A vanishes at any closed point of X, then the algebraic variety Uf = X − (f )0 has
no closed point; hence it is void, and f vanishes on X.
7.6 Valuations and Dedekind Rings

Definitions: An injective morphism A → B of integral rings is birrational if it induces an
∼
isomorphism between the field of fractions, ΣA − → ΣB .
If (O, m) is a local ring, an injective morphism O → B is dominant if m = m0 ∩ O for some
maximal ideal m0 of B; i.e., mB 6= B.
A domain V, of field of fractions Σ, is a valuation ring of Σ when f ∈ V or f −1 ∈ V, ∀f ∈ Σ.
The ring V = Σ is the trivial valuation. Ideals of a valuation ring V are totally ordered (in
particular V is local) because if I, J is an incomparable pair, then

f ∈ I, f ∈ /J f h
⇒ ∈/ V, ∈/ V.
h∈ / I, h ∈ J h f
Proposition: Any regular local ring (O, m) of dimension 1 is a principal ideal domain, and a
valuation ring (named discrete valuation ring).
Proof: Since m = tO, the vector spaces mn /mn+1 have dimension 1.

Hence (p. 193) if mn+1 is the first power not containing a given ideal I 6= 0, we have that I
generates mn /mn+1 and, by Nakayama’s lemma, I = mn = tn O.
Any function 0 6= f ∈ Σ is f = utn , where u ∈ O∗ , n ∈ Z; hence f ∈ O or f −1 ∈ O, and we
say that n = vx (f ) is the valuation of f at the point x ∈ Spec O defined by m. q.e.d.

1. We have vx (f g) = vx (f ) + vx (g) and vx (f + g) ≥ min vx (f ), vx (g) .
Moreover O = {f Σ : vx (f ) ≥ 0}.
A discrete valuation of afield Σ is any map v : Σ − {0} → Z such that v(f g) = v(f ) + v(g),
v(f + g) ≥ min v(f ), v(g) . When v is not trivial (i.e. v 6= 0), we may assume that it is
surjective, and O := {f ∈ Σ : v(f ) ≥ 0} is a discrete valuation ring with field of fractions
Σ and v is the corresponding valuation. Hence, the discrete valuations rings with field of
fractions Σ may be identified with the surjective discrete valuations v : Σ − {0} → Z.
2. Let A be a principal ideal domain and p an irreducible element. The local ring Ap is a discrete
valuation ring, and vp (a) = n when pn is the highest power of p dividing a. This is the p-adic
valuation.
3. A point of an algebraic variety is simple or non singular when the local ring is regular.
The local ring of a curve at a non singular point x is a discrete valuation ring, and vx (f ) is
the number of zeros of f at x (or poles, if it is negative).

4. The discrete valuation of k(t) at t = ∞ is defined to be v∞ p(t)
q(t) = deg q − deg p.
7.6. VALUATIONS AND DEDEKIND RINGS 203
Theorem: A noetherian local ring O of dimension 1 is regular if and only if it is normal.
Proof: If O is regular, it is a principal ideal domain; hence it is normal.

If O is normal, we consider f ∈ m non null, and the first power mn ⊆ f O (it exists since
m = rad f O, because dim O = 1). Take h ∈ mn−1 , h ∈ / f O, and we put x = h/f ,
xm ⊆ O, x ∈
/ O.
To conclude that m is a principal ideal it is enough to prove that xm = O.

Otherwise xm ⊆ m, and x defines an endomorphism of m, satisfying the corresponding
characteristic polynomial. Hence x is integral over O, and O is not a normal ring.
Lemma: A local integral ring V is a valuation ring if and only if any dominant birrational
morphism V → B is an isomorphism.
Proof: Let V be a valuation ring and V → B a birrational morphism.

If b ∈ B is not in V, then b−1 ∈ V is not invertible; hence b−1 ∈ mV , and mV B = B, so that
the morphism is not dominant.
Conversely, let f ∈ Σ, f ∈/ V. Since V → V[f ] is not dominant, then mV V[f ] = V [f ],
1 = a0 + a1 f + . . . + an f n ; ai ∈ mV ,
0 = (a0 − 1)f −n + a1 f n−1 + . . . + an ,
Since a0 − 1 ∈ V ∗ , the morphism V → V[f −1 ] is finite; hence dominant, and f −1 ∈ V.
Theorem: Let Σ be the field of fractions of a domain A. The integral closure B of A in a finite
extension L of Σ is the intersection of all the valuation rings Vi of L containing A.
Proof: Valuation rings are normal since finite birrational morphisms are dominant (p. 200);
hence B ⊆ Vi . On the other hand, f −1 ∈ L is invertible in A[f −1 ],
T
1 = f −1 (a0 + a1 f −1 + . . . + an f −n ),
f n = a0 f n−1 + a1 f n−2 + . . . + an ,
/ B, then f −1 is not invertible in A[f −1 ].

just when f is integral over A. Hence, if f ∈
Localizing A[f ] at a prime ideal containing f −1 , and considering a maximal ring for dom-
−1
ination (Zorn’s lemma), we obtain a valuation ring Vi not containing f .
Lemma: Let A → B be a ring morphism, and let C be the integral closure of A in B. If S is a

multiplicative system of A, then CS is the integral closure of AS in BS .
Hence, an integral ring A is normal if and only if Ax is normal at any point x ∈ Spec A.
Proof: Clearly CS = C ⊗A AS is integral over AS . Moreover, if sb ∈ BS is integral over AS , then

b n
n−1
+ as11 sb + . . . + asnn = 0, and we put t = s1 · · · sn . Multiplying by sn tn we obtain an

s
integral dependence relation of bt over A; hence bt ∈ C, and sb = st bt
∈ CS .
Finally, A is normal if and only if Ā/A = 0, and we conclude (p. 138).
Finiteness Theorem: Let A be a noetherian normal ring and L a finite separable extension of
the field of fractions ΣA . The integral closure B of A in L is a finitely generated A-module.
Proof: The trace metric is non singular (p. 145), and we have tr(β) = σ1 (β) + . . . + σd (β), where
σ1 , . . . , σd : L → L0 are the points of L with values in a trivializing extension L0 .
Hence tr(B) is integral over A, and tr(B) ⊆ B ∩ ΣA = A.

By the lemma, some b1 , . . . , bn ∈ B form a base of L.
Let b∗1 , . . . , b∗n be the dual base, tr(bi · b∗j ) = δij . Now, if b ∈ B,
b = tr(b1 b)b∗1 + . . . + tr(bn b)b∗n ∈ Ab∗1 + . . . + Ab∗n .
Hence B ⊆ Ab∗1 + . . . + Ab∗n and, A being noetherian, B is a finitely generated A-module.
Theorem: Let Σ be the field of fractions of an integral finitely generated k-algebra A. The
integral closure of A in any finite extension L of Σ is a finitely generated A-module.
Proof: We fix an injective finite morphism k[x1 , . . . , xn ] → A. The integral closure of A in L is

the integral closure of k[x1 , . . . , xn ] in L, so that we may assume that A = k[x1 , . . . , xn ].
If char k = 0, it follows from the above theorem.
If char k = p, we may assume that L is a normal extension of Σ (p. 154).
If G = Aut(L/Σ), then L is a separable extension of LG by Artin’s theorem, and LG is a
purely inseparable extension of Σ (p. 154). If we show that the integral closure B of A in LG is
a finitely generated A-module, then B is a noetherian normal ring, and the above theorem let
us conclude that the integral closure of B in L is a finitely generated B-module, hence a finitely
generated A-module.
We may assume that A = k[x1 , . . . , xn ] and L is a purely inseparable extension of Σ; hence
(p. 154) there is a power q = pr such that L = Σ(α1 , . . . , αr ), αiq ∈ Σ.
If αiq = Pi /Qi , then (Qi αi )q = Pi Qq−1 i , and we may assume that αiq ∈ A:
αiq = λi,j1 ...jn xj11 . . . xjnn ,

P
j1 ...jn
√
λi,j1 ...jn y1j1 . . . ynjn ,
P p
q
αi = yj = q xj .
j1 ...jn
p
Now L = k(x1 , . . . , xn , α1 , . . . , αr ) → K(y1 , . . . , yn ), where K = k( q λi,j1 ...jn ).
Since K[y1 , . . . , yn ] is normal, and it is finite over k[x1 , . . . , xn ] = k[y1q , . . . , ynq ], it is the
integral closure of A = k[x1 , . . . , xn ] in K(y1 , . . . , yn ), and it is a finitely generated A-module.
Definition: An integral noetherian ring A of dimension 1 is a Dedekind domain if all the

local rings Ax are regular (or normal).
In such a case, any valuation ring V of the field of fractions ΣA containing A is a discrete
valuation ring Ax : If mV ∩ A = px , then V dominates Ax , and V = Ax . Hence,
A = {f ∈ ΣA : vx (f ) ≥ 0, ∀x ∈ Spec A}.
An integral curve Spec A has no singular point when A is a Dedekind domain.

The integral closure A of Z in a finite extension L of Q is a noetherian ring by the finiteness
theorem, hence it is a Dedekind domain.
Theorem: Any non null ideal I of a Dedekind domain A uniquely decomposes as a product of
powers of prime ideals, I = pn1 1 . . . pnr r .
Proof: At any point of (I)0 = {x1 , . . . , xr } we have that Ixi coincides with some power pni i , since
Axi has no other non null ideals.
Now I = pn1 1 . . . pnr r since both ideals coincide at any point of Spec A.
Uniqueness is obvious.
7.6. VALUATIONS AND DEDEKIND RINGS 205
Modules over Dedekind Domains

Let A be a Dedekind ring. The support of a finitely generated torsion A-module T is formed
by a finite number of closed points; hence T decomposes as a direct sum of the localization at
such points and, the local rings Ax being principal ideal domains, T decomposes, uniquely up
to isomorphisms, as a direct sum of primary monogenous modules (p. 172),
n
T = ij A/pi ij .
L
If T is the torsion submodule of a finitely generated A-module M , then L = M/T is locally

free since it is torsion free (p. 171); hence it is projective (we shall prove it in p. 210), and
M = T ⊕ L since the following exact sequence splits
0 −→ T −→ M −→ L −→ 0.
If rk L = 1, localizing at the generic point we see that L is a submodule of the field of

fractions Σ; hence f L ⊆ A, and L is isomorphic to a non null ideal.
Definitions: An isomorphism I ' I 0 between non null ideals induces at the generic point an
isomorphism Σ ' Σ; hence an homothety, and we see that I ' I 0 ⇔ I 0 = f I for some f ∈ Σ.
Non null ideals I = pn1 1 . . . pnr r correspond to effective divisors D = n1 x1 +. . .+nr xr , ni ≥ 0,
and if we introduce the divisor of a rational function 0 6= f ∈ Σ,
P
D(f ) = vx (f ) x,
x∈Spec A
then I 0 ' I if and only if the corresponding divisors are linearly equivalent,
D0 = D(f ) + D for some f ∈ Σ.
The Picard group of isomorphism classes of non null ideals is Pic(A) = Div(A)/{D(f )}
since any divisor is equivalent to an effective divisor. In fact, by the Chinese theorem
A/pn1 1 . . . pnr r = (A/pn1 1 ) ⊕ . . . ⊕ (A/pnr r )
and we see that, given closed points x1 , . . . , xr and natural numbers m1 , . . . , mr , there exists a
function f ∈ A such that vxi (f ) = mi .
In general, when rk L = r > 1, we take m ∈ L non null, and we put
L0 = {m0 ∈ L : am0 ∈ Am for some a ∈ A},
so that L/L0 is a torsion free module of rank r − 1.

By induction on the rank, we see that L is a direct sum of ideals (not uniquely),
L = I1 ⊕ . . . ⊕ Ir .
Lemma: Given closed points x1 , . . . , xr ∈ Spec A, and natural numbers m1 , . . . , mr , there exists
f ∈ Σ such that vxi (f ) = −mi , and vx (f ) ≥ 0 at any other point x ∈ Spec A.
Proof: Let a ∈ A be such that vxi (a) = mi . If a−1 has other poles, we take b ∈ A vanishing at
them with equal order, and vxi (b) = 0. Now take f = b/a. q.e.d.
Given ideals I, J without common zeros, I + J = A, the exact sequence
0 −→ I ∩ J −→ I ⊕ J −→ A −→ 0
shows that I ⊕ J ' A ⊕ (I ∩ J). When J has common zeros with I, by the above lemma there
exists f ∈ Σ such that f J and I have no common zeros; hence I ⊕ J ' A ⊕ I 0 for some ideal I 0 ,
and we see that I1 ⊕ . . . ⊕ Ir ' I ⊕ Ar−1 for some ideal I.
Theorem: Any finitely generated A-module M decomposes, uniquely up to isomorphisms,

n
M = I ⊕ Ar−1 ⊕ ⊕ij A/pi ij .
Proof: We have just proved the existence.

The uniqueness follows from this fact: If L = I ⊕ Ar−1 , then I = Λr L.
Corollary: The K-group of finitely generated locally free A-modules is
K(A) = Z ⊕ Pic(A).
Proof: We have a natural morphism K(A) → Z ⊕ Pic(A), [L] 7→ (rk L, Λrk L L), the inverse being
the morphism Z ⊕ Pic(A) → K(A), (r, [I]) 7→ I ⊕ Ar−1 .
7.7 Birrational Finite Morphisms

Let Σ be the field of rational functions on an integral curve C = Spec A (i.e. the field of
fractions of an integral finitely generated k-algebra A of dimension 1).
The integral closure Ā of A in Σ is a finitely generated A-module (p. 204); hence it is a
finitely generated k-algebra, and C̄ = Spec Ā is a non singular curve (named desingularization
of C) endowed with a finite birrational morphism C̄ → C.
Definition: The finitely generated A-module C = Ā/A is the conductor, and it vanishes at a
point x just when Ax = Āx ; i.e., when x is a simple point.
Theorem: The number of singular points of any integral curve is finite.
Proof: Localizing at the generic point pg the exact sequence
0 −→ A −→ Ā −→ C −→ 0,
we obtain Cpg = Σ/Σ = 0, so that C is a finite torsion module; hence of finite support. q.e.d.
C̄ → C is an isomorphism at any simple point of C. Let us study it at a singular point x.

Let O = Ax be the local ring of C at x, and let m be the maximal ideal.
Theorem: O
b is reduced, and the minimal primes of O
b correspond to the maximal ideals of Ō.
Proof: C = Ō/O is complete, since it is annihilated by some power of m.
0 −→ O −→ Ō −→ C −→ 0
0 −→ O
b −→ O
b̄ −→ C −→ 0
Ō is a Dedekind domain; hence mŌ = mn1 1 . . . mnr r , where m1 , . . . , mr are the maximal ideals
of Ō. By the Chinese remainder theorem,
b̄ = lim Ō/mn Ō = lim(Ō/mn1 n ⊕ . . . ⊕ Ō/mnr n ) = O

O bx ⊕ . . . ⊕ O
bxr ,
←− ←− 1 r 1
7.7. BIRRATIONAL FINITE MORPHISMS 207
where Oxi , the localization of Ō at mi , is regular; hence Obx is regular, so that it is integral.
i
So we see that Ob̄ is reduced (hence so is the subring O)b and the minimal primes correspond
to the maximal ideals (corresponding to the maximal ideals of Ō).
Now, we have Cp = 0 at any minimal prime p of O, b since C is annihilated by a power of mO; b
hence Obp = Ob̄ , and Spec O
p
b̄ has a unique point over p.
Hence, the minimal primes of O b correspond to the minimal primes of O. b̄
Definitions: The minimal primes of O b are the analytical branches of the curve C at x, and
they correspond to the points xi of the fibre of C̄ over x.
The intersection multiplicity at x of C with a hypersurface H of equation f = 0 is defined
to be (C ∩ H)x = l(O/f O).
Lemma: lO (O/f O) = lO (Ō/f Ō).
f·
Proof: We have l(Ō/O) = l(f Ō/f O) since Ō/O − → f Ō/f O is an isomorphism.
The additive character of length, and the following commutative square let us conclude,
fO / f Ō

O / Ō
P
Theorem: (C ∩ H)x = vxi (f ) · [κ(xi ) : κ(x)].
xi →x
Proof: If f is invertible in O, it is obvious.

If f ∈ m, then Ō/f Ō has dimension 0, and it decomposes as a direct sum of localizations,
Ō/f Ō = Ox1 /f Ox1 ⊕ . . . ⊕ Oxn /f Oxn ,

(C ∩ H)x = lO (Ox1 /f Ox1 ) + . . . + lO (Oxn /f Oxn ).
Now, the length of the O-module Oxi /f Oxi is the product of the length vxi (f ) as an Oxi -
module, by the length [κ(xi ) : κ(x)] of the unique simple Oxi -module κ(xi ).
Quadratic Transformations
Now we assume that k is infinite, so that no vector space is a finite union of proper vector
subspaces (p. 144). If mi is the minimal value at m of the valuation vxi , xi ∈ Spec Ō, then
{a ∈ m : vxi (a) > mi } is a proper vector subspace of m; hence there exists f ∈ m of minimal
value at any valuation vxi . If m = (f1 , . . . , fd ), then vxi (fj /f ) ≥ 0 and fj /f ∈ Ō,
h i
f1 fd
O −→ O1 = O f ,..., f −→ Ō.
Definitions: The finite birrational morphism O → O1 is the quadratic transformation or

blowup of O at x. The ring O1 is semilocal (with a finite number of maximal ideals), and the
f
ideal mO1 = f O1 is principal, since fj = f fj ∈ f O1 .
This construction holds even if O is a semilocal subring of Ō, so that we may blow-up again
O1 at a point, and so on.
Lemma: O1 does not depend on the function f ∈ m of minimal valuation.

Proof: If f 0 ∈ m, and vxi (f 0 ) = vi (f ), then f 0 /f ∈ O1 does not vanish at any point of Spec Ō,
nor at any point of Spec O1 , since the morphism O1 ,→ Ō is finite.
Hence f 0 /f is invertible in O1 , and fj /f 0 = (fj /f )(f /f 0 ) ∈ O1 ,
h i h i
O ff10 , . . . , ffd0 ⊆ O1 = O ff1 , . . . , ffd .
By symmetry, we also have the reverse inclusion.
Theorem: Any curve C desingularizes with a finite number of quadratic transformations.
Proof: If O → O1 is an isomorphism, then x is simple point: m = mO1 = f O1 = f O.

Since the length of Ō/O is finite, after of a finite number of blow-ups
O −→ O1 −→ . . . −→ Or
we obtain a ring Or ⊆ Ō without singular points; hence it is normal, and Or = Ō.
Stability Lemma: mn = mn O1 , when n 0.

n1 n
f ...f d
Proof: The O-module O1 is generated by the elements f n11 +...+n
d
d
.
a
Since O1 is a finite O-module, when n 0, we may fix a generating system of the form fn ,
with a ∈ mn . Now mn O1 = f n O1 ⊆ mn , and we conclude.
Theorem: SO (n) = mn − c; where m = l(O/f O), c = l(O1 /O).
Proof: When n 0, we have mn = mn O1 = f n O1 . Hence we have an exact sequence
0 −→ O/mn −→ O1 /f n O1 −→ O1 /O −→ 0
SO (n) = l(O/mn ) = l(O1 /f n O1 ) − l(O1 /O) = l(O/f n O) − l(O1 /O) = nl(O/f O) − l(O1 /O).
Corollary: The multiplicity m is the intersection number of the blow-up of the curve with the
exceptional fibre f = 0, counting each point y with the degree [κ(y) : κ(x)],
m = (C1 ∩ E).
Proof: m = lO (O/f O) = lO (O1 /f O1 ).
Corollary: The multiplicity is 1 if and only if x is a simple point.
Proof: If 1 = m = l(O/f O), then m = f O is a principal ideal. q.e.d.
In this process, over a singular point x appear some points y, that we draw as a tree (the
end points being the branches at x) with certain multiplicity my at each node y.
Corollary: If k is algebraically closed, and C is a plane curve,

X my
dimk (Ō/O) = .
y
2
Proof: We have lO (M ) = dimk M when k = κ(x), and at a point of multiplicity m of a plane
curve (p. 197), the Samuel polynomial is n+1 n+1−m m

2 − 2 = mn − 2 .
7.7. BIRRATIONAL FINITE MORPHISMS 209
Example: Let k be algebraically closed, so that, after an axes change, we may assume that our
point is the origin. If C is a plane curve, and x = 0 is not tangent to C at the origin,
0 = P (x, y) = a0 y m + a1 xy m−1 + . . . + am xm + aij xi y j ,

P
a0 6= 0,
i+j>m
dividing by xm we obtain an integral dependence relation of xy over the local ring O of C at the
origin. Hence vi (x) ≤ vi (y) for any discrete valuation vi centered at the origin, and O1 = O xy .

We put z = xy , so that

0 = P (x, xz) = xm P1 (x, z) = xm a0 z m + a1 z m−1 + . . . + am + aij xi+j−m z j ,
P
i+j>m
and we see that O1 is the semilocal ring of the curve P1 (x, z) = 0 at intersection points with the
exceptional fibre x = 0, which are points x = 0, z = λ, where y = λx is a line of the tangent
cone a0 y m + a1 xy m−1 + . . . + am xm = 0.
For example, let C be the complex plane curve (integral by Eisenstein criterion)
P (x, y) = y 7 − x9 y − x10 y + x11 + x12 = 0.
The tangent cone is y 7 = 0 and, blowing-up with z = xy , appears the point x = z = 0,
0 = P (x, xz) =x7 (z 7 − x3 z − x4 z + x4 + x5 ),

0 = P1 (x, z) =x4 − x3 z + x5 − x4 z + z 7 .
The tangent cone is x3 (x − z) = 0, and blowing-up with s = xz ,
0 = P2 (z, s) = z 3 − s3 + s4 − zs4 + zs5 ,
appear the simple point z = 0, s = 1 ( ∂P ∂s does not vanishes), and the point z = s = 0,
2
of multiplicity 3. Blowing-up it appear 3 points, all simple since the blow-up of the curve
intersects the exceptional fibre with multiplicity 1. The blow-up tree is
•1 •1 •1
1• • 3 7
4
3

l(Ō/O) = 2 + 2 + 2 = 30
•4
•7
Theorem: The intersection multiplicity of C with a hypersurface H of multiplicity r is the
product of the multiplicities plus the intersection multiplicities of the blow-ups at the common
points of the exceptional fibre,
(C ∩ H)x = mr + (C1 ∩ H1 ).
Proof: If the equation of H is P (x1 , . . . , xn ) = 0, and f = x1 is of minimal valuation, then the

x
birrational transformation is zj = x1j , and we have
P (x1 , . . . , xn ) = xr1 P1 (x1 , z2 , . . . , zn ),
where P1 = 0 is the equation of the hypersurface H1 . Now
(C ∩ H)x = l(O/P ) = l(O1 /P ) = l(O1 /f r P1 ) = rl(O1 /f ) + l(O1 /P1 ) = mr + (C1 ∩ H1 ),
where (C1 ∩ H1 ) = l(O1 /P1 ) is just the sum of the intersection multiplicities of C1 and H1 at
the common points over x.
7.8 Faithfully Flat Morphisms

Theorem: On finitely generated modules over a local ring, the conditions of being free, projective
and flat are equivalent.
Proof: Always free ⇒ projective ⇒ flat, so we must prove flat ⇒ free.

Let {m1 , . . . , mn } be a minimal generating system of a finite flat O-module M .
If f1 m1 + . . . + fn mn = 0, we put I = (f1 , . . . , fn ).
By flatness I ⊗O M → M is injective, and f1 ⊗ m1 + . . . + fr ⊗ mr = 0.
The base change O → k = O/m shows that f¯1 ⊗ m̄1 + . . . + f¯r ⊗ m̄r = 0 in the vector space
(I/mI) ⊗k (M/mM ) = (I/mI) ⊗k (k m̄1 ⊕ . . . ⊕ k m̄n ).
Hence f¯i = 0, and I/mI = 0. By Nakayama I = 0, and m1 , . . . , mn is a base of M .
Lemma: HomA (M, N )x = HomAx (Mx , Nx ), for all x ∈ Spec A, when M is an A-module of
finite presentation (it admits an exact sequence Am → An → M → 0).
Proof: Just consider the following commutative diagram with exact rows
HomA (Am , N )x ←− HomA (An , N )x ←− HomA (M, N )x ←− 0

|| || ↓
m n
HomAx (Ax , Nx ) ←− HomAx (Ax , Nx ) ←− HomAx (Mx , Nx ) ←− 0
Theorem: On finitely presented A-modules, the conditions of being locally free, projective and
flat are equivalent.
Proof: Since flat modules are locally free by the above theorem, we only have to show that any
locally free module P is projective.
Now, if E is an exact sequence of A-modules, so is HomA (P, E)x = HomAx (Px , Ex ), since Px
is a free Ax -module; hence HomA (P, E) is exact (p. 138).
Theorem: If P is a projective finitely generated A-module, any point x ∈ Spec A has a basic
neighborhood U such that PU is a free AU -module.
Proof: Since Px is free, we have a commutative diagram with exact rows
0 / Ker /L /P / Coker /0

0 / (Ker)x / Lx ∼
/ Px / (Coker)x /0
where L is free. Since Coker is finitely generated and Cokerx = 0, in a neighborhood U we have
CokerU = 0 (the support of any element is closed, p. 138).
Since PU projective, LU = KerU ⊕ PU , and KerU is finitely generated.
Since Kerx = 0, we may fix U so that KerU = 0, and PU = LU is free.
Theorem: If P is a flat A-module, the following conditions are equivalent,
1. P is faithfully flat: any sequence of A-modules E is exact ⇔ E ⊗A P is exact.
2. P/mP 6= 0 for any maximal ideal m of A.
3. M ⊗A P = 0 ⇔ M = 0.
7.9. GALOIS THEORY OF RINGS 211
Proof: (1 ⇒ 2) 0 → A/m → 0 is not exact, hence 0 → P/mP → 0 is not exact.

(2 ⇒ 3) If m ∈ M is non null, and we take an epimorphism Am → A/m → 0, the epimor-
phism (Am) ⊗A P → (A/m) ⊗A P → 0 shows that (Am) ⊗A P 6= 0.
Since P is flat, (Am) ⊗A P is a submodule of M ⊗A P ; hence M ⊗A P 6= 0.
i p i⊗1 p⊗1
(3 ⇒ 1) Let us consider M 0 → − M→ − M 00 and M 0 ⊗A P −−→ M ⊗A P −−→ M 00 ⊗A P.
Since P is flat, (Im i) ⊗A P = Im (i ⊗ 1), (Ker p) ⊗A P = Ker (p ⊗ 1), and

Im i + Ker p Im (i ⊗ 1) + Ker (p ⊗ 1)
⊗A P =
Im i Im (i ⊗ 1)

Im i + Ker p Im (i ⊗ 1) + Ker (p ⊗ 1)
⊗A P =
Ker p Ker (p ⊗ 1)
hence Im i = Ker p if and only if Im (i ⊗ 1) = Ker (p ⊗ 1).
Corollary: Let A → B be a faithfully flat morphism. An A-module M is finitely generated

(resp. flat) if and only if the B-module MB is finitely generated (resp. flat).
Proof: If MB is finite, it is generated by some elements of M , defining a morphism An → M

such that B n → MB is surjective; hence so is An → M , and M is finitely generated.
If E is an exact sequence of A-modules, then EB is exact; and if MB is a flat B-module,
then EB ⊗B MB = (E ⊗A M )B is exact. Hence E ⊗A M is exact, and M is flat.
Corollary: A flat morphism A → B is faithfully flat if and only if φ : Spec B → Spec A is

surjective.
Proof: If A → B is faithfully flat, then the fibres of φ are non void by the fibre formula, since
B ⊗A κ(x) 6= 0. Conversely, if the fibres are non void and x = φ(y), then the flat morphism
Ax → By is faithfully flat by the theorem; hence so is A → B.
Theorem: If A → B is faithfully flat, then the sequence A → B ⇒ B ⊗A B is exact.
Proof: The argument of p. 189 holds.
7.9 Galois Theory of Rings

Let A be a noetherian ring such that S = Spec A is connected.
Definitions: A covering of S is an unramified finite flat morphism X = Spec B → S (i.e.
B is a finite flat A-module and ΩB/A = 0). Then the degree of the κ(x)-algebra B ⊗A κ(x) is
locally constant; hence constant if S is connected, and it is the degree of the covering.
The trivial covering of degree n is S⊕ . n. . ⊕S := Spec An → S.
Any covering of degree 1 is an isomorphism: it is the trivial covering of degree 1.
Theorem: The concept of covering is stable under base changes (if A → B is a covering, so is
C → B ⊗A C), and it is local with respect to faithfully flat morphisms (if A → C is faithfully
flat, then A → B is a covering if and only if so is C → B ⊗A C).
Proof: Finite flat morphisms, and the module of differentials, are stable under base changes.
Lemma: Any connected component of a covering also is a covering.

Proof: If X = X 0 q X 00 , then B = B 0 ⊕ B 00 (p. 190). Since B is finite and flat, so is B 0 , and it

is unramified since 0 = ΩB/A = ΩB 0 /A ⊕ ΩB 00 /A .
Lemma: Any section of a connected covering X → S is an isomorphism.

σ
Proof: Any section 0 −→ I −→ B −−→ A −→ 0 defines a derivation
D : B −→ I/I 2 , D(b) = b − σ(b) (modulo I 2 ).
Since ΩB/A = 0, we have I = I 2 ; and Ix = 0 or Ix = Bx at any point x ∈ X.

Hence I vanishes on a closed open set. Since X is connected, I = 0. q.e.d.
When X is connected, for any covering Y → S we obtain the Points Formula:

Connected components of X ×S Y
HomS (X, Y ) = HomX (X, X ×S Y ) =
isomorphic to X via the first projection
Hence the degree of Y bounds the number of S-morphisms X → Y , and both are equal if
and only if Y is trivial over X; i.e., X ×S Y = qX.
1×f 2 π
Any morphism f : X → Y also is a covering, since it is X −−→ X ×S Y −→ Y , where 1 × f
is an isomorphism onto a connected component, and π2 is a covering.
In particular, any morphism X → X is an automorphism, since the degree is 1. Hence,

Connected components of X ×S X
Aut(B/A) := AutA-alg (B) =
Definition: A connected covering X = Spec B → S is a Galois covering when X ×S X → X

is a trivial covering, B ⊗A B = ⊕G B,
X ×S X = Xq .G. . qX =: X × G,
where G = HomA-alg (B, B) = AutA-alg (B) is the Galois group. Since A → B is faithfully flat,
the exact sequence A → B ⇒ B ⊗A B = ⊕G B shows that A = B G ,
X/G := Spec B G = S.
Moreover, when a finite group G acts on an A-algebra B, the argument of page 146 shows
that B G ⊗A C = (B ⊗A C)G when the base change Y = Spec C → S is flat; i.e. (X/G) ×S Y =
(X ×S Y )/G. Therefore, when a covering Y = Spec C → S is trivial over a Galois covering
X → S of group G, it is fully determined by the finite G-set
F (Y ) = HomA-alg (C, B) = HomS (X, Y ) = [Connected components of X ×S Y ]
because X ×S Y = Xq F.(Y
. .) qX =: X × F (Y ) and we have
Y = (X/G) ×S Y = (X ×S Y )/G = (X × F (Y ))/G.
Hence, given a finite G-set ∆, we define the associated covering to be
R(∆) = (X × ∆)/G := (q∆ X)/G := Spec (⊕∆ B)G ,
which is a covering of S of degree |∆| trivial over X. In fact we have
X ×S R(∆) = X ×S (X × ∆)/G = ((X ×S X) × ∆)/G = (X × G × ∆)/G,

7.9. GALOIS THEORY OF RINGS 213
where G acts via X-morphisms, so that (X × G × ∆)/G = X × (G × ∆)/G = X × ∆.

Moreover, the points formula shows that ∆ = HomS (X, R(∆)) = F R(∆) and we obtain the
following result:
Galois Theorem: If X = Spec B → S = Spec A is a Galois covering of group G = Aut(B/A),

then the functors F and R define an equivalence of categories

Coverings of S Finite R ◦ F = Id
! ,
trivial over X G-sets F ◦ R = Id
Artin’s Theorem: If a connected covering Spec B → S admits a group of automorphisms G

such that A = B G , then it is a Galois covering of group G = Aut(B/A).
Proof: X ×S X has a connected component ' X for each element of G, and
X = X ×S (X/G) = (X ×S X)/G = (Xq .G. . qX)/G ⊕ (X1 q . . . q Xr )/G
is connected. The components Xi do not exist, and X ×S X = Xq .G. . qX.
Theorem: Any covering Y → S is trivial over some Galois covering X → S.
Proof: If X → S is a connected covering and X ×S Y = X q . . . q X q X 0 q . . . is not trivial,

then
X 0 ×S Y = X 0 ×X (X ×S Y ) = X 0 q . . . q X 0 q (X 0 ×X X 0 ) q . . .
and X 0 ×X X 0 = X 0 q . . . by the points formula; hence X 0 ×S Y has more trivial components
than X ×S Y . Starting with X = S, finally we obtain a connected component X of a product
Y ×S . . . ×S Y such that X ×S Y = qX is trivial.
Now, since X ×S X is a closed open subspace of (Y ×S . . . ×S Y ) ×S X = ⊕X, we conclude
that X is a Galois covering of S.
7.9.1 The Fundamental Group

Definition: Let s : Spec k̄ → S be a geometric point, where k̄ is an algebraically closed
field (for example an algebraic closure of a residue field). The geometric fibre of a covering
Y = Spec C → S over s is
HomS (Spec k̄, Y ) = HomSpec k̄ (Spec k̄, Spec k̄ ×S Y ) = Spec (C ⊗A k̄),
because C ⊗A k̄ = ⊕k̄ (it is a separable k̄-algebra), and it has as many points as the degree of
Y over S.
When X → S is a connected covering, if two S-morphisms X ⇒ Y coincide at point of the
geometric fibre, they coincide, since any two connected components of X ×S Y are disjoint.
Pairs Xx , where X = Spec B → S is a Galois covering and x : Spec k̄ → X is a geometric
point over s, form a filtering projective system with the morphisms
φji : (Xj = Spec Bj )xj −→ (Xi = Spec Bi )xi , φji (xj ) = xi ,
because, if Xx and X̄x̄ are Galois coverings, we take the connected component X 0 of X ×S X̄
passing through the geometric point x0 = (x, x̄), so that we have morphisms Xx0 0 → Xx , Xx0 0 →
X̄x̄ , and X 0 is a Galois covering. In fact, X and X̄ are trivial over X 0 ; hence so is X ×S X̄ and
its connected component X 0 ; i.e., X 0 ×S X 0 = qX 0 .
Definition: Put Gi = Aut(Bi /A). Any morphism φji : (Xj )xj → (Xi )xi induces an epimorphism
Gj → Gi , and the fundamental group of S at the geometric point s is
π1 (S, s) = lim
←−
Gi .
Any covering Y → S is trivial over some Galois covering (Xi )xi , and xi defines a canonical
bijection of the fibre F (Y ) = HomS (Spec k̄, Y ) of Y over s with the finite Gi -set, hence finite
π1 (S, s)-set, HomS (Xi , Y ) classifying Y . But so we only obtain finite π1 (S, s)-sets where the
action factors through some quotient π1 (S, s) → Gi .
To characterize these actions, we consider π1 (S, s) as a projective limit of discrete groups.
Any open subgroup U ⊂ π1 (S, s) contains the kernel Ki of a projection π1 (S, s) → Gi , so that
any continuous action π1 (S, s) × ∆ → ∆ on a finite discrete space is induced T by some action
Gi × ∆ → ∆. In fact, the isotropy subgroups Ix , x ∈ ∆, being open, so is x Ix ⊇ Ki , and the
action factors through Gi . Now, according to the Galois theorem,
Galois Theorem: The fibre functor F (Y ) = HomS (Spec k̄, Y ) defines an equivalence of the
category of coverings of S with the category of finite π1 (S, s)-sets (with continuous action).
Definition: Let G be a finite group. A G-principal covering is a covering X → S (not

∼
necessarily connected) with an action of G defining an isomorphism X × G := qG X − → X ×S X,
so that it is a Galois covering of group G when it is connected.
Isomorphisms of G-principal coverings are S-isomorphisms commuting with the action of G.
If we fix a geometric point of the fibre of s, we say that it as a pointed principal covering.
Now, for any finite (discrete) group G the argument of p. 244 gives

Pointed G-principal
Corollary: HomTopGr (π1 (S, s), G) = .
coverings of S
Finally, we show that the fundamental group is a functor:

Let φ : S 0 → S be a morphism (S 0 = Spec A0 connected), s0 : Spec k̄ → S 0 a geometric point,
and s = φ(s0 ) : Spec k̄ → S. Given a pointed Galois covering (Xi )xi of S, then Xi ×S S 0 is a
Gi -principal covering of S 0 , pointed with the geometric point x0 = (xi , s0 ). Hence it defines a
continuous group morphism π1 (S 0 , s0 ) → Gi , and we obtain a continuous group morphism
φ∗ : π1 (S 0 , s0 ) → lim
←−
Gi = π1 (S, s).
Example: In the case of a field k, coverings Spec A → Spec k are separable finite k-algebras A,
Galois coverings are Galois extensions, and to fix a geometric point s : Spec k̄ → Spec k is just to
fix an algebraically closed extension k → k̄. Any Galois extension k → L admits an embedding
L → k̄, so that sep = {α ∈ k̄ : α is algebraic and separable over k} is
sep
S the separable closure k̄
just k̄ = i Li , where Li runs over all Galois extensions of k. Since Li is invariant under any
k-automorphism of k̄ sep , the fundamental group of Spec k is just the absolute Galois group
π1 (Spec k, s) = lim
←−
Aut(Li /k) = Aut(k̄ sep /k).
When k = Fq is the finite field with q elements, then (p. 148) the fundamental group is
Q b
π1 (Spec Fq ) = lim
←−
Z/nZ = p Zp ,
and the Frobënius automorphism F (α) = αq generates a dense subgroup.

Now, any point p : Spec Fq → S defines a morphism π1 (Spec Fq , s) → π1 (U, s), where U is
any open neighborhood of p in S, and the image of F is the Frobënius automorphism Fp at
the point p, so generalizing the definition of p. 149.
Chapter 8
Topology
8.1 Lattice Semirings

Definition: A set endowed with two operations (A, +, ·) is a commutative semiring when
(A, +) and (A, ·) are abelian semigroups (associative and commutative operations with neutral
a·
element) and the maps A −−→ A are morphisms of semigroups,
a(b + c) = ab + ac , a · 0 = 0.
and a morphism of semirings is a map f : A → A0 such that
f (a + b) = f (a) + f (b) , f (0) = 0

f (a · b) = f (a) · f (b) , f (1) = 1
A semiring A is a lattice semiring if moreover a2 = a, 1 + a = 1; ∀a ∈ A.

In these notes all semirings will be assumed to be lattice semirings.
Theorem: If (A, +, ·) is a (lattice) semiring, then A∗ = (A, ·, +) is also a (lattice) semiring,

the dual semiring of A.
Proof: a + a = a(1 + 1) = a · 1 = a.
(a + b)(a + c) = a + a(b + c) + bc = a(1 + b + c) + bc = a + bc.
Examples: The closed sets of a topological space X form a semiring AX = A(X),
a+b=a∩b , 0=X
ab = a ∪ b , 1=∅
and any continuous map φ : Y → X induces a morphism φ−1 : A(X) → A(Y ).

A base of the topology of X is a family B ⊆ AX stable under finite unions and intersections
(in particular 0, 1 ∈ B, and B is a semiring) and such that any closed set in X is an intersection
of closed sets in the base (the complements form a base of open sets in the usual sense)
In general, any lattice of closed sets (or just subsets) of X is a semiring.
The dual semiring A∗X is canonically isomorphic to the lattice of open sets in X.
K = {0, 1}, with the operations 0 + a = a, 0 · a = 0, 1 · a = a and 1 + a = 1 is the unique
(lattice) semifield. In fact, if ab = 1, then a = a(ab) = a2 b = ab = 1.
Definition: If A is a semiring, I ⊆ A is an ideal when 0 ∈ I, I + I ⊆ I and A · I ⊆ I.
215
216 CHAPTER 8. TOPOLOGY
Any ideal I defines an equivalence relation on A,
a ≡ b (mod. I) ⇔ a + x = b + x, for some x ∈ I
and the quotient set A/I is a semiring with the operations
[a] + [c] = [a + c] , [a] · [c] = [ac].
It has the usual universal property (p. 46); but the isomorphism theorem fails.
For example, AY = AX /IY when IY is the ideal of all closed sets containing a closed set Y .
However, if i : Z → X is a dense subspace, then i∗ : A(X) → A(Z) is surjective with null kernel,
but it may be not injective.
Definition: A multiplicative system S of A (i.e., 1 ∈ S and s, t ∈ S ⇒ st ∈ S) defines an

equivalence relation on A × S,
(a, s) ≡ (b, t) ⇔ atu = bsu, for some u ∈ S.
The quotient set AS is the localization of A at S, and it is a semiring with the usual
universal property (p. 85) when endowed with the operations
a b at + bs a b ab
+ = , · = ·
s t st s t st
For example, if p is the prime ideal of all closed sets containing a given point x ∈ X, then
A(X)p is the semiring of germs at x of closed sets.
Definitions: Integral semirings, prime and maximal ideals, and (Krull) dimension are defined
as in the case of rings. The spectrum of a semiring A is the set Spec A of all prime ideals, with
the Zariski topology (closed sets are zeros (I)0 = {p ∈ Spec A : I ⊆ p} of ideals I ⊆ A),
(A)0 = ∅ , (0)0 = Spec A

P T
i Ii 0 = i (Ii )0
(I ∩ J)0 = (I)0 ∪ (J)0
and any morphism f : A → B induces a continuous map
f ∗ : Spec B −→ Spec A , f ∗ (p) = p ∩ A := {a ∈ A : f (a) ∈ p}.
Basic closed sets are (a)0 , a ∈ A, and basic open sets are Ua = Spec A − (a)0 .
Many properties of rings, and their proofs, hold in the case of (lattice) semirings:
1. Maximal ideals are prime ideals (p. 69). The subspace of all maximal ideals is the maximal
spectrum Specm A ⊆ Spec A.
2. Ideals of A/I correspond to ideals of A containing I (p. 69); hence Spec (A/I) = (I)0 .
3. Any ideal I 6= A is contained in a maximal ideal (p. 120).
4. The closure of a point x ∈ Spec A is (px )0 . Hence Spec A is a T0 space and Specm A is formed
by the closed points of Spec A (p. 135).
5. Spec A is a T0 compact space (p. 135).

8.1. LATTICE SEMIRINGS 217
6. Prime ideals of AS correspond to prime ideals of A not intersecting S, hence prime ideals of
Ap correspond to prime ideals contained in p (p. 136).
7. Spec (A ⊕ B) = (Spec A) q (Spec B), (p. 136).
8. Irreducible closed sets in Spec A correspond to prime ideals of A, and the dimension of A
is the supremum of the lengths of all chains of irreducible closed sets (or specializations
x0 > x1 . . . > xn ) in Spec A (p. 196).
but (lattice) semirings have some additional nice properties:
1. The unity is the unique invertible element. By duality, a + b = 0 ⇒ a = b = 0.

If ab = 1, then a = a · ab = a2 b = ab = 1.
2. Let I be an ideal. If a + b ∈ I, then a, b ∈ I.
3. Any finitely generated ideal is principal.

aA + bA = (a + b)A, since a = a(1 + b) = a2 + ab = a(a + b).
4. Any semiring is a union of finite semirings.

The subring generated by any finite family is finite since a2 = a and a + a = a.
5. Any nilpotent element is null.

If an = 0, then a = 0 because an = a.
6. The canonical localization morphism A → AS always is surjective.

a a
s = 1 , since as = as2 .
with surprising and agreeable consequences:
1. Principal ideals have a unique generator: aA = bA ⇒ a = b.
2. Any element a 6= 1 belongs to some maximal ideal.
3. Basic closed sets (hence basic open sets) are stable under finite unions and intersections,
(a + b)0 = (a)0 ∩ (b)0 , (ab)0 = (a)0 ∪ (b)0
4. An ideal I is maximal if and only if A/I = K = {0, 1}.
5. Any ideal is an intersection of prime ideals.
6. A prime ideal p is minimal if and only if Ap = K.
7. If (I)0 = Spec A, then the ideal I is the null, I = 0.
8. Any closed open set in Spec A is a basic closed set, (p. 190).
If Spec A = (I)0 ∪ (J)0 = (I ∩ J)0 and ∅ = (I)0 ∩ (J)0 = (I + J)0 , then I + J = A and
I ∩J = 0, so that we have a+b = 1, ab = 0 for some a ∈ I, b ∈ J, and we see that (I)0 = (a)0 .
9. Prime ideals of A correspond to prime ideals of A∗ , p ↔ p∗ = A − p.

Spectral Representation Theorem: Any semiring A is canonically isomorphic to the semir-

ing of basic closed sets in the spectrum Spec A.
Proof: If (a)0 = (b)0 , then aA = bA, since any ideal is an intersection of prime ideals, and a = b
since the generator of a principal ideal is unique.
Note: Now it is clear that our semirings are lattices with the order a ≤ b ⇔ a + b = a. The
first element is 1 and the last one is 0, while min(a, b) = a + b and max(a, b) = ab.
Theorem: The functor Spec is representable, Spec A = Hom(A, K).
Proof: The kernel of a morphism A → K always is a prime ideal, and each prime ideal p is the
kernel of the morphism f : A → K, f (p) = 0, f (A − p) = 1.
Corollary: Spec (lim

−→
Ai ) = lim
←−
(Spec Ai ), and dim (lim
−→
Ai ) ≤ sup{dim Ai }.
Proof: Any representable functor transforms inductive limits1 into projective limits.
If p is a prime ideal of A, we put pi = Ai ∩ p. If p 6= q, there exists an index i such that
pi 6= qi ; hence, if p0 ⊂ p1 . . . ⊂ pn is a chain of primes in A, there exists an index i such that
(p0 )i ⊂ (p1 )i . . . ⊂ (pn )i , and n ≤ dim Ai .
Corollary: Any spectral topological space (i.e. the spectrum of a semiring) is a projective limit
of finite T0 topological spaces. (The converse also holds, see p. 226).
Proof: Any semiring A is a union of finite semirings, A = lim

−→
Ai ; hence Spec A = lim
←−
(Spec Ai ).
Q
Tensor Product: Let {Ai } be a family of semirings, Xi = Spec Ai , X = i Xi .
By the spectral representation theorem, Ai is a family of closed sets in Xi , hence in X by
the inverse image of the projection X → Xi .
N
The lattice of closed sets in X
N generated by these families is denoted by i Ai , since we
have canonical morphisms Ai → i Ai with the universal property
N Q
Hom i Ai , B = i Hom(Ai , B).
In fact, given morphisms fi : Ai → B, the continuous

N maps fi∗ : Y = Spec B → Xi define a
continuous map φ : Y → X, and a morphism f : i Ai ⊆ A(X) → A(Y ).
By theN spectral representation theorem B ⊆ A(Y ), and by construction f (Ai ) = fi (Ai ) ⊆ B;
hence f : i Ai → B induces on each semiring Ai the morphism fi .
N Q N Q
Corollary: Spec i Ai = i (Spec Ai ), Specm i Ai = i (Specm Ai ).
Proof: The firstQformula holds because Spec is a representable functor, and the second one
because (xi ) ∈ i Xi is closed if and only if so is any point xi .
Corollary: dim A ⊗ B = dim A + dim B.
Proof: Put X = Spec A, Y = Spec B. In X × Y we have a specialization (x, y) ≥ (x0 , y 0 ) if and

only if x ≥ x0 and y ≥ y 0 . Now it is clear that the supremum of the lengths of all chains of
specializations in X × Y is dim A + dim B.
1
The inductive limit of semirings is just the inductive limit as sets, with the obvious operations. If Ai is a base
of Xi , in general lim
−→
Ai is not a base of lim
←−
Xi , since it may be void even if the spaces Xi are non void.
8.2. COMPACT SPACES 219
Theorem: A semiring A has dimension 0 (any prime ideal is maximal) if and only if A is a
Boolean algebra (for any a ∈ A there exists a0 ∈ A such that a0 + a = 1, a0 a = 0).
Proof: Assume that dim A = 0. If a = 1, we take a0 = 0. If a 6= 1, and a ∈ p, since any prime is

minimal, Ap = {0, 1}, and a is annihilated by some element of A − p.
That is to say, no prime ideal contains Ann (a) + aA; hence Ann (a) + aA = A.
There exists a0 ∈ Ann (a) such that 1 = a0 + ab for some b ∈ A, so that
a0 + a = a0 + a(1 + b) = a + a0 + ab = a + 1 = 1.
Now let us prove the converse. If p is a prime ideal and a ∈ p, then a0 ∈

/ p since a0 + a = 1,
a 0
and s = 0 in Ap since a is annihilated by a ∈ A − p.
Then Ap = {0, 1}, and p is a minimal prime. Since any prime is minimal, dim A = 0.
Corollary: In any Boolean algebra, the complement a0 is unique, and it defines an isomor-
phism onto the dual semiring; (a + b)0 = a0 b0 , (ab)0 = a0 + b0 , 00 = 1, 10 = 0.
Proof: The equalities a0 + a = 1, a0 a = 0, show that in the spectral representation a0 is the

complement of a, and now all is obvious.
Corollary: Any Boolean algebra is a union of finite Boolean algebras, and any finite Boolean
algebra is isomorphic to the semiring of all subsets of a finite set.
Proof: If A is a Boolean algebra, any finite family a1 , . . . , an is contained in the semiring generated
by a1 , . . . , an , a01 , . . . , a0n , which is a finite Boolean algebra.
If moreover A is finite, then A = A(Spec A), and closed sets in Spec A are open.
Since X = Spec A is T0 and any point is closed, the topology is discrete.
Hence A coincides with the semiring of all subsets of the finite set X.
Stone’s Representation Theorem: The functor Spec defines an antiequivalence of the cat-
egory of Boolean algebras with the category of profinite spaces (i.e. projective limits of finite
discrete spaces) the inverse functor being X {semiring of closed open sets in X}.
Proof: Any Boolean algebra A is isomorphic to the semiring of closed open sets in Spec A,
because such closed sets are basic.
Moreover, any profinite space is X = lim
←−
Xi = lim
←−
(Spec Ai ) = Spec (lim
−→
Ai ) = Spec A, where
A = lim
−→
Ai is a Boolean algebra, so that A is just the semiring of closed open sets in Spec A = X.
8.2 Compact Spaces

Let B be a base of closed sets of a T0 topological space X.
Any point x ∈ X defines a prime ideal px = {b ∈ B : x ∈ b} of B. This map
X −→ Spec B, x 7→ px ,
is injective because X is T0 , it is continuous since b = (b)0 ∩ X, and it is a homeomorphism onto

the image (B is a base), which is dense: X ⊆ (b)0 ⇒ b = 0 ⇒ (b)0 = Spec B.
Theorem: A T0 topological space X is compact if and only if Specm B ⊆ X.

Proof: If X is compact, and m is a maximal ideal of B, then the intersection of any finite family
in m is non void since 1 ∈/ m; hence the intersection of all closed sets in m is non void, m ⊆ px
for some x ∈ X, and m = px since m is maximal.
Conversely, if a closed set (I)0 of Spec B does not intersect Specm B, it is void since any
ideal of B is contained in some maximal.
Since Spec B is compact, any subspace containing Specm B so is compact.
Corollary: X is compact and T1 if and only if Specm B = X.
Proof: If X is compact and T1 , we have to show that prime ideals px are maximal.
If x is not in a closed set c ∈ B, by compactness x is in a closed set b ∈ B not intersecting
c; i.e., b ∈ px and b + c = 1. Hence px + cB = B, and px is maximal.
Q
Tychonoff ’s Theorem: Any direct product i Xi of compact spaces is compact.
N Q
Proof: Put Ai = A(Xi ). By definition A = i Ai is a base of closed sets in i Xi .
Now, Specm Ai ⊆ Xi when Xi is compact, and we conclude since (p. 218)
Q Q
Specm A = i Specm Ai ⊆ i Xi ⊆ Spec A.
Corollary: Any projective limit of non void compact separated spaces is a non void compact
separated space. In is a compact separated space.
T Q
Proof: The limit of a projective system {Xi , φji } is the subspace i,j Yji of i Xi , where
Q
Yji = {(xi ) ∈ i Xi : xi = φji (xj )},
and all finite intersections are non void since the set
T of indices is filtering.
Now, Yji is closed when Xi is separated; hence i,j Yji is a non void closed set.
Corollary: Any profinite space is a compact separated space.
Lemma: Let φ : X → Y be a closed continuous map. The morphism φ−1 : AY /(AY ∩I) → AX /I
is injective for any ideal I of AX .
Proof: Let a, b ∈ AY . If φ−1 (a) + c = φ−1 (b) + c, c ∈ I, then φ(c) ∈ AY , and
a + φ(c) = a ∩ φ(c) = φ(φ−1 (a) ∩ c) = φ(φ−1 (b) ∩ c) = b + φ(c),
where φ(c) ∈ AY ∩ I, because φ−1 (φ(c)) ∈ I since it contains c ∈ I.
Proposition: Let {Xi , φji } be a projective system. If the maps φji are closed, then
Specm (lim
−→
Ai ) = lim
←−
(Specm Ai ) , Ai = A(Xi ).
Proof: We have Spec (lim

−→
Ai ) = lim
←−
Spec Ai ; p = (pi )i∈I , where pi = Ai ∩ p.
By the lemma, the morphisms φ−1
ji : Ai /pi → Aj /pj are injective, so that
S
A/p = lim
−→
(Ai /pi ) = i Ai /pi
and now it is clear that A/p = K if and only if A/pi = K for any index i.
Theorem: Let {Xi , φji } be a projective system of non void compact T0 spaces. If the maps φji
are closed, then the projective limit is a non void compact T0 space.
8.2. COMPACT SPACES 221
Proof: Put Ai = A(Xi ). We have A = lim

−→
Ai 6= 0 since the spaces Xi are non void and
Specm A = lim
←−
(Specm Ai ) ⊆ lim
←−
Xi ⊆ Spec A.
Definitions: If X is a topological space, and Y a metric space, then HomSets (X, Y )c is the set
of maps f : X → Y with the compact convergence topology, defined by the ”pseudometrics2 ”

dK (f, h) = sup d f (x), h(x) ∈ [0, ∞]
x∈K
where K runs over the compact sets in X, and C(X, Y )c is the set of continuous maps, with the
induced topology. A base of neighborhoods of a map f are the balls BdK (f, ε).
When we consider X with the discrete topology, the compact sets are the finite subsets,
Q and
we obtain the weak or pointwise convergence topology on HomSets (X, Y )w = X Y : it is
just the product topology.
When we consider X with the trivial topology, all the subsets are compact, and we obtain
HomSets (X, Y )u , the set of all maps with the topology of uniform convergence on X.
A family of continuous maps F ⊆ HomSets (X, Y ) is equicontinuous when, given ε > 0,
any point p ∈ X has a neighborhood Up in X such that d f (x), f (p) < ε, ∀x ∈ Up , f ∈ F (in
particular, any map f ∈ F is continuous).
Then the closure F̄ in HomSets (X, Y )w also is equicontinuous: given ε > 0, for any function
¯
f ∈ F̄ and any point x ∈ Up , there exists f ∈ F such that d f¯(x), f (x) and d f¯(p), f (p) are

< ε; since d f (x), f (p) < ε, we conclude that d f¯(x), f¯(p) < 3ε.
Lemma: The compact and pointwise convergence induce the same topology on any equicontin-
uous family F.
Proof: Given f ∈ F ⊆ C(X, Y ), ε > 0 and a compact set K ⊆ X, there is a finite open cover
K ⊆ U1 ∪ . . . ∪ Un and points xi ∈ Ui such that d f (x), f (xi ) < ε, ∀x ∈ Ui .
Now, if x ∈ K, then x ∈ Ui for some index i, so that

d f (x), h(x) ≤ d f (x), f (xi ) + d f (xi ), h(xi ) + d h(xi ), h(x) < 2ε + d f (xi ), h(xi )
for every function h ∈ F. Hence Bdx1 ,...,xn (f, ε) ∩ F ⊆ BdK (f, 3ε) ∩ F and we conclude.
Ascoli’s Theorem: Let F be an equicontinuous family of maps from a topological space X to

a metric space Y . If F(x) = {f (x); f ∈ F} has compact closure in Y for any point x ∈ X, then
F has compact closure in C(X, Y ).
Q
Proof: The closure F̄ in HomSets (X, Y )w = X Y is compact because it is a closed subspace of
Q
the compact space x F(x).
Now, F̄ is an equicontinuous family; hence F ⊆ F̄ ⊆ C(X, Y ) and the topology of the
compact space F̄ is induced by the topology of C(X, Y )c .
8.2.1 Proper Morphisms

Theorem: A topological space K is compact if and only if π2 : K × T → T is a closed map for
every topological space T .
Proof: If C ⊆ K × T is closed, and t ∈ T − π2 (C), then any point (x, t) ∈ K × t has an open
neighborhood Ux × Vx not intersecting C. When K is compact, the fibre K × t admits a finite
2
In fact dK is not a true pseudometric, because it may take the value ∞, but however it defines a topology in
HomSets (X, Y ) in the usual way.
T
open cover K ×t ⊆ (U1 ×V1 )∪. . .∪(Un ×Vn ), so that i Vi is a neighborhood of t not intersecting
π2 (C), and we see that π2 (C) is closed.
Conversely, put T = Spec AK , with the topology generated by the sets (a)0 (the open sets
are arbitrary unions of sets (a)0 , it is just the spectrum of the dual lattice).
The map φ : K → Spec AK , φ(x) = px , is not continuous, but still has dense image.
Let us consider the graphic Γφ ⊆ K × T . If π2 : K × T → T is closed, then π2 (Γ̄φ ) is a closed
set containing π2 (Γφ ) = Im φ; hence π2 (Γ̄φ ) = T . For any prime ideal p ∈ T = Spec AK there is
a point x ∈ X such that (x, p) ∈ Γ̄φ , so that for any neighborhood U of x, and any closed set
a ∈ p we have (U × (a)0 ) ∩ Γφ 6= ∅; i.e. U ∩ a 6= ∅. Since a is closed, x ∈ a; hence p ⊆ px .
When p is maximal, we conclude that p = px , so that K is compact.
Definition: A continuous map f : X → S is proper if it is universally closed; i.e. for any

continuous map T → S the base change π2 : X ×S T → T is a closed map.
1. If f : X → S is a proper morphism, so is π2 : X ×S T → T for any base change T → S.

In fact (X ×S T ) ×T Z = X ×S Z.
2. If Y → X and X → S are proper morphisms, so is the composition Y → S.

The composition of closed maps also is closed.
S
3. Let S = i Ui be an open cover. A morphism f : X → S is proper if and only if so is
f −1 (Ui ) → Ui for any index i.
A subset Y ⊆ S is closed if and only every Ui ∩ Y is closed in Ui .
4. The above theorem states that a topological space K is compact if and only if the projection
X → pt onto a point is a proper morphism.
5. If f : X → S is a proper morphism, and K ⊆ S is compact, then f −1 (K) is compact.

In fact, f −1 (K) → K and K → pt are proper; hence so is f −1 (K) → pt.
6. If K is compact, and S is separated, then any morphism X → S is proper.

For any base change T → S, the subspace K ×S T ⊆ K × T is closed, because it is the
inverse image of the diagonal by the natural map K × T → S × S; hence the composition
K ×S T ,→ K × T → T is a closed map.
7. Let S be a separated locally compact space, and f : X → S a continuous map. If any point
p ∈ S has a compact neighborhood Kp such that f −1 (Kp ) is compact, then f is proper.
The morphism f −1 (Kp ) → Kp is proper (f −1 (Kp ) is compact
S and Kp is separated); hence so
−1
is f (Up ) → Up , where Up is the interior of Kp , and S = p Up .
Lemma: Let f : X → S be a morphism. If f × Id : X × T → S × T is closed for any topological

space T , then f is a proper morphism.

Proof: Any morphism h : T → S is the composition of Γh : T → S × T , Γh (t) = h(t), t , with
the projection S × T → S.
The map X × T = X ×S (S × T ) → S × T is closed by hypothesis; hence so is after the base
change Γh : T → S × T , because Γh is an isomorphism onto a subspace of S × T .
Characterization: Proper morphisms are just closed morphisms with compact fibres.
8.3. SEPARATION PROPERTIES 223
Proof: The necessity is clear. Conversely, if f : X → S is a closed morphism with compact fibres,
let us see that f × Id : X × T → S × T is closed for any topological space T .
Let C ⊆ X × T be a closed set and put C 0 = (f × Id)(C) ⊆ S × T . If (s, t) ∈ / C 0 , then the
−1
compact set f (s) × t does not intersect the closed set C; hence there is an open set U × Vt in
X × T such that f −1 (s) × t ⊆ U × Vt and C ∩ (U × Vt ) = ∅. Then V = S − f (X − U ) is an open
set, and V × Vt is a neighborhood of (s, t) not intersecting C 0 . We see that C 0 is closed.
8.3 Separation Properties

Definition: Two points p1 6= p2 of the spectrum of a semiring A hybridize if both are in the
closure of some point; i.e., if p1 ∩ p2 contains a prime ideal.
In terms of the primes p∗i = A − pi of A∗ , we have that p1 and p2 do not hybridize if no
prime of A∗ contains them; i.e., p∗1 + p∗2 = A∗ .
In terms of A, there exist a1 , a2 ∈ A such that a1 a2 = 0, ai ∈
/ pi ; i.e., both points have
disjoint neighborhoods in Spec A.
Proposition: X is Hausdorff if and only if any two points of X ⊆ Spec B do not hybridize.
Proof: Let x1 , x2 ∈ X. If X is Hausdorff, there are b1 , b2 ∈ B such that bi ∈

/ pxi , b1 b2 = 0; hence
px1 , px2 do not hybridize.
Conversely, if px1 and px2 do not hybridize, there are b1 , b2 ∈ B such that bi ∈/ pxi , b1 b2 = 0;
and X is Hausdorff.
Proposition: A T1 space X is regular (open sets separate points and closed sets) if and only
if no point of X hybridize with a point of Specm AX .
Proof: Let x ∈ X, and let m be a maximal ideal of AX such that some closed set c ∈ m does not
pass through x. If X is regular, there are a, b ∈ AX such that
ab = 0 , a+c=1 , b∈
/ mx
hence a ∈/ m, since c ∈ m, and we see that mx , m do not hybridize.
Conversely, if a closed set c of X does not pass through x, by hypothesis for any maximal
ideal mi containing c there are closed sets ai , bi in X such that
ai ∈
/ mi , bi ∈ / mx , ai bi = 0.
T
Now (c)0 ∩ ( i (ai )0 ) does not intersect Specm AX , so that it is void.
Since Spec AX is compact, there exists a finite family a1 , . . . , an such that
∅ = (c)0 ∩ (a1 )0 ∩ . . . ∩ (an )0 = (c + a1 + . . . + an )0 ,
1 = c + a1 + . . . + an .
If we put a = a1 + . . . + an , b = b1 . . . bn , then ab = 0, a + c = 1, and x ∈
/ (b)0 .
Now Ua ∩ X and Ub ∩ X are disjoint neighborhoods of c and x.
Definition: A semiring A is normal if for each pair a1 , a2 ∈ A with sum a1 + a2 = 1 there is

some pair b1 , b2 ∈ A such that
a1 + b1 = 1 , a2 + b2 = 1 , b1 b2 = 0
and a space X is normal if so is AX (disjoint closed sets have disjoint neighborhoods).
Examples: Any Boolean algebra is normal (take bi = a0i ).
K = {0, g}, with a dense point g, is a normal space; but it is not T1 .
Theorem: If A is a semiring, the following conditions are equivalent,

1. A is a normal semiring.
2. Spec A is a normal space.
3. Any prime ideal of A is contained in a unique maximal ideal.

Proof: (1 ⇒ 2) If ∅ = (I1 )0 ∩ (I2 )0 = (I1 + I2 )0 , there are ai ∈ Ii such that a1 + a2 = 1.
By hypothesis we have
(ai )0 ∩ (bi )0 = (ai + bi )0 = ∅,

(b1 )0 ∪ (b2 )0 = (b1 b2 )0 = Spec A,
for some bi ∈ A. Now Ub1 , Ub2 are disjoint neighborhoods of the given closed sets.
(2 ⇒ 3) Closed points of Spec A have disjoint neighborhoods; hence they do not hybridize.
(3 ⇒ 1) By hypothesis the closure of each point has a unique closed point, and any two closed
points have disjoint neighborhoods; hence any two points with disjoint closure have disjoint
(basic) neighborhoods.
If a1 + a2 = 1, then C1 = (a1 )0 and C2 = (a2 )0 are disjoint.
If x ∈ C1 , any point of C2 and x have disjoint basic neighborhoods.
Since C2 is compact (and basic open sets are stable under finite unions and intersections)
C2 and x have disjoint basic neighborhoods.
Since C1 is compact, C1 and C2 have disjoint basic neighborhoods Ub1 , Ub2
(ai )0 ⊆ Ubi , ∅ = (ai )0 ∩ (bi )0 = (ai + bi )0 , ai + bi = 1

∅ = Ub1 ∩ Ub2 , Spec A = (b1 )0 ∪ (b2 )0 = (b1 b2 )0 , b1 b2 = 1
Corollary: Any base B of a compact separated space X is normal.
Proof: X = Specm B, and any two points of X do not hybridize.
Lemma: If A is normal, disjoint closed sets in Specm A have disjoint neighborhoods in Spec A.
Proof: If two closed sets (I1 )0 , (I2 )0 in Spec A intersect Specm A at disjoint closed sets, then
(I1 )0 ∩ (I2 )0 = (I1 + I2 )0 = ∅ because any ideal 6= A is contained in some maximal ideal, and
(I1 )0 , (I2 )0 have disjoint neighborhoods since Spec A is normal.
Corollary: If A is normal, Specm A is a separated space.
Corollary: If A is a Boolean algebra, then Spec A is a comapct Hausdorff space.
Proof: Any Boolean algebra is normal; hence Spec A = Specm A is T2 .
Continuous Retract Theorem: If A is a normal semiring, the map r : Spec A → Specm A,

where r(p) is the unique maximal ideal containing p, is a continuous retract.
Proof: Remark that each point x ∈ Specm A is in the closure of any point of the fibre r−1 (x);
hence any neighborhood of x in Spec A contains r−1 (x).
Let U be an open set in Specm A and C the complement.
By the lemma, any point x ∈ U and C have disjoint neighborhoods Vx , Wx in Spec A.
Hence r−1 (x) ⊆ Vx , r−1 (C) ⊆ −1
SWx , so that Vx ⊆ r (U ).
−1
We conclude that r (U ) = x∈U Vx is open, and r is continuous.
8.4. NOETHERIAN AND FINITE SPACES 225
8.4 Noetherian and Finite Spaces

Definition: A semiring A is noetherian if any ideal is finitely generated (hence principal).
A space X is noetherian if any descending chain of closed sets is finite (for example, the
spectrum of a noetherian ring). Any noetherian space X is compact, and
1. Any subspace Y of X is noetherian (and compact).

An infinite chain of closed sets C0 ⊃ C1 ⊃ . . . in Y would define an infinite chain of closed
sets C 0 ⊃ C 1 ⊃ . . . in X, because Ci = Y ∩ C i .
2. X has a finite number of irreducible components.

If X is not a finite union of irreducible closed sets, then X = Y1 ∪ Y2 , where some of the twon
closed sets Yi 6= X also has such property, so obtaining an infinite decreasing chain of closed
sets in X.
3. AX is a noetherian semiring, and it is the unique base of the topology of X.

An element c of an ideal I of AX generates I when no element of I is strictly contained in c,
so that a non principal ideal would define an infinite decreasing chain of closed sets. Any base
of X is AX since the intersection of any family of closed sets coincides with the intersection
of a finite subfamily.
4. Irreducible closed sets in X correspond to prime ideals in AX .

A closed set c of a space X is irreducible if and only if the ideal cAX is prime. If X is
noetherian, any prime ideal of AX is principal, and we conclude.
5. The spectrum of a noetherian ring is a noetherian topological space.

Any infinite chain of closed sets Y0 ⊃ Y1 ⊃ . . . in Spec A defines an infinite chain of ideals
I0 ⊂ I1 ⊂ . . . in A, where Ii = IYi (the inclusions are strict because Yi = (Ii )0 ). Hence, the
finiteness of the number of irreducible components gives another proof of the existence of a
finite number of minimal primes in a noetherian ring, p. 135.
If A is a noetherian semiring, then Spec A is a noetherian space since ideals of A correspond

to closed sets in the spectrum, (I)0 = (J)0 ⇔ I = J; hence (by 3)
A = A(Spec A).
However, if a T0 space X is noetherian, the inclusion X ⊆ Spec AX is strict when some

irreducible closed set in X has no dense point,
Theorem: The functors A Spec A, X AX define an antiequivalence of the category of

noetherian semirings with the category of noetherian T0 spaces where irreducible closed sets have
a dense point,
op
Noetherian Noetherian T0 spaces A = A(Spec A)
! ,
semirings with generic points X = Spec (AX )
Finite Spaces: Any finite ordered set X has a topology: a subset C is closed if it is stable
under specialization, x ≤ y ∈ C ⇒ x ∈ C. This topology determines the order, and a map
f : X → Y is order-preserving, x ≤ y ⇒ f (x) ≤ f (y), just when it is continuous.
Conversely, in a finite T0 topological space X closed sets are (finite) unions of closures of
points, so that the topology of X is fully determined by the order relation x ≤ y ⇔ x ∈ {y}.
Theorem: The functors A Spec A, X AX define an antiequivalence of categories

op
Finite Finite T0 Finite
! =
semirings spaces orders
Corollary: Projective limits of non void finite T0 spaces are non void spectral spaces.
Proof: If Xi = Spec Ai , then lim

←−
Xi = lim
←−
−→
Ai ).
Definitions: The barycentric subdivision βX of a finite order X is the space of all chains
x0 < x1 < . . . < xn , ordered by inclusion.
We have a natural continuous map βX → X transforming any chain into the last element,
and the inverse image of any closed set Y is βY . Inductively we define the iterated subdivisions,
β n X = β(β n−1 X), and we obtain a projective system
. . . −→ β n+1 X −→ β n X −→ . . . −→ βX −→ X
The geometric realization |X| of a finite order X is the subspace of all closed points of
lim
←−
β n X, and it is clear that |X| = |βX| = |β n X|.
If An denotes the topology of β n X, the continuous maps β n+1 X → β n X induce morphisms
An → An+1 , and we have lim
←−
β n X = lim
←−
(Spec An ) = Spec (lim
−→
An ). Hence
|X| = Specm (lim

−→
An ).
A continuous map f : X → X 0 induces a closed map
βf : βX −→ βX 0
transforming any chain x0 < x1 < . . . < xn of X into the greatest chain of βX contained in
f (x0 ) ≤ f (x1 ) ≤ . . . ≤ f (xn ).
We have closed maps β n f : β n X → β n X 0 and, by the lemma of p. 220, the continuous map
Spec (lim
−→
An ) = lim
←−
β n X −→ lim
←−
β n X 0 = Spec (lim
−→
A0n )
transforms closed points into closed points, hence it defines a continuous map
|f | : |X| −→ |X 0 |.
When Y → X is a closed embedding, |Y | → Y ×X |X| is a homeomorphism.
Theorem: The geometric realization of the finite order ∆n = {0, 1, . . . , n} is the n-tetrahedron,
i.e. the subspace {(t0 , . . . , tn ) ∈ Rn+1 : t0 + . . . + tn = 1, ti ≥ 0}.
Proof: Let A1 be the lattice of all finite unions of closed simplices (vertices, edges, faces,...) of
the n-tetrahedron. Simplices are just generators of prime ideals in A1 , so that Spec A1 has a
point for each simplex, and the order is the incidence relation.
Once we number the vertices of the tetrahedron, we have that Spec A1 = β∆n , where each
simplex is identified with the sequence formed by the vertices.
Let Ai+1 be the lattice of all finite unions of closed simplices of the i-th barycentric subdi-
vision of the tetrahedron (the new vertices being the barycentres of the former simplices).
8.4. NOETHERIAN AND FINITE SPACES 227
Now Spec Ai = β i ∆n , and we have commutative squares
Spec Ai+1 −→ Spec Ai

|| ||
i+1
β ∆n −→ β ∆n i
lim
←−
β i ∆n = lim
←−
Spec (Ai ) = Spec (lim
−→
Ai )
|∆n | = Specm (lim
−→
Ai ) = Specm B
The tetrahedron is a compact T1 space, hence it coincides (p. 220) with the maximal spec-
trum of any base of closed sets, and B is a base. q.e.d.
Remark that B is a base of a compact Hausdorff space; hence it is normal (p. 224) and we
have a continuous retract r : lim
←−
β i ∆n = Spec B −→ Specm B = |∆n |.
Theorem: The geometric realization of any finite T0 space X is a finite polyhedron (i.e., we
have a homeomorphism |X| ' P with a finite union P of closed simplices of a tetrahedron, and
such homeomorphism is named a triangulation of |X|).
Proof: If we number the points of X, then βX is a closed set in β∆n .

Hence |βX| = |X| is a closed set in |β∆n | = |∆n | formed by some simplices.
Examples: The above proof shows that the polyhedron |X| has a vertex for each point of X,
an edge for any chain x0 < x1 (joining the vertices x0 , x1 ), a face for any chain x0 < x1 < x2
(joining the edges x0 < x1 , x1 < x2 , x0 < x2 ), and so on.
We represent a finite space with a diagram of points at different levels, with an edge joining
a point y with an inferior point x when x < y; i.e., the closure of a point is formed by the points
under it (and closed points are at the bottom).
For example, the geometric realization of the following finite spaces
• • •
• • • • • •
• • • • • •
are a closed interval, a triangle, a circle and a sphere respectively.

Any semiring is an inductive limit of finite semirings, so that, in a certain sense, any space
X may be approximated by finite polyhedrons, since it is a dense subspace of a projective limit
of finite spaces, X ⊆ Spec AX = Spec (lim −→
Ai ) = lim
←−
(Spec Ai ).
Uryshon’s Lemma: If C0 and C1 are disjoint closed sets in a normal space X, there exists a
continuous function f : X → [0, 1] such that f (C0 ) = 0, f (C1 ) = 1.
Proof: I = β∆1 = {0, 1, g} is a finite space with two closed points 0,1 and a generic point g.
A continuous map φ : X → I is just a pair of disjoint closed sets C0 = φ−1 (0), C1 = φ−1 (1),
and the existence of disjoint open neighborhoods U1 , U2 in X amounts to lift φ to a continuous
map φ1 : X → βI,
x0g • • x1g φ1 •g
βI = −−−−−−→ = I
0• g• •1 0• •1
where U1 = φ−1 −1 −1
1 {0, x0g }, U2 = φ1 {1, x1g }, and therefore φ1 (g) = X − (U1 ∪ U2 ).
Since βI is formed by two copies of I, reiterating the argument we may lift φ1 to a continuous
map φ2 : X → β 2 I, and so on.
Now the continuous function f is given by the continuous retract theorem,
r
X −→ lim
←−
β n ∆1 −−−→ |∆1 | = [0, 1].
Spectral Spaces: A topological space X is a spectral (resp. Stone) space when it is homeo-
morphic to the spectrum of a semiring (resp. Boolean algebra), X = Spec A.
1. Any spectral space is a compact T0 space. If it is T1 , then it is a Stone space (p. 219); hence
it is a compact T2 space (p. 224) with a base of closed open sets. Conversely, if the Boolean
algebra B of closed open sets in a compact separated space K is a base of the topology, then
the map K → Spec B = Specm B is a homeomorphism, and K is a Stone space.
2. Since any semiring (resp. Boolean algebra) is an inductive limit of finite semirings (resp.
finite Boolean algebras) we see that spectral (resp. Stone) spaces are just the projective
limits of finite (resp. finite discrete) T0 spaces.
Hence any spectral space lim
←−
Xi is a continuous image lim
←−
(Xi )dis → lim
←−
Xi of a Stone space
and, by the Continuous Retract Theorem, any compact separated space is a continuous image
of a Stone space.
3. Compact open sets U ⊆ Spec A are basic because Ua1 ∪ . . . ∪ Uan = Ua1 +...+an ; hence any
spectral space X is homeomorphic to the spectrum of a unique (up to isomorphisms) semiring:
the family of all closed sets with compact complement.
Since compact sets in a Stone space are just closed sets, any Stone space is the spectrum of
a unique Boolean algebra: the algebra of all open closed sets. Hence any continuous map
Spec B → Spec A between Stone spaces is induced by a semiring morphism A → B, and we
see that the category of Boolen algebras is dual to the category of Stone spaces.
4. The spectrum of any commutative ring is a spectral space:
Theorem: A topological space is a spectral space if and only if it admits a base of compact open
sets (closed under finite unions and intersections) and any irreducible closed set has a unique
dense point.
Proof: Let B be a base of closed sets in X with compact complement.

T Any ideal I of B is just
the ideal of all closed sets b ∈ B containing the closed set c = a∈I a because, the complement
of b being compact, b contains a finite intersection of elements of I, so that b ∈ I.
Since B is a base, the lattice of closed sets in X is dual to the lattice of ideals of B.
Hence any prime ideal p ⊂ B corresponds to an irreducible closed set c = x̄, so that p = px
and the natural map X → Spec B is bijective. It is an homeomorphism because B is a base.
8.5 Compactifications
Definition: A continuous map X → K is a compactification of X if it is a homeomorphism
onto a dense subspace of a compact space K.
Any T0 space X admits the compactification X → Spec AX and, when it is T1 , the T1
compactification X → Specm AX . Now we study Hausdorff compactifications.
8.5. COMPACTIFICATIONS 229
Zeros z(f ) = {x ∈ X : f (x) = 0} of continuous functions f : X → R form a subring Z(X) of

AX (but it may be not a base of the topology of X) because
z(f ) + z(h) = z(f 2 + h2 ) , z(0) = 0

z(f )z(h) = z(f h) , z(1) = 1
Lemma: The semiring Z(X) is normal.
Proof: If z(f ) and z(h) are disjoint, the continuous function g = |f | − |h| is positive on z(h) and
negative on z(f ). If we put f 0 = 12 (g − |g|), h0 = 21 (g + |g|), then
z(f ) + z(f 0 ) = 1,
z(h) + z(h0 ) = 1,
z(f 0 ) · z(h0 ) = 0. q.e.d.
βX = Specm Z(X) is a compact Hausdorff space (p. 224) and we have a continuous retract
r : Spec Z(X) −→ Specm Z(X).
Composing it with the continuous map of dense image X → Spec Z(X) (it may be not
injective if Z(X) is not a base of the topology of X) we obtain a canonical map of dense image
j : X → βX, which is universal3 :
Theorem: Hom(βX, K) = Hom(X, K), f 7→ f j, for any compact Hausdorff space K.
Proof: A continuous map φ : X → K defines a morphism φ−1 : Z(K) → Z(X), and by Uryshon’s
lemma Z(K) is a base of K; hence K = Specm Z(K).
The required extension to βX (unique since the image of X → βX is dense and K is T2 ) is
r
βX = Specm Z(X) −→ Spec Z(X) −→ Spec Z(K) −−→ Specm Z(K) = K.
Definition: A T0 space X is completely regular when Z(X) is a base of the topology of X

(continuous functions separate points and closed sets).
Any subspace of a completely regular space so is completely regular, and by Uryshon’s lemma
any normal T1 space is completely regular.
Any completely regular space is regular, hence it is Hausdorff.
Theorem: If X is a topological space, the following conditions are equivalent,
1. X is completely regular.
2. βX is a compactification of X (the Stone-Cěch compactification).
3. X admits a Hausdorff compactification.
Proof: (1 ⇒ 2) If X is completely regular, j : X → βX is a compactification since Z(X) is a

base of the topology of X.
(2 ⇒ 3) is obvious, and finally, if X admits a Hausdorff compactification X → K, since K
is completely regular, so is the subspace X.
3
The existence of a universal continuous map into compact Hausdorff spaces also follows from the representabil-
ity theorem.
Proposition: A T1 topological space X is normal if and only if βX = Specm AX .
Proof: If X is normal, the extension to Specm AX of a continuous map X → K (unique since

the image of X → Specm AX is dense and K is T2 ) is
r
X ,→ Specm AX −→ Spec AX −→ Spec AK −−→ Specm AK = K.
If βX = Specm AX , then Specm AX is Hausdorff, so that maximal ideals of AX do not

hybridize; hence AX is normal. q.e.d.
C(X) will denote the ring of real continuous functions on a topological space X, and C ∗ (X)
the ring of bounded real continuous functions.
Any point x ∈ X defines a maximal ideal mx = {f ∈ C(X) : f (x) = 0}, and this map
X −→ Specm C(X)
is continuous and has dense image since (f )0 ∩ X = z(f ).

Analogously we have a continuous map of dense image X −→ Specm C ∗ (X).
Theorem: βX = Specm C ∗ (X).
Proof: C ∗ (X) = C(βX) by the universal property of βX, and Specm C(βX) = βX, since βX is
a compact Hausdorff space (p. 250).
Theorem: βX = Specm C(X).
Proof: Any ideal I of the semiring Z(X) defines an ideal I 0 = {f ∈ C(X) : z(f ) ∈ I} of C(X),
and any ideal J of C(X) defines an ideal z(J) = {z(f ) : f ∈ J} of Z(X).
Now, if m is a maximal ideal of Z(X), so is m0 since for any ideal m0 ⊂ J ⊂ C(X) we have
that m = z(m0 ) ⊂ z(J) ⊂ Z(X) (if z(f ) = ∅, then f is invertible).
Analogously, if m is a maximal ideal of C(X), so is z(m) since for any ideal z(m) ⊂ I ⊂ Z(X)
we obviously have m ⊂ I 0 ⊂ C(X).
So we obtain a natural bijection Specm Z(X) = Specm C(X), and it is a homeomorphism
since the zeros of z(f ) correspond to the zeros of f .
Definition: We say that a subalgebra B ⊆ C(X) separates points of X if for each pair of points
x 6= y of X there is a continuous function f ∈ B such that f (x) 6= f (y). In such a case, for any
pair of real numbers a 6= b there exists h ∈ B such that h(x) = a, h(y) = b.
Stone-Weierstrass Theorem: Let K be a compact separated space. A subalgebra B of C(K)

is dense if and only if it separates points of K.
Proof: If B is dense, and we take f ∈ C(K) such that f (x) = 0, f (y) = 1 (it exists by Urysohn’s
lemma), any function h ∈ B such that kf − hk < 21 separates the points x, y.
Conversely, if B separates points, considering the closure we may assume that B is closed,
and we have to show that for all f ∈ C(K), ε > 0, there is h ∈ B such that kf − hk < ε.
Given x ∈ K, for any y ∈ K there is hy ∈ B such that hy (x) = f (x), hy (y) = f (y), and on
an open neighborhood Uy we have h < f + ε. S
There are points y1 , . . . , yn such that K = i Uyi , and we put hx = min(hy1 , . . . , hyn ).
It is clear that hx (x) = f (x) and hx < f + ε, and by the following lemma hx ∈ B.
On an open neighborhood Vx of x we have fS− ε < hx .
There are points x1 , . . . , xm such that K = i Vxi .
8.6. DIMENSION THEORY 231
Now h = max(hx1 , . . . , hxm ) ∈ B, and f − ε < h < f + ε.
Lemma: Let B be a closed subalgebra of C(K). If f, h ∈ B, then max(f, h), min(f, h) ∈ B.
Proof: Since max(f, h) = 21 (f + h + |f − h|) and min(f, h) = f + h − max(f, h), it is enough to

show that |f | is in the closure of B whenever f ∈ B. Now, the series
√
1 − t = 1 − 12 t + . . .
uniformly converges on |t| ≤√1. If Pn (t) is the Taylor polynomial of degree n, for any ε > 0 there
exists n such that |Pn (t) − 1 − t | < ε on |t| ≤ 1.
If f ∈ B, dividing by a constant we have that |f | ≤ 1, and if we put t = 1 − f (x)2 ,
p
|Pn (1 − f 2 ) − |f || = |Pn (1 − f 2 ) − 1 − (1 − f 2 ) | < ε ,
hence |f | is adherent to B, since Pn (1 − f 2 ) ∈ B.
Corollary: Let X be a σ-compact space. A subalgebra B of C(X) is dense (with the topology of
uniform convergence on compact sets) if and only if it separates points of X.
Proof: If B separates points of X, for any compact set K ⊂ X we have that the image of the
restriction morphism B → C(K) separates points of K.
Hence it is dense in C(K), and B is dense in C(X).
Tietze’s Extension Theorem: Let X be a T1 normal space. If Y is a closed set in X, then

the restriction morphism C ∗ (X) → C ∗ (Y ) is surjective.
Proof: Since X is normal and T1 , the Stone-Cěch compactification is βX = Specm AX .

Since Y is closed, it is normal and T1 , so that βY = Specm AY ⊂ Specm AX , because
AY = AX /IY .
Replacing X, Y by βX, βY , we may assume that X is compact.
Let us consider the image of the restriction morphism C(X) → C(Y ). It is a subalgebra
which separates points of Y , so that it is dense.
If h ∈ C(Y ), there are functions fn ∈ C(X) such that |h − fn | < 2−n on Y .
Let φn : X → [0, 1] be a continuous function vanishing on Y and with value 1 on the closed
set |fn+1 − fn | ≥ 21−n (both sets do not intersect since the functions fn , fn+1 differ from h less
that 2−n on Y ). Now the series
∞
P
f = f1 + (fn+1 − fn )φn
n=1
uniformly converges, since |(fn+1 − fn )φn | < 21−n−1 , and it defines a continuous function on X
coinciding with h on Y , since on Y the n-th partial sum is fn , and fn → h.
8.6 Dimension Theory

Definition: The dimension of a topological space is the minimal (Krull) dimension of a base
of closed sets. The dimension may be infinite, and dim X = −1 if and only if X = ∅.
Theorem: dim X = 0 if and only if X 6= ∅ and it admits a base of open closed sets.
Proof: The open closed sets in X form a base if and only if some base of X is a Boolean algebra;
i.e., a semiring of dimension 0 (p. 219).
Examples: Rational numbers and irrational numbers define two spaces of dimension 0; but the
union R is connected, hence of dimension > 0.
Any space X of dimension 0 admits a base B which is a Boolean algebra; hence X is a
subspace of the compact Hausdorff space Spec B (p. 224), and X is completely regular.
Products and projective limits of spaces of dimension 0 have dimension 0. In particular any
profinite space is a Stone space (a compact separated space of dimension 0). Conversely, any
Stone space is K = Specm B = Spec B for some Boolean S algebra; hence it is a profinite space.
Let K be a compact Hausdorff space. Then AK = i Ai , where the semirings Ai are finite.
Let X i be the finite set Spec Ai with the discrete topology. By the continuous retract theorem,
K is a continuous image of a compact 0-dimensional space:
r
lim
←−
X i −→ lim
←−
−→
Ai ) = Spec AK −−−→ Specm AK = K
Lemma: If we have a surjective morphism A → A0 , then dim A0 ≤ dim A.
Proof: Any chain of prime ideals p0 ⊂ p1 ⊂ . . . ⊂ pn in A0 defines a chain of prime ideals

p0 ∩ A ⊂ p1 ∩ A ⊂ . . . ⊂ pn ∩ A in A.
Theorem: If Y is a subspace of X, then dim Y ≤ dim X.
Proof: If B is a base of X, the image of the restriction morphism B → AY is a base of Y .

Hence dim Y ≤ dim B.
Theorem: dim (Spec A) = dim A, for any semiring A.
Proof: dim (Spec A) ≤ dim A since A defines a base of Spec A.

On the other hand, if B is a base of Spec A, then Spec A is a subspace of Spec B.
Since the dimension of a semiring is the supremum of the lengths of all chains of specializa-
tions in the spectrum, we conclude that dim A ≤ dim B.
Corollary: The dimension of a noetherian space X is the supremum of the lengths of all chains
of irreducible closed sets4 in X.
Proof: If X is noetherian, it has a unique base AX , and prime ideals in AX correspond to

irreducible closed sets in X (p. 225).
Corollary: The dimension of a finite T0 space X is the maximal length of a chain in X (con-
sidered as a finite order), hence dim X = dim (βX).
Theorem: dim (X × Y ) ≤ dim X + dim Y .
Proof: If A and B are bases of X and Y , the image of the morphism A ⊗ B → A(X × Y ) is a
base of X × Y ; hence dim (X × Y ) ≤ dim A ⊗ B = dim A + dim B (p. 218).
Theorem: dim (lim

←−
Xi ) ≤ sup{dim Xi }, when the spaces Xi are finite and T0 .
Proof: If Xi = Spec Ai , then lim

←−
Xi = Spec (lim
−→
Ai ), and (p. 218)
dim Spec (lim

−→
Ai ) = dim (lim
−→
Ai ) ≤ sup{dim Ai }.
4
The supremum of the lengths of all chains of irreducible closed sets in a space X is the combinatorial or
Grothendieck dimension of X.
8.7. UNIFORM SPACES 233
Corollary: dim |X| ≤ dim X, for any finite T0 space X.
Proof: dim |X| ≤ dim (lim

←−
β n X) ≤ sup{dim β n X} = dim X.
Fundamental Theorem of Dimension Theory: dim Rn = n.
Proof: dim Rn ≤ dim |∆n | ≤ n; but the proof of the equality is postponed to p. 331.
Corollary: Any non empty open set U ⊆ Rn has dimension n. Hence, it is not homeomorphic
to an open subset of Rm when n 6= m.
Proof: Take an open ball B ⊆ U .

Since B is homeomorphic to Rn , we have n = dim B ≤ dim U ≤ dim Rn = n.
Corollary: dim |X| = dim X, for any finite T0 space X.
Proof: If dim X = n, then |X| contains a n-tetrahedron; hence Rn , so that n ≤ dim |X|.
Lemma: Let K be a compact separated space. If A is a dense subalgebra of C(K),

dim K ≤ dim A.
Proof: Let B be the semiring generated by the closed sets f −1 (I), where f ∈ A and I is a closed
interval (bounded or not). Since A is dense, B is a base of X. If p is a prime ideal of B, then
p0 = {f ∈ A : z(f ) ∈ p} is a prime ideal of A, and we have to show that p0 ⊂ q0 when p ⊂ q.
The closed sets f −1 [0, ∞) generate B, so that q contains a closed set f −1 [0, ∞) ∈ / p.
Since p is prime, and 0 = f −1 (−∞, 0] · f −1 [0, ∞), we have that f −1 (−∞, 0] ∈ p ⊂ q, and
z(f ) = f −1 (−∞, 0] + f −1 [0, ∞) ∈ q.
/ p, because f −1 [0, ∞) ⊇ z(f ) is not in p, we see that f ∈ q0 , and f ∈
Since z(f ) ∈ / p0 .
Theorem: The dimension of any finite polyhedron P is the minimal dimension of a dense
subalgebra of C(P ).
Proof: P is a finite union of some simplices σ of a tetrahedron in Rn .

The algebra A of polynomial functions on P separates points of P , and A ' R[x1 , . . . , xn ]/I,
where I is theTideal of all polynomials vanishing on P .
Since I = σ pσ , where pσ is the ideal of all polynomials vanishing on a given simplex σ of
P , the dimension of A is the maximal dimension of R[x1 , . . . , xn ]/pσ .
Now, R[x1 , . . . , xn ]/pσ ' R[y1 , . . . , yd ], d = dim σ, because pσ is the ideal of the linear
subvariety spanned by σ, so that dim A = dim P .
8.7 Uniform Spaces

Let X be a set. Given subsets ε, δ ⊆ X × X we put:
ε = {(x, y) ∈ X × X : (y, x) ∈ ε},
∗
ε ◦ δ = {(x, z) ∈ X × X : (x, y) ∈ ε, (y, z) ∈ δ for some y ∈ X},

nε = ε◦ . n. . ◦ε,
ε(x) = {y ∈ X : (x, y) ∈ ε} , x ∈ X.
A filter in a set X is a proper ideal F of the algebra PX of subsets of X: it is closed under
finite intersections, if A ∈ F then A ∪ B ∈ F for any subset B ⊆ X, and X ∈ F, ∅ ∈ / F.
Definition: A uniform space is a set X endowed with a filter of subsets of X × X, named

entourages satisfying the following conditions:
1. Any entourage contains the diagonal ∆ = {(x, x); x ∈ X}.
2. If ε is an entourage, then so is ε∗.

3. If ε is an entourage, there is an entourage δ such that 2δ ⊆ ε.
A family of entourages is a base of the uniformity when any entourage contains a member
of such family. The symmetric (i.e. ε∗ = ε) entourages form a base, because ε∗ ∩ ε is
symmetric. Moreover, if {εi } is a base of entourages, then so is {nεi } for any n ∈ N.
A map f : X → Y between uniform spaces is uniformly continuous when, for any en-
tourage ε in Y , we have that (f × f )−1 (ε) is an entourage in X; i.e. there is an entourage δ in
X such that (f × f )(δ ) ⊆ ε.
Induced Topology: A uniform structure defines a topology on X, the filter of neighborhoods
of a point x ∈ X being {ε(x)}, where ε runs on the entourages of X.
In fact, these are the filters of neighborhoods of a topology because, if 2δ ⊆ ε, then for any
y ∈ δ (x) ⊆ ε(x) we have δ (y) ⊆ (2δ )(x) ⊆ ε(x), so that δ (x) is in the interior of ε(x).
Let us consider on X ×X the product topology. If ε is symmetric, then 3ε is a neighborhood
of ε, because ε(x) × ε(y) ⊆ 3ε, ∀(x, y) ∈ ε. Hence, the open entourages form a base of the
uniformity, and any entourage is a neighborhood of the diagonal. Also ε̄ ⊆ 3ε, so that the closed
entourages form a base of the uniformity, and any point admits
a base of closed
neighborhoods.
Uniformly continuous maps are continuous: f −1 ε(y) = (f × f )−1 (ε) (x), y := f (x).
The T0 Quotient: The topology of a uniform space X is T0 when, for any pair of points x 6= y,
there is an entourage ε such that (x, y) or (y, x) is not in ε. In such case, if δ is a symmetric
uniform neighborhood such that 2δ ⊆ ε, then δ (x) ∩ δ (y) = ∅, and we see that X is separated,
and the intersection of all entourages is just the diagonal.
In general, the intersection of all entourages is the set of pairs (x, y) ∈ X ×X such that x̄ = ȳ.
Hence, if we identify such points, the canonical projection π : X → Xs induces an isomorphism
A(Xs ) = AX between the lattices of closed sets, and we see that any point x ∈ X defines a
maximal ideal of AX , that we denote mx = {A ∈ AX : x ∈ A}.
Now εs ⊆ Xs × Xs is defined to be an entourage when (π × π)−1 (εs ) is an entourage in X;
i.e. when εs = (π × π)(ε) for some entourage ε in X. So we obtain a uniformity on Xs because
(π × π)−1 (εs ) ⊆ 3ε, so that 2δ s ⊆ εs when 6δ ⊆ ε.
The uniform space Xs is separated, π : X → Xs is uniformly continuous, and any uniformly
continuous map X → Y to a separated uniform space Y uniquely factors through π.
Examples: (1) Any pseudometric d on a set X defines a uniform structure on X, such that a
base of entourages is defined by the sets Uε = Ud,ε := {(x, y) ∈ X × X : d(x, y) < ε}, ε ∈ R+ ,
and it is a uniform structure because Uε∗ = Uε and Uε ◦ Uε ⊆ U2ε . The corresponding topology
is just the topology defined by d. In general, any family {di } of pseudometrics on X defines a
uniform structure, a base of entourages being the finite intersections Udi1 ,ε1 ∩ . . . ∩ Udin ,εn .
(2) Any topological vector space E is a uniform space, a base of entourages being the sets
Ũ = {(x, y) ∈ E × E : y − x ∈ U }, where U runs over the neighborhoods of 0 in E. In fact
(Ũ )∗ = (−U )∼ , and Ṽ ◦ Ṽ ⊆ Ũ when V is a neighborhood of 0 such that V + V ⊆ U .
(3) The initial uniformity of a family of maps fi : X → Yi to uniform spaces Yi is defined
by the filter generated by the sets (fi × fi )−1 (εi ) ⊆ X × X, where εi runs over the entourages
of the spaces Yi , so that a map f : T → X is uniformly continuous if and only if so are the maps
fi ◦ f . The sets (fi1 × fi1 )−1 (εi1 ) ∩ . . . ∩ (fin × fin )−1 (εin ) form a base of entourages, and the
induced topology is just the initial topology of the continuous maps fi .
In particular, any subset Y of a uniform space X inherits a uniform structure.
Theorem: A compact separated space K admits a unique uniformity inducing its topology, the
entourages being all the neighborhoods of the diagonal.
Proof: Any continuous function f : K → R defines a pseudometric df (x, y) = |f (y) − f (x)|,

and the uniformity defined by these pseudometrics induces the topology of K because K is a
completely regular space (p. 229).
Finally, T
given
a uniformity inducing the topology of K, if U is a neighborhood of the diagonal,
c
then U ∩ ε = U ∩ ∆ = ∅, where ε runs over the closed entourages and U c := K − U .
c
Since K × K is compact, there is a finite empty intersection U c ∩ ε1 ∩ . . . ∩ εn = ∅.

Hence U ⊇ ε1 ∩ . . . ∩ εn is an entourage of the considered uniformity.
Corollary: Any continuous map f from a compact separated space K to a uniform space Y is
uniformly continuous.
Proof: If ε is an entourage in Y , then ε is a neighborhood of the diagonal in Y × Y ; hence

(f × f )−1 (ε) is a neighborhood of the diagonal in K × K, so that it is an entourage in K.
Definitions: A filter F in a topological space X converges to a point x ∈ X (unique when X

is separated) if it contains the filter Nx of neighborhoods of x. In particular, the limit point x
is adherent to any set A ∈ F.
A filter F in a uniform space X is a Cauchy filter if for any entourage ε there is a ε-small
set A ∈ F (i.e. A × A ⊆ ε); hence a closed ε-small set: if δ ⊆ ε is a closed entourage and A is
¯ = δ ⊆ ε. So, if a point x ∈ X is adherent to any set in F then
a δ -small set, then Ā × Ā ⊆ δ
F converges to x, because x ∈ Ā ⊆ ε(x), so that ε(x) ∈ F: A Cauchy filter F converges to x
when the maximal ideal mx of the semiring AX contains the ideal AX ∩ F.
1. Given a map f : X → Y and a filter F in X, then f (F) := {B ⊆ Y : f −1 (B) ∈ F } is a filter

in Y . When f is continuous and F converges to x ∈ X, then f (F) converges to f (x). When
f is uniformly continuous and F is a Cauchy filter on X, then f (F) is a Cauchy filter on Y .
2. In a uniform space X, the filter Nx of neighborhoods of x ∈ X, and the filter Fx of all subsets
containing x, are always Cauchy filters.
3. Given a sequence (xn ) in a topological space X, the filter F(xn ) of subsets containing all the
terms xn up to a finite number converges to a point x ∈ X if and only if xn → x.
A sequence (xn ) in a uniform space X is defined to be a Cauchy sequence when F(xn ) is a
Cauchy filter: for any entourage ε there is an index k such that (xn , xm ) ∈ ε, ∀n, m ≥ k.
Definition: A uniform space X is complete when it is T0 (hence separated) and any Cauchy
filter F on X converges to a point x ∈ X.
In the case of metric spaces, this definition coincides with the older one (p. 25):
Lemma: A separated uniform space X with a countable base of entourages {εn } is complete if
and only if any Cauchy sequence in X converges.
Proof: If (xn ) is a Cauchy sequence in X, then F(xn ) is a Cauchy filter. If X is complete, then
F(xn ) converges to a point x ∈ X, so that xn → x.
Conversely, replacing εn by ε1 ∩ . . . ∩ εn , we may assume that εn+1 ⊆ εn .
Let F be a Cauchy filter and let An be a closed set in F such that An × An ⊆ εn .
Replacing An by A1 ∩ . . . ∩ An , we may assume that An+1 ⊆ An . Pick xn ∈ An .
Then (xn , xm ) ∈ Ak × Ak ⊆ εk , ∀n, m ≥ k, so that (xn ) is a Cauchy sequence and it

converges to a point x ∈ Ān = An . Since An ⊆ εn (x), we conclude that F converges to x.
Proposition: Any compact separated uniform space K is complete.
Proof: If F is a Cauchy filter in K, then AK ∩ F is a proper ideal of AK ; hence it is contained

in some maximal ideal mx of AK (p. 219) so that F converges to x ∈ K.
Proposition: Any closed subspace i : Y ,→ X of a complete uniform space is complete.
Proof: If G is a Cauchy filter on Y , then i(G) is a Cauchy filter on X; hence i(G) converges to
some point x ∈ X, so that x ∈ Ȳ = Y and G converges to x.
Proposition: Any complete subspace Y of a separated uniform space X is a closed subspace.
Proof: If x ∈ Ȳ , then ∅ ∈
/ Nx ∩ Y , so that Nx ∩ Y is a Cauchy filter on Y .
Hence Nx ∩ Y converges to some point y ∈ Y , and y = x because X is separated.
Lemma: Let ι : Y → X be a dense subspace of a uniform space X. If any Cauchy filter G on

Y converges in X, in the sense that ι(G) = {A ⊆ X : A ∩ Y ∈ G} converges, then any Cauchy
filter F on X converges.
o
Proof: The filter F o = {A ⊆ X : A∈ FS} is a Cauchy filter: if ε is a symmetric entourage and
A ∈ F is a ε-small, then the open set a∈A ε(a) is 3ε-small.
Since F o ⊆ F, it is enough to see that F o converges.
Now F o ∩ Y is a filter because Y is dense, and it is a Cauchy filter on Y .
Hence ι(F o ∩ Y ) converges to some point x ∈ X, adherent to any subset in F o ⊆ ι(F o ∩ Y ),
and we see that F o converges to x.
Completion of a Uniform Space: Let X be a uniform space, and X̃ the set of Cauchy filters
on X. If ε is a symmetric entourage in X, then ε̃ ⊆ X̃ × X̃ denotes the set of pairs (F, F 0 ) of
Cauchy filters with a common ε-small set. The sets ε̃ form a base of entourages for a uniformity
on X̃: Clearly ε̃ contains the diagonal, it is symmetric, (ε ∩ δ )∼ ⊆ ε̃ ∩ δ ˜ , and ε̃ ◦ δ
˜ ⊆ (ε ◦ δ )∼ .
In fact, if (F, F 0 ) have a common ε-small set A and (F 0 , F 00 ) have a common δ -small set B,
then ∅ =6 A ∩ B ∈ F 0 , so that A ∪ B is a (ε ◦ δ )-small set, and A ∪ B is in F and F 00 .
Now, X has the initial uniformity of the map ι : X → X̃, ι(x) = Fx = {A ⊆ X : x ∈ A},
because δ ⊆ (X × X) ∩ ε̃ ⊆ ε whenever δ is a closed symmetric entourage and 2δ ⊆ ε.
In fact, if (x, y) ∈ δ , then x, y ∈ δ (x) and δ (x) is a closed ε-small set in Fx and Fy .
Let us see that ι(X) is dense and that any Cauchy filter F on X converges in X̃:
For every symmetric entourage ε in X, there is a ε-small set A ∈ F. If x ∈ A, then
ι(x), F) ∈ ε̃, so that ι(A) ⊆ ε̃(F); hence F ∈ X̃ is in the closure of ι(X), and the filter ι(F)
contains any neighborhood ε̃(F) of F in X̃.
By the lemma, the separated quotient X b = (X̃)s is a complete space, named completion
of the uniform space X, endowed with a canonical uniformly continuous map i = πι : X → X b
with dense image, such that X has the initial uniformity.
When X is separated, i : X → X b is injective; hence it is an isomorphism onto a dense
subspace of X. If moreover X is complete, then any Cauchy filter F on X converges to a point
b
x ∈ X, so that (F, Fx ) ∈ ε̃ for any symmetric entourage ε in X; hence π(F) = π(Fx ) = i(x)
and we see that i : X → X b is an isomorphism.
If f : X → Y is uniformly continuous and F is a Cauchy filter in X, then f (F) is a Cauchy

filter in Y , so that f induces a natural uniformly continuous map f˜: X̃ → Ỹ ; hence a uniformly
continuous map fb: X b = (X̃)s → (Ỹ )s = Yb and a commutative square
f
X /Y
i i

fb
X
b /Y
b
Universal Property: If f : X → Y is a uniformly continuous map and Y is complete, then

b → Y such that f = fb ◦ i.
there exists a unique uniformly continuous map fb: X
Homunif (X,
b Y ) = Homunif (X, Y ) , φ 7→ φ ◦ i.
Proof: The existence of fb follows form the above commutative square, because Y = Yb , and
this continuous map fb is unique since it is fully determined on the dense set i(X), and Y is
separated.
Theorem: If X is a subspace of a complete uniform space, then the completion of X is just the
closure, X
b = X̄.
Proof: Let us see that the inclusion map i : X → X̄ has the above universal property.
Given a point x ∈ X̄, the Cauchy filter f (Nx ∩ X) converges to a unique point φ(x) ∈ Y ,
so that φ(x) is in the closure of f (Ux ∩ X) for any neighborhood Ux of x in X̄, and we only
have to show that such extension φ : X̄ → Y (obviously unique since X is dense in X̄ and Y is
separated) is uniformly continuous.
Given a closed entourage ε in Y , there exists a symmetric δ in X̄ such that
entourage
(f × f ) (3δ ) ∩ X × X ⊆ ε. If (x, x0 ) ∈ δ , then δ (x) ∩ X × δ (x0 ) ∩ X ⊆ (3δ ) ∩ X × X and,

ε being closed, we conclude that (φ × φ)(δ ) ⊆ ε:

φ(x), φ(x0 ) ∈ f δ (x) ∩ X δ (x0 ) ∩ X ⊆ (f × f ) (3δ ) ∩ X × X ⊆ ε = ε.

×f
Theorem: A uniform space X is precompact (with compact completion) if and only if it

admits, for every entourage ε, a finite cover by ε-small sets.
Proof: Consider anSentourage ε0 in X b such that (X × X) ∩ ε0 ⊆ ε. If X b is compact, it admits

n Un by ε -small sets; hence X = n (Un ∩ X), and Un ∩ X is ε-small.
0 0 S 0 0
a finite cover Xb=
Conversely, to prove that X b is compact, since it is complete, it is enough to show (p. 219)
that any maximal ideal m of the semiring AXb generates a Cauchy filter. For any closed entourage
ε0 in X,
b put ε := (X × X) ∩ ε0 . There is a finite cover X = S Un by ε-small sets, and each
n
b is ε0 -small because ε0 is closed. Now X
closure i(Un ) in X b = i(X) = S i(Un ), and i(Un ) ∈ m
n
for some index n because m is a prime ideal.
Corollary: Any direct product of precompact spaces also is precompact.
Corollary: A complete uniform space is compact if and only if it admits a finite cover by ε-
small sets for every entourage ε. Hence, in a complete uniform space, compact sets are just
closed sets admitting a finite cover by ε-small sets for every entourage ε.
8.8 Galois Theory of Coverings

Now we consider the category of spaces over a fixed topological space S.
Definition: A continuous map X → S is a trivial covering if there is a discrete space F and
an isomorphism X ' S × F = qF S, and it is a covering if we may cover S with open sets U
where X ×S U → U is a trivial covering.
Examples: The maps e2πit : R → S1 , ez : C → C∗ = C − 0, z n : C∗ → C∗ , π : Sn → Pn (R), and

π : C → C/Γ, where Γ = Ze1 + Ze2 is discrete, are coverings.
If X → S is a covering, X inherits the local structure of S. When S is a smooth manifold,
a Riemann surface, a riemannian manifold, etc., so is X.
1. Coverings are stable under base changes. If X → S is a covering, so is T ×S X → T for any

continuous map T → S.
2. The concept of covering is local. If R → S is an open cover (or even a covering), then X → S
is a covering if and only if R ×S X → R is a covering.
3. If some maps Xi → S are coverings, so is qi Xi → S.
4. The fibres of a covering X → S have locally constant cardinal; hence it is constant (the
degree of the covering) when S is connected. The degree may be infinite, contrary to what
has been assumed in the case of fields (p. 147) and noetherian rings (p. 213).
5. If X → S is a surjective covering, any continuous map X → T constant on the fibres induces

a continuous map S → T . In particular, the sequence X ×S X ⇒ X → S is exact.
The induced map S → T is continuous because X → S is a local homeomorphism.
6. The functor “connected components” defines an equivalence of the category of trivial coverings
of a given connected space S with the category of sets.
7. When S is a locally connected space, if a group G acts on a covering X → S, then the quotient
space X/G is stable under base changes, T ×S (X/G) = (T ×S X)/G.
To show that the natural bijection (T ×S X)/G → T ×S (X/G) is a homeomorphism we may
assume that S is connected and X = S × F is a trivial covering, so that G acts through
an action on the discrete space F and X/G = S × (F/G). In such a case, (T ×S X)/G =
(T × F )/G = T × (F/G) = T ×S (X/G).
From now on we assume that S is connected and locally connected.

Then coverings of S are locally connected, morphisms between coverings of S are coverings,
and any connected component of a covering of S also is a covering.
Moreover, any section of a connected covering X → S is an isomorphism; hence morphisms
of a connected covering X into another covering Y are given by the Points Formula:

Connected components of X ×S Y
HomS (X, Y ) = HomX (X, X ×S Y ) =
where each S-morphism f : X → Y correspond to its graphic Γf := {(x, f (x))}x∈X .

Hence, if two morphisms X ⇒ Y coincide at a point, then they coincide (if two connected
components have a common point, they coincide).
8.8. GALOIS THEORY OF COVERINGS 239
Definition: A connected covering X → S is a Galois covering when π1 : X ×S X → X is a

trivial covering,
X ×S X = qG X = X × G , G := HomS (X, X).
Since π2 : X ×S X → X also is a trivial covering, we see that π2 : Γf → X is an isomorphism

for any S-morphism f : X → X. Hence f is an automorphism and G = Aut(X/S) is a group,
named Galois group.
Now, the exact sequence X × G = X ×S X ⇒ X → S shows that
S = X/G.
Therefore, when a covering Y → S is trivial over a Galois covering X → S of group G, it is

fully determined by the G-set
F (Y ) = HomS (X, Y ) = [Connected components of X ×S Y ]
because we have Y = (X/G) ×S Y = (X ×S Y )/G = (X × F (Y ))/G.

Hence, given a G-set ∆, we define the associated covering to be
R(∆) = (X × ∆)/G
where ∆ is considered with the discrete topology.

The natural map R(∆) → S is a covering trivial over X since
X ×S R(∆) = X ×S (X ×∆)/G = ((X ×S X)×∆)/G = (X ×G×∆)/G = X ×(G×∆)/G = X ×∆.
Moreover, the points formula shows that ∆ = HomS (X, R(∆)) = F R(∆) and we obtain the
following result:
Galois Theorem: Let S be a connected and locally connected space. If X → S is a Galois

covering of group G, then the functors F and R define an equivalence of categories

Coverings of S R ◦ F = Id
! G-sets ,
trivial over X F ◦ R = Id
Definition: Let S be a connected and locally connected space. A connected covering S̄ → S is

a universal covering when any covering of S̄ is trivial.
In particular S̄ ×S S̄ → S̄ is a trivial covering, so that S̄ → S is a Galois covering and the
Galois group Aut(S̄/S) classifies coverings of S:
Theorem: If S admits a universal covering S̄ → S, then the category of coverings of S is

equivalent to the category of G-sets, where G = Aut(S̄/S).
If it exists, the universal covering is unique up to (non canonical) isomorphisms: If S̄1 → S is

another universal covering, then S̄ ×S S̄1 is a trivial covering of both factors; hence any connected
component is isomorphic to S̄ and S̄1 .
The isomorphism is unique once we fix points of S̄ and s̄1 over a given point of S.
8.8.1 The Fundamental Group

Definitions: A continuous map γ : I = [0, 1] → X is a path of end points p = γ(0), q = γ(1),
and if γ 0 is a path of end points q, r, the composition γγ 0 is the path
(
0 γ(2t) 0 ≤ t ≤ 21
(γ · γ )(t) =
γ 0 (2t − 1) 12 ≤ t ≤ 1
Two paths γ0 , γ1 from p to q are homotopic, γ0 ≡ γ1 , if there is a continuous map H : I ×I → X,

named homotopy and we put Hs (t) = H(t, s), such that
H0 = γ0 , H1 = γ1 , Hs (0) = p, Hs (1) = q,
and this is an equivalence relation, compatible with the composition of paths. It is clearly reflexive
and symmetric, and if H : γ0 ≡ γ1 , H 0 : γ1 ≡ γ2 , then H 00 : γ0 ≡ γ2 , where
(
H2s 0 ≤ s ≤ 21
Hs00 = 0 1
H2s−1 2 ≤s≤1
and if H : γ ≡ γ 0 , H̄ : γ̄ ≡ γ̄ 0 , then H 0 : γγ̄ ≡ γ 0 γ̄ 0 , where

(
0 Hs (2t) 0 ≤ t ≤ 21
Hs (t) =
H̄s (2t − 1) 12 ≤ t ≤ 1
If q is the constant path q(t) = q, then H : γq ≡ γ (analogously pγ ≡ γ), where
( t

2t
γ 1+s 2t ≤ 1 + s q
H(t, s) = γ
q 2t ≥ 1 + s γ
s
If γ −1 is the inverse path γ −1 (t) = γ(1 − t), then H : γγ −1 ≡ p, where


γ(2t)
 2t ≤ 1 − s t
γ −1
H(t, s) = γ(1 − s) 1 − s ≤ 2t ≤ 1 + s p
γ

 −1
γ (2t) 2t ≥ 1 + s s
and composition of paths is associative, H : (γ1 γ2 )γ3 ≡ γ1 (γ2 γ3 ), where


4t
 γ 1 1+s t ≤ 1+s
4 γ 3

γ3

1+s 2+s γ2
H(t, s) = (4t − s −
γ2 1) 4 ≤t≤ 4 γ2
γ1

γ 4t−s−2
 2+s
≤t γ1
3 2−s 4
When p = q, the path is a loop at p. Homotopy classes of loops at p, with the composition
of maps, form a group π1 (X, p), the fundamental group of X at p.
Any continuous map f : X → Y naturally induces a morphism of groups

f∗ : π1 (X, p) −→ π1 Y, f (p) , f∗ [σ] = [f ◦ σ],
well defined: if H : σ0 ≡ σ̄, a homotopy f ◦ σ ≡ f ◦ σ̄ is H 0 (t, s) = f (H(t, s)).

Any path γ from p to q induces a group isomorphism (non canonical, it depends on the path,
except when the group is abelian)
∼
γ : π1 (X, p) −−→ π1 (X, q), [σ] 7→ [γ −1 σγ].
Two continuous maps f 0 , f : X → Y are said to be homotopic if there is a continuous map

H : X × I → Y such that
f (x) = H(x, 0) , f 0 (x) = H(x, 1).
This is an equivalence relation compatible with the composition of maps, so that topological
spaces, with homotopy classes of continuous maps, form a category.
Isomorphisms in this category are named homotopical equivalences, and a space is con-
tractible if it is homotopically equivalent to a point.
Lemma: If γ is the path γ(s) = H(p, s), the following square is commutative
f∗
π1 (X, p) −−→ π Y, f (p)
|| o|γ
f∗0
π1 (X, p) −−→ π Y, f 0 (p)

σ×1 H
Proof: If we consider F : I × I −−−−→ X × I −−→ Y , then
F (t, 0) = f (σ(t)) , F (t, 1) = f 0 (σ(t)),

F (0, s) = F (1, s) = γ(s).
Since the upper side of the square I × I is homotopic to the path formed by the three
remaining sides, composing with F we obtain that γ −1 (f ◦ σ)γ ≡ f 0 ◦ σ.
Theorem: If f : X → Y is a homotopical equivalence, then f∗ : π1 (X, p) → π1 (Y, f (p)) is an

isomorphism. In particular π1 (X, p) = 0 when X is contractible.
Definition: X is simply connected if it is path-connected (paths join any pair of points) and
π1 (X, p) = 0, and it is locally simply connected if any point has a base of simply connected
open neighborhoods (for example, any topological manifold).
Lemma: Let X → S be a trivial covering, and let f : T → S be a continuous map, T connected.

Any continuous lifting f˜: C → X of f , defined on a connected subspace C ⊂ T , admits a unique
extension to T .
Proof: Extensions of f correspond to continuous sections of X ×S T → T ; but the image of a

continuous section defined on a connected subspace is contained in a connected component of
X ×S T = qT , and now the statement is obvious.
Lemma: Let π : X → S be a covering, and let γ : I → S be a path with origin at p. If π(x) = p,

there exists a unique continuous lifting γ̃ : I → X of γ with origin at x.
Moreover, if γ ≡ γ 0 , then γ̃ ≡ γ̃ 0 .
Proof: The existence and uniqueness of the lifting γ̃ follows from the above lemma, considering
a partition of I with images contained in open sets of S where X is trivial.
Moreover, the existence of a lifting of any homotopy H : I × I → S between γ and γ 0 also

follows from the above lemma, considering a partition of I ×I into squares with images contained
in open sets of S where X is trivial, since the intersection of any square with the union of the
former squares is connected.
Lemma: Any covering X → S of a simply connected locally path-connected space is trivial.
Proof: It is enough to show that any connected covering X has degree 1.

If x, y are two points of the fibre of p, projecting onto S a path joining x to y we have a loop
at p which is not homotopic to a point, since the lifting is not a loop.
Theorem: Let S be a connected and locally path-connected space. If S̃ → S is a simply con-

nected covering, then it is the universal covering, and any point p̃ of the fibre of p defines an
isomorphism
π1 (S, p) = Aut(S̃/S).
Proof: Any loop σ has a lifting σ̃ with origin at p̃ and end point at τ (p̃), τ ∈ Aut(S̃/S).
The map π1 (S, p) → Aut(S̃/S), [σ] 7→ τ , is well defined (by the former lemma), it is surjective
(S̃ is path-connected, and any two points of the fibre of p may be joined by a path), it is injective
(any loop in S̃ is homotopic to a point) and it is a group morphism: If σ̃1 joins p̃ with τ1 (p̃),
and σ̃2 joins p̃ with τ2 (p̃), then the lifting of σ1 σ2 is σ̃1 · τ1 (σ̃2 ), with end point τ1 (τ2 (p̃)).
Corollary: π1 (S1 , p) = Z, generated by the loop e2πti : [0, 1] → S1 .
Proof: The map e2πti : R → S1 is a simply connected covering. q.e.d.
1. There is no continuous retract r : D2 → S1 of a disc onto the boundary.

i r
Z = π1 (S1 ) −−∗→ π1 (D2 ) = 0 −−∗→ π1 (S1 ) = Z would be the identity.
2. Brouwer’s Theorem: Any continuous map f : D2 → D2 has a fixed point.

Otherwise we have a continuous retract r : D2 → S1 , where r(x) is the intersection point of
S1 with the half-line with origin at x passing through f (x).
3. Any non constant polynomial with complex coefficients has a complex root.
If P (z) = z n + a1 z n−1 + . . . + an has no complex root, it defines a homotopy
n n n−1 n−1
H(z, s) = sn P 1−s + . . . + an s n

s z = (1 − s) z + a1 s(1 − s) z
between the continuous maps h0 (z) = z n and h1 (z) = an : S1 → C∗ . But h1,∗ = 0, while
h0,∗ : Z = π1 (S1 ) → π1 (C∗ ) = π1 (S1 ) = Z is the multiplication by n.
4. Two toruses C/Γ, C/Γ0 are analytically isomorphic if and only if Γ0 = aΓ, a ∈ C.
An isomorphism C/Γ0 ' C/Γ induces an isomorphism τ : C ' C of the universal coverings,
and we may assume that τ (0) = 0. Now, τ (z) = az since any analytic automorphism of C is
an affinity (p. 269).
The Universal Covering: Assume that S is a connected and locally simply connected space,
and let S̃ be the set of homotopy classes of paths with origin at p ∈ S, and put π : S̃ → S,
π(γ) = γ(1). Over any simply connected neighborhood Uq of q ∈ S we have a bijection

−1 homotopy classes
π (Uq ) = Uq × Fq , where Fq =
of paths from p to q
transforming [γ] into the pair q 0 = γ(1), [γα−1 ] , where α is a path in Uq joining q with q 0 (it

does not depend on α since all choices are homotopic).
These bijections define topologies on the open sets π −1 (Uq ), considering Fq as a discrete
space, so that π is a covering (trivial over Uq ). In fact, when Uq ⊆ Uq̄ , the topology defined on
π −1 (Uq ) coincides with the topology induced by π −1 (Uq̄ ); i.e., the composition
Uq × Fq = π −1 (Uq ) = Uq × Fq̄
is a homeomorphism, since it transforms the component Uq × [γ] into Uq × [γβ], where β is a

path in Uq̄ joining q with q̄. Moreover, the constant loop p̃ is in the fibre of p.
Galois Theorem: Let S be a connected and locally simply connected space. The space π : S̃ → S
of paths with origin at p is a simply connected covering, with a fixed point p̃ over p; hence it is
the universal covering of S, and the functor “fibre over p” defines an equivalence of the category
of coverings of S with the category of π1 (S, p)-sets.
Proof: If γ is a path in S with origin at p, the lifting γ̃ with origin at p̃ is γ̃(s) = [γs ],
(
γ(t) t ≤ s
γs (t) =
γ(s) t ≥ s
and the end point is γ̃(1) = [γ1 ] = [γ]; hence S̃ is path-connected.

Moreover, any loop σ̃ at p̃ is the lifting σ = πσ̃, so that [σ] = σ̃(1) = p̃, and the lifting σ̃ is
homotopic to the constant loop; i.e., S̃ is simply connected.
Finally, since we have fixed a point p̃ over p, the group of automorphisms is canonically
isomorphic to π1 (S, p), and HomS (S̃, X) may be identified with the fibre Xp of any covering X
over p: for any x ∈ Xp there is a unique morphism f : S̃ → X such that f (p̃) = x.
Corollary: If a connected locally simply connected space S = U1 ∪ U2 is the union of two simply
connected open sets and U1 ∩ U2 is connected, then S is simply connected.
Proof: Let us show that any covering of S admits a continuous section.

We have some section on U1 since it is simply connected, and the induced section on the
connected set U1 ∩ U2 may be extended to U2 by the lemma of p. 241.
Corollary: π1 (Sn , p) = 0, and π1 (Pn,R , p) = Z/2Z; when n ≥ 2.
Proof: We have a covering Sn → Pn,R of degree 2, and Sn is simply connected because it is the
union of two disks with connected intersection when n ≥ 2. q.e.d.
1. R2 is not homeomorphic to Rn when n 6= 2.

π1 (R2 − p) = π1 (S1 ) = Z, and π1 (Rn − p) = π1 (Sn−1 ) = 0 when n ≥ 3.
2. There is no continuous map f : Sn → S1 , n ≥ 2, preserving antipodes.

If f (−x) = −f (x), it induces a continuous map h : Pn → P1 and a commutative square
f
Sn / S1

Pn
h / P1
If we take a path γ̄ joining two antipodal points x, −x ∈ Sn , the image γ is a loop in Pn ;

hence h(γ) is homotopic to a point, since h∗ : Z/2Z = π1 (Pn ) → π1 (P1 ) = Z is null, and the
lifting f (γ̄) is a loop; absurd since f (−x) = −f (x) 6= f (x).
3. Any continuous map Sn → S2 , n ≥ 2, preserving antipodes is surjective.

If the north pole is out of the image, so is the south pole, and projecting onto the equator
along the meridians we obtain a continuous map Sn → S1 preserving antipodes.
4. Borsuk-Ulam Theorem: Any continuous map f : Sn → R2 , n ≥ 2, coincides on two

antipodal points.
f (x)−f (−x)
Otherwise the continuous map |f (x)−f (−x)| : Sn → S1 preserves antipodes.
5. Any continuous map f : Sn → R2 , f (−x) = −f (x), vanishes at some point.
6. No compact set in R2 is homeomorphic to S2 .
Definition: A principal covering of group G is a covering P → S (not necessarily connected)

with a free G-action transitive on any fibre; i.e., the action defines an isomorphism
∼
G × P −−→ P ×S P, (g, x) 7→ (x, gx).
Isomorphisms of principal coverings are isomorphisms of coverings which are also isomor-
phisms of G-sets.
If we fix a point of the fibre of p ∈ S, we say that it as a pointed principal covering.
When S is connected, isomorphisms of pointed principal coverings are unique.
Proposition: If S is a connected and locally simply connected space,

Pointed principal
Homgr (π1 (S, p), G) =
G-coverings of S
Proof: Principal G-coverings of S correspond to (say right) π1 (S, p)-sets F with a free transitive
G-action, g(xγ) = (gx)γ. Once we fix a point x ∈ F , the group G may be identified with F ,
and such actions correspond to morphism of groups f : π1 (S, p) → G, xγ = f (γ)x,
f (γ1 γ2 )x = xγ1 γ2 = f (γ1 )xγ2 = f (γ1 )f (γ2 )x.

Principal coverings
Corollary: Homgr (π1 (S, p), A) = , when A is an abelian group.
of S of group A
Proof: Principal G-coverings of S correspond to morphisms π1 (S, p) → G, modulo inner auto-

morphisms of G. In fact, if we fix another point y = gx in the fibre of p, the corresponding
morphism is gf g −1 ,
yγ = gxγ = gf (γ)x = gf (γ)g −1 y.
Van-Kampen Theorem: If U1 , U2 are two connected and locally simply connected open sets,
and U1 ∩ U2 is connected, then the fundamental group of the union is the coproduct (in the
category of groups) of the fundamental groups,
π1 (U1 ∪ U2 , p) = π1 (U1 , p) ∗π1 (U1 ∩U2 ,p) π1 (U2 , p).
Proof: Since isomorphisms of pointed principal coverings are unique, to give a pointed principal
G-covering over U1 ∪U2 is just to give one over each open set Ui , coinciding over the intersection.
That is to say, we have a fibred product
Homgr (π1 (U1 ∪ U2 , p), G) / Homgr (π1 (U1 , p), G)

Homgr (π1 (U1 , p), G) / Homgr (π1 (U1 ∩ U2 , p), G)
8.8.2 Triangulated Compact Surfaces

A triangulated compact surface X is a connected 2-dimensional polyhedron (a connected finite
union of vertices, closed edges and faces of a tetrahedron |∆n |) such that each edge only belongs
to two faces and the faces with any given vertex admit a cyclic ordering such that each face has
a common edge with the following one (hence with the former one). Given two triangulated
compact surfaces X, Y we put X ≡ Y when there is some homeomorphism X → Y .
Since X is connected, we may order the faces T0 , . . . , Tn so that each triangle Ti , i ≥ 1, has
a common edge with a former one. Inductively, one sees that T0 ∪ . . . ∪ Ti is homeomorphic
(with a homeomorphism preserving the triangulations) to a triangulated plane polygon with
an even number of sides, where the sides are identified by pairs. If we fix an orientation of
the boundary of such polygon, put a letter a±1 on each pair of identified sides (with the same
exponent when the identification preserves the orientation, and different exponent otherwise)
and read the letters starting from a vertex of the boundary, we see that X is fully determined
by a symbol a±1 ±1
1 . . . an , where each letter appears exactly two times.
A pair of first kind is a pair . . . x . . . x−1 . . . ≡ . . . x−1 . . . x . . ., and a pair of second kind is a
pair . . . x . . . x . . . ≡ . . . x−1 . . . x−1 . . .
For example, the symbol aa−1 represents a sphere, the symbol aa a projective plane, the
symbol aba−1 b−1 a torus, and the symbol aba−1 b a Klein bottle.
But different symbols may define homeomorphic compact surfaces. Apart of the rule that
we may replace a letter by any new symbol, we have the following transformation rules:
1. Cyclic permutation: a1 a2 . . . an ≡ a2 . . . an a1 .
2. Reverse orientation of a side: AxBxC ≡ Ax−1 Bx−1 C, AxBx−1 C ≡ Ax−1 BxC.
3. The pair xx−1 may be simplified : Axx−1 B ≡ AB.

4. Cyclic permutation inside a pair of first kind : AxBCx−1 D ≡ AxCBx−1 D. Hence, also cyclic
permutations outside a pair of first kind:
1 4 1
AxBCx−1 D ≡ x−1 DAxBC ≡ x−1 ADxBC ≡ DxBCx−1 A.
5. Grouping a pair of second kind : AxBxC ≡ AxxB −1 C, AxBxC ≡ AB −1 xxC, where we put
−1
(b1 . . . bs )−1 = b−1
s . . . b1 .
Given two symbols A and B, we say that the surface AB is the connected sum of A and
B, and it is obtained removing a small open disk in each surface A, B and then identifying the
boundaries of the disks (and rule 3 states that the connected sum of a surface A with a sphere
is just the surface A):
Lemma: The connected sum of a projective plane and a torus is the connected sum of three
projective planes,
Axxaba−1 b−1 B ≡ AxxbbaaB.
5 5
Proof: Axxaba−1 b−1 B ≡ Axa−1 xba−1 b−1 B ≡ Axa−1 a−1 b−1 x−1 b−1 B
5 5
≡ Axa−1 a−1 b−1 b−1 xB ≡ AxxbbaaB.
Theorem: Any triangulated compact surface is homeomorphic to one (and only one) of the
following surfaces:
1. A sphere, aa−1 .
2. A connected sum τg of g toruses, a1 b1 a−1 −1 −1 −1

1 b1 . . . ag bg ag bg .
3. A connected sum πg of g projective planes, a1 a1 . . . ag ag .
Proof: By the lemma, we only have to show that any symbol is equivalent to a sphere xx−1 or
to a connected sum of projective planes and toruses:
xx . . . zzaba−1 b−1 . . . cdc−1 d−1 .
1. All pairs of second kind may be grouped at the beginning, so that the symbol is xx . . . zzA,
where A is a symbol with no pairs of second kind.
5 5 5
AxBxC ≡ AxxB −1 C ≡ xA−1 xB −1 C ≡ xxAB −1 C.
Now, if there is another second kind pair, we iterate:

5 5 5
xxAyByC ≡ xxAyyB −1 C ≡ xxyA−1 yB −1 C ≡ xxyyAB −1 C.
2. If a symbol A has no pairs of second kind, then XA ≡ Xaba−1 b−1 . . . cdc−1 d−1 .
If the length l of A is 2, then A = aa−1 , and XA ≡ X. If l ≥ 4, consider a pair a . . . a−1
at minimal distance. If they are adjacent, cancel them with rule 3. Otherwise, there is some
letter b between them, a . . . b . . . a−1 , and the other letter b±1 in A is not between the pair
a . . . a−1 , since they are at minimal distance. There are two possible cases
X . . . a . . . b . . . a−1 . . . b−1 . . . or X . . . b−1 . . . a . . . b . . . a−1 . . .
and both are equivalent by rule 2 (change b by b−1 ). Now we repeatedly apply rule 4,
underlining the cyclic permutation:
4 4
X . . .a . . . b . . . a−1 . . . b−1 . . . ≡ . . . Xa . . . b. . . a−1 . . .b−1 . . . ≡ . . . Xa. . . b . . .a−1 b−1 . . .
4 4 1
≡ . . . Xa . . .ba−1 . . . b−1 . . . ≡ . . . Xaba−1 b−1 . . . ≡ Xaba−1 b−1 . . .
Finally, let us see that all these surfaces S2 , τg and πg are not homeomorphic, because they
have different fundamental group. Consider a small open disk D around the center p of the
polygon, and let U be the complement of p. Then U is homotopic to the image of the boundary
of the polygon, and the image of the natural map Z = π1 (U ∩ D) → π1 (U ) is generated by the
loop defined by the symbol.
Since π(D) = 0, the Van-Kampen theorem shows that the fundamental group of a surface
of symbol A is isomorphic to the quotient of the fundamental group of the boundary (with the
sides identified according to A) by the normal subgroup generated by the symbol.
In the case of τg (resp. πg ), the boundary are 2g (resp. g) circles with a common point.
In particular, the universal abelian quotient (the quotient by the normal subgroup generated
by the commutators ghg −1 h−1 ) is
π1 (S2 )ab = 0,
π1 (τg )ab = Z2g /( + bi − ai − bi )Z = Z2g ,
P
i ai
π1 (πg )ab = Zg /(2a1 + . . . + 2ag )Z ' Zg−1 ⊕ (Z/2Z).
Example: The Klein bottle K has symbol aba−1 b; hence π1 (K)ab = Z2 /2bZ ' Z ⊕ (Z/2Z), so
that K is homeomorphic to the connected sum π2 of two projective planes.
Chapter 9
Analysis III
9.1 Rings of Smooth Functions

Definitions: Let X be a topological space. Two continuous real functions, defined on some
neighborhoods of x ∈ X, have equal germ at x if they coincide on a neighborhood of x.
The ring Ox of germs at x of continuous functions is defined to be
Ox = lim
−→
C(U )
x∈U
where U runs over the open neighborhoods of the point x, and the germ at x of a continuous
function f ∈ C(U ) will be denoted by fx . The support of a function f : X → R is
supp f = {x ∈ X : f (x) 6= 0} = {x ∈ X : fx 6= 0}.

S
A partition of unity subordinate to an open cover X = i Ui is a family of continuous
real functions φi ∈ C(X) such that
1. For any index i, we have supp φi ⊆ Ui , and φi ≥ 0.
2. The family {supp φi } is locally finite (finite in a neighborhood of each point).

P
3. 1 = i φi , (the sum is senseful by condition 2).
Lemma: Let K be a compact Hausdorff space. If V is a neighborhood of p ∈ K, there is a non

negative continuous function f : K → R such that f (p) > 0 and supp f ⊆ V .
Proof: By Urysohn’s lemma, if Q ⊆ V is a compact neighborhood of p, there exists f ∈ C(K)

such that f (Qc ) = 0, f (p) > 0.
Theorem: Any open cover of a σ-compact space admits a subordinate partition of unity.
S o o
Proof: Put X = n Kn , with Kn compact and Kn ⊆K n+1 (p. 94), and Qn := Kn − K n−1
o
If p ∈ Qn , there is a continuous function f ≥ 0 with support in Ui ∩ (K n+1 −Kn−2 ) not
vanishing at p, and we choose a finite number of these functions not vanishing simultaneously
at any point of Qn .
P
Varying n weS obtain functions fj such that the family {supp fj } is locally finite
P and h = j fj
has no zero in n Qn = X. Replacing fj by fj /h, we may assume that 1 = j fj .
P
Now, for any index j, fix σ(j) such that supp fj ⊆ Uσ(j) , and put φi = fj .
σ(j)=i
249
250 CHAPTER 9. ANALYSIS III
Lemma: If V is a neighborhood of the origin in Rn , there is a smooth function f ∈ C ∞ (Rn ),

non negative and with compact support contained in V , such that f (0) > 0.
Proof: The neighborhood of radius ε is the support of f (x1 , . . . , xn ) = e(x21 + . . . + x2n − ε2 ),

where e(t) is the C ∞ function
(
e1/t t < 0
e(t) =
0 t≥0
Now, the above proof (smooth manifolds are always assumed to be σ-compact) gives:
Theorem: Any open cover of a smooth manifold admits a C ∞ subordinate partition of unity.
Corollary: Let Y1 , Y2 be disjoint closed sets of a smooth manifold X. There exists a smooth
global function 0 ≤ f ≤ 1 such that f (Y1 ) = 0, f (Y2 ) = 1.
Proof: Let φ1 , φ2 be a partition of unity subordinate to the open cover X = U1 ∪ U2 , where

Ui = X − Yi . Now f = φ1 .
Corollary: Let U be a neighborhood of x ∈ X. There is a function φ ∈ C ∞ (X), such that φ = 1

on a neighborhood of x, supp φ ⊆ U , and 0 ≤ φ ≤ 1.
Proof: There exists φ ∈ C ∞ (Rn ), 0 ≤ φ ≤ 1, such that φ = 1 on the closed neighborhood of

radius ε, and vanishing on the complement of the open neighborhood of radius 2ε.
Corollary: If f ∈ C ∞ (U ) and x ∈ U , there exists F ∈ C ∞ (X) with equal germ, fx = Fx .
Proof: If φ is a plateau function, and we extend φf by 0 on X − U , we obtain a global smooth

function F with equal germ at x.
Corollary: There is a smooth function f : X → R such that f −1 ([a, b]) is compact, ∀a, b ∈ R.
S o P
Proof: Put X = n Kn , Kn ⊂K n+1 , and take f = n hn , where hn is a non negative smooth
o
function such that hn (Kn−1 ) = 0 and hn (X− K n+1 ) = 1.
Definitions: The maximal spectrum of a ring A is the subspace Specm A ⊆ Spec A of all
maximal ideals. The real spectrum of a R-algebra A is the subspace SpecR A ⊆ Specm A of
all real maximal ideals m (the natural morphism R → A/m is an isomorphism).
When A is a subalgebra of C(X), we have a surjective morphism A → R, f 7→ f (x), for any
point x ∈ X; hence the kernel mx = {f ∈ A : f (x) = 0} is a maximal ideal of residue field R.
This map X → SpecR A is continuous, since zeros of continuous functions
z(f ) = (f )0 ∩ X = {x ∈ X : f (x) = 0}
are closed sets. It is injective when functions in A separate points (if x 6= y, then f (x) 6= f (y)
for some f ∈ A), and it is a homeomorphism onto the image when moreover any closed set in X
is an intersection of zeros; i.e., when functions in A separate points from closed sets: if x ∈/ Y,
then there is f ∈ A such that f (Y ) = 0, f (x) 6= 0.
Theorem: K = Specm C(K), when K is a compact Hausdorff space.

9.1. RINGS OF SMOOTH FUNCTIONS 251
Proof: Let m be a maximal ideal of C(K). Finite intersections of zeros of functions in m are non
void since z(f1 ) ∩ . . . ∩ z(fn ) = z(f12 + . . . + fn2 ), and f12 + . . . + fn2 ∈ m is not invertible.
Hence there is a point x ∈ K where all the functions in m vanish, m ⊆ mx .
Finally, m = mx since m is a maximal ideal.
Theorem1 : X = SpecR C ∞ (X), when X is a smooth manifold.
Proof: Let us consider f ∈ C ∞ (X) with compact fibres. If C ∞ (X)/m = R, then f − a ∈ m for
some a ∈ R. Since finite intersections of zeros of functions in m are non void, and z(f − a) is
compact, all the functions in m vanish at some x ∈ z(f − a); hence m = mx .
Theorem: Hom(X, Y ) = HomR-alg C ∞ (Y ), C ∞ (X) , for any two smooth manifolds X, Y .

Proof: A smooth map φ : X → Y induces a morphism of R-algebras φ∗ : C ∞ (Y ) → C ∞ (X),

φ∗ (f ) = f φ. On the other hand, any morphism of R-algebras C ∞ (Y ) → C ∞ (X) induces a
continuous map φ : X = SpecR C ∞ (X) → SpecR C ∞ (Y ) = Y , and it is smooth.
In fact, φ transforms smooth functions on Y into smooth functions on X, and if f is a smooth
function on an open set V ⊂ Y , and x ∈ φ−1 (V ), then f coincides on a neighborhood of φ(x)
with a smooth function F ∈ C ∞ (Y ); hence φ∗ (f ) coincides with the smooth function φ∗ (F ) on
a neighborhood of x, and φ∗ (f ) ∈ C ∞ (φ−1 V ).
Definition: Let X be a smooth manifold. If K is a compact set contained in a coordinate open

set (U ; x1 , . . . , xd ), we may consider on C m (X), 1 ≤ m ≤ ∞, the seminorms

1
kf kK,r = max k∂α f kK ; r ≤ m, r < ∞,
|α|≤r α!
(see p. 98 for notations). The seminorms pn of a countable family of compact sets contained in
open neighborhoods, whose interior sets cover X, define a linear topology on C m (X), independent
of the compact cover, and C m (X) is metrizable and complete, 1 ≤ m ≤ ∞ (p. 115).
Replacing pn by max{p1 , . . . , pn }, we may always assume that pn−1 ≤ pn .
Theorem: Let U be an open set in a σ-compact space X. Any function f ∈ C(U ) is a quotient,
f = g/h, of two continuous global functions, z(h) = X − U . Hence C(U ) = C(X)S , where S is
the multiplicative system of all functions without zeros in U .
S o
Proof: Put U = n Kn , Kn ⊆K n+1 , and consider the seminorms pn = k kKn on C(X).
o
Let 0 ≤ φn ∈ C(X) such that φn (X− K n+1 ) = 0, φn (Kn ) = 1.
Since supp φn f ⊆ Kn+1 , the function φn f is continuous when extended by 0 outside of U ,
and we may consider the convergent series
∞
X 1 φn f
g= ·
2n (1 + pn (φn )) 1 + pn (φn f )
n=1
∞
X 1 φn
h= ·
2n (1 + pn (φn f )) 1 + pn (φn )
n=1
1
This theorem shows that the R-algebra C ∞ (X) reconstructs the topological space X; but it also determines
the sheaf of smooth functions, since it clearly determines global smooth functions, and smooth functions on an
open set U ⊂ X are just continuous functions locally coinciding with global smooth functions. Hence, the functors
C ∞ and SpecR define an antiequivalence of the category of smooth (σ-compact) manifolds with a certain category
of R-algebras (namely, R-algebras isomorphic to some C ∞ (X)). And the former theorem shows that the category
of compact Hausdorff spaces is antiequivalent to a certain category of R-algebras.
where h vanishes exactly at X − U . Now, this epimorphism C(X)S → C(U ) is injective:

If a fraction as vanishes on U , then ha = 0, so that as = 0 in C(X)S . q.e.d.
This argument also proves the analogous result for smooth manifolds:
Theorem: Let U be an open set in a smooth manifold X. Any smooth function f on U is a

quotient, f = g/h, of two smooth global functions, z(h) = X − U . Hence C ∞ (U ) = C ∞ (X)S ,
where S is the multiplicative system of all smooth functions without zeros in U .
9.2 Ordinary Differential Equations

Definition: If D is a continuous vector field on a smooth manifold X, a derivable curve σ : I → X
is an integral curve of D if the tangent vector Tt ∈ Tσ(t) X coincides with Dσ(t) at any t ∈ I.
P
In a coordinate neighborhood we have D = i fi (x1 , . . . , xn )∂i , where the functions fi are
continuous, and σ(t) = (x1 (t), . . . , xn (t)) is an integral curve if the functions xi (t) define a
solution of the system of differential equations σ 0 (t) = f (σ(t)), where f = (f1 , . . . , fn ),

0
x1 = f1 (x1 , . . . , xn )

............

 0
xn = fn (x1 , . . . , xn )
In particular σ 0 is continuous, and if f is C m , so is σ 0 , and the solutions are C m+1 .

If we impose an initial condition σ(t0 ) = x, the equality σ 0 = f (σ) states that
Z t
σ(t) = x + f (σ(t))dt.
t0
Lemma: Any contractive map T of a complete metric space E (i.e., there is a constant k < 1
such that d(T σ, T ϕ) ≤ kd(σ, ϕ) for all σ, ϕ ∈ E) has a unique fixed point.
Proof: If σ0 ∈ E, then σn = T n (σ0 ) is a Cauchy sequence,
d(σn , σn+1 ) ≤ kd(σn−1 , σn ) ≤ . . . ≤ k n d(σ0 , σ1 ) = k n c,

n
d(σn , σn+m ) ≤ c(k n + . . . + k n+m−1 ) ≤ c 1−k
k
< ε when n 0,
and σ = lim σn is a fixed point, T (σ) = lim T (σn ) = lim σn+1 = σ.

There is no other fixed point since d(σ, T ϕ) = d(T σ, T ϕ) < d(σ, ϕ).
Theorem: If a vector field is C 1 , then through any point x passes an integral curve σ such that
σ(t0 ) = x, and it is unique (any two coincide on the common definition interval).
Proof: If X = Rn and supp D is compact, the continuous functions ∂fi /∂xj have compact
support and by the mean value theorem there is a constant k such that |f (x) − f (y)| ≤ k|x − y|.
We put I = [t0 − ε, t0 + ε] and we consider on E = C(I, Rn ) the maximum module norm.
Solutions are just fixed points of the following operator T : E → E,
Z t

T (σ) = x + f (σ(t) dt,
t
Z 0 t

kT (ϕ) − T (σ)k = sup [f (ϕ) − f (σ)] dt ≤ εk sup |ϕ(t) − σ(t)| ≤ kεkϕ − σk,

t∈I t0 t∈I
9.2. ORDINARY DIFFERENTIAL EQUATIONS 253
and it is contractive when kε < 1. This proves the existence of integral curves, and that any
two coinciding at some instant, coincide on a neighborhood.
Since this statement is local, it also holds in any smooth manifold X, and since the coinci-
dence locus is closed, any two solutions coincide on the common definition interval.
Definition: Through any point x ∈ X passes a maximal integral curve σx : Ix → X, σx (0) = x.

The flow τ of the vector field D is well-defined on the subspace W = {(t, x) : t ∈ Ix } ⊆ R × X,
τ : W −→ X, τ (t, x) = τt (x) = σx (t).
We have τ0 (x) = x, and τt+s (x) is an integral curve passing by y = τs (x) at t = 0, so that
Ix ⊆ Iy + s. Since x = τ−s (y), then Iy ⊆ Ix − s; hence Iy + s = Ix , and
τt (τs x) = τt+s (x).
Lemma: The flow of a C 1 vector field is continuous on a neighborhood of 0 × X.
Proof: Since it is a local question, we may assume that supp D is compact and X = Rn .
We fix a compact Λ ⊂ Rn (a cube, ball,...) and we put I = [−ε, ε].
We consider on E = C(I × Λ, Rn ) the maximum module norm, and we repeat the argument
of the previous theorem with the initial condition σ(0, x) = x,
Z t
T (σ) = x + f (σ(t, x))dt,
0
Z t

kT ϕ − T σk = sup [f (ϕ) − f (σ)] dt ≤ kεkϕ − σk.

(t,x)∈I×Λ 0
When kε < 1, the map T is contractive, and the fixed point provides a continuous family of
solutions τ : (−ε, ε) × Λ → Rn , such that τ (0, x) = x.
Theorem: If a vector field is C 1 , then W ⊆ R × X is open, and τ is continuous.
Proof: Let us show that τ is defined and it is continuous on a neighborhood of any point of
W . Otherwise, we fix a point p ∈ X and the first positive instant c ∈ Ip where it is false (the
negative case being analogous), and we put q = τc (p).
By the above lemma, τ is continuous on a neighborhood (−2ε, 2ε) × V of (0, q).
On a neighborhood U of p, the map τc−ε : U → X is continuous, and we may assume that
τc−ε (U ) ⊆ τ−ε (V ); hence (c − ε, c + ε) × U ⊂ W , and τ is continuous on this open set (against
the choice of c) since τc = τε τc−ε : U → V is continuous, and
τ (t, x) = τ (t − c, τc (x)).
t-eV V teV
U
p q
tc-e (p)
Lemma: The flow of a C m+1 vector field, m < ∞, is C m on a neighborhood of 0 × X.
Proof: Since it is a local question, we may assume that supp D is compact and X = Rn .
Let Λ ⊂ Rn be an open ball of radius r 0, I = [−ε, ε], and let E be the complete metric
space of all C m maps σ : I × Λ → Rn such that
kσk = sup |Dα σ| ≤ 2r.

(t,x)∈I×Λ
0≤|α|≤m
Let us see the case m = 1. We put x0 = t, and for any i = 0, . . . , n we have

n
X ∂f ∂σj
Di (f (σ1 , . . . , σn )) = (σ) · ,
∂xj ∂xi
j=1
and, since the maps ∂f /∂xj are C1, there is a constant k such that
1. kf (σ)k ≤ 2kr, when kσk ≤ 2r.
2. kf (ϕ) − f (σ)k ≤ k kϕ − σk, when kϕk, kσk ≤ 2r.

1
Now we repeat the argument of the contractive map, and we put ε = 2k ,
Z t Z t

kT (σ)k ≤ r + f (σ) dt ≤ r +
kf (σ)k dt ≤ r + 2rkε = 2r,
0 0
Z t Z t
kf (ϕ) − f (σ)k dt ≤ kεkϕ − σk = 12 kϕ − σk.

kT ϕ − T σk =
[f (ϕ) − f (σ)] dt ≤

0 0
In general, when m > 1, we have that Dα (f (σ)) is a polynomial on the derivatives of σ of

order ≤ m with coefficients derivatives of f of order ≤ m (which are C 1 -functions with compact
support), and there is also a constant k with properties 1 and 2.
Theorem: If a vector field is C m+1 , 1 ≤ m ≤ ∞, then W ⊆ R × X is open and the flow is C m .
Proof: We may repeat the argument of the continuous case.
Corollary: Tangent vector fields with compact support are complete, W = R × X.
Proof: If Dx = 0, the constant map x : R → X is an integral curve, and Ix = R.

If D = 0 outside a compact set, W being open, then (−2ε, 2ε) × X ⊂ W for some ε > 0, and
any integral curve may be extended for a time ε; hence Ix = R at any point x.
9.2.1 Uniparametric Groups and Lie Derivative

Definitions: A smooth map τ : R × X → X, (t, x) 7→ τt (x), is a uniparametric group if
1. τ0 (x) = x.
2. τt (τs x) = τt+s (x).
(It is a smooth action of the group R on X). If it is only defined on an open neighborhood of
0 × X, intersecting any line R × x in an interval, we say that it is a local uniparametric group
(condition 2 is assumed only whenever both terms are defined).
The infinitesimal generator of τ is the vector field D such that Dp is the tangent vector
to the curve τt (p) at the instant t = 0, so that it is defined by the derivation

∂(f ◦ τ ) f (τt p) − f (p)
Df = , Dp f = (Df )(p) = lim ·
∂t t=0 t→0 t
Hence, the flow of any vector field D is a local uniparametric group with infinitesimal genera-
tor D. As well, any local uniparametric group is (an open subset of) the flow of the infinitesimal
generator D, since the curve σ(t) = τt (p) is an integral curve of the vector field D passing
through p at the instant t = 0. In fact, since σ(t + ε) = τε (σ(t)), the tangent vector to σ at
the instant t is Dσ(t) . Vector fields correspond to (maximal) local uniparametric groups and, on
compact manifolds, to uniparametric groups.
∂
Local Classification of Vector Fields: If Dp 6= 0, then we have D = ∂u1 in some local
coordinate system (u1 , . . . , un ) at p.
Proof: Since it is a local question, we may assume that p is the origin of Rn and that supp D is
compact. We may also assume that Dp is not tangent to the hyperplane H of equation x1 = 0.
Now, the smooth map
R × H −→ Rn , (t, x) 7→ τt (x),
∂ ∂
transforms integral curves of ∂t into integral curves of D; hence it transforms ∂t into D.
We conclude since this map is a local diffeomorphism at p: the tangent map at p transforms
∂
( ∂t )p into Dp , and it is the identity on Tp H.
Definition: The Lie bracket of two vector fields D, D0 on a manifold X is the vector field
defined by the derivation (check that so it is)
[D, D0 ] = D ◦ D0 − D0 ◦ D
P P P
i fi ∂i , i hi ∂i = ij (fj ∂j hi − hj ∂j fi )∂i
1. [D, D0 ] = −[D0 , D], and therefore [D, D] = 0.
2. [D, f D0 ] = (Df )D0 + f [D, D0 ].
3. [[D1 , D2 ], D3 ] + [[D2 , D3 ], D1 ] + [[D3 , D1 ], D2 ] = 0 (Jacobi’s Identity)
Definition: The Lie derivative of a tensor field T with a vector field D of flow {τt } is the
following tensor field (a calculation in coordinates shows that it is C ∞ , since so is the flow)
(τt∗ T )x − Tx
(DL T )x = lim ·
t→0 t
Theorem: If DL T = 0, then the tensor field T is invariant, τt∗ T = T .

If DL T = f T , then τt∗ T and T are proportional, hτt∗ T i = hT i.
Proof: Let us consider in Tx X the curve σ(t)= (τt∗ T )x . The tangent vector at t = 0 is (DL T )x ,
and the tangent vector at any instant t is τt∗ (DL T )τt x since
∗
T )x = τt∗ (τε∗ T )τt x .

σ(t + ε) = (τt+ε
If DL T = 0, then σ 0 (t) = 0, and the curve is constant, τt∗ T = T .

If DL T = f T , then σ 0 (t) = h(t)σ(t), where h(t) = f (τt x). Hence τt∗ T is proportional to T ,
Rt
σ(t) = eH(t) σ(0), where H(t) = 0 h(t)dt. q.e.d.
1. DL f = Df .
2. DL (T + T 0 ) = DL T + DL T 0 .
3. DL (T ⊗ T 0 ) = (DL T ) ⊗ T 0 + T ⊗ (DL T 0 ).
τt∗ T ⊗τt∗ T 0 −T ⊗T 0 τt∗ T ⊗τt∗ T 0 −T ⊗τt∗ T 0 +T ⊗τt∗ T 0 −T ⊗T 0
DL (T ⊗ T 0 ) = lim t = lim t
t→0 t→0
τt∗ T −T τt∗ T 0 −T 0
= lim t ⊗ τt∗ T 0 + lim T ⊗ t = (DL T ) ⊗ T 0 + T ⊗ (DL T 0 ).
t→0 t→0
4. DL (ωp ∧ ωq ) = (DL ωp ) ∧ ωq + ωp ∧ (DL ωq ).
5. DL (Cij T ) = Cij (DL T ).
6. (DL T )(D1 , . . . , ωq ) = D(T (D1 , . . . , ωq )) − T (DL D1 , . . . , ωq ) − . . . − T (D1 , . . . , DL ωq ).

Just derive T (D1 , . . . , ωq ) = C11 . . . C11 (D1 ⊗ . . . ⊗ Dp ⊗ T ⊗ ω1 ⊗ . . . ⊗ ωq ).
7. DL ω = D ◦ ω − ω ◦ DL ; that is to say, (DL ω)(D0 ) = D(ω(D0 )) − ω(DL D0 ).
8. DL (df ) = d(Df ).
τt∗ df −df dτt∗ f −df
∗
τt∗ f −f

τ f −f
DL (df ) = lim t = lim t = lim d t t = d lim t = d(Df ).
t→0 t→0 t→0 t→0
9. DL D0 = [D, D0 ].
(DL D0 )f = (df )(DL D0 ) = D((df )(D0 )) − (DL (df ))(D0 )
= D(D0 f ) − (d(Df ))(D0 ) = D(D0 f ) − D0 (Df ) = [D, D0 ]f .
10. Jacobi’s Identity: DL [D1 , D2 ] = [DL D1 , D2 ] + [D1 , DL D2 ].
11. (D1 + D2 )L T = D1L T + D2L T .

(D1 + D2 )L D = [D1 + D2 , D] = [D1 , D] + [D2 , D].
(D1 +D2 )L ω = (D1 +D2 )◦ω −ω ◦(D1 +D2 )L = (D1 +D2 )◦ω −ω ◦(D1L +D2L ) = D1L ω +D2L ω.
We conclude since both (D1 + D2 )L and D1L + D2L derive tensor products.
12. [D1 , D2 ]L = [D1L , D2L ] on tensor fields: [D1 , D2 ]L T = D1L (D2L T ) − D2L (D1L T ).
When T is a vector field, it is just Jacobi’s identity. Now, when T is a 1-form, it follows from
property 7, and we conclude since both [D1 , D2 ]L and [D1L , D2L ] derive tensor products.
Corollary: [D, D̄] = 0 if and only if τt τ̄s = τ̄s τt ; (D, D̄ complete vector fields).
Proof: If DL D̄ = [D, D̄] = 0, then τt D̄ = D̄, and τt transforms integral curves of D̄ into integral
curves of D̄. The converse is obvious.
9.2.2 Pfaff Systems

Definitions: Let O be the local ring of germs at a point x of smooth functions on a smooth
manifold, and let T be the module of germs at x of vector fields. A distribution of rank r is a
free submodule D = hD1 , . . . , Dr i ⊆ T generated by r vector fields, linearly independent at x.
Once we fix representants of the germs Di , they are linearly independent on a neighborhood
U of x, and they define at any point y ∈ U a vector subspace of dimension r,
∆y = {λ1 D1,y + . . . + λr Dr,y } ⊆ Ty X.
So the distribution may be viewed as a germ at x of a smooth family of r-planes.

We say that D is integrable if D = h∂1 , . . . , ∂r i in some local coordinate system.
Let Ω be the O-module of germs of 1-forms. A Pfaff system of rank r is a free submodule
P = hω1 , . . . , ωr i ⊆ Ω generated by r forms, linearly independent at x, and it is the germ at x
of a smooth family of vector subspaces of dimension r,
Py = {λ1 ω1,y + . . . + λr ωr,y } ⊆ Ty∗ X.
We say that P is integrable if P = hdx1 , . . . , dxr i in some

P local coordinate system, and
projectable to a subring B ⊂ O if it is generated by 1-forms i fi dhi ; with fi , hi ∈ B.
Lemma: If DL P ⊆ P, then we have τt (Py ) = Pτt (y) in a neighborhood of x.
Proof: Let P = hω1 , . . . , ωr i. If DL P ⊆ P, then DL (ω1 ∧ . . . ∧ ωr ) = f ω1 ∧ . . . ∧ ωr and Λr P is

invariant under the flow {τt } of D by the former theorem; hence so is P.
Definition: The characteristic system of P is formed by the incident vector fields D such
that DL P ⊆ P (i.e., iD ω = 0, iD dω ∈ P, for all ω ∈ P).
Projection Theorem: If Dx 6= 0, a Pfaff system P is projectable to the ring of first integrals

of a field D if and only if D is in the characteristic system of P.
Proof: If Dx 6= 0, by the local classification of vector fields, on a small neighborhood U of x we

have a projection π : U → V admitting a smooth section σ : V → U passing through x, and first
integrals of D are just smooth functions on V . If P is incident to D and DL P ⊆ P, let us see
that P = hω1 , . . . , ωr i coincides with hπ ∗ σ ∗ ω1 , . . . , π ∗ σ ∗ ωr i.
Both systems coincide at the points of the section, since both coincide on the hyperplane
tangent to the section and both vanish on the supplement defined by D.
Since both are invariant under the flow of D, they coincide on a neighborhood of x.
Conversely, if P is projectable, then the vector field D is in the characteristic system since,
whenever Dfi = Dhi = 0, we have
P
0= i fi dhi (D),
P
0=D L
i fi dhi .
Definition: A distribution D is involutive when D1 , D2 ∈ D ⇒ [D1 , D2 ] ∈ D.

If P = Do is the incident Pfaff system, this condition states that the characteristic system
of P is just P o , since (D1L ω)(D2 ) = D1 (ω(D2 )) − ω([D1 , D2 ]).
Theorem: If P is a Pfaff system, the following conditions are equivalent,
1. P is integrable.
2. The ideal generated by P in the algebra of differential forms is stable under the exterior
differential, dP ⊆ P ∧ Ω• .
3. The incident P o is an involutive distribution.
Proof: (1 ⇒ 2) If P = hdx1 , . . . , dxr i, then d( i fi dxi ) = i dfi ∧ dxi ∈ P ∧ Ω• .

P P
(2 ⇒ 3) Let P = hω1 , . . . , ωr i. If D ∈ P o , and ω ∈ P, then DL ω ∈ P,
DL ω = iD dω + diD ω = iD ( i θi ∧ ωi ) = i (iD θi )ωi − (iD ωi )θi = i (iD θi )ωi .

P P P
(3 ⇒ 1) By induction on n = dim X. If the incident of P is involutive, it coincides with the

characteristic system of P, and by the projection theorem there is a projection π : U → V and
a Pfaff system P 0 = hω10 , . . . , ωr0 i on V such that P = hπ ∗ ω10 , . . . , π ∗ ωr0 i.
Moreover, any vector field D0 on V is the projection of some vector field D on U .
If D10 , D20 ∈ (P 0 )o are the projections of D1 , D2 ∈ P o , then [D10 , D20 ] is the projection of
[D1 , D2 ] ∈ P o , and [D10 , D20 ] ∈ (P 0 )o . By induction P 0 is integrable; hence so is P.
Frobënius Theorem: Involutive distributions are integrable.
Proof: If D = (Do )o is involutive, then Do = hdxr+1 , . . . , dxn i, and D = h∂x1 , . . . , ∂xr i.
Corollary: ω = f dh if and only if ω ∧ dω = 0. (Assuming ωx 6= 0).
Definition: A germ ω ∈ Ω is regular of class m if it does not vanish at the considered point x
and the module of incident vector fields preserving it
Dω = {D ∈ T : iD ω = 0, DL ω = 0} = {D ∈ T : iD ω = 0, iD dω = 0}
is a distribution of codimension m, necessarily integrable: If D, D̄ ∈ Dω , then
ω([D, D̄]) = D(ω(D̄)) − (DL ω)(D̄) = 0,
[D, D̄]L ω = DL D̄L ω − D̄L DL ω = 0.
By Frobënius theorem Dω = h∂m+1 , . . . , ∂n i in a local coordinate system, and ω is projectable
to a germ ω 0 in dimension m such that Dω0 = 0.
Darboux’s Theorem: Let ω ∈ Ω be a regular germ of class m.
If m = 2k + 1 is odd, there are local coordinates (z, x1 , . . . , xk , p1 , . . . , pk , . . .) where
ω = dz − p1 dx1 − . . . − pk dxk .
If m = 2k is even, there are local coordinates (x1 , . . . , xk , p1 , . . . , pk , . . .) where
ω = p1 dx1 + . . . + pk dxk .
Proof: We may assume that dim X = m and that Dω = 0.
We proceed by induction on m, and the case m = 1 is trivial.
If m is odd, then dω has non null radical, defining a distribution hZ i of rank 1 transversal
to ω = 0. Dividing Z by ω(Z), we may assume that ω(Z) = 1.
Let us fix a local coordinate system where Z = ∂z .
Adding to z a first integral of Z we may assume that θ = dz − ω does not vanish at x.
Now iZ θ = 0, iZ dθ = iZ dω = 0, and θ projects onto a 1-form θ0 in dimension m − 1, and
Dθ0 = 0, since the radical of dθ = dω is generated by Z.
By induction θ0 = p1 dx1 + . . . + pk dxk and we conclude.
If m is even, the radical of dω is null (otherwise the dimension is ≥ 2, and Dω 6= 0), and
vectors D such that iD dω = λω (hence iD ω = 0) define a distribution of rank 1, which is the
characteristic system of hωi.
Hence hωi is projectable, ω = p1 (π ∗ ω 0 ), to dimension m − 1.
Clearly p1 does not vanish at x and ω 0 does not vanish at π(x).
Moreover, if iD0 ω 0 = 0, iD0 dω 0 = 0, and we take a vector field D projecting onto D0 , we have
that iD ω = 0, iD dω ∈ hωi; hence D is in the characteristic system of hωi, and D0 = 0.
By induction ω 0 = dx1 + p̄2 dx2 + . . . + p̄k dxk , and if we put pi = p1 p̄i ,
ω = p1 dx1 + . . . + pk dxk .
Since the radical of dω = dp1 ∧ dx1 + . . . + dpk ∧ dxk is null, dp1 , dx1 , . . . , dpk , dxk are linearly
independent at x.
9.3. INTEGRATION OF DIFFERENTIAL FORMS 259
9.3 Integration of Differential Forms

Definitions: The volume forms on a smooth manifold X of dimension n are the n-forms not
vanishing at any point, and X is orientable if it admits a volume form.
Two volume forms, ω, ω 0 define the same orientation of X if ω 0 = f ω for some smooth
function f > 0, and orientations of X are just equivalence classes.
Proposition: Let {Ui , [ωi ]} be an oriented open cover of X. If the orientations coincide on
intersections, [ωi |Ui ∩Uj ] = [ωj |Ui ∩Uj ], then there is a unique orientation [ω] of X inducing on
each open set Ui the given orientation, [ω|Ui ] = [ωi ].
P
Proof: Just take ω = i φi ωi , where {φi } is a partition of unity subordinate to {Ui }.
Definition: A closed set Ω ⊆ X is a manifold with boundary when
1. The boundary ∂Ω is void or a smooth submanifold of dimension n − 1.

o
2. ∂ Ω= ∂Ω.
Lemma: If p ∈ ∂Ω, in a coordinate neighborhood (U ; x1 , . . . , xn ) we have that
U ∩ Ω = {p ∈ U : x1 (p) ≤ 0}.
Proof: Let U be a coordinate neighborhood where the equation of U ∩ ∂Ω is x1 = 0, and

U − (U ∩ ∂Ω) has two connected components,
U − (U ∩ ∂Ω) = U+ ∪ U− ,
o
of equations x1 > 0, and x1 < 0. Since X − ∂Ω =Ω ∪Ωc , intersecting with U ,
o
U+ ∪ U− = U − (U ∩ ∂Ω) = (U ∩ Ω) ∪ (U ∩ Ωc ).
o o
Hence U− = U ∩ Ω, and in this case U ∩ Ω is x1 ≤ 0, or U+ = U ∩ Ω, and in this case U ∩ Ω
is x1 ≥ 0, and we change the sign of x1 .
Definitions: A vector Dp ∈ Tp X, p ∈ ∂Ω, points outside of the manifold with boundary Ω if,
in the coordinate system of the above lemma,
Dp = λ1 ∂1 + . . . + λn ∂n , λ1 > 0 ;
i.e., for any curve σ : I → X, tangent to Dp at t = 0, there is ε > 0 such that σ(−ε, 0) ⊆ Ω,
σ(0, ε) ⊆ X − Ω.
Now each orientation [ω] of X induces an orientation on the boundary ∂Ω, considering on
Tp (∂Ω) the orientation defined by iDp ωp = ωp (Dp , . . .), where Dp points outside of Ω.
In the local coordinate system of the lemma, if [ω] = [dx1 ∧ . . . ∧ dxn ], then the orientation
induced on the boundary is [ω] = [dx2 ∧ . . . ∧ dxn ], so that it does not depend on the vector Dp
and, on a neighborhood of any point, orientations are defined by a differential form; hence we
obtain an intrinsic orientation on ∂Ω.
Let ω be a smooth differential n-form with compact support on an oriented smooth manifold
X of dimension n. We shall always consider local coordinate systems (u1 , . . . , un ) where the
fixed orientation is [du1 ∧ . . . ∧ dun ].
If the support of ω is contained in a coordinate neighborhood (U ; u1 , . . . , un ), then

ω = f (u1 , . . . , un )du1 ∧ . . . ∧ dun ,
where f is a smooth function with compact support on an open subset of Rn , and we put
Z Z
ω := f du1 . . . dun .
X Rn
This definition is intrinsic (does not depend on coordinates, nor on U ) because, if (x1 , . . . , xn )
is another coordinate system and ui = hi (x1 , . . . , xn ) = hi (x), we have
ω = f (u1 , . . . , un )du1 ∧ . . . ∧ dun = f (h1 (x), . . . , hn (x))Jdx1 ∧ . . . ∧ dxn
∂hi

(where the jacobian J = det ∂x j
is positive since du1 ∧ . . . ∧ dun and dx1 ∧ . . . ∧ dxn define the
same orientation) and the change of variables formula states that
Z Z
f du1 . . . dun = f (h1 (x), . . . , hn (x))Jdx1 . . . dxn .
Rn Rn
S
In general we consider a partition of unity {φi } subordinate to a cover X = i Ui by coor-
dinate open sets, and we put Z X Z
ω= φi ω,
X i X
where the sum is finite since the support of ω is compact and the family {supp φi } is locally
finite. This definition does not depend on the cover nor the partition of unity: S
In fact, if {ϕj } is a partition of unity subordinate to another cover X = j Vj , we have
Z Z Z Z Z
P P P P P P P
φi ω = φi ϕj ω = φ i ϕj ω = ϕj φ i ω = ϕj ω.
i X i X j i,j X j X i j X
Even if we have assumed that ω is smooth, these definitions are full sense when the local ex-
pression of ω in any local coordinate system is f (u1 , . . . , un )du1 ∧. . .∧dun , where f is integrable;
for example when ω is a continuous n-form. Now, any submanifold Y ⊂ Rn of codimension r ≥ 1
has null measure, because locally it is U ∩ Y = {x ∈ U : u1 (x) = . . . = ur (x) = 0}. Hence, we
may define the integral of a smooth n-form ω on any manifold with boundary Ω by (the indicator
function IΩ vanishes outside of Ω, where the value is 1)
Z Z
ω= IΩ ω.
Ω X
Z Z
Stokes Theorem: dω = ω, for any (n − 1)-form ω of class C 1 with compact support.
Ω ∂Ω
Proof: Using partitions of unity, it is enough to show that any point p ∈ X has a neighborhood
U where the theorem holds for any form with compact support contained in U .
We distinguish 3 cases.
1. If p ∈ Ωc , we put U = X − Ω, and Ω dω = 0 = ∂Ω ω when supp ω ⊆ U .
R R
o
2. If p ∈Ω, we may assume that Rn = X = Ω = U . In this case ∂Ω ω = ∅ ω = 0, and we may
R R
assume that ω = f dx2 ∧ . . . ∧ dxn , where the support of f is compact,

Z Z Z Z ∞
∂f ∂f
dω = dx1 . . . dxn = dx1 dx2 . . . dxn
Rn n ∂x1 Rn−1 −∞ ∂x1
ZR
= (f (b, x2 , . . . , xn ) − f (−a, x2 , . . . , xn )) dx2 . . . dxn = 0.
Rn−1
3. If p ∈ ∂Ω, we may assume that Rn = X = U , and that Ω is the closed set x1 ≤ 0, and ∂Ω is
the hyperplane x1 = 0.
R R
If ω = f dx1 ∧ . . . dx
ci . . . ∧ dxn , the argument of case 2 shows that
Ω dω = 0, and ∂Ω ω = 0,
since dx1 vanishes on ∂Ω. When ω = f dx2 ∧ . . . ∧ dxn ,
Z Z Z Z 0
∂f ∂f
dω = dx1 . . . dxn = dx1 dx2 . . . dxn
Ω Ω ∂x1 Rn−1 −∞ ∂x1
Z Z
= f (0, x2 , . . . , xn ) dx2 . . . dxn = ω.
Rn−1 ∂Ω
Definitions: The volume form of an oriented riemannian manifold X is the smooth n-form
dX such that dX(D1 , . . . Dn ) = 1 for any direct orthonormal base (p. 67). R
The integral of aRfunction f on a compact manifold with boundary Ω ⊆ X is Ω f dX, and
the volume of Ω is Ω dX.
The divergence of a vector field D is defined to be DL (dX ) = (div D)dX, and the gradient
of a function f is the vector field grad f such that Df = (grad f ) · D for any vector field D.
Divergence Theorem: The integral of the divergence of a vector field D of class C 1 on a

compact manifold with boundary Ω is the flux through the boundary S = ∂Ω; that is to say, if
Np is the unique unitary orthogonal vector to Tp S pointing outside of Ω,
Z Z
(div D)dX = (D · N )dS.
Ω S
Z Z Z Z Z
L
Proof: div D = D dX = (diD + iD d)dX = diD (dX) = iD dX
Ω Ω Ω Ω S
and we have to show that the restriction of iD dX to S coincides with (D · N )dS.
If (D2 , . . . , Dn ) is a direct orthonormal base in Tp S, then (N, D2 , . . . , Dn ) is a direct or-
thonormal base in Tp X, and
(iD dX)(D2 , . . . , Dn ) = dX(D, D2 , . . . , Dn ) = dX((D · N )N + . . . , D2 , . . . , Dn )
= (D · N )dX(N, D2 , . . . , Dn ) = D · N = (D · N )dS(D2 , . . . , Dn ).
Z Z
Corollary: (div grad f )dX = (N f )dS.
Ω S
9.3.1 Harmonic Functions

Definition: Let U ⊆ Rn be an open set, n ≥ 2. A function u ∈ C 2 (U ) is harmonic if it has
null laplacian:
0 = ∆u = (∂12 + . . . + ∂n2 )u = div(grad u).
An harmonic function on Rn − {p} is kx − pk2−n when n ≥ 3, and ln kx − pk when n = 2.
Lemma: If f is a continuous function on a closed ball B = B(x, r), then

Z Z r Z !
f dx1 . . . dxn = f dSρ dρ , Sρ := S(x, ρ).
B 0 Sρ
o
Proof: Let S be the unit sphere. We have a diffeomorphism (0, r) × S 'B (x, r) − {x}, and by
Fubini’s theorem
Z Z r Z !
f dx1 . . . dxn = iN (f dx1 ∧ . . . ∧ dxn ) dρ,
B 0 Sρ
where N is the unitary orthogonal vector to the spheres Sρ . We conclude because the restriction
of iN (dx1 ∧ . . . ∧ dxn ) to Sρ is just dSρ .
Theorem: Let u be an harmonic function on an open set U ⊆ Rn . The value u(p) at any point
p ∈ U is the mean value on any closed ball Br ⊂ U of radius r with center at p, and the mean
value on the sphere Sr = ∂Br :
Z Z
1 1
u(p) = udSr = udx1 . . . dxn .
Vol Sr Sr Vol Br Br
R
Proof: Let S be the unit sphere and consider the function M (r) = S u(p + rx)dS.
By the differentiation rule under the integral sign and the divergence theorem,
Z Z Z
dSr 1
M 0 (r) =

∂r u(p + rx) dS = (N u) n−1 = n−1 (∆u)dx1 . . . dxn = 0.
S Sr r r Br
Hence, M (r) is constant. Now, M (r) → u(p)(Vol S) when r → 0, and we conclude that
Z Z
1 1
udSr = u(p + rx)dS = u(p).
Vol Sr Sr Vol S S
Now, u and the constant function u(p) have the same integral on any sphere Sρ , 0 < ρ ≤ r;
hence, by the lemma
Z Z
u(p)(Vol Br ) = u(p)dx1 . . . dxn = udx1 . . . dxn .
Br Br
Corollary: If an harmonic function u on a connected open set U ⊆ Rn attains a maximum (or

a minimum), then it is constant.
Proof: The points where u attains Rthe maximum M form an open set because if u is not the
constant M on a ball B ⊂ U , then B udm < M .
Maximum Principle: Let U ⊂ Rn be an open set. If Ū is compact and u ∈ C(Ū ) is harmonic

on U , then the maximum and minimum of u are attained at some points of ∂U .
Corollary: Let U ⊂ Rn be a bounded open set. If u ∈ C(Ū ) is harmonic on U and it vanishes

on ∂U , then u = 0.
Corollary: If an harmonic function u on Rn vanishes at infinity (for any ε > 0 there is a

compact set K such that |u(x)| < ε, ∀x ∈ Rn − K), then u = 0.
Proof: Let Sr be the sphere of radius r with center at a given point p ∈ Rn . When r 0, we
have |u(x)| < ε, ∀x ∈ Sr ; hence |u(p)| < ε by the theorem.
Theorem: A continuous function u on an open set U ⊆ Rn is an harmonic function of class

C ∞ when for any closed ball B(p, r) ⊂ U we have
Z
1
u(p) = udS , S = S(p, r).
Vol S S
Proof: Let h(x) = φ(kxk2 ) be a smooth function with supp h ⊆ B(0, ε) and Rn hdx1 . . . dxn = 1.
R
Let V be the open set of all points x ∈ U such that B(x, ε) ⊆ U . On the spheres with centre
at x ∈ V , the functions u(y)h(y − x) and u(x)h(y − x) have the same integral, so that
Z Z Z
u(y)h(y − x)dy1 . . . dyn = u(x)h(y − x)dy1 . . . dyn = u(x) h(y − x)dy1 . . . dyn = u(x).
Rn Rn Rn
By the differentiation rule under the integral sign, the left hand side is a C ∞ function of x,
because so is h. Hence u ∈ C ∞ (V ).
Now, given a point x ∈ U , it is in V when ε is small; hence u ∈ C ∞ (U ).
Finally, given a point x ∈ U and a closed ball B(x, r) ⊂ U , in the proof of the above theorem
we have seen that
Z Z
(∆u)dx1 . . . dxn = rn−1 M 0 (r) , where M (r) = u(p + rx)ωS ,
B(x,r) S
and by hypothesis M (r) is constant.

Since ∆u has null integral on any ball B(x, r) ⊂ U , we conclude that (∆u)(x) = 0.
Corollary: Any harmonic function is of class C ∞ .
Corollary: The family H(U ) of all harmonic functions is closed in C(U ).
Proof: If f ∈ C(U ) is the limit of a sequence un ∈ H(U ), then for any closed ball B(p, r) ⊂ U
we have
Z Z Z
1 1 1
f (p) = lim un (p) = lim un dS = (lim un )dS = f dS.
Vol S S Vol S S Vol S S
9.3.2 De Rham Cohomology

Definitions: A differential p-form ω is closed if dω = 0, and it is exact if ω = dω 0 for some
(p − 1)-form ω 0 . Any exact form is closed since ddω 0 = 0, and the p-th De Rham cohomology
group of a smooth manifold X is the real vector space
p {closed p-forms}
HDR (X) = ·
{exact p-forms}
The cohomology ring of X is the anticommutative graded R-algebra
• L p
HDR (X) = p HDR (X), [ωp ] · [ωq ] = [ωp ∧ ωq ],
and any smooth map f : X → Y naturally induces a morphism of R-algebras
f ∗ : HDR
• •
(Y ) −→ HDR (X), f ∗ [ω] = [f ∗ ω].
Example: Let X be a compact oriented manifold X of dimension n. If [ω] is the orientation,

n (X) 6= 0, because exact n-forms have null integral,
R
then X ω > 0. Hence HDR
Z Z Z
0
dω = ω = ω 0 = 0.
0
X ∂X ∅
Definitions: Two smooth maps f0 , f1 : X → Y are homotopic, and we put f0 ≡ f1 , if there is

an interval I = (−ε, 1 + ε) and a smooth map H : X × I → Y such that
f0 (x) = H(x, 0) , f1 (x) = H(x, 1).

A smooth map f : X → Y is said to be a homotopical equivalence if there is a smooth

map g : Y → X such that f g ≡ IdY , gf ≡ IdX . A smooth manifold X is contractible to a
point p ∈ X if the inclusion p → X is a homotopical equivalence.
P
If ω is a p-form on X × I, in local coordinates ω = α fα (x, t) dxα
+ (terms withdt), where
P R1
dxα = dxi1 ∧ . . . ∧ dxip , and we consider on X the p-form I(ω) = α 0 fα (x, t)dt dxα .
By a direct calculation, I(dω) = d(I(ω)).
Lemma: I(∂tL ω) = j1∗ ω − j0∗ ω; where jt : X → X × I, jt (x) = (x, t).
Proof: Barrow’s rule and the equality ∂tL ω =

P
α (∂t fα )(x, t) dxα + (terms with dt).
Theorem: If f0 ≡ f1 , then f0∗ = f1∗ : HDR

• (Y ) −→ H • (X).
DR
Proof: Let H : f0 ≡ f1 . If ω is a closed p-form closed on Y , then ω̄ = H ∗ ω is closed, and
f1∗ ω − f0∗ ω = j1∗ H ∗ ω − j0∗ H ∗ ω = j1∗ ω̄ − j0∗ ω̄ = I(∂tL ω̄)

= I(di∂t ω̄ − i∂t dω̄) = I(di∂t ω̄) = dI(i∂t ω̄).
Corollary: If f : X → Y is a homotopical equivalence, f ∗ : HDR

• (Y ) −−
∼ • (X).
→ HDR
p
Poincaré’s Lemma: HDR (Rn ) = 0, p ≥ 1.
Proof: The homotopy H(x, t) = tx shows that Rn is contractible. q.e.d.
1. Any closed form is locally exact.
2. Let (D1 , . . . , Dn ) be a base of vector fields on a manifold X. If [Di , Dj ] = 0, any point has a
coordinate neighborhood where D1 = ∂1 , . . . , Dn = ∂n .
Proof: If (ω1 , . . . , ωn ) is the dual base, by Cartan’s formula
dωk (Di , Dj ) = Di (ωk (Dj )) − Dj (ωk (Di )) − ωk ([Di , Dj ]) = 0 − 0 − 0,
and by Poincaré’s lemma, on a neighborhood of any point x the 1-forms ωi are exact, ωi = dui ,
and u1 , . . . , un form a local coordinate system at x since the differentials define a base of Tx∗ X.
In these coordinates Di = ∂ui .
3. If n is even, any vector field D on the sphere Sn vanishes at a point.

If D has no zero, then we may assume that |D| = 1, so that it defines a smooth map
φ : Sn → Sn such that φ(x) is orthogonal to x. Now H(x, t) = cos(πt) x + sin(πt) φ(x) defines
a homotopy between the automorphism τ : Sn → Sn , τ (x) = −x, and the identity of Sn ; so
that τ ∗ : HDR
n (S ) → H n (S ) is the identity. Absurd, τ inverts orientations when n is even.
n DR n
When n is odd, φ(x0 , x1 , . . . , xn ) = (−x1 , x0 , . . . , −xn , xn−1 ) defines a tangent vector field on
Sn not vanishing at any point.
4. Put Bn,1+ε = {(x1 , . . . , xn ) ∈ Rn :

P 2
i xi < 1 + ε}. The inclusion i : Sn−1 → Bn,1+ε has no
smooth retract, not even up to homotopies.
If r : Bn,1+ε → Sn−1 is smooth and ri ≡ Id, then the composition
n−1 r∗
n−1 n−1 i∗
HDR (Sn−1 ) −−→ HDR (Bn,1+ε ) −−→ HDR (Sn−1 )
n−1 n−1
is the identity. Absurd since HDR (Bn,1+ε ) = 0, and HDR (Sn−1 ) 6= 0.
9.4. FUNCTIONS OF A COMPLEX VARIABLE 265
5. If a vector field D points outside at any point of the sphere Sn−1 , then it vanishes at a point
of the ball Bn .
Otherwise r : Bn,1+ε → Sn−1 , r(x) = Dx /|Dx |, is smooth and r(x) 6= −x for all x ∈ Sn .
Hence, up to a homotopy, r is a retract of the inclusion i : Sn−1 → Bn,1+ε , a homotopy
between ri and the identity being
t · r(x) + (1 − t)x
H(x, t) = ·
|t · r(x) + (1 − t)x|
1 (X) = 0, when X is a simply connected smooth manifold.
Proposition: HDR
Proof: Let ω be a closed 1-form on X, and let Px be the non empty set (Poincaré’s lemma) of
germs at x of primitives of ω (functions such that df = ω), so that such germs always differ in
a constant. We put (as in p. 189)
`
π : Pe = x Px −→ X, π(fx ) = x.
Any primitive f ∈ C ∞ (U ) defines a local section U → Pe of π, and if two sections coincide at

a point, fx = hx , they coincide on a neighborhood.
The images of these sections form a base of a topology on Pe, so that π is a covering (with
uncountable fibres), and continuous sections of π are just primitives of ω.
If X is simply connected, π admits a continuous global section, and ω is exact.
9.4 Functions of a Complex Variable

Let U ⊆ C be an open set. A function f = u + iv : U → C is C m if so are u and v,
C m (U, C) = C m (U ) ⊗R C = C m (U ) ⊕ i C m (U ).
Complex p-forms at a point z0 ∈ U are alternate R-multilinear maps of Tz0 U into C; i.e.,
ωp = ωp0 + iωp00 , where ωp0 and ωp00 are ordinary p-forms. We put
dωp = dωp0 + idωp00

Z Z Z
ωp = ωp + i ωp00
0
Ω Ω Ω
so that Stokes theorem holds. Moreover, we have
(Tz∗0 U )C := (Tz∗0 U ) ⊗R C = Cdx + Cdy = Cdz + Cdz̄,
where z = x + iy, z̄ = x − iy, and we define ∂f ∂f ∂f ∂f

∂z , ∂ z̄ by the equality df = ∂z dz + ∂ z̄ dz̄.
So we have C-derivations ∂z := 12 (∂x − i∂y ), ∂z̄ := 21 (∂x + i∂y ) : C ∞ (U, C) −→ C, and
R∂x + R∂y = Tz0 U ,→ DerC C ∞ (U, C), C = (Tz0 U )C = C∂x + C∂y = C∂z + C∂z̄ .

Theorem: If f = u + iv is C 1 on U , the following conditions are equivalent,

f (z0 +ε)−f (z0 )
1. At any point z0 ∈ U there exists the limit f 0 (z0 ) = lim ε , ε ∈ C.
ε→0
2. f∗ : C = Tz0 U −→ C is C-linear at any point z0 ∈ U ; i.e., ∂z̄ f = 0,

ux = vy
(Cauchy-Riemann Equations)
uy = −vx
3. The complex 1-form f (z)dz is closed.
4. If a circle γ surrounds a point z0 ∈ U , and it is in a disk contained in U ,

Z
1 f (z)
f (z0 ) = dz. (Cauchy’s Formula)
2πi γ z − z0
∞
an (z − z0 )n , (f is analytic).
P
5. Locally f is a power series, f (z) =
n=0
Proof: (1 ⇒ 2) As ε goes to 0 along the real axis or the imaginary axis,
f 0 (z0 ) = ∂x f = ux + ivx ,
f 0 (z0 ) = 1
i ∂y f = −iuy + vy .
(2 ⇒ 3) d(f dz) = df ∧ dz = ( ∂f
∂z dz +
∂f
∂ z̄ dz̄) ∧ dz = ∂f
∂ z̄ dz̄ ∧ dz = 0.
(3 ⇒ 4) When γ is centered at z0 , in polar coordinates z = z0 + reiθ we see that
Z 2π
f (z0 + reiθ )
Z Z
1 f (z) 1 iθ 1
dz = d(re ) = f (z0 + reiθ ) dθ
2πi γ z − z0 2πi γ reiθ 2π 0
is the mean value of f (z) on γ; hence it goes to f (z0 ) as r → 0.

But the integrals coincide2 when two circles surrounding z0 are in a disk D ⊂ U and (with
reverse orientations) they form the boundary of an annulus: just apply Stokes theorem to the
f (z)
1-form 2πi(z−z 0)
1
dz, which is closed because so is f (z)dz and d z−z 0
= −(z − z0 )−2 dz.
o
(4 ⇒ 5) Let D ⊂ U be a closed disc centered at z0 . If z ∈D, η ∈ ∂D,
∞
1 1 1 1 X (z − z0 )n
= = . z−z0 =
η−z (η − z0 ) − (z − z0 ) η − z0 1 − η−z (η − z0 )n+1
0 n=0
∞
f (η)(z − z0 )n
Z Z X
1 f (η) 1
f (z) = dη = dη
2πi ∂D η − z 2πi ∂D (η − z0 )n+1
n=0
and we may integrate term by term because the series uniformly converges on ∂D, since it is
bounded by a convergent geometric series (K is the maximum of |f (z)| on D)
f (η)(z − z0 )n K |z − z0 |n

|z − z0 |
(η − z0 )n+1 ≤ R ; < 1, (R the radius of D).

R n R
Therefore, on the interior of the disc D we have a power series expansion
∞
Z
P n 1 f (η)
f (z) = an (z − z0 ) , where an = dη.
n=0 2πi ∂D (η − z0 )n+1
MR
Cauchy’s Inequalities: |an | ≤ Rn , where MR is the maximum of |f (η)| on ∂D.
In fact, in polar coordinates η = z0 + Reiθ ,

Z 2π iθ )Rieiθ
Z 2π
1 f (z 0 + Re 1
|f (z0 + Reiθ )| dθ.

|an | = dθ ≤
2πi 0 Rn+1 eiθ(n+1) 2πRn 0
2 1
If a non constant polynomial P (z) has no complex root, then f (z) = P (z) is derivable at any point and the
mean value of f (z) on the circle |z| = r goes to 0 as r → ∞. Hence f (0) = 0, a contradiction proving again
D’Alembert’s theorem.
n
P
(5 ⇒ 1) Any series n an (z P− z0 ) is derivable in the interior of the disc of convergence, and
the derivative coincides with n an n(z − z0 ) n−1 , with equal radius of convergence.
Since the derivative is again a power series, analytic functions are infinitely derivable (and,
by the Cauchy-Riemann equations, the real and imaginary part u, v are harmonic functions).
Liouville’s Theorem: Any bounded analytic function on C is constant.

M
Proof: If |f (z)| ≤ M on C, then |an | ≤ Rn for any R > 0. Hence an = 0 when n ≥ 1.
Cauchy-Goursat Formula: If U simply connected, and f : U → C is analytic, then for any

closed curve γ in U we have Z
f (z) dz = 0.
γ
Proof: Since U is simply connected and f (z) dz is closed, it is exact (p. 265).
Definition: A topological space X, with a sheaf OX of complex valued continuous functions,

is a Riemann surface if it is locally isomorphic to an open set of C with the sheaf of ana-
lytic functions. We say that OX (U ) is the ring of analytic functions on U , and morphisms
(Y, OY ) → (X, OX ) between Riemann surfaces (see p. 275) are said to be analytic maps. An
analytic functions on U is just an analytic map U → C.
An analytic function f ∈ OX (U ) is a coordinate if it defines an isomorphism of U onto an
open set of C, and it is a local coordinate at p ∈ U if it is a coordinate on some neighborhood
of p (i.e., f 0 (p) 6= 0 when U is an open set of C).
Examples: Open sets in C, P1,C , toruses C/(Ze1 + Ze2 ), coverings of a Riemann surface,...
We shall always assume that all Riemann surfaces are connected and σ-compact3 .
Theorem: Let f : X → Y be a non constant analytic map. If p ∈ X, there are coordinate

neighborhoods (U, u) and (V, v) of p = 0 and f (p) = 0 such that f (u) = un , for some natural
number n ≥ 1, the ramification index indp f of f at p.
Proof: Let us fix local coordinates u, v so that p = f (p) = 0. Then the map is v = z n h(z), where
h(0) 6= 0.
In a neighborhood
p of p = 0 the n-th root of h(z) is analytic, and changing the coordinate z
by u = z n h(z) we have that v = z n h(z) = un . q.e.d.
This local classification of morphisms has some important and obvious consequences:
1. The fibres of any non constant analytic map are discrete. In particular the zeroes of a non
constant analytic function are discrete.
2. If two analytic functions coincide on a non void open set, then they are equal.
3. Any non constant analytic map is open.
4. Maximum Principle: If f ∈ O(X) is not constant, then |f | has no local maximum.
5. Any analytic function on a compact Riemann surface is constant.
6. Any injective morphism X → Y is an isomorphism onto an open subset of Y . Hence, any

bijective analytic map is an isomorphism.
3
In fact, according to a Radó’s theorem, any separated Riemann surface is σ-compact.
9.4.1 Meromorphic Functions

Let f : U → C be analytic, except at a point z0 ∈ U ; i.e., analytic on U − z0 . In calculations we
assume that z0 = 0; but we state the results in general.
Let us consider an annulus Ω = {r ≤ |z| ≤ R}. The boundary is formed by the circles γ and
Γ of radius r and R. The argument of Cauchy’s formula now gives
Z Z
1 f (η) 1 f (η)
z ∈ Ω; f (z) = dη − dη,
2πi Γ η − z 2πi γ η − z
∞ ∞
zn
Z Z
1 X 1 f (η) X 1 f (η)
η ∈ Γ; = , dη = dη z n ,
η−z η n+1 2πi Γ η − z 2πi Γ η n+1
n=0 n=0
∞
X ηn Z −∞ Z
1 1 f (η) X 1 f (η)
η ∈ γ; = , dη = dη zn,
z−η z n+1 2πi Γ z − η 2πi γ η n+1
n=0 n=−1
and the function f (z) may be expanded as a Laurent series on the annulus Ω,
∞
Z
P n 1 f (η)
f (z) = an (z − z0 ) , an = dη,
n=−∞ 2πi σ (η − z0 )n+1
where σ is γ, Γ or an intermediate circle, since the 1-form f (η)(η − z0 )−n−1 dη is closed on

U − z0 . Hence the expansion holds on a neighborhood (except z0 ), and Cauchy’s inequalities
hold, |an | ≤ r−n Mr , where Mr = max |f (z)|.
|z−z0 |=r
Definitions: If the coefficients an are null when n < 0, we say that z0 is a removable sin-
gularity of f , if they are null up to a finite number, z0 is a pole of f , (and the order is the
greatest m such that a−m 6= 0), and if there are infinite non null coefficients, z0 is an essential
singularity. A function is meromorphic if it is analytic up to a discrete set of poles.
Examples: If f has a pole of order m, on a neighborhood we have f = h/z m , where h is

analytic, so that meromorphic functions locally coincide with quotients of analytic functions;
hence the meromorphic functions on X form a field.
If f is a meromorphic function on P1 , subtracting the singular part of the Laurent expansion
at each pole (a rational function with a unique pole) we obtain an analytic function on P1 ; hence
it is constant. Meromorphic functions on P1 are rational functions.
Removable Singularity Theorem: If |f | is bounded on a neighborhood of z0 , it is a removable

singularity. Hence, if f may be continuously extended to z0 , the extension is analytic.
Proof: If Mε ≤ M for any small ε, then |a−n | ≤ M εn , and a−n = 0.
Weierstrass Theorem: A singularity is essential if and only if the image of any neighborhood
is dense in C.
Proof: If the origin is a pole, f (z) = h(z)z −m , where h(0) 6= 0.

Hence f (U − {0}) is not dense in C when U is small, since limz→0 f (z) = ∞.
Conversely, if there is a neighborhood U such that f (U − {0}) is not dense in C, we may
assume that the disc of radius R centered at the origin does not intersect f (U − {0}).
The module of f (z)−1 is bounded by R−1 ; hence h(z) = f (z)−1 is analytic on U , and
f (z) = h(z)−1 is meromorphic. The singularity is a pole.
Theorem: The meromorphic functions on a Riemann surface X are the analytic maps X → P1
(except the constant map ∞).
Proof: If f : X → P1 is a morphism, and we consider the usual coordinate neighborhoods (U0 ; z),
(U∞ ; z1 ) of the origin and the point at infinity, the function f is analytic on V0 = f −1 (U0 ), and
1/f is analytic on V∞ = f −1 (U∞ ). Hence f is meromorphic on V0 ∪ V∞ = X.
Conversely, if we extend a meromorphic function f it with the value ∞ at any pole, the map
f : X → P1 is analytic because f1 is an analytic function on a neighborhood of each pole.
az+b
Corollary: The analytic automorphisms of P1 are just4 the homographies τ (z) = cz+d ·
Corollary: The analytic automorphisms of C are just the affinities τ (z) = az + b.
Proof: If we extend an automorphism τ : C → C to P1 , τ (∞) = ∞, we obtain a meromorphic

function on P1 , since the singularity at ∞ is not essential.
Hence it is a homography fixing the point at infinity.
Definitions: A complex 1-form ω on a Riemann surface is analytic if locally ω = f (z)dz, where

f (z) is an analytic function (hence dω = 0). If ω is analytic on U − z0 , the integral along a small
closed curve γ around z0 (oriented as boundary of the region containing z0 ) does not depend on
1
the curve and, with a factor 2πi , it is the residue of ω at z0 ,
Z
1
Res(ω, z0 ) = ω.
2πi γ
If ω = f (z)dz, with f (z) meromorphic and Laurent expansion f (z) = n an (z − z0 )n , we
P
may integrate term by term since the series uniformly converges on γ,
Z Z
n 1
(z − z0 ) dz = d(z − z0 )n+1 = 0, when n 6= −1.
γ n+1 γ
Z
a−1 dz
Res(ω, z0 ) = = a−1 .
2πi γ z − z0
Residue Theorem: Let X be a Riemann surface, and let ω be an analytic 1-form up to a
discrete set. If Ω ⊆ X is a compact manifold with boundary and ω has no singular point on ∂Ω,
Z
1 P
ω= Res(ω, zi ).
2πi ∂Ω zi ∈Ω
a small disc Di containing each singularity zi ∈ Ω, then ω is closed in a

Proof: If we consider S
neighborhood of Ω − i Di , and by Stokes theorem
Z X Z Z X
0= ω− ω= ω − 2πi Res(ω, zi ).
∂Ω i ∂Di ∂Ω i
P
Corollary: zi ∈X Res(ω, zi ) = 0, when X is a compact Riemann surface.
Corollary: Let f be a meromorphic function on a Riemann surface X. If Ω ⊆ X is a compact

manifold with boundary and f has no zero or pole on ∂Ω, then the number of zeroes of f in Ω,
minus the number of poles, all counted with multiplicity, is just
Z
1 df
·
2πi ∂Ω f
4
so that the concept of Riemann surface captures the elusive structure of the complex projective line.
Proof: If f (z) = am z m + am+1 z m+1 + . . ., am 6= 0, then

df (mam z m−1 + . . .)dz m
ω= = = + . . . dz,
f am z m + am+1 z m+1 + . . . z
and the residue of ω at z = 0 is just m. The sum of the residues of ω is the number of zeros of
f minus the number of poles, all counted with multiplicity.
Corollary: The number of poles of a non constant meromorphic function f on a compact Rie-
mann surface coincides with the number of zeros5 .
9.4.2 Convergence in O(X)

If X is a Riemann surface and p ∈ X, the differential dp z : Tp X ' C of a local coordinate
z = x + iy defines a complex structure on Tp X, and it does not depend on the local coordinate:
if u is another local coordinate, then dp u = adp z, a 6= 0. As any complex structure on a real
vector space, it is defined by a R-linear isomorphism J : Tp X → Tp X such that J 2 = −1; and
we have J(∂x ) = ∂y , J(∂y ) = −∂x . The Riemann-Cauchy equations state that a smooth map
ϕ : X → Y between Riemann surfaces is an analytic morphism if and only if ϕ∗,p : Tp X → Tϕp Y
is C-linear (i.e. commutes with J).
On Tp X we have an orientation, requiring that D, J(D) are oriented pairs, and a scalar
product, well-defined up to a positive factor, such that D · J(D) = 0, ∀D ∈ Tp X.
In particular X inherits a natural conformal structure (a riemannian metric g, well-defined
up to the product by a positive function) and we say that g is a compatible metric, or that X
admits the metric g). When two Riemann surfaces X, Y are endowed with compatible metrics,
any oriented isometry X ' Y is an analytic isomorphism.
Now, if ϕ : Y → X is an analytic map, then ϕ∗ g is a compatible metric on Y .
In a local coordinate z = x + iy, any compatible metric (viewed as a quadratic form) is just
g = f 2 (dx2 + dy 2 ) = f 2 dzdz̄ for some smooth function f without zeroes.
The curvature K of such riemannian metric is just
∆(ln f ) (∂x2 + ∂y2 )(ln f )
K=− = − ·
f2 f2
Definition: A hyperbolic metric on a Riemann surface is a compatible riemannian metric
with constant curvature K = −1 (i.e., if and only if ∆(ln f ) = f 2 ).
Definition: The Poincaré metric on the open disk Dr ⊂ C of radius r ∈ R+ with center 0 is
the hyperbolic metric
2
2r
gr = gDr := dzdz̄.
r2 − |z|2
The disks Dr are isometric, by means of a homothety. The unit disk is D = D1 , gD = g1 .
Lemma: Let ϕ : (X, gX ) → (Y, gY ) be an analytic map between Riemann surfaces endowed with
hyperbolic metrics, so that ϕ∗ gY = F gX for some F ∈ C ∞ (X). If F attains a maximum M at
some point p ∈ X, then M ≤ 1.
Proof: If F (p) = M 6= 0, then ϕ∗ gY 6= 0 at p, so that ϕ is a local diffeomorphism at p, and ϕ∗ gY

is hyperbolic in a neighborhood of p. In a local complex coordinate z we have
ϕ∗ gY = h2 dzdz̄ , gX = f 2 dzdz̄ , F = h2 /f 2 .
5
Since a polynomial of degree n defines a meromorphic function on P1 with a pole of order n at z = ∞, this
result turns obvious D’Alembert’s theorem.
Since ln F attains a maximum at p, we have ∂x2 (ln F ) ≤ 0 and ∂y2 (ln F ) ≤ 0 at p; hence
0 ≥ (∆ ln F )(p) = 2(∆ ln h)(p) − 2(∆ ln f )(p) = 2h2 (p) − 2f 2 (p),
and we conclude that M = h2 (p)/f 2 (p) ≤ 1.
Ahlfors lemma: If gY is a hyperbolic metric on a Riemann surface Y , then any analytic map
ϕ : D → Y is contractive: ϕ∗ gY ≤ gD .
Proof: Since gD = limr→1 gr , we only have to show that the restriction ϕ : Dr → Y is contractive,
for all r < 1. Now we have
−2
∗ 2 2 2r
ϕ gY = h dzdz̄ = F gr , F = h ,
r2 − |z|2
where F ∈ C ∞ (Dr ) clearly goes to 0 when z → ∂Dr . Hence F attains a maximum M at a point
of Dr . By the lemma M ≤ 1, and we conclude.
Corollary: Any analytic map D → D is contractive with respect to the Poincaré metric.
Corollary: The Poincaré metric gD is the biggest hyperbolic metric on D.
Corollary: The analytic isomorphisms of D are the oriented isometries of gD :

z−a
τ (z) = eiθ ; θ ∈ R, |a| < 1.
1 − āz
Proof: We have τ (D) = D because τ (∂D) = ∂D and τ (a) = 0.
In fact |τ (1)| = |τ (−1)| = |τ (i)| = 1 and any homography transforms a circle into a circle (or
a line, a circle through ∞) because so do any similarity az + b and the inversion z1 , and these
homographies generate the group P GL(1, C).
Moreover, with a suitable value of θ, we obtain any oriented isometry τ∗,a : Ta D → T0 D, and
any isometry (p. 290) τ : D → D is fully determined by τ (0) and the oriented isometry τ∗,0 .
Schwarz’s Lemma: Let f : D → D be an analytic function. If f (0) = 0, then
|f (z)| ≤ |z| , |f 0 (0)| ≤ 1.

Proof: Since f is contractive, we have d 0, f (z) ≤ d(0, z), with the Poincaré metric.
Since d(0, z) is an increasing function of the Euclidean distance |z|, we have |f (z)| ≤ |z|.
Finally, by Ahlfors lemma f∗ : T0 D → T0 D is contractive.
Since f∗ is the multiplication by f 0 (0), we conclude that |f 0 (0)| ≤ 1.
Corollary: If f (0) = 0 and |f 0 (0)| = 1, then f (z) = f 0 (0)z is a rotation.
Proof: We have f (z) = zh(z) and, by the Schwarz’s lemma, |h(z)| attains a maximum at z = 0.
Hence h(z) is constant and f (z) = az = f 0 (0)z is a rotation.
Corollary: If a morphism f : D → D is an isometry at a point, then it is a global isometry.
Proof: Composing with an isometry of D we may assume that f is an isometry at z = 0 and

that f (0) = 0, so that |f 0 (0)| = 1, and f is a rotation.
Lemma: Let U ⊆ C be an open set. If a sequence (fn ) of analytic functions on U uniformly

converges on the compact sets K ⊂ U to a limit continuous function f , then f also is analytic,
and the derivatives fn0 uniformly converge on any compact set to f 0 .
Proof: Take an open disk D with center at z0 ∈ U , such that D̄ ⊂ U . If z ∈ D, then

Z
1 fn (η)
fn (z) = dη.
2πi ∂D η − z
Since the functions fn uniformly converge on the compact set ∂D, we also have
Z
1 f (η)
f (z) = dη,
2πi ∂D η − z
and the argument of p. 266 shows that f is analytic on D. Hence f is analytic.

Now take a closed disk of smaller radius D̄0 , also with center at z0 . If z ∈ D̄0 , then
Z
0 1 fn (η)
fn (z) = dη,
2πi ∂D (η − z)2
Z
0 1 f (η)
lim f (z) = dη = f 0 (z),
n→∞ n 2πi ∂D (η − z)2
and the limit clearly is uniform on the compact set D̄0 .

Since any compact set K ⊂ U may be covered with a finite number of such closed disks, we
conclude that fn0 → f 0 uniformly on any compact set K.
Weierstrass Theorem: The ring O(X) of analytic functions on a Riemann surface X is closed
in the ring C(X, C) of continuous complex-valued functions (with the compact convergence).
Montel’s Theorem: Let F be a family of analytic functions on a Riemann surface X. If for

any compact set K ⊆ X there is a constant c such that |f (x)| ≤ c, ∀f ∈ F, x ∈ K, then F has
compact closure in O(X).
Proof: Since O(X) is closed in C(X, C), we have to show that F has compact closure in C(X, C).
Now, F(x) = {f (x); f ∈ F } is a bounded set and, by Ascoli’s theorem (p. 221) we only
have to show that F is equicontinuous.
Let D be a relatively compact open disk in X. By hypothesis, there is an open disk Dr ⊂ C
2
such that f (D) ⊆ Dr , ∀f ∈ F. Now, dx2 + dy 2 ≤ r4 gr , and by Ahlfors lemma,
r2 ∗ r2
f ∗ (dx2 + dy 2 ) ≤ 4 f (gr ) ≤ 4 gD .
r2
Hence, |f (x1 ) − f (x2 )| ≤ 4 dD (x1 , x2 ) for any x1 , x2 ∈ D, f ∈ F, and F is equicontinuous.
Corollary: The family F = Homan (X, D) has compact closure in O(X).
By the maximum principle, any function in F̄ − F is constant; therefore:
Corollary: Fix p ∈ X and α ∈ D. The family {f ∈ Homan (X, D) : f (p) = α} is compact.
Lemma: Fix α ∈ D − 0. The family F = {f ∈ Homan (D, D − 0) : f (0) = α} is compact.

exp
Proof: Let h : D ' {x < 0} −−−→ D − 0 be the universal covering and fix a point α̃ ∈ D over α.
Since D is simply connected, any function f ∈ F admits a lifting f˜: D → D such that f˜(0) = α̃.
Hence, the continuous map (with the compact-open, hence the compact convergence, topologies)
h◦
Homan (D, D) −−−−→ Homan (D, D − 0)
transforms the family F̃ = {f˜ ∈ Homan (D, D) : f˜(0) = α̃} onto F. Since F̃ is compact, so is F.
Proposition: Let U = C − D̄ be the complement of a closed disk, and α ∈ U . Then the family
{f ∈ Homan (D, U ) : f (0) = α} is compact.
Proof: We may assume that D = D. Now the isomorphism z1 : U = C − D̄ → D − 0 induces a

homeomorphism Homan (D, U ) ' Homan (D, D − 0) and we conclude by the previous lemma.
Corollary: If the complement of an open set U ⊂ C contains a disk and α ∈ U , then the family
{f ∈ Homan (D, U ) : f (0) = α} has compact closure in O(D).
Koebe’s Theorem: The family of all injective analytic functions f on the unit disk D such
that f (0) = 0 and f 0 (0) = 1 is compact.
Proof: Let us see that in such family any sequence (fn ) admits a convergent subsequence.
For each index n, put rn = sup{r ∈ R+ : Dr ⊆ fn (D)}, so that Drn ⊆ fn (D and there exists
a point zn ∈
/ fn (D) with |zn | = rn . Remark that rn ≤ 1, just applying Schwarz’s lemma to the
composition (fn is injective)
r · fn−1
ϕ : D −−n−→ Drn −−−→ D , ϕ(0) = 0, ϕ0 (0) = rn .
Now the functions gn = fn /zn fulfill that 1 ∈

/ gn (D) and D ⊆ gn (D).
The covering z 2 + 1 : C − 0 → C − 1 is √
trivial over the simply connected set gn (D); hence it
admits a well-defined inverse function u = z − √1 on gn (D) such that u(0) = i.
Since the two determinations u and −u of z − 1 have disjoint images, the functions hn =
√
u◦gn = gn − 1 don’t attain values in the image of −u and, by the previous corollary, considering
a subsequence, we may assume that (hn ) converges to a limit function h : D → C.
Then g := h2 + 1 = lim(h2n + 1) = lim gn and considering a subsequence, since |zn | ≤ 1, we
may assume that zn → z. Then f := zg = lim(zn gn ) = lim fn is injective (f is not constant
because f 0 (0) = lim fn0 (0) = 1):
Lemma: If a sequence (fn ) of injective analytic functions on a connected open set U ⊆ C

converges in O(U ), then the limit function f is injective or constant.
Proof: Assume that f is not constant and f (z1 ) = f (z2 ) = a, where z1 6= z2 .

Let us consider a compact manifold with boundary Ω ⊂ U containing z1 , z2 such that f does
not attain the value a on the boundary ∂Ω. When n 0, we have
f 0 (z) fn0 (z)
Z Z
1 1
dz = dz,
2πi ∂Ω f (z) − a 2πi ∂Ω fn (z) − a
because these integrals have integer values (p. 269). Now, the left integral is ≥ 2, while the
right integral is ≤ 1, because fn is injective. Absurd.
Lemma: Let f be an analytic function without zeroes 1 (X) = 0,

on a Riemann surface X. If HDR
√
then there exist the analytic functions ln f and f .
df √
Proof: The closed 1-form f is exact, so that ln f exists, and so does f = e(ln f )/2 .
1 (U ) = 0 is isomorphic to an open subset of D.

Corollary: Any open set U ⊂ C with HDR
Proof: We may assume that 0 ∈ / U . The covering z 2 : C − 0 → C − 0 is trivial over U , because

√
it admits the analytic sections ±h(z) := ± z by the lemma. Since both sections have disjoint
images, U is isomorphic to an open set h(U ) ⊂ C whose complement contains a disk, that we
may assume to be D. Now f (z) = z1 transforms h(U ) onto an open subset of D.
Lemma: Let X be a Riemann surface with HDR 1 (X) = 0. If an injective morphism ϕ : X ,→ D
is not surjective, there is an injective morphism ϕ̄ : X ,→ D such that ϕ̄∗ gD > ϕ∗ gD .

√
Proof: We may assume that 0 ∈/ ϕ(U ). The function ϕ̄ = ϕ exists by the former lemma, so
that we have a commutative diagram
>D
ϕ̄
π
ϕ
X /D
where π(z) = z 2 , and ϕ̄ is injective because so is ϕ. By Ahlfors lemma, π ∗ gD ≤ gD , and if we

have π ∗ gD = gD at a point of D, then π is a global isometry (p. 271); absurd.
Hence ϕ̄∗ gD > ϕ̄∗ (π ∗ gD ) = ϕ∗ gD . q.e.d.
Now let X be a Riemann surface with a fixed vector 0 6= Dp ∈ Tp U , and let us consider
pointed morphisms φ : X ,→ Dr ; i.e. φ(p) = 0 and φ∗ (Dp ) = ∂x . Any pointed morphism induces
a hyperbolic metric φ∗ gr on X, that fully determines the radius r because |Dp | = 2/r2 . In
particular, a bigger hyperbolic metric determines a smaller radius.
Given a morphism ϕ : X ,→ D, composing it with a convenient isometry D ' Dr , we obtain
a pointed morphism φ : X ,→ Dr inducing the same hyperbolic metric, φ∗ gr = ϕ∗ gD . Hence, the
above lemma may be restated as follows:
Lemma: Let X be a Riemann surface with HDR 1 (X) = 0. If an injective morphism ϕ : X ,→ D

r
is not surjective, there is an injective morphism ϕ̄ : X ,→ Dr̄ with r̄ < r.
Riemann Mapping Theorem: If U ⊂ C is a connected open set with HDR 1 (U ) = 0, then U
is isomorphic to the unit disk D. In particular, U is simply connected.
Proof: We may assume that 0 ∈ U , and point U with ∂x ∈ T0 U . The family of all injective
analytic functions f : U ,→ Dr such that f (0) = 0 and f 0 (0) = 1 (i.e. pointed morphisms) is not
empty. Let R be the infimum of all r ∈ R such that there is a pointed morphism U ,→ Dr , and
let fn : U → Drn be a sequence of pointed immersions such that (rn ) is a decreasing sequence
with limit R. By Montel’s theorem, we may assume that (fn ) converges to an analytic function
f ∈ O(U ), injective (p. 273) because so are the functions fn and f 0 (0) = lim fn0 (0) = 1.
We have f (U ) ⊆ DR because f (U ) is an open set contained in all the disks Drn , and by the
lemma f : U ,→ DR is surjective. We conclude that f : U → DR is an isomorphism.
Chapter 10
Differential Geometry I
10.1 Smooth Manifolds

Definitions: To give a sheaf of (real valued) continuous functions OX on a topological space
X is Sto give a subalgebra OX (U ) ⊆ C(U ) for any open set U ⊆ X, so that for any open cover
U = i Ui we have that a continuous function f ∈ C(U ) is in OX (U ) if and only if the restriction
f |Ui ∈ OX (Ui ) for any index i. That is to say, the following sequence is exact,
Q Q
OX (U ) → OX (Ui ) ⇒ OX (Ui ∩ Uj ).
i i,j
Given two topological spaces endowed with a sheaf of continuous functions (X, OX ) and
(Y, OY ), a morphism (Y, OY ) → (X, OX ) is defined to be a continuous map φ : Y → X
transforming OX (U ) into OY (φ−1 U ),
f ∈ OX (U ) ⇒ φ∗ (f ) = f ◦ φ ∈ OY (φ−1 U ).
Examples: Given an open set V ⊆ Rn , the C ∞ -functions define a sheaf of continuous functions
CV∞ . If W is an open set in Rm , morphisms (V, CV∞ ) → (W, CW
∞ ) are just maps φ : V → W of C ∞
class.
A sheaf of continuous functions OX on a topological space X induces a sheaf of continuous
functions OU on any open set U ⊆ X. Just put (OU )(V ) = OX (V ) ⊆ C(V ), where V ⊆ U is
open.
Definitions: A smooth manifold is a topological space X endowed with a sheaf of continuous

functions OX such that any point of X has an open neighborhood U isomorphic to an open subset
V of some Rn with the sheaf of C ∞ functions, (U, OU ) ' (V, CV∞ ). In such case, the structural
∞ or C ∞ , the smooth functions on an open set U ⊆ X are defined
sheaf OX will be denoted by CX
to be the continuous functions in CX∞ (U ) := O (U ), and the morphisms (Y, C ∞ ) → (X, C ∞ )
X Y X
between smooth manifolds are said to be the smooth maps. The smooth manifolds with the
smooth maps define a category, and the isomorphisms are named diffeomorphisms.
In these notes all smooth manifolds are assumed to be σ-compact (hence separated).
The ring of germs at p ∈ X of smooth functions is denoted by OX,p or just Op ,
OX,p = lim ∞ (U ).
CX
−→
p∈U
A coordinate open set in X is an open set U diffeomorphic to an open set in Rn . If a

continuous map (x1 , . . . , xn ) : U → V ⊆ Rn is a diffeomorphism, we say that (x1 , . . . , xn ) is a
coordinate system on U ; then any smooth function on U is f (x1 , . . . , xn ) for a unique function
275
276 CHAPTER 10. DIFFERENTIAL GEOMETRY I
f ∈ C ∞ (V ). Some functions x1 , . . . , xn ∈ C ∞ (U ) define a local coordinate system at a point

p ∈ U if they form a coordinate system on a neighborhood of p.
Lemma: Op is a local ring, the maximal ideal mp = {f ∈ Op : f (p) = 0} being generated by

x1 − a1 , . . . , xn − an ; where x1 , . . . , xn are local coordinates at p = (a1 , . . . , an ).
Proof: The epimorphism Op → R, f → f (p), shows that mp is a maximal ideal.

If f ∈/ mp , then f has no zero on a neighborhood U of p; hence f is invertible in C ∞ (U ), and
the germ fp is invertible in Op . The unique maximal ideal of Op is mp .
It is clear that (x1 − a1 , . . . , xn − anP
) ⊆ mp . Conversely, if f ∈ mp , by the key lemma (p. 97)
in a neighborhood U of p we have f = i fi (xi − ai ), where fi ∈ C ∞ (U ). Considering germs we
see that f ∈ (x1 − a1 , . . . , xn − an ).
Definition: Tp X = DerR (Op , Op /mp ) is the tangent space to X at p. If D ∈ Tp X,
D(f g) = (Df ) · g(p) + f (p) · (Dg); f, g ∈ Op .

∂

If (U ; x1 , . . . , xn ) is a coordinate neighborhood, ∂x i p
= (∂xi )p = (∂i )p denotes the vector
∂ ∂f
(fp ) = (p).
∂xi p ∂xi
∂ ∂

Theorem: ∂x1 p , . . . , ∂xn p define a base of Tp X.
P
Proof: They are linearly independent. If i λi (∂i )p = 0, then
P P
0 = i λi (∂i )p (xj ) = i λi δij = λj .
Moreover, if D ∈ Tp X and Dxi = 0 for anyP index i, then D = 0.

In fact, by the lemma any germ is f = b + i hi (xi − ai ); hence
P P
Df = i (Dhi )(xi (p) − ai ) + i hi (p)Dxi = 0 + 0 = 0.
Now it is clear that for any vector D ∈ Tp X we have

∂ ∂
D = (Dx1 ) + . . . + (Dxn ) . (10.1)
∂x1 p ∂xn p
Definition: The dimension dimp X of X at p is the dimension of the vector space Tp X, and
by the above theorem it is locally constant. If it does not depend on the point p we say that it
the dimension dim X of X.
Example: If (A, E) is a real affine space of dimension n and p ∈ A, then the natural linear map
E → Tp A, transforming v ∈ E into the directional derivative ∂v,p ,
f (p + tv) − f (p) d(f ◦ σ)

∂v,p f = lim = , σ(t) = p + tv,
t dt

t→0 t=0
∂

is an isomorphism: If x1 , . . . , xn are coordinates in a base v1 , . . . , vn ; then ∂vi ,p = ∂xi p .
Definition: Let ϕ : X → Y be a smooth map, and put q = ϕ(p), where p ∈ X.

The tangent linear map of ϕ at p is the linear map
ϕ∗ : Tp X −→ Tq Y, (ϕ∗ D)(f ) = D(ϕ∗ f ) = D(f ◦ ϕ).

10.1. SMOOTH MANIFOLDS 277
The chain’s rule directly follows from the definition: (ψ ◦ ϕ)∗ = ψ∗ ◦ ϕ∗ .

If the equations of ϕ on some coordinate neighborhoods of p and P q are yi = fi (x1 , . . . , xn ),
∗
i.e. ϕ yi = fi (x1 , . . . , xn ), applying 10.1 we see that ϕ∗ ((∂xj )p ) = i (∂xj fi )(p) · (∂yi )q , so that
the matrix of ϕ∗ in the bases {(∂xj )p } and {(∂yi )q } is the jacobian matrix

∂fi
(p)
∂xj
and the inverse mapping theorem states that a smooth map ϕ : X → Y is a local diffeomor-
phism at p if and only if ϕ∗ : Tp X → Tq Y is an isomorphism.
Definitions: The dual Tp∗ X of the tangent space Tp X is the cotangent space.
If f ∈ Op , the differential of f at p is the 1-form
(dp f )(D) = Df, D ∈ Tp X.
If ϕ : X → Y is a smooth map and q = ϕ(p), then the transpose map of ϕ∗ : Tp X → Tq Y is

denoted by ϕ∗ : Tq∗ Y → Tp∗ X, (ϕ∗ ω)(D) = ω(ϕ∗ D), D ∈ Tp X.
1. The differential is a derivation:
(a) dp λ = 0, λ ∈ R.
(b) dp (f + g) = dp f + dp g.
(c) dp (f g) = g(p)dp f + f (p)dp g.
2. If (x1 , . . . , xn ) is a local coordinate system at p, then the 1-forms dp x1 , . . . , dp xn define a base

of Tp∗ X, the dual base of (∂x1 )p , . . . , (∂xn )p , and
∂f ∂f
dp f = ∂x1 (p) · dp x1 + . . . + ∂xn (p) · dp xn .
3. ϕ∗ (dq f ) = dp (ϕ∗ f ), f ∈ OY,q .

∼
Proposition: The differential defines an isomorphism dp : mp /m2p −
→ Tp∗ X.
Proof: The differential vanishes on m2p , since dp (f g) = g(p)dp f + f (p)dp g; hence it defines a
linear map mp /m2p → Tp∗ X, and it is an isomorphism since it transforms the generating system
[x1 − a1 ], . . . , [xn − an ] into the base dp x1 , . . . , dp xn (see also p. 139).
Theorem: If dp u1 , . . . , dp un form a base of Tp∗ X, then the functions u1 , . . . , un define a local

coordinate system at p.
Proof: Let us consider the smooth map ϕ = (u1 , . . . , un ) : X → Rn , q = ϕ(p).

Since ϕ∗ (dq xi ) = dp ui , then ϕ∗ transforms a base into a base; hence ϕ∗ : Tp X → Tq Rn is an
isomorphism, and ϕ is a local diffeomorphism by the inverse function theorem.
10.1.1 Tensor Fields

Definition: A vector field is a family of vectors {Dx }x∈X , where Dx ∈ Tx X, and it is smooth
(resp. of class C m ) if for any f ∈ C ∞ (U ) we have that the function (Df )(x) = Dx f so is smooth
(resp. of class C m ), so that the vector field defines derivations D : C ∞ (U ) → C m (U ) compatible
with restrictions. Unless otherwise stated, we assume that all vector fields are smooth, and the
C ∞ (X)-module of all smooth vector fields on X is denoted by D(X).
Theorem: D(X) = DerR (C ∞ (X), C ∞ (X)).
Proof: The inverse map assigns to any derivation D the vector field {Dx }x∈X defined by the
following vectors (any germ fx ∈ Ox has a global representant f , p. 250)
Dx fx := (Df )(x), fx ∈ Ox .
We only have to prove that (Df )(x) does not depend on the representant f .
If f, g ∈ C ∞ (X) have equal germ (coincide on a neighborhood U of x) we take (p. 250) a
function φ ∈ C ∞ (X), supp φ ⊆ U , φ(x) = 1, and we have φ(f − g) = 0,
0 = D(φ(f − g)) = (Dφ)(f − g) + φ(Df − Dg),
and taking values at x we see that (Df )(x) = (Dg)(x).
Definition: A 1-form on X is a family of 1-forms {ωx }x∈X , where ωx ∈ Tx∗ X, and it is smooth
(resp. of class C m ) if the function ω(D)(x) = ωx (Dx ) is smooth (resp. of class C m ) for any vector
field D ∈ D(U ), so that it defines C ∞ (U )-linear morphisms ω : D(U ) → C m (U ) compatible with
restrictions. Unless otherwise stated, we assume that all 1-forms are smooth, and the C ∞ (X)-
module of all smooth 1-forms on X is denoted Ω(X).
Theorem: Ω(X) = HomC ∞ (X) (D(X), C ∞ (X)).
Proof: The inverse map assigns to any morphism of modules ω : D(X) → C ∞ (X) the following
family {ωx }x∈X of 1-forms,
ωx (Dx ) := ω(D)(x),
where D is any vector field on X extending Dx (on a coordinate neighborhood U it clearly exists
and, multiplying by a plateau function, it may be extended by 0 on X − U ).
We have to show that ω(D)(x) does not depend on the vector field D.
If D0 is another vector field and Dx = Dx0 , just apply the following lemma to D0 − D.
Lemma: If D is a vector field on X and Dx = 0, then ω(D)(x) = 0.

P
Proof: We have D = i fi ∂i , fi (x) = 0, on a coordinate neighborhood
P U.
Let φ ∈ C ∞ (X) such that supp φ ⊆ U , φ(x) = 1. Now φ2 D = i (φfi )(φ∂i ), where φfi , φ∂i
are extended by 0 on X − U . We conclude taking values at x in the equality
φ2 ω(D) =
P
i (φfi )ω(φ∂i ).
Definition: The differential of f ∈ C ∞ (X) is the 1-form {dx f }x∈X . In the dual of D(X),
(df )(D) = Df.
Definition: A (p, q)-tensor field on X is a family of (p, q)-tensors {Tx }x∈X on Tx X, and it is
smooth (resp. of class C m ) if for any (D1 , . . . , Dp , ω 1 , . . . , ω q ) ∈ D(U )p × Ω(U )q , we have that
f (x) = Tx (Dx1 , . . . , Dxp , ωx1 , . . . , ωxq ) is a smooth (resp. of class C m ) function on U , so that it
defines C ∞ (U )-multilinear maps T : D(U )p × Ω(U )q → C ∞ (U ) compatible with restrictions.
Unless otherwise stated, any tensor field is assumed to be smooth, and the above argument
shows that the module T p,q (X) of smooth (p, q)-tensor fields on X is isomorphic to the module of
C ∞ (X)-multilinear maps D(X)p × Ω(X)q → C ∞ (X), and the module Ωp (X) of smooth p-forms
is isomorphic to the module of alternate C ∞ (X)-multilinear maps D(X)p → C ∞ (X).
10.1. SMOOTH MANIFOLDS 279
Operations with tensors and p-forms (tensor and exterior products, contraction of indices,...)
are pointwise extended to tensor fields and differential p-forms.
∂
In a coordinate neighborhood P (U ; x1 , . . . , xn ), we ∞
have the vector fields ∂i = ∂xi = ∂x i
, and
any vector field on U is D = i fi ∂i , where fi ∈ C (U ), since fi = Dxi is smooth. Hence,
D(U ) is a free module, with base ∂1 , . . . , ∂n , the module Ω(U ) is free, with base dx1 , . . . , dxn , the
module T p,q (U ) is free, with base {dxi1 ⊗ . . . ⊗ dxip ⊗ ∂j1 ⊗ . . . ⊗ ∂jq }, and the module Ωp (U ) is
free, with base {dxi1 ∧ . . . ∧ dxip }.
Definition: In a coordinate neighborhood, the exterior differential of p-forms is

P
ωp = α fα dxi1 ∧ . . . ∧ dxip , α = (i1 < . . . < ip )
P
dωp = α (dfα ) ∧ dxi1 ∧ . . . ∧ dxip .
It is easy to check that it is R-linear, d ◦ d = 0, and that it is an antiderivation,
d(ωp ∧ ωq ) = (dωp ) ∧ ωq + (−1)p ωp ∧ (dωq ).
Moreover, we prove by induction on p that d(df1 ∧ . . . ∧ dfp ) = 0.

If p = 1, it follows from d2 = 0, and in the general case we have:
d(df1 ∧ . . . dfp ) = (ddf1 ) ∧ (df2 ∧ . . . dfp ) − df1 ∧ d(df2 ∧ . . . dfp ) = 0 + 0 = 0.
Now let us prove

P that the exterior differential is intrinsic, independent of the coordinate
system. If ωp = α gα dui1 ∧ . . . ∧ duip , then, by the above properties,
P P
dωp = α dgα ∧ dui1 ∧ . . . ∧ duip + α gα d(dui1 ∧ . . . ∧ duip )
P
= α dgα ∧ dui1 ∧ . . . ∧ duip + 0.
If we put Ω• (X) = ⊕p Ωp (X), then d : Ω• (X) → Ω• (X) is an antiderivation, and d(dω) = 0.

Moreover, for any smooth map ϕ : Y → X we have ϕ∗ (dωp ) = d(ϕ∗ ωp ).
Lemma: d ◦ iD + iD ◦ d is a derivation of the algebra Ω• (X).
Proof: (diD + iD d)(ωp ∧ ωq ) = d((iD ωp ) ∧ ωq + (−1)p ωp ∧ iD ωq ) + iD ((dωp ) ∧ ωq + (−1)p ωp ∧ dωq )

= (diD ωp ) ∧ ωq + (−1)p−1 (iD ωp ) ∧ dωq + (−1)p (dωp ) ∧ iD ωq + ωp ∧ diD ωq +
+(iD dωp ) ∧ ωq + (−1)p+1 (dωp ) ∧ iD ωq + (−1)p (iD ωp ) ∧ dωq + ωp ∧ iD dωq =
= ((diD + iD d)(ωp ) ∧ ωq + ωp ∧ (diD + iD d)(ωq ).
Cartan’s Theorem: DL ωp = diD ωp + iD dωp .
Proof: By the lemma, we may assume that ωp = f or ωp = df ,
(diD + iD d)f = d0 + (df )(D) = Df = DL f,

(diD + iD d)df = diD df + iD 0 = d(Df ) = DL (df ).
Corollary: The Lie derivative commutes with the exterior differential, [DL , d] = 0.
Cartan’s Formula: (dω)(D, D̄) = D(ω(D̄)) − D̄(ω(D)) − ω([D, D̄]).
Proof: (dω)(D, D̄) = (iD dω)(D̄) = (DL ω − diD ω)(D̄) = D(ω(D̄)) − ω([D, D̄]) − d(ω(D))(D̄)
= D(ω(D̄)) − ω([D, D̄]) − D̄(ω(D)).
10.1.2 Smooth Submanifolds

Definitions: A smooth map ϕ : Y → X is a local embedding (resp. regular projection) at
a point q ∈ Y if ϕ∗ : Tq Y → Tp X, p = ϕ(q), is injective (resp. surjective).
Theorem: If ϕ : Y → X is a local embedding at q, then there are coordinate neighborhoods of p

and q where ϕ(y1 , . . . , ym ) = (y1 , . . . , ym , 0 . . . , 0).
Proof: Let (U ; x1 , . . . xn ) be a coordinate neighborhood of p, and put yi = ϕ∗ xi .

We may assume that dq y1 , . . . , dq ym form a base of Tq∗ Y , since ϕ∗ : Tp∗ X → Tq∗ Y is surjective.
Now y1 , . . . , ym are coordinates on a neighborhood of q, where we have ym+j = fj (y1 , . . . , ym ),
and all the functions fj (x1 , . . . , xm ) are defined on a small neighborhood of p. Now
ui = xi , i = 1, . . . , m
um+j = xm+j − fj (x1 , . . . , xm ), j = 1, . . . , r
have linearly independent differentials at p, so that they are coordinates on a smaller neighbor-
hood of p. In these coordinates ϕ(y1 , . . . , ym ) = (y1 , . . . , ym , 0 . . . , 0).
Theorem: If ϕ : Y → X is a regular projection at q, then there are coordinate neighborhoods of

p and q where ϕ(y1 , . . . , yn ) = (y1 , . . . , ym ), m ≤ n.
Proof: Let (U ; x1 , . . . xm ) be a coordinate neighborhood of p, and put yi = ϕ∗ xi .

The differentials dq y1 , . . . , dq ym are linearly independent, since ϕ∗ : Tp∗ X → Tq∗ Y is injective.
If we complete them so as to obtain a base dq y1 , . . . , dq yn of Tq∗ Y , then y1 , . . . , yn are coor-
dinates on a neighborhood of q, and in these coordinates ϕ(y1 , . . . , yn ) = (y1 , . . . , ym ).
Definition: Let Y be a subspace of a smooth manifold X. A continuous function f : Y → R

is smooth if it locally coincides with smooth functions on X: any point y ∈ Y has an open
neighborhood U in X such that f |U ∩Y = F |U ∩Y for some function F ∈ C ∞ (U ).
So we have a sheaf CY∞ of continuous functions on Y
CY∞ (V ) = {f : V → R smooth},
and Y is a (smooth) submanifold of X when (Y, CY∞ ) is a smooth manifold.

For example, Y = 0 × Rm is a submanifold of Rn × Rm , since in this case the sheaf CY∞ is
just the sheaf of smooth functions on Rm in the usual sense.
Lemma: If Y is a submanifold of X, the inclusion i : Y → X is a local embedding.
Proof: The linear tangent map i∗ : DerR (OY,q , R) → DerR (OX,q , R) is injective, since the restric-
tion morphism i∗ : OX,q → OY,q is surjective by definition.
Theorem: Let Y = {x ∈ X : f1 (x) = . . . = fr (x) = 0}; where f1 , . . . , fr ∈ C ∞ (X). If the

differentials dp f1 , . . . , dp fr are linearly independent at any point p of Y , then Y is a submanifold
of X of codimension r, and the tangent space at p is
Tp Y = hdp f1 , . . . , dp fr io .
Proof: We complete the differentials so as to obtain a base dp f1 , . . . , dp fn of Tp∗ X. Then
ϕ = (f1 , . . . , fn ) : U −→ Rn
10.2. LINEAR CONNECTIONS 281
defines a diffeomorphism of U onto an open set U 0 of Rn , and an isomorphism of ringed spaces

of (U ∩ Y, CU∞∩Y ) onto the open set ϕ(U ∩ Y ) = U 0 ∩ (0 × Rn−r ) of 0 × Rn−r .
Hence (Y, CY∞ ) is a manifold of codimension r.
Moreover, if D ∈ Tp Y , then (dfj )(i∗ D) = (i∗ D)fj = D(fj ◦ i) = D(0) = 0.
Hence Tp Y ⊆ hdp f1 , . . . , dp fr io , and they coincide since both have dimension n − r.
Theorem: Let Y be a submanifold of X of codimension r. Any point q ∈ Y has a coordinate

neighborhood (U ; u1 , . . . , un ) in X such that
U ∩ Y = {x ∈ U : u1 (x) = . . . = ur (x) = 0}.
Proof: Since the inclusion i : Y → X is a local embedding, there is a coordinate neighbor-

hood (V ; y1 , . . . , ym ) of q in Y and a coordinate neighborhood (U ; u1 , . . . , un ) of q in X, where
i(y1 , . . . , ym ) = (y1 , . . . , ym , 0 . . . , 0); hence yi = ui |Y .
Since Y is a subspace of X, replacing U by a smaller neighborhood, V = U ∩ Y .
We may assume that U is an open set of Rn , and U ∩ Y = V × 0, where V is an open set of
m
R ; now the statement is obvious.
Corollary: Any submanifold is a locally closed subspace (a closed set in an open subset).
10.2 Linear Connections

Definition: A linear connection on a smooth manifold X assigns to any pair of vector fields
D1 , D2 a vector field D1∇ D2 , the covariant derivative of D2 in the direction D1 , so that
1. D∇ (D1 + D2 ) = D∇ D1 + D∇ D2 ,
D∇ (f D̄) = (Df )D̄ + f D∇ D̄.
2. (D1 + D2 )∇ D = D1∇ D + D2∇ D,

(f D)∇ D̄ = f (D∇ D̄).
Lemma: ∇ may be uniquely extended to a derivation of tensors preserving the type and
1. D∇ f = Df .
2. D∇ (Cij T ) = Cij (D∇ T ).
Proof: To prove the uniqueness, we derive ω(D̄) = C11 (ω ⊗ D̄),
(D∇ ω)(D̄) = D(ω(D̄)) − ω(D∇ D̄)

1...p+q
and deriving T (D1 , . . . , ωq ) = C1...p+q (D1 ⊗ . . . Dp ⊗ T ⊗ ω1 ⊗ . . . ⊗ ωq ), we see that
(D∇ T )(D1 , . . . , ωq ) = D(T (D1 , . . . , ωq )) − T (D∇ D1 , . . . , ωq ) − . . . − T (D1 , . . . , D∇ ωq ).
To prove the existence, just take these formulae as definition.
Definition: The covariant differential ∇T of a (p, q)-tensor T is the (p + 1, q)-tensor
(∇T )(D, D1 , . . . , Dp , ω1 , . . . ωq ) = (D∇ T )(D1 , . . . , Dp , ω1 , . . . ωq )
and we say that T is parallel or constant when ∇T = 0; i.e., D∇ T = 0 with any vector field D.
Lemma: If D or D̄ vanishes on an open set U , so does D∇ D̄.
Proof: If x ∈ U , take φ ∈ C ∞ (X) such that supp φ ⊆ U , and φ(x) = 1.

If φD̄ = 0, then 0 = D∇ (φD̄) = (Dφ)D̄ + φ(D∇ D̄), and 0 = λD̄x + (D∇ D̄)x = (D∇ D̄)x .
If φD = 0, then 0 = (φD)∇ D̄ = φ(D∇ D̄), and (D∇ D̄)x = 0. q.e.d.
This lemma shows that ∇ induces a linear connection on any open set U of X, so that
(D∇ D̄)|U = (D|U )∇ (D̄|U ).
On a coordinate neighborhood (U ; x1 , . . . , xn ) the connection is given by the Christoffel

symbols Γkij ∈ C ∞ (U ),
∂i∇ ∂j = k Γkij ∂k .
P
Example: On any real vector space of finite dimension E there is a unique linear connection
∇ such that the fields De defined (p. 276) by the vectors e ∈ E are parallel, D∇ (De ) = 0.
The uniqueness is obvious, and to prove the existence, just consider the connection with null
Christoffel symbols in the coordinate system (x1 , . . . , xn ) defined by some base of E,
D∇ (f1 ∂1 + . . . + fn ∂n ) = (Df1 )∂1 + . . . + (Dfn )∂n .
Definition: A vector field on X with support on a curve σ : I → X is a family of vectors

{Dt }t∈I , where Dt ∈ Tσ(t) X, and it is smooth (as we shall always assume) if the function
(Df )(t) = Dt f is smooth for any smooth function f on an open set U ⊆ X, so that it defines a
derivation D : C ∞ (X) → C ∞ (I).
The C ∞ (I)-module of fields with support on σ is denoted by Dσ = DerR (C ∞ (X), C ∞ (I)).
P
In a coordinate neighborhood, any vector field with support is Dt = i fi (t)∂i .
Any vector field T on I defines vector field with support on σ
σ∗ T
σ∗ (Tt ) : C ∞ (X) −−−→ C ∞ (I) −−→ C ∞ (I),
and any vector field D on X also defines a vector field with support on σ
D σ∗
Dσ(t) : C ∞ (X) −−→ C ∞ (X) −−−→ C ∞ (I).
and the following lemma states that T ∇ D is well defined as a vector field with support on σ:
Lemma: Given a linear connection ∇, the vector (D∇ D̄)p only depends on Dp and the value of
D̄ along a curve tangent to Dp .
P P
Proof: In a coordinate neighborhood, D = i fi ∂i , D̄ = j gj ∂j ,
D∇ D̄ = (Dgj )∂j + gj D∇ ∂j = (Dgj )∂j + gj fi Γkij ∂k ,

P P P P
j j j i,j,k
∇
gj (p)fi (p)Γkij (p)(∂k )p ,
P P
(D D̄)p = (Dp gj )(∂j )p +
j i,j,k
and (D∇ D̄)p is determined by the values fi (p), i.e. Dp , the values gj (p), i.e. D̄p , and the values
Dp gj , fully determined by the values of gj on a curve tangent to Dp .
Lemma: ∂t∇ : D(X) → Dσ may be uniquely extended to a covariant derivation of vector fields
with support ∂t∇ : Dσ → Dσ ,
∂t∇ (D + D0 ) = ∂t∇ D + ∂t∇ D0 ; D, D0 ∈ Dσ ,

∂t∇ (f (t)D) = f 0 (t)D + f (t)(∂t∇ D); f ∈ C ∞ (I), D ∈ Dσ .
Proof: Locally, if σ(t) = (x1 (t), . . . , xn (t)), the unique possible extension is
P P P
∂t∇ fi (t)∂i = (fi0 (t)∂i + fi (t) ∂t∇ ∂i ) = fi0 (t)∂i + fi (t) x0j (t)Γkji (σ(t))∂k
P P
i i i i j,k
Definition: A vector field D with support on a curve σ is parallel or constant if ∂t∇ D = 0,

and the curve is a geodesic of ∇ if the tangent vector is parallel, ∂t∇ ∂t = 0.
Theorem: Given a curve σ : I → X and a vector Dp ∈ Tp X, p = σ(t0 ), there is a unique

parallel vector field D with support on σ such that Dt0 = Dp , and this parallel transport of
vectors Tp X → Tσ(t) X, Dp 7→ Dσ(t) is a linear isomorphism (depending on the curve σ).
Proof: In a coordinate neighborhood, ∂t∇ ∂i = j hij (t)∂

P
P j.
The condition of a vector field with support D = i fi (t)∂i being parallel,
0 = ∂t∇ D = 0
P P
i fi (t)∂i + i,j hij (t)fi (t)∂j ,
states that the functions fi define a solution of the linear differential equation
fi0 = −
P
j hji (t)fj ,
P
with the initial condition fi (t0 ) = ai , when Dp = i ai ∂i . Hence (p. 252) the parallel transport
of Dp , and of a base of Tp X, exists, and it is unique, on a neighborhood V of t0 .
Since the parallel transport is linear, it is defined on V for any vector of Tp X.
Now it is clear that the parallel transport is defined on the whole interval I.
Theorem: Given Dp ∈ Tp X, there exists a geodesic σ : I → X passing through p at t = 0 with

tangent Dp , and any two coincide on the common definition interval.
Proof: Locally, σ(t) = (x1 (t), . . . , xn (t)) is a geodesic if
∂t = x01 (t)∂1 + . . . + x0n (t)∂n ,

∂t∇ ∂t = x00i (t)∂i +
P 0
xi (t)x0j (t)Γkij ∂k = 0,
P
i i,j,k
x00k (t) = − i,j x0i (t)x0j (t)Γkij (x1 (t), . . . , xn (t)),

P
and the theorem follows from the existence and uniqueness of the solution of a differential
equation (p. 252).
10.2.1 Torsion and Curvature

Definition: The torsion Tor∇ of a linear connection ∇ is the (2, 1)-tensor
Tor∇ (D1 , D2 ) = D1∇ D2 − D2∇ D1 − [D1 , D2 ]
and ∇ is symmetric if the torsion is null, [D1 , D2 ] = D1∇ D2 − D2∇ D1 , Γkij = Γkji .
Since Tor∇ (D, D) = 0, we only have to show that Tor∇ is C ∞ (X)-linear in the first variable,
Tor∇ (f D1 , D2 ) = f D1∇ D2 − (D2 f )D1 − f D2∇ D1 + [D2 , f D1 ]

= f D1∇ D2 − (D2 f )D1 − f D2∇ D1 + (D2 f )D1 + f [D2 , D1 ]
= f D1∇ D2 − f D2∇ D1 − f [D1 , D2 ] = f Tor∇ (D1 , D2 ).
Definition: The curvature R of ∇ is the (3, 1)-tensor (alternate in the two first indices)
R(D1 , D2 , D3 ) = D1∇ (D2∇ D3 ) − D2∇ (D1∇ D3 ) − [D1 , D2 ]∇ D3 ,
Let us see that R is C ∞ (X)-linear in the third index. Given vector fields δ1 , δ2 , δ3 ,
R(δ1 , δ2 , f δ3 ) = δ1∇ ((δ2 f )δ3 + f δ2∇ δ3 ) − δ2∇ ((δ1 f )δ3 + f δ1∇ δ3 ) − ([δ1 , δ2 ]f )δ3 − f [δ1 , δ2 ]∇ δ3 =
= (δ1 δ2 f )δ3 + (δ2 f )δ1∇ δ3 + (δ1 f )δ2∇ δ3 + f δ1∇ δ2∇ δ3 − −(δ2 δ1 f )δ3 − (δ1 f )δ2∇ δ3 −
− (δ2 f )δ1∇ δ3 − f δ2∇ δ1∇ δ3 − (δ1 δ2 f )δ3 + (δ2 δ1 f )δ3 − f [δ1 , δ2 ]∇ δ3 =
= f δ1∇ δ2∇ δ3 − f δ2∇ δ1∇ δ3 − f [δ1 , δ2 ]∇ δ3 = f R(δ1 , δ2 , δ3 ).
Definition: A linear connection ∇ is flat if there are local bases D1 , . . . , Dn of parallel vector
fields, ∇Di = 0; and it is locally Euclidean if any point admits a local coordinate system
(x1 , . . . , xn ) with null Christoffel symbols, ∂i∇ ∂j = 0.
Theorem: A linear connection ∇ is flat if and only if the curvature tensor R is null.
Proof: If ∇ is flat, then R(Di , Dj , Dk ) = Di∇ Dj∇ Dk − Dj∇ Di∇ Dk − [Di , Dj ]∇ Dk = 0, since
D∇ Dk = 0 for any vector field D; hence R = 0.
Conversely, if R = 0, we may assume that we are at the origin of Rn .
Take a vector D0 ∈ T0 Rn , and extend it by parallel transport along the axis OX1 ; then along
the lines parallel to OX2 , so that we get a vector field on the plane OX1 X2 , and so on.
This field D is smooth (the solutions of a differential equation smoothly depend on the initial
conditions, p. 254) and ∂1∇ D = 0 on OX1 , ∂2∇ D = 0 on OX1 X2 , . . . , ∂n∇ D = 0 on Rn .
Let us show ∂r∇ D = 0, by descending induction. If ∂r+1 ∇ D = . . . = ∂ ∇ D = 0, then
n
∇
0 = R(∂r+1 , ∂r , D) = ∂r+1 ∂r∇ D − ∂r∇ ∂r+1
∇ ∇
D = ∂r+1 ∂r∇ D,
so that ∂r∇ D is constant along the lines parallel to OXr+1 .

Since ∂r∇ D vanishes on the submanifold OX1 . . . Xr , it is null on OX1 . . . Xr+1 .
Repeating the argument with the equality R(∂r+2 , ∂r , D) = 0 we obtain that ∂r∇ D vanishes
on OX1 . . . Xr+2 , and finally we see that ∂r∇ D = 0 on Rn . The vector field D is parallel.
Repeating the argument with a base at the origin, we obtain a base of parallel vector fields.
Corollary: ∇ is locally Euclidean if and only if R = 0 and Tor∇ = 0.
Proof: If ∇ is locally Euclidean, it is clear that R = 0 and Tor∇ = 0.

Conversely, if R = 0, there are local bases of parallel vector fields D1 , . . . , Dn .
If moreover the torsion is null, [Di , Dj ] = Di∇ Dj −Dj∇ Di = 0, and there exist local coordinate
systems where Di = ∂i (p. 264). Now ∂i∇ ∂j = Di∇ Dj = 0.
Bianchi’s Identity: R(D1 , D2 , D3 ) + R(D2 , D3 , D1 ) + R(D3 , D1 , D2 ) = 0, when Tor∇ = 0.

Proof: Since R is a tensor, we may check the identity at a point, and assume that [Di , Dj ] = 0,
D1∇ D2∇ D3 − D2∇ D1∇ D3 + D2∇ D3∇ D1 − D3∇ D2∇ D1 + D3∇ D1∇ D2 − D1∇ D3∇ D2 =
= D1∇ (D2∇ D3 − D3∇ D2 ) + D2∇ (D3∇ D1 − D1∇ D3 ) + D3∇ (D1∇ D2 − D2∇ D1 ) = 0.
¯
Difference of Two Connections: The difference T (D1 , D2 ) = D1∇ D2 − D1∇ D2 of two linear
¯ ∇ is a (2, 1)-tensor. It is clearly C ∞ (X)-linear in D1 , and
connections ∇,
¯
T (D1 , f D2 ) = (D1 f )D2 + D1∇ D2 − (D1 f )D2 − D1∇ D2 = f T (D1 , D2 ).
¯ and ∇ have equal torsion if and only if T is symmetric.

1. ∇
¯ and ∇ have the same geodesics if and only if T is alternate.
2. ∇
¯
On a common geodesic, T (Dp , Dp ) = (D∇ D)p − (D∇ D)p = 0 + 0.
¯
Conversely, if T is alternate and D∇ D = 0, so is D∇ D = T (D, D) + D∇ D = 0.
¯ and ∇ have equal geodesics and torsion, they coincide.
3. If ∇
If T is symmetric and alternate, it vanishes.
4. The geodesics of ∇ are the geodesic lines of a unique symmetric connection.

The uniqueness follows from statement 3, and to prove the existence, we consider the con-
¯ ¯ have
nection D1∇ D2 = D1∇ D2 − 12 Tor∇ (D1 , D2 ). Since T = − 12 Tor∇ is alternate, ∇ and ∇
¯
the same geodesics. Finally, ∇ is symmetric:
¯ ¯
D1∇ D2 − D2∇ D1 − [D1 , D2 ] = D1∇ D2 − D2∇ D1 − [D1 , D2 ] − Tor(D1 , D2 ) = 0.
Newtonian Gravitation: A Galilean spacetime is a real affine space A4 whose space of

free vectors V is endowed with a with a 1-dimensional vector space Q of symmetric metrics,
generated by a metric g of type (+, 0, 0, 0), so that g = ω ⊗ ω for a 1-form ω (and g is named
the time metric because the time interval between two events p and q = p + v is defined to
√ p
be v · v = g(v, v) = |ω(v)|); and a Euclidean structure on the subspace E = rad g = ker ω of
spatial (or simultaneity) vectors, defined by a contravariant metric h∗ (ω1 , ω2 ) := hω1 |E , ω2 |E i
of type (0, +, +, +), named space metric1 .
An inertial reference system is defined to be an event p ∈ A4 , the origin, and a base
e0 , ~e1 , ~e2 , ~e3 of V such that ω(e0 ) = 1 and ~e1 , ~e2 , ~e3 is an orthonormal base of E. Now any event
is q = p + te0 + x1 e1 + x2 e2 + x3 e3 , and we say that (t, x1 , x2 , x3 ) are the inertial coordinates
of q. The time function t is well-defined up to the addition of a constant, because ω = dt,
and the trajectories of punctual bodies are defined to be the smooth sections σ : R → A4 of
t (curves where g is positive-definite), so that σ(t) = t, x1 (t), x2 (t), x3 (t) . The gravitational
force intensity (force per mass unit) is given by a field of spatial vectors F~ = F1 ∂1 + F2 ∂2 + F3 ∂3 ,
and freely falling trajectories are given by the newtonian law of motion x00i (t) = Fi .
There exists a unique symmetric linear connection ∇, the Cartan connection, such that
the freely falling trajectories are just the geodesic trajectories of ∇:
The difference tensor T of two such connections fulfills that T (D, D) = 0 when ω(D) = 1;
hence when ω(D) 6= 0 and, by continuity, for any tangent vector D, so that T is alternate. Since
T is symmetric (both connections are torsionless) we see that T = 0.
1
Both metrics g and h∗ are well defined up to a positive factor, and to fix g and h is just to fix time and length
units. Then the 1-form ω is well-defined up to a sign, and to fix it is just to fix the time orientation.
To prove the existence, just consider the following linear connection:
∂t∇ ∂t = −F~ , ∂t∇ ∂i = ∂i∇ ∂t = ∂i∇ ∂j = 0.
This connection ∇ (with the time and space metrics g, h) enables us to express all the
elements of the newtonian theory of gravitation, so rendering superfluous the flat connection of
A4 (the inertial reference systems):
1. Freely falling trajectories are just geodesic trajectories of ∇.
2. The curvature tensor R of ∇ defines a tensor R2,2 on Λ2 (T X), uniquely determined by the
condition R2,2 (D1 ∧ D2 , V~3 ∧ D4 ) = hR(D1 , D2 , D4 ), V
~3 i, because R(D1 , D2 , D4 ) always is
spatial. The conservative character of gravitational forces, ∂i Fj = ∂j Fi , is just the symmetry
condition2 R2,2 (D1 ∧ D2 , D3 ∧ D4 ) = R2,2 (D3 ∧ D4 , D1 ∧ D2 ).
3. When the matter distribution is given by a mass density function ρ on A4 , Gauss equation
∂1 F1 + ∂2 F2 + ∂3 F3 = −4πρ, the field equation encoding the inverse square gravitation law,
states that the Ricci tensor R2 of ∇ (the contraction R2 = C11 R of the first indices of the
curvature tensor) is3 just R2 = 4πρ dt ⊗ dt.
Physical Meaning of the Curvature: If D is a geodesic vector field, D∇ D = 0, the integral
curves should be understand as trajectories of inertial bodies. The apparent position of nearby
bodies is represented by a vector field E such that DL E = 0, the relative velocity by D∇ E and
the relative acceleration by D∇ D∇ E. If ∇ is symmetric, the curvature appears as a non null
relative acceleration of inertial bodies (in absence of forces!)
R(D, E, D) = D∇ E ∇ D − E ∇ D∇ D − [D, E]∇ D = D∇ E ∇ D = D∇ D∇ E.
10.3 Riemannian Manifolds

Definition: A semiriemannian metric g is a non-singular symmetric 2-covariant (smooth)
tensor field, and it is riemannian when it is positive-definite.
A riemannian manifold is a smooth manifold X with a riemannian metric4 , and isome-
tries are diffeomorphisms ϕ : X → Y preserving the metrics, gX = ϕ∗ (gY ).
In a riemannian manifold (X, g), we put D · D̄ = g(D, D̄).
Proposition: Any smooth manifold X admits a riemannian metric.

S
Proof: Let {φi } be a partition of unity subordinated to a cover X = i Ui by coordinate open
sets, and let us fix a riemannian
P metric gi on each open set Ui . P
Now the metric g = i φi gi is positive definite: if 0 6= D ∈ Tx X, then 0 < i φi (x)gi (D, D)
because some term is positive and no term is negative.
Fundamental Theorem of Riemannian Geometry: If g is a semiriemannian metric, there

exists a unique symmetric linear connection ∇ (the Levi-Civita connection of g) such that
∇g = 0.
Proof: The condition ∇g = 0 states that the connection derives the product defined by g,
D(D1 · D2 ) = (D∇ D1 ) · D2 + D1 · (D∇ D2 ),

2
typical of the Levi-Civita connections, even if ∇ does not preserve any non-singular metric.
3
so that the matter distribution is fully determined by the geometry (g, h, ∇) of spacetime.
4
In fact, it is only well-defined up to a positive constant factor, and to fix it is just to fix the length unit.
10.3. RIEMANNIAN MANIFOLDS 287
so that for any three vector fields X, Y, Z we have
X(Y · Z) = (X ∇ Y ) · Z + Y · (X ∇ Z),
Z(X · Y ) = (Z ∇ X) · Y + X · (Z ∇ Y ),
Y (X · Z) = (Y ∇ X) · Z + X · (Y ∇ Z).
Adding them with alternate signs, and using that [D1 , D2 ] = D1∇ D2 − D2∇ D1 , we obtain
(∗) 2(X ∇ Y ) · Z = X(Y · Z) + Y (X · Z) − Z(X · Y ) − X · [Y, Z] − Y · [X, Z] + Z · [X, Y ].
Since g is non singular, this equality determines X ∇ Y , so proving the uniqueness of ∇.

To see the existence, just define ∇ by the above equality, and it is easy to check that it is a
symmetric connection and ∇g = 0. q.e.d.
In a local coordinate system, if we put gij = ∂i · ∂j and (g ij ) denotes the inverse matrix of
(gij ), then (∗) let us express the Christoffel symbols of ∇ in terms of the metric g,

1 X kh ∂gjh ∂gij ∂ghi
Γkij = g − + .
2 ∂xi ∂xh ∂xj
h
Corollary: The geodesic curves have tangent vector of constant module.
Proof: ∂t (∂t · ∂t ) = (∂t∇ ∂t ) · ∂t + ∂t · (∂t∇ ∂t ) = 0 · ∂t + ∂t · 0 = 0.
Definition: A riemannian
P metric g is locally Euclidean if any point admits a coordinate
neighborhood where g = i dxi ⊗ dxi (i.e., gij = ∂i · ∂j = δij ).
Theorem: A metric g is locally Euclidean if and only if the curvature tensor R is null.
Proof: If g = i dxi ⊗ dxi , then we have Γkij = 0, and R = 0.

P
Conversely, if R = 0, then (p. 284) each point p has a neighborhood with a base of parallel
vector fields (D1 , . . . , Dn ), and we may assume that it is an orthonormal base at p.
They form an orthonormal base on a neighborhood since Di · Dj is locally constant,
Dk (Di · Dj ) = (Dk∇ Di ) · Dj + Di · (Dk∇ Dj ) = 0 + 0 = 0.
Moreover [Di , Dj ] = Di∇ Dj − Dj∇ Di = 0, because ∇ is symmetric, and on a coordinate

neighborhood we have Di = ∂i (p. 264), so that gij = Di · Dj = δij .
Definition: The Riemann-Christoffel tensor R2,2 of a riemannian metric g is
R2,2 (D1 , D2 ; D3 , D4 ) = R(D1 , D2 , D4 ) · D3 .
Theorem: 1. R2,2 (D2 , D1 ; D3 , D4 ) = −R2,2 (D1 , D2 ; D3 , D4 ).

2. R2,2 (D1 , D2 ; D4 , D3 ) = −R2,2 (D1 , D2 ; D3 , D4 ).
3. R2,2 (D3 , D4 ; D1 , D2 ) = R2,2 (D1 , D2 ; D3 , D4 ).
4. The cyclic sum on any three indices is null.
Proof: (1) R is alternate in the first two indices.

(2) Since R2,2 is a tensor, we may assume that [Di , Dj ] = 0,
0 = [D1 , D2 ](D3 · D4 ) = (D1 D2 − D2 D1 )(D3 · D4 ) =

= D1 (D2∇ D3 · D4 + D3 · D2∇ D4 ) − D2 (D1∇ D3 · D4 + D3 · D1∇ D4 ) =
= D1∇ D2∇ D3 · D4 + D2∇ D3 · D1∇ D4 + D1∇ D3 · D2∇ D4 + D3 · D1∇ D2∇ D4 −
− D2∇ D1∇ D3 · D4 − D2∇ D3 · D1∇ D4 − D1∇ D3 · D2∇ D4 − D3 · D2∇ D1∇ D4 =
= R2,2 (D1 , D2 ; D4 , D3 ) + R2,2 (D1 , D2 ; D3 , D4 ).
(3) Bianchi’s identity states that the cyclic sum, fixing the third index, is null, hence
R2,2 (D1 , D3 ; D2 , D4 ) + R2,2 (D4 , D1 ; D2 , D3 ) + R2,2 (D3 , D4 ; D2 , D1 ) = 0

Adding them, and using 1 and 2, the two first columns cancel and we obtain
R2,2 (D1 , D2 ; D3 , D4 ) = −R2,2 (D3 , D4 ; D2 , D1 ) = R2,2 (D3 , D4 ; D1 , D2 ).
(4) The cyclic sum, fixing the third index, is null by Bianchi’s identity; hence so is fixing any
other index since R2,2 is invariant under the Klein group according to 1, 2, 3.
Definition: In a riemannian manifold (X, g), the codifferential δω and the laplacian ∆ω of
a differential p-form ω are defined to be (the contraction of indices C12 with g)
δω = C12 (∇ω),
∆ω = (dδ + δd)ω.
If g is a semiriemannian metric, = dδ + δd is said to be the lambertian operator.
Sectional Curvatures: The tensor R2,2 may be viewed as a symmetric metric on Λ2 Tp X,
R2,2 (D1 ∧ D2 , D3 ∧ D4 ) = R2,2 (D1 , D2 ; D3 , D4 ),
where we also have (p. 67) the metric Λ2 g induced by the riemannian metric g,
(Λ2 g)(D1 ∧ D2 , D3 ∧ D4 ) = (D1 · D3 )(D2 · D4 ) − (D1 · D4 )(D2 · D3 ).
Definition: If Π ⊆ Tp X is a plane, then dim Λ2 Π = 1 and both metrics are proportional, with
a factor KΠ , the sectional curvature of Π. If D1 , D2 is a base of Π,
R2,2 (D1 , D2 ; D1 , D2 )
KΠ = ·
(D1 · D1 )(D2 · D2 ) − (D1 · D2 )2
and g is a metric of constant curvature if it does not depend on Π nor p.
1. (K = 0). The Euclidean space En = (Rn , i dx2i ) has null curvature. It is simply connected
P
and the connection is complete (the geodesics are defined for any value −∞ < t < ∞).
2. (K > 0). The sphere Sn of radius r in En+1 has constant curvature K = 1/r2 . It is
simply connected and complete, and the geodesic lines are just the intersections with the
planes through the center of the sphere. The stereographic projection from a pole onto the
equatorial hyperplane gives the metric
4(dx21 + . . . + dx2n )
2 , (K = 1/r2 ).
1+ K(x21 + ... + x2n )
on Rn , so that the geodesics are the circles and straight lines intersecting at opposite points
the sphere of radius r with center at the origin of Rn .
3. (K < 0). Let us consider on Rn+1 the metric g = dt2 − dx21 − . . . − dx2n . The hyperbolic
space Hn of curvature K = −1/r2 is the hypersurface
t2 − x21 − . . . − x2n = r2 , t > 0,
endowed with the restriction of −g. It is simply connected and complete, the geodesics being
the intersections with the planes through the origin of Rn+1 .
Klein’s Projective Model : Via the projection Rn+1 → Pn , the hyperbolic space is identified
with an open set in Pn bounded by a non-singular quadric of index 1, the geodesics being the
lines. In affine coordinates the quadric is i yi2 = 1, and the metric is
P
P 2 P 2
yi )( dyi ) + ( yi dyi )2
P
2 (1 −
r ·
(1 − yi2 )2
P
Poincaré’s Disk : If we project the hyperboloid Hn from the point (−r, 0, . . . , 0) onto the
equatorial hyperplane t = 0, we may identify Hn with the open ball in Rn of radius r, the
geodesics being the lines through the origin and the circles orthogonally intersecting the
boundary of the ball. The metric is
4(dx21 + . . . + dx2n )
2 , (K = −1/r2 ).
1 + K(x21 + . . . + x2n )
Poincaré’s Half-plane: Apply to the Poincaré’s disk an inversion and a translation, we may
identify it with the semiplane xn > 0, the geodesics being the lines and the circles orthogonally
intersecting the hyperplane xn = 0. The metric is
r2
(dx21 + . . . + dx2n ).
x2n
In the case of surfaces, the unique sectional curvature is the curvature K of the surface at the
given point, and it determines the tensor R2,2 ; hence the curvature tensor R. A riemannian
surface of null constant curvature is locally Euclidean.
10.3.1 Normal Coordinates

Let X be a smooth manifold, and let us consider natural projection
`
π : T X := Tx X −→ X , π(Dx ) = x.
x∈X
If (x1 , . . . , xn ) are coordinates on an open set U ⊆ X, then any vector Dx ∈ π −1 (U ) is fully

determined by the coordinates (x1 , . . . , xn ) of x and the coordinates yi = Dx (xi ) of Dx in the
base (∂1 )x , . . . , (∂n )x of Tx X. On T X we have a unique structure of smooth manifold such that
π −1 (U ), x1 , . . . xn , y1 , . . . , yn ) are local coordinate systems, and we say that the smooth map
π : T X → X is the tangent bundle of X. The smooth vector fields D on any open set U ⊆ X
correspond to the smooth sections s : U → T X of π (i.e. π ◦ s = IdU ).
Analogously we may construct the cotangent bundle T ∗ X → X and the bundles of tensors
q
Tp X → X, so that the smooth sections are the smooth 1-forms and the smooth tensor fields of
type (p, q) respectively.
Theorem: Let p be a point of a riemannian manifold X. The tangent vectors D ∈ Tp X such

0 (0) = D, is defined at t = 1 form an open neighborhood
that the geodesic curve σD (t), where σD
U0 of 0 in Tp X, and the exponential map
expp : U0 −→ X , D 7→ σD (1),
is a smooth map. Moreover, it is a local diffeomorphism at 0.
Proof: Any smooth curve γ : I → X has a canonical lifting γ̃ : I → T X, γ̃(t) = Tt = γ∗,t (∂t ), so
that the geodesic curves define a vector field Z on T X, and it is smooth because in any local
coordinate system (x1 , . . . xn , y1 , . . . , yn ) we have (p. 283)
 
n n n
X ∂ X X ∂
Z= yi −  Γkij yi yj  ·
∂xi ∂yk
i=1 k=1 i,j=1

If τt is the flow of Z, then π τt (D) = σD (t), so that expp = π ◦ τ1 ; hence (p. 254) U0 is
open and expp is smooth.
Moreover, since exp tD = σtD (1) = σD (t), the linear tangent map Tp X = T0 (U0 ) → Tp X is
just the identity, and we conclude by the inverse mapping theorem.
Proposition: Let ϕ, φ : X → Y be local isometries, X connected. If ϕ(p) = φ(p) and ϕ∗,p = φ∗,p ,
then ϕ = φ.
Proof: The isometries commute with the exponential maps, so that we have commutative squares
ϕ∗,p φ∗,p
Tp X / Tϕ(p) Y , Tp X / Tφ(p) Y
expp expϕ(p) expp expφ(p)

ϕ φ
X /Y X /Y
Hence the set of points x ∈ X where ϕ(x) = φ(x) and ϕ∗,x = φ∗,x is an open set.
Since it is always closed, we conclude.
Corollary: Let M = En , Sn or Hn . Given p, q ∈ X and a linear isometry φ : Tp X → Tq X, there

exists a unique isometry ϕ : M → M such that ϕ(p) = q and ϕ∗,p = φ.
Proof: We only have to prove the existence, the Euclidean case being obvious.
In the case of the sphere Sn of radius r in Rn+1 , just consider the restriction ϕ of the isometry
1⊥φ
Rn+1 = Rp ⊥ Tp Sn −−−−−→ Rq ⊥ Tq Sn = Rn+1 ,
and the case Hn is totally analogous: Rn+1 = Rp ⊥ Tp Hn → 1 ⊥ φRq ⊥ Tq Hn = Rn+1 .
Definition: Let p be a point of a riemannian manifold X. The exponential map defines a

diffeomorphism of an open ball B(0, r) in Tp X onto a neighborhood U of p in X. Once we fix
an orthonormal base in Tp X, we may identify it with Rn , and so we obtain a local coordinate
system (x1 , . . . , xn ) in U , named normal coordinates at p, such that
1. The coordinates of the point p are just (0, . . . , 0).
2. The curves xi = ai t are geodesic lines.
3. The vectors ∂x∂ 1 p , . . . , ∂x∂n p define an orthonormal base of Tp X.

qP
2
Gauss Lemma: Let (U ; x1 , . . . , xn ) be normal coordinates at p and put r = i xi . On U − p
P xi
the gradient of r is the tangent vector field N = i r ∂i P defined by the geodesic curves with
unitary tangent vector at p. In particular |grad r| = 1 and i xi gij = xj .
Proof: By definition N ∇ N = 0, |N | = 1, and a direct calculation shows that N r = 1.

For any vector D ∈ Tx X, we have to prove that
N · D = Dr
and it is clear when D = Nx , because N · N = 1 = N r. When D is tangent to a ”sphere” r = a,

so that Dr = 0 and we have to prove that N · D = 0, we may extend it so as to obtain a vector
field D such that [N, D] = 0. Now, N · D is constant along the geodesics through p:
N (N · D) = (N ∇ N ) · D + N · (N ∇ D) = 0 + (D∇ N ) · N = 12 D(N · N ) = 12 D(1) = 0.
Since the uniparametric group of N shrinks the spheres as we approach p, and it leaves
D invariant, we have that D → 0 as we approach p along the geodesic joining x with p. We
conclude that N · D = 0.
P
Finally, |grad r| = |N | = 1, and the equality N · ∂j = ∂j r gives that i xi gij = xj .
∂gij
Corollary: In normal coordinates, (p) = 0 and Γkij (p) = 0.
∂xk
P
Proof: Deriving h xh ghj = xj with ∂k ∂i at x = p, we obtain (∂k gij )(p) = −(∂i gkj )(p).
Since (∂k gij )(p) is symmetric in i, j and alternate in i, k, it is also alternate in j, k; hence in
i, j, and it vanishes.
Definition: A map γ : [a, b] → X is C ∞ when it is the restriction of a map (a − ε, b + ε) → X

of class C ∞ , and it is piecewise C ∞ when there is a partition a = x0 < x1 < . . . < xn = b such
that the restriction γ : [xi−1 , xi ] → X is C ∞ for any index i. Then the tangent vector Tt = γ∗ (∂t )
is well-defined up to a finite number of points, and the length of γ is
Z b
|Tt |dt.
a
Theorem: Let (U ; x1 , . . . , xn ) be normal coordinates at p. For any point x ∈ X there is a unique

(up to reparametrizations) piecewise C ∞ curve of pminimal length joining p and x, namely the
geodesic in U joining them, whose length is just x21 + . . . + x2n .
Proof: Since |N | = 1, we have (grad r) · T = N · T ≤ |T | for any tangent vector T , and the
equality holds if and only if T = λN , λ ≥ 0.
Now, let γ : [a, b] → U be a curve joining p and x ∈ U . Then
Z b Z b Z q
γ ∗ (dr) = r(x) − r(p) = x21 + . . . + x2n ,

length γ = |Tt | t ≤ (grad r) · Tt dt =
a a γ
and the equality holds just when Tt = λ(t)Nγ(t) , λ(t) ≥ 0; i.e. γ is a reparametrization of the
geodesic joining p and x in U .
Finally, if a curve γ joining p and x is not contained in U , it intersects some ”sphere”
x1 + . . . + x2n = r2 of radius r > r(p), so that the length is ≥ r.
2
Corollary: Any riemannian manifold is a metric space (with the same underlying topology):
d(p, q) = inf{length γ; γ a piecewise C ∞ curve joining p and q}.
Note: The normal neighborhood U of p is the image, by the exponential map, of an open ball
B(0, R), and the theorem shows that U is just the open ball B(p, r) of the metric d.
The theorem also shows that the function f (x) = d(p, x)2 is C ∞ in a neighborhood of p, has
null differential and the hessian is just 2gp . Hence the metric d (and the differentiable structure)
determine the riemannian metric g.
Proposition: Any local isometry ϕ : X → Y between complete and connected riemannian man-
ifolds is a covering.
Proof: Let B(y, r) be a normal neighborhood of y ∈ Y . It is enough to show that ϕ−1 B(y, r)

is just the disjoint union of the balls (they exist because X is complete) B(x, r), where ϕ(x) = y,
and that the maps ϕ : B(x, r) → B(y, r) are diffeomorphisms.
When ϕ(x) = y, the commutative square
Tx X ⊃ B(0, r) B(0, r) ⊆ Ty Y
 
expx  expy
y y
ϕ
X ⊇ expx B(0, r) −−−−→ B(y, r) ⊆ Y
shows that the surjective maps expx : B(0, r) → expx B(0, r) and ϕ : expx B(0, r) → B(y, r) are
in fact homeomorphisms.
Since ϕ : X → Y is a local homeomorphism, it follows that expx B(0, r) is open in X.
Now the commutative square shows that both homeomorphisms, being smooth maps, are
diffeomorphisms, so that expx B(0, r) = B(x, r), and ϕ : B(x, r) → B(y, r) is a diffeomorphism.
Finally, let us show that ϕ B(y, r) = qB(x, r). If y 0 = ϕ(x0 ) ∈ B(y, r), then the geodesic
−1

joining y 0 with y in B(y, r), of length < r, may be lifted to a geodesic in X, of length < r,
joining x0 with some point x ∈ ϕ−1 (y). Hence x0 ∈ B(x, r).
10.3.2 Surfaces of Constant Curvature

Minding’s Theorem: All surfaces of constant curvature K are locally isometric.
Proof: Let (X, g) be a surface, fix a point p ∈ X, and consider the geodesic curve γ, γ(0) = p,
tangent to a unitary vector Dp ∈ Tp X. Let Np , Dp be an orthonormal base in Tp X and consider
the parallel transport N of Np along γ, so that N · T = 0. Let σy be the geodesic curve tangent
to Nγ(y) and such that σy (0) = γ(y).
Then σ : R2 → X, σ(x, y) = σy (x), is a smooth map defined on a neighborhood of 0, and it
is a local diffeomorphism at 0 because σ∗ : T0 R2 → Tp X transforms ∂x , ∂y into the base Np , Dp .
Hence σ defines a local coordinate system (x, y) on a neighborhood of p = (0, 0).
By construction, ∂x∇ ∂x = 0 and |∂x | = 1, because ∂x is tangent to the geodesics σy . Moreover,
on the curve x = 0 we have
∂y∇ ∂y = ∂y∇ ∂x = 0 , |∂y | = 1 , ∂x · ∂y = 0.

10.4. RIEMANNIAN EMBEDDINGS 293
Now, ∂y · ∂x = 0 because it vanishes on x = 0 and, deriving ∂x · ∂x = 1, we have
0 = (∂y∇ ∂x ) · ∂x = (∂x∇ ∂y ) · ∂x = ∂x (∂y · ∂x ).
Hence the riemannian metric is g = dx2 + f 2 dy 2 , where f = |∂y |.

We have f (0, y) = 1, because |∂y | = 1 on x = 0, and deriving f 2 = ∂y · ∂y with ∂x we obtain
that 2f (∂x f ) = 2(∂x∇ ∂y ) · ∂y vanishes on x = 0, because so does ∂x∇ ∂y ; hence (∂x f )(0, y) = 0.
On the other hand, a direct calculation shows that the curvature K of dx2 + f 2 dy 2 is
∂x2 f
K=− , f 00 + Kf = 0.
f
When K is constant, integrating this ordinary differential equation, with the initial conditions
f (0, y) = 1, f 0 (0, y) = 0, we conclude

1 √
 when K = 0,
f (x, y) = cos K x when K > 0,
 √
cosh −K x when K < 0.

Theorem: Any simply connected riemannian surface X of constant curvature K admits a local
isometry X → M, unique up to an isometry of M (where M = E2 , S2 or H2 , depending on K).
Proof: Let Ix be the non empty set (by Minding’s theorem) of germs ϕx at x ∈ X of isometries
ϕ : U ,→ M of a connected neighborhood of x onto an open set of the model M. Now we put (as
in p. 189 and p. 265) `
π : Ie = x Ix −→ X, π(ϕx ) = x.
Any local isometry U → M defines a local section U → Ie of π. The images of these sections
generate a topology on I, e so that the continuous sections U → Ie of π correspond to the local
isometries U → M.
Moreover, π is in fact a covering of X, and even a principal bundle when endowed with the
natural left action of the group G of all isometries of M, because any germ ϕx is fully determined
(p. 290) by ϕ(x) and ϕ∗,x , so that any two such germs differ in a unique isometry of M.
If X is simply connected, it is a trivial principal bundle, Ie ' G × X, and it admits a
continuous section X → I, e unique up to the action of G.
Theorem: Any simply connected complete riemannian surface of constant curvature is isometric
to E2 , S2 or H2 .
Proof: The local isometry X → M is a covering, because X is complete.

Hence it is a trivial covering, because M is simply connected.
10.4 Riemannian Embeddings

Let (X; g, ∇) be a riemannian manifold, X̄ a submanifold of X, and ḡ the restriction of g to X̄.
If p ∈ X̄, we put Tp X = Tp X̄ ⊥ Np . If D1 , D2 are vector fields on X̄, in general D1∇ D2 has
a tangential component D1∇ D2 , and a normal component Φ(D1 , D2 ) = (D1∇ D2 )⊥ ,
D1∇ D2 = D1∇ D2 + Φ(D1 , D2 ).
Theorem: ∇ is the Levi-Civita connection of ḡ, and Φ is a symmetric C ∞ (X̄)-bilinear map,

(hence it defines symmetric bilinear maps Φp : Tp X̄ × Tp X̄ → Np ).
Proof: To see that ∇ is a symmetric connection and D(D1 · D2 ) = (D∇ D1 ) · D2 + D1 · (D∇ D2 ),

just project orthogonally onto Tp X̄ the corresponding properties of ∇, using that [D1 , D2 ] is
tangent to X̄ when so are D1 , D2 .
Moreover Φ is symmetric and bilinear,
0 = D1∇ D2 − D2∇ D1 − [D1 , D2 ]

= D1∇ D2 + Φ(D1 , D2 ) − D2∇ D1 − Φ(D2 , D1 ) − [D1 , D2 ]
= Φ(D1 , D2 ) − Φ(D2 , D1 ),
Φ(f D1 , D2 ) = (f D1∇ D2 )⊥ = f (D1∇ D2 )⊥ = f Φ(D1 , D2 ).
Weingarten’s Formula: Φ(D1 , D2 ) · N = −(D1∇ N ) · D2 ; where N is normal to X̄.
Proof: Φ(D1 , D2 ) · N = (D1∇ D2 ) · N = D1 (D2 · N ) − D2 · (D1∇ N ).
Gauss Equation: If R and R̄ are the Riemann-Christoffel tensors of X and X̄,
R̄(D1 , D2 ; D3 , D4 ) = R(D1 , D2 ; D3 , D4 ) + Φ(D1 , D3 ) · Φ(D2 , D4 ) − Φ(D1 , D4 ) · Φ(D2 , D3 ).
Proof: Since R, R̄ and Φ · Φ are tensors, we may assume that [Di , Dj ] = 0.
R(D1 , D2 ; D3 , D4 ) = (D1∇ D2∇ D4 − D2∇ D1∇ D4 ) · D3

= D1∇ D2∇ D4 − D2∇ D1∇ D4 + D1∇ (Φ(D2 , D4 )) − D2∇ (Φ(D1 , D4 )) · D3

= R̄(D1 , D2 ; D3 , D4 ) − Φ(D1 , D3 ) · Φ(D2 , D4 ) + Φ(D2 , D3 ) · Φ(D1 , D4 ).
Curves
Definitions: In the case of a curve of unitary tangent vector T , the curvature of the curve is
κ = |T ∇ T |, and it is null just when the curve is a geodesic. If the curve lies in a submanifold
X̄, it also has a geodesic curvature κg = |T ∇ T |, and a normal curvature κn = |Φ(T, T )|,
κ2 = κ2g + κ2n .
If the curve lies in an oriented Euclidean space of dimension 3, and the curvature does not
vanish at any point, the principal normal N = κ1 T ∇ T is orthogonal to T ,
0 = T (T · T ) = (T ∇ T ) · T + T · (T ∇ T ) = 2(T ∇ T ) · T = 2κ(N · T ),
and at any point of the curve we have the binormal B, so that (T, N, B) is a direct orthonormal
base. The torsion of the curve is τ = (T ∇ N ) · B.

∇
 T T = κN

Frénet Formulae: T ∇ N = −κT + τ B

 ∇
T B = −τ N
Proof: The first equality is just the definition of N , and the second one follows from the equality
(T ∇ N ) · T = T (N · T ) − N · (T ∇ T ) = −N · (κN ) = −κ.
Now, deriving with T the equality B · T = 0, we see that (T ∇ B) · T = 0; deriving B · N = 0,

that (T ∇ B) · N = −τ , and deriving B · B = 1, that (T ∇ B) · B = 0.
10.4. RIEMANNIAN EMBEDDINGS 295
Theorem: Fix a direct orthonormal base (Tp , Np , Bp ) at a point p ∈ R3 . Given smooth functions
κ(t) > 0, τ (t) on an interval I, then there exists a unique curve I → R3 of curvature κ and
torsion τ , such that the Frénet frame at t0 ∈ I is (Tp , Np , Bp ).
Proof: The Frénet frame T = (f1 , f2 , f3 ), N = (g1 , g2 , g3 ), B = (h1 , h2 , h3 ) of the sought curve
σ = (σ1 , σ2 , σ3 ) would define a solution of Frénet formulae and equations σi0 = fi ; which form a
system of 12 linear differential equations.
With the initial conditions σ(t0 ) = p, (Tt0 , Nt0 , Bt0 ) = (Tp , Np , Bp ), there exists a unique
solution on I (pp. 252, 283).
We only have to show that this solution defines an orthonormal base at each point, since
(Tp , Np , Bp ) is direct. If A = (T, N, B), we have to prove that At A = I, or AAt = I.
That is to say fi fj + gi gj + hi hj = δij . Now,
(fi fj + gi gj + hi hj )0 = fi0 fj + fi fj0 + gi0 gj + gi gj0 + h0i hj + hi h0j

= κgi fj + κgj fi + (τ hi − κfi )gj + (τ hj − κfj )gi − τ gi hj − τ gj hi = 0,
and the function fi fj + gi gj + hi hj is constant. Since the value at p is δij , we conclude.
Hypersurfaces
Definition: Let (X̄; ḡ, ∇ ) be a hypersurface of an Euclidean space. We may fix a unitary
normal vector field N on a neighborhood of any point of X̄, so that we have a true metric φ2 ,
the second fundamental form of X̄, such that (p. 293)
D∇ D0 = D∇ D0 + φ2 (D, D0 )N,
φ2 (D, D0 ) = (D∇ D0 ) · N = Φ(D, D0 ) · N.
The Weingarten endomorphism is the endomorphism φ attached to (ḡ, φ2 ),
φ2 (D, D0 ) = φ(D) · D0 .
Weingarten Formula: φ(D) = −D∇ N .
Proof: (D∇ N ) · N = 12 D(N · N ) = 0, and −(D∇ N ) · D0 = φ2 (D, D0 ), (p. 294).
Gauss Equation: The Riemann-Christoffel tensor R̄2,2 of X̄ is
R̄2,2 (D1 , D2 ; D3 , D4 ) = φ2 (D1 , D3 )φ2 (D2 , D4 ) − φ2 (D1 , D4 )φ2 (D2 , D3 ).
Proof: R2.2 = 0, and Φ(Di , Dj ) · Φ(Di0 , Dj0 ) = φ2 (Di , Dj )φ2 (Di0 , Dj0 ), (p. 294).
Codazzi-Mainardi Equation: D1∇ (φ(D2 )) − D2∇ (φ(D1 )) − φ([D1 , D2 ]) = 0.
Proof: Just take the tangential component in the following equality (due to the vanishing of the
curvature tensor of the Euclidean space),
0 = D1∇ D2∇ N − D2∇ D1∇ N − [D1 , D2 ]∇ N = D1∇ (φ(D2 )) − D2∇ (φ(D1 )) − φ([D1 , D2 ]).
Definitions: The endomorphism φ diagonalizes in an orthonormal base (p. 162), and the
principal curvatures of X̄ are the eigenvalues κi of φ (the sign changes if we replace N by
−N ). The curvature lines are the curves tangent to eigenvectors of φ (hence non null).
Euler Theorem: The principal curvatures κi are the normal curvatures of curvature lines.
Proof: If T is unitary and φ(T ) = κi T , then κi = φ(T )·T = φ2 (T, T ) = κn .
Proposition: Any closed connected hypersurface of the Euclidean space where every point is
umbilical, φ = λId, is a hyperplane or a hypersphere.
1
Proof: Put n = dim X̄. If φ = λId, then λ = n trφ is a smooth function, and by the Codazzi-
Mainardi equation,
0 = D1∇ (λD2 ) − D2∇ (λD1 ) − λ[D1 , D2 ] = (D1 λ)D2 − (D2 λ)D1 .
Taking D1 , D2 linearly independent we see that Dλ = 0 for any vector field D on X̄.
Hence λ is constant, since X̄ is connected.
φ = 0, D∇ N = 0, and N is locally constant, N = ai ∂i . Let us consider the
P
If λ = 0, thenP
vector field H = i xi ∂i , so that D∇ H = D. When D is tangent to the hypersurface,
D(H · N ) = (D∇ H) · N + H · D∇ N = D · N + H · 0 = 0.
P
That is to say, H · N = i ai xi is constant, and any point has a neighborhood contained in
a hyperplane. Since it is connected, the hypersurface is contained in a hyperplane, and since it
is closed, it is the whole hyperplane.
λ 6= 0, we have D∇ (H + λ1 N ) = D + λ1 D∇ N = 0; hence H + λ1 N =
P
If P i ai ∂i , and
1
|H − i ai ∂i | = |λ| on the hypersurface.
It is locally contained in the hypersphere i (xi − ai )2 = λ−2 , and we conclude as above.
P
Surfaces
If X̄ is a surface in an Euclidean space of dimension 3, at any point we have two principal
curvatures κ1 , κ2 (coincident at an umbilical point).
Let D1 , D2 be an orthonormal base such that φ(Di ) = κi Di .
If T = D1 cos ϑ + D2 sin ϑ is a unitary vector, the normal curvature of curves tangent to T is
κn = φ2 (T, T ) = φ(T ) · T = (κ1 D1 cos ϑ + κ2 D2 sin ϑ) · T = κ1 cos2 ϑ + κ2 sin2 ϑ.
The extremal values of the normal curvature are the principal curvatures.
Gauss Theorema Egregium: The product κ1 κ2 of the two principal curvatures is an intrinsic
invariant of the riemannian surface (X̄, ḡ).
Proof: By Gauss equation, κ1 κ2 is the curvature K of the surface,
K = R̄2,2 (D1 , D2 ; D1 , D2 ) = φ2 (D1 , D1 )φ2 (D2 , D2 ) − φ2 (D1 , D2 )2 = κ1 κ2 − 0.
10.5 Lie Groups

Definitions: A Lie group is a smooth manifold G, with a smooth group law G × G → G,
(x, y) 7→ xy, such that G → G, x 7→ x−1 is smooth.
The neutral or identity element is denoted by e.
A morphism of Lie groups is a smooth morphism of groups.
10.5. LIE GROUPS 297
Any element x of a Lie group G defines a diffeomorphism
Lx : G −→ G, Lx (g) = xg,
and a vector field D is (left) invariant when Lx D = D, ∀x ∈ G; i.e., Lx,∗ Dy = Dxy .

The invariant vector fields are stable under the Lie bracket, since Lx [D, D0 ] = [Lx D, Lx D0 ],
and they form the Lie algebra g of G.
Theorem: The map g → Te G, D 7→ De , is an isomorphism.
Proof: It is injective because any invariant vector field D is determined by the value at a point,
Dx = Lx (De ).
Now, given De ∈ Te G, we have to show that the vector field {Dx = Lx De } is smooth, i.e.,
that the following function is smooth for any f ∈ C ∞ (G),
h(x) = Dx f = De (f ◦ Lx ).
Let D be a vector field on G extending De , and let us consider on G × G the vector field
e = (0, D) and the function f˜(x, y) = xy. We conclude since h = (D
D e f˜ ) ,
G×e
e (a,e) f˜.
h(a) = De (f (ay)) = D
Corollary: Any Lie group admits a global base of vector fields, and it is orientable.
Theorem: If ψ : G → G0 is a morphism of Lie groups, then ψ∗ : g = Te G → Te G0 = g0 preserves

the Lie bracket, and ψ∗ (Dx ) = (ψ∗ D)ψ(x) , where D ∈ g.
Proof: Let us show that D0 = ψ∗ D when D0 is invariant and De0 = ψ∗ (De ).

Now, since ψLx = Lψ(x) ψ, we have
ψ∗ (Dx ) = ψ∗ (Lx De ) = Lψ(x) (ψ∗ De ) = Lψ(x) De0 = Dψ(x)

0
.
That is to say, ψ ∗ (D0 f ) = D(ψ ∗ f ), and therefore ψ ∗ ([D10 , D20 ]f ) = [D1 , D2 ](ψ ∗ f ).
Hence ψ∗ [D1 , D2 ]x = [D10 , D20 ]ψ(x) , and we conclude when x = e.
Lemma: Any invariant vector field is complete (the flow is defined on R × G).
Proof: The integral curve of an invariant vector field D passing through x at t = 0 is just
σx (t) = Lx (σe (t)) = xσe (t).
If σe (t) is defined on (−ε, ε), so is any integral curve σx (t), and D is complete. q.e.d.
A morphism of Lie groups gt : R → G defines a uniparametric group on G,
τ (t, x) = xgt , τs+t (x) = xgt+s = xgt gs = τs (τt x),
and Lx transforms integral curves ygt into integral curves xygt ; hence the infinitesimal generator
D is invariant (and De is the tangent vector at t = 0 to the curve gt ).
Theorem: Hom(R, G) = g.
Proof: It is enough to show that the integral curve gt : R → G of D ∈ g passing through e is a

morphism of groups. Since the integral curve passing through x is τt (x) = xgt ,
gs+t = τs+t (e) = τt (τs e) = τs (e)gt = gs gt .
Definition: The exponential map of G is exp : g → G, exp(D) = g1 , where gt is the morphism

corresponding to D. The tangent vector at t = 0 to the morphism φ(t) = gλt is λDe , so that it
corresponds to the vector field λD,
gt = exp(tD).
Theorem: The exponential map is a smooth map, and the tangent linear map at the origin
g = T0 g → Te G = g is the identity.
Proof: D e (x,D) = (Dx , 0) ∈ Tx G × TD g is a vector field on G × g, and the integral curve through
(x, D) is t →
7 (x · exp(tD), D); hence the flow is
τt (x, D) = (x · exp(tD), D),
and we see that the map τ1 (e, D) = (exp(D), D) is smooth.

Now, in g, the tangent vector at t = 0 to the curve tD is D; hence the tangent linear map
tangent sends it to the tangent vector at t = 0 to the curve exp(tD), which is De .
Theorem: If ψ : G → G0 is a morphism od Lie groups, the following square commutes,

exp
g /G
ψ∗ ψ
exp

g0 / G0
Proof: The tangent vector at t = 0 to the morphism ψ(gt ) : R → G0 is ψ∗ (De ); hence it corre-
sponds to the invariant vector field ψ∗ D, and exp(ψ∗ D) = ψ(g1 ) = ψ(exp(D)).
Corollary: Let G be a connected Lie group. If two morphisms of Lie groups ψ, ϕ : G → G0

coincide on the Lie algebra, ψ∗ = ϕ∗ : g → g0 , then ψ = ϕ.
Proof: Since the exponential map is a local diffeomorphism at the origin, by the above theorem
we have ψ = ϕ on a neighborhood of the identity. Now we conclude by
n.
S
Lemma: If G is connected, and U is a neighborhood of the identity, then G = nU Hence,
any open subgroup coincides with G.
−1 −1
S U nby U ∩ U we may assume that U = U .
Proof: Replacing
Now H = n U is a subgroup, and it is open since hU ⊆ H when h ∈ H.
Hence all the classes gH are open, and therefore closed; H = G.
Example: The Lie algebra of the linear group Gl(n, R) is Mn (R), each matrix A corresponds
to the morphism of groups t 7→ eAt , the Lie bracket is [A, B] = AB − BA, and the exponential
map is exp(A) = eA .
Lemma: If G is abelian and connected, exp : g → G is a surjective morphism of groups.
Proof: Let us consider the product µ : G × G → G, µ(x, y) = xy. The tangent linear map
µ∗ : Te G × Te G → Te G is µ∗ (De , De0 ) = De + De0 , since it is the identity on each factor.
10.5. LIE GROUPS 299
Take D, D0 ∈ g, and let gt , gt0 : R → G be the corresponding morphisms. Since G is abelian,

gt gt0 = µ(gt , gt0 ) : R → G is a morphism, corresponding to µ∗ (De , De0 ) = De + De0 ; hence
exp(D + D0 ) = g1 h1 = exp(D) exp(D0 ).
Now, since it is a morphism of groups, and it is a local diffeomorphism at the origin, so it is

at any point, and the image is an open subgroup of G.
Lemma: Any discrete subgroup H of a real vector space E of finite dimension is generated by
a family of linearly independent vectors,
H = Ze1 ⊕ . . . ⊕ Zer .
Proof: Any discrete subgroup is closed, because any Cauchy sequence stabilizes.
Replacing E by the vector subspace that H spans, we may assume that H contains a base
v1 , . . . , vr of E. Let us consider the open projection π : E → E/(Zv1 + . . . + Zvr ) ' S1r .
The subgroup π(H) ⊂ S1r is closed, because so is π −1 (π(H)) = H +Zv1 +. . .+Zvr = H, and it
is discrete: if U is a neighborhood of the origin intersecting H at no other point, π(U )∩π(H) = 0.
Since S1r is compact, π(H) is finite, and H is a finitely generated group of rank r.
Now, H is torsion free, since so is E, then (p. 171) H is a free group, H = Ze1 ⊕ . . . ⊕ Zer ,
and e1 , . . . , er are R-linearly independent since they span E.
Theorem: Any abelian connected Lie group G is isomorphic to (R/Z)r × Rn .

Proof: The kernel H of the exponential exp : g → G is a discrete subgroup because the ex-
ponential is a local diffeomorphism; hence the exponential factors through an isomorphism of
groups φ : E/Ze1 + . . . + Zer → G, and it is smooth since so is φ ◦ π = exp and π is a local
diffeomorphism. Moreover, E/Ze1 + . . . + Zer ' (R/Z)r × Rn .
Corollary: Any abelian compact connected Lie group is a torus (R/Z)r .
Corollary: Any abelian Lie group is G ' (R/Z)r × Rn × D, where D is discrete.
Proof: Let us consider the connected component Ge of the identity, and the exact sequence
0 −→ Ge −→ G −→ G/Ge −→ 0
Since Ge ' (R/Z)r × Rn is divisible, it is an injective Z-module (p. 123) and the sequence
admits a section s : G/Ge → G (smooth since G/Ge is discrete) defining an isomorphism
G ' Ge × (G/Ge ), g 7→ (g − s(ḡ), ḡ).

Chapter 11
Algebraic Geometry I
11.1 Sheaves and Presheaves

Definition: A presheaf P of sets on a topological space X is given by a set P(U ) for any open
set U ⊆ X, and a map ρU
V : P(U ) → P(V ) whenever V ⊆ U , so that
1. ρU
U = Id.
2. ρU V U
W = ρW ◦ ρV , when W ⊆ V ⊆ U .
(a contravariant functor from the category of open sets in X and inclusion morphisms, to the
category of sets). The elements s ∈ P(U ) are named sections of P on U , and we say that
s|V := ρU
V (s) is the restriction of s to V .
Given two presheaves P, P 0 on X, a morphism of presheaves f : P → P 0 is a morphism of
functors; i.e., a family of maps fU : P(U ) → P 0 (U ) compatible with the restriction morphisms,
in the sense that the following squares commute
fU
P(U ) / P 0 (U )
ρU
V ρU
V
fV

P(V ) / P 0 (V )
S
Definition: A presheaf F is a sheaf when, for any open cover U = i∈I Ui of an open set
U ⊆ X, we have that
1. If two sections s, s0 ∈ F(U ) coincide on the cover, s|Ui = s0 |Ui , then s = s0 .
2. Given sections si ∈ F(Ui ) coinciding on the intersections, si |Ui ∩Uj = sj |Ui ∩Uj , there exists a
section s ∈ F(U ) such that si = s|Ui for any index i ∈ I.
That is to say, for any open cover U = i∈I Ui we have an exact sequence of sets1
S
ρ Q ρ1 ,ρ2 Q
(*) F(U ) −−→ F(Ui ) ⇒ F(Ui ∩ Uj )
i∈I i,j∈I

where ρ(s) = (s|Ui )i∈I , and ρ1 (si )i∈I = si |Ui ∩Uj )i,j∈I , ρ2 (si )i∈I = sj |Ui ∩Uj )i,j∈I .
Morphisms of sheaves are just morphisms of presheaves.
A sheaf F 0 is a subsheaf of F if F 0 (U ) is a subset of F(U ) for any open set U , and the
inclusion F 0 ,→ F is a morphism of sheaves.
1 Q
In particular F(∅) is a one point set, because an open cover of ∅ is the empty family of open sets, and i∈I Xi
is a one point set when I = ∅.
301
302 CHAPTER 11. ALGEBRAIC GEOMETRY I
The restriction of F to an open set U ⊆ X is the sheaf (F|U )(V ) = F(V ), V ⊆ U .

Analogously we may define presheaves of groups, rings (or with values in a category C), and
the corresponding morphisms;
S and sheaves by requiring that the above sequence (*) is exact in
C for any open cover U = i∈I Ui .
Definition: Two sections s0 , s of a presheaf of sets P, defined on neighborhoods of a point x,

have equal germ at x if they coincide on a neighborhood of x. The germ of s is denoted sx , and
the stalk or fibre of P at x is the set Px = lim
−→
P(U ) of all germs. Put (pp. 189, 265, 293)
x∈U
P et
`
π: = Px −→ X, π(sx ) = x.
x∈X
When x ∈ U , we have a map P(U ) → Px , so that s ∈ P(U ) defines a section s̃ of π,
s̃ : U −→ P et , s̃(x) = sx .
If two sections s̃0 , s̃ coincide at a point, sx = s0x , by definition they coincide on a neighborhood
of x, so that the images of these sections s̃ form a base of a topology on P et , such that π is a
local homeomorphism, and these sections s̃ are continuous.
Moreover, any continuous section of π locally coincides with some of these sections s̃, so that
Px is also the set of germs at x of continuous sections of π.
This map π : P et → X is the étalé space of the presheaf P, and the sheaf P ] of continuous
sections of π is the sheafification of P or the associated sheaf to P. The maps P(U ) → P ] (U ),
s 7→ s̃, define a canonical morphism P → P ] inducing bijections on stalks, Px = Px] .
Any morphism of presheaves f : P1 → P2 induces maps fx : P1,x → P2,x and a continuous
map f et : P1et → P2et , hence a morphism of sheaves f ] : P1] → P2] .
Universal Property: If F is a sheaf of sets, then the morphism F → F ] is an isomorphism,

and for any presheaf of sets P,
Hom(P, F) = Hom(P ] , F).
Proof: The maps F(U ) → F ] (U ) are injective: if s0 , s ∈ F(U ) have equal germ at any point of
U , then they coincide on a cover of U ; hence s0 = s.
They are surjective: any continuous section σ : U → F et coincides on an open cover U = Ui
S
with some sections s̃i , si ∈ F(Ui ). Since F(Ui ∩Uj ) → F ] (Ui ∩Uj ) is injective, si = sj on Ui ∩Uj ,
and there is s ∈ F(U ) such that si = s|Ui , hence σ = s̃.
Now any morphism f : P → F factors through the canonical morphism P → P ] ,
f
P /F
o

f]
P] / F]
uniquely since sections of P ] locally coincide with sections s̃, where s ∈ P(U ).
Corollary: Hom(F, G) = HomX (F et , G et ), when F and G are sheaves of sets.
Corollary: A morphism f : F1 → F2 of sheaves of sets is an isomorphism if and only if the

map fx : F1,x → F2,x is bijective at any point x ∈ X.
Proof: If the maps fx are bijective, then the continuous map f et : F1et → F2et is bijective; hence
an homeomorphism since the projections Fiet → X are local homeomorphisms. q.e.d.
11.1. SHEAVES AND PRESHEAVES 303
1. If F is a set, the constant sheaf F is the sheafification of the constant presheaf U F.

The fibres are Fx = F , and the étalé space is the trivial covering F × X → X, where F is
discrete: the constant sheaf F is the sheaf of continuous maps U → F .
A sheaf of sets F is locally constant if it is constant on an open neighborhood of any point
x ∈ X; i.e., when the étalé space F et → X is a covering.
2. If ω is a closed 1-form on a smooth manifold X, then F(U ) = {f ∈ C ∞ (U ) : ω = df } is a

sheaf on X. It is locally constant, of fibre R, by Poincaré’s lemma, and the covering F et → X
was used in p. 265 to prove that ω is exact when X is simply connected.
When P is a presheaf of groups (rings, A-modules, ...) the fibres Px inherit a group structure,
so that P ] is a sheaf of groups, with the obvious universal property in the category of sheaves
of groups (resp. ...). Unless otherwise stated, from now on in these notes all presheaves and
sheaves are assumed to be of abelian groups, and we put Γ(U, F) := F(U ).
Definition: A morphism of sheaves F → G is injective (resp. surjective) if so are the

morphisms Fx → Gx , and in general we say that a sequence F 0 → F → F 00 is exact if so is the
sequence of stalks Fx0 → Fx → Fx00 at any point x ∈ X.
It is easy to check that if a sequence 0 → F 0 → F → F 00 is exact, so is the sequence
0 −→ F 0 (U ) −→ F(U ) −→ F 00 (U )
for any open set U ; but F(U ) → F 00 (U ) may be not surjective even if so is the morphism
of sheaves F → F 00 . For example, if O is the sheaf of smooth functions on S1 and Ω is the
sheaf of 1-forms, the differential d : O → Ω is surjective (any 1-form is locally exact); but
d : O(S1 ) → Ω(S1 ) is not surjective: exact 1-forms have null integral,
Z Z Z
df = f = f = 0.
S1 ∂S1 ∅
Definitions: If OX is a sheaf of rings on a topological space X, an OX -module is a sheaf

of abelian groups M such that M(U ) has a structure of OX (U )-module compatible with the
restriction morphisms, in the sense that, for any open set V ⊂ U ,
(f m)|V = f |V m|V ; f ∈ OX (U ), m ∈ M(U ),

r (we may cover X with open
and it is locally free of rank r if it is locally isomorphic to OX
r
sets U where M|U ' OX |U ) and it is a line sheaf when r = 1.
∞ is the sheaf of smooth functions on a smooth manifold X of dimension
For example, if CX
∞ -module of rank n, and the sheaf
n, then the sheaf of smooth vector fields is a locally free CX
n
ΩX of smooth n-forms is a line sheaf.
Operations with Sheaves: If F 0 is a subsheaf of a sheaf F on X, {Fi } is a family of sheaves

on X, and f : F → G is a morphism of sheaves,
1. F/F 0 is the sheafification of U F(U )/F 0 (U ), and (F/F 0 )x = Fx /Fx0 .
2. Ker f is the subsheaf (Ker f )(U ) = Ker [F(U ) → G(U )], and (Ker f )x = Ker fx .
3. Im f is the sheafification of U Im [F(U ) → G(U )], and (Im f )x = Im fx .

L L L L
4. i Fi is the sheafification of U i Fi (U ), and ( i Fi )x = i (Fi )x .
Q Q Q Q
5. i Fi is the sheaf ( i Fi )(U ) = i Fi (U ), and the stalk may be not i (Fi )x .
6. lim
−→
Fi is the sheafification of U lim
−→
Fi (U ), and (lim
−→
Fi )x = lim
−→
(Fi )x .
7. lim
←−
Fi is the sheaf (lim
←−
Fi )(U ) = lim
←−
Fi (U ), and the stalk may be not lim
←−
(Fi )x .
8. If M, N are O-modules, M ⊗O N is the sheafification of U M(U ) ⊗O(U ) N (U ), and

(M ⊗O N )x = Mx ⊗Ox Nx .
9. The sheaf of morphisms Hom(F, G) is the sheaf Hom(F, G)(U ) = Hom(F|U , G|U ).
10. The direct image φ∗ F by a continuous map φ : X → Y is the following sheaf on Y
(φ∗ F)(V ) = F(φ−1 V ), V ⊆ Y.
11.2 Sheaf Cohomology

Definitions: A differential A-module is a module K with a null square endomorphism d,
d : K −→ K, d2 = 0, Im d ⊆ Ker d,
and the cohomology of (K, d) is the quotient module H(K) = (Ker d)/(Im d).
A morphism of differential modules f : K → L is a morphism of modules commuting
with the differentials, f d = df , so that f induces a morphism in cohomology,
f : H(K) −→ H(L), f [c] = [f (c)].
i p
Theorem: If 0 → K 0 → − K→ − K 00 → 0 is an exact sequence of differential modules, we have a
connecting morphism δ and an exact triangle (the image of each morphism coincides with the
kernel of the following one)
H(K 0 )
i / H(K)
b
δ } p
H(K 00 )
Proof: Once we define the connecting δ, the exactness is easy to check.
[m00 ] ∈ H(K 00 ), where dm00 = 0,

m00 = p(m), hence p(dm) = d(p(m)) = dm00 = 0,
dm ∈ K 0 , and d(dm) = 0,
δ([m00 ]) := [dm] ∈ H(K 0 ).
It does not depend on the representant m:

If p(m) = p(n), then n − m ∈ K 0 , dn − dm ∈ dK 0 , and [dn] = [dm] in H(K 0 ).
Definitions: If the module is graded, K = ⊕n∈Z K n , and d(K n ) ⊆ K n+1 , it may be identified
with a complex K • of modules; i.e. a sequence of A-modules {K n }n∈Z with A-linear
L morphisms
n n
d :K →K n+1 n
such that d ◦ d n−1 = 0, so that the cohomology is H(K) = n∈Z H n (K • ),
n • n
where H (K ) := Ker d /Im d n−1 . A morphism of complexes f : K • → L• is a sequence of
11.2. SHEAF COHOMOLOGY 305
A-linear morphisms f n : K n → Ln commuting with the differentials (a homogeneous differential

morphisms f : ⊕n K n → ⊕n Ln ),
...
d / K n−1 d / Kn d / K n+1 d / ...
f n−1 fn f n+1

...
d / Ln−1 d / Ln d / Ln+1 d / ...
so that f induces morphisms f : H n (K • ) → H n (L• ). We say that f is a quasi-isomorphism

when f : H n (K • ) → H n (L• ) is an isomorphism for all n ∈ Z, and we put f : K • −∼
→ L• .
•
Hence, a resolution 0 → M → R (an infinite exact sequence 0 → M → R → R1 → . . .)
0
defines a quasi-isomorphism M − →∼
R• when M is viewed as a complex with null differential:
0 n
K = M and K = 0 when n 6= 0.
A sequence of morphisms of complexes K • → L• → M • is exact when K n → Ln → M n is
an exact sequence of A-modules for any n ∈ Z; hence, by the above theorem, any exact sequence
0 → K • → L• → M • → 0 induces a cohomology exact sequence
δ δ
. . . −→ H n−1 (M • ) −−→ H n (K • ) −→ H n (L• ) −→ H n (M • ) −−→ H n+1 (K • ) −→ . . .
When the complex has components of negative degree, sometimes it is convenient to put
Kn = K −n , dn = d−n : Kn → Kn−1 , and Hn (K• ) = H −n (K • ).
Snake’s Lemma: Any morphism f : M → N may be viewed as a complex K • , where K 0 = M ,

K 1 = N and d0 = f , so that H 0 (K • ) = Ker f and H 1 (K • ) = Coker f .
Hence any commutative diagram with exact rows
0 / M0 /M / M 00 /0
f0 f f 00

0 / N0 /N / N 00 /0
induces a cohomology exact sequence

δ
0 −→ Ker f 0 −→ Ker f −→ Ker f 00 −−→ Coker f 0 −→ Coker f −→ Coker f 00 −→ 0
Definition: A sheaf of abelian groups F on a topological space X is flasque when the restriction
morphism F(X) → F(U ) is surjective for any open set U in X.
i p
Lemma: If 0 → F 0 −−→ F −−→ F 00 → 0 is an exact sequence of sheaves and F 0 is flasque, then
for any open set U , the following sequence is also exact
i p
0 −→ F 0 (U ) −→ F(U ) −−→ F 00 (U ) −→ 0.
If moreover F is flasque, then so is F 00 .
Proof: It is enough to show that p is surjective. If s00 ∈ F 00 (U ), we consider pairs (V, s), where
s ∈ F(V ), p(s) = s00 |V , and by Zorn’s lemma, there is a maximal pair (V, s).
If V 6= U , and x ∈ U − V , since Fx → Fx00 is surjective, there is a neighborhood W of x, and
w ∈ F(W ), such that p(w) = s00 . Since F 0 is flasque, and p(s − w) = 0 on V ∩ W , there is a
section s0 ∈ F 0 (W ) coinciding with s − w on V ∩ W .
Now w + s0 ∈ F(W ) and s ∈ F(V ) coincide on V ∩ W , and define a section of F on U ∪ W
projecting onto s00 , against the maximal character of (V, s).
Finally, if F is flasque, we have a commutative square F(X) / F 00 (X)
epi
epi

F(U ) / F 00 (U ) q.e.d.
Definitions: If F is a sheaf, the Godement sheaf C 0 F is the flasque sheaf
(C 0 F)(U ) =
Q
x∈U Fx .
F is a subsheaf of C 0 F, and we put F1 := (C 0 F)/F, so that we have an exact sequence
0 −→ F −→ C 0 F −→ F1 −→ 0
and we repeat the process, 0 → F1 → C 0 F1 → F2 → 0, and so on. We put C n F := C 0 Fn , so

that we have an exact sequence of sheaves
0 −→ F −→ C 0 F −→ C 1 F −→ C 2 F −→ . . . −→ C n F −→ . . .
The complex of sheaves C • F = {C 0 F → C 1 F → C 2 F → . . .} is the Godement canonical

resolution of F, and taking global sections we obtain a complex of abelian groups
d d d
Γ(X, C • F) = {Γ(X, C 0 F) −−0→ Γ(X, C 1 F) −−1→ Γ(X, C 2 F) −−2→ . . .}
The n-th cohomology group of X with coefficients in the sheaf F is
H n (X, F) = H n Γ(X, C • F) = Ker dn /Im dn−1

and we say that F is an acyclic sheaf if H n (X, F) = 0, n ≥ 1.

Any morphism of sheaves f : F → G induces morphisms fx : Fx → Gx , hence a morphism
f0 : C 0 F → C 0 G, inducing a morphism F1 = (C 0 F)/F → (C 0 G)/G = G1 , and so on,
0 /F / C 0F / C 1F / C 2F / ...
f f0 f1 f2

0 /G / C 0G / C 1G / C 2G / ...
Taking global sections we have a morphism of complexes Γ(X, C • F) → Γ(X, C • G) inducing

morphisms on the cohomology groups f : H n (X, F) −→ H n (X, G).
Theorem: H 0 (X, F) = Γ(X, F).
Proof: Since 0 → F → C 0 F → C 1 F is exact, so is 0 → F(X) → (C 0 F)(X) → (C 1 F)(X).
Theorem: Any flasque sheaf is acyclic.
Proof: 0 −→ F −→ C 0 F −→ F1 −→ 0 is exact, and F is flasque; hence,

0 −→ F(X) −→ C 0 F(X) −→ F1 (X) −→ 0 is exact, and F1 is flasque,
0 −→ F1 (X) −→ C 1 F(X) −→ F2 (X) −→ 0 is exact, and F2 is flasque,...
0 −→ F(X) −→ C 0 F(X) −→ C 1 F(X) −→ C 2 F(X) −→ . . . is exact.
11.2. SHEAF COHOMOLOGY 307
i p
Theorem: Any exact sequence of sheaves 0 → F 0 → − F 00 → 0 induces a long cohomology
− F →
exact sequence,
i p δ i p δ
0 → H 0 (X, F 0 ) → − H 0 (X, F 00 ) →
− H 0 (X, F) → − H 1 (X, F 0 ) → − H 1 (X, F 00 ) →
− H 1 (X, F) → − ...
Proof: The sequences 0 → Fx0 → Fx → Fx00 → 0 are exact; hence so are the sequences
0 −→ Γ(U, C 0 F 0 ) −→ Γ(U, C 0 F) −→ Γ(U, C 0 F 00 ) −→ 0,
so that 0 → C 0 F 0 → C 0 F → C 0 F 00 → 0 is exact, and so is 0 → F10 → F1 → F100 → 0 by the

snake’s lemma. We obtain an exact sequence 0 → C • F 0 → C • F → C • F 00 → 0 of complexes of
flasque sheaves; hence an exact sequence of complexes of abelian groups
0 −→ Γ(X, C • F 0 ) −→ Γ(X, C • F) −→ Γ(X, C • F 00 ) −→ 0
inducing the required exact sequence.
De Rham’s Theorem: If 0 → F → R0 → R1 → R2 → . . . is an acyclic resolution of F

(the sequence is exact and the sheaves Rn are acyclic), then we have isomorphisms
H n (X, F) ' H n Γ(X, R• ) .

Proof: We have exact sequences 0 → F → R0 → C1 → 0 and 0 → C1 → R1 → C2 → 0.

The corresponding cohomology long exact sequences show that
d δ
0 −→ F(X) −→ R0 (X) −−0→ C1 (X) −−→ H 1 (X, F) −→ 0 is exact,
∼
δ : H n−1 (X, C1 ) −−→ H n (X, F) is an isomorphism,
0 −→ C1 (X) −→ R1 (X) −→ C2 (X) is exact,
H 0 R• (X) = Ker d0 = F(X),

H 1 R• (X) = C1 (X)/Im d0 = H 1 (X, F),

Finally, H n−1 (X, C1 ) ' H n R• (X) by induction on n.

q.e.d.
1. If a sheaf F is supported at a finite number of closed points x1 , . . . , xn (that is to say, Fx = 0

when x 6= xi ), then F(U ) = ⊕xi ∈U Fxi , and the sheaf F is flasque.
2. Let Z be the constant sheaf on the finite space X = {x1 , x2 , y1 , y2 }, xi < yj , representing a
circle (p. 227). The stalks of 0 → Z → C 0 Z → F1 → 0 are
Z• •Z Z• •Z 0• •0
0 −→ −→ −→ −→ 0
Z• •Z Z3 • • Z3 Z2 • • Z2
and F1 is flasque since it is supported in two closed points. Hence we have H i (X, Z) = 0,
i ≥ 2, and H 0 (X, Z) = H 1 (X, Z) = Z since these groups are the kernel and cokernel of the
morphism
d
Z4 = (C 0 Z)(X) −−→ F1 (X) = Z4 , d(x1 , x2 , y1 , y2 ) = (y1 − x1 , y2 − x1 , y1 − x2 , y2 − x2 ).
Theorem: Let A be a noetherian2 lattice semiring and let F be a sheaf on Spec A. If Fx = 0 at

any point x ∈ Spec A of dimension > d, then H p (Spec A, F) = 0 for any p > d. In particular,
H p (Spec A, F) = 0, p > dim A.
Proof: We proceed by induction on d, and it is obvious when d = −1.

Let {xi } be the family of all points of X = Spec A of dimension d.
The presheaf Fd (U ) = ⊕xi ∈U Fxi is a (flasque) sheaf since any open set is compact, and we
have a natural morphism φ : F → Fd since the support of a section of F, being closed, has a
finite number of irreducible components, hence a finite number of points of dimension d.
Moreover Fd has null stalk at any point of dimension > d, and (Fd )xi = Fxi , so that Ker φ
and Coker φ have null stalk at any point of dimension > d − 1, and
H p (X, Ker φ) = H p (X, Coker φ) = 0, p ≥ d

0 −→ Ker φ −→ F −→ Im φ −→ 0 is exact (11.1)
p p
H (X, F) = H (X, Im φ), p ≥ d,
0 −→ Im φ −→ Fd −→ Coker φ −→ 0 is exact (11.2)
p
H (X, Im φ) = 0, p > d.
Theorem: If A is a ring, any A-module M defines a sheaf M f on Spec A, attached to the

localization presheaf U MU , and Γ(Spec A, M ) = M , (p. 189). The sheaves M
f f are acyclic,
H n (Spec A, M
f ) = 0, n ≥ 1.
Proof: Proof: If j : Uf → X = Spec A is a basic open set, we put Ff = j∗ j ∗ F, so that we have

Ff (U ) = F(Uf ∩ U ), and Ff is flasque when so is the sheaf F.
Moreover, (Mf)f is the sheaf associated to the A-module Mf .
We proceed by induction on n and, given a cohomology class c ∈ H n (X, M f), first we see
n
that any point x has a basic neighborhood Uf such that c = 0 in H (X, Mf ).f
We truncate a flasque resolution 0 → M f → C • at the n-th step, so that we have a commu-
tative diagram with exact rows, where K is the image of Cfn−1 → Kf ,
0
0 /M
f / C0 / ... / C n−1 /K /0

/M / C0 / ... / C n−1 / K0 /0 K0 ⊆ Kf
0 ff
f f
In fact the bottom row is exact because, when 0 < p < n, the cohomology groups of the complex
of sections over any basic open set Ug are H p (Uf ∩ Ug , M
f) = H p (Spec Af g , M
f) = 0.
n
Hence c ∈ H (X, M f) = K(X)/C n−1 (X) is represented by a section s ∈ K(X), and on some
neighborhood Uf of x it comes from a section in C n−1 (Uf ) = Cfn−1 (X).
We conclude that c = 0 in H n (X, Mff ) = K0 (X)/C n−1 (X), as claimed.
f
Now, X being compact, X = Uf1 ∪ . . . ∪ Ufr , where c = 0 in H n (X, M ff ), and the long
i
cohomology exact sequence exact given by (the sheafification of) the exact sequence
0 −→ M −→ Mf1 ⊕ . . . ⊕ Mfr −→ N −→ 0
2
See p. 331 for the case of an arbitrary lattice semiring.
11.3. SCHEMES AND COHERENT SHEAVES 309
shows that the morphism H n (X, Mf) → ⊕i H n (X, Mff ) is injective (and we see that c = 0).
i
When n = 1, because the morphism ⊕i Mfi = H 0 (X, ⊕i M ff ) → H 0 (X, N

i
e ) = N is surjective,
and when n > 1, because H n−1 (X, N
e ) = 0 by induction.
Note: In the case of curves, the above results are much easier to prove.
Let X be a topological space with a dense point x whose open neighborhoods are just sets
with finite complement. Constant sheaves on X are flasque, as well as sheaves F with null fibre
at x because the support of any local section s ∈ F(U ), being a closed set, is a finite set of
closed points, and s may be extended by 0 out of U .
In general, the exact sequences 11.1 and 11.2 of the natural morphism φ : F → Fx let us
conclude that H p (X, F) = 0, p > 1, since Ker φ and Coker φ have null fibre at x.
Moreover, if X = Spec A, where A is a noetherian domain of dimension 1, let us prove that
H p (X, Mf) = 0, p ≥ 1. The torsion submodule T defines a flasque sheaf, because it has null stalk
at the generic point x, and the long cohomology exact sequence given by (the sheafification of)
the exact sequence 0 → T → M → M/T → 0 shows that we may assume that M is torsion free.
Now, the long cohomology exact sequence given by (the sheafification of) the exact sequence
0 −→ M gx −→ (Mx /M )∼ −→ 0
f −→ M
gx is flasque, the sheaf (Mx /M )∼ has null fibre at x, and the
let us conclude because the sheaf M
morphism Mx = H 0 (X, M gx ) → H 0 (X, (Mx /M )∼ ) = Mx /M is surjective.
Note: Let 0 → F → R• be a flasque resolution, and let f : X → Y be a continuous map.

The sheaves f∗ Rn are flasque; but the sequence 0 → f∗ F → f∗ R• may be not exact.
If it is exact, then the direct image f∗ preserves the cohomology groups,
H n (Y, f∗ F) = H n Γ(Y, f∗ R• ) = H n Γ(X, R• ) = H n (X, F).

Theorem: H n (X, i∗ F) = H n (Y, F), when i : Y → X is a closed embedding.
Proof: The functor i∗ is exact since (i∗ F)x = Fx when x ∈ Y , and 0 otherwise.
Theorem: If Y admits a base of open neighborhoods V such that F is acyclic on f −1 V , then
H n (Y, f∗ F) = H n (X, F).
Proof: By hypothesis the following sequences are exact
0 −→ F(f −1 V ) −→ R• (f −1 V )
|| ||
0 −→ (f∗ F)(V ) −→ (f∗ R• )(V )
and taking the inductive limit on the neighborhoods of y, we see that the sequence of stalks
0 → (f∗ F)y → (f∗ R• )y is exact; i.e., f∗ R• is a flasque resolution of f∗ F and we conclude.
11.3 Schemes and Coherent Sheaves

Definitions: A ringed space (X, OX ) is a topological space X with a sheaf of rings OX , and
it is a locally ringed space if the fibres OX,x are local rings.
The elements of the ring OX (U ) are said to be the functions on the open set U .
A morphism of ringed spaces (Y, OY ) → (X, OX ) is a continuous map φ : Y → X with

a morphism of sheaves of rings ψ : OX → φ∗ OY (morphisms of rings OX (U ) → OY (φ−1 U )
compatible with the restriction morphisms), and it is a morphism of locally ringed spaces if
moreover the morphisms OX,φ(y) → OY,y are local, my ∩ OX,φ(y) = mφ(y) (i.e. a function
f ∈ OX (U ) vanishes at a point x = φ(y) ∈ U if and only if ψ(f ) ∈ OY (φ−1 U ) vanishes at
y ∈ φ−1 U ).
If A is a ring, then (Spec A, Ã) is a locally ringed space, and any ring morphism φ : A → B
defines a morphism of locally ringed spaces
(f, φ) : (Spec B, B̃) −→ (Spec A, Ã),
where f : Spec B → Spec A is the continuous map induced by φ, and φ : Ã → f∗ B̃ is the

morphism of sheaves induced by the morphism of presheaves
φ
AU −−→ Bf −1 (U ) −→ B̃(f −1 U ) −→ (f∗ B̃)(U ).
Theorem: Homrings (A, B) = Homloc. ring. sp. (Spec B, Spec A).
Proof: Any morphism of locally ringed spaces (f, φ) : (Spec B, B̃) → (Spec A, Ã) induces a ring
morphism φ : A = Γ(Spec A, Ã) → Γ(Spec B, B̃) = B and, Af (y) → By being local, the following
commutative square shows that f is the map induced by φ,
φ
A /B
φ
Af (y) / By
Definition: A ringed space (X, OX ) is an affine scheme if it is isomorphic to (Spec A, Ã) for
some ring, necessarily A = OX (X), and it is a scheme if any point has an open neighborhood
U (hence a base of open neighborhoods) such that (U, OX |U ) is an affine scheme.
Morphisms of schemes are morphisms of locally ringed spaces.
If you fix a base scheme S, then a S-scheme is a scheme X endowed with a morphism of
schemes πX : X → S and a morphism of S-schemes is a morphism of schemes f : X → Y such
that πX = πY ◦ f .
1. The above theorem shows that the functors A (Spec A, Ã) and X OX (X) define an
antiequivalence of the category o rings with the category of affine schemes; hence an antiequiv-
alence of the category of k-algebras with the category of affine k-schemes.
2. Smooth manifolds and Riemann surfaces are locally ringed spaces, and the argument given
in the proof of the above theorem shows that smooth maps and analytic morphisms are just
the morphisms of locally ringed spaces.
3. If A is a domain, then the local

T rings Ax are subrings of the field of fractions Σ, and the
morphism of presheaves AU → x∈U Ax is an isomorphism on stalks; hence
T
Ã(U ) = Ax .
x∈U
Proposition: Any irreducible closed set C in a scheme is the closure of a unique point, named
generic point point of C.
Proof: If we consider an affine open set U intersecting C, then U ∩ C is irreducible, so that it

has a dense point x (p. 135). Now C = x, because C is irreducible.
Definition: A scheme X is noetherian if it is a finite union of affine open sets Ui = Spec Ai ,

where the rings Ai are noetherian.
In such a case any affine open set is U = Spec A, with A noetherian. In fact, we may assume
that X = Spec A, and a chain of ideals of A stabilizes just when it stabilizes in any ring Ai .
Moreover, any noetherian scheme X is a noetherian space where closed irreducible sets have
a dense point; hence X = Spec AX (p. 225) and H p (X, F) = 0 when p > dim X (p. 308).
Definitions: If X is a scheme, an OX -module M is quasi-coherent if we may cover X by
affine open sets U = Spec A where M|U = M̃ for some A-module M (hence on any affine open
set V ⊂ U ). When the scheme X is noetherian, M is coherent if moreover M is a finitely
generated A-module.
Locally free sheaves are quasi-coherent, and isomorphism classes of line sheaves form an
abelian group with the tensor product ⊗OX , the Picard group Pic(X).
The unity is OX , and the inverse of a line sheaf L is L−1 = HomOX (L, OX ).
Theorem: Any quasi-coherent sheaf M on Spec A is M = M

f for some A-module M .
Proof: Put M = M(Spec A).

Then the natural morphisms MU → M(U ) induce an isomorphism M f → M.
In fact, we may cover Spec A by basic open sets Ui where M|Ui = M
fi , and the following
commutative diagram with exact rows shows that Mf = M(Uf ),
L L
0 −→ Mf −→ i M(Ui )f ⇒ i,j M(Ui ∩ Uj )f
↓ L || L ||
0 −→ M(Uf ) −→ i M(Ui ∩ Uf ) ⇒ i,j M(Ui ∩ Uj ∩ Uf )
Corollary: Let X be an affine scheme. The functor M M(X) defines an equivalence of the
category of quasi-coherent OX -modules with the category of OX (X)-modules.
Definitions: If U is an open set in X, we say that the scheme (U, OX |U ) is a open subscheme
of X. If I is a quasi-coherent sheaf of ideals, the support Y of the sheaf of rings OX /I is closed
in X, and we say that the scheme (Y, OX /I) is the closed subscheme of X defined by I.
A morphism Y → X is an open or closed embedding if it defines an isomorphism of Y onto
an open or closed subscheme of X.
Definitions: A scheme over a field k is a morphism X → Spec k (a structure of k-algebra on

OX (U ), so that restriction morphisms are morphisms of k-algebras), and it is a scheme of finite
type over k if it is a finite union of affine open sets Ui = Spec k[ξ1 , . . . , ξn ].
If moreover it is of dimension 1, it is a curve, and it is a non singular curve if any local
ring OX,x is regular (a discrete valuation ring).
A scheme X is reduced when so is the ring OX (U ) for any open set U , and it integral
when so is the ring OX (U ) for any open set U , so that X is irreducible (with generic point pg )
and OX (U ) ⊆ Σ, where the generic fibre Σ = OX,pg is a field. A integral scheme X of finite
type over a field k is complete if any valuation ring V of Σ, containing k, centers at a unique
point x ∈ X (i.e., V dominates the local ring OX,x ⊂ Σ).
The Riemann variety

Definition: The Riemann variety of a finite extension Σ of k(t) is the ringed space
X = {discrete valuation rings of Σ containing k}

T
OX (U ) = Ox
x∈U
where Ox is the valuation ring of x ∈ X, and closed sets 6= X are finite subsets not containing
the generic point pg defined by the trivial valuation ring Σ. Any complete non singular curve C
over k is the Riemann variety of the field of rational functions ΣC = OC,pg .
A projective line over k is any k-scheme isomorphic to the Riemann variety P1 of k(t).
Theorem: The Riemann variety (X, OX ) is a non singular complete curve.
Proof: If B (resp. B 0 ) is the integral closure of k[t] (resp. k[ 1t ]) in Σ, the morphisms k[t] → B
and k[ 1t ] → B 0 are finite (p. 204) and
(
0 U = {x ∈ X : vx (t) ≥ 0} = Spec B
X =U ∪U ,
U 0 = {x ∈ X : vx (t) ≤ 0} = Spec B 0
where U, U 0 are open sets in X, since X − U = ( 1t )0 , and X − U 0 = (t)0 .

We conclude since B and B 0 are Dedekind domains (p. 204).
Definitions: The group Div(X) of divisors of X is the free abelian group on all closed points
P : k] of a closed point x ∈ X is finite by the nullstellensatz, and
of X. The degree deg x = [κ(x)
the degree of a divisor D = x nx · x is
P
deg D = x nx (deg x).
Any rational function 0 6= f ∈ Σ has a finite number of zeros and poles (in a Dedekind domain,
any non null function has a finite number of zeros) and the divisor of f is
P
D(f ) = x vx (f ) · x,
D(f h) = D(f ) + D(h).
Two divisors D, D0 are linearly equivalent, and we put D0 ∼ D, when D0 = D + D(f ) for
some f ∈ Σ, and linear equivalence classes of divisors form a group Div(X)/ ∼.
Theorem: deg D(f ) = 0.
Proof: If f is algebraic over k, then D(f ) = 0. Otherwise, since f is algebraic over k(t), then t
is algebraic over k(f ), and Σ is a finite extension of k(f ).
The integral closure B of k[f ] in Σ is a torsion free finite k[f ]-module; hence it is free,
k[f ]n ' B,
and localizing at the generic point we see that n = [Σ : k(f )].

Moreover, the ring B/f B, being of dimension 0, decomposes as a direct sum (p. 190)
k n ' B/f B = (Bx1 /f Bx1 ) ⊕ . . . ⊕ (Bxr /f Bxr )

vxi (f ) = l(Bxi /f Bxi )
P
and we see that n = vxi (f )(deg xi ), where xi runs over the zeros of f .
The number of zeros of f is n = [Σ : k(f )], and the number of poles of f (the number of
zeros of f −1 ) is [Σ : k(f −1 )]. Both coincide since k(f −1 ) = k(f ). q.e.d.
Any divisor D defines a line subsheaf LD of the (constant) sheaf of rational functions Σ,
LD (U ) := {f ∈ Σ : D + D(f ) ≥ 0 on U }.
In fact, at any point x we may fix a parameter t ∈ Ox and a neighborhood U where t has
no other zeros or poles, nor the divisor D has other points, so that if n is the coefficient of x in
D, we have that LD is free on U ,
LD |U = t−n OX |U .
If D0 = D + D(h), then φ : LD0 → LD , φ(f ) = hf , is an isomorphism, so that the sheaf LD
only depends on the class [D].
Moreover, the morphism φ : LD1 ⊗OX LD2 → LD1 +D2 , φ(f1 ⊗ f2 ) = f1 f2 , is an isomorphism
since so is on stalks,
LD1 +D2 = LD1 ⊗OX LD2 .
Theorem: Pic(X) = Div(X)/ ∼.
∼
Proof: An isomorphism LD0 − → LD defines an isomorphism Σ ' Σ on the generic stalks, hence
a homothety of ratio h.
On any open set we have 0 ≤ D0 + D(f ) if and only if 0 ≤ D + D(hf ) = D + D(h) + D(f );
hence D0 = D + D(h), and D0 ∼ D.
∼
Conversely, given a line sheaf L, if we fix an isomorphism Lpg −
→ Σ, then we may view inside
Σ the sections of L and stalks Lx , which are free Ox -modules of rank 1.
Hence Lx = mnx x for some integer nx .
We have nx = 0 at any point, up to a finite number, since L(U ) = f OX (U ) on an affine
open set U , and f has a finite
P number of zeros and poles.
Now, if we put D = − x nx · x, the natural morphism L → LD is an isomorphism.
Corollary: Pic(P1 ) = Z, and we put OP1 (n) = Lnp∞ .
Proof: We have to show that any divisor D of degree 0 is D = D(f ), f ∈ k(t).

Since both have degree 0, it is enough to see that they coincide in Spec k[t].
If D = n1 x1 + . . . + nr xr in the affine part, we take polynomials pi (t) generating the maximal
ideals mxi of k[t], and D coincides with the divisor of f (t) = p1 (t)n1 . . . pr (t)nr .
Morphisms: Any k-morphism Σ0 → Σ induces a morphism of k-schemes π : X → X 0 between

the Riemann varieties. In fact, if a discrete valuation ring Ox of Σ contains k, then Ox0 = Ox ∩Σ0
is a discrete valuation ring of Σ0 containing k, and we put x0 = π(x).
Since any fibre of π is finite, π is continuous, and the morphism of sheaves OX 0 → π∗ OX is
defined by the inclusions
Ox0 −→ OX (π −1 U ) =
T T
OX 0 (U ) = Ox .
x0 ∈U π(x)∈U
If U = Spec A is affine, then −1
π (U ) = Spec B, where the integral closure B of A in Σ is a
0
locally free A-module of rank d = [Σ : Σ ], the degree of the morphism.
Therefore, any fibre Spec B/mx0 B of π defines a divisor of degree d(deg x0 ),
π ∗ (x0 ) =
P
l(Bx /mx0 Bx ) · x
π(x)=x0
so that we have an inverse image π ∗ D0 of divisors, and Lπ∗ D0 = π ∗ (LD0 ).

So is any non constant k-morphism of schemes X → X 0 .
11.4 Riemann-Roch Theorem

(
p n+1 p=0
Theorem: dimk H (P1 , O(n)) =
0 p 6= 0
(
n−1 p=1
dimk H p (P1 , O(−n)) =
0 p 6= 1
Proof: Once we prove that the structural sheaf OP1 is acyclic, the exact sequences
0 −→ O(n) −→ O(n + 1) −→ k∞ −→ 0
where k∞ is the sheaf k supported at infinity, let us conclude.

•
Now, the projection π : P1 → •/\• , transforming the origin and the infinity onto the two closed
points, and any other point onto the dense point, preserves the cohomology of quasi-coherent
sheaves (pp. 308, 309), and π∗ OP1 admits the flasque resolution,
k[t, 1t ] k[t, 1t ] 0
0 −→ k[t] • k[ 1 ] −→ k[t, 1 ] • k[t, 1 ] −→ k[ 1 ]/k • k[t]/k −→ 0
• • t t • • t t • •
0 −→ H 0 (P1 , OP1 ) −→ k[t, 1t ] −→ k[ 1t ]/k ⊕ k[t]/k −→ H 1 (P1 , OP1 ) −→ 0

We conclude since the morphism k[t, 1t ] → (k[ 1t ]/k) ⊕ (k[t]/k) is surjective.
Theorem: The cohomology groups of any coherent sheaf M on a non singular complete curve
X are finite dimensional vector spaces.
Proof: Any morphism k(t) → Σ defines a projection π : X → P1 preserving the cohomology of

coherent sheaves, and π∗ M is coherent when so is M, because π −1 (Spec k[t]) = Spec B, where
B is a finite k[t]-module (p. 204).
Hence we may assume that X = P1 .
The torsion of M is supported at a finite number of closed points, where the stalk has finite
dimension (in particular it is a flasque sheaf); hence we may assume that it is null, so that M
is a subsheaf of the constant sheaf Mg .
Now a non null section s ∈ Mg defines a sheaf of ideals I
I(U ) = {f ∈ OP1 (U ) : f s ∈ M(U )}.
The theorem holds for I ' O(−n), and for M/Is by induction on the rank, and the following
exact sequence let us conclude:
0 −→ I −→ M −→ M/Is −→ 0
Definitions: We put hp (M) = dimk H p (X, M), and the Euler-Poincaré characteristic
χ(M) = χ(X, M) = p (−1)p hp (M) = h0 (M) − h1 (M)

P
is an additive function on the category of coherent sheaves.

The genus of a non singular complete curve X is g = h1 (OX ).
According to the above theorem, χ(P1 , O(n)) = n + 1, n ∈ Z.
The algebraic closure of k in Σ is H 0 (X, OX ). It is a finite extension k̄ of k, and from now
on, we always assume that k̄ = k (or replace k by k̄), so that χ(OX ) = 1 − g.
11.4. RIEMANN-ROCH THEOREM 315
Riemann-Roch Theorem (weak): χ(LD ) = χ(OX ) + deg D.
Proof: The cohomology exact sequence of the exact sequence
0 −→ LD −→ LD+x −→ κ(x) −→ 0
shows that χ(LD+x ) = χ(LD ) + deg x.

Hence the theorem holds for a divisor D if and only if it holds for D + x.
Since it obviously holds when D = 0, we conclude.
Corollary: A non singular complete curve X is rational, ΣX ' k(t), if only if the genus is 0
and it has some rational point.
Proof: If g = 0 and X has a rational point x, then χ(Lx ) = 1 − 0 + 1 = 2, so that h0 (Lx ) ≥ 2,

and there is a non constant rational function f ∈ Σ with a unique simple pole at x.
Hence [Σ : k(f )] = number of poles of f = 1, and Σ = k(f ).
Corollary: Given a closed point x ∈ X, there is f ∈ Σ with a unique pole at x, so that the open
set U = X − x is affine.
Proof: We have h0 (Lnx ) ≥ 1 − g + n.

When n ≥ g + 1, there exists a non constant morphism f : X → P1 with a unique pole at x;
hence U = f −1 (A1 ) is affine.
Definition: Let X be a curve over a field k. The k-linear functor F (M) = H 1 (X, M)∗ is left
exact on the category of coherent sheaves, because H 2 (X, M) = 0, and any pair is dominated
by a minimal pair. In fact, any sequence of epimorphisms
p1 p2 p3
Mξ −−→ M0ξ0 −−→ M00ξ00 −−→ . . .
stabilizes since Ker p1 ⊆ Ker (p2 p1 ) ⊆ Ker (p3 p2 p1 ) . . . and X is noetherian. By the representabil-
ity theorem F is an inductive limit of representable functors: there is an inductive system of
coherent sheaves {Mi } such that, for any coherent OX -module M, we have natural linear iso-
morphisms (where the quasi-coherent sheaf ωX/k := lim −→
Mi is the dualizing or canonical sheaf
of X)
H 1 (X, M)∗ = lim
−→
HomOX (M, Mi ) = HomOX (M, ωX/k ),
Theorem: ωP1 = OP1 (−2).

∗
Proof: We have H 1 (P1 , O(−2)) 6= 0, so that O(−2), with any non null ξ ∈ H 1 P1 , O(−2) , is a
minimal pair (quotients of a line sheaf are of torsion sheaves, hence acyclic).
Hence O(−2) is a submodule of ωP1 , and if we put M(n) = M ⊗OP1 O(n), n ≥ 1,
0 −→ O(−2) −→ ωP1 −→ K −→ 0 is exact

0 −→ O(n − 2) −→ ωP1 (n) −→ K(n) −→ 0 is exact

0 −→ Γ P1 , O(n − 2) −→ Γ P1 , ωP1 (n) −→ Γ P1 , K(n) −→ 0 is exact
Γ(P1 , ωP1 (n)) = HomOX (O(−n), ωP1 ) = H 1 (P1 , O(−n))∗
dim Γ(P1 , O(n − 2)) = n − 1 = dim H 1 (P1 , O(−n))∗
Γ(P1 , K(n)) = 0
so that K = 0, because torsion elements of stalks define global sections and elements of the
generic stalk define (p. 314) morphisms O(−n) → K; hence sections of K(n).
Theorem: The dualizing sheaf ωX of any non singular complete curve X is a line sheaf.
Proof: Let us consider a morphism k(t) → Σ and the corresponding projection π : X → P1 .

The direct image π∗ preserves the cohomology of coherent sheaves, and π∗ M is coherent
when so is M, so that
H 1 (X, M)∗ = H 1 (P1 , π∗ M)∗ = HomOP1 (π∗ M, ωP1 )

= Homπ∗ OX π∗ M, HomOP1 (π∗ OX , ωP1 ) ,
H 1 (X, M)∗ = HomOX (M, ωX ) = Homπ∗ OX (π∗ M, π∗ ωX ), (11.3)
π∗ ωX = HomOP1 (π∗ OX , ωP1 ), (11.4)
hence ωX is torsion free and it has rank 1. It is a line sheaf.
Definition: A divisor K of the curve X is canonical if ωX = LK .
Riemann-Roch Theorem: If K is a canonical divisor of a non singular complete curve X of

genus g, then for any divisor D,
dim H 0 (X, LD ) = χ(OX ) + deg D + dim H 0 (X, LK−D ).
Proof: We have H 1 (X, LD )∗ = HomOX (LD , LK ) = H 0 (X, LK−D ), and we conclude by the weak
Riemann-Roch theorem.
Corollary: h1 (LK ) = 1, h0 (LK ) = g, deg K = 2g − 2.
Proof: H 1 (X, LK )∗ = HomOX (LK , LK ) = H 0 (X, OX ) has dimension 1.

Moreover H 0 (X, LK ) = HomOX (OX , LK ) = H 1 (X, OX )∗ has dimension g, and
g = h0 (LK ) = 1 − g + deg K + h0 (OX ) = 2 − g + deg K.
11.4.1 Calculation of the Canonical Sheaf

Definition: If X is a k-scheme, the sheafification of U ΩOX (U )/k is the sheaf of differentials
ΩX/k . Since the module of differentials localizes (p. 140), on any affine open set U = Spec A it
is the sheaf defined by ΩA/k , and it is coherent when X is of finite type.
Proposition: If X is a non singular curve over a perfect field, ΩX/k is a line sheaf.
Proof: If t is a local parameter, mx = tOx , the exact sequence
mx /m2x −→ ΩOx ⊗Ox κ(x) −→ Ωκ(x)/k = 0
shows that ΩOx ⊗Ox κ(x) = hdti, and by Nakayama’s lemma ΩOx = Ox dt.
If it is not torsion free, then ΩΣ/k = 0, and Σ is (p. 144) a separable extension of k(t) such
that no k-derivation of k(t) may be extended to Σ. Absurd.
Theorem: The dualizing sheaf of a non singular complete curve X over an algebraically closed
field is the sheaf of differentials, ωX/k = ΩX/k .
First Proof: The theorem holds in P1 since ΩP1 = O(−2).

In fact, dt has no zeros nor poles in the affine part, and it has has two poles at infinity,
dt = d(u−1 ) = −u−2 du.
In general, we take a separable morphism k(t) → Σ, so that the trace metric (p. 145) is non
singular, and it defines an exact sequence
0 −→ π∗ OX −→ HomOP1 (π∗ OX , OP1 ) −→ C −→ 0
where C is supported at a finite number of closed points, since the generic stalk is null.
Hence C ⊗OP1 L = C for any line sheaf L, and we have exact sequences
0 −→ ΩP1 ⊗OP1π∗ OX −→ HomOP1 (π∗ OX , OP1 ) ⊗OP1ΩP1 −→ C −→ 0

0 −→ ΩP1 ⊗OP1π∗ OX −−−−−−−−→ π∗ ΩX −−−−−−−−→ ΩX/P1 −→ 0
where ΩX/P1 (Spec A) = ΩB/A , and B is the integral closure of A in Σ. By 11.4 we have
HomOP1 (π∗ OX , OP1 ) ⊗OP1ΩP1 = HomOP1 (π∗ OX , ΩP1 ) = π∗ ωX ,
and to conclude that ωX ' ΩX we have to show that B ∗ /B and ΩB/A have equal length at any
point of Spec B, where we view B inside B ∗ = HomA (B, A) via the trace metric.
The trace metric and the module of differentials are stable under base changes A → A0 .
Hence we may prove the theorem after localization and completion at a point y ∈ Spec A,
by −→ lim By /mny By = lim By /(mnx 1 . . . mnx r )n = B
A bx ⊕ . . . ⊕ B
bxr
←− ←− 1 r 1
where π −1 (y) = {x1 , . . . , xr }, and it is enough to prove it for the morphisms A

by → B
bx .
Since k is algebraically closed, both are rings of formal series (p. 198),
b −→ B
k[[t]] = A b = k[[x]]
tBb = (xn )
B/t
b B b = k[x]/(xn )
and Nakayama’s lemma shows that (1, x, . . . , xn−1 ) is a base of the A-module
b B,
b
B b ⊕ Ax
b=A b n−1 = A[x]/(x
b ⊕ . . . ⊕ Ax b n
+ . . .) = A[x]/(P
b ).
Lemma: Let A be a Dedekind domain and B the integral closure of A in a finite separable
extension of the field of fractions. If B = A[ξ] = A[x]/(P ) = A[x]/(xn + . . .), then
n−1
B ∗ = A P 01(ξ) ⊕ A P 0ξ(ξ) ⊕ . . . ⊕ A Pξ 0 (ξ)
and therefore B ∗ /B ' B/(P 0 (ξ)) ' ΩB/A .

1
Proof: If ξ = α1 , . . . , αn are the roots of the separable polynomial P (x), decomposing P (x) into
simple fractions, and expanding as a power series of x−1 , we see that
n
X 1 1
= = x−n (1 + a1 x−1 + . . .)
P 0 (αi )(x
− αi ) P (x)
i=1
i (
ξ 0 0≤i≤n−2
tr =
P 0 (ξ) 1 i=n−1
ξi
Hence P 0 (ξ) ∈ B ∗ , since ξ i+j ∈ A ⊕ Aξ . . . ⊕ Aξ n−1 , and the matrix
 
0 0 1
ξiξj •
 

tr 0 = 
P (ξ)  0



1 • •
ξ ξ n−1
is invertible: 1
P 0 (ξ) , P 0 (ξ) , . . . , P 0 (ξ) form a base of B ∗ , and B ∗ /B ' B/(P 0 (ξ)). q.e.d.
Second Proof: Let us show that HomO (ΩX , ωX ) is a trivial line sheaf :
Let X = U ∪ U 0 = (Spec B) ∪ (Spec B 0 ) be the open cover of p. 312.
For any k-algebra A, the base change XA is the scheme
XA = X ×k (Spec A) = UA ∪ UA0 = Spec (B ⊗k A) ∪ Spec (B 0 ⊗k A)
and the direct product X ×k X is the scheme (integral since k is algebraically closed)
X ×k X = XB ∪ XB 0 = (X ×k Spec B) ∪ (X ×k Spec B 0 ).
The diagonal morphism X → X ×k X is a closed embedding since the natural morphism

B ⊗k B 0 −→ OX (U ∩ U 0 ) = B 1t = B 0 [ t ] is surjective, and the sheaf of differentials ΩX = ∆/∆2
is a line sheaf; hence the sheaf of ideals of the diagonal ∆ is locally principal.
Now, given a closed point x ∈ X, we restrict to X ×k x the following exact sequence
0 / ωX ⊗k OX / Hom(∆, ωX ⊗k OX ) / Hom(∆/∆2 , ωX ) /0

0 / ωX / Hom(mx , ωX ) / Hom(mx /m2 , ωX /mx ωX ) /0
x
and we take the direct image by the second projection π : X ×k X → X,
HomO (ΩX , ωX )
δ / R1 π∗ (ωX ⊗k OX ) (11.5)

δx / H 1 (X, ωX ) / H 1 (X, LK+x ) = 0
Hom(mx /m2x , ωX /mx ωX )
where HomO (ΩX , ωX ) is a line sheaf, and R1 π∗ (ωX ⊗k OX ) = H 1 (X, ωX ) ⊗k OX is a trivial line
sheaf (according to the following lemma). This commutative diagram shows that δ is surjective
at any closed point x ∈ X; hence it is an isomorphism.
Lemma: Let X be a non singular complete curve over a field k. If M is a quasi-coherent

OX -module and A is a k-algebra, we have natural isomorphisms
H p (XA , M ⊗k A) = H p (X, M) ⊗k A.
Proof: Comparing the Mayer-Vietoris sequences of M and MA = M ⊗k A,
0 −→ H 0 (X, M) −→ Γ(U, M) ⊕ Γ(U 0 , M) −→ Γ(U ∩ U 0 , M) −→ H 1 (X, M) −→ 0

0 −→ H 0 (XA , MA ) −→ Γ(U, M)A ⊕ Γ(U 0 , M)A −→ Γ(U ∩ U 0 , M)A −→ H 1 (XA , MA ) −→ 0
∼
we see, the functor ⊗k A being exact, that H p (X, M) ⊗k A −−→ H p (XA , M ⊗k A).
Residues: Let X be a complete non singular curve over an algebraically closed field k, and Σ
the function field. The sheaf of differentials ΩX/k admits the flasque resolution
0 −→ ΩX/k −→ ΩΣ/k −→ ΩΣ /ΩX −→ 0, (11.6)
where ΩΣ is the constant sheaf and ΩΣ /ΩX the sheaf of principal parts of meromorphic forms:
{an tdtn + . . . + a1 dtt } ,

L L
(ΩΣ /ΩX )(U ) = (ΩΣ /ΩOx ) = tOx = mx .
x∈U x∈U
δ
We have Hx0 (X, ΩΣ ) = Hx1 (X, ΩΣ ) = 0; hence ΩΣ /ΩOx = Hx0 (X, ΩΣ /ΩX ) −→
x
Hx1 (X, ΩX ) is
an isomorphism. On the other hand,
Homk (mx /m2x , ΩX /mx ΩX ) = (m−1

x ΩOx )/ΩOx ⊂ ΩΣ /ΩOx ,
where the identity mx /m2x = ΩX /mx ΩX corresponds to dtt ∈ (m−1

x ΩOx )/ΩOx .
Now take global sections in 11.5, with ωX = ΩX , and with support in x in the exact sequence
0 → ΩX → Hom(mx , ΩX ) → (m−1 x ΩOx )/ΩOx → 0. We obtain a commutative diagram
Hom(ΩX , ΩX ) / (m−1 ΩO )/ΩO o (m−1

x x x
Id x ΩOx )/ΩOx
δ o δx o δx

H 1 (X, ΩX )
Id / H 1 (X, ΩX ) o Hx1 (X, ΩX )
and we see that δx dtt ∈ H 1 (X, ΩX ) does not depend on the point x, since it is just δ(Id).

∼
Therefore, we have a global residue Res : H 1 (X, ΩX ) −
→ k, hence a local residue
Res
Resx : ΩΣ /ΩOx = Hx1 (X, ΩX ) −→ H 1 (X, ΩX ) −−−→ k
at any closed point x, such that Resx dtt = 1 when tOx = mx . But the fully determination of

the local residue, Resx tdtn = 0, n > 1, is postponed to p. 419.

Theorem: A finite family (θx ) ∈ x∈X {an tdtn + . . . + a1 dtt } of principal

L
P parts of meromorphic
forms is defined by a global meromorphic form ω ∈ ΩΣ if and only if x∈X Resx (θx ) = 0.
δ
− H 1 (X, ΩX ) → 0.
L
Proof: By 11.6, we have an exact sequence ΩΣ → x∈X (ΩΣ /ΩOx ) →
Hurwitz’s Formula: If π : X → X 0 is a morphism between non singular complete curves over

an algebraically closed field, defined by a separable extension Σ0 → Σ of degree d, then
2g − 2 = d(2g 0 − 2) +
P
ex , ex = l ΩOx /Ox0 .
x∈X
Proof: We have deg π ∗ (K 0 )

= d(deg K 0 )
= d(2g 0 − 2), and the stalk of ΩX/X 0 at the generic point
of X is ΩΣ/Σ0 = 0, so that we have an exact sequence
0 −→ π ∗ ΩX 0 = Lπ∗ K 0 −→ ΩX = LK −→ ΩX/X 0 −→ 0
Definition: A curve X over a field k is smooth if ΩX/k is a line sheaf.
Examples: When k is algebraically closed, ΩOx /mx ΩOx = mx /m2x (p. 141), and smooth curves
are just non singular curves.
The curve y 2 = xp − t over k = Fp (t) is non singular; but it is not smooth, since it is singular
over the algebraic closure k̄.
Proposition: Let X be a smooth complete curve over a field k. For any extension k → K we
have that the dualizing sheaf is stable under base change,
ωX/k ⊗k K = ωXK /K .
Proof: The cohomology group H 1 (XK , ωX ⊗k K) = H 1 (X, ωX ) ⊗k K is non null, so that there
exists a non null morphism DX ⊗k K → ωXK , and both are line sheaves.
Now we have exact sequences
0 −→ ωX ⊗k K −→ ωXK −→ C −→ 0
0 → H 0 (X, ωX )K → H 0 (XK , ωXK ) → H 0 (XK , C) → H 1 (X, ωX )K → H 1 (XK , DXK ) → 0
|| || || ||
H 1
(X, OX )∗K 1
H (XK , OXK ) ∗
H 0
(X, OX )∗K 0 ∗
H (XK , OX K
)
and we see that H 0 (XK , C) = 0. Since it is flasque, C = 0 and ωX ⊗k K = ωXK .
Theorem: The dualizing sheaf of a smooth complete curve X over a field k is the sheaf of
differentials ΩX/k .
Proof: The dualizing sheaf and the sheaf of differentials are stable under base changes k → K,
and the theorem holds over the algebraic closure k̄ (p. 316).
11.5 The Projective Spectrum

Definitions: The projective spectrum of a graded ring R = R0 ⊕ R1 ⊕ . . . is the subspace
X = Proj R of Spec R formed by all homogeneous prime ideals p = ⊕n pn not containing the
irrelevant ideal R+ = ⊕n≥1 Rn (the closed sets are (I)0 = {x ∈ X : I ⊆ px }, where I = ⊕n In
is a homogeneous ideal, and the open sets Uf = X − (f )0 , f homogeneous, form a base) with
the sheaf of rings OX attached to the presheaf of homogeneous localization
n o
U R(U ) = afnn : an , fn ∈ Rn ; fn without zeros on U
where R(U ) is the degree 0 component of the localization of R by all homogeneous elements
without zeros on U . The stalk OX,x is the homogeneous localization R(x) of R by all homogeneous
elements not vanishing at x.
We always assume that R1 generates R+ , so that the open sets Uf = Proj R − (f )0 , with
deg f = 1, cover X.
A projective space over A is any A-scheme isomorphic to some Pd,A = Proj A[x0 , . . . , xd ],
and projective schemes over A are closed subschemes of projective spaces.
Theorem: Uf = Spec R(f ) , when deg f = 1, so that Proj R is a scheme.
Proof: Homogeneous prime ideals of R not containing f correspond to homogeneous primes

q = ⊕∞ n n
−∞ f q0 of Rf = ⊕n f R(f ) , where q0 is a prime ideal of R(f ) .
This bijection Uf = Spec R(f ) is a homeomorphism since (an )0 ∩ Uf = fann 0 .

11.5. THE PROJECTIVE SPECTRUM 321
It is an isomorphism of ringed spaces since we have an isomorphism of presheaves
an /f n f m an
U ⊆ Uf (R(f ) )U −→ R(U ) , →
7 ·
bm /f m f n bm
Proposition: Any integral projective scheme Proj k[ξ0 , . . . , ξn ] over a field is complete.
n o
Pm (ξ0 ,...,ξn )
Proof: Let Σ = Q m (ξ0 ,...,ξn )
be the field of rational functions.
We have to show that any valuation ring V of Σ containing k centers at a unique point.
Take a quotient ξξsi of maximal valuation, so that all the fractions ξξri = ξξrs · ξξsi ∈ V,
ξ0 ξn
Ai = k ξi , . . . , ξi ⊂V
and V centers at the point of Uξi defined by Ai ∩ mV .

ξj
If V centers at a point of Uξj = Spec Aj , then Aj ⊂ V, and ξi is invertible in V, so that
ξi ξj
Aij = Ai ξj = Aj ξi ⊂ V.
Both centers of V are in Ui ∩ Uj = Spec Aij ; hence they coincide since V may not center at
two different points of an affine open set.
Definitions: If M = ⊕∞ −∞ Mn is a graded R-module, then M denotes the sheaf on Proj R

f
induced by the presheaf of homogeneous localization
n o
U M(U ) = m n
fn : mn ∈ M n , f n ∈ R n without zeros on U
and, as in the case of the sheaf of local rings, it coincides on Uf = Spec R(f ) with the sheaf
defined by M(f ) , when deg f = 1.
M (n) is the graded R-module M (n)d = Mn+d , and we put OX (n) = R(n)∼ .
OX (n) is a line sheaf since on Uf we have isomorphisms f n : OX |Uf → OX (n)|Uf .
If M is an OX -module, we put
M(n) = M ⊗OX OX (n).
Since Mf = ⊕n f n M(f ) = M(f ) ⊗R(f ) Rf , we have M(f ) ⊗R(f ) N(f ) = (M ⊗R N )(f ) .

Hence, the natural morphism M f ⊗O N e → (M ⊗R N )∼ is an isomorphism,
X
M (n)∼ = M
f ⊗O OX (n) = M
X
f(n),
OX (n) ⊗OX OX (m) = OX (n + m).
Example: Since k[x0 , . . . , xd ] is a unique factorization domain, the presheaf is a sheaf

n o
OPd (n)(U ) = PQ n+m (x0 ,...,xd )
m (x0 ,...,xd )
: Qm without zeros on U
On Pd,A we have that xi defines a global section of OPd,A (1) not vanishing at any point of
Ui = Spec A xx0i , . . . , xxdi . Hence the sections x0 , . . . , xd generate the stalk of OPd,A (1) at any

point, and they define an epimorphism OPd+1 d,A

→ OPd,A (1) that, after transposing it, should be
viewed as a line sub-bundle of the trivial bundle of rank d + 1.
This is the universal line sub-bundle: for any A-scheme X, the X-points of the projective
space are families, parameterized by X, of lines of the trivial bundle of rank d + 1.

Line quotients
Universal Property: HomA (X, Proj A[x0 , . . . , xd ]) = d+1 .
of OX
Proof: Let s0 , . . . , sd be sections of a line sheaf L on X generating the stalk at any point.
s s
On the open set Vi where si generates, we have sj = sji si , where sji ∈ OX (Vi ).
By the following lemma, the morphism of A-algebras
x s
A xx0i , . . . , xxdi −→ OX (Vi ), xji 7→ sji ,

induces a morphism of A-schemes φi : Vi → Ui , and we have φ∗i O(1) = L|Vi , φ∗i (xj ) = sj .
These morphisms φi coincide on intersections, hence they define a morphism φ : X → Pd,A
and an isomorphism φ∗ O(1) = L, φ∗ (xj ) = sj . The uniqueness is obvious.
Lemma: Hom(X, Spec A) = Hom(A, OX (X)), for any scheme X.
Proof: The natural map Hom(U, Spec A) → Hom(A, OX (U )) is bijective when U is affine (p.
310). Since both terms are sheaves of sets, it is bijective when U = X.
Corollary: Let E be a finite dimensional k-vector space. The group of all automorphisms of the
scheme P(E) = Proj S • E ∗ is the group P Sl(E) of projectivizations of semilinear automorphisms
of E, and the subgroup of all k-automorphisms is the subgroup P Gl(E) of projectivities.
Proof: Any automorphism σ : P(E) → P(E) defines an automorphism σ ∗ of the ring of global
sections Γ(P(E), OP(E) ) = k. Hence we are reduced to the case of a k-automorphism, which
directly follows from the universal property.
Definition: Let X be a non singular complete curve over an algebraically closed field. The
global sections of a line sheaf LD separate points and infinitely near points when for any
pair of closed points p, q there is a section of LD−p which is not a section of LD−p−q ; i.e., a
global section of LD vanishing at p and not at q (if p = q, not vanishing twice at p).
If X → Pd is the
morphism defined by a base s0 , . . . , sd of Γ(X, LD ), and Bi is the integral
closure of Ai = k ss0i , . . . , ssdi in ΣX , the global sections of LD separate points when the rings
Bi /mBi of the fibres are local, and separate infinitely near points when they are reduced; hence
Bi /mBi = Ai /m since k is algebraically closed, and Nakayama’s lemma shows that Ai = Bi .
That is to say, the morphism X → Pd is a closed embedding.
Corollary: If deg D > 2g, the global sections of LD separate points and infinitely near points.
Any non singular complete curve over an algebraically closed field is projective.
Proof: The divisors K − (D − p) and K − (D − p − q) are of negative degree and, by the

Riemann-Roch theorem, h0 (LD−p ) = h0 (LD−p−q ) + 1.
11.5.1 Projective Morphisms

Let R = A[ξ0 , . . . , ξd ], with A a noetherian ring.
We put X = Proj R, and we consider the open embedding j : Ui = X − (ξi )0 ,→ X.
The morphisms ξi : R(n) → R(n + 1) induce morphisms OX (n) → OX (n + 1).
If M is a quasi-coherent sheaf on X, we have morphisms M(n) → M(n + 1), and
fr+n fr+n
lim
−→
M(n) −→ j∗ (M|Ui ), m ⊗ hr 7→ ξin hr m
is an isomorphism. In fact, on the affine open sets Ui we have that lim

−→
Mn = Mξi when Mn = M
and the transition morphisms are ξi : M = Mn → Mn+1 = M .
11.5. THE PROJECTIVE SPECTRUM 323
Theorem: lim
−→
H p (X, Fi ) = H p (X, lim
−→
Fi ), when X is a noetherian topological space.
Proof: The natural morphism lim

−→
Fi (X) → (lim
−→
Fi )(X) is injective since X is compact (p. 331).
Any section s ∈ (lim−→
Fi )(X) is locally defined by some sections si ∈ Fi (U ).
These sections si do not coincide on intersections but, the intersections being compact, they
coincide as sections of Fk for a big index k, and they define a global section of Fk inducing s.
Therefore lim
−→
C • Fi is a flasque resolution of lim
−→
Fi , and
H p (X, lim p Γ(X, lim C • F ) = lim H p Γ(X, C • F ) = lim H p (X, F ).

−→
F i ) = H −→ i −→ i −→ i
Theorem: Any quasi-coherent sheaf M on X is the homogeneous localization M = M f of a

graded R-module M , finitely generated if M is coherent.

Proof: We put M = ⊕n Γ X, M(n) . The natural morphism M f → M is an isomorphism
f(Ui ) = S Γ(X,M(n))
M n ξin = lim
−→
Γ X, M(n) = Γ X, lim
−→
M(n) = Γ X, j∗ M|U i = M(Ui ).
When M is coherent, Lthe modules M(Ui ) are finitely generated, and there is a finitely
generated submodule N = n Nn ⊆ M such that N e → M is surjective.
e ⊆M
Hence an isomorphism since N f = M.
Theorem: Any coherent sheaf M on X admits a finite presentation

⊕j OX (nj ) −→ ⊕i OX (ni ) −→ M −→ 0
Proof: The homogeneous localization preserves exact sequences, and any finitely generated
graded R-module M admits a finite presentation ⊕j R(nj ) → ⊕i R(ni ) → M → 0.
n+d

free A-module of rank d n ≥ 0, p = 0

Theorem: H p (Pd,A , OPd,A (n)) = free A-module of rank −n−1
d n < −d, p = d

0 otherwise

Proof: Once we prove the statement for the sheaf of local rings OPd , it follows, by induction on
n and d, for OPd (n) and OPd (−n) according to the exact sequences
0 −→ OPd (n − 1) −→ OPd (n) −→ OPd−1 (n) −→ 0
0 −→OPd (−n) −→ OPd (−(n − 1)) −→ OPd−1 (−(n − 1)) −→ 0
The first one also shows that H p (Pd , O(n)) = H p (Pd , O(n + 1)) for all p ≥ 1, n ≥ 0.
Therefore, when p ≥ 1, we have
H p (Pd , O) = lim
−→
H p (Pd , O(n)) = H p (Pd , lim −→
O(n)) = H p (Pd , i∗ OU0 ) = H p (U0 , OU0 ) = 0.
Finally, since OPd (Ui ) = A xx0i , . . . , xxdi , it is clear that H 0 (Pd , OPd ) = R0 = A.

Example: If pC is the sheaf of ideals of a conic C of equation q(x0 , x1 , x2 ) = 0, then we have

an isomorphism q : OP2 (−2) → pC , and the exact sequence
0 −→ pC = OP2 (−2) −→ OP2 −→ OC −→ 0
shows that H 0 (C, OC ) = k, H 1 (C, OC ) = 0; hence (p. 315) any non-singular conic with a
rational point is isomorphic to the projective line (Steiner’s Theorem).
Serre’s Theorem: Let M be a coherent sheaf on X = Proj A[ξ0 , . . . , ξd ].

1. H p (X, M) is a finitely generated A-module, null when p > d.
2. There is an integer n0 such that, for any n > n0 , the sheaves M(n) are acyclic, generated by
the global sections,
Γ(X, M(n)) ⊗A OX −→ M(n) −→ 0
Proof: We may assume that X = Pd,A . The semiring of closed sets generated by the complements
of U0 , . . . , Ud defines a projection π : X → β∆d onto a finite space of dimension d, and π∗
preserves the cohomology of quasi-coherent sheaves (p. 309), hence (p. 308)
H p (X, M) = H p (β∆d , π∗ M) = 0, p > d.
On the other hand, if we consider a presentation
0 −→ K −→ ⊕i OX (ni ) −→ M −→ 0
the kernels and cokernels of the connecting morphisms H p (X, M) → H p+1 (X, K) are finitely
generated A-modules and, by descending induction, H p (X, M) is a finitely generated A-module.
0 −→ K(n) −→ ⊕i OX (ni + n) −→ M(n) −→ 0
⊕i OX (ni + n) is acyclic when ni + n 0; hence H p (X, M(n)) = H p+1 (X, K(n)); p ≥ 1.

By descending induction we conclude that H p (X, M(n)) = 0, n 0.
Moreover ⊕i OX (ni + n) is generated by global sections; hence so is M(n).
f(n)) when n 0.
Theorem: If M is a finitely generated graded R-module, then Mn = Γ(X, M
Proof: We have exact sequences 0 → K → ⊕i R(ni ) → M → 0, 0 → K 0 → ⊕j R(nj ) → K → 0,
0 −→ K(n) −→ ⊕i OX (ni + n) −→ M(n) −→ 0

0 −→ K0 (n) −→ ⊕j OX (nj + n) −→ K(n) −→ 0
When n 0, they remain exact on global sections, and the following commutative diagram
with exact rows let us conclude,
⊕j R(nj )n −→ ⊕i R(ni )n −→ Mn −→ 0
|| || ↓
H 0 X, ⊕j OX (nj + n) −→ H 0 X, ⊕i OX (ni + n) −→ H 0 (X, M) −→ 0

Bézout’s Theorem: Let R = k[x0 , x1 , x2 ], and let C, C 0 be two plane projective curves of
degrees n, m and equations Pn = 0, Pm 0 = 0, without common irreducible components.
If p, p0 are the respective sheaves of ideals, the intersection multiplicity at a point z is

the length of the OP2 ,z -module OP2 ,z /pz + p0z ,
(C 0 ∩ C)z = l OP2 ,z /pz + p0z ,

so that the dimension of the k-vector space Γ(P2 , O/p + p0 ) = ⊕z Oz /pz + p0z is the intersection
number, counting points with degree and multiplicity.
Since R is a unique factorization domain, and Pn , Pm0 have no common irreducible factor,
we have exact sequences

φ ϕ 0
0 −→ R(−n − m) −−→ R(−n) ⊕ R(−m) −−→ R −→ R/(Pn , Pm ) −→ 0
0 −→ OP2 (−n − m) −→ OP2 (−n) ⊕ OP2 (−m) −→ OP2 −→ OP2 /p + p0 −→ 0
11.6. COMPLETE CURVES 325
where φ(Q) = (Pm 0 Q, P Q) and ϕ(A, B) = P A − P 0 B. Now, the Euler-Poincaré characteristic

n n m
is additive, so that the intersection number is the product of the degrees:
dimk Γ(P2 , OP2 /p + p0 ) = χ(P2 , OP2 /p + p0 )

= χ(OP2 ) − χ(OP2 (−n)) − χ(OP2 (−m)) + χ(OP2 (−n − m))
= 1 − n−1 m−1 n+m−1

2 − 2 + 2 = nm.
11.6 Complete Curves

Let Σ be the field of rational functions on a complete curve X over a field k, and let X̄ be the
Riemann variety of Σ.
Since any discrete valuation centers at a unique point of X, we have a natural morphism
p : X̄ −→ X
and p−1 (Spec A) = Spec Ā, where Ā is the integral closure of A in Σ, so that the direct image
p∗ preserves (p. 309) the cohomology of quasi-coherent sheaves.
(∗) 0 −→ OX −→ p∗ OX̄ −→ C −→ 0
where C = (p∗ OX̄ )/OX is supported at the singular points of X.
Definition: The arithmetic genus of X is π = dimk H 1 (X, OX ) and the geometric genus
of X is g = dimk H 1 (X̄, OX̄ ).
in Σ, we have k = H 0 (X, OX ) = H 0 (X̄, OX̄ ), and the exact

When k is algebraically closedP
sequence (∗) shows that g = π − dimk (Ōx /Ox ).
x∈X
Theorem: If M is a coherent sheaf on a complete curve X, the k-vector spaces H n (X, M) have
finite dimension.
Proof: Let φ : M → p∗ p∗ M be the natural morphism. The theorem holds for Ker φ and Coker φ
since both are coherent sheaves supported at the singular points of X.
It also holds for p∗ p∗ M since H n (X, p∗ p∗ M) = H n (X̄, p∗ M), and we conclude,
0 −→ Im φ −→ p∗ p∗ M −→ Coker φ −→ 0
0 −→ Ker φ −→M −→ Im φ −→ 0
Proposition: The dualizing sheaf of a plane curve X of degree n is
ωX = OX (n − 3).
Proof: We have an exact sequence 0 → OP2 (m − n) → OP2 (m) → OX (m) → 0.

Hence h1 (OX (n − 3)) = h2 (OP2 (−3)) = 1, and OX (n − 3) is a minimal pair with any non
null element of H 1 (X, OX (n − 3))∗ , since torsion sheaves are acyclic.
If it is dominated by another minimal pair Mξ , when m 0,
0 −→ OX (n − 3) −→ M −→ K −→ 0 is exact
0 −→ Γ(X, OX (m + n − 3)) −→ Γ(X, M(m)) −→ Γ(X, K(m)) −→ 0 is exact
h0 (M(m)) ≤ dim Hom(OX (−m), ωX ) = h1 (OX (m))

2 2 m+n−1 m−1
= h (O(−m − m)) − h (O(−m)) = −
2 2

m+n−1 m−1
h0 (OX (m + n − 3)) = h0 (O(m + n − 3)) − h0 (O(m − 3)) = −
2 2
h0 (K(m)) = 0
and we conclude that K = 0; hence OX (n − 3) = M, and OX (n − 3) = ωX .
Corollary: If X is a plane curve of degree n and LK is the canonical line sheaf of the non
singular model X̄, we have an isomorphism of p∗ OX̄ -modules
p∗ (LK ) = HomOX (p∗ OX̄ , OX ) ⊗OX OX (n − 3).
Proof: The argument of p. 316 shows that p∗ ωX̄ = HomOX (p∗ OX̄ , ωX ).
Since ωX is a line sheaf, we see that Hom(p∗ OX̄ , ωX ) = Hom(p∗ OX̄ , OX ) ⊗OX ωX .
Chapter 12
Algebraic Topology I
12.1 Cohomology with Supports

Definitions: Let F be a sheaf on a topological space X. The support of a section s ∈ F(U ) is
supp(s) = {x ∈ U : sx 6= 0}
and it is closed in U . We put Γ(U, F) = F(U ), the subgroup of sections with compact support
is denoted by Γc (U, F) and, given a closed set Y in X, the subgroup of sections with support
in Y is denoted by ΓY (X, F). The cohomology groups with compact supports and the
local cohomology groups with support in Y are defined with the Godement resolution
Hcn (X, F) = H n Γc (X, C • F)

HYn (X, F) = H n ΓY (X, C • F)

and the proofs of p. 306 remain valid because flasque sheaves are Γc -acyclic and ΓY -acyclic
since, if 0 → F 0 → F → F 00 → 0 is an exact sequence and F 0 is flasque, any section s00 ∈ F 00 (U )
comes from a section s ∈ F(U ), and we may fix s so that supp(s) = supp(s00 ).
In fact, s defines on V = U − supp(s00 ) a section of F 0 , which may be extended to a section
s on U , and now s − s0 vanishes on V , so that s − s0 maps to s00 and supp(s − s0 ) = supp(s00 ).
0
S
Definition: A sheaf of rings O admits partitions of unity if for any open cover X P= i Ui
there are fi ∈ O(X) such that the family {supp fi } is locally finite, supp fi ⊆ Ui , and i fi = 1.
The sheaf of smooth functions CX∞ on a smooth manifold, and the sheaf of continuous func-
tions CX on a σ-compact space admit partitions of unity (pp. 250, 249).

On any σ-compact space X, the sheaf S of integer discontinuous functions C 0 Z admits parti-
tions of unity, since any open cover X = i Ui admits a partition of unity {φi } by continuous
functions. If for any point x ∈ X we fix an index i such that φi (x) 6= 0, and we take fP
i: X → Z
with value 1 at any point where we choose the index i, and 0 at any other point, then i fi = 1
and the family supp fi ⊆ supp φi ⊆ Ui is locally finite.
Lemma: If a sheaf of rings O admits partitions of unity, any O-module M is acyclic (and
Γc -acyclic when X is σ-compact).
Proof: H p (X, M) is the p-th cohomology group of the complex of O(X)-modules

d0 d1 d2
Γ(X, C 0 M) −−→ Γ(X, C 1 M) −−→ Γ(X, C 2 M) −−→ . . .
If dp+1 z = 0, there is an open cover X = i Ui and zi ∈ Γ(Ui , C p M) such that z|Ui = dp zi .

S
327
328 CHAPTER 12. ALGEBRAIC TOPOLOGY I
If {fi } is a partition of unity subordinated to {Ui }, then
dp p
P P P
i fi zi = i fi d zi = i fi z = z.
If X is σ-compact and supp z is compact, we may P take U0 = (supp z)c , U1 , . . . , Un , with

U 1 , . . . , U n compact, and z0 = 0, so that the support of i fi zi is compact.
Theorem: The De Rham cohomology groups of a smooth manifold X are topological invariants,
p
HDR (X) = H p (X, R).
Proof: Let Ωp be the sheaf of differential p-forms on X.

Any closed form is locally exact by Poincaré’s lemma, so that we have an exact sequence
∞ d d d
0 −→ R −→ CX −−→ Ω1 −−→ Ω2 −−→ . . .
∞ -module.
which is an acyclic resolution of the constant sheaf R since Ωp is a CX
p
By De Rham’s theorem, H p (X, R) = H p Γ(X, Ω• ) = HDR (X). q.e.d.
We have an analogous result for the cohomology groups with compact support,
Γc (X, Ωp )
Hcp (X, R) = ·
d(Γc (X, Ωp−1 ))
Definition: If Y is a subspace of X, the restriction F|Y of a sheaf F on X is the sheaf of

continuous sections of the local homeomorphism (F et )|Y → Y .
The stalk at any point coincides with the stalk of F, and we put H n (Y, F) = H n (Y, F|Y ).
When i : Y → X is a closed subset, the direct image i∗ preserves the cohomology groups (p.
309), and if we put FY = i∗ (F|Y ), then we have
H n (X, FY ) = H n (Y, F).
Restriction of continuous sections defines an epimorphism F → FY , since (FY )x = Fx or 0,

according to x ∈ Y or not, and the kernel is denoted by FU , where U = X − Y ,
0 −→ FU −→ F −→ FY −→ 0
Taking sections on an open set V we see that
FU (V ) = {s ∈ F(V ) : supp s ⊆ U ∩ V }
= {s ∈ F(U ∩ V ) : supp s is closed in V }
Γc (X, FU ) = Γc (U, F) when X is separated.
For example, Zet

U , apart from the zero section, has a copy of U for each non null integer.
Hom(ZU , F) = F(U ), and any sheaf F admits an epimorphism ⊕i ZUi → F → 0.
Lemma: Hcn (X, FU ) = Hcn (U, F), when X is σ-compact.
Proof: If M is an O-module, so is MU since supp(f m) ⊆ supp m.

Hence (C n F)U is a C 0 Z-module and (C • F)U is a Γc -acyclic resolution of FU ,
Hcn (X, FU ) = H n Γc (X, (C • F)U ) = H n Γc (U, C • F) = Hcn (U, F).

12.1. COHOMOLOGY WITH SUPPORTS 329
Mayer-Vietoris Exact Sequences: Let F be a sheaf on a topological space X. If U1 , U2 are

open subsets of X, we have an exact sequence
δ δ
− H n (U1 ∪ U2 , F) → H n (U1 , F) ⊕ H n (U2 , F) → H n (U1 ∩ U2 , F) →
... → − H n+1 (U1 ∪ U2 , F) → . . .
and if moreover X is σ-compact, we also have an exact sequence

δ δ
− Hcn (U1 ∩ U2 , F) → Hcn (U1 , F) ⊕ Hcn (U2 , F) → Hcn (U1 ∪ U2 , F) →
... → − Hcn+1 (U1 ∩ U2 , F) → . . .
If Y1 , Y2 are closed subsets of X, we have an exact sequence

δ δ
− H n (Y1 ∪ Y2 , F) → H n (Y1 , F) ⊕ H n (Y2 , F) → H n (Y1 ∩ Y2 , F) →
... → − H n+1 (Y1 ∪ Y2 , F) → . . .
Proof: We have exact sequences
0 −→ Γ(U1 ∩ U2 , C • F) −→ Γ(U1 , C • F) ⊕ Γ(U2 , C • F) −→ Γ(U1 ∩ U2 , C • F) −→ 0.

0 −→ FU1 ∩U2 −→ FU1 ⊕ FU2 −→ FU1 ∪U2 −→ 0, and the above lemma.
0 −→ FY1 ∪Y2 −→ FY1 ⊕ FY2 −→ FY1 ∩Y2 −→ 0, and H n (X, FY ) = H n (Y, F).
Closed Subspace Exact Sequence: If Y is a closed set in a σ-compact space X, and we put
U = X − Y , we have an exact sequence
δ δ
. . . −−→ Hcn (U, F) −→ Hcn (X, F) −→ Hcn (Y, F) −−→ Hcn+1 (U, F) −→ . . .
Proof: The cohomology exact sequence of 0 → FU → F → FY → 0 and the above lemma.
Local Cohomology Exact Sequence: If Y is a closed set in X, and we put U = X − Y , we

have an exact sequence
δ δ
. . . −−→ HYn (X, F) −→ H n (X, F) −→ H n (U, F) −−→ HYn+1 (X, F) −→ . . .
Proof: The sequence 0 −→ ΓY (X, C • F) −→ Γ(X, C • F) −→ Γ(U, C • F) −→ 0 is exact.
Excision: HYn (X, F) = HYn (U, F), when U is an open neighborhood of Y .
Proof: We have ΓY (X, F) = ΓY (U, F) for any sheaf F on X.
Theorem: Any constant sheaf G on the cube C = [0, 1]n is acyclic,
H p (C, G) = 0, p ≥ 1.
Proof: By induction on n, and it is obvious when C is a point.

If n ≥ 1, we decompose C = P1 ∪ P2 as a union of two parallelograms intersecting at a cube
of dimension n − 1. Since the morphism
G ⊕ G = H 0 (P1 , G) ⊕ H 0 (P2 , G) −→ H 0 (P1 ∩ P2 , G) = G, (g1 , g2 ) 7→ g2 − g1 ,
is surjective, by Mayer-Vietoris H p (C, G) = H p (P1 , G) ⊕ H p (P2 , G), p ≥ 1.

Iterating the subdivisions, C = P1 ∪ . . . ∪ Pr , we obtain that
H p (C, G) = H p (P1 , G) ⊕ . . . ⊕ H p (Pr , G), p ≥ 1.
Since any cohomology class vanishes on a neighborhood of each point, and C is compact, we
conclude that H p (C, G) = 0, p ≥ 1. q.e.d.
(
G p = 0, n
1. Cohomology of Spheres: H p (Sn , G) =
0 p 6= 0, n
The Mayer-Vietoris exact sequence for two hemispheres intersecting at the equator.
(
G p=n
2. Hcp (Rn , G) =
0 p 6= n
The closed subspace exact sequence when Y is a point of Sn , and U = Rn .
3. Rn and Rm are not homeomorphic when n 6= m.
4. Let X be a compact Hausdorff space, and let ∅ = X−1 ⊂ X0 ⊆ X1 . . . ⊆ Xd = X be closed sets

such that Xi − Xi−1 = qni Ri . The cohomology groups H p (X, k) with coefficients in a field
k have finite
P dimension, they vanish when p > d, P
and the Euler-Poincaré characteristic
χ(X) := p p i
p (−1) dimk H (X, k) of X is χ(X) = i (−1) ni .
Let U = Rd be a connected component of Xd − Xd−1 . The exact sequence of the closed

subspace Y = X − U shows that H p (X, k) = H p (Y, k)
Pwhen p 6= d, d − 1, and that we have
an exact sequence (and we conclude by induction on i ni )
0 −→ H d−1 (X, k) −→ H d−1 (Y, k) −→ k −→ H d (X, k) −→ H d (Y, k) −→ 0.
(
G p = 0, 2, . . . , 2n
5. Cohomology of Complex Projective Spaces: H p (Pn,C , G) =
0 otherwise
Induction on n and the exact sequence of the closed subspace Y = Pn−1 , U = Cn .
6. If π : X̄ → X is a covering of a topological manifold, H p (X̄, G) = H p (X, π∗ G).

We must show that the sequence 0 → π∗ G → π∗ (C • G) is exact (p. 309). Now, any point of
X admits a base of neighborhoods Ci which are cubes and
H p (π −1 Ci , G) = H p qCi , G = H p (Ci , G) = 0, p ≥ 1.
Q
(
F2 0 ≤ p ≤ n
7. Cohomology of Real Projective Spaces: H p (Pn,R , F2 ) =
0 p>n
By induction on n, the exact sequence of the closed subspace Y = Pn−1 , U = Rn , shows that
it is enough to check that H n (Pn , F2 ) 6= 0. Now, if π : Sn → Pn is the universal covering, we
have an exact sequence of sheaves on Pn
tr
0 −→ F2 −→ π∗ F2 −−→ F2 −→ 0
where trf = f + τ f , and {Id, τ } is the Galois group of π. Hence H n (Pn , F2 ) 6= 0 by the
cohomology exact sequence H n (Pn , F2 ) → H n (Pn , π∗ F2 ) = F2 → H n (Pn , F2 ).
Cohomology and Dimension

Lemma: Let A be a lattice semiring, and let C be a sheaf on X = Spec A. If the restriction
morphism C(X) → C(U ) is surjective for any basic open set U , then C is acyclic.
Proof: If 0 → F 0 → F → F 00 → 0 is an exact sequence and F 0 (X) → F 0 (U ) is surjective for any

basic open set U , let us see that we have exact sequences
i p
0 −→ F 0 (U ) −→ F(U ) −−→ F 00 (U ) −→ 0
12.1. COHOMOLOGY WITH SUPPORTS 331
If s00 ∈ F 00 (U ), since U is compact, it admits a finite basic cover U = U1 ∪ . . . ∪ Un where s00 |Ui
comes from of a section of F, and the argument of p. 305 proves that s00 comes from a section of
F on the basic open set U1 ∪ U2 (in the spectrum of a semiring, basic open sets form a lattice!).
Hence s00 comes from a section of F on U = U1 ∪ . . . ∪ Un .
Now, if F(X) → F(U ) also is surjective, so is F 00 (X) → F 00 (U ), and the argument of p. 306
let us conclude (replacing flasque sheaves by sheaves C such that C(X) → C(U ) is surjective for
any basic open set U ).
Theorem: Let A be a lattice semiring, and let F be a sheaf on X = Spec A. If Fx = 0 at any

point x ∈ X of dimension > d, then H p (X, F) = 0 for all p > d. In particular,
H p (X, F) = 0, p > dim X.
Proof: Let i : Xd → X be the subspace of points of dimension ≥ d. The natural morphism
φ : F −→ i∗ (F|Xd )
induces an isomorphism on stalks at any point of Xd and, if i∗ (F|Xd ) is acyclic, the argument
of p. 308 let us conclude. By the lemma, it is enough to show that the morphism
(F|Xd )(Xd ) = (i∗ F|Xd )(X) −→ (i∗ F|Xd )(U ) = (F|Xd )(Xd ∩ U )
is surjective on any basic open set U .

If the support |s| of s ∈ (F|Xd )(Xd ∩ U ) is closed in Xd , then s may be extended by 0.
Now, |s| is closed in Xd ∩ U ; hence the closure Y of |s| in U has no point of dimension > d,
and |s| is closed in Xd by the following lemma.
Lemma: If a point x ∈ X − Uf is adherent to a closed set Y of a basic open set Uf , then the
dimension of x is smaller than that of some point of Y .
Proof: Since any closed set in Spec A is the spectrum of a semiring, we may assume that Y = Uf .
If p is the prime ideal of x, there is no h ∈ A − p such that Uh ∩ Uf = ∅.
Hence p contains the annihilator of f , and f is not zero in Ap .
Since the intersection of all prime ideals of a semiring is null, f is not in some prime ideal q
of Ap , different from p since x ∈ (f )0 , and q defines a point of Uf of bigger dimension than the
dimension of x.
Theorem: If X is a compact separated space, we have H p (X, F) = 0, p > dim X.
Proof: If B is a base of the topology of X of dimension d, then X = Spec m B and the inclusion
j : X → §B admits a continuous retract r : Spec B → X (p. 224).
Moreover, r−1 (U ) is contained in any open set in Spec B intersecting X at U .
Hence (r∗ C)x = Cx , and r∗ preserves cohomology. Since F = r∗ j∗ F,
H p (X, F) = H p (Spec B, j∗ F) = 0, p > d.
Fundamental Theorem of Dimension Theory: dim Rn = n.
Proof: On a closed ball Bn we have H n (Bn , ZU ) = Hcn (U, Z) = Z when U ' Rn .

Hence dim Rn ≥ dim Bn ≥ n, and we already know that dim Rn ≤ n (p. 233).
Lemma: lim
−→
Γ(X, Fi ) = Γ(X, lim
−→
Fi ), when X is a compact Hausdorff space.
Proof: The natural morphism lim

−→
Γ(X, Fi ) → Γ(X, lim
−→
Fi ) is injective:
If s ∈ Fi (X) vanishes as a section of the inductive limit, then it vanishes on a neighborhood
of any point x as a section of Fj for some index j ≥ i, independent of x since X is compact.
Hence s = 0 in Fj (X).
The morphism is surjective: S
Given a global section s of the inductive limit, on some finite open cover X = r Ur it
comes from some sections sr ∈ Fi (Ur ), which do not coincide on intersections; but any point x
has an open neighborhood Wx where all the sections sr (with x ∈ Ur ) define the same section
sx ∈ Fj (Wx ), j i. S
If we consider an open cover X = r Vr , where V r ⊆ Ur , we may assume that Wx ⊆ Vr when
x ∈ Vr , and Wx ∩ V r = ∅ when x ∈ / V r ; so that, when Wx and Wy intersect, we have x, y ∈ Ur
for some index r, and the sections sx , sy coincide on Wx ∩ Wy , when j i. A finite number of
such open sets Wx cover X, and the sections sx define a global section of Fj inducing s.
Theorem: Hcn (X, lim

−→
Fi ) = lim
−→
Hcn (X, Fi ), when X is σ-compact.
Proof: The above argumrnt shows that the morphism lim −→

Γc (X, Fi ) → Γc (X, lim
−→
Fi ) is injective.
Now, given any global section s of the inductive limit with compact support, contained in
the interior of a compact K 0 , contained in the interior of a compact K 00 , by the former lemma
it comes from a section s00 of some sheaf Fj on K 00 , vanishing outside of the interior of K 0 when
j is big enough. Extending s00 by 0 outside from K 0 , we obtain a global section of Fj , with
support contained in K 0 , inducing s.
Any C 0 Z-module is Γc -acyclic, because C 0 Z admits partitions of unity (p. 327); hence
Hcn (X, lim Fi ) = H n Γc (X, C • Z ⊗Z (lim Fi )) = H n lim Γc (X, C • Z ⊗Z Fi )]

−→ −→ −→
H n Γc (X, C • Z ⊗Z Fi ) = lim Hcn (X, Fi ).

= lim
−→ −→
Theorem: If X is a topological manifold of dimension n, we have Hcp (X, F) = 0, p > n.
Proof: Hcp (X, F) = Hcp (X, lim

−→
FU ) = lim
−→
Hcp (X, FU ) = lim
−→
Hcp (U, F),
where U runs over finite unions U = V1 ∪ . . . ∪ Vm of open sets contained in compact sets of
dimension n, so that Hcp (V, F) = H p (K, FV ) = 0, p > n, for any open set V ⊆ Vi .
Now, if Hcp (V1 ∪ . . . ∪ Vr , F) = 0, p > n, the Mayer-Vietoris exact sequence shows that also
Hc (V1 ∪ . . . ∪ Vr+1 , F) = 0, p > n. Hence Hcp (U, F) = 0, p > n, and we conclude.
p
12.2 Homological Algebra

Definitions: A bicomplex (K •• , d1 , d2 ) is a family of A-modules {K p,q }p,q∈Z with morphisms
of A-modules dpq
1 : K
p,q → K p+1,q , dpq : K p,q → K p,q+1 such that the following diagram is
2
commutative, and all the rows and columns are complexes:
↑ d2 ↑ d2
d1 d1 d
. . . −−→ K p,q+1 −−→ K p+1,q+1 −−1→ . . .
↑ d2 ↑ d2
d d d
. . . −−1→ K p,q −−1→ K p+1,q −−1→ . . .
↑ d2 ↑ d2
A morphism of bicomplexes f : K •• → L•• is a family of morphisms f pq : K pq → Lpq

commuting with the differentials, f d1 = d1 f , f d2 = d2 f .
12.2. HOMOLOGICAL ALGEBRA 333
We always see a bicomplex K •• as a complex (we introduce the sign so that d ◦ d = 0)
(K •• )n = K p,q , d(mpq ) = d1 mpq + (−1)p d2 (mpq )

L
p+q=n
and we say that K •• has bounded diagonals if all the direct sums ⊕p+q=n K p,q are finite.
A complex K • is viewed as a bicomplex with K 0n = K n and K pn = 0 when p 6= 0.
If r ∈ Z, then K • [r] denotes the complex K • [r]n = K n+r with the differential (−1)r d, so
that H n (K • [r]) = H n+r (K • ).
A morphism of complexes f : K • → L• may be viewed as a bicomplex with K n at degree
(−1, n), Ln at degree (0, n), and d1 = f . The cone Cone• f of f is the associated simple complex
Conen f = K n+1 ⊕ Ln , d(a, b) = (−da, f (a) + db).
We have an exact sequence of complexes
0 −→ L• −→ Cone• f −→ K • [1] −→ 0
and the connecting induced on cohomology is just f .

Hence, f is a quasi-isomorphism if and only if Cone• f is acyclic, H n (Cone• f ) = 0.
Lemma: Let K •• be a bicomplex with bounded diagonals. If the columns K p• are exact sequences,
then H n (K •• ) = 0, ∀n ∈ Z.
Proof: If [m] ∈ H n (K •• ) and m = mp,q + mp+1,q−1 + mp+2,q−2 + . . ., where mp,q 6= 0, then the
cycle condition dm = 0 implies that d2 mp,q = 0.
Hence mp,q = d2 np,q−1 , with np,q−1 ∈ K p,q−1 , since the columns are exact, and [m] = [m0 ],
where m0 = m − (−1)p dnp,q−1 has lower height than m in the bicomplex.
Hence [m] has representants of arbitrarily low height.
The diagonals being bounded, we conclude that [m] = 0.
Bicomplex Theorem: Let f : K •• → L•• be a morphism of bicomplexes with bounded diag-

onals. If f : K •n −
∼
→ L•n (resp. f : K n• −
∼
→ Ln• ) is a quasi-isomorphism for all n ∈ Z, then
••
f: K − ∼ ••
→ L is a quasi-isomorphism.
Proof: If f is injective, the exact sequences 0 → K •q → L•q → C •q = L•q /K •q → 0 show that

H n (C •q ) = 0 and, by the lemma, H n (C •• ) = 0.
Now the exact sequence 0 → K •• → L•• → C •• → 0 shows that H n (K •• ) − ∼
→ H n (L•• ).
•q •q •q
In the surjective case, consider the kernel C of K → L , and in the general case (not
used in these notes) consider the cone C •q of K •q → L•q .
12.2.1 The Functors TorA n

n and ExtA
Definition: Let P• → M → 0, P•0 → N → 0 be projective resolutions. Then we have quasi-

isomorphisms Pn ⊗A P•0 −
∼
→ Pn ⊗A N , P• ⊗A Pn0 −
∼
→ M ⊗A Pn0 because projective modules are flat.
By the bicomplex theorem P• ⊗A P•0 −
→ P• ⊗A N , P• ⊗A P•0 −
∼ ∼
→ M ⊗A P•0 , and we put
0 0
TorA
n (M, N ) = Hn (P• ⊗A N ) = Hn (P• ⊗A P• ) = Hn (M ⊗A P• ),
so that these modules do not depend on the resolutions, and TorA A

n (M, N ) = Torn (N, M ).
Since M ⊗A (−) is right exact, TorA
0 (M, N ) = M ⊗A N .
If 0 → N 0 → N → N 00 → 0 is an exact sequence, then the exact sequence of complexes

0 → P• ⊗A N 0 → P• ⊗A N → P• ⊗A N 00 → 0 induces an exact sequence
0 00 0
. . . −→ TorA A A A
n (M, N ) −→ Torn (M, N ) −→ Torn (M, N ) −→ Torn−1 (M, N ) −→ . . .
Definition: If 0 → N → I • is an injective resolution, then HomA (Pn , N ) − ∼

→ HomA (Pn , I • )
n ∼ n
and HomA (M, I ) − → HomA (P• , I ). By the bicomplex theorem we have quasi-isomorphisms
→ HomA (P• , I • ), HomA (M, I • ) −
∼
HomA (P• , N ) − ∼
→ HomA (P• , I • ), and we put
ExtnA (M, N ) = H n HomA (P• , N ) = H n HomA (P• , I • ) = H n HomA (M, I • ) ,

so that these modules do not depend on the resolutions.

Since HomA (M, −) is left exact, Ext0A (M, N ) = HomA (M, N ).
If 0 → N 0 → N → N 00 → 0 is an exact sequence, then the exact sequence of complexes
0 → HomA (P• , N 0 ) → HomA (P• , N ) → HomA (P• , N 00 ) → 0 induces an exact sequence
δ
. . . −→ ExtnA (M, N 0 ) −→ ExtnA (M, N ) −→ ExtnA (M, N 00 ) −−→ Extn+1 0
A (M, N ) −→ . . .
If 0 → M 0 → M → M 00 → 0 is an exact sequence, then the exact sequence of complexes

0 → HomA (M 00 , I • ) → HomA (M, I • ) → HomA (M 0 , I • ) → 0 induces an exact sequence
δ
. . . −→ ExtnA (M 00 , N ) −→ ExtnA (M, N ) −→ ExtnA (M 0 , N ) −−→ Extn+1 00
A (M , N ) −→ . . .
Proposition: Any module M over a principal ideal domain A admits a projective resolution
0 −→ P1 −→ P0 −→ M −→ 0.
Proof: Quotients of injective (= divisible, p. 123) modules are injective; hence any module N
admits an injective resolution 0 → N → I → I/N → 0, and Extn (−, N ) = 0, n ≥ 2.
Now, if 0 → K → L → M → 0 is an exact sequence, where L is a free module, then
Ext1 (K, N ) = Ext2 (M, N ) = 0
for any module N . Hence the functor HomA (K, −) is exact and K is projective.
12.2.2 Derived Functors

Definition: A covariant functor F : A B from the category of A-modules into the category
of B-modules is additive when the maps F : HomA (M, N ) −→ HomB (F (M ), F (N )) are group
morphisms, F (f + g) = F (f ) + F (g).
In such a case, if K • = {K n , dn } is a complex of A-modules, then F (K • ) = {F (K n ), F (dn )}
is a complex of B-modules, since F (d)2 = F (d2 ) = F (0) = 0.
Moreover, additive functors transform split exact sequences into split exact sequences, hence
they preserve finite direct sums, F (M ⊕ N ) = F (M ) ⊕ F (N ), and commute with the formation
of the simple complex when the diagonals are bounded.
Lemma: If F is an additive functor, and f : I • − ∼

→ J • is a quasi-isomorphism of bounded below
injective complexes, then F (f ) : F (I • ) −
∼
→ F (J • ) is a quasi-isomorphism.
Proof: If J • = 0, then I • = 0 → In → In+1 → . . . is an exact sequence of injective modules;

hence it splits and F (I • ) is also an exact sequence. In the general case, Cone• f −
∼
→ 0; hence
Cone• F (f ) = F (Cone• f ) −∼
→ 0, and F (f ) is a quasi-isomorphism. q.e.d.
12.2. HOMOLOGICAL ALGEBRA 335
Any A-module M is a quotient, L0 M → M → 0, of the free module generated by the elements

of M , modulo the submodule generated by the 0 of M (so that the functor L0 preserves the
morphism 0 and complexes); hence M admits a functorial projective (in fact free) resolution
L• M → M → 0 preserving complexes.
Now (p. 134) M is a submodule of the injective module I 0 M = (L0 M ∗ )∗ , and we see that
M also admits a functorial injective resolution 0 → M → I • M preserving complexes.
Definition: If an additive covariant functor F : A B is left exact (if 0 → M 0 → M → M 00 is

0 00
exact, so is 0 → F (M ) → F (M ) → F (M )) the right derived functor Rn F : A B is
Rn F (M ) = H n [F (I • M )] ,
and M is F -acyclic when Rn F (M ) = 0 for all n > 0. Injective modules are F -acyclic, and the
condition of F being left exact states that R0 F = F .
Now we extend this definition to bounded below complexes, and we say that a complex K • is
injective, F -acyclic,...., when so are all the modules K n . By the bicomplex theorem K • → I • K •
is a quasi-isomorphism when K • is bounded below, and we put
RF (K • ) = F (I • K • ),
Rn F (K • ) = H n [F (I • K • )] .
By the lemma, if f : K • −
∼
→ L• is a quasi-isomorphism of bounded below complexes, then so
• → RF (L ), so that f : Rn F (K • ) −
∼
is f : RF (K ) − • ∼
→ Rn F (L• ) is an isomorphism.
Lemma: If a bounded below complex A• is F -acyclic, then F (A• ) −

∼
→ RF (A• ).
Proof: The sequence 0 → F (Aq ) → F (I • Aq ) is exact because Aq is F -acyclic, and the bicomplex
theorem shows that F (A• ) −
∼
→ F (I • A• ).
Corollary: If A• , Ā• are F -acyclic bounded below complexes, any quasi-isomorphism A• −

∼
→ Ā•
induces a quasi-isomorphism F (A• ) − → F (Ā• ).
∼
Definition: The morphism L• → I • L• induces a morphism F (L• ) → RF (L• ), hence morphisms

H n [F (L• )] → Rn F (L• ). So, any quasi-isomorphism K • −
∼
→ L• induces natural morphisms1
n • n • n •
DR : H [F (L )] → R F (L ) = R F (K ), and any commutative square
K•
∼
/ L•
f t

K̄ •
∼
/ L̄•
induces commutative squares H n [F (L• )]

DR / Rn F (K • )
F (t) f

H n [F (L̄• )]
DR / Rn F (K̄ • )
De Rham’s Theorem: Let K • , A• be bounded below complexes. If A• is F -acyclic, any quasi-

isomorphism K • −
∼
→ A• induces isomorphisms DR : H n [F (A• )] → Rn F (K • ).
1
Coinciding up to a sign with the connecting iteration of p. 307; but, in the case of the quasi-isomorphism
M → I • M , the morphism DR : H n [F (I • M )] → Rn F (M ) is the identity (p. 531).
Proof: We have quasi-isomorphisms F (A• ) −

∼
→ RF (A• ) ←
− RF (K • ).
∼
Derived Functors Exact Sequence: Any exact sequence 0 → K • → L• → M • → 0 of

bounded below complexes induces a long exact sequence
δ δ
. . . −−→ Rn F (K • ) −→ Rn F (L• ) −→ Rn F (M • ) −−→ Rn+1 F (K • ) −→ . . .
Proof: We have a commutative diagram with exact rows
0 / K• / L• / M• /0

0 / I •K • / I • L• / I • L• /I • K • /0
where I • L• /I • K • is an injective resolution of M • . We conclude considering the cohomology

exact sequence induced by the exact sequence of complexes
0 −→ F (I • K • ) −→ F (I • L• ) −→ F (I • L• /I • K • ) −→ 0.
Theorem: Let 0 → M → A0 → A1 → . . . → An−1 → Mn → 0 be an exact sequence. If the

modules Ai are F -acyclic, we have canonical isomorphisms
Rp F (Mn ) = Rp+n F (M ), p ≥ 1.
Proof: Any F -acyclic resolution 0 → Mn → B 0 → B 1 . . . defines an F -acyclic resolution of M ,
0 −→ M −→ A0 −→ . . . −→ An−1 −→ B 0 −→ B 1 −→ . . .
and we have isomorphisms Rp F (Mn ) = H p [F (B • )] = Rp+n F (M ), p ≥ 1. q.e.d.

1. The right derived functors of HomA (M, −) are the functors ExtnA (M, −).
2. Right exact covariant additive functors may be derived on the left using projective resolutions
instead of injective resolutions, and may be extended to bounded above complexes. The left
derived functors of M ⊗A (−) are the functors TorA n (M, −).
3. Left exact contravariant additive functors may be derived on the right using projective res-
olutions instead of injective resolutions, and may be extended to bounded above complexes.
The right derived functors of HomA (−, M ) are the functors ExtnA (−, M ).
4. Categories of modules may be replaced by categories of sheaves since any sheaf admits an
injective functorial resolution by injective sheaves (see below), and the right derived functors
of Γ, Γc and ΓY coincide with the cohomology groups defined with the Godement resolution,
since any injective sheaf I is flasque,
I(X) = Hom(ZX , I) −→ Hom(ZU , I) = I(U ) −→ 0.
Hence, for any bounded below complex of sheaves K• we have hypercohomology groups
Hn (X, K• ), Hnc (X, K• ), HnY (X, K• ), and it is immediate to generalize to this case the forth-
coming inverse image and cup product.
Definition: Let O be a sheaf of rings on X. If we fix at any point x ∈ X an Ox -module Mx ,
the corresponding Godement sheaf C is the flasque sheaf
Q
C(U ) = Mx ,
x∈U
12.3. INVERSE IMAGE 337
and for any O-module N we have a natural isomorphism

Q
HomO (N , C) = HomOx (Nx , Mx ).
x∈X
Hence, C is an injective O-module when Mx is an injective Ox -module for any x ∈ X.

In fact, Mx defines an O-module concentred at the closure of x,
(
Mx x ∈ U
Mx (U ) =
0 x∈/U
Q
and the Godement sheaf C is the direct product sheaf, C = x Mx .
Moreover, it is easy to check that HomO (N , Mx ) = HomOx (Nx , Mx ), so that
Q Q Q
HomO (N , C) = HomO (N , Mx ) = HomO (N , Mx ) = HomOx (Nx , Mx ).
x∈X x∈X x∈X
Now, if M is an O-module, and for any stalk Mx we consider the injective Ox -module
I 0 Mx , the corresponding Godement sheaf I 0 M is an injective O-module, and we have an
injective morphism of O-modules M → I 0 M.
Therefore, any O-module M admits an injective functorial resolution 0 → M → I • M.
12.3 Inverse Image

Definition: If f : Y → X is a continuous map, the inverse image of a sheaf F on X is the
following sheaf f −1 F on Y
(f −1 F)(V ) = HomX (V, F et ) = HomY (V, F et ×X Y ).
The étalé space of f −1 F is F et ×X Y → Y , and the stalks are (f −1 F)y = Ff (x) ; hence f −1
is an exact functor, it preserves inductive limits and f −1 (ZU ) = Zf −1 U .
If s ∈ F(U ), then f ∗ s = s ◦ f ∈ (f −1 F)(f −1 U ), and we have an inverse image of sections
f ∗ : F(U ) −→ (f −1 F)(f −1 U ).
Now, if G is a sheaf on Y , any morphism of sheaves f −1 F → G induces a morphisms

F(U ) → (f ∗ F)(f −1 U ) → G(f −1 U ), so defining a morphism of sheaves F → f∗ G,
Adjunction Formula: Hom(f −1 F, G) = Hom(F, f∗ G).
Proof: Since F admits (p. 328) a presentation ⊕j ZUj → ⊕i ZUi → F → 0 and both functors are
left exact, we may assume that F = ZU . Now,
Hom(f −1 ZU , G) = Hom(Zf −1 U , G) = G(f −1 U ) = (f∗ G)(U ) = Hom(ZU , f∗ G).
Note: In the case of a morphism of ringed spaces (f, φ) : Y → X, the morphism of sheaves of
rings φ : OX → f∗ OY corresponds to a morphism of sheaves of rings f −1 OX → OY , and the
inverse image of OX -modules is defined to be
f ∗ M = (f −1 M) ⊗f −1 OX OY ,
so that the adjunction formula HomOY (f ∗ M, N ) = HomOX (M, f∗ N ) also holds.

In this course topological spaces are assumed to be ringed with the constant sheaf Z, and
we put f ∗ F or f −1 F indistinctively (but on schemes f ∗ M =
6 f −1 M).
Definition: The sequence 0 → f ∗ F → f ∗ (C • F) is exact, since the functor f ∗ is exact. Hence

the inverse image of sections induces an inverse image of cohomology classes (compatible with
morphisms of sheaves in an obvious sense)
f∗
f ∗ : H n (X, F) = H n Γ(X, C • F) −−−→ H n Γ(Y, f ∗ C • F) −→ H n (Y, f ∗ F)

Lemma: If 0 → F → R• is a resolution, the following square is commutative,
f∗
H n [Γ(X, R• )] / H n [Γ(Y, f ∗ R• )]
DR DR
f∗
H n (X, F) / H n (Y, f ∗ F)
Proof: The following diagram is commutative (p. 335)
H n [Γ(X, C • F)]
∼
/ H n [Γ(X, C • R• )] o H n [Γ(X, R• )]
f∗ f∗ f∗

H n [Γ(Y, f ∗ C • F)] / H n [Γ(Y, φ∗ C • R• )] o H n [Γ(Y, φ∗ R• )]
Corollary: If f : Y → X is a smooth map, the morphism f ∗ : H p (X, R) → H p (Y, R) is defined

by the inverse image of differential forms, f ∗ [ ωp ] = [ f ∗ ωp ].
g f
Theorem: If Z → → X are continuous maps, (f g)∗ = g ∗ f ∗ .
− Y −
Proof: Since f ∗ (C • F) is a resolution of f ∗ F, the following diagram commutes

f∗ g∗
H n [Γ(X, C • F)] −−−→ H n [Γ(Y, f ∗ C • F)] −−→ H n [Γ(Z, (f g)∗ C • F)]
k ↓DR ↓DR
f∗ g∗
H n (X, F) −−−→ H n (Y, f ∗ F) −−→ H n (Z, g ∗ (f ∗ F))
Proposition: The inverse image preserves the connecting, f ∗ (δcn ) = δ(f ∗ cn ).
Proof: If 0 → F 0 → F → F 00 → 0 is an exact sequence of sheaves, just take cohomology in the

following commutative diagram with exact rows,
0 / Γ(X, C • F 0 ) / Γ(X, C • F) / Γ(X, C • F 00 ) /0
f∗ f∗ f∗

0 / Γ(Y, C • (f ∗ C • F 0 )) / Γ(Y, C • (f ∗ C • F)) / Γ(Y, C • (f ∗ C • F 00 )) /0
O O O
0 / Γ(Y, C • f ∗ F 0 ) / Γ(Y, C • f ∗ F) / Γ(Y, C • f ∗ F 00 ) /0
Lemma: If a sheaf of rings O on X admits partitions of unity, then the restriction O|Y to any
closed set Y also admits partitions of unity.
Proof: If {Vi } is an open cover of Y , and {f0 , fi } is a partition of unity of O subordinated to

the cover {U0 = X − Y, Ui }, where Vi = Ui ∩ Y , then {fi |Y } is a partition of unity of O|Y
subordinated to {Vi }.
12.3. INVERSE IMAGE 339
n n
f : Y• →X is a continuous map and F is a sheaf on Y , the higher direct
Definition: If
•
images
R f∗ F = H f∗ (C F) are the cohomology sheaves of the complex of sheaves f∗ (C F).
That is to say, Rn f∗ F is the associated sheaf of the presheaf
U H n (f −1 U, F).
When Rn f∗ F = 0, n ≥ 1, we have that f∗ (C • F) is a resolution of f∗ F, and
H n (Y, F) = H n (X, f∗ F).
Cohomology of the Fibre: Let f : Y → X be a proper morphism between σ-compact spaces.

If F is a sheaf on Y , and x ∈ X, we have natural isomorphisms
(Rn f∗ F)x = H n (f −1 x, F).
Proof: Put C = C p F. The restriction of sections defines an isomorphism
(f∗ C)x = lim

−→
Γ(f −1 U, C) −−
∼
→ Γ(f −1 x, C).
x∈U
It is injective: if the support of a section s ∈ Γ(f −1 U, C) does not intersect the (compact)
fibre of x, then f (supp s) is closed in U , and s vanishes on f −1 (U − f (supp s)).
It is surjective: if s̄ ∈ Γ(f −1 x, C), then for any point y ∈ f −1 x there is a neighborhood
Ui of compact closure and si ∈ Γ(Ui , C) extending s̄, and a finite number U1 , . . . , Ur cover the
fibre f −1 (x). If f0 , f1 , . . . , fr is a partition of unity of C 0 ZY subordinated to the open cover
{U0 = Y − f −1 (x), U1 , . . . , Ur }, then f1 s1 + . . . + fr sr ∈ Γ(Y, C) extends s̄.
Now, by the lemma, (C • F)|f −1 x is an acyclic resolution of F|f −1 x , and
(Rn f∗ F)x = H n (f∗ C • F)x = H n Γ(f −1 x, C • F) = H n (f −1 x, F).

Base Change Theorem: Let us consider a fibred product of continuous maps between σ-
compact spaces,
φ̄
X ×S T /X
f¯ f
φ
T /S
If f is proper, then we have isomorphisms φ∗ (Rn f∗ F) = Rn f¯∗ (φ̄∗ F).
Proof: The inverse image φ̄∗ induces morphisms
H n (f −1 U, F) −→ H n (φ̄−1 f −1 U, φ̄∗ F) = H n (f¯−1 φ−1 U, φ̄∗ F) −→ (Rn f¯∗ φ̄∗ F)(φ−1 U )
defining a morphism of sheaves Rn f∗ F → φ∗ (Rn f¯∗ φ̄∗ F), and the corresponding morphism
φ∗ (Rn f∗ F) → Rn f¯∗ (φ̄∗ F) is an isomorphism. In fact, if s = f (t), then
φ̄ ∗
(φ∗ Rn f∗ F)t = (Rn f∗ F)s = H n (f −1 s, F) −−−→ H n (f¯−1 t, φ̄∗ F) = (Rn f¯∗ φ̄∗ F)t
is an isomorphism because φ̄ : f¯−1 (t) → f −1 (s) is a homeomorphism.
Corollary: π ∗ : H n (X, G) −−
∼
→ H n (X × [0, 1], G), when X is σ-compact.
Proof: Let C • be the Godement resolution of the constant sheaf G on X × [0, 1].
The cohomology of the fibre shows that π∗ C • is a (flasque) resolution of π∗ G = G; hence it

calculates the inverse image. We conclude since we have a canonical morphism π ∗ π∗ C • → C •
and the following composition is the identity,
π∗
Γ(X, π∗ C • ) −−−→ Γ(X × [0, 1], π ∗ π∗ C • ) −→ Γ(X × [0, 1], C • ).
Corollary: If two continuous maps φ, ψ : X → Y between σ-compact spaces are homotopic, then
φ∗ = ψ ∗ : H n (Y, G) → H n (X, G). Therefore, if φ : X → Y is a homotopical equivalence, then
φ∗ : H n (Y, G) → H n (X, G) is an isomorphism.
Proof: If H : X × [0, 1] → Y is a continuous map and φ = Hi0 , ψ = Hi1 , where it (x) = (x, t),
then φ∗ = i∗0 H ∗ , ψ ∗ = i∗1 H ∗ . Now i∗0 = i∗1 , both being the inverse of π ∗ . q.e.d.
1. Cohomology of Affine Spaces: H p (Rn , G) = 0, p ≥ 1.
(
G p=n
2. If X is a n-dimensional topological manifold, Hxp (X, G) =
0 p 6= n
Excision and the local cohomology sequence, since Rn − x is homotopic to Sn−1 .
3. The boundary of a manifold with boundary (any point has a neighborhood homeomorphic
to an open set in [0, ∞) × Rn−1 ) is a topological concept.
If x is in the boundary of P = {(x1 , . . . , xn ) ∈ Rn : x1 ≥ 0}, then P and P −x are contractible,
and Hxp (P, G) = 0, p ≥ 0, by the local cohomology sequence.
4. If the spheres Sn and Sm are homotopically equivalent, then n = m.
5. The inclusion i : Sn−1 ,→ Bn of the sphere in the ball has no continuous retract r.
Otherwise r∗ would be a section of i∗ : 0 = H n−1 (Bn , Z) → H n−1 (Sn−1 , Z) = Z.
6. Any continuous map φ : Bn → Bn has some fixed point (Brouwer’s Theorem).

Otherwise we have a continuous retract r : Bn → Sn−1 , where r(x) is the intersection point
of Sn−1 and the half-line with origin at φ(x) passing through x.
7. Any continuous tangent vector field to Sn vanishes at a a point when n is even.

If a continuous vector field D on Bn does not vanish at any point, then it points outside at
some point x ∈ Sn−1 .
We may repeat the proofs of the differentiable case (p. 264).
Finiteness Theorem: Let A be a noetherian ring and F a sheaf of A-modules on a separated
locally compact space X. If any point has a base of neighborhoods U in X such that the modules
H p (U, F) are finitely generated, then for any compact set K contained in the interior of a
compact set L we have that the images of the restriction morphisms H p (L, F) → H p (K, F) also
are finitely generated A-modules.
Proof: By induction on p. We fix L and we consider the family of all compact sets K with some
compact neighborhood K̄ contained in the interior of L such that H p (L, F) → H p (K̄, F) has
finitely generated image (so that any compact set contained in K is also in the family).
By hypothesis any point in the interior of L has a compact neighborhood in the family; hence
we only have to show that the family is stable under finite unions.
If K1 , K2 are in the family, by definition Ki has a compact neighborhood K̄i contained in
the interior of L such that the images of H p (L, F) → H p (K̄i , F)are finitely generated.
12.4. CUP PRODUCT 341
We may choose a compact neighborhood Ki0 of Ki contained in the interior of K̄i .

We have a commutative diagram
H p (L, F) / H p (L, F) ⊕ H p (L, F)
ρ

H p−1 (K̄1 ∩ K̄2 , F) / H p (K̄1 ∪ K̄2 , F) / H p (K̄1 , F) ⊕ H p (K̄2 , F)
γ

H p−1 (K10 ∩ K20 , F) / H p (K 0 ∪ K 0 , F)
1 2
where the central row is exact by Mayer-Vietoris, and the image of ρ is finitely generated, and
the image of γ by induction. Now it is easy to see that the image of H p (L, F) → H p (K10 ∪K20 , F)
is finitely generated, so that K1 ∪ K2 is in the considered family.
Corollary: If F is a locally constant sheaf of stalk A on a compact manifold with boundary X,

then the A-modules H p (X, F) are finitely generated. In particular so are H p (X, A).
Proof: The A-modules H p (Rd , A) and H p (Rd , A) are finitely generated.

Hence, any point x ∈ X has a base of neighborhoods U such that the A-modules H p (U, F)
are finitely generated. Now just put K = L = X in the above theorem.
12.4 Cup Product

Definition: If (K • , d), (L• , d) are complexes of A-modules (or sheaves of A-modules), we have
a bicomplex K • ⊗A L• = {K p ⊗A Lq } with the differentials d1 = d ⊗ 1, d2 = 1 ⊗ d, and the
∼
natural isomorphisms K p ⊗A Lq − → Lq ⊗A K p , with a factor (−1)pq , define an isomorphism of
•
complexes K ⊗A L − • ∼ •
→ L ⊗A K , •
(d ⊗ 1 + (−1)q 1 ⊗ d)((−1)pq bq ⊗ ap ) = (−1)(p+1)q bq ⊗ dap + (−1)p(q+1) (−1)p dbq ⊗ ap .
The tensor product of two cycles is a cycle of K • ⊗A L• , so that we have morphisms

⊗
H p (K • ) ⊗A H q (L• ) −−→ H p+q (K • ⊗A L• ).
Definition: If M is a sheaf of A-modules on a topological space X, the exact sequence
0 −→ M −→ C 0 M −→ M1 −→ 0
splits on stalks, a retract (C 0 M)x → Mx transforming any germ of discontinuous section into
the value at x ∈ X. Hence, for any sheaf of A-modules N , the sequence
0 −→ M ⊗A N −→ C 0 M ⊗A N −→ C 1 M ⊗A N −→ C 2 M ⊗A N −→ . . .
is exact, M ⊗A N −→∼
C • M ⊗A N , so that hence M ⊗A C q N −
∼
→ C • M ⊗A C q N .
•
By the bicomplex theorem M ⊗A C N − ∼ • •
→ C M ⊗A C N ; hence M ⊗A N − ∼
→ C • M ⊗A C • N ,
and we get a cup product ∪ : H p (X, M) ⊗A H q (X, N ) → H p+q (X, M ⊗A N ),
⊗ DR
H p Γ(X, C • M) ⊗ H q Γ(X, C • N ) −−→ H p+q Γ(X, C • M ⊗ C • N ) −−−→ H p+q (X, M ⊗ N ).

Proposition: If 0 → M → R• and 0 → N → S • are resolutions such that R• ⊗A S • is a

resolution of M ⊗A N , then the following square is commutative
⊗ / H p+q [Γ(X, R• ⊗A S • )]
H p [Γ(X, R• )] ⊗A H q [Γ(X, S • )]
DR DR

H p (X, M) ⊗A H q (X, N )
∪ / H p+q (X, M ⊗A N )
Proof: By the bicomplex theorem we have quasi-isomorphisms

∼
M ⊗A N −−→ R• ⊗A S • −−
∼
→ C • R• ⊗A S • −−
∼
→ C • R• ⊗A C • S •
and the following commutative diagram let us conclude (p. 335)
H p [Γ(X, C • M)] ⊗ H q [Γ(X, C • N )] −→ H p+q [Γ(X, C • M ⊗ C • N )]

k ↓
H p [Γ(X, C • R• ) ⊗ H q [Γ(X, C • S • )] −→ H p+q [Γ(X, C • R• ⊗ C • S • )]
↑ ↑
H p [Γ(X, R• )] ⊗ H q [Γ(X, S • )] −→ H p+q [Γ(X, R• ⊗ S • )]
Corollary: (cp ∪ cq ) ∪ cr = cp ∪ (cq ∪ cr ).

cp ∪ cq = (−1)pq cq ∪ cp .
f ∗ (cp ∪ cq ) = (f ∗ cp ) ∪ (f ∗ cq ).
Corollary: If X is a smooth manifold, the cup product of HDR • (X) = H • (X, R) is defined by
the exterior product of differential forms, [ωp ] ∪ [ωq ] = [ωp ∧ ωq ].
12.4.1 Universal Coefficients and Künneth’s Theorem

Now A will denote a principal ideal domain, so that torsion free A-modules are flat.
Definition: A sheaf of A-modules M Q

is flat if so is any stalk Mx , or equivalently any module
M(U ) since M(U ) is a submodule of x∈U Mx , and Mx = lim −→
M(U ).
Lemma: If M is a sheaf of A-modules on a σ-compact space X, and P is projective,
Hcn (X, M) ⊗A P = Hcn (X, M ⊗A P ).
Proof: It holds when P is free (p. 332) and P is a direct summand of a free module.
Universal Coefficients Formula: If M is a flat sheaf of A-modules on a σ-compact space X,

for any A-module N we have exact sequences
0 −→ Hcn (X, M) ⊗A N −→ Hcn (X, M ⊗A N ) −→ TorA n+1

1 (Hc (X, M), N ) −→ 0,
and if moreover N is finitely generated, we have exact sequences
0 −→ H n (X, M) ⊗A N −→ H n (X, M ⊗A N ) −→ TorA

1 (H
n+1
(X, M), N ) −→ 0
Proof: Let 0 → P1 → P0 → N → 0 be a projective resolution (p. 334). The kernel and cokernel
of φn : Hcn (X, M) ⊗A P1 → Hcn (X, M) ⊗A P0 are Tor1 (Hcn (X, M), N ), Hcn (X, M) ⊗A N .
12.4. CUP PRODUCT 343
Since M is flat, the sequence 0 → M ⊗A P1 → M ⊗A P0 → M ⊗A N → 0 is exact, and we

obtain exact sequences
φ φ
Hcn (X, M)⊗P1 −
→ Hcn (X, M)⊗P0 → Hcn (X, M⊗N ) → Hcn+1 (X, M)⊗P1 −
→ Hcn+1 (X, M)⊗P0
0 −→ Coker φn −→ Hcn (X, M ⊗A N ) −→ Ker φn+1 −→ 0 .
If N is finite, it admits a resolution 0 → Am → An → N → 0. When P = Ar , the equality

H n (X, M ⊗A P ) = H n (X, M) ⊗A P is clear and we repeat the above proof.
Corollary: 0 −→ Hcn (X, Z) ⊗Z G −→ Hcn (X, G) −→ TorZ1 (Hcn+1 (X, Z), G) −→ 0 is exact.
Projection Formula: Let f : X → Y be a proper morphism of σ-compact spaces. If M is a

flat sheaf of A-modules on X, for any sheaf of A-modules N on Y we have isomorphisms
(Rn f∗ M) ⊗A N = Rn f∗ (M ⊗A f ∗ N ).
Proof: The morphism F degree(Rn f∗ M) ⊗A N → Rn f∗ (M ⊗A φ∗ N ) given by the universal

coefficients formula is an isomorphism according to the cohomology of the fibre,
((Rn f∗ M) ⊗A N )y = (Rn f∗ M)y ⊗A Ny = H n (f −1 y, M) ⊗A Ny

= H n (f −1 y, M ⊗A Ny ) = (Rn f∗ (M ⊗A f ∗ N ))y .
Lemma: If K • , L• are complexes of A-modules and L• is flat, we have exact sequences
H p (K • ) ⊗A H q (L• ) → H n (K • ⊗A L• ) → TorA p • q •
L L
0→ 1 (H (K ), H (L )) → 0
p+q=n p+q=n+1
Proof: Since L• is flat, cycles Zq and boundaries Bq are torsion free, and
0 −→ B q −→ Z q −→ H q (L• ) −→ 0
is a flat resolution of H q (L• ); hence the cokernel and kernel of φpq : H p (K • )⊗B q → H p (K • )⊗Z q
are H p (K • ) ⊗ H q (L• ) and Tor1 (H p (K • ), H q (L• )).
On the other hand, we have an exact sequence
d
0 −→ Z • −→ L• −−→ B • [1] −→ 0
where the differentials of Z • and B • [1] are null, and the connecting
δ : B q = H q−1 (B • [1]) −→ H q (Z • ) = Z q
is just the inclusion. Since the modules B q are flat, we have an exact sequence
1⊗d
0 −→ K • ⊗A Z • −→ K • ⊗A L• −−−→ K • ⊗A (B • [1]) −→ 0
inducing exact sequences

δ δn+1
. . . −−n→ H n (K • ⊗ Z • ) −→ H n (K • ⊗ L• ) −→ H n (K • ⊗ (B • [1])) −−−−→ H n+1 (K • ⊗ Z • ) −→ . . .
0 −→ Coker δn −→ H n (K • ⊗A L• ) −→ Ker δn+1 −→ 0
Moreover, since Z q and B q are flat modules, we have that
H n (K • ⊗A Z • ) =
L p • L p •
H (K ⊗A Z q ) = H (K ) ⊗A Z q
p+q=n p+q=n
• • • •
n−1 p q
H p (K • ) ⊗A B q
L L
H (K ⊗A B [1]) = H (K ⊗A B [1] ) =
p+q=n−1 p+q=n
L
so that δn = φpq , and we conclude,
p+q=n
H p (K • ) ⊗A H q (L• ),
L L
Coker δn = Coker φpq =
p+q=n p+q=n
• •
TorA p q
L L
Ker δn+1 = Ker φpq = 1 (H (K ), H (L )).
p+q=n+1 p+q=n+1
Note: On σ-compact spaces we also consider the direct image with proper supports
(f! F)(U ) = {s ∈ F(f −1 U ) : the support of s is proper over U }
so that Γc (X, f! F) = Γc (Y, F), and we put Rn f! F = Hn f! (C • F) .

If Rn f! F = 0, n ≥ 1, then f! (C • F) is a Γc -acyclic resolution of f! F, because the sheaves

f! (C p F) are C 0 Z-modules, so that Hcn (Y, F) = Hcn (X, f! F).
The Cohomology of the Fibre, Base Change and Projection formulas, and their proofs,
remain valid for (non proper) continuous maps between σ-compact spaces if we replace the
direct images Rn f∗ by Rn f! , and the cohomology groups H n by Hcn .
If a sheaf M is flat, so is C 0 M since (C 0 M)(U ) = x∈U Mx is torsion free.

Q
Since the exact sequence 0 → M → C 0 M → M1 → 0 splits on stalks, the sheaf M1 is flat,

and we conclude that the sheaves C p M are also flat.
Künneth’s Theorem: Let X, Y be σ-compact spaces and p1 : X × Y → X, p2 : X × Y → Y

the natural maps. If M, N are flat sheaves of A-modules on X, Y , we have exact sequences
Hcp (X, M) ⊗ Hcq (Y, N ) → Hcn (X ×Y, p∗1 M ⊗ p∗2 N ) → Tor1 (Hcp (X, M), Hcq (Y, N )) → 0
L L
0→
p+q=n p+q=n+1
Proof: According to the former lemma, it is enough to prove the following two statements.
1. p∗1 ⊗ p∗2 : Γc (X, C • M) ⊗A Γc (Y, C • N ) = Γc (X × Y, p∗1 C • M ⊗A p∗2 C • N ).

p1
X ×Y /X
p2 q1
q2
Y /•
Γc (X × Y, p∗1 C i M ⊗ p∗2 C j N ) = q1! p1! (p∗1 C i ⊗ p∗2 C j ) = q1! (C i ⊗ p1! (p∗2 C j ))
= q1! (C i ⊗ q1∗ q2! (C j )) = (q1! C i ) ⊗ (q2! C j )
= Γc (X, C i M) ⊗ Γc (Y, C j N ).
2. p∗1 C • M ⊗A p∗2 C • N is a Γc -acyclic resolution of p∗1 M ⊗A p∗2 N .

0 → p∗1 M → p∗1 C • M and 0 → p∗2 N → p∗2 C • N split on stalks; hence p∗1 C • M ⊗ p∗2 C • N is a
resolution of p∗1 M ⊗ p∗2 N , and the sheaves p∗1 C i M ⊗ p∗2 C j N are Γc -acyclic,
Rn p1! (p∗1 C i ⊗ p∗2 C j ) = C i ⊗ Rn p1! (p∗2 C j ) = C i ⊗ q1∗ (Rn q2! C j ) = 0, n ≥ 1,
and the sheaves p1! (p∗1 C i ⊗ p∗2 C j ) = C i ⊗ q1∗ (q2! C j ) areΓc -acyclic,
Hcn (X, C i ⊗ q1∗ (q2! C j )) = Rn q1! (C i ⊗ q1∗ (q2! C j )) = (Rn q1! C i ) ⊗ (q2! C j ) = 0, n ≥ 1.
Example: No continuous map µ : Sn × Sn → Sn defines a group structure when n is even.

12.5. LOCALLY TRIVIAL STRUCTURES 345
In fact, the cohomology ring of the sphere is H • (Sn , Z) = Z[tn ]/(t2n ), so that
H • (Sn × Sm , Z) = H • (Sn , Z) ⊗Z H • (Sm , Z) = Z[xn , ym ]/(x2n , ym

2
),
where p∗1 (tn ) = tn ⊗ 1 = xn , p∗2 (tm ) = 1 ⊗ tm = ym . Put µ∗ (tn ) = axn + byn .

If e is the identity of Sn , the continuous map j : Sn → Sn × Sn , j(p) = (p, e), fulfills that µj
and p1 j are the identity, and that p2 j is constant,
tn = j ∗ µ∗ (tn ) = j ∗ (axn + byn ) = j ∗ (ap∗1 (tn ) + bp∗2 (tn )) = atn + 0,
and we obtain a = 1. Analogously b = 1, and we obtain a contradiction if n is even
0 = µ∗ (t2n ) = (µ∗ tn )2 = (xn + yn )2 = xn yn + yn xn = 2xn yn .
12.5 Locally Trivial Structures

Lemma: If G is an abelian group, the sheaf of automorphisms G of the trivial principal covering
X × G → X is the constant sheaf G.
Proof: Any element h ∈ G defines an automorphism τh (y) = hy since G is abelian,
τh (gy) = hgy = ghy = gτh (y),
and this injective morphism of sheaves G → G is surjective.

In fact, if τ : U × G → U × G is an automorphism and τ (x, 1) = (x, h(x)), then τ = τh ,
τh (x, g) = (x, h(x)g) = g(x, h(x)) = gτ (x, 1) = τ (x, g).

1 Principal coverings of X
Theorem: H (X, G) = (up to isomorphisms).
of abelian group G
Proof: Let us fix an open cover R = {Ui }i∈I of X.

∼
If a principal covering P → X is trivial on it, φi : PUi −
→ Ui × G, we have automorphisms
−1 ∼
gij = φi φj : Uij × G −→ Uij × G, Uij = Ui ∩ Uj , and on any open set Uijk = Ui ∩ Uj ∩ Uk ,
gij gjk = gik .
Families {gij ∈ G(Uij )}i,j∈I satisfying such condition are named construction data, since
they let us reconstruct the principal covering as a quotient
`
P ' i∈I Ui × G / ≡
by the following equivalence relation: (x, gi ) ≡ (x, gj ) when x ∈ Uij , gi = gij (x)gj .
The construction datum depends on the trivialization φi . If we fix other isomorphisms
∼
φ̄i : PUi −
→ G × Ui , we obtain another construction datum {ḡij }.
Now, gi = φ̄i φ−1 ∼
i : Ui × G −
→ Ui × G is an automorphism, so that gi ∈ G(Ui ) and
ḡij = gi gij gj−1 .
Therefore, if two construction data {gij } and {ḡij } are said to be equivalent when there are
sections gi ∈ G(Ui ) such that ḡij = gi gij gj−1 , we have a natural bijection

Principal coverings {Construction data}
=
of X trivial over R Data equivalence
Now let us consider the Godement resolution of G,
0 −→ G −→ C 0 G −→ G1 −→ 0
0 −→ G1 −→ C 1 G −→ G2 −→ 0
Γ(X, G1 )
H 1 (X, G) =
Γ(X, C 0 G)
If H 1 (X, G)R denotes the subgroup of all cohomology classes in H 1 (X, G) vanishing on the
cover R, and Γ(X, G1 )R denotes the subgroup of all global sections of G1 coming from sections
of C 0 G on the cover R, we have
Γ(X, G1 )R
H 1 (X, G)R = ·
Γ(X, C 0 G)
If f 0 ∈ Γ(X, G1 )R , there are fi ∈ Γ(Ui , C 0 G) such that f 0 |Ui = fi ; hence fi = gij fj on Uij
for some construction datum gij ∈ G(Uij ). If we consider other sections f¯i representing f 0 , and
defining a datum {ḡij }, then f¯i = gi fi , where gi ∈ G(Ui ), and
ḡij = f¯i /f¯j = gi fi /gj fj = gi gij gj−1 .

{Construction data}
Γ(X, G1 )R −→
Data equivalence
This group morphism is not injective, the kernel being the image of Γ(X, C 0 G) since condition
gij = 1 signifies that sections fi coincide on intersections Uij .
Finally let us see that this morphism is surjective. Given a construction datum {gij }, we fix
at any point x ∈ X an index kx ∈ I such that x ∈ Ukx and we put
fi (x) = gikx (x), fi ∈ Γ(Ui , C 0 G),
so that on Uij we have fi = gikx = gij gjkx = gij fj .

These sections of C 0 G coincide as sections of G1 , so that they define a global section f 0 of
G1 inducing the construction datum {gij }. In summary,

{Construction data} Principal coverings
H 1 (X, G)R = =
Data equivalencia of X trivial over R
and we conclude since any cohomology class vanishes on some open cover and any principal
covering is trivial over some open cover.
Note: This argument shows that any kind of locally trivial structure is classified, modulo iso-
morphisms, by the cohomology group H 1 (X, G), where G is the sheaf of automorphisms of the
corresponding trivial structure, provided that G is abelian.
Hurewicz’s Theorem: If X is a connected and locally simply connected space, then for any
abelian group G we have a group isomorphism
H 1 (X, G) = Homgr (π1 (X, x), G).
Proof: Principal G-coverings are classified (p. 244) by Homgr (π1 (X, x), G), and we must show
that the bijection H 1 (X, G) = Homgr (π1 (X, x), G) is in fact a group morphism.
Now, π1 = π1 (X, x) is the group of automorphisms of the universal covering X̃ → X, and
any group morphism φ : π1 → G correspond to the principal covering
P = (G × X̃)/π1 −→ X̃/π1 = X,
where the action of π1 on P is σ(g, x̃) = (g · φ(σ −1 ), σx̃).

Let us fix an open cover of X by simply connected open sets Ui .
We have π1 -isomorphisms X̃Ui ' π1 × Ui inducing trivializations
∼
φi : PUi −−→ G × Ui , φi [(g, σ, x)] = (g · φ(σ), x)
and on Uij the isomorphism (π1 × Uj )Uij ' X̃Uij ' (π1 × Ui )Uij transforms (1, x) into (σij (x), x).
Hence a construction datum of P is precisely φ(σij ) because
(φi φ−1
j )(g, x) = φj [(g, σij , x)] = (gφ(σij ), x).
Now it is clear that the product of morphisms corresponds to the data product.

Z
 n = 0, 2
n
1. H (τg , Z) = Z2g n = 1 where τg is the connected sum of g toruses.

0 n>2

H 1 (τg , Z) = Hom(π1 (τg )ab , Z) = Hom(Z2g , Z) = Z2g .

The complement of an open disc is homotopic to 2g circles identified at a point, and the
closed subspace exact sequence let us conclude,
0 −→ Z2g −→ Z2g −→ Z −→ H 2 (τg , Z) −→ 0

F2
 n = 0, 2
2. H (πg , F2 ) = Fg2
n
n=1 where πg is the connected sum of g projective planes.

0 n>2

H 1 (πg , F2 ) = Hom(π1 (πg )ab , F2 ) = Hom(Zg−1 ⊕ F2 , F2 ) = Fg2 .

The complement of an open disc is homotopic to g circles identified at a point, and the closed
subspace exact sequence let us conclude,
0 −→ Fg2 −→ Fg2 −→ F2 −→ H 2 (πg , F2 ) −→ 0
12.5.1 Vector Bundles

Definitions: A real vector space over a topological space X is a continuous map π : E → X
+ .
endowed with continuous operations E ×X E −−→ E and R × E − → E satisfying the vector space
axioms, including the existence of a continuous zero section 0 : X → E (any fibre Ex = π −1 (x)
inherits a structure of real vector space). A morphism between two vector spaces E → X and
E 0 → X over X is a continuous map φ : E → E 0 over X such that all maps φx : Ex → Ex0 are
linear. The trivial vector space of rank n over X is Rn × X → X, and a real vector bundle
of rank n is a locally trivial vector space E → X of rank n; i.e., locally E|U ' Rn × U .
It is a line bundle when n = 1.
Analogously we define complex vector bundles replacing R by C, and smooth vector bundles
replacing topological spaces and continuous maps by smooth manifolds and smooth maps.
1. If X is a smooth manifold, the tangent and cotangent bundles are smooth vector bundles.
2. The sheaf of continuous sections of a real vector bundle of rank n is a locally free sheaf of
rank n over the sheaf of real continuous functions CX , and so we obtain an equivalence of
` inverse functor transforming any locally free CX -module E into the vector
categories, the
bundle E = x∈X (Ex /mx Ex ) → X (with the obvious topology).
Analogously complex vector bundles correspond to locally free modules over the sheaf of
complex continuous functions, and smooth vector bundles correspond to locally free modules
over the sheaf of smooth functions.
3. The graph ξ ⊂ Pn × E of the incidence (i.e., (p, e) ∈ ξ when e ∈ E is in the line represented
by p ∈ Pn ) defines a line bundle π : ξ → Pn , the tautological bundle of Pn . It is not trivial
since any continuous section s : Pn → ξ vanishes at some point: otherwise, considering a
scalar product on E and dividing s(p) by the module, we obtain a continuous section of the
covering Sn → Pn , absurd since the sphere Sn is connected.
Lemma: The sheaf of automorphisms of a sheaf of rings O is the sheaf O∗ of invertible sections.
Proof: O(U ) = HomO|U (O|U , O|U ).
Classification of Line Sheaves: H 1 (X, O∗ ) = [ Line O-modules ] (up to isomorphisms).

Locally constant
Corollary: H 1 (X, F2 ) = .
sheaves of stalk Z
Corollary: If X is a σ-compact space (or a smooth manifold and the line bundles are C ∞ ),

1 Real line 2 Complex line
H (X, F2 ) = , H (X, Z) =
bundles over X bundles over X
Proof: The sheaf of real continuous (resp. smooth) functions O is Γ-acyclic, and the following
exact sequence shows that the natural morphism H 1 (X, O∗ ) → H 1 (X, F2 ) is an isomorphism,
ef
0 −→ O −−→ O∗ −→ Z/2Z −→ 0
The sheaf of complex continuous (resp. smooth) functions O is also Γ-acyclic, and we have
an exact sequence
2πi ef
0 −→ Z −−−→ O −−→ O∗ −→ 0
Hence (p.336) we have a canonical isomorphism H 1 (X, O∗ ) = H 2 (X, Z).
Definition: If X is σ-compact, real line bundles L → X are classified, up to isomorphisms, by

the obstruction class δ(L) ∈ H 1 (X, F2 ), and complex line bundles by δ(L) ∈ H 2 (X, Z).
Proposition: If φ : Y → X is a continuous map, δ(φ∗ L) = φ∗ (δ(L)).
Proof: If {gij } is a construction datum of L on an open cover {Ui }, then {gij ◦φ} is a construction
datum of φ∗ L on the cover {φ−1 Ui }; hence if c ∈ H 1 (X, OX ∗ ) corresponds to L, then φ∗ L
corresponds to the image of c by the morphism
∗ φ∗
H 1 (X, OX ) −−−→ H 1 (Y, φ∗ OX
∗
) −→ H 1 (Y, OY∗ ).
We conclude by the compatibility of φ∗ with the morphism H 1 (X, O∗ ) → H 1 (X, F2 ) in the

real case, and with δ : H 1 (X, O∗ ) → H 2 (X, Z) in the complex case. q.e.d.
1. A smooth manifold X of dimension n is orientable if and only if the line sheaf ΩnX is trivial.
Hence, if H 1 (X, F2 ) = 0, then X is orientable.
2. If Y is a closed hypersurface of a smooth manifold X, then the sheaf of ideals pY (U ) = {f ∈

CX∞ (U ) : f |
U ∩Y = 0} is a line sheaf, and it is trivial if and only if Y admits a global equation
f = 0 with non null differential at any point of Y .
If H 1 (X, F2 ) = 0, any closed hypersurface Y of X is orientable and admits a global equation.
In fact, if f = 0 is a global equation of Y , then N = grad f does not vanish at any point of
Y , and an orientation [ωX ] of X defines an orientation [iN ωX ] of Y .
Definition: Let X be a smooth manifold, and π : E → X a smooth vector bundle. A smooth

section s : X → E is transversal to the null section s0 at a point p ∈ X when s(p) 6= 0 or

Ts(p) E = Ts(p) s0 (X) + Ts(p) s(X) ,
and s is transversal to S0 when so it is at any point of X, so that the zero set s−1 s0 (X) is

empty, or a smooth submanifold of codimension the rank of E.

Any trivialization E|U = U × Rr defines a linear topology on the vector space of smooth
sections Γ(U, E) = Γ(U, U × Rr ) = C ∞ (U )r , and we equip the vector space of smooth global
sections with the initial topology of the restriction maps Γ(X, E) → Γ(U, E), so that Γ(X, E) is
metrizable, complete and with a countable base of open sets (p. 115).
Theorem: The smooth sections transversal to 0 are dense in Γ(X, E).
Proof: When X is an open set in Rd and E = X × Rr , the condition of s : X → X × Rr ,

s(x) = x, f (x) , not being transversal to 0 means that 0 ∈ Rr is a critical value of f : X → Rr .
Given ε > 0, by Sard’s theorem there exists e ∈ Rr , kek < ε, such that e is not a critical value
0

of f , so that s (x) = x, f (x) + e is transversal to 0 and the theorem is proved.
In the general case, given a coordinate open set U ⊆ X, a trivialization E|U = U × Rr , and a
compact set K ⊂ U , we pick φ ∈ C ∞ (X) with compact support ⊂ U and φ = 1 on a neighborhood
of K. If s ∈ Γ(X, E), there is a sequence σn in Γ(U, E) such that lim(s|U + σn ) = s|U in Γ(U, E)
and s|U + σn is transversal to 0.
Then s + φσn ∈ Γ(X, E) is transversal to 0 on K and lim(s + φσn ) = s in Γ(X, E). Hence
the open set USK ⊂ Γ(X, E) of all global sections transversal to 0 at any point of K is dense.
Put X = K Tn where Kn is a compact set contained in a coordinate open set where E is
trivial. The set n UKn of all sections transversal to 0 is dense in Γ(X, E) by Baire’s theorem.
Lemma: Let s be a global section of E, and U ⊆ X a connected open set. If s has a unique zero
x on U , then for any point x0 ∈ U there is a section s0 ∈ Γ(X, E) such that s0 = s on X − U
and x0 is the unique zero of s0 on U .
Proof: When E = X ×Rr is trivial, so that s(x) = x, f (x) , consider a neighborhood x ∈ V ⊂ U

such that for any point x0 ∈ V there is a vector field D with compact support ⊂ V whose flow
{τt } fulfills τ1 (x0 ) = x. Then s0 (x) = x, f (τ1 x) coincides with s on X −U , because supp D ⊆ U ,
and x0 is the unique zero of s0 on U .
Hence, in general, the points x0 ∈ U where the lemma holds (resp. is false) is an open set.
Since U is connected and the lemma holds when x0 = x, we conclude.
Theorem: Let X be a connected non-compact smooth manifold of dimension n. Any smooth

real vector bundle E → X of rank n admits a smooth global section without zeros.
S o
Proof: Put X = n Kn , where Kn is compact, Kn ⊆K n+1 , and X −Kn has no relatively compact
connected component (p. 94). Pick a global section s transversal to 0, so that the zeros are
discrete (hence finite on each Kn ).
Inductively, we construct global sections sn with a discrete zero-set and no zero in Kn , such
that sn+1 = sn on Kn . Let x ∈ Kn+1 be a zero of sn , and V the connected component of
X − Kn containing x. Since V is not relatively compact, we may pick x0 ∈ V − Kn+1 , and we
may assume that sn (x0 ) 6= 0, because sn has discrete zeros.
Let U ⊆ V be a connected open set containing x0 , x and no other zero of sn .
By the lemma, there is a global section s0 coinciding with sn on X − U (hence on Kn ) and
with the same zeros as sn , except that x ∈ Kn+1 is replaced by x0 ∈ / Kn+1 .
Since the zeros of sn on Kn+1 are finite, iterating we obtain the required section sn+1 . Now
the section s such that s|Kn = sn is smooth (so it is on the interior of Kn ) and has no zero.
Corollary: Any connected non-compact smooth manifold admits a vector field without zeros.
Corollary: Any complex line bundle L on a connected non-compact smooth surface is trivial.
Proof: As a real vector bundle, L has rank 2; hence it has a global section without zeros.
Corollary: H 2 (X, Z) = 0 when X is a connected non-compact smooth surface.
12.6 Local Cohomology

Definition: Let Y ⊂ X be a closed set. The sheaf of sections of F with support in Y is ΓY F =
Hom(ZY , F), and HYp F = Rp ΓY (F) = Extp (ZY , F) is the p-th local cohomology sheaf ; i.e.,
(ΓY F)(U ) = ΓY ∩U (U, F), and HYp F is the associated sheaf of the presheaf U HYp ∩U (U, F),
(HYp F)(U ) = Hp [ΓY (C • F)] .
Lemma: If C is a Godement sheaf, then ΓY C is a flasque sheaf.

Q Q
Proof: If C(U ) = x∈U Mx , then ΓY ∩U (U, C) = x∈Y ∩U Mx .
Theorem: If Y is a closed submanifold 2 of codimension d in a topological manifold X, then the

local cohomology sheaves HYp Z are null, except the normal orientation sheaf TY /X = HYd Z,
which is a sheaf concentred on Y , locally constant of stalk Z.
Proof: The statement is local, and we may assume that X = Rm × Rd , Y = Rm × 0, and in

such a case we conclude by the local cohomology exact sequence, X − Y = Rm × (Rd − 0) being
homotopic to the sphere Sd−1 .
Corollary: HYp (X, Z) = H p−d (Y, TY /X ).

∼
Proof: If Z −−→ C • is the Godement resolution of the constant sheaf Z on X, we have quasi-
isomorphisms TY /X [−d] ←−∼
− Z • −−
∼
→ ΓY C • , where Z • is the complex
ΓY C 0 −→ ΓY C 1 −→ . . . −→ ΓY C d−1 −→ Z d −→ 0 −→ . . .
Since TY /X is concentred on Y , and ΓY C • is a complex of flasque sheaves,
H p−d (Y, TY /X ) = Hp (X, TY /X [−d ]) = Hp (X, Z • ) = Hp (X, ΓY C • ) = HYp (X, Z).

2
In the sense that any point of Y has an open neighborhood U in X such that the inclusion Y ∩ U → U is
homeomorphic to the obvious inclusion Rm × 0 → Rm × Rd .
12.6. LOCAL COHOMOLOGY 351
Definitions: A closed submanifold Y is normally orientable in X if TY /X is isomorphic to

∼
the sheaf ZY , and the normal orientations of Y in X are the isomorphisms ZY − → TY /X .
Once we fix a normal orientation, the morphism H p−d (Y, Z) − ∼
→ HYp (X, Z) → H p (X, Z) is
denoted by i∗ , and the local cohomology exact sequence gives the Gysin Exact Sequence
i δ
. . . −→ H p−d (Y, Z) −−∗→ H p (X, Z) −→ H p (X − Y, Z) −−→ H p−d+1 (Y, Z) −→ . . .
and the cohomology class of Y in X is [Y ] = i∗ (1) ∈ H d (X, Z).
Projection Formula: i∗ (i∗ (a) ∪ b) = a ∪ i∗ (b); a ∈ H • (X, Z), b ∈ H • (Y, Z).
Proof: That is to say, i∗ : H • (Y, Z) → H • (X, Z) is a morphism of H • (X, Z)-modules, where the
structure of module on H • (Y, Z) is induced by the ring morphism i∗ : H • (X, Z) → H • (Y, Z).
Now, up to a degree change, the direct image i∗ is a composition of morphisms of modules
(all are compatible with the cup product)
i∗
H• (Y, Z[−d]) ←−− H• (X, ZY [−d]) ←−
∼
− H• (X, Z • ) −−
∼
→ H• (X, ΓY C • ) −→ H• (X, C • ).
Finally, the degree change H • (Y, Z) −

→∼
H• (Y, Z[−d]) is an isomorphism of modules, when
we consider the left module structure, since the total differential of a bicomplex is not affected
when we change the second degree. q.e.d.
1. If H 1 (Y, F2 ) = 0, any line sheaf over the constant sheaf Z is trivial; hence Y is normally
orientable regardless the ambient manifold X and the codimension d.
2. If Y is connected and HYd (X, Z) 6= 0, then TY /X |Y has a non null section, hence it is trivial
and Y is normally orientable.
If a connected hypersurface Y admits a connected open neighborhood U such that U − Y
disconnects, then HY1 (X, Z) = HY1 (U, Z) 6= 0 by the local cohomology exact sequence, and Y
is normally orientable.
3. Any line sheaf over F2 is trivial since F∗2 = 1. If we use cohomology with coefficients in F2 ,
any closed submanifold is normally orientable, with a unique normal orientation, and the
cohomology class always is well defined.
4. If Y is a connected closed submanifold in Rn of codimension 1, then the complement U =

Rn − Y has two connected components.
Gysin’s exact sequence shows directly that dim H 0 (U, F2 ) = 2,
0 −→ F2 = H 0 (Rn , F2 ) −→ H 0 (U, F2 ) −→ H 0 (Y, F2 ) = F2 −→ H 1 (Rn , F2 ) = 0.
12.6.1 Topological Intersection Theory

Let Y1 , Y2 be closed submanifolds of codimension d1 , d2 of a topological manifold X. If Y1 ∩ Y2
is a submanifold of codimension d1 + d2 , on any open set U ⊆ X we have a group morphism
∪ : HYd11∩U (U, Z) ⊗Z HYd22∩U (U, Z) −→ HYd11∩Y

+d2
2 ∩U
(U, Z),
so that the cup product defines a morphism of sheaves
∪ : TY1 /X ⊗Z TY2 /X −→ TY1 ∩Y2 /X .

Definition: If we fix normal orientations ξ1 , ξ2 , ξ of Y1 , Y2 and Y1 ∩ Y2 at a point y ∈ Y1 ∩ Y2 ,

the intersection multiplicity of Y1 and Y2 at y is the integer number m such that
ξ1 ∪ ξ2 = mξ ∈ TY1 ∩Y2 /X,y .
It is locally constant, and the sign changes if we change some normal orientation.
Theorem: Let Y1 , Y2 be normally oriented closed submanifolds of codimension d1 , d2 . If Y1 ∩ Y2

is a normally oriented submanifold of codimension d1 + d2 with a finite number of connected
components C1 , . . . , Cr , and mi is the intersection multiplicity of Y1 with Y2 along Ci , then
[Y1 ] ∪ [Y2 ] = m1 [C1 ] + . . . + mr [Cr ].
Proof: HYd11∩Y
+d2
(X, Z) = ⊕j HCd1j+d2 (X, Z). Once we fix a normal orientation of Y1 ∩ Y2 , we have
P2
ξY1 ∪ ξY2 = j mj ξCi , and the following commutative square let us conclude,
HYd11 (X, Z) ⊗Z HYd22 (X, Z)

∪ / H d1 +d2 (X, Z) = ⊕j H d1 +d2 (X, Z)
Y1 ∩Y2 Cj

H d1 (X, Z) ⊗Z H d2 (X, Z)
∪ / H d1 +d2 (X, Z)
Definition: We say that Y1 and Y2 transversally intersect at a point y ∈ Y1 ∩ Y2 when there

∼
is an open neighborhood U of y in X and a homeomorphism φ : U − → Rd1 × Rd2 × Rm such that
d m d
φ(Y1 ∩ U ) = 0 × R × R and φ(Y2 ∩ U ) = R × 0 × R .
2 1 m
Theorem: The morphism TY1 /X ⊗Z TY2 /X → TY /X defined by the cup product is an isomorphism
when Y1 , Y2 transversally intersect; i.e., the intersection multiplicity is ±1.
Proof: Since the intersection multiplicity is a local topological invariant we may assume that
X = Sd1 × Sd2 × Sd3 , Y1 = p1 × Sd2 × Sd3 , Y2 = Sd1 × p2 × Sd3 .
By Künneth’s theorem we have an isomorphism of graded rings
Z[x1 ]/(x21 ) ⊗Z Z[x2 ]/(x22 ) ⊗Z Z[x23 ]/(x23 ) = H • (X, Z),
where [Y1 ] = x1 ⊗ 1 ⊗ 1 and [Y2 ] = 1 ⊗ x2 ⊗ 1.

Now, Y1 ∩ Y2 is a connected submanifold of codimension d1 + d2 and
x1 ⊗ x2 ⊗ 1 = [Y1 ] ∪ [Y2 ] = m[Y1 ∩ Y2 ]
is not divisible in H d1 +d2 (X, Z) by a natural number m 6= ±1. Hence m = ±1.
Corollary: The group H 2 (Pn,C , Z) is generated by the cohomology class x of any hyperplane
and we have an isomorphism of graded rings
H • (Pn,C , Z) = Z[x]/(xn+1 ), deg x = 2.
Proof: Pn−1 is normally orientable in Pn since HP2n−1 (Pn , Z) 6= 0, and by the Gysin exact sequence
we have isomorphisms
i∗ : H p−2 (Pn−1 , Z) −−
∼
→ H p (Pn , Z), p ≥ 1.
In particular x = i∗ (1) generates H 2 (Pn , Z). If y is a generator of H 2 (Pn−1 , Z), then i∗ (y p−1 )
generates H p (Pn , Z) by induction on n. The closed subspace exact sequence
i∗ δ
H 2 (Pn , Z) −−→ H 2 (Pn−1 , Z) −−→ H 3 (Cn , Z) = 0
12.7. DUALITY THEOREM 353
shows that y = i∗ (x), and by the projection formula H p (Pn , Z) is generated by
i∗ (y p−1 ) = i∗ ((i∗ x)p−1 ) = i∗ (1)xp−1 = xp .
Corollary: The group H 1 (Pn,R , F2 ) is generated by the cohomology class x of any hyperplane,
and we have an isomorphism of graded rings
H • (Pn,R , F2 ) = F2 [x]/(xn+1 ), deg x = 1.
Proof: The same argument of the complex case holds. q.e.d.
1. The inclusion i : Pm ,→ Pn , m < n, has no continuous retract.

The epimorphism of rings i∗ : A[x]/(xn ) → A[x]/(xm ) has no section (A = Z or F2 ).
2. Any diffeomorphism φ : P2n,C → P2n,C preserves orientations.

Since φ∗ (x) = ±x, then φ∗ (x2n ) = (±x)2n = x2n .
3. Pn,R can not be covered with n open sets homeomorphic to affine spaces.
If a topological space admits an open cover X = U1 ∪ . . . ∪ Un , Ui ' Rn , and we put
Yi = X − Ui , then the morphisms HYpi (X, A) → H p (X, A) are surjective for all p ≥ 1, and the
following commutative square shows that, in the ring H • (X, A), any product of n elements
of positive degree is null,
HY•1 (X, A) ⊗ . . . ⊗ HY•n (X, A)

∪ / H•
Y1 ∩...∩Yn (X, A) =0

H • (X, A) ⊗ . . . ⊗ H • (X, A)
∪ / H • (X, A)
4. Borsuk-Ulam Theorem: Any continuous map φ : Sn → Rn identifies some pair of antipodal

points.
φ(x)−φ(−x)
Otherwise ϕ : Sn → Sn−1 , ϕ(x) = kφ(x)−φ(−x)k , is continuous and ϕ(−x) = −ϕ(x).
Now ϕ gives a continuous map ϕ̄ : Pn → Pn−1 not trivializing the covering Sn−1 → Pn−1 , so
that the following ring morphism fulfills ϕ̄∗ (x) = x; absurd:
ϕ̄∗ : F2 [x]/(xn ) = H • (Pn−1 , F2 ) −→ H • (Pn , F2 ) = F2 [x]/(xn+1 ).
12.7 Duality Theorem

Let X be a σ-compact space such that Hcp (X, F) = 0 for all p > n and all sheaves F (as a
topological manifold of dimension n, p. 332), and let C • be the Godement resolution of the
constant sheaf defined by a principal ideal domain A, truncated at the n-th step:
0 −→ A −→ C 0 A −→ C 1 A −→ . . . −→ C n−1 A −→ C n −→ 0.
Since this resolution 0 → A → C • splits on stalks, for any sheaf of A-modules M we have a
resolution 0 → M → M⊗A C • , which is Γc -acyclic since the sheaves M⊗A C i A are C 0 Z-modules,
and Hcp (X, M ⊗A C n ) = Hcp+n (X, M) = 0 (p. 336).
Let us fix an injective resolution 0 → A → I 0 → I 1 → 0 of the ring A.
The contravariant functor F (M) = HomA (Γc (X, M ⊗A C p , I q ) is exact (sheaves C p are flat,
sheaves M ⊗A C p are Γc -acyclic and A-modules I q are injective) and F transforms inductive
limits into projective limits since Γc preserves inductive limits (p. 332). By Grothendieck’s
representability theorem3 there is a sheaf of injective A-modules D−p,q such that
HomA (M, D−p,q ) = HomA (Γc (X, M ⊗A C p ), I q ),

Hom•A (M, D) = Hom•A (Γc (X, M ⊗A C), I).
where Homp,q
A (K, L) = HomA (K
−p , Lq ), with the differentials induced by those of K • and L• .
Definition: The dualizing complex of X with coefficients in a principal ideal domain A is

the simple complex DXA defined by the injective bicomplex D •• and, by the following lemma, we
have exact sequences
0 −→ Ext1A (Hcp+1 (X, M), A) −→ Ext−p (M, DX

A
) −→ HomA (Hcp (X, M), A) −→ 0.
Lemma: If K • is a complex of A-modules, we have exact sequences
0 −→ Ext1A (H p+1 (K • ), A) −→ H −p Hom•A (K, I) −→ HomA (H p (K • ), A) −→ 0.

Proof: HomA (H p (K • ), A) and Ext1A (H p (K • ), A) are the kernel and cokernel of
ϕp : HomA (H p (K • ), I 0 ) −→ HomA (H p (K • ), I 1 ).
On the other hand, we have an exact sequence of complexes
0 −→ Hom•A (K, I 1 [−1]) −→ Hom•A (K, I) −→ Hom•A (K, I 0 ) −→ 0,
and the connecting δ−p of the induced cohomology exact sequence
δ−(p+1) δ−p
. . . −−−−−→ H −p Hom•A (K, I 1 [−1]) −→ H −p Hom•A (K, I) −→ H −p Hom•A (K, I 0 ) −−→ . . .

coincides with the morphism ϕp ; i.e. the following square is commutative,

δ−p
H −p Hom•A (K, I 0 ) −−−→ H −p+1 Hom•A (K, I 1 [−1])

|| ||
• ϕp
p 0
HomA (H (K ), I ) −−−→ HomA (H p (K • ), I 1 )
Duality Theorem: If X is a topological manifold of dimension n, all the cohomology sheaves

H−p (DXA ) are null, except TA = H−n (D A ), which is a locally constant sheaf of stalk A, and for
X X
any sheaf of A-modules M we have exact sequences:
0 −→ Ext1A (Hcp+1 (X, M), A) −→ Extn−p A p

A (M, TX ) −→ HomA (Hc (X, M), A) −→ 0.
Proof: If U is an open set in X, when M = AU we have exact sequences
0 −→ Ext1A (Hcp+1 (U, A), A) −→ H −p [D(U )] −→ HomA (Hcp (U, A), A) −→ 0.
If U ' Rn , then Hcn (U, A) ' A and Hcp (X, A) = 0, p 6= n; hence H −n [D(U )] ' A and
H −p [D(U )]
= 0, p 6= n. Therefore H−p (D) = 0, p 6= n, and the sheaf H−n D is locally constant
because, when X = Rn and U is an open ball, we have Hcn (X − U, A) = 0, and the natural
morphism Hcn (U, A) → Hcn (X, A) is an isomorphism.
3 L L
Any O-module M admits an epimorphism I P → M, where P = U OU , because HomO (OU , M) =
M(U ). Hence (p. 134) any minimal pair Qξ of F is fully determined by the subset HomO (P, M) ⊆ F (P).
Now, since D has no term of degree < −n, we have an exact sequence
−n
0 −→ TA
X −→ D −→ D−n+1 −→ . . . −→ D1 −→ 0,
−p n−p
which is an injective resolution of TA A
X ; hence ExtA (M, D) = ExtA (M, TX ).
Corollary: TA n A A
X (U ) = HomA (Hc (U, A), A), and therefore TU = TX |U .
Corollary: 0 −→ Ext1A (Hcp+1 (X, A), A) −→ H n−p (X, TA p

X ) −→ HomA (Hc (X, A), A) −→ 0.
Corollary: H n−p (X, F2 ) = Hcp (X, F2 )∗ .
Proof: When A = F2 , any locally constant sheaf is trivial, so that TFX2 = F2 .
Definitions: The orientation sheaf of a topological manifold X is TX = TZX , and X is

orientable if Z ' TX . The orientations of X at a point x are the two generators of the stalk
TX,x , and the orientations of X are the isomorphisms Z ' TX .
If X is connected of dimension n, the condition of being orientable is equivalent to
0 6= Γ(X, TX ) = HomZ (Hcn (X, Z), Z),
and on any orientable manifold X of dimension n we have exact sequences
0 −→ Ext1Z (Hcp+1 (X, Z), Z) −→ H n−p (X, Z) −→ HomZ (Hcp (X, Z), Z) −→ 0.
Proposition: TA A
X = TX ⊗Z A; hence, TX ' A when X is orientable.
Proof: The natural morphism Hcn (U, Z) ⊗Z A → Hcn (U, A) is an isomorphism by the universal
coefficients theorem. So we have a morphism
TX (U ) ⊗Z A = HomZ (Hcn (U, Z), Z) ⊗Z A −→ HomA (Hcn (U, A), A) = TA

X (U ),
which is an isomorphism when U ' Rn , and it induces an isomorphism TX ⊗Z A = TA

X.
Corollary: H n−p (X, k) = Hcp (X, k)∗ , when X is an orientable manifold of dimension n and k
is a field.
Corollary: If char k 6= 2, a manifold X is orientable if and only if k ' TkX .
Proof: The sheaf ∆X of orientations of X defines an orientation covering ∆et X → X of degree

2, and it is trivial if and only if X is orientable.
When char k 6= 2, the natural morphism ∆et et k et k et
X ,→ TX → (TX ) is injective, and if (TX ) → X
et
is a trivial covering, so is ∆X , and X is orientable.
Corollary: On any smooth manifold X, the topological and smooth orientations coincide.
Proof: Let us consider the sheaf ∆dif

X (U ) = {smooth orientations of U }.
Any smooth orientation of U defines an integration of n-forms with compact support and,
by Stokes theorem, for any (n − 1)-form with compact support ωn−1
Z Z Z
dωn−1 = ωn−1 = ωn−1 = 0.
U ∂U ∅
So, any smooth orientation of U defines a non null linear form

Z
: Hcn (U, R) −→ R,
U
and we obtain an injective morphism of sheaves ∆dif R

X ,→ TX .
R
Now, if X admits a smooth orientation, TX has a global section not vanishing at any point;
hence R ' TR R et
X , and X is topologically orientable. Conversely, if (TX ) → X is a trivial covering,
dif et
so is (∆X ) → X, and X admits a smooth orientation. q.e.d.
1. If X is a compact orientable manifold of odd dimension n, then χ(X) = 0.

χ(X) = p (−1)p dim H p (X, Q) = − p (−1)n−p dim H n−p (X, Q) = −χ(X).
P P
2. If X is a connected manifold of dimension n and Hcn (X, Q) 6= 0, then TQ

X has a non zero
section and X is orientable; hence Hcn (X, Q) ' Q.
3. Any closed submanifold Y in Rn of codimension 1 is orientable. Hence, the real projective

plane is not a closed submanifold of R3 .
We may assume that Y is connected, so that (p. 351) U = Rn − Y has two connected
components (obviously orientable) and the closed subspace exact sequence let us conclude
0 −→ Hcn−1 (Y, Q) −→ Hcn (U, Q) = Q2 −→ Hcn (Rn , Q) = Q −→ 0.
12.7.1 Degree Theory

Theorem: Hcn (X, TX ) = Z, when X is a connected manifold of dimension n.
Proof: If k is field, by duality and universal coefficients,
k = Homk (TkX , TkX ) = Hcn (X, TX ⊗Z k)∗ = (Hcn (X, TX ) ⊗Z k)∗ ,
hence Hcn (X, TX ) ⊗Z k = k and, if Hcn (X, TX ) is finitely generated, Z = Hcn (X, TX ).
On the other hand, if a connected open set is a finite union U = V1 ∪ . . . ∪ Vr of open sets
Vi ' Rn , the Mayer-Vietoris exact sequence
Hcn (V1 , TX ) ⊕ Hcn (V2 ∪ . . . ∪ Vr , TX ) → Hcn (U, TX ) → Hcn+1 (V1 ∩ (V2 ∪ . . . ∪ Vr ), TX ) = 0
shows, by induction on r, that Hcn (U, TX ) is finitely generated; hence Z = Hcn (U, TX ).
Such open sets U cover X, and we have (p. 332)
Hcn (X, TX ) = Hcn (X, lim

−→
(TX )U ) = lim
−→
Hcn (U, TX ) = Z,
because any morphism Z = Hcn (U, TX ) → Hcn (U 0 , TX ) = Z is an isomorphism, since so is the

dual morphism (U and U 0 are connected):
Γ(U 0 , Z) = Hom(Hcn (U 0 , TX ), Z) −→ Hom(Hcn (U, TX ), Z) = Γ(U, Z).
Lemma: If π : Y → X is a covering, then TY = π ∗ TX .
Proof: If U ⊂ Y is an open set such that π : U → π(U ) is a homeomorphism, then we have a

natural isomorphism ϕU : (π ∗ TX )|U → (TY )|U , and ϕV = (ϕU )|V when V is an open set in U ;
hence it defines an isomorphism π ∗ TX → TY .
Theorem: If X is a connected manifold of dimension n,

(
n Z if X is orientable
Hc (X, Z) '
Z/2Z if X is not orientable
Proof: If X is orientable, Hcn (X, Z) ' Hcn (X, TX ) = Z.

If X is not orientable and π : Y → X is the orientation covering, TY = π ∗ TX has a canonical
non null section and Y is orientable.
We have an epimorphism tr : π∗ Z → Z, tr(f )(x) = f (x1 ) + f (x2 ), where π −1 (x) = {x1 , x2 },
and the cohomology exact sequence
Hcn (Y, Z) = Hcn (X, π∗ Z) −→ Hcn (X, Z) −→ Hcn+1 (X, Ker tr) = 0
shows that Hcn (X, Z) ' Z/mZ, where m > 0 since X is not orientable.
π∗ tr
The morphism Z −→ π∗ Z −
→ Z is multiplication by 2; hence so is the composition
π∗ tr
Z/mZ = Hcn (X, Z) −−−→ Hcn (X, π∗ Z) = Z −−→ Hcn (X, Z) = Z/mZ,
and we see that m = 1 or 2. The case m = 1 is impossible since the universal coefficients formula
gives Hcn (X, F2 ) = 0, while by duality Hcn (X, F2 )∗ = H 0 (X, F2 ) = F2 .
Corollary: If X is a connected oriented smooth manifold of dimension n, the integration of

n-forms induces an isomorphism
Z
: Hcn (X, R) −−
∼
→ R.
X
Proof: The linear map X : R = Hcn (X, R) → R is non null (just integrate forms with compact
R
support contained in a coordinate neighborhood).
Definition: If X is a connected orientable manifold of dimension n, any orientation defines

an isomorphism Z = Hcn (X, Z), hence a generator εX of this group. If π : Y → X is a proper
morphism between connected oriented manifolds of dimension n, it induces a morphism
π ∗ : ZεX = Hcn (X, Z) −→ Hcn (Y, Z) = ZεY ,
and the degree of π is the integer number deg π such that π ∗ (εX ) = (deg π)εY .
Theorem: If X is a connected orientable manifold of dimension n and p ∈ X, then the natural

morphism Hpn (X, Z) −→ Hcn (X, Z) is an isomorphism, so that any orientation εX ∈ Hcn (X, Z)
of X defines a normal orientation εp ∈ Hpn (X, Z) at any point p.
Proof: If U is a connected open neighborhood of p, the morphism Hcn (U, Z) → Hcn (X, Z) is an
isomorphism, since so is the dual morphism, and Hpn (X, Z) = Hpn (U, Z).
Hence it is enough to check the theorem on one orientable manifold of dimension n.
When X = Sn , the theorem follows from the local cohomology exact sequence,
Hpn (Sn , Z) −→ H n (Sn , Z) −→ H n (Rn , Z) = 0.
Definition: Let π : Y → X be a continuous map between manifolds of dimension n. If p = π(q),

we have a morphism
π ∗ : Zεp = Hpn (X, Z) −→ Hqn (Y, Z) = Zεq ,
and the degree of π at q ∈ Y is the integer number deg q π such that π ∗ (εp ) = (deg q π)εq .
Theorem: Let π : Y → X be a proper morphism between connected oriented manifolds of di-

mension n. If the fibre of p ∈ X is finite, π −1 (p) = {q1 , . . . , qr }, then
deg π = deg q1 π + . . . + deg qr π.
Proof: It follows directly from the commutative square,
π∗ / H n−1 (Y, Z) = Zεq ⊕ . . . ⊕ Zεq
Hpn (X, Z) π p 1 r

π∗ / H n (Y, Z)
Hcn (X, Z) c
since π ∗ εX = (deg π)εY , π ∗ εp = ((deg q1 π)εq1 , . . . , (deg qr π)εqr ) and the vertical morphisms
transform εp into εX , and εqi into εY , according to the above theorem.
Corollary: If the degree of π is not zero, then π is surjective.
Theorem: Let π : Y → X a continuous map between manifolds of dimension n. If π is a local

homeomorphism at q ∈ Y , then deg q π = ±1.
Proof: The definition of deg q π is local on X and Y , by excision; hence we may assume that π
is a homeomorphism, an obvious case.
Corollary: The degree of z n : C → C at the origin is n.
Proof: It has degree 1 at any point p 6= 0, because it preserves the orientation. q.e.d.
1. Any non constant proper morphism X → Y between Riemann surfaces has positive degree;
hence it is surjective. If the degree is 1, then it is injective, so that it is an isomorphism. In
particular any non constant polynomial P : C → C is surjective: it has a complex root.
2. Let D be a vector field on Rn . If Ω ⊂ Rn is a compact manifold with boundary and D does not
vanish on the boundary, we may consider the degree of the proper morphism φ : ∂Ω → Sn−1 ,
φ(p) = Dp /|Dp |. This degree is zero when D does not vanish on Ω, since φ may be extended
to a neighborhood of Ω and
Z Z Z Z
φ∗ ωn−1 = d(φ∗ ωn−1 ) = φ∗ (dωn−1 ) = φ∗ 0 = 0.
∂Ω Ω Ω Ω
Therefore, when Ω is a small ball centred at an isolated singularity, Dp = 0, the degree does
not depend on the ball and we name it index of the vector field D at p.
In general, if all singularities of the vector field are isolated, considering small open balls Bi
around each singularity p1 , . . . , pr contained in Ω, we have that D does not vanish on the
manifold with boundary Ω − (B1 ∪ . . . ∪ Br ), so that the degree of φ coincides with the sum
of the indices of the vector field at all singularities contained in Ω.
3. If π : Sn → Pn (R), n ≥ 2, is a continuous map, there is no non empty open set U ⊂ Pn (R)
such that π defines a homeomorphism π −1 (U ) −−
∼
→ U.
In fact, otherwise we have an isomorphism
π ∗ : Hpn (Pn , F2 ) = Hpn (U, F2 ) −→ Hpn (π −1 (U ), F2 ) = Hpn (Sn , F2 ),
and so is also π ∗ : H n (Pn , F2 ) → H n (Sn , F2 ) since both groups are generated by the cohomol-
ogy class of a point. Absurd: π ∗ (xn ) = (π ∗ x)n = 0, since π ∗ x ∈ H 1 (Sn , F2 ) = 0.
Lefschetz’s Theorem
The duality isomorphisms HomA (M, D−p,q ) = HomA (Γc (X, M ⊗A C p ), I q ) are given by mor-
phisms Γc (X, D−p,q ⊗A C p ) → I q , defining a morphism ξ : Γc (X, D ⊗A C) → I.
In the case of a manifold of constant dimension n, the morphism
ξ : Hcn (X, TA 0
X ) = Hc (X, D) −→ A
is surjective since otherwise there is no surjective morphisms Hcn (X, M) → A, and there are
when M = TA X . Hence it is an isomorphism.
The isomorphism Hom•A (M, D) = Hom•A (Γc (X, M ⊗A C), I) is induced by the following
pairing (where λ is the obvious morphism),
⊗ λ ξ
Γ(Hom• (M, D)) ⊗ Γc (M ⊗ C • ) −−→ Γc (Hom• (M, D) ⊗ (M ⊗ C • )) −−→ Γc (D ⊗ C • ) −−→ I.
When M = A, we have that Hom• (M, D) = D is an injective resolution of TA •
X, M⊗C = C
•
is a Γc -acyclic resolution of A, Hom• (M, D) ⊗ (M ⊗ C • ) is a Γc -acyclic resolution of TA

X ⊗ A,
p
and the epimorphism H n−p (X, TA X ) → Hom A (Hc (X, A), A) is induced by the pairing
∪ λ ξ
H n−p (X, TA p
X ) ⊗A Hc (X, A) −−→ Hcn (X, TA −→ Hcn (X, TA
X ⊗A A) − X) −
−→ A.
When X is an oriented compact manifold and A = k is a field, the cup product defines a
metric identifying H • (X) = H • (X, k) with the dual H • (X)∗ ,
hc0 , ci = ξ(c0 ∪ c).
By Künneth’s theorem H • (X × X) = H • (X) ⊗ H • (X), so that any orientation εX of X
defines an orientation εX ⊗ εX of X × X, and we fix the normal orientation of the diagonal
∆ : X → X × X so that ∆∗ preserves orientations, and by the projection formula
h∆∗ a, bi = ha, ∆∗ bi.
Theorem: The cohomology class of the diagonal defines the metric of the cup product,
ha ⊗ b, pX×X (∆)i = ha, bi.
Proof: ha ⊗ b, pX×X (∆)i = ha ⊗ b, ∆∗ (1)i = h∆∗ (a ⊗ b), 1i = ha ∪ b, 1i = ha, bi.
Corollary: If we fix a base (ai ) of H • (X), and (bi ) is the dual base, hai , bj i = δij ,
pX×X (∆) = i (−1)deg ai ai ⊗ bi .
P
Proof: hbj ⊗ ai , i (−1)deg ai ai ⊗ bi i = (−1)deg ai δij hbi ⊗ ai , ai ⊗ bi i = δij hbi , ai ihai , bi i

P
= (−1)(deg ai )(deg bi ) δij = hbj , ai i = hbj ⊗ ai , pX×X (∆)i.
Definition: If X is an oriented compact manifold, the Lefschetz number Λf of a continuous

map f : X → X is the global intersection number of the graph Γf = f × 1 : X → X × X with
the diagonal,
Λf = hpX×X (∆), pX×X (Γf )i.
n
(−1)p trfp∗ , fp∗ : H p (X) → H p (X).
P
Lefschetz’s Formula: Λf =
p=0
Proof: Λf = hpX×X (∆), (f × 1)∗ 1i = h(1 × f )∗ pX×X (∆), 1i

= h(1 × f )∗ i (−1)deg ai ai ⊗ bi , 1i = i (−1)deg ai hf ∗ ai , bi i = p (−1)p trfp∗ .
P P P
Corollary: The self-intersection of the diagonal is the Euler-Poincaré characteristic,

∆∗ pX×X (∆) = χ(X) · εX .

1. If X is a compact orientable manifold of dimension n = 4d + 2, then χ(X) is even.

Since (−1)p dim H p (X, Q) = (−1)n−p dim H n−p (X, Q), we only have to show that the dimen-
sion of H 2d+1 (X, Q) is even. Now, the cup product defines a non singular alternate metric
∪
H 2d+1 (X, Q) × H 2d+1 (X, Q) −−→ Q.
2. If a compact orientable smooth manifold X admits a continuous vector field without zeros,
then χ(X) = 0.
If the tangent bundle π : T X → X admits a continuous section s without zeros, then we have
0 = s∗ (pT X (s0 X)) = s∗0 pT X (s0 X) . Now, since X admits a riemannian metric (p. 286)

and T X is the normal bundle of the diagonal embedding ∆ : X → X × X, by the tubular

neighborhood lemma
0 = s∗0 pT X (s0 X) = ∆∗ pX×X (∆) = χ(X) · εX .

3. If τ is a homography of the complex projective line, the degree is 1 since it preserves orien-
tations; hence Λτ = 2. When it is parabolic, the topological intersection multiplicity of the
diagonal and the graph at the unique fixed point is 2.
4. Any continuous map f : S2n → S2n of degree 6= −1 has Lefschetz number Λf 6= 0, and f has
some fixed point. Analogously, any continuous map S2n+1 → S2n+1 of degree 6= 1 has some
fixed point.
5. If X = R2 /Z2 , any matrix A ∈ M2×2 (Z) induces a continuous map f : X → X. The action
of f ∗ on H 1 (X, R) = R2 is defined by A, and on H 2 (X, R) = R by the determinant. Hence
Λf = 1 − trA + det A.
6. If f is an analytic endomorphism of a complex torus C/(Zα + Zβ), the lifting of f to the

universal covering is an endomorphism C → C fixing the origin; hence it is the product by a
2 2 2 2
√ the degree of f is d = a + b , and Λf = 1 − 2a + a + b . So
number complex a + bi, so that
we see that |Λf − d − 1| ≤ 2 d.
12.8 Characteristic Classes

Definition: If E → X is a real or complex (K = R or C) vector bundle, P(E) is the set of
all vector subspaces of dimension 1 of the fibres Ex , with the quotient topology, and we have a
natural projection π : P(E) → X. The vector bundle π ∗ E = E ×X P(E) has a tautological line
sub-bundle ξE ,→ π ∗ E, the fibre over a point p ∈ P(E) being the corresponding line of Eπ(p) .
The tautological line bundle of the projective space Pd is denoted by ξd .
Lemma: If L is a line bundle over a separated compact space X, then there exists a continuous
map f : X → Pd such that L = f ∗ ξd .
Proof: If we see that the dual line bundle L∗ is generated by a finite number of global sections
{s0 , s1 , . . . , sd }, then we have an epimorphism
X × Kd+1 −→ L∗ , (x, λ0 , . . . , λd ) 7→ λ0 s0 (x) + . . . + λd sd (x),
and the injection L → X × Kd+1 defines a continuous map f : X → Pd such that L = f ∗ ξd .

Now, since X is completely regular, for any point of X there is a global section generating
the fibre on a neighborhood and, X being compact, a finite number of global sections generate
the fibre at any point.
12.8. CHARACTERISTIC CLASSES 361
Corollary: The obstruction class of the tautological line bundle generates the group H 2 (Pd , Z)
in the complex case, and the group H 1 (Pd , F2 ) in the real case.
Proof: Let i : Pd−1 → Pd be a hyperplane. In the complex case, the closed subspace exact
sequence shows that i∗ : H 2 (Pd , Z) → H 2 (Pd−1 , Z) is an isomorphism. Since
i∗ (δ(ξd )) = δ(i∗ ξd ) = δ(ξd−1 ),
the index m of the subgroup generated by δ(ξd ) in H 2 (Pd , Z) does not depend on d.
The above lemma shows that any line bundle on P1 is a m-th power of a line bundle.
Absurd when m 6= 1, since H 2 (P1 , Z) ' Z.
The argument also holds in the real case. q.e.d.
From now on, in the complex case we shall always consider cohomology groups with coeffi-
cients in Z, and in the real case with coefficients in F2 , and we fix the orientations so that the
obstruction class δ(ξd ) is just the cohomology class of a hyperplane.
Hirsch-Leray Theorem: Let E be a vector bundle of rank r over a σ-compact space X and
let xE = δ(ξE ) be the obstruction class of the tautological bundle of P(E). Then H • (P(E)) is a
free H • (X)-module of base {1, xE , x2E , . . . , xr−1
E }.
Proof: Let Z → C • be the Godement resolution of the constant sheaf Z on P(E).

Any cohomology class xjE , 0 ≤ j ≤ r − 1, is represented by a global section of C 2j , and these
sections define a morphism of complexes
⊕j Z[−2j] −→ π∗ C •
which is a quasi-isomorphism (cohomology of the fibre, p. 339, and projective spaces, p. 352).
Now, since π∗ C • is a complex of flasque sheaves, we have a group isomorphism
⊕j H • (X, Z)[−2j] = H • Γ(X, π∗ C • ) = H • Γ(P(E), C • ) = H • (P(E), Z)

transforming (a0 , . . . , ar−1 ) into π ∗ (a0 ) + π ∗ (a1 )xE + . . . + π ∗ (ar−1 )xr−1

E .
π∗
In fact Γ(X, π∗ C • ) −−−→ Γ(P(E), π ∗ π∗ C • ) → Γ(P(E), C • ) is just the identity.
In the case of real vector bundles we replace Z by F2 and 2j by j.
Definitions: Let X be a σ-compact space. The Chern classes of a complex vector bundle
E → X of rank r are the coefficients ci (E) ∈ H 2i (X, Z) of the characteristic polynomial of the
endomorphism of the free H • (X)-module H • (P(E)) defined by xE = δ(ξE ),
xrE + c1 (E)xr−1
E + . . . + cr (E) = 0.
P
We agree that c0 (E) = 1 and ci (E) = 0, i > r. The total Chern class is c(E) = i ci (E).
Analogously, when E → X is a real vector bundle, we have Stieffel-Whitney classes
wi (E) ∈ H i (X, F2 ), and we agree
P that w0 (E) = 1 and wi (E) = 0, i > r. The total Stieffel-
Whitney class is w(E) = i wi (E), and the Stieffel-Whitney classes wi (X) of a smooth
manifold X are those of the tangent bundle, wi (X) = wi (T X).
From now on we shall give statements and proofs only for the Chern classes, but they also
hold for the Stieffel-Whitney classes.
Functoriality: ci (f ∗ E) = f ∗ (ci (E)), for any continuous map f : T → X.

Proof: Let us consider the continuous map 1 × f : P(f ∗ E) = P(E) ×X T −→ P(E) ×X X = P(E).
We have ξf ∗ E = (1 × f )∗ ξE , where xf ∗ E = (1 × f )∗ xE , and in H • (P(f ∗ E)),
0 = (1 × f )∗ (xrE + a1 xr−1 r ∗ r−1 ∗

E + . . . + ar ) = xf ∗ E + (f a1 )xf ∗ E + . . . + f ar .
Theorem: c1 (L) = −δ(L), for any line bundle L. Hence c1 (L ⊗ L0 ) = c1 (L) + c1 (L0 ).
Proof: When L is a line bundle, P(L) = X, ξL = L, xL = δ(L), and xL + c1 (L) = 0.
Theorem: The class cr (E), r = rk E, are the zeros of any continuous section s : X → E,
cr (E) = s∗ pE (s0 (X)) .

Proof: Let x̄ be the obstruction class of the tautological bundle over Ē = P(E ⊕ 1).
By the Hirsch-Leray theorem, the cohomology class of the null section is
pĒ (s0 (X)) = a0 x̄r + a1 x̄r−1 + . . . + ar , ai ∈ H 2i (X).
Its restriction to any fibre is the cohomology class of a point (vector bundles are locally
trivial), which is just the restriction of x̄r .
Since the restriction of ai , i > 0, is null, we see that a0 = 1.
Now, the restriction of this class to the infinity j : P(E) → P(E ⊕ 1) is null since the zero
section s0 does not intersect the infinity,
0 = j ∗ (x̄r + a1 x̄r−1 + . . . + ar ) = xrE + a1 xr−1

E + . . . + ar ,
hence ai = ci (E). Since the tautological bundle is trivial on the affine part E,
s∗ pE (s0 (X)) = s∗ pĒ (s0 (X)) = s∗ (x̄r + a1 x̄r−1 + . . . + ar ) = ar = cr (E).

Splitting Principle: Let E → X be a vector bundle of rank r. There exists a base change
π : Y → X such that π ∗ E admits a filtration 0 = Er ⊂ . . . ⊂ E1 ⊂ E0 = E whose quotients
Ei−1 /Ei are line bundles and π ∗ : H • (X) → H • (Y ) is injective.
Proof: On P(E) we have that ξE ,→ π ∗ E is a line bundle and π ∗ E/ξE is a vector bundle of rank
r − 1. Since π ∗ : H • (X) → H • (P(E)) is injective by the Hirsch-Leray theorem, we conclude by
induction on the rank.
Additivity: If 0 → E 0 → E → E 00 → 0 is an exact sequence of vector bundles, then
c(E) = c(E 0 ) · c(E 00 ).
Proof: By the splitting principle there is a base change p : Y → X such that p∗ : H • (X) → H • (Y )
is injective and p∗ E 0 , p∗ E 00 admit filtrations with line quotients. Since p∗ is injective and Chern
classes are functorial, we are reduced to show that for any filtration 0 = Er ⊂ . . . ⊂ E1 ⊂ E0 = E
with line quotients Ei−1 /Ei we have c(E) = (1+α1 ) . . . (1+αr ), where we put αi = c1 (Ei−1 /Ei ).
We proceed by induction on rm and it is an identity when r = 1.
Let us consider the projection π : P(E) → X and the inclusion j : P(E1 ) → P(E).
The natural morphism ξE → π ∗ (E/E1 ) defines a global section of π ∗ (E/E1 ) ⊗ ξE ∗ vanishing
on P(E1 ), and transversally intersecting the zero section (check it locally). Hence
j∗ (1) = c1 (π ∗ (E/E1 ) ⊗ ξE
∗
) = xE + α1 .
12.8. CHARACTERISTIC CLASSES 363
By induction j ∗ (xE + α2 ) . . . (xE + αr ) = 0, and applying j∗ , the projection formula shows

that ci (E) is just the i-th elementary symmetric function of α1 , . . . , αr ,
0 = j∗ j ∗ (xE + α2 ) . . . (xE + αr ) = (xE + α2 ) . . . (xE + αr )j∗ (1) = (xE + α1 ) . . . (xE + αr ).

Roots of a Vector Bundle: If 0 → E1 → E → E2 → 0 is an exact sequence of vector bundles

and we fix a hermitian metric (scalar product in the real case) on E, then E = E1 ⊕ E1⊥ , and
E1⊥ ' E2 ; hence E ' E1 ⊕ E2 .
Now, by the splitting principle, there is a base change π : Y → X such that π ∗ : H • (X) →
H • (Y ) is injective and π ∗ E is a direct sum of line bundles, π ∗ E = L1 ⊕ . . . ⊕ Lr , and we say
that α1 = c1 (L1 ), . . . , αr = c1 (Lr ) are the ”roots” of E, since ci (E) is just the i-th elementary
symmetric function of α1 , . . . , αr . We also put Lαi = Li .
1. ci (E ∗ ) = (−1)i ci (E).
If E = Lα1 ⊕ . . . ⊕ Lαr , then E ∗ = L−α1 ⊕ . . . ⊕ L−αr .
2. c1 (E) = c1 (Λr E).

If E = Lα1 ⊕ . . . ⊕ Lαr , then Λr E = Lα1 ⊗ . . . ⊗ Lαr = Lα1 +...+αr .
3. If pY is the sheaf of ideals of a hypersurface Y of a smooth manifold X, then w1 (pY ) = [Y ];

hence Y admits a global equation f = 0 with non null differential at any point of Y if and
only if its cohomology class [Y ] ∈ H 1 (X, F2 ) is null.
The inclusion pY → CX ∞ induces a section s : C ∞ → p∗ transversally intersecting the zero
X Y
section along Y ; hence [Y ] = w1 (p∗Y ) = w1 (pY ).
4. If N is the normal bundle of a compact smooth submanifold i : Y → X of codimension d,

then wd (N ) = i∗ [Y ].
Since X admits a riemannian metric (p. 286) there is a neighborhood U of Y in X and a
diffeomorphism U ' N transforming i into the zero section (p. 410).
5. Let X be a compact smooth manifold of dimension n. If X is a smooth submanifold of Rn+d ,

then w(X)−1 has degree < d.
We have an exact sequence 0 → T X → (T Rn+d )|X → N → 0 and he tangent bundle T Rn+d
is trivial; hence w(X)−1 = w(N ) = 1 + w1 (N ) + . . . + wd (N ), and wd (N ) = 0 since the
cohomology class of X in Rn+d is null.
6. w(Pn,R ) = (1 + x)n+1 .
If E = Rn+1 , the projection E − {0} → Pn induces, at any vector e 6= 0, an identification
of E/Re with the tangent space to Pn at the point hei, so defining a canonical isomorphism
T Pn = Hom(ξn , E/ξn ), and we have an exact sequence
0 −→ Hom(ξn , ξn ) −→ Hom(ξn , E) −→ T Pn −→ 0,
w(Pn ) = w(ξn∗ )n+1 = (1 + δ(ξn ))n+1 = (1 + x)n+1 .
7. P4 is not a smooth submanifold of R7 .

w(P4 )−1 = (1 + x + x4 )−1 = 1 + x + x2 + x3 .
8. If Pn is parallelizable, then n + 1 is a power of 2.

If n + 1 = 2r m, with m odd, then the tangent bundle T Pn is not trivial because w2r (Pn ) 6= 0,
r m r
(1 + x)n+1 = 1 + x2 = 1 + mx2 + . . .
9. If there is a bilinear product on Rn without zero divisors, then n is a power of 2.

If a1 , . . . , an is a base of Rn and we put vi (x) = xa−1
1 ai , then v1 (x) = x, v2 (x), . . . , vn (x) are
linearly independent for any x 6= 0. Hence v2 , . . . , vn define linearly independent sections of
T Pn−1 = Hom(ξ, Rn /ξ), and Pn−1 is parallelizable.
12.9 Spectral Sequences

We work in the category of modules over a ring (or a sheaf of rings). Any exact triangle
i1
C1[ / C1
δ1 j1
E1
defines a differential d1 = j1 δ1 : E1 −→ E1 , d21 = j1 δ1 j1 δ1 = 0, and we put:
C2 = Im i1 , E2 = H(E1 ) = Ker d1 /Im d1 .

i2 : C2 → C2 , the restriction of i1 a C2 .
δ2 : E2 → C2 , the morphism induced by δ1 : Ker d1 → C2 on the quotient.
j2 : C2 → E2 , the morphism defined by the commutative diagram
i1
0 / δ1 (E1 ) / C1 / C2 /0
j1 j1 j2

0 / j1 δ1 (E1 ) / Ker d1 / E2 /0
i2
Theorem: So we obtain a derived exact triangle C2[ / C2
δ2 j2
E2
Proof: The equality i2 (C2 ) = i1 (Ker j1 ) = Ker j2 follows from the snake’s lemma, applied to the
above diagram. The other two are immediate. q.e.d.
Iterating we obtain derived exact triangles and differentials

ir / Cr ,
Cr[ dr = jr δr : Er −→ Er
δr jr
Er
p, F p+1 ⊆ F p , dF p ⊆ F p .
S
Now let (M, d) be a filtered differential module: M = pF
The exact sequence
i
0 −→ ⊕p F p+1 −→ ⊕p F p −→ ⊕p F p /F p+1 −→ 0
i1
induces (p. 304) an exact triangle C1 g / C1 = ⊕p H(F p )
j1
δ1
E1 = ⊕p H(F p /F p+1 )
12.9. SPECTRAL SEQUENCES 365
Let ipk : H(F p ) → H(F p−k ) be the natural morphism. In any derived triangle we have
p+r−1
H(F p ).
L L
Cr = Im ir−1 = Im ir−1 ⊆
p p
Moreover Er = ⊕p Erp , and the derived triangles decompose as exact sequences
ir / Cr = Im ir−1
Cr[
δr jr
Er
δ i jr δ
. . . −→ Erp −−r→ Im ir−1
p+r
−−r→ Im ip+r−1
r−1 −−→ Erp+r−1 −−r→ . . .
and the differential dr is given by morphisms
dr = jr δr : Erp −→ Erp+r .
Now we assume that M is a complex, M = ⊕n M n , dM n ⊆ M n+1 , with a compatible

filtration, F p = ⊕n (F p ∩ M n ) = ⊕n M p,n−p .
Let ikp,n−p : H n (F p ) → H n (F p−k ) be the natural morphism. Then
p+r−1,n+1−p−r
Cr = Im ir−1 = ⊕p,n Im ir−1 ⊆ ⊕p,n H n (F p ).
Moreover Er = ⊕p,n Erp,n−p , the triangles decompose as exact sequences
δ p+r,q+1−r i
p+r−1,q+2−r jr
. . . −→ Erp,q −−r→ Im ir−1 −−r→ Im ir−1 −−→ Erp+r−1,q+2−r −→ . . .
and the differential dr is given by morphisms
dr = jr δr : Erp,q −→ Erp+r,q−r+1 .
On the other hand, the images of natural morphisms ip,n−p∞ : H n (F p ) → H n (M ) define a

filtration of H n (M ), and we shall study whether the spectral sequence converges to
p,n−p p,n−p
GH n (M ) = = Im ip,n−p p+1,n−p−1
L
E∞ , E∞ ∞ /Im i∞ .
p
Definition: The filtration is regular if for all n we have H n (F p ) = 0 when p 0.
Theorem: If the filtration is regular, then the spectral sequence converges,

p,q
E∞ = lim
−→
Erp,q
and in such a case we put E2p,q ⇒ H p+q (M ).
Proof: Once we fix p and n, the morphism δr : Erp,n−p → Im ip+r,n+1−p−r

r−1 is null when r 0,
p+r,n+1−p−r n+1 p+r n+1 p+1
because ir−1 :0=H (F )→H (F ),
Im ip+1,n−p−1
r−1 −→ Im ip,n−p p,n−p
r−1 −→ Er −→ 0.
Moreover dr = jr δr vanishes on Erp,n−p , and we have epimorphisms Erp,n−p → Er+1

p,n−p
.
Taking inductive limit on r, we conclude,
Im ip,n−p
∞ −→ Im ip+1,n−p−1
∞ −→ lim
−→
Erp,n−p −→ 0.
Theorem: Let φ : M → M̄ be a morphism of filtered complexes, φ(F p ) ⊆ F̄ p . If both filtrations

are regular and φ induces isomorphisms Erp,q −
∼
→ Ērp,q for some index r, then φ is a quasi-
n ∼ n
isomorphism, φ : H (M ) −
→ H (M̄ ).
∼
Proof: Since the spectral sequences converge, φ induces an isomorphism GH n (M ) − → GH n (M̄ );
n
hence also on completions (p. 195), but H (M ) is complete when the filtration is regular because
p,n−p
the morphisms i∞ : H n (F p ) → H n (M ), p 0, are null.
Bicomplex Spectral Sequence: A bicomplex K •• admits the filtration F p =

L L i,j
K ,
i≥p j
E1p,q = Hdq2 (K p• ),
E2p,q = Hdp1 (Hdq2 (K •• )).
If the filtration is regular (for example, if K •• has bounded below diagonals) we have a
spectral sequence converging to the cohomology of the bicomplex
E2p,q = Hdp1 (Hdq2 (K •• )) ⇒ H p+q (K •• ).
Analogously, if the bicomplex K •• has bounded above L Ldiagonals, we obtain a convergent

spectral sequence when considering the filtration F p = K i,j ,
i j≥p
E2p,q = Hdp2 (Hdq1 (K •• )) ⇒ H p+q (K •• ).
Hypercohomology Spectral Sequence: Let K • be a bounded below complex.

p,• p,•
If we fix injective resolutions IB and IH of the boundaries B p and the cohomology H p of
•
K , then we have exact sequences
p,•
0 −→ IB −→ IZp,• −→ IH
p,•
−→ 0,
where IZp,• is an injective resolution of the cycles Z p , and exact sequences

p+1,•
0 −→ IZp,• −→ I p,• −→ IB −→ 0,
where I p,• is an injective resolution of K p . By the bicomplex theorem, K • → I •• is a quasi-

isomorphism, and moreover cycles, boundaries and cohomology of I •• respect to the differential
d1 are injective resolutions of cycles, boundaries and cohomology •
n
••
ofnK . •
Now, if F is a left exact additive functor, then H F (I ) = R F (K ), and
Hdp1 F (I •• ) = F Hdp1 (I •• ) = F (IH

p•

),
since the cycles, boundaries and cohomology of the complex (I •q , d1 ) are injective, so that any
additive functor preserves them. Hence Hdp2 (Hdq1 (I •• )) = Rp F (H q (K • )), and the second spectral
sequence of the bicomplex F (I •• ) is
E2p,q = Rp F (H q (K • )) ⇒ Rp+q F (K • ).
Grothendieck’s Spectral Sequence: Let F : A B and G : B C be left exact covariant

functors. If F transforms injective objects into G-acyclic objects, then for any bounded below
complex K • of A we have
R(GF )(K • ) −
∼
→ RG(RF (K • )),
E2p,q =Rp G(Rq F (M )) ⇒ Rp+q (GF )(M ).
12.9. SPECTRAL SEQUENCES 367
Proof: Let K • −
∼
→ I • be an injective resolution. Since F (I • ) is G-acyclic,
R(GF )(K • ) = GF (I • ) −
∼
→ RG(F (I • )) = RG(RF (K • )),
and we have the hypercohomology spectral sequence
E2p,q = Rp G(Rq F (K • )) ⇒ Rp+q G(F (I • )) = Rp+q (GF )(K • ).
Leray’s Spectral Sequence: If f : X → Y and g : Y → Z are continuous maps, then we have

spectral sequences
E2p,q = Rp g∗ (Rq f∗ F) ⇒ Rp+q (gf )∗ F.
Proof: The functor f∗ transforms flasque sheaves into flasque sheaves; hence g∗ -acyclic. q.e.d.
1. E2p,q = H p (Y, Rq f∗ F) ⇒ H p+q (X, F), (when f : X → Y is a continuous map).
2. E2p,q = Hcp (Y, Rq f! F) ⇒ Hcp+q (X, F), (if moreover X, Y are σ-compact).
Γc (X, −) = Γc (Y, −) ◦ f! , and f! (C p F) is a C 0 Z-module; hence it is Γc (Y, −)-acyclic.
3. E2p,q = H p (X, HYq F) ⇒ HYp+q (X, F), (when Y is a closed set in X).
ΓY = Γ ◦ ΓY and, if C is a Godement sheaf, then ΓY C is flasque (p. 350).
4. E2p,q = H p (X, ExtqO (M, N )) ⇒ Extp+q

O (M, N ).
HomO (M, −) = Γ ◦ HomO (M, −) and, if I is an injective O-module, then HomO (M, I) is a
flasque sheaf since all the restriction morphisms are surjective,
HomO (M, I) −→ HomO (MU , I) = HomO (M|U , I|U ).

Chapter 13
Analysis IV
13.1 Dirichlet Problem

Lemma: Let U ⊆ Rd be an open set. If V is a connected open set with compact closure V̄ ⊂ U ,
then there is a constant c such that for any harmonic function u > 0 on U we have
u(x) ≤ cu(y) , ∀x, y ∈ V.
Proof: Fix 0 < r < 21 d(V, ∂U ). If x, y ∈ V and |y − x| < r, then B(x, r) ⊂ B(y, 2r) b U , so that
Z Z
1 Vol B2r 1
u(x) = udm ≤ udm = 2d u(y).
Vol Br B(x,r) Vol Br Vol B2r B(y,2r)
Since V̄ is compact, it admits a cover by a finite number m of balls of radius r/2.

Since V is connected, given x, y ∈ V there is a sequence x = x0 , . . . , xk = y with |xi −xi−1 | < r
and k ≤ m + 1. Hence
u(x) ≤ 2d u(x1 ) ≤ 22d u(x2 ) ≤ . . . ≤ 2kd u(y) ≤ 2(m+1)d u(y).
Harnack’s Theorem: Let (un ) be an increasing sequence of harmonic functions on a connected

open set U ⊆ Rd . If there is a point p ∈ U such that the sequence un (p) is bounded, then (un )
uniformly converges on any compact set to an harmonic function u on U .
Proof: Given ε > 0, there is an index k such that un (p) − um (p) < ε, for any n ≥ m ≥ k.
If V is a connected bounded neighborhood of p with V̄ ⊂ U , by the lemma

un (x) − um (x) ≤ c un (p) − um (p) < cε,
so that (un ) uniformly converges on V , and the limit function is harmonic (p. 263).
Proposition: Let Pk be the vector space of all homogeneous polynomials of degree k on Rd , and
Hk ⊂ Pk the vector subspace of harmonic polynomials. If we put r2 = x21 + . . . + x2d , then
Pk = Hk ⊕ r2 Hk−2 ⊕ r4 Hk−4 ⊕ . . .
λα xα defines a differential operator DP =

P P
Proof: Any homogeneous polynomial P = λα ∂α ,
and P · Q := DP (Q) defines a scalar product on Pk since xα · xβ = (α!)δαβ . Moreover,
(∆Pk ) · Qk−2 = Pk · (r2 Qk−2 ),
so that Hk = (r2 Pk−2 )⊥ . Hence Pk = Hk ⊥ (r2 Pk−2 ), and we conclude by induction on k.
369
370 CHAPTER 13. ANALYSIS IV
Dirichlet Problem (on Balls): Let B ⊂ Rd be an open ball. If f ∈ C(∂B), then there is a
unique extension u ∈ C(B̄) harmonic on B.
Proof: By the maximum principle, the extension is unique, and to prove the existence we may
assume that B = B(0, r0 ).
If f is the restriction of a homogeneous polynomial Pk , then Pk = Hk +r2 Hk−2 +r4 Hk−4 +. . .
for some harmonic polynomials Hi , and uk = Hk + r02 Hk−2 + r04 Hk−4 + . . . is the extension.
P P
When f is the restriction of a polynomial P = k Pk , just put u = k uk .
Finally, by the Stone-Weierstrass theorem, f is a uniform limit f = lim pn of polynomial
functions pn on ∂B. If un is the harmonic extension of pn , then (un ) converges because by the
maximum principle kun − um kB̄ ≤ kpn − pm k∂B , and u = lim un is the required extension.
Corollary: Let p be a point of an open set U ⊆ Rd . If an harmonic function u on U − p is

bounded in a neighborhood of p, it is a removable singularity. Hence, if u may be continuously
extended to p, the extension is harmonic.
Proof: We may assume that p = 0 and B̄ ⊂ U , where B = B(0, 1).

By the above result, there exists an extension v ∈ C(B̄) of u|∂B , harmonic on B, and we
conclude if we show that v = u on B − 0.
Given ε > 0, the function u − v + ε 1 − |x|2−d , resp. u − v + ε ln |x| when d = 2, is continuous

on B̄ − 0, harmonic on B − 0, null on ∂B and it goes to −∞ at the origin. By the maximum

principle, it is negative on B − 0. When ε → 0, we see that u − v ≤ 0 on B − 0.
The same argument, with ε < 0, shows that u − v ≥ 0 on B − 0.
Definition: Let U ⊆ Rd be an open set, d ≥ 2. A function u ∈ C(U ) is subharmonic if any

point p ∈ U has a neighborhood V such that for any closed ball B̄ ⊂ V with centre at p we have
Z
1
u(p) ≤ u dm.
Vol B B
The sum and the maximum of two subharmonic functions also are subharmonic functions.
Maximum Principle: If a subharmonic function u on a connected open set U ⊆ Rd attains a

maximum, then it is constant.
Proof: The points p ∈ U where u attains the maximum M form R an open set because if u is not
the constant M on a ball B ⊂ U with centre at p, then Vol1 B B u dm < M = u(p).
Characterization: If u ∈ C(U ), then the following conditions are equivalent:
1. The function u is subharmonic on U .
2. If V is a connected open set with compact closure V̄ ⊂ U and v ∈ C(V̄ ) is harmonic on V

and u ≤ v on ∂V , then u ≤ v on V .
Z
1
3. For any closed ball B̄ = B̄(p, r) ⊂ U we have u(p) ≤ u dS, where S = ∂ B̄.
Vol S S
Z
1
4. For any closed ball B̄ = B̄(p, r) ⊂ U we have u(p) ≤ u dm.
Vol B̄ B̄
13.1. DIRICHLET PROBLEM 371
Proof: (1 ⇒ 2) The function u−v is subharmonic on V . Since u−v ≤ 0 on ∂V , by the maximum

principle u − v ≤ 0 on V .
(2 ⇒ 3) Let us consider an extension v ∈ C(B̄) of u|S , harmonic on B. By hypothesis u ≤ v
on B, so that Z Z
1 1
u(p) ≤ v(p) = v dS = u dS.
Vol S S Vol S S
(3 ⇒ 4) Since u(p) − u has non-negative integral on any sphere S(p, ρ), 0 < ρ ≤ r, it also
has (p. 261) non-negative integral on B̄(p, r). Finally, (4 ⇒ 1) is obvious.
Lemma: Let B̄ ⊂ U be a closed ball. If u ∈ C(U ) is subharmonic, then the function uB ∈ C(U ),
harmonic on B and such that uB = u on U − B, also is subharmonic.
Proof: We have u ≤ uB by the point 2 of the characterization.

Hence on any small sphere S with centre at p ∈ U − B we have
Z Z
1 1
uB (p) = u(p) ≤ u dS ≤ uB dS.
Vol S S Vol S S
Perron Method: Let U ⊂ Rd be a connected open set with compact closure. If f ∈ C(∂U ),
then F = {u ∈ C(Ū ) : u|∂U ≤ f and u subharmonic on U } defines an harmonic function on U :
h(x) = sup {u(x)}.

u∈F
Proof: By the maximum principle u(x) ≤ max∂U u ≤ max∂U f < ∞; hence h is well-defined.
Given p ∈ U , we consider a sequence un ∈ F such that h(p) = lim un (p). Replacing un by
max{u1 , . . . , un }, we may assume that un ≤ un+1 . Fix a closed ball B̄ ⊂ U with centre at p.
Replacing un by (un )B we may also assume that un is harmonic on B (if u ≤ u0 , then uB ≤ u0B
by the maximum principle).
Now, by Harnak’s theorem, u = lim un is harmonic on B, and u(p) = h(p).
To conclude, it is enough to show that u = h on B:
If u(q) < h(q) at some q ∈ B, then there is g ∈ F such that u(q) < g(q) ≤ h(q). Let
ūn = (max{un , g})B , so that ū = lim ūn is an harmonic function on B such that u ≤ ū and
u(p) = ū(p) = h(p). By the minimum principle u = ū on B. Absurd, u(q) < ū(q). q.e.d.
The problem is to determine whether h defines a continuous extension of f .
Definition: A barrier for U at y0 ∈ ∂U is a function b ∈ C(Ω̄), harmonic on U , such that b > 0

except b(y0 ) = 0. A boundary point y0 is regular if there is a barrier for U at y0 .
Examples: If there is a closed ball B̄ = B̄(p, r) such that B̄ ∩ Ū = y0 , then y0 is a regular point.
A barrier is b(x) = r2−d − |x − p|2−d , resp. b(x) = ln |x − p| − ln r when d = 2.
If U is convex (contains the segment joining any two points of U ), the boundary is regular.
If the boundary of U is a smooth manifold, the boundary is regular.
Lemma: Let h be the harmonic function provided by the Perron method. If y0 ∈ ∂U is a regular
point, then lim h(x) = f (y0 ), (x ∈ U ).
x→y0
Proof: Given ε > 0, since f and b are continuous on ∂U , there is λ 0 such that
|f (y) − f (y0 )| < ε + λb(y) , ∀y ∈ ∂U,

f (y0 ) − λb(y) − ε < f (y) < f (y0 ) + λb(y) + ε , ∀y ∈ ∂U.
Hence, all the harmonic functions f (y0 ) − λb(x) − ε are in F, while f (y0 ) + λb(x) + ε bounds
above any function in F. Therefore
f (y0 ) − λb(x) − ε ≤ h(x) ≤ f (y0 ) + λb(x) + ε , ∀x ∈ U.
When x → y0 , we conclude that f (y0 ) − ε ≤ lim h(x) ≤ f (y0 ) + ε.
x→y0
Theorem: Let U ⊂ Rd
be a connected open set with compact closure and regular boundary. Any
function f ∈ C(∂U ) admits a unique extension h ∈ C(Ū ) harmonic on U .
13.1.1 Uniformization Theorem

Definitions: Let X be a Riemann surface. The laplacian of u ∈ C 2 (X) is the continuous
2-form ∆u := d(∗du), where ∗ is the transpose of J, and u is harmonic when ∆u = 0.
In a local analytic coordinate z = x + yi we have ∗dx = −dy, ∗dy = dx, so that
2
∂ u ∂2u

∆u = − + 2 dx ∧ dy.
∂x2 ∂y
A function u ∈ C(X) is subharmonic when so it is in a coordinate neighborhood of each
point of X. In fact, this condition does not depend on the fixed coordinates, by the point 2 of
the characterization of subharmonic functions.
Examples: If f = u + vi is an analytic function, then u and v are harmonic, by the Cauchy-
Riemann equations. Conversely, if u is an harmonic function and H 1 (X, R) = 0, then there is a
(unique up to the addition of a constant) function v ∈ C 2 (X) such that ∗du = dv, and f = u+vi
is analytic because locally it satisfies the Cauchy-Riemann equations.
If (U, z) is coordinate neighborhood of a point p ∈ X and z(p) = 0, then ln |z| is harmonic
on U − p. Locally, it is the real part of the analytic function ln z = ln |z| + (arg z)i.
If f is an analytic function and f (p) 6= 0, then ln |f | is harmonic in a neighborhood of p,
because it is the real part of ln f .
Definition: Let U ⊂ X be a relatively compact open set. A Green’s function on U with pole
at p ∈ U is a function g ∈ C(Ū − p) such that
1. g = 0 on the boundary ∂U , and g is harmonic on U − p.
2. g − ln |z| has a removable singularity at p, where z is a local coordinate such that z(p) = 0.
(This condition does not depend on the fixed local coordinate).
If it exists, such function g is unique, since the difference of two Green’s functions would be
harmonic on U and null on ∂U .
If f is an analytic function, then ln |f | is the Green’s function when |f | = 1 on ∂U and f
has no zero on U , except a simple zero at z.
Imagine that the electrostatical force between punctual charges in C is proportional to the
charges and inversely proportional to the distance (and fix the charge unit so that Coulomb’s
constant is 1). Then the electric potential of a charge e is just e ln r. If C − U is filled with a
conductor material and we put a positive unit charge at p, so that the negative charges in C − U
move and attain an equilibrium position, the electric potential is just the Green’s function of U
with pole at p.
Theorem: If the boundary ∂U is regular, then the Green’s function with pole at p exists.
Proof: We may assume that, in the local coordinate z, the unit disk D̄ ⊂ U . To construct −g
with Perron method we consider the family F of functions u ∈ C(Ū − p) such that
13.1. DIRICHLET PROBLEM 373
1. u = 0 on ∂U , and u is subharmonic on U − p.
2. u + ln |z| is subharmonic in a neighborhood of p.
This family F is not empty, because it has the function

(
− ln |z| when |z| ≤ 1
u=
0 otherwise
Let us see that the functions u ∈ F are uniformly bounded on U − Dε , ε < 1:

Let h be the continuous function on Ū − Dε , harmonic on U − D̄ε , such that h = 1 on ∂Dε
and h = 0 on ∂U , and let uε be the maximum of u on ∂Dε . Applying the maximum principle
to the subharmonic function u + ln |z| on the unit disk D, we obtain
uε + ln ε ≤ u1 + ln 1 = u1 .
Moreover, uε h is harmonic on U − D̄ε and u ≤ uε h on the boundary; hence u ≤ uε h on

U − Dε . In particular u1 ≤ uε h1 . Hence uε + ln ε ≤ uε + h1 , and we conclude that
− ln ε
max {u(x)} = uε ≤ ·
x∈U −Dε 1 − h1
Now the proof of the Perron method shows that the function −g(x) := supu∈F {u(x)} is
harmonic on U − p, and may be continuously extended by 0 on ∂U .
Finally, let us see that p is a removable singularity of the harmonic function −g + ln |z|:
If u ∈ F, then on Dε we have
− ln ε −h1 ln ε
u + ln |z| ≤ uε + ln ε ≤ + ln ε =
1 − h1 1 − h1
so that −g + ln |z| is bounded above on Dε . On the other hand, the function
(
ln ε − ln |z| when |z| ≤ ε
u=
0 otherwise
is in the family F; hence u ≤ −g, so that ln ε ≤ −g + ln |z| on Dε , and −g + ln |z| also is bounded
below on Dε . We conclude that p is a removable singularity of −g + ln |z|.
Lemma: Let U ⊂ X be a connected open set with regular boundary, and g the Green’s function
on U with pole at p ∈ U . If H 1 (U, R) = 0, then there exists f ∈ O(U ) such that |f | = eg .
Proof: Given an harmonic function u, let us denote ũ an analytic function (it exists locally) such
that u = Re(ũ). The equation |f | = eg admits the local solution f = eg̃ at any point, except
at p, where locally we have g = u + ln |z| for some harmonic function u, and a local solution
is f = zeũ . Moreover, any two local solutions have equal modulus, so that both differ in a of
modulus 1. That is to say, the etalé space of the sheaf of solutions of |f | = eg is a principal
covering of U of group T = {z ∈ C : |z| = 1}, and we conclude if we show that it is a trivial
covering (i.e. it admits global sections).
Now, the principal coverings of U of group T are classified (p. 345) by H 1 (U, T), and the
cohomology exact sequence of the following exact sequence of sheaves shows that H 1 (U, T) = 0,
because H 1 (U, R) = 0 and H 2 (U, Z) = 0 (p. 350),
2π eti
0 −→ Z −−−−→ R −−−−→ T −→ 0.
Proposition: Let U ⊂ X be a connected open with regular boundary. If H 1 (U, R) = 0, then U

is isomorphic to the unit disk D.
Proof: Let g be the Green’s function with pole at p ∈ U , and fix f ∈ O(U ) such that |f | = eg .
Since eg = 1 on ∂U , by the maximum principle we have a commutative square
f
U /D
|−|

eg / [0, 1]
Ū
which shows that f −1 (Dr ), r < 1, is compact; hence f is a proper map, of degree 1 because f
has a unique simple zero at p. It is an isomorphism (p. 358).
Lemma: Let 1
S X be a non-compact Riemann surface such that H (X, R) = 0. There is a countable
cover X = n Un by connected open sets Un such that
1. Ūn is a compact manifold with boundary and Ūn ⊂ Un+1 .
2. H 1 (Un , R) = 0.
Proof: Fix a point p ∈ X and a proper map f : X → R (p. 250). By Sard’s theorem there is an
increasing sequence of regular values an → ∞, and we may assume that f (p) < a0 .
Let Yn be the connected component of {x ∈ X : f (x) ≤ an } containing p, and let V1 , . . . , Vm
be the relatively compact connected components of X − Yn . We put (p. 94)
o
Kn = Yn ∪ V̄1 ∪ . . . ∪ V̄m , Un =K n .
The open sets Un fulfill the first condition. Let us see the second one:
The connected components of X − Kn have non-compact closure; i.e. the connected compo-
nents of X − Un are non-compact, so that Hc0 (X − Un , R) = 0.
The closed subspace exact sequence (by duality Hc1 (X, R)∗ = H 1 (X, R) = 0)
δ
0 = Hc0 (X − Un , R) −−−→ Hc1 (Un , R) −→ Hc1 (X, R) = 0
shows that Hc1 (Un , R) = 0, and again by duality H 1 (Un , R) = Hc1 (Un , R)∗ = 0.
Uniformization Theorem: Let X be a Riemann surface. If H 1 (X, R) = 0, then X is isomor-

phic to the complex plane C, the complex projective line P1 or the unit disk D.
In particular, X is simply connected.
S
Proof: If X is not compact, fix the cover X = n Un of the lemma and a local coordinate at a
point p ∈ U0 . By the above proposition, we have an isomorphism fn : Un → Drn , unique with
the conditions f (p) = 0, f 0 (p) = 1, which determine the radius rn .
Moreover rm < rn when m < n. Just apply Schwarz’s lemma to the composition
−1
fm fn
ϕ : Drm −−−−−→ Um −−−−→ Drn , ϕ(0) = 0, ϕ0 (0) = 1.
By Koebe’s theorem, the injective analytic functions fn : U0 → DR , where R = lim rn (and

DR := C when R = ∞), admit a subsequence converging to an injective analytic function
f : U0 → DR . Applying again Koebe’s theorem to such subsequence on U1 , we may extend f
13.2. FRÉCHET SPACES 375
to U1 , and so on. Finally we obtain an injective analytic function f : X → DR , so that X is

isomorphic to an open subset of DR ⊆ C.
By the Riemann mapping theorem, we conclude that X is isomorphic to C or D.
If X is compact, the exact sequence of the closed subspace shows that Hc1 (X − p, R) = 0
and, by duality, H 1 (X − p, R) = 0. Hence X − p ' C or X − p ' D by the non-compact case.
In the first case, the isomorphism extends to a homeomorphism X → P1 , which is an
isomorphism by the removable singularity theorem.
In the second case, the isomorphism X − p ' D is a bounded analytic function; hence it
extends to a (non constant) analytic function X → C. Absurd, X is compact.
Corollary: The fundamental group of any Riemann surface X is torsion free.
Proof: The universal covering X̃ → X is a Galois covering of group π1 (X).

If H ⊆ π1 (X) is a finite subgroup, then π1 (X̃/H) = H, and by Hurewicz’s theorem
H 1 (X̃/H, R) = Homgr (π1 (X̃/H), R) = Homgr (H, R) = 0.
We conclude that X̃/H is simply connected, so that H = 0.
Note: If X̃ → X is the universal covering of a Riemann surface, then X = X̃/G, where G is

a group of automorphisms of X̃ without fixed points. If X̃ = P1 , then X = P1 , because any
homography has a fixed point. If X̃ = C then G is a discrete group of translations; hence (p.
299) X is C, a cylinder C/Ze or a torus C/(Ze1 + Ze2 ). In particular π1 (X) is abelian.
The universal covering of any other Riemann surface is D.
Picard Little Theorem: Any non constant analytic function f : C → C omits at most one
value.
Proof: The universal covering of C − {p, q} is D, because C − {p, q} is not compact and has non
abelian fundamental group. Since C is simply connected, any analytic function f : C → C−{p, q}
may be lifted to the universal covering
7D
f˜
π
f
C / C − {p, q}
and f˜: C → D is a bounded analytic function. Hence it is constant and so is f .
13.2 Fréchet Spaces

Let us consider the category of topological R-vector spaces and continuous R-linear maps.
1. If the origin of E is closed, then E is a separated space.

y−x
The diagonal is the inverse image of 0 by the continuous map E × E −−→ E.
2. If a linear map f is continuous at 0, then it is continuous.

Translations are continuous, and f (x) = f (x − e) + f (e).
3. The closure of a vector subspace also is a vector subspace.

4. Any neighborhood U of 0 is absorbing (if e ∈ E, then λe ∈ U for any λ sufficiently small)

and there is a neighborhood V of 0 such that V̄ ⊆ V + V ⊆ U .
The continuity of the product K × E → E and the addition E × E → E at 0.
5. There is a base of balanced (λU ⊆ U for any scalar |λ| ≤ 1) open neighborhoods of 0.
If U is a neighborhood of 0, by theScontinuity of the product, there is an open neighborhood
V of 0 and ε > 0 such that U 0 := |λ|<ε λV ⊆ U , and U 0 is a balanced open set.
6. A family F of continuous linear maps E → F is equicontinuous when for every neighborhood

V of 0 in F there is a neighborhood U of 0 in E such that f (U ) ⊆ V , ∀f ∈ F. In particular,
a family of linear forms on E is equicontinuous when there is a neighborhood U of 0 in E
such that every ω ∈ F is absolutely bounded by 1 on U .
7. Let B be the unit ball of a seminormed space E. A linear form ω : E → R is continuous if

and only if kωk = supB |ω(e)| < ∞, and so we obtain a norm on the topological dual E 0 .
Proposition: If V is a vector subspace of a topological vector space E, then Ē = E/V , with the
quotient topology, is a topological vector space and the canonical map π : E → Ē is an open map
such that any continuous linear map vanishing on V uniquely factors through π. The space Ē
is separated if and only if V is closed. If the topology of E is defined by a family of seminorms
{qi }, then the topology of Ē is defined by the seminorms q̄i (ē) = inf v∈V qi (e + v).
Proof: The map π is open because π −1 (πU ) = U + V = v∈V v + U is open when so is U .

S
Hence E × E → Ē × Ē and R × E → R × Ē are open surjective continuous maps, so that
the addition Ē × Ē → Ē and product R × Ē → Ē are continuous because so are the induced
maps E × E → Ē and R × E → Ē. The other statements are easy to check.
Definition: A topological vector space E is Fréchet when it is complete (hence separated)

and the topology may be defined by aPcountable family {qn } of seminorms (hence also by the
translation-invariant metric d(x, y) = n 2−n min{1, qn (y − x)}).
Replacing qn by max{q0 , . . . , qn }, we may always assume that qn−1 ≤ qn .
1. If V is a closed vector subspace of a Fréchet space E, then Ē = E/V also is Fréchet. In fact,
the translation-invariant metric d̄(x̄, ȳ) = inf u,v∈V d(x+u, y +v) defines the uniform structure
of Ē. Now, if (x̄n ) is a Cauchy sequence in Ē, we may assume that d(x̄ ¯ n−1 , x̄n ) < n+1
1
, and
2
1 1
fix xn ∈ E such that x̄n = π(xn ), d(xn−1 , xn ) < 2n . Hence d(xn , xn+m ) < 2n , ∀m ≥ 0, so
that xn → e ∈ E, and x̄n → π(e).
2. C(X) when X is a σ-compact space (p. 114), C m (X) when X is a smooth manifold (p. 251),
Γ(X, E) when E is a smooth vector bundle over a smooth manifold X (p. 349), and O(X)
when X is a Riemann surface (p. 114) are Fréchet spaces.
3. Let E be a Fréchet space, K a compact space. The vector space of continuous maps
C(K, E) is Fréchet with
the topology of uniform convergence, defined by the seminorms
q(f ) = maxx∈K q f (x) , where q runs over the continuous seminorms on E.
fn fn+1
Definition: An exact sequence . . . → En −→ En+1 −−−→ En+2 → . . . of continuous linear
maps between topological vector spaces is topologically exact if the induced continuous linear
bijections En /Ker fn = Ker fn+1 are isomorphisms (i.e. open maps).
Lemma: Let φ : E → F be a continuous linear map. If E is Fréchet and, for every neighborhood
U of 0 in E, the closure φ(U ) is a neighborhood of 0 in F , then φ is open.
Proof: As U varies, U + U runs over a base of neighborhoods

of 0 in E; hence it is enough to
show that any point y ∈ φ(U ) is contained in φ U + U .
Let {Un } be a base of neighborhoods of 0 in E such that U0 = U and Un + Un ⊆ Un−1 ,
Un + . . . + Un+k ⊆ Un−1
U0 + U1 + . . . + Uk ⊆ U + U
Since y − φ(U1 ) is a neighborhood of y, it intersects φ(U0 ) = φ(U ) at some point y0 , so that

y − y0 ∈ φ(U1 ). Recursively, we construct a sequenceP yn in F such that yn = φ(xn ) ∈ φ(Un ) and
y − (y0 + . . . + yn ) ∈ φ(Un+1 ). Now the series x = n xn converges, because E is complete and
P
xn + . . . + xn+k ∈ Un−1 , and we have
P x ∈ U + U . Finally, y = n yn because φ(Un ) is a base of
neighborhoods of 0 in F , and y = n φ(xn ) = φ(x).
Open Mapping Theorem: Any exact sequence of continuous linear maps between Fréchet
spaces is topologically exact.
Proof: Since En /Ker fn and Ker fn+1 are Fréchet, it is enough to prove that any continuous
linear bijection φ : E → F between Fréchet spaces is open.
If U is a neighborhood of 0 in E, then
S S
F = φ(E) = φ nU = nφ(U ).
n n
Baire theorem states that some nφ(U ) = nφ(U ) contains an open set nV , so that the closure
of φ(U − U ) = φ(U ) − φ(U ) contains the neighborhood V − V of 0.
As U varies, U − U runs over a base of neighborhoods of 0. We conclude by the lemma.
Closed Graph Theorem: Let f : E → F be a linear map between Fréchet spaces. If the graph
Γf = { e, f (e) ; e ∈ E} is closed in E × F , then f is continuous.
Proof: The first projection π1 : Γf → E is a continuous linear bijection; hence an isomorphism,

and f = p2 ◦ p−1
1 is continuous.
Theorem: If K is a compact topological space, the functor HomTop (K, −) preserves exact se-
quences of Fréchet spaces.
Proof: If E → Ē is a separated quotient of a Fréchet space E, we have to show that the induced
map π : C(K, E) → C(K, Ē) is surjective. Let B the open unit ball of a continuous seminorm q
on E, and B̄ the open ball of the induced seminorm q̄ on Ē.
Let f ∈ C(K, B̄). Given ε > 0 and a S continuous seminorm q 0 on E, since f is uniformly
K = i Ui and points ti ∈ Ui such that q̄ 0 f (t)−f (ti ) < ε,

continuous there is a finite open cover

∀t ∈ Ui . Moreover, since q̄ f (ti ) < 1, we have f (ti ) = ēi , where q(ei ) < 1.
P
Let {φi } be a partition of unity subordinated to {Ui }. We have q i φi ei < 1 and
q̄ 0 f (t) − i φi (t)ēi = q̄ 0
P P
i φi (t)(f (t) − f (ti ) < ε,
since we can restrict the sum to those i such that t ∈ Ui . Hence C(K, B̄) ⊆ π(C(K, B)).
By the lemma π is open, and therefore surjective.
Banach-Steinhaus Theorem: A family F of continuous linear maps from a Fréchet space

E to a topological vector space F is equicontinuous if and only if for every x ∈ E the set
F(x) = {f (e); f ∈ F } is bounded in F . (A set B ⊂ F is bounded if for every neighborhood V

of 0 in F there is a scalar λ 6= 0 such that λB ⊆ V ).
Proof: If F is equicontinuous, there is a neighborhood U of 0 in E such that f (U ) ⊆ V , ∀f ∈ F.

Now, given x ∈ E, we have λx ∈ U for some λ 6= 0; hence λF(x) = F(λx) ⊆ V . T
Conversely, let V be a closed neighborhood of 0 in F . Then U := F −1 (V ) = f ∈F f −1 (V )
is a closed set, and it is absorbing when F(x) is bounded, ∀x ∈ E. By Baire theorem, U has
non-empty interior; so that F −1 (V − V ) contains the neighborhood U − U of 0 in E.
As V varies, F −1 (V − V ) runs over a base of neighborhoods of 0 in F , and we conclude.
Theorem: Let E be a topological vector space. If the kernel of a linear form ω : E → R is

closed, then ω is continuous.
Proof: Replacing E by E/Ker ω, we may assume that E is separated of dimension 1 and ω is

the inverse of a linear bijection R → E, λ 7→ λe.
Given ε > 0, there is a balanced neighborhood U of 0 such that εe ∈
/ U . Hence U ⊆ (−ε, ε)e,
so that ω(U ) ⊆ (−ε, ε) and we conclude that ω is continuous at 0.
Corollary: Any separated topological vector space of finite dimension n is isomorphic to Rn .
Proof: By induction on n, and it is obvious when n = 0. If n > 0, the kernel N of any linear
form 0 6= ω : E → R is separated, of dimension n − 1; hence N ' Rn−1 is complete, so that it is
closed in E, and ω is continuous. We conclude that any linear bijection E → Rn is continuous,
and the inverse also is continuous, because so is any linear map Rn → E.
Corollary: Any finite dimensional subspace of a separated topological vector space is closed.
Corollary: A separated topological vector space E is finite dimensional if and only if some
neighborhood P of 0 is precompact.
Proof: If a vector subspace H ⊂ Rn does not contain a bounded set P , then H + 21 P does not
contain P (in Rn /H it is clear that 12 P does not contain P ).
Hence, if E were infinite dimensional, we could construct a sequence xn in P such that
/ hx1 , . . . , xn−1 i + 12 P , so that xn − xm ∈
xn ∈ / 12 P whenever n 6= m. Absurd since (xn ) has an
adherent point in the completion Pb and 21 P is a neighborhood of 0.
Lemma: Let C be a convex set in a topological vector space E. If x is an interior point of C

and y ∈ C̄, then the segment [x, y) is interior to C.
o
λ
Proof: Put U =C, and let us show that λx + (1 − λ)y ∈ U , 0 < λ < 1. Now, y + 1−λ (x − U ) is
open and contains y (since x ∈ U ); hence it contains z ∈ C (since y ∈ C̄). Since C is convex,
λ
λU + (1 − λ)z ⊆ U , and λx + (1 − λ)y ∈ λU + (1 − λ)z because z ∈ y + 1−λ (x − U ).
Hahn-Banach Theorem: Let E be a topological vector space, U a convex open set, V a linear
subvariety not meeting U . There is a closed hyperplane containing V and not meeting U .
Proof: We may assume that V is a vector subspace. Let W be a maximal vector subspace
containing V and not meeting U (it exists by Zorn’s lemma). W is closed, because W̄ is a vector
subspace not meeting U , and we must show that dim E/W > 1 is absurd.
Replacing E by SE/W we may assume that V = W = 0 and that E ' R2 .
Replacing U by λ>0 λU , we may assume that U is a cone (λU ⊆ U for every λ > 0).
Pick 0 6= y ∈ ∂U (it exists, R2 − {0} is connected). Since U is an open cone, λy ∈ ∂U when

λ > 0; hence λy ∈/ U.
Moreover λy ∈ / U when λ < 0, since otherwise by the lemma 0 ∈ [λy, y) ⊆ U .
We conclude that Ry ∩ U = ∅, against the maximal character of W = 0.
Hahn-Banach Theorem: Let V be a vector subspace of a seminormed space E. Any linear

form ω on V of seminorm ≤ 1 is the restriction of a linear form on E of seminorm ≤ 1.
Proof: The linear subvariety L = {v ∈ V : ω(v) = 1} does not meet the unit ball B of E, a
convex open set. Hence there is a hyperplane ω 0 (x) = 1 containing L (so that ω is the restriction
of ω 0 ) and not meeting B (so that kω 0 k ≤ 1).
Corollary: If 0 → V → E → W → 0 is a topologically exact sequence of locally convex spaces,

then the induced sequence 0 → W 0 → E 0 → V 0 → 0 is exact, and every equicontinuous set in V 0
is the image of an equicontinuous set in E 0 .
Proof: A set F of linear forms on a locally convex space F is equicontinuous if and only if there
is a continuous seminorm q on F such that every ω ∈ F has seminorm ≤ 1.
Corollary: Let E be a separated locally convex space. Any vector subspace V ⊆ E of finite
dimension admits a topological supplement (a vector subspace W such that the natural map
V × W → E is an isomorphism).
Proof: The inclusion V ,→ E admits a continuous retraction E → V .
Theorem: Let E be a separated locally convex space, e ∈ E. If ω(e) = 0, ∀ω ∈ E 0 , then e = 0.
Corollary: Let V be a vector subspace of a locally convex space E, and e ∈ E. If ω(e) = 0 for
any continuous linear form ω ∈ E 0 vanishing on V , then e ∈ V̄ .
13.2.1 Duality
When we endow a real vector space E with the weak topology of a certain vector subspace
E 0 ⊆ E ∗ (the initial topology of the maps ω : E → R, ω ∈ E 0 , defined by the seminorms
qω (e) = |ω(e)|, ω ∈ E 0 ) then any continuous linear form ω on Ew is bounded on some vector
subspace V = hω1 , . . . , ωn io , where ωi ∈ V . Hence ω(V ) = 0 and ω ∈ hω1 , . . . , ωn i ⊆ E 0 , so that
E 0 is the topological dual of Ew .
From now on E, E 0 will be a pair of real vector spaces with a non-singular bilinear pairing
E ×E 0 → R, identifying each factor with a space of linear forms on the other one. Then we endow
both factors with the weak topology of the pointwise convergence on the other one, so obtaining
separated locally convex spaces1 Ew , Ew 0 in a fully symmetric situation: any statement has an
equivalent dual statement, interchanging E and E 0 . Moreover, we have on E a finer locally

convex topology such that the topological dual is E 0 , so that weakly closed sets are closed,
compact sets are weakly compact, continuous linear maps are weakly continuous, etc.
Definition: A disk D ⊆ E is a convex balanced set, and the polar of D is the disk
Do = {ω ∈ E 0 : |ω(D)| ≤ 1} = {ω ∈ E 0 : ω(D) ≤ 1} ⊆ E 0 .
When V ⊆ E is a vector subspace, V o = {ω ∈ E 0 : ω(V ) = 0} is just the incident.

1 0
Our basic objects of study Ew , Ew are not complete nor metrizable spaces even in the most simple cases.
Any absorbing disk D ⊆ E defines a seminorm qD (x) = inf{|λ| : x ∈ λD} on E, and the unit
ball B of a seminorm q on E is an absorbing disk such that q = qB .
Lemma: If U, V are disjoint open convex sets, some closed hyperplane separates U and V .
Proof: By the Hahn-Banach theorem, some closed hyperplane ω(x) = 0 does not meet U − V .
Hence ω(U ) and ω(V ) are disjoint intervals in R, and we conclude.
Theorem: Any closed convex subset C of E is an intersection of closed half-spaces (subsets

ω(x) ≥ a, where ω ∈ E 0 , a ∈ R). In particular C is weakly closed.
Proof: If p ∈
/ C, there is a convex neighborhood U of 0 such that p + U and C + U do not meet.
By the lemma there is a closed half-space H such that C ⊆ H, p ∈ / H.
Corollary: Any closed disk in E is weakly closed.
Bipolar Theorem: If D is a weakly closed disk in E, then D = (Do )o . Hence, the polarity
defines a lattice anti-isomorphism

(Weakly) closed ∼ Weakly closed
−−−→
disks in E disks in E 0 .
where closed vector subspaces of E correspond to weakly closed vector subspaces of E 0 , and disked
closed neighborhoods of 0 in E correspond to equicontinuous weakly closed disks in E 0 .
Proof: If p ∈
/ D, some closed hyperplane ω(x) ≤ 1 contains D and not p; hence p ∈ / (Do )o .
Finally, a family F ⊆ E 0 is equicontinuous if and only if there is a disked neighborhood D
of 0 such that F ⊆ Do .
Corollary: The topology of E is the topology of uniform convergence on the equicontinuous

subsets of E 0 (or the equicontinuous weakly closed disks in E 0 ).
Proof: If D = (Do )o is a disked closed neighborhood of 0, then qD (x) = supω∈Do |ω(x)|.
Theorem: Any equicontinuous family in E 0 has compact closure in Ew 0 . Hence, the topology of
E is a locally convex topology intermediate between the weak topology and the weakly compact-
convergence topology. Moreover, the dual of E for any such topology is E 0 .
Proof: If D is a disked closed neighborhood of 0 in E, we have to show that
F = {ω ∈ E 0 : |ω(D)| ≤ 1} = {ω ∈ E ∗ : |ω(D)| ≤ 1}
has compact closure in Ew 0 . Now, F(x) = {ω(x); ω ∈ F } ⊂ R is bounded (D is absorbing) and

∗
Q
F = {ω ∈ E : |ω(D)| ≤ 1} is closed in the compact space x∈E F(x) ⊂ C(Edis , R) (with the
pointwise convergence topology). Hence F is compact.
Finally, if ω ∈ E ∗ is continuous for the weakly compact-convergence topology, there is a
weakly compact disk K ⊂ Ew 0 ⊆ E ∗ such that |ω(K o )| ≤ 1.
w
By the bipolar theorem, applied to Ew ∗ , we have ω ∈ (K o )o = K ⊂ E 0 .
Example: Let E be a Fréchet space. By the Banach-Steinhaus theorem, any weakly bounded
(hence any weakly compact) set in E 0 is equicontinuous, so that the topology of E is the weakly
compact-convergence topology.
13.2.2 The Transpose Linear Mapping

Let E, F be separated locally convex spaces. Any weakly continuous linear map f : E → F
induces a weakly continuous linear map f ∗ : F 0 → E 0 , f ∗ (ω) = ω ◦ f , and f ∗∗ = f : E → F .
Definitions: A continuous linear map f : E → F is a homomorphism when the induced

continuous linear bijection E/Ker f → Im f is an isomorphism (when E/Ker f is endowed with
the quotient topology and Im f with the topology induced by F ).
Proposition: The functors E Ew and E Ew0 preserve topologically exact sequences of
separated locally convex spaces: If V ⊆ E is a closed vector subspace, we have topologically exact
sequences
i π
0 −→ Vw −−→ Ew −−→ (E/V )w −→ 0
π∗ i∗
0 −→ (E/V )0w −−−→ Ew
0
−−→ Vw0 −→ 0
Proof: The topology of Vw is induced by the topology of Ew because E 0 → V 0 is surjective.

If ω ∈ E 0 , then the seminorm qω (e) = |ω(e)| induces on E/V the null seminorm when
ω(V ) 6= 0, and the seminorm qω̄ when ω(V ) = 0, where ω̄(ē) = ω(e). Hence, the weak topology
of E/V is just the quotient topology of Ew .
The topology of (E/V )0w is induced by the topology of Ew 0 because E → E/V is surjective.
If e ∈ E, then the seminorm qe (ω) = |ω(e)| induces on V 0 the null seminorm when e ∈ / V
o
(because inf η∈V o |(ω + η)(e)| = 0, since η(e) 6= 0 for some η ∈ V ), and the seminorm qe when
e ∈ V . Hence, the weak topology of V 0 is just the quotient topology of Ew 0 .
Proposition: Let f : Ew → Fw be a continuous linear map. The closure of Im f in Fw is the

incident of Ker f ∗ , and Ker f is the incident of Im f ∗ :
w w
Im f = (Ker f ∗ )o , Ker f = (Im f ∗ )o = (Im f ∗ )o .
In particular, f is injective if and only if Im f ∗ is dense in Ew

0 .
w
Proof: It is clear that Ker f = (Im f ∗ )o ; hence Ker f ∗ = (Im f )o = (Im f )o .
w
By the bipolar theorem we conclude that Im f = (Ker f ∗ )o .
Proposition: A continuous linear map f : Ew → Fw is a homomorphism if and only if Im f ∗ is

0 . Hence, if Im f ∗ = E 0 , then f is an injective weak homomorphism.
closed in Ew
Proof: Replacing Ew by (E/Ker f )w = Ew /(Ker f )w , and Fw by (Im f )w , we may assume that

f is bijective, an obvious case (if Im f ∗ is closed, then Im f ∗ = 0o = E 0 ).
Proposition: A continuous linear map f : E → F is a homomorphism if and only if it is a

weak homomorphism (Im f ∗ is weakly closed in E 0 ) and every equicontinuous family in Im f ∗ is
the image by f ∗ of an equicontinuous family in F 0 .
Proof: Replacing Ew by (E/Ker f )w = Ew /(Ker f )w , and Fw by (Im f )w , we may assume that

f is bijective. Now, two (separated) locally convex topologies on a vector space are identical if
and only if they have the same dual and equicontinuous families in the dual.
Definition: A linear map f : E → F is compact if there is a neighborhood U of 0 in E such

that f (U ) has compact closure in F ; a fortiori it is continuous, since for every neighborhood V
of 0 in F there is an integer n such that f (U ) ⊆ nV , i.e. f n1 U ⊆ V .
Lemma: If i : E → F is an isomorphism onto a closed subspace of F and f : E → F is compact,

then h = i + f is a homomorphism onto a closed subspace, with finite dimensional kernel.
Proof: Put N = Ker h and let U be a disked neighborhood of 0 such that f (U ) has compact
closure. Then V = N ∩ U is a neighborhood of 0 in N such that
i(V ) = −f (V ) ⊆ −f (U ) = f (U )
has compact closure, and therefore is precompact. Since i is an isomorphism, V is precompact,

and N admits a precompact neighborhood of 0: it is finite dimensional.
To prove that f is a homomorphism onto a closed subspace of F , we may restrict i and f to a
topological supplement of N (it exists because N is finite dimensional), so that we may assume
that h is injective, and we have to prove that h is an isomorphism onto a closed subspace.
This means that any ultrafilter2 m in E converges whenever h(m) converges in F : the
convergent ultrafilters determine the neighborhoods of 0 because (p. 217) any filter is an inter-
section of ultrafilters, and if p ∈ Ē, any ultrafilter containing the intersections of E with the
neighborhoods of p converges to p.
Let q be the seminorm of U . Then limm q(x) = a ∈ [0, ∞] exists because [0, ∞] is a compact
separated space. If a < ∞, there exists A ∈ m such that A ⊆ (a + 1)U , then f (A) has compact
closure, so that limm f (x) exists, and so does limm i(x) = limm h(x) − limm f (x) = y ∈ F because
limm h(x) exists by hypothesis. But i(E) being closed, we have y = i(e), e ∈ E, and m converges
to e because i : E → i(E) is an isomorphism.
But a = ∞ is impossible:
x
= limp h(x)

Otherwise m comes from an ultrafilter p in {q(x) > 0} and limp h q(x) q(x) = 0.
x x

Moreover, limp q q(x) = 1, and by the preceding argument, limp q(x) = e ∈ E exists.
x x

Now h(e) = limp h q(x) = 0, q(e) = limp q q(x) = 1; absurd, h is injective.
Proposition: If f : E → F is compact, so is f ∗ : Fc0 → Ec0 when we endow the duals with the
compact convergence topology.
Proof: Let D be a closed disked neighborhood of 0 in E such that the disk K = f (D) has compact
closure, then f ∗ (K o ) ⊆ Do . Now, K o is a neighborhood of 0 in Fc0 , and Do is weakly compact
in E 0 . Since the family Do is equicontinuous, Do also is compact for compact convergence: Do
is compact in Ec0 .
Schwartz Theorem: Let p : E → F be a surjective linear map between Fréchet spaces. If

f : E → F is compact, then h = p + f is a homomorphism onto a closed subspace of finite
codimension in F .
Proof: Let Ec0 , Fc0 be the duals with the compact convergence topology (their duals are then E
and F , because compact sets are weakly compact), so that f ∗ is compact.
Im p∗ is closed because p is a homomorphism; hence a weakly homomorphism.
Any compact disk in F is the image by p∗∗ = p of a compact disk in E (because any compact
set is the image of a compact set, p. 377). Since the equicontinuous sets in E are just the subsets
of the compact disks, we see that the injective continuous linear map p∗ is an isomorphism onto
2
An ultrafilter in a set X is a maximal (or prime, since P(X) is a Boolean algebra) ideal m of the algebra
P(X) of subsets of X. Given a map f : X → Y , then f (m) := {B ∈ P(Y ) : f −1 (B) ∈ m}, is an ultrafilter in Y and
(when Y is a separated topological space) limm f (x) denotes the limit, if it exists, of f (m); i.e. limm f (x) = y ∈ Y
means that for every neighborhood V of y we have f (A) ⊆ V for some A ∈ m.
13.3. COMPACT RIEMANN SURFACES 383
a closed subspace of Ec0 and, by the lemma, h∗ = p∗ + f ∗ is a homomorphism onto a closed

subspace, with finite dimensional kernel.
The image of h is closed in F because h∗ is a (weak) homomorphism, and of finite codimension
w
because Im h = Im h = (Ker h∗ )o .
Finally, h is a homomorphism by the open mapping theorem.
13.3 Compact Riemann Surfaces

Let ϕ : X → Y be an analytic map between Riemann surfaces, x ∈ X, and y = ϕ(x) ∈ Y .
When ϕ is not constant, there are (p. 267) coordinate neighborhoods (U, u) and (V, v)
centered at x and y respectively, such that ϕ has the normal form v = un , where n is the
ramification index of ϕ at x. Hence, if OX,x denotes the local ring of germs at x of analytic
functions, then indx ϕ = dim (OX,x /my OX,x ), where my is the maximal ideal of the local ring
OY,y , y = ϕ(x).
When f : X → P1 is a non null meromorphic function, indx f is the number of zeros of f at
x, when f (x) = 0, and the number of poles when f (x) = ∞.
Theorem: Let ϕ : X → Y be a non constant analytic map. If X is compact, then Y is compact,

ϕ is surjective, with finite fibres, and the number
P
d(y) = indx ϕ
ϕ(x)=y
does not depend on y ∈ Y , and it is named degree of the morphism ϕ.
Proof: Since ϕ(X) is a compact open set in Y , we have that Y = ϕ(X) is compact.
The fibres are discrete; hence finite because X is compact.
Put ϕ−1 (y) = {x1 , . . . , xn }, ni = indxi ϕ, and fix coordinate neighborhoods U1 , . . . , Un , V
centered at x1 , . . . , xn , y where ϕ : Ui → V has normal form v = uni i (to obtain the normal form,
only a change of coordinate in Ui is required, not in V ), so that the fibre of ϕ : Ui → V over any
point y 0 ∈ V , y 0 6= y, is formed by ni points with ramification index 1.
S
We may assume that U1 , .P . . , Un are disjoint and, replacing V by Y − ϕ(X − i Ui ), that
ϕ−1 (V ) = qi Ui . Now d(y 0 ) = i ni = d(y), so that d is locally constant.
The Main Example: If X is a C-scheme, then Xan will be the set of rational points, so that
any f ∈ OX (U ) defines a map fan : Uan → C, and we endow Xan with the initial topology of
the maps fan when U runs over the open subsets and f ∈ OX (U ); hence, any morphism of
C-schemes ϕ : X → Y induces a continuous map ϕan : Xan → Yan .
Let C be the Riemann variety of a finite extension Σ of C(t), and C[x1 , . . . , xn ]/I the ring of
some affine open neighborhood U of a closed point p ∈ C. Since OC,p is a regular ring, we may
assume that the ideal I is generated by n − 1 polynomials with linearly independent differentials
at any point of the closed set Uan ⊂ Cn . Hence Uan is a closed smooth surface and Tp∗ Uan is a
complex vector subspace of Tp∗ Cn , isomorphic to mp /m2p , the algebraic cotangent space. So we
see that the topological space Can has a natural smooth structure with a complex structure on
the cotangent spaces.
Any generator z of mp has non-null differential in a neighborhood V ⊂ Can , and dx z is
C-linear at any point x ∈ V . Hence, z is a local isomorphism (preserving the complex structure
of the tangent spaces) so that Can is a Riemann surface and z is a local coordinate at p.
Finally, any morphism C(t) → Σ defines a morphism of C-schemes C → P1 ; hence an
analytic map ϕ : Can → (P1 )an such that the number of points of the fibres ϕ−1 (t) (counted
with the ramification index) is constant (p. 313). Hence, any point t ∈ (P1 )an has a compact
neighborhood K such that ϕ−1 (K) is compact, and we conclude that Can is a compact Riemann
surface. We shall see that there are no other compact Riemann surfaces, but first let us extend
some algebraic constructions used in the course Algebraic Geometry I:
Definitions: A divisor on a Riemann surface X is a function D : X → Z with support on a

discrete closed set, hence finite if X is compact. The divisors form an abelian group Div(X),
0 0
P
and we putP D ≤ D when D(x) ≤ D (x), ∀x ∈ X. The degree of a divisor D = x∈X nx · x is
deg D = x nx , whenever the sum only has a finite number of non null terms.
Any meromorphic function 0 6= f has a discrete set of zeros and poles and the divisor of f
is defined to be
X X X
D(f ) = vx (f ) · x := (indx f ) · x − (indx f ) · x
x∈X f (x)=0 f (x)=∞
and two divisors D, D0 are linearly equivalent, and we put D0 ∼ D, when D0 = D + D(f ) for
some meromorphic function f 6= 0.
We have D(f h) = D(f ) + D(h), D(1/f ) = −D(f ) and D(f + h) ≥ min{D(f ), D(h)}.
Moreover, D(f ) ≥ 0 if and only if f ∈ OX (X).
Hence, when X is compact, any meromorphic function f is fully determined, up to a constant
factor, by the divisor D(f ), and moreover, deg D(f ) = 0 by the above theorem.
Definition: Any divisor D defines a subsheaf LD of the sheaf MX of meromorphic functions:
LD (U ) = {f ∈ MX (U ) : D|U + D(f ) ≥ 0} , by convention 0 ∈ LD (U ),

P
and it is a line OX -module. In fact, if D = x nx · x, and (U, z) is a coordinate neighborhood
centered at x such that ny = 0 at any point x 6= y ∈ U , then LD |U = z −nx OX |U .
Hence, LD(f ) = f1 OX , and LD0 ⊆ LD if and only if D0 ≤ D.
Moreover, it is easy to check that LD ⊗OX LD0 = LD+D0 and HomOX (LD , LD0 ) = LD0 −D .
Proposition: The line OX -modules LD and LD0 are isomorphic if and only if D and D0 are
linearly equivalent.
Proof: We may assume that D0 = 0. Now, if D = D(f ), then LD(f ) = f1 OX ' OX .

Conversely, if ϕ : OX → LD is an isomorphism, put f = ϕ(1). Then LD = f OX = L−D(f )
as subsheaves of MX , so that D = −D(f ).
Definitions: At any point x ∈ X the complex structure of Tx X defines a R-linear automorphism

J : Tp X → Tp X such that J 2 = −1, inducing C-linear automorphisms of Tx X⊗R C and Tx∗ X⊗R C.
A smooth 1-form ω is of type (1,0) when J(ω) = iω, and of type (0, 1) when J(ω) = −iω. The
1,0
sheaves of smooth 1-forms of type (1,0) and (0,1) are denoted ΩX and Ω0,1
X respectively. Both
∞
are modules over the sheaf CX of complex smooth functions; hence acyclic sheaves (p. 327).
If f ∈ C ∞ (X), the decomposition df = ∂f + ∂f¯ of df as a sum of a 1-form ∂f of type (1,0)
¯
and a 1-form ∂f of type (0,1) defines derivations
∞ 1,0 ∞ 0,1
∂ : CX −→ ΩX , ∂¯ : CX −→ ΩX .
In a coordinate open set (U, z), 1-forms of type (1,0) (resp. (0,1)) are just f dz (resp. f dz̄),
f ∈ C ∞ (U ). Hence,
∂f ¯ = ∂f dz̄
∂f = dz , ∂f
∂z ∂ z̄
and the Cauchy-Riemann equations state that we have an exact sequence of sheaves
∂¯
∞
0 −→ OX −→ CX −−−→ Ω0,1
X .
Note: Let Cc∞ (C) (resp. CK ∞ (C)) be the complex vector space of all complex smooth functions
ρ : C → C with compact support (resp. with support in a given compact set K ⊆ SC). Then
CK∞ (C) is a closed subspace of C ∞ (C), hence a Fréchet space, and we equip C ∞ (C) = ∞
c K CK (C)
∞
with the inductive limit topology. The space D of C-linear continuous maps Cc (C) → C, is the
space of complex distributions on C. Any point p ∈ C definesR a distribution, the Dirac delta
δp (ρ) = ρ(p), and any smooth function f also, Tf (ρ) = C f ρdxdy. Now ∂z̄ acts on D by the
transpose map, (∂z̄ T )(ρ) = −T (∂z̄ ρ), up to a sign introduced so that ∂z̄ Tf = T∂z̄ f .
1
The following lemmas state that a solution of the equation ∂z̄ f = δ0 is just f = πz , and that
1
the convolution product f = πz ∗ ρ is a solution of ∂z̄ f = δ0 ∗ ρ = ρ.
Z
1 ∂ρ(u)
Lemma: dū ∧ du = 2πiρ(z) , ρ ∈ Cc∞ (C).
C z − u ∂ ū
Proof: Let Cε be the boundary of the disk Dε of radius ε with center at z. Then
Z Z
1 ∂ρ(u) ∂ ρ(u) 1
dū ∧ du = dū ∧ du (because z−u is analytic)
C z − u ∂ ū C ∂ ū z−u
Z Z Z
ρ(u) ρ(u) Stokes ρ(u)
= d du = lim d du == − lim du
C z − u ε→0 C−Dε z − u ε→0 Cε z −u
Z 2π Z 2π
u=z+εeiθ ρ(z + εeiθ ) iθ
== lim d(εe ) = lim iρ(z + εeiθ ) dθ = 2πiρ(z).
ε→0 0 εeiθ ε→0 0
Lemma: If ρ ∈ Cc∞ (C), then a smooth solution of the equation ∂z̄ f = ρ is

Z
1 ρ(u)
f (z) = dū ∧ du.
2πi C z − u
Proof: Use polar coordinates u = z + reiθ centered at z. Then dū ∧ du = 2irdr ∧ dθ, and
Z Z
1 ρ(u) 1
f (z) = dū ∧ du = ρ(z + reiθ )e−iθ dr ∧ dθ.
2πi C z − u π C
Since ρ has compact support, we may differentiate under the integral sign; hence (using that
∂z̄ = ∂ū and the above lemma)
∂ρ(z + reiθ ) −iθ ∂ρ(u) dū ∧ du
Z Z
∂f 1 1
(z) = e dr ∧ dθ = = ρ(z).
∂ z̄ π C ∂ z̄ 2πi C ∂ ū z − u
Dolbeault Exact Sequence: The sheaf OX of analytic functions on a Riemann surface X
admits the acyclic resolution
∞ 0,1∂¯
0 −→ OX −→ CX −−−→ ΩX −→ 0.
Proof: Any germ of Ω0,1

X is the germ of a section with compact support.
Theorem: If E is a locally free OX -module, then H p (X, E) = 0, p ≥ 2.

∞ -modules;
Proof: Applying ⊗OX E to the above exact sequence, we obtain a resolution of E by CX
hence an acyclic resolution,
∞ 0,1
0 −→ E −→ CX ⊗OX E −→ ΩX ⊗OX E −→ 0.
Lemma: Let K ⊂ C be a compact set, and ω : C(K) → C a continuous linear form. Then the
1
function h(a) = ω z−a , a ∈ C − K, is analytic on C − K.
1 1 P (a−z0 )n
Proof: If a ∈ B(z0 , r) ⊂ C − K, then z−a = (z−z0 )−(a−z 0)
= n (z−z 0)
n+1 ·
The series converges uniformly on z ∈ K, because |a − z0 | < r ≤ |z − z0 |. Hence

∞ ∞
!
(a − z0 )n

1 X X 1
(∗) h(a) = ω =ω = ω (a − z0 )n .
z−a (z − z0 )n+1 (z − z0 )n+1
n=0 n=0
Lemma: Let K be a compact subset of an open set U ⊆ C. If no connected component of U − K

has compact closure in U , then the restriction morphism R(U ) → R(K) has dense image (where
R(Z) ⊆ C(Z) denotes the subalgebra of rational functions without poles on Z).
Proof: Since the image is a subalgebra, so is the closure, and it is enough to show (p. 379) that
1
ω z−a = 0, a ∈ C − K, for any continuous linear form ω : C(K) → C vanishing on R(U ).
1

Consider the function h(a) = ω z−a on C − K, and use that it is analytic:
1. h(a) = 0 when a is in the unbounded connected component of C − K.

1 P zn
If |a| > maxK |z|, then z−a = − n an+1 , and the series converges uniformly on K. Since
n n+1 1

ω(z /a ) = 0 and ω is continuous, we see that 0 = ω z−a = h(a).
2. h = 0 on any connected component W of C − K intersecting C − U .

/ U , then ω (z−z10 )n+1 = 0, and (∗) shows that h = 0 on a disk D ⊆ W .

If z0 ∈ W , z0 ∈
3. h = 0 on C − K. If W is a bounded connected component of C − K, then W̄ is compact,

so that W̄ intersects C − U (if W̄ ⊂ U , any connected component of U − K contained in W
would be relatively compact in U ), and W intersects C − U because ∂W ⊆ K.
Theorem: If U ⊆ C is an open set, then R(U ) is dense in O(U ).
Proof: Any compact subset of U is contained in a compact set K such that U − K has no
connected component with compact closure in U (p. 94) and it is enough to show that ω(f |K ) =
0, ∀f ∈ O(U ), for any continuous linear form ω : C(K) → C vanishing on R(U ).
1

By the lemma, ω z−a = 0, ∀a ∈ C − K.
Pick a smooth function ρ with compact support ⊂ U and such that ρ = 1 on a neighborhood
of K, so that ∂z̄ ρ has a compact support L not intersecting K.
Since ∂ū f = 0 and supp (∂ū ρ) = L, when z ∈ K we have
Z Z
1 1 ∂(f ρ) 1 f (u) ∂ρ(u)
f (z) = f (z)ρ(z) = (u) dū ∧ du = dū ∧ du.
2πi C z − u ∂ ū 2πi L z − u ∂ ū
Since K × L and K is compact, the integrand R(u, z) := h(u)

z−u is uniformly continuous; hence,
given ε > 0, each point a ∈ L has a neighborhood C where |R(z, u)−R(z, a)| < ε, ∀z ∈ K, u ∈ C,
so that we may find a finite partition L = qLn by measurable sets, and points an ∈ Ln , such
that the Riemann sums of R(u, z) uniformly approximate the integral; i.e. for every z ∈ K,
X X cn
(Vol L)ε > f (z) − R(an , z)(Vol Ln ) = f (z) − , cn = h(an )(Vol Ln ),

n n
z − an
P cn
so that the functions n z−a uniformly converge to f (z) in C(K).
P cn P n cn
Since ω n z−an = n ω z−an = 0, because an ∈ / K, we have ω f |K = 0.
Corollary: Let K be a compact subset of an open set U ⊆ C. If no connected component of U −K

has compact closure in U , then the restriction morphism R(U ) → O(K) has dense image (where
O(K) = {f ∈ C(K) : f may be extended to an analytic function on a neighborhood of K}).
Proof: If V is an open neighborhood of K, in C(K) we have

th. lem.
O(V ) ⊆ O(V ) === R(V ) ⊆ R(K) === R(U ).
Dolbeault-Grothendieck Lemma: If U ⊆ C is an open set, we have an exact sequence

∂¯
0 −→ O(U ) −→ C ∞ (U ) −−−→ Ω0,1 (U ) −→ 0.
S o
Proof: Put U = n Kn with Kn compact, Kn−1 ⊆ Un :=K n , and no connected component of
U − Kn with compact closure in U (p. 94).
We have to show that ∂z̄ : C ∞ (U ) → C ∞ (U ) is surjective. Pick f ∈ C ∞ (U ).
1. There exist un ∈ C ∞ (U ) such that ∂z̄ un = f on Un and |un+1 − un | < 2−n on Kn−1 .
If u1 , . . . , un are constructed, pick ũn+1 ∈ C ∞ (U ) such that ∂z̄ ũn+1 = f on Un+1 (it exists
because f coincides on Un+1 with some ρ ∈ C ∞ (U ) with compact support). Now ũn+1 − un
is analytic on Un , because ∂z̄ (ũn+1 − un ) = f − f = 0 on Un . Since no connected component
of U − Kn−1 has compact closure in U , there is an analytic function R(z) on U such that
|ũn+1 − un − R| < 2−n on Kn−1 . Put un+1 := ũn+1 − R.
2. The function u(z) = lim un (z) ∈ C ∞ (U ) is a solution of ∂z̄ u = f .

P∞
u = un + (um+1 − um ).
m=n
On Un , the right term is an absolutely convergent series of analytic functions; hence it con-
verges to an analytic function h, so that u = un + h is smooth on Un and
∂z̄ u = ∂z̄ un − ∂z̄ h = f + 0 = f .
Theorem: If U ⊆ C is an open set, then H 1 (U, OU ) = 0.
Corollary: H 1 (U, O∗ ) = 0; hence any line OU -module is trivial (p. 348).
Proof: H 1 (U, OU ) = 0 and H 2 (U, Z) = 0 (p.350). We conclude by the exact sequence of sheaves
exp
0 −→ Z −→ OU −−−→ OU∗ −→ 0.
Weierstrass Theorem: Any divisor in U ⊆ C is the divisor of some meromorphic function.
Corollary: Any meromorphic function f on U ⊆ C is the quotient of two analytic functions.
Proof: The divisor of poles of f is the divisor of some h ∈ O(U ). Then f h ∈ O(U ).
Definition: The principal parts at a point x of a Riemann surface X are the elements of the
quotient group MX,x /OX,x = { az nn + . . . + az1 }, where z is a local coordinate centered at x.
Mittag-Leffler Theorem: Given a principal part at any point of a discrete closed set in an
open set U ⊆ C, there is a meromorphic function on U with the prescribed principal parts and
no other pole.
Proof: We have an exact sequence 0 → M → O → P → 0, where P is the sheaf of principal

parts. Now take sections on U and use that H 1 (U, O) = 0.
13.3.1 Riemann-Roch Theorem

Let L be a free A-module, eventually of infinite rank, let {dxi }i∈I denote a base, and fix a good
order on the set of indices I. Then {dxi0 ∧ . . . ∧ dxip }i0 <...<ip is a base of Λp+1 L, and we have
an exact sequence of A-modules
i i i
. . . Λ3 L −−D
−→ Λ2 L −−D
−→ L −−D
P
−→ A −→ 0 , D= i ∂i .
In fact, Im iD ⊆ Ker iD becauseiD ◦ iD = 0, and Ker iD ⊆ Im iD because
iD (dx1 ∧ ω) = ω − dx1 ∧ (iD ω).

L
Let U = {Ui }i∈I an open cover of X. The stalks of L = i ZUi are free Z/modules; hence,
if dxi denotes a generator of the stalk of ZUi , we have an exact sequence
i i i
. . . Λ3 L −−D
−→ Λ2 L −−D
−→ L −−D
P
−→ Z −→ 0 , D= i ∂i .
The Cech complex of a sheaf F on X is Č • (U, F) = Hom(Λ• L, F):

L
Č p (U, F) = Hom
Q
ZUi0 ∩...∩Uip , F = F(Ui0 ∩ . . . ∩ Uip )
i0 <...<ip i0 <...<ip
p+1
d : Č p (U, F) −→ Č p+1 (U, F) , (−1)k si0 ...ibk ...ip+1 |Ui0 ∩...∩Uip+1 .
P
(ds)i0 <...<ip+1 =
k=0
and the Cech cohomology groups are Ȟ p (U, F) = H p Č • (U, F) .

Applying Hom(−, F) to the exact sequence ΛL → L → Z → 0 we obtain an exact sequence

d
0 → F(X) → Č 0 (U, F) −
→ Č 1 (U, F) so that
Ȟ 0 (U, F) = F(X).
Leray’s Theorem: If F is acyclic on any intersection Ui0 ∩ . . . ∩ Uip , then
Ȟ p (U, F) = H p (X, F) , p ≥ 0.
Proof: Consider an injective resolution 0 → F → I • and the bicomplexes with bounded diagonals
Hom(Z, I • ) −→ Hom(Λ• L, I • ) ←− Hom(Λ• L, F)
Now Hom(Z, I p ) −→ Hom(Λ• L, I p ) is a quasi-isomorphism because Λ• L → Z → 0 is an

exact sequence and Hom(−, I p ) is an exact functor.
By hypothesis F(U ) = Hom(ZU , F) → I • (U ) = Hom(ZU , I • ) is a quasi-isomorphism when
U = Ui0 ∩ . . . ∩ Uip ; hence Hom(Λp L, F) → Hom(Λp L, I • ) is a quasi-isomorphism.
By the bicomplex theorem,
H p (X, F) = H p Hom(Z, I • ) = H p Hom(Λ• L, I • ) = H p Hom(Λ• L, F) = Ȟ p (U, F).

Lemma: Let V ⊆ U be open sets in a Riemann surface X. If V has compact closure V̄ ⊆ U ,

then the restriction morphism OX (U ) → OX (V ) is compact.
Proof: Let B ⊂ OX (U ) be the unit ball of the seminorm defined by the compact set K = V̄ .
For any sequence (fn ) in B we have |f | ≤ 1 on V ; hence, by Montel’s theorem, some
subsequence uniformly converges on the compact subsets of V .
We conclude that the image of B in OX (V ) has compact closure.
Finiteness Theorem: Let X be a compact Riemann surface, and E a locally free OX -module of
finite rank. All the cohomology groups H p (X, E) are complex vector spaces of finite dimension.
Proof: Let U = {Ui } be a finite open cover of X by open sets Ui isomorphic to disks Dri such
that the disks Dri /2 also define an open cover V = {Vi } of X, and such that E|Ui ' OX n . By
Leray’s theorem, Ȟ p (U, E) = H p (X, E) = Ȟ p (V, F).

E(Ui0 ∩ . . . ∩ Uip ) ' OX (Ui0 ∩ . . . ∩ Uip )n inherit a natural Fréchet topology; hence so do
the vector spaces Č p (U) := Č p (U, E), and the linear maps d : Č p (U) → Č p+1 (U) are continuous
because so are the restriction maps, so that Z p (U) := Ker d is a Fréchet space. Moreover, the
linear maps Č p (U) → Č p (V) are compact by the lemma; hence so are Z p (U) → Z p (V).
(Case p = 0) The compact map Z 0 (U) → Z 0 (V) is bijective; hence the Fréchet space Z 0 (V) =
H 0 (X, E) admits a compact neighborhood of 0: it is finite-dimensional (p. 378).
(Case p = 1) The restriction morphism f : Z 1 (U) → Z 1 (V) is compact, and the cokernel of
d : Č 0 (V) → Z 1 (V) is just Ȟ 1 (V, E) = H 1 (X, E). Now h = d + f : Č 0 (V) ⊕ Z 1 (U) → Z 1 (V) is
surjective, because Ȟ 1 (U, E) = H 1 (X, E) = Ȟ 1 (V, E), and Schwartz theorem let us conclude.
Definition: We put hp (E) := dim H p (X, E), and g = h1 (OX ) is the genus of X.
Corollary: Meromorphic functions separate points on any compact Riemann surface X.
Proof: If x ∈ X, then dim Ox /mnx = n. When n > g, the exact sequence
0 −→ OX −→ Lnx −→ Ox /mnx −→ 0
shows that h0 (Lnx ) > 1; hence there is a meromorphic function with a unique pole at x.
Corollary: The topology of X is the initial topology of the meromorphic functions X → P1 .
Proof: The continuous bijection X → Xin is a homeomorphism because Xin is compact (so is
X) and separated, because the meromorphic functions separate points.
Corollary: If x ∈ X, there is a meromorphic function on X with a simple pole at x.
Proof: Since Ox /mx is flasque, we have h1 (Ox /mx ) = 0, and the exact sequence
0 −→ Lnx −→ L(n+1)x −→ Ox /mx −→ 0
shows that h1 (L(n+1)x ) ≤ h1 (Lnx ). Now the exact sequence
0 −→ OX −→ Lnx −→ Ox /mnx −→ 0
shows that h0 (Lnx ) = n − const, n 0. Hence h0 (L(n−1)x ) < h0 (Lnx ) < h0 (L(n+1)x ) when
n 0, so that there are meromorphic functions f, h ∈ M(X) with a poles at x of order n and
n + 1 respectively, and f /h has a simple pole.
Corollary: Any line sheaf L is isomorphic to the line sheaf of some divisor.
Proof: If we have an injective morphism L → OX , then OX /L is supported on a finite number

of points, so that it defines a positive divisor D and L ' L−D .
In the general case, we consider a point x ∈ X and the exact sequence
0 −→ L−nx −→ OX −→ Ox /mnx −→ 0.
After applying ⊗Lnx ⊗ L, the cohomology exact sequence
0 −→ H 0 (X, L) −→ H 0 (X, Lnx ⊗ L) −→ Ox /mnx −→ H 1 (X, L)
shows that h0 (Lnx ⊗ L) 6= 0 when n > h1 (L).

Hence there is a non null morphism OX → Lnx ⊗ L, and dualizing we obtain an injective
morphism L−nx ⊗ L∗ → OX . We conclude that L−nx ⊗ L∗ ' L−D and L ' LD−nx .
Lemma: Let L be a line sheaf such that h1 (L) 6= 0. For any line sheaf L0 we have a natural
injective morphism
HomOX (L0 , L) −→ HomC H 1 (X, L0 ), H 1 (X, L) .

Proof: If ϕ : L0 → L is non-null, then it is injective, and the exact sequence
0 −→ L0 −→ L −→ T −→ 0
(where T is flasque, because it has finite support) induces an exact sequence

ϕ
H 1 (X, L0 ) −−−→ H 1 (X, L) −→ H 1 (X, T ) = 0
and ϕ : H 1 (X, L0 ) → H 1 (X, L) is non-null, because H 1 (X, L) 6= 0.
Lemma: If Dn is a strictly increasing sequence of positive divisors, then h1 (LDn ) = 0, n 0.
Proof: The cohomology exact sequence of the exact sequence of sheaves
0 −→ LDn −→ LDn+1 −→ T −→ 0
where T is flasque (because of finite support) shows that h1 (LDn+1 ) ≤ h1 (LDn ), so that h1 (LDn )
stabilizes when n 0. Moreover, the exact sequences
0 −→ OX −→ LDn −→ ODn −→ 0
show that h0 (LDn ) → ∞ when n → ∞. The above lemma shows that h1 (LDn ) = 0, n 0.
Theorem: If X is a compact Riemann surface, H p (X, MX ) = 0, p ≥ 1.
Proof: We have MX = lim

−→
LD , where D runs over the positive divisors, and we conclude by the
above lemma, because the cohomology commutes with inductive limits (p. 332)
H p (X, MX ) = lim
−→
H p (X, LD ) = 0 , p ≥ 1.
Theorem: Let X be a compact Riemann surface. The field M(X) is a finite extension of C(t),
and the points of X correspond to the discrete valuations of M(X) trivial over C.
Proof: Let t : X → P1 be a non constant meromorphic function, of degree d, and let us see
that C(t) → M(X) is a finite extension. It is enough to show that any meromorphic function
f ∈ M(X) fulfills a relation of degree d
f d − s1 (t)f d−1 + . . . ± sd (t) , si (t) ∈ C(t).
Out of the ramification points, the definition of si (t) is obvious: it is the i-th elementary
symmetric function of the values of f on the points of the fibre of t ∈ P1 . Let us see that the
functions si (t) may be extended to meromorphic functions on P1 :
If t0 ∈ P1 is a ramification point, and f has no pole in the fibre of t0 , then f is bounded on a

neighborhood of the fibre, so that si is bounded on a neighborhood of t0 and by the removable
singularity theorem si may be extended to an analytic function. If f has poles in the fibre of
t0 , we pick h ∈ C(t) with many so zeros at t0 , that f¯ = h(t)f has no poles in the fibre. Now
s̄i = hi si is analytic on a neighborhood of t0 , and we conclude that si is meromorphic.
Now, any point x ∈ X defines a discrete valuation vx : M(X) − {0} → Z, trivial over C:
vx (f ) = number of zeros or poles of f at x
(the poles counted with negative sign) and vx 6= vy when x 6= y because meromorphic functions
separate points. Finally, if some discrete valuation v were not defined by a point of X, the
theory of algebraic curves (p. 315) shows the existence of a non constant f ∈ M(X) with a
unique pole at v, so that f has no pole on X. Absurd. q.e.d.
If X is a compact Riemann surface, then Xalg will be the Riemann variety of the field
M(X). Any non constant analytic morphism ϕ : X → Y induces a morphism of C-algebras
ϕ∗ : M(Y ) → M(X), ϕ∗ (f ) = f ◦ ϕ; hence a morphism of C-schemes ϕalg : Xalg → Yalg .
The C-scheme Xalg is a complete non singular curve, and conversely, recall that any complete
non singular curve C over C defines a compact Riemann surface Can .
Theorem: The functors X Xalg and C Can define an equivalence of categories

Compact Riemann Complete and non-singular
!
surfaces algebraic curves over C
Proof: If X is a Riemann surface, the natural map X → (Xalg )an is an homeomorphism by the
above results, and it is analytic; hence an isomorphism.
If C is the Riemann variety of a finite extension Σ of C(t), then the elements of Σ define
meromorphic functions on Can , because locally they are quotients of analytic functions. If the
extension Σ → M(Can ) were of degree > 1, then the meromorphic functions f ∈ Σ do not
separate the discrete valuations of M(Can ); i.e. do not separate points of Can . Absurd. Hence
Σ = M(Can ) and C = (Can )alg .
Corollary: The non constant analytic morphisms ϕ : X → Y correspond to the morphisms of

C-algebras ϕ∗ : M(Y ) → M(X), and deg ϕ = [M(X) : M(Y )].
Corollary: H p (X, LD ) = H p (Xalg , LD ), for any divisor D.
Proof: Let us consider the following exact sequence of sheaves on Xalg

alg
0 −→ LD −→ M(X) −→ PD −→ 0
where M(X) is the constant sheaf (hence flasque, because all Zariski open sets are connected),
and the exact sequence of sheaves on X
0 −→ LD −→ MX −→ PD −→ 0
The cohomology exact sequences define a commutative diagram with exact rows
M(X) / H 0 (Xalg , P alg ) / H 1 (Xalg , LD ) /0

D

H 0 (X, MX ) / H 0 (X, PD ) / H 1 (X, LD ) /0
alg
We conclude because M(X) = H 0 (X, MX ), and also H 0 (Xalg , PD ) = H 0 (X, PD ) since
alg
both sheaves PD and PD have the same stalk at any point x ∈ X and any global section is zero
everywhere, up to a finite number of points. q.e.d.
Now all the results on the cohomology of line sheaves on complete non-singular algebraic
curves (pp. 314-324) also hold for line sheaves on compact Riemann surfaces. In particular the
Riemann-Roch theorem: On any compact Riemann surface X there is a canonical line sheaf
LK such that H 1 (X, L)∗ = HomOX (L, LK ), and for any divisor D,
h0 (LD ) = 1 − g + deg D + h0 (LK−D ).
But the calculation (p. 316) of the dualizing sheaf ωX = LK may be greatly simplified:
Recall that, since ωX represents the functor H 1 (X, −)∗ , it comes equipped with a linear
map ξ ∈ H 1 (X, ωX )∗ such that for any linear form η ∈ H 1 (X, L)∗ there is a unique morphism
ϕ : L → ωX such that ϕ∗ (ξ) = η. Let ΩX be the line sheaf of analytic 1-forms.
The exact sequence 0 → OX → MX → PX → 0 induces an exact sequence
0 −→ ΩX −→ MX ⊗OX ΩX −→ PX ⊗OX ΩX −→ 0
where the sheaf MX ⊗ ΩX = (lim

−→
LD ) ⊗ ΩX = lim
−→
LDn is acyclic. Now the exact sequence
(MX ⊗OX ΩX )(X) −→ (PX ⊗OX ΩX )(X) −→ H 1 (X, ΩX ) −→ 0
shows that H 1 (X, ΩX ) is just the group (PX ⊗OX ΩX )(X) = x {an zdzn + . . . + a1 dz
L
z } of principal
parts of meromorphic forms (where z is a local coordinate centered at x ∈ X), modulo the
subgroup defined by the global meromorphic forms (MX ⊗OX ΩX )(X).
P
Now, we have a linear map Res = x Resx : (PX ⊗OX ΩX )(X) → C, and the residue theorem
states that it factors through a (non null, since Res dz
z = 1) linear map
Res : H 1 (X, ΩX ) −→ C.
Theorem: The dualizing sheaf is (ΩX , Res).
Proof: There is a (non null, hence injective) morphism ϕ : ΩX → ωX such that ϕ∗ (ξ) = Res;
hence a commutative diagram with exact rows
(MX ⊗OX ΩX )(X) / (PX ⊗O ΩX )(X) Res / C

X
ϕ ϕ Id
ξ
(MX ⊗OX ωX )(X) / (PX ⊗O ωX )(X) /C
X
If ϕ : ΩX,x → ωX,x were not surjective at some point x ∈ X, then dz ∈ Dx , so that dz

has
dz
dz
z z
null image by ϕ in (PX ⊗OX ωX )(X). Absurd, 0 = ξ ϕ( z ) = Res z = 1.
L dz dz
Corollary: A finite family (ωx ) ∈ x {an z n + . . . + a1 z } of
P principal parts of meromorphic
forms is defined by a global meromorphic form if and only if x Resx (ωx ) = 0.
Corollary: The genus g = h1 (OX ) is the maximal number of linearly independent analytic
forms on X, and it coincides with the topological genus, 2g = dim Q H 1 (X, C).
Proof: By duality we have
g = h1 (OX ) = h0 HomOX (OX , ΩX ) = h0 (ΩX ),

1 = h0 (OX ) = h0 HomOX (ΩX , ΩX ) = h1 (ΩX ).

Moreover, since X is a compact oriented surface, H 0 (X, C) = H 2 (X, C) = C.

We conclude by the cohomology exact sequence of the exact sequence of sheaves
d
0 −→ C −→ OX −−−→ ΩX −→ 0.
Chapter 14
Differential Geometry II
14.1 Valued Differential Calculus

Let CX∞ be the sheaf of smooth functions on a smooth manifold X of dimension n, and recall
∞ -module of rank n.
that the sheaf D of (smooth) vector fields is a locally free CX
∞ -module E of rank r.
Definitions: Let us fix a locally free CX
An E-valued differential p-form is an alternate CX ∞ -multilinear morphism of sheaves
ω : D× . p. . ×D −→ E
and we agree that valued 0-forms are just global sections of E.

The interior contraction of ω with a vector field D is the E-valued (p − 1)-form
(iD ω)(D2 , . . . , Dp ) = ω(D, D2 , . . . , Dp ).
∞ -bilinear morphism of sheaves E ×E 0 → ·

Given a CX − E 00 , the exterior product of an E-valued
0 0 00
p-form ω with an E -valued q-form ω is the E -valued (p + q)-form
(ω ∧ ω 0 )(D1 , . . . , Dp+q ) = 1
sgn (σ) ω(Dσ(1) , . . . , Dσ(p) ) · ω 0 (Dσ(p+1) , . . . , Dσ(p+q) ).
P
p!q!
σ∈Sp+q
The proofs of p. 65 show that
iD (ω ∧ ω 0 ) = (iD ω) ∧ ω 0 + (−1)p ω ∧ (iD ω 0 )

ω ∧ ω 0 = (−1)pq ω 0 ∧ ω
when ω 0 ∧ ω is considered respect to the product e0 · e = e · e0 .

In particular, if E 0 = E, we have ω ∧ ω 0 = (−1)pq ω 0 ∧ ω when the product is commutative.
Examples: Ordinary p-forms are just CX ∞ -valued p-forms.

q ∞ -module; hence any (p, q)-tensor
The sheaf Tp of tensor fields of type (p, q) is a locally free CX
q
field may be viewed as a Tp -valued 0-form.
An ordinary p-form ωp and a global section e of E, define an E-valued p-form ωp ⊗ e,
(ωp ⊗ e)(D1 , . . . , Dp ) = ωp (D1 , . . . , Dp ) e.
The identity I : D → D, I(D) = D, is a D-valued 1-form.

The torsion tensor Tor(D1 , D2 ) = D1∇ D2 − D2∇ D1 − [D1 , D2 ] of a linear connection ∇ is
a vector valued 2-form, and the curvature R(D1 , D2 ) = D1∇ D2∇ − D2∇ D1∇ − [D1 , D2 ]∇ is an
End(D)-valued 2-form.
395
396 CHAPTER 14. DIFFERENTIAL GEOMETRY II
Lemma: If (U ; u1 , . . . , un ) is a coordinate open set in X where E is trivial, and {e1 , . . . , er } is

a base of E(U ), then E-valued p-forms on U form a free C ∞ (U )-module of base
{ (dui1 ∧ . . . ∧ duip ) ⊗ ej }1≤i1 <...<ip ≤n,1≤j≤r .
Definition: A morphism of sheaves ∇ : D × E −→ E is a linear connection in E if
1. D∇ (e1 + e2 ) = D∇ e1 + D∇ e2 ,
(f D)∇ e = f D∇ e.
2. (D1 + D2 )∇ e = D1∇ e + D2∇ e,

D∇ (f e) = (Df )e + f D∇ e.
Examples: In CX ∞ we have a natural connection D ∇ f = Df , implicitly used in the differential
calculus with ordinary forms.

Connections in D are just linear connections (p. 281) on the manifold X.
If we have connections in E and E 0 , then we have natural connections in
E ⊕ E0 , D∇ (e + e0 ) = D∇ e + D∇ e0
∞ E0
E ⊗CX , D∇ (e ⊗ e0 ) = (D∇ e) ⊗ e0 + e ⊗ (D∇ e0 )
E∗ , (D∇ ω)(e) = D(ω(e)) − ω(D∇ e)
Hom(E, E 0 ) , (D∇ T )(e) = D∇ (T e) − T (D∇ e)
Definition: The Lie derivative of an E-valued p-form ω with a vector field D is the E-valued
p-form
p
(DL ω)(D1 , . . . , Dp ) = D∇ (ω(D1 , . . . , Dp )) −
P
ω(D1 , . . . , [D, Di ], . . . , Dp ).
i=1
·
If we have a bilinear product E × E 0 → − E 00 , and compatible connections in the sense that
∇ 0 ∇ 0 ∇ 0
D (e · e ) = (D e) · e + e · (D e ), it is easy to check that
DL (ω ∧ ω 0 ) = (DL ω) ∧ ω 0 + ω ∧ (DL ω 0 ).
Theorem: Let ΩpX E be the sheaf of E-valued p-forms. There are unique R-linear morphisms
N
of sheaves d : ΩpX E −→ Ωp+1

N N
X E such that for any vector field D,
DL = d ◦ iD + iD ◦ d.
Proof: Let us show the uniqueness by induction on p.

If p = 0, then D∇ e = DL e = iD (de); hence (de)(D) = D∇ e.
If p ≥ 1, then iD (dω) = DL ω − d(iD ω) and, by induction, d(iD ω) is fully determined.
To prove the existence, we define recursively the exterior differential. If p = 0,
(de)(D) = D∇ e
and, once we have defined d on (p − 1)-forms, for any p-form ω we put
(dω)(D, D1 , . . . , Dp ) = (DL ω)(D1 , . . . , Dp ) − (diD ω)(D1 , . . . , Dp ).
We must show that dω is a (p + 1)-form. It is R-multilinear and alternate.

14.1. VALUED DIFFERENTIAL CALCULUS 397
In fact, it is clear that (dω)(D, . . . , Di , . . . , Di , . . .) = 0, and
(dω)(D, D, D2 , . . . , Dp ) = (DL ω)(D, D2 , . . . , Dp ) − (diD ω)(D, D2 , . . . , Dp ) =

p
= D∇ (ω(D, D2 , . . . , Dp )) − ω(D, . . . , [D, Di ], . . .) − (iD diD ω)(D2 , . . . , Dp ) = 0
P
i=2
since (iD diD ω)(D2 , . . . , Dp ) = DL D(iD ω)(D2 , . . . , Dp ) − (diD iD ω)(D2 , . . . , Dp )

p
= D∇ (ω(D, D2 , . . . , Dp )) − ω(D, . . . , [D, Di ], . . .).
P
i=2
Finally, let us show that it is CX∞ -multilinear. Linearity in the variables D , . . . , D is clear;
1 p
hence also in the first variable, since it is alternate.
Cartan’s Formula: (dω)(D1 , D2 ) = D1∇ (ω(D2 )) − D2∇ (ω(D1 )) − ω([D1 , D2 ]).
Proof: (dω)(D1 , D2 ) = (D1L ω)(D2 ) − (diD1 ω)(D2 )

= D1∇ (ω(D2 )) − ω([D1 , D2 ]) − D2∇ (ω(D1 )).
Corollary: The exterior differential of the identity is the torsion 2-form,
Tor∇ = dI.
Proof: (dI)(D1 , D2 ) = D1∇ (I(D2 )) − D2∇ (I(D1 )) − I([D1 , D2 ]) = Tor∇ (D1 , D2 ).

·
Theorem: If we have a bilinear product E × E 0 →
− E 00 , and compatible connections in the sense
∇ 0 ∇ 0 ∇ 0
that D (e · e ) = (D e) · e + e · (D e ), then
d(ωp ∧ ωq0 ) = (dωp ) ∧ ωq0 + (−1)p ωp ∧ (dωq0 ).
Proof: Since iD d = DL − diD , it follows directly from the equalities
iD (ωp ∧ ωq0 ) = (iD ωp ) ∧ ωq0 + (−1)p ωp ∧ (iD ωq0 ),

DL (ωp ∧ ωq0 ) = (DL ωp ) ∧ ωq0 + ωp ∧ (DL ωq0 ).
Example: In a galilean space-time

X, a punctual body of mass m is described by a trajectory
σ(t) = t, x1 (t), x2 (t), x3 (t) with a tangent vector field I, the impulse, such that ω(I) = m,
I = mU = m(∂t + x01 (t)∂1 + x02 (t)∂2 + x03 (t)∂3 ) = m∂t + p~,
where U = ∂t + ~v is the 4-velocity of the particle (an intrinsic concept) and p~ = m~v is the
momentum (it depends on the inertial reference system). Hence, once an orientation [dX] of
X is fixed, a continuous matter distribution is described by a vector-valued 3-form Π3 , the
impulse 3-form, whose intuitive meaning is that Π3 (D1 , D2 , D3 ) is the sum, with the sign
of dX(I, D1 , D2 , D3 ), of the impulse of the particles crossing the infinitesimal parallelogram
determined by D1 , D2 , D3 . As any other vector-valued 3-form, we have Π3 = C11 (dX ⊗ T 2 )
for a unique 2-contravariant tensor T 2 , named matter tensor, which does not depend on the
orientation. In an inertial reference system, we have
T 2 (dt, dt) = dt Π3 (∂1 , ∂2 , ∂3 ) = mass density.

T 2 (dt, dxj ) = dxj Π3 (∂1 , ∂2 , ∂3 ) = j-momentum density.

T 2 (dxi , dt) = mass flux acroos a unit area perpendicular to ∂i .

T 2 (dxi , dxj ) = j-momentum flux acroos a unit area perpendicular to ∂i .
When T 2 = ρU ⊗ U , where ω(U ) = 1, the matter is a dust of mass density ρ and mean
velocity U . When T 2 = ρU ⊗ U + ph, the matter is a perfect fluid of mass density ρ, mean
velocity U and pressure p. We always assume that the matter tensor T 2 is symmetric.
When we consider the exterior differential defined by the Cartan connection ∇, the equation
dΠ3 = 0 encodes the mass conservation law and the newtonian motion law (the equation of
continuity and the Euler equations in presence of a gravitational force). In terms of the matter
tensor, this equation dΠ3 = 0 states that div∇ T 2 = 0, where the divergence of a tensor T
(with some contravariant index) is div∇ T := C11 (∇T ).
14.1.1 Curvature
Definition: In general d2 6= 0. For example, for any section e we have
(d2 e)(D1 , D2 ) = D1∇ ((de)(D2 )) − D2∇ ((de)(D1 )) − (de)([D1 , D2 ])

= D1∇ (D2∇ e) − D2∇ (D1∇ e) − [D1 , D2 ]∇ e = R(D1 , D2 )(e)
where the curvature R is an End(E)-valued 2-form; i.e., d2 e = R ∧ e, the exterior product being
considered with respect to the product End(E) × E → E, T · e = T (e).
Theorem: d2 ω = R ∧ ω.
Pfree sheaf EP
Proof: If the locally is trivial on an open set U , and {e1 , . . . , er } is a base of E(U ),
then we have ω = i ωi ⊗ ei = i ωi ∧ ei , for some ordinary p-forms ωi , and
dω = i (dωi ) ∧ ei + (−1)p ωi ∧ (dei ) ,

P
d2 ω = i (d2 ωi ) ∧ ei + (−1)p+1 (dωi ) ∧ (dei ) + (−1)p (dωi ) ∧ (dei ) + ωi ∧ (d2 ei )

P
= i ωi ∧ (d2 ei ) = i ωi ∧ (R ∧ ei ) = i (−1)2p R ∧ ωi ∧ ei = R ∧ ω.
P P P
Bianchi’s Differential Identity: dR = 0.
Proof: d3 e = d2 (de) = R ∧ (de) for any section e of E, and
d3 e = d(d2 e) = d(R ∧ e) = (dR) ∧ e + R ∧ (de);
hence (dR) ∧ e = 0. That is to say, (dR)(D1 , D2 , D3 )(e) = 0, and dR = 0.
Parallel Transport: The sheaf of smooth sections E of a real vector bundle E → X is a locally
free CX∞ -module, and E-valued forms are just forms with values in the fibres of E.
Given a connection ∇ on E, and a smooth map φ : Y → X, there is a unique connection

φ∗ ∇ on the sheaf of smooth sections of the vector bundle φ∗ E = E ×X Y → Y such that, for
any local section e ∈ E(U ),
φ∗ (de) = d(φ∗ e).
When E admits a base e1 , . . . , er we have D∇ ei = j ωij (D)ej for some ordinary 1-forms
P
ωij , and the only possibility is just to consider the connection given by the 1-forms φ∗ ωij ,
∗
D̄φ ∇ (φ∗ ei ) = j (φ∗ ωij )(D̄)(φ∗ ej )
P
so that clearly φ∗ (dei ) = d(φ∗ ei ); hence also for any other local section e. By the uniqueness,
these local connections coincide on intersections and they define a global connection.
For example, when E = T X is the tangent bundle and σ : I → X is a smooth curve, fields
with support on σ are just smooth sections of σ ∗ T X, and σ ∗ ∇ is just the covariant derivative
of fields with support introduced in p. 283.
14.1. VALUED DIFFERENTIAL CALCULUS 399
In general, given a smooth curve σ : I → X and a vector e0 ∈ Ex0 in the fibre of x0 = σ(t0 ),
there exists a unique parallel section e : I → E, de = 0, such that e(t0 ) = e0 , and so we obtain
a parallel transport (depending on σ, not just on the end points)
∼
Eσ(t0 ) −−→ Eσ(t1 ) ; t0 , t1 ∈ I.
Cartan Structure Equations: Let ∇ be a linear connection on X. If {D1 , . . . , Dn } is a local
base of vector fields, and {θ1 , . . . , θn } is the dual base, we have
n
ωij ⊗ Di , ωij (D) = θi (D∇ Dj ).
P
1. dDj =
i=1
n
2. d2 Dj = R ∧ Dj = Ωij ⊗ Di , Ωij (D, D0 ) = θi (R(D, D0 )Dj ).
P
i=1
n
Θi ⊗ Di , Θi (D, D0 ) = θi (Tor(D, D0 )).
P
3. Tor = dI =
i=1
P P
Calculating the exterior differential of the identity I = j θj ⊗ Dj = j θj ∧ Dj ,
P P P P
dI = j dθj ∧ Dj − j θj ∧ dDj = i dθi ⊗ Di − i,j (θj ∧ ωij ) ⊗ Di
P P
= i (dθi + j ωij ∧ θj ) ⊗ Di
and, comparing with 3, we obtain the first structure equation,
P
dθi + j ωij ∧ θj = Θi .
Differentiating 1 and comparing with 2, we obtain the second structure equation,
P
dωij + k ωik ∧ ωkj = Ωij .
If we consider θ = (θi ), Θ = (Θi ) as Rn -valued forms, and ω = (ωij ), Ω = (Ωij ) as forms
with values in the n × n matrices, these equations are
dθ + ω ∧ θ = Θ
dω + ω ∧ ω = Ω
where the exterior products are considered with respect to the matrix product.
If we differentiate the first structure equation,
0 + (dω) ∧ θ − ω ∧ (dθ) = dΘ
and we replace dθ and dω by the values given by the structure equations, we obtain
Bianchi’s Identity: Ω ∧ θ − ω ∧ Θ = dΘ.
If we differentiate the second structure equation,

0 + (dω) ∧ ω − ω ∧ (dω) = dω
and we replace dω by the value given by the second structure equation, we obtain
Bianchi’s Differential Identity: Ω ∧ ω − ω ∧ Ω = dΩ.
In the case of a symmetric connection (such as the Levi-Civita connection of a riemannian

metric), we have Θ = 0 and the structure equations and Bianchi’s identities are

dθ + ω ∧ θ = 0 Ω∧θ =0
,
dω + ω ∧ ω = Ω Ω ∧ ω − ω ∧ Ω = dΩ
14.2 Calculus of Variations

Definition: Two smooth maps s̄, s : X → Y defined on certain neighborhoods of a point x ∈ X
have equal k-jet at x when s∗ (f ) ≡ s̄∗ (f ) (mod. mk+1 ∞
x ) for all f ∈ C (Y ). In particular, they
have equal 1-jet when s̄(x) = s(x) and the tangent linear maps at x coincide.
Equivalence classes are k-jets at x of maps, and the k-jet at x of a map s is denoted by jxk s.
The set of 1-jets of local smooth sections of a regular projection π : Y → X is denoted by
1
J Y , and it is endowed with natural projections
p
J 1Y /Y p(jx1 s) = s(x)
π
π̄

X π̄(jx1 s) = x
If (x1 , . . . , xn ) are coordinates on an open set U ⊆ X, and (x1 , . . . , xn , y1 , . . . , ym ) on an

open set V ⊆ π −1 (U ), then in p−1 (V ) the 1-jet jx1 s of a section s is fully determined by the
coordinates (x1 , . . . , xn , y1 , . . . , ym ) of s(x) and the coefficients of the jacobian matrix
∂yj
yj,i = (x).
∂xi
On J 1 Y we have a unique structure of smooth manifold such that (xi , yj , yj,i ) are local
coordinate systems. Hence p and π̄ are regular projections, and any section s : X → Y has a
1-jet extension s̄ : X → J 1 Y , s̄(x) = jx1 s, which is smooth since the 1-jet extension s̄ of a local
section yj = yj (x1 , . . . , xn ) is given by the equations
yj = yj (x1 , . . . , xn )
∂yj (x1 , . . . , xn )
yj,i =
∂xi
Moreover, on J 1 Y we have a structure 1-form θ with values in the inverse image p∗ T v Y of
the vertical bundle (the bundle of vectors tangent to Y with null projection on X),
θ(Djx1 s ) = p∗ D − s∗ π̄∗ D,
P P
θ = j (dyj − i yj,i dxi ) ⊗ ∂yj .
Definitions: The kernel of θ defines the structural Pfaff system P of J 1 Y , and it is locally
generated by the structure 1-forms
P
θj = dyj − i yj,i dxi ; j = 1, . . . , m,
and the 1-jet extension s̄ of a section s is characterized by the condition of being tangent to the
structural Pfaff system, s̄∗ P = 0.
The contact ideal I is the ideal generated by P in the exterior algebra p ΩpJ 1 Y .
L
A vector field D̃ on J 1 Y is a infinitesimal contact transformation if it preserves the

structural Pfaff system, D̃L P ⊆ P , or equivalently the contact ideal, D̃L I ⊆ I.
Proposition: Any vector

P field D on Y is the projection of a unique contact infinitesimal trans-
formation D̃. If D = j hj ∂yj is vertical,
X ∂ X ∂hj X ∂hj

∂
D̃ = hj + + yk,i .
j ∂yj j,i ∂xi k ∂yk ∂yj,i
14.2. CALCULUS OF VARIATIONS 401
Proof: If f ∈ C ∞ (Y ), then df ≡
P
i fi dxi
(mod P ), for some functions fi , since
P
dyj ≡ i yj,i dxi (mod. P ).
P P
Now, if D = i gi ∂xi + j hj ∂yj , then D̃xi = gi , D̃yj = hj , and the functions D̃yj,i are fully
determined by the following congruences (mod. P )
0 ≡ D̃L θi = D̃L (dyj − yj,i dxi ) = dhj − (D̃yj,i )dxi − yj,i dgi ,
P P P
i i i
P P P
(D̃yj,i )dxi ≡ dhj − yj,i dgi ≡ ui dxi .
i i i
Definitions: Given a n-form Ln on J 1Y

, the critical sections of the variational problem defined
by Ln are the smooth sections s : X → Y such that
Z
D̃L Ln = 0
s̄
for any vertical vector field D on Y with compact support. If τt is the flow of D, this condition
states thatR the derivative at t = 0 of the integral of Ln on the 1-jet extension of τt (s) is 0.
Since s̄ D̃L ωn = 0 for any n-form ωn ∈ I, two n-forms Ln , L0n on J 1 Y define the same
variational problem when
L0n ≡ Ln (mod. I).
We always assume that locally Ln = Ldx1 ∧ . . . ∧ dxn for some smooth function L on J 1 Y ,
named lagrangian, and we put dX = dx1 ∧ . . . ∧ dxn .
Lemma: If (iD dLn )|s̄ = 0 for any vector field D on J 1 Y , then s is a critical section.
The converse holds when dLn ≡ 0 (mod. I).
Proof: For any vector field D on J 1 Y , with compact support on s̄, we have
Z Z Z Z
L Stokes
D Ln = iD dLn + diD Ln = diD Ln === 0
s̄ s̄ s̄ s̄
since iD Ln has compact support on s̄, and Ps is a critical section.

Conversely, if dLn ∈ I, locally dLn = j θj ∧ ωj for some n-forms ωj .
Now, if a section s is critical, for any vertical vector field D with compact support we have
Z Z Z X Z
L
0 = D̃ Ln = iD̃ dLn + diD̃ Ln = θj (D̃)ωj .
s̄ s̄ s̄ j s̄
When D = ρ∂yj , where ρ ≥ 0 has compact support,

Z Z
0 = θj (D̃)ωj = ρωj .
s̄ s̄
Since the support of ρ is arbitrarily small, ωj |s̄ = 0.

Since also θj |s̄ = 0, then for any vector field D on J 1 Y we have
P P
(iD dLn )|s̄ = iD j θj ∧ ωj |s̄ = i θj (D)ωj − ωj (D)θj |s̄ = 0.
Fundamental Lemma: There is a n-form Θ on J 1 Y , defining the same variational problem

as LdX, which is closed modulo the contact ideal,
Θ ≡ LdX (mod. I),

dΘ ≡ 0 (mod. I).
n−1
Moreover Θ is unique if we require that Θ ≡ LdX (mod. P ∧ ΩX ).
Proof: On J 1 Y , a local base of (n + 1)-forms which are multiple of dX is
θj ∧ dX = dyj ∧ dX, dθj ∧ i∂xi dX = −dyj,i ∧ dX.
Therefore, locally there are functions fji , gj on J 1 Y such that

P P
d(LdX) = dL ∧ dX = j,i fji dθj ∧ i∂xi dX + j gj θj ∧ dX,
P
d(LdX) ≡ j,i fji dθi ∧ i∂xi dX (mod. I),
P
Θ = LdX − j,i fji θj ∧ i∂xi dX
and we have Θ ≡ LdX, dΘ ≡ 0 (mod I).

The local uniqueness is clear since {θj , dyj,i , dxj } is a local base of 1-forms, and
d(f θj ∧ i∂xi dX) ≡ −f dyj,i ∧ dX.
Finally, the local uniqueness let us conclude the global existence of the n-form Θ on J 1 Y ,
named Poincaré-Cartan form of the variational problem.
Its local expression (where we put Lyji = ∂L/∂yj,i ) is
i
P P
Θ = LdX − j,i (−1) Lyji dyj − k yj,k dxk ∧ dx1 ∧ . . . dx
ci . . . ∧ dxn .
Theorem: A smooth section s is critical if and only if for any vector field D on J 1 Y
(iD dΘ)|s̄ = 0.
Definitions: An infinitesimal symmetry of the variational problem defined by LdX is a

contact infinitesimal transformation D on J 1 Y such that DL (LdX) ≡ 0 (mod. I). The Noether
invariant of a symmetry D is the (n − 1)-form −iD Θ.
Since Θ ≡ LdX (mod. I), and DL I ⊆ I, D is an infinitesimal symmetry if and only if
DL Θ ≡ 0 (mod. I).
Noether’s Theorem: The Noether invariant of an infinitesimal symmetry D is a closed form

on the 1-jet extension s̄ of any critical section,
(diD Θ)|s̄ = 0.
Proof: Since DL Θ ≡ 0 (mod. I), on any section s we have (DL Θ)|s̄ = 0.

If moreover s is critical, then (iD dΘ)|s̄ = 0, and
0 = (DL Θ)|s̄ = (diD Θ + iD dΘ)|s̄ = (diD Θ)|s̄ .
Note: When X = R, Noether invariants are functions on J 1 Y , constant on any critical section.
In general, if X = R × S, where we identify the first factor with time, ωn−1 = (iD Θ)|s̄ is a closed
(n − 1)-formRon X and it gives all kind of conservation laws. For example, if the support of ωn−1
is compact, t×S ωn−1 does not depend on the instant t.
14.2. CALCULUS OF VARIATIONS 403
14.2.1 Problems in Dimension 1

Now we assume that X = R, and we put t = x1 , yj0 = yj,1 .
J 1 Y has odd dimension 2m + 1, so that the 2-form dΘ has non null radical (p. 163) and
condition (iD dΘ)|s̄ = 0 states that s̄ is tangent to the radical of
P
dΘ = j (dLyj0 − Lyj dt) ∧ θj
or equivalently that the 1-forms dLyj0 − Lyj dt vanish on s̄.
Euler-Lagrange Equations: A smooth section s is critical if and only if on s̄ we have

d

dt L yj
0 − Lyj = 0.
Definitions: A lagrangian L is regular if det(Lyi0 yj0 ) 6= 0 at any point of J 1 Y , so that
dt, θj = dyj − yj0 dt, dLyj0 − Lyj dt

P
define a local base of 1-forms, or equivalently the radical of dΘ = j (dLyj0 − Lyj dt) ∧ θj has
dimension 1 and it is not vertical. In such a case, the radical hZi of dΘ is incident to P = hθj i,
and the integral curves of the lagrangian field Z, normalized with condition Zt = 1, are just
the 1-jet extensions of critical sections.
Lemma: Let D be a vector field on a manifold X. If there exists a function t ∈ C ∞ (X) such
that Dt > 0, then the integral curves of D are the fibres of a regular projection X → W onto a
(eventually non separated) manifold.
Proof: The integral curves of D are the fibres of a surjective map π : X → W , and we consider
on W the quotient topology (U ⊆ W is open if and only if so is π −1 U ) and the sheaf
O(U ) = {f ∈ C(U ) : π ∗ f ∈ C ∞ (π −1 U )}.
Let us show that (W, O) is a smooth manifold and π is a regular projection.

By hypothesis the hypersurfaces Ha of equation t = a transversally intersect integral curves,
and at a unique point since t is an increasing function on any integral curve.
Each point of X admits an open neighborhood V such that V → π(V ) = Ha ∩ V is a regular
projection for some smooth structure on π(V ), coinciding with the structure induced by W .
In fact, if U is an open set in Ha ∩ V , then
π −1 (U ) = τt (V 0 )
S
V 0 ,t
0
is an open set in X, where V runs over all open subsets of U .
By the same reason, if f is smooth on U , then π ∗ f is smooth on π −1 U . q.e.d.
Since Zt = 1, then the 1-jet extensions of critical sections are the fibres of a regular projection
J 1 Y → W , and the 2-form dΘ is projectable onto W since
iZ dΘ = 0, Z L dΘ = diZ dΘ + iZ ddΘ = 0 + 0 = 0.
Definition: The projection ω2 of dΘ onto W is a non singular closed 2-form, and the (eventually
non separated) symplectic manifold (W, ω2 ) is the variety of solutions of the variational
problem.
Proposition: Any infinitesimal symmetry D of the variational problem, and the Noether in-
variant f = −Θ(D), are projectable onto the variety of solutions W , and iD ω2 = df on W .
Proof: We have to show that D preserves the sheaf of first integrals of Z. If Zf = 0,
Z(Df ) = [Z, D]f − D(Zf ) = −(DL Z)f.
Deriving with DL the equality 0 = C11 (Z ⊗ dΘ), and using that DL (dΘ) = 0, we obtain that
C11 (DL Z ⊗ dΘ) = 0. Hence DL Z is in the radical of dΘ and it is proportional to Z, so that
(DL Z)f = 0 and we conclude that Z(Df ) = 0.
Finally, f = −Θ(D) is constant on any solution, and it defines a smooth function on W .
Condition DL Θ = 0 shows that iD dΘ = df ; hence iD ω2 = df on W .
Hamilton’s Equations: Assume that Y = R × F and that L does not depend on t.

In this case ∂t is an infinitesimal symmetry of the variational problem, and P the Noether
invariant h = −Θ(∂t ) is the energy or hamiltonian. In coordinates h = −L − j yj0 Lyj0 .
Now J 1 Y = R × T F , and the lagrangian field Z does not depend on t since ∂t is an infinites-
imal symmetry, Z = ∂t + Z̄ where Z̄ is a vector field on the tangent bundle T F .
The 2-form dΘ is invariant by ∂t , and its restriction ω2 to any fibre T F is non singular since
fibres are transversal to hZi = rad dΘ.
(T F, ω2 ) is a symplectic manifold.
Since iZ dΘ = 0 and i∂t dΘ = ∂tL Θ − di∂t Θ = dh, we have
Hamilton’s Equations: iZ̄ ω2 = dh.

P
The restriction of dΘ to any fibre is ω2 = j dLyj0 ∧ dyj ; hence the functions
pj = yj , qj = Lyj0
P
are canonical coordinates on T F , in the sense that ω2 = j dqj ∧ dpj .
In canonical coordinates, Hamilton’s equations are
(
qj0 = hpj
p0j = −hqj
14.3 Natural Bundles

Let us fix a smooth manifold X of constant dimension n.
Definitions: A natural bundle over a smooth manifold X is a regular projection π : F → X,
endowed with a lifting τ∗ : FU = π −1 (U ) → FV = π −1 (U ) of any diffeomorphism τ : U → V
between open sets in X, such that the following squares commute
τ∗ / FV
FU
π π

U
τ /V
1. Functoriality: Id∗ = Id , and (τ ◦ τ 0 )∗ = τ∗ ◦ τ∗0 .
2. Local Character: For any diffeomorphism τ : U → V and any open set U 0 ⊂ U , the
restriction of τ∗ to FU 0 is the lifting of τ|U 0 : U 0 → τ (U 0 ).
14.3. NATURAL BUNDLES 405
3. Regularity: If {τt : Ut →SVt }t∈T is a smoothSfamily of diffeomorphisms, parameterized by

a manifold T (i.e., U = t Ut × t and V = t Vt × t are open sets of X × T , and U → V ,
(x, t) 7→ (τt x, t), is a diffeomorphism), then {τt∗ : FUt → FVt }t∈T is a smooth family of
diffeomorphisms between open sets of F .
A morphism of natural bundles is a smooth map ϕ : F → F 0 commuting with the lifting of
diffeomorphisms, ϕτ∗ = τ∗ ϕ,
ϕ
FU / F0
U
τ∗ τ∗
ϕ

FV / F0
V
A natural bundle F → X is of order1 ≤ k if for any pair of diffeomorphisms τ 0 , τ : U → V ,

and any point x ∈ U where jxk τ 0 = jxk τ , we have τ∗0 = τ∗ on the fibre Fx of x.
Examples: The tangent bundle T X → X, the cotangent bundle T ∗ X → X (where τ∗ = (τ ∗ )−1 ),
and all tensor bundles Tpq X → X, are natural vector bundles of order 1.
If F → X is a natural bundle of order k, then J 1 F → X is a natural bundle of order k + 1.
If F → X is a natural bundle of order ≤ k and we fix a point p ∈ X, the Lie group Gp = Gkp
of k-jets at p of diffeomorphisms U → V fixing p, acts on the fibre Fp ,
g · e = τ∗ (e); g = jpk τ ∈ Gkp , e ∈ Fp ,
and this action is smooth (Fp is a Gkp -manifold).

In fact, in a coordinate neighborhood of p it is clear the existence of a smooth family of
diffeomorphisms τg : Ug → Vg , parameterized by g ∈ Gp , such that g = j k τg . By regularity, (τg )∗
is a smooth family of diffeomorphisms of Fp , and g · e = (τg )∗ (e) is a smooth action.
Galois Theorem: The fibre functor F Fp defines an equivalence of categories

Natural bundles
! Gkp -manifolds .

over X of order ≤ k
Proof: The similarity with the Galois theory of coverings (p. 243) is clear; but paths joining
two points p and x (defining an identification of the fibres) are replaced by diffeomorphisms
σ : U → V such that σ(x) = p. So we shall obtain a universal natural bundle P → X of order
k, and it is a Galois bundle, P ×X P = Gp × P , of group Gp . Moreover P/Gp = X, and P
trivializes any natural bundle F of order ≤ k, F ×X P = P × Fp , so that the action of Gp on Fp
reconstructs the bundle, F = (P × Fp )/Gp . Let us see the details:
The manifold P of k-jets jxk σ of diffeomorphisms σ : U → V , σ(x) = p, admits a regular
projection P → X, jxk σ 7→ x, and the group Gp acts on P ,
(jpk g) · (jxk σ) = jxk (gσ).
This universal bundle P is a smooth principal bundle of group Gp (in the sense that locally
we have diffeomorphisms PU = Gp × U over X preserving the action of the Lie group Gp ) and
it is a natural bundle of order k.
The lifting of a diffeomorphism τ : U → V is defined as follows
τ
P|U −−∗→ P|V , τ∗ (jxk σ) = jτk(x) (στ −1 ).
1
In fact, any natural bundle has finite order; but in this notes we shall not prove this result.
Now, the associated bundle of a Gp -manifold Fp is defined to be
F = (P × Fp )/G −→ X, [(jxk σ, e)] 7→ x,
and it is a natural bundle of order ≤ k, where the lifting of τ : U → V is

τ
F|U = (PU × Fp )/Gp −−∗→ (PV × Fp )/Gp = F|V , τ∗ [jxk σ, e] = [τ∗ (jxk σ), e].
This lifting is well defined since the actions of τ∗ and Gp on P commute:
τ∗ (jp g · jx σ) = τ∗ jx (gσ) = jτ (x) (gστ −1 ) = jp g · jτ (x) (στ −1 ) = jp g · τ∗ jx σ,
and we get a functor from the category of Gp -manifolds into the category of natural bundles,
any Gp -morphism f : Fp → Fp0 inducing the morphism of natural bundles
Id×f
F = (P × Fp )/Gp −−−−→ (P × Fp0 )/Gp = F 0 , [jxk σ, e] 7→ [jxk σ, f (e)].
Let us see that this functor and the fibre functor define an equivalence of categories.
We have a Gp -diffeomorphism
Fp = [(P × Fp )/Gp ]p , e 7→ [(jpk Id, e)],

jp g · [jp Id, e] = [g∗ jp Id, e] = [jp g −1 , e] = [jp g · (jp g −1 , e)] = [jp Id, jp g · e].
On the other hand, if F → X is a natural bundle of order ≤ k, we have an isomorphism of

F onto the associated bundle of the fibre Fp ,
(P × Fp )/Gp = F, [jx σ, e] 7→ σ∗−1 e,
and it is well defined since, if we replace (jx σ, e) by the equivalent pair
jp g · (jx σ, e) = (jx (gσ), jp g · e) = (jx (gσ), g∗ e),
we have that [jx (gσ), g∗ e] 7→ (gσ)−1 −1

∗ g∗ e = σ∗ e.
Definition: Replacing p by the origin of Rn we shall see that a natural bundle over X defines
a natural bundle over any manifold of dimension n.
References of order k at x ∈ X are k-jets jxk σ of diffeomorphisms σ of a neighborhood of
x ∈ X onto a neighborhood of the origin 0 ∈ Rn such that σ(x) = 0.
References of order k at the origin of Rn form a Lie group Gkn , acting on the bundle RXk →X
of references of order k at points of X, which is a natural bundle of order k.

The associated bundle of a Gkn -manifold F0 is the natural bundle (RX k × F )/Gk → X,
0 n
and it is of order ≤ k.
Corollary: The functor “associated bundle” defines an equivalence of categories

k Natural bundles
Gn -manifolds ! .
over X of order ≤ k
Proof: Once we fix local coordinates at p, we have isomorphisms Gkn ' Gkp , RX k ' P , any
k k k
Gn -manifold F0 inherits a structure of Gp -manifold, and (RX × F0 )/Gn ' (P × F0 )/Gp .
Now the result follows from the above theorem. q.e.d.
14.4. CHERN CLASSES AND CURVATURE 407
1. An inverse functor is the fibre functor F Fp , where Fp is considered with the structure of
k
Gn -manifold defined by a choice of coordinates. To obtain an intrinsic inverse functor, just
take, as in the Galois theory of coverings (p. 239), the funtor
F k , F ) = Hom ((Rk ) , F ) ' F .
Homnat (RX Gkp X p p p
2. Natural vector bundles of order ≤ 1 correspond to smooth linear representations of the group
G1n = Gl(Rn ), the tangent bundle corresponding to the obvious action of Gl(Rn ) on Rn .
3. The sign of the determinant defines an action of Gl(Rn ) on {±1}, corresponding to the
orientation covering ∆X → X.
4. Natural coverings of order ≤ k correspond to actions of Gkn on discrete manifolds. Such

actions factor through the quotient by the connected component of the identity and, since
Gkn has two connected components, we see that natural coverings of finite order are disjoint
unions of the trivial covering X → X and the orientation covering ∆X → X.
14.4 Chern Classes and Curvature

Let ∇ be a linear connection on a complex vector bundle E → X of rank r.
Any homogeneous polynomial P (xij ) of degree p on the matrices Mr×r corresponds to a
symmetric linear map P̃ : p Mr×r → C, P (A) = P̃ (A⊗ . p. . ⊗A).
N
When P is invariant (under the action of the linear group) P̃ is well defined on the endo-
morphisms of any complex vector space of dimension r.
Hence it induces a morphism P̃ : p End(E) → CX , and we obtain an ordinary 2p-form
N
P (Ω) = P̃ (R∧ . p. . ∧R).
If we fix a local base of sections, this complex 2p-form is P (Ωij ), where Ωij are the curvature
2-forms (p. 399) and the product of 2-forms is the exterior product.
As an example we have the invariant polynomials cp (A) = tr(Λp A), and
|I + A| = 1 + c1 (A) + . . . + cr (A).
Theorem: If a polynomial P (xij ) is invariant under the linear group, then the differential form
P (Ω) is closed, and its class in H • (X, C) does not depend on the linear connection ∇.
Proof: By the Bianchi’s identity dR = 0, we have
d(R∧ . p. . ∧R) = (dR)∧ . p. . ∧R + . . . + R∧ . p. . ∧(dR) = 0.
Since P is invariant, the morphism P̃ : p End(E) → CX commutes with the parallel trans-
N
port; hence with the covariant derivative of sections and the exterior differential,
dP (Ω) = d(P̃ (R∧ . p. . ∧R)) = P̃ (d(R∧ . p. . ∧R)) = 0.
Now, given two connections ∇0 , ∇1 on E, we consider the projection π : X × R → X and

∗ ∗ ∗ ∗

the connection ∇ = tπ ∇1 + (1 − t)π ∇0 on π E, so that P (Ωi ) = si P (Ω) , i = 0, 1.
We conclude because s∗0 = s∗1 , since π ∗ : H • (X, C) → H • (X × R, C) is an isomorphism.
Lemma: If Ω is the curvature 2-form Ωof a linear connection on a complex line bundle L, then
the obstruction class of L is δ(L) = 2πi ∈ H 2 (X, C).
Proof: Let O be the sheaf of complex smooth functions, and Z p the sheaf of closed p-forms.
According to the following commutative diagram with exact rows, where dl(f ) = f −1 df ,
/Z 2πi /O ef / O∗ /0
0
2πi dl

0 /C /O d / Z1 /0
we must prove that dl : H 2 (X, Z) = H 1 (X, O∗ ) → H 1 (X, Z 1 ) = H 2 (X, C) takes δ(L) into [Ω].
Let us consider the commutative diagram with exact rows
0 / O∗ / C 0 O∗ /F /0
dl dl

0 / Z1 / C 0 Ω1 /G /0
O O X O
0 / Z1 / Ω1 d / Z2 /0
X
Let {Ui } be an open cover of X where there are continuous sections ei : Ui → L not vanishing
at any point, and we put ei = gij ej on Ui ∩ Uj .
Once we fix generators of the fibres, ei defines a section fi of C 0 O∗ on Ui , and fi /fj = gij
in O∗ (Ui ∩ Uj ), so that the sections fi define a global section f of F, representing δ(L).
We must prove that f and Ω define sections of G differing in a global section of C 0 Ω1X .
Now, we have the connection 1-forms θi , and Ω = dθi −θi ∧θi = dθi by the Cartan’s structure
equations; hence we must show that the germ of θi − dl(fi ) at a point does not depend on the
index i; i.e., that we have θi − θj = dl(fi /fj ) = dl(gij ),
D∇ ei = θi (D)ei = gij θi (D)ej ,

D∇ ei = D∇ (gij ej ) = D(gij )ej + gij D∇ ej = (dgij + gij θj )(D)ej ,
−1
θi = θj + gij dgij .
Theorem: The Chern classes of any complex vector bundle E are

−Ω
cp (E) = cp .
2πi
Proof: Two connections ∇0 , ∇00 on certain vector bundles E 0 , E 00 , with curvature 2-forms Ω0 , Ω00 ,
induce a connection ∇ on E 0 ⊕ E 00 with curvature 2-form
0
Ω 0
Ω=
0 Ω00
−1 −1 0 −1 00

det I + 2πi Ω = det I + 2πi Ω ∧ det I + 2πi Ω
−1 −1 −1
P
cn 2πi Ω = cp 2πi Ω ∧ cq 2πi Ω
p+q=n
and if the theorem holds for E 0 and E 00 , then it also holds for E 0 ⊕ E 00 by Whitney’s formula,
since the exterior product represents the cup product (p. 342).
After a base change Y → X, injective at the cohomology level, we may assume that E
decomposes as a direct sum of line bundles (p. 363) and we conclude by the former lemma.
14.4. CHERN CLASSES AND CURVATURE 409
Theorem: If X is a connected orientable manifold of dimension n, then integer cohomology

classes in Hcn (X, R) are just classes of integer integral,
Z
εX = 1.
X
Proof: If U is a connected open subset, the natural morphism Hcn (U, Z) → Hcn (X, Z) is an
isomorphism, since the dual is the restriction morphism TX (X) → TX (U ).
Hence we are reduced to the case X = Rn .
Now, by the Künneth and Fubini’s theorems,
Z Z Z n
εRn = εR ∧ . . . ∧ εR = εR
Rn Rn R
and we are reduced to one of the cases X = R, or R2 , or the sphere S2 (remark that X εX is
R
always positive since we integrate with the orientation defined by εX ).

Let us prove it on the complex projective line P1 , that we identify with the sphere of radius
1 in R3 through the stereographic projection.
If ξ is the tautological line bundle on P1 , the tangent bundle is ξ ∗ ⊗ ξ ∗ (p. 316) and
c1 (ξ ∗ ⊗ ξ ∗ ) = −2c1 (ξ) = 2εP1 .
The Levi-Civita connection of the sphere is a connection on the complex vector bundle ξ ∗ ⊗ξ ∗
since it preserves rotations.
The curvature 2-form is Ω = −iω2 , where ω2 is the area form; hence
i 1
c1 (ξ ∗ ⊗ ξ ∗ ) = Ω = ω2 ,
Z 2πZ 2π
1 4π
2εP1 = ω2 = = 2.
P1 2π S2 2π
Corollary: If π : Y → X is a proper morphism between connected oriented smooth manifolds of

dimension n, and ωn is a n-form with compact support in X,
Z Z
∗
π ωn = (deg π) ωn .
Y X
Proof: The integration of n-forms defines isomorphisms X : Hcn (X, R) ' R, Y : Hcn (Y, R) ' R
R R
(p. 357), and π ∗ (εX ) = (deg π)εY by the definition of deg π (p. 357).
Gauss-Bonnet Theorem
The tangent bundle of a compact oriented riemannian surface X is a complex line bundle with
the structure defined by the automorphism J attached to the area form ω2 ,
J(D) · D0 = ω2 (D, D0 ).
and the Levi-Civita connection ∇ preserves the scalar product and the area form; hence the
complex structure, and it is a connection on this complex line bundle.
If K is the scalar curvature of the surface, the curvature 2-form of this connection is
Ω = −iKω2 .
Z
1
Gauss-Bonnet Theorem: Kω2 = χ(X).
2π X
Proof: Since the tangent bundle T X is the normal bundle of the diagonal map X → X × X, by
the following lemma we have a diffeomorphism of a neighborhood of the zero section s0 in T X
onto a neighborhood of the diagonal ∆ in X × X; hence
c1 (T X) = s∗0 (s0,∗ (1)) = ∆∗ (∆∗ (1)) = χ(X)εX ,

Z Z Z
iΩ 1
χ(X) = c1 (T X) = = Kω2 .
X X 2π 2π X
Tubular Neighborhood Lemma: Let N be the normal bundle of a compact submanifold

i : Y → X of a riemannian manifold X. There is an open neighborhood U of Y in X and a
diffeomorphism ϕ : N → U transforming the zero section into the embedding, i = ϕ ◦ s0 .
Proof: Let us consider the neighborhoods Uε = {Dy ∈ N : kDy k < ε} of the zero section s0 (Y ).
By the theorem of smooth dependence on the initial conditions of the solutions of a differential
equation (p. 254) the exponential map
exp : Uε −→ X, exp(Dy ) = σ(1),
where σ(t) is the geodesic curve tangent to Dy at t = 0, is well-defined and it is smooth if ε is

sufficiently small.
Moreover, at any point y ∈ Y , the tangent linear map
exp∗ : Ty (N ) = Ty Y ⊕ Ny −→ Ty X = Ty Y ⊕ Ny
is the identity. In fact, it is obvious on Ty Y , and on Ny we have that the line γ(t) = tDy is
tangent to Dy at t = 0, and exp γ(t) = σ(t) is tangent to Dy at t = 0.
Hence we may fix ε so that exp is a local diffeomorphism at any point, and even injective.
Otherwise for any natural number n there are vectors Dn , Dn0 at some points yn , yn0 ∈ Y , of
modulus < 1/n, such that exp(Dn ) = exp(Dn0 ).
Since Y is compact, we may assume that the sequences converge
(y, 0) = lim(Dn )yn ,

(y 0 , 0) = lim(Dn0 )yn0 .
Since exp is continuous, we have y = y 0 , so that the equality exp(Dn ) = exp(Dn0 ), for large
n, contradicts the statement that exp is a local diffeomorphism at y.
p
Finally, the diffeomorphism Uε → N , Dy 7→ 1 − kDy k2 /ε2 Dy is the identity on s0 (Y ).
Chapter 15
Algebraic Geometry II
15.1 Injective Modules

All rings are assumed to be noetherian.
Definitions: An injective morphism of A-modules M ,→ Me is an essential extension if any
non null submodule of Me has non null intersection with M , and we name it an injective hull
of M if moreover Me is an injective A-module.
Lemma: If M ,→ Me is an essential extension and M ,→ I is an injective morphism, then any

extension Me → I of j is injective (since the kernel intersects M at 0, it is null).
Theorem: Any A-module M admits a unique injective hull E(M ).
Proof: We know that M is a submodule of an injective module I (p. 134) and by Zorn’s lemma
there is a maximal essential extension M ,→ Me ,→ I. Again Zorn’s lemma shows the existence
of a maximal submodule N ⊂ I such that Me ∩ N = 0.
Since Me ,→ I/N is essential, by the above lemma any extension I/N → I is injective; hence
Me ,→ I/N ,→ I, and Me ,→ I/N is an isomorphism.
Then Me is a direct summand of I; hence Me is injective, and it is an injective hull of M .
If M ,→ Me0 is another injective hull, by the above lemma Me ,→ Me0 .
If it is not surjective, Me being injective, it has a non null supplement in Me0 ; absurd since
Me0 is an essential extension of M . Hence Me ' Me0 .
Theorem: Any injective A-module is I ' ⊕j E(A/pj ), where pj are prime ideals.
P
Proof: Let {Ij } be a maximal family of submodules Ij ' E(A/pj ) such that the sum J = j Ij
is direct (it exists by Zorn).
Direct sums of injective modules are injective by the ideal criterion (A is noetherian).
Hence I = J ⊕I 0 . If I 0 6= 0, some element has prime annihilator p (p. 192) and E(A/p) ,→ I 0 ,
against the maximal character of the family.
Lemma: The annihilator of any non null element of E(A/p) is a p-primary ideal.
Proof: If A/I ,→ E(A/p) and p̄ is an associated prime of the ideal I, then A/p̄ ,→ A/I (p. 192),
and the annihilator of 0 6= m ∈ (A/p) ∩ (A/p̄) is p̄ = p. Hence I is p-primary.
Theorem: The sheaf I˜ induced by an injective module I is flasque.
411
412 CHAPTER 15. ALGEBRAIC GEOMETRY II
Proof: Since Spec A is noetherian, direct sums of flasque sheaves are flasque (p. 323), and we
may assume that I = E(A/p).
If a prime ideal p0 does not contain p, by the lemma Ip0 = 0, and otherwise I = Ip0 , since the
kernel of the epimorphism (injective modules are divisible) I → Ip0 intersects A/p at 0.
Hence I˜ is the constant sheaf I concentred on (p)0 , which has a dense point, so that it is a
flasque sheaf. q.e.d.
1. This theorem shows again that the sheaves M̃ are acyclic (p. 308), at least when the ring A
is noetherian: if 0 → M → I • is an injective resolution, the flasque resolution 0 → M̃ → I˜•
calculates H p (Spec A, M̃ ), and Γ(Spec A, I˜• ) = I • .
2. If Σ is the field of fractions of a principal ideal domain A, then we have E(A) = Σ and
E(A/m) ' Σ/Am , where A/m ' m−1 /A ,→ Σ/Am .
In particular, the injective hull of the k[x]-module k[x]/(x) is n≥1 kx−n .
L
15.2 Local Algebra

In this section O is a local noetherian ring, m is the maximal ideal, X = Spec O, x ∈ X is the
closed point, k = O/m is the residue field, and M is a finitely generated O-module.
d
If a ∈ m, the complex K1 = Oe −−→ O = K0 , d(e) = a, is denoted by K(a).
If M• is a complex, we put M• (a) = M• ⊗O K(a), and we have exact sequences
π
0 −→ M• −→ M• (a) −−→ M• [1] −→ 0 , π(m + m0 ⊗ e) = m0 ,
·a
. . . −→ Hp+1 (M• (a)) −→ Hp (M• ) −−→ Hp (M• ) −→ Hp (M• (a)) −→ . . . (15.1)
Definitions: The Koszul complex of M and a sequence a1 , . . . ar ∈ m is
KM (a1 , . . . , ar ) = M ⊗O K(a1 ) ⊗O . . . ⊗O K(ar ),
L p1 , . . . , ar )M . If we put L = Oω1 ⊕ . . . ⊕ Oωr , the Koszul

and H0 (KM (a1 , . . . , ar )) = M/(a
complex is just the complex p Λ L ⊗O M , with the differential
j−1 a
P
d(ωi1 ∧ . . . ∧ ωip ⊗ m) = j (−1) ij ωi1 ∧ . . . ∧ ω
bij ∧ . . . ∧ ωip ⊗ m.
A sequence a1 , . . . , ar ∈ m is M -regular (or just regular when M = O) if ai does not divide

0 in M/(a1 , . . . , ai−1 )M ; i.e., for any index i we have an exact sequence
i a
0 −→ M/(a1 , . . . , ai−1 )M −−→ M/(a1 , . . . , ai−1 )M
Theorem: The following conditions are equivalent,
1. The sequence a1 , . . . , ar ∈ m is M -regular.
2. The complex KM (a1 , . . . , ar ) is a free resolution of M/(a1 , . . . , ar )M .
3. H1 (KM (a1 , . . . , ar )) = 0.
Proof: (1 ⇒ 2). When p > 1, we have the exact sequence (15.1)
0 = Hp (KM (a1 , . . . , ar−1 )) −→ Hp (KM (a1 , . . . , ar )) −→ Hp−1 (KM (a1 , . . . , ar−1 )) = 0

15.2. LOCAL ALGEBRA 413
and we see that Hp (KM (a1 , . . . , ar )) = 0 by induction on r. If p = 1, we conclude since ar does

not divide 0 in M/(a1 , . . . , ar−1 )M and we have an exact sequence
·a
0 −→ H1 (KM (a1 , . . . , ar )) −→ M/(a1 , . . . , ar−1 )M −−−r→ M/(a1 , . . . , ar−1 )M (15.2)
(3 ⇒ 1). By the Nakayama’s lemma and the exact sequence (15.1)

·a
H1 (KM (a1 , . . . , ar−1 )) −−−r→ H1 (KM (a1 , . . . , ar−1 )) −→ H1 (KM (a1 , . . . , ar )) = 0
we have H1 (KM (a1 , . . . , ar−1 )) = 0; hence a1 , . . . , ar−1 is M -regular by induction on r.

Now the exact sequence (15.2) shows that a1 , . . . , ar is a M -regular sequence.
Example: If O is a regular ring and df1 , . . . , dfr ∈ m/m2 are linearly independent, then the
ring O/(f1 , . . . , fi ) is regular; hence integral, and the sequence f1 , . . . , fr is regular.
Corollary: If an ideal is generated by a regular sequence, I = (a1 , . . . , ar ), then I/I 2 is a free

O/I-module of rank r.
Proof: Put L = ri=1 Oωi and tensor the exact sequence Λ2 L → L → I → 0 with O/I.
L
Since the differential (Λ2 L) ⊗O O/I → L ⊗O O/I is null, we have L ⊗O O/I ' I/I 2 .
Theorem: If an ideal I is generated by a regular sequence, then the graded ring O[I] = ⊕n I n
is the symmetric algebra of I, and the graded ring GI O = ⊕n I n /I n+1 is the symmetric algebra
of the free O/I-module I/I 2 ,
•
(O/I)[x1 , . . . , xr ] ' SO/I (I/I 2 ) = GI O.
Proof: If I is generated by a regular sequence a1 , . . . , ar , the Koszul complex shows that I is

the quotient of Ox1 ⊕ . . . ⊕ Oxr by the submodule generated by the elements yij = ai xj − aj xi ;
hence S • I is the quotient of the polynomial ring A[x1 , . . . , xr ] by the ideal J = (yij ), and we
must prove that J is just the kernel of the morphism A[x1 , . . . , xr ] −→ ⊕n I n , xi 7→ ai .
Let Pn (x1 , . . . , xr ) be a homogeneous polynomial of degree n such that Pn (a1 , . . . , ar ) = 0.
To show that Pn ∈ J, we proceed by induction on r and n, and we put Ō = O/a1 O.
The reduction P̄n (0, x2 , . . . , xr ) ∈ Ō[x2 , . . . , xr ] fulfills P̄n (ā2 , . . . , ār ) = 0; hence it is in the
ideal generated by āi xj − āj xi . Since Ō[x2 , . . . , xr ] = O[x1 , . . . , xr ]/(a1 , x1 ),
Pn (x1 , . . . , xr ) ≡ a1 Sn (x1 , . . . , xr ) + x1 Tn−1 (x1 , . . . , xr ) (mod. J).
Since a1 xi ≡ ai x1 (mod. J), we see that Pn ≡ x1 Qn−1 (mod. J).

Now, a1 is not a zero divisor and 0 = Pn (a1 , . . . , ar ) = a1 Qn−1 (a1 , . . . , ar ).
Hence Qn−1 (a1 , . . . , ar ) = 0, and Qn−1 ∈ J by induction on n, so that Pn ∈ J.
Finally, GI O = O[I] ⊗O (O/I) = (SO • I) ⊗ (O/I) = S • (I/I 2 ).
O O/I
15.2.1 Regular Rings

Lemma: A finitely generated O-module M is free if and only if TorO
1 (M, k) = 0.
∼
Proof: Let us consider an epimorphism L → M , where L is free and L ⊗O k −
→ M ⊗O k.
The exact sequence 0 → N → L → M → 0 induces an exact sequence
∼
0 = Tor1 (M, k) −→ N ⊗O k −→ L ⊗O k −−→ M ⊗O k −→ 0
so that N ⊗O k = 0. By Nakayama’s lemma N = 0, and M is free.

Definitions: In general, if we have a resolution by free modules Li of finite rank
0 −→ N −→ Ln−1 −→ . . . −→ L0 −→ M −→ 0
and Torn+1 (M, k) = 0, then (p. 336) 0 = Tor1 (N, k) and N is free. The projective dimension
of M is the minimal length of a projective resolution of M . It is the least number n such that
Torp (M, N ) = 0, p > n, for any module N (or just Torn+1 (M, k) = 0). The global dimension
of O is the supremum of the projective dimensions of all finite O-modules, and it is the first
number n such that Torn+1 (k, k) = 0 (or Torp (M, N ) = 0, p > n, for any module N ).
Lemma: If f ∈ O does not divide 0 in O nor in M , then for any O/f O-module N ,
TorO O/f O
p (M, N ) = Torp (M/f M, N ), p ≥ 0.
Proof: TorO
p (O/f O, M ) = 0, p ≥ 1 since by hypothesis we have exact sequences
·f
0 −→ O −−→ O −→ O/f O −→ 0
·f
0 −→ M −−→ M −→ M/f M −→ 0
Now, if L• → M → 0 is a free resolution, then L• /f L• → M/f M → 0 is a resolution by free

O/f O-modules, and
TorO O/f O

p (M, N ) = Hp (L• ⊗O N ) = Hp (L• /f L• ) ⊗O/f O N ) = Torp (M/f M, N ).
Serre’s Theorem: The ring O is regular if and only if it has finite global dimension.
Proof: If O is a regular ring of dimension n, and m = (f1 , . . . , fn ), then the Koszul complex
K(f1 , . . . , fn ) is a free resolution of k, and the differential of K(f1 , . . . , fn ) ⊗O k is null.
Hence Torn (k, k) ' k and Torn+1 (k, k) = 0, and the global dimension of O is n.
We prove the converse by induction on n = dim O.
There is no 0 6= f ∈ m such that f m = 0; in fact, if 0 → Ld → . . . → L0 → k → 0 is a free
resolution, we may assume that Ld ⊆ mLd−1 , and 0 6= f Ld ⊆ f mLd−1 = 0.
Hence m is not one of the associated primes p1 , . . . , pr of the ideal 0.
There exists f ∈ m − (p1 ∪ . . . ∪ pr ∪ m2 ). In fact, if f1 ∈ m − (p2 ∪ . . . ∪ pr ∪ m2 ) and f1 ∈ p1 ,
we take f2 ∈ (p2 ∩ . . . ∩ pr ∩ m2 ) − p1 , and f = f1 + f2 .
Now f ∈ m − m2 and dim (O/f O) = n − 1.
Let us see that O/f O has finite global dimension (hence it is regular).
Since 0 → m/f O → O/f O → k → 0 is exact, we are reduced to show that the projective
dimension of m/f O is finite. Since the exact sequence
0 −→ hf¯i −→ m/f m −→ m/f O −→ 0
splits (a retract is the composition m/f m → m/m2 → hf¯i defined by any supplement of hf¯i in
m/m2 ), we are reduced to see that m/f m has finite projective dimension.
O/f O
We conclude by the lemma, since f is not a zero divisor, Torp (m/f m, k) = TorO
p (m, k).
Corollary: If O is a regular local ring, then Op is regular for any prime ideal p.
O
Proof: The Tor functors localize, Torn p (Op /pOp , Op /pOp ) = TorO
n (O/p, O/p)p = 0.
Definition: A noetherian ring A is regular if so is any local ring Ap .

If the dimension of A is finite, by Serre’s theorem it is equivalent to the existence of a number

n such that TorA
p (M, N ) = 0, p > n, for any pair of finitely generated A-modules M, N .
Corollary: Let A be a finite type k-algebra and k → L an extension. If AL is regular, so is A.
Proof: TorA AL
p (M, N )L = Torp (ML , NL ).
Definition: The height of a prime ideal p of a ring A is the dimension of the local ring Ap .
Lemma: A noetherian domain is a UFD if and only if any height 1 prime ideal p is principal.
Proof: If A is a DFU and p ∈ p is irreducible, then the ideal pA ⊆ p is prime, and pA = p.

Conversely, by noetherianity any element is a product of irreducible elements, and uniqueness
is obvious if we prove that any irreducible element p generates a prime ideal.
Any minimal prime p containing pA has height 1 because dim Ap /pAp = 0.
Hence p = aA, and p = ab. Since p is irreducible, b is invertible, and pA = p is prime.
Theorem: Any regular local ring O is a unique factorization domain.
Proof: We proceed by induction on n = dim O.

We take f ∈ m − m2 , so that O/f O is regular, and f O is a prime ideal.
(1) Of is a DFU: If p ⊂ O is a prime of height 1 and f ∈ / p, since Of is a regular ring of
dimension < n, by induction pf is a line Of -module.
Since p admits a finite resolution by free O-modules, pf admits a finite resolution by free
Of -modules, and pf = Of in the K-group of locally free Of -modules. Since L 7→ Λrk L L is an
additive function with values in Pic(Of ), we see that pf ' Of is principal.
(2) O is a DFU: Let p ⊂ O be a prime of height 1. If f ∈ p, then f O = p. If f ∈ / p, then pf
is principal, pf = pOf , and by noetherianity we may assume that p ∈ p is not multiple of f .
If a ∈ p is not a multiple of p, we have af r = pb, r > 0, b ∈
/ f O, contradicting that f O is
prime. Hence p = pO is principal.
15.2.2 Depth
When A is a noetherian ring, the primary decomposition of ideals may be easily extended to
a·
finitely generated modules: a submodule N ⊂ M is primary if any homothety M/N − → M/N
is injective or nilpotent, and in such a case Ann (M/N ) is a p-primary
T ideal.
Any submodule is an intersection of primary submodules, N = i Ni , the associated primes
pi = Ann (M/NS i ) being the prime ideals coinciding with the annihilator of some element of
M/N , and i pi is formed by the elements of A dividing 0 in M/N .
Lemma: There is a M -regular element if and only if HomO (k, M ) = 0.
Proof: The existence of a M -regular element states that m is not an associated prime of the
ideal 0. Hence no element of M has annihilator m, and HomO (O/m, M ) = 0.
Theorem: There is a M -regular sequence f1 , . . . , fr if and only if ExtpO (k, M ) = 0, 0 ≤ p < r.
Proof: By induction on r. If the sequence f1 , . . . , fr is regular, we have exact sequences

·f1
0 −→ M −−−→ M −→ M/f1 M −→ 0
·f
0 = Extp−1 (k, M/f1 M ) −→ Extp (k, M ) −−−
→ Extp (k, M ), p < r.
1
Since Extp (k, M ) is annihilated by m (consider an injective resolution of M ), it is null.

Conversely, if HomO (k, M ) = 0, by the lemma there is a M -regular element f1 , and the
exact sequence
0 = Extp (k, M ) −→ Extp (k, M/f1 M ) −→ Extp+1 (k, M ) = 0
shows that Extp (k, M/f1 M ) = 0, p < r − 1.
Hence there exists a (M/f1 M )-regular sequence f2 , . . . , fr ; and f1 , f2 , . . . , fr is M -regular.
Definitions: The depth of M is the first integer p such that Extp (k, M ) 6= 0, and the above
argument shows that any M -regular sequence may be extended so as to have length p.
M is Cohen-Macaulay if the depth coincides with the dimension of supp M .
So, O is a Cohen-Macaulay ring when Extp (k, O) = 0, p < dim O.
For example, regular rings are Cohen-Macaulay.
If f1 , . . . , fr is a regular sequence, O is Cohen-Macaulay if and only if so is O/(f1 , . . . , fr ).
O is Cohen-Macaulay if and only if so is O, b because Extp (k, O) ⊗O Ob = Extp (k, O).
b
O b O
Ischebeck’s Theorem: If M is a finitely generated module of depth p and N is a finitely

generated module with support of dimension d, then
ExtnO (N, M ) = 0, n < p − d.
Proof: By induction on d. Taking a filtration of N with quotients ' O/pi (p. 192) we are
reduced to the case N = O/p, and when d = 0 it follows from the above theorem.
If d = dim O/p > 0, and we take f ∈ m − p, then dim O/(p, f ) < d, and by induction
Extn (O/(p, f ), M ) = 0, n < p − d + 1.
·f
Now the exact sequence 0 → O/p −
→ O/p → O/(p, f ) → 0 induces an exact sequence
·f
0 = Extn (O/(p, f ), M ) −→ Extn (O/p, M ) −−→ Extn (O/p, M ) −→ Extn+1 (O/(p, f ), M ) = 0
where n < p − d. Now Extn (O/p, M ) = 0 by Nakayama’s lemma. q.e.d.
1. If O is Cohen-Macaulay, and p is an associated prime of 0, then dim O = dim O/p, and p is
a minimal prime.
HomO (O/p, O) 6= 0; hence 0 ≥ depth O − dim O/p = dim O − dim O/p.
2. If O is Cohen-Macaulay, then dim O = dim (O/p) + dim Op for any prime ideal p.
If p is a minimal prime, then dim O = dim O/p, and we conclude because dim Op = 0. If p is
not minimal, there exists f ∈ p not dividing 0. Hence Ō is a Cohen-Macaulay local ring od
dimension dim O − 1, and we have O/p = Ō/p̄, and dim Ōp̄ = dim Op − 1. We conclude by
induction on dim O.
3. If O is Cohen-Macaulay, so is Op for any prime ideal p; and we may define Cohen-
Macaulay rings to be noetherian rings A such that Ax is Cohen-Macaulay ∀x ∈ Spec A.
If dim Op = 0, then Op is Cohen-Macaulay. If p is not minimal, by (1) there exists f1 ∈ p not
dividing 0. By induction (O/f1 O)p = Op /f1 Op is Cohen-Macaulay; hence so is Op .
4. If O is Cohen-Macaulay of dimension n, and dim O/(f1 , . . . , fn ) = 0, then f1 , . . . , fn is a
regular sequence.
Since dim O/(f1 , . . . , fi ) = n − i, then f1 does not divide 0 and O/f1 O is Cohen-Macaulay.
We conclude by induction on n.
5. Let A → B an injective finite morphism of integral rings. If A is regular and B is Cohen-

Macaulay, then the morphism is flat.
We may assume that A is local of dimension n, and m = (f1 , . . . , fn ). Since the morphism
is finite, dim B/(f1 , . . . , fn ) = 0, and f1 , . . . , fn is a regular sequence at any closed point y
of Spec B since By is Cohen-Macaulay of dimension n (p.201). We conclude that B is a free
A-module: TorA

1 (A/m, B) = H 1 K (f
B 1 , . . . , f n ) = 0.
15.2.3 Local Cohomology

Lemma: Hxp (X, N ) = lim
−→
ExtpO (O/mn , N ), for any O-module N .
Proof: Since the sheaves I˜p are flasque (p. 411), the local cohomology groups may be calculated
with an injective resolution 0 → N → I • . Moreover Γx (X, Ñ ) = lim
−→
HomO (O/mn , N ),
Hxp (X, Ñ ) = H p Γx (X, I˜• ) = lim p Hom (O/mn , I • ) = lim Extp (O/mn , N ).

−→
H O −→ O
Theorem: The depth of M is the first integer p such that Hxp (X, M ) 6= 0, and in such a case
we have that ExtpO (k, M ) ,→ Hxp (X, M ).
Proof: If Exti (k, M ) = 0, then Exti (mn /mn+1 , M ) = 0.
0 −→ mn /mn+1 −→ O/mn+1 −→ O/mn → 0
and the Ext exact sequence shows that Exti (O/mn , M ) = 0, and Hxi (X, M ) = 0.
Moreover, Exti+1 (O/mn , M ) ,→ Exti+1 (O/mn+1 , M ), and Exti+1 (O/m, M ) ,→ Hxi+1 (X, M ).
Corollary: O is Cohen-Macaulay if and only if Hxp (X, O) = 0, p 6= dim O.
Proof: Since H p (X, M ) = 0, p ≥ 1, and H p (X − x, M ) = 0, p ≥ dim (X − x) = dim O − 1, we

always have Hxp (X, M ) = 0, p > dim O.
Lemma: Let Cfl be the category of finite length O-modules. An O-linear contravariant functor
Cfl Cfl , M M ∗ , is representable by an injective hull I of the residue field k if and only if
it is exact and k ' k. In such a case the natural map M → M ∗∗ is an isomorphism for any
∗
O-module M of finite length, we have HomO (I, I) = O b and
I = lim
−→
(O/mr )∗ .
Proof: If M ∗ is exact, by the representability theorem it is representable by an inductive limit

I of modules of finite length, I = lim
−→
HomO (O/mr , I) = lim
−→
(O/mr )∗ .
Let us see, using the ideal criterion, that I is injective. If J is an ideal, any morphism J → I
factors through J/mr J; hence through J/ms ∩ J by Artin-Rees. Since the functor is exact on
finite length modules, the morphism may be extended to O/ms , hence to O.
Now, I is a direct sum of injective hulls of k because supp I = x.
The sum has a unique term when k ∗ ' k.
Conversely, if I is an injective hull of k, it is clear that M ∗ = HomO (M, I) is an exact functor
and k ∗ ' k. Hence k = k ∗∗ and, by induction on the length, M = M ∗∗ for any finite length
module. Finally,
HomO (O/mr , I)∗ , I = lim(O/mr )∗∗ = lim O/mr = O.

HomO (I, I) = lim
←− ←− ←−
b
Theorem: If O is a regular local ring of dimension n, then Hxn (X, O) is an injective hull of the
residue field k.
Proof: Using the Koszul complex, one may easily see that
(
k p=n
ExtpO (k, O) =
0 p 6= n
and, by induction on the length, we have ExtpO (M, O) = 0, p 6= n, for any module of finite length
M . Hence, the functor F (−) = ExtnO (−, O) is exact on the category of finite length O-modules
and F (k) ' k, and it corresponds to an injective hull of k,
I = lim
−→
ExtnO (O/mr , O) = Hxn (O).
Example: If O is a local ring, there is no canonical injective hull of the residue field O/m.
But when O is a local k-algebra and the extension k → O/m is finite, then M ∗ = Homk (M, k)
is an exact functor on the category of finite length O-modules, and we have an isomorphism
O/m ' (O/m)∗ , because the O-module (O/m)∗ is annihilated by m. Hence this exact functor
corresponds to an injective hull lim
−→
(O/mn )∗ of O/m.
So, if A is a local finite k-algebra, the injective hull of A/m is just A∗ = Homk (A, k).
Residues on Curves
Let C be a non singular complete curve over an algebraically closed field k, and O the local
ring at a closed point x ∈ C. Since O is regular, an injective hull of the residue field O/m = k
is I := Hx1 (ΩO ) ' Hx1 (O) ' lim
−→
(O/mn )∗ , and I ∗ ' (lim
−→
(O/mn )∗ )∗ = lim
←−
O/mn = O.
b
In the surface S = C ×k C we have Ω2S = ΩC ⊗k ΩC ; hence, considering the diagonal
embedding ∆ : C → S, we have (Ω2S ⊗ ∆−1 ) ⊗ OC = ΩC ⊗ ΩC ⊗ (ΩC )∗ = ΩC , so that, tensoring
with Ω2S ⊗ ∆−1 the exact sequence 0 → ∆ → OS → OC → 0, we obtain an exact sequence
0 −→ Ω2C×C −→ Ω2C×C ⊗ ∆−1 −→ ΩC −→ 0;
hence a commutative square, where δx is an injective morphism of O ⊗k O-modules because

1
O ⊗k O has depth 2 at (x, x), so that H(x,x) (Ω2C×C ⊗ ∆−1 ) ' H(x,x)
1 (O ⊗k O) = 0,
H 1 (ΩC )
δ / H 2 (Ω2 ) o ∼
H 1 (ΩC ) ⊗k H 1 (ΩC ) (15.3)
O O S O
δx
Hx1 (ΩC ) / H 2 (Ω2 ) o ∼
Hx1 (ΩC ) ⊗k Hx1 (ΩC )
(x,x) S
Lemma: If µ : I ,→ I ⊗k I is an injective morphism of O ⊗k O-modules, then µ∗ : I ∗ ⊗k I ∗ → I ∗

defines a structure of k-algebra on the k-vector space I ∗ . In fact, (O,
b ·) ' (I ∗ , µ∗ ).
Proof: We have µ(k) ⊆ k ⊗k k because µ(k) is annihilated by m ⊗k O + O ⊗k m.

Hence we have commutative squares
µ µ∗
IO / I ⊗k I O b ' I ∗ ⊗k I ∗
b ⊗k O / I∗ ' O
O
b
µ µ∗
k / k ⊗k k k ⊗k k /k
∼ ∼
15.3. QUASI-COHERENT SHEAVES 419
and µ∗ : O
b ⊗k O
b→O b is O
b ⊗k O-linear;
b hence µ∗ (a ⊗ b) = abu for some u ∈ O.
b
u·
Now, u is invertible because µ∗ (k ⊗k k) = k, so that (O,
b µ∗ ) −−→ (O,
b ·) is a k-linear isomor-
phism preserving the product. q.e.d.
Hence δx∗ : Hx1 (ΩO )∗ ⊗k Hx1 (ΩO )∗ → Hx1 (ΩO )∗ defines a canonical ring structure on Hx1 (ΩO )∗ ,
with a unity r : Hx1 (ΩO ) → k providing a purely local definition of the residue. A tedious
calculation of δx would show that it coincides with the local residue defined in p. 319, but at
least it is easy to see that both coincide up to a non null factor:
Corollary: Resx = λr, where 0 6= λ ∈ k.
Proof: The image of Res ∈ H 1 (ΩC )∗ by the natural map H 1 (ΩC )∗ → Hx1 (ΩC )∗ is Resx , by
definition, and dualizing 15.3 we obtain a commutative square1
δ∗
H 1 (ΩC )∗ o H 1 (ΩC )∗ ⊗k H 1 (ΩC )∗
δx∗

Hx1 (ΩO )∗ o Hx1 (ΩO )∗ ⊗k Hx1 (ΩO )∗
so that Res2x = λResx , λ ∈ k, in the ring I ∗ ' O.

b Since O
b is an integral local ring, and Resx 6= 0
dt
(recall that Resx t = 1), we conclude that Resx = λ.
Theorem: Let π : C̄ → C be a morphism between non singular complete curves over k. If π is

unramified at a point x̄ ∈ C̄, for any meromorphic differential form ω on C we have
x = π(x̄) Resx (ω) = Resx̄ (π ∗ ω).
Proof: Let Ō be the local ring of C̄ at x̄.

Since π is unramified, the natural morphisms O/mn → Ō/m̄n , ΩO ⊗O Ō → ΩŌ are isomor-
phisms. Hence so is the morphism Hx1 (ΩO ) → Hx̄1 (ΩŌ ), which is compatible with the connecting,
so that Hx̄1 (ΩŌ )∗ → Hx1 (ΩO )∗ is an isomorphism preserving the product, hence the unity.
It follows that Resx̄ = αResx , 0 6= α ∈ k.
Now, if m = tO, then m̄ = tŌ, so that Resx̄ dtt = 1 = Resx dtt , and α = 1.

dt

Theorem: If t is a local parameter at x ∈ C, then Resx tn = 0, n ≥ 2.
Proof: The morphism t : C → P1 is unramified at x, so that we may assume that x is the origin
of P1 . Now, tdtn has a unique pole at the origin; hence the residue is 0, because the sum of all
residues of a meromorphic form is 0.
15.3 Quasi-Coherent Sheaves

Definition: If j : U → X is an open set and F is a sheaf on X, we put F U = j∗ j ∗ F; i.e.
0
F U (V ) = F(U ∩ V ), so that we have a natural map F U → F U whenever U 0 ⊂ U . Q
Let U = {U1 , . . . , Un } be a finite family of open sets in X. We put Č 0 (U, F) = i F Ui , so
that we have a natural morphism F → Č 0 (U, F). In general we put
Č p (U, F) = F Ui0 ∩...∩Uip

Q
i0 <...<ip
where δ ∗ defines in fact an algebra structure on H 1 (ΩC )∗ , the global residue Res is the unity, and the natural
1
morphism H 1 (ΩC )∗ → Hx1 (ΩC )∗ is a morphism of k-algebras.

p+1
d : Č p F −→ Č p+1 F, (ds)i0 ...ip+1 = (−1)k si0 ...ibk ...ip+1 |Ui0 ∩...∩Uip+1 ; d2 = 0.
P
k=0
and we say that Č • F = Č • (U, F) is the Cech complex of F attached to U.
Theorem: If U = {U1 , . . . , Un } is a finite open cover of X, then the Cech complex defines a
finite resolution of any sheaf of abelian groups F; i.e. we have an exact sequence
d d
0 −→ F −→ Č 0 (U, F) −−−→ . . . −−−→ Č n−1 (U, F) −→ 0.
Proof: When X = Ui (and we may assume that i = 1), we put U0 = {U2 , . . . , Un }, so that
Y Y
Č p (U, F) = F Ui1 ∩...∩Uip × F Ui0 ∩...∩Uip = Č p−1 (U0 , F) × Č p (U0 , F),
1=i0 <i1 <...<ip 16=i0 <...<ip
the differential (in germs) being d(a, b) = (b − d0 a, d0 b), where d0 is the differential of Č • (U0 , F).
If d(a, b) = 0, then b = d0 a and (a, b) = d(0, a). The sequence is exact.
In the general case, since Č • (U, F)|V is just the Cech complex of the sheaf F|V associated
to the open cover {U1 ∩ V, . . . , Un ∩ V } of V , the sequence is exact on any open set Ui .
Definition: A morphism of schemes X → S is said to be separated when the diagonal mor-

phism ∆ : X → X ×S X is a closed embedding, and a scheme X is separated when so is the
morphism X → Spec Z. In such a case, any intersection U ∩ V of affine open sets is affine, since
it is a closed subscheme of the affine scheme U ×Z V .
For example, Spec A and Proj R always are separated schemes.
Theorem: Let M be a quasi-coherent sheaf on a noetherian separated scheme X. For any

cover X = U1 ∪ . . . ∪ Un by affine open sets, we have that the Cech complex Č • M is a finite
quasi-coherent acyclic resolution of M. Moreover, if
0 −→ M0 −→ M −→ M00 −→ 0
is an exact sequence of quasi-coherent sheaves, then we also have an exact sequence
0 −→ Č • M0 −→ Č • M −→ Č • M00 −→ 0.
Proof: When i : U → X is an affine open set, i∗ preserves the cohomology of quasi-coherent

sheaves (pp. 308, 309), so that MU is acyclic, and the functor M MU is exact.
U
Moreover M is quasi-coherent since the restriction to any affine open set V coincides with
j∗ (M|U ∩V ), where j : U ∩ V → V is the inclusion, and U ∩ V is affine.
Corollary: If f : X → S is a morphism of noetherian separated schemes, and M is a quasi-

coherent sheaf on X, the sheaves Ri f∗ M are quasi-coherent.
Proof: We may assume that S = Spec A.

If i : U → X is an affine open set, then the sheaf f∗ MU = (f i)∗ (M|U ) is quasi-coherent
(because f i is a morphism of affine schemes), and MU is f∗ -acyclic.
Hence f∗ (Č p M) is quasi-coherent, and so is Rp f∗ M = Hp f∗ (Č • M) .

Definition: A morphism of schemes φ : T → S is flat when so is any morphism OS,s → OT,t ,

where s = φ(t), so that the functor φ∗ : OS -mod OT -mod preserves exact sequences.
15.3. QUASI-COHERENT SHEAVES 421
Theorem: Given a fibred product of morphisms of noetherian separated schemes
φ̄
XT /X
f¯ f
φ
T /S
and a quasi-coherent sheaf M on X, if the base change φ is flat, then
φ∗ (Ri f∗ M) = Ri f¯∗ (φ̄∗ M).
Proof: To show that the natural morphism φ∗ (Ri f∗ M) → Ri f¯∗ (φ̄∗ M) defined by the inverse
image is an isomorphism, we may assume that S = Spec A and T = Spec B, and we must prove
that H i (X, M) ⊗A B = H i (XB , φ̄∗ M).
Let {Ui } be a finite cover of X by affine open sets.
Then {φ̄−1 (Ui )} is a cover of XB by affine open sets, and
Γ(X, Č • M) ⊗A B = Γ(XB , Č • φ̄∗ M).
Taking cohomology we conclude since B is a flat A-module.
Deligne’s Formula: Let p be a coherent sheaf of ideals on a noetherian scheme X and let
U = X − (p)0 . If N is coherent and M is quasi-coherent,
lim
−→
HomX(pn N , M) = HomU (N |U , M|U ).
Proof: First we assume that X = Spec A, M = M f, N = N e , and p = f A.

If N = A, the annihilator ideal of f stabilizes when n ≥ r, and any element m/f i =
n
f m/f i+r of Mf comes from the (well defined) morphism fmi : f i+r A → M .
r
Hence lim
−→
HomA (f n A, M ) = Mf , and the formula holds when N is free.
If N is not free, N = L/K where L is a free module of finite rank. By the Artin-Rees lemma,
there is an exponent r such that K ∩ f n+r L = f n (K ∩ f r L), and replacing N by f r N we may
assume that there is a presentation L0 → L → N → 0 such that f n L0 → f n L → f n N → 0
remains exact. Taking HomA (−, M ) and lim
−→
, we conclude.
If p = (f1 , . . . , fr ), we put p1 = (f1 ), p2 = (f2 , . . . , fr ). We have the exact sequence
0 −→ pn1 N ∩ pn2 N −→ pn1 N ⊕ pn2 N −→ pn1 N + pn2 N −→ 0.
The filtration pn1 N + pn2 N is equivalent to pn N , and the filtration pn1 N ∩ pn2 N is equivalent
to (p1 p2 )n N , because pn1 pn2 N ⊇ pn+r
1 N ∩ pn2 N ⊇ pn+r
1 N ∩ p2n+r N by the Artin-Rees lemma.
Taking HomA (−, M ) and inductive limit, by induction on r we obtain the following exact
sequence, which let us conclude, where Ui = X − (pi )0 ,
2
HomA (pn N, M ) −→
L
0 −→ lim
−→
HomUi (N |Ui , M|Ui ) −→ HomU1 ∩U2 (N |U1 ∩U2 , M|U1 ∩U2 )
i=1
In general, when X is not affine, the morphism of sheaves

lim
−→
HomX (pn N , M) −→ HomX (N , M)U
induces an isomorphism on sections over any affine open set.

Hence it is an isomorphism of sheaves, and we conclude taking global sections.
Note: Again we obtain (see p. 337) that any injective module I over a noetherian ring A defines
a flasque sheaf, because the morphisms I → lim
−→
HomA (pn , I) = I(U
e ) are surjective.
If X is a scheme, any OX -module M admits an injective morphism M → I into an injective

OX -module, and the functor HomX (−, I) is representable on the category of quasi-coherent
sheaves: there exists a quasi-coherent sheaf Iqc such that for any quasi-coherent sheaf N
HomX (N , Iqc ) = HomX (N , I).
Therefore, Iqc is injective in the category of quasi-coherent sheaves.

When M is quasi-coherent, the morphism M → Iqc is injective, and we see that M admits
a resolution by injective quasi-coherent sheaves.
Lemma: If X is noetherian, any injective quasi-coherent sheaf I is flasque. Moreover, the sheaf
HomX (M, I) is flasque for any quasi-coherent sheaf M.
Proof: Let U be an open set in X. If M is coherent, we have an epimorphism
HomX (M, I) −→ lim

−→
HomX (pn M, I) = HomU (M|U , I|U ).
In general, given a morphism s : M|U → I|U , let N ⊆ M be a maximal submodule so that

there exists a morphism t : N → I such that t|U = s|N |U (it exists by Zorn’s lemma).
If N 6= M, we take N ⊂ N 0 such that N 0 /N is coherent, and an extension t0 : N 0 → I of t.
Since the morphism s − t0 : N 0 |U → I|U vanishes on N |U , and N 0 /N is coherent, there is a
morphism t̄ : N 0 → I such that t̄|U = s − t0 .
Now t0 + t̄ : N 0 → I coincides with s on U , against the maximal character of N .
15.4 K-Theory
All schemes are assumed to be noetherian and separated, K(X) is the K-group (p.177) of
coherent OX -modules and K 0 (X) is the K-group of locally free coherent OX -modules.
If Y is a closed subscheme of X, we put Y = OY = OX/pY ∈ K(X).
If L is locally free, then L0 7→ L ⊗ L0 ∈ K 0 (X) is additive and defines a group morphism
hL : K 0 (X) → K 0 (X), hL (L0 ) = L ⊗ L0 . Now L 7→ hL ∈ End(K 0 (X)) is additive and defines a
group morphism K 0 (X) → End(K 0 (X)); hence K 0 (X) is a ring with the product
L · L0 = L ⊗OX L0 .
Analogously, K(X) is a K 0 (X)-module with the product L · M = L ⊗OX M.

If f : X → S is a morphism of schemes, then L 7→ f ∗ L ∈ K 0 (X) is additive and it defines a
ring morphism
f ! : K 0 (S) −→ K 0 (X), f ! (L) = f ∗ L.
When f is flat, it also defines a group morphism f ! : K(S) → K(X), f ! (M) = f ∗ M.
i π
If f : X → S is a projective morphism (it factors X −→ Pn × S −−2→ S, where i is a closed
embedding) and M is a coherent OX -module, p
P then pthep sheaves R f∗ M are coherent, and nulls
when p 0 (p. 323). The function M 7→ p (−1) R f∗ M ∈ K(S) is additive, and it defines
Grothendieck’s admirable direct image
f! : K(X) −→ K(S), f! (M) = p (−1)p Rp f∗ M.

P
Theorem: (f ◦ g)! = f! ◦ g! .
15.4. K-THEORY 423
P S
Proof: In K-theory, we have M = n M n /M n+1 for any finite filtration M = n Mn , and
p p p p • •
P P
p (−1) M = p (−1) H (M ) for any bounded complex M .
Now the result follows from Leray’s spectral sequence Rp f∗ (Rq g∗ M) ⇒ Rp+q (f g)∗ M,
f! (g! M) = (−1)p+q Rp f∗ (Rq g∗ M) = (−1)n Rn (f g)∗ M = (f g)! M.

P P
p,q n
f ! (s)

Projection Formula: f! · x = s · f! (x).
Proof: The natural morphism L ⊗OS Rp f∗ M → Rp f∗ (f ∗ L ⊗OX M) is an isomorphism when L

is locally free; since it is a local question and the case L = OS is obvious.
Theorem: φ! f! = f¯! φ̄! : K(X) → K(T ), for any flat base change φ : T → S.
Proof: We have φ∗ (Rp f∗ M) = Rp f¯∗ (φ̄∗ M), (see p. 421).
Theorem: The natural morphism K 0 (X) → K(X) is an isomorphism when X is a (noetherian

and separated) regular scheme of finite dimension.
Proof: First we show that any coherent sheaf M is a quotient of a locally free sheaf.
In fact, if U is an affine open set, y is a generic point of Y = X − U and O = OX,y , then
Spec O − y = U ∩ §O = Spec A is affine (X is separated) and the exact sequence
0 = Hy0 (Spec O, O)
e −→ O = H 0 (Spec O, O)
e −→ A = H 0 (Spec O − y, O)
e −→ Hy1 (Spec O, O)
e
shows that Hy1 (Spec O, O) e 6= 0, and O is a regular local ring of depth ≤ 1.

Hence dim O ≤ 1, and the ideal p of Y is a line sheaf (where non null).
There is a morphism ⊕i pni → M surjective on U , since the stalk of M at any point of U is
generated by M(U ) = lim −→
HomX (pn , M). Taking an affine open cover of X we conclude.
Now, by Serre’s P theorem, M admits a finite resolution L• → M → 0 by locally free sheaves,
and the element p (−1)p Lp of K 0 (X) defined by L• does not depend on the resolution.
In fact, if L0• → M → 0 is a resolution and there is an epimorphism L0• → L• (inducing the
p
P
identity on M), the kernel N• is an exact sequence; hence p (−1) Np = 0 and
p 0 p p p
P P P P
p (−1) Lp = p (−1) Lp + p (−1) Np = p (−1) Lp .
In general we take a resolution L00• with an epimorphism onto L• and L0• .

Once we construct the step p − 1, the next one is (taking Zp00 → Zp and Zp00 → Zp0 surjective)
Lp / Zp / Lp−1 d / Lp−2 Zp = Ker d

O O O O
epi
/ Lp ×Z Z 00 ×Z 0 L0 / Z 00 / L00 d00 / L00
L00p p p p p p p−1 p−2 Zp00 = Ker d00

/ Z0 / L0 d0 / L0
L0p p p−1 p−2 Zp0 = Ker d0
Now, given an exact sequence 0 → M0 → M → M00 → 0, we take epimorphisms L00 → M0 ,

L000 → M, and we have a commutative diagram with exact rows,
0 / L0 / L0 ⊕ L00 / L00 /0
0 0 0 0

0 / M0 /M / M00 /0
Repeating the argument with the kernels of the vertical morphisms we obtain resolutions
L0• , L• , L00•
such that Lp = L0p ⊕ L00p .
Hence M 7→ p (−1)p Lp ∈ K 0 (X) is additive, and it defines a morphism K(X) → K 0 (X),
P
inverse of the natural morphism K 0 (X) → K(X). q.e.d.
p q 0 p+q L ⊗ L0q .
P P P
1. If M = p (−1) Lp and N = q (−1) Lq , then M · N = p,q (−1) p
Since L• ⊗ L0• coincides with the alternate sum of the cohomology sheaves in K-theory, when
X is regular of finite dimension, the product of coherent sheaves in K(X) is
M · N = p (−1)p TorO
P
p (M, N ),
X
O (U )
where TorOp (M, N ) is the associated sheaf of the presheaf U
X Torp X (M(U ), N (U )).
On any affine open set U = Spec A it is the sheaf defined by TorA
p (M(U ), N (U )).
2. Analogously f ! M = p (−1)p Lp ⊗OX OY and, when X, Y are regular of finite dimension, the
P
inverse image f ! : K(X) → K(Y ) of coherent sheaves is (and this formula let us also define
the inverse image j ! : K(X) → K(Y ) for any regular closed embedding j : Y → X)
f ! M = p (−1)p TorO
P
p (OY , M).
X
3. If k is a field, the dimension defines an isomorphism K(Spec k) = Z.

When π : X → Spec k is a projective variety,
P π! : K(X) → K(Spec k) = Z coincides with the
Euler-Poincaré characteristic, π! (M) = p (−1)p dim H p (X, M) = χ(X, M).
When π : X → Spec k is a regular projective variety, given closed subschemes Y, Z of com-
plementary codimension, we have a global intersection number
(Y ∩ Z) = π! (Y · Z) = p (−1)p χ(X, TorO
P
p (OY , OZ )),
X
and when Y ∩ Z is 0-dimensional, it is just the number of common points, counting any point
x with the degree [κ(x) : k] and Serre’s intersection multiplicity
O
(Y ∩ Z)x = p (−1)p l(Torp X,x (OY,x , OZ,x )).
P
4. The ideal of a projective plane curve Cn of degree n is ' OP2 (−n). Hence Cn = 1 − OP2 (−n)
in K(P2 ), and we obtain Bézout’s Theorem (p. 323):
(Cn ∩ Cm ) = χ(Cn · Cm ) = χ(1 − O(−n) − O(−m) + O(−n − m))

= 1 − n−1 − m−1 + n+m−1

2 2 2 = nm.
5. If C is a smooth projective curve of genus g, in K(C × C) we have O∆ = 1 − L−∆ and

ΩC = L−∆ − L−2∆ , where ∆ is the diagonal; hence
∆ · ∆ = (1 − L−∆ )2 = 1 − 2L−∆ + L−2∆ = O∆ − ΩC ,

(∆ ∩ ∆) = χ(OC ) − χ(ΩC ) = 2 − 2g.
In higher dimensions, using the Koszul complex, one may see that
O O
Torp X×X (O∆ , O∆ ) = Λp Tor1 X×X (O∆ , O∆ ) = Λp (p∆ /p2∆ ) = ΩpX ,
∆ · ∆ = p (−1)p ΩpX ,
P
(∆ ∩ ∆) = p (−1)p χ(ΩpX ) = p,q (−1)p+q dimk H p (X, ΩqX ).

P P
15.4. K-THEORY 425
Lemma: The sheaves OY , with Y an irreducible closed set, generate the group K(X).
Proof: If M is a coherent sheaf, we proceed by induction on the support and the lengths of
stalks Mx at the generic points x of Z = supp M.
Let p be the ideal of x. If Z has some other generic point, by induction the lemma holds for
M/pM and pM (since px Mx 6= Mx ); hence for M.
If Z = x, then M is annihilated by some power of p, and we only have to prove the lemma
for the quotients pi M/pi+1 M. When pM = 0, we may assume that X = x is integral.
If Mx is the constant sheaf of stalk Mx , the lemma holds for the kernel of M → Mx by
induction, and we may assume that M is torsion free.
Now a rational section 0 6= s ∈ Mx defines an exact sequence
0 −→ I −→ M −→ M/IM −→ 0 , I(U ) = {f ∈ OX (U ) : f s ∈ M(U )},
where the lemma holds for M/IM and OX /I by induction.

Since it holds for OX , we conclude that it also holds for I and M.
Example: When X is a nonsingular curve, for any divisor D we have D = LD − 1 in K(X). In

fact this equality is obvious when D = 0, and if it holds for D so it does for D ± x, as the exact
sequence 0 → LD → LD+x → OX /mx → 0 shows.
Now, the morphism Z ⊕ Pic(X) → K(X), (n, D) 7→ n + D, is an isomorphism, the inverse
being K(X) → Z ⊕ Pic(X), L 7→ (r, Λr L), where r = rk L. In fact, by the lemma (or p. 314)
the line sheaves generate the group K(X). Hence K(k[t]) = Z and K(P1 ) = Z ⊕ Z.
Theorem (Gysin): If j : Y → X is a closed subscheme, and i : U → X is the complementary

open subscheme, we have an exact sequence
!j i!
K(Y ) −−→ K(X) −−→ K(U ) −→ 0
Proof: It is obvious that i! j! = 0, and i! is surjective since any coherent sheaf M on U is the
restriction of a coherent sheaf f on X. In fact, if i∗ M = S Mk , with Mk coherent, then
M
S k
M = (i∗ M)|U = k Mk |U , and M = Mk |U for some index k.
Let us show that Ker i! ⊆ Im j! .
Two coherent extensions M f and Mf0 of M differ in K(X) in sheaves concentred on Y .
In fact, the kernels of M
f → i∗ M and M f0 → i∗ M are concentred on Y , and the images differ
from the sum in sheaves concentred on Y .
Now, if FY (X) is the subgroup of K(X) generated by all sheaves concentred on Y , the
morphism s : K(U ) → K(X)/FY (X), M 7→ M, f is well defined, hence Ker i! ⊆ FY (X).
Finally, Im j! = FY (X) since any coherent sheaf N concentred on Y coincides, in K-theory,
with the graded module of the finite filtration {pnY N }, which is an OY -module.
Definition: The vector and projective bundles of a locally free OX -module E are
π : E = Spec S • E ∗ −→ X
π : P(E) = Proj S • E ∗ −→ X
and the universal property (see p. 322) of the projective bundle P(E) is just

Line quotients Line sub-bundles
HomX (T, P(E)) = =
of E ∗ ⊗OX OT of E ⊗OX OT
An affine bundle of associated vector bundle E is a morphism of schemes P → X endowed

with an action E ×X P → P locally (on X) isomorphic to the additive action of E on E.
Theorem: If E → X is a vector bundle, then π ! : K(X) → K(E) is an isomorphism.
Proof: The morphism π ! is injective since the zero section s : X → E induces a morphism
s! : K(E) → K(X), and s! π ! = Id since (just take a free resolution of M )
TorpA[t1 ,...,tn ] (M ⊗A A[t1 , . . . , tn ], A) = 0, p ≥ 1.
To show that π ! is surjective, by noetherian induction and Gysin exact sequence, we may
assume that X = Spec A, E = AnX , and n = 1.
Let Z be a irreducible closed set in A1X , let q be the corresponding prime ideal in A[t], and
put p = q ∩ A. If q = pA[t], we conclude: A[t]/q = π ! (A/p).
Otherwise q defines a non null prime ideal in (A/p)[t], and it becomes principal after localizing
by some function 0 6= f¯ ∈ A/p, so that q = (p, Q(t)) in Af [t].
Then Af [t]/q = 0 in K-theory, as shows the following exact sequence (where B = Af /pAf )
·Q(t)
0 −→ B[t] −−−−→ B[t] −→ Af [t]/q −→ 0
and Z = 0 in K(A1 × Uf ). By Gysin, Z comes from K(A1Y ), where Y = (f )0 .

By noetherian induction K(Y ) → K(A1Y ) is surjective, and we conclude.
Periodicity Theorem: Let E be a locally free OX -module of rank r + 1, and let us consider
the class xE = 1 − OP(E) (−1) in K 0 (P(E)). We have an isomorphism,
L ∼ P ! i
r+1 K(X) −→ K(P(E)), (a0 , . . . , ar ) 7→ i π (ai )xE .
Proof: If M is a coherent OX -module, then Rp π∗ (π ∗ M ⊗ OP(E) (n)) = M ⊗ Rp π∗ (OP(E) (n)).

According to the cohomology (p. 323) of the sheaves OPr (n), if we put ξ = O(−1),
(
a n=0
π! (π ! (a)ξ n ) =
0 1≤n≤r
Let us see that 1, xE , . . . , xrE are “linearly independent”:

Since xE = 1 − ξ, it is enough to prove that so are 1, ξ, . . . , ξ r .
If π ! (a0 ) + π ! (a1 )ξ + . . . + π ! (ar )ξ r = 0, applying π! we see that a0 = 0.
Now, multiplying by ξ −1 and applying π! , we see that a1 = 0, and so on.
Now let us prove that 1, xE , . . . , xrE “generate”:
By noetherian induction and Gysin we are reduced to an open set where E is trivial.
We consider a hyperplane j : Pr−1 × X → Pr × X = P(E) and we put xr = 1 − OPr (−1).
Now we proceed by induction on r, and we have a Gysin exact sequence
j i!
!
K(Pr−1 × X) −−→ K(Pr × X) −−→ K(Ar × X) −→ 0
By the projection formula, j! (xdr−1 ) = j! (j ! (xdr )) = j! (1)xdr = xd+1 2 r

r , so that xr , xr , . . . , xr
r
“generate” the image of j! , and we conclude because K(A × X) = K(X).
Corollary: If π : P → X is an affine bundle, then π ! : K(X) → K(P ) is an isomorphism.
Proof: If V is the associated vector bundle, the construction of the projective closure of an affine
space (pg. 158, and note that the function 1 defines a linear 1-form on E vanishing on V ) shows
15.4. K-THEORY 427
that we have an exact sequence of locally free OX -modules 0 → V → E → OX → 0 such that

P is just the complement of the closed embedding i : P(V ) → P(E).
The natural morphism OP(E) (−1) → E ⊗OX OP(E) → OP(E) identifies OP(E) (−1) with the
ideal of P(V ). Hence i! (1) = xE , and the projection formula shows that
i! (1) = xE , i! (xV ) = x2E , . . . , i! (xr−1 r

V ) = xE .
! j i!
Now the Gysin exact sequence K(P(V )) −−→ K(P(E)) −−→ K(P ) → 0 and the Periodicity
theorem turn obvious the statement.
Corollary: K(Pr ) = Z[x]/(xr+1 ), where xd = Pn−d .
Proof: By induction on r, since j! ((j ! x)d ) = j! (j ! (xd )) = xd+1 . q.e.d.

d
1. If C is a plane curve of degree d, then C = 1 − (1 − x)d = dx − x2 .

2
2. In Pn we have χ(xr ) = 1, 0 ≤ r ≤ n. Now, if C is a curve of degree d in P3 , we have

χ(C · x) = d, χ(C · x2 ) = χ(C · x3 ) = 0, and C = dx2 + ax3 . Since d + a = χ(C) = 1 − π,
where π is the arithmetic genus, we see that C = dx2 + (1 − d − π)x3 .
If S is a surface of degree d in P3 , then we have χ(S · x2 ) = d and χ(S · x3 ) = 0; hence
S = dx + ax2 + bx3 . Since 1 − π = χ(S · x) = d + a, where π is the arithmetic genus of a
hyperplane section, and 1 − pa = χ(S) = d + a + b, where pa is the arithmetic genus of S, we
see that S = dx + (1 − d − π)x2 + (π − pa )x3 .
3. A rational 1-form ω on P2 defines a line sheaf (ω)(U ) = {f ω ∈ ΩP2 (U ) : f ∈ k(x, y)} and, if
(ω) ' O(−n), we have an exact sequence
∧ω
(*) 0 −→ (ω) −→ Ω1P2 −−−→ Ω2P2 ⊗ O(n) −→ C −→ 0
If the local equation at a point p is ω = h(f dx + gdy), where f, g ∈ Op have no common

factor, then Cp = Op /(f, g). In K(P2 ),
C = δx2 , where δ = p dimk Cp ,
P
Ω1P2 = 3t − 1 = 3(1 − x) − 1 = 2 − 3x,

Ω2P2 ⊗ O(n) = O(n − 3) = (1 − x)3−n ,
since there is an exact sequence 0 → Ω1P2 → O(−1)3 → O → 0. Now (∗) gives
2 − 3x + δx2 = (1 − x)n + (1 − x)3−n 2 − 3x + (n2 − 3n + 3)x2 ,
3 = δ − n2 + 3n.
15.4.1 Graded K-Theory

All schemes will be algebraic varieties over a field k (separated k-schemes of finite type).
Let F p (X) be the subgroup of K(X) generated by coherent sheaves supported on points x
of codimension dim OX,x ≥ p. So we get a filtration of K(X) and we put
GK • (X) = p GK p (X) = p F p (X)/F p+1 (X).

L L
Any irreducible closed set Y of codimension p defines a class [Y ] = [OY ] ∈ GK p (X), and
the class in GK p (X) of a coherent sheaf M with support of codimension p is
lOX,y (My ) [Y ] ∈ GK p (X),

P
[M] =
cod y=p
where Y is the closure of the point y. Hence GK p (X) is a quotient of the group cod y=p Zy of
P
cycles of codimension p, so that it is a group of equivalence classes of cycles.
If f : X → S is a projective morphism and X, S are irreducible, then f! F p (X) ⊆ F p+d (S),
d = dim S − dim X, and it induces a morphism f∗ : GK • (X) → GK • (S) of degree d.
Now we study the compatibility of this filtration with inverse images (obvious in the case of
open embeddings, vector bundles and projective bundles).
Lemma: If L is a line sheaf, then 1 − L ∈ F 1 (X) and (1 − L) · F p (X) ⊆ F p+1 (X), so that it
induces a homogeneous morphism [1 − L] : GK • (X) → GK • (X) of degree 1.
Proof: If codim (supp M) = p, then M and L ⊗OX M coincide in GK p (X) since both have equal
length at any point of codimension p; hence (1 − L)M ∈ F p+1 (X).
Example: Let X be a projective variety of dimension d overa field k, and π : X̄ = Proj ⊕I n → X

the blowup of X along an ideal I ⊂ OX . If E = π −1 (I)0 is the exceptional fibre, defined by
the ideal IOX̄ = OX̄ (1), by the lemma we have E i = 0, i > d, in K(X̄).
Moreover, since I n OX̄ = OX̄ (n), when n 0 we have
π∗ (I n OX̄ ) = I n
Rp π∗ (I n OX̄ ) = 0, p ≥ 1,
so that χ(X, I n ) = χ(X̄, I n OX̄ ). Hence the Samuel function SI (n) = χ(X, OX /I n ) and the
Hilbert function HI (n) = χ(X, I n /I n+1 ) are polynomial functions when n 0,
d

n n n i n
Ei,
P
I OX̄ = (IOX̄ ) = (1 − E) = 1 + (−1)
i=1 i
SI (n) = χ(X, OX /I n ) = χ(X, OX ) − χ(X, I n ) = χ(X, OX ) − χ(X̄, I n OX̄ )
d

n
= χ(X, OX ) − χ(X̄, OX̄ ) − (−1)i χ(X̄, E i )
P
.
i=1 i
d−1

P i i+1 n
HI (n) = ∆SI (n) = SI (n + 1) − SI (n) = (−1) χ(X̄, E ) .
i=0 i
and we obtain a geometric interpretation of the coefficients of the Samuel and Hilbert polyno-
mials S(n) and H(n), the independent term of S(n) being χ(X, OX ) − χ(X̄, OX̄ ).
Moreover, when n 0 we have
n−1
P n−1
P
HI (i) = SI (n) = S(n) = χ(X, OX ) − χ(X̄, OX̄ ) + P (i),
i=0 i=0
P
χ(X, OX ) − χ(X̄, OX̄ ) = HI (n) − P (n) .
n≥0
and we see that χ(X, OX ) = χ(X̄, OX̄ ) when the Hilbert function is polynomial for all n ≥ 0.
For example, if we blow-up a rational point x of a regular surface,
S(n) = dimk (OX,x /mn ) = n+1 = 0 + n1 + n2

2
and we see that χ(X, OX ) = χ(X̄, OX̄ ), that χ(E, OE ) = 1 and that the self-intersection of the
exceptional fibre is (E ∩ E) = −1.
If we blow-up a rational point x of multiplicity m of a hypersurface X of a projective space
n+d−1 n+d−m−1

Pd , the Samuel polynomial (p. 197) is d − d , and the coefficient of degree zero
is (−1)d m

d . In particular, χ(X, OX ) = χ(X̄, O X̄ ) when m < d.
15.4. K-THEORY 429
Periodicity Theorem: Let E be a locally free OX -module of rank r + 1, and let us consider
the morphism xE = [1 − OP(E) (−1)] : GK • (P(E)) → GK • (P(E)). We have an isomorphism
• •
L ∼ P ∗ i
r+1 GK (X) −→ GK (P(E)), (a0 , . . . , ar ) 7→ i π (ai )xE .
Proof: We put x = 1 − OP(E) (−1) ∈ K 0 (P(E)), so that π! (xi ) = 1 when 0 ≤ i ≤ r.

If a ∈ F p (X), then π∗ (π ∗ (a)·xiE ) is the class of π! (π ! (a)xi ) = aπ! (xi ) = a mod. F p+i−r+1 (X),
so that it is null if i < r, and (
0 i<r
π∗ (π ∗ (a) · xiE ) =
a i=r
Now the argument of p. 426 shows that 1, xE , . . . , xrE are “linearly independent”.
Now we take 0 6= [M] ∈ GK p (P(E)), and we know that M = π ! (a0 ) + . . . + π ! (ar )xr .
If 0 6= āi ∈ GK ni (X), and m = min{n
Pi +∗ i} < p,i then the class ofp−iM in GK m (P(E)) is not
null, absurd. Hence m ≥ p, and [M] = i π (āi ) · xE , with āi ∈ GK (X).
Lemma: If i : U → X is an open subscheme, then i! F p (X) = F p (U ); so that i! induces an

epimorphism i∗ : GK • (X) → GK • (U ).
Proof: In the argument of p. 425, if a coherent sheaf M on U has support of codimension

≥ p, then the coherent extension M f ⊆ supp (i∗ M) ⊆ supp M of
f ⊂ i∗ M has support supp (M)
codimension ≥ p, and we conclude.
Corollary: If P → X is an affine bundle, then π ∗ : GK • (X) −→ GK • (P ) is an isomorphism.
Proof: Since π ! : K(X) → K(P ) is an isomorphism compatible with filtrations, it is enough to

show that π ∗ : GK • (X) → GK • (P ) is surjective.
The projective closure i : P → P(E) is an open embedding, so that
i∗ : GK • (P(E)) = GK • (X) ⊕ GK • (X)xE ⊕ . . . ⊕ GK • (X)xrE −→ GK • (P )
is surjective. Since i∗ (xE ) = . . . = i∗ (xrE ) = 0, because i∗ OP(E) (−1) is trivial, we conclude.
Corollary: If s : X → E is the zero section of a vector bundle, then s! (F p (E)) = F p (X), so

that s! induces a morphism s∗ : GK • (E) → GK • (X).
Proof: By the above corollary, π ! F p (X) = F p (E), and s! is the inverse of π ! .
Lemma: If i : H → X is a hypersurface of locally principal ideal p, then i! F p (X) ⊆ F p (H), so

that i! induces a morphism i∗ : GK • (X) → GK • (H). Moreover, i∗ i∗ (x) = [1 − p]x.
Proof: Let Y be a irreducible closed set in X of codimension ≥ p.

By definition i! (OY ) = OY ⊗OX OH − TorO 1 (OY , OH ), and the locally free resolution
X
0 −→ p −→ OX −→ OH −→ 0
shows that i! (OY ) = OY ∩H ∈ F p (H) when H does not contain Y .

Moreover, i! (i! OY ) = OY ∩H = OY − p ⊗ OY = (1 − p) · OY .
If H contains Y , then OY ⊗OX OH = OY and
TorO
1 (OY , OH ) = p ⊗OX OY ,
X
so that i! (OY ) = (1 − p|H ) · OY ∈ F p (H), and i! (i! OY ) = OY − p ⊗ OY = (1 − p) · OY .

Deformation to the Normal Cone: Let Y → X be a closed embedding of ideal p. We put
OX [p] = OX ⊕ p ⊕ . . . ⊕ pn ⊕ . . .
Gp OX = OX /p ⊕ p/p2 ⊕ . . . ⊕ pn /pn+1 ⊕ . . .
and we say that C = Spec Gp OX is the normal cone of Y in X.

The natural morphism Gp OX → OX /p defines the zero section Y → C.
If the embedding Y → X is regular, then (p. 413) we have Gp OX = S • (p/p2 ), and C is the
normal bundle NY /X → Y defined by the locally free OY -module (p/p2 )∗ .
The blowup X̄ = Proj OX [p] → X of X along Y is an isomorphism over X − Y , and the
fibre over Y is just Proj OX [p]/pOX [p] = Proj Gp OX = P(C).
The natural morphism (Gp OX )[x0 ] → Gp OX defines a closed embedding
P(C) → P(1 ⊕ C) = Proj (Gp OX )[x0 ],
with complement Ux0 = C, and we say that P(1 ⊕ C) is the projective closure of C.
Let Z̄ be the blowup of X ×k A1 along Y × 0, and π : Z̄ → A1 the natural morphism.
π −1 (A1 − 0) = X ×k (A1 − 0) since Z̄ → X ×k A1 is an isomorphism out of Y × 0,

π −1 (0) = Proj OX×A1 [p + (t)]/tOX×A1 [p + (t)] = Proj (OX [p] ⊗OX [t])/(pt)
= Proj (OX [p] ⊗OX [t])/(p) ∪Proj (OX [p] ⊗OX [t])/(t)
= Proj (Gp OX ) ⊗OY OY [t] ∪ Proj OX [p] = P(1 ⊕ C) ∪ X̄.
Moreover, P(1 ⊕ C) ∩ X̄ = P(C). If we remove X̄, the fibre over 0 is the normal cone C.
Hence, if we put Z = Z̄ − X̄, we have a commutative triangle (an algebraic variant of the
Tubular Neighborhood Lemma) where π is flat and j is a closed embedding,
j
Y × A1 /Z
π
A1
1. π −1 (A1 − 0) = X × (A1 − 0), and j : Y × (A1 − 0) → X × (A1 − 0) is the obvious morphism.
2. π −1 (0) = C, and the embedding j : Y × 0 → C is the zero section.
Lemma: If i : Y → X is a regular closed embedding, then i! F p (X) ⊆ F p (Y ), so that i! induces

a morphism i∗ : GK • (X) → GK • (Y ).
Proof: j1 : Y → Z is the composition of i : Y → X, the section X × 1 → X × (A1 − 0), and the

open embedding X × (A1 − 0) → Z. Since the maps F p (Z) → F p (X × (A1 − 0)) → F p (X) are
surjective, we are reduced to prove the lemma for j1! : K(Z) → K(Y ).
But the morphisms j1! , j0! : K(Z) → K(Y × A1 ) ⇒ K(Y ) coincide since both morphisms
∼
K(Y × A1 ) ⇒ K(Y ) are just the inverse of the isomorphism π ! : K(Y ) −
→ K(Y × A1 ).
Hence we are reduced to prove the lemma for j0! : K(Z) → K(Y ); but j0 : Y → Z is the
composition of the zero section Y → NY /X with the closed embedding NY /X → Z, defined by a
locally principal ideal, and the lemma holds for both morphisms.
Definition: An algebraic variety X of dimension d over a field k is smooth if the diagonal

∆ : X → X ×k X is a regular embedding of codimension d.
15.4. K-THEORY 431
Theorem: If X is a smooth variety, the product of K(X) is compatible with the filtration,
F p (X) · F q (X) ⊆ F p+q (X), and it induces a ring structure on GK • (X).
If f : Y → X is a morphism of smooth varieties, then f ! (F p (X)) ⊆ F p (Y ), so that f ! induces
a ring morphism f ∗ : GK • (X) → GK • (Y ).
Proof: If Y, Z are integral subvarieties of codimension p, q, we must show that Y · Z ∈ F p+q (X).
If ∆ : X → X ×k X is the diagonal morphism, then Y · Z = ∆! (Y ×k Z) since the formula
L0 ⊗OX L = ∆∗ (L0 ⊗k L) is immediate for locally free sheaves, which generate K(X).
Since Y ×Z is of codimension p+q and ∆ is a regular embedding, the lemma let us conclude.
Finally, f : Y → X is the composition of the graph 1 × f : Y → Y ×k X with the projection
π : Y ×k X → X, and both (1 × f )! and π ! are compatible with the filtrations.
The first since 1 × f is a regular closed embedding, and the second since the codimension of
Y ×k Z in Y ×k X coincides with the codimension of Z in X.
Definition: Let X be a smooth variety. The Chern classes ci (E) ∈ GK i (X) of a locally free
OX -module E of rank r are the coefficients of the unique relation
xrE + c1 (E)xr−1
E + . . . + cr (E) = 0
in GK • (P(E)), and the Chern classes of X are those of the tangent bundle TX = (Ω1X )∗ .
The proofs given in the topological case (p. 361), but now ξE = OP(E) (−1), show that the
Chern class of a line sheaf is c1 (L) = −[1 − L] = [1 − L∗ ], that Chern classes are functorial,
ci (f ∗ E) = f ∗ ci (E), that the total class c(E) = 1+c1 (E)+. . .+cr (E) satisfies Whitney’s formula,
and that the last Chern class is cr (E) = s∗ [s0∗ (1)] for any global section s : X → E.
1. If π : X → Spec k is a projective smooth variety, we have a degree morphism

deg = π∗ : GK d (X) = F d (X) −→ Z, deg [M] = χ(X, M),
where d = dim X; hence a pairing GK p (X) ⊗ZGK d−p (X) → Z, hz1 , z2 i = deg(z1 z2 ), defining
global intersection numbers of cycles of complementary codimension,
h[M], [N ]i = i (−1)i χ(X, TorO
P
i (M, N )).
X
2. We have a ring isomorphism GK(Pd ) = Z[x]/(xd+1 ), where x = [1 − O(−1)] ∈ GK 1 (Pd ), and
the exact sequence 0 → Ω1Pd → OPd (−1)d+1 → OPd → 0 shows that ci (Pd ) = d+1
i x .
i
3. If E is a locally free OX -module of rank r, then c1 (E) = c1 (Λr E) and ci (E ∗ ) = (−1)i ci (E).
4. deg cn (X) = p,q (−1)p+q dim H p (X, ΩqX ) for any smooth projective variety X of dimension
P
n, and cd (NY /X ) = i∗ (i∗ (1)) for any smooth closed subvariety i : Y → X of codimension d.
In fact, the Koszul complex shows that TorO i 2
i (OY , OY ) = Λ (pY /pY ); hence
X
i∗ (i∗ (1)) = i (−1)i TorO i i 2 2 ∗

P P
i (OY , OY ) =
X
i (−1) Λ (pY /pY ) = cd ((pY /pY ) ).
15.4.2 Cohomology theories and Chern Classes

Jouanolou’s Trick: Let X be a quasi-projective k-variety (an open subscheme of a projective
k-scheme). There exists an affine bundle π : P → X such that P is an affine scheme. Hence,
for any line sheaf L on X, we have π ∗ L ' f ∗ OPn (1) for some k-morphism f : P → Pn .
Proof: The incidence divisor H in P(E) × P(E ∗ ), of equation i ui xi = 0, is an hyperplane

P
section of the natural embedding P(E) × P(E ∗ ) → P(E ⊗ E ∗ ); hence the complement U is an
affine scheme, and the natural projection π : U → P(E ∗ ) is an affine bundle of associated vector
bundle O(−1) ⊗ O(−1)o . (See ex. 7 in p. 492.)
Alternative Proof : The rank 1 idempotent endomorphisms, T 2 = T and cT (x) = xn (x − 1),
define a closed (hence affine) subscheme P of Endk E such that the morphism π : P → P(E),
π(T ) = Im T , is an affine bundle of associated vector bundle Hom(E/O(−1), O(−1)).
If X → P(E) is a closed subscheme, then PX = π −1 (X) → X is the required affine bundle.
Now, if U → X is an open subscheme, blowing up the complement, we may assume that
Y = X − U is defined by a locally principal ideal; hence so is the ideal of PY in PX , so that
PU → PX is an affine morphism, and we see that PU is an affine scheme.
Finally, π ∗ L is generated (as any coherent sheaf on an affine scheme) by a finite number of
global sections, and we have π ∗ L ' f ∗ OPn (1) by the universal property of Pn .
Definition: A cohomology theory is a contravariant functor A from the category of smooth

quasi-projective k-varieties into the category of commutative rings, endowed with a functorial
morphism of A(X)-modules f∗ : A(Y ) → A(X) for projective morphisms f : Y → X (Id∗ = Id,
(f g)∗ = f∗ g∗ and the projection formula f∗ (f ∗ (x)y) = xf∗ (y) holds); hence a Chern class
cA ∗
1 (L) = s0 (s0∗ (1)) ∈ A(X) of line bundles L → X (where s0 : X → L is the zero section), and
a cohomology class [Y ]A = i∗ (1) ∈ A(X) of smooth closed subvarieties i : Y → X, such that
1. i∗1 + i∗2 : A(X1 ⊕ X2 ) → A(X1 ) ⊕ A(X2 ) is an isomorphism; hence A(∅) = 0.
2. π ∗ : A(X) → A(P ) is an isomorphism for any affine bundle π : P → X.

Hence c1 (L) = s∗ (s0∗ (1)) for any section s of a line bundle L, because s∗ = s∗0 , both being
the inverse of π ∗ : A(X) → A(L).
3. For any smooth closed subvariety i : Y → X, with complement j : U → X, we have an exact

sequence
∗ i j∗
A(Y ) −−− → A(X) −−−−→ A(U )
4. If E → X is a vector bundle of rank r + 1, and we put yE = cA

1 (OP(E) (−1)), then
r.
A(P(E)) = A(X) ⊕ A(X)yE ⊕ . . . ⊕ A(X)yE
5. For any projective bundle π : P(E) → X and any morphism f : Y → X, we have a commuta-
tive square
f∗
A(P(E)) / A(P(f ∗ E))
π∗ π∗
f∗
A(X) / A(Y )
6. If f : X̄ → X is transversal to a smooth closed subvariety i : Y → X, in the sense that

Ȳ = Y ×X X̄ is smooth and the natural morphism f ∗ NY /X → NȲ /X̄ is an isomorphism, we
have a commutative square (in particular, f ∗ [Y ] = 0 when f −1 (Y ) = ∅)
f∗
A(Y ) / A(Ȳ )
i∗ i∗
f∗

A(X) / A(X̄)
15.4. K-THEORY 433
and a morphism of cohomology theories ch : A → Ā is a natural transformation preserving

direct images. That is to say, ch : A(X) → Ā(X) is a ring morphism preserving inverse images,
ch(f ∗ (a)) = f ∗ (ch(a)), and direct images, ch(f∗ (a)) = f∗ (ch(a)). Remark that:
1. Both i1 , i2 are transversal to i1 , i2 ; hence i∗2 i1∗ = 0, i∗1 i2∗ = 0, i∗1 i1∗ = Id, i∗2 i2∗ = Id,
so that the inverse of i∗1 + i∗2 : A(X1 ⊕ X2 ) − →∼
A(X1 ) ⊕ A(X2 ) is just i1∗ + i2∗ . Hence
f∗ = f1∗ + f2∗ : A(X1 ) ⊕ A(X2 ) → A(S) when f = f1 ⊕ f2 : X1 ⊕ X2 → S is projective.
2. The zero section s0 : X → L of a line bundle L is transversal to the natural morphism

f ∗ L → L, for any morphism of schemes f : X̄ → X; hence f ∗ c1 (L) = c1 (f ∗ L).
3. In the projective bundle π : P(E) → X of a locally free OX -module of rank r we have that
OP(E) (−1) is a submodule of π ∗ E, and Ē = (π ∗ E)/O(−1) is locally free of rank r − 1. Since
π ∗ : A(X) → A(P(E)) is injective by condition 4, by induction on r we obtain the Splitting
Principle: There is a base change π : X̄ → X such that π ∗ : A(X) → A(X̄) is injective and
π ∗ E = L1 + . . . + Lr is a sum of line sheaves in K(X̄).
4. Given a smooth closed hypersurface Y → X, the line sheaf LY admits a section transversal
to the zero section and vanishing at Y . Hence c1 (OX ) = 1, and c1 (LY ) = [Y ].
Examples: Let Lx be a line sheaf of Chern class x. We say that A follows the additive law x + y
when c1 (Lx ⊗ Ly ) = x + y, and the multiplicative2 law x + y − xy when c1 (Lx ⊗ Ly ) = x + y − xy.
1. The K-theory K(X) is a cohomology theory. Condition 6 holds since Y is locally defined by
a regular sequence which is a regular sequence in X̄, so that for any locally free OY -module
L we have TorOn (OX̄ , L) = 0, n ≥ 1.
X
The cohomology class of a smooth closed subvariety i : Y → X is i! (1) = OY ∈ K(X).

If L is a line OX -module, then cK ! ∗
1 (L) = s0 (s0! (1)) = 1 − L ∈ K(X). The K-theory follows
the law x + y − xy since 1 − (L ⊗ L) = (1 − L ) + (1 − L) − (1 − L0 )(1 − L).
0 0
2. The rational graded K-theory GK • (X) ⊗Z Q is a cohomology theory, but we shall not prove
it, since the proofs of condition 3 that we know involve the theory of the Chow ring, that we
deliberately avoid in these notes.
The Chern class of a line bundle L is cGK ∗ 1
1 (L) = [1 − L ] ∈ GK (X), and the theory follows
0 2
the additive law x + y because (1 − L )(1 − L) ∈ F (X).
3. In the complex case, A(X) = H 2• (Xan , Q) = ⊕p H 2p (Xan , Q) defines a cohomology theory

following the additive law x+y (where Xan denotes the topological space of all closed points of
X, with the coarsest topology such that the maps f : Uan → C, f ∈ OX (U ), are continuous).
For an arbitrary projective morphism f : Y → X, the projection formula states that the direct
image f∗ may defined to be the adjoint of the inverse image f ∗ : Hc2• (Xan , Q) → Hc2• (Yan , Q)
by Poincaré’s duality. Condition 6 holds since the morphism f ∗ : f ∗ TY /X → TȲ /X̄ defined
by the inverse image is an isomorphism (use the argument of p. 352, showing that the
multiplicity is ±1 in any transversal intersection).
4. If A is a cohomology theory defined on the k-varieties, so is X A(X) ⊗Z Q, and we denote

it A ⊗ Q. Moreover, for any field extension k0 → k we have that X A(Xk ) is a cohomology
theory on the k0 -varieties.
2
The group law of the multiplicative group, since we have (1 − x)(1 − y) = 1 − (x + y − xy).
Definition: The Chern classes of a locally free OX -module E of rank r are the coefficients
cA
i (E), or simply ci (E), of the characteristic polynomial cE (y) of the endomorphism of the free
A(X)-module A(P(E)) defined by the product with yE = c1 (OP(E) (−1)),
cE (t) = y r − cA
1 (E)y
r−1
+ . . . + (−1)r cA
r (E) ∈ A(X)[t],
up to a sign that we introduce so that the first Chern class of any line sheaf L coincides with
the previous one. In fact, P(L) = X and OP(L) (−1) = L.
Functoriality: ci (f ∗ E) = f ∗ (ci (E)) for any morphism f : X̄ → X.
Proof: We have f ∗ c1 (OP(E) (−1)) = c1 (f ∗ OP(E) (−1)) = c1 (OP(f ∗ E) (−1)).
Additivity: For any exact sequence 0 → E1 → E → E2 → 0 of locally free OX -modules,

P
cn (E) = ci (E1 ) · cj (E2 ).
i+j=n
Proof: If rk E1 = 1, then i : X = P(E1 ) → P(E) is a section of P(E) → X and i∗ is injective.

Let us consider the complement j : U → P(E) of P(E1 ). The map U → P(E2 ) is an affine
bundle of the vector bundle Hom(OP(E2 ) (−1), E1 ), so that j ∗ : A(P(E)) → A(U ) = A(P(E2 )) is
surjective (in fact, j ∗ (yE
n ) = y n ). Therefore we have a commutative diagram with exact rows
E2
(since yE1 = i∗ yE , the first square commutes by the projection formula)
i∗ j∗
0 / A(P(E1 )) / A(P(E)) / A(P(E2 )) /0
yE1 yE yE2
i∗
j∗
0 / A(P(E1 )) / A(P(E)) / A(P(E2 )) /0
and the additivity of the characteristic polynomial shows that cE (y) = cE1 (y)cE2 (y).
Now we proceed by induction on rk E1 , and by the splitting principle we may assume the
existence of a line sheaf L ⊂ E1 such that Ē1 = E1 /L and Ē = E/L are locally free.
We have an exact sequence 0 → Ē1 → Ē → E2 → 0 and we conclude,
cE (y) = cL (y)cĒ (y) = cL (y)cĒ1 (y)cE2 (y) = cE1 (y)cE2 (y).
Definition: Since 1 + c1 (E)t + . . . + cr (E)tr is an additive function with values in the multi-
plicative group of invertible formal series with coefficients in A(X), it induces a group morphism
on K(X), so defining Chern classes ci (x) ∈ A(X) for every x ∈ K(X).
Corollary: The cohomology ring of the projective space is A(Pd ) = A(Spec k)[x]/(xd+1 ), where
x corresponds to the cohomology class xd = [Pd−1 ] of any hyperplane.
Proof: Put yd = c1 (OPd (−1)). By additivity, trivial bundles have null Chern classes.
Hence in A(Pd ) = A(Spec k) ⊕ A(Spec k)yd ⊕ . . . ⊕ A(Spec k)ydd we have ydd+1 = 0.
In A(P1 ), the exact sequence 0 → OP1 (−1) → OP2 1 → OP1 (1) → 0 shows that x1 = −y1 .
Considering a line in Pd , we see that xd = −yd + a2 yd2 + . . . + ad ydd in A(Pd ).
We conclude that xd+1
d = 0 in A(Pd ).
Corollary: Chern classes are always nilpotent.

15.4. K-THEORY 435
Proof: Let L → X be a line bundle. By Jouanolou’s trick there is an affine bundle π : P → X

such that π ∗ L = f ∗ (O(1)) for some morphism f : P → Pd . Therefore π ∗ c1 (L) = f ∗ (xd ) is
nilpotent, since so is xd , and c1 (L) is nilpotent because π ∗ is an isomorphism.
We conclude since any Chern class, after a base change injective in cohomology, is a sum of
products of Chern classes of line bundles.
Example: To compute the Chern classes in the K-theory of a locally free OX -module E of rank
r, we may assume that it is a sum of line sheaves, E = L1 + . . . + Lr ∈ K(X); hence,
∗
cK ∗
P K P
1 (E) = i c1 (Li ) = i (1 − Li ) = rk E − E , (15.4)
∗ 2 ∗ r r ∗
cK K ∗
Q Q
r (E) = i c1 (Li ) = i (1 − Li ) = 1 − E + Λ E + . . . + (−1) Λ E .
Definition: Given a locally free OX -module E, there is a base change π : X̄ → X such that
π ∗ : A(X) → A(X̄) is injective and π ∗ E = Lα1 + . . . + Lαr is a sum of line sheaves in K(X̄), so
that ci (E) is the
P i-th elementary symmetric function of the ”roots” α1 , . . . , αr . For any formal
series F (t) = n an tn with coefficients in A(Spec k) we put
F+ (E) = F (α1 ) + . . . + F (αr ) ∈ A(X),
(where we view the sum as a power series in the elementary symmetric functions ci (E), and the
sum is finite because Chern classes are nilpotent) and F+ is an additive function, so defining a
functorial group morphism F+ : K(X) → A(X) named additive extension of F . Analogously,
when a0 is invertible, we have a multiplicative extension F× : K(X) → A(X)∗ such that
F× (E) = F (α1 ) · . . . · F (αr ) ∈ A(X)∗ .
15.4.3 Grothendieck’s Riemann-Roch Theorem

Universal Property: If a cohomology theory A follows the group law x+y−xy of the K-theory,
then there is a unique morphism of cohomology theories ch : K → A.
Proof: If E is a locally free OX -module, by 15.4 we have E = rk E − cK ∗

1 (E ) in K(X).
Hence the unique possible morphism of cohomology theories ch : K(X) → A(X) is
∗
ch(E) = rk E − cA
1 (E ). (15.5)
This function ch is additive on the locally free OX -modules, since so are the rank and cA
1;
hence it defines a group morphism ch : K(X) → A(X), and it commutes with inverse images
because so do the rank and cA 1.
It preserves products of line bundles because A follows the law x + y − xy:
∗ ∗ ∗ ∗ ∗ A ∗
ch(L1 · L2 ) = 1 − cA A A A
1 (L1 ⊗ L2 ) = 1 − c1 (L1 ) − c1 (L2 ) + c1 (L1 )c1 (L2 )
∗ ∗
= 1 − cA A

1 (L1 ) 1 − c1 (L2 ) = ch(L1 ) · ch(L2 ).
Now, by the splitting principle, ch preserves arbitrary products.

We only have to prove that ch preserves direct images, and the theorem follows from the
following lemma, since ch preserves Chern classes of line sheaves,
∗ ∗
ch cK A A

1 (L) = ch(1 − L ) = 1 − ch(L ) = 1 − (1 − c1 (L)) = c1 (L).
Panin’s Lemma: If a functorial ring morphism ch : A → Ā between cohomology theories pre-

serves the cohomology class of hyperplanes, ch[cA Ā
1 (OPd (1))] = c1 (OPd (1)), then it preserves direct
images; i.e., for any projective morphism f : Y → X,
ch(f∗ (a)) = f∗ (ch(a)) , a ∈ A(Y ).
Proof: By Jouanolou’s trick, the natural transformation ch preserves Chern classes of line
sheaves; hence it also preserves cohomology classes of smooth hypersurfaces.
By definition, any projective morphism f : Y → X is the composition of a closed embedding
Y → Pn × X with the natural projection π2 : Pn × X → X.
If the lemma holds for two morphisms, so it holds for the composition; hence we must prove
the lemma for a closed embedding i : Y → X and the projection π2 : Pn × X → X.
1. If the lemma holds for the zero section s : Y → N̄ = P(1 ⊕ NY /X ) of the projective closure of
the normal bundle, then it holds for i : Y → X.
The deformation Z̃ to the normal cone, without removing the blowup X̃ of X (p. 413), gives
a (commutative by condition 6) diagram, where U = Z̃ − (Y × A1 ),
Ā(U )
O
j∗
i∗0
Ā(N̄ ) o Ā(Z̃ )
O O
s∗ i∗
i∗0
Ā(Y ) o ∼ Ā(Y × A1 )
So we see that (Ker i∗0 ) ∩ (Ker j ∗ ) = 0, because the column is exact by condition 3 and s∗
is injective (since p∗ s∗ = Id, where p : N̄ → Y is the natural projection). Now we have a
commutative diagram, where the lemma states the coincidence of the vertical pairs,
Ā(U )
↑j ∗
0 i∗ i∗
Ā(N̄ ) ←−− Ā(Z̃ ) −−1→ Ā(X)
↑↑ ↑↑ ↑↑
A(Y ) == A(Y × A1 ) == A(Y )
The difference of the first par is null by hypothesis; hence so is the difference of the central
pair as we have seen, and we conclude that the difference of the last pair is null.
2. If L is a line sheaf over Y , the lemma holds for the zero section s : Y → L̄ = P(1 ⊕ L); hence
it holds for every closed embedding Y → X of codimension 1.
We have ch(s∗ (1)) = s∗ (1), because Y is an hypersurface in L̄. Now, if a ∈ A(Y ), we may
put a = s∗ b because s∗ : A(L̄) → A(Y ) is surjective, and
ch(s∗ a) = ch(s∗ s∗ b) = ch(bs∗ (1)) = ch(b)s∗ (1) = s∗ (s∗ ch(b)) = s∗ (ch(a)).
3. If E is a vector bundle on Y , the lemma holds for the zero section s : Y → Ē = P(1 ⊕ E);
hence it holds for every closed embedding Y → X.
If E admits a filtration {Ei } with quotients Ei /Ei−1 of rank 1, the lemma holds for the zero
section Y → Ē1 and for the morphisms Ē1 → Ē2 → . . . → Ēr = Ē; hence it holds for the
composition s : Y → Ē.
In general, we have a morphism π : Y 0 → Y such that π ∗ is injective and E 0 = π ∗ E admits
such filtration. The lemma holds for the zero section s0 : Y 0 → Ē 0 , and we conclude applying
condition 6 to the morphisms π : Ē 0 → Ē and s : Y → Ē,
π ∗ s∗ (ch(a)) = s0∗ π ∗ ch(a) = s0∗ ch(π ∗ a) = ch(s0∗ π ∗ a) = ch(π ∗ s∗ a) = π ∗ ch(s∗ a).
4. If the lemma holds for Pn → Spec k, so it does for π2 : Pn × X → X, by condition 5.
5. The lemma holds for the projection p : Pn → Spec k onto a point, and we conclude.
15.4. K-THEORY 437
Proof: We consider the embedding i : Pn−1 → Pn , and we put
A = A(Spec k), xn = c1 (OPn (1)) = i∗ (1) ∈ A(Pn )

Ā = Ā(Spec k), x̄n = c̄1 (OPn (1)) = ī∗ (1) ∈ Ā(Pn )
By hypothesis ch(xn ) = x̄n ; hence ch(xrn ) = x̄rn , and the ring morphism ch : A(Pn ) → Ā(Pn )
induces an isomorphism of Ā-algebras A(Pn ) ⊗A Ā = Ā(Pn ). We have to prove that the 1-form
p̄∗ : Ā(Pn ) → Ā is just the base change of the 1-form p∗ : A(Pn ) → A.
Now, if we consider the cohomology class ∆n = ∆∗ (1) ∈ A(Pn × Pn ) = A(Pn ) ⊗A A(Pn ) of
the diagonal embedding ∆ : Pn → Pn × Pn , we have
(p∗ ⊗ 1)(∆n ) = π∗ ∆∗ (1) = Id∗ (1) = 1,
where π : Pn ×Pn → Pn stands for the second projection. That is to say, by means of the polarity
ω 7→ (ω ⊗ 1)(∆n ) defined by the diagonal, p∗ corresponds to the unity.
According to the following proposition, this equality fully determines the 1-form p∗ , and
since the cohomology class of the diagonal is stable under the base change A → Ā (because the
lemma holds for the diagonal embedding), we conclude that p∗ also is stable.
Proposition: The polarity A(Pn )∗ → A(Pn ), ω 7→ (ω ⊗ 1)(∆n ), is an isomorphism.
Proof: By induction on n we shall prove that

 
0 0 1
n
•
 
X
ars xrn xsn

∆n = ⊗ =
 0


r,s=0  
1 • •
where ars = 0 when r + s < n, and ars = 1 when r + s = n. In fact,
i∗ (xrn−1 ) = i∗ i∗ (xrn ) = xrn · i∗ (1) = xr+1

n ,
∗ ⊗1)(∆ ) is the cohomology class in P
and by condition 6 we have that (iP n n−1 ×Pn of the diagonal
of Pn−1 . Now, if we put ∆n−1 = rs a0rs xrn−1 ⊗ xsn−1 , then
(i∗ ⊗ 1)(∆n ) = r,s ars xrn−1 ⊗ xsn

P
(1 ⊗ i∗ )(∆n−1 ) = r,s a0rs xrn−1 ⊗ xs+1

P
n
By induction, we obtain the stated result for the coefficients ars , r < n.
By symmetry it also holds for ars , s < n, and we conclude. q.e.d.
Now, given the functor A, the direct image may be modified: Let F× be the multiplicative
extension of an invertible series F = a0 + a1 t + . . .. For any projective morphism f : Y → X we
consider the virtual relative tangent bundle Tf = TY − f ! TX ∈ K(Y ) and we put
f∗new (a) = f∗ F× (−Tf ) a = F× (TX )f∗ F× (TY )−1 a ∈ A(X)

(15.6)
so that the following square (with vertical isomorphisms) is commutative:

f∗
A(Y ) / A(X)
·F× (TY ) o o ·F× (TX )

f∗new
A(Y ) / A(X)
Lemma: (A, f∗new ) is also a cohomology theory, and cnew

1 (Lx ) = xF (x).
Proof: In the case of the zero section s0 : X → Lx of a line bundle, we have −Ts0 = Lx , and s∗0
is surjective; hence:
∗ new ∗ ∗
cnew
1 (Lx ) = s0 (s0∗ (1)) = s0 [s0∗ (F× (Lx ))] = s0 [s0∗ (F (x))] = F (x) · s0∗ (1) = F (x) · x.
Moreover, all the conditions of a cohomology theory are easy to check, except condition 4:
Put y = yE ∈ A(P(E)) and z = yE new = yF (y) = a y + a y 2 + . . . ∈ A(P(E)).
0 1
Then z = a0 y + . . . and we have z d = ad0 y d when y d+1 = 0.
n n n
Since the powers of y generate the A(X)-module A(P (E)), so do the powers of z.
Now, A(P (E)) is a free A(X)-module of rank r + 1; hence 1, z, . . . , z r define a basis (consider
the characteristic polynomial of the endomorphism of A(P(E)) defined by z). q.e.d.
Now, if A follows the additive law x + y, we may modify the direct image of A ⊗ Q with an
exponential so that it follows the multiplicative law x + y − xy of the K-theory.
Since eax = 1 − (1 − eax ) and 1 − eax = −ax + . . ., it is convenient to fix a = −1.
Hence we modify the direct image of A ⊗ Q with the invertible series
1 − e−t t2 t3 t4

t
F (t) = = 1 − + − + − ...
t 2! 3! 4! 5!
new −x
c1 (Lx ) = xF (x) = 1 − e
and we obtain a new cohomology theory (A⊗Q)new following the x+y−xy law. By the universal
property of the K-theory we have a morphism of cohomology theories ch : K → (A ⊗ Q)new and,
by 15.5 (pg. 435) it is just the additive extension of the series et , named Chern character,
∗
ch(Lx ) = 1 − cnew new x x
1 (Lx ) = 1 − c1 (L−x ) = 1 − (1 − e ) = e .
Grothendieck’s Riemann-Roch Theorem: If a cohomology theory A on the smooth quasi-

projective varieties over a field follows the additive law x + y, then for any projective morphism
f : Y → X we have the following commutative square, where the Todd class Td is the multi-
t2 t4
plicative extension of the formal series F (t)−1 = 1−et −t = 1 + 2t + 12 − 720 + . . .,
f!
K(Y ) / K(X)
Td(TY )·ch Td(TX )·ch

f∗
A(Y ) ⊗ Q / A(X) ⊗ Q

Td(TX ) · ch(f! (y)) = f∗ Td(TY ) · ch(y)
15.6
Proof: ch(f! (y)) = f∗new (ch(y)) = F (TX )f∗ F (TY )−1 ch(y) .

Definition: A graded cohomology theory A• L is a cohomology theory with values in the

category of graded commutative rings, A• (X) = n
n≥0 A (X), such that for any projective
morphism f : Y → X between connected varieties, the direct image f∗ : An (Y ) → An+d (X)
changes the degree in the codimension d = dim X − dim Y .
Remark that elements of negative degree are assumed to be null, and that the cohomology
class of any hypersurface has degree 1, so that ci (x) ∈ Ai (X) for any x ∈ K(X).
For example, GK • (X) ⊗Z Q and H 2• (Xan , Z) are graded cohomology theories.
15.4. K-THEORY 439
Lemma: Any graded cohomology theory A• follows the additive law c1 (Lx ⊗ Ly ) = x + y.
Proof: Since A• (Pm × Pn ) = A• (Spec k)[xm , xn ]/(xm+1 m ,x

n+1 ), there exist a, b ∈ A0 (Spec k) and
n
c ∈ A1 (Spec k) such that c1 π1∗ OPm (1)⊗OPm ×Pn π2∗ OPn (1) = ac1 π1∗ OPm (1) +bc1 π2∗ OPn (1) +c.

Restricting to Pm × pt, pt × Pn and pt × pt we see that a = b = 1, c = 0.

By Jouanolou’s trick there is an affine bundle π : P → X and a morphism f : P → Pm × Pn
such that π ∗ Lx = f ∗ π1∗ OPm (1) and π ∗ Ly = f ∗ π2∗ OPn (1).
Hence c1 (π ∗ Lx ⊗ π ∗ Ly ) = c1 (π ∗ Lx ) + c1 (π ∗ Ly ), so that π ∗ (c1 (Lx ⊗ Ly )) = π ∗ x + π ∗ y.
We conclude because π ∗ : A• (X) → A• (P ) is an isomorphism.
Theorem: If A• is a graded cohomology theory on the smooth quasi-projective varieties over a

perfect field, there is a natural homogeneous ring morphism GK • (X) → A• (X) ⊗ Q preserving
inverse and direct images (hence Chern classes and cohomology classes).
Proof: Let i : Y → X be a closed subvarietyL of codimension d.

p
If Y is smooth, ch(OY ) = [Y ] + . . . ∈ p≥d A (X) ⊗ Q by the Riemann-Roch theorem.
In general Y is smooth outside a closed set Ysing of codimension > d, when the base field is
perfect, and if we put j : U = X − Ysing → X, repeatedly applying condition 3 we see that we
have injective morphisms (recall that Ai (Z) = 0 when i < 0)
j ∗ : Ap (X) ⊗ Q −→ Ap (U ) ⊗ Q , p ≤ d.
Since j ∗ (ch(OY )) = ch(j ! OY ) = [Y ∩ U ] + ..., we see that ch : K(X) → A• (X) ⊗ Q preserves

filtrations; hence it induces a ring morphism ϕ : GK • (X) → A• (X) ⊗ Q, ϕ([x]d ) = [ch(x)]d when
[x]d ∈ GK d (X), and it preserves inverse images since so does ch.
By the Riemann-Roch theorem, it preserves direct images, since for any projective morphism
f : Y → X of codimension d and any element y ∈ F n (Y ) we have
ϕf∗ ([y]n ) = ϕ[f! (y)]n+d = [ch(f! (y))]n+d = [f∗ (Td(Tf )ch(y))]n+d

= f∗ [(1 + . . .)ch(y)]n = f∗ [ch(y)]n = f∗ ϕ([y]n ).
1. The Chern character and Todd class of a locally free sheaf E = Lα1 + . . . + Lαr are
= rk E + c1 + 21 (c21 − 2c2 ) + 16 (c31 − 3c1 c2 + 3c3 ) + . . .
P αi
ch(E) = ie
Td(E) = i (1 + 21 αi + 12
1 2 1
αi4 + . . .) = 1 + 21 c1 + 12
1 1
(c21 + c2 ) + 24
Q
αi − 720 c1 c2 + . . .
2. If C is a smooth projective curve, then K = c1 (ΩC ) is the class of any canonical divisor,
and the Riemann-Roch theorem for the projection C → Spec k onto a point gives, for any
locally free sheaf E of rank r,
χ(C, E) = deg [Td(L−K )ch(E)] = deg [(1 − 12 K)(r + c1 (E))] = deg (c1 − 2r K).
3. If S is a smooth projective surface, and we put K = c1 (Ω1S ) and χtop = deg c2 (S), then we
1
have (when E = OS , we obtain Noether’s identity χ(S, OS ) = 12 (K 2 + χtop )):
2
Td(TS ) = 1 − 12 K + 1
12 (K + c2 (S)),
rk E + c1 (E) + 21 c1 (E)2 − 2c2 (E) ,

ch(E) =
rk E 2 1 1 2
χ(S, E) = 12 (K + χtop ) − 2 K · c1 + 2 c1 − deg c2 .
4. The Riemann-Roch theorem for a smooth closed hypersurface i : Y → X gives the ad-
junction formula, i∗ KY = Y (KX + Y ). Just compare terms of degree 2 in
i∗ (Td(TY )) = ch(OY ) · Td(TX ) = ch(1 − L−Y ) · Td(TX )
i∗ (1) − 21 i∗ KY + . . . = (Y − 12 Y 2 + . . .)(1 − 12 KX + . . .)
5. Let ω be a rational 1-form on a smooth projective surface S and let us consider the line
sheaf (ω)(U ) = {f ω ∈ ΩS (U ) : f ∈ ΣS } = LD (U ), where D is the divisor of zeros and
poles of ω. We have an exact sequence
∧ω
(∗) 0 −→ (ω) −→ Ω1S −−−→ Ω2S ⊗ L−D −→ C −→ 0
If x, y are parameters at a point p, and ω = h(f dx+gdy), where f, g ∈ Op have no common

factor, a local equation of D is h = 0, and Cp ' OpP /(f, g). Now, ch(OS /mp ) = p by the
Riemann-Roch theorem for p ,→ S, so that ch(C) = p∈S l(Cp ) · p. If we put K = c1 (Ω2S ),
the exact sequence (∗) gives the Zeuthen-Segre invariant:
ch(LD ) + ch(LK−D ) = ch(Ω1S ) + ch(C),

eD + eK−D = 2 + K + 21 K 2 − c2 (S) + ch(C),
P
c2 (S) = D(K − D) + p l(Cp ) p.
6. Let X be a smooth projective variety. If i : Y → X is a closed smooth subvariety of

codimension d, then ch(OY ) = i∗ (Td(NY /X )) = Y +. . . in A• (X)⊗Q. Since (see problem 25
in page 475) [ch(OY )]r = P (c1 (OY ), . . . , cr−1 (OY )) − (−1)r rcr (OY ), we see that c1 (OY ) =
0, . . . , cd−1 (OY ) = 0, cd (OY ) = (−1)d (d − 1)! Y in A• (X) ⊗ Q.
7. The universal property of GK • ⊗ Q shows that the topological and algebraic genus of a
smooth projective curve C coincide, 21 dimQ H 1 (Can , Q) = dimC H 0 (C, ΩC ), since for any
complex smooth projective variety X we have h∆, ∆itop = h∆, ∆ialg ,
i i p+q dim H p (X, Ωq ).
P P
i (−1) dimQ H (Xan , Q) = p,q (−1) C X
15.5 Duality Theory

Now all schemes will be noetherian and separated, and f : X → S is a projective morphism.
If M is a coherent OX -module, then the sheaves Rp f∗ M are coherent (p. 323) and we fix a
finite affine cover of X, affine over S (with images contained in affine open sets of S).
On the category of quasi-coherent OX -modules, the functor M f∗ Č p M is exact and it
commutes with inductive limits since global sections commute with inductive limits (p. 323).
Now the representability theorem3 directly proves (p. 354) the
Duality Theorem: For any bounded below complex I of injective quasi-coherent OS -modules
there exists a bounded below complex f ! I of injective quasi-coherent OX -modules such that, for
any quasi-coherent OX -module M we have a natural OS (S)-linear isomorphism
Hom•X (M, f ! I) = Hom•S f∗ (Č • M), I .

Definition: When I is an injective resolution of OS , we say that DX/S = f ! I is the dualizing

complex of f and, up to quasi-isomorphisms, it does not depend on the affine cover of X (any
two affine covers admit a common affine refinement) or the injective resolution of OS (the cone
of a quasi-isomorphism I ' J is a split exact sequence).
When S = Spec k is the spectrum of a field, we may take I = k, and the dualizing complex
DX = f ! k is a bounded complex of injective quasi-coherent sheaves.
3
In the case of a contravariant functor F on the category of quasi-coherent sheaves, minimal pairs Qξ form
a set because, according to Deligne’s formula, anyone is fully determined by the set of all elements η ∈ F (pn )
admitting a morphism of pairs (pn )η → Qξ .
15.5. DUALITY THEORY 441
∼
Theorem: If U is an open set in S, then DX/S |f −1 U −
→ Df −1 U/U .
For any coherent OX -module M we have a quasi-isomorphism
f∗ Hom•X (M, DX/S ) ' Hom•S (f∗ Č • M, I).
Proof: Let U be an affine open set in S, and p the sheaf of ideals of S − U .

Since the fixed open cover of X is affine over S, we have f∗ Č • (f ∗ pn ⊗ M) = pn ⊗ f∗ Č • M,
and by Deligne’s formula (the quasi-isomorphism is due to the following lemma)
Γ(U, f∗ Hom•X (M, DX/S )) = lim

−→
Hom•X (pn , f∗ Hom•X (M, DX/S ))
= lim
−→
Hom•X (f ∗ pn ⊗ M, DX/S ) = lim
−→
Hom•S (f∗ Č • (f ∗ pn ⊗ M), I)
= lim
−→
Hom•S (pn ⊗ f∗ Č • M, I) −−
∼
→ Γ(U, Hom•S (f∗ Č • M, I)).
Hence f∗ Hom•X (M, DX/S ) ' Hom•S (f∗ Č • M, I). Moreover, if we put V = f −1 U , then
Hom•V (M|V , DX/S |V ) −−

∼
→ Hom•S ((f |V )∗ Č • (M|V ), I|U ) = Hom•V (M|V , DV /U )
since I|U is an injective resolution of OU . Now, if q is the sheaf of ideals of a closed subscheme
∼
in V , we put M|V = qn , and taking inductive limit, we see that DX/S |V − → DV /U .
Lemma: If K is a bounded quasi-coherent complex

of coherent cohomology sheaves Hp (K), then
• n • •
the natural morphism lim
−→
Hom p , Hom (K, I) → Hom (K|U , I|U ) is a quasi-isomorphism.
Proof: First we see the existence of a quasi-isomorphic coherent subcomplex K 0 ,→ K.

If K i is coherent for all i > p, we take a coherent submodule M ⊂ K p such that dp (M) =
dp (K p ), and a coherent submodule N ⊂ Ker dp such that N → Hp (K) → 0.
Replacing K p and K p−1 by M + N and d−1 p−1 (M + N ), the complex is coherent in degree p.
• •
Now we put R = Hom (K, I), R = Hom (K 0 , I), and we have a commutative square
0
lim Hom• (pn , R) / Γ(U, R)

−→
o qis

lim Hom •
(pn , R0 )
iso / Γ(U, R0 )
−→
where the quasi-isomorphism is due to the fact that R → R0 is a quasi-isomorphism of flasque

complexes (hence the cone is a flasque acyclic complex).
To conclude, we must show that lim
−→
Hom• (pn , R) → lim
−→
Hom• (pn , R0 ) is a quasi-isomorphism.
If S is affine, R(S) ' R0 (S) induces a quasi-isomorphism (U is affine)
Hom• (pn , R) = lim Hom• pn , R(S) 0 (S) = lim Hom• (pn , R0 ).

∼
lim ] = Γ U, R(S) ] − ^
→ Γ U, R
−→ −→ −→
In general it is a quasi-isomorphism since so it is on sections over any affine open set in S.

Since both are flasque complexes, we conclude taking global sections.
Theorem: If φ : T → S is flat, we have a quasi-isomorphism φ̄∗ DX/S ' DXT /T .
Proof: If φ∗ I → J is an injective quasi-coherent resolution, then J is a resolution of OT because

φ is flat, and for any coherent OX -module M
f¯∗ Hom•XT (φ̄∗ M, φ̄DX/S ) = f¯∗ φ̄∗ Hom•X (M, DX/S ) since φ is flat
= φ∗ f∗ Hom•X (M, DX/S ) (p. 421)
' φ∗ Hom•S (f∗ Č • M, I) by the previous theorem
' Hom•S (φ∗ f∗ Č • M, J) by the following lemma
= Hom•S (f¯∗ φ̄∗ Č • M, J) (p. 421)
= Hom•S f¯∗ Č • (φ̄∗ M), J)
' f¯∗ Hom•XT (φ̄∗ M, DXT /T ) by the previous theorem
Lemma: If K is a bounded quasi-coherent complex of coherent cohomology sheaves H p (K), and

φ is flat, then φ∗ Hom• (K, I) −
∼
→ Hom• (φ∗ K, J); i.e.,
φ∗ RHom• (K, I) ' RHom• (φ∗ K, φ∗ I).
Proof: Let K 0 ⊂ K be a quasi-isomorphic coherent subcomplex.

Since I is injective, Hom• (K, I) ' Hom• (K 0 , I), and since it is a local question, we may
assume that we have a resolution L → K 0 by finite free modules.
Now, using that φ is flat and the bicomplex spectral sequence, we have quasi-isomorphisms
φ∗ Hom• (K, I) ' φ∗ Hom• (K 0 , I) ' φ∗ Hom• (L, I) = Hom• (φ∗ L, φ∗ I)

' Hom• (φ∗ L, J) ' Hom• (φ∗ K 0 , J) ' Hom• (φ∗ K, J).
15.5.1 Calculation of the Dualizing Complex

Theorem: When i : Y → X is a closed embedding, DY /S ' Hom•X (OY , DX/S ).
Proof: We consider in Y the restriction of the fixed affine cover of X.
Then we have i∗ Č • (i∗ M) = Č • (i∗ i∗ M), and we conclude,
Hom•X (M, i∗ DY /S ) = Hom•Y (i∗ M, DY /S ) = Hom•S (f∗ Č • (i∗ i∗ M), I)

= Hom•X (i∗ i∗ M, DX/S ) = Hom•X (M, HomX (i∗ OY , DX/S )).
Lemma: If B = A/I, where the ideal I is generated by a regular sequence, I = (f1 , . . . , fd ),

then ExtpA (B, A) = 0, p 6= d, and we have a canonical isomorphism
ExtdA (B, A) = HomB (Λd (I/I 2 ), B).

p p
If moreover TorA
p (B, M ) = 0, p > 0, then ExtA (B, M ) = ExtA (B, A) ⊗A M .
Proof: The Koszul complex K of f1 , . .. , fd is a freeresolution of B, and the dual complex only
has a non null cohomology group, H d HomA (K, A) ' B.
P
If a regular sequence gi = j aij fj generates I, we have an isomorphism K(f1 , . . . , fd ) '
K(g1 , . . . , gd ), given in degree p by Λp (aij ), and a commutative diagram
ExtdA (B, A)
f1 ,...,fd g1 ,...,gd
{ det(aij ) #
B /B
The same holds with the dual bases of f1 ∧ . . . ∧ fd and g1 ∧ . . . ∧ gd in Λd (I/I 2 )∗ , so that
the isomorphism ExtdA (A/I, A) = Λd (I/I 2 )∗ is intrinsic.
Moreover Hom•A (K, A) ⊗A M = Hom•A (K, M ) is a free resolution of B; hence

H p Hom•A (K, A) ⊗A M = H p Hom•A (K, M ) .

Definition: A closed embedding Y → X is a regular embedding of codimension d if the sheaf

of ideals pY /X is locally generated by regular sequences of length d.
Theorem: Let p be the ideal of a regular embedding i : Y → X of codimension d. The cohomology

sheaves of DY /X are null, except ωY /X = Hd (DY /X ) = Λd (p/p2 )∗ . If moreover the sheaves
Hp (DX/S ) are i∗ -acyclic, then
Hp+d (DY /S ) = ωY /X ⊗OY i∗ Hp (DX/S ).

Proof: The first statement follows directly from the lemma because
Hp (DY /X ) = Hp RHomX (OY , OX ) = ExtpX (OY , OX ).

Moreover, DY /S = HomX (OY , DX/S ), and we have a spectral sequence (p. 366)
E2p,q = ExtpX (OY , Hq (DX/S )) ⇒ Hp+q (DY /S ).
If the sheaves Hq (DX/S ) are i∗ -acyclic, by the lemma E2p,q = 0, p 6= d, and
Hd+q (DY /S ) = ExtdX (OY , Hq (DX/S )) = ExtpX (OY , OX ) ⊗OX Hq (DX/S ).

Definition: A flat morphism of finite type X → S is a smooth morphism of dimension d if the
diagonal ∆ : X → X ×S X is a regular embedding of codimension d.
Theorem: If X → S is a projective smooth morphism of dimension d, then the dualizing complex

DX/S has a unique non null cohomology sheaf, H−d (DX/S ) = Λd ΩX/S .
Proof: Since the morphism X → S is flat,

Hp (DX×S X/X ) = Hp (π1∗ DX/S ) = π1∗ Hp (DX/S ) = Hp (DX/S ) ⊗OX OX×X ,
and these sheaves are ∆∗ -acyclic because, if A is the local ring of X ×S X at a point of the
diagonal, with the notations of the lemma, we have
TorA p p B
p (B, M ⊗B A) = H [(M ⊗B A) ⊗A K] = H (M ⊗B K) = Torp (M, B),
K being a complex of flat B-modules. We conclude because the sheaves Hp (DX/X ) are null,
except H0 (DX/X ) = OX , and according to the above theorem,
Hp+d (DX/X ) = ω∆ ⊗ ∆∗ Hp (DX×S X/X ) = (Λd ΩX/S )∗ ⊗ Hp (DX/S ).

Theorem: If an open set U → X admits a closed embedding in AnS ,
Hp (DX/S )|U ' Extp+n
An (OU , OAS ).
n
S
Proof: The ideal of the closed embedding i : U → AnS ×S X = AnX is defined in the open set AnU
by a regular sequence of length n, so that ωU/AnX ' OU is trivial.
As before (now A is the local ring of AnX at a point of U ), the sheaves
Hp (DAnX /AnS ) = Hp (π2∗ DX/S ) = π2∗ Hp (DX/S ) = Hp (DX/S ) ⊗OX OAnX
are i∗ -acyclic, and the lemma let us conclude,
i∗ π2∗ Hp (DX/S ) ' ωU/AnX ⊗ i∗ Hp (DAnX /AnS ) = Hp+n (DU/AnS ) = Extp+n

An (OU , OAS ).
n q.e.d.
S
1. If X is a projective scheme over a field, then we have HomX (M, DX ) = Γ(X, Č • M)∗ . Hence
Ext−p p ∗
X (M, DX ) = H (X, M) , and if the dualizing complex DX has a unique non null coho-
d−p
mology sheaf ωX = H−d (DX ), then ExtX (M, ωX ) = H p (X, M)∗ .
2. In the case of a finite k-algebra A, we have HomA (M, ωA ) = M ∗ for any finitely generated
A-module M . When M = A, we see that the dualizing module is ωA = A∗ .
3. The dualizing sheaf of Pd is a line sheaf, ωPd ' OPd (r), and ωPd ' OPd (−d − 1) since
H d (Pd , O(n))∗ = Hom(O(n), ωPd ) = Γ(Pd , O(r − n)).
4. If Y is a hypersurface of a regular projective variety X, defined by a locally principal ideal

p = L−Y , then ωY = (p/p2 )∗ ⊗ ωX , and the canonical divisor KY of Y is given by the
adjunction formula, KY = Y · (KX + Y ).
5. If X is a smooth projective variety of dimension n, then H p (X, ΩqX )∗ = H n−p (X, Ωn−q
X ). In
q ∗ n n−q
fact, (ΩX ) ⊗ ΩX = ΩX , and for any locally free OX -module L we have
H p (X, L)∗ = Extn−p (L, ΩnX ) = H n−p (X, L∗ ⊗ ΩnX ).
6. If K is the canonical divisor of an irreducible smooth projective surface S, by duality we have

h2 (D) = h0 (K − D) for any divisor D, and by the Riemann-Roch theorem (p. 439)
h0 (nD) + h0 (K − nD) ≥ χ(LnD ) = 21 (D · D)n2 − 12 (K · D)n + χ(OS ).
When D2 > 0, if n 0 we see that h0 (nD) 0 or h0 (K − nD) 0, in which case

h0 (K + nD) = 0 (so that h0 (−nD) 0) because any non-null section of LK+nD gives
h0 (K − nD) ≤ h0 (K − nD + K + nD) = h0 (L2K ).
Hence h0 (nD) > 1 for some n ∈ Z and, S being irreducible, nD is linearly equivalent to
an effective divisor, and we conclude that D · H > 0 for any hyperplane section H. That
is to say, the symmetric bilinear form Pic(S) × Pic(S) → Z, (D0 , D) 7→ D0 · D, has index 1
(Hodge’s index theorem): H · H > 0, and if D · H = 0, then D2 ≤ 0 (and D2 = 0 only if
D is numerically equivalent to 0, in the sense that D · E = 0 for any divisor E; that is to
say, D is in the radical of the intersection metric).
7. When the surface is a direct product, S = C 0 × C, of smooth projective curves with some
rational point, the divisors p0 × C and C 0 × p clearly form an hyperbolic pair, orthogonal to
any divisor D̄ = D − d(p0 × C) − d0 (C 0 × p), where d0 , d are the degrees of the divisor D over
both factors. Hence D̄2 ≤ 0, and we obtain Castelnuovo’s inequality, D · D ≤ 2d0 d, and
the equality only holds if D is numerically equivalent to d(p0 × C) + d0 (C 0 × p).
8. In the particular case of the graph Γ of a morphism C → C of degree d, the adjunction

formula Γ · (KS + Γ) = 2g − 2, where g is the genus of C, states that Γ2 = (2 − 2g)d. Now,
Castelnuovo’s inequality (nΓ + m∆)2 ≤ 2(n + m)(dn + m) gives
q(n, m) = gdn2 + (1 + d − Γ · ∆)nm + gm2 ≥ 0
because ∆2 = 2 − 2g. The discriminant of q(n, m) is ≤ 0, and we obtain an√estimate of the

number Γ · ∆ of fixed points, counted with multiplicity, |1 + d − Γ · ∆| ≤ 2g d.
9. If moreover k = Fq is a finite field, we have a k-morphism C → C, the identity on the

underlying topological space and OC (U ) → OC (U ), f 7→ f q , on any open set U . By base
change to an algebraic closure k → k̄ we obtain a Frobënius morphism F : C̄ → C̄, the
fixed points being the rational points of C. This Frobënius morphism has degree q and it is
purely inseparable (so that the graph transversally intersects the diagonal) and we obtain an
√
estimate of the number N of rational points, |1 + q n − N | ≤ 2g q, equivalent to the Riemann
hypothesis for the zeta function of C (see [27]).
15.5.2 Local Duality

Let Ob be a complete noetherian local ring and let x be the closed point of X = Spec O.b We put
U = X −x, and for any O-module
b ∗
M we put M = HomOb (M, I), where I is a fixed injective hull
of the residue field O/m.
b Now we try to determine the dual Hxp (X, M )∗ of the local cohomology
groups in terms of morphisms to a local dualizing complex D. b
We have the local cohomology exact sequence
0 −→ Hx0 (X, M ) −→ M −→ M
f(U ) −→ Hx1 (X, M ) −→ 0
and Hxp (X, M ) = H p−1 (U, M

f), p ≥ 2. Hence, if we consider a finite open cover of U by basic
open sets, the corresponding Cech complex Č • (M
f|U ) is an acyclic resolution of M
f|U , and the
•
cohomology groups of the following complex K (M ):
f|U ) → . . . → K p (M ) = Γ(U, Č p−1 M

K 0 (M ) = M → K 1 (M ) = Γ(U, Č 0 M f|U ) → . . .
are just the local cohomology groups Hxp (X, M ). Moreover, K p (M ) is a linear exact functor
and commutes with inductive limits (p. 323); hence, by the representability theorem, K p (M )∗
is (linearly) representable by an injective O-module
b b −p .
D
Local Duality Theorem: There is a bounded complex of injective O-modules

b D,
b named local
dualizing complex of O, such that for any O-module M , we have natural isomorphisms
b b
(K • M )∗ = Hom•Ob (M, D).

b
Corollary: Hxp (X, M )∗ = Ext−p

b (M, D).
b
O
Theorem: The local dualizing complex of a complete regular local ring of dimension n has a
unique non zero cohomology module, H −n (D)
b ' O.
b
b = Ext−p (O,
Proof: We have H −p D b = Hxp (O)
b D) b ∗ = 0 when p 6= n.
Ob
Moreover, Hxn (O) b ∗ ' (O)
b is an injective hull of O/m, so that H n (O) b ∗∗ = O
b (p. 417).
x
b0 is a quotient of O,
Lemma: If O b0 is Hom b (O
b an injective hull of the residue field of O b0 , I).
O
Proof: On the category of Ob0 -modules of finite length, the functor F (M ) = Hom b (M, I) is exact
O
and F (k) ' k; hence it corresponds to an injective hull of the residue field,
0
lim b0 /m r , I) = Hom b (O
HomOb (O b0 , I).
−→ O
b0 is a quotient of O,
Theorem: If O b 0 of O
b the local dualizing complex D b0 is
b 0 ' RHom b (O
D b0 , D).
b
O
Proof: The basic open cover of U = Spec O b0 − x, and

b − x induces a basic open cover of Spec O
HomOb (O 0
b , I) is the injective hull of the residue field of O 0 b -module M 0 ,
b . Now, for any O 0
(K • M 0 )∗ = Hom•Ob0 K • M 0 , HomOb (O b0 , I) = Hom• (K • M 0 , I)

O
b
• 0 b • • b0 b

= HomOb (M , D) = HomOb M, HomOb (O , D) .
Corollary: Let DX be the dualizing complex of a projective variety X over a field. At any point
x ∈ X, the completion of the stalk DX,x is quasi-isomorphic to the local dualizing complex Dbx
of the complete local ring O
bX,x .
Proof: If X is a closed subscheme of a projective space Pd , then DX = RHomOP (OX , ΩdPd ) and
d
DX,x = RHomOPd ,x (OX,x , ΩdPd ,x ) ' RHomOPd ,x (OX,x , OPd ,x ),
since Dx is an injective resolution of ΩdPd ,x ' OPd ,x when D is an injective resolution of ΩdPd .
Hence the completion of DX,x is just RHom b (O bX,x , O
OPd ,x
bP ,x ) = D
d
b x.
Corollary: If Ob0 is a quotient of a complete regular local ring O, b0

b the local dualizing complex D
is a bounded injective complex with finitely generated cohomology modules.
b0 -modules Extp (O
Proof: The O b = Extp (O
b0 , D) b0 , O)
b are finitely generated.
b O b O
15.5.3 Biduality
Definition: Let O be a local noetherian ring. A bounded complex D of injective O-modules
(or any quasi-isomorphic bounded complex of O-modules) with finitely generated cohomology
modules H p (D) is a biduality complex if the natural morphism M • → DD(M • ) is a quasi-
isomorphism for any bounded complex of O-modules M • with finitely generated cohomology
modules H p (M • ), where we put D(M • ) = RHom•O (M • , D).
Theorem: A bounded complex D of injective O-modules, with finitely generated cohomology

modules, is a biduality complex if and only if the natural morphism O → DD(O) is a quasi-
isomorphism, O ' Hom•O (D, D).
Proof: If D is a biduality complex, just put M = O in the definition.

Conversely, if M is a finitely generated O-module and L• → M → 0 is a resolution by
free modules of finite rank, by hypothesis the morphism Lp → Hom• (Hom• (Lp , D), D) is a
quasi-isomorphism.
Hence, by the bicomplex theorem, L• − ∼
→ Hom• (Hom• (L• , D), D), and M ' DD(M ).
Finally, we proceed by induction on the number of non null componentes of the bounded
complex M • and, if M p is the non null component of highest degree, we have a commutative
diagram with exact rows
0 / Mp / M• / M̄ • /0

0 / DD(M p ) / DD(M • ) / DD(M̄ • ) /0
Since M p → DD(M p ), M̄ • → DD(M̄ • ) are quasi-isomorphisms, so is M • → DD(M • ).

Corollary: If O is regular local ring, a biduality complex is O.
Proof: O has a finite injective resolution since ExtnO (M, O) = 0, n > dim O, and the quasi-
isomorphism O ' RHom• (RHom• (O, O), O) is obvious.
Corollary: Any biduality complex of a field k is quasi-isomorphic to k, up to non canonical

isomorphisms and re-grading.
Proof: A complex D of vector spaces is quasi-isomorphic to H(D) = {H p (D), d = 0}, and
k ' Hom•k (D, D) ' Hom•k (H(D), H(D)).
Theorem: If D is a biduality complex of O, then RHomO (Ō, D) is a biduality complex of any

quotient ring Ō.
Proof: If Q is an injective O-module, then HomO (Ō, Q) is an injective Ō-module.

We conclude since Hom•Ō (M̄ , Hom•O (Ō, D)) = Hom•O (M̄ , D) for any Ō-module M̄ .
Corollary: Any quotient of a regular local ring posses a biduality complex.
Corollary: Let DX be the dualizing complex of a projective variety X over a field. The stalk
DX,x at any point is a biduality complex of OX,x -modules, and we have a quasi-isomorphism
OX ' RHom•OX (DX , DX ).
Proof: We have seen (p. 446) that DX,x ' RHomOPd ,x (OX,x , OPd ,x ).
Hence it is a biduality complex of OX,x -modules.
Finally, DX is quasi-isomorphic to a coherent subcomplex (p. 441), so that
RHom•OX (DX , DX ) x = RHom•OX,x (DX,x , DX,x ) = OX,x .

Corollary: The local dualizing complex D b of a complete local ring O,

b quotient of a complete
regular local ring, is a biduality complex of O-modules.
b
Note: We have shown (p. 446) that the completion of D = DX,x is quasi-isomorphic to the
b of the completion O
local dualizing complex D b of the local ring O = OX,x ; but now, using that
•
OX = HomOX (DX , DX ), we may exhibit a canonical quasi-isomorphism,
b = RHom b (O,
D b ∗ = lim(RHomO (O/mn , O))∗
b = (RΓx O)
b D)
O ←−
= lim
←−
(RHomO (O/mn , RHom•O (D, D)))∗ = lim
←−
(RHom•O (O/mn ⊗ D, D))∗
= lim
←−
(D/mn D)∗∗ = lim
←−
D/mn D.
Theorem: A bounded complex D of injective O-modules, with finitely generated cohomology

modules, is a biduality complex if and only if the natural morphism k → DD(k) is a quasi-
isomorphism, where k = O/m is the residue field.
Proof: If D is a biduality complex, just put M = k in the definition.

Conversely, by induction on the length, we see that M → DD(M ) is a quasi-isomorphism
for any finite length O-module M ; just consider an exact sequence 0 → M 0 → M → M 00 → 0,
where l(M 0 ), l(M 00 ) < l(M ).
Now, by induction on the dimension of supp M , we shall see that M → DD(M ) is a quasi-
isomorphism for any finitely generated O-module M . If M 0 ⊂ M is the submodule of elements
with support (m)0 , replacing M by M/M 0 we may assume that m is not an associated prime of
M , so that there is a M -regular element t ∈ m, and we have an exact sequence
t
0 −→ M −→ M −→ M/tM −→ 0
t
When p 6= 0, the exact sequence H p (DD(M )) −
→ H p (DD(M )) → H p (DD(M/tM )) = 0
p
and Nakayama’s lemma show that H (DD(M )) = 0. Now the commutative diagram
0 /M t /M / M/tM /0
α α ᾱ

0 / H 0 (DD(M )) t / H 0 (DD(M )) / H 0 (DD(M/tM )) /0
t t
has exact rows and ᾱ is an isomorphism; hence Ker α −
→ Ker α and Coker α − → Coker α are
surjective by the Snake’s lemma. Nakayama’s lemma shows that α is an isomorphism.
b of a complete local ring O

Corollary: The local dualizing complex D b is a biduality complex.
b ' (RΓx (k))∗ ' k ∗ = k; hence DD(k) ' D(k) ' k.

Proof: We have D(k) = RHomOb (k, D)
Theorem: Put k = O/m. A bounded complex D of O-modules, with finitely generated cohomol-
ogy modules, is a biduality complex if and only if there is an integer n such that
(
p k p=n
ExtO (k, D) =
0 p 6= n
Proof: If D is a biduality complex, then RHomO (k, D) is a biduality complex of k; hence there
is some integer number n such that ExtnO (k, D) = k, and ExtpO (k, D) = 0 when p 6= n.
For the converse, it is clear that k → DD(k) is a quasi-isomorphism, so we only have to
show that D is quasi-isomorphic to a bounded complex of injective O-modules.
First we prove that ExtpO (M, D) = 0, p ∈
/ [n − d, n], by induction on d = dim (supp M ).
We may assume that m is not an associated prime of M , so that we have exact sequences
t
0 −→ M −→M −→ M/tM −→ 0
t
ExtpO (M, D) −→ ExtpO (M, D) −→ Extp−1
O (M/tM, D) = 0
whenever p ∈ / [n − d, n]. Hence ExtpO (M, D) = 0 by Nakayama’s lemma.

Now, if D ' I • is an injective resolution, we conclude if we prove that the cycles Z r = B r
are an injective module when r 0. Let us consider K • : . . . → I r−1 → I r → 0 → . . . and
K̄ • : . . . → I r−1 → B r → 0 → . . ., so that we have exact sequences
0 −→ K̄ • −→K • −→ B r+1 [−r] −→ 0

• •
Extr+1 r+1
O (M, K ) −→ ExtO (M, B
r+1
[−r]) −→ Extr+2
O (M, K̄ )
where M is any O-module. When r 0, we have Extr+2 • r+2

O (M, K̄ ) = ExtO (M, D) = 0 and
r+1 • •
ExtO (M, K ) = 0 because K is an injective complex without terms of degree > r.
Hence 0 = Extr+1
O (M, B
r+1 [−r]) = Ext1 (M, B r+1 ) and B r+1 is an injective module.
O
15.6. FORMAL FUNCTIONS THEOREM 449
Corollary: A bounded complex D of O-modules is a biduality complex if and only if D ⊗O O

b is
a biduality complex of O-modules.
b
b = Extp (k, D) ⊗O O
Proof: Extpb (k, D ⊗O O) O
b = Extp (k, D).
O
O
Theorem: If it exists, the biduality complex D of a local noetherian ring O is unique up to

(non-canonical) isomorphisms and regrading.
Proof: Let D0 be another biduality complex of O-modules, and put D0 (M • ) = Hom•O (M • , D0 ).

Let us consider the natural morphism (where the tensor product is derived: factors must be
replaced by bounded above flat resolutions)
M• ⊗
=
D0 (D) = M • ⊗
=
Hom• (D, D0 ) −→ Hom• (Hom• (M • , D)D0 ) = D0 D(M • ).
Since it is a quasi-isomorphism when M • is a bounded above complex of free modules of

finite rank, so it is when M • is a bounded complex with finitely generated cohomology modules,
because M • is quasi-isomorphic to a finitely generated subcomplex (p. 441). Now,
D(D0 ) ⊗
=
D0 (D) ' D0 DD(D0 ) = D0 (D0 ) ' O,
and the following lemma shows that Hom• (D, D0 ) = D0 (D) ' O[n].
We conclude that D0 ' D0 D(D) ' D ⊗=
D0 (D) ' D ⊗=
O[n] ' D[n].
Lemma: Let K, L be two bounded above complexes with finitely generated cohomology modules.
If K ⊗
=
L ' O, then K ' O[n] for some integer n.
Proof: We may assume that p = 0 and q = 0 are the largest integers such that H p (K) 6= 0 and
H q (L) 6= 0, so that H 0 (K ⊗
=
L) = H 0 (K) ⊗ H 0 (L) 6= 0.
Therefore H 0 (K) ⊗ H 0 (L) = O and we see that H 0 (K) = H 0 (L) = O.
It follows that K ' K1 ⊕ O and L ' L1 ⊕ O.
Since O ' K ⊗ =
L ' O ⊕ K1 ⊕ L1 ⊕ (K1 ⊗=
L1 ), we have K1 ' 0; hence K ' O.
15.6 Formal Functions Theorem

In this section we shall omit many details and we don’t distinguish between derived Rn F and
hyperderived Rn F functors. Let f : X → S = Spec O be a projective morphism, where O is a
local noetherian ring, and we assume the existence of a biduality complex of O-modules DS (it
exists when O is a quotient of a regular local ring).
Now DX = f ! DS is a biduality complex at any closed point x ∈ X, because
RHomOX,x (κ(x), DX,x ) = RHomO (κ(x), DS ) = (RΓs κ(x))∗ ' κ(x),
where s is the closed point of S. Hence, for any bounded complex of OX -modules M• , of
coherent cohomology sheaves Hp (M• ), we have a natural quasi-isomorphism
M• −→ Hom•X (Hom•X (M• , DX ), DX ).
Theorem: If Y is the fibre of the closed point s of S, we have canonical isomorphisms (the dual
* is considered with respect to the injective hull of the residue field of s)
HYp (X, M• )∗ = Ext−p •

X (M , DX ) .
b
Proof: HYp (X, M• )∗ = Hsp (S, Rf∗ M• )∗ = Ext−p •

O (Rf∗ M , DS )
b by local duality,
= Ext−p •
X (M , DX )
b by duality.
Corollary: (Rp f∗ M• ) b = HY−p (X, Hom•X (M• , DX ))∗ .
Proof: (Rp f∗ M• ) b = Rp f∗ Hom•X (Hom•X (M• , DX ), DX ) = HY−p (X, Hom•X (M• , DX ))∗ .
b
Corollary: If π : Z → X is a projective morphism and E is the fibre of a closed point x ∈ X,
(R−p π∗ DZ )xb = HEp (Z, OZ )∗ ,

(Rp π∗ OZ )xb = HE−p (Z, DZ )∗ .
Proof: Let Zx = Z ×X Spec OX,x → Spec OX,x be the morphism induced by π.

Since DX,x is a dualizing complex of the local ring OX,x ,
(Rp π∗ OZ )xb = HE−p (Zx , DZ )∗ = HE−p (Z, DZ )∗ ,

∗
(R−p π∗ DZ )xb = HEp Zx , Hom•Y (DZ , DZ ) = HEp (Z, OZ )∗ .
Formal Functions Theorem: If p is the ideal of the fibre Y of the closed point, then
(Rp f∗ M• ) b = lim
←−
H p (X, M• ⊗
=
(OX /pn )).
Proof: Take a quasi-isomorphism P → M• , where P is a bounded above complex of flat OX -

modules (for example, direct sums of sheaves OU ),
(Rp f∗ M• ) b = HY−p (X, Hom• (M• , DX ))∗ = lim

←−
Ext−p (OX /pn , Hom• (P, DX ))∗
= lim
←−
Ext−p (P ⊗ (OX /pn ), DX )∗ = lim
←−
H p (X, P ⊗ (OX /pn )).
Note: Let us fix p, n ≥ 1. If M is a coherent OX -module, then, by Artin-Rees, the morphisms

Tori (M, OX /pn+r ) −→ Tori (M, OX /pn ) are null when r 0. Now, since the sheaves M/pn M
are supported on Y , we have
H p X, M ⊗ (OX /pn ) = lim H p (X, M/pn M) = lim H p (Y, M/pn M),

lim
←− = ←− ←−
and we see that (Rp f∗ M)b = lim

←−
H p (Y, M/pn M).
1. Rp f∗ M = 0 when p is greater than the dimension of the fibre Y . In fact Rp f∗ M is a finitely

generated O-module and (Rp f∗ M) b = 0, because H p (Y, M/pn M) = 0 when p > dim Y .
2. If f∗ OX = O, then the fibre Y is connected, because otherwise the local ring O b would
decompose as a direct sum of two rings,
b = (f∗ OX ) b = lim H 0 (Y1 ⊕ Y2 , O/pn ) = lim H 0 (Y1 , O/pn ) ⊕ lim H 0 (Y2 , O/pn ) .

O ←− ←− ←−
3. Assume now that both X and S are integral and that the field ΣS is algebraically closed in
the field ΣX (for example if f is birrational). If O is normal, then the fibre Y is connected
because, f∗ OX being a finite O-algebra contained in ΣX , we have f∗ OX = O.
4. If we put X 0 = Spec f∗ OX , we see that f factors into the composition of a projective morphism
Z → X 0 with connected fibres and a finite morphism X 0 → S, because f∗ OX is a finitely
generated O-module.
15.7. GROTHENDIECK TOPOLOGIES 451
5. Now we assume that X = Spec O, where O is a Cohen-Macaulay local ring of dimension d

with a dualizing module ωX , so that DX ' ωX [−d].
If π : X̄ = Proj ⊕n I n /I n+1 → X = Spec O is the blow-up along an ideal I ⊂ O, we have
a cone C = Spec GI O with vertex (the closed subscheme defined by the irrelevant ideal)
V = Spec O/I, the center of explosion, and with directrix E = Proj GI O, the exceptional
fibre E = π −1 (V ). Let Y = π −1 (x) be the fibre of the closed point x ∈ V . We have natural
projections C → V and V − C → E, and the fibre Y0 of C over x is a cone with directrix Y .
Since V and C are affine, and C −V is the complement of the zero section of a line bundle over
E, we have a local cohomology exact sequence (where λ is in fact an homogeneous morphism)
λ
Hxi (C, OC ) −→ HYi 0 (C, OC ) −−→ HYi 0 (C − x, OC ) −→ Hxi+1 (C, OC )
|| ||
i n n+1 ) HYi 0 (C − V, OC )
L
n Hx (V, I /I
||
i (E, O (n))
L
n∈Z YH E
If C = Spec GI O is Cohen-Macaulay, then Hxi (C, OC ) = 0, i < dimx C = d, and
HYi (E, OE (−n)) = 0, n > 0, i < d − 1.
Moreover, if C is Cohen-Macaulay, so is E because C − V → E is a line bundle (without

the zero section) and, E being an hypersurface of X̄, we see that X̄ also is Cohen-Macaulay,
with a dualizing sheaf ωX̄ such that DX̄ ' ωX̄ [−d]. Now the exact sequence
0 −→ OX̄ (1) −→ OX̄ −→ OE −→ 0
induces local cohomology long exact sequences, where n > 0, i < d,
0 = HYi−1 (E, OE (−n)) −→ HYi (X̄, OX̄ (−n + 1)) −→ HYi (X̄, OX̄ (−n))
and HYi (X̄, OX̄ (−n))∗ = Rd−i π∗ ωX̄ (n) b = 0, n 0, i < d (p. 323). By descending

induction on n, we conclude that HYi (X̄, OX̄ (−n)) = 0, n ≥ 0, i < d. In particular
HYi (X̄, OX̄ ) = 0, i < d,

Ri π∗ ωX̄ = 0, i > 0.
6. If the ideal I is generated by a regular sequence, then GI O = (O/I)[x1 , . . . , xr ] is Cohen-

Macaulay; hence Ri π∗ ωX̄ = 0, i > 0.
7. Let X be an hypersurface (i.e., defined by a locally principal ideal) of an ambient regular

variety Z. If we blow-up X along a regular center C, and we denote by p and p0 the ideals
of C in X and Z respectively, then Spec Gp OX is an hypersurface of the regular variety
Spec Gp0 OZ ; hence it is Cohen-Macaulay and again Ri π∗ ωX̄ = 0, i > 0.
15.7 Grothendieck Topologies

Definition: A presheaf of sets on a category C is a contravariant functor P : C Ens, and
morphisms of presheaves are just natural transformations. Hence, any morphism f : Y → X
induces a map f ∗ : P(X) → P(Y ) and, given a section s ∈ P(X), we put s|Y = f ∗ (s) when no
confusion is possible.
Definition: A sieve of an object X of a category C is a family morphisms Y → X stable under

specialization; i.e. a subfunctor of the functor of points Hom(−, X).
So any family of morphisms {Ui → X} generates a sieve R:
R(T ) = {morphisms T → X factoring through some Ui → X}.
If P is a presheaf of sets, then a morphism of presheaves R → P is defined by a compatible

family of sections si ∈ P(Ui ), in the sense that fi∗ (si ) = fj∗ (sj ) when a morphism T → X admits
two factorizations (eventually i = j)
fi
T / Ui
fj

Uj /X
From now on, C will be a category with fibred products X ×S Y .

Then the compatibility condition means that p∗1 (si ) = p∗2 (sj ), where p1 , p2 are the natural
projections of Ui ×X Uj onto the factors; i.e. we have an exact sequence
Q Q
Hom(R, P) −→ P(Ui ) ⇒ P(Ui ×X Uj )
i i,j
Definitions: A pretopology on C is given by some sets of morphisms {Ui → X}i∈I for any
object X, named covers of X, so that
1. If {Ui → X}i∈I is a cover, then {Ui ×X Y → Y }i∈I is a cover for any morphism Y → X.
2. If {Ui → X}i∈I is a cover and for any index i we have a cover {Vij → Ui }j∈Ji , then
{Vij → Vi → X}i∈I,j∈Ji is a cover.
3. The identity X → X is a cover of any object X.
A presheaf of sets F is separated when the natural mapQF(X) −→ Hom(R, F) is injective

for any cover {Ui → X}, i.e. when so is the map F(X) → i F(Ui ), and it is a sheaf when
F(X) −→ Hom(R, F) is bijective, i.e. we have exact sequences
Q p∗1 ,p∗2 Q
(*) F(X) −→ F(Ui ) ⇒ F(Ui ×X Uj )
i i,j
Morphisms of sheaves are just morphisms of presheaves.
Theorem: Let F be a separated presheaf (resp. sheaf ) of sets. If a sieve C of X contains some
cover {Ui → X}, then the natural map F(X) −→ Hom(C, F) also is injective (resp. bijective).
Proof: Let R be the sieve generated by {Ui → X}, and consider the natural maps
f
F(X) / Hom(C, F)
h g
&
Hom(R, F)
If F is separated, then h is injective; hence so is f . Moreover g also is injective, since any

morphism Y → X in C admits a cover {Yi → Y } such that the morphisms Yi → Y → X are in
R, and F is separated.
If F is a sheaf, then h is moreover bijective; hence so is f .
Definitions: A sieve C of X is a covering sieve if it contains some cover of X. Two pretopolo-

gies on C are equivalent if both have the same covering sieves (any cover {Vα → X} is refined
by some cover {Ui → X} of the other pretopology, in the sense that any morphism Ui → X
factors through some morphism Vαi → X). A topology is a pretopology such that any set of
morphisms {Vα → X} refined by a cover also is a cover, so that any pretopology is equivalent to
a unique topology (defined by all the sets {Vα → X} generating a covering sieve; i.e. refined by
a cover). A situs or site (with finite fibred products) is a category with finite fibred products
endowed with a topology.
Given sites C, D, a covariant functor f −1 : C D preserving fibred products is continuous
when {f −1 Ui → f −1 X} is a covering in D for any covering {Ui → X} in C, so that, if F is a
sheaf on D, then the presheaf (f∗ F)(U ) = F(f −1 U ) is in fact a sheaf, and any sheaf morphism
F → G induces a natural sheaf morphism f∗ F → f∗ G.
1. Let X be a topological space. The open sets, with the inclusion maps, form
S a category Op(X),
and covers are defined to be surjective families {Ui → U }; i.e. U = i Ui . Sheaves are just
sheaves in the ordinary sense, because Ui ×U Uj = Ui ∩ Uj . Moreover, any continuous map
φ : X → Y between topological spaces induces a continuous functor φ−1 : Op(Y ) → Op(X)
preserving the final object, and φ∗ is the usual direct image of sheaves (p. 304).
If U ⊂ X is an open set, then the inclusion Op(U ) ,→ Op(X) is a continuous functor (not
preserving the final object) and the corresponding direct image is just F F|U .
2. Any type of morphisms stable under base change and compositions, and including isomor-
phisms, defines a pretopology, where covers are families {R → X} of a single morphism of
the considered type.
3. In the category of topological spaces, the surjective families {Ui → X} of open embeddings
define a pretopology (hence a topology). Sheaves are just presheaves defining a classical sheaf
on any topological space.
Analogously the Zariski topology on the category of schemes Sch is defined by the pre-
topology given by the (set-theoretically) surjective families {Ui → X} of open embeddings.
4. The coarsest topology on C is the trivial topology: the total sieve is the unique covering
sieve, so that any presheaf P is a sheaf. If a morphism R → X admits a section, it is a cover
of X; hence the sequence P(X) → P(R) ⇒ P(R ×X R) is exact.
The finest topology is the discrete topology: any sieve is a covering sieve. The empty sieve
covers any object X; hence F(X) = Hom(∅, F) is a one point set for any sheaf F.
5. The canonical topologyQon C is defined by Qthe families {Ui → U } such that we have exact
sequences Hom(V, X) → i Hom(Vi , X) ⇒ i,j Hom(Vij , X) for any object X and any base
change V → U ; where Vi = Ui ×U V , Vij = Vi ×V Vj . It is the finest topology on C such that
the representable functors Hom(−, X) are sheaves.
6. The canonical topology on the category of sets Ens is defined by the surjective families
{Ui → X}.
7. Let G be a group. The canonical topology on the category of G-sets is defined by the surjective
families {Ui → X}. Now G → G/H is a cover, and G ×G/H G = qH G, so that we have
F(G/H) = F(G)H for any sheaf F, where the action of g ∈ G on F(G) is induced by the
right multiplication by g −1 on G. Any sheaf is representable, F(U ) = HomG (U, F(G)).
8. Given a finite Galois extension k → L of group G, let TrivL/k the category of finite k-algebras
trivial over L. According to Galois theorem, the opposite category Trivop L/k is equivalent to
op
the category of finite G-sets. Hence, if we consider on TrivL/k the toplogy defined by the
surjective families {Spec Bi → Spec A}, we have that the category of sheaves is equivalent to
the category of G-sets, where any sheaf F corresponds the the G-set F(L). Moreover, any
sheaf is an inductive limit of representable sheaves.
9. Let S be a connected and locally connected space admitting a universal covering S̄ → S.

According to Galois theorem, the category RevS of coverings X → S is equivalent to the
category of G-sets, where G = Aut(S̄/S). Hence the canonical topology on RevS is defined
by the surjective families {Ui → X}, and any sheaf is representable. Any sheaf F is fully
determined by the G-set F(S̄).
10. Let C be a situs. Given a covariant functor f −1 : C0 C preserving finite fibred products,
the families {Ui0 → X 0 } such that {f −1 Ui0 → f −1 X 0 } is a cover in C define on C0 the initial
topology: the coarsest topology such that f −1 is continuous.
Now, given S ∈ Ob C, let C|S be the category of S-objects X → S and S-morphisms. The
natural functor C|S C, (X → S) X, preserves finite fibred products; hence it induces
a topology on C|S , so that any sheaf F on C induces a sheaf F|S on C|S . Any S-object T
induces a natural continuous functor C|T = (C|S )|T C|S , and (F|S )|T = FT .
Let R be a full subcategory of C (some objects of C, with the same morphisms and
composition of morphisms, HomR (X, Y ) = HomC (X, Y )) stable under fibred products of C.
The natural functor R ,→ C preserves finite fibred products; hence it induces a topology on
R, so that any sheaf F on C induces a sheaf (F|R )(X) = F(X) on R.
11. Given topologies

T Ti on C, the families {Uj → X} which are covers for any Ti define W a
topology i Ti on C: the finest topology coarser than any Ti . We also have the topology i Ti
generated by the topologies Ti , the coarsest topology finer than any Ti : given a pretopology
Pi of each Ti , the families {Uj1 ...jn → . . .W→ Uj1 → X}, where {Uj1 ...jr → Uj1 ...jr−1 } is a cover
of some Pi , give a pretopology defining i Ti .
W
A presheaf F is a sheaf (resp. separated) for i Ti if and only if it is a sheaf (resp. separated)
for any Ti . In fact,Q check that the families {Ui Q
it is easy to Q → X} such that we have an exact
sequence F(Y ) → i F(Yi ) ⇒ ij F(Yij ) (resp. F(Y ) → i F(Yi ) is injective) for any base
change Y → X define a pretopology; hence a topology by the above theorem.
12. If a situs C admits a set {Gα } of topological generators

Q (any object X admits a cover
{Gαi → X}), then the natural map Hom(F, G) → α Hom F(Gα ), G(Gα ) is injective for
any two sheaves F, G; hence Hom(F, G) is a set and the sheaves on C form a category C.
e
Moreover, a sheaf morphism F → F̄ is an isomorphism if and only if F(Gα ) → F̄(Gα ) is

covers {Gαi αjQ
bijective for any index α: Just considerQ β → Gαi ×X Gαj }, so that for any sheaf
G we have an exact sequence G(X) → i G(Gαi ) ⇒ ijβ G(Gαi αj β ).
Recollement Theorem: let F, G be sheaves on a situs C, and fix a cover {Ui → X}.
1. The “presheaf” Hom(F, G)(X) := Hom(F|X , G|X ) is a “sheaf”: given sheaf morphisms
φi : F|Ui → G|Ui such that φi = φj on Uij = Ui ×X Uj (i.e. on C|Uij ) ¡even when i = j!, there
is a unique sheaf morphism φ : F|X → G|X such that φi = φ|Ui . Hence, φ is an isomorphism
when so is φ|Ui for any index i.
2. Given sheaves Fi on C|Ui and isomorphisms φji : Fi |Uij → Fj |Uij , such that φki = φkj ◦ φji
on Ui ×X Uj ×X Uk , then there is a sheaf F on C|X with isomorphisms F|Ui ' Fi such that
φji is the composition Fi |Uij ' F|Uij ' Fj |Uij , and it is unique up to an isomorphism by (1).
Proof: Let T → X be a morphism in the sieve R of X generated by {Ui → X}.

We have a well-defined map φ : F(T ) → G(T ), independent of the factorization through a
morphism Ui → X, because φi = φj on Uij . Hence, if we consider R as a full subcategory of
C|X , with the induced topology, we have a sheaf morphism φ : (F|X )|R → (G|X )|R , and (1)
follows from the comparison lemma (p. 461); but let us give a direct proof:
For any morphism Y → X we have a cover {Yi → Y }, hence a commutative diagram
F(Y ) / F(Yi ) −−−→/ F(Yij )
φ φ

G(Y ) / G(Yi ) −−−→ / G(Yij )
with exact rows, inducing a map φ : F(Y ) → G(Y ). So we obtain a sheaf morphism φ : F|X →
G|X such that φi = φ|Ui , and it is clearly unique.
On the other hand, for any factorization t : T → Ui we have a set Fi (t) and a bijection
φt0 t : Fi (t) ' Fj (t0 ) for any other factorization t0 : T → Uj , so that φt00 t = φt00 t0 φt0 t for any
third factorization t00 : T → Uk (in particular φtt = Id, since it is bijective and φtt = φtt φtt );
hence we may identify such sets, and we obtain a well-defined set F(T ), endowed with bijections
F(T ) ' Fi (t) such that φt0 t is just the composition Fi (t) ' F(T ) ' Fj (t0 ). Moreover, we have
a well-defined map F(T ) → F(T 0 ) for any X-morphism T 0 → T , so that we have in fact a sheaf
on R, and (2) also follows from the comparison lemma; but let us give a direct proof:
For any morphism Y → X, we define F(Y ) to be the kernel of F(Yi ) ⇒ F(Yij ).
So we obtain a presheaf F on C|X , and it is a sheaf:
For any cover {Vα → Y } we have a commutative diagram with exact rows
/
Q −−−−−−→/ Q
F(Y ) i F(Yi ) ij F(Yij )
−−−−→/

/
Q Q Q
α F(Vα ) αi F(Vαi ) αij F(Vαij )
↓ ↓

/
Q Q
αβ F(Vαβ ) αβi F(Vαβi )
Q Q
where the middle column is exact and ij F(Yij ) → αij F(Vαij ) is injective (because F|Ui ' Fi
is a sheaf). A diagram chasing shows that the first column is exact. So we obtain a sheaf F on
C|X , with isomorphisms F|Ui ' Fi such that φji is the composition Fi |Uij ' F|Uij ' Fj |Uij .
15.7.1 The Faithfully Flat Topology

Lemma: Let M be an A-module and A → B a faithfully flat morphism. We have an exact
sequence
d1 ,d2
M → M ⊗A B ⇒ M ⊗A B ⊗A B ; d1 (m ⊗ b) = m ⊗ 1 ⊗ b, d2 (m ⊗ b) = m ⊗ b ⊗ 1.
Proof: It is clear when Spec B → Spec A admits a section (see p. 189, or p. 453).
In general, after the base change A → B, we obtain the sequence corresponding to the B-
module MB = M ⊗A B and the morphism B → B ⊗A B, b 7→ 1 ⊗ b, which admits the section
B ⊗A B → B, b1 ⊗ b2 7→ b1 b2 . Since A → B is faithfully flat, we conclude.
Definition: QLet A be a ring. The finite surjective families of flat morphisms {Spec Bi → Spec B}
(i.e. B → i Bi is faithfully flat) define a pretopology on the category Aff|A of affine A-
schemes; hence a topology: the fp topology (faithfully Q flat topology).
Q A presheaf F is a
sheaf if and only if it preserves finite direct products F( n Bn ) = n F(Bn ) and the sequence
F(B) → F(C) ⇒ F(C ⊗B C) is exact for any faithfully flat morphism B → C.
Theorem: If M is an A-module, then M̃ (B) = MB = M ⊗A B is a sheaf on (Aff|A )fp .
Proof: Clearly M̃ (B1 × B2 ) = M̃ (B1 ) × M̃ (B2 ) and, if B → C is a faithfully flat morphism of

A-algebras, then MB → MC ⇒ MC⊗B C is exact by the lemma.
Definition: On (Aff|A )fp we have a sheaf of rings O(B) = B, and M̃ is an O-module. An

O-module M is quasi-coherent when it is defined by some A-module, clearly M(A).
Moreover HomA (M, N ) = HomO (M̃ , Ñ ); hence, by the recollement of sheaf morphisms
Corollary: If M, N are A-modules, then B HomB (MB , NB ) is a sheaf.
Theorem: Any locally quasi-coherent O-module M (i.e. the sheaves M|Ui are quasi-coherent
in some cover {Ui = Spec Bi → S = Spec A}) is quasi-coherent.
Proof: Let us prove that the natural sheaf morphism M(S)∼ → M is an isomorphism on the
cover {Ui → S}; i.e. M(S) ⊗A Bi = M(Ui ), both sheaves being quasi-coherent on Ui .
For any flat morphism T = Spec B → S we have a commutative diagram with exact rows
(the two last vertical morphisms being isomorphisms because M is quasi-coherent on Ui )
/
Q −−−→/ Q
M(S) ⊗A B i M(Ui ) ⊗A C ij M(Uij ) ⊗A B
o o
−−−−−−−−−→/

/
Q Q
M(T ) i M(Ti ) ij M(Tij )
so that M(S) ⊗A B = M(T ). When B = Bi , we conclude that M(S) ⊗A Bi = M(Ui ).
Corollary: Let U = Spec B → Spec A be faithfully flat and let N be a B-module. If we have an
isomorphism φ : p∗1 N ' p∗2 N such that p∗13 φ = p∗23 φ ◦ p∗12 φ, then there is an A-module M such
that N = MB , (where pij : U ×S U ×S U → U ×S U is the projection on the factors i, j).
Proof: By the recollement of sheaves, we have an O-module M such that M|U ' N ∼ , and M
is quasi-coherent by the theorem.
Theorem: The functor of points HomA (−, X) of any A-scheme X is a sheaf on (Aff|A )fp .
Proof: When X is affine, the result follows from the exact sequence of rings B → C ⇒ C ⊗B C
for any faithfully flat morphism B → C.
In general, if two morphisms φ, ψ : Spec B → X coincide on a cover S Spec C → Spec B, then
they coincide topologically (covers are surjective). Hence Spec B = i Vi , where Vi is a basic
open set such that φ(Vi ) = ψ(Vi ) is contained in an affine open subset of X.
By the previous case, φ and ψ coincide on Vi , so that φ = ψ, and X • is separated.

Now, let φ : Spec C → X be a morphism such that p∗1 φ = p∗2 φ : Spec (C ⊗B C) → X.
Then, set theoretically, φ is constant on the fibres of Spec C → Spec B, because for any two
points y1 , y2 of a fibre there is a point z ∈ Spec (C ⊗B C) such that p1 (z) = x1 , p2 (z) = x2
(remark that κ(y1 ) ⊗κ(x) κ(y2 ) 6= 0 when y1 and y2 are in the fibre of x).
By the following S lemma, φ factors through a continuous map φ̄ : Spec B → X.
Hence Spec B = i Vi , where Vi is a basic open set such that φ̄(Vi ) is contained in an affine
open subset of X. By the affine case, we obtain scheme morphisms φ̄i : Vi → X, compatible on
the intersections Vi ∩ Vj because HomA (−, X) is separated, and we conclude.
Lemma: If the image of a morphism of schemes φ : Spec B → Spec A is dense, then it contains
any point x ∈ Spec A defined by a minimal prime p of A. Moreover
1. If φ is a flat morphism and y ∈ Spec B is defined by a minimal prime of B, then φ(y) is

defined by a minimal prime of A.
2. If φ is faithfully flat, then Spec A has the quotient topology (a subset Z ⊆ Spec A is closed
if and only if φ−1 Z ⊆ Spec B is closed).
/ Im φ, then Bp = 0; hence f = 0 in B for some f ∈ A − p, and φ−1 (Uf ) = ∅.

Proof: If x ∈
(1) Since Aφ(y) → By is flat and local, it is faithfully flat; hence y = Spec By → Spec Aφ(y)
is surjective, so that Spec Aφ(y) = φ(y).
(2) Replacing Z by Z, we may assume that Z is dense.
If φ−1 Z is closed (hence affine), applying (1) to φ−1 Z → Z ⊆ Spec A we see that Z contains
any point of Spec A defined by a minimal prime.
Now, by (2), the closed set φ−1 Z contains any point of Spec B defined by a minimal prime;
hence φ−1 Z = Spec B, and Z = Spec A because φ : Spec B → Spec A is surjective.
Corollary: If X, Y are A-schemes, then B HomB (XB , YB ) is a sheaf.
Proof: It is clear that HomA (X, Y ) = Hom(X • , Y • ), and the result follows from the recollement
of sheaf morphisms.
The fpqc Topology:

A morphism of schemes φ : Y → X is quasicompact when the inverse image of any compact
open set also is compact (i.e. the inverse image of any affine open set is a finite union of affine
open sets) and it is faithfully flat when it is flat (if x = φ(y), then OY,y is a flat OX,x -module)
and set-theoretic surjective. If S is a scheme, the fpqc topology on the category Sch|S of
S-schemes is the topology generated by the Zariski topology and the topology defined by the
faithfully flat quasicompact S-morphisms Y → X.
Lemma: If a presheaf F on Sch|S is separated (resp. a sheaf ) in the Zariski topology and
F(A) → F(B) is injective (resp. F(A) → F(B) ⇒ F(B ⊗A B) is exact) for any faithfully flat
S-morphism Spec B → Spec A, then F is separated (resp. a sheaf ) in the fpqc topology.
Proof: We have to show that F is separated (resp. a sheaf) in the pretopology defined by the
faithfully flat quasicompact S-morphisms f : R → X.
−1
S S
Fix an affine open cover X = Q U
i i , and finite
Q affine open covers f (Ui ) = α Uiα .
Now, the composition F(X) → i F(Ui ) → i F(qα Uiα ) is injective and it factors through
F(X) → F(R), so that F(X) → F(R) is injective and F is separated.
Moreover, if s ∈ F(R) is in the kernel of F(R) ⇒ F(R ×X R), put siα = s|Uiα .
Since F(qα Uiα ) = Πα F(Uiα ) and qα Uiα is affine, there is si ∈ F(Ui ) such that siα = si |Uiα .
Now, si = sj on Ui ∩Uj = Ui ×X Uj , since both coincide on the cover {Uiα ×X Ujβ → Ui ×X Uj }
and F is separated, so that there is t ∈ F(X) such that si = t on Ui .
Hence s = si = t on Uiα , and s = t|R since F is separated. q.e.d.
The above results show that in the category of S-schemes T → S with the fpqc topology
1. O(T ) = OT (T ) is a sheaf of rings, and any quasi-coherent OS -module M defines a sheaf of

O-modules M(T ) = (M ⊗OS OT )(T ) (also named quasi-coherent).
2. If M, N are quasi-coherent OS -modules, T HomOT (M ⊗OS OT , N ⊗OS OT ) is a sheaf.
3. Any locally quasi-coherent O-module also is quasi-coherent.
4. The functor of points X • = HomS (−, X) of any S-scheme X is a sheaf.
5. If X, Y are S-schemes, then T HomT (X ×S T, Y ×S T ) is a sheaf.
6. If a presheaf of sets is locally representable by affine morphisms, then it is representable by

an affine morphism X → S.
In fact (f : X → S) f∗ OX is an equivalence of the category of affine morphisms with the
(opposite) category of quasi-coherent OS -algebras.
Cech Cohomology: Let F be an abelian presheaf on a situs C. For any cover {Ui → X} we
have a complex of abelian groups Č • ({Ui → X}, F), where Ui0 ...ip = Ui0 ×X . . . ×X Uip ,
Č p ({Ui → X}, F) =
Q
i0 ,...,ip F(Ui0 ...ip )
p+1
d : Č p F −→ Č p+1 F, (ds)i0 ...ip+1 = (−1)k si0 ...ibk ...ip+1 |Ui0 ...ip ,
P
k=0
Ȟ p ({Ui → X}, F) = H p Č • ({Ui → X}, F) .

In particular, Ȟ 0 ({Ui → X}, F) = F(X) when F is a sheaf.

If {Vj → X} is a finer cover, we may fix X-morphisms Vj → Uθ(j) , hence a morphism of
complexes φ : Č • ({Ui → X}, F) → Č • ({Vj → X}, F) and group morphisms
Ȟ p ({Ui → X}, F) −→ Ȟ p ({Vj → X}, F).
If we fix other X-morphisms Vj → Uθ̄(j) we obtain a homotopic morphism φ̄ (i.e. we have

φ̄ − φ = dh + hd for some group morphisms h : Č p+1 ({Ui → X}, F) → Č p ({Vj → X}, F)) so
that it induces the same morphism on the cohomology groups:
p
(−1)k sθ(j0 )...θ(jk )θ̄(jk )...θ̄(jp ) |Vj0 ...jp
P
h(s)j0 ...jp =
k=0
So we obtain an inductive system of groups and (assuming that any object X admits a cofinal
set of covers) the Cech cohomology groups are defined to be
Ȟ p (X, F) = lim
−→
Ȟ p ({Ui → X}, F).
Note: In multiplicative notation we have
Ȟ 1 ({Ui → X}, F) = (sij ) ∈ ij F(Uij ) : ski = skj sji / ≡,

Q
where (sij ) ≡ (tij ) when tij = si sij s−1

Q
j for some (si ) ∈ i F(Ui ).
When the sheaf of automorphisms of a sheaf T (of rings, modules,...) is commutative, then
Ȟ 1 ({Ui → X}, Aut) classifies, up to isomorphisms, the sheaves on X locally isomorphic to T on
the cover {Ui → X}, and Ȟ 1 (X, Aut) classifies the sheaves on X locally isomorphic to T .
1 (S, G ) for any scheme S.

Theorem: Pic(S) = Ȟfpqc m
Proof: Locally free sheaves of rank n in the fpqc topology are locally free in the Zariski topology
(p. 211), and the sheaf of automorphisms of the trivial line sheaf is just the functor of points of
the multiplicative group, Gm (X) = OX ∗ .
Example: When a cover X → S admits a finite group of automorphisms G such that the
morphism G × X := qG X → X ×S X, (g, x) 7→ (x, gx), is an isomorphism (a Galois covering,...),
then F(X) is a G-module and we have isomorphisms G× . p. . ×G × X = X×S . .p+1 . . . . ×S X,
(g1 , . . . , gp , x) 7→ (x, g1 x, g1 g2 x, . . . , g1 . . . gp x), so that the Cech complex is
Č p (X/S, F) = HomEns G× . p. . ×G, F(X)

p
(−1)k f (. . . , gk gk+1 , . . .) + (−1)p+1 f (g1 , . . . , gp )
P
(df )(g1 , . . . , gp+1 ) = g1 f (g2 , . . . , gp+1 ) +
k=1
and we put Ȟ p (X/S, F) = H p G, F(X) because these groups only depend on the G-module

F(X). In multiplicative notation, we see that
H 1 (G, A) = (ag ) ∈ G A : agh = ag · g(ah ) / ≡

Q
where ag ≡ bg when bg = a−1 ag · g(a) for some a ∈ A.

Hence H 1 (G, K ∗ ) = Ȟ 1 (K/k, Gm ) = 0 when k → K is a Galois extension of group G.
When G = (σ) is cyclic, a cocycle (ag ) is fully determined by an element β = aσ ∈ K ∗ of
norm 1, so that β = σ(α)/α for some α ∈ K ∗ (Hilbert’s theorem 90).
15.7.2 Sheafification of a Presheaf

(1) Given a presheaf of sets P, on P(X) we may consider an equivalence relation: s ∼ s0 when
both coincide on some cover {Ui → X}.
Now Ps (X) := P(X)/∼ is a separated presheaf, and the natural morphism P → Ps has a
universal property: any morphism to a separated presheaf uniquely factors through Ps .
(2) Given a separated presheaf of sets S, for any object X we han an injective natural map
S(X) −→ LS(X) := lim

−→
Hom(R, S),
where R runs over the covering sieves of X, and the maps Hom(R, S) → Hom(R0 , S) are
injective. Assume that this inductive limits is a set for any object X (for example, when the
situs admits a set of topological generators).
Any morphism Y → X induces a natural map LS(X) → LS(Y ), so that LS is a presheaf of
sets, separated because so is S, and we have a canonical injective morphism S → LS.
Theorem: Let S be a separated presheaf. If LS(X) is a set for any object X, then LS is a sheaf
of sets, and any morphism S → F to a sheaf F uniquely factors trough LS.
Hom(S, F) = Hom(LS, F).

Proof: Let us consider a cover {Ui → X} and a LS-compatible family of sections si ∈ LS(Ui ).
By definition si is given by a S-compatible family of sections sij ∈ S(Vij ) on a cover {Vij → Ui }.
On the cover {Vij → X} the LS-compatible sections sij ∈ S(Vij ) are S-compatible, because
S is separated, and define the required section s ∈ LS(X).
Such section s is unique because LS is a separated presheaf.
Now let us see the universal property:
Any morphism S → F into a sheaf F induces natural maps
LS(X) = lim
−→
Hom(R, S) −→ lim
−→
Hom(R, F) = F(X)
so that it factors trough LS. The factorization is unique because, by construction, any section
s ∈ LS(X) comes, on a cover of X, from sections of S.
Definition: When L(Ps ) is a sheaf of sets (v.gr. when C admits a set of topological generators)
P ] = L(Ps ) is the sheafification of P or the associated sheaf to P.
Any morphism P → F into a sheaf of sets F uniquely factors trough P ] , any section
s ∈ P ] (X) is defined on some cover of X by sections of P, and two sections s, s0 ∈ P(X)
coincide in P ] (X) if and only if both coincide on some cover of X.
Note: Even if a category C has not a final object p, the sections of any presheaf P : C Ens
on it may be defined to be the projective limit (it may be not a set!)
lim
←−
P = lim
←−
P(X) := Hompresh (p, P)
C X∈Ob C
where p denotes the constant presheaf defined by a one-point set, the functor of points of the
hypothetical final object (analogously, we may define the projective limit of a covariant functor
C Ens). Hence, a section s of the projective limit is just a family of elements sX ∈ P(X),
compatible in the sense that sY = f ∗ (sX ) for any morphism f : Y → X.
The inductive limit of P over C is defined so that for any set T

Hom lim
−→
P, T = lim ←−
Hom P(X), T .
C X∈Ob C
(the inductive limit may be not a set!). A section s of the inductive limit is just an element
sX ∈ P(X), identifying sX with f ∗ (sX ) for any morphism f : Y → X.
Now, given a functor f −1 : C D and a presheaf of sets P on C, for any object V in
D we consider the category of pairs (U, V → f −1 U ), where U ∈ Ob C, the morphisms of
(U, V → f −1 U ) into (U 0 , V → f −1 U 0 ) being the morphisms φ : U → U 0 such that f −1 φ is a
V -morphism, and we put
(f ∗ P)(V ) = lim
−→
P(U ).
V →f −1 U
So we obtain a “presheaf” f ∗ P on D. When it is a true presheaf of sets, for any presheaf Q

on D we have the adjunction formula
Hom(f ∗ P, Q) = Hom(P, f∗ Q)
since any morphism f ∗ P → Q induces natural maps P(U ) → (f ∗ P)(f −1 U ) → Q(f −1 U ) =

(f∗ Q)(U ), and any natural transformation P(U ) → Q(f −1 U ) induces a natural transformation
(f ∗ P)(V ) = lim
−→
P(U ) −→ lim
−→
Q(f −1 U ) −→ Q(V ),
V →f −1 U V →f −1 U
When f −1 is continuous and F is a sheaf on C, if the “sheafification” of f ∗ F is a sheaf of

sets (also denoted f ∗ F), we also have Hom(f ∗ F, G) = Hom(F, f∗ G) for any sheaf G on D.
Examples: If C is a small category, then f ∗ F is always a presheaf of sets.
In the case of the full subcategory f −1 : R ,→ C|X defined by a crible R ⊆ X • , we have
f∗ Q = Q|R , and (f ∗ P)(U ) = P(U ) when the X-object U is in R, and (f ∗ P)(U ) = ∅ otherwise,
so that f ∗ P is a presheaf of sets.
In particular, if U is an open set in a topological space X, and we consider the inclusion
f −1 : Op(U ) ,→ Op(X), for any sheaf F on U , we have that f ∗ F is just the sheaf of continuous
sections of F et → U ,→ X.
Comparison Lemma: Let C be a situs and R a full subcategory, stable under fibred products
in C, with the induced topology. If any object of C admits a cover by objects of R (and for
any sheaf G on R the inverse image “presheaf” is a presheaf of sets), then the “categories” of
sheaves on C and R are equivalent, where a sheaf F on C corresponds to F|R :
1. We have Hom(F, F̄) = Hom(F|R , F̄|R ) for any two sheaves F, F̄ on C: given a morphism
φ : F|R → F̄|R , there is a unique morphism f : F → F̄ such that φ = f |R .
2. Given a sheaf G on R, we have G = F|R for some sheaf F on C.
Proof: Let f −1 : R ,→ C, f −1 V = V , be the inclusion functor.

First we show that the “sheaf” f ∗ G is in fact a sheaf of sets:
The presheaf P = f ∗ G fulfils that P(V ) = G(V ) when V ∈ Ob R and, since any cover of V in
C may be refined by a cover in R, we also have Ps (V ) = G(V ). Now, any object U in C admits
a covering crible R generated by a family {Vi → U } with Vi ∈ Ob R, so that any covering crible
of U contains a covering crible C generated by a family {Viα → Vi → U } with Viα ∈ Ob R.
Hence, any Ps -compatible family siα ∈ Ps (Viα ) defines, G being a sheaf, sections si ∈ Ps (Vi ),
and we have si = sj on Vi ×U Vj because both coincide on the cover {Viα ×U Vjβ → Vi ×U Vj }
and Ps is separated. That is to say, Hom(R, Ps ) = Hom(C, Ps ), so that Ps] (U ) = Hom(R, Ps ) is
a set, and we see that the “sheaf” f ∗ G = Ps] is a sheaf of sets.
Moreover, (f ∗ G)|R = G, because when U ∈ Ob R, we may take R = U • .
Finally, the natural map f ∗ (F|R ) → F is an isomorphism because (p. 454) any object of C
admits a cover by objects of R, and f ∗ (F|R ) (V ) = (F|R )(V ) = F(V ) when V ∈ Ob R.
Corollary: Let C be a situs and R a full subcategory, stable under fibred products in C, with
the induced topology. If R contains a set of topological generators of C, then the categories of
sheaves C
e and R e are equivalent, where a sheaf F on C corresponds to F|R .
Proof: By hypothesis we have full subcategories G ,→ R ,→ C, where Ob G is a set of topological

generators of C, that we may assume to be stable under fibred products.
By the comparison lemma, the categories C e and Re are equivalent to G.
e
Example: Let G be a group and let G-Sets be the situs of G-sets. The object G is a topological
generator; hence it defines a full subcategory fulfilling the above hypotheses, so that any sheaf F
on G-Sets is fully determined by the G-set F(G). Analogously, given a finite Galois extension
k → L of group G, the object Spec L is a topological generator of the category Trivop L/k , so that
any sheaf F is fully determined by the G-set F(Spec L).
Injective Resolutions:
Let C be a situs with a set of topological generators {Gα }.
In the category of sheaves of abelian groups on C, images, quotients, direct sums and induc-
tive limits are defined to be the sheafification of the corresponding presheaves, so that the usual
universal properties and isomorphism theorem F/Ker f = Im f hold.
Moreover, inductive limits preserve exact sequences, and for any sheaf of abelian groups F,
HomAbSh (Z[G•α ]] , F) = HomAbPrsh (Z[G•α ], F) = HomPrsh (G•α , F) = P(Gα )
so that the category of sheaves of abelian groups admits a generator U = ⊕α Z[G•α ]] , and any
subsheaf F 0 ⊆ F is fully determined by the set of all morphisms U → F factoring through F 0 .
In particular, the subsheaves of an abelian sheaf F form a set.
Lemma: An abelian sheaf I is injective if and only if the natural map Hom(U, I) → Hom(V, I)
is surjective for any subsheaf V of the generator U .
Proof: If F 0 is a subsheaf of a sheaf F, we must show that any morphism f : F 0 → I is the

restriction of a morphism F → I. Take a morphism ρ : U → F which does not factor trough
F 0 , put V = ρ−1 (F 0 ) and consider the morphism φ : V → I, φ = f ρ. By hypothesis it admits
an extension φ0 : U → I, and the argument of the ideal criterion (replacing the ring A by U and
the ideal I by V ) let us extend f to a morphism F 0 + Im ρ, and we conclude.
Theorem: Any abelian sheaf F admits a natural monomorphism F → I into an injective sheaf
I, so that we have a funtorial injective resolution F → I • F.
Proof: Let I be the set of all morphisms fi : Vi → F, where Vi is a subsheaf of the generator U .
The natural morphisms (1i , −fi ) : Vi → (⊕I U ) × F, where 1i : Vi → ⊕I U is the inclusion into
the i-th term, induce a natural morphism ⊕i∈I Vi → (⊕I U ) × F.
Let I1 F be the cokernel, so that we have a natural monomorphism F → I1 F and, for any
morphism fi : Vi → F, we have a commutative square
Vi /U
fi 1i

F / I1 F
Now, inductively, we define a sheaf Iα F for any ordinal number α, and a natural monomo-
morphism Iβ F → Iα F whenever β ≤ α.
When α = β + 1, put Iα F = I1 (Iβ F); and put Iα F = lim −→ β<α β
I F when α is a limit ordinal.
Let σ be the first ordinal number with infinite cardinal strictly greater than the cardinal ℵ
of the set of subsheaves of U , and let us see that Iσ F is an injective sheaf (so conluding).
Let f : V → Iσ F be a morphism. If we show that f (V ) ⊆ Iα F for some α < σ, we conclude.
Now, since σ is a limit ordinal, we have Iσ F = lim I F, so that
−→ α<σ α
V = f −1 lim f −1 (Iα F).

I F = lim
−→ α<σ α −→ α<σ
If the chain of subsheaves f −1 (Iα F) of V stabilizes at some index, the result is obvious.
Otherwise, there is a set Ω = {αi } of ordinal numbers αi < σ, with limit σ, and such that
|Ω| ≤ ℵ, against the following result of the theory of ordinal numbers:
Lemma: Let ℵ be an infinite cardinal, and σ the first ordinal with ℵ < |σ|. If Ω = {αi } is a set
of ordinal numbers αi < σ, with limit σ, then ℵ < |Ω|.
Proof: Otherwise |Ω| ≤ ℵ. Since |αi | ≤ ℵ, the cardinal of the limit of Ω is bounded above by the
cardinal of a union of ℵ sets of cardinal ℵ; i.e. by ℵ2 = ℵ. Absurd. q.e.d.
Definition: The right derived functors of a left exact functor F , defined on the category of
abelian sheaves over a situs with a set of topological generators, are (Rn F )(F) = H n [F (I • F)].
In particular, we have the cohomology groups H n (X, F) = H n [(I • F)(X)] of an object X
with coefficients in an abelian sheaf F.
Exercises
1. Analysis I
1. If a, b, c are cardinal numbers, show that 0 + a = a, 0 · a = 0, 1 · a = a, a0 = 1 (in particular 00 = 1),
a1 = a, 1a = 1, a + b = b + a, (a + b) + c = a + (b + c), a(b + c) = ab + ac, ab = ba, (ab)c = a(bc),
(ab)c = (ac )(bc ), ab+c = (ab )(ac ), (ab )c = abc .
Moreover, if a ≤ b, prove that a + c ≤ b + c and ac ≤ bc.
2. If a is a non null cardinal, show that 0a = 0.
If a1 ≤ a2 , b1 ≤ b2 are non null cardinals, show that ab1 1 ≤ ab2 2 . (Remark that 0 ≤ 1, while
1 = 00 > 01 = 0).
3. If 2 ≤ n ∈ N, prove that n + ℵ0 = ℵ0 , ℵ0 + ℵ0 = ℵ0 , nℵ0 = ℵ0 , ℵn0 = ℵ0 , n2ℵ0 = 2ℵ0 , nℵ0 = 2ℵ0 ,
2ℵ0 2ℵ0 = 2ℵ0 , (2ℵ0 )ℵ0 = 2ℵ0 , ℵ0 2ℵ0 = 2ℵ0 , ℵℵ0 0 = 2ℵ0 .
4. Let F be the set of all finite subsets of N. Prove that |F | = ℵ0 .
5. Prove that the rings of polynomials Q[x1 , . . . , xn ] are countable.
6. Let a = |A| be an infinite cardinal. Prove that ℵ0 ≤ a, ℵ0 + a = a and a + a = a. (Hint: By Zorn’s
lemma, A is a disjoint union of countable subsets, so that a = ℵ0 b, and a + a = ℵ0 b + ℵ0 b = ℵ0 b = a.)
Also prove that ℵ0 a = a and aa = a. (Hint: Consider pairs (B, φ) where B ⊆ A and φ : B → B × B
is a bijection. By Zorn’s lemma there is a maximal pair (B, φ). If b := |B| < a, take a subset C ⊆ B c
of cardinal b, so that we have a bijection C → (B × C) q (C × B) q (C × C); hence a bijection
B q C → (B × B) q (B × C) q (C × B) q (C × C) = (B q C) × (B q C) extending φ.)
If moreover 2 ≤ b ≤ 2a , prove also that ba = 2a .
7. If Y is a subset of a well-order X, show that Y is isomorphic to an initial ray of X.
8. Show that any well-order X coincides with the set of its incomplete initial rays, ordered by inclusion.
9. Let α, β be the ordinal numbers defined by two well-orders A, B respectively. Then α + β is defined
to be the ordinal of the disjoint union A q B, where a < b for any a ∈ A, b ∈ B; and αβ is defined to
be the ordinal of A × B, with the lexicographical order: (a1 , b1 ) < (a2 , b2 ) when a1 < a2 , or a1 = a2
and b1 < b2 . Show that ω < ω + 1 while 1 + ω = ω, and that ω < 2ω = ω + ω, while ω2 = ω.
10. Let x, y ∈ R. If x ≤ y, prove that −y ≤ −x. If 0 < x < y, prove that y −1 < x−1 .
11. If a Cauchy sequence (qn ) represents x ∈ R, show that (|qn |) represents |x|.
12. For all x, y ∈ R, show that |x + y| ≤ |x| + |y|.
13. Let z = a + bi be a complex number, b 6= 0, and ρ = |z|. Show that (the sign is the sign of b)
r r
√ √ ρ+z ρ+a ρ−a
z= ρ = ±i ·
|ρ + z| 2 2
14. If there is a map f : X → N with finite fibres, show that |X| ≤ ℵ0 .
Prove that the set of algebraic complex numbers (root of a non null polynomial with rational co-
efficients) is countable. Conclude that the set of transcendent (non-algebraic) real numbers is un-
countable. (In particular, they exist!). (Hint: Exercise 5).
15. Prove that any compact connected set in R is a closed interval [a, b], where a ≤ b.
465
466 EXERCISES
x
16. Prove that f : (−1, 1) → R, f (x) = 1−x2 is a homeomorphism. Conclude that any open ball in Rn is
homeomorphic to Rn .
17. Show that a map f : X → Y between metric spaces is continuous at a point x ∈ X if and only if for
any ε > 0 there exists δ > 0 such that f B(x, δ) ⊆ B(f (x), ε).
18. Prove that two continuous maps f, g : X → Y between metrizable spaces coincide whenever both
coincide on a dense subset of X.
19. If p is a point of a metric space X, show that f : X → R, f (x) = d(p, x), is continuous.
20. If f, h : X → R are continuous functions, prove that so are min(f, h) and max(f, h).
21. Prove that the cardinal of the ring of continuous functions C(R) is 2ℵ0 .
22. If X, Y are metric spaces, show that d (x, y), (x0 , y 0 ) := d(x, x0 ) + d(y, y 0 ) is a metric on X × Y , and

that we have a canonical isometry of the completion of X × Y onto X b × Yb .
23. Let f : U → V be a bijection between two open subsets of R. If f is differentiable, with non null
derivative at any point, prove that so is the inverse g : V → U , and that g 0 (y) = f 01(x) , where y = f (x).
R R
Moreover, show that f (x)dx + g(y)dy = xf (x) = g(y)y, (i.e.; if F (x) is a primitive of f (x), then
G(y) = g(y)y − F g(y) is a primitive of g(y)), and conclude that (draw a picture)
Z f (b) Z b
g(y)dy = f (b)b − f (a)a − f (x)dx.
f (a) a
0 0
24. Let f, g : (a, b) → R be differentiable functions. If |f (x)| ≤ g (x), ∀x ∈ (a, b), prove that |f (y)−f (x)| ≤
|g(y) − g(x)|, ∀x, y ∈ (a, b).
25. Let f : (a, b) → R be a differentiable function. Prove that f 0 (x) ≥ 0, ∀x ∈ (a, b), if and only if f is an
increasing function. Conclude that f 0 (x) ≤ 0, ∀x ∈ (a, b), if and only if f is a decreasing function.
26. Given a function f : U − {a} → R, we say that c ∈ R is the limit of f (x) as x → a when for any
ε > 0 there exists δ > 0 such that B(a, δ) ⊆ U and 0 < |x − a| < δ ⇒ |f (x) − c| < ε, and we put
lim f (x) = c or we say that f (x) → c as x → a. Prove the following assertions:
x→a
(a) If it exists, the limit is unique.

(b) The limit of f (x) as x → a exists if and only if f admits an extension f˜: U → R continuous at
x = a. Moreover, f˜(a) = lim f (x).
x→a
(c) A function f : U → R is continuous if and only if lim f (x) = f (a), ∀a ∈ U .
x→a
(d) A function f : U → R is differentiable at x = a if and only if the limit
f (x) − f (a) f (a + t) − f (a)
lim = lim
x−a
x→a t→0 t
0
exists, and in such a case such limit is just f (a).
(e) If a continuous function f is differentiable at x = a, then f (a + t) = f (a) + f 0 (a)t + ũ(t)t for some
continuous function ũ(t) defined on a neighborhood of t = 0 and vanishing at t = 0.
(f) L’Hospital’s Rule: Let f, g : U → R be differentiable functions such that f (a) = g(a) = 0. If g
0
and g 0 do not vanish on U − {a}, and the limit of fg0 (x)
(x)
as x → a exists, then so does the limit of
f (x)
g(x) as x → a, and both coincide. (Hint: Use Cauchy’s theorem to prove that in a neighborhood
f (x) f 0 (ξ)
g(x) = g 0 (ξ) , where 0 < |ξ − a| < |x − a|, so that ξ → a when x → a).
27. Let f : U → R be a function of class C n , n≥ 2, such that f 00 (a) = . . . = f (n−1 (a) = 0.

If n is even and f (n (a) > 0, prove that x = a is a point of strict convexity (in a neighborhood of a
we have f (x) > f (a) + f 0 (a)(x − a) when x 6= a.)
If n is even and f (n (a) < 0, prove that x = a is a point of strict concavity (in a neighborhood of a
we have f (x) < f (a) + f 0 (a)(x − a) when x 6= a.)
ANALYSIS I 467
If n is odd and f (n (a) > 0, then in a neighborhood of a we have that f (x) > f (a) + f 0 (a)(x − a) when
x > a, and f (x) < f (a) + f 0 (a)(x − a) when x < a.
If n is odd and f (n (a) < 0, then in a neighborhood of a we have that f (x) < f (a) + f 0 (a)(x − a) when
x > a, and f (x) > f (a) + f 0 (a)(x − a) when x < a.
28. Prove that the Taylor polynomial Tan f is the unique polynomial P (x) of degree ≤ n such that P (a) =
f (a), P 0 (a) = f 0 (a), P 00 (a) = f 00 (a), . . . , P (n (a) = f (n (a).
29. Let f be a function of class C n on an set U ⊆ R and a ∈ U . Prove that f (x) = (Tan f )(x)+O(x)(x−a)n ,
where O(x) is a continuous function on U such that O(a) = 0.
Ra Rb Ra Rb Rc Ra
30. If we put a f dx = 0, and a f dx = − b f dx when b < a, show that a f dx + b f dx + c f dx = 0
for any three real numbers a, b, c.
Rb
31. Let f : [a, b] → R be a continuous function. Show that a f dx = f (ξ)(b − a) for some a < ξ < b.
Rb
If morover a |f |dx = 0, prove that f = 0.
Rb
Conclude that d(f, h) = a |h − f |dx defines a metric on the ring C([a, b]).
32. Let ln : R+ → R be the inverse of the exponential ex : R → R+ . Prove that (ln x)0 = x1 , and determine
a primitive of the function x1 : R− → R.
R∞
33. For any positive real number t, show that 0 e−tx dx = t−1 . Differentiate under the integral sign to
R∞ R∞ R∞
obtain 0 xe−tx dx = 1/t2 , 0 x2 e−tx dx = 2/t3 and in general 0 xn e−tx dx = n!/tn+1 . Then obtain
Euler’s formula Z ∞
n! = xn e−x dx.
0
∞ ∞ 2 2
e−t (1+x )/2
Z Z
−x2 /2
34. Let us consider the integral I = e dx. For any t ∈ R set f (t) = dx.
0 0 1 + x2
Show that f (0) = π/2 and f (t) → 0 as t → ∞. Differentiate under the integral sign to obtain
Z ∞ Z ∞
0 −t2 /2 −(tx)2 /2 −t2 /2 2 2
f (t) = −e e tdx = −e e−y /2 dy = −Ie−t /2 ,
0 Z a Z0 a
2
0
f (a) − f (0) = f (t)dt = −I e−t /2 dt.
0 0
∞ 2
e−t /2
Z
When a → ∞, conclude that I 2 = π/2, so obtaining the gaussian integral √ dt = 1.
−∞ 2π
R∞ −tx2 /2
p
Now, for any positive real number t, show that −∞ e dx = 2π/t, and differentiate under the
integral sign to obtain the moments of the gaussian:
Z ∞ √
2n −tx2 /2 1 · 3 · 5 · . . . · (2n − 1) 2π
x e dx =
−∞ t(2n+1)/2
Z ∞ 2n −x2 /2
x e
√ dx = 1 · 3 · 5 · . . . · (2n − 1).
−∞ 2π
R∞ √
35. Consider the Euler interpolation n! = 0 xn e−x dx of the factorial. Use the substitution y = x to
√
show that 12 ! = 12 π.
36. Assume that there is a non constant polynomial P (x) ∈ C[x] without complex roots, so that we have
1
a well-defined continuous function P (z) : C → C. For any positive r ∈ R consider the integral
Z 2π
dθ
I(r) = iθ )
·
0 P (re
Differentiate under the integral sign to obtain that I(r) is constant:
∂ 1 P 0 (reiθ ) iθ ∂ 1
iθ
= − iθ 2
ire = ir
∂θ P (re ) P (re ) ∂r P (reiθ )
468 EXERCISES
Z 2π θ=2π
1 ∂ 1 1
I 0 (r) = dθ = = 0.

ir ∂θ P (reiθ ) irP (reiθ ) θ=0

0
R 2π 1
R 2π 1
Now, 0 P (re iθ ) ≤ goes to 0 as r → ∞. Hence I(r) = 0. Obtain a contradiction, since

0 |P (reiθ )|
I(r) → 2π/P (0) as r → 0, so obtaining a new proof of D’Alembert’s theorem.
37. If 0 6= z ∈ C, show that the argument of z/z̄ doubles the argument of z.
Show that there are infinite angles with rational sine and cosine.
Show that the equation x2 + y 2 = z 2 has infinite integer solutions with x, y, z coprime.
38. Prove that the centers of the exterior squares drawn on two sides of a triangle and the middle point
of the third side determine a rectangle isosceles triangle.
ln 2
39. Where does the following argument fail? 2 = eln 2 = e2πi 2πi = (e2πi )(ln 2)/(2πi) = 1(ln 2)/(2πi) = 1.
x −x x −x
40. Consider the hyperbolic functions cosh x = e +e 2 and sinh x = e −e
2 . Show that (cosh x)0 =
0 2 2
sinh x, (sinh x) = cosh x, cosh x − sinh x = 1, cosh(x + y) = (cosh x)(cosh y) + (sinh x)(sinh y),
sinh(x + y) = (sinh x)(cosh y) + (cosh x)(sinh y).
2. Linear Algebra
1. Does the operation x ∗ y = (x + y)/(1 + xy) define a group structure on the interval (−1, 1)?
2. Given a group G, the opposite group Gop is defined to be the set G with the operation a ∗ b := ba.
Show that we have a natural group isomorphism G −∼
→ Gop , a 7→ a−1 .
3. Prove that the set Aut(G) of all automorphisms of a group G, with the composition of maps, is a
group, that any element a ∈ G defines an inner automorphism, τa : G → G, τa (x) = axa−1 , and
that so we obtain a morphism of groups τ : G → Aut(G), τ (a) = τa .
4. Prove that a group G is commutative if and only if the product G × G → G is a group morphism.
5. Prove that no group is a union of two smaller subgroups.
6. Let H be a subgroup of a group G. If aHa−1 ⊆ H, ∀a ∈ G, prove that aHa−1 = H, ∀a ∈ G.
7. If G is a group, prove that ord(ab) = ord(ba), ∀a, b ∈ G. (Hint: ord(a) = ord(bab−1 ).)
8. Prove that any subgroup of index 2 is a normal subgroup.
Q
9. If H0 ⊂ H1 ⊂ . . . ⊂ Hn is a sequence of normal subgroups, show that |Hn /H0 | = i |Hi /Hi−1 |.
10. Prove that any group of prime order is cyclic.
11. Prove that a finite group G is a cyclic group if and only if for any divisor d of |G| there is a unique
subgroup of G of order d.
12. Let G be a finite group of order n and a ∈ G. Prove that the equation xm = a admits a unique
solution in G when n, m are coprime. (Hint: nZ + mZ = Z.)
13. Show that the center Z(G) = {g ∈ G : gx = xg, ∀x ∈ G} of a group G is a normal subgroup of G.
If G/Z(G) is a cyclic group, prove that G is an abelian group.
14. Prove that any group of order 4 is isomorphic to Z/4Z or to (Z/2Z) × (Z/2Z).
15. If H1 , H2 are subgroups of a group G, we put H1 H2 = {h1 h2 : hi ∈ Hi }.
Prove that H1 H2 is a subgroup if and only if H1 H2 = H2 H1 . Conclude that H1 H2 is a subgroup
whenever H1 is a normal subgroup of G.
If G is a finite group, prove that |H1 | · |H2 | = |H1 H2 | · |H1 ∩ H2 |.
16. Show that A4 has no subgroup of order 6. (The converse of Lagrange’s theorem fails).
17. Prove that the symmetric group Sn has trivial center, Z(Sn ) = 1, when n ≥ 3.
18. Show that An is generated by the 3-cycles. (Hint: (12)(23) = (123) and (12)(34) = (123)(234)).
LINEAR ALGEBRA 469
19. Show that any non trivial normal subgroup of S4 is {Id, (12)(34), (13)(24), (14)(23)} or A4 .
20. If n 6= 4, prove that An is the unique non trivial normal subgroup H of Sn .
(Hint: Let σ = σ1 . . . σr be the decomposition as a product of disjoint cycles, of decreasing order, of
an element of H. If σ1 = (1, . . . , d), d ≥ 3, take σ̄ = (d, . . . , 3, 1, 2)σ2−1 . . . σr−1 , so that σ̄ ∈ H and
(1, d, 2) = σ̄σ ∈ H. If σ = (12)(34)σ3 . . . σr , take σ̄ = (13)(24)σ3−1 . . . σr−1 , so that (14)(23) = σ̄σ ∈ H
and conclude that (154) = (14)(23) · (23)(15) ∈ H).
21. Prove that the sign defines the unique epimorphism Sn → {±1} = S2 , and show the existence of an
epimorphism S4 → S3 . (Hint: A tetrahedron has 3 pairs of opposite edges).
If n > 4, prove that there is no epimorphism Sn → Sm , where 2 < m < n.
22. Prove that the group of automorphisms of a cyclic group of order n is isomorphic to the group (Z/nZ)∗
of invertible elements of the ring Z/nZ.
23. If d divides n, show that the number of elements of Z/nZ of order d is φ(d).
P
Conclude that n = d|n φ(d) for any natural number n. (We agree that φ(1) = 1).
24. If A is a subset of a set X, there is an indicator function a : X → F2 such that a(A) = 1, a(Ac ) = 0.
So we may identify P(X) with the ring of F2 -valued functions on X, the null function corresponding
to ∅ and the function 1 corresponding to X. Prove that
(a) We have a + a = 0, a2 = a and, if b corresponds to a subset B ⊆ X, then
Ac = X − A corresponds to 1 + a,
A ∩ B corresponds to ab,
A ∪ B corresponds to a + b + ab.
A M B = (A ∩ B c ) ∪ (Ac ∩ B) corresponds to a + b.
A − B = A ∩ B c corresponds to a + ab.
(b) The symmetric difference A M B defines a commutative group structure on P(X).
(c) A ∩ (B M C) = (A ∩ B) M (A ∩ C) and (A M B) − C = (A − C) M (B − C).
25. Let 0 and 1 represent the logical values true and false respectively, so that the propositional calculus
may be viewed as a calculus with F2 -valued functions, and tautologies are just logical operations
corresponding to the zero function. Prove the following correspondence between logical operations
and operations with F2 -valued functions:
Logical Operation Functional Operation
¬p 1+p
p∨q pq
p∧q p + q + pq
p⇒q (1 + p)q
p⇔q p+q
If a1 , . . . , an ∈ F2 , check that (p1 + a1 + 1) · · · (pn + an + 1) vanishes everywhere, except at p1 =
a1 , . . . , pn = an . Determine a logical operator with a given truth table (Post’s problem).
a·
26. Prove that any finite integral ring A is a field. (Hint: Consider the group morphisms A −−→ A).
27. If A is a ring, prove that there is a unique ring morphism Z → A.
28. Let End(E) be the (non commutative) ring of all endomorphisms of an abelian group E. Prove that
the structures of k-vector space on E correspond to the ring morphisms k → End(E).
29. Prove that the direction of a linear subvariety is well-defined: if x + V = y + W , then V = W .
30. Let k be a field where 1 + 1 6= 0. If a non empty subset X of a k-vector space contains the line passing
through any two points of X, prove that X is a linear subvariety.
Does this statement hold when k = F2 ?
31. Let V1 , V2 be two vector subspaces of a vector space E. Prove that (V1 + V2 )/V2 ' V1 /(V1 ∩ V2 ).
If moreover V1 ⊆ V2 , prove that (E/V1 )/(V2 /V1 ) ' E/V2 .
470 EXERCISES
32. If f : E → E is a linear map and f 2 = f , prove that E = (Ker f ) ⊕ (Im f ).

33. Let V, W be two vector subspaces of a vector space E. Prove that W is a supplement of V in E if and
only if the canonical projection π : E → E/V induces an isomorphism W ' E/V .
34. Let f : E → F be a linear map. Let V ⊂ E be a vector subspace and put U = {e ∈ E : e ∈ / V }. If
f (V ) = F , prove that f (U ) = F .
P
35. If V0 ⊂ V1 ⊂ . . . ⊂ Vn is a sequence of vector subspaces, show that dim (Vn /V0 ) = i dim (Vi /Vi−1 ).
36. If 0 → E0 → E1 → . . . → En → 0 is an exact sequence of finite-dimensional vector spaces, show that
dim E0 − dim E1 + . . . + (−1)n dim En = 0.
37. Let f : E → E 0 be a linear map between finite dimensional vector spaces. Prove that the matrix of f
in some bases has all the coefficients 0 except some ones in the diagonal, which are just 1.
38. If two linear forms ω, ω 0 : E → k have equal kernel, show that ω 0 = λω for some 0 6= λ ∈ k.
f h h∗ f∗
39. If E1 − → E3 is an exact sequence of linear maps, prove that so is E3∗ −→ E2∗ −→ E1∗ .
→ E2 −
40. The cokernel of a linear map f : E → F is Coker f := F/Im f . Prove that (Ker f )∗ = Coker f ∗ and
(Coker f )∗ = Ker f ∗ . (Hint: 0 → Im f → F → Coker f → 0 is exact).
41. If E, F are finite-dimensional k-vector spaces, show that the map Homk (E, F ) → Homk (F ∗ , E ∗ ),
f 7→ f ∗ , is an isomorphism.
42. If E = V ⊕ W , show that we have a natural isomorphism E ∗ = V ∗ × W ∗ .
P
43. Show that the trace tr(aij ) = i aii of square matrices satisfies tr(AB) = tr(BA); hence
tr(A1 . . . An−1 An ) = tr(An A1 . . . An−1 ).
Moreover, hA|Bi := tr(Āt B) is a scalar product on the vector space Mn×n (K).
v·e
44. Show that the polynomial kte + |e·v| vk2 = kek2 t2 + 2|e · v|t + kvk2 , where t ∈ R and e, v are two vectors
in a Euclidean vector space, has discriminant ≤ 0, and obtain another proof of the Cauchy-Schwarz
inequality.
45. Let e, v be non null vectors in a Euclidean vector space.
If |he|vi| = kek · kvk , prove that e, v are linearly dependent.
If ke + vk = kek + kvk, prove that v = λe, ∃λ ∈ R+ .
46. If two linear subvarieties X = p + V , Y = q + W of a vector space E are perpendicular (i.e. V and
W ⊥ are incident), prove that V ∩ W = 0 or V + W = E.
47. Prove the following statements in any real Euclidean vector space:
(a) (Pons asinorum): Any isosceles triangle has two equal angles.
(b) The sum of the squares of the sides of a warped quadrilateral abcd is the sum of the squares of
the two diagonals ac, bd plus 4 times the square of the segment joining their middle points.
(c) Let g, h, f be the barycenter, orthocenter and circumcenter of a triangle abc. Prove that the 3
middle points of the sides, the 3 foots of the heights, and the 3 middle points of the segments
determined by the orthocenter and the vertices, are all inscribed in a circle, centered at the middle
point between the orthocenter and the circumcenter, h+f 2 =
3g+h
4 = a+b+c+h
4 ·
48. Let E be a Euclidean vector space. A linear map f : E → E is said to be an isometry if it preserves
the scalar product: e · v = f (e) · f (v), ∀e, v ∈ E. Prove the following statements:
(a) Any isometry is a linear isomorphism.

(b) If A is the matrix of f in any orthonormal basis of E, then Āt A = I.
(c) For any vector subspace V ⊆ E, there is an isometry sV : E → E such that sV is the identity on
V and −Id on V ⊥ .
LINEAR ALGEBRA 471
(d) In the real case, the determinant of the matrix A of f in any basis of E is ±1, (and isometries of
determinant 1 are named rotations).
(e) In the real case, any bijection f : E → E preserving the origin, f (0) = 0, and distances, |f (v) −
f (e)| = |v − e|, is an isometry. (Hint: It preserves
the quadratic form q(e) = e · e; hence the
scalar product, so that f (e + v) − f (e) − f (v) · f (w) = 0.)
49. Prove that the group RT of rotations of a tetrahedron (rotations transforming the figure into itself)
is isomorphic to A4 , and that the group IT of isometries is isomorphic to S4 .
Prove that the group RC of rotations of a cube (hence of the dual octahedron) is isomorphic to S4 ,
and the group IC of isometries is isomorphic to S4 × {±1}.
Prove that the group RD of rotations of a dodecahedron (hence of the dual icosahedron) is isomorphic
to A5 , and the group ID of isometries is isomorphic to A5 × {±1}.
Check that the four vertices of a tetrahedron and the four vertices of the symmetric tetrahedron (with
respect to the center) determine a cube, while the middle points of the edges of a tetrahedron are the
vertices of an octahedron.
However, prove that from a tetrahedron you can not define a dodecahedron in a natural way; from
a cube you can not define a tetrahedron nor a dodecahedron; and from a dodecahedron you can not
define a tetrahedron nor a cube. (Hint: If you naturally define a platonic solid B from another one
A, show that you have injective group morphisms RA → RB and IA → IB ).
50. Let T be an endomorphism of a vector space E.
If all vectors e ∈ E are eigenvectors of T , prove that T = λId for some scalar λ.
If e ∈ E and T (e) are linearly independent, show that ST 6= T S for some S ∈ Endk (E).
If T commutes with all the endomorphisms of E, prove that T = λId for some scalar λ.
51. If two selfadjoint endomorphisms S, T of an Euclidean vector space E commute, ST = T S, prove that
E admits an orthonormal base of common eigenvectors.
52. If f is an isometry of an hermitian vector space E, show that any eigenvalue α ∈ C of f has modulus 1,
and that E admits an orthonormal base of eigenvectors of f . (Hint: If f (V ) = V , then f (V ⊥ ) = V ⊥ .)
53. Show that any linear map f : Rn → Rn , f (X) = AX, admits an invariant vector subspace of dimension
≤ 2. (Hint: If f has no eigenvector, there exists 0 6= X ∈ Cn such that AX = αX; hence AX̄ = ᾱX̄,
and X, X̄ are linearly independent because ᾱ 6= α. Now X + X̄, 1i (X − X̄) ∈ Rn generate an invariant
plane.)
If f is an isometry of a real Euclidean vector space E, conclude that E admits a decomposition
E = V1 ⊕ . . . ⊕ Vn as a direct sum of invariant vector subspaces of dimension 1 or 2.
54. Let T be an endomorphism of a k-vector space of finite dimension E. Prove that the eigenvalues of
hT : Endk (E) → Endk (E), hT (S) = T S, are just the eigenvalues of T .
55. If α ∈ k is a root of multiplicity m of the characteristic polynomial cT (x) of an endomorphism T ,
prove that 1 ≤ dim Vα ≤ m, where Vα = Ker (αId − T ).
56. Let T be an endomorphism of a finite-dimensional vector space E. If V ⊂ E is an invariant vector
subspace, T (V ) ⊆ V , prove the existence of a unique endomorphism T̄ : E/V → E/V such that
T̄ (ē) = T (e), ∀e ∈ E. If c̄(x) is the characteristic polynomial of T̄ , and c0 (x) is the characteristic
polynomial of the restriction T |V : V → V , T |V (v) = T (v), prove that cT (x) = c̄(x)c0 (x).
If T is diagonalizable, conclude that so are T |V and T̄ .
Q
57. Let cT (x) = i (x − αi ) be the characteristic
Q polynomial of an endomorphism T of a k-vector space
E. If P (x) ∈ k[x], show that cP (T ) (x) = i (x − P (αi )). (Hint: By Kronecker’s theorem, you may
assume that T has an eigenvalue).
Conclude that T is nilpotent if and only if tr(T r ) = 0, 1 ≤ r ≤ dim E.
58. If T is a nilpotent endomorphism of a vector space of dimension n, show that T n = 0.
59. Prove that the characteristic (resp. annihilator) polynomial of an endomorphism T coincides with the
characteristic (resp. annihilator) polynomial of the transpose endomorphism T ∗ .
472 EXERCISES
60. If two diagonalizable endomorphisms of a finite-dimensional vector space E commute, prove that both
diagonalize in a common basis of E.
61. If two endomorphisms of a finite-dimensional complex vector space commute, prove that they have a
common eigenvector.
62. Let G be a finite group of automorphisms of a finite-dimensional complex vector space E. Prove that
any automorphism τ ∈ G is diagonalizable. (Hint: Look at the annihilator of τ ).
63. Let J be an endomorphism of a real vector space E 6= 0 of dimension n. If J 2 = −Id, prove that n is
even and any vector is in an invariant plane. (Hint: E admits a complex structure).
64. Prove that a class [Q(x)] is invertible in k[x]/(P (x)) if and only if the polynomials P (x) and Q(x)
have no common root (in any extension of k).
Let d ≥ 2 be a natural number. Show that 1 + t = S(t)d for some formal power series S(t) ∈ C[[t]],
and prove that any invertible element of C[x]/(xn ) is a d-th power. Obtain that any invertible element
in C[x]/(P (x)) is a d-th power. (Hint: The Chinese remainder theorem).
If A is an invertible matrix with complex coefficients, conclude the existence of a polynomial Q(x) ∈
C[x] such that A = Q(A)d . (Hint: In C[x]/(φA (x)), the class [x] is invertible).
j ...j
65. Let Ti11...ipq be the coordinates of a (p, q)-tensor T in some base e1 , . . . , en of E, and let us consider
a new base ē1 , . . . , ēn of E. Let B = (bij ) = (bij ) be the base change matrix, ēj = i bij ei , and put
P
j ...j
B −1 = (cij ) = (cij ). Prove that the coordinates T̄i11...ipq of T in the new base are just
n
j ...j P k ...k j ...j l ...l
T̄i11...ipq = bi11...ipp cl11...lqq Tk11 ...kqp
k1 ...kp l1 ...lq =1
66. If char k 6= 2, show that any 2-covariant tensor on a finite-dimensional k-vector space decomposes
uniquely as a sum of a symmetric tensor and an alternate tensor.
67. If E is a finite-dimensional k-vector space, show that T 2 E ∗ = Homk (E, E ∗ ).
68. Let f ∗ : T p E ∗ → T p F ∗ and f∗ : T p F → T p E be the linear maps induced by a linear map f : F → E
between vector spaces of finite dimension, p ≥ 1. Prove that
(a) f is injective ⇔ f ∗ is surjective ⇔ f∗ is injective.
(b) f is surjective ⇔ f ∗ is injective ⇔ f∗ is surjective.
69. When dim E = 3, prove that any 2-form Ω2 ∈ Λ2 E ∗ is Ω2 = ω1 ∧ ω2 for some linear forms ω1 , ω2 ∈ E ∗ .
When dim E > 3, and ω1 , ω2 , ω3 , ω4 ∈ E ∗ are linearly independent, show that Ω2 = ω1 ∧ ω2 + ω3 ∧ ω4
is not an exterior product Ω2 = ω ∧ ω 0 of linear forms.
70. Prove that any non null vector e ∈ E induces exact sequences (where n = dim E)
e∧ e∧ e∧ e∧ e∧
0 −→ Λ0 E −−−→ Λ1 E −−−→ Λ2 E −−−→ . . . −−−→ Λn−1 E −−−→ Λn E −→ 0
i i i i i
0 −→ Λn E ∗ −−e→ Λn−1 E ∗ −−e→ . . . −−e→ Λ2 E ∗ −−e→ Λ1 E ∗ −−e→ Λ0 E ∗ −→ 0
71. Show that a family of vectors e1 , . . . , ep ∈ E is linearly independent if and only if Ωp (e1 , . . . , ep ) 6= 0
for some p-form Ωp on E.
72. Show that any linear map f : F → E induces a natural linear map Λp f : Λp F → Λp E, and prove that
Λp f = 0 if and only if p > dim (Im f ).
73. For any matrix (aij ) ∈ Mn×n (k), let Aij denote the minor obtained by removing the i-th row and the
j-th column, multiplied by (−1)i+j . Given 1 ≤ i, j ≤ n, prove that ai1 Aj1 + . . . + ain Ajn defines a
n-form Mn×n (k) = k n × . . . × k n → k, and conclude that ai1 Aj1 + . . . + ain Ajn = |A| when i = j, and
it is 0 otherwise, so that
   |A| . . . 0 
a11 . . . a1n A11 . . . An1
. .. .. 
... ... ...  ... ... ...  =   .. . . 
an1 . . . ann A1n . . . Ann 0 . . . |A|
LINEAR ALGEBRA 473
Conclude that this formula holds for any square matrix (aij ) with coefficients in a ring. (Hint: Any
matrix is the specialization of the generic matrix (xij ) with coefficients in the polynomial ring Z[xij ],
which is contained in a field).
74. Let e1 , . . . , en be a basis of a real Euclidean vector space E, and
√ let ω1 , . . . , ωn be the dual basis. If
we put G = det(ei · ej ), show that the volume form is ΩE = ± G ω1 ∧ . . . ∧ ωn .
Conclude that the area S of the parallelogram determined by two vectors e, v ∈ E is
s
e·e e·v p
S= = kek2 kvk2 − (e · v)2 = kek · kvk · |sen α|,
v·e v·v
where α is the angle between e and v, and obtain Heron’s formula for the area of a triangle whose
sides have lengths a, b, c
p p
S = 41 4a2 b2 − (a2 + b2 − c2 )2 = s(s − a)(s − b)(s − c) , s = 21 (a + b + c).
75. Let E be a real Euclidean vector space. If e1 , . . . , en ∈ E, we put G(e1 , . . . , en ) = det(ei · ej ). Prove
that
(a) G(e1 , . . . , en ) ≥ 0, and G(e1 , . . . , en ) = 0 if and only if e1 , . . . , en are linearly dependent.
(b) The volume of the parallelepiped determined by a basis e1 , . . . , ed of E is G(e1 , . . . , ed )1/2 .
(c) If G(e1 , . . . , en ) 6= 0, the distance of e ∈ E to the vector subspace he1 , . . . , en i is just
s
G(e, e1 , . . . , en )
·
G(e1 , . . . , en )
76. Let ΩE be the volume form of a 3-dimensional Euclidean R-vector space E. If e, v ∈ E, then the linear
form iv ie ΩE = ΩE (e, v, −) corresponds to a vector e × v, named cross product of e and v, so that
(e × v) · u = ΩE (e, v, u). Prove that it is bilinear, e × v = −v × e, and that e × v 6= 0 if and only if e
and v are linearly independent, and in such a case e × v is orthogonal to both factors, the module of
e × v is the area of the parallelogram determined by e and v, and e, v, e × v is a direct base. (Hint:
iu ΩE is the area form of the plane he, vi, and the area of the parallelogram determined by e and v is
|(iu ΩE )(e, v)| = |ΩE (u, e, v)| = |(e×v)·u| = |λu·u| = λ. Moreover, ΩE (e, v, e×v) = (e×v)·(e×v) > 0).
77. Let e1 , e2 , e3 be a direct basis
P of an oriented Euclidean R-vector space, and let e1 , e2 , e3 be the dual
i
P
basis, e · ej = δij . If u = i ui ei , v = i vi ei , show that

1 u2 u3
e1 + u3 u1 e2 + u1 u2 e3 .

(a) u × v =
ΩE (e1 , e2 , e3 ) v2 v3
v3 v1 v1 v2
(b) (u × v) · w = u · (v × w).
(c) (u × v) × w = (u · w)v − (v · w)u. (Hint: Consider a direct orthonormal base u1 , u2 , u3 such that
u ∈ Ru1 , v ∈ Ru1 + Ru2 ).
(d) (u × v) × w + (v × w) × u + (w × u) × v = 0.
(e) ku × vk2 = kuk2 kvk2 − (u · v)2 . (Hint: Use ex. 74 or the following formula.)

u · ū v · ū
(f) (u × v) · (ū × v̄) =
.
u · v̄ v · v̄
(Hint: Use formula (c), or remark that both terms are covariant tensors symmetric in u, ū and
v, v̄, defining the same quadratic form according to the former formula.)
78. Let T be an endomorphism of a 3-dimensional oriented Euclidean R-vector space E. If e · (T e) = 0
for all e ∈ E, prove the existence of a vector v ∈ E such that T (e) = e × v for all e ∈ E.
79. Let u be a non null vector of modulus ω in a real Euclidean vector space of dimension 3. Prove that
v = u × ~r is the velocity of a body at the position ~r, tourning around us with angular velocity ω
(radians per time unit) and axis u. If we consider the endomorphism T : E → E, T (~r) = u × ~r, the
equations of such motion, with initial position ~r0 , are just
~r = etT ~r0 , ~r 0 = T etT ~r0 ) = T (~r ).
474 EXERCISES
When ω = 1, we have u × (u × v) = −v (remark that ΩE (u, u × v, v) = −ΩE (u, v, u × v) < 0).

Hence T 3 = −T (give an alternative proof showing that cT (x) = x3 + x).
Conclude that T 4 = −T 2 , T 5 = T, . . . and that the rotation of t radians around the axis u is
t2 2 t3 3 t4 4 t5 5 t6 6 t7 7
etT = 1 + tT + T + T + T + T + T + T + ...
2! 3! 4! 5! 6! 7!
2 3 4 5 6 7
t t t t t t t8
= 1 + tT + t2 − T − T 2 + T + T 2 − T − T 2 . . .
2! 3! 4! 5! 6! 7! 8!
= 1 + (sin t)T + (1 − cos t)T 2 .
When t = 2π, we obtain e2πT = 1 (an equation close to e2πi = 1, since in the complex plane ωiz is
also the velocity of a body, placed at the position z, turning around us with angular velocity ω).
3. Algebra I
1. Let k be a field. If a ∈ k, show that we have a ring isomorphism k[x]/(x − a) ' k, [P ] 7→ P (a).
Given different elements a1 , . . . , an ∈ k, show that we have a ring isomorphism

k[x]/ (x − a1 ) . . . (x − an ) ' k ⊕ . . . ⊕ k, [P ] 7→ (P (a1 ), . . . , P (an )).
2. Let k be a field and a1 , . . . , an ∈ k. Prove that m = (x1 − a1 , . . . , xn − an ) is a maximal ideal of
k[x1 , . . . , xn ] and that the natural morphisms k → k[x1 , . . . , xn ]/m is an isomorphism.
If P ∈ k[x1 , . . . , xn ], show that the following generalized Ruffini’s rule holds:
P
P (a1 , . . . , an ) = 0 if and only P = i (xi − ai )Qi .
3. Let a, b be two ideals of a ring A, and p a prime ideal. If ab ⊆ p, show that a ⊆ p or b ⊆ p.
4. Let p be a prime number, and let d be a divisor of p − 1. If p does not divide n ∈ Z, prove that n is a
p−1
d-th residue modulo p if and only if n d ≡ 1 (mod. p).
5. Let Q(x) = (x2 − 2)(x2 + 1)(x2 + 2). Show that the reduction Q̄(x) ∈ Fp [x] has some root in Fp at
any prime number p, while Q(x) has no rational root.
6. Show that the congruence Q(x) = (2x − 3)(3x − 2) ≡ 0 (mod. n) has integer solution for any natural
number n ≥ 2, while Q(x) has no integer root. (Hint: The chinese remainder theorem).
7. Show that the number 17 is a quadratic residue modulo 2n for any exponent n ≥ 1.
Let p be an odd prime. If an integer a is a quadratic residue modulo p, prove that a is a quadratic
residue modulo pn for any exponent n ≥ 1.
Show that Q(x) = (x2 − 17)(x2 + 1)(x2 + 17) ≡ 0 (mod. n) has integer solution for any natural number
n ≥ 2, while Q(x) has no rational root.
8. Prove that any ideal of a ring A × B is a × b, where a is an ideal of A and b is an ideal of B. Moreover,
we have a ring isomorphism (A × B)/(a × b) = (A/a) × (B/b), [(a, b)] 7→ ([a], [b]).
9. Prove that any ideal a 6= 0 of the ring k[[t]] of formal power series with coefficients in a field k is
a = (tn ) for some natural number n.
10. If a ring A is not a field, show that the ring A[x] is not a principal ideal domain. (Hint: Consider the
ideal (a, x), where a is not a unit).
11. Let A be a domain. If a map δ 0 : A − {0} → N satisfies condition 1 of the definition of Euclidean ring
(p. 73) and we put δ(a) = min δ 0 (af ), f ∈ A − {0}, prove that (A, δ) is a Euclidean ring. (Hint: Put
b = af c + r with δ(r) < δ(af )).
12. If (A, δ) is a Euclidean ring, prove that a ∈ A − {0} is invertible if and only if δ(a) = δ(1).
Show that the invertible gaussian integers are ±1 and ±i.
13. Prove that the following conditions on a prime number p are equivalent:
ALGEBRA I 475
(a) p is not a sum of two perfect squares.

(b) p is irreducible in the ring Z[i].
(c) p ≡ 3 (mod. 4). (Hint: We have a ring isomorphism Z[i]/pZ[i] ' Fp [x]/(x2 + 1)).
14. Prove that a + bi, b 6= 0, is irreducible in Z[i] if and only if a2 + b2 is a prime number.
15. Let n = pn1 1 . . . pnr r , where p1 , . . . , pr are different prime numbers. Prove that n is a sum of two perfect
squares if and only if the exponents ni are even whenever pi ≡ 3 (mod. 4).
16. Prove that no prime number p admits two decompositions as a sum of two perfect squares. That is
to say, if p = a2 + b2 = c2 + d2 , then a2 = c2 or a2 = d2 .
17. If p is an odd prime factor of n2 + 1, show that p ≡ 1 (mod. 4).
Conclude the existence of infinite prime numbers of the form 4m + 1.
18. Let n, m ∈ N be coprime. Prove that any prime dividing n2 + m2 is a sum of two perfect squares.
√ √ √
19. If d ≥ 3 is odd, show that Z[ −d ] is not a principal ideal domain. (Hint: (1 + −d )(1 − −d ) is
even).
2πi √
20. Show that Z e 3 and Z[ −2 ] are Euclidean rings.
21. If A is a Euclidean ring, prove the existence of a maximal ideal m such that any non-zero element of
the residue field A/m is represented by a unit. (Hint: Consider the maximal ideal generated by an
irreducible element p of minimal norm δ(p)).
√
22. Put A = Z 1+ 2−19 and show that A∗ = {±1}. Conclude that A is not a Euclidean ring. (Hint: We

have A ' Z[x]/(x2 − x + 5), so that A/2A and A/3A are fields of 4 and 9 elements).
23. Put A = R[x, y]/(x2 + y 2 + 1) and show that A∗ = R − {0}. Conclude that A is not a Euclidean ring.
(Hint: A = R[x] + R[x]y, and if p(x) + q(x)y is a unit, then (p + qy)(p − qy) = p2 + q 2 (x2 + 1) is
invertible in R[x]; hence it is constant).
24. Express 4a2 b2 − (a2 + b2 − c2 )2 as a polynomial in the elementary symmetric function in a, b, c. (Hint:
See Heron’s formula, p. 473).
25. Prove that xd1 + . . . + xdn = P (s1 , . . . , sd−1 ) − (−1)d dsd for some polynomial P , where si is the i-th
elementary symmetric function in x1 , . . . , xn . (Hint: Consider the roots of xd − 1).
√ √
26. If a, b ∈ Q, prove that Q( a ) = Q( b ) if and only if b/a is a square in Q.
27. Let us consider an intermediate field k ⊂ L ⊆ k(x). Show that the degree [k(x) : L] is finite.

28. If α ∈ C is a root of a unitary irreducible polynomial P (x) ∈ Z[x], prove that Z[α] ' Z[x]/ P (x) .
29. Prove that any finite extension of R is isomorphic to R or C.
30. Let k → L be an extension. If α1 , . . . , αn ∈ L are algebraic over k, prove the rationalization of
algebraic expressions: k(α1 , . . . , αn ) = k[α1 , . . . , αn ].
31. Prove the following properties of the cyclotomic polynomials:
Φ2n (x) = Φn (−x) when n is odd.

pn−1

Φpn (x) = Φp x when p is prime.
n
Φr x p
Φpn r (x) = when r is not a multiple of the prime p.
Φr xpn−1
2πi
√
32. Prove that the real part of e 5 is 5−1
4 . Give a ruler-and-compass construction of a regular pentagon,
and of a regular 15-sided polygon, inscribed in a given circle.
33. If P (x) ∈ Q[x] has degree d and P (n) ∈ Z for any integer n 0, prove that
P (n) = md nd + . . . + m2 n2 + m1 n1 + m0 ,

where m0 , . . . , md ∈ Z. Conclude that P (n) ∈ Z for all n ∈ Z. (Hint: ∆ nd = n

d−1 ).
476 EXERCISES
34. Define e∇ (s) and e∆ (s) rigorously when s = (n3 ), and give sense to the following calculation:
X n3
∇ ∇2 ∇n

s + . . . = e∇ s 0

= s+ s+ s + ... +
n n! 1! 2! n! 0
2
∆n

∆+1
∆
∆ ∆
= e s 0 =e e s 0 =e s+ s+ s + ... + s + ...
1! 2! n! 0
∆2 ∆3

∆
=e s+ s+ s+ s = e(0 + 1 + 3 + 1) = 5e .
1! 2! 3! 0
35. Give sense and rigor to the following calculation:
e2t
Z
1 2t 3 1 1
e2t t3 dt = e t = e2t t3 = t3
D D+2 2 1 + 12 D
e2t D D2 D3
3
3t2

t 3t 3
= 1− + − + . . . t3 = e2t − + − ·
2 2 4 8 2 4 4 8
00
P bP
36. Prove that 0 ≤ Eab+ E a for any polynomial P ∈ R[x].
P0 P0
Number of roots of P
Obtain Budan-Fourier theorem: ≤ Vab (P, P 0 , P 00 , . . . , P (n ).
between a and b
37. Let k = Fp (t). Show that xp − t is irreducible in k[x], with a unique root of multiplicity p.
38. Imagine that the gravitational force between punctual bodies in the complex plane C is proportional to
the masses and inversely proportional to the distance (and fix the mass unit so that the gravitational
constant is 1). Prove that
(a) A body at α ∈ C attracts a body at z (both of mass 1) with a force Fz = (ᾱ − z̄)−1 .
(b) If we place bodies of mass 1 at the roots α1 , . . . , αn of a polynomial P (z) with simple roots, then
the force Fz acting on a body of mass 1 at a point z is given by the formula
1 1 P 0 (z)
−F̄z = z−α1 + ... + z−αn = P (z)
so that the gravitational force vanishes just at the roots of the derivative P 0 (z).
(c) Obtain Gauss-Lucas theorem: The roots of P 0 (z) are in the convex envelope of the roots of
P (z).
(d) Prove this theorem when P (z) has multiple roots. (Hint: Assume that the mass placed at any
root αi is just the multiplicity of αi in P (z)).
39. If A is a principal ideal domain, show that so is any localization AS .
40. Prove that (AS )T = AST , where we put ST = {st : s ∈ S, t ∈ T }.
ni
41. If n ≥ 2, show that Z[ni] is not a unique factorization domain. (Hint: n is a root of x2 + 1).
2 2
42. If d ∈ Z is not a square,
√ but it is a square in Fp and the equation x − dy = ±p has no integer
solution, show that Z[ d ] is not a unique factorization domain.
43. Let Q(x) = a0 xn + . . . + an ∈ Z[x]. If a, b ∈ Z are coprime and a/b is a root of Q(x), show that
Q(x)/(bx − a) has integer coefficients, so that a − b divides Q(1) and a + b divides Q(−1).
44. If the product (xn + . . .)(xm + . . .) of two unitary polynomials with rational coefficients has integer
coefficients, show that both factors have integer coefficients.
45. Show that the discriminant of a cubic x3 − px2 + qx − r with coefficients in a field is
σ1 = p, σ2 = p2 − 2q, σ3 = p3 − 3pq + 3r, σ4 = p4 − 4p2 q + 4pr + 2q 2 .

3 σ1 σ2
∆ = σ1 σ2 σ3 = −4p3 r − 27r2 + 18pqr − 4q 3 + p2 q 2 .

σ2 σ3 σ4
ANALYSIS II 477
46. Prove that the discriminant of a quartic x4 − px3 + qx2 − rx + s = (x − α1 )(x − α2 )(x − α3 )(x − α4 )
with coefficients in a field is just the discriminant of the cubic resolvent
(y − ϑ1 )(y − ϑ2 )(y − ϑ3 ) = y 3 − qy 2 + (pr − 4s)y − (s(p2 − 4q) + r2 ),
ϑ1 = α1 α2 + α3 α4 , ϑ2 = α1 α3 + α2 α4 , ϑ3 = α1 α4 + α2 α3 .
47. Prove that some complex root of x4 − x3 + x2 + 1 is not a quadratic irrational. (Hint: the cubic
resolvent is irreducible).
48. Prove that any element of a field of rational functions k(x1 , . . . , xn ) algebraic over k is in k.
49. Let k be a field. The polynomial xn + c1 xn−1 + . . . + cn−1 x + cn with coefficients in the field of rational
functions Σ = k(c1 , . . . , cn ) is said to be the generic unitary polynomial of degree n with coefficients
in k. Prove that it has simple roots and that it is irreducible in Σ[x].
50. Prove that a cubic with real coefficients has a unique real root when the discriminant ∆ is negative,
and that it has three real roots when ∆ is positive.
51. If ∆ is the discriminant of P (x), show that P (a)2 ∆ is the discriminant of (x − a)P (x).
52. Let ∆(a1 , . . . , an ) be the discriminant of the polynomial xn + a1 xn−1 + . . . + an−1 x + an with real
∂∆ ∂∆
coefficients. Prove that n + (n − 1)a1 ∂a 2
+ . . . + an−1 ∂a n
= 0.
(Hint: The discriminant remains invariant when we replace αi by αi + t).
53. Let P (x, y) = 0, Q(x, y) = 0 be an algebraic system of two equations with rational coefficients. If it
only has a finite number of complex solutions, prove that the resultants R(y) and R̄(x) do not vanish,
so that the rational solutions may be determined in a finite number of steps.
54. Let us express the Bézout resultant as a determinant when m = n.
If r < n, then (a0 xr + . . . + ar )Q − (b0 xr + . . . + br )P has degree < n,
a0 Q − b0 P = A11 xn−1 + A12 xn−2 + . . . + A1n

(a0 x + a1 )Q − (b0 x + b1 )P = A21 xn−1 + A22 xn−2 + . . . + A2n
.................. ...........................
In k[x]/(P ) we have hQ (a0 ) = A11 xn−1 + A12 xn−2 + . . . + A1n ,

hQ (a0 x + a1 ) = A21 xn−1 + A22 xn−2 + . . . + A2n ,
...... ........................
n−1
hQ (a0 x + . . . + an−1 ) = An1 xn−1 + An2 xn−2 + . . . + Ann .
Use these equalities to conclude that

A11 A12 . . . A1n a0 0 . . . 0

A21 A22 . . . A2n a1 a0 . . . 0

...
= · |hQ | = an0 · |hQ | = Rb (P, Q).
. . . . . . . . . , .. .. . . . ..
An1 An2 . . . Ann an−1 an−2 . . . a0
4. Analysis II
Q Q
1. In a directQproduct i XQ i of topological spaces, show that the closure of a direct product i Yi of
subsets is i Ȳi . Hence, i Yi is closed if and only if Yi is closed in Xi for any index i.
2. Prove that a metrizable space X admits a countable base of open sets if and only if it is separable
(it admits a countable dense set).
3. Let X be a topological space. If we consider on X the equivalence relation x ≡ y when x̄ = ȳ, prove
that X0 = (X/ ≡) is a T0 -space, and that any continuous map X → Y into a T0 -space uniquely factors
through X0 ; i.e. HomTop (X, Y ) = HomTop (X0 , Y ).
If we put x ≡ y when there is a finite sequence x = x0 , x1 , . . . , xn = y such that xi ∈ x̄i+1 or xi+1 ∈ x̄i ,
prove that X1 = (X/ ≡) is a T1 -space, and that any continuous map X → Y into a T1 -space uniquely
factors through X1 ; i.e. HomTop (X, Y ) = HomTop (X1 , Y ).
478 EXERCISES
If we put x ≡ y when f (x) = f (y) for any continuous map from X into a separated space, prove that
XH = (X/ ≡) is a separated space, and that any continuous map X → Y into a separated space
uniquely factors through XH ; i.e. HomTop (X, Y ) = HomTop (XH , Y ).
4. Let R ⊆ X × X be an equivalence relation in a topological space. Show that X/R is separated if and
only if R is closed in X × X.
5. Given equivalence relations R, R0 on two topological spaces X, X 0 , show that we have a homeomor-
phism (X × X 0 )/(R × R0 ) = (X/R) × (X 0 /R0 ).
6. Let ∅ =
6 V ⊂ R be an open set, and put X = (R × {0, 1})/ ≡, where (x, 0) ≡ (x, 1) when x ∈ V .
Show that X is a non separated topological manifold. (Hint: (x, 0) and (x, 1) do not have disjoint
neighborhoods when x ∈ ∂V ).
7. If a T0 space X admits a countable base of open sets, prove that |X| ≤ 2ℵ0 .
8. Prove that any countable intersection of open subsets of a compact separated space is Baire.
9. Prove that the concept of differentiable map does not depend on the scalar product that we fix on the
space of free vectors E.
10. Let ϕ : U → E 0 be a differentiable map and let f be a differentiable function on U . Show that
f ϕ : U → E 0 is differentiable and (f ϕ)∗ = f (ϕ∗ ) + df ⊗ ϕ.
11. Show that any R-multilinear map E1 × . . . × En → F is a map of class C ∞ .
12. Give an alternative proof of Schwarz’s
R a+ε theorem, using
R a+ε Barrow’s rule an the differentiation rule under
the integral sign to show that a (∂1 ∂2 f )dx1 = a (∂2 ∂1 f )dx1 .
13. Use the mean value theorem to give an alternative proof of Schwarz’s theorem: Near the origin, given
δ1 , δ2 , show that there exist |χ1 | < |δ1 |, |χ2 | < |δ2 | such that

f (δ1 , δ2 ) − f (δ1 , 0) − f (0, δ2 ) − f (0, 0) = ∂1 f (χ, δ2 ) − ∂1 f (χ1 , 0) δ1 = ∂2 ∂1 f (χ1 , χ2 )δ1 δ2 .
f (δ1 , δ2 ) − f (0, δ2 ) − f (δ1 , 0) − f (0, 0) = ∂1 ∂2 f (χ01 , χ02 )δ1 δ2 , with |χ01 | < |δ1 |, |χ02 | < |δ2 |.

Hence ∂2 ∂1 f (χ1 , χ2 ) = ∂1 ∂2 f (χ01 , χ02 ), and conclude when (δ1 , δ2 ) → (0, 0).
14. If f ∈ C m (U ) and a ∈ U , show that f (n) : E× . n. . ×E → R, f (n) (v1 , . . . , vn ) = (∂v1 . . . ∂vn f )(a) is a
symmetric tensor. Moreover, if f (1) = . . . = f (m−1) = 0, prove that
(a) If f (m) (v, . . . , v) > 0 for any vector v 6= 0, then f has a local minimum at x = a.
(b) If f (m) (v, . . . , v) < 0 for any vector v 6= 0, then f has a local maximum at x = a.
(c) If f (m) (v, . . . , v) > 0 for some vector v, and f (m) (e, . . . , e) < 0 for another vector e, then f has
not an extremum at x = a.
15. Let f1 , . . . , fn be functions of class C m on an open set U of a n-dimensional affine space, and assume
that dp f1 , . . . , dp fn are linearly independent at any point p ∈ U . If these functions separate points
in U (for any two points x 6= y of U , we have fi (x) 6= fi (y) for some index i) and we consider
the map ϕ = (f1 , . . . , fn ) : U → Rn , prove that ϕ(U ) is an open set and that ϕ : U → ϕ(U ) is a
C m -diffeomorphism.
16. Prove the key lemma along a linear subvariety: Let V be an open set in Rd and let U be a tubular open
neighborhood of V × 0 in Rd × Rn (if (x, y) ∈ U , then (x, ty) ∈ U , for any 0 ≤ t ≤ 1). If f ∈ C m (U )
vanishes on V × 0, then there are functions f1 , . . . , fn ∈ C m−1 (U ) such that f = f1 y1 + . . . + fn yn .
17. Let (X, A, µ) be a measure and A1 ⊇ . . . ⊇ An ⊇ . . . a decreasing sequence of sets in A. If
T space
µ(A1 ) < ∞, prove that µ n An = lim µ(An ).
18. Let I ⊆ P(X) be a family closed under finite intersections (in particular ∅ ∈ I) such that I c is a finite
disjoint union of elements of I, for any I ∈ I. Show that the finite disjoint unions of elements of I
form an algebra. Conclude that the finite unions of disjoint (eventually unbounded) rectangles in Rn
form an algebra.
19. If A ⊂ Rn , prove the existence of a Borel set B ⊇ A such that me (A) = me (B).
ANALYSIS II 479
20. Prove that the category of measurable spaces has direct products.
Let X, Y be topological spaces and T a measurable space. If f : T → X and h : T → Y are measurable
maps, and X, Y have a countable base of open sets, show that f × h : T → X × Y is measurable.
Conclude that X × Y also is the direct product in the category of measurable spaces.
.
21. If f, h : X → [0, ∞] are measurable, prove that so is f ·h. (Hint: [0, ∞]×[0, ∞] →
− [0, ∞] is measurable).
22. Prove that the cardinal of B(Rn ) is 2ℵ0 .
23. Consider the real vector space with base the vertices e1 , e2 , e3 , e4 of a regular tetrahedron, and let Mi
be the matrix of the linear automorphism induced by the reflection on the i-th face. Show that the
coefficients of Mi are in 13 Z, and that the reduction Si modulo 3 of 3Mi vanishes on such base, except
Si (ei ) = −(e1 + . . . + ebi + . . . + e4 .
Show that Si Si = 0 and that, when ij 6= ij+1 for any index j, we have
(
(−1)k−1 Si1 when ik = i1
Sik . . . Si1 =
(−1)k Sik Si1 when ik 6= i1
(in particular Sik . . . Si1 6= 0). If R1 , . . . , R4 are the reflections on the faces of a regular tetrahedron
and a product Rik . . . Ri1 is a translation, conclude that ij = ij+1 for some index j. Hence the linear
components ri of such reflections generate a free group (with the only relations ri2 = 1) and the
rotations a = r1 r2 , b = r3 r4 generate a free group G, so that
G = Ga q Ga−1 q Gb q Gb−1 q {Id} = Ga q aGa−1 = Gb q Gb−1 ,
where Gx = {x . . .}. Now consider the natural action of G on a sphere S2 , let M be a subset of
S2 with a unique point in each orbit, and put Mx = Gx M . Conclude that we may cover a sphere
S2 = GM = Ma ∪ Ma−1 ∪ Mb ∪ Mb−1 ∪ M with 5 subsets with countable intersections, so that, after
composing with motions, we may form two spheres S2 = Ma ∪ aMa−1 = Mb ∪ bMb−1 of equal radius
(Banah-Tarski paradox).
24. Prove that any Riemann-integrable bounded function f : [a, b] → R is Lebesgue-measurable, and the
Lebesgue integral of f coincides with the Riemann integral.
25. Let C ⊂ [0, 1] be the cantor set. Show that some subset N ⊂ C is not a Borel set, and conclude that
IN is a Riemann-integrable function, but not a Borel-measurable function.
26. If a function f : Rn → R is almost everywhere continuous, prove that the set of points where f is
continuous is a Borel set.
27. Let f, g be two continuous functions with compact support in Rn . Prove the following statements
about the convolution product
Z
(f ∗ g)(x) = f (x − t)g(t)dt.
Rn
(a) f ∗ g is a continuous function in Rn and supp (f ∗ g) ⊆ (supp f ) ∪ (supp g).

(b) The set C0 (Rn ) of continuous functions with compact support, with the convolution product and
the usual sum, is a commutative ring (without unit).

∞ ∂(f ∗ g) ∂f
(c) If f is of class C , so is f ∗ g, and = ∗ g.
∂xi ∂xi
28. Let I ⊆ P(X) be a family closed under finite intersections (in particular ∅ ∈ I) such that I c is a
countable disjoint union of elements of I, for any I ∈ I. Prove that the σ-algebra generated by I
is the least family I ⊆ M closed under complements and countable disjoint unions. (Hint: Use the
argument of the monotone class theorem). If I ⊂ Rn is a rectangle, show that I c is a countable
disjoint union of rectangles. Conclude that B(Rn ) is the least family containing the rectangles and
closed under complements and countable disjoint unions.
29. Give an alternative proof of the change of variables formula as follows:
Show that the formula holds for ϕ1 ◦ ϕ2 whenever it holds for ϕ1 and ϕ2 .
Prove that the formula holds when n = 1 (Hint: Barrow’s formula shows that it holds on any finite
interval) and then proceed by induction on n:
480 EXERCISES
(a) The formula holds when ϕ(x1 , . . . , xn ) = (y1 , . . . , yn−1 , xn ). (Hint: Show that in this case Jϕ (p)
is the jacobian at p = (a1 , . . . , an ) of ϕ : U ∩ {xn = an } → V ∩ {xn = an }, and conclude using
Fubini’s theorem).
(b) In general, if yi = yi (x1 , . . . , xn ) are the equations of ϕ and we fix a point p ∈ U , show that
the jacobian at p of ϕ1 (x1 , . . . , xn ) = (y1 , . . . , yn−1 , xn ) is non zero, relabelling x1 , . . . , xn if
necessary. Hence there is an open neighborhood U 0 of p such that ϕ1 : U 0 → V 0 = ϕ1 (U 0 )
is a diffeomorphism, and ϕ2 = ϕϕ−1 0 0
1 : V → ϕ(U ) is a diffeomorphism and ϕ2 (y1 , . . . , yn ) =
y1 , . . . , yn−1 , zn (y1 , . . . , yn ) . Conclude that the formula holds for ϕ|U 0 : U 0 → ϕ(U 0 ).
S
(c) If U = i Ui is an open cover and the formula holds for ϕ|Ui for any index i, prove that the
formula holds for ϕ.
30. Consider Rn with the σ-algebra L of all measurable sets and the Lebesgue measure m, so that a
function f is L-measurable when f −1 (B) is a measurable set for any Borel set B ⊆ R, and prove
the following results:
(a) If f : Rn → [0, ∞] is L-measurable, Rn f dm = m{(x, y) ∈ Rn × R : 0 ≤ y < f (x)}.

R
(b) If N ⊂ Rn+m has null measure, then R m(Ny ) = 0 almost everywhere. (Hint: Take a Borel set
N ⊆ B with m(B) = 0, show that hB dm = m(B) = 0 and conclude that m(By ) = 0 almost
everywhere).
(c) If f : Rn+m → [0, ∞] is L-measurable, then hf (y) = Rn f (x, y)dm is L-measurable and Fubini’s
R
formula holds for f . (Hint: Take a disjoint union LR= B ∪ N , where

R m(N ) = 0 and B is a Borel
set, show that hN (y) R= m(Ny ) isRL-measurable and hN dm = 0 = IN dm. Hence hL = hB +hN
is L-measurable and hL dm = IL dm, and conclude with the monotone convergence theorem).
(d) If f : Rn+m m
R → R is L-integrable, and fy (x) = f (x, y) is L-integrable for any y ∈ R , then
hf (y) = Rn f (x, y)dm is L-integrable and Fubini’s formula holds for f .
(e) If f : V → [0, ∞] is L-measurable, then (f ◦ ϕ)|Jϕ | is L-measurable and the change of variables
formula holds for f . (Hint: If L = B ∪ N is a disjoint union of a Borel set and a set of null
measure, the formula holds for IL = IB + IN , and conclude with the monotone convergence
theorem).
(f) If f : V → R is L-integrable, then (f ◦ ϕ)|Jϕ | is L-integrable and the change of variables formula
holds for f .
31. Prove that any direct product of topological vector spaces, with the product topology, also is a topo-
logical vector space.
32. If we endow the spaces of continuous maps with the compact-open topology, prove that
(a) If f : X → Y is continuous, so are f∗ : C(T, X) → C(T, Y ) and f ∗ : C(Y, T ) → C(X, T ).

(b) C(T × X, Y ) = C T, C(X, Y ) when T is a separated space and X is a locally compact separated
space. Moreover, the natural map C(T, X) × C(X, Y ) → C(T, Y ) is continuous.

33. Let q1 , . . . , qn be seminorms on E, and p a seminorm on Rn . Show that q(e) = p q1 (e), . . . , qn (e) is
a seminorm on E, defining the topology induced by {q1 , . . . , qn } when p is a norm.
Let E1 , . . . , En be seminormable spaces (topological vector spaces such that the topology is defined
by a seminorm). Given linear maps fi : E → Ei , show that the initial topology is defined by a
seminorm. Conclude that E1 × . . . × En also is seminormable.
34. If q : E → K is a seminorm, show that |q(y) − q(x)| ≤ q(y − x), ∀x, y ∈ E. Conclude that q is
continuous, when we consider on E the linear topology defined by q.
Let q 0 : E → K be another seminorm. If q ≤ cq 0 for some constant c ∈ R, prove that q also is continuous
when we consider on E the linear topology defined by q 0 .
35. Prove that two seminorms q 0 , q on a vector space E define the same topology if and only if there exist
m, M ∈ R+ such that mq(e) ≤ q 0 (e) ≤ M q(e), ∀e ∈ E.
ALGEBRA II 481
36. Let e1 , . . . , en be an orthonormal base of a real Euclidean space E. If p is a seminorm on E and we

put c = max{p(e1 ), . . . , p(en )}, show that p(e) ≤ nckek, ∀e ∈ E.
When p is a norm, prove that p is bounded below on the unit sphere {e ∈ E : kek = 1} by a constant
m > 0, so that mkek ≤ p(e), ∀e ∈ E. Conclude that, in a finite dimensional real vector space, all the
norms define the same linear topology.
37. Prove that the product C m (U ) × C m (U ) → C m (U ), 0 ≤ m ≤ ∞, is continuous.
5. Algebra II
1. Show that the actions of a group G on a given set with n elements correspond to the group morphisms
G → Sn . Conclude that any finite group G of order n is isomorphic to a subgroup of Sn .
2. Show that G = P Gl(2, F5 ) is a subgroup of S6 of index 6. Obtain an automorphism τ : S6 → S6 such
that τ (G) = {σ ∈ S6 : σ(1) = 1}, and conclude that τ is not an inner automorphism of S6 . (Hint:
Consider the action of S6 on the set of conjugate subgroups of G).
3. If p is the least prime number dividing the order of a finite group G, prove that any subgroup H of G
of index p is a normal subgroup. (Hint: G acts on G/H).
4. If p is a prime number, prove that any group of order p2 is abelian, and that any non-trivial normal
subgroup of a non abelian group G of order p3 contains the center Z(G).
5. If H is a normal subgroup of a p-group G, prove that H ∩ Z(G) 6= 1.
6. If a finite group G has a unique Sylow p-subgroup for any prime p, prove that G is solvable.
7. If n 6= 4, prove that An is a simple group (any normal subgroup is trivial). (Hint: If H is a non-trivial
normal subgroup of An , it is not a normal subgroup of Sn , p. 469, so that it has a unique conjugate
subgroup H 0 = τ Hτ −1 , where σ is any odd permutation. Now H 0 ∩ H = Id and H 0 H = An , because
they are normal subgroup of Sn , so that An = H 0 × H. Hence |H| is even, and H contains a product
σ = τ1 . . . τr of disjoint transpositions, and we get a contradiction σ = τ1 στ −1 ∈ H 0 ).
8. If n ≥ 5, prove that An is the unique non trivial subgroup of Sn of index < n. Moreover, An has no
non-trivial subgroup of index < n.
9. If H is a non-trivial subgroup of a simple group G, prove that [G : H] ≥ 5.
Let n be the number of Sylow p-subgroups of a simple group G 6= Z/pZ. Prove that n ≥ 5.
10. Show that the A-module structures on a given abelian group M correspond to the ring morphisms
A → End(M ).
11. If M is an A-module, show that we have a natural isomorphism HomA (A, M ) = M , f 7→ f (1).
P
12. If M0 ⊂ M1 ⊂ . . . ⊂ Mn is a sequence of submodules, show that l(Mn /M0 ) = i l(Mi /Mi−1 ).
13. If 0 → M0 → M1 → . . . → Mn → 0 is an exact sequence of A-modules of finite length, show that
l(M0 ) − l(M1 ) + . . . + (−1)n l(Mn ) = 0.
14. Let 0 = M0 ⊂ M1 ⊂ . . . ⊂ Mn = M be a flag of an A-module M of finite length. Given a maximal
ideal m of A, prove that the number of indices such that Mi /Mi−1 ' A/m does not depend on the
flag. (Hint: Use induction on n, considering a quotient M/N by a simple submodule).
15. Show that any abelian group of finite length is a finite group.
16. If Σ is the field of fractions of a domain A, show that Σ and Σ/A are divisible A-modules.
17. Let k be a field and let A = k[x]/(p(x)), where p(x) 6= 0. Show the existence of an isomorphism of
A-modules A ' A∗ = Homk (A, k), and conclude that A is an injective A-module.
Moreover, any projective A-module is an injective A-module.
18. Prove the universal property of the polynomial ring A[x1 , . . . , xn ]:

HomA-alg (A[x1 , . . . , xn ], B) = B n , φ 7→ φ(x1 ), . . . , φ(xn ) .
482 EXERCISES

19. Prove that A[x1 , . . . , xn ]/(p1 , . . . , pr ) ⊗A B = B[x1 , . . . , xn ]/(p1 , . . . , pr ).
20. If E is a finite-dimensional k-vector space, show that there is a canonical isomorphism
Tpq E = E ∗ ⊗k . p. . ⊗k E ∗ ⊗k E⊗k . q. . ⊗k E.
21. If E, F are finite-dimensional k-vector spaces, show that Homk (E, F ) = E ∗ ⊗k F .
22. If E, F are finite-dimensional k-vector spaces, prove that dim (E ⊗k F ) = (dim E)(dim F ), and that
dimk E = dimK (E ⊗k K) for any extension k → K.

23. Prove that Z[x]/ Q(x) ⊗Z Fp = Fp [x]/ Q̄(x) , where Q̄(x) is the reduction of Q(x) modulo p.
24. Show that there is an isomorphism of C-algebras C ⊗R C = C ⊕ C, z1 ⊗ z2 7→ (z1 z2 , z̄1 z2 ).
25. If d = g.c.d.(m, n), show that (Z/nZ) ⊗Z (Z/mZ) → Z/dZ, [a] ⊗ [b] 7→ [ab], is a ring isomorphism.
26. Let a, b be ideals of a ring A and let M be an A-module. If a + b = A, prove that we have an
isomorphism of A-modules M/abM = (M/aM ) ⊕ (M/bM ), [m] 7→ ([m], [m]).
27. Let A, B 6= 0 be algebras over a field k. Prove that A → A ⊗k B, a 7→ a ⊗ 1, is injective.
28. Show that an extension k → L is trivial if and only if L ⊗k L is a field. (Hint: The product defines a
morphism of k-algebras L ⊗k L → L).
S T
29. If {Mi }i∈I is a chain of submodules of a module, show that lim
−→
Mi = i Mi and lim←−
Mi = i Mi .
30. Prove that any A-module is an inductive limit of finitely generated A-modules, and that any ring is
an inductive limit of finitely generated Z-algebras.
31. Prove that the inductive limit of an inductive system {Mi , φij } of A-modules is isomorphic to the
quotient of ⊕i Mi by the submodule N generated by the elements mi − φij (mi ), endowed with the
natural morphisms φi : Mi → (⊕i Mi )/N .
32. Let M be an A-module and f ∈ A. If we consider the inductive system of A-modules {Mn }n∈N , where
Mn = M and φnn+1 (m) = f m for all n ∈ N, show that Mf = lim
−→
Mn .
33. If M is a module over a domain A, show that Mf = lim

−→
HomA (f n A, M ).
34. Let {Xi , φij }, {Yi , ϕij } be inductive (resp. projective) systems of sets with the same index set. If we
have maps fi : Xi → Yi commuting with the transition maps, fj φij = ϕij fi , prove the existence of a
unique map f : lim −→
Xi → lim−→
Yi (resp. f : lim
←−
Xi → lim
←−
Yi ) such that f φi = ϕi fi (resp. ϕi f = f φi ).
35. Let {Mi0 }, {Mi }, {Mi00 } be inductive systems of A-modules, with the same index set. If we have exact
sequences 0 → Mi0 → Mi → Mi00 → 0, where the morphisms commute with the transition morphisms,
prove that they induce an exact sequence 0 → lim −→
Mi0 → lim
−→
Mi → lim
−→
Mi00 → 0.
36. Let {Mi0 }, {Mi }, {Mi00 } be projective systems of A-modules, with the same index set. If we have exact
sequences 0 → Mi0 → Mi toMi00 , where the morphisms commute with the transition morphisms, prove
that they induce an exact sequence 0 → lim ←−
Mi0 → lim
←−
Mi → lim
←−
Mi00 .
However, show that the natural morphism Z = lim
←−
Z → lim
←−
(Z/2n Z) is not surjective.
1
37. Show that 1 + 999 . . . = 0 and 1−10 = 1 + 10 + 102 + . . . = 111 . . . in the ring lim
←−
(Z/10n Z).
38. If M is an A-module and S ⊂ A is a multiplicative system, show that MS = lim

−→
Mf .
f ∈S
39. Prove that lim

−→
(Mi ⊕ Ni ) = (lim
−→
Mi ) ⊕ (lim
−→
Ni ) and lim
←−
(Mi × Ni ) = (lim
←−
Mi ) × (lim
←−
Ni ).
Moreover, prove that inductive limits commute with arbitrary direct sums, and that projective limits
commute with arbitrary direct products.
40. Given an inductive system of rings {Ai }i∈I , show that rad(lim
−→
Ai ) = lim
−→
(rad Ai ).
41. Prove that any flat module P over a domain A is torsion free.
ALGEBRA II 483
42. Prove that any inductive limit of flat A-modules is a flat A-module.
43. Prove that a direct sum ⊕i Mi of A-modules is flat if and only if so are the A-modules Mi .
44. If M, N are flat A-modules, prove that so is M ⊗A N .
If B is a flat A-algebra and M is a flat B-module, prove that M is a flat A-module.
45. If M is a finitely generated (resp. flat, projective, free) A-module, prove that MB is a finitely generated
(resp. flat, projective, free) B-module for any base change A → B
46. Where does the following argument fail? Let k be a field. If a k-algebra A admits a morphism of
k-algebras A → k, then any A-module M is free, because M = M ⊗A A = (M ⊗A k) ⊗k A, and M ⊗A k
is a free k-module, so that (M ⊗A k) ⊗k A is a free A-module.
47. Given any object M of a category, prove that the composition of morphisms defines a group structure
on the set of automorphisms of M .
48. Show that a group G may be viewed as a category C with a unique object where any morphism
is an isomorphism, and prove that the opposite category Cop is defined by the opposite group Gop .
Moreover, if C0 is the category defined by another group G0 , show that the covariant functors C C0
0
are just the group morphisms G → G .
49. Prove that the construction of the dual space E E ∗ defines an equivalence of the category of
finite-dimensional k-vector spaces and linear maps with the opposite category.
50. Let F : C C0 be a covariant functor such that the maps F : HomC (M, N ) → HomC0 (F (M ), F (N ))
are bijective and any object C0 is isomorphic to F (M ) for some object M of C. If the objects of C0
form a set (so that for any object M 0 of C0 we may choose an object G(M 0 ) of C and an isomorphism
F (G(M 0 )) ' M 0 ) prove that F defines an equivalence of categories.
51. Let D be the discrete (identities are the only morphisms) category of pointed sets (X, x) and let us
consider the covariant functors S, p : D Sets such that S(X, x) = X and p(X, x) = {x}. Show that
for any set X we have an injective map X → Homnat (p, S), so that the functorial morphisms p → S
do not form a set. (In general, the ”category” Hom(C, C0 ) of covariant functors of a category C into
a category C0 is not a category, unless you impose some conditions guaranteeing that Homnat (F, G)
is a set for any two functors F, G : C C0 .)
52. Let U → X, V → X be two subspaces of a topological space X. Show that U ×X V = U ∩ V .
53. Prove that a topological space X is connected if and only if the functor Hom(X, −) preserves coprod-
ucts (so explaining why connected spaces are assumed to be non empty):
Hom(X, Y q Z) = Hom(X, Z) q Hom(X, Z).
54. Let k be a field and fix a natural number n. Let C be the category of n-dimensional k-vector spaces
and linear isomorphisms, and let us view the dual E ∗ as a covariant functor F : C C, F (E) = E ∗ ,
−1 ∗
F (τ ) = (τ ) . If n ≥ 2 (or k has more than 3 elements) prove that there is no natural isomorphism
E → E ∗ defined on the n-dimensional k-vector spaces; i.e., F is not isomorphic to the identity functor.
(Hint: A natural isomorphism E → E ∗ defines a non-singular intrinsic covariant tensor T2 on E).
If n = 1 and k = F2 or F3 , show the existence of a natural isomorphism E ' E ∗ .
55. Using Yoneda’s lemma, prove that HomC (X, Y ) = Homnat (X • , Y • ), for any objects X, Y of C, and
answer the following questions:
(a) What natural maps G → G may be defined on groups? (What are the natural transformations
from the forgetful functor Groups Sets into itself?)
What natural morphisms G → G may be defined on groups? (What are the natural transforma-
tions from the identity functor Groups Groups into itself?)
What natural maps G → G × G may be defined on groups?
(b) Let k be a field. Determine the natural maps A → A × A and operations A × A → A that may
be defined on k-algebras. Obtain the natural ring morphisms A → A that may be defined on
k-algebras, when char k = 0 and when char k = p > 0.
484 EXERCISES
(c) Find the natural maps X → X × X and operations X × X → X on topological spaces.

(d) Find the natural maps M → M × M and operations M × M → M on A-modules.
Q
56. In the category of sets, determine the direct product i∈I Xi when Xi = ∅ for any index i (resp.
when I = ∅, and when Xi = I = ∅). (Hint: The direct product of a family of sets {Xi }i∈I such that
Xi = X for any index i is Hom(I, X) = X I .)
Given a set S, determine the fibred product over S of a family of maps {fi : Xi → S}i∈I when Xi = ∅
for any index i (resp. when I = ∅, when Xi = I = ∅, and when Xi = I = S = ∅).
57. Show that the covariant functor ∅ : A-mod Sets preserves kernels, direct products and projective
limits (with non-empty set of indices), but not the final object, so that it is not representable.
58. Let WOrd be the category of well-orders, the morphisms X → Y being the isomorphisms onto an
initial ray, so that Hom(X, Y ) is a one-point set when X ≤ Y and Hom(X, Y ) = ∅ otherwise. Prove
that WOrd has not a final object, so that the constant contravariant functor F : WOrd Sets is
not representable, even if it is left exact and transforms inductive limits into projective limits. Show
that the minimal pairs of F do not form a set.
59. Let I be a category (its objects are named indices or vertices, and its morphisms are named arrows).
An I-diagram in a category C is a covariant functor I C; i.e. an object Xi for any index i ∈ I and a
morphism f˜: Xi → Xj for any arrow f from the vertex i to j, preserving compositions and identities.
The index category I is viewed as the generic diagram with indices in I, given by the identity functor
I I. A diagram is small (resp. finite) when the indices and arrows form a set (resp. a finite set).
The limit of an I-diagram is an object limi∈I Xi endowed with compatible morphisms πi : lim Xi →
Xi , in the sense that πj = f˜πi for any arrow f from the vertex i to j, and such that for any compatible
family (pi : T → Xi )i∈I there is a unique morphism p : T → lim Xi such that pi = πi p, ∀i ∈ I. It is the
universal left cone. The colimit colimi∈I Xi is the universal right cone, i.e. a universal compatible
family of morphisms (λi : Xi → colim Xi )i∈I , in the sense that λi = λj f˜.
(a) If the limit or the colimit of a diagram exists, prove that it is unique up to a canonical isomorphism.
{(xi ) ∈ i Xi : f˜(xi ) = xj for any arrow f }, and the
Q
(b) The limit of a small diagram in Sets is just `
colimit is the quotient of the disjoint union i Xi by the smallest equivalence relation such that
xi ≡ f˜(xi ) for any xi ∈ Xi and any arrow f with origin at the i-th vertex. Hence, for any small
diagram in an arbitrary category, we have
HomC (T, lim Xi ) = lim HomC (T, Xi ) , HomC (colim Xi , T ) = lim HomC (Xi , T ).
(c) Show that direct products (hence the final object), kernels, fibred products and projective limits
are limits, while coproducts (hence the initial object), cokernels and inductive limits are colimits.
(d) Given a finite diagram (X1 , . . . , Xn ) with morphisms (φ1 : Xi1 → Xj1 , . . . , φm : Xim → Xjm ),
show that the limit is just the kernel of the morphisms
Q Q
f, h : i Xi ⇒ k Xjk , f (xi ) = (φ1 (xi1 ), . . . , φm (xim )), h(xi ) = (xj1 , . . . , xjm ).
Moreover, the kernel of f, h : X ⇒ Y is just the fibred product of the graphics 1 × f : X → X × Y

y 1 × g : X → X × Y , and the fibred product of f : X → S, h : Y → S is just the kernel of the
morphisms f π1 , hπ2 : X × Y → S.
Conclude that a category has finite limits if and only if it has finite products and kernels (or
fibred products), and that a covariant functor preserves finite limits if and only if it is left exact
(preserves finite products and kernels).
Analogously,Qwhen I is Qa small category, the limit of a diagram (Xi )i∈I is the kernel of a pair of
morphisms i Xi ⇒ φ Xj , where φ runs over all the morphisms φ : Xi → Xj in the diagram.
Hence a category has small limits if and only it has products and kernels (or finite products,
projective limits and kernels, or products and fibred products, or...), and a covariant functor
preserves small limits if and only if it is left exact and preserves products (or projective limits).
State the dual results.
ALGEBRA II 485
60. Prove that the limit of all the objects and morphisms of a category C, if it exists, is just the initial
object4 : lim X = ∅.
X∈C
(Hint: Consider the morphisms πX : lim X → X, so that πX = f πY for any morphism f : Y → X;
hence πX = πX πlim X , and πlim X = Idlim X by the uniqueness of the factorization through lim X.
Conclude that any morphism f : lim X → X is πX = f πlim X = f .)
Conversely, if C has an initial object ∅, prove that ∅ = limX∈C X.
(Hint: Clearly the family {iX : ∅ → X}X∈C is compatible, and we have pX = iX p∅ for any compatible
family {pX : T → X}X∈C . If f : T → ∅ is another factorization, then p∅ = i∅ f = f .)
61. If I has a final index ∞, show that the colimit of any I-diagram is just X∞ = colimi∈I Xi .
62. Let Ω ⊂ I be a subcategory such that any vertex Xi admits some morphism g : Uα → Xi with Uα ∈ Ω,
and assume that for any other morphism h : Uβ → Xi , Uβ ∈ Ω, we have a commutative square (with
φ and ϕ in Ω)
Uγ
φ
/ Uα
ϕ g

Uβ
h / Xi
Prove that limi∈I Xi = limα∈Ω Xα for any I-diagram.
Hint: Any Ω-compatible family (pα : T → Uα ) admits a unique extension to a I-compatible family
(pi : T → Xi )i∈I , since pi must be the composition of pα : T → Uα with g̃ : Uα → Xi . The above
square shows that it does not depend on g and that the family (pi )i∈I is I-compatible, since for any
two arrows f : Xi → Xj , h : Uβ → Uj , replacing g by f g we have a commutative diagram:
6 Uα
g̃
/ Xi
φ̃
Uγ f˜
(
ϕ̃
Uβ
h̃ / Xj
Conclude that if a category C has small limits and there is a set of objects {Ui }i∈I such that any
object X admits some morphism Ui → X, then C has an initial object.
63. Let C be a category with small limits and F : C Sets a covariant functor preserving small limits.
(a) Show that the category of pairs also has small limits. (Hint: The limit of a small diagram of
pairs (Xi )ξi is just (lim Xi )(ξi ) , where (ξi ) ∈ lim F (Xi ) = F (lim Xi ).)
(b) Prove a Representability Theorem: If there is a set of pairs {(Ui )αi }i∈I such that any other
pair Xξ admits some morphism of pairs (Ui )αi → Xξ , then F is representable.
(c) Prove that F (Y ) → F (X) is injective when Y ,→ X is a monomorphism5 (Hint: If α, β ∈ F (Y )
have equal image ξ ∈ F (X), consider a left cone of the diagram of pairs Yα → Xξ ← Yβ .)
Let Xξ be a pair, assume that
T T the subobjects Yi ,→ X such that ξ ∈ T F (Yi ) form a set,
T and put
Y
i i := lim Yi . Show that Y
i i is a subobject of X and ξ ∈ F ( Y
i i ), so that ( i Yi )ξ is a
minimal pair dominating Xξ .
(d) Prove another Representability Theorem: If the subobjects of any object form a set, and the
minimal pairs of F form a set, then F is representable.
64. If m1 , . . . , mr are maximal ideals of a ring A, prove that A/(mn1 1 . . . mnr r ) = (A/mn1 1 ) ⊕ . . . ⊕ (A/mnr r ).
65. If a prime ideal p of a ring A is contained in a finite union of prime ideals, p ⊆ p1 ∪ . . . ∪ pn , show that
p ⊆ pi for some index i. (Hint: If the closures of some points x, x1 , . . . , xn ∈ Spec A are not incident,
show the existence of a function fi ∈ A vanishing at all these points, except for xi . Conclude the
existence of a function f ∈ A such that f (x) = 0 and f (xi ) 6= 0 for any index i).
4
This formula states that the points T → ∅ are just the compatible families of points {pX : T → X}X∈C !
5
In the sense that Y • (T ) → X • (T ) is injective for any object T , and we say that two monomorphisms define
the same subobject of X when both define the same subfunctor of X • . Dually we define epimorphisms and
quotients of X.
486 EXERCISES
66. If a finite group G acts on a ring A, prove that the fibres of the natural map Spec A → Spec AG are just
G G
the orbits of the action of G on Spec S A ∩p = A ∩q
Q A. (Hint: If p, q are prime ideals in A such that
and we take f ∈ p, then N (f ) = τ ∈G τ (f ) ∈ AG ∩ p ⊆ q. Conclude that p ⊆ τ ∈G τ (q) and apply
the above problem).
67. Show that Spec Ax is the intersection of all the neighborhoods of x in Spec A.
68. Let f, h ∈ A. Prove that Uf ⊆ Uh if and only if h divides some power f n .
Conclude that Uf = Uh if and only if there is an isomorphism of A-algebras Af ' Ah .
69. Prove that any element of a minimal prime ideal of a ring is a zero divisor.
70. If there is a function f ∈ A such that f 2 = f and f 6= 0, 1, show that Spec A is not connected.
71. Let I be an ideal of a ring A. Prove that 1 + I is a multiplicative system such that Spec A1+I is the
intersection of all open neighborhoods of (I)0 .
72. Prove that Spec (lim
−→
Ai ) = lim
←−
(Spec Ai ).
73. If M is an A-module and S ⊂ A a multiplicative system, show that (AnnA (M ))S = AnnAS (MS ).
74. Prove that a ring A is reduced if and only if so is Ax at any point x ∈ Spec A.
75. Let (O, m) be a local k-algebra. If M is an O-module, show that dimk M = l(M ) · [O/m : k].
76. Let M, N be finitely generated modules over a local ring O.
If M ⊗O N = 0, prove that M = 0 or N = 0.
If M ⊗O N = O, prove that M = N = O. (Hint: M and N are monogenous modules).
77. If M, N are finitely generated A-modules, show that supp (M ⊗A N ) = (supp M ) ∩ (supp N ).
78. If M is a finitely generated A-module, show that supp (M/IM ) = (I + Ann M )0 for any ideal I.
79. Let φ : Spec B → Spec A be the continuous map induced by a ring morphism A → B.

If J ⊂ B is an ideal, prove that the closure of φ (J)0 is just (A ∩ J)0 .
If M is a finitely generated A-module, show that supp (MB ) = φ−1 (supp M ).
80. Let I be an ideal of a ring A and x ∈ Spec A. Prove that Ix = Ax if and only if x ∈
/ (I)0 .
81. Prove that any closed open set U ⊆ Spec A is a basic open set: U = Uf for some function f ∈ A.
Moreover, (Af )x = Ax when x ∈ U , and (Af )x = 0 otherwise. (Hint: Show that U = (I)0 and
U c = (J)0 , where I + J = A, so that U ⊆ (h)0 and U c ⊆ (f )0 , where f + h = 1).
82. Let N 0 , N be submodules of an A-module M . Prove that N 0 ⊆ N if and only if Nx0 ⊆ Nx , ∀x ∈ Spec A.
T
83. If A is a domain, prove that in the field of fractions we have A = x Ax , x ∈ Spec A.
84. Prove that any A-linear epimorphism An → An is an isomorphism.
85. Let M, N be two A-modules. If M is finitely generated and M ' M ⊕ N , prove that N = 0.
86. Prove that an A-module M is flat if and only if Mx is a flat Ax -module at any point x ∈ Spec A.
87. Let k̄ be an algebraic closure of a field k. If k → K is an algebraic extension, prove the existence of
an injective k-morphism K → k̄.
88. Let k[ε] = k[t]/(t2 ). Show that to give a morphism of k-algebras A → k[ε] is just to give a rational
point x ∈ Spec A and a k-derivation D : A → A/mx = k.

89. Let A, B be two k-algebras. Prove that ΩA⊗k B/k = ΩA/k ⊗k B ⊕ A ⊗k ΩB/k .
90. Let p(x) ∈ A[x]. If we put B = A[x]/(p(x)), show that ΩB/A = B/p0 (x)B dx.

91. If A = k[ξ1 , . . . , ξn ] is a finitely generated k-algebra, show that ΩA/k = A dξ1 + . . . + A dξn .
92. If A = lim
−→
Ai , prove that ΩA/k = lim
−→
ΩAi /k .
ALGEBRA II 487
93. Let B → C a morphism of A-algebras. Prove that 0 → ΩB/A ⊗B C → ΩC/A → ΩC/B → 0 is a split
exact sequence if and only if any A-derivation B → M into a C-module M may be extended to an
A-derivation C → M .
94. Let A be a ring and let M be an A[x]-module. If D : A → M is a derivation, show that for any element
m ∈ M there is a unique derivation D̄ : A[x] → M extending D such that D̄(x) = m.
95. Let k → k(α) be an extension and let D : k → E be a derivation into a k(α)-vector space E.
If α is transcendent over k, show that for any vector e ∈ E there is a unique derivation D̄ : k(α) → E
extending D such that D̄α = e.
If α is algebraic over k and p(x) = i ai xi ∈ k[x] is the irreducible polynomial of α over k, prove the
P
following statements:
(a) If p0 (x) 6= 0, there is a unique derivation D̄ : k(α) → E extending D.
(b) If p0 (x) = 0 and i (Dai )αi 6= 0, there is no derivation D̄ : k(α) → E extending D.
P
(c) If p0 (x) = 0 and i

P
i (Dai )α = 0, for any e ∈ E there is a unique derivation D̄ : k(α) → E
extending D such that D̄α = e.
96. Let k → L be an extension. If char k = 0, prove that α ∈ L is transcendent over k if and only if there
is a k-derivation D : L → L such that Dα 6= 0.
97. If a ring A has a finite discrete spectrum Spec A = {x1 , . . . , xn }, prove that the natural map A →
Ax1 ⊕ . . . ⊕ Axn is an isomorphism.
98. If A is a finite k-algebra, prove that the following conditions are equivalent:
(a) A is reduced, rad A = 0.
(b) Any A-module is injective.
(c) Any A-module is projective.
(d) Any A-module is flat.
99. Prove that a finite k-algebra A is separable if and only if A is a projective A ⊗k A-module (with the
structure defined by the natural morphism µ : A ⊗k A → A, µ(a ⊗ b) = ab).
100. Prove that two finite extensions k → L1 and L1 → L2 are separable if and only if so is k → L2 .
If k → L1 and k → L2 are separable finite extensions, show that so is any composite k → L1 L2 .
If k → L is a separable finite extension, prove that so is any composite K → KL with an arbitrary
extension k → K.
101. Let Q → K be a finite extension of degree d, and let us consider a separable polynomial P ∈ K[x]. Show
that A = Q[a] = Q[y]/(Q) for almost all a ∈ A = K[x]/(P ). If Q = Q1 . . . Qr is the irreducible factor
decomposition in Q[y], prove that P = P1 . . . Pr , where Pi is irreducible in K[x] and d(deg Pi ) = deg Qi .
Moreover, Pi (x) is the annihilator of y, viewed as a K-endomorphism of Q[y]/(Qi ). In particular, P (x)
is irreducible in K[x] if and only if so is Q(y) in Q[y], and the irreducible factor decomposition in K[x]
may be reduced to the case K = Q, solved in p. 71.
102. If k → L is a separable finite extension, prove that the following conditions are equivalent:
(a) L is the decomposition field of some polynomial with coefficients in k.
(b) Any composite of L with itself is isomorphic to L.
(c) L is a Galois extension: L ⊗k L = ⊕L.
(d) Any polynomial p(x) ∈ k[x] with a root in L has all the roots in L.
103. If k → L is a Galois extension, show that any composite K → KL is a Galois extension, where K is
an arbitrary extension of k.
If k → L0 also is a Galois extension, prove that L0 ⊗k L ' ⊕L00 for some Galois extension L00 of k.
104. If a k-algebra A is trivial over an extension k → L and Homk-alg (A, L) = {p1 , . . . , pn }, show that we
have an isomorphism of L-algebras A ⊗k L = L⊕ . n. . ⊕L, a ⊗ λ 7→ λp1 (a), . . . , λpn (a) .
488 EXERCISES
105. Consider the functors F (A) = Spec AL and R(∆) = HomG (∆, L) involved in Galois theorem. Prove
that L = RF (L) and G = F R(G), and that both F and R preserve coproducts and cokernels. (Hint:
R is representable, and F is just the composition of a base change A AL with an equivalence of
categories, see p. 143).
Show that any finite G-set ∆ admits a presentation qi G ⇒ qj G → ∆ (Hint: If H is a subgroup of G,
we have an exact sequence G × H = qH G ⇒ G → G/H, where (g, h) 7→ g and (g, h) 7→ gh), and that
any finite k-algebra A trivial over L is a kernel A → A ⊗k L ⇒ A ⊗k L ⊗k L of morphisms between
trivial L-algebras (Hint: The sequence k → L ⇒ L ⊗k L is exact).
Finally, show that ∆ = F R(∆) and A = RF (A), (so obtaining a new proof of Galois theorem) using
the following commutative diagrams with exact rows
qi G ⇒ qj G → ∆ A → ⊕j L ⇒ ⊕i L
k k ↓ ↓ k k
F R(⊕i G) ⇒ F R(⊕j G) → F R(∆) RF (A) → RF (⊕j L) ⇒ RF (⊕i L)
106. Prove Grothendieck’s characterization of the forgetful functor G-Setsf Setsf on the category
G-Setsf of finite G-sets: If a covariant functor F : C Setsf is representable, F (Y ) = Hom(X, Y ),
then the finite group G := Aut(F ) = Aut(X)op acts on F (Y ), τ f := f ◦ τ , and the induced functor
C G-Setsf is an equivalence of categories when the following conditions hold:
(a) The category C admits finite coproducts and cokernels and F preserves them.
(b) The functor F is conservative (a morphism f is an isomorphism whenever so is F (f ))
(c) Any endomorphism of F is an automorphism: G = F (X).

(Hint: Consider the functor R : G-Setsf C, R(∆) = q∆ X /G, where g ∈ G acts on q∆ X via the
g i i
morphisms Xx − → Xgx →− q∆ X, x ∈ ∆. The obvious morphisms Xx → − q∆ X → R(∆) and qF (Y ) X →
Y induce (the second being G-invariant) natural morphisms t∆ : ∆ −→ F (R(∆)), tY : R(F (Y )) −→ Y .
Show that t∆ is bijective because F preserves coproducts and cokernels, hence quotients by G. On
tF (Y ) F (tY )
the other hand, we have a natural endomorphism F (Y ) −−−→ F RF (Y ) −−−−→ F (Y ), so that it is
a bijection by 3 (in fact, it is the identity). Since tF (Y ) is bijective, so is F (tY ); hence tY is an
isomorphism by 2.)
Obtain an alternative proof of Galois theorem, showing that the involved functor F fulfills the above
three conditions. (Hint: F is the composition of a base change A AL with an equivalence of
categories, see p. 143.)
107. Let F : C Setsf be a covariant functor. Prove that there exists a finite group G, an equivalence of
categories Φ : C G-Setsf and an isomorphism of functors F ' i ◦ Φ (where i : G-Setsf Setsf is
the forgetful functor) if and only if C admits finite coproducts and cokernels, any endomorphism of
F is an automorphism, and F is representable, exact and conservative. Moreover, in such case, G is
isomorphic to the group of automorphisms of F .
108. Let k → L be a Galois extension, char k = 0. If H = {τ1 , . . . , τn } is a subgroup of the Galois group,
prove that the k-linear map s : L → LH , s(α) = τ1 (α) + . . . + τn (α), is surjective.
109. Prove that any finite separable extension k → K only has a finite number of intermediate fields.
110. Let K1 , K2 be Galois extensions of a field k. If we fix a composite L of K1 and K2 , prove that
K̄ = K1 ∩K2 is a Galois extension of k. Moreover, if G1 , G2 , G, Ḡ are the Galois groups of K1 , K2 , L, K̄
over k, show that G = G1 ×Ḡ G2 = {(g1 , g2 ) ∈ G1 × G2 : g1 |K̄ = g2 |K̄ }.
111. Where does the following argument fail? If i : k → K and j : K → L are Galois extensions, then
j ◦ i : k → L also is a Galois extension because
K ⊗k L = (K ⊗k K) ⊗K L = (⊕K) ⊗K L = ⊕(K ⊗K L) = ⊕L,
L ⊗k L = L ⊗K (K ⊗k L) = L ⊗K (⊕L) = ⊕(L ⊗K L) = ⊕L.
(Hint: If k → K is a Galois extension, then we have an isomorphism of K-algebras K ⊗k K = ⊕K,
α ⊗ λ 7→ (τ1 (α)λ, . . . , τ1 (α)λ), where Autk-alg (K) = {τ1 , . . . , τn }).
ALGEBRA II 489
112. Let k → L0 and L0 → L be Galois extensions. If any automorphism of L0 over k may be extended to
an automorphism of L over k, prove that k → L is a Galois extension.
113. Prove Artin’s theorem when the extension k → L is not finite, but so is H. (Hint: If α1 , . . . , αr ∈ L,
then k(. . . , τi αj , . . .) is a Galois extension of fixed degree n = |H|).
114. Galois Group of Cubics: Let p(x) ∈ k[x] be a separable polynomial of degree n. If char k 6= 2, prove
that the Galois group G of p(x) over k is contained in the alternate subgroup An if and only if the
discriminant ∆ of p(x) is a square in k.
In the case of an irreducible cubic, if ∆ is a square, then G = A3 . Otherwise, G = S3 .
115. Let p4 (x) ∈ k[x] be a separable reducible quartic. If p4 (x) has no root in k, show that the Galois
group is G = {id, (12)(34)} or G = {id, (12), (34), (12)(34)}.
116. Galois Group of Quartics: Let p4 (x) = x4 + px3 + qx2 + rx + s ∈ k[x] be a separable irreducible
quartic, char k 6= 2. Let L = k(α1 , α2 , α3 , α4 ) be the decomposition field and G ⊆ S4 the Galois
group. Let r(y) be the cubic resolvent (p. 477), with roots ϑ1 = α1 α2 + α3 α4 , ϑ2 = α1 α3 + α2 α4 ,
ϑ3 = α1 α4 + α2 α3 , and let d = [k(ϑ1 , ϑ2 , ϑ3 ) : k]. Put V = {id, (12)(34), (13)(24), (14)(23)}.
Prove that the Galois group of L over k(ϑ1 , ϑ2 , ϑ3 ) is G ∩ V , so that |G ∩ V | = 2, 4 and |G| = 2d, 4d.
(a) If d = 6, then G = S4 .
(b) If d = 3, then G = A4 .
(c) If d = 1, then G = V .
(d) If d = 2, then |G| = 4 or 8. Up to conjugation there are only two possibilities:
G = D8 = { id , (1234) , (13)(24) , (1432) , (12)(34) , (14)(23) , (13) , (24) }
G = C4 = { id , (1234) , (13)(24) , (1432) }
and the cubic resolvent r(y) has a unique root in k, namely a = α1 α3 + α2 α4 . Moreover, G = C4
if and only if the polynomials
(x − α1 α3 )(x − α2 α4 ) = x2 − ax + s

x − (α1 + α3 ) x − (α2 + α4 ) = x2 + px + q − a
√
have the roots in the decomposition field of r(y), which is just k(ϑ1 , ϑ2 , ϑ3 ) = k( ∆), where ∆
is the discriminant of r(y) = (y − a)p2 (y). Hence, in the case d = 2, we have
√
i. G = C4 , when x2 − ax + s and x2 + px + q − a decompose over k ∆ .
ii. G = D8 otherwise.
117. Give quartics with rational coefficients and Galois groups S4 , A4 , D8 , C4 and V .
118. If an irreducible quartic with rational coefficients has just two real roots, prove that the Galois group
over Q is S4 or D4 .
119. Let α ∈ C be a root of an irreducible quartic p(x) ∈ Q[x]. If the Galois group of p(x) is S4 or A4 ,
show that there is no non-trivial intermediate field between Q and Q(α).
120. Prove that the Galois group of the generic polynomial of degree n is the symmetric group Sn .
121. Let Fq be the finite field with q elements. Prove that any finite extension Fq → L is a Galois extension,
of cyclic Galois group, generated by the automorphism F (α) = αq .
122. Let p 6= q be two prime numbers. Prove that xq−1 + . . . + x + 1 is irreducible in Fp [x] if and only if p
generates the group (Z/qZ)∗ .
123. Let Q ∈ Z[x] be separable. Prove that the reduction Q̄ ∈ Fp [x] is separable at any prime p, up to a
finite number. (Hint: Q̄ is inseparable if and only if p divides the discriminant of Q).
124. Determine a polynomial P (x) ∈ Q[x] of degree 5 and Galois group Z/5Z over Q.
For any prime number p, prove the existence of P (x) ∈ Q[x] with Galois group G ' Z/pZ.
√ √
125. We have 2 = e2πi/8 + e−2πi/8 , but prove that 3 2 is not a rational function, with rational coefficients,
of some complex roots of unity.
490 EXERCISES
√
126. Let p be an odd prime. If ε8 is a primitive 8-th root of unity over Fp , prove that 2 = ε8 + ε−1
8 .
p −p
Show that 2 is a square in Fp if and only if ε8 + ε−1
8 = ε 8 + ε8 .
p2 −1

Conclude that p = 1 if and only if p ≡ ±1 (mod. 8); that is to say, p2 = (−1) 8 .
2
127. Let p > 3 be a prime number and let ε12 be a primitive 12-th root of the unity over Fp .
√
Show that 3 = ε312 + 2ε712 , and obtain that p3 = 1 if and only if p ≡ ±1 (mod. 12).
128. Determine the natural numbers n such that cos 2π

n is a rational number.
129. Let us consider an extension k → k(α), αn ∈ k, where n 6= 0 in k and k contains the n-roots of unity.
If d > 0 is the first exponent such that αd ∈ k, show that xd − αd is irreducible in k[x].
130. Let Q → L be a Galois extension of group G. If G is not a cyclic group, show that L is the
decomposition field of a polynomial q(x) ∈ Q[x] without rational roots such that the reduction q̄(x) ∈
Fp [x] has some root in Fp for all prime numbers p, up to a finite number.
131. Let G be the Galois group over Q of a polynomial q(x) ∈ Q[x] of degree n without rational roots. If
any element τ ∈ G fixes some root of q(x), show that n > 4 and q(x) is not irreducible in Q[x].
132. Let G be the Galois group over Q of xn − a ∈ Q[x]. If any element τ ∈ G fixes some root of q(x), and
n is prime or n ≤ 7, show that a is a n-th power in Q.
133. Prove that any extension of Q contained in an extension of Q by quadratic radicals also is an extension
of Q by quadratic radicals.
134. Let k be a field of null characteristic containing the n-th roots of unity {ε, ε2 , . . . , εn = 1}. If k → L
is a cyclic extension of degree n and σ is a generator of the Galois group, prove that
(a) The endomorphism σ of L is diagonalizable, of eigenvalues ε, ε2 , . . . , εn = 1.
(b) Any element α ∈ L decomposes uniquely as a sum α = α1 + . . . + αn , where σ(αi ) = εi αi , so
√
that αi = n ai for some ai ∈ k.
(c) In fact αi = n1 R(α, ε−i ), where R(α, ε) = α +P
εσ(α) + ε2 σ 2 (α) + . . . + εn−1 σ n−1 (α) is the
Lagrange’s resolvent of α by ε, so that α = n1 i
i R(α, ε ).
135. In a finite k-algebra A, the norm N (a) and trace tr(a) of a ∈ A are defined to be the determinant
a·
Q endomorphism A −−→PA. In the case of a Galois extension k → L of group G, prove
and trace of the
that N (α) = g∈G g(α) and tr(α) = g∈G g(α).
136. Let k → L be a Galois extension of cyclic group {σ, σ 2 , . . . , σ n = Id}, and β ∈ L.
Multiplicative Hilbert’s theorem 90: We have N (β) = 1 if and only if β = α/σ(α) for some
0 6= α ∈ L (Hint: Fix θ ∈ L such that 0 6= α := θ + β(σθ) + . . . + β . . . (σ n−2 β)(σ n−1 θ).)
Additive Hilbert’s theorem 90: We have tr(β) = 0 if and only if β = α −σ(α) for some α ∈ L.
(Hint: Put α = tr1θ β(σθ) + (β + σβ)(σ 2 θ) + . . . + (β + . . . + σ n−2 β)(σ n−1 θ) , the trace tr : L → k
being non null because L is a separable extension.)
137. Let q(x) ∈ Q[x] be an irreducible polynomial of prime degree p. If q(x) only has 2 imaginary roots,
show that the Galois group is Sp ; hence q(x) is not solvable by radicals when p ≥ 5.
138. The Metacyclic Group: Let p be a prime number and let Aff p (1) be the group of affinities ax + b of
an affine line over Fp , viewed as a subgroup of order p(p − 1) of the group Sp of all permutations of
the affine line. Let T be the subgroup of all translations x + b, so that it is the subgroup of order p
generated by the cycle (1, 2, . . . , p). Prove the following statements:
(a) Aff p (1) is a solvable group. (Hint: Aff p (1) → F∗p , ax + b 7→ a, is a group morphism).
(b) Aff p (1) is the normalizer N (T ) of T in Sp . (Hint: The index of N (T ) in Sp is the number
number of p-cycles
p−1 = (p−1)!
p−1 = (p − 2)! of subgroups of order p).
(c) Let G ⊆ Sp be a solvable transitive subgroup. Any non trivial normal subgroup N ⊂ G is
transitive. (Hint: The isotropy subgroups I1 , . . . , Ip of G are conjugated, hence so are the
subgroups I1 ∩ N, . . . , Ip ∩ N , and all the orbits of the action of N have equal cardinal).
PROJECTIVE GEOMETRY 491
(d) Any solvable transitive subgroup G ⊆ Sp is, up to conjugation, a subgroup of Aff p (1) containing
T . Hence |G| = pd, where d divides p − 1. (Hint: If 1 H1 . . . Hn = G is a resolution, show
that, up to conjugation, H1 = T . Then prove inductively that T is the unique subgroup of of Hi
of order p, so that T Hi ).
(e) Let G ⊆ Sp be a solvable transitive subgroup. If τ ∈ G has 2 fixed points, then τ = id.
(f) A transitive subgroup G ⊂ Sp is solvable if and only if |G| = pq for some q < p. (Hint: Prove
that, up to conjugation, T is the unique subgroup of G of order p).
(g) The Galois group of an irreducible polynomial q(x) ∈ Q[x] of prime degree p is solvable if and
only if the decomposition field of q(x) is generated by any two roots.
(h) (Galois): If an irreducible polynomial q(x) ∈ Q[x] of prime degree p has solvable Galois group
and 2 real roots, then all the roots are real.
139. Prove that π0k (A ⊕ B) = π0k (A) ⊕ π0k (B) and π0k (A ⊗k B) = π0k (A) ⊗k π0k (B).
140. If k → K is a finite extension and A is a finite K-algebra, prove that [A : k]s = [A : K]s · [K : k]s .
141. Prove that subalgebras, quotients and tensor products of purely inseparable k-algebras are purely
inseparable k-algebras, and that the concept of purely inseparable k-algebra is geometric and local.
142. Let B be a subalgebra of a finite k-algebra A. If A is rational over some extension k → L, prove that
any morphism of k-algebras B → L may extended to a morphism of k-algebras A → L.
143. If k → L is a finite extension, prove that the following conditions are equivalent:
(a) L is a normal extension of k (i.e., L ⊗k L is a rational L-algebra).
(b) If an irreducible polynomial p(x) ∈ k[x] has a root in L, then it has all the roots in L.
(c) L is the decomposition field over k of some polynomial p(x) ∈ k[x].
(d) Any composite of L with itself is isomorphic to L.
144. Show that any purely inseparable extension is a normal extension.
145. Show that any extension of degree 2 is a normal extension.
146. Prove that any finite normal extension k → L of group G = Aut(L/k) decomposes as a tensor product
of a Galois extension and a purely inseparable extension, L = π0k (L) ⊗k LG .
147. Let k → L be a normal extension. If K is an intermediate field, prove that the group Aut(L/k) acts
transitively on the set Homk-alg (K, L).
148. Let Tr be the trace metric of a finite k-algebra A. If char k = 0, show that A = π0 (A) ⊕ (rad Tr).
6. Projective Geometry
1. Determine the cardinal of a projective space of dimension n over a finite field of q elements.
2. Show that a subset X ⊆ P(E) is a linear subvariety if and only if it contains the line passing through
any two different points of X.
3. Pappus theorem: In a projective plane, let A1 , A2 , A3 and B1 , B2 , B3 be two triples of collinear
points. Prove that the three pairs of lines Ai Bj and Bj Ai , i 6= j, intersect at collinear points.
4. Let λ = (P1 , P2 ; P3 , P4 ) be the cross-ratio of 4 points of a projective line P1 over a field k. Show that
unordered sets of 4 different points of P1 are projectively classified by the sum of the double products
of the 6 possible cross-ratios (up to a sign and a constant)
(λ2 − λ + 1)3
j(λ) = ∈ k ∪ {∞}.
λ2 (λ − 1)2
5. Show that an homography τ is an involution, τ 2 = Id, if and only if the linear representant T is
traceless, tr T = 0.
492 EXERCISES
6. Prove the universal property of the vector extension An → E of an affine space (p. 158): We have
Homaff (An , E 0 ) = Homk-lin (E, E 0 ) for any k-vector space E 0 .
7. Given a projective space π : E → P(E) and an hyperplane π(V ), show that P(E) − π(V ) has a natural
structure of affine space with space of free vectors V o ⊗k V . (Hint: Put π(e) + ω ⊗ v := π e + ω(e)v .)
8. If a bijection τ : P(E) → P(E 0 ) between projective lines preserves harmonic quadruples, prove that it is
induced by a semilinear transformation (σ, T ) : (k, E) → (k 0 , E 0 ). (Hint: A bijection k → k 0 preserves
squares if it preserves sums and inverses, λ2 = λ − (λ−1 + (1 − λ)−1 )−1 ).
9. Let P(E) be a projective space of dimension n. When n ≥ 2, show that straight lines may be defined
in terms of projectivities. In fact, if OP1 P2 (P3 ) denotes the orbit of P3 under the action of the group
of projectivities fixing P1 and P2 , minus P3 , prove that 3 different points are collinear if and only
if OP1 P2 (P3 ) = OP1 P3 (P2 ) = OP2 P3 (P1 ). Conclude that any bijection τ : P(E) → P(E) preserving
projectivities, τ P Gl(E)τ −1 = P Gl(E), is a collineation. (Hint: In the case of non collinear points, we
have P ∈ / OP1 P2 (P3 ) and P ∈ OP1 P3 (P2 ) when P is a third point in the line P1 + P2 .)
When n = 1, consider an affine reference P0 , P1 , P∞ in P(E) and the image P00 , P10 , P∞0
by a bijection
0
τ : P(E) → P(E) preserving homographies, so that τ : A1 = P(E) − {P∞ } → P(E) − {P∞ } = A01 is a
bijection such that τ (0) = 0, τ (1) = 1. Prove that τ is a ring automorphism and conclude that τ is
induced by a semilinear automorphism T : E → E. (Hint: τ preserves the addition and the product
because it preserves translations and homotheties with center at the origin.)
10. Prove the Fundamental Theorem of Affine Geometry: A bijection An → A0n between affine
spaces of dimension n ≥ 2 is an affinity if and only if it induces an isomorphism between the lattices
of affine subvarieties.
Prove the Fundamental Theorem of Euclidean Geometry: An affinity between Euclidean spaces
of dimension n ≥ 2 is a similarity if and only if it preserves the perpendicularity of lines.
11. Let us consider an affine space An of dimension n ≥ 2 over a field k. When k 6= F2 , prove that X ⊆ An
is an affine subvariety if and only if it contains the line passing through any two different points of
X. (Hint: Given two vectors e, v in a k-vector space, show that the equation xe + z(yv − xe) = e + v
admits a solution in k.) Conclude that a bijection An → A0n onto another affine space of dimension n
is an affinity if and only if it preserves lines.
When k = F2 (so that any pair of points is a line), show that X ⊆ An is an affine subvariety if and
only if it contains the plane passing through any three different points of X. Conclude that a bijection
An → A0n onto another affine space of dimension n is an affinity if and only if it preserves planes.
12. Let Aff be the category of affine spaces and affinities over a fixed base field k, and let PAff be the
category of pairs (P(E), H), where H is a hyperplane of a k-projective space P(E), and projectiv-
ities preserving the hyperplane. Show that we have an equivalence of categories F : Aff PAff,
F (An , V ) = (P(E), π(V )), where E is the vector extension of An . (Hint: If you are willing to assume
that the objects form a set, use problem 50 in p. 483. Otherwise put An = P(E) − H and reconstruct
V as the group of translations, well identified with the linear representant of H up to a factor).
Let Euclid the category of Euclidean affine spaces (An , V, hΩ2 i) and similarities, and let PEuclid be
the category of sequences (P(E), H, Q), where Q is a non singular quadric of index 0 in a hyperplane
H of P(E), and projectivities preserving the absolute. Show that we have an equivalence of categories
Euclid PEuclid.
13. Let T 0 , T be two totally isotropic subspaces of a metric vector space E. If dim T 0 > dim T , prove
the existence of a vector 0 6= e ∈ T 0 such that hei ⊕ T is totally isotropic. (Hint: The natural map
T 0 /(T 0 ∩ T ) → (T /T 0 ∩ T )∗ is not injective.)
Conclude (without using Witt’s theorem) that all maximal totally isotropic subspaces of E have equal
dimension).
14. The signature of a (symmetric) metric S on a finite dimensional real vector space E is defined to be
the pair (r+ , r− ) where r+ (resp. r− ) is the maximal dimension of a vector subspace of E where S
is positive definite (resp. negative definite). In the real case, show that the dimension and signature
classify metrics. Moreover, if A is the matrix of S in some basis of E, show that r+ (resp. r− ) is the
number of positive (resp. negative) roots of |xI − A|, counted with multiplicity.
Given a symmetric n × n matrix (aij ) with real coefficients and signature (n+ , n− ), prove that there
are vectors e1 , . . . , en ∈ E such that S(ei , ej ) = aij if and only if n+ ≤ r+ and n− ≤ r− .
15. Let E be a real vector space of dimension 2. Show that q : End(E) → R, q(T ) = det T , is a quadratic
form of signature (2, 2).
16. Let Ω2 be an alternate metric on a k-vector space E. If e ∈ / rad Ω2 , show that e · e0 = 1 for some
0 0 0 ⊥
vector e ∈ E, so that E = he, e i ⊥ he, e i . Obtain, by induction on the dimension, a direct proof of
the classification of 2-forms (p. 163), even when char k = 2.
17. Let φ : E → E ∗ be the polarity of a non-singular metric S on a vector space E. The dual metric S ∗
is defined to be the unique metric on E ∗ such that S ∗ (φ(e), φ(v)) = S(e, v), i.e., φ(e) · φ(v) = e · v.
If we multiply S by a non-null factor λ, show that the dual metric is multiplied by λ−1 , and we define
Q∗ = hS ∗ i to be the dual quadric of the non-singular quadric Q = hSi.
Prove that the points of Q∗ are just the tangent hyperplanes to Q, and that two points are conjugated
with respect to Q if and only if the polar hyperplanes are conjugated with respect to Q∗ .
Prove that φ−1 : E ∗ → E is just the polarity of the non-singular metric S ∗ , and conclude that S is
the dual metric of S ∗ , so that Q is just the dual quadric of Q∗ .
18. Let Q = hSi be a quadric in a projective space P(E) of vertex X = π(V ) and let S̄ be the projection
of S by the epimorphism E → Ē = E/V . Since S̄ is non-singular, it defines a dual metric S̄ ∗ in
Ē ∗ ,→ E ∗ ; hence a dual quadric hS̄ ∗ i in P(Ē ∗ ) = X o = π(V o ) ,→ P(E ∗ ).
Prove that the points of the dual quadric are just the tangent hyperplanes to Q.
Show that quadrics in P(E) with vertex X naturally correspond with non-singular quadrics in X o .
19. Let Q = hSi be a quadric and P = hei a non-incident point. Put ω = ie S, and show that in the pencil
λS + µω ⊗ ω, the quadric C passing through P is represented by ω(e)S − ω ⊗ ω.
Prove that P is in the vertex of C, and that any line joining P to an intersection point of Q and C is
tangent to Q (so that C deserves the name of tangent cone to Q from the point P ).
20. (Resolution of cubics and quartics with radicals): Let δ be the discriminant of a cubic p(x) ∈ Q[x]
with three different complex roots α1 , α2 , α3 . Prove that a homography τ (x) = ax+b cx+d permuting the
√
roots, τ (α1 ) = α2 , τ (α2 ) = α3 , τ (α3 ) = α1 , may be determined as a rational function of ∆. If we fix
2πi
a projective parameter θ so that the fixed points √ of τ are θ = 0 and θ = ∞, show that τ (θ) = e 3 θ,
3
and our cubic equation reduces to θ = β, β ∈ Q( ∆ ).
The roots of a quartic x4 + c1 x3 + c2 x2 + c3 x + c4 are given by the intersection points of two conics
0 = y 2 + c1 xy + c2 y + c3 x + c4

y = x2
If we find a singular conic in the pencil (y 2 + c1 xy + c2 y + c3 x + c4 ) + λ(y − x2 ) = 0, intersecting it
with y = x2 we obtain the roots of the quartic. Use this fact to show how the roots of a quartic may
be calculated by means of the roots of a cubic equation.
21. Prove the following statements in the hyperbolic plane (P2 , Q = hΩi):
(a) Lines have infinite length.
(b) For any two pairs r, s and r0 , s0 of perpendicular lines, there is a projectivity τ : P2 → P2 such
that τ (Q) = Q, τ (r) = r0 and τ (s) = s0 . (All right angles are equal to one another).
√
(c) The height of any equilateral right triangle is bounded above by arccosh 2.
(d) The circles with center at a point O = π(e) are just the conics hΩ − µω ⊗ ωi bitangent at the
intersection points with the polar line of O, such that 0 < µ ≤ 1 (where ω = ie Ω and we fix Ω so
that Ω(e, e) = 1).
22. In a projective line, show that (x, y; p, q)(y, z; p, q) = (x, z; p, q).
Let x, y, z be non-collinear points in the hyperbolic plane, and let us consider the points at infinity
{a1 , a2 }, {b1 , b2 } and {c1 , c2 } of the lines xy, yz and xz respectively. Prove the existence of collinear
points c1 < p < x < z < q < c2 such that
494 EXERCISES
1 < (x, z; c2 , c1 ) < (x, z; q, p) = (x, y; a2 , a1 )(y, z; b2 , b1 ),

and conclude that the hyperbolic distance satisfies the triangle inequality.
23. Special Theory of Relativity: Now we consider an inertial reference (p0 ; e0 , e1 , e2 , e3 ) in a Minkowski
spacetime (A4 , V ; g, h = −c2 g), and the dual basis ω0 , ω1 , ω2 , ω3 of e0 , e1 , e2 , e3 . Recall that we put
e · v = g(e, v). Prove the following statements:
(a) If two events are joined by a trajectory (a connected smooth curve where the time metric g is
positive-definite) then they may be joined by an inertial (= straight line) trajectory.
(b) A vector e is timelike (e · e >√0) if and only if p and q = p + e occur at the same place for some
inertial observer. In this case e · e is the time interval between p and q for such inertial observer,
and it bounds below the time interval between p and q for any other inertial observer.
(e · e < 0) if and only if p and q = p + e are simultaneous for some inertial
(c) A vector e is spacelike p
observer. In this case h(e, e) is the distance between p and q for such inertial observer, and it
bounds below the distance between the places where p and q occur for any other inertial observer.
(d) If a voyager moves away with velocity ve1 during a time interval p t, and then he returns with
velocity −ve1 , check that the proper time of the voyager is τ = 2t 1 − (v/c)2 < 2t. If v c,
the difference between the observed time 2t and the proper time τ is ≈ t(v/c)2 .
(e) If we consider a rod of length le1 at rest in an inertial
p reference, check that its length for another
inertial observer moving with speed ve1 is l0 = l 1 − (v/c)2 . When v c, lengths contract
≈ lv 2 /2c2 .
(f) Now assume that c = 1 (i.e. h = −g) and let us consider a non-zero 2-form F on V and the
endomorphism F̃ : V → V , such that F (e, v) = h(F̃ (e), v). If in an inertial reference e0 , e1 , e2 , e3
we have
F = ω0 ∧ (E1 ω1 + E2 ω2 + E3 ω3 ) − (B1 ω2 ∧ ω3 + B2 ω3 ∧ ω1 + B3 ω1 ∧ ω2 ),
check that the matrix of the endomorphism F̃ is just
 
0 E1 E2 E3
E1 0 B3 −B2 
A= E2 −B3

0 B1 
E3 B2 −B1 0
4 P 2 P 2 2 2
= (x2 − α2 )(x2 + β 2 ).
P
|xI − A| = x + i Bi − i Ei x − i E i Bi
~ = E1 e1 + E2 e2 + E3 e3 , B
Even if the electric and magnetic fields E ~ = B1 e1 + B2 e2 + B3 e3 depend
on the inertial observer, the invariants
~ 2 − kEk
kBk ~ 2 = B 2 + B 2 + B 2 − E 2 − E 2 − E 2 = β 2 − α2
1 2 3 1 2 3
~ ~
hE|Bi = (E1 B1 + E2 B2 + E3 B3 ) = β 2 α2
2 2
do not depend, and these invariants classify the pair (g, F ):

i. If α = 0 and β = 0 (i.e., E ~ and B~ are orthogonal vectors of equal length), there exists an
inertial reference where F = ω0 ∧ ω1 + ω1 ∧ ω3 .
ii. Otherwise, there exists an inertial reference where the vectors E ~ and B
~ are parallel, of
lengths α and β respectively, and F = αω0 ∧ ω1 + βω2 ∧ ω3 .
Hence we have that B ~ = 0 for some inertial observer if and only if β = 0, and that E ~ = 0 for
some inertial observer if and only if α = 0.
(g) Show that the electric field E ~ only depend on the velocity e0 of the inertial observer, not on the
~ = F̃ (e0 ), or that ie F = i ~ Eh.)
spatial axes e1 , e2 , e3 . (Hint: Show that E 0 E
(h) Once we fix an orientation of V , prove that there is a unique 4-form ΩV (the hypervolume form)
such that the hypervolume of any oriented inertial reference frame e0 , e1 , e2 , e3 is 1. Moreover,
show that the restriction of F to E = (Re0 )⊥ is just −iB~ ΩE , where ΩE := ie0 ΩV , and conclude
that the magnetic field B ~ only depend on the velocity e0 of the oriented inertial observer, not on
the spatial axes e1 , e2 , e3 .
24. Assume, for the sake of simplicity that c = 1 and that free vectors in spacetime form a vector space V
of dimension 2, so that the time metric g has signature (1, 1), and let {e0 , e1 } be a basis of V where
the matrix of the time metric g is diag(1, −1).
If e00 is another vector such that e00 · e00 = 1, prove that the coordinates of e00 are t = cosh α, x = sinh α,
where α is the length (with respect to −g) of the arc joining e0 with e00 in the hyperbola t2 − x2 = 1.
Hence the apparent speed v of any body with velocity e00 is just v = tanh α := (sinh α)/(cosh α).
Let us consider inertial observers with velocities e0 , e00 , e000 and let vi/j be the speed of the i-th observer
for the j-th observer. If α is the length of the arc joining e0 with e00 and β is the length of the arc
joining e00 with e000 , then we have v2/1 = tanh α and v3/2 = tanh β. Conclude (and compare with
tanh α+tanh β v3/2 +v2/1
problem 1 in p. 468) that v3/1 = tanh(α + β) = 1+(tanh α)(tanh β) = 1+v3/2 v2/1 .
25. Given a n-dimensional

P P space (An , V ), let us consider a polynomial q of degree ≤ P
affine 2 on An . If
we have q = aij xi xj + ai xi + c in an affine reference, prove that the leading term aij xi xj is
a well-defined quadratic form on V , i.e. that it does not depend on the fixed affine reference. (Hint:
Show that q is the restriction of a unique quadratic form on the vector extension An ,→ En+1 defined
in p. 158, and consider its restriction to V ⊂ En+1 ).
26. Let us consider an Euclidean plane (A2 , V ), where we have fixed the unit length (i.e. the scalar
product on V ). The 4-dimensional real vector space E = {d(x2 + y 2 ) + ax + by + c} of polynomial
functions on A2 of degree ≤ 2 with leading term proportional to the (quadratic form of the) scalar
product is named the space of circles since it includes real circles (x − x0 )2 + (y − y0 )2 − r2 = 0,
imaginary circles (x − x0 )2 + (y − y0 )2 + r2 = 0, points (x − x0 )2 + (y − y0 )2 = 0, lines ax + by + c = 0
and the empty set c = 0. Show that in this space E the coefficient d defines a well-defined linear
form (it does not depend on the Euclidean reference) and prove that there is a unique quadratic
form q : E → R such that q (x − x0 )2 + (y − y0 )2 − r2 = r2 , so that the modulus of a real circle
√
C = (x − x0 )2 + (y − y0 )2 − r2 = 0 is just the radius r = C · C.
Show that the bilinear metric h corresponding to q has signature (3, 1), so that (E, Rh) is a Minkowski
spacetime. Fix g := −h, and prove that equations of real circles are spacelike vectors (C · C > 0),
equations of imaginary circles are timelike vectors (C · C < 0), equations of points are light vectors
(P · P = 0), while equations of lines are just vectors in the hyperplane d = 0 tangent to the light cone
at the generatrix defined by the empty set d = a = b = 0. Moreover:
(a) Two real circles C1 , C2 are orthogonal, C1 · C2 = 0, when both circles intersect orthogonally.
(b) In the case of a real circle C1 and an imaginary circle C2 of radius r2 i, we introduce the concentric
real circle C̄2 of radius r2 , and condition C1 · C2 = 0 means that C1 intersects C̄2 in a pair of
diametrally opposite points.
(c) A real circle C and a point P are orthogonal when C passes through P .
(d) A circle C and a line L are orthogonal when L passes through the center of C.
(e) Two lines are orthogonal, L1 · L2 = 0, when both are perpendicular.
(f) No circle is orthogonal to the empty set, while so are all lines.
(g) If two different real circles C1 , C2 intersect and φ is the angle between the tangent lines at a
common point, then C1 · C2 = |C1 | · |C2 | · cos φ.
(h) If we represent any real circle by the equation x2 + y 2 + ax + by + c = 0 with d = 1, check that
we have −2C1 · C2 = D2 − r12 − r22 , where ri is the radius of Ci and D is the distance between
the centers of both circles. When C2 is a point (r2 = 0), up to a factor, C1 · C2 is just the power
of a point with respect to the circle C1 .
Finally, we represent real circles by the equation C of modulus 1 and d > 0,

(x − x0 )2 + (y − y0 )2 − r2
C= ·
r
496 EXERCISES
(a) Prove that the product C1 · C2 of two real circles has the following geometrical meaning:
C1 · C2 >1 non-intersecting interior circles

C1 · C2 =1 interior tangent circles
C1 · C2 = cos φ where φ is the angle formed by the radius at a common point
C1 · C2 = −1 exterior tangent circles
C1 · C2 < −1 non-intersecting exterior circles
(b) There exist equations e1 , . . . , en ∈ E with a given configuration matrix (gij ) = (ei · ej ) if and only
if this matrix is symmetric, of signature ≤ (3, 1).
(c) If we look for 4 mutually orthogonal circles, the configuration matrix (Ci · Cj ) is the unit matrix,
so that this configuration is impossible.
(d) If the configuration matrix (gij ) = (Ci · Cj ) of 4 real circles has rank 4, then C1 , C2 , C3 , C4 is a
basis of E and the inverse matrix (g ij ) is just the matrix of the dual metric in the dual basis.
P 1-form, ω · ω = 0, determines a relation between the radii and the centers of
Hence, any isotropic
the circles: 0 = i,j g ij ω(Ci )ω(Cj ).
(e) Isotropic 1-forms are ω(C) = d and the complex 1-form ω 0 (C) = − 21 (a + bi). If κ = r−1 is the
curvature of a real circle C and z ∈ C is the center, then ω(C) = κ and ω 0 (C) = κz.
(f) The curvatures κi and centers zi of 4 mutually exterior tangent circles satisfy
2(κ21 + κ22 + κ23 + κ24 ) = (κ1 + κ2 + κ3 + κ4 )2 ,
2(κ21 z12 + κ22 z22 + κ23 z32 + κ24 z42 ) = (κ1 z1 + κ2 z2 + κ3 z3 + κ4 z4 )2 .
27. Try to generalize these results to the space {d(x21 + . . . + x2n ) + a1 x1 + . . . + an xn + c} of equations of
spheres in a n-dimensional Euclidean space.
28. Pair of points p + q in a real projective line P1 = P(E), including coincident points 2p and pairs of
complex conjugate points, are just points of the real plane P2 = P(S 2 E ∗ ). Show that the pairs 2p of
coincident points define a real non-singular quadric Q in P2 (S 2 E ∗ ), and that
(a) The tangent line to Q at a point 2p is just the line Lp of all pairs p + x.
(b) The exterior points are pairs p + q of different real points, the tangent lines to Q trough p + q
being Lp and Lq .
(c) The interior points are pairs of conjugate complex points (they form an hyperbolic plane!).
(d) The polar of a pair p + q is the line of all pairs of harmonic conjugates with respect to p, q.
(e) Three pairs p1 + q1 , p2 + q2 , p3 + q3 are collinear if and only if there is an involutive homography
τ such that τ (pi ) = qi , i = 1, 2, 3.
29. Prove that central conics have two axes, while parabolas have a unique axis.
30. Prove that any central conic (not a circumference) has two real focuses on an axis, and two imaginary
focuses on the other axis, while parabolas have a unique proper focus (and an improper focus), and it
is on the axis. Moreover, the points of the directrix of a parabola are just points with perpendicular
tangent lines.
31. Given two points in a plane, prove that the points such that the sum of the distances to such points
is a fixed constant 2a define an ellipse with focuses at the fixed points, and semi-major axis a.
32. Let us consider a line and an exterior point in a plane. Prove that the points such that the distances
to the fixed point and line have a given proportion e > 0 (named eccentricity) define an ellipse when
e < 1, a parabola when e = 1 and an hyperbola when e > 1, such that the fixed point is a focus
and the line is the corresponding directrix. In polar coordinates (ρ, θ) with center at the focus and
vertical directrix, this conic is ρ(1 + e cos θ) = p, where p is named semi latus rectum. In an ellipse
of semi-major axis a, semi-minor axis b and distance c between the center and the focus (so that
a2 = b2 + c2 ) show that
r
c b2
e = = 1 − 2 , ap = b2 .
a a
33. Assume that the position ~x(t) of a planet with respect to the Sun fulfills the differential equation
m~x
~x00 = − ·
|~x|3
If ~x and ~v := ~x0 are linearly independent at some instant, show that the angular momentum ~h = ~x ×~v ,
hence h := |~h|, is constant along the trajectory, so that ~x lies in a plane.
Consider polar coordinates, so that ~x = (ρ cos θ, ρ sin θ, 0). Prove that hdt = ρ2 dθ (a line segment
joining a planet and the Sun sweeps out equal areas during equal intervals of time), and conclude that
d~v m
= − (cos θ, sin θ, 0)
dθ h
m m
so that the hodograph ~v = h (− sin θ, cos θ, 0) + ~c lies in a cercle of radious R := h (the center ~c
being essentially the Runge-Lenz vector ~h × ~c ).
Put e = |~c|/R, and assume that ~c is in the positive y-axis, so that ~v = R(− sin θ, e + cos θ, 0). Using
that ~h = ~x × ~v , conclude that the orbit of a planet is an ellipse with the Sun at a focus:
ρ(1 + e cos θ) = p , p := h2 /m.
p
Finally, relate the area A = πab = a3 p of the ellipse to the period T , and conclude that the square
of the period of a planet is proportional to the cube of the semi-major axis of its orbit:
Z Z
1 2 h hT
A= ρ dθ = dt = , mT 2 = 4π 2 a3 .
2 2 2
34. In the case of the positions ~x1 (t), ~x2 (t) of two punctual masses m1 , m2 moving according to the
newtonian gravitational attraction,
m1 m2 m1 m2
m1 ~x001 (t) = (~x2 − ~x1 ), m2 ~x002 (t) = (~x1 − ~x2 ),
|~x1 − ~x2 |3 |~x1 − ~x2 |3
1
show that the barycenter m1 +m 2
(m1 ~x1 + m2 ~x2 ) moves uniformly. In the case that m1 ~x1 + m2 ~x2 = 0,
so that the position of both particles is fully determined by the relative position ~x1 − ~x2 ,
m2 m1
~x1 = (~x1 − ~x2 ), ~x2 = − (~x1 − ~x2 ),
m1 + m2 m1 + m2
prove that ~x(t) := ~x1 (t) − ~x2 (t) fulfills the differential equation ~x00 = − (m1|~x+m
|3
2)
~x and ~x(t) follows a
conic with a focus at the barycenter, so that the trajectories of both particles are homothetic conics,
the center of homothety being a common focus and the ratio being −m1 /m2 .
35. Let T (M ) be the torsion submodule of a module M over a domain A. If S is a multiplicative system of
A, show that T (MS ) = (T M )S . Conclude that M is torsion free if and only if so is Mx , ∀x ∈ Spec A.
36. Put B = k[x]/(pn ), where p(x) ∈ k[x] is irreducible. Show that the annihilator of B ∗ = Homk (B, k)
is pn , and obtain the existence of a B-linear isomorphism B ' B ∗ = Homk (B, k).
If we have an exact sequence 0 → B → M → M̄ → 0 of finitely generated B-modules, prove that
the dual sequence 0 → M̄ ∗ → M ∗ → B ∗ → 0 splits, and conclude that M ' B ⊕ M̄ ; so obtaining
an elementary (without using the ideal criterion on injectivity) proof of the second decomposition
theorem for k[x]-modules and the classification of endomorphisms.
37. Let T be an endomorphism of a finite-dimensional vector space E. Prove that E is a monogenous
k[x]-module if and only if so is E ∗ (with the module structure defined by T ∗ : E ∗ → E ∗ ).
Conclude that T and T ∗ have the same invariant factors.
38. Prove the following properties of the tensor algebra of a module:
(a) T • (M ⊕ N ) = (T • M ) ⊗A (T • N ).
(b) T • (M/N ) = (T • M )/(N ), where (N ) is the two-sided ideal generated by N ⊆ M = T 1 M .
(c) (TA• M )B = TB• (MB ), for any base change A → B.
39. Prove the following properties of the symmetric algebra of a module:
498 EXERCISES
(a) S • (M ⊕ N ) = (S • M ) ⊗A (S • N ).
(b) S • (A⊕ . n. . ⊕A) = A[x1 , . . . , xn ].
(c) S • (M/N ) = (S • M )/(N ), where (N ) is the ideal generated by N ⊆ M = S 1 M .
• •
(d) (SA M )B = SB (MB ), for any base change A → B.
40. Let L ' An be a free A-module of rank n. Prove that Λp L is a free A-module of rank np , so that

any endomorphism T : L → L induces an endomorphism Λn : Λn L → Λn L of a free module of rank 1:

the product by an element det T ∈ A, named determinant of T . Show that
(a) We have det(T ⊗ 1) = det T for any ring morphism A → B.

(b) The determinant is multiplicative: det(T ◦ S) = (det T )(det S).
(c) T is an automorphism if and only if det T is invertible in A.
41. Let E be a finite-dimensional k-vector space. Show that Λp E ∗ = (Λp E)∗ is canonically isomorphic to
the vector space Alt(E, . p. ., E; k) of p-covariant alternate tensors, that (S p E)∗ is canonically isomorphic
to the vector space Sym(E, . p. ., E; k) of p-covariant symmetric tensors, and that S 2 E ∗ is canonically
isomorphic to the vector space of quadratic forms (maps q : E → k such that q(λe)P = λ2 q(e) and
g(e, v) = q(e + v) − q(e) − q(v) is a bilinear map, so that in coordinates q(x1 , . . . , xn ) = i≤j aij xi xj ).
When char k = 0, show that the composition Sym(E, . p. ., E; k) ,→ T p E ∗ → S p E ∗ is an isomorphism;
but, when char k = 2, there is no natural (i.e. functorial) isomorphism Sym(E, E; k) = S 2 E ∗ . (Hint:
When dim E = 2, the natural map S 2 E ∗ → Sym(E, E; k) has 1-dimensional image).
42. If E is a finite-dimensional k-vector space, show that we have natural exact sequences
0 −→ Λ2 E −→ E ⊗k E −→ S 2 E −→ 0,
h
0 −→ Sym(E, E; k) −→ Bil(E, E; k) −−→ Alt(E, E; k) −→ 0,
0 −→ Alt(E, E; k) −→ Bil(E, E; k) −→ S 2 E ∗ −→ 0.
43. Let us consider the polarity E → E ∗ defined by a symmetric metric g, where dimk E < ∞.
Prove that the induced linear maps Tpq E → Tpq E ∗ = (Tpq E)∗ and Λp E → Λp E ∗ = (Λp E)∗ define
symmetric metrics Tpq g and Λp g on Tpq E and Λp E.
If g is non singular, prove that so are Tpq g and Λp g, and that Tqp g is the dual metric of Tpq g, via the
natural isomorphism Tqp E = (Tpq E)∗ .
If char k = 0 and we view Λp E as a vector subspace of T p E, formed by the alternate tensors, show
that the restriction of T p g to Λp E is just p!(Λp g).
n
(−1)i tr(Λi T )xn−i .
P
44. If T is an endomorphism of a vector space of dimension n, show that cT (x) =
i=0
45. Classify the group of invertible elements in the ring Z[x]/(xn ).

46. Let I be the annihilator of a finitely generated module M over a principal ideal domain A. Show that
the ring of all endomorphisms of M commuting with all the idempotent endomorphisms of M is A/I.
r
d2r−2i+1
Q
47. Let d1 , . . . , dr be the invariant factors of a finite abelian group G. Prove that |Aut G| = i .
i=1
48. Let E be a finite dimensional k-vector space. Let φ1 , . . . , φr be the invariant factors of an endomor-
phism T : E → E, and put di = deg φi . Prove that
Pr
(a) dimk Endk[x] E = i=1 (2r − 2i + 1)di .
(b) Any endomorphism of E commuting with the endomorphisms commuting with T is in k[T ].
(c) ET is monogenous if and only if the ring Endk[x] E is commutative.
49. Let T be an isometry, (T e) · (T v) = e · v, of a non singular metric

n on a k-vector space, char k 6= 2.
Prove that T has as many elementary divisors (x − a)n as x − a1 .
50. Let S be a non-singular symmetric metric on a finite dimensional k-vector space E, car k 6= 2. If
S 0 is another symmetric metric on E and we put e · v = S(e, v), e ∗ v = S 0 (e, v), then the linear
endomorphism T : E → E such that (T e) · v = e ∗ v = e · (T v) defines a structure of k[x]-module on
E. Prove the following statements:
(a) The polarity φ : E → E ∗ , φ(e) = S(e, −), is an isomorphism of k[x]-modules.
(b) The S-orthogonal of a submodule also is a submodule.
(c) If V1 , V2 are submodules and V1 · V2 = 0, then V1 ∗ V2 = 0.
(d) The radical of S 0 is a submodule.
(e) If another pair (S̄, S̄ 0 ) defines T̄ : Ē → Ē, then a k-linear isomorphism ET −∼
→ ĒT̄ transforms
0 0
(S, S ) into (S̄, S̄ ) if and only if it is k[x]-linear and transforms S into S̄.
(f) First Theorem: Let p = pn1 1 . . . pns s be the decomposition into irreducible factors of the annihila-
tor polynomial of T . Then the decomposition E = E1 ⊥ . . . ⊥ Es , Ei = Ker pni i , is orthogonal for
both metrics. (Hint: The polarity φ : E → E ∗ = E1∗ ⊕ . . . ⊕ Es∗ is a k[x]-linear and pni i annihilates
Ei∗ = {ω ∈ E ∗ : ω(Ej ) = 0, j 6= i}. Hence φ(Ei ) = Ei∗ , and Ei · Ej = 0 when j 6= i).
(g) Second Theorem: E is an orthogonal sum of homogeneous modules (i.e. ' A/pn ⊕ . . . ⊕ A/pn
with p irreducible). (Hint: Assume that E = Hn ⊕ . . . ⊕ H1 , where Hi is homogeneous, with
annihilator pi . Show that Hn is non singular: since rad Hn is a submodule, otherwise there is
0 6= pn−1 e ∈ rad Hn , so that pn−1 e · Hi = e · pn−1 Hi = 0, i ≤ n − 1, and pn−1 e ∈ rad E. Hence
E = Hn ⊥ Hn⊥ ).
Now assume that E is homogeneous, with annihilator pn , and consider the filtration
E = En ⊃ En−1 ⊃ . . . ⊃ E1 ⊃ E0 = 0, where Ei = pn−i E.
(a) dim Ei = i(dim E1 ). (Hint: Show that p : Ei → Ei−1 is an epimorphism with kernel E1 ).
(b) Ei · Ej = 0, when i + j ≤ n. (Hint: Show that pn−i E · pn−j E = E · p2n−i−j E = 0).
⊥ ⊥
(c) E1 = En−1 = rad En−1 . (Hint: Show that E1 ⊂ En−1 , and compare dimensions).
Third Theorem: There are vector subspaces Fi ⊂ Ei such that

(a) E = Fn ⊕ Fn−1 ⊕ . . . ⊕ F1 .
(b) We have isomorphisms p : Fi → Fi−1 , i ≥ 2.
(c) Fi · Fj = 0, when i + j 6= n + 1.
(Hint: When n = 1, put F1 = E1 . When n = 2, show that F1 = E1 is totally isotropic, so that there is
another totally isotropic subspace F2 such that E = F2 ⊕ F1 . When n ≥ 3, consider the projections S̄,
S̄ 0 , T̄ of S, S 0 and T by π : E → Ē := En−1 /E1 , so that S̄ is non singular and T̄ is the endomorphism
associated to the pair (S̄, S̄ 0 ). Show that the annihilator of T̄ is pn−2 . Hence, by induction, you have
a decomposition Ē = F̄n−2 ⊕ . . . ⊕ F1 , p : F̄i ' F̄i−1 and F̄i · F̄j = 0, when i + j 6= n − 2 + 1 = n − 1.
Now fix a subspace Fn−1 ⊂ En−1 such that π : Fn−1 ' F̄n−2 and put F1 = E1 , Fi = pn−i−1 Fn−1 , so
that all the morphisms in the following commutative diagram
Fn−1
p
/ Fn−2 p
/ ... p
/ F2 p
/ F1
π π π

F̄n−2
p
/ F̄n−3 p
/ ... p
/ F̄1
are in fact isomorphisms. Using that π : Fn−2 ⊕ . . . ⊕ F2 → Ē is an isometry, show that the subspaces
Fn−1 , . . . , F1 satisfy the required conditions. Then find a totally isotropic subspace Fn such that
∼ ∼
p : Fn − → pFn and π : pFn −→ F̄n−2 (and put Fn−1 := pFn ), so that Fn · Fi = 0 when i 6= 1): fix a
∼
subspace Un such that p : Un − → Fn−1 , and consider on Un ⊕ F2 the non singular metric e ◦ v = e · pv;
in fact, if R is the radical, you have
0 = R ◦ F2 = R · pF2 = R · F1 ,
500 EXERCISES
0 = R ◦ Un = R · pUn = R · Fn−1 ,
these equalities stating that R ⊆ En−1 (hence R ⊆ F2 ) and π(R) ⊆ rad Ē = 0. Hence R = 0 because
π : F2 → F̄1 is an isomorphism. Now show that F2 is totally isotropic for this metric ◦
F2 ◦ F2 = F2 · pF2 = F2 · F1 = 0, (n ≥ 3)
hence there is a totally isotropic subspace Vn ⊂ Un ⊕F2 (i.e. Vn ·pVn = 0) such that Un ⊕F2 = Vn ⊕F2 .
∼
Moreover π(pVn ) = π(pUn ) = F̄n−2 , and you may fix Fn−1 = pVn , so that p : Vn − → Fn−1 , and
Vn · Fn−1 = 0. Now Vn ⊕ F1 is non singular for S and F1 is totally isotropic; hence there is a totally
∼
isotropic subspace Fn with Fn ⊕ F1 = Vn ⊕ F1 . Check that E = Fn ⊕ En−1 , p : Fn − → Fn−1 , it is
isotropic, Fn · Fn−1 = (Vn ⊕ F1 ) · Fn−1 = 0, and Fn · Fi = Fn · pFi+1 = pFn · Fi+1 = Fn−1 · Fi+1 = 0
when n + i 6= n + 1.)
(a) When k is algebraically closed, E is an orthogonal sum of monogenous submodules. (Hint: The
radical of pn−1 e · v is Ker pn−1 = En−1 ; hence there is e ∈ Fn such that pn−1 e · e 6= 0. Now
V = he, pe, . . . , pn−1 ei is non singular, and a submodule because p = x − α; hence E = V ⊥ V ⊥ ).
(b) When k is algebraically closed, pairs of symmetric metrics (the first non singular) are classified
by the elementary divisors of T .
51. Now assume that S is symmetric and non singular, and S 0 = H ∈ Λ2 E ∗ is hemisymmetric:
(T e) · v = e ∗ v = e · (−T v)
qe · v = e · q̄v, where q̄(x) = q(−x).
In this case T and −T have equal annihilator p(x), so that p(−x) = ±p(x). If p = pn1 1 . . . pns s is the
decomposition into irreducible factors, for any factor pi we have (p̄i ) = (pi ), or (p̄i ) = (pj ), (p̄j ) = (pi )
with j 6= i. Prove the following statements:
(a) First Theorem: In the decomposition E = E1 ⊕ . . . ⊕ Es , each term Ei = Ker pni i is (for S)
1.- non singular and orthogonal to the remaining terms, when (p̄i ) = (pi ),
2.- or totally isotropic, and Ei ⊕ Ej is non singular and orthogonal to the remaining terms, when
(p̄i ) = (pj ).
(Hint: The polarity φ : E → E ∗ of S is a semilinear isomorphism, φ(qe) = q̄φ(e), so that it
transforms Ei into the p̄i -primary component of E ∗ = E1∗ ⊕ . . . ⊕ Es∗ . Hence φ(Ei ) = Ei∗ when
(p̄i ) = (pi ), and φ(Ei ) = Ej∗ , φ(Ej ) = Ei∗ when (p̄i ) = (pj ), so that φ(Ei ⊕ Ej ) = Ei∗ ⊕ Ej∗ =
(Ei ⊕ Ej )∗ ).
(b) In case 1, the second and third decomposition theorems hold. (Hint: You may repeat the proofs
of the symmetric case, pn e · v = e · p̄n v = ±e · pn v).
(c) In case 2, the pair of metrics in Ei ⊕ Ej is unique up to an automorphism. (Hint: If S̃, H̃ is other
pair of metrics defining the same module structure, S̃(T e, v) = H̃(e, v), and ϕ : E −∼
→ E ∗ is the
−1
polarity of S̃, then τ = φ ϕ : Ei → Ei is a module isomorphism such that S̃(ei , ej ) = τ ei · ej ;
hence τ ⊕ Id : Ei ⊕ Ej → Ei ⊕ Ej transforms S into S̃, and H into H̃).
(d) When k is algebraically closed, pairs of metrics symmetric and hemisymmetric (the first non
singular) are classifies by the elementary divisors of T . If α 6= 0, there are as many elementary
divisors (x − α)n as (x + α)n , and the number of elementary divisors equal to x2n is even.
If p = x − α, then p̄ = −(x + α) and we are in case 2. Consider the sum V ⊕ V̄ of two monogenous
submodules of annihilator (x − α)n and (x + α)n . Fix a base e, pe, . . . , pn−1 e in V . The image
by the polarity is a base of V̄ ∗ that, in the reverse order, is dual of a base ē, p̄ē, . . . , p̄n−1 ē in V̄ ,
(
i j i+j 1 when i + j = n − 1
(*) p e · p̄ ē = e · p̄ ē =
0 when i + j 6= n − 1

α when i + j = n − 1

i j i j i i+1 j
(*) p e ∗ p̄ ē = T p e · p̄ ē = (αp e + p e) · p̄ ē = 1 when i + j = n − 2

0 otherwise

If p = x, and p̄ = −x, the second decomposition theorem let us reduce to the case of a ho-
mogeneous module of annihilator xn . If n is even, the metric pn−1 e · v is hemisymmetric and
pn−1 e · e = 0, ∀e ∈ Fn ; but pn−1 e · ē = 1 for some ē ∈ Fn , and the monogenous submodules

V =< e, pe, . . . , pn−1 e >, V̄ =< ē, p̄ē, . . . , p̄n−1 ē > are totally isotropic, with null intersection,
and V ⊕ V̄ is non singular. The former calculation shows that the matrices of the metrics are
(*), with α = 0. Since E = (V ⊕ V̄ ) ⊥ (V ⊕ V̄ )⊥ , we conclude.
If n is odd, the metric pn−1 e · v is symmetric, and the argument of the pair of symmetric metrics
shows that the space is an orthogonal sum of monogenous submodules where the pair of metrics
is fully determined. In fact, by the third decomposition theorem, there is a base e, T e, . . . , T n−1 e
such that T n−1 e · e 6= 0 and, k being algebraically closed, you may assume that T n−1 e · e = 1,
(
(−1)j when i + j = n − 1
T i e · T j e = (−1)j T i+j e · e =
0 when i + j 6= n − 1
(
(−1)j when i + j = n − 2
T i e ∗ T j e = T i+1 e · T j e =
0 when i + j 6= n − 2
52. If S is symmetric and non singular, for any 2-covariant tensor S 0 we have
T e · v = S 0 (e, v) = e · T̄ v
for some endomorphisms T, T̄ : E → E. We have studied the cases T̄ = T and T̄ = −T ; but show that
the above decomposition theorems may be extended to any involution relating T̄ with T .
(a) If T is an isometry of S, then T e · v = e · T −1 v, so that E is a k[x, x−1 ]-module and the polarity
φ : E → E ∗ is semilinear, of automorphism p(x) 7→ p̄(x) = p(x−1 ).
(b) If S is hemisymmetric and non singular, then T̄ = T when S 0 is hemisymmetric, and T̄ = −T
when S 0 is symmetric.
(c) Since any 2-covariant tensor T2 decomposes uniquely as a sum T2 = S + H of a symmetric and a
hemisymmetric tensors, obtain a classification, when k is algebraically closed, of the 2-covariant
tensors with non singular symmetric or hemisymmetric component.
53. If S is a simple A-module, show that S ' A/m for some maximal ideal m of A.
54. Let A be a ring (eventually non-commutative). Use Zorn’s lemma to prove the following statements
about arbitrary A-modules:
P
(a) If N ⊂ M is a submodule and M = N + i∈I Si , where the submodules Si are simple, then
M = N ⊕ (⊕j∈J Sj for some subset J ⊆ I.
(b) Every semisimple module is a direct sum of simple modules.
(c) Every submodule of a semisimple module also is semisimple.
(d) Every semisimple module admits a unique decomposition as a direct sum (eventually infinite) of
homogeneous submodules of different types.
55. Show that any left ideal of EndD (E) is the set of all endomorphisms vanishing on a certain vector
subspace V ⊆ E, and that any right ideal of EndD (E) is the set of all endomorphisms with image
contained in a certain vector subspace V ⊆ E.
56. Prove that a ring A is semisimple if and only if it is artinian (strictly decreasing sequences of ideals are
finite) and any nilpotent ideal is null. (Hint: If I is a minimal ideal and I 2 6= 0, then I 2 = I, so that
Ib = I for some b ∈ I, and b = ab = a2 b for some a ∈ I. Show that a2 = a, so that A = Aa ⊕ A(1 − a).
Iterate the argument with A(1 − a), and conclude because A is artinian).
57. Let G be a group of automorphisms of a finite extension k → L. Prove that the natural k-linear map
L ⊗k k[G] → Endk (L) is an isomorphism if and only if k → L is a Galois extension of group G.
58. Clifford Algebras: Let q be a quadratic form on a n-dimensional k-vector space, char k 6= 2, and let
h , i : V × V → k be the corresponding symmetric bilinear map. The Clifford algebra of q is the
quotient of the tensor algebra T • V by the two-sided ideal generated by the elements v ⊗ v − q(v),
v ∈ V , and it is endowed with a canonical map V ,→ C(q).
(a) We have he, vi = 21 (e · v + v · e), for all e, v ∈ V .
502 EXERCISES
(b) Let A be a k-algebra. For any k-linear map f : V → A such that f (v)2 = q(v), there exists a
unique morphism of k-algebras φ : C(q) → A such that φ(v) = f (v), v ∈ V .
(c) C(q) admits an involutive
P automorphism sucht that P v 7→ −v for all v ∈ V , and an involutive
anti-automorphism a = λi1 ...ip vi1 . . . vip 7→ a = λi1 ...ip vip . . . vi1 .
(d) The natural map Λ• V ,→ T • V → C(q) is a linear isomorphism; hence dim C(q) = 2n , and any
element a ∈ C(q) defines p-vectors ap ∈ Λp V . (Hint: Consider an orthogonal basis).
(e) The metric may be extended to the Clifford algebra, ha, bi = (at b)0 , and hax, yi = hx, at yi.
59. Let A be the Clifford algebra of a Minkowski space V , of signature (1, 3). Consider the matrices

0 1 i 0 j 0 ij 0
v0 = , v1 = , v2 = , v3 =
1 0 0 −i 0 −j 0 −ij
and conclude that A = EndH E, where E = H ⊕ H is a right H-vector space, named space of spinors
(recall that H denotes the quaternions). Prove the following statements:

0 1
(a) The volume form Ω is Ω = v0 v1 v2 v3 = , and Ω2 = −1.
−1 0
(b) If Rtr is the real part of the trace, then (a)0 = 21 Rtr(a), and hv, v 0 i = (vv 0 )0 = 12 Rtr(vv 0 ).
(c) Up to a real factor, there exists a unique H-bilinear map h : E × E → H, in the sense that
h(e1 q1 , e2 q2 ) = q̄1 h(e1 , e2 )q2 , such that (ae) · e0 = e · (at e0 ). Moreover, e · u = u · e, and
0
0 1 0 1 q1
h= , e · e0 = h(e, e0 ) = (q̄1 , q̄2 ) = q̄2 q10 + q̄1 q20 .
1 0 1 0 q20
(d) The metric h defines a H-linear isomorphism Ē = E ∗ of left H-vector spaces, hence a R-linear
isomorphism E ⊗H Ē = E ⊗H E ∗ = EndH E = A = Λ• V , and we have a R-bilinear metric
H• : E × E → Λ• V such that H• (eq, u) = H• (e, uq̄). Hence we have a Λ• V -valued quadratic form
Q• (e) = H• (e, e) = e ⊗ (e·). In particular, any spinor e ∈ E defines a vector Q(e) = Q1 (e) ∈ V ,
so that any spinor field may be viewed as a tangent vector field, and
H• (e, u) = H• (u, e)t ,
Q• (eq) = |q|2 Q• (e),
Q2 (e) = Q3 (e) = 0,
Q1 (e) = Q1 (Ωe) = 21 [e ⊗ (e·) + (Ωe) ⊗ (Ωe·)],
Q0 (e) = 12 (e · e).
(e) If T is a H-linear endomorphism and T (e) = eα, then T (eq) = (eq)(q −1 αq), so that eq has proper
value q −1 αq. Now, the proper values of Ω are the quaternions ε of square −1, and the subspace
of proper vectors Sε = {(q, qε)} is a 2-dimensional vector space over the field R + Rε.
(f) Every spinor in Sε is isotropic for h, while Si · S−i ⊆ C and Si · Si ⊆ Cj.
(g) We have E = Si ⊕ S−i , and the elements of A of even degree preserve Si and S−i , while the
elements of odd degree interchange them.
(h) If e = s + s0 is the decomposition of a spinor as a sum of proper spinors of values i and −i, then
Q0 (e) = Re(s · s0 ),
Q4 (e) = Im(s · s0 )Ω,
|Q(e)|2 = |s · s0 |2 = Q0 (e)2 + Q4 (e)2 , where we see Q4 with values in R.
(i) If e 6= 0, then Q(e) 6= 0 always points to the future (changing the sign of h if necessary), and the
following conditions are equivalent:
i. Q(e) is an isotropic vector.
ii. Q0 (e) = Q4 (e) = 0.
iii. s · s0 = 0.
iv. e is a proper spinor, Ωe = eε.
COMMUTATIVE ALGEBRA 503
(j) A vector v ∈ V is isotropic if and only if Ker v = {e : ve = 0} 6= 0, and in such case Ker v = eH,
where Q(e) = v. A spinor e ∈ E is proper if and only if Ann e = {v ∈ V : ve = 0} 6= 0, and in
such case Ann e = RQ(e). Conclude that the projective space of the (R + Rε)-vector space Sε is
just the space of future light directions.
Using the following calculations in coordinates (where we see a pure quaternion xi + yj + zij as a
vector xv1 + yv2 + zv3 ), prove that two spinors e0 , e represent the same vector, Q(e) = Q(e0 ), if and
only if e0 = (a + bΩ)eq, where a2 + b2 = 1 = |q|.

q1 q̄2 q1 q̄1
Q• (e) = Q(q1 , q2 ) = (q1 e1 + q2 e2 ) ⊗ (q̄1 ω2 + q̄2 ω1 ) = ∈ EndH E,
q2 q̄2 q2 q̄1
Q0 (e) = 12 RtrQ• (e) = 21 (q1 q̄2 + q2 q̄1 ) ∈ R,
Q1 (e) = 21 [Q• (e) + ΩQ• (e)Ω] = 21 [(|q1 |2 + |q2 |2 )v0 + (q1 q̄2 − q2 q̄1 )] ∈ V,
Q4 (e) = Q• (e) − Q0 (e) − Q(e) = 12 (|q1 |2 − |q2 |2 )Ω ∈ Λ4 V.
7. Commutative Algebra
1. Boolean algebras: A Boolean algebra is a set A endowed with two associative and commutative
operations ∧, ∨ satisfying the following axioms (compare with definition in p. 219, where the operations
are denoted + and · respectively):
1. Existence of neutral elements: a ∧ 0 = a , a ∨ 1 = a.
2. Distributive: a ∨ (b ∧ c) = (a ∨ b) ∧ (a ∨ c) , a ∧ (b ∨ c) = (a ∧ b) ∨ (a ∧ c).
3. Absorption: a ∨ 0 = 0 , a ∧ 1 = 1.
4. Idempotence: a ∨ a = a , a ∧ a = a.
5. Complement: If a ∈ A, there is ā ∈ A such that ā ∧ a = 1 and ā ∨ a = 0.
The fundamental examples are the algebra P(X) of all subsets of a set X (where ∧ is the intersection
and ∨ is the union, so that 0 = X, 1 = ∅ and ā is the complement of a) and the statements of any
formalized logical theory including the logical operations and, or, no (identifying equivalent statements,
so that 0 are tautologies). Now, the implication a ⇒ b is defined to be ā ∨ b and the equivalence a ⇔ b
is defined to be (ā ∨ b) ∧ (a ∨ b̄). Prove the following statements:
(a) Any boolean algebra A is a ring with the operations a + b := (a ⇔ b), a · b := a ∨ b, and any
element is idempotent, a2 = a (such rings are named Boole rings).
Conversely, any Boole ring (A; +, ·) is a boolean algebra with the operations a ∧ b := a + b + ab,
a ∨ b := a · b, and ā := 1 + a. In particular, (a ⇒ b) = (1 + a)b.
(b) Any finitely generated ideal of a Boole ring is a principal ideal.
(c) Any Boole ring A has characteristic 2 and, if it is a domain, then A = F2 .
(d) Any prime ideal p of a Boole ring is a maximal ideal of residue field A/p = F2 .
(e) If A is a Boole ring, the points x ∈ Spec A correspond to the ring morphisms vx : A → F2 (the
coherent interpretations of the logical theory as true=0 or false=1 statements),
Spec A = Hom (A, F2 ).
(f) Any element f ∈ A defines a function f : X = Spec A → F2 , f (x) = vx (f ), vanishing just on
(f )0 . If we consider F2 with the discrete topology, these functions are continuous. So we obtain
any continuous function X → F2 , and we have a ring isomorphism A = C(X, F2 ).
(g) A is isomorphic to a ring of subsets of X, the ring of all closed open subsets.
Show that any ideal I of a Boole ring is formed by all the continuos functions vanishing on the closed
set (I)0 ; that is to say, given a family of axioms B ⊂ A, the ideal I generated by B is formed by all
the statements f which are true in any coherent interpretation satisfying the given axioms: v(f ) = 0
whenever v(b) = 0, ∀b ∈ B. At first, this fact does not mean that the statement f may be derived from
the given axioms and tautologies (the null element 0) only using the Modus Ponens; such statements
forming the least subset B 0 , containing B and 0, such that
504 EXERCISES
If a ∈ B 0 and (a ⇒ b) ∈ B 0 , then b ∈ B 0 .
Prove that the ideal of A generated by B contains B 0 (Hint: (a ⇒ b) = b + ab).
In fact B 0 is the ideal generated by B (if a statement is true in any coherent interpretation, it may be
obtained from the axioms and tautologies using the Modus Ponens):
(a) If a ∈ A and b ∈ B 0 , since b ⇒ (a ∨ b) is null, we have ab = (a ∨ b) ∈ B 0 .
(b) If a, b ∈ B 0 , since a ⇒ (b ⇒ (a ⇔ b)) is null, we have that b ⇒ (a ⇔ b) is in B 0 , and we conclude
that a + b = (a ⇔ b) ∈ B 0 .
2. Prove that any proper element of a noetherian domain is a product of irreducible elements.
3. If A is a noetherian ring, show that so is any localization AS .
4. If A is a noetherian ring, prove that (rad A)n = 0 for some n ∈ N.
Q
5. Show that the ring ∞ R = C(N) is not noetherian.
6. Prove that any surjective endomorphism f : M → M of a noetherian A-module is an isomorphism.
(Hint: Consider the submodules Ker f n ).
7. Let I be the annihilator of a noetherian A-module M . Show that there is an injective morphism
A/I → M ⊕ . . . ⊕ M and conclude that A/I is a noetherian ring.
8. Let {Ufi } be a basic open cover of Spec A. Prove that an A-module M is noetherian if and only if
Mfi is a noetherian Afi -module for any index i.
9. Prove that any noetherian ring A of dimension 0 has finite length, finite and discrete spectrum,
Spec A = {x1 , . . . , xn }, and that A = Ax1 ⊕ . . . ⊕ Axn .
10. Let A be a noetherian ring. Prove that finite length A-modules are just noetherian A-modules M
such that supp M is a finite set of closed points.
11. Let A 6= 0 be a ring. If there is an injective A-linear morphism Ar → An , prove that r ≤ n. (Hint:
If A is noetherian, localize at a minimal prime. In the general case, consider the noetherian subring
Z[aij ], where (aij ) is the matrix of the morphism).
12. Let I be an ideal of a noetherian ring A. If m = rad I is a maximal ideal, prove that I is a m-primary
ideal; hence any proper ideal containing some power mn is a m-primary ideal.
13. Let A = k[x, y], p = (x), m = (x, y). Show that I = m2 ∩ p is not a primary ideal, and that it admits
another reduced primary decomposition, I = (x2 , y) ∩ p. (Hint: xy ∈ I, x ∈
/ I, y ∈
/ rad I = p).
14. Let A be a noetherian UFD. If p ∈ A is irreducible, show that p = pA is a prime ideal, that the ideals
pn A are p-primary, and that these are the only p-primary ideals of A.
15. Let I = q1 ∩ . . . ∩ qr be a reduced primary decomposition of an ideal I of a noetherian ring A, and
put pi = rad qi . Once we fix an associated prime p, prove that
(a) The intersection of the primary components with radical contained in p is just Ip ∩ A, so that it
does not depend on the decomposition. Hence, the intersection of the primary components with
radical strictly contained in p neither depend on the decomposition.
(b) If I = q01 ∩ . . . ∩ q0r is another reduced primary decomposition, pi = rad q0i , then we also have
I = q1 ∩ . . . ∩ qi−1 ∩ q0i ∩ qi+1 ∩ . . . ∩ qr . (Hint: When p1 , . . . , pi are just the associated primes
contained in pi , then q1 ∩ . . . ∩ qi = q01 ∩ . . . ∩ q0i−1 ∩ q0i = q1 ∩ . . . ∩ qi−1 ∩ q0i ).
(c) If q001 , . . . , q00r are primary ideals, pi = rad q00i , each one appearing in some reduced primary decom-
position of I, then I = q001 ∩ . . . ∩ q00r .
(n) (n)
(d) If pn ⊆ qi (so that p = pi ) and we put qi := (I + pn )p ∩ A, then I = q1 ∩ . . . qi . . . ∩ qr .
(n )
(e) If ni is the first exponent such that qi i appears in some reduced primary decomposition of I,
(n ) (n )
then I = q1 1 ∩ . . . ∩ qr r .
16. Let B = ⊕n≥0 Bn be a graded ring, Bi · Bj ⊆ Bi+j , (in particular B0 is a subring). An ideal I ⊆ B
is homogeneous when I = ⊕n (Bn ∩ I). If B is a noetherian ring, prove that
(a) If p is a prime ideal, then ph = ⊕n (Bn ∩ p) is a prime ideal.

(b) Any minimal prime ideal is a homogeneous ideal.
(c) If q is a primary ideal, then qh = ⊕n (Bn ∩ q) is a primary ideal. (Hint: Assume qh = 0).
(d) Any homogeneous ideal I admits a homogeneous primary decomposition, and the associated
primes of I are homogeneous prime ideals.
(e) If any homogeneous element of positive degree is a zero divisor, then some associated prime
contains the irrelevant ideal ⊕n≥1 Bn .
17. Let us consider an action of a group G on a ring A; i.e., a group morphism G → Autrings (A). We say
that the ring of invariants AG = {a ∈ A : g · a = a, ∀g ∈ G} is stable under base change when, for any
ring morphism AG → B, we have that the natural morphism B → (A ⊗AG B)G is an isomorphism.
In such case, if A is a noetherian ring, prove that so is AG . (Hint: If I is an ideal of AG and we put
I 0 = AG ∩ IA, show that I = I 0 using the base change AG → AG /I).
Moreover, if A is a graded finitely generated algebra over a field k and G acts by homogeneous k-
automorphisms, conclude that AG is a finitely generated k-algebra (Hint: AG is a graded k-algebra
and the irrelevant ideal is finitely generated).
When G is a finite group of automorphisms of a Q-algebra, show that the algebra of invariants is
1
stable under base change (Hint: The arithmetic mean m : A → AG , m(a) = |G|
P
g ga).
b is the I-adic completion of a noetherian ring A, prove that Specm A
18. If A b = Specm (A/I).
19. Let M be a finitely generated module over a noetherian ring A. Prove that M = 0 if and only if M
has null completion at any point of Spec A.
20. Let (O, m) be a noetherian local ring. If t ∈ m and dim O/tO = dim O − 1, prove that
mult(O/tO) ≥ mult O.
21. Let I ⊆ m be an ideal of a noetherian local ring (O, m). If f¯ ∈ I r /I r+1 , r ≥ 1, is not a zero-divisor in
the graded ring GI O, prove that
(a) f ∈ I r is not a zero-divisor in O.
(b) f I n = I r+n ∩ f O for any n ≥ 0.
(c) GI¯Ō = (GI O)/(f¯), where I¯ is the image of I in Ō = O/f O.
22. If P2 (x) ∈ R[x] has degree 2 and no real root, prove that R[x]/(P2m ) ' C[t]/(tm ), and that the trace
metric (p. 145) of this R-algebra has rank 2 and index 1. (Hint: Use the proposition of p. 199, and
then consider the base 1, i, t, it, . . . , tm−1 , itm−1 ).
Show that the trace metric of R[x]/(x − a)m ' R[t]/(tm ) has rank 1 and sign +.
Prove that the rank of the trace metric Tr in R[x]/(P ) is the number of different roots of P (x), and
the index is half the number of imaginary roots. Conclude that all the roots of P (x) are real if and
only if Tr is non-negative, Tr(a, a) ≥ 0.
Show that the matrix of Tr in the base 1, x, . . . , xn−1 of R[x]/(P ), where n = deg P (x), is just (the
sums σr of powers of roots are given by the Newton and Girard formulae)
 
n σ1 . . . σn−1
 σ1 σ2 . . . σn 
Tr =   ...
.
... ... ... 
σn−1 σn . . . σ2n−2
23. Let A = C[x, y]/(y 2 − x) and let C = Spec A be the complex plane curve of equation x = y 2 . Prove
that the complement of the point x = 1, y = 1 is a basic open set Uf , and that the first projection
π : Uf → A1 (given by the morphism of C-algebras C[x] → Af , x 7→ x) is not a finite morphism, even
if it is a surjective closed and open map with finite discrete fibres.
24. Let G be a finite group of automorphisms of a finitely generated k-algebra A = k[ξ1 , . . . , ξn ]. Prove
that AG is a finitely generated k-algebra. G
Q (Hint: Consider the subalgebra B ⊆ A generated by the
coefficients of the polynomials Pi (x) = g (x − gξi )).
506 EXERCISES
25. Let A = k[ξ1 . . . , ξn ] be a finitely generated k-algebra. When k is infinite, show that some morphism
k[ξ10 . . . ξn−1
0
, ξn ] ,→ A is finite, where ξi0 = ξi − λi ξn , λi ∈ k (i.e. some linear projection Spec A → Ad
is finite).
26. If A is a finitely generated k-algebra and k → L is an extension, prove that dim A = dim AL .
27. If A, B are finitely generated k-algebras, prove that dim A ⊗k B = dim A + dim B.
28. Prove that any affine algebraic variety with a finite number of closed points has dimension 0.
29. Let Σ be the field of fractions of an integral finitely generated k-algebra A. Prove that the dimension
of A is the transcendence degree of Σ over k (the maximal number of algebraically independent
elements in Σ).
30. Let f : X → Y be a morphism between algebraic varieties. If C is an irreducible closed set in X, prove
that dim C ≥ dim f (C).
31. Let X = Spec A be an irreducible algebraic variety of dimension n. If f ∈ A is not invertible nor
nilpotent, show that any irreducible component of (f )0 has dimension n − 1.
32. Let X = Spec A be an irreducible algebraic variety. Prove that all maximal chains of irreducible closed
subsets have equal length. (Hint: If X ⊃ Y ⊃ . . . is a maximal chain, take f ∈ A vanishing on Y .
Prove that Y is an irreducible component of (f )0 and apply induction on dim X).
Prove that all irrefinable chains of irreducible closed subsets Y ⊃ . . . ⊃ Z, with fixed extremes, have
equal length, namely dim Y − dim Z.
If x is a closed point of X, conclude that dim X = dim Ax .
33. Let Y1 , Y2 be irreducible closed sets in an irreducible algebraic variety X. If Y is an irreducible

component of Y1 ∩ Y2 , show that codim Y ≤ codim Y1 + codim Y2 , where codim Z = dim X − dim Z.
34. Let f : X → Y be a morphism between irreducible algebraic varieties. If a closed point y ∈ Y is in

the image of f , prove that dim f −1 (y) ≥ dim X − dim Y.
35. Let Y, Z be two closed sets in an affine algebraic variety X = Spec A over an algebraically closed field
k. If all the rational points of Y are in Z, prove that Y ⊆ Z.
36. Let X = Spec A, Y = Spec B be affine algebraic varieties over an algebraically closed field k. If X
and Y are reduced (resp. connected, irreducible, integral) prove that so is X ×k Y .
P
(Hint: Take a basis {hi } of B as a k-vector space and prove that if i fi ⊗ hi ∈ A ⊗k B is nilpotent,
then so are the functions fi ∈ A. On the other hand, if X ×k Y = C1 ∪ C2 , where C1 and C2
are closed sets, and Yr stands for the topological subspace of all rational points of Y , prove that
Yi = {y ∈ Yr : X × y ⊆ Ci } is a closed set in Yr ).
37. Let k be a field and let A = ⊕n An = k[ξ1 , . . . , ξd ] be a graded k-algebra, deg ξi = 1. Prove that
dim A = dim Am , where m = (ξ1 , . . . , ξd ), and that dim A − 1 coincides with the degree of the Hilbert
polynomial H(n) = dimk An .
38. Prove that any maximal ideal of k[x1 , . . . , xn ] is generated by n elements.
39. If a system of polynomial equations p1 (xP

1 , . . . , xn ) = 0, . . . , pr (x1 , . . . , xn ) = 0 with real coefficients
has no complex solution, prove that 1 = i qi pi for some qi ∈ R[x1 , . . . , xn ].
40. Prove that any noetherian valuation ring is a discrete valuation ring.
41. Prove that any Dedekind domain with a finite number of maximal ideals is a principal ideal domain.
42. Prove that any ideal of a Dedekind domain is generated by two elements.
43. Even if R[x, y]/(x2 + y 2 + 1) is not a Euclidean ring (p. 475), prove that it is a principal ideal domain.
(Hint: Any imaginary point x = α, y = β is in some real line ax + by = c).
√
44. Let d be a square-free integer
h √ number.
i Prove that A = Z[ d ] is a Dedekind domain when d ≡ 2 or 3
(mod. 4), and that A = Z 1+2 d is a Dedekind domain when d ≡ 1 (mod. 4).
Put d = −19, so that A ' Z[x]/(x2 − x + 5). According to a fundamental result of number theory, any
√ group Pic(A) is represented by some ideal I such that |A/I| is bounded above
element of the Picard
by the square root 19 of the absolute value of the discriminant. Assuming this result, conclude that
A is a principal ideal domain (even if it is not a Euclidean ring, p. 475).
45. Let (O, m) be the local ring of an integral plane curve at a rational point x of multiplicity m. If
O → O1 is the quadratic transformation at x, show that mm−1 = mm−1 O1 .
46. Let A be a noetherian domain of dimension 1. Prove that l(A/f hA) = l(A/f A) + l(A/hA) whenever
f, h ∈ A are nonzero.
47. Let C1 , C2 be two plane curves over a field k of equations P1 (x, y) = 0, P2 (x, y) = 0. If they have no
common irreducible component, prove the following statements:
P
(a) dimk k[x, y]/(P1 , P2 ) = (C1 ∩ C2 )z · deg z.
z∈C1 ∩C2
(b) Let us introduce the homogenization Q̃(x0 , x1 , x2 ) = xd0 Q( xx01 , xx20 ) of any polynomial Q(x, y) of
degree d. We say that the curves C1 , C2 do not intersect at infinity when x0 = 0, x1 = 0, x2 = 0
is the unique solution (in arbitrary extensions of k) of the system

x0 = 0 
P̃1 (x0 , x1 , x2 ) = 0
P̃2 (x0 , x1 , x2 ) = 0

We put k[x0 , x1 , x2 ] = ⊕Hn , A = k[x0 , x1 , x2 ]/(P̃1 , P̃2 ) = ⊕n An , and we have exact sequences
(where φ(Q̃) = (P̃2 Q̃, −P̃1 Q̃), ψ(Q̃1 , Q̃2 ) = P̃1 Q̃1 + P̃2 Q̃2 )
φ ψ π
0 −→ Hn−d1 −d2 −−→ Hn−d2 ⊕ Hn−d1 −−→ Hn −−→ An −→ 0.
(c) dimk An = 2 − n−d21 +2 − n−d22 +2 + n−d1 −d
n+2 2 +2

2 = d1 d2 , when n ≥ d1 + d2 .
(d) We have an isomorphism (the union is considered in the ring Ax0 )

k[x, y]/(P1 , P2 ) = n An x−n x1 x2
S
0 , Q(x, y) 7→ Q x0 , x0 .
(Hint: If Q( xx10 , xx20 ) = 0, then Q̃(x0 , x1 , x2 ) = xd0 Q( xx10 , xx20 ) = 0 in Ax0 ; hence xr0 Q̃(x0 , x1 , x2 ) is
null in A, so that xr0 Q̃(x0 , x1 , x2 ) = Q̃1 (x0 , x1 , x2 )P̃1 (x0 , x1 , x2 ) + Q̃2 (x0 , x1 , x2 )P̃2 (x0 , x1 , x2 ) and
Q(x, y) = Q̃1 (1, x, y)P1 + Q̃2 (1, x, y)P2 ).
(e) dimk k[x, y]/(P1 , P2 ) = dimk (An x−n
0 ) ≤ dimk An = d1 d2 , n 0.
(f) If the curves do not intersect at infinity, then the radical of k[x0 , x1 , x2 ]/(x0 , P̃1 , P̃2 ) is the maximal
ideal (x0 , x1 , x2 ). It follows that An is null in k[x0 , x1 , x2 ]/(x0 , P̃1 , P̃2 ) = A/x0 A when n 0,
and we have epimorphisms x0 : An → An+1 ; hence isomorphisms.
No element of An is annihilated by a power of x0 , and we have isomorphisms An → An x−n 0 ,
n 0, and dimk k[x, y]/(P1 , P2 ) = dimk An = d1 d2 .
Obtain Bézout’s theorem: The number of common points, counted with degree and multiplicity, is
≤ d1 d2 (and the equality holds when both curves do not intersect at infinity)
P
(C1 ∩ C2 )z · deg z ≤ d1 d2 .
z∈C1 ∩C2
48. A singular point of a curve is a cuspidal point if it has a unique analytical branch. Let us consider
a cuspidal point x of multiplicity m over an algebraically closed field. After a finite number h of
quadratic transformations, the multiplicity will decrease, m0 < m. Prove that the intersection number
with a simple curve R is (C ∩ R)x = km if both curves separate after k ≤ h quadratic transformations;
but, if they do not separate after h quadratic transformations, then they intersect transversally and
(C ∩ R)x = hm + m0 . Show that there is a maximal contact hm + m0 with simple curves.
508 EXERCISES
49. Let C = Spec A be a curve and let m ⊂ A the maximal ideal of a point x ∈ C. Fix f ∈ m = (f1 , . . . , fd )
of minimal value at any discrete valuation centered at x, and without zeros at any other singular point
of C. The functions fj /f may be not integral over A, since f may have other zeros z1 , . . . , zr on
C; but take g ∈ A not vanishing at any singular point of C, and vanishing at z1 , . . . , zr with equal
multiplicity than f . Prove that the functions fj g/f are integral over A,
h i
A −→ A1 = A ff1 g , . . . , ffd g ⊆ Ā
If A = A1 , show that x is a simple point. (Hint: If fj g ∈ f A, then mAx = f Ax since g ∈ A∗x ).

50. If M is a finitely generated module over a domain A, prove the existence of a non empty basic open
set Uf such that Mf is a free Af -module.
51. Let M be a finitely generated A-module. If the natural morphism I ⊗A M → M is injective for any
finitely generated ideal I ⊂ A, show that M is a flat A-module.
52. Prove that any A-module is an inductive limit of finitely presented A-modules. (Hint: Consider the
set of all finite families m1 , . . . , mn ∈ M and relations (a11 , . . . , a1n ), . . . , (ar1 , . . . , arn ) ∈ An ).
Conclude that HomA (M, −) preserves inductive limits if and only if M is finitely presented.
53. Prove that a ring morphism A → B is faithfully flat if and only if it is a flat morphism and the map
Spec B → Spec A is surjective.
54. Let M be a finitely generated A-module. If any point x ∈ Spec A has a basic neighborhood Uf such
that Mf is a free Af -module, prove that M is a projective A-module.
S
55. If Spec A = i Ui is a finite basic open cover, prove that A → ⊕i AUi is faithfully flat.
56. Show that any finite flat morphism A ,→ B is faithfully flat. Hence, if A is a Dedekind domain and B
is integral, any finite injective morphism A → B is faithfully flat.
57. If O is a local noetherian ring, prove that O → O
b is a faithfully flat morphism.
8. Topology
1. In a lattice semiring A, we put a ≤ b whenever aA ⊆ bA (i.e. ab = a). Show that so we obtain an order
relation on A, and that A is a distributive lattice with first and last element. In fact, max(a, b) = a + b
and min(a, b) = ab.
2. Let Y be a subspace of a topological space X. Show that the subset S ⊆ A(X) of all closed sets which
does not intersect Y is a multiplicative system of A(X) and that A(X)S is just the semiring of germs
of closed sets along Y
3. If B is a base of closed sets of a topological space X and b ∈ B, prove that S = {1, b} is a multiplicative
system of B and that BS is isomorphic to the basis of X − b induced by B.
4. Prove that the non zero divisors of a semiring A form a multiplicative system of A.
Show that a closed subset Y of X is not a zero divisor in AX if and only if Y has empty interior.
5. Let P(X) be the semiring of all subsets of a set X (i.e. P(X) is just the semiring of closed sets in the
discrete space Xdis ). Show that the elements of X correspond to the principal prime ideals of P(X).
Conclude that any semiring isomorphism P(X) → P(Y ) is induced by a bijection Y → X.
6. If A is a semiring, prove that Homtop (X, Spec A) = Homsrings (A, AX ) for any topological space X.
7. Show that the semiring K[x] := {0, x, 1}, with the obvious operations (1 + x = 1, x2 = x) has the
universal property of a ”ring of polynomials”
Homsrings (K[x], A) = A
and conclude that Spec K[x] = K, where K = {0, 1} is considered as a topological space with a unique
closed point 0, and a dense point 1. Moreover, the ”affine space” Kn has the universal property
Homtop (X, Kn ) = AnX , and the open subset Kn − {0} represents the functor
Homtop (X, Kn − {0}) = {(f1 , . . . , fn ) ∈ AnX : f1 + . . . + fn = 1} = {open covers X = U1 ∪ . . . ∪ Un }.
TOPOLOGY 509
8. Describe the elements of the finite semiring K[x1 , . . . , xn ] := K[x]⊗ . n. . ⊗K[x] and the natural homeo-
morphism Kn = Spec K[x1 , . . . , xn ].
9. Let A → B be a semiring morphism and let p be a prime ideal of A. Prove that the fibre of
f ∗ : Spec B → Spec A over the point defined by p is Spec Bp /pBp , where Bp := Bf (A−p) .
10. If f : A → B is an injective semiring morphism prove that the continuous map f ∗ : Spec B → Spec A
is surjective. (Hint: Show that Ap → Bp also is injective, and in the case of a maximal ideal m, prove
that mB = B is contradictory).
11. Let I = [a, b] be a closed interval in the real line and let B be the base of all finite unions of points and
closed subintervals. Let Bx be the localization of B at the maximal ideal of all closed sets containing
a given point x ∈ I. If a < x < b, show that Bx = {0, 1, x, x+ , x− } (where x, x+ , x− are the germs
of x, [x, b] and [a, x] respectively) and that Bx has a unique maximal ideal (x) and two prime ideals
(x+ ) and (x− ).
If x = a or x = b, show that Bx has a maximal ideal and a prime ideal.
12. Show that any T0 topological space is a dense subspace of a projective limit of finite topological spaces.
13. Prove that inductive limits and tensor products of boolean algebras are boolean algebras.
14. If some base of the topology of a T0 space X is a Boolean algebra, show that X is totally disconnected
(any connected subset is a point), and if the topology AX is a Boolean algebra, show that X is discrete.
15. Let A be a semiring. If Spec A = (a)0 q (b)0 , prove that A = Aa × Ab = (A/bA) × (A/aA).
16. If Y → X is a closed continuous map, prove that Spec AY → Spec AX also is a closed map.
17. Prove that a topological space X is T1 if and only if X ⊆ Specm (AX ).
18. Prove that any base of closed sets of a compact separated space is a normal semiring.
19. Let B be a basis of a locally compact space X. Prove that X ∩ Specm B is open in Specm B.
Conclude that a normal T1 -space X is locally compact if and only if X is open in Specm AX .
20. If M is a noetherian A-module, prove that supp M is a noetherian subspace of Spec A.
21. A topological space X is said to be sober if any irreducible closed subset is the closure of a unique
point. Prove that the functor Spec defines an equivalence of the (opposite) category of noetherian
semirings with the category of noetherian sober spaces.
Conclude that any projective limit of noetherian sober spaces is a compact space.
22. Let A be a semiring. Prove than an open set U ⊆ Spec A is basic, U = Ua , if and only if it is compact.
Conclude that semirings with homeomorphic spectra are isomorphic.
23. Prove that a topological space is a spectral space if and only if it is sober and the closed sets with
compact complement form a base of the topology.
If A is a ring, conclude that Spec A = Spec B for some semiring B.
24. If X is a finite order, and X ∗ is the dual order, show that |X| and |X ∗ | are homeomorphic.
25. Let A be a ring. If any prime ideal of A is contained in a unique maximal ideal, prove that Specm A
is a separated space. (Hint: (Ax )y = 0 when x, y are different closed points).
If Specm A is a separated space, and the intersection of all maximal ideals of A is 0, prove that any
prime ideal of A is contained in a unique maximal ideal.
26. Let A be a semiring such that the intersection of all maximal ideals is the zero ideal. If Specm A is a
separated space, prove that A is normal.
27. Prove that the spectrum of any ring of null dimension is a Stone space.
Let k be a field. If X is a Stone space, show the existence of a k-algebra A of dimension 0 such
that X = Spec A. (Hint: K is a projective limite of discrete finite spaces, spectra of trivial finite
k-algebras).
510 EXERCISES
28. Let X be a locally compact non-compact separated space. Prove that compact sets and complements
of relatively compact open sets form a base B of X, such that the complements of the relatively
compact open sets form a maximal ideal, and X → X ∗ = Specm B is the one point compactification
of X. Moreover, for any separated compactification X → K there is a unique continuous extension
f : K → X ∗.
29. Prove that SpecR C ∗ (X) = Specm C ∗ (X).
30. Show that any non-empty open set in Spec C ∗ (X) intersects Specm C ∗ (X).
Prove that any two disjoint closed sets in Specm C ∗ (X) have disjoint neighborhoods in Spec C ∗ (X).
Prove that any prime ideal of C ∗ (X) is contained in a unique maximal ideal, and that the retraction
Spec C ∗ (X) → Specm C ∗ (X) so obtained is continuous.
31. Show that any real maximal ideal of C(R) is the ideal of all continuous functions vanishing at a certain
point a ∈ R. Conclude the existence of non-real maximal ideals in C(R).
32. Prove that any two disjoint closed sets in Specm C(X) have disjoint neighborhoods in Spec C(X). (Hint:
If f1 + f2 = 1 for two non-negative functions, and you put h = f1 − f2 , h1 = |h| − h, h2 = |h| + h,
then h1 h2 = 0, f1 + h1 > 0 and f2 + h2 > 0).
Prove that any prime ideal of C(X) is contained in a unique maximal ideal, and that the retraction
Spec C(X) → Specm C(X) so obtained is continuous.
33. Prove that any projective limit of finite non-empty sets is non-empty.
34. Let X be a metrizable space without isolated points. Considering a sequence (xn ) with a unique limit
point x, show that we have a subspace X1 = {x, x1 , x2 , . . .} such that dim A(X1 ) = 1. Now X1 is the
set of limit points of a sequence (yn ), and we put X2 = X1 ∪ {y1 , y2 , . . .}. Show that dim A(X2 ) = 2,
and so on. Conclude that dim AX = ∞.
35. Let C(X) be the ring of real continuous functions on a metrizable space X. Prove the following
statements:
(a) If S ⊂ C(X) is the multiplicative system of all continuous functions without zeros on a given
open subset U ⊆ X, we have a canonical ring isomorphism C(X)S = C(U ).
(b) If m ⊂ C(X) is the maximal ideal of all continuous functions vanishing at a given point x ∈ X,
then C(X)m is isomorphic to the ring O of germs at x of continuous functions. Hence, the
localization morphism C(X) → C(X)m is surjective, and the kernel is the ideal n of all continuous
functions with null germ at x.
(c) m = m2 .
(d) If x is not an isolated point, then mO is not a finitely generated ideal, so that O and C(X) are
not noetherian rings. Moreover n 6= m, and Ox is a finitely generated flat C(X)-module, but not
a projective C(X)-module.
T
(e) Let a ⊆ C(X) be an ideal. The C(X)-module C(X)/a is flat if and only if a = y∈Y ny for some
closed subset Y ⊆ X. (Hint: Take Y = (a)0 ∩ X. If C(X)/a is flat, then aOx = 0 when x ∈ Y ,
while aOx = Ox when x ∈ / Y ).
(f) Any R-derivation D : C(X) → C(X)/m = R is null.
(g) Any R-derivation D : C(X) → C(X) is null.
(h) If a germ f ∈ O is not constant, there is a prime ideal p ⊂ O such that f¯ ∈ κ(p) = (O/p)p is
transcendent over R, and there is a R-derivation D : O → κ(p) such that Df 6= 0.
(i) The kernel of the differential d : O −→ ΩO/R is defined by the constant germs.
36. Let βXdis = Specm P(X) = Spec P(X) be the space of ultrafilters in a set X, and B a base (stable
under finite unions and intersections) of closed sets of a topology on X. Show that any ultrafilter M
defines a prime ideal p = B ∩ M in B and that so we obtain a continuous surjective map
Spec P(X) → Spec B.
Moreover M converges to a point x ∈ X if and only if p ⊆ px := {b ∈ B : x ∈ b}.
TOPOLOGY 511
Let I ⊂ B be a proper ideal. If the filter IPX converges to x ∈ X, prove that I ⊆ px . (Hint: If b ∈
/ px ,
then the complement of b contains some a ∈ I; hence a + b = 1, and b ∈ / I).
If m ⊂ B is a maximal ideal, prove that mPX converges to x ∈ X if and only if m = px .
37. Show that any topological group G (a group endowed with a topology such that the product
G × G → G and the inverse G → G are continuous) is a uniform space, where a base of entourages is
defined by the sets Ũ = {(x, y) ∈ G × G : xy −1 ∈ U }, where U runs over the neighborhoods of 1 in G.
38. Prove that a Cauchy filter F in a separated uniform space X can not converge to two points x 6= y.
39. Let X be a uniform space. Show that the filter of neighborhoods Nx of a point x ∈ X (hence any
convergent filter) is a Cauchy filter. (Hint: V (x) × V (x) ⊆ U when V is symmetric and V 2 ⊆ U ).
If a Cauchy ideal I ⊂ AX has a zero x ∈ X, prove that the Cauchy filter IPX converges to x. Conclude
that a point x ∈ X is a zero of a Cauchy ideal I ⊂ AX if and only if IPX converges to x. (Hint: If
ε ε ε
A ∈ I is -small, then x ∈ A, so that A ⊆ (x), and (x) ∈ IPX ).
40. Put S1 = {z ∈ C : |z| = 1}, and X = S1 − {1}. Let I+ (resp. I− ) be the ideal of AX of all closed
sets in X containing some arc {eti ; t ∈ (0, ε)}, ε > 0 (resp. ε < 0), and let mi be a maximal ideal of
AX containing Ii . Prove that m+ and m− are different Cauchy maximal ideals containing a common
ε -small set for every entourage ε of X. Conclude that the topology of the space X̃ of all Cauchy
maximal ideals of AX is not the Zariski topology.
41. Let Ē be the completion of a topological vector space (E, p). Show that the addition, E × E → E and
the products λ × E → E, λ ∈ K, are uniformly continuous, so that they extend to the completion,
and the maps Ē × Ē → Ē, K × Ē → Ē, define a structure of topological vector space on Ē.
42. Show that any locally compact space X is a k-space (it is the inductive limit of its compact subspaces).
If any point of a topological space X has a countable base of neighborhoods, show that X is a k-space.
(Hint: If Y ∩ K is a closed set in K for any compact set K ⊆ X and yn → x, then K = {x, y1 , y2 , . . .}
is compact, so that x ∈ Y ).
ε
43. Let X be a topological space, and Y a uniform space. If K ⊆ X is a compact set and is an entourage
ε ε
in Y , let W (K, ) be the set of pairs (f, h) in HomSets (X, Y ) such that f (x), h(x) ∈ , ∀x ∈ K.
ε
Prove that the sets W (K, ) form a base of entourages for a uniformity, the compact-convergence
uniformity: HomSets (X, Y )c . When we consider X with the discrete
Q topology, we obtain the weak or
pointwise convergence uniformity. Show that HomSets (X, Y )w = X Y , with the product uniformity.
When we consider X with the trivial topology, we obtain HomSets (X, Y )u , the set of all maps with
the uniformity of uniform convergence on X.
A family F ⊆ HomSets (X, Y ) is equicontinuous ε
if, given a entourage of Y , any point p ∈ X has a

ε
neighborhood Up in X such that f (x), f (p) ∈ , ∀x ∈ Up , f ∈ F. Then prove that the closure F̄ in
HomSets (X, Y )w also is equicontinuous, and that the compact and weak convergence induce the same
uniformity on F. Conclude that F̄ also is the closure of F in C(X, Y )c .
Prove Ascoli’s Theorem: If F ⊆ HomSets (X, Y ) is equicontinuous and F(x) = {f (x); f ∈ F} is
precompact (resp. has compact closure in Y ) for any point x ∈ X, then F is precompact (resp. has
compact closure) in C(X, Y )c .
If K is a compact separated space, prove that any compact family F ⊂ C(K, Y ) is equicontinuous.
ε
(Hint: Given a point p ∈ K and an entourage in Y , there is a finite cover of F by sets W (Ki , ), ε
where Ki are compact neighborhoods of p).
Prove that F ⊆ HomSets (X, Y ) is equicontinuous if and only if the map δ : X → HomSets (F, Y )u ,
δx (f ) = f (x), is continuous. When X is a k-space, conclude that F is equicontinuous if and only if so
is the restriction F|K to any compact set K ⊆ X.
If X is a separated k-space, prove the converse of Ascoli’s theorem: If a family F ⊆ C(X, Y )c has
compact closure (resp. is precompact) then F is equicontinuous and F(x) has compact closure (resp.
is precompact) in Y , ∀x ∈ X.
44. Let X be a topological space, and C(X, Y ) the set of continuous maps to a complete uniform space Y ,
with the compact-convergence uniformity. Prove that C(X, Y ) is complete when X is compact; hence
also when X is a k-space.
512 EXERCISES
45. Let S be a smooth manifold. If π : X → S is a covering, prove the existence of a unique structure of
smooth manifold on X such that π is a local diffeomorphism.
46. Following the proof of ex. 106 in p. 488, prove Grothendieck’s characterization of the forgetful functor
G-Sets Sets on the category of G-sets:
Let F : C Sets be a representable covariant functor, F (Y ) = HomC (X, Y ). Then the group G :=
AutC (X)op naturally acts on the sets F (Y ), and the induced functor C G-Sets is an equivalence
of categories when the following conditions hold:
(a) The category C admits coproducts and quotients by groups, and F preserves them.
(b) The functor F is conservative.
(c) Any endomorphism of F is an automorphism: G = F (X).
Obtain an alternative proof of Galois theorem for coverings of a connected and locally connected space.
47. Let S be a connected and locally connected topological space. If G is a group of automorphisms of a
connected covering P → S and P/G = S, show that it is a Galois covering of group G.
48. Let S be a locally connected space with a countable base of open sets. If π : X → S is a connected
covering, prove that S admits a countable base {Un } of connected open sets where π is trivial, and
that the family B of connected components of the open sets π −1 (Un ) is a base of the topology of X.
If V ∈ B, show that {V 0 ∈ B : V 0 ∩V 6= ∅} is countable. (Hint: Otherwise, V intersects an uncountable
family of disjoint open sets of some π −1 (Un )).
Fix an open set V0 ∈ B, and prove that the family B 0 of open sets V ∈ B such that there is a finite
sequence V0 , V1 , . . . , Vn = V with Vi−1 ∩ Vi 6= ∅ for any i = 1, . . . , n is countable.
Prove that V ∈B0 V is a closed open set, so that X = V ∈B0 V and B 0 = B.
S S
Conclude that X admits a countable base of open sets and π has countable fibres.
49. Show that any continuous map f : [0, 1] → [0, 1] has a fixed point.
50. Prove that any continuous map f : Sn → R, n ≥ 1, coincides on two antipodal points.
51. Prove that any non-surjective continuous map f : S2 → S2 has a fixed point.
52. Prove that any 3 × 3 matrix with positive coefficients has a proper vector with positive coordinates
(and positive proper value).
53. Let S2 = U1 ∪ U2 ∪ U3 be an open cover of a sphere. Prove that some open set Ui contains two
antipodal points. (Hint: Consider a subordinated partition of unity (f1 , f2 , f3 ), so that the image of
(f1 , f2 , f3 ) : S2 → R3 is in the plane x + y + z = 1).
Let S2 = Y1 ∪ Y2 ∪ Y3 be a closed cover of a sphere. Prove that some closed set Yi contains two
antipodal points. (Hint: Consider the map (f1 , f2 ) : S2 → R2 , fi (x) = d(x, Yi )).
54. Given two bounded measurable sets A1 , A2 in a plane, prove the existence of a line dividing both in
two sets of equal area. (Hint: Parametrize the half-planes P by a sphere S2 , and consider the map
(f1 , f2 ) : S2 → R2 , fi (P ) = area of Ai ∩ P ).
55. Show that the fundamental group of a 1-dimensional connected polyhedron with v vertices and w
wedges is a free6 group on w − v + 1 elements.
Conclude that any subgroup of a free group Ln of index m is a free group on nm − m + 1 elements.
56. Show that the fundamental group of a torus τg of genus g is a free group on 2g generators.
57. Let X, Y be topological manifolds of dimension n ≥ 3. Show that the fundamental group of the
connected sum X#Y is just the coproduct π1 (X, p) ∗ π1 (Y, q).
Conclude the existence of a connected compact manifold X of dimension n with fundamental group a
free group on a given finite number of generators.
6
The free group Ln on n elements is the coproduct, in the category of groups, of n infinite cyclic groups.
ANALYSIS III 513
9. Analysis III
1. Prove that a smooth manifold X is compact if and only if X = Specm C ∞ (X).
2. Let X be a smooth manifold and let m be the maximal ideal of C ∞ (X) defined by a point p ∈ X.
Prove that C ∞ (X)m is isomorphic to the ring of germs at p of smooth functions.
3. Let O be the local ring of germs at x = 0 of smooth functions on R. If I is the ideal of O defined
by the germs with null Taylor expansion at x = 0, prove that xI = I, and conclude that O is not a
noetherian ring. Conclude that C ∞ (Rn ), n ≥ 1, is not noetherian.
4. Prove that the ideal n of C ∞ (R), defined by the smooth functions with null germ at x = 0, is not
finitely generated.
5. Let U ⊆ Rn be an open set and let p = (a1 , . . . , an ) ∈ U . Prove the following statements:
(a) The diagonal ideal ∆ = {f (x, y) ∈ C ∞ (U × U ) : f (x, x) = 0} is ∆ = (y1 − x1 , . . . , yn − xn ).

(b) The maximal ideal mp = {f (x) ∈ C ∞ (U ) : f (p) = 0} of p is mp = (x1 − a1 , . . . , xn − an ).
(Hint: Using a partition of unity, you may assume that U is convex).
6. If X is a topological space, prove that

p no element
p of C(X)pis irreducible. (Hint: For any real-valued
p
function f we have f = f+ − f− = ( f+ + f− )( f+ − f− ), where f+ = max(f, 0)).
If O is the ring of germs at the origin p of smooth functions on R2 , show that the coordinate function
x1 is irreducible; while the ideal (x1 ) is not prime. (Hint: dp x1 6= 0).
7. Show
R that in Rn there is a sequence of functions ρm ≥ 0 of class C ∞ with compact support, such that
ρ (x)dx = 1 for any index m, and, given ε > 0, we have supp ρm ⊆ Bε (0) when m 0. Conclude
Rn m
that any continuous function f on Rn with compact support is a uniform limit of functions ρm ∗ f of
class C ∞ . Moreover, given ε > 0, we have supp (ρm ∗ f ) ⊆ (supp f ) + Bε (0) when m 0.
8. Let D be a vector field on a smooth manifold X. Given a topological space Λ, and continuous maps
a : Λ → X, t0 : Λ → R, prove the existence of an open set U ⊆ Λ × R and a continuous family of
solutions ϕλ (t) : U → X such that ϕλ (t0 (λ)) = a(λ). Moreover, if Λ is a smooth manifold, then ϕ is
of class C m when so are a(λ), and t0 (λ). (Hint: Put ϕλ (t) = τ (t − t0 (λ), a(λ)), where τ is the flow).
9. Prove that any vector field i fi ∂i on Rn with bounded components fi is complete.

P
If D is a vector field on Rn , show that f D is complete for some positive function f ∈ C ∞ (Rn ).
If D is a vector field on a smooth closed submanifold X ⊂ Rn , conclude that f D is complete for some
positive function f ∈ C ∞ (X).
10. Let X be a smooth manifold. Show that any vector field D on an open set U ⊂ X is D = D∗ /h, where
D∗ is a vector field on X (vanishing on X − U ) and h ≥ 0 has no zeros onPU . (Hint: If KPis a compact
set contained in a coordinate neighborhood, we have the seminorms k i fi ∂i kK,r = i kfi kK,r on
the module of vector fields on X, so that we may repeat the proof of p. 251).
When X is compact, prove that there is a positive smooth function h ∈ C ∞ (U ) such that hD is
complete. (Hint: D∗ is complete, and integral curves of points of U do not reach X − U ).
11. If τt (x1 , . . . , xn ) = (f1 (t, x), . . . , fn (t, x)) is a local uniparametric group in Rn , show that the infinites-
imal generator is D = (∂t f1 |t=0 ) ∂x1 + . . . + (∂t fn |t=0 ) ∂xn .
Obtain the infinitesimal generator of the most common uniparametric groups of R2 :
Translations τt (x, y) = (x + at, y + bt) D = a∂x + b∂y

cos αt − sin αt x
Rotations τt (x, y) = D = α(−y∂x + x∂y )
sin αt cos αt y
Homotheties τt (x, y) = (eat x, eat y) D = a(x∂x + y∂y )
At
P P
Linear maps τt X = e X D = i ( j aij xj )∂i
514 EXERCISES
12. Let X be a totally ordered set, and consider the order topology generated by the open intervals
(a, b) := {x ∈ X : a < x < b} , (a, ∞) := {x ∈ X : a < x} , (−∞, b) := {x ∈ X : x < b}.
Prove that the open intervals define a base of the order topology.
Let us consider the first uncountable ordinal ω1 , and the lexicographical order on the long ray
X = ω1 × [0, 1) = {λi ; λ ∈ [0, 1), i ∈ ω1 . Prove the following statements:
(a) X does not admit a countable base of open sets. (Hint: Find an uncountable disjoint family of
open sets).

(b) If 0 < i ∈ ω1 then there is an order-preserving homeomorphism 00 , 0i ) ' [0, 1). (Hint: Let
i ∈ ω1 be the first index where the result fails. If i = j + 1, then [00 , 0i ) = [00 , 0j ) ∪ [0j , 0j+1 ) =
[0, 1) ∪ [1, 2) = [0, 2). Otherwise i is the limit of a sequence 0 = i0 < i1 < i2 < . . ., then show
that [0in , 0in+1 ) = [n, n + 1) and conclude that [00 , 0i ) = [0, ∞). Absurd).
(c) For any b ∈ X there is an order-preserving homeomorphism [00 , b) = [0, 1), and for any a, b ∈ X,
a < b, there are order-preserving homeomorphisms (a, b) = (0, 1) and [a, b] = [0, 1].
(d) X is connected and path-connected.
(e) The long line L = X − {00 } is a connected separated topological manifold, but not σ-compact.
13. Let X be a connected and separated smooth manifold (eventually not σ-compact). If there is a smooth
vector field D on X (a tangent vector Dx at each point x ∈ X such that (Df )(x) := Dx f is a smooth
function for any f ∈ C ∞ (U )) without zeroes, prove that X is σ-compact. (Hint: Show that the
maximal integral curves of D define open subsets of X).
Prove that any connected and separated riemannian manifold (X, g) is in fact σ-compact. (Hint:
Show that the maximal geodesic lines define open subsets of X).
14. Prove that i[D,D0 ] = [DL , iD0 ] on p-forms: i[D,D0 ] ωp = DL (iD0 ωp ) − iD0 (DL ωp ).
15. Let τt , τ̄t be the flows of two vector fields D, D̄ on a smooth manifold X and let p ∈ X. Prove that
the curve γ : I → X, γ(t) = τ̄−t τ−t τ̄t τt (p), is well-defined
on an open interval I = (−ε, ε), and that,
f γ(t) −f (p)
for any smooth function f on X,we have lim t2 = [D, D̄]p f .
t→0
16. Let X be a smooth manifold. Prove that any homogeneous derivation of the algebra Ω• (X) of differ-
ential forms is a Lie derivative DL for a unique vector field D on X.
17. Show that the vector fields D on R3 such that DL (dx2 + dy 2 + dz 2 ) = 0 are spanned by the vector
fields ∂x , ∂y , ∂z , x∂y − y∂x , x∂z − z∂x and y∂z − z∂y .
18. Let ω1 , . . . , ωr be 1-forms on a smooth manifold X, linearly independent at any point. Show that a
p-form Ωp belongs to the ideal of Λ• (X) spanned by ω1 , . . . , ωr if and only if Ωp ∧ ω1 ∧ . . . ∧ ωr = 0.
19. Let ω2 be a closed 2-form. If the radical {D ∈ T : iD ω2 = 0} is a distribution (i.e., it has constant
rank), show that it is integrable.
20. First Order Partial Differential Equations: Given a smooth function F (x1 , . . . , xn , z, p1 , . . . , pn ), a
classical solution of the partial differential equation F = 0 is a smooth function z(x1 , . . . , xn ) such
that F (x1 , . . . , xn , z, ∂1 z, . . . , ∂n z) = 0. Let us study the problem on germs at a fixed point of R2n+1 .
Consider the 1-form ω = dz − p1 dx1 − . . . − pn dxn , and let us call (generalized) solution any germ
of submanifold of dimension n where F and ω specialize to 0. Prove that classical solutions are just
solutions where x1 , . . . , xn define a local coordinate system.
Now assume that dF and ω are linearly independent, and consider the Pfaff system P = hω, dF i. Show
that the characteristic system of P is generated by the following vector field (so that P is projectable
to the subring B of first integrals of DF ):
n n n
!
X ∂F ∂ X ∂F ∂F ∂ X ∂F ∂
DF = − pi + + pi ·
i=1
∂pi ∂xi i=1 ∂z ∂xi ∂pi i=1
∂pi ∂z
If we are at the origin of R2n+1 , the hypersurface xn = 0 is not tangent to the characteristic field,
∂F
∂pn (0) 6= 0, and we choose XP
1 , . . . , Xn−1 , Z, P1 , . . . , Pn ∈ B specializing at xn = 0 into the coordinates,
prove that hω, dF i = hdZ − i Pi dXi , dF i. (Hint: F ∈ B and the proof of the projection theorem).
ANALYSIS III 515
Conclude that the following equations (whenever they define a germ of submanifold of dimension n)

 F = 0,

Z = f (X1 , . . . , Xn−1 ),

Pi = (∂i f )(X1 , . . . , Xn−1 ), i = 1, . . . , n − 1

define a solution coinciding on xn = 0 with a given smooth function f (x1 , . . . , xn−1 ). Prove that other
solutions of the equation F = 0 are defined by the equations (λ1 , . . . , λn ∈ R)

F =0

Z = λn

Xi = λi i = 1, . . . , n − 1

Lagrange-Charpit Method: When n = 2, let G be a first integral of the characteristic field DF ,

and assume that Q = hω, dF, dGi is a Pfaff system of rank 3. Prove that
(a) DG F = 0, so that DF and DG are in the characteristic system of Q. (Hint: DF G = −DG F ).

(b) DF and DG are linearly independent. (Hint: So are ω, dF, dG and iDF dω = dF − (∂z F )ω,
iDG dω = dG − (∂z G)ω).
(c) Qo = hDF , DG i and Q = hdF, dG, dHi, so that ω = udH + vdF + wdG.
(d) The manifolds F = 0, G = λ1 , H = λ2 , are solutions of the considered equation.
Jacobi’s Method: In the case of an equation F (x1 , . . . , xn , p1 , . . . , pn ) = 0 without the unknown z,

show that a classical solution z(xP 1 , . . . , xn ) defines a germ of submanifold pi = (∂i z)(x1 , . . . , xn ) of
R2n of dimension n, where ω = i pi dxi specializes to dz, so that F and dω specialize to 0. If we call
(generalized) solution any germ of submanifold of dimension n where F and dω specialize to 0, prove
that classical solutions are just solutions where x1 , . . . , xn define a local coordinate system.
Show that (R2n , ω2 := dω) is a symplectic manifold, so that for any function f ∈ C ∞ (R2n ) we
have a vector field Df such that df = iDf ω2 . Prove that [Df , Dg ] = D{f,g} , where the Poisson
bracket is defined to be {f, g} := ω2 (Df , Dg ) = Df g = −Dg f . (Hint: Show that DfL ω2 = 0;
hence (i[Df ,Dg ] ω2 )(D) = ω2 ([Df , Dg ], D) = Df (ω2 (Dg , D)) − ω2 (Dg , [Df , D]) = Df (Dg) − [Df , D]g =
D(Df g) = D{f, g} = (d{f, g})(D)).
Put F1 = F , and assume that dF1 does not vanish at the considered point. Show that there is a
first integral F2 of DF1 such that dF1 , dF2 are linearly independent. Then prove that the distribution
hDF1 , DF2 i is involutive (Hint: {F1 , F2 } = DF1 F2 = 0), and show that there is a common first integral
F3 such that dF1 , dF2 , dF3 are linearly independent. Conclude that so we obtain germs F1 , . . . , Fn ,
with linearly independent differentials, such that 0 = DFi Fj = ω2 (DFi , DFj ).
Show that DF1 , . . . , DFn define a distribution, totally isotropic for ω2 , whose incident is generated
by dF1 , . . . , dFn . Conclude that the submanifolds F = 0, F2 = λ2 , . . . , Fn = λn are solutions of the
considered equation F = 0.
21. Let X be an oriented smooth manifold of dimension n and let ω be a differential (n + 1)-form on
R × X. Use Fubini’s theorem to prove that
Z Z Z
ω= i∂t ω dt.
R×X R t×X
2
22. Gauss-Green Formula: If Ω ⊂ R is a compact region bounded by a curve C, show that
ZZ Z
∂g ∂f
− dxdy = (f dx + gdy).
Ω ∂x ∂y C
23. If ω and θ are closed forms, prove that so is ω ∧ θ. If moreover θ is exact, so is ω ∧ θ.

24. Let H = i xi ∂i be the vector field on Rn defined by the flow τt (x1 , . . . , xn ) = (et x1 , . . . , et xn ). Prove
P
the following equalities for any p-form ω on Rn :
∂ ∗
τt∗ (H L ω) = τ ω ,
∂t t
516 EXERCISES
Z 0 Z 0 Z 0
∂ ∗
ω= τ ω dt = τt∗ (H L ω)dt = H L (τt∗ ω)dt ,
−∞ ∂t t −∞ −∞
L p n p n
and conclude that the Lie derivative H : Ω (R ) → Ω (R ) is surjective.
If dω = 0, show that ω = H L ω̃, where dω̃ = 0, so obtaining another proof of Poincaré’s lemma:
ω = H L ω̃ = diH ω̃ + iH dω̃ = d(iH ω̃).
1
25. Prove that HDR (S2 ) = 0.
1
Prove that HDR (S1 ) = R. (Hint: Consider the differential dθ of the argument).
26. If n is even, show that any vector field D on an open neighborhood of the sphere Sn is normal at some
point of the sphere, Dx ∈ (Tx Sn )⊥ .
Given a smooth map f : Sn → Sn , show that f (x) = ±x at some point x ∈ Sn .
Prove that there is no Pfaff system on rank 1 on Sn .
Given a Pfaff system P of rank 1 on an open neighborhood of Sn , show that Px is normal to the
sphere at some point x ∈ Sn .
Conclude that any smooth map Pn,R → Pn,R has a fixed point.
27. Let us consider an integrable Pfaff system P = hωi on a smooth manifold X. Prove that
(a) We have dω = ω ∧ ω1 for some 1-form ω1 on X.

(b) The 3-form ω3 = ω1 ∧ dω1 is closed.
3
(c) The cohomology class [ω3 ] ∈ HDR (X) only depends on the Pfaff system P, not on ω1 nor the
generator ω.
28. Incompressible Stationary Fluids: Let us consider an incompressible stationary perfect fluid in the
Euclidean space R3 ; i.e., the density ρ is constant, and the pressure p(x, y, z) and mean velocity
~v = v1 (x, y, z)∂x + v2 (x, y, z)∂y + v3 (x, y, z)∂z don’t depend on time and satisfy the equations
(
div ~v = 0 (Mass conservation law)
~v ∇~v = −grad (p/ρ) (Law of motion)
If we put v 2 = |~v |2 , and ω~v is the 1-form corresponding to ~v , prove that

2
(a) The 1-form corresponding to ~v ∇~v is just i~v dω~v + d v2 .
2
(b) The law of motion may be restated as i~v dω~v = −d ρp + v2 .
(c) The flow of ~v preserves dω~v ; i.e., the Lie derivative ~v L (dω~v ) vanishes.
(d) The integral of dω~v on any bounded 2-dimensional region of the fluid remains constant, and so
does the integral of ω~v on the boundary (named circulation of ~v along the boundary).
(e) The flow of ~v also preserves the rotational of ~v , which is defined to be the vector field rot ~v such
that irot ~v ω3 = dω~v , where ω3 is the volume form of R3 .
p v2
(f) The function ρ + 2 is constant along any trajectory of the fluid.
p v2
(g) In the case of an irrotational fluid, dω~v = 0, the function ρ + 2 is constant (on the whole space,
not just on each trajectory).
29. Plane Stationary Fluids: Let us consider a complex 1-form on an open set U ⊆ C,
ω = f (z) dz = (v1 − v2 i)(dx + idy) = (v1 dx + v2 dy) + (−v2 dx + v1 dy)i,
and put ~v = v1 + v2 i = f (z), so that f (z) = |~v |e−iθ , where θ is the angle between the x-axis and ~v .
Prove that
(a) ω is holomorphic if and only if dω~v = 0 and div ~v = 0. In such case, if we fix the pressure to be
2 2
p = ρ(a − v2 ), where a and ρ are constants, we have i~v dω~v = −d ρp + v2 .
ANALYSIS III 517
(b) Once we fix the mass density ρ and the pressure aρ at the equilibrium points, holomorphic 1-forms
ω in U correspond with incompressible stationary irrotational fluids, the real part of ω being the
1-form ω~v of the fluid.
(c) The 1-form ω = λz −1 dz, with λ real, defines a fluid whose trajectories are the half-lines with end
point at the origin, which is a source when λ > 0 and a drain when λ < 0.
(d) The 1-form ω = iλz −1 dz defines a fluid whose trajectories are concentric circles, the sign of λ
determining the sense of rotation.
30. Let X be a Riemann surface. If π : Y → X is a covering, show that Y admits a unique structure of
Riemann surface such that π is a local analytic isomorphism.
31. Let f be an analytic function on a Riemann surface X. Show that any point p ∈ X has a coordinate
neighborhood (U, z) such that z(p) = 0 and f |U = a + z n for some n ∈ N, a ∈ C.
32. Let p be a point of a Riemann surface X, and let Op be the ring of germs at p of analytic functions
defined on a neighborhood of p. Prove that
(a) Op is a discrete valuation ring, of maximal ideal mp = zOp , where z is any local coordinate at p
such that z(p) = 0
(b) Any R-derivation D : Cp∞ → R induces a C-derivation D̃ : Op → C, D̃(u + iv) = Du + iDv, so
that we have a R-linear isomorphism Tp X = DerC (Op , C).
(c) For any non constant analytic map f : X → Y , we have indp f = dim C (Op /mq Op ); q = f (p).
33. An almost complex structure on a smooth manifold X is a (1,1)-tensor field J (a field of endomor-
phisms) such that J 2 = −Id, and it is a complex structure if at any point p ∈ X there exist local
coordinate systems (x1 , y1 , . . . , xn , yn ) where J(∂xi ) = ∂yi , J(∂yi ) = −∂xi .
The automorphism J ⊗ 1 : (Tp X)C → (Tp X)C has the eigenvalues ±i, so that we have a decomposition
(1,0) (0,1)
(Tp X)C = Tp ⊕ Tp where the complex vectors of type (1, 0) have eigenvalue i and the complex
vectors of type (0, 1) have eigenvalue −i. We also have an automorphism J ∗ ⊗ 1 : (Tp∗ X)C → (Tp∗ X)C ,
and a decomposition (Tp X)C = Ω(1,0) ⊕ Ω(0,1) , where the (1, 0)-type 1-forms have eigenvalue i and
(1,0) (0,1)
the (0, 1)-type 1-forms have eigenvalue −i. Moreover, Ω(1,0) and Ω(0,1) are C-dual to Tp and Tp
respectively. Prove that
(a) An almost complex structure is a complex structure if and only if the complex Pfaff system Ω(1,0)
of (1, 0)-forms is integrable, in the sense that locally Ω(1,0) = hdz1 , . . . , dzn i for some complex
smooth functions z1 , . . . zn
(b) The complex distribution T (0,1) of complex (0, 1)-vector fields is involutive (closed under the Lie
bracket) if and only if the following Nijenhuis tensor N vanishes:
N (D1 , D2 ) = [JD1 , JD2 ] − J[JD1 , D2 ] − J[D1 , JD2 ] − [D1 , D2 ].
34. In this exercise we assume, without proof, the Newlander-Nirenberg theorem: An almost complex
structure is complex if and only if the Nijenhuis tensor vanishes. If X is a surface, prove that the
Nijenhuis tensor is always null. Moreover, if (X, g, ω2 ) is an oriented riemannian surface, prove the
following results:
(a) The endomorphism J attached to the area 2-form, J(D) · D0 = ω2 (D, D0 ), defines an almost
complex structure on X.
(b) X, endowed with the sheaf O of complex smooth functions f with C-linear differential df , is a
Riemann surface.
(c) Locally, there exist isothermal coordinates (x, y) such that g = h2 (dx2 +dy 2 ) for some positive
function h.
(d) If u is an harmonic function with non null differential at a given point p ∈ X, then in a certain
open neighborhood U of p there is a complex coordinate z = u + iv. Moreover, u, v are isothermal
coordinates on U .
(e) Two riemannian metrics g 0 , g on an oriented surface induce the same complex structure if and
only if they are conformal: g 0 = h2 g for some positive function h.
518 EXERCISES
(f) Let (X, gX ) and (Y, gY ) be two oriented riemannian surfaces. An oriented smooth map φ : X → Y
is analytic if and only if φ∗ gY = f gX for some non-negative smooth function f .
35. Prove that any Riemann surface admits a compatible riemannian metric.
36. Let (X, gX ) and (Y, gY ) be two Riemann surfaces with a compatible metric. Prove that an oriented
smooth map ϕ : X → Y is analytic if and only if ϕ∗ gY is proportional to gX .
37. Let us consider the unit disk D, with the Poincaré metric. Prove the following statements:
(a) The geodesic lines are the circles orthogonally intersecting ∂D (including the diameters of D).
Hence, given two different points z1 , z2 ∈ D, there is a unique geodesic line joining them.
(b) The length of any curve γ : [a, b] → D joining two points z1 , z2 ∈ D is bounded below by the
length d(z1 , z2 ) of the geodesic joining them. (Hint: Assume that z1 = 0 and z2 ∈ [0, 1), and
compare the length of γ(t) = x(t) + y(t)i with the length of x(t)).
(c) The hyperbolic distance d(0, z) is an increasing function of |z|.
(d) If an analytic function f : D → D is an isometry at a point, then it is a global isometry.
38. If a compact Riemann surface X admits a hyperbolic metric gX , prove that it is unique and the
analytic automorphisms of X are just the oriented isometries of gX .
39. Show that the complex plane C does not admit a hyperbolic metric.
40. Let X be a simply connected Riemann surface and fix a point p ∈ X and a vector 0 6= Dp ∈ Tp X.
Show that we have a bijection

pointed local analytic hyperbolic metrics on X
= , ϕ→ 7 ϕ∗ (gr ).
isomorphisms ϕ : X → Dr such that |Dp | = 2/r2
(Hint: Any metric of constant curvature K = −1 defines (p. 293) an oriented local isometry X → D,
unique up to an automorphism of D).
1
41. Prove that z + z defines an analytic isomorphism of D with P1 − [−2, 2].

Find isomorphisms of D with C − [0, ∞) and with C − (−∞, −1] ∪ [1, ∞) .
10. Differential Geometry I

1. Prove that any open ball in Rn is diffeomorphic to Rn .
2. Define the obvious structure of smooth manifold of the real and complex projective spaces.
3. Let X be the non separated topological manifold of exercise 6 in p. 478, and put U0 = R×0, U1 = R×1,
so that U0 and U1 have an obvious smooth structure. Show that (X, OX ) is a non-separated smooth
manifold when, for any open set U ⊆ X, we put
OX (U ) = {f ∈ C(U ) : f |U0 ∩U ∈ C ∞ (U0 ∩ U ) and f |U1 ∩U ∈ C ∞ (U1 ∩ U )}.
4. Prove that the category of smooth manifolds has finite products and coproducts.
5. Let (x1 , . . . , xn ) be a local coordinate system of a smooth manifold X at a point p = (a1 , . . . , an ).
Let (Op , mp ) be the local ring of germs at p of smooth functions on X. If jpr f stands for the Taylor
expansion of f at p up to degree r, show that we have an isomorphism of R-algebras
Op /mr+1
p ' R[x1 − a1 , . . . , xn − an ]/(x1 − a1 , . . . , xn − an )r+1 , f 7→ jpr f.
Moreover, if m is the maximal ideal of C ∞ (X) defined by p, then C ∞ (X)/mr+1 = Op /mr+1
p .
6. If ϕ : X → Y is a surjective smooth map, prove that dim X ≥ dim Y . (Hint: Use Srad’s theorem).
7. If i : Y → X is a closed smooth submanifold, prove that i∗ : C ∞ (X) → C ∞ (Y ) is surjective.
8. Let i : Y → X be a submanifold and y ∈ Y . Prove that the kernel of i∗ : OX,y → OY,y is an ideal
generated by a finite number of germs with linearly independent differentials at y.
DIFFERENTIAL GEOMETRY I 519
9. Prove that most local properties of smooth manifolds are in fact local with respect to surjective regular
projections π : R → X. For example:
(a) A map φ : X → Y into a smooth manifold Y is continuous (resp. smooth, a regular projection)
if and only if so is φ ◦ π.
(b) A subset Y ⊆ X is closed (resp. open, a submanifold) if and only if so is π −1 (Y ) ⊆ R.
(c) If φ : Y → X is a smooth map, then Y ×X R is a smooth submanifold of Y × R and it is the fibred
product in the category of smooth manifolds. Moreover, φ is a local embedding (resp. regular
projection) if and only if so is φ × Id : Y ×X R → X ×X R = R.
10. Let R ⊆ X × X be an equivalence relation on a smooth manifold X. If there exists a structure
of (separated) smooth manifold on the quotient set X̄ = X/R such that the canonical projection
π : X → X̄ is a regular projection, show that R is a closed submanifold of X × X and that both
projections π1 , π2 : R → X are regular projections. Moreover, dim (X/R) = 2dim X − dim R.
11. Two smooth maps φ : X → S, ψ : Y → S are said to be transversal at a point (x, y) of the topological
fibred product X×S Y when Im φ∗,x +Im ψ∗,y = Ts S, where s = φ(x) = ψ(y). If φ and ψ are transversal
at any point, prove that the fibred product X ×S Y exists in the category of smooth manifolds, and
that dim (X ×S Y ) = dim X + dim Y − dim S.
12. Prove the following formula for the exterior differential of p-forms:
(dωp )(D0 , . . . , Dp ) = i (−1)i Di ωp (. . . , D
b i , . . .) + P (−1)i+j ωp ([Di , Dj ], . . . , D
P b i, . . . , D
b j , . . .).
i<j
13. (Lagrange multipliers): Let Y = {x ∈ X : f1 (x) = . . . = fr (x) = 0},where X is a smooth manifold

and f1 , . . . , fr ∈ C ∞ (X). If dp f1 , . . . , dp fr are linearly independent at any point p ∈ Y , prove that the
restriction h|Y of h ∈ C ∞ (X) has null differential at p if and only if dp h ∧ dp f1 ∧ . . . ∧ dp fr = 0.
14. Show that the following subgroups of Gl(n, C) = {A ∈ Mn×n (C)} are closed submanifolds:
U (n) = {AĀt = I}, Sl(n, C) = {det A = 1}, O(n, C) = {AAt = I},
Gl(n, R) = Gl(n, C) ∩ Mn×n (R), Sl(n, R) = Sl(n, C) ∩ Gl(n, R), O(n, R) = O(n, C) ∩ Gl(n, R).
15. Let f : X → Y be a smooth map and y = f (x). Prove that f∗ : Tx X → Ty Y is surjective if and only
if f ∗ : OY,y → OX,x is injective and m2x ∩ OY,y = m2y .
16. Let f : Y → X be a smooth map and x = f (y). Prove that the following conditions are equivalent:
(a) f∗ : Ty Y −→ Tx X is injective.
(b) f ∗ : OX,x −→ OY,y is surjective
(c) f ∗ : OX,x −→ OY,y is surjective, the kernel I is a finitely generated ideal and m2x ∩ I = mx I.
17. Prove the key lemma along a smooth subvariety: If Y is a submanifold of codimension r of a smooth
manifold X, any point q ∈ Y has a coordinate neighborhood (U ; u1 , . . . , un ) in X such that the ideal
of all smooth function f ∈ C ∞ (U ) vanishing on Y ∩ U is just (u1 , . . . , ur ).
18. Let Γkij be the Christoffel symbols of a linear connection ∇ in a local coordinate system (x1 , . . . , xn ).
Check that the torsion Tor∇ and curvature R of ∇ are
P k
Tor∇ = Tij dxi ⊗ dxj ⊗ ∂k , Tijk = Γkij − Γkji ,
i,j,k
l l
= ∂i Γljk − ∂j Γlik + (Γhjk Γlih − Γhik Γljh ).
PP P
R= Rij,k (dxi ∧ dxj ) ⊗ dxk ⊗ ∂l , Rij,k
i<j kl h
19. Let ∇ be a linear connection on a smooth manifold and let p be a point of X. Let σD (t) be the
geodesic line tangent to D ∈ Tp X at t = 0. Prove that
(a) The exponential map exp : Tp X → X, exp(D) = σD (1), is a well-defined smooth map on a
neighborhood U of the origin in Tp X.
(b) The tangent linear map exp∗ : Tp X = Tp U → Tp X is the identity, and the exponential map defines
a diffeomorphism of an open neighborhood V of the origin in Tp X onto an open neighborhood of
p in X.
520 EXERCISES
(c) If we consider a local coordinate system on a neighborhood of p, corresponding to linear coordi-

nates in V , then Γkij (p) + Γkji (p) = 0.
(d) A linear connection is symmetric if and only if any point p ∈ X admits a coordinate neighborhood
where the Christoffel symbols vanish at the given point, Γkij (p) = 0.
(e) If ∇ is symmetric, then dω is the hemisymmetrization of ∇ω for any p-form ω.
20. If a symmetric linear connection ∇ on a smooth manifold X preserves a volume form, ∇(dX) = 0,
prove that DL (dX) = (C11 ∇D)dX for any vector field D.
21. If σ : I → X is a geodesic line, and φ : I 0 → I is a diffeomorphism such that σ φ(t) also is a geodesic

line, show that φ(t) = at + b for some a, b ∈ R.

22. Prove that the Cartan connection ∇ preserves the time and space metrics, ∇g = 0, ∇h = 0, and that
in any inertial coordinate system, ∇∂t = −dt ⊗ F~ , ∇∂i = 0, ∇(dt) = 0, ∇(dxi ) = Fi dt ⊗ dt. Moreover,
the curvature tensor R and the Ricci tensor R2 of ∇ are just
P P
R = i (dt ∧ dFi ) ⊗ (∂i ⊗ dt) , R2 = − i (∂i Fi )dt ⊗ dt.
23. Let ∇ be a symmetric linear connection on a smooth manifold X and let ω be a 1-form on X not
vanishing at any point. If ∇(ω ⊗ ω) = 0, show that ∇ω = 0 and dω = 0.
24. Let (t, x, y, z) be inertial coordinates in the Minkowski spacetime. Fix the year as the time unit, and
the light-year as the length unit, so that c = 1 and the gravitational acceleration on the earth surface
(the most comfortable acceleration for us, 980 cm/s2 ) is ≈ 1. Check that
t = sinh τ, x = cosh τ − 1, y = 0, z = 0,
is a trajectory of proper time τ . Calculate the unitary tangent vector, U · U = 1, and the accel-
eration U ∇ U , and check that it is a trajectory of acceleration g = 1, starting at rest at τ = 0.
After traveling τ = 2.4 years, we reach Proxima Centauri (≈ 4.5 light-years), after τ = 11 years,
the center of the Milky Way (≈ 29.000 light-years), after τ = 15.5 years, the Andromeda Galaxy
(≈ 2.7 million light-years) and after τ = 25 years we arrive at the most remote observable parts of the
Universe (≈ 35 billion light-years); even if we never reach the light speed!
25. Let us consider in R6 the cone t2 − x2 − y 2 − z 2 − u2 − v 2 = 0, defining a quadric Q in P5 . Prove that
the intersection of the cone with the hyperplane u − v = 1, endowed with the metric
g = dt2 − dx2 − dy 2 − dz 2 − du2 − dv 2
induced by the cone, is isometric to the Minkowski space via the coordinates (t, x, y, z).
So Q is a compactification of the Minkowski space. Check that different light directions are compact-
ified with different points, while all non-light directions are compactified with a unique point.
26. Let T3 and T21 be tensors on a riemannian manifold (X, g) satisfying the following conditions:
(a) T3 (D1 , D2 , D3 ) = T3 (D1 , D3 , D2 ).
(b) T21 (D1 , D2 , ω) = −T21 (D2 , D1 , ω).
Prove the existence of a unique linear connection ∇ on X such that ∇g = T3 and Tor∇ = T21 . (Hint:
Repeat the proof of the fundamental theorem of riemannian geometry).
27. Let g be a covariant symmetric metric of constant rank on a smooth manifold X. If there exists
a symmetric linear connection ∇ such that ∇g = 0, prove that the radical V of g is an involutive
distribution. Moreover g is projectable by V; i.e., V L g = 0 whenever V is a vector field in V. (Hint:
D∇ V is in V for any vector field D).
28. Let h be a contravariant symmetric metric of constant rank on a smooth manifold X, let V be the
incident of the radical of h, and let h , i denote the metric induced by h on the distribution V. Given
a linear connection ∇ on X, prove that ∇h = 0 if and only if
(a) D∇ V ∈ V, whenever V ∈ V.
(b) DhV1 , V2 i = hD∇ V1 , V2 i + hV1 , D∇ V2 i, whenever V1 , V2 ∈ V.
DIFFERENTIAL GEOMETRY I 521
(Hint: If we put V1 = h(ω1 ), then (D∇ ω1 )(V2 ) = D(hV1 , V2 i) − hV1 , D∇ V2 i).

29. Let us consider the Euclidean space R3 . Using that the classical newtonian potential N (ρ) of a
function with compact support ρ ∈ C ∞ (R3 ) is a solution, commuting with partial derivatives, of
Poisson’s equation ∆u = 4πρ, give a new proof of Poincaré’s lemma in germs:
(a) For any p-form ω with compact support in R3 , define a p-form ∆−1 ω, commuting with d and δ,
such that ∆∆−1 ω = ω.
(b) Prove Poincaré’s lemma in R3 for closed germs, dω = 0,
ω = ∆∆−1 ω = dδ∆−1 ω + δ∆−1 dω = d(δ∆−1 ω).
(c) In R3 , given a closed (p + 1)-form, dω = 0, and a coclosed (p − 1)-form, δω 0 = 0, prove that at
any point there is a germ of a p-form α such that dα = ω, δα = ω 0 .
(Hint: ω + ω 0 = ∆∆−1 (ω + ω 0 ) = (d + δ)(d + δ)∆−1 (ω + ω 0 ) = (d + δ)(δ∆−1 ω + d∆−1 ω 0 ) = (d + δ)α).
30. Prove the following formulae in the Euclidean space of dimension 3:
rot(grad f ) = 0 , div(rot D) = 0 , grad(f h) = h(grad f ) + f (grad h)
rot(f D) = (gradf ) × D + f (rot D) , div(f D) = (grad f ) · D + f (div D)
rot(rot D) = grad(div D) − ∆D
div(D1 × D2 ) = (rot D1 ) · D2 − D1 · (rot D2 ) , grad(D1 · D2 ) = (∇D1 ) · D2 + (∇D2 ) · D1
rot(D1 × D2 ) = D2 · (∇D1 ) − (div D1 )D2 − D1 · (∇D2 ) + (div D2 )D1 .
2 n k
P
31. Let ∇ be the Levi-Civita connection of the Euclidean metric g = i dxi in R . If Γij are the
Christoffel symbols of ∇ in a local coordinate system (U ; u1 , . . . , un ), prove that
2
(a) ∇2 f = ∇(∇f ) = ij ∂x∂i ∂x f
dxi ⊗ dxj , for any smooth function f ∈ C ∞ (U ).
P
j
(b) ∇2 uk = − ij Γkij dui ⊗ duj .

P
(c) If Γkij = 0, then ui = bi + j aij xj , where A = (aij ) is invertible. (U connected).

P
(d) If g = i du2i , then ui = bi + j aij xj , where At A = I. (U connected).

P P
(e) Any diffeomorphism Rn → Rn preserving ∇ is an affinity.

(f) Any diffeomorphism Rn → Rn preserving g is a motion.
32. Let C ∞ Aff be the category of pairs (X, ∇), where ∇ is a linear connection on a smooth manifold X,
and diffeomorphism preserving the linear connection. Show that the category Aff of real affine spaces
and affinities is equivalent to the subcategory of C ∞ Aff defined P
by the pairs (X, ∇) isomorphic to
some Rn with the Levi-Civita connection of the Euclidean metric i dx2i .
If we consider the category Riemann of riemannian manifolds and diffeomorphisms preserving the
metric, show that the category of Euclidean spaces (An , V, Ω2 ) with fixed unit length is equivalent to
the subcategory ofPRiemann defined by the riemannian manifolds isomorphic to some Rn with the
Euclidean metric i dx2i .
(Hint: If you are willing to assume that the objects form a set, you may use problem 50 in p. 483 in
both statements. Otherwise some additional work is required).
33. Prove that any harmonic coordinate system (u1 , . . . , un ) in Rn , such that g ii = g(dui , dui ) goes to
P ∂ui 2
some constant ci at infinity, is ui = bi + j aij xj . (Hint: g ii = j ∂x
P
j
).
34. Let F = F1 ∂1 + F2 ∂2 + F3 ∂3 be a force vector field in the Euclidean space (R3 , g = i dx2i ), and let
P
us consider the trajectory σ : I → R3 of a punctual particle satisfying the newtonian law of motion
mT ∇ T = F , where T = σ∗ (∂t ) is the velocity and the mass m ∈ R+ . If we consider the work 1-form
ωF = g(F, −) and the momentum 1-form θ = g(mT, −), show that T ∇ θ = ωF .
If the forces are conservative, dωF = 0, prove that the forces derive from a potential, ωF = du for
some smooth function u ∈ C ∞ (R3 ). Moreover,
m
(a) The energy e = 2T · T − u is constant along the trajectory of the particle.
(b) The punctual particle follows a geodesic line of the metric ḡ = 2m(e + u)g.
522 EXERCISES
(c) Let L = m2 T ·T +u be the lagrangian. If we fix two points p1 = σ(t1 ), p2 = σ(t2 ) of the trajectory
and a vector field D vanishing at both, and we extend T to a neighborhood so that [D, T ] = 0,
then we have Lagrange’s equations T (mT · D) − DL = 0, not involving ∇, and the principle
of least action
Z t2
0= (DL)dt.
t1
35. Let (X, g, ωX ) be an oriented semiriemannian manifold of dimension n. Given a p-vector D1 ∧. . .∧Dp ,
we put iD1 ∧...∧Dp ωX := iDp . . . iD1 ωX , and we define the Hodge star operator ∗ : Ωp (X) → Ωn−p (X)
by the formula ∗ωp = iωp ωX , where a p-form ωp is viewed as a p-vector by means of the isomorphisms
g : Tx X → Tx∗ X induced by the metric g. If the metric g has s negative signs, prove that ∗1 = ωX ,
∗ωX = (−1)s , ∗ ∗ ωp = (−1)p(n−p)+s ωp , g(ωp , ω̄p ) = (−1)s g(∗ωp , ∗ω̄p ), ωp ∧ ∗ω̄p = g(ωp , ω̄p )ωX ,
∗δωp = (−1)p+1 d ∗ ωp .
Now we consider on the space of p-forms with compact support Ωpc (X) the symmetric metric
Z
hωp , ω̄p i = g(ωp , ω̄p )ωX .
X
Z
Prove that hωp , ω̄p i = ωp ∧ ∗ω̄p , hdωp , ωp+1 i = −hωp , δωp+1 i, and hωp , ω̄p i = hωp , ω̄p i.
X
36. Put K(D1 , D2 ) = R2,2 (D1 , D2 , D2 , D2 ), where R2,2 is the Riemann-Christoffel tensor of a riemannian
metric. Prove that
1 ∂ 2

R2,2 (D1 , D2 , D3 , D4 ) = K(D1 + tD3 , D2 + sD4 ) − K(D1 + tD4 , D2 + sD3 )
6 ∂t∂s t=0,s=0

and conclude that the curvature tensor R vanishes at a point p whenever all the sectional curvatures
KΠ at p are zero.
37. A symplectic manifold is a smooth manifold X, endowed with a closed non-singular 2-form ω2 . If
D is a tangent vector field, we put h(D) = iD ω2 , and we say that D is a hamiltonian vector field
when DL ω2 = 0.
(a) Prove that D is a hamiltonian vector field if and only if h(D) is a closed 1-form.
(b) If D1 , D2 are hamiltonian vector fields, prove that so is [D1 , D2 ] and that, in fact, it corresponds
to an exact 1-form. (Hint: i[D,D̄] = [iD , D̄L ]).
(c) The cotangent bundle π : T ∗ X → X of a smooth manifold has a structural 1-form θωx = π ∗ (ωx );
hence a 2-form ω2 = dθ. Prove that (T ∗ X, ω2 ) is a symplectic manifold.
(d) In a riemannian manifold (X, g) we have a canonical isomorphism T X → T ∗ X, D 7→ iD G; hence
the tangent bundle T X inherits a 1-form θ and a 2-form ω2 = dθ. Prove that the geodesic vector
field Z is the hamiltonian vector field corresponding to e(Dx ) = 21 Dx · Dx , in the sense that
iZ ω2 = −de. Moreover, θ(Z) = 2e.
38. Meusnier’s Theorem: Show that all curves passing through a point with common tangent have
equal normal curvature.
39. Let q(x, y) = d(x2 + y 2 ) − 2ax − 2by + c and let us consider, on the open subset of R2 where q(x, y) 6= 0,
the riemannian metric g = 4q(x, y)−2 (dx2 + dy 2 ).
(a) When q(x, y) = x2 + y 2 − 1, the metric g has constant negative curvature (the unit disk, with
this metric, is the Poincaré’s disk).
(b) When q(x, y) = x2 + y 2 + 1, the metric g has constant positive curvature (it is the stereographic
projection, onto the equator plane, of the metric of the unit sphere).
(c) When q(x, y) = x2 + y 2 , the metric g has null curvature (it is just the inversion of the Euclidean
metric with respect to the unit circle).
(d) When q(x, y) = x, the metric g has constant negative curvature (the semiplane x > 0, with this
metric, is the Poincaré’s half-plane).
(e) When q(x, y) = 1, the metric g is the Euclidean metric.
ALGEBRAIC GEOMETRY I 523
(f) In general, the riemannian metric g has constant curvature −R2 (q), where R2 is the quadratic
form on the space of circles (p. 495) and the geodesics of g are just real circles and lines orthogonal
to q(x, y) with respect to R2 .
(Hint: If 0 < f (x, y), the curvature of f −2 (dx2 + dy 2 ) is K = f ∆f − kgrad f k2 . To find the geodesics,
use that 2(T ∇ T ) · N = T (T · N ) + T (N · T ) − N (T · T ) when [T, N ] = 0).
40. If q(x1 , . . . , xn ) = d(x21 + . . . + x2n ) − 2a1 x1 − . . . − 2an xn + c, prove that 4q −2 (dx21 + . . . + dx2n ) has
constant curvature K = cd−(a21 +. . .+a2n ). (Hint: If f (x1 , . . . , xn ) is positive, the sectional curvatures
of f −2 (dx21 + . . . + dx2n ) are Kij = f (fii + fjj ) − kgrad f k2 , where fii = ∂i ∂i f ).
∇(df )
41. Prove that the second fundamental form of a surface f (x, y, z) = 0 in R3 is φ2 = − ·
kgrad f k
42. Let D0 , . . . , Dn be a local orthonormal basis of vector fields in a riemannian manifold X.
(a) Show that the laplacian of a smooth function u is just ∆u = i Di2 u − (Di∇ Di )u .
P
(b) Let φ the Weingarten endomorphism of an hypersurface X̄ → X, and let us extend the unitary
normal vector N to a neighborhood so that N ∇ N = 0. Check that the relation of ∆ and the
¯ of X̄ is given by the formula
laplacian ∆
¯ = ∆u + (trφ)(N u) − N (N u) ,
∆u u ∈ C ∞ (X).
(Hint: Fix a local orthonormal basis D1 , . . . , Dn in X̄, extend it to a neighborhood so that
N ∇ Di = 0, and check that N, D1 , . . . , Dn is an orthonormal basis).
(c) In the Euclidean space X = R3 , prove that a surface X̄ is minimal (trφ = 0) if and only if the
cartesian coordinates x, y, z are harmonic functions on X̄. (Hint: The condition N ∇ N = 0 states
that N 2 x = N 2 y = N 2 z = 0).
43. Show that any Lie group G admits a unique linear connection ∇ such that the left invariant vector
fields are parallel. Moreover:
(a) The morphisms of Lie groups σ : R → G are the geodesic lines passing through the neutral element
e ∈ G at t = 0.
(b) exp : g = Te G → G is just the exponential map of ∇ at e.
(c) ∇ has null curvature, but Tor∇ (D, D0 ) = −[D, D0 ] = 0 for all D, D0 ∈ g.
(d) ∇ is complete.
44. Prove that a connected Lie group G is abelian if and only if [D0 , D] = 0 for all D0 , D ∈ g.
45. Let g be the Lie algebra of a Lie group G. Show that left invariant linear connections on G, in the
sense that Lx (D∇ D0 ) = (Lx D)∇ (Lx D0 ), correspond to bilinear maps g × g → g.
11. Algebraic Geometry I

1. Prove that any open subset U ,→ X is the étalé space of a sheaf of sets U on X. Moreover, for any
sheaf of sets F on X we have Hom(U, F) = F(U ).
2. If X is a connected topological space, show that the category of constant sheaves of sets over X is
equivalent to the category of sets.
3. A continuous map π : P → X is said to be an étalé space if it is a local homeomorphism: any
point p ∈ P has an open neighborhood V such that π : V → π(V ) is a homeomorphism onto an open
subset π(V ) of X. Prove that the étalé space π : P et → X of a presheaf of sets P is in fact an étalé
space, and that the construction of the étalé space defines an equivalence of categories

Sheaves of Étalé spaces
sets over X over X
524 EXERCISES
4. Show that, in the category of étalé spaces over a topological space X, the final object is the identity
map X → X, and prove that the open embeddings i : U ,→ X are just the étalé spaces such that the
unique morphism to the final object is a monomorphism.
Conclude that two T0 topological spaces are homeomorphic if and only if their categories od sheaves
of sets are equivalent.
5. Prove that a sequence F 0 → F ⇒ F 00 (resp. F 0 ⇒ F → F 00 ) is exact in the category of presheaves
of sets on a topological space X if and only if so is the sequence F 0 (U ) → F(U ) ⇒ F 00 (U ) (resp.
F 0 (U ) ⇒ F(U ) → F 00 (U )) in the category of sets for any open set U ⊆ X.
6. Prove that a sequence F 0 → F ⇒ F 00 (resp. F 0 ⇒ F → F 00 ) is exact in the category of sheaves of sets
on a topological space X if and only if so is the sequence Fx0 → Fx ⇒ Fx00 (resp. Fx0 ⇒ Fx → Fx00 ) in
the category of sets for any point x ∈ X. (Hint: Consider the sheaf associated to the presheaf of sets
U Coker (F 0 (U ) ⇒ F(U )) and look at the fibres).
7. Prove that a sequence F 0 → F ⇒ F 00 is exact in the category of sheaves of sets on a topological space
X if and only if so is the sequence of sets F 0 (U ) → F(U ) ⇒ F 00 (U ) for any open set U ⊆ X.
8. Given a sequence F 0 → F → F 00 → 0 of morphisms of sheaves of abelian groups on X, show that the
induced sequence of abelian groups Fx0 → Fx → Fx00 → 0 is exact at any point x ∈ X if and only if
0 → Hom(F 00 , G) → Hom(F, G) → Hom(F 0 , G) is exact for any sheaf of abelian groups G on X.
Given a sequence 0 → F 0 → F → F 00 of morphisms of sheaves of abelian groups on X, show that the
induced sequence of abelian groups 0 → Fx0 → Fx → Fx00 is exact at any point x ∈ X if and only if
0 → Hom(G, F 0 ) → Hom(G, F) → Hom(G, F 00 ) is exact for any sheaf of abelian groups G on X.
9. Let ∇ be a flat connection on a smooth manifold X and let F be the sheaf of bases of parallel vector
fields. Show that the étalé space F et → X is a covering. When X is simply connected, conclude that
it is parallelizable (there exists a global base of parallel vector fields).
10. Let f (z) be an analytic function on a connected open set U ⊂ C, let O be the sheaf of analytic
functions on C, and let us consider the connected component X of the étalé space of O containing the
germ of f at a point of U . Show that X admits a unique structure of Riemann surface such that the
natural map π : X → C is a local isomorphism. Moreover, we have an analytic section i : U → X of π
and an analytic continuation F of f (z) on X (an analytic function F on X extending f ).
Show that for any other unramified analytic map π 0 : X 0 → C, endowed with an analytic section
i0 : U → X 0 of π 0 and an analytic continuation F 0 of f (z) on X 0 , there exists a unique analytic map
h : X 0 → X such that π 0 = π ◦ h, i = i0 ◦ h and f 0 = F ◦ h.
11. Let P1 , P2 be two presheaves over a topological space X. Prove that the associated sheaf of the
presheaf U P1 (U ) ⊗Z P2 (U ) is just P1] ⊗Z P2] .
12. In the category of sheaves on a topological space X, show that Hom(F ⊗ G, L) = Hom(F, Hom(G, L)).
13. Let (X, OX ) be a ringed space. If M is an OX -module, show that HomOX (OX , M) = M.
If E, E 0 are two locally free OX -modules of finite rank, prove that
E ⊗OX HomOX (M, N ) = HomOX (M, N ⊗OX E),
E ∗ ⊗OX E 0 = HomOX (E, E 0 ).
If 0 → E 0 → E → E 00 → 0 is an exact sequence of locally free OX -modules of finite rank, prove that
p 0 n−p
L maps Λ E ⊗OX Λ
the images of the natural E → Λn E define a filtration of Λn E, and the associated
graded module is just p (Λp E 0 ⊗OX Λn−p E 00 ).
14. If a topological space X is discrete, show that any sheaf on X is a flasque sheaf.
If a topological space X has a dense point, show that any constant presheaf on X is a flasque sheaf.
15. Let F be a sheaf on a topological space X and x ∈ X. Show that lim
−→
H p (U, F) = 0, p ≥ 1.
x∈U
16. If F, G are sheaves on a topological space X, show that H (X, F ⊕ G) = H p (X, F) ⊕ H p (X, G).
p
If F is a sheaf on a disjoint union X ⊕ Y , show that H p (X ⊕ Y, F) = H p (X, F) ⊕ H p (Y, F).

17. If M is a torsion free A-module, prove that H 1 (X, M ) is a torsion free A-module.
18. Let R be a flat Z-module and let F be a sheaf on a topological space X. Prove that
(a) The presheaf (F ⊗Z R)(U ) = F(U ) ⊗Z R is a sheaf, and it is flasque when so is F.
(b) 0 → F ⊗Z R → (C • F) ⊗Z R is a flasque resolution of F ⊗Z R.
(c) H p (X, F ⊗Z R) = H p (X, F) ⊗Z R; hence, H p (X, R) = H p (X, Z) ⊗Z R.
19. If f : X → S is a continuous map, and 0 → F 0 → F → F 00 is an exact sequence of sheaves on X,
prove that the sequence 0 → f∗ F 0 → f∗ F → f∗ F 00 also is exact.
20. Let φ : Spec B → Spec A be a morphism of schemes, defined by a ring morphism A → B.
If M is an A-module, show that φ∗ (M
f) is the sheaf defined by the B-module MB = M ⊗A B.
If N is a B-module, show that φ∗ (N
e ) is the sheaf defined by the A-module N .
21. Let us consider a morphism of ringed spaces Φ : Y = Spec B → X = Spec A, defined by a continuous
map φ : Spec B → Spec A and a sheaf morphism ψ : OX → OY . If Φ preserves quasi-coherent modules
(i.e. φ∗ M
f is a quasi-coherent A-module,
e for any B-module M ) show that Φ is a morphism of locally
ringed spaces. (Hint: Let us consider the ring morphism ψ : A → B. We have to show that a function
f ∈ A vanishes at a point x = φ(y) if and only if ψ(f ) vanishes at y ∈ Y . If f (x) 6= 0, then
ψ(f )(y) 6= 0 because Φ is a morphism of ringed spaces. If f (x) = 0, then the open set φ−1 (Uf )
does not contain the closure of y, i.e. the support of B/p, where p is the prime ideal of y. Hence
g −1 Uf ) = (φ∗ B/p)(U
0 = B/p(φ g f ), and (φ∗ B/p)(Uf ) = (B/p)f = Bf /pf because φ∗ B/p is quasi-
g g
coherent. Now, Bf /pf = 0 means that ψ(f )(y) = 0.)
22. Let us consider the continuous map φ : Y = Spec Q → X = Spec Z, φ(Y ) = 2, and the sheaf
morphism ψ : OX → OY corresponding to the inclusion map φ∗ OX = Z2 →= OY = Q, so that
Φ = (φ, ψ) : Spec Q → Spec Z is a morphism of ringed spaces. Show that Φ is not a morphism of
locally ringed spaces.
23. Let Rings be the category of rings and ring morphisms, Aff the category of affine schemes and
morphisms of locally ringed spaces, and C the category of affine schemes and morphisms of ringed
spaces. Since any morphism of locally ringed spaces is a morphism of ringed spaces, the spectrum
defines a contravariant functor Spec Rings C. Since any morphism of ringed spaces Y → X induces
a ring morphism OX (X) → OY (Y ), we also have a contravariant functor C Rings, X OX (X).
For any ring A we have a natural ring isomorphism A = OSpec A (Spec A), and for any affine scheme
X we have a natural isomorphism of locally ringed spaces X = Spec OX (X), hence an isomorphism of
ringed spaces. We conclude that the functor Spec Rings C defines an antiequivalence of categories.
Where does this argument fail?
24. Let A2 = Spec k[x, y]. Show that the complement U of the origin is not an affine scheme.
(Hint: Γ(U, OA2 ) = k[x, y] = Γ(A2 , OA2 )).
25. Let U, V be two affine open sets in an affine scheme X = Spec A. Prove that U ∩ V also is an affine
scheme and that the natural morphism OX (U ) ⊗A OX (V ) → OX (U ∩ V ) is an isomorphism.
26. Prove that the underlying topological space of a noetherian scheme is a projective limit of finite
topological spaces. (Hint: It is a spectral space, p. 228).
27. Let X be a scheme and let OXred be the sheaf of rings defined by the presheaf U OX (U )red . Show
that Xred = (X, OXred ) is a reduced closed subscheme of X with the following universal property: We
have Hom(Y, X) = Hom(Y, Xred ) for any reduced scheme Y .
28. Let X be a scheme. Prove that a sequence 0 → M0 → M → M00 of quasi-coherent sheaves is exact if
and only if so is, for any quasi-coherent sheaf N , the sequence
0 −→ HomOX (N , M0 ) −→ HomOX (N , M) −→ HomOX (N , M00 ).
Prove that a sequence M0 → M → M00 → 0 of quasi-coherent sheaves is exact if and only if so is, for
any quasi-coherent sheaf N , the sequence
0 −→ HomOX (M00 , N ) −→ HomOX (M, N ) −→ HomOX (M00 , N ).
526 EXERCISES
29. Let (X, OX ) be a ringed space. An OX -module M is quasi-coherent if any point admits an open
neighborhood U and a presentation (I, J are arbitrary sets of indices)
⊕J OX |U −→ ⊕I OX |U −→ M|U −→ 0.
Prove that this definition coincides with the definition of p. 311 when X is a scheme.
An OX -module M is of finite type if any point admits an open neighborhood U and an epimorphism
n
OX |U −→ M|U −→ 0,
where n is a natural number, and M is coherent if it is of finite type and, for any open set V and
n
any morphism f : OX |V → M|V , the kernel of f is of finite type. Prove that this definition coincides
with the definition of p. 311 when X is a noetherian scheme.
30. Let X be a noetherian scheme. If M is a coherent sheaf and N is a quasi-coherent sheaf, prove that
HomOX (M, N ) is quasi-coherent. (Hint: HomA (M, N )S = HomAS (MS , NS ) when M is a finitely
generated module over a noetherian ring A).
31. If M is a coherent sheaf on a noetherian scheme X and N is an OX -module, show that
HomOX (M, N )x = HomOX,x (Mx , Nx ).
32. Let f : Y → X be a morphism of noetherian schemes. Prove the following statements:
(a) If M is a quasi-coherent OX -module, then f ∗ M is a quasi-coherent OY -module.

(b) If M is a coherent OX -module, then f ∗ M is a coherent OY -module.
(c) If N is a quasi-coherent OY -module, then f∗ N is a quasi-coherent OX -module. (Hint: When
X is affine, cover Y with a finite number of affine open sets Vi , and cover Vi ∩ Vj with a finite
number of affine open sets Vijk .)
33. Let A be a noetherian ring and put X = Spec A. If M is a finitely generated A-module, show that
HomOX (M e ) = HomA (M, N )∼ .
f, N
34. A morphism of schemes f : X → S is said to be affine if any point s ∈ S has an open affine
neighborhood U = Spec A such that f −1 (U ) = Spec B is affine. If f : X → S is an affine morphism,
prove that,
(a) If M is a quasi-coherent OX -module, then f∗ M is a quasi-coherent OS -module and

H p (X, M) = H p (S, f∗ M) , p ≥ 0.
(b) If g : Y → S is another affine morphism, then HomS (X, Y )(U ) = HomU (X|U , Y |U ) is a sheaf of
sets on S, and the natural morphism HomS (X, Y ) → HomOS -alg (f∗ OX , g∗ OY ) is an isomorphism
of sheaves.
(c) The natural map HomS (X, Y ) → HomOS -alg (f∗ OX , g∗ OY ) is bijective, so that a S-morphism
X → Y is an isomorphism if and only so is the induced morphism f∗ OX → g∗ OY .
(d) Conclude that f −1 (V ) is affine for any affine open set V ⊆ X. In fact, if X → Spec A is an
affine morphism, then X is an affine scheme, so that the category of affine schemes over Spec A
is equivalent to the opposite category of A-algebras.
35. If V ⊂ U are affine open sets in a scheme X, prove that the restriction morphism OX (U ) → OX (V )
is flat, and the natural morphism M(U ) ⊗OX (U ) OX (V ) → M(V ) is an isomorphism for any quasi-
coherent OX -module M.
36. If x1 , . . . , xn are closed points of an affine curve C, prove that C − {x1 , . . . , xn } is affine.
37. If x is a closed point of a non singular complete curve X, show that the line sheaf L−x is just the
sheaf of ideals of the closed subscheme Spec κ(x) → X.
38. Let M be a torsion-free coherent sheaf of rank r on a non singular complete curve X. If D is a divisor
of the line sheaf Λr M, prove that χ(X, M) = r · χ(X, OX ) + deg D.
39. Let (X, OX ) be a ringed space. If M is an OX -module and I is an injective OX -module, prove that
the sheaf HomOX (M, I) is flasque. (Hint: The sheaf MU is an OX -module).
Conclude that ExtnOX (E, M) = H n (X, HomOX (E, M)) when E is a locally free OX -module.
40. If E is locally free sheaf on P1 , prove the following statements:
(a) There is an exact sequence 0 → O(−1) → E(n) → Ē → 0, where H 0 (P1 , E(n)) = 0 and Ē is
locally free.
(b) H 0 (P1 , Ē) = 0, so that Ē = ⊕i O(−mi ), mi ≥ 1, by induction on the rank.
(c) Ext1O (Ē, O(−1)) = 0, so that the exact sequence 0 → O(−1) → E(n) → Ē → 0 splits.
(d) Conclude that E is a direct sum of line sheaves, E ' ⊕i O(ni )
Moreover, show that Ext1O (E, T ) = for any torsion coherent sheaf T , so that any coherent sheaf on P1
is a direct sum of line sheaves and a torsion sheaf.
41. Let Σ be the field of rational functions on a complete non singular curve X over a field k.
If x ∈ X is a rational point and mx = tOx , show that Σ/Ox = kt−1 ⊕ . . . ⊕ kt−n ⊕ . . .
Given closed points x1 , . . . , xr of an affine open set U and Laurent series fi ∈ Σ/Oxi , 1 ≤ i ≤ r, show
the existence of a rational function f ∈ Σ with Laurent expansion fi at xi and no other pole in U .
42. Let π : X → X 0 be a morphism of degree d between non singular complete curves. A closed point
x ∈ X is said to be a ramification point when mx0 OX,x = mrx , r ≥ 2, and rx = r is defined to be the
ramification index of π at x.
(a) Show that d = π(x)=x0 rx at any point x0 ∈ X 0 .
P

(b) If the base field k has null characteristic, prove that l ΩOx /Ox0 = rx − 1.
43. Let C be a complete non singular curve over a field k of null characteristic. Prove that
(a) Any non-constant k-morphism C → P1,k without ramification points is an isomorphism.
(b) If there is a non constant k-morphism P1,k → C, then the genus of C is 0.
(c) Any subfield of k(t) strictly containing k is k(q(t)) for some rational function q(t).
(d) Let C 0 be another complete non singular curve. If there is a non constant k-morphism C 0 → C
without ramification points and C has genus g = 1, then the genus of C 0 is g 0 = 1.
44. Prove that the canonical series K of a complete non singular curve C has no fixed point; that is to
say, h0 (K) > h0 (K − x) for any closed point x ∈ C.
45. Let C be a complete non singular curve over a field k (algebraically closed in the field of rational
functions ΣC , as we always assume). If D is an effective divisor, show that h0 (D) ≤ 1 + deg D, and
that the equality holds if and only if C has genus g = 0.
46. Let D0 , D be two effective divisors on a complete non singular curve C. Prove that the morphism
P(H 0 (C, LD )) × P(H 0 (C, LD0 )) −→ P(H 0 (C, LD+D0 )), (D1 , D2 ) 7→ D1 + D2 ,
has finite fibres, and obtain that h0 (D) + h0 (D0 ) ≤ h0 (D + D0 ) + 1.
If D is a special divisor, H 1 (C, LD ) 6= 0, conclude that h0 (D) ≤ 1 + 21 deg D.
47. Let π : X̄ → X be a non constant morphism between non singular complete curves over an algebraically
closed field k. Since (ΩX , ResX ) represents the functor M H 1 (X, M)∗ and we have the residue map
1 1
ResX̄ : H (X, π∗ ΩX̄ ) = H (X̄, ΩX̄ ) → k, then
we have a morphism of OX -modules tr : π∗ ΩX̄ → ΩX ,
the trace morphism, such that ResX tr(θ) = ResX̄ (θ), ∀θ ∈ H 1 (X, π∗ ΩX̄ ) = H 1 (X̄, ΩX̄ ).
If π −1 (x) = {x̄1 , . . . , x̄n }, prove that for any meromorphic 1-form ω̄ on X̄ we have

Resx tr(ω̄) = Resx̄1 ω̄ + . . . + Resx̄n ω̄.
48. Show the existence of a natural morphism of schemes Proj A → Spec A0 , p 7→ A0 ∩ p, for any graded
ring A = A0 ⊕ A1 ⊕ . . ..
528 EXERCISES
49. Let Aff be the category of affine spaces and affinities over a field k, and let Affsch be the cate-
gory of pairs (P, H), where H is a hyperplane of a k-projective space P ' Proj k[x0 , . . . , xd ], and k-
isomorphisms preserving the hyperplanes. Show that F : Aff Affsch , F (An , V ) = (Proj S • E ∗ , (ω)0 )
is an equivalence of categories, where E is the vector extension of An and ω ∈ E ∗ = F is the linear
form defined by the constant function 1. (Hint: If you are willing to assume that the objects form a
set, you may use problem 50 in p. 483. Otherwise you must define An to be the set of rational points
of P − H and V to be the kernel of the transpose of k = Γ(P, OP ) → Γ(P, LH ) = F ).
50. Let C be a complete non-singular conic over a field k. If C has no rational point, prove that any closed
point x ∈ C has even degree [k : κ(x)], and that deg : Pic C → 2Z is an isomorphism.
51. Let X be a closed subscheme of Pn,k defined by an homogeneous ideal (p1 , . . . , pn ) of k[x0 , . . . , xn ],
di = deg pi . If dim X = 0, prove Bézout’s theorem: dimk H 0 (X, OX ) = d1 . . . dn .
52. Let X = Proj k[x0 , . . . , xd ]/(Pn ) be an hypersurface of degree n in Pd , d ≥ 2. Prove that
(a) H 0 (X, OX ) = k, so that X is connected.
(b) dim H d−1 (X, OX ) = n−1

d .
(c) H p (X, OX ) = 0 when p 6= 0, d − 1.
53. Max Noether’s Theorem: Let Pn (x0 , x1 , x2 ) = 0 and Qm (x0 , x1 , x2 ) = 0 be two plane curves with
no common irreducible component, and let I be the sheaf of ideals of the intersection. If a curve
Rd (x0 , x1 , x2 ) = 0 passes through any common point and it is locally in the ideal of the intersection,
in the sense that Rd ∈ Γ(P2 , I(d)), prove that Rd ∈ (Pn , Qm ).
54. Let X be a projective variety over a field. If M is a coherent OX -module, prove the existence of a
polynomial H(n) with rational coefficients such that H(n) = χ(X, M(n)). (Hint: Induction on the
dimension of supp M).
55. Now we put R = k[x0 , . . . , xd ], we consider the irrelevant ideal m = (x0 , . . . , xd ), and M will be a
finitely generated graded R-module. We say that M is a free graded module if it admits a basis of
homogeneous elements, M ' ⊕i R(ni ). Prove the following statements:
(a) If mM = M , then M = 0.
(b) If TorR
1 (R/m, M ) = 0, then M is a free graded R-module.
(c) M admits a free graded resolution 0 → Ld+1 → . . . → L0 → M → 0.
(d) Let A = k[x1 , . . . , xd ] = R/(x0 − 1) and let M̄ be a finitely generated A-module. We have
M̄ ' M ⊗R A for some finitely generated graded R-module M . (Hint: Applying the base change
R → A to graded morphisms between free graded R-modules we may obtain any morphism
between free A-modules, just homogenize the matrix).
(e) TorR
p (R/(x0 − 1), M̄ ) = 0, p ≥ 1.
(f) M̄ admits a free resolution 0 → L̄d+1 → . . . → L̄0 → M̄ → 0.

(g) The K-group of finitely generated A-modules is K(k[x1 , . . . , xd ]) = Z.
56. Let (O, m) be an integral noetherian local ring of dimension 1, and put
Rm O = O ⊕ m ⊕ m2 ⊕ . . .
Gm O = (O/m) ⊕ (m/m2 ) ⊕ . . .
Prove that Proj Gm O is finite, and that there exists ξ ∈ md not vanishing at any point of it.
Prove that the blow-up C1 = Proj Rm O is an affine scheme, C1 = Spec O1 .
h i
Moreover, if md = (ξ, ξ1 , . . . , ξr ), then O1 = O ξξ1 , . . . , ξξr is a finite O-algebra.
57. Let k be a noetherian ring. Prove the following statements:

(a) If M is a finitely generated k-module and N is a finitely generated flat k-module, then there is a
k-scheme Homk (M, N ) such that for any k-scheme X we have

HomOX (M ⊗k OX , N ⊗k OX ) = Homk X, Homk (M, N ) .
(b) If A is a finite flat k-algebra, the functor X HomOX -alg (A ⊗k OX , A ⊗k OX ) is representable

by a k-scheme Endk-alg (A),

HomOX -alg (A ⊗k OX , A ⊗k OX ) = Homk X, Endk-alg (A) ,
and the functor X AutOX -alg (A ⊗k OX , A ⊗k OX ) is representable by an open subscheme
Autk-alg (A) of Endk-alg (A),

AutOX -alg (A ⊗k OX , A ⊗k OX ) = Homk X, Autk-alg (A) .
√ √
58. Determine Autk-alg (A) when k = Q, A = Q( n 2 ); and also when k = F2 (t), A = k( t ).
59. Let C be an elliptic curve (complete non singular curve of genus 1) over an algebraically closed field
k. Prove the following statements (where all the considered points are closed):
(a) C is isomorphic to a non singular plane cubic.
(b) Given two points x, y ∈ C, we have y = τ (x) for some involution τ : C → C. (Hint: Construct a
morphism π : C → P1 of degree 2 such that π(x) = π(y)).
(c) Given morphisms π1 , π2 : C → P1 of degree 2, prove the existence of automorphisms τ : C → C
and σ : P1 → P1 such that the following square commutes:
C
τ /C
π1 π2

P1
σ / P1
(d) If char k 6= 2, any morphism C → P1 of degree 2 ramifies at 4 points P1 , P2 ; P3 , P4 ∈ P1 and, if
we put λ = (P1 , P2 ; P3 , P4 ), the elliptic curve is classified by the j-invariant (p. 491)
(λ2 − λ + 1)3
j(λ) = ·
λ2 (λ − 1)2
(e) If we fix a point x ∈ C, any line sheaf of degree 0 is Ly−x for a unique point y ∈ C.
60. Let C be a non singular plane cubic over an algebraically closed field. Prove that the tangent lines at
three collinear points intersect C at collinear points.
Moreover, if a straight line passes through two inflexion points (points where the tangent intersects
with multiplicity greater than 2) of C, so is the third
61. Let U = Spec A = C − {p1 , . . . , pr } be an affine open set of a complete non singular curve C over an
algebraically closed field k. Prove that the following conditions are equivalent:
(a) The ring A is a principal ideal domain.
P
(b) Pic(C) = i Zpi .
(c) The curve C has genus 0.
Conclude that the following conditions are equivalent:
(a) The ring k[x, y]/(p(x, y)) is a principal ideal domain.
(b) The polynomial p(x, y) is irreducible and p(x, y) = 0 is a non singular curve of genus 0.
(c) The ring k[x, y]/(p(x, y)) is a Euclidean ring.
62. Show that no real maximal ideal of R[x, y]/(x2 + y 2 − 1) is a principal ideal.
63. Let X be a non singular complete curve of genus g ≥ 2 over an algebraically closed field k. Prove that
(a) The canonical divisors of X define a closed embedding X ,→ Pg−1 if and only if for any pair of
closed points p, q we have h0 (LK−p−q ) = g − 2, or equivalently, h0 (Lp+q ) = 1.
(b) This canonical embedding identifies X with a curve of degree 2g − 2 in Pg−1 , well defined up to
projectivities (so reducing the classification of curves of genus g to the projective classification of
such canonical curves), except when X is hyperelliptic (it admits a projection π : X → P1 of
degree 2).
530 EXERCISES
p k 6= 2, this projection π : X → P1 of degree 2 ramifies at 2g + 2

(c) If X is hyperelliptic and char
points, so that ΣX = k x, P (x) for some polynomial P (x) of degree 2g + 2. (Hint: Hurwitz’s
formula).
64. Let X = Proj k[x0 , x1 , x2 ]/(Pn ) be a plane curve of degree n. Show that we have an exact sequence
n P ·
0 −→ OP2 (−n) −−− → OP2 −→ OX −→ 0
n−1

and obtain that π := dimk H 1 (X, OX ) = dimk H 2 (P2 , OP2 (−n)) = 2 .
When k is algebraically closed, if my denote the multiplicity of any point y of the blow-up tree Ax (p.
208) at a singular point x, conclude that the geometric genus g of X is
X X
n−1 my
g= − .
2 2
x∈X y∈Ax
65. Show that the dualizing sheaf DX of a complete curve X is a torsion free coherent sheaf of rank 1.
(Hint: Given a non-singular closed point x, show that S(n) = dimk H 0 (X, DX ⊗OX Lnx ) goes to ∞
as a polynomial of degree 1). Conclude that, if ωX is a torsion free coherent sheaf of rank 1 such that
h0 (X, ωX ) = h1 (X, OX ) and h1 (X, ωX ) = h0 (X, OX ), then ωX ' DX .
12. Algebraic Topology I

1. Let U be an open set in a topological space X and put Y = X − U . If F is a sheaf on X, show that
FY = ZY ⊗Z F and FU = ZU ⊗Z F.
2. Let F1 (t, x1 , x2 , x3 )∂1 + F2 (t, x1 , x2 , x3 )∂2 + F3 (t, x1 , x2 , x3 )∂3 be a smooth vector field on the comple-
ment U ⊂ R4 of a finite number of non-intersecting trajectories (smooth sections of the first projection
t : R4 → R). If it is a conservative force field, in the sense that ∂j Fi = ∂i Fj , i, j = 1, 2, 3, prove that
Fi = ∂i u, i = 1, 2, 3, for some smooth function u(t, x1 , x2 , x3 ) on U .
3. Given a closed cover X = Y1 ∪ Y2 , show that χ(Y1 ∪ Y2 ) = χ(Y1 ) + χ(Y2 ) + χ(Y1 ∩ Y2 ), in case that all
the considered Euler-Poincaré characteristics are finite.
If X, Y are connected compact manifolds of dimension n, conclude that χ(X#Y ) = χ(X) + χ(Y ) − 2
when n is even, and χ(X#Y ) = χ(X) + χ(Y ) when n is odd.
4. Let ωp be a closed differential p-form
R on the sphere Sn . If p < n, show that ωp is exact. If p = n,
show that ωp is exact if and only if Sn ωp = 0.
5. If A is a boolean algebra, prove that any sheaf on X = Spec A is acyclic.
6. Let f : S1 → S1 be the covering f (z) = z 2 . Show that the sheaf f∗ Z is not constant, even if any stalk
of f∗ Z is isomorphic to Z2 .
7. If I, J are two ideals of a ring A, show that
(a) TorA
1 (A/I, A/J) = (I ∩ J)/IJ.
(b) TorA A
2 (A/I, A/J) = Tor1 (I, A/J).
(c) TorA A
n+1 (A/I, A/J) = Torn (I, A/J) = Torn−1 (I, J), n ≥ 2.
8. Prove that Ext1Z (Z/nZ, Z/mZ) ' Z/dZ, where d = m.c.d.(n, m), and Ext1Z (Z/nZ, Z) ' Z/nZ.
9. Prove that the functor TorA
n (−, N ) preserves inductive limits.
10. Prove that the Tor functors localize, TorA AS

n (M, N )S = Torn (MS , NS ).
11. If A → B is a flat morphism, show that TornB (M ⊗A B, N ⊗A B) = TornA (M, N ) ⊗A B.

12. Let 0 → M → R• be a resolution of an A-module, and F : A-mod B-mod a left-exact additive
functor. If 0 → M → A• is a F -acyclic resolutions, and f, g : R• → A• are morphisms of complexes
such that we have commutative diagrams
ALGEBRAIC TOPOLOGY I 531
M
∼
/ R• M
∼
/ R•
o o
g
} f
}
A• A•
prove that the induced morphisms f, g : H n [F (R• )] → H n [F (A• )] coincide. (Hint: Both coincide with
the composition of DR : H n [F (R• )] → Rn F (M ) with the inverse of DR : H n [F (A• )] −∼
→ Rn F (M )).
Conclude that, in the case of the functorial injective resolution 0 → M → I • M , the natural morphism
DR : H n [F (I • M )] → Rn F (M ) is the identity.
Given an exact sequence 0 → M → M0 → M1 → 0, resolutions 0 → M0 → R0• , 0 → M1 → R1• and a
commutative diagram with exact rows
0 / R• / R0• / R1• /0
O O O
0 /M / M0 / M1 /0
prove that we have a commutative square
H n [F (R• )]
δ / H n+1 F (R1• )]
DR DR

Rn F (M )
δ / Rn+1 F (M1 )
so that the connecting morphism δ : Rn F (M1 ) → Rn+1 F (M ) is independent of the F -acyclic resolu-
tions used to define it. (Hint: Apply F to the following commutative diagram with exact rows:)
0 /M / M0 / M1 /0
'/ '/ /' R1• /0
0 R• R0•

0 / I •M / I • M0 / I • M1 /0
' ' '
0 / I • R• / I • R0• / I • R1• /0
State and prove analogous results when M is replaced by a bounded below complex K • .
i p
Given an exact sequence of complexes 0 → K • → − K0• −→ K1• → 0, prove that we have a natural
•
quasi-isomorphism p : Cone i − ∼ •
→ K1 such that the connecting morphism δ : H n (K1• ) → H n+1 (K • )
is just the opposite of the morphism H n (K1• ) = H n (Cone• i) → H n (K • [1]) = H n+1 (K • ). Use this
fact to prove again that the connecting morphism δ : Rn F (K1• ) → Rn+1 F (K • ) is independent of the
F -acyclic resolutions used to define it.
13. If f : Y → X is a continuous map, show that f ∗ ZX = ZY .
14. If f : Y → X is a continuous map and F is a sheaf on X, show that f ∗ F is the associated sheaf of the
presheaf V lim
−→
F(U ).
f (V )⊆U
15. Let X̄ be a topological space X endowed with the discrete topology, and let j : X̄ → X be the obvious
continuous map. Show that C 0 F = j∗ j ∗ F for any sheaf F on X.
16. Prove that the cohomology groups H 1 (X, Z), Hc1 (X, Z) and HY1 (X, Z) are always torsion-free. (Hint:
The exact cohomology sequence of the exact sequence 0 → Z → Q → Q/Z → 0).
17. If Y ⊆ X is a locally closed subspace and F is a sheaf on Y , prove that there is a unique sheaf F X on
X such that F X |Y = F and F X |X−Y = 0, namely F X (U ) = {s ∈ F(U ∩ Y ) : supp (s) is closed in U }.
18. If j : Y → X is a closed subspace and F is a sheaf on X, show that FY = j∗ j ∗ F. If j : U → X is an
open subset of a σ-compact space X and F is a sheaf on X, show that FU = j! j ∗ F.
19. Excision: Let Y ⊆ X be a closed subset and let F be a sheaf on X. If U is a neighborhood of Y ,
show that HYp (X, F) = HYp (U, F|U ).
532 EXERCISES
20. Let O be a sheaf of rings on a topological space X. If M is an O-module, show that the cohomology
groups H p (X, M), Hcp (X, M) and HYp (X, M) have a natural structure of O(X)-module.
21. Let j : U → X be an open subset of a topological space X and let i : Y = X−U → X be the complemen-
tary closed set. If O is a sheaf of rings on X, show that OU and OY are O-modules, and prove that for
any O-module M we have natural isomorphisms HomO (OU , M) = M(U ), HomO (OU , M) = j∗ j ∗ M,
HomO (OY , M) = ΓY (X, M), HomO (OY , M) = ΓY M.
22. Let (X, O) be a ringed space. If M, N are O-modules, then TorO n (M, N ) will denote the associated
TorO(U )

sheaf of the presheaf U n M(U ), N (U ) , and it is endowed with a natural structure of O-
module. If 0 → M0 → M → M00 → 0 is an exact sequence of O-modules, prove that we have a long
exact sequence of O-modules
δ
. . . −→ TorO 00
−→ M0 ⊗O N −→ M ⊗O N −→ M00 ⊗O N −→ 0.
1 (M , N ) −
23. Let (X, O) be a ringed space. Show that the product

∪
H p (X, O) ⊗ H q (X, O) −−→ H p+q (X, O ⊗ O) −→ H p+q (X, O)
defined by the cup product and the natural morphism O ⊗ O → O defines a structure of anticom-
mutative graded O(X)-algebra on H • (X, O) = ⊕p H p (X, O). Moreover, if M is an O-module, the
following product defines a structure of graded H • (X, O)-module on H • (X, M) = ⊕p H p (X, M):
∪
H p (X, O) ⊗ H q (X, M) −−→ H p+q (X, O ⊗ M) −→ H p+q (X, M)
24. Show that the tangent bundle T X and the cotangent bundle T ∗ X of a smooth manifold X are always
orientable manifolds.
25. If M is a finitely generated module over a local ring (O, m), prove that all minimal generating systems of
M have the same number of elements g(M ) = dimO/m (M/mM ). If M is a finitely generated A-module,
show that the function f (x) = g(Mx ) is semicontinuous on Spec A; i.e., {x ∈ Spec A : g(Mx ) < n} is
an open subset for any n ∈ N. Moreover, the following conditions are equivalent:
(a) M is a projective A-module.
(b) Any point of Spec A has a basic neighborhood Uf such that Mf is a free Af -module.
(c) M is a flat A-module and the function f (x) = g(Mx ) is locally constant on Specm A.
26. Let E → X be a real vector bundle of rank n over a compact separated space X. Prove that the
C(X)-module of all continuous sections of E is a finitely generated projective C(X)-module.
27. Prove that a real line bundle L → X over a (separated) topological manifold X is trivial if and only
if for any continuous map γ : S1 → X we have that the line bundle γ ∗ L → S1 is trivial.
28. Let τg be a torus of genus g. Prove that the cup product defines a non-singular alternate metric
H 1 (τg , Z) × H 1 (τg , Z) −→ H 2 (τg , Z) = Z.
Let πg be the connected sum of g real projective planes. Prove that the cup product defines a non-
singular metric H 1 (πg , F2 ) × H 1 (πg , F2 ) → H 2 (πg , F2 ) = F2 .
29. Prove that g is the maximum number of disjoint circles in τg (and in πg ) with connected complement.
30. Let Y be a closed set of a topological space X. If F is a flasque sheaf on X, prove that so is ΓY F.
31. Let j : U → X be an open subset of a topological manifold. If Y is a normally orientable closed
submanifold of X, prove that pU (Y ∩ U ) = j ∗ pX (Y ) .

32. Check that the topological intersection number of a real conic with a tangent line in R2 is 0, while the
topological intersection number of a conic with a tangent line in the complex plane C2 is 2.
33. If X, Y are topological varieties, prove that TX×Y = (π1∗ TX ) ⊗Z (π2∗ TY ).
34. If j : Y → X is a closed submanifold of a topological manifold X, prove that TY = TY /X ⊗Z j ∗ TX .
35. Prove that any topological manifold of dimension 1 is orientable. Conclude that any non-compact
connected smooth curve is diffeomorphic to R. (Hint: Consider the integral curves of a vector field
without zeros).
ALGEBRAIC TOPOLOGY I 533
36. Define a direct image i∗ : H p (Y, TY ) → H p+d (X, TX ) for any closed submanifold i : Y → X of constant
codimension d, and prove that (ij)∗ = i∗ j∗ for any other closed submanifold j : Z → Y of constant
codimension.
37. Let (X, OX ) be a smooth manifold of dimension n. Prove that TX ⊗Z OX = ΩnX .
If p is the sheaf of ideals of a closed smooth submanifold Y → X of codimension d, and N = (p/p2 )∗
is the sheaf of smooth sections of the normal bundle N , show that TY /X ⊗Z OY ' Λd N .
Conclude that Y is normally orientable in X if and only if N admits a smooth orientation.
38. Let I = [a, b] be a closed interval and put U = (a, b). Show that the dualizing complex DI has null
cohomology sheaves, except H−1 (DI ) = ZU .
39. Determine the cohomology sheaves Hp (DX ) when X is the union of a plane and an intersecting line.
40. Let Y be a closed set in a sphere Sn , n ≥ 2. Prove the following statements:
(a) If Y is homeomorphic to a ball, Y ' Br , then Sn − Y is connected.
(b) If Y is homeomorphic to sphere, Y ' Sr , then r ≤ n, and Y = Sn when r = n.
(c) If Y ' Sn−1 , then Sn − Y has two connected components.
(d) If Y ' Sr , r < n − 1, then Sn − Y is connected.
41. Let Y be a closed set in Rn , n ≥ 2. If Y ' Sn−1 , show that Rn − Y has two connected components,
one bounded and the other not. If f : Bn → Rn is an injective continuous map, prove that f maps
the interior of Bn onto the bounded connected component of Rn − f (Sn−1 ).
42. Prove that any continuous injective map f : Rn → Rn is an open map.
43. Let X be a compact oriented smooth manifold of dimension n. Prove that we have a perfect pairing
(identifying each factor with the dual of the other factor)
Z
H p (X, R) ⊗R H n−p (X, R) −→ R , hω, ηi = ω ∧ η.
X
44. Let X be a smooth manifold of dimension n. Prove that X admits a metric g of constant signature
(1, n−1) if and only if X admits a distribution of rank 1. (Hint.: Consider the endomorphism attached
to g and a riemannian metric, and the eigenvectors with positive eigenvalue.)
Conclude that the sphere S4 admits no Lorentz metric. (Hint.: χ(S4 ) 6= 0).
45. Let us consider the roots c(E) = (1 + α1 ) . . . (1Q + αr ) and c(F ) = (1 + β1 ) . . . (1 + βQ
s ) of two complex
vector bundles E, F . Prove that c(E ⊗ F ) = (1
i,j Q + αi + β j ), c(Hom(F, E)) = i,j (1 + αi − βj ),
c(Λp E) = i1 <...<ip (1 + αi1 + . . . + αip ), c(S p E) = n1 +...+nr =p (1 + n1 α1 + . . . + nr αr ).
Q
46. If X is a closed smooth hypersurface in Rn , prove that wi (X) = 0. (Hint: wi (X) = w1 (X)i and X is
orientable).
47. Prove that Pn,R admits a tangent vector field without zeroes if and only if n is odd.
48. Let Y be a d-dimensional connected compact topological manifold with boundary. Prove that the
connecting morphism δ : H d (∂Y, F2 ) → H d+1 (Y − ∂Y, F2 ) is an isomorphism, and that the restriction
morphism j ∗ : H d (Y, F2 ) → H d (∂Y, F2 ) is null.
Given a sequence I = (n1 , . . . , nr ) ∈ Nr , if we put n = n1 + 2n2 + . . . + rnr , we have a Stieffel-Whitney
number wI (X) = w1 (X)n1 . . . wr (X)nr ∈ H n (X, F2 ) = Z/2Z for any connected compact smooth
manifold X of dimension n. Prove that all Stieffel-Whitney numbers are 0 when X is the boundary
of a connected compact smooth manifold Y . (Hint: Prove that wi (X) = j ∗ wi (Y ), using the exact
sequence 0 → T X → j ∗ (T Y ) → NX/Y → 0).
49. Any complex vector bundle V → X of rank r has an underlying oriented real vector bundle VR → X
of rank 2r, and a conjugate vector bundle V̄ → X, where the new complex structure is α · v = ᾱv.
Moreover, any real vector bundle E → X of rank r defines a complex vector bundle EC = E ⊕ iE → X
of rank r. Prove the following statements:
(a) EC = EC .
534 EXERCISES
(b) (VR )C = V ⊕ V̄ .
(c) If E is an oriented vector bundle of rank r, the natural isomorphism (EC )R = E ⊕ E preserves
the orientation if and only if 2r is even.
(d) ci (V̄ ) = (−1)i ci (V ).
(e) 2ci (EC ) = 0 when i is odd.
50. The Pontrjagin classes of a real vector bundle E → X are defined to be
pi (E) = (−1)i c2i (EC ) ∈ H 4i (X, Z),
P
and p(E) = i pi (E) is the total Pontrjagin class. For a smooth manifold X, we put pi (X) = pi (TX ).
Prove the following properties of these classes:
(a) p(f ∗ E) = f ∗ p(E) for any continuous map f : Y → X.

(b) 2 p(E ⊕ F ) − p(E) · p(F ) = 0.
(c) pi (VR ) = j (−1)j ci−j (V )ci+j (V ) for any complex vector bundle V .
P
(d) pi (Pn,C ) = n+1

i
i x , where x is the cohomology class of an hyperplane.
51. Prove that the category of sheaves on a finite topological space X is equivalent to the category
of systems with indices in the (eventually non-filtering) ordered set X (families of abelian groups
{Fx }x∈X endowed with morphisms ρxy : Fx → Fy when x ≤ y, so that ρxx = Id and ρxz = ρyz ρxy
whenever x ≤ y ≤ z). Moreover, show that Γ(X, F) = lim ←−
Fx .
x∈X
52. Derived Functors of the projective limit: Endow N with the open sets Un = {0, 1, . . . , n}.
Show that any sheaf F on N defines a projective system Fn = F(Un ), where Fn+1 → Fn are the
restriction morphisms, and that the global sections are Γ(N, F) = lim
←−
Fn .
Prove that so we obtain an equivalence of the category of sheaves of abelian groups on N with the
category of projective limits of abelian groups with indices in N. Moreover:
(a) F is flasque if and only if the morphisms Fn+1 → Fn are surjective for all n ∈ N.
p
(b) The derived functors lim
←−
of lim
←−
vanish when p > 1.
(Hint: In the Godement resolution 0 → F → C 0 F → G → 0, check that G is flasque).
1
(c) Any projective system (Fn ) with a finite number of non-null terms is acyclic, lim
←−
(Fn ) = 0.
(Hint: Γ(N, C 0 F) = Γ(N, G)).
53. Let {Xn } be an inductive system of topological spaces satisfying one of the following conditions:
(a) The morphism Xn → Xn+1 is an open embedding for all n ∈ N.
(b) Xn is a compact separated space, and Xn → Xn+1 is a closed embedding, for all n ∈ N.
Prove that for any sheaf F on X = lim
−→
Xn we have a convergent spectral sequence
p,q p

E2 = lim
←−
H q (Xn , F|Xn ) ⇒ H p+q (X, F).
(Hint: Γ(X, F) = lim
←−
Γ(Xn , F|Xn ) and the projective system is surjective when F is flasque).
Calculate the cohomology groups H p (R∞ , Z), H p (P∞,R , F2 ), H p (P∞,C , Z) and H p (S∞ , Z); where we
put R∞ = lim
−→
Rn , P∞,R = lim
−→
Pn,R , P∞,C = lim
−→
Pn,C and S∞ = lim −→
Sn .
13. Analysis IV
1. In Rn , put B = B(0, 1), S = S(0, 1) and U = B − {0}, so that ∂U = S ∪ {0}. Show that the
continuous function f : ∂U → R such that f (S) = 0, f (0) = 1, admits no continuous extension
u : Ū → R harmonic on U .
2. Prove that a function u ∈ C 2 (U ) is subharmonic if and only if ∆u ≥ 0.
ANALYSIS IV 535
3. Show that the function |x|a is subharmonic when a ≥ 1.

If u > 0 is an harmonic function, show that ua , a ≥ 1, and − ln u are subharmonic.
4. Let f ∈ C 2 (R) such that f 00 ≥ 0. If u is an harmonic function, prove that f (u) is subharmonic.
5. Let Ω ⊂ R2 be a relatively compact connected open set. If there is a closed segment I such that Ω̄ ∩ I
is just a point y0 , prove that y0 is a regular point.
6. Let U ⊂ Rd be a relatively compact connected open set. If any continuous function f ∈ C(∂U ) has an
extension u ∈ C(Ū ) harmonic on U , show that any boundary point y0 ∈ ∂U is a regular point. (Hint:
Consider f (x) = |x − y0 |).
7. Show that any two cylinders C/Ze, C/Zv are isomorphic Riemann surfaces.
8. Show that any analytic automorphism τ : D → D of finite order has a fixed point. (Hint: Brouwer’s
fixed point theorem).
9. If the universal covering of a connected open set U ⊂ C is isomorphic to C, prove that U = C − p for
some point p ∈ C.
10. Let T = C/(Ze + Zv) be a torus. Prove that any analytic morphism C → T is constant or surjective.
11. If a Riemann surface X admits a non-constant analytic morphism C → X, show that X is isomorphic
to C, P1 , C − 0 or a torus.
12. Let N be the closure of 0 in a topological vector space E. Prove that E/N is separated, and that any
linear continuous map f : E → F into a separated topological vector space F uniquely factors through
the canonical projection π : E → E/N .
13. Prove that arbitrary direct products of locally convex spaces are locally convex.
Moreover, the topology of any locally convex space E is the initial topology of some family of linear
maps E → Bi to Banach spaces (complete normable spaces). In particular, if E is separated, it is
isomorphic to a vector subspace of a direct product of Banach spaces.
14. If E is a seminormed space, prove that a set is bounded if and only if it is contained in a ball. Moreover,
a locally convex space is seminormable if and only if it admits a bounded neighborhood of 0.
15. Prove that a linear map f of a seminormable space E to a topological vector space F is continuous if
and only if f transforms the unit ball of E onto a bounded set of F .
16. Let E, F be seminormed spaces. Prove that a linear map f : E → F is continuous if and only if
kf k := supkek≤1 kf (e)k < ∞, and that so we obtain a seminorm on the space L(E, F ) of continuous
linear maps. Moreover, L(E, F ) is separated (resp. complete) when so is F .
17. Prove that any finite-dimensional real vector space E admits a unique separated linear topology: the
initial topology of the linear maps E → R. Conclude that, for any vector subspace V ⊆ E, there
is a unique linear topology on E such that 0̄ = V : the initial topology of the canonical projection
π : E → E/V , when we consider on E/V the unique separated linear topology.
18. Prove that countable direct products and projective limits of Fréchet spaces are Fréchet.
19. If two closed subspaces V1 , V2 of a Fréchet space E are algebraic supplements (V1 ∩V2 = 0, V1 +V2 = E),
prove that they are topological supplements (the natural map V1 × V2 → E is an isomorphism).
20. If 0 → E1 → E → E2 → 0 is a topologically exact sequence of metrizable locally convex spaces,
prove that so is the induced sequence of Fréchet spaces 0 → E b1 → E b → Eb2 → 0. (Hint: The
topology of E2 = E/E1 is induced by E/E1 , so that we have a continuous section s : E
b b b2 → E/
b Eb1
of p : E/E1 → E2 . The continuous linear bijection s + Id : E2 × (Ker p) → E/E1 is an isomorphism;
b b b b b b
hence s(Eb2 ) is closed and Ker p = 0).
21. Prove that a continuous linear map f from a separated locally convex space E onto a Fréchet space
F is a homomorphism if and only if it is a weak homomorphism. (Hint: Any closed equicontinuous
family is weakly compact and, when F is Fréchet, any weakly compact set in F 0 is equicontinuous).
536 EXERCISES
22. Let E, F, G be Fréchet spaces, f : E → G a continuous linear map, and h : F → G a compact linear
map. If f (E) + h(F ) = G, show that f is a homomorphism onto a closed subspace of G of finite
codimension.
23. Let f : E → F be a weak homomorphism between separated locally convex spaces, such that every
compact disk in F is the image by f of a compact disk in E. If h : E → F is a compact linear map,
prove that f + h is a weak homomorphism onto a closed subspace of F of finite codimension. (Hint:
Follow the proof of Schwartz theorem).
24. If f : E → F is an isomorphism of separated locally convex spaces, and h : E → F is a compact linear

map, show that f + h : E → F is a homomorphism onto a closed subspace of F of finite codimension,
and the kernel is finite dimensional.
25. Let E be a complex topological vector space.

If E is separated and finite-dimensional, show that E is isomorphic to Cn .
If U ⊂ E is a convex open set, and V a linear subvariety not meeting U , show that there is a closed
hyperplane containing V and not meeting U .
If E is separated and locally convex, and 0 6= e ∈ E, show that ω(e) 6= 0 for some continuous linear
form ω : E → C.
If E is locally convex, show that any closed vector subspace is weakly closed.
26. If ω is a closed smooth complex 1-form on an open set U ⊆ C, prove that there is an analytic function
f on U such that [f (z)dz] = [ω] ∈ H 1 (U, C).
27. Show that the sheaf ΩX of analytic 1-forms on a Riemann surface X admits the acyclic resolution
d
0 −→ ΩX −→ Ω1,0 1,1
X −−−→ ΩX −→ 0,
where Ω1,1
X denotes the sheaf of complex smooth 2-forms. Moreover, if D ⊆ C is an open disk, we have
an exact sequence
d
0 −→ Ω(D) −→ Ω1,0 (D) −−−→ Ω1,1 (D) −→ 0.
S
28. Given an open cover X = i Ui of a Riemann surface, and meromorphic functions fi ∈ M(Ui )
such that fj − fi ∈ O(Ui ∩ Uj ), the additive Cousin’s problem asks for a meromorphic function
f on X such that f − fi ∈ O(Ui ); i.e. whether the natural map H 0 (X, MX ) → H 0 (X, MX /OX )
is surjective. Show that the answer is positive when H 1 (X, OX ) = 0. The multiplicative Cousin’s
problem assumes that fj /fi is a non-vanishing analytic function and asks for a meromorphic function
f on X such that f /fi is a non-vanishing analytic function on Ui ; i.e. whether the natural map
H 0 (X, M∗X ) → H 0 (X, M∗X /OX ∗ ∗
) is surjective. Show that the answer is positive when H 1 (X, OX ) = 0.
S
29. Let X = i Ui be an open cover of a topological space X, and F a sheaf on X. Prove the existence
of a natural morphism φ : Ȟ 1 (U, F) ,→ H 1 (X, F).
(Hint: Consider the Godement resolution 0 → F → C • F, and a Cech 1-cycle (sij ). Inductively we
construct fi ∈ (C 0 F)(Ui ) such that sij = fj − fi . Now dfj = dfi + dsij = dfi , and we have a section
ω ∈ (C 1 F)(X) such that dω = 0; hence a cohomology class φ([sij ]) = [ω] ∈ H 1 (X, F). If we fix other
fi0 ∈ (C 0 F)(Ui ) such that sij = fj0 − fi0 , then fj0 − fj = fi0 − fi , and we have f ∈ (C 0 F)(X) such that
f = fi0 − fi ; hence ω 0 = ω + df and [ω 0 ] = [ω]. Finally, when (sij ) is a boundary, sij = sj − si , we take
fi = si , so that ω = 0.)
Prove that φ : Ȟ 1 (U, F) ,→ H 1 (X, F) always is injective.
(Hint: If ω = df , f ∈ (C 0 F)(X), then d(fi − f ) = 0, so that si := fi − f ∈ F(Ui ). Then f = fi − si =
fj − sj , and sij = fj − fi = sj − si is a boundary).
Conclude that Ȟ 1 (U, F) = H 1 (X, F) when H 1 (Ui , F) = 0, ∀i ∈ I.
(Hint: Given [ω] ∈ H 1 (X, F), we have ω = dfi , for some fi ∈ (C 0 F)(Ui ). Then d(fj −fi ) = ω −ω = 0,
so that sij := fj − fi ∈ F(Ui ∩ Uj ), and sij + sjk = fj − fi + fk − fj = fk − fi = sik ).
DIFFERENTIAL GEOMETRY II 537
30. Let Ω(1,1) be the sheaf of complex smooth 2-forms on a compact Riemann surface X. Prove that we
have Ω(1,1) = Ω(1,0) ⊗OX Ω(0,1) , and H 1 (X, ΩX ) = Ω(1,1) (X)/dΩ(1,0) (X).
Conclude that we have a non-null linear map X : H 1 (X, ΩX ) → C, [ω2 ] 7→ X ω2 , and use it to prove
R R
that ΩX is isomorphic to the dualizing sheaf D. (Hint: Consider a local coordinate z centered at a
point x ∈ X and a function f (r2 ), r2 = z z̄, with compact support, such that f (0) = 1, f 0 (r2 ) ≤ 0.
Show that ω = f (r2 ) dz
z ∈ (Ω
(1,0)
⊗ Lx )(X) and
(1,1)
dω = f 0 (r2 )dz̄ ∧ dz ∈ Ω(1,1) (X) ⊂ (ΩX ⊗ Lx )(X),
so that ξ(dω) = 0, while X dω = X f 0 (r2 )dz̄ ∧ dz = 2i X f 0 (r2 )dx ∧ dy 6= 0).
R R R
14. Differential Geometry II

1. Let X be a (eventually non σ-compact) smooth manifold. Prove that the module D(X) of smooth
∞
tangent vector fields is just the module of R-derivations of the structural sheaf7 CX ; i.e., of morphisms
∞ ∞ ∞ ∞
of sheaves D : CX → CX such that D : C (U ) → C (U ) is an R-derivation for any open set U ⊆ X.
∞
Prove that the sheaf of 1-forms Ω1X on X is HomCX ∞ (DX , C
X ).
Show that the sheaf of (p, q)-tensor fields is Ω1X ⊗CX p
∞ . . . ⊗C ∞ Ω
X
1 q
X ⊗CX DX ⊗CX . . . ⊗CX DX .
∞ ∞ ∞
2. Let A be a k-algebra, char k = 0, and put D = Derk (A, A). A connection on an A-module M is a
group morphism ∇ : D → Endk (M ) such that (aD)∇ = a(D∇ ) and D∇ (am) = (Da)m + a(D∇ m).
Prove the following statements:
(a) A natural connection on M = A is D∇ a = Da. Moreover, given connections (Mi , ∇i ), so are
⊕i Mi , D∇ ( i mi ) = i D∇i mi .
P P
M1 ⊗A . . . ⊗A Mn , D∇ (m1 ⊗ . . . ⊗ mn ) = i m1 ⊗ . . . ⊗ D∇i mi ⊗ . . . ⊗ mn .
P
HomA (M1 , M2 ), (D∇ f )(m) = D∇2 f (m) − f (D∇1 m).

M ∗ = HomA (M, A), (D∇ ω)(m) = D(ω(m)) − ω(D∇ m).

0
MultA (M1 , . . . , Mn ; M 0 ), (D∇ T )(m1 , . . . , mn ) = D∇ (T (m1 , . . . , mn )) − i T (. . . , D∇i mi , . . .).
P
(b) Consider the A-module of A-multilinear maps Tp : D× . p. . D → M and the Lie derivative
(DL Tp )(D1 , . . . , Dp ) = D∇ (Tp (D1 , . . . , Dp )) − i Tp (D1 , . . . , [D, Di ], . . . , Dp ).

P
Then, (D1 + D2 )L = D1L + D2L , DL (Tp + Tp0 ) = DL Tp + DL Tp0 , DL (aTp ) = (Da)Tp + a(DL Tp ),
(λD)L = λ(DL ), ∀λ ∈ k, and i[D,D̄] = iD ◦ D̄L − D̄L ◦ iD = [iD , D̄L ].
(c) There is a unique k-linear operator d on the M -valued p-forms such that DL = iD ◦ d + d ◦ iD :
i ∇
P P i+j

dωp (D0 , . . . , Dp ) = i (−1) Di ωp (. . . D̂i . . .) + i<j (−1) ωp ([Di , Dj ] . . . D̂i . . . D̂j . . .) .
(d) If we have an A-bilinear map M × M 0 → M 00 , compatible with the connections in the sense that
00 0
D∇ (m · m0 ) = (D∇ m) · m0 + m · (D∇ m0 ), then d(ωp ∧ ωq ) = (dωp ) ∧ ωq + (−1)p ωp ∧ (dωq ).
(Hint: Proceed by induction on p + q, using DL = iD ◦ d + d ◦ iD ).
(e) Consider the curvature EndA (M )-valued 2-form K(D1 , D2 ) = D1∇ D2∇ − D2∇ D1∇ − [D1 , D2 ]∇ .
Then, on the p-forms we have:
i. [D1L , D2L ] = [D1 , D2 ]L + K(D1 , D2 )◦
ii. [DL , d] = (iD K)∧
iii. d2 = K∧
iv. dK = 0 (Bianchi’s Identity).
(Hint: The second and third identities,
P by∇induction on p. The last one, using Jacobi’s iden-
tity, the formula dK(D0 , D1 , D2 ) = D0 K(D1 , D2 ) − K([D0 , D1 ], D2 ) , where the summa-
tion extends over the three circular permutations, and the derivation law of endomorphisms
D0∇ K(D1 , D2 ) = D0∇ ◦ K(D1 , D2 ) − K(D1 , D2 ) ◦ D0∇ ).
7
In fact this is the sensible definition of vector field when there is not enough global functions: non-separated
smooth manifolds, analytic manifolds, schemes over a field, etc., so that any global vector field induces, by its
very definition, a vector field on any open subset.
538 EXERCISES
(f) If h ∈ HomA D, EndA (M ) , then ∇0 = ∇ + h also is a connection on M , and so we obtain any

connection on M . Moreover:
0
i. DL = DL + h(D),
ii. d0 = d ∧ h,
iii. K 0 = K + h ∧ h + dh.
3. If ∇ is a linear connection on a smooth manifold X check that ∇T is just the differential dT , where
the tensor field T is viewed as a section of the sheaf of (p, q)-tensors on X.
4. Prove the following formulae of the differential calculus with E-valued p-forms:
[iD1 , D2L ] = i[D1 ,D2 ] , [D1L , D2L ] = [D1 , D2 ]L + R(D1 , D2 ), [DL , d] = (iD R)∧.
5. Let R be the curvature 2-form of a linear connection on a smooth manifold X of dimension n. Prove
that the ordinary 2-form ω2 (D, D0 ) = tr(R(D, D0 )) is just the curvature of the induced linear connec-
tion on the sheaf ΩnX of volume forms.
6. A riemannian manifold (X, g) is isotropic at p ∈ X when the metric R2,2 induced (p. 288) by the
Riemann-Christoffel tensor on Λ2 Tp X is proportional to the metric Λ2 g,

R2,2 (D1 , D2 ; D3 , D4 ) = K (D1 · D3 )(D2 · D4 ) − (D1 · D4 )(D2 · D3 ) ,
and X is isotropic when so it is any point (but the factor K may depend on the point). In an isotropic
manifold, use Bianchi’s differential identity to prove that
0 = (D0 K)(Λ2 g)(D1 , D2 , D3 , D4 ) − (D1 K)(Λ2 g)(D0 , D2 , D3 , D4 ) + (D2 K)(Λ2 g)(D0 , D1 , D3 , D4 ).
Considering an orthonormal basis (and D3 = D1 , D4 = D2 ) conclude that any isotropic connected
manifold of dimension n ≥ 3 has constant curvature (Schur’s Theorem).
7. Let dX be a volume form on a smooth (1 + n)-dimensional manifold, and let Πn = C11 (dX ⊗ T 2 ) be
the vector-valued n-form corresponding to a 2-contravariant tensor T 2 .
(a) If g is a covariant metric, and T (D) := T 2 · D is the endomorphism associated to T 2 , prove that
for any vector field D we have Πn · D = iT (D) dX, and d(Πn · D) = div T (D) dX.
(b) If a symmetric linear connection ∇ on X preserves the volume form, ∇(dX) = 0, show that
dΠn = dX ⊗ (div∇ T 2 ). (Hint: At any point p ∈ X, consider a local base of vector fields
D0 , . . . , Dn such that (∇Di )p = 0).
8. Let ∇ be a linear connection on a smooth manifold X. If ρ ∈ C ∞ (X) and U is a vector field, show
that div∇ (ρU ⊗ U ) = (U ρ + ρdiv∇ U )U + ρU ∇ U . In the case of a perfect fluid of mass density ρ, mean
velocity U and pressure p, show that the condition div∇ (ρU ⊗ U + ph) = 0 is equivalent to the pair of
equations
U ρ + ρdiv U = 0 (Mass conservation law)

ρU ∇ U = −grad p (Law of motion)
9. Let R2 be the Ricci tensor of a riemannian 3-manifold (X, g). Prove that R2 = 0 implies that R = 0,
and that R2 = f g implies that X has constant curvature.
10. The Einstein tensor G2 of a riemannian (or Lorentz) metric g is defined to be G2 = R2 − 21 κg,
where κ = C12 R2 is the scalar curvature (contraction of the two indices of R2 with g). Check that
C11 (∇G2 ) = 0, and that G2 = 0 if and only if R2 = 0.
11. Weyl’s geometry: Let X be a smooth manifold endowed with a riemannian (or Lorentz) metric g
well-defined up to a constant factor at any point (i.e., up to a positive smooth function, so that we
don’t assume the existence of a length or time unit universally defined) and a linear connection ∇
preserving it: dg = ∇g = ω1 ⊗ g for some 1-form ω1 (obviously depending on the representant g, i.e.,
on the fixed units). Prove the following statements:
(a) d2 g = ω2 ⊗ g, where ω2 is a closed 2-form not depending on the fixed representant g.
(b) ω2 = 0 if and only if locally there exists a representant g such that ∇g = 0 (so that we have in
fact a riemannian geometry).
ALGEBRAIC GEOMETRY II 539
(c) The Einstein tensor R2 − 21 κg does not depend on the representant g, where R2 is the Ricci tensor
of ∇, and κ = C12 R2 is the contraction of the two indices of R2 with g.
12. In case that you know what an holomorphic vector bundle E → X over an analytic manifold X is,
prove that if we fix a smooth hermitian metric on E, then there is a unique linear connection ∇ on
the sheaf of smooth sections satisfying the following equivalent conditions:
(a) D(e · v) = (D∇ e) · v + e · (D∇ v), for any smooth tangent vector field D and any pair e, v of smooth
sections.
(b) D(e · v) = (D∇ e) · v + e · (D̄∇ v), for any smooth complex vector field D and any pair e, v of
smooth sections.
(c) d(e · v) = (de) · v + e · (dv), for any pair e, v of smooth sections.
∂ ∇
(d) ∂ z̄i e = 0, for any holomorphic section e and any holomorphic local coordinates (z1 , . . . , zn ).
(e) In a local basis (eP 1 , . . . , er ) of holomorphic sections, the structure 1-forms ωij are of (1,0) type
(recall that dei = j ωij ej ).
(Hint: Put hij = ei · ej . Then ∂hij + ∂h ¯ ij = dhij = P ωik hkj + P hik ω̄jk , so that if ω = (ωij ) is of
k k
(1,0) type, it is fully determined, ω = (∂h)h−1 ).
13. If π : E → X is a smooth vector bundle, show that π̄ : J 1 E → X also has a natural structure of smooth
vector bundle.
14. If π : Y = R × X → R is the natural projection, show that J 1 Y = R × (T X), where T X is the tangent
bundle of the smooth manifold X.
If π : Y = X × R → X is the natural projection, show that J 1 Y = R × (T ∗ X), where T ∗ X is the
cotangent bundle of the smooth manifold X.
15. The Takens elementary complex of a real vector space V is S • ⊗ Λ• V ∗ , with the grading induced
from Λ• V ∗ and the differential d defined by the product with the identity Id ∈ V ⊗ V ∗ ,
d d d
· · · −−−→ S • V ⊗ Λp V ∗ −−−→ S • V ⊗ Λp+1 V ∗ −−−→ · · ·
If we consider the contraction of indices c : S • V ⊗ Λp V ∗ → S • V ⊗ Λp−1 V ∗ , show that
(cd + dc)(x) = (r + n − p)x, n = dim V,
r p ∗
whenever x ∈ S V ⊗ Λ V , and conclude that the cohomology groups of the Takens elementary
complex are zero except for H n (S • V ⊗ Λ• V ∗ ) = Λn V ∗ .
16. Let S2n+1 ⊂ Cn+1 be the sphere of unitary vectors (with the usual hermitian metric). Show that the
natural map S2n+1 → Pn,C is a smooth principal bundle of group S1 = {z ∈ C : |z| = 1}, and that it
is not a trivial bundle.
17. Let L → X be a smooth line bundle, and let L∗ be the complement of the zero section. Show that
L∗ → X is a smooth principal bundle of group K ∗ = K − {0} (where K = R or C), and that so we
obtain any smooth principal bundle of group K ∗ .
15. Algebraic Geometry II

1. Decompose Q/Z as a direct sum of injective hulls E(Z/pZ).
2. If (O, m) is a finite local k-algebra, prove that the O-module O∗ = Homk (O, k) is the injective hull
E(O/m) of the residue field. (Hint: (O/m)∗ ' O/m).
3. Let x be a point of a noetherian scheme X, and let i : Spec OX,x → X be the natural morphism. If
M is a OX,x -module, show that HomOX (N , i∗ M
f) = HomO (Nx , M ) for any OX -module N .
X,x
If I is the injective hull of the residue field κ(x) of the local ring OX,x , prove that the quasi-coherent
sheaf i∗ Ie is an injective OX -module.
540 EXERCISES
4. Let I be an injective module over a noetherian ring A. Prove that Ix is an injective Ax -module at any
point x ∈ Spec A.
5. Let A = C[∂1 , . . . , ∂n ] be the ring of linear differential operators with constant coefficients, and let M
be an A-module. Given operators pi ∈ C[∂1 , . . . , ∂n ] and elements mi ∈ M , i = 1, . . . , r, we consider
the system of differential equations

p1 (∂1 , . . . , ∂n )u = m1

..................

pr (∂1 , . . . , ∂n )u = mr

(qij ) (pi )
Put I = (p1 , . . . , pr ) and fix a free resolution As −−−→ Ar −−−→ A → A/I → 0.
(a) If Ext1A (A/I, M ) = 0, P

prove that the system admits some solution u ∈ M if and only the
integrability conditions i qij mi = 0, i = 1, . . . , r, hold.
(b) Show that M = C[x1 , . . . , xn ] is an injective A-module. In fact, it is the injective hull of the
residue field of A at the origin.
Moreover, if A/I is a finite C-algebra, then HomA (A/I, M ) = (A/I)∗ and the dimension of the
space of solutions of the homogeneous system pi (∂1 , . . . , ∂n )u = 0 is the degree of A/I.
(c) When M is the ring of complex smooth functions on Cn and A/I is a finite C-algebra, prove
that ExtpA (A/I, M ) = 0, p > 0. (Hint: When I is a maximal ideal, take the Koszul resolution of
A/I = C. In general, proceed by induction on the degree of A/I).
(d) Show that the ring of formal series C[[x1 , . . . , xn ]] is an injective A-module. (Hint: If A is a
k-algebra, then A∗ is an injective A-module because it represents the exact functor N N ∗,
and HomC (C[∂1 , . . . , ∂n ], C) = C[[x1 , . . . , xn ]]).
6. Let P1 , . . . , Pr be a regular sequence of homogeneous polynomials in k[x0 , . . . , xd ], r < d, and let Y be

the closed subscheme of Pd defined by the ideal (P1 , . . . , Pr ). If we put di = deg Pi , prove the following
n

statements (we agree that m = 0 when n < m):
(a) H p (Y, OY (n)) = 0, when p 6= 0, d − r.

(b) dim H d−r (Y, OY (d1 + . . . + dr + n)) = −n−1

d , when n > −d − 1 − inf(d1 , . . . , dr ).
(c) dim H 0 (Y, OY (n)) = n+d

d , when n < inf(d1 , . . . , dr ).
7. Let C be a curve in Pr defined by a regular sequence P1 , . . . , Pr−1 of homogeneous polynomials. If we

put di = deg Pi , prove that
r−1
P d −1
χ(C, OC (n)) = d1 . . . dr−1 (n + 1) − d1 . . . dr−1 j2 ·
j=1
8. Prove that a ring A is regular if and only if so is the ring A[x].

9. Let O be a noetherian local ring. If the annihilator of f ∈ O is a prime ideal p, prove that
depth O ≤ dim (O/p).
10. If O is a local Cohen-Macaulay ring, prove that f ∈ O is an algebraic zero divisor if and only if (f )0
is a topological zero divisor (it has non empty interior).
11. Let I ⊆ m be an ideal of a noetherian local ring (O, m). Let us consider the cone C = Spec GI O and
the closed point x of the vertex V = Spec (O/I) ⊆ C.
(a) If I is m-primary, show that dimx C = dim O.

(b) In general, if y is the generic point of an irreducible component of the vertex V , show that the
local ring of C at y is just the local ring of the cone Spec (GIy Oy ) at the closed point y, and
conclude that dimy C = dim Oy .
(c) If O is Cohen-Macaulay and the cone C is Cohen-Macaulay at x, prove that dimx C = dim O.
12. Let A → B be a finite faithfully flat morphism. Prove that A is Cohen-Macaulay if and only if so is
B. Conclude that an affine algebraic variety X = Spec A over a field is Cohen-Macaulay if and only
if it admits a finite flat morphism X → An onto the affine space.
13. Let E be a k-vector space. We say that a coproduct linear map µ : E → E ⊗k E defines a structure
of (commutative) coalgebra on E, with counity a linear map u : E → k, if the diagrams
E
µ
/ E ⊗k E E
µ
/ E ⊗k E E
µ
/ E ⊗k E
µ µ⊗Id s u⊗Id
µ Id
# #
E ⊗k E
Id⊗µ
/ E ⊗k E ⊗k E E ⊗k E k ⊗k E
commute, where s(e ⊗ v) = v ⊗ e. Prove that (E, µ, u) is a coalgebra if and only if µ∗ : E ∗ ⊗k E ∗ → E ∗
and u∗ : k → E ∗ define a structure of k-algebra on E ∗ , the unity being u ∈ E ∗ .
With the notations of p. 418, show that any injective morphism of O ⊗k O-modules δ : I ,→ I ⊗k I
defines a structure of coalgebra on I. Moreover, I ⊗k I is an injective hull of the residue field of O ⊗k O,
and the counity r : I → k is the unique linear map such that r(k) = k and the diagram
I
µ
/ I ⊗k I
r⊗r
r
"
k
commutes. Conclude that the local residue Resx coincides, up to a factor 0 6= λ ∈ k, with the counity
of the coproduct δx : Hx1 (ΩC ) → H 1 (ΩC ) ⊗k H 1 (ΩC ).
14. Let S be a scheme. Show that the functor of points X • = HomS (−, X) of a S-scheme X is always a
sheaf, in the sense that U X • (U ) is a sheaf on any S-scheme.
Let F : S-schemes Sets be a contravariant functor. Given a S-scheme R → S, any R-scheme
is obviously a S-scheme, so that F induces a contravariant functor FR : R-schemes Sets. If F
is a sheaf, and it is locally representable, in the sense that FR is representable for some open cover
R = ⊕i Ui → S, prove that F is representable.
(Hint: Take X 0 → R representing FR with certain ξ 0 ∈ FR (X 0 ) = F (X 0 ). Since FR comes from F , it
is endowed with an isomorphism φ : π1∗ FR − ∼
→ π2∗ FR , where πi : ⊕i,j Ui ∩ Uj = R ×S R ⇒ R are the
canonical projections, with obvious transitivity conditions on R ×S R ×S R = ⊕i,j,k Ui ∩ Uj ∩ Uk . Hence
we have an isomorphism ϕ : π1∗ X 0 −∼
→ π2∗ X 0 , φ(π1∗ ξ 0 ) = ϕ∗ (π2∗ ξ 0 ), with the same transitivy conditions.
Prove that there is a S-scheme X and an isomorphism X 0 = X ×S R such that ϕ corresponds to the
canonical isomorphism π1∗ XR = π2∗ XR , and we have π1∗ (ξR ) = π2∗ (ξR ), where ξR ∈ F (XR ) corresponds
to ξ 0 ∈ F (X 0 ). Since F is a sheaf, show that ξR comes from a unique ξ ∈ F (X). Conclude that
ξ : X • → F is an isomorphism, since both are sheaves).
(a) Show that fibred products X ×S Y exist in the category of schemes.

(b) Let S be a scheme and let A be a quasi-coherent OS -algebra. Show the existence of a S-scheme
Spec A representing the functor F (T ) = HomOT -alg (A ⊗OS OT , OT ).
(c) If E is a locally free OS -module (of finite rank), show the existence of a S-scheme E = Spec (S • E ∗ )
representing the functor F (T ) = Γ(T, E ⊗OS OT ).
15. Let S be a scheme and let A• = A0 ⊕ A1 ⊕ . . . ⊕ An ⊕ . . . be a quasi-coherent graded OS -algebra. If

the natural morphism S • A1 → A• is surjective, prove the existence of a S-scheme Proj A• such that
for any affine open set U → S we have a natural isomorphism (Proj A• ) ×S U = Proj Γ(U, A• ).
16. Let U be an open subset of a ringed space (X, OX ). If I is an injective OX -module, prove that I|U
is an injective (O|U )-module.
17. Let X be a noetherian scheme. If I is an injective OX -module, prove that any stalk Ix is an injective
OX,x -module.
18. If X is a noetherian scheme, prove that any direct sum of injective OX -modules is injective.
542 EXERCISES
19. Let A be a noetherian ring, and put X = Spec A. If I is an injective A-module, prove that Ie is an
injective OX -module.
If M is a finitely generated A-module, conclude that Extn (M e ) = Extn (M, N )∼ .
f, N
OX A
20. Let X be a noetherian scheme. Prove that any quasi-coherent OX -module is the inductive limit of its
coherent submodules. Moreover, if I is an injective quasi-coherent sheaf, prove that I|U is an injective
quasi-coherent OU -module for any open set U ⊂ X. (Hint: Deligne’s formula).
21. Let Y be a closed subscheme of a noetherian scheme X. If M is a coherent OX -module, prove that
HYp (X, M) = lim
−→
ExtpOX (OX /pnY , M).
22. Show that K 0 (Z) = K(Z) = Z.

23. If O is a local ring, show that rk : K 0 (Spec O) → Z is an isomorphism.
If O is a regular local ring, show that rk : K(Spec O) → Z is an isomorphism.
Lr i
24. Let E be a locally free OX -module of rank r + 1. Prove that K(P(E)) = i=0 K(X)yE , where we
put yE = 1 − OP(E) (1). (Hint: yE = −O(1)xE and O(1) is invertible).
25. Let m be the maximal ideal of a rational point x of a smooth surface X, and let
π : X̄ = Proj (OX ⊕ m ⊕ m2 ⊕ . . .) −→ X
be the blow-up of X at x. Let E be the exceptional fibre, defined by the ideal pE = mOX̄ .
Show that pE = OX̄ (1), and pnE = OX̄ (n) for all n ≥ 1.
Show that OE · OX̄ (1) = OE (1) in the ring K(X̄), and (E ∩ E) = −1.
26. Let X → P2 be the blow-up of a projective plane at a rational point x, and let E be the exceptional
fibre. Prove that
(a) GK 1 (X) = ZH ⊕ ZE, where H is a line not incident to x.
(b) GK 2 (X) = Zp, where p is any rational point.
(c) H · H = p, E · E = −p and H · E = 0.
27. Let X be a smooth variety. If p is the ideal of a smooth closed subvariety Y → X, prove that
Y · Y = i (−1)i Λi (p/p2 ) ∈ GK • (X).
P
28. If E is a locally free OX -module of rank r and L is a line sheaf, prove that cr (E − L) = cr (E ⊗OX L∗ )
in GK • (X).
r
29. If c(E) = (1 + α1 ) . . . (1 + αr ) ∈ GK • (X), show that c(−E) = (1 − αi + αi2 − αi3 + . . .).
Q
i=1
30. Let X be a smooth variety. If N is the normal bundle of a smooth closed subvariety i : Y → X of
codimension d, prove that i∗ i∗ (y) = y · cd (N ) ∈ GK • (Y ), where y ∈ GK • (Y ).
31. Let j : Y → X be a smooth closed hypersurface of a smooth variety X. Prove the existence of an
exact sequence 0 → j ∗ L−Y → j ∗ ΩX → ΩY → 0, and conclude that in GK • (X) we have
j
(−1)m cm (X) · Y j+1−m .
P
j∗ ci (Y ) =
m=0
32. If 0 → E 0 → E → E 00 → 0 is an exact sequence of locally free OX -modules, then the morphisms

Λi E 0 ⊗ Λn−i E P → Λn E define a filtration of Λn E, the graded module being ⊕i (Λi E 0 ⊗ Λn−i E 00 ).
Hence
P i Λ E
n
= i (Λi E 0 )(Λn−i E 00 ) in K 0 (X). We put λi (E) = Λi E ∈ K 0 (X). The function λt (E) =
i
i λ (E)t , valued in the multiplicative group of formal series Pwith coefficients in K 0 (X) and constant
term 1, is additive, and we have a well-defined series λt (x) = i λi (x)ti for any x ∈ K 0 (X).
If L is a line sheaf, show that λt (−L) = (1 + Lt)−1 = i (−1)i Li ti and λi (−L) = (−1)i Li .
P
33. If a locally free OX -module E of rank r has a section without zeros, show that 0 = i (−1)i Λi E in
P
K 0 (X). (Hint: E − OX is the class of a locally free sheaf of rank r − 1, so that 0 = λr (E − 1)).
34. If E is a locally free OX -module of rank r, prove that 0 = i (−1)r−i λi (E ∗ )OP(E) (r − i) in K 0 (P(E)).
P
(Hint: The exact sequence 0 → Ē → π ∗ E ∗ → O(1) → 0 shows that λr E ∗ − OP(E) (1) = 0).
35. If X is a smooth variety, we define the γ-operations in K(X) as
n
γ n (x) = λn (x + n − 1) = (−1)n−i λi (x + n), n ≥ 0.
P
i=0
γ n (x)tn , prove that

P
If we put γt (x) = n
t (x),
(a) γt (x) = λ 1−t
λt (x) = γ 1+t
t (x).
(b) γ 0 (x) = 1,
γ 1 (x) = x,
γ n (x + y) =
P i
γ (x)γ j (y).
i+j=n
(c) If E is a locally free sheaf of rank r, then γ r+1 (E − r) = 0.

(d) If L is a line sheaf, then γ n (L) = L and γ n (1 − L) = (1 − L)n .
(e) Let E be a locally free sheaf of rank r and let ξ = OP(E) (−1) be the tautological line sheaf of the
projective bundle π : P(E) → X. If we put x = 1 − ξ ∈ K(P(E)), prove that
r
π ! γ i (E − r) (1 − ξ)r−i = 0
P
i=0
and conclude that cK

i (E) = γ i (E − r) ∈ K(X) and that ci (E) = [γ i (E − r)] ∈ GK i (X).
36. If E is aPlocally free sheaf on a smooth variety X, we put σ i (E) = S i E ∈ K(X). Check that
σt (E) = i σ i (E)ti is an additive function, valued in the multiplicative group of formalP series with
coefficients in K(X) and constant term 1, and we have a well-defined series σt (x) = i σ i (x)ti for
any x ∈ K(X).
If L is a line sheaf on X, show that σ i (−L) = 0, i ≥ 2, and σ i (1 − L) = 1 − L, i ≥ 1.
Finally, prove that σt (x)λ−t (x) = 1. (Hint: Reduce to the case of a line sheaf).
37. Let D be a tangent vector field to a smooth variety X of dimension n. If D has a finite number of
zeros x1 , . . . , xr , show that cn (X) = µ1 x1 + . . . + µr xr ∈ GK n (X), where µi = l OX,xi /(f1 , . . . , fn )
and D = f1 ∂1 + . . . + fn ∂n in a neighborhood of xi .
38. Let X, Y be smooth algebraic k-varieties. Prove that any closed embedding Y → X is regular. (Hint:
The ideal py of Y at a point y ∈ Y contains a sequence of parameters of length codimy Y , generating
py , according to the exact sequence py /mX,y py → mX,y /m2X,y → mY,y /m2Y,y → 0).
39. Let E → X be a vector bundle, and let U be the complement of the incidence relation in P(E)×X P(E ∗ ).
Show that both π1 : U → P(E) and π2 : U → P(E ∗ ) are affine bundles, and that the natural map
π1∗ OP(E) (−1) ⊗OU π2∗ OP(E ∗ ) (−1) → OU is an isomorphism, so that π1∗ OP(E) (−1) = π2∗ OP(E ∗ ) (1).
Conclude that in any cohomology theory A(P(E)) = A(X) ⊕ A(X)xE ⊕ . . . A(X)xr−1 E , where we put
xE = c1 (OP(E) (1)) and r = rk E. Moreover, xrE − c1 (E ∗ )xr−1 E + . . . + (−1)r
c r (E ∗
) = 0.
40. Let F (t) = i ai ti be a formal series with coefficients in A(Spec k). If E is a locally free sheaf, by
P
definition F+ (E) = (rk E)a0 + S(c1 (E), . . . , cn (E), . . .) for a well-defined power series S. Show that
the additive extension F+ : K(X) → A(X) is just F+ (x) = (rk x)a0 + S(c1 (x), . . . , cn (x), . . .).
Prove an analogous statement for the multiplicative extension F× : K(X) → A(X).
41. Given a cohomology theory (A, f∗ ) on the smooth quasi-projective k-varieties, let us consider a new
direct image f∗new so that (A, f∗new ) also is a cohomology theory. Prove the existence of an invertible for-
mal series F (t) = a0 +a1 t+. . . with coefficients in A(Spec k) such that f∗new (a) = f∗ (F× (−Tf )a). (Hint:
Show that cnew new
1 (OP1 (1)) = a0 c1 (OP1 (1)), where a0 ∈ A(Spec k) is invertible. Hence c1 (OPd (1)) =
c1 (OPd (1)) · F [c1 (OPd (1))] for some invertible series F (t), and conclude by Panin’s lemma).
42. In the case of a smooth projective variety of dimension g with trivial tangent bundle, show that for
1
any divisor D we have χ(LD ) = g! deg Dg .
544 EXERCISES
43. Let E be a locally free sheaf of rank r on the projective space Pd . If x = c1 (O(1)) ∈ GK • (Pd ), then
c(E) = 1 + m1 x + . . . + mr xr = (1 + α1 x) . . . (1 + αr x)
for some integer numbers m1 , . . . , mr and some complex numbers α1 , . . . , αr . Prove that
r
! r
X
(αi +n)t dt X n + d + αi
χ(Pd , E(n)) = Res e , t = 0 = .
i=1
(1 − e−t )d+1 i=1
d
(Hint: Use the variable change u = 1 − e−t to calculate the residue). Conclude that
(a) When r = 2, d = 3, the number m1 m2 is even. (Hint: d+α + d+α

d
1
d
2
is integer).
(b) When r = 2, d = 4, we have m2 (1 + m2 − 3m1 − 2m21 ) ≡ 0 (mod. 12).
(c) There is no vector bundle E on P3 such that E|P2 is the tangent bundle of P2 .
44. Assuming that GK • ⊗ Q is a cohomology theory, prove that ch : K(X) ⊗Z Q → GK • (X) ⊗Z Q is an
isomorphism for any smooth projective variety X.
45. If A is a cohomology theory, prove that any functorial group morphism φ : K(X) → A(X) is the
additive extension of a formal power series F (t) with coefficients in A(Spec k).
If A follows the law x + y, then φ is a ring morphism when F (x + y) = F (x)F (y) and F (0) = 1.
If A follows the law x + y − xy, then φ is a ring morphism when F (x + y − xy) = F (x)F (y) and
F (0) = 1. For example F (t) = (1 − t)k , k ∈ Z.
46. Define the Adams operations Ψk : K(X) → K(X) so that Ψk (L1 + . . . + Ln ) = Lk1 + . . . + Lkn when
L1 , . . . , Ln are line sheaves, k ∈ Z. Prove that Ψk is in fact a ring morphism.
State and prove a Riemann-Roch type theorem for Ψk .
47. Let D be a bounded complex of injective quasi-coherent sheaves over a noetherian scheme X. If we
put D(M) = Hom•OX (M, D), prove that the following conditions are equivalent:
(a) The natural morphism M• → DD(M• ) is a quasi-isomorphism for any bounded complex M• of
quasi-coherent sheaves, with coherent cohomology sheaves Hp (M• ).
(b) The natural morphism OX → DD(OX ) is a quasi-isomorphism.
(c) The natural morphism κ(x) → DD(κ(x)) is a quasi-isomorphism at any closed point x.
48. Let O be a Cohen-Macaulay local ring of dimension d with residue field k = O/m, and let ω be a
biduality module (a biduality complex with a unique nonzero term). Prove that
(a) ExtdO (k, ω) = k and ExtpO (k, ω) = 0, p 6= d.
(Hint: RHomO (k, O) = RHomO (k, RHomO (ω, ω)) = RHomO (k ⊗
=
ω, ω)).
(b) If ω1 , ω2 are two biduality O-modules, then they are (non-canonically) isomorphic. (Hint: If
t ∈ m is regular, by induction on d show that ωi /tn ωi is a biduality (O/tn O)-module, so that
HomO (ω1 /tn ω1 , ω2 /tn ω2 ) = O/tn O and we have compatible isomorphisms ω1 /tn ω1 ' ω2 /tn ω2
defining an isomorphism ω b1 ' ω
b2 . Conclude that HomO (ω1 , ω2 ) ' O, any generator defining an
isomorphism ω1 ' ω2 ).
(c) Hxd (Spec O, ω) is an injective hull of k, where x is the closed point of Spec O.
(d) Hxp (Spec O, M )∗ = Extd−p (M, ω) b for any finitely generated O-module M .
49. Prove that a sheaf morphism F → G is an isomorphism if and only if F(X) → G(X) is injective for
any object X, and for any section t ∈ G(X) there is a cover {Ui → X} where the sections t|Ui come
from some sections si ∈ F(Ui ).
50. Show that the fpqc topology on the category of S-schemes isSdefined by the families {Vα → X}
refined by some family {Spec Bij → Spec Ai ,→ X}, where X = i Spec Ai is an affine open cover and
{Spec Bij → Spec Ai } is a finite surjective family of flat morphisms. (Hint: These families {Vα → X}
define a pretopology because any cover {Vα → U } of an affine scheme U is refined, U being compact,
by a finite surjective family of flat morphisms {Spec Bi → U }.)
Conclude that, for any morphism T → S, the induced topology on Sch|T is just the fpqc topology of
Sch|T .
f1 ,f2
51. Show that a sequence F 0 ⇒ F → F 00 of sheaves on a situs is exact if and only if
(a) Two sections s, s̄ ∈ F(X) coincide in F 00 (X) if and only if there is a cover {Ui → X} sch that
for any index i we have that s|Ui and s̄|Ui coincide in the cokernel of F 0 (Ui ) ⇒ F(Ui ).
(b) Any section s00 ∈ F 00 (X) comes, on some cover of X, from sections of F.
(Hint: The sheafification Coker of the presheaf U Coker(F 0 (U ) ⇒ F(U )) is the cokernel of f1 , f2 ,
and it fulfills both conditions. Conversely, condition 2 shows that F → F 00 factors through a morphism
φ : Coker → F 00 , and if two sections s1 , s2 ∈ Coker(X) coincide in F 00 (X), then condition 1 states
that they coincide on a cover of X, so that s1 = s2 . Now condition 2 states that locally any section
s00 ∈ F 00 (X) is defined by sections of F, hence of Coker, so that φ is an isomorphism.)
52. If M is an A-module and A → B is a faithfully flat morphism, show that we have an exact sequence
d d
M → M ⊗A B − → M ⊗A B ⊗A B − → M ⊗A B ⊗A B ⊗A B . . .
Pp
where d(m ⊗ b1 ⊗ . . . ⊗ bp ) = k=0 m ⊗ b1 ⊗ . . . bk ⊗ 1 ⊗ . . . ⊗ bp . (Hint: After a faithfully flat base
change, we may assume that Spec B → Spec A admits a section).
Now, if we consider the functor of points of the additive group, Ga (Spec B) = B, conclude that
Ȟ p (B/A, Ga ) = 0, p ≥ 1.
In the case of a Galois extension of cyclic group G = (σ), obtain the additive Hilbert’s theorem 90:
Any element β ∈ K of null trace is β = σ(α) − α for some α ∈ K.
53. Let k be a field. For any k-algebra A, let ℵA be the supremum of the cardinals of the subalgebras of
A which are fields, so that ℵA ≤ ℵB whenever there is a morphism of k-algebras f : A → B. Let P(A)
be the first well-order of cardinal ℵA and let P(f ) : P(A) → P(B) be the unique isomorphism of P(A)
onto an initial ray of P(B), so that P(f ) = P(h) for any other morphism h : A → B. Show that P is
a separated presheaf of sets on (Aff|k )fp = (k-alg)opf p such that the “sheafification” LP is not a sheaf
of sets.
54. Let f −1 : Aff ,→ Sch be the inclusion of the category of affine schemes into the category of schemes.
If P is a presheaf of sets on Aff, show that f ∗ P is a presheaf of sets. Conclude that the “categories”
of sheaves on Schfpqc and Afffpqc are equivalent.
55. In an arbitrary category C, a topology is defined by a family of sieves for any object X, named
covering sieves of X, such that
(a) (Stability by base change) If R is a covering sieve of X, then R ×X • Y • is a covering sieve of Y
for any morphism Y → X.
(b) (Local character) Let C be a sieve of X. If R is a covering sieve of X and for any morphism
U → X in R we have that C ×X • U • covers U , then C covers X.
(c) X • is a covering sieve of X, for any object X.
(and it is sensible to assume that moreover any covering sieve contains a set-generated covering sieve).
Prove that any sieve C of X containing a covering sieve R also is a covering sieve. (Hint: For any
morphism U → X in R we have that C ×X • U • = U • covers U . Moreover, the intersection R ∩ R0
of two covering sieves of X also covers X, because for any morphism U → X in R we have that
(R ∩ R0 ) ×X • U • = R0 ×X • U • covers U .)
56. Let C be a category with finite fibred products. Show that any pretopology defines a topology in the
above sense. (Hint: To prove the second condition, consider a cover {Ui → X} generating a sieve
contained in R. For each index i there is a cover {Vij → Ui } generating a sieve contained in C ×X • Ui• ,
so that the morphisms Vij → X are in C. Hence C contains the sieve of X generated by the morphisms
Vij → X, defining a cover of X.)
Given a topology on C show that the sets {Ui → X} generating a covering sieve define a pretopology
on C. (Hint: If {Ui → X} generates a sieve R, then {Ui ×X Y → Y } generates R ×X • Y • , for any
morphism Y → X. Moreover, if {Ui → X} generates a covering sieve R, and for any index i we
have a family {Vij → Ui } generating a covering sieve of Ui , then the sieve R0 of X generated by the
546 EXERCISES
morphisms Vij → X fulfills that R0 ×X • Ui• covers Ui (it contains the morphisms Vij → Ui ). Hence
for any morphism Y → Ui → X in R we have that R0 ×X • Y • = (R0 ×X • Ui• ) ×Ui• Y • covers Y , so
that R0 covers X.)
57. Show that the opposite category of fields, with ring morphisms, has not fibred products, and that
the sieves containing a finite (resp. separable, algebraic, purely inseparable,...) extension define a
topology.
58. Show that the category of smooth manifolds has not fibred products, and that the sieves containing a
surjective regular projection define a topology, finer than the classical topology. (Hint: If P → X is a
regular projection, then P ×X Y exists for any smooth map Y → X and the natural map P ×X Y → Y
is a regular projection).
Index
action , dual, 53
, transitive, 117 , orthonormal, 57
of a Galois group, 186 change, 51, 127
of a group, 117 of a topology, 91
acyclic, 335 of closed sets, 215
affinity, 158 of neighborhoods, 91
algebra, 74, 101, 128 usual, 50
, σ-, 101 bicomplex, 332
, Azumaya, 185 binormal, 294
, Borel σ-, 101 blowup, 207, 428, 430
, Clifford, 501 bound
, Lie, 297 , lower, 15
, anticommutative, 173 , upper, 15
, boolean, 219, 503 boundary, 22
, central, 183 bracket
, exterior, 173 , Lie, 255
, finite, 74 , Poisson, 515
, purely inseparable, 153 branches, analytical, 207
, rational, 141 bundle
, separable, 144 , affine, 426
, symmetric, 173 , associated, 406
, tensor, 172 , cotangent, 290
, trivial, 143 , line, 347
, trivial Azumaya, 185 , natural, 404
algorithm, Euclid’s, 73 , normal, 430
angle, 167 , principal, 405
annihilator, 171 , projective, 425
anti-isomorphism, 15 , relative tangent, 437
anti-symmetrization, 64 , tangent, 290
antiequivalence of categories, 129 , tautological, 348, 360
argument, 38 , vector, 347, 425
asymptote, 165
automorphism, 42, 45, 47, 129 cardinal, 14
, Frobënius, 148, 149, 214 category, 129
, inner, 468 , opposite, 129
axiom of choice, 16 center, 118, 165, 183, 468
axis, 158, 170 chain, 16
character, Chern, 438
ball, 21 characteristic
barrier, 371 , Euler-Poincaré, 314, 330
base, 50, 119 of a ring, 83
, direct, 67 characterization of
547
548 INDEX
proper morphisms, 222 , Koszul, 412

subharmonic functions, 370 , Takens elementary, 539
class , biduality, 446
, Chern, 361, 431, 434 , bounded diagonals, 333
, Pontrjagin, 534 , dualizing, 354, 440
, Stieffel-Whitney, 361 , local dualizing, 445
, Todd, 438 component
, cohomology, 351 connected, 93
, obstruction, 348 irreducible, 135
classification of composite, 142
abelian groups, 174 composition, 14, 129
analytic morphisms, 267 concavity, strict, 466
compact surfaces, 246 concept
cyclic groups, 44 , geometric, 142
endomorphisms, 175 , local, 144
line sheaves, 348 conductor, 206
metrics, 162 cone, 378
metrics on modules, 180 , normal, 430
modules, 172 , tangent, 195, 198
pairs of metrics, 180 of a morphism, 333
principal coverings, 345 conformal metrics, 517
projectivities, 176 congruence, 14, 43
vector fields, 255 , Euler’s, 70
closure, 22 , Fermat’s, 70
, algebraic, 143 , Wilson’s, 70
, projective, 430 , Fermat’s, 84
, separable, 214 conjugate, 20, 41
coalgebra, 541 connection, 537
codifferential, 288 , Cartan, 285
codimension, 155, 438 , Levi-Civita, 286
cohomology, 304 , complete, 288
, Cech, 388, 458 , flat, 284
, De Rham, 263 , linear, 281, 396
, local, 327 , locally Euclidean, 284
, sheaf, 306 , symmetric, 283
of the fibre, 339 contraction
theory, 432 , interior, 66, 395
theory, graded, 438 , of indices, 63
with compact supports, 327 convexity, strict, 466
cokernel, 132, 470 coordinate, 267
colimit, 484 , local, 267
collineation, 159 coordinates, 50
commutator, 152 , affine, 158
compactification, 228 , canonical, 404
, Stone-Cěch, 229 , homogeneous, 156
, one-point, 95 , inertial, 165, 285
completion, 27, 193, 236 , isothermal, 517
complex, 304 , normal, 290
, Cech, 388, 420 coproduct, 130, 541
INDEX 549
counity, 541 , small, 484

cover, 23, 452 diameter, 165
covering, 211, 238 diffeomorphism, 99, 275
, Galois, 212, 239 , local, 99
, associated, 146, 212, 239 differential, 95, 139, 277, 278
, orientation, 355 , covariant, 281
, principal, 214, 244 , exterior, 279
, trivial, 211, 238 differentials, module of, 139
, universal, 239 differentiation under the integral sign, 33, 108
criterion dimension, 49, 231, 276
, Eisenstein’s, 72 , Grothendieck, 232
, diagonalization, 60 , Krull, 196
, ideal, 122 , combinatorial, 232
, reduction, 71 , global, 414
cross-ratio, 156 , projective, 414
curvature, 289, 294, 398, 496, 537 direction, 48
, geodesic, 294 directrix, 164, 170
, normal, 294 discriminant, 88, 162
, principal, 295 disk, 379
, scalar, 538 , Poincaré’s, 289, 522
, sectional, 288 , unit, 270
line, 295 distribution, 256, 385
curve, 206, 311 , integrable, 256
, elliptic, 529 , involutive, 257
, hyperelliptic, 529 divergence, 261, 398
, integral, 252 divisor, 177, 312, 384
, non singular, 311 , canonical, 316
, rational, 315 , effective, 205
, smooth, 319 , elementary, 172
cycle, 41, 428 , special, 527
, zero, 46
degree, 74, 357, 431 domain, 46
, separability, 153 , Dedekind, 204
, transcendence, 506 , principal ideal, 69
of a covering, 211, 238 , unique factorization, 86
of a divisor, 312, 384 dual
of a morphism, 313, 383 , dual, 155
delta, Dirac, 385
depth, 416 eccentricity, 496
derivation, 138 eigenspace, 57
derivative, 28, 82 eigenvalue, 57
, Lie, 255, 396, 537 eigenvector, 57
, covariant, 281 electromagnetic field, 165
, directional, 96 element
, partial, 96 , algebraic, 76
determinant, 66, 67, 498 , first, 15
, Vandermonde, 86 , identity, 45
diagonal, 234 , integral, 199
diagram, 484 , inverse, 42
, finite, 484 , invertible, 45
550 INDEX
, last, 15 , closed subspace, 329

, maximal, 15 , derivations, 139
, minimal, 15 , derived functors, 336
, neutral, 42 , differentials, 140
, opposite, 42 , local cohomology, 329
, proper, 46 , short, 49
, separable, 144 , split, 121
, unit, 45 , topologically, 376
embedding of complexes, 305
, closed, 311 of sheaves, 303
, local, 280 exact triangle, 304
, open, 311 excess, 81
, regular, 443 excision, 531
endomorphism, 57 extension, 74
, Weingarten, 295 , Galois, 146
, diagonalizable, 59 , additive, 435
, hermitian, 58 , algebraic, 76
, selfadjoint, 58 , essential, 411
, symmetric, 58 , finite, 74
energy, 404 , multiplicative, 435
entourage, 233 , normal, 153
, symmetric, 234 , radical, 151
envelope, Galois, 146 , trivial, 74
epimorphism, 485 by quadratic radicals, 77
equation fibre, 302
, Codazzi-Mainardi, 295 , geometric, 213
, Gauss, 294, 295 field, 45
, first structure, 399 , algebraically closed, 143
, second structure, 399 , decomposition, 143
equations , lagrangian, 403
, Cauchy-Riemann, 265 , perfect, 145
, Euler-Lagrange, 403 , residue, 120, 134
, Hamilton’s, 404 of fractions, 85
, Lagrange’s, 522 filter, 233
equicontinuous, 221, 511 , Cauchy, 235
equivalence , convergent, 235
, affine, 165 filtering, 124
, homotopical, 241, 264 filtration, 195
, linear, 205, 312, 384 , regular, 365
, numerical, 444 , stable, 196
, projective, 164 flag, 49
class, 13 flow, 253
of categories, 129 focus, 170
of projectivities, 176 form
relation, 13 , Poincaré-Cartan, 402
exact sequence, 49, 132 , analytic, 269
, Dolbeault, 385 , closed, 263
, Gysin, 351 , exact, 263
, Mayer-Vietoris, 329 , fundamental, 295
INDEX 551
, linear, 53 , harmonic, 261, 372

, quadratic, 164, 498 , hyperbolic, 468
, regular, 258 , increasing, 32
, structure, 400 , integrable, 31, 108
, volume, 67, 259, 261 , meromorphic, 268
of a permutation, 41 , primitive, 31
1-form, 278 , simple, 106
formula , smooth, 275
, Cartan’s, 279, 397 , strictly decreasing, 29
, Cauchy’s, 266 , strictly increasing, 29
, Cauchy-Goursat, 267 , subharmonic, 370, 372
, Deligne’s, 421 , universal additive, 177
, Gauss-Green, 515 of class C m , 30
, Girard, 83 functor
, Heron’s, 473 , (linearly) representable, 133
, Hurwitz’s, 319 , additive, 334
, Lagrange’s interpolation, 47 , conservative, 488
, Lefschetz’s, 359 , continuous, 453
, Taylor’s, 30, 98 , contravariant, 129
, Weingarten, 295 , covariant, 129
, Weingarten’s, 294 , derived, 335
, adjunction, 337, 439, 444 , exact, 132
, change of variable, 33 , left exact, 132
, change of variables, 110 , linear, 133
, class, 117 , representable, 132
, commutation, 80, 81 , right exact, 132
, fibre, 137 of points, 130
, graph, 131 fundamental theorem of
, integration by parts, 31 affine geometry, 492
, integration by substitution, 31 dimension theory, 233, 331
, points, 142, 212, 238 Euclidean geometry, 492
, projection, 343, 351, 423 projective geometry, 159
riemannian geometry, 286
, universal coefficients, 342
formulae gaussian integers, 73
, Cardano, 74 generators
, Frénet, 294 , topological, 454
, Newton, 83 genus, 314, 389
fraction, simple, 79 , arithmetic, 325
function, 134, 309 , geometric, 325
, Green’s, 372 geodesic, 283
, Hilbert, 196, 428 geometric
, Lebesgue measurable, 480 realization, 226
, Samuel, 196, 428 geometry
, additive, 177 , Euclidean, 167
, analytic, 36, 266, 267 , Weyl’s, 538
, bounded, 31 , elliptic, 170
, decreasing, 32 , hyperbolic, 170
, differentiable, 28 , non Euclidean, 170
, elementary symmetric, 74 germ, 249, 302
552 INDEX
gradient, 261 , irreducible, 191

greatest common divisor, 72 , irrelevant, 320, 505
group, 41 , maximal, 69
, K-, 177 , minimal, 184
, p-, 117 , primary, 191
, Brauer, 186 , prime, 69
, Galois, 146 , principal, 69
, Klein, 157 , real maximal, 250
, Lie, 296 generated, 69
, Picard, 205, 311 identity, 129
, abelian, 42 , Bézout’s, 72
, absolute Galois, 214 , Bianchi’s, 284, 399, 537
, cyclic, 44 , Bianchi’s differential, 398, 399
, fundamental, 214, 240 , Jacobi’s, 255, 256
, local uniparametric, 254 , Noether’s, 439
, opposite, 468 image, 43
, quotient, 43 , admirable direct, 422
, simple, 481 , direct, 304, 344
, solvable, 118 , higher direct, 339, 344
, topological, 511 , inverse, 337
, uniparametric, 254 impulse 3-form, 397
incident, 53
half-plane, Poincaré’s, 289, 522
index, 43, 161, 358
half-space, 380
, ramification, 267, 527
hamiltonian, 404
indicator, Euler, 70
vector field, 522
inequality
harmonic conjugate, 157
, Castelnuovo’s, 444
height, 415
, Cauchy’s, 266
hessian, 99
, Cauchy-Schwarz, 55, 470
Hodge star, 522
, triangle, 21, 55
homeomorphism, 24
infimum, 16
, local, 523
infinitesimal generator, 254
homography, 157
initial ray, 17
homomorphism, 381
homothety, 158 injective hull, 411
homotopy, 240, 241, 263, 458 integral, 106, 260
hybridize, 223 closure, 199
hyperbolic pair, 160 dependence relation, 199
hypercohomology, 336 interior, 22
hyperplane, 155 interval, 22
, polar, 164 invariant
, tangent, 164 , Noether, 402
at infinity, 158 , Zeuthen-Segre, 440
factors, 173
ideal, 45, 215 involution, 491
, Fitting, 173 irrational, quadratic, 77
, annihilator, 171 irreducible, 46
, contact, 400 isometry, 21, 159, 286, 470
, diagonal, 139 isomorphism, 119, 129
, homogeneous, 504 , group, 42
INDEX 553
, linear, 47 line, projective, 312

, metric, 21 linear representation, 182
, ring, 45 , irreducible, 182
of algebras, 74 linearly independent, 50
of functors, 129 localization, 216
of ordered sets, 15 of a module, 123
isotypic decomposition, 183 of a ring, 85
long
jacobian, 96 line, 514
jet, 400 ray, 514
Jouanolou’s trick, 431 loop, 240
kernel, 43, 132 manifold

, almost complex, 517
lagrangian, 401 , complex, 517
, regular, 403 , conformal, 270
lambertian, 288 , isotropic, 538
laplacian, 288, 372 , parallelizable, 524
lattice, 16 , riemannian, 286
law, quadratic reciprocity, 150 , smooth, 275
least common multiple, 72 , symplectic, 522
lemma , topological, 95
, Ahlfors, 271 with boundary, 259, 340
, Artin-Rees, 192 map, 14
, Caratheodory’s, 102 , affine, 158
, Dolbeault-Grothendieck, 387 , analytic, 267
, Euclid’s, 69, 72 , bijective, 14
, Gauss, 71, 291 , compact linear, 381
, Nakayama’s, 138 , continuous, 24
, Panin’s, 435 , differentiable, 95
, Poincaré’s, 264, 516, 521 , exponential, 290, 298, 410, 519
, Schur’s, 182 , injective, 14
, Schwarz’s, 271 , linear, 47
, Uryshon’s, 227 , measurable, 101
, Yoneda’s, 130 , proper, 222
, Zorn’s, 16 , semilinear, 55
, comparison, 461 , smooth, 275
, key, 97, 99 , surjective, 14
, normalization, 201 , tangent linear, 95, 276
, snake’s, 305 , transversal, 519
, stability, 208 , uniformly continuous, 25, 234
, tubular neighborhood, 410 of class C m , 96
length, 49 matrix
of a curve, 291 , Jordan, 175
of a module, 120 , jacobian, 96, 277
limit, 19, 22 of a linear map, 51
, inductive, 124, 131, 460 of a metric, 162
, projective, 124, 131, 460 maximum, 15
of a diagram, 484 , local, 29
of a function, 466 measure, 101
554 INDEX
, Lebesgue, 103 , base change, 127

, exterior, 102 , birrational, 202
, null, 34, 102 , dominant, 202
method , faithfully flat, 457
, Jacobi’s, 515 , finite, 199
, Lagrange-Charpit, 515 , flat, 420, 457
, Perron, 371 , localization, 85, 123
metric, 21, 159 , metric, 159
, Lorentz, 165 , projective, 422
, Poincaré, 270 , proper, 222
, alternate, 163 , quasicompact, 457
, dual, 493 , smooth, 443
, hermitian, 163 , unramified, 211
, hyperbolic, 270 in a category, 129
, locally Euclidean, 287 of A-modules, 119
, riemannian, 286 of G-sets, 117
, semiriemannian, 286 of S-schemes, 310
, space, 165, 285 of algebras, 74, 128
, time, 165, 285 of bicomplexes, 332
, trace, 145 of cohomology theories, 433
minimum, 15 of complexes, 304
, local, 29 of differential modules, 304
model, Klein’s projective, 289 of functors, 129
module, 119 of groups, 42
, Cohen-Macaulay, 416 of Lie groups, 296
, differential, 304 of locally ringed spaces, 310
, divisible, 123 of natural bundles, 405
, faithfully flat, 210 of ordered sets, 15
, finite presentation, 210 of presheaves, 301, 451
, finitely generated, 119 of ringed spaces, 310
, flat, 127 of rings, 45
, free, 119 of schemes, 310
of semirings, 215
, graded, 196
of sheaves, 301, 452
, homogeneous, 179, 182
morphism of
, injective, 122
metric spaces, 21
, noetherian, 190
pairs, 125
, primary, 172
motion, 167
, projective, 122
multilinear, 61
, semisimple, 182
multiplicity, 74
, simple, 182
, Serre’s intersection, 424
, torsion, 171
, intersection, 207, 324, 352
, torsion free, 171
of a local ring, 197
modulus of a
complex number, 20 neighborhood, 22
vector, 55 norm, 113, 187, 490
monomorphism, 485 normal, principal, 294
morphism normalizer, 118
, Frobënius, 145, 445 number
, affine, 526 , Lefschetz, 359
INDEX 555
, algebraic, 465 , generic, 130, 135, 310

, cardinal, 18 , geometric, 213
, complex, 20 , inflexion, 529
, integer, 13 , middle, 158
, natural, 13 , non singular, 202
, ordinal, 17 , ramification, 527
, rational, 14 , rational, 141
, real, 19 , regular, 371
, transcendent, 465 , simple, 202
, singular, 164
object , umbilical, 296
, final, 130 , unit, 158
, initial, 130 of a quadric, 164
operations, Adam’s, 544 of an algebra, 141
orbit, 117 polar, 379
order, 268 polarity, 55, 159
, dual, 16, 156 pole, 81, 268
, total, 16
polyhedron, 227
, well, 16
polynomial
of a group, 43
, Samuel, 196
of a natural bundle, 405
, Taylor, 30, 98
of an element, 44
, annihilator, 59
relation, 15
, characteristic, 57, 175
orientable, 259, 355
, cyclotomic, 77
orientation, 67, 259, 355
, generic, 477
, normal, 351
, minimal, 76
origin, 158
, separable, 144
orthogonal, 55, 160
, symmetric, 74
pair, 125 , unitary, 73
, minimal, 125 power of a cardinal, 15
paraboloid, 165 presheaf, 301, 451
paradox, Banach-Tarski, 479 , constant, 303
parallel, 48, 158, 281, 283 , separated, 452
transport, 283, 399 pretopologies, equivalent, 453
parameter, projective, 156 pretopology, 452
part, principal, 387 principle
partition, 31, 106 , Cavalieri’s, 110
of unity, 249, 327 , duality, 156
path, 240 , maximum, 262, 267, 370
permutation, 41 , splitting, 362, 433
perpendicular, 167, 470 of least action, 522
p-form, 63, 279, 395 problem
point, 130, 134, 158 , Cousin’s, 536
, adherent, 22 , Dirichlet, 370
, base, 180 , Post’s, 469
, conjugate, 164 product
, critical, 99, 112 , convolution, 479
, cuspidal, 507 , cross, 473
, fundamental, 180 , cup, 341
556 INDEX
, direct, 119, 130 retract, 121

, exterior, 65, 395 ring, 45
, fibred, 131 , Boole, 503
, scalar, 55 , Cohen-Macaulay, 416
, tensor, 62, 218 , Euclidean, 73
of cardinals, 15 , artinian, 501
of complex numbers, 20 , division, 183
of ideals, 69 , graded, 196, 504
of integers, 13 , integral, 46
of rationals, 14 , local, 138
projectable, 160, 257 , noetherian, 190
projection , normal, 200
, canonical, 13 , opposite, 182
, regular, 280 , quotient, 45
projectivity, 157 , reduced, 136
pseudometric, 113 , regular, 198, 414
, semilocal, 207
quadric, 164 , semisimple, 183
, absolute, 167 , simple, 183
, dual, 164, 493 root, 74
quasi-isomorphism, 305 of unity, 77
quaternions, 188 of unity, primitive, 77
quotient, 485 roots of unity, 150
rotation, 471
radical of rotational, 516
a metric, 160 rule
a ring, 136 , Barrow’s, 32
an ideal, 191 , Descartes, 84
radius of convergence, 36 , L’Hospital’s, 466
rank of a , Ruffini’s, 46, 474
matrix, 51 , chain’s, 28, 96, 277
metric, 160
module, 170 scalar, 45
ratio, simple, 158 scheme, 310
rectangle, 101 , affine, 310
reference , complete, 311
, Euclidean, 167 , integral, 311
, affine, 158 , noetherian, 311
, inertial, 165, 285 , projective, 320
, projective, 156 , reduced, 311
residue, 269, 319 , separated, 420
resolution, 305, 307 of finite type, 311
, Godement, 306 over a base scheme, 310
resolvent section, 16, 52, 121, 290
, Lagrange’s, 490 , critical, 401
, cubic, 477 semigroup, 215
resolvent, Lagrange’s, 490 semilinear, 159
resultant, 87 seminorm, 113
, Bézout, 88 semiring, 215
, Euler, 88 , dual, 215
INDEX 557
, lattice, 215 , covering, 453, 545

, noetherian, 225 sign of a permutation, 41
, normal, 223 signature, 492
sequence similarity, 167
, Cauchy, 18, 19, 25, 114, 235 ratio, 167
, convergent, 19, 22 simple, 49
, null, 18 singularity
, regular, 412 , essential, 268
series, 26 , removable, 268
, Laurent, 268 site, 453
set situs, 453
, G-, 117 solvable in radicals, 151
, ε-small, 235 space
, Cantor, 34 , T0 , 92
, absorbing, 376 , T1 , 92
, balanced, 376 , T3 or regular, 92
, basic closed, 216 , T4 or normal, 92
, basic open, 135, 216 , σ-compact, 94
, bounded, 23, 378 , k-, 511
, closed, 21 , Baire, 26
, countable, 14 , Banach, 535
, dense, 22 , Euclidean vector, 55
, measurable, 103 , Fréchet, 376
, open, 21 , Stone, 228, 232
, quotient, 13 , étalé, 302, 523
sheaf, 301, 452 , affine, 158, 201
, Godement, 306, 336 , compact, 23
, acyclic, 306 , complete metric, 25
, associated, 302, 460 , complete topological vector, 114
, canonical, 315 , complete uniform, 235
, coherent, 311, 526 , completely regular, 229
, constant, 303 , connected, 22
, dualizing, 315 , contractible, 241, 264
, flasque, 305 , cotangent, 198, 277
, flat, 342 , disconnected, 22
, line, 303 , dual, 53
, local cohomology, 350 , elliptic, 160
, locally constant, 303 , hyperbolic, 160, 289
, locally free, 303, 311 , irreducible, 135
, normal orientation, 350 , locally connected, 93, 94
, orientation, 355 , locally convex, 114
, quasi-coherent, 311, 456, 458, 526 , locally ringed, 309
, restriction of a, 302 , locally simply connected, 241
of continuous functions, 275 , measure, 101
of differentials, 316 , metric, 21
of finite type, 526 , metric vector, 159
of modules, 303 , metrizable, 22
sheafification, 302, 460 , noetherian, 225
sieve, 452 , non singular, 160
558 INDEX
, normal, 223 , topological, 350

, path-connected, 241 submodule, 119
, precompact, 237 , torsion, 171
, profinite, 219 generated, 119
, projective, 155, 320 subobject, 485
, quotient vector, 48 subring, 45
, regular, 223 subscheme
, ringed, 309 , closed, 311
, seminormable, 480 , open, 311
, separable, 477 subsequence, 18
, separated or Hausdorff, 92 subsheaf, 301
, simply connected, 241 subspace, 21, 22
, sober, 509 , locally closed, 281
, spectral, 218, 228 , relatively compact, 94
, tangent, 276 , vector, 47
, topological, 22 subvariety
, topological vector, 113 , affine, 158
, totally disconnected, 509 , linear, 48, 155
, totally isotropic, 160 sum
, uniform, 233 , Riemann, 31
, vector, 47 , connected, 246
measurable, 101 , direct, 52, 119, 130
spacetime , orthogonal, 160
, Galilean, 285 of cardinals, 15
, Minkowski, 165 of complex numbers, 20
specialization, 130, 196 of ideals, 69
spectral sequence of integers, 13
, Grothendieck’s, 366 of rationals, 14
, Leray’s, 367 of vector subspaces, 48
, bicomplex, 366 supplement, 52
, convergent, 365 , topological, 379
, hypercohomology, 366 support, 137, 249, 327
spectrum, 134, 216 supremum, 15
, maximal, 216, 250 surface
, projective, 320 , Riemann, 267
, real, 250 , minimal, 523
spinors, 502 symbol, 245
stalk, 302 , Christoffel, 282
subcategory, full, 454 , Legendre’s, 70
subdivision, barycentric, 226 symmetry, 160
subgroup, 42 , infinitesimal, 402
, Sylow, 118 system
, alternate, 44 , Pfaff, 257
, isotropy, 117 , characteristic, 257
, normal, 43 , coordinate, 275
, trivial, 42 , inductive, 124
generated, 42 , integrable Pfaff, 257
submanifold , local coordinates, 276
, smooth, 280 , multiplicative, 85
INDEX 559
, projective, 124 , Hodge’s index, 444

of generators, 50 , Hurewicz’s, 346
of parameters, 197 , Ischebeck’s, 416
, Künneth’s, 344
tensor, 61
, Koebe’s, 273
, Einstein, 538
, Kronecker’s, 75, 142
, Nijenhuis, 517
, Krull’s, 198
, Ricci, 286
, Lagrange’s, 43, 147
, Riemann-Christoffel, 287
, Lagrange’s mean value, 29
, alternate, 63
, Lebesgue’s, 34
, contravariant, 61
, Leray’s, 388
, covariant, 61
, Liouville’s, 267
, curvature, 284
, Maschke’s, 182
, matter, 397
, Max Noether’s, 528
, torsion, 283
, Mertens’, 37
field, 278
, Meusnier’s, 522
theorem
, Minding’s, 292
, Abel’s, 36
, Artin’s, 148, 213 , Noether’s, 402
, Ascoli’s, 221, 511 , Pappus, 491
, Bézout’s, 90, 324, 424, 507, 528 , Picard little, 375
, Baire, 26 , Rank, 67
, Banach-Steinhaus, 377 , Riemann mapping, 274
, Bolzano’s, 25 , Riemann-Roch, 315, 316, 392
, Borsuk-Ulam, 244, 353 , Rolle’s, 29
, Brouwer’s, 242, 340 , Rouché-Frobënius, 51
, Budan-Fourier, 476 , Schröder-Bernstein, 14
, Cantor’s, 15 , Schur’s, 538
, Cartan’s, 279 , Schwartz, 382
, Cauchy’s, 118 , Schwarz’s, 98
, Cauchy’s mean value, 29 , Serre’s, 323, 414
, D’Alembert’s, 39, 76, 151, 242, 266, 270, , Skolem-Noether, 183
358, 468 , Steiner’s, 323
, Darboux’s, 258 , Stokes, 260
, De Rham’s, 307, 335 , Stone’s representation, 219
, Desargues, 156 , Stone-Weierstrass, 230
, Euler, 296 , Sturm’s, 82
, Frobënius, 54, 258 , Sylow’s first, 118
, Frobenius, 188 , Sylow’s second, 118
, Galois, 147, 213, 214, 239, 243, 405 , Sylow’s third, 118
, Gauss egregium, 296 , Tietze’s, 231
, Gauss-Bonnet, 409 , Tychonoff’s, 220
, Gauss-Lucas, 476 , Van-Kampen, 245
, Grothendieck’s Riemann-Roch, 438 , Wedderburn’s, 185, 188
, Gysin, 425 , Weierstrass, 268, 272, 387
, Hahn-Banach, 378 , Witt’s, 161
, Hamilton-Cayley, 59 , Zermelo’s, 17
, Harnack’s, 369 , base change, 339
, Heine-Borel, 23 , bicomplex, 333
, Hirsch-Leray, 361 , bipolar, 380
560 INDEX
, chinese remainder, 70, 135 , final, 91

, closed graph, 377 , fp, 456
, continuous retract, 224 , fpqc, 457
, decomposition, 73 , initial, 91, 454
, degree, 76 , linear, 113
, divergence, 261 , order, 514
, division, 46 , pointwise convergence, 221
, dominated convergence, 108 , trivial, 22, 453
, duality, 354, 440 , uniform convergence, 221
, finiteness, 203, 340, 389 , weak, 221, 379
, first decomposition, 171, 179 coarser, 91
, formal functions, 450 finer, 91
, going down, 201 generated, 91, 454
, going up, 200 torsion, 294
, independence, 151 trace, 63, 145, 490, 527
, intermediate value, 25 trajectory, 285
, inverse mapping, 100, 277 , freely falling, 285
, isomorphism, 44, 46, 48 transformation
, local duality, 445 , infinitesimal contact, 400
, monotone class, 109 , natural, 129
, monotone convergence, 107 , quadratic, 207
, open mapping, 377 translation, 158
, periodicity, 426, 429 transpose, 54
, primitive element, 144 transposition, 41
, projection, 257 transversal, 349, 352
, pythagorean, 55 triangulation, 227
, recollement, 454 type of a 1-form, 384
, reduction, 149 ultrafilter, 382
, reflexivity, 53 uniform convergence, 35
, removable singularity, 268 uniformity
, representability, 133, 485 , compact convergence, 511
, residues, 269 , initial, 234
, second decomposition, 172, 179 , weak convergence, 511
, spectral representation, 218 of uniform convergence, 511
, symmetric functions, 74 universal property of the
, uniformization, 374 K-theory, 435
90, Hilbert’s, 459, 490, 545 base change, 127
of zeros, Hilbert’s, 201 completion, 28, 237
time differentials, 139
, interval, 285 direct product, 122, 130
, proper, 165 direct sum, 122, 130
topology, 22, 91, 545 exterior power Λp E, 66
, I-adic, 192 fibred product, 131
, Grothendieck, 453 graded K-theory, 439
, Zariski, 135, 216, 453 inductive limit, 125
, canonical, 453 localization, 85, 123
, compact convergence, 114, 221 projective limit, 125
, compact-open, 113 projective space, 322
, discrete, 22, 95, 453 quotient group, 44
INDEX 561
quotient ring, 46
quotient vector space, 48
sheafification, 302
symmetric algebra S • M , 173
tensor algebra T • M , 172
tensor product, 126, 128
tensor space Tpq E, 63
valuation, 202
, discrete, 202
, trivial, 202
ring, 202
ring, discrete, 202
value
, absolute, 19
of a function, 134
variation, 81
variety
, Riemann, 312
, affine algebraic, 201
, algebraic, 427
, polar, 164
, quasi-projective, 431
, smooth, 430
, tangent, 164
of solutions, 403
vector, 47
, Runge-Lenz, 497
, free, 158
, isotropic, 160
, light, 165
, spatial, 165, 285
extension, 158
field, 277
field with support, 282
field, complete, 254
vertex, 164, 170
volume, 67, 261
zeros of an ideal, 135

562 INDEX
Bibliography
[1] V. Arnold: Équations Différentielles Ordinaires, Ed. Mir, Moscou (1974)
[2] E. Artin: Geometric Algebra, Interscience Publ., New York (1957)
[3] E. Artin: Galois Theory, Univ. of Notre Dame Press, Notre Dame (1942)
[4] M. Artin: Grothendieck Topologies, Harvard Math. Dept. Lecture Notes (1962)
[5] M. Atiyah, I. G. Macdonald: Introduction to Commutative Algebra, Addison-Wesley, Read-

ing (1969)
[6] A. Borel, J. P. Serre: Le Théorème de Riemann-Roch, Bull. Soc. Math. France Tome 86
(1956), pp. 97–136
[7] N. Bourbaki: Éléments de Mathématique, Ed. Hermann, C.C.L.S., Masson, Springer-Verlag,

Paris (1940-1998)
[8] H. Cartan, S. Eilenberg: Homological Algebra, Princeton Univ. Press, Princeton (1956)
[9] M. Demazure, A. Grothendieck: Schémas en Groupes I (SGA 3), Lecture Notes in Math.
151, Springer-Verlag, Heidelberg (1970)
[10] R. Godement: Théorie des Faisceaux, Act. Sci. Ind. 1252, Ed. Hermann, Paris (1964)
[11] A. L. Gorodentsev: Algebra I, II, Textbook for Students of Math., Springer (2016)
[12] A. Grothendieck: Topological Vector Spaces, Ed. Gordon and Breach, New York (1973)
[13] A. Grothendieck: Sur quelques points d’algèbre homologique, Tohoku Math. Jour. Vol. 9,
No. 2-3 (1957), pp. 119–221
[14] A. Grothendieck: La Théorie des Classes de Chern, Bull. Soc. Math. France Tome 86
(1958), pp. 137–154
[15] A. Grothendieck: Fondements de la Géométrie Algébrique, Séminaire Bourbaki 149, 182,

190, 195, 212, 221, 232, 236, Institut H. Poincaré, Paris (1957-1962)
[16] A. Grothendieck: Éléments de Géométrie Algébrique I, II, III, IV, Publ. Math. I.H.E.S. 4,
8, 11, 17, 20, 24, 28, 32, Paris (1960-1967)
[17] A. Grothendieck: Local Cohomology, Lecture Notes in Math. 41, Springer-Verlag, Heidel-
berg (1967)
[18] A. Grothendieck: Revêtements Etales et Groupe Fondamental (SGA 1), Lecture Notes in
Math. 224, Springer-Verlag, Heidelberg (1971)
563
564 BIBLIOGRAPHY
[19] R.C. Gunning: An Introduction to Analysis, Princeton Univ. Press, Princeton (2018)
[20] R. Hartshorne: Algebraic Geometry, Graduate Texts in Math. 52, Springer-Verlag, Heidel-
berg (1977)
[21] R. Hartshorne: Residues and Duality, Lecture Notes in Math. 20, Springer-Verlag, Heidel-
berg (1966)
[22] N. J. Hicks: Notes on Differential Geometry, Van Nostrand, London (1965)
[23] B. Iversen: Generic Local Structure in Commutative Algebra, Lecture Notes in Math. 310,
Springer-Verlag, Heidelberg (1973)
[24] B. Iversen: Cohomology of Sheaves, Universitext, Springer-Verlag, Heidelberg (1986)
[25] H. Matsumura: Commutative Algebra, Benjamin, New York (1970)
[26] J. S. Milne: Course Notes

https://www.jmilne.org/math/CourseNotes/index.html
[27] J. S. Milne: The Legacy of Bernhard Riemann after One Hundred and Fifty Years (Lizhen
Ji, Frans Oort, Shing-Tung Yau Editors), ALM 35 (2015), pp. 487–565
http://www.jmilne.org/math/
[28] J. Milnor, J. Stasheff: Characteristic Classes, Annals Math. Studies 76, Princeton Univ.
Press, Princeton (1974)
[29] A. Navarro: On Grothendieck’s Riemann-Roch theorem, Exp. Math. 35 (2017), pp. 326–342
[30] K. Nomizu: Lie Groups and Differential Geometry, Math. Soc. Japan, Tokyo (1956)
[31] M. T. Sancho: Methods of Commutative Algebra for Topology, Publ. Dpto. Mat. 17, Univ.
Extremadura, Badajoz (1987)
[32] F. Sancho, P. Sancho: Álgebra Conmutativa y Geometrı́a Algebraica,

http://matematicas.unex.es/˜sancho/LibroGeometriaAlgebraica/geometria0.pdf
[33] J. P. Serre: Faisceaux Algébriques Cohérents, Annals of Math. 2nd Ser. Vol. 61, No. 2.
(1955), pp. 197–278
[34] J. P. Serre: Algèbre Locale. Multiplicités, Lecture Notes in Math. 11, Springer-Verlag,
Heidelberg (1965)
[35] R. Y. Sharp: Steps in Commutative Algebra, London Math. Soc. Student Texts 51, Cam-
bridge Univ. Press., Cambridge (2001)
[36] S. S. Shatz: Profinite Groups, Arithmetic, and Geometry, Annals Math. Studies 67, Prince-
ton Univ. Press, Princeton (1972)
[37] R. G. Swan: The Theory of Sheaves, Chicago Univ. Press, Chicago (1964)
Colophon
And what does it mean the term canonical or natural that we so often use in these notes
without a precise formal definition?
In our daily life we use canonical as authorized and recognized, specially in the catholic
ecclesiastical law; and natural is the Latin translation of the Greek term ϕυσιζ, a central
concept of the ancient Greek philosophy that nowadays we use in the word Physics to point that
this science studies Nature.
In Heidegger’s words (Einführing im die Metaphysic I §3):
Die Φυσιζ ist das Sein selbst, kraft dessen das Seiende erst beobachtbar wird und
bleibt... Als Gegenerscheinung tritt heraus, was die Griechen θσιζ, Setzung, Satzung
nennen oder νoµoζ, Gesetz, Regel im Sinne des Sittlichen... dem Gegensatz zu τ χνη
– was weder Kunst noch Technik besagt, sondern ein Wissen... 8
Briefly, where we most clearly perceive His aroma.
8
Φυσιζ means the force that prevails, brings-forth and remains regulated by itself... As an opposite manifes-
tation the Greeks introduced what they called θσιζ, what you put, or νoµoζ, custom, human convention... or
τ χνη, which means production from a knowledge...

Degree

Uploaded by

Document Information

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

Degree

Uploaded by

Copyright:

Available Formats

Notes for a Licenciatura

Based on Lectures of J. Sancho

July 12, 2021

Si quelque chose pourtant est saccagé et mutilé, et desamorcé de sa force originelle,

Vous soulignez, à juste titre, la difficulté psychique de la création solitaire... C’est

Juan A. Navarro González

6 Projective Geometry 155

Third Year 189

7 Commutative Algebra 189

7.7 Birrational Finite Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

9 Analysis III 249

10 Differential Geometry I 275

Fourth Year 300

11 Algebraic Geometry I 301

11.4.1 Calculation of the Canonical Sheaf . . . . . . . . . . . . . . . . . . . . . . 316

12 Algebraic Topology I 327

14 Differential Geometry II 395

Fifth Year 411

15 Algebraic Geometry II 411

15.4.1 Graded K-Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427

Index of Terms 547

• Algebra II: Actions of a group. Modules. Categories. Spectrum of a ring. Module of

• Commutative Algebra: Noetherian rings. Primary decomposition. Completion. Dimen-

• Topology: Lattice semirings. Separation properties. Noetherian spaces. Compactifications.

• Algebraic Topology I: The cohomology ring. Homological algebra. Local cohomology.

• Analysis IV: Uniformization theorem. Fréchet spaces. Compact Riemann surfaces.

• Differential Geometry II: Valued differential calculus. Calculus of variations. Natural

6. The module of differentials, giving an algebraic treatment of the differential calculus.

The central result is Galois theorem, stated as an antiequivalence of the category13 of k-

1. Primary decomposition of ideals and completion.

2. Krull dimension theory and regular rings.

0+0=0 0+g =g+0=g g+g =g

1. The functor Spec clarifies the theory of compactifications.

Differential Geometry I: We introduce smooth manifolds as ringed spaces X locally isomor-

H p (X, F) = 0, p > dim X,

dim H 0 (X, LD ) − dim H 1 (X, LD ) = 1 − g + deg D.

H 1 (X, M)∗ = HomOX (M, DX )

for some quasi-coherent sheaf DX , and the problem is to determine DX .

multiplicative group. Remark that (1 − x)(1 − y) = 1 − (x + y − xy).

2. Given a projective morphism f : X → S, the representability theorem gives the existence of

RHomOS (Rf∗ (M), OS ) = RHomOX (M, DX/S )

OX = DX/X = DX/X×X ⊗ ∆∗ DX×X/X = Λd N ⊗ ∆∗ (π1∗ DX/S ) = Λd N ⊗ DX/S

However, along the five years, an increasing maturity level is assumed.

Notations: Exceptionally we use some non-standard notations. The covariant derivative of a

X ∩ Z := f −1 (Z) = {x ∈ X : f (x) ∈ Z},

so that X ∩ Z is a subset of X but not a subset of Z.

Acknowledgements: My debt and gratitude to my teachers at the University of Salamanca

1.1 Integer and Rational Numbers

Theorem: In X/ ≡ , only equivalent elements are identified; [x] = [y] ⇔ x ≡ y .

Proof: If [x] = [y], then y ∈ [y] = [x], and x ≡ y.

Corollary: Each element of X is in a unique equivalence class.

Proof: If x ∈ [y], then y ≡ x; hence [y] = [x].

(m, n) ≡ (m0 , n0 ) when m + n0 = m0 + n,

a + b = (m + r) − (n + s) , a · b = (mr + ns) − (nr + ms)

and both are well-defined. In fact, if a = m0 − n0 , then m + n0 = m0 + n, and

and we also have a + b = m0 + r − (n0 + s), ab = m0 r + n0 s − (n0 r + m0 s).

(a, s) ≡ (b, t) when at = bs,

(at + bs)s0 t = a0 tst + bss0 t = (a0 t + bs0 )st,

1.2 Cardinal and Ordinal Numbers

If A ⊆ X, we put f (A) = {f (a) : a ∈ A} ⊆ Y .

Schröder-Bernstein Theorem: If |X| ≤ |Y | and |Y | ≤ |X|, then |X| = |Y |.

Proof: Let f : X → Y and g : Y → X be injective maps, so that h = gf : X0 := X → X2 := h(X)

Cantor’s Theorem: |X| < 2|X| .

First proof : Obviously |X| ≤ |P(X)|.