Grundlehren der

mathematischen Wissenschaften 267

A Series of Comprehensive Studies in Mathematics

Editors

M. Artin S. S. Chern J. M. Frohlich E. Heinz

H. Hironaka F. Hirzebruch L. Hormander S. Mac Lane
W. Magnus C. C. Moore J. K. Moser M. Nagata
W. Schmidt D. S. Scott Ya. G. Sinai J. Tits
B. L. van der Waerden M. Waldschmidt S. Watanabe

Managing Editors

M. Berger B. Eckmann S. R. S. Varadhan

Grundlehren der mathematischen Wissenschaften
A Series of Comprehensive Studies in Mathematics
A Selection
180. Landkof: Foundations of Modem Potential Theory
181. Lions/Magenes: Non-Homogeneous Boundary Value Problems and Applications I
182. Lions/Magenes: Non-Homogeneous Boundary Value Problems and Applications II
183. Lions/Magenes: Non-Homogeneous Boundary Value Problems and Applications Ill
184. Rosenblatt: Markov Processes, Structure and Asymptotic Behavior
185. Rubinowicz: Sommerfeldsche Polynommethode
186. Handbook for Automatic Computation. Vol. 2. Wilkinson/Reinsch: Linear Algebra
187. Siegel/Moser: Lectures on Celestial Mechanics
188. Warner: Harmonic Analysis on Semi-Simple Lie Groups I
189. Warner: Harmonic Analysis on Semi-Simple Lie Groups II
190. Faith: Algebra: Rings, Modules, and Categories I
191. Faith: Algebra II, Ring Theory
192. Mallcev: Algebraic Systems
193. P61ya/Szego: Problems and Theorems in Analysis I
194. Igusa: Theta Functions
195. Berberian: Baer*-Rings
196. Athreya/Ney: Branching Processes
197. Benz: Vorlesungen iiber Geometric der Algebren
198. Gaal: Linear Analysis and Representation Theory
199. Nitsche: Vorlesungen iiber Minimalfliichen
200. Dold: Lectures on Algebraic Topology
201. Beck: Continuous Flows in the Plane
202. Schmetterer: lntroduct'ion to Mathematical Statistics
203. Schoeneberg: Elfiptic Modular Functions
204. Popov: Hyperstability of Control Systems
205. Nikollskii: Approximation of Functions of Several Variables and Imbedding Theorems
206. Andre: Homologie des Algebres Commutatives
207. Donoghue: Monotone Matrix Functions and Analytic Continuation
208. Lacey: The Isometric Theory of Classical Banach Spaces
209. Ringel: Map Color Theorem
210. Gihman/Skorohod: The Theory of Stochastic Processes I
211. Comfort/Negrepontis: The Theory of Ultrafilters
212. Switzer: Algebraic Topology-Homotopy and Homology
213. Shafarevich: Basic Algebraic Geometry
214. van der Waerden: Group Theory and Quantum Mechanics
215. Schaefer: Banach Lattices and Positive Operators
216. P6Iya/Szego: Problems and Theorems in Analysis II
217. Stenstrom: Rings of Quotients
218. Gihman!Skorohod: The Theory of Stochastic Process II
219. Duvant/Lions: Inequalities in Mechanics and Physics
220. Kirillov: Elements of the Theory of Representations
221. Mumford: Algebraic Geometry 1: Complex Projective Varieties
222. Lang: Introduction to Modular Forms
223. Bergh/Lofstrom: Interpolation Spaces. An Introduction
224. Gilbarg/Trudinger: Elliptic Partial Differential Equations of Second Order

Continued after Index

E. Arbarello
M. Cornalba
P. A. Griffiths
J. Harris

Geometry of Algebraic Curves

Volume I

With 10 Illustrations

Springer Science+Business Media, LLC

E. Arbarello M. Cornalba
Dipartimento di Matematica Dipartimento di Matematica
Istituto "Guido Castelnuovo" Universita di Pavia
Universita di Roma "La Sapienza" 27100 Pavia
00187 Roma Italia
Italia

P. A. Griffiths J. Harris
Office of the Provost Department of Mathematics
Duke University Brown University
Durham, NC 27760 Providence, RI 02912
U.S.A. U.S.A.

AMS Subject Classification: 14-XX

Library of Congress Cataloging in Publication Data

Main entry under title:
Geometry of algebraic curves.
(Grundlehren der mathematischen Wissenschaften; 267)
Bibliography: v. 1, p.
lncludes index.
1. Curves, Algebraic. I. Arbarello, E. II. Series.
QA565.G46 1984 512'.33 84-5373

© 1985 by Springer Science+Business Media New York

Originally published by Springer-Verlag New York Inc. in 1985
Softcover reprint ofthe hardcover 1st edition 1985
Ali rights reserved. No part of this book may be translated or reproduced in any form
without permission from Springer Science+Business Media, LLC.

Typeset by Composition House Ltd., Salisbury, England.

987654 32 1
ISBN 978-1-4419-2825-2 ISBN 978-1-4757-5323-3 (eBook)
DOI 10.1007/978-1-4757-5323-3
To the memory of Aldo Andreotti
Preface

In recent years there has been enormous activity in the theory of algebraic
curves. Many long-standing problems have been solved using the general
techniques developed in algebraic geometry during the 1950's and 1960's.
Additionally, unexpected and deep connections between algebraic curves
and differential equations have been uncovered, and these in turn shed
light on other classical problems in curve theory. It seems fair to say that the
theory of algebraic curves looks completely different now from how it
appeared 15 years ago; in particular, our current state of knowledge repre-
sents a significant advance beyond the legacy left by the classical geometers
such as Noether, Castelnuovo, Enriques, and Severi.
These books give a presentation of one of the central areas of this recent
activity; namely, the study of linear series on both a fixed curve (Volume I)
and on a variable curve (Volume II). Our goal is to give a comprehensive
and self-contained account of the extrinsic geometry of algebraic curves,
which in our opinion constitutes the main geometric core of the recent
advances in curve theory. Along the way we shall, of course, discuss appli-
cations of the theory of linear series to a number of classical topics (e.g., the
geometry of the Riemann theta divisor) as well as to some of the current
research (e.g., the Kodaira dimension of the moduli space of curves).
A brief description of the contents of the various chapters is given in the
Guide for the Reader. Here we remark that these volumes are written in the
spirit of the classical treatises on the geometry of curves, such as Enriques-
Chisini, rather than in the style of the theory of compact Riemann surfaces,
or of the theory of algebraic functions of one variable. Of course, we hope
that we have made our subject understandable and attractive to interested
mathematicians who have studied, in whatever manner, the basics of curve
theory and who have some familiarity with the terminology of modern
algebraic geometry.
We now would like to say a few words about that material which is not
included. Naturally, moduli of curves play an essential role in the geometry
of algebraic curves, and we have attempted to give a useful and concrete
discussion of those aspects of the theory of moduli that enter into our work.
However, we do not give a general account of moduli of curves, and we say
nothing about the theory ofTeichmiiller spaces. Secondly, it is obvious that
Vlll Preface

theory of abelian varieties and theta functions is closely intertwined with

algebraic curve theory, and we have also tried to give a self-contained
presentation of those aspects that are directly relevant to our work, there
being no pretense to discuss the general theory of abelian varieties and
theta functions. Thirdly, again the general theory of algebraic varieties
clearly underlies this study; we have attempted to utilize the general methods
in a concrete and practical manner, while having nothing to add to the
theory beyond the satisfaction of seeing how it applies to specific geometric
problems. Finally, arithmetical questions, as well as the recent beautiful
connections between algebraic curves and differential equations including
the ramifications with the Schottky problem, are not discussed.
These books are dedicated to Aldo Andreotti. He was a man of tremendous
mathematical insight and personal wisdom, and it is fair to say that
Andreotti's view of our subject as it appears, for example, in his classic paper
"On a theorem of Torelli," set the tone for our view of the theory of algebraic
curves. Moreover, his influence on the four of us, both individually and
collectively, was enormous.
It is a pleasure to acknowledge the help we have received from numerous
colleagues. Specifically, we would like to thank Corrado De Concini, David
Eisenbud, Bill Fulton, Mark Green, Steve Kleiman, and Edoardo Sernesi
for many valuable comments and suggestions. We would also like to express
our appreciation to Roy Smith and Harsh Pittie, who organized a conference
on Brill-Noether theory in February, 1979 at Athens, Georgia, the notes of
which formed the earliest (and by now totally unrecognizable) version of
this work.
Also, it is a pleasure to thank Steve Diaz, Ed Griffin, and Francesco
Scattone for excellent proofreading.
Finally, our warmest appreciation goes to Laura Schlesinger, Carol
Ferreira, and Kathy Jacques for skillfully typing the various successive
versions of the manuscript.
Contents

Guide for the Reader xii

List of Symbols XV

CHAPTER I
Preliminaries I
§1. Divisors and Line Bundles on Curves I
§2. The Riemann-Roch and Duality Theorems 6
§3. Abel's Theorem I5
§4. Abelian Varieties and the Theta Function 20
§5. Poincare's Formula and Riemann's Theorem 25
§6. A Few Words About Moduli 28
Bibliographical Notes 30
Exercises 31
A. Elementary Exercises on Plane Curves 3I
B. Projections 35
C. Ramification and Plucker Formulas 37
D. Miscellaneous Exercises on Linear Systems 40
E. Weierstrass Points 4I
F. Automorphisms 44
G. Period Matrices 48
H. Elementary Properties of Abelian Varieties 48
APPENDIX A
The Riemann-Roch Theorem, Hodge Theorem, and
Adjoint Linear Systems 50
§1. Applications of the Discussion About Plane Curves with Nodes 56
§2. Adjoint Conditions in General 57
CHAPTER II
Determinantal Varieties 6I
§I. Tangent Cones to Analytic Spaces 6I
§2. Generic Determinantal Varieties: Geometric Description 67
§3. The Ideal of a Generic Determinantal Variety 70
§4. Determinantal Varieties and Porteous' Formula 83
(i) Sylvester's Determinant 87
(ii) The Top Chern Class of a Tensor Product 89
(iii) Porteous' Formula 90
(iv) What Has Been Proved 92
X Contents

§5. A Few Applications and Examples 93

Bibliographical Notes 100
Exercises 100
A. Symmetric Bilinear Maps 100
B. Quadrics 102
C. Applications of Porteous' Formula 104
D. Chern Numbers of Kernel Bundles 105
CHAPTER III
Introduction to Special Divisors 107
§1. Clifford's Theorem and the General Position Theorem 107
§2. Castelnuovo's Bound, Noether's Theorem, and Extremal Curves 113
§3. The Enriques-Babbage Theorem and Petri's Analysis of the Canonical Ideal 123
Bibliographical Notes 135
Exercises 136
A. Symmetric Products of IP' 1 136
B. Refinements of Clifford's Theorem 137
C. Complete Intersections 138
D. Projective Normality (I) 140
E. Castelnuovo's Bound on k-Normality 141
F. Intersections of Quadrics 142
G. Space Curves of Maximum Genus 143
H. G. Gherardelli's Theorem 147
I. Extremal Curves 147
J. Nearly Castelnuovo Curves 149
K. Castelnuovo's Theorem 151
L. Secant Planes 152
CHAPTER IV
The Varieties of Special Linear Series on a Curve 153
§1. The Brill-Noether Matrix and the Variety q 154
§2. The Universal Divisor and the Poincare Line Bundles 164
§3. The Varieties w;(C) and G;(C) Parametrizing Special Linear Series on a
Curve 176
§4. The Zariski Tangent Spaces to G:i(C) and w;(C) 185
§5. First Consequences of the Infinitesimal Study of G;j(C) and W:i(C) 191
Biographical Notes 195
Exercises 196
A. Elementary Exercises on /lo 196
B. An Interesting Identification 197
C. Tangent Spaces to W1 (C) 197
D. Mumford's Theorem for _qJ's 198
E. Martens-Mumford Theorem for Birational Morphisms 198
F. Linear Series on Some Complete Intersections 199
G. Keem's Theorems 200

CHAPTER V
The Basic Results of the Brill- N oether Theory 203
Bibliographical Notes 217
Exercises 218
Contents XI

A. W}(C) on a Curve C of Genus 6 218

B. Embeddings of Small Degree 220
C. Projective Normality (II) 221
D. The Difference Map rjJd: Cd x Cd-+ J(C) (I) 223

CHAPTER VI
The Geometric Theory of Riemann's Theta Function 225
§1. The Riemann Singularity Theorem 225
§2. Kempf's Generalization of the Riemann Singularity Theorem 239
§3. The Torelli Theorem 245
§4. The Theory of Andreotti and Mayer 249
Bibliographical Notes 261
Exercises 262
A. The Difference Map ¢d (II) 262
B. Refined Torelli Theorems 263
C. Translates of~-~> Their Intersections, and the Torelli Theorem 265
D. Prill's Problem 268
E. Another Proof of the Torelli Theorem 268
F. Curves of Genus 5 270
G. Accola's Theorem 275
H. The Difference Map ¢d (III) 276
I. Geometry of the Abelian Sum Map u in Low Genera 278

APPENDIX B
Theta Characteristics 281
§1. Norm Maps 281
§2. The Wei! Pairing 282
§3. Theta Characteristics 287
§4. Quadratic Forms Over 71./2 292
APPENDIX C
Prym Varieties 295
Exercises 303

CHAPTER VII
The Existence and Connectedness Theorems for ~(C) 304
§1. Ample Vector Bundles 304
§2. The Existence Theorem 308
§3. The Connectedness Theorem 311
§4. The Class of W;( C) 316
§5. The Class of q 321
Bibliographical Notes 326
Exercises 326
A. The Connectedness Theorem 326
B. Analytic Cohomology of Cd, d :s; 2g - 2 328
C. Excess Linear Series 329

CHAPTER VIII
Enumerative Geometry of Curves 330
§1. The Grothendieck-Riemann-Roch Formula 330
Xll Contents

§2. Three Applications of the Grothendieck-Riemann-Roch Formula 333

§3. The Secant Plane Formula: Special Cases 340
§4. The General Secant Plane Formula 345
§5. Diagonals in the Symmetric Product 358
Bibliographical Notes 364
Exercises 364
A. Secant Planes to Canonical Curves 364
B. Weierstrass Pairs 365
C. Miscellany 366
D. Push-Pull Formulas for Symmetric Products 367
E. Reducibility of Wg_ 1 n (Wg_ 1 + u) (II) 370
F. Every Curve Has a Base-Point-Freeg:_ 1 372

Bibliography 375
Index 383
Guide for the Reader

This book is not an introduction to the theory of algebraic curves. Rather, it

addresses itself to those who have mastered the basics of curve theory and
wish to venture beyond them into more recently explored ground. However,
we felt that it would be useful to provide the reader with a condensed account
of elementary curve theory that would serve the dual purpose of establishing
our viewpoint and notation, and furnish a handy reference for those results
which are more frequently used in the main body of our work. This is done
in the first chapter; quite naturally, few proofs are given in full or even
sketched. One notable exception is provided by the theorem of Riemann,
describing the theta divisor of the Jacobian of a curve of genusg as a translate
of the image under the Abel-Jacobi map of the (g - 1)-fold symmetric
product of the curve.
The reader is assumed to have a working knowledge of basic algebraic
geometry such as is given, for example, in the first chapter of Hartshorne's
book Algebraic Geometry. Occasionally, however, we have been com-
pelled to make use of relatively more advanced results, such as the theory
of base change or the Grothendieck-Riemann-Roch theorem. Our policy,
in this situation, has been to give complete statements and adequate refer-
ences to the existing literature when the results are first used. The main
exception to this rule is provided by Chapter II, which contains a down-to-
earth and utilitarian presentation, with complete proofs, of the first and
second fundamental theorems of invariant theory for the general linear
group, the local structure of determinantal varieties, and their global
enumerative properties such as Porteous' formula. The main reason for this
exception is, of course, that the varieties of special divisors, which form one
of the main objects of study in this book, have a natural determinantal
structure and many of their properties are essentially direct consequences
of general facts about determinant varieties. We also felt that most readers
would prefer a unified account of the results rather than a sequence of
references to scattered sources in the literature.
The first two chapters are thus of a preliminary nature, and most readers
will probably want to use them primarily for reference purposes. The main
theme of the book, that is, the study of special divisors and the extrinsic
geometry of curves, is introduced in Chapter III. Here will be found, beside
XIV Guide for the Reader

elementary facts such as Clifford's theorem (which are discussed here and
not in Chapter I simply because they are close in nature to some of the results
that will be encountered in later chapters), Castelnuovo's description of
extremal curves, Noether's theorem, and the theorems of Enriques-Babbage
and Petri on the canonical ideal.
Chapter IV is of a foundational nature. In it the varieties of special divisors
and linear series on a fixed curve-the main characters of this book-are
defined, and the functors they represent are identified. This is where the
results of Chapter II are first applied in a systematic way. Although con-
taining no major results of independent interest, except for Martens'
improvement of Clifford's theorem and its subsequent refinement by
Mumford, this chapter is, in a sense, the cornerstone on which most of the
later chapters rest.
In Chapter V the main theorems of Brill-Noether theory are stated and
illustrated by means of examples drawn from low-genus cases. In fact most
of the theorems are proved by ad hoc arguments, for genus up to six.
Chapter VI is probably the most geometric in nature and collects many of
the central results about the geometry of the theta divisor of a Jacobian.
The main topics touched upon are Riemann's singularity theorem and its
generalization by Kempf, Andreotti's proof of the Torelli theorem for curves,
and Andreotti and Mayer's approach to the Schottky problem via the heat
equation for the theta function.
Chapter VII contains the proofs of some of the results stated in Chapter V,
notably those of the existence and connectedness theorems, and of the
enumerative formulas for the classes of the varieties of special linear series
and divisors. The enumerative geometry of these varieties, and related ones,
is further investigated in the eighth and final chapter of this volume.
The second volume will contain an exposition of the fundamentals of
deformation theory and of the main properties of the moduli space of curves,
the proof of the remaining results of Brill-Noether theory, a presentation
of the basic properties of the varieties of special linear series on a moving
curve with special attention to series of dimension one and two (that is, to
Hurwitz spaces and varieties of plane curves), and a proof of the theorem that
the moduli space of curves of sufficiently high genus is of general type.
List of Symbols

X 1 rg 22
Pa(C) 1 (J23
g(C) 1 8 23
h;(V, :#') wd 25
multp(D) 2 cd 25
deg(D) 2 (J 25
Div(C) 2 K 27
/lp 2 j(C) 28
(¢) 2 Ag 28
ResP 2 d; 29
!f(D) 3 t(j 29
D ~ 0 3 cp(k) 37
D"'O 4 A(C) 41
IDI 4 Tp(Y) 61
g~ 4 ~(X) 61
¢~ 5 !P~(X) 62
¢L 6 Xk(¢) 83
c/JIDI 6 -F 85
Kc 7 E- F 85
E>c 7 dp,q(a) 86
l(D) 7 f.lo 108, 159
i(D) 7 n(d, r) 116
r(D) 7 c~ 153, 177
f*(D) 8 W~(C) 153, 177
cp(D) 12 G~(C) 153
l5 12 p 159
P1• · · ·, Pd 12 f.lo, w 187
J(C) 17 ¢d 223
u 17, 18 l-d 223
Divd(C) 18 /g 249
cd 18 Nm 281
Pic(C) 18 !PE 304
J"fy 22 {!)IP'E(l) 305
dg 22 K(X) 331
XVI List of Symbols

ch(E) 331 (g(t))ty• ... t~" 341

n 1 331 rif0) 341
td(E) 332 V:i 345

C) 341
Jl.(a, b, m, i, {3)
cPa 358
352

Note. Throughout this book, if V is a vector space (resp. if E is a vector

bundle) we will denote by IP'V (resp. IP'E) the space of one-dimensional
subspaces of V (resp. of the fibers of E); thus
IP'V = Proj(Eij Symn V*).
More generally, if Cis a cone, IP'C will stand for its projectivization. Similarly,
by G(k, V) (resp. G(k, E)) we will mean the space of k-dimensional subspaces
of V (resp. of the fibers of E).
Chapter I

Preliminaries

This book will be concerned with geometric properties of algebraic curves.

Our central problem is to study the various projective manifestations of a
given abstract curve. In this chapter we shall collect various definitions,
notations, and background facts that are required for our work.

§1. Divisors and Line Bundles on Curves

By curve we shall mean a complete reduced algebraic curve over IC; it may be
singular or reducible. When speaking of a smooth curve, we shall always
implicitly assume it to be irreducible. Sometimes, when no confusion is
possible we sha,ll drop the adjective "smooth"; this will only be done in
sections where exclusively smooth curves are being considered.
We shall assume known the basic properties of sheaves and line bundles
on algebraic varieties and analytic spaces, and shall make the usual identi-
fication of invertible sheaves with line bundles and oflocally free sheaves with
vector bundles. When tensoring with line bundles we shall often drop the
tensor product symbol. If ff is a sheaf of IC-vector spaces over a topological
space V, we shall set, as customary

hi(V, ff) = dime Hi(V, ff),

x(ff) = I (-lYhi(V, ff).
The basic invariant of a curve Cis its genus. To be more precise, we shall use
the words arithmetic genus of C to denote the integer

pu(C) = 1 - x(C9c).

Of course, when Cis connected, the arithmetic genus of C equals h 1 (C, C9c).
On the other hand, when C is irreducible, we shall denote by g(C) its geo-
metric genus, which is defined to be the (arithmetic) genus of its normalization.
In this book we shall usually talk about the genus of a curve without further
specification; hopefully, it will always be clear from the context which genus
we are referring to. A very basic but non-elementary fact, which can be proved
2 I. Preliminaries

via potential theory, is that the genus of a smooth curve Cis one-half of the
first Betti number of the underlying topological surface; in symbols

g(C) = t rank(H 1 (C, Z)).

Throughout this chapter we shall fix a smooth curve C.

To understand the geometry of C it is essential to study its meromorphic
functions; this is best done in the language of divisors and line bundles. A
divisor

ni E Z and Pi E C,

is a formal linear combination of points on C. We may assume that the Pi

are distinct, and then ni is the multiplicity multP,(D) of D at Pi· The divisors
form a group Div(C), and the degree homomorphism

deg: Div(C) ~ Z

is defined by

deg(D) =I ni.
i

The group of divisors of degree zero is denoted by Div0 (C).

If¢ is a meromorphic function (resp., a meromorphic differential) on C,
then in terms of a local holomorphic coordinate z on C,

<P = f(z) (resp., ¢ = f(z) dz),

wheref(z) is a meromorphic function. If the point p E C corresponds to the
origin in the z-plane, and if we write

f(z) = z 11g(z), g(O) =!= 0, oo,

then the order 11p( ¢) = J1 of ¢ at p is well defined. The divisor (¢) associated
to ¢ is defined to be

<<P> = LJ1p(</J)p.
peC

In case ¢ is a meromorphic differential, its residue at p is

Resp(cP) =
2n
p Jet>,
-1 y
§1. Divisors and Line Bundles on Curves 3

where y is any curve homotopic to {I z I = B} in a small punctured neighbor-

hood of p. A simple but basic fact is the residue theorem

I Resp(</>) = 0.
peC

This is a straightforward consequence of Stokes' theorem. In fact, suppose

4> has poles at Pt. ... , Pd and let U; be a small parametric disc around P; such
that U; n Ui = 0 if i # j. Setting C* = C- U;
U;, we obtain

I
i
Resp;(</>) =I 2n
i
pr -1 Jou,
4>

= - 1
2nyl=i Jc•
r
d<J>

= 0.
When applied to the logarithmic differential 4> = dflf of a meromorphic
function/, the residue theorem gives

deg((f)) = 0.

If we view f as a holomorphic map

this expresses the well-known topological fact that the degree of the divisor
f - 1(q) is independent of q E IP>\ and agrees with the degree (also called sheet
number) off
A divisor D is said to be effective, and we write D ~ 0, if all points of D
appear with non-negative multiplicity. We shall write D ~ D' to mean
D - D' is effective.
To any divisor Done can attach the sheaf (!)(D) defined by the prescription

r(U, (!)(D))= {merom~rphic functions on U}

that satisfy (f) + D lu ~ 0.

Actually, (!)(D) turns out to be a line bundle, since it is generated, over any
sufficiently small open set, by 1/g, where g is a local defining equation for D.
It is customary to write

!l'(D) = H 0 (C, (!)(D)).

Conversely, given a line bundle L and a non-zero meromorphic section s,

we may define, in complete analogy with the case of meromorphic functions,
4 I. Preliminaries

the divisor D = (s) of s, and division by s yields an isomorphism

L ~(!)(D).

It will be an easy consequence of the Riemann-Roch theorem that any line

bundle on C has a non-zero meromorphic section, thus showing that every
line bundle on Cis of the form (!)(D), up to isomorphism.
The following formal rules are clear from the definitions:

(!)(D) (8) (!)(D') ~ (!)(D + D'),

(!)(D)- 1 ~ (!)(-D).

A basic notion in the study of divisors is the one of linear equivalence. A

divisor D is linearly equivalent to zero, and we write D ~ 0, if

D =(f)

for some meromorphic functionf Two divisors D and D' are linearly equivalent
if

D- D' ~ 0,

and the linear equivalence class of a divisor D is called the divisor class of D
and denoted by [D]. Clearly, two divisors D and D' are linearly equivalent
if and only if there is an isomorphism between (!)(D) and (!)(D').
To each projective manifestation of C there is attached a linear series of
divisors on the curve itself. First of all, given a divisor D, the complete linear
series (or system) ID I is the set of effective divisors linearly equivalent to
D. Given two meromorphic functions f and g, notice that (f) = (g) if and
only if there is a non-zero constant A. such that f = A.g. We then have an
identification

IDI = IP-P(D)

obtained by associating to each non-zero jE £>(D) the divisor (f) + D. A

complete linear series is therefore a projective space. More generally, any
linear subspace of a complete linear series is called a linear series (or system).
A linear series !!fi = IPV, where Vis a vector subspace of £>(D), is said to be a
9d if
deg(D) = d; dim(V) = r + 1.
A gJ is called a pencil, a g~ a net, and a g3 a web. By a base point of a linear
series !!fi we mean a point common to all divisors of !!fi. If there are none we say
§1. Divisors and Line Bundles on Curves 5

that the linear series is base-point1ree. The base locus of !7fi is defined to
be the divisor

where
nP = min{multp(D): DE !!fi}.
As we mentioned before, there is a fundamental relation between linear
series on C and maps of C to projective spaces. This is best expressed in the
language of line bundles. As we already said, given a non-zero holomorphic
section s of a line bundle L, and denoting by D the divisor of s, there is an
isomorphism between L and (!)(D). It is therefore natural to use the symbol
IL I to denote the complete linear series

of all divisors of section of L. Now, given a g:i !7fi = IPV, where Vis a sub-
space of H 0 (C, L), notice that a base point of !7fi is a point of C where all
sections in V vanish. If !7fi is base-point-free, we can then define a map

by
¢~(p) = {s E V: s(p) = 0}.
When, on the other hand, !7fi has a non-empty base locus B, defining, as is
customary
L(- B) = L ® (!)(-B),

we may view Vas a subspace of H 0 (C, L( -B)) with the property that the
corresponding linear series !!fi( -B) is base-point-free; we may then set

It is useful to express ¢~ in homogeneous coordinates on IP'. If we choose

a basis s0 , s 1 , ... , s, of V, we have

The notation means this. Given a point q E C, choose a local coordinate z

centered at q and a local trivialization of L near q. In this trivialization we can
write
s; = zllg;(z),
6 I. Preliminaries

where 11 is the multiplicity of q in B (of course, 11 = 0 if q is not a base point)

and the g;'s are holomorphic functions such that g;(O) #- 0 for at least one
value of i. We then set

[s 0 (p), ... , s,(p)] = [g 0 (z(p)), ... , gr(z(p))].

In view of the local description of¢~ we have isomorphisms

¢;cmw(l)) ~ L( -B),
¢;: H 0 (1P', C91P'r(l)) ~ V.

We may also notice that the image of¢~ is a non-degenerate curve C' in
IP'. ·This means that C' is not contained in any hyperplane. Moreover, it
follows from the above formulas that

d = deg(L) = deg(C') · deg(f) + deg(B),

where f is just ¢~ viewed as a map from C to C'.
If Dis a divisor and Lis a line bundle we shall write c/Jn instead of ¢ 1n 1 and
¢ L instead of ¢ 1LI·
A fundamental example is provided by the so-called rational normal curve
of degree n, which by definition is the image of IP 1 under the embedding

In terms of an affine coordinate z on IP 1, the mapping vn is given by

z ~ [1, z, ... , zn].

The symbol vn is suggestive of the fact that such a map is a special case of a
Veronese map.

§2. The Riemann-Roch and Duality Theorems

A basic problem concerning linear series on curves is to compute the dimen-

sion of !l'(D) and for this the main tool is the Riemann-Roch theorem. To
state it we need a few definitions. Recall that the group of isomorphism
classes of line bundles on C can be identified with H 1 (C, C9D. From the
fundamental exponential sequence

we get the coboundary map

§2. The Riemann-Roch and Duality Theorems 7

If L is a line bundle the integer deg(L) is called the degree of L. It is an easy

consequence of Stokes' theorem that this definition agrees with the one
previously given for divisors, that is

deg((!)(D)) = deg(D).

We also recall that the canonical line bundle Kc is defined to be the dual of
the tangent bundle 0c or, which is the same, the sheaf whose sections over an
open set U are the holomorphic differentials on U. Thus the vector space
H 0 (C, Kc) is the space of holomorphic differentials on C; these are also
called abelian differentials. For brevity we shall often write K and 0 instead
of Kc and 0c. We can now state the

Riemann-Roch Theorem. For any line bundle L on a smooth curve C of genus g

From a cohomological point of view the main ingredient of the proof, in

one way or another, is the duality theorem. Recalling that each class in
H 1 (C, K) is represented in Dolbeault cohomology by a (1, 1)-form, this can
be stated as follows.

Duality Theorem. For any line bundle Lon a smooth curve C the pairing

(, ):H 1 (C,L) x H 0 (C,KL- 1 )--+ IC

given by <e. 11> = Sc e 1\ 11 is a duality.

The Riemann-Roch theorem can in fact be gotten by combining the

duality theorem with the easily proved formula

h 0 (C, L) - h 1 (C, L) = deg(L) - g + 1.

As a first consequence we find that for any point p on C and any large enough
integer n the line bundle L(np) has a non-zero section, thus showing that L
has a non-zero meromorphic section and hence is of the form (!)(D) for suit-
able D. Without losing any information we can therefore state the Riemann-
Roch theorem in terms of divisors.
Historically this is the form in which the theorem was originally stated
(cf., for instance, Enriques-Chisini [1, Vol. III]) and is also the form in which
we will most often use it. A proof of it is sketched in the exercises. It is custom-
ary to write
l(D) =dim 2(D) = h0 (C, (!)(D)),
i(D) = h0 (C, K(- D)),
r(D) = l(D) - 1.
8 I. Preliminaries

Of course, if ID I is non-empty

r(D) = dimiDI.

The Riemann-Roch formula can now be rephrased in either one of the

following two ways:

l(D) - i(D) = deg(D) - g + 1,

r(D) - i(D) = deg(D) - g.

The integer i(D) is called the index of speciality of the divisor D and is equal
to the number of linearly independent abelian differentials w such that

(w):::::: D.

The divisor D is said to be special if it is effective and i(D) > 0. In a sense it

may be said that the entire book is devoted to the theory of special divisors.
The simplest case of the Riemann-Roch formula arises when D is the
trivial divisor, in which case we obtain

This remarkable equality due to Riemann is the first instance of Hodge

theory on an algebraic variety (cf., for instance, Griffiths-Barris [1]).
Next, applying the Riemann-Roch formula to the canonical bundle, we find

deg(K) = 2g - 2.

Since K is the dual of the tangent bundle to C, this formula is nothing but the
Hopf index theorem for compact Riemann surfaces. As a corollary we can
prove the classical Riemann-Hurwitz formula. Let

f: c-+ C'

be a non-constant (and her.ce surjective) holomorphic map between smooth

curves C and C' of genera g and g'. If Dis any divisor on C, we take the inverse
image f*(D) of D to be the divisor on C whose local defining equations
are obtained by composing with f the local defining equations of D. For
any q on C and p = f(q) on C' we choose local coordinates z and w centered
at q and p such thatfhas the standard form w = z•<qJ. Then, for any p on C',

f*(p) = L v(q)q.
qEf-l(p)
§2. The Riemann-Roch and Duality Theorems 9

The degree n of the divisorf*(p) is well known to be independent of p and is

called the degree or sheet number of the map f For general pit is simply the
number of points in the preimage of p. We define the ramification divisor R
on Cby

R = I (v(q) - 1)q.
qEC

The integer v(q) - 1 is called the ramification index off at q. Since, around q,
the mapfis of the form w = zv<ql, for any meromorphic differential¢ on C' we
have

llif*¢) = 11J<ql(¢)v(q) + v(q)- 1.

Thus

{f*(¢)) = !*((¢)) + R.
Counting degrees and recalling that deg Kc = 2g - 2 we obtain the
Riemann-Hurwitz formula

2g - 2 = n(2g' - 2) + deg R.
As a further straightforward application of the Riemann-Roch formula
we may compute l(D) in extreme degree ranges. It suffices to notice that if
d = deg(D) < 0 then l(D) = 0, and similarly i(D) = 0 when d > deg(K) =
2g - 2, to conclude that

0 if d < 0,
{
l(D) = d - g +1 if d > 2g - 2.

For divisors in the intermediate range

O~d~2g-2

the situation is more interesting. It will immediately follow from the geo-
metric version of the Riemann-Roch theorem, to be discussed later, that if d
is in the above degree range and D consists of d general points of C then

r(D) = {0d- g
if d ~ g,
if d?: g.

We may call an effective divisor D, for which the above fails to hold, an
exceptional special divisor.
10 I. Preliminaries

The admissible values (d, r) for a complete g:i on Call lie above the heavy
lines in the diagram below. The exceptional special divisors are those for
which (d, r) is strictly above the heavy lines. It is, of course, understood that
when d > 2g - 2 the dimension r is forced to equal d - g.

,f]-1

We may note in passing that the only g~;! 2 is the canonical series. This
follows, by the Riemann-Roch theorem, from the remark that the only line
bundle of degree zero with non-zero sections is the trivial one.
A useful concept that emphasizes the duality implicit in the Riemann-
Roch formula is the concept of residuation. A divisor D' is said to be residual
to an effective divisor D in case

D' E IK( -D) I.

We note that, conversely, Dis residual to D' and the relations

i(D) = l(D'),
l(D) = i(D')

hold. Using these, and the fact that the degree of the canonical series is
2g - 2, it follows that the Riemann-Roch formula forD is equivalent to that
forD'.
As a first geometric application of the Riemann-Roch theorem we are
going to study the canonical map

and this in turn will provide us with a geometrical interpretation of the

theorem itself.
We first need to recall that a smooth curve C of genus g > 1 is said to be
hyperelliptic if it carries a gi (since g > 0 this has to be base-point-free).
§2. The Riemann-Roch and Duality Theorems II

Equivalently, there should be a degree two map

The corresponding involution of Cis called the hyperelliptic involution. By

the Riemann-Hurwitz formulafhas 2g + 2 ramification points and hence,
being of degree two, has 2g + 2 branch points a 1 , •.. , a 29 + 2 • In other
words, the curve C can be thought of as the Riemann surface of the algebraic
function

A distinguishing feature of such a curve is that one can be totally explicit in

describing the abelian differentials on it. If suffices to notice that, in case p(z)
is a polynomial of degree not greater than g - 1,

dz
p(z)-
w

represents a holomorphic differential on C, and therefore an explicit basis

for H 0 ( C, K) is

_ dz _ dz _ 9 _ 1 dz
w1 - - , w 2 - z -, ... , w 9 - z -.
w w w

Historically the integrals of these differentials, that is the abelian integrals,

were the subject of extensive study and it was only much later that their
investigation led to the very notion of the Riemann surface of an algebraic
curve.
It is apparent that, with the choice of coordinates in IP 9 - 1 = IP H 0 ( C, K)*
given by w 1 , ... , w 9 , the canonical map c/Jx is simply the composition of the
two-sheeted covering f with the Veronese embedding v9 _ 1 which we
described at the end of Section 1 :

Thus c/Jx represents the hyperelliptic curve C as a ramified two-sheeted

covering of a rational normal curve in IP 9 - 1 •
On the other hand we are now going to show that, when C is a non-
hyperelliptic curve ofgenus g > 2 (any curve of genus 2 is hyperelliptic), the
canonical map c/Jx is an embedding. First of all notice that IK I has no base
point. If it had one, call it p, we would have h0 (C, K(- p)) = g and therefore,
12 I. Preliminaries

by the Riemann-Roch theorem, there would be a meromorphic function

having a simple pole at p and no other singularity, implying that g = 0. This
shows, incidentally, that for any p on C, h0 (C, K(- p)) = g - 1. Now notice
that, for any choice of p and q on C, l(p + q) = 1 since there are no g~ 'son C.
By the Riemann-Roch theorem this means that

h0 (C, K( -p- q)) = g- 2.

When p # q this means that there are abelian differentials vanishing at p

but not at q, showing that ¢K is one-to-one. When p = q we find that there
are abelian differentials vanishing simply at p and this exactly means that the
differential of ¢Kat p is not zero.
The image of ¢K, in case Cis non-hyperelliptic, is called a canonical curve.
It is a smooth non-degenerate curve in IP 9 - 1 of degree 2g - 2. Conversely,
any non-degenerate, irreducible curve in IP 9 - 1 of geometric genus g and
degree 2g - 2 is a canonical curve. In fact, we have already observed that
the only g~;! 2 on a genus g curve is the canonical series.
Using the canonical image of a (possibly hyperelliptic) curve we want to
give a geometric interpretation of the index of speciality of an effective divisor
on a smooth curve of genus g > 1 and then use it to restate the Riemann-
Roch theorem in a geometric form. For this we need a notation that will be
very often used throughout this book.

Let C be a smooth curve and let¢: C--+ !Pr be a holomorphic mapping. For
any effective divisor D on C we denote by ¢(D) the intersection of the hyper-
planes H such that either ¢(C) c H or cjJ*(H) ~ D.

If D = p 1 + · · · + Pa and in case¢ is an embedding we shall often write l5

or p 1 , ... , Pa for ¢(D). If we make the further assumption that the P; are
distinct, then l5 is the ordinary linear span in IPr of these points; in general, if
L
we write D = Jlaqa, where the qa are distinct, then Dis the linear span of
the appropriate osculating spaces to Cat the points qa.
With this notation it is clear that

i(D) = g - 1 - dim cPK(D).

Substituting this in the Riemann-Roch formula gives the

Geometric Version of the Riemann-Roch Theorem. For D an effective divisor

of degree d on a smooth curve of genus g > 1

r(D) = d - 1 -dim cPK(D).

When the points of D = p 1 + · · · + Pa are distinct we may restate this as

saying that r(D) is the number of independent linear relations among the
§2. The Riemann-Roch and Duality Theorems 13

points ¢K(P;) on the canonical curve. The geometric version of the Riemann-
Roch theorem implies that the exceptional special divisors are those divisors
whose points (counted with multiplicities) fail to be independent on the
canonical image. This, together with the fact that the canonical image is non-
degenerate, gives, as previously announced, that for a general effective
divisor D of degree d

r(D) = {0 if d ~ g,
d- g if d ~g.

Residuation may be interpreted as follows. If D is a special effective

divisor on C, and D' is residual to D, the complete linear series ID I is the mov-
ing part of the series cut out on C by the hyperplanes through ¢K(D'). In
other words, ID I consists of all divisors DH such that

DH + D' = </Jt(H),

where His a hyperplane through ¢K(D').

In passing we note that canonical curves have the remarkable property
that if they admit one d-secant (d - r - 1)-plane then there are rxf such
planes. For example, if a canonical curve C c IP 3 of genus 4 has one trisecant
line, then there are oo 1 such trisecant lines. In a sense that is difficult to make
precise, the above seems to be the characteristic property of the canonical
curve.
A useful consequence of the geometric version of the Riemann-Roch
theorem is the following. Let C be a hyperelliptic curve of genus g. Let d be
an integer such that 0 ~ d ~ g. Then any complete g'd on C is of the form

rg~ + P1 + · · · + Pa-zn
where no two of the p;'s are conjugate under the hyperelliptic involution. In
fact, the Riemann-Roch theorem implies that any series of this form is
complete, by the well-known fact that any effective divisor of degree fJ ~ g on
a rational normal curve in IP 9 - 1 spans a IP 15 - 1 • (When the points of the
divisor are distinct, this amounts to the non-vanishing of the Vandermonde
determinant.) On the other hand, any degree d effective divisor on C belongs
to one of these series.
To conclude this section we want to deduce from the duality theorem a
constructive version of the Riemann-Roch theorem for effective divisors.
Let D be such a divisor; the Mittag-Leffler sequence attached to Dis

Setting
14 I. Preliminaries

where n" > 0 and the q" are distinct, and choosing a local coordinate z"
centered at p", a section of mv(D) can be written as

where
-1

'1a = L akz~
k= -n.x

is a Laurent tail (or principal part) centered at q". The Mittag-Leffler problem
consists in deciding which collections of Laurent tails come from a global
merom orphic function on C. This is equivalent to solving the equation DIJ = 0,
where iJ is the co boundary map

gotten from the Mittag-Leffler sequence. In view of the duality theorem

this amounts to solving the system of linear equations

(DIJ,W)=O forevery wEH 0 (C,K).

To make this pairing computable we will show that

(2.1)

By definition

where¢ is a Dolbeault representative ofthe cup product DIJ · w. To compute¢

choose, for each ll(, a small disc U" centered at q" and set

'1o = 0.

We may find coo (1, D)-forms~" on U", il( = 0, ... , m, such that

then ¢ is the global (1, 1) form defined by

§3. Abel's Theorem 15

Denoting by T;.(e) a disc of radius e centered at qa, we have

<JtJ, w) = {¢
=lim r
e--+0 Jc-(UT«(<))
¢

=-lim I l
oT«(r.)
~0

= lim I f ~a - ~o
JiJT«(<)

=lim I r
JiJT«(<)
IJaW

= 2nj=l I ResqJIJaW),
a

as desired. Summarizing we can say that a collection of Laurent tails '1 comes
from a global meromorphic function f E !l'(D) if and only if
L Resq.(IJaW) = 0 for every wE H 0 (C, K).
a

As a system of linear equations for '1 this is a system in d unknowns con-

sisting of g equations, among which those corresponding to differentials w
in H 0 (C, K(- D)) are trivial. A moment of reflection will convince the reader
that, in fact, there are exactly g - i(D) independent equations in the above
system. Since two meromorphic functions having the same principal parts
differ by a constant we recover the Riemann-Roch theorem.
Finally, formula (2.1), when applied to the divisor consisting of a single
point p on C, tells us that the image of

represents the point ¢K(P) on the canonical curve, thus providing another
description of the canonical map.

§3. Abel's Theorem

We have already remarked on the central role played by the abelian dif-
ferentials in the study of linear systems of divisors on an algebraic curve.
These differentials also define cohomology classes, as well as systems of
16 I. Preliminaries

differential equations on the curve, and we shall now recount the relevant
definitions and facts in this development.
Let w and w' be holomorphic differentials on a smooth curve C. Locally
w = f(z) dz, wheref(z) is a holomorphic function, and consequently

w A w' = 0,
j=lw " w= lf(zW(j=ldz" dz),

which implies

few w' = 0,
A

j=1 L w " w> o if w "" o.

Since dw = 0, the form w defines a cohomology class in the deRham co-
homology group HbR(C) ~ H 1 (C, C). We shall now define the period matrix
of the curve, and translate the above relations into the famous Riemann
bilinear relations. For this we let w 1, .•• , w 9 be a basis of H 0 (C, K), and
y 1 , ... , y29 a basis for H 1 (C, £:). The period matrix Q is then the g x 2g
matrix whose jth column nj is given by

Using the standard fact that cup product in cohomology is Poincare dual to
intersection in homology we obtain the Riemann bilinear relations

nQtn = o,
j=lnQtn > o,
where Q is an integral skew-symmetric matrix whose transposed inverse is
equal to the intersection matrix (yi, yj). It is an elementary fact that we may
always choose the basis Yt. ... , y29 to be symplectic, meaning that the inter-
s"'ction matrix is of the form

The Riemann bilinear relations that one encounters in the classical literature
are written in terms of such a basis.
A well-known consequence of these relations is that the 2g column
vectors of the period matrix generate a lattice A in C9 , so that the quotient
§3. Abel's Theorem 17

C9/A is a complex torus. This torus is called the Jacobian variety of C and
denoted by the symbol J(C).
To say that A is a lattice is equivalent to saying that integration over
cycles defines an injection

This gives rise to a coordinate-free descripton of the Jacobian variety as

The Riemann bilinear relations also imply that the Jacobian variety of C
has a principal polarization. This aspect, which is unimportant here, will be
dealt with extensively in the next section.
The geometry of the curve C interacts with the geometry of J(C) via a
basic group homomorphism

(3.1) u: Div(C) ~ J(C),

where, as is customary, Div(C) denotes the group of all divisors on C. This

homomorphism depends on the choice of a base point p0 . Having fixed this,
we first define
u:C~J(C)

by setting

u(p) = (JP w 1 , ... , JP w 9).

Po Po

The right-hand side of this equality depends on the choice of a path from p0
top, but the ambiguity involved in this choice disappears when we pass to the
quotient modulo A. The general mapping (3.1) is then obtained by linearity:
explicitly, for

we set
u(D) = L u(p;) - L u(qi).
i j

The individual components

of u(D) are frequently called abelian sums.

18 I. Preliminaries

As we already remarked, the definition of u depends on the choice of a

base point. However, if we denote by Divd( C) the set of all divisors of degree
d on C, the restriction of u to the subgroup Div 0 ( C) is independent of p0 , as
is apparent from the definitions.
The version of the basic mapping u which we will find most useful in the
sequel involves the symmetric products of the curve. The d-fold symmetric
product of Cis, by definition, the set of effective divisors of degree don C, and
is denoted by Cd. Clearly, Cd is the quotient of the ordinary d-fold product
Cd by the natural action ofthe symmetric group Sd; as such, Cd is a projective
variety. Moreover, the map
(3.2) u: cd-+ J(C)

is holomorphic, since its composition with the quotient map from cd to cd

is both holomorphic and invariant.
Let U be a coordinate open subset of C. We take, as coordinates on the
d-fold product of U, the compositions of the projections onto the factors
with the coordinate on U. Recall then that any Sd-invariant holomorphic
function on ud depends holomorphically on the d elementary symmetric
functions of the local coordinates. A consequence is that Cd is smooth.
The main purpose of this section is to discuss Abel's theorem, which is the
basic result relating the algebro-geometric concept of linear equivalence and
the transcendental concept of abelian sum.

Abel's Theorem. Let D and D' be effective divisors of degree d on a smooth

curve C. Then D is linearly equivalent to D' if and only if u(D) = u(D').

Abel's theorem may be rephrased as saying that the (set-theoretic) fibres

of the mapping u from Cd to J(C) are complete linear series; more precisely,
for each DE Ca, we have

Another useful interpretation of Abel's theorem involves the Picard

group

Pic(C) = H 1 (C, (9*).

In more detail, we shall denote by Picd(C) the subset of Pic( C) consisting of

isomorphism classes of degree d line bundles. As we already observed in
Section 1, Picd(C) is just the quotient of Divd(C) modulo linear equivalence.
The exponential sequence provides an identification

thus exibiting Pic 0 (C) as a complex torus.

§3. Abel's Theorem 19

Abel's theorem then states, first of all, that for each d there is a factorization

and, secondly, that ¢ is injective.

Indeed, even more is true, namely that ¢ is a bijection. The statement that
¢is surjective is equivalent to the Jacobi inversion theorem, stating that if g
is the genus of C,

u: C 9 -> J(C)

is onto. To see this, since C 9 and J(C) have the same dimension, it suffices to
show that u is generically finite. By Abel's theorem, it then suffices to produce
a divisor DE C9 such that r(D) = 0. Since the image of C under the canonical
mapping is non-degenerate, the geometric version of the Riemann-Roch
theorem tells us that a general D will do.
As for the proof of Abel's theorem, here we shall only sketch the easy part,
namely that, for any DE Divd(C), u is constant on ID 1. Identify J( C) with
C 9/A and denote by w1 , ... , w9 linear coordinates on C9 ; then dwa descends
to a holomorphic differential on J(C). Since ID I is a projective space, u*(dwa)
vanishes on ID I for every IX. This proves that u is constant on ID 1.
An argument based on the "easy part" of Abel's theorem provides one of
the classical proofs of the Riemann-Roch theorem, which we now sketch.
Let D = p 1 + · · · + Pd be an effective divisor of degree d; for simplicity
of notation we shall assume that D consists of d distinct points. We set
r = r(D) and let A. I> ••• , A., be local coordinates in the projective space ID I
nearD;weshallwriteD-< = p 1 (A.) + · · · + piA.)todenotethepointofiDicor-
responding to the value A. of the coordinates. Further, for each i we let z; be a
local coordinate on C around P; and write

z;(p;(A.)) = z;(A.),
IX= 1, ... , g.

Since the points p;(A.) move with r degrees of freedom, for general A. we have
that

Now, the classical way of writing the easy part of Abel's theorem is

L
I
J
z;(A)

PO
fa;(z;) dz; =constant (modulo periods),
20 I. Preliminaries

and differentiating with respect to A., we obtain

OZ·
I !~;(z;(A.)) aA.' CA.) = o.
' v

This relation, together with the assertion on the rank of the matrix oz;/oA.v(A.),
tells us that

(3.3) dim <h(D) ~ d - 1 - r.

To show that, in this relation, we have an equality, we proceed by contradic-

tion. Suppose (3.3) is a strict inequality. Then, given D' E IK(- D) I, we would
have

dim ID' I > g - d +r- 1.

Now, applying (3.3) to D' we would get

dim </JK(D') < 2g - 2 - d - 1 - (g - d +r- 1)

= g- 2- r.

But then, again by residuation, we would deduce that dim ID I > r, which
is absurd. We therefore have equality in (3.3), proving the Riemann-Roch
theorem in its geometric version.

§4. Abelian Varieties and the Theta Function

In this section we shall briefly discuss principally polarized abelian varieties
and Riemann's theta function. The reason for doing this is that the theta
divisor on the Jacobian of a curve C very closely reflects the geometric prop-
erties of the curve itself; many beautiful and classical results such as Riemann's
theorem, the Riemann singularity theorem, and Torelli's theorem deal with
this interplay.
Given a general complex torus

then, upon making the natural identification

we may view a skew-symmetric bilinear form

E:A X A-+7l..
§4. Abelian Varieties and the Theta Function 21

as an element

~ E Hom( A2 H 1(A, £:), Z) ~ H 2 (A, £:).

Now we choose generators eb ... , e 29 for A and, regarding each e; as a

column vector, we define the period matrix to be

This terminology is justified by the fact that

where yi E H 1 (A, Z) corresponds to ei and u 1, ... , u9 are the coordinates on

IC 9 • Let Q be the 2g x 2g matrix representing E with respect to the basis
e 1 , ... , e29 • To say that ~ is a (1, 1)-form is equivalent to saying that Q and
Q satisfy the first Riemann bilinear relation

On the other hand, the second Riemann bilinear relation

is equivalent to the assertion that~ is a positive (1, 1)-form. Now any such~
is the first Chern class of an ample line bundle L on A; moreover, this line
bundle is uniquely determined up to translation (see the bibliographical
notes at the end of the chapter). The Riemann-Roch theorem for L states that

A principally polarized abelian variety is a pair consisting of a complex torus

A together with the Chern class ~ of an ample line bundle on A such that ·

I_
g!
f ~g =
A
1.

This simply means that the skew-symmetric matrix Q representing ~ is

unimodular. Thus a principal polarization ~ on A is the fundamental class of
a divisor
0cA,
22 I. Preliminaries

which is unique up to translation and is called the theta divisor. We may

observe, in passing, that the Riemann bilinear relations satisfied by the period
matrix of a smooth curve C mean that the intersection pairing defines a
canonical principal polarization on J(C).
We will now explicitly construct the theta divisor of a principally polarized
abelian variety. In particular, this construction applies to the Jacobian variety
of a smooth curve.
To begin with, it is a standard linear algebra lemma that, since the pairing

E:A X A-+71.

is unimodular, we may choose generators for A such that the matrix of E has
the standard form

Q= ( 0
-1 9
lg)
0 .

Writing the period matrix Q in block form

Q = (B, B'),

where B and B' are g x g matrices, the Riemann bilinear relations imply that
Band B' are both non-singular. If we change our basis for IC9 by B- 1, then n
becomes

and the Riemann bilinear relations are now, respectively

Z= 1Z,

ImZ > 0.

We shall refer to such a period matrix as a normalized period matrix. The

Siegel generalized upper half plane is defined to be the set of all g x g sym-
metric matrices with positive definite imaginary part, and is denoted by the
symbol Jlfg. We also let .919 denote the set of isomorphism classes of principally
polarized abelian varieties of dimension g. Clearly there is a surjective map

For g = 1 this is the well-known parametrization of smooth elliptic curves by

the upper half plane in IC. In general there is an identification
§4. Abelian Varieties and the Theta Function 23

where

ro ~ Sp(2g, 71.)/ ± 120

is the Siegel modular group operating on ~ by

T(Z) = (AZ + B)(CZ + D)-I,

where

T = (~ ~) E Sp(2g, 71.).
On the product C 0 x Jlfy we define Riemann's theta function

e(u, Z) = L e"J=1<m,Zm)+21tj=1(m,u>,
me7L9

where the bracket denotes the standard duality pairing (m, v) = m«v«La
on C 0 • Since Im Z > 0, the above series converges normally, and hence
uniformly on compact sets, on C0 x ~ to a holomorphic function. One
easily checks that the theta function satisfies the functional equations

O(u + ea, Z) = O(u, Z),

O(u + e0 +a• Z) = e- 2 "J=1<u~+z~~>e(u, Z),

where Z = (Zap) and eh ... , e20 are the 2g column vectors in the period
matrix Q. It follows from these equations that for each fixed Z E ~' the
divisor (0) of Riemann's theta function is invariant under translation by
elements of the lattice A generated by eh ... , e20 , and therefore induces a
divisor E> on A = CO/A. It is clear from its definition that the theta function is
even in u and therefore E> is symmetric, i.e.

-e =e.
We now want to verify that E> is the principal polarization of A. This amounts
to showing that the fundamental class of E> is represented by the differential
form
g

~ = L dxl%
a=l
1\ dxo+«•

where x 1 , •.• , x 29 are the real linear coordinates on C 9 corresponding to the

basis e 1 , ... , e20 • Since this is a purely topological statement and since ~
24 I. Preliminaries

is connected (in fact, convex), it is enough to give the proof for a particular
Z in .Yfg. It will be convenient to choose Z to be a diagonal matrix

Z= (
r1 •
..
0) ' Im ri > 0.
0 r9

Geometrically Z corresponds to a product of g elliptic curves. It is apparent

from the definition of the theta function that for such a Z
g

8(u, Z) = fl 8(ui, rJ
i= 1

We are thus reduced to the one variable case, where all we have to show is that
the divisor of (J consists of a single point on the elliptic curve. In fact, using the
functional equation for e, we find

deg(8) =
2n
p Jd
-1
log (J

=
2n
p (t
-1 Jo
d log (J + r d log e)
Jt+t

+ 1r----1
2n....; -1
(Jt
1
+td log (J + fo
1 +t
d log (J )
= 1,
where the first integral is taken around the fundamental parallelogram

Im u

'
······························.:.::.:··1 +'

0 Reu

We conclude this section by observing that term-wise differentiation of the

theta series shows that Riemann's theta function is a solution of the heat
equation
iJ2(J iJ(J
~ (u, Z) = 2nJ-l (1 + ~~p) "'Z ,
uU~ uUp u ~fJ
§5. Poincare's Formula and Riemann's Theorem 25

where Z = (Zap) and bap is the Kronecker symbol. This equation, as we shall
see in Chapter VI, provides a tight link between the intrinsic geometry and the
infinitesimal deformations of an algebraic curve.

§5. Poincare's Formula and Riemann's Theorem

We now return to the abelian sum mapping

u: Ca ~ J(C)

defined in Section 3. We shall denote by Wa(C) the image of Ca under u.

As usual we let E> be the theta divisor on J(C). A natural problem is to try
and relate the subvarieties E> and ~(C). We shall use the symbols 8, wa,
and ca to denote the fundamental classes of E>, Wa(C), and Ca, respectively.
The fundamental relation between e and Wa is

Poincare's Formula. When 0::::;; d::::;; g,

Proof If we can prove the right-hand side equality it follows, in particular,

that ~(C) is d-dimensional (this, by the way, is just the content of the
Jacobi inversion theorem). Therefore, by Abel's theorem, u is generically
1-1, which proves the left-hand side equality as well. Clearly d!u*(ca) is the
image of the fundamental class of the d-fold cartesian product ca via the
mapping

v: ca ~ J(C)

obtained by composing u with the d!-sheeted covering of Ca by ca.

We now choose a symplectic basis 1'1> .•. , y29 of H 1 (C, Z) =A, think of
J( C) as U /A, and let x 1o ••• , x 29 be the real coordinates on U corresponding
to the basis y 1 , ... , y29 • Denoting by

I = (i 1, ... , i2d),
the exterior monomials that generate

we have to show that

(5.1) d\ • f
Cd
v*(dxl) =(
g
~ d)l· f
J(C)
eg-a A dx 1
26 I. Preliminaries

for any index set I. Since

u*(dx;) is Poincare dual to 'Y;· Therefore

L u*(dxa 1\ dxg+a) = - L u*(dx 9 +a 1\ dxa)

= (ya, 'Yo+a) = 1,
and

Ju*(dx; 1\ dxj) = (y;, Yi) = 0 if li- jl =F g.

Thus, by Fubini's theorem, the left-hand side of (5.1) can be non-zero only
when dx 1 is a product of monomials

The same is true for the right-hand side, since the intersection matrix is in
standard form, i.e.

For every index set B = (/3~> ... , f3a), where 1 :::;; {3 1 < · · · < f3a :::;; g, we set

so that

eo-d = (g - d)! I I'JA,

IAI=g-d

and therefore

r (}g-d 1\ I'JB = (g - d)!.

JJ(C)

On the other hand, denoting by U;_ the composition of the projection of cd

on the A.th factor with u: C -4 J(C), we have

v*(dx;) = I;_ uf(dx;),

§5. Poincare's Formula and Riemann's Theorem 27

and

= d!,
where the last sum is over all permutations of {1, ... , d}. Q.E.D.

The first and most important corollary of Poincare's formula is

Riemann's Theorem. For a suitable point K of J(C)

l¥g-1(C) = 0- K.

By Poincare's formula, Wg-1 = e. Since, as we already observed, 0 is

determined by 0 up to translation, Riemann's theorem follows.
The point K E J(C) is called Riemann's constant. Of course, it depends on
the base point introduced to define the map u. A simple residue computation
(cf. Lewittes [1], for example) shows that

-2K = u(K),

where K is a canonical divisor. It is often convenient to define a new map

n: C9 _ 1 --+ 0 c J(C)

by setting n(D) = u(D) + K. This has the property that

- n(D) = n(K - D),

which gives a geometrical content to the symmetry property of the theta

divisor.
When g = 2 the map n provides an isomorphism between C and the
theta divisor. This is the first instance of Torelli's theorem, which states that
the curve C can be recovered from the pair (J(C), 0), i.e., from the knowledge
of J(C) as a principally polarized abelian variety. Torelli's theorem will be
proved in Chapter VI.
In the course of the proof of Poincare's formula we have remarked that
part of the statement really boils down to the Jacobi inversion theorem.
In fact, using Poincare's formula, one may describe Jacobi inversion ex-
plicitly as follows (see Lewittes [1] for details). Consider the map

u: C--+ J(C).
28 I. Preliminaries

For any A E J(C)denote by e,l thetranslateofthethetadivisor by A. Poincare's

formula implies that, when u(C) is not contained in 8-l, the pull-back u*(8,l)
is an effective divisor of degree g. It can be proved that the rational map

defined by

1/J(A.) = u*(8H,.),

where K is Riemann's constant, is an inverse of

u: C9 --+ J(C).

§6. A Few Words About Moduli

In this work we shall deal primarily not with a fixed curve, but with one
which is allowed to "vary with moduli." By definition, the moduli space A 9 of
curves of genus g is the set of isomorphism classes of smooth, genus g curves.
For genus one, it is well known that any smooth elliptic curve has a plane
model

Denoting this curve by C we set

j(C) = 1728g~/(g~ - 27gD.

It is classical that j( C) depends only on the isomorphism class of C and

induces a 1-1 map

j: Al--+ IC.

The set A 1 is thus endowed with a complex structure. That this is true for
higher genera as well was long taken for granted, and Riemann already
indicated how to compute the dimension of A 9 , which he showed to be
3g - 3 (when g > 1, of course). In a transcendental setting all this can be
rigorously established via Teichmiiller theory.
It was only much later that Baily showed that A g has a natural structure of
quasi-projective normal variety of dimension 3g - 3. The most natural way
of proving this is probably by means of geometric invariant theory: this is
due to Mumford. It turns out that A 9 is also irreducible. This was already
observed by Klein and follows from results of Liiroth and Clebsch.
§6. A Few Words About Moduli 29

Following a classical idea, Mumford-Deligne and Mayer proposed a

natural compactification Jly of .A9 • This consists of the isomorphism
classes of the so-called stable curves of (arithmetic) genus g. A stable curve
is a curve whose only singularities are nodes and whose smooth rational
components contain at least three singular points of the curve. It was later
proved by Knudsen that J!t9 is in fact a projective variety. The complement of
.A9 in .A9 turns out to be a codimension one subvariety -1 which is the union

of [g/2] + 1 irreducible subvarieties. The general point of -1 0 is an irreducible

stable curve of genus g with one node, while the general point of di, i > 0,
consists of a smooth curve of genus i joined at one point to a smooth curve of
genus g- i.

~ ... ~
·~

In general, both .A9 and .il9 are singular. All the singularities arise from
curves which have a non-trivial group of automorphisms, and we shall
denote by .il~ the smooth open subset of .il9 consisting of automorphism-
free curves. Over .A~ there is a universal family

with the property that, for each p E .A~, the fiber f- 1 (p) is a stable curve
whose isomorphism class is p, and, most important, that each flat family of
stable, automorphism-free, genus g curves parametrized by a variety S
is induced by a unique morphism from S to .A~. No such object can
exist on .il9 (or on .A9 ). However, the structure of .A9 is natural in the sense
that, for each flat family

g: X--+ S
30 I. Preliminaries

of genus g stable curves, there is assigned a morphism from S to Jig which is

compatible with base change and is such that, when Sis a point, the image of S
is the isomorphism class of X.
It was shown by Severi that .itg is a unirational variety when g :-=:;; 10,
and it was for a long time an open problem whether this held true for higher
genera as well. Sernesi showed that .it 12 is in fact unirational. The problem
was finally given a negative answer by Mumford and Harris who proved that
.itg is of general type whenever g is sufficiently large.
In this volume we shall make free use of those few basic properties of .itg
that we will occasionally need. A more complete discussion of moduli space
will appear in the first chapters of the second volume. In the same volume
proofs of the irreducibility of .itg, of Severi's unirationality result, and of the
theorem of Harris and Mumford will also be given.

Bibliographical Notes
We give a few sources for background reading, from various points of view,
in the theory of algebraic curves.
For a classical Riemann surface viewpoint we suggest Farkas and
Kra [1], Siegel [1], Springer [1], and Weyl [1].
For a modern Riemann surface point of view; i.e., the study of one-
dim~nsional compact, complex manifolds, the reader may consult Gunning
[1].
Algebraic curves are treated in Coolidge [1], Enriques-Chisini [1],
Fulton [1], Severi [2], and Walker [1].
The theory of curves and abelian varieties is discussed in Fay [1],
Lang [1], Mumford [1], [3], [5], and [6].
Curves, as part of general algebraic geometry, are treated in Griffiths-
Harris [1], Hartshorne [2], and Shafarevich [1], among many other
places.
Deformation theory and moduli appear in Kodaira-Spencer [1] (local
theory) and Mumford [3] and [4].
The quasi-projectivity of Ag is proved in Baily [1]. For the invariant-
theoretic approach we refer to Mumford-Fogarty [1]. Klein's remark
on the irreducibility of Ag appears in Klein [1, no. 19]: his argument is
based on earlier work by Liiroth [1] and Clebsch [1]. The com-
pactification of Ag by stable curves was introduced in Deligne-Mumford
[1]. The projectivity of Jig was established by Knudsen [1]. The
main results concerning the unirationality (or non-rationality) of Ag can
be found in Severi [2], Sernesi [1], Harris-Mumford [1], and
Harris [3].
A more complete discussion of the literature on the moduli space of
curves will be found in the bibliographical notes to the second volume.
Exercises 31

Exercises

A. Elementary Exercises on Plane Curves

Algebraic curves are frequently given by irreducible equations

f(x,y) = 0.

By this we shall always mean the irreducible affine algebraic curve

given by the complex solutions to the above equation. Associated to r 0 are the following
objects:

(i) The projective curve

given by the zeros

of the homogeneous form

d = degf,

associated to f. We think of r as adding to r 0 the points on the line (X 0 = 0)

at infinity.
(ii) The normalization

4>: c __. r c IP' 2

ofr.

Sometimes we refer to C itself as the normalization of r. It is a compact Riemann

surface that classically was thought of as "the Riemann surface associated to the algebraic
function y(x) where f(x, y(x)) = 0." What this means is that, via the projection r 0 --+ C
on the x-axis, we may picture r 0 as a branched covering of the complex x-plane. Then
C is realized as a branched covering of the x-sphere IP' 1 = C u { oo} that agrees with
r 0 --+ C outside the singularities of r 0 and the points lying over x = oo. The con-
struction of C --+ IP' 1 that is pertinent to the following exercises is given, e.g., in Weyl
[1], we shall refer to it as completing r 0 to a compact Riemann surface.
32 I. Preliminaries

We consider the following affine algebraic curves:

(a) y2 = x3 + 1.
(b) y3 + x3 = 1.
(c) x2y2 + x2 + y2 = 0.
(d) y2 = x6 - 1.
(e) y2 = xs - 1.
(f) l = x 5 - 1.
(g) y3 = (x2 + 1)2(x3 - 1).
(h) y 7 = x 2 (x- 1).
(i) y + x 3 + xy 3 = 0.

A-1. Show that the affine curves (a), (b), (d), (e), (f), and (i) are smooth. Show that (a)
may be completed to a compact Riemann surface by adding one point to the
affine curve, (b) by adding three points, (d) by adding two points, (e) and (f) by
adding,one point, and (i) by adding two.

A-2. Show that the curves (c) and (h) are singular exactly at the origin, while (g) is
singular at (±i, 0) (i = j=l). Show that in the normalizations of (g) and (h)
there is exactly one point lying over each singular point, while for (c) there are
two over each singular point.

A-3. Let r = C be the compact Riemann surface assoc1.-ced to the curve in (a) above.
The curve C can be realized as a branched coverin~;, n: C-> IP' 1 of the x-sphere.
Set, respectively

P1, P2, P3 = ( -1, 0), ( -w, 0), ( -w 2, 0),

q!, q2 = (0, 1), (0, -1),

r = n- 1 (oo),

s 1 , s2 = (2, 3), (2, - 3).

(i) Find the branch points of n: C ---> IP' 1 and show that the genus of C is
equal to one.
(ii) Establish the linear equivalences

2p 1 - 2p 2 - 2p 3 - q1 + q2 - 2r - s 1 + s2 ,
P1 + P2 + P3 - 3r,

(iii) Determine .!l'(p 1 + q 1 ) and describe the complete linear system jp 1 + q 1 j.

(iv) Find the points t E C such that, respectively,

t - P1 + q2- r,
t - 2s 1 - r.
Exercises 33

A-4. Let r = C be the compact Riemann surface associated to the curve in (b) above.
The curve C can be realized as a branched covering n: C -> IP' 1 of the x-sphere.
(i) Find the branch points and ramification points of n: C -> IP' 1 and show that
the genus of C is equal to one.
(ii) For any two points p, q E C find the complete linear series IP + ql. (Hint:
Draw lines.)
(iii) Find a map ry: C-> IP' 1 of degree two such that ry((l, 0)) = ry((O, 1)), and
determine the branch points of rJ.
(iv) Using A-3(i) and (ii) show that the curves in (a) and (b) are isomorphic,
and find an explicit isomorphism.

A-5. Let C be the compact Riemann surface associated to the curve in (d) above,
realized as a branched covering n: C ..... IP' 1 of the x-sphere. We shall denote
also by (x, y) the point on C corresponding to a smooth point (x, y) E r 0 . Set

P + q = n-l(oo),

s ~> s2 = (0, i), (0, - i), respectively.

(i) Show that

p +q ~ 2r.,
5
I'·~ 3p + 3q.
a=O

(ii) Determine the complete linear system lr0 + r2 + r4 1. Find the divisor D
on C such that D + r 1 ~ r0 + r2 + r 4 .
(iii) ForD = p + q + r0 + r 3 find !l'(D). Describe the map

given by IDI, find the equation of c/>n(C), and determine the singularities
of c/>n(C).
(iv) Determine the necessary and sufficient conditions that the Laurent tails

f. = a. at r., C( = 0, ... ,5,

y

at s2 ,

should be the principal parts of a global meromorphic function on C.

34 I. Preliminaries

A-6. Let C be the compact Riemann surface associated to the curve in (f) above,
realized as a branched covering n: C--> IP' 1 over the x-sphere. Set

r. = c,·, 0),
(i) Find the ramification divisor of n and compute the genus of C.
(ii) Establish the linear equivalences

3p ~ 3r.,
4
L r. ~ 5p.
rX=O

(iii) Determine the space H 0 (C, K) ofholomorphic differentials on C.

(Hint: Find !i'(D) where D = (dx).)
(iv) Describe the canonical map

and determine the equations of its image.

(v) Using part (iii) show that, forD = L~=o r.
h0 (C, K( -D))= 1.

From the Riemann-Roch theorem conclude that r(D) = 2, and by using

residuation determine the complete linear system ID 1.
(vi) ForE= 3p show that lEI is the unique g~ on C, and that 12EI = IKI.

A-7. Let C be the compact Riemann surface associated to the curve in (g) above with
map n: C--> IP' 1 onto the x-sphere. Set

p = n-l(oo),

q 1 , q2 =points on C lying over (i, 0), ( -i, 0) on r,

r. = (w", 0),

(i) Find the ramification divisor of nand show that g(C) = 4.

(ii) Show that

2 2
I r. +I q. +sp.
0 1

(iii) Determine H 0 (C, K), describe the canonical map

and find the equations of cfJK(C).

Exercises 35

(iv) For
E = 3p,
{
D = K- E,

show that
EfD.

From the Riemann-Roch theorem, conclude that r(D) = 1.

(v) Using residuation, find the pencil IDI and describe the corresponding
map¢n·
(vi) Using part (vi) of A-6 and part (iv) above, show that the normalizations of
the curves in (f) and (g) are not isomorphic.

A-8. Let r = C be the compact Riemann surface associated to the curve in (i) above,
and denote by nx, ny: C--> [pi the branched coverings obtained by projecting
r 0 on the x- andy-axes, respectively. Compute the genus g(C) by:
(i) the Riemann-Hurwitz formula applied to nx;
(ii) the Riemann-Hurwitz formula applied tony;
(iii) the genus formula for C c IP' 2 (you must show that C is smooth).

Note. The formula for the genus of a smooth plane curve of degree d IS
g = (d- l)(d - 2)/2; this is discussed in Appendix A below.

A-9. Let C be the compact Riemann surface associated to the curve in (h) above.
(i) Compute the genus g(C).
(ii) Determine H 0 (C, K), and describe the canonical map ¢K and its image.
(iii) Show that the normalizations of the curves in (h) and (i) above are not
isomorphic.

A-10. Show that the curve r 0 in (c) above is rational by

(i) the Riemann-Hurwitz formula applied to C--> [pi;
(ii) giving an explicit rational parametrization.
(Hint: Look at the conics through the three nodes of r.)

B. Projections

The following sequence of exercises concerns projections, and for this we will use the
notations:
L --> C is a line bundle on a smooth curve;
V c H 0 (C, L) is a linear subspace and f» = IP'V the corresponding linear system
with associated map

for any point x E IP'V* we set

f»x = IP'(Ann x) c f»,
36 I. Preliminaries

and for p E C we set

~(- p) = { D - p: D E~ and D ~ p}.

We assume that ~ has no base points.

B-1. Show that p is a base point of~ <I><Pl and

~( -p) = ~<f>(p)( -p).

(This is just to get familiar with the notation.)

B-2. Recall that the projection

from x E lfl>V* to a hyperplane H not containing xis defined by

nx(y) = xyn H,

where xy is the line joining x to y. Show that

(this is true even when x E r/J~'~(C)).

B-3. Let r = r/J!B(C), X E lfl>V*, and define the multiplicity multx(r) of rat X to be the
degree of the base locus of ~x· Assuming that nx: r-+ nx{r) is generically one-to-
one, show that

deg nx(r) = deg(r) - multx{r).

B-4. Show that~= IK(p + q)l has the property that r/J~'~ is not an embedding but
rp~'~x is if Cis non-hyperelliptic and xis the singular point of r/J~'~(C).

B-5. Let C c lfl>' (r ~ 3) be a smooth curve. Show that a general projection

maps C biregularly to a smooth curve.

B-6. Let C c lfl> 3 be a non-degenerate smooth curve. Show that a general projection
with center p E lfl> 3

n: lfl>3 _ {p}-+ lfl>z

maps C birationally to a plane curve having ordinary double points as its only
singularities.
(Hint: Show that the set of trisecant lines to C and the set of tangent lines to C
describe two surfaces SandS'. Then take p ¢SuS'.)
Exercises 37

Note. It may be reassuring to the reader to know that the image of a complex analytic
variety under a proper holomorphic map is again a complex analytic variety of the
appropriate dimension. (Remmert's proper mapping theorem, cf. Gunning-Rossi
[1].)

C. Ramification and Pliicker Formulas

In the following sequences of exercises, we will describe the notion of ramification of
a map C...,. IJl>', or, more generally, of a linear series on C. For the following, then, C
will be a smooth curve of genus g, La line bundle of degree d, V c H 0 (C, L) an (r + i)-
dimensional vector space of sections, and p E C a point.

C-1. Show that the set

l:p,V = {ordP u}ueV-{OJ

consists of exactly (r + 1) non-negative integers.

The sequence IX 0 ~ IX 1 ~ • • • ~ IX, defined by

is called the ramification sequence of Vat p; we say that pis an inflectionary point
for V if (1X 0 , ..• , IX,) =1- (0, ... , 0). We shall sometimes write rti(p, V) for !Xi defined
above.

C-2. Show that only finitely many points of C are infl.ectionary points for a given
series V.

C-3. Show that a point p E Cis a Weierstrass point (cf. Exercises E-1 to E-13) if and
only if it is an infl.ectionary point for the canonical series V = H 0 ( C, K), and that
the weight of p is
w(p) = L !Xi.
C-4. If cJ>: C...,. IJl>' is any map, given locally by a vector-valued function

c/>(z) = [v(z)],
we may define a map

cJ><kl: C...,. G(k + 1, r + 1) c.. IJl>Ak+ 11C'+ 1

by setting
cJ><kl(z) = [v(z) 1\ v'(z) 1\ · · · 1\ v<kl(z)].
Show that:
(a) the vector v(z) " · · · 1\ v<kl(z) does not vanish identically, and hence
cJ><•l(z) is defined for all z;
(b) cJ><kl(z) is independent of the choice of function v(z); and
(c) cJ><kl(z) is independent of the choice of local coordinate z on C.
The map cJ><kl is called the kth associated map of ¢; the k-plane A c IJl>' given
by the image cp<kl(p) is called the osculating k-plane to ¢(C) at ¢(p).
38 I. Preliminaries

C-5. If ¢ = ¢v for V a linear system as above, show that the osculating k-plane at
¢(p) is the unique k-plane having maximal order of contact with ¢(C) at ¢(p),
and that this order of contact is k + cxk.

C-6. Show that the kth associated map cp<kJ is an immersion at p if and only if cxk+ 1 = cxk;
and that, in general, the differential dcp<kJ vanishes to order exactly cxk+ 1 - cxk
at p.

C-7. In terms of the identifications

G(k + 1, V) = G(r - k, V*),

prove the duality

and hence

C-8. In terms of the identification of the tangent space to the Grassmannian G(k, V)
at A with Hom(A, VjA), show that the image of the differential dcp<kJ at each
uninflected point has rank 1, with kernel ¢<k- 11(p) and image ¢(k+ 11(p). Using
this, show that ¢ is determined by cp<kJ for any k.

C-9. Using the preceding exercise, show that in general a map t/J: C --> G(k + 1, r + 1),
such that the planes t/J(p) do not contain a fixed vector, is the kth associated
map of some map ¢: C--> IP' if and only if rank dt/J ~ 1 at each point (cf.
Griffiths-Harris [3]).
(Hint: Do the case k = 2 first.)

There are a series of formulas relating the ramification indices of a linear system
and the degrees of the associated maps, called the Plucker formulas; the next exercises
give a derivation of these formulas.

C-10. Let E~+ 1 be the vector bundle of rank k + 1 on C whose fiber at a point pis
the space of sections of L near p modulo those vanishing to order k + 1-that
is,

(E~+ 1 )p = r(L/L(-(k + l)p)).

(For a formal definition, let Ll c C x C be the diagonal, n 1 and n 2 : C x C --> C

the projection maps, and take

Show that the first Chern class of E~+ 1 is

Exercises 39

C-11. Let V ® ((jc be the trivial vector bundle on C with fiber V, and rk: V ® ((jc -+ E1+ 1
the natural evaluation map. If k ~ r, show that Coker rk is a sheaf supported
on the inflectionary points of V, and that, in general, the dimension of the stalk
of Coker rk at pis the sum of the ramification indices L~=o rx;.

C-12. Show that the kernel of rk is locally free; and that as a subbundle of V ®(()cit
has fiber at each p given by

Conclude that if dk is the degree of

then

C-13. From the preceding three exercises, and the exact sequence of sheaves

deduce the basic relation

ana, in case k = r, the relation

L rx;(p, V) = (r + 1)d + ( r + 1) (2g- 2).

~· 2
Taking second differences in the relation (*),we have what are usually referred
to as the Plucker formulas:

L(cxk+1- rxk) = -dk+l + 2dk- dk- 1 + 2g- 2.

PEC

C-14. From the relation(**), deduce that the only nondegenerate curve C c IP' with no
iriflectionary points is the rational normal curve.

C-15. Combining the relation(**) with exercise C-3 above, prove again (cf. Exercise
E-8 of this chapter) that the sum of the weights of the Weierstrass points on C
is g 3 - g.

In case r = 1, the Plucker formula ( **) is just the Riemann-Hurwitz formula. In

the following exercise we look at the case r = 2.
40 I. Preliminaries

C-16. Let C c IP 2 be a plane curve with only ordinary singularities-that is, nodes and
cusps-and assume that the same is true for the dual curve C* = qP>(C) c IP 2 *.
Let d, o, K, b, and f be the degree and number of nodes, cusps, bitangents and
flexes of C, respectively, and d*, o*, K*, b*, and f* the corresponding quantities
for C*. Show that

o* = b, K* = j, b* = 0 and j* = K,

and, using (**) above and the genus formula g = ( d- 1) - (o + K), derive
the classical Plucker formulas for plane curves 2

d* = d(d- 1)- 2o - 3K,

d = d*(d* - 1) - 2b - 3f

D. Miscellaneous Exercises on Linear Systems

In the following series of exercises, C is a smooth curve of genus g and D is a divisor of

degree d. Assume that ID I is base-point-free and let c/Jv be the map associated to ID 1-

D-1. Show that the map c/Jv is an embedding if, and only if,

r(D - p - q) = r(D) - 2

for all p, q E C (including the case p = q).

D-2. Show that c/Jv is an embedding if d ;:::: 2g + 1. If d = 2g, then show that c/Jv will fail
to be an embedding if, and only if, D ~ K + p + q for some p, q E C.

D-3. Show that if Dis a general divisor of degree d ;:::: g + 3, then c/Jv is an embedding.
Hint: Choose D so that u(D)¢ (u(K)- J¥g_ 3 (C)) + JV;z(C). (Here, for A, B c
J(C), we set A + B = {a + b: a E A and bE B} and -A = {-a: a E A}.)

D-4. Suppose that g = 2. Show that C cannot be embedded in IP' as a curve of degree
d < 5. If d = 5, then describe the image of C under the map

How many quadric and cubic surfaces contain c/Jv(C)? Among the cubics, how
many are "new" (i.e., not in the ideal generated by the quadric)?

D-5. Show that the quadric containing the curve c/Jv(C) c IP 3 of the preceding exercise
is singular if D ~ 2K + p for some p E C, and smooth otherwise.

D-6. Suppose that g = 3. Show that the smallest degree of an embedding of C in a

projective space is 4 if Cis non-hyperelliptic and 6 if Cis hyperelliptic.

D-7. Suppose that g = 4. Show that the smallest degree of an embedding of C in a

projective space is 6 if C is non-hyperelliptic and 7 if C is hyperelliptic.
Exercises 41

D-8. Show that if m > 1 and g > 1

Use this to conclude that a canonical curve of genus 4lies on one quadric surface
Q c IP' 3 and on one cubic surfaceS not containing Q. Conclude that the canonical
curve is equal to Q (") S.

D-9. Let C be a hyperelliptic curve of genus g ~ 2. Show that every special linear series
g~ is a sum of r copies ofthe hyperelliptic g~ plus d - 2r fixed points. Conclude that
no curve of genus g ~ d can have a gi and a base-point-free gJ other than the gi.
(Hint: Using the canonical model of C show that any collection p 1 , •.. , Pd of
distinct points that fails to impose independent conditions on IK I must contain a
pair of conjugates under the hyperelliptic involution. Bear in mind that any set of
points on a rational normal curve is in linear general position.)

E. Weierstrass Points

The following series of exercises deals with the Weierstrass points on a smooth curve C
of genus g. We let vlt(C) denote the field ofmeromorphic functions on C.

Definition. We set N = {n E Z: n ~ 0} and, for any point p E C, define the Weierstrass

semigroup H P c Z by

HP = {n:thereexistsfeA(C)with(f)00 = np},

and the Weierstrass gap sequence GP c N by

GP = N- Hv
= {n: there is no f E A(C) with Cf)ao = np}.
E-1. Show that, for any point p E C, either:

r(np) = r((n - 1)p) and n E GP,

or
r(np)=r((n-1)p)+1 and nEHP.

Using the Riemann-Roch theorem conclude that

cardinality of Gv = g.

E-2.
(i) Show that

GP = {n EN: there exists wE H 0 (C, K) with Jl.p(w) = n - 1}.

Conclude again that Gv has cardinality g.

(ii) Show that for any point p E C, 2g - 1 E GP if and only if {!)((2g - 2)p) ~ K.
42 I. Preliminaries

E-3. If Cis hyperelliptic with n:C-+ IP 1 the two-sheeted map, show that:
(i) pis a branch point of n if and only if GP = {1, 3, 5, ... , 2g - 1}; and
(ii) pis not a branch point of n if and only if GP = {1, 2, 3, ... , g}.

Definitions.

(i) A point p E Cis a Weierstrass point if its gap sequence is not {1, 2, ... , g}.
(ii) p E Cis a norma/Weierstrass point if its gap sequence is {1, 2, ... ,g - 1, g + 1}.
(iii) In general, if H c r'\J is a semigroup such that G = r'\J - H has cardinality g
define its weight w(H) by setting

w(H) = I n - g(g + 1)/2.

neG

Finally, for p E C, we set

w(p) = w(Hp).

E-4. Show that the following are equivalent:

pis a Weierstrass point,

{ w(p) ¥- 0,
r(gp) ¥- 0.

Show that pis a normal Weierstrass point if and only if w(p) = 1.

E-5. If H = {a 1 , a2 , •. •} c r'\J is any semigroup such that G = r'\J - H has g elements,

show that
<Xl

w(H) = I (g - a. + n)
n=1

(here we assume that a 1 < a2 < · · ·).

E-6. Show that

<Xl

w(p) = I (r(np)- max{O, n- g})

n=l

(thus the weight w(p) measures the overall failure of the sequence of divisors {np}
to adhere to the generic behavior r(np) = max{O, n- g}).

Note. This problem may be notationally messy. It may be helpful to make an array
of dots

in which the lengths ofthe columns are the numbers r(np) - max(O, n - g); the lengths
of the rows will then be the gap values.
Exercises 43

E-7.

(i) Show that, if H c !\1 is any semigroup with G = !\1 - H having g elements,
then

w(H) :s; g(g - 1)/2,

with equality holding if and only if 2 E H.

(ii) If g 2::: 4 and w(H) = 2, then show that either

H = {g - 1, g + 2, g + 3, ... },
or

H = {g, g + 1, g + 3, g + 4, ... }.
E-8. Choose a basis w 1 , .•• , w 9 for H 0 (C, K) and, in terms of a local coordinate
z on C, set

w. = J.(z)dz, 01. = 1, ... ,g.

Consider the Wronskian

Jiz) )
J~(z)

... ~~-:'>(z)
(i) Show that the local expressions

W(z)(dz)g(g+ l)/l
(9+ 1)
give a global section WE H 0 (C, K® 2
).

(ii) Show that

,Up(W(z)) = w(p),

and conclude that

I w(p) = (g - 1)g(g + 1).

peC

E-9. Conclude from the preceding two exercises that every curve of genus g possesses
at least 2g + 2 distinct Weierstrass points, and that Cis hyperelliptic if and only
if it has exactly 2g + 2 distinct Weierstrass points.

E-10. For each of the curves (c)-(i) in exercise batch A, locate the Weierstrass points
and determine their weights.
44 I. Preliminaries

E-ll. Let C c IP' 2 be the Fermat curve given by

in homogeneous coordinates. Let ~ be a dth root of - 1 and

p = [1, ~. 0].

Show that pis a Weierstrass point with gap sequence

1, 2, 3, ............ ,d- 2,
d + 1, d + 2, ......... '2d - 3,

2d + 1, 2d + 2, ... ' 3d - 4,

(d - 4)d + 1, (d - 4)d + 2,
(d- 3)d + 1.

(Hint: Observe that the line L = {X 1 - ~X 0 = 0} has contact of order d with

Cat p (i.e., L· C = dp), and hence any curve of degree k < d having contact of
order k + 1 with pat d must contain L. Apply this to IKcl = II!Jc(d- 3)1 using
Exercise E-2.)

E-12. Continuing the preceding exercise, show that

e; 1)).
and that the total weight of all points

3d((d: 1)-
xi = 1, xj = ~. xk = 0 is thus

E-13. Using Exercises E-8 and E-12, show that if d = 4 (in which case Cis a canonical
curve), then every Weierstrass point is of the form Xi= 1, Xi= ~. Xk = 0, while
if d ~ 5 there are other Weierstrass points of C. In cased = 5, locate these others.

F. Automorphisms

F-1. Show that every automorphism of IP' 1 is of the form l/J(z) = (az + b)j(cz +d),
i.e., that

Aut(IP' 1 ) = PGL(2, C).

F-2. Let C be a curve of genus 1. Show that there is an exact sequence

0 --+ T--+ Aut(C) --+ F --+ 0,

where Tis the group of translations in the torus C, and F is a finite group of order
2, 4, or 6. (In fact, there is a unique curve of genus 1 such that 91= F = 4 (resp., 6).)
Exercises 45

Exercises F-3 through F-10 contain a proof of the following:

Theorem. Let C be a smooth curve of genus g ;::: 2 with automorphism group Aut( C). Then
Aut( C) is a .finite group of order at most 84(g - 1).

F-3. Let C be a hyperelliptic curve of genus g ;::: 2 and denote by n: C---> IP' 1 the two-
sheeted covering map given by the hyperelliptic gi. Denote also by -r the hyper-
elliptic involution (sheet interchange) and by p 1 , ... , p 2 g+l E IP' 1 the branch points
of n. Show that any element of Aut( C) commutes with -r, and from this conclude
that Aut( C) is a Z/2-extension of the group of automorphisms of IP' 1 preserving
{p 1 , .•• , p2 g+ 2 }. In partic11lar, in this case Aut Cis finite.

F-4. Let C be a curve of genus g, ¢ E Aut( C), and p 1 , •.• Pg+ 1 general points of C. Show
that there exists a meromorphic function f on C with polar divisor (/) 00 =
p 1 + · · · + Pg+ 1 , and, counting zeros and poles off- c/J*f, that ¢ has at most
2g + 2 fixed points. Using Weierstrass points, deduce that Aut( C) is finite.

F-5. For a simpler, albeit less elementary argument, show that if¢ is any automor-
phism of C, the eigenvalues of

are all roots of unity, and use this and the Lefschetz fixed point theorem to prove
that any automorphism of C fixing all the Weierstrass points of C is the identity.
Again, conclude that Aut( C) is finite.

F-6. For a Weierstrass-point-free argument, show that:

(i) any automorphism of C acting trivially on H 1 (C, Z) is the identity (use the
Lefschetz fixed point theorem); and
(ii) if¢ E Aut C, then the automorphism ¢*: H 1(C, Z)---> H 1(C, Z) is both
integral and unitary with respect to the Hermitian form H(w, w') = Jc
w A w'
on H 1 • 0 (C).

Conclude once more that Aut( C) is finite.

F -7. Let ¢: C ---> C be an automorphism of finite order n. Show that the orbit (or
quotient) space Cj {¢} has naturally the structure of a compact Riemann surface
such that C---> Cj{¢} is ann-sheeted covering totally ramified at the fixed points
of¢. Show that if n is prime then these are all the ramification points.

F -8. Retaining the notations from the preceding exercise, and using the Riemann-
Hurwitz formula show that the number rx of fixed points of¢ satisfies

2g-2+2n
('J. ~ 1 ,
n-

with inequality only if the quotient Cj{ ¢} ~ IP' 1 •

46 I. Preliminaries

F-9. Show more generally that if G is any group of automorphisms of C of order n,

GP c G the subgroup of elements fixing p E C, then the quotient CjG has naturally
the structure of a compact Riemann surface such that C---+ CfG is an n-sheeted
cover with ramification indices

for each p E C.

F-10. Continuing the preceding exercises, using again the Riemann~Hurwitz formula
show that Aut( C) has order at most 84(g - 1), and that if equality holds then the
quotient

CjAut(C)

is isomorphic to 1? 1 and there are exactly three points of C fixed under non-
trivial subgroups of Aut( C) (this proof requires non-trivial fiddling).

This completes the proof of the theorem stated above. In the remaining exercises
on automorphisms we shall give applications and special cases of this theorem.

F-11. Show that a curve of genus g = 2 cannot have an automorphism of order 7.

Conclude that the bound 84(g - 1) is not always sharp.

F -12. Show that a curve of genus g = 3 cannot have an automorphism of order 5.

F -13. Show that if a curve of genus g = 3 has an automorphism of order 7, then it must
be the normalization of one of the two curves

F -14. Let C be the normalization of the plane curve

Show that C is isomorphic to the normalization of the curve

but is not isomorphic to the normalizations of either y 2 = x6 - x or y 2 = x(x 4 - 1).

(Hint: use Exercise F-3.)

F -15. Show that if g( C) = 2, then Aut( C) has order at most 48, with equality if and only
if C is the normalization of

y2 = x(x 4 - 1).
Exercises 47

F-16. Show that if g(C) = 4, then Aut( C) has order at most 120, with equality if and
only if Cis the curve given in homogeneous coordinates X 0 , ••• , X 4 in IP' 4 by

l;X;=O,
l;Xf = 0,
L:xr = o.
F -17. Show that if g( C) = 3, then the order of Aut( C) is 168 if and only if Cis the curve

(cf. Exercise F-13).

F-18. Show that, if an automorphism cp has order nand at least (4n- 2)/(n- 1) fixed
points, then these fixed points are Weierstrass points of C. In particular, this will
be the case if C has five or more fixed points. (J. Lewittes.)

Note. Recall that an automorphism cp: C'---> C' is said to be a deck transformation
for the covering C' ~ C if cp commutes with a, i.e., a o cp = a, and the covering C' ~ C
is called Galois (or normal) if the group of deck transformations acts transitively on
the points of a general fiber.

F -19. Show that a covering C' ~ C is Galois if and only if the corresponding extension
of function fields A( C)~ A(C') is normal.

F-20. Suppose the covering C' ~ C is unramified. Show that a is Galois if and only if
for p E C' the image

is a normal subgroup; and, conversely, every normal subgroup of finite index in

n 1(C) arises in this way.

F-21. Let C' ~ C be an unramified Galois covering of Riemann surfaces and cp E Aut C.
Show that cp lifts to an automorphism of C' if and only if

preserves the subgroup a*n 1 (C') c n 1 (C) associated to the cover C'.

F-22. Using the preceding exercise, show that for any curve C of genus g ~ 2 and any n,
there exists an n29-sheeted unramified Galois cover of C to which every auto-
morphism of C lifts. Conclude, in particular, that if the estimate in the theorem
preceding exercise F -3 is sharp for curves of genus g, it is sharp for curves of genus
h = n29(g - 1) + 1; and hence, in particular, that it is sharp for infinitely many g.

Note. To see that the estimate is not sharp for infinitely many genera, in fact that the
inequality Aut( C) :s; 8(g + 1) holds for infinitely many genera-see Accola [1].
48 I. Preliminaries

G. Period Matrices
The periods of abelian integrals may always be given as ordinary calculus integrals.
Sometimes, e.g., if the curve has a large symmetry group, they may even be evaluated.

G-1. Let C be the curve (a) of exercise batch A. Recalling the notations from Exercise
A-3 let A1 be the (real) line segment joining -1 and -win the complex x-plane,
and A2 the segment joining-wand -w 2 • Show that theinverseimagesn- 1 (A;) c C
are closed curves that (with suitable orientations) give a symplectic basis for
H 1 (C, Z).

G-2. Show that

f -ro dx 1 J-"' 2
dx
-1 Jx3+1 = ;;; -ro Jx3+l
and conclude that C is isomorphic to Cj(Z + Zw). (Z + Zw is the lattice generated
by 1 and w.)

G-3.
(i) Let C now be the compact Riemann surface corresponding to the curve (d)
of exercise batch A, and recall the notations used in Exercise A-5. Let A; be
the line segment joining ~;- 1 to~; in the complex x-plane and set r; = n- 1(A;).
Orient the loops r; and compute the intersection matrix Q(r;, rj). Conclude
that the r; generate H 1 (C, Z).
(ii) In terms of the r; find a normalized basis {y;} for H 1(C, Z).

G-4. Using the relations

f~· dx 1 J~<+' dx
~·-I Jx6=1 = ~ ~· Jx6-=l'
f ~' xdx 1 J~ .. , xdx
~·-' -}x6 - 1= ~2 ~· Jx6=1'
find a basis for H 0 (C, K) normalized with respect to the basis {y;} in the preceding
exercise. From this write down the period matrix of C.

H. Elementary Properties of Abelian Varieties

A trivial but important fact is that any holomorphic mapping

f:X--> y

between compact, complex manifolds induces a mapping

on holomorphic q-forms.
Exercises 49

H-1. Show that any holomorphic mapping between complex tori is a translation
followed by a group homomorphism.

We recall that, if A c A' are lattices of rank 2n inC" and A= C"/A, B = C"/A', then
the induced map

is an isogeny, and A and B are said to be isogenous.

H-2. Show that an equidimensional map between complex tori is an isogeny (equi-
dimensional means that dim A = dim Band the Jacobian ofthe map is generically
of maximal rank).

An abelian variety A is simple if it has no non-trivial abelian subvarieties (non-

trivial means of dimension not equal to 0 or dim A).

H-3. Show that any non-constant holomorphic mapping between simple abelian
varieties is an isogeny.

H-4. (Poincare's Complete Reducibility Theorem). Show that a non-simple abelian

variety is isogenous to a non-trivial product of abelian varieties. (Note: This
requires essential use of the polarization.)

H-5. Iff: C 1 ---+ C 2 is a non-constant holomorphic mapping between compact Riemann

surfaces where C 2 has positive genus, show that J(C 1 ) is isogenous to the product
of J(C 2 ) with another abelian variety.

H-6. Let E be the elliptic curve

and C the hyperelliptic curve

Show that J(C) is isogenous toE x E.

(Hint: Consider the automorphisms (x,y)---+(-x,y) and (x,y)---+(-x, -y)
of C.)

H-7. Let E be the elliptic curve

and C the Fermat quartic given in affine coordinates by

Show that J(C) is isogenous toE x E x E.

Remark. For C a smooth curve, the principally polarized abelian variety J(C) is
never a non-trivial product of principally polarized abelian varieties (here product
means with polarizations). However, J(C) may be a non-trivial product of complex tori.
Appendix A

The Riemann-Roch Theorem,

Hodge Theorem, and Adjoint
Linear Systems

The Riemann-Roch theorem is the basic result in the theory of algebraic

curves. The usual proofs of this theorem are either potential theoretic,
algebraic using repartitions, or a mixture of each using sheaf cohomology
and Serre duality. In the first sequence of exercises below we will give an
alternate proof, one that has proven useful pedagogically and that gives as a
by-product the Hodge theorem and the completeness of adjoint series for
plane curves with nodes.
The proof will be divided into two major steps, the first of which is the
following existence result:

(*) Theorem. Let C be a compact Riemann surface. Then there exists a

holomorphic mapping

f: c--+ [p>2

whose imager is a plane curve with nodes and wheref: c--+ r is birational.

The second step will be to prove the Riemann-Roch theorem for C by

studying r. The reader may omit the first step and begin with Exercise 5
below.
To establish Theorem ( *) we shall assume the following basic analytic
fact, an elementary proof of which may be found in the appendix to Gunning-
Rossi [1]:

(**) Let C be a compact Riemann surface and L--+ C a holomorphic line

bundle. Then

1. Let p 1 , p 2 , p 3 , ... be a sequence of points (not necessarily distinct), and set

Dn = p 1 + · · · + Pn· Using(**) show that for n large enough,
Appendix A. Riemann-Roch Theorem, Hodge Theorem, Adjoint Linear Systems 51

and that

for some constant c.

(Hint: Consider the exact cohomology sequence of

0 ~ L(Dn-1) ~ L(Dn) ~ L(Dn) ® (!) Pn ~ 0

for n = I, 2, .... Here L(Dn) = L ® (!)(Dn) and (!)P" is the skyscraper sheaf
(!)cj(!)c( -pn).)

2. Using the preceding exercise show that there exists a holomorphic em-
bedding

¢: C ~ IP'.

3. Let C c IP' be a one-dimensional complex submanifold (e.g., the image

¢(C) in the preceding problem). Show that the arguments used in Exercises
B-5 and B-6 of Chapter 1 may be adapted to show that:
(i) under a general projection IP' - IP'- 4 ~ IP\ C maps biholomor-
phically to its image; and
(ii) under a general projection IP' - IP'- 3 ~ IP 2 , C maps in a generically
one-to-one fashion onto an analytic curve r c IP 2 (i.e., r is locally
the zeros of an analytic function in open subsets of IP 2 ) with the
following property: every point of r is either smooth or is the union
of two smooth branches having distinct tangents (thus, C ~ IP 2 is
an immersion). We shall refer to the image of such an immersion as
an analytic plane curve with nodes, and we shall call the original
Riemann surface C (always connected) its normalization.

4. Let r c IP 2 be an analytic plane curve with nodes. By considering the

projection
n: r~ IP 1
onto a general line, show that r is an irreducible algebraic curve
F(x 0 , x 1 , x 2 ) = 0.
(Hint: In a suitable affine coordinate system the projection is (x, y) ~ x.
Both x andy are meromorphic functions on C, and we let y 1 (x), ... , yix)
be the values of yon the points in n- 1 (x). Now consider
d
f(x, y) = L (y - Yv(x)),
v=l

and show that F(x 0 , x 1 , x 2 ) = x~f(xdx 0 , x 2 jx 0 ) has the required proper-

ties.)
With this exercise we have completed the proof of Theorem ( *).
52 Appendix A. Riemann-Roch Theorem, Hodge Theorem, Adjoint Linear Systems

Let r c IP' 2 be an algebraic curve with nodes. We say that r is in general

position with respect to a coordinate system [x 0 , x 1 , x 2 ] in case the following
conditions are satisfied:
(i) [O, o, lJ ¢ r;
(ii) the line Loo = {x 0 = 0} is neither tangent to r nor passes through
a node;
(iii) set x = xdx 0 andy = x 2 jx 0 and consider the map

induced by the projection (x, y)-+ x. Then over each point x E IP' 1
there is at most one simple branch point (two sheets interchanging)
or one node (note that n;; 1 ( oo) = L 00 • r).

5. Show that r is in general position with respect to a general coordinate

system.
(Hint: It is useful to interpret the branch points of n: r-+ IP' 1 as the
vertical tangents, i.e., the points of r whose tangent line passes through the
center [0, 0, 1] of projection.)

From now on we will assume that r c IP' 2 is an irreducible plane curve

with nodes having homogeneous equation

affine equation

f(x, y) = 0,
and in general position with respect to the coordinate system. We denote by
¢: c -+ r the normalization of r, and assume that r has degree d.

6. Let D c IP' 2 be a plane curve of degree n given by a homogeneous poly-

nomial G(x 0 , xh x 2 ) and set g(x, y) = G(l, x, y). Assuming that r is not
contained in D, define the divisor ¢*D to be the zero divisor of the mero-
morphic function ¢*(g) on C. Show that

deg(¢*D) = nd

(Bezout's theorem).
(Hint: Note that

and that
deg(nL 00 • r) = nd.
Appendix A. Riemann-Roch Theorem, Hodge Theorem, Adjoint Linear Systems 53

Now use the fact that the number of zeros equals the number of poles of a
meromorphic function on C.)

7. Denote by

n: C ~ IP 1

the composition of ¢: c~r with the projection of r on the x-axis, and by

.L\ c C the divisor lying over the nodes of r (thus deg .L\ = 2£5 where £5 is the
number of nodes). Let E be the plane curve with affine equation
(of foy)(x, y) = 0. Show that the ramification divisor R of n is given by

R = cp*E- .L\.

Using the Riemann-Hurwitz formula conclude that

g = (d - l)(d - 2)/2 - £5

(genus formula).

8. Let D be the plane curve given by a homogeneous polynomial G(x 0 , x 1 , x 2 )

of degree d - 3 and set g(x, y) = G(l, x, y). Define the Poincare residue to be

(i) w_
-
¢*( g(x, y) dx ) _
(of foy)(x, y) -
¢*(- g(x, y) dy )
(of /ox)(x, y)

(why is the equality true?). Show that w is a merorriorphic differential with

divisor

(ii) (w) = cp*D - .L\.

Using this prove the adjunction formula

(iii) Kc ~ C9c(d - 3)(- .L\),

and the inequality

h0 (C, Kc) ~ h0 (1P 2 , C9(d - 3)) - t5

= (d - l)(d - 2)/2 - £5
=g.

9. Show that, with w defined as in (i), statement (ii) is true even if r is not in
general position with respect to the coordinate system; conclude (iii)
directly.
54 Appendix A. Riemann-Roch Theorem, Hodge Theorem, Adjoint Linear Systems

10. Denote by HbR(C) the first deRham cohomology (with complex co-
efficients) of C and by H 1 • 0 (C) c HbR(C) the image of the natural map

Prove the Hodge theorem

l
(Hint: For w, 1/J E H 0 (C, Kc) it is clear that

F;I L w A w> 0, w ;f-0,

few A 1/1 = 0.

From this conclude that

is injective, and then by the inequality in Exercise 8 that j E8 J is an iso-

morphism.)

11. Let Pl, ... ' PiJ E IP> 2 be the nodes of r:

(i) Show that p 1 , ... , Pil impose independent conditions on the linear
system I(!)IF'2(d - 3) I of plane curves of degree d - 3.
(ii) Show that the system 1: c 1 (!)IF'2(d - 3) 1 of curves passing through
P1> ... , Pil cuts out the complete canonical series IKc I; i.e.

IKe I= {¢*D- A: DE 1:}.

(Hint: By the Hodge theorem equality must hold in the inequality in

Exercise 8.)

Remark. Assertions (i) and (ii) are the cornerstones of the classical theory
of plane curves; we have established them by a transcendental argument.

Let D be any divisor of degree m on C. In the remaining exercises of this

group we will construct the complete linear system ID I and establish the
Riemann-Roch theorem and completeness of the adjoint system.

12. Write D = D' - D" where D', D" are effective divisors of respective
degrees m', m". Let G be a plane curve of degree n containing the nodes
Appendix A. Riemann-Roch Theorem, Hodge Theorem, Adjoint Linear Systems 55

P1o ... , Po of rand the image of D'; i.e.

¢*G = D' + Ll + E
for some effective divisor E with

deg E = nd - 2<5 - m'.

Show that the plane curves H of degree n satisfying

¢*H;;::: Ll + E + D"
form a linear system~ c I@IJ>2(n)l with

dim ~ ;;::: (n + l)(n + 2)/2 - 1 - nd + b + m.

13. Given H E ~ write
¢*H = il + E + D" + B,
and show that

From this conclude that

r(D) ;;::: dim~ - h0 (1P> 2 , @(n - d))

;;::: (n + l)(n + 2)/2 - 1 - nd + [) + m - (n - d + l)(n - d + 2)/2
= -(d- l)(d- 2)/2 + [) + m
=m-g.

In particular, if m ;;::: g then r(D) ;;::: 0. The assertion

r(D) ;;::: deg D - g

is generally known as Riemann's inequality.

14. Assuming that D = p 1 + · · · + Pm is effective use the residue theorem in

the form

I Resp,(fw) = 0, f E !l'(D) and wE H 0 (C, K)

i

to show that

r(D) ~ m- g + h0 (C, K( -D)).

56 Appendix A. Riemann-Roch Theorem, Hodge Theorem, Adjoint Linear Systems

15. Using the preceding two exercises, establish the Riemann-Roch theorem

r(D) = deg D- g + h0 (C, K( -D))

for any divisor D.
(Hint: Treat the cases h0 (C, (!}(D))# 0, h 0 (C, K( -D))# 0, and
h 0 (C, (!}(D))= h0 (C, K( -D))= 0 separately.)

16. Show that the system Ln c I(!}o:>,(n) I of curves passing through P1o ... , p~
cuts out the complete linear system IKc(n - d + 3) I for all n (this is called
the completeness of linear system of adjoint curves).

§1. Applications of the Discussion About

Plane Curves with Nodes
This series of exercises will give some properties of plane curves r c IP' 2 ,
possibly with nodes. We recall that a setS = {p 1, ... , pd} of distinct points
in IP' 2 is said to impose independent conditions on curves of degree n if

where f s c (!}o:>2 is the ideal sheaf of the zero-dimensional variety S.

17. Show that any n + 1 distinct points impose independent conditions on

curves of degree n. Show that n + 2 distinct points fail to impose independent
conditions on curves of degree n if, and only if, they lie on a line.

18. Let C c IP' 2 be a smooth plane curve of degree d ~ 4. Using the preceding
exercise and the geometric form of the Riemann-Roch theorem, show that:
(i) C does not have a g~ form :::;; d - 2;
(ii) if lEI is a gJ_lo then E = D- p where p E C and DE l(!}c(1)1;
(iii) h0 (C, (!}c(1)) = 3 (i.e., the g~ on C cut out by the lines is complete),
and I(!}c(l) I is the unique g~ on C.

19. Improve Exercise 17 by showing that if a setS of 2n + 1 distinct points in

IP' 2 fail to impose independent conditions on curves of degree n, then S must
include n + 2 collinear points.

20. Now suppose that r c IP' 2 is an irreducible curve of degree d with b

nodes and no other singularities and C its normalization. Show that:
(i) if b :::;; d - 2, then C does not have a g~ for m :::;; d - 3;
(ii) if b:::;; d- 3, then any gJ_ 2 on Cis cut out by lines through a node;
(iii) if b :::;; d - 3, then the g~ cut out on r by lines is complete, and
if b < d - 3 it is the unique g~ on the normalization C of r.
§2. Adjoint Conditions in General 57

21. Let r c IP' 2 be the plane quintic

I XfXJ = 0.
i*i

Determine the genus g(C) and the space H 0 (C, K) of holomorphic differ-
entials for the normalization C of r. Then give the canonical map and the
equations of its image. (Upon doing this exercise you will know the standard
Cremona transformation of IP' 2 , if you don't already.)

22. Let r c IP' 2 be an irreducible quintic curve with <5 nodes, C its normaliza-
tion. Show that the g;
cut out on C by the lines is complete if and only if
<5 :s; 3.

23. Let r c IP' 2 be an irreducible sextic curve with <5 nodes, C its normaliza-
tion. Show that the g~ cut out on C by lines is complete if <5 :s; 5, incomplete
if <5 ~ 7, and if <5 = 6 then it is incomplete if and only if all six nodes lie on a
conic.

24. More generally, let r be an irreducible plane curve of degree d

c IP' 2
with nodes r 1 , ••• , r~, C its normalization. Show that the g~ cut out on C by
lines is complete if and only if r 1 , ... , r0 impose independent conditions on
curves of degree d - 4.

§2. Adjoint Conditions in General

The analysis of plane curves carried out in the above exercises under the
assumption that r has only nodes can be carried out in more general cir-
cumstances.
Maintaining the notations of those exercises, if r E r is any singularity in
the finite plane, p 1 , ... , Pk the points of C lying over r, we define the adjoint
divisor L1r of r to be the divisor on C

where

We define the adjoint ideal Ir c (!)lf'2.r to be the ideal of functions g(x, y) such
that the divisor

(cf>*g) ~ L1r;
58 Appendix A. Riemann-Roch Theorem, Hodge Theorem, Adjoint Linear Systems

that is, the ideal of functions g such that the differential

_ *( g(x, y) dx )
wg - ¢ (of foy)(x, y)

is regular near ¢ - 1 (r), and we say that g satisfies the adjoint conditions at r
if g E J,. Globally, the definitions are analogous, e.g., the adjoint divisor A of
¢: c--+ r c IP 2 is the sum
A= I A,.
rersing

Finally, the number {J. of adjoint conditions at r is the index in (91P'2, r of

the ideal I.-that is, the dimension of (91P'2,r/I, as a complex vector space-
and the number of adjoint conditions overall is the sum of these numbers; i.e.,

{J = I J,.
rersing

25. Show that the ramification divisor of the projection map n: C--+ IP 1 is
R - A; similarly, show that the divisor of the meromorphic 1-form n* dx is

(n* dx) = R- 2D 00 -A,

i.e.

Kc = fjl*(9r(d - 3)(- A).

Either way, deduce the formula

_ (d - 1)(d - 2) l d
g(c) -
A
2 -2 ~~

26. Show that forgE C[x, y] the 1-form

wg -
_¢ *( g(x,y)dx )
(of /oy)(x, y)

is holomorphic on C if and only if deg g ::::;; d - 3 and g satisfies the adjoint

conditions at every singularity of r. Conclude that

ho(c K) > (d - 1)(d - 2) - {J

' - 2
and hence that
{J ~ 1 deg A.
§2. Adjoint Conditions in General 59

27. Suppose now that r E r lies in the finite plane, with p 1, ... , Pk lying over
r; let

_ <!>*(8f/8y
w- dx )

and let V be the (deg 8,)-dimensional vector space

Define a symmetric bilinear pairing

1/J:VxV--d::
by

1/J(g, h) = L Resp,(ghw).
i

Show that 1/J is well defined and non-degenerate (i.e., the induced map
I[!: V ~ V* is an isomorphism).
28. Let g(x, y) be any polynomial. Show by the residue theorem that

L Resp;(</J*gw) = 0.
i

29. Let I, be as above the adjoint ideal at r, W the •\-dimensional vector

space

and W = </J*W c V the image of W under the injection

Show by Exercises 27 and 28 that W is isotropic for 1/J, i.e., 1/J(W, W) =0 or,
equivalently

1/J*W c Ann(W).

30. Deduce from the preceding exercise and Exercise 26 above the Gorenstein
relation
deg 8, = 2<5,
and likewise the global relation

g(C) = (d - l~d - 2) _ J.
60 Appendix A. Riemann-Roch Theorem, Hodge Theorem, Adjoint Linear Systems

31. Using Exercises 26 and 30, deduce that:

(a) the adjoint ideals of r impose independent conditions on curves of
degree d- 3;
(b) the linear system of plane curves of degree d - 3 satisfying the
adjoint conditions cuts out the complete canonical series on C; and
(c) more generally, the linear system of plane curves of degree n satisfying
the adjoint conditions cuts out on C the complete series
IKc(n- d + 3)1.

32. Examples. Find the adjoint divisor and conditions for the following
singularities; verify in each of these cases that deg A,= 2c5,:
(a) a node: two smooth arcs meeting transversely (e.g., y 2 - x 2 = 0);
(b) an nth-order node: two smooth arcs having contact of order n
(e.g., yz - xzn = 0);
(c) an nth-order cusp (e.g., y 2 - x 2 n + 1 = O);
(d) an ordinary triple point: three smooth arcs meeting transversely at
r (e.g., xy(x - y) = 0);
(e) a triple point with an infinitely near double point: three smooth arcs,
two of which are simply tangent (e.g., (y + x 2 )(y - x 2 )x = 0);
(f) y 3 - x 5 = 0;
(g) yP - xq = 0 for p, q any pair of relatively prime integers.
Chapter II

Determinantal Varieties

In this chapter we have collected a few foundational results about determin-

antal varieties that we will need in later chapters. The main application of the
results in this chapter will be to the varieties of special divisors on curves.
These varieties have in fact a natural determinantal structure defined in terms
of the Brill-Noether matrix, and their study is the central theme of this book.
Another ubiquitous example of determinantal variety is the one of rational
normal scrolls.

§1. Tangent Cones to Analytic Spaces

We first need to review a number of ancillary results dealing with the notion
of tangent cone to an analytic space at a point.
If Y is a complex manifold and X c Y an effective divisor (i.e., an analytic
subspace given locally by one equation), then it is easy to define the tangent
cone to X at a point p E X. In holomorphic coordinates z 1> ••• , zn on Y
centered at p we expand a local defining equation f(z) of X as a series of
homogeneous forms

00

f(z) = L J;,(z).
k=O

Regarding z 1 , ..• , zn as coordinates on the tangent space Tp(Y), the tangent

cone is the algebraic subvariety

defined by the equation

fh(z) = 0,
62 II. Determinantal Varieties

where his the minimum integer k such thatj~ # 0. For an arbitrary analytic
space X, denoting by m the maximal ideal in the local ring (f)X,p• the tangent
cone to X at p is defined to be

~(X) = Spec(~ mi/mi+ 1 ).

The natural surjective ring homomorphism

EB Symi(m/mz) ~ EB mi/mi+ 1
i

defines an inclusion

of the tangent cone in the Zariski tangent space to X at p. The projectivized

tangent cone to X at p is defined by

In other words, the projectivized tangent cone is the exceptional divisor in

the blow-up of X at p, and it is naturally an algebraic subvariety of the
projectivized Zariski tangent space IPTP(X). The multiplicity multp(X) of X
at pis then defined to be the degree of IPg;;(X) in IPTP(X). It is important to
notice that IPg;;(X) is a Cartier divisor on the blow-up X of X at p. In other
terms, IPg;;(X) is an analytic subspace of X defined, locally, by the vanishing
of a single holomorphic function.
In case X is given as an analytic subspace of a complex manifold Y there is
another convenient description of the tangent cones to X which parallels the
one we have given for divisors. Let p be a point of X. Denote by I the ideal of
X in (f)Y,p• and by M and m = Mjl the maximal ideals in (f)Y,p and (f)X,p•
respectively. We set

It is straightforward to check that

is a graded ideal. This is usually called the ideal of initial forms of I. We claim
that J is, indeed, the ideal of ~(X) in Tp(Y). In fact, for any k,
mk/mk+ 1 ~ + /)/(Mk+ 1 +I)
(Mk
~ Mk/(Mk+1 + Mk n I)
~ (Mk/Mk+1)/lk>
§1. Tangent Cones to Analytic Spaces 63

and therefore,

which is what we had to prove.

Another useful remark, in the above situation, is the following. Let Y be
the blow-up of Y at p, and denote byE = IPTv(Y) the exceptional divisor on
Y. Then IP~(X) is the intersection of E with the blow-up X of X at p.
In particular, the line bundle associated to IPg;,(X) is the restriction of
(!)(E) to X.
There is a standard situation where projectivized tangent cones can be
easily computed; this will occur time and again in the following chapters.
Specifically, suppose we are given a proper holomorphic map

f: X-> Y

between complex manifolds. We shall assume thatfmaps X bimeromorphi-

cally onto its image. Let p be a point ofj(X); we wish to compute the pro-
jectivized tangent cone to f(X) at p under the additional assumption that the
(scheme-theoretic) fiberf- 1 (p) is smooth. This assumption implies that the
vector bundle homomorphism

is injective, where N is the normal bundle tof- 1 (p) in X. On the other hand
dj, being injective, induces a well-defined morphism

g: lPN-> IPTp(Y).

The lemma we want to prove is the following:

(1.1) Lemma. Let f: X-> Y be a proper holomorphic map between complex

manifolds which maps X bimeromorphically onto Z = f(X). Let p be a point of
Z such that the fiber f - 1 (p) is smooth. Denote by N the normal bundle to f- 1 (p)
in X and let g be as above. Denote by y and 17 the analytic cycles associated to
lPN and IP~(Z), respectively. Then g*(y) = 17·

To prove this we first blow up Z and Y at p, and X alongf- 1 (p). We then

get a commutative diagram

G----->EcF
("\ ("\ ("\

x____i___,zc¥

j j j
X~ZcY
64 II. Determinantal Varieties

where G, E, and Fare the exceptional divisors. As we shall presently show the
existence of J is ensured by the smoothness of f- 1 (p). Choose coordinates
z 1 , .•. , zn on X in such a way that G is locally defined by the equations
z1 = · · · = zk = 0; likewise, choose coordinates w 1 , ... , wm on Y, centered
at p. In these coordinates the map f can be written

oc = 1, ... , m.
Let

i,j = 1, ... ' k,

be the local equations of X in UJ>k-l x X. Similarly, let

(1.2) oc, f3 = 1, ... , m,

be the local equations of Yin IP>m- 1 x Y. In these coordinates the lifting J is

given by

A trivial computation shows that (1.2) is satisfied. On the other hand, to say
thatf- 1 (p) is smooth means that the Jacobian matrix

(iJz.
of~ (0, ... ' 0, zk+1' ... ' zn)) -
1 a-l, ... ,m
i= 1, ... ,k

has rank k, and therefore the 11~'s cannot simultaneously vanish.

By what we already remarked we have

E = IP>~(Z),

F = IP>J;,(Y);
furthermore,
G= IP>N.

With these identifications it is apparent from the local expression of]that the
restriction
§1. Tangent Cones to Analytic Spaces 65

is nothing but g. Notice, moreover, that Z is reduced and irreducible and],

being proper, maps X bimeromorphically onto Z. It follows, in particular,
that g = J 16 maps G onto E. Assuming, for the time being, that f - 1 (p), and
hence G, is connected, this implies that

where Ais a rational number. We are thus reduced to showing that the degrees
of g*(y) and 17 are the same. We recall that, if CD 6 (1) denotes the tautological
line bundle on G = IP'N,

CDF(l) ~ CDv( -F)® CDF,

CDG(l) ~ CDx(- G)® CDG,

while the local expressions for Jshow that

]*CDv(F) ~ CDx(G).

Therefore, we have

deg(17) = ic 1(CDil)t- 1

= ic1(CDv(-F)t- 1

= J/1(CDv( -F))"- 1 · c1(CDz(E))

= - J/1(CDv( -F))"

= - f/1(CDx( -G))"

= Lc1(CD 6 (1)t- 1.

The next to last equality follows from the fact that] maps X bimeromorphi-
cally onto Z. On the other hand, we have

deg(g*(y)) = Lg*c 1(CDF(l))"- 1

= Lc1 (CDG(l))n- 1.
66 II. Determinantal Varieties

This disposes of the case when f - 1 (p) is connected. In general, notice that,
upon shrinking X and Y if necessary, we may suppose that each connected
component of X contains exactly one component of G. The general case
follows at once by applying to each connected component of X the special
case we have just proved. The proof of Lemma (1.1) is thus completed.

Corollary. With the hypotheses and the notation of Lemma (1.1)

g(IP>N) = P~(Z)

as sets.
To prove the corollary it suffices to observe that Py;;(Z), being a Cartier
divisor in the irreducible n-dimensional variety Z, has pure dimension n - 1,
so that the set-theoretical equality follows from the equality of cycles in the
statement of the lemma.
In special cases the conclusion of the above corollary can be considerably
sharpened.

(1.3) Lemma. Suppose the hypotheses of Lemma (1.1) are met. Assume more-
over that

h = df: N --+ Tp(Y)

is a birational map of N onto its image h(N), and that h(N) is normal. Then

h(N) = ~(Z)

as schemes.

Proof By the corollary we just proved h(N) and ff;,(Z) have the same sup-
port; in terms of their affine coordinate rings this means that

ffi mi/mi+ 1 ~ r(h(N), (!)h<N>) = r(N, (!)N)

i~O

is onto. Herem stands for the maximal ideal in (!)z,p and the equality on the
right-hand side follows from the normality of h(N) and the fact that Nand
h(N) are birational. We let I c (!)x be the ideal sheaf off- 1 (p). Thus,

r(N, (!)N) = EB r(X, Ii/Ii+ 1 ),

i;:::O

and all we have to prove is that

ffi mi/mi+ 1 ffi r(X, JijJi+ 1)

--+
i;,:O i;,:O
§2. Generic Determinantal Varieties: Geometric Description 67

is injective. Suppose not: this means that there is a function f 1 on Z vanishing

at p to order exactly equal to i, whose pull-back to X vanishes to order at
least i + 1 along! - 1 (p ). By the surjectivity of rx we can inductively construct,
foranyh?: l,functionsg11 Emi+hsuchthatthepull-backtoXoff1 + g 1 + ···
+ gh vanishes to order at least i + h + 1 alongf-\p). Passing to comple-
tions, this shows that

is not mJective. By the theorem on formal functions (cf. Hartshorne [2,

page 277]; here the result is proved in an algebraic setting, but the argument
works also in an analytic one) the right-hand side is isomorphic to <J:ii;J;,.
Thus the homomorphism

is not injective. This is absurd since, by the birationality assumption on f,

mz.p injects into (f*mx)p. Q.E.D.

§2. Generic Determinantal Varieties: Geometric Description

Let M(m, n) = M be the variety of m x n complex matrices, and, for

0::;; k::;; min(m, n), denote by Mk(m, n) = Mk the locus of matrices of rank
at most k. This is called a generic determinantal variety. We also set

Mk = Mk(m, n) ={(A, W)EM x G(n- k, n): A· W = 0}.

Projection onto G(n - k, n) exhibits Mk as an algebraic vector bundle over

G(n- k, n) of rank mk. Therefore Mk is smooth, connected, and has di-
mension k(m + n - k). Now let

n: M x G(n - k, n)--> M

be the projection. Clearly n maps Mk properly onto Mk, showing that Mk is

an irreducible algebraic subvariety of M. Moreover, when A E Mk - Mk-l•
there is exactly one point of Mk lying above A, namely the pair (A, ker A).
This shows

Proposition. Mk is an irreducible algebraic subvariety of M of codimension

(m - k)(n - k).

The next step will be to compute the tangent space to Mk

at any point
(A, W). We shall do this by "differentiating" the relation A· W = 0. We
begin by recalling the standard identification between the tangent space to
68 II. Determinantal Varieties

G(n - k, n) at W and Hom(W, C"/W), which can be given as follows. A

tangent vector to G(n - k, n) at W is a morphism from Spec (C[e]) to
G(n - k, n) centered at W. Upon choosing a basis w1 , .•• , wn-k for W, this
can be lifted to a morphism

into the variety of (n - k)-frames in C". The v/s are uniquely determined
modulo W, and the homomorphism

¢: w~C"/W

corresponding to the given tangent vector is defined by

¢(w;) = class of V; in C"jW.

Now, a tangent vector to Mk at (A, W) is a couple (B, ¢),where B is an

m x n matrix and ¢ is as above, such that

(A + eB)(w; + ev;) = 0, i = 1, ... , n- k.

This amounts to saying that A · W = 0 and, moreover,

Av; + Bw; = 0, i = 1, ... , n- k,

that is,

(A¢ + B) · W = 0.

Thus

(2.1)

which has dimension equal to

(n - k)(m - rank(A)).

In particular, when A E Mk - Mk_ 1 , the differential of n at (A, ker A) is

injective, showing that A is a smooth point of M k and that

In general, the set-theoretic fiber of

§2. Generic Determinantal Varieties: Geometric Description 69

over any A E Mk can be identified with G(n- k, ker A). The above dimen-
sion count shows that, if W is an (n - k)-plane in ker A, then the differential
of n maps the normal space at (A, W) to the set-theoretic fiber over A in-
jectively into TA(M). Thus the scheme-theoretic fiber n- 1 (A) is smooth, and
therefore the hypotheses of Lemma (1.1) are met by

at any A E Mk. Hence, by the corollary to Lemma (1.1) and formula (2.1), as
a set, the tangent cone to Mk at A is

(2.2) !§A(Mk) = {BE M: for some WE G(n- k, ker A), B · W c A ·IC"}

(actually we will see later that this is an equality of schemes). A first conse-
quence is that, when A EMk-b !fA(Mk) spans Mas a vector space. In fact,
given any non-zero vector v E IC", and any vector wE em, there is aBE !fA(Mk)
such that Bv = w. That this is true follows from the remark that there is
certainly an element WE G(n - k, ker A) such that v ¢ W; for such a W, the
conditions

B· W c A·Cn,
B·v =w

can be simultaneously met. Recalling that, as we proved above, Mk - Mk- 1

is smooth, we conclude

Proposition. The singular locus of Mk is exactly equal to Mk_ 1 •

A priori, formula (2.2) only describes the tangent cones toMk as sets. What
we would like to remark here is that, as we announced above, these tangent
cones are indeed reduced. We shall prove this by showing that the hypotheses
of Lemma (1.3) are satisfied by

at any A E Mk. The point is that the support of !JA(Mk), namely

{BE M: for some WE G(n - k, ker A), B · W c A· IC"}

is nothing but the product of a linear space times a generic determinantal

variety. To see this we choose complementary subspaces H 2 to H 1 = ker A
and K 2 to K 1 = A ·IC", and notice that to give an m x n matrix B such that,
70 II. Determinantal Varieties

for some WE G(n - k, ker A), B · W c A· IC", is the same as giving arbitrary
linear mappings

al: Hl--+ Kl,

az: Hz--+ Kz,
a3: Hz--+ Kb

plus a linear mapping

of rank at most k - h, where h = rank(A). Thus the support of $A(Mk) can

be identified with the product

V x Mk-h(m - h, n - h),

where V is a linear space of dimension h(m + n- h). What is more, the

normal bundle to the fiber n- 1 (A) = G(n - k, ker A), that is,

{(B, W): BEM, WE G(n- k, ker A), B · W c A ·IC"},

is nothing but the product

V x Mk-h(m - h, n - h),

and hence maps birationally to the support of $A(Mk). At this point, to

finish with the proof that $A(M k) is reduced, it would be enough to show
that generic determinantal varieties are normal. However, the elementary
linear algebra that we have been using so far is not sufficient to prove this.
What is needed is a rather more sophisticated analysis of the ideal of Mk; this
will be carried out in the next section, where a proof of the normality of
generic determinantal varieties will also be found.

§3. The Ideal of a Generic Determinantal Variety

The main purpose of the present section is to study the ideal of a generic
determinantal variety. The first results to be proved are quite classical and
go under the names of first and second fundamental theorem of invariant
theory for the general linear group (cf. Weyl [2]). These will be proved by
making a systematic use of standard bases as introduced by Hodge and
others. In the same vein, we shall next prove the more recent result, due to
Eagon-Hochster, asserting that generic determinantal varieties are
Cohen-Macaulay.
§3. The Ideal of a Generic Determinantal Variety 71

Departing from tradition we shall prove the second fundamental theorem

of invariant theory first. Using the notation of the previous section this can
be stated as follows.

Second Fundamental Theorem of Invariant Theory. The ideal of Mk in the

affine coordinate ring of M is generated by the (k + 1) x (k + 1) minors.

We shall denote the ideal generated by the (k + 1) x (k + 1) minors by

the symbol Ik+l· From the very definition of Mk it is clear that the support
of the variety defined by I k + 1 isMk. Therefore it suffices to prove the following.

Proposition. For any k the ideal Ik is equal to its radical.

The proof of this fact relies on a detailed analysis of the homogeneous

coordinate ring of the Grassmannian G(k, N) under the Plucker embedding.
This ring will be denoted by the symbol R.
To set up, to any N x k, rank k matrix A we associate the point of G(k, N)
corresponding to the vector subspace of ICN spanned by the columns of A.
Given a matrix

of indeterminates, the Plucker coordinates for G(k, N) are the k x k minors

of this matrix. Given indices i 1, ... , ik, with 1 ~ ii ~ N, for any j, we shall
denote by

the determinant of the k x k matrix whosejth row is the i}hrowof(X.,,J We

also introduce the following notation:

S(al> ... , at) = Symmetric group on the letters a 1 , ••• , at.

The first basic results we need are the famous

Pliicker Relations. Let s ~ k, and set

S' = S(i., ... , ik) X S(jl> ... ,js) C S.

Then

L e(a)[i1 · · · is-1a(is) · · · a(ik)] · [a(j1) · · · a(js)js+ 1 · · · jk] = 0.

ifeS/S'
72 II. Determinantal Varieties

Proof First observe that the sum is well defined since each summand de-
pends only on the class (j of (J modulo S'. Writing

the above sum may be viewed as an alternating (k + 1)-form in v;,, ... , v;.,
ov -- 0, tt
f\k + 1 tf'k
vil, ... , vj.,· smce
· . must vams . 11y. Q.E.D .
. h 1"d enttca

Using the Plucker relations we are now going to exhibit an explicit basis
for R, the homogeneous coordinate ring of G(k, N), as a complex vector
space (actually the same computation works over the integers, too).
Given a monomial

we shall find it convenient to write it in matrix notation as

Such a monomial will be said to be standard 1 if the rows are increasing, i.e.,

i11 < it 2 < ··· < itk for any t,

and the columns are non-decreasing, i.e.,

for any s.

Lemma. The standard monomials generate R as a complex vector space.

1 The reader will notice the slight ambiguity in this terminology. To be precise we should

have defined standard tableaux and called standard the monomials associated to such tableaux.
For our modest purposes, however, such a subtle distinction would prove more of a hindrance
than an advantage.
§3. The Ideal of a Generic Determinantal Variety 73

To prove this we begin by ordering monomials lexicographically. First

of all we order Plucker coordinates as follows. We say that

if there exists s such that 1 :::;; s :::;; k and

This is a total order on the Plucker coordinates, with minimal element

..
[1 · · · k]. We then order monomials lexicographically, i.e.,

[ i~l: ... i~kl

: < :
[j~l j~k]
:
ihl ihk jhl jhk

if there is t such that 1 :::;; t :::;; h and

Uvl · · · ivk] = Uvl · · · ivk] if V < t,

[itl · · · irk] < Utl · · ·irk].

To prove our lemma is clearly suffices to show that any non-standard mono-
mial can be expressed as a linear combination of monomials which precede it
in the lexicographical order. Indeed it suffices to check this for a non-standard
monomial of degree two

[h~1 . ••
... 1k
~k]·

Clearly, up to a change of sign, we may assume that the rows are increasing.
Now, to say that this monomial is non-standard means that there is an s
such that

But now the Plucker relation

expresses our monomial as a linear combination of "smaller" ones. Our

second basic step is

Lemma. The standard monomials give a basis of R as a complex vector space.

74 II. Determinantal Varieties

We shall in fact prove a more general result. Consider the flag

where F; is spanned by the first i vectors in the canonical basis of CN. Con-
sider also the Schubert varieties

V(;, ... ;kl ={WE G(k, N): dim W n F;. z s, 1::; s::; k}.

Clearly, Vu, ... ikl is contained in V(;, ... ikl if and only if i 1 z j 1, ... , ik z jk.
Moreover,

Vi1-··kl = {Fk},
ViN-k+ 1 ... N] = G(k, N).

We shall denote by R 1;, ... ikl the homogeneous coordinate ring of V(;, ... ikJ·
We will prove our lemma by proving more generally the following:

Lemma. Those standard monomials T such that T · [i 1 • • · ik] is not standard

form a basis for the ideal of V[;, ... ikl· Thus the restrictions to V[;,···ikl of those
standard monomials T such that T · [i 1 • • • ik] is standard form a basis for
R[i,···ikJ·

Proof A simple computation shows that if

[~1
11
••.
. . .
~k]
!k

is non-standard then [s 1 • · ·sk] vanishes identically on V(;, ... ;k1. Therefore

our lemma will be proved if we can show that the restrictions to V(;, ... ikl of
those standard monomials T such that T · [i 1 • • • ik] is standard are linearly
independent. We proceed by induction. The statement is trivial for V( 1 ... kJ·
Assume it is true for all Schubert varieties contained in V(;, ... ikl and suppose
we have a relation

A simple computation shows that [i 1 • • • ik] does not vanish on the irreducible
variety V(;, ... ik]· We may therefore assume that one of the monomials T,,
say T1 , is of the form

... i~k]
. '
irk
§3. The Ideal of a Generic Determinantal Variety 75

with

By the induction hypothesis T1 does not vanish on V[ir, ... irk]· On the other
hand, we know that ~vanishes on V[ir, ... irkl if~· [i, 1 · · · i,k] is not standard.
Thus

where the ~. are those among the ~ such that ~ · [i, 1 · · · i,k] is standard.
Since T1 is one of these monomials, the above is a non-empty relation. This
contradicts the induction hypothesis. Q.E.D.

We are now in a position to prove the proposition and therefore our

theorem. To each matrix (ai) EM we may associate the element of
G(n, n + m) spanned by the columns of

0 1

1 0

This yields on isomorphism between M and the open subset of G(n, n + m)

where [m + 1 · · · n + m] does not vanish. The affine coordinate ring C[xii]
of M is therefore isomorphic to a quotient of R, namely

C[xii] = Rj([m + 1 · · · m + n] ± 1).

Thus a basis for C[xii] is given by all the standard monomials

[: : :J
such that [ih 1 · · · ihnJ #- [m + 1 · · · m + n]. On the other hand, any s x s
minor of (xii) is a Plucker coordinate. More precisely, let us consider the
minor constructed out of rows i 1 , . . . , i. and columns r 1 · • · r•. It is straight-
forward to see that this minor is nothing but the Plucker coordinate
76 II. Determinantal Varieties

where {is+l• ... , in} is the complement in {m + 1, ... , m + n} of

{n + m- r. + 1, ... , n + m- r1 + 1}.
Therefore a basis of the determinantal ideal Ik is given by all standard
monomials

[ i~l. ··· i~n]

. '
ill itn

such that i 1 k ::::;; m and [i11 • • • i 1m] =F [m + 1 · · · m + n]. Reciprocally, a basis

for the quotient ring Dk = C[xuJ/h consists of all those standard monomials

[ i~l. ··· i~n]

. '
itl itn

such that ilk > m and [i11 • • • imJ =F [m + 1 · · · m + n].

To conclude the proof of the proposition suppose

where the T, are standard monomials not belonging to Ik. We may and will
assume that T1 is the maximum of the 7;, in the lexicographical order. But
then

where L stands for a linear combination of monomials which precede T[ in

the lexicographical order. Using the Plucker relations we can rewrite Las a
linear combination of standard monomials which also precede T[. Since T[
is a standard monomial which does not belong to h, the only way out is that
a 1 = 0. Repeating this argument, one shows that all the ah must be zero. This
concludes the proof of the proposition and hence of the theorem.

We now turn to the first fundamental theorem of invariant theory. The

general linear group G = GL(k, C) acts on

M(m, k) x M(k, n)

by

g(A, B) = (Ag-1, gB).

§3. The Ideal of a Generic Determinantal Variety 77

Also, matrix multiplication yields a map

M(m, k) x M(k, n) ~ M(m, n).

Clearly, the image of this map is Mk(m, n). Let S be the coordinate ring of
M(m, k) x M(k, n) and D = Dk+l the coordinate ring of Mk(m, n). We want
to prove the

First Fundamental Theorem of Invariant Theory. The ring D is the ring of

invariants ofS under the action of G. In symbols

Corollary. The variety Mk is normal.

To prove the corollary observe that since M(m, k) x M(k, n) is smooth,

it is enough to show that D is integrally closed in S. Suppose a E S satisfies a
relation of integral dependence over D; then all of its translates by the action
of G satisfy the same relation. Hence there is a finite number of these. But G
is connected so that a must be G-invariant.
We now proceed to prove the first fundamental theorem. First of all
notice that, if the theorem holds for a given choice of m and n, it also holds
for any smaller values of m and n. We may therefore assume that m and n
are large, in particular, larger than k. We shall often write pairs of
matrices (A, B), of size m x k and k x n, respectively, in block form as
follows:

where A' and B' are k x k matrices, and we set

d(A, B)= det(A'B').

We need the following:

Lemma. D[l/d] = S[l/d]G.

Proof One inclusion is trivial. We denote by W the open subset of M(m, k) x
M(k, n) where d does not vanish. We also denote by V the linear space of
those pairs (A, B) such that A' is the identity matrix. The group G acts
on G x Vby left multiplication on the first factor. We then have a G-in variant
isomorphism
G X v --dV,
(g, u) ~ gu,
78 II. Determinantal Varieties

an explicit inverse of which is given by

Clearly the ring of invariants of G · x Vis just the coordinate ring of V. Hence
any G-in variant regular function f on W, i.e., any element of S[1/d] 6 , is a
regular function of A" A'- 1, A'B', A'B". But

A"A'- 1 = ~ A"B'
d '

proving thatfE D[1jd]. Q.E.D.

To conclude the proof of the first fundamental theorem we consider a G-

in variant element a E S. By the lemma we just proved that a belongs to
D[1/d], and therefore there is an integer k such that dka ED. Thus it suffices
to show that, wheneverfE Sand dfE D, thenfE D. We will do this by using
the explicit basis for D constructed in the proof of the second fundamental
theorem. Recall that we have identified M(m, n) with an open subset of
G(n, m + n) by associating to a matrix (a;) the n-plane spanned by the
columns of

0 1

1 0

We then have

d = [1 · · · k m + 1···m + n - k].

Recalling the standard basis we constructed for D in the course of the proof
of the second fundamental theorem, let us now write

df= I il(;T;,

..
where each T; is a standard monomial of type

[ i~l. i~·]
. '
itl itn
§3. The Ideal of a Generic Determinantal Variety 79

such that i 1 ,k+ 1 > m and [i, 1 · · · i,.] # [m + 1 · · · m + n]. In particular, the
regular function L cxi T; vanishes on the hypersurface d = 0 and therefore
there is a positive integer h such that

where y ED. Now write y as a linear combination of standard monomials

We may assume that T1 is the maximum among the T;'s in the lexicographical
order. We may then write, as in final step of the proof of the second funda-
mental theorem,

(L cxi T;)h = cx1 Tf + L

= L pj dSj,
where L stands for a linear combination of standard monomials which
precede the standard monomial Th in the lexicographical order. Since dSi
is a standard monomial for any i, we conclude that Tt and hence T1 , is
divisible by d. Repeating this argument shows that all the T;'s are actually
divisible by d. Therefore

belongs to D. This concludes the proof of the first fundamental theorem of

invariant theory.

The last results of this section are due to Eagon-Hochster and to

Hochster, Laksov, and Musili. Our proof will closely follow Musili's.

(3.1) Theorem. Mk is Cohen-Macaulay.

As we have already done, we identify a matrix (aij) EM with the point of

+ m) spanned by the columns of the matrix
G(n, n

0 1

0 1
1 0 0
80 II. Determinantal Varieties

In this fashion M is identified with one of the standard coordinate patches of

G(n, n + m). Let now F be the span of the last n vectors in the canonical
basis of cn+m. The determinantal variety Mk is described inside M by the
Schubert condition

Mk = {AEM: dim(A n F);;::: n- k},

so that M k can be thought of as an open subset of a Schubert variety. Thus our

theorem follows from the following more general one.

(3.2) Theorem. The cone over a Schubert variety is Cohen-Macaulay. In fact,

if V is a union of Schubert varieties of codimension one in a fixed Schubert
variety, the cone over Vis Cohen-Macaulay.

This theorem will be proved by a double induction on the dimension and

number of components of V. We fix the flag

F 1 c F 2 c ... c F N = eN,

where Fi is spanned by the first i vectors in the canonical basis of eN. We

denote by V[i, ... ikl the Schubert variety of those points Win G(k, N) such that

dim(W n F;.);;::: s, s = 1, ... ' k,

and by R[i, ... ikl the homogeneous coordinate ring of V[i, ... ikl' as we have
already done in the proof of the second fundamental theorem of invariant
theory. As we observed there, the ideal of V[i, ... ikl is generated (as an ideal) by
those Pliicker coordinates [j 1 · · · jk] such that

[~111 . . .
...
~k]
lk

is not standard, that is, such that i 1 < j 1 for some l.

We begin our inductive proof ofthe theorem by noticing that its conclusion
is trivially true for V[ 1 ... k1, which consists of a single point. We next fix a
Plucker coordinate

and set

! 5 = [i 1 · · ·is- 1 is - 1 is+ 1 · · · ik],

r.,=[i1···is-1 i.-1 is+1"""it-1 i,-1 it+1···ik].
§3. The Ideal of a Generic Determinantal Variety 81

Clearly the Schubert varieties of codimension one in V. are exactly the varieties
V.•. Since the ideal of each V.. is generated by standard monomials, the ideal of
U~= 1 V.. is also generated by standard monomials, and hence is generated
(as an ideal) by Plucker coordinates. The coordinate [j 1 · · ·jk] belongs to the
ideal of V.. if and only if it belongs to the ideal of V. or else j. = i •. Thus
[j 1 · · · AJ belongs to the ideal of U~= 1 V.)f and only if it belongs to the ideal of
V., or else it equals r; in particular, R</(r) is the homogeneous coordinate
ring of U~ = 1 V. •. We also notice that

We would like to prove that the cone over any union of Schubert varieties
of codimension one in V. is Cohen-Macaulay, assuming the analogous
statement to be already proved for all Schubert varieties of dimension lower
than that of V.. We begin with a single Schubert variety V..· By one of our
previous remarks, the ring R<)(r .) is the homogeneous coordinate ring of the
union of all Schubert varieties contained in V. •• hence it is Cohen-Macaulay
by induction hypothesis. Since '• is not a zero divisor, R<. is also Cohen-
Macaulay. We next consider

v = v.•• u ... u v..h.

We may assume, inductively, that the cone over

is Cohen-Macaulay. Notice that

is a union of Schubert varieties of codimension one in V..h' and hence the cone
over it is Cohen-Macaulay by induction hypothesis. Denoting by R', R", R 111
the homogeneous coordinate rings of V, V', V n V..h, respectively, we have
an exact sequence of R-modules

0 ---+ R' ---+ R" EfJ RTsh ---+ R 111

---+ 0.

Since the image of '•h in R" EB R<•h is not a zero divisor, part (c) of the following
elementary result implies that R' is also Cohen-Macaulay, finishing the proof
of our theorem.
82 II. Determinantal Varieties

Lemma. Let

be an exact sequence of modules over a Noetherian ring A. Let fo, ... , fd be

elements of A. Then:
(a) If f 0 , . . . , fd is a regular sequence forM and N, it is a regular sequence
for L.
(b) If f 0 , .•. , !tJ is a regular sequence for L and N, it is a regular sequence
forM.
(c) Iff0 , .•. , fd is a regular sequence for M,f1 , ... , fd is a regular sequence
for Nand f 0 N = {0}, then f 0 , ..• , fd is a regular sequence for L.

Proof. When d = 0, (a) is obvious; to prove (b), suppose f 0 x = 0, for some x

in M. Then a(x) = 0 by hypothesis, therefore x ELand, again by our assump-
tion, X = 0. Now, to prove (a)and (b) by induction ond,it suffices to show that

0--+ Llf0 L 4 M/f0 M ~ N/f0 N--+ 0

is exact, that is, that {3' is injective. To do this, suppose x is an element of L

such that {3(x) belongs to f 0 M. Thus there is a yin M such that {3(x) = f 0 y.
Therefore f 0 a(y) = 0, a(y) = 0, y belongs to L, and x belongs to f 0 L.
Part (c) of the lemma now follows from parts (a) and (b). Clearly

Tod(M, A/f0 A) = 0; N/f0 N ~ N.

Moreover, from the long exact sequence of Tor's for N and

0--+ A/1 ~A--+ A/f0 A--+ 0,

where J is the annihilator of fo, we deduce that

Tod(N, A/f0 A) ~ NjJN = N.

In fact, since fo is not a zero divisor in M, JM = 0, and hence JN = 0. Thus

we deduce from our original exact sequence another exact sequence

0--+ N--+ Llf0 L--+ M/f0 M 4 N--+ 0.

Denoting by K the kernel of y, we conclude, by (a), that f 1 , ••• , ftJ is a regular

sequence for K and, by (b), that it is regular for L/ fo L as well. Since fo is not
a zero divisor in L, this concludes the proof of the lemma.
§4. Determinantal Varieties and Porteous' Formula 83

§4. Determinantal Varieties and Porteous' Formula

Generic determinantal varieties are the prototypes of what one calls a
determinantal variety. We proceed with a formal definition. Let

¢: E-+ F

be a homomorphism between holomorphic vector bundle of ranks n, m over

an analytic space X. Upon choosing local trivializations of E and F over an
open set U, the homomorphism¢ is represented by an m x n matrix A of
holomorphic functions. This corresponds to a morphism

f: U-+ M = M(m, n).

We denote by Uk the preimage of Mk = Mk(m, n). In other terms, the ideal of

Uk is generated by the (k + 1) x (k + 1) minors of A. It is immediate to see
that U k does not depend on the choice oftrivialization, and therefore there is a
well-defined analytic subspace

such that Xk(¢) n U = Uk for any U. The variety Xk(¢) is what we call the
kth determinant a/ variety or kth degeneracy locus associated to ¢: it is sup-
ported on the set

{p E X: rank(¢ p) ~ k}.

It is clear from the definition that Xk(¢), when non-empty, has codimension
at most (m - k)(n - k).
The generic determinantal variety M k is nothing but the kth determinantal
variety associated to the morphism

where

It is convenient to consider, beside the variety X k( ¢),the analogue Xk( ¢)of

the desingularization Mk of M k defined in the previous section. This can be
intrinsically defined as follows. We let

n: G(n - k, E)-+ X

be the Grassmann bundle of (n - k)-planes in the fibers of E. We also denote

by S and Q, respectively, the universal subbundle and quotient bundle on
84 II. Determinantal Varieties

G(n - k, E). Thus if xis a point of X and W is an (n - k)-plane in Ex, i.e., a

point of n- 1 (x), the fibers of Sand Q over Ware W itself and Ex/W, respec-
tively. The bundles S and Q sit in an exact sequence

0 ~ S ~ n* E ~ Q ~ 0.

Composing the lifted homomorphism

n*(¢): n*E ~ n*F,

with the inclusion of S in n* B gives a homomorphism

r/}:s~n*F.

We shall denote by Xk(¢) the subvariety ofG(n - k, E) defined by the vanish-

ing of$. Thus the suppo,rt of Xk(¢) is simply the set of all couples (x, W),
where xis a point of Xk and W is an (n - k)-plane contained in the kernel of
¢x· It is a simple exercise in words to check that, if U and fare as in the
definition of Xk(¢), then Xk(¢) n n- 1 (U) is nothing but the fiber product
Mk X MU. In what follows we shall often write xk, xk instead of Xk(¢), Xk(¢),
when no confusion is likely.
The theorem stating that generic determinantal varieties are Cohen-
Macaulay (Theorem (3.1)) implies that the varieties Xk are also Cohen-
Macaulay, provided they have the" correct" codimension. More precisely, we
have:

(4.1) Proposition. Let X be a complex manifold and let

be a homomorphism ofholomorphic vector bundles ofranks nand m, respectively.

If Xk(¢) has codimension (m- k)(n - k), it is Cohen- Macaulay.

It follows, in particular, that ifthe hypotheses of the above proposition are

satisfied, then Xk(¢) has no embedded components. The proposition is an
immediate consequence of Theorem (3.1) plus the following simple algebraic
statement.

Lemma. Let f: X~ Y be a morphism between complex manifolds. Suppose Z is a

Cohen-Macaulay analytic subspace of Y of pure codimension k. Then, iff- 1 (Z)
has pure codimension k, it is Cohen-Macaulay.

Proof. Let r c X X y be the graph of f. Projection onto X induces an

isomorphism between rand X, and f- 1 (Z) is just the projection of the sub-
space Z' of r cut out by X x Z. On the other hand, since X is smooth, X x Z
§4. Determinantal Varieties and Porteous' Formula 85

is Cohen-Macaulay, while r is locally a complete intersection, by the smooth-

ness of Y. In fact, if x 1 , ••• , xn are local coordinates for X and y 1 , ... , Ym local
coordinates for. Y, the equations of r are, locally,

i = 1, ... ,m,

where (!1 , ... , j~) is a local coordinate expression of f. Now our hypothesis
on the codimension off- 1 (Z) implies that

codim(Z', X x Z) =dim X+ dim Y- k- dim(Z')

= n + m - k - (n - k)
= m.

Thus Z' is of pure codimension m and locally defined by m equations in the

Cohen-Macaulay space X x Z. This implies that Z', and hence f- 1 (Z), are
Cohen-Macaulay as well (cf. Matsumura [1, pp. 107-108]). Q.E.D.

The main purpose of this section is to prove a general formula, due to

Porteous [1] and Kempf-Laksov [1] that expresses the fundamental
classes of the determinantalloci of any holomorphic bundle mapping E --+ F
as a polynomial in the Chern classes of the bundles E and F. To explain this we
first recall a few notations from the theory of Chern classes. In our discussion
we shall use integral cohomology throughout; however, it is important to
notice that, in an algebraic setting, Porteous' formula and its proof go through
verbatim if one uses, instead, Chern classes with values in the Chow ring of
algebraic cycles modulo rational equivalence.
For any complex vector bundle F over a space Y,

is the Chern polynomial of F. By the Whitney product formula, Cr extends to a

homomorphism from the Grothendieck group K(Y) to the multiplicative
group of the invertible elements in the power series ring H*(Y) [[t]]. We shall
sometimes improperly write - F to denote the negative of the class ofF in
K(Y). Then we have

Similarly, we write E - F for the difference of the classes of E and Fin K(Y).
We are now in the position to explain Porteous' formula. Let E and F be
holomorphic vector bundles of respective ranks n and m over a complex
manifold X. If
86 II. Determinantal Varieties

is any holomorphic bundle mapping, we denote by xk the fundamental class of

X k( ¢ ). Porteous' formula expresses xk in terms of the Chern classes of E and
F. To state it, we introduce a piece of notation. For any formal series

+oo
a(t) = L akt\
k=- 00

we set

We then have

(4.2) Porteous' Formula. Under the assumption that Xk(¢) is empty or has the
expected dimension dim X- (n- k)(m- k), we have

xk = ilm-k,n-kCcr(F - E))
= ( -l)(m-k)(n-k)fln-k,m-k(cr(E - F)).

As we already announced, when X is a smooth quasi-projective variety,

and ¢ is a homomorphism of algebraic vector bundles, Porteous' formula
remains valid in the Chow ring of X. We also remark that, by an argument
using universal bundles, it is not difficult to establish the existence of a
formula expressing xk as a polynomial P(c(E), c(F)) in the Chern classes of E
and F; what requires effort is determining this polynomial. A hint is provided
by Giambelli's formula from Schubert calculus (cf. page 205 of Griffiths-
Barris [1]) which gives Porteous' formula in case F is trivial. We will
concern ourselves first with finding the polynomial P and then, in the final
discussion, consider what exactly has been proved.
From this point of view, the difficulty in establishing Porteous' formula is
primarily technical, and readily overcome. The Chern classes of a vector
bundle E ~ X describe the loci where one or more sections of X will be
linearly dependent, and in some cases this is sufficient to solve our problem:
i.e., the locus X 0 is the zero locus of¢, considered as a section of Hom(E, F), so
that

Xo = Cmn(E* (8) F);

and, in case m = n, we have that Xm-l is the zero locus of 1\m¢, so that
Xm _ 1 = C1 ( / \m E* ® 1\m F)
= c 1(F)- c 1(E).
§4. Determinantal Varieties and Porteous' Formula 87

The problem is that, in general, X k is not so readily expressed as such a locus-

or, rather, since we can always write

that, as a section of 1\k+l E* ® 1\k+l F, J\k+ 1 4> vanishes in the wrong

codimension.
The solution to this problem is straightforward and elegant; we must ask
not only for the fundamental class of X k, but also for the class x of Xk in
G(n - k, E). Keeping the notations introduced at the beginning of this
section, recall that Xk is just the zero locus of the bundle homomorphism ¢,
which we may view as a section of Hom(S, n* F). Accordingly

X= Cm(n-klS* ® n*F),

and we may compute the class of Xk as

where
n*: H*(G(n- k, E), Z) ~ H*(X, Z)

is the Gysin homomorphism.

Our work is now clearly cut out for us: in the following two discussions,
we derive a formula for the top Chern class ofthe tensor product of two vector
bundles; and in the third we apply this formula to the bundles S* and n*F on
G(n - k, E), and evaluate the image of the class we get under the Gysin map
n* to arrive at Porteous' formula.

(i) Sylvester's Determinant

We recall here a classical formula for the resultant of two polynomials. If k is

a field, p(t) = Lf=o p;t; and q(t) = L7'=o q;t; are two polynomials with
coefficients ink, then p and q have a common root if and only if the (m + n) x
(m + n) determinant

Po P1 Pn 0 0
0 Po P1 Pn 0 0

0 0 Po Pn
R(p, q) =
qo qm 0 0
0 qo qm 0 0

0 0 qo qm
88 II. Determinantal Varieties

vanishes. The proof is clear: recognizing the rows of this determinant as the
coefficients of the polynomials p, tp, ... , tm- 1p, q, tq, ... , tn- 1q, we see that
if p and q have a common root, then these polynomials do not span the
(m + n)-dimensional vector space of polynomials of degree not exceeding
m + n - 1, and hence must be linearly dependent; while a linear relation
among the rows says that there exist polynomials a(t), b(t) of degrees m - 1
and n - 1 such that
a(t)p(t) + b(t)q(t) = 0;

since a(t) can vanish at at most m - 1 of the roots of q, p(t) must vanish at the
remaining one(s). Alternatively, we can argue that the first m and last n rows
of this matrix generate the subspace of pm+n- 1 spanned by the roots of p
and q, respectively, on a rational normal curve C; since any m + n points
on such as curve are independent, two such secant planes can fail to span
pm+n- 1 only if they have a point of C in common~that is, if p and q have
a common root.
We note that, if p0 = 1, the above determinant can be reduced somewhat
by multiplying on the right by the unipotent matrix

Sm+n-1
Sm+n-2

0 0

where I s;ti = 1/p(t); we obtain that R(p, q) equals

0 0
0 1

0 0 1 0 0
ro r1
0 ro

0 ro
rm rm+n-1

= L'!m,n(q(t)jp(t)),

where I r;ti = q(t)/p(t).

§4. Determinantal Varieties and Porteous' Formula 89

We can also exchange the first m and the last n rows of our determinant to
find
J1.m,n(q(t)jp(t)) = ( -l)mnt,.n,m(p(t)jq(t));

the analogous operations in general yield the formula

for any monic power series f(t).

We note also that the polynomials t.m,n provide a criterion for the ration-
ality of a power series: f(t) = I~o a;ti is the power series of a rational
function if and only if

for all sufficiently large m and n.

(ii) The Top Chern Class of a Tensor Product

We can rephrase the above discussion in terms of symmetric functions, as

follows. If oc 1 , . . . , ocn and {3 1 , ... , !3m are indeterminates, p 1 , ... , Pn, and
q 1 , •.• , qm their elementary symmetric functions, so that
n
p(t) = 1+ LP;ti = TI (1 + OC;t),
i= 1

I q;ti = TI o + /3;t),
m

q(t) = 1 +
i= 1

then the identity

fl ({3j - 0:;) = 11m, n(q(t)/p(t))

i,j

holds. In fact, we already have

TI ({3j- oc;) = A(pb ... ' Pn, q1, ... ' qm)
i,j

for some polynomial A of weighted degree mn; since t.m,n(q(t)/p(t)) is such a

polynomial vanishing whenever A does, it follows that

J1.m,n(q(t)jp(t)) = AA

for some scalar .A., which may be seen to be 1 by setting o: 1 = ··· = o:n = 0.
90 II. Determinantal Varieties

This statement in turn gives us directly the

Lemma. If E and Fare complex vector bundles on a space X, of ranks n and m

respectively, then

Cmn(E* ®F)= llm,n(cr(F)jc,(E))

= ( -l)mnlln, m(c,(E)jc,(F)).

Proof We may assume that E and F split into direct sums of line bundles:

E= $L;,
F = $M;,

so that

c,(E) = fl (1 + Ct;t); c,(F) = fl (1 + [3;t).

We have

E* ® F = (J} Lf ® Mi
i,j

and correspondingly

CmnCE* ®F) = TI cl(Lt ® Mj)

i,j

= TI Cf3j - a;)
i,j

= llm,n(c,(F)jc,(E)). Q.E.D.

(iii) Porteous' Formula

We now have all the tools necessary to derive Porteous' formula; all we
require to make the computation is a little luck.
As in the definition of the loci Xk( ¢ ), we let

n: G(n - k, E) -+ X

be the Grassmann bundle of(n - k)-planes in the fibers of E, and denote by S

and Q the universal subbundle and quotient bundle on G(n - k, E), re-
spectively. As we already observed we have

xk = n*(cm<n-kj(S* ® n:*F))
= n* llm,n-k(n*c,(F)/c,(S)),
§4. Determinantal Varieties and Porteous' Formula 91

by our formula for the top Chern class of a tensor product. To express this
quantity in terms of the Chern classes of E and F, we need to notice that, for
any monomial f1 C;(QY' we have

The above direct image is a multiple of the fundamental class of X when

L icx; = k(n - k), while the Gysin images of monomials in C;(Q) of(weighted)
degree greater than k(n - k) are, of course, more complicated; and here is
where the luck comes in: since

c(S)c(Q) = n*c(E),

we may write

and now we notice that, inasmuch as the entries of the (n - k) x (n - k)

determinant A are all linear combinations, with coefficients in H*(X, Z),
of the Chern classes c0 (Q), ... , ck(Q), by our previous remark, only those terms
in the entries of the determinant A which contain the factor ck(Q) will contribute
to its Gysin imagen* A. Thus, by the push-pull 1 formula we have

and, to conclude, we only have to show that

Since Q restricts to the universal quotient bundle on each fiber of n, this will
follow from the

Lemma. If Q is the universal quotient bundle on the Grassmannian G(n-k, n),

then

1 We recall the content of the push-pull formula. This states that, for any map f: X --> Y

such that f* is defined, and for any integral cohomology classes IX on X and fJ on Y, one has:
fiiX. f*fJ) =MIX). {J.
92 II. Determinantal Varieties

Proof This is a fairly standard result in the Schubert calculus of the Grass-
mannian. In fact the integral on the left is just the number of zeros common
to n - k general sections of Q. Any vector v E C" determines a section of Q,
which vanishes precisely at those (n - k)-planes W which contain v. If we
choose n - k linearly independent vectors v1, ... , vn-k E C", the correspond-
ing sections of Q vanish simultaneously only at the span of vl> ... , vn-k·
Q.E.D.

(iv) What Has Been Proved

We want to make one observation here about what has been established by
the preceding computation. The content of Porteous' formula is that the
class

f1m-k,n-k(cr(F - E))

is supported on Xk> and agrees with the fundamental class of Xk when this
locus is empty or has the right codimension. In fact, what we have established
are the analogous statements for the locus Xk, i.e.,

In case the locus Xk is of the "right" dimension or empty, its fundamental

class is:

L1m,n-b;*c,(F- E)· cr(Q)),

and since if xk is of the right dimension, xk must also be, and the map
xk-+ xk is generically one-to-one, it follows that:

In case Xk is of the right dimension or empty, the fundamental class of Xk is:

Since X k is empty if and only if Xk is, this firmly establishes Porteous' formula
when Xk is empty, or, which is the same, shows that L1m-k,n-k(c,(F- E))
is supported on Xk. In general the problem is that, if Xk is of the wrong
dimension, our formula for its fundamental class is completely without
meaning, as is then the remainder of the calculation. But it is possible that Xk
have the wrong dimension, and Xk the right dimension, even so. This will occur,
in fact, exactly when the general fiber of xk-+ xk is positive-dimensional,
i.e., when a component of xk is contained in xk-1•
To actually prove Porteous' formula, in the sense given above, by means
of the calculation made here, one needs to use the powerful refined inter-
section theory developed by Fulton [2] and MacPherson which is
admirably suited to problems of "excess intersection." Alternatively, one
§5. A Few Applications and Examples 93

can employ a different approach altogether, working directly on X, as in

Kempf-Laksov [1]. As it stands, however, our calculation establishes
Porteous' formula only under the additional assumption that, when Xk is
non-empty, none ofits components be entirely contained in Xk-l· Incidentally,
we may notice that this extra hypothesis will always be satisfied in the cases
in which we will apply Porteous' formula. ·

§5. A Few Applications and Examples

As a first illustration of the usefulness of Porteous' formula we shall evaluate
the degree of the generic determinantal variety Mk(m, n). Since this is a
cone with vertex at the zero matrix we may, equivalently, evaluate the degree
of the projectivization V of Mk(m, n) in lPN, where N = mn - 1. Denoting
by xii• i = 1, ... , m, n,
j = 1, ... , the homogeneous coordinates in lPN, the
matrix (x;) can be viewed as a matrix with entries in H 0 (1PN, (1)(1)), and
hence gives a vector bundle homomorphism

The variety V is just the kth determinantal variety associated with ¢. Since
it has the "correct" dimension k(n + m - k) - 1, its fundamental class can
be computed by means of Porteous' formula. Replacing ¢ with its transpose,
if necessary, we may and will assume that m ~ n. Denoting by ~ the class
of a hyperplane we have

cc((l)m(1)) = cc((IJ(l)r

= o + ~tr
f
;~o
(~)~iti.
I

Therefore the fundamental class of V is

m )~m-k
( m-k ( m )~m+n-2k-l
m+n-2k-1
det

( m )~m-n+l ( m )~m-k
m-n+1 m-k
94 II. Determinantal Varieties

or, which is the same, the degree of Vis

det

To evaluate this determinant, we first replace the second column with the
sum of the first two columns, the third column with the sum ofthe second and
third columns, etc., then in the resulting matrix we replace the third column
with the sum ofthe second and third columns, the fourth column with the sum
of the third and fourth columns, and so on, until we get

( m+1 ) (m+n-k-1)
m-k+1 m+n-2k-1

= det

(
m + n-
k- 1)
m-k

1 1
(m- k)! (m + n- 2k- 1)!
n-flk-1 (m + i)! d
= et
i=O (k + i)!
1 1
(m- n- 1)! (m- k)!

= n-fr (k + (m + i)!

)
i=O i)!(m- k + i)!
1
...
...
x det ( m- k m-k+l
(m - k)(~ - k - 1) (m - k + l)(m - k)
§5. A Few Applications and Examples 95

The determinant in the last expression is nothing but the Vandermonde

determinant

' ')
1 1
...
( m-k m-k+1
det m ~ k)z (m- k + 1?

in disguise, hence its value is

n-k-l
fl [(m - k + i) - (m - k + j)] = fl i!.
n-k-l?.i>j?.O i=O

In conclusion we find the following formula for the degree of Mk(m, n):

Il
n-k-l (m+i)!i!
(5.1) deg(Mk(m, n)) = (k + i)!(m- k + i)!

Recalling the determinantal description of the tangent cones to Mk(m, n)

we gave in Section 2, the above formula also makes it possible to compute
the multiplicity of Mk(m, n) at any one of its points. The result is

(5.2) Proposition. If A is an m x n matrix of rank h :s; k, the multiplicity of

Mk(m, n) at A is

multA(Mk(m, n)) = deg Mk-h(m - h, n - h)

= fl
n-k-l (
m
_ h + z.z.
')I 'I

i=o (k- h + i)! (m- k + i)!

In this book a recurrent example of determinantal variety is provided by
the so-called rational normal scrolls. These can be introduced in three
equivalent ways. The first one has a geometrical flavor and generalizes the
classical construction of a quadric in three-space.
Take k complementary linear subspaces

L; c IP>", i = 1, .. . ,k,

with dim L; = a; and such that not all the a;'s are equal to zero. If a; =I= 0
choose a rational normal curve C; c L; and an isomorphism
96 II. Determinantal Varieties

If a;= 0, we set C; = L; and let¢; be the constant map. The variety

Xa 1 , ••• ,ak = U c/J1(t), · · ·' c/Jk(t)

te[JJ> 1

swept out by the (k- 1)-planes spanned by the corresponding points of the
C;'s is then called a rational normal scroll.
Alternatively the variety Xa,, ... ,ak may be described as the image of the
projective bundle

under the map given by <PIP'(EJ(l). Then, for each i, the image of the direct
summand <PIP'.( -a;) of E maps to the rational curve C;. It is not hard to see
that the degree of the scroll Xa,, ... ,ak is given by
k
deg X a,, ... ,ak = L ai = n -
i= 1
k + 1.

This is the smallest possible degree of an irreducible non-degenerate k-fold

IPn. Conversely, we have the following classical result

(5.3) Theorem. Any irreducible non-degenerate kjold of degree n - k + 1

in IPn is either a rational normal scroll, a cone over the Veronese surface in IPS,
or a quadric of rank greater than 4.

The reader may find a proof of this theorem in Harris [2] or, for the
case k = 2, which is the relevant one, on p.525 of Griffiths-Harris [1].
For us the interest of scrolls, at this junction, stems from the fact that they
admit a very nice and classical determinantal description which has the
advantage of clearly exhibiting their defining ideal. Assume that a 1 = · · · =
ah_ 1 = 0, a; =1= 0, i = h, ... , k. Choose in IPn homogeneous coordinates

in such a way that xgl, ... , X~l are homogeneous coordinates in L;, i =
1, ... , k. Consider the matrix

x<h) ... x<k)

ak-1 )
M - ( o
a,, ... , ak - X\h) ... X~~ .

We then have the following

Proposition. The scroll Xa,, ... ,ak is the determinantal variety whose ideal is
generated by the 2-by-2 minors of the matrix Ma,, ... ,ak·
§5. A Few Applications and Examples 97

Let Y be the determinantal variety defined by the vanishing of the

2-by-2 minors of M.,, ....••. First of all observe that the set-theoretic equality
between x.,, ....•• and Y follows immediately from the geometrical descrip-
tion of x.,, ... ,a, and from the very well-known fact that, up to a change of
coordinates, the rational normal curve

i = h, ... ,k,
is given by

x<i) . . . x<i) ) }
ci = { rank ( x~> . . . x~}- 1 ::;; 1 .

Now the set-theoretical equality we just proved implies, in particular, that

Y has the "correct" codimension. By (4.1) this implies that Y (and also the ·
cone over Y) are Cohen-Macaulay, so that Y has no embedded components.
Therefore to prove the proposition it suffices to show that the degree of Y
is equal to the degree of x.,, ... , .,, i.e., equal to n - k + 1. On the other
hand, Y can be thought of as a linear section of the generic determinantal
variety M 1 (n - k + 1, 2), and formula (5.1) exactly tells us that

deg(M 1(n - k + 1, 2)) = n - k + 1,

proving the proposition.

It may be instructive to introduce the matrix defining a scroll X =

x.,, ... ,a, in a more intrinsic way. Let L
be the restriction to X of the hyper-
plane bundle on IPn. The scroll X is ruled by a pencil of (k - 1)-planes which
we denote by lEI. We then have

h 0 (X, L( -E))= n- k + 1.
The second equality follows from the first and from the linear normality of X,
which we shall prove in the next proposition. Let us consider the multiplica-
tion map

It is then easy to show that, with a suitable choice of bases, the transpose of
the matrix M.,, ... ,a, represents the dual map

As we just mentioned, one of the basic properties of rational normal

scrolls is given by the following.
98 II. Determinantal Varieties

Proposition. Rational normal scrolls are projectively normal.

A direct proof is as follows. Let X be a rational normal scroll of degree k

in IP>n. We proceed by induction on k. The case k = 1 is well known. Given an
integer v the cohomology sequence of

shows that the projective normality of X is equivalent to the vanishing

statement

v = 1, 2, ...

Let H be a general hyperplane in IP>n; then, by Theorem (5.3) H n X is

again a rational normal scroll, therefore our vanishing statement follows,
from the induction hypothesis, by looking at the cohomology sequence of

0-> .f"x(v- 1)-> .f"x(v)-> .f"xnn(v)-> 0.

Another approach to the projective normality of scrolls is by means of

Serre's normality criterion, which reads as follows:

Serre's Criterion (cf. Serre [1, p. III-13]). A Noetherian domain R is integrally

closed ifand only if the following two conditions hold:
(i) For every prime P of height one in R, Rp is a discrete valuation ring.
(ii) The associated primes of any principal ideal in R all have height one.

We shall prove only the part of the criterion which we shall need here and
in the following chapters, namely that (i) and (ii) imply that R is integrally
closed. If we can show that

n
ht(P)= 1
Rp = R,

since the Rp are integrally closed, the same will be true of R. Suppose then that
a and {3 are elements of R such that a/{3 belongs to Rp for every prime P of
height one. In particular, if {3R = nQ; is a primary decomposition of {3R,
and P; stands for the radical of Q;, condition (ii) implies that a belongs to
{3Rp,, for every i. This means that there is a Y; not belonging toP; such that
ay; belongs to {3R; thus a belongs to {3R, that is, a/{3 belongs toR. This finishes
the proof.

In the context of analytic spaces, Serre's criterion can be formulated as

follows.
§5. A Few Applications and Examples 99

(5.4) Proposition. Let X be a reduced, irreducible analytic space. Then X

is normal if and only if the following conditions hold:
(i) The singular locus of X has codimension two or more.
(ii) For every analytic function g on an open subset of X, the subspace
defined by g has no embedded components.

As we did for Serre's criterion, we shall only prove that (i) and {ii) imply
normality. To use Serre's criterion we simply have to show that, for any point
x on X, and for any height one prime ideal P in the local ring R of X at x,
Rp is a discrete valuation ring. Let ffbe the conductor sheaf of the normaliza-
tion of X. Then (!)x/ff is supported on an analytic subset of X of codimension
two or more. IfF is the stalk of ff at x, this implies that F is not contained in
P, thus FRp = Rp. But FRp is the conductor of the integral closure of Rp.
This means that Rp is integrally closed, hence a discrete valuation ring, since
P has height one. This finishes the proof of Proposition (5.4).

To show that rational normal scrolls are projectively normal, we may

proceed as follows. First of all it is clear, either from the determinantal or
geometric description of a rational normal scroll X, that its singularities
occur in codimension two or more; clearly, this is also true of the cone
over X. We also remarked that the cone over X is Cohen-Macaulay. In
particular, complete intersections in the cone over X do not have embedded
components. Normality of the cone over X, that is, projective normality of
X, follows from Proposition (5.4).
Let us consider again the exact cohomology sequence of:

where X is a rational normal scroll and H is a hyperplane. Since X is non-

degenerate, we have that h0 (X, 5x(l)) = 0. On the other hand
h 1 (1Pn, 5x(l)) = 0, so that we get an isomorphism

I x(2) .§4 I x n n(2)

(here, and in the sequel, we denote by I y the homogeneous ideal of the pro-
jective variety Y and by Ir(h) its degree h summand). By taking the inter-
section of X with a general (n - k)-plane we then see that the dimension of
Ix(2) equals the number of linearly independent quadrics in IPn-k passing
through n - k + 1 points in general position. Hence

.
dtm Ix(2) = (n- + 1)
k
2 .

Combining this with the determinantal description of X we obtain the

following
100 II. Determinantal Varieties

Theorem. Let X be a k-dimensional rational normal scroll contained in IP'".

Then there exists a matrix of linear forms in IP'":

ln--k+1)
1~-k+ 1

such that the ideal lx is generated by the 2-by-2 minors of M. Moreover the
quadrics
("- ~ + 1 )

rx < [3, rx, f3 = 1, ... , n - k + 1,

are linearly independent and form a basis of I x(2).

Bibliographical Notes
Lemma (1.3) appears in Kempf [2]. The approach to the first and second
fundamental theorems of invariant theory has been suggested to us by C.
De Concini. Both theorems appear in Weyl's book [2]. A modern treatment,
ialid also in characteristic p, can be found in De Concini-Procesi [1], or
else in De Concini-Eisenbud-Procesi [1]. Fundamental papers in the
theory are the one by Hodge [1] and the one by Doubilet-Rota-Stein [1].
The first proof of the Cohen-Macaulay property of determinantal varieties
is to be found in Eagon-Hochster [1]. The analogous result for Schubert
varieties is due to Musili [1], Laksov [1], and Hochster [1]. For Porteous'
formula we refer to the original paper of Porteous [1] and to Kempf-
Laksov [1], where it is proved under less restrictive hypotheses. Finally,
a discussion of rational normal scrolls can be found in Harris [2].

Exercises

A. Symmetric Bilinear Maps

We shall use the following notations:

IP'N = ifl'<" + 1Hn + 2li 2 - 1 is the projective space of symmetric quadratic forms on
en+ 1;
Q will denote either a point in IP'N thought of as a quadric in IP'", or a point in
c<n+ 1Hn+Z)/Z - {0} thought of as a symmetric linear map from e+ 1 to its dual (the
context will make clear which interpretation to use);
~ c IP'N is the determinantal variety of forms of rank :-:;; k;
Gk is the Grassmannian of (n - k)-planes in IP'", or equivalently the Grassmannian
of (n - k + I)-planes in e+ 1 ;
Exercises 101

A E Gk will denote either a [p>•-k in IP'" of the corresponding cn-k+ I in c•+ I (again
the context will make clear which interpretation to use);
for Q E PN, ker Q will be either the kernel of the symmetric map

or else the projection of this linear space to IP' •;

is the incidence variety with indicated projections.

A-1. Show that X k is irreducible and smooth; and that

dim xk = k(n +1- k) + (k +

2 1) - 1.

A-2. Show that n 2 is biholomorphic over l-Ie - l-Ie-band from this conclude that

codim l-Ie = (
n-k+2) .
2

A-3. In this exercise we will identify the tangent spaces to Xk, and for this we use the
identifications (cf. Section 2 of Chapter II)

TA(Gk) ~ Hom(A, 1["+ 1/A),

TQ(IP'N) ~ cN+I;cQ,

where CQ c eN+ I is the line over Q E IP'N. Points in 'I(A, Ql Gk X IP'N) will be denoted
by ( ¢, R) where ¢ E Hom(A, e+ I I A) and R E i[N +I /ICQ. Show that

1(A,Qj(Xk) = {(c/J, R): R(v, w) + Q(c/Jv, w) = Ofor all VEA, wEIC"+ 1}.

A-4. Show that the projectivized tangent cone to l-Ie at Q E v; - VI- 1 has support given by
supp IP'YQ(l-Jc) = {R:rank(RikerQ)::::;; /- k}

and has multiplicity one.

A· 5. A point Q E l-Ie - l-Jc_ 1 may be pictured as a cone over a smooth quadric in IP'k-l
with vertex A ~ [p>•-k where A = ker Q. Show that

where tJ·,e right-hand side is the vector space of quadrics in IP'" that contain A.
102 II. Determinantal Varieties

B. Quadrics

In this sequence of exercises we will adopt the following notations:

Vis an m-dimensional complex vector space and

Q:VxV->C

a non-degenerate bilinear form;

we denote by

Q: V--> V*

the isomorphism corresponding to Q;

a k-plane A c Vis isotropic for Q if

QIAxA = 0;

G(k, V) = Gk denotes the Grassmannian of k-planes in V and

the locus of isotropic subspaces (we write :Ek(Q) if there is a possibility of confusion).

In Exercises B-1 through B-7 we will assume Q to be symmetric.

B-1. Show that :Ek is empty fork> m/2 and non-empty fork= [m/2]. Show also that
fork' < k ~ [m/2] every k' -dimensional isotropic subspace lies in a k-dimensional
isotropic subspace.

B-2. Making the identification

show that, fork ~ [m/2], :Ek is smooth of codimension e; l) in Gk with tangent

space given by

TA(:Ek) = {</> E Hom( A, VjA): Q(</> · v, w) + Q(v, </> · w) = 0 for all v, wE A}.

B-3. Show that, if k < m/2, the locus :Ek is irreducible.

(Hint: If v E V is an isotropic vector for Q, set W = Q(v).L/C · v (thinking
of Q(v) as a vector in V*) and note that Q induces a non-degenerate form Qw on W
such that

Now use induction on k.)

Exercises 103

B-4. Denote the automorphism group of Q by

Aut(Q) = {A E GL(V): Q(Av, Aw) = Q(v, w) for all v, w}.

Note that Aut(Q) acts on l:k, and show that this action is transitive (i.e., for
A, A' E l:k there exists A E Aut(Q) such that

AA =A';

note that this proves that l:k is smooth).

B-5. Keeping the notations of the preceding exercises, assume that m = 2n is even and
k = n. Show that

dim(A n A') = n (mod 2)

if and only if
det A= + 1.
Conclude that, if m = 2n, then l:" has two irreducible components r 1, r 2 such that
the condition (*) is equivalent to A, A' being in the same component.

B-6. From the above deduce the

Corollary. For A" A2 , A3 E l:",

dim A1 n A2 + dim A 2 n A3 + dim A1 n A3 =n (mod 2).

B-7. Assuming still that m = 2n, let A0 E l:". Show that

V,.(A 0 ) = {A E l:": dim An A0 ;::o: k and dim An A0 = k (mod 2)}

is a closed subvariety of codimension k(k - 1)/2 in l:".
(Suggestion: express V,.(A 0 ) as the image of an incidence correspondence
fibered over G(k, A0 ).)

Now let V be an m-dimensional complex vector space,

Q; V X V-> I[

a non-degenerate alternating (skew-symmetric) bilinear form.

B-8. Show that m is even, and that l:k =F 0 if and only if k ~ m/2.

B-9. When k = m/2, show that l:k is smooth and irreducible of dimension k(k + 1)/2.
B-10. Let Aut(Q) be the group of linear transformations A: V-> V such that
Q(v, w) = Q(Av, Aw), for all v, wE V. Show that det A = 1 for all A E Aut(Q).
Hint: Consider Q as an element of A2 V*, and look at Q A Q A ···A Q.
m/2times
104 II. Determinantal Varieties

C. Applications of Porteous' Formula

Most of the following exercises will involve the Grassmannian G = G(2, 4) of lines in IP' 3 .
We will denote by S the universal sub-bundle on G, and by c 1 and c 2 its Chern classes.

C-1. Let L be a linear functional on C4 . Show that we may define a global section
II£ of S* by setting

for A E G(2, 4), and that these are all the global section of S*.

C-2. Using the preceding exercise, show that the classes -c 1 and c 2 are represented
by the cycles

:E 1 = {A: An A 0 =f. (0)},

and
:E 2 = {A:A c W 0 },

respectively, where A0 is a fixed two-plane in C4 , and Wo a fixed three-plane. In

particular, conclude that d = cfc 2 = 1 in H 8 (G, Z) ~ Z.

C-3. If Q is the universal quotient bundle on G, use the relation

c(S)c(Q) = 1

to express c;(Q) in terms of c 1 and c2 , and to obtain two relations on the classes
c 1 and c2 • In particular, deduce that c1 = 2.

C-4. Using the splitting principle, find the Chern classes of Sym• S* for n = 2, 3, 4, and 5.

C-5. Let FE Sym 3 (C 4 )* be a homogeneous cubic polynomial on C4 • Show that we

may define a global section Up of Sym 3 S* by

Use this and the preceding exercise to determine the number of lines on a cubic
surface.

Remark. In case the cubic surface {F = 0} is smooth-and only in this case-the

zeros of uF are all simple.

C-6. In a similar vein, let {F 1j AeiP" be a general pencil of quartic surfaces. Representing
the pencil {F.} as a bundle map

on G, use Exercise C-4 above and Porteus' formula to determine the number of
surfaces FA in the pencil that contain a line. (Answer: 320)
Exercises 105

C-7. Similarly, find the number of quintic surfaces in a general net (two-dimensional
linear system) that contain a line. (Answer: ?)

C-8. Using the same techniques as in the preceding problems, find the number of conic
curves in a pencil that contain a line.

C-9. Find the number of reducible curves in a general net of plane cubic curves.

C-10. Find the number of reducible surfaces in a general web (three-dimensional linear
system) of quadric surfaces in IP 3 • Compare this to the answer obtained by
representing the locus of reducible quadrics as the image of the multiplication map

C-11. Let~ c G(2, 4) be a curve. Show that the degree of the surface X c IP 3 swept out
by the corresponding one-parameter family of lines is given by

deg X = - ~ · c1 •

C-12. Let {FJ,ell'' be a general pencil of cubic surfaces. Using the preceding exercise,
find the degree of the surface swept out by the. lines of the surfaces F,.

D. Chern Numbers of Kernel Bundles

In the following, we will be dealing with a map of vector bundles 4>: E ---> F on a space X,
and its kth degeneracy locus X k. We will consider the case where the expected (and actual)
dimension of Xk is 1, that is, rank E = m, rank F = n, and dim X = (m - k)(n - k) + 1;
in particular, Xk+ 1 = 0. In this case the kernel and cokernel of 4> are bundles on Xk>
denoted K and K'; we will be concerned with their first Chern classes.

D-1. Let G = G(m - k, E) -.". X be the Grassmannian bundle, S ---> G the universal
sub-bundle, and
¢:s---> n*F

the induced map (cf. Section 4 of Chapter II). Assuming Xk is reduced, show that

c 1 (K) = c 1 (S) · Xk

= c 1 (S) · Cn(m-k)(n*F- S),

where xk is the zero locus of cf>.

D-2. Evaluate the product in(*) above to obtain the formula

Cn-r+l Cn-r+2
Cn+m-2r )
( Cn-r-1 Cn-r Cn+m~2r-2
c 1(K) = -det :

Cn-m+ 1 Cn-r
106 IT. Determinantal Varieties

D-3. Assuming Xk is smooth and using the tangent space computations of Section 2,
show that the normal bundle to X k in X is

N = K*® K'
and compute its first Chern class accordingly.

D-4. Let A= (L;) be a general n x (n + 1) matrix of homogeneous linear forms on

IP 3 , and C c IP 3 the locus where rank A ~ n - 1. Find the genus of C.

D-5. Using the notations of Exercise Sequence C, show that the first Chern class of the
tangent bundle to G = G(2, 4) is

Finally, if {F.} is a pencil of cubic surfaces, use this and the set-up of Exercise
C-12 to find the genus of the curve C formed by the lines on the surfaces F;,.
Compare this result with the one obtained by representing Cas a 27-sheeted cover
of IP 1 and using the Riemann-Hurwitz formula.

Note. For a further application of the formulas above, see Exercise C-7 of Chapter
VIII.
Chapter III

Introduction to Special Divisors

In this chapter we shall begin our study of linear series on smooth curves.
Our central question is: What are the limitations on the dimension r(D) of a
complete linear series ID I ? The first result in this direction is the classical
Clifford theorem. After discussing this, in a somewhat different vein, we shall
prove Castelnuovo's bound on the genus of a curve in projective r-space. This
will lead to Max Noether's theorem on the projective normality of canonical
curves, to a detailed study of extremal curves in r-space and to a brief presenta-
tion of Petri's theory.

§1. Clifford's Theorem and the General Position Theorem

We begin by recalling some definitions and results from Chapter I. Let D be

an effective degree d divisor on a smooth curve C of genus g. By the Riemann-
Roch theorem we always have

d - g ::;; r(D) ::;; d,

and as we explained in Section 2 of Chapter I,

If d > 2g - 2 then r(D) = d - g.

r(D) - 0 fior d ::;; g,

If d ::;; 2g - 2 and D is general then { ( ) - d
rD = - g for d ?: g.

In geometry one seldom encounters general divisors, and our main object of
study will be the behavior of r(D) for non-general D when 1 ::;; d ::;; 2g - 2.
The main general fact concerning special divisors is

Clifford's Theorem. (i) If D is an effective divisor of degree d on C with d ::;;

2g- 1, then

r(D) ::;; d/2;

108 III. Introduction to Special Divisors

(ii) if equality holds then either D is zero, D is a canonical divisor, or C is

hyperelliptic and D is linearly equivalent to a multiple of a hyper-
elliptic divisor.

Proof. We note that, by the Riemann-Roch theorem and the assumption that
d .::; 2g - 1, the Clifford inequality is trivially satisfied unless D is special;
we then assume that i(D) # 0. We first observe that r(D) ~ r if and only if
there is a divisor in ID I containing any r given point> of the curve. An im-
mediate consequence is that

r(D + D') ~ r(D) + r(D')

for any two effective divisors D and D' (indeed given r(D) + r(D') general
points of C we may find E E ID I containing the first r(D) and E' E ID'l con-
taining the last r(D'); then E + E' E ID + D' I contains the given set of points).
Since D is special we may find an effective divisor D' such that D + D' E IK 1.
Noting that i(D') = r(D') + 1 we may combine the above inequality with the
Riemann-Roch theorem to obtain

r(D) + r(D') .::; g - 1,

r(D) - r(D') = d - g + 1.
The first part of Clifford's theorem follows by adding these relations.
The basic inequality above, applied to a pair of residual divisors D and D',
may be interpreted as saying that the image of the bilinear map

has dimension at least equal to

l(D) + i(D) - 1,

that is, the sum of the dimensions of the two factors of the domain of f.Lo,
minus 1.
In fact, it is a remarkable theorem of H. Hopf that, given any setup of a
linear map

v: A® B-+ C,

where A, B, C are complex vector spaces and v is injective on each factor

separately, then

dim v(A ® B) ~ dim A + dim B - 1.

§1. Clifford's Theorem and the General Position Theorem 109

This result, which is false over IR, was one of the earliest applications of
topology to algebra.
The analysis of the map J.Lo in terms of the geometry of C is, perhaps, the
central theme of this book. The first case of this is the second half of Clifford's
theorem, in which we describe the circumstances under which equality
may hold, that is, when J.Lo may have minimal rank l(D) + i(D) - 1. The main
tool here will be the classical:

General Position Theorem. Let C c IP>', r 2 2, be an irreducible non-degener-

ate, possibly singular, curve of degree d. Then a general hyperplane meets C in d
points any r of which are linearly independent.

Assuming the general position theorem we now complete the proof of

Clifford's theorem.

Proof of(ii) in Clifford's Theorem. If equality holds in (i) then we must have

r(D) + r(D') = g - 1,

and then every canonical divisor is a sum E + E', with E E ID I and E' E ID' 1.
Since deg D + deg D' = 2g - 2, we may assume that deg D ::::;; g - 1, and by
the inequality of the first part either D = 0 or r(D) > 0. Assuming the latter, if
C were non-hyperelliptic then, by the geometric Riemann-Roch theorem,
every hyperplane section E + E' of the canonical curve C c IP 9 - 1 would
contain a set E of d ::::;; g - 1 dependent points. This contradicts the general
position theorem. Suppose now that C is hyperelliptic. In Chapter I, after
proving the geometric version of the Riemann-Roch theorem, we showed
that any complete g~ on C is of the form

rgi + P1 + · · · + Pd-2n
provided that d ::::;; g. Since we are assuming that 2r(D) = d, it follows im-
mediately that ID I is a multiple of the hyperelliptic gi on C. Q.E.D.

We now go back to the general position theorem. When r = 2 there is

nothing to prove: therefore, we assume from now on that r 2 3. Due to the
importance of the theorem we shall give two arguments. The first is based on
the following:

Lemma. Let C c: !Pr, r 2::: 3, be as in the statement of the general position

theorem. Then a general hyperplane meets C in d points, no three of which are
collinear.

We shall first show, in three steps, that the lemma implies the theorem.
(i) For a general point p E C, the projection
110 III. Introduction to Special Divisors

from pis birational onto its image. Indeed, by the lemma we may find a pair of
points p, q E C that do not lie on a trichord, and then nP is one-to-one over
np(q).
(ii) Next we let U c (IP')* be the open set of hyperplanes transverse to C,
and consider the incidence correspondence

IcCxU

consisting of pairs (p, H) where p E H n C. We claim that, if the general

position theorem is false, then for a general pair (p, H) E I there is a dependent
set of points

P = P1• pz, ... , p, E H n C.

Indeed, we first observe that I is irreducible of dimension r. Now we consider

the subvariety I 0 c I consisting of pairs (p, H) where p E H n C is part of a
dependent set of points p =Pt. ... , p,.It is clear that, if the general position
theorem is false, then dim I 0 = r. By irreducibility we must have I 0 = I as
desired.
(iii) Finally, for p E C a general point we consider the projection

By step (i), nP is birational onto the image curve C'. By step (ii), if the general
position theorem fails for C c IP', then it also fails for C' c IP'- 1 . But the
lemma is equivalent to the general position theorem in the case r = 3, and
this contradiction shows that the lemma implies the general position theorem.

Proof of the Lemma. We first observe that the lemma is equivalent to the
statement that there are at most oo 1 trichords to C, which is in turn equivalent
to the assertion that not every pair of points of C lie on a trichord; it is this last
statement we shall prove. To see the equivalence of the last two italicized
statements let J c C 2 x U be defined by

J = {(Pt. Pz, H): P~> p 2 E Hand p 1 + p2 is a trichord}.

Since C 2 is irreducible and the fiber dimension of J ~ C 2 is always r - 2, we

see that the surjectivity of J ~ U is equivalent to dim J ;::: r, which is in turn
equivalent to the surjectivity of J ~ C 2 . To prove that not every chord ofC is a
trichord, we argue by contradiction. If the statement is false, then given
general points p, q E C the chord pq will meet C in a third point v = u(p, q).
We choose small arcs around each of the points p, q, and v, local parameters s
on the first arc, t on the second, u on the third, and imagine the three arcs as
given by the projections to IP' of e+ 1 -valued functions p(s), q(t), v(u). Then
we may determine u(s, t) such that

p(s) A q(t) A v(u(s, t)) =0.

§1. Clifford's Theorem and the General Position Theorem Ul

Differentiating this relation with respect to s and t gives

1
p 1\ q 1\ v +p 1\ q 1\ V1 • ~
0U
OS
=0,
pAq
I I au 0
1\V+pAq/\V·~=.
at
Since it is clearly the case that oujos ¥=. 0 and 8uj8t ¥- 0, we infer that
P1 1\ q 1\ v = ll(p 1\ ql 1\ v),

This in turn implies that the vectors p, p q, ql lie in a C 3 ; i.e., we have shown
1
,

that if the lemma is false then any two tangent lines to C meet in a point. Now,
since r 2 3 and Cis non-degenerate, it follows that the tangent lines do not lie
in a fixed plane and hence this point must be the same for any pair of tangent
lines. So that, if the lemma is false, all tangent lines to C pass through a fixed
point p. But now projecting C from p would give a map with vanishing
differential and therefore C would be a line through p. This contradiction
establishes the lemma. Q.E.D.

Second Proof of the General Position Theorem. This approach introduces a

new idea but yields a much broader statement. Briefly, the idea is that a
general hyperplane section r = H n C cannot possess both dependent and
independent sets of r points, because there is no way to distinguish, uniformly
asH varies over all hyperplanes transverse to C, one subset ofr points ofr from
another. Since C c IP' is non-degenerate, r will contain some subset of r
independent points. The general position theorem will follow once we make
the italicized statement precise. As before we consider the incidence corre-
spondence I c C x U. The projection

is a d-sheeted topological covering. If we fix a base point H 0 E U and let

r 0 = C n H 0 be the corresponding hyperplane section, then as is the case for
any topological covering space there is the monodromy map

the image of which is called the monodromy group M of the covering I~ U.

Our general principle can be expressed by the following:

Lemma. The monodromy group of I ~ U is the full symmetric group.

For any positive integer m, we set

I(m) = {(p 1 ,
... , Pm, H): HE U,

and all the p;'s are distinct and belong to H n C}.

112 III. Introduction to Special Divisors

A statement equivalent to the above lemma is that the associated corre-

spondence I(d) c ca x U is connected. This is because an arc in I(d)
connecting two points in I(d) lying over H 0 will project to a loop in n 1 (U, H 0 )
whose corresponding monodromy transformation takes one of the two
orderings of r 0 into the other. Similarly, to say that I(m) is connected is
equivalent to saying that the monodromy group is m-times transitive; i.e.,
it acts transitively on the set of ordered subsets of r 0 consisting of m distinct
points.
To show that M is the full symmetric group it will suffice to show that
M is twice transitive and contains a simple transposition. We proceed in two
steps. We first show that M is twice transitive. We set

7(2) = {(p 1 , p2 , H): p 1 , p2 E H n C and p 1 #- p 2 }

c C X C X lfll'*;

this is a slightly enlarged version of I(2). Then, since any two distinct points of
C span a line, we see that l(2) maps onto (C x C) - A, with all fibers being
pr- 2 's. Consequently l(2) is irreducible, and the Zariski open subset I(2) c
l(2) is connected.
We now show that M contains a simple transposition. As is customary,
we denote by C* the dual of C, that is, the set of hyperplanes that are tangent
to C. If H 1 E C* - (C*)sing is a hyperplane that is simply tangent to Cat one
point, and if

is a one-parameter family of hyperplanes with H, E U for t #- 1, and H 1 as

as above, meeting C* transversely at H t. then it is easy to see that H, n C
contains two points that come together to the point of tangency as t ~ 1.
These points interchange as t turns once around 1. Thus M contains a simple
transposition. This proves the lemma.

Having established that M is the full symmetric group we know that I(r) is
connected. The subvariety

J = {(p1, ... , p,; H): Pt. ... , p, are dependent} c I(r)

is therefore a proper closed subvariety of I (r ), and, by irreducibility, dim J < r.

Thus J projects onto a proper subvariety of U and this proves the general
position theorem.

As promised, this version yields a more general result.

Uniform Position Theorem. Let C c lfll', r 2 3, be an irreducible, non-degener-

ate, possible singular, curve of degree d. If ::0 is any linear system on C, and
r = H n C a hyperplane section general with respect to ::0, then all subsets of
§2. Castelnuovo's Bound, Noether's Theorem, and Extremal Curves 113

m points of r impose the same number of conditions on ~. Equivalently, if

r' c r is any subset which fails to impose independent conditions on~. then
every divisor in ~ containing r' contains r.

This property of r is called uniform position. Of course, applying the above

theorem when ~ is the series cut out by hyperplanes in IP', we obtain the
general position theorem.

§2. Castelnuovo's Bound, Noether's Theorem, and

Extremal Curves
In this work we are primarily interested in exceptional special divisors. When
Cis non-hyperelliptic, these are the divisors whose points fail to be independent
on the canonical curve C c IP 9 - 1 • Before proceeding we make one remark. If
Dis a special divisor and D' is residual to D, then, in view of the Riemann-
Roch theorem:

r(D) + g = r(D') + d + 1,

it will, at least in principle, suffice to consider divisors of degree d ::::; g - 1. It

is natural to pose the following question: What is the "most special" a divisor
D can be? Because of the Riemann-Roch theorem, maximizing i(D) =
r(D') + 1 is the same as maximizing r(D). Taken literally, the above question
has already been answered by Clifford's theorem: the bound in part (i) is,
after all, sharp. But, as we saw in part (ii) of Clifford's theorem, the existence of
g'/s on a hyperelliptic curve C that achieve this bound is simply a reflection of
the fact that C possesses a g~. Accordingly, we may refine our question by
restricting our attention to linear series that are not composed with an
involution 1 - i.e., those with the property that the mapping to projective
space given by the linear series is birational onto the image curve-and ask
for an upper bound r(D) ::::; R(g, d) under this additional hypothesis. For
these reasons we consider an irreducible non-degenerate projective algebraic
curve C 0 in IP' of degree d. If

¢: C-+ C0 c IP'

is the normalization map and DE Icp*@c 0 (1) I is the pull-back of a hyperplane

section, then the mapping ¢ is given by a linear subsystem of the complete
system ID I; in particular, r(D) 2 r. Since we may expect the function R(g, d)
to be monotone increasing with d and decreasing with g, then our original
question should be, and, as it will turn out, is equivalent to the problem: What

1 Classically, the word "involution" means expressing C as a many-sheeted covering of

another curve.
114 III. Introduction to Special Divisors

is the maximum genus of a curve C that admits a birational mapping ¢ onto

a non-degenerate curve C 0 c IP' of degree d? When r = 2, there is a very
classical answer provided by the genus formula

g= <a - 1)(a - 2)/2 - I Jp,

peCo

where bP is a measure of the singularity of C 0 at p. In particular, since bP = 0 if

and only if pis a smooth point of C 0 we have that

g ::; (d - 1)(d - 2)/2,

with equality holding only for smooth plane curves. When r 2:: 3, even for a
smooth curve there is no formula for its genus in terms of its degree. The
situation is therefore more complicated. However there is an inequality
(known to Halphen [1] among others) for space curves, and this was ex-
tended to curves in IP' by G. Castelnuovo in 1889 [1]. We shall now derive
his bound.
Castelnuovo's idea is to estimate the dimension

rx 1 = r(ID)

of the complete linear systems associated to multiples of the hyperplane

divisor D on C, and then to apply Clifford's theorem. To do this let

be the linear subseries cut out on C by hypersurfaces of degree l and set

Let r be the pull-back to C of a general hyperplane section r 0 of C 0 . Then,

since E 1 certainly contains the series r + E1_ t. we have

f3t - /31- 1 2:: dim E 1 - dim El- r),

where E 1( - r) is the subseries of divisors in E1 containing r. Hence, denoting

by f Co the ideal sheaf of C 0 c IP', we have

f3t - f3t-1 2:: (h 0 (1P', f9(l)) - h0 (1P', J c0 (l)))

- (h 0 (1P', Jro(l))- h0 (1P', Jc0 (l)))
= h0 (1P', @(l)) - h0 (1P', Jr0 (l)),

which is just the number p1 of conditions imposed by r 0 on hypersurfaces of

degree l in IP' or, equivalently, on hypersurfaces of degree l contained in the
§2. Castelnuovo's Bound, Noether's Theorem, and Extremal Curves 115

hyperplane H ~ lfll'- 1 which cuts out r 0 • What is needed then is an estimate

on p1; and this is provided by

Lemma. lfp 1 , ••• , Pais a set of points in lfll'- 1 such that any r among them are
linearly independent, then p 1 , ... , Pa impose at least

min(d, k(r - 1) + 1)
independent conditions on the homogeneous polynomials of degree k.

Proof We first assume that k(r - 1) < d and label the points as

P1, · · ·, Pr-1; · · ·; P(k-1Hr-1J+~> · · ·, Pk(r-1J; · · ·; Pa·

'-v-'

Let Li be a non-zero linear function vanishing at the points in the group Hi.
By the assumption of the lemma Li will be non-zero at the remaining points,
and in particular Lj(pa) #- 0. Then the homogeneous polynomial of degree k

will vanish at p 1, ... , Pk(r- 1) but not at Pa· Since our labeling of the points P; is
completely arbitrary, we may find a homogeneous form assuming preassigned
values at any of k(r - 1) + 1 points selected from D, which implies the lemma
when k(r - 1) + 1 :::;; d. The remaining case is similar (and easier). Q.E.D.

According to the lemma, then, we have

(2.1) f3t - P1-1 2 Pt 2 min(d, l(r - 1) + 1).

Thus, if we set

[~]
m = r- 1 '

we have

IJ(o 2 Po= 0,
1)( 1 2 /3 1 2 r,

(2.2)
1)( 2 2 /3 2 2 r + 2(r - 1) +1= 2r - 1,

IJ(m 2 /3m m
2 1 ~/(r - 1) + 1 = (m +2 1) (r - 1) + m.
116 III. Introduction to Special Divisors

Applying Clifford's theorem to the series 1mD 1 we see that mD is non-special;

hence by the Riemann-Roch theorem,

(2.3) g = deg(mD) - r(mD)

=md-IXm

::::;; md - ( m +
2
1) (r - 1) - m

= (;)(r- 1) +me,

where

d - 1 = m(r - 1) + e.
We have proved, then

Castelnuovo's Bound. Let C be a smooth curve that admits a birational mapping

onto a non-degenerate curve of degree d in IP>'. Then the genus of C satisfies the
inequality

g(C) ::::;; n(d, r),

where Castelnuovo's number n(d, r) is defined by

m(m- 1)
n(d, r) = 2 (r - 1) + me,
where

m=[~J
r- 1
d - 1 = m(r - 1) + e.
Observe that, for fixed r and large d, asymptotically

d2
n(d, r) - 2(r - 1)

A closer look at the argument for Castelnuovo's bound gives more than the
bound itself. Since E 1 is a subseries of I(!)c(ID) I, r imposes at least as many
conditions on l@c(ID)I as on E 1, so that we have

IX1- IX1-1 ~ P1 ~ min(d, l(r- 1) + 1)

§2. Castelnuovo's Bound, Noether's Theorem, and Extremal Curves 117

as well as (2.1). Thus if equality holds in (2.3), it follows that we must have

IXz - IXz-1 = /3 1 - /31- 1 = min(d, l(r - 1) + 1).

In particular, we have ak = f3k for all k; from this we infer that if equality holds
in Castelnuovo's bound then the mappings

are surjective for all k 2 1. Equivalently, the image of

is a projectively normal curve that is biholomorphic to C under the mapping </J.

(Recall that for curves, and only for curves, projective normality implies
smoothness.) We shall refer to curves whose genus equals Castelnuovo's
number as extremal curves. Since ¢L maps C biholomorphically to its image
C 0 , we may identify C with C 0 . To obtain a clearer picture of extremal curves
we shall divide them in three cases according to the possibilities

d < 2r, d = 2r, d > 2r.

(i) When d < 2r it follows from Castelnuovo's bound-or better from

Clifford's theorem-that Dis non-special and consequently r = d - g. Then
d < 2r = 2d - 2g implies that d > 2g, and in this case it is well known that
the map <P is an embedding.
(ii) When d = 2r, Castelnuovo's formula gives us r = g - 1, and therefore
C c lfl> 9 - 1 is a canonical curve. A consequence of the projective normality of
extremal curves is then the very famous

Max Noether's Theorem. If C is a non-hyperelliptic curve, then the homo-

morphisms

are surjective for l 2 1.

In particular, since

dim Sym 2 H 0 (C, K) = g(g + 1)/2,

dim H 0 (C, K 2 ) = 3g- 3,

we have that a canonical curve of genus g lies on exactly (g - 2)(g - 3)/2

linearly independent quadrics. These quadrics will be the subject of consider-
able discussion in the next section and in Section 4 of Chapter VI. Here we
shall give two examples.
118 III. Introduction to Special Divisors

When g = 4 we find that a canonical curve lies on a unique quadric in IP 3 • A

similar count shows that there are five independent cubics containing the
curve, hence at least one cubic containing the curve but not the quadric, so that
a non-hyperelliptic curve of genus 4 is the complete interesection of a quadric
and a cubic in IP 3 .
When g = 5 we see that a canonical curve C c IP 4 lies on exactly three
linearly independent quadrics and is generally the complete intersection of
these quadrics. More precisely, we recall that a curve C is said to be trigonal
in case it has a base-point-free gL i.e., there is a map f: C-+ IP 1 of degree 3,
and we will later show that a non-hyperelliptic, non-trigonal curve of genus 5
is the complete intersection of three quadrics in IP 4 and conversely.
(iii) When d 2: 2r + 1, we are in the range of exceptional special divisors;
we shall first make a few remarks about these beautiful curves when r ::::;; 3, and
then we shall discuss these in general.
First, when r = 2 Castelnuovo's bound gives the previously noted
inequality

g ::::;; (d - 1)(d - 2)/2;

in this case extremal curves are smooth plane curves. The hyperplane section
consists of d points on a line and there is not much more to be said about these
here.
When r = 3 the estimate gives

{
(k - 1) 2 when d = 2k is even,
(2.4)
g ::::;; k(k - 1) when d = 2k + 1 is odd.

Moreover, if equality holds, then a general hyperplane section gives a

configuration D of d 2: 5 points in IP 2 that imposes only min(d, 2 · 2 + 1) = 5
conditions on the linear system of plane conics. In other words, any conic that
passes through five points of D must contain all of D, and from this it follows
that the points of D lie on a unique plane conic. Finally, since equality holds in
(2.1) we have
h0 (C, (l)c(2)) = h0 (C, (l)c(1)) +5
=9
= h0 (1P 3 , (I)IP',(2)) - 1,
i.e., if Cis an extremal curve of degree d 2: 7, then C lies on a unique quadric Q.
In fact, we can say more: Assuming that the quadric is smooth we recall that
it has two rulings IL 1 I, IL 2 1by straight lines, and consequentely any curve C
on Q has two numerical characters

m = C·Lt.
n = C·L 2 ,
§2. Castelnuovo's Bound, Noether's Theorem, and Extremal Curves 119

whose sum m + n is the degree of C. The canonical bundle KQ is represented

by the divisor - 2L 1 - 2L 2 , and by the adjunction formula the genus of Cis

g = !(KQ · C + C · C) + 1
=mn-m-n+1
= (m - 1)(n - 1).

Assuming m ;?: n, this is maximized when

m = n = k, in case d = 2k,
m - 1 = n = k, in case d = 2k + 1.

Moreover, in these cases equality holds in (2.4). In case the quadric Q is

smooth we may draw the following conclusion.

Lemma. An extremal space curve of degree d = 2k is the complete intersection

of a quadric and a surface of degree k. When d = 2k + 1 then,for some line L
lying on the quadric, the sum C + Lis the complete intersection ofa quadric and
a surface of degree k + 1.

If Q is not smooth, a similar argument can be used to prove the lemma.

We note that if Cis any curve lying on a smooth quadric Q in 1Pl 3 , the points
of a general hyperplane section lie on a conic and so impose the minimum
number of conditions on the curves of degree k in that hyperplane, i.e.,
min {d, 2k + 1}; the difference between extremal curves, i.e., those of
"balanced type" (m, m) or (m, m- 1), and the others, i.e., those of un-
balanced type (m, n), where I m - n 1 ;?: 2, is that the latter are not projectively
normal.
Although the geometry of extremal curves in jpl' for r ;?: 4 is analogous to
that of extremal curves in 1Pl 3 , it is certainly more delicate. In 1Pl 3 , from the
string of inequalities (2.2) used to derive Castelnuovo's bound, we deduced
that an extremal curve lay on a quadric surface. Knowing all curves on such a
surface, we were able to characterize geometrically the curves of maximum
genus. In the general case we will again use the same string of inequalities to
infer that an extremal curve C c jpl' must lie on a large number of quadric
hypersurfaces. Then, under the hypothesis

deg C = d ;?: 2r + 1,
we shall argue that C lies on a surface of a very special type, and finally we will
be able to characterize extremal curves. Let us then consider an extremal curve
C c jpl' of degree d and maximal genus g = n(d, r). As in the case of space
120 III. Introduction to Special Divisors

curves, equalities must hold throughout (2.2). In particular,

h0 (C, (!}(D)) = r + 1,

and

h0 (C, (!}(2D)) = 3r,

where, as usual, D denotes a hyperplane section of C. Since

ho
(IPr' (!}i"r(2)) = (r+2)
2 '

we conclude that C lies on (r - 1)(r - 2)/2linearly independent quadrics. Of

course, in case r = 3, we get back the fact that C lies on a unique quadric. The
question we must ask is: What is the intersection of the quadrics that contain
C? The answer to this question is a consequence of the following beautiful
result.

Castelnuovo's Lemma. Let d and r be integers such that r ~ 3 and d ~ 2r + 1.

Let r c IP'- 1 be a collection of d points in general position which impose only
2r - 1 conditions on quadrics. Then r lies on a unique rational normal curve X r.

Castelnuovo's ingenious argument, which actually gives a synthetic

construction of Xr. is presented in Griffiths-Harris [1, pages 528-531],
and we will not reproduce it here.
Consider then our extremal curve C c IP'. Let H be a general hyperplane in
IP' and set r = C n H. By the general position theorem r consists of d
distinct points in general position in H ~ IP'- 1• As we already noticed,
h0 (C, (!}(D))= r + 1andh 0 (C, (!}(2D)) = 3r.WecanthenapplyCastelnuovo's
lemma and conclude that r lies on a unique rational normal curve X r c H.
Since

d = deg(r) ~ 2r +1~ 2 deg X r,

every quadric in IP' containing C, and hence r, also contains the rational
normal curve Xr. Conversely, as we noticed in Section 5 of Chapter II, Xr is
the intersection of the (r - 1)(r - 2)/2 quadrics containing it, hence the
intersection of the quadrics containing C meets H exactly in Xr. Thus the
intersection of the quadrics containing Cis a surface S whose general hyper-
plane section is a rational normal curve; in particular deg S = r - 1. As a
special case of Theorem (5.3) of Chapter II, we recall that, with the exception
of the Veronese surface in IP 5 , the non-degenerate surfaces of minimal degree
r - 1 in IP' are the rational normal scrolls.
§2. Castelnuovo's Bound, Noether's Theorem, and Extremal Curves 121

Let us consider the case of a smooth rational normal scroll S and let us
find, on it, extremal curves. The Picard group of S is easy to describe. In fact
Pic(S) is freely generated, over Z, by the classes H of a hyperplane section
of S, and L of a line of the ruling. The intersection products are given by

H ·H = r- 1, H·L = 1; L·L =0.

Each of the linear systems IHI and IL I contains smooth rational curves, and
the adjunction formula gives

0 = (L) = L . L + Ks . L 1
Pa 2 + '

O = p.(H) = H · H ~ K 8 · H + 1.

From this we obtain

Ks = - 2H + (r - 3)L.

We may now compute the (arithmetic) genus p.(C) of any curve

CEiexH + f3LI. Namely,

_ C ·C + Ks · C
Pa (c) - 2 + 1

= (ex - 1 ~(ex - 2) (r - 1) + (r - 2 + {3)(ex - 1).

We write the degree d of C as

d = m(r - 1) + 1 + e,
and also as

d = (r - 1)ex + {3.
Comparing these we see that the maximal genus

m(m- 1)
n(d, r) = 2 (r - 1) + me

of Cis attained exactly when

ex= m + 1, {3 = e - r +2 for any e,

ex= m, {3 = 1 in case e = 0.
122 III. Introduction to Special Divisors

A similar analysis may be carried out in the case in which S is a cone over a
rational normal curve, and an even easier argument may be given for curves
on a Veronese surface. The final conclusion is this:

(2.5) Theorem. Let d and r be integers such that r 2 3, d 2 2r + 1. Set

m = [(d - 1)/(r - 1)] and d = m(r - 1) + 1 + e. Then extremal curves
C c P' of degree d exist and any such curve is one of the following:
(i) The image of a smooth plane curve C c P 2 of degree k under the
Veronese map P 2 --+ P 5 . In this case r = 5, d = 2k.
(ii) A non-singular member of the linear system ImH + L I on a rational
normal scroll. In this case e = 0.
(iii) A non-singular element of the linear system I(m + 1)H - (r - e - 2)L I
on a rational normal scroll.

Remark that in case (iii), when e = r - 2, the curve is the complete

intersection of a scroll and a hypersurfa:ce of degree m + 1; in case (iii), for any
e, adding r - e - 2 lines of the ruling to an extremal curve we get a complete
intersection.
The characterization of extremal curves in terms of the surface S yields
some corollaries. For example, since the lines of the ruling of a rational normal
scroll will cut on C a pencil of degrees m or m + 1, we have the

(2.6) Corollary. If C c P' is an extremal curve of degree

d = m(r - 1) + 1 + e 2 2r + 1

and is not a plane curve, then either:

(i) C possesses a complete 2 g~, call it IE I, satisfying

(r- 2)E E IK( -(m- 2)H) I,

where H is the hyperplane divisor, in case e = 0; or,

(ii) C possesses a complete 2 g~+ 1 , call it IE I, satisfying

(e- 1)£ E IK( -(m- 1)H)I,

in general; and finally,

(iii) in case e = 1, we have

K = C9((m- l)H),
2 That the g~, or the g~+ ~> is complete, does not follow directly from the above analysis.
A proof is outlined in one of the exercises.
§3. The Enriques-Babbage Theorem and Petri's Analysis of the Canonical Ideal 123

i.e., any extremal curve C c [FD' of degree

d =2 (modulo r - 1), d "2:. 2r

is a Klein canonical curve.

We note that one may also use this description of extremal curves to give
particularly nice plane models for them; cf. Accola [2].
We close this section by observing that an analysis similar to Castelnuovo's
may also be carried out for curves C c IP'' whose genus is close to, but not
necessarily equal to n(d, r). Specifically, we can go back to the inequalities
(2.1), (2.2), and (2.3) and ask under what circumstances some may fail; the
point is that they do not fail independently (thus, for example, by Castelnuovo's
lemma strict inequality in (2.1) for l = 2 implies at least strict inequality for all
l < m). One conclusion we may arrive at is the following: for any d and r, set

m1- -[~]
r
,

1 if 8 1 = r - 1,
f1.1= {
0 if 8 1 i= r - 1,

and

Then we have the statement (cf. Eisenbud-Harris [3]).

(2.7) Theorem. Any irreducible, reduced, and non-degenerate curve C c IP'' of

degree d "2:. 2r + 3 and genus g > n 1(d, r) lies on a surface of degree r- 1.

A complete proof of this theorem is beyond our present scope; a proof

under the additional hypothesis that d ~ r is outlined in Exercise batch 31
below.

§3. The Enriques-Babbage Theorem and Petri's Analysis of

the Canonical Ideal
The statement of Castelnuovo's lemma, which we recalled in the previous
section, requires that the set r c IP'' contain at least 2r + 1 points. This
condition cannot be improved: the intersection of three general quadrics in
IP' 3 is a collection of eight points in general position imposing only seven
conditions on quadrics, and it does not lie on a twisted cubic curve. Thus our
124 III. Introduction to Special Divisors

analysis of extremal curves does not apply to canonical curves, which are
extremal curves of degree 2g - 2 in IP'g- 1 . In the case of a canonical curve C
we can prove the following:

(3.1) Proposition. If the intersection of the quadrics containing a canonical

curve C contains one additional point p ¢ C, then C lies on either the Veronese
surface (in case g = 6) or on a rational normal scroll.

Proof. We first argue that we may assume p does not lie on infinitely many
chords of the canonical curve C c IP'g- 1 ; or, equivalently, that the projection
of C from p to IP'g- 2 is birational. For if p did lie on infinitely many chords, then
every such chord would lie in all the quadrics passing through C. Clearly a
general point of the surface S swept out by those chords does not lie on oo 1
chords; if it did, the union of all the chords to C would be contained in S. We
may accordingly replace p by a general point of Sand proceed. Thus we may
assume that the projection of C from p is birational onto a non-degenerate
curve C' c IP'g- 2 . The hyperplane sections of C' are cut out on C by hyper-
planes H c IP'g- 1 passing through p. Applying the uniform position theorem
we see that the points of r = H n C, together with p, form a collection of
2(g- 1) + 1pointsingeneralpositioninH ~ IP'Y- 2 thatimposeonly2g- 3
conditions on quadrics. Castelnuovo's lemma then applies and we conclude
that the base S of the linear system of quadrics through C intersects H in a
rational normal curve. Thus S is either a Veronese surface on a rational
normal scroll. This scroll cannot be a cone over a rational normal curve since
C is non-hyperelliptic. Hence such a scroll must be smooth. Q.E.D.

The preceding proposition is the central step in the proof of a nice theorem
due to Enriques and Babbage stating that, with only few exceptions, a
canonical curve is, set-theoretically, an intersection of quadrics. To char-
acterize those canonical curves that are not intersections of quadrics we use
the proposition we just proved and notice that, exactly as in Corollary (2.6),
one can easily show that if Cis any extremal curve of degree d = m(r - 1) +
1 + c; in IP'r lying on a scroll, then C possesses a g~ + 1 . For canonical curves we
have that r = g - 1, d = 2g - 2, and m = 2. Consequently Cis trigonal; i.e.,
it has a gj. In fact the gj is the linear system cut out on C c S by the ruling of
the scroll. Finally, if the canonical curve C c IP' 5 lies on the Veronese surface
the curve must be a smooth plane quintic. Summarizing we have proved the

Enriques-Babbage Theorem. If C c IP'g- 1 is a canonical curve, then either C

is set-theoretically cut out by quadrics, or Cis trigonal or is isomorphic to a plane
quintic.

In a very beautiful and fundamental paper Petri goes.far beyond this set-
theoretical description. He constructs the first few steps of a resolution of the
ideal of a canonical curve by constructing a basis of its homogeneous co-
§3. The Enriques-Babbage Theorem and Petri's Analysis of the Canonical Ideal 125

ordinate ring and a set of generators for the first module of syzygies. In the
remaining part of this section we shall carry out Petri's analysis.
Let us consider a smooth non-hyperelliptic curve C of genus g, which we
imagine embedded in its canonical space rr:w- 1 . The first step in Petri's
analysis is to construct an explicit basis for the vector space H 0 (C, Kn) of n-
canonical differentials on C. Let p 1 , ... , p9 be g general points on C. By the
uniform position theorem, which we proved in the first section of this chapter,
and by the geometric version of the Riemann-Roch theorem, the dimension
of the vector space H 0 (C, K(- p 1 - · · · - pi - · · · - p9 )), i = 1, ... , g, is
equal to one. Let wi be a generator of this vector space. Thus, as a section of
K, wi vanishes at all the p/s, with the sole exception of Pi. It is then clear that
w 1 , •.. , w 9 is a basis of H 0 (C, K); in fact, for any relation 2: A.iwi = 0,
evaluating at pj gives A.j = 0. Moreover, by general position, we may assume
that the divisors (wi) consist of 2g - 2 distinct points and that their supports
are pairwise disjoint. We shall adopt the following notation. We let

be the natural map. We also let X 1 , •.. , X 9 be the basis of H 0 (P 9 - \ {!)IP'9 -t(1))
defined by

i = 1, ... ' g.

The X;'s can be thought of as homogenous coordinates in P 9 - 1• Given a

homogeneous polynomial

we set

so that, for instance, we have: Xf X 3 = wfw 3 • We also denote by I = EB I k

the homogeneous ideal of C and by

R(C) = (~ H 0 (P 9 - 1, {!)IP'•-t(n)~/1
the homogeneous coordinate ring of C. Now set D = p 3 + · · · + p9 • Again
by general position the dimension of the vector space H 0 (C, K(- D)) is equal
to two and its basis is {w 1 , w2 }. Moreover, by our assumption that the
divisors (wi) have no point in common, the pencil IK(- D) Iis base-point-free.
We next consider the tower
126 III. Introduction to Special Divisors

When n - 1 2 s 2 1, by the Riemann-Roch theorem we get

h0 (C, K"( -sD)) = (2n - 1)(g- 1) - s(g - 2),

so that each vector space in the above filtration has codimension g - 2 in the
preceding one. The goal is to exhibit, for each s, a set of n-canonical forms in
H 0 (C, K"( -sD)) which are linearly independent modulo

H 0 (C, K"(( -s- 1)D))

and give a step-by-step construction of a basis for H 0 (C, K"). To find the basis
of H 0 (C, K( -sD)) we proceed inductively by considering, for each nand s
such that n - 1 2 s 2 1, the cup-product mapping

The kernels of these maps can be computed by means of the following lemma.

Base-Point-Free Pencil Trick. Let C be a smooth curve, L an invertible sheaf

on C, and :F a torsion~free {!)cmodule. Let s 1 and s 2 be linearly independent
sections of L, and denote by V the subspace of H 0 (C, L) they generate. Then
the kernel of the cup-product map

is isomorphic to H 0 ( C, :F ® L - 1 (B)), where B is the base locus of the pencil

spanned by s 1 and s 2 .

Proof.Lets 1 ® t 2 - s 2 ® t 1 beanelementofthekernel;thuss 1 t 2 - s 2 t 1 = 0.
Write

where sis a section of {!)(B) vanishing on B, and r 1 , r 2 are sections of L(- B).
Since :F is torsion-free we have that r 1 t 2 - r 2 t 1 = 0; as r 1 , r 2 have no
common zeros, this means that t; is a section of :F(- D;), i = 1, 2, where D; is
the divisor of zeros of r;. In other words, there are sections r 1 , r 2 of :F ®
L - 1(B) such that t; = r;r;. Plugging this into r 1 t 2 - r2 t 1 == 0 we get, using
again the fact that :F is torsion-free, that r 1 = r 2 = r. In conclusion

i = 1, 2.

Conversely, it is clear that any element r of H 0 (C, :F ® L - 1 (B)) gives rise, by

the above prescription, to a relation s 1 t 2 - s 2 t 1 = 0. Q.E.D.
§3. The Enriques-Babbage Theorem and Petri's Analysis of the Canonical Ideal 127

Applying the base-point-free pencil trick to (3.3) yields

Combining this with the Riemann-Roch theorem and making a straight-

forward computation, we get the

Lemma. Let n - 1 ~ s ~ 1. Then ¢n,s is surjective except if(n, s) = (3, 2). In

this case Im ¢ 3 • 2 is of codimension one in H 0 ( C, K 3( - 2D)).

Let us now begin Petri's construction.

Basis for H 0 (C, K 2 ). Consider the cup-product mapping

By the lemma this mapping is surjective and it is immediate to check that the
elements ·

i = 3, ... ' g,

form a basis of H 0 (C, K 2 ( -D)). Now consider the two-step tower

It is then clear that wL ... , w; are differentials in H 0 (C, K 2 ) which are

linearly independent modulo H 0 (C, K 2 ( -D)). Since the latter space is of
codimension g- 2 in the former one, we get a basis of H 0 (C, K 2 ):

:~~~1(1);~(1)!, ... 'g]l

w 2 wi, ~: 3, ... , g b~sis of ~ 0 (C, ~ 2 ( -D))
wf, 1 - 3, ... , g basrs of H (C, K ).

Basis for H 0 (C, K 3 ). This one is, so to speak, the only anomalous case. The
difference comes from the fact that, as stated in the lemma, the map

is not surjective. We set W = Im ¢ 3 • 2 . By the lemma we know that W is of

codimension one in H 0 (C, K 3 ( -2D)). Let us then consider the filtration
128 III. Introduction to Special Divisors

Again, with the exception of W, each vector space of this tower is of codimen-
sion g - 2 in the preceding one. We first claim that a basis of W is given by

wi, wfw 2 , w 1 w~, wL

WfW;, W1W2 W;, W~W;, i = 3, ... ,g.

Since W is (3g - 2)-dimensional it suffices to show that the above differentials

are linearly independent. Suppose there is a relation

with a, /3, y E H 0 (C, K). Setting

g g-1
(wz) = LP; + L q;,
i= 1 i= 1
i*2

and recalling that (w 1) and (w 2 ) have disjoint supports, we see that a belongs
to the one-dimensional vector space H 0 (C, K( -p 1 - Ir:f q;)), which is
generated by w 2 • Therefore a = A.w 2 , and similarly y = f.iWz, with A., f.i E IC.
We then get

Writing f3 = IJ=
3 /3;w;, and looking at the order of zero at p;, i = 3, ... , g,
we see that /3; = 0, for all i. This leads to a relation A.w 2 wf = f.iW 1wL and
considering the order of zero at p 1 , we see that A. = f.i = 0. We next try to
construct, as canonically as possible, an element of H 0 ( C, K 3 ( - 2D)) which,
together with the basis of W, generates H 0 ( C, K 3 ( - 2D)), or, what is the same,
an element of H 0 (C, K 3 ( -2D))- W. Actually, we shall exhibit one such
element for each index i = 3, ... , g. For an index i in this range, we consider
the unique divisor in the pencil IK(- D) I having a zero of order two at P;,
and let fi.; be a differential having this as its zero divisor. We claim that
fi;WI E H 0 (C, K\ -2D))- W Suppose not; writing fi.; = A.;w 1 + f.i;Wz we
then have

with 1/J, ¢, x E H 0 (C, K). A moment of reflection shows that, for some 1/J', ¢',
x' E H 0 (C, K), one may write the above relation in the form

Let (fi.;) = p3 + · · · + 2p; + · · · + p9 + D;; since, by hypothesis, IK( -D) I is

base-point-free, for each i, the two divisors D; and (w 2 ) are disjoint. The above
§3. The Enriques-Babbage Theorem and Petri's Analysis of the Canonical Ideal 129

relation shows then that x' E H 0 ( C, K(- D;)), which is a one-dimensional

vector space generated by ii;. Our relation reduces then to the form

and looking at the order of vanishing at Pk we easily get a contradiction,

proving that, for each i = 3, ... , g, the differential ii;wf generates

Altering each ii; by a suitable non-zero constant, it is possible to write

(3.4) i = 3, . .. ,g,

so that if too generates H 0 ( C, K 3( - 2D)) modulo W. Let us now choose

[3; E H 0 ( C, K(- D)) in such a way that {ii;, [3;} is a basis for H 0 ( C, K(- D)).
Using the fact that each step in the filtration

is of codimension g - 2 in the preceding one, we can finally construct a basis

of H 0 (C, K 3 ) in the following way:

wtw;, i = 3, .. . ,g,

OJ1 OJ2 W;' i = 3, ... ,g,

OJ~ W;' i = 3, .. . ,g, basis of W

if basis of H 0 (C, K\ -2D))
Piwf, i = 3, .. . ,g, basis of H 0 (C, K 3 ( -D))
wf, i = 3, .. . ,g, basis of H 0 (C, K 3 ).

Basis for H 0 (C, Kn), n ~ 4. To build a basis for H 0 (C, Kn) we use the filtra-
tion (3.2). First of all we construct a basis for H 0 (C, Kn(( -n + l)D)) by using
the basis of H 0 (C, K 3 ( -2D)), which we just constructed, and an induction
procedure based on the surjectivity of the maps

¢m,m- 1 : H 0 (C, Km- 1 (( -m + 2)D)) ® H 0 (C, K( -D))

~ H 0 (C, Km(( -m + l)D))
130 III. Introduction to Special Divisors

form~ 4. Specifically one checks that a basis for H 0 (C, K"(( -n + 1)D)) is
given by

l+m=n,
s + t = n- 1, i = 3, ... ,g,
h + k = n- 2.

To get a basis for H 0 ( C, K") we now climb the ladder and notice that a basis of
H 0 (C, K"(( -s + l)D)) modulo H 0 (C, K"( -sD)), s::; n- 1, is given by the
differentials f1f- 1 w?-s+t, i = 3, ... , g. In conclusion, a basis for H 0 (C, K") is
given by

w 11w';,, l+m=n,
s +t= n- 1,

t.,;,
w~wiwi,
i = 3, 0 0 0 'g,
h k-
WlW21'f, h+k=n-2, fm H 0 (C, K'(( -n + I)D))
fJ'/- 2Wf, i = 3, 0 0 0' g, basis fo.r H 0 (C, K"(( -n + 2)D))

Piw'/- 1, i = 3, 0 0 0' g, basis for H 0 ( C, K"(- D))

OJ'/, i = 3, 0 0 0 'g, basis for H 0 (C, K").

The second step in Petri's analysis is to construct quadratic and cubic

generators for the homogeneous ideal of C. First of all, we notice that for
3 ::; i, k ::; g, and i # k the quadratic differential wiwk belongs to

and vanishes at p 1 and p 2 . Looking at the basis of H 0 (C, K 2 ( -D)), we see that
we can write

wiwk = L iisikws + bikW1W2,

s=3

where asik = A.sikw 1 + 11sikw 2 and A.sik• 11sik• bik are complex numbers. It then
follows that the quadratic polynomials

~;k = xixk- I asikxs- bikxlx2, 3 ::; i, k ::; g' i # k,

s=3

all vanish on C. We obtain in this way (g - 2)(g - 3)/2linearly independent

elements in I 2 • But, by Max Noether's theorem in Section 2, the dimension of
§3. The Enriques-Babbage Theorem and Petri's Analysis of the Canonical Ideal 131

I 2 is exactly (g - 2) (g - 3)/2. Thus the J;k's form a basis for I 2 . Let us now
consider cubic relations. Looking at (3.4) we define cubic polynomials Gk 1 by
setting

3 ~ k, l ~ g, k =F l.

These polynomials obviously belong to I 3 and, moreover, satisfy the cocycle

relations

(3.5)

We can now state

Petri's Theorem. Let C be a canonical curve ofgenus g 2 4. Then the ideal ofC
is generated by the J;k's, unless C is trigonal or isomorphic to a smooth plane
quintic, in which cases it is generated by the J;k's and the Gk 1's.

To prove Petri's theorem we first show that the J;k's and the Gkt's generate I.
Let us show that they generate I 3 • Consider then an element

Since the wf's, i = 3, ... , g,arelinearlyindependent modulo H 0 (C, K 3 ( - D)),

by restricting R to C we get Yiii = 0. Using the expression of the J;k's we can
write
g

R = L bukXthk + L (Jl;!Y.; + v;/3;)X? + w,

i=3

where w belongs to the vector space W generated by xi, xi x 2 , x 1 x~, x~, xi X;,
x 1 x 2 x;, x~x;, i = 3, ... , g. Restricting to C and using (3.4), we get L/1; = 0
and V; = 0, for all i. But then we can write

with w' E W. Restricting again to C we find that w' E I. Since, by the definition
of W, we have W n I= {0}, our claim is proved.
Let us now consider a relation

As above we immediately get Yiiii = 0 and we can write

132 III. Introduction to Special Divisors

where w belongs to the subspace W' of H 0 (1l:w-I, @1Jl>9 -,(4)) generated by the
monomialsX~X~,s + t = 4,andX1X~Xi,h + k = 3,i = 3, ... ,g.Againit
is immediate to check that W' n I = {0}. Restricting to C and using the
explicit basis of H 0 (C, K 4 ), we easily get ri = /1-i = 0. We then write

where the Q's are quadratic in X 1 and X 2 • Using again the expression of the
fik's we can write

R L p;kiik + L T,.lGkl + L A.;(J(f Xf

=
+ 'f. !1-;(J(;/3iXf + w',
with w' E W'. Restricting R to c and using the fact that fi.iwf = r; + ei> we
get I (A.;(J(i + 11-JJi) = 0, so that

with w" E W'. But, as we observed, W' n I= {0}, proving our claim. The
general case of I. is handled in exactly the same way, and the first part of
Petri's theorem is proved.

To prove the second half of his theorem Petri exhibits a system of syzygies
among the quadratic and cubic relations. First of all we can assume that
g ~ 5: When g = 4 we already know that the canonical curve is the complete
intersection of a quadric and a cubic in IP' 3 and that the rulings of the quadric
cut out on C a gj. We also observed that, if C is trigonal or isomorphic to a
smooth plane quintic, then its ideal cannot be generated by quadrics only.
Before writing down the syzygies among the fik's and Gk/s we need the follow-
ing notation. Consider the relations
g

wiwk = I a:.ikw. + bikwlwz;

s=3

we then see that, when g ~ 5, for any triple of distinct integers i, l, k the
differential Ziuk vanishes doubly at p 1, so that there are scalars Pikl such that

We can now write Petri's syzygies: for any triple of distinct indices i, k, l
such that 3 :-::;; i, k, l :-::;; g, we have
g

(3.6) fikXl- Ji1Xk = I (a.ufsk- asikfsl) + PiklGkl•

s=3

where, by convention, j~k = .1;1 = 0. To prove that indeed these relations

§3. The Enriques- Babbage Theorem and Petri's Analysis of the Canonical Ideal 133

hold we proceed as follows. First of all, one checks easily that

hkXI- .htXk = I (a.uf~k - asiklst) + akitXf - aukXf + w,

s= 3

where w E W. Restricting to C we get the relation

which can also be written as

w'EW.

Looking at the basis of H 0 (C, K 3 ) and keeping in mind that W n I= {0},

we obtain that w' = 0 and that Piki = Pitk· This proves that Petri's syzygies
hold and that the Pikl are symmetric in i, k, and l.
To prove his theorem Petri proceeds as follows. Fix an integer k,
3 ~ k ~ g, and look at the variety Vk defined by the equations

{_hk = 0}, i = 3, ... 'g, i # k.

Then write these equations in the following way

( b;,Xk - st3
asik)x• = ak;kXk + b;kX!Xz.
•*k
Let 11k be the determinant of the (g - 3) x (g - 3) matrix

and let 11~ be the determinant of the matrix obtained by substituting the sth
column of Mk with the column 1(akikXk + b;kX 1 X 2 ), i = 3, ... ,g, i # k.
Both 11k and L1k are homogeneous polynomials in X 1 , X 2 , and Xk. By
Cramer's rule we have

s = 3, ... , g, s # k.

More geometrically, we can define a (rational) surface Fk whose parametric

equations, away from the hypersurface 11k = 0, are

X 1 = X 1,
Xz = Xz,
(3.7) xk = xb
11s
A'.= 11> s = 3, ... 'g, s # k.
134 III. Introduction to Special Divisors

By construction, the surface Fk is the only component of fk which is not con-

tained in the hypersurface dk = 0. Now observe that, on one hand C is
contained in J.-k, and on the other, the homogeneous polynomial dk does not
vanish at the point Pk of C. We can therefore conclude that Fk is the only
component of fk containing C. Parenthetically, it could be instructive to
notice that, if we denote by Ck the projection of C from the vertex

into the plane {X. = 0}, s = 3, ... , g, s # k, then Ck is a degree g + 1 plane

curvewhilethecurvesX 1 dk = O,X 2 dk = O,Xkdk = O,d~ = O,s = 3, ... ,g,
s # k, generate the system of degree g - 2 adjoint curves to Ck; so that the
surface F k is nothing but the birational image of IP 2 under the map associated
to this linear system.
Petri's key observation is the following:

Lemma. For 3 :::; k, l:::; g, k # l, the surface Fk coincides with the surface F1
if and only if Pikt = 0 for all i = 3, ... , g, i # k, i # l.

Before proving the lemma let us see how the theorem follows from it.
We first assume that all the P;k/s vanish. By the lemma the surfaces F 3 , ... , F 9
all coincide. Writing F 3 = · · · = F 9 = F, we have that all the quadrics
through C contain F and we may use Proposition (3.1) to conclude that C
is either trigonal or isomorphic to a smooth plane quintic. It now remains to
show that, if for some triple i, s, t the complex number Pist is non-zero, then the
ideal I is generated by the /;/s. Suppose not and let J c I be the ideal
generated by the /;k's. Since, as we know, I is generated by the /;k's and Gk/s,
there must be a k and an l such that Gk 1 ¢ J. Looking at the syzygies (3.6),
this implies that Pikt = 0 for every i, and by the lemma we get Fk = F 1 •
Now, given any j such thatj # k,j # l, 3 :::; j:::; g, write

Therefore either Gki ¢ J or Gi1 ¢ J. Arguing exactly as in the case of Gkt>

this says that either Fk = Fi or F 1 = Fi. Since Fk = F 1 we get F 3 = · · · = F 9 •
In particular F. = F 1 , and by the lemma this implies that Pist = 0, contrary
to our hypothesis.
It now remains to prove the lemma. Suppose first that Pikt = 0 for every i.
Then Petri's syzygies restricted to F k tell us that

L
•*l
(b;sXk - asik)f.t =0 (mod Fk), i = 3, ... ,g,
•*k i # l, i # k.
Bibliographical Notes 135

Let Ll 1k be the determinant of the (g - 4) x (g - 4) matrix (b;sXs - asik),

= 3, ... , g, i =!= l, k, s =1= l, k. By Cramer's rule
i, s

i = 3, ... 'g, i =I= l, i =I= k.

Since Ll 1k(Pk) =!= 0, the above equations tell us that F k is contained in ~.

But, as we remarked, F 1 is the only component of ~ containing the curve
C, so that Fk = F 1 , proving the first part of the lemma. Suppose, conversely,
that Fk = F~> and assume that Pikl =1= 0 for some index i. Then Petri's syzygies
imply that Gk 1 vanishes on Fk. Write Gk 1 in the form

Gkt = akX~- a1Xf + F(X;Xi, X;X1X2, X;XD

+ AXkXi + pXkX 1 X 2 + vXkXL
where F is a linear combination of the monomials XtX 1 , X;X 1X 2, X;XL
i = 3, ... , g, i =1= k. Multiplying by (Llk) 2 and using (3.7), we get

(Llk) 2 Gkt = ak(Llk) 2 X~ - a1(Llk) 2 Xf

+ ilkF(LlkXiXf, LlkXiX 1 X 2 , ilkX;XD
+ A.(ilk) 2 X kXf + p(Llk) 2 X kX 1 X z + v(Llk) 2 X kXL
so that,

(Llk) 2 Gktl Fk = ak(Llk) 2 X~ - rxtCLli) 2

+ LlkF(Lli Xi, ili X 1 X 2, iliX~)
+ A.(Llk) 2 XkXf + f1(Llk) 2 XkX 1 X 2 + v(Llk) 2 XkX~.
Since Gk 1 vanishes on Fk, the polynomial on the right-hand side of the last
equality is identically zero. Looking at the term akX~ 9 - 4 coming from
ak(Llk) 2 X~ and recalling that ak = c5X 1 + yX 2 , with y =I= 0, c5 =!= 0, we get a
contradiction. This concludes the proof of Petri's theorem.

Bibliographical Notes
Clifford's theorem is, of course, classical; the original statement is in Clifford
[1]. The trilinear algebra set up of Clifford's theorem was used, over !R,
by H. Hopf to study the question of existence of division algebras in di-
mension 1, 2, 3, and 8. The argument in Hopf [1] appears to be one of the
first applications to geometry ofthe cup-product in cohomology. The general
position theorem is also classical and may be found for instance in Severi
[2]. It is perhaps worth remarking that the result is false in characteristic
p =I= 0. A recent general discussion of monodromy questions is given in Harris
[1].
136 III. Introduction to Special Divisors

Castelnuovo's bound appears in Castelnuovo [1]. The present discus-

sion of the bound and its consequences is extracted from Griffiths-Barris
[1] which contains a proof of Castelnuovo's lemma and a discussion of
rational normal scrolls. An analytic proof of Castelnuovo's lemma is given
in Griffiths-Harris [2].
The Enriques-Babbage theorem appears in Enriques [1] and Babbage
[1]. The ideal theoretical study of the canonical curve was done by K. Petri
in [1], and modern versions of this approach can be found in Saint-Donat
[1], Arbarello-Sernesi [1], and Mumford [3]. For a somewhat different
approach, see Shokurov [1].

Exercises

A. Symmetric Products of 1Jll 1

A-1. Show that the dth symmetric product of P 1 is isomorphic to pd, and that under this
identification the locus of divisors of the form dp, pEP', is rational normal curve
c c pd,
For the remainder of these exercises we wiii implicitly make the identifications
(P 1)d = pd and {dp}peiJ>' = C.

A-2. Show that the tangent line to Cat the point dp is the locus of divisors

{(d- l)p + q}qeiJ>'·

More generally, show that the osculating k-plane to C at dp is the locus of divisors

{(d - k)p + q, + ... + qk},

(Note that in the case k = d - 1 this shows that the locus of osculating hyper-
planes to a rational normal curve is a rational normal curve in the dual projective
space.)
Observe that if p 1 , ..• , Pk are distinct points of P 1 , the span of the points
dp" ... , dpk is just the (k - 1)-dimensionallinear series spanned by

(this is tautologous). We wiii call such a (k - I)-plane a strict secant plane to C.

(For example, a strict secant line is not a tangent line.)

A-3. Show that a polynomial P(X) E H 0 (P 1 , (I)(d)) is expressible as a sum of k dth

powers of linear polynomials if and only if the corresponding point in
Exercises 137

lies on a strict secant (k - 1)-plane. Conclude the

Theorem. A general polynomial of degree d is expressible as a sum of k dth powers

of linear forms if and only if k ~ (d + 1)/2.

Hint: Use the independence of points on a rational normal curve to conclude that
the locus of secant k-planes to C contains an open set of dimension min(d, 2k + 1).

A-4. Show that there exist polynomials of degree d which are not expressible as a sum
of fewer than d dth powers of linear forms.
Hint: Look at a point on a tangent line to C.

B. Refinements of Clifford's Theorem

In the following series of exercises we will obtain sharper versions of the basic inequality
r(D + D') ~ r(D) + r(D') in Section 1 of Chapter III.
One notational convention :if fZ c ID I and & c IE I are linear series, we will denote
by fZ + & the subseries of ID + El spanned by sums of divisors in fZ and divisors in&,
and by & - fZ the subseries of & containing a general divisor D of !:Z. Also, we will write
r(!:Z) for the dimension of a linear series fZ.

B-1. Suppose that fZ, & are linear series on a smooth curve C of genus g > 0, and assume
that fZ + & is base-point-free and c/>1Hc is birational. Show that

r(!:Z + &) ~ r(!:Z) + r(&) + 1.

(Hint: Using the uniform position lemma, show that a general Fe fZ + tff is not
the sum of a divisor in fZ and a divisor in&.)

B-2. Suppose fZ, & are base-point-free pencils and that cf>!!i! =F cf>c· Show that

r(!:Z + &) ~ 3.

(Hint: Let the pencils be given by u 0 , u 1 e H 0 (C, (!)(D)) and r 0 , r 1 e H 0 (C, (!)(E))
and, under the assumption that r(!:Z + &) = 2, show that the image of C under the
map [a 0 r 0 , u 0 rl> u 1 r 0 , a 1rd lies in a hyperplane section of the quadratic

from this obtain a contradiction to cf>!!i! =F c/>r,.)

B-3.
(i) As an application of the preceding exercise, show that a smooth, non-hyper-
elliptic curve of genus g ~ 5 cannot have two distinct g~'s.
(ii) Show by example that there exist non-hyperelliptic, non-trigonal curves of all
genera g ~ 1 which possess infinitely many distinct gi's (consider two-sheeted
branched covers of elliptic curves).
138 III. Introduction to Special Divisors

B-4. Let !21, rff be linear series on a smooth curve and assume that !21 is base-point-free.
Show that

r(E21 + rff) ~ 2r(rff) - r(rff - !21).

(Hint: For disjoint divisors D 0 , D 1 E !21, show that the intersection of the linear
spaces D 0 + rff, D 1 + rff has dimension at most r(rff- !21). Alternatively (and, in
fact equivalently), look up the base-point-free pencil trick.)

B-5. As an application of the preceding exercise, suppose that C is a smooth, non-

hyperelliptic curve having a g~ = !21 and that rff is a g;_
1 on C. Show that either
rff or K- rff contains the g~.
(Hint: Estimate dim !21 + rff and use Riemann-Roch.)

B-6. Let !21, rff be linear series on a smooth curve and assume that !21 is base-point-free of
degree d and c/>,_,1 is birational. Show that

r(E21 + rff) ~ min(r(E21) + 2r(rff) - r(rff - !21), d + r(rff)).

In particular, if d ~ r(E21) + r(rff) - r(rff - !21) then

r(E2! + rff) ~ r(E21) + 2r(rff) - r(rff - !21).

(Hint: Use the uniform position lemma to argue that the number of conditions
imposed by Don !21 + rff is (degree permitting) the number of conditions imposed
by D on !21 plus the number of conditions imposed by D on rff.)

B-7. Using Castelnuovo's theorem, show that if a smooth curve C of genus g has a
g'i,+-, say !21, with 0 ::o; IX ::o; r- 2 then either:
(i) ¢qp is not birational and C is a two-sheeted covering of a curve of genus at
most IX/2: or
(ii) ¢~is birational and either:
(a) g ::o; r + 21X; or
(b) g = r + 21X + 1 and C is trigonal if IX > 0 while !21 = IK I if IX = 0
(cf. Comessatti [2], Beauville [2]).

C. Complete Intersections

In some of the remaining exercises we will use the following standard facts:
For X c Ya smooth divisor on a smooth variety, the adjunction formula

holds; an equivalent and perhaps more familiar notation is

Kx = Kr ® [XJix·
Exercises 139

It is also well known that

H 1(1P'', (!J(k)) = 0, 1 ::s;; i ::s;; r - 1 and any k;

H 1(1P'', .fi'(k)) = 0, i > 0 and k ~ 0

for any coherent sheaf .fi' on IP''.

C-1. Suppose that C = S n Tis the transverse intersection of two smooth surfaces of
degrees m, n in IP' 3 (in particular, Cis a smooth complete intersection). Show that

and that the genus of C is

g = tmn(m +n- 4) + 1.

C-2. Suppose that F 1 , ••• , F 1 are homogeneous polynomials on IP'' whose locus of
common zeros has dimension r - l. Let H be any homogeneous polynomial that
is locally in the ideal generated by F 1, •.. , F 1 (more precisely, the forms F 1 , •.. , F 1
generate a subsheaf J c (!JIJ>,, and H should be a section of J Q9{91J>,(d) where
deg H =d). Show that

H = IA.F.

for suitable homogeneous forms A •.

(Hint: In the local ring at each point of IP'' the localizations of the F. give a
regular sequence, and the Koszul complex (cf. Griffiths-Harris [1]) gives a
canonical resolution

•<P
where deg F. = d•. An argument using (*) then shows that the global sections of
this sequence also give an exact sequence.) When r = l = 2 this result is Noether's
"AF + BG" theorem.

C-3. Let C c IP' 3 be a smooth complete intersection of two smooth surfaces of degrees
m and n. Using the preceding exercise, compute the number oflinearly independent
surfaces of degree k containing C. Deduce that the surfaces of degree m + n - 4
cut out the complete canonical series on C.

C-4. Show that the curve C in the preceding exercise does not have a g~+•- 3 •
(Hint: Show that any set of d ::s;; m + n - 3 distinct points in IP' 3 impose
independent conditions on surfaces of degree m + n - 4.)
140 III. Introduction to Special Divisors

D. Projective Normality (I)

Definitions. A smooth curve C c IP' is k-normal if the hypersurfaces of degree k cut out
the complete linear series 1(!Jc(k) 1. A 1-normal curve is usually said to be linearly normal,
and C c IP' is projectively normal if it is k-normal for every k.

D-1. Show that a smooth irreducible curve C c IP' is k-normal if and only if

H 1 (1P', Jc(k)) = 0
where,$ c c (!)",is the ideal sheaf of C. Deduce that Cis k-normal fork ~ 0, and
from this infer that every meromorphic function on C is the restriction of a
rational function on IP' (this exercise is a consequence of(*) and (**) in the
preceding batch of exercises).

D-2. Let C c IP' be a smooth irreducible curve and D = H · C a hyperplane section.

Show that if C is linearly normal then every quadric in H ~ IP'- 1 containing D
is the restriction to H of a quadric in IP' containing C.

D-3. Keeping the notations of the preceding exercise, show that C is projectively
normal if and only if every hypersurface in H containing D is the restriction to H
of a hypersurface in IP' containing C.

D-4. Let C c IP' be a smooth irreducible curve of degree d < 2r. Show that C is pro-
jectively normal if and only if it is linearly normal.

D-5. Using Exercise B-6, show that if C c IP' is smooth, irreducible, and k-normal,
and (!Jc(k - 1) is non-special, then Cis (k + 1)-normal (and hence [-normal for
allL > k).

D-6. Let C c IP' be a smooth irreducible curve of degree d. Using the preceding
exercise, show that if Cis k-normal fork :::;; (d + r)f(r - 1), then Cis projectively
normal.

D-7. Let C c Q c IP 3 be a smooth curve of type (m, n) on a smooth quadric surface Q.

Show that C is projectively normal if and only if Im - n I :::;; 1 (i.e., C is as close
to a complete intersection as its degree m + n will allow).

D-8. With C as in the preceding exercise, if m < n show that C fails to be k-normal
exactly for k satisfying m :::;; k :::;; n - 2.

D-9. More generally, let S and T be smooth surfaces in IP 3 whose intersection is the
sum of smooth curves C and D. Show that Cis projectively normal if and only if
Dis.

D-10. Let C c IP' be a smooth irreducible curve with defining homogeneous ideal
I c S = C[X 0 , ••• , X,]. Show that C is projectively normal if and only if the
homogeneous coordinate ring Sf! is integrally closed (this is frequently taken as
the definition of projective normality).
(Hint: Show that the integral closure of Sf! is R = ffineo H 0 (C, (!J(n)).)
Exercises 141

D-11. Show that a smooth complete intersection curve C c IP'' is projectively normal.

D-12. Let Q c IP' 3 be a smooth quadric surface, L 1 and L 2 lines from the two distinct
rulings, and (r)(m, n) = (r)Q(mL 1 + nL 2 ). Show that

ho(Q, (r)(m, n)) = {(m + 1)(n + 1), m,n;;::: -1,

0, m :s; - 1 or n :s; - 1,
0, m,n;;::: -1,
{
h (Q, (r)(m, n)) = -(m
1 + 1)(n + 1), m;;::: -1,n :s; -2or
m :s; -2, n;;::: -1.

(Suggestion: Use the Kiinneth formula.) Using these computations give

another proof of Exercise D-7.

D-13. Suppose that ID I is any complete linear system on a smooth, linearly normal
curve C c IP''. Show that, for A E ID I, dim ~ is independent of the divisor A.
(Hint: Relate dim~ to h0 (C, (9(1)( -A)).)
This result is false if we do not assume C to be linearly normal (why?).

D-14. Show that the following curves in IP' 3 (assumed to be smooth) are projectively
normal:
(1) A twisted cubic.
(2) A quartic of genus 1.
(3) A quintic of genus 2.
(4) A non-hyperelliptic sextic of genus 3.
(5) A sextic of genus 4.
(6) A septic of genus 5.
(7) A septic of genus 6.

(Hint: It may help to observe that, in all the above cases, (r)c(k) is non-special
fork;;::: 2.)

D-15. Show that the curves listed in the preceding exercise are all the projectively
normal non-degenerate curves of degree :s;7 in IP' 3 •

E. Castelnuovo's Bound on k-Normality

This series of exercises will establish the following result due to Castelnuovo:

Theorem. Let C c IP'' be a smooth curve of degree d. Then the hypersurfaces in IP'' of
degree d - 2 cut out on C a complete linear series.

Recently, this theorem has been strengthened by Gruson, Lazarsfeld and Peskine,
who replace the d- 2 in the statement with d- r + 1 (assuming Cis nondegenerate,
of course); cf. Gruson-Lazarsfeld-Peskine [1].

The following notations will be used.

142 Ill. Introduction to Special Divisors

If r ;:::>: 3, n: C---+ C c IP' 2 will denote a general projection of C onto a plane curve
having only nodes as singularities.
We let {r.}.~ 1 •...• d be the set of nodes of C and we set

n-1(r.) = Pa + q.;
d

r = I (p. + q.):
CJ:==l

d-1

r' = I (p. + q.).

a=l

E-1. Using Appendix A, show that the linear system cut out on C by hypersurfaces of
degree d - 2 through r is the complete series I(9c(d - 2)(- r) 1.

E-2. Using Appendix A again, show that there exists a hypersurface of degree d- 3
containing r' but not Pd or qd, and hence a hypersurface of degree d - 2 containing
r' and Pd but not qd.

E-3. Conclude from the preceding exercise that r imposes independent conditions on
hypersurfaces of degree d - 2 in IP'', and then from this complete the proof of the
theorem.

F. Intersections of Quadrics

F -1. Show that a collection r = p ~> ... , p2 • of 2n points in general position in IP'" is cut
out by quadrics.
(Hint: Ifq liesoneveryquadriccontainingp~> ... ,p 2 .,1etp~> ... ,pk be a minimal
subset of p 1 , .•• , p 2• such that q E p 1, ••. , pk; consider quadrics of the form

Q = {Ph••·,f};", ... ,pbpap•••,Pan-k+)

u (r- {pb ... , P:; · · ·, Pk, P., · · ·, Pan-k+ ,}).

F-2. Let L---+ C be a line bundle of degree d ;:::>: 2g + 2 over a smooth curve of genus g
with corresponding embedding

Using the preceding exercise and Exercise D-2, show that (f>L(C) is the intersection
of quadrics.

F-3. Show that the statement of the preceding exercise is false when d = 2g + 1. Is there
a counterexample with g ;:::>: 3?
Exercises 143

The remaining exercises are motivated by the discussion in Chapter VI concerning

quadrics of minimal rank through a canonical curve, and the reader may wish to post-
pone them until reading that section.

Let X c IP" = IP'V be a quadric hypersurface given by a symmetric bilinear form

Q:V XV---> IC,

or equivalently
Q:V---> V*.

Recall that the rank of X is the rank of Q.

F-4. Show that X has rank kif and only if the singular locus of X is an (n- k)-plane.
(Hint: X,;ng = IP(ker Q), and X has rank k exactly when X is a cone over a
smooth quadric of dimension k - 1.)

F-5. Retaining the preceding notations, show that the following are equivalent:

rank X::;; 2m; and

X contains an (n - m)-plane.

F -6. Let IP'k c IP" be any linear subspace and X' = X n IP'k. Show that

rank X ;;:: rank X' ;;:: rank X - 2(n - k).

Hint: If IP'k = IP'W for W c V, consider the sequence

F-7. By refining the arguments of F-1 and F-2 show that any linearly normal curve of
degree d;;:: 2g + 2 in IP' is cut out by quadrics of rank 4 (in particular, this applies
to curves of degree d < 2r).

G. Space Curves of Maximum Genus

In the following series of exercises, taken from Harris [5], we will prove the

Theorem. Let C c IP 3 be a smooth curve of genus g(C) and degree d > k(k- 1) lying on
an irreducible surface of degree k. Then

d2
g(C) ::;; nM) = 2k + d(k 2- 4)
- :2e(k- e- 1 + ke) ,
where

{
=
e -d modulo k,
O::;;e::;;k-1.
144 Ill. Introduction to Special Divisors

Before doing the general case we will, in the first two exercises, derive the theorem
when k = 3. In doing both this and the general case we will use the Riemann-Roch
theorem for an effective divisor r on an irreducible plane curve H (the reason for the
notation "H" will appear below), where r is supported away from the singularities of H.
The statement is

where

d = deg r,
{k = degH,

wH is the dualizing sheaf of H.

We remark that

so that ( *) applies just as if H were smooth.

In general, a linear system on any curve His defined to be IP'Vwhere V c H 0 (H, L)
is a linear subspace of the global sections of a line bundle L ---> H. Any divisor r supported
on the smooth points of H defines a line bundle (!) H(r).

G-1. Let r c IP' 2 be a collection of d distinct points lying in the smooth locus of an
irreducible cubic plane curve H, and denote by p 1 the number of conditions that
r imposes on curves of degree I. Show that:
(a) if I < d/3, then

PI
=(' +2)
2 _(1-2 I)= 31.
'

(b) if I ~ d/3, then

p1 = d - I if I = d/3 and r is a complete intersection,

{
Pz = d otherwise.

G-2. Let C 0 c IP' 3 be an irreducible curve of degree d lying on an irreducible cubic

surfaceS, and denote by g the genus of the normalization C of C 0 . Show that

d = 0 (3),

d ~ 0 (3).

(Suggestion: Use the preceding exercise applied to a general hyperplane

section H of S to estimate the dimension h 0 (C 0 , (!)eo( I)), and then proceed exactly
as in the proof of Castelnuovo's bound.)
Exercises 145

G-3. Using the preceding exercise and Castelnuovo's bound, show that there does not
exist a smooth curve C c IP' 3 of degree 9 and genus 11.

G-4. Give a proof of

Gieseker's Lemma. Let V c H 0 (1P' 1 , (')(d)) be any linear subspace and W the
image of the map

Assume that the linear series IP' V has no .base points. Then either

or

dim W 2 dim V + 2.
(Hint: Use the base-point-free pencil trick to deduce an exact sequence

defining a vector bundle E, and then estimate h0 (1P' 1 , E) using the Grothendieck
decomposition

dimV-1
E = EB (')ll',(e;)
i = 1

of E. Alternatively, show that if dim W = dim V + 1, then every divisor DE IP'W

is the sum of a divisor D' E IP'V and a point p E 1(')11',(1) 1; use a monodromy
argument then to conclude that for any DE IP'Wand p E Supp D, D' = D - p E IP'V
and hence that V = H 0 (1P'\ (')(d)).)

G-5. Let r c IP' 2 be a collection of points in uniform position and let i 1(r) be the
smallest integer i such that there exists a curve of degree i in IP' 2 containing r.
Show that every curve of degree i 1 (r) containing r is irreducible.
(Hint: Show that if C = C 1 u C 2 is a plane curve of degree i 1 containing r,
then r n c. fails to impose independent conditions on curves of degree deg c. for
IX = 1 or 2; then apply the uniform position statement.)

G-6. Continuing with the notations of the preceding exercises, let H be a curve of
degree i 1 (r) containing r, let iir) be the smallest integer i such that there exists
a curve of degree i containing r but not containing H, and let
146 III. Introduction to Special Divisors

Show that

(a) I u, = d;
I

(b) u 1 =I+ 1,

(c)

Having done this, define m to be the least positive integer such that

(d) u, = 0, l~m.

G-7. By applying Gieseker's lemma to the linear series IJl>JI! cut out on a general line
L c 1Jl> 2 by l...f r ® lD"2(/) I, show that

(e) u 1 < u 1_ 1,

G-8.
(i) Compute the sequences p1 = p 1(i 1 , d) and u1 = u 1(i 1 , d) in case r is the
complete intersection of a pair of curves of degrees i 1 and i2 .
(ii) Similarly, compute p1 = p1(i 1 , d) and u 1 = u 1(i 1, d) in case r is residual to
i 1 i 2 - d collinear points in an intersection of curves of degrees i 1 and i 2 •

(Suggestions : One way to do (i) is by Noether's AF + BG theorem. One way

to do (ii) is by appealing to the Riemann-Roch formula ( *).)

G-9. Now fix i 1 and d > i 1 (i 1 - 1). Show that, if {u1} is any sequence satisfying (a)-(e)
above, then

Moreover, if equality holds for a sequence {u1} arising from r as in G-5 then r is
residual to i 1 i2 - d collinear points in an intersection of two curves of degrees i 1
and i2 •

G-10. Finally, let C 0 c 1Jl> 3 be an irreducible curve of genus d > k(k - 1) lying on an
irreducible surface S of degree k. Applying the preceding two exercises to a
general plane section H, estimate the differences

and deduce the bound in the theorem.

Remark. Using G-9 it can be shown that, if equality holds in the estimate in the
theorem, then C 0 is residual to a plane curve in a complete intersection of S with a surface
of degree [(d- 1)/k] + 1.
Exercises 147

H. G. Gherardelli's Theorem

In these three exercises we will retain the notation of the preceding batch and prove the

Theorem (G. Gherardelli). Let C c lfD 3 be a projectively normal curve such that
Kc ~ (!!c(m) for some m. Then Cis a complete intersection of two surfaces.

H-1. Setting

and

show that

f11 + f3m-1+1 = d,
Y1 = Ym-1+2•

and

I Yi = d = deg(C).
i

H-2. Let k be the smallest degree of a surface S c lfD 3 containing C, and let n be the
smallest degree of a surface T c lfD 3 containing C but not S. Show that

1'1 = 1 + 1, 0 ~ 1 ~ k- 1,
1'1 = k, k~l~n-1.

H-3. Conclude that

Y1 ;?: k - 1 + n - 1, n~l~n+k-1.

and hence that

Ll'i;?: nk.
i

This implies the theorem.

I. Extremal Curves

1-1. Let X c n:o• be a smooth surface scroll, and denote by Land H the divisors of a line
of the ruling and a hyperplane section, respectively. Show that

Kx = (!!x( -2H + (r- 3)L).

148 III. Introduction to Special Divisors

I-2. Let C E laH + PLibe a smooth curve. Show that

a(a - 1)
g(C) = 2 (r- 1) +(a- 1)(p- 1).

I-3. Show that

H\X, f9x(aH + PL)) = 0 for a > 0, P~ -a.

By using the Riemann-Roch theorem for surfaces

x(C9x(D)) = x(C9x) + !(D · D - Kx ·D),

where D is a divisor on X, deduce that

a(a + 1)
h0 (X, f9x(aH + PL)) = - 2- (r - 1) + (a + 1)(p + 1).

(Hint: Use the exact sequence

I -4. Show that a smooth curve C E IaH + PL I is projectively normal if and only if
- (r - 1) $; P $; 1.

I-5. Let C c IP'' be an extremal curve of degree

d = m(r - 1) + 1+e> 2r,

and let I c C[ X 0 , . . . , X,] be the homogeneous ideal defining C. Show that a

minimal set of generators for I consists of (r - 1)(r - 2)/2 quadrics and r - 1 - e
polynomials of degree m + 1.
(Sketch: In case C E l(m + 1)H- (r- e- 2)Lishow that:
(i) Every hypersurface of degree l $; m containing C contains X.
(ii) Modulo the ideal of X there are exactly

h0 (X, f c(m + 1)) = h0 (X, r9((r - e - 2)L)) = r - e - 1

linearly independent hypersurfaces of degree m + 1 containing C.

(iii) H 0 (X, r9((r- e- 2)L)) ® H 0 (X, C9(aH))-+ H 0 (X, r9(aH + (r - e - 2)L))
is surjective.)
Perform a similar analysis in case C- mH +Lor in case C lies on a Veronese
surface.

I-6. Show that the ideal of a trigonal canonical curve of genus g is generated by the
(g - 2)(g - 3)/2 quadrics containing it plus g - 3 additional cubics.
Exercises 149

I-7. Let C E lo:H + {JL I be a smooth curve on a scroll X with - (r - 2) :-:;:; {3. Show that
the linear system l@x((o:- 2)H + (r- 3 + {J)L)I cuts out the complete canonical
series on C.
(Hint: Use I-1, I-2, I-3, the adjunction sequence

0--+ Kx--+ Kx(C)--+ Kc--+ 0,

and the fact that

h 1(X, Kx) = h 1 (X, {!)) by Kodaira-Serre duality,

= 0 (since X is rational).)

I-8. Let C c IP'' be an extremal curve of degree d > 2r given by(*) above. Show that
any m - 1 points of C impose independent conditions on 1Kc I, and m points fail
to do so only if they are collinear. Conclude that C possesses no g~_ 1 , and that the
g~ cut out by the lines on X is unique. (Use the preceding exercise.)

J. Nearly Castelnuovo Curves

The following series of exercises will give a proof of theorem (2.7) of Chapter III under
the additional assumptions:

(i) C = C 0 is smooth;
{
(ii) deg C > 2'- 1 •

The first assumption is primarily a technical convenience. However, the second assump-
tion allows a much simpler argument than is required in the general case; for example,
it is not too difficult to deduce Castelnuovo's lemma on quadrics from general theorems
when d > 2'- 1 (cf. Exercise J-4 below).

J-1. This exercise is combinatorial. Setting

lr, lr < d,
pf = { lr- 1, /r = d,
d, lr > d,

show that

l
n 1 (d, r) = ld- L pf,
i=l

where n 1 (d, r) is defined as in (2.7).

In the next two exercises we let X c IP'' be an irreducible non-degenerate variety of

dimension k and codimension c = r - k, Y be a general hyperplane section of X, and
150 III. Introduction to Special Divisors

J-2. Show that

p,(X) ~ Pz-1 (X) + p,(Y).
J-3. By repeatedly applying the preceding exercise, deduce that

(i) p,(X) ~ (I+ k- I) + (I + k)

k c k '

(ii) for 0 :s; IX :s; c, if d ~ c+ IX + 1 then

p,(X) ~ (I+ 1)c + (I + 2) + (I +

k-
k
k-
k IX k
k)
.

The next exercise will use the following general result (cf. Fulton [2]).

Theorem. If Wi c IP'' are irreducible varieties of degree d; and X= nWf, then the sum
of the degrees of the irreducible components of X is at most d;. n;
J-4. Suppose that C c IP'' is a non-degenerate curve of degree d > 2'- 1 and r is a
general hyperplane section of C. Suppose that

pz(r) :s; 2r - 1 + IX

for some IX with 0 :s; IX < r - 2. Show that r lies on a curve of degree at most
r- 1 +IX.
(Hint: If X is the intersection of the quadrics containing r use d > 2'- 1 to
show that dim X ~ 1; p 2 (r) = p 2 (X) < 3r - 3 and uniform position to show
dim X = 1 and estimate its degree.)

J-5. Suppose now that Cis smooth of genus g, and show that for l sufficiently large
l

9 :s; td - I p;(r).
i=1

J-6. Suppose now that r lies on an elliptic normal curve H, that is, an irreducible,
non-degenerate curve of arithmetic genus 1 and degree r in IP"- 1 • With the
notations of J-1 show that

with equality holding if and only if d '!- 0 (r) or r is a complete intersection of H

with a hypersurface in IP''- 1 .

J-7. Using the inequalities in J-3, assuming that

pz(r) > 2r - 1,
show that

for all/.
Exercises 151

J-8. Suppose that r lies on a smooth rational normal curve in IP''- 1 . Show that

pz(r) = 2r - 1.

J-9. Combining Exercises J-1 and J-5, show that if r does not lie on a rational normal
curve, then pz(r) ~ p{ for all I, with equality holding only if r lies on an elliptic
normal curve.

J-10. With Castelnuovo's number rc(d, r) as defined in Section 2 of Chapter III, show that

n 1 (d, r) > rc(d, r + 1).

Using the above exercises show that if

then C is linearly normal.

Now complete the proof of the theorem under the additional assumptions ( *).

K. Castelnuovo's Theorem

In this sequence of exercises we will prove the following generalizations of Max Noether's
theorem.

Theorem. Let ID I be a complete base-point1ree linear series of dimension r = r(D) ~ 3

on a smooth curve C, and assume that the mapping

rPv: C--> IP''

is birational onto its image. Then the natural map

is surjective for 1 ~ 0.

Before beginning we recall

Castelnuovo's Lemma. Let L--> C be a line bundle such that

H 1 (C, L( -D))= 0,

and let f!fi c ID I be a base-point1ree pencil given by a two-dimensional linear subspace

V c H 0 (C, (I}(D)). Then the natural map

is surjective.
152 III. Introduction to Special Divisors

This follows immediately from the base-point-free pencil trick which gives the exact
sheaf sequence

0--> L(-D)--> V® L-> L(D)--> 0.

K-1. Let !?2 c ID I be a base-point-free pencil corresponding to

Show that the image of the natural map

is of codimension lr - I in H 0 (C, K(ID)).

(Hint: Do first the cases I = I, 2 using the base-point-free pencil trick and the
Riemann-Roch theorem. By Castelnuovo's lemma applied to L = K(ID) deduce
that every element in H 0 (C, K(ID)) is, for I;:::: 3, of the form

I P.w. + Qry,

where w~> ... , w 9 is a basis for H 0 (C, K), rJ E H 0 (C, K(2D)), P. E Sym 1 V, and
Q E Sym1- 2 V.Then argue by induction on /(thecase l = 2 shows you what to do).)

K-2. Supposethatp 1 + · · · + pdis ageneraldivisorin IDI and set£= p 1 + · · · + P,- 2 .

Show that ID- Elisa base-point-free pencil, and that in the exact cohomology
sequence of

0--> {!}(D - E) --> {!}(D)--> {!}(D)® {!}E--> 0

the map

is surjective (use the general position theorem).

K-3. By tensoring the exact cohomology sequence of the preceding exercise with
H 0 (C, K((/- I)D) and arguing by induction on /, complete the proof of the
theorem.

L. Secant Planes

L-1. Using the general position theorem, show that a curve C c IP' cannot have
cod (d + I)-secant (d- I)-planes spanned by their intersection with C, for any
d < r.
Chapter IV

The Varieties of Special Linear

Series on a Curve

As we have said, our approach to the study of curves is to analyze their

projective realizations, and to speak of these is to speak of linear series. A
more in-depth reflection leads to the appreciation of the fact that it is not
only the single linear series which are important, but rather the configuration
of all linear series of given type that a curve carries. To make precise what
these "configurations" are we shall introduce three main kinds of varieties,
which we now describe set-theoretically.
The first one, q, is the subvariety of the d-fold symmetric product Ca
parametrizing effective divisors of degree d on C moving in a linear series
of dimension at least r:

c~ = {DE Ca: dim IDI ~ r}.

The second, W~(C), is the subvariety of Pica( C) parametrizing complete

linear series of degree d and dimension at least r:

W:i(C) = {IDI: deg D = d, r(D) ~ r}.

The third one, G:i(C), parametrizes linear series of degree d and dimen-
sion exactly r on C:

G:i(C) = {g~'s on C}.

The link between C~ and W~(C) is given by the abelian sum mapping

in fact

u( C~) = W~( C).

An essential feature of these subvarieties is that, on one hand, C~ and W~(C)

carry natural determinantal structures while, on the other, G~(C) turns out
154 IV. The Varieties of Special Linear Series on a Curve

to be the canonical blow-up of the determinantal variety W:i(C) as defined

in Chapter II, Section 4.
The reader should be aware of the fact that in the next two chapters we
shall not need the results of this one in their full generality; a more naive
approach would have been sufficient. At the cost of disregarding history
and of rendering the chapter technically rather heavy, we have chosen to
collect in a unique body all the various constructions and foundational
results concerning the varieties of special divisors, in order to avoid repeti-
tions and possible confusion in the sequel. In our opinion the sections that
should get the most attention are Section 1, which is the closest in spirit to
the origin of the subject; Section 4, which introduces one of the most widely
used technical tools of the book; and Section 5, which contains a few practical
applications. On a first reading a more superficial look at the remaining
sections should be sufficient.

§1. The Brill-Noether Matrix and the Variety Ca

Let C be a smooth genus g curve. As usual we denote by Cd its d-fold symmetric
product. Our goal is to define a subvariety Cd of Cd parametrizing effective
divisors of degree d on C that vary in a linear series of dimension at least r.
In symbols, we would like to have

Supp(C:i) = {DE Cd: r(D) ~ r}.

As with so much of the theory, the key tool to define Cd as a variety is the
Brill-Noether matrix which we are now going to introduce. Let D = P; L
be an effective degree d divisor on C. Choose a local coordinate z on C in a
neighborhood of {p 1 , . . . , pd} and write

a= 1, .. . ,g,

where w 1 , ... , wg form a basis for the vector space of holomorphic differen-
tials on C and the f.'s are holomorphic functions in a neighborhood of
{p 1 , ... , Pd}. Assuming that Pt. ... , Pd are distinct, it follows from the
Riemann-Roch theorem that r(D) ~ r if and only if

The above matrix, which is usually written

§1. The Brill-Noether Matrix and the Variety C~ 155

is called the Brill-Noether matrix: the important role it plays in the funda-
mental paper [1] by Brill and Noether justifies this terminology. The
fact that r(D) ~ r if and only if the rank of the Brill-Noether matrix does not
exceed d - r shows that, in a neighborhood of Din Cd, the subset

{D' E Cd: r(D') ~ r}

is the common zero locus of the (d - r + 1) x (d - r + 1) minors of this

matrix. Clearly, the ideal generated by these minors does not depend on the
choice of local coordinate z and of the differentials w 1 , ••. , w9 • In this way
we may define q determinantally, at least away from the diagonals.
We now wish to extend the above determinantal description by means of
an intrinsic interpretation ofthe Brill-Noether matrix which holds regardless
of whether D consists of distinct points or not. There is really nothing to this.
Just recall the basic abelian sum mapping

u: cd--+ J(C)

defined by

u(D) = (... , ua(D), ... ),

where
d
D = LPi,
i= 1

and p0 is a base point, chosen once and for all. It is apparent that, in case D
is the sum of d distinct points, the Brill-Noether matrix is just the Jacobian
matrix of u relative to the coordinates ua on J(C) and the coordinates

i = 1, ... , d,

on Cd. Thus the Brill-Noether matrix can be thought of as giving a local

coordinate representation of the homomorphism of locally free sheaves

We now define C;j as the (d- r)th determinantal variety attached to u*, the
one whose support is

As we just remarked, away from the diagonals, we have

supp(C:i) = {DE Cd: r(D) ~ r},

156 IV. The Varieties of Special Linear Series on a Curve

as desired. We now show that this is the case everywhere by proving the
following fundamental

(1.1) Lemma. r(D) ~ r if and only (f the rank of u* at D does not exceed
d - r. In particular, the support of C:i is precisely the set {DE Cd: r(D) ~ r},
and the fibers of u are linear series (scheme-theoretically).

Here we shall proceed by direct computation; a more intrinsic proof will

be found in the third section of this chapter. One implication is trivial: since
u - 1 (u(D)) = ID I is a projective space of dimension r(D), the rank of u* at D
does not exceed d - r(D). The other implication has already been proved
when D consists of d distinct points.

Before treating the general case we shall recall some elementary properties
of symmetric functions. Consider the elementary symmetric functions in h
variables:

aj(x 1 , ... , xh) = I

it< ... <ij
xi,Xi 2 • • • xii• j = 1, ... ' h.

It is well known that these functions generate the ring of all symmetric
functions in h variables. For our particular purposes we shall need another
set of generators: the Newton functions. These are symmetric functions
defined by

i = 1, 2, ...

We shall show that the first h Newton functions generate the ring of all
symmetric functions, by showing that each ai, i = 1, ... , h, can be expressed
as a polynomial in the v/s, i = 1, ... , h. We have

h h
1+ I ajtj = f1 (1 + xjt).
j= 1 j= 1

Hence

log(1 + J jt J
1
ajtj) = 1 1
(-1Y+ 1 ~} ti
= I<-1r1 ~ ti.
i= 1 I

Therefore

(1.2)
§1. The Brill-Noether Matrix and the Variety C:i 157

Let us now use the following notation. Given a set y of infinitely many
variables

Y = (y1, Y2, YJ, · · .),

define functions ~j(y) = ~j{y 1 , y2 , ••• , yj) by setting

1
~j(y) = I h 1 '. . . .
h
h.,Y1'···Y/·
J •
h·

h,+2h2+ ... +jhj=j

A straightforward formal computation shows that

Looking at (1.2) we can then conclude that

j = 1, ... , h,

proving that each aj can be expressed as a polynomial in v1 , ... , vh.

Let us make a further observation. Consider the Newton functions
vn = vn(x 1 , ••• , xh) for n ~ h + 1. By what we just proved these are poly-
nomials in v1 , ... , vh. We claim that these polynomials contain no linear
term. In symbols

(1.3) n~h+l.

By recursion it really suffices to show that vn belongs to (v 1 , ... , vn_ 1 ) 2 , for

n ~ h + 1. Looking at the coefficient of tn in (1.2) we get

On the other hand, the very definition of ~n tells us that

A ( V1•···•-
tln
( l)n+ 1 -Vn) 1 (-
-- l)n+ 1 Vn ( )2
- E V1, ... ,Vn-1'
n n! n

proving our claim.

We shall finally need -the following remark.

158 IV. The Varieties of Special Linear Series on a Curve

Lemma. Let g(w) be a holomorphic function in a neighborhood of 0 E C.

Consider the Newton functions

i = 1, ... , h.

Write y(vl> ... , vh) = If= 1 g(w;). Then

Proof. We have

so that

and the lemma follows from (1.3). Q.E.D.

We now go back to the proof of Lemma (1.1) and prove that the rank of
u* at DE Cd equals d - r(D). Write D = L
hipi, where p 1 , ... , Pn are distinct.
For each i, let zi be a local coordinate centered at Pi· We take as local co-
ordinates on Cd, in a neighborhood of D, the functions

i=ht+l, ... ,ht+h2,

where vi stands for the ith Newton function. Write

where f~i is a holomorphic function in a neighborhood of Pi. We now compute

the Jacobian matrix (ou~fo(;) at D, applying the lemma to the abelian sums
§1. The Briii-Noether Matrix and the Variety Cd 159

viewed as functions of ( 1 , ... , (d. Since near P; the derivative of ua.(q) with
respect to z; is /a.;, the Jacobian (oua./8(;) at Dis

1
!tt(O) JW(O) · · · (il; ~ 1)! /~ht'- 1)(0) ftz(O) JW(O) (h 2 -1)! JW-I>(O)

1 (h,- I) 1 (h2- I)
.fgl (0) .f~\l(O) (hi -1)! Jgl (O) .fgz(O) .f~WO) (hz -1)! fgz (0)

"· h,

To say that the rank of this matrix is d - r(D) is just to say that there are
exactly g - d + r(D) independent relations amongst its rows, that is, there
are exactly g - d + r(D) linearly independent holomorphic differentials
vanishing at D: this is the Riemann-Roch theorem.
The above matrix will also be called the Brill-N oether matrix. Sometimes
we shall improperly use this name to denote the sheaf homomorphism u*
as well.
The determinantal description of the variety C:i provides a lower bound
on its dimension. Since Cd is locally defined by the simultaneous vanishing
of all (d - r + 1) x (d - r + 1) minors of a d x g matrix of holomorphic
functions, as we observed in Chapter II, Section 4, the codimension of C:i
can be at most

[d - (d - r)} [g - (d - r)] = r(g - d + r).

Thus every component of C:i has dimension at least

d - r(g - d + r) = r + g - (r + l)(g - d + r).

The number

p(g, r, d) = g - (r + l)(g - d + r),

which is of fundamental importance in the study of special divisors, is called

the Brill-Noether number and will usually be denoted simply by p, unless
confusion is likely. Thus we can say that every component of C:i has dimension
at least equal to the "expected" dimension p + r, where pis the Brill-Noether
number.
It is now instructive to determine the Zariski tangent space to q at a
divisor D not belonging to c;;+ 1, for this computation brings to the fore-
ground again the basic cup-product mapping Jlo, which we already
encountered in the proof of Clifford's theorem. Here we shall give only an
L
elementary argument covering the caseD = p;, where the p;'s are distinct;
160 IV. The Varieties of Special Linear Series on a Curve

the whole question will be reconsidered, from a slightly less elementary, but
more natural, point of view, in Section 4 of this chapter.
We first choose local coordinates zl> ... , zd on C, centered at Pt> ... , pd,
respectively. The z;'s can also be viewed as local coordinates on Cd, centered
at D. Next, when I q; is close to D, we denote by ¢(I q;) the transpose of the
Brill-Noether matrix at I q;, i.e., we set

Denoting by M the variety of d x g matrices, we shall view¢ as an M-valued

mapping defined in a neighborhood of D. By definition, in this neighborhood,
q is just the pull-back, via ¢, of the determinantal subvariety Md-r of all
matrices of rank not exceeding d - r. In particular, the Zariski tangent
space to q at D is the pull-back, under the differential of¢, of the tangent
space to Md-r at A = ¢(D). This has been described in Chapter II, Section
2, where it is shown that

under the assumption that A does not belong to Md-r-l· This is true in our
case, since we are assuming that D does not belong to C:i+ 1 . We then have
to describe, in intrinsic terms,

Tv(q) = ¢;;: 1 (TA(Md-r))

= {v E Tv(Cd): ¢*(v) · ker A c Image A},

where A = ¢(D). To begin with, let v = I v; ojoz; be a tangent vector to

Ca at D. To v we may associate

viewed as a section of cPv(D). This sets up an isomorphism

The important fact which can be checked by a trivial calculation, is that this
isomorphism is intrinsic, i.e., does not depend on the choice of coordinates.
Actually Tv(Cd) and H 0 (C, cPv(D) are intrinsically isomorphic even if D
contains multiple points. In keeping with our "naive" approach, we shall
§1. The Brill-Noether Matrix and the Variety Cd 161

postpone a proof of this, together with a global description of the tangent

bundle to cd• to a later section of this chapter. If v = LV;
ojoz; is as above,

where w;(P) stands for the derivative of w;/dzi at Pi with respect to zi. On
the other hand, if we identify the CO on which A operates with H 0 (C, K), it
is clear that A "is" the restriction mapping

Thus, recalling the identification given above between Tv(Cd) and

H 0 (C, mv(D)), we may identify ¢*(v) · ker A with the image, under the cup-
product homomorphism

of

v ® H 0 (C, K( -D)).

It follows that

¢*(v) · ker A c Image(A)

if and only if for any w belonging to H 0 (C, K(- D)),

f3(v ® w) = a(w')

for some w' in H 0 ( C, K). In other terms, if we let

be the composition of f3 with the coboundary mapping

v belongs to Tv(C~) if and only if

{3'(v ® w) = 0 for any wE H 0 (C, K( -D)).

162 IV. The Varieties of Special Linear Series on a Curve

This condition is most conveniently expressed by saying that

(s, [3'(v ® m)) =0

for any m belonging to H 0 (C, K( -D)) and any s belonging to H 0 (C, (!)(D)),
where< , ) stands for the Serre duality pairing. Now look at the commuta-
tive diagram

where the horizontal rows are cup-product homomorphisms. Clearly,

(s, [3'(v ® m)) = (s, [3"(15v ® m))

= (,u 0 (s®m),l5v)
= (rt.,u 0 (s®m),v),

where

is the cup-product mapping and

is the restriction mapping. Putting everything together, we conclude

(1.5) Lemma. If D belongs to Cd but not to c:/ 1, the tangent space to C:i at
Dis

where ,u 0 is the basic cup-product mapping (1.4) and rt. is the restriction mapping
from H 0 (C, K) to

Of course, this lemma has been proved, so far, only under the additional
assumption that D has no multiple points; it is, however, valid in general.
§1. The Brili-Noether Matrix and the Variety C:i 163

It is now very easy to say when Dis a smooth point of C~. Notice that the
kernel of r:x is just H 0 (C, K(- D)), so that, in particular, it is contained in the
image of Jlo. Thus

dim Tv( C;j) = d - dim(Image r:xJ1 0 )

= d- dim(lmage Jlo) + h0 (C, K( -D))
= d- (r + l)(g- d + r) + dim(ker Jlo) + (g- d + r)
= p + r + dim(ker J1 0 ).
As we already observed that every component of q has dimension at least
p + r, we reach the following conclusion.

(1.6) Lemma. The variety q is smooth and has the "expected" dimension

p +r= g - (r + 1)(g - d + r) + r
at DE q - c~+ 1 if and only if the cup-product homomorphism

is injective.

At this point, one might ask under what circumstances is C:i different
from q+ 1 • The answer is that this is always the case, unless the contrary is
forced "by the numbers." More exactly we have the simple but basic

(1.7) Lemma. Assume g - d + r 2: 0. Then no component of C;j is entirely

contained in C:i + 1 .

Proof. We shall argue by contradiction. Let D be a general point of a com-

ponent of q and assume r(D) > r. If q is a point of C there is an effective
divisor E such that Dis linearly equivalent toE + q. Moreover, if q is general
(more exactly, if q is not a base point of 1D 1),

r(E) = r(D) - 1 2: r,

and, by the Riemann-Roch theorem,

i(E) = i(D) = g- d + r(D) 2: 1.

Thus, if p is a general point of C, pis not a base point of 1K(- E) I, and hence

i(E + p) = i(E) - 1,
r(E + p) = r(E) = r(D) - 1 2: r,
164 IV. The Varieties of Special Linear Series on a Curve

by the Riemann-Roch theorem. On the other hand, by the generality of p

and q, E + p belongs to the same component of q as D. This contradicts
the generality of D. Q.E.D.

§2. The Universal Divisor and the Poincare Line Bundles

In the previous section we have introduced the Brill-Noether matrix and
explained, in elementary terms, its significance in the study of special divisors.
Now the time has come to give more solid foundations to our study of the
symmetric products of a curve and their relation to the Jacobian variety.
The first new concept we wish to introduce is the one of universal divisor.
As usual, we let C be a fixed smooth curve of genus g. The universal effective
divisor of degree d on C is the divisor

which, for any Din Cd, cuts on

C ~ C x {D}

exactly the divisor D. To convince ourselves that this prescription really

defines a divisor, and for later use, we shall now give explicit local equations
for A. Let D = IJ= 1 P; be a point of Cd. Choose a coordinate ( on C in a
neighborhood U of {p 1 , ••. , Pd}. In a neighborhood V of D there are co-
ordinates

defined by the property that the natural morphism

is given, in a neighborhood of (p 1 , ... , Pd), by

u;(y(q 1, ... , qd)) = ith symmetric function of (((q 1), ... , ((qd)).

Then A is described, in U x V, by the equation

F((, u) = 0,

where
§2. The Universal Divisor and the Poincare Line Bundles 165

We would now like to explain the adjective "universal" attached to the

divisor ~- First of all, we recall what is meant by a relative divisor in C x S
over an analytic space S. This is simply an effective (Cartier) divisor on
C x S which does not contain fibers of the projection onto S. We shall say
that such a divisor has degree d if it cuts a degree d divisor on each fiber of
this projection. Then the universal property of~ is the following:

(2.1) Lemma. For any analytic spaceS, any relative degree d divisor inC x S
over S is the pull-back (lc x f)*(~) of the universal divisor for an unique
morphism

In other words, Cd represents the functor

G: {analytic spaces} ~ {sets}

defined by

G(S) = {relative degree d divisors in C x S overS}.

The proof of the lemma is very simple. Let D be a relative degree d divisor
in C x S over S. First of all, since the problem is local on S, we may assume
that D is supported in U x S, where U is a coordinate open subset of C,
and that D is defined, in U x S, by a global equation ¢. Next, if ~ is a co-
ordinate in U, using the Weierstrass preparation theorem we may write, in
a unique way,
¢ = hg,

where h and 1/h are holomorphic and

is a Weierstrass polynomial in ~-It is then clear that the only map

such that
(lc x f)*(~) = D

is defined, in the local coordinates a; on Cd introduced in the definition of~.

by
i = 1, ... ' d.
This completes the proof of the lemma.
166 IV. The Varieties of Special Linear Series on a Curve

As we know Pied( C) parametrizes degree d line bundles on C. The first

use of the universal divisor .1: will be in showing that actually there is, on
C x Pied( C), a universal line bundle, in a sense which we shall now make
precise. By a Poincare line bundle of degree dfor C we shall mean a line bundle
ff> on C x Pied( C) which, for each Lin Pied( C), restricts exactly to Lon

C ~ C x {L}.

If !f>, ff>' are two Poincare line bundles, we may write, in a unique way,

ff>' = ff> ® v*~,

where

is the projection, and f!ll is a line bundle on Pied(C). It suffices to take

In fact, by the very definition of Poincare line bundle, ff>' ® !f>- 1 is trivial
on each fiber of v, hence f!ll is a line bundle. Moreover,

is trivial. Hence

has a global section which does not vanish identically on any fiber of v. Since
the above line bundle restricts to the trivial bundle on every fiber of v, this
global section never vanishes. This shows that ff> ® ff>'- 1 ® v*f!ll is trivial,
as desired.
Of course, it is not clear, at this point, that Poincare line bundles exist.
To show that this is the case is our next task. We begin by noticing that it
suffices to construct a Poincare line bundle for one d. In fact, let ff> be one
such, and let e be an integer. Fix a line bundle La on C of degree d - e. We
get an identification

a: Pice( C) - Pied( C)

by setting

a(L) = L ®La.
§2. The Universal Divisor and the Poincare Line Bundles 167

Then, clearly,

is a Poincare line bundle of degree e, where

is the projection.
In constructing Poincare line bundles, we may therefore limit our attention
to large values of d. In particular, we shall assume that d ;;::: 2g - 1, so that the
fibers of

are all projective spaces of dimension d - g. For any point q in C we consider

the divisor in Ca:

Xq = {q + D': D' E Ca-d·

We then set

where

is the projection. We claim that 2 is a Poincare line bundle of degree d. To

see this, fix a point, say (!)(D), in Pica(C); all we have to do is show that
(!)(d - n* X q) restricts, on C x ID I, to the pull-back of (!)(D) from C. But
this is clear, since on one hand, for every q' in C, d and n* X q cut on the
projective space

IDI ~ {q'} X IDI

the divisors X q' n ID I and X q n ID I, which are both hyperplanes, while

(!)(d- n*Xq) restricts to (!)(D) on C x {D}.
Having established that there exist Poincare bundles, we may notice, in
passing, that for any q in C there is a unique degree d Poincare line bundle
whose restriction to {q} x Pica( C) is trivial.
Poincare line bundles have a universal property, which is expressed by
the following
168 IV. The Varieties of Special Linear Series on a Curve

(2.2) Lemma. Let 2'd be a degree d Poincare line bundle for the smooth curve
C. Let S be an analytic space and let 2' be a line bundle over C x S such that,
for each s in S, 2' lc x {sJ has degree d. Then there exists a unique map

such that the pull-back, via lc x f, of 2'd is of the form 2' ® ¢*~, where ~
is a line bundle on S and ¢ is the projection of C x S onto S.

Proof. As in the construction of the Poincare line bundle, we may and will
assume that d ~ 2g - 1, so that Cd is fibered over Picd(C) in projective
spaces of dimension d - g. The problem is local on S. In fact, assume the
lemma holds for any "sufficiently small" S; we can find covers {SJ of S by
arbitrarily small open subsets such that Pic(Si) is trivial. Then if we can find
morphisms

and isomorphisms

then the/;'s patch together into a morphism f from S to Picd(C), by unique-

ness, and moreover, since Cis complete,

is a never vanishing holomorphic function gii on Si n Sj alone. Furthermore,

so that the gii's are the transition functions of a line bundle ~ on S. The
~/scan then be viewed as local expressions of a global isomorphism

We may then assumeS to be "small," and, in particular, that Pic(S) is

trivial. The existence off is then clear, for we may assume that 2' has a
global section s which does not vanish identically on any fiber of¢. Then s
defines a relative degree d divisor, which corresponds, by the universal
property of the universal divisor ~. to a morphism of S into Cd. Composing
this morphism with u gives f.
As for the uniqueness off, this is clear over S,ea by the very definition of
Poincare bundle. Thus iff' is another map of S into Picd(C) under which
2'd pulls back to 2', we may assume that both f and f' map into an open
§2. The Universal Divisor and the Poincare Line Bundles 169

subset U ofPicd(C) over which Cd is a trivialiP'd- 9-bundle. Let g, g' be liftings

off, f' to maps into Cd. Also, let D, D' be the corresponding liftings of Sf to
relative divisors on C x S. Now, to give a lifting of either for Sf corresponds
to choosing a section over S of

Thus there is a lifting g" off whose corresponding relative divisor is D'. But
then, by the universal property of the universal divisor, g" = g', and hence
f = f'. Q.E.D.
By a family of degree d line bundles over C parametrized by an analytic
space S we shall simply mean a line bundle on C x S which restricts to a
degree d line bundle on C x {s }, for any sinS. We shall say that two families
Sf and Sf'' of degree d line bundles on C parametrized by S are equivalent if
there is a line bundle !!It on S such that

where¢ stands for the projection of C x S onto S. In this language the lemma
we just proved simply says that Picd(C) represents the functor

F: {analytic spaces}---+ {sets},

F(S) = {equivalence classes of families of degree d

line bundles on C parametrized by S}.

We would finally like to give an intrinsic description of the tangent bundle

to Cd and of the Brill-Noether homomorphism. The first step is to make
explicit the canonical identification between the tangent spaces to Picd(C)
and H 1 (C, CD) provided by the (non-canonical) isomorphism between
Pica( C) and H 1 (C, C9)/H 1 (C, Z).
Let L be an element of Picd(C). Recall that TL(Picd(C)) can be described
as the set of morphisms of pointed analytic spaces

Spec C[s] = S---+ (Picd(C), L),

where IC[s] = IC[T]/(T 2 ) is the ring of dual numbers. In other words

Now let us define a .first-order deformation of L to be a family Sf of line

bundles on C parametrized by S = Spec IC[a] such that

j*(!l') = L,
170 IV. The Varieties of Special Linear Series on a Curve

where j is the inclusion of C in C x S. By the universal property of the

Poincare bundle there is a natural 1-1 correspondence between TL(Picd( C))
and the set of equivalence classes of first-order deformations of L.
Now suppose that Lis given by transition data ({U ~}, {g~p}), where

is an open cover of C and the functions

are transition functions for L. Thus, in triple intersections the coqrcle rule

holds.
We may now write a first-order deformation !l' of L as being given by
transition functions

defined in (U ~ n Up) x S, where ¢~/l is holomorphic on U ~ n Up. Again we

have the cocycle rule

It is easy to see that this cocycle rule is equivalent to

This just says that the 1-cochain {¢~p} is in fact a cocycle, and hence defines
a class

Next, recall that the identification between Picd(C) and H 1(C, f9)/H 1 (C, Z) is
provided by the exponential sequence

It follows that the element of H 1 (C, @) associated to the tangent vector

corresponding to !l' is simply the class of the logarithmic derivatives, with
§2. The Universal Divisor and the Poincare Line Bundles 171

respect to s, of the transition data of ff, divided by 2nj=l. But

1 d - 1 -ld
2nj=l ds log gap = 2nj=l gap ds (gap(1 + s</>ap))
1

Thus the element of H 1 (C, CD) corresponding to the tangent vector defined
by ff is <J>/2nj=l. In what follows, however, we will find it more convenient
to associate to this tangent vector the class </>, which will be called the
Kodaira-Spencer class of the first-order deformation .!l'.
We have finally assembled all the elements which are necessary to describe
the Brill-Noether homomorphism in global, intrinsic terms. As usual, we
denote by

the projection and by d the universal divisor inc X cd. Consider the exact
sequence of coherent sheaves on C x Cd

0 --> (9 --> IP(d) --> (9 8 (d) --> 0,

and the corresponding exact sequence of direct image sheaves

We then have the following result, the embryo of which we encountered in

the first section of this chapter.

(2.3) Lemma. There are isomorphisms,</> and 1/J, making the following diagram
commutative

so that the Brill-N oether homomorphism may be identified with the coboundary
homomorphism b.

Proof. The identification </> between u*E>Picd(C) and R 1 n* (9 is given by the

Kodaira-Spencer class. As for 1/J, let X be a vector field on an open subset
172 IV. The Varieties of Special Linear Series on a Curve

U of Cd. The field X lifts, via n, to a vector field X on C x U. Now, ifF 1 = 0,

F 2 = 0 are local equations for ~ in an open subset of C x U, then

where G is a non-vanishing function. Thus

X(F 1 ) X(F 2 ) . .
(2.4) - - = - - + (a holomorphtc functiOn).
F1 Fz

Therefore, if {t;,.} is an open cover of C x U and, for each ()(, F ~ = 0 is a local

equation for~ in t;,., the cocycle

{----p:;-
X(F~)} E C ({t;,.}, (f)(~)lcxu)
0

determines, by restriction to ~. a section

t/Ju(X) E r(C x U, (f)d(M).

Formula (2.4) also shows that the collection of homomorphisms {t/Ju}

defines a sheaf homomorphism

t/1: E>c. ~ n* (f)d(~).

We now show that <f>u* = Jtf;. Let D = Lf= 1 P; be a point of Cd, and let
F = F((,o') = 'd- (J'1'd-1 + ... + (-l)dO'd
be a local equation for~ near {p 1 , ... , pd} x {D}, as in the formal definition
of~- Consider a tangent vector

a
X='[_a.-
' 80';

to Cd at D. This corresponds to a morphism

~:Spec C[t:] ~ (Cd, D).

By definition, <f>u*(X) is the Kodaira-Spencer class of the pull-back bundle

Oc x ~)*(()(~).To compute this class consider the pull-back ofF via lc x ~.
namely

let U be a small neighborhood of {pb ... , pd}, and set V = C - {pb ... , Pd}.
Then G is the transition function of the bundle Oc x ~)*(f)(M relative to the
§2. The Universal Divisor and the Poincare Line Bundles 173

cover {U', V'}, where U' = U x Spec C[a], V' = V x Spec C[a]. Therefore
the Kodaira-Spencer class ¢u*(X) of Clc x ~)*C9(M is the class of

{G- ~GI } = {F- XCF>I }

1
8 UnV
1
UnV
E C 1 ({u, v}, (9).

On the other hand, it follows directly from the definitions that

This proves the commutativity of our diagram.

It remains to show that ljJ is an isomorphism. Let g be a holomorphic
function in a neighborhood of 11 n (C x U), where U is an open set in Cd
containing D. Using Weierstrass' division theorem, we may write

g = hF + R,
where h is holomorphic and
d
R = LC-lYb;(oXd-i.
i= 1

Then the section of n*@&(M over U determined by g/F is equal to l/J(Y),

where
a
Y= Lb;(o)-.
o<J;

This shows that ljJ is onto. Conversely, if Y is as above,

Y(F) L1=1( -1Yb;(<J)(d-i

F F
can be holomorphic only when all the b;'s vanish. This completes the proof
of the lemma, achieving the description of the Brill-N oether homomorphism
in cohomological terms. Q.E.D.

A final and important comment is in order. Lemma (2.3) shows that the
co kernel of u* is just R 1 n* C9(M. In fact, since R 1 n* (9&(11) is zero, a piece of
the exact cohomology sequence of

lS

(2.5)
174 IV. The Varieties of Special Linear Series on a Curve

This provides a global presentation of R 1 n*(!)(d), i.e., exhibits R 1n*(!)(d)

as the co kernel of a homomorphism of vector bundles. It is crucial to notice
that the above presentation is functorial in the following sense. Let

be a morphism, let d' be the corresponding relative divisor in C x S and

denote by ¢ the projection of C x S onto S. Then we have an analogue of
(2.5), namely

Using the Weierstrass preparation theorem, as we did for the universal

divisor d, it is easy to show that ¢* (!),.,{d') is locally free and that, for any s
inS,

where d~ is the divisor cut on C ~ C x {s} by K On the other hand, by

pull-back, we get another presentation

Clearly the natural morphism

is an isomorphism. As for

it is also an isomorphism, since the two sheaves in question are locally free
and, for any s in S,

Summarizing, we have natural identifications

such that b' = f*(b). It follows, in particular, that

§2. The Universal Divisor and the Poincare Line Bundles 175

An important special case is the one in which f is the inclusion in Cd of a

point D. What we have just said means that H 1 (C, (!)(D)) is the cokernel of the
homomorphism

This provides another proof of the statement that the rank of u* at D does
not exceed d - r if and only if

i(D) ~ g- d + r,
or, which is the same, if and only if

r(D) ~ r

(cf. Section 1 of this chapter).

To put the functorial presentation (2.5) in proper perspective, we recall
the basic result about base change in cohomology

(2.6) Theorem. Let f: X--+ Y be a proper morphism of analytic spaces, and

let fF be a coherent (!)x-module. Suppose fF is (!)y-jlat. Then,for every point of
Y there are a neighborhood U and a finite complex

of free finitely generated {!}u-modules, with the following property. For every
morphism of analytic spaces g: V--+ U, there are functorial isomorphisms

where f3 is the projection of f- 1(U) xu V onto V and rx is the composition of

the projection off - 1 (U) x u V onto f- 1 (U) with the inclusion off - 1 (U) in
X. Moreover, if the dimension of all the fibers off does not exceed n, we may
assume that Km = 0 when m > n.

It is important to notice that Theorem (2.6) is still true (and has, in fact, a
much simpler proof) if we replace the words "analytic space" with "algebraic
variety," and "neighborhood" with "Zariski neighborhood."
Applying Theorem (2.6) to the morphism

and to the sheaf (!)(A), we get, locally on Cd, a functorial presentation of

R 1 n* (!)(A) (and one whose kernel is n* (!)(A), as well). In a sense, then, at
least for what concerns R 1n* (!)(A), the functorial presentation (2.5) is a
global and explicit version of the functorial cohomology complex whose
existence is asserted by (2.6).
176 IV. The Varieties of Special Linear Series on a Curve

§3. The Varieties WJ(C) and Gd(C) Parametrizing Special

Linear Series on a Curve
Our first goal, in this section, is easily stated. We fix a smooth genus g curve
C and two non-negative integers r, d; we would then like to construct a sub-
variety W~( C) of Pied( C) whose support is the set of complete linear series
of degree d and dimension at least r. In other words, we would like to have

Supp(W~(C)) = {L E Picd(C): h 0 (C, L) 2: r + 1}.

This is certainly possible since the above set is the image of q under the
proper map

the problem is to choose the "right" scheme structure.

We shall denote by!!' a fixed Poincare line bundle of degree d for C, and
by v the projection

Also, we choose, once and for all, an effective divisor E of degree

m = deg(E) 2: 2g - d - 1

on C, and let

be the product divisor inC x Pied( C). By Theorem (2.6), the hypothesis on
the degree of E guarantees that

and that v* !f'(r) is locally free of rank

n = rank(v*!f'(r)) = d + m- g + 1.
Thus the exact higher direct image sequence of

is
§3. The Varieties W~(C) and G~(C) Parametrizing Special Linear Series on a Curve 177

The two middle terms are locally free of ranks n and m, and it is natural to
define W:i(C) as the locus "where the fiber of R 1 v*f!! has dimension at least
g - d + r," or, more precisely, as the (m + d - g - r)th determinantal
variety attached to y. In the language of Chapter II, Section 4, and setting,
for brevity

we have

W:i(C) = Xm+a- 9 -rCy).

We shall normally write lifd(C) instead of W~(C). We shall later see that this
lifd(C) is the same as the one already defined in Chapter I. We also let

be the canonical blow-up of W:i(C), which was also defined in Section 4 of

Chapter II.
We shall return to G:i(C) later in this section. For the moment, we have
to justify our definition of W:i(C), in the sense that the support of W:i(C) is
precisely the set of those L in Pica( C) such that h0 ( C, L) is at least r + 1. We
can do much better. For brevity, we set

We have seen that the kernel and cokernel of

are v*(f/!) and R 1 v* !!!, respectively. The analogue ofthis is true for any family
of degree d line bundles on C. Namely, letS be any analytic space and let L
be a family of line bundles on C parametrized by S. Denote by ¢ the projec-
tion of C x S onto S, and set r' = E x S. As before, there is an exact sequence

Now, by the universal property of Poincare bundles, there are a unique map

and a unique line bundle fJl on S such that

(lc x f)*!!! ~ L @ ¢* !Jl.

We then have the following

178 IV. The Varieties of Special Linear Series on a Curve

(3.1) Lemma. There are canonical isomorphisms between the kernel and
cokernel of

and 4>*(L) ® f!Jt, R 1 ¢*(L) <8) f!Jt, respectively. In particular

The proof of this lemma is straightforward. Applying Theorem (2.6) to

the projection

and to the sheaf 2(r) shows that the natural homomorphism fromf*(K 0 )
to

is in fact an isomorphism. On the other hand, it is clear (or, for skeptics,

follows from (2.6)) that the natural homomorphism fromf*(K 1 ) to

is also an isomorphism. We then have a commutative diagram

f*(K 0 ) ~ f*(K 1 )

~,(L(r')) l ~ ~.lr')/L)
111 ® 111.

Since the kernel and cokernel of y' ® 1~ are, respectively, ¢*(L) <8) f!Jt and
R 1 ¢*(L) ® f!Jt, the lemma follows.

As a special case, we look at the one when S is a single point. Thus L is

simply a degree d line bundle on C, and f maps S to the corresponding point
of Picd(C). Then the lemma simply says that

h1 (C, L) = m - rankL(y).
Therefore we have

Supp(W~(C)) = {LE Picd(C): rankL(Y) ::::;; m + d - g - r}

= {L E Picd(C): h 1(C, L) ~ g- d + r}
= {L E Picd(C): h 0 (C, L) ~ r + 1},
as desired.
§3. The Varieties W;j(C) and G;j(C) Parametrizing Special Linear Series on a Curve 179

There seems to be a certain degree of arbitrariness in the definition of

W:i( C), coming from the fact that we had to choose both a Poincare line
bundle and an effective divisor of high degree on C. This is not a serious
problem, however. Changing the Poincare line bundle has the effect of
tensoring both K 0 and K 1 with the same line bundle. This clearly does not
change the determinantal varieties attached to y. As for our second choice,
let q be a point of C, set E' = E + q, r' = E' x Picd(C), and look at the
homomorphism

y': v* 2'(r') ~ v*(2'(r')/2').

Choose a local frame for v* 2'(r') by adding to a frame for v* 2'(r) a section
s ofv* 2'(r') which induces a nowhere vanishing section ofv* 2'(r')/v* 2'(r).
Then choose a frame for v*(2'(r')/2') by adding to a frame for v*(2'(r)/2')
the image of s under y'. Relative to these frames, the homomorphism y'
is represented by a matrix of holomorphic functions

, (1 B)
A= 0 A'

where A represents y. Thus the minors of size m + d - g - r + 1 of A are

minors of size (m + 1) + d - g - r + 1 of A'. Conversely, every minor of
size (m + 1) + d - g - r + 1 of A' is a linear combination, with holo-
morphic coefficients, of minors of size m + d - g - r + 1 of A. Thus the
varieties W:i(C) defined by choosing, as fixed effective divisor, E orE', are
the same. Adding one point at a time, we can thus show that, for any effective
divisor E" of degree at least 2g - d - 1, we get the same W:i( C) if we choose,
as fixed divisor, E orE + E". Reversing the roles of E and E" we conclude
that the definition of W:i( C) is independent of the choice of E.

(3.2) Remark. The independence of W:i( C) of the choices made, which we

checked by direct computations, also follows from general properties of
Fitting ideals. We shall describe them here, without proofs, although they
will not be strictly necessary in the sequel. For more details the reader is
referred to Fitting [1], Northcott [1].

Let X be an analytic space and let Jii be a coherent analytic sheaf on X.

A presentation of Jii is an exact sequence

The homomorphism rx is represented by an m x n matrix A whose entries

are holomorphic functions on X. The hth Fitting ideal of the above presenta-
tion is the ideal in {!Jx generated by the (m - h + 1) x (m - h + 1) minors
of A in case h :::;; m, and is defined to be {!Jx when h > m.
180 IV. The Varieties of Special Linear Series on a Curve

The basic property of Fitting ideals is that any two presentations of :!i'
have the same Fitting ideals. This will enable us to speak of the Fitting ideals
of :!i' without reference to any particular presentation, and also makes it
possible to define the Fitting ideals in the absence of a presentation for :!i'.
In fact, :!i' will always have local presentations, and the hth Fitting ideals of
these patch together to yield a global ideal in (!)x, by the basic property of
Fitting ideals. This ideal will be called the hth Fitting ideal of :!i'.
An easy property of Fitting ideals is that they are compatible with base
change. This means that, whenever,

f: y--+ X

is a morphism of analytic spaces, the hth Fitting ideal off* :!i' is generated,
as an {!}y-module, by the hth Fitting ideal of :!i'. This is simply a reflection of
the fact that the pull-back of a presentation of :!i' is a presentation off*(:!i'),
since the tensor product is a right exact functor.
In the language of Fitting ideals, then, W~( C) is simply the subvariety of
Picd(C) defined by the (g- d + r)th Fitting ideal of R 1 v*!£.
Let :!i' be a coherent analytic sheaf over the analytic space X. The Fitting
rank of :!i' is defined to be the largest integer h such that the hth Fitting ideal
of :!i' vanishes. A moment of reflection will then convince the reader that
Lemma (3.1) essentially says that W~(C) represents the functor

equivalence classes of families L of degree d line bundles on}

{
S--+ C x S ~ S such that the Fitting rank of R 1 ¢*L is at least .
g-d+r

The variety W~(C) has been defined as the (m + d - g - r)th deter-

minantal variety attached to a homomorphism between suitable vector
bundles of ranks m + d - g + 1 and m, respectively, over Picd(C). Thus
(cf. Chapter II, Section 4) every component of W~(C) has dimension at least
equal to

g - [(m +d- g + 1) - (m +d - g - r)][m - (m +d- g - r)]

= g - (r + 1)(g - d + r),
provided, of course, that
r 2 d- g.

We then see that the "expected" dimension of W~(C) is exactly the Brill-
Noether number
p = g - (r + 1)(g - d + r),
which we first met in the previous section. Because of its importance, we
express this as a formal lemma.
§3. The Varieties W:i(C) and G:i(C) Parametrizing Special Linear Series on a Curve 181

(3.3) Lemma. Suppose r ~ d- g. Then every component of W~(C) has

dimension greater or equal to the Brill-Noether number

p(g, r, d).

We remark, in passing, that, when L belongs to W~(C), but not tow~+ 1 (C),
the Brill-Noether number is just

g- l(L)i(L) = g- h0 (C, L)h 0 (C, KL - 1 ).

In a sense, then, the Brill-Noether number is invariant under the operation

of passing from a linear series to the residual one.
We may now ask what is the precise relation between c:; and W:i(C). Of
course, as sets, the second is just the image of the first under the abelian sum
mapping

The question then is, what can we say if we also take the scheme structures
of c:; and W:i( C) into account? The following result clarifies the situation.

(3.4) Proposition. u- 1(W:i(C)) = c:;.

Proof. We let n be the projection of C x Cd onto the second factor, and let
!l' be a Poincare line bundle of degree d for C. We set

L = Clc x u)*(!l').

By Remark (3.2) and Lemma (3.1), the ideal sheaf of u- 1 (W:i(C)) is the
(g- d + r)th Fitting ideal of R 1 n*(L). Now let A be the universal divisor
in C x Cd. By the definition of Poincare line bundle, L ®(!)(-A) is trivial
on the fibers of n, and hence there is a line bundle f7l on Cd such that

Since tensoring with a line bundle obviously has no effect on the Fitting
ideals, this in turn means that the ideal sheafofu- 1 (W:i(C))is the(g- d + r)th
Fitting ideal of R 1 n* @(A). On the other hand, the discussion at the end of
Section 2 shows that this is precisely the ideal of c:; (see, in particular, the
presentation (2.5)). Q.E.D.

The preceding result shows, in particular, that the restriction of u to

C:i factors through the inclusion of W:i( C) in Pied( C).
Lemma (1.6) has an exact analogue for W;j(C).
182 IV. The Varieties of Special Linear Series on a Curve

(3.5) Lemma. Suppose g- d + r ~ 0. Then no component of W~(C) is

entirely contained in w~+ 1 (C).

This is just a corollary of (1.6). In fact, if a component X of W:i(C) were

contained in w:;+ 1 (C), u- 1(X) would contain a component of q, which
would then be contained ih u- 1 (w:;+ 1 ( C)) = q+ 1 .

We now turn to the variety G:i(C) which we introduced, along with

W:i(C), as the canonical blow-up of W:i(C) itself. It will turn out that G:i(C)
parametrizes linear series of degree d and fixed dimension r on C, so that
we will find it much more manageable than W:i(C) in studying mappings
from C to IP''.
To see that the points of G:i(C) are, in fact, the g:i's on C, recall the two-
stage complex

of locally free sheaves over Picd(C) used in the definition of W:i(C) and G:i(C).
The points of G:i(C) are the couples (x, V), where x belongs to Picd(C), and
Vis an (r + i)-dimensional subspace of the kernel of

Now, x corresponds to a degree d line bundle L on C, while, by Lemma

(3.i), the kernel of Yx is canonically isomorphic to H 0 (C, L). Thus G:i(C)
parametrizes couples (L, V), where L is a degree d line bundle on C and V
is an (r + i)-dimensional vector subspace of H 0 (C, L). In symbols

Supp(G:i(C)) = {(L, V): L E Picd(C), V E G(r + i, H 0 (C, L))}.

In other words, as we announced above G:i(C) parametrizes the g:i's (complete
or not) on C.
We now wish to show how the above description of the points of G:i(C)
has an exact analogue for S-valued points of G:i(C), for any analytic spaceS.
To this end, we introduce the cumbersome but useful concept of family of
g:i's on C (parametrized by an analytic spaceS). By this we mean the datum of

(i) A family L of degree d line bundles on C, parametrized by S.

(ii) A locally free, rank(r + i) subsheaf F of cj>*L' where cJ> is the projec-
tion of C x S onto S, with the property that, for each s E S, the
homomorphism

is injective.
§3. The Varieties W~(C) and G~(C) Parametrizing Special Linear Series on a Curve 183

Condition (ii) means that, writing

L ® ¢*:Yi = (lc X f)*(Sf),

where .se is a degree d Poincare line bundle and f is a suitable morphism of

S into Picd(C), we require that F ® :Yt be a vector subbundle of f*(K 0 ).
At least when S is reduced, a family of gd's on C parametrized by Scan be
thought of as a holomorphically varying family

Ls-+ C, S E S,

of degree d line bundles on C, together with a holomorphically varying

family !?fis of gd's

!?fisciLsl, sES.

If we are given a family of gd 's, (fj = (L, F) on C parametrized by S, and a

morphism

f: T-+ S,

we can define the pull-back

f*((fj) = ((lc x f)* L, f*(F)),

which is a family of gd's on C parametrized by T.

Two families (L, F), (L', F') of gd's on C parametrized by S are said to be
equivalent if there exist a line bundle :Yt on S and an isomorphism

L' ~ L ® ¢*i?ll

such that F' is identified with F ® :Yt. We are now going to construct a
universal family of gd's on C parametrized by Gd(C). We recall that G:i(C)
is naturally a subvariety of the Grassman bundle G(r + 1, K 0 ) over Picd(C).
We denote by

the restnctwn of the projection from G(r + 1, K 0 ). Then the universal

family of g:i's on C parametrized by G:i(C) is (c*(.st), :17"), where .se is a
Poincare line bundle of degree d and :17" is the restriction to G:i( C) of the
universal subbundle on G(r + 1, K 0 ) (of course, to say "the" universal
family is somewhat improper, for there is one for each Poincare line bundle,
but we shall ignore this). We are now going to explain in what precise sense
the family we just constructed is universal.
184 IV. The Varieties of Special Linear Series on a Curve

(3.6) Theorem. For any analytic spaceS and any family'§ of g~'s on C para-
metrized by S, there is a unique morphism from S to Ga(C) such that the pull-
back of the universalfamily parametrized by Ga(C) is equivalent to'§.

Proof. Write '§ = (L, F). By the universal property of the Poincare line
bundle, there are a unique morphism

and a line bundle fYi on S such that, denoting by

</J: C X S--+ S

the projection,

(lc x f)*(!l') ~ L (8) <P*!Yi.

As we already noticed, F (8) fYi can be viewed as a vector sub bundle off*(K 0 )
contained in

where

is the functorial complex constructed at the beginning of this section. By the

universal property of Grassmann bundles, F ® fYi is the pull-back of the
universal subbundle via a unique section of

over S. This section corresponds to a unique morphism of analytic spaces

over Pied( C)

which factors through the inclusion

Ga(C) c G(r + 1, K 0 ),
since F ® fYi is annihilated by f*(y). Q.E.D.

A moment of reflection will convince the reader that to speak of families

of g~'s, up to equivalence, is just to speak of relative divisors. Thus a first
consequence of the theorem we just proved (and of the universal property of
the universal divisor) is that G~(C) is nothing but Cd.
§4. The Zariski Tangent Spaces to G:i(C) and W;j(C) 185

§4. The Zariski Tangent Spaces to G:i(C) and Wci(C)

We shall now begin a study of the infinitesimal structure of G:i(C) and Wd(C)
by describing their tangent spaces. Throughout the discussion, C will stand
for a fixed smooth genus g curve.
Let w be a point of G:i(C), corresponding to a degree d line bundle L on
C plus an (r + I)-dimensional subspace W of H 0 (C, L). The Zariski tangent
space to G:i(C) at w is the set ofmorphisms of analytic spaces from Spec(C[s])
to Gd(C) sending the unique point of Spec(C[s]) tow. In symbols

Tw(Gd(C)) = Hom(Spec(C[s]), (G:i(C), w)).

The universal property of Gd(C) expressed by Theorem (3.6) implies that

the tangent space to Gd( C) at w is just the set of isomorphism classes offamilies
of g:i's on C parametrized by Spec(C[s]) which restrict to (L, W) on C. Such a
family will be called afirst-order df!formation qf (L, W).
Our task is then to describe the isomorphism classes of first-order deforma-
tions of (L, W). This could be done directly, but we choose instead an indirect
approach, relying in part on the explicit construction of Gd(C). This was
based on the functorial cohomology complex

constructed in the previous section, and G:i(C) was described as a subvariety

of the Grassmann bundle G(r + I, K 0 ) over Pica(C). As we did there, we
let

be the restriction of the projection of G(r + I, K 0 ) onto Pica( C). We then

have an exact sequence

By the universal property of G:i(C), c- 1(L) parametrizes gd's in ILl, i.e.,

(r + I)-dimensional subspaces of H 0 (C, L). In symbols,

But then

Tw(c- 1(L)) = Tw(G(r + I, H 0 (C, L)))

= Hom(W, H 0 (C, L)/W).
186 IV. The Varieties of Special Linear Series on a Curve

Having described the kernel of c*, we now turn to its image. This requires
an explicit computation. Suppose L is defined by the transition data { V,},
{g,11 }. Let L' be a first-order deformation of L, defined by the transition data

where

A holomorphic sections E H 0 (C, L) is given by a collection {s,} offunctions,

satisfying

An extension of s to a section of L' is given locally by

where s, = g,11 s11 • But this is equivalent to saying, again, that

and that, moreover,

We observe that the left-hand side of the above inequality is a cocycle

representing the cup-product

under the natural pairing

here, of course, 4> stands for the cohomology class of {</>, 11 }, i.e., it is the
Kodaira-Spencer class of the first-order deformation L'. The right-hand
side is the coboundary bs', where s' = {s~} belongs to C0 ({U,}, L). Our
computation establishes the following

Lemma. Given L in Pied( C) and a section s of L, the set of tangent vectors

§4. The Zariski Tangent Spaces to G:i(C) and W:i(C) 187

such that s may be extended to a section of the corresponding first-order

deformation of L is given by

It is important to remark that, when ¢ · s = 0, the extensions of s corre-

spond to the different ways of writing the cocycle representing ¢ · s as a
co boundary.
Returning to the problem of computing the image of c*, notice that the
image of the tangent vector to G~(C) represented by the first-order deforma-
tion (L', W') of (L, W) is represented by the first-order deformation L' of
L. Now, letting L" be a first-order deformation of L with Kodaira~Spencer
class ¢, clearly (L, W) has a first-order deformation (L", W") if and only if
every section of W extends to a section of L", i.e., by the preceding lemma, if
and only if¢ · W = 0. In other terms the image of c* is the set of those ¢ in
H 1 ( C, CD) such that¢ · W = 0. The result of our computations is incorporated
in the following, more general, proposition.

(4.1) Proposition. (i) Every component of G~(C) has dimension at least equal
to the Brill~ N oether number

p =g- (r + 1)(g - d + r ).
(ii) Let w be a point of G~(C), corresponding to a line bundle L and an
(r + i)-dimensional vector subspace W of H 0 (C, L). Then the
tangent space to G~(C) at w fits into an exact sequence

where

Dually

where

is the cup-product homomorphism.

(iii) We have

dim Tw(G~(C)) = p + dim(ker Jlo,w)-

188 IV. The Varieties of Special Linear Series on a Curve

In particular, G~(C) is smooth of dimension pat w if and only if

ker Jlo, w = 0.

Part (ii) has just been proved, except for the dual description of the image
of c*. Here the basic cup-product mapping

encountered in the proof of Clifford's theorem and in the first section of this
chapter appears again. In fact /lo, w is just its restriction to W ® H 0 ( C, KL - 1 ).
From now on /lo, w will often be written Jlo, unless confusion is likely. To
show that

it suffices to notice that, writing < , ) for the Serre duality pairing, ¢ · W = 0
if and only if

(¢,J1 0 (s®r)) = (¢·s,r) =0

for every s in W and every r in H 0 ( C, KL- 1 ).

Part (i) of the proposition follows from Lemma (3.3) when r 2: d - g.
When r < d- g, G~(C) maps onto Picd(C) and the fiber over a point L of
Picd(C) is the Grassmannian of (r + 1)-planes in H 0 (C, L) which, by the
Riemann-Roch theorem, has dimension at least (r + l)(d - g - r). Thus

dim G:i(C) 2: g + (r + l)(d - g - r) = p.

Part (iii) of the proposition follows from part (ii) by a trivial computation.
In fact we have

dim Tw(G:i(C)) = (r + 1)(h 0 (C, L)- r- 1) + g- dim(lmage Jlo.w)

= g + (r + l)(h 0 (C, L)- r- 1)- (r + l)h 0 (C, KL - 1 )
+ dim(ker Jlo, w)
= p + dim(ker Jlo, w).
This ends the proof of Proposition (4.1).

It is now very easy to deduce from (4.1) a description of the tangent spaces
to W~(C).
§4. The Zariski Tangent Spaces to G:i(C) and W:i(C) 189

(4.2) Proposition. (i) Let L be a point of W~(C) not belonging to w;/ 1(C)
(thus r ;;:::: d - g). The tangent space to W~(C) at Lis

where

is the cup-product mapping. Thus W~(C) is smooth of dimension pat

L if and only if flo is injective.
(ii) Let L be a point of W~+ 1 (C). Then

In particular, if W;j(C) has the expected dimension p and r > d - g

(i.e., p < g), Lis a singular point of W~(C).

There is very little, here, that needs proof. Part (i) follows from part (ii)
of (4.1) once we notice that

c: G~(C)-> W~(C)

is biregular off w~+ 1 (C). To prove (ii), first recall that, for suitable integers
n, m, k, W~(C) is locally the pull-back, via a mapping¢ from an open subset
ofPicd(C) to the variety M(n, m) ofn x m matrices, ofthe subvariety Mk(n, m)
of matrices of rank at most k. Moreover, to say that L belongs to W:i+ 1 ( C)
means that c/J(L) belongs to Mk_ 1(n, m). On the other hand, we showed in
Chapter II, Section 2 that the tangent space to Mk(n, m) at a point of
Mk_ 1 (n, m) is the tangent space to all of M(n, m). Thus

TL(W~(C)) = ¢;; 1 (Tq,<LlMk(n, m)))

= ¢;; 1 (Tq,(L)(M(n, m)))
= TL(Picd(C)),

which finishes the proof of (4.2).

To give a more geometrical description of the tangent spaces to Wd(C) -
W:i+ 1 (C), consider the canonical map

Given L in Wd(C) - W:i+ 1 (C), Proposition (4.2) states that the Zariski
tangent space TL(W:i(C)) is defined by the simultaneous vanishing of the
190 IV. The Varieties of Special Linear Series on a Curve

differentials in Image Jl.o. On the other hand, given any non-zero section s in
H 0 (C, L), the image of
s ® H 0 (C, KL - 1) ~ H 0 (C, K)

is equal to H 0 (C, K(- D)), where D = (s), and the linear subspace of lfll 9 - 1
defined by the simultaneous vanishing of the forms in H 0 (C, K(- D)) is, by
definition, ¢K(D). We therefore have the following geometric description of
the projectivized Zariski tangent space to W~( C) at L:

(4.3) lfll'Ji(W~(C)) = n
DelLI
¢K(D) c lfll 9 - 1•

The subvariety W~( C) of Pied( C), as a set, is the image of G~( C) via the
mapping c. It is natural to ask whether this is also an equality of schemes. We
shall now settle the question in the affirmative when G~(C) is smooth and has
the "correct" dimension.

(4.4) Proposition. Suppose G~( C) is smooth of dimension

p = g - (r + l)(g - d + r).
Then the variety W~(C)is Cohen- Macaulay, reduced, and normal. If d < g + r
the singular locus of W~(C) is w~+ 1 (C).

Proof We can limit ourselves to the case d < g + r. The variety W~( C) -
w~+ 1 (C) is isomorphic to G~(C)- c- 1 (W~+ 1 (C)), and hence smooth.
Moreover Lemma (3.5) shows that it has non-empty intersection with every
component of W~(C). In particular, to show that W~(C) is reduced, it suffices
to show that it has no embedded components. This is true since W~(C), being
a determinantal variety of the" correct" dimension p, is Cohen-Macaulay. It
remains to show that it is normal. To this end, notice, first of all, that every
component of c~+ 1 has codimension at least one in C~, which is of pure
dimension p + r, and, secondly, that a general fiber of

is (r +I)-dimensional. Thus w~+ 1 (C) has codimension at least two in

W~(C). Since W~(C) is Cohen-Macaulay, this suffices to show that it is
normal (cf. Proposition (5.4) in Chapter II).

Since G~(C) = Cd is smooth and connected of dimension d = p(g, 0, d),

we find, as a consequence of the result we just proved, the following

(4.5) Corollary. The variety WiC) is reduced, irreducible, normal and Cohen-
Macaulay. If d < g the singular locus of ltd( C) is WJ(C).
§5. First Consequences of the Infinitesimal Study of G:i(C) and W:i(C) 191

To finish the section, we shall reconsider, in the light of the results proved
in later sections, the naive computation of the tangent space to C:i at a point
D not belonging to C:i+ 1, which we gave in the first section (Lemma (1.5)).
First of all, by Proposition (3.4), the Zariski tangent space to C:i at D is the
pull-back, via

of T(!)(vlW:i(C)). Secondly, by the basic Lemma (2.3), u* can be identified

with the coboundary mapping

gotten from the exact sheaf sequence

Thus, by Serre duality, the transpose of u* is the restriction homomorphism

Combining the above remarks, and letting

be the cup-product mapping, we find

Tv(C:i) = u;; 1(T(!)<vlW:i(C)))

= u;; 1 ((1mage Jlo)j_)
= Image(ll(Jlo)j_.

This is precisely the statement of Lemma (1.5).

§5. First Consequences of the Infinitesimal Study of

G:i( C) and Wei( C)

As an application of the study of Zariski tangent spaces to G:i( C) and W:i( C)

that we did in the previous section, we shall present Saint-Donat's proof and
Mumford's subsequent refinement of a nice theorem of Martens which
provides an upper bound for the dimension of W:i(C).

(5.1) Theorem (Martens). Let C be a smooth curve of genus g 2 3. Let d be

an integer such that 2 s d s g - 1 and let r be an integer such that 0 < 2r s d.
192 IV. The Varieties of Special Linear Series on a Curve

Then if Cis not hyperelliptic, every component of W~(C) has dimension at most
equal to d - 2r - 1; in symbols

dim W~( C) ~ d - 2r - 1.

If C is hyperelliptic

dim W~(C) = d - 2r.

Proof. By the remark immediately following the proof of the geometric
version of the Riemann-Roch theorem, any complete g~ on a hyperelliptic
C is of the form

rgi + P1 + · · · + Pd-2r·

This takes care of the hyperelliptic case. Now assume C to be non-hyper-

elliptic and suppose there is ad, 2 ~ d ~ g - 1, such that W~( C) is not empty
and
dim W~( C) 2:: d - 2r.

Choose the minimum d such that this holds. Notice that, by Clifford's theorem,
the number d - 2r must be positive. Let L be a point belonging to a com-
ponent of W~(C) of maximal dimension. We may assume that h0 (C, L) =
r + 1 = l. Set i = h1 (C, L). Since dis minimum we may also assume that
IL I has no base points. By Proposition (4.2) we know that
dim TL(W~(C)) =g- li + dim(Ker f.lo).
Therefore, by our assumption, we get

dim(Ker f.lo) 2:: li - l - i + 1.

Let s 1 , ... , s1 be a basis for H 0 (C, L). Since ILl has no base points we may
assume that s 1 , s2 have no common zeros. Set

h = 1, ... ' l.
Obviously

dim(Ker J.lo, wh) - dim(Ker f.lo, wh_,) ~ i,

and therefore the last two inequalities and the base-point-free pencil trick
give us
h0 (C, KL - 2 ) = dim(Ker J.lo,w 2) 2:: i - l + 1= g- d,

so that
§5. First Consequences of the Infinitesimal Study of G~(C) and W~(C) 193

If d ~ g - 2, then 4 ~ 2d ~ 2g - 4, so, by Clifford's theorem, C is hyper-

elliptic, a contradiction. Suppose now that d = g - 1. Since W:i(C) has
positive dimension, we may assume that Lis not a theta characteristic, i.e.,
that L 2 ;t. K. This contradicts the last inequality. Q.E.D.

Let us now turn to Mumford's refinement of Martens' theorem. We shall

say that a smooth curve C is hi-elliptic if it can be represented as a ramified
double covering of an elliptic curve.

(5.2) Theorem (Mumford). Let C be a smooth non-hyperelliptic curve of

genus g ; : : : 4. Suppose that there exist integers rand d such that 2 ~ d ~ g - 2,
d ;;::::: 2r > 0 and a component X of W:i( C) with

dim X = d - 2r - 1.

Then C is either trigonal, hi-elliptic or a smooth plane quintic.

Proof (Mumford). In what follows, to simplify the notation we shall always

write the symbol W:i(C) to mean "a component of W:i(C)." Assume then
that
dim W:i( C) = d - 2r - 1.

Notice that by imposing r - 1 general base points to the series belonging

to W:i( C) we get
dim WJ_r+ 1 (C) ; : : : (d- r + 1)- 3.
On the other hand, since C is not hyperelliptic, equality holds by Martens'
theorem. We may therefore limit ourselves to the case r = 1. Let d be the
minimum integer such that

dim WJ(C) = d- 3.

Arguing as in the proof of Martens' theorem we conclude that

and therefore

(5.3) dim Wti 1 (C) ; : : : d - 3.

We want to apply Martens' theorem to this situation. There are three possible
cases:
(a) 2d ~ g - 1. Then
dim W~d' 1 (C) ~ 2d - 2(d - 1) - 1,
and hence, by (5.3),
d ~ 4.
194 IV. The Varieties of Special Linear Series on a Curve

(b) g - 1 < 2d < 2g - 4. Then, passing to the residual series, (5.3) gives

Again by Martens' theorem, we get

d ~ 4.

(c) 2d = 2g - 4. Then, passing to the residual series, (5.3) gives

dim W~(C);;::.: d- 3,

and therefore

d ~ 5, g= d + 2.
In conclusion, in every cased ~ 5, and if d = 5, g = 7. If d = 3, Cis trigonal.
Assume now that Cis not trigonal and d = 4: in particular, then, g ;;::.: 6. Our
hypothesis that the dimension of WJ(C) is equal to d - 3 gives

dim W,i(C) = 1.

Let Land L' be two distinct points of W,i(C). We notice that

(5.4) h 0 (C, L ® L') = 4.

In fact, by Clifford's theorem, h0 ( C, L ® L') is less than or equal to 4. On

the other hand, since L has no base points, by the base-point-free pencil
trick the kernel of the multiplication map

is isomorphic to H 0 ( C, L - l ® L'). Since L and L' are distinct points of

W,i(C) this implies that vis injective, proving (5.4). By duality

If p 1 , •.. , P9 - 6 are general points on C,

therefore IL' I is contained in

Bibliographical Notes 195

On the other hand, dim IM I = 2. Let

¢: c----> [p>2

be the morphism defined by I M 1. The morphism

defined by IL' I is obtained by composing ¢ with a projection from a point

of ¢(C). Since W!(C) has positive dimension, we may assume that the center
of projection is a smooth point of ¢(C). Therefore

4 = d = (deg ¢) · (deg ¢(C) - 1).

Two cases are possible. Either deg ¢ = 1, in which case ¢(C) is a plane
quintic, which has to be smooth since g ~ 6, or else C is a double covering
of a plane cubic, which has to be smooth since C is not hyperelliptic.
There remains to examine the cased= 5, g = 7. We wish to show that
this cannot occur. If it did, on C there would be a two-dimensional family of
degree five line bundles L such that h0 ( C, L) ~ 2 and IKL- 2 1#- 0: in
other terms

(5.5) KL- 2 ~ (!}(p + q)

for some points p, q E C. Moreover, by our choice of d, for a general L the
linear series IL I would have no base points. But now, for any divisor DE IL I
not containing p or q, we would have:

By the Riemann-Roch theorem, then, IL(p + q)l would be a g~ and

IKL- 1( - p
- q) I a g~, which cannot exist on a non-hyperelliptic curve of
genus 7.

Bibliographical Notes
The Brill-Noether number and the Brill-Noether matrix first appear in
Brill-Noether [1]. A standard reference for Poincare line bundles, both in
the case at hand and in the one of a family of curves is Grothendieck's lecture
[1]. The basic theorems about base change in cohomology in an analytic
setting are to be found in Grauert [1]; for a lucid account of the algebraic
case we strongly recommend Mumford's book [1, Section 5].
Our definition of the loci W~( C) as the determinantal varieties associated
to a functorial cohomology complex for the Poincare line bundle is as in
196 IV. The Varieties of Special Linear Series on a Curve

Kleiman-Laksov [1], or Kempf [1]. The variety Gd(C) is introduced

in Arbarello-Cornalba [1].
For the basic properties of Fitting ideals we refer to Northcott [1], as
well as to Fitting's original paper [1].
The first part of (4.2), in a more general context, can be found in Griffiths
[1]. Corollary (4.5) is in fact part of Kempf's singularity theorem, to be
proved in Chapter VI. The original proof is in Kempf [2].
Martens' theorem first appeared in Martens [2]. Our proof is adapted
from Saint-Donat [2]. Mumford's refinement of Martens' theorem is
taken from Mumford's paper on Prym varieties [2].

Exercises

A. Elementary Exercises on f.lo

A-1. Let C be a curve of genus g, L a line bundle of degree g - 1 on C such that L 1 1

is a base-point-free pencil. Show that the map

will fail to be injective if and only if L 2 ~ K. (More generally, suppose that IL I

is a pencil with base divisor D. Show that flo will fail to be injective if and only if
h0 (C, KL - 2 (D)) > 0.)

A-2. In general, suppose that Lis any theta characteristic; that is, a line bundle such
that L 2 = K. In terms of the corresponding identification H 0 (C, L) = H 0 (C, KL - 1 )
we may then write

Show that f.1. 0 (A 2 H 0 (C, L)) = 0, and conclude that if r = r(L)

dimTL(W~_ 1 (C))2g- (r +
2
1) .
A-3. Suppose now that C is a non-hyperelliptic curve of genus 4. The canonical model
of Cis then the intersection of a quadric Q and a cubic Fin 1Jl> 3 . Show directly:

(i) If Q is smooth, W~(C) consists of two reduced points.

(ii) If Q is singular, W~(C) consists of one non-reduced point, whose tangent
space is one-dimensional. Identify this tangent space.

A-4. With C as in A-3 above, let D E q be any point. Identify explicitly the tangent
spaces to q at D in case (i) Q is smooth, and (ii) Q is singular.
Exercises 197

B. An Interesting Identification

Let DE Cd. In this chapter we encountered the identification

0( ( ) _ {first-order deformations of}

H C, C2v D) 7 . . ·
D as a diVIsor on C

Recall that the right-hand side are relative divisors on C x S (S = Spec IC[e]/(~: 2 )) whose
restriction to C x Spec IC is D.

B-1. Make the mapping Cl. explicit. For example, if D = dp where p E C and if z is a
local coordinate on C centered at p, then v E H 0 (C, C2v(D)) is

What is the corresponding divisor on C x S?

B-2. We also encountered the identification

{first-order deformations of D as a divisor on C} jt Tv(Ca).

Combining this with ( *) gives

(this is tjJ - I in (2.3)). Using B-1 you may make f3 o Cl. explicit. For example, what is

(/3 o e~.)- 1 (tangent space to the diagonal)?

B-3. Use these explicit identifications to verify the commutativity of the diagram in
Lemma (2.3).

C. Tangent Spaces to W1 (C)

In the following three exercises C will be a non-hyperelliptic curve.

C-1. Show that for p E C a general point there do not exist an integer m and a point
q #- p such that mq ~ mp.
(Hint: Use a tangent space computation to show that

v E J(C),

unless m =nand v = 0. (Here mW1 (C) = {mu(p), p E C}.))

198 IV. The Varieties of Special Linear Series on a Curve

C-2. Let A c C2 be an algebraic curve giving a one-dimensional family {D ... } of divisors

of degree 2 on C, and assume that

u(A) = - W1 (C) + v.

Show that C has a gL denoted by ID I, and A = { D - p} P e c consists of all divisors

of degree 2 subordinate to the g~ (again use tangent spaces).

C-3. Retaining the assumptions of the preceding exercise, assume that

u(A) = W1 (C) + v,

and show that A = {p + q}qec for some fixed p E C.

D. Mumford's Theorem for g;'s

The proof of Mumford's theorem proceeds by first reducing to the case r = 1. As

we will see in the following exercises, this means that it is not necessarily sharp in case
r > 1.

D-1. Show that if Cis hi-elliptic and d:::; g- 2, then dim w;(c) = d- 5.

D-2. Show that if Cis trigonal and d:::; g- 2, then dim w;(C) = d- 6.

D-3. Show that if C is a two-sheeted cover of a curve of genus 2 and d :::; g - 2, then
dim w;cc) = d- 6.

D-4. Use Mumford's theorem and Exercise D-2 above to conclude that for any C, if
d:::; g- 2 and dim w;(c) = d- 5, then Cis hi-elliptic.

D-5. More generally, show that if C is hi-elliptic (resp., trigonal; a two-sheeted cover
of a curve of genus 2) and d:::; g- 2, then dim W~(C) = d- 2r- 1 (resp.,
d- 3r; d- 2r- 2). Use this to give a sharpened form of Mumford's theorem
when r;:::: 2.

D-6. Can you prove this result directly, by giving an argument that uses g~'s rather
that first reducing to gJ's? (We don't know how this might go.)

E. Martens-Mumford Theorem for Birational Morphisms

E-1. Let C be any curve of genus g, L ={!)(D) a special line bundle on C of degree d
with h0 (C, L) = r + 1. Show that

d;:::: g- d + 2r + h0 (Kc( -2D)).

Exercises 199

E-2. Now assume that (his birational. Use Exercise B-6 of Chapter III and the estimate
in the preceding excercise to conclude that

rank (J.l 0 : H 0 (C, L) ® H 0 (C, Kc ® L*) ..... H 0 (C, Kc))

;;::: 2g- 2d + 3r- 1- h0 (C, Kc(-2D))

and hence prove the

Theorem. IfiLl is a g:i on C with rfJL birational, then ina neighborhood ofu(L) E J(C),

dim W~(C) s h 0 (C, (!)(2D))- 3r.

Conclude in particular that if d < g - 1 or d = g - 1 and L 2 =F Kc, then

dim W~(C) s d - 3r.

F. Linear Series on Some Complete Intersections

In these exercises we will examine the varieties W~(C) (written W~, for brevity) for two
special types of curves, both complete intersections C = S n Tin IP' 3 . In each case we
will use the geometry of the surfaces containing C, the fact that Kc ~ (!Jc(2) (because
in both examples deg S + deg T = 6), and the following superabundance lemma:

F-1. Let r c IP' 3 be a collection of d s 7 points. Assuming that r fails to impose inde-
pendent conditions on quadrics, show that either:
(i) r contains four collinear points;
(ii) r contains six points on a conic; or
(iii) r contains seven coplanar points.

F -2. Suppose that C = S n T is the smooth complete intersection of a smooth quadric

Sand a smooth quartic T (thus Cis a curve of type (4, 4) on S). Using the preceding
exercise show that set-theoretically:
(i) w~ = 0;
(ii) Wi consists of the two points corresponding to the two pencils cut out by
the rulings of S (in particular, these gi's are distinct);
(iii) W~ = Wi + W1 (i.e., every g~ on C has a base point), and that W~ consists
of two disjoint copies of C;
(iv) W~ = {(!Jc(1)} (i.e., the hyperplane series is the unique g~);
(v) W~ = (Wi + W2 ) u (W~ - W2 ). Conclude that W~ consists of three
copies of C 2 , with the two components of Wi + W2 being disjoint, and each
component meeting W~ - W2 in a curve;
(vi) W~ = W~ - W1 (thus, W~ is set-theoretically isomorphic to C);
(vii) W~ = (Wi + W3 ) u (W~ - W2 + W1 ); and
(viii) finally, show that W~ is not contained in

(Hint: dim w:_ (C);;::: g- 4 for any curve C.)

1
200 IV. The Varieties of Special Linear Series on a Curve

F-3. In parts (ii), (iii), and (iv) of the preceding exercise, show that all schemes are
reduced, and thus all set-theoretic equalities are in fact scheme-theoretic
equalities.

F -4. Suppose that C = S n Tis the smooth intersection of two smooth cubic surfaces.
We consider C as the base locus of the pencil {S ;.} ._If' of cubic surfaces spanned by
Sand T. Using the analysis of cubic surfaces show that:

(i) w~ = 0;
(ii) dim W~ = 1 (even better, show that set-theoretically W~ ~ {(L, A.): L c S;.}
c G(2, 4) X IFD 1 ;
(iii) w~ = 0 and w~ is scheme-theoretically one point (i.e., the hyperplane
series is the unique g~ on C);
(iv) w~ = 0 and scheme-theoretically w~ = w~ - WI;
(v) w~ = (W~ + WI) u (W~ - Wz). Show that w~ thus reduces to a union of
two surfaces meeting along a curve ; and
(vi) finally, show that W~ is not contained in

In both these examples, the g:_ 1 's are the only special linear series not arising
from linear subspaces of IFD 3 .

G. Keem's Theorems

In this batch of exercises we will look at one of several extensions of Mumford's theorem
due to Keem.

(*)Theorem. Let C be a smooth algebraic curve of genus g 2 11, and suppose that for
some integers d and r satisfying d :$ g + r - 4, r 2 1, we have

dim W~(C) 2 d - 2r - 2.

Then C possesses a g!.

For further results along these lines, see Keem [1].

G-1. Suppose C satisfies the conditions of Theorem (*) above, but does not possess a
g!. Show that for some e such that 5 :o;; e :o;; g- 3, we have

dim W!(C) 2 e - 4,

and a general element of a component of maximal dimension in W!(C) has no

base points.
Exercises 201

G-2. Show that with e as in the preceding exercise, we must have

and, applying Mumford's theorem, deduce that we must have either

e = 5 or 6,
or

e= 8 and g = 11.

In the following exercises, we consider the cases e = 5, e = 6, and e =8

allowed above.

G-3. Suppose that we have e = 5, and let L 0 and L be general members of W~(C).
Show that the map 4>~.~ associated to the complete series 1L~ 1 is a birational
embedding of C in IP 3 .

G-4. Use the base-point-free pencil trick in the situation of the preceding exercise to
conclude that

so that

and

Now apply Exercise L-1 of Chapter III to <f>KLo-z to obtain a contradiction.

G-5. Suppose now that in the situation of Exercise G-1 we have e = 6. Show that for
L 0 , L general points of W~(C), we have

and hence, by the base-point-free pencil trick, that

G-6. Using the preceding exercise and Exercise L-1 of Chapter III, show that the map
4>~.{ C --> IP 4 is not birational, and derive a contradiction.

G-7. Finally, suppose that e = 8 and g = 11. Show that for general L E W~(C) we have

L 2 ~K(-p-q-r-s)

for some p, q, r, s E C.
202 IV. The Varieties of Special Linear Series on a Curve

G-8. Let J(C) ~ J(C) be the map given by multiplication by m. Show that m- 1 (Jif.'k(C))
is still irreducible. Conclude that in the circumstances of the preceding exercise,
we must have

for every line bundle L such that L 2 ~ K(- p - q - r - s).

G-9. Now let M be any line bundle on C with M 2 ~ K. Show that h 0 (C, M) :o;; 3,
and conclude that for general p, q E C, h0 (C, M(- p - q)) :o;; 1 to obtain a contra-
diction with the preceding exercise.
Chapter V

The Basic Results of the

Brill-N oether Theory

As we recalled in Chapter I, a genus g curve depends on 3g - 3 parameters,

describing the so-called moduli of the curve. Our goal in this chapter is to
describe how the projective realizations of a curve vary with its moduli,
and what it means, from this point of view, to say that a curve is "general"
or "special." Accordingly, we would like to know, first of all, what linear
series can we expect to find on a general curve and, secondly, what the sub-
varieties of the moduli space corresponding to curves possessing linear
series of specified type look like. In the previous chapter we introduced the
varieties

Wd(C) = {L E Pica(C): h0{C, L) ~ r + 1}

G:i(C) = {g:i's on C},

referring to them as "configurations of linear series on C," and it is on these

that we will focus our attention. A natural question is, how can we tell one
curve from another by looking at these configurations, or more precisely,
what do these look like in general, and how-and where-can they
degenerate? Clearly, from the point of view of non-special linear series,
all curves look alike. It is at the level of special linear series that differences
appear: this is what was classically called the problem of special linear
series. In this chapter we shall deal only with one aspect of the problem;
namely, we shall state and discuss a few fundamental results describing what
happens on a general curve (and, in some instances, on every curve). All of
these results will be proved in Chapters VII and XI of this book.
In describing the configurations of linear series of given dimension r, the
essential step consists in studying the complete ones; in symbols, this amounts
to dealing with
204 V. The Basic Results of the Brill-Noether Theory

As we already pointed out, the really interesting case is the one of special
linear series or, more precisely, the case of exceptional linear series; these are
the complete, degree d, series ID I for which

r = r(D) > 0, g- d +r= i(D) > 0.

The cases in which these inequalities are not satisfied are very easily dealt
with. Consider, in fact, the abelian sum mapping

If r = 0, by Abel's theorem, u is an isomorphism between an open set of

Ca and Wa(C)- WJ(C). If, on the other hand, r > 0 and g - d + r = 0,
then, by the Riemann-Roch theorem, u maps Ca onto W~(C) and, again by
Abel's theorem, the general fiber of u is a (d - g)-dimensional projective
space, showing that, in this case, W~( C) and Pica( C) coincide.
The study of complete (exceptional) special linear series can be slightly
simplified by means of residuation. In fact, passing to the residual series
provides an isomorphism

given by

Thus we can restrict our attention to complete linear series of degree less
than g. A final consideration comes from Clifford's theorem. According to
this, any linear series ID I of degree d less than 2g - 1 satisfies the inequality

d 2 2r(D),

and equality holds only in a few cases: when D = 0, D = K or else when the
curve C is hyperelliptic and ID I is a multiple of the hyperelliptic involution
on C. Therefore, when 0 < d < 2g - 2, W~' 2 ( C) either consists of a single
point, when C is hyperelliptic, or else is empty.
In the light of these considerations we emphasize that the results that we
shall discuss and illustrate in this chapter, although stated in general, really
acquire a non-trivial meaning when thought of as theorems about complete
linear series g~ such that

r > 0, g > d > 2r.

In particular, these limitations already take care of the cases for which
g ~ 3.
V. The Basic Results of the Brill-Noether Theory 205

For a fixed value of the genus it is useful to represent on a diagram the

values of d and r that we are interested in:

The union of the two shaded parts, with the exclusion of the lines r = 0 and
r = d - g, constitutes the region of "exceptional special divisors." The two
shaded parts are divided by the vertical line d = g about which residuation
produces a diffracted reflection, and the region we are interested in is the
open triangle: r > 0, g > d > 2r.
Let us now bring into the picture the Brill-Noether number

p = g - (r + l)(g - d + r).
In the diagram below, the shaded area is that part of the region of exceptional
special divisors for which p ;;:::: 0.

d
206 V. The Basic Results of the Brill-Noether Theory

As we explained in Chapter IV, Section 3, the varieties W~(C) and G~(C),

if non-empty, have dimension at least equal to p. The first natural question
is whether the condition p ~ 0 suffices to conclude that W~(C) and G~(C)
are non-empty. This is answered by the following existence theorem, in-
dependently proved by Kempf and Kleiman-Laksov.

(1.1) Existence Theorem. Let C be a smooth curve of genus g. Let d, r, be

integers such that d ~ 1, r ~ 0; then if

p = g - (r + 1)(g - d + r) ~ 0,

G~(C), and hence W~(C), are non-empty. Furthermore, every component of

G~(C) has dimension at least equal top; the same is true for W~(C) provided
r ~ d- g.

We now wish to offer, as an illustration, a direct analysis of the varieties

of special divisors for curves of genus g ::::;; 6. We shall not use the Existence
Theorem, but will rely on ad hoc arguments. As we already explained, we
shall limit ourselves to the case of complete linear series g~ for which

r > 0, g > d > 2r.

The first case is g = 4. Here the only values of d and r to be considered

are d = 3 and r = 1, and for these the Brill-Noether number p is equal to
zero. Two possibilities are to be examined. First assume that the curve C is
hyperelliptic. Then, as we observed in Chapter I, Section 2, any g~ on C is
obtained by adding an arbitrary point to the g1. Thus, in this case, W~(C)
(or at least its support) is isomorphic to C. Suppose now that Cis not hyper-
elliptic. Then its canonical model, r, is a smooth complete intersection of
a quadric and a cubic in IFD 3 . Consider a g~ on C. By the geometric version
of the Riemann-Roch theorem, the linear span in IFD 3 of a divisor D belonging
to the g~ is a line, which is thus a trisecant to r. By Bezout's theorem this
line lies in the quadric Q containing r. This shows that a g~ on C is cut
out by the lines of a ruling of Q. Two cases can then occur. Either Q is smooth,
hence doubly ruled, and then W~(C) consists of two distinct points, or
else Q is a cone and then W~(C) consists of a single point.

The next case is g = 5. Here the values of d and r to be considered are

d = 3, r= 1 (p < 0),
d = 4, r= 1 (p = 1).

When d = 3 and r = 1 the Brill-Noether number pis negative. In this case

the existence theorem doesn't give any information, but we still try and
V. The Basic Results of the Brill-Noether Theory 207

describe W~(C), when non-empty. First of all, if Cis hyperelliptic, then, as

before, any g~ is obtained by adding an arbitrary point to the g~, and W~(C)
is thus isomorphic to the curve itself. If C is non-hyperelliptic, but possesses
a g~, then the residual of this g~ is a g; which, by Clifford's theorem, cannot
have base points. The curve C can then be realized as a plane quintic r,
and the genus formula tells us that r has, as its sole singularity, a node or an
ordinary cusp, the g~ being cut out on C by the pencil of lines through
this double point. Conversely, any plane quintic r having as its only singular-
ity a node or a cusp is a genus 5 curve possessing a g~. We also claim that,
in this case, W~(C) consists of a single point. In fact a g~ on C must be cut
out, on r, by adjoint curves of degree 2 (i.e., conics through the singular point)
passing through five residual points. But six points belong to the base locus
of a pencil of conics only if five of them are on a line I, the pencil of conics
being the sum of the pencil of lines through the remaining point plus I.
On the other hand, the only way for a pencil of lines to cut a g~ on r is to be
centered at the singular point of r, proving our claim.
Let us now examine the cased = 4, r = 1. Here the Brill-Noether number
p is equal to 1. Let us first assume that C is hyperelliptic. In this case any
complete gl is obtained from the g~ by adding two base points, so that Wi(C)
is isomorphic to C 2 •
Suppose now that Cis non-hyperelliptic. We shall prove that, in this case,
Wi(C) is one-dimensional. By Martens' theorem (Theorem (5.20) in Chapter
IV) we already know that it is at most one-dimensional. Consider then a
canonical model r of C. As we have seen, r lies on three independent quadrics
in IP> 4 . Let us consider a complete gl,IDI, on C. Its residualiK( -D) I is also
a gl, and we denote by F and F' the fixed divisors of ID I and IK(- D) I,
respectively. We will distinguish two cases according to whether ID Icoincides
with its residual or not. Suppose first that ID I # IK(- D) I; choose hyper-
planes Hii in IP> 4 such that

Hii · r = D; + F + F' + Dj, 1 s i,j s 2,

where D; +FE IDI, i = 1, 2, D~; + F' E IK( -D)I,j = 1, 2, and whereD 1 and
D 2 (resp. D't. D].) have disjoint supports. If Lii = 0 is the defining equation
of H;j, then

defines a non-zero meromorphic function on C which is, in fact, holomorphic.

This means that, for some non-zero constant k, the canonical curve r is
contained in the singular quadric
208 V. The Basic Results of the Brill-Noether Theory

This quadric is ruled by two systems of two-planes. One of the rulings cuts
out the series ID I, the other one cuts out the series IK(- D) 1. In the case
where ID I = IK(- D) I we just consider hyperplanes Hii such that

where D 1 and D 2 have disjoint supports and Di +FE IDI, i = 1, 2. Using

the same notation as before we see that the canonical curve r is contained in
the rank 3 quadric

This quadric is simply ruled by a system of two-planes cutting out, on r,

the given gJ.. Also observe that in both cases the gi, being complete, uniquely
determines the quadric. In fact, by the geometric version of the Riemann-
Roch theorem, any divisor D belonging to the gJ. spans a IP 2 so that, as D
varies in the gi, its linear span sweeps out the quadric. If C is not trigonal
the above construction can be reversed. Namely, the two rulings of a
quadric of rank 4 containing r cut out, on r, two mutually residual gJ.'s;
while the single ruling of a rank 3 quadric containing r cuts out an auto-
residual gJ.. It is also clear that autoresidual gJ.'s are in finite number, any
two of them differing by a two-torsion point in the Jacobian variety of C.
To describe W,l(C) in the non-trigonal case we now proceed as follows.
Consider three linearly independent quadrics containing the canonical curve
r, and denote by Q0 , Q1 , Q2 the symmetric matrices defining them. The
locus of singular quadrics containing Cis then described, in the homogeneous
coordinates Jc 0 , Jc 1 , A2 , by the equation

This equation cannot be identically satisfied; otherwise, there would be a

two-plane of singular quadrics containing rand hence at least oo 2 g,l's on C.
But this possibility, as we already pointed out, is ruled out by Martens'
Theorem. It follows that the locus of singular quadrics containing r is a
quintic curve in the plane with homogeneous coordinates Jc 0 , Jc 1 , Jc 2 • We can
then conclude that W,l(C) is a two~sheeted ramified cover of a plane quintic,
the involution of this covering being given by residuation. An interesting
situation occurs when C is trigonal and non-hyperelliptic. As we already
observed when studying W~(C), the residual of the g~ on C is then a
g; which exhibits C as a plane quintic C having, as unique singularity,
a node or a cusp. It is a useful exercise to check that, in this case, W,l(C)
consists of two copies of the curve C meeting in two points, or doubly
at a point if C has a cusp, the two components being interchanged by
residuation. One copy of C is the locus of gJ.'s of type g~ + p, p E C. The
other is the locus of gJ.'s which are cut by lines through a fixed point of C.
V. The Basic Results of the Brill-Noether Theory 209

Let us now examine the last case: g = 6. Here the values of d and r to be
considered are
d= 3, r = 1 (p < 0),
d = 4, r= 1 (p = 0),
d = 5, r= 1 (p = 2),
d = 5, r=2 (p < 0).

The cased = 3, r = 1 is dealt with exactly as before. Namely, if Cis hyper-

elliptic, then any gj is obtained from the gi by adding an arbitrary base
point so that Wi(C) is isomorphic to the curve itself. If C is non-hyper-
elliptic but, nevertheless, contains a gi, then WMC) reduces to a single
point. Suppose in fact that C possesses two distinct gi's. By Clifford's
theorem the sum of the two gfs is a complete g~. This g~ defines a morphism
¢ of C into IP> 2 , which cannot be a triple covering of a conic, otherwise the
two g~'s would coincide. On the other hand ¢ cannot be a two-sheeted
covering of an elliptic cubic (the g~ would not contain a gD, nor can it be
birational, otherwise the image curve would be a sextic with two triple
points, in contrast with the genus formula. This is absurd and our claim is
proved.
Let us now consider the other case in which the Brill-Noether number p
is negative, namely, when d = 5, r = 2. Again, when p is negative, the
existence theorem doesn't say anything, but, as before, we shall describe what
W~(C) looks like, when non-empty. First of all, when C is hyperelliptic,
any g~ on C can be obtained from the double of the gi by adding an arbitrary
base point. This shows that, in this case, W~( C) is isomorphic to C itself.
Suppose now that C is not hyperelliptic. A g~ on C must then define a hi-
rational morphism of C onto a plane quintic, which, by the genus formula,
must be smooth. We claim that the g~ is unique. Suppose in fact that another
g~ exists. Then it must be cut out, on the plane quintic, by a two-dimensional
linear system of conics passing through five points. The only way this can
happen is for this system of conics to be the union of the system of lines in
IP> 2 and a fixed line. This exactly means that the given g~ is the one cut out
by the lines in [JJ> 2 • In conclusion, for a smooth plane quintic, W~( C) consists of
exactly one point.
We now come to the most interesting case; the one for which d = 4,
r = 1. Here the Brill-Noether number is equal to zero. Let us dispose of
two special cases. First of all, if Cis hyperelliptic then any complete gl on C
is obtained by adding two base points to the gi, so that W!(C) is birationally
equivalent to the second symmetric product of the curve. If C is isomorphic
to a smooth plane quintic then, as in the case of g~'s, one shows that any gl
is obtained from the g~ by subtraction of a point, so that W!(C) is isomorphic
to the plane quintic itself. Suppose now that C is neither hyperelliptic nor
isomorphic to a plane quintic. The existence theorem assures us that there
210 V. The Basic Results of the Brill-Noether Theory

are g!'s on C: a direct argument is as follows. We can obviously assume that

C is not trigonal. Let p. q, r be any three points of C. Since C is not trigonal,
H 1 (C, K(- p - q - r)) maps isomorphically to H 1 (C, K(- p)). Thus the
evaluation map

is onto, and IK( -p)l embeds C in IP> 4. On the other hand

so there is a pencil of quadrics in IP> 4 containing C. Among these there is

at least one (five, in general) quadric of rank at most four. One ruling of
this quadric cuts a g! on C.
We shall now give an idea of what the various possibilities for W!(C)
are. The residual ofag! is ag~ which defines a morphism¢ onto a plane curve
r. Excluding the cases where Cis hyperelliptic or else isomorphic to a smooth
plane quintic, we see that only few cases can occur: r is a conic and ¢ a
three-sheeted covering; r is a smooth cubic and ¢a two-sheeted covering;
r is a plane sextic and has a triple point; r is a sextic having only double
points as singularities. In the first three cases W!(C) contains a one-di-
mensional component. In fact, in the first and third case, C is trigonal and
therefore infinitely many g!'s can be obtained from the gj by adding an
arbitrary base point. In the second case Cis hi-elliptic, i.e., a ramified double
covering of an elliptic curve, and infinitely many g!'s on C can be obtained
by varying the g~ on the elliptic curve rand composing this with ¢. As far
as the fourth case is concerned, we shall limit ourselves to the general case
of a plane sextic with four distinct double points p 1 , p 2 , p 3 , p 4, no three of
them collinear. To see that this does indeed occur, given four points p 1 , •.. , p4
in the plane, no three of them collinear, set

F = Ql + Qz + L1 + Lz,
G = Q3 + Q4 + L3 + L4,
whereQ 1 , ... , Q4 aredistinctsmooth conics through pi> ... , p4 andL 1 , •.• , L 4
are distinct lines with no points in common, and not passing through any
one of the points Pt> ... , p4 . By Bertini's theorem, a general member of the
pencil spanned by F and G is a sextic r with the required properties. We shall
prove that, in this case, W!(C) consists of exactly five points. These cor-
respond to the four g!'s cut out on r by the pencils of lines though p 1 , ... , p4
and to the one cut out by the conics through p 1 , ... , p 4 . To see that indeed
these are the only g!'s on C,just observe that a g! must be cut out on r by a
one-dimensional system of cubics through p 1 , ... , p4 and through six
residual points p 5 , ••• , p 10 . By Bezout's theorem this is possible only if this
system possesses a base curve, which must then be a line or a conic. From this
V. The Basic Results of the Brill-Noether Theory 211

it is immediate to deduce that the given g! must coincide with one of the five
we previously described.

The final case is d = 5, r = 1. Obviously, so far as existence is concerned,

we have nothing left to prove. If Cis hyperelliptic, any complete g~ is obtained
from the gi by adding three base points, so that W~(C) is birationally
equivalent to the third symmetric product of C. If C is not hyperelliptic,
then Martens' theorem says that the dimension of W~(C) does not exceed
two. Since p = 2, W~(C) is two-dimensional. What W~(C) actually looks
like depends on the particular nature of C, as may be seen from the following
two examples. Suppose first that C is a smooth plane quintic. Then, as the
reader may easily check, any complete g~ on Cis of the form I(f.Jc(l)(- p + q) I,
where p and q are distinct points. Thus W~(C) - w;(c) is isomorphic to
the product of C with itself minus the diagonal. On the other hand, if C is
trigonal, W~(C) has two components. One component is made up of the
series g~ + p + q, where p and q are points of C; the other component is
made up of the residuals of the series in the first component.
From the preceding discussion it appears that, whenever p ;;:::: 0, and with
the exception of cases that can be geometrically described, and directly
dealt with, the variety W~(C) has dimension equal top. Moreover, whenever
p = 0, we were able to count the points of W~(C). The results of these direct
computations are particular cases of a formula, discovered by Castelnuovo,
stating that, when p = 0, the number of g~'s is
.,
(1.2) 'Tl
g · o (g -
i= d
l.
+ r + i)! ·
Both Kempf and Kleiman-Laksov generalized this formula and obtained,
along with the existence theorem, an explicit expression for the class of
W:i(C) in the cohomology ring of the Jacobian. As we shall see in Chapter
VII, their proofs are based on the determinantal nature of W:i(C) and on
Porteous' formula.
212 V. The Basic Results of the Brill-Noether Theory

(1.3) Theorem. Let C be a smooth curve of genus g. Let be the class of the e
theta divisor in the cohomology ring of the Jacobian J( C). Let w:i be the class of
W:i(C). Then, if W:i(C) is of the expected dimension p,

W~= ( Ilr '!

l.
)
.()9-p.l
;= o (g - d + r + i) !
Recalling that the theta divisor provides a principal polarization of the
Jacobian, and hence its g-fold intersection number is g!, we get, as a special
case, Castelnuovo's count for the number of g:i's when p = 0.
It was remarked by Fulton and Lazarsfeld that the existence theorem is,
in a sense, a reflection of an ampleness property of the explicit complex we
used to define W:i( C) itself. From this remark they went on to prove the
following fundamental Lefschetz (or Bertini) type of result.

(1.4) Connectedness Theorem. Let C be a smooth curve of genus g. Let d, r

be integers such that d ;:::>: 1, r ;:::>: 0. Assume that

p = g - (r + 1)(g - d + r) ;:::>: 1.

Then G:i(C), and hence w;(C), are connected.

The reader is invited to go back to the examples we treated and directly

show that, whenever p ;:::>: 1, the variety W:i(C) is connected.
The existence theorem gives a lower bound for the dimension of G:i(C).
We may now ask whether this bound is sharp and, as a special case, whether
W:i( C) is empty for negative p. In this generality the answer to this question is
obviously no. Hyperelliptic curves in genus g ;:::>: 3, trigonal curves in genus
g ;:::>: 5, and smooth plane quintics in case g = 6, give examples for which
p < 0 but W:i(C) is non-empty. To get some feeling for which kind of
geometrical implications the condition p < 0 gives, we shall offer two
examples. The first is the one of gJ's. For these the Brill-Noether number is
given by

p = 2d- g- 2.

A base-point-free gJ on C defines ad-sheeted ramified covering of IP 1 which,

by Hurwitz' formula, has 2d + 2g - 2 ramification points. Since, up to
projectivities, we can always assume that the d-sheeted covering in question
is ramified over the points 0, 1, oo, we conclude that a curve of genus g which
1 It is useful to observe that the enumerative formula for the class of W~(C) also holds in

the Chow ring of Picd(C) with minor formal modifications. Namely, one has to substitute
W.- 1 (C) for the theta divisor, and also one has to be careful in choosing the right identification
between Picd(C), where W~(C) lives, and Pic•- 1 (C), which is the natural ambient space for
W.- 1(C).
V. The Basic Results of the Brill-Noether Theory 213

can be expressed as a d-sheeted covering of IP 1 depends on no more that

2d + 2g - 5 parameters. Now the condition p < 0 can be rewritten in this
way:

2d + 2g - 5 < 3g - 3.

We can therefore assert that, if Cis a general curve of genus g, and if p < 0,
then WJ(C) is empty. Another example of this phenomenon is provided by
smooth plane curves of degree d. Let C be one such; then the genus of Cis
given by

(d- l)(d- 2)
g = -----=---
2

On the other hand, the Brill-Noether number for g~'s is

p=3d-2g-6
= d(6- d)- 8.

For g and p given as above, the condition p < 0 can also be written as

d(d ; 3) - 8 < 3g - 3.

Now the number on the left-hand side is clearly an upper bound for the
number of moduli of curves having smooth, degree d plane models. Here
again we see that, if p < 0, then a general curve of genus g cannot be realized
as a smooth plane curve of degree d. Now let us go back for one moment to
the example following the statement of the Existence Theorem. Let us restrict
our attention to the cases for which

dim Wd( C) > max(p, - 1).

One then observes that these are

g = 4, 5, 6 C hyperelliptic,
g = 5, 6 C trigonal,
g = 6, C a smooth quintic; Chi-elliptic.

In each of these cases a direct. parameter count shows that C cannot be a

general curve of genus g. All these considerations point in the direction of
the following dimension theorem, which was first stated by Brill and Noether,
and proved by Griffiths-Harris.
214 V. The Basic Results of the Brill-Noether Theory

(1.5) Dimension Theorem. Let C be a general curve of genus g. Let d and r

be integers such that d ~ 1, r ~ 0. Then if

p = g - (r + 1)(g - d + r) < 0,

G:i( C) is empty. If p ~ 0, then G:i is reduced and of pure dimension p.

The study of Zariski tangent spaces that we carried out in Section 4 of

Chapter IV makes it natural to ask what can be said about the infinitesimal
structure of G~( C) and W~( C) when C is a general curve. This question is
answered by the following result, whose proof is due to D. Gieseker.

(1.6) Smoothness Theorem. Let C be a general curve of genus g. Let d, r

be integers such that d ~ 1 and r ~ 0. Then G~(C) is smooth of dimension p.

Corollary (Fulton-Lazarsfeld). If p ~ 1, G:i(C), and hence W:i(C), are

irreducible.

The corollary follows at once from Gieseker's theorem by Fulton and

Lazarsfeld's connectedness theorem. It is noteworthy that, for special
curves C, W:i(C) may well be reducible even if it has the expected dimension
p and p > 0. The first instance of this phenomenon occurs for g!'s on curves
of genus 5. As we already observed, if Cis a non-hyperelliptic, trigonal curve
of genus five, W !C C) is the union of two components, each of them isomorphic
to C, meeting at two points. By contrast, for a general genus 5 curve, W!(C)
is an irreducible smooth double cover of a smooth plane quintic.
The Smoothness Theorem can also be seen as a singularity theorem for
W:i(C) on a general curve C. In fact, by Proposition (4.4) of Chapter IV, the
smoothness result for G:i(C), when Cis general, makes it possible to conclude
that, when r > d - g, then for a general curve C

W~(C)sing = W~+ 1 (C).

To better understand this result we recall the expression of the Zariski

tangent spaces to G:i(C) given in Proposition (4.1) of Chapter IV. Let w be
a point of G:i(C); assume that w corresponds to an (r + !)-dimensional
subspace W of H 0 ( C, (!)(D)), for some effective degree d divisor D on C; then

dim Tw( G~( C)) = p + dim(ker f..lo. w ),

where

is the cup-product mapping. It follows that an equivalent version of the

preceding theorem is
V. The Basic Results of the Brill-Noether Theory 215

(1.7) Smoothness Theorem (Second Version). Let C be a general curve ofgenus

g. Let D be an effective divisor on C. Then the cup-product homomorphism

is injective.

This is the version that Gieseker proved and this is the one that K. Petri
stated, in a very parenthetical way, in a long forgotten paper [2]. 1
We shall now illustrate in a couple of examples the content of the
above theorem, or more precisely the role and geometrical meaning of
the map J.lo. The first example is the one of g~'s on curves of genus 4. Here we
know that, for a non-hyperelliptic curve, W~(C) consists of either two points
or of a single point, and that the second case occurs exactly when the given g~
is autoresidual and cut out by the ruling of a quadric cone Q containing a
canonical model of C. Let ID I be a g~ on C. By the base-point-free pencil
trick, the map

IS InJective if ID I is not autoresidual and has a one-dimensional kernel

otherwise. In the latter case, since the range and the domain of f.lo are both
four-dimensional, the cokernel of f.lo is also one-dimensional, and the reader
will observe that its projectivized dual can be canonically identified with the
vertex of Q.
The second example is the one of gi's on a curve C of genus 6. Let us put
ourselves in the case when C can be realized as a plane sextic with at most
double points as singularities, and more precisely with four nodes (or cusps)
p 1 , ... , p4 , no three of them collinear. For such a sextic we already noticed
that the only gi's are the ones cut out by the pencils of lines centered at
p 1 , .•. , p4 , respectively, and by the pencil of conics through p 1 , ... , p4 .
All of this is in accordance with Castelnuovo's count (1.2) which says that
Wi(C) consists of five points. To completely check Castelnuovo's count it
would of course be necessary to verify that the five gi's on Care all simple

1 "Haben 4>g>, 4>g>, ... , <P~'l eine zu G•. korresiduale Gruppe gemein, wahrend sie in der

gleichen G. treffen wie </> 0 , </> 1 , . . . , </>,so liefert der Riemann-Rochsche Satz weiter die Re-
lationen fiir r' > 0:

(2)

wenn g~: die zu g~ residuale Schar bedeutet.

Im allgemeinen Kerper sind diese (r + l)(r' + I) Formen 4> bekanntlich linear unabhangig,
bei speziellen Korpern braucht dies nicht der Fall zu sein.
Jede solche Relation legt den Parametern des Korpers eine Bedingung auf und die Zahl der
Parameter wird im allgemeinen gerade un die Zahl A. 1 dieser Relationen verkleinert." (loc. cit.,
p. 184)
216 V. The Basic Results of the Brill-Noether Theory

points of W!(C). This is what we are now going to do. Consider one of the
five g!'s, for example the one corresponding to lines through the double point
p. Let us examine f.lo in this case. An element of kerf-1 0 would correspond to
a relation

where lt. 12 are distinct lines through p, and Q 1 , Q2 are conics through the
remaining double points. If such a relation held there would exist a line I
such that Q; = ll;, i = 1, 2. This is absurd since I would then have to contain
the remaining three double points. This argument breaks down precisely
when the three remaining double points are collinear. Suppose, for example,
that our plane sex tic has four nodes and that three of them are collinear;
let p be the remaining one. In this case W !(C) is no longer reduced. In fact
it has exactly one double point, corresponding to the coming together of
two of the previously described g!'s, namely the one cut out by conics and the
one cut out by lines through p. The above argument also shows that, for this
g!, the kernel of f.lo is one-dimensional. This situation may degenerate further
and the most degenerate case occurs when all the g!'s come together. In
this case W!(C) consists of a single five-fold point. An example of this is
given by the curve whose affine equation is

It can be easily checked however, that, in all these cases the kernel of f.lo
is at most one-dimensional.
The theorems we have illustrated in this chapter, while constituting the
cornerstone of the study of the varieties G:;(C) and w:;(C), give only a faint
idea of the richness and diversity of the results that have been obtained in
this direction. Some of these, like the Martens and Mumford theorems of
Chapter IV, or like the singularity theorems of Riemann and Kempf, to
be proved in the next chapter, deal mainly with the structure of the varieties
w:;( C) themselves. Others are concerned with the geometry of the linear
series these varieties parametrize. As an illustration, we mention, among
these, the following result.

(1.8) Theorem. Let C be a general curve of genus g, and f0 E G:;(C) a general

point. Ifr ~ 3 then the map if>!!P: C--+ P' is an embedding.

Observe that by this theorem for r ~ 3 a general curve of genus g may be

represented as a smooth non-degenerate curve of degree d in P' if and only
if p = g - (r + 1)(g - d + r) ~ 0.
Bibliographical Notes 217

Bibliographical Notes

The first instance of the theory of special divisors can be found in Riemann's
famous work on abelian functions [1], but the first extensive study of
this theory appears in the fundamental paper of Brill and Noether [1].
It is there that the Brill-Noether matrix makes its first appearance and so
does the Brill-Noether number. The influence of this paper has been vast.
Enriques and Chisini devote a large part of the third volume of their treatise
[1, Ch. I, 1h; Ch. III:30, 31, 32, 33] to the problem of special linear series
and so does Severi in his Lectures [2]. The same attention to the problem
is testified to by Castelnuovo's paper [2], which contains the count of
g:i's when p = 0.
In the early sixties, A. Mayer [1], D. Mumford, and J. Lewittes [1]
revisited Riemann's singularity theorem, bringing back attention to the
theory of special divisors.
The existence theorem for grs was first proved by Meis [1]. Later, at
a time when the techniques of enumerative geometry were better understood,
the first fundamental theorem of the theory was established with a com-
pletely modern approach. In fact (partly under the influence of unpublished
work of Mumford) simultaneously Kempf and Kleiman-Laksov gave the
first rigorous proof of the Existence Theorem, and of Theorem (1.3). (See
Kempf [1], Kleiman-Laksov [1, 2].)
The Connectedness Theorem is totally new; it has no classical origin and
is completely due to the joint efforts of Fulton and Lazarsfeld [1].
The Dimension Theorem, on the contrary, has a long history, one which
cannot be recounted in these short notes. Brill, Noether, Enriques, and
Severi had a clear idea of the statement and of the reasons why it holds true,
but were never able to produce an acceptable proof. Maybe the only one
who somehow had in mind a convincing strategy was Severi [1].
The first results in this direction (dealing essentially with the case
of gJ's) were obtained by Farkas [1] and Martens [2], [3]. It was then
Kleiman who, in [1], going back to a degeneration argument ofCastelnuovo
[2], was able to reduce the Dimension Theorem to an enumerative problem
on a rational curve.
The first proof of the Dimension Theorem comes a few years later in
Griffiths-Barris [4], where a more direct degeneration argument is given
and where the enumerative problem is solved. A substantially simpler
proof of the Dimension Theorem is presented in Eisen bud-Harris [1].
As we pointed out in the text, the Smoothness Theorem was stated, in a
very indirect way, by Petri in [2]. When Petri's study of the ideal of the
canonical curve was revisited, this statement was brought to light. The
first proof of the Smoothness Theorem is due to Gieseker [1] and a more
direct and simpler proof appears in Eisenbud-Harris [2]. It should
also be mentioned that the Smoothness Theorem for GJ first appears in
218 V. The Basic Results of the Brill-Noether Theory

Lax [1], while the one for G~ is to be found first in Arbarello-Cornalba

[1]. Theorem (1.8) is proved in Eisenbud-Harris [1].
The bibliographical notes to Chapters VII, VIII, and XI will be a natural
complement to these.

Exercises
A. W!(C) on a Curve C of Genus 6

We have seen that every smooth curve C of genus 6 has a g~; in this sequence of exercises
we will let .{j; be one such.

A-1. Show that, if@ has two base points, then¢,. is a 2-to-1 map onto a conic.

A-2. Show that, if 9- has exactly one base point, then ¢,.embeds Cas a smooth plane
quintic.

A-3. Assume that q; has no base points and show that one of the following occurs:
(a) ¢,.maps C in a 3-to-1 manner onto a conic (thus Cis trigonal);
(b) ¢,. maps C in a 2-to-1 manner onto a smooth plane cubic (thus C is hi-
elliptic);
(c) ¢,.maps C in a 2-to-1 manner onto a singular plane cubic (again, Cis hyper-
elliptic);
(d) ¢,.maps C birationally to a plane sextic curve C0 . In this case C 0 cannot
have a point of multiplicity ~4, and if C0 has a triple point, then Cis trigonal.

A-4. Show that, conversely, if C is hyperelliptic then either ¢,. has two base points
or else condition (c) in the preceding exercise holds.

A-5. Show that if Cis trigonal, then either condition (a) or condition (d) in Exercise
A-3 holds, and if the latter, then C0 has a triple point.

A-6. Show that if C is a smooth plane quintic, then condition (a) in Exercise A-3
holds.

A-7. Show that if Cis hi-elliptic, then condition (b) in Exercise A-3 holds.

A-8. Suppose now that C is hyperelliptic and let ID I be the g1 on C. Show that set-
theoretically

W!(C) = u(D) + lfl(C),

and that W!(C) is singular at the point u(2D). In addition, show that W!(C)
is reduced away from u(2D). (We would guess that it is reduced at u(2D) as well,
but we haven't checked this.)
Exercises 219

A-9. Suppose that Cis trigonal and let ID I be the g~ on C. Show that set-theoretically

Wi(C) = (u(D) + W1(C)) u {K( -2D)},

where K( -2D) may or may not be in u(D) + W1 (C) (thus, the above decomposi-
tion may or may not be redundant). Show that <Ps» is of type (a) in Exercise A-3
iff»= J2DI, while it is of type (d) if i?fl = JK( -D)( -p)l # 12D 1for some p E C.
(Hint: IfiEJ, IFI are ag~ and agi, respectively, show that either IFI = lEI + p
or the map

is an embedding and use adjunction to show that Kc ~ {!)c(2E + F).)

A-10. Show that if K( -2D) rt u(D) + W1 (C), then Wi(C) is reduced; while if

K( -2D) E u(D) + W1(C),

then K( -2D) is an embedded point of W.l.(C). In the latter case, identify the
tangent space to W.l.(C) at K( -2D).

A-11. Suppose Cis hi-elliptic with two-sheeted map n: C--+ E onto an elliptic curve E.
Show that W.l.(C) is reduced and

W.i.(C) = n*Wi(E) ~ E,

where

n*: J(E)--+ J(C)

is the induced map.

A-12. Suppose that C c IP' 2 is a smooth plane quintic with {!)c(1) = {!)c(D). Show that
set-theoretically

Wi(C) = u(D)- W1 (C).

In addition, show that W.l.(C) is reduced.

Combining the preceding four exercises, we have established the following:

Suppose that C is a smooth curve of genus 6. Then the following are equivalent:
(i) Cis not hyperelliptic, trigonal, hi-elliptic, or a smooth plane quintic;
(ii) C is birationally equivalent to a plane sextic curve C 0 having double points
as singularities;
(iii) dim W.i.(C) = 0

For the remainder of this sequence of exercises we will restrict our attention to
curves satisfying the conditions of the theorem and we will denote by <f>: C--+ C 0 a
birational map of C to a plane sextic C 0 having only double points.
220 V. The Basic Results of the Brill-Noether Theory

A-13. Show that, if p 1 , ••• , p 7 E IP' 2 are seven distinct points that fail to impose in-
dependent conditions on the linear system I(!)lf'2(3) I of plane cubics, then five
of the P; are collinear, and conversely.

A-14. Using the preceding exercise, show that if eight distinct points p 1 , ••• , p 8 E IP' 2
fail to impose independent conditions on I(!)lf'2(3) 1, then either five are collinear
or all eight lie on a conic, and conversely.

A-15. Using the preceding exercise, show that the only g!'s on Care those cut out on C
by pencils of lines through double points of C0 , and the one cut out on C by conics
passing through all four nodes of C0 (more generally, satisfying the adjoint
conditions on C, cf. Appendix A).

A-16. Compute ker Jl.o for the g.J. cut out on C by lines through a double point p of C 0
in the following cases:
(i) pis a node and the remaining double points are not collinear;
(ii) p is a node and the remaining double points are collinear (we assume that
the double points are distinct):
(iii) p is a tacnode:
(iv) p is the intersection of two smooth branches having contact of order 3
(we call this an oscnode).

A-17. Show that a plane sextic with four distinct nodes, of which no three are collinear,
can never be birational to a plane sextic with a tacnode.

A-18. Show that a plane sextic C 0 with four distinct nodes, three of which are collinear,
is birational to a plane sextic C0 with one tacnode and two ordinary nodes.
Show that the tangent line to C0at the tacnode cannot pass through either of the
remaining nodes.

A-19. Let C0 be a plane sextic having two tacnodes such that the tangent line to C0
at each tacnode does not pass through the other. Show that C0 cannot be hi-
rational to a plane sextic with an oscnode (cf. A-16) and a node.

A-20. Show that the support of W,l.(C) consists of one point if and only if C0 has a
singularity consisting of two smooth arcs meeting with contact of order 4 at a
point which is a flex of each arc.

B. Em beddings of Small Degree

In this batch of exercises we will not assume Theorem (1.8) of Chapter V.

B-1. Let C be any curve of genus g and D a general effective divisor of degree g + 3.
Show that:

r(D) = 3;

IDI is base-point-free;
c/Jn: C -> IP' 3 is an embedding.
Exercises 221

B-2. Show that a hyperelliptic curve C of genus g ~ 2 can never be embedded in IP'
with degree at most g + 2.

B-3. Suppose that Cis trigonal and Dis a general effective divisor of degree d. Show that
¢K<-DJ is not an embedding unless D = 0. (Thus we cannot biregularly project
¢K(C) from a linear space spanned by general points on the canonical curve.)

B-4. Similarly, suppose that C is a smooth hi-elliptic curve of genus g ~ 6 and D a

general effective divisor of degree don C. Show that ¢K<-DJ is not an embedding
unless d ~ 1.
Conversely, we will now see that, if C is not hyperelliptic, trigonal, or hi-
elliptic, then ¢K(C) may be biregularly projected to a curve of degree g + 2.

B-5. Let C be a curve of genus g ~ 4 not hyperelliptic, trigonal, or hi-elliptic, and D a

general effective divisor of degree g - 4. Show, using Mumford's theorem, that

cPK(-D); C-+ iP 3

embeds C as a curve of degree g + 2.

B-6. Let C be a general curve of genus g with 4 ~ g ~ 7. Show that g + 2 is the smallest
degree of an embedding of C in IP'. On the other hand, for g = 8 show that

W~(C) = 0
and conclude that C may be embedded as a curve of degree 9 in IP 3 •

C. Projective Normality (II)

This series of exercises will investigate the question: What is the smallest r such that a
general curve C of genus g can be embedded in IP' as a projectively normal curve? By
Max Noether's theorem on the projective normality of the canonical curve we know
that r ~ g- 1.

C-1. Show that a general curve of genus g ~ 6 cannot be realized as a projectively

normal curve in IP 3 •
(Hint: Use (1.5) to estimate the degree of C, and then study H 0 (C, CDc(2))).

C-2. More generally, by a similar argument show that, if C c IP' is a projectively normal
general curve of genus g, then
g ~ r(r + 1)/2.
C-3. Using properties of scrolls show that if C c IP' is linearly normal and hyper-
elliptic, then the linear series cut out on C by quadrics has vector space dimension
~ 3r, and hence that if C c IP' is projectively normal and hyperelliptic then

{
d ~ 2g + 1,
r?.g+l.
222 V. The Basic Results of the Brill-Noether Theory

C-4. Suppose that Cis a smooth curve of genus g and Dis a non-special effective divisor
of degree d 2': g + 1 with ID I base-point-free. Show that the map

is surjective for k 2': 2.

(Hint: Let V c H 0 (C, (I)(D)) correspond to a base-point-free pencil in IDI and
use the base-point-free pencil trick to deduce the exact sequence

0--> (l)((k - l)D)--> V ® (l)(kD)--> (l)(kD)--> 0.

Now use the Riemann-Roch theorem together with the non-speciality of D.)
A consequence of this exercise is the assertion:

( *) IfC c 'P'' is any smooth curve for which (l)c(l) is non-special, then Cis projectively
normal if, and only if, it is linearly and quadratically normal.

In the remaining series of exercises we will prove the following:

(**)Theorem. If C is a general curve of genus g 2': 3 and D a general divisor of degree

d = [3g/2] + 2, then ¢v(C) is projectively normal.

Corollary. A general curve of genus g may be realized as projectively normal curve in

'P'' for

C-5. Let C be a general curve of genus g and D a general divisor of degree d.

(i) Show that, if d < 3g/2 + 2, then D is not linearly equivalent to a sum of
divisors D 1 , D 2 with r(D 1 ), r(D 2 ) 2': 1.
(ii) Show that, if d = 3g/2 + 2, then D is the sum of at most finitely many such
pairs.

C-6. Let C be a smooth curve and D a divisor such that there exist linearly independent
sections

satisfying

Show that D is linearly equivalent to a sum of divisors that move in pencils.

Conclude, using the preceding exercise, that if C is a general curve of genus g and
D a general divisor on C of degree d < 3gj2 + 2 (resp., d = 3gj2 + 2), then the
image ¢v(C) c pd- 9 lies on no (resp., at most finitely many) quadrics of rank ::;; 4.
Exercises 223

C-7. Using Exercise A-2 of Chapter II, show that if a curve C c IP' lies on no quadrics

of ranks 4, then it lies on at most(' ; 2) linearly independent quadrics. Similarly,

if C lies on finitely many quadrics of rank s 4, then it lies on at most (' ; 2) +1

linearly independent quadrics. The theorem now follows.
Referring to the question posed at the beginning of this series of exercises,
we have

+ J89+1
-1
2 s rs z3g + 2.
So far as we know, it is an open question just what the best estimate for r is.

D. The Difference Map ¢d: Cd x Cd--+ J(C) (I)

In this sequence of exercises we shall study the map

defined by

cf>iD, E) = u(D - E).

The image will be denoted by ~(C), or simply by~ if no confusion is possible. It is

an analytic variety whose fundamental class we denote by vd.

D-1. Show that the differential of cf>d has maximal rank min(2d, g) at a general point,
and hence that dim(~)= min(2d, g).

D-2. Show that

is birational if C is non-hyperelliptic and has degree 2 if C is hyperelliptic. More

generally, show that if Cis non-hyperelliptic then cf>d is birational ford < gj2.

D-3. Show that, if Cis non-hyperelliptic and d < gj2, then

Similarly, show that for any smooth curve of genus g = 2d, the mapping cf>d has

degree(~).
224 V. The Basic Results of the Brill-Noether Theory

(Remark. In proving this by a computation similar to the proof of Poincare's

formula you need not assume anything about the rank of (c/> 912 )*. Having proved
that the degree is non-zero, you may conclude that every divisor of degree 0 is the
difference of two effective divisor classes of degree g/2.)

D-4. Suppose that C is a non-hyperelliptic curve of genus 4 and suppose that D is a

divisor of degree 0 on C. Show that

ifu(D) = 0,

dim <I> '(«(D)) ~ {; if u(D) E V1 u (K - 2Wi(C)),

otherwise,

(the middle equation means that either D - p - q or else D - E - E' where E, E'
are divisors from the two gi's on C).

D-5. Keeping the notations of the preceding problem, assume that [E] # [E'] and
show that

l-Yg_ 1 (C) n (l-Yg_ 1 (C) + u(E- E')) = V2 (C) + u(E- E').

Chapter VI

The Geometric Theory of

Riemann's Theta Function

The results of the Brill-Noether theory stated in the preceding chapter have
been proved in full generality only fairly recently. However, special but
important cases of them were classically known and, in a sense, provided a
motivation for the entire theory. What we have in mind here are the classical
theorems concerning the geometry of Wy _ 1(C), that is, the geometry of
Riemann's theta function. Of course, these results are more than mere
exemplifications of the general ones of Chapter V. Rather, they are to be
viewed as illustrations of how those general results can be used in the study of
concrete geometrical problems. Our analysis will be carried out partly by
means of classical methods, and partly in the language of Chapter IV.
In the first section we shall be mostly concerned with the proof of the
beautiful Riemann singularity theorem, which gives a precise geometrical
interpretation of the multiplicities of the singular points of the theta divisor.
In the second, following Kempf, we shall show that Riemann's singularity
theorem, as well as a generalization of it, due to Kempf himself, are indeed
formal consequences of the general theory exposed in Chapters II and IV.
In the third section we shall give Andreotti's proof of Torelli's theorem.
In the fourth, and final, section we shall present the theory of Andreotti
and Mayer, whose goal is to try and geometrically solve the Schottky
problem of giving explicit equations for the locus of Jacobians inside the
generalized Siegel upper half-plane. Our presentation will make use of a
theorem, whose proof is due to Mark Green, to the effect that the vector
space of quadrics through a canonical curve is spanned by the tangent cones
to the theta divisor at double points.

§1. The Riemann Singularity Theorem

Throughout this section, C will denote a fixed smooth curve of genus g. We

fix a basis w 1 , ... , wg for the space H 0 (C, K) of abelian differentials on C. As
usual, we shall denote by
226 VI. The Geometric Theory of Riemann's Theta Function

the canonical map, and we recall the abelian sum map

u: cd--+ J(C)

given, forD = "'t; 1 p;, by

One of the most beautiful results concerning special divisors relates the
dimension r(D) of the linear system in which D moves to the multiplicity of the
point u(D) on lfd(C), where d is the degree of the divisor in question. This
theorem dates from Riemann, and, in its modern form, is due to Mayer,
Lewittes, and Mumford in case the degree is g - 1, and to Kempf in general.
We shall begin by proving the special case when the degree is g - 1, and then,
in the next section, shall prove the general result.
We choose on C a base point that will be kept fixed throughout the dis-
cussion. This allows us to identify the Jacobian J(C) with Picd(C). Recall that,
by Riemann's theorem, the divisors 0 and J¥g _1 (C) are translates of one
another. More precisely

where K is Riemann's constant. The result we wish to prove is the following.

Riemann Singularity Theorem. Let C be a smooth curve of genus g. Then for

every effective divisor D of degree g - 1,

Before proving the theorem we would like to make a couple of remarks.

The first one concerns notation. It will be convenient to use, instead of u, the
map

n: C9 _ 1 --+ J(C),

which we defined at the end of Section 5 of Chapter I by letting n(D) be equal

to u(D) + K. As we explained there, one nice feature of n is that

- n(D) = n(K - D).

The Riemann singularity theorem can now be restated as saying that

§I. The Riemann Singularity Theorem 227

If h 0 (C, {I)( D)) = r + 1 and 8 is Riemann's theta function, this amounts to

saying that

ase
(1.1) au ... au (n(D)) = 0 for s ~ r,
a1 IXs

ar+ le
(1.2) au ... au (n(D)) # 0 for some multi-index (a 0 , ..• , a,).
IXQ ar

We shall begin by proving the first relation.

The geometric idea is that the whole projective space 1D 1 is contracted to
the point n(D) on e, and we shall explore the consequences obtained by
"differentiating" this fact. Choose any divisor D 0 = L PiE ID I and let p;(t),
i = 1, ... , g - 1, be holomorphic arcs in C such that Pi(O) = Pi· Set

Then, by Riemann's theorem,

We shall successively differentiate this equation with respect to t and then

evaluate at t = 0. The first step gives

L ~ (n(D)) aua (0) = 0.

a aua at
In a neighborhood of Pi we choose a local coordinate zi and write

p;(t) = z;(t).

We then have

and since the z;(O) are arbitrary we infer that

(1.3)
ae (n(D))gai(Pi) = 0
L-;- for i = 1, ... , g - 1.
:x uUa
228 VI. The Geometric Theory of Riemann's Theta Function

This says that, in IP 9 - 1 with homogeneous coordinates [wb ... , w 9], the
equation

is satisfied by all points

Equivalently, the hyperplane above contains all the points cfJK(p;) on the
canonical curve.
Thus, if r(D):;::: 1, so that there is a divisor D 0 E ID I containing any given
point of C, the above linear relation holds on the entire canonical curve.
Since this curve is non-degenerate we conclude that n(D) is a singular point
of E>. Of course this we already knew, even in greater generality (cf. Corollary
(4.5) in Chapter IV). All we have done here is to present the classical
approach.
The remainder of the proof of (1.1) proceeds similarly. Assuming that
r(D) :;::: 1 we take the second derivative of 8(n(D,)) =
0 and evaluate at t = 0,
using the vanishing of the first derivatives, to obtain

Since the z;(O)'s are arbitrary this implies that

(1.4)

for any i andj. Ifr(D) :;::: 2, we may find a divisor D 0 E ID I passing through any
given pair of points on the curve. Then, because the canonical image of C is
non-degenerate, the bilinear form

Q(w, w') = I aazae (n(D))wawp

a,p Ua Up

must vanish identically. More generally, if r(D) :;::: k, the same kind of argu-
ment shows that the k-linear form

vanishes identically. This concludes the proof of (1.1).

§I. The Riemann Singularity Theorem 229

We shall now prove (1.2). The case when r(D) = 0 is a special case of (4.5)
in Chapter IV, but it is immediate to give an elementary proof. Notice that, if
h0 (C, <P(D)) = 1, then the rank of the Brill-Noether matrix at Dis exactly
g - 1, so that the map n has maximal rank at D. Since, by Abel's theorem,
n is one-to-one at D, the only thing that might prevent E> from being smooth
at n(D) would be if E> were non-reduced. But this is forbidden by Riemann's
theorem.
The proof of ( 1.2) uses the fact that, if D belongs to Cd, then

(1.5)

This is easy. In local coordinates, u* is represented by the Brill-Noether

matrix, whose columns therefore span u* Tv(Cd). Recalling the explicit
calculation of the Brill-Noether matrix in Section 1 of Chapter IV, the
formula follows.
We will now prove (1.2), and we begin by establishing some notations.
Consider a complete linear series ID I of degree g - 1 and dimension r. Given
a general effective divisor E = p 1 + · · · + Pr + 1 of degree r + 1, we set

E; = P1 + · · · + Pi + · '· + Pr+ 1;
we then have
(i) ID + El = IDI + E (i.e., r(D +E)= r),
(ii) ID-£1=0,
(iii) ID - Ei I # 0 for i = 1, ... , r + 1
(the third condition is automatic; we state it only for reference). Consider the
divisor

X Pi = {E' E cr+ 1: E' - Pi ~ 0}

in C,+ 1 . This divisor is clearly isomorphic to C, and hence smooth. We now

consider the holomorphic mapping

v: c,+ 1 ~ J(C)
defined by

v(E') = u(D + E- E') + K.

The proof of (1.2) will consist in examining the divisor v*E> on C,+ 1 . We first
claim that the support of this divisor is given by

supp(v*E>) = ( yxp,) u {E': ID- E'l # 0}.

230 VI. The Geometric Theory of Riemann's Theta Function

Indeed, by Riemann's theorem and condition (i) above, the support of v*e
consists of all E' E C,+ 1 such that, for some D 0 E ID I,

D0 + E- E' :2': 0.

Our claim follows from conditions (i) and (iii) above. In a moment we will
show that X Pi occurs with multiplicity one in v*e. Assuming this, (1.2)
follows by observing that, by condition (ii) above, E does not belong to
{E': ID - E' I =1= 0} and consequently, in a neighborhood of E E C,+ 1, the
divisor v*e is equal to xp, + + XPr+l' Thus
0 0 0

r +1= multE(v*0)
:2': multv<EJ(0)
= mult"<DJ(0),

which proves (1.2).

It remains to prove that X Pi occurs with multiplicity one in v*e. For this it
suffices to show that, for a general point

E; = Pi + q 1 + ... + q,
of X Pi' the tangent spaces to e and v( c,+ 1) intersect transversely at v(Ei) =
u(D + E - E;) + K, i.e., that

To show this we will use (1.5). We first choose general points q 1 , ... , q, so
that r(D - q 1 - · · · - q,) = 0, and then for a unique D 0 E ID I we will have

Do = D'o + q 1 + ... + q" D0 :2':0.

Having chosen the q;s, we then choose the p;'s to satisfy

as well as the conditions (i), (ii), (iii) above. It is clear that all these conditions
are satisfied in a non-empty Zariski open subset of C,+ 1 . By (1.5) we have that

On the other hand,

IP>v* TE;(Cr+l) = IP>u* TE;(Cr+l)

= (!JK(E;).
§I. The Riemann Singularity Theorem 231

But then
¢K(P;) ¢ PI;,(E;l8),
¢K(P;) E Pv* TE;(Cr+ 1),

proving that v* TE;(Cr+ 1 ) is not contained in 7;,<E;J(8). This concludes the

proof of the Riemann singularity theorem.

The Riemann singularity theorem gives us information about the first

non-vanishing term in the power series expansion of (}(u) around a point
n(D) E e. Specifically, if DE C9 _ 1 and r(D) = r, then (}(u + n(D)) = V.+ 1 (u)
+ V.+ 2 (u) + · · · where Vm(u) is a homogeneous polynomial of degree m in
u E CB and V.+ 1 i= 0. We will now give a result concerning the terms

V.+1, ... , Vzr+1·

To explain the theorem we write

and define a symmetric, m-linear form vm on C 9 by

IZt, ... ,IZm

Then Vm is just the polarization of Vm; i.e.,

We observe that V.+ 1 , viewed as a polynomial on T,(v)(J(C)), is well defined up

to scalar multiples (here, well defined means allowing arbitrary local holo-
morphic changes of coordinates). In general, Vm is well defined, up to scalar
multiples, modulo the ideal in C[u1, Ug] generated by v.+b
0 0 0, 0 0 vm-1·
0,

By abuse oflanguage we will refer to "the hypersurface Vm in P9 - 1 ,"by which

we mean the hypersurface in PH 0 (C, Kc)* defined by the form vm. By our
remarks above, the intersections

are all intrinsically defined subvarieties of canonical space pg- 1 .

For our last piece of terminology we recall that the k-secant variety to a
curve r c pn is the closure of the union of the linear subspaces in pn spanned
by a k-tuple of distinct points of r. Thus, the 1-secant variety is the curve
itself, and in general the k-secant variety contains the locus of osculating
(k - 1)-planes to the curve r.
232 VI. The Geometric Theory of Riemann's Theta Function

(1.6) Theorem. Assume that ID I is a base-point-free linear system of degree

g - 1 and dimension r. Then:

(i) ForeverydivisorAEIDI, V.+ 1 containsthe(g- r- 2)-planecfJK(A).

In particular, V.+ 1 contains the r-secant variety of ¢K(C).
(ii) For s :2: 0, V.+2+s contains the (r- s)-secant variety of ¢K(C). In
particular, V.+ 1 , •.. , V2 ,+ 1 all contain ¢K(C).
(iii) V.+ 2 is zero modulo V.+ 1 if and only if D is semi-canonical.
Informally, (ii) states that "¢K(C) is very much contained in E>."
For an application we consider a non-hyperelliptic curve C of genus four.
As we know, the canonical curve

is the complete intersection of a quadric Q and cubicS in IP' 3 • In general, the

quadric is smooth and its two rulings cut out the two distinct g~'s on C; Q is
singular exactly when C has a semi-canonical g~.
Let ID Ibeag~ on C. By the Riemann singularity theorem and part (i) of the
theorem, the tangent cone to E> at n(D) is the quadric Q. From (iii) we then
have the

Corollary. Let C be a genus four canonical curve having no semi-canonical

pencil, 8(u) its theta function, and WE IC 4 a point at which 8(w) = o8joua(w) = 0
(rx = 1, ... , 4). Writing

8(u + w) = V2 (w) + Viw) + ···,

the curve C is the locus in IP' 3 defined by Viu) = Viu) = 0.

We observe that this corollary gives, in this case, an explicit reconstruction

of the curve from its polarized Jacobian.
The proof of theorem (1.6) will occupy the remainder of this section. Being a
generalization of the Riemann singularity theorem, the proof will be an
extension of the argument used there.
If A = p 1 + · · · + p0_ 1 E C0_ 1, with the points Pi being distinct, if ti is a
local coordinate centered at Pi, and Pi(t;) stands for the point corresponding to
the value ti of the coordinate, we set

n(t 1, ... , t0 -l) = n(pl (t 1) + · · · + P0-1 (to-1)),

on
w(pi) = -a (0),
ti
ak+ ln
w<k>(p;) = otk + 1 (0).
'
§I. The Riemann Singularity Theorem 233

lthasalreadybeenobserved thatw(p;) E 1[:9 - {0} projects onto <f>x(P;) E IP> 9 - 1.

More generally, w(p;), w'(p;), ... , w<kl(p;) span in 1[:9 a linear space that
projects onto the osculating space <f>K((k + 1)p;) to <f>K(C) at <f>K(p;).
We now prove part (i) of the theorem. By Riemann's theorem, O(n(t)) = 0.
Taking derivatives, for every collection k 1 , ... , k 9 _ 1 with k; = r + 1 we Li
have

since all lower-order derivatives of(} are zero, by (1.1 ). The restriction of the
multilinear form V,.+ 1 to the (g - r - 2)-plane <f>K(/'1) is thus identically zero,
I.e.,

Since every r-secant linear space to <f>K(C) is contained in <f>x(A) for some
~E ID I, this proves (i).
Part (ii) will be established by a similar argument together with the
following

Lemma. Let x(t) = (x 1(t), ... , xit)) E IC"- {0} define an analytic arc in
IP>"- 1, and let Q(v 1 , ... , v1) be a symmetric /-linear form with Q(v) = Q(v, ... , v)
the corresponding polynomial. Suppose that Q vanishes on the r-secant plane
locus of the arc, i.e.,

(1.7)

for all t t. ... , t, and alll-tuplesj 1 , ••• , j 1 in {1, ... , r}. Then for any t 1 , .•. , t,_.,
k1, •.. , k 1 with L
k; :::;; s + 1, and i 1 , ... , i 1 E {1, ... , r- s}, we have

Informally, if the number N of derivatives taken plus the number P of

points of the arc involved in the argument of Q satisfy N + P :::;; r + 1, then
Qis zero. Observe that this is one better than the obvious: ifl: k; :::;; s then the
argument of Qwould lie in a r-secant plane to the arc, and by assumption Q
would vanish.

Proof of the Lemma. We use induction on s, the chain rule

iJQ (vl (t), ... , vlt)) = "L- Q(v

~
- 1(t), ... , V;(t), I
... , v1(t)),
ut i
234 VI. The Geometric Theory of Riemann's Theta Function

and the symmetry of Q. Differentiating (1.7) with respect to ti, gives

for all tt. ... , tr and j 1 , ••. ,j1 E {1, ... , r}, which is the lemma for s = 0.
If tt.···,tr-., kt.···,k 1, and it.····i1 E{l, ... ,r-s} are as in the
statement of the lemma, and if k 1 > 0, then by induction

for any t. Differentiating this equation with respect to t and then setting
t = ti, yields the result. Q.E.D.

We now turn to the proof of (ii). In this and in the following computations
use will be made of the chain rule, which as we now explain simplifies con-
siderably since 8 is a variety of translation type. Namely, recalling that

n(t 1• ... , t 9 -1) = n(p1 (t 1) + · · · + P9 -1Ct9 -1))

= u(p1(t 1)) + · · · + u(p 9 -1(t9 -1)) + K,
it follows that

so that in applying the chain rule all crossed mixed partials are zero.
Thus, for integers / 1 , ... , lr such that L li = r + 2, differentiation of
8(n(t)) = 0 gives

w(p), ... , w(pi), ... , w'(pi), ... , w(pr), ... , w(p,))

~
~-1 ~
+ V,.+ 2(w(p1), ... , w(p1), ... , w(pr), ... , w(pr)),
'-v----'
I,

again since the lower-order derivatives of 8(u) are zero at u = n(M. By the
case s = 0 of the lemma, each term in the first sum is zero. Since the points
p 1 , •.. , Pr may be chosen arbitrarily we conclude that V,:+ 2 contains the r-
secant variety to ¢K(C), which is the cases = 0 of (ii).
§I. The Riemann Singularity Theorem 235

We shall now do the case s = 1; the computation will make the general
pattern clear. For integers 11 , ... , 1,_ 1 with L
l; = r + 3, differentation of
8(n(t)) = 0 gives

I,

where

A1 = i~i ~ ([.) (l·)-

~ V,.+ 1(w(p 1), ... , w(p 1 ), ••• , w(p;), ... , w(p;), w'(p;), ... ,
4 4- I

w(p), ... , w(p), w'(p), ...)

li- I

A2 = ~3 (d[.)-V,.+ 1(w(p1), ... , w(p 1 ), ••• , w(p;), ... , w'(p;), w'(p;), ... ,
/1 I,- 2
W(P,-1), · · ·, w(p,_ 1))
I,_,

A3 =~ (~) V,.+ 1(w(p 1), ... , w(p 1), ... , w(p;), ... , w(p;), w"(p;), ... ,
11 I,- 2
w(p,_ 1), ... , w(p,_ 1))

([.)-
A4 = ~ ~ V,.+iw(p 1 ), ••• , w(p 1), ... , w(p;), ... , w(p;), w'(p;), ... ,

4 4- I
W(Pr-1), · · ·, w(p,-1)).

By the lemma in the case s = 1, each term in the sums A 1 , A 2 , and A 3 is zero.
Moreover, each term in the sum A4 is zero because of the cases = 0 of(ii) and
the fact that w(p 1 ), ••• , w(p,_ 1 ), w'(p;) lie in the r-secant space

to 4>K(C). This shows that V,.+ 3 contains 4>K(p 1 + · · · + p,_ 1 ), and since the P;
were arbitrary this gives the case s = 1 of (i) in the theorem.
236 VI. The Geometric Theory of Riemann's Theta Function

We now turn to part (iii). One direction is elementary from symmetry

considerations: by Exercise 44 of Appendix B, if D is semi-canonical then
the function

where l(u) is a suitable linear form on C 9 , is either odd or even. In either case, if
V0 = · · · = V,. = 0 and V,.+ 1 # 0, then V,.+ 2 = 0. As in the proof of (1.2), the
other half of (iii) involves new considerations, but the idea is simple. In parts
(i) and (ii) we proved that both V,.+ 1 and V,.+ 2 contain the r-secant variety but,
as we proved in (i), the distinguishing feature of V,.+ 1 is that

V..+ 1 ::::J U cfJK(/1).

deiDI

By Clifford's theorem, except in the case when C is hyperelliptic and ID I is

multiple of the gL this is a stronger statement than just saying that V,.+ 1
contains the r-secant variety. Assuming that 2D ¢ IK I (which in particular
rules out the hyperelliptic case), to prove that V,.+ 2 if= 0 modulo V,. + 1 , it will be
sufficient to show that

(1.8) V..+z P cfJK(D).

Since D is base-point-free, we may assume that D = p 1 + · · · + p 9 _ 1o

where the P; are distinct. Since 2D ¢ IK I we have

i.e., the tangent lines to the canonical curve at the points c!JK(p;) span iP> 9 - 1•

Renumbering if necessary, we may assume that

By the geometric form of the Riemann-Roch theorem this is equivalent to

(1.9) ID + P1 + ·· · + Pr+11 = IDI + P1 + ··· + Pr+1·

We define
v: C,+ 1 ~ J(C)
by

v(E) = u(D + p 1 + · · · + Pr+ 1) - u(E).

As in the proof of (1.2), the argument for (iii) will consist in examining the
divisor v*(E>).
§I. The Riemann Singularity Theorem 237

We set
Xp, = C, + P; c Cr+1•

and will prove the following

Lemma. (i) l: is smooth at p 1 + · · · + Pr+ 1 .

(ii) in a neighborhood of p 1 + · · · + Pr+ 1 , we have
r+ 1
v*(e) = L xp, + l:;
i= 1

moreover, l: and X p, intersect transversely.

Assuming the lemma we may complete the proof of (iii) as follows. Using
the notations from the proof of (ii), we set

v(t) = u(D + P1 + · · · + Pr+1- P1(t1)- · · · - Pr+1(tr+1)).

Then the function (}(v(t)) is defined in a neighborhood of the origin in e+ 1,
and

J.l.o(}(v(t)) = r + 2.
More precisely, the lemma tells us that, up to a change of coordinates, the
tangent cone to the divisor of (}(v(t)) consists of the union of the coordinate
hyperplanes in e+ 1 together with a hyperplane not containing any of the
coordinate axes. In particular,

By the computation in the proof of (ii),

Thus
238 VI. The Geometric Theory of Riemann's Theta Function

On the other hand, again by the computation in the proof of (ii), differentia-
=
tion of 8(n(t)) 0 gives

0- ar+ 28(n(t)) (0)

- oti ot 2 • · · ot,+ 1
= V,.+ 1(w'(p1), w(pz), · · ·, w(p,+ 1))
+ V,.+ z(w(pd, w(p1), w(pz), . · . , w(p,+ 1)).

Subtracting the last two relations we obtain

i.e., the hypersurface V,+ 2 does not contain the r-secant (g - r - 2)-plane
¢K(D), so that (1.8) holds. The proof of(iii) will be complete once we establish
the lemma.

Proof of Part (i) of the Lemma. Let s 1 , ... , s,+ 1 be a basis for H 0 (C, lP(D)) and,
by choosing local trivializations of lP(D), represent Sj by a holomorphic
function fPJ in a neighborhood of Pi. Setting f(t;) = (f1 (tJ, ... , f..+ 1 (tJ) E
cr+ 1, L has local defining function
a{t 1 , ..• , t,+ 1 ) = det(,f;{ti))
= f(t1) 1\ • • • 1\ f(tr+ 1).

Thus, for example,

Since the unique section of lP(D) vanishing at p 2 , ••• , Pr+ 1 vanishes to order
exactly one at p 1 , this is non-zero at the origin. Therefore, L is smooth at
P1 + · · · + Pr+1·
Proof of Part (ii) of the Lemma. This is very much like the proof of(1.2). First,

by (1.9) we have
supp v*(0) = ( yX P.) u L.
If E + Pi is a general point of X Pi,
r(D + P1 + · · · + Pi + · · · + Pr+ r - E) = 0,
§2. Kempf's Generalization of the Riemann Singularity Theorem 239

so that 8 is smooth at v(E + pJ Moreover, if D 0 + E is the unique divisor in

ID I containing E, the tangent plane to e at v(E + P;) is given by

Since, by hypothesis, this does not contain ¢K(P;), whereas

does, we conclude that 8 meet v(C,+ 1 ) transversely at v(E + p;); in particular

multxp, v*(E>) = 1.

Similarly, if EEL is near p 1 + · · · + Pr+ 1 we have

r(D- E)= 0.

If D 0 + E is the unique divisor in ID I containing E, then 8 is smooth at v(E)

with tangent plane

On the other hand,

v* IPTiC,+ 1) = cPK(E).
Since, by (1.9). D 0 , E, and p 1 , ..• , Pr+ 1 do not all lie in a hyperplane, near
P1 + ··· + Pr+1

multl: v*(E>) = 1.

This completes the proof of the theorem.

§2. Kempf's Generalization of the Riemann

Singularity Theorem

In this section we momentarily interrupt our exposition of the classical

geometric theory of the Riemann theta function to briefly talk about an
important generalization of the Riemann singularity theorem, due to Kempf.
The result is an exact analogue for ~(C) of what Riemann did for the theta
divisor, or, what is the same, for Jify_ 1 (C). One of the main difficulties that
Kempf had to overcome is that ~(C) is no longer a divisor, so that the
expression of its tangent cones cannot be given by taking derivatives of a
single function. The proof will be a fairly easy application of the machinery
240 VI. The Geometric Theory of Riemann's Theta Function

developed in Chapters II and IV. Actually, we will find it convenient to

prove a singularity theorem for W~( C) which reduces to Kempf's when r = 0.
We begin by recalling the content of Proposition (4.4) in Chapter IV.
Suppose C is a smooth genus g curve and let d, r be non-negative integers
such that d < g + r. Assume that G~(C) is smooth of the "correct" dimen-
Sion

p = g - (r + 1) (g - d + r).
Then W;j(C) is a Cohen-Macaulay, reduced, normal variety and its singular
locus is precisely W:i+ 1 (C).
The problem now is to describe the tangent cone, 5L(W:i(C)), to W~(C) at a
point£, and to compute its degree in terms of h0 (C, L). Of course, this tangent
cone will be viewed as a subvariety of H 0 (C, K)*, which is the Zariski tangent
space to Picd(C) at L. In order to state the result, we have to choose bases
x1, ... , Xt, Y1> ... , Yi for H 0 (C, L) and H 0 (C, KL - 1 ), respectively. For
simplicity, we shall write x.y, to denote the image of x. ® y, under the cup-
product mapping

As usual, we shall denote by

the natural mapping. The result we wish to prove is the following.

(2.1) Theorem. Let L be a point of W~( C). Assume that G~( C) is smooth of the
"correct" dimension

p = g - (r + l)(g - d + r)

in a neighborhood of c- 1 (L), and that d < g + r. Then:

(i) ~(W:i(C)) is Cohen-Macaulay, reduced, normal, and, set-theo-

retically

.9-;JW;j(C)) = U Jl.o(W ® H 0 (C, KL- 1 ))1_

WeG(r+ 1,H0(C,L))

(ii) The ideal of ~(W:i(C)), as a subvariety of H 0 (C, K)*, is generated by

the (l- r) x (l- r) minors of the matrix

(x.y,)s= 1, ... ,I·

t= 1, ... , i
§2. Kempf's Generalization of the Riemann Singularity Theorem 241

(iii) The degree of 5i(W~(C)) is

rl (i + h)! h! .
h=o (/- 1 - r +h)! (i- l + r + 1 +h)!

Since G~(C) = Ca is smooth of dimension d = p(g, 0, d), when r = 0 we

get, as a special case,

Kempf's Singularity Theorem. Let L be a point of Wa(C) with d < g. Then

.Yr.(~( C)) is a Cohen-Macaulay, reduced, normal subvariety of H 0 (C, K)* of
degree

( i ) =
l- 1
(g - 1-
d+l-
1
1) ,

whose ideal is generated by the maximal minors of the matrix (XsYr)s=l, ... , 1,
and is supported on r=l, ... ,i

U .Uo(W ® H 0 (C, KL -t)).L.

We!Jl'HD(C,L)

From a geometrical point of view it is interesting to look at the projectiviza-

tion of the tangent cones to W:i( C). These sit naturally in the space of the
canonical curve. Under the hypotheses of Theorem (2.1 ), we get

In Kempf's case, we find:

IP'§"L(~(C)) = U (fJK(D).
DelLI

We shall now prove Theorem (2.1). To describe the tangent cone to W:i(C)
at L we shall use Lemma (1.3) in Chapter II. This requires computing the
normal bundle to c- 1 (L), which we shall denote by N. The answer is provided
by Proposition (4.1) in Chapter IV. By the smoothness assumption on G:i(C),
we find that, for any (r + i)-dimensional subspace W of H 0 (C, L), the
restriction of .Uo to W ® H 0 ( C, KL- 1) is injective. Since the normal space to

at W is nothing but the image of

242 VI. The Geometric Theory of Riemann's Theta Function

we also find that

N = {(W, v): v _l_,u 0 (W ® H 0 (C, KL - 1 ))}
c G(r + 1, H 0 (C, L)) x H 0 (C, K)*,

and that the differential

is just projection onto the second factor. The rest of the proof is reduced to a
linear algebra result. In order to state it, we let A, B, E be complex vector spaces
of dimensions a, b, e, and let

</>:A® B--+ E

be a linear map. We choose bases xi> ... , x. and Y! •... , Yb of A and B,

respectively, and write, for simplicity

Lemma. Let w be an integer such that 1 ~ w ~ a. Assume that a < b + w and

that,Jor each Win G(w, A), the restriction of¢ toW (8) B is injective. Set

I= {(W, y): y _l ¢(W ®B)}

c G(w, A) x E*.
Let

p: G(w, A) x E* --+ E*

be the projection. Then:

(i) p(I) is Cohen-Macaulay, reduced and normal.
(ii) The ideal of p(I) is generated by the (a - w + 1) x (a - w + 1)
minors of the matrix (X;Y)i=l, ... ,a·
j= 1, ... ,b
(iii) The degree of p(I) is

w- 1 (b+h)!h!
}lo (a- w +h)! (b- a+ w +h)!
(iv) p maps I birationally onto p(I).

Before proving the lemma, we deduce from it the theorem. In our situation,
A is H 0 (C, L), B is H 0 (C, KL - 1 ), E is H 0 (C, K), ¢is ,u0 , and w is r + 1. The
hypothesis that a < b + w is an exact translation of d < g + r, by the
Riemann-Roch theorem. Also, I is just N, and the restriction of p to I is c*.
§2. Kempf's Generalization of the Riemann Singularity Theorem 243

To apply Lemma (1.3) of Chapter II, we first need to know that N maps
birationally to c*(N), which is part (iv) of the lemma, and secondly that c*(N)
is normal, which follows by part (i). Thus we conclude that ffL(W:i(C)) is
reduced and supported on

c*(N) = U ,u0 (W ® H 0 (C, KL - 1 )).L.

WeG(r+ 1,HD(C,L))

That it is Cohen-Macaulay and normal follows from part (i) of the lemma,
while parts (ii) and (iii) of the theorem are just translations of the correspond-
ing parts of the lemma.

Proof of the Lemma. We begin by noticing that the injectivity hypothesis on

¢implies that I is a rank (e - bw) vector bundle over G(w, A), and hence is
smooth of dimension w(a - w) + e - bw. We let J be the subvariety of E*
whose ideal is generated by the (a- w + 1) x (a- w + 1) minors of(xiy).
To say that y belongs to J means that the matrix (xiY/Y)) has rank at most
L
a- w, that is, that the space of vectors (~ 1 , ... , ~a) such that ~iXiY/Y) = 0
has dimension at least w. This amounts to saying that there is a w-dimensional
vector subspace W of A such that y annihilates </J(W (8) B), that is, that y
belongs to p(I). Thus the support of J is p(I). Now the dimension of p(I), and
hence of J, does not exceed

dim I = e - w(b + w - a).

On the other hand, since J is determinantal and its "expected" dimension is
also equal to
e - (b - (a - w))(a - (a - w)) = dim I,
it turns out that J is a determinantal variety of the "correct" dimension, and
hence, in particular, is Cohen-Macaulay. Thus, to prove parts (i) and (ii) of
the lemma it will be enough to show that J is generically reduced and normal.
By Serre's criterion (cf. Proposition (5.4) in Chapter II) this will follow if we
can show that J is smooth outside a codimension two subset. We define a
map
IX: E* ~ M

into the variety of b x a matrices by letting 1X(y) be the transpose ofthe matrix
(xiyj(y)). Thus J is the pull-back, via IX, of the variety Ma-w of all b x a
matrices of rank at most a - w. We also let Ma-w be the canonicaldesingular-
ization of Ma-w defined in Chapter II, Section 2. In other words, Ma-w is the
variety of couples ((mii), V), where mii is a b x a matrix and Vis a w-dimen-
sional vector subspace of ca satisfying the "equations"
244 VI. The Geometric Theory of Riemann's Theta Function

for every i and every (e1o ••• , ea) in V. It is clear from the definitions of I and rx
that I is just the fiber product of Ma-w and E* over M. Since I is smooth and
since Ma-w and Ma-w are isomorphic off Ma-w- 1, J must be smooth off
rx- 1(Ma-w- 1). But now, to say that y belongs to rx- 1(Ma-w- 1) means that
there is a vector subspace W' of A of dimension strictly greater than w such
that y annihilates lj>(W' ® B). Thus p - 1 (y) has dimension at least w, so, if the
codimension of rx- 1(Ma-w- 1) in J is not two or more,

dim I ::2:: dim rx- 1(Ma-w-1) +W

;:::: dim I.

This implies that J = rx- 1(Ma-w- 1); but then the above inequalities cannot
hold, since w ;:::: 1 in any case. The only way out is for rx- 1 (Ma-w- 1 ) to have
codimension two or more. This finishes the proof of (i) and (ii).
To prove (iv), suppose two points of I project to the same one in p(I). In
other words suppose there are yin E* and distinct elements W, W' of G(w, A)
such that

y l_ 1/>(W ®B),
y l_ 1/>(W' ®B).

Thus y annihilates 1/>((W + W') ®B), and therefore belongs to rx- 1(Ma-w- 1).
Since we just showed that rx- 1(Ma-w- 1) is strictly smaller than p(I), this
proves that I maps birationally onto p(I).
As for part (iii) of the lemma, it follows from formula (5.1) in Chapter II
giving the degree of a generic determinantal variety and from the remark that
p(I) is the pull-back of Ma-w via a linear map. This concludes the proof the
lemma and of Theorem (2.1 ).

The following example is a nice illustration of the Riemann singularity

theorem for a g;_
1 . Suppose that C c IP> 2 is a smooth plane quintic curve.
Then the genus of C is g = 6 and C is non-hyperelliptic. The linear series
Imc(1) Icut out on C by the lines in IP> 2 is a g;_ 1 ; accordingly, we expect to find
a triple point on e.
In fact we may see that there is a unique triple point, as follows. By the
Riemann singularity theorem any triple point on 0 comes from a g; on the
curve. Since, by Clifford's theorem, there is no g~ on the curve, the g; must be
free of base points. The holomorphic map

g;
given by the will map onto a plane curve C' of degree d' = 5/f> where fJ is the
degree of C --* C'. It follows that fJ = 1, C' is biholomorphic to C, and then,
g;
as was shown in Chapter V, that the is 1@c(1) 1.
§3. The Torelli Theorem 245

To describe the tangent cone to e at the triple point we first make a

remark concerning the Veronese surface 3, defined to be the image of the
standard mapping
cj;I'J<Zl: !Pz --+ IPs.

It is a smooth surface of degree 4, whose chordal and tangential varieties we

shall denote by a(S) and r(S), respectively. By an obvious dimension count,
a(S) is supposed to fill out IP 5 • The Veronese surface has the remarkable
property of being the only smooth and non-degenerate surface for which
a( B) =F !P 5 . For the Veronese surface the varieties a(B) and r(S) coincide, and
the tangential variety r(S) is a cubic hypersurface in IP 5 that is singular along

The canonical bundle of C is (9c(2), and the canonical mapping on the

curve extends to IP 2 to give a diagram

Thus the canonical curve lies on the Veronese surface. For each A. E IP 2 * we
denote by 1A the corresponding line in IP 2 , and by D;. the intersection 1A • C.
Then the divisor DA maps to five points on the conic cj; 19<2j(IA), and hence

dim cj;K(D A) = 2

(as must be the case, by the geometric Riemann-Roch theorem). In particular,

each cj;K(DA) lies in the chordal variety to the Veronese surface, and since both
the tangent cone to e at u(g;) and the chordal-tangential variety are cubic
hypersurfaces in IP 5 we conclude that the projectivized tangent cone to the
theta divisor at the g; on a smooth plane quintic is the tangential variety to the
Veronese sw:face.
In Exercise B-2 we will return to a discussion of this example.

§3. The Torelli Theorem

To each smooth curve C of genus g there is attached the abelian variety J(C)
together with the principal polarization given by the theta divisor e. It is a
fundamental result, due to Torelli, that the curve C can actually be recon-
structed from the abstract principally polarized abelian variety (J(C), 8).
This can be formally stated as follows.

Torelli's Theorem. Let C, C' be smooth curves. Let (J( C), 8) and (J( C'), 8') be
the corresponding Jacobians. Assume there is an isomorphism of principally
246 VI. The Geometric Theory of Riemann's Theta Function

polarized abelian varieties between the Jacobians ofC and C'. Then C and C' are
isomorphic.

The proof that we shall present below is due to Andreotti. Of course, when
the genus of C and C' is zero or one there is nothing to be proved. Accordingly,
we shall assume C and C' to have genus at least 2 throughout. We denote by
E>reg the set of smooth points of E>. Consider the Gauss map

y: ereg -+ (IP'g-1 )*
that assigns to each point u E E>reg the translate to the origin of the pro-
jectivized tangent hyperplane IP'T.,(E>). Let r be the closure of the graph of y
and denote by f its normalization. Let

cc f-+ (IP>9 - 1 )*

be the induced morphism. Clearly IX is completely determined by (J(C), E>).

We are going to show, first of all, that IX is finite. Next, denoting by Ba the
branch locus of IX, we shall prove that:
(i) If C is not hyperelliptic,

where C* stands for the dual hypersurface to the canonical curve

C c po-1.
(ii) If C is hyperelliptic,

where C0 is the rational normal curve which is the image ofC under the
canonical mapping, C~ is its dual, and pf, ... , p~g+ 2 are the stars of
hyperplanes in IP'9 - 1 through the branch points p 1, ... , Pzu+z of the
canonical map.

We shall then complete the proof ofTorelli's theorem by indicating how C can
be reconstructed from C* or C~ u pf u · · · u p~9 +z·
Since the Gauss map remains unchanged if the theta divisor is replaced by
a translate, we are allowed to identify E> with Wg_ 1(C). Accordingly, we shall
consider the theta divisor as consisting of all abelian sums u(D), where D varies
over C9 _ 1. The next remark to be made is the following. Let u be a point in
E>reg· Then u is of the form u(D) for a unique Din C9 _ 1, and the projectivized
tangent space toE> at u is the linear span of cPK(D), so that we have

y(u(D)) = (IJK(D).
Let us now assume that C is not hyperelliptic. Then

r = {(u(D), H): 15 c H} c ex (IP' 9 - 1)*

§3. The Torelli Theorem 247

(here, as usual, we are writing 15 for ¢K(D)). To see this it suffices to notice
that, by the general position theorem, any point in r is the limit of points
(u(D), y(u(D))), where u(D) is a smooth point ofE>.It is clear from the explicit
description of r that the projection from r to (IP 9 - 1 )* is a finite map;
therefore a is also finite. Now set

r = {(D,H):15 c H} c C9 _ 1 x (IP 9 - 1 )*,

and let a' be the projection from r' to (IP 9 - 1 )*.The variety r' is irreducible and
the map (u, 1): r' ~ r is finite and generically one-to-one. By the very
definition of normalization, we thus get a commutative diagram

The map a' is clearly a finite map of degree c: =~).and its branch locus is
the locus of tangent hyperplanes to C, that is, the dual hypersurface C*; thus
B, c C*. On the other hand, the ramification locus of a', which we shall
denote by R, is the set of couples (D, H), where 15 c Hand D contains multiple
points. Thus, if n is the projection from r' onto C9 _ 1 , n(R) is the irreducible
codimension one subvariety of C9 _ 1 consisting of all points of the form
2p + p 1 + .. · + p9 _ 3 . If Dis a point of C 9 _ ~> 15 is a hyperplane if and only if
D does not belong to C!_ 1 , hence n is biregular off n- 1 (C!_ 1 ), while it has
positive-dimensional fibers over C!_ 1 . Hence C!_ 1 has codimension at least
two in C9 _ 1 and a general point of R is a smooth point of r'. In particular, at
such a point, r' and fare isomorphic. Thus B, contains an open subset of C*.
As we already observed that C* contains B., and C* is irreducible, we
conclude that B, is equal to C*.
Let us now deal with the case when Cis hyperelliptic. We first claim that
the rational map y au from C9 _ 1 to (IP 9 - 1 )* extends to a regular map [3.
We may in fact set

We thus get a commutative diagram

248 VI. The Geometric Theory of Riemann's Theta Function

where c5(D) = (u(D), f3(D)). Since it is obvious from its definition that c5 is
finite and generically one-to-one, this allows us to identify C0 _ 1 with f and f3
with a. But then f3 is a finite map of degree 2°- 1 whose branch locus is the
locus of tangent hyperplanes to C 0 , plus the locus of hyperplanes passing
through some branch point of the canonical map.
Properties (i) and (ii), whose proof we just finished, allow us to reconstruct
C from the knowledge of (J(C), 0). It is, in fact, an elementary general local
result that the double duality

(C*)* = C

is valid (cf. Griffiths-Harris [1, page 397]), thereby establishing the

Torelli theorem in the non-hyperelliptic case. In the hyperelliptic case one
need only notice that C is entirely determined by the position of the branch
points of ¢K on C 0 •
A direct way of reconstructing C (or C0 ) from its dual is the following: for
each point p E C the locus T,( C)* of hyperplanes containing the tangent line
T,(C) is a (g - 3)-plane in (lfl> 0 - 1 )*.Conversely, if A c (lfl> 0 - 1 )*is any (g - 3)-
dimensionallinear system of hyperplanes in lfl>9 - 1 all of which are tangent to
C, then by Bertini's theorem the general one can only be tangent at a base
point p of A, so that A must be contained in -and hence equal to-T,( C)*. Of
course, it may happen that T,(C)* = Yq(C)* for some pair p # q of points on
C, but this can only happen finitely many times. In this way we have now
identified the curve C as being the normalization of the variety of (g - 3)-
planes contained inC*.
If one wishes to "picture" Andreotti's proof we recommend the case of a
non-hyperelliptic curve of genus 3. The canonical model is then a smooth
plane quartic curve, and there are two kinds of hyperplane sections-those
that are general such as I, and those that are tangent such as 10 :
§4. The Theory of Andreotti and Mayer 249

Of course, there are degenerate forms of 10 , such as hi-tangents and flex-

tangents, but these need not concern us since they vary in a lower-dimen-
sional family. The point is that, thinking of e as all sums u(p + q) where
p, q E C, the Gauss mapping

is everywhere defined by

y(u(p + q)) = pq,

where it is understood that 2p is the tangent line at p. The Gauss mapping
ramifies exactly over the dual curve C* c (IP 2 )* of tangent lines to C.
In the hyperelliptic case the picture is

pf

Here C0 is the canonical curve and B, is the reducible curve whose compon-
ents are the curve C~ consisting of the lines 10 tangent to C 0 , and the pencils
pj of lines lj through pi, where p 1 , .•• , p 8 are the branch points of the hyper-
elliptic involution.

§4. The Theory of Andreotti and Mayer

The classical Schottky problem is the one of finding explicit equations for
the locus f 9 of Jacobians in the Siegel upper-half plane £ 9 in terms of the
theta function. The problem has been solved by means of techniques that
bear no relationship to the theory of special divisors.
Among the classical approaches to the Schottky problem, which, although
so far unsuccessful, have been quite fruitful in a number of ways, there is
however one which is close to the ideas we have been developing. This is due
to Andreotti and Mayer.
250 VI. The Geometric Theory of Riemann's Theta Function

The essential tool used by Andreotti and Mayer in their beautiful paper
[1] is the heat equation

satisfied by the theta function. This will come into play later.
The starting point of their analysis is the following classical remark. Let
0 be the theta divisor in the Jacobian J(C) of a smooth genus g curve. Then
the dimension of the singular locus of 0 is given by

dim(0sing) = g - 4 if Cis not hyperelliptic,

dim(0sing) = g - 3 if C is hyperelliptic.

w: _
The proof of this fact is as follows. By Riemann's theorem, 0 is a translate of
~ _ 1 (C) and by Riemann's singularity theorem 0sing is a translate of 1 (C).
If C is hyperelliptic, then the assertion that 0sing is (g - 3)-dimensional
is a special case of the last part of Martens' theorem (Theorem (5.1) in
Chapter IV). Again by Martens' theorem, the dimension of 0sing cannot
exceed g- 4 when Cis not hyperelliptic. On the other hand, for w:_
1 (C),
the Brill-Noether number p equals g - 4. Hence, by Lemma (3.3) in Chapter
IV, the result follows, provided we can show that, for g ~ 4, w:_
1 (C) is not
empty. Of course, this is a special case of the Existence Theorem (cf. Chapter
V), but one which can be proved by an "ad hoc" argument. We shall actually
show that any effective divisor p 1 + · · · + p9 _ 3 of degree g - 3 on C is
contained in a degree g - 1 divisor D that moves in a pencil. In proving this,
we shall view C as being canonically embedded in !P0 - 1 • If the secant plane
p 1 + · · · + p0 _ 3 meets Cat a further point or is tangent to Cat one of the
Pi• then we can find a further point q such that p 1 + · · · + p0 _ 3 + q moves in
a pencil and we are done. Consequently, we may assume that we have p 1 , ••• ,
p9 _ 3 such that the secant plane

P1 + · · · + Po- 3 = IP 9 - 4

does not meet C away from the points Pi and is not tangent to Cat any of the
Pi· We claim that the linear projection

from this secant plane cannot be an embedding. For, if n were an embedding

of C as a smooth plane curve then we would have

deg n(C) = 2g - 2 - (g - 3) = g + 1,
and hence g would equal g(g - 1)/2, which is a contradiction if g ~ 4. In
this case, then, n cannot be an embedding and it follows that either there
§4. The Theory of Andreotti and Mayer 251

mustbedistinctpointsp, qsuchthatn(p) = n(q),orapointp = qwherenhas

vanishing differential. Then

D = P1 + · · · + Po-3 + P + q
moves in a pencil, since 15 is a IP>g- 3 •
We have thus seen that the theta divisor of any Jacobian is singular in
dimension g - 4 or more. Andreotti and Mayer's idea is to attempt to
characterize Jacobians by this condition. This is motivated by the fact that,
as they observe, one can write down explicit equations defining the locus
N 0 _ 4 c Yf0 corresponding to principally polarized abelian varieties whose
theta divisor is singular in dimension g - 4 or more.
The formal set-up is as follows. One may construct a universal family of
principally polarized abelian varieties n: :![- Yf0 by setting

where p is the representation of Z 20 in the automorphisms group of C 0 x Yfg

defined by

p(y)(u, Z) = (u + (I, Z)y, Z).

We then have a commutative diagram

where 11 is the quotient map (a local homeomorphism), pis the projection and
n is a proper map with the property that

where Az is the lattice generated by the columns of(/, Z). We now define, in
terms of 8, a number of analytic subvarieties of C 0 x Yf0 • We set

A = {(u, Z) E C0 x ;Yt'0 : 8(u, Z) = 0},

B = {(u,Z)EA: ::~ (u,Z) = O,foralla}

We then set r = PIB and

252 VI. The Geometric Theory of Riemann's Theta Function

Since n is proper and 11 is a local homeomorphism, p(Bk) is an analytic sub-

variety of £ 9 • Following Andreotti and Mayer we set

k ~ 0,

so that N k is the sublocus of £ 9 consisting of the points that correspond to

principally polarized abelian varieties whose theta divisor has a singular
locus of dimension at least k. Clearly,

· · · c Nk c Nk-l c · · · c N 0 •

We first want to show that N 0 , and therefore N b is a proper subvariety of £ 9 •

This is a nice and elementary consequence of the properties of the solu-
tions of the Cauchy problem for the heat equation. Suppose, in fact, that B 0
surjects onto £ 9 • Then there is an open subset U x V contained in CO x £ 9
and a section

s: V-+ U X V

such that
e(s(Z), Z) = o,
{ ae
-a (s(Z), Z) = o, Z E V, IX = 1, ... , g.
u"

By shrinking V and changing coordinates, if necessary, the theta function

transforms into a function v(u, Z) satisfying an equation

We may also assume that our change of coordinates makes s(Z) equal to
zero, so that the preceding conditions read as follows

v(O, Z) = 0,
{
av
-a co, Z) = o.
u"
But now the differential equations above give, by iteration, that all partials of
v vanish at (0, Z), which is absurd.
We shall denote by f'o the analytic Zariski closure of the Jacobian locus
/ 9 in £ 9 • We have already observed that
§4. The Theory of Andreotti and Mayer 253

The main result of Andreotti and Mayer is the following.

Andreotti-Mayer Theorem. }q is the unique irreducible component of Ng _ 4

that contains /g·

It is known that, for g ;;::: 4, the locus Ng_ 4 has other components besides
A. However the analysis of the "ramification loci" N k introduced by
Andreotti and Mayer has proved to be an essential tool in the study of the
moduli space of principally polarized abelian varieties.
The proof of Andreotti and Mayer's theorem uses the fact that /g is
irreducible of dimension at least 3g - 3. This follows, for example, from the
fact that the Teichmiiller space g; of all marked Riemann surfaces of genus g
is a bounded domain in IC 3 g- 3 and that the map from g; to Jll'g associating
to each curve its Jacobian has discrete fibers. We refer to Bers [1] for these
statements. An overview of the moduli theory of curves will also be found
in the second volume.
To prove the theorem all we have to show is that the dimension of any
component of Ng_ 4 containing :/y is at most 3g - 3. We shall need the
following.

Lemma. Let M be an irreducible component of N k. Let L\ be a component of Bk

such that p(L\) = M. Let Z 0 be a simple point of M. Let L qap(o/oZap) E TzoCM),
Then for every (u 0 , Z 0 ) E p- 1(Z 0 ) n L\ we have

Proof Set

L\' = L\ - P- 1(Msing) - L:\sing - {x E L:\: rankx(p) is not maximum}.

Clearly L:\' is open in L\ and p(L\') is dense in M. Therefore it suffices to prove

the lemma with (u 0 , Z 0 )EL:\'. Locally L:\' is parametrized by

u = u(y, t), Z = Z(t),

t EIC 5 , s =dim M, yEIC•, r = diml\'- rank pi,..··

Since L\' c Bk c B 0 we have

e(u(y, t), Z(t)) = 0,

ae
-0 (u(y, t), Z(t))
Ua
=o.
254 VI. The Geometric Theory of Riemann's Theta Function

Differentiating the first of these equations we find

"L. az
ae (u(y, t), Z(t)) - 0-
az~/i
= o.
a,{i a{i ti

Since I qap(ojoZap) E Tz 0 (M), there exist constants A.i such that q~fl =
I J..ioZap/otJ Q.E.D.

Returning to the proof of the theorem, we let M be an irreducible com-

ponent of N 9 _ 4 containing :/g. In order to show that the dimension of M
is at most 3g - 3 it suffices, by the lemma, to find a point Z 0 EM such that:
(1) Z 0 is a simple point of M;
(2) in p- 1(Z 0 ) n ,1 there are N = (g - 2)(g - 3)/2 points

such that the vectors

V·
, = (~
aza{i (u(i), Z o)) E ([;g(g+ 1 )/ 2 , i = 1, ... , N,

are linearly independent.

Andreotti and Mayer's beautiful idea is the following. Choose as Z 0 a point
corresponding to a Jacobian J(C), where C is a non-hyperelliptic smooth
curve, and take u0 >, ... , u<N> to be general points in E>sing (these have to be
double points). To show that the vectors vi above are linearly independent is
the same as showing that the quadrics

l L. ~(u(i)
Q. = {"
a{i aza{i ' z 0
)w a w /i = o} c [p>Y- 1

are linearly independent. Now use the heat equation to get

These are exactly the tangent cones toE> at the points u<i), which, as we saw in
the proof of Riemann's singularity theorem, all contain the canonical image
of C (cf. (1.4)).
Denote then by I z(E>) the linear span of the tangent cones to double points
of E> inside the vector space I 2 of all quadrics through the canonical curve.
Since, by Noether's theorem, the dimension of I 2 is N = (g - 2)(g - 3)/2,
the fact that the vectors vl> ... , vN are linearly independent, and hence the
§4. The Theory of Andreotti and Mayer 255

Andreotti-Mayer theorem, are a consequence of the following result, whose

proof is due to Mark Green.

(4.1) Theorem. For any non-hyperelliptic smooth curve C of genus g 2 4,

lz(E>) = lz.

The problem of deciding whether the above is true or not originated with
Andreotti and Mayer's paper: in it the result is proved for general C, and
that, of course, is all they needed to conclude the proof of their theorem.
In proving (4.1) we shall need the following general remark. Let C be any
smooth genus g curve, let

D = Pt + · · · + Pa-h• h2 1

be an effective divisor on C, and consider the morphism IX from C to Ca

defined by

1X(p) = hp + D.
Then IX induces isomorphisms

H 0 (C, Kc) ~ H 0 (Ca, OL),

(4.2)
H 1 (C, <Pc) ~ H 1 (Ca, <Pc)·

We may argue as follows. The map IX factors

where [3 is the quotient morphism, and

y(p) = (pt, · · ·, Pa-h• p, · · ·, p).

To prove the first part of (4.2) we have to show that the map

is an isomorphism (here, and in what follows, ( )sd stands for the invariants
under the action of the symmetric group Sa, and the isomorphism on the right
follows from Kiinneth's formula). Injectivity is obvious. As for surjectivity,
we must show that any invariant holomorphic 1-form won ca descends to
Ca. This is clear away from the diagonals: by Hartogs' theorem it suffices
256 VI. The Geometric Theory of Riemann's Theta Function

then to show that OJ descends to Cd near a general point of the diagonals. At

such a point, relative to suitable local coordinates on Cd, the morphism f3 is

If we write

we have that, for any i;;:::: 3, interchanging z 1 and z 2 leaves a; unchanged.

Therefore a; is a function of u, v, z 3 , ... , zd. On the other hand, the invariance
of OJ implies that

hence

OJ = a du + b dv + a3 dz 3 + · · ·,
where

Since a, bare holomorphic and invariant under the interchange of z 1 , z 2 ,

this shows that OJ is holomorphic on cd.
The second part of (4.2) follows from the first by Hodge theory. Alter-
natively, we may proceed as follows. As before, it is clear that

On the other hand, since f3 is finite,

The projection onto the invariants

1
f(a) = d' L a(a)
"<rESd

Hl(Cd, f9dsd = f(Hl(Cd, f9cd))

= H 1(Cd,f(f3*(f9cd)))

= H 1 (Cd, f9c).
§4. The Theory of Andreotti and Mayer 257

Another general remark that we shall need is the following. Let

¢:£-+ F

be an injective homomorphism of locally free sheaves of ranks n and n + 1

on an irreducible variety X. One may identify F with N + 1 F ® N F*, hence
one gets a homomorphism

We claim that 1/J a ¢ = 0. This is best seen by taking duals; in other words we
have to show that the composite homomorphism

vanishes. But this is clear, since

acts on F by

which is zero whenever f comes from E. The above formula also makes it
clear that, if e 1 , ... , en andfi> ... ,J, + 1 are local frames forE and F and we
write

then

where {e{} is the dual frame to {eJ and M; stands for the minor obtained by
deleting the ith column from the matrix (mhk). In particular, by Cramer's rule,
the sequence

(4.3)

is exact off the locus where (mhk) is not of maximal rank. In case this locus is of
codimension two or more, and X is normal, we get exactness everywhere, by
Hartog's theorem. This applies in particular to the Brill-Noether homo-
morphism
258 VI. The Geometric Theory of Riemann's Theta Function

for a non-hyperelliptic curve C. In fact, since the singular locus of the theta
divisor has dimension g - 4, c:-1> which is the degeneracy locus of the

c:_
Brill-Noether matrix, has codimension two in C9 _ 1 . From now on, for
simplicity, we write D instead of 1 ; we also identify u(C 9 _ 1 ) with (a
suitable translate of) the theta divisor 0 in J( C), and write L for the pull-back
of @(0) to C 9 _ 1 • By adjunction

on C9 _ 1 - D = 0 - E>sing; since D has codimension two, this isomorphism

holds in fact on all of C 9 _ 1 , again by Hartogs' theorem. With these conven-
tions, (4.3) yields an exact sequence

where .fv stands for the ideal sheaf of Din C9 _ 1 : in fact .fv is locally generated
by the maximal minors Mi of the Brill-Noether matrix.
We shall next construct a diagram

(4.4)

t
H 1(C 9 _ 0c 9 _)!.., Hom(H 0 (C 9 _ Q~g- H 1(C9 _ (()c 9 _ ) ) !_. H 1(C 9 _ 1..f0 ® L)

,J
1, 1, ), 1,

H 0c) ---'--. Sym' J( C, me) ___}___. H 0 ( C,- ® mol

commutative up to constants, where ¢, t/J are as defined above, (j is a co-

boundary map, a is defined via the identification (4.2), and his the transpose
of the cup-product map

Sym 2 H 0 (C, Kc)--+ H 0 (C, K~).

A moment of reflection will convince the reader that ¢ and

are induced by the contraction homomorphisms

eC 9 -1
® n1C9 -1 --+ m .
Cg-1'

The only maps in (4.4) that have not been defined so far arefand y. We shall
postpone the definition of y, as well as the proof that y is an isomorphism,
until the second volume. Here we shall only define f and show that the
rightmost rectangle in (4.4) commutes. In doing so we shall use the same
symbols to denote the theta function or its derivatives and their pull-backs
to C9 _ 1 • It will be clear from the context which ones we are talking about.
§4. The Theory of Andreotti and Mayer 259

We begin by choosing coordinates u 1 , ... , ug in J(C) in such a way that the

Chern class of <Y(0) is represented by

F1 L du;
- 2- 1\ du;,-
and denote by W; the pull-back of du; to C (or to Cd, when needed), and by
w{ the pull-back of dii;, viewed as a class in H 1 (C, <Yc) or H 1 (Cg_ 1 , <Yc.- J
We then set

where() is the Riemann theta function and lv stands for pull-back to D. The
meaning of the right-hand side is the following. If we view() as a section of
(9(0), that is as a collection {()a} oflocal holomorphic functions subject to the
relations

where the ~pa are transition functions for <Y(0), then, on the theta divisor,
and hence, by pull-back, on Cu_ 1 , one has

o2 ()p o2 ()a a (}()a

~ap~ = ~ +~(log~pa):;-
uU; uUi uU; uui uU; uui

a (}()a
+ ~(log
uui
~pa) :;-·
uU;

Since the pull-back of () vanishes to second order on D, this shows that

{o 2()afou; ouiiv} represents a section of L ® <Yv. The formula also shows that
[>j(w{ ® wj + wj ® wn is represented by the cocycle

Up to constants, the Chern class of <Y(0) is represented by the cocycle

and by our choice of coordinates, we conclude that, up to constants, the

Cech cocycle representing dii; is the pull-back to Cg-1 of o/ou;(log ~pa). It
follows that

[Jj(w,_ ® w'!'
I J
+ w'!'J ® w,_) =I
((}()
k -w'!'
OU; J
(}()
+ -w,_
ouj I
)
'
260 VI. The Geometric Theory of Riemann's Theta Function

where k is a constant. On the other hand

where the M; are the maximal minors of the Brill-Noether matrix (mhk)
relative to the frame {wt} in H 1 (C, (!)c)® mc9 _ 1 and any fixed frame in
E>c 9 _ 1 • To conclude commutativity, up to constants, of the rightmost rect-
angle in (4.4), we have to show that there is a constant k 1 such that

i = 1, 0 0 0' g,

where both sides are viewed as s.ections of L. To do this, notice that

ae
I 8-
U;
mji = o, j = 1, 0 0 0' g- 1.

This is the content of formula (1.3). Then, by Cramer's rule, off D we have

ae = F(-1)i-1M. i = 1, g,
a
U; " 0 0 0'

where F is a never-vanishing function. By Hartogs' theorem, F must be

constant, and the above holds on all of C9 _ 1 •
We are now ready to prove theorem (4.1). As we have said, the crucial
fact, to be proved in the second volume, is that the map y in (4.4) is an
isomorphism. This immediately implies that

Ker(f) c Im(h).

Suppose now that I 2 (E>) #- I 2 • Since Sym 2 H 1(C, (!)c) is the dual to the
vector space of quadrics in canonical space, there must be an element

Q= L %(W~ ® wj + wj ® w{)E Sym 2 H 1(C, (!)c)

i~j

which does not annihilate I 2 , but is such that

This means that f(Q) = 0. As a consequence, Q belongs to the image of h,

and hence annihilates the kernel of h*, that is I 2 , a contradiction. This ends
the proof of Theorem (4.1).
Bibliographical Notes 261

We may observe that this theorem gives another constructive proof of the
Torelli theorem, at least for curves whose canonical ideal is generated by
quadrics, namely for curves which are neither hyperelliptic, non-trigonal,
a
nor isomorphic to smooth plane quintic.
Let us finish by taking a closer look at the tangent cones to the theta
divisor at double points. As we know (cf. Kempf's singularity theorem), the
tangent cone to such a point u(D) is a quadric Qn whose equation can be
obtained as follows. Choose bases x 1 , x 2 and y 1 , y 2 for H 0 (C, (!)(D)) and
H 0 (C, K( -D)) and set wij = X;Yj· Then Qn is given by

It is clear from this that Qn = QK-D; from a different point of view, this is a
reflection of the evenness of the theta function. It is also clear that Qn has
rank at most four (we recall that the rank of a quadric is the rank of the cor-
responding symmetric matrix). When ID I = IK - D I we may choose
x 1 = Yt. x 2 = y 2 so that w 12 = w21 , and Qn has rank three. Geometrically
we know that

QD = uX = u
~E IDI ~· E IK-DI
~I c IPg-1;

this explicitly exhibits, in this particular case, the rulings in IP 9 - 3 's that exist
on any quadric of rank three or four. The two rulings are distinct, and hence
Qn has rank four, exactly when ID I # IK - D 1.

Bibliographical Notes

The original version of Riemann's singularity theorem is to be found in

Riemann [1]. Modern proofs have been provided by Mayer [1], Lewittes
[1], and Mumford (unpublished). Kempf proves his singularity theorem in
[2]. The generalization to W:i(C), which closely follows Kempf's ideas, is
adapted from Arbarello-Cornalba [1].
Proofs of Torelli's theorem abound. There is of course Torelli's original
argument [1]. Two proofs are due to Andreotti [1, 2]: his second proof is
essentially the one we have given in the text. Other proofs are due to
Comessatti [2] (for a modern version see Ciliberto [1]), Wei! [1], Matsusaka
[1], Martens [1], Saint-Donat [2] and Little [1]. As we observed in the
text, theorem (4.1), whose original proof is in Green [1], also leads to a
constructive proof of Torelli's theorem. See also exercise batches C and E.
The literature on the Schottky problem is vast. We do not even attempt
to give an overview of it, since the problem is only cursorily touched upon in
the text. Any reader interested in learning more about the subject would
262 VI. The Geometric Theory of Riemann's Theta Function

do well to consult Mumford's lectures [3], Mumford's two volumes [5, 6],
and Appendix D to Chapter VII in Mumford-Fogarty [1]. Solutions to the
Schottky problem have been provided by Arbarello-De Concini [1, 2] (using
results of Gunning [2] and Welters [1]), by Mulase [l] and Shiota [1].
Our main concern in this chapter is with Andreotti and Mayer's approach
as found in their paper [1]. It is a known fact that the Andreotti-Mayer
locus N 9 _ 4 contains other components beside the Jacobian locus if g 2:: 4.
In his paper [1], Beau ville analyzes in detail the cases g = 4, 5. His analysis is
completed, for g = 5, by Donagi [1]. A more direct proof, for the case
g = 4, has been provided by Smith-Varley [1].

Exercises

A. The Difference Map ¢d (II)

In this sequence of exercises we will retain the notations from exercise batch D of
Chapter V.

A-1. Let C be a curve of genus g and D = p 1 + · · · + pd, E = q 1 + · · · + qd general

divisors on C. Show that the projectivized image of the differential

is the join tPK( D), ¢K(E) of the subspaces ¢K(D) and </JK(E) in the projective space
of the canonical curve (as usual, we identify all tangent spaces to J(C) with C9 ).

A-2. Now suppose that Cis non-hyperelliptic and let H · tPK(C) = p 1 + · · · + p 29 _ 2 be

a general hyperplane section. Show that the osculating spaces to </JK(C) at the
points P; are in general position; i.e., for any integers a 1 , •.. , a 29 _ 2 with
L;a; = g - 1 and D = L;
a;p;,

dim tPK(D) = g - 2.

Conclude that no two distinct divisors of degree d ~ g - 1 supported on H · </JK(C)

are linearly equivalent to one another.
(Suggestion: See the uniform position theorem.)

A-3. Suppose that Cis non-hyperelliptic and d < g/2. Using the preceding two exercises
show that

is birational onto its image Va-

A-4. In case C is hyperelliptic and d < g/2, show that the mapping tPd has degree 2d
onto its image.
Exercises 263

A-5. In the non-hyperelliptic case show that the projectivized tangent cone to V1 at the
origin 0 E J(C) is

Note. You must show that c/>1 1 (0) is the diagonal d c C x Cas a scheme.

A-6. More generally, in case C does not have a g} show that the projectivized tangent
cone to Vd at 0 E J(C) is the d-secant variety to the canonical curve.

A-7. Assuming that C has no g}'s, show that

multo(Vd) = I (g - d -
i=O d-
~ - i)(~)
l l

(cf. Exercise A-3 of Chapter VIII).

A-8. Suppose that C has a single g} (necessarily complete), say JD J. Show that the fiber
c/>i 1(0) consists set-theoretically of the diagonal d c Cd X Cd together with the
set of pairs

{(E, E'): E, E' E ID I}.

Conclude that the support of the projectivized tangent cone IPffo(J-'d) contains the
d-secant variety to 4>K(C) together with the union

.:::. = U c/>K(E + E').

E,E'eJDJ

A-9. Retaining the notation of the preceding exercise, assume that

H 0 (C, K( -2D)) = 0

and show the support of IPffo(Vd) is exactly the d-secant variety to 4>K(C) together
with E.

B. Refined Torelli Theorems

Let C be a smooth curve with a special linear series g~. By a refined Torelli theorem we
shall mean a geometric construction of C and g~ from the polarized Jacobian (J(C), E>).
In particular, this means that the g:i is unique. In the following sequence of exercises
we shall give refined Torelli theorems for the following curves with g:i's:
(a) smooth plane quintics;
(b) smooth plane sextics;
(c) extremal curves of degree 3r - 1 in lfD'.
Note that case (c) includes case (a).
264 VI. The Geometric Theory of Riemann's Theta Function

In Exercises B-1 and B-2, C c IP' 2 will be a smooth quintic with hyperplane bundle
L = (Dc(l) = (Dc(D).

B-1. Using Exercise 18(iii) of Appendix A, show that set-theoretically

W~(C) = {u(D)},

W~(C) = {u(D + p- q): p, q E C}

= V1 + u(D),
where V1 =image ¢ 1 , with rjJ 1 as in exercise batches C of Chapter V and A of
Chapter VI.

B-2. Using the Riemann singularity theorem and the preceding exercise show that

(1--Yg-I (C)),;ng = V1 + u(D),

((1--Yg-l(C)),;ng)sing = {u(D)},

[p>§"u(D)((0sing)sing) = rPK(C).

The last equality implies the refined Torelli theorem for smooth plane quintics.

In Exercises B-3 through B-5, C c IP' 2 will be a smooth plane sextic (genus = 10),
L = (Dc(l) = (Dc(D) the hyperplane bundle, and S = rP~t 3 /IP' 2 ) c IP' 9 the cubic Veronese
surface. Note that Kc = (9c(3) and

B-3. Show that set theoretically

W~(C) = 0,
W~(C) = {u(D + P1 + Pz + P3): PiE C} u {u(2D - P1 - Pz - PJ): PiE C}
= (W3 (C) + u(D)) u (- W3 (C) + u(2D)),

W~(C)sing = {u(D + p 1 + p 2 + p 3 ); PiE C and p 1 , p2 , p 3 collinear}.

(Hint: Show that nine points in IP' 2 impose seven or fewer conditions on cubics
if and only if six lie on a line or all nine lie on a conic. For this prove that the
net of cubics through the nine points has a fixed component.)

B-4. Using the preceding exercise show that rPK(C) c IP' 9 is characterized by the
following construction: Denoting by 0trl the points of multiplicity ;: : : r on 0
(thus 0(2) = @sing), for V E iP' 9 set

Then
Exercises 265

B-5. Setting
~ = n IP'ffu(E>)
ue8(3)

establish the set-theoretic equality

where the right-hand side is the union of the IP 3 's in IP 9 spanned by the images
under c/J(I)Il' 2 <3 J of the lines A. in IP 2 • Show that also

~ = U IPT.(E>< 3 J),
ue(9(3))stng

where IPT is the projectivized Zariski tangent space.

Combining B-4 and B-5 we may reconstruct both cPK(C) and the g~ on C from
(J(C), 0), thereby establishing the refined Torelli theorem for plane sextics.

B-6. Finally, suppose that C c IP' is an extremal curve of degree 3r - 1 with hyperplane
bundle L = @c(1) = @c(D).
(i) Show that
Kc = @c(2).

(ii) Show that

w;_,(C) = {u(D)},
w;:::l(C) = {u(D + p- q): p, q E C}
= V1 + u(D)
(cf. Exercise B-1). As in the case r = 2, conclude that

C. Translates of l¥g_ 1 , Their Intersections, and the Torelli Theorem

The following series of exercises investigates the relationships among the subvarieties
W~ = W~(C) (we will use the abbreviated notation throughout) of the Jacobian of a
curve under sums and translates.

C-1. Let D be any divisor on a smooth curve C of genus g. Show that, for p E C a
general point,

r(D) if Dis special,

r(D + p) = { . . .
r(D) +1 tf D ts non-specml,
and
r(D - p) = r(D) - 1 if r(D) 2:: 0.
266 VI. The Geometric Theory of Riemann's Theta Function

C-2. Establish the set-theoretic equality

~ = n
EeCe-d
(W:. - u(E)), d:s;e:s;g-1.

(Hint: Use the preceding exercise to show that, if D is a divisor of degree d

with d ::;; e ::;; g - 1, then the following are equivalent:

r(D) ~ 0,
{
r(D + E) ~ 0 for all E E ce-d·)

Show also that at a smooth point of ~ equality holds scheme-theoretically,

i.e., that the intersection of the tangent spaces on the right is the tangent space
to~.

C-3. Generalize the previous exercise to the following set-theoretic equalities:

W~= n
EeC,.
(W:, + u(E) - u(F)), d::;; e::;; g- 1,
FeCe -d +r

W~= n
EeC,. - 5
(W~ + u(E) - u(F)), r:s;s:s;e-g+l.
FeCe -d +r -s

C-4. Let E be a divisor of degree zero. If

l-Yg-1 = l-Yg- 1 - u(E)

show that E ~ 0.

(Suggestion: Use C-1 to show that

h 0 (C, K(E - p 1 - · •• - Pg- 1 )) ~ 1 for all p1, ..• , Pg-t•

if and only if E ~ 0.)

C-5. Use the preceding exercise and the symmetry of E> to show that Riemann's
constant K satisfies

2K = u(K)

independently of the choice of base points.

C-6. Let p, q E C, and establish the set-theoretic equalities

l-Yg- 1 n (l-Yg-t + u(p- q)) = (l-Yg- 2 + u(p)) u (W: - u(q))

= n ((l-Yg-1 + u(p - r)) u (l-Yg-1 + u(s - q))).
r,seC
Exercises 267

The next few exercises will establish the following converse to the preceding exercise:
if C is non-hyperelliptic, then for any point u E J(C) there exist w, w' E J(C) with w,
w' #- 0, u and such that

~-! n <~-! + u) c <~-! + w) u en~-!+ w'),

if and only if
u = u(p- q)

for some p, q E C. This in turn will yield the

(*) Theorem Let C be a smooth non-hyperelliptic curve with Jacobian J(C) and theta
divisor 9. Then:
(i) there exist points v E J(C) such that

e n (9 + v) c (9 + w) u (9 + w')
for some pair {w, w'} #- {0, v} of points in J(C);
(ii) for any such v, the intersection

X= en (9 + v)
has two irreducible components X 1 and X 2 ; and
(iii) the curve u(C) is (up to±) a translate of the locus

{wEJ(C): 9 + w => X 1 }.
(As suggested in Mumford [3], this theorem has a corollary the Torelli theorem.)

We adopt the following notation: for~ = IP'V a linear system corresponding to a

linear subspace V c H 0 (C, L) and points Pt> ... , PeE C, set

V(- Pt - · · · - P.) = V n H 0 (C, L(- p 1 - · · • - p.)),

~(-Pt- · · · - P.) = IPV(-Pt- · ··- p.).

C-7. Suppose that ~ and tff are distinct g~'s on a curve C. Show that there exists points
Pt• ... , Pr+ 1 on C such that

~(-pl- · ··- Pr+l) oF 0,

tff(-pl- ···- Pr+l) = 0.
C-8. We keep the above notations and let ~. tff 1 , ... , tff m be distinct g~'s on C, with ~
base-point-free and cPrB birational. Using the uniform position property, show
that there exists points p 1 , ... , Pr+ 1 on C such that

~(-pl- ·· · - Pr+l) oF 0,
tffl-p 1 -···-p,+ 1 )=0 foralli.
268 VI. The Geometric Theory of Riemann's Theta Function

C-9. Show that the conclusion of the preceding exercise is false if we drop the assump-
tion that ~ is base-point-free, and likewise if we drop the assumption that </>!!d is
birational.
(Hint: For the latter, suppose, for example, that 7Lj2 x 7Lj2 acts on C with
quotient IP'l, and that 4>!B is the quotient map C-> IP' 1 .)

C-10. Using C-7, show that, for any divisor D of degree zero on C,

X= ~- 1 n (~_ 1 + u(D))
is contained in ~- 1 + u(E) where deg E = 0 if, and only if,

E ~ D or E ~ 0.

C-11. Using C-8 again show that if C is non-hyperelliptic, the variety X of the pre-
ceding exercise is not contained in any finite union of translates of ~- 1 other
than ~- 1 and ~- 1 + u(D) unless D ~ p - q for points p, q E C.

D. Prill's Problem

The following question was posed by David Prill:

Question Let f: C -> B be a branched covering of a curve of genus g(B) ~ 2, and for
p E B set DP = f- 1(p). Then is it possible that r(Dp) ~ 1 for all p E B?

D-1. Show that not all the divisors DP are linearly equivalent.

D-2. Show that, for general p E B, there does not exist q # p such that DP ~ Dq.
(Hint: Use Exercise C-1 of Chapter IV.)

D-3. Let G be a finite group acting on IP' 1 . Show that if G has a fixed point (i.e., a point
p E IP' 1 such that gp = p for all g E G) then it has two.

D-4. Using the above, conclude that if the mapping f: C-> B in the question is a
Galois covering, then r(Dp) # 1 for general p E B.

E. Another Proof of the Torelli Theorem

In this sequence of exercises we let C c IP'g- 1 be a canonical curve of genus g with

Jacobian J(C) and theta divisor E> c J(C). As usual we identify all the projectivized
tangent spaces to J(C) with IP'g- 1 = IP'T0 (J(C)), and for p E IP'g- 1 we set

rp = {uEE>:pEIP'T.(E>)};

the following will be a consequence of the first four exercises below.

Exercises 269

Theorem. If g ;:::: 5, then the set-theoretic equality

{p E n::w- 1 : rp is reducible} = C U {finite set}

is valid.

Corollary. A non-hyperelliptic curve of genus g;:::: 5 may be reconstructed from the data
(J(C), E>).

A variant of the argument below gives the corollary in the hyperelliptic case, and a
closer inspection of the proof will also allow the extension to the cases g = 3, 4.

E-1. Suppose that Cis hi-elliptic with n: C-+ E a degree two mapping onto an elliptic
curve E. Show that the chords n*q (q E E) all pass through a common point p E IP' 9 - 1,
pfC.
(Hint: Show that for any q, q' E E, r(n*(q + q')) = 1 and use the geometric
form of the Riemann-Roch theorem.)

E-2. Show that, conversely, if g ;:::: 5 and a point p E IP'9 - 1 - C lies on oo 1 chords to C,
then C is hi-elliptic and the chords to C through p meet C at the fibers of the hi-
elliptic pencil. In case g = 4 show that, if Cis not hi-elliptic then C lies on a quadric
cone and p is the vertex of that cone.
(Hint: Consider the projection of C from p to IP'9 - 2 .)

E-3. Let p E IP' 9 - 1 be any point and define the incidence correspondence I P c C 9 _ 1 x
(IP'g-1 )*by

IP = {(p 1 + · · · + P9 -1• H): P1 + · · · + P9 - 1 c H, pEH}.

Show that I P is reducible if and only if p lies on w 1 chords to C.

(Hint: Use the uniform position lemma.)

E-4. Complete the proof of the above theorem.

E-5. In this exercise we let C be hyperelliptic with 2-to-1 map n: C-+ IP' 1 ramified at
p 1 , .•• , p29 +z· Let (h(C) c 1P'9 - 1 be the canonical image of C and q, = cjJK(p;).
With the above notations, show that

From this you may conclude the Torelli theorem in this case also.

E-6. Show that for any p E IP' 9 - 1 and d with 2 < d < g - 1, p lies on oo d- 1 (d - 1)-
planes spanned by their intersection with a canonical curve C if and only if p E C.
(Hint: Look at the projection of C from p.)

E-7. Using the preceding exercise, show that a non-hyperelliptic curve C may be
recovered from the data (J(C), lt;,(C)) for 3 :-:;; d :-:;; g - 2 by the set-theoretic
relation
C = {p E 1Jl> 9 - 1 : dim rp = d- J}.

where rp = {u E WiC): p E IP'I;,(WiC))}

270 VI. The Geometric Theory of Riemann's Theta Function

F. Curves of Genus 5

In this sequence of exercises we will use the following notations:

C is a non-hyperelliptic curve of genus 5 identified with its canonical model

cc 1Jl> 4 ;

I...fc(2) I ~ 1Jl> 2 is the linear system of quadrics in 1Jl>4 containing C, r c Ifc(2) I

is the locus of quadrics of rank ~ 4, and 1' c r is the locus of quadrics of rank ~ 3.

F -1. Show that, if C is non-trigonal, then no quadric Q E I...fc(2) I may be singular at

a point of C. Using Bertini's theorem conclude that the general Q E I...fc(2) I is
smooth in the non-trigonal case.

F-2. On the other hand, if C is trigonal show that every quadric containing C is
singular, and that there is exactly one quadric of rank 3 containing C. (Cf. Section 3
of Chapter III.)

From these exercises we conclude that:

C is non-trigonal if and only if it lies on a smooth quadric.

From now on we will assume that C is non-trigonal.

F-3. Suppose that A c 1Jl>(H0 (1Jl> 4 , (!)(2))) is a general 2-plane. Show that the inter-
section of the quadrics Q E A is a smooth canonical curve of genus 5. Conclude
that for a general curve of genus 5, r is a smooth plane curve.

F-4.
(i) Show that the singular locus of r is exactly 1', and that the singularities of
rare ordinary nodes.
(Suggestion: Use the description of tangent cones to symmetric deter-
minantal varieties given in Exercise A-4 of Chapter II.)
(ii) Conclude that r is a plane quintic curve with no multiple components and,
in particular, r spans I...fc(2) 1.

F-5. Let DE c4 and suppose that u(D) E W!(C). Show that the quadric

and conclude that for C a non-hyperelliptic and non-trigonal curve of genus 5

cpK(C) = n 1Jl>ff,(0).
ue9sing

This is, of course, a special case of Theorem (4.2) in the text.

Exercises 271

F-6. With the preceding notations, show that the following are equivalent:

QvEr',
(!)c(2D) ~ Kc·

From this conclude that a non-hyperelliptic curve of genus 5 can have at most
10 semi-canonical pencils. In the hyperelliptic case there are 56 semi-canonical
pencils; while a trigonal curve has at most one.
(Hint: Show that a configuration of lines maximizes the number of double
points.)

F-7. Show that the map

c{J: W!(C)--> r
given by

has degree two and is branched exactly over r·. From this conclude that
(W!(C))sing = QJ- 1 (r')
= {u(D): DE q and (!)c(2D) ~ Kc}.

In the next exercise we will use the following result of Fulton-Hansen [1]:

Theorem. Let X c IP' be an irreducible subvariety of codimension d, Y an irreducible

variety of dimension d', and

f: y--> IP"

a finite map. If d' > d, then f- 1(X) is connected.

This theorem will be proved in exercise batch A of Chapter VII.

We will also use the scheme Jik c IP 9 of quadrics in IP 4 of rank ~ k. There is a double-
covering

corresponding to the two rulings of a rank 4 quadric; this covering is ramified over V3 •

F -8. Show that V4 is irreducible.

F-9. Show that, if r is smooth (which, by F-3, is the case for C a general curve of
genus 5), then W!(C) is irreducible.
(Hint: Show that the double covering in F-7 is the pull-back of

and now use the Fulton-Hansen theorem).

272 VI. The Geometric Theory of Riemann's Theta Function

Remark. This is a special case of the corollary to the Connectedness Theorem (1.4)
of Chapter V.

F-10. Using F-7 and an examination of the branching behavior of W}(C)-+ r, show
that W}(C) is irreducible if and only if r is irreducible.

For the next two exercises we recall that C is hi-elliptic if there is a 2-to-1 map
n: C -+ E onto an elliptic curve. Assuming that C has genus 5 and is hi-elliptic, there i~
an irreducible component u(E) of W}(C) corresponding to the pull-back of the gi's
onE.

F-11. Retaining the notation of F-7 show that ¢(u('J:.)) is a line component of r, and
that ¢(u('J:.)) meets the remainder of r transversely in four points corresponding
to the four semi-canonical pencils on C pulled back from E (cf. Exercise F-4).

F-12. Conversely, show that, if L c r is a line component and E-+ Lis the double
cover branched at the four points of intersection of L with r - L, then C is a
double cover of the elliptic curve E and L = ¢(u('J:.)).

If we agree to call n: C -+ E a hi-elliptic pencil, then by the preceding two exercises

there is a bijection
{lines in r} -+ {hi-elliptic pencils}.

In particular, C can have at most five hi-elliptic pencils.

F-13. Show that C has five hi-elliptic pencils if and only if it has 10 semi-canonical
pencils.

F-14. Let (A..,;) (1 :::; IX:::; 3, 1 :::; i:::; 5) be a 3 x 5 matrix with all 3 x 3 minors non-
zero. Show that the intersection of the quadrics

Q. = L: A.•• ixf
i

in IP 4 is a smooth canonical curve with 10 semi-canonical pencils, and that

conversely any such curve can be represented in this way.
(Hint: If C has 10 semi-canonical pencils then by the preceding exercise there
are five hi-elliptic pencils ni: C-+ Ei. Show that the corresponding sheet inter-
changes must all commute, and consider the representation of the group they
generate on IP 4 (the elements of this group are represented by simultaneously
diagonalizable matrices.))

In the following few exercises, we investigate the geometry of C and r in case C is a

double cover of a curve of genus 3.
Suppose now that B is a curve of genus 3, C ~ B a two-sheeted unramified map,
and n*: J(B) -+ J( C) the pull-back map. Let u 0 E Ker n* c J(B) be the divisor class of
order 2 on B associated ton (cf. Exercises 13 and 14 in Appendix B).
Exercises 273

F-15. Let

Show that

E = n*D c W!(C).

F-16. Using the adjunction formula on J(B), show that the arithmetic genus of Dis

p.(D) = 7.

Note. A count of constants indicates that D should be smooth for B outside a co-
dimension 1 subset of moduli.

F-17. Show that the map n*: D-+ E is an unramified double cover, and conclude that
the arithmetic genus

p.(E) = 4.

F-18. Show that if d, w1 are the classes of D and W1(B) in J(B), then

and using Exercises D-10 and D-11 of Chapter VIII, conclude that if e is the
class of E in J(C),

In particular, conclude that E =f. W!(C).

F-19. Using the description of the theta characteristics on B given in Exercises 47-50
of Appendix B, show tl:lat D contains 12 theta characteristics of B, and E corre-
spondingly contains 6 theta characteristics of C.

F-20. Using the preceding two exercises, show that the map 4>: W!(C)-+ r expresses
E as a double cover of a cubic curve, branched at six points. Conclude that r
consists of a cubic curve G and a conic curve H, with E the double cover of G
branched over the intersection G n H.

F-21. Now let F = lj>- 1 H be the remaining component(s) of W!(C) besides E. Show
that F spans a two-dimensional abelian subvariety A of J(C), and hence that
A= J(F).

F -22. Conclude from the preceding exercise that the Prym variety of the double cover
C ~ B is the Jacobian J(F) of the component F of W!(C) (cf. Appendix C).

F-23. Is the converse of the above analysis correct-i.e., if r consists of a cubic curve
Hand a conic curve G, is C then a double cover of a curve of genus 3? (Show that
this is the case if r is a sum of five lines.)
274 VI. The Geometric Theory of Riemann's Theta Function

We may collect some of the results of the above exercises in the following table:

canonical curve C c IP' 4 locus r c IP' 2 = IP'Jc(2) curve E>sing c J(C) with
of singular quadrics involution -1.
through C.

C is trigonal Every quadric through C esing is the union of two

is singular (i.e., all copies of C conjugate
quadrics containing the under the involution and
scroll Shave rank ::; 4.) meeting at two points
In this case r = IP' 2
<~

C non-trigonal r is a plane quintic E>,ing/- I ~ r. In

having at most nodes particular, every singular
(the nodes of rare in a quadric containing C is
1-1 correspondence with a tangent cone to esing
semi-canonical pencils on
C). In particular, r has
no multiple components

C hi-elliptic r contains a line (the line esing contains elliptic

(=>non-trigonal) components of r are in a curve components, each
1-1 correspondence with mapping 2 to 1 onto the
the elliptic pencils of lines in r
degree 2 on C)
<I>

C non-trigonal and there r contains a conic (the E>,ing = L 1 u L 2 has two

is a two-sheeted converse is not known) components of genera
unramified mapping p.(L 1 ) = 2, p.(L 2 ) = 4
C -+ C' where C' has meeting at six points
genus 3

C non-trigonal and with r is a smooth plane esing is a smooth

no semi-canonical pencils quintic irreducible curve of
genus 11
(=> C not hi-elliptic)
<~

A general question of some interest is: For what g can we find a linear system on an
algebraic surface that includes the general curve of genus g? This is elementary to do
for g ::; 6, g # 5. The following sequence of exercises will sketch a proof of such a
theorem when g = 5. The essential step is the following
Exercises 275

Theorem. A general curve of genus five may be represented as a plane sextic having two
tacnodes and one ordinary node.

F -24. Suppose that r is smooth and let I c IP' 2 be a line cutting out a divisor D = 1· r
where

D= P1 + · · · + Ps E i@r(l)l
with the Pi distinct. Show that there is a birational plane model C 0 of C where
C 0 is a sextic having five nodes ri with the following property: The gl's on C 0
cut out by the lines through ri correspond to the points Pi E r (recall that each
point of r corresponds to two gl's on C).
(Hint: Look at the surface obtained by intersecting the pencils of quadrics
corresponding to I c IP' 2 = j..fc(2) 1. For this it may be helpful to recall the
representation of the transverse intersection of two quadrics in IP' 4 as IP' 2 blown
up at five points (cf. Griffiths-Harris [1])).

F-25. Suppose now that I becomes tangent tor with pi, Pi coming together. Show that
the nodes ri, ri come together to form a tacnode.

F-26. Prove the theorem above by letting I become an ordinary bitangent tor. (Here,
you must show that a general plane quintic r has at least one ordinary bitangent-
in fact, this is true whenever r is smooth (why?).)

F-27. Using the above theorem, show that there exists a linear system of plane quintics
containing a general curve of genus five.

Note. We don't know if this is possible for any g with 7 ~ g ~ 22 (for g ;;:: 23,
cf. Chapter XIII).

G. Accola's Theorem

In the following sequence of exercises, we will prove a theorem of Accola's, relating

three semi-canonical pencils on a curve of genus 5 and the quadrics Q c IP' 4 that they
determine. Specifically, let C c IP' 4 be a canonical curve, Q1, Q2 , Q3 distinct quadrics of
rank 3 containing C, Ai a 2-plane of the ruling of Qi, and Di = Ai · C so that IDd is a
semi-canonical pencil. We will prove the

Theorem. (a) if Q1 , Q2 , Q3 are dependent, C is a two-sheeted cover of an elliptic curve E,

the pencils IDil are pull-backs from E, and correspondingly

(b) If conversely the Qi are independent, then r(2K - D 1 - D2 - D 3 ) = 0.

The conclusion of (b) means that if A.: J 2 x J 2 -+ 71./2 is the Wei! pairing (cf. Appendix B),
A.(D 2 - D 1 , D 3 - D 1 ) = 1.
276 VI. The Geometric Theory of Riemann's Theta Function

G-1. (i) Show that Cis not trigonal (cf. Exercise F-2 above).
(ii) Show that, if Q1 , Q2 , Q3 E IP' 2 ~ l..fc(2) I are not linearly independent, then
Cis hi-elliptic (cf. Exercise F-11).
Now deduce part (a) of the theorem.

G-2. Suppose now that Q1 , Q2 , Q3 are independent. Show that Q1 is the unique quadric
containing C and A1 , i.e., no linear combination of Q2 , Q3 contains A 1 .
(Hint: If Q2 contained A1 , then Q 1 n Q2 n Q 3 would contain a conic.)

G-3. Show that, fori =F j, Ai meets A; in one point Pu·

(Hint: (!)(Di +D) =F K.)

G-4. Let Di = C · Ai be as above. Using G-2, show that the six points of the divisor
D + p 12 + p 13 do not lie on a conic in A1 .
(Hint: Show Q 2 , Q3 cut out distinct conics on A 1 .)

G-5. Show that no quadric in IP' 4 contains A 1 u A2 u A3 •

(Hint: Show that such a quadric would be singular at Ai n Ai.)

G-6. From the preceding two exercises show that the 12 points D 1 u D 2 u D 3 impose
independent conditions on quadrics. Conclude that

and from this deduce Accola's theorem.

Here is a variant of Accola's theorem: ¢, rare as in exercise batch F.

G-7. Suppose that D 1 , D 2 , D 3 , D 4 are semi-canonical pencils on a canonical curve

C c IP' 4 , and set Pi = rjJ(u(D,)) E r. Show that the points Pi are collinear if and
I,
only if DiE 12Kcl·

H. The Difference Map ¢d (III)

We retain the notations from exercise batches D of Chapter V and A of Chapter VI.
This sequence of exercises will give a proof of the following:

Theorem. Let C be a curve of genus g and u E J(C). If d < g- 1 and

dim r/Ji 1 (u);::: d- 1,

then either:

(i) u = u(p - q) where p, q E C;

(ii) C is hi-elliptic with 1t: C --> E the two-sheeted map to an elliptic curve and
u E n*J(E); or
(iii) C has genus 4 and u is the difference of the two grs.
Exercises 277

H-1. Suppose that dim cf>i 1 (u) = k. Show that, for some e ~ 0, the fiber 4>i-1.(u) has
dimension k - e and the general point (D, E) E cf>i.!.(u) consists of a pair of
divisors with disjoint support.

H-2. Suppose now that dim cf>i 1(u) = k, and that for (D, E) a general point in cf>i 1 (u)
the divisors D, E have disjoint support. Show that
r(D +E)~ k.

H-3. Combining the preceding exercises conclude that

dim c/>i 1 (u) :::;; d, d:::;; g- 1,

with equality if and only if u = 0.

H-4. Let ul, u2 and w: cd X cd--+ J(C) be defined by

u 1 (D, E)= u(D), u 2 (D, E) = u(E),

w(D, E) = u(D + E).

Show that for any u E J(C),
dim u 1 (c/>i\u)) =dim uicf>i 1(u))
= dim w(cf>i 1 (u));

call this numbers.

H-5. Now suppose that for some u E J(C) and d :::;; g - 2, we have

dim c/>i 1 (u) = d- 1;

let (D, E) be a general point of cf>i 1 (u). Show that

r(D) + r(E) + s ~ d - 1.

H-6. With the hypotheses of the preceding problem conclude that either:

(i) C is hyperelliptic, with ID I and IE I multiples of the hyperelliptic gi plus base

points and u = u(p - q) for some p, q E C; or
(ii) C is not hyperelliptic, in which case
d-1-s
r = r(D) = r(E) = .
2

H-7. In case (ii) above, show, by the Martens-Mumford theorem, that if r ~ 1 then
conditions (ii) or (iii) of the above theorem hold. Show, on the other hand, that
ifr = 0 then
dim Wti 1 (C) ~ d- 1
and conclude that either d = 1-i.e., condition (i) of the theorem holds; d = 2;
or d = 3 and g = 5. The latter two cases will be dealt with in the following exercises
to conclude the proof of the theorem.
278 VI. The Geometric Theory of Riemann's Theta Function

H-8. Suppose now that d = 2; let {(D., E.)}.. 8 = </Ji 1 (u). Show that if for general A.,
ID.+ pi is a g~ for some p then the same is true of E• (for a different gD, and we
must have condition (iii) of the theorem.
(Hint: Use Exercise D-2 of Chapter IV).
Assuming this is not the case, then, show in case g 2 5 that for general p E C
there is a unique divisor D• (resp. E •. ) containing p, and then that D• =E •. ;
conclude that the divisors D •, E • are all fibers of an elliptic penciL
(Hint: Consider the diagram

where

rx(J.l, A.) = Dll + E. E c4,

{ ~ = image of rx,
u is the Abelian sum mapping.

Show that:
(i) rx cannot be birational (i.e., B x B cannot be fibered by IP 1 's);
(ii) D" = EA!Il) for some map A.: B--+ B;
(iii) a general pair of points of B moves in a penciL
From this conclude the result)

H-9. To remove the final possibility mentioned in H-7 above, suppose now that r = 0,
d = 3, g = 5, and s = 2. Conclude that

w(<f>i 1 (u)) = u(Kc) - W2 (C),

and hence that for general p, q E C there exists divisors D, E having disjoint support
such that
Kc = (9(D + E + p + q) and u = u(D - E).

Use the uniform position lemma (monodromy statement) to derive a contradiction.

A similar argument will dispose of the remaining case g = 4 in the preceding
exercise.

I. Geometry of the Abelian Sum Map u in Low Genera

I-1. Let C be a curve of genus 2. Show that the map

expresses C 2 as the blow-up of Pic 2 (C) at the point u(Kc).

Exercises 279

1-2. Let C be a hyperelliptic curve of genus 3. Show that W2 (C) c Pic 2 (C) has an
ordinary double point at u(g1).
(Note: You could also show that the graph of the hyperelliptic involution in
C2 is a rational curve of self-intersection - 2.)

1-3. Let C be any curve of genus 3. Show that the map

u: C 3 -+ Pic\C)

expresses C 3 as the blow-up of Pic 3 (C) along the curve W~(C) = u(Kc)- W1 (C).

1-4. Let C be a non-hyperelliptic curve of genus 4. If C has two distinct grs, show that
the map

blows down these grs to ordinary double points of W3 (C). What happens if C
instead has just one, semi-canonical g~?

1-5 (Suggested by A. Collino). Let C be a curve of genus 4 with two grs, and let W3
be the blow-up of W3 (C) at the nodes of W3 (C) coming from the two grs. Show
that If; can be realized as the variety

and that the projection n: W3 -+ W3 (C) factors through the projection n 1 : W3 -+ C 3

on the first factor: n = u o n 1 •
(Note: If we blow down opposite rulings on these two quadrics, then we get a
smooth non-projective three-fold. This follows from the fact that the two g~'s are
homologous in C 3 for a general curve C, a fact which in turn follows by computing
the intersection numbers of the two grs with the classes x and (} in Chapter VII
below.)

1-6. Let C now be a trigonal curve of genus 5. Show that the projectivized tangent cone
IP.r,.<gJ/W3 (C)) to W3 (C) at its singular point is the rational normal scroll containing
the canonical curve C c IP 4 .

1-7. Let C be an arbitrary curve of genus 5, ID I a gl on C. Show that the projectivized

tangent cone IP.r,.<Dl(W4 (C)) is a singular quadric, with vertex the projectivized
Zariski tangent space to Wi(C).

1-8. Now let C be a general curve of genus 6, ID 1 1, ... , ID 5 1 its gl's. Show that the
projectivized tangent cone IP.r,.<Dil WiC) is a three-dimensional rational normal
scroll X; containing the canonical curve C c IP 5 ; and that the intersectionS= X; n
280 VI. The Geometric Theory of Riemann's Theta Function

is a smooth del Pezzo surface (surface of degree 5) in IPs. More generally, show that
if C has between two and five g_l's, the intersectionS = n X; of the corresponding
scrolls is a surface of degree 5 in IPs, singular if and only if C has fewer than five
g_l's (cf. Arbarello-Harris [1]).

1-9. Let C be a hi-elliptic curve of genus 6 or more. Show that the intersection of the
projectivized tangent cones to W4 (C) at points of W_l(C) is a cone over an elliptic
normal curve. What is the situation in genus 5?
Appendix B

Theta Characteristics

In the following series of exercises we will develop the algebraic theory of

theta characteristics on smooth curves (for the case of singular curves, cf.
Harris [ 4]).

§1. Norm Maps

If n: C'--+ C is any branched cover of Riemann surfaces, fa meromorphic

function on C', we may define the meromorphic function Nm,J (or just
Nmf) by

(Nm,J)(p) = TI f(q)v(q)'
qen- 1 (p)

where v(q) is the multiplicity with which q appears in the fiber of n over p.

1. Show that Nm"f is indeed a well-defined meromorphic function on C,

and that the map

Nm": .A(C')* --+.A( C)*

is a (multiplicative) group homomorphism.

2. Suppose that C' ~ C is a Galois extension, i.e., that the group of deck
transformations of C' acts simply transitively on the sheets. Show that the
extension of function fields .A( C) --+.A( C') is a Galois extension, and that

Nm" = Nm.AI(C')/AI(C):.A(C')* --+.A(C)*

is the algebraic norm.

282 Appendix B. Theta Characteristics

3. Letting n: C'--+ C again be an arbitrary cover, we may define a map

Nm" = n*: Div(C')--+ Div(C),

by
Nm('f. q;) = I, n(q;).
Show that for any function f E .$/(C'),

Nm((f)) = (Nnif)

and conclude that Nm induces a map

Nm": J(C)--+ J(C').

Show that N m" o n*: J( C) --+ J( C) is just multiplication by n = deg n.

We now want to prove the following assertion:

(*) If C' ~ Cis any branched cover of Riemann surfaces, the map

is surjective.

To do this, we make one auxiliary definition: We say that a field K is

quasi-algebraically closed if every polynomial F(X 1 ,
••. , Xn) of degree din
n variables over K has a root in K whenever n > d.

4. Show that .$/(lfll 1 ) = IC(t) is quasi-algebraically closed.

(Hint: If FE IC[t] [X 1 · · · X nJ, assume the solution is a collection of
polynomials of degree m ~ 0, and solve for their coefficients.)

5. Show that a finite extension of a quasi-algebraically closed field is again

quasi-algebraically closed.

6. If Cis any Riemann surface, show that .$/(C) is quasi-algebraically closed.

7. Use the preceding exercise to conclude statement(*).

(Note: It suffices to prove this for Galois covers.)

§2. The Weil Pairing

In this sequence of exercises we will use the following notations:
J 2 = J 2 ( C) is the group of points of order 2 on J( C), sometimes thought
of as the line bundles L such that L ® L = (1)c;
§2. The Wei! Pairing 283

Since J(C) = H 0 (C, K)*/H 1 (C, Z) we also make the identification

for f a meromorphic function on C and D a divisor of degree 0 with

support disjoint from (f), we define

J(D) = [1 J(p)multp(D)•
peC

Given meromorphic functions f, g on C whose divisors D, E have disjoint

support, the W eil reciprocity law is

f(E) = g(D).

8. Verify the Weil reciprocity law in case C = IP' 1 by writing/, g explicitly in

terms of a euclidean coordinate on IP' 1 .

9. If C' ~ Cis a branched covering of curves, and the norm maps Nm = Nm,
are as defined above, show that for any functions f on C', g on C and divisors
D on C', Eon C,

(Nmf)(E) = f(n* E),

g(NmD) = (n*g)(D).

Conclude, in particular, that

f((n*g)) = (Nmf)(g).

10. Using the previous two exercises, prove the Weil reciprocity law for a
pair of functions f, g on any curve C by representing f as the pull-back of a
function on IP' 1 .

Note. For a direct proof of the Weil reciprocity law using the residue
theorem, see Griffiths-Harris [1].

11. Suppose that '7, w E J 2 and let D, E be divisors with disjoint support
representing '7, w, respectively. Write

2D =(f), 2E =(g),

and show that

f(E)
g(D) = ± 1.
284 Appendix B. Theta Characteristics

12. Retaining the notation from the preceding exercise define the W eil
pairing

by

1 f(E)
A.(f/, w) = 11 log - () .
ny -1 g D

Show that this is well defined.

13. Given 17 E J 2 represented by L E Pic 0 (C) with L 2 ~ (!Jc, define the double
cover

n: c~--+ c
by

C~ = {sEL:s 2 = 1},

where 1 is the constant section of (!Jc (alternatively, if D is a divisor repre-

senting 17 with 2D = (f), then C~ is isomorphic to the normalization of the
curve {(x, y) E C x IP 1 : y 2 = f(x)}). Check that this double cover is smooth
and unramified.

14. Show that the kernel of

is {0, 17 }.
(Suggestion: Consider the mapping induced on the first homology
groups by

n*: J(C)--+ J(C~)

by making the identification

and similarly for C~.)

15. If r: C--+ C denotes the sheet-interchange involution for the double

cover constructed above, show that there exist meromorphic functions h on
C such that
h(q) = -h(rq).
§2. The Wei! Pairing 285

(Suggestion: For any effective divisor Don C invariant under r, consider

the representation ofr on H 0 (C, (()(kD)) for large k.)

16. If r is as in the preceding exercise and h is any meromorphic function on

C with h = -h or, show that (Nmh) is divisible by 2; i.e., that we can write

(Nmh) = 2D,

and that then

where

17. Show that the image of the map

1 - r: J(C)--+ J(C),

u f--+ u- ru,

is an abelian subvariety of dimension g - 1 in J(C).

18. Show that the kernel of the map

Nm: J(C)--+ J(C)

consists of a finite union of cosets of the image of 1 - r by points of order 2

on J 2 (C) (i.e., that Im(1- r) = Prym(C/C); cf. Appendix C).
(Hint: Use the fact that n*(NmD) = D + rD.)

19. Using the identifications Jz(C) = H 1 (C,7l./2) and J 2 (C) = H 1 (C,7l./2),

show that

order((Ker Nm) n J z(C)) = 22 o+ 1,

and conclude that Ker(Nm) has exactly two connected components.

20. Show that every divisor Eon C with NmE "' 0 on Cis linearly equivalent
on C to D - rD for some DE Div(C).
(Hint: Write NmE =(f) and (using (*)) f = Nmg; and consider the
divisor E + (g).)
286 Appendix B. Theta Characteristics

21. Let

be defined by

</liD)= D- rD.

Show, using the preceding exercise, that the images of </Jd and </Jd' are disjoint
if and only if d + d' is odd (i.e., that for k large, the images of </J 2k and </J 2k+ 1
are the connected components of Ker(Nm)).

22. Prove the preceding exercise directly, arguing that if D is a divisor on C

such that

D- rD =(f),

then for any g E ~(C) such that g = -go r we have,

1 = f((g)) = g(rD)
g((f)) g(D)
so that
deg D =0 mod(2).

23. Let 7!; cq -> c be the double cover associated to YJ E J i C), and for
wE J 2 (C) write
n*E ~ D- rD,
where E represents w (this uses Exercise 20 above). Show that the Weil
pairing is
A.(ry, w) = deg D mod(2)
24. Consider the identifications

H 1 (C, 71./2)

!\
J 2 (C) ~ {unramified double covers of C},

where IX is the obvious map, f3 is the map in Exercise 13, andy associates to
any class YJ E H 1(C, 71./2) the double cover corresponding to the subgroup
YJ.L in H 1 ( C, 71./2) with respect to the intersection pairing. Show that these
identifications comiT''lte.
§3. Theta Characteristics 287

25. Using the preceding exercise, show that via the identification ex the Weil
pairing on J 2 corresponds to intersection of cycles on H 1 ( C, 71./2).

26. Suppose thatf: C-> IP' 1 is a smooth hyperelliptic curve with ramification
divisor R = p 1 + · · · + p 29 + 2 and branch divisor B = x 1 + · · · + x 29 + 2
where X; = f(p;). Show that every 11 E J z(C) can be represented by a divisor

where the ia and j p are distinct.

(Hint: Cis the Riemann surface of y 2 = Tit~i 2 (x- x;).)

27. Keep the notations of the preceding problem and let ID I be the hyper-
elliptic gi. Then,

E "' kD - Pz, - · · · - Pzzk'

Show that any two points 17, w E J 2 ( C) may be represented by divisors of the
form in the preceding exercise with disjoint support, and use this representa-
tion to compute }.,(17, w).

28. Let }.,ii be an arc in IP' 1 joining X; to X; (and missing the other branch
points), Yii its inverse image in C, and 11ii the class of Yii in H 1 (C, 71./2). Show
that the intersection number mod 2 of 11ii and 11kz is

.. . = {0 if {i,j} n {k, I} = 0 or {i,j},

(11 '1 11 kz) 1 otherwise.

29. Show that 11ii is represented by the divisor

E;i = P;- Pi
- D- P;- Pi-
30. Comparing the above with Exercise 27, verify directly that the Weil
pairing corresponds to the intersection pairing on cycles mod(2).

Remark. By continuity plus irreducibility of the moduli space of curves

of genus g this implies the same assertion for any curve of genus g.

§3. Theta Characteristics

Definition. A theta characteristic on a smooth curve C is a line bundle L
with L ® L ~ Kc.

31. If C has genus g, show that there are 229 theta characteristics (cf. the
definition of J z( C) given above).
288 Appendix B. Theta Characteristics

32. Suppose that C is hyperelliptic with two-sheeted covering n: C ~ IP 1

given by the hyperelliptic series ID I and with ramification divisor
P1 + · · · + Pz 9 +z·
(i) Show that every theta characteristic on C is of the form

L = (!)c(E),

where

E = mD +Pi, +···+Pig-1-2m'

with -1 ~ m ~ (g - 1)/2, and the Pi, are distinct.

(ii) Show that the above representation is unique if m :2': 0, while if
m = - 1 there is a single relation

-D +Pi, +···+Pig +t ~ -D +Pit + · · · + Pjg+t

(This shows again that the number of theta characteristics is 2 29 .)

33. For the theta characteristic E in Exercise 32 show that

In the next three exercises we will prove the following result, whose
formulation requires the concept of a "family of line bundles over a family
of curves" that we will make precise in the second volume.
Intuitively, a family of curves {Cr} is a map n: rt' ~ B of analytic spaces
whose fibers Cr = n- 1(t) are all smooth curves; a family of line bundles
{Lr} on {Cr} are the restrictions Lr = .Pic, of a line bundle .P on rc. Finally, a
family of theta characteristics is a family {Lr} of line bundles such that
L~ 2 ~ Kc, for all t.
With this said, our goal is to prove the

( **) Theorem. Let Lr ~ Ct be a holomorphic family of theta characteristics.

Then h 0 (Cr. Lr) is constant modulo 2.

34. Let L be a theta characteristic on C, and D = p 1 + · · · + Pd any divisor

of degree d > g - 1 ; let V be the 2d-dimensional vector space

V = H 0 (C, L(D)/L( -D)).

§3. Theta Characteristics 289

Via the identifications L 2 ~ Kc, L(D) 2 ~ Kc(2D) we may define a quadratic

form Q on V by

Q(a, r) = L Resp,(ar).
i

Show that Q is well defined and non-degenerate.

35. Let A1 c V be the subspace H 0 (C, L/L( -D)). Show that A1 is a d-

dimensional isotropic subspace for Q.

36. Let A2 c V be the image, under the restriction map, of H 0 ( C, L(D)). Show
that dim A2 = d and that A2 is an isotropic subspace for Q (use the Riemann-
Roch and residue theorems).

37. Show that

From Exercises 34 through 37 together with statement ( *) of Exercise

B-5 in Chapter II you may complete the proof of Theorem ( ** ).

The following exercise also requires concepts that will be formulated

precisely in the second volume.

38. Let g-; c .A9 denote the family of smooth curves of genus g having a
theta characteristic L with h0 (C, L) ~ r + 1. If L 0 -+ C 0 is a theta charac-
teristic with

then convince yourself that

codim g-; s r(r + 1)/2

in a neighborhood of C 0 .
(Hint: Use Exercises 34-37 and Exercise B-7 of Chapter II.)

Definition. A theta characteristic L is even or odd according to the parity of

h 0 (C, L).

From Theorem (**) it follows that the configuration of even and odd
theta characteristics is invariant under deformation of the curve, and this
290 Appendix B. Theta Characteristics

configuration will be described in the following exercises. More precisely,

keeping our notations, for each theta characteristic L ~ C we define

by

and will prove the

Riemann-Mumford Relation. For w, 11 E J 2 (C)

where A(l], w) is the Wei/ pairing.

39. Let 1'/ E J 2 ( C) and n: C~ ~ C the double covering associated to 11 and

r: c~ ~ c~ the sheet exchange involution. Show that, for any wE JiC),

(Hint: Consider the representation of ron H 0 (C~, n* L).)

40. Let E be any divisor on C representing w and write

n*E ~ D- rD

with D, rD disjoint (cf. Exercise 20). Set

V = H 0 (C~, n*L(D)/n*L) EB H 0 (C~, n*L/n*L( -rD))

= H 0 (C, L(NmD)/L) ® H 0 (C, L/L( -NmD)).

Note that V ~ C2d, where deg D = d. Using the second expression for V and
an isomorphism L® 2 ~ Kc, define a quadratic form

Q: v x v~ c
by

Q(a EB a', A.® A') =I Res(aA') +I Res(a'A.),

§3. Theta Characteristics 291

where the sum is over the points of NmD. Show that Q is well defined and
non-degenerate.

41. Denote by A 1 and A 2 the direct summands of Vand set

Show that
(i) A1 , A2 , A3 are d-dimensional isotropic subspaces for Q;
(ii)

and, using Exercise B-6 of Chapter II, that

By combining this with Exercise 23 you may complete the proof of

the Riemann-Mumford relation.

There is another version of the Riemann-Mumford relation due to

Thurston.

42. Let L ~ (9c(D) be a theta characteristic and ¢ a meromorphic 1-form

with(¢) = 2D. Let 17 E J 2 (C) ~ H 1 (C, Z/2) and

y(t): [0, 1] --> C - support D

be a smooth, non-self-intersecting loop representing 17· Show that

qL(17) =
1
r--1
f! d log(y'(t), ¢(y(t))).
rcy -1 o

(Hint: Using Exercises 32 and 33 do this explicitly for a hyperelliptic curve,

and using Theorem(**) deduce it for all C.)

43. Use Exercises 42 and 30 to deduce the Riemann-Mumford relation.

(Note. If y1 , y2 are two smooth loops, then y 1 + y2 may have self-inter-

sections and will have to be smoothed.)

In the next three exercises we will give a smattering of the classical ap-
proach to theta characteristics.
292 Appendix B. Theta Characteristics

44. Let C be a compact Riemann surface of genus g with normalized basis

A. 1 , ... , A. 29 for H 1 (C, Z), period matrix Z and associated Riemann theta
function 8(u) as defined in Chapter I. For a, bE Z 9 show that

e
a,b
-
(u)=e-",;-l(b,u>e ( u+-+-
a
2
Zb)
2

is odd or even depending on the parity of (a, b).

45. Using the preceding exercise and the Riemann singularity theorem show
that if Lis the line bundle of degree g - 1 on C with u(L) = K (=vector of
Riemann constants),

-"' ~ 'b~+i H 1(C,tZ)

w-L...a;2+L...; 2 E Hl(C,Z)

~ H1(C, tz)
~ J2(C),

then
h 0 (L ® w) =(a, b) mod(2).
Use this to conclude also the invariance of h0 (C 1 , L 1) mod(2).

We return now to the theta characteristics on a hyperelliptic curve,

where we will verify directly the Riemann-Mumford relation.

46. Let C and L be as in Exercise 32, and let Eii be the divisor representing
the homology class n;i of Exercises 28 and 29. If I = {i 1 , ... , i9 _ 1 _ 2 m}, show
that if m ~ 0,

-1 if i, j ¢:I,
h 0 (L(Eii)) = h 0 (L) +{ + 1 if i, j E 1,
0 otherwise.
Derive the corresponding expression in case m = -1.

47. Using Exercises 46 and 43 or 44, verify the Riemann-Mumford relation

for C.

§4. Quadratic Forms Over 7Lj2

Let V be a vector space over the field Z/2. By a bilinear form on V we will
mean a symmetric, bilinear pairing

A.: v X v -> Z/2

§4. Quadratic Forms Over 71./2 293

with the additional requirement that A(v, v) = 0 for all v E V (for a general
L
symmetric bilinear pairing A(x, y) = auX;Yi• the restriction of A to the
diagonal is in fact linear:

here we require that this linear form be zero). Given a quadratic form (i.e., a
homogeneous polynomial of degree two)

q: v--+ ll./2
we have the associated bilinear form

A(x, y) = q(x + y) + q(x) + q(y).

48. Show that if dim V = 2n there is a unique non-degenerate symmetric
bilinear form on V up to isomorphism, which can be written as

A(X, y) = L (X;Yn+i + Xn+iYJ

i

49. Show that there are exactly two quadratic forms having a given non-
degenerate associated bilinear form A, and that if A is as in the preceding
exercise, these can be written as

and

Counting the number of zeros of each of these, show that they are not
isomorphic.

50. Interpret the Riemann-Mumford relation above to say that qL is a

quadratic form whose associated bilinear form is the Wei! pairing. Show that
qL ::;:; q+ if Lis even, while qL ::;:; q- if Lis odd.

In the following exercises we will consider the action of monodromy on

the set of theta characteristics, and for this we will use the results of Chapter
X together with the additional notations:
s (resp. s+' s-) denotes the set of all (resp. even, odd) theta charac-
teristics;
Sp(J 2 ) is the group Sp 2 g(7l./2) of symplectic transformations mod(2),
defined by

Sp(J 2 ) = {A E GL 2g(7l./2): A(A17, Aw) = A('1, w) for 17, wE 1 2 },

294 Appendix B. Theta Characteristics

where J 2 = J 2 (C) and ll is the Weil pairing. Using the terminology from
Chapter X, we will prove the

(***) Theorem. The global monodromy group acts transitively on s+ and s-.
51. By arguments similar to those in Chapter X show that Sp(J 2 ) is generated
by Picard-Lefschetz transformations

'1 --+ '1 + ll(IJ, w)w.

Conclude that the monodromy is Sp(J 2 ) and that the monodromy action is
transitive on non-zero elements of J 2 •

52. For any theta characteristic L and A E Sp(J 2 ), show that

is linear in '1· Conclude that there exists a unique w E J 2 with

Denoting by O(S) the group of affine linear transformations of J 2 that

preserve qL, this shows that the projection map

O(S) --+ Sp(J 2 )

is an isomorphism.

53. Show that every wE J 2 with qL(w) = 0 occurs for some A in Exercise 52.
(Hint: Show that any two quadratic forms mod 2 on (7l./2) 2 " with the same
number of zeros are conjugate under Sp 2 n(7l.j2)).

From Exercises 51 through 53 you may complete the proof of Theorem

( *** ).
Appendix C

Prym Varieties

In this appendix we will give a brief account of the theory of Prym varieties.
Prym varieties are abelian varieties associated to unramified double covers
of curves; in studying them we will correspondingly have a number of
occasions to refer back to the preceding appendix.
To begin with, our basic objects are as follows: a curve C of genus g
and an unramified double cover
f: C--> C,
with involution

r: C--> C
exchanging sheets of C over C. We will also denote by r the induced in-
volution on the groups H 1 ( C, 1:'), H 0 ( C, K), and J( C); in the first two cases
we will denote the + 1- and - 1-eigenspaces of r by a superscript + or -.
To begin with, we represent the map n topologically as in the figure, and
choose normalized bases for H 1 (C, 1:') and H 1 (C, 1:') as drawn.

Note that we have:

f*ao = ao,
f*b 0 = 2b 0 ,

f*(a;) = ai + ag-1+i} .
f *(b·)' = b-.,+ Eg-1+•. 1 = 1' · · ·' g - 1.

A basis for the lattice H 1 ( C, Z)- of the anti-invariant cycles is then given by

i = 1, ... ' g - 1.

Let w 0 , w 1 , ... , w 29 _ 2 be a normalized basis for H 1 • 0 (C), i.e., one for which
296 Appendix C. Prym Varieties

It is then easy to see that a basis for H 1 • 0 (C)+ is given by

while a basis for H 1 • 0 (C)- is given by

{wi- wi+g- 1 : i = 1, ... ,g- 1}.

It is also easy to see that

is a lattice of maximal rank so that one can define a complex torus

Appendix C. Prym Varieties 297

of dimension g - 1. The intersection matrix of .the anti-invariant cycles is of

the form

and this suggests that the principal polarization on J(C) restricts to twice a
principal polarization on Prym(C, r). Equipped with this principal polariza-
tion the complex torus is called the Prym variety of the two-sheeted covering
f: C -4 C. To better understand, let us consider the three basic maps Nm,
n*, and r,

where the norm mapping is defined as in Exercise 3 of Appendix B by

NmC[, m;P;) = L mJ(p;).

Let 0 and 0 denote the theta divisors on J(C) and J(C) respectively. Recall
(Exercise 14 of Appendix B) that

Ker n* = {0, L,},

where L, is a point of order 2 in J(C). Moreover, it is easy to check that

Pr m(C r) = Im(l _ r) = {conn~ct~d com~one~t of Nm- 1 (0)}·

Y ' contammg the 1denttty

As it turns out it is more convenient to look at the norm map Nm as defined

in Pic 29 - 2 (C)

The two basic results in Mumford's approach are the following:

(a) Nm- 1 (Kc) is the disjoint union of two translates of Prym(C, r), p+,
and p-. Moreover, L E p+ if and only if NmL = Kc and h 0 (C, L)
is even.
(b) 8 · P + = 28 where 8 is a principal polarization on P +.
298 Appendix C. Prym Varieties

The first result, and especially the statement about the parity of h0 ( C, L), is
a crucial one and is based on Exercises 35-37 of Appendix B. Namely one
observes that for each L E Pic 2 g- 2 (C) such that NmL = Kc, the rank 2
vector bundle n* L is equipped with a quadratic form

Thus, by an argument analogous to the one in Exercises 34-37 of Ap-

pendix B,

is constant modulo 2 on p+ (resp. p-). From (a) and (b) it follows that

0 n p+ = {L E Pic 2 Y- 2 (C): NmL = Kc, h0 (C, L) is even and at least 2}

C <§sing'

3sing = {L E p+: multL E) is even and at least 4}

u {L E p+: multL e
= 2 and TL(P+) c 3(0)}
_,
=
-~~

~sing U ~sing·

We are now going to show that

Claim. Ifg :2: 5 and dim ssing :2: g - 5, then almost all points LEssing are of
the form L = (f*M)(I'i) where NmL = Kc, M is a line bundle on C, !'!an
effective divisor on C, h 0 (C, M) :2: 2 and h 0 (C, L) is even.

In this assertion "almost all" means a union of Zariski open subsets in

each component of maximum dimension in Ssing· To prove this we recall the
description of Zariski tangent spaces:

TL(E'>) = (lm(J1 0 : H 0 (C, L) ® H 0 (C, KcL - 1 )--> H 0 (C, Kc)))-1,

TL(P+) = CH1·occ)-)*.

Notice that since NmL = Kc we have Kc = n*NmL. On the other hand,

n*NmL = L ® r*L. It then follows that

Kc:L - 1 = r*L.

Assume now that L E s;ing so that

NmL = Kc,
h0 (C, L) = 2,
TL(P+) c §"L(E>).
Appendix C. Prym Varieties 299

By the Kempf singularity theorem, the equation in TL(J(C)) of the tangent

cone :YL(0) is given by

where {s 1 , s 2 } and {t 1 , t 2 } are bases of H 0 (C, L) and H 0 (C, KcL - 1 ), respec-

tively. By the above, we may assume that

j = 1, 2,

and we decompose the holomorphic differential s;ti in its invariant and anti-
invariant parts

It then follows that

But this in turn means that s2 /s 1 defines a meromorphic function h on C and

letting

D = (h) 0 , M = (!)c(D),

wegetL = (n*M)(A),provingourclaimincaseL E 3~ing· We can now assume

that the point L E 3sing belongs to an irreducible component Z of 3~ing' with
dim Z ~ g - 5. Suppose now that for a general L E Z

so that

It follows that
300 Appendix C. Prym Varieties

In view of the identifications of TL(e) and TL(P+) above, this translates into

where v0 (s 1 1\ s 2 ) = s 1 (r*s 2 ) - s 2 (r*s 1 ). This means that codim(Ker v0 )::;; 4.

But in A 2 H 0 (C, L) the decomposable vectors form a non-degenerate sub-
variety of dimension at least five, so that there exists a decomposable vector
s 1 1\ s 2 in Ker v0 • This means that(*) holds and therefore we can proceed
as before.

We want to apply the above discussion to the double covering

f: esing-> r
described in exercise batch F of Chapter VI, where esing is the singular locus of
the theta divisor of a general canonical curve C of genus 5 and r is a smooth
plane quintic realized as the locus of rank 4 quadrics through C. Let r be
the - 1 involution on esing. By Exercise F -7 of Chapter VI and the Riemann
singularity theorem, if Q is a general quadric through C, f- 1 (Q) is the pair
of g!'s determined by the two distinct rulings of Q.
Before restricting to this situation let us consider any two-sheeted un-
branched covering

f: r _, r
of a smooth plane quintic, and denote by r the sheet-interchange involution
on f. Let L, E J(r) be the point of order 2 determining f, and let S c
Prym(f, r) be the theta divisor of the Prym variety off (Prym(f, r) is a
principally polarized abelian variety of dimension 5). We will show that:

(**) The following conditions are equivalent:

(i) dim ssing = 1;
(ii) h0 (r, @r(1) ® L,) is even.
If(i) or (ii) holds then
(iii) Ssing = {(j*@r(1))(p- rp): p E f} U {finite set}
~ ru {finite set}.
If (i) or (ii) are not satisfied then the following equivalent conditions are met

(iv) dim ssing = 0;

(v) h0 (r, @r(1) ® L,) is odd.
If (iv) or (v) holds, then
(vi) f*@r(l) E Ssing·
Appendix C. Prym Varieties 301

The proof of ( **) is a consequence of the claim above. In fact, from the
proof of the claim we deduce that, if there exists an irreducible component Z
of 3sing with dim Z ~ g - 5 = 1 then a general point L E Z is of the form

L = (f* M)(A),
with

NmL = Kr,
{
h0 (r, M) even and at least 2.

Since

2 deg M = deg Nmf*M ~ deg Kr = 10,

we have that IM I is a g~ with d ~ 5 and r ~ 1. As follows from Exercise

18 of Appendix A, only two cases are possible: either M = (T!r(l) or
M = (T!r(1)( -p). Since (T!r(2) = Kr, a line bundle L of the form (f*M)(A)
satisfies the condition above only if L = j*(T!r(l) or L = (f*(T!r(1))(p - rp),
p E f. Now for any point p E f, h0 (f, (f*(T!r(l))(p - rp)) is even if and only
if h0 (f, j*(T!r(l)) is odd. Since

H 0 (f, j*(T!r(l)) ~ H 0 (r, (T!r(l)) EB H 0 (r, (T!r(l) ® L,),

we conclude that the following are equivalent:

(a) dim Z 2 1;
(b) a general point of Z is of the form j*(Dr(P - rp);
(c) Z = {(f*(T!r(l))(p- rp):pEf};
the equivalence of(a), (b), and (c) establishes(**).
A covering/: f ~ r is called even (resp. odd) if h0 (r, (T!r(O ® L,) is even
(resp. odd). We consider first the even case. Then Prym(f, r) is a five-
dimensional principally polarized abelian variety for which

3sing =f u {finite set};

moreover, the -1 involution on corresponds to the r involution on f.

3sing
®sing ~ r above.
It is then natural to ask if we are in the situation f:
The following theorem is the result of published and unpublished results
of Clemens, Tjurin, Masiewicki, Donagi, and Smith and it says that the
answer is yes, and in the strongest sense.

Theorem. Let C be a general canonical curve of genus 5 and r the locus of

rank 4 quadrics containing C. Then the two-sheeted unramified covering

f: esing ~ r
302 Appendix C. Prym Varieties

is even and every unramified two-sheeted covering of a smooth plane quintic

arises in this way. Moreover

(J(C), 8) ~ (Prym(8sing• -1), 3).

(In particular, the "finite set" in ssing is the empty set.)

As a corollary of this theorem and the Torelli theorem we infer that the
pairs (r, L,) where

{ r is a smooth plane quintic,

Lz ~ (l)r and h0 (r, (l)r(1) ® L,) =0 mod(2),

are in 1-1 correspondence with general genus 5 canonical curves C.

The case of odd coverings is not less harmonious. Consider a cubic
threefold, i.e., a non-singular cubic hypersurface

Such a threefold determines a five-dimensional principally polarized abelian

variety (see Clemens-Griffiths [1])

(J(X), 8),

the intermediate Jacobian of X. On the other hand, take a line

leX

and consider all 2-planes n passing through I. Any such 2-plane intersects
X along a curve of the form

n n X= I u C,

where C is a conic. These 2-planes can be viewed as points in a IP 2 and the

2-planes n for which

C = 11 u /2 ={union of2lines}

describe a smooth plane quintic r in this IP 2 • This quintic comes naturally

equipped with a two-sheeted unramified covering

{([A], [n]) E G(2, 5) X r: A c n, A =f. I} = r~r c IP 2 ,

where .h is the projection. In [1], Tjurin proves

Theorem. Let X be a cubic threefold and let

Ji:f--+r
Exercises 303

be the two-sheeted covering defined by a line l c X. Then this is an odd covering

and every odd covering arises in this way. Moreover

(J(X), 0) = (Prym(f', r), 3).

This theorem leads to a proof that the cubic threehold X is not rational, a
result that was first established in Clemens-Griffiths [1]. In fact one
first establishes the general result that the rationality of a smooth threehold
X implies that the intermediate Jacobian J(X) is a direct sum of Jacobian
varieties of smooth curves and then one uses Theorem (**) above to argue
that, since J(X) is the Prym variety of an odd two-sheeted covering of a
smooth plane quintic, 0sing is zero-dimensional so that J(X) cannot be a
direct sum of J acobians of smooth curves.

Exercises

1. Let rbeacurve of genus 2, expressed as a doublecoverof1P' 1 ramified at p 1 , .•• , p6 E r,

and let 17 = p; - Pi· If f 4 r is the unramified double cover associated to 17 (cf.
Exercise 13 of Appendix B) and r the sheet interchange, identify the elliptic curve
Prym(f, r).

2. More generally, suppose r is any hyperelliptic curve, expressed as a double cover

rr: r--> IP' 1 ramified at p 1 , ... , Pz.+z; let 17 = P;- Pi and let f--> r be the correspond-
ing double cover. Show that Prym(f, r) = J(r'), where r' is the hyperelliptic curve
given as the double cover of1P' 1 branched over {n(pk):k i= i,j}.

3. In the situation of the preceding exercise, if 17 is a point of order 2 in J(r) not of the
form P;- Pi• is Prym(f, r) a Jacobian?

4. Let r now be a canonical curve of genus 3, 17 the point of order 2 in J(r) given as the
difference of the divisors of degree 2 cut by a pair of bitangent lines. Let f ..... r be the
associated double cover and r the sheet interchange. Of what curve of genus 2 is
Prym(f, r) the Jacobian?

5. Let f ..... r be a ramified double cover, r: f ..... f the sheet interchange, and let P
be the complex torus

(the minus superscripts again refer to the (-!)-eigenspace under the action of r).
Show that P does not admit a principal polarization in general.

Note. You may want to use the fact, to be established in the second volume, that
if r is a general curve of genus g, the standard polarization of J(r), given by the inter-
section form on r, is the unique polarization of J(r) up to multiplication by scalars.
Chapter VII

The Existence and Connectedness Theorems

for w:i(C)

In this chapter we will prove some of the basic results of Brill-N oether theory,
as described in Chapter V. These results fall into two main categories: those
which apply to an arbitrary curve, and those which are true only for a
general curve. In this chapter we will be concerned with the former, leaving
the latter to the second volume.
With the theory developed thus far, the arguments in this chapter are
straightforward. In each case, we take the construction of the variety in
question (W:i(C) or C:i) as a determinantal variety, given in Chapter IV, and
analyze its particular circumstances. In the case of W:i(C), the key feature
is the ampleness of one of the two bundles involved in the construction, the
direct image of the Poincare bundle. This allows us to conclude our two
qualitative theorems, the existence and connectedness theorems. Our
"quantitative" results, the determination of the expected fundamental classes
of W:i(C) and q, follow when we apply Porteous' formula (Chapter II) to
their determinantal constructions, once we compute the Chern classes of the
bundles involved.
It should be remarked that the proofs of the results in this chapter appear
nearly in reverse historical order. Thus, the first proofs of the existence
theorem were via the computations of the classes of W:i(C) and Cd in
Kleiman-Laksov [1], Kempf [1], and Kleiman-Laksov [2]. Some 10 years
later, Fulton and Lazarsfeld [1] proved the connectedness theorem (at the
same time proving existence); finally, Lazarsfeld noted that an even simpler
proof of the existence theorem could be given.

§1. Ample Vector Bundles

Let us first fix our notation. In what follows, we will denote by E ~ X a

vector bundle E on a variety X. By the projective bundle IP E ~ X associated
to E, we will mean the bundle over X whose fiber over a point x E X is the
projective space oflines in the fiber Ex of E over x; scheme-theoretically,

IPE = Proj(~ Symn E* )·

§I. Ample Vector Bundles 305

The tautological sheaf (DIPE(l) will be the sheaf of sections of the line bundle
on lfl> E whose fiber over a point (x, ~) E lfl> E is the space of linear functionals
on the line ~ c E,; thus the space of sections of (DIPE(l) over a fiber lfl>Ex is
the dual vector space E~, and the sheaf map n*E*---+ (DIPE(l) induces an iso-
morphism between n*lDIPE(l) and E*. More generally, we see that

and that for any coherent sheaf§' on X,

(1.1)

What is the correct notion of ampleness for vector bundles? A priori

there may be several answers to this, since the direct generalizations of the
various definitions or criteria for ampleness in the case of line bundles do
not turn out to be equivalent for vector bundles. What turns out to be the
correct notion is the following. A vector bundle E is ample if the tautological
line bundle lD( 1) is ample on [JJ> E*. An equivalent definition is that E is ample
if tensoring with high symmetric powers of E kills the higher cohomology of
coherent sheaves §' on X; i.e., if for any ff, there exists an n0 such that
Hi(X, §' ® Sym" E) vanishes for all n ~ n0 and i > 0. That this vanishing
result holds for any ample E follows from (1.1) by the Leray spectral sequence
and the vanishing of Rin* (DIPE*(n) fori > 0 and n ~ 0. The opposite implica-
tion follows from the remark that for any E with the vanishing property above
and large enough n we can find a section of Sym" E assuming any pre-
assigned values at any two points of X.
We observe that if we have an exact sequence of vector bundles

0 ---+ E ---+ F ---+ G --> 0,

then F ample implies that G is ample; this follows from the fact that (DIPG*(l)
is just the restriction of <VIPF*(l) under the inclusion lfl>G* ~ lfl>F* dual to the
quotient map F---+ G. It is also the case that if E and G are both ample, then
F must be; this can be proved as follows. Filter Symk F by the images of
Sym 1 E Q9 Symk-t F, l = 0, ... , k; the successive quotients of this filtration
are the bundles Sym 1 E Q9 Symk-t G. Therefore, for any coherent sheaf§' on
X, if we can show that, for high enough k, and any I, the higher cohomology
groups of Sym 1 E Q9 Symk-t G Q9 §' vanish, this will imply that the higher
cohomology groups of Symk F Q9 §' also vanish for large k. By what we just
observed, to say that this holds for any §' is the same as saying that F is
ample. In case F is the direct sum of E and G, to say that F is ample is actually
equivalent to the vanishing of the higher cohomology groups of Sym1 E Q9
Symk-t G ® ff, for any ff, large enough k, and any/. In fact, in this particular
case, Symk F is the direct sum of the bundles Sym 1 E Q9 Symk-t G.
To prove our claim, it is thus sufficient to examine the case when F is the
direct sum of E and G. This can be done by direct computation. What has
306 VII. The Existence and Connectedness Theorems for w:;( C)

to be shown is that the sections of Symk(F) separate points of IFDF* for some
large k. In other terms, given vectors e + g, e' + g' in F* which are not
proportional, we must be able to find a section of Symk F that vanishes
at e + g but not at e' + g' (here, of course, we are taking e and e' to be in
E* and g, g' to be in G*). We first remark that, since E, G are ample, there
are sections of Symk F that do not vanish at e + g or e' + g'. If e is not
proportional to e' there are sections of Symk F that vanish at e but not ate',
since E is ample, and the same can be said of g, g', and G. The only case to
be examined is then the one when e, e' and g, g' are proportional. We can
certainly find a section P of Symk E and a section Q of Symk G such that
P(e) = Q(g) but P(e), P(e'), Q(g), Q(g') are not all zero. Then P - Q, viewed
as a section of Symk F, vanishes at e + g, but does not vanish at e' + g'
unless e, g are not zero and there are constants rx, [3 such that

(J. =f. [3, e' = rxe, g' = [Jg,

The same can be done for another large exponent k' prime to k. Thus, if
every section of Symk+k' F vanishing at e + g also vanishes at e' + g', the
above must be satisfied, and in addition one must also have that rxk' = [Jk'.
Since k and k' are relatively prime this implies that rx = [3, a contradiction.
We conclude that the sections ofSymk+k' F separate points ofiFDF*, as desired.
Finally, we note that while the condition of ampleness of line bundles on a
fixed variety is both open and closed, the condition of ampleness for vector
bundles is open but not closed (for example, the ample bundle QIP',(l) EB
QIP',(l) on IFD 1 may specialize to QIP', EB (91P',(2)). What is true is that forE a fixed
vector bundle and L a line bundle, the condition of ampleness of E ® L is
both open and closed in L, since IFDE = IFD(E ® L) for all L. In particular, if
L is algebraically equivalent to zero, i.e., if the topological first Chern class
c 1(L) E H 2 (X, Z) vanishes, then E ®Lis ample if and only if E is.
Ample vector bundles have some topological properties analogous to
those of ample line bundles. For example, the famous hyperplane theorem
of Lefschetz states that if Y is an affine variety of dimension n, i.e., if X is any
n-dimensional complete variety, L an ample line bundle on X, (J a global
section of L with zero divisor Z, and Y = X - Z, then

Hi( Y, Z) = 0 for i > n.

The corresponding statement for vector bundles is the

(1.2) Proposition. Let E-+ X be an ample vector bundle of rank r, (J a global

section of E, Z the zero locus of (J, and Y =X- Z. Then

Hi( Y, Z) = 0 for i 2 dim X + r.

§I. Ample Vector Bundles 307

Proof. Let a be the global section of @IP'E*(1) corresponding to CJ via (1.1) and
let 2 c [JJ> E* be its zero locus. Then via the projection, [JJ> E* - 2 is a bundle
of affine spaces over X - Z = Y; in particular, for all i

But by the Lefschetz hyperplane theorem applied to @IP'E*(l),

Hi(IP>E* - 2, Z) = 0 for i ~dim IP>E* + 1 =dim X+ r. Q.E.D.

One immediate consequence of this proposition is that if E ~ X is an

ample vector bundle with X projective and dim X ~ rank E, then every
global section of E must have a zero. To broaden this slightly, we may say that
if Lis a line bundle on X, E a vector bundle with L * ® E ample, and dim X ~
rank E, then every homomorphism¢: L ~ E must have rank equal to zero
somewhere. This in turn may be generalized to the

(1.3) Proposition. Suppose X is a projective variety, E, F vector bundles on

X of ranks m and n with E* ® F ample, and ¢: E ~ F any vector bundle map.
If

dim X ~ (m - k)(n - k),

then somewhere ¢ must have rank at most equal to k.

Proof. We will actually prove something more: it will be shown that if

Lk c X is any component of the locus X k of points x E X such that
rank(¢x) ~ k, then

(here codim(A, B) denotes the minimum codimension of any component

of A in B); in fact, this bound holds for the codimension of any component
ofLk n Xk_ 1 • In particular, this says that if dim Lk ~ m + n- 2k + 1 then
X k _ 1 n Lk f= 0; the result will of course follow.
To prove this statement, we may as well restrict to X = Lb and assume
rank (¢x) = k everywhere. We now shift attention to IP>E ~X, and to the
vector bundle map

given as the composition of n*¢: n* E ~ n* F with the inclusion @IP'E( -1) c

n*E. Then we observe that the bundle @IP'E(l) ® n*F is ample over IP>E. To
see this remark that the tautological bundle on IP>(@IP'E( -1) ® n*F*) =
iP> E x x [JJ> F* is just the restriction of the ample line bundle @IP'<E®F*)(1) under
308 VII. The Existence and Connectedness Theorems for W.(C)

the Segre inclusion IPE x x IPF*--+ IP(E ® F*). It follows that if Z c IPE is
the zero locus of $, then

Hi(IPE - Z, 1':) = 0 for i 2 dim X +n+m- 1.

But now if I c F is the image bundle of¢, then the induced map IP E - Z --+
IP'I expresses IP'E - Z as a bundle of affine spaces over IP'I. Thus, since the
fundamental class of IP' I is non-zero,
H2dimX+2k-2(1P£ _ z, £:) = H2dimX+2k-2(1PI, £:) i= 0,

and we conclude that

dim X+ n + m- 1 > 2 dim X+ 2k- 2;

i.e.,

dim X < n +m- 2k + 1. Q.E.D.

§2. The Existence Theorem

The relevance of the above to the study of special linear systems comes from
the determinantal description of the varieties W;j(C) of linear systems on a
curve C given in Chapter IV, which we briefly recall. Let Y' be a Poincare
line bundle on C x Pied( C), D = L Pi a divisor of large degree m on C, r'
the divisor D x Pied( C) on C x Pied( C), and v' the projection from
C x Picd(C) to Picd(C). Then v~(Y'(r')) and v~(Y'(r')/Y') are vector
bundles of ranks m + d - g + 1 and m over Pied( C), and we defined W;j( C)
to be the locus where the evaluation map

y: v~(Y'(r')) ___. v~(Y'(r')/Y)

has rank m - g + d - r or less. Here we will find it convenient to adopt a

slightly different point of view. We begin by recalling that, if we let a be the
isomorphism

given by

a(L) = L ® (!)(D)

and if ff is a Poincare line bundle on Picm+d(C), then we may choose for

ff' the bundle

(lc X a)*(ff (8} (I)(- r')).

§2. The Existence Theorem 309

Thus, if we let r be the divisor D X Picm+d(C) inc X Picm+d(C) and denote

by v the projection from C x Picm+d(C) to Picm+d(C), the image a(W:i(C))
of W;i(C) in Picm+d(C) is the locus where the evaluation map

has rank at most m - g + d- r. We will study this map, and for this we will
write n for d + m, and take as our Poincare line bundle !l' on Picn( C) the
bundle constructed in Section 2 of Chapter IV, after fixing a point q E C.
Next, we observe that if we choose the points Pi of the divisor D = I: Pi
to be distinct and let r P be the divisor p x Picn( C) on C x Picn( C), then
v*(!l' /!l'(- r)) is a direct sum of line bundles

v*(!l'/!l'(-r)) = ffiv*(!l'/!l'(-rp)).
i

Now, v*(!l'/!l'( -rp)) is just the restriction of !l' to the section {p} x
Picn( C) ~ Pic"( C). By the construction of !l', if u: Cn -+ Picn( C) is the natural
map, and we denote by X P the subset of Cn consisting of all divisors whose
support contains the point p E C, we have

v*(!l'/!l'( -rp)) = u*({!)dXP- Xq)),

since{!)cxdM <8J {!)(p)xc" = {!)c"(Xp).Ifp = q,thelinebundlev*(!l'/!l'(-rp))
is trivial. If pi= q, v*(!l'/!l'( -rp)) is no longer trivial, but it remains alge-
braically equivalent to the trivial bundle. Our conclusion, then, is that
v*(!l'/!l'( -r)) is a direct sum of line bundles algebraically equivalent to the
trivial line bundle.
We note in passing that if we do not take the points Pi of D to be distinct,
for example, if we take D = m · q, this is no longer true. What is true in this
case, as the reader may verify, is that v*(!l'/!l'( -r)) has a .filtration by sub-
bundles whose successive quotients are trivial line bundles. For the purposes
of the succeeding argument this will work just as well.
We turn now to the bundle v*!l', which for brevity we will call En or just
E if no confusion is likely. This turns out to have a very nice geometric
description; namely for n ~ 2g - 1, we have the

(2.1) Proposition. (i) The associated projective bundle !fl>E-+ Picn(C) is

isomorphic to the nth symmetric product en, with projection map
corresponding to u: Cn-+ Picn(C); and
(ii) in terms of this isomorphism,

Proof. The first statement is basically tautological: since the fiber of E = v* !l'
over a point L E Picn(C) is the vector space H 0 (C, L), the points of !fl>E
310 VII. The Existence and Connectedness Theorems for W.(C)

correspond to pairs (L, ~) where Lis a line bundle of degree n and ~ is a one-
dimensional vector space of sections of L; i.e., a divisor D E IL 1. As L varies
these correspond to all effective divisors of degree n on C.
As for the second statement, to see this we consider the natural map

obtained by evaluation and restriction, and the induced map on IP>E

By definition, 1/1 is zero on pairs (L, ~) such that rr(q) = 0 for all rr E ~ c
H 0 (e, L); that is, in terms of the above isomorphism IP>E ~ en, 1/J vanishes
on the divisor Xq c en. But we have seen that v*(!l'/!l'( -rq)) is the trivial
bundle on Picn(e); thus the transpose map ti(J: (!)c"-+ (!)IJI>il) is a section of
(!)IJI>il) with zero locus Xq. Finally, since (!)(Xq) and (!)IJI>il) both restrict to
the same bundle (!)(1) on the fibers of E-+ X, ti(J must vanish to order 1 on
Xq. Q.E.D.

The identification given in Proposition (2.1) will be used many times.

What makes it useful here is that it allows us to establish the

(2.2) Proposition. E* is an ample bundle.

Proof. By the previous proposition this is equivalent to the statement that X q

is an ample divisor on en.
We may see this by using Nakai's criterion
(Hartshorne [3]), which states that if X is any projective variety and D a
Cartier divisor on X, then Dis ample if and only iffor every effective algebraic
k-cycle Z on X the intersection number (Z · Dk) is positive.
To apply this to the divisor Xq on en, since all the divisors {Xp}pEC are
algebraically and hence numerically equivalent, it suffices to show that for
any irreducible k-dimensional subvariety Z c en, for a general pEe the
intersection Z n X P is non-empty of dimension k - 1. But this is clear: if
D= L PiE Z is a general point we may take p to be any of the points Pi not
held in common by all D' E Z.
Alternatively, one can use Nakai's criterion to establish the general fact
that iff: X-+ Y is any finite surjective map and L any line bundle on Y,
thenf*L is ample if and only if Lis ample. We then apply this to the map
r: en-+ en from the ordinary to the symmetric product, observing that
n
r*(!)en(Xq) = Q9 n{(!)c(q)
i= 1

is clearly ample (where then;: en-+ e are the projection maps). Q.E.D.
§3. The Connectedness Theorem 311

We may now combine what we know to conclude the Existence Theorem.

We have seen that W:i(C) is isomorphic to the locus in Pic"(C) where the
bundle map

has rank at most m- g + d - r. Since E* is ample and v*(.!e/.!e( -r)) =

EB~=l (.!e/.!e(-rp,)), with each line bundle v*(.!e/.!e(-rp,)) algebraically
equivalent to the trivial bundle, we may conclude that E* ® v*(.!e/.!e(- rp,))
is ample and hence that E* ® v*(.!e/.!e( -r)) is. Applying Proposition (1.3)
we deduce the Existence Theorem of Chapter V, i.e.,

(2.3) Theorem. If Cis any smooth curve of genus g, and d and r non-negative
integers such that p = g - (r + l)(g - d + r) ~ 0, then W:i(C) # 0-

§3. The Connectedness Theorem

A fairly elementary examination of the properties of ample vector bundles
has yielded the existence theorem for special divisors. We will now go back
and establish a stronger statement about ample bundles, which will in
particular imply the connectedness theorem for W:i( C).
The statement to be proved is the following:

(3.1) Proposition. Let X be a connected variety, E and F vector bundles of

ranks m and n on X with E* ® F ample, and ¢: E -+ F any bundle homo-
morphism. Then for any k such that

dim X ~ (m - k)(n - k) + 1,
the kth degeneracy locus

Y = {x EX: rank ¢x :s;; k}

is connected.

Proof. Replacing¢ with its transpose if necessary, we can and will require that
n ~ m. The proof uses one general construction that we shall describe in
advance. Let A and B be vector bundles over a space W and let IP =
IP Hom(B, A)~ W. Let O": A-+ B be a bundle map; by composition O"
induces a bundle map

8: Hom(B, A)-+ Hom(A, A).

312 VII. The Existence and Connectedness Theorems for lfd(C)

We consider the composition of bundle maps on iP'

(!)IP'( -1) 4 n* Hom(B, A)~ n* Hom(A, A) ~ (!)IP',

where i is the standard inclusion, and tr is the trace functional. This map,
viewed as a section of the line bundle (!)IP'(1), will be denoted tr(a). The key
point here is that if Y c W is the zero locus of a and YIP' c iP' the zero locus
of tr(a), then YIP' meets a fiber iP' w of iP' over a point wE Win a hyperplane if
w ¢ Y, and contains the fiber if w E Y. Thus iP' - YIP' is a bundle of affine
spaces over W - Y, and in particular for all i

The proof of Proposition (3.1) is now accomplished by constructing the

following configuration of spaces and bundle maps.

(1) Let cjJ: E --+ F be the given bundle map, Y c X the locus
{rank c/Jx ~ k}, and set V = X - Y.
(2) Let G = G(m - k, E)~ X be the Grassmannian bundle of (m- k)-
planes in E; and let cp 6 be the composition

where S is the universal subbundle on G and i the standard inclusion.

Let Y6 be the zero locus of cp 6 and set V6 = G - Y6 . Y6 is the standard
blow-up of Y as described in Chapter IV, Section 4.
(3) Set iP' = iP' Hom(F, E) and let {): iP' --+ X be the projection. Let
c/JIP' = tr(c/J) be the trace of c/J, YIP' the zero divisor of c/JIP' and set VIP'=
iP'- YIP'.
(4) Finally, set Q = iP' Hom(cx*F, S), let {3: Q--+ G be the projection, set
c/JQ = tr(cp 6 ), let YQ be the zero divisor of c/JQ and set VQ = Q - YQ.
Note that Q may also be described as a subvariety of the fiber product
iP' x x G: if we write

then we can describe Q by

Q = {(x, a, A): Im(a) c A}.

Let y: Q --+ iP' be the projection.

§3. The Connectedness Theorem 313

Summing up, we have the diagram of spaces and bundle maps:

/ Q = P Hom( a* F. S)

.,~"'•:
P = iP' Hom(F, E)
rPIP = tr(¢)

We now proceed clockwise from IP>. To begin with, our basic hypothesis
that E* (8) F is ample implies that (IJIP'(1) is ample, and hence that VIP' is an
affine variety. Now, since the section ¢1P' of (IJIP'(1) pulls back via y to the section
¢Q of (I) Q(1) (the trace of a map ¢x: F x -> F x whose image lies in A is the same
as that of ¢xiA), we see that y- 1(VIP') = VQ; we want to analyze the cohomology
of VQ via this map. For this we observe that the fiber of y: VQ -> VIP' over
(x, a) E VIP' is the Grassmannian of (m- k- rank(a))-dimensional subspaces
of E/Im a; in particular, the fiber dimension of y is lk over the locus V~ =
{(x, a): rank(a) = m- k -l}, which has codimension (k + l)(n- m + k + l)
in IP>.
The next step requires the notion of constructible sheaf. A sheaf of abelian
groups fF on an algebraic variety Z is said to be constructible if Z can be
expressed as a finite disjoint union of locally closed algebraic subvarieties
on each one of which fF is locally constant. It is a fundamental result, to be
discussed later, that the direct images of a constructible sheaf under a
morphism of algebraic varieties are constructible. This applies, in particular,
to the sheaves Riy* 7L., which, by what we just said, are concentrated on the
affine variety v~+ 1 whenever i > 2/k. It is another fundamental property of
constructible sheaves, also to be discussed later, that the cohomology of such
a sheaf on an affine variety vanishes above the dimension of the variety
itself. In our special case, we then conclude that Hi([P>, Riy*7L.) vanishes
whenever there is an l such that

i > 2/k,
j > dim v~+ 1 = dim X + mn - 1 - k(n - m + k)
- (l + 1)(n - m + 2k + l + 1).
314 VII. The Existence and Connectedness Theorems for w;;(C)

Since we are assuming that n ~ m, this is always the case as soon as

i +j > dim X + mn - 1 - k(n - m + k).

Using the Leray spectral sequence for y and 7l., we conclude that

Hi(VQ, 7l.) = 0 if i ~ dim X + mn - k(n - m + k).

Moving along, since the map f3: VQ ~ VG expresses VQ as a bundle of
affine spaces over VG, we have

We now apply Lefschetz duality on G. Since G is a rea12(dim X + (m- k)k)-

dimensional manifold, we have an isomorphism

it follows that H 1 (G, YG; 7l.) = 0 if

2 dim X + 2(m - k)k - 1 ~ dim X + mn - k(n - m + k);
i.e., that YG is connected if dim X ~ (m - k)(n - k) + 1. Since the map a
carries YG onto Y, this concludes the proof. Q.E.D.

As we saw in Section 2 that W~(C)can be described as the (m- g + d- r)th

degeneracy locus of a bundle map¢: E ~ F, where Eisofrankm + d- g + 1,
F is of rank m, and E* ® F is ample. Applying (3.1), we obtain the Connected-
ness Theorem of Chapter V, that is

(3.2) Theorem. If Cis any smooth curve of genus g, and

p =g - (r + 1)(g - d + r) ~ 1,

then W~( C) is connected.

We note that in fact (3.1) says something more: it implies that w:;(C) is
"connected in dimension p - 1 ":i.e., that for any L, ME W:;(C) there is a
chain of irreducible components X 0 , •.. , X n of W:;( C) with L E X 0 , ME X n,
and dim X; n X;+ 1 ~ p - 1. This follows if we apply (3.1) to the restriction
of the bundle map defining W~(C) to a sufficiently general subvariety
X c Pic"( C) of codimension p - 1 passing through Land M.

To conclude this section, recall that the proof of Proposition (3.1), and
hence of the Connectedness Theorem, are based on two properties of con-
§3. The Connectedness Theorem 315

structible sheaves that we have not proved yet. Here we are going to partially
fill this gap.
The first, and basic, property we used is that higher direct images of
constructible sheaves are constructible. We are unable to give a self-contained
proof of this fact in this book (cf. the bibliographical notes at the end of this
chapter); instead, we shall assume it and give a proof of the second property
of constructible sheaves that is used. This is expressed by the following result.

(3.3) Proposition. Let $' be a constructible sheaf on an affine n-dimensional

variety X. Then Hi( X, $') vanishes for i > n.

This result is a direct generalization of the Lefschetz hyperplane theorem;

this theorem, or an equivalent version of it, is obtained by taking$' equal
to the constant sheaf?...
To prove the proposition we consider X as being embedded in some
affine space. By generic projection we may then represent X as a finite
covering of en; let? stand for the projection. By the Leray spectral sequence,

for every i. Since ?* $' is constructible, we are thus reduced to proving the
theorem for X = en. This will be done by induction on n, the case n = 1
being trivial. The statement that needs to be proved is that, iff: en --+ en- 1
is a suitable linear projection, then Rif*$' vanishes for i larger than one.
Once we know this, assuming the statement of the proposition to have been
proved for en- 1 ' we get that

The result then follows from the Leray spectral sequence for f.
It remains to find an f with the property that Ri* $'vanishes fori greater
than one. Notice that$' is locally constant off a closed subvariety W of en.
We choose fin such a way that its restriction to W is a finite map. We shall
show that for any X in en - 1 the stalk (Rif* ff)x is isomorphic to
Hi(f- 1 (x), $');this will be enough for our purposes, since.f- 1 (x) is isomor-
phic to C and the restriction of$' to f - 1 (x) is constructible. Since

Hi(f- 1 (x), $')=lim H;(N, $'),

N

where N runs through all neighborhoods off- 1 (x), we will be done if we can
find, for each such N, a neighborhood U of x and a neighborhood N' of
f- 1(x) such that N' c: N n f- 1(U) and
316 VII. The Existence and Connectedness Theorems for w;;(C)

We view en as the product c X e- 1 and f as the projection onto the second

factor. If U is a sufficiently small ball centered at x there is an open disc A in
C such that

W nf- 1(U) c Ax U,

A X [J c N.

Let t, z = (z b ... 'Zn- d be linear coordinates on c and cn- 1 such that u

is the unit ball centered at the origin. We may find a continuous function

such that 0 ::::;; ¢(t) ::::;; 1 for every t, ¢ is identically equal to one on A, and
(t, z) belongs toN whenever

We then set

N' = {(t, z): I lz;l 2 ::::;; ¢(t)}.

To see that the cohomology groups of ff on f - 1 (U) and N' coincide we

compare the two Mayer-Vietoris sequences

and notice that H;(f- 1(U)- W, ff) is isomorphic to H;(N' - W, $')since

$' is locally constant off Wand f - 1 (U) - W retracts onto N' - W.

§4. The Class of WJ( C)

We would now like to give a refinement of the existence theorem proved in

the second section ofthis chapter; namely, we will determine the fundamental
class of the cycle W:i(C) in Picd(C). To do this, we will again use the represen-
tation of W:i(C) as a determinantal variety associated to the bundle map

over Pic"( C). Given this description of W:i(C), we can use Porteous' formula
to determine its class; what we need to know are the Chern classes of the
§4. The Class of w;;( C) 317

two bundles involved. Moreover, since we have seen that v*(!£>/ .2(- r)) is a
direct sum of line bundles with first Chern class 0, it has trivial Chern class;
consequently, it only remains to compute the Chern class of E. Porteous'
formula will then give the fundamental class w:i of W:i( C) (in case W;j( C) is
of the expected dimension p) as

w:i = ( -l)<r+ 1)(g-d+r>!lr+ 1.g-d+r(cr(E))

(4.1)
= Llg-d+r.r+1(ctC -E)).

We will compute the Chern class of E by using again the isomorphism

IP'E ~ Cn. We begin by recalling some generalities on the cohomology of
projective bundles.
Let E be any complex vector bundle of rank k over a space X. On the
projective bundle IP'E ~ X, denote by S and Q the universal sub- and
quotient bundles. Thus S is the line bundle whose fiber over a point (x, ~) E
IP'E is the one-dimensional subspace~ c Ex and we have the exact sequence

0 ~ S ~ u* E ~ Q ~ 0.
Notice that, in the analytic case, Sis the dual of (!)IP'E(1). Setting x = c 1 (S*),
the Whitney product formula, applied to the above exact sequence, gives

(1 - x) · c(Q) = u*c(E);

in other words,

(4.2) c(Q) = u*c(E) · (1 + x + x 2 + . · ·).

Taking the homogeneous terms of degree k on both sides, we have (since Q
has rank k - 1),

Indeed, once one has defined the first Chern class of a line bundle, this
relation may be used as the definition of the Chern class of E: an elementary
spectral sequence argument shows that as additive groups
k-1
H*(IP'E, 1::) = EB xi· u*H*(X, 1::);
i=O

consequently, xk must be uniquely expressible as a linear combination

k-1
xk = I u*ak-i ·xi
i=O

with a; E H 2 ;(X, 1::), and we may define c;(E) = -a;.

318 VII. The Existence and Connectedness Theorems for W.(C)

There is another way of extracting the Chern classes of E from the multi-
plicative structure of the cohomology ring H*(PE, Z) that is perhaps more
direct. Recall that the Gysin homomorphism

u*: H*(PE, Z)-+ H*-zk+ 2 (X, Z),

sometimes called integration over the fiber, satisfies the push-pull formula:
for any classes a E H*(P E, Z), f3 E H*(X, Z),

u*(a · u*/3) = u*a · /3.

Applying u* to (4.2), we obtain

Since u*c;(Q) = 0 fori< k- 1, and since ck_ 1(Q) restricts to the generator
of H 2 k- 2 (Pk-1, Z) in each fiber, u* c(Q) = 1. Thus we find

u*(1 + x + x 2 + · · ·) = c(- E),

where by c( -E) we mean the formal inverse 1/c(E); specifically,

(4.3)

We may take this as a definition of the Segre class s(E) = 1/c(E), and thereby
as a definition of c(E).
In any event, it is the relation (4.3) that we will use to calculate the Chern
class of the bundle E = v* 2 on Pic"( C). Recall that we have P E ~ Cn, and
that via this isomorphism the first Chern class of the tautological bundle
(DIP'E(1) is the class x of the divisor X q on Cn. By (4.3), if u: Cn -+ Pic"( C) is
the abelian sum map we have

To evaluate the right-hand side, note that all the divisors {Xp}pec are
homologous; so that in cohomology

Now, since Xq ={DE en: D- q ~ 0}, if the qi are distinct the intersection

X q1 n · · · n X qr:x = {D E C n·· D - "i...J q.l >

- 0}

is just the image of en-~ in en via the inclusion D-+ D + L qi. Moreover,
this intersection is transverse: in terms of the identification
§4. The Class of W.(C) 319

we have, forD E Xq

so forD En Xq;
n• Tn(Xq)
i= 1
= H 0 (e, (!)(D- L q)j(!))

Thus the class xa is represented by the image of en-a in en, and the Gysin
image u*x• in Picn(e) is represented by a translate of W..-a; i.e.,

By Poincare's formula

u*x a ___
=-,-
1_--:-:- ()9- n +a
(g-n+a)! '

and consequently

Finally, we arrive at the formula for the Chern class of E:

c(E) = e- 8 •

To obtain the formula for the class w~ of W~(e) in case W~(e) has the
right dimension, we apply Porteous' formula:

w:i = Llg-d+r.r+l(c,(-E))
= Llg-d+r,r+ l(ete)
eo-d+r
(g- d + r)! (g- d + r + 1)!

= det (g - d +r- 1)! (g - d + r)!

(g- d + r)!
= e<r+ l)(g-d+r). Ll(g - d + r, g - d + r - 1, ... ' g - d),
320 VII. The Existence and Connectedness Theorems for Wd(C)

where in general we set

1 1
a 1 ! (a 1 + 1)! (a 1 +n-1)!
1 1
L'.(a 1 , .•• , an) = det a 2 ! (a 2 + 1)!

1
(an+ n- 1)!

It remains now only to evaluate the determinant L'.. This is straightforward:

if we multiply the ith row by (ai + n - 1) !, we obtain the determinant

. · · (a 1 + n - 1)(a 1 + n - 2)
(a 2 + n - 1)(a 2 + n - 2)
(a 1 + n - 1)
(a 2 + n - 1)
1)
1
d et (
.. . ..
.
··· (an + n - 1)(an + n - 2) (an + n - 1) 1

c·
This may in turn be column-reduced to the Vandermonde determinant

af

~
a! a!
n-!
det a 2 :
a~ az
:) f1 (a, -
. j>J
a)

ann-1 azn an 1
Thus,
L'.( )
al, ... ,an =Tinf1i>i(a;-
(a;+ n-a)1)1..
i=l

In the present case, we haven = r + 1 and a; = g - d + r + 1 - i, so that

f1 (a; - a) = f1 U- i)
j>i l~i<j~r+l

= TI IX!.
a=O

Assembling the pieces, we arrive at the

(4.4) Theorem. If Cis any smooth curve of genus g, and W~(C) is either empty
or of the expected dimension p, it has fundamental class

w~ = TI cd . e<r+ l)(g-d+r).
a=O (g- d + r +IX)!
§5. The Class of C~ 321

This enumerative formula is the same as Theorem (1.3) in Chapter V. We

may remark that the result we just proved implies the Existence Theorem for
special divisors, since it shows that the fundamental class of W~( C) is not
zero when p :2: 0.

§5. The Class of C:J

In this section we will combine Porteous' formula with the determinantal

description of C~ given in Section 1 of Chapter IV to compute the class c~
of q. This formula represents a refinement of the formula for the class of
W~(C) since, given the class of C~, we can evaluate w~ as

To set up our computation of cd, recall that C~ is defined to be the (d - r)th

determinantal variety associated to the differential

of the map u: cd--+ J(C) ~ Picd(C). Since the tangent sheaf eJ(C) to the
Jacobian of Cis trivial, Porteous' formula then tells us that

(5.1) c~ = tlg-d+r ••(c,(- E>cJ).

It remains to calculate the Chern classes of Cd. We do this first in case

d :2: 2g - 1, so that

is the projective bundle associated to the direct image E = v* 2' of the

Poincare line bundle on Picd(C) = J(C). As with any fiber bundle, we have
an exact sequence

(5.2)

where the" vertical tangent bundle" eu is the bundle whose fiber at any point
DE cd is the tangent space to the fiber through D of cd--+ J(C). Since
Cd = IP'E--+ J(C) is a projective bundle, we have the Euler exact sequence
for eu
(5.3)
322 VII. The Existence and Connectedness Theorems for w:;(C)

This sequence may be obtained by observing that for any vector space V,
and any elements v E V and l E V*, if n: V- {0} --+ IPV is the projection then

depends only on the image of A. in IPV. This defines a map V ® V*--+ 7[_. 11PV
whose kernel at each point [A.] of IP Vis generated by V ® A. J. and the element
"trace" in Hom(V, V)* = V ® V* (the Euler relation). As described, this
map then globalizes to give the exact sequence (5.3).
Together the two sequences (5.2) and (5.3) yield

c,(E>c) = c,(u*E ® C9o>i1)).

To evaluate the right-hand side we write (as before) x for c 1 (C9o>E(l)) (the
class of the divisor Xq;;; Cd_ 1 in Cd) and formally factor c,(u*E):

d-g+1
c,(u* E) = fl (1 + tf3;).
i= 1

Then

c,(u*E ® (9o>E(1)) = fl (1 + t(x + /3;))

i

= (1 + tx)d-g+ 1 fl
;
(1 + ___!fi_)
1 + tx
= (1 + tx)d-g+ 1cr(u*E),

where r = t/(1 + tx). Since we have seen that c.(u*E) = e•8 (here we are
writing () for u*8, a notational convention we will continue to follow for the
remainder of the book), we have the formula

(5.4)

ford:?: 2g- 1.
Finally, if we embed Cd_ 1 in Cd as the divisor Xq, the normal bundle

has Chern polynomial (1 + xt). From the sequence

§5. The Class of C~ 323

we conclude that

c,(ec._) = c,(8c)/(1 + xt)

= ( 1 + xt)d-ge-•Of(l +tx>.

Consequently ifformula (5.4) is valid for Cd it holds for Cd_ 1 as well; and so
we conclude that it must hold for all d.
According to (5.1) to compute c~ we must evaluate Ag-d+r,r for the power
series

1
c,(-8c.) = ___ = (1 + xt)g-d-1etllf(l+xt)
c,(8c.)
tk(Jk
= '\' (1 + )g-d-1-k ~
00
(5.5) ~ xt k' .
k=O ·

Noting that, since r 2 1, the determinant Ag-d+r,r involves only terms of

degree at least g - d + 1 we see that terms in the sum (5.5) corresponding to
k ::;; g - d do not appear in the determinant. Setting

y = -x, m = g- d,

and using

(1 7 i y't',
+ xt)- a- 1 = "(a+i) ..

we have

L Ln [(n) yien-i+g-d J
t•+g-d
n;o:Oi=O i (n-i+g-d)! '

and correspondingly

= det
1 ( 1) yieg-d+1-i ... I (,.) yieg-d+r-i
; ~0 i (g - d + 1 - i) ! ; = 0 i (g - d + r - i)!
324 VII. The Existence and Connectedness Theorems for ~(C)

Rotating the above matrix clockwise through 90°, we get

Llg-d+r,r((1 + tx)g-d-1et6)
= ( -l)r(r-1)/28'(g-d)

y () y' ry'-1() ()'

-+-:------:-:-:- -+ + .. ·+-:---:-:-
m! (m+ 1)! m! (m+1)! (m+r)!
y2 2y() ()2
- + + -,----,-,--
xdet m! (m+1)! (m+2)!

y' ry'-1() ()' y2r- 1 ()2r-1

-+ + .. ·+-:---:-:- - - + ... + -,-----::------:-:-:-
m! (m+1)! (m+r)! m! (m+2r-1)!
= ( -l)r(r+ 1)f2()r(g-d)
1 yr-1 er-1
m!
--+
m!
.. ·+----
(m+r-1)!
y () y' ()'
- +--- - + ... + ...,-------,--,. 0
x det m! (m+1)! m! (m+r)!

y' 8' Y2r-1 ()2r-1

- + ... + -,---,-- --+ .. ·+ (m+2r-1)! 0
m! (m+r)! m!
(5.6)

We will evaluate the determinant in the last expression of (5.6) by a sequence

of row and column operations.
First we subtract from each row y times the row above, thereby eliminating
the first term of each entry in all but the first row. The (i + 1)st column is now

yi iyi-1() ()i
m! + (m + 1)! + ... + (m + i)!'
yi() iyi- 1()2 ()i+ 1
---+ + .. · + - - - -
(m + 1)! (m + 2)! (m + i + 1)!'
yi+ 1() (i + 1)yi()2 ()i +2
-=-=-----:-:--:- + + ... + -,-----~
(m + 1)! (m + 2)! (m + i + 2)!'

yr+i-1() (r + i- 1)yr+i-2()2 er+i

_::_____ _ + + ... + - - - -
(m + 1)! (m + 2)! (m + r + i)!'
§5. The Class of Cd 325

for 0 ::;; 1 ::;; r - 1; the last column is '(1, x, 0, ... , 0). We now repeat the
operation, beginning this time with the third row. The (i + 1)st column is now

yi iyi-18 8i
m! + (m + 1)! + · ·. + (m + i)!'

yi8 iyi-182 8i+ 1

_::....___+ +···+------,----
(m+1)! (m+2)! (m+i+1)!'
i 82 iyi-1 83 8i+2
(m + 2)! + (m + 3)! + · .. + (m + i + 2)!'

yr+i-282 (r + i- 2)yr+i-383 8r+i

:....___ + + ... + - - - -
(m + 2)! (m + 3)! (m + r + i)!'

and the last column is '(1, x, x 2 , 0, ... , 0). Continuing in this fashion the
matrix in (5.6) becomes

(5.7)

1 Y 8 yr-1 8'-1
-+ -+···+ 1
m! m! (m+1)! m! (m+r-1)!

8 y8 82 y'-18 (}'
+ +···+ X
(m+1)! (m+1)! (m+2)! (m+1)! (m+r)!

8' y8' w+ 1 y'-18' 82r-1

+ +···+ x'
(m + r)! (m+r)! (m+r+ 1)! (m+r)! (m+2r-1)!

We now perform the same sequence of operations on the first r columns

of the matrix (5. 7). When this is done we arrive at the matrix

1 8
1
m! (m + 1)! (m + r- 1)!
8 82
X
(m + 1)! (m + 2)!

8' 8'+ 1
x'
(m + r)! (m + r + 1)! (m + 2r- 1)!
326 VII. The Existence and Connectedness Theorems for Wd(C)

Expanding this determinant by cofactors of the last column, we obtain

c~ = ±(
a~o
-l)r(r-1)/2 ~(m, m + 1, ... , ~•... , m + r)ya(Jr(g-d+r)-a,

where ~(al> ... , ak) is defined as in the last section; evaluating~ yields the

Theorem. If C is any smooth curve of genus g, and C~ is either empty or has

the expected dimension p + r, it has fundamental class

c:i=(rl i!. )±<-1Y(g-d+r+o:-1)!

i~o (g- d + r + z- 1)! a~o o:! (r- o:)!

Bibliographical Notes

The history of the theorems proved in this chapter is recounted in the

bibliographical notes for Chapter V; we will not repeat it here. A more
general treatment of ample vector bundles may be found in Hartshorne
[I] and in Griffiths [2]. The fact that a section of an ample vector bundle
must have a zero if the dimensions so indicates was first proved in Bloch-
Gieseker [1]; the stronger connectedness theorem for degeneracy loci of
bundle maps was found by Fulton and Lazarsfeld [1], who in the same paper
applied it to deduce the connectedness of W:i(C). The computations of the
expected classes of the loci W~(C) c J(C) and C:i c Cd were first made
in Kempf [1] and Kleiman-Laksov [1], and in Kleiman-Laksov [2],
respectively. Finally, sources for the material on constructible sheaves are
Thorn [1], Mather [1], and Artin [1].

Exercises

A. The Connectedness Theorem

In this series of exercises we will derive and then apply the basic connectedness theorem
of Fulton and Hansen [1]:

(*)Theorem. Let X, Y c IP'" be irreducible varieties with dim X+ dim Y > n. Then
X n Y is connected.

A-1. Show that Theorem(*) is equivalent to the

(**)Theorem. If~ c IP'" x IP'" is the diagonal, and Z c IP'" x IP'" any irreducible
subvariety of dimension greater than n, then Z n ~ is connected.
Exercises 327

A-2. Let Y c IP'" be any irreducible variety of dimension k ~ 2 and {X;.Lel'' any
pencil of hyperplane sections of Y with base locus of dimension k - 2. Show that
if there is a point pEn X.< of the base of the pencil such that the general X;. is
smooth at p, then the general X;. is irreducible.

A-3. Let Y c IP'" be as in the preceding exercise, and let p, q be any points of Y such
that the line pq ¢ Y. Let i\ c IP'" be a general!-plane containing p and q, I > n - k.
Show that, except possibly in the case k = 2, n = 3, i\ n Y is irreducible. (Also,
find a counterexample in case Y is a surface in IP' 3 .) Deduce, without any
hypotheses on Y. p, and q, that p and q lie in the same connected component of
i\ n Y.

A-4. Let Y again be an irreducible, k-dimensional variety in IP'", p, q any points of Y.

Suppose that {A,} is any family of /-planes through p and q. Show that if, for
t t= 0, p and q lie on the same connected component of Y n i\, then the same is
true fort= 0. Use this and the preceding exercise to deduce Theorem(*) in case
X is an arbitrary linear subspace of IP'".

A-5. Consider now the birational map

</>:IP'" X IP'"--~IP'2n

given by

</>([Xo, ... ,X.],[Y0, ... , Y,,])= [1,Xx o1 , ... , x.,Y1, ... ,

X 0 Y0 Y0
Y..]
= [X 0 Y0 , X 1 Y0 , •.• , X. Y0 ,
X 0 Y1 , •.. , X 0 Y,],

Let~ be the graph of</>, with projections rr 1 , rr 2 onto IP'" x IP'" and IP' 2". Show that:

(a) rr 1 : ~ -+ IP'" x IP'" is simply the blow-up of the locus X 0 = Y0 = 0.

(b) rr 2: ~ -+ IP' 2" is the blow-up of the union of the planes i\ 1 , i\ 2 given by

(Z 0 = Z 1 = · · · = z. = 0)

and

(Z 0 = z.+ 1 = · · · = Z 2• = 0).

(c) The image in IP' 2" of the proper transform in ~ of the diagonal ~ c IP'" x IP'"
is the plane i\ given by

(d) i\ is disjoint from i\ 1 and i\ 2.

328 VII. The Existence and Connectedness Theorems for Wd(C)

A-6. Now let Z c IP'" x IP'" be any irreducible subvariety, and set Z = nz(n! 1 (Z)) c
IP' 2 ". Using the preceding exercise, show that if Z n A is connected, then Z n ~
is; using Exercise A-3 above, deduce Theorem(**).

A-7. By a similar argument, prove the stronger

Theorem. Let Y c IP'" be any subvariety, X any irreducible variety and f: X -+ IP'"
any regular map with dim f(X) + dim Y > n. Then f- 1 (Y) is connected.

A-8. Let X be any smooth irreducible variety, f:X-+ IP'm any map, and assume
m < 2 · dim X. Show that iff is an immersion (df is nowhere zero) then f is an
embedding.
(Hint: Look at f X f:X X X-+ IP'm X IP'm.)

A-9. Deduce from the preceding exercise that if X c IP'" is any smooth, irreducible
subvariety and the tangential variety

T(X) = U Tv(X) c IP'"

peX

has dimension < 2n, then the secant variety

S(X) = ( U
p,.q
pq)
p,tJEX

equals T(X).

A-10. Using Exercise A-8 above, show that if X c IP'm is any smooth, irreducible variety
with dim X > m/2, then X has no finite unramified coverings, and hence that
H 1(X, £'.) = 0

B. Analytic Cohomology of Cd, d s 2g - 2

We consider here that part of H*(Cd, IIJ) generated by x and (}when d ~ 2g - 2, and
recall (cf. Section 5 of Chapter VIII) that when Cis general, this is the entire analytic
part H!.(Cd, IIJ) of H*(Cd, IIJ).

B-1. Let C be a smooth curve of genus g, d ~ 2g- 2, and m ~ d/2. Show that the
(m + 1) x (m + 1) matrix with entries

has determinant

det ~ = (g!)m+l n
m
__a'·_ # 0.
•=O(g-o:)!

From this conclude that xm, xm- 1 (}, •.. , em are independent in H*(Cd, IIJ).
(Hint: use the Poincare formula to show that (xd-•(f')[Cd] = g!f(g- o:)!)
Exercises 329

B-2. Show that when d :::; 2g - 2 and m ;;::: d/2, the classes xm, xm- 18, ... , em span a
subspace of dimension d - m + 1 in H 2 m(C4 , II)). In this case, show that any
d - m + 1 of these classes are independent.

B-3. Show that when C is general and d :::; 2g - 2

2m(C , """
. H ••
d 1m 4 'lol!) =
{md + 1, m:::; d/2,
-m+ 1, m;;::: d/2.

B-4. When d :::; 2g - 2 and 0 :::; k :::; 2m - d - 1, show that the relations

k
J;,(x, (}) = L (-1)"
d-m+1
g-
(
- IX •
)I
xm-k-•(Jk+a
a=O 1X!(d-m+1-1X)!

=0
are satisfied. From the preceding exercise, conclude that these constitute a basis
for the relations satisfied by the classes x•(JP (IX + f3 :::; g - 1).

B-5. Show that for 1 :::; IX :::; d- 1

g + -(g-
- - -'. - - - l X. X d- 2(J .
(1 ) d-1 1)1
X d-1-•e• - -lXX
(g-:x)! (g-IX)!

C. Excess Linear Series

C-1. Suppose that E, F,and F' are vector bundles of ranks m, n, and n + 1 on a projective
variety, and

a pair of bundle maps with n surjective and E* ® Ker n ample. Using (1.2) show
that if~· (resp. ~) is the locus where a (resp. no a) has rank k or less,

dim~';;::: dim~- (m- k).

C-2. Apply the preceding exercise to conclude that on any curve C,

dim W~_ 1 (C) ;;::: dim W~(C) - r - 1,

and, in particular, that if C possesses oo 2 gJ's then it possesses a gJ_ 1 •

C-3. Deduce as corollaries of the preceding exercise that

dim W~tl(C);;::: dim W~(C)- (g- d + r),

and

dim w~+ 1 (C);;::: dim W~(C)- (g- d + 2r + 2).

Chapter VIII

Enumerative Geometry of Curves

In this chapter we will continue in the direction suggested by Sections 4 and 5

of the previous chapter: that is, we will try to solve some of the enumerative
problems that arise in the theory of curves and linear systems. While this
is in some sense a quantitative approach, qualitative results may also emerge.
For example, the answer to the enumerative question: "How many g~'s
does a curve C possess" (Theorem (4.4) in Chapter VII) implies the existence
theorem (Theorem (2.3) in Chapter VII).
Before we begin, it is necessary to add one major tool to our kit. In Sections
4 and 5 of Chapter VII, the computation of the classes of W~( C) and C~
came down to computations of the Chern classes of vector bundles (the
push forward E = v* 2' of the Poincare line bundle, and the tangent bundle
Gcd to Ca). These were determined on an ad hoc basis, using special circum-
stances in each case. To proceed much further, we need a more general
technique for making such computations, and this is the Grothendieck-
Riemann-Roch formula. In the first section of this chapter, this formula is
introduced and, to get started, the Chern classes of E and 8c dare re-computed.
Following this, we use the Grothendieck-Riemann-Roch formula to
establish the secant plane formula. This may be thought of as an extension of
formula (5.6) in Chapter VII giving the class of C~: just as this counts the
divisors that fail to impose independent conditions on the canonical series,
the secant plane formula counts divisors failing to impose independent
conditions on an arbitrary linear series.
Finally, we introduce and compute the classes of the various diagonals in
the symmetric product of a curve, and combine this with the secant plane
formula to answer some additional enumerative questions.

§1. The Grothendieck-Riemann-Roch Formula

In general, the Grothendieck-Riemann-Roch formula may be used to

compute the Chern classes of the alternating sum of the direct images of a
coherent sheaf :#' on a variety X under a proper morphism n: X -+ Y. To
explain this we introduce some terminology. For simplicity, we assume that
X and Yare smooth, projective varieties. In addition to the Grothendieck
§I. The Grothendieck-Riemann-Roch Formula 331

group K(X) constructed from the holomorphic vector bundles on X we

may also consider the analogously defined group K(X) constructed from the
coherent analytic sheaves on X. There is an obvious map

(1.1) oc: K(X)--+ K(X),

and, since any coherent sheaf on a smooth projective variety has a global
finite resolution by locally free sheaves, oc is surjective; in fact, it is an iso-
morphism.
Given a proper morphism

n: X--+ Y

and a coherent analytic sheaf .fi' on X, we define

n 1(ff) =I (-lYRin*.fi' E K(Y).

i

Next, for E --+ X any holomorphic vector bundle with Chern polynomial

c1(E) =I c;(E)t;
i
(1.2)
= fl(l + OC;t)
i

we define the Chern character to be

ch(E) = I e"' E H*(X, Q).

i

This makes sense, since the kth homogeneous term

1 "L.
kf k
OC;
• I

on the right is expressible as a polynomial in the elementary symmetric

functions of the oc;. Thus, if we write ch; for the ith homogeneous term of ch(E),

ch 0 = rank(£)
ch 1 =I oc; = c (E) 1

ch2 = ~I ocf = ~ [ci oc;)2 - 2 i~i oc;ocj]

c 1 (E) 2 - 2ciE)
2
332 VIII. Enumerative Geometry of Curves

The Chern character has the important properties that, for

an exact sequence of holomorphic vector bundles,

ch(F) = ch(E) + ch(G);

and for any bundles E, F

ch(E ®F)= ch(E) · ch(F).

In other words, if we consider the ring structure on K(X) induced by

tensor products of bundles, then the Chern character induces a ring homo-
morphism

ch: K(X) ~ H*(X, Q).

Conversely, ch(E) may be defined to be the value onE on the unique functorial
ring homomorphism K(X) ~ H*(X, Q) that maps every line bundle L
to e<•<LJ.
We observe that, because of the isomorphism (1.1), the definition of the
Chern classes and Chern character may be extended to coherent sheaves. If
V c X is a divisor, then from the exact sequence:

we infer that

c,(@v) = (1- tv)- 1

= 1 + tv + t 2 v 2 + · · · ,

where v E H 2 (X, Q) is the fundamental class of V.

The last definition we require is that of the Todd class, td(E), given for a
vector bundle E with Chern polynomial (1.2) by

td(E) = n -a~-a;
'
1 E H*(X, Q).

As with the Chern character, this is a symmetric power series in the a;,
and so may be expressed in terms of the Chern classes of E. Thus, since

a a a2
-,---------,=1+-+-+ ...
1- e • 2 12
§2. Three Applications of the Grothendieck-Riemann-Roch Formula 333

we have

TI r:x; a
OC· r:J.~
= 1 + L: _!_ + L:--'- + L: _,_)
OC·r:J. ·
+ ···
; 1- e ' 2 12 i<j 4

= 1 + tc 1 (E) + l 2 (ci(E) + ciE)) + · · ·.

The notation td(X) is used when E is the tangent bundle of X. The Todd
class is multiplicative in the sense that, for an exact sequence

o~E~F ~ G ~o,

we have

td(F) = td(E) td(G).

With these explanations, the Grothendieck-Riemann-Rochformula is

ch(n 1 ~) • td(Y) = n*(ch(ff) · td(X)).

The simplest case of this is the case in which Y is a point. Here, using the
identification H 0 (Y, Q) =II), we have that td(Y) = 1 and n 1ff = x(ff)
= L: (-l);h;(X, ff), so that the formula reduces to the Hirzebruch-Riemann-
Roch formula

x(ff) = (ch(ff) · td(X))[X].

Thus, if X is a curve of genus g

x(ff) = c 1 (ff) + rank(ff) · (1 -g),

and if X is a surface

which reduces, in case ff is a line bundle, to

( a;;)_ ci{ff)- Ct(ff) · Kx (En )

X:#' - 2 + X lYX •

§2. Three Applications of the Grothendieck-Riemann-Roch

Formula
Our first application of the Grothendieck-Riemann-Roch theorem will
be to rederive the formula of Section 4 of Chapter VII for the Chern classes
of the bundle Eon Picn(C) .To do this, recall that E is defined to be the direct
334 VIII. Enumerative Geometry of Curves

image v* ft' of a Poincare line bundle ft' on C x Pic"( C) under the pro-
jection v: C x Pic"( C)--> Pic"( C); we will apply the Riemann~Roch formula
to the map v.
We begin by developing some notation for the cohomology of C and
Pic"( C), as follows. First, we choose a symplectic basis c5 1 , ... , c5 29 for
H 1 (C, l'.) (e.g., the dual of the basis y1 , ... , y29 for H 1 (C, l'.) introduced in
Section 5 of Chapter I); we will also denote by c5 1 , ••• , c5 29 the classes in
Pic"( C) = J(C) corresponding to the above via the isomorphism

We will denote by c5'f, ... , c5~ 9 (resp. c5'1 , .•. , c5~ 9 ) the pull-backs to C x Pic"( C)
of these classes from Pic"( C) (resp. C). Finally, we will write (}for the pull-
back to C x Pic"( C) of the class (} E H 2 (Pic"( C), l'.), and we will denote by '1
the pull-back of the class of a point on C. Summarizing, we have

r:t. = 1, .. . ,g,
c5~c5p = 0 if f3 =f. r:t. ± g,
g

e= Ia= 1
c5~c5;+~·

We now make two observations. First, obviously

td(Pic"( C)) = 1,

and almost as obviously

td(C x Pic"(C)) = 1 + (1 - g)'7.

Secondly, since we have taken n ;;=:: 2g - 1, the higher cohomology of the

Poincare bundle ft' on the fibers of v vanishes. Since v is flat, it follows that

Putting these together with the Riemann~Roch formula yields a formula

for the Chern character of E = v* ft':

(2.1) ch(E) = v*((1 + (1 - g)l'/) · ch(f£')).

It remains thus to determine the Chern class of ft'. This may itself be found
by applying the Riemann~Roch formula to the original construction of ft'
(cf. Section 2 of Chapter IV); we leave this as an exercise. It turns out to be
substantially simpler in this case to proceed directly. Set
§2. Three Applications of the Grothendieck-Riemann-Roch Formula 335

where ci.i is the component of c 1 (2') in the (i,j)th term of the Kunneth
decomposition (integral coefficients throughout)

H 2 (C x Pic"(C)) = (H 2 (C) ® H 0 (Pic"(C)))

E9 ® H 1 (Pic"(C)))
(H 1 (C)
E9 (H 0 (C) ® H 2 (Pic"(C))).

Since 2' is trivial on {q} x Pic"( C) we infer that c 0 • 2 = 0, and since 2'
has degree n on a fiber C x {L} of v it follows that c2 • 0 = nry. To determine
cL \note that if w: C ~Pic"( C) is the map

w: p 1----> u(p + (n - l)q),

then the universal property of the Poincare line bundle implies that

Oc X w)*!l' = mcxc(L\ + (n- l)rq- r~),

where rq = {q} X C, r~ = c X {q} and L\ c c X cis the diagonal. Since

lc x w induces an isomorphism on the (1, 1) part of H 2 , we conclude from
the standard formula for the class of the diagonal that
g
c1.1 = L\1.1 = _ ~
'\' ((J"(j'
Ct. g+a
_ (j"
g+a
{j')
a •
a=1

Let us call this class y. Note that

')12 = - I ((j~(j~+.[J;+.[J~ + (j;+.(J~(j~(j~+·)

•= 1
= -2ry8,
while evidently

')1 3 = 11' ')I = 0.

Summarizing,

and therefore

ch(!l') = ec•<LJ = I (n. 11 + y)k

k=O k!
= 1+n ° 11 + ')I - rye.
336 VIII. Enumerative Geometry of Curves

Plugging this into (2.1), we have

ch(E) = v*((1 + (1 - g)r/)(1 + n · '7 + y - 1]8))

= v*((n - g + 1)'7 - 1]8)
= (n - g + 1) - e.
We now solve this for the Chern classes of E. Writing
n-g+ 1
cc(E) = L
~=1
(1 + A.~t)
we have

LA~= -8
L A.! = 0 for k > 1.

Then

~>{J

More generally, the formulas above for the Newton functions in the A's show
that
cc(E) = elog(ll( 1 + ;.~t))

= eEiog(1+J.~t>

= e-EJ.~t

= e-te,

or, which is the same, that

Our second application ofthe Grothendieck-Riemann-Roch formula will

be to redo the computations of Section 5 of Chapter VII for the Chern classes
of the tangent bundle E>cd to the symmetric product Cd. To do this, recall
from Chapter IV that if .1 is the universal divisor of degree d on c X cd,
the pointwise identification

of Laurent tails with tangent vectors to Cd globalizes to a sheaf isomorphism

§2. Three Applications of the Grothendieck-Riemann-Roch Formula 337

where 1t1, 1t2 Stand for the projection maps on C X Cd. It is to this situation
that we apply the Grothendieck-Riemann-Roch formula, together with the
vanishing of the higher direct images of(!}&(8), to obtain

Omitting rrf, rrf, the right-hand side is

The td(Cd) factors cancel out and, recalling that 1J denotes the pull-back of the
class of a point on C, we have

From the usual exact sequence

we obtain

where (J E H 2(C x Cd, Z) is the class of 8, and so our problem is simply to

determine this class.
To begin with, we note that the inclusion j: C--+ Cd sending p to
p + (d- l)q induces an isomorphism

(2.2)

To see thatj* is surjective, we note that, as already observed, the composition

is an isomorphism; to see that j* is injective, we may apply the Lefschetz

hyperplane theorem to the string of smooth ample divisors

As before, we will denote by () 1 , •.. , () 29 a symplectic basis for H 1 (C, Z) :;;;

H 1(Cd, Z) and by (Jl_, .•. ' ()Zg (resp. ()'{, ... ' (J2g) its pull-back to c X cd
from C (resp. from Cd); also we will continue to write ()for the pull-back
of the class () E H 2(J(C)) to Cd, and x for the class of the divisor Xq =
q + cd-1 c cd.
With this understood, we compute the class (J of the universal divisor 8 by
breaking it up into its components under the Kunneth decomposition of
H 2 (C X Cd):
338 VIII. Enumerative Geometry of Curves

with (ji.j E H;(C) ® Hj(Cd). Clearly, since~ restricted to {q} x Cd is just the
divisor Xq,

(j0,2 =X.

Equally clearly, since~ restricts on each fiber C x {D} of n 1 to a divisor of

degree don C,

(j2, 0 = dry.

Finally, since the map

1c X j: C X C-+ C X Cd

induces an isomorphism on the (1, 1) part of the cohomology, and since

Clc X j)*~ = ~c + (d- 1)rq,

where rq = {q} x C and ~c is the diagonal in C x C, we may calculate

as before that

g
(jl, 1 = - L...
" (()'a (j"g+a - (j'g+a (j")
a ·
a=l

Calling this classy, we combine the preceding formulas to obtain

(j = dry + ')! + X;

as before we may observe that y2 = - 2ry8, while ry 2 = 1JY = y3 = 0.

Returning to the computation of ctCE>c.), we have

(2.3) ch(E>c.) = (n 2 )*((1 + (1 - g)ry) · (edn+y+x - 1)).

On the right-hand side we have

edn+y+x = L (dry+ y + x)k

k k!
§2. Three Applications of the Grothendieck-Riemann-Roch Formula 339

Using the fact that y2 = - 2rye this gives

Plugging into (2.3) we find

ch(0c) = g - 1 + ((d - g + 1) - e)ex.

To see from this what the Chern class of 0ca is, note that, by the previous
argument, (d - g + 1) - e is the Chern character of a vector bundle of
rank d- g + 1 having Chern polynomial e-' 8• Consequently, ((d- g + 1)
- e)ex is the Chern character of the tensor product of such a vector bundle
with a line bundle having first Chern class x, and by the computation of
Section 5 of Chapter VII it follows that

Our third application of the Grothendieck-Riemann-Roch theorem

will yield a new formula. Recall that Cd is the cycle of divisors of degree d
on C that fail by r to impose independent conditions on the canonical series
IK I; equivalently, these correspond to d-secant (d - r - 1)-planes to the
canonical curve cjJ K( C) c IP9 - 1 . We would like now to derive a similar
formula, but with the complete canonical series replaced by an arbitrary
linear series f!.C = IP V c IL 1. Our approach will be to construct a bundle
EL on Ca whose fiber (EL)v over a point DE Ca is the space of sections of
Lover D: e.g., if D = p 1 + · · · + Pa, with the P; distinct,

(EL)D = EB
i
LPi'

We should then have a natural map from the trivial bundle V ® (!)ca on
Ca to EL given by evaluation, and the degeneracy loci of this bundle map will
be the cycles in question.
We begin by defining, for any line bundle L on C, a vector bundle EL
on Ca by

Here again n 1 , n 2 are the projection maps on C x Ca and d c C x Ca is

the universal divisor. Note that since dis flat over Ca and ref Lis locally free,
the sheaf(!).!..® nfL is flat over Ca; thus EL is locally free and we have the
identification
340 VIII. Enumerative Geometry of Curves

Moreover, the natural map of sheaves on C x Cd

ret L --> (I) Ll. ® ret L

given by restriction pushes forward to a map

from the trivial bundle H 0 (C, L) ® (l)cd on cd to EL. This in turn may be
restricted to a map

(2.4)

for any subseries V c H 0 (C, L) to give the desired evaluation map.

It remains to determine the Chern class of EL. The result is expressed in the

(2.5) Lemma. If L has degree n,

Proof. It will suffice to sketch the derivation, which is essentially the same as
in the last application. From the vanishing of the higher direct images and
the Grothendieck-Riemann-Roch theorem we obtain

Cancelling out the terms td(Cd) and using the exact sequence

we have

ch(EL) = (re 2 )*((1+ (1 - g)ry) · (enq - enq-o))

= (re 2 M(l + (1 - g)ry) · (1 + n ·11 - (1 + (n - d)ry - Ory)e-x))
= n- g + 1 + (d- n + g - 1 + O)e-x,

from which the lemma follows.

§3. The Secant Plane Formula: Special Cases

Before deriving the secant plane formula in general, it is worthwhile to
mention one special case in which the calculations are not so cumbersome.
In order to facilitate the computations involved, we pause briefly to intro-
duce and illustrate two useful notations.
§3. The Secant Plane Formula: Special Cases 341

We denote by

the coefficient oft~' · · · t;b in a formal Laurent series g(t 1 , ..• , tb).
We shall define the binomial coefficients (:), regardless of the sign
of the integers n and i, by the following convention

n(n - 1) · · · (n - i + 1) . .
.1 Ifl > 0,
G) = {
1 l. ifi = o,
0 if i < 0.

For example,

(1 + t)a = ~ (;)ti
= ~(-lY(-a -i 1 + }i
regardless of the sign of a. Moreover, for any formal series c(t) = I citi,

[(1 - xt)~c(t/(1 - xt))],P = [~ citi(l - xtY- i]

J tP

"(-oc+j+i-1)
= [ L...
i,j
. cix't'
l
··+·]1
tP

that is, we have

(3.1) [(1- xt)~ · c(t/(1- xt))],P = [(1 + xt)P-~- 1 c(tHp.

We now consider a linear series !::0 of degree nand dimension r. This means
that we are given a line bundle L ~ C of degree n, an (r + !)-dimensional
linear subspace V c H 0 (C, L), and then !::0 = IP'(V) consists of all divisors
of sections in V. We do not assume that !::0 is complete or non-special. The
cycle rlV) of all divisors of degree d that are subordinate to the linear
series V is defined to be

rif0) = {DE Cd: E - D 2 0 for some E E !::0}.

342 VIII. Enumerative Geometry of Curves

As will be seen momentarily, there is a natural scheme structure on rlS&),

and we will prove the

(3.2) Lemma. Suppose that n :?: d :?: r. Then ri~) is r-dimensional and its
fundamental class Yl~) is given by

Proof. As a set, ri~) is the locus of those D such that there exists a section
s E V vanishing in D; i.e., the locus where the bundle map (2.4)

has a kernel. We then define ri~) to be the rth determinantal variety of IXy
and compute its class accordingly. By a relatively standard formula in the
theory of Chern classes (or by Porteous' formula), we have

Using (2.5), cd_,(EL) is the coefficient of td-r m the power series

(1- xt)d-n+g- 1e8'1( 1-tx>. By (3.1), we have

cd_,(EL) = ((1- xt)d-n+g-leOt/(1-xtJ],•-•

= [(1 + xt)n-g-rete],•-··

and hence

Remark. In case d = n, we are computing the class v of the linear system V

itself, and since eg+ 1= 0, (3.2) gives

X 089
if b=n-g-r>O
v=-
'
g. - '

where i5 is the difference between the "expected dimension" n - g of the

complete linear system IL I and the actual dimension r of V. Of course, in
§3. The Secant Plane Formula: Special Cases 343

case n 2 2g - 1, we knew this already, since Vis a linear space of codi-

mension b in a fiber of Cd -+ J( C), and the class of a general fiber is 89 /g!.

As a second application of (2.5) we will now prove the following result.

Proposition. Let d, r ~, r.x. = 1, ... , 1 be integers such that ax = d - r~ 2 0 and

I~ a~ = d. Suppose that there are finitely many effective divisors of degree d
subordinate to each of l series ~. of respective degrees n~ and dimension r~.
Then the number of these divisors, counted with appropriate multiplicity, is

Proof. The virtual number of effective divisors of degree d subordinate to

the integer
~ 1 , ... , ~~is

l
TI Yi~~) E H 2 d(Cd, Z) ~ Z,
~=1

where Yi~~) is the class of ri~,). By (3.2) this is

(3.3)

To evaluate this we use the Gysin mapping

By Poincare's formula and the push-pull formula we have

u (xd-fl(JP) = {g!j(g- /3)!~ if f3 s min(d, g),

* 0 if f3 > min(d, g),

where ~ is the positively oriented generator of H 29(J(C), Z) ~ Z. By our

convention this can also be written

if f3 s d,

if f3 > d.
344 VIII. Enumerative Geometry of Curves

Effectively, then, in (3.3) we may set x = 1 and

if {3 :::;; d,

if {3 > d,

to have

pto (~) (~ tay

=(1 +at/ar (modulo{tl> ... ,tl)d+l),

which implies the lemma. Q.E.D.

Example. We shall calculate the number yz{!') 1)y 2 (!') 2 )[Cd] of pairs of points
common to two pencils !') 1 and !') 2 of degrees n 1 and n2 • According to the
proposition this is the coefficient of t 1 t 2 in

which turns out to be

(n 1 - 1)(n 2 - 1) - g.

It is interesting to verify this by a geometric argument. The pair of pencils

gives a map

that has bidegree (n 1, n 2 ). If we assume that the image curve is birational to

C and has /J ordinary double points, then by the genus formula for curves
on a quadric

/J = (n 1 - 1)(n2 - 1) - g,

which is the formula above.

§4. The General Secant Plane Formula 345

Example. The number of (r + 1)-tuples of points common to a pencil

~ 1 of degree n 1 and an r-dimensional series ~ 2 of degree n 2 is

Yr+1(~1)'Yr+1(~z)[Cd] = [(1 + t1)n,-g- 1(1 + tzt 2 -g-r(l + t1 + tz)9]t~t 2

= (n 2 - g- r)[(1 + td'- 9 - 1 (1 + t 1 + t 2 )9]r~
+ g[(1 + t1t·-g- 1(1 + t1 + t2)9 - 1 ]t~

= (n 2 - g - r) (n r
1 - 1) + (n r 2)
g
1 -

-(
- n r)(n -1) - g (n -2) .
2 -
1 1
r r- 1

§4. The General Secant Plane Formula

As indicated earlier, we will be concerned with the following problem:
given a linear system~= IJl>V c ILl of degree nand dimensions, find the
class of the cycle V~ consisting of all divisors of degree d that impose at most
d - r conditions on .@
Of course, the formulas in Section 3 are examples of this when d - r = s.
Another case we have already dealt with is when ~ = IK 1; here

(this is clear, at least set-theoretically), and our formula for the class v~ of
Vd will yield yet another proof of the Existence Theorem (2. 3) of Chapter VII
for special linear systems.
To derive the secant plane formula, we observe that set-theoretically the
locus V~ is the (d - r)th degeneracy locus of the map (2.4); we take this as
the definition of V:i as an analytic space (note that by relative duality and
Lemma (2.3) in Chapter IV the map IXIKI is the transpose of the Brill-Noether
homomorphism, proving in this case the scheme-theoretic equality V:i = C:i).
By Porteous' formula and the formula (2.5) for the Chern class of EL we
have the following preliminary

(4.1) Lemma. Suppose that s + 1 ~ d - r, and that V:i is either empty or of

pure dimension d - r(s + 1 - d + r). Then the fundamental class v:i of V:i
is equal to

vd = ( -1y(s+ 1-d+r)~r.s+ 1-d+r((1 + xt)d-n+g-1e-8t/(1 +xt))

= ~.+ 1-d+r,r((1 + xtt-d-g+ 1e8t/(1 +xt)).
 346 VIII. Enumerative Geometry of Curves

There now arises the major problem of simplifying the determinant in this
expression. We shall do this in three steps, the first of which is

Reduction A. The identity

is valid.

Proof To begin with, let us write out the determinant A.,b((1 + xt)<e9'i(l +xtl)
explicitly. We have

(1 + xt)<e9 •1< 1 +xt) = (1 + xtY L ~ (Jiti(1 + xt)- i

j J!

= I~ eiti(1 + xtr j
j J!

= L ~ (c ~ j)eixiti+ i
i,j]! l

= I (I ~ (cn-- !)eixn- i)tn,

n~o j~ol· J

so that

a-h+l(
L _
)(Ji
,
C-)·
Ia
( C - J') -X
(Ji a-j
.. j~O a- b + 1 - j j! X
a-h+l-j

j~o a - } J.
= det
a+h-1(
I C- ·
1 )(Ji
-xa+h-1-j I
a (c _ -
1.) (Ji a-}
.
i~o a+ b- 1 - j j! j~o a-j j!x

Now, we can try to perform the same reduction on this matrix that we
did in the special case a = g - d + r, b = r, c = g - d - 1. The problem
is that the success of that sequence of row and column operations in reducing
to a matrix of monomials depended on the original matrix having the form
above and starting in one corner with a monomial; since in the original case
we had c = g - d - 1, a - b + 1 = g - d + 1, so that the corner entry

a-b+ 1 ( C _ 1. ) (Ji a-b+ 1 ( C _ 1. ) (Ji

"
L. -xa-b+1-J. -_ "L. -x a-b+1-j
j~o a- b + 1 - j j! j~a-b a- b + 1 - j j!
e•-bx ea-b+ 1
= - + -:-----:;----:-::--:-
(a - b)! (a - b + 1)!
§4. The General Secant Plane Formula 347

had only two terms, we had only to add one extra row to the matrix to obtain
the desired form. To carry this out more generally, at least for c ::; a - b, we
would have to add a - b - c extra rows. At the conclusion of the reduction,
then, we would be left with an (a - c) x (a - c) matrix, with b columns
consisting of monomials in 8 and a - b - c consisting of monomials in x.
We would then expand not by cofactors along one column, but by minors
along a - b - c. This approach will, it may be seen, ultimately yield the
secant plane formula below; in the present general context, however, the
following reduction seems more straightforward.
The first step is essentially the same as before. The first column of the
matrix above is

(ac) xa + (ca-- 1)1 xa- 18 + ... + (c -0 a) -,

8
a!
2

1)
( a +c 2) xa + 2 + (ca -+ 1 X a+ 1 e + ... + (c -a0 - 2) (aea++25!.
2

Adding to each row from the second to the last x times the row above it
thus has the effect of replacing c by c + 1 in all but the first row, and similarly
in the remaining columns. Performing the same operation from the third
row on down then replaces c + 1 by c + 2 in the third to last rows; and
continuing in this fashion we arrive at a matrix with first column

(:)xa + (c-
a-
1)xa-1e
1
+ ...

( c + 1)xa+ 1 + (:)x"8 + ...

a+ 1

c + 2)xa+2
a+ 2
+ (c + 1)xa+1e
a+ 1
+ ...

( c + b-
a+b-1
1) a+b- 1 + (ca+b-2
X
+ b- 2) a+b- 28 + ...
X
 348 VIII. Enumerative Geometry of Curves

and (i + 1)st row

( c + ~)xa+i
a+z + (ca+z-1
+ ~- 1)xa+i-le + ...

( c :- i
a+ z- 1
)xa+i-1 + (ca++ z-
~- 1)xa+i-ze
2
+ ...

( c+i )xa-b+i+2 + ( c + i- 1 )xa-b+i+1e + ...

a-b+i+2 a-b+i+1

c+i
( a-b+i+1 )xa-b+i+1 + (ca-b+z
+ i- 1.)xa-b+ ie
+ ...
Now adding to each of the first b - 1 columns x times the column on its
right replaces c by c + 1 in all but the right-hand column; repeating this
for the first b - 2 columns, the first b - 3 columns, and so on, replaces c
by c + b - i in the ith column; we have then the matrix

L
a-b+1 ( C - )· ) ulJj a-b+1-j
j;O a - b + 1- j j! X

a+b-
~
£-
1 (c + 2b - 2 - j') ej a+b-1-J.
-X
a (c+b-1-)')0j a-J.
~
£- -X
j;o a +b - 1- j j! j;o a-j j!

By the relation(:)= ( -1r( -n +mm- 1). we may rewrite this as

I (
a-b+l n )ej a-b+l-j
j;o a- b +1- j j! y

I (
a+b-1 n ) ulJj a+b-1-j \...,
a
'-
( n )
. ~y
lJj
(7 a- j
j;o a+ b- 1 - j j! y j;o a - J J.

where n = a - c - b and y = - x. The point of the reduction so far is that

now the binomial coefficient part of each term in the entries depends only
on the power of y involved, while the straight factorial part depends only
on the power of 8; so that the entries are the coefficients of t in the product
of the power series
§4. The General Secant Plane Formula 349

and

I
i= 1
(~)yiti = (1 + yt)" = (1
I
- xt)a-c-b.

We have accordingly proved the formula

more generally, since the coefficients of e6110 +xt) were irrelevant to the
reduction, we have, for any power series f(t),

Reduction B. MacDonald's Formula

The last expression for Aa,b((1 + xt)<e611 <l+xt)) transposes readily into a
(special case of) a formula of MacDonald. To derive this, note that for any
L
power series f(t) = c;t; in one variable,

= L
O'ES&
sgn(O")ca+l-u(l) ... Ca+b-u(b)

where
A(t) = TI (t; - t)
i>j

= L sgn(O")ti<1>-tt2<ZJ-1 ... tb"<b)-t

aeS&

is the Vandermonde determinant in the variables t 1 , ... , tb. To obtain an

expression more symmetric in the variables t;, we can reorder the t; by a
permutation r E Sb and write

Aa,b(f) = sgn(r)[f1 f(t;) · A(t)] 1y1 +<<1H ... tgb+<<bl-1

= sgn(r)[f1 f(t;) · L\(t) · t~-•Ol · · · tg-•<hlJmr,)a+b-1·
350 VIII. Enumerative Geometry of Curves

Summing this over all r E Sb and noting that if r'(IX) = b +1- r(IX) then
sgn(r') = sb · sgn(r), where sb = ( -1)b(b- 1 l 12 , we infer that

In our present circumstance, this yields the formula

L1a.b((1 + xt)cet6/(1+xt)) = L1a,b((1- xtt-c-bet6)

= s~
b•
[n
i; 1
(1 - Xt;t-c-be6It'i1(t?] .
(It ..... lb)" + b- 1

Since L1(t) 2 contains a sum over all permutations, the expression on the
right in Reduction B does not represent an essential simplification over the
original determinant. However, in some cases the formula is more manage-
able; for example, let us prove the following result (cf. MacDonald [1]):

(4.2) Proposition. The virtual number of divisors of degree d imposing only

d - r conditions on a given linear series g~ and containing d - r(s + 1 - d + r)
given points is

(s +1- d + r)!

X [+:.crr(l + t;)n-g-s(l + L t;)9l1(t)2l, .. ·ts+l-d+r)s+d+2r.

Proof. Recalling that x E H 2 ( Ca, 7L.) is the class of Ca_ 1 + p c Ca for any
p E C, we must evaluate the cup-product

Using (4.1) and the two reductions above we find

v:i = 8r1
r.
[n
i=l
(1 _ xt;)g-n+se<Itil6t1(t)2]
(l!'''lr)s-d+2r
.

Since

if IX :::;; d,

if IX> d,
§4. The General Secant Plane Formula 351

where ~ is the positively oriented generator of H 2 a(Ca, Z), to evaluate

the cup-product above we may replace x by 1 and, after expanding e<IrilB,
replace (p by (!)ad when rx ::::; d and by 0when rx > d. When this is done
e<Iti)8 = L (L t;)~e~
~ rx!

~to (!)(L t;)~

=(1 + L t;) 9 (modulo(t 1 , ••• , t,)a+ 1 ).

Substituting in the expression above gives the first expression of (4.2). The
second one is obtained similarly.

Example. The simplest (non-trivial) case of the proposition we just proved

is to compute the number of double points of a plane curve of degree n;
that is, divisors of degree 2 imposing only one condition on a given two-
dimensional linear series. Here we take d = 2, r = 1, and s = 2; trivially
L1(t) = 1, and hence the number of double points is

[(1 - t)g+Z-n(l + t) 9],2 = (g + ~- n) - g(g + 2- n) + (~)

(n - 1)(n- 2)
2 - g,
as expected.

Example. To generalize the above example, we find the number vm of(m + 1)-
secant (m - 1)-planes to a curve C c IP 2 m of degree n. Here r = 1, s = 2m,
and d = m + 1, and plugging in we have
Vm = [(1 - t)9 + Zm- 1 (1 + t)9],m+ 1

= mf(-1Y(g
~~o
+2m-
rx
n)(m+1-rx
g )·

It may be an interesting problem to determine when this number is

positive, negative or zero.

Example. The next simplest example is the formula for the number of 4-
secant lines to a space curve C c IP 3 of degree n. Here we take d = 4, r = 2,
and s = 3; then L1(t) 2 = (t 2 - t 1 ) 2 , and consequently the number is

- [((1 - t 1 )(1 - t 2 ))9 + 3 -n(l + t 1 + tz)9(d - 2t1 tz + d)]rrr~

(n - 2)(n - 3) 2 (n - 4) g(n 2 - 7n + 13 - g)
12 2
352 VIII. Enumerative Geometry of Curves

We now return to the derivation of the secant plane formula. For the
third stage in the reduction we will use the following additional notational
convention

J.l(a, b, m,
. /3) = (ma++ i-
1'
i- 1) (a+ i- /3)!
f3 (a+ b- /3)!(/3- 1)!"
We have then the
Reduction C.
b
L1a, b((1 + xt)ne'o) = '\'
L... A(/3)2
il f1 J.la,
( b,n,z,;
. f3 )
l,;,p,<· .. <Pb-5.a+b i=l

Proof. To simplify the determinant

~•• 1,((1 + xt)"e0')
I ( n )f!_xa-j
j=o a - j j!
I
a+h-I(

j=O a+ b- 1 -
n )Oi_
j j! X
a+b-1-j

= det

a-I+I (a- b +n 1 - j )~j! a-h+I-j

L• ( n )(}i .
-x•- 1
j=O X i=O a- j j!
we consider the defining expansion of the determinant
b
det A = L sgn( o") f1 ai,
aESb i=l
a(i)

and separate out those terms in the b-fold products f1 ai,a(i) which take
from the polynomial ai, a<i>(x, (}) the term involving ea•. The determinant
may in this way be expressed as a sum

L
cq, ... ,ab
det Aa,, ... ,a••

where

( n ) 8"' a+b-1-cz 1
a+ b- 1- ()(' (l(,!x

( n ) (}"• a-h+!-••
a- b + 1- ()(" rx,!x
Setting IX = (1X 1 , ••• , 1Xb) we have then
§4. The General Secant Plane Formula 353

where

To evaluate c:5n;k,, ... ,k.• add to each column of the above matrix (except the
first) the column on its left; repeat the process for all but the first two columns,
and so on. We have then

( n+b-1)
k1 + b- 1

(k1 + b- 1)!
Ti
b (n +i-1)! d
= et
;= 1 (n-k;)!
1 1
kb! (kb +b- 1)!

b ((n + i -k.)'1) ') .fl

.fl (n -
b ( 1 )
1)'· A(kb, ... 'k1).
=
•=1 , · •=1 ,+ b -
(k.

Here, by the expression fl ((n + i - 1) !j(n - k;) !) we mean the product

TI ((n + i - 1) ..... (n - ka(i) + 1)), where 0" E sb and ka(1) 2: ka(2) 2: ... ka(b)•
if k.,(i) + i > 0 for all i; and 0 if k.,u> + i < 0 for any i (the need for this
cumbersome convention will disappear shortly). We have then

det A =
a
rl (
i=t (n- a+ rx;
(n +i - 1)! ) 1
+ i - 1)! (a+ b- rx;- i)! rx;!
X A(rxl + 1, !Xz + 2, ... ' (Xb + b)(Jf.a;Xab-L,a;
= det Bp = rl (
;= 1
(n- i- 1)! 1
(n- a+ /3;- 1)! (a+ b- /3;)!({3;- i)!
)

X A(/3 1, •.. , {3b). ()L(P;-i)Xab-L,(P;-i)'

354 VIII. Enumerative Geometry of Curves

where /3; = rx; + i. The reason for the shift from the index set rx to the index set
f3 is this: for each choice of /3 1 < /3 2 < · · · < f3b and a E Sb, we may set

noting that

11(/3") = sgn( a )11(/3)

we can sum the quantities det Bp over all rearrangements of f3 to obtain

L
aES&
det Bpa = TI (
i= 1
(n + i - 1)! . 1
(n - a + /3; - 1)! (a -r b - /3;)!
)11(/3). ()L(fi,-i)xab-L.(p,-i)

1
x L
<TESb
sgn(a)
Il (f3a(i)-
.
!).
1•

The sum in this expression may then be evaluated as

1 1
(/31 - 1)! (/31 -b)!
11(/3 b f3 b)
= det
Il (/3;- 1)!
0 0 0 '

1 1
(f3b- 1)! (f3b- b)!

and we have

11a,b((1 + xt)"e8 r) = L 11(f3) 2

n
1 ,;p, <liz<··· <{i&5.a+b

X (n + i- 1)!
i= 1 (n - a + /3; - 1)! (a + b - f3;)! (/3; - 1)!

where again the quotient (n + i - 1)!/(n- a+ /3;- 1)! is taken to be the

product (n + i - 1) · ... · (n - a + /3;) (note that, now that the f3;'s are
ordered, we have i - 1 ~ /3; - a always). If we wish to avoid this convention,
we may rewrite this last formula as

x (a+ i - /3;)! ()L.<P,-i)xab-L.<P<-il. Q.E.D.

(a+ b- /3;)! (/3;- 1)!
§4. The General Secant Plane Formula 355

We now apply Reductions A and C to the formula for v~ given by Lemma

(4.1). Reduction A gives

v~ = 11k,r((1 - xt)-"e 16)

= ( -l)'kl1,,k((1 - xt)Je- 16 ),

where

6 = n- g- s,
k= s + 1- d + r,
while Reduction C yields the formulas

I TI f.l.(r, k, 6, i, f3J()Y:.<tJ,- i)xrk- I.<tJ,-

k
(4.3) v~ = 11(/3)2 i)
L5,fJ,< .. ·<iJk:Sk+r i=l

k
I 11(/3? TI f.l.(k, r, -6, i, {3Jer.<tJ,-i)( -xyk-r.<tJ,-i>.
-1:5/J!< .. ·</J,:Sk+r i=l

This formula is, clearly, a mess. Nonetheless, it does represent a substantial

simplification of (4.1 ). For example it has C: k) terms rather than the
roughly r! k' terms involved in the expansion of (4.1). Most importantly,
(4.3) displays much more clearly the qualitative behavior of the class v~:
for example, we notice at once that in case 6 = n - g - s :2: 0, all terms in
the first expression of (4.3) have a positive sign. To motivate this condition,
recall that our linear system !!fi = PV is a g~ corresponding to a subspace
V c H 0 (C, L), and by the Riemann-Roch theorem we have

s =dim PV ~dim ILl = n- g + h 1(C, L);

thus the following conditions are equivalent:

If we define the relative deficiency of a linear series of dimensions and degree n

on a genus g curve to be 6 = n - g - s, then the positivity of the terms in
(4.3) has the following implication:

If the series !!fi has non-negative relative deficiency and if r(s + 1- d + r) ~ d,

then the cycle v~ is non-empty.
356 VIII. Enumerative Geometry of Curves

Somewhat more geometrically:

If C c ps has genus g and degree n, then C possesses d-secant (d - r - I)-

planes provided that

n ~ s +g and r(s + 1- d + r) s d.

Unfortunately, the hypothesis n - g - s ~ 0 excludes the most interesting

case, namely when ffi is complete and is a special linear series. The difficulty
here is not merely apparent but is real: the conclusion above is not always
true, since, as we shall see by example, there are cases in which v:i is zero
and still others where it is non-zero but is not the class of an effective cycle
(what can be concluded in these cases is that V:i is non-empty and has the
wrong dimension). From a technical standpoint the difficulty is that the
coefficients of (4.3), viewed as a polynomial in x, may have alternating signs,
and since there are relations among x and 8 in H*(Cd, Q) it is not possible
to conclude that v:i #- 0.
In borderline cases, however, we are able to extract information from
(4.3):

Proposition. lfs + 1 ~ d- r ~ 0 and r(s + 1- d + r) s d, and ifs = n-

g + 1 {i.e., the series ffi has relative deficiency -1) then v:i = 0 if and only if
g < (r + l)(s- d + r + l).Inparticular,ifg ~ (r + l)(s- d + r + l),then
V:i #- 0; while if g < (r + l)(s- d + r + 1) then Vd is either empty or of
the wrong dimension.

We observe that the condition {J = -1 is satisfied when

ffi=IKI

and that in this case the inequality g ~ (r + l)(s - d + r + 1) amounts to

saying that p ~ 0, so that this proposition implies the existence theorem
for special divisors (Theorem (2.3) in Chapter VII).

Proof of the Proposition. In cased - r = s + 1, V:i = Cd and there is nothing

to prove. Thus we may assume that s- d + r ~ 0. In the sum (4.3) we
must have

1s /3 1 < · · · < /3, s s + 1 - d + 2r,

and also, since {J = -1, we must have

0 s {3; - (s + 1- d + r)
for all i. The only index sets yielding non-zero terms are

pu> = (s + 1 - d + r, ... , ~ ... , s + 1 - d + 2r)

 §4. The General Secant Plane Formula 357

where 0 ~ j ~ r. Then we have

A(f3(j))2 = (P~=O ~! )2,

Jl(r-J)!
' (g+s-n+i-1)! _. 1
)] (g - n + d - r - 2 + f31i>)! - 1 .,
' 1 (r - j)!
TI (s + 1 - d + 2r- f3U>)
i=l ; ·' = TI'i=O z.
.''
nr . 1

; - 1)'.
i=l (f3u>
= (s - d
TI'i=l cs- d +r+z.
+ r + j)!
·)',

so that by (4.3)

v' = flr "I

z. Ir ( s- d + r + J")I. ( -x)iw<s+l-d+r>-i.
d i=o(s-d+r+i)!i=O j!(r-j)!

Since eg+ 1 = 0, v~ = 0 when g < r(s - d + r) (which, incidentally, implies

that g < (r + 1)(s + 1 - d + r)). Therefore we may assume that g z
r(s - d + r). To see when the expression for v~ is non-zero we multiply by
the monomial

(}"xd-r(s+ l-d+r)-o
0 ~ rx ~ d - r(s + 1- d + r)
of complementary degree. Ignoring the non-zero factor

rr
i=O (s - d
i!
+r+
~
i)! '

where ~is the positive generator of H 2d(Cd, Z), this gives

~ (-1)i(s -d+r+j)! (
L...
g )crs-
( d +r) -J+r+rx.
. )'
i=O j!(r-j)! r(s-d+r)-j+r+rx

(s-d+r)!g! ~ .(s-d+r+j)(g-r(s-d+r)-rx)
= L... ( -1)1
(g-r(s-d+r)-rx)!i=O j r-j

= (s- d + r)! g! ~ [
(g- r(s- d + r)- rx)! i~O (1 + t)'
1 J
d+r+ 1 11
.[(1 + t)g-r(s-d+r>-•],r-j

(s- d + r)! g! [(1 + t)g-a-(r+ l)(s-d+r+ l)+r],r

(g - r(s- d + r)- rx)!

= __(.~-=-~~ r)~!-
(g- r(s- d + r)- rx)!
(g- (r + l)(s + 1- d + r)- rx + r)
r ·
358 VIII. Enumerative Geometry of Curves

This is zero for all rJ. such that

0 ::::;; rJ. ::::;; d - r(s + 1- d + r)

exactly when

g < (r + 1)(s - d + r + 1),

so that we know that v~ is not zero in case g ;;::: (r + 1)(s - d + r + 1). In

fact, this computation establishes the proposition once we realize that the
class v~ is a polynomial in x and () and that the intersection pairing is non-
degenerate on the subring of H*(Cd, (ll) generated by X and e.

§5. Diagonals in the Symmetric Product

Our last application of the topology of the symmetric products to enumera-

tive geometry will be concerned with divisors with multiplicities, and to set
this up we define the diagonal mappings for Cd:

to be given by

where a = (a 1 , •.. , ak) and I~= 1 a;n; = d.

To state the result we shall retain the notations used previously in this
chapter; in addition we set
k
Ra(t) = 1 + I aft;,
i= 1

Pa(t) = 1 + I a;t;,
i= 1

The main result is the

(5.1) Proposition. The image, via the diagonal mapping cPa, of the fundamental
claSS of Cn, X · · · X Cnk is
§5. Diagonals in the Symmetric Product 359

As an application of (5.1) we will be able to derive the classical

De Jonquieres' Formula. Let a 1 , ... , ak> n1o ... , nk> d be positive integers. Let
r be a non-negative integer. Suppose the a;'s are distinct, L: n; = d - r, and
L: a;n; = d. Set a = (al> ... , ak), n = (n 1 , .•. , nk). Then the virtual number
Jla," of divisors having n; points of multiplicity a; in a given linear series of
dimension r and degree d is

The proof of the proposition will be given in several steps, and will be
based on the following

Lemma. Let C be a curve with general moduli. The subring of H*(Ca, IIJ)
generated by the fundamental classes of algebraic cycles on cd is generated
by X and 8.

A few words of explanation are in order. The sentence, "If C is a curve

(of genus g) with general moduli, then ... " is shorthand for "There are
denumerably many proper subvarieties of the moduli space of genus g
curves such that, if the point corresponding to the curve C does not belong
to any one of them, then ... ".

Proof of the Lemma. When d;::: 2g - 1 we have seen in Chapter VII (cf.
Proposition (2.1)) that there are a vector bundle Ed on J(C) and a biregular
morphism Cd :::::-. IP'(Ed) such that the diagram

is commutative. Letting A*(Ca), A*(J(C)) be the Chow rings of Cd = IP'(Ed)

and J(C), respectively, we have

A*(Ca) = A*(J(C))[x],

where x is the cycle corresponding to X c Cd and x satisfies the minimal

equation (4.2) of Chapter VII. Of course, one should now read the Chern
classes in this formula as having values in the Chow ring of J(C). The lemma
follows, when d ;::: 2g - 1, from this remark together with Poincare's
formula and the assertion that, when C has general moduli, the algebraic
part of H 2 q(J(C), IIJ) is generated by eq (this will be proved in Chapter X).
360 VIII. Enumerative Geometry of Curves

Assuming the result for Cd+ 1 we will deduce it for Cd. Let Z c Cd be an
analytic subvariety; we want to express the fundamental class z of Z as a
polynomial in X and e. For this we define

by

t/JiD,p) = D + p,
and we denote by X q the image of the inclusion

given by

j(D) = D + q.
If we define the subvariety Zq c Cd- 1 by

Zq = {D- q: DE Z and D- q;;::.: 0},

then

(this means that a point in Xq n {D + p: DE Z and p E C} arises either when

qED or when p = q). Passing to fundamental classes, this gives

z =j*[t/JiZ X C)]- [t/Jd-l(Zq X C)].

By our induction assumption,j*[t/JiZ x C)] is a polynomial in x and e.

Repeating the argument with two points q 1 , q 2 where q = q 1 and with
tj;d_ 1 (Zq, x C) replacing Z, we are reduced to establishing the result for
t/Jd- 1 (t/Jd- 2 (Zq,q 2 x C) x C) where

Continuing in this way we are eventually reduced to establishing the result

for the image of the standard inclusion

given by

D--> D + ql + qz + ... + qd-k•

The fundamental class of this cycle is xd-k. Q.E.D.
§5. Diagonals in the Symmetric Product 361

By the lemma the class on the left-hand side of the formula in (5.1) is a
polynomial in x, 8, and to determine the coefficients we will evaluate the
pull-backs cfJ:cxa()P), r:t. + p = e, in H 2 e(en, X •.• X enk' Z).
We begin by showing that

(5.2) _;,*((J•)[e nt X ••• X

'1-'a
enk] = (g)e n I(e !)2 I
nk. ;
Il a;2n·'.
1. · •·

When e > g, both sides are zero; when e ::::; g, the verification is straight-
forward but requires introducing some notation. As in Section 2 of this
chapter we let <5 1 , ... , <5 29 be a symplectic basis for

and we denote by<)~), x<i), ... the pull-backs of <>a, x, ... from the ith factor of
en, X •.. X en.· We also set

'11 = '1;, .. 0 '1;,,

where I = (i 1, ••• , i1), 1 ::::; i 1 < · · · < i1 ::::; g, and recall that

Note that

()l = l! L '1 I'

(5.3) { III=l
'1 1 is the positively oriented generator of H 2a(ea, Z) in case 111 =d.

The reason for the second statement is that on the ordinary cartesian product
ea, '7 1 is d! times the generator of H 2a(ea, Z), and the mapping ed ~ ed has
degree d!. Moreover, any other 2d-fold product of the <>a's is zero.
For the multiplication map

P.a,: J(e) ~ J(e)

given by

P.a.(u) = a;u,

we have
362 VIII. Enumerative Geometry of Curves

(think of J( C) as IR 29/1' 29 and of ba as dxa). It follows that

rl.*(t:5 a) = "a.c:)<il
o/a ~ t a'
i

rl. *( ) - " ~(i) ~(j)

'f'a 1'/a - L, aiajua Ug+a•
i, j

Even though this last expression is a little messy, by the last remark in the
preceding paragraph in the final evaluation we may discard all the terms
such that i # j in the above. Thus, when I = (i 1 , ••• , ie), e =In;, we may
write

¢:(1'/I) = JJ 1( ~ a~I'/~~J
This, together with (5.3), gives

where }' is the positively oriented generator of H 2 e(cn, X ••• X cnk• Z).
When combined with the first equation in (5.3) this proves (5.2).
For the next step, by a similar argument we will show that, when p + q = e

Again the verification is straightforward. We have

ri.*x = "a.x<i)
o/a L... t '
i

and then

Since (x<l)r· ... (x(k)rk E H*(IICn,, <Q>) is dually represented by cn,-m, X ..•
X Cnk-mk' we may Combine this with (5.3) to have

¢:(xq8P)[Cn, x · · · x CnJ

I I q! I
m,+···+mk=q ml .... mk. i
TI
a;"'¢:(eP)[Cn,-m, X .•• X cnk-mJ

= p!(g)[(1 +
p
l.=afti)P(l
'
+ l.=aiti)q]
' t71 ···tf:k
,

where p +q= Ii n;. This establishes (5.4).

§5. Diagonals in the Symmetric Product 363

Completion of the Proof of Proposition (5.1). The expression on the right-

hand side of the identity to be proved is

I [( -1YPa(tY- 9 Ra(t)9 -a(Ra(t)- P~(t)Yxd-e-aeajct!]t?'···t~k·

•
Multiplying this by 8Pxq and evaluating on the fundamental class of Cd gives

Comparing once more with (5.4) gives the result.

Proof of DeJonquii~res' Formula. We must derive the formula for the virtual
number of divisors

degDi = ni

in a linear series f!fi of degree d and dimension r, assuming that the dimensions
are such that there are a finite number of such divisors to be expected; i.e.,
that

I ni = d- r.
i

To do this we evaluate on f!fi ~ IP' the formula (5.1) for the fundamental
class of c/Ja(Cn, x · · · x CnJ· Since

81f» = 0,
{
x'[f!fi] = 1,

the result is

Example. We will compute the number r(d, g) oftritangent planes to a space

curve C c IP 3 of degree d and genus g. This is obtained by taking

n 2 = d- 6,
364 VIII. Enumerative Geometry of Curves

in DeJonquieres' formula, which gives

r(d, g)= s(d; 3) + s(~) + 4g(d- 5)(d + g- 5).

Note that for a canonical curve this is

r(6, 4) = 120,
which is the number of odd theta characteristics on a curve of genus 4.

Example. By combining the diagonal formula with the secant-plane formula

it is possible to determine the virtual number of k-planes meeting a given
curve C c IP'' with given multiplicities of intersection. The simplest case of
this is finding the number T(d, g) of tangential trisecants to a curve C c IP' 3
of degree d and genus g (a tangential trisecant is a tangent line meeting the
curve in one additional point). To calculate T(d, g) we intersect, in C 3 , the
class of the locus of divisors D = p 1 + p2 + p 3 that fail to impose indepen-
dent conditions on the hyperplane series g~ with the diagonal ¢2.1(C x C).
The answer is

T(d, g) = 2(d - 2)(d - 3) + 2g(d - 6).

For a canonical curve this is 2 · 12, in accordance with the Riemann-Hurwitz

formula for the number of branch points of each of the g~'s.

Bibliographical Notes

The original proof of the Grothendieck-Riemann-Roch formula appears

in Borel-Serre [1]; a more recent version may be found in Fulton [2].
Original attempts to derive enumerative formulas for projective curves did
not use it, but rather relied on Schubert calculus (as in Giambelli [1])
or on the theory of correspondences, as in MacDonald [1]. The use of
Riemann-Roch formulas was first suggested in this setting by Schwarzen-
berger. Similarly, earlier proofs of formulas such as DeJonquieres' (as, for
example, in Coolidge [1]) did not use the topology of the symmetric prod-
uct; the cohomology of Cd and its use in establishing formulas for curves
appears first in MacDonald [2] and Mattuck [1].

Exercises
A. Secant Planes to Canonical Curves

In this sequence of exercises C c IP'9 - 1 will be a smooth canonical curve with d-secant
variety
:EdciP'g-1
Exercises 365

defined as follows: In cd X G(d, g) we let rd be the closure of the subvariety {(D, A):
¢K(D) = IP A}, fd the inverse image of rd in the projectivized universal subbundle IPS
over G(d, g), and :Ed the image of fd under the natural map IPS--> IP9 - 1 .

A-1. Let E~--> Cd be the bundle defined in Section 2 of Chapter VIII and <!!(1) the
tautological line bundle over IP(E~). Show that :Ed is the image of IP(E~) under the
map to IP9 - 1 given by the sections of <!!(1) (in this exercise we are only interested
in equality of varieties, not of schemes).

A-2. Show that, ford < g/2, the map

is birational (use the uniform position lemma).

A-3. Using the preceding two problems, show that

deg :Ed= Id (g - d - .
1- i) (g). , d < gj2.
i=O d- l I

A-4. Alternatively, show that deg :Ed is the number of divisors of degree d failing to
impose independent conditions on a general (2d- i)-dimensional subspace
V c H 0 (C, K). Using the secant plane formula show that this number is

(1 + t)9
deg :Ed = [ (1 - t)g- 2d ,, .
J
(It may be reassuring to verify that the answers obtained in these two problems
do in fact coincide.)

B. Weierstrass Pairs

Definition. Let C be a smooth curve of genus g and IX, {3 a pair of positive integers with
IX+ f3 = g- 1. We say that p, q e Cis a Weierstrass pair of points (of type (IX, {3)) if
r(1Xp + f3q) ;;:>: 1.

B-1. Show that not every pair of points p, q e Cis a Weierstrass pair.

B-2. Show that a non-hyperelliptic curve of genus 4 can have at most finitely many
Weierstrass pairs of type (1, 2), and use the result at the end of Chapter VIII to
count them.

B-3. Show that a hi-elliptic curve of genus 5 may have oo 1 Weierstrass pairs of type
(2, 2). Can a non-trigonal curve of genus 5 have oo 1 Weierstrass pairs of type (3, 1)?

B-4. By combining (5.3) and the theorem at the end of Chapter VII find an enumerative
formula for the number (assumed finite) of Weierstrass pairs of type (IX, {3).
(Answer: g1X{3((g - l)IX/3 - IX - {3).)
366 VIII. Enumerative Geometry of Curves

B-5. Expressing the condition r(r:xp + {3q) ~ 1 determinantally, find an enumerative

formula for the number (assumed finite) of Weierstrass pairs of type (r:x, {3).
(Warning: Watch out for the diagonal inC x C.)

For p, q E C the Weierstrass semigroup H(p, q) c N x N is defined by

H(p, q) = {(m, n): there exists f E .A( C) with (j) 00 = mp + nq}.

B-6. Show that H(p, q) c N x N is a sub-semigroup with index

card(N x N- H(p, q)) ~ (g +

2
2) - 1.

(Hint: Show that for any fixed r:x,

card{/3:/3 ~ 1, (r:x, {3)¢H(p, q)} ~ r(r:xp)- (r:x- g),

and use Exercise E-6 of Chapter 1.).

B-7. Show that if p and q are general, then equality holds in ( *); that if either p or q
is a Weierstrass point for C then equality fails in(*); and that equality may fail
in (*) even if neither p nor q is a Weierstrass point of C.

C. Miscellany

C-1. Let C, B 1 , B 2 be curves of respective genera g, g 1 , g 2 • Assume that

i = 1, 2,

is a d;-sheeted mapping such that

is birational to its image. Show that

g:::; (dl- 1)(d2- 1) + dlgl + d2g2.

(Suggestion: Estimate the self-intersection of 4J(C) in B 1 x B 2 by taking the
Kiinneth decomposition of its class and applying the Hodge index theorem
(cf. page 126 of Griffiths-Harris [1]).)

C-2. As applications of the preceding exercise observe that a curve of genus g ~ 3

cannot be both hyperelliptic and hi-elliptic, that a curve of genus g ~ 5 cannot
be both trigonal and hi-elliptic; and that a curve of genus g ~ 6 cannot be hi-
elliptic in more than one way.

C-3. Let C be a smooth curve of genus g ~ 2. By a Chern class argument show that the
bundle E 29 _ 1 -+ J(C) introduced in Section 2 of Chapter VII does not have any
sub-line bundles. Conclude that there does not exist a section ofu:Cd-+ J(C) for
d::;2g-1.
Exercises 367

Remark. On the other hand, it is a general fact about projective bundles that, for
d;?: 2g, there is always a section of u: Cd---> J(C).

C-4. Let V denote the vector space of homogeneous polynomials f(x 0 , x 1 ) of degree d,
and for any collection of positive integers n 1 , ..• , nk with L:
i · n; = d, let V, c V
be the variety of polynomials having n; roots of multiplicity i. Using DeJonquieres'
formula, show that

C-5. Suppose that f: C ---> B is a branched covering of a curve C of genus g ;?: 2 over a
curve B of genus h ;?: 1. Let A c J(C) be the inverse image of J(B) under the pull-
back map. Show that the fundamental class of A is not a rational multiple of (} 9 -h.

C-6. From the preceding exercise and the theorem on global monodromy proved in
Chapter X conclude that a general curve of genus g ;?: 2 does not admit a non-
constant map to a curve of positive genus.

D. Push-Pull Formulas for Symmetric Products

Let C be a smooth curve of genus g. Given an algebraic cycle Z in Cd we define the

following associated cycles

A(Z) = {D + p: DE Z}
= {EECd+I:E- D;?: 0 for some DEZ} c cd+l;

B(Z) = {D - p: DE Z and D - p;?: 0}

= {EECd_ 1 :D- E;?: OforsomeDEZ} c Cd-l·

In both of these we have only described the cycle set-theoretically, and it is understood
that multiplicities are to be attached appropriately, e.g., the multiplicity of a component
W of A(Z) is card {DE Z: E-D;?: 0} where Eisa general point of W. Similarly, we
define cycles
Ak(Z) = {E E cd+k: E-D;?: 0 for some DE Z},

Bk(Z) = {E E cd-k: D-E;?: 0 for some DE Z}.

The assignments Z---> Ak(Z), Z---> Bk(Z) induce maps on the analytic part of the rational
cohomology

Ak: H;;:'(Cd, CD)---> H;;:'+ 2 k(Cd+k, CD),

Bk: H;;:'(Cd, CD)---> H;;:'- 2 k(Cd-k, CD).

(Recall that for a curve C with general moduli, H:"(Cd, Q) is generated by x and 6.)
In this series of exercises we will derive formulas for Ak> Bk and will give some
applications of these formulas.
368 VIII. Enumerative Geometry of Curves

D-1. Define

by

'I'(D, p) = (D, D + p),

and let ljJ denote the fundamental class of the image. Show that for z E H!n(Cd, Q)

Where 71:1, 1tz are the projections of Cd X Cd+ 1 tO Cd, Cd+ 1·

Remark. This involves relating cup-product to intersection via Poincare duality.

An alternate approach to these exercises would be to define A, B by the above formulas,
and then interpret these maps geometrically by our defining relations.

A(z)·w = z·B(w).

D-3. Show that A, B take the subring of EfJd H!.(Cd, Q) generated by x, 0 to itself
(cf. the lemma in Section 5 of Chapter VIII).

D-4. Denote again by x, 0 the pull-backs to Cd x C of x, 0 on Cd, and by 11 the pull-back

to Cd x C of the class of a point on C. For 'I' as in D-1 show that
'l'*nfx = x,

'l'*nfO = 0,

'l'*n!x = x + 11,
'l'*n!O = 0 + g ·11 + y,

where y is the class in Section 2 of Chapter VIII.

D-5. Continuing the preceding exercise, if ex + f3 + A. + .5 = d + 1 show that

(B(x"OP) · x;.06 )c. = (njx;.0 6 · n! x•oP · 1/J)c. x ca.,

= ('l'*njx 1·06 · 'l'*n! x•OP)c. x c

= cx(x•+A-1. OP+ii)c. + fJ(g- f3 + 1)(x"H. OPH-l)c.
(recall that l = - 211fJ).

D-6. Conclude from D-3 and D-5 that

Exercises 369

D-7. Similarly, show that when Ll + P+ i1. + b = d +1

(A(x•eP). xAe~)c._' = il.(x•+AePH)c.+'
+ fJ(g- {J + l)(Xa+H I, ePH-!)cd+I'

From this conclude that

D-8. Show that

and using the preceding two exercises conclude that

D-9. Suppose that n ;::: 2g- 1 and let !!fi c en beag~. Denoting by rd(fifi) c ed the
cycle of divisors of degree d subordinate to the g~, rederive the formula (3.20)
of Chapter VIII for the fundamental class yififi) of rififi) by noting that

eg
[fifi] = xn-g-r l'
g.

rififi) = An-i!!fi),

and applying the preceding exercise.

The remaining exercises will give applications of the above formulas, and for these
we will use the following notations:

e--> His ann-sheeted covering of e over a curve H of genus h;

f 0 = {n- 1 q: qEH} c en;

fd = {n- 1q + D: qEH and DEed}

= Aifo);
~d = u(fd)
= u(f 0 ) + J.t-d(e);
if 0 • ad, and ad are the fundamental classes off 0 , fd, and ~d;
2 c en is the main diagonal. i.e.,

2 = {p, + ··· + Pn-2 + 2p:p,, ... ,pn_ 2 ,pEe};and

~ is the fundamental class of 2.
370 VIII. Enumerative Geometry of Curves

D-10. Show that

(fl0 ·x)=1,

(fl 0 · ~) = 2g - 2 - n(2h - 2),

and use (5.1) in Chapter VIII to conclude that

D-11. Use D-8 and the relation

(fld. ed+ t)ca+n = (Aiflo). ed+ t)c.+n

=(flo. BiOd+!))cn

to show that

(g- 1)!
(g _ d _ 1)! (d + 1)-nh.

E. Reducibility of l¥g_ 1 n (l¥g_ 1 + u) (II)

The following sequence of exercises will establish the following refinement of Theorem
( *) in exercise batch C of Chapter VI.

(**) Theorem. Let C be a non-hyperelliptic curve of genus g ;;:: 4 and v E J(C). Then the
intersection

w-;,_t(C) n (w-;,_ 1 (C) + v)

is reducible if and only if either:

(a) v = u(p- q) where p, q E C; or

(b) C is a two-sheeted covering n: C --> E of an elliptic curve and v E n* J(E)

(cf. Wei! [1] and Mumford [3]).

These exercises will use the following general result:

Let X be a smooth irreducible variety of dimension at least three and V1, V2 ample divisors
on X. Then V1 n V2 is connected in codimension one.

Sketch of Proof. For n ~ 0 we may assume that In V1 I, In V2 1 are very ample. If D 1 E

In V1 I, D 2 E In V2 are smooth, then D 1 n D 2 is connected by the Lefschetz theorem. Thus
1

is the limit of connected varieties, and is therefore connected.

Exercises 371

To see that it is connected in codimension 1, we intersect V1 , V2 with a smooth

threefold X' c X and use the same argument to conclude that V1 11 V2 11 X' is connected.

E-1. We have already seen (cf. Exercise C-6 in Chapter VI) that exception (a) in Theorem
( **) actually occurs, and this exercise is to show that (b) occurs. Specifically,
suppose n:: C--+ E is a two-sheeted cover of an elliptic curve and define

by

~ = { D: D - n:- 1 p ;;::: 0 for some p E E}

(thus, ~ consists of all effective divisors of degree g - 1 that contain a fiber of

n:: C--+ E). Show that

dim(u(~)) = g- 2,
and

for any Z' En:* J(E).

E-2. Let u*(u) be the fundamental class of u(~). Using Exercise D-11 above show that
the intersection number (u*(u) · 89 - 2 ) equals (g - 2)(g - 1)! and conclude that

for v En:* J(E). Hence, the intersection in the statement of Theorem (**)is reducible
in case (b).

E-3. Suppose now that C is a non-hyperelliptic curve of genus g and that

is reducible for some v i= 0. Using the connectedness result above show that some
pair of components E 1 , E 2 c a-;,_ 1 (C) 11 (»-;,_ 1 (C) + v) contains an irreducible
variety r of dimension g - 3.

E-4. Retaining the notations from the preceding exercise, let u(D) be a general point
of r, and write also
u(D) = u(F) + v,
where FE cg-1• Using Martens' theorem show that

r(D) = r(F) = 0,

and using the Riemann singularity theorem together with

372 VIII. Enumerative Geometry of Curves

show that

cPK(D) = cPK(F);

i.e., the' divisors D, F span the same hyperplane in IP' 9 - 1.

E-5. Continuing the notations of the preceding exercise, write

D = D' + G,
F = F' + G,

where all divisors are effective and supp(D') is disjoint from supp(F'). Show
that G =F 0.
(Hint: if G = 0 then by the preceding exercise

u(D- F)= v,

u(D + F) = u(Kc)

for all DE r where dim r = g - 3. From this obtain a contradiction when

dim r > 0; i.e., g ~ 4.

E-6. Assuming that deg D' = deg F' = d, on a sufficiently small open set U c r
define a map

by

n(D, F) = (D', F').

Using Clifford's theorem show that

dim cPK(D' + F') ~ d,

and conclude that a general fiber of n has dimension at most g - d - 2.

(Hint: Consider hyperplanes in IP'9 - 1 containing cPK(D' + F').)

E-7. Finally, using the preceding exercise show that dim(Im n) ~ d - 1, and using
exercise sequence H of Chapter VI conclude the proof of Theorem (** ).

F. Every Curve Has a Base-Point-Free g:_ 1

In this sequence of exercises we will prove the following result:

( *) Let C be a smooth, non-hyperelliptic curve of genus g 4. Then there exists a base-

0:-
~
point free 1 On C.
Exercises 373

F-1. Suppose that C is a non-hyperelliptic curve of genus g and l: is an irreducible

component of w;_,(C) such that, for a general u(D)el:, the complete linear
series ID has a base point. Show that one of the following holds (cf. the proof of
1

Mumford's theorem in Chapter IV):

(i) C c IP' 2 is a smooth plane quintic and

l: = W~(C) = {u(D +p - q): p, q E C},

where (!)c(D) = (!)c(l).

(ii) C is trigonal with g~ = IF I , and
l; = {u(F + p 1 + ... + p9 _ 4 ): p 1 , •.• , pg- 4 E C},
(iii) C ~ E is hi-elliptic and

l: = {n* J(E) + u(p 1 + · · · + p9 _ 5 ): p 1 , .. , p9 _ 5 E C}.

F-2. Assuming still that C is non-hyperelliptic, show that if g ~ 6 then not every
component of w:_,(C) is of the form (ii) or (iii) in the preceding exercise.
(Hint: Use the computation of intersection numbers in exercise batch D
above.)

F-3. Using these two exercises conclude the proof of (*).

Bibliography

Accola, R. D. M.
[1] On the number of automorphisms of a closed Riemann surface. Trans. Amer.
Math. Soc. 131 (1968), 398-408.
[2] Plane models for Riemann surfaces admitting certain half-canonical linear
series, part I, in Riemann Surfaces and Related Topics. Proceedings of the 1978
Stony Brook Conference (1. Kra and B. Maskit, eds.), Princeton University
Press, Princeton, 1981, pp. 7-20.
Andreotti, A.
[1] Recherches sur les surfaces algebriques irregulieres, Acad. Roy. Belg. Cl. Sci.
Mem. Col/. in 8°, 27 (1952), no 7.
[2] On a theorem o[Torelli, Amer. J. Math. 80 (1958), 801-828.
Andreotti, A. and Mayer, A.
[1] On period relations for abelian integrals on algebraic curves, Ann. Scuola
Norm. Sup. Pisa 21 (1967), 189-238.
Arbarello, E. and Cornalba, M.
[1] Su una congettura di Petri, Comment. Math. Helv. 56 (1981), 1-38.
Arbarello, E. and De Concini, C.
[1] On a set of equations characterizing Riemann matrices, Ann. of Math. 120
(1984), 119-140.
[2] An analytical translation of a criterion of Welters and its relation with the K-P
hierarchy, in Proceedings of the Conference "Week of Algebraic Geometry"
(Sitges, 1983), to appear in Lecture Notes in Mathematics.
Arbarello, E. and Harris, J.
[1] Canonical curves and quadrics of rank 4, Compositio Math. 43 (1981), 145-179.
Arbarello, E. and Sernesi, E.
[1] Petri's approach to the study of the ideal associated to a special divisor, Invent.
Math. 49 (1978), 99-119.
Artin, M.
[1] Theon!me de finitude pour un morphisme propre: dimension cohomologique
des schemas algebriques affines, in Theorie des topos et cohomologie etale des
schemas (SGA IV), tome 3, Lecture Notes in Mathematics 305, Springer-
Verlag, Berlin-Heidelberg-New York, 1973, pp. 145-167.
Babbage, D. W.
[1] A note on the quadrics through a canonical curve, J. London Math. Soc. 14
(1939), 310-315.
Baily, W.
[1] On the theory of ll-functions, the moduli of abelian varieties, and the moduli
of curves, Ann. of Math. 75 (1962), 342-381.
376 Bibliography

Beauville, A.
[1] Prym varieties and the Schottky problem, Invent. Math. 41 (1977), 149-196.
[2] L'application canonique pour les surfaces de type general, Invent. Math. 55
(1979), 121-140.
Bers, L.
[1] Uniformization, moduli, and Kleinian groups, Bull. London Math. Soc. 4
(1972), 257-300.
Bloch, S. and Gieseker, D.
[1] The positivity of the Chern classes of an ample vector bundle, Invent. Math.
12 (1971), 112-117.
Borel, A. and Serre, J. P.
[1] Le theon!me de Riemann-Roch, Bull. Soc. Math. France 86 (1958), 97-136.
Brill, A. and Noether, M.
[1] Uber die algebraischen Functionen und ihre Anwendungen in der Geometrie,
Math. Ann. 7 (1873), 269-310.
Castelnuovo, G.
[1] Ricerche di geometria sulle curve algebriche, Atti R. Accad. Sci. Torino 24
(1889), 196-223.
[2] Numero delle involuzioni razionali giacenti sopra una curva di dato genere,
Rendiconti R. Accad. Lincei (4) 5 (1889), 130-133.
Ciliberto, C.
[I] On a proof of Torelli's theorem, in Algebraic Geometry-Open Problems (C.
Ciliberto, F. Ghione, F. Orecchia, eds.), Lecture Notes in Mathematics 997,
Springer-Verlag, Berlin-Heidelberg-New York-Tokyo, 1983, pp. 113-123.
Clebsch, A.
[1] Zur Theorie der Riemann'schen Fltichen, Math. Ann. 6 (1873), 216-230.
Clemens, C. H. and Griffiths, P.
[1] The intermediate iacobian of the cubic threefold, Ann. of Math. 95 (1972),
281-356.
Clifford, W. K.
[1] On the classification ofloci, in Mathematical Papers, Macmillan, London, 1882.
Comessatti, A.
[1] Limiti di variabilita della dimensione e dell'ordine d'una g~ sopra una curva
di dato genere. Atti R. /st. Veneto Sci. Lett. Arti 74 (1914/15), 1685-1709.
[2] Sulle trasformazioni hermitiane della varieta di Jacobi, Atti R. Accad. Sci.
Torino 50 (1914-15), 439-455.
Coolidge, J. L.
[1] Algebraic Plane Curves, Oxford University Press, Oxford, 1931.
De Concini, C., Eisenbud, D. and Procesi, C.
[1] Hodge algebras, Asterisque 91 (1982).
De Concini, C. and Procesi, C.
[1] A characteristic-free approach to invariant theory, Adv. in Math. 21 (1976),
330-354.
Deligne, P. and Mumford, D. .
[1] The irreducibility of the space of curves of given genus, lnst. Hautes Etudes Sci.
Pub!. Math. 36 (1969), 75-110.
Donagi, R.
[1] The tetragonal construction, Bull. Amer. Math. Soc. (N.S.) 4 (1981), 181-185.
Bibliography 377

Doubilet, P., Rota, G.~c. and Stein, J.

[1] On the foundations of combinatorial theory IX, Studies in Applied Math. 53
(1974), 185-216.
Eagon, J. A. and Hochster, M.
[1] Cohen-Macaulay rings, invariant theory, and the generic perfection of deter-
minantalloci, Amer. J. Math. 93 (1971), 1020-1058.
Eisenbud, D. and Harris, J.
[1] Divisors on general curves and cuspidal rational curves, Invent. Math. 74
(1983), 371--418.
[2] A simpler proof of the Gieseker-Petri theorem on special divisors, Invent.
Math. 74 (1983), 269-280.
[3] Curves in Projective Space, Les Presses de l'Universite de Montreal, Montreal,
1982.
Enriques, F.
[1] Sulle curve canoniche di genere p dello spazio a p -1 dimensioni, Rend. Accad.
Sci. /st. Bologna 23 (1919), 80-82.
Enriques, F. and Chisini, 0.
[1] Teoria Geometrica delle Equazioni e delle Funzioni Afqebriche, 4 vols., Zanichelli,
Bologna, 1915, 1918, 1924, 1934.
Farkas, H. M.
[1] Special divisors and analytic subloci of the Teichmuller space, Amer J. Math.
88 (1966), 881-901.
Farkas, H. M. and Kra, I.
[1] Riemann Surfaces, Springer-Verlag, New York-Heidelberg-Berlin, 1980.
Fay, J.D.
[1] Theta Functions on Riemann Surfaces, Lecture Notes in Mathematics 332,
Springer-Verlag, Berlin-Heidelberg-New York, 1973.
Fitting, H.
[1] Die Determinantenideale eines Moduls, Jahresber. Deutsch. Math. Verein.
46 (1936), 195-228.
Fulton, W.
[1] Algebraic Curves, Benjamin, New York, 1969.
[2] Intersection Theory, Springer-Verlag, Berlin-Heidelberg-New York-Tokyo,
1984.
Fulton, W. and Hansen, J.
[1] A connectedness theorem for projective varieties, with applications to inter-
sections and singularities of mappings, Ann. of Math. 110 (1979), 159-166.
Fulton, W. and Lazarsfeld, R.
[1] On the connectedness of degeneracy loci and special divisors, Acta Math.
146 (1981), 271-283.
Giambelli, G. Z.
[1] Ordine di una varieta pili ampia di quella rappresentata coll'annullare tutti i
minori di dato ordine estratti da una data matrice generica di forme, Mem.
R. /st. Lombardo (3) 11 (1904), 101-135.
Gieseker, D.
[1] Stable curves and special divisors, Invent. Math. 66 (1982), 251-275.
Grauert, H.
[1] Ein Theorem der analytischen Garbentheorie und die Modulraume komplexer
Strukturen, lnst. Hautes Etudes Sci. Pub!. Math. 5 (1960), 233-292.
378 Bibliography

Green, M.
[1] Quadrics of rank four in the ideal of the canonical curve, Invent. Math. 75
(1984), 85-104.
Griffiths, P.
[1] The extension problem for compact submanifolds of complex manifolds I, in
Proc. Conf Complex Analysis (Minneapolis 1964) (A. Aeppli, E. Calabi,
H. Rohrl, eds.), Springer-Verlag, New York, 1965, pp. 113-142.
[2] Hermitian differential geometry, Chern classes, and positive vector bundles, in
Global Analysis (D. Spencer, S. Iyanaga, eds.), Princeton University Press and
University of Tokyo Press, Tokyo, 1969, pp. 185-251.
Griffiths, P. and Harris, J.
[1] Principles of Algebraic Geometry, Wiley-lnterscience, New York, 1978.
[2] Residues and zero-cycles on algebraic varieties, Ann. of Math. 108 (1978),
461-505.
[3] Algebraic geometry and local differential geometry, Ann. scient. Ec. Norm. Sup.
(4) 12 (1979), 355-432.
[ 4] The dimension of the variety of special linear systems on a general curve,
Duke Math. J. 47 (1980), 233-272.
Grothendieck, A.
[1] Technique de descente et theoremes d'existence en geometrie algebrique V.
Les schemas de Picard: theoremes d'existence, Seminaire Bourbaki, 14e annee,
1961-62, n" 232.
Gruson, L., Lazarsfeld, R., and Peskine, C.
[1] On a theorem of Castelnuovo, and the equations defining space curves, Invent.
Math. 72 (1983), 491-506.
Gunning, R. C.
[1] Lectures on Riemann Surfaces, Princeton University Press, Princeton, 1966.
[2] Some curves in abelian varieties, Invent. Math. 66 (1982), 377-389.
Gunning, R. C. and Rossi, H.
[1] Analytic Functions of Several Complex Variables, Prentice-Hall, Englewood
Cliffs, N.J., 1965.
Halphen, G. H.
[1] Memoire sur Ia classification des courbes gauches algebriques, J. Ecole Poly-
technique 52 (1882), 1-200.
Harris, J.
[1] Galois groups of enumerative problems, Duke Math. J. 46 (1979), 685-724.
[2] A bound on the geometric genus of projective varieties, Ann. Scuola Norm. Sup.
Pisa (4) 8 (1981 ), 35-68.
[3] On the Kodaira dimension of the moduli space of curves, II: the even genus
case, Invent. Math. 75 (1984), 437-466.
[4] Theta characteristics on algebraic curves, Trans. Amer. Math. Soc. 271 (1982),
611-638.
[5] The genus of space curves, Math. Ann. 249 (1980), 191-204.
Harris, J. and Mumford, D.
[1] On the Kodaira dimension of the moduli space of curves, Invent. Math. 67
(1982), 23-86.
Hartshorne, R.
[1] Ample vector bundles, Inst. Hautes Etudes Sci. Pub/. Math. 36 (1969), 75-110.
[2] Algebraic Geometry, Springer-Verlag, New York-Heidelberg-Berlin, 1977.
[3] Ample Subvarieties of Algebraic Varieties, Lecture Notes in Mathematics 156,
Springer-Verlag, Berlin-Heidelberg-New York, 1970.
Bibliography 379

Hochster, M.
[1] Grassmannians and their Schubert subvarieties are arithmetically Cohen-
Macaulay, J. Algebra 25 (1973), 40-57.
Hodge, W. V. D.
[1] Some enumerative results in the theory of forms, Proc. Cam. Phil. Soc. 39
(1943), 22-30.
Hopf, H.
[1] Ein topo1ogischer Beitrag zur reellen Algebra, Comment. Math. Helv. 13
(1940/41), 219-239.
Keem, C.
[1] A remark on the variety of special linear systems on an algebraic curve, Ph.D.
Thesis, Brown University, 1983.
Kempf, G.
[1] Schubert methods with an application to algebraic curves, Publications of
Mathematisch Centrum, Amsterdam, 1972.
[2] On the geometry of a theorem of Riemann, Ann. of Math. 98 (1973), 178-185.
Kempf, G. and Laksov, D.
[1] The determinantal formula of Schubert calculus, Acta Math. 132 (1974),
153-172.
Kleiman, S.
[1] r-special subschemes and an argument of Severi's, Adv. in Math. 22 (1976),
1-23.
Kleiman, S. and Laksov, D.
[1] On the existence of special divisors, Amer. J. Math. 94 (1972), 431-436.
[2] Another proof of the existence of special divisors, Acta Math. 132 (1974),
163-176.
Klein, F.
[1] Ober Riemann's Theorie der Algebraischen Functionen und Ihrer Integrate,
Teubner, Leipzig, 1882.
Knudsen, F. F.
[1] The projectivity of the moduli space of stable curves III, Math. Scand. 52
(1983), 200-212.
Kodaira, K. and Spencer, D. C.
[ 1] On deformations of complex analytic structures I, II, Ann. of Mat h. 67 (1958),
328-466.
Laksov, D.
[1] The arithmetic Cohen-Macaulay character of Schubert schemes, Acta Math.
129 (1972), 1-9.
Lang, S.
[1] Abelian Varieties, lnterscience, New York, 1959.
Lax, R. F.
[1] On the dimension of varieties of special divisors, Trans. Amer. Math. Soc.
203 (1975), 141-159.
Lewittes, J.
[1] Riemann surfaces and the theta function, Acta Math. 111 (1964), 37-61.
Little, J.
[1] Translation manifolds and the converse of Abel's theorem, Compositio Math.
49 (1983), 147-171.
380 Bibliography

Liiroth, J.
[1] Note iiber Verzweigungsschnitte und Querschnitte in einer Riemann'schen
Flliche, Math. Ann. 4 (1871), 181-184.
MacDonald, I. G.
[1] Some enumerative formulae for algebraic curves, Proc. Camb. Phil. Soc. 54
(1958), 399--416.
[2] Symmetric products of an algebraic curve, Topology 1 (1962), 319-343.
Martens, H. H.
[1] A new proof of Torelli's theorem, Ann. of Math. 78 (1963), 107-111.
[2] On the varieties of special divisors on a curve, J. Reine Angew. Math. 227
(1967), 111-120.
[3] Varieties of special divisors on a curve II, J. Reine Angew. Math. 233 (1968),
89-100.
Mather, J.
[1] Stratifications and mappings, in Dynamical Systems (M. M. Peixoto, ed.),
Academic Press, New York, 1973.
Matsumura, H.
[1] Commutative Algebra, second edition, Benjamin, Reading, Mass., 1980.
Matsusaka, T.
[1] On a theorem of Torelli, Amer. J. Math, 80 (1958), 784-800.
Mattuck, ·A.
[1] Symmetric products and Jacobians, Amer. J. Math. 83 (1961), 189-206.
Mayer, A. L.
[1] Special divisors and the Jacobian variety, Math. Ann. 153 (1964), 163-167.
Meis, T.
[1] Die minimale Bliitterzahl der Konkretisierungen einer kompakten Riemann-
schen Fliiche, Schr. Math. Inst. Univ. Munster No. 16 (1960).
Mulase, M.
[1] Cohomological structure of solutions of soliton equations; isospectral de-
formations of ordinary differential operators and a characterization of Jacobian
varieties. Preprint, 1984.
Mumford, D.
[1] Abelian Varieties, Oxford University Press, Bombay, 1970.
[2] Prym varieties I, in Contributions to Analysis (L. V. Ahlfors, I. Kra, B. Maskit,
L. Niremberg, eds.), Academic Press, New York, 1974, pp. 325-350.
[3] Curves and Their Jacobians, The University of Michigan Press, Ann Arbor,
1975.
[4] Stability of projective varieties, L'Ensei,qnement Mathern. 23 (1977), 39-110.
[5] Tata Lectures on Theta I, Birkhliuser, Boston-Basel-Stuttgart, 1983.
[6] Tata Lectures on Theta II, Birkhliuser, Boston-Basel-Stuttgart, 1984.
Mumford, D. and Fogarty, J.
[1] Geometric Invariant Theory, second enlarged edition, Springer-Verlag, Berlin-
Heidelberg-New York, 1982.
Musili, C.
[1] Postulation formula for Schubert varieties, J. Ind. Math. Soc. 36 (1972),
143-171.
Northcott, D. G.
[1] Finite Free Resolutions, Cambridge University Press, Cambridge, 1976.
Bibliography 381

Petri, K.
[1] Uber die invariante Darstellung algebraischer Funktionen einer Veriinder-
lichen, Math. Ann. 88 (1922), 242-289.
[2] Uber Spezialkurven I, Math. Ann. 93 (1925), 182-209.
Porteous, I. R.
[1] Simple singularities of maps, in Proceedings of Liverpool Singularities-Sym-
posium 1 (C. T. C. Wall, ed.), Lecture Notes in Mathematics 192, Springer-
Verlag, Berlin-Heidelberg-New York, 1971, pp. 286-307.
Riemann, B.
[1] Theorie der Abel'schen Functionen, in Mathematische Werke, Teubner,
Leipzig, 1876.
Saint-Donat, B.
[1] On Petri's analysis of the linear system of quadrics through a canonical curve,
Math. Ann. 206 (1973), 157-175.
[2] Varietes de translation et theon!me de Torelli, C.R. Acad. Sci. Paris Ser. A
280 (1975), 1611-1612.
Sernesi, E.
[1] L'unirazionalita dei moduli delle curve di genere dodici, Ann. Scuola Norm.
Sup. Pisa (4) 8 (1981), 405-439.
Serre, J.P.
[1] Algebre Locale-Multiplicites, Lecture Notes in Mathematics 11, Springer-
Verlag, Berlin-Heidelberg, 1965.
Severi, F.
[ 1] Sulla classificazione delle curve algebriche e sui teorema di esistenza di Riemann
II, Rendiconti R. Accad. Lincei (5) 241 (1915), 1011-1020.
[2] Vorlesungen uber Algebraische Geometrie, Teubner, Leipzig, 1921.
Shafarevich, I. R.
[1] Basic Algebraic Geometry, Springer-Verlag, Berlin-Heidelberg-New York,
1977.
Shiota, T.
[1] Characterization of Jacobian varieties in terms of soliton equations, Ph.D.
Thesis, Harvard University, 1984.
Shokurov, V. V.
[1] The Noether-Enriques theorem on canonical curves, Math. USSR Sbornik
15 (1971), 361-403.
Siegel, C. L.
[1] Topics in Complex Function Theory, 3 vols., Wiley-Interscience, New York,
1969, 1971, 1973.
Smith, R. and Varley, R.
[1] On the geometry of .h;;, to appear in Rend. Sem. Mat. Univers. Politecn.
Torino.
Springer, G.
[1] Introduction to Riemann Surfaces, Addison-Wesley, Reading, Mass., 1957.
Thorn, R.
[1] Ensembles et morphismes stratifies, Bull. Amer. Math. Soc. 75 (1969), 240-284.
Tjurin, A. N.
[1] The geometry of the Poincare theta-divisor of a Prym variety, Math. USSR
Izvestija 9 (1975), 951-986.
382 Bibliography

Torelli, R.
[1] Sulle varieta di Jacobi, Rendiconti R. Accad. Lincei. Ct. Sci. Fis. Mat. Nat. (5)
22 (1913), 98-103.
Walker, R. J.
[I] Algebraic Curves, Princeton University Press, Princeton, 1950.
Wei!, A.
[1] Zum Beweis des Torellischen Satzes, Nachr. Akad. Wiss. Gottingen. Math.
Phys. Kl. lla (1957) 33-53.
Welters, G. E.
[1] A criterion for Jacobi varieties, to appear.
Weyl,H.
[1] Die Idee des Riemannschen Flache, Teubner, Leipzig, 1913.
[2] The Classical Groups, Princeton University Press, Princeton, 1946.
Index

Abel's Theorem 18 Chern

Accola's theorem 275 character 331
Abelian polynomial 85
differentials 7 Chern classes
sums 17 in G(2,4) 104
variety 21 of kernel bundles 105
Adjoint of a symmetric product 336
conditions 58 of a tensor product 89
divisor 57 Clifford's theorem 107
ideal 57 refinements of 137
Adjunction formula Complete interjections 138
in the plane 53 Complete linear series 4
on a general surface 138 Completeness of the adjoint series 54,
Ample vector bundle 305 56, 60
Andreotti-Mayer Theorem 253 Complex torus 20
Associated map 37 Connectedness theorem
Automorphisms 44 of Fulton-Hansen 271, 326
for degeneracy loci 311
for W~(C) 212, 314
Base locus 5 Constructible sheaf 313
Base-point-free c~
linear series 5 definition of 154
pencil trick 126 dimension of 159
Bi-elliptic curves 198, 210, 269, 276, in genus 4 196
280, non-emptiness 206
of genus 5 272 tangent space to 162, 163, 197
of genus 6 218 Cubic threefold 302
embeddings of 221 Curve 1
Brill-Noether Curves of genus 2 33, 40, 46, 48, 278
matrix 154, 159 Curves of genus 3 35, 40, 46, 248,
number 159, 205 279, 303
Canonical Curves of genus 4 34, 40, 47, 196,
curve 12, 117-118 206, 276, 279
line bundle 7 Curves of genus 5 206, 270, 279, 300
map 10, 15 Curves of genus 6 209, 218, 279
Castelnuouvo' s
bound 116
lemmas 120, 151 Deck transformation 47, 281
number 116 Degeneracy locus 83
theorem 151 existence theorem 307
theorem on k-normality 141 connectedness theorem 311
384 Index

Degree Gorenstein relation 59

of a divisor 2 G;, 153, 177' 187' 188
of a line bundle 7 Grothendieck
ofamap 9 decomposition 145
De Jonquieres' formula 359 group 85, 331
Determinantal varieties 67, 83 Riemann-Roch formula 333
Diagonal mappings 358
Difference map 223, 262, 276
Dimension theorem 214 Heat equation 24, 250
Divisor 2 Hodge theorem for curves 54
Duality theorem 7 Hurwitz formula, see Riemann-Hurwitz
formula
Hyperelliptic curve 10, 41, 43
Effective divisor 3 embeddings of 221
Embedding in genus 6 218
criteria for 40 Prym varieties of 303
of degree g + 3 40 theta-characteristics on 287-292
theorem 216 Torelli theorem for 247, 269
Enriques-Babbage theorem 124 W,(C) 279
Euler exact sequence 321
Exceptional divisor 9, 63, 205
Excess linear series 329 Ideal of initial forms 62
Existence theorem 206, 311 Index of speciality 8
Extremal curves 117, 147, 263 lnflectionary point 37
Isogeny 49

Families
of curves 288 Jacobian variety 17
of linear systems 182 Jacobi inversion theorem 19, 27
of line bundles 169
of theta characteristics 288
Fermat curve 44 Keem's theorems 200
First fundamental theorem of invariant Kempf's singularity theorem 241
theory 77 Klein canonical curve 123
First-order deformation k-normal 140, 141
of a linear series 185 Kodaira-Spencer class 171
of a line bundle 169 Koszul complex 139
Fitting ideal 179
Fundamental theorems of invariant
theory 71, 77 Laurent tail 14
Lefschetz hyperplane theorem 306, 315
Linear equivalence 4
Galois Linearly normal 140
covering 47 Linear series 4
extension 281 Linear system 4
Gap sequence 41
General position theorem 109
Generic determinantal variety 67 MacDonald's formula 350
Genus 1 Martens' theorem 191
Castelnuovo's bound on 116 for birational maps 198
formula for plane curves 53, 58, 60 Mittag-Leffler problem 14
of space curves 143 Mittag-Leffler sequence 13
Gherardelli' s theorem 147 Moduli space 28
Gieseker's lemma 145 Monodromy group 111
Index 385

Multiplicity Principal part (of a meromorphic

of a divisor 2 differential) 14
of a point in a variety 62 Projections 35, 51
of a point on a curve 36 Projective bundle 304, 317
Mumford's theorem 193 Projectively normal 140, 221
for birational maps 198 Projectivized tangent cone 62
for g~'s 198 Prym varieties 297
Push-pull formula 91, 318
for symmetric products 367
Nearly extremal curves 149
Net 4
of quadrics in IP' • 270 Quadrics 102
Newton functions 156 containing a curve 142
Noether, Max canonical 255
"AF+BG" theorem 139 of genus 5 270, 275
theorem 117, 151 in IP' 3 140
Non-degenerate 6 linear spaces on 102
Normal covering 47 spaces of 100
Normalized period matrix 22 Quasi-algebraically closed 282
Normalization 31 Quintic plane curve 244, 264, 270-274,
Normal Weierstrass point 42 300
Norm map 281

Order 2 Ramification
Oscnode 220 divisor 9
Osculating k-plane 37 index 9
sequence, of a map 37
Rational normal curve 6, 39
Pencil 4 Rational normal scroll 95, 121, 147
Period matrix 16, 21 Relative divisor 165
normalized 22 Residual 10
Petri's construction 127 Residue 2
Petri's theorem 131 theorem 3
Picard group 18 Riemann
Plane curves 31 bilinear relations 16, 21
genus formula for 53, 59 constant 27
linear series on 56 inequality 55
Plane sextic curve (see Sextic plane singularity theorem 226
curve) theorem 27
Plane quintic curve (see Quintic plane theta function 23
curve) Riemann-Hurwitz formula 8, 39
Pli.icker Riemann-Mumford relation 290
formula 39 Riemann-Roch theorem 4, 7
relations 71 for an abelian variety 21
Poincare geometric version 12
complete reducibility theorem 49 proof via adjoint series 55-56
formula 25 for surfaces 148
line bundle 166 Riemann surface associated to an alge-
residue 53 braic function 31
Postulation 56, 115, 120, 142, 144,
149, 199, 220
Prill's problem 268 Schottky problem 249
Principally polarized abelian variety 21 Schubert varieties 74
386 Index

Secant plane 12, 152 Torelli's theorem 27, 245, 263, 265,
formula for 340, 345 268
Secant variety 231 Trigonal curves 118, 198
of a canonical curve 364 embeddings of 219, 221
Second fundamental theorem of invariant of genus 5 270, 279
theory 71 of genus 6 218
Segre class 318 linear series on 138
Semicanonical divisors (see Theta
characteristics)
Serre's criterion 98 Uniform position theorem 112
Sextic plane curves 209, 220, 264, 275 Universal divisor 164
Sheet number 9 cohomology class of 336
Siegel modular group 23 Universal family
Siegel upper half plane 22 of abelian varieties 251
Simple abelian variety 49 of curves 29
Smoothness theorem 214, 215
Special divisor 8
Stable curve 29 Veronese map 6
Standard monomials 72 Veronese surface 245
Superabundance 56, 115, 120, 142,
144, 149, 199, 220
Sylvester's determinant 87 Web 4
Symmetric product 18, 136, 309 Weierstrass
Chern classes of 322, 336 gap sequence 41
cohomology of 328 pairs 365
tangent spaces to 171, 197 points 41
semigroup 41
Weight (of a Weierstrass point) 42
Tangent cone 61 W~(C)
to the theta divisor 226, 255 connectedness of 212
to W~(C) 240 definition of 17 6, I 77
Theta characteristics dimension of 189, 192, 193, 198,
definition of 287 200
in genus 5 270-274, 275 examples of 196, 199, 206-211, 218
Theta divisor 22 existence of 206
intersection of translates of 267 irreducibility of 214
singular locus of 250 multiplicity of 240-241
tangent cones to 226, 255, 270, tangent spaces to 189, 197, 269, 240
tangent planes to 268 Weil
Theta function 23 pairing 284
Todd class 332 reciprocity 283
Grundlehren der mathematischen Wissenschaften
Continued from page ii

225. Schutte: Proof Theory

226. Karoubi: K-Theory, An Introduction
227. Grauert/Remmert: Theorie der Steinschen Raume
228. Segal/Kunze: Integrals and Operators
229. Hasse: Number Theory
230. Klingenberg: Lectures on Closed Geodesics
231. Lang: Elliptic Curves: Diophantine Analysis
232. Gihman/Skorohod: The Theory of Stochastic Processes III
233. Stroock/Varadhan: Multi-dimensional Diffusion Processes
234. Aigner: Combinatorial Theory
235. Dynkin/Yushkevich: Markov Control Processes and Their Applications
236. Grauert/Remmert: Theory of Stein Spaces
237. Kothe: Topological Vector-Spaces II
238. Graham/McGehee: Essays in Commutative Harmonic Analysis
239. Elliott: Probabilistic Number Theory I
240. Elliott: Probabilistic Number Theory II
241. Rudin: Function Theory in the Unit Ball of C"
242. Blackburn/Huppert: Finite Groups I
243. Blackburn/Huppert: Finite Groups II
244. Kubert/Lang: Modular Units
245. Cornfeld/Fomin/Sinai: Ergodic Theory
246. Naimark: Tl\eory of Group Representations
247. Suzuki: Group Theory I
248. Suzuki: Group Theory II
249. Chung: Lectures from Markov Processes to Brownian Motion
250. Arnold: Geometrical Methods in the Theory of Ordinary Differential Equations
251. Chow/Hale: Methods of Bifurcation Theory
252. Aubin: Nonlinear Analysis on Manifolds, Monge-Ampere Equations
253. Dwork: Lectures onp-adic Differential Equations
254. Freitag: Siegelsche Modulfunktionen
255. Lang: Complex Multiplication
256. Hormander: The Analysis of Linear Partial Differential Operators I
257. Hormander: The Analysis of Linear Partial Differential Operators II
258. Smoller: Shock Waves and Reaction-Diffusion Equations
259. Duren: Univalent Functions
260. Freidlin/Wentzell: Random Perturbations of Dynamical Systems
261. Rernmert/Bosch/Giintzer: Non Archemedian Analysis-A Systematic Approach to Rigid
Analytic Geometry
262. Doob: Classical Potential Theory & Its Probabilistic Counterpart
263. Krasnosegcskii/Zabreiko: Geometrical Methods of Nonlinear Analysis
264. Aubin/Cellina: Differential Inclusions
265. Grauert/Remmert: Coherent Analytic Sheaves
266. de Rham: Differentiable Manifolds
267. Arbarello/Cornalba!Griffiths: Geometry of Algebraic Curves, Vol. I