Graduate Texts in Mathematics 213

Editorial Board
S. Axler F.W. Gehring K.A. Ribet

Springer
New York
Berlin
Heidelberg
Barcelona
Hong Kong
London
Milan
Paris
Singapore
Tokyo
Graduate Texts in Mathematics
TAKEtrrIlZARiNG. Introduction to 34 SPITZER. Principles of Random Walle
Axiomatic Set Theory. 2nd ed. 2nded.
2 OXTOBY. Measure and Category. 2nd ed. 35 ALExANDERlWERMER. Several Complex
3 ScHAEFER. Topological Vector Spaces. Variables and Banach Algebras. 3rd ed.
2nded. 36 KEu.sy/NAMIOKA et al. Linear
4 Hn.TONISTAMMBACH. A Course in Topological Spaces.
Homological Algebra. 2nd ed. 37 MONK. Mathematical Logic.
5 MAc LANE. Categories for the Working 38 GRAUERTIFRrrzsam. Several Complex
Mathematician. 2nd ed. Variables.
6 HUGlIBSIPtPER. Projective Planes. 39 ARVESON. An Invitation to c*-Algebras.
7 SERRE. A Course in Arithmetic. 40 KEMENY/SNEUiKNAPP. Denumerable
8 TAKEtrrIlZARiNG. Axiomatic Set Theory. Markov Chains. 2nd ed.
9 HUMPHREYS. Introduction to Lie Algebras 41 APosroL. Modular Functions and
and Representation Theory. Dirichlet Series in Number Theory.
IO COHEN. A Course in Simple Homotopy 2nded.
Theory. 42 SERRE. Linear Representations of Finite
11 CONWAY. Functions of One Complex Groups.
Variable I. 2nd ed. 43 GIU.MANlJERISON. Rings of Continuous
12 BEALS. Advanced Mathematical Analysis. Functions.
13 ANDERSONlFtJu.EJt. Rings and Categories 44 KENDIG. Elementary Algebraic Geometry.
of Modules. 2nd ed. 45 LoEVE. Probability Theory I. 4th ed.
14 GoLUBITSKy/GUIU.EMIN. Stable Mappings 46 LotiVE. Probability Theory D. 4th ed.
and Their Singularities. 47 MOISE. Geometric Topology in
IS BERBERIAN. Lectures in Functional Dimensions 2 and 3.
Analysis and Operator Theory. 48 SAOfs/WU. General Relativity for
16 WINTER. The Structure of Fields. Mathematicians.
17 ROSENBLAtT. Random Processes. 2nd ed. 49 GRUENBERGlWEIR. Linear Geometry.
18 HALMos. Measure Theory. 2nd ed.
19 HALMos. A Hilbert Space Problem Book. 50 EDWARDS. Fermat's Last Theorem.
2nd ed. 51 KuNGENBERG. A Course in Differential
20 HUSEMOU.ER. Fibre Bundles. 3rd ed. Geometry.
21 HUMPHREYS. Linear Algebraic Groups. 52 HARTSHORNE. Algebraic Geometry.
22 BARNESIMAcK. An Algebraic Introduction 53 MANJN. A Course in Mathematical Logic.
to Mathematical Logic. 54 GRAVERIWATKINS. Combinatorics with
23 GREUB. Linear Algebra. 4th ed. Emphasis on the Theory of Graphs.
24 HOLMES. Geometric Functional Analysis 55 BRoWNIPEARCY. Introduction to Operator
and Its Applications. Theory I: Elements of Functional Analysis.
25 HEwrrr/SmOMBERG. Real and Abstract 56 MASSEY. Algebraic Topology: An
Analysis. Introduction.
26 MANES. Algebraic Theories. 57 CRoWElllFox. Introduction to Knot
27 KElLEy. General Topology. Theory.
28 ZARIsKIlSAMUEL. Commutative Algebra. 58 KOBUTZ. p-adic Numbers. p-adic
Vol.l. Analysis. and Zeta-Functions. 2nd ed.
29 ZARIsKIlSAMUEL. Commutative Algebra. 59 LANG. Cyclotomic Fields.
V01.ll. 60 ARNOLD. Mathematical Methods in
30 JACOBSON. Lectures in Abstract Algebra I. Classical Mechanics. 2nd ed.
Basic Concepts. 61 WHITEHEAD. Elements of Homotopy
31 JACOBSON. Lectures in Abstract Algebra D. Theory.
Linear Algebra. 62 KARGAPOwvIMERuJAKOV. Fundamentals
32 JACOBSON. Lectures in Abstract Algebra of the Theory of Groups.
m. Theory of Fields and Galois Theory. 63 BOu.DBAS. Graph Theory.
33 HIRsoi. Differential Topology.
(continued after index)
Klaus Fritzsche
Hans Grauert

From Holomorphic
Functions to
Complex Manifolds

With 27 Illustrations

Springer
Klaus Fritzsche Hans Grauert
Bergische Universitiit Wuppertal Mathematisches Institut
GauBstra6e 20 Georg-August-Universitiit G6ttingen
0-42119 Wuppertal Bunsenstra6e 3-5
Gennany 0-37073 G6ttingen
[email protected] Gennany

Editorial Board
S. Axler F.W. Gehring K.A. Ribet
Mathematics Department Mathematics Department Mathematics Department
San Francisco State East Hall University of California,
University University of Michigan Berkeley
San Francisco, CA 94132 Ann Arbor, MI 48109 Berkeley, CA 94720-3840
USA USA USA
[email protected] [email protected]. [email protected]
umich.edu

Mathematics Subject Classification (2000): 32-01, 32Axx, 32005, 32Bxx, 32Qxx, 32E35

Library of Congress Cataloging-in-Publication Data

Fritzsche. Klaus.
From holomorphic functions to complex manifolds 1 Klaus Fritzsche, Hans Grauert.
p. cm. - (Graduate texts in mathematics; 213)
Includes bibliographical references and indexes.
ISBN 978-1-4419-2983-9 ISBN 978-1-4684-9273-6 (eBook)
DOl 10.1007/978-1-4684-9273-6
I. Complex manifolds. 2. Holomorphic functions. I. Grauert. Hans. 1930- n. Title.
III. Series.
QA331.7 .F75 2002
515'.98---4c21 2001057673
© 2002 Springer-Verlag New York. Inc.
All rights reserved. This work may not be translated or copied in whole or in part without the written
permission of the publisher (Springer-Verlag New York. Inc .• 175 Fifth Avenue. New York. NY
10010, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in
connection with any form of information storage and retrieval, electronic adaptation, computer soft-
ware, or by similar or dissimilar methodology now known or hereafter developed is forbidden.
The use in this publication of trade names, trademarks, service marks, and similar terms, even if they
are not identified as such. is not to be taken as an expression of opinion as to whether or not they are
subject to proprietary rights.
9 8 7 654 3 2 1

SPIN 10857970

Springer-Verlag New York Berlin Heidelberg

A member of BertelsmannSpringer Science+Business Media GmbH
This reprint has been authorized by Springer-Verlag (Berlin/Heidelberg/New York) for sale in the
Mainland China only and not for export therefrom.
Preface

The aim of this book is to give an understandable introduction to the the-

ory of complex manifolds. With very few exceptions we give complete proofs.
Many examples and figures along with quite a few exercises are included.
Our intent is to familiarize the reader with the most important branches and
methods in complex analysis of several variables and to do this as simply as
possible. Therefore, the abstract concepts involved with sheaves, coherence,
and higher-dimensional cohomology are avoided. Only elementary methods
such as power series, holomorphic vector bundles, and one-dimensional co-
cycles are used. Nevertheless, deep results can be proved, for example the
Remmert-Stein theorem for analytic sets, finiteness theorems for spaces of
cross sections in holomorphic vector bundles, and the solution of the Levi
problem.
The first chapter deals with holomorphic functions defined in open sub-
sets of the space en. Many of the well-known properties of holomorphic
functions of one variable, such as the Cauchy integral formula or the maxi-
mum principle, can be applied directly to obtain corresponding properties of
holomorphic functions of several variables. Furthermore, certain properties of
differentiable functions of several variables, such as the implicit and inverse
function theorems, extend easily to holomorphic functions.
In Chapter II the following phenomenon is considered: For n 2: 2, there
are pairs of open subsets H c Peen such that every function holomorphic
in H extends to a holomorphic function in P. Special emphasis is put on
domains G c en for which there is no such extension to a bigger domain.
They are called domains of holomorphy and have a number of interesting
convexity properties. These are described using plurisubharmonic functions.
If G is not a domain of holomorphy, one asks for a maximal set E to which all
holomorphic functions in G extend. Such an "envelope of holomorphy" exists
in the category of Riemann domains, i.e., unbranched domains over en.
The common zero locus of a system of holomorphic functions is called
an analytic set. In Chapter III we use Weierstrass's division theorem for
power series to investigate the local and global structure of analytic sets.
Two of the main results are the decomposition of analytic sets into irreducible
components and the extension theorem of Remmert and Stein. This is the
only place in the book where singularities play an essential role.
Chapter IV establishes the theory of complex manifolds and holomorphic
fiber bundles. Numerous examples are given, in particular branched and un-
branched coverings of en, quotient manifolds such as tori and Hopf manifolds,
projective spaces and Grassmannians, algebraic manifolds, modifications, and
toric varieties. We do not present the abstract theory of complex spaces, but
do provide an elementary introduction to complex algebraic geometry. For
example, we prove the theorem of Chow and we cover the theory of divi-
vi Preface

sors and hyperplane sections as well as the process of blowing up points and
submanifolds.
The present book grew out of the old book of the authors with the ti-
tle Seveml Complex Variables, Graduate Texts in Mathematics 38, Springer
Heidelberg, 1976. Some of the results in Chapters I, II, III, and V of the old
book can be found in the first four chapters of the new one. However, these
chapters have been substantially rewritten. Sections on pseudoconvexity and
on the structure of analytic sets; the entire theory of bundles, divisors, and
meromorphic functions; and a number of examples of complex manifolds have
been added.
Our exposition of Stein theory in Chapter V is completely new. Using only
power series, some geometry, and the solution of Cousin problems, we prove
finiteness and vanishing theorems for certain one-dimensional cohomology
a
groups. Neither sheaf theory nor methods are required. As an application
Levi's problem is solved. In particular, we show that every pseudoconvex
domain in en is a domain of holomorphy.
Through Chapter V we develop everything in full detail. In the last two
chapters we deviate a bit from this principle. Toward the end, a number of
the results are only sketched. We do carefully define differential forms, higher-
dimensional Dolbeault and de Rham cohomology, and Kahler metrics. Using
results of the previous sections we show that every compact complex mani-
fold with a positive line bundle has a natural projective algebraic structure. A
consequence is the algebraicity of Hodge manifolds, from which the classical
period relations are derived. We give a short introduction to elliptic opera-
tors, Serre duality, and Hodge and Kodaira decomposition of the Dolbeault
cohomology. In such a way we present much of the material from complex
differential geometry. This is thought as a preparation for studying the work
of Kobayashi and the papers of Ohsawa on pseudoconvex manifolds.
In the last chapter real methods and recent developments in complex an-
alysis that use the techniques of real analysis are considered. Kahler theory is
carried over to strongly pseudoconvex subdomains of complex manifolds. We
give an introduction to Sobolev space theory, report on results obtained. by
J.J. Kohn, Diederich, Fornress, Catlin, and Fefferman (a-Neumann, subeUip-
tic estimates), and sketch an application of harmonic forms to pseudoconvex
domains containing nontrivial compact analytic subsets. The Kobayashi met-
ric and the Bergman metric are introduced, and theorems on the boundary
behavior of biholomorphic maps are added.
Prerequisites for reading this book are only a basic knowledge of calculus,
analytic geometry, and the theory of functions of one complex variable, as
well as a few elements from algebra and general topology. Some knowledge
about Riemann surfaces would be useful, but is not really necessary. The
book is written as an introduction and should be of interest to the specialist
and the nonspecialist alike.
 Preface vii

We are indebted to many colleagues for valuable suggestions, in particular

to K. Diederich, who gave us his view of the state of the art in a-Neumann
theory. Special thanks go to A. Huckleberry, who read the manuscript with
great care and corrected many inaccuracies. He made numerous helpful sug-
gestions concerning the mathematical content as well as our use of the English
language. Finally, we are very grateful to the staff of Springer-Verlag for their
help during the preparation of our manuscript.
Wuppertal, Gottingen, Germany Klaus Fritzsche
Summer 2001 Hans Grauert
Contents

Preface v

I Holomorphic FUnctions 1
1. Complex Geometry .......... . 1
Real and Complex Structures. . . . . 1
Hermitian Forms and Inner Products 3
Balls and Polydisks 5
Connectedness . . . 6
Reinhardt Domains 7
2. Power Series .. 9
Polynomials 9
Convergence 9
Power Series 11
3. Complex Differentiable Functions 14
The Complex Gradient . . . . 14
Weakly Holomorphic Functions, 15
Holomorphic Functions 16
4. The Cauchy Integral . . . . . . . 17
The Integral Formula . . . . . 17
Holomorphy of the Derivatives 19
The Identity Theorem . . . . . 22
5. The Hartogs Figure . . . . . . . . 23
Expansion in Reinhardt Domains 23
Hartogs Figures . . . . . . . . 25
6. The Cauchy-Riemann Equations 26
Real Differentiable Functions . 26
Wirtinger's Calculus . . . . . . 28
The Cauchy-Riemann Equations. 29
7. Holomorphic Maps 30
The Jacobian. . 30
Chain Rules .. 32
Tangent Vectors 32
The Inverse Mapping 33
8. Analytic Sets . . . . . . . 36
Analytic Subsets . . . 36
Bounded Holomorphic Functions . 38
Regular Points . . . . . . . . . . 39
Injective Holomorphic Mappings . 41
x Contents

II Domains of Holomorphy 43
l. The Continuity Theorem 43
General Hartogs Figures 43
Removable Singularities . 45
The Continuity Principle 47
Hartogs Convexity . . . . 48
Domains of Holomorphy 49
2. Plurisubharmonic Functions 52
Subharmonic Functions . 52
The Maximum Principle 55
Differentiable Subharmonic Functions 55
Plurisubharmonic Functions 56
The Levi Form . . . . 57
Exhaustion Functions 58
3. Pseudoconvexity . . . . . 60
Pseudoconvexity . . . 60
The Boundary Distance 60
Properties of Pseudoconvex Domains 63
4. Levi Convex Boundaries 64
Boundary Functions 64
The Levi Condition 66
Affine Convexity . . 66
A Theorem of Levi. 69
5. Holomorphic Convexity 73
Affine Convexity . . 73
Holomorphic Convexity 75
The Cartan-Thullen Theorem 76
6. Singular Functions. . . . . . . . . 78
Normal Exhaustions . . . . . . 78
Unbounded Holomorphic Functions 79
Sequences . . . . . . . . . 80
7. Examples and Applications . . . . 82
Domains of Holomorphy 82
Complete Reinhardt Domains 83
Analytic Polyhedra .. 85
8. Riemann Domains over en . . . 87
Riemann Domains . . . . . . 87
Union of Riemann Domains . 91
9. The Envelope of Holomorphy. . 96
Holomorphy on Riemann Domains . 96
Envelopes of Holomorphy 97
Pseudoconvexity . 99
Boundary Points 100
Analytic Disks .. 102
 Contents Xl

III Analytic Sets 105

l. The Algebra of Power Series 105
The Banach Algebra B t . 105
Expansion with Respect to Zl 106
Convergent Series in Banach Algebras . 107
Convergent Power Series 108
Distinguished Directions .. . 109
2. The Preparation Theorem . . . . 110
Division with Remainder in B t 110
The Weierstrass Condition . . 113
Weierstrass Polynomials. . . . 114
Weierstrass Preparation Theorem 115
3. Prime Factorization . . . 116
Unique Factorization 116
Gauss's Lemma .. 117
Factorization in H n . 119
Hensel's Lemma . . . 119
The Noetherian Property 120
4. Branched Coverings .. 123
Germs. . . . . . . . 123
Pseudopolynomials 124
Euclidean Domains 125
The Algebraic Derivative 125
Symmetric Polynomials 126
The Discriminant .. 126
Hypersurfaces . . . . 127
The Unbranched Part 130
Decompositions .. . 130
Projections . . . . . . 132
5. Irreducible Components. 135
Embedded-Analytic Sets 135
Images of Embedded-Analytic Sets 137
Local Decomposition 138
Analyticity . . . . . . . 140
The Zariski Topology . 141
Global Decompositions 141
6. Regular and Singular Points 143
Compact Analytic Sets . 143
Embedding of Analytic Sets 144
Regular Points of an Analytic Set 145
The Singular Locus . . . 147
Extending Analytic Sets. 147
The Local Dimension . . 150
xii Contents

IV Complex Manifolds 153

1. The Complex Structure . 153
Complex Coordinates 153
Holomorphic FUnctions 156
Riemann Surfaces . . . 157
Holomorphic Mappings 158
Cartesian Products " 159
Analytic Subsets . . . . 160
Differentiable Functions. 162
Tangent Vectors . . . . . 164
The Complex Structure on the Space of Derivations 166
The Induced Mapping. . . . 167
Immersions and Submersions 168
Gluing . . . . . . . . . . . . 170
2. Complex Fiber Bundles . . . . . 171
Lie Groups and 'Transformation Groups 171
Fiber Bundles . . . . . . 173
Equivalence. . . . . . . . 174
Complex Vector Bundles 175
Standard Constructions . 177
Lifting of Bundles . . ... 180
Subbundles and Quotients 180
3. Cohomology....... 182
Cohomology Groups . 182
Refinements . . . 184
Acyclic Coverings .. 185
Generalizations. . . . 186
The Singular Cohomology. 188
4. Meromorphic FUnctions and Divisors 192
The Ring of Germs .. 192
Analytic Hypersurfaces 193
Meromorphic Functions 196
Divisors . . . . . . . . . 198
Associated Line Bundles 200
Meromorphic Sections. . 201
5. Quotients and Submanifolds 203
Topological Quotients . . 203
Analytic Decompositions 204
Properly Discontinuously Acting Groups 205
Complex Tori . . . . . . . . . 206
Hopf Manifolds. . . . . . . . . 207
The Complex Projective Space 208
Meromorphic FUnctions . 210
Grassmannian Manifolds . . . 211
 Contents xiii

Submanifolds and Normal Bundles. 214

Projective Algebraic Manifolds 216
Projective Hypersurfaces 219
The Euler Sequence . . . 222
Rational FUnctions. . . . 223
6. Branched Riemann Domains 226
Branched Analytic Coverings 226
Branched Domains. . . . . . 228
Torsion Points . . . . . . . . 229
Concrete Riemann Surfaces . 230
Hyperelliptic Riemann Surfaces 231
7. Modifications and Toric Closures 235
Proper Modifications .. 235
Blowing Up . . . . . . . . . 237
The Tautological Bundle . 237
Quadratic Transformations 239
Monoidal Transformations 241
Meromorphic Maps 242
Toric Closures . . . . . . . 244

v Stein Theory 251

1. Stein Manifolds 251
Introduction . . . . . . 251
Fundamental Theorems '252
Cousin-I Distributions. 253
Cousin-II Distributions 254
Chern Class and Exponential Sequence 255
Extension from Submanifolds . . . '. . 257
Unbranched Domains of Holomorphy 257
The Embedding Theorem . 258
The Serre Problem. . . . . 259
2. The Levi Form . . . . . . . . . 260
Covariant Tangent Vectors 260
Hermitian Forms . . . . . . 261
Coordinate Transformations 262
Plurisubharmonic Functions 263
The Maximum Principle .. 264
3. Pseudoconvexity......... 266
Pseudoconvex Complex Manifolds 266
Examples . . . . . 267
Analytic Tangents . . . 274
4. Cuboids . . . . . . . . . . 276
Distinguished Cuboids. 276
Vanishing of Cohomology 277
Vanishing on the Embedded Manifolds 278
xiv Contents

Cuboids in a Complex Manifold 278

Enlarging U' .. 280
Approximation .. 281
5. Special Coverings .. 282
Cuboid Coverings 282
The Bubble Method 283
Frechet Spaces . . . 284
Finiteness of Cohomology . 286
Holomorphic Convexity . . 286
Negative Line Bundles. . . 287
Bundles over Stein Manifolds . 288
6. The Levi Problem'. . . . . . . . . 289
Enlarging: The Idea of the Proof . 289
Enlarging: The First Step . . . 290
Enlarging: The Whole Process 292
Solution of the Levi Problem 293
The Compact Case ...... 295
VI Kahler Manifolds 297
1. Differential Forms .... 297
The Exterior Algebra 297
Forms of Type (p, q) . 298
Bundles of Differential Forms . 300
2. Dolbeault Theory . . . . . . . . . 303
Integra;tion of Differential Forms 303
. The Inhomogeneous Cauchy Formula 305
The a-Equation in One Variable 306
A Theorem of Hartogs . 307
Dolbeault's Lemma 308
Dolbeault Groups 310
3. Kahler Metrics . . . . . 314
Hermitian metrics . 314
The Fundamental Form 315
Geodesic Coordinates . 316
Local Potentials .... 317
Pluriharmonic Functions 318
The Fubini Metric 318
Deformations . . . . . 320
4. The Inner Product . .. 322
The Volume Element 322
The Star Operator . . 323
The Effect on (p, q)-Forms 324
The Global Inner Product 327
Currents ...... 328
5. Hodge Decomposition . . . . . 329
 Contents xv

Adjoint Operators . 329

The Kiihlerian Case 331
Bracket Relations 332
The Laplacian . 334
Harmonic Forms 335
Consequences. . 338
6. Hodge Manifolds .. 341
Negative Line Bundles. 341
Special Holomorphic Cross Sections 342
Projective Embeddings 344
Hodge Metrics . 345
7. Applications . . . . . . . . 348
Period Relations . . . . 348
The Siegel Upper Halfplane . 352
Semi positive Line Bundles 352
Moishezon Manifolds 353

VII Boundary Behavior 355

1. Strongly Pseudoconvex Manifolds 355
The Hilbert Space . . 355
Operators . . . . . . . 355
Boundary Conditions 357
2. Subelliptic Estimates .. 357
Sobolev Spaces . . . . 357
The Neumann Operator . 359
Real-Analytic Boundaries . 360
Examples . . . . . 360
3. Nebenhiillen.......... 364
General Domains . . . . . 364
A Domain with Nontrivial Nebenhiille . 365
Bounded Domains . . . . . . . . . . . . 366
Domains in ((:2 . . . . . . . . . . . . . . 366
4. Boundary Behavior of Biholomorphic Maps. 367
The One-Dimensional Case . . . . . 367
The Theory of Henkin and Vormoor 367
Real-Analytic Boundaries 369
Fefferman's Result . . 369
Mappings . . . . . . . 371
The Bergman Metric 371
References 375

Index of Notation 381

Index 387
Chapter I

Holomorphic Functions

1. Complex Geometry
Real and Complex Structures. Let V be an n-dimensional com-
plex vector space. Then V can also be regarded as a 2n-dimensional real
vector space, and multiplication by i := yCT gives a real endomorphism
J : V , V with J2 = -id v . If {al, ... ,an} is a complex basis of V, then
{al' ... ' an, ial, ... , ian} is a real basis of V.
On the other hand, given a 2n-dimensional real vector space V, every real
endomorphism J : V , V with J2 = -id v induces a complex structure on V
by
(a + ib) . v := a ·'v + b· J(v).
We denote this complex vector space also by V, or by (V, J), if we want to
emphasize the complex structure.
If a complex structure J is given on V, then -J is also a complex structure.
It is called the conjugate complex structure, and the space (V, J) is sometimes
denoted by V. A vector v E V is also a vector in V. If z is a complex number,
then the product z· v, formed in V, gives the same vector as the product z· v
in V.
Our most important example is the complex n-space

en := {z := (z}, ... , zn) : Zi E e for i = 1, ... , n},

with the standard basis

el:= (1,0, ... ,0), ... ,en := (0, ... ,0,1).

We can interpret en as the real 2n-space

]R2n = {(x,y) = (x}, ... ,xn,Y}' ... ,Yn) : Xi,Yi E]R for i = 1, .. . ,np,

together with the complex structure J : ]R2n , ]R2n, given by

These considerations lead naturally to the idea of "complexification."

1 A row vector is described by a bold symbol, for instance v, whereas the corre-
sponding column vector is written as a transposed vector: v t.
2 1. Holomorphic Functions

Definition. Let E be an n-dimensional real vector space. The com-

plexification of E is the real vector space Ee := EffiE, together with the
complex structure J : Ee ~ E e , given by

J(v, w) := (-w, v).

Furthermore, conjugation ( in Ee is defined by

C(v, w) := (v, -w).

Since (oj = -Jo(, it is clear that (defines a complex isomorphism between

Ee and Ee·
The complexification of lR n is the complex n-space en identified with 1R 2n in
the way shown above. In this case the conjugation ( is given by

and will also be denoted by z t-+ z.

If V = Ee is the complexification of a real vector space E, then the subspace

Re(V) := {(v,O) : VEE} C V

is called the real part of V. Since it is isomorphic to E in a natural way, we

can write V ~ E ffi iE. If V is an arbitrary complex vector space, then V is
the complexification of some real vector space as well, but this real part is
not uniquely defined. It is given by the real span of any complex basis of V.
Example
Let E be an n-dimensional real vector space and E* := HomlR(E,lR) the
real dual space of linear forms on E. Then the complexification (E*)e can be
identified with the space HomlR(E, q of complex-valued linear forms on E.
In the case E = lR n , a linear form A E E* is always given by

with some fixed vector a E lRn. An element of the complexification (E*)e is

then given by v t-+ v· zt with z = a + ib E (lRn)e = en.
Now let T be an n-dimensional complex vector space and F(T) := HomlR(T, q
the space of complex-valued real linear forms on T. It contains the subspaces
T I := Homc(T, e) of complex linear forms and T I := Homc(T, q of complex
antilinear forms 2.
2 A real linear map>. : T -t C is called complex antilinear if >.( c . v) = c· >.( v) for
c E C. Therefore, T' can be viewed as the set of complex antilinear forms on T.
 1. Complex Geometry 3

Let {al, ... ,an } be a complex basis of T, and bi := iai, for i = 1, ... ,n.
Let {Ol, ... , on,{31 , ... ,,on} be the basis of T* = HomlR(T, JR.) that is dual to
{al, ... , an, bl ,···, bn }. Then we obtain elements

Ai := 0i + i,8i E F(T), i = 1, ... , n.

Claim. The forms Ai are complex-linear.

PROOF: Consider an element Z = Zlal +.. +znan E T with Zi = Xd..jYi E C.

Then
n n
Ak(Z) Ak ( L Xiai + LYibi)
i=l i=l
n n
= LXiAk(ai) + LYiAk(bi )
i=l i=l

Now the claim follows. •

It is obvious that the Ai are linearly independent. Therefore, {AI, ... , An} is
a basis of T/, and P:I, ... , Xn} is a basis of 'fl.
Since it is also obvious that T'n'f ' = {O}, we see that every element A E F(T)
has a unique representation
n n
A = LCiAi + LdiXi , with Ci,di E C.
i=l i=l

Briefly,
A = A' + A", with A' E T' and )," E 'f'.
Here A is real; i.e., A E HomJR(T,R) if and only if A" = N.

Hermitian Forms and Inner Products

Definition. Let T be an n-dimensional complex vector space. A Her-
mitian form on T is a function H : TxT -+ C with the following
properties:
1. v f-t H(v, w) is C-linear for every wET.
2. H(w,v) = H(v,w) for v,w E T.

It follows at once that w f-t H (v, w) is C-antilinear for every VET, and
H(v,v) is real for every vET. If H(v,v) > 0 for every v f. 0, H is called an
inner product or scalar product.
4 I. Holomorphic Functions

There is a natural decomposition

H(v, w) = S(v, w) + iA(v, w),

with real-valued functions S and A. Since

S(w, v) + iA(w,v) = H(w,v) = H(v,w) = S(v,w) - iA(v,w),

it follows that S is symmetric and A antisymmetric.

Example
If k is a field, the set of all matrices with p rows and q columns whose elements
lie in k will be denoted by Mp,q(k) and the set of square matrices of order n
by Mn(k). Here we are interested only in the cases k = JR and k = IC.
A Hermitian form on en is given by
H: (z, w) 1--+ zHw t ,

where H E Mn(C) is a Hermitian matrix, i.e., Ht = H.

The associated symmetric and antisymmetric real bilinear forms S and A are
given by

and
1
A(z,w)=Im(zHwt) = 2i(zHw t -wHz t ).

If H is an inner product, then S is called the associated Euclidean inner

product.
The identity matrix En yields the standard Hermitian scalar product
n

(zlw) =z·w t = Lzvwv.

/./=}

I
Its symmetric part (z w)2n:= Re«(zlw» is the standard Euclidean scalar
product. In fact, if we write z = x + iy and w = u + iv, with x, y, U, v E JRn,
then
1 -t -t
2(z.w +w·z)
n 1
= L 2(vvwv + wvzv)
v=}
n
= L(xvuv + YvVv).
v=1
 1. Complex Geometry 5

If the standard Euclidean scalar product on lRn is denoted by (-. ·1·· ·)n' we
obtain the equation

Balls and Polydisks

Definition. The Euclidean norm of a vector z E en is given by

the Euclidean distance between two vectors z, w by

dist(z, w) := liz - wll.

An equivalent norm is the sup-norm or modulus of a vector:

Izl:= max Izvl·

v=l, ... ,n

This norm is not derived from an inner product, but it defines the same topo-
logy on en as the Euclidean norm. This topology coincides with the usual
topology on ]R2n. We assume that the reader is familiar with it and mention
only that it has the Hausdorff property.

Definition. Br(zo):= {z E en : dist(z,zo) < r} is called the (open)

ball of radius r with center zo0

A ball in en is also a ball in lR 2n , and its topological boundary

8Br(zo) = {z E en : dist(z,zo) = T}

is a (2n - I)-dimensional sphere.

Definition. Let r = (rl, ... , Tn) E ]Rn, all Tv > 0, Zo = (ziO), ... , z~») E
en. Then
pn(zo, r) := {z E en : Izv - z~O)1 < Tv for v = 1, ... , n}

is called the (open) polydisk (or polycylindeT) with polyradius r and center
zoo If r E lR+ and r := (r, ... , r), we write P~(zo) instead of pn(zo, r).
Then P~(zo) = {z E en : Iz - zol < r}.
If 0 denotes the open unit disk in e, then pn := Pf(O) = 0 x ... x 0 is
~
n times
called the unit polydisk around o.
6 I. Holomorphic Functions

We are not interested in the topological boundary of a polydisk. The following

part of the boundary is much more important:
Definition. The distinguished boundary of the polydisk pn(zo, r) is
the set

Tn(zo,r) = {z E en : Izv - z~O)1 = rv for 1/ = 1, ... ,n}.

The distinguished boundary of a polydisk is the Cartesian product of n circles.

It is well known that such a set is diffeomorphic to an n-dimensional torus.
In the case n = 1 a polydisk reduces to a simple disk and its distinguished
boundary is equal to its topological boundary.

Connectedness. Both the Euclidean balls and the polydisks form a base
of the topology of en. By a region we mean an ordinary open set in en. A
region G is connected if each two points of G can be joined by a continuous
path in G. A connected region is called a domain.
If a real hyperplane in IRn meets a domain, then it cuts the domain into
two or more disjoint open pieces. For complex hyperplanes in the complex
number space (which have real codimension 2) this is not the case:
1.1 Proposition. Let G c en be a domain and

E := {z = (ZI,"" zn) E en : Zl = O}.

Then G' := G - E is again a domain.

PROOF: Of course, E is a closed set, without interior points, and G' = G - E

is open. Write point,s of en in the form z = (Zl' z*), with z* E en-I. Given
two points v = (VI, v*) and w = (WI, w*) in G', it must be shown that y
and w can be joined in G' by a continuous path. We do this in two steps.
Step 1: Let G = pn(zo, c) be a small polydisk. Then G' is the product of
a punctured disk and a polydisk in n - 1 variables. Define := (WI, Y*).z
z
Clearly, E G', and we can join VI and WI within the punctured disk, and
y* and w* within the polydisk. Therefore, y and w can be joined within G '.
Step 2: Now let G be an arbitrary domain. There is a path <p : I --+ G joining
y and w. Since <p(l) is compact, it can be covered by finitely many polydisks
U 1 , ... , Ul such that U).. c G for A = 1, ... , l.

It is easy to show that there is a 0 > 0 such that for all t', t" E I with
It' - t"l < 0, <pet') and <p(t") lie in the same polydisk Uk. Then let a = to <
tl < ... < tN = b be a partition of I with Itj - tj-Ii < 0 for j = 1, ... , N.
Let Zj := <p(tj) and A(j) E {I, ... , l} be chosen such that U)..(j) contains Zj
and Zj-l (it can happen that A(jr) = A(h) for jl =I- h). By construction Zj-l
lies in U)..(j) n U)..(j-l) , and thus U>'(j) n U>'(j-l) - E is always a nonempty
open set.
 1. Complex Geometry 7

We join v = Zo E U>'(l) and some point Zl E U>.(l) n U>'(2) - E by a path

CPl interior to U>'(l) - E. By (1) this is possible. Next we join Zl and a point
Z2 E U>'(2) n U>'(3) - E by a path CP2 interior to U>'(2) - E, and so on. Finally,
CPN joins ZN-1 and W = ZN within U>'(N) -E. The composition of cpI, ... , CPN
connects v and w in G'. •

Reinhardt Domains
Definition. The point set

will be called absolute space, the map r : en ~ JI with r(zl, ... , zn) :=
(lz11, .. . ,lznl) the natural projection.

The map r is continuous and surjective. For any r E "jI, the preimage
r- 1 (r) is the torus Tn(o, r) . For Z E en, we set Pz := pn(o, r(z)) and
T z := Tn(O,r(z)) = r-1(r(z)) (see Figure 1.1).
Definition. A domain G c en is called a Reinhardt domain if for
every z E G the torus T z is also contained in G.

Figure 1.1. A polydisk in absolute space

Reinhardt domains G are characterized by their images in absolute space:

r-1r(G) = G. Therefore, they can be visualized as domains in JI. For exam-
ple, both balls and polydisks around the origin are Reinhardt domains.
8 I. Holomorphic Functions

Example
Let Zo E en, v = 1, ... ,no Then T(e i9 . zo) = T(ZO), but
with Iz~O)1 > 1 for
11 . Izol > le i9 - 11, and for suitable () this expression
le i9 • Zo - zol = le i9 -
may be greater than C. SO pn (zo, c) is not a Reinhardt domain.

Definition. Let G c en be a Reinhardt domain.

1. G is called proper if 0 E G.
2. G is called complete if Vz E G n (C*)n : Pz c G (see Figure 12).

Later on we shall see that for any proper Reinhardt domain G there is a
smallest complete Reinhardt domain 8 containing G.

(a) (b)

IZII
>
Figure I.2. (a) Complete and (b) noncomplete Reinhardt domain

Exercises
e
1. Show that there is an open set B c 2 that is not connected but whose
image T(B) is a domain in absolute space.
2. Which of the following domains is Reinhardt, proper Reinhardt, complete
Reinhardt?
(a) a 1 := {z E e2 : 1 > IZII > IZ21},
(b) G2:= {z E e 2 : IZII < 1 and IZ21 < l-Izll},
(c) G 3 is a domain in e 2 with the property •
Z E G ==:} e it . Z E G for t E R
3. Let a c en be an arbitrary set. Show that G is a Reinhardt domain
<=> 39 c r
open and connected such that G = T- 1 (9).
4. A domain G c en is called convex, iffor each pair of points z, wE G the
line segment from Z to w is also contained in G. Show that an arbitrary
domain a is convex if and only if for every Z E 8a there is an affine
linear function A : en --* lR with A(Z) = 0 and Ale < O.
 2. Power Series 9

2. Power Series
Polynomials. In order to simplify notation, we introduce multi-indices.
For v = (Vi, ... , V n ) E zn and Z E en define

zr
n
Ivi := L Vi and zV:= 1 ••• z~n.
i=i

The notation V 2: 0 (respectively V > 0) means that Vi > 0 for each i

(respectively V 2: 0 and Vi > 0 for at least one i).
A function of the form

z r--+ p(z) = L avz v , with av E e for Ivl ~ m,

Ivl~m

is called a polynomial (of degree less than or equal to m). If there is a v

with Ivl = m and a v =1= 0, then p(z) is said to have degree m. For the
zero polynomial no degree is defined. An expression of the form avz v with
a v =1= 0 is called a monomial of degree m := Ivl. A polynomial p(z) is called
homogeneous of degree m if it consists only of monomials of degree m.
2.1 Proposition. A polynomial p(z) =1= 0 of degree m is homogeneous if
and only if
p(>.z) = >.m . p(z), for all >. E c.
PROOF: Let p(z) = avz v be a monomial of degree m. Then

The same is true for finite sums of monomials.

On the other hand, let p(z) = Llvl~N avz v be an arbitrary polynomial with
p(>.z) = >.m . p(z). Gathering monomials of degree i, we obtain a polyno-
mial Pi(Z) = Llvl=i a"z" with Pi(>'Z) = >.i . Pi(Z). Then for fixed z the two
polynomials
N
>. r--+ p( AZ) = L Pi (z) . Ai and A r--+ Am. p( Z)
i=O

are equal. This is possible only if the coefficients are equal, i.e., Pm(z) = p(z)
and Pi(Z) = 0 for i =1= m. So P = Pm is homogeneous. _

Convergence. If for every v E No

a complex number CII is given, one
can consider the series Lv>o CII and discuss the matter of convergence. The
trouble is that there is no canonical order on No.
10 I. Holomorphic Functions

Definition. The series Lv>o Cv is called convergent if there is a bi-

jective map cp : N -t No such that L~llc<p(i)1 < 00. Then the complex
number L~l C<p(i) is called the limit of the series.

It is clear that this notion of convergence is independent of the chosen map

cp, and that it means absolute convergence.
2.2 Proposition. Lv;::>:o C v is convergent if and only if

{ Llcvl: IeNg finite}

vEl
is a bounded set.
The proof is trivial.
2.3 Proposition. If the series Lv;::>:o Cv converges to the complex number
c, then for each E > a there exists a finite set 10 e No such that:
1. L jcvl < for any finite set KeNo with K n 10 = 0.
E,
vEK
2. IL Cv - I < for any finite set I with 10 e I e No·
C E,
vEl

PROOF: We choose a bijective map cp : N -t No. Then E~l c<p(i) = c, and

the series is absolutely convergent. For a given E > a there exists an io E N
I
such that L~io IC<p(i) I < E and L;~l c<p(i) - C < E. I
Setting 10 := 'P( {1, 2, ... , io}), it follows that LVEKlcvl < E for any finite set
I
K with K n 10 = 0, and LVElo Cv - C < E. I
Then for any finite set I with 10 e I e No,

ILCv-CI=I(LCv-C)+
vEl vElo
L cvl::;ILcv-cl+ L Icvl<2E.
vEl vElo vEl -10 -10

•
Example
Let ql, ... , qn be real numbers with 0 < qi < 1 for i = 1, ... , n, and q :=
(ql, ... , qn). Then for any v E No, qV = qr 1 ••• q~n is a positive real number.
If I e No is a finite set, then there is a number N such that I e
{a, 1, ... , N}n, and therefore
 2. Power Series 11

Since the partial sums are bounded, the series is convergent. It is absolutely
convergent in any order, and the limit is
1
"'qV-II-
n

~ - l-q··
V~O i=l •

We call this series the generalized geometric series.

Now let M c en be an arbitrary subset, and {tv : v E No} a family of
complex-valued functions on M. We denote by IlfvllM the supremum of Ifvl
onM.
Definition. The series 2:v>o fv is called normally convergent on M
if the series of positive real numbers 2:v~ollfvIlM is convergent.

2.4 Proposition. Let the series 2:v>o fv be normally convergent on M.

Then it is convergent for any Z EM, and for any bijective map <p : N -+ No
the series 2:: 1 f'P(i) is uniformly convergent on M.

PROOF: If the series is normally convergent, then 2:v>ollv(z)I is convergent

for any z E M. But then 2:v>o fv(z) is also convergent, and there is a complex
number f(z) such that 2::1 f'P(i)(Z) converges to f(z), for every bijective
map <p : N -+ No.
If an c > 0 is given, there is an io such that 2::io III'1'( i) I M< c. Then
m m
IL f'P(i)(Z) I S Lllf'P(i) 11M < c, for m > k 2: io and z E M.
i=k i=k

Therefore,
k
IL I= I L IS c, for k 2: i o·
~

f'P(i)(Z) - f(z) f'P(i)(Z)

i=1 i=k+1

This proves the uniform convergence. •

Power Series. Let {a v : v E No} be a family of complex numbers, and
Zo E en a point. Then the expression

is called a (formal) power series about zoo It is a series of polynomials. If

this series converges normally on a set M to a complex function f, then as a
uniform limit of continuous functions f is continuous on M.
12 I. Holomorphic Functions

2.5 Abel's lemma. Let pI ee p e C n be polydisks around the origin. 3 If

the power series Z:v>o a"z" converges at some point of the distinguished
boundary of P, then it converges normally on P'.

PROOF: Let w E ooP be a point where ~,,>o a"w" is convergent. Then

there is a constant c such that la"w"l :::; c for an v E No.
We choose real numbers qi with 0 < qi < 1 such that IZil :::; qilwil for any
Z = (Zl,"" zn) E pI and i = 1, ... , n. It follows that

Then lIa"z"lIp' :::; q" . c as well, and from the convergence of the generalized
geometric series it follows that E,,~o avz" is normally convergent on P'. •

Definition. We say that a power series ~v>o a,,(z - zo)" converges

compactly in a domain G if it converges normally-on every compact subset
KeG.

2.6 Corollary. Let P e C n be a polydisk around the origin and w be a point

of the distinguished boundary of P. If the power series E,,>o a"z" converges
at w, then it converges compactly on P. -

PROOF: Let K e P be a compact set. Then there is a q with 0 < q < 1

such that K e q . P ee P. Therefore, the series is normally convergent on
K. •

Let S(z) = E,,~o a"z" be a formal power series about the origin, and
B := {z E C n : S(z) convergent}.

2.7 Proposition. The interior BO is a complete Reinhardt domain, and

S(z) converges compactly in BO.

PROOF: Let w be a point of BO. There is a polydisk pn (w, c) e BO and

a point v E pn(w,c:) n (C*)n such that w E Pv(O). Then Tw e B O , and if
wE (C*)n, then also Pw(O) e BO.

To see that BO is a complete Reinhardt domain, it remains to show that it

is connected. But this is very simple. Every point of BO can be connected to
a point in BO n (c*)n, and then within a suitable polydisk to the origin.
3 The notation U cc V means that U lies relatively compact in Vj i.e., V is a
compact set which is contained in V.
 2. Power Series 13

From these considerations it follows that BO is the union of relatively compact

polydisks around the origin. Therefore, S(z) converges compactly on BO. •
The set BO is called the domain of convergence of S(z).
2.8 Proposition. Let G be the domain of convergence of the power series
S(z) = 2:v~o avz v . Then

also converges compactly on G.

PROOF: Let w be any point of (c*)nnG, and lavwvl ::; c for every v E No.
H0<q < 1 and z = q . w, then

Ia v Vi
. VjZl
Vj
... Zj '"
Vn
zn
I = IZjl
Vj
. Iavz VI <- IZjl
c
. Vj . q
Ivl
.

Now,

is convergent. Therefore, SZj(z) is convergent, and it follows that SZj is nor-

mally convergent on Pz(O). Since every compact set KeG can be covered
by finitely many polydisks of this kind, SZj is compactly convergent on P .•

Definition. Let B c C n be an open set. A function f : B ---t C is called

holomorphic if for every Zo E B there is a neighborhood U = U(zo) C B
and a power series S(z) := 2:v>o av(z - zo)V that converges on U to
f(z). -
The set of holomorphic functions on B is denoted by O(B).

It is immediately clear that every holomorphic function is continuous.

Exercises
1. Let f, 9 be two nonzero polynomials. Prove that

deg(f· g) = deg(f) + deg(g).

2. Let f = fr "'!k be a homogeneous nonzero polynomial. Show that fi is
homogeneous, for i = 1, ... , k.
14 1. Holomorphic Functions

3. Find the domain of convergence for the following power series:

00

I(z,w) = LZw\ g(z,w) = L(zW)k,

k~O k=O

4. Determine the limit and the domain of convergence of the series

F(z,w) = L(2z)V + L zl' )w v .

v~O I'~O

5. A polyradius r = (rt. ... ,rn ) E "f/ is called a radius 01 convergence for

the power series I(z) = Ev>o avz v if I(z) is convergent in P = pn(o, r),
but not convergent in any polydisk pI = pn(o, r') with P cc P'.
Prove the following generalization of the root test:
r is a radius of convergence for I(z) if and only if lim ''\Ilavlr v = 1.
v~O

3. Complex Differentiable Functions

The Complex Gradient
Definition. Let Been be an open set, Zo E B a point. A function
I : B -+ e is called complex differentiable at Zo if there exists a map
a : B -+ en such that the following hold:
1. a is continuous at Zo.
2. 1(.,.) = I(zo) + (z - zo) . a(z) t for z E B.

Complex differentiability is a local property: For I to be complex differen-

tiable at Zo it is sufficient that there is a small neighborhood U = U(zo) c B
such that the restriction Ilu is complex differentiable at zoo
3.1 Proposition. II I is complex differentiable at zo, then the value 01 the
function a at Zo is uniquely determined.

PROOF: Assume that there are two maps al and a 2 satisfying the condi-
tions of the definition. Then

(z - zo)' (a1(z) - a 2 (z» t = 0 for every z E B.

In particular, there is an e > 0 such that the equation holds for z = Zo + tei,
with tEe, ItI < e, and i = 1, ... , n. If a A = (ai).), ... , a~», then

t· (a~l)(z) - a~2)(z» = 0, for It I < e, z = Zo + tei, and i = 1, ... ,n.

Because the a~).) are continuous at Zo, it follows that al(zo) = D. 2 (zo). •
 3. Complex Differentiable Functions 15

Definition. Let J : B -+ C be complex differentiable at Zoo If there

exists a representation

J(z) = !(zo) + (z - zo) . A.(z) t,

with A. continuous at Zo, then the uniquely determined numbers

are called the partial derivatives of ! at Zo. The vector

yo !(zo) := (fZl (zo),· .. , JZn (zo)) = A.(zo)

is called the complex gradient of ! at Zo.

Remarks
1. If ! is complex differentiable at Zo, then J is continuous there as well.
2. A function J is caUed complex differentiable in an open set B if it is
complex differentiable at each point of B. Then the partial derivatives of
! define functions !zv on B. If each of these partial derivatives is again
complex differentiable at Zo, then J is called twice complex differentiable
at Zo, and one obtains second derivatives

By induction, partial derivatives of arbitrary order may be defined.

3. Sums, products, and quotients (with nonvanishing denominators) of com-
plex differentiable functions are again complex differentiable.

Weakly Holomorphic Functions. Let Been be an open set,

Zo E B a point, and! a complex-valued function on B. For w =1= 0 let
'Pw : e -+ en be defined by

'Pw() := Zo + (w.
Then! 0 <Pw () is defined for sufficiently small (. If J is complex differentiable
at Zo, then we have a representation J(z) = !(zo) + (z - zo) . A.(z) t, with A.
continuous at zo0 It follows that

and! 0 'Pw is complex differentiable at ( = 0, with

16 I. Holomorphic Functions

This is the complex directional derivative of 1 at Zo in the direction w. We

denote it by Dw/(zo). In particular, Iz" (zo) = De.J(zo} for v = 1, ... ,n.
An arbitrary function I is called partially differentiable at Zo if all partial
derivatives D e ,,1(zo} exist for v = 1, ... ,n.
A function I is called weakly holomorphic on B if it is continuous and partially
differentiable on B. Then for Z = (Zl,".' zn) E B and v = 1, ... , n the
functions
(H I(Zl,." ,Zv-ll(,zvH, ... ,zn)
are holomorphic functions of one variable.
If I is complex differentiable on B, then I is also weakly holomorphic on B.
Later on we shall see that weakly holomorphic functions are always complex
differentiable, in contrast to the behavior of real differentiable functions.

Holomorphic Functions
3.2 Proposition. Let Peen be a polydisk around the origin, and S(z) =
Ev~o ayz V a power series that converges compactly on P to a function 1.
Then I is complex differentiable at 0, with

IZI (0) = al.O •...• O,· •.• Izn (0) = ao ..... O.I·

PROOF: We choose a small polydisk PE C C P around the origin such that

S(z) is normally convergent on PE. But then the series obtained by any
rearrangement of the terms is also normally convergent, and it converges to
the same limit. We write

I(z) = Lavz v
v~O

= ao,o •...•o + ZI • L avzr1 - 1 Z~2 ••• z~n

Vl>O
V2, •.• ,lIn :2:::0

Vl=O.V2>O Vl=· .. =Vn_l=O

113,···,Vn~O vn>o

Since the series Al (z), ... , An (z) converge normally on PE to continuous func-
tions, I is complex differentiable at 0, with 1z,,(0) = Ay(O). •

3.3 Corollary. II Been is an open set, and 1 :B ~ e a holomorphic

lunction, then I is complex differentiable on B.
 4. The Cauchy Integral 17

PROOF: Let Zo E B be an arbitrary point. There is a power series S(w)

converging compactly near 0 to a holomorphic function 9 such that

f(zo + w) = g(w) = g(O) + WI' .ell(w) + ... + Wn . .eln(w),

with continuous functions .ell, ... '~n' It follows that f is complex differen-
tiable at zoo •

Exercises
1. Show that there is a function f : C n ~ C that is complex differentiable
at every point z = (Zl,"" zn) with Zn = 0, but is nowhere holomorphic.
2. Prove the following chain rule: If G c C n is a domain, f : G ~ C a
complex differentiable function, and <P = (<pi> ... , <Pn) : ~ ~ G a map
with holomorphic components <Pi, then f 0 <P : ~ ~ C is a holomorphic
function, with (f 0 <p)'«() = Vf(<p«()). <p'«() t.

4. The Cauchy Integral

The Integral Formula. Let r = (rb ... , rn) be an element of IR+.,
P = pn(O,r), T = Tn(O,r), and f a continuous function on T. Then

defines a continuous function k f : P x T ~ C.

Definition.

is called the Cauchy integral of f over T.

Obviously, Cf is a continuous function on P.

4.1 Theorem (Cauchy integral formula). Let P, T be as above, and
U = U(P) be an open neighborhood of the closure of P. If f is weakly holo-
morphic on U, then CflT(z) = fez) for any z E P.

PROOF: If P = Drl (0) x· .. x Drn (0), we may assume that U = U I X· •• X Un,

with open neighborhoods Ui = Ui (Dr' (0)), fori = 1, ... , n.

18 I. Holomorphic FUnctions

Since I is weakly holomorphic, we can fix z' = (Zl' ... , Zn-l) E U1 X .•. x Un-l
and apply the Cauchy integral formula in one variable to (n ~ I(z', (n). For
Zn E Drn (0) it follows that

I~z',zn) = 2:i !
I<nl=rn

Similarly, for the penultimate variable Zn-l and z" = (Zl,"" Zn-2) E U1 X
••• X Un -2 we obtain

I (z" ,Zn-1! Zn ) 1
= 211"i !
I~n-d=r .. - l

and after n steps, I(z) = CfIT(Z), for z E P. •

4:2 Theorem (power series expansion). Let P = pn(o, r) C en be
a polydisk and T its distinguished boundary. II I : T -+ e is a continuous
junction, then there is a power series Ev>o avzv that converges to C,(z) in
all 01 P. -
The coefficients av 01 this series are given by

PROOF: Setting 1 := (1, ... ,1) E N~, for z E P and ( E T it follows that
1
« - Z)l

If r = (Tl." ., Tn), then for fixed z E P and arbitrary ( E T we have

z-I ='qj:= Iz-I

I-2 _1 < 1, for j = 1, ... ,n.
(j Tj

Since T is compact and I continuous on T, there is a constant M with

I/«)I ~ M on T. Then Ev;::o (J«)/(V+1 )zv is dominated on T by the
 4. The Cauchy Integral 19

convergent series (M/~l) LII>oqll, where q = (qt. ... ,qn), and therefore it
is normally convergent as a series of functions on T with limit I(t;)/(t; - Z)l.
We can interchange summation and integration:

Gj(z)
1
= ( 21Ti
)n iT(I(t;) "II
(t; _ Z)l dt; = ~ allz ,

l {~~!
with
a v := (2~i) n dt;.
The series converges for each Z E P.
-
4.3 Osgood's theorem. Let Been be an open set. The following state-
ments about a function I : B -? e are equivalent:
1. f is holomorphic.
2. f is complex differentiable.
3. f is weakly holomorphic.

PROOF: We already know that a holomorphic function f is complex differ-

entiable, and it is trivial that then f is weakly holomorphic.
On the other hand, let f : B -? e be weakly holomorphic, and Zo E B
an arbitrary point. There is a small polydisk P around Zo that is relatively
compact in B. If T is its distinguished boundary, then lip = GilT, and the
Cauchy integral is the limit of a power series. So f is holomorphic. _
In addition, if f is weakly holomorphic on B, Zo E B a point, and Pee B
a polydisk around Zo, then there is a power series S(z) = Lv>o av(z - zo)V
that converges to f on all of P. -

Holomorphy of the Derivatives

4.4 Weierstrass's convergence theorem. Let G c en be a domain, and
(ik) a sequence of holomorphic functions on G that converges uniformly to a
function f. Then f is holomorphic.

PROOF: The limit function is continuous. Let Zo E G be a point, Pee G

a polydisk around Zo, and T its distinguished boundary. Then

Since T is compact, we can interchange integral and limit. Thus, for any fixed
z E P,
20 I. Holomorphic Functions

Since 1 is continuous on T, the Cauchy integral CIIT has a power series

expansion in P. Therefore, 1 is holomorphic at ZOo •

4.5 Proposition. Let S(z) = 2:1I~0 allz ll be a power series and G its do-
main 01 convergence. Then the limit lunction 1 01 S(z) is holomorphic on G,
and the lormal derivative
SZ; () '"""' all' VjZ!VI ... ZjII; -1 ... ZnVn
Z = ~
11>0
IIj>O

converges to Iz j • In particular, all partial derivatives 01 1 are likewise holo-

morphic.

PROOF: Since S(z) converges compactly on G, 1 is locally the uniform limit

of a sequence of polynomials. Then it follows from Weierstrass' theorem that
1 is holomorphic. But also SZj (z) converges compactly on G, and its limit
function gj must be holomorphic on G.
Now let Zo be an arbitrary point of G. Since G is a complete Reinhardt
domain, there is a polydisk P around the origin with Zo EPee G. We
define

f*(z):= 1 z
; gj(z!"", Zj-1, (, Zj+l,"" zn) d( + I(z!, ... ,0, ... , zn).

For the path of integration we take the connecting segment between 0 and
Zj in the zrplane. Then f* is defined on P.
Let S (z) = 2:: 0Pi (z) be the expansion into a series of homogeneous polyno-
mials. Then SZj (z) = 2::0 (Pi)zj (z), and this series converges uniformly on
the compact path of integration we used above. Therefore, we can interchange
summation and integration, and consequently,

00

LPi(Z) = I(z),
i=O

for z E P. Hence IZ j (zo) = I:

j (zo) = gj(zo), •
4.6 Corollary. Let G c en be a domain and f : B -+ e a holomorphic
lunction. Then 1 is infinitely complex differentiable in G.

Let v = (VI, ... , v n ) be a multi-index. Then we use the following abbrevia-

tions:
 4. The Cauchy Integral 21

1. v! := VI!··· v n !.
2. If f is sufficiently often complex differentiable at zo, then
olvlf
D V f(zo) := 8 zlV1 ... 8 Znv" (zo).

4.7 Identity theorem (for power series).

Let fez) = 2::v>o avz v and g(z) = 2::v>o bvz v be two convergent power series
in a neighborhood U = U(O) c en. Ifthere is a neigborhood V(O) C U with
flv = glv, then av = bv for all v.

PROOF: We know that f and 9 are holomorphic. Then DV 1(0) = DVg(O)

for all v, and successive differentiation gives
D V f(O) = v! . a v and D V g(O) = v! . by.

•
4.8 Corollary. Let G c en
be a domain, Zo EGa point, and I : B -t e
a holomorphic function. If I(z) = 2::v>o av(z - zo)'" is the (uniquely deter-
mined) power series expansion near Zo -E G, then

av = ~
v.
.D V f(zo), for each v E No.
4.9 Corollary (Cauchy's inequalities). Let G c en be a domain, I :
G -t e holomorphic, Zo EGa point, and P = pn (zo, r) C eGa polydisk
with distinguished boundary T. Then
vI
ID v l(zo)1 :S --=-
rV
. suplfl·
T

PROOF: Let fez) = 2::v>o av(z - zo)V be the power series expansion of f
at zoo Then DV I(zo) = v!av and

av = (2~i) n h«( !~:~V+l d(,

and therefore

ID V f(zo)1 <

(2:!)n I
[0,211")"
vI
< --=-.
rV
suplfl·
T

•
22 I. Holomorphic Functions

The Identity Theorem. Let G c C n be always a domain. The con-

nectedness of G will be decisive in the following.
4.10 Identity theorem (for holomorphic functions).
Let h, h be two holomorphic junctions on G. Ij there is a nonempty open
subset U c G with It/u = hlu, then h = h·

PROOF: We consider I := h - h and the set

N:={ZEG: D"/(z) =0 for all v}.

Then N =f. 0, since U c N. Let Zo E G be an arbitrary point, and

j(Z) = L ~D" I(zo)(z - zo)"

,,<::0 v.
the power series expansion of j in a neighborhood V = V(zo) c G. If Zo
belongs to N, then Ilv == 0, and also V C N. This shows that N is open.
Because all derivatives D" I are continuous, N is closed. Since G is a domain,
we get N = G and h = h. •
Remark. In contrast to the theory of one complex variable, it is not suf-
ficient that 11 and h coincide on a set M that has a cluster point in G.
Consider for example, G = C 2 and M = {(ZI, Z2) : Z2 = o}. The holomor-
phic functions II(ZI,Z2):= Z2(Zl-Z2) and h(zI,z2):= Z2(ZI +Z2) are equal
on M, but h(O, 1) = -1 and 12(0,1) = 1.

4.11 Theorem (maximum principle).

Let j : G -t C be a holomorphic function. Ij there is a point Zo E G such
that III has a local maximum at zo, then I is constant.

PROOF: We consider the map t.pw : C -t cn with t.pw«) = Zo + (w, for

an arbitrary w =f. o. Then I 0 <Pw is a holomorphic function of one complex
variable, defined near (" = O. Now, since Ij 0 <Pwl has a local maximum at
the origin, this function must be constant in a neighborhood of the origin.
But the direction w was chosen arbitrarily, so I also has to be constant in a
neighborhood of 0 E C n . The identity theorem implies that I is constant on
G. •

Exercises
1. Prove Liouville's theorem: Every bounded holomorphic function on en
is constant.
2. Prove that if
1/ E NO', then
IE O(C n ) and If(z)1 :s c· IIz"lI for some C>
I is a polynomial of degree at most II/I·
° and some
 5. The Hartogs Figure 23

3. Let G c en be a domain and f E O(G) not constant. Prove that then

f(U) c e is open for any open subset U C G.
4. Let G c en be a domain. A set :F of holomorphic functions on G is called
locally bounded, if for every z E G there is a neighborhood U(z) C G such
that { II f II u : f E :F} is bounded. Prove the following:
(a) (Lemma of Ascoli) If A eGis a dense subset and (fn) is a locally
bounded sequence of holomorphic functions in G which converges
pointwise on A, then (fn) is compactly convergent on G.
(b) (Theorem of Montel) Every locally bounded sequence of holomorphic
functions in G has a comp~tly convergent subsequence.
Hint: More or less, you can use the well-known proof from the 1-
dimensional theory.

5. The Hartogs Figure

Expansion in Reinhardt Domains. Let r~, r~ be real numbers with
o < r~ < r~ for 1 :5 v :5 n. We define
P .- {z E en : Izvl < r~ for all v},
Q .- {z E en : r~ < Izvl < r~ for all v}.
Clearly, P and Q are Reinhardt domains. Let f be a holomorphic function in
Q. Then for all r E 1"(Q), the Cauchy integral GilT. is a holomorphic function
in P r , and therefore a fortiori in P (see Figure 1.3).

T2 ------~

--77-r~~
P // I·
I
r I
I

Figure 1.3. Expansion in the polydisk

5.1 Proposition. The function fr : P ~ C given by fr(z) := GIlT. (z) is

independent of r.

PROOF: We have
24 I. Holomorphic Functions

In each variable (v the integrand I(,)/((v-zv) is holomorphic on the annulus

{(v : r~ < I(vl < r~}. From the Cauchy integral formula for one variable it

r r
follows that
1(,) d(v = 1(,) d(v
JI(vl=r v (v - Zv JI(vl=r~ (v - Zv
if r~ < rv :S r~ < r~. This yields the proposition. •
5.2 Proposition. Let G c C n be a proper Reinhardt domain, I holomor-
phic on G. Then lor every z E G n (C*) n the Cauchy integral GliTz coincides
with I in a neighborhood 01 the origin.

PROOF: G n (c*)n is a Reinhardt domain. Therefore, Go := T(G n (c*)n)

is a domain in the absolute space.
Let B := {r E Go : CIIT • coincides with I in the vicinity of O}. Then B 'f
0, because there is a small r E Go such that Pr(O) c G.
B is open: If ro E B, we can find sets P, Q as we did at the beginning of this
section such that ro E Q c Go. Then for r E Q, Ir = GilT. is a holomorphic
function on P, and independent of r. But Iro coincides with I near the origin.
Therefore, Q c B.
Also, Go - B is open. The proof goes as above. Since Go is connected, that
implies that B = Go. •

5.3 Corollary. Let G be a proper Reinhardt domain, I holomorphic in G.

Then there is a power series S(z) which converges in G to I.

PROOF: Let Zo G be arbitrarily chosen. Then there is a point w E G n

E
(C*)n with Zo E Pw . The holomorphic function g := GIlT", has a power
series expansion g(z) = Lv>o avz v in Pw ' Since g coincides with I in a small
neighborhood of the origin,-the coefficients a v are those of the Taylor series
of I about O. Since Zo was arbitrary, it follows that the series converges in
all of G. By the identity theorem its limit is equal to I. •

Definition. If G is a proper Reinhardt domain, then

.-
~·-
G
u
zEGn(C*)"

is called the complete hull of G.

 5. The Hartogs Figure 25

Remarks
1. Every complete Reinhardt domain is proper, but the opposite is in general
false. For n = 1, Reinhardt domains are open disks around 0, and there
is no difference between proper and complete domains.
2. The complete hull G of a proper Reinhardt domain G is again a domain
containing G. And it is Reinhardt: For z E G there is some Zl with
Z E P Z1 C G. But then also Tz C P Z1 C G. The same argument shows
that G is complete.
3. Let G I be another complete Reinhardt domain with G C G I . For z E
G n (c*)n, z also lies in G I , and by the completeness of G I it follows
that P Z C G 1. So G c G I, and we see that G is the smallest complete
Reinhardt domain containing G.
An immediate consequence is the following:
5.4 Theorem. Let G be a proper Reinhardt domain and f be holomorphic
1
in G. Then there is exactly one holomorphic function in G with = f. fiG
Hartogs Figures. In tl;1e case n = 1 the situation above cannot appear.
For n ;::: 2 we can choose sets G and Gin Cn such that G ::f G. This reflects an
essential difference between the theories of one and several complex variables.
Now let n 2 2, pn the unit polydisk, ql, ... , qn real numbers with 0 < q" < 1
for v = 1, ... ,n, and

H = H(q) := {z E pn : IZ11 > ql or Iz~1 < qlJ for I-L = 2, ... ,n}.

Then (pn, H) is called a Euclidean Hartogs figure (see Figure 1.4). H is a

proper Reinhardt domain and pn its complete hull.

(a)

q21------'
H

I ...,

Figure 1.4. (a) 2-dirnensional, and (b) 3-dimensional Hartogs figure

26 I. Holomorphic Functions

5.5 Hartogs' theorem. Let (pn, H) be a Euclidean Hartogs figure. Then

any holomorphic function f on H has a holomorphic extension on pn. 1
The theorem follows immediately from our considerations above.

Exercises
1. For 0 < q < 1 let G l , G 2 c e2 be defined by
Gl .- {(z,w): q < Izl < 1 and Iwl < I},
G2 .- {(z,w): Izl < 1 and Iwl < q}.
(a) Prove that every holomorphic function f on G 1 has a unique repre-
sentation

L
00

f(z,w) = an(w)zn, with an E 0(0).

n=-oo

(b) Prove that every holomorphic function 9 on G 2 has a unique repre-

sentation
00

g(z, w) = I>n(W)zn, with bn E O(Oq(O».

n=O

(c) Use (a) and (b) to prove that every holomorphic function f on G l UG2
has a unique holomorphic extension to the unit polydisk.
2. Let G c en be an arbitrary Reinhardt domain, f E O(G). Show that
there exists a uniquely determined "Laurent series" LVEZn avz v converg-
ing compactly in G to f.

6. The Cauchy-Riemann Equations

Real Differentiable Functions. Recall the following from real an-
alysis:
Let Been be an open set and Zo a point of B. A function f : B -+ IR is called
differentiable (in the real sense) if there is a real linear form L : en -+ IR and
a real-valued function r with:
1. fez) = f(zo) + L(z - zo) + r(z - zo)·
. r(w)
2. ll~O /lw/l = O.
The real linear form D f(zo) := L is called the (total) derivative of f at zoo
It can be given in the form
 6. The Cauchy-Riemann Equations 27

L :u + iv t-+ u· \7 xf(zo) t + v· \7yf(zo) t,

with \7 xf(zo) = (fXl (zo), ... , fXn (zo)) and \7 yf(zo) = (fYl (zo), ... , fYn (zo)).
We call C\7xfCzo), \7yfCzo)) the real gradient of fat zo0
If f = 9 + ih : B ~ e is a complex-valued function, then f is called differen-
tiable (in the real sense), if g and h are differentiable. The (real) derivative
of f at Zo is defined to be the complex-valued real linear form
D f(zo) := Dg(zo) + iDh(zo).

6.1 Proposition. A function f : B ~ e is (real) differentiable at zo if and

only if there are maps 1).',1)." : B ~ en such that:
1. 1).' and 1)." are continuous at Zo.
2. fez) = f(zo) + (z - zo) . I).'(z) t + (z - zo) . I)."(z) t for Z E B.
The values I).'(zo) and I).II(ZO) are uniquely determined.

PROOF: (1) Let f be differentiable at Z00 Then there is a complex linear

form A' and a complex antilinear form A" such that
D f(zo) = A' + A".
The decomposition is uniquely determined, and there are vectors I).'(zo) and
I)."(zo) such that
A'(w) = w· I).'(zo) t and A"(w) = w· I)."(zo) t.
Now we define
, r(z - zo) _ _
S(z) .- I). (zo) + 211z _ zoll2 . (z - zo),

"( r(z - zo) ( )

1)." (z) .- I). zo) + 211z _ ZOll2 . Z - Zo .

It is easy to see that

(a~' and 1)." are continuous at Zo,
(bY(z) = f(zo) + (z - zo) . I).'(z) t + (z - zo) . I)."(z) t.

(2) Now let the decomposition be given, and define

L(w) .- w· 1).' (zo) t + W . 1)." (zo) t,
r(w) := w· (I).'(z) - 1).' (zo)) t + w· (I).II(Z) - 1)." (zo)) t.

Since
I~I~~I ~ 1II).'(z) - I).'(zo)II + III).II(Z) - 1)." (zo)II,

it follows that f is differentiable at Zo with derivative L. •

28 I. Holomorphic Functions

Wirtinger's Calculus
Definition. Let f :B ~ e be real differentiable at Zo0 If we have a
representation

fez) = f(zo) + (z - zo) . !::J.' (z) t + (z - zo) . !::J." (z) t,

with!::J.' and!::J." continuous at zo, then the uniquely determined numbers

and
~ (zo) = /z., (zo) := e v • !::J." (zo) t
uZv

are called the Wirtinger derivatives of f at Zo.

The complex linear (respectively antilinear) forms (8f)zo : en ~ e and
(8f)zo : en ~ e are defined by
n n
(8f)zo(w):= I:fz~(zo)wv and (8f)zo(w) := I:fz~(zo)wv,
v=l v=l

and the differential of f at Zo by (df)zo := (8f)zo + (8f)zo .

Obviously, Df(zo) = (df)zo.
If we introduce the holomorphic (respectively antiholomorphic) gradient

'If:=(fZl,···,fZn) and 'If:=(h1, ... ,hJ,

then (8f) zo(w) = W· 'If(zo)t and (8f) zo(w) = W· 'If(zo)t.
6.2 Proposition. Let f be a (complex-valued) function that is real differ-
entiable at Zo. Then

fz~(zo) = ~(fx.,(zo) - ify.,(zo»,

h~(zo} = ~(fx,,(zo) + ify.,(zo».

PROOF: Let be L := D f(zo). Then

fx.,{zo) = L(ev } = (8f)zo(ev) + (8f)zo(ev) = fz.,(zo) + h~(zo)

and
 6. The Cauchy-Riemann Equations 29

Putting things together we obtain

Ix,,(zo) - i/ll.,(zo) = 2/z.,(zo) and Ix.,{zo) + i/ll" (zo) = 2!z,,(zo).

Remark. Use these formulas with care! The derivatives Ix" and in I""
-
general are complex-valued. So the equations do not give the decomposition
of Iz" and /Z" into real and imaginary parts, respectively!

The Cauchy-Riemann Equations

6.3 Theorem. Let I : B 4 C be a continuously real differentiable function.
Then I is holomorphic il and only il /Z" (z) == 0 on B, lor v = 1, ... , n.

PROOF: (a) If I is holomorphic, then I is complex differentiable at every

point Zo E B. Comparing the two decompositions

I(z) = I(zo) + (z - zo) . a(z) t

and
I(z) = I(zo) + (z - zo) . a'(z) t + (z - zo) . a"(z) t
we see that a'(zo) = a(zo) and a"(zo) = O. The latter equation means that
/z.,(zo) == 0 for v = 1, ... ,n.
(b) If /z.,(z) == 0, then I is holomorphic in each variable and is consequently
holomorphic. _

Remark. Now the following is clear: If I is holomorphic near zo, then

n

(8f)zo =0 and DI(zo)(w) = (df)zo(w) = (tJf)zo(w) = Llz"(zo)w,,.

,,=1

The equation (af)z = 0 is the shortest version of the Cauchy-Riemann dil-

lerential equations. In greater detail, these are the equations

/z,,(z) == 0, for v = 1, ... ,no

Finally, if I = 9 + i h, then we can write the Cauchy-Riemann equations in

their classical form:

gx" = hy" and hx" = -gil,,' for v = 1, ... , n.

30 I. Holomorphic Functions

Exercises
1. Derive the Cauchy-Riemann equations in their classical form.
2. Let f: G -t e be real differentiable. Prove the formulas

(fz,,) = Clh" and fz"z,. = /z,.z" for 1/, J.L = 1, ... ,n.
3. Let G c en be a domain and h, ... ,!k : G -t e holomorphic functions.
Show that if E;=l kl
j is constant, then all fJ are constant.

~ = I-a I·
aa2jhl2 ah 2
Hint: If h is holomorphic, then
ZjVZj Zj

7. Holomorphic Maps
The Jacobian. Let Been be an open set. A map
f = (h, ... , fm) : B -t em
is called holomorphic (respectively real differentiable) if all components fi
are holomorphic (respectively real differentiable).
1.1 Proposition. The map f : B -t em is holomorphic if and only if
for any Zo E B there exists a map ~ : B -t Mm,n(C) with the following
properties:
1. ~ is continuous at zoo
2. f(z) = f(zo) + (z - zo) . ~(z) t, for z E B.
The value ~ (zo) is uniquely defined.

PROOF: The map f is holomorphic if there are decompositions

with ~jJ continuous at zo, for f.L = 1, ... , m.

Then ~ is given by ~(z) t = (~1 (z) t, ... ,~m(z) t). We leave the further
details to the reader. •

Definition. If f : B -t em is holomorphic, then Jr(zo) := ~(zo) is

called the complex Jacobian (matrix) of f at zoo The associated linear
map f'(zo) : en -t em is called the (complex) derivative of fat zoo It is
given by
 7. Holomorphic Maps 31

Explicitly, we have

)
This matrix is also defined for differentiable maps.
Definition. If f = g + ih : B -+ C m is a differentiable map, then the
real Jacobian matrix JlR,r(zo) E M 2m ,2n(lR) is the real matrix associated
to the real linear map
(Dg(zo), Dh(zo)) : C n = ]R2n -+ ]R2m.

The real Jacobian of f = g + ih is given by

Jlft,r =

The ]R-linear map Df(z) : C n -+ C n is defined by Df(z) := Dg(z) + i Dh(z).

Setting (8f)z := ((8fdw'" (8fm)z) and (8f)z := ((8fdz,"" (8fm)z) , we
obtain
Df(z) = (8f)z + (8f)z.

7.2 Theorem. A differentiable map f = g + ih : B -+ C m is holomorphic

if and only if Df(z) is C-linear for every z E B.
If f is holomorphic and n = m, then det(JR,r(z)) = Idet Jr(z)j2.

PROOF: The map f is holomorphic if and only if (8f)z = 0 for every z. Then
Df(z) = (8f)z, which is complex linear. In this case we have the Cauchy-
Riemann equations

and therefore

If n = m, then Jlft,r = ( AC ~ ) with B = -C and A = D, and Jr =

A+iC.
32 I. Holomorphic Functions

By elementary transformations,

= det ( A~iC
det ( A~iC
= Idet(A + iCW.

•
It follows that holomorphic maps are orientation preserving!

Chain Rules. Let Been be an open set, f : B --t em a differentiable

map, and 9 a complex-valued differentiable function that is defined on the
image of f. Then 9 0 f : B --t e is differentiable, and the following holds:
7.3 Proposition (complex chain rule).
m m
(gof)z., = L(gw,. of)· (f/-&)z., + L(gw,. of)· (7,. tv ,
/-&=1
m m
(g 0 fhv = L(gw,. 0 f) . (f/-&hv + L(gw,. 0 f) . (7,.)zv·
/-&=1

One can use the well-known proof for the chain rule in real analysis, consid-
ering ZII and "'ill as independent variables.
7.4 Corollary. If f and 9 are holomorphic, then
(gofhv(z) - 0 (i.e., gof is holomorphic),
m
(g 0 f)zv (z) = L 9w,. (f(z» . (f,.)zv (z).
/-&=1

The second equation can be abbreviated as

V(g 0 f)(z) = Vg(f(z» . Jr(z).

Tangent Vectors. In this paragraph we use the term differentiable for

infinitely differentiable.
Definition. A tangent vector at a point z E en is a pair t = (z, w),
where the direction w of t is an arbitrary vector of en. If the base point
z is fixed, we simply write w instead of tor (z, w).
The set T. of all tangent vectors at z is called the tangent space (of en)
at z.
 7. Holomorphic Maps 33

The notation "tangent vector" is motivated by the following:

Let Been be an open set and I c lR an interval containing 0 as an interior
point. If a = (a1,"" an) : I -t B is a differentiable path, then a'(O) is the
direction of the tangent to the curve a at the point a(O). Therefore,

0(0) := (a(O), a'(O))

is called the tangent vector of a at z = a(O). Each tangent vector (z, w) E Tz

can be written in the form 0(0), e.g., a(t) := z + two
The tangent space Tz carries in a natural way the structure of a complex
vector space:

(z, wd + (z, W2) (z, WI + W2),

>. . (z, w) (z, >. . w), for>. E e.
Every tangent vector t = (z, w) operates linearly on the algebra C(B) of
differentiable functions on B by

t(JJ = Df(z)(w).
This is the directional derivative, also denoted by Dwf(z). If t = 0(0), for
some differentiable path a, then t[fJ = (J 0 a)'(O), due to the chain rule.
The operator t: 6"(B) -t lR satisfies the product rule:

t[f . gJ = t(JJ . g(z) + f(z) . t[gJ.

In general, a linear operator satisfying the product rule is called a derivation.
In Chapter IV we will show that the tangent space is isomorphic to the vector
space of derivations.

The Inverse Mapping. Let B1. B2 c en be open sets, and r : B1 -t B2

a holomorphic map.
Definition. The map r is called biholomorphic (or an invertible holo-
morphic map) if r is bijective and r- 1 holomorphic.

7.5 Inverse mapping theorem. Consider a point Zo E B1 and its image

Wo = f(zo). Then the following are equivalent:
1. There are open neighborhoods U = U(zo) C B1 and V = V(wo) C B2
such that r : U -t V is biholomorphic.
2. det Jr(zo) =I- O.
34 I. Holomorphic FUnctions

PROOF: If flu : U -+ V is biholomorphic, then (flu )-1 0 f = idu and

1 = det(En ) = det(J(rlu)-l(wo} . Jr(zo» = det(J(rlu)-l(wo» . det(Jr(zo»,
and therefore det(Jr(zo» i- o.
If det(Jr(zo» i- 0, then also det(Ja,r(zo) = Idet Jr(zo)12 i- O. It follows from
real analysis that there are open neighborhoods U = U(zo) C B1 and V =
V(wo) C B2 such that flu: U -+ V is bijective and g := (flu )-1 : V -+ U
a continuously differentiable map (in the real sense). Then fog = id v is a
holomorphic map, and if we write f = (/1, ... , / n) and g = (91, ... , gn), then
n
0= (fll 0 g)W,. = L«(fIl)ZA 0 g) . (9.x)w,., for v, f.L = 1, ... , n .
.x=1
In the language of matrices this means that

O=Jr' ( V~1 ).
V9n

Since Jr is invertible, it follows that V g.x = 0 for each A. Therefore, the map
g is holomorphic. _

7.6 Implicit function theorem. Let Been x em be an open set, f =

(/1, ... , /m) : B -+ em a holomorphic mapping, and (zo, wo) E B a point
with f(zo, wo) = 0 and

det (O/p. (zo, wo) I f.Lv=n+1,

OZII
= 1, ... , m
... ,n+m
) i- O.

Then there is an open neighborhood U = U' X U" c B and a holomorphic

map g : U' -+ U" such that

{(z, w) E u' X u" : fez, w) = O} = {(z,g(z») : z E U' }.

PROOF: We write Jr(zo, wo) = (J' I J"), with J' E Mm,n(C) and J" E
Mn(C), and define F: B -+ en x em by F(z, w) := (z, fez, w». Then

det JF(zo, wo) = det (~~ J,,) i- O.

Therefore, there are open neighborhoods U = U(zo, wo) c B and V =
V(zo,O) C en +m such that Flu: U -+ V is biholomorphic. Obviously,
F-1(u, v) = (u, h(u, v». We may assume that U = U' X U" c en x em
and V = U' x W, with some open neighborhood W = W(O) c em. Defining
g: U' -+ U" by g(z) := h(z,O), it follows that
 7. Holomorphic Maps 35

fez, w) = 0 {:==> F(z, w) = (z,O)

{:==> (z, w) = F-l(z, 0)
{:==> w = h(z, 0) = g(z).
This completes the proof. •
Remark. We can exchange the coordinates in the theorem. If rk Jr(zo, wo) =
m, then there are coordinates Zi" .•. , Zi n such that f-l(O) is the graph of a
map g = g(Zi 1 , • . . , Zi n ) near (zo, wo).

Exercises
i 1. Let G= pn C e2 be the unit polydisk and f = (/1, h) : G -4 G a
holomorphic map with £(0) = O.
(a) Show that iff(z) = z+ Ln>2 Pn(z) with pairs Pn(Z) = (pin>Cz),p~n)(z))
of homogeneous polynomi81s of degree n, then fez) == z. Hint: Use
Cauchy's inequalities and consider the iterated maps fk = f 0 ••. 0 f
(k times).
(b) Show that if f is biholomorphic, then /1, 12 are linear.
2. Let G l , G 2 c en be two domains. A continuous map f : G l -4 G 2 is
called proper if for every compact subset K c G 2 the preimage f-l(K)
is a compact subset of G l .
(a) Show that every biholomorphic map is proper. Give an example of a
proper holomorphic map that is not biholomorphic.
(b) Let G l and G 2 be bounded. Show that a continuous map f : G l -4 G 2
is proper if and only if for every sequence (Zk) in G 1 tending to aG 1 ,
the sequence (f(Zk)) tends to aG 2.
(c) Let G', G" c e be bounded domains and f : G' X G" -4 G 2 a
proper holomorphic map onto a bounded domain G 2 c e 2 . Show
that Z r-t fw (z, w) cannot vanish identically on G'. Let Zo E ae' be an
arbitrary point and (Zk) a sequence in G' tending to ZOo Show that the
sequence of holomorphic maps 'Pk : Gil -4 G 2 with 'Pk(W) := f(Zk, w)
has a subsequence converging compactly on Gil to a holomorphic
map 'Po : G" -+ ((:2 with 'Po(G") c aG 2. Show that there must exist
at least one point Zo E aG' such that the corresponding map 'Po is
not constant.
3. Use the results of the last exercise to prove that there is no proper map-
ping from the unit polydisk to the unit ball in e 2 .
4. Let G l C en and G2 c em be domains and f: Gl -4 G 2 a biholomorphic
map. Show that m = n.
5. Let G c en be a domain and D : C(G) -4 IR a derivation, Le., an IR-linear
map satisfying the product rule at Zo E G. Show that D[f) depends only
on flu, U an arbitrary small neighborhood of Z00
36 I. Holomorphic Functions

6. Let G c en be a domain, f = (ft, ... , fm) : G -+ em a holomorphic

mapping, and M := Hz, w) E G x em : w = fez)}. Prove the following:
If 9 : G x em -+ e is a holomorphic function with glM = 0, then for
every point (zo, wo) E M there is a neighborhood U and an m-tuple
(at, ... , am) of holomorphic functions in U such that
m

g(z, w) = L a,..(z, w)· (wI' - f,..(z» for (z, w) E U.

,..=1

8. Analytic Sets
Analytic Subsets. Let Been be an arbitrary region. If U c B is
an open subset, and ft, . .. ,fq are holomorphic functions on U, then their
common zero set is denoted by

N(ft, .. ·, fq) = {z E U : ft(z) = ... = fq(z) = OJ.

Definition. A subset A c B is called analytic if for every point Zo E B

there exists an open neighborhood U = U(zo) c B and holomorphic
functions ft, ... , fq on U such that UnA = N(ft,···, fq).

If Zo is a point of B-A, then we can choose an open neighborhood U = U(zo)

and holomorphic functions ft, ... ,fq on U such that

Zo E U' := U - N(ft, .. ·, fq) cUe B.

Since the zero set N (ft, ... , fq) is closed in U, it follows that B - A is open
and A closed in B. Therefore, an analytic set in B could have been defined as
a closed subset A C B such that for any Zo E A there exists a neighborhood
U and functions ft, ... , fq E O(U) with An U = N(ft,···, fq)·
Example
In general, analytic sets cannot be given by global equations. We consider
the domain G := G l U G 2 with

Gl .- {z = (Zl,Z2) E e2 : IZll < ~ and IZ21 < I},

1
G 2 .- {z = (Z},Z2) E e2 : IZll < 1 and 2' < IZ21 < I}.

For the analytic set we take A := {(Zl' Z2) E G2 : Zl = Z2} (see Figure 1.5).
The sets G}, G 2 give an open covering of G with An G 1 =0 and An G2 =
{(z}, Z2) : Zl - Z2 = oJ. So A is an analytic subset of G.
 8. Analytic Sets 37

IZ21
I - -......---~-:JI

1Ioo.._ _ _.......' - -_ _ _ -'---~

IZll

Figure 1.5. A not globally defined analytic set

If f is a holomorphic function in G that vanishes on A, then f can be analyt-

ically continued to the unit polydisk p2, since (P2, G) is a Euclidean Hartogs
figure (up to the order of the coordinates). Let jbe the continuation. Since
g{z) := j(z,z) vanishes for! < Izi < 1, it also vanishes for 0 ~ Izl ~ !.
This means that f vanishes on A = {( Zl , Z2) E G : Zl = Z2}. Any zero set
of finitely many holomorphic functions in G that vanish on A must contain
A. So A itself cannot be given by global holomorphic functions. In the next
chapter we define special domains in en
each of which possesses a holomor-
phic function that cannot be analytically extended to a larger domain. Those
domains are called domains of holomorphy. On .such domains the global rep-
resentation of analytic sets is possible. The proof of this fact is not contained
in this book, because it requires sheaf theory. One has to show that the
sheaf of germs of holomorphic functions that vanish on A is "coherent" (cf.
[GrRe84], Section 4.2). Then every stalk of this sheaf is generated by global
sections (Cartan's theorem A, cf. Chapter V in this book, and [GrRe79], Sec-
tion IV.5). From that it can be proved that A is the zero set of finitely many
global holomorphic functions.
Definition. A subset M of a domain G is called nowhere dense in G
if the closure of M in G has no interior points.

Since an analytic set A eGis always closed in G, it is nowhere dense if in

every neighborhood of every point z E G there are points outside of A.
8.1 Proposition. Assume that A is an analytic set in a domain G c en.
If A has an interior point, then A = G. If A is nowhere dense in G, then
G - A is connected.

PROOF: To start with we assume that G = B is a ball and that there are
holomorphic functions It, ... , f q on B with A = N (It, ... , fq).
38 I. Holomorphic Functions

If Zo E B is an interior point of A, we consider an arbitrary complex line L

through zoo By the identity theorem the functions fi all vanish on L n Band
therefore in B.
If A is nowhere dense in Band L an arbitrary complex line, then either
L nBc A or A has only isolated points on L n B. So .any two points of L n B
outside of A can be connected in L n (B - A).
Now let G be an arbitrary domain. If Zo EGis an interior point of A, and
Wo EGan arbitrary point, then we can join these points by a continuous
path Q: : [0,1] -+ G. The compact image ofthis path can be covered by finitely
many balls BeG such that B n A is the zero set of holomorphic functions
on B. Successively it follows that every ball is contained in A. So A = G.
If A is nowhere dense in G, then we consider zo, Wo E G - A and use the
same continuous path. It is clear from above that any point z in the first ball
B that is not an element of A can be joined in B - A to zoo Applying this
successively we obtain a curve between Zo and Wo in B - A. •
If n = 1, then a nowhere dense analytic set consists only of isolated points.

Bounded Holomorphic Functions. Assume that G c en is a do-

main and A eGa proper analytic subset.
8.2 Riemann extension theorem. If f is a holomorphic function in
G - A that is bounded in a neighborhood of every point of A, then f can be
holomorphically extended to G.

PROOF: Since A =f. G, A is nowhere dense in G. Let Zo E A be an arbitrary

point. Then there is a complex line L through Zo that in a neighborhood of
Zo intersects A only in Zo.
After a linear change of coordinates we may assume that Zo = 0 and that
L = Cel is the zl-axis. We can find a polydisk
P = {z = (Zl'Z') E e x Cn - l : IZll < rlllz'l < r} cc G
such that An {z : IZll = rl, Iz/l < r} is empty. For any c' E cn-I with
Ic'l < r, the set D = {z : IZII SrI and Zl = c / } is a I-dimensional disc such
that DnA contains only isolated points, since otherwise DcA (see Figure
1.6). By the classical Riemann extension theorem in one variable f can be
extended to a function j(ZI, Z/) that is holomorphic in Zl. By the classical
Cauchy integral formula we have

~
f(ZI,Z ') = -1.
2m
1
/(/=r,
( z')
f((,
- Zl
de, for IZI I < rl and Iz 'I < r.

The integrand on the right side is holomorphic on P. Consequently, the left

side is differentiable (in the real sense), and since integration and differenti-
ation by Zi can be exchanged, 1
is holomorphic on P. If we carry this out
 8. Analytic Sets 39

Iz'l

A
T

c' 1----+-------1 ~ D

.
Tl IZll
Figure 1.6. Riemann extension theorem

at every point 210 E A, by the identity theorem we obtain the desired global
extension of f to G. •

Regular Points. Let G c en be a domain, and z EGa point. If b, ... , fq

are holomorphic functions in a neighborhood of z, then we define

Definition. An analytic set A eGis called regular of codimension

q at z E A if there is a neighborhood U = U(z} c G and holomorphic
functions b, ... , fq on U such that:
1. An U = N(b, .. · ,fq}.
2. rkz(b,· .. , fq} = q.
The number n - q is called the dimension of A at z.
The set A is called singular at z if it is not regular at that point. The
set of regular points of A is denoted by Reg(A} or A, the set of singular
points by Sing(A}.

It is clear that A is open in A, and therefore Sing(A} C A closed.

8.3 Theorem (local parametrization of regular points). Let A C G
be analytic, Zo E A a point. A is regular of codimension q at Zo if and only
if there are open neighborhoods U = U(zo} c G and W = W(O} c en and a
biholomorphic map F : U -t W such that F(zo} = 0 and
F(U n A} = {w = (Wl, ... , w n ) E W : wn-q+l = ... = Wn = o}.

PROOF: Let A be regular at zoo There is an open neighborhood U = U(zo}

such that A n U = N (b, ... , fq) and rkzo (b, ... ,/q) = q. By renumbering
the coordinates we can achieve that
40 I. Holomorphic Functions

J (h , ... ,/.) (z 0 ) -- (J' I JII) ,

with J' E Mq,n-q(C), J" E Mq(C), and det J" =1= O. Then define F : U -+ en
by

F(Zl, ... , zn) := (Zl -ziO), ... , Zn_q -z~o!..q, h(Zl,"" zn),"" fq(zl,"" zn»).
Consequently, the Jacobian has the form

and therefore det J F =1= O. Shrinking U if necessary, we have our biholomorphic

map F : U -+ W, with F(zo) = 0 and

w = F(z) for some z E UnA -<==> wn-q+l = ... = Wn = O.

The other direction of the proof is trivial. II

Up to this point it is not clear whether or not there exist regular points. In
Chapter III we will show that the set of singular points of an analytic set
A is a nowhere dense analytic subset of A. At the moment we want only to
demonstrate that the zero set of a single holomorphic function contains at
least one regular point (and then, of course, a nonempty open set of regular
points).
8.4 Proposition. Let G c en be a domain, and f a nonconstant holomor-
phic function on G. Then the analytic set N(J) contains a regular point.

PROOF: The case n = 1 is trivial. Therefore, we assume n > 1.

If every point of A := N(J) is singular, then 'V fez) == 0 on A. Since f is not
constant, it is impossible that there is a point z such that DV fez) = 0 for
every multi-index v. Therefore, we can find a point Zo E A, an integer no, a
multi-index vo, and some>. E {I, ... ,n} such that
1. Ivol = no and (Dvo fk (zo) =1= 0,
2. D V fez) = 0 for every z E A and every II with 1111 ~ no.
The set M := {z E G : DVo fez), = O} is analytic in G and regular of
codimension 1 at zoo We may assume that Zo = 0 and M = {z = (Zl' z') E
G : Zl = O}, making G sufficiently small.
We have A eM, and we want to show equality near Zo. It is clear that the
function ( t-+ f«(,O') has exactly one zero at ( = 0, and it follows easily
from Rouche's theorem that for z' sufficiently close to 0' the functions ( t-+
f«(, z') also have exactly one zero. This means that there is a neighborhood
V = V (0) c U such that V n A = V n M. In particular, Zo is a regular point
of A. •
 8. Analytic Sets 41

Definition. A k-dimensional complex submanifold of a domain G C

en is an analytic set A c G such that A is regular of codimension n - k
at every point.

If A eGis a k-dimensional complex submanifold, then for every point Z E A

there is an open neighborhood U = U(z) c G, an open set week, and a
holomorphic map <p: W -t U such that:
1. rkJ<p(w) = k for W E W.
2. <p(W) = UnA.
3. <p: W -t UnA is a topological map.4
The proof follows immediately from the local parametrization theorem. The
map <p is called a local parametrization.

Injective Holomorphic Mappings. Let G c en be a domain, and

f = (ft, ... , fn) : G -t en a holomorphic map.
8.5 Theorem. If f is injective, then det Jr(z) =F 0 everywhere.

PROOF: We use induction on n. The case n = 1 is well known. We consider

the case n > 1 and define h := det Jr.
Assume that N(h) =F 121. Then there exists an open subset U C G such that
M := Un N(h) is a nonempty (n - I)-dimensional complex submanifold of
U.
We claim that Jr\M == O. To prove this, we assume that there is a point
Zo E M with Jr(zo) =F O. Without loss of generality, we may assume that

{}fn ( )
-{} Zo =F
Zn
o.
Let F : G -t en be defined by F(z', zn) := (z', fn{z', zn)). Then detJF(zo) =F
0, and there are connected open neighborhoods U of Zo and V ofwo := F(zo)
such that F : U -t V is biholomorphic. There is a holomorphic map f : V -t
en - 1 such that
and we define

g = (91, ... ,9n-d : W:= {W' E en - 1 (w',w~») E V} -t en - 1

by g(w') := f(w',w~»).
4 A map tp : X ~ Y between topological spaces is called topological or a homeo-
morphism if it is continuous and bijective and the inverse mapping tp -1 : Y ~ X
is also continuous.
42 I. Holomorphic Functions

Since g is injective, we can apply the induction hypothesis and conclude that
det Jg(wh) :f; O. Now

Jf(zo) . JF-l (wo) = ( Jf~:O) ) = (Jg~~o) ~).

Therefore, h(zo) = det JF(ZO) ·det Jg(wo) :f; 0 as well. This is a contradiction.
We have demonstrated that Jf(Z) = 0 for every Z E M. Since f is holomor-
phic, also Df(z) == 0 on M, and using a local parametrization of M we obtain
that flM is locally constant. But this is impossible, since f is injective. The
set N (h) must be empty. _

8.6 Corollary. If G c en is a domain, and f : G -t en an injective

holomorphic mapping, then also f( G) is a domain, and f : G -t f( G) is
biholomorphic.

PROOF: Let Wo := f(zo) be a point of G' := f(G). Then det Jf(ZO) :f; 0,
and there are open neighborhoods U = U(zo) c G and V = V(wo) c en
such that f : U -t V is biholomorphic. It follows that Wo is an interior point
of G' and that f- I is holomorphic at Wo0 _

Exercises
1. Prove the following properties:
(a) Finite intersections and unions of analytic sets are analytic.
(b) If f: G I -t G 2 is a holomorphic map between domains and A C G 2
an analytic set, then f-I(A) c G I is analytic as well.
(c) If Al C G I and A2 C G 2 are analytic sets, then Al XA2 is an analytic
subset of G l x G 2 •
2. Let U C c;n be an open neighborhood of the origin and A cUbe
an analytic subset containing the origin. For 1 ::; k ::; n - 1 and I =
{i l , ... , ik} C {I, ... , n} let PI : en -t e k be defined by

Prove: If A is regular of codimension n - k at the origin, then there exists

an I and open neighborhoods V = V(O) C U, W = W(O) C e k such
that PI : A n V -t W is bijective.
3. Show that A := {(w, Zl, Z2) E e 3 : w 2 = ZIZ2} is an analytic set that is
regular of codimension 1 outside the origin and singular at O.
4. Let AI, A2 be two analytic sets in a neighborhood of the origin in en
such that 0 E A := Al n A 2. Suppose that U n Al :f; U n A2 for every
neighborhood U of o. Show that A is singular at o.
Chapter II

Domains of Holomorphy

1. The Continuity Theorem

General Hartogs Figures. The subject of this chapter is the contin-
uation of holomorphic functions. We consider domains in en, for n ~ 2. A
typical example is the Euclidean Hartogs figure (pn, H), where pn = pn(o, 1)
is the unit polydisk, and

H={ZEpn: IZ 11>Q1 0r lzlIl<Qllforv=2, ... ,n}.

Here Q1, .•. , qn are real numbers with 0 < Qll < 1 for v = 1, ... , n. Every
i
holomorphic function f on H has a holomorphic extension on pn.
Definition. Let g = (91, ... , 9n) : pn -t en be an injective holo-
morphic mapping, P := g(pn) and ii := g(H). Then (p, ii) is called a
general Hartogs figure.

We use the symbolic picture that appears as Figure II.l

Z2, ..• , Zn

Figure ILL General Hartogs figure

1.1 Continuity theorem. Let G c en be domain, (P, ii) a general Har-

togs figure with ii c G, f a holomorphic function on G. If GnP is connected,
then f can be continued uniquely to G U j5.
44 II. Domains of Holomorphy

PROOF: Let g : pn -t en be an injective holomorphic mapping such that

P := g(pn) and ii := g(H). The function h := log is holomorphic in
H. Therefore, there exists exactly one holomorphic function Ii on pn with
hlH = h. Since g : pn -t P is biholomorphic, the function 10 := Ii 0 g-l is
defined on P, and it is a holomorphic extension of IIR- We define

j(z) := {/(Z) for Z E q,

lo(z) for z E P.

Since Gnp is connected and I = 10 on ii, it follows from the identity

1
theorem that is a well-defined holomorphic function on G UP. This is the
desired extension of I. •

Example
Let n ~ 2 and pI C C P be polydiscs around the origin in en. Then every
holomorphic function f on P - pI can be extended uniquely to a holomorphic
function on P.
For a proof we may assume that P = pn is the unit polydisk, and pI =
pn(O,r), with r = (TI! ... ,Tn) and 0 < Tv < 1 for v = 1, ... ,n. It is clear
that G := P - pI is a domain.
Given a point Zo = (z~O), ... , z~O») E G with Iz~)1 > Tn, we choose real num-
bers ql, ... , qn as follows: For v = 1, ... , n - 1, let qv be arbitrary numbers,
with Tv < qv < 1. To obtain a suitable qn, we define an automorphism T of
the unit disk 0 by
(_Z(O)
T«() := z~)( ~ 1 .
This automorphism maps z~O) onto 0 and a small disk D c {( E e : Tn <
1(1 < I} around z~) onto a disk KeD with 0 E K. Notice that 0 need not
be the center of K. We choose qn > 0 such that Dqn (0) C K.
If we define H := {z E pn : IZll > ql or Izvl < qv for v = 2, ... ,n}, then
(pn, H) is a Euclidean Hartogs figure. The mapping g : pn -t pn defined by

is biholomorphic, and (p,ii) = (pn,g(H» is a general Hartogs figure, with

ii C {z E pn : IZII > Tl or IZnl > Tn} C G.

Since Pn G = G is connected, the continuity theorem may be applied. The
preceding example is a special case of the so-called Kugelsatz which we shall
prove in Chapter VI.
 1. The Continuity Theorem 45

ZIO)
n

Figure 11.2. A Hartogs figure for concentric polydises

Removable Singularities. Let G c en be a domain. If A eGis an

analytic set and I a holomorphic function on G - A that is locally bounded
along A, then by Riemann's extension theorem I has a holomorphic extension
to G. If n 2:: 2 and A is a complex linear subspace of codimension greater
than or equal to 2, then every function holomorphic on G - A has such an
extension.
1.2 Theorem. Let pn = pn(o, 1) be the unit polydisk in en, n 2:: 2, k 2:: 2,
and
E := {z = (z}, ... , zn) E en : Zn-k+l = ... = Zn = O}.
Then every holomorphic lunction I on pn - E can be holomorphically ex-
tended to pn.

PROOF: Set P' := {z' := (Zl' ... ' Zn-k) : Iz'l < I}, and for 0 < r :s 1
define P;':= {z" = (Zn-k+l, ... ,Zn) : Iz"l < r}.
Let p" := P{' and fix an c with 0 < c « 1. Then pn nEe P' X P~', and for
W E P' the function Iw(z") := I(w, z") is holomorphic on P" - p:,. From
the example above we know that I w has a holomorphic extension fw to P".
Now define f: pn -+ e by i(w, i
z") := iw(z"). On pn - E, is equal to I
and is therefore holomorphic.
For wE P' take a small open neighborhood U = U(w) cc P'. Then K :=
U x OP:' is compact. By the maximum principle we conclude that

Ifez', z") I = Ih' (z") I :s Il/z" Ilap:, :s II/IIK < 00, for (z', z") E U X P;' - E.

From Riemann's extension theorem it follows that i is holomorphic on pn .•

46 II. Domains of Holomorphy

1.3 Corollary. For n 2 2, every isolated singularity of a holomorphic func-

tion of Zl, ... ,Zn is removable.
Riemann's extension theorem is false if we drop the condition "f bounded
along the analytic set." For example, let G c en be a domain, 9 : G -+ e a
holomorphic function, and let f : G-N(g) -+ e be defined by fez) := l/g(z).
Then f is holomorphic on G - N (g) but cannot be extended to any point of
N(g).
Things look quite different if there is a little hole in the hypersurface:
1.4 Proposition. Let n 22, Go C C n - 1 a domain, g: Go --+ C a contin-
uous function, and r:= {z = (z',zn) E Go xC: Zn = g(z')} the graph of
g in G := Go x e. In addition, let Zo be a point of rand U = U(zo) eGa
small neighborhood.
If f is a holomorphicfunction on (G-r)UU, then f has a unique holomorphic
extension to G.

PROOF: The uniqueness of the extension follows from the identity theorem.
For the proof of existence (which is only a local problem) we may assume
that Co = {z' E en-I: Iz'l < I} and that there is a q with 0 < q < 1 such
that Ig(z')1 < q for z' E Co. It also may be assumed that U is connected.
Then it is clear that G' := (G - r) U U C pn = pn(o, 1) is connected.
Since g : z' t-+ (z', g(z')) is continuous, U' := g-l (U) is an open neighborhood
of z~ with (U' x D) nrC U and therefore U' x 0 C C/. For v = 1, ... , n - 1
let Tv be the automorphism of 0 defined by
<: _ z(O)

Tv(() := z~O)<: .::. 1 .

Then h: pn --+ pn with h(Zl,,,,,Zn):= (T1(Z2), ... ,Tn- 1(zn),zd is holo-
morphic, h(O) = (z~,O), and h({z E pn : IZ11 > q}) C {w E pn : Iwnl > q}.

Iznl r--,-""""",,-::::.,-....,....-=_
q H

r'
U
'----------_ _-+-- L __
z~ Iz'l
Figure 11.2. Extending a holomorphic function across a hypersurface
 1. The Continuity Theorem 47

We define ql := q, and for v = 2, ... , n choose qll such that

b(D x Dq2 (O) x ... x Dqn (0» C u' x D.
Then (pn,H) with H:= {z E pn : IZll > ql or IZIII < qll for v = 2, ... ,n}
is a Euclidean Hartogs figure, and (P, ii) = (pn, b(H» is a general Hartogs
figure, with ii c G' (see Figure II.3). Since P n G' = G' is connected, the
proposition follows from the continuity theorem. _

The Continuity Principle. Sometimes we wish to use a family of

analytic disks instead of a Hartogs figure.
Definition. A family of analytic disks is given by a continuous map
IP : 5 x [0,1] --+ en such that IPt«() := rp«(, t) is holomorphic in D,
for every t E [0,1]. The set St := IPt(D) is called an analytic disk, and
bSt := IPt(8D) its boundary.

Observe that in general bSt is not the topological boundary of St.

Definition. A !1omain G C en is said to satisfy the continuity prin-
ciple if for any family {St, t E [0, 1]} of analytic disks in en with
U 099 bSt C G and So C G, it follows that U099 St C G.

Example
Let pn be the unit polydisk and {St, t E [0, I]} a family of analytic disks
in en with UO<t<l bSt C pn and So C pn. Because So and the union of all
boundaries bSt- are compact sets, there is an c: > 0 such that
U bSt C pn(O, 1 - c) and So c pn(O, 1 - c:).
0::;t9
We assume that UO::;t::;l St is not contained in pn, and define
to := inf{t E [0,1] : St ¢. pn}.
It is clear that to > 0, Sto ¢. pn, and St C pn for 0 :5 t < to. Then Sto
contains a point zo = (zlO), ... , z~») E 8pn. If the family of analytic disks is
given by the map IP : 5 x [0,1] --+ en, and wI' denotes the J.'th coordinate
function, then fl',t«() := wI' 0 IP«(, t) is continuous on 5 and holomorphic in
D. Choosing J.' such that Iz~O)1 = 1, there is a (0 E D with fl',to«(o) = zt» and
Ifl',to«(o)1 = 1. But by the maximum principle we have
Ifl',t«(o)1 :5 suplfl',tl :5 1 - c, for t < to.
aD
Since t t-t il',t«(o) is continuous, a contradiction is reaclled, and therefore pn
satisfies the continuity principle.
48 II. Domains of Holomorphy

Hartogs Convexity.
Definition. A domain G c en
is called Hartogs convex if the following
holds: If (p, H) is a general Hartogs figure with He G, then PeG.

An immediate consequence of the definition is the following:

The biholomorphic image of a Hartogs convex domain is again Hartogs
convex.

1.5 Theorem. Let G c en be a domain that satisfies the continuity prin-

ciple. Then G is Hartogs convex.

PROOF: Let (p, H) be a general Hartogs figure with H C G. We assume

that it is the biholomorphic image (g(pn),g(H» of a Euclidean Hartogs figure
(pn, H) with

H = {z : IZll > ql or Iz,.1 < qjJ for IJ. = 2, ... ,n}.

In order to define analytic disks we choose some r with ql < r < 1 and
introduce the affine analytic disks

Dw := {z = (Zl,Z") E pn = pI X p" : IZll < r and z" = w}.

Since Dw C pn for every W E P", we can define CPw : 0 x [0, 11 --t en by
setting CPw«, t) := g(r(, tw). Then a family {Stew) : 0 ~ t ~ I} of analytic
disks in P is given by

St(w):= CPw(D x {t}) = g(Dtw ).

It follows that bSt(w) C G for every w E P" and every t E [0,11, and in
addition, So(w) = g(Do) c G. The situation is illustrated in Figure 11.4.

z"

w
St{w} bSt{w}

...
Izd
~

Figure 11.4. Analytic disks in a Hartogs figure

 1. The Continuity Theorem 49

Since G satisfies the continuity principle, we obtain that g(Dw) = SleW) is

contained in G. This is valid for every W E Pll. Therefore, PeG, and G is
Hartogs convex. _

1.6 Corollary. The unit polydisk pn is Hartogs convex.

Domains of Holomorphy
Definition. Let G c en be a domain, J holomorphic in G, and Zo E
BG a point. The function J is called completely singular at Zo if for
every connected neighborhood U = U(zo) c en and every connected
component C of Un G there is no holomorphic function g on U for which
glc = Jlc.

Example
Let G ;= e- {x E R ; x ::; O} and let J be a branch of the logarithm on
G. Then J is completely singular at z = 0 but not at any point x E R with
x < O.
Definition. A domain G c en is called a weak domain oj holomorphy
if for every point Z E BG there is a function J E O( G) that is completely
singular at z.
The domain G is called a domain oj holomorphy if there is a function
J E O(G) that is completely singular at every point z E BG.

Examples
1. Since en has no boundary point, it trivially satisfies the requirements of
a domain of holomorphy.
2. It is easy to see that every domain Gee is a weak domain of holomor-
phy; If Zo is a point in BG, then J(z) ;= l/(z - zo) is holomorphic in G
and completely singular at Z00
For G = D we can show even more! The function J(z) ;= L~o zu! is
holomorphic in the unit disk and becomes completely singular at any
boundary point. Therefore, D is a domain of holomorphy. At the end of
this chapter we will see that every domain in e is a domain of holomorphy.
3. If J ; D --+ e is a holomorphic function that becomes completely singular
at every boundary point, then the same is true for f; pn = D x ... x D --+
e, defined by f(Zl, ... , zn) ;= J(zd+···+ J(zn). In fact, ifzo is a bound-
ary point of pn, then there exists an i such that the ith component z1°) is
f
a boundary point of D. If could be extended holomorphically across Zo,
50 II. Domains of Holomorphy

then h«() := j(ziO), ... , (, ... , z~») would also have a holomorphic ex-
tension. But then J could not be completely singular at z~O). Therefore,
the unit polydisk is a domain of holomorphy.
4. If (pn, H) is a Euclidean Hartogs figure, then H is not a domain of holo-
morphy.

1. 7 Proposition. Let e c en be a domain. IJ Jor every point Zo E 8e

there is an open neighborhood U = U (zo) c en and a holomorphic Junction
J : e u U ~ e with J(zo) = 0 and J(z) i= 0 Jor z E e, then e is a weak
domain oj holomorphy.

PROOF: We show that 1/ J is completely singular at zo0 For this assume that
there is a connected open neighborhood V = V(zo), a connected component
C c V n e, and a holomorphic function F on V with Fie = (1/1) Ie'
The set V' := V - N(f) is still connected and contains C. By the identity
theorem the functions F and 1/ J must coincide in V'. Then F is clearly not
holomorphic at Zo. This is a contradiction. _

1.8 Corollary. Every convex domain in en is a weak domain oj holomor-

phy.

PROOF: If Zo E 8e, then because of the convexity there is a real linear

form>. on en with >.(z) < >,(zo) for z E e. We can write>. in the form
n n
>.(z) = I: G:"Z" + I: QI/ZI/, with a := (G:I, ... , Cl: n ) i= O.
,,=1 1/=1

So ,\ = Re h(z), where h(z) := 2 . L:~=1 G:"Z" is holomorphic on en.

Since the function J(z) := h(z) - h(zo) is holomorphic on en, J(zo) = 0, and
J(z) i= 0 on e, the proposition may be applied. -
We will show that every weak domain of holomorphy is Hartogs convex. As
a tool we need the following simple geometric lemma, which will be useful in
other situations as well.
1.9 Lemma (on boundary components). Let e c en be a domain,
U c en an open set with U n e i= 0 and (en - U) n e i= 0.
Then en 8C n 8U i= 0 Jor any connected component C of U n e.
PROOF: We choose points Z1 E C C Un e and Z2 E (en - U) n e. There
is a continuous path "I : [0, 1] ~ e with "1(0) = Z1 and "1(1) = Z2' Let
to := sup{t E [0,1] : "I(t) E C} and Zo := "I(to). Clearly, Zo E 8C n e,
but Zo f/. C. Since C is a connected component of Un G, Zo cannot lie in
Un e and therefore even not in U. Since "I(t) E U for t < to, it follows that
Zo E 8U. -
 1. The Continuity Theorem 51

1.10 Theorem. Let G c C n be a weak domain of holomorphy. Then G is

Hartogs convex.

PROOF: Assume that G is not Hartogs convex. Then there is a general

Hartogs figure (P, H) with H c G but P n G f P. We choose an arbitrary
Zo in H and set C := Cpna(zO).l Since H lies in P n G and is connected, it
follows that He C. FUrthermore, C ~ P.
Since P n G f 0 and (C n - G) n P f 0, by the lemma there is a point
Zl E BC n BG n P (see Figure II.5).

Figure II.5. G is not Hartogs convex

Let f be an arbitrary holomorphic function in G. Then fie is also holomor-

phic, and by the continuity theorem it has a holomorphic extension F on P.
Since P is an open connected neighborhood of Zl, we obtain that f is not
completely singular at Zl' This completes the proof by contradiction. _
It follows, for example, that every convex domain is Hartogs convex. As a
consequence, we see that every ball is Hartogs convex.
1.11 Theorem. Every domain of holomorphy is Hartogs convex.
The proof is trivial.
For the converse of this theorem one has to construct on any Hartogs convex
domain a global holomorphic function that becomes completely singular at
every boundary point, something that is rather difficult. It was done in 1910
by E.E. Levi in very special cases. The general case is called Levi's problem.
In 1942 K. Oka gave a proof for n = 2. At the beginning of the 19506 Oka,
Bremermann, and Norguet solved Levi's problem for arbitrary n. It was gen-
1 We denote by GM(Z) the connected component of M containing z.
52 II. Domains of Holomorphy

eralized for complex manifolds (H. Grauert, 1958) and complex spaces (R.
Narasimhan, 1962). Finally, in 1965 L. Hormander published a proof that
used Hilbert space methods and partial differential equations.

Exercises
1. Prove the following statements:
(a) Finite intersections of Hartogs convex domains are Hartogs convex.
(b) If G 1 C G 2 C G 3 C ... is an ascending chain of Hartogs convex
domains, then the union of all G; is also Hartogs convex.
2. Let G c en be a domain, 0 S r < R, and a EGa point. Let U = U(a) C
G be an open neighborhood and define Q := {w E em : r < Iwl < R}.
Prove that every holomorphic function on (G x Q) U (U x pm(o, R)) has
a unique holomorphic extension to G x pm(o, R).
3. Let 0 < r < R be given. Use Hartogs figures to prove that every holo-
morphic function on BR(O) - Br(O) has a unique holomorphic extension
to the whole ball BR(O).
4. For c ~ 0, consider the domain

Prove that G c is Hartogs convex if and only if c = O.

5. Let G c en be a domain and f : G -+ DR(O) c e a function, r =
{(z, w) E G x DR(O) : w = fez)} its graph. Sow that if there is a
holomorphic function F in G x D R(O) that is completely singular at every
point of r, then f is continuous. (With more effort one can show that f
is holomorphic.)
6. Show that the "Hartogs triangle" {(z, w) E e 2 : Iwl < Izl < I} is a weak
domain of holomorphy.

2. Plurisubharmonic Functions
Subharmonic Functions. Recall some facts from complex analysis of
one variable. A twice differentiable real-valued function h on a domain Gee
is called harmonic if hzz(z) == 0 on G. The real part of a holomorphic function
is always harmonic, and on an open disk every harmonic function is the real
part of some holomorphic function.
If D = Dr(a) C e is an open disk and (J : IR -+ IR a continuous periodic
function with period 21T, then there is a continuous function h : D -+ IR that
is harmonic on D such that h(re it ) = (J(t) for every t (Dirichlet's principle).
An upper semicontinuous function cp : G -+ IR U { -oo} is said to satisfy the
weak mean value property if the following holds:
For every a E G there is an r > 0 with Dr(a) cc G and
 2. Plurisubharmonic Functions 53

for 0 < (} :s; r.

Remarks
1. If <p : G -+ lR U { -oo} is an upper semicontinuous function, then the sets
UI/ := {z E G : <p(z) < v} are open, and therefore <p is bounded from
above on every compact subset KeG. It follows that the integral in the
definition always exists.
2. Harmonic functions satisfy the weak mean value property (even the
stronger mean value property with "=" instead of ":S;").
3. If f : G -+ IC is a nowhere identically vanishing holomorphic function,
then loglfl satisfies the weak mean value property. In fact, the function
<p := 10gIJI is harmonic on G - N(J), because it can be written locally
as Re(log 1), with a suitable branch of the logarithm. And at any point
Zo E N(J) we have <p(zo) = -00, so the inequality of the weak mean
value property is satisfied.

2.1 Proposition. Let <p : G -+ lR satisfy the weak mean value property. If
<p has a global maximum in G, then <p is constant.

PROOF: Let a E G be any point with c := <pea) > <p(z) for z E G. We

choose an r > 0 such that

Dr(a) cc G and <pea) :s; 2~ 121r <p(a + (}e it ) dt for 0 < (}:s; r.
Assume that there is abE Dr(a) with <pCb) < <pea). We write b = a + (}e ito
and get

1 {21r 1 (21r
<pea) :s; 27f Jo <p(a + (}e it ) dt < 27f Jo <pea) dt = <pea).

This is a contradiction, so <p must be constant on Dr(a). Now we define the

set M := {z E G : <p(z) = c}. Obviously, M is closed in G and not empty,
and we just showed that M is open. So M = G, and <p is constant. _

Definition. Let G C IC be a domain. A function s : G -+ lR U { -oo}

is called subharmonic if the following hold:
1. s is upper semicontinuous on G.
2. If Dec G is a disk, h: D -+ lR continuous, hiD harmonic, and h ~ s
on {)D, then h ~ s on D.

2.2 Proposition. Let SI/ : G -+ lR U { -oo} be a monotonically decreasing

sequence of subharmonic functions. Then s := lim v -+ oo Sv is subharmonic.
54 II. Domains of Holomorphy

PROOF: The limit s = lim v-4oo Sv = inf{sv} is upper semicontinuous. Let

Dec G be a disk, h : D 4 IR continuous and harmonic on D, with s :S h
on aD. For fixed c we consider the compact sets

Kv := {z E aD : sv(z) ;::: h(z) + c}.

Then Kv+l C K" and n~=l K" = 0. Therefore, there is a Vo E N with
K" = 0 for v 2: Vo. This means that for v 2: 110, S" < h + e on aD, and
therefore the same is true on D. Since the Sv are decreasing, s < h + c on D.
This holds for every e > 0, and consequently s :S h on D. _

2.3 Proposition. Let (So)oEA be a family of subharmonic functions on

G. If s := sup So ist upper semicontinuous and finite everywhere, then s is
subharmonic.

PROOF: If s :S h on aD, where D cc G and h : D 4 IR is continuous

and harmonic on D, then 8 0 :S h on aD for every Q E A. Since the 8 0 are
subharmonic, it follows that 8 0 :S h on D for every Q E A. But then 8 :S h
on D as well. _

Examples
1. Clearly, every harmonic function is subharmonic.
2. Let s : G 4 IR be a continuous subharmonic function such that -s is also
subharmonic. Then 8 is harmonic. To show this, we look at an arbitrary
point a E G and choose an r > 0 such that D := Dr(a) cc G. Then there
is a continuous function h : D 4 IR with hlaD = 81aD that is harmonic
on D (Dirichlet's principle). It follows that s :S h on D. But because -h
is also harmonic, we have -s :S -h on D as well. Together this gives
s = h on D.
3. Let J : G 4 IC be a holomorphic function. Then s := loglJI is subhar-
monic. In fact, if J(z) == 0 on G, then we have 8(Z) == -00, and there is
nothing to prove. Otherwise, s is harmonic on G - N(J), and we have
only to look at an isolated zero a of f. We choose D = Dr(a) cc G and
a function h that is continuous on D and harmonic on D, with s :S h
on aD. We know that s, and therefore also 8 - h, has the weak mean
value property on D, and it is certainly not constant. So it must take its
maximum on the boundary aD. This means that s :S han D.
4. Let G C IC be an arbitrary domain. The boundary distance 6a : G 4
1R+ U { +oo} is defined by

oa(z) := sup{r E IR : Dr(z) C G}.

Claim: s := -log 6a is subharmonic on G.

PROOF: If G = IC, then s(z) == -00 and there is nothing to prove.

If G =1= IC, then s is real-valued and continuous. For W E BG we define
 2. Plurisubharmonic Functions 55

Sw : G --* lR. by setting sw(z) := -loglz - wi. Then s(z) = sup{sw(z) :

w E aG}. By Proposition 2.3 the claim follows. •

The Maximum Principle

2.4 Theorem. Let s : G --* lR. U { -oo} be a subharmonic function on a
domain G C C. If s takes its maximum on G, then it must be constant.

PROOF: Assume that c := sea) ~ s{z) for every z E G. As in the case of

functions that have the weak mean value property it suffices to show that s
is constant in a neighborhood of a. If this is not the case, there is a small disk
D = OrCa) cc G and bEaD with sea) > s{b). Since s is upper semicontin-
uous, there is a continuous function h on aD with s :5 h:5 c and h{b) < c.
Solving Dirichlet's problem we can construct a harmonic continuation of h
on D. Now
h{a) = - 1 1
211" 0
2
" h{a + re it ) dt < c = sea).

This is a contradiction. •
For later use we give the following criterion for a function to be subharmonic:

2.5 Theorem. Let G C C be a domain and s : G --* lR. U { -oo} an upper

semicontinuous function. Suppose that for every disk Dec G and every
function f E OeD) with s < Re(f) on aD it follows that s < Re(f) on D.
Then s is subharmonic.

PROOF: Let D = OrCa) cc G, h : D --* IR continuous and harmonic on D,

and s :5 h on aD. For simplicity we assume a = O.
For v E N, a harmonic function hv on Dv := O(v/(v-l))r(O) :,) D is given by

Then (hv) converges on D uniformly, increasing monotonically to h. Further-

more, for every v there is a holomorphic function fv on Dv with Re(fv) = h v '
Let c > 0 be given. Then there is a Vo such that Ih - hvl < con D for v ~ Vo.
Therefore, s < hv +c = Re(fv +c) on aD for v ~ Vo. By definition it follows
that s < h", + con D. Since (hv) is increasing, it follows that s < h + c and
therefore s :5 h on D. •

Differentiable Subharmonic Functions

2.6 Lemma. Let s : G --* lR. be a «j'2 function such that Szz > 0 on G. Then
s is subharmonic.
56 II. Domains of Holomorphy

PROOF: Let D = Dr(a) cc G and let a continuous function h: D -+ IR be

given such that h is harmonic on D and s $ h on aD. We define !p := s - h.
Assume that cp takes its maximum at some interior point Zo of D. Then we
look at the Taylor expansion of!p at Zo in a small neighborhood about Zo:

!p(ZO + z) = cp(zo) + 2 ReQ(z) + !Pzz(zo)zz + R(z),

where Q(z) := CPz(zo)z + ~!Pzz(ZO)Z2 is holomorphic and R(z)jlzI2 -+ 0 for
Z -+ O. The function t/J(z) := 2ReQ(z) is harmonic, with t/J(O) = O. Since
it cannot assume a maximum or a minimum, it must have zeros arbitrarily
close to but not equal to O. On the other hand, !p(zo + z) - !p(zo) $ 0 and
CPzz(zo)zz > 0 outside z = O. This is a contradiction. Thus r.p must assume
its maximum on the boundary of D, and s $ h on D. _

2.7 Theorem. Let s : G -+ IR be a ~ function. Then s is subharmonic if

and only if Szz ::::: 0 on G.

PROOF: (a) Let szz(z) ::::: 0 for every z E G. Then we define Sv on G by

setting Sv := s + (ljv)zz. Obviously, (sv)zz = Szz + (l/v) > O. Then Sv
is subharmonic by the above lemma. Since (sv) converges, monotonically
decreasing, to s, it follows that 8 is subharmonic.
(b) Let s be subharmonic on G. We assume that szz(a) < 0 for some a E G.
Then there is a connected open neighborhood U = U(a) c G such that
8 zz < 0 on U. By the lemma it follows that -8 is subharmonic on U. Then
8 must be harmonic on U. So 8 zz (a) = 0, contrary to assumption. _

Plurisubharmonic Functions. We return to the study of domains in

arbitrary dimensions. Let G c en be a domain and (a, w) a tangent vector
at a E G. We use the holomorphic mapping O!a,w : e -+ en defined by
O!a,w«) := a + (w.
Definition. Let G c en be a domain. An upper semicontinuous func-
tion p : G -+ lR U { -oo} is called plurisubharmonic on G if for every
tangent vector (a, w) in G the function

Pa,w«) := p 0 O!a,w«) = p(a + (w)

is subharmonic on the connected component G(a, w) of the set O!;,!...(G) c

C containing O.

Remarks
1. Plurisubharmonicity is a local property.
 2. Plurisubharmonic Functions 57

2. If 1 E O(G), then logl/l is plurisubharmonic.

3. If Pl, 112 are plurisubharmonic, then so is Pl + P2.
4. If P is plurisubharmonic and c > 0, then c . P is plurisubharmonic.
5. If (Pv) is a monotonically decreasing sequence of plurisubharmonic func-
tions, then P := lim v --+ oo Pv is also plurisubharmonic.
6. Let (PQ)QEA be a family of plurisubharmonic functions. If P := sup(pQ)
is upper semicontinuous and finite, then it is also plurisubharmonic.
7. If a plurisubharmonic function p takes its maximum at a point of the
domain G, then P is constant on G.

The Levi Form

Definition. Let U c en
be an open set, 1E ~2(U; JR.), and a E U.
The quadratic form 2 Lev(f): Ta ~ JR. with

Lev (f) (a, w) := L Iz,;%,. (a)wvw~

v,~

is called the Levi form of 1 at a.

Obviously, Lev(f) is linear in I.

Examples
1. In the case n = 1 we have Lev(s)(a,w) = szz{a)ww. So s is subharmonic
if and only if Lev(s)(a, w) ~ 0 for every a E G and wEe.
2. Let I(z) := IIzll2 = E~=l ZiZi. Then Lev(f)(a, w) = IIwll 2 for every a.
3. If 1 E ~2(U; JR.) and e : JR. ~ JR. is twice continuously differentiable, then

Lev({? 0 I)(a, w) = e"(f(a» '1(8I)a(wW + l(f(a» . Lev (f) (a, w).

4. If F : U --t V c em is a holomorphic map and 9 E ~(V; JR.), then

Lev(g 0 F)(a, w) = Lev(g) (F(a), F'(a)(w».

5. For 1 E ~2(U; JR) the Taylor expansion at a E U gives

I(z) = I(a) + 2 Re(Qf(z - a» + Lev(f)(a, z - a) + R(z - a),

where Qf(w) = E~=l /z., (a)wv +! Ev,~ Iz"z" (a)wvw~ is a holomorphic

quadratic polynomial, and

lim R(z - a) = O.
z--+a liz - all 2
2 If H : TxT ~ C is a Hermitian form on a complex vextor space, then the
associated q'IJ.admtic form Q: V ~ R is given by Q(v) := H(v, v).
58 II. Domains of Holomorphy

2.8 Theorem. A function I E lf2(U; JR) is plurisubharmonic il and only if

Lev(f)(a, w) ~ 0 lor every a E U and every WE Ta.

PROOF: Let (a, w) be a tangent vector in G and a := aa,w' Then loa(O) =

I(a) and
(f 0 a),«O) = Lev(f 0 a)(O, 1) = Lev(f)(a, w).
Now, I is plurisubharmonic if and only if loa is subharmonic near 0 for any
a = aa,w' Equivalently, (f 0 a)a(O) ~ 0 for any such a. But this is true if
and only if Lev (f) (a, w) ~ 0 for any tangent vector (a, w) in G. _

2.9 Corollary. Let G 1 C en and G 2 c em be domains, F : G 1 -t G 2

a holomorphic map, and 9 E lf2 (G 1 ; JR) plurisubharmonic. Then 9 0 F is
plurisubharmonic on G 1.

PROOF: This is trivial, because of the formula in Example 4 above. _

Exhaustion Functions. For every domain Gee the function - log 8c

is subharmonic. In higher dimensions it is in general not true that this func-
tion is plurisubharmonic for every domain G.
Definition. Let G c en be a domain. A nonconstant continuous func-
tion I : G -t R is called an exhaustion function for G if for c < SUPa (f)
all sublevel sets
Gc(f) := {z E G : I(z) < c}
are relatively compact in G.

Example
For G = en, the function I(z) := IIzll2 is an exhaustion function. For G -# en,
we define the boundary distance 8a by

8c (z) := dist(z, en - G).

Then -8a is a bounded, and - log 8c an unbounded, exhaustion function.

We only have to show that 8c is continuous:
For every point z E G there is a point r(z) E en - G such that

8c(z) = dist(z, r(z» ~ dist(z, w) for every wEen - G.

Then for two arbitrary points u, v E G we have

 2. Plurisubharmonic Functions 59

l5G(u) = lIu - r(u)1I < lIu - r(v)1I ~ lIu - vII + l5G(v),

and in the same way l5G(v) < lIu - vII + 5G(u).
Therefore, Il5G(u) -l5G(v)1 ~ lIu - vII·
Definition. A function I E 'C2 ( G; JR) is called strictly plurisubhar-
monic if Lev (f) (a, w) > 0 for a E G, wE Ta , and w:I= o.

For a proof of the following result we refer to [Ra86], Chapter II, Proposition
4.14.
2.10 Smoothing lemma. Let G c en be a domain, I: G -+ JR a continu-
ous plurisubharmonic exhaustion junction, KeG compact, and c > O. Then
there exists a 'Coo exhaustion junction g : G -+ JR such that:
1. g ~ I on G.
2. g is strictly plurisubharmonic.
3. Ig(z) - l(z)1 < c on K.

Exercises
1. Let Gee be a domain. Prove the following statements:
(a) If I: G -+ e is a holomorphic function, then Ilia is subharmonic for
a> O.
(b) If u is subharmonic on G, then uP is subharmonic for pEN.
(c) Let u ~ -00 be subharmonic on G. Then {z E G : u(z) = -oo}
does not contain any open subset.
2. Let Gee be a domain, s t= -00 a subharmonic function on G, P :=
{z E G : s(z) = -oo}. Show that if u is a continuous function on G and
subharmonic on G - A, then u is subharmonic on G.
3. Let U c en be open, f: U -+ e k a holomorphic map, and A E Mk(JR) a
positive semidefinite matrix. Show that r,o(z) := fez) . A . fez) t is pluri-
subharmonic.
4. Let G = {( z, w) E e 2 : Iwl < Izl < 1} be the Hartogs triangle. Prove that
there does not exist any bounded plurisubharmonic exhaustion function
on G.
5. Are the following functions plurisubharmonic (respectively strictly pluri-
subharmonic) ?

Pl(Z) .- log(1 + IIzII2), for Z E en,

P2(Z) .- -log(1 - IIzII2), for IIzll < 1,
P3(Z) .- IIzIl2e-Re(z,,), for z E en.

6. Consider a domain G c en and a function I E 'C2 (G). Prove that I is

strictly plurisubharmonic if and only if for every open .set U C C G there
is an c > 0 such that I(z) - cllzll 2 is plurisubharmonic on U.
60 II. Domains of Holomorphy

3. Pseudoconvexity
Pseudo convexity
Definition. A domain G c c;n is called pseudoconvex if there is a
strictly plurisubharmonic Cfloo exhaustion function on G.

Remarks
1. By the smoothing lemma the following is clear: If - log 8G is plurisub-
harmonic, then G is pseudoconvex.
2. Pseudoconvexity is invariant under biholomorphic transformations.

3.1 Theorem. If G c c;n is a pseudoconvex domain, then C satisfies the

continuity principle.

PROOF: Let p : G -t IR. be a strictly plurisubharmonic exhaustion function.

Suppose that there exists a family {St : 0 :::; t :::; I} of analytic disks given
bya continuous mapping i.p: 5 x [0, 1]-t c;n such that So c G and bSt c G
for every t E [0,1]' but not all St are contained in G.
The functions po i.pt : 0 -t G are subharmonic for every t with St C G. It
follows by the maximum principle that plSt :::; maxbS. P for all those t.
We define to := inf{t E [0,11 : St ~ G}. Then to > 0, Sto c G, and Sto
meets BC in at least one point zoo We can find an increasing sequence (tv)
converging to to and a sequence of points Zv E Stv converging to Zoo So
p(zv) -t Co := sUPG(p), but there is a c < Co such that plbst :::; c for every
t E [0,11. This is a contradiction. -

3.2 Corollary. If G is pseudoconvex, then G is Hartogs convex.

The Boundary Distance

3.3 Theorem. If G c c;n is a Hartogs convex domain, then -log 8G is
plurisubharmonic on G.

PROOF: For Z E G and u E en with lIull = 1 we define

8G,u(z) := sup{t > 0 : Z + TU E C for ITI :::; t}.

Then oG(z) = inf{8G,u(z) : lIull = I}, and it is sufficient to show that
- log 8G,u is plurisubharmonic for fixed u.
(a) Unfortunately, 8G,u does not need to be continuous, but it is lower semi-
continuous:
 3. Pseudoconvexity 61

Let Zo E G be an arbitrary point and c < 6G,u(zo). Then the compact set
K := {z = Zo + ru : Irl :5 c} is contained in G, and there is a 6 > 0 such
that {z : dist{K, z) < 6} c G.
For z E B.s(zo) and Irl :5 c we have

lI{z + ru) - (zo + ru)1I = liz - zoll < 6, and therefore 8G,u(z) 2:: c.
(b) The function -log6G,u is upper semicontinuous, and we have to show
that
sec) := -log8G,u(zo + (b)
is subharmonic for fixed u, zo, b. First consider the case that u and b are
linearly dependent: b = AU, A =F o.
Let Go be the connected component of 0 in {( E C : Zo + (b E G}. Then

6G,u(zo + (b) = sup{t> 0 : Zo + (b + ru E G for Irl :5 t}

= sup{t > 0 : (+ riA E Go for Irl :5 t}
= IAI· sup{r > 0 : (+ a E Go for lal :5 r}
= IAI·8Go{(),

and this function is in fact subharmonic.

(c) Now assume that u and b are linearly independent. Since these vectors
are fixed, we can restrict ourselves to the following special situation:

n=2, zo=O, b=el, and u=e2.

Then s«() = -logsup{t > 0 : «(, r) E G for Irl :5 t}. We use holomorphic
functions to show that s is subharmonic. Let R > r > 0 be real numbers such
that {(, 0) E G for I( I < R, and let f : DR (0) -t <C be a holomorphic function
such that s < h:= Ref on 80 r (0). We have to show that s < h on OrCO).
We have the following equivalences:

s«() < h{() ¢=> sup{t > 0 : {(, r) E G for Irl :5 t} > e-h(C)
¢=> ((,c.e-f(C»)EGforcED.

Cd) Define a holomorphic map F by

Then F is well defined on a neighborhood of the unit polydisk p2 = p2{O, 1).

It must be shown that F{P2) c G. We already know the following:
1. F(zl, Z2) E G for IZII = 1 and IZ21:5 1, because set) < h{t) on 80r (0).
2. F{zl,O) E G for IZll :5 1, because «(,0) E G for 1(1 :5 r.
62 II. Domains of Holomorphy

These facts will be used to construct an appropriate Hartogs figure. First,

note that

JF(Zt, Z2) = ( : e-f~rzl»)' so det JF(Zl, Z2) -=f O.

By the inverse function theorem it follows that F is biholomorphic.

For 0 < a < 1 we define h.s : ((:2 -+ ((:2 by h.s(Zl' Z2) := (Zl, OZ2) and apply h.s
to the compact set

Consequently,
Co := ho(C) = {(Zt,Z2) E ((:2 : (lzll::; 1, Z2 = 0) or (IZll = 1, IZ21 ::; o)}.
Then F(Co) C G, as we saw above, and therefore C.s C F-l(G).
For 0 < e < min(o, 1 - 0) we define a neighborhood U€ of C.s by U€ :=
{(Zl,Z2) E ([2 : (IZll < l+e, IZ21 < e) or (l-e < IZII < l+e, IZ21 < o+e)}.
If we choose e small enough, then U€ C F-l(G).
Finally, we define H€ := hil(U€ n P2) n p2 (see Figure II.6). Then
H€ = {(Z},Z2) E p2 : (Zl,OZ2) E u€np2}
= {(Zl, Z2) E (:2 : (IZll < 1, IZ21 < i) or (1 - e < IZII < 1, IZ21 < I)}.

l-e

Figure 11.5. Construction of the Hartogs figure

Since (P2,H€) is a Euclidean Hartogs figure, (F 0 h.s(p2),F o--h.s(H€» is a

general Hartogs figure with F 0 ho(H€) C F(U€ n p2) C G. Since Gis Hartogs
 3. Pseudoconvexity 63

pseudoconvex, it follows that F 0 h.;(P2) C G. This is valid for every < 1. a

But p 2 = UO<';<1 h.;(P2). Therefore, F(p2) C G, which was to be shown. _

3.4 Theorem. The following properties of a domain G c en are equivalent:

1. G satisfies the continuity principle.
2. G is H artogs pseudoconvex.
3. - log Ja is plurisubharmonic on G.
4. G is pseudoconvex.

PROOF:

(1) ~ (2) is Theorem 1.5,

(2) ~ (3) is Theorem 3.3,
(3) ~ (4) follows from the smoothing lemma,
(4) ~ (1) was proved in Theorem 3.1.
-
Properties of Pseudoconvex Domains
3.5 Theorem. If G I , G 2 c en are pseudoconvex domains, then G I n G 2 is
pseudoconvex.

PROOF: The statement is trivial if one uses Hartogs pseudo convexity. _

3.6 Theorem. Let G I C G 2 C ... c en be an ascending chain of pseudo-

convex domains. Then G:= U::I G v is again pseudoconvex.

PROOF: This follows immediately from the continuity principle. _

3.7 Theorem. A domain G c en is pseudoconvex if and only if there is an

open covering (U.).EI of G such that U. n G is pseudoconvex for every LEI.

PROOF:

"~" is trivial. The other direction will be proved in two steps. At first, we
assume that G is bounded.
For any point Zo E 8G there is an open set U. such that Zo E U. and Gnu.
is pseudoconvex. If we choose a neighborhood W = W(zo) C U. so small
that dist(z, 8U.) > dist(z, zo) for every z E WnG, then oa(z) = 6anu.(z) on
WnG. This shows that there is an open neighborhood U = U(8G) such that
-log6a is plurisubharmonic on Un G (we use the fact that 8G is compact).
Now, G - U cc G. We define
64 II. Domains of Holomorphy

c:=sup{-log8G(z): zEG-U},

and
p(Z) := max ( -log 8G(z), IIzl12 + C + 1).
Then p is a plurisubharmonic exhaustion function, and by the smoothing
lemma, G is pseudoconvex.
If G is unbounded, we write it as an ascending union of the domains
G v := Bv(O) n G. Each G v is bounded and satisfies the hypothesis, so is
pseudoconvex. Then G is also a pseudoconvex domain. _

Exercises
1. Suppose that G1 c en and G2 c em are domains.
(a) Show that if G1 and G2 are pseudoconvex, then G1 x G2 is a pseu-
doconvex domain in n +m e
(b) Show that if there is a proper holomorphic map f : G1 -+ G 2 and G2
is pseudoconvex, then G 1 is also pseudoconvex.
2. Let G c en be a domain and (! : G -+ JR a lower semicontinuous positive
function. Prove that

G := {(z', w) E G x e : Iwl < (!(z')}

is pseudo convex if and only if - log (! is plurisubharmonic.
3. A domain G c en is pseudoconvex if and only if for every compact set
KeG the set

Kpi := {z E G: p(z) ~ s~pp for all plurisubharmonic functions p on G}

is relatively compact in G.

4. Levi Convex Boundaries

Boundary Functions
Definition. Let G c en be a domain. The boundary of G is called
smooth at Zo E BG if there is an open neighborhood U = U(zo) c en
and a function (! E 'ff'OO(U;lR) such that:
1. Un G = {z E U : (!(z) < O}.
2. (d(!)z =1= 0 for z E U.
The function (! is called a local defining function (or boundary function).

Remark. Without loss of generality we may assume that (!Yn 1= O. Then

by the implicit function theorem there are neighborhoods
 4. Levi Convex Boundaries 65

U ' o f (ZO'
' Xn(0) -_ (
ZI(0) , ... , Zn-l'
(0)
Xn(0)) E trn-l
~
ltD
X n., utI of Y~) E JR,

and a «foo function "( : U' -+ UtI such that {(z', Xn , Yn) E U' X UtI : [!(z', Xn +
iYn) = O} = {(z', Xn , "(z', xn)) : (z',x n ) E U'}.
Making the neighborhood U:= {(z',x n + iYn) : (z',x n ) E U' and Yn E UtI}
small enough and correcting the sign if necessary, one can achieve that

Un G = {(z', Xn + iYn) EU : Yn < "(z', xn)}.

In particular, Un BG = {z E U : [!(z) = O} is a (2n - I)-dimensional

differentiable sub manifold of U.

4.1 Lemma. Let BG be smooth at Zo, and let [!1, [!2 be two local defining
functions on U = U (zo). Then there is a «foo junction h on U such that:
1. h > 0 on U.
2. [!1 = h . [!2 on U.
3. (d[!dz = h(z) . (d[!2)z for z E Un BG.

PROOF: Define h := eI/ [!2 on U - BG. After a change of coordinates, we

.have Zo = 0 and [!2 = Yn. Then get) := [!1(Z',X n + it) is a smooth function
that vanishes at t = o. Therefore,

[!1 (z', zn) g(Yn) - g(O)

= IoYn g'(8) d8 = Y: . 10 1 g'(tYn) dt

= [!2(Z',X n + iYn)· h(z',zn),

where
h( Z,Xn+IYn
I .)
=
11 B[!1 ( , . ) dt
~ Z,Xn+ltYn
o UYn
is smooth.

°
For Z E BG we have (d[!dz = h(z) . (de2)z. Therefore, h(z)
greater than 0, since h(z) 2': by continuity.
=f. 0, and even
_

4.2 Theorem. Let G cc en be a bounded domain with smooth boundary.

Then BG is a differentiable submanifold, and there exists a global defining
function.

PROOF: We can find open sets Vi CC Ui c en, i = 1, ... , N, such that:

1. {Vi, ... , VN } is an open covering of BG.
2. For each i there exists a local defining function [!i for G on Ui .
66 II. Domains of Holomorphy

3. For each i there is a smooth function 'Pi : Ui -+ IR. with 'PilVi == 1,

'Pilcn-u; == 0, and 'Pi :::=: 0 in general.
Define 'P := Li 'Pi (so 'P > 0 on 8G) and 'l/Ji := 'Pi/'P. Then Li 'l/Ji == 1 on
aG. The system of the functions 'l/Ji is called a partition of unity on aG.
The function e := L;:1 'l/Jiei is now a global defining function for G. We
leave it to the reader to check the details. _

The Levi Condition. For the remainder of this section let Gee en
be a bounded domain with smooth boundary, and e : U = U( 8G) -+ IR. a
global defining function. Then at any Zo E aG the real tangent space of the
boundary
Tzo(8G) := {v E Tzo : (de)zo(v) = O}
is a (2n - 1)-dimensional real subspace of Tzo' The space

is called the complex (or holomorphic) tangent space of the boundary at Zoo
It is a (2n - 2)-dimensional real subspace of Tzo, with a natural complex
structure, so an (n - I)-dimensional complex subspace3 .
Definition. The domain G is said to satisfy the Levi condition (respec-
tively the strict Levi condition) at Zo E 8G if Lev(e) is positive semidef-
inite (respectively positive definite) on Hzo(8G). The domain G is called
Levi convex (respectively strictly Levi convex) if G satisfies the Levi con-
dition (respectively the strict Levi condition) at every point Z E aG.

Remark. The Levi conditions do not depend on the choice of the boundary
function, and they are invariant under biholomorphic transformations.
If el = h· e2, with h> 0, then for Z E 8G,

So on Hz(8G) the Levi forms of el and e2 differ only by a positive constant.

Affine Convexity. Recall some facts from real analysis:

A set M c IR.n is convex if for every two points x, y EM, the closed line
segment from x to y is contained in M. In that case, for each point Xo E
IR.n - M there is a real hyperplane H c IR.n with Xo E Hand M n H = 0.
This property was already used in Section 1.
3 Hz(8G) is often denoted by T;,O(8G).
 4. Levi Convex Boundaries 67

If a E ]Rn, U = U(a) is an open neighborhood and <p: U ~ lR is at least lf2,

then the quadratic form

Hess(<p)(a, w) := L <Pxvx,. (a)wVwtl

is known as the Hessian of <p at a.
4.3 Proposition. Let G cc ]Rn be a domain with smooth boundary, and
e a global defining function with (de)x i= 0 for x E BG. Then G is convex if
and only if Hess(e) is positive semidefinite on every tangent space Tx (BG).

PROOF: Let G be convex, and Xo E BG an arbitrary point. Then T :=

Txo(BG) is a real hyperplane with TnG = 0. For wET and a(t) := xo+tw
we have
(e 0 a)"(O) = Hess(e)(Xo, w).
Since e(xo) = 0 and eo a(t) ~ 0, it follows that {! 0 a has a minimum at
t = O. Then ({! 0 a)"(O) ~ 0, and Hess(e) is positive semidefinite on T.

Now let the criterion be fulfilled, assume that 0 E G, and define {!e by

For small c: and large N the set Go := {x : (!e(x) < O} is a domain. We have
Go C Gol C G for c:' < c:, and Uoo Go = G. Therefore, it is sufficient to
show that Go is convex.
The Hessian of (!o is positive definite on Tx(aG) for every x E BG. Thus this
also holds in a neighborhood U of BG. If c: is small enough, then BGe C U.
We consider

S:= {(x, y) E Go X Ge : tx + (1 - t)y E Go, for 0 < t < I}.

Then S is an open subset of the connected set Go xG e . Suppose that S is not a

closed subset. Then there exist points xo,yo EGg and a to E (0,1) with toxo+
(1- to)yo E aGo. So the function t H eg 0 a(t), with a(t) := txo + (1- t)yO,
has a maximum at to. Then (eooa)"(to) :S 0 and Hess(eo)(a(to),xo-Yo) :S O.
This is a contradiction. _
A domain G = {e < O} is called strictly convex at Xo E BG if Hess({!) is
positive definite at Xo. This property is independent of {! and invariant under
affine transformations.
Now we return to Levi convexity.
4.4 Lemma. Let U c en be open and <p E ce'2 (U j ]R). Then

Lev(<p)(z, w) = ~ (Hess(<p)(z, w) + Hess(<p)(z, iw».

68 II. Domains of Holomorphy

PROOF: This is a simple calculation! •

4.5 Theorem. Let Gee en
be a domain with smooth boundary. Then the
following statements are equivalent:
1. G is strictly Levi convex.
2. There is an open neighborhood U = U (oG) and a strictly plurisubhar-
monic function (J E 'i&'OO(U;lR) such that Un G = {z E U : (J(z) < O}
and (d(J)z =I 0 for Z E U.
3. For every z E oG there is an open neighborhood W = W(z) c en,
an open set V C en, and a biholomorphic map F : W -+ V such that
F(WnG) is convex and even strictly convex at every point ofF(WnoG).

PROOF:

(1) => (2) : We choose a global defining function e for G, and an open
neighborhood U = U(oG) such that e is defined on U with (de)z =I 0 for
z E U. Let A > 0 be a real constant, and eA := e Ae - 1. Then eA is also a
global defining function, and

The set K := oG X s2n-l is compact, and

Ko := {(z, w) E K : Lev(e)(z, w) ::::; O}

is a closed subset. Since Lev(e) is positive definite on Hz(oG), we have

(oe)z(w) =I 0 for (z, w) E Ko. Therefore,

M .- min Lev(e)(z, w) > -00,

K
C minl(oe)z(wW > O.
Ko

We choose A so large that A . C + M > O. Then

Lev(eA)(Z, w) = A· [Lev(e)(z, w) + AI(oe)z(w)1 2 ] 2 A· (M + AC) >0

for (z, w) E Ko, and

for (z,w) E K - K o.
So Lev(eA)(Z, w) > 0 for every z E oG and every wEen - {OJ. By conti-
nuity, eA is strictly plurisubharmonic in a neighborhood of aGo
(2) => (3) : We consider a point Zo E oG and make some simple coordinate
transformations:
 4. Levi Convex Boundaries 69

By the translation Z H W = Z - Zo we replace Zo by the origin, and a

permutation of coordinates ensures that (JWl (0) i= O.
The linear transformation

gives Ul = W· Y' (J(O) t, and therefore

(J(U) 2Re(u·Y'(e ow )(0)t)+ terms of degree ~2

2 Re (u . Jw(O) t . Y' (J(O) t) + terms of degree ~ 2
2 Re (w . Y' (J(O) t) + terms of degree ~ 2
2 Re( ud + terms of degree ~ 2.

Finally, we write (J(u) = 2 Re(ul + Q(u» + Lev({J)(O, u) +"', where Q is a

quadratic holomorphic polynomial, and make the biholomorphic transforma-
tion

It follows that

(J(v) = 2 Re(vd + Lev({J)(O, v) + terms of order ~ 3.

By the uniqueness of the Taylor expansion

1
e(v) = D{J(O)(v) + 2Hess({J)(O, v) + terms of order ~ 3,

and therefore Hess({J)(O, v) = 2· Lev(e)(O, v) > 0 for v i= (in the new

coordinates). Everything works in a neighborhood that may be chosen to be
°
convex.
(3) =:} (1) : This follows from Lemma 4.4:

The latter property is invariant under biholomorphic transformations. •

A Theorem of Levi. Let G cc en be a domain with smooth boundary.

If G is strictly Levi convex, then it is easy to see that G is pseudoconvex.
We wish to demonstrate that even the weaker Levi convexity is equivalent
to pseudo convexity. For that purpose we extend the boundary distance to a
function on en.
for Z E G,
for Z E BG,
for z ~ G.
70 II. Domains of Holomorphy

4.6 Lemma. -de is a smooth defining function for G.

PROOF: We use real coordinates x = (Xl"," XN) with N = 2n. It is clear

that G = {x : -dc(x) < a}.
Let Xo E DG be an arbitrary point and (l : U(xo) -t IR a local defining
function. We may assume that {lXN (xo) f. O. Then by the implicit function
theorem there is a product neighborhood U' x U" of Xo in U and a smooth
function h : U' -t IR such that

{(X',XN) E U' x U" : (2(X',XN) =O} = {(x',h(x')) : x' E U'}.

It follows that 0 = V'x /{2(x', hex')) + (2xN(X', hex'»~ . V'h(x').

At the point (x', h(x'» E DG the gradient V' (l(x', h(x'» is normal to DG
and directed outward from G. Every point y in a small neighborhood of the
boundary has a unique representation y = x + t . V' (2(x) , where t = -dc(Y)
and x is the point where the perpendicular from y to DG meets the boundary.
Therefore, we define the smooth map F : U' x IR -t IRN by

y = F(x', t) := (x', hex'»~ + t· V' (l(x', h(x'».

Then there are smooth functions A and b such that

, ( E N - l + t· A(x') V'X/{l(X', hex'»~ t )
JR.,F(X ,t) = V' hex') + t . b(x') &.>XN (x', h(x'» ,

and therefore

det J]R,F(X', 0) det ( EN -1 -{lxN (x', h(x'» . V' hex') t )

V' hex') /?XN (x', h(x'»

=
{lxN
('
x,hx
('» d (E
. et
N- l
0'
-V'h(x') t
1+IIV'h(x')1I2
)

/?XN (x', h(x'»(l + IIV'h(x')II2) f. O.

It follows that there exists an c: > 0 such that F maps U' x (-c:, c:) difIeo-
morphically onto a neighborhood W = W(xo), and U' x {a} onto DG n W.
Moreover, since dc(x + t . V' (l(x» = -t for It I < c: and c small enough, it
follows that de = (-t) 0 F- l is a smooth function near DG. If p' is defined
by p'(x',t) := (x',O), then the projection

p=p'oF- l :x+t'V'{l(X)HX, forxEDG,

is a smooth map, and de is given by dc(Y) = (j. IIy - p(y)II, where (j = 1

for y E G and (j = -1 elsewhere.
For y (j. DG we have
 4. Levi Convex Boundaries 71

lIy - :(y) II . t;(Yk - Pk (y)) (8 kv - (Pk)y,,(y))

= lIy - :(Y)II . [Yv - Pv(y) - (y - p(y) IPy,,(Y») N] ,

and therefore
(T
V'de(y) = lIy _ p(y)1I . [y - p(y) - Dp(y)(y - p(y»].

Since (J(p(y» == 0, it follows that Dp(y)(V' (J(p(y))) = o. But y - p(y) is a

multiple of V' (J(p(y». Together this gives
y - p(y) V' e(p(y»
V'de(y) = (T. lIy - p(y) II = ± IIV' e(p(y»U·
If y tends to 8G, we obtain that V'dG (y) =/: o. •
E.E. Levi showed that every domain of holomorphy with smooth boundary is
Levi convex, and locally the boundary of a strictly Levi convex domain G is
the "natural boundary" for some holomorphic function in G. Here we prove
the following result, which is sometimes called "Levi's theorem".
4.7 Theorem. A domain G with smooth boundary is pseudoconvex if and
only if it is Levi convex.

PROOF:

(1) Let G be pseudoconvex. The function -de is a smooth boundary function

for G, and - log dG = -log 8e is plurisubharmonic on G, because of the
pseudoconvexity. We calculate
1 1 2
Lev( -logde)(z, w) = dG(z) . Lev( -dG)(z, w) + dG(z)2 ·1(8(dG»z(w)1 .

This is nonnegative in G. If Z E G, W E Tz , and (8(dG»z(w) = 0, it follows

that Lev( -dG)(z, w) :::=: O. This remains true for Z ~ 8G, so -dG satisfies
the Levi condition.
(2) Let G be Levi convex, and suppose that G is not pseudoconvex. Then in
any neighborhood U of the boundary there exists a point Zo where the Levi
form of - log 8e has a negative eigenvalue. This means that there is a vector
Wo such that
ep((O) = Lev(log8G)(zo, wo) > 0, for ep(e) := log8G(zo + (wo)·
Consider the Taylor expansion
1
ep(O) + 2 Re( ep<C 0)( + 2ep(d O)( ) + epa(O) 1(1 + ...
2 2
ep«()
ep(O) + Re(A( + B(2) + >'1(1 2 + ... ,
72 II. Domains of Holomorphy

with complex constants A, B and a real constant A > o.

We choose a point Po E 8G with dG(Zo) = IIpo - zoll, and an arbitrary c > O.
Then an analytic disk t/J : De(O) ---t en can be defined by
t/J«) := Zo + (wo + exp(A( + B(2)(po - zo).
We have t/J(O) = Po, and we wish to show that t/J«) E G, for 0 < 1(1 < c and
e sufficiently small.
Since c,o«) ~ c,o(O) + Re(A( + B(2) + (A/2)1(1 2 near ( = 0, it follows that

dG(zo + (wo) = exp(c,o«))

> exp(c,o(O))·1 exp (A( + B(2) I· exp(~1(12)

> dG(zo)·1 exp (A( + B(2) 1
= lIexp (A( + B(2) (Po - zo)lI,
for ( small and # O. This means that we can choose the c in such a way
that t/J«) E G, for 0 < 1(1 < e. The analytic disc is tangent to 8G from the
interior of G.
Now I«) = da(t/J«)) is a smooth function with a local minimum at ( = O.
Therefore (8dG)po(t/J'(O)) = (81)0(1) = 0, and
I«) = Re U<dO)(2) + 1<,1(1 2 + terms of order ~ 3.
Since Re U«(0)e 2it ) + 1<, ~ 0 for every t, it follows that

Lev(dG)(Po,t/J'(O)) = la(O) > O.

This is a contradiction to the Levi condition at Po, because -dG is a defining
function for G. •

Exercises
1. Prove Lemma 4.4.
2. Assume that G CC (:2 has a smooth boundary that is Levi convex outside
a point a that is not isolated in {)G. Show that G is pseudoconvex.
3. Assume that G C (:2 is an arbitrary domain and that S eGis a smooth
real surface with the following property: In every point of S the tangent
to S is not a complex line. Prove that for every compact set KeG there
are arbitrarily small pseudoconvex neighborhoods of S n K.
4. Assume that G cc (:2 is a domain with smooth boundary. Then G
is strictly Levi convex at a point Zo E {)G if and only if the following
condition is satisfied:
There is a neighborhood U = U(zo), a holomorphic function c,o : 0 ---t U
with c,o(O) = Zo and c,o'(O) # 0, and a local defining function (J on U such
that «(J 0 c,o)«) > 0 on 0 - {O} and (e 0 c,o)a(O) > O.
 5. Holomorphic Convexity 73

5. Let Gee en be a domain with smooth boundary. If G satisfies the strict

Levi condition at Zo E 8G, then prove that the following hold:
(a) There is no analytic disk <p: D ~ en with

<p(O) = Zo and lim 6G(<p«(» = O.

(~o 1I<p«() - <p(O)1I2
(b) There are a neighborhood U = U(zo) and a holomorphic function I
in U with Gn {z E U : I(z) = O} = {zo}.
6. A bounded domain G c en is called strongly pseudoconvex if there are
a neighborhood U = U(8G) and a strictly plurisubharmonic function
(! E ~2(U) such that GnU = {z E U : {!(z) < O}. Notice that a strongly
pseudoconvex domain does not necessarily have a smooth boundary!
Prove the following results about a strongly pseudoconvex bounded d~
main G:
(a) Gis pseudoconvex.
(b) If G has a smooth boundary, then G is strictly Levi convex.
(c) For every z E 8G there is a neighborhood U = U(z) such that UnG
is a weak domain of holomorphy.
7. Let G c en be a pseudoconvex domain. Then prove that there is a family
of domains G v C G such that the following hold:
(a) Gv CC Gv+ 1 for every 1/.
(b) U~l G v = G.
(c) For every 1/ there is a strictly plurisubharmonic function Iv E
~OO(Gv+l) such that G v is a connected component of the set

{z E GV+l : Iv(z) < O}.

5. Holomorphic Convexity
Affine Convexity We will investigate relationships between pseudocon-
vexity and affine convexity. Let us begin with some observations about convex
domains in RN.
Let .!L' be the set of affine linear functions I : RN ~ R with

I(x) = alXl + ... + aNxN + b, all ... ,aN, bE R.

If M is a convex set and Xo a point not contained in M, then there exists

a function I E .!it with I(xo} = 0 and 11M < O. For any c E R, the set
{x E RN : I(x) < c} is a convex half-space.
Definition. Let M C RN be an arbitrary subset. Then the set

H(M) := {x E RN : I(x) :5 s"'ff I, for all f E .!L'}

is called the affine convex hull of M.
74 II. Domains of Holomorphy

5.1 Proposition. Let M,M1 ,M2 C]RN be arbitrary subsets. Then

1. Me H(M).
2. H (M) is closed and convex.
3. H(H(M)) = H(M).
4· If Ml C M 2 , then H(Md C H(M2)'
5. If M is closed and convex, then H(M) = M.
6. If M is bounded, then H (M) is also bounded.

PROOF: (1) is trivial.

(2) If Xo (j. H(M), then there is an f E .!£ with J(xo) > sUPM J. By conti-
nuity, J(x) > SUPM J in a neighborhood of Xo. Therefore, H(M) is closed.
If Xo, Yo are two points in H(M), then they are contained in every convex
half-space E = {x : J(x) < SUPM f}, and also the closed line segment from
Xo to Yo is contained in each of these half-spaces. This shows that H(M) is
convex.
(3) We have to show that H(H(M)) C H(M). If x E H(H(M)) is an arbi-
trary point and f an element of .!£, then J(x) ~ sUPH(M) J ~ sUPM J, by
the definition of H (M).
(4) is trivial.
(5) Let M be closed and convex. If Xo (j. M, then there is a point Yo E M
such that dist(xo, M) = dist(xo, Yo) (because M is closed). Let Zo be a point
in the open line segment from Xo to Yo. Then Zo (j. M, and there is a function
J E .!£ with J(zo) = 0 and JIM < O. Since t M f(txo + (1 - t)yo) is a
monotone function, it follows that J(xo) > 0 and therefore Xo (j. H(M).
(6) If M is bounded, there is an R > 0 such that M is contained in the closed
convex set BR(O). Thus H(M) c BR(O). •

Remark. H(M) is the smallest closed convex set that contains M.

5.2 Theorem. A domain G C ]RN is convex if and only if K cc G implies

that H(K) cc G.

PROOF: Let G be a convex domain, and M CC G a subset. Then H(M) is

closed and contained in the bounded set H(M). Therefore, H(M) is compact,
and it remains to show that H(M) C G. If there is a point Xo E H(M) - G,
then there is a function f E .!£ with J(xo) = 0 and fie < O. It follows that
sUPM f < 0, and f(xo) > SUPM f· This is a contradiction to Xo E H(M).
 5. Holomorphic Convexity 75

On the other hand, let the criterion be fulfilled. If Xo, Yo are two points of G,
then K := {Xo, Yo} is a relatively compact subset of G. It follows that H(K)
is contained in G. Since H (K) is closed and convex, the closed line segment
from Xo to Yo is also contained in G. Therefore G is convex. _

Holomorphic Convexity. Now we replace affine linear functions by

holomorphic functions.
Definition. Let G C c;n be a domain and KeG a subset. The set

K = KG := {z E G : /f(z)/ :5 s~/f/, for all f E O(G) }

is called the holomorphically convex hull of Kin G.

5.3 Proposition. Let G C c;n be a domain, and K, Kl, K2 subsets of G.

Then
1. KCK.
2. 8 is closed in G.
3. K=K.
4. If Kl C K 2 , then Kl C K;.
5. If K is bounded, then K is also bounded.

PROOF: (1) is trivial.

(2) Let Zo be a point of G - K. Then there exists a holomorphic function
f on G with If(zo)/ > sUPKIf/. By continuity, this inequality holds on an
entire neighborhood U = U(zo) c G. So G - K is open.
(3) sUPR/f/ :5 sUPK/f/·
(4) is trivial.
(5) If K is bounded, it is contained in a closed polydisk pn(o, r). The coordi-
nate functions Zv are holomorphic in G. For z E K we have /zv/ :5 SUPK/zv/ :5
r. Hence K is also bounded. _

Definition. A domain G C c;n is called holomorphically convex if

K cc G implies that K cc G.
76 II. Domains of Holomorphy

Example
In C; every domain is holomorphically convex:
Let K cc G be an arbitrary subset. Then K is bounded, and it remains
to show that the closure of K is contained in G. If there is a point Zo E
K - G, then Zo lies in aR naG. We consider the holomorphic f~nction
f(z) := l/(z - zo) in G. If (ZII) is a sequence in K converging to zo, then
If(zlI)1 ~ sUPKlfl ~ sUPKlfl < 00. This is a contradiction. For n ~ 2, we
will show that there are domains that are not holomorphically convex. But
we have the following result.
5.4 Proposition. If G c c;n is an affine convex domain, then it is holo-
morphically convex.

PROOF: Let K be relatively compact in G. Then H(K) cc G. If Zo is a

point of G - H(K), then there exists an affine linear function A E C with
A(ZO) > sUPK A. Replacing A by A - A(O) we may assume that A is a homo-
geneous linear function of the form

A(Z) = 2 Re(alzl + ... + anzn ).

Then f(z) := exp(2 . (alzl + ... + anzn )) is holomorphic in G, and If(z)1 =
exp(-X(z)). Therefore, If(zo)1 > sUPKlfl, and Zo E G - K. This proves K cc
G. •
In general, holomorphic convexity is a much weaker property than affine
convexity.

The Cartan-Thullen Theorem. Let G C c;n be a domain, and

c > 0 a small real number. We define

G/Z := {z E G : 8a (z) ~ c}.

Here are some properties of the set G£:

1. If z E G, then there is an c > 0 such that 8G(z) ~ c.
Therefore, G = U£>o G/Z.
2. If Cl ~ C2, then GEl:::> G£2'
3. G E is a closed subset of c;n. In fact, if Zo E c;n - GE , then 8G(zo) < c
or Zo ¢ G. In the latter case, the ball B£{zo) is contained in c;n - G E • If
Zo E G - G/Z and 8 := 8a(zo), then B£-.s(zo) C c;n - G/Z. So c;n - G E is
open.

5.5 Lemma. Let G C c;n be a domain, KeG a compact subset, and f a

holomorphic function in G. If KeG £, then fOT any 8 with 0 < 8 < c there
exists a constant C > 0 such that the following inequality holds:
 5. Holomorphic Convexity 77

PROOF: For 0 < 6 < E, G' := {z E G : dist(K,z) < 8} is open and

relatively compact in G, and for any z E K the closed polydisk pn(z,6) is
contained in G' c G. If T is the distinguished boundary of the polydisk and
III ~ Con G', then the Cauchy inequalities yield

5.6 Theorem (Cartan-Thullen). IIG is a weak domain 01 holomorphy,

-
then G is holomorphically convex.

PROOF: Let K cc G. We want to show that K cc G. Let E .-

dist(K, en - G) ~ dist(K, en - G) > O. Clearly, K lies in Ge .
We assert that the holomorphically convex hull K lies even in Ge . Suppose
this is not so. Then there is a Zo E K - G e . Now let I be any holomorphic
function in G. In a neighborhood U = U(zo) c G, I has a Taylor expansion

I(z) =L all(z - zot, with all = ~DII

1/.
I(zo).
"2':0

The function z H a,,(z) := ~D" I(z) is holomorphic in G. Therefore,

lall(zo)1 ~ sUPKlall(z)l. By the lemma, for any 6 with 0 < 6 < E there
exists a C > 0 such that sUPKlall(z)1 ~ CI6 1"1, and then

On any polydisk pn(zo,8) the Taylor series is dominated by a geometric

series. Therefore, it converges on P = pn (zo, E) to a holomorphic function f
We have I = 1 near zo, and then on the connected component Q of Zo in
P n G. Since P meets G and en - G, it follows from Lemma 1..9 that there
is a point Zl E P n aQ naG. Then I cannot be completely singular at Zl.
This is a contradiction, because I is an arbitrary holomorphic function in G,
and G is a weak domain of holomorphy. _

Exercises
1. Let G 1 C G 2 c en be domains. Assume that for every I E O( Gd there
is a sequence of functions I" E O(G 2 ) converging compactly on G l to I.
Show that for every compact set K C G 1 it follows that KG 2 nG l = KG\.
78 II. Domains of Holomorphy

2. Let F : G 1 -+ G2 be a proper holomorphic map between domains in en,

respectively em. Show that if G 2 is holomorphically convex, then so is
G1 .
3. Let G c en be a domain an~ c G be a closed analytic disk with
boundary bS. Show that S C (bS) G.
4. Define the domain G c e 2 by G := p2(0, 1) - P2(0, 1/2). Construct the
holomorphically convex hull KG for K := {(z!, Z2) : ZI = 0 and IZ21 =
3/4}. Is KG a relatively compact subset of G?
5. Let F be a family of functions in the domain G. For a compact subset
KeG we define

KF := { z E G : If(z)1 ~ s~plfl for all f E F} .

The domain G is called convex with respect to F, provided that KF is
relatively compact in G whenever K is. Prove:
(a) Every bounded domain is convex with respect to the family 'tfO(G)
of all continuous functions.
(b) The unit ball B = B1 (0) is convex with respect to the family of
holomorphic functions zi· z~ with V,j-t = 1, ... ,n and k,l E No.

6. Singular Functions
Normal Exhaustions. Let G c enbe a domain. If G is holomorphically
convex, we want to construct a holomorphic function in G that is completely
singular at every boundary point. For that we use "normal exhaustions."

Definition. A normal exhaustion of G is a sequence (Kv) of compact

subsets of G such that:
1. K" cc (K"+lt, for every v.
2.U:: 1 K" = G.

6.1 Theorem. Any domain G in en admits a normal exhaustion. If Gis

holomorphically convex, then there is a normal exhaustion (Kv) with K" =
K" for every v.
PROOF: In the general case, K" := pn(o, v) n GIl" gives a normal exhaus-
tion. If Gis holomorphically convex, K" cc G for every v. We construct a
new exhaustion by induction.
Let Ki := R1. Suppose that compact sets Ki, ... , K~_l have been con-
structed, with Rj = Kj for j = 1, ... , v-I, and Kj cc (K/+lt. Then
there exists a >,(v) EN such that K~_l C (K).(,,)t. Let K~ := K).(,,).
 6. Singular FUnctions 79

It is clear that (K~) is a normal exhaustion with K~ = K~. •

Unbounded Holomorphic Functions. Again let G c en be a

domain.
6.2 Theorem. Let (Kv) be a normal exhaustion 01 G with Kv = K v ,
)..(/1) a strictly monotonic increasing sequence 01 natural numbers, and (ZIl)
a sequence 01 points with ZIl E K)..(Il)+l - K)..(Il).
Then there exists a holomorphic lunction I in G such that I/(zll)1 is un-
bounded.

PROOF: The function I is constructed as the limit function of an infinite

series I = E:=l Iw By induction we define holomorphic functions III in G
such that:
1. 1IIlIK).(I') < 2- 1l for J.L ~ 1.
,,-I
2. 111l(zll)1 > /1 + 1 + LlfJ(zll)1 for /1 ~ 2.
j=l

Let II := o. Now for /1 ~ 2 suppose that 11, ... ,1,,-1 have been constructed.
Since ZIl E K)..(Il)+l - K)..(Il) and K)..(Il) = K)..(Il) , there exists a function 9
holomorphic in G such that Ig(zll)1 > q := sUPK).(I') Igl. By multiplication by
a suitable constant we can make

Ig(z,,)I> 1 > q.
If we set III := gk with a sufficiently large k, then I" has the properties (1)
and (2).
We assert that E" II-' converges compactly in G. To prove this, first note
that for KeG an arbitrary compact subset, there is a /10 E N such that
K C K)..("o). By construction SUPK if" 1 < 2-" for /1 ~ J.Lo. Since the geo-
metric series E,. 2-" dominates E I" in K, the series of the I" is normally
f
convergent on K. This shows that = EI-'I" is holomorphic in G. Moreover,

I/(z,,)1 > 1/,,(zJ1.)I- Lllv(zJ1.)1

v¥"

> J.L +1- L TV (since zJ1. E K)..(v) for v> J.L)

> J.L (since LTV = 1).

v~l

It follows that I/(z,,)1 -+ 00 for J.L -+ 00. •

80 II. Domains of Holomorphy

The following is an important consequence:

6.3 Theorem. A domain G is holomorphically convex if and only if for
any infinite set D that is discrete in G there exists a junction f holomorphic
in G such that If I is unbounded on D.

PROOF: (1) Let G be holomorphically convex, D c G infinite and discrete.

Moreover, let (Kv) be a normal exhaustion of G with Kv = Kv. Then Kv nD
is finite (or empty) for every v E N. We construct a sequence of points z/A E D
by induction.
Let Zl ED - Kl be arbitrary, and >'(1) EN minimal with the property that
Zl lies in K A(1)+l' Now suppose the points Zl, •.• , Z/A-l and the numbers
>'(1), ... , >'(J.L - 1) have been constructed such that

Zv E K A(v)+l - KA(v), for v = 1, ... , J.L - 1.

Then we choose z/A E D - K A(/A-l)+1 and >'(J.L) minimal with the property
that z/A lies in K A(/A)+1' By the theorem above there is a holomorphic function
fin G such that If(z/A)I -+ 00 for J.L -+ 00. Therefore, If I is unbounded on D.
(2) Now suppose that the criterion is satisfied, and K cc G. Then KeG,
and we have to show that K is compact. Let (zv) be any sequence of points
in K. Then

sup{lf(Zv)1 : v E N} $ suplfl
, K
< 00, for every f E O(G).

Therefore, {zv : v E N} cannot be discrete in G. Thus the sequence (zv)

has a cluster point Zo in G. Since K is closed, Zo belongs to K. SO G is
holomorphically convex. _

Sequences. For a domain G c en we wish to construct a sequence that

accumulates at every point of its boundary.
6.4 Theorem. Let (Kv) be a normal exhaustion of G. Then there exists
a strictly monotonic increasing sequence >'(J.L) of natural numbers and a se-
quence (z/A) of points in G such that:
1. z/A E KA(/,)+l - KA(/A) , for every J.L.
2. If Zo is a boundary point of G and U = U(zo) an open connected neigh-
borhood, then every connected component of U n G contains infinitely
many points of the sequence (z/A)'

PROOF: This is a purely topological result, since we make no assumption

about G. The proof is carried out in several steps.
 6. Singular Functions 81

(1) Let B = {B II : v E N} be the countable system of balls with rational

center and rational radius meeting 8G. Every intersection BllnG has at most
countably many connected components. Thus we obtain a countable family

C = {C,.. : 3v E N such that C,.. is a connected component of BII E 8}.

(2) By induction, the sequences )"(J.l) and (z,..) are constructed. Let ZI be
arbitrary in C1 - K 1 • Then there is a unique number ),,(1) such that Zl E
K>.(l)+I - K>'(l)'

Now suppose Zl, ... ,Zj.I-l have been constructed such that

Zj E Cj n (K>'(j)+l - K>.(j))' for j = 1, ... , J.l- 1.

We choose zp. E Cp. - K>.(,..-l)+I and )"(J.l) as usual. That is possible, since
there is a point W E B II (,..) n 8C,.. n 8G if C,.. is a connected component of
B II (,..) n G. Then en - K>'(j.I-l)+I is an open neighborhood of wand contains
points of Cw
(3) Now we show that property (2) of the theorem is satisfied. Let Zo be a
point of 8G, U = U(zo) an open connected neighborhood, and Q a connected
component of Un G. We assume that only finitely many z,.. lie in Q, say
Zl, ... , Zm. Then

are open connected sets that contain no Zw Obviously, Q* is a connected

component of G n U* .
There is a point Wo in U* n 8Q* n 8G, and a ball BII C U" with Wo E B II .
Then BII n G c U* n G. Moreover, BII n G must contain a point Wl E Q*.
The connected component C* of WI in BII n G is a subset of the connected
component of Wl in U* n G. But C* is an element C"'o of C. By construction
it contains the point zj.lo' That is a contradiction. Infinitely many members
of the sequence belong to Q. •

6.5 Theorem. If G is holomorphically convex, then it 'is a domain of holo-

morphy.

PROOF: Let (KII ) be a normal exhaustion of G with ](11 = KII and choose
sequences )"(J.l) EN and (z,..) in G such that zj.I E K>.(,..)+l - K>.(j.I)' We may
assume that for every point Zo E 8G, every open connected neighborhood
U = U(zo), and every connected component Q of Un G there are infinitely
many z,.. in Q.
Now let f be holomorphic in G and unbounded on D := {Zp. : J.l EN}. It is
clear that f is completely singular at every point Zo E 8G. •
82 II. Domains of Holomorphy

Remark. It is not necessary that a completely singular holomorphic func-

tion is unbounded. In 1978, D. Catlin showed in his dissertation that if
Gee en is a holomorphically convex domain with smooth boundary, then
there exists a function holomorphic in G and smooth in a neighborhood of G
that is completely singular at every point of the boundary of G.

Exercises
1. A domain Gee en
is holomorphically convex if and only if for every
z E 8G there is a neighborhood U(z) such that Un G is a domain of
holomorphy.
2. Let G 1 C enand G 2 c em be domains of holomorphy. If f : G 1 -+ em is
a holomorphic mapping, then f- 1 (G 2 ) n G 1 is a domain of holomorphy.
3. Find a bounded holomorphic function on the unit disk 0 that is singular
at every boundary point.

7. Examples and Applications

Domains of Holomorphy
7.1 Proposition. Every domain in the complex plane e is a domain of
holomorphy.

PROOF: e
We have already shown that every domain in is holomorphically
convex. Therefore, such a domain is also a domain of holomorphy. _

7.2 Theorem. The following statements about domains G E en are equiv-

alent:
1. G is a weak domain of holomorphy.
2. G is holomorphically convex.
3. For every infinite discrete subset D C G there exists a holomorphic func-
tion f in G such that If I is unbounded on D.
4. G is a domain of holomorphy.
The equivalences have all been proved in the preceding paragraphs. Fur-
thermore, we know that every domain of holomorphy is pseudoconvex. Still
missing here is the proof of the Levi problem: Every pseudoconvex domain
is holomorphically convex. We say more about this in Chapter V.

Every affine convex open subset of en is a domain of holomorphy. The n-fold

Carte:lian product of plane domains is a further example.
7.3 Proposition. If G b ... ,Gn C e are arbitrary domains, then G :=
G 1 X ... x G n is a domain of holomorphy.
 7. Examples and Applications 83

PROOF: Let D = {Zp. = (zf, ... , zt:) : J1. E N} be an infinite discrete subset
of G. Then there is an i such that (zf) has no cluster point in Gi , and there
I
is a holomorphic function f in G i with liml'-tool f{zf) = 00. The function i
in G, defined by i(Zb"" zn) := f{zi), is holomorphic in G and unbounded
on D. •

Remark. The same proof shows that every Cartesian product of domains
of holomorphy is again a domain of holomorphy.

Complete Reinhardt Domains. Let G c C n be a complete Rein-

hardt domain (see Section I.1). We will give criteria for G to be a domain of
holomorphy. For that purpose we define a map log from the absolute value
space "f/ to IRn by

Definition. A Reinhardt domain G is called logarithmically convex if

10gr{Gn (c*)n) is an affine convex domain in IRn :

Remark. For z = (Zl,"" zn) E G we have logr(z) = (IOgIZll, ... , 10gIZn/).

If z E (C*)n, then IZil > 0 for each i, and logr(z) is in fact an element of ]Rn.

7.4 Proposition. The domain of convergence of a power series S(z) =

EII~o allz ll is logarithmically convex.

PROOF: Let G be the domain of convergence of S(z), and M := 10gr(G n

(c*)n) c IRn. We consider two points x, y E M and points z, w E G n (c*)n
with logr(z) = x and logr(w) = y. If A> 1 is small enough, AZ and AW still
belong to G n (c*)n. Since S(z) is convergent in AZ, AW, there is a constant
C > 0 such that

lalll· AIIII 'Izill :::; C and lalll· AIIII 'Iwlll :::; C, for every 1I E No.
Thus
la,,1 . AI"I . Izilit ·lw"1 1- t :::; C, for every 1I and 0 :::; t :::; 1.
It follows from Abel's lemma that S{z) is convergent in a neighborhood of

Zt:= (lzlltlwill-t, ... , IZnltlwnll-t).

This means that Zt E G and tx + (1 - t)y = logr{zt) E M, for 0 :::; t :::; 1.
Therefore, M is convex. _
84 II. Domains of Holomorphy

7.5 Proposition. Let G be a complete Reinhardt domain. If G is logarith-

mically convex, then it is holomorphically convex.

PROOF: Let K be a relatively compact subset of G. Since G is a complete

Reinhardt domain and K a compact subset of G, there are points Zl, ... ,Zk E
G n (c*)n such that
k
KeG' := U pn(O, qi) C G, where qi := r(zd·
i=l
We consider the set .L = {m(z) = ZV : v E No} of monomials, which is a
subset of O(G). For z E pn(O,qi) and m E.L we have

Im(z)l = Izvl < qf = Im(qi)l·

Let Z := {ZIt ... ,Zk}. Then for z E R it follows that

Im(z)l ~ suplml ~ suplml ~ suplml, for every m E .L.

K G' Z

Suppose that R is not relatively compact in G. Then R has a cluster point

Zo in aG, and it follows that Im(zo)1 ~ supzlml, for every m E .L.
Let h(z) := logr(z), for z E (c*)n. Since G is logarithmically convex, the
domain Go := h(G n (c*)n) c an is affine convex. For the time being we
assume that Zo E (c*)n. Then:xo := h(zo) E aGo, and there is a real linear
function 'x(x) = alXl + ... + anXn such that 'x(x) < 'x(xo) for x EGo.
Let x = logr(z) be a point of Go, and U E an with Uj ~ Xj for j = 1, ... , n.
Then eUj ~ e:l: j = IZjl, and therefore (since G is a complete Reinhardt do-
main) w = (e Ul , ... , eUn ) E G n (c*)n and U E Go. In particular,
'x(x) - naj = 'x(x - nej) < 'x(:xo), for every n E N.
Therefore, aj ~ 0 for j = 1, ... , n.
Now we choose rational numbers Tj > aj and define X(x) := rlXl + .. ·+rnxn.
If we choose the rj sufficiently close to ai, the inequality X(qi) < X(:xo)
holds for i = 1, ... , k, and it still holds after multiplying by the common
denominator of the r j. Therefore, we may assume that the r i are natural
numbers, and we can define a special monomial mo by mo(z) := Z~l ... z~n.
Then
( -)1 --e);(qil < e);("0) -- Imo (z)1
Imoz. 0 , 1:0 • - 1 , ... , k .
~,r~-

So Imo(zo)1 > supzlmol, and this is a contradiction.

If Zo ¢ (C*)'\ then after a permutation of the coordinates we may assume
that zi°) ... zfO) :/= 0 and zf~l = ... = z~) = O. We can project on the space
 7. Examples and Applications 85

e! and work with monomials in the variables ZI, ... , Z!. Then the proof goes
through as above. _
Now we get the following result:
7.6 Theorem. Let G c en be a complete Reinhardt domain. Then the
following statements are equivalent:
1. G is the domain of convergence of a power series.
2. G is logarithmically convex.
3. G is holomorphically convex.
4. G is a domain of holomorphy.

PROOF: We have only to show that if G is a complete Reinhardt domain

and a domain of holomorphy, then it is the domain of convergence of a power
series. By hypothesis, there is a function J that is holomorphic in G and
completely singular at every boundary point. In Section I.5 we proved that
for every holomorphic function in a proper Reinhardt domain there is a power
series S(z) that converges in G to f. By the identity theorem it does not
converge on any domain strictly larger than G. _

Analytic Polyhedra. Let G c en be a domain.

Definition. Let U C G, VI"'" Vk c e open subsets, and !I, ... , Jk
holomorphic functions in G. The set

P:={ZEU: Ji(Z)EVi, fori=l, ... ,k}

is called an analytic polyhedron in G if Pee U.

If, in addition, VI = ... = Vk = 0, then one speaks of a special analytic
polyhedron in G.

Remark. An analytic polyhedron P need not be connected. The set U

in the definition ensures that each union of connected components of P is
also an analytic polyhedron if it has a positive distance from every other
connected component of P.

7.7 Theorem. Every connected analytic polyhedron P in G is a domain of

holomorphy.

PROOF: We have only to show that P is a weak domain of holomorphy.

If Zo E {)P, then there is an i such that Ji(ZO) E {)Vi, Therefore, fCz) .-
(fi(Z) - Ji(ZO))-1 is holomorphic in P and completely singular at Zo0 _
86 II. Domains of Holomorphy

Example
Let q < 1 be a positive real number, and

P:= {z = (ZbZ2) E e 2 : Izd < 1, IZ21 < 1 and IZI' z21 < q}.
Then P (see Figure II.7) is clearly an analytic polyhedron, but neither affine

q
1 1---....-

Figure 11.1. An analytic polyhedron

convex nor a Cartesian product of domains. So the analytic polyhedra enrich

our stock of examples of domains of holomorphy.
We will show that every domain of holomorphy is "almost" an analytic poly-
hedron.
7.8 Theorem. If G c en
is a domain of holomorphy, then there exists
a sequence (P,,) of special analytic polyhedra in G with P" cc P,,+l and
U::l P" = G.
PROOF: Let (K,,) be a normal exhaustion of G with K" = K". If z E
8K,,+1 is an arbitrary point, then z does not lie in Kv C (K"+lt, and
therefore not in K". Hence there exists a function f holomorphic in G for
which q := SUPK If I < If(z)l. By multiplication by a suitable constant we
obtain q < 1 < If(z)l, and then there is an entire neighborhood U = U(z)
such that If I > 1 on U.
Since the boundary 8K"+1 is compact, we can find finitely many open neigh-
borhoods U",j of Z",j E 8K"+1, j = 1, ... , k", and corresponding functions
f",j holomorphic in G such that If",jl > 1 on Uv,j, and 8K,,+1 C U~:l U",j'
We define

P" := {z E (K,,+lt : If",j(z)1 < 1 for j = 1, ... , kv}.

Clearly, Kv C P" c (Kv+1t. Furthermore, M := K,,+l - (U",l U'" U Uv,kJ

is a compact set with P" C M c (K"+1t. Consequently, P" cc K v+1' Thus
P" is a special analytic polyhedron in G. It follows trivially that the sequence
(P,,) exhausts the domain G. •
 8. Riemann Domains over en 87

In the theory of Stein manifolds one proves the converse of this theorem.

Exercises
1. If R is a domain in the real number space lRn, then
TR = R+ ilR n := {z E en : (Re(Zl)' . .. ,Re(Zn» E R}
is called the tube domain associated with R. Prove that the following
properties are equivalent:
(a) R is convex.
(b) TR is (affine) convex.
(c) TR is holomorphically convex.
(d) TR is pseudoconvex.
Hint: To show (d) => (a) choose Xo, Yo E R. Then the function ep«) :=
-lnb"TR(xo+«Yo -xo) is subharmonic in D. Since b"TR(X+iy) = b"R(X),
one concludes that t ~ -lnb"R(xo + t(yO - xo)) assumes its maximum
at t = 0 or t = 1.
2. Let G c en be a domain. A domain G c en is called the envelope 01
holomorphy of G if every holomorphic function 1 in G has a holomorphic
extension to G. Prove:
(a) If R c lR n is a domain and H(R) its affine convex hull, then G :=
H(R) + ilR n is the envelope of holomorphy of the tube domain G =
R+ ilRn.
(b) If G c en is a Reinhardt domain and G the smallest logarithmi-
cally convex complete Reinhardt domain containing G, then Gis the
envelope of holomorphy of G. Hint: Use the convex hull of logr(G).
3. Construct the envelope of holomorphy of the domain
Gq := p2(O, (l,q» U p2(O, (q, 1»).
4. A domain G c en is called a Runge domain if for every holomorphic
function 1 in G there is a sequence (PI') of polynomials converging com-
pactly in G to I. .
Prove that the Cartesian product of n simply connected subdomains of
e is a Runge domain in en.
5. A domain G c en is called polynomially convex if it is convex with
respect to the family of all polynomials (cf. Exercise 5.5). Prove that
every polynomially convex domain is a holomorphically convex Runge
domain.

8. Riemann Domains over en

Riemann Domains. It turns out that for general domains in en the
envelope of holomorphy (cf. Exercise 7.2) cannot exist in en. Therefore, we
have to consider domains covering en.
88 II. Domains of Holomorphy

Definition. A (Riemann) domain over Cn is a pair (G,1I') with the

following properties:
1. G is a connected Hausdorff space. 4
2. 11' : G -+ Cn is a local homeomorphism (that is, for each point x E G
and its "base point" z := 1I'(x) E C n there exist open neighborhoods
U = U(x) c X and V = V(z) C C n such that 11' : U -+ V is a
homeomorphism) .

Remarks
1. Let (G, 11') be a Riemann domain. Then G is pathwise connected, and
the map 11' : G -+ C n is continuous and open. The latter means that the
images of open sets are again open.
2. If (G v , 11'v) are domains over Cn for v = 1, ... ,l, and Xv E G v are points
over the same base point zo, then there are open neighborhoods Uv =
=
Uv(xv) c Gv and a connected open neighborhood V V(zo) C C n such
that 1I'vlu" : Uv -+ V is a homeomorphism for v = 1, ... ,l.

Examples
1. If G is a domain in cn, then (G,idG) is a Riemann domain.
2. The Riemann surface of .,fi. (without the branch point) is the set

G:= {(z,w) EC* xC: w 2 =z}.

If G is provided with the topology induced from C.. x C, then it is a
Hausdorff space. The mapping cp : C.. -+ G defined by ( H «(2, () is
continuous and bijective. Therefore, G is connected. The mapping cp is
called a uniformization of G.
Now let 11' : G -+ C be defined by 1I'(z, w) := z. Clearly, 11' is continuous. If
(zo, wo) EGis an arbitrary point, then zo =I 0, and we can find a simply
connected neighborhood V(zo) C C ... Then there exists a holomorphic
function f in V with j2(z) == z and f(zo) = woo We denote fez) by .,fi..
The image W := f(V) is open, and the set 1I'-1(V) can be written as the
union of two disjoint open sets

U± := {(z, ±f(z)) : z E V} = (V x (±W)) n G.

Let j(z) := (z, fez)). Then i:
V -+ G is continuous, and 1r 0 j(z) == z.
The open set U := U+ is a neighborhood of (zo, wo), with i(v) = U and
i 0 1I'(z, w) = (z, w) on U; that is, 1I'lu : U -+ V is topological. Hence

(G, 11') is a Riemann domain over C.

4 A general topological space X is said to be connected if it is not the union of
two disjoint nonempty open sets. A space X is called pathwise connected if each
two points of X can be joined by a continuous path in X. For open sets in en
these two notions are equivalent.
 8. Riemann Domains over en 89

The space G can be visualized in the following manner: We cover e with

two additional copies of e, cut both these "sheets" along the positive
real axis, and paste them crosswise to one another (this is not possible
i~ 1R 3 without self intersection, but in higher dimensions, it is). This is
illustrated in Figure II.S.

Figure II.S. The Riemann surface of Vi

8.1 Proposition (on the uniqueness oflifting). Let (G, 11') be a domain
over en and Y a connected topological space. Let Yo E Y be a point and
'l/Jl,'l/J2 : Y ~ G continuous mappings with 'l/Jl(YO) = 'l/J2(YO) and 11' 0 'l/Jl =
11' 0 'l/J2. Then 'l/Jl = 'l/J2'

PROOF: Let M := {y E Y : 'l/Jl(Y) = 'l/J2(Y)}' By assumption, Yo E M,

so M =I 0. Since G is a Hausdorff space, it follows immediately that M is
closed. Now let Y E Y be chosen arbitrarily, and set x := tPl (y) = tP2(Y) and
z := lI'{x). There are open neighborhoods U = U{x) c G and V = V(z) c
en such that 11' : U ~ V is topological, and there is an open neigborhood
W = W(y) with 'I/J>.{W) c U for ,\ = 1,2. Then

'l/Jllw = (lI'IU)-l 011' 0 'l/Jllw = (lI'IU)-l 011' 0 'l/J21w = tP2lw,

and therefore W c M. Hence M is open, and since Y is connected, it follows
that M = Y. •

Definition. Let zo E en be fixed. A (Riemann) domain over en with

distinguished point is a triple 9 = (G, 11', xo) for which:
1. (G,lI') is a domain over en.
2. Xo is a point of G with lI'(xo) := Zoo
90 II. Domains of Holomorphy

Definition. Let 9j = (G j ,7rj,Xj) be domains over en

with distin-
guished point. We say that {h is contained in 92 (denoted by {h -< 92 )
if there is a continuous map cp : G1 -+ G 2 with the following properties:
1. 7r2 0 cp = 7rl (called "cp preserves fibers").
2. cp(xt} = X2·

8.2 Proposition. If 91 -< 92, then the fiber preserving map cp : G 1 -+ G 2

with CP(X1) = X2 is uniquely determined.
This follows immediately from the uniqueness of lifting.
8.3 Proposition. The relation "-<" is a weak ordering; that is:
1. 9 -< 9.
2. 91 -< 92 and 92 -< 93 :=} 91 -< 93·
The proof is trivial.

Definition. Two domains 91, 92 over en

with fundamental point are
called isomorphic or equivalent (symbolically 91 ~ 92) if 91 -< 92 and
92 -< 91.

8.4 Proposition. Two domains 9j = (Gj , 7rj,Xj), j = 1,2, are isomorphic

if and only if there exists a topological 5 fiber preserving map cp : G1 -+ G 2
with cp(xt} = X2·

PROOF: If we have fiber preserving mappings CPl : G 1 -+ G2 and CP2 : G 2 -+

G}, with CP1(X1) = X2 and CP2(X2) = Xl, it follows easily from the uniqueness
of fiber preserving maps that CP2 0 CPl = idOl and CPl 0 CP2 = idG2. The other
direction of the proof is trivial. _

Definition. A domain 9 = (G, 7r, xo) with 7r( xo) = Zo is called schlicht
if it is isomorphic to a domain 90 = (Go, idoo, zo) with c Go en.

8.5 Proposition. Let 9j = (G j , idG; ,Xj), j = 1,2, be two schlicht domains

with G1 , G 2 c en. Then 91 -< 92 if and only if G 1 C G 2 ·

Example
5 Recall that a "topological map" is a homeomorphism!
 8. Riemann Domains over en 91

Let G l := {(Z,W) E C 2 : W2 = zandz =f O} and 71'l(Z,W):= z. Then

Ql = (Gr,7I'1,(1, 1)) is the Riemann surface of..;z, with distinguished point
(1,1). The domain Ql is contained in the schlicht domain Q2 = (C,ide, 1), by
'P(z, w) := z. But the two domains are not isomorphic.

Union of Riemann Domains. We begin with the definition of the

union of two Riemann domains. Let Qj = (G j , 71'j, Xj), j = 1,2, be two Rie-
mann domains over cn with distinguished point, and Zo := 71'l(xd = 71'2(X2).
We want to glue G l , G 2 in such a way that Xl and X2 will also be glued.
To get a rough idea of the construction, assume that we already have a
Riemann domain Q = (G, 71', xo) that is in some sense the union of Ql and
02' Then there should exist continuous fiber preserving maps 'PI : G l -t G
with 'Pl(Xl) = xo, and <P2 : G 2 -t G with 'P2(X2) := Xo. If a : [0,1] -t G l
and {J : [0,1] -t G 2 are two continuous paths with 0'(0) = Xl, {J(O) = X2 and
71'1 0 a = 71'2 0 {J, then 1'1 := <PI 0 a and 1'2 := 'P2 0 {J are continuous paths in G
with 71' 0 1'1 = 71' 0 1'2 and 1'1 (0) = 1'2(0) = Xo. Due to the uniqueness of lifting,
it follows that 1'1 = 1'2. This means that aCt) and {J(t) have to be glued for
every t E [0,1]. Unfortunately, this is an ambiguous rule. For example, we
could say that X E G l and Y E G 2 have to be glued if 71'1 (X) = 71'2(Y). Then
the desired property is fulfilled, but it may be that there are no paths a from
Xl to X and {J from X2 to y with 71'1 0 a = 71'2 0 {J.

Therefore, we proceed in the following way: Start with the disjoint union
G l UG 2 , and take the "finest" equivalence relation'" on this set with the
following property:
1. Xl '" X2.
2. If there are continuous paths a : [0,1] -t G l and {J : [0,1] -t G 2 with
0'(0) = Xl, {J(O) = X2, and 71'1 0 0'= 71'2 0 {J, then 0'(1) '" {J(1).

One can equip G := (G l U G 2 )/ '" with the structure of a Riemann domain.

This will now be carried out in a more general context.

Let X be an arbitrary set. An equivalence relation on X is given by a partition

!!C = {Xv: v E N} of X into subsets with:
1. UVEN XII = X.
2. XII n XJL = 0 for v =f J1..

The sets Xv are the equivalence classes.

Now let a family (3r;,)"El of equivalence'relations on X be given with 3r;, =
{X~, : v" EN"} for LEI. We set N := TI"El No and

Xv := nX~"
"El
for v := (VJ"El EN.
92 II. Domains of Holomorphy

Then &: = {Xv: l/ E N} is again an equivalence relation (simple exercise),

and it is finer than any ~. This means that for every LEI and every v E N,
there is a v. EN. with Xv eXt,.

We apply this to the disjoint union X = UAEL G A , for a given family (QA)AEL
of Riemann domains gA = (G A' 7r A, X A) over en with distinguished point. An
equivalence relation on X is said to have property (P) if the following hold:
1. X A '" XII' for A, (} E L.
2. If 0 : [O,lJ -* G A and {3 : [O,lJ -* G g are continuous paths with 0(0) '"
{3(0) and 7rA 00 = 7rg 0 {3, then 0(1) '" {3(1).
We consider the family of all equivalence relations on X with property (P).
It is not empty, as seen above in the case of two domains. Therefore we can
construct an equivalence relation (as above) that is finer than any equivalence
relation with property (P). We denote it by "'p. It is clear that 7r A(x) = 7r g (Y)
if X EGA' yEGg, and X "'P y. The relation "'P also has property (P), and
the elements of an equivalence class Xv all lie over the same point z = z(Xv).
We define G := X! "'P and 7f(Xv) := z(Xv). The equivalence class of all X A
will be denoted by x.
8.6 Lemma. Let y E G A and X E Gil be given with 7rg (x) = 7rA(Y) =: z. If
we choose open neighborhoods U = U(y) eGA' V = Vex) eGg, and an open
connected neighborhood W = W (z) such that 7r A : U -* Wand 7r11 : V -* W
are topological mappings, then for cP := (7r g IV)-1 07r A : U -* V the following
hold:
1. cp(y) = x.
2. If x '" p y, then cp(y') '" p y' for every y' E u.

PROOF: The first statement is trivial. Now let 0 : [O,lJ -* W be a con-

tinuous path with 0(0) = z and 0(1) = 7r A(y') for some y' E U. Then
{3 := (7r>.IU)-1 00 and "I := cp 0 {3 are continuous paths in U and V with
{3(0) = y "'p x = cp(y) = "1(0) and 7rA 0 {3 = 7r g 0 cp 0 {3 = 7r g 0 'Y. Therefore,
y' = {3(1) "'p 'Y(1) = cp(y'). •

8.1 Theorem. There is a topology on G such that

g:= (G,7f,x)
is a Ri~mann domain over en
with distinguished point X, and all maps CPA :
G A -* G with
CPA(X) := equivalence class of x
are continuous and fiber preserving.
 8. Riemann Domains over C n 93

PROOF: (1) Sets of the form cp,\(M) for M open in G,\ together with G
constitute a base of a topology for G. To see this it remains to show that the
intersection of two such sets is again of this form.
Let McG,\ and NeG (! be open subsets. Then

cp,\(M) n cp(!(N) = CPo(N n cp;l (cp,\ (M))).

But cp;l(cp'\(M)) is open in G(}. In fact, let x E cp;l(cp'\(M)) be given, and
Y E M be chosen such that cp,\(y) = CPo(x) (and therefore y I'Vp x). Let z :=
1I"'\(Y) = 1I"(}(x). Then there exist open neighborhoods U = U(y) and V = V(x)
and an open connected neighborhood W = W(z) such that 11",\ : U ~ Wand
V ~ W are topological mappings. Let cp:= (1I"oIV)-1 011",\: U ~ V. By
11"(}:
the lemma, cp(y) = x and cp(y') "'P y' for every y' E U.
So V' := cp(MnU) is a neighborhood of x in Go, and since cp,,(cp(y')) = cp),(y')
for every y' E U, it follows that V' c cp;l(cp,\(M».
Consequently, every CP)' is a continuous map.
(2) Remark: Since every y EGis an equivalence class cp),(x), we have

M = Ucp,\(cp~l(M» for any subset MeG.

),EL

(3) 1r : G ~ en is continuous: Let V c en be an arbitrary open set, and

M := 7f-l(V). Then cp~l(M) = 1I"~I(V) is open in G)" and therefore M =
U),EL cp,\(cp~I(M)) is open in G.

(4) G is a Hausdorff space: Let Yt. Y2 E G with Yl :f: Y2, and Zl := 7f(yt},
Z2 := 7f(Y2)'
There are two cases. If Zl :f: Z2, then there are open neighborhoods V1(Zl)
and V2(Z2) with VI n V2 = 0. Then 7f- I (Vl) and 7f- I (V2) are disjoint open
neighborhoods of YI and Y2. If Zl = Z2, then we choose elements Xl E G,\,
X2 E G(} with Cp,\(XI) = YI and cp(}(X2) = Y2, and since Xl and X2 are not
equivalent, the above lemma implies that there are disjoint neigborhoods of
YI and Y2·
=
(5) G is connected: Let Y cp,\(x) be an arbitrary point of G. Then there is
a continuous path Q : [0, 1] ~ G). that connects the distinguished point x).
to x. Then CP)' 0 Q connects x to y.
(6) 7f is locally topological: Let Y = cp).(x) be a point of G, and z = 7f(y) =
Then there exist open neighborhoods U = U(x) c G). and W =
11"). (x).
W(z) c en such that 11"), : U -+ W is a topological mapping. if := cp).(U)
is an open neighborhood of y, with 7f(if) = 1I"),(U) = W. In addition, 7fl u is
injective, since 1r 0 cp). = 7r). and 1I").lu is injective.
94 II. Domains of Holomorphy

(7) Clearly, the maps cp).. : G).. -+ aare fiber preserving, and it was already
shown that they are continuous. _

Now 9 has the following properties:

1. {h -< 9, for every A E L.
2. If 9* is a domain over en with 9).. -< 9* for every A, then 9 -< 9*.
PROOF: (of the second statement)
If 9* is given, then there exist fiber preserving mappings cpl : G).. -+ G*. We
introduce a new equivalence relation ~ on the disjoint union X of the G).. by

x ~ x' : ~ x E G).., x' E G u and cpl(x) = cp~{X').

It follows from the uniqueness of lifting that ~ has the property (P). Now we
a
define a map cp : -+ G* by

cp(cp)..{X)) := cpl{x).
Since'" p is the finest equivalence relation with property (P), cp is well defined.
Also it is clear that cp is continuous and fiber preserving. _
Therefore 9 is the smallest Riemann domain over en that contains all do-
mains 9)...
Definition. The domain 9 constructed as above is called the union of
the domains 9).., and we write 9 = U)..EL 9)...

Special cases:
1. From 91 -< 9 and 92 -< 9 it follows that 91 u 92 -< 9 ..
2. From 91 -< 92 it follows that 91 u 92 ~ 92.
3. 9u9~9.
4. 91 U 92 ~ 92 U 91.
5. 91 U (92 U 93) ~ (91 U 92) U 93'

Example
Let ~h = (G b 7r1, xd be the Riemann surface of .jZ with distinguished point
Xl = (1, 1) and 92 = (G2' id, X2) the schlicht domain

G2 = { Z E e : ~ < Izl < 2 }

with distinguished point X2 = 1.
Then 91 U 92 = (a, 7r, xo), where a = (G l UG2)/ "'p.
Let Y E 7r11 (G 2 ) C G 1 · Then we can connect y to the point Xl by a path
a in 7r11 (G 2 ), and 7rl{Y) to X2 by the path 7rl 0 a in G 2 . But Xl "'P X2, so
 8. Riemann Domains over en 95

Y P 11"1 (y) as well. This shows that over each point of G 2 there is exactly
I'V

one equivalence class.

Now let Z E e - {O} be arbitrary. The line through z and 0 in e contains a
segment Q : [0,1) -+ e* that connects z to a point z· E G2. There are two
paths QllQ2 in G l with 11"1 OQl = 11"1 OQ2 = Q. Since Ql(l) "'P Q2(1), it
follows that Ql(O) "'p Q2(O).
Then it follows that {h u g2 = (e - {O}, id, 1).

Exercises
1. For t = (t l , ... , t n) E "f/ define iP t : en -+ en by
.= (itl
n.. ( Zl,.··, Zn ).
'J.'t e Zl, ... , e itn)
Zn.

A Riemann domain g = (G, 11", xo) is called a Reinhardt domain over

en if 1I"(xo) = 0 and for every t E "f/ - (C'")n there is an isomorphism
CPt : g -+ g with 11" 0 CPt = iP t 0 11". Prove:
(a) If G c en is a proper Reinhardt domain, then g = (G, id, 0) is a
Reinhardt domain over en.
(b) Let G ll G2 c e 2 be defined by

Gl .- p2(0, 1) - {(z,w) : Izi = ~ and Iwl S ~},

G2 .- {(z,w) E p2(0, 1) : Iwl < ~}.
Gluing G l and G2 along {(z, w) : ~ < Izl < 1 and Iwl < ~} one
obtains a Riemann domain over e 2 that is a Reinhardt domain over
e 2 , but not schlicht. Show that this domain can be obtained as the
union of gl = (G
l , id, (~,~)) and g2 = 2, id, (~,~»). (G
2. Let J = {O, 1,2,3, ... } c No be a finite or infinite sequence of natural
numbers and Pi = pn (Zi' ri), i E J, a sequence of polydisks in en.
Assume that for every pair (i,j) E J x J an "incidence number" Cij E
{O, I} is given such that the following hold:
(a) Cij = Cji and Cii = 1.
(b) Cij = 0 if Pi n Pj = 0.
(c) For every i > 0 in J there is a j < i with Cij = 1.
(d) If Pi n P j n Pk # 0 and Cij = 1, then Cik = Cjk.
Points Z E Pi and w E Pj are called equivalent (z '" w) if z = wand
Cij = 1. Prove that G := U Pi / '" carries in a natural way the structure
of a Riemann domain over en.
Let 11" : G -t en be the canonical projection and suppose that there is a
n
point Zo E iEJ Pi· Is there a point Xo E G such that (G, 11", xo) can be
written as the union of the Riemann domains (Pj , id, zo)?
96 II. Domains of Holomorphy

9. The Envelope of Holomorphy

Holomorphy on Riemann Domains
Definition. Let (G, 7r) be a domain over en. A function I : G -+ e
is called holomorphic at a point x E G if there are open neighborhoods
U = U(x) c G and V = V(7r{x» C en such that 7rlv : U -+ V is
topological and 1 0 (7rlv)-1 : V -+ e is holomorphic. The function I is
called holomorphic on G if I is holomorphic at every point x E G.

Remark. A holomorphic function is always continuous. For schlicht do-

mains in en the new notion of holomorphy agrees with the old one.

Definition. Let (ij = (G j ,7rj,Xj), j = 1,2, be domains over en with

distinguished point, and (il -< (i2 by virtue of a continuous mapping
<.p : G l -+ G 2 . For every function I on G 2 we define IIG 1 := 1 0 <.p.

9.1 Proposition. II I: G 2 -+ e is holomorphic and (il -< 92, then IIG1 is

holomorphic on G I .

PROOF: Trivial, since <.p is a local homeomorphism with 11"2 0 <.p = 7rl. •

Definition.
1. Let (G, 7r) be a domain over en and x EGa point. If I is a holo-
morphic function defined near x, then the pair (I, x) is called a local
holomorphic lunction at x.
2. Let (Gl,7rI), (G2,7r2) be domains over en, and Xl E G I , X2 E G 2
points with 7rl {xt} = 7r2(X2) =: z. Two local holomorphic functions
(II, Xl)' (12, X2) are called equivalent if there exist open neighbor-
hoods Ul(Xl) C G l , U2(X2) C G2, V(z), and topological mappings
7rl : UI -+ V, 7T2: U2 -+ V with II 0 (7rllvJ-l = h 0 (7r2Iu2)-1.
3. The equivalence class of a local holomorphic function (I, x) is denoted
by Ix.

Remark. If (II)X1 = (h)x2' then clearly, II(XI) = h(x2)' In particular,

if Gl = G 2 , 7Tl = 7r2, and Xl = X2, then it follows that II and 12 coincide in
an open neighborhood of Xl = X2.
 9. The Envelope of Holomorphy 97

9.2 Proposition. Let (Gl. 7rd, (G2,7r2) be domains over en, a,\ : [0, 1] ~
G,\ continuous paths with 7r} oat = 7r2oa2. Additionally, let f'\ be holomorphic
on G,\, for.x = 1,2. If (ft)OII(O) = (12) 01 2(0), then also UdOl I (1) = (12) 01 2(1)'

°
PROOF: Let M := {t E [O,IJ : UdOI(t) = (h)o2(t)}. Then M",
E M. It is easy to see that M is open and closed in [0,1], because of the
identity theorem for holomorphic functions. So M = [0,1].
0, since

9.3 Proposition. Let 9j = (Gj ,7rj,Xj), j = 1,2, be domains over en with

distinguished point, and 91 -< 92. Then for every holomorphic function f on
G 1 there is at most one holomorphic function F on G 2 with FIG I = f, i.e.,
a possible holomorphic extension of f is uniquely determined.

PROOF: Let F}, F2 be holomorphic extensions of f to G 2. We choose neigh-

borhoods U,\(x,\) c G,\ such that the given fiber-preserving map cp : G 1 ~ G z
maps U} topologically onto U2. We have Fj °cplul = flu l , for j = 1,2, and
therefore F11u2 = F21u2. It follows that (Fdx2 = (F2)X2' Since each point of
G 2 can be joined to X2, the equality FI = F2 follows. •

Envelopes of Holomorphy
Definition. Let 9 = (G, 7r, xo) be a domain over en
with distinguished
point and § a nonempty set of holomorphic functions on G.
Let (Q,\hEL be the system of all domains over en with the following
properties:
1. 9 -< 9'\ for every .x E L.
2. For every f E § and every .x E L there is a holomorphic function
F,\ on G,\ with F'\IG = f.
Then H~(Q) := U'\EL 9'\ is called the § -hull of 9.
If $ = a(G) is the set of all holomorphic functions on G, then H(9) :=
HO(G)(Q) is called the envelope of holomorphy of g. If $ = {J} for
some holomorphic function f on G, then H,(9) := H{!}(Q) is called the
domain of existence of the function f.

9.4 Theorem. Let 9 = (G,7r,xo) be a domain over en, $ a nonempty set

of holomorphic functions on G, and H~(Q) = (G,rr,Io) the §-hull. Then
the following hold:
1. 9 -< H~(Q).
2. For each function f E § there exists exactly one holomorphic function
F on G with FIG = f.
98 II. Domains of Holomorphy

3. /191 = (G 1,71"1, xd is a domain over en such that 9 -< 91 and every lunc-
tion lEg: can be holomorphically extended to G 1 , then 91 -< H§(Q).

PROOF: H$(Q) is the union of all Riemann domains 9>. = (G>., 7f>., x>.) to
which each function lEg: can be extended. We have fiber-preserving maps
i.p>. : G -t G>. and CP>. : G>. -t C.

Let "'P be the finest equivalence relation on X := U>'EL G>. with property
(p).6 Then C is the set of equivalence classes in X relative to "'p. We define
a new equivalence relation ~ on X by

x ~ x' : ~ x E G>., x' E G/J' 7f>.(x) = 7f/J(x' ), and for each IE:F
and its holomorphic extensions F 1 , F2 on G>., respectively G/J'
we have (F>.)x = (F/J)x l •

Then ~ has property (P):

(i)For any A we can find open neighborhoods U = U(xo), V = V(x>.), and
W = W(7f(xo» such that all mappings in the following commutative dia-
gram are homeomorphisms:

Then for lEg: and its holomorphic extension F>. on G>. we have that
F>. 0 (7f>.IV)-l = F>. 0 i.p>. 0 (7fIU)-l = 10 (7fIU)-l is independent of A.
Therefore, all distinguished points x>. are equivalent.
(iiJf a : [0,1] -t G>. and {3 : [0,1] -t G /J are continuous paths with a(O) :::= {3(O)
and 7f>. 0 a = 7fe 0 {3, then (F>.)o(o) = (Fe )!3(o), It follows that (F>')o(l) =
(Fe )!3(l) as well, and therefore a(l) ~ {3(1).
Since 9 -< 9>. and 9>. -< H§(Q), it follows that 9 -< H:F(Q). Furthermore, the
fiber preserving map cp := CP>. 0 i.p>. does not depend on A.
Now let a function lEg: be given. We construct a holomorphic extension
F on C as follows:
If Y E C is an arbitrary point, then there is a A ELand a point V>. E G >.
such that y = CP>.(y>.), and we define

If y = CP/J(Ye) as well, then V>. "'p Y/J, and therefore V>. ~ Yi' as well. It follows
that F>.(y>.) = Fi'(Ye), and F is well defined.
6 For the definition of property (P) see page 92.
 9. The Envelope of Holomorphy 99

We have F 0 cp = F 0 cp).. 0 '{J).. = F).. 0 '{J).. = f on G. This shows that F is

an extension of f, and from the equation F 0 cp).. = F>. it follows that F is
holomorphic (since cp>. is locally topological).
The maximality of H.?(9) follows by construction. •
The $-hull H$(9) is therefore the largest domain into which all functions
f E $ can be holomorphically extended.
9.5 Identity theorem. Let 9j = (Gj , 7rj, Xj), j = 1,2, be domains over
en, and 9 = (G, Jr, x) the union of 91 and 92. Let Ii : G j -t e be holomor-
phic functions and 9 = (G, 7r, x) a domain with 9 ~ 9j for j = 1, 2 such that
hlo = 1210. Then there is a holomorphic function 1 on G with 110; = Ii,
for j = 1,2.

PROOF: Let f := hlo = 1210, and $ := {fl· Since 91 ~ H$(Q) and

92~ H$(9), it follows that 91 u 92 ~ H$(Q).

Let lbe a holomorphic extension of f to 8 (where H$(Q) = (8,1i',x), and

1:= l1 G· Then
(110;) 10 = 110 = (JIG) 10 = flo = f·
Therefore, 110; is a holomorphic extension of f to G j • Due to the uniqueness
of holomorphic extension, lIo;
= Ii for j = 1,2. •

Pseudo convexity. Let pn C en be the unit polydisk, (pn, H) a Eu-

clidean Hartogs figure, and <P : pn -t en an injective holomorphic map-
ping. Then (<p(pn), <P(H» is a generalized Hartogs figure. l' = (pn, <P, 0) and
1£ = (H, <P, 0) are Riemann domains with 1£ ~ 1'. We regard the pair (1',1£)
as a generalized Hartogs figure.
9.6 Proposition. Let (G, 7r) be a domain over en, (1',1£) a generalized
Hartogs figure, and Xo EGa point for which 1£ -< 9 := (G, 7r, xo).
Then every holomorphic function f on G can be extended holomorphically to
9U'P.
The proposition follows immediately from the identity theorem.
Definition. A domain (G,1I") over en is called Hartogs convex if the
fact that (1', 1i) is a generalized Hartogs figure and Xo EGa point with
1£ -< 9 := (G, 7r, xo) implies 9 U l' ~ 9.
A domain 9 = (G, 7r, xo) over en is called a domain of holomorphy ifthere
exists a holomorphic function f on G such that its domain of existence
is equal to 9.
100 II. Domains of Holomorphy

Remark. If G c en is a schlicht domain, then the new definition agrees

with the old one.

9.7 Theorem.
1. If 9 = (G, 11", xo) is a domain over en and $ a non empty set of holo-
morphic functions on G, then H,(9) is Hartogs convex.
2. Every domain of holomorphy is Hartogs convex.

PROOF: Let (P,1I.) be a generalized Hartogs figure with 11. -< H,(9). Then
every function f E $ has a holomorphic continuation to H,(9) UP. There-
fore, H,(Q) UP -< H,(9). On the other hand, we also have H ~(9) -<
H,(9) UP. So H$(9) U P ~ H$(9). •
A Riemann domain (G, 11") is called holomorphically convex if for every infinite
discrete subset D C G there exists a holomorphic function f on G that is
unbounded on D.
9.8 Theorem (Oka, 1953). If a Riemann domain (G, 11") is Hartogs pseu-
doconvex, it is holomorphically convex (and therefore a domain of holomor-
phy).
This is the solution of Levi's problem for Riemann domains over en. We
cannot give the proof here. .
It seems possible to construct the holomorphic hull by adjoining Hartogs
figures (cf. H. Langmaak, [La60]). It is conceivable that such a construction
may be realized with the help of a computer, but until now (spring 2002) no
successful attempt is known. We assume that parallel computer methods are
necessary.

Boundary Points. In the literature other notions of pseudoconvexity

are used. We want to give a rough idea of these methods.
Definition. Let X be a topological space. A filter (basis) on X is a
nonempty set 'R of subsets of X with the following properties:
1. 0 ¢ 'R.
2. The intersection of two elements of 'R contains again an element of
the set 'R.

Example
1. If Xo is a point of X, then every fundamental system of neighborhoods
of Xo in X is a filter, called a neighborhood filter of Xo.
 9. The Envelope of Holomorphy 101

2. Let (xn) be a sequence of points of X. If we define SN := {xn n ~

N}, then 'R := {SN : N E N} is the so-called elementary filter of the
sequence (x n ). A filter is therefore the generalization of a sequence.

Definition. A point Xo E X is called a cluster point of the filter 'R if

Xo E A, for every A E 'R.. The point Xo is called a limit of the filter 'R if
every element of a fundamental system of neighborhoods of Xo contains
an element of 'R.

For sequences the new notions agree with the old ones.
If f : X ~ Y is a continuous map, then the image of any filter on X is a
filter on Y, the so-called direct image.
Definition. Let (G, 11") be a Riemann domain over c;n. An accessible
boundary point of (G, 11-) is a filter 'R on G with the following properties:
1. 'R. has no cluster point in G.
2. The direct image 1f('R.) has a limit Zo E c;n.
3. For every connected open neighborhood V = V(zo) C c;n there is
exactly one connected component of 1f- 1(V) that belongs to 'R.
4. For every element U E 'R there is a neighborhood V = V(zo) such
that U is a connected component of 1f-1(V).

Remark. For a Hausdorff space X the following hold:

1. A filter in X has at most one limit.
2. If a filter in X has the limit xo, then Xo is the only cluster point of this
filter.
(for a proof see Bourbaki, [Bou66J, §8.1)
Therefore, the limit Zo in the definition above is uniquely determined.
There is an equivalent description of accessible boundary points that avoids
the filter concept. For this consider sequences (X.,) of points of G with the
following properties:
1. (X.,) has no cluster point in G.
2. The sequence of the images 1f(X.,) has a limit Zo E en.
3. For every connected open neighborhood V = V(zo) C c;n there is an
no E N such that for n, m ~ no the points Xn and Xm can be joined by a
continuous path 0: : [0, 1J -+ G with 1f 0 0:([0, 1]) C V.
Two such sequences (X.,), (y.,) are called equivalent if:
1. lim.,-too 7r(X.,) = lim.,-too 1f(Y.,) = Zoo
2. For every connected open neighborhood V = V(zo) there is an no su<:h
that for n, m ~ no the points Xn and Ym can be joined by a continuous
path a : [O,lJ -+ G with 1f 0 aerO, 1]) C V.
102 II. Domains of Holomorphy

An accessible boundary point is an equivalence class of such sequences.

Let 8G be the set of all accessible boundary points of G. Even if G is schlicht,
this set may be different from the topological boundary aG. There may be
points in aG that are not accessible, and it may be happen that an accessible
boundary point is the limit of two inequivalent sequences.
We define G := GU8G. If ro = [xnl is an accessible boundary point, we define
a neighborhood of ro in G as follows: Take a connected open set U c G
such that almost all Xn lie in U and 7r(U) is contained in a neighborhood
of Zo := limn-too 7r(xn). Then add all boundary points r = [Ynl such that
almost all Yn lie in U and limn-too 7r(Yn) is a cluster point of 7r(U). With this
neighborhood definition G becomes a Hausdorff space, and *:
G -+ en with
7r(X) ifxEG,
{
*(x) := lim 7r(xn) if x = [xnl E 8G,
n-too

is a continuous mapping.

Definition. A boundary point r E 8G is called removable if there is

a connected open neighborhood U = U(r) C G such that (U, *) is a
schlicht Riemann domain over en and 8G n U is locally contained in a
proper analytic subset of U.
A subset M C 8G is called thin if for every ro E M there is an open
neighborhood U = U(ro) C G and a nowhere identically vanishing holo-
morphic function f on U n G such that for every rEM n U there exists
a sequence (xn) in Un G converging to r such that limn--->oo f(xn) = O.

Example
Let G c enbe a (schlicht) domain and A eGa nowhere dense analytic
subset. Then every point of A is a removable boundary point of G' := G - A.
The points of the boundary of the hyperball Br(O) c en are all not removable.
Let B be a ball in the affine hyperplane H = {(zo, ... , zn) E en+! : Zo = I},
and G c e n + 1 - {O} the cone over B. Then every boundary point of G is
not removable, since locally the boundary has real dimension 2n + 1. The set
M := {O} is thin in the boundary, as is seen by choosing f(zo, . .. , zn) := Zoo

Analytic Disks. Let (G,7r) be a Riemann domain over en.

If rp: D -+ G
* (*
is a continuous mapping, 0 rp : D -+ en holomorphic, and 0 rp)' «() # 0 for
( E D, then S := rp(D) is called an analytic disk in G. The set bS := rp(oO)
is called its boundary.
 9. The Envelope of Holomorphy 103

Let I := [0,1] be the unit interval. A family (St)tEI of analytic disks cpt(D)
in G is called continuous if the mapping (z, t) f-t CPt(z) is continuous. It is
called distinguished if St C G for 0 ~ t < 1 and bSt C G for 0 ~ t ~ 1.
Definition. The domain G is called pseudoconvex if for every distin-
guished continuous family (8t )tEI of analytic disks in G it follows that
8 1 C G.
The domain G is called pseudoconvex at r E 8G if there is a neighborhood
U = U (r) C G and an c > 0 such that for every distinguished continuous
family (St)tEI of analytic disks in G with *(St) C BE(*(r)) it follows
that St n U c G for tEl.

As in C n one can show that a Riemann domain is pseudoconvex if and only

if it is Hartogs pseudoconvex.
9.9 Theorem (Oka). A Riemann domain (G,7T) is pseudoconvex if and
only if it is pseudoconvex at every point r E 8G.

9.10 Corollary. If (G, 7T) is a domain of holomorphy, then G is pseudo-

convex at every accessible boundary point.
The converse theorem is Oka's solution of Levi's problem.
Finally, we mention the following result:
9.11 Theorem. Let (G,7I") be a Riemann domain over cn, and M C 8G
a thin set of nonremovable boundary points. If G is pseudoconvex at every
point of 8G - M, then G is pseudoconvex.

PROOF: See [GrRe56J, §3, Satz 4. •

Exercises
1. Prove that a Reinhardt domain 9 over C n must be schlicht if it is a
domain of holomorphy.
2. Prove that if (G,7T) is a Reinhardt domain, then for every J E O(G)
there is a power series 8(z) at the origin such that f(x) = S(7T(X» for
xE G.
3. Prove that the envelope of holomorphy of a Reinhardt domain is again a
Reinhardt domain.
4. Prove that the Riemann surface of the function J(z) = log(z) has just
one boundary point over 0 E C.
5. Find a schlicht Riemann domain in C 2 whose envelope of holomorphy is
not schlicht.
6. Construct a Riemann domain 9 = (G, 7T, xo) over C 2 such that for all
X,Y E 7T- 1 (7T(XO» and every J E O(G) the equality J(x) = J(y) holds.
104 II. Domains of Holomorphy

7. Let (G,7I') be a Riemann domain and (G,7r) its envelope of holomorphy.

If f is a holomorphic function on G and Fits holomorphic extension to
G, then f(G) = F(C).
8. Consider

G .- {(z, w) : ~ < Izi < 1, Iwl < 1 }

- {(reit,w) : ~ < r < 1, t > 0 and lwl = _t_}.
2 - l+t
Determine the envelope of holomorphy of G.
9. Let G c en be a domain and p : G -+ lit a plurisubharmonic function.
If Zo is an accessible boundary point of B := {z E G : p(z) < c} cc G,
then B is pseudoconvex at Zo, in the sense of the last paragraph.
Chapter III

Analytic Sets

1. The Algebra of Power Series

The Banach Algebra B t . In this chapter we shall deal more exten-
sively with power series in C n • Our objective is to find a division algorithm
for power series that will facilitate our investigation of analytic sets.
We denote by C[z] the ring of formal power series 2:v>o avz v about the
origin. Let R+ be the set of n-tuples of positive real numbers.

Definition. Let t = (tl, ... , tn) E R+ and f = 2:v>o avz v E C[zl We

define the "number" IIflit by -

if this series converges,

IIflit := { ~v2:olav/t" otherwise.

Let B t := {J E C[z] : IIflit < oo}.

Remark. One can introduce a weak ordering on R+ if one defines

(tl, ... ,tn)S(t~, ... ,t~):¢::=:> t i St;fori=I, ... ,n.

For fixed f, the function t H IIflit is monotone: If t S t*, then IIflit S IIflit o •

Definition. A set B is called a complex Banach algebra if the following

conditions are satisfied:
1. There are operations

+ :B xB ~ B, .: C x B ~ Band 0 : B x B ~ B

such that
(a) (B, +, .) is a complex vector space,
(b) (B, +, 0) is a commutative ring with 1,
(c) c· (J 0 g) = (c . J) 0 9 = f 0 (c . g) for all f, 9 E Band c E C.
2. To every fEB a real number Ilfll ~ 0 is assigned that has the
properties of a norm:
(a) lie· fll = Icl·llfll, for c E C and fEB,
(b) Ilf + gil S Ilfll + IlglI, for f, 9 E B,
106 III. Analytic Sets

(c) IIfll = 0 ~ f = O.
3. IIf 0 gil :s: IIfll· IIgll. for f, g E B.
4. B is complete; i.e., every sequence in B that is Cauchy with respect
to the norm has a limit in B.

1.1 Theorem. B t = {J E C[z~ : IIflit < oo} is a complex Banach algebra

for any t E 1R+.

PROOF: Clearly, C[z] is a commutative C-algebra with 1. Straightforward

calculations show that II ... lIt satisfies the properties (2a), (2b), (2c) and (3). It
follows that B t is closed under the algebraic operations, and all that remains
to be shown is completeness.
Let (1),) be a Cauchy sequence in B t with 1>., = E,,>o a~~)zv. Then for every
c > 0 there is an n = n(c) E N such that for all >., J.L-? n,

~)a~~) - a;:)lt" = IIfA - fl-'lIt < c.

,,~o

Since tV = tr 1 ••• t!;.n # 0, it follows that

c
I a~'\) - a;:) I < t" for every v E No·

For fixed v, (a~'\» is therefore a Cauchy sequence in C which converges to a

complex number a".
Let f := Ev~o avz v . This is an element of C[z].
Given J > 0, it follows that there exists an n = n(8) such that
LI a~'\) - a~'\+") It" < ~ for>. ? nand J.L E N.
,,~o

Let I c No be an arbitrary finite set. For any>. ? n there exists a J.L = J.L(>') E
N such that EVElla~'\+") - a"lt" < 8/2, and then

LI a~'\) - a v It" < 8, for>. ? n.

vEl

In particular, 111>.. - flitS: 8. Thus 1>., - f (and then also J) belongs to B t ,

and (f>..) converges to f. •

Expansion with Respect to Zl. For the following we require some

additional notation:
If v E No, t E 1R+, and z E cn, write
 1. The Algebra of Power Series 107

V=(Vl,V'), t=(tl,t'), and Z=(ZI,Z').

An element I = E",~o a",z'" E C[z] can be written in the form

L f>..z~, L a(.~.vl){z'V'.
00

1= with f>..(z') =
A=O v' ~o

The series f>.. are formal power series in the variables Z2, ... ,Zn' We call this
representation of I the expansion 01 I with respect to ZI' Now the following
assertions hold:
00

1. IE B t {=} f>. E Bt' for all ~, and LIlf>..lltd~ < 00.

2. IIzl' flit = tJ. . IIfllt, for s E No-

PROOF:

(1) Since we are dealing with absolute convergence, it is clear that

00

IIflit = 2)f>.lkt~.
A=O

(2) We have z]'· I = E~=o f>..z~+s. The right side is the unique expansion of
I with respect to ZI. Now the formula can be easily derived.
Z]' . _

Convergent Series in Banach Algebras. Let B be a complex

Banach algebra and (fA) a sequence of elements of B. The series E~I!.x
converges to an element fEB if the sequence Fn := E~=l IA converges to
I with respect to the given norm.
1.2 Proposition. Every I E B with III - III < 1 is a unit in B with

1-1 = ~(1 - f)A and III-III ::; 1- II~ _ III'

PROOF: Let c := III - III. Then 0 ::; c < 1, and the convergent geometric
series E~=o c A dominates the series L~=o(1 - f)A. As usual, it follows that
this series converges to an element 9 E B. We have
n n
(1 - (1 - f)) . 2)1 - f)A
A=O
n+l
2:::(1- f)A - 2:::(1- f)A
n

A=O A=1
1- (1 - f)n+l.
As n tends to infinity we obtain f . 9 = 1 and 11911 ::; L~o c A = 1/(1 - c:). _
108 III. Analytic Sets

Convergent Power Series. A formal power series I = Lv>o avz v is

called convergent if it is convergent in some polydisk P around the origin. In
that case there exists a point t E P n 1R+., and since I converges absolutely
at t, it follows that I E B t . On the other hand, if I E B t , then by Abel's
lemma I converges in P = pn(o, t).
1.3 Theorem.
1. Hn ;= {f E C[z] 3t E 1R+. with 1I/IIt < oo} is the set 01 convergent
power series.
2. Hn is a C-algebra.
3. There is no zero divisor in Hn: II I· g = 0 in H n , then 1=0 or 9 = O.

We have already proved the first part, and then the second part follows easily.
The last part is trivial, since C[z] contains no zero divisors.
Remark. If I is convergent and 1(0) = 0, then for every E > 0 there is
atE 1R+. with II/lIt < E. In fact, since 1(0) = 0, we have a representation
1= Zd1 + ... + znln. If 1I/IIt < 00, then also Il/illt < 00 for i = 1, ... , n, and
n n
1I/IIt = L tilililit :s; max(tb· .. , t n) . Lilli lit-
i=1 i=1

This expression becomes arbitrarily small as t -t o.

When we go from B t to H n , we lose the norm and the Banach algebra
structure, but we gain new algebraic properties:
1. I E Hn is a unit {::::::} 1(0) '" O.
PROOF: One direction is trivial. For the other one suppose that 1(0) '" o.
Then 9 ;= 1·/(0)-1 - 1 is an element of Hn with g(O) = O. So there
exists a t with IIglit < 1, and 1·1(0)-1 is a unit in B t . Thus I is a unit
in Hn. •
2. The set m ;= {f E Hn : 1(0) = O} of all nonunits in Hn is an ideal:
(a) h,h Em=> h + hEm.
1 E m.
(b) IE m and hE Hn => h·
An ideal a in a ring R is called maximal if for every ideal b with a C b C R
it follows that a = b or b = R.
One can show that in any commutative ring with 1 :j:. 0 there exists a maximal
ideal. If a c R is maximal, then R/ a is a field.
1.4 Theorem. The set m 01 nonunits is the unique maximal ideal in H n ,
and Hn/m ~ C.

PROOF: If a c Hn is a proper ideal, then it cannot contain a unit. Therefore,

it is contained in m. The homomorphism <p : Hn -t C given by 1 t--t 1(0) is
surjective and has m as kernel. •
 1. The Algebra of Power Series 109

Distinguished Directions. An element I E Hn is called Zl -regular 01

order k if there exists a power series lo(zt} in one variable such that:
1. I(Zl, 0, ... ,0) = zf . lo(zt}.
2. 10(0) i: O.
If I is zl-regular of some order, I is called zl-regular.
Let I(z) = E~=o f>.zr be the expansion of I with respect to Zl. Then I is
zl-regular of order k if and only if 10(0') = ... = lk-l(O') = 0 and Ik(O') i: o.
I is zl-regular if and only if I(Zl, 0, ... ,0) ¢ O.
We often need the following properties:
1. I is a unit in Hn <==* I is zl-regular of order O.
2. If />.. is zl-regular of order k)." for>. = 1,2, then It . h is zl-regular of
order kl + k 2 •
There are elements I i: 0 of Hn that 1(0) = 0 which are not zl-regular, even
after exchanging the coordinates.
Definition. Let e = (C2,"" Cn) be an element of en-I. The linear
map ffc : en ~ en with

is called a shear.

The set E of all shears is a subgroup of the group of linear automorphisms

of en, with ffO = idcn.
We can write ffc(Z):= Z + Zl' (0, e). In particular, we have ffc(et} = (l,e).
1.5 Theorem. Let IE Hn be a nonzero element. Then there exists a shear
ffsuch that I 0 ff is Zl -regular.

PROOF: Assume that I converges in the polydisk P. If we had I(Zb z') = 0

for every point (Zb z') E P with Zl i: 0, then by continuity we would have
1= 0, which can be excluded. Therefore, there exists a point a = (aI, a') E P
with al i: 0 and I(a) i: O. We define e := (al)-l . a' and ff := ffc . Now,

So f 0 ff(Zl. 0') ¢ 0, and f 0 ff is zl-regular. •

Remark. If It, ... ,I, are nonzero elements in H n , then I := It ... II i: 0
converges on a polydisc P, and there exists a point a E P with I(a) i: 0 and
al ¥ O. As in the proof above we obtain a shear ff such that It 0 ff, ••• , b 0 ff
are zl-regular.
110 III. Analytic Sets

Exercises
1. Let I(z) = Lv2:o avz v be a formal power series.
(a) Prove the "Cauchy estimates": I E B t ~ lavl ::; IIIIIt/t V for
almost every v.
(b) Prove that if there is a constant C with lavls v ::; C, then IE B t for
t ::; s.
(c) Let In(z) = Lv>o av,nzv be a sequence of power series with IIIniis ::;
C. If every sequence (av,n) converges in C to a number a v , then
show that (fn) is a Cauchy sequence in Bt converging to I(z) =
LU2:o avz u , for every t ::; s.
2. The Krull topology on Hn is defined as follows: A sequence (fn) converges
in Hn to I if for every kEN there is an no with I - In E mk for n ~ no.
What are the open sets in Hn? Is Hn with the Krull topology a Hausdorff
space?
3. Let B be a complex Banach algebra with 1. Show that for every I E B
r
the series exp(f) = L;=:'=o In! is convergent, and that exp(f) is a unit
in B.
4. If I is a formal power series and I = L~=OPA its expansion into homo-
geneous polynomials, then the order of I is defined to be the number

ord(f) := min{s E No : Ps t= O}.

Now let (fn) be a sequence of formal power series such that for every
kEN there is an no with ord(fn) ~ k for n ::::: no· Show that E;=:'=1 In is
a formal power series. Use this technique also for the following:
If 91, ... , 9m are elements of Hn with ord(9d ~ 1, then

Lap. wp. I--t L aP.(91 (z), ... ,9m(Z))P.

p.2:0 p.2:0

defines a homomorphism <p: Hm -+ Hn of complex algebras.

2. The Preparation Theorem

Division with Remainder in B t . Let a fixed element t E lR+. be
chosen. When no confusion is possible we write B in place of B t , B' in place
of B t " and II I II in place of II I lit- The ring of polynomials in ZI with coefficients
in B' is denoted by B'[ZI].
2.1 Weierstrass Formula in B t • Let I and 9 = L~=OgAZf be two
elements 01 B. Assume that there exists an s E No and a real number e with
o < e < 1 such that 98 is a unit in B' and II zf - 9 . 9;; 111 < e . tf .
 2. The Preparation Theorem 111

Then there exists exactly one q E B and one r E B'[ZIJ with deg(r) < s such
that

f = q. 9 + r,
with

and

1
IIrll < Ilfll· 1_ EO·

PROOF: Let us first try to explain the idea of the proof. If h E B, then there
is a unique decomposition h = qh ·zf +rh, where rh E B'[ZIJ and deg(rh) < s.
lf 9 is given, we define an operator T = Tg : B -+ B by

T(h) := 9 . g;1 . qh + rho

lfT were an isomorphism, then f = T(T- 1 f) = g. (g;l· qr_1 f )+rr- 1 f would
be the desired decomposition. One knows from Banach space theory that T is
an isomorphism if idB - T is "small" in some sense. Since lI(idB - T)(h)1I =
II zf - gg; 111 . IIqh II, one can, in fact, conclude from the hypothesis of the
theorem that idB - T is "small." Now T- 1 = E~o{idB - T)A. Since (idB-
T)0 f = f and (idB - T) 1 f = (zf - gg; 1 )qf, we obtain the following algorithm:
Inductively we define sequences fA' qA, r A beginning with fo = f = ziqo + ro·
lf fA = ZfqA + rA has been constructed for some ), ~ 0, then we define

and obtain qA+l and r A+l by the unique decomposition

1>.+1 = zfqA+l + r>.+l, r>.+1 E B'[ztJ with deg(T>.+d < s.

If we define q := E~=o g;lqA and r := E~=o TA, then

00 00

00

L (gg;lqA + rA)
00

= g. q + T.
A=O
112 III. Analytic Sets

When using this algorithm we do not need the abstract transformation T

and the Banach theory of such transformations. However, it is necessary to
prove the convergence of all of the series that were used.
For this let h := -(zl- gg;I). Then IIhll < c' ti and gg;1 = zl + h.
From J>.. = zlq).. + r).. it follows that Ilr)..1I ~ Ilf)..1I and IIq)..1I ~ tIs. IIf)..lI·
Furthermore, from f)..+! = -h· q).. it follows that

11/>-.+111 ~ Ilhll·lIq)..1I < c ·llf)..ll·

Thus IIf)..11 < c).. ·/1111 and "£'::=0 f).. converges.
Since
IIg;1q)..11 < c)..t I llg;111·llfll
S and Ilr)../i < c).. IIfll ,
the series q = "£'::=0 g;1q).. and r = "£'::=0 r).. also converge.
The estimates for /lgsql/ and 11111 follow readily:
1
Lc).. =
00 00

~ Lllq)..11 ~ tlSllfll' tlSllfl/· 1-c'

)..=0 )..=0
00 1
I/rl/ < Ll/r)..1/ ~ Ilfl/· 1-c'
)..=0
It still remains to show uniqueness. Assuming that there are two expressions
of the form

it follows that

and

lI(ql - q2)gszfll < lI(ql - q2)gszf + (rl - r2)11

I/(ql - q2)gsh/l
< c' tf ·1I(q1 - q2)gsll
= c ·1I(ql - q2)gszfll·

2.2 Corollary. If the assumptions of the theorem are satisfied and if in

addition f E B'[ZlJ, 9 E B'[ZlJ, and deg(g) = s, then q E B'[ZlJ with q = 0
or deg(q) = deg(f) - s.

PROOF: Let d := deg(f). For d < s we have the decomposition f = O· g+ f·

Hence we have to consider only the case d ~ s.
 2. The Preparation Theorem 113

We assume that deg(fJ') :::; d for f.J. = 0, ... ,A. Then deg(q~) :::; d - s, and
therefore

deg(f~+d = deg(J~ - r~ - gg;lq~) :::; max(d, s - 1, s + (d - s» = d.

Hence deg(!A) :::; d and deg(q~) :::; d-s for all A. It follows that deg(q) :::; d-s,
and from f = g. q + r we can conclude that deg(q) = d - s. •

The Weierstrass Condition. We use the notation from above.

Definition. Let s E No. An element 9 = E~o g~z~ E B satisfies the
Weierstrass condition (or W-condition) at position s if:
1. g8 is a unit in B'.
2. /lzi - gg;I/1 < !ti.

Let R be an integral domain.! A polynomial f(u) = fsu s + fs_1Us-1 + ... +

It u + fo E R[u] is called monic or normalized if fs = 1. A polynomial
f E B'[zd is normalized if and only if it is zl-regular of some order k :::; s.
2.3 Weierstrass preparation theorem in B t • If 9 E B satisfies the W-
condition at position s, then there exists exactly one normalized polynomial
wE B'[zd of degree s and one unit e E B such that 9 = e . w.

PROOF: We apply the Weierstrass formula to f = zi. There are uniquely

determined elements q E B and r E B'[ZI] with zi = q. 9 + r and deg(r) < s
!
(we choose an c < such that IIzi - gg;111 < ctv.
But then zi - gg;1 = (q - g;l)g + r is a decomposition in the sense of the
Weierstrass formula. Therefore, we have the estimate

IIgsq -111 :::; t1Sllzi - gg;I/1' -11 < -Ie < l.

-c -c
That means that gsq and hence q is a unit in B. Let e:= q-l and w:= zi -r.
Then w is a normalized polynomial of degree s, and e· w = q-l(zi - r) = g.
If there are two decompositions 9 = el(zi - rt} = e2(zi - r2), then

From the uniqueness condition in the Weierstrass formula it follows that

el = e2 and rl = r2' •

2.4 Corollary. If 9 is a polynomial in ZI, then e is also a polynomial in

ZI·
1 An integral domain is a commutative nonzero ring in which the product of two
nonzero elements is nonzero.
114 III. Analytic Sets

PROOF: We use the decomposition zf - gg-;1 = (q - g-;l)g + r. From the

Weierstrass formula it follows that

Ilrll ~ IIzi - gg;II1' _1_ < tf. _c:_ < tf.

1-c: 1-c:
Since Ws = 1, it is also true that

IIzi - ww;111 = IIzi - wll = IIrll < tf·

Therefore, 9 = °
e . w + is a decompositionin the sense of the Weierstrass
formula, and the proposition follows from Corollary 2.2. •
The Weierstrass preparation theorem serves as a "preparation for the exami-
nation of the zeros of a holomorphic function." If the function is represented
by a convergent power series g, and there exists a decomposition 9 = e·w with
a unit e and a "pseudopolynomial" W(ZI' z') = zf + Al (z')zf-l + ... + As{z'),
then 9 and w have the same zeros. However, the examination of w is simpler
than that of g.

Weierstrass Polynomials. Now we turn to the proof of the Weierstrass

formula and the preparation theorem for convergent power series.
The ring Hn is an integral domain with 1. If IE Hn and I(z) = I:r=o f>..(z')zt
with f>.. = 0 for A > s, then I is an element of the polynomial ring Hn-dzl].
If Is =I 0, then deg(f) = s. If I is normalized and I>.{O') = 0 for A < s, then
I is zl-regular exactly of order s, and l(z},O') = zf,
Definition. A normalized polynomial w E Hn-dzl] with deg{w) =s
and w{z}, 0') = zfis called a Weierstrass polynomial.

We have seen that a normalized polynomial wE H n - 1 [zI! with deg{w) = s is

a Weierstrass polynomial if and only if it is zl-regular of order s. It follows
easily that the product of two Weierstrass polynomials is again a Weierstrass
polynomial.
If 9 = e· w is the product of a unit and a Weierstrass polynomial of degree s,
then we also have that 9 is zl-regular of order s, since the unit e is zl-regular
of order O. We now show that conversely, every zl-regular convergent power
series is the product of a unit and a Weierstrass polynomial.
2.5 Theorem. Let 9 E Hn be zl-regular 01 order s. Then lor every c: > 0
and every to E R+ there exists at:::; to such that 9 lies in B t , gs is a unit
in Bt" and IIzf - gg-;ll1t :::; c:. tf·

PROOF: Let 9 = 'Er=og>.zt be the expansion of 9 with respect to Zl. Then

g>.(O') = 0 for A = 0,1, ... , s - 1 and gs(O') =I- O.
Since 9 is convergent, there exists a tl :::; to with IIglitl < 00. Then g>. E Bt~
for all A, and in particular,
 2. The Preparation Theorem 115

fez') := gs(O')-lgs(Z') - 1 E Bt~.

Now, f(O') = 0. Thus there exists a t2 :::; t1 such that IIflitl < 1 for all t :::; t2'
Therefore, g8 is a unit in Bt" and 9 an element of B t .
Let h := zf - gg;l. Then h E B t for all t :::; t 2, and we have an expansion
h = L~=o h).z; with hs = 0, h). = _g).g;l for A =I: s, and h).(O') = for °
=
A O,l, ... ,s - 1.
If t1 > °is sufficiently small, then
00 (Xl

I L h).z~ lit ~ t1+1 ·11 L h).zt- s - 1

1It2 < tf . ~,
).=s+l ).=8+1

for all t = (t1' t') :::; t2. And since h).(O') =

small t1 there exists a suitable t' such that
° for A = 0, ... , s - 1, for every

8-1 8-1
II L h).z~ lit = Lllh).lkt~ < tf . ~.
).=0 ).=0

Consequently, /lhll t :::; € . tf. •

Remark. In a similar manner one can show that if gl, ... , gN E C[z] are
convergent power series and each gi is zl-regular of order Si, then for every
€ > 0 there is an arbitrary small t E lR~ for which

gi E B t , (gi)s; is a unit in Bt' and IIZ~i - gi(gi);.lli :::; €. t~;.

Weierstrass Preparation Theorem

2.6 Theorem (Weierstrass division formula). Let 9 E Hn be zl-regular
of order s. Then for every f E Hn there are uniquely determined elements
q E Hn and r E Hn-dzd with deg(r) < s such that

f = q. 9 + r.
If f and 9 are polynomials in Zl with deg(g) = s, then q is also a polynomial.

PROOF: There exists atE lR+ and an E with 0 < E < 1 such that f and 9
lie in B t , g8 is a unit in Bt', and IIzf - gg;ll1t :::; €. tf, It then follows from
the division formula in B t that there exist q and r with f = q. 9 + r.
Let two decompositions of f be given:
116 III. Analytic Sets

We can find atE IR+. such that f, qI, q2, r}, r2 lie in B t and 9 satisfies the
W-condition in B t . From the Weierstrass formula in B t it follows that ql = q2
and rl = r2. •

2.7 Theorem (Weierstrass preparation theorem). Let 9 E Hn be Zl-

regular of order s. Then there exists a uniquely determined unit e E Hn and
a Weierstrass polynomial wE Hn-dzll of degree s such that

9 = e ·w.

If 9 is a polynomial in Zl, then e is also a polynomial in Zl.

PROOF: There exists atE IR+. such that 9 satisfies the W-condition in B t .
The existence of the decomposition g = e . w with a unit e and a normalized
polynomial w of degree s therefore follows directly from the preparation the-
orem in B t . Since 9 is zl-regular of order s, the same is true for w. So w is a
Weierstrass polynomial.
Now, w has the form w = zl-r, where r E Hn-I[Zll and deg(r) < s. Thus, if
there exist two representations 9 = el(zi - rl) = e2(zi - r2), it follows that
zf = ell. 9 + rl = e21 . g + r2· The Weierstrass formula implies that el = e2,
rl = r2 and therefore WI = W2. •

Exercises
1. Write a computer program to do the following: Given two polynomials
f(w, x, y) (of degree n in wand degree m in x and y) and g(w, x, y) with
g(w, 0, 0) = w S , the program uses the Weierstrass algorithm to determine
q and r (up to order m in x and y) such that f = q . 9 + r.
2. Let f : pn-l X D -+ C be a holomorphic function and 0 < r < 1 be a real
number such that ( H f (z' , () has no zero for z' E pn-l and r :S I(I < l.
Then prove that there is a number k such that for every z' E pn-l
the function ( H f(z', () has exactly k zeros (with multiplicity) in D.
Use this statement to give an alternative proof for the uniqueness in the
Weierstrass preparation theorem.
3. Show that the implicit function theorem for a holomorphic function f :
enxC -+ e with f(O) = 0 and fZn (0) =f. 0 follows from the Weierstrass
preparation theorem.

3. Prime Factorization
Unique Factorization. Let I be an arbitrary integral domain with
1. Then 1* := 1- {O} is a commutative monoid with respect to the ring
multiplication, and the set IX of units of I is an abelian group.
 3. Prime Factorization 117

Let a, b be elements of 1*. We say that a divides b (symbolically a Ib) if there

exists e E 1* with b = a . e. We can also allow the case b = O. Then every
element of 1* divides 0, and a unit divides every element of I.
Definition. Consider an element a E 1* - I x .
1. a is called irreducible (or indecomposable) if from a = al . a2 (with
at, a2 E 1*) it follows that al E IX or a2 E IX.
2. a is called prime if alaIa2 implies that alaI or ala2·

Irreducible and prime elements can be defined in an arbitrary commutative

monoid. In 1* every prime element is irreducible, and in some rings (for
example, in Z or in lR[X]) it is also the case that every irreducible element is
prime. In zlRJ one can find irreducible elements that are not prime.
Definition. I is called a unique factorization domain (UFD) if every
element a E ]X can be written as a product of finitely many primes.

One can show that the decomposition into primes is uniquely determined up
to order and multiplication by units. In a UFD every irreducible element is
prime and any two elements have a greatest common divisor (gcd).
Every principal ideal domain 2 is a UFD, and in this case the greatest common
divisor of two elements a and b can be written as a linear combination of a
and b. For example, Z and K[X] (with an arbitrary field K) are principal
ideal domains. So in particular, qX] is a UFD.

Gauss's Lemma. Let I be an integral domain. Two pairs (a, b), (e, d) E
I x 1* are called equivalent if ad = be. The equivalence class of a pair (a, b)
is called a fraction and is denoted by a/b. The set of all fractions has the
structure of a field and is denoted by Q(1). We call it the quotient field of I.
The set of polynomials f (u) = ao +al u+· .. +an un in u with coefficients ai E I
constitutes the polynomial ring Ilu]. The set IO[u] of monic polynomials in
Ilu] is a commutative monoid. Therefore, we can speak of factorization and
irreducibility in 1° [u].
3.1 Gauss's lemma). Let I be a unique factorization domain and Q =
Q(I). If WI, W2 are elements of QOlu] with WIW2 E IO[u], then w).. E IOlu] for
A = 1,2.

PROOF: For A = 1,2, w).. = a)..,o + a)..,lu + ... + a)..,s"._IUs.>.-l + us.>. with
E Q.
a)..,1/ Therefore, there exist elements d).. E I such that d).. . w).. E Ilu]. We
can choose d).. in such a way that the coefficients of d).. . w).. have no common
divisor (such polynomials are called primitive).
2 A principal ideal domain is an integral domain in which every ideal is generated
by a single element.
118 III. Analytic Sets

We define d := d1 d2 and assume that there is a prime element p with pi d.

Then p doesn't divide all coefficients d)..a)...v of d>. . w>.. Let /-L>. be minimal
such that p t d>.a>..I'Jo.' Then

(d1wJ)(d2W2) = ... + U1'1+1'2 (dal,l'l a2.1'2 + something divisible by p) + ....

Since J is a UFD, p doesn't divide (d1al.I'J(d2a2,!'J. So the coefficient of
Ul'l +1'2 is not divisible by p. But since WIW2 has coefficients in J, every divi-
sor of d must divide every coefficient of d . WIW2 = (d 1wt}(d2 w2)' This is a
contradiction!3
When d has no prime divisor, it must be a unit. But then d 1 and d2 are also
units, and w>. = d-;:l(d>.w>.) belongs to JO[u]. _

3.2 Corollary. Let J be a unique factorization domain.

1. Jfa E JO[u] is prime in Q[uJ, then it is also prime in [O[u].
2. Jf a E [O[u] is reducible in Q[uJ, then it is reducible in JO[u].
3. Every element of JO[u] is a product of finitely many prime elements.
4. Jf a E [O[u] is irreducible, it is also prime.

PROOF: 1. Let a E JO[u] be a prime element in Q[u]. If a divides a product

a'a" in [O[u], then it does so in Q[u]. Therefore, it divides one of the factors
in Q[u]. Assume that there is an element bE Q[u] with a' = abo By Gauss's
lemma b E JO[u]. This shows that a is prime in [O[u].
2. Let a E JO[u] be a product of nonunits all a2 E Q[u]. If Ci E Q is the
highest coefficient of ai, then CIC2 = 1, c;lai E QO[u] and a = (c 11 at}(c21 a2)'
By Gauss c;lai E [O[u], and these elements cannot be units there. So a is
reducible in JO[u].
3. Every element a E JO[u] is a finite product a = al ... al of prime elements
of Q[u]. One can choose the ai monic, as in (2). Using Gauss's lemma several
times one shows that the ai belong to [O[u]. By (1) they are also prime in
JO[u].
4. Let a E [O[u] be irreducible. Since it is a product of prime elements, it
must be prime itself. _

Remark. In the proof we didn't use that [ is a unique factorization do-

main. We needed only the fact that Q[u] is a UFD (since Q is a field) and
the statement of Gauss's lemma: If all a2 E QO[u] and ala2 E [O[uj, then
ai E [O[uj for i = 1,2.

3 The original version of Gauss's lemma states that the product of primitive poly-
nomials is again primitive. The reader may convince himself that this fact can
be derived from our proof.
 3. Prime Factorization 119

Factorization in Hn. Now the above results will be applied to the case
I=Hn.
Definition. Let 1 E H n , f = E~op~ be the expansion of f as a
series of homogeneous polynomials. One defines the order of 1 by the
number

ord(f) := min{A E No : p~ =F O} and ord(O) := 00.

(See also Exercise 1.4 in this chapter)

Then the following hold:

1. ord(f) ~ 0 for every 1 E Hn.
2. ord(f) = 0 ~ f is a unit.
3. ord(JI . h) = ord(JI) + ord(12)·

3.3 Theorem. Hn is a unique factorization domain.

PROOF: We proceed by induction on n.

e
For n = 0, Hn = is a field, and every nonzero element is a unit. In this
case there is nothing to show.
Now suppose that the theorem has been proved for n - 1. Let f E Hn he a
nonunit, f =F o. If f is decomposable and 1 = h ·12 is a proper decomposition,
then ord(f) = ord(ft} + ord(12), and the orders of the factors are strictly
smaller than the order of f. Therefore, f can be decomposed into a finite
number of irreducible factors.
It remains to show that an irreducible f is prime. Assume that f Ihh, with
fA E (Hn)* for A = 1,2. There exists a shear (j such that h 0 (j, 12 0 (j and
f 0 (j are zl-regular. If we can show that f 0 (j divides one of the 1~ 0 u, then
the same is true for f and fA. Therefore, we may assume that h, 12, and f
are zl-regular.
By the preparation theorem there are units el, e2, e and Weierstrass polyno-
mials WI,W2,W such that h = el·WI, h = e2·W2, and f = e·w. Thenw divides
WIW2. If WIW2 = q . w with q E Hn, then the division formula says that q is
uniquely determined and a polynomial in Zt. So w divides WtW2 in H~_t[zll.
Since w is irreducible in H n , it must also be irreducible in H~_dzl]. By the
induction hypothesis H n - 1 is a UFD, and therefore w is prime in H~_l[Zl]. It
follows that wlwi or wlw2 in H~_t[zt] and consequently in Hn. This means
that fiJI or 1112 in Hn. •

Hensel's Lemma. Let w E Hn[u] be a monic polynomial of degree s.

There is a polydisk P around 0 E en
where all the coefficients of w converge to
120 III. Analytic Sets

holomorphic functions. Therefore, we can look at W as a parametrized family

of polynomials in one. variable u. By the fundamental theorem of algebra
w(O, u) splits into linear factors, and the same is true for every w(z, u) with
z E P. We now show that such splittings are coherently induced by some
splitting of W in H~[u], at least in a neighborhood of O.
3.4 Hensel's lemma. Let w(O, u) = TI~=l (u - CA)S). be the decomposition
into linear factors (with CII f CIJ for v f JL and SI + ... + Sl = s). Then there
are uniquely determined polynomials WI,'" ,WI E H~[ul with the following
properties:
1. deg(w A ) = SA, for>. = 1, ... , l.
2. W>..(O,u) = (u-c>..)s)..
3. W = WI· ... ·WI.

PROOF: We proceed by induction on the number l. The case 1 = 1 is trivial.

We assume that the theorem has been proved for l - 1.
First consider the case w(O, 0) = O. Without loss of generality we can assume
that Cl = O. Then w(O, u) = us, . h(u), where h is a polynomial over C with
deg(h) = S - Sl and h(O) f O. So W is u-regular of order SI, and there exists
a unit e E H~[ul and a Weierstrass polynomial WI with W = e . WI. Since
WI (0, u) = us', it follows that
I
e(O, u) = h(u) = II (u - cAY>'·
>"=2

By induction there are elements Wz, ... ,WI E H~[ut with deg(w>..) = S>..,
w>..(O,u) = (u - c>..)s>. and e = Wz" ·WI. Then W = WIWZ" 'WI is the de-
sired decomposition ..
If w(O,O) f 0, then we replace W by w'(z,u) := w(z,u + Cl) and obtain a
decomposition w' = w~ ... wf as above. Define

This gives a decomposition w = WI ... WI in the sense of the theorem. The

uniqueness statement also follows by induction. •

The Noetherian Property. Let R be a commutative ring with 1.

An R-module is an abelian group M (additively written) together with a
composition R x M -+ M that satisfies the following rules:
1. r(xt + X2) = rXl + rX2 for r E Rand Xl,X2 EM.
2. (rl + r2)x = rlX + r2X for rl, rz E R and x E M.
3. rl(rZx) = (rlr2)x for rl, rz E R and x E M.
 3. Prime Factorization 121

4. 1· x = x for x E M.
These are the same rules as those for vector spaces (and the elements of
a module are sometimes also called vectors). However, it may happen that
rx = 0 even if r #- 0 and x #- O. Therefore, in general, an R-module has no
basis. So-called free modules have bases by definition. -An example is the free
module Rq := R x ... x R (q times), with a basis of unit vectors. An example
of a nonfree module is the Z-module M := Zj6Z, where 2 . 3 = 2 . 3 = O.
If M is an R-module, then a submodule of M is a subset N c M with the
following properties:
1. x,yEN ==} x+yEN.
2. r E R and x EN==} rx E N.
A submodule of an R-module is itself an R-module.
Example
The ring R is also an R-module. The composition is the ordinary ring mul-
tiplication. In this case the submodules of R are exactly the ideals in R. An
R-module M is called finite if there is a finite set {Xl, ... , X n } C M such
that every x E M is a linear combination of the Xi with coefficients in R. The
free module W is obviously finite. But Zj6Z is also finite, being generated
by the class T.
Definition. An R-module M is called noetherian if every submodule
N C M is finite.
A ring R is called noetherian if it is a noetherian R-module. This means
that every ideal in R is finitely generated (in the sense of a module).

3.5 Proposition. Let R be a noetherian ring. Then any ascending chain

of ideals
10 C II C 12 C ... c R
becomes stationary, i.e., there is a ko such that h = Iko for k ;::: ko.

PROOF: The set J := U~o Ik is obviously an ideal. Since R is noetherian,

J is generated by finitely many elements II, ... , IN. Each Iv lies in an ideal
h". If ko = max(kl"" ,kN), then all Iv are elements of Iko' So Ik = Iko for
k;::: ko. •

3.6 Theorem. If R is a noetherian ring, then Rq is a noetherian R-module.

PROOF: We proceed by induction on q.

122 III. Analytic Sets

The case q = 1 is trivial. Assume that q 2 2 and the theorem has been proved
for q - 1. Let M C RQ be an R-submodule. Then
1:= {r E R : 3r' E RQ-l with (r,r') EM}

is an ideal in R and as such is finitely generated by elements rl, ... ,rl. For
every r), there is an element r~ E RQ-l such that r), := (r)" r~) lies in M.
The set M' := M n ({O} x RQ-l) can be identified with an R-submodule
of Rq-l, and by the induction assumption it is finite. Let r), = (0, r~), A =
l + 1, ... ,p, be generators of M'.
An arbitrary element x E M can be written in the form x = (Xl, x') with
Xl E I. Then Xl = I:~=l a),r)" a), E R, and
1 1
x- La)'r)' = (O,x' - La),r~) EM'.
),=1 ),=1

That is, there are elements al+l> ... , a p E R such that

I p

x-La)'r)'= L a),r),.
),=1 ),=1+1

Hence {rl' ... , r p} is a system of generators for M. •

3.7 Ruckert basis theorem. The ring Hn 01 convergent power series is
noetherian.

PROOF: We proceed by induction on n. For n = 0, Hn = C, and the

statement is trivial. We now assume that n 2 1 and that the theorem has
been proved for n - 1. Let I C Hn be a nonzero ideal and 9 =J 0 an element
of I. Without loss of generality we can further assume that 9 is zl-regular of
order s.
Let 1> = 1>g : Hn -+ (Hn_d S be the Weierstrass homomorphism, which is
defined in the following manner: For every 1 E Hn there are uniquely defined
elements q E Hn and r = ro + rlzl + ... + rs_lzf- l E H n- 1[Zl) such that
1 = q . g + r. Let 1>(1) := (ro, ... ,rs-r).
Now, 1> is an Hn_l-module homomorphism. By the induction hypothesis
H n- 1 is noetherian, and so (Hn_I)S is a noetherian Hn_I-module. Since M :=
1>(1) is an Hn_1-submodule, it is finitely generated. Let r), = (r~t), ... , r~~..\),
A = 1, ... , l, be generators of M.
If 1 E I is arbitrary, then 1 = q·g+r with r = rO+rlzl + .. . +rs-lz:-l, and
there are elements al, ... , al E H n - l such that (ro, rl, .. ·, rs-r) = 1>g(l) =
I:~=l a),r),. Hence we obtain the representation
 4. Branched Coverings 123

The set {g, r(l}, ••• , r(l)} with r(A) = r6 +

A) riA) Zl + ... + r~~.\ z:-l is a system
of generators of I. •

Exercises
1. Prove that O(C) is not a UFD.
2. Let M be a finite Hn-module, and m c Hn the maximal ideal. If M =
m . M, then M = O.
3. Let f : pn-l X D ---+ e be a holomorphic function such that for every
z' E pn-l there is a unique solution Zn = cp(z') E D of the equation
f(z', zn) = O. Use function theory of one variable to show that cp is
continuous, and use Hensel's lemma to show that cp is holomorphic.
4. Let f E Hn be zl-regular, f = e . w with a unit e and a Weierstrass
polynomial w E Hn-dzlJ. Prove that f is irreducible in Hn if and only
if w is irreducible in H n - l [ZlJ.
5. Show that f(z, w) := z2 - w 2 (1- w) is irreducible in the polynomial ring
C[z, wJ and reducible in Hn.
6. Let f E Hn be given with fZi (0) =I- 0 for some i. Prove that f is irreducible
in Hn.

4. Branched Coverings
Germs. Let Been be an open set and Zo E B a fixed point. A local
holomorphic function at Zo is a pair (U, f) consisting of an arbitrary neigh-
borhood U = U(zo) c B and a holomorphic function f on U. Two such
functions f : U ---+ e and 9 : V ---+ e are called equivalent if there is a neigh-
borhood W = W(zo) c Un V such that flW = glW. The equivalence class
of a local holomorphic function (U, f) at Zo is called a germ (of holomorphic
functions) and is denoted by fzo- The value f(zo) as well as all derivatives of
f at Zo (and therefore the Taylor series of f at zo) are uniquely determined
by the germ. On the other hand, if a convergent power series at Zo is given,
then this series converges in an open neighborhood of Zo to a holomorphic
function f, and the germ of f determines the given power series. So the set
Ozo of all germs of holomorphic functions at Zo can be identified with the e-
algebra of all convergent power series of the form L..,>o(Z-zo)v. This algebra
is isomorphic to the algebra Hn and has the same algebraic properties.
Let fzo =I- 0 be any element of Ozo with f(zo) = O. Then there are a neigh-
borhood U(zo) c B, a neighborhood W(O) C e n - l , a holomorphic function
e on U, and holomorphic functions al, ... , as on W such that after a suitable
change of coordinates the following hold:
124 III. Analytic Sets

1. e(z) -=I 0 for every z E U.

2. fez) = e(z)· w(z - zo) for W(WI' w') = wi + al(w')zf- 1 + ... + as(w').

Pseudopolynomials. A pseudopolynomial of degree s over a domain

G c en is a holomorphic function W in G x e that is given by an expression

with hI. ... , hs EO, where 0 = O( G) denotes the ring of holomorphic

functions on G. The set of pseudopolynomials of any degree over G will be
written as OO[u).
We begin with several remarks on the algebraic structure.
4.1 Proposition. If G is a domain, i.e., a connected open set, then the
ring 0 = O( G) is an integral domain.

PROOF: We need to show only that 0 has no zero divisors. Assume that
II, h are two holomorphic functions on G with both fi 1= O. Since G is a
domain, their zero sets are both nowhere dense in G, and there is a point
z E G with lI(z)· h(z) -=I O. So II· h 1= O. •
It also follows that OO[u) is free of zero divisors. We denote by Q the quo-
tient field of O. Then the group Q[u)X of units in the integral domain Q[u)
consists of the nonzero polynomials of degree O. If 0* c 0 is the multiplica-
tive subgroup of not identically vanishing holomorphic functions on G, then
Q[u)X nO = 0*.
4.2 Proposition. If WI,W2 E QO[u) are pseudopolynomials with WI . W2 E
OO[u), then Wl,W2 E OO[u).

PROOF: If W = US + (fI/gdu s - 1 + ... + (fs/gs) is an arbitrary element of

QO[u), then for all z E G the germs gi,z are not O.
For a moment we omit the i. If the quotient of fz and gz is holomorphic, i.e.,
fz = h z · gz with h z E Oz, then hz is uniquely determined and there is a ball
B around z in G such that hz comes from a holomorphic function h on B
and the equation f = h . 9 is valid in B. If we take another point z' E B,
the germ of h at this point is the quotient of the germs of f and 9 at this
point. So z H hz(z) defines a global holomorphic function h on G. We write
h = fig.
Thus, if Wz := US + ((fdz/(gdz)u s - I + ... + ((fs)z/(gs)z) lies in O~[u) for
every z E G, then W E OO[u].
Now we apply Gauss's lemma in the unique factorization domain Oz ~ Hn.
Let W := WI . W2. Then (wdz . (W2)z = Wz E O~[u] for every z E G. Conse-
quently, the coefficients of (Wi)z are holomorphic, and by the remarks above
this means that Wi E OO[u). •
 4. Branched Coverings 125

The following is an immediate consequence of the above two propositions:

4.3 Theorem. Let G E en be a domain. Then OO[u] is a factorial monoid;
i.e., every element is a product of finitely many primes.

Euclidean Domains
Definition. An integral domain I is called a Euclidean domain if there
is a function N : I* -+ No with the following property (division with
remainder): For all a, bEl, b =I- 0, there exist q, rEI with
1. a=q·b+r,
2. r = 0 or N(r) < N(b).
The function N is called the norm of the Euclidean domain.

Examples
1. Z is a Euclidean domain with N(a) := lal.
2. If k is a field, then k[x] is a Euclidean domain, by N(f) := deg(f).
Every Euclidean domain I is a principal ideal domain and thus factorial. If
a, b are elements of I, then the set of all linear combinations

r . a + s . b =I- 0, r, s E I,

has an element d with N(d) minimal. The element d generates the ideal
a = {ra + sb : r, s E I} and is a greatest common divisor of a and b. It is
determined up to multiplication by a unit.
Now assume again that G is a domain in en, 0 = O(G), and Q = Q(O}. Then
Q[u] is a Euclidean domain. If Wt,W2 are pseudopolynomials in OO[u], there
is a linear combination in W = rIwI + r2w2 t= 0 in Q[u] with minimal degree.
It can be multiplied by the product of the denominators of the coefficients in
rl and r2. Then rt, r2, and ware in O[u], and W is a greatest common divisor
of WI, W2.

The Algebraic Derivative. Let 0 and Q be as above. If W E OO[u]

has positive degree, then it has a unique prime decomposition W = WI ... W,.
The degree of each Wi is positive. We say that W is (a pseudopolynomial)
without multiple factors if all the Wi are distinct.
The (algebmic) derivative of a pseudopolynomial is defined as follows. If
W = E~=oa"u", then D(w}:= E~=I v· a,,· u,,-I. Thus
D(WI +W2) D(wt} + D(W2),
D(WI ·W2) D(wt} . W2 + WI . D(W2}.
126 III. Analytic Sets

4.4 Theorem. An element W E OO[uJ is without multiple factors if and

only if a greatest common divisor of wand D{w) is a function h E 0·.

PROOF: If W has the irreducible Wi as a multiple factor, then D{w) is also

divisible by Wi. This is also true in Q[u]. So a greatest common divisor is
certainly not a function h E 0·.
Assume now that W = TIi Wi has no multiple factor. Then

D{w) = LWl···D(Wi)···WI.
i

If the degree of the greatest common divisor 'Y of wand D{w) is positive, then
'Y is a product of certain Wi. So at least one Wi divides both wand D{w). Then
Wi divides WI ... D(wd··· WI and hence D(wi). This is not possible, because
D{wi) has lower degree. So the degree of the greatest common divisor is 0,
and therefore it is a function h E 0·. •

Symmetric Polynomials.
Definition. A polynomial p E Z[Ul, ... , us] is called symmetric if for
all i, j we have P(Ul, ... , Ui,·.· Uj, ... , us) = p(Ul' ... ' Uj, ... , Ui,· .. , us)·

There are the elementary symmetric polynomials 0"1, ... , 0" s defined as follows:
O"l(Ul, ... ,Us ) Ul+···+U s,
0"2{Ut. . .. , us) = Ul (U2 + ... + us) + U2{U3 + ... + us) + ... + Us-l us,

The following result is proved, e.g., in the book [vdW66].

4.5 Theorem. If p E Z[Ul, ... , us] is symmetric, then there is exactly one
polynomial Q{yt. ... , Ys) E Z[Yl, ... , Ys] such that p = Q(O"l, ... , O"s).

The Discriminant. Consider the special symmetric polynomial

pv(ut. ... ,us ) = II(ui _Uj)2
i<j
(square of the Vandermonde determinant). Since it is symmetric, there is
a uniquely determined polynomial QV(Yl, ... ,Ys) with integral coefficients
such that
 4. Branched Coverings 127

Definition. If w = US + hI (Z)u s- I + ... + hs{z) is a pseudopolynomial

in OO[u], then ~w = Qv( -hI, h2,"" (-l)Shs) is called the discriminant
of w. It is a holomorphic function in G, and we denote its zero set by Dw'

It is well known from the theory of polynomials that

where WI, ... , Ws are the zeros of the polynomial u H w(u, z). So ~w(z) = 0
if and only if there is a pair i =I- j with Wi = Wj.
Assume now that w is without multiple factors. Then there is a linear combi-
nation of wand D{w) that is a function hE 0·. We restrict to a point z E G
with h{ z) =I- O. Then the greatest common divisor of w{ u, z) and D{w)( u, z)
is 1. This means that w(u, z) has no multiple factors; i.e., the zeros of w(u, z)
are all distinct. So ~w(z) =I- 0, and Dw is nowhere dense.
Example
Let G c en be a domain, a, b holomorphic functions in G, and w{u, z) :=
u 2 - a(z) . u + b(z). In this case

PV(UI' U2) = II (Ui - Uj)2 = (UI - U2)2 = (UI + U2)2 - 4· UI • U2·

i<j

6 w (z) = Qv(a(z), b(z» = a(z? - 4b{z).

If z E G and ~w(z) =I- 0, there are two different solutions of w(u,z) = O.

Hypersurfaces. We use the theory of pseudopolynomials to study ana-

lytic hypersurfaces. Such analytic sets are locally the zero set of one holomor-
phic function. Assume that 1 is a holomorphic function in a connected neigh-
borhood of the origin in en+! that is not identically O. Without loss of gener-
ality we may assume that A = N(f) contains the origin. Then a generic com-
plex line f through 0 meets A in a neighborhood of 0 only at the origin. After
a linear coordinate transformation, f = {( U, z) : z = O} is the first coordinate
axis. By the Weierstrass preparation theorem 1(0,0) = e(O,O) . W(O,O) in the ring
Hn+I' where e(O,O) is a unit in Hn+!, and w(O,O) E Hn[u] a Weierstrass poly-
nomial. We can represent the germs locally by holomorphic functions. Thus
there is a domain G c en containing 0, and a disk D = {u E e : lui < r}
such that in U := D x G there are a holomorphic function e that does not
vanish in U and a pseudopolynomial w over G with 1 = e· win U. We may
assume that A n (aD x G) = 0. Therefore, the zero set of 1 in U is that of
w.
128 III. Analytic Sets

We can decompose w into prime factors. Using the fact that any power of a
prime factor vanishes at the same points as the prime factor does, we may
assume that w is without multiple factors. Then the discriminant ~w is not
identically zero in G. We set Dw = {z E G : ~w(z) = O}.
4.6 Theorem (on branched coverings). If Zo E G - DW1 there are a
neigborhood W = W(zo) c G - Dw and holomorphic functions JI, ... , fs in
W with fi(z) =I- h(z) for i =I- j and z E W such that
w(u, z) = (u - JI(z))··· (u - fs(z)) in ex w.
There are fewer than s points over any point Zo E Dw (see Figure III. 1).

Figure 111.1. A branched covering over G

A point z E G above which there are fewer than s points is called a branch
point. All points of the discriminant set Dw are branch points. Over all other
points our set A is locally the union of disjoint graphs of holomorphic func-
tions, and is therefore regular.
PROOF: For Zo E G - Dw the polynomial w(u, zo) has s distinct roots.
We write w(u, zo) = (u - Cl)··· (u - c s ), where the Ci all are distinct. If
w(u, z) = US + h1(z)U s - 1 + ... + hs(z), then the germ
wZo := u2 + (hdzou s - 1 + ... + (hs)zo
is a polynomial over Ozo ~ Hn. By Hensel's lemma it has a decomposition
wZo = Wl,zo ... ws,zo with the following properties:
 4. Branched Coverings 129

1. Wi,zO(U, ZO) = U - Ci for i = 1, ... , s.

2. deg(wi,zo) = 1.
We have Wi,zo = U - ri, with ri E Hn. There are a connected neighborhood
W(zo) c G - Dw and holomorphic functions It, ... , fs in W such that the
power series ri converge to Ii- Since the germs of wand (u - It)··· (u - fs)
coincide at Zo, it follows from the identity theorem that

wlcxw = (u - It)··· (u - fs),

and since We G-Dw, it also follows that fi(Z) =I- fj(z) for i =I- j and Z E W.
•
Examples
1. Let G = C and W = zI -
Z2. Then the discriminant is given by ~w(Z2) =
4z2 , and Dw = {O} C C. If Z2 E C*, there is a neighborhood W C C - Dw
where ..jZ2 is well defined. There we have W = (Zl - ..jZ2) . (Zl + ..jZ2).
This gives a surface above C that is a connected unbranched 2-sheeted
covering over C-{O}. The point 0 is a branch point. This is the (branched)
Riemann surface of "fZ. The unbranched part was already discussed in
Section 11.8.
2. A completely different situation is obtained if we take W = zI -
z~
(Zl - Z2) . (Zl + Z2). The discriminant is 4z~ in this case, and the discrim-
inant set Dw is again the origin in C. The set A consists of two distinct
sheets, which intersect above 0, and both are projected biholomorphically
onto C. The set A - {O}, i.e., A without the branch point, is no longer
connected.
3. In higher dimensions the situation is even more complicated. Let us con-
sider the analytic set A = N(f), where f(z!, . .. ,zn) = Z:l + ... + z~n
with Si ~ 2 for i = 1, ... , n. This is a very simple holomorphic function.
The derivatives are fZi = Si . z;i- 1 , and their joint zero set consists only
of 0 E cn. So all other points of A are regular.
Every line £ through the origin lies completely in A, or f has a zero of
order S with s ~ min(s!, ... , Sn) on £ at the origin. Therefore, there is
no line that intersects A in 0 transversally. From this one can conclude
that 0 is in fact a singular point of A (see, for example, Exercise 8.2 in
Chapter I).
Now we look on f: cn -+ C as a fibration with general fiber
At = {z E C n : Zfl + ... + z~n = t}.
Then A = Ao has an isolated singularity, while all other sets At are
regular everywhere. We call the family (AdtEC a deformation of A.
130 III. Analytic Sets

The Unbranched Part. We assume that G c en is a domain and

w(u, z) a pseudopolynomial over G of degree s. We set G' = G - D w ,

A = ((u,z) E ex G : w(u,z) = O},

and A' = AIG' the part of A over G'. Then A' is an unbranched covering of
G'. It is an n-dimensional submanifold of e x G', and we have the canonical
projection 7r : A ~ G. If (uo, zo) E A' is a point, there is a small neighborhood
B = B(uo, zo) c A' that is mapped by 7r holomorphically and topologically
onto a ball around Zo in G'. We also call B a ball. The holomorphic map
(7rIB)-I : 7r(B) ~ en+! is a local parametrization of A'. A complex function
fin B is called holomorphic if f 0 (7rIB)-I is holomorphic. In particular, the
components of 7r itself are holomorphic functions on B.
For holomorphic functions in B we have the same properties as for holomor-
phic functions in a domain of en. For example, the identity theorem remains
valid, and we obtain the following results:
4.7 Proposition. Assume that Al is a connected component of A' and that
M is an analytic subset of AI. Then M = AI, or M is nowhere dense in AI.

4.8 Proposition. Iff is a holomorphic function on A', and Al a connected

component of A', then either f vanishes identically on Al or its zero set is
nowhere dense in AI.

4.9 Proposition. Let Al again be a connected component of A'. Assume

that M is a nowhere dense analytic set in Al and that f is a holomorphic
function in Al - M that is bounded along M. Then f has a unique holomor-
phic extension to AI.

Decompositions. We consider the interaction between the decomposi-

tion of a pseudopolynomial into irreducible factors and the decomposition of
its zero set into "irreducible" components.
4.10 Proposition. Let G c en be a domain and w(u,z) a pseudopolyno-
mial over G without multiple factors. Then w is irreducible if and only if the
intersection of its zero set A with e x (G - E) is connected for every nowhere
dense analytic subset E C G which contains the discriminant set Dw.

PROOF: Since locally over G' = G - Dw the set A' = AIG' looks like a
domain in en, a nowhere dense analytic set does not disconnect A', locally
and globally. Therefore, we may assume that E = Dw.
If w is not irreducible, every factor Wi defines an analytic set Ai over G - Dw·
The intersection of different Ai is empty. SO AI(G - Dw) is not connected.
 4. Branched Coverings 131

If, on the other hand, AIG' has the connected components Ai, i = 1, ... , s,
with s > 0, then for any point z E G - Dw there is a ball BeG - Dw around z
such that AilB splits into graphs of holomorphic functions 1; : j = 1, ... , Si·
In each case we form the pseudopolynomial Wi = (u - II) ... (u - fs i )' The
zero set of this Wi is exactly AiIB, and it determines Wi and vice versa. So
over the intersection of two different balls the pseudopolynomials must be
the same, and thus we obtain global holomorphic functions Wi in G - Dw' If
z E Dw, then there is a neighborhood W of z such that Ail(W - Dw) c AIW
is a bounded set. So the coefficients of Wi are bounded over this neighborhood
and extend holomorphically to G. We also denote this extension by Wi, and
for reasons of continuity it follows that W = WI' .. WS' -

If the Wi are the irreducible factors of w, we call their zero sets Ai the irre-
ducible components of A. The sets A~ = AilG' are the connected components
of AIG'.
4.11 Proposition. Assume that W* ,ware pseudopolynomials without mul-
tiple factors over a domain G and that A* = {w· = O} c A = {w = O}. Then
w' is a factor of w.

PROOF: Let D denote the union of the discriminants of w· and w. It is

a nowhere dense analytic set in G. Over G - D we decompose the two
unbranched coverings into connected components. There we have A· =
Al U··· U As· and A = Al U'" U As with s* S s. This yields pseudopolyno-
mials over G - D that extend to pseudopolynomials WI, ... , Ws over G, with
w· = WI ... Ws. and W = WI ... Ws' This implies the result. _
The following result is proved analogously.
4.12 Proposition. Assume that W is free of multiple factors and that A =
{w = O} is the disjoint union of two nonempty sets M', M" that are closed in
C x G. Then there are pseudopolynomials w' , w" over G with M' = {w' = O},
M" = {w" = O}, and w' . w" = w.

PROOF: The construction is first carried out outside Dw. We set G' =
G - Dw and use the fact that every nonempty open subset of A' = AIG'
must be a union of connected components of A'. If W = WI'" Ws is the
decomposition into irreducible factors, then we may assume that there is an
s· with 0 S s· S s such that for w' = WI" .ws* and w" = Ws*+1 .,. Ws we
have M'IG' = {(u, z) E C X G' : w'(u, z) = O} and M"IG' = {(u, z) E
ex G' : w"(u, z) = O}.
It is now essential that in a continuous family f (u, z) of holomorphic functions
of one variable u the zeros depend continuously on the family parameter z
("continuity of roots"). We omit the proof here. If we apply this fact (and
the equations A = M' U M", W = w' . w"), we get that the sets M'IG' and
 132 III. Analytic Sets

M"IG' are not empty (i.e., 0 < s* < s) and that their closures in C x G are
M', respectively Mil. •

Projections. In the next section we will investigate zero sets of several

holomorphic functions. Here we begin with the simplest case, the common
zero set N of a pseudopolynomial w over G c en and an additional holomor-
phic function f in a neighborhood of A = {(u, z) E ex G : w(u, z) = O}.
Our method involves the projection of N to G.
4.13 Proposition. Assume that f = f( u, z) is a continuous function on A
that is holomorphic outside of e x Dw and does not vanish identically in a
neighborhood of any point of A. Then the projection of N = {j = w = O} to
G is an analytic set N' = {f = O}, where f is a holomorphic function in G
that does not vanish identicclly. -

PROOF: If z E G - D w, we have a ball BeG - Dw around it such that

over B our w has the form w( u, z) = (u - h (z)) ... (u - fs (z)). The function
f does not vanish identically on any graph u = k Consequently,
l(z) := f(h(z), z)··· f(Js(z), z)

does not vanish identically. In the usual way we obtain the holomorphic
function f in the entire set G - Dw' It is bounded along Dw' So it extends
to a holomorphic function in G. •

Now consider the following situation: Assume that G is a domain in and en

that w is a pseudopolynomial over G without multiple factors. Let f be a
holomorphic function in a neighborhood of A = {w(u, z) = O} c e x G that
does not vanish identically on any open subset of A and define

N:= ((u,z) E e x G : w(u,z) = f(u,z) = O}.

Denote by N' the projection of N to G.
We want to give a definition for "unbranched points" of N. The difficulty is
that there may exist such unbranched points of N lying in the set of branch
points of w.
4.14 Proposition. For any point Zo E Nt there is an arbitrarily small
linear coordinate change in Zl, ... ,Zn such that thereafter the line parallel to
the Zl -axis through Zo intersects N' in Zo as an isolated point.
In such coordinates there is a neighborhood U(zo) C G, a domain G' in
e
the space n- 1 of the variables z' = (Z2,"" zn), and a pseudopolynomial
W'(Zl, z') over G' such that {(Zl' z') E C xG' : W'(Zl, z') = O} = N' n U.
PROOF: A "small" linear change of the coordinates Zl,"" Zn means here
. that the transformation is very near to the identity. Since L does not vanish
 4. Branched Coverings 133

identically, after a small generic coordinate transformation the line parallel

to the zl-axis through Zo intersects N' in Zo as an isolated point. And then
it is also clear that U, G', and w' with the desired properties exist. _

Let us now assume that we have chosen a point Zo E N' and suitable coordi-
nates as above, and that U, G', and w' have also been chosen.
Definition. In the given situation, a point (u, z) E N n (C x U) is
called an unbranched point of N if zEN' - C x D w ' and there is a
neighborhood V = V(z) c N' - C x D w ' with a holomorphic function 9
on V such that N n (C x V) is the graph {u = g(w) : w E V}. (Figure
111.2 shows the situation.)

Z2 .

......
o examples of unbranched points ~
ex D.."
Figure 111.2. Branched and unbranched points of N

4.15 Theorem. In the given situation, in every neighborhood of an arbi-

trary point (Ul,Zl) of N n (C x U) there are unbranched points of N.

PROOF: We may assume that Zl EN' - (C x D w ')' Then we take a small

neighborhood W = E X U1 of (Ul, Zl) such that the following hold:
1. U1 c U - (C x Dw')'
2. (Ul, Zl) is the only point of A above Zl in W.
134 III. Analytic Sets

3. The set A n W is defined by a pseudopolynomial w· over UI.

Setting N{ c N' n U{ to be the image set of W n N under the canonical
projection 11' : W -t UI , take U1 so small that N{ = N' nUl.
We restrict w· to N~ and replace possible multiple factors by one factor at a
time. So we obtain a new pseudopolynomial WI without multiple factors over
N~ such that

NnW = {( u, z) E E x Nf : WI (u, z) = O}, E a suitable disk.

Arbitrarily near to (UI, zt) we can find points (U2, Z2) E NnW lying over
N~ - D W1 ' All of these points are unbranched points of N, since we can find
neighborhoods W 2(U2, Z2) C Wand U2(Z2) C U1 with the same properties as
Wand Ul. Now choose U2 so small that it contains no point of DWl and that
wIIN' n U2 decomposes into linear factors. Then every sheet of AI(N' n U2 )
with the property that it contains points of N is a graph over N' n U2 • •

Exercises
1. Prove that every symmetric polynomial in UI,.'" Us can be written as a
polynomial in the power sums Sk := u~ + ... + u!.
2. Calculate the discriminant of a cubic polynomial.
3. Let D := Dr(O) C e and I be a holomorphic function in an open neigh-
borhood of D without zeros in aD. If I has in D the ~eros Cl,.'" Cs
(some of them may be equal), then

4. Let I : D x pn -t e be a holomocphic function in the variables

u, z}, ... , Zn' Assume that I is u-regular of order s and that for every
z E pn the function U M I(u, z) has exactly s zeros udz), ... , us(z)
(with multiplicity) in D. Show that the coefficients of the "pseudopoly-
nomial" w(u,z) := rr;=l(U - Uj(z» are holomorphic.
5. Prove the "continuity of roots": Let I (u, z) be u-regular at the origin.
Show that there is an r > 0 such that if g(z) is a function defined in a
neighborhood of 0 with Ig(z)1 < rand I(g(z), z) = 0, theng is continuous
at O.
6. A complex function I on an analytic set A in a domain G c enis called
holomorphic if it is locally the restriction of a holomorphic function in
the ambient space.
(a) Let A := ({w, z) E e 2 : w 2 = Z3} be the Neil parabola. There is
a bijective holomorphic parametrization of A given by w = t 3 and
Z = t 2 . Describe the holomorphic functions on A as functions of the
parameter t. Is there a meromorphic function on A that has a pole
at 00 with main part t m ?
 5. Irreducible Components 135

(b) Show that the analytic set A = {(w, ZI, Z2) E C 3 : W2 = ZIZ2} is
not regular at the origin. Consider the holomorphic map (t1, t2) t-+
(tIt2, ti, t~). It is a "two-to-one" map. Describe the local holomorphic
functions on A in tl, t2.
7. Prove that there is a topological holomorphic map from C* onto A :=
{(w, z) E C2 : w 2 = ZIZ2}.
8. Prove that there is no topological holomorphic map from C· onto the
"elliptic surface" A:= {(w,z) E C 2 : w 2 = (z2 -1)(z2 - 4)}.
9. Define the pseudopolynomial wE O(C 2 )[u) by w(u, Zl, Z2) := u2 - U· Zl
and determine the discriminant set and the irreducible components of
A := {w = O}. Let! on A be defined by !(U,Zl,Z2) := Z2 . (u - 1).
Consider N:= {J = w = O} C C3 and determine the projection N' C C 2
and the set of unbranched points of N.

5. Irreducible Components
Embedded-Analytic Sets. We wish to study general analytic sets.
Since it is easier to work with pseudopolynomials than with arbitrary holo-
morphic functions, we introduce the notion of "embedded-analytic sets."
These are subsets of the common zero set of finitely many pseudopolyno-
mials, and they are not a priori analytic by definition. But later on, it will
turn out that they are indeed analytic.
Assume that G c C n - d = {z' = (Zd+ 1, ... , Zn)} is a domain and that
Wi(Zi;z'), i = 1, ... , d, are pseudopolynomials over G without multiple fac-
tors. The zero sets of the single Wi intersect transversally4 in Cd x G. We
denote by D C G the union of the d discriminant sets belonging to the Wi.
We call it the union discriminant set. We put

A := {(z!, ... ,Zd, z') : Wi(Zi; z') = 0, for i = 1, ... ,d and z' E G}.

Over any ball BeG - D the set AlB consists of finitely many disjoint
holomorphic graphs. Every graph is contained in a connected component Z
of AI(G - D). We call the closure of Z in A an irreducible embedded-analytic
component of .Ii.

Definition. If A is defined as above, any union of finitely many ir-

reducible embedded-analytic components of A is called an embedded-
analytic set of dimension n - d.
4 Two submanifolds M, N C en intersect transversally at a point z E M n N if
the entire space is spanned by vectors that are tangent to M or N at z. In our
case the common zero set of the pseudopolynomials Wi contains enough vectors
to span en. Therefore, we say that these zero sets intersect transversally.
136 III. Analytic Sets

By definition any embedded-analytic set A can be decomposed into finitely

many irreducible components.
The surrounding set A :J A is not uniquely determined. Sometimes we can
make A smaller by throwing away those irreducible factors of Wi that do not
vanish identically on A. Then the Wi are uniquely determined by A.
5.1 Proposition. Assume that A is an embedded-analytic set in A C Cd X G
and that f is a holomorphic function in a neighborhood of A that does not
vanish identically on any open subset of A. If N = {z E A : fez) = oJ,
then for any point Zo E N there is an arbitrarily small linear change of the
coordinates z' such that the affine space parallel to the (Zl,"" Zd+t)- "axis"
through Zo intersects N in an isolated point. If Zl E N is any point near
zo, then there are unbranched points of N arbitrarily near Zl. At all of these
points N is a submanifold of dimension n - d - 1.

PROOF: We proceed as in the proof for the last theorem of the previous
section. The procedure to find unbranched points will be denoted by (*).
First we construct the projection f of f, which is holomorphic in G. For this
observe that if D is the union discriminant set of the polynomials WI, ... ,Wd
and z' E G - D, we always have the same number of points ZI, ... ,Zs in A
over z'. We set fez') = f(zl)'" f(zs) and obtain f, which is holomorphic on
G - D. Since itis bounded along D, we can extend it holomorphically to G.
Therefore, the projection set is N' = {[(z') = OJ.
Assume now that z~ EN'. Then, after an arbitrarily small linear coordinate
change in the variables z', the line L parallel to the Zd+l-axis through z~
intersects N' in an isolated point. Then by Weierstrass's theorem we can
find a neighborhood U = U(z~) c G, a domain G', and a pseudopolynomial
w' (Zd+1, z") over G' without multiple factors such that Un N' is equal to the
set {z' = (Zd+1' z") E C X G' : W'(Zd+l, z") = OJ, with z" = (Zd+2,"" zn).
Since the space Cd x L intersects N at Zo in an isolated point, it remains to
prove the existence of unbranched points. We apply (*) to N n (Cd x U) and
prove in the same way as before that for points ZI EN n (Cd x U) there are
unbranched points of N arbitrarily near to ZI' Of course, at these points N
is a submanifold of dimension n - d - 1. •

For the following we use the same notation and hypotheses as above.
5.2 Theorem. Let N' c G be the projection of N. For every point Zo =
(zo, z~) ENe Cd X G, after a suitable linear change of the coordinates z'
there is a neigborhood U = U(z~) c G, a domain G' in the space of the
variables Zd+2, ... ,Zn, and a pseudopolynomial w' over G' without multiple
factors such that;
1. N' n U = {w' = O}.
 5. Irreducible Components 137

2. N n (Cd x U) is an embedded-analytic set of dimension n - d - 1.

PROOF: We use the notation and results from the proof above. Thus the
first statement is clear. We set ~+! := w'. Restricting the Wi to Cd x (N'nU)
and projecting them down to Cd X G', we get pseudopolynomials ~i(Zi;Z")
over G'. Then N n (Cd x U) is in the joint zero set of ~l, .. · .!l::!d+l. Let
A be the union of those irreducible components of this set that contain the
unbranched points of N. Since N is the closure of unbranched points, it
follows that N n (Cd+! x G') cA. By the mapping theorem that we prove
in the next paragraph, every irreducible component of A is in N. So we have
the desired equality. -

Images of Embedded-Analytic Sets. Assume that G = Cd X G' c

Cn is a domain and A eGan irreducible embedded-analytic set over G'.
5.3 Mapping theorem. Let G 1 = C dl x G t c C nl be a domain, Al C G l
t,
an embedded-analytic set over G F a holomorphic mapping from a neigh-
borhood of A C G into G 1 such that F(U) C Al for some nonempty open
subset U C A. Then F(A) CAl.

PROOF: We denote by D c G the union of the discriminant sets of the

Wi(Zi,z') that define the surrounding set A for A. It is sufficient to prove
that F(A n (Cd x (G' - D») c AI. Since we can connect two points of
A n (Cd x (G' - D» by a chain of arbitrarily small balls, it is enough to give
the proof for such a ball. So we may replace A by a ball in Cn - d and may
assume that F is defined in a neighborhood of B.
Let Al be an embedded-analytic set in

Al = {Wl(Wl, w') = ... = Wd l (Wdp w') = OJ.

Then Wi oFlu == 0, and by the identity theorem Wi oFIB == O. So F(B) CAl.
t
For an arbitrary point v E G we choose a small transformation of the
coordinates in C dl such that Al is also embedded in a set Ar
= {wr = ... =
wdl = OJ. The transformation can be made arbitrarily small, and we can do
it so that
Al n AI n (C dl x {v}) = Al n (Cdl x {v}).
Then Al is given by the infinite set of holomorphic equations Wi = 0, wi = 0,
v E Gt . If F maps a nonempty open part of B into AI, then by the identity
theorem Wi 0 F = wi 0 F == 0, and consequently F(B) C AI. This completes
the proof. _

Remark. Assume that A is an analytic set in a domain G C cn

and that
Zo is a point of A. If Zo is a regular point of dimension n - d, then there is a
neighborhood U = U(zo) c G such that An U is an embedded-analytic set.
138 III. Analytic Sets

In fact, there is a neighborhood U with holomorphic functions 11, ... ,/d such
that N(/I, ... , Id) = An U and the rank of the Jacobian is d everywhere.
We may assume that

det (fi)Zj (zo) I J == 11'",,,.,. , dd ) =F O.

i.

Then thetransformationF(zl,'" ,zn) = (/I(z), ... ,/d(Z),Zd+l,'" ,zn) maps

a neighborhood of Zo biholomorphically onto a neighborhood of the image
point. If the inverse is given by

z = F-l(w) = (gl (w), ... , gd(W), Wd+lo' .. ,wn ),

then A is given by the equations

Zl = gl (0, ... ,0, Zd+l,"" zn),

So A n U is an embedded-analytic set.

Local Decomposition. We use embedded-analytic sets to show that an

arbitrary intersection of analytic sets is again an analytic set.
First we consider the following situation. Let G c en
be a domain and Zo E G
a point. Assume that at Zo a set ,9' of local analytic functions I is given such
that
1. For every lEY there is a connected open neighborhood U(zo) C G
with I E O(U) and I ~ O.
2. I(zo) = O.
We want to construct a "maximal" analytic set S* in a neighborhood
U*(zo) C G such that for each zero set N of finitely many elements I E ,9'
there is a neighborhood V = V(zo) c G with S* n V c N n V. Then S* is
uniquely determined near Zo and can be considered as the common zero set of
the functions lEY. It may be nontrivial even if the domains of definition of
the functions I tend to the point zoo For example, if,9' is the set of the func-
tions In(z) := zi'/(l-nzIl, defined on Un := {z E en : Re(zl) < lin}, then
S* is the analytic set {Z1 = O} in an arbitrary neighborhood of 0, whereas
the intersection of the Un does not contain any neighborhood of the origin.
We employ the results from the beginning of this section several times and
carry out an induction on the codimension of the embedded-analytic sets
obtained from the functions lEY.
 5. Irreducible Components 139

(A) We begin with one arbitrarily chosen function IE .7. The equation 1= 0
gives an analytic setS S of codimension 1. We decompose S into irreducible
components Si in a neighborhood U(zo) (given by pseudopolynomials in C x
C'), and we choose the neighborhood U so small that the Si stay irreducible
in the whole neighborhood.
(B) Next we try to obtain codimension 2. If every function lEY vanishes
identically near Zo on Si, we leave Si unchanged (and have it as a codimension
1 component for our S*). Otherwise, there is an l' E Y that does not vanish
identically in any small neighborhood of Zo on Si. We apply Theorem 5.2:
After an arbitarily small linear change of the coordinates z' the set Si n {I' =
O} is a finite union of irreducible embedded-analytic sets Sij of codimension
two, which stay irreducible if we pass to some smaller neighborhood ofzo. The
Sij are embedded in the zero set of two pseudopolynomials w~j (Zl j Z3, •.. , zn)
and W~(Z2j Z3, •.. , zn).
(C) Now codimension 3 follows. For this we need consider only the Sij.
Leave Sij unchanged if every I vanishes .on Sij (and get codimension 2
components for S"). Otherwise, find an 1" E .7 not vanishing identically
on Sij, and (after an arbitrarily small coordinate change of the variables
z" = (Z3, •.. , zn)) the set Sij n {I" = O} is the union of a finite set of
irreducible embedded-analytic sets Sijk, given in the zero set of three pseu-
dopolynomials W~k(ZA; Z4, .. " zn), A = 1,2,3.
(D) Continuing, it is possible to obtain components of codimension 1, 2,
... , n - 1, and finally one reaches dimension O. If there is a I-dimensional
component S = Sil ... i n _ 1 such that not every I E .7 vanishes on S, we
have to replace S by the one-point set {zo}. Then the procedure stops. Only
finitely many steps were necessary.
We obtained a finite system .70 of local holomorphic functions I, 1', I", .. .
and a finite system A of irreducible embedded-analytic sets Si, Sij, Sijk, .. .
and may assume that they all are defined in one neighborhood U(zo) c C,
that every SEA of dimension d is embedded in a set C n - d X Cd' and that
the union discriminant sets Ds C Cd C Cd belong to the embedding of S.
The necessary linear coordinate change in z, can be made at the beginning
of the procedure, i.e., once for all steps of the procedure.
If SEA is an irreducible embedded-analytic set that has an open part in the
union of the other sets of A, then it also has an open part in an irreducible
S' E A, S' =1= S. It follows by the mapping theorem that it is completely
contained in S'. Then we simply throw it away and denote the new system
again by A. After finitely many steps we have that the intersection of every
5 An exact definition of dimension and codimension of analytic sets will be given
later. Here we use embedded-analytic sets, for which the dimension has already
been defined. An analytic hypersurface is obviously embedded-analytic.
140 III. Analytic Sets

S with the union of the rest of A is nowhere dense in S. Moreover, the points
of S over Ds are nowhere dense in S.
We denote by S* = USE.A S the union of all remaining components. Then
S* is given by the finitely many holomorphic functions / E .9'0. Therefore,
it is an analytic set. If .Y C .9' is an arbitrary finite subset, then in a small
neighborhood of Zo every / E .Y vanishes at every z E S*. So S* C N(.Y).
Obviously, S* is uniquely determined by this property.
Finally, we want to show that the decomposition into irreducible embedded-
analytic sets is unique. For that we use the notion of regularity for points of
embedded-analytic sets just as in the analytic case. Clearly, the intersection
of two different Si E A contains no regular point. So the points of every
. . .
SEA are regular if they are not in such an intersection and not over D A.
We denote the set o/regular points of S* by S* and set S = SnS* for SEA.
Then for SEA the sets S are the connected components of S*. Since the
set S* is uniquely determined in a neighborhood of zo, this is also true for
its connected components. And since the closure of S is S, the irreducible
embedded-analytic components S are .also uniquely determined near zoo
5.4 Theorem. The intersection 0/ (even infinitely many) analytic sets is
an analytic set and is locally a finite union 0/ components that are irreducible
embedded-analytic sets. This decomposition is locally uniquely determined.

PROOF: Let {A. : tEl} be a family of analytic sets in a domain G C c;n,

and Zo E A := n.EIA. an arbitrary point. We consider the system .9' of all
local holomorphic functions / such that:
1. / is defined in an open neighborhood U of Zo (depending on f).
2. / '1= 0 near zoo
3. There is an tEl such that / vanishes near Zo on A •.
As above, Zo is contained in an analytic set S* that is the union of irreducible
embedded-analytic sets S and that is given by a finite subsystem .9'0 C .9'.
If z is a point of A that is sufficiently near zo, then every / E .9'0 is defined at
z and vanishes on some A. and consequently at z. This shows that A C S* in
a neighborhood of zoo On the other hand, let z be a point in the intersection
of S" with a small neighborhood of zoo Any analytic set A. is given by finitely
many holomorphic functions It, ... '/N that belong to the system .9'. Then
by construction every /~ vanishes on every embedded-analytic component S
of S .. , in particular at z. Therefore, S* C A. for all t. Thus S* is contained
in the intersection A of the A., and we have the equality A = S" near zoo
Since S" is an analytic set that has a unique decomposition into irreducible
embedded-analytic sets, this completes the proof. _
 5. Irreducible Components 141

Analyticity. Now we are able to prove the following result, which we

announced at the beginning of the section:
5.5 Proposition. Every embedded-analytic set A in a domain Cd x G c en
is an analytic set.

PROOF: As in the last part of the proof of the mapping theorem, it follows
that the embedded-analytic set A is given as the joint zero set of infinitely
many holomorphic functions. Theorem 5.4 shows that A is an analytic set. -
Consequently, every analytic set has locally a unique decomposition into ir-
reducible analytic components.

The Zariski Topology. We prove that the system of all analytic sets
has the properties of the system of closed sets of a topology.
5.6 Theorem. The system A of all analytic sets in a domain G c en has
the following properties:
1. G and the empty set belong to A.
2. If A l , ... , Al E A, then also A = U!=l Ai E A.
3. If I is an index set and {A. : £ E I} a collection of analytic sets in G,
then A = n. El A. is also an analytic set in G.

PROOF:

(1) G is defined by the zero function, and 0 by the constant function 1.

(2) Let z E A = Al U ... U AI. Then in a neighborhood U(z) there are
holomorphic functions Ai : i = 1, ... , I j = 1, ... , di , such that for all i we
have
Un Ai = N(Al,"" Ad;)'
It follows that UnA = N(Jl,il ... Ail: ji = 1, ... , di ).
3) This is Theorem 5.4.
So the analytic sets are the closed sets of a topology in G. We call this
-
topology the (analytic) Zariski topology of G. It plays an important rule in
complex algebraic geometry.

Global Decompositions. Assume that G c en is a domain and A c G

an analytic subset. We call A irreducible if the set of regular points A C A is
connected. It follows that A has the same dimension d at all regular points.
This number d is called the dimension of A and is denoted by dim(A). Every
irreducible embedded-analytic set is also an irreducible analytic set.
5.7 Theorem. Every analytic set A has a unique decomposition into count-
ably many irreducible analytic subsets Ai' The covering .rd = {Ai : i =
1,2,3, ... } is locally finite.
142 III. Analytic Sets

PROOF: We decompose A into connected components. Let A' be such a

component. It has dimensiond in all of its points.
We consider a point Zo E A that lies in A'. In a neighborhood U = U(zo) c
G we have a decomposition of A into finitely many irreducible embedded-
analytic components A l , ... , Am. By Ai we denote the set of points of Ai
that are not over the union discriminant set. Some d-dimensional Ai meet
A'. Their union A * is contained in A' and dense in A' n U. Hence, the closure
of A* in U is equal to A' n U. But A* is an analytic set.
From this it follows that A' is an analytic set, that only finitely many A'
intersect U, and that the union of all A' is A (as it is locally). Since the
topology of G is countable, it follows that the set of the A' is countable. _

5.8 Corollary. If A is irreducible and A = Al U A 2 , where A l , A2 are

arbitrary analytic sets, then A = Al or A = A 2 .
Sometimes this condition is used as the definition of irreducibility.
5.9 Proposition. Let A, BeG be irreducible analytic sets. If there is an
open set U c G such that A n U =f. 0 and A n U c B n U, then A C B.

PROOF: This is an immediate consequence of the mapping theorem. _

Another corollary is the following:
5.10 Identity theorem (for analytic sets). Let A, BeG be irreducible
analytic sets. If there is a point Zo E A n B and an open neighborhood U =
U (zo) c G with A n U = B n U, then A = B.

5.11 Proposition. Let A, BeG be analytic subsets with A c B. If A is

irreducible, then A is contained in some irreducible component of B.

PROOF: Let B = UAEA BA be the unique decomposition into irreducible

components. We can choose an open set U C G and a finite set {>'l, ... , AI} c
A such that UnA =f. 0 is irreducible and UnB = (UnBA\)U·· ·u(UnBA,).
Then UnA = UnA n B = (U nAn BAJ u ... u (U nAn BAI)' Thus there
is an index j such that UnA = UnA n B Aj , so UnA c un B Aj • It follows
that A is contained in B Aj ' -

Now we can generalize the notion of the dimension to arbitrary analytic sets.

Definition. If A eGis an analytic set with irreducible components

Ai, then dim(A) := SUPi dim(Ai} is called the (complex) dimension of A.
 6. Regular and Singular Points 143

In general, the dimension of an analytic set can be 00. But if G I cc G is a rel-

atively compact subdomain, then only finitely many irreducible components
intersect GI. So the dimension of A n G I is finite.
An analytic set is called pure-dimensional of dimension d if all its irreducible
components have the same dimension d.

Exercises
1. Let A be an analytic set near the origin in en. Assume that every ir-
reducible component of A has dimension ~ 1. Show that there exists
a neighborhood U = U(O) such that An U is the union of irreducible
one-dimensional analytic sets containing o.
2. Consider A := {(z}, Z2) E e 2 : z~ = z~ + zn. Show that A is irreducible,
but A has a nontrivial decomposition into irreducible components in a
small neighborhood of the origin.
3. Let AI, A2 c en be analytic sets. Prove that Al - A2 is analytic.
4. Let {Ai : i E N} be a locally finite family of irreducible analytic. sets in
a domain G c en. Suppose that Ai <t. Aj for i =F j and prove that the
Ai are the irreducible components of their union.

6. Regular and Singular Points

Compact Analytic Sets. Our goal is to prove the following simple
proposition.
6.1 Proposition. IIG c en is a domain and A eGan irreducible compact
analytic set, then A consists 01 a single point.
We first prove a lemma.
6.2 Lemma. II I is a holomorphic junction in a neighborhood 01 A, then
IIA is constant.
PROOF: We assume that the dimension of A is n - d. Since A is compact,
there is a point Zo E A where III takes its maximum. After a linear coordinate
change there is a neighborhood U = U(zo) c G and a domain G' c en - d
such that A n U is an embedded-analytic set over G'. Denote by D c G' the
union discriminant set. Over every z" E G' - D our A n U has s points. The
point Zo lies over some z~ E G' and we may assume that it is the only point
of A n U over z~.
It remains to construct the elementary symmetric functions associated with
IIAnu over G' - D. For this, if Zl, .•• , Z" are mapped onto z", we define
li(z") := O'i(f(ZI), ... ,/(z,,». These functions are holomorphic on G' - D
and bounded along D. So they extend to holomorphic functions in G'. The
144 III. Analytic Sets

absolute value of every extension takes on its maximum at z~. By the max-
imum principle each such is constant in G'. Since the values of f can be
reconstructed from the values of the fi, it follows that f is constant over G',
. .
in particular in some open subset of A. Since A is connected, it follows that
f is constant on A and then by continuity also on A. _

PROOF of the proposition: All coordinate functions Zi must be constant on

A. So A is a single point. _

A consequence is that every compact analytic subset A c G consists of only

finitely many points.

Embedding of Analytic Sets. Assume that A eGis an analytic set

in a domain G c en, that 0 E A, and that the plane P = {Zd+ 1 = ... =
Zn = O} intersects A in an isolated point.
6.3 Theorem. In a neighborhood U(O) the set A is an analytic subset of
an embedded-analytic set of dimension n - d that is defined over a domain G'
in the space of variables z' = (Zd+1,' .. ' zn). If the set of (n - d)-dimensional
regular points is dense in A, then A is itself an embedded-analytic set.

Remark. No coordinate transformation is necessary for this statement!

PROOF: By definition, A is the zero set N(/t, ... , iN) of finitely many
holomorphic functions in a neighborhood of o. Since An P = {O} is isolated,
there is an i such that Ii does not vanish identically in any neighborhood
of 0 on the zl-axis. Consequently, Ii is zl-regular, and we can apply the
Weierstrass preparation theorem, which implies that A is locally contained
in the zero set of a pseudopolynomial W(Zl; Z2, ... ,Zn)'
Now we proceed by induction on d. In the case d = 1 there is nothing to prove.
If d > 1 we consider the projection 11" : en -t en- 1 with z 1-+ Z' = (Z2, • •. ,zn).
N
To h, ... , f there are associated as usual holomorphic functions 2 , ••• , L LN
of z' such that

1I"({W = h = ... = fN = OJ) = A := {z' : -2

f (z') = ... = f (z')
-N
= O}

in some neighborhood of O.
The intersection of P' = {z' : Zd+l = ... = Zn = O} with A contains 0 as an
isolated point. So we have for A the same situation, but with one dimension
fewer. By the induction hypothesis it follows that there are pseudopolynomi-
als W2,'" ,Wd such that A locally is contained in the set {W2(Z2;Z") = ... =
Wd(Zd; z") = OJ. By exchanging ZI with Z2, we obtain a pseudopolynomial
WI(ZI; z") such that A is contained in {WI = W2 = ... = Wd = OJ.
 6. Regular and Singular Points 145

If the regular points of dimension n - d are dense in A, we can take the union
A * of those irreducible components ofthe embedded analytic set that contain
such a regular point. Then A * and A are identical. This completes the proof.
•
Remark. This result is also known as the "embedding theorem of Remmert
and Stein."
Again we consider a domain G c cn, an analytic set A c G, and a domain
G' c C n- d such that 7r(G) C G' (where 7r : z t-t z' = (Zd+1,"" zn»). Suppose
that there exists a domain G* c G such that for every z' E G' there exists a
neighborhood U = U(z') cc G' with (Cd x U) n G* CC (Cd X U) n G and
(Cd x U) n A C G*. Then the following holds.
6.4 Proposition. If the set of regular (n - d)-dimensional points of A is
dense in A, then A is an embedded-analytic set over G'.

PROOF: We take an arbitrary point Zo E G'. It follows from the hypotheses

that the set (Cd x {zo}) n A is compact and analytic. Therefore, it consists
of finitely many points. Each of these points has a neighborhood such that
the restriction of the set A to this neighborhood is an embedded-analytic set
over a neighborhood of zoo By multiplying the pseudopolynomials by the same
distinguished variable belonging to our various points we obtain d uniquely
determined pseudopolynomials over a neighborhood U'(zo) C G' such that
A n (Cd x U') is just their joint zero set. But these pseudopolynomials glue
together to form global pseudopolynomials over G'. •

Regular Points of an Analytic Set. Assume again that G C C n is

a domain, and A eGan analytic set.
6.5 Theorem. For any Zo E A there is a fixed neighborhood U(zo) C G
with finitely many holomorphic functions iI, ... , IN whose joint zero set is
An U such that for all d at every regular point z E An U of dimension n - d
the rank of their Jacobian at z is equal to d.
It is remarkable that this statement is also true for a singular point Zo of A.
PROOF: After applying a linear coordinate transformation in en we can
find a neighborhood U = U(zo) c G such that An U is a finite union of
irreducible embedded-analytic components. To give these in the canonical
form a further coordinate transformation is not necessary. We denote by
A' the union of all (n - d)-dimensional irreducible components of A n U
and choose pseudopolynomials Wl, ... ,Wd of minimal degree such that A' is
contained in the common zero set of the Wi'
Let jj C A' be the set of points that lie over the union discriminant set
D A'. Its dimension is equal to n - d - 1. Now we carry out the proof in
146 III. Analytic Sets

several steps and construct sets Db D 2 , ... , Dn-d+l = 0 with D i + 1 C Di

and dim(13 i ) = n-d-i. We begin with 13 1 := D. The Jacobian of WI, ... ,Wd
has rank d in A' - 13 1 (implicit function theorem).
Next we decompose Dl into irreducible components ex and choose, if possi-
ble, in each ex a point Zx where A' is regular. We can make U so small that
only finitely many ex occur in U. Then we apply another linear transforma-
tion inen that is near the identity such that in a small neighborhood of any
z.\ the set A' can be written as an (n - d)-dimensional holomorphic graph

{z : Zi = h,i(Zd+l,"" zn) for i = 1, ... , d}.

Finally, we apply a linear transformation that is very near the identity to
the variables Z1, ... ,Zd such that for every A and every point z in A' above
(Z~I" .. ,z~) the first d coordinates of z are distinct.
Now we use Proposition 6.4. In the new coordinates (and in a slightly
smaller neighborhood, which we again denote by U) the set A' is again an
embedded-analytic set contained in the common zero set of pseudopolynomi-
als WI, ... , Wd. We choose Wi with minimal degree. We can assume that the
components of 13 1 are still irreducible in U, and that the points z.\ are still
in U. In a neighborhood of Zx we have a decomposition

with wA,i(Z.\) =I- O. So the Jacobian determinant of W1"",Wd with respect

to the variables Z1,' .. ,Zd does not vanish at any z).. We denote the zero
set of this Jacobian in 13 1 n U by D2 • It has dimension n - d - 2, and
WI, .. ' ,Wd,Wl, ... ,Wd have rank d on A' - 132 •
Now apply the same procedure to 13 2 and obtain an (n - d - 3)-dimensional
133 and continue in this way until reaching 13n-d+l = 0.
By putting all of the pseudopolynomials together, in a small neighborhood
U of Zo we get holomorphic functions ft,.··, Id, Id+l,' .. , IN (with N =
(n - d + 1) . d) whose rank is d in every regular point of A' n U. Since the
pseudopolynomials always were chosen with minimal degree, it follows that
A' = N(ft, ... , IN) near zo°
Now set A + = A n U - A'. It is the union of the remaining irreducible com-
ponents of AnU. We may assume that U is so small that A+ is the common
zero set of finitely many holomorphic functions 91, ... ,98 in U. Multiplying
the h by the 9j yields finitely many holomorphic functions in U that de-
scribe the set An U. No point of A' n A+ is a regular point of A. For every
zEA' - A+ there is a 9j that does not vanish there. So the rank of the
Jacobian of the Ii . 9j is equal to d at every nonsingular point of A' - A + .
 6. Regular and Singular Points 147

The same procedure can be used for every d and the corresponding A', and
consequently, in finitely many steps we obtain a neighborhood of Zo and
holomorphic functions in this neighborhood with the desired properties. _

The Singular Locus. From the preceding theorem we conclude the

following:
6.6 Theorem. The set Sing(A) of singular points of an analytic set A is
again an analytic set.

PROOF: The intersection of two irreducible components of A belongs to

Sing(A). The union S of all these intersections is an analytic set.
Assume that Zo is a point of an irreducible component A' of A and dim(A') =
n - d. Then there is a neighborhood U = U(zo) c G with holomorphic
functions iI, .. . , f N vanishing exactly on A' n U such that their Jacobian
has rank d in each of the regular points. Let S* be the analytic set of all
points of A' n U where all d x d minors of the Jacobian vanish. Clearly, S* is
contained in Sing(A') n U. On the other hand, at any point of Sing(A') n U
the Jacobian of iI, ... , fN cannot have rank d. So Sing(A') n U = S*, and
Sing(A') is analytic in G.
The union of S and the sets Sing(A') for all irreducible components A' is the
set Sing(A). It is analytic, since the union is locally finite. _
The set Sing(A) is called the singular locus of A.

Extending Analytic Sets. Let G c cn be a domain.

6.7 Lemma. Let Zo = (ziO), ... , z~O») C C n be an arbitrary point and E =
{z : Zi = z;O) for i = 1, ... , d} an affine plane of codimension d containing
Zo. If A eGis an irreducible analytic set of positive dimension that is not a
subset of E, then there is an open dense subset C C Cd such that

fe(zl, ... , Zn) := Cl (Zl - ziO») + ... + Cd(Zd - z~O»)

does not vanish identically on A for every c = (Cl' ... , Cd) E C. In particular,
for any hyperplane Ho C C n containing E there is a hyperplane H arbitrarily
close to Ho and also containing E such that dim(Ai) :s; dim(A) - 1 for every
irreducible component Ai of A n H.

PROOF: We define cp : Cd ---+ o(cn) by cp(c) := fe. This is a C-linear map,

and V := {c E Cd : felA == O} is a linear subspace. Suppose that V = Cd.
Then (Zi - ziO»)IA == 0 for i = 1, ... , d, and therefore ACE. This is a
contradiction, and consequently, V must be a proper subspace of Cd. For any
c in the open dense subset C := Cd - V, fe does not vanish identically on A.
Thus He := {z : fe(z} = O} is a hyperplane containing E. _
148 III. Analytic Sets

Our main tool for extending analytic sets is the following.

6.8 Proposition. If E = {z E en : ZI = ... = Zd = o} is an (n - d)-
dimensional plane and A an analytic set in G - E, whose irreducible compo-
°
nents all have dimension n - d + l with < I < d, then the closure A of A in
G is an analytic set in G.

PROOF: The proposition is of a local nature. We may assume that A is

irreducible and 0 E EnG. It is enough to construct a continuation of A into
a neighborhood of o.
Let c = (0, ... ,0, Cd+!, ... , en) be an arbitrary point of EnG. We consider
the following family of (d - I)-dimensional planes through c: For

A = ( aij I i = 1, ... ,n - d+ l) E M (If")

J. -_ 1, ... , d _ l n-d+l,d-l IV

we have the linear map LA : ed - l -+ e n - d +l and define

P(c, A) := c + {(w', LA(W')) : w' E Cd-I}.

So P( c, A) consists of vectors w = (WI, ... ,Wd-l, Wd-/+1, ... ,wn ) with

d-l
Wd-l+i L a i j · Wj for i = 1, ... ,l,
j=1 d-l
Wd-/+i Cd-/+i + Laij . Wj for i = l + 1, ... ,n - d + l.
j=1
Then every P(c, A) meets E exactly in c = (0, ... ,0, Cd+l, ... , cn ). If 0 is the
zero matrix, then P(c) := P(c, 0) is the plane ed-1x{(O, ... ,0,Cd+b ... ,cn )}.
In the next chapter we will introduce Grassmannian manifolds and a topology
on the set of linear subspaces (with fixed dimension) of a given vector space.
In our case it follows that a neighborhood of c + Po is given by the set of all
e
planes P = c+P with PEB(O x n - d +l ) = en. This shows that every (d-l)-
dimensional plane through c that is near P(c, 0) is of the form P(c, A).
We choose real numbers °< Tl < T2 and T > ° so small that the "shell"

s = {z = (z',z") E ed - l x en - d +1 : Tl < IIz'll < T2 and Iz"l < T}

is a relatively compact open subset of G. Only finitely many irreducible com-
ponents Ai of A enter S. We can find a hyperplane HI containing E that
intersects the Ai not at all or in codimension 1. Let Aij be the finitely many
irreducible components of HI n Ai that enter S. We can find another hyper-
plane H2 containing E that intersects all Aij at most in codimension 2. We
continue this procedure running through the irreducible components of the
Aij n H 2 . After finitely many steps we have n - d + l hyperplanes such that
their intersection P is a plane P( c, A) that meets An S in at most finitely
 6. Regular and Singular Points 149

many points. By the above lemma we can choose this plane arbitrarily near
Po := P(c).
Now apply a linear coordinate transformation very near to the identity that
leaves E invariant and maps our plane P(c, A) onto Po and replace the
transformed shell S by a new shell S' (in the new coordinates) that is a little
bit smaller such that S' is contained in the old (transformed) S. This can be
done so that as' n Po n A = 0.
Earlier we proved that A is an embedded-analytic set in a neighborhood of
the points of the intersection A n Po n S' over a domain G' in the space of
variables Zd-l+l, . .. ,Zn' So there is a small closed ball B C C n - d + 1 around
the origin such that (C d - l x B) n as' n A remains empty, every irreducible
component of (C d - 1 x B) n S' n A enters Po n S', and every plane through
a point of B and parallel to Po intersects A n S' in at most finitely many
points.

E
S'

An Bo

Figure III.3. Intersecting A with Bo and Po

Now, the set Bo := {z' E e d - l : IIz'lI < r2} x B is a neighborhood of

the origin in en, and each of our parallel planes through points of B - E
meets A n Bo in a compact analytic set and therefore in at most finitely
many points (cf. Figure III.3). At all of these points the set A is locally
an embedded-analytic set over B. By multiplying the pseudopolynomials by
the same distinguished variable that we-obtained for the different intersection
points over the same base point in B - E, we have that An (e d - 1 x (B - E)) is
150 III. Analytic Sets

an embedded-analytic set over B - E. The coefficients of the corresponding

pseudo polynomials over B - E are bounded along E. Hence, they can be
analytically extended to B. This means that A n (Bo - E) has a unique
analytic continuation to Bo. •
The proposition just proved is also true if A c G - E is an analytic set, whose
irreducible components all have dimension greater than n - d, since we can
write A as a finite union of pure-dimensional analytic sets. As a consequence
we have the following theorem.
6.9 Theorem of Remmert-Stein. Assume that G c en is a domain,
KeG an (n - d)-dimensional analytic subset, and A an analytic subset of
G - K all components of which have dimension> n - d. Then the closure A
of A in G is an analytic set in G.

PROOF: If Zo E K is a regular point, then K can be transformed in a

neighborhood of Zo to a plane E. So A is analytic in a neighborhood of Zo
and therefore in all regular points of K. We can replace K by the set KI of
singular points of K, which is analytic again and has lower dimension. By
the same argument we show that A is analytic at all regular points of K I .
Continuing in this way we prove that A is analytic in G. •
This theorem first was proved by R. Remmert and K. Stein; see [ReSt53].

The Local Dimension. We show how our results are linked with the
classical dimension theory of analytic sets.
Let G c en be a domain, A eGan analytic set, and Zo E A a point. There
is an open neighborhood U = U(zo) c G such that UnA is a finite union
of irreducible analytic components AI, ... , AI. If we choose U small enough,
then the Ai are uniquely determined.

Definition. In the given situation the uniquely determined number

dimzo(A):= max dim(A.x)

,\=1, ... ,1

is called the (local) dimension of A at Zoo

The set A has dimension 0 at Zo E A if and only if there is an open neigh-

borhood U = U(zo) c G such that UnA = {zo}.
6.10 Proposition. Let k := dimzo (A) be positive. Then k is the smallest
number with the property that there are holomorphic functions II, ... ,ik in
a small neighborhood U of Zo such that Zo is isolated in AnN (II, ... , ik).

PROOF: If dimzo (A) = k, then there must be at least one k-dimensional

irreducible component A' of A at Zo.
 6. Regular and Singular Points 151

If f is any holomorphic function near zo, then either flA' == 0 (and therefore
AnN (I) still k-dimensional) or A' n N (I) is (k - 1)-dimensional. So at least
k functions are required.
On the other hand, by Lemma 6.7 we can find a holomorphic function h
near Zo that does not vanish identically on any irreducible component A' of
dimension k at Zoo It follows that A' n N(h) has dimension k - 1 for all
such components A'. We can repeat this process, and after k steps we reach
dimension zero, so that Zo is isolated in An N(h,···, /k). _

Definition. If A has dimension k at zo, then any system {h,·.·, /k} of

holomorphic functions with AnN (h, ... , /k) = {zo} is called a pammeter
system for A in Zo.

6.11 Ritt's lemma. Let B c A be closed analytic sets in a domain G c

en. Then B is nowhere dense in A if and only if dimz (B) < dimz (A) for
every Z E B.

PROOF: Let the criterion be fulfilled, and Zo be an arbitrary point of B.

Then there exists an open neighborhood U of Zo in G and a parameter system
{II, ... ,fk} on U for B at Zoo Since dimzo (A) > k, it is not possible that Zo is
isolated in AnN(h, ... , fk). This means that (A-B)nN(h, .. ·, fk)nW =f:. 0
for every neighborhood W = W(zo). So B is nowhere dense in A.
On the other hand, let B be nowhere dense in A, and Zo a point of B. In a
small neighborhood U of Zo we have unique decompositions into irreducible
components:

B n U = BI U ... U Bm and A = Al U ... U A l .

Every component Bi is contained in a component Aj(i) , and for any open

neighborhood W = W(zo) we have (Aj(i) - B i ) n W =f:. 0, because otherwise,
there would exist points Z E W n Bi where B is dense in A. So dim(Bi ) <
dim(Aj(i) for all i. It follows that dimzo(B) < dimzoCA). _

Let G c en be a domain, A eGan analytic set, and Zo E A a point. If

dimzo CA) = k, then the number n - k is called the codimension of A at zoo
6.12 Second Riemann extension theorem. Suppose that n 2:: 2 and
that the analytic set A c G has everywhere at least codimension 2. Then any
holomorphic function f on G - A has a holomorphic extension to G.

PROOF: We may assume that A is irreducible of codimension d 2:: 2. If Zo

is a regular point of A, then there is a neighborhood U of Zo such that UnA
152 III. Analytic Sets

is biholomorphically equivalent to an open subset of a linear subspace E of

codimension d. By the theorem on removable singularities (see Section 11.1)
J can be holomorphically extended to Zoo
We repeat this procedure. Beginning with the set Sing(A), which has codi-
mension d + 1, after finitely many steps only a set of isolated points remains.
Since I can also be extended to these points, we obtain the desired result. _

Exercises
1. Let A l , A2 c en be analytic sets. Show that dim(A 1 n A 2) ;:::: dim(Al) +
dim(A 2 ) - n.
2. Assume that G c en is a domain and I a nonconstant holomorphic
function on G. Prove that there is an at most countably infinite set
Z C e such that Ac := {z E G : J(z) = c} is regular for c E e - z.
3. Let G c en be a domain and A eGan analytic set. Prove that for any
k ;:::: 0 the closure of the set Ak := {z E A : dimz(A) = k} is either
empty or is a pure k-dimensional analytic subset of G.
4. Let G c en be a domain and ft, ... , 1m holomorphic functions on G.
Denote by N the common zero set N(ft, . .. ,1m). Show that if the rank
of the Jacobian of ft, ... , 1m at some point Zo E N is equal to r, then
there is a neigborhood U = U(zo) c G and a closed submanifold M C U
of dimension less than or equal to n - r such that UnA eM.
5. Let A be an analytic set in a domain G c en. For every point Zo E A
the set Izo(A) := {(f)zo E Ozo : IIA == O} is an ideal in Ozo. Show that
there are a neighborhood U of Zo and holomorphic functions ft, ... , fk
on U such that:
(a) AnU=N(ft, ... ,fk).
(b) Izo(A) is generated by the germs (fdzo,···, (fk)zo.
Show that the vector space
n
Tzo(A) = {w E en : Lw,Jz,,(zo) = 0 for every I E Izo(A)}
v=l
has dimension n - rkzo (f1, ... , I k). It is called the Zariski tangential space
of A at zoo
6. Let A be defined as in Exercise 5, and suppose that Zo is a regular point
of A. Show that Tzo(A) is the set of tangent vectors nCO) (see Section
1.7), where a : I -+ en is differentiable, a(I) C A, and a(O) = zoo
7. Let A be an analytic set in a domain G c en, and let Zo E A be an
arbitrary point. The embedding dimension of A at Zo is the smallest
integer e such that there is an open neighborhood U = U(zo) and a
closed submanifold M C U of dimension e with An U eM. It is denoted
by embdimzo(A). Prove that Z H embdimz(A) is upper semicontinuous
on A, and that embdimzo(A) = dimdTzo(A)).
8. Consider the analytic set A = {w = exp(l/z)} C {(w, z) E e 2 : z -IO}.
Determine the closure of A in e 2 .
Chapter IV

Complex Manifolds

1. The Complex Structure

Complex Coordinates. Let X ~e a Hausdorff space, i.e., a topological
space satisfying the Hausdorff separation axiom. Sometimes such a space is
also called a separated space or a T 2 -space. Hausdorff spaces are the most
common in topology (for example, every metric space is a Hausdorff space),
but non-Hausdorff spaces do arise, in particular in algebraic geometry. The
space en with the Zariski topology is not Hausdorff.
We think that a space X is too big if there exists a discrete subset with the
cardinality of the continuum. Therefore, we demand that the topology of X
have a countable base. In this case X is said to satisfy the second axiom
of countability. Obviously, en has a countable basis. A metric space has a
countable basis if and only if it contains a countable dense subset.
A Hausdorff space X is called locally compact if every point x E X has a
compact neighborhood. ~f X is compact, then it is also locally compact. If
X is locally compact, but not compact, then X can be made compact by
adjoining just one point (Alexandrov's one-point compactification). Every
Hausdorff space that is locally homeomorphic to an open subset of en is
locally compact. So, for example, every Riemann domain over en is locally
compact.
Definition. An open covering "1/ = {Vv : 1/ E N} of a Hausdorff space
X is called a refinement of the covering %' = {U, : tEl} of X if there
is a map r : N -t I (the refinement map) with
Vv C Ur(v) for every 1/ E N.

The refinement map is not uniquely determined, but we can fix it once and
for all.
A covering "1/ = {Vv : 1/ E N} is called locally finite if each x E X has a
neighborhood U = U(x) such that U meets only finitely many Vv '
Definition. A Hausdorff space X is called paracompact if every open
covering %' of X has a locally finite open refinement "1/.

Every compact space is paracompact. Furthermore, every locally compact

space with countable basis is paracompact.
154 IV. Complex Manifolds

For the moment we only assume that X is a Hausdorff space.

Definition. An n-dimensional complex coordinate system (U, <p) in X
consists of an open set U C X and a topological map <p from U onto an
open set Been.

If p E X is a point, then every coordinate system (U, <p) in X with p E U

is called a coordinate system at p. The entries Zi in z = <p(p) are called the
complex coordinates of p (with respect to (U, <p)).
If f is a complex function in U, we can consider it as a function of the complex
coordinates Zl, ... , Zn, by

Two (n-dimensional) complex coordinate systems (U, <p) and (V, 'Ij!) in X are
called (holomorphically) compatible if either Un V = 0 or the map

is biholomorphic (see Figure IV.I).

Figure IV.I . Change of coordinates

The sets B1/J := 'Ij!(U n V) and Bf{J := <p(U n V) are open subsets of en. If Zi
(respectively Wj) are the complex coordinates with respect to <p (respectively
'Ij!), then compatibility of the coordinate systems means that the functions
Zi = Zi(Wl, ... , w n ) and Wj = Wj(Zb .. " zn) are holomorphic.
 1. The Complex Structure 155

A covering of X with pairwise compatible n-dimensional complex coordinate

systems is called an n-dimensional complex atlas on X. Two such atlases Al
and A2 are called equivalent if any two coordinate systems (U, cp) E Al and
(V, 'l/J) E A2 are compatible. An equivalence class of (n-dimensional) complex
atlases on X is called an n-dimensional complex structure on X. It contains
a maximal atlas that is the union of all atlases in the equivalence class.
Definition. An n-dimensional complex manifold is a Hausdorff space
X with countable basis, equipped with an n-dimensional complex struc-
ture.

Every complex manifold is locally compact and paracompact.

Examples
1. The complex n-space en is an n-dimensional complex manifold. The
complex structure is given by the coordinate system (en, id).
2. If X is an arbitrary n-dimensional complex manifold, then any nonempty
open subset B c X is again an n-dimensional complex manifold. For
p E B there is a coordinate system (U, cp) in X at p. Then (UnB, cplunB)
is a coordinate system in B at p. All of these coordinate systems are
compatible.
3. Let G c en be a domain and X eGa k-dimensional complex sub-
manifold. Of course, X is a Hausdorff space (in the relative topol-
ogy) with countable basis. For Zo E X there are open neighborhoods
W = W(zo) c G and B = B(D) c en and a biholomorphic map
F : W --+ B such that

F(WnX)={(Wl, ... ,Wn)EB: Wk+l=···=Wn=O}.

Let pr' : en --+ e k be the projection (WI. .. " w n ) t-+ (WI. ... , Wk). We
define U := W n X and cp := pr' 0 F : U --+ e k . Then (U, cp) is a k-
dimensional complex coordinate system in X at Zo.
If (V, 'l/J) is another coordinate system, with 'l/J = pr' 0 P, then

cp 0 'l/J-l(Wl, .. : ,Wk) = pr' 0 F 0 p- l (Wl' ... ,Wk, 0, ... ,0)

is holomorphic. So we get a complex structure on X.
4. Finally, let (G, 7r) be a Riemann domain (over en). Then G is a connected
Hausdorff space, and for every pEG there is an open neighborhood
U = U(p) such that B := 7r(U) is open and cp := 7rlu : U --+ B is
topological. Then (U, cp) is a complex coordinate system. If'l/J = 7rlv is
another coordinate system, then for x E Un V we have cp(x) = 'l/J(x) =
7r(x) =: z and
cp 0 'l/J-l(Z) = cp(x) = z.
156 IV. Complex Manifolds

Therefore, the coordinate systems are compatible. We get a complex

structure on G. One can prove that G has a countable basis (d. [Gr55],
§2]). So every Riemann domain over C n is an n-dimensional complex
manifold.

Holomorphic Functions. Let X be an n-dimensional complex mani-

fold.
Definition. A complex function I on an open subset B c X is called
holomorphic if for each p E B there is a coordinate system (U, cp) at p
such that 10 cp-I : cp(U n B) -+ Cis holomorphic.

If Zl, ... , Zn are the complex coordinates with respect to (U, cp), then

(ZI, ... , Zn) H 1 0 cp-I(ZI, ... , zn)

is a holomorphic function in the ordinary sense. If z.., = z..,(Wl. ... , w n ), where
WI, ... ,Wn are the complex coordinates with respect to a coordinate system
(V, '1/1), then
1 0 'I/I-l(Wll ... , W n ) = 10 cp-I(ZI(Wl, ... , w n ), ... , Zn(WI, ... , w n ))
is also holomorphic. So the definition of holomorphy is independent of the
coordinate system. We denote the set of holomorphic functions on B by O(B).
It is a C-algebra with unit element.
Example
Let G c cnbe a domain and X eGa k-dimensional complex submanifold.
We consider a complex coordinate system (U, cp) in X, where U is the inter-
section of X with an open set W C G and cp = pr' 0 F, with a biholomorphic
map F : W -+ B c cn
such that F(U) = {w E B : Wk+l = ... = Wn = O}.
If I is a holomorphic function on G, then

Ilx 0 cp-I(Wl. ... , Wk) = 10 F-I(Wl, ... , Wk, 0, ... ,0)

is holomorphic. Therefore, Ilx is a holomorphic function on the complex
manifold X.
1.1 Identity theorem. Let X be connected. II I, 9 are two holomorphic
functions on X that coincide in a non empty open subset U eX, then I = g.

PROOF: Let W = {x EX: I(x) = g(x)}. Then WO =/: 0, since U C W.

Assume that there exists a boundary point Xo of WO in X and let (U, cp) be a
coordinate system at Xo with cp(xo) = o. Then all derivatives of I 0 cp-l and
go cp-l must coincide at o. It follows that the power series of these functions
around the origin are equal. But then I = 9 in a whole neighborhood of xo,
and this is a contradiction.
 1. The Complex Structure 157

If there were a point x E M := X - WO that was not an interior point of M,

then x would be a boundary point of WO. This shows that M must be open.
Since X is connected, M has to be empty. -

1.2 Maximum principle. Let X be connected, IE O(X), and Xo E X a

point such that III has a local maximum at Xo· Then I is constant.
PROOF: The functions I and 9 := I(xo} are both holomorphic on X. If
(U, cp) is a coordinate system at Xo and B := cp(U), then 10 := 10 cp-l is
holomorphic on B, and 1/01 has a local maximum at Zo := cp{xo). Thus there
is an open neighborhood B' = B'(zo) c B such that 10 is constant on B' and
I is constant on U' := cp-l(B'}. So Ilu = glu, and by the identity theorem
I = g; i.e., I is constant. -

1.3 Corollary. II X is compact and connected, then every holomorphic

junction on X is constant.
PROOF: The continuous function III takes its maximum at some point of
X. Now the corollary follows from the maximum principle. _

1.4 Corollary. There is no compact complex submanilold 01 positive di-

mension in en.

PROOF: Let X c en be a compact connected submanifold. Then the stan-

dard coordinate functions z" Ix must be constant, for v = 1, ... , n. This
means that X is a single point. If X is not connected, it is a finite set of
~~. -
Remark. Another proof is given in Section 111.6.

Riemann Surfaces. An (abstract) Riemann surlace is by definition a

I-dimensional connected complex manifold.
Example
The complex plane e and every domain in e are Riemann surfaces. Recall
the Riemann surface of Vz,

X = {(w, z) E e2 : w 2 = z, Z f; O}.
Since X is a Riemann domain over e, it is a I-dimensional connected complex
manifold. From the projection 'Tr ;= pr21x : X -+ e we get complex coordinate
systems (U,cp) with cp:= 'Trlu and sufficiently small U.
158 IV. Complex Manifolds

The function 1:X ~ C with I(w, z) := w is a global holomorphic function

with 12 = z.
Example
The Riemann sphere C = c U {oo} is a compact connected Hausdorff space.
We have two coordinate systems (C,cp) and (C - {O},'1jI) with IO(Z) = z and
'IjJ(z) = liz. On C* = C n (C - {O}) we have I" 0 'IjJ-1(Z) = liz, and this is
holomorphic. So C is a compact Riemann surface. Every global holomorphic
function on C is constant, but there are nontrivial meromorphic functions,
for example I(z) = z (with one pole at (0). Here a meromorphic function on
X is a function f that is holomorphic outside a discrete subset P c X and
satisfies
lim If(x)1 = 00 for every pEP.
x-+p

The points of P are called the poles of I.

Holomorphic Mappings. Let F : X ~ Y be a continuous map be-
tween complex manifolds.
Definition. The map F defined above is called holomorphic if for any
p E X there is a coordinate system (U,cp) in X at p and a coordinate
system (V, 'IjJ) in Y at F(p) with F(U) c V such that

'IjJ 0 F 0 10- 1 : cp(U) -4 'IjJ(V)

is a holomorphic map.

1.5 Proposition. The map F : X ~ Y is holomorphic if and only il for

any open subset V C Y and any 1 E O(V) it follows that loF E O(F-l(U)).
The proof is an easy exercise.
The category 01 complex manifolds consists of a class of objects, the complex
manifolds, and a class of sets such that to any pair (X, Y) of objects there is
assigned a set O(X, Y) (which may be empty), the set of holomorphic maps
between X and Y. In a general category this set would be called the set of
morphisms from X to Y.
For (G,F) E O(Y,Z) x O(X,Y) we always have the composition GoF E
O(X, Z) such that the following axioms hold:
1. If HoG and Go F are defined, then (H 0 G) 0 F = H 0 (G 0 F).

2. For any manifold X we have the identity map idx E O(X, X) such that
idy 0 F = F and F 0 idx = F, if the compositions are defined.
 1. The Complex Structure 159

Another example for a category is given by the topological spaces and contin-
uous mappings. If we replace in our definitions above the field e by IR and the
word "holomorphic" by "differentiable", we get the category of differentiable
manifolds and differentiable mappings. From every n-dimensional complex
manifold we obtain a 2n-dimensional differentiable manifold by "forgetting
the complex structure."
A holomorphic function f : X ~ e is obviously a holomorphic mapping.
More generally, in the case of a Riemann surface a meromorphic function f
on X may be viewed as a holomorphic mapping f : X ~ c.
Definition. A biholomorphic map F : X ~ Y is a topolc.gical map
such that F and F- I are holomorphic. If there exists a biholomorphic
map between X and Y, then the manifolds are called isomorphic or
biholomorphically equivalent, and we write X ~ Y.

Remark. If X is a complex manifold and (U, cp) a complex coordinate

system with cp(U) = Been, then cp : U ~ B is a biholomorphic map.

Cartesian Products. Assume that Xl, ... ,Xm are complex manifolds
of dimension nl, . .. ,nm . Then the set X = Xl X ..• X Xm carries a natural
topology generated by the sets U = UI X ... X Um, Ui C Xi open for i =
1, ... , m. One sees immediately that X is a Hausdorff space with countable
basis.
Given complex coordinate systems (Ui , CPi) in Xi, for i = 1, ... ,m, one defines
a coordinate system (U, cp) for X by

It is clear that two such coordinate systems are compatible. So we obtain an

n-dimensional complex atlas and a complex structure on X. The projections
Pi : X ~ Xi are holomorphic maps for i = 1, ... ,m.
A simple example is en = ex· .. x Co
-----------
n times

The Cartesian product of two complex manifolds Xl, X 2 satisfies the follow-
ing universal property:
Given any complex manifold Y and any two holomorphic maps F : Y ~ X I
and G: Y ~ X 2 , there exists exactly one holomorphic' map H: Y ~ Xl X X 2
with F = PI 0 Hand G = P2 0 H.
Although trivial in our case (we set H := (F, G)), this property becomes
important in more general categories.
160 IV. Complex Manifolds

Analytic Subsets. Let X be an n-dimensional complex manifold.

Definition. A subset A c X is called analytic if for each point p E X

there are a (connected) open neighborhood U = U(P) and finitely many
holomorphic functions iI, ... ,Jm on U such that

unA={qEU: fJq)=Ofori=l, ... ,m}.

We call A an analytic hypersurface if we can always take m = 1.

From the definition it follows that A is a closed subset of X. Locally, an

analytic set in X is the same as an analytic set in an open set Been. So
most properties of analytic sets in en can be transferred.
1.6 Proposition. If X is connected and A c X analytic, then either A =
X or A is nowhere dense and X - A is connected.
PROOF: Assume that A #: X. If A is somewhere dense in X, then A contains
interior points (because it is closed in X). Since X is connected, the interior
of A has a boundary point p E X - A (same argument as in the proof of
the identity theorem). We take a connected neighborhood U = U(p) such
that An U = {q E U : iI(q) = ... = fm(q) = O}. Then U contains an
open subset V (consisting of interior points of A) where iI, ... , fm vanish
identically. By the identity theorem they vanish on the whole set U, and p
cannot be a boundary point of the interior of A. This is a contradiction, and
it follows that A is nowhere dense.
If X - A is not connected, it can be decomposed into two nonempty open
subsets UI, U2 • The function f : X -A -+ e with f(x) == 0 on U1 and f(x) == 1
on U2 is holomorphic and bounded. By Riemann's extension theorem (which
can be applied locally) there exists a holomorphic function on X that i
i
coincides with f outside A. Since can take only the values 0 and 1, it is
locally constant. But on the connected manifold X every locally constant
function is constant. This is a contradiction. _

Let iI, ... , f m be holomorphic functions that are defined on an open subset
U eX. Let p E U be a point and (V, 'ljJ) a complex coordinate system in X
at p. The mapping f = (iI, ... , fm) : U -+ em is holomorphic, and we define

Jr(Pi'ljJ) :=
•
(8(fi 8z
0 'ljJ-l) ('ljJ(p»)
j
I i.==-
1, ... , m ) .
J - 1, ... , n

This is something like a Jacobian matrix of f at p, but it depends on the

coordinate system 'ljJ. Since
 1. The Complex Structure 161

we have

This shows that

is independent of the chosen coordinate system.

Definition. An analytic set A c X is called regular (of codimension
d) at a point pEA if there are an open neighborhood U = U (p) c X
and holomorphic functions iI, ... , fd on U such that:
1. An U = {q E U : iI(q) = ... = fd(q) = O}.
2. rkp(iI, ... ,jd) = d.
The number n - d is called the dimension of A at p.
If A is regular at every point, A is called a complex submanifold of X.

1.7 Proposition. An analytic set A is regular of codimension d at pEA

if and only if there is a complex coordinate system (U, <p) in X at p with
<p(U) = Been and <p(UnA) = {w E B : Wn-d+l = ... = Wn = O}.
If A is a complex submanifold of X, then A itself is a complex manifold.

PROOF: Let (U, 'ljJ) be an arbitrary coordinate system at p and W := 'IjJ(U).

Then A := 'ljJ(AnU) is an analytic subset of W that is regular of codimension
d at Zo := 'ljJ(p), and there exists a biholomorphic map £ from W onto an open
neighborhood B = B(O) c en with £(zo) = 0 and £(A) = {w : Wn-d+l =
... = Wn = O}. We take <p := £ 0 'ljJ.

If A c X is a submanifold, then A inherits a natural complex structure from

X. This can be demonstrated in the same way as in the case X = en. •

Example
Let F : X ~ Y be a holomorphic map from an n-dimensional manifold into
an m-dimensional manifold. Then

GF:= {(x,y) E X xY : y = F(x)}

is called the graph of F.
162 IV. Complex Manifolds

Let Po E X be a point and qo ::;= F(po) E Y. We choose coordinate systems

(U, cp) in X at Po and (V, t/J) in Y at qo, with F(U) c V. Then (U x V, cp x 1/J) is
a coordinate system in X x Y at (Po, qo) E Gp. Writing t/J 0 F = (II,·.·, 1m)
we get

Gp n (U x V) = {(cp x t/J)-l(Z, w) : Ii 0 cp-1(Z) - Wi = 0 for i = 1, ... , m}.

So Gp is locally defined by the functions 9i(P, q) := Ii(p) - Wi 0 t/J(q), for i =

1, ... ,m. Since rk(pO,qo) (91, ... ,gm) = m, we see that Gp is an n-dimensional
submanifold. ..
The diagonal ~x C X x X is a special case, which is given as the graph of
the identity:
~x = {(x, x') E X xX : x = x'}.

Example
Let A = {(W, Zl, Z2) = ZlZ2}. The projection p : (w, Zt. Z2) t-+
E (:3 : w 2
(zt. Z2) realizes A as a branched covering over (:2 that is the zero set of
the pseudopolynomial w(w; Zl, Z2) = w 2 - ZlZ2. Outside the discriminant set
Dw = {(Zt.Z2) : ZlZ2 = O} it always has two regular leaves over (:2. So A is
everywhere 2-dimensional and regular outside Dw. It is even regular outside
the origin, since V'w( w, Zl, Z2) vanishes only at (0,0,0). One can show that 0
is, in fact, a singular point; e.g., by using Exercise 8.2 in Chapter 1.
The map cp: (:2 -+ A with cp(t1,t2) := (t1t2,t~,t~) is surjective. We call it a
unilonnization. The Jacobian

vanishes exactly at (0,0). The image point (0,0,0) = cp(O,O) is then called a
nonuniformizable point.

Analogously to the situation in en one proves that the set Sing(A) of singular
points of an analytic set A is a nowhere dense analytic subset. The set A is
called irreducible if A - Sing(A) is connected. To every analytic set A C X
there is a uniquely determined locally finite system of irreducible analytic
sets (A'\)'\EA such that A is the union of all these irreducible components A,\.

Differentiable Functions. Let X be an n-dimensional complex man-

ifold, B c X an open set. A function I : B -+ C is called differentiable
(respectively smooth), if for every complex coordinate system (U, cp) with
Un B :f 0 the function 10 cp-1 is differentiable (respectively infinitely dif-
ferentiable) on cp(U n B). We denote the JR.-algebra of real-valued smooth
 1. The Complex Structure 163

functions on B by <C(B) and the (>algebra of complex-valued smooth fuc-

tions by <C(B, q.

1.8 Proposition. In every complex manifold X there is a sequence of com-

pact subsets (Kn) with Kn C (Kn+1t and U:=l Kn = X.

PROOF: Since X is locally compact with countable basis, we can find a

countable basis (Bv)VEN of the topology of X such that each Bv is compact.
We take Kl := B 1 . If nl is the minimal number such that Kl C Bl u·· ·UBn"
then kl ? 2, and we take K2 := Bl U ... U B n" and so on. _
We call (Kn)nEN a compact exhaustion of X.
1.9 Proposition. Let an open covering %' = {UL : ~ E I} of X be given,
and two real numbers r, r' with 0 < r' < r. Then there is a locally finite open
refinement "j/ = {VA : ,\ E L} of %' such that the following hold:
1. For each ,\ E L there exists a complex coordinate system (VA' CPA) in X
with CPA(VA) = Br(O).
2. The open sets cp.;:l(Br/(O)) also cover X.

PROOF: We use a compact exhaustion (Kn) and define Ml := Kl and

Mn := Kn - (Kn-1t for n ? 2. Then (Mn) is a covering of X by compact
sets.
We consider a fixed M = Mn. For each x E M there is an index L = ~(x) E I
and an open neighborhood V = V(x) C ULn«Kn+1)O -Kn - 2). We can make
V so small that there is a complex coordinate system cP : V -+ Br(O) with
cp(x) = 0, and we define V' := cp-l(Br/(O)). The set M is covered by finitely
many neighborhoods V~,l"'" V~,mn like our V'. Then

"j/ := {Vn,i : n E N, i = 1, ... , m n }

is the desired covering.

-
Definition. A (smooth) partition of unity on X is a family (CPL)LEI of
smooth real-valued functions such that:
1. CPL ? 0 everywhere.
2. The system of the sets supp( CPL) is locally finite.
3. LLEl CPL = 1.

1.10 Theorem. For any open covering %' = {UL ~ E I} of X there is a

partition of unity (CPL) with SUPP(CPL) C UL.
164 IV. Complex Manifolds

PROOF: We have a locally finite refinement )/ = {VA : oX E L} of %' and

complex coordinates <PA : VA -+ Br(O) as in Proposition 1.9. If"p : en -+ R is
a smooth function with 0 :::; "p(z) :::; 1, "p(z) == 1 on Brl(O) and "p(z) == 0 on
en - Br(O), we define a smooth function "pA on X by WA = 'l/J 0 <P>. on VA and
'l/JA(X) == 0 otherwise.
Let T : L -+ I be a refinement map (with VA C Ur (>.»). Then 'W = {WL :
LEI} with WL := UAEr-1(L) VA is an open refinement of %' with W. CU•.
In addition it is locally finite, because for x E X there is a neighborhood
P = P(x) such that P n VA oj:. 0 only for oX E L o, Lo c L finite. But then
P n W. oj:. 0 only for L = T(oX), oX E L o.
We define <P. := L>'Er-1(.) 'l/J>.. The sum is finite at every point. So <P. is
smooth and has its support in W.. Every x E X lies in a set <P';: 1 (Brl (0)),
where 'l/JA is positive. Therefore, <P := L. <P. is well defined and everywhere
positive. Now we can define the partition of unity by <P. := <pj <p. •

1.11 Corollary. Let U C X be an open set and V cc U an open subset.

Then there exists a function f E G"(X) with flv = 0 and flex-v) = 1.

PROOF: The system {U, X - V} is an open covering of X. Let {<PI, <P2} be a

partition of unity for this covering. Then supp(<pd C u, SUpp(rp2) eX - V,
and <PI + <P2 = 1. We take f := <P2· •

Tangent Vectors. Let X be an n-dimensional manifold and a E X an

arbitrary point.

Definition. A derivation on X at a is an R-linear map v : G"(X) -+ R

such that

v[J . g] = v[JJ . g(a) + f(a) . vlgJ for f, 9 E G"(X).

If c is constant, then vIc] = 0 for every derivation v.

1.12 Proposition. If f E G"(X) and flv = 0 for some open neighborhood
U = U(a) C X, then v[f] = 0 for every derivation v at a.
PROOF: We choose a function g E G"(X) such that glv = 0 for some open
neighborhood V = V(a) CC U and glex-v) = 1. Then g. f = f, and from
the derivation rule it follows that

v[J] = v[g . fJ = vlg] . f(a) + g(a) . v[J] = O.

•
 1. The Complex Structure 165

1.13 Corollary. If f, 9 are two functions of 6"(X) with flu = glu for some
open neighborhood U = U(a) C X, then v[J] = v[g] for every derivation v at
the point a.
It follows from the corollary that we can restrict derivations to locally defined
functions and work with coordinates. If cp = (Zl, ... , zn) : U ----t en is a
coordinate system at a and z., = Xv + i Yv, then we define partial derivatives
at a by

for i = 1, ... , n. The partial derivatives depend on the chosen coordinate

system, but once we have made our choice, every derivation v at a has a
unique representation l

v tai (a~) tbi(aa.) ,

=
i=l 1 a
+
i=l Y. a

with ai = v[xil and bi = v[Yil for i = 1, ... ,n.

In en the space of derivations is isomorphic in a natural way to the space of
tangent vectors. But what is a tangent vector on a complex manifold X? We
start with a differentiable path a : I ----t X, where I C lR is an interval with
o E I, and a(O) = a. Let (U, cp) be a coordinate system in X at a. Then we
can write cp 0 a = (al, ... , an) and get the tangent vector

( cp 0 a)' (0) = (a~ (0), ... , a~ (0) ).

Unfortunately, this vector depends on the coordinate system. But a tangent
vector at a should somehow be completely determined by a pair (cp, c), where
cp is a coordinate system at a and e = (Cl, ... ,cn ) E en an arbitrary vector.
In this sense the tangent vector to a is given by the pair (cp, (cp 0 a)'(O)). If
we take another coordinate system 'l/J, then

Therefore, we call two pairs (cp, c) and ('l/J, e') equivalent if the Jacobian of
'l/J 0
cp-l at cp(a) transforms e into e', i.e., if

c = c' . J",01/J-l ('lj;(a)) t.

An equivalence class is called a tangent vector at a. The set Ta(X) of all
tangent vectors at a is called the tangent space. It carries the structure of a
complex vector space, which can be defined on representatives:
1 For the proof use the fact that every sm~oth function f on a domain G C
en has near Zo EGa unique representation fez) == L:~=l gv(z)(x.., - xe) +
L~=l h..,(z)(y.., - ye) with g..,(zo) == fx,,(zo) and h..,(zo} = fy,,(zo).
166 IV. Complex Manifolds

(cp,ct)+ (cp,C2) .- (CP,Cl+C2),

A· (cp, c) .- (cp, A· c).

The Complex Structure on the Space of Derivations. If

f = 9 + ih is a complex-valued smooth function on the open set B c X and
v a derivation at a E B, then we define

v[J) := v[g) + i v[h).

1.14 Proposition. For every e E en

and every coordinate system cp at a
there is a unique derivation v at a such that

v[J) := e· \l(J 0 cp-l )(cp(a)) t for every holomorphic function f.

The derivation v depends only on the equivalence class of (cp, c).

PROOF: If a coordinate system cp = (ZI, ... , zn) with Zv = Xv + iyv and a

vector c = a + ib are given, then v can be defined by

v:= 1'tal'
=1
(8=1') + tb (881')
1'=1 a
v
y a
.
If f is a holomorphic function, then fyv = ifxv and fxv = fzv. Consequently,
n
v[f) = L(av + ibv)(f 0 cp-l )xv (cp(a)) = c· \l(J 0 cp-l)(<p(a)) t •
1'=1
The uniqueness follows from the equations v[xv) + i v[Yv) = vlzv) = Cv·
If the pair (cp,c) is equivalent to ('!/J,e'), then

e' . \l(J 0 '!/J-l)('!/J(a)) t = c'· Jcpo,p-l ('!/J( a)) t . \l(J 0 cp-l)( cp(a)) t
= c· \l(J 0 cp-l)(cp(a)) t.

Therefore, v is determined by the equivalence class of (cp, e). •

The assignment (cp,c) ~ v induces a real vector space isomorphism between
the tangent space Ta(X) and the space of derivations at a. It follows that the
tangent space has complex dimension n. The pair (cp, e v ) is mapped onto the
derivation (8j8xv)a, and (cp, ieI') onto (8/{)yv)a.
1.15 Proposition. A complex structure on the space of derivations at a is
given by

J(vHf) = i . vlf), for every holomorphic function f·

 1. The Complex Structure 167

PROOF: Obviously, J is lR-linear, and J 0 J(v) = -v. •

Ifv corresponds to the tangent vector given by (<p,c), then J(v) corresponds
to (<p, ic). One must distinguish carefully between the real number J(v)(J]
and the complex number i . v[f] if f is a real-valued smooth function.

The differential operators ({)/{)Zv)a and ({)/OZv)a are not real-valued deriva-
tions, and therefore they do not correspond to tangent vectors. But they are
nevertheless useful. If v is a derivation, then
n n
v(J] = L av(f 0 <p-l)xv +L bv(f 0 <p-l )Yv
v=l v=l
n
= La v · ((fo<p-lk + (fo<p-lhvl
v=l
n
+ Lbv · i· ((f 0 <p-l)zv - (fo<p-lhJ
v=l
n n
= Lcv(f 0 <p-l)zv + LCv(f 0 <p-lhv'
v=l v=!

if Cv := av + ib v for v = 1, ... ,n.

Therefore, every (real-valued) derivation can be written in the form

v= I:c ({)~v) + I:c (~v) .

v=l
v
a v=l
v
a

The Induced Mapping. Let F : X ~ Y be a holomorphic map be-

tween complex manifolds. Let x E X be an arbitrary point, y := F(x) E Y.

Definition. The tangential map F. = (F.)x : Tx(X) ~ Ty(Y) is

defined by

(F.v)[g] := v(g 0 F], for derivations v and functions 9 E C(Y).

The map F. is linear, acting on tangent vectors as follows:

if cp is a coordinate system at x, and 1/J is a coordinate system at y.

Now we have an assignment between the category of complex manifolds (with
a distinguished point) and the category of vector spaces. To any manifold
168 IV. Complex Manifolds

X and any point a E X there is associated the tangent space Ta(X). To

any holomorphic map F : X -t Y with F(a) = b there is associated the
homomorphism F. : Ta(X) -t n(Y). This assignment has the following
properties:

(idx ). idTx(x) ,
(G 0 F). G. 0 F. (if G : Y -t Z is another holomorphic map).

Such an assignment is called a covariant functor. If it interchanged the order

of the maps, it would be called a contravariant functor.
Remark. For historical reasons the elements of the tangent space are called
contravariant vectors and the elements of its dual space covariant vectors.
But the tangent functor behaves covariantly on the tangent vectors and con-
travariantly on the covariant tangent vectors in Ta(X)'. One should keep this
in mind.

Immersions and Submersions. We are particularly interested in

the case where the (local) Jacobian of a holomorphic map F : X -t Y has
maximal rank. If n = dim(X) and m = dim(Y), then the rank is bounded
by min(n, m). Only two cases are possible:
Definition. The holomorphic map F is called an immersion at x if
rk(F.) = n :'S m, and F is called a submersion at x if rk(F.) = m 2: n.
In the first case (F.)x is injective; in the second case it is surjective.

We call F an immersion (respectively submersion) if it is an immersion (re-

spectively submersion) at every point x E X.
Remark. If F : X -t Y is an injective immersion, then for every x E X
there are neighborhoods U(x) c X and V(F(x)) C Y such that F(U) is a
submanifold of V. In addition, if X is compact, then F(X) is a submanifold
of Y. We omit the proof here.

1.16 Theorem. Let Xo E X be a point, Yo := F(xo). The following condi-

tions are equivalent:
1. F is a submersion at Xo.
2. There are neighborhoods U = U(xo) C X and V = V(yo) C Y with
F(U) c V, a manifold Z, and a holomorphic map G : U -t Z such that
x H (F(x), G(x)) defines a biholomorphic map from U to an open subset
of V x Z.
3. There is an open neigborhood V = V(Yo) C Y and a holomorphic map
s : V -t X with s(Yo) = Xo and F 0 s = idv. (Then s is called a local
section for F.)
 1. The Complex Structure 169

PROOF: (1) ==> (2) : We can restrict ourselves to a local situation and
assume that U = U(O) c en
and V = V(O) c em are open neighborhoods,
and F: U -+ V a holomorphic map with F(O) = 0 and rk(JF(O» = m.
We write JF(O) = (J;..(O), J;J.(O», with J;"(O) E Mm.m(C) and J;J.(O) E
Mm.n-m(C). Choosing suitable coordinates we may assume that det J;"(O) f=
o. We define a new holomorphic map F : U -+ V X en-
m c by en
F(z',z"):= (F(z',z"),z"), for z' E em, z" E en - m .
Then

JF-(O) _- (J;..(O) J;J.(O»), and t h erelore

1: d et JF
(0) r 0
J.. .
o En - m

By the inverse function theorem there are neighborhoods U(O) C U and

W(O) c ensuch that F: if -+ W is biholomorphic.
We observe that Z := e n - m is a complex manifold, and G := pr2 : U -+ Z
with (z', z") I-t Z" is a holomorphic map such that (F, G) = F is biholomor-
phic near o.
(2) => (3) : If U, V, Z, and G are given such that F(U) C V and (F, G) :
U -+ W c V x Z is biholomorphic, then s : V -+ X can be defined by

s(y):= (F,G)-I(y,G(xo».

Then (F, G)(s(yo» = (Yo, G(xo» = (F, G)(xo), and therefore s(yo) = Xo·
Furthermore, (F, G) 0 s(y) = (F,g) 0 (F, G)-l(y, G(xo» = (y, G(xo». Thus
Fos(y)=y.
(3) => (1) : If s is a local section for F, with s(Yo) = xo, then F. os. (v) = v
for every v E TyO (Y). Thus it follows immediately that F .. is surjective. _

1.17 Corollary. If F : X -+ Y is a submersion, then for each y E Y the

fiber F- 1 (y) is empty or an (n - m)-dimensional submanifold of X. In the
latter case T x (F- 1 (y» = Ker«F.)x) for all x E F- 1 (y).
PROOF: We consider a point Xo E X. Let M := F-l(yo) be the fiber over
Yo := F(xo)· Then we can find neighborhoods U = U(xo) C X, V = V(yo) C
Y, a manifold Z, and a holomorphic map G : U -+ Z such that (F, G) : U -+
W c V x Z is biholomorphic. It follows that M n U = (F, G) -1 ( {yo} x Z) n U
is a manifold of dimension n - m.
Since FIM is constant, we have F.ITxo(M) == O. This means that Txo(M) C
Ker(F.). Since these spaces have the same dimension, they must be equal. _
170 IV. Complex Manifolds

Gluing. Assume that X is a set that is the union of a countable collection

(X" )"EN of subsets such that the following hold:
1. For every v E N there is a bijection 'P" : X" ~ M", where M" is an
n-dimensional complex manifold.
2. For every pair (v,J.£) E N x N the subset cp"(X,, n XJL) is open in M", and
the map
'P" 0 'P;.l : 'PJL(X" n XI') ~ 'P"(X,, n XI')
is biholomorphic.
3. For every pair of points a E X" and b E XI' with a f= b there are
open neighborhoods U(cp,,(a» c M" and V('PI'(b» c MI' with 'P;;l(U) n
cp;l(V) = 0.

1.18 Proposition. Under the above conditions there is a unique complex

structure on X such that the X" are open in X and the cp" : X" ~ M" are
biholomorphic.
PROOF: We give only a sketch of the proof and leave the details as an
exercise for the reader.
A subset U C X is called open if 'P,,(UnX,,) is open in M" for every v. Then
the collection of open sets has the properties of a topology on X. In addition,
for every open set W C Mil the set 'P;;l(W) is open in X. Consequently,
the maps CPII : XII ~ Mil are homeomorphisms. From the last hypothesis it
follows that the topology on X is Hausdorff, and since the collection of the
X II is countable, it has a countable basis.
If U c M" is open and t/J : U ~ en a coordinate system, then -if; := t/J 0 'PII :
'P;; 1 (U) ~ en is a coordinate system for X. One checks easily that two
such coordinate systems are biholomorphically compatible. So we obtain a
complex structure on X. •
One says that X is obtained by gluing the manifolds M". Another way
to describe this process is the following. Let there be given a collection of
complex manifolds M", open subsets Mill' eM", and biholomorphic maps
'PilI' : MI''' ~ Mill' (including 'Pw = idMJ. Consider pairs (x, v) with
x E Mil' Then (x, v) is called equivalent to (Y,J.£) if

x E M"I" Y E MI''' and x = 'P"JL(Y)'

The set X of equivalence classes is the result of the gluing process. Of course,
one has to add a condition that ensures the Hausdorff property.

Exercises
1. Let M be a compact connected complex manifold with dim(M) ~ 2 and
N C M a closed submanifold of codimension greater or equal to 2. Show
that every holomorphic function f : M - N ~ e is constant.
 2. Complex Fiber Bundles 171

2. Let X be a Riemann surface.

(a) If f is a meromorphic function on X, and P the set of poles, show
that
J(x) := {f(X) for x E X - P,
00 for x E P
defines a holomorphic map 1: X -t C.
(b) Prove that every holomorphic map f : X -t C that is not identically
00 defines a meromorphic function on X.
3. Let f : X -t Y be a nonconstant holomorphic map between Riemann
surfaces, Xo E X, and Yo := f(xo) E Y. Prove that there is a k 2: 1 such
that there are complex coordinates cp : U(xo) -t C and 1jJ : V(yo) -t C
with:
(a) cp(xo) = 0, 1jJ(yo) = O.
(b) feU) c v.
(c) 1jJ 0 f 0 cp-1(Z) = zk.
4. The general linear group GLn(C) is an open subset of the vector space
Mn(C). Prove that the special linear group SLn(C) := {A E Mn(C) :
det(A) = I} is a submanifold of GLn(C). Calculate the tangent space
TE(SLn(C)) C TE(GLn(C)) = Mn(C), where E = En is the identity
matrix.
5. Let f : X -t Y be a holomorphic map and Z C Y a closed submanifold.
Show that if
Im((f.)x) + Tf(x)(Z) = Tf(x)(Y)
for every x E f-1(Z), then f- 1 (Z) is a submanifold of X.
6. The holomorphic maps f : X -t Z and 9 : Y -t Z are called transversal
iffor every (x, y) E X x Y with f(x) = g(y) =: z the following holds:

Prove that the fiber product

X Xz Y:= ((x,y) E X x Y : f(x) = fey)}

is a complex submanifold of X x Y.

2. Complex Fiber Bundles

Lie Groups and Transformation Groups. Assume that G is a
set that has the structure of a group and at the same time that of an n-
dimensional complex manifold. The inverse of 9 E G will be denoted by g-1,
the identity element bye, and the composition of two elements gl, g2 E G by
g1g2·
Definition. We call G a complex Lie group if the following two prop-
erties hold:
172 IV. Complex Manifolds

1. The mapping 9 rl g-l (from G to G) is holomorphic.

2. The mapping (gl,g2) ~ g1g2 (from G x G to G) is holomorphic.

There are many examples of complex Lie groups. The simplest one is the
space C n , where the composition is vector addition. Another example is the
group C* with respect to ordinary multiplication of complex numbers.
The most important example is the general linear group

GLn(C):= {A E Mn(C) : detA f:. OJ.

2
Its complex structure is obtained by considering it as an open subset of C n .
The multiplication of matrices is bilinear, and the determinants appearing in
the calculation of the inverse of a matrix A are polynomials in the coefficients
of A.

Every matrix A E GLn(C) defines a linear and therefore holomorphic map

1> A : cn -+ C n by
1>A(Z) := z· At.
Then 1>AB(Z) = z· (AB)t = z· (BtAt) = (z .Bt). At = 1>A(<I>B(Z)). If En
is the identity matrix, then 1> E n = id. Furthermore, if A is any matrix with
1> A = id, then A must be the identity matrix, since 1> A (ei) = ei . A t is the
transpose of the ith column of A.
We want to generalize this situation. Let X be a complex manifold and G a
complex Lie group.

Definition. We say that G acts analytically on X (or is a complex

Lie transformation group on X) if there is a holomorphic mapping 1> :
G x X -+ X with

The holomorphic map x r l 1>(g, x) is denoted by 1>g. We say that G acts

effectively or faithfully on X if 1>g = id x implies that 9 = e.
Often we write gx instead of 1>(g, x) or 1>g(x). A point x E X with gx = x
is called a fixed point of g. We say that G acts freely if only the identity
element e E G has fixed points in X. The general linear group GLn(C) acts
analytically and faithfully on cn,
but not freely.
Let {WI, ... , W2n} be any basis of C n over lR. Then

is a subgroup of the (additive) group C n generated by WI, ... , W2n. The group
r acts on C n by 1>(w, z) := z + w. This is an example of a free action.
 2. Complex Fiber Bundles 173

Fiber Bundles. Let X and F be complex manifolds and G a complex

Lie group acting analytically and faithfully on F.

Definition. A topological (respectively holomorphic) fiber bundle over

X with structure group G and typical fiber F is given by a topological
space (respectively a complex manifold) P and a continuous (respectively
holomorphic) map 7r : P -+ X, together with
1. an open covering all = {U. : tEl} of X,
2. for any tEl a topological (respectively biholomorphic) map

<P. : 7r- 1 (U.) -+ U. x F

with pr1 0 <P. = 7r,

3. for any pair of indices (t, K.) E I x I a continuous (respectively holo-
morphic) map g.K. : U. n UK. -+ G with

for x E U.K. := U. n UK. and p E F.

The maps <P. are called local trivializations and the maps g.K. a system of
transition functions.

Since G is acting faithfully, we get the following compatibility condition:

Then gu =e and gK. = g;:;/.

Now let a system of transition functions (g.K) be given such that the com-
patibility condition is satisfied. Using the gluing techniques mentioned at the
end of Section 1, a suitable bundle space P, a projection 7r : P -+ X, and
local trivializations can be constructed as follows:
Identifying (x,p) and (x,g.K.(x)p), we can glue together the Cartesian prod-
ucts UK. x F and U. x F over U'K.' Due to the compatibility condition this
works in a unique way over U'''A' The obvious projection from P to X is
continuous. Therefore, P is a fiber bundle over X with structure group G
and typical fiber F. If the transition functions gL" are holomorphic, then P
carries the structure of a complex manifold, and the projection and the local
trivializations are holomorphic. So P becomes an analytic fiber bundle in
that case.
Example
Let X be an n-dimensional complex manifold. There is an open covering
all = {U. : t E I}, together with complex coordinate systems
174 IV. Complex Manifolds

We take C as typical fiber and C' as structure group, acting on C by multi-

plication. Then we can define transition functions g.,. : Ut ,. -+ C* by

The compatibility conditions are satisfied because of the chain rule and the
determinant product theorem. So by the gluing procedure described above
we get a holomorphic fiber bundle over X that is called the canonical bundle
and is denoted by K x .

Equivalence. Let 1fp : P -+ X and 1fQ : Q -+ X be two topological or

holomorphic fiber bundles over the same manifold X, with the same fiber F
and the same structure group G. We assume that there is an open covering
'fI = {UL : ~ E I} of X such that there are trivializations ipL : 1fpl(UL) -+
UL X F and 1/J. : 1fQl(U.) -+ UL X F.
Definition. A fiber bundle isomorphism between P and Q is a topo-
logical (respectively biholomorphic) map h : P -+ Q with 1fQ 0 h = 1rp
such that for any ~ E I there is a continuous (respectively holomorphic)
map hL : U. -+ G with

The bundles are called equivalent in this case.

We give a description of bundle equivalence in the context of transition func-

tions. Let (g~,,J be the system of transition functions for P with respect to
'fI, and (g~~) the corresponding system for Q. Then we have

(x,h.(X)g~K(X)P) = 1/JLohoip;l(x,g~K(x)p)
= 'l/JLohOip-;;,l(X,p)
1/J. 0 1/J-;;,l(X, h,.(x)p)
(x, g~~ (x )hK (x)p).

Since G is acting faithfully, it follows that

Two systems of transition functions (g:,.) and (g~~) with respect to the
same covering are called topologically (respectively analytically) equivalent
or cohomologous if there are continuous (respectively holomorphic) maps
hL : U. -+ G satisfying condition (C).
Equivalent fiber bundles have equivalent systems of transition functions. On
the other hand, it is easy to see that fiber bundles constructed from equiv-
alent systems of transition functions are themselves equivalent. Now we will
demonstrate that the latter remains valid in passing to finer coverings.
 2. Complex Fiber Bundles 175

Let us assume that there are given systems of transition functions g' = (g:l<)
and gil = (g:~) with respect to a covering 'W = {U, : £ E I}. We call them
equivalent as well if there is a refinement "f/ = {Vv : v E N} of 'W (with
refinement map T : N -t J) and a collection of maps hv : Vv -t G such that
h ",g7(V)7(/1-)
I
= /I
g7(V)7(/1-)
h f.L

To show that the systems are equivalent in the old sense, we define h, : U, -t
Gby
on U, nV/l-.

In fact, h, is well defined, since on U, n VV/1- we have

h
v =
/I ( ,
g7(V)7(/1-)h/1- g7(V)7(/1-)
)-1

and therefore g:~(V)hv(g:7(V))-1 = g:~(/1-)h/l-(g:7(/1-))-1. Then on U'I< n V/l- we

get

The bundles are equivalent!

If two bundles are given with respect to two different coverings, then they are
called equivalent if they are equivalent with respect to a common refinement,
for example, the intersection of the coverings. Then everything works as above.

Complex Vector Bundles. Let X be an n-dimensional complex

manifold.
Definition. A complex topological (respectively holomorphic) vector
bundle of rank r over X is a topological (respectively holomorphic) fiber
bundle V over X with C r as typical fiber and GLr(C) as structure group.
In the case r = 1 we are speaking of a complex line bundle.

If 11" : V -t X is the bundle projection, then we denote by V", the fiber 7J"-l(X).
It has the structure of an r-dimensional complex vector space. A trivialization
cP : 7J"-l(U) -t U X C r is also called a vector bundle chart. For any x E U the
induced map <I>", : V", -t C r is a vector space isomorphism.
Definition. Let V be a holomorphic vector bundle over X. If U c X
is an open subset, then a continuous (differentiable, holomorphic) cross
176 IV. Complex Manifolds

section (or simply section) in V over U is a continuous (differentiable,

holomorphic) map s : U -t V with 7r 0 s = idu.
We denote by r(U, V) (or O(U, V)) the vector space of holomorphic cross
sections in V over U, by tC(U, V) the space of differentiable sections, and
by «fO(U, V) the space of continuous sections.
The vector bundle V is called 910bally generated if the canonical map
r(X, V) -t Vx with S H sex) is surjective for every x E X.

Let (U" <1>.).Ei be a collection of vector bundle charts <p. : 7r- 1(U.) -t U. XC r
for V, and 9",< : U."" ~ GLr(C) the system of transition functions, given by
<1>.0 <1>;;:1 (X, z) = (x, z· 9."" (x) t) for (x, z) E U. K X cr.
If s is a holomorphic section in V, then

<p. 0 slu, (x) = (x, s.(x))

defines a system of holomorphic maps s. : U. -t C r , and we obtain the
compatibility condition
s.(x) = SK(X) . 9.K(X) t on U.K.
On the other hand, any such system (sJ defines a global section s.
Example
We can define vector bundles by giving a system of transition functions. The
construction of the bundle space is carried out with the same gluing technique
as for general fiber bundles.
If X is an arbitrary n-dimensional complex manifold, and (U., 'P.).El a com-
plex atlas for X, then

9.",,(X) := J<p,o<p;l('PK(X)) E GLn(C)

defines a system of transition functions with respect to all = {U., ~ E I}.
The corresponding vector bundle T(X) is called the tangent bundle of X.
It results from gluing (x,c) E U"" x Cn with (x,c· 9.",,(X)t) E U. x cn, for
x E U."". Therefore, we can identify the fiber (T(X))x with the tangent space
Tx(X). The local trivializations <1>. : T(X)lu, -t U. XCn are given as follows:
For a E U. the trivialization <1>. maps a tangent vector v E Ta(X), represented
by ('P., c), onto the pair (a,c) E U. x cn. If we denote the equivalence class
of ('P" c) at a by ['P., c], we obtain
<p. 0 <p;;:l(a, c) = <p. (['P"", cD = <p.(['P., c· 9."" (a) t]) = (a, c· 9."" (a) t).
A holomorphic section in T(X) is also called a holomorphic vector field.
 2. Complex Fiber Bundles 177

Definition. Let V, W be two holomorphic vector bundles over X. A

vector bundle homomorphism between V and W is a fiber preserving
holomorphic map 'l/ : V -+ W such that for any x E X a linear map
'f}x : Vx -+ Wx is induced.
The map 'l/ is called a vector bundle isomorphism if 'f} is bijective and
'f}, 'f}-l both are vector bundle homomorphisms.

A holomorphic vector bundle V of rank r over X is called trivial if it is

isomorphic to the bundle X x cr. This is equivalent to the existence of a
frame {6, ... ,~r} of holomorphic sections ~i E f{X, V) such that for every
x E X the elements 6 (x), ... , ~r (x) E Vx are linearly independent. In this
case V is globally generated. But there are also nontrivial bundles that are
globally generated.
2.1 Proposition. A holomorphic map 'f} : V -+ W is a vector bun-
dle homomorphism if and only if for each pair of vector bundle charts
<I> : Vlu -+ U x ((7 and lit : Wlu -+ U x CS there is a holomorphic map
h : U -+ Ms,r{C) with

lit-I 0 'f/ 0 <I> {x, z) = (x, z· h{x) t).

We omit the elementary proof.

Standard Constructions. Let X be an n-dimensional complex man-

ifold. We can think of a vector bundle over X as a parametrized family of
vector spaces. Therefore, numerous constructions from linear algebra carry
over to the theory of vector bundles.
1. The direct sum: If V, Ware two vector bundles over X, then the direct
sum, or Whitney sum, V EEl W := V XX W carries a vector bundle structure
that is defined as follows:
Let %' = {U. : LEI} be an open covering of X such that there are vector
bundle charts <I>. : Vlu, -+ u. x C r and lit. : Wlu, -+ U. x C k • Then a vector
bundle chart

{V EEl W)lu, = ({v, w) E V x W : 1l"v{v) = 1l"w{w) E UJ -+ U. x cr+ k

can be defined by

If g.K, respectively h. K, are transition functions for V, respectively W, then

the matrices
178 IV. Complex Manifolds

are transition functions for V EEl W. The fiber of the Whitney sum is given by
(V EEl W)., = V., x W.,.
2. The dual bundle: Let 7r : V ~ X be a holomorphic vector bundle of
rank r with trivializations cI>. : 7r- 1 (U.) ~ U. X C r and transition functions
g.... Gluing the sets U. x C r together by means of the transition functions
g:.. := g!. leads to the dual bundle 7r' : V' ~ X. Denote the associated
trivializations over U. by \II.,
We will show that for every x E X there is a natural isomorphism

Given elements x E U., v E Vx , and .x E (V')x, we define

using the vector space isomorphisms (cI>.)., : Vx ~ C r and (\II.)., : (V')., ~

Cr ,
For x E U. n UK. we obtain

(cI>K.).,(v) , (\II,.)., (.x) t

= (cI>K.)x 0 (cI>.);1 0 (cI>.).,(v) , [(\IIK.)" 0 (\11.);1 0 (\11.).,(.x)1 t
(cI>.)x(v) 'gK..(x) t , [(\II.)xC.x) . g.K.(x») t
= (cI>.).,(v)· g,..(x)t. g.K.(x)t. (\II.)x(.x)
= (cI>.).,(v) , (1lI.)x(.x).

This shows that the definition of .x(v) is independent of the trivializations.

3. Tensor powers of a line bundle: Let 7r : F ~ X be a line bundle with

transition functions g... : U.K. ~ C* ,
Definition. For kEN, the tensor power Fk is the line bundle defined
by the transition functions g~,.,

We give an interpretation of Fk using the dual bundle 7r' : F' ~ X, As-

sume that there is an open subset U C X and a holomorphic function
f : (7r,)-1(U) ~ c. If 1/J. : (F')lu. ~ U. x C are trivializations (with
1/J. 0 1/J;1 (x, z) = (x, z ' g.K.(x)-1», then we have a power series expansion

L av,.(x)zV
00

f 0 1/J;-1(X, z) =
1.1=0

on 1/J.(F'lu.nu) with holomorphic functions av,. on Un U•. Over U. n U,. n U

the following holds:
 2. Complex Fiber Bundles 179

00

jo'l/J;l(X,Z) = jo'l/J;-l(X,Z'gLlt(X)-l) = L(av,,(x)g,.. (x)-V)zV,

v=o
and therefore av" = av,.. . g~... This means that av = (a v,,) is a cross section
in Fk over U. So every holomorphic function j on F' that is homogeneous of
degree k on the fibers is a section in Fk. In particular, F: can be identified
with the space Lk(F~, C) of k-linear functions j : F~ x ... x F~ --t C-
For et, ... , ek E Fx the tensor product el ® '" ® ek E (Fk)x is defined by

The tensor power e ® ... ® e of an element e E Fx is denoted by e k for short.

Finally, we define F- k := (F')k ~ (F k )'.

4. The tensor product: Let p : V --t X be a vector bundle of rank r with

transition functions G, .. : U, .. --t GLr(C), and 11" : F --t X a line bundle with
transition functions g, .. (with respect to the covering %' = {U, : tEl}).
Definition. The tensor product V ® F is the vector bundle of rank r
given by the transition functions

Let <1>, : Vlu, --t U, X cr and 'l/J, : F'lu, --t U, x C be local trivializations. If
j: F'lu --t Vlu is a holomorphic map which is linear on the fibers, then over
U, n U we have
<1>,0 j 0 'l/J;-l(X, z) = (x, z· "1,(x»,
where "1, : U, n U --t cr is a holomorphic map. Over U, n u.. n U we calculate

and on the other hand,

<1>,0 j 0 'l/J;l(x, z) = <1>,0 <1>;1 (x, "1.. (x) . z) = (x, 1].. (x) . G, .. (x) t . z).
It follows that

Consequently, "1 = ("',) is a cross section in V ® F, and we obtain for every x

an isomorphism
(V 0 F)x ~ HomdF~, Vx ).
For v E Vx and e E Fx the tensor product v ® e E (V ® F)x is defined by

(v 0 e)(>.) := >.(e) . v.
180 IV. Complex Manifolds

Lifting of Bundles. Let f : X --* Y be a holomorphic map between

complex manifolds and p : V --* Y a vector bundle of rank r.
Definition. The lifted bundle, or pullback, j*V over X is defined by

rV:= X Xy V:= ((x, v) E X x V : f(x) = p(v)}.

The bundle projection p: j*V --* X is given by jJ(x, v) := x.

The fiber of j*V over x E X is given by (J*V)x = Vf(x). Therefore, the lifted
bundle is trivial over the preimage sets f-l(y).
One has the following commutative diagram:

j*V ~ V
p 4. 4. p
X -4 Y

If %' = {U, : ~ E I} is an open covering of Y such that V is trivial over U"

then ~:= {U, = f-l(U,) : ~ E I} is an open covering of X such that j*V
is trivial over U,.
If iP, : Vlu, --* u, x C r is a trivialization for V, then we can define a trivial-
ization ~, : rVlu, --* U, x C r by

If G u :. are transition functions for V with iP, 0 iP;l(X, w) = (x, w· Gu.:(x) t),
then

i.e., rV is given by the transition functions Gu .: 0 f.

If ~ E r(u, V) is a holomorphic section over some open subset U c Y, then
r
~ can be lifted to a section [ E r( fj, V) given by

[(x) := (x, ~(J(x))).

Subbundles and Quotients.

Definition. Let 7r : V --* X be a vector bundle of rank r. A subset
We V is called a subbundle (of rank p) of V if there is a p-dimensional
linear subspace E c C r , and for any x E X an open neighborhood U =
U(x) and a trivialization iP : Vlu --* Uxcr such that iP-l(Ux E) = Wlu.

One sees immediately that:

 2. Complex Fiber Bundles 181

1. Wx = W n Vx is always a p-dimensional linear subspace of Vx '

2. W is a submanifold of V.
3. W is itself a vector bundle.

Example
Let V be a vector bundle of rank r on X and Y C X a submanifold. If we
denote the natural injection Y Y X by j, then Vly := j*V is a vector bundle
of rank r on Y. We apply this to the tangent bundle T(X).
Choose an open covering %' = (U~)~EI such that there are complex coordinate
systems 'P~ = (zt, ... , z~) for X in U~ with the following properties:
1. U, n Y = {zd+! = ... = z~ = O}.
2. zt, ... , zd are complex coordinates for Y.
A trivialization <P, : j*T(X)lu,nY -t (U, n Y) X en is given by
(y, ['P~, cD H (y, c).

A tangent vector v belongs to Ty(Y) C Ty(X) if and only if there is a differ-

entiable path a: I -t Y with a(O) = x and ('P~oa)'(O) = c. This is equivalent
to the statement that c = (Cl,"" Cd, 0, ... ,0). Therefore

cp;l(U, X {c E en : Cd+! = ... = Cn = O}) = T(Y)lu"

and T(Y) is a subbundle of j*T(X).

If G'I< := (()z~ /8z~ I v, Il = 1, ... , n) are the transition functions for T( X),
then

where g'l< = (8z~/8z~)IY lv,

Il = 1, ... , d), are the transition functions for
T(Y), and g:I«Z', 0) = «8zU8z~)(z',O) Iv,1l = d+ 1, ... ,n).
Let V be a vector bundle of rank r. If W C V is a subbundle, then we define
the quotient bundle V/W by (V/W)x := Vx/Wx . We have to show that
there are trivializations for V /W. If E C e r is a subspace such that there
are trivializations cp : Vlu -t U X e r with Wlu = <p- 1 (U x E), then we can
choose a subspace F C e r with E tB F = e r and define ~ : (V/W) Iu -t U x F
by
~(v mod W) := prp(cp(v»,

wherepr p : E tB F -t F is the canonical projection.

Using trivializations as above one obtains transition functions
182 IV. Complex Manifolds

for V such that 9£1< are transition functions for W, and h,,,. are transition
functions for V IW.
Example
The quotient bundle Nx(Y) := j*T(X)IT(Y) is called the normal bundle of
YinX.

Exercises
1. Let G be a complex Lie group that acts analytically on a complex man-
ifold X. Then for every x E X the stabilizer G x := {9 E G : 9X = x}
is a closed Lie subgroup of G, i.e., a subgroup and a (closed) complex
submanifold.
Prove that there is a unique complex structure on G IGx such that the
canonical projection 7r : G -t GIG x is a holomorphic submersion.
2. Let 7r : V -t X be a complex vector bundle of rank r. For x E X let fAx
be the set of bases of Vx ' Prove that the disjoint union of the fAx, x EX,
carries the structure of a fiber bundle over X with structure group and
typical fiber equal to GLr(C).
3. Let cp : V -t W be a vector bundle homomorphism over X. Suppose that
rk(cpx) is independent of x E X. Prove that

Ker(cp) := U Ker(cpx) and Im(cp):= U Im(<px)

xEX xEX

are subbundles of V, respectively W. Show that Im(cp) ~ VI Ker(cp).

4. Let X = C be the Riemann sphere. Determine the transition functions
for T(X) for the canonical bundle Kx and for the normal bundle of {oo}
in X.
5. Let f : X -t Y be a holomorphic map. Prove that there is a uniquely
determined vector bundle homomorphism l' : T(X) -t J*T(Y} with
(f'}x = (f*)x for x E X.
6. Let p : V -t X be a holomorphic vector bundle. Show that there are
vector bundle homomorphisms h : p*V -t T(V) and k : T(V) -t p*T(X)
over V with Im(h) = Ker(k).

3. Cohomology
Cohomology Groups. Let X be an n-dimensional complex manifold
and 7r : V -t X a complex vector bundle of rank r. Assume that there is an
open covering 'F/ = {U, : LEI} of X.
We consider sections in V on the sets U, and on intersections U'I< := U, n U".,
respectively U,,,.A := U, n U". n UA.
 3. Cohomology 183

Definition. A O-dimensional cochain with values in V (with respect

to %') is a function s that assigns to every LEI a section s. E r(u., V).
The set of O-dimensional cochains is denoted by CO (%' , V).
A I-dimensional cochain (with values in V) is a function ~ that assigns to
every pair (L, 1\:) E IxI a section ~LK E nULK , V). The set of I-dimensional
cochains is denoted by C 1 (%', V).
Finally, a 2-dimensional cochain (with values in V) is a function A that
assigns to every triple (L, 1\:, v) E I x I x I a section A LKv E nULKV , V),
and the set of all these 2-dimensional cochains is denoted by C 2 (%" V).

Assume that a O-dimensional cochain S is given. One may ask whether the
sections SL E nULl V) can be glued together to a global section s E rex, V).
For that it is necessary and sufficient that SL = SK on U'K' This can be
expressed in another way: If we assign to every O-cochain s a l-cochain 8s by
(8S)LK := SK - s,' then S defines a global section if and only if 88 = O.
Definition. The coboundary operators

are defined by

.- SK - S, (on ULK ),
.- ~KV - ~'V + ~LK (on U'KV ).
A cochain s E CO (%', V) (respectively ~ E C 1 (%', V) is called a cocycle
if 8s = 0 (respectively 8f, = 0). The sets of cocycles are denoted by
ZO(%" V) (respectively Zl(%" V».

Remarks
1. The sets of cochains and the sets of cocycles are all complex vector spaces.
2. We can identify ZO(%" V) with rex, V).
3. A l-cochain f, is a co cycle if and only if the following compatibility con-
dition holds:
f,w = ~LK + f,KV on ULKV '
4. Sometimes we need co cycles of degree 2. We call an element A E
C2 (%', V) a cocycle if the following compatibility condition holds:

The set of all these co cycles is denoted by Z2 (%', V).

184 IV. Complex Manifolds

A coboundary is an element of the image of the co boundary operator. The sets

of coboundaries are again vector spaces, denoted by Bl(%" V) = 6CO(%" V)
and B2(%" V):= 6C1 (%" V). For completeness we define BO(%" V):= o.
3.1 Proposition. Bi(%" V) c Zi(%" V) for i = 0,1,2.

PROOF: The case i = 0 is trivial. For ~ = ~s E BI (%', V) we have

For A = ~TJ E B2(%" V) we have

(TJvj.£ - TJ,,. + TJw)

- (TJltl-' - TJ,I-' + TJut)
+ (TJltV - TJ.v + TJ,It)

•
Definition. Hi(%" V) := Zi(%" V)jBi(%" V) is called the ith coho-
mology group of V with respect to %'.

We have HO(%" V) = reX, V), independently of the covering, and we have

Hl(%" V) = {e E Cl(~, V) : ~,~ = e.K. +eK.v} .

{~ : 3s E C (%', V) wIth e,K. SK. - s.} =
The canonical map from ZI(%" V) to HI(%" V) will be denoted by q.
We do not want to elaborate on H2, because we need it only in very special
cases.

Refinements. Let "f/ = {Vn : n E N} be a refinement of %'. Then

there is a refinement map r : N ~ I with Vn C Ur(n). It induces maps
ro : CO (%', V) ~ CO ("f/, V) and rl : C l (%', V) ~ C l ("f/, V) by

(roS)n := (sr(n»)!Vn and (rle)nm := (er(n)r(m»)lvnm •

Then 6(ros) = rl(~s), and if 6~ = 0, then also 6(rl~) = O. Therefore, r

induces a map r* : Hl(%" V) ~ HI("f/, V) by

r*(q(e)) := q(rle)·

By the remarks above it is clear that r* is well defined, and it is a vector

space homomorphism.
 3. Cohomology 185

3.2 Proposition. The map T* is independent of the refinement map T and

is injective.

PROOF: Let a : N -t [ be another refinement map and ~ a cocycle with

respect to %'. We have to show that Tl~ - ale is a coboundary:

(Tl( - al~)nm ~T(n)r(m)IVm" - el7(n)<7(m) Iv" ...

= (~T(n).,.(n) + ~.,.(n)r(m») - (e.,.(n)T(m) + er(1Jl)I7(m»)
~T(n).,.(n)IV"m - ~T(m)<7(m)IV"m .

We define 'r/ E coer, V) by'

'r/n := ~T(n)<7(n) Iv".

Then Tl~ - (Tle = 6'r/.

e
Now we consider a cocycle with respect to %' such that Tle = 68 for some
8 e
E coer, V). We want to see that itself is a coboundary.

On Vnm n U. we have

~T(n)T(m) = eT(n). + e.T(m) = ~.r(m) - e.r(n)'

Since eT(n)T(m) Iv"", = (81Jl - 8n )lv" ... , it follows that

e.T(m) - 8m = e.T(n) - Sn on Vnm n U •.

Therefore, we can define h. on U. by hLlu,nv" := ~.T(n) - Sn. This gives an
element hE CO(%', V), and on U. K n Vn we have

h. - hK = (~.T(n) - 8n ) - (eKT(n) - sn) = ~.T(n) + ~T(n)K = ~.K'

This means that e= 6( -h). SO T* is injective. •
We have seen that if "f/ is a refinement of %' , then Hl (%' , V) can be identified
with a subspace of Hl("f/, V). Therefore, we form the union of the spaces
Hl(%" V) over all coverings %' and denote this union by Hl(X, V). We call
it the absolute, or tech, cohomology group of X with values in V.

Acyclic Coverings. Let X be a complex manifold and p : V -t X a

vector bundle over X.
The covering %' is called acyclic, or a Leray covering for V, if Hl(U., V) = 0
for every t E [.
If "f/ = {Vn : n EN} is another covering, then U. n "f/ := {U. n Vn : n E N}
is a covering of U., and since Hl(U. n "f/, V) is a subspace of Hl(U., V), we
have also Hl(U. n "f/, V) = O.
186 lV. Complex Manifolds

3.3 Theorem. If %' = {Ut : LEI} is acyclic and "I' = {Vn : n E N}

is a refinement, with refinement map r : N --t I, then r* : HI (%', V) --t
HI ("I', V) is bijective.

PROOF: We start with an TJ E ZI("I', V) and define TJ(t) E ZI(UL n"l', V) by

TJ!:),. := TJnm IU.nVnm , for all n, m with U. n Vnm #- 0.

Since %' is acyclic, there is an element g(.) E CO(Ut n "1', V) with 1](L) = 8g<t).
Then TJnm + g!:) = g~) on U. n Vnm ' and therefore g~) - g~) = g!:) - g~~) on
u.",nVnm . .
Now we define ~ E Zl(%" V) by ~.~Iu.",nvn := g!:) - g~"'), and hE CO("I', V)
by hn := g~T(n» (on Vn = Vn n UT(n»'
Then on Vnm we have

(rI~ -1])nm = ~T(n)T(m) -1]nm

= g~(n)) _ g~(m)) _ (g~(n» _ g~T(n»)

g~T<n» _ g~(m»

h n -hm ·
So 1] = rI~ + 8h, and r* is surjective. •
3.4 Corollary. If %' is an acyclic covering of X, then
HI(X, V) = HI(%" V).

PROOF: We have HI(%" V) c HI(X, V). If a is an element of HI(X, V),

then there is a covering "I' with a E HI ("I', V). Now we can find a common
refinement "/II' of %' and "1'. Then HI ("I', V) c HI ("/II', V) = HI(%" V), and
therefore a E HI(%" V). •

Generalizations. The simplest case of a vector bundle over X is the

trivial line bundle Ox := X x C. Therefore, the associated eech cohomology
group HI (X , Ox) plays an important role for the function theory on X.
The trivial fiber bundle X x C* is not a vector bundle,but it is not so far from
that. If 1r : P --t X is a general analytic fiber bundle, then a section in P over
an open set U C X is a holomorphic map s : U --t P with 1r 0 s = idu. If the
typical fiber of P is an abelian complex Lie group, then the set r(u, P) of all
sections in P over U carries in a natural way the structure of an abelian group.
In the case of the bundle Ox := X x C* we have a canonical isomorphism

r(U, Ox) ~ O*(U) := {f E O(U) : f(x) -I- 0 for every x E U}.

 3. Cohomology 187

This is a multiplicative abelian group.

If %' = {U~ : LEI} is an open covering of X, then we can form cochains
~ : (L, "') H ~~I< E 0* (Uu<). The set C 1 (%' , Ox) of all these cochains forms a
(multiplicative) abelian group, and we can define the subgroups ZI(%" Ox)
and B 1 (%' , Ox) of cocycles and coboundaries: A cochain ~ is called a cocycle
if ~.V = ~.I<~I<V on U.I<V, and ~ is called a coboundary if there are functions
s. E O*(U.) such that ~~I< = SI<S;:1 on U~K.
Since all the groups are abelian, we have the quotient group

which we call the first cohomology group with values in Ox (with respect to
the covering %'). Just as in the case of vector bundles we can pass to finer
coverings and finally form the Cech cohomology group Hl(X, Ox).
There is a nice interpretation for the elements of Hl(X, Ox). Every co cycle
with values in Ox defines a line bundle over X, and this bundle is independent
of the covering. Two co cycles f.' and ~" define equivalent line bundles if and
only if there are functions h~ with ~:I< = ~:~hl<h;:l, Le., if and only if the
cohomology classes of (,' and (," are equal. Therefore, HI (X, Ox) is the set of
isomorphy classes of line bundles over X. This group is also called the Picard
group of X and is denoted by Pic(X). The group structure is induced by the
tensor product. The identity element corresponds to the trivial bundle Ox
and the inverse to the dual bundle.
In the same way as above we can form cohomology groups of any fiber bundle
P whose typical fiber is an abelian group. If the fiber of P is a nonabelian
group, things become a little bit more complicated. We can define co cycles
with values in P, but they do not form a group. In the set ZI (%', P) of
co cycles we can introduce an equivalence relation by
t:'
'"
t:"
rv '" : {::=:}
3 hL . h "'£1<
WIt t:' = h- 1 g£l<
£
" h
1<.

The set HI (%' , P) of all equivalence classes is called the cohomology set with
values in P (with respect to %'). Usually it is not a group, but there is a
distinguished element, represented by the cocycle ~ with ~£I< = 1 for all L, "'.
Passing to finer coverings and forming the cohomology set Hl(X, P) causes
no problems. The most important nonabelian case is the cohomology set

whose elements correspond to the isomorphy classes of vector bundles of

rank r over X. The distinguished element is represented by the trivial bundle
X x ([7.

In the definition of the cohomology groups of fiber bundles P over X with

an abelian group as typical fiber we used only the fact that feU, P) is an
188 IV. Complex Manifolds

abelian group for· every open set U eX, and that we can restrict sections
over U to open subsets V C U. Having this in mind we can define another
sort of cohomology group.
Let K C C be a subgroup with respect to the addition, for example K = Z, JR,
or C. For an open set U C X we define the (abelian) group K(U) by

K(U) := {f : U -t K : f is locally constant }.

Then K(U) is an abelian group. If V c U is an open subset and f an element
of K(U), then flv E K(V). If U is connected, then K(U) ~ K.
We can define groups of cochains CO (0//, K), C 1 (0//, K), and C 2 (0//, K) just
as we did it for vector bundles. If ~ is an element of C 1 (0//, K), then ~L" E
K(UL,,). Cocycles, coboundaries, and cohomology groups are defined in the
usual way. For example, we have

The Singular Cohomology. We want to give a short overview of co-

homology groups of topological spaces and their relation to Cech cohomology
as defined above. For proofs see [Gre67].
For q E No the set

"x·
n
b. q := {x = (xo, . .. ,xn ) E JRq+1 ~ 1, = 1, X·
t_ > o}
i=O
is called the q-dimensional standard simplex. The O-dimensional simplex is a
point, ~1 a line segment, ~2 a triangle, and so on.
Let X be a topological space. We assume that all spaces here are connected
and locally connected. A singular q-simplex in X is a continuous map a :
~q -t X. If X is a complex manifold and if there is an open neighborhood
U = U(~q) and a smooth map a : U -t X with alAq = a, then a is
called a differentiable q-simplex. A singular q-chain in X is a (formal) linear
combination nlal + ... + nkak of singular q-simplices with ni E Z. The set
Sq(X) of all singular q-chains in X is the.free abelian group generated by the
singular q-simplices.
For a q-simplex a and i = 0, ... ,q the (q - I)-simplex ai : ~q-l -t X is
defined by

ai(XO, . .. , xq-d := a(xo, ... , Xi-I, 0, Xi,·· ., xq-d·

It is called the ith face of a.

 3. Cohomology 189

The (q-1)-chain 8a:= L:?=O(-1)i ai is called the boundary of a. The bound-

ary operator 8 induces a homomorphism

a : Sq(X) -+ Sq-l (X),

and it follows easily that 80 8 = o.

Definition. The group Hq(X) := {e E Sq(X) 8e = 0}/asq+1(X) is
called the qth singular homology group of X.

In general, Ho(X) is isomorphic to the free abelian group generated by the

connected components of X. Since we assume all spaces to be connected, we
have Ho(X) ~ Z. If X is an n-dimensional complex manifold, then Hq(X) = 0
for q > 2n.
Now for q ~ 0 we define the group of singular q-cochains to be

Then the coboundary operator 8 : sq(X) -+ sq+l(X) is defined by

8f[c] := f[ae], for f E sq(X) and c E Sq+l(X).

Obviously, we have 808 = o. We can define cocycles (elements f of sq(X)

with of = 0) and coboundaries (elements of the form og with 9 E Sq-l(X)).

Definition. The group Hq(X) := {J E sq(X) : 8f = O}/8S q- 1(X)

is called the qth singular cohomology group of X.

From above it is clear that HO(X) ~ Z.

A topological space X is called contractible if there is a point Xo E X and a
continuous map F : [0,1] xX -+ X with F(O, x) = x and F(l, x) = Xo. In that
case H I ( X) = {O}. The space X is called locally contractible if every point
of X has arbitrarily small contractible neighborhoods. Among the connected
topological spaces there is a big class of spaces (including the so-called CW-
complexes) that are locally contractible and have the following properties:
1. Hq(X) ~ Hq(X, Z) for q = 0,1,2, ....
2. Every open covering of X has a refinement %' = {UL : I- E I} that is
acyclic in the sense that HI (UL , Z) = 0 for every I- E I.
We call such spaces good topological spaces. For example, every (connected)
complex manifold is a good topological space, and also every irreducible an-
alytic set.
Finitely generated abelian groups are classified as follows:
190 IV. Complex Manifolds

If G is a finitely generated abelian group, then there are uniquely deter-

mined numbers rENo and nl, ... , ns E N with ni ~ 2 and ni Ini+ 1 for
i = 1, ... ,s - 1 such that

The number r is called the rank of G, and zr the free part of G. The sum
of the Zin/i. is called the torsion part of G.
In many cases cohomology can be computed from homology:
3.5 Theorem. If X is a compact good topological space (for example, a
compact connected complex manifold), then Hq(X) is finitely generated for
every q, and for q ~ 1 there is an isomorphism

Hq(X) ~ (free part of Hq(X» EB (torsion part of Hq-1(X)).

The rank of Hl(X) is called the first Betti number of X, and is denoted by
b1(X).
Now we give an application of Cech cohomology methods to singular coho-
mology. Let X and Y be good topological spaces.
3.6 Theorem (Kiinneth formula).

PROOF: We choose open coverings %' = {U. : /, E I} of X and "f/ = {Vn

n E N} of Y such that all the U., VII and all their pairwise intersections are
connected. We write Wm := U. X Vn and Wm,l<m := wmnwl<m = U.I< X Vnm -
A cocycle '1/1 E Zl(%, x 1', Z) is given by constant maps 'I/1m,l<m : Wm,l<m -t Z.
We identify any cocycle ~ E ZI (%' ,Z) with a cocycle f E ZI (%' x 1', Z) by
e..n,l<m = ~.I<' and also any TJ E Zl(1',Z) with an 1j E ZI(%, x 1', Z). This
induces natural injections

with Im(jd n Im(h) = {O}, and therefore an injective map

j : Hl(%"Z) x Hl(1',Z) -t Hl(%, x 1',Z)

by j(a, b) := jl(a) + h(b).

Given a cocycle '1/1 E ZI(%, x 1',Z), for any /, E I we define '1/1. E ZI(1',Z) by
('I/1Jnm := 'I/1m,.m- The cohomology class of '1/1. in Hl(1', Z) is independent of
the index L. We see this as follows_
 3. Cohomology 191

(a) If ULK # 0, then by the cocycle property we have

1/Jtn,Lm - 'l/JKn,Km

Setting (r,oLK)n := ¢Ln,Kn we get a O-cochain r,oLK E CO(1', IE) with Or,oLK =
WK - ¢L' SO the cohomology classes of ¢L and ¢K are equal in this case.
(b) If ULI< = 0, one can find a chain of sets U).." ... , U).N with UL).l =I- 0,
U).i).i+1 # 0, and U).Nt< # 0 (since X is connected), and from (a) it
again follows that the cohomology classes of WL and 'l/Jt< are equal.
Let Wo be a representative of the common cohomology class of the 'l/JL' The
assignment 'I/J f--+ 'l/Jo induces a map p: HI(o// x 1',IE) --t HI(1', IE). We will
prove that for every class e E HI (cJU X 1', IE) there is a class a E HI (0//, IE)
with jl (a) = e - j2(p(e», and consequently j(a, p(e» = c.
For each t there is an 1h E CO(1',IE) with 'l/Jo = ¢L - 01/(7]J, and we define
7] ECO(o// x 1', IE) by 7]Ln := (7]Lk

Then 1:= 'I/J - 'l/Jo - 07] E ZI(cJU x -r,IE) and

IL = WL - ('l/JL - 01/(7]L» - 01/(7]L) = 0 for every t.

Now we construct a e E Zl(cJU,lE) with {j= I:

For n E N define en E ZI (0//, IE) by

Since ILn,Lm = 0 for all n, m, and l, in the case Vnm #0 we have

and therefore em = en. If Vnm = 0, we can argue in the same way as above
in (b), because Y is connected.
So we have a e E Zl (cJU ,IE) with en = e for every nand

eLI< = ILn,Kn = ILn,Kn + It<n,Km = ILn,Km (for arbitrary n, m).

Therefore, {j = 1 and 'I/J = {j + ~ + 07]. This shows that j is surjective. _

Exercises
1. Let all = {Ub U2, U3} be the covering of X := {z E ((: : 1 < Izl < 2}
given by U1 := {x + iy : y < x}, U2 := {x + iy : y < -x}, and
U3 := {x + iy : y> a}. Calculate Hl(cJU, IE).
2. Prove that dime HI (1(:2 - {O}, 0) = 00, where 0 denotes the trivial line
bundle over ((:2. Hint: Use Laurent series.
192 IV. Complex Manifolds

3. Show that HI (C, 0) = O.

4. Show that all = {Ui : i = 1, ... ,n} with Ui := {z : Zi =f. O} is an acyclic
covering of en - {O} for O.
5. Let X be an n-dimensional complex manifold such that Hq(X) = Z for
q even, 0 ~ q ~ 2n, and Hq(X) = 0 otherwise. Calculate the singular
cohomology of X.

4. Meromorphic :Functions
and Divisors
The Ring of Germs. Let X be an n-dimensional complex manifold
and x E X a point. Two holomorphic functions 1,9 defined near x are called
equivalent at x if there exists a neighborhood U = U(x) with Ilu = glu. The
equivalence class of I at x is called the germ of I at x. We denote the germ
by Ix and the set of all germs by Ox'
Having fixed a complex coordinate system r.p : U -t Been at x, we may
identify the set Ox with the ring Hn of convergent power series by

Ix t-t Taylor series of I 0 r.p-I at r.p(x).

So Ox has the structure of a local e-algebra. 2 An element Ix E Ox is a

unit if and only if f(x) =f. O. Of course, Ox is also noetherian and a unique
factorization domain.
4.1 Proposition. Let I, g be holomorphic functions near Xo EX. II the
germs Ixo,gxo are relatively prime in OXO} then there is an open neigborhood
U = U (xo) C X such that Ix, gx are relatively prime in Ox lor x E U.

PROOF: We can work in en and assume that Xo = 0 and that 10 and 90

are Weierstrass polynomials in Zl. Since 10 and go are relatively prime in Hn,
they are also relatively prime in H~_t!ZI]' If Q is the quotient field of H n - I ,
it follows from Gauss's lemma that 10 and go are relatively prime in Q[ZI]'
We can find a linear combination

h = a· 10 + b· 90,
where a, bE Hn-dzl], and h E (Hn-d* is the greatest common divisor of 10
and go. If U = U' X U" c e x e n - l is a sufficiently small neighborhood of
the origin, the power series a, b converge to pseudopolynomials over U" and
h to a holomorphic function on U" that does not vanish identically.
2 A commuta.tive C-algebra A with unity is called a. local C-algebra if the set m of
nonunit!'l forms an ideal in A and the composition of canonical homomorphisms
C,-+ A ...:.. Aim is surjective.
 4. Meromorphic Functions and Divisors 193

Let w = (WI, w') be a point in U. If 'Pw is a common factor of fw and gw,

then it divides hw, which has degree 0 as a polynomial in Zl. So 'Pw does not
depend on Zl and does not vanish identically in Z2, ... ,Zn. Therefore, 'Pw is
zl-regular of order 0, and by the preparation theorem it is, up to a unit, a
Weierstrass polynomial of degree 0; i.e., 'Pw is already a unit.
This shows that f wand gw are relatively prime. •
Analytic Hypersurfaces. Let X be an n-dimensional complex man-
ifold. We consider analytic hypersurfaces A c X. Then locally A is given
as the zero set of one holomorphic function f. We always assume that f
does not vanish identically and therefore A is nowhere dense in X. If locally
A = {(Zl, ... , zn) : Zn = O}, then every holomorphic function g vanishing
on A is of the form g(Zl' ... ,zn) = Zn . g(Zl' . .. , zn). We will generalize this
result to the arbitrary case.
4.2 Proposition. Every hypersurface A C X is a pure-dimensional ana-
lytic set of dimension n - 1.

PROOF: After choosing appropriate coordinates, we may assume that A is

contained in an open set U c en and f is a Weierstrass polynomial w in Zl
without multiple factors such that

A = N(w) = {z E U : w(z) = O}.

Since N (w) is a branched covering over some domain G c en - l , every irre-

ducible component of N(w) has dimension n - 1. •

4.3 Theorem (Nullstellensatz for hypersurfaces). Let A C X be an

analytic hypersurface and Xo an arbitrary point of A.
1. There exists an open neighborhood U = U(xo) C X and a holomorphic
function f on U such that:
(aj UnA = {XE U : f(x) = O}.
(bj If h is a holomorphic function on a neighborhood V = V(xo) C X
with hlA = 0, then there is a neighborhood W = W(xo) C Un V and
a holomorphic function q on W such that hl w = q. Ulw).
2. If i is any holomorphic function defining A in U and h is again a holo-
morphic function on a neighborhood V = V(xo) vanishing on A, then
there exists a kEN and a holomorphic function ij on a neighborhood
W = W(xo) C un V such that hkl w = ij. (ilw).

PROOF: Again we work in en and assume that Xo = o. Let 1

be an ar-
bitrary defining function for A near the origin. After choosing appropriate
194 IV. Complex Manifolds

coordinates we find a unit e and a Weierstrass polynomial W in Zl such that

1= e· w. We choose a neighborhood U = U' x U" of the origin such that:
1. AnU = ((Zl,Z') E U : W(Zb Z') = Ok.
2. There is a prime factorization W = WI 1 ••• W~l on U.
We define W := WI··· WI. This is a pseudopolynomial without multiple factors
that also defines A in U.
If h is a function vanishing on A, which we may assume to be defined on U,
then by the division formula there is a holomorphic function q near the origin
and a pseudopolynomial r with deg(r) < deg(w) such that near 0,

h=q·w+r.

Since w has no multiple factors, the greatest common divisor of wand 8w / 8z I

is a not identically vanishing holomorphic function 9 of Z2, ... , Zn, and we can
find pseudopolynomials ql, q2 with

8w
9 = ql . W + q2 . -8 .
ZI

We may assume that everything is defined on U. Suppose that there is a

z~ E U" such that w((, z~) E er(] has a multiple zero (0. Then w((o, z~) =
8wj8z l ((0, z~) = 0, and therefore g(z~) = O. Hence, if g(z') i= 0, then w((, z')
has exactly s := deg(w) distinct zeros. Since hIN(w) = 0, h((, z') has at least
these s distinct zeros. Using this fact and the division formula, it follows that
r(z') = 0 for z' E U" - N(g). Therefore, by the identity theorem r = 0 and
h = q. w. Taking f = w yields the first part of the theorem.
Let k := max(k l , ... , kl). Then

This proves the second part of the the\>rem. •

Every local holomorphic function f that satisfies the conditions of the first
part of the Nullstellensatz will be called a minimal defining function for A.
Now let A be an irreducible analytic hypersurface in X, and h a holomorphic
function on some open subset U c X with hl(AnU) = O. For Xo E UnA there
exists a neighborhood V = V(xo) c U and a minimal defining function f for
A on V. Then

ordA,xo(h):= max{m E Ii:{ : 3q with h = fm. q near xo}.

It follows from the Nullstellensatz that ordA,xo(h) 2: 1, and from the unique
factorization into primes that it is finite. Furthermore, it is independent of f,
because if h, h are two minimal defining functions, then we have equations
 4. Meromorphic Functions and Divisors 195

It = qi h and h = q21t· It follows that It = qi q21t and therefore It (1 -

qIq2) = o. So qi and q2 must be units.
For every point Xo E A there is a neighborhood U = U(xo) such that
ordA,x(h) ~ ordA,xo(h) for x E UnA.
4.4 Proposition. If A is irreducible, h holomorphic in a neighborhood of
A, and h/A = 0, then the number ordA,x(h) is independent of x EA.

PROOF: Let Xo E A be an arbitrary point. In a neighborhood U of Xo there

exist a decomposition An U = Al U ... U Al into irreducible components
and minimal defining functions 1>.. for A,\. Since h vanishes on every A,\,
there exist k I , ... , kl E N and a holomorphic function q on a neighborhood
V = V(xo) c U such that
h = f~l ... ft' . q, and (h)xo f qxo for .A = 1, ... ,l.
Since (It)xo' ... ,(ft)xo are irreducible, it follows that (h)xo and qxo are rel-
atively prime for .A = 1, ... ,l. But then (h)x and qx remain relatively prime
for x sufficiently close to xo: say in a neighborhood W(xo) c V.
Let n(x) := ordA,x(h). It is necessary to consider two cases:
(a) If A is irreducible at Xo, then 1 = 1, and it is clear that n(x) = n(xo) for
x E WnA.
Consequently, x H n(x) is a locally constant integer-valued function on the
. .
set A of regular points of A. Since A is globally irreducible, A is connected
and n(x) globally constant on A. Let n* EN be the value of this function.
(b) If l > 1, then f := It··· fl is a minimal defining function for A at Xo.
With m := min(k 1 , ... , kt) we have
h = f~l ... flkl . q = fm . fl,

where fl is a holomorphic function near Xo. Therefore, n(xo) 2: m.

We assume that m = k,\. In every small neighborhood of Xo there are regular
points x E A,\ that do not belong to AI' for Mi= .A. Then n(x) = n*, and f'\ is
a minimal defining function for A at x. Since h = f~). . q, with a holomorphic
function q, and (h)x f qx, it follows that n(x) = k,\.
So m:S n(xo) :S n(x) = n* = k,\ = m, and therefore n(xo) = n*. •
Now we define
ordA(h) ;= { coonstant value of ordA,x(h) if h/A = 0,
otherwise.
One easily sees that
196 IV. Complex Manifolds

Meromorphic Functions. Let X be an n-dimensional complex mani-

fold. We consider holomorphic functions that are defined outside of an ana-
lytic hypersurface. In the I-dimensional case these are holomorphic functions
with isolated singularities.
Definition. Let A c X be an analytic hypersurface. A complex-valued
function m on X - A is called a meromorphic function on X if for any
point x E X there are holomorphic functions g, h on an open neighbor-
hood U = U(x) c X such that N(h) cAn U and m = g/h on U - A.

Obviously, m is holomorphic on X - A. In particular, every holomorphic

function f on X is also meromorphic on X.
Different meromorphic functions may be given outside of different analytic
hypersurfaces. If m).. : X - A).. -+ C are meromorphic functions on X, then
ml ± m2 and ml . m2 are meromorphic functions on X, given as holomorphic
functions on X - (AI U A2)'

If m : X - A -+ C is a meromorphic function, for pEA we have two

possibilities:
(a) There is a neighborhood U = U(p) c X such that m is bounded on U -A.
Then there is a holomorphic function in on U with injU-A = mju-A, and
p is called a removable singularity for m.
(b) For any neighborhood V = yep) c X and any n E N there is a point
x E V - A with jm(x)j > n. If m = g/h near p, then h must vanish at
p, because otherwise we would be in situation (a). Now there are again
two possibilities:
(i) If g(p) =I 0, then limx-+pjm(x)j = +00, and we have a pole at p.
(ii) The other possibility is g(p) = O. This cannot occur in the case n = 1,
since it may be assumed that the germs gp and hp are relatively
prime, but it is possible for n > 1. The behavior of m is extremely
irregular in that case: We take any c E C. Then gp - c . hp and hp are
relatively prime, and therefore there exists a sequence (xv) of points
in N(g-ch)-N(h) with limv-+oo Xv = p. This means that m(x v ) = c
for every v. We call p a point of indeterminacy in this case.
In the case n = 1 a meromorphic function is a function that is holomorphic
except for a discrete set of poles. For n > 1 we have the polar set
P(m) := {p EX: m is unbounded in any neighborhood of p}.

The polar set consists of poles and points of indeterminacy. We show that
P(m) is an analytic hypersurface.
Let p E X be an arbitrary point and U = U(p) c X a connected neighbor-
hood, where m is the quotient of 9 and hand N (h) c A. We may assume
that pEA and gp, hp are relatively prime. Then
 4. Meromorphic Functions and Divisors 197

m bounded near p <==> 3 <p (holomorphic near p) with 9 = <p • h

<==> hplgp
<==> hp is a unit
<==> h(P) =I: O.
So P{m) n U = {x E U : hex) = OJ.
If Z c X is an irreducible hypersurface, then we define ordz(m) as follows: If
m = glh near x, then ordz,x{m) := ordz,x(g) -ordz,z{h).1f we choose gx. hx
relatively prime, this definition is independent of 9 and h. Now it follows
exactly as above that ordz,x(m) is constant on Z.
4.5 Identity theorem for meromorphic functions. Let X be connected,
m : X - A -+ C a meromorphic junction, and U C X a nonempty open set
such that mlu-A = O. Then P(m) = 0 and m = O.
PROOF: The set X - P{m) is connected, m is holomorphic there, and U-
(AUP(m» is a nonempty open subset of X -P(m). By the identity theorem
for holomorphic functions it follows that m = 0 on X - P(m). But then m
is globally bounded and P(m) = 0. •

The set .4l(X) of meromorphic functions on X has the structure of a ring

with the function m = 0 as zero element. We set

.4l(X)* .- .4l(X) - {OJ

= {m E .4l(X) : m vanishes nowhere identically}.
If m E .4l(X)" has a local representation m = glh, the zero set N{g) is
independent of this representation. Therefore, we can define the global zero
set N(m), which is an analytic hypersurface in X. Outside of P{m)UN(m), m
is holomorphic and without zeros. Therefore, 11m is also holomorphic there
and has local representations 11m = hlg. So 11m is also meromorphic, and
consequently .4l(X) is a field. For this it is essential that X is connected!
4.6 Levi's extension theorem. Let A C X be an analytic set that has at
least codimension 2, and let Tn be a meromorphic function on X-A. Then
there exists a meromorphic function m on X with mix -A = m.
PROOF: Since the statement is true for holomorphic functions, we may
assume that P(m) =I: 0. So it is an analytic hypersurface in X - A. By the
theorem of Remmert-Stein, Q := P( m) is an analytic set in X. By Riemann's
second extension theorem the holomorphic function m on (X - Q) - A has
a holomorphic extension mto X - Q.
Let pEA n Q be a point. We have to show that mis locally meromorphic at
p. We choose an open neighborhood U = U(P) c X and a function 9 E O(U)
such that:
198 IV. Complex Manifolds

1. Q n U c N(g).
2. N(g) = Nl U··· U Nk is a decomposition into irreducible components.
Since dim(A) < dim(Ni ), there are points ai E Ni - A and neighborhoods
m
Vi = Vi(ai) C U -A such that = Pi/qi on Vi-Q and N(qi) C QnVi c N(g);
i.e., gIN(q;) = O.
From the Nullstellensatz it follows that there is a number Si E N and a
holomorphic function ri such that gSi = ri . qi. Then m
= Pirig-Si near ai'
This means that there exists an sEN such that gS . m is holomorphic near
al> ... ,ak·
Thus N := (U - A) n P(gSm) is empty or an analytic hypersurface that is
contained in N (g) - A. In the latter case it is a union of irreducible components
of N (g) - A, and this is impossible, since every such component contains
a point ai' So N must be empty, and gSm is holomorphic in U - A. By
Riemann's second extension theorem there is a holomorphic extension h of
gSm on U. Then g-Sh is meromorphic on U with (g-Sh)lu_A = m. •

Divisors. Let X be a connected complex manifold, m E .4l(X)*, and

Z C X an irreducible analytic hypersurface. If Z c P(m), then ordz(m) is a
negative integer, and if Z c N(m), then ordz(m) E N. In all other cases we
have ordz(m) = O.
If P(m) = ULEI p. and N(m) = u.>'EL N>. are the decompositions of the polar
set and the zero set into irreducible components, then the formal sums

(m)oo := I) - ord p, (m)) . PL and (m)o:= L ordNJm) . N>.

LEI >'EL

are called respectively the divisor of poles and the divisor of zeros of m.
Finally, div(m) := (m)o - (m)oo is called the divisor of m. From the remarks
above it is clear that

div(m) = L ordz(m) . Z,
zcx
where the sum is over all irreducible hypersurfaces Z in X.
Definition. Let (ZL)LEI be a locally finite system of irreducible ana-
lytic hypersurfaces ZL EX. If for every ~ E I a number n L E Z is given,
then the formal linear combination

D = Ln.. Z.
LEI

is called a divisor on X.
The divisor is called positive or effective if n. ?:: 0 for every ~ E I.
 4. Meromorphic Functions and Divisors 199

Divisors can be added or multiplied with integer constants in an obvious way.

Therefore, the set ~(X) of all divisors on X has the structure of an abelian
group.
As we just have seen, there is a map div : Al(X)· -+ ~(X). Since we have
div(mlm2) = div(mt} + div(m2), div is a group homomorphism.
Sometimes it is useful to generalize the notion of a divisor a little bit. Let
(A")"EI be a locally finite system of (arbitrary) analytic hypersurfaces in X,
and (n")"EI a system of integers. Then for every t E I we have a decomposition

into irreducible components. The system {A~, : t E I,.A" E L"} is again

locally finite, and we define E"EI n" . AL := E"EI EA,EL, n, . A~,.
With this notation it is possible to restrict divisors to open subsets: If U c X
is open and D = E"EI n" . Z" a divisor on X, then

Du:=
" with Z,nUi'0

4.7 Proposition. If A c X is an analytic hypersurface and f a minimal

defining function for A in an open set U with UnA :f. 0, then

div(Jk) = k· A n U.

The proof is more or less straightforward.

Now let an arbitrary divisor D on X be given. Then for any point p E X there
is an open neighborhood U = U(p) eX, a finite system {Zi : i = 1, ... , N}
of irreducible hypersurfaces Zi C U, and a system of numbers ni E Z such
that
N
Diu = L:: ni . Zi.
i=1

In addition, there is a neighborhood V = V (p) C U such that there exist

minimal defining holomorphic functions fi for Zi in V. Then
N N
div(IT r:i) = L:: ni' Zi n V = Dlv.
i=1 i=1

In this way every divisor is locally the divisor of a meromorphic function.

200 IV. Complex Manifolds

Associated Line Bundles. Let X be a connected n-dimensional com-

plex manifold. If Z c X is an analytic hypersurface, then there is an open
covering tfI = {U. : i E I} of X with the following property:
If Ut n Z -I- 0, then there is a minimal defining function i. E O(U.) for Z.
Setting ft := 1 if U. n Z = 0, in UtK we get the two relations

with suitable holomorphic functions g'K and gK •. Then

Since f. does not vanish identically, it follows that g.KgK. = 1 on UtK . This
shows that gLI< E O*(ULK ) and gl<L = g;;/. Furthermore, on ULI<). we have the
compatibility condition
g.Kgl<). = gL).·
The system of the nowhere vanishing functions 9.1< = f'/ fK defines a holo-
morphic line bundle on X, which we denote by [Z]. It is easy to show that
this definition does not depend on the covering and the functions f •.
4.8 Proposition.

1. There is a section Sz E reX, [Z]) with Z = {x EX: sz(x) = O}.

2. [Z] is trivial over X - Z.

PROOF: The system of holomorphic functions f. defines a global section Sz

with {x E UL : sz(x) = O} = {x E UL fL(X) = O} = U. n Z. Then it is clear
that [Z]lx-z is trivial. -

We can generalize the concept of associated line bundles to the case of divi-
sors. If D is a divisor on X, then there is an open covering tfI = {U. : i E I}
of X, and meromorphic functions m. on UL with Diu, = div(m.). It follows
that the functions

are nowhere vanishing holomorphic functions on ULI<. They define a line bun-
dle, which we denote by [D). If D = k· Z, then [Dj = [Zjk. If D = Dl + D 2 ,
then [D] = [Dd ® [D 2 ]. Thus the map

15 : ~(X) -t Pic(X), 15(D):= isomorphy class of [D],

is a homomorphism of groups.
4.9 Theorem. The sequence of group homomorphisms

.4l(X)* ~ ~(X) ~ Pic(X)

 4. Meromorphic Functions and Divisors 201

is exact.

PROOF: (1) Let m i= 0 be a meromorphic function. Then [div(m)] is given

by only one transition function m/m = 1. Therefore, 80 div(m) = 1.
(2) Let D be a divisor on X with 8(D) = 1. We assume that Diu, = div{mJ
and [D] is represented by g~", = m~/m",. Since [D] is trivial, there are nowhere
vanishing holomorphic functions h~ with h • . g.", = 1 . h", on U.",. Then

Therefore, a meromorphic function m on X can be defined by mlu, := h. ·m•.

Obviously, div(m)lu, = div(m.) = Diu" and therefore div(m) = D. •

Meromorphic Sections. Let X be a connected n-dimensional complex

manifold. Any analytic hypersurface Z C X leads to a line bundle [Z], to-
gether with a global holomorphic section Sz that vanishes exactly on Z. The
construction of Sz fails in the case of an arbitrary divisor D and its associ-
ated line bundle [D]. Therefore, we introduce the notion of a meromorphic
section.
Definition. Let 7r : L -+ X be an analytic line bundle and A C X an
analytic hypersurface. A holomorphic section s E r(X - A, L) is called
a meromorphic section over X in L if for every point x E X there is
an open neighborhood U = U(x) C X, a function h E O{U), and a
holomorphic section t E r(U, L) such that:
l. h· s = t over U - A.
2. N(h) cAn U.

If we have a system of trivializations CPt : 7r-l(U.) -+ U. X C and transition

functions 9."" then we have the following description of s.
For x E U. - A we define s.(x) by 'P. os(x) = (x, s.(x)). Then s. = g,,,,' s'" on
U,'" - A. If we choose the U, small enough, there are holomorphic functions
h., t. on U, such that h• . s. = t. over U. - A and N(hJ cAn U•. This
means that s. is a meromorphic function on U•. We could have as well said
that a meromorphic section is a system (s.) of meromorphic functions with
s. = g.", . s"'.
If Z is an irreducible hypersurface, then ordz(s.) = ordz(s",), and we denote
this number by ordz(s). The sum

div(s) := ordz(s) . Z
zcx irreducible hypersurface

is called the divisor of the meromorphic section s.

202 IV. Complex Manifolds

Now we have the solution for our problem: Let D be the divisor on X given
by Diu, = div(m.). Then [D] is described by the transition functions gu< =
m.m;l, and the system of the m. defines a global meromorphic section SD
of [D] with div(sD) = D.
So far our definitions seem to be purely tautological. But for example, they
allow us to determine the space of holomorphic sections of [D] in terms of
meromorphic functions on X. For that we need the following notation: If
D 1 , D2 are two divisors on X, then Dl 2: D2 if and only if Dl - D2 is a
positive divisor.
4.10 Theorem. Let D = Eznz . Z be a divisor on X. Then there is a
natural isomorphism {m E ..4t(X)* : div(m) 2: -D} ~ reX, [DJ).

PROOF: Let SD be the global meromorphic section of [D] with div(sD) = D.

Then for any meromorphic function m on X also t := m·SD is a meromorphic
section of [D].
If m is a meromorphic function with div(m) 2: -D, then

div(t) = div{m) + div{SD) = div(m) + D 2: -D +D = O.

This means that t is a holomorphic section. The map m H m· SD is obviously

injective.
Let t E f(X, [D]) be given. If Diu, = div(m.), then

Hence t.m;:l = t",m;l on U."" and there exists a meromorphic function m on

X with mlu. = t.m;:l. Therefore, div(m)lu. = div(t.) - div(m.) 2: -Diu,.
So the map is an isomorphism. _

Example
Let X = C be the Riemannian sphere and D = n . 00. Then
r(x, [D]) = {m E .A(X)* : ordoo(m) 2: -n and ordp(m) ~ 0 otherwise}.

The holomorphic sections in [D] are just the meromorphic functions on X

that have a pole of order at most n at 00 .

Exercises
1. Let x be a point in a complex manifold X and let lx, gx be nonunits in Ox.
Prove that Ix and gx are relatively prime if and only if dim(N(f, g)) ::;
n-2.
 5. Quotients and Submanifolds 203

2. Let G c en be a domain, A eGan analytic hypersurface, U eGan

e
open subset with UnA f=. 0, and f : U ~ a minimal defining function
for A. Prove that Sing(A) n U = {z E U : fez) = 0 and '\If(z) = O}.
3. Consider the meromorphic function m(zl' Z2) := z21z1 on e 2 . Show that
the closure X of the graph {(Zl' Z2, w) : Zl f=. 0 and w = m(zl' Z2)} in
e 2 xC is an analytic hypersurface. Determine a minimal defining function
for X at (0,1,00).
4. Classify the singularities of m(zl' Z2) := sin(zl)1 sin(zlz2).
5. Let L -+ X be a holomorphic line bundle. Prove that L = [DJ for some
divisor D on M if and only if L has a global meromorphic section s f=. O.
6. Let X be a compact Riemann surface. Show that for every nonconstant
meromorphic function f on X the numbers of zeros and poles are equal
(counted with mUltiplicity).

5. Quotients and Submanifolds

Topological Quotients. Let X be an n-dimensional complex manifold
and "-' an equivalence relation on X. If x, Y E X are equivalent, we write
x "-' y or R(x, y). For x E X let

X(X):={YEX: y"-'X}={yEX: R(y,x)}

be the equivalence class of x in X. These classes give a decomposition of X

into pairwise disjoint sets. The set XI R of all equivalence classes is called the
topological quotient of X modulo R.
Let 7r : X -+ XI R be the canonical projection given by 7r : x r-+ X(x). Then
XI R will be endowed with the finest topology such that 7r is continuous. This
means that U C XI R is open if and only if 7r-l(U) C X is open. We call this
topology the quotient topology.
A set A C X is called saturated with respect to the relation R if

5.1 Proposition.

1. A saturated {:::::=} A = UXEA X(x).

2. If U C XI R is open, then 7r- 1 (U) is open and saturated.
3. If We X is open and saturated, then 7r(W) C XI R is open.

The proof is trivial.

204 IV. Complex Manifolds

5.2 Proposition. Let Z be an arbitrary topological space. A map f

XI R -t Z is continuous if and only if f 0 7r : X -t Z is continuous.
This statement is also trivial, since (f 0 7r)-l(U) = 7r-l(f-l(U)).

Analytic Decompositions. If X is an n-dimensional complex mani-

fold and R an equivalence relation on X, one can ask whether XI R carries
the structure of a complex manifold such that 7r is a holomorphic map. As-
sume that such a structure exists. Then XI R must be a Hausdorff space. If
cp : U -t C k is a complex coordinate system for XI R, then D := 7r- l (U)
is an open saturated set in X, and f := cp 0 7r : D -t C k a holomorphic
map with f-l(f(x)) = 7r-l(7r(X)) = X(x). So the fibers of f are equivalence
classes, and the equivalence classes must be analytic sets. If additionally 7r is
a submersion, then rkx(f) = k for every xED, and the fibers are (n - k)-
dimensional manifolds. We now show that these conditions are also sufficient
for the existence of a complex structure.
Let X be an n-dimensional complex manifold and Z = {ZL : ~ E I} a decom-
position of X into d-dimensional analytic sets. For x E X let ~(x) E I be the
uniquely determined index with x E ZL(X). Then there is an equivalence rela-
tion R on X such that the equivalence class X(x) is exactly the analytic set
ZL(X). We consider the topological quotient XI R and the canonical projection
7r : X -t XI R and assume that the following conditions are fulfilled:

1. XI R is a Hausdorff space.
2. For any Xo E X there exists a saturated open neighborhood D of X(xo)
D
in X and a holomorphic map f : -t C n - d such that
(a) f-l(f(x)) = X(x) for all x E D.
(b)' rkx(f) = n - d for x E D.

5.3 Theorem. Under the conditions above, XI R carnes a unique structure

of an (n - d) -dimensional complex manifold such that 7r : X -t XI R is a
holomorphic submersion.

PROOF: Let Xo E X be given. Then there is an open neighborhood U of

X(xo) in X with 7r- 1 C7r(D)) = D,
and a submersion f: D
-t C n - d whose
fibers are equivalence classes X(x). If Zo := f(xo), then there is an open
neighborhood W = W(zo) C C n - d and a holomorphic section s : W -+ D
(with s(zo) = Xo and f 0 s = idw). For z E W we have f-l(z) = X(s(z)),
and therefore

7r-I(7r(S(W))) = U X(s(z)) = U f-l(z) = {-leW).

zEW

This is an open set, so 7r(s(W)) C XI R is open as well. We define a complex

coordinate system cp : 7r(s(W)) -t C n - d by
 5. Quotients and Submanifolds 205

cp(1T(S(Z») := z.

Then cp(7r(x)) = f(x). This shows that cp is well defined and continuous. It is
also bijective, with cp-l(z) = 7r(s(z», and therefore a homeomorphism.
Now let 'l/J be another coordinate system given by 'l/J(7r(t(z»)) := z, where t is
a local section for some suitable submersion g. Then

cp 0 'l/J-l(z) = cp(7r(t(z») = f(t(z)).

The coordinate transformations are holomorphic. •

Properly Discontinuously Acting Groups. Let G be a complex
Lie group acting analytically on an n-dimensional complex manifold X. Then

R(x, y) : {:::::::> :3 9 E G with y = gx

defines an equivalence relation on X. The equivalence class X(x) = {y EX:

:3 9 E G with y = gx} is called the orbit of x under the group action and is
also denoted by Gx. The topological quotient X/ R is called the orbit space
and is also denoted by X / G.
We consider a very special case.
Definition. The group G acts properly discontinuously if for all x, y E
X there are open neighborhoods U = U(x) and V = V(y) such that

{g E G : gU n V =I 0}
is empty or a finite set.

Here the orbits Gx are discrete subsets of X and are therefore O-dimensional
analytic subsets. If the action is free, we want to show that all conditions are
fulfilled for X/G to be a complex manifold and 7r : X --+ X/G a holomorphic
submersion (which means in this case that 7r is an unbranched covering).
5.4 Lemma. Let G act freely and properly discontinuously on X and let
Xo, Yo E X be given.
1. If there is a go E G with Yo = goxo, then there are neighborhoods U =
U(xo) and V = V(Yo) such that gU n V = 0 for 9 =I go. In the case
Yo = Xo and go = e one can choose V = U.
2. If gxo =I Yo for every 9 E G, then there are neighborhoods U = U(xo)
and V = V(yo) such that gU n V = 0 for every 9 E G.

PROOF: At first we choose neighborhoods Uo(xo) and Vo(Yo) such that

M := {g E G : gUo n Vo =I 0}
206 IV. Complex Manifolds

is finite or empty. There is nothing to prove if M = {go} in the first case

or M = 0 in the second case. Therefore we assume that there are elements
gl,···,gN, N ~ 1, with M = {go,gl, ... ,gN} in the first case and M =
{gl, ... , gN} in the second case. Then we define Y>.. := g>..xo, for)' = 1, ... , N.
Since G acts freely, Y>. =f Yo for ). = 1, ... , N.
We choose neighborhoods W>.. = W>..(y>..) and V = V(yo) C Vo such that
W>.. n V = 0, and we choose a neighborhood U = U(xo) c Uo such that
g>.U C W>., for)' = 1, ... , N. Then gU n V = 0 for 9 =f go in the first case,
and 9 E G in the second case. •

5.5 Theorem. Let G act freely and properly dicontinuously on X. Then

X / G has the unique structure of an n-dimensional complex manifold, so that
7[ : X -+ X/G is an unbranched holomorphic covering.

PROOF: Let U C X be an open set. Then 7[-l(7[(U)) = UgEG gU is an

open set, and therefore 7[(U) is also open. For Xo E X we can choose an open
neigborhood U = U(xo) such that gUnU = 0 for g =f e. Then 7[ : U -+ 7[(U)
is bijective.
(1) We have to show that X/C is a Hausdorff space. Let Xl>X2 E X be
given, with 7[(xt} =f 7[(X2). Then gX1 =f X2 for every g E C. There are open
neighborhoods U = U(X1) and V = V(X2) with gU n V =f 0 for every 9 E G.
Then 1I"(U) and 7[(V) are disjoint open neighborhoods of 1I"{xt} and 1I"(X2).
(2) We verify the other conditions. Let Xo E X be given and choose a small
open neigborhood U = U(xo) C X such that 7[ : U --+ 7[(U) is a homeomor-
phism and such that there exists a complex coordinate system <p : U -+ en.
Then f : fl := 7[-1(1I"(U)) -+ en can be defined by f(gx) := <p(x), for x E U
and 9 E G. It is clear from above that f is well defined. The fibers of f are
the G-orbits, and on gU we have f(y) = <p(g-l(y)). This shows that f is
holomorphic, and rky(f) = n for every y E fl.
If U is small enough, then 7[-I(7[(U)) = U gEG gU, with pairwise disjoint sets
gU that are topologically equivalent to 7[(U). So 7[ is an unbranched covering .
•
Complex Tori. Let {WI, ... ,W2n} be a real basis of en. Then the discrete
group r := Zwl + ... + Zw2n acts freely on en by translation. The set

Aw := r +w = {w + w : wE r}
is the orbit of w.
The group r acts properly discontinuously on en: Let zo, Wo E en be given.
If Wo = Wo + Zo for some Wo E r, choose
 5. Quotients and Submanifolds 207

c< ~ . inf{lIwll : wE r- {OJ }.

Then (w + Bt:(zo)) n Bt:(wo) = 0, unless w = woo

lf Wo - Zo IlT and
c< ~ . dist(wo, r + zo),
then (w + Bt:(zo)) n Be (wo) = 0 for every w.
The n-dimensional complex manifold Tn = TF := c;n Ir is called a complex
torus, and r is called the lattice of the torus.
The set P := {z = t1wl + ... + t2nw2n : 0 ~ ti ~ I} contains a complete
system of representatives for the equivalence classes. Therefore, Tn = 7r(P)
is a compact space. The map

induces a homeomorphism Tn --t 8 1

5.6 Proposition. HI (8 1 , '1.) =

--------
z.
X ... X

2n times
8 1.

PROOF: Let U1 := {e 211"it : -:1

< t < H,
U2 := {e211"it : 0 < t < n,
and U3 := {e211"it : :1
< t < I}. Then au := {UI,U2 ,U3 } is an acyclic open
covering for 8 1 with U123 = 0. Therefore, every triple ~ = (a, b, c) E '1.3
is a co cycle in Zl (au, '1.). It is a coboundary if and only if there is a triple
(u,v,w) EZ3 with

a = v - ·u, b = w - u, and c = w - v.

This is the case if and only if a + c= b. Since every co cycle has the form
(a, b, c) = (0, b - a - c,O) + (a, a + c, c) = (b - a - c) . (0,1,0) + 15(0, a, a + c),
it follows that Hl(au, '1.) is generated by the class of (0, 1,0). •
From the Kiinneth formula it follows that Hl(Tn,z) ~ z2n, and therefore
the first Betti number of Tn is equal to 2n.

Hopf Manifolds. Let {! > 1 be a fixed real number and n > 1. Then
the (multiplicative) group r := {{!k : k E Z} acts freely on c;n - {OJ by
z t--t {!k . z.
The action is properly discontinuous. To see this, we define the sets

Ur := {z E c;n : r < IIzlI < {!T}, for r > O.

Then the sets (lur are pairwise disjoint. If two points Zl, Z2 E c;n - {OJ
are given, one can find an r > 0 and a k E '1. such that Zl E U := Ur and
Z2 E V := {!kUr . The case k = 0 is allowed. Now flU n V = 0, unless s = k.
208 IV. Complex Manifolds

So H = Hr := (C n - {OJ )jr is an n-dimensional complex manifold, and

the canonical projection 1f : C n - {OJ -+ H is an unbranched holomorphic
covering. H is called a Hopf manifold.
The map
( .lnllzll) z )
z f--7 exp ( 21f1 In (J '11zll

induces a diffeomorphism H -+ Sl X s2n-1. Here s2n-1 is the (2n - 1)-

dimensional sphere in JR2n = C n with 2n - 1 ~ 3.
5.7 Proposition. For k ~ 2, Hl(Sk,Z) = O.

PROOF: We have Sk = {x = (Xl, ... ,Xk+d Ilxll I}. Then all

{Ul , U2} with

-c < Xk+l :s:

I},
-1 :s: Xk+l < c}
is an open covering of Sk with contractible sets, and

U12={XES k : -c<Xk+! <c}

is connected. Therefore, Cl(o//, Z) = Zl(o//, Z) = Z and CO (0//, Z) = Z2. The

coboundary map ~ : CO (0// , Z) -+ C1(o//, Z) is given by ~(a, b) := b - a. Then
obviously, Bl (0//, Z) = Z and HI (0//, Z) = O. •
It follows that H1(H,Z) = Hl(Sl x S2k-l,Z) = Z, and bl(H) = 1.
The Complex Projective Space. In X:= C n +! - {OJ we consider
the equivalence relation

R(z, w) : {=:} 3), E C* with w = ).z.

The equivalence class L z of z is the set Lz = Cz - {O}, the complex line
through z and 0 without the origin. So we have a decomposition of X into
I-dimensional analytic sets. We can also look at these sets as the orbits of
the canonical action of C* on X by scalar multiplication.

Definition. The topological quotient lpm := XjR = (cn+1 - {O})jC*

is called the n-dimensional complex projective space.

Let 1f : X = C n + 1 - {OJ -+ IPn be the canonical projection, with 1f(z) := L z ,

and let two points z = (zo, ... , zn), w = (wo, ... , w n ) be given. We have

1f(z) = 1f(w) {=:} 3), E C* with Wi = ).Zi for i = 0, ... ,n

{=:} Wi = Zi for all i, j where the fractions are defined.
Wj Zj
 5. Quotients and Submanifolds 209

So 7r(z) does not determine the entries Zj, but the ratios Zi : Zj. Therefore,
we denote the point x = 7r(zo, ... ,zn) by (zo : ... : zn) and call zo, •.• , Zn the
homogeneous coordinates of x. If Zo, ... , Zn are homogeneous coordinates of
x, then so are AZo, •.. , AZn for every A E C·.
If We X is an open set, then 7r- 1(7r(W)) = U,xEC' A' W is a saturated open
set in X, and therefore 7r(W) is open in Ipm. For example, this is true for

fJ; := {z = (zo, ... , zn) E Cn +! - {O} : Zi "# O} eX, i = 0, ... , n.

The sets Ui : = 7r ( fJi ) form an open covering of Ipm.

We show that IP'n is a Hausdorff space: Let z, w E X be given, with Lz "# Lw'
Then
*
Z :=
Z
W
and w* := IIwllw

are distinct points of s2n+! = {x E ]R2n+2 = C n +! : IIxll = I}. Therefore,

we can find an € > 0 such that Bt:(z·) n Bt:(w*) = 0. Then U := 7r(Bt:(z*))
and V := 7r(Be:(w*)) are disjoint open neighborhoods of 7r(z), respectively
7r(w).

Now let a point Zo = (zbO), ... , z~O») E X be given. Then there exists an index
i with z~O) "# 0, and Zo lies in fJi . We define fi : fJi ~ C n by

ZO Zi-l Zi+l Zn)

fi ( ZO"",Zn ) := ( - , ... , - - , - - , ... , - .
Zi Zi Zi Zi

Then
z·
{ w E Ui
~
:
W·
-1.. = .2. for j "# i }
Wi
W.}
Zi

{WEUi: ~
W= Zi'·Z

7r- 1 (7r(Z)).

If a point u = (Uo, ... , un) E fJ; is given, we define a holomorphic section

S : C n ~ fJ; by S(Zl"'" zn) := (UiZl, ... , UiZi, Ui, UiZi+l, ... , UiZn). Then
UO Ui-l Ui+l Un)
S ( - , ... , - - , - - , .•• , - = (uo, ... ,un ),
Ui Ui Ui Ui

and
fi 0 S(Zl"'" Zn) = (Zl, ... , zn).

Therefore, fi is a submersion, and rkz(fi) = n for every z.

Altogether this shows that IP'n is an n-dimensional complex manifold and
7r : C n +!
- {O} ~ IP'n a holomorphic submersion. Since every equivalence
class Lz has a representative in the sphere S2n+l, IP'n = 7r(S2n+!) is compact.
210 IV. Complex Manifolds

Local coordinates are given by the maps 'Pi : Ui -t en with 'Pi 0 tr = f i . This
means that
.
'Pi (Zo ...... Zn ) -_ (ZO -
Zi ... , -Zn) . 3
-, ... , -,
Zi Zi Zi
The set

is biholomorphically equivalent to en. We call it an affine part of ]pn. If we

remove Uo from F, we get the so-called (projective) hyperplane at infinity
Ho = {(zo: . .. : Zn) E]pn : Zo = O}
= {(O: tl : ... : tn) : (tb ... , tn) E en - {O} }.
It has the structure of an (n - 1)-dimensional complex projective space. If we
continue this process we get
U ]pn-l,
U pn-2,

=
It remains to study ]pI = {(zo : Zl) : (zo, zt) E e 2 - {O} }. But this is the
union of e = {(I : t) : tEe} and 00 := (0 : 1), with t = zdzo. In a
neighborhood of 00 we have the complex coordinate zof ZI = 1ft. So we see
that ]pI = C = e u {oo} is the well-known Riemann sphere.
The hyperplane Ho is a regular analytic hypersurface, given by

u n Ui
no = {( Zo:· .. : Zn ) f u·.
i Zo =
Zi
O}.
Therefore, Uo is dense in ]pn.
It should be remarked that there is no reason to distinguish between Uo and
the other sets Ui . Everything above could have been done as well with the
affine part Ui and the hyperplane Hi := {tr(z) E]pn : Zi = O}.

Meromorphic Functions. On a compact complex manifold every

global holomorphic function is constant. But we know already from the ex-
ample of the Riemann sphere that there may exist nonconstant meromorphic
functions. In this regard we consider the compact manifolds defined above,
beginning with the complex projective space F.

A nonconstant polynomial p(t) = E~I=o a"t" is a holomorphic function on

the set Uo = {(I: tl : ... : tn) It = (t l , ... ,tn) E en}. In fact, it defines a
meromorphic function on ]pn with polar set Ho. We see this as follows:
3 The hat signalizes that the ith term is to be left out.
 5. Quotients and Submanifolds 211

The functions t/l- = z/l-/ Zo, J.L ? 1, are holomorphic coordinates on Uo, and
w>. := Z>./Zi, oX i= i likewise on Ui . Therefore, on Ui - Ho = Ui n Uo we have

L
k
( ZO)k. av(ZI)VI ... (Zn)vn
Zi Zo Zo
Ivl=O
k
~ a wk-lvlwVI .. ·w vn .
~ V 0 In'
Ivl=O

.
l.e., p -- g /h on U·t - H 0, were
h .- "k
g (w) .- L.,.,lvl=O avwok-Ivl WIVI... WnVn an d
h(w) := wa are holomorphic functions on Ui with

N(h) = {w E Ui : Wo = O} = Ui n Ho.
So there are numerous global meromorphic functions on projective space.

Now let T = en /r be an n-dimensional complex torus, and 7r : en -+ T

the canonical covering. If m is a meromorphic function on T, then m 0 7r is a
merom orphic function on en, which is periodic with respect to the generators
WI, ... ,W2n of the lattice r. In the case n = 1 such meromorphic functions
always exist; they are the r-elliptic functions. We shall later see that for
n ? 2 the existence of r -periodic functions depends on the lattice r. In fact
there are complex tori with no nonconstant meromorphic functions.

Finally, consider the Hopfmanifold H = (Cn-{O})/r with r = {gk : k E Z}

and n > 1. Let m be a meromorphic function on H. Since the canonical
projection 7r : en - {O} -+ H is a covering, m := m 0 7r is meromorphic on
C n - {O}. Since n > 1, it follows from Levi's extension theorem that can m
be extended to a meromorphic function on C n . On any line L through the
origin in en, m must have isolated poles or be identically 00. But since m
comes from H, poles on L must have a cluster point at the origin, which
is impossible unless m is constant on L. The same argument works for any
other value of m. A meromorphic function on C n - {O} that is constant on
every line through the origin comes from a meromorphic function on the
projective space lPn-I. This means that if h : H -+ IPn - 1 is the canonical
map, then a bijection Al(IPn-l) -+ Al(H) is defined by m I--t m 0 h. On the
n-dimensional Hopf manifold there are not "more" meromorphic functions
than on (n - I)-dimensional projective space.

Grassmannian Manifolds. The set of I-dimensional complex subvec-

tor spaces of e n + l can be identified with the n-dimensional projective space,
and we have given it a complex structure. Now we do the same for the set
G k,n of k-dimensional subspaces of en. The idea is the following: If Vo c en
is a fixed element of Gk,n, then we choose an (n - k)-dimensional subspace
212 IV. Complex Manifolds

Wo c en such that Vo EB Wo = en. We are looking for a topology on Gk,n

such that the set of all k-dimensional subspaces V with V EB Wo = en is a
neighborhood of Vo in Gk,n'
But how to get complex coordinates? In the case G 1 ,n+1 = lP'n we consider, for
example, Vo = Ceo with eo = (1,0, ... ,0) and Wo = {(zo, ... , zn) : Zo = O}.
Then Vo EB Wo = e n+ 1 , and a vector z = (zo, ... , zn) =I- 0 generates a 1-
e
dimensional space V with V EB Wo = n + 1 if and only if Zo =I- O. Multiplication
by a nonzero complex scalar does not change the space V. Therefore, V is
uniquely determined by

Zo-1 . z = z = (1 , Zo-1 . -)
Zo-1 . ( Zo, -) z

The map f : V r-t Zo 1 . z E en gives the familiar local coordinates.

When we try to transfer this procedure to higher k, we use anoth~r viewpoint.
Every V with V 63 Wo = en+! has the form Graph( rpv) of a linear map
e
rpv : -+ en given by f(V) = rpv(l). If Vo c en is a k-dimensional subspace
and Vo EB Wo = en, then every other k-dimensional subspace V c en with
V EB Wo = en has the form Graph(rpv) for rpv E Homc(Vo, W o ). Fixing
bases of Vo and W o, the matrix of rpv with respect to these bases gives local
coordinates in Mk,n-k(C) ~ ek(n-k).
Now we will do this job in detail. An ordered k-tuple of linearly independent
vectors a1,"" ak E en can be combined in a matrix

with rk(A) = k. The set

St(k,n) := {A E Mk,n(C) : rk(A) = k}

is called the complex Stiefel manifold of type (k, n). Since its complement in
Mk,n(C) ~ e kn is an analytic set given by the vanishing of all (k x k) minors
of A, St(k,n) is an open set in Mk,n(C) and therefore a complex manifold.
The group GLk(C) acts on St(k, n} by multiplication from the left, and every
orbit of this group action represents exactly one k-dimensional subspace of
en. The topological quotient

is called the complex Grassmannian of type (k, n).

If, for example, Wo = {w = (W1, ... , w n ) : W1 = ... = Wk = O}, then a
matrix A E St(k, n} represents a basis of a k-dimensional space V with V EB
Wo = en if and only if A = (AoIA) with Ao E GLk(C) and A E Mk,n-k(C)'
 5. Quotients and Submanifolds 213

In this case V has the form Graph (c.py) for a linear map cpv : C k -+ C n - k .
Of course, V is uniquely represented by the matrix AOl . A = (EklAol . A),
and AOl . A is the matrix of cpv with respect to the standard bases.
Now we consider the set of multi-indices

Jk,n := {I = (i l , ... ,ik) E N k : 1::; i l < ... < ik ::; n}.

For any A E St(k, n) there exists an 1= (i l , ... , ik) E Jk,n such that

alii
ali" )
( E GLk(C).
A I := : :
aki l aki"
Then there is a permutation matrix PI E GLn(C) such that A . PI =
(AIIAI).
For fixed I we define

VI := {A E St(k,n) : detAI =I- O}.

We remark that (G· A)J = G· AI and (G:A)I = G· AI for G E GLk(C).

Therefore, VI is invariant under the action of GLk(C).
5.8 Lemma. Let 1rk,n : St(k, n) -+ Gk,n be the canonical projection. Then

7l"k.~(7l"k,n(VI)) = VI for every I E Jk,n.

PROOF: Let A E 7l"k.~(7l"k,n(VI)) be given. Then there is an A* E VI with

7l"k,n(A) = 1rk,n(A *). This means that there is a matrix G E GLk(C) with
A = G· A *. Since VI is invariant under the action of GLk(C), A lies in VI.
The converse inclusion is trivial. _
So VI is a saturated open subset of St(k, n), and UI := 1rk,n(VI) is open in
Gk,n. We leave it to the reader to show that Gk,n is a Hausdorff space.
If EI C C n is generated by ei 1 , ••• , eiA:' and FI C C n by the remaining
ej, then EI E9 FI = cn, and every k-dimensional subspace V C C n with
V E9 FI = C n is represented by a matrix A E VI. The uniquely determined
matrix All. AI E Mk,n-k(C) describes the linear map cpv : EI -+ Fl.
Therefore, we define the holomorphic map fI : VI -+ Mk,n-k(C) ~ ck(n-k)
by
1 -
fI(A):= AI . AI.
It is clear that fI is holomorphic, and since
214 IV. Complex Manifolds

fI respects the fibers of 1I"k,n' It remains to show that fI has maximal rank.
For that we construct sections:
Let A E VI be given. Then we define s : Mk,n-k(C) -+ VI by

s(B) := AI' (EkIB) . PII.

This is a holomorphic map with S(fI(A» = (AIIAI)' p I I = A and

flO s(B) = fI(A I · (EkIB) . PIl) = All. (AI' B) = B.

So fI is a submersion and Gk,n a complex manifold of dimension k(n - k).

Complex coordinates are given by <PI : UI -+ Mk,n-k(C) with

-t
Let Sk,n := {A E St(k, n) : A· A = Ek} be the set of orthonormal systems
of k vectors in en. Then Sk,n is a compact set, and 1I"k,n : Sk,n -+ Gk,n is
surjective. So Gk,n is compact.

Submanifolds and Normal Bundles. Let X be an n-dimensional

complex manifold. A holomorphic map f : Y -+ X is called an embedding
if there is a submanifold Z C X such that f induces a biholomorphic map
from Y onto Z. Every embedding is an immersion, but in general not every
(injective) immersion is an embedding. The following proposition has been
already mentioned in Section 1.
5.9 Proposition. Let f : Y -+ X be a holomorphic map between com-
plex manifolds. If Y is compact and f an injective immersion, then f is an
embedding.

PROOF: Every immersion defines a local embedding. To see this, we may

consider a holomorphic map f from a neighborhood V = V(O) c em into a
neighborhood U = U(O) c em x e n - m with f(O) = (0,0) and rk Jc(O) = m.
We write f = (fl' f2) and assume that already rk JCl (0) = m. Then for the
map F : V x e n - m -+ en with

F(z, w) := (fl(z), f2(z) + w)

we have det JF(O, 0) = det JCl (0) =f O. So there exist neighborhoods V* =
V*(O) C V and W = W(O, 0) c U such that F : V* x V* -+ W is biholomor-
phic. Since F(z, 0) = f(z), the image f(V*) = F(V* x {O}) is a submanifold
ofW.
We have proved that for every point Xo E Y there are neighborhoods
V = V(xo) C Y and W = W(J(xo» c X such that f(V) is a closed
submanifold of W. We have to show that there is a small open neighborhood
U = U(J(xo» c W such that f(Y) n U = f(V) n U. Suppose that there is
 5. Quotients and Submanifolds 215

a sequence Xn E Y - V with f(xn) -t f(xo). Since Y is compact, we can

assume that (xn) converges to an element x* E Y - V. Since f is injective,
f(x*) =I- f(xo)· But f(x n ) must converge to f(x*). This is impossible. -

If Y is a regular hypersurface in a complex manifold X, then there are two

line bundles associated to Y, namely the normal bundle N x (Y), which is
defined on Y, and the bundle [Y] on X. We will show that these bundles
coincide on Y.
Choose an open covering t1/ = (U.)tEl of Y in X such that there are complex
coordinates zt, ... , z~ for X in Ut with the following properties:
1. Y n U. = {x E U. : z~ (x) = O}.
2. zt, ... , z~_l are complex coordinates for Y.
We have already seen in Section 2 that the normal bundle N x (Y) is given
with respect to Ut n Y by the transition functions

h tl< ( zl,···,
I<
zn-l'
I< {) z~I< ( zl,···,
0) = {) I< I<
zn-l, 0) .
zn
On the other hand, the line bundle [Y] is defined by the transition functions
f ." .-
.- Znt / Zn· BU t Zn• ( Zl'···' Zn1<) -- f tl< (Zl'···' zn1<) . zn1<.lmp1·les
I< I< I<

{) z~ ( f (zl,···,zn_l'O.
0) =.1<
I< zl,···,zn_l'
I< I< I< I< )
{)
Zn
This yields the so-called first adjunction formula.
5.10 First adjunction formula.

[Yliy = Nx(Y) for every regular hypersurface Y c X.

The normal bundle Nx(Y) is naturally related to the canonical bundles Kx

and Ky. If gtl<, respectively Gtl< are the transition functions for T(Y), re-
spectively T(X), then

Therefore, det G.I< 0 j = det gtl< . h.l<, and det g;;/ = (det G;;.? 0 j) . ht'" which
is equivalent to the second adjunction formula:
5.11 Second adjunction formula.

Ky = j* Kx ® Nx(Y) for every regular hypersurface Y eX.

This formula can be generalized to higher codimensionsj see, for example,

[GriHa78].
216 IV. Complex Manifolds

Projective Algebraic Manifolds. If 11" : en+! - {O} --+ ]p>n is the

canonical projection and x E ]p>n, we define

e(x) := 11"-1 (x) U {O}.

This is a complex line through the origin in e n +1 , and we have e(11"(z)) = ez

e
for z E n + 1 - {O}.
e
A set X c n +1 is called a conical set or a cone if it is the union of a family
of complex lines through the origin. This means that

z EX===} ).z E X for ). E C.

If X is an arbitrary subset of ]p>n, then

X:= U e(x) = 11"-1 (X) U {O}

xEX

is a conical set.
5.12 Lemma. Let X c e n +1 be a conical set, f a holomorphic function
near the origin, and f = 2:::"=0 PII its expansion into homogeneous polynomi-
als. If there is an e > 0 such that fIB.co)nx == 0, then Pvlx = 0 for every
//.

PROOF: Let z f:. 0 be an arbitrary point of Bc(O) n X. Then

00

). H f().z) = LPII(Z).II
11=0

vanishes identically for 1).1 < 1. So Pv(z) = 0 for every //, and since X is
conical, Pvlx == 0 for every //. •
Now let F 1, . .. ,Fk be homogeneous polynomials in the variables zo, ... ,Zn.
Then the analytic set

is a cone. If we set X' := X - {O}, then the image X := 11"(X') C JP>n is the
set

In Ui = {(zo : ... : zn) : Zi f:. O}, we can define holomorphic functions All
by
fi,1I (Zo : ... : zn) := Fv ( -,
~~)... , - .
Zi Zi

Then X n Ui = {x E Ui : fi,l(x) = ... = fi,k(X) = O}, and consequently X

is an analytic set.
 5. Quotients and Submanifolds 217

Definition. An analytic set X c pn that is the zero set of finitely

many homogeneous polynomials is called a (projective) algebraic set. The
subsets X n Ui are called (affine) algebraic.
A complex manifold X is called projective algebraic if there is an N E N
and a holomorphic embedding j : X -+ pN such that j(X) is a regular
algebraic set.

5.13 Theorem of Chow. Every analytic set X in projective space is the

zero set of finitely many homogeneous polynomials Fl"'" Fs such that if
x E X is a regular point of codimension d, then rkz(Fl, ... , Fs) = d for every
z E 7r- I (x).

PROOF: If X c pn is a nonempty analytic set, then X' = 7r- 1 (X) is also

analytic. Since dimz(X' ) ~ 1 for all z E en + 1 _ {O}, by the extension theorem
of Remmert-Stein its closure X = X' U {O} is analytic in en+!.
By Theorem 6.5 in Chapter III we can find an open neighborhood U =
U(O) c en+! and finitely many holomorphic functions /I, ... ,1m on U with
N(/I, ... ,lm) = unx ~nd rkz(/I, ... ,lm) = d at any regular point z of
dimension n+ 1-d in U nX. Now we expand Ii into homogeneous polynomials
Pi,v' Then Pi,vlx == 0 for all i, v.
Let h c 0 0 S:! Hn+1 be the ideal that is generated by all Pi,v, V $ k,
i = 1, ... , m. Since
II C 12 C ... c 00
is an ascending chain of ideals in a noetherian ring, it must become station-
ary. Thus there are homogeneous polynomials F 1 , ... , Fs such that every Pi,v
is a finite linear combination of the Fu' But then every Ii is also a linear
combination of the Fu:
s

Ii = L ai,uFu'
u=1

It is clear that N (/I, ... , 1m) = N (F1 , ... , Fs) near the origin. But since X
is a cone, even X = N(Fl,"" Fs). Setting

A(z) = (ai,u(z) Ii==- 1, ... , m

a-I, ... ,s
) ,

we have

Ju" ... ,!m.)(Z) = A(z) . J(F" ... ,F.l(z) for z E X near O.

Therefore, d = rkz(h, ... , fm) $ rkz(FI, ... , Fs). If X is regular of codimen-

sion d at x, then X is also regular of codimension d at every Z E 7r- 1 (x),
218 IV. Complex Manifolds

because 7r is a submersion. It follows that rkz(F}, ... , Fs) cannot be greater

than d at these points. Since the rank is constant along 7r- 1 (x), it must be
equal to d at every Z E 7r- 1 (x). •

Examples
1. Let L c C n + 1 be a complex linear subspace of co dimension q. Then there
are linear forms 'PI, ... , 'Pq on Cn+ 1 such that

L = {z E Cn+l : 'Pl(Z) = ... = 'Pq(z) = OJ.

Since the linear forms are homogeneous polynomials of degree 1,

JlI'(L) := {(zo : ... : zn) E JlI'n : 'P/l(zo,.", zn) = 0 for J.L = 1, ... , q}

is a regular algebraic set. We call JlI'(L) a (projective) linear subspace. It

is isomorphic to IP'n-q.
2. We now show that the Grassmannian manifolds are projective algebraic.
Let 0 s: k s: n be given, and N := (~) - 1. We identify A. kcn with
CN+l and define the Plucker embedding pI : Gk,n --+ JlI'N as follows: If a
subspace V c C n has the basis {aI, ... , ad, then

It is clear that this is a well-defined injective map. To see that it is a

holomorphic immersion, we choose another descripton. As above, when
we introduced the Grassmannian, we use the set of multi-indices

To any I E fk,n there corresponds a permutation matrix PI E GLn(C)

such that for A E St(k,n) we get A· PI = (AIIAI)' We define p:
St(k, n) --+ CN+l - {OJ by

p(A):= (detAIII E fk,n)'

Then p(G . A) = det G . p(A), so p induces a map p : Gk,n --+ IP'N such
that the following diagram commutes:
p
St(k, n) ---7 C n+1 - {OJ

7rk,n 1 17r
p
Gk,n ---7 JlI'n
 5. Quotients and Submanifolds 219

al/\···/\ak= L detA/·eI,
IEJk,n

and therefore p = pI.

Let 'PI : UI ~ Mk,n-k(CC) be a complex coordinate system for Gk,n'
Then 'Pil(B) = 7rk,n(EkIB) . Pil) and

po 'Pil(B) = 7r 0 p(EkIB) ,Pi l ) = 7r(det(Ek IB) . Pil) J J E fk,n)'

Obviously, p is a holomorphic map.

Every matrix B E Mk,n-k(CC) can be written in the form

with bJL E CC k for J-L = k+ 1, ... ,no We have det(EkIB) 'Pi 1)/ = 1, and
if J = (I - {iv}) U {J-L} for some II E {I, ... , k} and J-L E {k + 1, ... , n},
then

and therefore det(EnIB) . Pi1)J = bvw So po 'Pil(B) contains the

components 1, bVJL for all II, J-L and some other components. It follows
that pI is an immersion. Since Gk,n is compact, pI is an embedding.

Projective Hypersurfaces. The simplest example of an analytic hy-

°
persurface in ]p>n is the hyperplane H = {zo = O}.
IfS E M n+1 (CC) is a symmetric matrix, then qs(z) := z·S·zt is a homogeneous
polynomial of degree 2. The hypersurface

Qs := ({zo : ... : zn) : qs(zo, ... , zn) = O}

is also called a hyperquadric. It follows easily from the classification of sym-
metric matrices that Qs ~ QT (biholomorphic) if and only if rk(S) = rk(T).
In particular, every quadric of rank n + I is biholomorphically equivalent to
the standard hyperquadric Qn-l = ({zo : ... : zn) : z5 + ... + z; = a}. Since

Qn-l n Uo = ({I : tl : ... : t n ) : ti + ... + t~ = -I},

Qn-l has no singularity in Uo. The same works in every Ui , so Qn-l is a
projective algebraic manifold.
Now consider the Grassmannian G 2 ,4 y ]p>5. Since dim(G 2,4) = 4, it is a
hypersurface. From multilinear algebra one knows that a 2-vector W E /\ 2CC 4
is decomposable (i.e., of the form W = a /\ b) if and only if w /\ w = O. So

G 2 ,4 = {7r(w) : W12W34 - Wl3W24 + Wl4W23 = o}

220 IV. Complex Manifolds

is a hyperquadric that is isomorphic to Q 4. This means that there is a I to 1

correspondence between the set of projective lines in p3 (which is the same as
the set G 2 ,4 of planes in ( 4 ) and the set of points of a 4-dimensional quadric.

5.14 Theorem. Every analytic hypersurface Z c pn is the zero set of a

single homogeneous polynomial.

PROOF: If Z c pn is an analytic hypersurface, then Z = rr-l(Z) U {O} C

en+! is also an analytic hypersurface. Therefore, there exists an open neigh-
borhood U = U(O) c e n + 1 , a point Z E un Z, and a holomorphic function
f : U --* e with "\7 fez) =1= 0 and znu c N(f). Making U smaller if necessary,
we can find a holomorphic function 9 dividing f such that Un Z = N(g).
Without loss of generality we may assume that U = U' X U" c e x en and

9 ( ZI, Z ') = zlk + ak-l ( Z ') Zlk-l + ... + ao (Z ')

is a Weierstrass polynomial.
If aK,1/ is the homogeneous part of a K of degree v, then

zf + qk-l,l (z')z;-l + ... + qO,k(Z')

Pk(ZI, z') :=

is the homogeneous part of g of degree k. Since Z is conical, it follows that

Pki z = O.
Now there is a dense open subset V C U" such that {t E U' : g(t,z') = O}
consists of exactly k points, for every z' E V. Since N (g) n U c N (Pk) n U
and deg(Pk) = k, we have 9 = Pk over V and then everywhere in U, by the
identity theorem. So Z = N (Pk). •
We can choose a polynomial P with minimal degree such that
Z = {(zo: ... : zn) E pn : p(zo, ... ,zn) = O}.
Then by the degree of Z we understand the number deg(p). For example,
deg(H) = 1 for any hyperplane, and deg(Q) = 2 for any hyperquadric.
Now, let Z C pn be an arbitrary hypersurface of degree k, defined by some
homogeneous polynomial P of degree k. Then
Z n Ui = {(zo : ... : zn) E Ui : zi k • p(zo, ... , zn) = O},
and the line bundle [Z] is given by the transition functions gij = (ZjjZi)k.
In particular, for every hyperplane H we have the same line bundle [H] with
transition functions Zj j Zi.
Definition. If He pn is a hyperplane, then the line bundle 0(1) :=
[H] is called the hyperplane bundle.
The kth tensor power of the hyperplane bundle is denoted by O(k).
 5. Quotients and Submanifolds 221

If Z C pn is a hypersurface of degree k, then [Z] = O(k).

A homogeneous polynomial F of degree k induces a global section SF E
f(lPn,O(k» by
(SF)i(ZO:"': zn):= z;kF(zo, ... ,zn) for Zi =I O.
In fact, (SF)i is a holomorphic function on Ui , with

Obviously,
{x E pn : SF(X) = O} = {(zo : ... : zn) : F(zo, ... , zn) = O}.
On the other hand, let S be an arbitrary global holomorphic section of O(k).
Then s can be represented by holomorphic functions Si : Ui --* e with
z; . Sj(zo : ... : zn) = zf . Si(ZO : ... : zn) on Ui n Uj ,
and we obtain a holomorphic function f : e n + 1 - {O} --* e with
fez) = zf . Si(7r(Z» on 7r- 1(Ui ).
There is a holomorphic continuation F of f on en + 1 with F(>.z) = >.k. F(z).
This means that"F is a homogeneous polynomial of degree k, with SF = s.
Consequently, we have the following result.
5.15 Proposition. For kEN the vector space r(pn, O(k» is isomorphic to
the space of h0rrl:0geneous polynomials of degree k in the variables Zo, ... , Zn·
Analytic hypersurfaces in pn are exactly the zero sets of global holomorphic
sections of O(k).
For any Zo E en + 1 - {O} and any kEN there is a homogeneous polynomial
F of degree k with F(zo) =I O. Therefore O(k) is generated by global sections.
The bundle 0(1) can be described in purely geometric terms. For this we
note that if
Po := (0: ... : 0 : 1) E pn+l,
then a projection 7ro : pn+l - {Po} --* pn is given by
7ro(zo : ... : Zn : Zn+l) := (zo : ... : Zn).
We obtain local tri vializations <Pi : 7r0" 1 (Ui ) --* Ui X e by
<Pi(ZO: ... : Zn: zn+!) := (zo: ... : zn), Z:;l),

with <p;l«zO : ... : zn), c) = (zo : ... : Zn : CZi). Obviously this gives the
transition functions ZjZ;l. If F is a linear form on e n +!, then the section SF
is given at x = 7r(z) = (zo : ... : zn) by
SF(X) = <p;l(X, z;l F(z» = (zo : ... : Zn : F(z».
222 IV. Complex Manifolds

The Euler Sequence. Here we discuss the tangent bundle of the pro-
jective space in more detail. Let 71' : en+! - {O} -+ JP'n be the canoni-
e
cal projection. Then for any point Z E n + 1 - {O} we have a linear map
e
i.pz : n+ 1 -+ T1I'(z) (lP'n) defined by

i.pz(w)[JJ := ! 10 (f 0 1r(z + tw».

The vector w can be interpreted as tangent vector 0(0) = a*.o«d/dt)o) with

aCt) := z + two It follows that
1r*.z(w)[fJ = (71' 0 a)*.o(d/dt)o[J) = (d/dt)o[f 071' 0 aJ = i.pz(w)[J].
Since 71' is a submersion and i.pz(z) = 0, the linear map i.pz is surjective with
Ker(i.pz) = ez.
If we use the local coordinates t" := Z" / Zi in Ui = {( Zo : Zl : ... : zn) I Zi f O},
then for any z with Zi =1= 0 it follows that

Thus a trivialization '¢i : T(JP'n)lu. -+ Ui X en is given by

'¢i(i.pz(W)):= (7I'(z), :/ao, ... ,lii, ... ,an»)',

with a" = W" - (WiZi-1 )z" for v f- i.

Let O(l)EIl(n+l) be the direct sum of n + 1 copies of the hyperplane bundle.
Recall that every linear form F on e n +1 defines a global section SF of 0(1) by
(s F)i = z;1 F. Then two vector bundle homomorphisms j : OlP" = Jpm X e -+
O(l)EIl(n+l) and q : O(l)EIl(n+l) -+ T(JP'n) can be defined by

j(x, c) := c· (szo(x), ... , Sz" (x))

and
q(SFo(1r(Z», ... , SF" (1r(z») := i.pz(Fo(z), ... , Fn(z»),
where Fo, . .. ,Fn are linear forms on e n+ 1 •
Using the canonical trivializations, j and q can be given over Ui by
j: «zo : ... : zn), c) H «zo : ... : zn), c· (zoz;l, ... , ZnZ;l))
and
q: «zo : ... : zn), (wo, ... , wn » H «zo : ... : zn), (ao,· .. , a;, ... ,an»),
with aLI = W" - (WiZ;l)Z". This shows that both of these homomorphisms
have constant rank. Since there is always an index i such that S Zi (x) i- 0, j
is injective. Since 0(1) is globally generated, q is well defined and surjective.
 5. Quotients and Submanifolds 223

Definition. A sequence 0 -+ V' .4 V 4 V" -+ 0 of vector bundles is

called exact if the following hold:
1. j and q have constant rank.
2. j is injective and q is surjective.
3. Im(j) = Ker(q).

5.16 Theorem. We have an exact sequence of vector bundles

The sequence is called the Euler sequence.

PROOF: It is clear that q 0 j(n(z), c) = <pz(c· z) = O. If, on the other hand,
x = n(z) and q(SFo(x), ... , SF" (x)) = 0, then (Fo(z), ... , Fn(z» must be
a multiple of z. So there exists acE C; such that (SFo(X), ... ,SF,,(X» =
c· (szo(x), ... , sz" (x». •

Rational Functions. We consider some connections with complex alge-

braic geometry.
A meromorphic function m on lPn is called rational if m = 0, or if there are
homogeneous polynomials F and G of the same degree such that F =I 0 and

F(zo, ... ,zn)

m (Zo : ... : Zn ) = G( ).
Zo,··· ,Zn

5.17 Theorem. Every meromorphic junction on lPn is rational.

PROOF: If mE .A"(lPn), then its divisor has a finite representation

div(m) =L ni . Zi ,
i

with ni E Z and irreducible hypersurfaces Zi C pn. By the theorem of Chow

every Zi is the zero set of a homogeneous polynomial Fi of degree d i ~ 1.
Then F := TIi Ft; is a rational function on c;n+l that is homogeneous of
degree d := Li nidi E Z.
Since div(m 0 n) = div(F) on c;n+l - {O}, the function f := (m 0 n) . F-l is
holomorphic there without zeros. It has a holomorphic extension to c;n+!1
1
with 1(0) =I 0, because cannot have isolated zeros.
For c E C·, the equation l(c. z) = c- d ·l(z) is valid on an open subset of
cn+!, and then (by the identity theorem) everywhere on cn+l. Thus d ~ 0
224 IV. Complex Manifolds

and c- d = 1. From f(c. z) = j(z) it follows that fez) is a constant wo, and
m 07r(z) = Wo . F(z) a rational function. _

A rational function on a submanifold X C IPn is the restriction mix of a

rational function m on ]pm. We have already seen that every polynomial p
on Uo ~ en can be extended to a rational function m on IPn. Therefore, on
every algebraic manifold there is a large number of rational functions.
If A c IPn is a projective algebraic set, then Ai := An Ui is called an affine
algebraic set. A complex manifold is called an affine algebraic manifold if it is
biholomorphically equivalent to an affine algebraic set. A regular function on
an affine algebraic manifold j : X '-t Ui is a holomorphic function f : X -+ e
such that there exists a polynomial p on Ui ~ en with f = p 0 j. It can be
shown that a rational function on an affine algebraic manifold is always a
quotient of regular functions.
In algebraic geometry there is a more general definition for regular functions,
which coincides with our notation in the case of affine algebraic manifolds.
On projective algebraic manifolds all regular functions are constant, whereas
there are many rational functions. In the affine case the field of rational
functions is exactly the quotient field of the ring of regular functions.
If Z c IP n is a hypersurface of degree k, then X := IP n - Z is an affine
algebraic manifold. We can see this as follows:
Let I be the set of multi-indices v = (vo, . .. , v n ) with Vo + ... + Vn = k. Then
#(1) = (ntk) is the number of monomials z" = z~o ... z~n, v E I. We set
N := #(1) - 1 and define the Veronese map Vk,n : IPn -+ IPN by

One can show that Vk,n is an embedding, and so its image is Vk,n = Vk,n(lP n )
an algebraic sub manifold of IPN. If p is a homogeneous polynomial of degree
k with zero set Z, then there are complex numbers a", v E I, such that
p = E"EI a"z". It follows that

Vk,n(Z)=Vk,nn{(W")"EI: La"w,,=O}
"EI
is the intersection of Vk,n with a hyperplane H C IPN. Therefore, IPn - Z ~
vk,n(lPn - Z) = Vk,n n (IPN - H) is affine algebraic.

Exercises
1. Show that M := {( Zo : Zl : Z2 : Z3) E 1P3 : z8 + ... + z~ = O} is a
projective manifold. Consider the group G := {gm : 0::; m ::; 4}, where
9 : 1P3 -+ ]p3 is defined by
 5. Quotients and Submanifolds 225

g(ZO : Zl : Z2 : Z3) := (ezo : e2Z1 : e3Z2 : e4Z3), e = e21ri / 5 •

Prove that X := MjG is a complex manifold (called a Godeaux sur/ace).

2. Let F : C3 ~ C 3 be defined by

Set G = {Fm : mE Z} and prove that X := (Cx(C 2 -{O}»jG is a com-

pact complex manifold. Consider the holomorphic map 7r : X ~ C that
is induced by the projection onto the first component. Then X t := 7r- I (t)
is a 2-dimensional complex manifold. Show that X t is biholomorphically
equivalent to Xl for every t -:f:. 0, but that there is no biholomorphic map
between Xo and Xl' Show that Xo is a Hopf manifold.
3. For W}, ... ,W2n E en define 0 := (wi, ... ,win) E M n,2n(C) and prove
that
WI, ... ,W2n linearly independent -¢=} det (~) -:f:. O.

Let Tn be the torus defined by the lattice r := Zwl + ... + Zw2n . Prove
that Tn is biholomorphically equivalent to Tn' if and only if there are
matrices G E GLn(C) and M E GL 2n (Z) such that 0' = G· o· M.
4. The Segre map Un : pI X pn ~ p2n+l is defined by

Un«XO : Xl), (yo: ... : Yn)) := (XOYO : ... : XOYn : XIYO : ••• : XIYn) .
Show that En := Un(pl X pn) is a complex manifold and Un an injective
holomorphic immersion.
5. Show that for n ~ 2, two irreducible hypersurfaces in pn always have a
nonempty intersection.
6. Decompose

into irreducible components.

7. Let d> 0 and n > O. Prove the following theorem of Bertini. If Hn+l(d)
is the vector space of homogeneous polynomials of degree d in zo •...• Zn.
then there is a dense open subset U C Hn+l(d) such that

N(F) = {(zo : ... : zn) E pn : F(zo, ... , zn) = O}

. is a projective manifold for every FE U.

8. Use the notation from the introduction to Grassmannian manifolds. For
IE .Fk,n set VI:= {A E St(k,n) : detAI #- O} and Ur:= 7rk,n(Vr) C
Gk,n' Then
glJ(7rk,n(A» := A/I ·.AJ on Vr n VJ
give the transition functions of a vector bundle Uk on Gk,n' Show that
there are n independent global holomorphic sections in Uk.
226 IV. Complex Manifolds

9. Show that HI (]p>n, 0) = 0 for n 2:: 1 and HI(]p>l,O(k» = 0 for k 2:: o.

Prove that there is an exact sequence of vector spaces

for n 2:: 2 and k 2:: 1. Conclude that HI (]p>n ,

k 2:: o.
°(
k » = 0 for n 2:: 1 and

10. Let X be a compact complex manifold and 1f : V -+ X a holomorphic

vector bundle of rank r 2:: 2. Show that there is a holomorphic fiber
bundle P -+ X with typical fiber ]p>'"-l and a line bundle L -+ P such
that Px = ]P>(Vx ) and LIp:. ~ Op",(I).
11. Prove that for every automorphism f of pn (Le., every biholomorphic
map f : ]p>n -+ ]p>n) there is an A E GLn+I(1C) with f(1f(z)) = 1f(z· At).
This is formulated as Aut(]p>n) = PGL n+1(1C) := GL n+1(IC)/C".
12. Let r = WI + W2 be a lattice in C and p the Weierstrass function
for this lattice. Show that the map Z f-+ (1 : p(z) : p'(z» induces an
embedding <p of the torus T = c/r into ]p>2. Determine the equation of
the image <peT) in ]p>2. Determine transition functions for <p"0(1).

6. Branched Riemann Domains

Branched Analytic Coverings. Let f : X -+ Y be a continuous map
between Hausdorff spaces. Then the image of compact sets is compact, and
the preimage of closed sets is closed.

Definition. The map f is called closed if the image of closed subsets

of X is closed in Y. It is called proper if the preimage of compact subsets
of Y is compact.

6.1 Proposition.
1. If f : X -+ Y is closed and A C X a closed subset, then flA : A -+ Y is
closed..
2. Iff is closed, y a point of f(X), and U an open neighborhood of the fiber
f-l(y), then there is an open neighborhood W = W(y) C Y such that
f-I(W) c U.
3. Let X and Y be additionally locally compact.
(a) If every point y E Y has a neighborhood W such that f-I(W) is
compact, then f is proper.
(b) If f is proper, then f is closed.
(c) If f is closed and every fiber f-l(y) compact, then f is proper.

PROOF: (1) is trivial.

 6. Branched Riemann Domains 227
;,

(2) Let I be closed and U be an open neighborhood of some nonempty fiber

I-I(y). Then W := Y - I(X - U) is an open subset of Y. If y were not
in W, then there would be a point x E I-I(y) - U. This is impossible, and
therefore W is an open neighborhood of y with 1-1 (W) cU.
(3a) Let Y be locally compact and KeY compact. It follows from the
criterion that there are finitely many sets WI"'" Wm such that I-I(K) is
contained in the compact set I-I(WI ) U ... U I-I(Wm ). Since l-l(K) is
closed, it must itself be compact.
(3b) Let I be proper and A c X be a closed set. For every y E I(A) there is a
sequence of points Xv E A with limv I(x v ) = y. The set N := {y} U {f(xv) :
v E N} is compact (using the local compactness of Y), and consequently
I-I(N) is compact. Thus there is a subsequence (xv,,) converging to some
point x E I-I(N) n A. It follows that y= I(x) E I(A).
(3c) Let I be closed and assume that every fiber is compact. For y E Y
choose a compact neighborhood U of I-I(y). This is possible, since X is
locally compact. By (2) there exists an open neighborhood W of yin Y with
I-I(W) c U. Then l-l(W) c U as well. As a closed subset of a compact
set it is itself compact, and from (3a) it follows that I is proper. _

Definition. A continuous map I : X ~ Y between locally compact

Hausdorff spaces is called finite if it is closed and if each fiber has only
finitely many elements.

Obviously, every finite map is proper. Conversely, if a proper map has only
discrete fibers, then it must be finite.
Definition. A holomorphic map 7r : X ~ Y between n-dimensional
complex manifolds is called a branched (analytic) covering if the following
hold:
1. The map 7r is open, finite, and surjective.
2. There is a closed subset DeY with the following properties:
(a) For every y E Y there is an open neighborhood U = U(y) c Y
and a nowhere dense analytic subset A c U with D n U c A.
(b) 7r: X - 7r- l (D) ~ Y - D is locally biholomorphic.
The set D is called the critical locus. A point x E X is called a branch
point if 7r is not locally biholomorphic at x. The set B of branch points is
called the branch locus. The covering is called unbranched if B is empty.

The branch locus B is nowhere dense in X.

Now we consider a domain G c en and d pseudopolynomials Wi (Wi, z) over

G and define
228 IV. Complex Manifolds

~ d .
A:={(w,Z)EC xG: Wi(Wi,z)=Oforz=l, ... ,d}.
Let 7r : A -+ G be the restriction of the projection pr2 : Cd x G -+ G.
6.2 Proposition. The map 7r is surjective, finite, and open.

PROOF: (1) The surjectivity follows from the fact that, for fixed Z E G,
each Wi (Wi, z) has at least one zero.

(2) Let MeA be a closed subset and Zo a point of 7r(M). Then there
is a sequence (wv,zv) in M with Zv -+ Zoo Since the coefficients of the Wi
are bounded near zo, the components of the zeros wv are also bounded.
Consequently, there is a subsequence (wVI') of (w v) that converges in Cd to
some woo It is clear that (wo, zo) lies in M. Thus Zo belongs to 7r(M), and 7r
is closed.
Since 7r has finite fibers, it is a finite (and in particular a proper) map.

(3) It remains to show that 7r is open. Let (wo, zo) E A be an arbitrary point.
If V(wo) C Cd and W(zo) c G are open neighborhoods, we have to find an
open neighborhood W'(zo) C 7r(A n (V x W)).
By Hensel's lemma there are pseudopolynomials wi and wi* such that

Define

A:= {(w,z) E V x W : W;(Wl'Z) = ... = w.i(Wd'Z) = O}.

Then A c An (V x W), and (7rIA)-l(zO) = {(wo,zo)}. Since 7rIA is a closed

map, there is a neighborhood W'(zo) C W such that A n (Cd x W') =
(7rIA)-l(W') c An(V x W). Since also 7rIA: A -+ W is surjective, it follows
that
W' = 7r(A n (Cd x W')) c 7r(A n (V x W)).
Thus 7r is open. •
Remark. If X c A is an irreducible component that is everywhere regular
of dimension n, then the same proof shows that 7rlx is also surjective, finite,
and open. If D eGis the union discriminant set for A, then 7r : X -+ G is a
branched analytic covering with critical locus D.

Branched Domains. A continuous map f : X -+ Y between Hausdorff

spaces is called discrete at x E X if the fiber f-l(f(X)) is a discrete subset
of X. The map f is called discrete if it is discrete at every point x E X.
A holomorphic map f : X -+ Y between complex manifolds is discrete at
Xo if and only if there is an open neighborhood U = U(xo) CC X with
f(xo) f/- f(8U).
 6. Branched Riemann Domains 229

Now let M be an n-dimensional complex manifold.

Definition. A branched domain over M is a pair (X, 11") with the fol-
lowing properties:
1. X is an n-dimensional connected complex manifold.
2. 71": X -+ M is a discrete open holomorphic mapping.

Let 11" : X -+ M be a branched domain and Xo E X an arbitrary point.

We can choose a coordinate neighborhood B of 1I"(xo) in M such that B is
biholomorphically equivalent to a ball in en and 1I"(xo) = O. Since 11" is discrete
and open, there is a connected open neighborhood U = U(xo) CC X such
that 7I"(U) C B is open and 71"-1(0) n U = {xo}. We assume that 1I"(U) = B.
Define j : U -+ U x B by j(x) := (x,1I"(x». Then j is a holomorphic embed-
ding, and
A := j(U) = {(x, z) E U x B : z = 1I"(x)}
is a regular analytic set in U x B. We have a factorization 11" = 7f 0 j, with
7f := pr2lA. Let WI,." ,wn be complex coordinates near Xo in U. It follows
from the results of Section III.6 that over a small neighborhood of 0 in B there
are pseudopolynomials Wi(Wi,Z) such that A is an irreducible component of
the embedded analytic set

A = ((w,z) : Wi(Wi,Z) = 0 for i = 1, ... ,n}.

If DeB is the union of the discriminant sets of the Wi, then A (and therefore
also A) is unbranched over the complement of D. Therefore, if 11' : X. -+ M is
a branched domain, then for every x E X there is a neighborhood U (x) C X
and a neighborhood W(f(x» c M with 1I'(U) = W such that 1I"Iu : U -+ W
is a branched analytic covering.
If M = en or M = IP'n, then a branched domain over M is also called a
branched Riemann domain.

Definition. Two branched domains (Xb 1l"I) (over M) and (X2,1I"2)

(over N) are called equivalent if there are holomorphic maps <p : M -+ N
and rp : Xl -+ X 2 such that the following diagram commutes:

M ~ N

Torsion Points. As an example we consider a polydisk P = pn(o, r) C

en, a positive integer b, and the set

Xo := {(z, W) E P x e : wb - Z1 = OJ,
230 IV. Complex Manifolds

together with the projection 11"0 : (z, W) H z. Then Xo is a regular analytic

hypersurface in P x e and consequently an n-dimensional complex manifold,
e
which is the Cartesian product of a polydisk pI C n - l and a neighborhood
of the origin in the Riemann surface of W = {/Z. It is unbranched outside
{z : Zl = O}.

Definition. Let (X, 11") be a branched domain over some manifold M.

A point Xo E X is called a torsion point (or winding point) of order b if
there is an open neighborhood U = U(xo) c X such that 11" : U -+ 1I"(U)
is equivalent to 7fo : Xo -+ P. In this case we say that 7flu is a winding
covering.

We consider the local case, i.e., a branched analytic covering 7f : X -+ Q over

some polydisk Q c en around the origin, and we assume that it is branched
of order b over 0 and unbranched outside {z E Q : Zl = O}. Then X' :=
X -7f- I ( {Zl = O}) is still connected, and the covering 7f : X -+ Q is equivalent
to 7fo : Xo -+ P. For a proof see [GrRe58], §2.5, Hilfssatz 2. One shows that the
two coverings outside the branching locus have the same fundamental group.
Therefore, they must be equivalent there. Using the theorem of Remmert-
Stein it follows that X and Xo are equivalent.
The following is now immediate.
6.3 Proposition. Let (X,7f) be a branched analytic covering over M, with
branching locus D. If Xo E X is a point such that Zo = 7f(xo) is a regular point
of D, then there are neigborhoods W = W(xo) C X and U = U(zo) c M
such that 7f : W -+ U is a winding covering.
The proof follows simply from the fact that one can find coordinates such
that locally the situation above is at hand, and then X is locally equivalent
to Xo.

Concrete Riemann Surfaces. A concrete Riemann surface is a

branched Riemann domain 7f : X -+ e
or 7f : X -+ JlDI. It is an abstract
connected Riemann surface, and in the latter case it may be compact. The
infinitely sheeted Riemann surface of w = log Z is not a concrete Riemann
surface in our sense.
Here is a method for constructing concrete Riemann surfaces. Let X be a
Hausdorff space, M = e or M = JlDI, and 11" : X -+ M a continuous mapping
such that for every Xo E X there is an open neighborhood U = U(xo) C X,
a domain Vee (in the case M = e), respectively a domain V in an affine
part ~ e of M (in the case M = JlDI) and a topological map 'IjJ : V -+ U with
the following properties:
I. 11" 0 'IjJ : V -+ IC is holomorphic.
2. (7f 0 'IjJ)' does not vanish identically.
 6. Branched Riemann Domains 231

The map 'IjJ is called a local unijormization.

Obviously, (U, 'IjJ-l) is a complex coordinate system. We have to show that
two such systems (Ul, 'ljJ1l) and (U2' 'IjJ"2l) are holomorphically compatible.
We consider only the case M = C. The map

'l/J := 'l/J 1 l 0 'l/J2 : 'l/J"2l(Ul n U2) --7 'l/J1l(U1 n U2)

is a homeomorphism. We denote the set 'IjJ); 1 (Ul nU2) by V;. Then Dl := {t E
Vi* : (7r 0 'ljJl)'(t) = O} is discrete in Vt, and therefore D2 := 'l/J-l(Dt} C V2*
is also discrete. Let to be a point of V2* - D 2. Then

(7r 0 'ljJd('IjJ(to» f= o.
Therefore, there are open neighborhoods U = U('IjJ(to» c Vi* and W =
W(zo) C C (for Zo := 7r('l/J2(tO))) such that 7rO'l/Jl : U --7 W is biholomorphic.
Then 7r1'lPl(u) : 'ljJl(U) --7 W is a homeomorphism, and

'l/J = 'ljJll 0 (7rI"'l(U»-l 0 7r 0 'ljJ2 = (7r 0 'ljJl)-l 0 (7r 0 'l/J2), on V := 'IjJ-l(U).

Since V is an open neighborhood of to, we see that 'IjJ is holomorphic on
V2* -D 2· As a continuous map it is bounded at the points of D 2, and therefore
it must be holomorphic everywhere in V2'".
From the above it follows that X is a Riemann surface, and obviously 7r is a
holomorphic map. We still want to see that it is a finite branched covering.
If Xo E X, then there is an open neighborhood U = U(xo) C X and a local
uniformization 'IjJ : V --7 U. We define <I> : U --7 V X 7r(U) C C2 by

<I>(x) := ('l/J-l(X), 7r(x».

Then <I>(U) = {(z, w) E V X 7r(U) : 7ro'IjJ(z) = w}. Restricting the projection

(z, w) H W, we obtain a branched covering p : iJ!(U) --7 7r(U) that is equiva-
lent to 7rlu : U --7 7r(U). Let be j := 7r 0 'l/J and Zo := 'IjJ-l(XO). If !'(zo) f= 0,
then p is unbranched. If f' has an isolated zero of order k at Zo, then

fez) = 7r(xo) + (z - zo)k . h(z),

where h is a holomorphic function with h(zo) f= O. It is clear that then p is

branched of order k, equivalent to the Riemann surface of if'i.

Hyperelliptic Riemann Surfaces. For g ;::: 2 choose 2g + 2 different

points Zl, Z2, .. . , Z2g+1, Z2g+2 in Xo := pl. We assume that they are all real,
in natural order, and not equal to 00.
Now we take two copies X l ,X2 of pl regarded as lying above Xo and cut
them along the lines fi between Z2i+1 and Z2i+2, i = 0, ... ,g. We define

X;: := {t E X.x : Im(t) ::; O} and xt:= {t E X.x : Im(t) > O}

232 IV. Complex Manifolds

~ ~
_____:_1____t_O__________•__________•____Z_2_:+_1____ ~
tg____• 2_:_+-J2

Figure IV.2. The hyperelliptic surface

and glue xi to Xl along the cuts ii, and similarly Xi to Xt.

We obtain a topological (Hausdorff) space X that lies as a branched cover-
ing of order 2 over ]pI (see Figure N.2). It is branched only at the points
Zl> .•• , Z2g+2, and at each of these points it looks exactly like the Riemann
surface of viz over the origin. Therefore, it is clear that X, together with the
canonical projection 11" : X ~ ]pI, is a concrete Riemann surface. It is called
a hyperelliptic surface. 4
A curve going in Xl from Z2i+1 to Z2i+2 and then in Xi back to Z2i+1
defines a cycle C: in X, i = 0, ... , g. Similarly a curve in Xl which goes
from Z2i+2 to Z2i+3 and then in X 2 back to Z2i+2 gives a cycle C:, in X,
i = 0, ... ,9 - 1. Finally we define C;
by going in Xl from Z2g+2 thro\.l.gh
00 to Zl and then in X 2 back to Z2g+2' So HI (X, Z) is generated by the
29+2 cycles C6, ... , C~, C~, ... ,c;. But C' := c6+'" +C~ is the boundary
of X2, and C" := C~ + ... + C; is the boundary of xt U Xi. Therefore,
HI(X,Z) = ZI(X,Z)jBl(X,Z) ~ Z2g (and Ho(X,Z) = Z, because X is
connected). It follows that HI(X,Z) ~ Z29. The number
1 1
9 = 9(X) := "2bl(X) = "2 rk(HI(X,Z))

is called the genus of X. Recalling that ]pI is the Riemann sphere, we replace
X 2 by the complex conjugate X 2 and glue Xl and X 2 inserting tubes into
the cuts ii' Thus we realize X as a sphere with 9 handles.
Now we give a more analytical description of the hyperelliptic surface. Define
29+2
X' := {(s, t) E C2 : s2 = II (t - Zi)},
i=O
4 In the case 9 = 1 we would get an elliptic surface that is isomorphic to a 1-
dimensional complex torus.
 6. Branched Riemann Domains 233

and 1r' : X' -t C by 1r(s, t) := t. This is a branched covering with branching

locus D = {Zl,"" Z2g+2}. For F(t) := I1;!62(t-zd, the function s = JF(t)
is well defined on X'. Over {t E C : It I > R}, X' consists of two disjoint
punctured disks. Therefore, we can complete X' to a compact Riemann sur-
face X by filling the points at infinity, i.e., the punctures. The map 1r' can
be extended continuously to a map 1r : X -t pl. It follows by the Riemann
extension theorem that 1r is holomorphic, and X is a branched covering of
order 2 over pI with branching locus D.
Remark. The Riemann surface X is not the completion of X' in p2.
The projective analytic set X', which has {s2 = F(t)} as affine part, has a
singularity at infinity.
The functions t and s = JF(t) extend to meromorphic functions f and 9 on
X that have poles of order 1, respectively 9 + 1, over 00.
We now calculate Hl(X,O). For this the following is an essential tool.
6.4 Serre's duality theorem. If X is a compact Riemann surface and V
an analytic vector bundle over X, then Hl(X, V) and HO(X, Kx 0 V') are
finite-dimensional vector spaces of equal dimension.

PROOF: See [Nar92J, page 47. •

In the case V = Ox = X x C we have Hl(X, 0) ~ HO(X,Kx). So we need
some remarks about the canonical bundle Kx.
1. Let U = (U.).EI be an open covering of X such that there are complex
coordinates 'P. : U. -t C. We denote the complex coordinate on U. by t •.
Then the line bundle K x is given by the transition functions
dtl<
g.1< = di'
•
A holomorphic 1-form on X is a global section w E r(X, T(XY) =
HO(X, Kx). We can write wlu, = w.dt •. Then w.dt. = wl<dt", on U.I<'
and therefore w. = g.1< . WI<'
2. On pI we have two systems of complex coordinates, namely, to := zr/ Zo
on Uo = {(zo : Zl) : Zo =f. O} and tr := zolzl on U1 . We denote to by t
and -tl by s. Then on U0 1 we have s = -lit. Using sand t as complex
coordinates, the canonical bundle on pI is defined by

gOl = ~; = ! (-~) = t~ = C~f·

This is also the transition function of the bundle [DJ associated to the
divisor D = -2· 00. Therefore,
HO(pl, Kpl) ~ HO(pl, [DJ)
{m E M(pl)* : ordoo(m) 2: 2 and ordp(m) 2: 0 otherwise}.
234 IV. Complex Manifolds

But such a function m is holomorphic and vanishes at 00, and conse-

quently is identically O. Thus HO(PI, Kpl) = O.
3. Let X be a compact Riemann surface, and WI,W2 E HO(X,Kx) two
holomorphic I-forms. In local coordinates w>.lu. = w~;'ldtL' for>. = 1,2.
Then /Iu. := W?l/w?l defines a global meromorphic function / on X
such that W2 = /. WI.
Now we return to the hyperelliptic surface X. A holomorphic map j : X --+ X
is defined by (s, t) 1---+ (-s, t). Clearly, j permutes the two leaves, and P = id.
This map is called the hyperelliptic involution, and it induces a linear map
j* : HO(X, Kx) --+ HO(X, Kx) with (j*)2 = id. We get a decomposition of
HO(X, Kx) into the eigenspaces associated with the eigenvalues ±l. But if
j*w = w, then w = 11"* (cp), where cp is a holomorphic I-form on pl. Since
H°(JP>I, KII>1) = 0, it follows that w = O. Thus j*w = -w for every w E
HO(X, Kx).
The equation S2 = F(t) implies that 2s ds = F'(t) dt. If set) = 0, then t = Zk
for some k. But F'(Zk) = I1i#(Zk - Zi) '# O. Therefore,

dt 2ds
Wo := --; = F'(t)

is a holomorphic I-form on X'. Near 00 we have set) ~ t9+l and therefore

Wo ~ (I/t)9- l d(l/t). This shows that Wo is a holomorphic I-form on X that
has a zero of order g -1 over 00. The same argument shows that the I-forms
Wv := tV . (dt/s) are holomorphic on X, for v = 1, ... ,g - l.

Now let w be an arbitrary element of HO(X, Kx). Then there is a meromor-

phic function / on X with w = / .Wo. Since Wo has no zero over pI - {00 }, / is
holomorphic there. And since j*w = -wand j*wo = -Wo, we have j* / = /.
This means that / = h 0 11" for some meromorphic function h on pI that is
holomorphic outside 00. Consequently, h is a polynomial, say of degree d. It
has d zeros in C and a pole of order d at 00. It follows that / also has a pole
of order d at every point above 00. Since w = / .Wo is holomorphic, d ~ g - 1.
Therefore, w is a linear combination of the I-forms
dt dt
t-, ... ,
s s
It follows that dime HO(X, Kx) = g.
6.5 Proposition. I/ X is a hyperelliptic Riemann sur/ace, then the genus
g = g(X) is equal to dimHO(X, Kx).
We will later see that the same result is true for arbitrary compact Riemann
surfaces.
 7. Modifications and Toric Closures 235

Exercises
1. Let rr : X -+ M be a branched analytic covering with critical locus
D eM. Show that if there is a point Yo E D such that D has codimension
2 at Yo, then there is no branch point in rr-1(yo).
2. Let rr : X -+ M, and rr' : X' -+ M, be branched analytic coverings
and D, respectively D', the critical locus. Prove that if X - rr-1(D) is
equivalent to X' - (rr')-l(D'), then X is equivalent to X'.
3. Consider the holomorphic map F : C 2 -+ C3 defined by

(w, Zl, Z2) = F(tl' t2) = (tl . t2, ti, t~).

It is the uniformization of a 2-dimensional analytic set A C C3 . Show
that A - {O} is a branched domain over C 2 - {O}. What are the torsion
points? Describe A as the zero set of a pseudopolynomial w(w, z). Is A
also a branched domain in our sense?
4. Let F be a homogeneous polynomial of degree k and

X = {(zo: ... : Zn : zn+d E Ipm+l : Z~+l - F(zo, ... ,zn) = o}.

Assume that D := {(zo : '" : zn) E ]p>n : F(zo, ... , zn) = o} is regular
of co dimension 1 and prove that the canonical projection rr : X -+ ]p>n
is a branched Riemann domain. Determine the torsion points and their
orders.
5. Define cp : ]p>l -+ ]p>n by cp(zo : Zl) := (zo : Z~-lZl : ... : zf) and prove
that the image C := cp(]P>l) is a regular complex curve in ]p>n. Define

X:= {(p,H) E C x Gn,n+l : p E H},

where an element HE Gn,n+l is identified with a projective hyperplane

in ]p>n. Show that the projection rr = pr2 : X -+ G n,n+1 is an n-sheeted
branched covering.
6. Let X c ]p>n be a d-dimensional connected complex submanifold. Prove
that there is an (n - d - 1)-dimensional complex projective plane E c ]p>n
not intersecting X, a d-dimensional complex projective plane F C ]p>n
with En F = 0, and a canonical projection rr : ]p>n - E -+ F such that
rrlx : X -+ F is a branched domain.

7. Modifications and Toric Closures

Proper Modifications. Assume that X is an n-dimensional connected
complex manifold and A c X a compact submanifold. Is it possible to cut
out A and replace it by another submanifold A' such that X' = (X - A) uA'
is again a complex manifold? In general, the answer is no, but sometimes such
complex surgery is possible. Then we call the new manifold X' a modification
of X.
236 IV. Complex Manifolds

Definition. Let f : X -+ Y be a proper surjective holomorphic map

between two n-dimensional connected complex manifolds. The map f is
called a (proper) modification of Y into X if there are nowhere dense
analytic subsets E c X and S c Y such that the following hold:
1. f(E) c S.
2. f maps X - E biholomorphically onto Y - S.
3. Every fiber f-l(y), yES, consists of more than one point.
The set S is called the center of the modification and E = f- l (S) the
exceptional set.

Assume that we have a proper modification f :X -+ Y with center Sand

define

X := {(y,x) E Y x X : y = f(x)} and 7r:= prllx : X -+ Y.

Then X is an n-dimensional connected closed submanifold of Y x X. If KeY

is compact, then 7r- 1 (K) = (K xX)nX is a closed subset of K x f-l(K) and
therefore compact; 7r : X - 7r- 1 (S) -+ Y - S is biholomorphic, with inverse
y H (y,f-l(y)); and for Yo E S we have 7r- 1 (yo) = {(Yo,x) : f(x) = Yo} ~
f-l(yO). This is a set with more than one element.
Therefore, we can generalize the notion of a proper modification in the fol-
lowing way.
Let X and Y be two connected complex manifolds. A generalized (proper)
modification of Y in X with center S is given by an irreducible analytic subset
X of the Cartesian product Y x X and a nowhere dense analytic subset S c Y
such that
1. 7r:= prllx : X -+ Y is proper.
2. 7r- l (S) is nowhere dense in X.
3. X - 7r- 1 (S) is a complex manifold.
4. 7r maps X - 7r- 1 (S) biholomorphically onto Y - S.
5. X has more than one point over each point of S.
The center S is sometimes also called the set of indeterminacy.
If X is a manifold, then 7r : X -+ X is an ordinary modification of X into X
with center S.
7.1 Proposition. 7r- l (S) has codimension 1 in X.

PROOF: We consider only the case where X is regular. A proof of the general
case can be found in [GrRe55].
Assume that Xo E 7r- l (S) is isolated in the fiber of 7r over Yo := 7r(xo). Then
there is a neighborhood U of Xo in Y x X such that U n X is a branched
covering over Y. Since Yo lies in S, there must be at least one additional
 7. Modifications and Toric Closures 237

point Xl E X in the fiber over Yo. But then in any small neighborhood
of Yo there are points y E Y - 8 with several points in 11'-1 (y). This is a
contradiction. Therefore, dimxo 11'-1 (1I'(xo)) > O. This is possible if and only
if rk xo (1I') < n := dim(Y), and this shows that 11'-1(8) in local coordinates is
given as the zero set of det J rr • Therefore, it has co dimension 1. •

Blowing Up. Let U = U(O) be a small convex neighborhood around the

e
origin in n + 1 • We want to replace the origin in U by an n-dimensional
complex projective space. If 11' : en+!_ {O} -+ pn is the canonical projection,
then every line Cv through the origin determines an element X = 1I'(v) in the
projective space, and X determines the line lex) = 11'-1 (x) U {O} such that
ev = l(1I'(v)). Now we insert pn in such a way that we reach the point x by
approaching the origin along lex).
We define
X := {(w,x) E U x pn : wE lex)}.
This is a so-called incidence set (another example is considered in Exercise
6.5). We first show that it is an (n + I)-dimensional complex manifold. In
fact, we have

(w,7l'(Z» EX{::::=:} Z =f. 0, and 3 A E e with w = AZ

{::::=:} 3 i with Zi =f. 0 and Wj = Wi • Zj for j =f. i
Zi
{::::=:} Z =f. 0 and ZiWj - WiZj = 0 for all i, j.

So X is an analytic subset of U x IP'n, with

Xn (U x Uo ) ~ {(w,t) E U x en : Wj = wotj for j = I, ... ,n}.

In U x Ui (for i > 0) there is a similar representation. It follows that X is

a submanifold of co dimension n in the (2n + I)-dimensional manifold U x
IP'n. The map q := prllx : X -+ U is holomorphic, mapping X - q-l(O)
biholomorphically onto U - {O} by q : (w, x) f-+ wand q-l : w f-+ (w,1I'(w)).
Obviously, q is a proper map.
The preimage q-l(O) is the exceptional set ((O,x) : 0 E lex)} = {O} x IP'n.
So q : X -+ U is a proper modification. It is called HopI's (1-process or the
blowup of U at the origin.
If w =f. 0 is a point of U and (An) a sequence of nonzero complex numbers
converging to 0, then q-l(AnW) = (AnW,1I'(W) converges to (0,1I'(w». This
is the desired property.

The Tautological Bundle. We consider the case U = e n +!, i.e., the

manifold
F:= ((w,x) E Cn+ 1 X pn : wE lex)}
238 IV. Complex Manifolds

and the second projection

p:= p r 21F : F -+ lPn.

Thenp-l(x) = lex) is always the complex line determined by x. The manifold

F looks like a line bundle over the projective space. In fact, we have local
e
trivializations <Pi : p-l(Ui ) -+ Ui X defined by <Pi(W,X) := (X,Wi)' Clearly,
<Pi is holomorphic. In fact, it is biholomorphic with inverse map

So over Uij ,

Hence F has the transition functions 9ij = zd Zj, and consequently, it is

the dual bundle of the hyperplane bundle 0(1). It is denoted by O( -1) and
is called the tautological bundle, because the fiber over x E jpn is the line
lex), which is more or less the same as x. Sometimes F is also called the
Hopi bundle, because it lies in en+! x F and the projection onto the first
component is Hopf's a-process (see Figure IV.3).
Let j: F -+ F x en +1 be defined by j(w,x) := (x,w), and J i : Ui x e-+
Ui X en+! by
J i (lr(Z),c):= (lr(z), ~ .z).
Then we have the following commutative diagram:

Flu; ~ Ui x e
j.!. .!. J i
Ui X en+! rv
Ui X e n +1
This shows that F is a subbundle of the trivial vector bundle lPn x en+!. In
particular, it follows that r(lPn,O(-I)) = 0; Le., there is no global holomor-
phic section in the tautological bundle.
There is an interesting geometrical connection between F and F'. Consider
the point Xo := (0: ... : 0 : 1) E jpn+l and the hyperplane

Ho := {lr(z) E jpn+l : Zn+! = o}.

Then the hyperplane bundle F' = 0(1) is given by the projection 11'+
jpn+! - {xo} -+ Ho ~ F, with

lr+(Zo : ... : Zn : zn+d := (zo : ... : Zn : 0).

 7. Modifications and Toric Closures 239

f(x)

11'

/ UC Cn + 1

/
Figure IV.3. Hopf's u-process

Here Ho is the zero section of 0(1). Removing the zero section gives a man-
ifold isomorphic to C n +l - {O}, and the fibers of the hyperplane bundle
correspond to the lines through the origin. Blowing up the point Xo gives
us an additional point to each fiber and so a projective bundle F' over ]pn
with fiber ]pl. But looking in the other direction, F' - Ho is nothing but the
tautological bundle O( -1) over the exceptional set of the blowing up.

Quadratic Transformations. We consider the case n = 1. Let M

be a 2-dimensional connected complex manifold and p E M a point. Let
U = U(p) c M be a small neigborhood with complex coordinates z, w such
that (z(p), w(p)) = (0,0). Let X c U X ]pI be the blowup of U at the origin.
Then
Qp(M) := (M - U) uX = (M - {p}) U]pl
is again a 2-dimensional complex manifold, with a nonsingular compact an-
alytic subset N ~ ]pI and a proper holomorphic map qp : Qp(M) -+ M such
that q; 1 (p) = Nand qp : Qp( M) - N -+ M - {p} is biholomorphic. We call
Qp(M) the quadratic transformation of M at p.
Let F : MI --+ M2 be a biholomorphic map between 2-dimensional complex
manifolds and F(Pl) =P2. Let ql : Qpl(Md --+ MI and q2: Qp2(M2) --+ M2
240 IV. Complex Manifolds

be the quadratic transformations. Then there exists a biholomorphic map

if' : QPl (Md -+ Qp2(M2 ) such that q2 0 if' = F 0 ql. This follows directly
from the construction, and it shows that the quadratic transformation is a
canonical process.
In [H055) Hopf proved that every proper modification with a single point as
center is a finite sequence of quadratic transformations. 5
We consider the quadratic transformation q : Qo(U) -+ U, where

U = {(z,w) E C 2 : Iii < rl and Iwl < r2}.

Then Qo(U) = {(z,w, «: 7]) E U X pI : Z7] - w( = o} = Uo U Ul, with

Uo ~ {(7], z) E C 2 : IZ7]1 < r2 and Izl < rd,
U1 ~ {«, w) E C 2 : Iw(1 < rl and Iwl < r2}
and
w = Z7] and (=_1 on vo
TT
n U1.
7]
Thus q is given by

(7],Z) t-t (z,z7]) and «,w) t-t (w(,w).

Consider, for example, the curve C := {(z, w) E U : w 2 = Z3}, which has a

singularity at the origin. Then

q-l(C - {O}) n Uo {(7], z) E Uo : z =F 0 and z - 7]2 = O},

q-l(C - {O}) nUl = {«,w) E U1 : w =F 0 and 1- W(3 = O}.

In U1 - Uo we have ( = o. So q-l(C - {O}) lies completely in Uo, and its

closure in Qo(U) is the singularity-free curve C' := {(7], z) : z = 7]2}. We
call it the strict transform of C. The map q : C' -+ C is called a resolution
of the singularities of C. One can show that for any curve in a 2-dimensional
manifold the resolution of the singularities is obtained by successive quadratic
transformations. A proof can be found, for example, in [Lau71).

If a 2-dimensional manifold M contains a compact submanifold Y ~ ]PI, one

can ask whether Y is the exceptional set of a blowup process (then one says
that Y can be "blown down" to a point). Here we give a necessary condition.
We determine the transition functions for the normal bundle of the fiber
Fa := q-l(O) in Qo(U). The divisor Fo is the zero set of the functions 10 :
Ua -+ C with 10(7], z) := z and It : U1 -+ C with It «, w) := w. Therefore,
the associated line bundle [Fa] on Qo is given by the transition function
5 In algebraic geometry this was already known at the beginning of the twentieth
century, for example by the italiens G. Castelnuovo and F. Enriques.
 7. Modifications and Toric Closures 241

z (
90t{Z,W, «(: 77)) =- =- on Uo nUl·
W 77
But (,77 are the homogeneous coordinates on Fo ~ ]pl. So NQo(Fo) = [FollFo
is described by the same transition function as the tautological bundle O( -1).
It follows that a submanifold Y,~]pl in a complex surface M can be excep-
tional only if NM(Y) ~ 0(-1).
We give a topological interpretation of this criterion. We set

So := 1 on Uo and

which defines a meromorphic section s in 0(-1) that has no zero and

only one pole at 00 = (0 : 1). We can use this section to calculate the self-
intersection number Y . Y. Since more knowledge about topology is needed
for the complete calculation, we can give only a sketch.
The self-intersection number of a compact submanifold Y <= M can .be de-
scribed as follows. Take a copy Y' of Y (which need be only a topological or
piecewise differentiable submanifold of M) such that it intersects Y transver-
sally. Then the intersection Y n Y' is a zero-dimensional submanifold, Le., a
finite set. Now count the number of these intersection points respecting orien-
tation. The result is the self-intersection number. Furthermore, one can show
that if Z c NM(Y) is the zero section, then y. Y = Z· Z, and the number
Z . Z can be calculated by intersecting Z with some not identically vanishing
(continuous) section Z'. We obtain such a Z' in the following way. Let s be
the meromorphic section in N = N M (Y) mentioned above. Choosing appro-
priate coordinates near 00, we can there write N = ({z,w) E e 2 : Izl < I}
and s(z) = (z, liz) (since s has a pole of order one). For 0 < e < 1 and
Izl = e we have s(z) = (z, zl( 2 ). Then we define by s
s(z) for Izl > e,
s(z):= (z,;)
{
for Izl ::; e.
So s is a piecewise differentiable section, homologous and transversal to the
zero section with one zero of order 1. Since z gives an orientation opposite to
that defined by z, we have the self-intersection Y . Y = -1.
If Y ~ ]pI is exceptional, then Y . Y = -1. It is a deeper result that the
converse of this statement is also true (cf. [Gr62]).

Monoidal Transformations. We consider a domain G c en

and
holomorphic functions fo, ... , /k on G such that A := N(lo, ... , /k) c G
is a singularity-free analytic subset of codimension k + 1. Then we define
f := (10, ... , fk) : G -r e k + l and
242 IV. Complex Manifolds

x .- {(w, x) E G X jp'k : few) E lex)}

((w,1I"(z)) E G X jp'k : zdj(w) - Zjfi(W) = 0 for i,j = 0, ... ,k}.

The map p := prIlx : X --+ Gis holomorphic and proper. For Wo E G - A we

have p-I(WO) = {wo} xjp'k, whereas p : X _p-I(A) --+ G-A is biholomorphic
with p-I : W H (w, Uo(w) : ... : hew»~).
The set X is called the monoidal transformation of G with center A. The
exceptional set E(f) := p-I(A) is the Cartesian product A x jp'k. In the case
k = n - 1 and fi = Zi for i = 0, ... , k this is the O"-process.
One can show that the monoidal transformation is also a canonical process.
Therefore, it can be generalized to an n-dimensional connected complex man-
ifold M and a sub manifold A C M of codimension k + 1. Then the ~onoidal
- -
transformation of M with center A is an n-dimensional manifold M with a
hypersurface E C M such that M - E ~ M - A. The exceptional set E is
a fiber bundle over A with fiber jp'k. The fiber over x E A is the projective
space jp'(Nx ), where N = NM(A) is the normal bundle of A in X. The divisor
E determines a line bundle L := [E] on M with Llp(N.,) = Op(N.,)(-1). We
leave the details to the reader (see also Exercise 5.10).

Meromorphic Maps. Let X be an n-dimensional complex manifold

and f a meromorphic function on X with polar set P and S C P its set of
indeterminacy points. Then f is holomorphic on X - P, and we can extend
it to a holomorphic map f : X - S --+ jp'1 by setting f(x) := 00 for x E P - S.
The set
G f := {(x,y) E (X - S) X jp'1 : y = f(x)}
is called the graph of f. Let Gf C X X jp'l be the closure of the graph in
X x jp'1.
7.2 Proposition. The graph Gf is an irreducible analytic subset of X x jp'l
and defines a proper modification of X in jp'l with S as set of indeterminacy.

PROOF: Let Xo E P be an arbitrary point. Then there is a neighborhood

U = U (xo) C X and holomorphic functions g, h : U --+ C such that f (x) =
g(x)/h(x) for x E U - P. We can assume that gx and hx are relatively prime
forxEU.ThenPnU={xEU: h(x)=O}andSnU={xEU: g(x) =
hex) = OJ. Hence P is a hypersurface and S has codimension 2 in X. Then
dime S x jp'l) < n and G f C (X X jp'1) - (S X jp'l) is an n-dimensional analytic
set. By the theorem of Remmert and Stein its closure G f in X x jp'1 is also
analytic. In fact, we have

The map 11" := prl : Gf --+ X is obviously a proper holomorphic map, with
1I"-I(X) = {x} X jp'1 for xES and
 7. Modifications and Toric Closures 243

7r- l (x) _ { (x, (1 : I(x))) for x E X - P,

- (x, (0: 1)) for x E P - s.

So 7r : G f - 7r- l (8) --+ X - 8 is biholomorphic, and G f defines a proper

modification of X in pI with indeterminacy set 8. As usual, X is assumed
to be connected. Then Reg( G f) is also connected, and G f irreducible. -

Now we consider the notion of a meromorphic map. To develop the correct

point of view, we start with a very special case. Let
om := pI X . . . X pI
~
m times

be the so-called Osgood space. If II, ... , 1m are meromorphic functions on

X without, points of indeterminacy, then they define a holomorphic map
f = (II, ... , 1m) : X --+ om. If Ii has a set 8 i as set of indeterminacy, we
define 8 := 8 1 u· .. u 8 m . Then f is a holomorphic map from X - 8 into om.
Again we have its graph:

Gr = {(X;Yb ... ,Ym) E (X - 8) X Om : Yi = li(X) for i = 1, ... ,m}.

For Xo E 8 we define

Tr(Xo) := {y E om : 3xv E X - 8 with Xv --+ Xo and f(xv) --+ y}.

Then Tf(XO) is a nonempty compact subset of om. Setting Tf(X) := {f(x)}

for X E X - 8, we get a map Tf from X into the power set P( om). 6 We define

G f := {(x,y) E X x om : y E Tf(X)}.

One sees easily that Gf is the topological closure of Gf in X x om. We cannot

apply the Remmert-Stein theorem to show that Gf is an analytic set, but
locally it is an irreducible component of the analytic set A with

{(x; (zJ : zi), ... , (zo : zl)) E U x om :

ZJgl(X) - zth l (x) = ... = Zo9m(X) - zlhm(x) = O},
where Ii = gdhi on U - Pi, for i = 1, ... , m. We could carry out the simple
proof here, but we refer to [Re57], Satz 33.
Definition. A meromorphic map between complex manifolds X and
Y is given by a nowhere dense analytic subset 8 eX, a holomorphic
map I : X - 8 --+ Y, and an irreducible analytic set X c X x Y such
.that
1 0 prl = pr2 on X n s) x Y) «X -
and 7r := prllx : X --+ X is proper and surjective. The set 8 is called the
. set 01 indeterminacy of I. We also write I: X ~ Y.
6 As usual, the power set P(M) of a set M is the set of all subsets of M.
244 IV. Complex Manifolds

We define Tf : X ~ P(Y) by Tf(x) := pr2(7r-1(X)). Then Tf(x) i§ always a

nonempty compact set, and

x = ((x,y) E X x Y : y E Tf(x)}.
For x E X - S we have 7r- 1(x) = {(x,/(x))}, and for xES we have
7r- 1 (x) = X n ({x} x Y). This is a compact analytic set. If S is minimal, then
7r : X ~ X is a proper modification with center S.
So alternatively we could have defined a meromorphic map between X and
Y to be a map T : X ~ P(Y) such that:
1. X := {(x, y) E X x Y : y E T(X)} is an irreducible analytic subset of
XxY.
2. 7r:= pr11x : X ~ X is a proper modification.
This definition is due to Remmert.
A meromorphic map is called surjective if pr2(X) = Y. If I is a surjective
meromorphic map between X and Y and 9 a meromorphic map between Y
and Z, then it is possible to define the composition 9 0 I as a meromorphic
map between X and Z (cf. [Ku60]). But it is not a composition of maps in
the set-theoretic sense!
Any meromorphic function I on X defines a meromorphic map between X
and pI, any m-tuple (/1, ... ,1m) of meromorphic functions a meromorphic
map between X and om.
7.3 Proposition. Let I: X ~ Y be a proper modification olY with center
S. Then 1-1 : Y -->.X is a surjective meromorphic map with S as set 01
indeterminacy.

PROOF: We use the analytic set

Y:= ((y,x) E Y x X : y = I(x)} = ((y,x) E Y x X : x E 1- 1 (y)}.

Then 1-1: Y - S ~ X is holomorphic with 1-1 oprl(Y'X) = 1- 1(y) = X =
pr2(Y' x) for (y, x) E Y n «Y - S) x X). Obviously, 7r := prIll' : Y ~ Y is
a proper holomorphic map. It is surjective, since I is surjective. Finally, we
have pr2(Y) = X. So 1-1 defines a surjective meromorphic map. _

Toric Closures. Let X be an n-dimensional complex manifold. A clo-

sure of X is an n-dimensional compact complex manifold M such that X is
biholomorphically equivalent to an open subset of M. We are interested in
closures of X = en.
In the case n = 1 the situation is simple. Assume that we have a compact
Riemann surface M that is a closure of X = C. Then we define I : M - {O} ~
C by
 7. Modifications and Toric Closures 245

{ 0liz
for z E C - {O},
I(z) :=
otherwise.
The function I is continuous and is holomorphic at every point z with I(z) #
O. By the theorem of Rad6-Behnke-Stein-Cartan (cf. [Hei56]) I is everywhere
holomorphic. But then the zero set N(f) is discrete. We can consider I as
a meromorphic function on M with exactly one pole of order one at Z = O.
So I can have only one zero (cf. [F081J, Section I.4, Proposition 4.24 and
corollary), and M must be the Riemann sphere pI, the well-known one-point
compactification of C.
In higher dimensions one can find several closures of cn. In [Bie33) an injective
holomorphic map {3 : C 2 -+ C 2 is constructed whose functional determinant
equals 1 everywhere and whose image U = (3(C 2 ) has the property that there
exist interior points in C 2 - U. We can regard U as an open subset of p2.
Then p2 is a closure of C2 ~ U, but this closure contains interior points in
its boundary. We want to avoid such "pathological" situations, even if they
may be interesting for those working in dynamical systems.

Definition. A closure of cn is a triple (X, U, 11» where:

1. X is an n-dimensional connected compact complex manifold.
2. U C X is an open subset with U = X.
3. 11>: C n -+ U is a biholomorphic map.
The closure (X, U, 11» is called regular if for any polynomial p the holo-
morphic function poll> -1 extends to a meromorphic function on X.

7.4 Proposition. II (X, U,II» is a regular closure olcn, then Xoo := X-U
is an analytic hypersurlace in X.

PROOF: Let /; be the meromorphic extension of Zi 011>-1, for i = 1, ... , n.

The polar set Si of Ii is a hypersurface in X. We set P := PI u··· U Pn and
claim that P = X - U.
One inclusion being trivial, we consider a point Xo E X - U. There is a
sequence of points Xv E U with limv~oo Xv = Xo. If we write Zv := 11>-1 (xv) as
Zv = (zt),··., z;;-» , there must be an index k with IZkv)1 -+ 00. So 1!k(xv)1 -+
00 for v -+ 00. This means that Xo E Pk C P. •

Example
The manifold X = pn is a closure of C n . As open subset U with U = X we
can take the set U := Uo = {(zo : ... : zn) I Zo # OJ. Then 11>0 : C n -+ U with
11>0 (t 1, ... , t n ) : = (1 : t 1 : ... : t n ) is biholomorphic. We have seen in Section
5 that any polynomial on U extends to a meromorphic function I on pn with
246 IV. Complex Manifolds

polar set Ho = {(zo : ... : zn) I Z{) = o}. So (pn, Uo, CPo) is a regular closure of
C n , with Xoo = Ho·
Another example is the Osgood space on, which contains C n via

Also, (on, C n , CP) is a regular closure, but here

Xoo = (£oo} X pl X ... X pl) U ... U (pl X ... X pl X {oo})

is an analytic hypersurface with singularities. The complex Lie group G n :=

C* x ... x C* (n times) acts in a natural way on Cn and is contained in Cn
as an open orbit of this action.
Definition. A complex manifold X is called a torie variety or torus
embedding if G n acts holomorphically on X and there is an open dense
subset U C X biholomorphically equivalent to G n by 9 f-+ gx for each
xEU.

Remark. In algebraic geometry the group Gn is called a torus. This is the

reason for the notion "toric variety." A ''variety'' is an irreducible algebraic
subset of C n or pn. It may have singularities. Here we do not allow singular-
ities, but we allow arbitrary manifolds that do not have to be algebraic.
More information about toric varieties can be found in [Oda88j, [Fu93], and
[Ew96j.
We wish to restrict the closures of C n as much as possible. If the axes of C n
can be extended to the closure, then the action of Gn should also extend to
the closure.
Definition. A closure (X, U, CP) of C n is called a torie closure if the
action ofGn on Cn extends to an action on X such that CP(g·z) = g.cp(z)
for 9 E Gn and z E Cn .

Examples
1. The projective space pn is a toric variety with

g. (zo : ... : zn) := (zo : gZl : ... : gZn)

and G n = (c*)n c C n = Uo C F. Since the embedding C n = Uo y lP'n
is "equivariant," pn is a toric closure of C n .
2. A further simple example is the Osgood space on. Here the toric action
is given by

g' ((zJ : zt), ... , (zo : zi)) := ((zJ : gz}), ... , (zo : gzi))·
 7. Modifications and Toric Closures 247

3. Let Ad C No be the set of multi-indices v with Ivl < d and define

CPd : G n -t pN by CPd(Z) := (ZV : v E Ad). Then Xd := CPd(Gn ) C pN is
the image of pn under the Veronese map Vd,n with

Vd,n(ZO: Zl : •.. : zn) = (zg-IVlzv : v E Ad).

Obviously, Xd is a toric closure of en.

7.5 Proposition. Let (X, U, CP) be a regular closure olen and 1: Z -t X a

proper modification with center S C Xoo. Then (Z, I-l(U),J-l 0 CP) is again
a regular closure.

PROOF: It is clear that (Z, I-l(U), 1-10 CP) is a closure. Now let p be a
polynomial. Then the holomorphic function po cp-l : U -t e has a mero-
morphic extension fi on X. But then fi 0 1 is a meromorphic extension of
po (I-I 0 CP)-1 = (po CP-l) 01. •
Now let eX, U, CP) be an arbitrary regular toric closure of en. The meromor-
phic extensions Ii of ZiOcp-1 define a meromorphic mapping f = (ft,···, In) :
X ........ on and therefore a generalized modification X of X in on. We know
that X is an irreducible component of the analytic set

A = {(XjZl. ... ,Zn) E X x on : z~It(x) - zUJ(x) = 0 for i = 1, ... ,n},

where we write Zi E pI in the form Zi = (z~ : zD and Ii in the form
Ii = (Jj: In·
The two projections 11' := prllx : X -t X and q := pr21x : X -t on
are proper holomorphic maps. It may be that X has singularities over the
indeterminacy set S of f. Nevertheless, we consider X as a closure of en.
We take fj := lI'-I(U) as dense open set, and the biholomorphic map i :=
11'-1 0 <P : en -t fj. This is possible, since flu = cp-l is holomorphic and
therefore SeX - U.
The preimage 11'-1 (X - U) = X - fj is a nowhere dense analytic subset, and
the meromorphic map f 0 11' : X . . . . on coincides on fj with the holomorphic
map q. By the identity theorem it follows that f 0 11' = q is holomorphic
everywhere on X. Since (foll')lo = cp-l 011' = i - I , every polynomial extends
to a meromorphic function (without points of indeterminacy) on X. SO the
new closure X is regular. Since qlo = i - I maps fj biholomorphically onto
en, the map q : X -t on is a proper modification of on (in a generalized
sense, since X may have singularities).
Let us now consider the q-fibers. If Zo E on is given, the fiber q-l(ZO) is
contained in the set
248 IV. Complex Manifolds

F:= (flu)-I(zo) U {x E X - U : 3z 11 E (:n with 4>(ZIl) --* x, ZII --* zo}.

For Zo E (:n we obtain q-l(zO) = {4>(Zo)}, whereas for Zo E on - en the

fiber is a compact subset of the hypersurface X - U.
It is easy to check that X is even a toric closure. One uses the fact that X
and on both are toric and that flu is Gn-equivariant.
Example
Let X = pn be projective space and carry out the above construction. The
functions Ii : pn ~ pI are given by

Ii: (Zo : ZI : ... : Zn) t-+ (zo : Zi).

These are meromorphic functions with polar set Ho and indeterminacy set
Ho n Hi· In the inhomogeneous coordinates wo, ... , iilj, ... , Wn with Wk =
Zk/ Zj the function Ii is equal to wdwo for i =1= j (respectively to 1/wo for
i = j).

Now X is given by

X = {(w; ZI, ... , zn) E pn x on : ZbWi - ziwo = 0 for i = 1, ... , n}.

This is built from pn by a finite sequence of quadratic transformations, and

therefore regular. The space X is a regular toric closure of en in the original
sense.
There are many toric closures, but pn and on are the most important ones.
More information about this topic is given in [BrM078) and [PS88).

Exercises
1. Calculate the strict transform of X = {(z, w) : w 2 = z2(z + 1)} when
blowing up (:2 at the origin.
2. Show that KlI'n ~ O( -n - 1) and K y ~ Oed - n - 1) for every regular
hypersurface Y C pn of degree d.
3. Let EI be the projective bundle associated to the vector bundle V =
011'1 E!1 011'1 (1) (cf. Exercise 5.10). Show that the bundle space of EI is
biholomorphically equivalent to the blowup Qp(p2) with p = (1 : 0 : 0).
4. Consider the meromorphic function I(z,w):= w/z on (:2 and show that
its graph G f C (:2 X pI is equal to the blowup of (:2 at the origin.
5. Let X be a complex manifold and L a holomorphic line bundle over X.
Show that any set of linearly independent global sections So,.··, SN E
r (X, L) defines a meromorphic map 4> : X --* pN by
4>(X) := (So (x) : ... : SN(X)).

What is the exact meaning of this definition?

 7. Modifications and Toric Closures 249

6. Let X be a compact connected complex manifold. A set of meromorphic

functions 11, ... , 1m E Al(X) is called analytically dependent if f(G f ) i
om, where f = (ft, ... ,Jm) is the second projection Gf C X x om -+- om.
Let P be the union of the polar sets of the Ii and show that the following
properties are equivalent:
(a) ft, ... , 1m are analytically dependent.
(b) The Jacobian of f : X - P -+- C m has rank less than m.
(c) There is a nonempty open subset U C X - P, a domain G C C m
with feU) c G, and a holomorphic function 9 on G such that
g(ft(x), ... , 1m (x)) == o.
The functions ft, .. . , 1m are called algebraically dependent if there is a
polynomial pi 0 such that p(ft(x), ... , 1m (x)) == 0 on X -Po Prove that
every set of analytically dependent functions is also algebraically depen-
dent. (Hint: The map G f -+- om is proper. Use without proof Remmert's
proper mapping theorem, which says that the image of an analytic set
under a proper holomorphic mapping is also an analytic set. Then use
Chow's theorem.)
7. Let 4> : G 2 -+- ]p3 be defined by 4>(ZI' Z2) := (1 : ZI : Z2 : Z1Z2). Show that
X = 4>(G2) C p3 is a closed submanifold that coincides with the image
of the Segre map a1,1 : ]pI x pI -+- p3 with a1,1 ((zo : zt), (wo : wt}) :=
(zowo: ZIWO: ZOWI : Z1WI). Prove that X is a toric closure ofC 2 .
Chapter V

Stein Theory

1. Stein Manifolds
Introduction
Definition. A complex manifold X is called holomorphically spread-
able if for any point Xo E X there are holomorphic functions h, ... , f N
on X such that Xo is isolated in the set

N(h, ... ,fN) = {x EX: hex) = ... = fN(X) = D}.

It is clear that the holomorphic map f := (h, ... ,fN) : X -+ eN is discrete

at Xo. In [Gr55] it is shown that if n = dim(X), then there is a discrete
holomorphic map 7r : X -+ en. Thus (X, 7r) is a branched domain over en.
Furthermore, it follows that the topology of X has a countable base. Note
that if X is holomorphically separable, i.e., for any x, y E X with x f:. y
there exists a holomorphic function f on X with f(x) =1= fey), then it is
holomorphically spreadable.
If X is holomorphically spreadable, A c X a compact analytic set, and
Xo E A, then there are holomorphic functions h, ... , fN on X (which then
must be constant on A) such that Xo is isolated in NUl, ... '!N). SO Xo must
be an isolated point of A, and it follows that every compact analytic subset
of X is finite!
Definition. A complex manifold X is called holomorphically convex
if for any compact set K C X the holomorphically convex hull

R := {x EX: If(x) I ::; s~plfl.for every f E O(X) }

is likewise a compact subset of X.

Every compact complex manifold is holomorphically convex. A complex man-

ifold with a countable base is holomorphically convex if and only if for any
infinite discrete subset D C X there exists a holomorphic function f on X
such that sUPDlf1 = 00. The proof is the same as in en.
Definition. A Stein manifold is a connected complex manifold that is
holomorphically spreadable and holomorphically convex.
252 V. Stein Theory

A Stein manifold of dimension n > 0 cannot be compact. In the converse

direction, Behnke and Stein proved in 1948 that every noncompact Riemann
surface is a Stein manifold (see [BeSt48]). For n ~ 2, a domain G c en
is Stein if and only if it is holomorphically convex. By the Cartan-Thullen
theorem this is the case exactly if G is a domain of holomorphy. This remains
true for (unbranched) Riemann domains over en, since Cartan-Thullen is
also valid for such domains. Hence for n ~ 2 there are many noncompact
domains in en and over en that are not Stein.
1.1 Proposition. Let f : X -t Y be a finite holomorphic map between
complex manifolds. If Y is a Stein manifold, then X is also Stein.
In particular, every closed submanifold of a Stein manifold is Stein.

PROOF: It is necessary to show only that X is holomorphically convex. For

this let K c X be a compact set and note that f(K) is also compact. We

- -
consider an arbitrary point x E R. For 9 E O(Y) we have go f E O(X)
and Ig 0 f(x)1 ~ sUPKlg 0 fl, so f(x) E f(K) and x E f-1(f(K». Since Y
is holomorphically convex an~ f proper, f-1(f{K) is compact as well. As a
closed set in a compact set, K is compact.
Since en is Stein, every closed submanifold of en is also Stein. •
Example
Every affine-algebraic manifold is isomorphic to a closed submanifold of en.
Therefore, it is a Stein manifold. On the other hand, we shall see later on
that there exists an algebraic surface that is Stein, but not affine-algebraic.
1.2 Proposition. If X is a Stein manifold and f E O(X), then X - N(f)
is Stein.
PROOF: If X is a Stein manifold, then it is immediate that X - N(f) is
holomorphically spreadable. Let D be an infinite discrete set in X - N(f).
If it is discrete in X, nothing remains to be proved. If it has a cluster point
Xo E N(f), then 9 := 1/ f is holomorphic on X - N(f) and unbounded on
D. •

FundaIllental Theorems. The following important theorems cannot

be proved completely within the constraints of this book. One needs deeper
tools, including sheaf theory.
1.3 Theorem A. Let 7r : V -t X be an analytic vector bundle over a Stein
manifold X. Then for any Xo E X there are global holomorphic sections
Sl, .. " SN E r(X, V) with the following property:
 1. Stein Manifolds 253

IfU = U(xo) c X is an open neighborhood and s E r(U, V), then there exists
an open neighborhood V = V(xo) C U and holomorphic functions It,···, fN
on V such that
slv = It· S1 + ... + IN' SN·

1.4 Theorem B. Let 7f : V ~ X be an analytic vector bundle over a Stein

manifold X. Then Hl(X, V) = O.
Moreover, if A c X is an analytic set, ~ an open covering of X and ~ E
Zl(~, V) such that ~vJ.tlu~"nA = 0, then we can find a cochain "I E CO(~, V)
with 77vlu~nA = 0 and &"1 =~.
The first part of this theorem will be proved at the end of this chapter.
1.5 Oka's principle. Let X be a Stein manifold.
1. Every topological fiber bundle over X has an analytic structure.
2. If two analytic fiber bundles over X are topologically equivalent, they are
also analytically equivalent.

Cousin-I Distributions. A Mittag-Leffler distribution on C consists of

a discrete set {zv : v E N} in C together with principal parts hv of Laurent
series at each Zv' A solution of this distribution is a meromorphic function
on C with the given principal parts and no other pole. We can express the
situation in the language of Cech cohomology. Define Uo := C - {zv : v E N}
and ho := 0, and let Uv be an open neighborhood of Zv that contains no zJ.t
with J.L f:. v. Then hvlu~ is meromorphic, and hv - hI-' is holomorphic on UVI-I"
A solution is a meromorphic function I such that I - hv is holomorphic on
Uv '
Definition. Let X be a complex manifold. A Cousin-I distribution
on X consists of an open covering ~ = {Ut : LEI} together with
meromorphic functions ft on U. such that I. - II< is holomorphic on U.I<'
A solution of the Cousin-I distribution is a meromorphic function I on
X such that f - I. is holomorphic on U•.

If a Cousin-I distribution is given, then a cocycle ~ E Zl(~, 0) is defined by

~Ut := (II< - f.)lu.",· If the distribution has a solution I, then "I. := (It - f)lu.
defines a cochain 11 E CO(~, 0) with 111< - "I. = '.1<' In other words, &"1 = ,.
Conversely, if there exists an "I with &"1 = " then Ilu. := It - "I. defines a
meromorphic function f on X with I. - I = "I. on U•. This means that f is
a solution.
The following result is a consequence of Theorem B for the trivial bundle
Ox =X xC.
254 V. Stein Theory

1.6 Proposition. On a Stein manifold every Cousin-I distribution has a

solution.
In particular, on an open Riemann surface every Mittag-Leffler distribution
has a solution.
Remark. The condition "Stein" is not necessary. If X is a complex mani-
fold with HI(X, 0) = 0 (for example X = pI), then every Cousin-I distribu-
tion has a solution. And even if HI (X, 0) =F 0, then the cohomology class of
a given distribution may be zero. This implies that this special distribution
has a solution.

Cousin-II Distributions. Recall the Weierstrass theorem: Let {z" :

V E N} be a discrete set in C, and (n,,) a sequence of positive integers. Then
there exists a holomorphic function f on C that has zeros exactly at the z",
and these are of orders n". The distribution (zv,n")"EN is a divisor D on C
with D ~ 0 and div(f) = D.
A divisor D on a complex manifold X is called a principal divisor if there
is a meromorphic function m on X with dive m) = D. If D ~ 0, then m is
holomorphic. So we are interested in conditions on X such that every divisor
is principal. From the exact sequence Al(X)* ~ ~(X) ~ Pic (X) it fol-
lows that Ker(6) = ~(X) is a necessary and sufficient condition. This is, for
example, the case if every analytic line bundle on X is trivial. Due to Oka's
principle, if X is Stein, it is sufficient that every topological line bundle be
trivial. One can prove that the topological line bundles on X are classified
by elements of H2(X, Z). Here we will show directly that the vanishing of
H2(X, Z) is sufficient for the solvability of the generalized Weierstrass prob-
lem.
Definition. Let X be a complex manifold. A Cousin-II distribution
on X consists of an open covering %' = {U. : (, E J} together with
meromorphic functions f. =F 0 on U. such that on U.I< there are nowhere
vanishing holomorphic functions g.1< with f. = g.1< . fl<'
A solution of the Cousin-II distribution is a meromorphic function f
on X such that flu. = h;1 . f. with a nowhere vanishing holomorphic
function h. on U•.

A Cousin-II distribution is therefore a meromorphic section in an analytic line

bundle L that is defined by the transition functions g.I<' If L is trivial, then
there are nowhere vanishing holomorphic functions h. on U. with g.1< = h.h; I ,
and flu. := h;l . f. defines a solution.
The Cousin-II distribution uniquely determines the cocycle 'Y = (g.l<) E
ZI(%" 0*). If f is a solution with flu. = h;1 . f., then
 1. Stein Manifolds 255

The following shows that the converse also holds.

1.1 Proposition. Let'Y E ZI(%"O*) be the cocycle of a Cousin-II distri-
bution. The distribution is solvable if and only if'Y E BI (%' ,0*).
If HI(X, 0*) = 0, every Cousin-II distribution on X has a solution.
Unfortunately, Theorem B cannot be applied to HI(X, 0*). So even for Stein
manifolds we need an additional topological condition.

Chern Class and Exponential Sequence. Let X be an arbitrary

n-dimensional connected complex manifold. We choose an open covering %' =
{UL ; LEI} of X such that the UL and all intersections Uu" are simply
connected and construct an exact sequence of abelian groups:

HI(%"Z) -4 HI(%"O) ~ HI(%"0*) ~ H2(%"Z).

1. If'; = (';u<) E ZI(%"Z), then also'; E ZI(%" 0). If'; is an integer-valued

coboundary, then it is also a holomorphic one. Thus j is well defined.
2. Let f = (fLI<) E ZI(%"O) be given. Then we define

e([f]) ;= [exp(211"iftl<)J.

Obviously, e is well defined, and eo j([W = [1]. On the other hand,

if e([f]) = [1], then there are nowhere vanishing holomorphic functions
hL with exp(211"ifut) = hLh;l. Since the UL are simply connected, there
are holomorphic functions gL with exp(gL) = h t . Then exp(gL - gKJ =
exp(211"iftl<), and it follows that there are integers ';LI< with
1 1
fu< = -2 . gL -
11" I
-2. gl<
11" I
+ ';LI<.
This means that [fJ = j([';]).
3. Here we define c; HI (%',0*) -t H2(%" Z). Let h = (hLI<) E ZI (%',0*)
be given. Since the UtI< are simply connected, there are holomorphic
functions gtl< on ULI< with exp(211"igLI<) = hLI<. They are determined only
up to some 211"i';LI< with ';LI< E Z, but

TJVI-'Il ;= (gl-'Il - gVIl + gVI-I) I

UVI"Q

uniquely defines a cohomology class [1]J E H2(%" Z), since

exp(211"i1]vl-'ll) = hl-'Ilh;;i hilI-' = 1

and
256 V. Stein Theory

Let c([h]) := [1]].

If [hI = e([J]), then there are nowhere vanishing holomorphic functions
h. with

where the g, are suitable holomorphic functions on U,. We have coe([J]) =

[1]] with 7]VJ.Le = (fJ.Le - fve + fVJ.L)luv..u = 0, since f is a co cycle.
Now let [hI be given with c([h]) = 0. Then there is an element ~ = (~u<) E
CI(tF/,Z) with

where exp(27rig.K) = h. K. So f'K := g'K - ~'I< is a co cycle with values in

0, and exp(27rif.K) = h. K • This means that e([f]) = [hI.
If the covering tF/ is fine enough, everything is independent of the covering.
By construction it is acyclic with respect to Z.

Definition. Let h = (h'K) E ZI (tF/ , 0*) be a cocycle defining an

analytic line bundle L E Pic(X) = HI(X, 0*). Then c(h) := c([h]) =
c(L) E H2(tF/,Z) = H2(X,Z) is called the Chern class of h (or of L).

Let us return to the Cousin-II distributions.

1.8 Proposition. Let X be a Stein manifold. Then a Cousin-II distribution
on X is solvable if and only if its Chern class vanishes.
If H2(X, Z) = 0, every Cousin-II distribution is solvable.

°
PROOF: If X is Stein, then HI(X,O) = and c: HI(X,O*) ~ H2(X,Z)
is injective. Let h be a Cousin-II distribution on X. Then h is solvable if and
only if [hI = 0, and that is the case if and only if c(h) = o.
If in addition H2(X, Z) = 0, then also HI(X, 0*) = {I}. •

Remarks As in the case of Cousin-I distributions, here it is sufficient that

HI (X, 0) = 0, and X need not be Stein. On the other hand, there are
examples of simply connected Stein manifolds where Cousin-II is not solvable.
The condition H2(X, Z) = 0 is essential, and for the solvability of any special
distribution the vanishing of the Chern class is necessary.
In the Stein case, using higher cohomology groups one can show that c :
HI(X,O*) ~ H2(X,Z) is even bijective.
For a noncompact Riemann surface X it is always the case that H2(X, Z) = o.
Therefore, in this case the Weierstrass problem is always solvable.
 1. Stein Manifolds 257

Extension from Submanifolds. Let X be a complex manifold and

A c X an analytic set. A continuous function f : A ~ e is called holomorphic
on A if for every x E A there is an open neighborhood U = U(x) C X and
a holomorphic function 10n U such that JiunA = flunA. One sees easily
that this notion is well defined, but the local continuation 1 is not uniquely
determined.
1.9 Theorem. Let X be a Stein manifold, A C X an analytic set, and
f :A ~ e a holomorphic function. Then there exists a holomorphic function
1 on X with ilA = f·

PROOF: In the differentiable category such a theorem would be proved with

the help of a partition of unity. Here in complex analytic geometry we use
cohomology. It is a typical application of Theorem B.
We can find an open covering %' = (U.).EI of X and holomorphic func-
tions 1..
on U. with 1..lu,nA = flu,nA. Then e'l< := defines a cocy- 1.. - 1..
de e E Z1(%,,0), with e'l<lu,,,nA = O. By Theorem B there is a cochain
'f/ E CO (6// ,0) with 'f/.lu,nA = 0 and 8'f/ = Then e.
f. - fl< = 'f/I< - 'f/. on U.I<,

and liu, := 1.. +'f/. defines a global holomorphic function 10n X. Obviously,
liA = f· •

Unbranched Domains of Holomorphy. Let X be an unbranched

domain of holomorphy over en. Obviously, X is holomorphically spreadable,
and since Cartan-Thullen holds for such domains, X is holomorphically con-
vex. It follows directly that X is a Stein manifold, and all theorems on Stein
manifolds are applicable in this case.
We consider now an unbranched domain of holomorphy p : X ~ pn. We will
>show that it is also a Stein manifold.
Let 11" : en+! - {OJ ~ IPn be the canonical projection. Then we obtain a new
manifold
X := {(x, z) E X x (C n +! - {O}) : p(x) = 1I"(z)}
together with two canonical projections 7f : X ~ X and p: X ~ C n +!_ {OJ.
Since (p,1I") : X X (C n + 1 - {O}) ~ pn x pn is a submersion, the fiber product
X = (p, 11") -1 (6p R) is in fact an (n + 1)-dimensional complex manifold, and
we have the following commutative diagram:

X L C n + 1 - {OJ
7f..j.. ..j..1I"
X ----t IPn
p
258 V. Stein Theory

For a point Po := (xo, zo) of X there exist open neighborhoods U = U(xo) C

X and V = V(1T(ZO» c IP'n such that p : U --+ V is biholomorphic. Then
1T- I (V) is a neighborhood of zo, if-l(U) = (U X (C n + 1 - {o}» n X a
neighborhood of Po, and p: if-I(U) --+ 1T- I (V) a biholomorphic map with
p-I : Z f-t «plu )-I(1T(Z», z). So p: X --+ M := C n +! - {o} is an unbranched
Riemann domain. It is fibered over X by if with fibers isomorphic to C· .
By hypothesis X is a domain of holomorphy. Therefore, it is pseudo convex.
Now we use notation and results from Section 11.8. If X has only removable
boundary points, then X has locally the form U - A, with an open set U C C n
and an analytic hypersurface A. But then X also has such a form and is
pseudoconvex. If X has at least one nonremovable boundary point, then X
(which is something like a cone over X) has only non-removable boundary
points over 0 E C n + l . The set of these boundary points over 0 is thin in
ax. Outside the origin X is pseudoconvex, since it locally looks like X xc.
By a theorem of Grauert/Remmert (see Section 11.8) it follows that X is
pseudoconvex everywhere. From Oka's theorem we know that then X is a
Stein manifold.
To show that X is also Stein, we have first to show that X is holomorphically
convex. If not, we would have an infinite sequence D = {Xi : i E N} in
X without a cluster point such that every holomorphic function f on X
is bounded on D. We have that i5 := if-I (D) is an analytic subset in X,
and fl;;:-l(x;) := i defines a holomorphic function f on D. Since X is Stein,
f
there exists a holomorphic function on X with li.B
= f. On every fiber
if-I (x) ~ C· we have a Laurent expansion

L
00

fl;;:-l(x) = av(x)zv.
v=-oo

Since if is a submersion, we can define a holomorphic function 9 on X by

g(x) := ao(x). Then g(Xi) = i, and this is a contradiction.
In the same way we can show that for x, y E X with X =I- y there exists a
holomorphic function f on X with f(x) =I- f(y). Thus X is holomorphically
separable and therefore Stein.

The Embedding Theorem. By a theorem of Whitney every differen-

tiable manifold can be embedded into a space ]RN of sufficiently high dimen-
sion. In general, this is false for complex manifolds, since, e.g., in C N there
are no positive-dimensional compact complex submanifolds. However, Stein
manifolds can be embedded.
1.10 Theorem. If X is an n-dimensional Stein manifold, then there exists
c
an embedding j : X '-+ 2n + I.
 1. Stein Manifolds 259

The proof is due to Remmert (see his Habilitationsschrift and [Re56]),

Narasimhan ([Nar60J), and Bishop ([Bi61)).
O. Forster ([F070)) proved the following theorem:
1.11 Theorem. Any 2-dimensional Stein manifold can be embedded into
C 4 , but there is a 2-dimensional Stein manifold that cannot be embedded into
C3 .
In general, for n ?: 2 every n-dimensional Stein manifold can be embed-
c
ded into C 2n , and for n ?: 6 even into 2n -[(n-2)/3). For arbitrary n there
always exists a proper immersion into C 2n - 1 . But there are examples of n-
dimensional Stein manifolds X that cannot be embedded into CN (and not
immersed into CN - 1) for N := n + [~J.
At the end of his paper Forster conjectured that every n-dimensional Stein
manifold can be embedded into CN+I and immersed into CN.
In 1992 J. Schiirmann showed that Forster's conjecture is true for n ?: 2 (see
[Schue92],[Schue97)). In 1970/71 Stehle had given an embedding of the unit
disk ~ into (:2 (see [St72]), and in 1973 Laufer an embedding of certain annuli
into C 2 • At the moment it is an open problem whether every noncompact
Riemann surface can be embedded into C 2 • There has been quite a bit of
progress on this by the work of J. Globevnik and B. Stens0nes (see [GISte95]),
but no final result.

The Serre Problem. Above it was proved that if f : X ---+ Y is a finite

holomorphic mapping and Y a Stein manifold, then X is also Stein.
In 1953 Serre posed the following problem: Is the total space of a holomorphic
fiber bundle with Stein base Y and Stein fiber F a Stein manifold? ([Se53]).
It is easy to show that the answer is positive in the case of an analytic vector
bundle.
Between 1974 and 1980 it was proved that the answer is positive in the
case of I-dimensional fibers (Siu, Sibony, Hirschowitz and Mok). In 1976 Siu
generalized this to the case where the fiber is a bounded domain G c cn
with
HI (G, q = 0, and in 1977 Diederich and Fornress showed that the Serre
conjecture is true if the fiber is a bounded domain in C n with C2 smooth
boundary.
In 1977 Skoda ([Sko77]) gave the first counterexample, a fiber bundle with
C 2 as fiber and an open (not simply connected) set in C as base. Demailly
improved the example; he used the complex plane or a disk as base. Finally,
in 1985 G. Coeure and J.J. Loeb presented a counterexample with C· as
base and a bounded pseudoconvex Reinhardt domain in C 2 as fiber. Over
the years a number of positive examples have been found, for example by
Matsushima/Morimoto, G. Fischer, Ancona, Siu, and Stehle. Examples where
260 V. Stein Theory

the fiber is a so-called Banach-Stein space play an important role, but we

cannot go into the details here.

Exercises
1. Show that the Cartesian product X x Y of two Stein manifolds is Stein.
2. Use Theorems A and B to prove that the total space of a vector bundle
over a Stein manifold is Stein.
3. Give an example of a Cousin-I distribution on ((:2 - {OJ that has no
solution.
4. Let Uj := {z E ((:3 : Zj =I- O} and au
= {Ub U2 , U3 }. Show that every
Cousin-I distribution with respect to au
has a solution.
5. Let X be a complex manifold, 7r : L -7 X a holomorphic line bundle, and
Z C L the zero section. Prove that if L - Z is Stein, then X is a Stein
manifold as well.

2. The Levi Form

Covariant Tangent Vectors. Let X be an n-dimensional complex
manifold and x E X a point. The elements of the tangent space Tx(X)
(abbreviated by T) are the (contravariant) tangent vectors at x. If Zl,"" Zn
are local coordinates at x, then every tangent vector can be written in the
form
v= LVv~+
a LVv a =: .
v uZv v uZv

Of course, T has a natural structure of an n-dimensional complex vector

space.
Now we consider the space F = F(T) of complex-valued real linear forms on
T. For example, if I is a local (real- or complex-valued) smooth function at
x, then its differential (df)x E F is defined by (df)x(v) := v[/]. It is uniquely
determined by the germ of I at x. In local coordinates we get

v v

It follows that
(dj)x(v) = (df)x(v).
In particular, we have the elements dz v , cLz v E F defined by
dzv(v) := vlzv) and cLzv(V):= vlzv].
Then dz v = dx v + idyv and cLz v = dzv = dx v - idyv, for v = 1, ... , n.
The space F is the complexification of the 2n-dimensional real vector space
T* = HomlR(T,lR). Therefore, F = T* EB i T* is a 2n-dimensional complex
vector space, with basis
 2. The Levi Form 261

We call the elements of F complex covariant tangent vectors or complex 1-

forms at x.
We have (df)x = Lv fz)x) dzv + Lv /z.,(x) azv'
In general, a complex covariant r-tensor at x is an IR-multilinear mapping

cp:Tx .. ·xT~C.
~
r-times

The tensor product cp ® 1/1 of an r-tensor and an s-tensor is the (r + s )-tensor

given by

(cp ® 1/1)(Vl, ... , vr, Vr+l.···, vr+s) := CP(Vl"'" vr) ·1/1(vr+1!"" vr+8 ).
The set of r-tensors carries the structure of a complex vector space, and the
assignment (cp,1/1) H cp ® 1/1 is C-bilinear. For example, (dz v ® azl')(v, w) =
vv,ww

Hermitian Forms. Let X be an n-dimensional complex manifold. The

notion of a plurisubharmonic function in a domain G C cn
was already
defined in Chapter II. Of course, a plurisubharmonic function on a com-
plex manifold is a real-valued function that is ('"COO) differentiable and pluri-
subharmonic with respect to all local coordinates belonging to the complex
structure of X. The notion of plurisubharmonicity is invariant with respect
to holomorphic coordinate transformations. So in order to prove plurisub-
harmonicity at some point of X it is enough to prove it with resepct to an
arbitrary coordinate system. However, here we wish to express the notion of
plurisubharmonicity in invariant terms. We do this by Hermitian forms.
Definition. A Hermitian form at Xo E X is a Hermitian form

The form H is called positive semidefinite if H (v, v) 2: 0 for all v, and it

is called positive definite if H(v,v) > 0 for v =F O.

A Hermitian form has a unique representation

n
H = L hijdzi ® azj,
i,j=l

where H := (h ij I i, j = 1, ... , n) is a Hermitian matrix; i.e., it satisfies the

equation H = H t. In the following we suppress the symbol ® and write
H = L:i,j hijdziazj'
262 V. Stein Theory

With respect to the local coordinates we can associate to any tangent vector
v= Li vi 8j8zi + Li vi 8 jOzi the corresponding vector v = (Vb .•. , V n ) E en
and write
n
H(v,w) = L hijViWj=v.H.w t .
i,j=l

Coordinate Transformations. Assume that F : G ~ B is a holo-

morphic map of domains in en given by equations
Wk = fk(Zl, ..• ,zn), for k = 1, ... ,no
If 9 is a differentiable function in B, by the complex chain rule it follows that

(goF)z. = ~)gwk of)· (fk)Zi and (goFh. = LC9wk of)· (ik)Zi'

k k

The transformation of a (contravariant) tangent vector (Le., a derivation) ~

is given by F ,,(~)[g] = ~[g 0 F]. This means that
n 8 n 8
F,,(~) = L~[Jk]a + L~[ik]Ow

t;n(n~~i' (fk)Zi )8 + t;n(n~~i'

k=l Wk k=l Wk

_ _ )8
= (ik)Zi Owk'
8Wk

or, if ~ "" e= (~1, ... , ~n), then

F,,~ "" (E~i' (ft)Zi"'" L~i' (fn)zi) = e· JJ.

i i

Now, a covariant tangent vector cp at wEB will be transported in the

opposite direction:
F"cp(~) := cp(F,,~).

In particular, F"(df)w(~) = (df)w(F ,,~) = (F ,,~)[f] = ~[J 0 F] = d(f 0 F)z(~)

for w = F(z). This gives us the formula
F"«df)F(z) = d(f 0 F)z·
Therefore we also write cp 0 F := F"cp for arbitrary covariant vectors cpo
If H is a Hermitian form at w = F{z) E B, then we can define a Hermitian
form F"H at z by
F"H(~,1J):= H(F,,~,F .. 1J).
If J = JF is the Jacobian of Fat z and H = Lk,l hk,l dWk dWI, then

F" H (~, 1J) = (e .J t) . H . ("7 . J t) t = e. (J t . H . J) . 11 t .

So F" H is given by the Hermitian matrix J t . H . J.
 2. The Levi Form 263

Plurisubharmonic Functions. Assume that p is a real-valued (~OO)

c c;n and wEB a point. We consider
differentiable function in a domain B
at w the Hermitian form

Hp = L a a2pa
k,l Wk W,
(W)dWkdwI'
_

given by the Hermitian matrix

If F : G -+ B is a biholomorphic transformation with F(z) = w, then a direct

calculation shows that

H(p 0 F, z) = JF(Z) t . H(p, w) . JF(Z),

Therefore, F· Hp will be described by the Hermitian matrix H(p 0 F, z), i.e.,

F·Hp = H poF .
Now let X be an n-dimensional complex manifold, p a real-valued smooth
function on X, and x E X an arbitrary point. Assume that cp : U -+ B c c;n
is a local coordinate system at x. Then a tangent vector ~ at x is uniquely
determined by a pair (cp,e), e E c;n.
We define the Hermitian form Hp : Tx(X) x Tx(X) -+ C; by

If t/J is another coordinate system at x and ~ '" (t/J,e), then

and

HpO",-1 (e, ij) = H(pO'l'-1 )0('1'0",-1) (e, ij)

= H pO'l'-1 «cpo t/J-l ).e, (cp 0 t/J-l ).ij)
= H pO 'l'-1(e,"1)·
We see that the definition of Hp is independent of the local coordinates.
Definition. Assume that p is a real-valued smooth function on the
complex manifold X. Then Lev(p)(x,~) := Hp(e,e) is called the Levi
form of pat x. It is the quadratic form on Tx(X) associated with Hp,
and it does not depend on local coordinates.
The function p is called plurisubharmonic on a subset M C X if the Levi
form of p is positive semidefinite at any point x E M. If the Levi form is
positive definite at every point, we say that p is strictly plurisubharmonic.
264 V. Stein Theory

Example
Assume that 11, ... '!m are holomorphic functions on X. We show that p:=
E~=l ik . 7k is plurisubharmonic in X. In fact, we have

Hp = LPZi,ZjdzicLzj
i,j

= f: (~(fk)Zi
k=l &,}
. (fk)zj dZi cLzj)

~ t, ((t,(t.).; dZi ) 0 (t, (J.).; dZ;) ) .

That means that

Lev(p)(x,~) = Hp(~,~) = L
m

k=l
IL(ik)Zi (x) ~i I 2: O.
i
2

The Maximum Principle. A nonconstant plurisubharmonic function

does not take on a maximum.
2.1 Theorem. Assume that A c X is a compact connected analytic set
and that P is a plurisubharmonic function on X. Then pi A is constant.

PROOF: We may assume that A is irreducible. If there is a point Xo E A

where piA takes its maximum value c, we shall prove:
(*) The function p is identically equal to c in a small neighborhood of Xo.
If we know (*), then we consider the set K of all points x E A with p( x) = c.
The open kernel KO is not empty. If KO =I A, there is a boundary point Xl
of K in A. The function p also takes its maximum at Xl' So by (*) it follows
again that p(x) = c in a neighborhood of Xl, which is a contradiction.
So we have only to prove (*). We may assume that A is an analytic subset
of en and Xo = O. If the codimension of A is equal to d, then there is an
(n - d)-dimensional domain G' c en - d with 0' E G' and pseudopolynomials
Wl (Zl j z'), ... , Wd (Zdj z') over G' such that A is an embedded-analytic subset
of the joint zero set of the Wi in a neighborhood of 0, and 0 is the only point
of A over 0'.
We take a ball BeG' around 0' and restrict everything to an arbitrary
complex line i C B through 0'. The restriction Ali is denoted by A'. Let Z
be a linear coordinate on i with origin 0' such that the embedding of i in B
 2. The Levi Form 265

is given by Z t-+ (ZI, ... ,Zn-d) = (alz, ... ,an-dz). The restriction Will may
contain multiple factors. We throwaway such superfluous factors, so we can
assume that every Wilt is free of multiple factors.
Let us first assume that the union of the discriminant sets of the pseudopoly-
nomials over i consists only of the point 0' and that A' is irreducible and has
s sheets over i. Then A' is the Riemann surface of VZ. We write A' in the
form
A' ~ {(t; aIZ, .. . ,an-dZ) : Z = tS}.
By F(t) := (t; alt S, ... , an_dtS) we have a local parametrization of A'. Then
(p 0 F)tf(t) = H(p 0 F, t) is given again by Jt . H(p, F(t)) . J, where J =
(1; aI, . .. , an-d) t denotes the Jacobian of the holomorphic map F. The proof
is the same as in the case of a biholomorphic map F. So (p 0 F)tf ~ 0, and
po F is a subharmonic function of t. We get p' := piA' == c on A'.
The same is true if A' is not irreducible but has 0' as the joint discriminant
set, since 0 is the only point of A' over 0'.
Now assume that the union of the discriminant sets is general. Every point in
A' can be connected with O. We introduce the subset K of all points x E A'
with p'(x) = c. If KO =I- A', there is a boundary point Xl of K in A'. We
know that A' is an embedded-analytic set. Then there is a neighborhood
U(XI) c A' that is embedded-analytic over a disk B'(ZI) C i around a point
Zl E B such that over Zl the only point of U is Xl and the union discriminant
set consists of Zl only. Then we get p'lu == c (by the same argument as above),
which is a contradiction to the property "boundary point." So p' == c follows.
This holds for all e, and therefore p == c over the whole ball B ,which is in a
full open neighborhood of Xo in A. So we have (*). •

Exercises
1. Assume that p is a real-valued smooth function on the complex manifold
e
X. If { '" is a tangent vector and cp a complex coordinate system at
Xo EX, then define

L(P
n
(op)xo({) := 0 cp-l)z,,(cp(xo))' {.., .
..,=1
Show that (op)xo : Txo --+ C is a complex-valued real linear form that
does not depend on the local coordinates. Prove the following formulas:
(a)

Lev(p· q)(x, {) = p(x)· Lev(q)(x, {) + q(x) . Lev(p)(x, {)

+ 2 Re (op)x({) . (Oq)x({)) .
266 V. Stein Theory

(b)
Lev(h 0 p)(x,~) = h"(P(x)) 'l(ap)x(~W + h'(p(x)) . Lev(p)(x, ~).
2. Let G cc X be a relatively compact domain with smooth boundary.
Show that there is an open neighborhood U of aG in X and a real-
valued smooth function <p on U such that GnU = {x E U : <p(x) < O}
and (a<p)x i= 0 for x E aGo Show that
Hx(aG) = {~ E Tx(X) : (a<p)x(~) = O}
is a well-defined subspace ofTx(X) that does not depend on the boundary
function <po
Show that if for every x E aG there is a local boundary function 7jJ such
that Lev(7jJ) is positive definite on Hx(aG), then rp can be chosen as a
strictly plurisubharmonic function.
3. Let G cc X be a relatively compact domain with smooth boundary,
and rp : U(aG) -T IR a global boundary function. If Lev(rp) has for every
x E aG at least one negative eigenvalue on Hx(aG), G is called pseudo-
concave.
Show that if X is connected and there is a nonempty pseudoconcave
domain in X, then every global holomorphic function on X is constant.

3. Pseudoconvexity
Pseudo convex Complex Manifolds. If X is an arbitrary complex
manifold, then there exists a sequence of compact subsets Ki E X with the
following properties:
1. The set K i - 1 is always contained in the open kernel (Kit of K i .
00

i=l
If X is holomorphically convex, then the Ki can be chosen in such a way that
the holomorphically convex hull

Ki = {X'E X : I/(x)1 ~ s;:;1/1 for all I E O(X) }

always equals K i . (One uses the same proof as for domains in en.)
Therefore, for every point x E X - Ki there is a holomorphic function I in
X such that I/(x)1 > 1 and III < Ion K i . By passing over to a multiple and
a power of I, we can make III arbitrarily small on Ki and arbitrarily big in
a fixed neighbor hood of x,
Since Ki+2 - (Ki+1t is compact, there are finitely many holomorphic func-
tions li(l), ' .. ,ft,) in X such that for Pi := 2:~~ll/t) 12 the following hold:
 3. Pseudoconvexity 267

l. sUPK Pi < 2- i .
~
2. Pi > 1 on Ki+2 - (Ki+d .
0

Of course, Pi is a nonnegative plurisubharmonic function in X.

The sum 2::1 Pi converges compactly on X to a nonnegative function P,
with {x EX: p(x) < c} CC X for every c > O.
If f is a holomorphic function on a domain G c en, and G' cc G a sub do-
main, then we have the Cauchy estimate

ID'" f(z)1 ::; OG~)I"'I . s~plfl, for z E G'.

Using this estimate in the intersection of a local coordinate system for X with
(Kit, one shows that all derivatives of 2:i Pi converge compactly in X to
the corresponding derivatives of p. So P is Cfj'oo and again a plurisubharmonic
function. One can even show that P is real-analytic (see [DoGr60]).
Definition. A complex manifold X is called pseudoconvex if there
exists a nonnegative smooth plurisubharmonic exhaustion function P on
X (Le., a Cfj'oo function P with {x EX: p(x) ::; r} CC X for all r > 0
such that the Levi form of P is everywhere positive semidefinite).
If we can find for P a strictly plurisubharmonic function in X, then X
is called stricly pseudoconvex or 1-complete 1 . If p is strictly plurisubhar-
monic only outside a compact set K eX, then X is called 1-convex or
strongly pseudoconvex (at infinity).
In the literature a pseudo convex manifold is often called weakly 1-
complete.

Above it was shown that

every holomorphically convex complex manifold is pseudoconvex.
We shall prove later on that every I-complete complex manifold is holomor-
phically convex, and even Stein (solution of the Levi problem). Also, strongly
pseudo convex manifolds are holomorphically convex. But in general this is
not true for weakly I-complete (Le., pseudoconvex) manifolds.

Examples. Strict pseudoconvexity (I-completeness) is one of the most

important notions in the analysis of complex manifolds. Many constructions
can be carried out only in the strict pseudoconvex case. Let us consider some
examples.
Example 1: The theory was inaugurated by the following result:
1 An n-dimensional complex manifold X is called q-complete if it has an exhaustion
function p such that at every point of X the Levi form of p has at least n - q + 1
positive eigenvalues.
268 V. Stein Theory

An unbranched domain Gover C n is strictly pseudoconvex if and only if

it is a domain of holomorphy.
For proofs see [Oka53], [Br54], [No54].

Example 2: Every unbranched Hartogs convex domain over C n is strictly

pseudoconvex. See [Ri68], where a smoothing procedure for strictly pseud~
convex functions is introduced.

Example 3: The proofs of the following statements are elementary.

Every compact complex manifold is pseudoconvex. The Cartesian product of
finitely many (strictly) pseudoconvex complex manifolds is (strictly) pseud~
convex. Any submanifold of a (strictly) pseudoconvex complex manifold is
(strictly) pseudovonvex.

Example 4: Assume that G c C n is a domain of holomorphy and that

A eGis an analytic hypersurface. Then G - A is Hartogs convex (and
therefore st(ictly pseudoconvex). For the proof we just take a Hartogs figure

in the unit polydisk pn = {t : Itil < 1 for i = 1, ... ,n} and a biholomorphic
mapping F : H -+ G - A. The mapping F extends to a holomorphic mapping
pn -+ G. If A = F-l(A) c pn ~s not empty, then it is an analytic hypersurface
in pn - H. Some lines L(t') = {t = (it, t') : tl E C} will intersect it in a
compact subset of pn - H; for other t' the intersection is empty (see Figure
V.I).

r ~' L{t')
A
H

Figure V.l. Hartogs convexity of G - A

There is a limit t~ with c: < It~1 < I such that the intersection of L(t~)
with A is not empty but in arbitrarily small neighborhoods of t~ there are
points t' for which the intersection is empty. If tEA n L(t~), then there is
a neighborhood U where A is given by a holomorphic equation f = O. The
function tl H f(tt, t~) has isolated zeros in U n L(t~), and by the theorem
 3. Pseudoconvexity 269

of Rouche tl H f(tl, t') also has to vanish at some points of un L(t') for t'
near to to. This is a contradiction.
Thus F(pn) c G- A, and the domain G - A is Hartogs pseudoconvex.

Example 5: Let A c IF be an analytic hypersurface. We know that there

is a homogeneous polynomial w such that

A = {(zo : Zl : ... : zn) E IP'n : w(zo, ... , zn) = o}.

Let w be homogeneous of order s. Then

n
p(zo: ... : zn) := log {( I>iZir /Iw(zo, ... , ZnW}
i=O
is a well-defined smooth exhaustion function for the affine algebraic manifold
IP'n - A. We calculate the Levi form in local coordinates tv = zv/zo, v =
1, ... , n using the properties of the logarithm. Since the Levi form of f + 7
vanishes for any holomorphic function f, it follows that

Lev(p) (to, e) = s· Lev(log(1 + IIt1l 2) )(to, e)

s· (1 + ~tloI12)2 ·I(to, eW + 1 + l~toll2 ,"eIl 2)

= (1 + 11:0112)2 . (lIell 2 + Cll t oll 2 . lie 112 -I(to, e}1 2)) ,

and this expression is positive for e =F O. So p is strictly plurisubharmonic

everywhere, and X = IP'n - A is I-complete. In this case we can show directly
that X is holomorphically convex:
Every function
n
f(z o·· ..... Z n ) '-
ZSO ••• zSn
0 n with LSi = S,
i=O
.- ( )'
W ZO"",Zn

is holomorphic in X, and the maximum of the absolute values of all these

functions tends to infinity as (zo : ... : zn) approaches A. So K cc X for
any subset K C eX. Consequently, X is holomorphically convex, and it is
even Stein, as one can see from the following theorem.
3.1 Theorem. Let X be a holomorphically convex connected complex man-
ifold that contains no compact analytic subset of positive dimension. Then X
is a Stein manifold.

PROOF: Let Xo E X be an arbitrary point. Then the set

A := {x EX: f(x) = f(xo) for every f E O(X)}

270 V. Stein Theory

is a closed analytic subset of X. Clearly, it is contained in the holomorphically

convex hull fxJ = {x EX: If(x)1 $ If(xo)1 for all f E O(X)}. Since ~
has to be compact, A is likewise compact. This is possible only if A consists
of isolated points. Then there exists an open neighborhood U = U(xo) and
holomorphic functions it, ... , f N in U such that

{xo}=AnU={xEU: h(x)=···=fN(X)=O}.

So X is holomorphically spreadable and therefore Stein.

-
3.2 Proposition. A I-complete complex manifold cannot contain compact
analytic subsets of positive dimension.

PROOF: Let p be a strictly plurisubharmonic exhaustion function in the

manifold X. Then p is plurisubharmonic, and by the maximum principle it
must be constant on any compact connected analytic subset A eX. If A has
positive dimension, then there is a point x E A and an open neighborhood
U = U(x) c X such that An U is a submanifold of U of positive dimen-
sion. The function plAnu is strictly plurisubharmonic and constant. That is
impossible. _

3.3 Corollary. Let X be a I-complete manifold that is holomorphically

convex. Then X is a Stein manifold.
At the end of this chapter we will see that the condition "holomorphically
convex" is not necessary.
3.4 Corollary. Let A c ]p>n be an analytic hypersurface. Then]p>n - A is a
Stein manifold. Every analytic subset B c ]p>n of positive dimension meets A
in at least one point.

Example 6: There is a famous theorem by H. Cartan:

A domain G C 1(:2 is a domain of holomorphy if and only if the first
Cousin problem is always solvable.
The solvability is also true for higher-dimensional domains of holomorphy,
but there is a greater class of domains with this property. Take, e.g., the
domain G c pn C I(:n (with n 2: 3) that is the union of the three open sets

U1 .- {z E pn IZll>g},
U2 .- {z E pn IZ21 > g},
U3 .- {z E pn I(Z3, ... , zn)1 < g}.
 3. Pseudo convexity 271

, ZI 31

Figure V.2. Cousin-I in a domain which is not Hartogs convex

The domain G = U 1 U U2 U U3 is not Hartogs pseudoconvex (see Figure V.2).

But all three covering elements Ui are domains of holomorphy. Therefore,
we have HI(Ui , 0) = 0 for i = 1,2,3, as will be proved in Section V.5. So
%' = {UI' U2, U3} is an acyclic covering for 0, and every Cousin-I distribution
in G can be given by a co cycle I E HI(%" 0), up to a coboundary. Such an
I consists of holomorphic functions lij in U ij with h2 + 123 + hI = 0 on
U 123 . This implies that the Laurent series of h2 (around the origin) in U l23
contains no powers zf z~ with i < 0, j < O. By the identity theorem this is true
on the whole set U12. Therefore, we can subtract a coboundary 8{gb g2, O}
from our co cycle such that thereafter h2 == 0.
Then the new 123, h3 coincide on U 123 • Together they give a holomorphic
function h in (U1 U U 2 ) n U 3 that extends holomorphically to U 3 • Then the
new cocycle I is equal to the coboundary 8(0, 0, h), and hence the old cocycle
also cobounds (Le., is a coboundary). Consequently the Cousin-I problem is
solvable.

Example 7: Assume that X is an n-dimensional complex manifold and that

Xo E X is a point. We can blow up X in Xo. Then we obtain an n-dimensional
complex manifold X, an (n - I)-dimensional complex submanifold A c X
that is isomorphic to jp>n-l, and a proper holomorphic map 1r : X ~ X that
maps A to Xo and X - A biholomorphically onto X - {xo}. We have a strictly
pseudo convex neighborhood U around Xo. We can lift the strictly plurisub-
harmonic exhaustion function p on U by 1r to 1r-l(U). This is a strongly
pseudoconvex neighborhood of A that is not, however, strictly pseudocon-
vex.

Example 8: For a similar example consider the analytic set

A:= {z E C 2n : ZlZn+j - ZjZn+1 = 0 for j =2, ... ,n}.

Outside of the origin A is regular of dimension n + 1. For example, if Zi =I 0,

then zn+j = Zj . Zn+i/ Zi for j = 1, ... , n. So A is there parameterized by
Zl, ... ,Zn and Zi+l. It follows that dim(A) = n + 1.
272 V. Stein Theory

c
We have a meromorphic map from 2n to the Osgood space on given by
Wi = zd Zn+i, i = 1, ... ,n. Its graph is the set

x = {(z,w) E C 2n X on : WiZn+i - Zi = 0 for i = 1, ... ,n}.

We denote by X the closure of the part of X that lies over A - {o}. Let
X -+ A be the restriction of the canonical projection from A x on
11" :
to A. Then 11" maps a 1-dimensional projective space onto 0 and the rest
biholomorphically onto A - {o}.
The space X is an (n + I)-dimensional complex manifold, locally given by
the equations
Zi = WiZn+i for i = 1, ... , n,
Wi = W1 for i = 2, ... , n.
The set E(s):= {(zb ... ,2n) E A : zdzn+l = ... = Zn/Z2n = s} is an
n-dimensional plane for every s E p1, and A is the union of all these planes.
Since E(S1) n E(S2) = {o} for S1 =f S2, A is singular at the origin. It follows
that X is a vector bundle of rank n over 11"-1 (0) ~ p1.
Now we use the function p(z) = Li ZiZi on c2n. It induces a strictly plurisub-
harmonic function on the complex manifold A - {o}. The (n + 1)-dimensional
complex manifold X is strongly pseudoconvex by po 11", but not 1-complete,
since it contains the compact analytic subset 11"-1(0).

Example 9: Consider the covariant tangent bundle T' of pn. If ~ is a global

holomorphic vector field on pn, then a plurisubharmonic function Pt; on T' is
defined by
pr;(wx ) := wx(~x) . wx(~x).
We consider local coordinates in a set Ui C pn, for example tv := zv/zo for
v = 1, ... , n in the case i = O. Then every w over Ui can be written uniquely
in the form w = Lv Wvdtv. So tl, ... ,tn, Wl, ... ,Wn are local coordinates in
T' over Ui' If ~ = L,,,f.v8/8zv, then

We have the following n + n2 holomorphic vector fields over Ui that extend

to pn:
8 8
'--8 = q(O, ... , Zo , ... ,0) and tP.-8 = q(O, ... , zp. , ... ,0),
tv --..,..., tv "-v-"
vth place vth place

where q : O(1)EB" 'EBO(1) -+ TPn is the canonical bundle epimorphism in the

Euler sequence. Plurisubharmonic functions pt, respectively ptp., are defined
in local coordinates by wvwv , respectively wvwv . tp.t,.. By adding them all we
obtain the plurisubharmonic function
 3. Pseudoconvexity 273

Pi = (t
v=1
wvwv) . (1 + t t,.,J".) .
".=1

Finally, if we add the Pi constructed for the various Ui , i = 0, ... , n, we

obtain a function p. If Z denotes the zero section in T', then P vanishes on
Z and is strictly plurisubharmonic and positive outside of Z. It tends to 00
for IIwll -+ 00. So the complex manifold T' is strongly pseudovonvex, but not
I-complete.

Example 10: This example is probably due to J.P. Serre.

Assume that E is an elliptic curve (Le., a compact Riemann surface of genus
1). Then E is a I-dimensional torus, given by a lattice of periods (1, e), where
the imaginary part of eis positive. We may write the elements of E as real
e,
linear combinations z = s· 1 + t . with s, t E [0,1]. The first cohomology
group HI (E, 0) is equal to C (this will follow from results of the next chapter,
but it is also a very classical result in the theory of Riemann surfaces).
We have a covering ~ of E consisting of the two elements

{z = s + te : 0 ~ s ~ 1/2, 0~t~1},
{z=s+te: 1/2~s~1, 0~t~1}.

Denote by C c E the circle {z = 1 + t . e:0 ~ t ~ 1} and define there

the function f == 1 (and 0 on the other component of Ul n U2 ). Then f
is a co cycle in ZI (~ , 0). It yields a nonvanishing cohomology class, since
otherwise we would obtain a nonconstant bounded holomorphic function on
C. We construct a fiber bundle A above E with fibers Ax = C by gluing
above C the point (z, w) E U2 X C with (z, w - 1) E U1 X C. We denote this
point in the bundle by [z, w].
Now, A is topologically trivial (since we can find a continuous function 9 on
U2 that coincides with f on C such that !5{0, g} = I), but it is not analytically
trivial, because it defines a nontrivial cohomology class. It follows from the
construction that the notions of real lines, planes, and convexity are well
defined in A.
Let A be the bundle with typical fiber pI that is obtained from A by adding
the point at infinity to each fiber. Then A is compact and has the infinite cross
section D = Doo over E. We put X := A = A-D. Then X does not contain
any compact analytic set of dimension 1. Otherwise, there would be a number
b such that the analytic set meets every fiber in exactly b points. We could
pass over to their barycenters and would obtain an ordinary holomorphic
cro~s section in A. This would imply the triviality of A.

We consider the real 3-dimensional surface

S = Sr = ([z, w] E A : z = s + te, w = S + rexp(21l'i . lJ) and 0 ~ lJ ~ I}.

274 V. Stein Theory

Here r > 0 is a very big number. The surface S bounds a tube G, which
converges to A as r tends to 00.
We look at the holomorphic tangent H in any point of S. We can consider H
as a real plane in A. By subtracting s in the fiber above s + t~ we pass over
from A to E x C. The hypersurface S is transformed diffeomorphically into
a convex cylinder over E and H into a contacting plane that does not enter
the interior of the cylinder. So H does not enter the interior of the tube G.
If H were contained in S, then the transformed H would lie in the boundary
of the cylinder and therefore be compact. So H itself would be a compact
analytic set. We saw that such a set does not exist. So there remains only
the possibility that H contacts S of first order; i.e., the intersection H n Sis
a real line. We can choose a complex coordinate ( = x + i y in H such that
our real line is exactly the x-axis.
If {! is a smooth defining function for S, then {! behaves on H like the function
y2. It follows that ({!iH )(~ > O. This means that the Levi form of {! is always
positive definite on the holomorphic tangent H. Now we can construct a
strictly plurisubharmonic function ~ in U - D (where U is a neighborhood
of D in .4) whose level sets are the manifolds Sr such that ~ converges to 00
when approaching D.2 It follows that X is strongly pseudoconvex.
Later on we shall prove that a strongly pseudo convex manifold is holomorphi-
cally convex, and since X contains no positive-dimensional compact analytic
subset, it is a Stein manifold. So there are many holomorphic functions in X.
Assume that there is a meromorphic function on A that has poles of order m
on D only. Then the coefficient of the highest polar part of f is a holomorphic
cross section 1J in the mth tensor power of the normal bundle of D. Because
this normal bundle is topologically trivial, 1J cannot have zeros on D. So f
tends to infinity approaching D, and no analytic set {f = canst} meets D.
This is a contradiction, since A is not analytically trivial. Every holomorphic
function on X must have essential singularities on D.

Analytic Tangents. Let G be a domain in en with n 2 and p a strictly

~
plurisubharmonic function in G. Denote by X the set {z E G : p(z) < O}.
Let B cc G be an open subset and w E aX n B an arbitrary point.
The expansion of p in w is given by

p(z) = pew) + 2ReQ(w,z - w) + Lev(p)(w,z - w) + R(w,z - w),

where
1
Q(w, h) = h· Vp(w) t + 2 h· Hess(p)(w) . h t,
2 We may assume that {! is globally defined and {! = T on Sr. Then we choose an
unbounded strictly monotonic smooth function h : R --+ R such that h" / h' is
very large. The function ~:= h 0 {! will do the job.
 3. Pseudoconvexity 275

with the complex Hessian Hess(p)(w) = (PZ,Zi (w) I i,j = 1, ... , n ), and
. R(w,h)
~~ IIhll 2 = O.

The map h H Q(w, h) is a holomorphic polynomial of degree 2, for every w.

The assignment w H Q(w, h) is smooth.
Let G' cc G be an open neighborhood of B. We can fl·,r' constants c, k > 0
such that

Lev(p)(w, z - w) + R(w, z - w) ~ kllz - wl1 2 cIIz - wll 3

for z E G' and arbitrary w.

We say that a real number c > 0 is sufficiently small with respect to B if
every ball U(w) with center w E aX n B and radius c belongs to G', and

is valid on these U ( w ).
Assuming that this is the case, for w E aX n B we define the analytic set

A(w) := {z E U(w) : Q(w,z - w) = O}

in U(w). On A(w) we have p(z) ~ O. Therefore, A(w) - {w} is outside of

X. Since w belongs to the boundary of X, this implies that A = A( w) has
codimension l. We call A an analytic tangent (or, in German, a StutzJliiche)
for X at w.
3.5 Proposition. Let G,p, X, B, G' be as above. There exists a differen-
tiable family A(w) of analytic tangents to aX at the points wE BnaX with
A(w) n X = {w}.
Here "differentiable" means that the defining quadratic polynomials for A(w)
depend smoothly on w.

Exercises
1. Let X be a complex manifold and Y C X a closed complex submanifold.
Construct a nonnegative smooth function f : X -+ lR with the following
properties:
(a) Lev(f)(x,~) ::::: 0 for every x E X, ~ E Tx(X).
(b) For every x E Y there is a linear subspace P x C Tx(X) such that
Px + Tx(Y) = Tx(X) and Lev(f) (x, ~) > 0 for ~ E P x , ~ f:. O.
2. Let X be a I-complete complex manifold and il, ... , fq E O(X). Show
that X - N(il, ... , fq) is q-complete.
3. Let Gee en be a strictly convex domain with smooth boundary. Con-
struct the differentiable family A(w) of analytic tangents to aG.
276 V. Stein Theory

4. Let 'P be a strictly plurisubharmonic smooth function defined in a neigh-

borhood U = U(O) c en with 'P(O) = 0, and 'I/J an arbitrary smooth
function in U. Prove that there are r > 0 and c > 0 sufficiently small
such that 'P + c . 'I/J is strictly plurisubharmonic in Br(O) and

is a Stein manifold.

4. Cuboids
Distinguished Cuboids. If Q is a closed subset in en, then in this
section we say that something is defined on Q if it is defined in a small
neighborhood of Q. Two objects are called equal on Q if they coincide in a
small neighborhood of Q. SO actually we consider "germs" along Q.
Definition. A cuboid is a closed domain

where ai < bi are real numbers for i = 1, ... , 2n.

If there are partitions

ai = aoi < ai1 < ... < am· b

i • = i, for i = 1, ... , 2n,
then we denote by A the system of sequences
j
ai' Z. -- 1, ••• , 2n, j=O, ... ,mi.

A closed covering %'A of Q is defined by the system of cuboids

Qjl •...•l2n = {z : a1,-1 ::; Xi ::; a1', for i = 1, ... , 2n}.

For any open covering ofQ there is a closed cuboid covering which is finer.
In the following it will be not enough to have a covering of Q. Additionally,
we need a system of complex submanifolds
Q = Xo ~ Xl ~ ... ~ X i- l ~ Xi ~ ... ~ X B, s::; n,
where Xi has dimension n - i, such that there is a holomorphic function Ii
in X i - 1 vanishing everywhere on Xi to first order (and maybe also vanishing
at points of X i - 1 - Xi).
Definition. A distinguished cuboid is a cuboid that is equipped with a
system {(Xi, Ii) : i = 1, ... , s}. The number n - s is called the manifold
dimension of the distinguished cuboid.
 4. Cuboids 277

Vanishing of Cohomology. As described above, everything on a

closed cuboid is assumed to be defined on an open neighborhood of the
cuboid. Therefore, we can consider co cycles and coboundaries with respect
to closed cuboid coverings.
4.1 Proposition. Assume that Q is a cuboid and c£rA a closed cuboid cov-
ering of Q. Then every cocycle ~ E Zl(c£rA , 0) cobounds.

PROOF: First we consider a very simple system A. We just take the case
where ml = 2 and all other mi = 1. Then

c£rA = {Qo = Ql,I, ... ,b Ql = Q2,1, ... ,d

has the minimal possible number of elements. The co cycle ~ is given by one
holomorphic function ~01 on Qo n Ql = {z E Q : Zl = Let c: be a small aD.
positive number. Then in the zl-plane we can choose two continuous paths
(Xj fromat at
+ i(a~+l - c:) to + i(a~+l + c:) (for j = 0 on the left side and
for j = 1 on the right side of the line Xl = at,
see Figure V.3) and define
11 = {110, 111} by the Cauchy integral

.(
11) Zl,···, Zn·-
)._-1-1211"1
.
O:j
~Ol (w, Z2, ... , Zn) d w.
W - Zl

Then we get 811 = ~.

~ 21-plane

_ _ ----i--- /
Ql

<: I 111'-. ____

I I -----

l----L---_ --- ---

I :

------~-
/ ~ --""-- - - - - - ---------
---._-.
-------- ~..
Qo
...
-._._..... Z2, ... , Zn
Figure V.3. Cohomology of a cuboid -.-__ ,.

The next step is an induction on the number m = l:i mi. We just han-
dled the case m = 2n + 1. Now we assume that m > 2n + 1 and that the
proposition already has been proved for any number smaller than m. We put
s: s:
io := max{ i : 1 i 2n and mi ~ 2} and define
278 V. Stein Theory

By the induction hypothesis, we have 'I'ft", = 6(11') and '1'ft"1I = 6(TJ") with
cuboid coverings %" for Q' and %''' for Q". Hence, , - 6{ TJ', TJ"} is a cocycle
with respect to the covering %' that is given by a holomorphic function 9 on
the surface
Q , n Q " = { z E Q : Xio = aio
mi-l}
0 •

As above, by integration we obtain holomorphic functions TJo on Q' and TJl

on Q" with 9 = TJl - TJo· So we have a cochain a = {TJ' + 1}0, TJ" + TJd on Q
with 6a = , (since the functions TJj are already cocycles). This completes the
induction. _

Vanishing on the Embedded Manifolds. We consider a distin-

guished cuboid Q in en
with manifold dimension n - 5, and for i = 0, ... ,5
we prove the following:
4.2 Proposition. Every cocycle of Zl(%'A n Xi, 0) cobound5, and every
holomorphic function on Xi+! can be extended to a holomorphic /unction on
Xi'
PROOF: We carry out an induction on i. In the case ·i = 0 the proposition
has been proved already. The induction hypothesis now states that it holds
for some i, 0 SiS 5 - 1, and we prove the extension property for i + 1.
Assume that 9 is the holomorphic function on Xi+!. Define g(z) = 0 on
the other connected components of N(Ji+d C Xi. If the covering %'A is
sufficiently small, we can extend this 9 to a cochain TJ E CO(%'A n Xi, 0).
The coboundary 6(TJ) vanishes on N(Ji+l). Therefore, there is a cochain a E
Cl(%'A nXi,O) with 6(TJ) = a· fi+l. It is clear that a is a cocycle, and by
the induction hypothesis there is a 'Y E CO(%'A n Xi, 0) with a = 6("(). Since
6(TJ - fi+l . 'Y) = 0, we get 9 = TJ - fi+l . 'Y as a holomorphic extension of g.
Now we prove that any' E Zl(%'A n X i + 1 , 0) cobounds. As in the proof of
the preceeding theorem we have to show this oLly in the case where ml = 2
and mi = 1 for i > 1. So, is simply a holomorphic function 9 on (Ql,l, ... ,l n
Q2,l, ... ,dnXi+l. We have to find holomorphic functions /' on Ql,l, ... ,l nXi+l
and f" on Q2,l, ... ,l n Xi+! with f" - f' = g. For that we first extend 9 to
(Ql,l, ... ,l nQ2,l, ... ,l) nxi , construct /', f" for Xi (induction hypothesis), and
then restrict them to Xi+l. That completes the proof. -

Cuboids in a Complex Manifold. We assume that X is an n-

dimensional complex manifold and p a smooth real-valued function on X
with p(x) -t 2 when x -t 8X.3 We assume further that p is strictly plurisub-
3 This means that for every e > 0 there is a compact subset K C X such that
p(x) > 2 - e for EX - K.
 4. Cuboids 279

harmonic on {p(x) = I} and define Y := {x EX: p(x) < I}. Then Y is a

relatively compact strongly pseudo convex open subset of X.
For an n-tuple z = (Zl. ... , zn) we use the norm

jzj := supmax(jRe(zi)j, jIm(ziD.

i

For an open neighborhood G C X of BY we want to apply the results from

the end of V.3. (It does not matter that there are no global coordinates on
G.) We consider a relatively compact open neighborhood B = B(BY) CC G
and choose a G' with B cc G' cc G.
For any point Xo E BY there is a compact cuboid U· with center Xo in
a coordinate neighborhood S around Xo. If a real number c: > 0 is given,
we can choose every U· so small that U· C U (w) C C S for every point
w E U*, where U(w) is the ball with center w and radius c: (with respect
to the local coordinates). Then for every w E U* n BY we have the analytic
tangent A(w) given by a quadratic polynomial fw in U(w). It follows that
A(w) n U· n Y = w, and the function pjU· n (A(w) - z) is positive. This is
illustrated by Figure V.4.
For an open subset Y' cc Y we have the following proposition.
4.3 Proposition. There is a distinguished cuboid Q. = U* x Q c eN with
the following property:
For s = N - n the submanifold Xs is projected biholomorphically onto a
compact set U' C U* with Y' n U* c U' c Y.

A(w)
U·
BY

&Y'

Figure V.4. Projection of the distinguished cuboid

PROOF: We may work in en and assume that Xo is the origin. We use the
differentiable family of analytic tangents A(w) = N(fw) with w E BY n U*.
280 V. Stein Theory

Take a positive number r so big that we always have 11/ fw(z)1 < r for
w E 8Y n U* and z E Y' n U*. In a small neighborhood of w we have
Il/fw(z)1 > r. Therefore, by compactness there are finitely many of these
fw, say It,···, fs, with maxd1/ h(z)1 > r for all z E 8Y n U*. For a fixed i
we denote by Q~ the cuboid {z' = (Zn+ 1, ... , Zn+i) : Iz'l ~ r} and by Q~' the
cuboid {z" = (Zn+i+l"'" zn+s) : Iz"l ~ r}. We put Q* = U* x Q~ X Q~'.
The sub manifolds Xi C Q* are obtained in the following way: Consider the
graph of the i-tuple (1/ It, ... , 1/ fd in U* x Q~, take the union of those
connected components that contain points over Y, and multiply this union by
Q~'. The manifold Xi has dimension N - i. In X i - 1 we have the holomorphic
function Zn+i 'h(z) - I, which vanishes on Xi to order 1.
Finally, the projection U' of Xs contains no point of 8Y. Hence, it is contained
in Y. Since U* nY' is contained in U', the proposition is proved. _

4.4 Corollary. In the above notation, H1(U', 0) = O.

Enlarging U'. We use the same notation as before and construct a set
fj' C U* n Y that is bigger than U' where the vanishing theorem still holds.
For that purpose we take an open set Y' instead of Y' with Y' cc Y' cc Y
and U' C Y'. We need a bigger r> r such that we still have 11/ fw(z)1 < ron
Y' for wE 8Ynu*. But now maxdl/ h(z)1 > rno longer is true on 8YnU*.
So we add some functions to the old ones, fsH,"" Is, such that we get the
old situation again.
We get the Q~ and the Q~', for i = 1, ... , s. But we write Q~ instead of Q~, Q~'
for Q~', and Xi for Xi. Then we have Q* = U* x Q~ X Q~' c eN with N < N.
The projection of Xs is a compact set fj' C Y n U* with Y' n U* c fj'. So
again we have the vanishing of the cohomology.
The following statement is proved in the next paragraph.
(*) Every holomorphic function on U' can be approximated arbitrarily
well by holomorphic functions on fj'.
Suppose all cochains are given with respect to a covering %' of fj' ~ Xs-. If
e E Zl(%" e e
0) is a cocycle over fj', we have = 8('/]) over U' and = c5(if)
over fj', with cochains '/] and Ti with respect to the covedngs Un U' and un fj' .
Then Ti - ", is a holomorphic function over U', which we approximate by a
sequence of holomorphic functions fiJ E O(fj'). We may replace the cochain
if by ifj := if -?iJ and obtain that Tij approximates,/] over U'.
Now we can prove the following theorem.
 4. Cuboids 281

4.5 Theorem. If Xo E aY and U· is a small cuboid in a local coordinate

system around Xo, then HI(y n U*, 0) = O.

PROOF: Take a fixed open covering <P/ of Y n U· and exhaust Y by a

sequence of domains Yi with Yi c c Yi+ lee Y. We get a sequence of compact
sets U: C U· n Y with Yi n U· C U:.
If eE Zl (%' , 0) is a cocycle, for every i we get a cochain "Ii on U: such that
e= 8("Ii) on U;' As we have seen above, we may assume that the difference
of "IHI and "Ii on U: is as small as we want. Then the sequence "Ii converges
to some T] E CO(%" 0) with o(T]) = e. •
Approximation. We first prove the following simple result.
4.6 Proposition. Assume that Q is a cuboid in eN. Then any holomorphic
function f on Q can be approximated arbitrarily well by polynomials.

PROOF: We make the induction hypothesis that f is already a polynomial

in the variables z' = (Zl, ... , zs). Let "f" be the boundary of the rectangle with
corners (as+l-E,aN+s+l-E), (bs+1 +E,aN+sH-€), (bs+l +E,bN+s+l +€),
and (as+l - E, bN+s+ 1 + E) in the (XsH, XN+s+l)-plane. We have the integral
representation

fe z' ,Zs+l,Z") = -1-1

2'
71"1 'Yo
f(z',
~
<, -
e, Zll) d~<,.
Zs+l

It is necessary to approximate 1/(e - zs+d with a fixed E 1"1,,1 by a poly-e

nomial. We consider only the case where Re(~) = bs + 1 + E. In this case there
exists a disk D (with a center far to the left) such that E aD and the e
rectangle with corners (asH, aN+s+d, (bsH,aN+s+t), (b sH , bN+sH), and
(asH, bN+s+1 ) is contained in D. Then we expand l/(~ - zs+t) into a power
series on the disk and get the desired approximation. So the induction step is
complete. Since nothing remains to be shown for s = 0, the proof is finished .
•
We have to consider the cuboids Q = U· x Q~ and Q = U· x Qi. For this
denote by M the Cartesian product of the last s-
s factors of the Cartesian
P!oduct Qi. It follows that Q~ x M c Qi, and S := (U· x Q~ x M) n
Xi is projected bihoiomorphically onto Xs and onto U'. Every holomorphic
function on S is the restriction of a holomorphic function in U* x Q~ x M
and can be approximated arbitrarily well l?Y polynomials, in particular by
holomorphic functions on the larger set U* x Qs' Therefore, every holomorphic
function on U' can be approximated by holomorphic functions on 0', and we
have proved (*).
282 V. Stein Theory

Finally, we wish to replace U* by its open kernel (U*t. For this we exhaust
U* by relatively compact cuboids U; with U; cc U;+1' and choose the
s
functions /i, i = 1, ... , (from the "enlarging" process), independently of j.
Then we can approximate cochains on Uj C U; by those on Uj C U;, and
we can approximate these again by cochains on U;+l. So for any l-cocycle ~
on (U*t n Y we obtain (as a limit) a cochain 7] on (U*)O n Y with 07] = ~.
Then we have the following theorem.
4.7 Theorem. If Xo E 8Y and U* is a small cuboid in a local coordinate
system around xo, then H1(y n (U*t, 0) = O.

Exercises
1. Let G c c en be a strictly convex sub domain with smooth boundary.
Use coverings with distinguished cuboids to prove that H1(G, 0) = o.
2. Prove for the same domain G that every holomorphic function in G can
be approximated by polynomials.

5. Special Coverings
Cuboid Coverings. Again let X be an n-dimensional complex manifold
andp a smooth real-valued function on X withp(x) -+ 2 for x -+ 8X.Assume
that p is strictly plurisubharmonic on {p(x) = I}, and let Y := {x EX:
p(x) < I}. We also consider open sets Y = {x EX: p(x) < 1 + a}, where
Q

o $; a < c and c is a very small positive number such that also Y is strongly
Q

pseudoconvex.
Since 8Y is compact, it can be covered by finitely many cuboids Ut around
points Xi E 8Y which are always contained relatively compactly in a local
coordinate neighborhood such that Xi is the origin of the coordinate system.
If U; = {z : Izl $; Ti} and 0 < ti < Ti, then we may also assume that the
open sets Ui = {z : Izl < til cover 8Y. By adding more cuboids, the ti can
be chosen as small as necessary.
We also need these properties for the sets 8Y and for sets 8Y, where Y is
Q ,

an open set between Y and Y For that we have to move the centers of the
Q •

cuboids a little bit. If a is very small, the Ui still cover 8Y.

There may exist a compact set K c Y that is not yet covered by the Ui ·
Then we construct a covering of K with additional cuboids Ui C C Ut c C Y.
Finally, we have the following result.
5.1 Proposition. For a small c > 0, fOT any a with 0 $; a < c arid any
open set Y with Y C·Y c Y there exist coverings %'* = {Ut : i = 1, ... , m}
Q
 5. Special Coverings 283

and au = {Ui : i = 1, ... , m} of Y with Ui cc Ut such that the following

hold:
1. Every Ut is a compact cuboid in a local coordinate system around a point
Xi that corresponds to the origin in this coordinate system. The set Ut
is small enough in the sense of Sections V.3 and V.4 to have families of
analytic tangents in Ut.
2. There are numbers 0 < ti < ri such that

in the chosen coordinates.

3. For each i , either Xi E Y or Ut c Y .
4· For each i, the "star" Si = Uj : Uj nUd f2J Uj is contained in Ut ·

We call (au, au·) a special pair of coverings for (Y, Y) (cf. Figure V.5).

Figure V. 5 .Special pairs of coverings

The B u h hIe Method. Let a small e > 0 be fixed in the sense of the pre-
ceding paragraph. We choose a special pair (au, au·) of coverings for (Y, V).
Since Hi (Ui n Y,O) = 0, it follows that %' n Y is acyclic, and every coho-
mology class of Hi (Y, 0) can be given by a cocycle ~ E Zi (%' nY, 0).
Let Xo E BY be an arbitrary point and i an index such that Xo E Ui . Then
we have a cochain TJ = {TJj} E CO(au n Ut nY, 0) with 6(TJ) = ~1(Ut n Y).
We change rlj to 0 if Uj r:t. Si . Then 17 is a cochain over Y, and the cocycle
( - 8(TJ) represents the same cohomology class as f But ~ - 8(17) vanishes on
UinY.
Now we use the bubble method. By altering the strongly plurisubharmonic
function p a little bit on some compact subset of Ui (let us say by subtracting a
smooth function with compact support) we enlarge Y a little bit to a strongly
pseudoconvex open subset Y ::> Y with Y - Y cc Ui . Then every cohomology
284 V. Stein Theory

class represented by some ~ E Zl(%, nY, 0) extends to Y. SO the restriction

HI (Y, 0) --7 HI (Y, 0) is surjective. We assume that for Y (after a small
movement) the conditions for Ui , Ut are still satisfied.
In the next step we go to another Xo E Y and apply the procedure to Y.
Continuing in this way, after finitely many steps we have the following result.

5.2 Proposition. There is a fundamental system of strongly pseudoconvex

neighborhoods Yo of Y such that the restriction HI(yo, 0) --7 HI(y,O) is
surjective.

Frechet Spaces. The following theory can be found in the famous multi-
graphed notes [CS53J. See also [GuR065J.
Assume that V is a complex vector space.

Definition. A seminorm in V is a map P : V --7 {x E IR : x;:::: O} with

the following properties:
1. p(a· f) = laJ . p(f) for a E C, f E V.
2. P(fl + h) :::; p(fd + p(12), if fl' 12 E V.

If a sequence of seminorms Pi, i EN, is given in V, we can define a topology

in V. If e > 0 and mEN, we call the set

u~m)(o) = {f E V : Pi(f) < e for i = 1, ... ,m}

an (E, m)-neighborhood of O. As m increases the (e, m)-neighborhoods, of

course, become smaller.
A subset W c V with 0 E W is called a neighborhood of 0 if there is an
(e,m)-neighborhood U~m)(O) C W. If fo E V is an arbitrary element and W
a neighborhood of 0, then

fo +W := {fo +f : fEW}

is a neighborhood of fo. Finally, a set MeV is called open if it is a neigh-

borhood of any f E M.

Definition. A sequence (fk) in V is called convergent to f if for any

neighborhood W(f) there is a number ko such that fk E W for k ;:::: ko-
The sequence (fl.,) is called a Cauchy sequence if for all e > 0 and mEN
there is a number ko such that Pi(fl - fk) < e for i = 1,.:., m and
k,l;::::ko:
 5. Special Coverings 285

Definition. A Prechet topology on a complex vector space V is a topol-

ogy defined by a sequence (Pi) of semi norms, with the following proper-
ties:
1. For every f E V - {O} there is an i with Pi (f) > 0; i.e., the topology
is Hausdorff.
2. The space V is complete in this topology: Every Cauchy sequence
converges.
A Prechet space is a complex vector space with a F'n~chet topology.

Every Frechet space possesses a metric that defines its topology.

The vector space VeX) of holomorphic functions on a complex manifold X
is a Frechet space. We just take an increasing sequence of compact subsets
Ki with Ki c (Ki+lt and U Ki = X, and we define Pi(f) := suplf(Ki)l·
If (fk) is a Cauchy sequence of holomorphic functions on X, then fkl(Kit
converges uniformely to a holomorphic function on (Kit. Hence (Ik) con-
verges compactly to a holomorphic function f on X. This means that (Ik)
converges to f in the topology of VeX).
Since every Frechet space V is a metric space, a subset MeV is compact
if and only if every sequence (fk) in M has a convergent subsequence with
limit in M.

Definition. A linear map v : E --t F between two Frechet spaces

is called compact or completely continuous if there is a neighborhood
U(O) C E such that v(U) is relatively compact in F.

Obviously, every completely continuous map is continuous.

We have the following famous theorem of Schwartz.
5.3 Theorem of L. Schwartz. Assume that E, F are two Pr&het spaces
and that u, v : E --t F are two continuous linear maps with the following
properties:
1. u is surjective.
2. v is completely continuous.
Then the quotient F / (u + v) (E) has finite dimension.
A typical example of a completely continuous map is given as follows:
Let X be a complex manifold and Y cc X an open subset. Using the
increasing sequence of compact subsets from above, we have Y C Ki for
some i. Now, U := {J E VeX) : Pi(f) < I} is a neighborhood of 0 in
E:= VeX). Let v : E --t F := V(Y) be defined by v(f) := fly. Then

v(U) C ~ := {g E V(Y) : {;uplgl ::; I}.

286 V. Stein Theory

Since PA is closed and bounded, it follows from Montel's theorem that it is

compact. This shows that the restriction map v is completely continuous.
Remark. It is clear that we can replace the space of holomorphic functions
by the space of holomorphic cross sections in an analytic vector bundle.

Finiteness of Cohomology. We wish to apply Frechet space theory

to our standard situation.
Let X be an n-dimensional complex manifold and Y = {x EX: p(x) < I},
where p is a smooth exhaustion function on X with sup(p) = 2. Assume, as
usual, that p is strictly plurisubharmonic on a neighborhood of BY. Let a
small real number E > 0 be fixed such that Ya is strongly pseudoconvex for
o :s: fr < E. We have shown that the restriction map Hl(Ya, 0) --* Hl(y, 0)
is surjective for sufficiently small fr.
We start with a special pair (tf/, tf/*) of cuboid coverings for (Y, Y",), choose
numbers ~ with ti < ~ < ri and define fJi = {z : Izl < ~}. This gives a
covering q;-
= {fJi : i = 1, ... , m} of Y a.

The spaces CO(tf/ n Y,O), Zl(tf/ n Y,O), and Zl(q;-n Ya,O) are Fr~chet
spaces, as is Zl(q;- n Ya , 0) EB CO(%, nY, 0). Since the cohomology can be
extended from Y to Ya , it follows that the map

with u(~, 1]) := ~lo//nY + 6(1]), is surjective. The map v : E -+ F is defined

as the negative of the restriction map from Zl(q;-n Ya , 0) to Zl(tf/ nY, 0),
Le., by v(~, 1]) := -~lo//nY. Then v is completely continuous.
The map u + v is the coboundary map 6 : CO(tf/ nY, 0) --* Zl(tf/ nY, 0),
since the first summand goes to o. So the quotient F/(u+v)(E) is the first co-
homology group of Y. Its complex dimension is finite by Schwartz's theorem.
Thus we have proved the following result.
5.4 Theorem. If Y is a strongly pseudoconvex relatively compact subset of
a complex manifold X, then dime Hl(y, 0) < 00.

Holomorphic Convexity. We consider the same situation as above

and choose an fr > 0 such that Y c C Ya and:
1. dime Hl(Ya, 0) < 00.
2. For every x E BY we have an analytic tangent A(x) c Y", to BY, with
A(x) n Y = {x}, given by a holomorphic function fx in a neighborhood
U of A(x).
 5. Special Coverings 287

Let x E ay be an arbitrary point, f = fx, A = A(x), and U a suitable

neighborhood of A. Then %'X = {U nYo, Yo - A} is a covering of Yo consisting
of only two elements. So Zl(%,x, 0) = O(UnYo -A), and every function f- k
belongs to Zl(%,x, 0). Since H1(Yo, 0) is finite-dimensional, for m» 0 there
are complex numbers a1, ... , am, not all zero, such that the cocycle given by
the function
9 = (a1 . f- 1 + ... + am· f-m)I(UnY",_A)
is cohomologous to o. Therefore, we find holomorphic functions hl in Yo - A
and h2 in Un Yo such that h1 - h2 = g. The function

h ._ { hl on Yo - A,
.- h2 +g onUnYo ,

is meromorphic in Yo with poles in A. So hlY is a holomorphic function, and

for y -t x it tends to 00.
If K is compact in Y, then the holomorphic convex hull K does not approach
x. This is true for every x E ay. So K is compact. As a consequence we have
the following result.
5.5 Theorem. If Y is a strongly pseudoconvex relatively compact subset of
a complex manifold X, then Y is holomorphically convex.
Moreover, we have a solution of the Levi problem in this special case.
5.6 Theorem. Assume that Y = {x EX: p(x) < I}, where p(x) is a
smooth function in X with p(x) -t 2 for x -t ax
that is strictly plurisubhar-
monic on ay. If Y contains no higher-dimensional compact analytic subsets,
then Y is a Stein manifold.

PROOF: In Section V.3 we showed that every holomorphically convex man-

ifold that contains no compact analytic subset of positive dimension is a Stein
manifold. _

5.7 Corollary. If p is a strictly plurisubharmonic exhaustion function of

X, then Y = {x EX: p(x) < I} is a Stein manifold.
This is clear, since under the assumption there are no compact analytic sub-
sets of positive dimension in X.

Negative Line Bundles. Assume now that Z is a compact n-dimensional

complex manifold and that F is a holomorphic line bundle on Z. Let
11" : F -t Z be the canonical projection.

Definition. The line bundle F is called negative if there is a strongly

pseudo convex neighborhood Y cc F of the zero section of F.
288 V. Stein Theory

We identify Z with the zero section in F, and we call Y a tube around Z.

For an arbitrary holomorphic vector bundle V of rank s on Z we consider the

pullback bundle 1I"*V on F. If U c Z is an open set, then the holomorphic
cross sections in 1I"*V over 1I"-1(U) are holomorphic maps f : 1I"-I(U) ~ V
such that for any z E U the restriction of f to the fiber Fz has values in Vz .
The trivial line bundle on F is denoted by OF. Its local holomorphic cross
sections are the local holomorphic functions on F. If Vlu ~ U x Cs, then
the holomorphic cross sections of 1I"*V over 1I"-I(U) are s-tuples of local
holomorphic functions. Since the cocycles in our theory are given by lo-
cal holomorphic functions, we may replace OF by 1I"*V on F and get the
same results for X = F and the strongly pseudo convex set Y. So we have
dime HI(y, 11"* V) < 00.

Among the holomorphic maps f : 1I"-I(U) ~ V that correspond to sections

in 11"* V we have the maps f that on every fiber Fz are homogeneous of degree
m (with values in Vz ). We denote the space of these maps by Om(1I"-I(U), V).
It can be identified with the space of holomorphic cross sections in the tensor
product V ® F-m (see Section IV.2).
We choose a finite open covering ~ = {Ui : i = 1,2, ... , l} of Z. Let q;-be
the covering of F given by the sets Ui = 1I"-I(Ui ) We denote by Z;,.(€, 11"* V)
the vector space of co cycles ~m = {~ij} with ~ij E Om(1I"-I(Uij), V). These
are homogeneous of degree m on the fibers of F. We also call them cocycles of
degree m. Every finite sum ~ = Li ~i of co cycles of distinct degrees mi is con-
tained in ZI(F, 11"* V) and can be restricted to ZI(y, 1I"*V). It is a coboundary
if and only if all ~i are coboundaries. Since the cohomology HI (Y, 11"* V) is
finite, we obtain the following theorem:
5.8 Theorem. Assume that F is a negative line bundle and V an arbi-
trary holomorphic vector bundle on the compact manifold Z. Then there is
an integer mo such that Hl(Z, V ® F- m ) = 0 for all m 2:: mo.

Remark. A line bundle F on Z is called positive if its complex dual F' is

negative. Since (F')' = F, the above result can be reformulated as follows:
If F is a positive line bundle on Z, then Hl(Z, V ® Fm) = 0 for m 2:: mo.

Bundles over Stein Manifolds. Now we again consider a general

n-dimensional complex manifold X and assume that we have a strictly
plurisubharmonic function p in X with p(x) ~ 2 when x ~ aX. Then
Z = {x EX: p( x) < I} is called a special Stein manifold.
We use similar methods to those in the last paragraph. Now let F be the
trivial line bundle X x C on X and the strictly plurisubharmonic function p
 6. The Levi Problem 289

on F be defined by
p(x, w) := p(x) + wW.
The manifold Y = {(x, w) E F : x E Z and p( x, w) < I} is a strictly
pseudoconvex tube around the zero section in Flz. However, Z is not compact
now; it is only relatively compact in X. The tube extends to a neighborhood
of Z. The boundary of the restriction to Z has an "edge" BY n (B Z x C).
But the proof of finiteness of cohomology goes through in the same way as if
there were no edge. We have to apply the method of analytic tangents to BY
and {)Z simultaneously. If V is a holomorphic vector bundle on X, we denote
the pullback of V to F by 8. It then follows that dime HI (Ylz, 8) < 00. But
in this situation all the tensor powers of F are again the trivial line bundle.
We have V ® F-m !2::! V for every m. From this we conclude the following:
5.9 Theorem. If Z is a special Stein manifold (as described above) and V
a holomorphic vector bundle on Z, then HI(Z, V) = O.

Exercises
1. Let E be a Frechet space and let U = U(O) c E be a neighborhood such
that U is compact. Prove that E is finite-dimensional.
2. Let u : E -+ F be a continuous linear map between Frechet spaces. Prove
that if u(E) is finite-dimensional, then u is compact.
3. Let E be a Frechet space. A subset M C E is called bounded if for every
neighborhood U = U(O) c E there is an ro E 1R. such that Mer· U for
every r ~ ro. Show that a compact set K C E is closed and bounded. Let
X be a complex manifold. Show that every closed and bounded subset
K C O(X) is compact.
4. Let X be an n-dimensional Stein manifold and Y C X a closed subman-
ifold of dimension n - 1. Prove that the line bundle N x (Y) is positive.
5. Let F = O( -1) be the tautological bundle over 1P'2. Prove that F is
negative and that

6. Let X C 1P'3 be a regular hypersurface and i : X y JP>3 the canonical

injection. Prove that HI(X, i*O(k)) = 0 for k E Z.

6. The Levi Problem

Enlarging: The Idea of the Proof. We wish to show that every
strongly pseudoconvex (i.e., I-convex) manifold is holomorphically convex,
and that every strictly pseudoconvex (i.e., I-complete) manifold is Stein.
This solves the Levi problem for complex manifolds.
290 V. Stein Theory

Recall that an n-dimensional manifold X is called strongly pseudoconvex if

there is a smooth exhaustion function p on X that is strictly plurisubharmonic
on {x EX: p( x) ~ I}.
For s ~ 1 the open sublevel set X s = {x EX: p( x) < s} is holomorphically
convex, and its cohomology with coefficients in 0 is finite. We will construct
an open set X with Xs cc X by extending Xs in several steps such that every
holomorphic function on X s can be approximated by holomorphic functions
onX.

In the first step we choose a point Xl E {x EX: p( x) = s}, a local coordinate

system around xl, a sufficiently small compact cuboid Ui around Xl in that
coordinate system, and an open concentric cuboid UI cc (Uit such that
every analytic tangent A(xo) with Xo E [It - Xs is defined in Ui.
By disturbing p a little bit in a neighborhood -of XI. we get a new function
P2 and a bigger set X2 = {P2 < s} with Xs c X 2 and Xl E X2. We do this
carefully enough to get X 2 strongly pseudoconvex again. Since everything
happens in a cuboid, we can apply the theory of Section V.4 and obtain the
desired approximation property for functions in Xl = Xs and X2. Then we
repeat this process and construct a bigger set X 2 , and so on.
Since X s is compact, finitely many set X I, X 2, X3, ... , X N already cover
X s, and we order them such that XHI is obtained bI enlarging Xi in a
neighborhood of some point Xi E Xs' Finally, XS CC X := X N , and every
holo!D-orphic function on Xs can be approximated by holomorphic functions
onX.

Enlarging: The First Step. We choose an s ~ 1. Then p is strictly

plurisubharmonic in a neighborhood of f)X s , and also on X - Xs' We put
axs = {x EX: p(x) = s} and get 8Xs c axs .
Assume that Xl E axs is a point that is the origin in a local coordinate
system, and that Ui = {z : Izl ::; r} is a compact cuboid in that coordinate
system. Furthermore, let UI CC Ui be a concentric open cuboid such that
every analytic tangent A(xo) with Xo E U I - Xs is defined in Ui. On A(xo)
we have p(x) > s outside xo, and there is an e > 0 with p(x) > s + e on
8Ui n A(xo)· We assume that p is strictly plurisubharmonic on Ui.
6.1 Proposition. We can add to p a function h ::; 0 with support U I such
that h(xt} < 0, p +h is still strictly plurisubharmonic, and the following
approximation property is satisfied:
If X := {x EX: p(x) + hex) < s}, then every holomorphic function {or
cross section in a holomorphic vectorbundle} over Xs can be approximated
over Xs (i.e., on every compact subset of Xs) arbitrarily well by holomorphic
functions {respectively holomorphic cross sections} over X.
 6. The Levi Problem 291

PROOF: Assume that K c Xs is compact. We choose an R » 0 and con-

sider the family A(x) of analytic tangents to the points of U 1 n axs . There
is a finite set of points Yi, ... , y;,., E U 1 n axs such that for the defining
functions It of A(yn and the corresponding vector g* := (II Ii, ... , 1/1;'1)
we have the estimate

Ig*(x)1 := maxl1/ ft(x)1 > Ron axs n U 1.

1

Now we replace the points Yi by points Yi very near to yi, with p(Yi) > S,
and the functions It by Ii (the defining function for A(Yi)) such that we still
have Ig(x)1 > Ron axs nUl, for g:= (l/h, ... ,l/Im,).
We choose Rand m1 so big that Ig(x)1 < R in K n Ui.
Now we add a smooth function h with support U 1 that is strictly negative on
Ul . We let X := {x EX: p(x) + hex) < s} and may assume that the points
Yi are not contained in X and that Ig(x)1 > R on X - XS. We can make h
and its derivatives so small that P + h is still strictly plurisubharmonic in Ui
(see Figure V.6).

U*1

Figure V.6. The enlarging process

Next we take an Ii > R and add further functions Ix with x E ax,

say
I ml+1,···,Im, such that for 9 = (l/h,.··,llfm) it follows that Ig(x)1 > R
at every point of ax n Ui, and Ig( x) I < Ii in K n Ui. Then we define
fj = {x E xnu; : Ig(x)l:S Ii}.
292 V. Stein Theory

We define U f by replacing the number R by R in the first ml inequalities.

We use the enlarging method of Section V.4 and prove in the same way that
every holomorphic function in U f can be approximated by those in U.
Now we let U and U' increase by making the number m bigger (i.e., adding
new functions) and, simultaneously, by also making R and it bigger with
R < R. So we can construct sequences Uj, Uj where the Uj converge to
Ui n X and the Uj to Ui n XS' Hence, we have the approximation of the
holomorphic functions in Xs n Ui by those in X n Ui.
Let 11' be the open covering of X consisting of the two elements WI = Xs
and W 2 = Ui n X. On the intersection W 12 we can approximate every hol(}-
morphic function by those on W 2 . Since the cohomology of X is finite, there
are co cycles 6, ... ,~k E ZI (11',0) such that the corresponding cohomology
classes ~ I'''''~ k form a basis of the image of Z1('W, 0) in HI(X, 0). The
mapping <p : C k x CO('W, 0) -+ Zl('W, 0) defined by

<p(aI, .. . , ak, "1) := aI6 + ... + ak~k + 6("1)

is a surjective mapping of Fn§chet spaces. By the open mapping theorem (see
[Ru74]) <p is an open map.
If 9 is a holomorphic function in X s , then 9 can be approximated on W 12 by
a sequence of holomorphic functions fJ E O(W2)' Since <p is open, the cocycle
9 - Ij (which is a small function on W 12 ) can be written as a sum Ii + 6("1j) ,
with a small function Ii E 0(W12 ) and a (small) cochain "1j. Replacing fJ - Ii
by a suitable approximation I, and "1j by a suitable (and also small) cochain
{{!I, {!2}, we get 9 - I = (J2 - (J}, with {!l E O(Wd and (J2 E O(W2 ). The
function
~(x)'- { g(x) + (JI(X)
9 .- I(x) + (J2(X)
is holomorphic on X and approximates 9 on X s as well as we want. _

Enlarging: The Whole Process. Let Xl = X s , PI = p, and h} = h.

Then we define X := X and P2 := P + h. Repeating the process from above
2
with a suitable point X2 E 8X2, cuboids U2 CC U2, and a suitable function
h2 we obtain a new set X = X3 that is defined by a new function P3 = P2+h2,
and every holomorphic function on X 2 can be approximated arbitrarily well
by holomorphic functions on X3.
Continuing in this way we can always choose the correction functions hi so
small that the strict plurisubharmonicity is never disturbed. After finitely
many steps we obtain the following:
 6. The Levi Problem 293

6.2 Proposition. There is a strictly pseudoconvex complex manifold X

with XS CC X such that every holomorphic function on Xs 5:,an be approxi-
mated over Xs arbitrarily well by holomorphic functions on X.

We have the analoguous result for cochains.

6.3 Proposition. Assume that "If' is an open covering of X and that ~ E
Zl("If',O) is a cocycle with ~IXs = J(1/s) , ~ = 0(17). Then, by changing 1/,
the difference 17 - 17s can be made arbitrarily small on Xs'

Remark. We can repl~ce X by a set Xs+e:, with an E > 0, since such a

manifold is contained in X.

Solution of the Levi Problem. Let X be an n-dimensional strongly

pseudo convex manifold that as above is endowed with a smooth exhaustion
function p that is strictly plurisubharmonic on {p ~ I}.
The solution of the Levi problem will be given in three steps. First we prove
an approximation theorem for Xs and X.
6.4 Theorem. Every holomorphic function in X s , s ~ 1, can be approxi-
mated by holomorphic functions in X.

PROOF: There is a maximal swith s < S:S: 00 such that every holomorphic
function on Xs can be approximated by those on Xs. We have to prove s = 00.
If this were not the case, we could approximate each holomorphic function on
Xs by those on X s+ c with E > O. This would imply the approximation of the
holomorphic functions on Xs by those on X s+c and contradict maximality. _

Assume now that K is a compact set in X s , s ~ 1. Then K is also contained

in X t for t > s. We denote the holomorphically convex hull of K in X t by
Kt . In the second step of our proof we show that Kt = Ks for every t 2: s.
~ ~

It is always the case that Ks C Kt, and, due to the approximation results,
Kt n Xs = Ks. Suppose that there exists at> s such that Kt is bigger
than Ks. Then we can find a minimal tf ~ s with Kt C X 1'. It follows that
tf< t (since X t is holomorphically convex), and there is at least one point
Xo E Kt n oXt ,. We choose a very small number E > 0 and consider the set

Xt'+c'
Let A(xo) C Xt'+c be the analytic tangent to Xt' at xo. We may assume
that there is a neighborhood U = U(xo) and a holomorphic function gin U
such that A(xo) = {x E Un Xt'+c : g(x) = O}, and we find a meromorphic
function f on Xt'+c with poles only on A(xo).
294 V. Stein Theory

Then we construct a sequence of points Xv outside of X t , converging to Xo

and meromorphic functions fv on Xt'+e with poles only on A(x v ). This can
be done in such a way that fv converges to f with respect to the Fubini metric
in C (which will be described in Section VI.3). Then f" is holomorphic in
a neighborhood of Xt', Ifv(xo)1 takes arbitrarily large values, and sUPKlfvl
stays bounded. By approximation we can find a holomorphic function on r
X t such that 1f*(xo)1 > sUPKIf*I, i.e., Xo ¢ Kt . This is a contradiction.
Let us give some more details for the construction of i and Iv. To begin with i,
we use the open covering %' = {UI , U2} of X t ,+< given by U1 = Un X t ,+< and
U2 = X t ,+< - A(xo). Since dim H1(Xt'+e, 0) < 00, there are complex numbers
al, ... ,am such that the function a := alg- 1 + ... + amg- m is a cocycle on U12
that is cohomologous to zero. Then a = 12 - it, with it E O(Ul) and 12 E 0(U2).
So we get the meromorphic function i that is equal to a + it on U1 , and equal to
h on U2. It has A(xo) as polar set.
We can find a holomorphic family of analytic tangents A(z), always given by one
holomorphic equation gz = 0, with z arbitrarily near to Xo. If m is big enough, then
for every z we have a vector space Pz of principal parts p = alg;l + ... + amg;m
and a linear map <flz : P z -+ H 1 (Xt '+e, 0). Let Ez C P z be the kernel of <flz. Then
dim(Ez ) ~ 1, and there is a minimal dimension mo :S m for Ez that will occur
at generic points. Taking Ez only at generic points and forming the closure for
z -+ xo, we get a regular holomorphic family of vector spaces. If we always take
meromorphic functions iz with principal part in E z, we get a continuous family
(fz) and the desired convergence fz -+ i for z -+ Xo·

The last step of our proof is simple:

Let K c X be any compact set. Then K is contained in some K s , s 2: 1, and
from above we know that Kt = Ks for every t > s. Let X be an arbitrary
point of the holomorphically convex hull K. Then x E Xt, for some t > s. If f
is a holomorphic function in Xt, then there is a holomorphic approximation
f E O(X). Since we have lj(x)1 ::; sUPKIf!, we can conclude that If(x)1 ::;
sUPKlfl as well. So x E Kt , and K = Ks. This proves the following theorem.

6.5 Theorem. If X is strongly pseudoconvex, then X is also holomorphi-

cally convex. It is a Stein manifold if and only if there are no compact analytic
subsets in X of positive dimension.

In addition, we have the following result.

6.6 Proposition. If X contains a compact analytic subset A of positive
dimension, then already A C Xl.

The PROOF is the same as that for K = Ki .

We also obtain the following theorem.
 6. The Levi Problem 295

6.7 Theorem. If X is a strictly pseudoconvex (i.e., I-complete) manifold

and V a holomorphic vector bundle on X, then Hl(X, V) = o.

PROOF: This has already been proved for every distinguished Stein mani-
fold X s , s 2: 1. By approximation of coboundaries we get the result for X .
•
The Compact Case. Let X be a compact complex manifold, and
V a holomorphic vector bundle on X. We can find a pair of finite open
coverings (%', %'*) for X consisting of concentric cuboids Ui cc Ut such
that H 1 (Ui , V) = Hl(Ut, V) = 0 for all i. Then CO(%" V), Zl(%" V) and
Zl (%' * , V) are Frechet spaces. The finiteness of the coverings is essential here!
Again we consider the linear mappings

such that u is surjective, v is completely continuous, and Im( u + v) =

Bl (%', V). Since the coverings are acyclic, it follows from the theorem of
Schwartz that dime Hl(X, V) < 00.

All restriction maps r(Ut, V) ~ f(U, V) are completely continuous. From

this it also follows that id : rex,
V) ~ rex,
V) is completely continuous,
and f(X, V) is finite-dimensional. So we have the following theorem.
6.8 Theorem. If X is a compact complex manifold, then

dime rex, V) < 00 and dime Hl(X, V) < 00.

Exercises
1. Apply the results of this section to domains in en.
What do strong and
strict pseudoconvexity mean in this case?
2. Let X be a strongly pseudoconvex complex manifold. Prove that there
is a compact subset K C X such that every irreducible compact analytic
subset A C X of positive dimension is contained in K.
3. Let X be a Stein manifold and V a holomorphic vector bundle over X.
Prove that for every Xo E X there exists an open neighborhood U =
U(xo) c X and holomorphic sections 6, ... , f.N E rex,
V) such that for
every x E U, Vx is generated by 6(x), ... ,f.N(X).
Chapter VI

Kahler Manifolds

1. Differential Forms
The Exterior Algebra. Let X be an n-dimensional complex manifold
and x E X a point. We consider complex-valued alternating multilinear forms
on the tangent space Tx(X) (abbreviated by T).
We assume that the reader is familiar with real multilinear algebra!
Definition. A complex r-form (or r-dimensional differential form) at
x is an alternating lR-multilinear mapping

cp:Tx···xT~C.
~
r-times

The set of all complex r-forms at x is denoted by pr.

Remarks
1. By convention, pO = c. pI = P(T) is the complexification of the 2n-
dimensional real vector space T; = HomJR(T,lR).
2. Since T is (2n )-dimensional over lR, every alternating multilinear form
on T with more than 2n arguments must be zero. So pr = 0 for r > 2n.
3. In general, pr is a complex vector space. We can represent an element
cp E pT uniquely in the form cp = Re(cp) + i Im(cp), where Re(cp) and
Im(cp) are real-valued r-forms at x. Then it follows directly that

dimcpr = C;)·
4. We associate with each element cp E pr a complex-conjugate element
rp E pr by setting c,o(v}, ... , vr ) := cp(v}, ... , v r ). We have:
(a) rp = Re(cp) - i Im(cp).
(b) 7p = cpo
(c) (cp + t/J) = -;p + ¢.
(d) cp is real if and only if c,o = cp.
Now let cp E pr and t/J E ps be given. The wedge product cp A t/J E pr+s is
defined by
298 VI. Kahler Manifolds

Then <p A1J!(v, w) = <p(v) ·1J!(w) - <p(w) ·1J!(v) for I-forms <p, 1J!, and in general:
=
1. <p A 1J! (-lt s1/J A <p (anticommutativity).
2. (<p A 1/J) A W=<p A (1J! A w) (associativity).
In particular, t.p A <p = 0 for every I-form <po
We also write /{ F instead of Fr. With the multiplication "A" the vector
space
00 2n
AF:= EBArF = EBF r

r=O r=O
becomes a graded associative (noncommutative) (:-algebra with 1. It is called
the exterior algebra at x.
For the moment let Wj := dZ j and wn+j := d:Zj for j = 1, ... ,n. Then Fr is
generated by the elements

W V1 A ... A W Vr , with 1 :::; Vl < ... < Vr :::; 2n.

The number of these elements is exactly (2;). So they form a basis, and every
<p E Fr has a uniquely determined representation

<p=

with complex coefficients a V1 .. 'Vr'

Forms of Type (p, q). Now we consider the influence of the complex
structure.

Definition. Let p, q E No and p + q ~ r. A form t.p E Fr is called a

form of type (p, q) if

1.1 Proposition. If <p E Fr is a nonzero form of type (p, q), then p and q
are uniquely determined.

PROOF: Suppose <p is of type (p, q) and of type (p', q'). Since <p =I 0 there
exist tangent vectors Vl, ... ,Vr such that <p( VI, •.• , vr ) i= O. Then
 1. Differential Forms 299

) cP(fl. rp( VI, ..• , V r ),

rp( CVl , ... ,CVr = { Cp' 7"/' (
C . rp VI,· .. ,Vr .
)

Therefore, cpcq = cP' c q' for each c E C. If c = e it , then eit(p-q) = eit(p' -q'),
for arbitrary t E IR. That can hold only when p :- q = p' - q'. Since p + q =
p' + q' = r, it follows that p = p' and q = q'. •

1.2 Proposition.
1. Ifrp is of type (p,q), then rp is of type (q,p).
2. If rp, 'IjJ are both of type (p, q), then rp + 'ljJ and>..· rp (with>" E C) are also
of type (p, q).
3. If rp is a form of type (p, q) and'IjJ of type (p', q'), then rp "'IjJ is of type
(p+p',q+q').

We leave the PROOF to the reader.

Example
We have dzv(cv) = (cv)[zvl = c· (v[zv)) = c· dzv(v), since Zv is holomorphic.
So dz v is a form of type (1,0).
From dZ v = dz v it follows that dZ v is a form of type (0,1).
Then dz·'1 " ... " dz''p "<lz'J1 " ... " dZ·Jq (with 1 -< il < ... < i P -< nand
1 S]1 < .,. < jq S n) is a form of type (p, q).
1.3 Theorem. Any r-form rp has a uniquely determined representation

rp = L rp(p,q) ,
p+q=r

where rp(p,q) E FT are forms of type (p, q).

PROOF: The existence of the desired representation follows from the fact
that the forms dZ i1 " .•. " dz ip "<lzj1 " ... " <lzjq constitute a basis of Fr.
For the uniqueness let

Then
L 'ljJ(p,q) = 0 for 'IjJ(p,q) := rp(p,q) - ;p(p,q).
p+q=r
It follows that

0= L 'ljJ(p,q)(CV}, ... ,CVr ) = L cP(fl.'ljJ(p,q)(v}, ... ,vr ).

p+q=r p+q=r
300 VI. Kahler Manifolds

For fixed (VI, ... ,VT ) we obtain a polynomial equation in the ring e[c, c], and
all coefficients 'ljJ(p,q) (VI, •.. , V T ) must vanish. Since we can choose VI, ... , V T
arbitrarily, we have cp(p,q) = ij5(p,q) for all p, q. •

Bundles of Differential Forms. Let X be an n-dimensional complex

manifold. We denote by F(X) the complexified cotangent bundle T*(X) 0iC.
It has the spaces F(Tx(X)) = Hom]R(Tx(X), q = TAX)* 0 e of complex
covariant tangent vectors as fibers, so it is a (topological) complex vector
bundle of rank 2n. I It even has a real-analytic structure, but not a complex-
analytic structure.
If E is a (topological) complex vector bundle of rank m over X, then for
o S r S m we can construct a bundle t{ E of rank (';) over X such that
(/\TE)X = /\T(Ex) for every x E X. If E is given by transition functions gij,
then /\T E is given by the matrices g;;)
whose entries are the (r x r) minors
of 9iJ.
In particular we have the vector bundles /\T F(X) of rank (~n), with a real-
analytic structure.
Definition. An r-form (or an r-dimensional differential form) on an
open set U C X is a smooth section wE r(U, /\T F(X)).

So an r-form w on U assigns to every x E U an r-form Wx at x. If Z1, ..• ,Zn

are local coordinates in a neighborhood of x, then Wj := dZ j and wn+j := Uzj
form a basis of the I-forms on this neighborhood, and there is a representation

Wx = a·lI .. ·t.r (x)w,1..1 1\ ... 1\ W l r '

where x H ai, ... ir(X) are smooth functions.

Henceforth, the set of all (smooth) r-forms on U will be denoted by gr(U),
and the subset of all forms of type (p, q) by JiI(p,q) (U).
If f is a smooth function on U, then its differential df E Jill (U) is given by
x H (df)x. In local coordinates we have
n n
df = L fzv dz.., + L hv Uz..,.
..,=1 1'=1

A (smooth) vector field is a smooth section of the tangent bundle T(X). So

in local coordinates it can be written in the form
1 The tangent bundle T(X) is a complex-analytic vector bundle of rank n. Here
we denote its real dual bundle by T*(X). It is a real-analytic bundle of rank
2n (over Ii). The complex dual bundle T'(X) of T(X) (with fiber T~(X)
Homc(T",(X),C) ) is a complex-analytic bundle of rank n over C.
 1. Differential Forms 301

where the coefficients ~II are smooth functions. Then we can apply df to such
a vector field and obtain
n n
df(~) = 2:~lIfzv + 2:Lhv.
11=1 11=1

For any open set U the differential can be generalized to the Poincare map
d = du : dr(U) -t d r + 1 (U) in the following way:

Ifw= ai, .. .ir Wi, /\ ... /\ Wir is the basis representation in a

1:::;:i,<···<i r :::;:2n
coordinate neighborhood U, then

du(w) :=

One can show that this definition is independent of the choice of the local
coordinates and that d has the following properties:
1. If f is a smooth function, then df is the differential of f.
2. dis IC-linear.
3. dod = O.
4. dv(wlv) = (duw)lv, for w E dr(U).
5. If <p E dr(U) and 'IjJ E dS(U), then d(<p /\ 'IjJ) = dip /\ 'IjJ + (-It<p /\ d'IjJ.
6. d is a real operator; that is, dfp = dip. In particular, dip = d(Re <p) +
i d(Im<p).
The differential dw is called the total derivative or exterior derivative of w.
Now we consider the decomposition of an r-form into a sum of forms of type
(p, q). We use some abbreviations. If I = (i 1 , ... , ip) and J = (jI, ... ,jq) are
multi-indices in increasing order, we write

instead of
ai1 ... i",j, ... jqdzi , /\ ... /\ dzi " /\ azj, /\ ... /\ azjq.
So a general r-dimensional differential form w has the unique representation

w = 2: L aIJdzI /\azJ,
p+q=r III=p
IJI=q

and the Poincare differential of w is given by

302 VI. Kiihler Manifolds

dw = L L daIJ 1\ dZI 1\ azJ.

p+q=r Ill=p
jJl=q

If f is a smooth function, then df = 8 f + 8 f, with

n n
8f = Lfz"dzv and 8f = L h" az v.
v=l v=l

Here of has type (1,0), 8f has type (0,1), and df type (1,1).
1.4 Proposition. If <p is a form of type (p,q), then d<p has a unique de-
composition d<p = 8<p + 8<p with a (p + 1, q)-form 8<p and a (p, q + I)-form
8<p.

PROOF: If <p = L-l,JaIJdzl 1\ azj, define

8<p := L 8al J 1\ dZI 1\ azJ and 8<p:= L 8a[ J 1\ dz[ 1\ azJ.
I,J 1,J

Then d<p = 8<p + 8<p is the unique decomposition of the (p + q + I)-form d<p
into forms of pure type. _
For general r-forms the derivatives with respect to z and z are defined in the
obvious way. (In the French literature one writes d' <p instead of 8<p and d" <p
instead of 8<p.)
1.5 Theorem.
1. 8 and 8 are C>linear operators with d = 8 + 8.
2. 88 = 0, 88 = 0, and 88 + 88 = o.
3. 8,8 are not real. We have
8<p = fi<p and 8<p = Ocp.
4. II <p is an r-Iorm and'lj; is arbitrary, then
8( <p 1\ 'I/J) = 8<p 1\ 'Ij; + (-1 t <p 1\ 'I/J,
8(<p 1\ 'I/J) = 8<p 1\ 'Ij; + (-It <p 1\ 8'1j;.

PROOF: It suffices to prove this for forms of pure type. Then the formulas
can be easily derived from the corresponding formulas for d and the unique-
ness of the decomposition into forms of type (p, q). _

Remark. Sometimes the operator de := i(8 - 8) will be used. Then de<p =

deep, so de is a real operator with dCd c = O. We have ddc = 2i 88.
A smooth function f is holomorphic if and only if 8f = o. Correspondingly,
it follows for a (P,O)-form <p = ~III=p aI dz[ that 8<p = 0 if and only if all
coefficients al are holomorphic. Hence we make the following definition:
 2. Dolbeault Theory 303

Definition. Let <p be a p-form on the open set U eX.

1. <p is called holomorphic if <p is of type (P,O) and 8<p = O.
2. <p is called antiholomorphic if and only if <p is of type (O,p) and
8<p = o.
The set of holomorphic p-forrns on U is denoted by nV(U).

Clearly, <p is antiholomorphic if and only if <j5 is holomorphic.

Exercises
1. Let X be an n-dimensional complex manifold, f a smooth function on
X, and w a smooth form of type (n - 1, n - 1) on X. Prove that

2. A real differential form of type (1,1) is called positive if iw{v, v) > 0 for
v
every tangent vector =1= o. Prove that i88/1z/lz is a positive form on en.
3. Consider the (n, n - I)-form "10 on endefined by
n
"10 := (_1)n(n-l)/2 2) -l)kZk dz1 /\···/\ dZ n /\ azl/\···/\ di;./\ ... /\ azn
k=l

and calculate 8"10. Let f be a holomorphic function on an open neigh-

borhood of the closed ball Br(O), and w := (J(z)/IIz/l 2n) . "10. Show that
dw =0.
4. Let X := lP'n. On Uo := {(zo : Zl : ... : zn) E lP'n : Zo =1= O} we use
the holomorphic coordinates tv = zv/zo. Then wo = dtl /\ ... /\ dt n is
a holomorphic n-form on Uo. Show that Ii holomorphic n-form on an
open subset U c X is the same as a holomorphic section over U in the
canonical bundle Kx. Prove that there is a meromorphic section s in Kx
with slua = Wo and dives) = -en
+ l)Ho, where Ho = {zo = O} is the
hyperplane at infinity.

2. Dolbeault Theory
Integration of Differential Forms. We recall some results from real
analysis.
Let G c en ~ ]R2n be a domain. A closed subset MeG is called an
r-dimensional submanifold of class <;fk if for each z E M there exists a neigh-
borhood U of z in G and a <;fk map f : U --+ jR2n-r such that:
1. Un M = f-l(O).
2. The rank of the (real) Jacobi matrix of f is equal to 2n - r.
304 VI. Kahler Manifolds

It follows from the (real) implicit function theorem that for any z E M there
is a neighborhood U of z, an open set W c IRr , and a '{/k map r.p : W -7 G
such that:
1. r.p maps W homeomorphically onto U.
2. The rank of the Jacobian matrix of r.p is equal to r.
Then r.p is called a local parametrization of M.
An r-dimensional differential form W = 2: p +q =r 2:[,J aIJ dz[ A cLzJ is called
an r-form of class '{j'k if all coefficients are functions of class k. Then dw is
an (r + I)-form of class 'ifk-l. We always assume that k ~ l.
If 4> : Bl -7 B2 is a 'ifk map between domains Bl c em and B2 c en, then
every r-form w of class '{j'k on B2 can be lifted back to Bl by

(4)*W)z(Vl,'''' vr ) := W4>(z)(4>.Vl, ... , ~*vr).

If UI, ... ,U2n are real coordinates in B2 then

4>* (,",a . . du 11 A".Adu.)

L...; 11···1r tr = '"'(a·
~ . o4»d(u·1} o4»A···Ad(u·tr 04».
'll···'lr

Therefore, 4>* W is again an r-form of class 'ifk. We have:

1. 4>*(WI A W2) = 4>*WI A 4>*W2'
2. 4>*(dw) = d(4)*w), in particular 4>*(df) = d(J 0 4».

Now let W = a dZ I A ... A dZ n A liZ l A ... A cLzn be an arbitrary (2n)-form.

Since dZ i A liZi = -2i dXi A dYi, it follows that
w a· (_1)n(n-l)/2dz 1 A cLzl A··· A dZ n A cLzn
= a· (_l)n(n-l)/2 . (-2)nj n dxl A dYl A ... A dXn A dYn
a· (-1)n(n+1)/2(2i)ndxl A dYl A··· A dXn A dYn.
We writ.e a := a· (_1)n(n+1)/2(2i)n. If Been is a bounded domain and a
continuous in a neighborhood of B, then

In In
w:= a(xl + iYI,"" Xn + jYn) dx1dYl ... dXndYn.
If MeG is an r-dimensional sub manifold of class C(?k, w an r-form of class
C(j'k, and K C M a compact subset that is contained in the range of a local
parametrization r.p : W -7 M, then we define

L w:= i-l(K) r.p*w.

The integral does not depend on the parametrization if we allow only changes
of parameters with positive Jacobi determinant. If K cannot be covered by
a single parametrization, then one uses a partition of unity.
 2. Dolbeault Theory 305

Finally we have the famous Stokes's theorem.

2.1 Stokes's theorem. Let G c en be a bounded domain with smooth
boundary and w a smooth (n - 1) -form in a neighborhood of G. Then

{ w = ( dw.
Jac Jc

The Inhomogeneous Cauchy Formula. We consider a generaliza-

tion of Cauchy's integral formula in the I-dimensional case. Let G cc be e
a domain with smooth boundary, and U = U (G) an open neighborhood of
the closed domain.
2.2 Theorem. Let f be a continuously differentiable (complex-valued) func-
tion on U. Then, for Z E G, we have

fez) = ~ ( f(() d( +~ { (af joc,)(() d( 1\ dC,.

2m Jac ( - z 2m Jc (- Z

PROOF: Let z E G be fixed, choose an r > 0 such that the disk Dr(z) lies
relatively compactly in G, and define Gr := G - Dr(z). The differential form

w(z) := ~. f(() d(
27T1 (- z
is of class 'ifl in an open neighborhood V = V(Gr), and its Poincare differ-
ential is
dw(z) = -~ . (of jac,)(() d( 1\ dC,.
27T1 (- z
By the theorem of Stokes we have

1Cr
dw(z) = (
Jacr
w(z) = (
Jac
w(z) - (
JaOr(Z)
w(z).

If 0 < c < rand Ac,r := Dr(z) - Dc(z), then for any continuous function 9
on Dr(z),

J g(() d( 1\ d(
A.,r ( - Z
= l r 121< g(z + e· eit ) ·2ledtde
.
e' e l
't

l
c 0

2i
r
10 2
1< g(z + e . e it ) . e- it dt de.
This integral exists and stays bounded if c tends to zero. So

( dw(z) = f w(z) - lim f w(z).

Je Jac r---'rO Jaor(z)
306 VI. Kahler Manifolds

In order to show that the limit on the right side is equal to fez) we estimate

Ilaor(z)
f w(z) - fez) I = 12:i 127r fez + re it ) • idt - :7r ! fez) dt I
= 12~ 127r (f(z + re it ) - fez)) dt I
< sup If(z + re it ) - f(z)1 -+ 0 for r -+ o.
[O,27r]

It follows that fez) = faa w(z) - fa dw(z). •

The a-Equation in One Variable. The Poincare lemma from real
analysis can be formulated as follows:
Let G c en be a star-shaped (e.g., a convex) region, <p E ,sdr(G), r > 0,
and d<p = O. Then there exists a 1/J E ,sdr-1(G) with dt/J = <po
We want to prove a similar theorem for the a operator. For this we start in
one variable using the integral

1.
Chf(z) := -2 f (f«) d( A i(,
7r1 la - z
for bounded functions f E '6'1 (G).

2.3 Theorem. Let f E '6'1(C) be a function with compact support. Then

there is a continuously differentiable function u on e with U z = f.

PROOF: Choose an R > 0 such that supp(f) cc D := DR(O). Then u :=

Chf is a continuously differentiable function on D, and we can assume that
the integral is taken over the whole plane. It follows that

u z
( ) = _1
2.
7r1
i C
f« + z) dl" A d7'
( " ",

and therefore (applying formulas for the derivative of parametric integrals)

Uz(z) =
~ f 8f /8((+z)d(Ad(
lc
27r1 (
~ f 8f /8(() d( A d(
2m lc (- z

= ~ f 8f/8(() d( Ad(
2mlD (-z

= fez) - ~ f f«() d( = fez),

2m laD (- z
 2. Dolbeault Theory 307

since IlaD == o. •

2.4 Corollary. Let G cc e be a bounded domain, and I E 'ifl(G) a

bounded lunction. Then there exists a continuously differentiable function u
on G with U z = I.

PROOF: Again we define u := ChI.

Let Zo E G be an arbitrary point and D := Dr(zo) a disk with D cc G.
There exists a smooth function {! on e with
1. O:S {! :s 1,
2. (!ID == 1,
3. supp({!) cc G.
We define II := {!. I and h := I-II. Then II is a continuously differentiable
function on e with IdD = liD, and h a function on G with hiD == o.
There exists a 'if 1 function U1 on e with (u1h = II. The function U2 := Ch h
is defined on G, and given by

U2(Z) =
1
-2 .
7rI
1 h(()
-;- d(
G-D" - Z
1\
-
d(.

The integrand is continuous and bounded on G - D, and holomorphic in z,

for zED.
Obviously, u = U1 + U2 on G, and on D we have Uz = (u1h = II = f. •

A Theorem of Hartogs. Now we are able to prove the so-called Kugel-

satz, also known as Hartogs' theorem:
2.5 Theorem. Let G c en be a domain, n ?: 2, KeG a compact subset. II
G - K is connected, any holomorphic lunction I on G - K has a holomorphic
1
extension on G.

PROOF: Choose an open neighborhood U = U(K) cc G and a smooth

function (! on en with supp(e) cc G and {!Iu == 1. Define
8{!
gk := I· ~, for k = 1, ... , n.
UZk
We have
8g k _ I 8 2{! _ 8g1
Ozl - . OzkOzI - Ozk·
For fixed z" = (Z2' ... , zn) we define
308 VI. Kahler Manifolds

Then U Z1 = g1 (we can apply the previous corollary, since g1 has compact
support), and for k ~ 2 we get

~ r (gdzk «(, Z2,"" zn) d( 1\ de,

21l",lc (- Zl
~ r
21l"1 lc
(gkhl «(, Z2,···, zn) d( 1\ de,
(- Zl
gk(Zl, Z2,···, zn).

For z ~ supp(e) we have UZk (z) = 0, k = 1, ... , n. Therefore, U is a holomor-

phic function outside supp(e). Since g1 «,
z') = 0 for arbitrary ( and large
IIz'lI, it follows that U(Zb z') = 0 for large liz' II , and so U is defined everywhere
and is identically zero on the unbounded connected component Z of G - K.
There exists a nonempty open subset V of Z n (G - supp(e» C G - K. The
function 1:= (1 - e) . f + u is defined on G and coincides with f on V. In
addition, we have

1- - { -f· l?Zk + U Zk = -gk + gk = 0 in G - K,

Zk - = gk = 0
U Zk in U.

So 1is holomorphic in G.
-
Dolbeault's Lemma Before proving the a analogue of Poincare's
lemma we need the following result.
2.6 Proposition. Let P cc Q be polydisks around the origin in en. Let
9 be a smooth function in Q that is holomorphic in Zk+b"" Zn, for some
k ~ l.
Then there exists a smooth function u in Q that is again holomorphic in
Zk+l,"" Zn such that U Zk = 9 on P.

PROOF: There are disks E cc D around 0 in the zk-plane such that P =

P' x E x P" and Q = Q' x D x Q". Let e be a smooth nonnegative function
on e with elE == 1 and elc-D == O. Then we define

(
U Zb"" )._ -1-1 e«() .
Zn .- 2 .
1l"1 C
g(Zb'" ,Zk-l, (, Zk+1,"" Zn) dl" d7'
I"
., - Zk
., 1\ .,.

The function U is defined on Q, and the integrand has compact support.

Therefore,
UZk(z) = (e' g)(z) = g(z) on P,
and by the rules for parametric integration we get U z , = 0 for i = k+ 1, ... , n.
So U is holomorphic in these variables. -
 2. Dolbeault Theory 309

2.7 Dolbeault's lemma. Let P cc Q be polydisks around the origin in

en. For each r.p E pfp,q ( Q) with q > 0 and 8r.p = 0 there exists a t/J E
PlP,q-l(P) with 8t/J = r.plp.

PROOF: Without loss of generality we may assume that p = O. Let

k ::; n be the smallest number such that r.p does not involve the differen-
tials cLzk+b"" cLzn . We carry out the proof by induction on k.
For k = 0 there is nothing to prove, since q > O.
Now we assume that k > 1 and the theorem has already been proved for
k - 1. Then we write
r.p = cLzk /\ a + (3,
where a E PlO,q-l(Q) and (3 E Plo,q(Q) are forms not involving cLzk, ... , cLzn .
If a = I:jJI=q-l aJcLzJ (where I:* means that cLzk, ... , cLzn do not occur),
then
o = 8r.p = -cLzk /\ 8a + 8(3

- 2:cLzk/\cLzv /\ ( 2:" (aJhVcLzJ) +8(3.

v#k IJI=q-l
Then it follows that (aJhv == 0 for 11 > k. By the above proposition there
are smooth functions AJ in Q that are holomorphic in Zk+l,"" zn such that
(AJ hk = aJ on P. We define 1 := I:jJI=q-l AJ cLzJ. Then

81 = 2:* 2:(AJhv cLzv /\ cLzJ

IJI=q-l v(/.J

2:* (ai cLzk /\ cLzJ + 2: (AJ hv cLz v /\ cLzJ )

IJI=q-l v<k
v(/.J
= cLzk /\ L" aJ cLzJ + ...
IJI=q-l
cLzk /\ a + Tf,
where Tf does not involve cLzk, .. ' ,cLzn • We see that also (3-Tf does not involve
cLzk, ... ,cLzn .
Since (3-Tf = (r.p-cLzk/\a)-(81-cLzk/\a) = r.p-81, we have 8(.8-Tf) = 8r.p = 0,
and the induction hypothesis implies that there is a form u E pfO,q-l(P) with
8u = ((3 - Tf)lp· Setting t/J := u + 1, it follows that
8t/J = ((3 - Tf) + (dZk /\ a + Tf) = r.p on P.
This completes the proof. _
We immediately obtain the following result for manifolds X:
310 VI. Kahler Manifolds

If U c X is an open set and cp E .Q(°,q(U) a smooth form with q ~ 1

and acp = 0, then for every x E U there exists an open neighborhood
V = Vex) c U and a form 1/J E .Q(0,q-l(V) with a1/J = cplv.
Without proof we mention the following result, which provides more precise
information (see [GrLi70] and [Li70]).
2.8 Theorem (I. Lieb). Let G cc C n be a Levi convex domain with
smooth boundary and w = LJ aJ azJ a smooth (0, q)-form on G with aw = o.
Moreover, suppose that there is a real constant M > 0 with

IIwl! := m;xs~plaJI ::; M.

Then there exists a constant k independent of wand a form 1/J of type (0, q -1)
on G with a1/J = wand 111/J1I ::; k· M.
Siu and Range generalized this theorem to domains with piecewise smooth
boundary (see [RaSi]). L. Hormander has given a quite different proof for the
solution of the a-equation with L2-estimates on pseudo convex domains using
methods of functional analysis (see [Hoe66]). In the meantime there exists a
big industry of solving a-equations, but we do not want to go here into detail.

Dolbeault Groups. Let X be an n-dimensional complex manifold and

U C X an open subset. Then we have natural inclusions2
E : C y .Q(°(U) = ¥iOO(U) and E: OP(U) y dP,o(U).
Together with the Poincare differential d and the a-operator we get the fol-
lowing sequences of C-linear maps:
1. The de Rham sequence

If U is diffeomorphic to a star-shaped domain, then the sequence is exact,

i.e., {f E d°(U) : df = O} = C and ZT(U) := Ker(d : dT(U) -t
dr+l(U» is equal to Br(u) := Im(d : dr-leU) -t dr(U». In general,
we have only Br(u) C ZT(U) (because dod = 0).
2. The Dolbeault sequence

We define
{cp E dP,q (U) : acp = O} for 0 ::; q ::; n,
.- 8(dp,q-l(U» for 1 ::; q ::; nand BP,o(U):= O.
Then Bp,q c Zp,q, because 8 0 8 = 0 (respectively 0 C OP (U».
2 Here C stands for the set of locally constant functions.
 2. Dolbeault Theory 311

Definition. The complex vector space Hp,q(U) := Zp,q(U)j Bp,q(U)

is called the Dolbeault (cohomology) group of type (p, q) of the set U, and
the complex vector space Hr(u) := zr(U)jBr(u) is called the rth de
Rham (cohomology) group of U.

We have

For q 2: 1 the condition Hp,q(U) = 0 means that for every 'P E .f2Ip,q(U) with
8'P = 0 there is a 7/J E .f2Ip,q-l(U) with 87/J = 'P. From the solution of the
a-equation (using, for example, (Hoe66], chapter IV, or Dolbeault's lemma
and an approximation theorem due to Oka) we obtain that HP,q(G) = 0 for
every strictly pseudoconvex domain G c en and q 2: 1.
If U is connected, then HO(U) = C. If U is star-shaped, then Hr(u) = 0 for
r > o.

Now let n~ := APT'(X) be the holomorphic vector bundle of holomorphic

p-forms on X (in particular, n~ = Ox, nk
= T'(X) and nx
= Kx). Then
np(U) = r(U, n~) for every open set U C X. We construct a linear map

For every ~ E Z1 (X, n~) there is an open covering %' = {Ui : i E I} of X

such that ~ is given with respect to this covering. Then we can pass to any
refinement. So we may assume that %' is as fine as we want.
Let ~ = (~ij) E Z1(%" n~) be given. Then ~ij E np(Uij ) and ~ij+~jk = ~ik on
Uij k. Let (ek) be a partition of unity for the covering %'. We define differential
forms
'Pi := Lek~ki E .f2IP'°(Ui), for i E I.
kE/

Here 'P = ('Pi) is an element of CO(%" AP'o F(X)) with J'P =~, since

'Pj - <Pi = L ek(~kj - ~kd = L ek~ij = ~ij.

k k

In Uij we have
8<pj - 8<Pi = 8~ij = o.
:!herefore, a global (p, I)-form w on X can be defined by wlu; := 8<pj. Now,
ow = o. So wE Zp,1(X).

If ~ is a coboundary, ~ = J'f/ for some 'f/ E CO(%" n>:-), then 'f/j - 'f/i =
~ij = <Pj - 'Pi on Uij , and a global (P,O)-form 7/J on X can be defined by
7/Jlu; := 'f/i - <Pi· We get
312 VI. Kahler Manifolds

8(-"p}lu; = 8CPi - 81/i = wlup since 1/i is holomorphic.

So the assignment ~ H W induces a map D : HI (X, n~) -+ HP,I (X) by
[~I H [wI. where [···1 denotes the residue class. One can show that this
definition is independent of the covering OU.
2.9 Dolbeault's theorem. The map D : Hl(X,n~} -+ HP,I(X} is an
isomorphism.

PROOF: We use the notation from above. To show that D is injective, we

suppose that there is a (P,O}-form e on X with 8{! = w. Then 8CPi = 8elu;,
for every i E I. It follows that 8(cpi - e) = 0 and therefore Ti := CPi - e is
holomorphic on Ui .
Then T = (Ti) is a cochain of holomorphic p-forms, with

So the class of ~ vanishes.

Now let an element w E Zp,l(X) be given. Since 8w = 0, we can find a
covering OU = {Ui : i E I} of small polydisks (in local coordinates) with
(p,O)-forms CPi such that 8CPi = WIUi' Then 8( CPj - CPi) = 0 for every pair
(i,j), and ~ij := CPj - CPi lies in np(Uij }. Clearly, ~ = (~ij) is an element of
ZI(OU, n~) whose cohomology class is mapped by D onto the class of w. This
shows that D is surjective. _
Dolbeault's theorem has an interesting consequence:
2.10 Theorem. Let X be a Stein manifold. If cP is a (p, I}-form on X with
8cp = 0, then there exists a form"p on X of type (P,O) with 8"p = cpo
PROOF: By Theorem B for Stein manifolds we have HI(X, n~) = 0 for
every p, and the result follows from Dolbeault's theorem. _

Remark. Dolbeault's theorem is also true for forms of type (p, q) with
q > 1. For the proof one has to introduce higher cohomology groups
Hq(X, n~), to generalize Theorem B for those cohomology groups and to
show that Hq(X, n~) ~ Hp,q(X).

In an analogous way one can prove de Rham's theorem:

where Hr(x, C) denotes the cohomology group with values in the set of
locally constant functions (isomorphic to the singular cohomology group with
values in C).
Let OU be an open covering of X such that all intersections ULO " ' Lk are con-
tractible. In the case r = 2 the de Rham isomorphism
 2. Dolbeault Theory 313

is given as follows:
Let wE d 2 (X) be given, with dJ.JJ = O. On every Uv there exists an element
(3v E .0"1 (Uv) with
wlu... = d{3v.
Then d({31J. - (3v) = 0 on Uvw Therefore, on every UVIJ. there exists a smooth
(complex-valued) function IvlJ. with

d IvlJ. = {31J. - (3v.

Now, d(flJ.>' - Iv>. + IvlJ.) = 0 on U J1,>,.

V It follows that

avlJ.>' := (f1J.>, - Iv>. + IVIJ.)lu...I'>'

is constant.
So we have a = (avlJ.>') E Z2(OU,q, and the de Rham class [w] E H2(X) will
be mapped onto the cohomology class [a] E H2 (OU , q.

Exercises
1. Let 'fJo be the (n, n-l)-form defined in Exercise 1.3, and I a holomorphic
function on an open neighborhood of the closed ball Br(O). Prove that

(n - I)!
1(0) = (2rri)n
r
JaBr(O)
I(z)
IIzll 2n . 'fJo

(integral formula of Bochner-Martinelli).

2. Let (rv) be a monotone increasing sequence tending to r > 0, and let <p
be a smooth (0, q)-form on pn(o, r) with q > 0 and 8<p = o. Construct a
sequence ('l/Jv) of smooth (0, q - I)-forms on pn(o, r) such that
(a) 8'I/Jv = <p on pn(o, rv).
(b) l'l/Jv+1 - 'l/Jvl < 2- 11 on pn(o, r v-1) in the case q = 1, and 'l/Jv+1 = 'l/Jv
on pn(o, r v -1) for q ~ 2.
Show that Hp,q (P) = 0 for q ~ 1 and every polydisk P in en.
3. Let X be a complex manifold and rr : E --t X a holomorphic vector
e
bundle of rank q over X. If <p : Elu --t U x q is a trivialization, there
are sections ~v : U --t E defined by r.po~v(x) = (x,e v ). The q-tuple
~ = (~1' ... ' ~q) is called a lrame for E. Use the concept of frames to
define differential forms of type (p, q) with values in E locally as q-tuples
of forms. Let dP,q(U, E) be the space of (p, q)-forms with values in E
over U. Show that there is an analogue of the Dolbeault sequence for
forms with values in E.
314 VI. Kahler Manifolds

3. Kahler Metrics
Hermitian metrics. Assume that X is an n-dimensional complex man-
ifold.

Definition. A Hermitian metric H on X is an assignment of a positive

definite Hermitian form Hx to each tangent space Tx(X) such that locally
in complex coordinates it can be written in the form
n

Hx = L 9ij(X) dziazj,
i,j=l

with smooth coefficients 9ij.

If a : [0, 1] -t X is a differentiable are, we can assign to it a length

LH(a) := 11 .J Ho.(t) (o(t), o(t)) dt,

where o(t) = a.8j8t = I:v a~(t)8j8zv+ I:v (i~(t)8jOzv and a = (al, ... , an)
in local coordinates. So

L 9ij(a(t))a~(t)Qj(t) dt.
i,j

The length is positive if a is not constant.

Now, for a tangent vector ~ E Tx(X) a norm can be defined by

Then LH(a) = Jollla.8j8tllH dt. There is a uniquely determined differen-

tiable I-form a along a defined by

It follows that a*a = adt, with a = a(a.8j8t) = lIa.8j8tIlH. Traditionally, a

is called the line element and is denoted by ds, though it is not the differential
of a function. We have

LH(a) = 11 a*ds = idS,

and a* Ha(t) = Ila*8j8tllt- dt 2 = (dS)2. Therefore, we also denote the Hermi-

tian metric H by ds 2 .
 3. Ka.hler Metrics 315

Let OJ! = (U.).EI be a (countable) open covering of X by local coordinate

charts with coordinates (zl, ... , z~). Then there exists a partition of unity
«(J.) for OJ!, and
n
L{h· Ldziaz~
ds 2 :=
.EI i=1
is a Hermitian metric on X. Thus we have the following result.
3.1 Proposition. On every complex manifold there exists a Hermitian met-
ric.

The Fundamental Form. In a local coordinate system we associate

to our Hermitian form H = ds 2 = Li,j gijdzidZj the differential form

W = WH := i· Lgijdzi 1\ dZj .
i,j

(In the literature one often uses W = i /2 . L gijdzi 1\ az j.) We have w=

-i· Li,j 9ijazi 1\ dZj = w. SoW is a real (1, 1)-form.

I
Let G = (gij i,j = 1, ... , n) be the (Hermitian) matrix ofthe coefficients of
e,
H, and 1] the coordinate vectors of two tangent vectors €,.,.,. Then H(€,.,.,) =
e·G·ijt.
If we apply a holomorphic coordinate transformation

z = F(w) = (h(w), ... , fn(w)),

we obtain

This can be written in matrix form as

G(w) = JF(W) t . G(F(w)) . JF(W),

which is the transformation formula for Hermitian forms. Thus F·WH =
WF' H·SO the correspondence H H W is independent of the coordinates. The
form W is globally defined. In fact,

i,j

i . (e· G . ijt - 1] . G . t)
-2Im(H(€,.,.,)).
The form WH is called the fundamental form of the metric H.
316 VI. Kahler Manifolds

Definition. A Kahler metric on X is a Hermitian metric H with

dwH = O. If X possesses a Kahler metric H, the pair (X,WH) is called a
Kahler manifold. In this case we also say that X is a Kahler manifold.

In general, a Hermitian metric H does not satisfy the condition dWH = O.

There are many complex manifolds that cannot have a Kahler metric. How-
ever, on a Riemann surface X every 3-form vanishes. So every Riemann sur-
face is a Kahler manifold.

Geodesic Coordinates. Assume that H is a Hermitian metric on the

manifold X and that z = (Zl, ... , zn) are holomorphic coordinates around
some point Xo EX.
Definition. The coordinates z are called geodesic in Xo (with respect
to H) if H = Li,j gij dzid:Zj and
1. gij(XO) = bij, i.e. G(xo) is the unit matrix,
2. (gij k (xo) = 0 and (gij hv (xo) = 0 for 1/ = 1, ... ,n, i.e., (dgij )xo = O.

We will show that the existence of geodesic coordinates is a characteristic

property of Kahler metrics. For that we need a lemma.
3.2 Lemma. H = Li,j gijdzid:Zj is a Kahler metric if and only if

(The first equation implies the second one.)

PROOF: Let w = WH be the fundamental form of H. Then we have

dw i· (2.:(9iJk dzv 1\ dZi 1\ d:Zj +

I,J,1.I
2::
'I.,J , 1./
(gijhv d:Zv 1\ dZi 1\ d:Zj )

i. (~)L«9ij)ZV - (gvj)z,) dzv 1\ dzi ) 1\ d:Zj

} v<.

+ L(L«9ivh j - (9iJhJ d:Zv 1\ d:Z j ) 1\ dZi ).

• v<}

3.3 Proposition. H is a Kahler metric if and only if there are geodesic

coordinates around every point in X.

PROOF: We consider a fixed point Xo EX. If geodesic coordinates exist

around Xo, then all first derivatives of the gij are zero there, and dw = 0 at
Xo·
 3. Kahler Metrics 317

Assume now that H is a Kahler metric. We take a local coordinate system

around Xo such that Xo is the origin of this coordinate system and apply a
unitary transformation z = w· U t such that U t • G(O)· U = En. We denote
the new coordinates by z again and the new coefficient matrix of H also by
G such that now G(O) = En.
Finally, we apply a transformation

1
Zi = Wi + 2"w t
. Ai . w, i = 1, ... , n,

where the Ai = (aL I k, l = 1, ... , n) are constant symmetric matrices.

Then

So the Jacobian J of the transformation is given by J(w) = En + Lk Ak ·Wk,

(i, I
with
A = aJk i = 1, ... , n )
- k .
J= 1 , ... ,n .

The coefficient matrix G of H in the new coordinates is therefore given up

to first order by

G(w) = J(w) t . G(z) . J(w)

= (En + LAkwk) t. G(z)· (En + LA!WI)

k I

~ G(z) + L Ak . G(Z)Wk + L G(z) . A!Wj.

k I

Since (Zi)wj (0) = Oij, we get (G 0 z)w.>. (0) = G z .>. (0) for all A. So Gw.>. (0) =
G z ". (0) + Ai. The first derivatives of the functions 9ij have to be zero at o.
We can achieve this by setting

a;.\ := -(gji)z". (0).

Since dJJJ = 0, it follows that the matrices Ai are in fact symmetric (use the
equations of the lemma). So geodesic coordinates exist at Xo. •

Local Potentials. We assume that H is a Kahler metric on X and that

U is a polydisk in a local coordinate system. We have the fundamental form
w = i . Lid gijdzi 1\ azj of H with dw = O. Then by the Poincare lemma we
can construct a realI-form <P with d<p = win U. We have the decomposition
<p = <Po + <PI = Li ai dZ i + Li ai azi. Comparing types, we get 8<po = 8<PI = 0
318 VI. Kahler Manifolds

and acpo + 8CPI = w. Then, by Dolbeault's lemma, there are functions h o, hI

in U with 8ho = CPo and ah l = CPl. It follows that

i8a(i(ho - hI)) = a8ho + 8ah l = acpo + 8CPI = w.

The operator iaa = ~ddc is real. So we can replace h* := i (h o - ht) by
the arithmetic mean h of h* and its conjugate. Therefore, i8ah = w. The
real-valued function h is called a local potential of the Kahler metric H. It is
strictly plurisubharmonic, because

wah = i L hz;zj dZ i /\ azj

i,j

is the fundamental form of a Hermitian metric; Le., the Levi form of h is

positive definite.
In the other direction, if local potentials h exist for a Hermitian metric H,
then w = Wah is the fundamental form to H, and dw = o. So H is a Kahler
metric.

Pluriharmonic Functions. A real-valued function h on an open subset

U C X is called pluriharmonic ifi8a = O. In this case the I-form cp = 8h-ah
is closed (Le., dcp = 0) and purely imaginary (Le., of the form cp = i'!j;, with a
real form '!j;). In fact, we have 7p = -cpo
If U is simply connected and Xo a fixed point of U, we can define a function
9 by g(x) := IXXo 'l/J. Then dg = 'l/J = -icp, so 8g =
-i8h and ag = iah. Let
f := h + ig. It follows that
af = ah + iag = ah - ah = 0,

so f is holomorphic. Furthermore, df = 8f = 28h.

The Fubini Metric. The complex projective space IPn is one of the most
important examples of Kahler manifolds. Let

7f : (zo, ... , zn) t-+ (zo : ... : Zn)

be the canonical projection from C n+ l - {O} onto IPn. The unitary linear
transformations LA(Z) = z·A t ofC nH (with A E U(n+I), i.e., A E GLn(C)
and At. A = En) give biholomorphic transformations cp = CPA: pn -+ pn, by
CPA(7f(Z)) = 7f(LA(Z)). The set of these is the group PU(n + 1) of projective
unitary transformations. If Xo, Xl are arbitrary points in the projective space,
then there exists a projective unitary transformation cp with cp(xo) = Xl·
On C n +l we have the canonical strictly plurisubharmonic function

p(Zo, . .. ,zn) := zozo + ... + znzn,

 3. Kahler Metrics 319

which is positive outside of O. It is invariant by unitary linear transformations

ofCn +!.
If Xo E pn is a point, we can find a simply connected neighborhood U(xo)
and a holomorphic cross section f : U --* Cn +! - {O} for the submersion 1r.
The cross section is determined up to multiplication by a nowhere vanishing
holomorphic function h. We use log(po f) (or equivalently any function log(po
(h· f)) as local potential for a Hermitian metric. Since
i0810g(p 0 (h· f) = ia8[log(p 0 f) + log(h) + log(ii)] = ia810g(p 0 f),
we obtain a well-defined Hermitian form H on pn. Its fundamental form is
WH = ia810g(p 0 f).
We now show that H is positive definite.
Everything is invariant under the group PU(n + 1). If A is a unitary matrix,
CPA (Xo) = Xl, and f a holomorphic cross section in a neighborhood of Xl, then
1:= L Al 0 f 0 CPA is a section in a neighborhood of Xo. So po 1=
(po f) 0 CPA
is a local potential of H near xo.1t follows that H xo ({,11) = H X1 (cp.{,CP.11)·
Therefore, it is sufficiept to prove only the positive definiteness at the point
Xo = (1 : 0 : .,. : 0).
At this point a holomorphic cross section is given by

f: (zo : ... : zn) I-t (1, Zo

Zl, ... , Zn).
Zo
Therefore, we may take as potential the function
log(p 0 f(zo, . .. , zn» = log(l + tltl + ... + tntn )
= tltl + ... + tntn + terms of higher order,
where ti = zdzo are the (inhomogeneous) local coordinates at Xo. Therefore,
H({,11) = 6i1t + ... + {nfin' and this is positive definite. The metric H
constructed in this way is called the Fubini metric (or Fubini-Study metric)
on 1pm. Since H is given by local potentials, it is a Kahler metric.
3.4 Proposition. If X is a Kahler manifold, then any closed submanifold
Y C X is also a Kahler manifold.

PROOF: We can restrict the given Kahler metric H on X to the submanifold

Y. Then the restricted fundamental form wHly corresponds to the restricted
Hermitian form Hly, which again is positive definite, and wHly is also closed.
So Y is a Kahler manifold again. _
The following is an immediate consequence.
3.5 Theorem. Every projective algebraic manifold is a Kahler manifold.
320 VI. Kiihler Manifolds

Definition. Let X be an arbitrary n-dimensional complex manifold.

We say that meromorphic functions It, ... ,!k on X are analytically de-
pendent if
rkx(fl, ... ,fk) < k
at every point where each of the functions is holomorphic. They are called
algebraically dependent if there exists a polynomial p(Wl,.'" Wk) that
is not the zero polynomial such that p(lt, ... ,fn) vanishes identically
wherever it is defined.
X is called a Moishezon manifold if it is compact and has n analytically
independent meromorphic functions.

On compact manifolds algebraic and analytic dependence are equivalent no-

tions. In this case one can show that the field of all meromorphic functions
on X has transcendence degree at most the dimension n of X. For projective
manifolds there is equality. But a Moishezon manifold is in general not a
Kahler manifold. Moishezon's theorem states that
A Moishezon manifold X is projective algebraic if and only if it is a
Kahler manifold (see [Moi67], cf. also the article "Modifications" of Th.
Peternell, VII.6.6 in [GrPeRe94]). .

Deformations. There are many compact Kahler manifolds that are not
projective algebraic. There are even examples of Kahler manifolds where
every meromorphic function on them is constant.
The Kahler property is invariant under small holomorphic deformations, but
not under large ones (see [Hi62], and also [GrPeRe94], VII.6.5).
Definition. Assume that X is a connected (n + m )-dimensional com-
plex manifold, Y an m-dimensional complex manifold, and 7r : X -t Y
a proper surjective holomorphic map that has rank m at every point.
Then (X, 7r, Y) is called a holomorphic family of compact n-dimensional
complex manifolds.

Indeed, since 7r is a proper submersion, the fibers Xy = 7r- l (y), y E Y,

are compact submanifolds of X. When y varies, the fibers may have different
complex structures. However, they are isomorphic as differentiable manifolds.

3.6 Theorem. If Xo is a Kahler manifold and 7r : X -t Y is a deformation

of Xo (i.e., a holomorphic family with Xo = 7r- 1 (yo) for some Yo E Y), then
there exists a neighborhood U(Yo) C Y such that every fiber over U is again
a Kahler manifold.
On the other hand, it may happen that all fibers Xy except for XyO are Kahler
manifolds.
 3. Kahler Metrics 321

The first statement can be found in many books on Kahler manifolds (for
example in [MoKo71J). The original proof of the second statement was given
by H. Hironaka.

Exercises
1. Let X be a complex manifold and E a holomorphic vector bundle over
X. Use local frames as described in Exercise 2.3. A Hermitian form H
on E is an assignment of a Hermitian form Hx to each fiber Ex such that
(~i I~j) := H(~i,~j) is smooth for every local frame ~ = (6,.·· q ). Let ,e
h~ be the matrix h~ := ( ~i Iej ) ). Calculate a transformation formula
e
of h~ for changing from to another frame e'.
We call H a Hermitian
scalar product (or a fiber metric) on E if Hx is positive definite for every
x EX. In this case, prove that if Xo E X is fixed, then there is a local
frame at Xo such that h~(xo) = Eq and (ah~)xo = o.
2. For z E e n +1 - {OJ consider the linear map
cpz : en+! -t T1r (z) (IPn) ,
which has been defined in Section IV.5. Show that the Fubini-Study
metric H is given by H(cpz(v), cpz(w» = (v Iw).
For Z E en +1 with IIzll = 1 let Hz := {w E en+! : (zjw) = OJ. For
w E Hz with IIwll = 1 define
a(t) := 1r«cost)z + (sint)w).
Prove that a is a closed path of length 7r. Using the fact that every
geodesic on IPn is of this form, prove the formula

dist(7r(z), 7r(w» = arctan I(z tw) w - z I .

3. For decomposable k-vectors A = a1/\· .. /\ ak and B = b 1 /\· .. /\ bk define

(A I B) := det( (3; j bj ))
and IAI = (A IA)1/2. Prove that there is a Kahler form won Gk,n locally
given by w = iaaloglAI.
4. Assume that X is a connected complex manifold and that A c X is a
nowhere dense analytic set. Prove that if h is a bounded pluriharmonic
function on X - A, then it can be extended as a pluriharmonic function
toX.
5. Let Been be a closed ball. Is there a plurisubharmonic function h on
en - B with limz-+B h(z) = -oo?
6. A Kiihler metric H on a complex manifold X is called complete if X is a
complete metric space with respect to the associated distance function.
Construct a complete Kiihler metric on en - {OJ.
322 VI. Kahler Manifolds

4. The Inner Product

The Volume Element. Let X be an n-dimensional complex manifold
with a Hermitian metric H, and w = WH the associated fundamental form.
For Xo E X it is possible to choose coordinates Zl,'" ,Zn such that Xo is
the origin and gij(XO) = bij . Therefore, Hxo = 2:i dZiazi. These are called
Euclidean coordinates. We calculate the 2n-form w n in these coordinates at
Xo·
We have w = i '2:" dz" 1\ liZ", and so
n
wn (i. 2:)dx" + idyv) 1\ (dx" _ idyv)) n
v=l

A simple induction on k shows that

(t dx v 1\ dYv ) k = k! l~Vl <~vk~n dX V1 1\ dYvl 1\ ... 1\ dX"k 1\ dY"k'

It follows that

and

Remarks
1. In general coordinates one would get
wn = 2n n! g dXl 1\ dY1 1\ . .. 1\ dX n 1\ dYn,

where g = det(gij Ii, j = 1, ... ,n).

2. In the literature one finds other formulas:
If w is defined by w = 42:" dz v 1\ liZ", then
wn = n! dXl 1\ dYl 1\ ... 1\ dX n 1\ dYn.
If, in addition, the wedge product is defined in such a way that cp 1\ 1j; =
~(cp ® 1j; -1j; ® cp) for I-forms cp and 1j;, then w = -~ 1m Hand
 4. The Inner Product 323

Definition. The 2n-form

1 1
dV:= __ w n = --(-2ImH)n
2nn! 2nn!
is called the volume element associated to H.

The traditional notation "dV" does not mean that the volume element is the
differential of some (2n-l)-form. It is easy to see that this is impossible, e.g.,
on compact manifolds. Moreover, in the case of a compact Kahler manifold
(X, w) it follows that f x wn = 2nn! f x dV > O.
4.1 Proposition. If (X,w) is a compact Kiihler manifold, then w k defines
a nonzero class in the de Rham group H2k(X) for 1 :::; k :::; n = dim(X).

PROOF: It is clear that d(w k ) = O. Suppose that there is some (2k-l)-form

cp with dcp = wk. Then d( cp 1\ w n - k ) = wn , and by Stokes's theorem

This is a contradiction. •
The Star Operator. Let Zl, •.. ,Zn be Euclidean coordinates at Xo EX.
We use real coordinates in the following form:

Then any r-form cp can be written as cp = 2:1 a1 dU1, where the summation
is over all ascending sequences 1= (i b ... , i r ), 1:::; il < ... < ir :::; 2n.
Definition. The (Hodge) star operator * : dr(X) -+ d 2n - r (x) is
the C-linear map defined by

where I U I' = {I, ... , 2n}, I n I' = 0, c1,1' = ±1, and

The star operator depends on the metric (which determines Euclidean coor-
dinates) and on the orientation (which is determined by the order of the real
coordinates), but on nothing else. So it is globally defined and invariant under
unitary coordinate transformations (in the sense that * (cp 0 F) = (* cp) 0 F
for these transformations).
324 VI. Kahler Manifolds

4.2 Proposition.
1. If cp is an r-form, then * * cp = (-It· cpo
dV if 1= J,
2. dUI 1\ * dUJ = { 0
o th enmse.
.
3. * is real; i. e., * (j5 = *cp.
PROOF: (1) We have * * dUI = cI',IcI,!' du], and

C]',]C],1' du], 1\ du] cI',!cI,!'( _lr(2n-r) dU]1\ dul'

= cI',l(-lt dV
= (-It dUI' 1\ dUI.
(2) dUI 1\ * dUJ = cJ,JldUI 1\ dUJI = 0 if I n JI =J 0, and = dV if JI = II.
(3) We have *(Re cp + i 1m cp) = *(Re cp) + i *(Im cp), by definition. _
If cp = LI aldul and .'IjJ = LJ bJduJ, then

cp 1\ *'IjJ = L a]bJdu]1\ * dUJ = (L alb] )dV.

],J ]

In particular, this expression is symmetric in cp and 'IjJ.

Now we define *(cp) := * (j5. This is a complex antilinear isomorphism between
$'r(x) and $'2n-r(x).

We get cp 1\ *'IjJ = (L] albl ) dV and cp 1\ *cp = (L]laI12) dV.

The Effect on (p, q)-Forms. We calculate"* cp for (p, q)-forms cpo
4.3 Lemma. If cp = dZ I 1\ cli J with III = p and 111 = q, then

*
cp 1\ cp = 2p +q dV.

PROOF: We write I = M U A and J = M U B, with pairwise disjoint sets

A,B,M c N:= {I, ... ,n} such that dZ]l\cli J = COdZA 1\ cli B I\wM, where
WM = TIJ.'EM dzJ.' I\cliJ.' and lcol = l.
Let m := IMI. Then WM = (-2i)mTIJ.'EMdu2J.'-1 1\ dU21-'" The remainder
dZ A 1\ cli B is a form of type (p - m,q - m).

If, for example, A = {al, ... , as} (with s =p - m), then

dZA (dU2al-l + idu2aJ 1\ ... 1\ (dU2as-l + idu2a.)

L L
S

is - v C1cduac 1\ dUa'c'
v=O ICI=v
 4. The Inner Product 325

where C C {I, ... , s} is an ascending sequence of length IGI, ac the sequence

of the 2ai - 1, i E C, and ao the sequence of the 2ai, i E C; ac is a number
equal to ±l. So we have L:~=o C) = 28 summands. The analogue will be
obtained for azB.
Altogether we see that cp = dz[ f\ azJ is equal to 2m times 2P+q - 2m different
monomials CK dUK with ICKI = 1. So

cp f\ "* cp = (2:)2m . CKI2) dV = 2p +q dV.

K

•
Again let cp = dz[ f\ azJ with III = p, IJI = q and p + q = r. Then * cp is a
(2n - r)-form, and it can be written as

v+Jl=r IKI=n-v
ILI=n-Jl

with certain constants aKL.

Now we apply the unitary transformation F(z) = z· C t , with

c= (e itl
'. 0), ti ElR for i = 1, ... ,n.
o e itn

For I = {il' ... , ip} C N = {I, ... , n} we set t J := til + ... + tip' Then
F*cp = exp(i(tJ - t.r))· cp

and

Since the star operator is invariant under unitary transformations, we have

[t follows that

* cp = exp(i(tJ - tJ))' L L aKL . exp(i(tK - tL)) azK f\ dz L ·

v+l-'=rK,L

The uniqueness of the representation implies that

exp(i(tJ + tK - tJ - td) = 1 for all K, L.

326 VI. Kahler Manifolds

Arbitrary ti are allowed. If we choose them all very small, then t1 + tK =

tJ + tL·
If there is a pair (K, L) with I U K i- J U L, then for example there is an
i E I that is not contained in JuL. Setting ti i- 0 and tj = 0 for j i- i gives
a contradiction. Therefore, p + (n - v) = q + (n - J.L). On the other hand, we
have v + J.L = p + q. This is possible if and only if q = J.L and p = v. * cp is a
form of type (n - q, n - p), and we can write

*CP= L aKLazKl\dzL ·
IKI=n-p
ILI=n-q

We consider an index i E I U J. There are three possibilities:

1. If i E I and it/. J, then i must lie in L, since we always have IuK = JUL.
We choose ti = e (positive, but very small) and tj = 0 for j i- i, and we
consider the equation t1 + tK = tJ + tL. Since t1 = e, tK ~ 0 and tJ = 0,
it follows that tL ~ e, and therefore even tL = e. But then tK = o. This
implies that i ELand i t/. K.
2. If i E J and i t/. I, then analogously i E K and i t/. L.
3. If i E In J, then cP contains the term dZi 1\ azi and therefore the term
dU2i-l 1\ dU2i. This implies that dU2i-l 1\ dU2i (and therefore dZ i 1\ azi)
cannot occur in * cpo
So I u K = J u L = {I, ... , n} and * cp = a azl' 1\ dZJ', with some constant
factor a. We may assume that e1,1' = eJ,J' = 1.
4.4 Proposition. In the above notation
"*(dZ1 1\ azJ) = adz!, 1\ dzJ' with a = 2P+q-nin(_1)q(n-p)+n(n+l)/2.

PROOF: For cp = dZ1 1\ azJ we have

cp 1\ "* cp dZ1 1\ azJ 1\ (a dzl' 1\ azJ')

= a( _1)q(n- p) dZ1 1\ dZ1' 1\ azJ 1\ azJ'
a( -1 )q(n- p)+n(n-l)/2 (dz 1 1\ azd 1\ ... 1\ (dzn 1\ azn )
a( _1)q(n- p)+n(n-l)/2( -2i)n dV
= a( _1)q(n- p)+n(n-l)/2+ n 2n i n dV.

On the other hand, we know from the lemma that cp 1\ "* cp = 2p+q dV. Com-
paring the coefficients, it follows that a = 2P+ q- n i -n( _1)q(n- p)+n(n+l)/2 .
•
Remark. The formula is valid only with respect to Euclidean coordinates.
And one has to observe the rule that e1,1' = eJ,J' = 1.
 4. The Inner Product 327

The Global Inner Product. We work on a compact Hermitian man-

ifold X. Let ~ = {U1 , ... , UN} be a finite open covering of X with coordi-
nates, and (l?d a partition of unity for ~. If t/J is any differentiable 2n-form
on X, we define the integral

(In local coordinates l?vt/J has the form C(ZI' ... ' zn) dV with a differentiable
function c with compact support. We already know the meaning of the inte-
gral over such a 2n-form.) One can show that this definition is independent
of the choice of the partition of unity.
*
If cp is an r-form, then t/J = cp /\ cP is a 2n-form that locally is equal to a form
cdV, with c :::: 0 and c(x) = 0 if and only if CPx = o.
Definition. The inner product of two r-forms cp, t/J on X is defined by

4.5 Proposition.

1. (CPI + CP2 , t/J) = (CPI , t/J) + (CP2 , t/J).

2. (c· cP, t/J) = c . (<p, t/J).
3. (t/J, cp) = (cp, t/J).
4· (cp, cp) :::: 0, and (cp, cp) = 0 if and only· if cP = o.

PROOF: (1) and (2) are trivial.

(3) We have *
t/J /\ cP = t/J /\ * cP = cP /\ * t/J.
°
(4) It is clear that (cp, cp) :::: always. If cP i 0, then there is a point Xo E
X with CPxo i O. If Xo E Uv and l?v(xo) > 0, then we can find a small
neighborhood V = V(xo) CC UV such that l?vCP = cdV on V, with c > 0
everywhere in V. It follows easily that (cp, cp) > o. _
Since ( ... , ... ) is a Hermitian scalar product on Jl1r (X), a norm on this space
is defined by IIcpli := (cp, cP //2.
4.6 Proposition. Two forms of different type are orthogonal to each other.

PROOF: Let cP, t/J be two forms of type (p, q), respectively (s, t), such that
*
p + q = s + t = r. Then cP /\ t/J is a form of type (p + n - s, q + n - t). If
s > p, then q > t and therefore q + n - t > n. If s < p, then p + n - s > n.
*
So cP /\ t/J = 0 in these cases. _
328 VI. Kahler Manifolds

Unfortunately, Jtlr(x) is not complete in the induced topology, so that we

cannot apply Hilbert space methods directly. We will show how to construct
a closure of Jtlr(x).

Currents. Let X be a compact Hermitian manifold. A sequence rpv =

EI aI,v dUI of (smooth) r-forms on X is said to be convergent to zero in
Jtlr(x) if for every a all sequences DaaI,v are uniformly convergent to zero.
Definition. A current of degree 2n-r is an JR.-linear map T : Jtlr(x) 4
C such that if (rpv) is a sequence of r-forms converging to zero, then T( rpv)
converges to zero in C.

The set of all currents of degree 2n - r is denoted by Jtlr(x)'.

Examples
1. A current of degree 2n is a distribution in the sense of L. Schwartz.
2. Let 'IjJ be a differential form of degree 2n - r. Then 'IjJ defines a current
T,p of degree 2n - r by

It is easy to see that T,p is, in fact, a current. So Jtl2n-r(x) C Jtlr(x),.

3. Let M c X be an r-dimensional differential submanifold. Then a current
TM is defined by TM [rp] := JM rp. Therefore, we say that a current of
degree 2n - r has dimension r.
Let (rpv) be a sequence in Jtlr(x) converging to some r-form rp (Le., rpv -rp 4
0). Then also *(rpv - rp) 40, and therefore T,p[*(rpv - rp)] 40 in C for every
'Ij; E Jtlr(x). It follows that

lim ('IjJ, rpv) = lim { 'IjJ 1\ *rpv = lim T,p[*rpv]

v~oo v~ooJx v~oo

= T,p(*rp] = Ix 'IjJ 1\ *rp = ('IjJ, rp).

Definition. For T E Jtlr(x)' and rp E Jtl2n-r(x) we define

In particular, we have (Tv>, rp) = ('IjJ, rp). If 'ljJu -t 'Ij;, then (T,p, rp) =
limv~oo ('ljJu, rp). This motivates the following.
 5. Hodge Decomposition 329

Definition. If (1/J,,) is a Cauchy sequence in .Q{2n-r(x), then T

lim v ---+ oo 1/Jv E .Q{r(x)' is defined by

The limit T is called a square integrable current.

We omit the proof that T is actually a current.

Before defining the notion of a limit of a sequence of currents we note that

T = lim 1/Jv {:=:} (T, 4') - (1/Jv, 4') -+ 0 for all 4' E .Q{2n-r(x)
1.'---+00

Definition. A sequence (Tv) of currents in jij'r(x)' is called convergent

to a current T E jij'r(x)' (written as T = lim v ---+ oo Tv) if

We state the following without proof.

4.7 Theorem. The space jij'r (X)' of currents of degree 2n - l' is complete;
i. e., every Cauchy sequence converges. The set of square integrable currents
-2n-r
forms a complete subspace of jij'r(x)', which we call the closure .Q{ (X)
of the space of (2n - r)-forms.

Exercises
1. Compute the volume of the complex projective space IPn with respect to
the Fubini metric, and the volume of a p-dimensionallinear subspace.
2. Let (X,w) be a compact Kahler manifold and Y c X a closed subman-
ifold with dim(Y) = m. Then vol(Y) = c . Jy
wm for some constant c.
Calculate this constant!
3. Try to define the product of a current and a differential form!
4. For T E jij'r(x)' define dT E jij'r+I(X)' by dT[4']:= (-It+ I T[d4']. Show
that dT is in fact a current with d(dT) = O. Calculate dT for the case
that T is a form or a submanifold.

5. Hodge Decomposition
Adjoint Operators. A complex vector space with a Hermitian scalar
product is called a unitary vector space. Let VI, V2 be two unitary vector
330 VI. Kahler Manifolds

spaces, and T : VI -+ V2 an operator (i.e., a C-linear mapping). If there

exists an operator T* : V2 -+ VI with

(T(v) , w) = (v, T*(w» for v E VI, wE V2 ,

then T* is called an adjoint operator for T. It is clear that T* is uniquely

determined, but in general it may not exist.
If Vl, V2 are unitary vector spaces and T : VI -+ V2 is a continuous operator,
then it follows from the Cauchy-Schwarz inequality that v r l (T(v) , w) is
continuous for any fixed w E V2 • If VI, V2 are Hilbert spaces, then it follows
by the Riesz representation theorem that there exists an element T* ( w) with
(T(v) , w) = (v, T*(w», and that w r l T*(w) is continuous.

Now let X be an n-dimensional complex manifold.

5.1 Lemma. If X is compact and cp a (2n-l)-form on X, then Ix dcp = O.

PROOF: Let %' = {Uv : v = 1, ... , N} be a finite open covering of X by

local coordinates and ([Iv) a subordinate partition of unity. We choose open
subsets U: c c Uv with piecewise smooth boundary and sUPP([lv) c U:.
Then

1 x
dcp = Lv 1 d([lvcp) = Lv 1 d([lvcp) = Lv 1
u~ u~ au~
[lvCP = 0,

by Stokes's theorem. •
5.2 Proposition. The operator 6 := -* d"* : .91r +1(X) -+ .91 r (X) is the
adjoint of d:.91 r (X) -+ .91r + 1 (X).

PROOF: Let cP E .91 r (X), 'IjJ E .91 r + I (X) be two arbitrary forms. Then
cP A *'IjJ is a (2n - I)-form, and

Therefore,

(dcp, 'IjJ) Ix dcp A *'IjJ = Ix *

d( cp A 'IjJ) - (-1 Ix
r cp A d* 'IjJ

(_lr+1+2n- r IxcpA**d"*'IjJ = - Ix cpA*(*d"*'IjJ)

(cp, -*d"*'IjJ).

•
 5. Hodge Decomposition 331

5.3 Proposition. The operator?J := -*

&* : d P + 1 ,Q(X) -t dp.q(X), re-
spectively -:a := -*En : d P ,Q+1(X) -t dp.q(X), is the adjoint of 0, respec-
tively 8.

PROOF: We consider only the first case. Let r.p be a form of type (p, q) with
p + q = r and 1/J a form of type (p + 1, q). Then r.p 1\ * 1/J is a form of type
(n - 1, n). Therefore, 8(r.p 1\ *1/J) = 0 and

d(r.p 1\ *1/J) = o(r.p 1\ *1/J) = or.p 1\ *1/J + (-lrr.p 1\ o(*1/J).

It follows that

(or.p, 1/J) = Ix or.p 1\ * 1/J = Ix d( r.p 1\ * 1/J) - (-1 r Ix <P 1\ 0(* 1/J )

= (_lr+ H2n - T
Ix r.p1\**o(*1/J) = (r.p, -*&*1/J).

•
5.4 Proposition. We have
1. t5 = 19 +-:a'
2. 8t5 = 0, iH} = 0 and 1919 = 0,
3. ?J-:a + -:a?J = 0,
4. ?Jr.p = -:a("fj5) and -:ar.p = ?J("fj5).
The proofs are trivial.

The Kahlerian Case. Now let X be an n-dimensional Kiihler manifold.

For the moment X is not assumed to be compact. A canonical operator
L: Ja"T(X) -t Ja"T+2(X) is defined by

where w is the fundamental form. If r.p is a (p, q)-form, then Lr.p is a (p+ 1, q+ 1)-
form. Since cU.,; = 0 and w is of type (1,1), we also have ow = 8w = O. It
follows that

and
- 1- 1- -
oLr.p = 2 o(r.p 1\ w) = 2 o<p 1\ w = Lor.p.
In addition, we have
Lr.p = Dfj5.
332 VI. Kahler Manifolds

5.5 Proposition. If X is a compact (Kiihler) manifold, then the adjoint

operator A: .rz(r+2(x) -+ .rz(T(X) of L is given by

A = (-lr"*L"*.

PROOF: If 'I/J is an (r + 2)-form, then w 1\ "*'I/J is a (2n - r)-form, and

(L'P, 'I/J) = rx 'P

~2 J 1\ w 1\ "* 'I/J r
= ~2 J (-1 r'P 1\
x
** w 1\ "* 'I/J
('P, (-lr"*L"*'I/J).

•
Note also that A is a real operator: A'P = Acp.
5.6 Proposition. The commutativity relations· iJ A AiJ and -:aA = A-:a
hold.
We leave the proof to the reader.

Bracket Relations. Let k,l be integers, and 8 : .rz(T(X) -+ .rz(r+k(x),

T : .rz(T(X) -+ .rz(r+l(x) linear operators that are defined for every r. Then

[8, T] := 80 T - T 0 8: .rz(r(x) -+ .rz(r+k+l

is called the bracket or the commutator of 8 and T.

Example
As we have seen above, iJ, -:a : .rz(r -+ .rz(r-l and A : .rz(r -+ .rz(r-2 have the
brackets [iJ, A] = [-:a, A] = 0 (as operators from .rz(r to .rz(r-3). Before we
calculate more brackets, we have to calculate iJ'P very explicitely for a form
'P of type (p, q). For that we need geodesic coordinates. So everything that
follows is valid only on Kahler manifolds.
Let 'P = ~/,J al J dZI 1\ azJ be an arbitrary (p, q)-form. We consider geodesic
coordinates at a point Xo E X such that w = j ~v dz v 1\ az v . Then

I,J I,J

where a(p,q) = 2P + Q - n jn(_1)q(n- p )+n(n+1)/2, and C/,l' = c/',J' = 1. In a

neighborhood of Xo the formula is more complicated, because the coefficients
gij of the metric are involved. We have to apply 7j in this neighborhood, but
because we are working with geodesic coordinates, we obtain the exact value
at Xo by simply applying 8 to our representation of"* 'P at Xo. It follows that

8* 'P = a(p, q) . (_1)n-p L L (aIJ )z, dzl' 1\ azv 1\ azJ',

I,J tEJ
 5. Hodge Decomposition 333

Now we need some notation.

If J = (j1, ... ,jq), then J(v) := (jI, ... ,iv, ...
,jq) for v = 1, ... , q. Further-
more, let I* and J*(v) be arrangements of I, respectively J(v), such that
el',!< = e(jv,J'),J*(v) = l.
Then dzlo = (-l)p(n- p) dZ I and clzr(v) = (-l)n-q+q(n-q)+v-lclzJ(v) , and
we get

-;gcp = -*8*cp
q

= -a(p, q) . a(n - p, n - q + 1) . (_l)n-p L L(aIJ )z;", dz l * A clzJ*(v)

I,J v=1
a(p, q) . a(n - p, n - q + 1) . (_1)n-p+p(n-p)+n-q+q(n-q)
q

x L L( -It(aIJ )z;v dZI A clzJ(v)

I,J v=1
q

= 2· (-l)P L L( -It(aIJ )z;v dZI A clzJ(v)'

I,J v=1

5.7 Theorem. On a compact Kiihler manifold [~, L] = i 8.

PROOF: In order to calculate -;gLcp and L-;gcp for a (p, q)-form cp we begin
with
1
Lcp = iW A cp (since cp A w = w A cp)

= (~t )..=1
dz).. A clz)..) A ( L aIJ dZI A clzJ)
1,J

= ~(-1)P L L aIJ dz).. A dZI A clz).. A clzJ.

I,J )..EI'nJ'
From that we get
-;g Lcp =
= i(-1)P+1 L (-l)p+1(aIJ)z>.dz).. AdzIAclzJ
L1,J )..EI'nJ'
q
+ i (_1)P+1 L L L( _l)p+v+l(aIJ )z;v dz).. A dZI II clz).. II clzJ(v)
I,J )..EI'nJ' v=1
iLL (aIJ )z). dz).. A dZI A clzJ + CPo,
I,J )..EI'nJ'

with
334 VI. Kahler Manifolds

CPo := i . L L L( -1)"(alJ )Zjv dz).. A dZI A az).. A azJ(v).

I,J )..EI'nJ' 1'=1

On the other hand,

1 -
L -:acp = iW A 1Jcp

( i tdZ)..l\az)..) A (-l)PLt(-I)"(aIJ)zjV dz] AazJ(v)

)..=1 I,J 1'=1
q
= iLL L (-I)"(alJ)Zjv dz).. AdzIAaz)..AazJ(v)
I,J 1'=1 )..El'nJ(v)'
q
= i L L( -1)"( _1),,-1 (aIJ )Zjv dz jv A dZ I A azJ
I,J 1'=1
q

+iL L L(-l)"(aIJ)Zjv dz).. AdzI Aaz)..AazJ(v)

I,J )..EI'nJ' 1'=1

= -iL L (aIJ)z>.dz).. AdzI AazJ + CPo·

I,J )..El'nJ

In total we obtain

-:aLcp - L-:acp = i L L (aIJ )Z>. dz).. A dZ I A azJ = i&cp.

I,J )..EI'

•
5.8 Corollary. [1J,L] = -i8, [&,A] = -i19 and [8, A] = i1J.
The proof is an easy exercise.

The Laplacian. Let X be an n-dimensional compact complex manifold.

Definition. The (real) Laplacian ~ : s;(r(x) -t s;(r(x) is defined by

~cp := (d <5 + <5 d)cp.

The complex Laplacian 0 : s;(p,q(X) -t s;(p,q(X) is defined by

o cp := (819 + -:a8)cp.
Since d and <5 are real operators, ~ is also real.
5.9 Proposition. On a compact Kahler manifold, 0 = D.
 5. Hodge Decomposition 335

PROOF: We have 8"J = -"J8 and 8iJ = -iJ8, as can be seen from the
bracket relations. Therefore,

-i 0 -i(8"J + "J8)
8( -i"J) + (-i "J)8
8(8A - A8) + (8A - A8)8.
From this one easily derives that - i 0 = i O. •
A further consequence is

i 0 = 8(8A - A8) + (8A - A8)8 = i(8iJ + iJ8).

Therefore, 0 = 8iJ + 08 and Ocp = (8iJ + iJ8)cp = Ocp; i.e., 0 is a real
operator.
5.10 Proposition. On a compact Kahler manifold, 0 = !.D..

PROOF: We have

.D. do + od
(8 + 8)(iJ +"J) + (iJ + "J)(8 + 8)
(8 iJ + iJ 8) + (8"J + "J 8)
20.

•
This formula is not valid on general compact manifolds!

Harmonic Forms. Let X be a compact Hermitian (not necessarily

Kahler) complex manifold. We define
8r(X) ._ d.f21 r - 1 (X),
~W(X) ._ o.f21 r + 1 (X),
£r(x) ._ {cp E .f21r(x) : .D.cp = a}.

The elements of £r (X) are called harmonic forms.

Example
For calculating the star operator on (p, q)-forms one needs the number
a(p,q) = 2P+Q-nin(_1)q(n-p)+n(n+l)/2. Ifn = 1, then a(l,O) = -i, a(O, 1) = i
and a ( 1, 1) = - 2 i. It follows that in geodesic coordinates

.D.1 = odf -*d*(/z dz + fzdZ)

-* d(ifzdZ - ifz dz)
-*(2ifzz dz /\ dZ)
-4fzz = - fxx - fyy·
336 VI. Kahler Manifolds

This motivates the name "harmonic."

5.11 Proposition.
1. cp E.fr {::::::} dcp = Jcp = o.
2. The vector spaces f!lJ r , ~r, and,yt'r are mutually orthogonal.

PROOF: (1) If D:.cp = 0, then (cp, dJcp) + (cp, Jdcp) = O. But (cp, dJcp) =
(Jcp, Jcp) and (cp, Jdcp) = (dcp, dcp). Therefore, dcp = Jcp = o.
(2) We have (dcp, J'fj;) = (cp, JJ'fj;) = 0 for all cp,'fj;. And if D:.'fj; = 0, then
(dcp, 'fj;) = (cp, J'fj;) = 0, and analogously, (JQ, 'fj;) = (Q, d'fj;) = 0 for all CP,Q.
So the spaces are mutually orthogonal. _

5.12 Hodge theorem. If X is a compact Hermitian manifold, then

PROOF: We give only a sketch of the proof. For more details see [dRh84],
[Schw50], or [GriHa78], and for a very explicit and rather elementary proof
see [War71].
(1) If we assume that .ll1 r is finite-dimensional, then the proof is very easy. We
let V be the orthogonal complement of f!lJr + ~r. Then .ll1r = f!lJr EB (iJr EB V,
and it follows immediately that D:.cp = 0 for every cp E V.
(2) If .ll1 r is infinite-dimensional, then we consider its closure, the Hilbert
space .z2 of square integrable currents. We can define derivatives of currents
by
aT [ acp ]
au)cp] := -T au v .
So it is possible to speak of harmonic currents (Le., currents T with D:.T = 0).
Now, the Laplacian is an elliptic operator, the definition of which we now
recall.
A linear differential operator P has locally (in real coordinates UI, . . . ,un)
the form
P = L aa(u)Da : 'i&'oo(U,RM) -+ 'i&'OO(U,R N ),
\a\9
with (N x M) matrices aaand Da = a\a\/(auf 1 •• ·au~n). Then the symbol
of P at u is the linear map

ap(u,e) := L aa(u)e a : RM -+ R N ,
\a\=k
 5. Hodge Decomposition 337

e
which depends on E IRn - {o}. The operator P is called elliptic if ap(u, e)
e
is injective for every u and every =/: O. Since aA(U,e) = -lIeIl 2 .id, it follows
that Ll is elliptic.
We state the regularity theorem:
Let P : dr(X) -+ dr(X) be an elliptic operator, and <p an element of
dr(X). If there is a square integrable current T with P(T) = <p, then
there is an r-form 'Ij; with T = T",.
If "Y is the orthogonal complement to ~r aJ ~r in .:£2, then simple Hilbert
space theory implies that .:£2 = !JjJr aJ ~r aJ "Y, and the sum is orthogonal.
Just as in the finite-dimensional case it follows that Ll(1") = 0, and using
approximation of currents by forms one sees that every harmonic current
belongs to "Y. From the regularity theorem it follows that "Y = ,Yt'r.
Finally, one shows that if 'Ij; is an r-form and T", = T + S + T<p, with a
harmonic form <p and currents T E ~r, S E ~r, then there are also forms
T E ~r, a E ~r such that 'Ij; = T + a + cp. _

5.13 Theorem. If X is a compact Hermitian manifold, then Hr(x) =

£,r(x).

PROOF: The de Rham group Hr(x) is the quotient of the closed forms
modulo the exact forms. Let <p E dr(X) be given with dcp = O. Then we
have a unique decomposition cp = d'lj; + 6{! + Hcp, with Ll(Hcp) = O. Now,

(6{!, 6(! ) (<p, 8 (! ) (because of the orthogonality)

= (dcp, (!) = O.

So cp = d'lj; + Hcp, and we define h : Hr(x) -+ ,Yt'r(x) by h : [cp] t--+ H<p.

This is well defined, and since cp = d'lj; ~ H cp = 0, it is injective. If a is
harmonic, then da = 0 and Ha = a. So h is surjective. _

Remark. By the theorem of de Rham we have Hr(x) ~ Hr(x, C). There-

fore, f3r(X) := dimc(,Yt'r(X)) is a topological invariant. We have

the rth Betti number. In Chapter IV only bi was introduced, since we did
not work with higher cohomology groups.

5.14 Theorem (Poincare duality). On every n-dimensional compact

Hermitian manifold we have Hr(x) ~ H2n-r(x).
338 VI. Kahler Manifolds

PROOF: It is easy to see that * ~ = ~ *. Therefore, * : £r(x) -+

£2n-r(x) is an isomorphism (depending on the metric). Then we apply
the previous theorem. _

Remark. The original Poincare duality states that there is an isomorphism

Hr(x,7L.) -+ H 2n - r (X,7L.). Our theorem here is a very weak version of this
topological result.

Consequences. Let X be an n-dimensional compact Hermitian manifold.

We define subspaces of dp,q(X) by
~p,q

~p,q .- 19dP,q+l,
£p,q {rp E dp,q : Orp = o}.
We have Orp = 0 ~ arp = 19rp = 0, and therefore the three spaces are
mutually orthogonal.
5.15 Decomposition theorem for (p, q)-forms. We have
dP,q = f!8 p,q EB ~p,q EB £P,q.

The proof is the same as in the case of r-forms. One needs the ellipticity of
0, which can be shown directly (and is trivially given on Kahler manifolds,
since then 0 = ~~).
5.16 Theorem .. Hp,q(X) ~ £p,q(X) for all p,q.
The proof is the same as for Hr(x).
5.17 Finiteness theorem. dime Hp,q(X) < 00 for all p, q.
Using the theory of elliptic operators one can prove that £P,q is finite-
dimensional. This implies our result. On the other hand, we know al-
ready from Dolbeault's theorem that HP,O(X) ~ np(X) = HO(X, n~) and
HP,l(X) ~ Hl(X,n~), where n~ is the complex analytic bundle of holo-
morphic p-forms. By the results of Chapter V these cohomology groups are
finite-dimensional. This can be generalized to higher cohomology. So we have
two different ways to prove the finiteness theorem.
5.18 Serre-Kodaira duality theorem. On an n-dimensional compact
Hermitian manifold Hp,q(X) ~ Hn-p,n-q(x) for all p, q.

PROOF: It is easy to see that *0 = 0*. Therefore, * defines an isomor-

phism from £P,q onto £n-p,n-q. _
As mentioned above, Poincare duality can be proved by topological means.
This is not possible for the Serre-Kodaira duality.
 5. Hodge Decomposition 339

Now we consider applications to Kahler manifolds.

5.19 Theorem. If X is an n-dimensional compact Kahler manifold, then
1. £r(x) = EB £p,q(X).

PROOF: (1) If <p E £r, then there is a unique decomposition

<p = L <p(p,q) ,
p+q=r

with <p(p,q) E JlIp,q. It follows that

o = t::.<p = L t::.<p(p,q).
p+q=r

If X is a Kahler manifold, then t::. = 20. Therefore, t::.<p(p,q) is a form of type

(p, q). The uniqueness of the decomposition then implies that t::.<p(p,q) = 0 for
all p,q. So <p(p,q) E £p,q(X).
(2) Since 0 is a real operator on Kahler manifolds, complex conjugation
defines an isomorphism from £P,q to £q,p. •

5.20 Decomposition theorem of Hodge-Kodaira. For every compact

K iihler manifold and every r,

Hr(x) ~ EB Hp,q(X).
p+q=r

In particular,

The theorem of Hodge-Kodaira follows immediately from the decomposition

theorem on harmonic forms.
We define f3p,q(X) := dime Hp,q(X). Then we have the equations
1. f3r = Lp+q=r f3q,Tl
2. f3p,q = f3q,p,
3. f3p,q = f3n-p,n-q,
4. f3r = f32n-r'

5.21 Corollary. If r is odd, then f3r(X) is even.

PROOF: We have
340 VI. Kiihler Manifolds

p+q=r p+q=r p+q=r

p<r/2 p>r/2 p<r/2

•
5.22 Proposition. If L : J2f'r(x) -+ J2f'r+2(x) is the map given by L(cp) =
~cpt\w, then U(I) E ,Yer,r.

PROOF: We have a(U(I)) = Lra(I) = 0, and

19(Lr(1» = (L19 + i a)U-1(I) = L19Lr- 1 (I) + i U-1a(I) = L19U- 1 (I).

Repeating this, we finally have 19(Lr(I» = L r19(I) = o. Thus U(I) is har-
monic. •
5.23 Corollary. If r is even, then {lr(X) 1= o.
PROOF: We have

Lr(1) = G) r wr = cdz 1 t\ clzl t\ ... t\ dZ r t\ clzr.

with c 1= 0 for 0 ~ r ~ n. Therefore, ,Yer,r 1= O. Since ,Yer,r C ,Ye2r, it follows

that {l2r(X) 1= o. •
This gives a topological condition for Kahler manifolds.
Examples
1. Every Riemann surface X is a Kahler manifold. If X is compact, then
9 := ~bl(X) is called the genus of X. For example, an elliptic curve (a
I-dimensional torus) has genus 1, since its homology is generated by two
independent cycles.
Now we have {l1,O + {lO,1 = bi = 29 and {l1,O = {lO,I. SO {l1,O = {lO,1 = g,
Le.,
9 = dime H1,o(X) = dime HO(X,K x ).
The genus is the number of independent holomorphic I-forms on X.
2. Let H be an n-dimensional Hopf manifold, n ~ 2. Then bi = 1, as we
have shown in Chapter IV. So H cannot be a Kahler manifold!

Exercises
1. Prove the commutativity relations iJA = AiJ and 19A = A19.
2. Prove the bracket relations [iJ,L] = -ia, [a, A] = -i19, and [a,A] =
i iJ.
 6. Hodge Manifolds 341

3. Let X be an n-dimensional compact complex manifold. Use the norm

11<p11 = (<p, <p)1/2 on dr(X) and prove for a given cohomology class u E
Hr(x) that among all representatives <p E dr(X) of u a representative
<Po has minimal norm if and only if b<po = O.
4. Calculate Hp,q(X) for X = pn and all p, q.

6. Hodge Manifolds
Negative Line Bundles. Assume that X is an n-dimensional compact
complex manifold and that F is a holomorphic line bundle on X. We define a
new notion of negativity for F. The new definition is formally more restrictive,
and if we would generalize it to arbitrary vector bundles, it would, in fact,
be different from our old notion. But it turns out that for line bundles the
new negativity is equivalent to the old one.
Let a system of trivializations <p, : Flu, -+ U, x C (with transition functions
gUt) be given. A fiber metric on F is given by a system h of positive smooth
functions h, such that

Then a smooth function Xh : F -+ IR can be defined by

Definition. A line bundle F over a compact manifold X is called

Griffiths negative if there exists a system of trivializations <p, : Flu, -+
u, x C (with transition functions gUi)' and a fiber metric h = (h,) on F
such that Xh is strictly plurisubharmonic on F - ZF (where ZF denotes
the zero section in F).

6.1 Proposition. The line bundle F is Griffiths negative if and only if

there exists a system of positive smooth functions e, on U, such that:
1. - log e, is strictly plurisubharmonic on U,.
2. e, = Igu< I . (j", on U,,,,.

PROOF: Let h be any fiber metric on F and consider a point Xo EX. There
is a trivialization <p at Xo such that h",(xo) = 1 and all first derivatives of h",
vanish at Xo. Then h",(x) . Iwl 2 is strictly plurisubharmonic at every (xo, w)
with w =1= 0 if and only if h", is strictly plurisubharmonic at Xo. Taking such
trivializations <p, we have only to set e, := (h,)-1/2. •
342 VI. Kahler Manifolds

The tube T := {v E F : Xh (v) < I} is a strongly pseudoconvex neighborhood

of the zero section in F. Locally, for T. := Tn rr-l(U.), we have

The boundary of T is given by the equation ww/ (}Z(x) = l.

6.2 Proposition. The line bundle F is Griffiths negative if and only if F
is negative (in the old sense).

PROOF: One direction is trivial. To show that every negative line bundle is
Griffiths negative one needs rotations in the fibers and a smoothing proce-
dure. For details see [Gr62]. _

Definition. The line bundle F on X is called Griffiths positive if the

complex dual F' is negative.

It is clear that F is Griffiths positive if and only it is positive in the old sense.

Special Holomorphic Cross Sections. We assume that F is a

positive line bundle on a compact manifold X. Then F' is negative. We
always denote the coordinate on the fibers of F' by w. Of course, the variable
w depends on the local trivialization.
The strongly pseudoconvex tube around the zero section of F' is locally given
by
T = {(z,w) E U x e : Iwl < e(z)},
where U is a domain in en.From the last chapter we know that dime HI (T, 0)
00. In this situation we prove the following result.

6.3 Proposition. If m is sufficiently large, the following two statements

are valid:
1. If
°
Xl, X2 E X are two different points, then there are holomorphic cross
sections so, SI E reX, Fm) with so(xd i- and SO(X2) i- 0 such that the
quotient Sl / So has different values at these points.
2. If Xo E X is a point, then there are holomorphic cross sections

PROOF: If So, Sl are two sections in a line bundle with so(xo) i- 0, then
there is a holomorphic function f in a neighborhood of Xo such that s 1 (x) =
f(x) . so(x) in that neighborhood. The quotient sI/so is defined to be equal
to f near Xo·
 6. Hodge Manifolds 343

We operate in several steps:

(1) Let Xo E X be a point. We assume that in local coordinates Xo corresponds
to the origin. We call a trivialization of F' around Xo usable if

log(e(z)) = Lgij(Z)ZiZj + IIz113 . a(z)

i,j

in a neighborhood of o. Here a is a smooth function and Ei,j gijZiZj is neg-

ative definite.
The existence of a usable trivialization follows easily. We take an arbitrary
trivialization and write

log(e(z)) = ao + Re(Q(z)) + Lgij(Z)ZiZj + I/z1/3. a(z),

i,j

with ao E IR and a quadratic polynomial Q(z) with Q(z) = O. Then we change

the trivialization by dividing w by exp(ao + Q(z)).
(2) Now let Xl, X2 be two different points in X. We choose usable trivializa-
tions simultaneously over neighborhoods U 1 (xd and U2(X2), and, moreover,
strictly increasing sequences 0 < r1 < r2 < ... < 1 and 0 < Sl < 82 < ... < 1
I I
such that 8d r i =f. 1 and {w = ri} nT, {w = sd n T are analytic sets
in T, contained in F'IU1 (respectively in F'1U2, cf. Figure VI.1). We get

? s F'
7 \

Figure VI.1. Construction of special cross sections

principal parts of meromorphic functions gi = 1/(1 - w/ri) in Tn (F'IU1 )

and g; = 1/(1 - w/ri) in Tn (F'1U2)' and hi = 1/(1- W!Si) in Tn (F'IU2)
as well. Because of the finiteness of cohomology there are complex numbers
ai that are not all zero such that there are meromorphic functions g, h in T
with prillcipal parts Ei ai(gi + g;) and Ei ai(gi + hi). We may assume that
the first nonvanishing coefficient ai is equal to l.
The functions g, hare holomorphic in a neighborhood of the zero section, and
locally they have power series expansions
344 VI. Kahler Manifolds

L L
00 00

g(x,w) = gm(X)W m and h(x,w) = hm(x)wm .

m=O m=O

The coefficients 9m (respectively hm) define global holomorphic sections SO,m

(respectively S1,m) in the tensor bundle Fm.
If over Xl the first nonvanishing part of 9 is aigi = 1/(1 - w/rd, then we
have
g(XI,w) = f
m=O
(r~ +cm)wm, with C
'
m -t 0 for m -t 00.

So for large m we have gm(XI) ~ l/(ri), and also 9m(X2) ~ l/(ri). Analo-
gously, hm(xd ~ l/(ri) and hm (X2) ~ l/{si). We take So = SO,m and
S1 = Sl,m for such a large m. Then the quotient s1I So at Xl has a value that
is nearly 1, and at X2 nearly (rdsi)m. These values are different for large m.
(3) Now we work at some point Xo, and over a neighborhood U of Xo we define
gi = l/{l-w/rd and hi,j(z, w) = 1/[1- (w/ri)· (l+gzj )], for j = 1, ... , n. If
U and g are sufficiently small, the polar sets of gi and hi,j are always analytic
sets in T. As above, we obtain meromorphic functions 9 and hj in T, for
j = 1, ... ,n, which have power series expansions g(x,w) = Lmgm(x)w m
and hj(x, w) = Lm hj,m(x)wm in a neighborhood of the zero section. Again
let gi be the first principal part with nonvanishing coefficient ai. For large m
we have gm(XO) ~ (l/ri)m and hj,m(xo) ~ U/ri)m(l + m . gZj). As above,
gm and hj,m define sections So and Sj in Fm with (Sj/so)(xo) ~ 1 +mgzj. So
the Jacobian of (s1/ So, ... , sn/ so) at Xo is approximately mngn and hence is
not zero for m large enough. _

Projective Embeddings. Let X be an n-dimensional compact complex

manifold and F a positive line bundle on X. Then for large m we have many
global holomorphic cross sections in Fm. If there are sections so, ... , SN that
do not vanish simultaneously, then

cp(x) := (so(x) : ... : SN(X»

defines a holomorphic map cP : X -t ]pn.

6.4 Kodaira's embedding theorem. For sufficiently large m, there are

holomorphic cross sections So, ... , SN in F m such that the map

cP : X -t (so(x) : ... : SN(X»

is an embedding of X in]PN.
PROOF: In the preceding paragraph it was shown that for any point Xo E X
there is a holomorphic cross section So in some tensor power of F that does
not vanish at Xo.
 6. Hodge Manifolds 345

The set Xl = {x EX: SO(x) = O} is a proper analytic set. We decompose it

into (finitely many) irreducible components, and for every irreducible compo-
nent we choose some point in that component and a holomorphic cross section
that does not vanish at this point. In this way we obtain finitely many holo-
morphic cross sections So, ... ,Sl whose joint zero set X 2 has dimension n - 2
at most. We decompose this again into irreducible components and continue
the procedure. After finitely many steps we obtain holomorphic cross sections
So, ... , Sq in some tensor power Fmo that do not vanish simultaneously.
For any two points Xl =1= X2 E X there are two holomorphic cross sections
to,tl such that (to(Xi): tl(Xi» are homogeneous coordinates of two different
points. The set YI = ((x,y) E X x X: (to (x) : tl(X» = (to(y) : tt(y»)} has
lower dimension. We decompose it into irreducible components and proceed
as above. After finitely many steps we get holomorphic cross sections to, . .. ,tp
such that x I-t (to(x) : ... : tp(x» is injective.
For any point Xo E X there are holomorphic cross sections Uo, Ul, .•. , U r such
that uo(xo) =1= 0 and the map x -+ (uo(x) : ... : ur(x» has a nonvanishing
Jacobian at Xo. Altogether we have holomorphic cross sections 10, ... ,IN that
do not vanish simultaneously such that the map ~ : x -+ Uo(x) : ... : IN(X))
is an injective immersion. Since X is compact, ~ is an embedding. _

It follows from Chow's theorem that X has a projective algebraic structure

and a genuine Zariski topology.

Hodge Metrics. We consider an n-dimensional complex Kahler manifold

X. Let H be the Kahler metric on X, and

W=WH = iLgijdziAdzj
i,j

the associated fundamental form. It is a closed real form of type (1,1).

We choose an open covering %' = {Uv : v E N} of X such that all in-
tersections UV1, ... ,Vk are contractible. If "f/ is of the same kind and a refine-
ment of %', and G is an abelian group, then the induced homomorphism
Hi(%" G) -+ Hi("f/, G) is an isomorhism, for i = 0,1,2. We choose %' to be
fine enough so that H has a potential hv in every Uv :

Here the hv are smooth real-valued functions, and the differences hvJl. =
h}" - hv are pluriharmonic in UVJ,£"
If we set {3v := -i oh v , then d{3v = i oahv = wluv • The closed form dhvJl. =
ohv}" + ahv}" (respectively i dChvJl. = ohv}" - ah v }") is real (respectively purely
imaginary). The sum
346 VI. Kahler Manifolds

is a closed holomorphic form of type (1,0). If Xo E Uvp. is a fixed point, then

gvp.(x) := l
Xo
x
i dChvp.

defines a purely imaginary function on UvjJ.' It is determined up to a purely

imaginary constant. Then fvp. := hl/p. + gvp. is a holomorphic function (since
dfl/p. = 28hvp.).
We have d(lp.).. - fv).. + fvp.) = 0, and therefore fp.).. - fv).. + fvp. E iR Then
f = (lij) E C 1 (%,,0) and 8f E Z2(%" ilR). On the other hand,
1 1
f3p. - f3v = -:-8hvp.
I
= -2'I dfvjJ.'
This shows that (1/2 i )8f defines a cohomology class c = c(H, %') E
H2 (%' , 1R) e:! H2 (X, 1R) that corresponds to w under the de Rham isomor-
phism (d. the description at the end of Section 2).
In the following we do not distinguish between H2(%" Z) and its image in
H2(%" 1R). If a cohomology class of H2(%" 1R) lies in H2(%" Z), then we call
such a class an integral class.
Definition. A Kahler metric H is called a Hodge metric if it is integral.
A Hodge manifold is a compact complex manifold that carries a Hodge
metric.

Clearly, every closed complex submanifold of a Hodge manifold is again a

Hodge manifold.
6.5 Theorem. Let X be a compact complex manifold. Every Hodge metric
on X defines via the de Rham isomorphism a cohomology class in H2(X, Z)
that is the Chern class of a negative line bundle F on X.

PROOF: Let w be the fundamental form of the Hodge metric. We take a

finite contractible covering %' = {Uk: k = 1, ... , q} that is so fine that in
every Uk we have a potential hk for w. As above, we construct the holo-
morphic functions fkl with dfkl = 28h kl . Then 8{Jkl} E Z2(%',2iZ), and
rkl := exp(7r1kl) defines a co cycle r = (rkt) E ZI(%"O*) that determines a
holomorphic line bundle F on X. If <Pk : Fluk -t Uk xC are the trivializations,
then

We define

Tk := {(x, w) E Uk xC: Iwl < exp( -7rhk(x»)},

and get IfJk 0 <pi 1 (11 IUkl ) = Tklukl' So we have a tube T around the zero
section in F with Tluk = Tk. Since -log(exp( -7rh k » = 7rh k is strictly pluri-
subharmonic, it follows that F is negative. _
 6. Hodge Manifolds 347

The above procedure can be reversed.

6.6 Theorem. Assume that X is a compact complex manifold, and F a
line bundle on X with a tube T around the zero section such that

Tluk = {(x, w) : Iwl < exp( -rrhk )}

with strictly plurisubharmonic functions h k . Then the hk are local potentials
of a Hodge metric on X.
All this finally implies that
Hodge manifolds are just projective algebraic manifolds.
This theorem was first proved by K. KODAIRA. We shall see in the next
section that it contains a generalization of the classical period relations for
the n-dimensional complex torus. These were originally derived by completely
different methods (see, e.g., [K054]).
Example
On jp'n we have the Fubini metric H and consider the metric (1/(2rr))H with

hk = 1 log
2rr (I Zo
Zk
12 + ... + 1Zn
Zk 12)
as local potentials, for k = 0, 1, ... , n. It follows that

and hi = ~ 10g(Zk/ZI) is a holomorphic function with Re(fkl) = h kl . There-

fore, the transition functions of the associated line bundle F are the functions
'Ykl = exp(rrfkl) = Zk/ZZ, and F is the (negative) tautological bundle.

Exercises
1. Assume that X is a compact Riemann surface and that Xo E X is a
point. Show that the bundle [-xol (associated to the divisor (-1)· xo) is
negative.
2. Compute the embedding of jp'n into jp'N defined by a basis of the vector
space r(jp'n, 0(2)) and determine the image.
3. Let X be a compact complex manifold and L -+ X a negative line bundle.
Prove that r(X, L) = {o}.
4. Let X be a compact complex manifold and 7f : E -+ X a holomorphic
vector bundle. Assume that there are trivializations <{). : Elu, -+ x cq u.
with transition functions gUt.
(a) A fiber metric on E is given by a system of smooth functions h. :
U. -+ Mn,n(C) such that h.(x} is a positive definite Hermitian matrix
 348 VI. Kahler Manifolds

for every x E U.) and hI< = g/I< . h • . 9.1< on U.I<' Prove that there is a
smooth nonnegative function Xh : E -+ lR with

Xh 0 cp;:-l(X, v) = v· h.(x) . v t .

Prove that if Xo E U.) then CPt can be replaced by a trivialization 'IjJ

such that h",(xo) = Eq (= identity matrix) and (8h",)xo = O. Such a
trivialization is called normal at Xo.
(b) The bundle E is called Griffiths negative if h can be chosen in such a
way that Xh is strictly plurisubharmonic on E-ZE (where ZE is the
zero section in E). As in the case of line bundles, E is called Griffiths
positive if the complex dual E' is Griffiths negative.
If cp : Elu -+ U x e q is a trivialization and cp(e) = (x, v), then the
tangential map Cp* : Te(E) -+ Tx(X) EBe q is an isomorphism. Denote
the space (cp*)-I(Tx(X)) by He(CP) (space of "horizontal" tangent
vectors at e with respect to cp). Prove that E is Griffiths positive
if and only if there is a fiber metric h on E such that for every
e E E - Z E with 11"( e) = x there is a normal trivialization cp at x such
that
Lev(Xh)(e,~) < 0 for ~ E He(CP), ~ =1= O.

(c) Prove that T(lfDn) is Griffiths positive with the Fubini metric.
(d) Let E be a Griffiths positive vector bundle and FeE a subbundle.
Show that ElF is Griffiths positive. What does this mean for the
normal bundle of a submanifold of pn?
5. Let X be a compact complex manifold and L -+ X a positive line bundle.
Show that there is an open neighborhood U = U (ZE) c c E such that
E - U is strongly pseudoconvex.

7. Applications
Period Relations. Recall the definition of the n-dimensional torus. We
take 2n vectors WI, ... ) W2nE en that are linearly independent over lR. The
additive subgroup

acts freely on en by translations. The quotient space Tn = en If is an n-

dimensional compact complex manifold and is called a torus. We say that
W = (wi, ... , win) E M n ,2n(C) is the period matrix of Tn.
For z E en we denote its equivalence class in Tn by [z]. By the quotient map
en -+ Tn) the affine space en becomes the universal covering space of Tn.
11" :
Since 11" is locally biholomorphic, we can use the affine coordinates Zl, ... , Zn
of en as local coordinates on Tn.
 7. Applications 349

Every torus has the structure of an abelian complex Lie group;

[v) + [W) ;= [v + W).

It is clear that this definition does not depend on the representatives of the
equivalence classes.
If wEen is a fixed vector, the translation Tw : en ~ en is defined by
Tw(Z) := Z + w. It induces a biholomorphic map Tn ~ Tn by [zll-+ [z + wl,
which we also denote by Tw. It depends only on the equivalence class of w.
From topology it is known that H 2 (rn, Z) is a free group spanned by the
e;) cycles
Cij(S,t):= SWi +tWj, 0:::; s:::; 1,0:::; t:::; 1,

with i < j. Since Tn is compact, every de Rham cohomology class [w) E

H2 (Tn) is uniquely determined by the periods

Vij ;= 1 CiJ
w, i < j.

On en we have the Euclidean metric ds 2 = dz 1 dz 1 + ... + dzndzn , which

is invariant under translations. It induces a Kahler metric Ho on Tn with
fundamental form
n
Wo = i E dZ II 1\ az ll •

v=1
It follows that every torus rn is a Kahler manifold. Using the Euclidean
volume element, for every differentiable function f on Tn we define the mean
value M[J) by

M[J)(z) ;= -II f f 0 Tw(Z) dV with 1:= f dV.

JWETn JTn

It is clear that M[Jl is invariant under arbitrary translations and therefore a

constant function.
7.1 Lemma. If w is the fundamental form of some Kiihler metric Ii on
Tn, then there is another Kahler metric H on Tn such that the associated
fundamental form W = WH has constant coefficients and the same periods as
w. If Ii is a Hodge metric, then H is also a Hodge metric.
PROOF: If w= i Lv,1' gvl' dzv 1\ azl" then we define gvl' := M[gvl') and

W := i E gvl' dzv 1\ azw

v,1'
The form W has constant coefficients. Since the gill' are periodic (i.e., invariant
under translations by elements of r), we have
350 VI. Kahler Manifolds

Iii Cii,,1J. 0 Tw) dz" 1\ cLzlJ. = Iii 9,,1J. dz" 1\ cLzlJ., for every wEen.

Using Fubini we compute

=
1
I
r-
iTn Vij dV =
-
Vij'

Now H is a Hodge metric if and only if W H defines an integral class. Since

a cohomology class in H2(Tn) is uniquely determined by the periods, the
second assertion also follows. •

W 1'th W.. -- (WI(i) , •.. (i))

, Wn lor z• -- 1, ... , 2n, we h ave
C

11 11 (W~i) ds + w;:) dt) 1\ (w~) ds + w~) dt)

W(i)W(i) - w(j)w(i)
v IJ. v IJ.'

and therefore

Vij i L gvlJ. (w~i)w~) - w;:)w~))

",IJ.
-2Im(Lg"ILw~i)w~») = -2Im(wi t .G'Wj),
V,IL
for G := (g"IJ.)'
7.2 Proposition. The torus Tn zs a Hodge manifold if and only if all
periods Vij are integral.

PROOF: Since every element of Tn has a unique representative

we can use u = (u 1, ... , U2n) as real coordinates on Tn.

We have a special contractible covering all = {Uk} of Tn . For k = (k I , ... , k2n)
with k i E {O, I} we define the open set

Uk = { u E Tn : ~. ki ~ Ui ~ ~ . (ki + I)} (see Figure VI.2).

 7. Applications 351

The boundaries of the Uk consist of parts of the "walls" 8 j and 8;, which
are defined by

8; = { u E Tn : Uj = ~} and 8j = {u E Tn : Uj = I}.
Since %' is contractible, the singular cohomology group H2(Tn) is isomorphic

Si

Figure VI.2. A special covering for the torus

to the cohomology group H2(%" Z). In particular, H2(o//, Z) = Z for the

torus Tl, which has real dimension 2. The class [dUl 1\ dU2] is integral, and
every integral class is an integral multiple of [dUll\du2]. If we denote by o//(n)
the covering of Tn, then an easy induction argument shows that a cohomology
class in H 2 (%,(n), Z) is integral (respectively real) if and only if it has the
form LV<1l a VIl duv 1\ dUll' with avlJ. in Z (respectively JR) :
For any v < 11 the class [du v 1\ dulJ.] is integral. Assume that the case n - 1
has already been proved. Then Tn = Tl X Tn-I, and after subtracting an
integral sum of [dUi 1\ dUj] with i < j < n the cohomology class is given by a
covering %,(1) X Tn-1 and is therefore a lifting of some integral multiple of
dU2n-l 1\ dU2n. So

[w] =L
i0
Vij[dui 1\ dUj] with Vij = 1
C0
wE JR,

and [w] is integral if and only if all periods Vij are integral. •
We call Tn an abelian variety if Tn is projective algebraic.
7.3 Proposition. A torus Tn is an abelian variety if and only if all periods
are integral (with respect to some Kahler form w).
More precisely, we have the following result on period relations.
7.4 Theorem. Let Tn be given by the period matrix W = (wf, .. . , win).
Then Tn is an abelian variety if and only if there is a positive definite Her-
mitian matrix G such that Im(W t • G . W) is integral.
352 VI. Kiihler Manifolds

The Siegel Upper Halfplane

Definition. The (generalized) Siegel upper halfplane (of degree n) is
the n( n + 1) /2-dimensional domain

1ln := {Z = X + iY E Mn(1C) : Z t = Z, and Y positive definite}.

7.5 Theorem. If Z E 1ln, then W = (En, Z) E M n ,2n(1C) is the period

matrix of an abelian variety Tn.

PROOF: We take for G the positive definite symmetric real matrix y-l.
Then

wt·G·W = ( En
X t + iy t
) . y-l . (En' X - iY)

( En
X t + iyt
) . (y-l, y-l. X - iEn)

(
y-l
x t . y-l + iEn
y-l .X- iEn
Xt.y-l·X+y ).
It follows that
Im(W t . G . -W) = (0En
is integral, and the period relations are satisfied. •
A converse of the theorem is also true, but we do not give the proof here.

Semipositive Line Bundles. We consider an n-dimensional compact

complex manifold X with a holomorphic line bundle F on X.
Definition. The bundle F is called seminegative if there is a tube
T around the zero cross section such that T = {(z,w) : Iwl < e(z)} in
local coordinates and -In(e) is everywhere plurisubharmonic and strictly
plurisubharmonic at at least one point of X. The bundle F is called
semipositive if the dual bundle F' is seminegative.

We give an example of a seminegative bundle F with dim HI (T, 0) = 00.

The boundary function - In(e) can be chosen to be strictly plurisubharmonic
outside a nowhere dense analytic set in X.

Here is the construction. We take a compact Riemann surface R of genus

g ;::: 1 and a negative line bundle L on R with H1(R, L') =1= O. The complex
dual L' is positive, and the cross sections in L -m, m > 0, are holomorphic
 7. Applications 353

functions on L that are homogeneous polynomials of order m on the fibers

on L.
We add a point at infinity to each fiber of L and obtain a compact algebraic
surface X. We denote by D the divisor of all points at infinity, and by F( m)
the line bundle over X, associated to the divisor mD. Then all cross sections
in L- m are sections in F(m) over X that vanish on the zero section Z =
Z L C LeX of order m. These sections are holomorphic functions on the
dual bundle F( -m) := F(m)' over X (associated to the divisor -mD) that
are linear on the fibers of F( -m), but vanish over Z. Let $ be the set of
all these functions. The function f(x) == 1 on X is also a cross section in
F(m) that vanishes on D to mth order, but it does not vanish on Z. It gives
a holomorphic function on F( -m) that is linear on the fibers, vanishes over
D, but does not vanish over Z. We denote it by h and add it to the system
§. Now, $ is a set of holomorphic functions on F( -m) that are linear on
the fibers and whose joint zero set is the zero cross section of F( -m).
Forming the sum of several products f J, f E $, we construct a plurisubhar-
monic function 't/J on F( -m) with the following properties:
1. The zero set of't/J is the zero cross section in F( -m).
2. 't/J is quadratic on the fibers.
3. 't/J(z) tends to 00 as z approaches of.
4. 't/J is strictly plurisubharmonic outside of F( -m)lz, but plurisubharmonic
everywhere.
So our tube T = ((w,x) : 't/J(w,x) < I} is pseudoconvex, and strongly pseu-
doconvex outside of F( -m) Iz.
The restriction F(-m)lz is trivial. Therefore, the holomorphic functions on
(F( -m)lz) n T consist of certain convergent power series Ei ai(x)w i , where
w is a coordinate on the fibers. All powers i appear. The first cohomology of
Z with coefficients in these power series is infinite-dimensional, since it con-
tains the direct sum of the cohomology with coefficients in those holomorphic
functions that are polynomials on the fibers with fixed degree i. We can ex-
tend these polynomials to F( -m) by replacing w by h. If we restrict to these
polynomials, we get an injection of the cohomology of Z with coefficients in
the polynomials on the fibers into H 1 (T, 0). So the cohomology Hl(T, 0) is
likewise infinite-dimensional.

Moishezon Manifolds. We previously mentioned Moishezon manifolds

(see Section VI.3). An n-dimensional compact complex manifold is called a
Moishezon manifold if it has n analytically independent meromorphic func-
tions. Without proof we state here a result that was originally called the
Grauert-Riemenschneider conjecture (cf. [GrRie70]).
7.6 Theorem. If a compact complex manifold X has a semipositive line
bundle, then it is a Moishezon manifold.
354 VI. Kahler Manifolds

The conjecture was proved by Y.T. Siu (see [Siu84] and the summary in
[GrPeRe94], VII.6).

To obtain a result in the other direction we have to blow up along a d-

codimensional submanifold Y C X with d ~ 2 (cf. Section IV.7, "Monoidal
Transformations"). This leads to a proper modification X of X, where Y is
replaced by a 1-codimensional submanifold Y and the map Y -+ Y is a fiber
bundle with typical fiber JP'd-l.
7.7 Theorem. If X is a Moishezon manifold, there is a finite sequence of
monoidal transformations with d-codimensional centers with various d that
lead to a proper modification X of X such that X is a projective algebraic
variety.
This statement is proved in [Moi67]. We know that X carries a negative line
bundle. By blowing down to X we get a "meromorphic" negative line bundle
on X. Now over certain points of X there is a family of lines instead of one
line.
In the case that X is a surface every Moishezon manifold is projective alge-
braic (see, e.g., [ChoK052]). But already in dimension three there are Moishe-
zon manifolds that are not projective algebraic. The first example was given
in Russia (the "Russian counterexample" or "the complex Sputnik," as it was
called at Princeton in 1958), but see also [Nag58] or better just the classical
example of Hironaka ([Hi60]), or [Pe93].
For a proof of the following important result see [Moi67] or [Pe86].
7.S Theorem. Assume that X is a Moishezon manifold which has a Kahler
metric. Then X is a projective algebraic manifold.

Exercises
1. Prove that the torus T2 given by the period matrix

w= (1 0 A
01 A
A)
R
has only constant meromorphic functions.
2. Prove that the Siegel upper halfplane is contractible.
3. For W E M n ,2n(C) let W := (::). Recall the results of Exercise 5.3 in
chapter IV and prove that for every n-dimensional torus T there is a
matrix Z E Mn(C) with detIm(Z) > 0 and T ~ T(E .. , Z).
4. Find a compact complex surface X with a seminegative line bundle F
such that F is not holomorphically convex.
Chapter VII

Boundary Behavior

1. Strongly Pseudoconvex Manifolds

The Hilbert Space. Assume that X is an n-dimensional complex man-
ifold that carries a (real analytic) Kahler metric ds 2 = Lgijdzidzj , where
the gij are real analytic functions of the local coordinates Zl, ... , Zn. We con-
sider a domain n cc X whose boundary an is 'jfoo-smooth and strictly Levi
convex. In this chapter such a domain is called strongly pseudoconvex.
If Xo E X is a point, we can introduce local coordinates in a neighborhood
of Xo such that G = (gij (xo)) is the identity matrix. We call them Euclidean
coordinates in Xo. They are determined up to holomorphic coordinate trans-
formations that at Xo are given to first order by a unitary matrix. We have
a Euclidean volume element dV in Xo, and also the complex star operator "*
given by
cp 1\ "* 'IjJ = 2p +q . (L
I,J
a lJ . blJ ) dV,

for forms cp = L alJ dZI 1\ azJ and'IjJ = L blJ dZI 1\ azJ of type (p, q). Using
the Euclidean coordinates we also have a scalar product at Xo that is given
by

This is invariant by unitary transformations and therefore independent of the

choice of the Euclidean coordinates.
Assume now that cp and 'IjJ are continuous on Ii". Then we have the inner
product
(cp, 'IjJ) = In I
(cp(x) 'IjJ(x)) dV = In cp 1\ "*'IjJ.

This is a pre-Hilbert space that completes to the Hilbert space L~,q = L~,q(n)
of (p, q)-forms with square integrable coefficients. It contains the space
dP,q(Ii") of '(foo_(p, q)-forms over Ii" as a dense subspace, and also the space
dcf,q(n) of such forms with compact support in n. Every element of L~,q is
a square integrable current of degree p + q (or bidegree (p, q)).

Operators. First, we shall extend the operator 8: dcf,q(n) ~ .ldcf,q+1(n)

to a bigger domain dom(8) c L;,q. For this we use the fact that there is a
uniquely determined adjoint operator 1J: dcf,q+I(n) ~ dcf,q(n) such that
356 VII. Boundary Behavior

The proof for this statement is the same as in the case of compact Kahler
manifolds.
Now the existence of the extension follows easily. If t.p E L~,q is given, we
define the current 8t.p by (8t.p,'lj;) = (t.p,fJ'lj;) for all 'lj; E dJ',q+l(n). We say
that t.p is an element of dom(8) if the current 8t.p is contained in L~,q+l' The
(unbounded) operator 8 is now a map from dom(8) to dom(8), with 808 = 0
(since fJ 0 fJ = 0).

Next we shall extend fJ to the conjugate operator 8* on a dense subset

dom(8) C L~,q+l' Assume that 'lj; E L~,q+l' If t.p E dJ',q(n), we define
(t.p, 8*'lj;) = (8t.p, 'lj;) and get a current 8*'lj; of bidegree (p, q) over n. For 'lj; an
element of dom(8) we require that the current 8*'lj; be contained in L~.q.
But this is not yet enough. We wish to have (t.p, 8*'lj;) = (8<p, 'lj;) for every
<p E dom(8): At the moment we have this only for <p E dJ',q(n). But a gen-
eral <p E dom(8) can be approximated by a sequence t.pi E dJ',q(n). Assume
now the validity of

(*) 1(8t.p,'lj;)1:::; c·II<pII, for all <p E dom(8),

where 1It.p1l = V(<p,t.p) and c > 0 is a constant that depends only on 'lj;. Then
we have also that (8<pi,'lj;) converges to (8t.p,'lj;), and (<pi,8*'l/J) to (t.p,8*'lj;).
This gives the required statement.
If, on the other hand, (8<p,'lj;) = (<p,8*'lj;) is always valid, then (*) follows. So
we define

Next we consider the operator 0 = 8*8 + 88* that maps a form in L~,q onto
a current of bidegree (p, q) over n. We define

dom(D):= {t.p E L;,q: <p E dom(8)ndom(8), 8<p E dom(8) , a7p E dom(8)}.

Then 0 maps domeD) to L~,q. For <p, 'lj; E domeD) we have the equality

(O<p, 'lj;) = (8t.p,8'lj;) + (a*<p,8*'lj;) = (t.p,O'lj;).

The space of harmonic forms is defined to be the space

Yt'p,q := {<p E domeD) : O(<p) = o}.

We get very simply that Yt'p,q is closed in dom(O) and that D(dom(D» is
orthogonal to Yt'p,q.
 2. Subelliptic Estimates 357

Boundary Conditions. We now consider forms 'P E dp,q(n). In this

case all derivatives 8'P, 8* 'P and also those of higher order are square inte-
grable. So domeD) ndp,q(n) consists of those forms 'P for which the equation
(*) is valid for 'P and 8'P. We assume that in a neigborhood U(on) there is
a smooth real function r such that unO. = {x E U : rex) < O} and that
dr(x) i:- 0 for x E U. Then by a computation (partial integration) one has
the following:
*
A form 1j; belongs to dom(8) if and only if or 1\ 1j; =' 0 on on. This
holds for arbitrary 1j;, not only for elements in the inverse image of the
space of smooth differential forms.
The following is then immediate.
1.1 Proposition. A form 'P belongs to domeD) if and only if the following
two boundary conditions are satisfied on on:
1.orl\*'P='O.
2. or 1\ *8'P =' O.

These conditions are called the 8-Neumann boundary conditions. For liter-
ature see [DL81] and [FoK072]. The main result of the next section is the
following:
Assume that q > O. Under the hypotheses given above, the harmonic
forms of type (p, q) (i. e., elements of .J't'p,q) represent the elements of the
cohomology group Hp,q (0.).
The first proof of this theorem was given in [Ko63].

Exercises
1. Give an explicit description of the operator 8* .
2. Prove that the boundary conditions for the operator 8 and C(j'<XJ functions
are always satisfied.
3. Prove that the boundary conditions are never satisfied if q = n and the
form is not identically zero on on.
4. Build a theory of harmonic forms for a "cohomology with compact sup-
port" HJ (0.,0). What are the boundary conditions?

2. Subelliptic Estimates
Sobolev Spaces. We look for an operator
N : dp,q(n) -+ dp,q(n) n domeD)
such that DoN = id. Finding such an operator is the so-called Neumann
problem. To solve this problem we need Fourier transforms and the so-called
subelliptic estimates.
358 VII. Boundary Behavior

We begin by introducing the notion of a Sobolev space. We start with the

vector space Y of rapidly decreasing '/tfoo-functions I : ]RN -+ C with

IxQ D{3 I(x) I -+ 0 for IIxll -+ 00 and for all a, {3 E NN.

Here

We have the Fourier transfomation F : Y -+ Y given by

F[J](x) = lex) := (21T~N/2 J I(y)e-i(x,y) dy.

For every s E ]R we define the norm II .. .lIs by

The Sobolev space of index s is defined as the completion Hs(]RN) of Y with

respect to this norm.

Now we work in the space ]RN+1 with coordinates (x,r) = (XI, ... ,xN,r),
and consider the half-space ]R~+1 = {(x, r) : r ~ O} and 'itfoo functions
I(x, r) with compact support in ]R~+1. Then for fixed r these functions are
in Y. We take the Fourier transforms with respect to x, for fixed r, and
define

This norm is called the tangential Sobolev norm.

Now let us return to our strongly pseudo convex domain n cc X. We put

N = 2n - 1, and take any Xo E an and a neighborhood U(xo) c X with
local coordinates

Zl = Xl + iX2, ... ,Zn-l = X2n-3 + iX2n-2, Zn = X2n-1 + ir.

Here r is the boundary function for an as in the preceding section. In the
following we will work only with respect to these local coordinates. If I E
do°,o(Unfi), then I is '/tfoo-smooth and has compact support in ]R~+1. Hence,
we can define the norm 11If11~. For <p E .0"cf,q(unfi) we put 1II<p~I~ = LJ H"I~,
where I runs through the coefficients of <po
Definition. The Neumann problem for (p, q)-forms is called subelliptic
on n if there is an e > 0 such that for every Xo E an there is a coordinate
neighborhood U(xo) and a constant c> 0 such that
 2. Subelliptic Estimates 359

for all <p E .!dri,q (u n 0) n dom(a*).

Here Q( <p, 1j;) = ('P, 1j;) + (a'P, a1j;) + (a"<p, a*1j;) is the s<rcalled energy form
(or the Dirichlet inner product).

2.1 Theorem. If f2 is strongly pseudoconvex, the Neumann problem is

subelliptic for all (p, q)-forms with q > O.
The proof of this theorem is rather difficult, of course. The s<rcalled Kohn-
Morrey inequality is involved. See [DL81]' for instance.

The Neumann Operator. There is an old theory on elliptic bound-

ary problems, beginning with Dirichlet's problem (see the old papers of L.
Bers and L. Nirenberg, for instance). Here the Garding inequality plays an
important role. If it holds for a problem, existence and regularity of solutions
of the problem can be proved. In our terminology it can be expressed in the
form

If c < 1, then the inequality for sub ellipticity is much weaker than the
Garding inequality. But nevertheless, in the papers of Kohn and Nirenberg
it was proved that there are results on the smoothness of the solution of the
Neumann problem that are similar to those in the old theory and that follow
from subellipticity alone. We state a first important result (see [DL81]):
2.2 Proposition. Assume that n is strongly pseudoconvex and that the
Neumann problem is subeUiptic for (p, q)-forms on f2. Then there is a unique
operator N : L;,q -+ domeD) with the following properties:
1. N(,Yep,q) = o.
2. DoN = id.
3. N (L;,q) is orthogonal to ,Yep,q.

This operator N is called the Neumann operator. Under the above conditions
the following theorem is true:
2.3 Theorem.
1. The harmonic space,Yep,q is finite-dimensional. It is contained in .!dp,q (n),
and for <p E ,Yep,q we have or 1\ "* <p == 0 on on.
2. The Neumann operator N is compact with respect to the L~,q -norm.
3. We have .!dp,q(n) = .!/ep,q EEl DN(.!dp,q(n)).
4· If'P E .!dp,q(n), a<p = 0, then <p is orthogonal to a"aN(.!dp,q(n)).
360 VII. Boundary Behavior

For a proof see [KoNi65] and [FoKo72]. Both papers are difficult to read.
In the case X = en there is a simpler proof. A domain nee en with smooth
boundary is called regular if there exists a finite number of subsets Si Can,
i = 0, ... , N, such that
1. 0 = SN C SN-1 C ... CSl C So = an.
2. If Z E Si, but z f/. Si+1, then there are a neighborhood U(z) and a
differentiable submanifold M C Un an with holomorphic dimension
equal to zero such that Si n U eM.
If n is pseudoconvex and regular, then a certain compactness estimate holds
for the Neumann problem for (p, q)-forms with q ~ 1 (cf. [Ca84]). Still using a
result of Kohn/Nirenberg (see [KoNi65]), from the compactness estimate the
existence and the compactness of the Neumann operator N follow. See also
[BoStr99] for further discussion. If n is an arbitrary pseudoconvex domain,
the Neumann operator exists as a continuous linear map L~,q -+ dom(O).
This is proved in [Ca83].
A strongly pseudo convex domain n c en is always regular, and therefore the
above theorem on the Neumann operator holds in this case. But in en the
harmonic space .)It'p,q is always equal to zero.

Real-Analytic Boundaries. We assume again that X is a complex

manifold, but we consider only a domain n with real-analytically smooth
boundary. Here we only assume that n is pseudoconvex. Then the following
holds:
2.4 Proposition. The Neumann problem is subelliptic on n for (p, q)-forms
with q ~ 1 if and only if all germs of local complex-analytic sets A c an have
dimension less than q.
Moreover, we have the following result:
2.5 Proposition. Assume that n cc X = en. Then there is no germ of
a local complex-analytic set A c ac of positive dimension.
For a proof see [DiFo78] and [Ko79].

Examples. We will construct two 2-dimensional examples nee X, where

n contains just one compact irreducible curve Y. From the existence of Y
it will follow that the cohomology group H 1 (n,0) is different from zero.
The cohomology classes are given by unique harmonic forms that satisfy the
boundary conditions. We can study these as if they were functions, which may
be useful for our investigations. Unfortunately, the forms in .)It'D,l are not in
general d-closed. So there is no connection between .)It'0,1 and the cohomology
with coefficients in e as in the case of compact Kahler manifolds. We will see
this more precisely in the examples.
 2. Subelliptic Estimates 361

Example 1
We take a compact Riemann surface Y of genus 9 > 1 and a line bundle
X on Y with Chern class -1. This line bundle is negative. There is a real-
analytic metric on the fibers of X such that 0 = {x EX: IIxll < I}
is a strongly pseudo convex relatively compact subdomain of X with real-
analytically smooth boundary. We get a uniquely determined Kahler metric
on X by using the metric of constant curvature on Y and the metric on the
fibers and requiring that the surfaces {x EX: IIxli < r} be orthogonal to
the fibers.
Over Y we have the vector bundle Oi of homogeneous polynomials of degree
i on the fibers of X. If i is sufficiently large, then Hl(y, Oi) vanishes. So the
finite-dimensional cohomology groap Hl(O, 0) is generated by finitely many
of these groups Hl(y, Oi). The cohomology classes are given by differential
forms r.p E Jd'O,l(Y,Oi) = r(Y,!\O,1(y) ® Oi). They all may be viewed as
differential forms on 0 whose coefficients are holomorphic polynomials of
degree i on the fibers. They vanish on Y to order i. We denote the set of these
forms by l2l?,1. Examples for forms of this type are given by the derivatives
8/, where / is a differentiable function in 0 whose restriction to the fibers is
a holomorphic polynomial of degree i. We take the orthogonal complement of
these forms and get £0,1 = ffii .Yt';,0,1, where .Yt';,0,1 C l2l?,1. This follows by a
direct computation. Thus we obtain a basis of finite length, and the restriction
of these forms to 80 satisfies the boundary conditions of harmonic forms.
Since 9 :::: 2, the space of harmonic forms .Jf{0,1 has positive dimension. We
can obtain a basis by explicit computation. The elements are harmonic forms
in 0 vanishing on YeO. If there were an isomorphism onto the cohomology
with coefficients in C (as in the compact case), they would have to be zero.

Example 2
We will construct another 2-dimensional example of a completely different
type. For this we use compact complex-analytic curves that are no longer reg-
ular. They are irreducible reduced I-dimensional complex spaces (Le., topo-
logical Hausdorff spaces that locally look like analytic sets in some complex
n-space). Here we go a bit beyond the scope of this book, but we hope that
the reader can nevertheless get an idea of the method.
We take the projective line ]p1 with the points 0 and 00. We identify these
two points, obtaining a compact irreducible I-dimensional complex curve that
has a "normal crossing" of I-dimensional domains in a neighborhood of the
gluing point O. This point 0 is the only singular point.
The projective line ]p1 is the so-called normalization of Y: There is a sur-
jective holomorphic mapping 7r : ]p1 -+ Y that maps 0 and 00 to 0 and
362 VII. Boundary Behavior

is biholomorphic elsewhere. The first cohomology group HI (Y, 0) of Y has

dimension 1 (as can be computed).
We build a 2-dimensional complex manifold X around Y. There is a neigh-
borhood U = U(O) in Y and a holomorphic embedding

j :U '-t X I = {( w, z) E C 2 : Iw I < I}

with j(U) = YI := {(w, z) E Xl : w 2 - Z2 = a}. We choose an c with

0< c < 1 and put P:= {(w,z) E Xl : Iwl:::; c} and Y{:= YI-P. We define

Y2 := Y - rl(YI n P) and Y;:= rl(yl ) n Y2 = rl(YI - P).

The manifold Y; consists of two disk shells

Yo rI({(w,z) E C 2 : c < Iwl < 1 and z = w}),
Yoo rl({(w,z) E C2 : c < Iwl < 1 and z = -w}).

On YI we have the holomorphic function WI := prIlv) , which can serve as

local (and even global) coordinate. The set Y 2 is the complement of the union
of two neighborhoods of 0 and of 00 in pI and therefore isomorphic to a disk
shell in Co = pI - {O,oo}. We have a global coordinate W2 on Y2, and a
holomorphic function WI = !(W2) on Y~ with

Now we put X 2 = Y2 X C and use in X 2 the product coordinates (W2' Z2).

We choose suitable small open neighborhoods WI = WI (Yd C Xl and W 2 =
W2(Y2) C X 2, and define the gluing map F : (Xl - P) n WI -+ (X2Iv;) n W 2
by
( ) F( ) {(J-l(w),Z - w) near z = W,
W2, Z2 = W, Z = (J-l(w), z + w) near z = -w.

So we obtain by this gluing procedure a complex manifold X around Y. The

singular curve Y is a hypersurface in X.

x
The covariant normal bundle N° = N (Y) is given by the adjunction formula

Nx(Y) = [-Yllv,

where [- Y] is the line bundle associated to the divisor - Y in X. The normal

bundle N = Nx(Y) is trivial over Y2. Therefore, the holomorphic function Z2
can be regarded as a holomorphic function on N that is linear on the fibers.
This defines a holomorphic cross section in N° over Y2 that is given by
the holomorphic function S2(W2) == 1 with respect to the local trivialization
over Y 2 . It extends to a global meromorphic section s over Y. In the local
trivialization over Y I , s is given by the meromorphic function
 2. Subelliptic Estimates 363

1
Sl(W,Z) = 2z =
{ll/(z
j (Z+W)
- w)
on z = w,
on z = -w,

since the transition function gi2 for N* is given by

* Z2 Z =f W
9 12 -- Z2 _ w2 = z2 - w2 .

Now we lift everything to pl. The lifted section s of 1r 0 (N°) has poles of first
order in 0 and 00, and no other poles and no zeros. So 1r°(N°) has Chern
class -2.
To construct a bundle with positive Chern class, we have to alter X 2 • For this
we take a point p E Y - Y 1 and two small disks D' c c D around p such that
- - I - I
D is still in Y - Y 1 C Y2 • Let S be the disk shell D - D . We glue (Y - D ) x C
with D x Cover S by means of the gluing map H : S x C -4 S X C with

We replace our old X 2 by a new X2 and get a new manifold X. Now the
lifting of the covariant normal bundle NJc (Y) to pI has Chern class 3. It
follows by Riemann-Roch that dimcHO(p1,1r*(NJc(Y») = 3 + 1 = 4. By
the projection pI --+ Y the fibers over 0 and 00 are identified. This gives an
additional condition for the existence of global holomorphic cross sections in
the covariant normal bundle of Y eX. So we have 3 linearly independent
holomorphic cross sections Sl, S2, S3 in N*.
One can see by computation that sl1Y1 may be assumed to be given by
(z - w)(z + w). The function IZ2 - w 212 is an extension of S181 to Xl' If we
add the function I(z - w)2(z + w)312, we obtain an extension that is strictly
plurisubharmonic outside of Y1 • We can also extend the other silYl to Xl
and the silY2 to X 2, form the SiBi, sum up, and glue together after a small
change in a neighborhood U(Y) cc X to obtain an extension h of SlB1 +
S282 + 8383 in U. The extension vanishes on Y and is positive outside. It is
strictly plurisubharmonic in a neighborhood U ' of Y. Then for very small
E > 0, the tube n = {h < E} C C U ' will have a smooth boundary. It is
strongly pseudoconvex.
We have dim HI (Y, 0) = 1. The cohomology class can be given by a cocy-
de with coefficients in C. It therefore extends to n. So there is a nonzero
harmonic form cp E £0,1 (n). It satisfies the boundary condition. So in this
example the cohomology of the curve Y, which has singularities, is given by
a smooth harmonic form in a tube around the curve. This shows that the
theory of harmonic forms on manifolds with boundary may be very useful.
364 VII. Boundary Behavior

Exercises
1. Show that the Sobolev space H- s may be identified with the dual of
H 8 • Show that the Sobolev norm II .. .lIs comes from a scalar product
(... , .. ')8' and prove the following generalized Schwarz inequality: If
f, 9 E Ho and f E Hs for some s > 0, then (f, g)8 :::; IIfll8 . IIglls.
2. It is a consequence of the famous "Rellich lemma" that the following
holds: If s > t, then for any c > 0 there is a neighborhood V(O) such
that IIflit :::; cllflls for all f with compact support in V. Use this to prove
the same estimate for the tangential Sobolev norm.
3. Assume that the Neumann problem is solvable for (p, q)-forms. Let a
form <p E L~,q with 8<p = 0 be given that is orthogonal to £P,q. Prove
that 1/J := {j* N<p is the only solution of the equation 81/J = <p that is
orthogonal to every 8-closed form.
4. Let n cc en be a strongly pseudo convex domain. Prove that for every
positive number C there is a smooth plurisubharmonic function e on n
with 0 :::; e:::; 1 such that Lev(e)(z, t) ~ Clltll 2 for every z E an.
5. Take Example 2 and prove that that the derivative d( <ply) is different
from 0 and that this form is not harmonic.
6. Take Example 1 and prove that £1,0 has infinite dimension.
7. The following problem requires some knowledge about sheaves and com-
plex spaces. Take integers r < s such that r, s are free of common divisors.
Consider at 0 E e the holomorphic functions that are given by conver-
gent power series in w r and w S • Then replace the ring of holomorphic
functions in 0 E PI by these holomorphic functions and repeat the pro-
cedure of Example 2. Since now the "structure sheaf' 0 of pI is smaller,
a new curve Y is obtained. Construct an X around Y such that there is
a 2-dimensional strongly pseudoconvex tube n around Y. Compute the
first cohomology of Y.

3. N ebenhiillen
General Domains. Assume that n c en is a general domain. We take
the open kernel of the intersection of all domains of holomorphy fi ::J nand
its connected component N(n) that contains n. (More precisely, we work in
the category of unbranched domains over en as in the case of the construction
of the hull of holomorphy).
Definition. The domain N(n) is called the Nebenhulle of n.

We can see very easily:

The Nebenhulle is a domain of holomorphy.
The NebenhUlle contains the envelope of holomorphy of n.
 3. Nebenhiillen 365

If 0 cc en
is a domain of holomorphy with (smooth) real-analytic boundary,
then 00 does not contain any germ of an analytic set of positive dimension.
Moreover, one can show that N{O) = O. That implies that one can approxi-
n
mate by Stein neighborhoods. See [DiFo77] for details.
The Hartogs triangle 0 := {(w, z) E e 2 : Iwl < Izi < I} is a domain of
holomorphy, but N{O) is the unit polydisk in e 2. In this case the envelope
of holomorphy is 0 and the Nebenhiille is much bigger. We will see that this
can happen even when 0 is a bounded domain with smooth boundary. The
following example was constructed in [DiFo77].

A Domain with Nontrivial Nebenhiille. We define a real-analytic

function (J : C* x e --t IR by
(J{w, z) = Iz + exp(i In{ww»1 2 - 1

and put
0:= {(w, z) E C· x e : (J(w, z) < o}.
A computation of the derivatives of (J gives
0(J
i z exp(i In(lwI2» _ ~ exp{ -i
ow = w
In(lwI2)),
w
0(J
OZ
= Z + exp{ -i In(lwI 2»,
02(J
owow = -1~12 exp{i In(lwI2)) - 1;1 2 exp{ -i In{lwI2)),
02(J I ~(J
OWOZ = ~ exp(i In{lwI2», and ozoz = 1.

3.1 Proposition. an is smooth.

PROOF: If there were a point Cwo, zo) E 00 where d(J = 0, then

0(J
oz (wo, zo) = O.

This gives Zo = - exp( i In{lwoI2» and hence (J(w!), Zo) = -1. But such a
point is not on the boundary. •

3.2 Proposition. The bmtndary of 0. is pseudoconvex. It is not strongly

pseudoconvex exactly along M = {(w, z) : z = O}.

PROOF: The complex dimension of the complex tangent space at every

point of 00 is 1. A vector ~ = aolow + boloz + aolow + bojoz at a point
366 VII. Boundary Behavior

of 8n lies in the complex tangent space if and only if (a8/8w+b8/8z)[u] = o.

So the vector space of complex tangents is spanned by the vector

~= _ 8u ~ + 8e ~.
8z 8w 8w 8z
The function e is the potential of a Hermitian form H. The boundary of n
is pseudo convex (strongly pseudoconvex) if and only if H ~ 0 (respectively
H > 0) on this vector. By a calculation we get

It vanishes only for z = o.

n
We take any neighborhood U of that is a domain of holomorphy. Then U
contains C· x {O}. If Wo E C., Zo E C and 0 < Izo + exp(i In(lwoI2))1 < 1,
then n n (e x {zo}) is the disjoint union of infinitely many disk shells in
C·. When we move Zo on a line to 0 the shells may become larger. So by the
theorem of continuity it follows that every holomorphic function in U can
be extended holomorphically to the point (wo,zo). So U and the Nebenhiille
N(n) contain the domain C· x {z : Izl < 2}, and consequently the Nebenhiille
is much bigger than n. •

Bounded Domains. The domain n of the preceding paragraph lies over

C· and is not bounded. But one can obtain a pseudoconvex bounded domain
by making n smaller. One takes an r > 371", leaves n unchanged over the
line segment [1, r], and deforms the circles n
n (C x {w}) slowly to 0 as
Iwl goes to 0 or to 00. The new domain is bounded and has a much bigger
Nebenhiillej the boundary is 'i!foo-smooth, but not real-analytic. This is the
"worm domain" constructed explicitly in [DiF077].

Domains in (:2. The domain n from above (with real-analytic boundary)

lies in C· x C. But we also can find such a domain in C 2. We have the universal
covering '(:2 -t C· X C by the holomorphic mapping (w, z) = F(w, Z) :=
n
(exp( w), Z). Then we lift n to C 2 and obtain a domain that has a smooth
real-analytic boundary and is not bounded.
The new domain n
is a "semitube": The translations in the imaginary w-
direction act on it. Hence, the Nebenhiille N(n) is also a semitube. So it can
be written as the set S(M) = {(w, Z) : (Re(w), Z) E M}, where M is the
smallest open set in {(u, Z) : u E JR., Z E C} with the following properties:
1. S(M) ::) n.
2. S (M) is pseudo convex.
3. JR. x {O} C M.
 4. Boundary Behavior of Biholomorphic Maps 367

We have F(S(M)) = {(w, z) : (Iwl, z) EM}, where M is the image of M

under the mapping (u, z) t--+ (exp( u), z). The domain S (M) is pseudoconvex
if and only if F (S (M)) is pseudoconvex, since F is locally biholomorphic
with respect to iii. So the Nebenhiille N(fi) is the set {iU E C : IZI < 2}, just
as N(n) is {w E C* : Izl < 2}.

Exercises
1. Let nee C 2 be a domain of holomorphy. If there is an analytic plane
E such that nnE has isolated boundary points, then !1 has a nontrivial
Nebenhiille.
2. Consider domains !1 c C. When is N(!1) = H(!1) (i.e., the envelope of
holomorphy)?
3. A family n(t) of domains in cn is called continuous at to if for every
compact set K C n(t o) and every domain n:) n
there is an € > 0 such
that K C n(t) c n
for It - tol < c.
(a) Construct a continuous family of domains !1(t) such that the family
of the envelopes of holomorphy H(!1(t» is not continuous.
(b) Define "upper semicontinuity" for families of domains. Find condi-
tions such that the upper semicontinuity of H(!1(t» follows from the
continuity of !1(t).

4. Boundary Behavior of
Biholomorphic Maps
The One-Dimensional Case. Assume that !1,!1' c e
are bounded
domains with 'ifoo-smooth boundary and that f: !1 -+ !1' is a biholomorphic
mapping. The following theorem is well known:
4.1 Theorem. The map f can be uniquely extended to an invertible 'if00
n
map j: -+ !1'.
In particular, if a!1, an' are real-analytic, then the extension is likewise real-
analytic.

The Theory of Henkin and Vormoor. Assume now that !1 is a

domain in cn. A holomorhic map F : n -+ eN is called Holder continuous
of index c > 0 if for all points w, z E !1 the estimate

IIF(w) - F(z)1I ::; M· IIw - zllc

holds with a constant M > O.

Here we state a special case of a theorem that was independently proved by
Henkin and by Vormoor (see [He73) and [Vo73]).
368 VII. Boundary Behavior

4.2 Theorem. Assume that 0 1 , O2 E c n are strongly pseudoconvex do-

mains with «/00 -smooth boundary and that F : 0 1 -+ O2 is a biholomorphic
map. Then F has an extension to a Holder continuous map F : 0 1 -+ O2
with index ~.
The PROOF of this theorem is very technical. The main ideas can be found
in the book of DiederichjLieb ([DL81]). An important tool is the notion of a
pseudodifferential metric. Since these metrics are interesting themselves, we
give the definition for general complex manifolds:
Definition. Assume that X is an n-dimensional complex manifold.
A pseudodifferential metric on X is an upper semicontinuous map Fx :
T(X) -+ lRt with the property

for all vectors ~ E T(X) and all a E C.

We do not require that Fx(~) be positive for ~ =I- 0, and we do not require
the triangle inequality.
As an example we consider the Kobayashi metric. It is of great importance,
not only within the scope of this section. A complex manifold is called hy-
perbolic if it has a complete Kobayashi metric. The word "complete" here
means that all metric balls are contained relatively compactly in X. It is
known that the set of holomorphic mappings into a hyperbolic manifold is
a normal family in the sense of Montel. A Riemann surface is hyperbolic in
this sense if and only if it is of hyperbolic type (in the classical sense). On
any other Riemann surface the Kobayashi metric degenerates to O.
But first let us give the definition of the Kobayashi metric. As usual, we
denote the unit disk by 0 = {z : Izl < I} C C. If f : 0 -+ X is a holomorphic
map and f(O) = xo, then we denote by 1'(0) the tangent vector 1'(0) := f.(e),
where f. : To(D) -+ Txo(X) is the induced mapping and e the canonical basis
of To(D) ~ C.
Definition. The Kobayashi metric is the pseudodifferential metric d x :
T(X) -+ lRt
defined by
dx(~) := inf{lal- 1 : :3 f : 0 -+ X holomorphic with f(O) = Xo, /,(0) = a~}.
The following statement is a very simple consequence of the definition.
4.3 Proposition. Assume that F : X -+ Y is a holomorphic map of com-
plex manifolds. Then dy(F*(~)) ::; dx(~).
So the Kobayashi metric is invariant under biholomorphic mappings. In cer-
tain cases it is completely degenerate. For example,
the Kobayashi metric on C n is identically O.
 4. Boundary Behavior of Biholomorphic Maps 369

For the Kobayashi metric in the unit disk we obtain

For a bounded domain nee en one shows that

lie II
daCe) ~ a· dist(z, on) ,

where ~ E Tz(n) ~ en, zEn, and a E e is a constant. Moreover, if n has a

'6'oo-smooth strongly pseudoconvex boundary, we even get

> II~II
daCe) - a· dist(z, on)1/2'

Real-Analytic Boundaries. We consider again the continuation of

biholomorphic maps F : n 1 -t n 2 to the boundary. But we assume now that
the boundaries are real-analytically smooth. In this case we can drop the
assumption of strong pseudoconvexity.
4.4 Theorem. IfF: n1 -t n2 is a biholomorphic map of pseudoconvex do-
mains with smooth real-analytic boundaries, then F extends to a (uniformly)
Holder continuous map F : 0 1 -t O2 .
This was proved by Diederich and Fornress (see [DiFo78]).

Fefferman's Result. In 1974 Charles Fefferman proved the following

theorem (see [Fe74]).
4.5 Theorem. Assume that n1. n 2 cc en are strongly pseudoconvex do-
mains with '6'00 -smooth boundaries and that F : n 1 -t n 2 is a biholomorphic
mapping. Then F extends to a '6'00 diffeomorphism F : 0 1 ~ n 2 .
For his proof Fefferman used the Bergman metric, which is invariant under
biholomorphic mappings. His proof was very difficult. There were many sim-
plifying methods written by authors like S.M. Webster, N. Kerzman, S.R.
Bell, E. Ligo<;ka, L. Nirenberg, P. Yang. Later, an interesting proof that uses
the Bergman projection was discovered by Catlin (see [Ca84]).
We wish to indicate some ideas concerning the Bergman kernel function. We
will define it in the general case of a strongly pseudoconvex domain n cc X
with smooth boundary, where X is an n-dimensional complex manifold. If F
is a complex analytic line bundle on X, one can prove as in Chapter V that
the cohomology HI (n, F) is finite. Then, following the ideas of Chapter V,
we can construct many holomorphic cross sections in F over n. We take for
F the line bundle of holomorphic n-forms X = h(ZI,"" zn) dZ l A .. , A dzn .
The norm II X /I is defined by
370 VII. Boundary Behavior

IlxII2:= Inx/\"*x.
Let Yt' be the space of holomorphic n-forms X on 0 that are square inte-
grable (i.e., with Ilxll < 00). The forms X E Yt' with Ilxll :::; 1 are uniformly
bounded (with respect to their coefficients in local coordinate systems) on
every compact set M c O. If A is a compact analytic subset of 0 of positive
dimension, all forms X may vanish there. This will lead to difficulties in defin-
ing the Bergman metric. The space Yt' is a Hilbert space and a subspace of
the Hilbert space £2(0) of arbitrary square integrable complex n-forms (with
the scalar product (<p, 'IjJ) = In
<p /\"* 'IjJ).

Next we define the kernel form. We take an orthonormal basis {Xi : i EN}
of Yt'. Then
L Xi(Z)Xi(W) = K(z, w) dZ l /\ ... /\ dZn /\ d'iih /\ ... /\ awn
is called the kernel form on 0, and the coefficient K is called the Bergman
kernel. The theory of K is well known for domains in en. We extend it
here to complex manifolds, but we give only an overview and avoid going
into detail. It is possible to prove that the kernel form is independent of
the choice of the orthonormal basis. The convergence of the defining series
is locally uniform, also for all derivatives. The kernel K is real-analytic on
0, holomorphic in z, and antiholomorphic in w. As (z, w) approaches 80,
K(z, w) tends to 00 to order n + 1 (i.e., its coefficients with respect to local
coordinates). When we pass over to the conjugate complex form, interchange
w with z, and multiply by (_1)n, we get the kernel form back. Moreover, the
kernel is invariant under biholomorphic mappings of 0 onto itself. In local
coordinates, but independent of these coordinates, we have
K(z, z) = sup If(zW.
x=/dz 1 1\"'l\dzn ,lIxll=1

We have the orthogonal projection P : L2(0) -+ Yt' with

= jn( _1)n(n-l)/2dz l /\ ... /\ dZ n . i a(w) /\ K(z, w) awl /\ ... /\ awn·

It is called the Bergman projection.
n
We denote by don,o (0) the set of '6'00 forms X of type (n, 0) in for which all
derivatives of any order of the coefficient of X vanish on 80, and by Aoo(O)
the set of '6'00 forms in n that are holomorphic in O.
4.6 Theorem. In the case X = en, for every regular pseudoconvex (and
therefore for every strongly pseudoconvex) domain 0 c en
the Bergman pro-
jection P maps don,o(fi) into Aoo(O).
 4. Boundary Behavior of Biholomorphic Maps 371

This was proved by Catlin in [Ca84]. The result is known as "0 satisfies
condition (R)." The worm domain is an example where condition (R) does
not hold.

Mappings. Now we consider another strongly pseudoconvex domain 0' cc

en with <t'oo smooth boundary, and a biholomorphic mapping F : 0 ~ 0'.
We can transform any form a of type (n, 0) on 0' to a form a 0 F of the same
type on O. We state without complete proof a lemma that is trivial in the
case of forms with compact support:
4.7 Lemma. For a E don,oCn') the transform a 0 F is an element of
don,o(O).
We give a brief idea of the proof. We represent al\a as a multiple of the kernel
form with a factor c(x) ~ O. The function c vanishes on 00' to arbitrarily
high order. We just have to prove that co F also has this property in O.
To do this we take in 0 and in 0' the metric <s(x, y) = suPjlf(x) - f(y)l.
Here f runs through the holomorphic functions in 0 (respectively 0') with
If(x)1 ::; 1 (this is the so-called Caratheodory distance). If Xo is a fixed point
of 0 (respectively 0'), then the balls around Xo approach 00 (respectively
00') to first order for r --+ OOj Le., the Euclidean distance of the ball to the
boundary of the domain behaves like exp( -r). If Xo E 0 and Yo = F(xo) EO',
then any ball of radius r around Xo is mapped onto a ball of the same radius
around Yo. This implies the vanishing of co F with all derivatives on 00.
We need the following <t'oo version of the Cauchy-Kovalevski theorem:
4.8 Lemma. Assume that hE AOO(O'). Then there is a form 9 E dn,O(O')
such that 8g = 0 on 00' and h - 6.g E don ,0 (n').

We use the lemma to complete the proof. It follows by partial integration that
6.g is orthogonal to all elements from AOO(O'). We can define an orthonormal
basis of £(0') consisting of such functions. So 6.g is orthogonal to £(0').
It follows that

h0 F = (PB(h - 6.g» 0 F = Pc«h - 6.g) 0 F) E AOO(O).

When we take for h the functions zlln" ... , Zn In" we get the differentiability
of F on n.The proof of Fefferman's theorem is completed.

The Bergman Metric. The Bergman metric is interesting in its own

right. It is a Kahler metric and depends only on the complex structure of the
manifold under consideration. Using its Riemannian curvature one can prove
interesting theorems that give a deeper insight into the complex structure.
The notion is well known for domains in en. Here we generalize it to complex
manifolds.
372 VII. Boundary Behavior

We will consider the situation where X is an n-dimensional complex manifold

and n cc X a subdomain with strongly pseudo convex boundary. Then n is
holomorphically convex. We can construct forms X belonging to.Ye. For that
we use the boundary of n. We consider the analytic subset E c n given by

E {x En: all X vanish at x or there is a tangent vector ~ at x

such that ~(b) = 0 for the coefficient b of every X}.

It can be proved that E is compact and has positive dimension everywhere.

So in the classical case of a domain in en
it is empty. This is no longer true in
complex manifolds. Therefore, we make the following additional assumption:
(*) The set E is empty.
Under this assumption we have the Bergman metric in n. In every local
coordinate system we just define the coefficients

.. ( )._ a2 log(K(z,
g.) Z.- a!l=
z»
ZiUZj
and the Bergman metric

ds 2 := Lgij(Z) dZ i dZj .
i,j
The definition is independent of local coordinates, and we obtain a Kahler
metric in n that is invariant by biholomorphic transformations of n.
If there are no compact analytic subsets of positive dimension in n, then n
is a Stein manifold. In general, n may contain compact analytic subsets of
positive dimension. For example, we may take for X a complex analytic line
bundle on ]P'I with Chern class c ::; -3. Then a strongly pseudoconvex tube
n around the zero section exists, and there are enough forms X on n. So in
this case as well we have the Bergman metric.
Near to the boundary the following estimates hold.
4.9 Proposition. If ~ is a tangent vector in Tz(n), zEn, II~II is the
Euclidean norm, and dist(z, an) is the Euclidean distance (everything with
respect to local coordinates), then there are positive constants a, b such that

a· dist(z,"'"
an)1/2 -< II~II -<.
b II~II
dist(z, an) .

The distance (with respect to the Bergman metric) between a fixed point Zo E
n and a point z (near to the boundary) is approximately

vn + 1 . (-In(dist(z, an»).
 4. Boundary Behavior of Biholomorphic Maps 373

Exercises
1. Prove the properties of the Bergman kernel. Compute the Bergman kernel
for the unit ball.
2. Assume that n c en is a bounded domain. Then n carries the Caratheo-
dory metric. This is a positive pseudodifferential metric on n and is
defined in the following way: If ~ is a vector at a point Zo E n, we take
all holomorphic functions f on n with If(z)1 < 1 in n and measure the
Euclidean length of the image f.(~). Then we take the supremum over
all these f. Prove that the Kobayashi length is at least as big as the
CaratModory length.
3. Consider the bicylinder pn(o,2) = {(w,z) : Iwi < 1, Izl < I} C e 2 and
a sequence of complex numbers Wi with IWil < 1 that converges to 1 and
take two different complex numbers Zl, Z2 in the open unit disk. Does
the Kobayashi distance between the points (Wi, Zl), (Wi, Z2) tend to oo?
4. Compute the Bergman metric and the Kobayashi metric for

n = {z E e : 0 < Izl < I}.

Are these metrics complete?
References

[BPV84] W. Barth, C. Peters, A. Van de Ven: Compact Complex Surfaces.

Erg. Math. 3.4, Springer, Berlin-Heidelberg (1984).
[BeSt48] H. Behnke, K. Stein: Entwicklungen analytischer Funktionen auf
Riemannschen Fliichen. Math. Ann. 120,430--461 (1948).
[BeTh70] H. Behnke, P. Thullen: Theorie der Funktionen mehrerer kom-
plexer Veriinderlichen, 2nd ed., Erg. Math. 51, Springer, Berlin-
Heidelberg-New York (1970).
[Bi61] E. Bishop: Mappings of partially analytic spaces. Am. J. Math.
83, 209-242 (1961).
[Bie33] L. Bieberbach: Beispiel zweier ganzer Funktionen zweier kom-
plexer Variablen, welche eine schlichte volumentreue Abbildung
des R4 auf einen Teil seiner selbst vermitteln. Sitzungsb. der
preuB. Akad. d. Wissensch. (1933).
[BoStr99] H.P. Boas, E.J. Straube: Global Regularity of the a-Neumann
Problem: A Survey of the L 2 -Sobolev Theory. Several Complex
Variables, MSRI Publications 31, 79-111 (1999).
[Bou66] N. Bourbaki: Elements of Mathematics, General Topology, part 1.
Hermann, Paris (1966).
[Br54] H. Bremermann: Uber die Aquivalenz der pseudokonvexen Ge-
biete und der Holomorphiegebiete im Raume von n komplexen
Veriinderlichen. Math. Ann. 128, 63-91 (1954).
[BrMo78] L. Brenton, J. Morrow: Compactijications of en. Trans. Am.
Math. Soc. 246, 139-153 (1978).
[CS53] H. Cartan: Seminaire E.N.S. 1953/54, exposes XVI and XVII.
[Ca83] D. Catlin: Necessary conditions for subellipticity of the a-
Neumann problem. Ann. Math. 111, 147-171 (1983).
[Ca84] D. Catlin: Global regularity of the a-Neumann problem. Symp.
Pure App!. Math. 41, 46-49 (1984).
[ChoKo52] W.L. Chow, K. Kodaira: On analytic surfaces with two indepen-
dent meromorphic functions. Proc. Nat. Acad. Sci. USA 38,319-
325 (1952).
[DL81] K. Diederich, I. Lieb: Konvexitiit in der komplexen Analysis.
Birkhauser Basel 1981, 1-150 (DMV Seminar 2).
[DiFo77] K. Diederich, J.E. Fornress: Pseudoconvex Domains: An Example
with Nontrivial Nebenhulle. Math. Ann. 225, 275-292 (1977).
[DiFo77b] K. Diederich, J.E. Fornress: Pseudoconvex domains. Existence of
Stein neighborhoods. Duke Math. J. 44, 641-662 (1977).
[DiFo77c] K. Diederich, J.E. Fornress: Pseudoconvex domains. Bounded
strictly plurisubharmonic exhaustion functions. Invent. Math. 39,
129-141 (1977).
376 References

[DiFo78] K. Diederich, J .E. Fornress: Pseudoconvex domains with real-

analytic boundary. Ann. Math. 107,371-384 {1978}.
[DiFo79] K. Diederich, J .E. Fornress: Proper holomorphic maps onto pseu-
doconvex domains with real-analytic boundary. Ann. Math. 110,
575-592 {1979}.
[DoGr60] F. Docquier, H. Grauert: Levisches Problem und Rungescher Satz
JUr Teilgebiete Steinscher Mannigfaltigkeiten. Math. Ann. 140,
94-123 {1960}.
[Ew96j G. Ewald: Combinatorial Convexity and Algebraic Geometry.
Springer, New York (1996).
[Fe74] C. Fefferman: The Bergman kernel and biholomorphic mappings
of pseudoconvex domains. Invent. Math. 26, 1-65 (1974).
[FoKo72] G.B. Folland, J.J. Kohn: The Neumann Problem for the Cauchy-
Riemann complex. Annals of Mathematics Studies 75, Princeton,
New Jersey (1972).
[Fo70] O. Forster: Plongements des varieUs de Stein. Comm. Math. Helv.
45, 17(}-184 (1970). .
[Fo81] O. Forster: Lectures on Riemann surfaces. Graduate texts in
Math., Springer, New York {1981}.
[FoSte87] J .E. Fornress, B. Stensr;mes: Lectures on counterexamples in sev-
eral complex variables. Math. Notes 33, Princeton university press
(1987).
[Fu93] W. Fulton: Introduction to Toric Varieties. Annals of Math. Stud-
ies 131, Princeton (1993).
[GISte95] J. Globevnik, B. Stens0nes: Holomorphic embeddings of planar
domains into C 2 • Math. Ann. 303, 579-597 (1995).
[Gr55] H. Grauert: Charakterisierung der holomorph vollstiindigen kom-
plexen Riiume. Math. Ann. 129, 233-259 {1955}.
[Gr62] H. Grauert: Uber Modifikationen und exzeptionelle analytische
Mengen. Math. Ann. 146, 331-368 (1962).
[GrLi70] H. Grauert, I. Lieb: Das Ramirezsche Integral und die Losung der
Gleichung '8 f = a im Bereich der beschriinkten Formen. Complex
Analysis, Rice university studies 56, 29-50 (1970).
[GrPeRe94} H. Grauert, Th. Peternell, R. Remmert: Encyclopaedia of Math-
ematical Sciences 74, Several Complex Variables VII. Springer,
Heidelberg {1994}.
[GrRe55} H. Grauert, R. Remmert: Zur Theorie der Modifikationen I.
Stetige und eigentliche Modifikationen komplexer Riiume. Math.
Ann. 129, 274-296 {1955}.
[GrRe56] H. Grauert, R. Remmert: Konvexitiit in der komplexen Analy-
sis (Nicht-holomorph-konvexe Holomorphiegebiete und Anwen-
dungen auf die Abbildungstheorie). Comm. Math. Helv. 31, 152-
183 (1956).
 References 377

[GrRe57] H. Grauert, R. Remmert: Singularitiiten komplexer Mannig-

faltigkeiten und Riemannsche Gebiete. Math. Zeitschr. 67, 103-
128 (1957).
[GrRe58] H. Grauert, R. Remmert: Komplexe Riiume. Math. Ann. 136,
245-318 (1958).
[GrRe79] H. Grauert, R. Remmert: Theory of Stein Spaces. Springer, Hei-
delberg (1979).
[GrRe84] H. Grauert, R. Remmert: Coherent Analytic Sheaves. Springer,
Heidelberg (1984).
[GrRie70] H. Grauert, O. Riemenschneider: Verschwindungssiitze fUr an-
alytische Kohomologiegruppen auf komplexen Riiumen. Invent.
Math. 11, 263-292 (1970).
[Gre67] M.J. Greenberg: Lectures on Algebraic Topology. Benjamin, New
York (1967).
[Gri69] Ph.A. Griffiths: Hermitian differential geometry, Chern classes,
and positive vector bundles. Global Analysis, papers in honor of K.
Kodaira, ed. by D.C. Spencer and S. Iyanaga, Princeton university
press (1969).
[GriHa78] Ph.A. Griffiths, J. Harris: Principles of Algebraic Geometry. John
Wiley & Sons, New York (1978).
[Gu90] R.C. Gunning: Introduction to Holomorphic Functions of Several
Variables I-III Wadsworth &Brooks/Cole, Belmont (Cal) (1990).
[GuRo65] R.C. Gunning and H. Rossi: Analytic Functions of Several Com-
plex Variables. Prentice-Hall (1965).
[Hei56] E. Heinz: Ein elementarer Beweis des Satzes von RadO-Behnke-
Stein-Cartan iiber analytische Funktionen. Math. Ann. 131, 258-
259 (1956).
[He73] G.M. Henkin: An analytic polyhedron is not holomorphically
equivalent to a strictly pseudoconvex domain. Dokl. Akad. Nauk
SSSR 210, 1026-1029 (1973), and Soviet. Math. Dokl. 14, 858-
862 (1973).
[He69] G.M. Henkin: Integral representations of functions holomorphic
in strictly pseudoconvex domains and some applications. Math.
USSR Sb. 7, 597-616 (1969).
[Hi60] H. Hironaka: On the theory of birational blowing-up. Thesis, Har-
vard (1960), unpublished.
[Hi62] H. Hironaka: An example of a non-Kiihlerian complex-analytic de-
formation of Kiihlerian complex structures. Ann. Math. 75, 190-
208 (1962).
[Ho55] H. Hopf: Schlichte Abbildungen und lokale Modifikationen 4-
dimensionaler komplexer Mannigfaltigkeiten. Comm. Math. Helv.
29, 132-155 (1955).
[Hoe66] L. H6rmander: An introduction to complex analysis in several
variables. van Nostrand, New York (1966).
378 References

[KaKa83] L. Kaup, B. Kaup: Holomorphic FUnctions of Several Variables.

Walter de Gruyter, Berlin (1983).
[Ko54] K. Kodaira: On Kiihler varieties of restricted type. Ann. Math.
60, No.1, 28-48 (1954).
[Ko63] J.J. Kohn: Harmonic integrals on strongly pseudoconvex mani-
folds. 1. Ann. Math. 78, 112-148 (1963) and II. Ann. Math. 79,
450-472 (1964).
[Ko79] J.J. Kohn: Subellipticity of the a-Neumann problem on pseudo-
convex domains: sufficient conditions. Acta math. 142, 79-122
(1979).
[KoNi65] J.J. Kohn, L. Nirenberg: Non-coercive boundary value problems.
Comm. Pure Appl. Math. 18, 443-492 (1965).
[Ku60] N. Kuhlmann: Projektive Modifikationen komplexer Riiume.
Math. Ann. 139, 217-238 (1960).
[La60] H. Langmaack: Zur Konstruktion von Holomorphiehiillen un-
verzweigter Gebiete iiber dem en. Schriftenreihe d. Math. Inst.
Munster 17, 1-88 (1960).
[Lau71] H.B. Laufer: Normal two-dimensional singularities. Annals of
Math. Studies 71, Princeton (1971).
[Li70] 1. Lieb: Die Cauchy-Riemannschen DGLn auf streng pseudokon-
vexen Gebieten (beschriinkte Losungen). Math. Ann. 190, 6-44
(1970).
[Moi67] B.G. Moishezon: On n-dimensional compact varieties with n al-
gebraically independent merom orphic functions. AMS Transl., II.
Ser., 63, 51-177 (1967).
[MoKo71] J. Morrow, K. Kodaira: Complex Manifolds. Holt, Rinehart and
Winston, New York (1971).
[Nag58] M. Nagata: Existence theorems for non-projective complete alge-
braic varieties. Illinois Journal of Mathematics 2, 490-498 (1958).
[Nar60] R. Narasimhan: Imbedding of holomorphically complete complex
spaces. Am. J. Math. 82, 917-934 (1960).
[Nar92] R. Narasimhan: Compact Riemann surfaces. Birkhauser, Boston
(1992).
[No54] F. Norguet: Sur les domaines d'holomorphie des fonctions uni-
formes de plusieurs variables complexes. Bull. Soc. France 82,
137-159 (1954).
[Oda88] T. Oda: Convex Bodies and Algebraic Geometry - An Introduction
to the Theory of Toric Varieties. Springer Heidelberg 1988.
[Oka53] K. Oka: Sur les fonctions analytiques de plusieurs variables, IX
Domaines finis sans point critique interieur. Jap. J. Math. 23,
97-155 (1953).
[Pe86] Th. Peternell: Algebraicity criteria for compact complex mani-
folds. Math. Annalen 275, 653-672 (1986).
 References 379

[Pe93] Th. Peternell: Moishezon manifolds and rigidity theorems.

Bayreuther Mathematische Schriften 54, 1-108 (1993).
[PS88] Th. Peternell, M. Schneider: Compactifications of e 3 - l. Math.
Ann. 280, 129-146 (1988).
[Ra86) R.M. Range: Holomorphic Functions and Integral Representations
in Several Complex Variables. Springer, Heidelberg (1986).
[RaSi] R.M. Range, Y.T. Siu: Uniform estimates for the a-equation on
domains with piecewise smooth strictly pseudoconvex boundaries.
Math. Ann. 206, 325-354 (1973).
[Ri68] R. Richberg: Stetige streng pseudokonvexe Funktionen. Math.
Ann. 175, 257-286 (1968).
[Re56] R. Remmert: Sur les espaces analytiques holomorphiquement con-
vexes. C. R. Acad. Sci. Paris 243, 118-121 (1956).
[Re57] R. Remmert: Holomorphe und meromorphe Abbildungen kom-
plexer Raume. Math. Ann. 133, 328-370 (1957).
[ReSt53] R. Remmert, K. Stein: Uber die wesentlichen Singularitaten ana-
lytischer Mengen. Math. Ann. 126, 263-306 (1953).
[dRh84] G. de Rham: Differentiable Manifolds. Grundlehren 266,
Springer, Berlin-Heidelberg-New York (1984).
[Ru74] W. Rudin: Functional Analysis. Tata McGraw-Hill (1974).
[Schue92] J. Schurmann: Einbettungen Steinscher Raume in affine Raume
minimaler Dimension. Preprint, Munster (1992).
[Schue97] J. Schiirmann: Embeddings of Stein spaces into affine spaces of
minimal dimension. Math. Ann. 307, 381-399 (1997).
[Schw50] L. Schwartz: Theorie des distributions I, II. Hermann, Paris
(1950/51).
[Se53] J.-P. Serre: Quelques problemes globaux relatifs aux varieUs de
Stein. Colloque sur les fonctions analytiques de plusieurs variables
complexes tenu a Bruxelles, Masson, Paris (1953).
[Siu84] Y.T. Siu: A vanishing theorem for semipositive line bundles over
non-Kahler manifolds. J. Diff. Geom. 19,431-452 (1984).
[Sko77] H. Skoda: Fibres holomorphes d base et d fibre de Stein. Invent.
Math. 43, 97-107 (1977).
[St72] J.-L. Stehle: Plongements du disque dans e 2 . Seminaire P. Lelong
1970/71. Springer Lecture Notes in Math. 275, 119-130 (1972).
[Vo73] N. Vormoor: Topologische Fortsetzung biholomorpher: Funktionen
auf dem Rande bei beschrankten streng-pseudokonvexen Gebieten
im en mit Coo-Rand. Math. Ann. 204, 239-261 (1973).
[vdW66] B.L. van der Waerden: Algebra l. Springer, Heidelberg (1966).
[War71] F.W. Warner: Foundations of differentiable manifolds and Lie
groups. Scott, Foresman and Co, Glenview, Illinois (1971).
[Wh36] H. Whitney: Differentiable manifolds. Ann. Math. 37, 64~80
(1936).
Index of Notation
natural numbers, integers, etc.
= Nu {O}
={xE.R:x>O}
= C - {O}

imaginary unit A
V* real dual space of a real (or complex) vector space
v' complex dual space of a complex vector space
F(V) C-valued real linear forms on a complex vector space

Mp,q(k) k-valued matrices with p rows and q columns

Mn(k) k-valued square matrices of order n
GLn(C) , general linear group
AB (or A· B) product of matrices

z, zt vector z = (ZI' ... ,zn) and transposed vector

(z IW)m standard Euclidean scalar product in am
Ilzll, dist(z, w) Euclidean norm and distance
(zlw) standard Hermitian scalar product
Izl sup-norm (or maximum-norm)

OrCa) open disk (in C) around a with radius r

o unit disk 0 1 (0)
Br(z) open ball with radius r around z
pn(z, r) polydisk in en around z with polyradius r
P~(z) poiydisk with polyradius r = (r, ... , r)
pn unit polydisk Pf{O)
r: en -+ 1" natural projection onto absolute value space
Pz polydisk pn{o, r(z»
Tn(z,r) distinguished boundary (torus)
Tz distinguished boundary Tn{o, r(z»

UccV U lies relatively compact in V

Cj(z) Cauchy integral of f : T -+ C at z E pn

'Vf(z) holomorphic gradient (/%1 (z), .. . ,1%" (z»
'Vf(z) antiholomorphic gradient (hi (z), ... , h" (z»
'Vxf(z) = (f"'1 (z), ... , I"'n (z»
'Vyf(z) = (fYl (z), ... , IYn (z»
382 Index of Notation

D V fez) higher partial derivative

Df(z) total real derivative of f at z
(&f)z(w) = fZl (Z)W1 + ... + fZn (z)w n
(8f)z(w) = h, (Z)W1 + ... + hn (z)w n
(df)z = (&f)z + (8f)z
Jr(z) complex (or holomorphic) Jacobian matrix
JR,r(z) real Jacobian matrix
g(G) space of smooth functions on G
O(G) space of holomorphic functions on G

t=(z,w) tangent vector at z with direction w

Tz tangent space at z
Q(O) tangent vector (0:(0),0:'(0»

N(It,.··, fq) common zero set of It, ... , fq

A (or Reg(A» set of regular points of A
Sing(A) set of singular points of A
rkz(It,···, fq) rank of the Jacobian JUl> ... ,Jq)

(pn,H) Euclidean Hartogs figure

(P,H) general Hartogs figure
bSt boundary of an analytic disk St = CPt (D)

connected component of M containing z

boundary distance (for a domain G)

Lev(f)(z, w) Levi form Lv,Jl. fzvz" (z)WVwJl.

Hz (&G) complex (or holomorphic) tangent space of &G
Hess(f)(z, w) Hessian Lv,Jl. fxvx" (z)WVwJl.
~

KG holomorphically convex hull of K in G

91 -< 92 the Riemann domain 91 is contained in 92

H§(9), H(9) $-hull of 9 and envelope of holomorphy
G set of accessible boundary points of G

ring of formal power series

= {J E c[zB : IIflit < oo}
ring of convergent power series
m maximal ideal in H n
set of nonzero element, set of units in an ideal I
set of monic polynomials with coefficients in I
germ of the holomorphic function f at z
 Index of Notation 383

w(u, z) pseudopolynomial
D(w) algebraic derivative of w
6""D", discriminant and discriminant set of w

dimz(A), dim(A) local and global dimension of an analytic set

Riemann sphere C U {oo}

tangent space of a manifold

tangential map

XxzY fiber product of manifolds

Kx, reX) canonical bundle and tangent bundle on X

r(u, V) = O(U, V) space of holomorphic sections in a bundle V

C(U,V) space of smooth sections in V

V$W, V®W Whitney sum and tensor product of bundles

V', pk complex dual bundle of V, tensor power of P
j*V, VjW pullback and quotient bundle
Nx(Y) normal bundle of Y in X
Ox,O:X trivial bundles X x C and X x C·

Ci(OU, V) Cech co chains with values in V

Zi(OU, V) Cech cocycles
Bi(OU, V) Cech coboundaries
Hi(OU, V) Cech cohomology group with values in V
Hi(X, V) cohomology group with values in V
Hq(X), Hq(X) singular homology and cohomology of X

Ox ring of germs of holomorphic functions

P(m) polar set of a meromorphic function
.A(X) set of meromorphic functions on X
div(m) divisor of a meromorphic function
[Z] line bundle associated to the divisor Z
Pic(X), 9J(X) Picard group and set of all divisors

rn = cnjr n-dimensional complex torus

pn complex projective space
St(k, n), Gk,n Stiefel manifold and Grassmannian
pI : Gk,n -+ pN Plucker embedding
0(1),0(-1) hyperplane bundle and tautological bundle

Osgood space pI x ... X pI (m times)

384 Index of Notation

e(h) = eeL) Chern class of a cocycle h (of a line bundle L)

I\r F(X) bundle of r-dimensional differential forms

n~ bundle of holomorphic p-forms on X
p;r(u) r-forms
p;p,q(U) forms of type (p, q) on U
1{)1\1/J wedge product

dcp, 01{), 81{) total, holomorphic and antiholomorphic differential

d C = i (8 - 0) (such that ddc = 2i08 )

Chj(z) = (1/{27ri)) fa f«()/{( - z) d( 1\ d(

Hp,q(X) Dolbeault group of type (p, q) on X
Hr(x) rth de Rham group of X

WH fundamental form of the Hermitian metric H

PU(n + 1) group of projective unitary transformations

dV volume element
*1{) Hodge star operator (p;r{x) --+ p;2n-r(x))
(I{), 1/J) inner product f x I{) 1\ "* 'I/J, with "* 1/J = *'I/J

current associated to a form 1/J

current associated to a submanifold M
space of currents of degree 2n - r

adjoint operators of d, 0, and 8

= (1/2)1{) 1\ w, with Kahler form W
adjoint operator of L

~ real Laplacian (eM + ~d)

o complex Laplacian {o{} + {}o)
.Jf'r{x) space of harmonic r-forms on X
.Jf'p,q{X) space of harmonic (p, q)-forms
f3r(X), f3p ,q(X) Betti numbers

zero section of the bundle F

generalized Siegel upper halfplane

L~,q{n) Hilbert space closure of "'M(n)

dom(T) domain of the operator T
N Neumann operator N : L~,q(n) --+ dom(O)
 Index of Notation 385

rapidly decreasing functions

Sobolev space of index s
tangential Sobolev norm

N(n) Nebenhiille of n

Fx· pseudo-differential metric on X

K(~,w) Bergman kernel
P: L2(n) ~£ Bergman projection
AOO(n) = Jd'n,O(O) n O(n)
Index
I-complete, 267 CaratModory distance, 371
I-convex, 267 Cart an-Thullen
- theorem of, 77
Abel's lemma, 12 Cauchy integral, 17
abelian variety, 351 Cauchy-Riemann equations, 29
absolute space, 7 tech cohomology, 185
accessible boundary point, 101 center
acyclic covering, 185 - of a modification, 236
adjoint operator, 330 chain rule, 32
adjunction formula, 215 Chern class, 256
affine algebraic, 217 Chow
affine algebraic manifold, 224 - theorem of, 217
affine algebraic set, 224 closed map, 226
affine part, 210 closure
algebraic derivative, 125 - of en, 245
analytic disk, 47, 102 - of a manifold, 244
analytic polyhedron, 85 coboundar~ 184, 187
analytic set, 36, 160 coboundary operator, 183
analytic tangent, 275 cochain, 183, 187
analytically dependent, 320 cocycle, 183, 187
antiholomorphic form, 303 codimension, 151
antiholomorphic gradient, 28 cohomologous, 174
Ascoli cohomology group, 184, 187
- lemma of, 23 commutator, 332
atlas, 155 compact exhaustion, 163
automorphism, 226 compact map, see completely
continuous
ball, 5 compactification
Banach algebra, 105 - one-point, 153
Bergman kernel, 370 compatibility condition, 173
Bergman metric, 372 compatible coordinate systems, 154
Bergman projection, 370 complete hull, 24
Bertini completely continuous, 285
- theorem of, 225 completely singular, 49
Betti number, 190 complex I-form, 261
biholomorphic, 33, 159 complex covariant r-tensor, 261
blowup, 237 complex differentiable, 14
boundary conditions, 357 complex gradient, 15
boundary distance, 54, 58 complex manifold, 155
boundary function, 64 complex n-space, 1
boundary operator, 189 complex structure, 1, 155
bracket, 332 complexification, 2
branch locus, 227 cone, 216
branch point, 128 conical set, 216
branched covering, 128, 227 conjugation, 2
branched domain, 229 connected, 6, 88
bubble method, 283 continuity of roots, 134
continuity principle, 47
canonical bundle, 174 continuity theorem, 43
388 Index

contractible space, 189 Dolbeault group, 311

convergence Dolbeault sequence, 310
- compact, 12 Dolbeault's lemma, 309
- domain of, 13 Dolbeault's theorem, 312
convergent series domain, 6
- of numbers, 10 domain of existence, 97
convex, 8, 66, 74 domain of holomorphy, 49, 99
- strictly, 67 - weak, 49
convex hull, 73 . dual bundle, 178
coordinate system, 154 duality theorem
Cousin-I distribution, 253 - of Poincare, 337
Cousin-II distribution, 254 - of Serre-Kodaira, 338
covariant tangent vector, 261
critical locus, 227 effective divisor, see positive divisor
cross section, 176 effective group operation, 172
cuboid, 276 elementary symmetric polynomial, 126
- distinguished, 276 elliptic operator, 337
current, 328 embedded-analytic
- harmonic, 336 - irreducible component, 135
- set, 135
embedding, 214
8-Neumann conditions, 357
embedding dimension, 152
de Rham group, 311
embedding theorem, 145, 258
de Rham isomophism, 312 energy form, 359
de Rham sequence, 310 envelope of holomorphy, 87, 97
decomposition theorem equivalence
- of Hodge-Kodaira, 339 - of bundles, 174
defining function, 64 - of systems of transition functions,
degree 174·
- of a hypersurface, 220 Euclidean coordinates, 322
derivation, 33, 164 Euclidean domain, 125
derivative Euler sequence, 223
- directional, 16 exact sequence, 223
- of a holomorphic map, 30 exceptional set, 236
- partial, 15 exhaustion function, 58
- real, 26 exterior algebra, 298
derivatives exterior derivative, 301
- Wirtinger, 28
diagonal, 162 factorial monoid, 125
differential, 260, 300 faithful, see effective group operation
differential form, 297, 300 Fefferman
dimension, 161 - theorem of, 369
- local, 150 fiber bundle, 173
- of a submanifold, 39 fiber bundle isomorphism, 174
- of an analytic set, 142 fiber metric, 341
- of an irreducible analytic set, 141 fiber product, 171
Dirichlet's principle, 52 filter basis, 100
discrete map, 228 finite map, 227
discriminant, 127 finite module, 121
discriminant set, 128 fixed point, 172
distinguished boundary, 6 form of type (p, q), 298
divisor, 198 frame, 177, 313
- of a meromorphic function, 198 Frechet space, 285
- of a meromorphic section, 201 Frechet topology, 285
 Index 389

free group operation, 172 homogeneous, 9

Fubini metric, 319 homogeneous coordinates, 209
fundar"nental form, 315 Hopf bundle, 238
Hopf manifold, 208, 225
Gauss's lemma, 117 Hopf's a-process, 237
general linear group, 172 hyperbolic manifold, 368
genus, 232 hyperelliptic involution, 234
- of a Riemann surface, 340 hyperelliptic surface, 232
geodesic coordinates, 316 hyperplane at infinity, 210
geometric series, 11 hyperplane bundle, 220, 221
germ, 123, 192 hyperquadric, 219
globally generated, 176 hypersurface, 127, 220
gluing, 170 - analytic, 160
- of fiber bundles, 173
- of manifolds, 170 ideal, 108
Godeaux surface, 225 identity theorem
good topological space, 189 for analytic sets, 142
graph,161 for holomorphic functions, 22
- of a meromorphic function, 242 for meromorphic functions, 197
Grassmannian, 212, 225 for power series, 21
greatest common divisor, 117 - on manifolds, 156
Griffiths negative, 341 immersion, 168
Griffiths positive, 342 implicit function theorem, 34
incidence set, 237
indeterminacy
harmonic form, 335, 356 - set of, 243
harmonic function, 52 inner product
Hartogs - Euclidean, 4
- theorem of, 307 - Hermitian, 3
Hartogs convex, 48, 99 integral class, 346
Hartogs figure integral domain, 113
- Euclidean, 25 inverse mapping theorem, 33
- general, 43, 99 irreducible analytic set, 141
Hartogs triangle, 365 irreducible component, 131
Hausdorff space, 153 irreducible element, 117
Hensel's lemma, 120 isomorphism
,Hermitian form, 3, 261, 321 of complex manifolds, 159
Hermitian metric, 314 - of vector bundles, 177
Hessian, 67
H n ,108 Jacobian
Hodge manifold, 346 - complex, 30
Hodge metric, 346 - real,31
Hodge star operator, 323
Holder continuous, 367 Kahler manifold, 316
holomorphic form, 303 Kahler metric, 316
holomorphic function, 13, 96, 257 kernel form, 370
- on a complex manifold, 156 Kobayashi metric, 368
- on an analytic set, 134 Kodaira's embedding theorem, 344
holomorphic gradient, 28 Krull topology, 110
holomorphic map, 30, 158 Kugelsatz, 44, 307
holomorphically convex, 75, 100, 251
holomorphically convex hull, 75 Laplacian, 334
holomorphically separable, 251 Leray covering, see acyclic covering
holomorphically spreadable, 251 Levi condition, 66
390 Index

Levi convex, 66 normal bundle, 182

- strictly, 66 normal exhaustion, 78
Levi form, 57, 263 normalized polynomial, 113
Levi's extension theorem, 197 normally convergent, 11
Levi's problem, 51 nowhere dense, 37
Lie group, 171
lifted bundle, 180 Oka
lifting - theorem of, 100
- of a differential form, 304 Oka's principle, 253
limit open covering, 153
- of a filter, 101 open map, 88
line bundle, 175 orbit space, 205
- associated to a hypersurface, 200 order
line element, 314 - of a power series, 119
linear subspace, 218 Osgood
Liouville - theorem of, 19
- theorem of, 22 Osgood space, 243
locall(>algebra, 192
local homeomorphism, 88 paracompact, 153
local parametrization, 39, 41 parameter system, 151
local potential, 318 partially differentiable, 16
local uniformization, 231 partition of unity, 163
locally compact, 153 pathwise connected, 88
locally finite covering, 153 period matrix, 348
logarithmically convex, 83 Picard group, 187
Plucker embedding, 218
pluriharmonic function, 318
maximal ideal, 108 plurisubharmonic, 56, 261, 263
maximum principle, 22 - strictly, 59
- for plurisubharmonic functions, 264 Poincare map, 301
- for subharmonic functions, 55 point of indeterminacy, 196
mean value property, 52 polar set, 196
meromorphic function, 196 pole, 196
- on a Riemann surface, 158 polydisk,5
meromorphic map, 243 polynomial,9
meromorphic section, 201 polynomially convex, 87
minimal defining function, 194 positive definite, 261
modification, 236 positive divisor, 198
- generalized, 236 positive form, 303
module, 120 positive line bundle, 288
Moishezon manifold, 320, 353 power series
monic polynomial, 113 - convergent, 108
monoidal transformation, 242 - formal, 11, 105
monomial,9 power sum, 134
Montel prime element, 117
- theorem of, 23 primitive polynomial, 117
principal divisor, 254
Nebenhulle, 364 principal ideal domain, 117
negative line bundle, 287 projective algebraic
neighborhood filter, 100 - manifold, 217
Neil parabola, 134 - set, 217
Neumann operator, 359 projective space, 208
Neumann problem, 357 projective unitary transformations, 318
noetherian, 121 proper map, 35, 226
 Index 391

properly discontinuous, 205 self-intersection number, 241

pseuqoconvex, 60, 103 seminegative line bundle, 352
- strongly, 73 seminorm, 284
pseudoconvex manifold, 267 semi positive line bundle, 352
pseudodifferential metric, 368 Serre duality, 233
pseudopolynomial, 124 Serre problem, 259
pullback, see lifted bundle shear, 109
pure-dimensional, 143 Siegel upper halfplane, 352
singular cochain, 189
quadratic transformation, 239 singular cohomology group, 189
quotient bundle, 181 singular homology group, 189
quotient field, 117 singular locus, 147
quotient topology, 203 singular point, 39
singular q-chain, 188
rational function, 223 singular q-simplex, 188
refinement, 153 smooth boundary, 64
refinement map, 153 smoothing lemma, 59
region, 6 Sobolev norm
regular closure, 245 - tangential, 358
regular domain, 360 Soholev space, 358
regular function, 224 standard simplex, 188
regular point, 39, 140, 161 star, 283
Reinhardt domain, 7 Stein manifold, 251
complete, 8, 85 Stiefel manifold, 212
- over en, 95
Stokes's theorem, 305
- proper, 8
strict transform, 240
Remmert-Stein
strictly pseudoconvex manifold, 267
- extension theorem, 150
strongly pseudoconvex, 267
removable boundary point, 102
structure group, 173
removable singularity, 196
StiitzfUi.che, 275
resolution of singularities, 240
subbundle, 180
Riemann domain
subelliptic, 358
- branched, 229
subharmonic function, 53
- with distinguished point, 89
submanifold, 41, 161
Riemann extension theorem
- first, 38 submersion, 168
symbol
- second, 151
Riemann surface - of a differential operator, 336
- abstract, 157 symmetric polynomial, 126
- concrete, 230
- of.,fi, 88 tangent bundle, 176
Ritt's lemma, 151 - of projective space, 222
Riickert basis theorem, 122 tangent space, 32, 165
Runge domain, 87 - holomorphic, 66
tangent vector, 32, 165
saturated set, 203 tangential map, 167
scalar product, 3 tautological bundle, 238
schlicht domain, 90 tensor power, 178
Schwartz tensor product, 179, 261
- theorem of, 285 Theorem A, 252
second axiom of count ability, 153 Theorem B, 253
section, see cross section topological map, 41
- in a fiber bundle, 186 topological quotient, 203
Segre map, 225 toric closure, 246
392 Index

toric variety, 246 Veronese map, 224

torsion point, 230 volume element, 323
torus, 207, 225, 348
transformation group, 172 weakly holomorphic, 16
transition functions,· i 73 wedge product, 297
transversal, 171 Weierstrass
trivial vector bundle, 177 - theorem of, 19
trivialization, 173 Weierstrass condition, 113
- of a vector bundle, 175 Weierstrass division formula, 115
tube, 288 Weierstrass formula, 110
tube domain, 87 Weierstrass function, 226
Weierstrass polynomial, 114
unbranched point, 133
uniformization, 88, 162 Weierstrass preparation theorem, 113,
116
union of Riemann domains, 94
unique factorization domain, 117 Whitney sum, 177
worm domain, 366
vector bundle, 175
vector bundle chart, 175 Zl -regular, 109
vector bundle homomorphism, 177 Zariski tangential space, 152
vector field, 176 Zariski topology, 141
Graduate Texts in Mathematics
(contiNud/rom page ii)

64 EDwARDS. Fourier Series. Vol. I. 2nd cd. 96 CONWAY. A Course in Functional

65 WELLS. Differential Analysis on Complex Analysis. 2nd cd.
Manifolds. 2nd cd. 97 KOBun. Introduction to Blliptic Curves
66 WA11!RHOUSE. Introduction to Affine and Modular Forms. 2nd ed.
Group Schemes. 98 BROCKERIfOM DIECK. Representations of
67 SERRE. Loca1 Fields. Compact Lie Groups.
68 WEIDMANN. Linear Operators in Hilbert 99 GRoVElBENSON. Finite Reflection Groups.
Spaces. 2nded.
69 LANG. Cyclotomic Fields U. 100 BERGICHlusTENSENIRBssEL. Harmonic
70 MAssEY. Singular Homology Theory. Analysis on Semigroups: Theory of
71 FARKASIKRA. Riemann Surfaces. 2nd ed. Positive Definite and Related Functions.
72 STIlLWEll.. Classical Topology and 101 EDWARDs. Galois Theory.
Combinatorial Group Theory. 2nd ed. 102 VAR.ADAR.AJAN. Lie Groups, Lie Algebras
73 HUNGERFORD. Algebra. and Their Representations.
74 DAVENPORT. Multiplicative Number 103 LANG. Complex Analysis. 3rd ed.
Theory. 3rd ed. 104 DUBROVINlFoMENKcVNoVIKOV. Modem
75 HOCHSOIlLD. Basic Theory of Algebraic Geometry-Methods and Applications.
Groups and Lie Algebras. PartU.
76 IrrAXA. Algebraic Geometry. 105 LANG. SLz(R).
77 HEcicE. Lectures on the Theory of 106 Sn.VERMAN. The Arithmetic of Elliptic
Algebraic Numbers. Curves.
78 BURRIsiSANKAPPANAVAR.. A Course in 107 OLVER. Applications of Lie Groups to
Universal Algebra. Differential Equations. 2nd ed.
79 WALTERS. An Introduction to Ergodic 108 RANGE. Holomorpbic Functions and
Theory. Integral Representations in Several
80 ROBINSON. A Course in the Theory of Complex Variables.
Groups. 2nd cd. 109 1BITo. Univalent Functions and
81 FORSlER. Lectures on Riemann Surfaces. TeicbmU11er Spaces.
82 Barrrru. Differential Forms in Algebraic 110 LANG. Algebraic Number Theory.
Topology. 111 HuSEMOlJ..ER. Elliptic Curves.
83 WASHINGTON. Introduction to Cyclotomic 112 LANG. Elliptic Functions.
Fields. 2nd cd. 113 KARATZASfSHRBVE. Brownian Motion and
84 IRELANIWROSEN. A Classical Introduction Stochastic Calculus. 2nd ed.
to Modem Number Theory. 2nd ed. 114 KOBun. A Course in Number Theory and
85 EDWARDS. Fourier Series. Vol. U. 2nd cd. Cryptography. 2nd ed.
86 VAN LlNT. Introduction to Coding Theory. 115 BERGERlGosnAUX. Differential Geometry:
2nded. Manifolds. Curves. and Surfaces.
87 BROWN. Cohomology of Groups. 116 KEu.l!y/SRlNlVASAN. Measure and
88 PIERCE. Associative Algebras. Integral. Vol. I.
89 LANG. Introduction to Algebraic and 117 SERRE. Algebraic Groups and Class Fields.
Abelian Functions. 2nd ed. 118 PEDERSEN. Analysis Now.
90 BR0NDSTED. An Introduction to Convex 119 ROTMAN. An Introduction to Algebraic
Polytopes. Topology.
91 BEARDON. On the Geometry of Discrete 120 ZIEMER. Weakly Differentiable Functions:
Groups. Sobolev Spaces and Functions of Bounded
92 D/ESTEL. Sequences and Series in Banach Variation.
Spaces. 121 LANG. Cyclotomic Fields I and U.
93 DUBRovINIFoMENKcVNoVIKOV. Modem Combined 2nd ed.
Geometry-Methods and Applications. 122 REMMERT. Theory of Complex Functions.
Part I. 2nd cd. Readings in Mathemlllics
94 WARNER. Foundations of Differentiable 123 BBBINGHAusIHERMEs et al. Numbers.
Manifolds and Lie Groups. Readings in MatMmatics
95 SHIRYAEV. Probability. 2nd cd.
124 DuBRovINIFoMENxolNoVIKOV. Modem 154 BROWNIPEARCY. An Introduction to
Geometry-Methods and Applications. Analysis.
PartID 155 KAssEL. Quantum Groups.
125 BERENSTEINIGAY. Complex Variables: 156 KEcHlUs. Classical Descriptive Set
An Introduction. Theory.
126 BOREL. Linear Algebraic Groups. 2nd ed. 157 MALLIAVIN. IntegIlltion and
127 MAsSEY. A Basic Course in Algebraic Probability.
Topology. 158 ROMAN. Field Theory.
128 RAuOI. Partial Differential Equations. 159 CONWAY. Functioos of One
129 PuLTONIHARRIs. Representation Theory: A Complex Variable n.
rmt Course. 160 LANG. Differential and Riemannian
Readings in Mathematics Manifolds.
130 DoDSONlPosToN. Tensor Geometry. 161 BORWElNIERDm.YI. Polynomials and
131 LAM. A First Course in Noncommutative Polynomial Inequalities.
Rings. 2nd ed. 162 Au>ERlNIBELL. Groups and
132 BBARDON. Iteration of Rational Functions. Representations.
133 HARRIs. Algebraic Geometry: A First 163 DIXONIMORTIMEIt. Permutation Groups.
Course. 164 NAniANSON. Additive Number Theory:
134 ROMAN. Coding and Information Theory. The Classical Bases.
135 ROMAN. Advanced Linear Algebra. 165 NAnlANSON. Additive Number Theory:
136 ADKJNslWElNTRAUB. Algebra: An Inverse Problems and the Geometry of
Approach via Module Theory. Sumsets.
137 AxLBRIBOORDON!RAMBY. Harmonic 166 SHARPE. Differential Geometry: Carlan's
Function Theory. 2nd ed. Generalization of Klein's Erlangen
138 COHEN. A Course in Computational ProgIam.
Algebraic Number Theory. 167 MORANDI. Field and Galois Theory.
139 BRBOON. Topology and Geometry. 168 EWALD. Combinatorial Convexity and
140 AUBIN. Optima and Equilibria. An Algebraic Geomeuy.
Introduction to Nonlinear Analysis. 169 BHATIA. Matrix Analysis.
141 BBCXERIWBlSPF6NNINGlKJumEL. GrtIbner 170 BREOON. ShcafTheory. 2nd ed.
Bases. A Computational Approach to 171 PE'reRSEN. Riemannian Geometry.
Commutative Algebra. 172 REMMERT. Classical Topics in Complex
142 LANG. Real and Functional Analysis. Function Theory.
3rded. 173 DIESTEL. Graph Theory. 2nd ed.
143 DooB. Measure Theory. 174 BRIDGES. Foundations of Real and
144 DENNISIFARB. Noncommutative Abstract Analysis.
Algebra. 175 UCKORISH. An Introduction to Knot
145 VICK. Homology Theory. An Theory.
Introduction to Algebraic Topology. 176 lB!. Riemannian Manifolds.
2nded. 177 NEWMAN. Analytic Number Theory.
146 BRIDGES. Computability: A 178 Cl.ARKEILEDYAEv/STERNIWOLENSKI.
Mathematical Sketchbook. Nonsmooth Analysis and Control
147 ROSENBERG. Algebraic K-Theory Theory.
and Its Applications. 179 DoUGLAS. Banach Algebra Techniques in
148 ROTMAN. An Introduction to the Operator Theory. 2nd cd.
Theory of Groups. 4th ed. 180 SRIVASTAVA. A Course OD Borel Sets.
149 RA~.Foundationsof 181 KRESs. Numerical Analysis.
Hyperbolic Manifolds. 182 WALlER.. Ordinary Differential
150 EISENBUD. Commutative Algebra Equations.
with a View Toward Algebraic 183 MEGGINSON. An Introduction to Banach
Geometry. Space Theory.
151 Su..VERMAN. Advanced Topics in 184 BOllOBAS. Modern Graph Theory.
the Arithmetic of Elliptic Curves. 185 COXlL1TlUJO'SHEA. Using Algebraic
152 ZIEGLER. Lectures on Polytopes. Geometry.
153 FULTON. Algebraic Topology: A
rll"StCourse.
186 RAMAKRlSHNANN ALENZA. Fourier 201 HlNDRY/Sn.VERMAN. Diophantine
Analysis on Number Fields. Geometry: An Introduction.
187 HARRISIMoRRlSON. Moduli of Curves. 202 l..I:E. Introduction to Topological
188 GoLDBLA1T. Lectures on the Hyperrea1s: Manifolds.
An Introduction to Nonstandard Analysis. 203 SAGAN. The Symmetric Group:
189 LAM. Lectures on Modules and Rings. Representations, Combinatorial
190 EsMONDE!MURlY. Problems in Algebraic Algorithms, and Symmetric Functions.
Number Theory. 204 EscoFlER. Galois Theory.
191 LANG. Fundamentals of Differential 205 FBJxIHAu>ERINlTHOMAS. Rational
Geometry. Homotopy Theory. 2nd cd.
192 HIRSCHILACOMBE. Elements of 206 MURlY. Problems in Analytic Number
Functional Analysis. Theory.
193 COHEN. Advanced Topics in Readings in Mathematics
Computational Number Theory. 207 Goosn.IROYLE. Algebraic Graph Theory.
194 ENGEIlNAGEL. One-Parameter Semigroups 208 CHENEY. Analysis for Applied
for Linear Evolution Equations. Mathematics.
195 NATIIANSON. Elementary Methods in 209 ARVESON. A Short Course on Spectral
Number Theory. Theory.
196 OSBORNE. Basic Homological Algebra. 210 ROSEN. Number Theory in Function
197 EISENBUDIHARRIs. The Geometry of Fields.
Schemes. 211 LANG. Algebra. Revised 3rd ed.
198 ROBERT. A Course in p-adic Analysis. 212 MATOUSEK. Lectures on Discrete
199 HEDENMALMlKORENBLUMIZHu. Theory Geometry.
of Bergman Spaces. 213 FRrrzsCHElGRAUERT. From Holomorphic
200 BAOICHERNISHEN. An Introduction to Functions to Complex Manifolds.
Riemann-Finsler Geometry.
 ~*.fr!l&t.fiij ~ (CIP) ~~
Jk~~pj§~ifJ~tliJ~ = From Holomorphic Functions to
Complex Manifolds: !Ii:)( / Oto !it ~ tJ] (Fritzsche, K.)
it. -::f~]j(: tttm-l!I~ifj!l&0~::I~/'R0~, 2009.5
ISBN 978-7-5100-0471-1

I. ]A ... II. !it ... lli. CD~M~pj§~-!li:)(cz)~t1EJf5-!Ii:

)( IV. 0174. 5
9"~!l&*I!I~i1r CIP ~tJ.ii~¥ (2009) :m 072986 %

* 4!; : From Holomorphic Functions to Complex Manifolds

11= 1!f: Klaus Fritzsche & Hans Grauert

J:j:I ~ 4!;: ]A~~£I§~ifJ~rJiEJf5

~tEt,fiij~: ~?8 xU~

ill !l& 1!f: tftm-l!I~ifj!l&0~::f~]j(0~

Ep ijlj 1!f: ~:ri1J ~!Ii: fp jj- :fH~0 ~
~ ~T: tftm-l!I~ifj!l&0~::I~]j(0~ (::f~]j(ljiApq*m 137 % 100010)
~~~iiS: 010~4021602, 010-64015659
~T1~fii: kjb@ wpcbj. com. cn

7f *: 243f
EP ~*: 17.5
!l& IX : 2009 ~ 06 J1
!l&*Jl~i2: I!I¥: 01-2009-"1082

.!f}: -978-7-5100-0471-1/0·686 ffi": 49. 00 j[;