(Graduate Texts in Mathematics 151) Joseph H. Silverman (Auth.) - Advanced Topics in The Arithmetic of Elliptic Curves-Springer-Verlag New York (1994) PDF

Graduate Texts in Mathematics 151
Editorial Board
S. Axler F.W. Gehring K.A. Ribet
Springer Science+Business Media, LLC

Graduate Texts in Mathematics
T AKEUTIIZARING. Introduction to 33 HIRSCH. Differential Topology.

Axiomatic Set Theory. 2nd ed. 34 SPITZER. Principles of Random Walk.
2 OXTOBY. Measure and Category. 2nd ed. 2nded.
3 SCHAEFER. Topological Vector Spaces. 35 ALEXANOERIWERMER. Several Complex
2nded. Variables and Banach Algebras. 3rd ed.
4 HILTON/STAMM BACH. A Course in 36 KELLEy/NAMIOKA et al. Linear
Homological Algebra. 2nd ed. Topological Spaces.
5 MAC LANE. Categories for the Working 37 MONK. Mathematical Logic.
Mathematician. 2nd ed. 38 GRAUERTIFRITZSCHE. Several Complex
6 HUGHEs/PIPER. Projective Planes. Variables.
7 SERRE. A Course in Arithmetic. 39 ARVESON. An Invitation to C*-Algebras.
8 T AKEUTIIZARING. Axiomatic Set Theory. 40 KEMENy/SNELUKNAPP. Denumerable
9 HUMPHREYS. Introduction to Lie Algebras Markov Chains. 2nd ed.
and Representation Theory. 41 APOSTOL. Modular Functions and Dirichlet
10 COHEN. A Course in Simple Homotopy Series in Number Theory.
Theory. 2nd ed.
11 CONWAY. Functions of One Complex 42 SERRE. Linear Representations of Finite
Variable I. 2nd ed. Groups.
12 BEALS. Advanced Mathematical Analysis. 43 GILLMAN/JERISON. Rings of Continuous
13 ANDERSON/fuLLER. Rings and Categories Functions.
of Modules. 2nd ed. 44 KENDIG. Elementary Algebraic Geometry.
14 GoLUBITSKy/GUILLEMIN. Stable Mappings 45 LoEVE. Probability Theory I. 4th ed.
and Their Singularities. 46 LoEVE. Probability Theory II. 4th ed.
15 BERBERIAN. Lectures in Functional 47 MOISE. Geometric Topology in
Analysis and Operator Theory. Dimensions 2 and 3.
16 WINTER. The Structure of Fields. 48 SACHS/WU. General Relativity for
17 ROSENBLATT. Random Processes. 2nd ed. Mathematicians.
18 HALMOS. Measure Theory. 49 GRUENBERG/WEIR. Linear Geometry.
19 HALMOS. A Hilbert Space Problem Book. 2nd ed.
2nded. 50 Eow AROS. Fermat's Last Theorem.
20 HUSEMOLLER. Fibre Bundles. 3rd ed. 51 KLINGENBERG. A Course in Differential
21 HUMPHREYS. linear Algebraic Groups. Geometry.
22 BARNEs/MACK. An Algebraic Introduction 52 HARTSHORNE. Algebraic Geometry.
to Mathematical Logic. 53 MANIN. A Course in Mathematical Logic.
23 GREUB. Linear Algebra. 4th ed. 54 GRAVERIWATKINS. Combinatorics with
24 HOLMES. Geometric Functional Analysis Emphasis on the Theory of Graphs.
and Its Applications. 55 BROWNIPEARCY. Introduction to Operator
25 HEWITT/STROMBERG. Real and Abstract Theory I: Elements of Functional
Analysis. Analysis.
26 MANES. Algebraic Theories. 56 MASSEY. Algebraic Topology: An
27 KELLEY. General Topology. Introduction.
28 ZARISKI/SAMUEL. Commutative Algebra. 57 CROWELLlFox. Introduction to Knot
Vol.I. Theory.
29 ZARISKIISAMUEL. Commutative Algebra. 58 KOBLITZ. p-adic Numbers. p-adic
Vol.lI. Analysis. and Zeta-Functions. 2nd ed.
30 JACOBSON. Lectures in Abstract Algebra I. 59 LANG. Cyclotomic Fields.
Basic Concepts. 60 ARNOLD. Mathematical Methods in
31 JACOBSON. Lectures in Abstract Algebra II. Classical Mechanics. 2nd ed.
Linear Algebra. 61 WHITEHEAD. Elements of Homotopy
32 JACOBSON. Lectures in Abstract Algebra Theory.
III. Theory of Fields and Galois Theory.
(continued after index)
Joseph H. Silverman
Advanced Topics in
the Arithmetic of
Elliptic Curves
With 17 Illustrations
t Springer
Joseph H. Silvennan
Mathematics Department
Brown University
Providence, RI 02912
USA
[email protected]
Editorial Board
S. Axler F.W. Gehring K.A. Ribet
Department of Mathematics Mathematics Department Department of Mathematics
San Francisco State University East HalI University ofCalifomia at
San Francisco, CA 94132 University of Michigan Berkeley
USA Ann Arbor, MI 48109 Berkeley, CA 94720-3840
USA USA
Mathematics Subject Classifications (1991): 14-01, llGxx, 14Gxx, 14H52
Library of Congress Cataloging-in-Publication Data

Silverman, Joseph H., 1955-
Advanced topics in the arithmetic of elliptic curves I Joseph H. Silverman .
p. cm. - (Graduate texts in mathematics; v. 151)
Includes bibliographical references and index.
ISBN 978-0-387-94328-2 ISBN 978-1-4612-0851-8 (eBook)
DOI 10. 1007/978-1-4612-0851-8
1. Curves, Elliptic. 2. Curves, Algebraic. 3. Arithmetic.
1. Title. II. Series.
QA567.S442 1994
516.3'52-dc20 94-21787
Printed on acid-free paper.
© 1994 Springer Science+Business Media New York

Originally published by Sprioger-Verlag New York, loc io 1994
All rights reserved. This work may not be translated or copied in whole or in part without the
written permission of the publisher Springer Scieoce+Busioess Media, LLC,
except for brief excerpts in connection with reviews or scholarly
analysis. Use in connection with any form of information storage and retrieval, electronic
adaptation, computer software, or by similar or dissimilar methodology now known or hereaf-
ter developed is forbidden .
The use of general descriptive names, trade names, trademarks, etc., in this publication, even
if the former are not especially identified, is not to be taken as a sign that such names, as
understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely
byanyone .
Production managed by Hal Henglein; manufacturing supervised by Vincent Scelta.

Photocomposed copy prepared from the author's TeX file.
9 8 7 6 5 4 3 2 (Corrected second printing, 1999)
ISBN 978-0-387-94328-2 SPIN 10727882

For Susan
Preface
In the introduction to the first volume of The Arithmetic of Elliptic Curves

(Springer-Verlag, 1986), I observed that "the theory of elliptic curves is
rich, varied, and amazingly vast," and as a consequence, "many important
topics had to be omitted." I included a brief introduction to ten additional
topics as an appendix to the first volume, with the tacit understanding that
eventually there might be a second volume containing the details. You are
now holding that second volume.
Unfortunately, it turned out that even those ten topics would not fit
into a single book, so I was forced to make some choices. The following
material is covered in this book:
I. Elliptic and modular functions for the full modular group.
II. Elliptic curves with complex multiplication.
III. Elliptic surfaces and specialization theorems.
IV. Neron models, Kodaira-Neron classification of special fibers,
Tate's algorithm, and Ogg's conductor-discriminant formula.
V. Tate's theory of q-curves over p-adic fields.
VI. Neron's theory of canonical local height functions.
So what's still missing? First and foremost is the theory of modular
curves of higher level and the associated modular parametrizations of ellip-
tic curves. There is little question that this is currently the hottest topic
in the theory of elliptic curves, but any adequate treatment would seem to
require (at least) an entire book of its own. (For a nice introduction, see
Knapp [lJ.) Other topics that I have left out in order to keep this book
at a manageable size include the description of the image of the £-adic
representation attached to an elliptic curve and local and global duality
theory. Thus, at best, this book covers approximately half of the material
described in the appendix to the first volume. I apologize to those who may
feel disappointed, either at the incompleteness or at the choice of particular
topics.
In addition to the complete areas which have been omitted, there are
several topics which might have been naturally included if space had been
available. These include a description of Iwasawa theory in Chapter II,
viii Preface
the analytic theory of p-adic functions (rigid analysis) in Chapter V, and

Arakelov intersection theory in Chapter VI.
It has now been almost a decade since the first volume was written.
During that decade the already vast mathematical literature on elliptic
curves has continued to explode, with exciting new results appearing with
astonishing rapidity. Despite the many omissions detailed above, I am
hopeful that this book will prove useful, both for those who want to learn
about elliptic curves and for those who hope to advance the frontiers of our
knowledge. I offer all of you the best of luck in your explorations!
Computer Packages
There are several computer packages now available for performing compu-
tations on elliptic curves. PARI and SIMATH have many built-in elliptic
curve functions, there are packages available for commercial programs such
as Mathematica and Maple, and the author has written a small stand-alone
program which runs on Macintosh computers. Listed below are addresses,
current as of March 1994, where these packages may be acquired via anony-
mous ftp.
PARI (includes many elliptic curve functions)
math.ucla.edu 128.97.4.254
megrez.ceremab.u-bordeaux.fr 147.210.16.17
(directory pub/pari)
(unix, mac, msdos, amiga versions available)
SIMATH (includes many elliptic curve functions)
ftp.math.orst.edu
ftp.math.uni-sb.de
apecs (arithmetic of plane elliptic curves, Maple package)
math.mcgill.ca 132.206.1.20
(directory pub / apecs)
Elliptic Curve Calculator (Mathematica package)
Elliptic Curve Calculator (stand-alone Macintosh program)
gauss.math.brown.edu 128.148.194.40
(directory dist/EllipticCurve)
A description of many of the algorithms used for doing computations on
elliptic curves can be found in H. Cohen [1, Ch. 7J and Cremona [1].
Acknowledgments
I would like to thank Peter Landweber and David Rohrlich for their care-
ful reading of much of the original draft of this book. My thanks also go
to the many people who offered corrections, suggestions, and encourage-
ment, including Michael Artin, Ian Connell, Rob Gross, Marc Hindry, Paul
Lockhart, Jonathan Lubin, Masato Kuwata, Elisabetta Manduchi, Michael
Rosen, Glenn Stevens, Felipe Voloch, and Siman Wong.
As in the first volume, I have consulted a great many sources while
writing this book. Citations have been included for major theorems, but
Preface ix
many results which are now considered "standard" have been presented as
such. In any case, I claim no originality for any of the unlabeled theorems
in this book, and apologize in advance to anyone who may feel slighted.
Sources which I found especially useful included the following:
Chapter I Apostol [1], Lang [1,2,3]' Serre [3], Shimura [1]
Chapter II Lang [1], Serre [6], Shimura [1]
Chapter IV Artin [1], Bosch-Liitkebohmert-Raynaud [1], Tate [2]
Chapter V Robert [1], Tate [9J
Chapter VI Lang [3,4]' Tate [3]
I would like to thank John Tate for providing me with a copy of his
unpublished manuscript (Tate [9]) containing the theory of q-curves over
complete fields. This material, some of which is taken verbatim from Pro-
fessor Tate's manuscript, forms the bulk of Chapter V, Section 3. In addi-
tion, the description of Tate's algorithm in Chapter IV, Section 9, follows
very closely Tate's original exposition in [2], and I appreciate his allowing
me to include this material.
Portions of this book were written while I was visiting the University
of Paris VII (1992), IRES (1992), Boston University (1993), and Harvard
(1994). I would like to thank everyone at these institutions for their hos-
pitality during my stay.
Finally, and most importantly, I would like to thank my wife Susan for
her constant love and understanding, and Debby, Danny, and Jonathan for
providing all of those wonderful distractions so necessary for a truly happy
life.
Joseph H. Silverman
March 27, 1994
Acknowledgments for the Second Printing

I would like to thank the following people who kindly provided correc-
tions which have been incorporated in this second revised printing: An-
drew Baker, Brian Conrad, Guy Diaz, Darrin Doud, Lisa Fastenberg, Benji
Fisher, Boris Iskra, Steve Harding, Sharon Kineke, Joan-C. Lario, Yihsiang
Liow, Ken Ono, Michael Reid, Ottavio Rizzo, David Rohrlich, Samir Sik-
sek, Tonghai Yang, Horst Zimmer.
Providence, Rhode Island February, 1999
Contents
Preface vii
Computer Packages viii
Acknowledgments viii
Introduction 1
CHAPTER I
Elliptic and Modular Functions 5
§l. The Modular Group 6
§2. The Modular Curve X(l) 14
§3. Modular Functions 23
§4. Uniformization and Fields of Moduli 34
§5. Elliptic Functions Revisited 38
§6. q-Expansions of Elliptic Functions 47
§7. q-Expansions of Modular Functions 55
§8. Jacobi's Product Formula for ~(T) 62
§9. Hecke Operators 67
§10. Hecke Operators Acting on Modular Forms 74
§11. L-Series Attached to Modular Forms 80
Exercises 85
CHAPTER II
Complex Multiplication 95
§l. Complex Multiplication over C 96
§2. Rationality Questions 104
§3. Class Field Theory - A Brief Review 115
§4. The Hilbert Class Field 121
§5. The Maximal Abelian Extension 128
§6. Integrality of j 140
§7. Cyclotomic Class Field Theory 151
§8. The Main Theorem of Complex Multiplication 157
§9. The Associated Gr6ssencharacter 165
§10. The L-Series Attached to a CM Elliptic Curve 171
Exercises 178
xii Contents
CHAPTER III
Elliptic Surfaces 187
§1. Elliptic Curves over Function Fields 188
§2. The Vv'eak Mordell-Weil Theorem 191
§3. Elliptic Surfaces 200
§4. Heights on Elliptic Curves over Function Fields 212
§5. Split Elliptic Surfaces and Sets of Bounded Height 220
§6. The Mordell-Weil Theorem for Function Fields 230
§7. The Geometry of Algebraic Surfaces 231
§8. The Geometry of Fibered Surfaces 236
§9. The Geometry of Elliptic Surfaces 245
§10. Heights and Divisors on Varieties 255
§11. Specialization Theorems for Elliptic Surfaces 265
§12. Integral Points on Elliptic Curves over Function Fields 274
Exercises 278
CHAPTER IV
The Neron Model 289
§1. Group Varieties 290
§2. Schemes and S-Schemes 297
§3. Group Schemes 306
§4. Arithmetic Surfaces 311
§5. Neron Models 318
§6. Existence of Neron Models 325
§7. Intersection Theory, Minimal Models, and Blowing-Up 338
§8. The Special Fiber of a Neron Model 350
§9. Tate's Algorithm to Compute the Special Fiber 361
§10. The Conductor of an Elliptic Curve 379
§11. Ogg's Formula 389
Exercises 396
CHAPTER V
Elliptic Curves over Complete Fields 408
§1. Elliptic Curves over C 408
§2. Elliptic Curves over lR 413
§3. The Tate Curve 422
§4. The Tate Map Is Surjective 429
§5. Elliptic Curves over p-adic Fields 438
§6. Some Applications of p-adic Uniformization 445
Exercises 448
Contents xiii
CHAPTER VI
Local Height Functions 454
§l. Existence of Local Height Functions 455
§2. Local Decomposition of the Canonical Height 461
§3. Archimedean Absolute Values - Explicit Formulas 463
§4. Non-Archimedean Absolute Values - Explicit Formulas 469
Exercises 476
APPENDIX A
Some Useful Tables 481
§l. Bernoulli Numbers and «(2k) 481
§2. Fourier Coefficients of ~(7) and j(7) 482
§3. Elliptic Curves over Q with Complex Multiplication 483
Notes on Exercises 484
References 488
List of Notation 498
Index 504
Introduction
In the first volume of The Arithmetic of Elliptic Curves, we pre-

sented the basic theory culminating in two fundamental global results,
the Mordell-Weil theorem on the finite generation of the group of rational
points and Siegel's theorem on the finiteness of the set of integral points.
This second volume continues our study of elliptic curves by presenting six
important, but somewhat more specialized, topics.
We begin in Chapter I with the theory of elliptic functions and modular
functions for the full modular group r(l) = SL 2 (Z)/{±1}. We develop this
material in some detail, including the theory of Hecke operators and the L-
series associated to cusp forms for r(l). Chapter II is devoted to the study
of elliptic curves with complex multiplication. The main theorem here
states that if K IQ is a quadratic imaginary field and if E IC is an elliptic
curve whose endomorphism ring is isomorphic to the ring of integers of K,
then K(j(E)) is the Hilbert class field of K; and further, the maximal
abelian extension of K is generated by j (E) and the x-coordinates t of the
torsion points in E(C). This is analogous to the cyclotomic theory, where
the maximal abelian extension of Q is generated by the points of finite
order in the multiplicative group C*. At the end of Chapter II we show
that the L-series of an elliptic curve with complex multiplication is the
product of two Hecke L-series with Grossencharacter, thereby obtaining at
one stroke the analytic continuation and functional equation.
The common theme of Chapters III and IV is one-parameter families
of elliptic curves. Chapter III deals with the classical geometric case, where
the family is parametrized by a projective curve over a field of characteristic
zero. Such families are called elliptic surfaces. Thus an elliptic surface
consists of a curve C, a surface £, and a morphism 7r : £ ---- C such that
almost every fiber 7r- 1 (t) is an elliptic curve. The set of sections
{maps (J : C ---- £ such that 7r 0 a(t) = t}

If j(E) = 1728 or j(E) = 0, one has to use x 2 or x 3 instead of x.
2 Introduction
to an elliptic surface forms a group, and we prove an analogue of the

Mordell-Weil theorem which asserts that this group is (usually) finitely
generated. In the latter part of Chapter III we study canonical heights
and intersection theory on e and prove specialization theorems for both
the canonical height and the group of sections.
Chapter IV continues our study of one-parameter families of ellip-
tic curves in a more general setting. We replace the base curve C by a
scheme S = Spec R, where R is a discrete valuation ring. The generic fiber
of the arithmetic surface e ----+ S is an elliptic curve E defined over the
fraction field K of R, and its special fiber is a curve £ (possibly singular,
reducible, or even non-reduced) defined over the residue field k of R. We
prove that if e ----+ S is a minimal proper regular arithmetic surface whose
generic fiber is E, and if we write e for the part of e that is smooth over S,
then e is a group scheme over S and satisfies Neron's universal mapping
property. In particular, E(K) ~ £(R); that is, every K-rational point on
the generic fiber E extends to an R-valued point of e. We also describe the
Kodaira-Neron classification of the possible configurations for the special
fibere and give Tate's algorithm for computing the special fiber. At the
end of Chapter IV we discuss the conductor of an elliptic curve and prove
(some cases of) Ogg's formula relati_ng the conductor, minimal discrimi-
nant, and number of components of e.
In Chapter V we return to the analytic theory of elliptic curves. We
begin with a brief review of the theory over C, which we then use to analyze
elliptic curves defined over TIt. But the main emphasis of Chapter V is on
elliptic curves defined over p-adic fields. Every elliptic curve E defined
over C is analytically isomorphic to C* j qZ for some q E C*. Similarly,
Tate has shown that if E is defined over a p-adic field K and if the j-
invariant of E is non-integral, then E is analytically isomorphic to K* j qZ
for some q E K*. (It may be necessary to replace K by a quadratic
extension.) Further, the isomorphism E(K) ~ K* jqZ respects the action
of the Galois group G K / K, a fact which is extremely important for the
study of arithmetic questions. In Chapter V we describe Tate's theory
of q-curves and give some applications.
The final chapter of this volume contains a brief exposition of the
theory of canonical local height functions. These local heights can be used
to decompose the global canonical height described in the first volume
[AEC, VIII §9]. We prove the existence of canonical local heights and give
explicit formulas for them. Local heights are useful in studying some of the
more refined properties of the global height.
As with the first volume, this book is meant to be an introductory text,
albeit at an upper graduate level. For this reason we have occasionally made
simplifying assumptions. We mention in particular that in Chapter II we
restrict attention to elliptic curves whose ring of complex multiplications
is integrally closed; in Chapter III we only consider elliptic surfaces over
fields of characteristic 0; and in Chapter IV we assume that all Dedekind
Introduction 3
domains and discrete valuation rings have perfect residue fields. Possibly
it would be preferable not to make these assumptions, but we feel that the
loss of generality is more than made up for by the concomitant clarity of
the exposition.
Prerequisites
The main prerequisite for reading this book is some familiarity with the ba-
sic theory of elliptic curves as described, for example, in the first volume.
Beyond this, the prerequisites vary enormously from chapter to chapter.
Chapter I requires little more than a first course in complex analysis. Chap-
ter II uses class field theory in an essential way, so a brief summary of class
field theory has been included in (II §3). Chapter III requires various clas-
sical results from algebraic geometry, such as the theory of surfaces and
the theory of divisors on varieties. As always, summaries, references, and
examples are supplied as needed.
Chapter IV is technically the most demanding chapter of the book.
The reader will need some acquaintance with the theory of schemes, such
as given in Hartshorne [1, Ch. II] or Eisenbud-Harris [1]. But beyond that,
there are portions of Chapter IV, especially IV §6, which use advanced
techniques and concepts from modern algebraic geometry. We have at-
tempted to explain all of the main points, with varying degrees of precision
and reliance on intuition, but the reader who wants to fill in every detail
will face a non-trivial task. Finally, Chapters V and VI are basically self-
contained, although they do refer to earlier chapters. More precisely, the
interdependence of the chapters of this book is illustrated by the following
guide:
/
ICh. I I - - - - - - - - - > ICh. VII ICh. III I n_+ ICh. Ivi
/
The dashed line connecting Chapter III to Chapter IV is meant to indicate
that although there are few explicit cross-references, mastery of the subject
matter of Chapter III will certainly help to illuminate the more difficult
material covered in Chapter IV.
References and Exercises
The first volume of The Arithmetic of Elliptic Curves (Springer-Verlag,
1986) is denoted by [AEC], so for example [AEC, VIII.6.7] is Theorem 6.7
in Chapter VIII of [AEC]. All other bibliographic references are given by
the author's name followed by a reference number in square brackets, for
example Tate [7, theorem 5.1]. Cross-references within the same chapter
are given by number in parentheses, such as (3.7) or (4.5a). References
from within one chapter to another chapter or appendix are preceded by
the appropriate Roman numeral or letter, as in (IV.6.1) or (A §3). Exercises
4 Introduction
appear at the end of each chapter and are numbered consecutively, so, for
example, exercise 4.23 is the 23 rd exercise at the end of Chapter IV.
Just as in the first volume, numerous exercises have been included at
the end of each chapter. The reader desiring to gain a real understanding of
the subject is urged to attempt as many as possible. Some of these exercises
are (special cases of) results which have appeared in the literature. A list
of comments and citations for the exercises will be found at the end of the
book. Exercises marked with a single asterisk are somewhat more difficult,
and two asterisks signal an unsolved problem.
Standard Notation
Throughout this book, we use the symbols
to represent the integers, rational numbers, real numbers, complex num-

bers, field with q elements, and p-adic integers respectively. Further, if R
is any ring, then R* denotes the group of invertible elements of R; and if A
is an abelian group, then A[m] denotes the subgroup of A consisting of all
elements with order dividing m. A more complete list of notation will be
found at the end of the book.
CHAPTER I
Elliptic and Modular Functions
In most of our previous work in [AEC], the major theorems have been of
the form "Let E / K be an elliptic curve. Then E / K has such-and-such
a property." In this chapter we will change our perspective and consider
the set of elliptic curves as a whole. We will take the collection of all
(isomorphism classes of) elliptic curves and make it into an algebraic curve,
a so-called modular curve. Then by studying functions and differential
forms on this modular curve, we will be able to make deductions about
elliptic curves. Further, the Fourier coefficients of these modular functions
and modular forms turn out to be extremely interesting in their own right,
especially from a number-theoretic viewpoint. We will be able to prove
some of their properties in the last part of the chapter.
This chapter thus has two main themes, each of which provides a
paradigm for major areas of current research in number theory and alge-
braic geometry. First, when studying a collection of algebraic varieties or
algebraic structures, one can often match the objects being studied (up
to isomorphism) with the points of some other algebraic variety, called a
moduli space. Then one can use techniques from algebraic geometry to
study the moduli space as a variety and thereby deduce facts about the
original collection of objects. A subtheme of this first main theme is that
the moduli space itself need not be a projective variety, so a first task is to
find a "natural" way to complete the moduli space.
Our second theme centers around the properties of functions and dif-
ferential forms on a moduli space. Using techniques from algebraic geom-
etry and complex analysis, one studies the dimensions of these spaces of
modular functions and forms and also gives explicit Laurent, Fourier, and
product expansions. Next one uses the geometry of the objects to define
linear operators (called Hecke operators) on the space of modular forms,
and one shows that the Hecke operators satisfy certain relations. One then
takes a modular form which is a eigenfunction for the Hecke operators
and deduces that the Fourier coefficients of the modular form satisfy the
same relations. Finally, one reinterprets all of these results by associating
an L-series to a modular form and showing that the L-series has an Euler
6 I. Elliptic and Modular Functions
product expansion and analytic continuation and that it satisfies a func-

tional equation.
§1. The Modular Group
Recall [AEC VI.3.6] that a lattice A <;;; C defines an elliptic curve E /C via
the complex analytic map
C/ A ---. EA(C) : y2 = 4x 3 - g2X - g3
Z f----> (p(z; A), p'(z; A)).
Here
p(z;A) = 12 +
Z
L ((
wEA
1 )2
Z -W
-~)
W
w¥O
is the Weierstrass p-function relative to the lattice A. (See [AEC VI,§3].)

Further, if Al and A2 are two lattices, then we have
if and only if Al and A2 are homothetic.

(See [AEC VI.4.1.1]. Recall Al and A2 are homothetic if there is a num-
ber c E C* such that Al = cA 2.)
Thus the set of elliptic curves over C is intimately related to the set
of lattices in C, which we denote by £.:
£. = {lattices in C}.
We let C* act on £. by multiplication,
cA = {cw : W E A}.
Then the above discussion may be summarized by saying that there is an
injection
£./C* L4 {elliptic curves defined over C}.
C-isomorphism
According to the Uniformization Theorem for Elliptic Curves (stated
but not proven in [AEC VI.5.1]), this map is a bijection. One of our goals
in this chapter is to prove this fact (4.3). But first we will need to describe
the set £./C* more precisely. We will put a complex structure on £./C*,
and ultimately we will show that £. /C* is isomorphic to C.
Let A E £.. We can describe A by choosing a basis, say
Switching WI and W2 if necessary, we always assume that the pair (W2, WI)
gives a positive orientation. (That is, the angle from W2 to WI is positive
and between 0 0 and 1800 • See Figure 1.1.)
An Oriented Basis for the Lattice A

Figure 1.1
Since we only care about A up to homothety, we can normalize our

basis by looking instead at
Our choice of orientation implies that the imaginary part of WdW2 satisfies
which suggests looking at the upper half-plane
H = {T E <C : Im( T) > O}.
We have just shown that the natural map
H --> f., j<C* ,
T f----> AT = ZT +Z
is surjective. It is not, however, injective. When do two T'S give the same
lattice? We start with an easy calculation.
8 1. Elliptic and Modular Functions
Lemma 1.1. Let a, b, c, dE lR, T E lC, T tJ- lR. Then
1m (aT + b) = (ad - bc) Im(T).

CT +d ICT + dl 2
PROOF. Let T = s+it. MUltiplying numerator and denominator by cf+d,

we find
aT+ b {aclTl2 + (ad + bc)s + bd} + {(ad - bc)t}i

CT + d ICT + dl 2
o
The ambiguity in associating aTE H to a lattice A lies in choosing an
oriented basis for A. Suppose that we take two oriented bases,
Then there are integers a, b, c, d, a', b', c', d' so that
w~ = aWl + bW2, Wl = a'w~ + b'w;,

w; = CWl + dW2, W2 = c'w~ + d'w;.
Substituting the left-hand expressions into the right-hand ones and using
the fact that Wl and W2 are lR-linearly independent, we see that
b) c(a'' d'b') = (10 0

(ac d )
1 .
Further, using Lemma 1.1 (with T = wl/w2) and the fact that our
bases are oriented, we find that
and so
ad - bc > O.
In other words, the matrix (~ ~) is in the special linear group over Z,
This proves the first half of the following lemma.

Lemma 1.2. (a) Let A c C be a lattice, and let Wl,W2 and w~,w& be two
oriented bases for A. Then
W~ = aWl + bw 2
for some matrix
w& = CWl + dW2
(b) Let Tl, T2 E H. Then ATl is homothetic to AT2 if and only if there is a
matrix
aTl + b
such that T2=---·
CTI + d
(c) Let A c C be a lattice. Then there is aTE H such that A is homothetic

to AT = ZT + Z.
PROOF. (a) This was done above.
(b) Using (a), we find that
ATl is homothetic to AT2
~ ZT2 + Z = ZO'TI + ZO' for some 0' E C*,
~ {T2 = aO'Tl + bO' for some (ac db) E SL 2(Z),

1 = CO'TI + dO'
aT! + b
=* T2 = CTI + d·
Conversely, if T2 = (aTl +b)/(CTI +d), let 0' = CTI +d. Then again using (a),
we find
O'A T2 = Z(aTl + b) + Z(CTI + d) = ZTl + Z = ATl .
Hence ATl and AT2 are homothetic.
(c) Write A = wlZ + W2Z with an oriented basis and take T = Wl/W2.
o
In view of Lemma 1.2(b), it is natural to define an action of SL 2(Z)
on H as follows:
,T = ~;:! for, = (::) E SL2(Z) and T E H.

The fact that ,Tis in H follows from Lemma 1.1, and the fact that this de-
fines a group action is an easy calculation. This action gives an equivalence
relation on the points of H, and Lemma 1.2(b) tells us what the cosets are.
There is a bijection
SL 2 (Z)\H one~one L/C,
T f----> AT.
We can actually do a little bit better, since the matrix
-1
-1 = (
o
acts trivially on H.
Definition. The modular group, denoted r(l), is the quotient group
r(l) = SL2 (Z)/{±1}.
Although r(l) is the quotient SL 2 (Z)/{±1}, we will generally just

write down matrices and leave it to the reader to remember that (r/ ~1)
is equal to (6 ~). For an explanation of the notation r(1), see exercise 1.6
where we define groups r(N) for all integers N ~ 1.
Remark 1.3. Note that ±1 are the only elements of SL2 (Z) which fix H.
For suppose that, = (~ ~) satisfies ,T = T for all T E H. This means
that
CT 2 - (d - a)T - b= 0 for all T E H,
from which we conclude that c = b = 0 and a = d. Hence, = ±1.

Remark 1.4. The group r(l) contains two particularly important ele-
ments, which we will denote
S= (0 -1)
1 0 '
Their action on H is given by
1
S(T) = --, T(T)=T+1.
T
Notice also that the elements Sand ST (~ ]1) have finite order,
r
=
and (ST)3 = (~ ~1 = 1,
so r(l) contains finite subgroups of order 2 and 3.

The next proposition provides us with a good description of the quo-
tient space f(l)\H.
Proposition 1.5. Let:r c H be the set
(See Figure 1.2 for a picture of:r and some of its translates by elements
ofr(l)')
(a) Let T E H. Then there is a, E f(l) such that ,T E:r.
(b) Suppose that both T and ,T are in :r for some, E f(l), , =I- 1. Then
one of the following is true:
11
§l. The Modular Group
-1 o 1
~ and Some of Its r(l)-nanslates
Figure 1.2
(i) Re(T) = -4 and iT = T + 1;

(ii) Re(T) =4 and iT = T - 1;
(iii) ITI = 1 and iT = -l/T.
(c) Let T E ~, and let
/(T) = bE r(l) : iT = T}
be the stabilizer of T. Then
{I, S} ifT = i;
/( ) = { {I, ST, (ST)2} if T = P = e 27ri / 3 ;
T {I, TS, (TS)2} if T = -15 = e 27ri / 6 ;
{I} otherwise.
PROOF. (a) We prove something stronger. Let r' be the subgroup of [(1)
generated by S = (? (/) and T = (6 i), and let T E H. We will prove
that there is a "( E r' such that "(T E 3".
For any "( = (~ ~) E r(l), Lemma 1.1 says that
Im(T)
ImbT) = ICT + d12'
Write T = S + it. Since t > 0, it is clear that

ICT + dl 2 = (cs + d)2 + (ct)2 ----> 00 as Ici + Idl ----> 00.
Hence, for our fixed T, there is a matrix "(0 E r' which maximizes the
quantity ImboT). Next, since TnT = T + n, we can choose an integer n so
that
IRe(Tn"(OT)I-:; ~.
We set "( = Tn,,(o and claim that "(T E T
Suppose to the contrary that "(T ~ T By construction, IRebT) I -:; ~,
so we must have hTI < 1. But then
ImbT)
Im(S"(T) = ~ > ImbT) = ImboT),
contradicting the choice of "(OT to maximize ImboT). This contradiction

shows that "(T E 3", which completes the proof of (a).
(b,c) We may assume that ImbT) ::::: Im(T), since otherwise we replace
the pair T,,,(T by the pair "(T,,,(-lbT). Writing,,( = (~~) as usual, we
have
Im(T)
Im(T)-:;ImbT)= ICT+dI 2 ' so ICT+dl-:;1.
Since 1m (T) ::::: ~ v'3, we must have Icl -:; 2/

v'3, so Ici -:; 1. Replacing "(
by -"( if necessary, it suffices to consider the cases c = and e = 1. °
ie = 01
Then a = d = 1 and "(T = T + b. Since
I Re( T) I -:; ~ and I RebT)1 = I Re(T + b)1 -:; ~,

it follows that
b = ±1 and Re(T) = =F~.
Ie = 11
By assumption, ITI ::::: 1 and IT + dl -:; 1. Writing T = S + it, this means that
and
so
§1. The Modular Group 13
1 :s; s2 + t 2 :s; 1 - 2ds - d2 = 1 - d(d ± 1) - d(2s =f 1).

Since d E Z, the quantity d(d± 1) is non-negative. Similarly since lsi :s; ~,
the quantity d(2s =f 1) is non-negative for one of the choices of + / - sign.
We conclude that
and d(2s + d) = 0.
We now look at several subcases.
Ic=l,d=OI
Then, = (1 (/ ), and since ITI = 1, we have
Hence one of the following three cases holds:
a = 0, Is I :s; 2'1 ITI = 1, ,=S, ,T= -liT;

a = 1, _1
s - 2' T= -p, ,=TS, ,(-p) = -p;
a = -1, s = -~, T = p, , = (ST)2, ,p =p.
Ic= 1,d= 1,s =-~I

Then T = P and , = (1 a 11 ), so
,T = a - 1
- - = a + p.
p+1
Since ,T E 3", this leads to two cases:
= 0, ,= ST, = p;
n,
a ,p
a = 1, ,= (i ,p = -po
Ic=l,d=-l,s=~1
Then T = -p, , = (1
-~11 ), and ,T = a + T, so just as in the previous
case there are two possibilities:
a=O, ,=(TS)2, ,( -p) = -p;
a=-l, '=(11~1)' ,( -p) = p.

o
The geometric description of the quotient space f(l)\H provided by
Proposition 1.5 can be used to give a quick proof of the following purely
algebraic fact.
Corollary 1.6. The modular group f(l) is generated by the matrices
S = (~ ~I ) and T = (~ ~).
PROOF. As in the proof of Proposition l.5(a), we let f' be the subgroup
of r(1) generated by Sand T. Fix some r in the interior :f, such as r = 2i.
Let"( E f(1). From the proof of (l.5a) there is a "(' E f' such that "'('("(r) E
:f. Thus r is in the interior of:f, and ("('''()r is in:f. We conclude from (1.5b)
that "("(' = 1. Therefore,,( = ,,('-1 E f/, which proves that f' = f(I). 0
Remark 1.6.1. It is in fact true that r(1) is the free product of its sub-
groups (S) and (ST) of orders 2 and 3. See exercise 1.1.
§2. The Modular Curve X(I)
The quotient space f(I)\H classifies the set of lattices in C up to homoth-

ety. Proposition 1.5 provides a nice geometric description of f(I)\H. The
vertical sides of the fundamental domain :f are identified by T, and the
two arcs of the circle Irl = I are identified by S, as shown in Figure 1.3.
Making these identifications, we see that as a topological space, r(1)\H
looks like a 2-sphere with one point missing. Our next tasks are to supply
that missing point, define a topology, and make the resulting surface into
a Riemann surface.
Rather than adding a single point to f(I)\H, we will give a more gen-
eral construction which is useful for generalizing the results of this chapter.
Definition. The extended upper half-plane H* is the union of the upper
half-plane H and the Q-rational points of the projective line,
H* = H U JlD1(Q) = H u Q U {oo}.
One should think of JlD1(Q) as consisting of the rational points on the real
axis together with a point at infinity. The points in JlD1(Q) are called the
cusps of H*.
There is a natural action of f(l) on JlD1(Q) defined by
(: ~) [~] = [::!~].
(Here we use [~] to denote homogeneous coordinates for a point in JlD1(Q).)
Thus r(1) acts on the extended upper half-plane H*. We define
Y(I) = f(I)\H and X(I) = f(I)\H*.
The points in the complement X(I) " Y(I) are called the cusps of X(I).
We now show that X(I) has only one cusp and calculate its stabilizer.
§2. The Modular Curve XCI) 15
--
The Geometry of r(1)\H
Figure 1.3
Lemma 2.1. (a)

X(1) " Y(1) = {oo}.
(b) The stabilizer in r(I) of 00 E H* is
1(00) = {(6 n E r(I)} = (the subgroup ofr(I) generated by T).
PROOF. (a) Let [=J E JPll(Q) be any point in H* "H. Since x and yare
homogeneous coordinates, we may assume that x, y E Z and gcd(x, y) = 1.
Choose a, b E Z so that ax + by = 1. Then
,= (a xb) E reI)
-y
and
Therefore every point in H* " H is equivalent (under the action of r(1»

to 00.
(b) We have (~ ~) [~] = [~] if and only if c = o. Hence (~ ~) has the
form (6 n· o
Topologically, XCI) looks like a 2-sphere. To make this precise, we

need to describe a topology on XCI). We start by giving a topology for H*.
Some Open Sets in H*

Figure 1.4
Definition. The topology of H* is defined as follows. For 7 E H, we take

the usual open neighborhoods of 7 contained in H. For the cusp 00, we
take as a basis of open neighborhoods the sets
{7 E H : Im( 7) > f\;} U {OO} for every f\; > O.

For a cusp 7 i 00, we take as a basis of open neighborhoods the sets
{the interior of a circle in H tangent to the real axis at 7} U {7}.
(See Figure 1.4.)
Remark 2.2.1. For any cusp 70 i 00, Lemma 2.1(a) says that there is
a transformation 'Y E r(l) with 'YOO = 70. Then one easily checks that 'Y
sends a set of the form {Im( 7) > f\;} to the interior of a circle in H tangent
to the real axis at 70. (See exercise 1.2.) In other words, the fundamental
neighborhoods of 00 and of the finite cusps are sent one-to-another by the
elements of r(1).
Remark 2.2.2. From the definition, it is clear that distinct points of H*
have disjoint neighborhoods. Hence H* is a Hausdorff space. It is also clear
from (2.2.1) that the elements of r(1) define homeomorphisms of H*.
The next lemma will help us describe the topology on the quotient
space X(l) = r(l)\H*. It will also be used later to define a complex
structure on X (1).
Lemma 2.3. For any two poipts TI,T2 E H*, let
and similarly, for any two subsets UI , U2 ~ H*, let
Then, for all Tl, T2 E H*, there exist open neighborhoods U l , U2 C H*

of TI, T2 respectively such that
(In other words, if "(UI and U2 have a point in common, then necessar-
ily "(TI = T2.)
PROOF. For any n, f3 E r(l) we have
and
It thus suffices to prove the lemma for any r(l)-translates of Tl and T2.
Using (1.5a) and (2.1a), we may assume that
From (1.5) and (2.1), we have a good description of how r(l) acts on H*
and 9'"*, as illustrated in Figure 1.2. We consider three cases, depending on
whether or not our points are at 00.
I TI,T2 E 9'"1
From (1.5) (or Figure 1.2) we see that 1(9'",9'") is finite; explicitly,
1(9'",9'") = {I, T, TS, TST, (TS)2, S, ST, STS, (ST)2, T- 1 }.

Let
9 = Interior ( U "(9'") .
,EI(:T,:T)
Then 9 is an open subset of H containing 9'". Further, 1(9,9) is finite, since
1(9, 9) ~ U
" ,,2EI(:T,:T)
Next we observe that if"( E 1(9,9)" 1(TI,T2), so "(Tl i- T2, then we

can find open sets V" W, in H satisfying
Let
-yEI(9,9) -yEI(9,9)
-yIi"I(T"T2) -yIi"I(T"T2)
By construction, 71 E U1 and 72 E U2 , so
1(71, T2) S;;; 1(U1, U2).

Suppose that they are not equal, say'Y E 1(U1, U2) '- l(Tl' T2). Then
and so
But V-y n W-y = 0, so 1 f{. l(V-y, W-y). This contradiction shows the other
inclusion and completes the proof that 1(Tl,T2) = 1(U1 ,U2).
IT1 E 3", T2 =
001
Let U1 be an open disk centered at T1. As in the proof of Proposition 1.5,
we observe that the quantity
Im(T)
'" = ",(U1 ) = sup Im("(T) = sup
TEU, TEU, Icr + dl 2
-yEr(I) (~ ~)Er(l)
+ it E U1, then sand t

is finite. (Note that if T = S are bounded, so
ICT+dI 2 = (CS+d)2+(ct)2 -> 00 as Icl+ldl -> 00 uniformly in T E Ud

Now
U2 = {T E H: Im(T) > "'} U {oo}
will be a neighborhood of 00 satisfying
for all 'Y E r(I).

Hence
IT1 = T2 = 001
Let
U= = {T E H: Im(T) > 2} U {oo}.
From (1.5) (or Figure 1.2) we see that the only elements of reI) which
take some point in U= to another point in U= are powers of T. Hence
from (2.1b) we conclude that
[(U=, U=) = {Tk E reI) : k E Z} = [(00,00).
o
Next we define a topology on XCI) and use Lemma 2.3 to show that
XCI) is a Hausdorff space. Note that this fact requires proof; it is not
immediate from the fact that H* is Hausdorff. (See exercise 1.3.)
Definition. Let
¢ : H* ------> r(l)\H* = X(l)
be the natural projection. The quotient topology on X(l) is defined by the
condition that U <:;;: X(l) is open if and only if ¢-l(U) is open. Equivalently,
it is the weakest topology for which ¢ is continuous. Note that ¢ is also an
open map, that is, it takes open sets to open sets. For if W c H* is open,
then so is
¢-l(¢W) = U 1'w.
'YEr(l)
Proposition 2.4. X(l) with its quotient topology is a compact Hausdorff

space.
PROOF. We start by checking that X(l) is compact. Let {UihEI be an
open cover of X(l). Then {¢-l(Ui)}iEI is an open cover of H*. In par-
ticular, some ¢-l(Ui ) contains 00, say 00 E ¢-l(Ui ,). By definition of the
topology on H*, there is a constant Ii > 0 so that
¢-l(Ui,) :2 {7 E H : Im(7) > Ii} U {oo}.
Hence the set ::r" ¢-l(Ui ,) is compact (it is closed and bounded), so there
is a finite subcover
Then Ui , U··· U Uin covers X(l).

Next we verify that X(l) is Hausdorff. Let X}'X2 E X(l) be distinct
points, and let 71,72 E H* be points with ¢( 7i) = Xi. Then 1'71 =1= 72 for
all l' E r(l), so in the notation of (2.3), I(7}, 72) = 0. From (2.3), there
are open neighborhoods U1, U2 <:;;: H* of 7}, 72 satisfying I(U 1, U2 ) = 0.
Then ¢(Ut), ¢(U2 ) are disjoint neighborhoods of Xl, X2. 0
Making X(l) into a compact Hausdorff space is a good start, but recall
that our ultimate goal is to give X(l) a complex structure. We recall what
this means.
Definition. Let X be a topological space. A complex structure on X is
an open covering {UihEl of X and homeomorphisms
'l/Ji : Ui ~ 'l/Ji (Ui ) c C

such that each 'l/Ji(Ui ) is an open subset of C and such that for all i,j E I
with Ui n Uj =1= 0, the map
'l/Jj 0 'l/J;1 : 'l/Ji(Ui n Uj ) ------> 'l/Jj(Ui n Uj )

is holomorphic. The map 'l/Ji is called a local parameter for the points in Ui .
A Riemann surface is a connected Hausdorff space which has a complex
structure defined on it.
Theorem 2.5. The following defines a complex structure on X(l) which

gives it the structure of a compact Riemann surface of genus 0:
Let x E X(l), choose Tx E H* with ¢>(Tx) = x, and let Ux C H* be a
neighborhood of Tx satisfying
(Such a Ux exists from Lemma 2.3 with T1 = T2 = Tx and Ux = U1 n U2 .)

Then
is a neighborhood of x, so {I(Tx)\Ux } xEX(l) is an open cover of X(l).
Ix # 001
Let r = #I(Tx), and let gx be the holomorphic isomorphism
gx : H ---+ {z E C : Izl < I},
Then the map
is well defined and gives a local parameter at x.

Ix = 001
We may take Tx = 00, so I(Tx) = {Tk}. Then
1/Jx (¢>( T)) = {oe27riT if ¢>( T) # 00,

if ¢>(T) = 00
is well defined and gives a local parameter at x.
Remark 2.5.1. If I(Tx) = {I}, then the natural map
is already a homeomorphism, so
is a local parameter at x. Thus the only real complication occurs when x

equals ¢>(i), ¢>(p), or ¢>(oo). (See also exercise 1.4.)
Remark 2.5.2. The following commutative diagrams illustrate the defi-

nitions of the local parameters 'l/Jx : I(Tx)\Ux '-> C.
4> 4>
Ux --; I(Tx)\Ux Ux --; I(Tx)\Ux
1 9x
z~zr
l 1Px 900 "" l 1PX
C --; C C
T - Tx
x =I- 00, gx(T) = --_ x = 00, goo(T) = e2rriT
T - Tx
PROOF (of Theorem 2.5). We already know that X(I) is a compact Haus-
dorff space (2.4), and it is clearly connected due to the continuous surjec-
tion <p : H* ......., X(I). Further, an inspection of Figure 1.2 shows that X(I)
has genus O. (For those who dislike such a visual argument, we will later
give an explicit map j : X(I) ......., pI(C). See (4.1) below. The interested
reader can check that our proof that j is analytic does not depend on the
a priori knowledge that X(I) has genus O. Then the elementary argument
described in exercise 1.11 shows that j is bijective, hence an isomorphism.)
By construction, the set
is a neighborhood of x. We must verify that the maps
are well-defined homeomorphisms (onto their images) and that they satisfy
the compatibility conditions for a complex structure.
We begin with a lemma which shows that the function gx(T) behaves
nicely with respect to the transformations in I (T x).
Lemma 2.6. Let a E H, let R : H ......., H be a holomorphic map with
R(a) = a, and let geT) = (T - a)/(T - a). Suppose further that
r times
~
Ro ... oR(T)=T
and that r :2': 1 is the smallest integer with this property. Then there is a
primitive rth-root of unity ( such that
for all T E H.
PROOF. Note that g is an isomorphism
g : H ....:::... {z E C: Izl < I}

with g(a) = 0, so the map
G =90 R 0 g-l : {z E IC : Izl < I} ~ {z E IC : Izl < I}
is a holomorphic automorphism of the unit disk with G(O) = O. It follows

that G(z) = cz for some constant c E IC. (See, e.g., Ahlfors [1].) Since
the r-fold composition Go··· 0 G(z) = z and r is chosen minimally, we
conclude that c is a primitive rth-root of unity. 0
We resume the proof of Theorem 2.5. Suppose first that x =I 00. Note
that from (1.5), I(Tx) is cyclic, say generated by R. Then (2.6) implies that
gX(RT) = (g(T) for all T E H,
where ( is a primitive rth-root of unity. Hence
so 'l/Jx is well defined on the quotient I(Tx )\Ux '

Next we check that 'l/Jx is injective. Let T1, T2 E Ux. Then
'l/Jx(¢(Td) = 'l/Jx(¢(T2)) ~ gx(Td T = gx(T2)'"

~ gx(T1) = (igx(T2) :s: i < r,
for some 0
~ gx(Td = gx(RiT2) for some 0 :s: i < r,
~ T1 = RiT2 for some 0 :s: i < r,
~ ¢(T1) = ¢(T2)'
Hence 'l/Jx is injective. Finally, it is clear from the commutative diagram
given in (2.5.2) that both 'l/Jx and 'l/J:;1 are continuous, since the maps ¢, gx,
and z I--t ZT are all continuous and open. Therefore 'l/Jx is a homeomorphism.
The case x = 00 is similar. From (2.1b) we know that 1(00) = {Tk}
consists of the translations T I--t T + k for k E Z. Hence 'l/Jx (¢( T)) = e 27riT
is well defined and injective on the quotient I (00) \ U00' And, as above, 'l/Jx
and 'l/J:;1 are continuous, since both ¢ and T I--t e27riT are continuous and
open. Hence 'l/Jx is a homeomorphism.
It remains to check compatibility. First let x, y E X(1) with x, y =I 00.
Then
Now gy and g:;l are holomorphic, so the only possible problem would be
the appearance of fractional powers of z. Let ( be the primitive r x th_root
of unity such that gx (Rx T) = (gx (T). Then using the fact that ¢ 0 I = ¢
for any I E r(l), we find
g~Y 0 g:;l((Z) = 'l/Jy 0 ¢ 0 Rx 0 g:;l(Z) = 'l/Jy 0 ¢ 0 g:;l(Z) = g~Y 0 g:;l(Z).

It follows that g~Y 0 g;;l(Z) is a power series in ZT x , which proves that the
composition 1/Jy 0 1/J;;l(Z) is holomorphic. (Note the importance of knowing
that ( is a primitive r x th_ root of unity.)
By exactly the same computation, taking g=(T) = exp(21TiT) , the
function
is holomorphic.
Finally, we note that
g~Y(T + 1) = 1/Jy 0 ¢ 0 T(T) = 1/Jy 0 ¢(T) = g~Y(T),
so g~Y (T) is a holomorphic function in the variable q = e2 71'iT. (Note T is
restricted to Uy n U=; it is not allowed to tend toward ioo.) Hence the
transition map
is holomorphic.
This completes the proof that the open sets I(Tx)\Ux and the maps
1/Jx : I(Tx)\Ux -> C
define a complex structure on X(l). D
§3. Modular Functions
In the previous section we showed that the quotient space X(l) = r(1)\H*
has the structure of a Riemann surface of genus O. It is natural to look at
the meromorphic functions on this Riemann surface.
Example 3.1. Recall that to each T E H we have associated a lattice AT =
ZT + Z and an elliptic curve CI AT' From Lemma 1.2(b) there is a well-
defined map (of sets)
r(l)\H --+ C
T j (CI AT) .
f---+
We will show later (4.1) that with the complex structure described in (2.5),
the j function is a meromorphic function on X(l) which gives a complex
analytic isomorphism
j : X(l) --.:::..., jp'1(C).
Every meromorphic function f on X(l) is thus a rational function
of j, that is, f E C(j). In order to have a richer source of functions, we will
study functions on H that have "nice" transformation properties relative
to the action of r(l) on H. Although these transformation properties may
look somewhat artificial at first, the corresponding functions actually define
differential forms on X(l), so they are in fact natural objects to study.
(See (3.5) below for further details.)
Definition. Let k E Z, and let f(7) be a function on H. We say that f is

weakly modular of weight 2k (for r(l)) if the following two conditions are
satisfied:
(i) f is meromorphic on H;
(ii) f(-y7) = (C7 + d)2k f(7) for all)' = (~ ~) E r(l), 7 E H.
Remark 3.2. Note that a function satisfying f(-y7) = (c7+d)K f(7) for an
odd integer r;, is necessarily the zero function, since taking), = (rl ~1)
yields f (7) = - f (7). This explains why we restrict attention to even
weights.
Remark 3.3. Since (1.6) says that r(l) is generated by the two matri-
ces S = (~ rl)
and T = (6 i),
a meromorphic function f on H is weakly
modular of weight 2k if it satisfies the two identities
f(7 + 1) = f(7) and
From the first it follows that we can express f as a function of
and f will be meromorphic in the punctured disk

{q: 0 < Iql < I}.
Thus f has a Laurent expansion J in the variable q, or in other words, f
has a Fourier expansion:
00
n=-oo
Definition. With notation as in (3.3), f is said to be
L
00
meromorphic at 00 if J= anqn for some integer no,

n=-no
L anqn.
00
holomorphic at 00 if J=
n=O
If f is meromorphic at 00, say J= a_noq-n O + ... with a_no f= 0, then the

order of f at 00 is
ordoo(f) = ordq=o(J) = -no·
If f is holomorphic at 00, its value at 00 is defined to be
f(oo) = ](0) = ao.
Definition. A weakly modular function that is meromorphic at 00 is called
a modular function.
Definition. A modular function that is everywhere holomorphic (i.e., ev-

erywhere on H and at 00) is called a modular form. If in addition f(oo) = 0,
then f is called a cusp form.
Example 3.4.1. Let A be a lattice. The Eisenstein series
is absolutely convergent for all integers k ;:::: 2. (See [AEC VI.3.1].) For T E
H we let
GZd T ) = GzdAr) = L 1
(mT + n)2k'
m.nEZ
(m,n)#(O,O)
By inspection,
for any c E C* ,
whereas
aT + b 1 1
kyr =Z--d +Z= --d(Z(aT+b)+Z(cT+d)) = --dAr'
CT+ CT+ CT+
Hence
G 2k hT) = G 2d A')'T) = G 2k ( (CT + d) -1 AT )

= (CT + d) 2k G Zk (Ar) = (CT + d) Zk G 2k (T).
Thus G 2k is weakly modular of weight 2k.
Proposition 3.4.2. Let k ;:::: 2 be an integer. The Eisenstein series G 2k
is a modular form of weight 2k. Its value at 00 is given by G 2k (00) =
2((2k), where ((s) is the Riemann zeta function. (For the complete Fourier
expansion ofG 2k , see (7.1).)
PROOF. We have just shown that G 2k is weakly modular, so it remains to

show that G 2k is holomorphic on H and at 00 and to compute its value at 00.
Note that if T is in the fundamental domain J' described in Proposition 1.5,
then
Hence the series obtained from G 2k (T) by putting in absolute values is

dominated, term-by-term, by the series obtained from Gzdp) by putting
in absolute values. Therefore G Zk is holomorphic on T But H is covered
by the r(1)-translates of J', and GzkhT) = (CT + d) 2k G 2k (T), so G 2k is
holomorphic on all of H.
Next we look at the behavior of G 2k (T) as T ----- ioo. Since the series
for G 2k converges uniformly, we can take the limit term-by-term. Terms
of the form (mT + n)-2k with m i- 0 will tend to zero, whereas the others
give n-2k. Hence
00
1
lim G 2k (T) = ----u; = 2((2k).
T~'lOO
n=-oo
n
n#D
This shows that G 2k is holomorphic at 00 and gives its value. o

Example 3.4.3. It is customary to let
and
(See [ABC VI.3.5.1].) The (modular) discriminant is the function
It is a modular form of weight 12, since from (3.4.2) we know that G 4(T)
and G 6(T) are modular forms of weights 4 and 6 respectively.
Using the well-known values (see (7.2) and (7.3.2))
4
((4) = ;0 and
we find that
47r4
92(00) = 120((4) = 3' ~(oo) = O.
Hence ~(T) is a cusp form of weight 12. We will see below (3.10.2) that it
is essentially the only one.
Remark 3.5. Let'"Y = (~ ~) E SL 2 (Z), and let dT be the usual differential

form on H. Then
aT+b) ad-bc -2
d("(T) =d ( CT +d = + d)2 dT
(IT = (CT + d) dT.
Thus dT has "weight -2." In particular, if J(T) is a modular function of

weight 2k, then the k-form
J(T) (dT)k
is r(l)-invariant. It thus defines a k-form on the quotient space r(l)\H, at

least away from the orbits of i and p, where the complex structure is a bit
more complicated.
We will soon show that f (T) (dT)k actually defines a meromorphic k-

form on X(I). We begin with a brief digression concerning differential
forms on arbitrary Riemann surfaces. In particular, formula (3.6b) below
will be crucial in our determination of the space of modular forms of a
given weight.
Definition. Let X/C be a smooth projective curve, or, equivalently, a
compact Riemann surface. Recall that Ox is the C(X)-vector space of
differential I-forms on X. (See [AEC II §4].) The space of (meromor-
phic) k-forms on X is the k-fold tensor product
0'X = O~k = Ox i8lqX) ... i8lqX) Ox.
0'X is a I-dimensional C(X)-vector space [AEC II.4.2a]. Notice that if we

00
set 01- = C(X), then E9 0'X has a natural structure as a graded C(X)-
k=O
algebra.
Let w E O'X, x E X, and choose a uniformizer t E C(X) at x. Then
w = g(dt)k
for some function 9 E C(X). We define the order of w at x to be
It is independent of the choice of t. (If t' is another uniformizer, then ap-

plying [AEC II.4.3b] we find that dt/dt' is holomorphic and non-vanishing
at x.) Just as with I-forms, we define the divisor of w by
div(w) = L ordx(w)(x) E Div(X);

xEX
we say that w is regular (or holomorphic) if
for all x E X.
Proposition 3.6. Let X/C be a smooth projective curve of genus g,

let k 2:: I be an integer, and let w E O'X.
(a) Let Kx be a canonical divisor on X [AEC II §4]. Then div(w) is
linearly equivalent to kKx .
(b)
deg(divw) = k(2g - 2).
PROOF. (a) Let r] E 01- be a non-zero I-form with divisor diver]) = Kx.
Then
F = w/r]k E 01- = C(X)
is a function on X, so
div(w) = k div(7]) + div(w/7]k) = kKx + div(F)
is linearly equivalent to kKx .
(b) From (a), deg(divw} = kdeg(Kx }. Now apply the Riemann-Roch
theorem [AEC II.5.4bJ, which says that deg(Kx) = 2g - 2. 0
The next proposition gives the precise relationship between a modular

function J of weight 2k and the corresponding k-form J(7} (d7)k.
Proposition 3.7. Let J be a non-zero modular function of weight 2k.
(a) The k-form J(7) (d7)k on H descends to give a meromorphic k-form wf
on the Riemann surface X(l). In other words, there is a k-form Wf E n~(1)
such that
¢*(wf) = J(7) (d7)k,
!
where ¢ : H -> X(l) is the usual projection.
(b) Let x E X(l), and let 7x E H* with ¢(7x ) = x. Then
ordr,,(J) ifx Ie ¢(i),¢(p),¢(oo);
~ordi(J)-~k ifx=¢(i);
ord (wf) =
x ! ordp(J) _ ~k if x = ¢(p);
ordoo(J) - k if x = ¢(oo).
Remark 3.7.1. If J is a modular function, then it is easy to see that the

order of vanishing of J at 7 E H depends only on the r(l)-equivalence class
of 7. The point is that since J(7) = (cr + d)2k J(7) and cr + die 0, we
have
ordr(J) = ord r (J 0 ')'-1) = ord"Yr(J).
Thus the expression in (3.7b) really does not depend on the choice of the
representative 7 x .
PROOF. (a) As we have seen, the k-form J(7) (d7)k is invariant for the
action of r(l) on H. We must show that for each x = ¢(7x ) E X(l),
the k-form J(7) (d7}k descends locally around x to a meromorphic k-form
on X(l), and that it vanishes to the indicated order. Clearly, we will
need to use the description of the complex structure on X(l) provided by
Theorem 2.5. We consider two cases.
Ix Ie 001
Using the notation from (2.5), there is a commutative diagram
q,
--->. I(7x )\Ux
1,p"
which defines a local parameter
at x. We write
and
-I() TxZ-Tx
T = gx Z = Z _ 1 '
so w = ZT is our local parameter.

Let R be a generator of 1(Tl;)' Then from (2.6),
and so
Here ( is some primitive rth-root of unity.

Now
J(T) (dT)k = J(g;;I(Z)) (dg;;I(Z))k

= J(g;;I(Z))g;;I'(z)k (dz)k = F(z) (dz)k,
where F(z) = J(g;I(Z))g;I'(z)k is a meromorphic function of z. Note

further that since gx is a local isomorphism, we have
We must show that F(z) (dz)k is a meromorphic function of w = ZT.

To do this, we use the fact (3.5) that J(T) (dT)k is r(l)-invariant. This
implies
F(z) (dz)k = J(T) (dT)k = J(RT) (dRT)k

= J(g;;I((Z)) (dg;;I((z))k = F((z) (d(z)k = F((z)(k (dz)k.
In particular, the function zk F(z) is invariant under the substitu-

tion z f-+ (z. Since ( is a primitive rth_root of unity, it follows that
for some meromorphic function FI (w). Hence
F(z) (dz)k = r-kzk(l-T)F(z) (d(ZT))k

= r- k z-1'k FI(zT) (d(ZT))k = r-kw-kFI(w) (dW)k,
which proves that J(T) (dT)k descends to a meromorphic k-form wf in a

neighborhood of x.
Finally, we compute
ordT=Tx J(T) = ordz=o F(z) = ordz=o z-k FI(zr) = -k + rordw=o FI(w);
ord x Wj = ordw=o r-kw- k FI(w) = -k + ordw=o FI(W).
Eliminating ordw=o FI (w) from these two equations yields
ord x Wj = ~r ord T
x
J- (1 - ~)r k.
It only remains to note that from (1.5),
1 if x -I ¢(i), ¢(p),
r= { 2 ifx=¢(i),
3 if x = ¢(p).
Ix = 001
Again using (2.5), we have a local parameter
'l/Jx : J(oo)\Uoo ---> C,
Let q = e 27riT be the local parameter at 00, and write J(T) /(q) as
in (3.3). Since dT = (27riq)-1 dq, we have
J(T) (dT)k = j(q)(27riq)-k(dq)k.

By definition, j is meromorphic at q = 0, so J(T) (dT)k descends to a
meromorphic k-form Wj in a neighborhood of 00. Finally,
ord oo Wj = ordq=o j(q) (27riq)-k = ordoo(f) - k.

o
Proposition 3.7 describes the local behavior of the k-form Wj E n~(1)"
The Riemann-Roch theorem, specifically Proposition 3.6(b), gives a global
description of its degree. Combining these results, we obtain the following
important formula.
Corollary 3.8. Let J be a non-zero modular function of weight 2k. Then
1 1 k
2 ord, (f) + "3 ord p (f) + ordoo (f) + L ordT (f) = "6.
TEr(I)\H*
r-=/=i,p,oo
(Here the sum is over any set of representatives for r(1)\H* excluding the
equivalence classes containing i, p, and 00.)
PROOF. First note that from (3.7.1), the sum is independent of the choice
of representatives for r(1)\H*. Let Wj E n~(1) be the k-form corresponding
to fer) (dr)k as in Proposition 3.7. By the Riemann-Roch theorem (3.6b)

and the fact that XCI) has genus 0 (2.5), we find that
deg(divwf) = -2k.
On the other hand, (3.7) gives
deg(divwf) = (~ordd - ~k) + Uordp f - ~k)

+ (ord oo f - k) + L ordT(f).
TEr(l)\H*
T,pi,p,oo
Equating these two expressions for deg( div W f) gives the desired formula.
o
Using Corollary 3.8, we can give a good description of the space of all
modular forms of a given weight. We set the notation
M2k = {modular forms of weight 2k for r(1)},
Mgk = {cusp forms of weight 2k for reI)}.

Note that both M2k and Mgk are C-vector spaces.
Example 3.9. For all k ~ 2, the Eisenstein series G 2k(r) is in M2k but
is not in Mgk • The modular discriminant 6(r) is in MP2. See (3.4.2)
and (3.4.3).
Theorem 3.10. (a) For all integers k ~ 2,
M2k ~ Mgk +CG 2k·
(b) For all integers k, the map
is an isomorphism of C- vector spaces.

(c) The dimension of M2k as a C-vector space is given by
if k < 0;
dim M2k ={ [~/6J if k ~ 0, k == 1 (mod 6);
[k/6 + IJ if k ~ 0, k =t 1 (mod 6).
(The square brackets denote greatest integer. For an alternative proof of (c)
using the Riemann-Roch theorem, see exercises 1.8 and 1.9.)
PROOF. (a) By definition, Mgk is the kernel of the map

so M2k/M8k has dimension at most 1. On the other hand, for k 2': 2, the
Eisenstein series G2k is in M2k and is not in M8k. (See (3.4.2).) Hence
for all k 2': 2.
(b) First we note that
which implies that G4(p) = 0 and G 6 (i) = O. Since G 4 and G 6 are modular
forms of weight 4 and 6 respectively (3.4.2), it follows from (3.8) that they
have no other zeros in r(I)\H. In particular,
so ~(T)is not identically zero.

Thus ~(T) is a non-zero modular form of weight 12 with ~(oo) = O.
It follows from (3.8) that
ordoo(~) = 1
and that ~(T) I- 0 for all T E H. (For an alternative proof that ~(T) I- 0
for all T E H, see [AEC VI.3.6a].) Therefore 1/~ has a simple pole at 00
and no other poles, so the map
M8k ~ M2k-12,
/ t----> / / ~
is well-defined. (The main point is that as long as / vanishes at 00,

then / / ~ will still be holomorphic at 00.) This gives an inverse to the
map in (b), so M8k ~ M 2k-12.
(c) If k < 0, then (3.8) implies immediately that M2k = o. (Note that all
of the terms in the left-hand sum are non-negative.) Similarly, if / E M o,
then (3.8) says that / has no zeros on H*. Thus / gives a holomorphic
non-vanishing function on X(l). But X(I) is a compact Riemann surface,
so an analytic map [I, 1] : X (1) ---> pi (C) is necessarily either constant or
surjective. Hence / is constant and Mo ~ C.
Next we use (3.8) to describe all functions / E M2k for small values
of k. Note that for small values of k, the equation
11k
-a+ -b+c=-
2 3 6
will have very few solutions in non-negative integers a, b, c. For example,

if k = 1, there are no solutions. We compile the results in Table 1.1.
Table 1.1
ordrf basis
k ordi f ordr f
7 # i,p for M2k
1 0
2 0 1 0 G4
3 1 0 0 G6
4 0 2 0 G~
5 1 1 0 G4 G6
Everything in Table 1.1 is clear except that the functions in the final
column actually form a basis. They are in M2k from (3.4.2), so we need to
show that M2k has dimension 1. But if II, h E M2k with 2 :S k :S 5, then
Table 1.1 shows that II and h have exactly the same zeros. Hence II / h E
M o = C, which proves that dim(M2k) = 1 for 2 :S k :S 5.
We have now verified (c) for all integers k :S 5. On the other hand,
if k 2:: 0, then using (a) and (b) we find that
dim M2k+12 = dim Mgk +12 + 1 from (a)

= dimM2k + 1 from (b).
Thus the left-hand side of (c) increases by 1 when k is replaced by k + 6.

Since the same is true of the right-hand side, an easy induction argument
completes the proof. 0
Example 3.10.1. Each of the vector spaces
has dimension 1. For example, since G~ E Ms and G s EMs, it follows

immediately that
Gs = cG~
for some constant c E Co Letting 7 -+ ioo and using (3.4.2), we can even
compute
2((8) 3
c = 4((4)2 = 7·
(See (7.2) for the calculation of ((8).) Similarly, GlO = t1 G 4 G 6 and G14 =
134°3 G~G6. More generally, M2k has a basis consisting of functions of the
form G 4Gg. (See exercise 1.10.) To appreciate the subtlety of identities
such as these, the reader might try to give a proof that G s = ¥G~ directly
from the series definition (3.4.1) of the G 2k 'S.
Example 3.10.2. Since M2k ~ Mgk+12 from (3.10b), the spaces
also have dimension 1. In particular, up to multiplication by a constant,

there is only one cusp form of weight 12, namely ~(7).
§4. Uniformization and Fields of Moduli
We begin by proving the Uniformization Theorem for elliptic curves, which

was stated but not proved in [AEC VI.5.1]. This theorem says that ev-
ery elliptic curve over IC is parametrized by Weierstrass elliptic functions.
Our main tool will be Theorem 3.7(a), which says in particular that every
modular function of weight 0 defines a meromorphic function on the Rie-
mann surface X(l). For a more elementary, but less intrinsic, proof of the
Uniformization Theorem, see exercise 1.11.
Definition. The modular j -invariant j (7) is the function
Thus j(7) is the j-invariant of the elliptic curve
and EAT (IC) has a parametrization using the Weierstrass p-function,
IC/AT --> EAT(IC),

z 1----+ (p(z;AT),p'(z;A T)).
(For details, see [AEC VI.3.6].)

Theorem 4.1. j(7) is a modular function of weight O. It induces a
(complex analytic) isomorphism
PROOF. From (2.4.2) and (2.4.3), both ~(7) and 92(7)3 = 26 33 53 G 4 (7)3
are modular forms, and both have weight 12, so their quotient is a modular
function of weight o. By (3.7a) with k = 0, j defines a meromorphic
function on X(l). (N.B. This means that j is meromorphic relative to
the complex structure on X(l) described by (2.5).) Hence j gives a finite

complex-analytic map
j : X(1) ---., JP'1(C).
Finally, we note that g2(ioo) = 120((4) cI 0 (3.4.2) and ~(ioo)

o (3.4.3). Since ~ has weight 12, (3.8) implies that
ord oo ~ = 1.
Thus j has a simple pole at the cusp 00 E X(l) and no other poles on X(l),
so the map j : X(l) ----+ JP'1(C) is an analytic map of degree 1 between
compact Riemann surfaces. It is therefore an isomorphism. 0
Corollary 4.2. Let 1 be a modular function of weight O.

(a) The function 1 is a rational function of j, that is, 1 E C(j).
(b) If in addition 1 is holomorphic on H, then 1 is a polynomial function
of j, that is, 1 E qj].
PROOF. (a) From (3.7a), 1 defines a meromorphic function on X(l), and

so by (4.1), 10 j-l is a meromorphic function on JP'l(CC). But the only
meromorphic functions on JP'1 (CC) are rational functions, so
for some P(T) E C(T).
Substituting t = j(x) with x E X(l) gives I(x) = P(j(x)).

(b) From (a), we know that 1 = P(j) for some rational function P(T) E
C(T). Suppose P is not a polynomial. Then there is a to E CC such
that P(to) = 00. The isomorphism j : X(l) -=:-. JP'1(C) from (4.1) sends H
to CC C JP'1(C), so we can find a TO E H with j(TO) = to. But then I(TO) =
P(j(TO)) = P(to) = 00, contradicting the assumption that 1 is holomorphic
on H. Hence P(T) must be a polynomial. 0
Corollary 4.3. (Uniformization Theorem For Elliptic Curves over CC)

Let A, B E CC satisfy 4A 3 + 27 B2 cI O. Then there is a unique lattice A c CC
such that
and
The map
CC / A ---., E: y2 = x 3 + Ax + B,
Z I-----> (p(z; A), ~p'(z; A))
is a complex analytic isomorphism.
PROOF. Using Theorem 4.1, we can choose aTE H such that
4A 3
j(T) = 1728 4A3 + 27B2'
Assume first that AB =I- O. It follows from this and the definition of jeT)
that
so
Let
a= and
Then
-4 B2g2(T)3
g2(A) = a g2(A T ) = A2g3(T)2 = -4A,
-6 B3g2 (T)3
g3(A) = a g3(A T ) = A3 g3 (T)2 = -4B.
Similarly, if A = 0, then jeT) = 0 and g2(T) = 0, whereas if B = 0,

then j (T) = 1728 and g3 (T) = O. Hence in these two cases it suffices to
take A = aA T with
a = {jg3(T) if A = 0, and a = «g2(T) if B = O.

-4B -4A
This gives the existence of A. Since we will not need the uniqueness of A
in our subsequent work, we will leave this fact to the reader. (See ex-
ercise 1.12.) Finally, we note that the second part of Corollary 4.3 is
essentially a restatement of [AEC VI.3.6b]. 0
We are now ready to relate the function jeT), defined as a meromor-

phic function on the Riemann surface X(I), to the j-invariant defined
in [AEC III §1] which classifies isomorphism classes of elliptic curves. We
let
£.G.G = {elliptic curves defined over C} .
c \l...-IS0mOrp h'Ism
If"' •
Thus an element of £.G.Gc is a C-isomorphism class of elliptic curves. We

also recall the notation
.G = {lattices in C}
from §1. Much of our preceding discussion is summarized in the following

proposition.
Proposition 4.4. There are one-to-one correspondences between the
following four sets, given by the indicated maps:
.GjC* +- r(1)\H
{A} = {AT} +- T
Here AT = ZT + Z, {EA} denotes the C-isomorphism class of the elliptic

curve EA : y2 = 4x 3 - g2(A)x - g3(A), and {A} is the homothety class of
the lattice A.
PROOF. Since j(ioo) = 00, the bijectivity of r(1)\H .L C is (4.1). The
bijectivity ofr(1)\H -+ £/C* is (1.2bc). Finally, the injectivity of £/C* -+
e££c is [AEC 4.1.1] and the surjectivity is (4.3). 0
Let us describe in a bit more detail the bijective map

e££c --+ C
given in Proposition 4.4. Let {E} E e££c be an isomorphism class of
elliptic curves, and choose a Weierstrass equation
E : y2 = x 3 + Ax + B
for some curve E in this class. Now take a basis 1'1,)'2 for the homology
group HI (E(C) , Z), and compute the periods
WI = 1,1 dx
Y
and W2 = 1,2 dx
Y
(See [AEC VI §1].) Switching WI and W2 if necessary, we may assume that
WI
TE =- E H.
W2
Then evaluate the holomorphic function j (T) at T = TE.
Thus the map
j : e££c --+ C, {E} I----t j (TE)
involves two transcendental (i.e., non-algebraic) operations, namely the
computation of the periods WI, W2 and the evaluation of the function j (T).
From this perspective, it seems unlikely that rationality properties of j (TE)
should have anything to do with rationality properties of E. To describe
the relationship that does exist, we make the following two definitions.
Definition. Let {E} E e££c, and let K ~ c.. We say that K is a field
of definition for {E} if there is an elliptic curve Eo in the isomorphism
class {E} such that Eo is defined over K. We say that K is a field of
moduli for {E} if for all automorphisms a E Aut(C/Q),
El7 E {E} if and only if a acts trivially on K.
Note that the field of moduli exists and is unique, since by Galois
theory an equivalent definition is that the field of moduli is the fixed field
of the group
{a E Aut(C/Q): El7 E {E}}.
From the complex analytic viewpoint described above, it is not clear that
the number j ( {E}) should have any relationship to fields of definition and
moduli for {E}. Note that there are lots of bijections e££c -+ c.. For
example, j' ({ E}) = e71"j ({ E}) + e- 7r is also a bijection. But clearly, it is
not possible for both j and j' to have good rationality properties.
Proposition 4.5. Let {E} E £££c.

(a) Q (j ({E})) is the field of moduli for {E}.
(b) Q (j ({E} )) is the minimal field of definition for {E}.
PROOF. The j-invariant j(E) of the elliptic curve
is
so for any (J E Aut(C/Q),
j(E<7) = j(E)<7.
(a) From [AEC III.1.4b] we have
if and only if
Since j(E<7) = j(E)<7, this shows that Q(j(E)) is the field of moduli
for {E}.
(b) We know from [AEC II1.1.4bc] that there exists an elliptic curve Eo
defined over Q(j(E)) with j(Eo) = j(E), and so satisfying Eo ~/c E. This
shows that Q(j(E)) is a field of definition for {E}.
On the other hand, if K is any field of definition for {E}, let EolK be
a curve in {E} given by an equation
with A,B E K.
Then
4A3
j(E) = j(Eo) = 1728 4A3 + 27B2 E K,
so Q(j(E)) s;;: K. o
Remark 4.6. The reader should note that the proof of Proposition 4.5 is
very elementary because we have explicit Weierstrass equations with which
to work. (This is how [AEC III.1.4bc] was proven.) For modular curves
of higher level the problem becomes considerably more difficult, since one
cannot rely on explicit equations. (See Shimura [1, §6. 7].) Finally, we
should mention that an analogous statement is false for abelian varieties of
higher dimension; the field of moduli for an isomorphism class of abelian
varieties need not be a field of definition.
§5. Elliptic Functions Revisited
Let A c C be a lattice. Our fundamental elliptic function is the Weier-

strass p-function,
p(z;A) = 1
2"
z + L ((z-w
1 )2 - 1 ) .
2"
W
wEA
w;iQ
As we have seen [AEC VI §3], p defines a meromorphic function on the

elliptic curve Cj A. It has a pole of order 2 at 0 E Cj A and no other poles.
We have also computed the Laurent series of p around z = 0 [AEC VI.3.5a],
p(z; A) = z2
1
+~ 2k
~(2k + 1)G2k+2(A)z ,
k=l
valid for Izlless than the smallest non-zero vector in A.

Since p(z; A) has no residues, we can integrate it to find a new function
which will almost be periodic for the lattice A. Note, however, that when
we integrate the series for p(z; A) term-by-term, it is necessary to adjust
the constant of integration in each term so as to ensure convergence.
Proposition 5.1. (a) The series
«(z; A) = -1
z
+" z-w w
z)
~' ( -1- + -1 + 2"
W
wEA
w;iQ
is absolutely and uniformly convergent on compact subsets of C "A. It

defines a merom orphic function on C with simple poles on A and no other
poles. «(z; A) is called the Weierstrass (-function (associated to the lat-
tice A).
(b) The Laurent series for ( around z = 0 is
«(z; A) = - -
1
Z
L G2k+2(A)Z2k+l.
00
k=l
PROOF. (a) Let C c C" A be a compact set, and let
E = inf{lz - wi: z E C, wE A} and M = sup{lzl : z E C}.

Since C is compact, we have E > 0 and M < 00.
Let z E C, and let w E A satisfy Iwl > 2M. Then
1_ 1_ + ~ +...:...1
Z - W w
W
= 2
Z2 . _1_ ::; 21wM132 .
w3 1_ ~
w
On the other hand, there are only finitely many terms in the sum with 0 <
Iwl : : : 2M, and for z E C those terms all satisfy
lIz I Izl -1
I--+-+-
z - w W w2 < -1- + -
- Iz - wi
1
Iwl+ -
Iwl<2 - C
+1-+
Iwl
M
- 2.
Iwl
Hence
We know [AEC VI.3.1a] that the last series converges, which proves the
series defining «(z; A) converges absolutely and uniformly on C.
It follows that «(z; A) is holomorphic on C" A, and an inspection of
the series defining ( shows immediately that it has simple poles at each
point of A.
(b) Let z be a complex number such that Izi < Iwl for all non-zero w E A.
Then
_1_ + ~ + ~2 = _~ {_1_ -1-~}

z - w W w W 1_ ~ w
w
so
«(z;A) = - +
1
Z wEA
L --1 L (-)
00
W k=2
Z
W
k
w#D
1
=- -
Z
LG k+1(A)zk.
00
k=2
This is the desired series once one notes that Gk(A) = 0 for odd k. 0
Differentiating the series (5.1a), we see that ('(z; A) = -p(z; A). Thus
the derivative of «(z; A) is periodic for the lattice A, so ( itself will have some
sort of "quasi-periodicity" property as explained in the following proposi-
tion.
Proposition 5.2. (a) For all z E C,
d
dz «(z; A) = -p(z; A).
(b) For all w E A and all z E C,
«(z + w; A) = «(z; A) + TJ(w),

where the number 7](w) is independent of z. The map
is called the quasi-period map associated to A. If W E A and W ~ 2A,

then 7](w) is given by the formula
7](W) = 2«(~w;A).
(c) The quasi-period map is a homomorphism of A into C.

(d) (Legendre Relation) Let A = ZWI + ZW2 be a lattice with basis satis-
fying Im(wdw2) > O. Then
PROOF. (a) The series (5.la) defining (converges absolutely and uniformly,
so it can be differentiated term-by-term. The result is the defining series
for -!J.
(b)
d d
dz ((z + w; A) = -!J(z + W; A) = -!J(z; A) = dz ((z; A).
Integrating, we find that the quantity
7](W) = «(z + W; A) - ((z; A)
is independent of z. If, further, W ~ 2A, then (does not have a pole at ±~'"-"
Putting z = -~w and using the fact (evident from the defining series)
that «(-z;A) = -«(z;A), we find in this case that T)(w) = 2((~w;A).
(c) We compute
T)(W + w') = ((z + W + w': A) - ((z; A)

= {((z + W + w';A) - ((z + w; A)} + {((z + '"-': A) - «(z: A)}
= 7](w / ) + 7](w).
(d) We integrate ((z; A) around a fundamental parallelogram offset slightly

so as not to contain points of A on its boundary. Thus let D be the region
and let
be its boundary as illustrated in Figure 1.5.

a
An Offset Fundamental Domain for IC / A
Figure 1.5
The only pole of ( in D is a simple pole of residue 1 at z = O. (Look

at the series (5.1a) defining (.) Hence
r «(z; A) dz
laD
= 27fi.
On the other hand, using (b) we get some cancellation when computing
the line integrals over opposite sides. Thus
j L1+L3
«(z; A) dz = r ((a + tW2; A) W2dt + fO ((a +
lo
l
1
WI + tW2; A) W2dt
= fo1 «(a + tW2; A) W2dt - fo1 ((a + tW2) + rl(wr)) W2dt

= -T)(Wr)W2'
Similarly,
Therefore
Remark 5.3. Let EIC be an elliptic curve given by a Weierstrass equa-

tion, and let
dx
WE=------
2y + alX + a3
be the associated invariant differential. The lattice A for E is the set of
periods
where'Y runs over all closed paths on E(C). (Equivalently, 'Y runs through
the cycles in Hl(E(C),Z). See [AEC VI §1].) The classical name for an
everywhere holomorphic differential such as WE on a Riemann surface such
as E(C) is a differential of the first kind.
Similarly, a differential of the second kind is a meromorphic differential
with no residues (i.e., with no simple poles), and a differential of the third
kind is a merom orphic differential with at worst simple poles.
The differential
8J(Z; A) dz = X WE
is thus a differential of the second kind on CIA, and its indefinite integral
is the multi-valued function -((z; A). The indeterminacy in ( is given by
the numbers
1, XWE = Ja
ra+w
8J(z; A) dz = -((a + w; A) + ((a; A) = -1](w),
where W = I, WE is the period associated to the closed path "f-

In terms of our original Weierstrass equation, there is the period map
whose image is the lattice A. Using this to identify A with the first ho-
mology of E(C), we see that the quasi-period map associates to a path the
negative of the corresponding period for the differential XWE:
1] : Hl (E(C), Z) ----7 C,
The last function we want to examine is essentially the integral of (.

To eliminate the indeterminacy caused by the simple poles of (, we take
the exponential of the integral. This leads to a familiar function which we
used in [AEC VI §3] to construct elliptic functions with a given divisor.
Proposition 5.4. (a) The infinite product
a(z; A) = z II (1 _ ~) e Z / W +(1/2)(z/w)2
wEA
w,eo
defines a holomorphic function on C with simple zeros on A and no other
zeros. It is called the Weierstrass a-function (associated to the lattice A).
(b)
d d2
dz log a(z; A) = ((z; A), dz 2 log a(z; A) = -p(z; A).
(c) For all z E C and wE A,
a(z + w; A) = 1jJ(w)e'7(w)(z+~w)a(z; A),
where T/ : A --. C is the quasi-period map for A, and 1jJ is defined by
1jJ : A -+ {±l}, 1jJ(w) = {I ifwE2A;

-1 ifwtJ- 2A.
PROOF. (a) This is a restatement of [AEC VI.3.3aJ.

(b) Taking the derivative of
loga(z; A) = log(z) + L {log ( 1 - ~) + ~ + ~ (~f}

wEA
w,eo
gives the defining series (5.la) for (, and then from (5.2a) we see that the
second derivative is -po Note that the logarithms are locally well defined
up to the addition of a constant which disappears when we differentiate
and also that we must take the principal branch of log (1 -
~) for almost
all w in order to ensure the convergence of the series.
(c) From (b) and (5.2b),
d a(z+w;A)
dz log a(z; A) = ((z + w; A) - ((z; A) = T/(w),
so
a(z + Wi A) = Ce'7(w)za(zi A)
for some constant C not depending on z. Note also that a is an odd

function, a fact that is clear from the product defining a.
We consider two cases. First, if w tJ- 2A, then a does not vanish at ±~w.
Hence putting z = - ~w gives
so
c = _e~1/(w)w.
Next, if W E 2A, then u has a simple zero at ±!w. Using L'Hopital's
rule yields
Ce-~1/(W)W = lim u(z + W; A) u' (lw·

2'
A) -1
z--+-w/2 u(z; A) u'(-!w;A) - .
(Note that u' is an even function, since u is odd.) Hence in this case we
find that
which completes the proof of (c). o

Any elliptic function can be factored as a product of Weierstrass u-
functions reflecting its zeros and poles. We give a general result and two
important examples. To ease notation, since the lattice A is fixed, we will
write u(z) and p(z) instead of u(z; A) and p(z; A).
Proposition 5.5. Let fez) be a non-zero elliptic function for the lat-
tice A. Write the divisor of f as
r
div(f) = L ni(ai)
i=1
for some ai E C, and let

r
b = Lniai.
i=1
(See [AEC VI §2] for the definition of the divisor of an elliptic function.)
Then there is a constant c E C* so that
u(z) rrr n·
fez) = c u(z _ b). u(z - ai) ' .
• =1
Corollary 5.6.
(a) ( )_ () __ u(z + a)u(z - a)

p z p a - u(z)2 u (a)2.
(b) '( ) __ u(2z)

P z - u(z)4·
PROOF (of Proposition 5.5). Let
From [AEC VI.2.2c] we know that bE A, so using (5.4c), we find that
O'(z) = ±eT/(b)(z-!b)
O'(z - b)
is holomorphic and non-vanishing on all of C. Since O'(z) has simple zeros
on A and no other zeros, it follows that 9 has exactly the same zeros and
poles as f. Hence f (z) / g( z) is everywhere holomorphic.
Next we verify that 9 is an elliptic function. Let w E A, and use (5.4c)
to write
O'(z + w) = Ae Bz
O'(z)
for certain constants A and B which depend on w but not on z. Then
g(z+w) = O'(z-b) O'(z+W)rr(O'(z+w-ai))n i

g(z) O'(z - b + w) O'(z) i=l O'(z - ai)
II (AeB(Z-ai)) n,
r
= e-B(z-b)e Bz
i=l
= eB(b-Enia;J (AeBZ)Eni = 1.
The last equality follows from the definition of b and the fact [AEC VI.2.2b]
that the divisor of an elliptic function has degree O.
This proves that g(z) is an elliptic function, and so f(z)/g(z) is an
everywhere holomorphic elliptic function. From [AEC VI.2.1] we conclude
that it is constant. 0
PROOF (of Corollary 5.6). (a) Since tl(z) is an even function of order 2,
we see immediately that the zeros of tl(z) - tl(a) are a and -a. Thus
div(tl(z) - tl(a)) = (-a) + (a) - 2(0).

Applying (5.5) we find
( ) _ () =CO'(z+a)O'(z-a)
tl z tl a 0'(z)2
for some constant C. Multiplying by z2 and using
and lim O'(z) = 1

z-+O z
gives the value of C,
(b) Divide (a) by z - a and let z ~ a. This yields
, l' a(z - a) a(z + a) __ '( )a(2a)

p (a) =- 1m Z - a a ( z )2 a ( a )2 -
z--->a
a 0 a(a)4'
Since a' (0) = 1, this is the desired result. o
§6. q-Expansions of Elliptic Functions
As we have seen in §§1-4, it is often convenient to use normalized lattices
with 7 E H.
We then use the obvious notation
p(z; 7), ((z; 7), a(z; 7) for
We will soon see that p, (, and a are quite well behaved when considered
as functions of two variables (z; 7) E C x H.
Note that since 1 E AT) the p function satisfies the relation
p(z + 1; 7) = p(z; 7).
This means that it is possible to expand p as a Fourier series in the vari-

able u = e2 :rriz. Similarly, since Ar+1 = AT) the p function satisfies
p(z; 7 + 1) = p(Z; 7).
Thus, as a function of 7, the p function should have a Fourier expansion

in terms of q = e 2:rrir.
This idea can be formulated more intrinsically as follows. Let
and
and let
be the cyclic subgroup of C* generated by q. Then there is a complex-

Note that this is an isomorphism of complex Lie groups, since it is clearly

a homomorphism.
Our first step is to express p(z; r) as a power series in the variables u =
e 21fiz and q = e 21fiT • The quickest way to do this is to write down (by
magic?!?) the correct expression, and then verify that it gives the same
function as p(z; r). We opt instead for a somewhat lengthier, but hopefully
more perspicuous, derivation.
Consider first the series
1 1
p(z;A) z + '"
= 2" 6 (z-w )2
wEA
w;iO
defining p. How does it arise? From [AEC VI.2.3] we know that any non-
constant elliptic function must have at least two poles, so we look for a
meromorphic function F( z) satisfying
(i) F(z +w) = F(z) for all z E C, w E A;
(ii) F(z) has a double pole at each point in A and no other poles.
The simplest function with a double pole at w is (z - W)-2. Byaver-
aging over w E A, we find a series
1
F(z) = 2: (z -
wEA
w)2
which Jormally satisfies (i) and (ii). The problem is that this series is not
absolutely convergent. However, by subtracting an appropriate constant
from each term, we can create a series which does converge and has the
desired properties. This is how we "discovered" p(z; A) in [AEC VI §3].
We apply the same principle to express p(z; r) as a function of u and q.
Exponentiating the conditions (i) and (ii), we look for a function F(u; q)
satisfying
(iii) F( qku; q) = F( u; q) for all u E C*, k E 2;
(iv) F(u; q) has a double pole at each u E qZ and no other poles.
As above, we look for F to be an average
F(u; q) = 2: J(qn u )
nEZ
for some elementary function J. Such an F will clearly satisfy the period-
icity condition (iii).
To obtain (iv), we need J(T) to have a double pole at T = 1. For
example, we might use J(T) = (1 - T)-2. But the series
does not converge, since iqi < 1. The terms with n --t -00 are all right,
since then qrt --t 00. But as n --t 00, the nth term goes to 1.
In order to get convergence, we want f(T) to have a double pole at T =
1 and also to satisfy
lim f(T)
T ..... O
=0 and lim f(T)
T ..... oo
= O.
The simplest such function is f(T) = T(1-T)-2, which leads us to consider

the function F(u; q) in the following lemma ..
Lemma 6.1. Let
(a) The series defining F, considered as a function of z, converges abso-

lutely and uniformly on compact subsets of C " AT'
(b) F is an elliptic function for the lattice AT' It has a double pole at
each Z E AT and no other poles.
(c) The Laurent series for F around Z = 0 begins
F(u; q) 1
= (21fi)2z2 - { 12
1 - 2 L
n;::':l
(1-qrt}
qrt)2 + (powers of z).
PROOF. (a) Note that
(1 - q rt u)2
We use this identity to rewrite the terms in F having n < O. This gives an
alternative expression for F,
Now let C c C " AT be a compact set. Then u = e 27riz is bounded away

from 0 and 00 uniformly for z E C. Since qrt --t 0 as n --t 00, it follows that
there are constants Cl and C2 so that
for all z E C, n ::::: C2.
This shows that except possibly for the terms with n ::; C2, the series is
absolutely and uniformly convergent on C.
Consider now one of the finitely many terms with n ::; C2. We know
that u of q" for z E C, so the compactness of C ensures that
inf
zEC
11 - qn u±ll > O.
Hence the terms with n ::; C2 are also uniformly bounded.

(b) From (a), F is a holomorphic function on rc " AT, and looking at the
series defining F. it is clear that F has a double pole at each point in AT'
Finally, since the transformations z f---> Z + 1 and Z f---> Z + T correspond
to U f---> II and u f---> qu respectively, it is clear again from the series that F
is an elliptic function for the lattice AT'
(c ) Note that u = e 27riz ---+ 1 as Z ---+ O. Hence the pole at Z = 0 in the
series for F comes from the term wiLh n = O. Now a liLLIe freshman calculus
yields
1 1
(1 - u)2
( . )2 - -
27r~z 12
+ (powers of z).
Hence using the alternative series for F given above, we find
lim {F(ll:q) - (
Z~O
1)2
27rzz
+~}
12
= z--->o
lim {F(u;q) - ( u
1- u
)2}
. {q"U q"U- 1
L + , }
=~n
1l~1 "2"1 (1 - q"U)2 (1 _ qnu-l )2
qn
=2
n2"l
L (1 _ qTl) 2 •
o
Theorem 6.2. Let U = e 27riz and q = e 27riT .
(a)
(b)
Remark 6.2.1. For IXI < 1 there is the elementary identity

X
(1 - X)2 = X dX
d(l)
1_ X
""'
~ mX =
m
.
m2:1
This is sometimes used to rewrite the final sum in (6.2a) as

qn mqm
L
n2:1
(1- qn)2 = L L mqmn = L
n2:1 m2:1 m2:1
1- qm·
PROOF (of Theorem 6.2). (a) Let F(u; q) be as in (6.1). Consider the
function
From (6.1b) we see that this expression is an elliptic function for the lat-
tice AT which is holomorphic on C '- AT. Further, comparing the Laurent
series for F given in (6.1c) with the known Laurent series for ~, we see that
it is also holomorphic at z = 0, and in fact it vanishes there. It thus repre-
sents an everywhere holomorphic elliptic function which vanishes at z = o.
Applying [AEC VI.2.1], we conclude that it is identically zero. This proves
the first equality in (a), and the second is an easy rearrangement of the
terms in the initial sum. (See the proof of (6.1a).)
d . d
(b) Apply -d = 27rZU- to (a). 0
z du
The next step is to find a q-expansion for ((z; T) analogous to (6.2).
By construction,
d
dz ((z; T) = -~(z; T),
so we try integrating the series (6.2a) for ~ term-by-term. Proceeding
blindly, we find
J qnu
(1 - qn u)2
dz = J qnu ~ = _1_ .
(1 - qn u)2 27riu
1
27ri (1 - qnu)
.
Unfortunately, the series
is clearly divergent, the nth term goes to 1 as n ---t 00. But just as in
the original definition of ~, we can improve the convergence by adding a
constant onto each term.
The second expression for ~ in (6.2a) has the form
1 ""' qnu qnu-l
(27ri)2~(Z;T) = ~ (1- qnu)2 +L (1- qnu-I)2 + GI ,
n2:0 n2:1
where
(We will soon identify C 1 more precisely.) Now integrate:
for some constant of integration C 2 = C 2 (q). Note that the last series is
absolutely and uniformly convergent on compact subsets of C " AT, so it
defines a meromorphic function on C. (The proof is identical to the proof
of (6.1a).) Further, the (d/dz)-derivative of this series can be computed
term-by-term and agrees with the series (6.2a) for SJ. This proves that it
equals ( for some choice of C2.
To find C 2 , we compute the first few terms of the Laurent series
around z = O. We already know (5.1b) that
1 .
((z; r) = - - G 4(r)z3 + hIgher powers of z.
z
On the other hand, the pole at z = 0 (i.e., at u = 1) in the above q-series
comes from the n = 0 term, so we find
-+L
-u { -qnu qnu- 1 }
-
1 - U n>l 1 - qnu
+ 1- qnu-
1 -271'iCl z + C2
- '- v .f
vanishes at z = 0 (u = 1)
e 27riz
-----;::2----:-·
1- e 7r'Z
+ C2 + (powers of z)
1 1
= -.- + - + C 2 + (powers of z).
271'lz 2
Since the Laurent series for ( has no constant term, we see that C 2 =- ~.
This proves part of the following theorem.
Theorem 6.3. Let ((z; r) be the Weierstrass (-function and T7 : AT --> C
the quasi-period homomorphism associated to A.
1 _qnu qnu-l 1 1
(a) 271'i((Z;T) = L
n2:0
1- qnu +L
n2:1
1- qnu- 1 + 27ri T7 (l)z - 2'
(b) 1 1 { qn }
-
27rZ( 12 -1+242:( 1 - qn )2.
·)21](1)=-
n2:1
PROOF. Let
We proved above that

1 1
- . ((z; T) = G(z; T) - 27riG1 (q) - -.
27rZ 2
Now evaluate at z = ~. From (5.2b), ((~;T) = 1](1). Further, z = ~
corresponds to u = e rri = -1, so all of the terms in G( ~; T) cancel except
the n = 0 term. Thus
Hence
so
1
G1(q) = - (27ri)21](1).
This completes the proof of both parts (a) and (b). D
Finally, we integrate the series for ((z; T) and exponentiate to obtain

an important q-product expansion for a(z; T).
Theorem 6.4. The Weierstrass a-function has the product expansion
. ) ___1_ ~'7(1)z2 -rriz(l_ )

a ( Z, T
II (1 - q n u)(l- qnu- 1 )
(
-
27rz. e e u 1-) qn 2 '
n2:1
where u = e 2rriz and q = e 2rriT as usual, and 1](1) is the quasi-period

associated to the period 1 E AT.
PROOF. By construction,
a'(z; T) d
( ) = -dZ loga(z; T) = ((z; T).
a Z;T
54 L Elliptic and Modular Functions
Using (6.3a) and integrating gives
loga(z;7) = J ((z;7)dz = J 1
-.((Z;7)-
du
= '~
" J _qnu du
1 - qnu -U + '"
~
J 27ft
qn u -1 du
1 - qnu -1 -U
u
+
J
(77(l)z - 7fi) dz
n;:o.O n;:o. 1
We claim that the series will converge provided we use the principal branch
of the logarithm when evaluating 10g(1 - qnu) and 10g(1 - qn u- 1 ). To see
this, note that for n sufficiently large we have Jq n u ±lJ ~ ~. So, for all but
finitely many n,
Hence the series will converge.

Exponentiating, we eliminate any ambiguity arising from the choice of
a branch of the logarithm and obtain the product representation
a(z; 7) = eh(l)z2-1Tiz+C3 IT (1 - qnu) IT (1 - qnu- 1 ).

TI.;:o.O n;:o. 1
It remains to find C 3 . Recall that a was normalized by the condition

that a(z;7)/Z -+ 1 as Z -+ o. It is the n = 0 term in the product which
vanishes at Z = 0, so we find
. a(z;7)
1 = 11m - - -
Z----""?O Z
= lim e~1)(1)z2-1Tiz+C3 (1 - u)
Z~O Z
IT (1 _ qTl. u )(l _ qnu- 1)
u~l n;:o.l
=e c3 (-27fi) IT(1- qn)2.

n;:o.l
Hence
c
e 3=-27fi
1 IT (1_qn)2'
1
n;:o.l
which gives the desired product formula for a(z; 7). o

§7. q-Expansions of Modular Functions
The Eisenstein series G2dT) is a modular function of weight 2k. It sat-

isfies G 2k (T + 1) = G 2k (T), so it has a Fourier expansion in terms of the
variable q = e 27riT . In this section we will compute the Fourier series of G 2k
and use it to deduce various properties of the Fourier expansions for ~(T)
and j( T).
Proposition 7.1. Let k :::::: 2. Then
where
1
((s) = ~- and
L...J nS
n~l
are respectively the Riemann (-function and the kth-power divisor func-
tion.
PROOF.
G 2k (T) = E (mT
1
+ n)2k
m,nEZ
(m,n);i(O,O)
1 1
E n2k +2EE(mT+n)2k'
00
nEZ m=l nEZ

n;iO
The first sum is just 2((2k). Notice that the rightmost (inner) sum is
clearly invariant under T t--> T + 1. We now compute its Fourier expansion.
Lemma 7.1.1. Let k :::::: 1 be an integer. Then for all T E H,
E ~--~~
1
= (21Ti)2k E(X) r 2k-l e 27rirT
(T+n)2k (2k-l)! .
nEZ r=l
PROOF. Ignoring questions of convergence, we have a formal identity
1 1 d2k
E
nEZ
(T + n)2k = E
nEZ
(2k _ I)! d 2kT 10g(T + n)
1 d 2k
= (2k _ I)! d2kT log II (T + n).
nEZ
Of course, this product does not converge. But we do get convergence if we

factor an n out of each term. (Remember this is just a formal manipulation
to see what the answer should be. Otherwise one might rightly object that
dividing by (00!)2 is a highly dubious procedure.) The product
T II (1 + ~) =
nEZ
T IT (1 - ::)
n=l
no;iO
converges to give a function that is holomorphic on C, has simple zeros

at each integer, and no other zeros. With this description, the reader will
undoubtedly recognize the usual product expansion for the sine function.
(See Ahlfors [1].)
sin(?TT) =?TT IT (1 -::) .

n=l
We now reverse our formal argument to produce a rigorous proof.

Starting with the product expansion of the sine function, we take the log-
arithmic derivative, yielding
d . 1 E<Xl -2T
-d log(sm ?TT) =- + 2 2
T T n=l n - T
--+1<Xl (
- T ~ - - +n-T
n+T -- -1 1) .
Now taking (2k - 1) more derivatives, we find
d 2k
d 2kT log(sin n) = -(2k - I)!
{I T2k
<Xl
+~ (1+ (n T)2k + (n _1) }
T)2k
= -(2k - I)! E (n+T1 )2k'

nEZ
Next we compute the Fourier series of (a branch of) log (sin ?TT). Writ-
ing
sin(?TT) = 1
2i (e" ")
1rtT - e- 1rtT =- 2i "( 1- e2")
1 e- 1rtT 1rtT ,
we find (for T E H)
log(sin ?TT) = - loge - 2i) - ?TiT + log (1 - e21riT )

= -loge -2i) - ?TiT - E<Xl _e21rtrT.
1 "
r=l r
Differentiating 2k times (with k ;:::: 1) yields

Equating this expression for d2k log(sin 7fT) / dT2k with the expression ob-
tained above gives the desired result.
o
We resume the proof of Proposition 7.1. Applying (7.1.1) with mT in
place of T, we find
1
LL
(X)
G 2k (T) = 2((2k) +2 (mT + n)2k

m=l nEZ
( 2 ·)2k
= 2((2k) + 2
(X) (X)
7fZ '"""' '"""' r2k-le27rirmT

(2k -I)! L L
m=l r=l
(2 ·)2k
= 2((2k) + 2 7fZ
(X)
'"""' '"""' r2k-le27rinT.

(2k -I)! L L
o
n=l rln
As is well known, ((2k) is a rational multiple of 7f2k. It is frequently

convenient to factor a 7f2k out of the Fourier series for G 2k , yielding a series
with rational coefficients. We briefly recall the details concerning special
values of the Riemann (-function at even integers.
Definition. The Bernoulli numbers Bk are defined by the power series
expansion
For example, one easily computes
Bo = 1,
and B2k+l = 0 for all k 2: 1.
For a longer table of Bk's and the corresponding values of ((2k), see (A §1).
Proposition 7.2. For all integers k 2: 1,
(X) 1 (27fi)2k
((2k) = ~ n 2k = - 2(2k)! B 2k.
PROOF. First we use the definition of the Bk's to write
7fX cot (7fx) e 7rix + e- 7rix

= 7fix. . = 7fix
e7r2X _ e- 7r2X
(
1+ 2)
--=-2- - : . - -
e 7rtx - 1
_. ~ B (27fix)k _ ~ B (27fix)2k
-7fZX+ L k k! - L 2k (2k)!
k=O k=O
Next we use the product expansion for sin(7rx) already considered in the
proof of (7.1),
sin(7rx) = 7rX ITDC (

1 ~ nx2)
2 .
n=l
Taking the logarithmic derivative yields
1 = 2x 1
7rcot(7rx) =;: + L ~ n2 . 1 ~ (x2/n2)
71,=1
1 2 = (X2) k }
= ;: + ~ ~;: {;
D0 {
n2
1
= ;:{ 1 ~ 2 L
x
((2k)X2k}.
k=l
Comparing the two Laurent series for 7rX cot( 7rx) gives the desired result.
D
Remark 7.3.1. We can now define a normalized Eisenstein series E 2k (T)

as the series
Using (7.1) and (7.2) we see that

G 2k (T) = 2((2k)E2k(T).
The fact that the E2k'S have leading coefficient 1 makes them particularly
easy to compare. For example, E'J and E8 are both modular forms of
weight 8. Since AI'll, has dimension 1 from (3.10.1), we know that they are
multiples of one another. But since they are normalized, we see on com-
paring their constant terms that E'J = E 8. Equating Fourier coefficients
gives the identity
n-l
0"7(n) = 0"3(n) + 120 L 0"3(rn)0"3(n ~ rn).
rn=l
The reader will be able to construct many more identities of this sort.
Remark 7.3.2. We can also write g2(T) and g3(T) in terms of normalized
Eisenstein series:
4 1
g2(T) = 60G4(T) = 120((4)E4(T) = (27r) 223E4(T);
6 1
g3(T) = 140G6(T) = 280((6)E6(T) = (27r) 23 3 3 E6(T).
These expressions are useful for computing the Fourier expansion of 6. ( T)

and j (T), as explained in the next proposition.
Proposition 7.4. (a) The modular discriminant has the Fourier expan-
sion
~(T) = (27r)12 LT(n)qn,
n21
where T(l) = 1 and T(n) E Z for all n. The arithmetic function n f---+ T(n)
is called the Ramanujan T-function.
(b) The modular j-function has the Fourier expansion
. 1
J(T) = - +~
~c(n)qn,
q n20
where c(n) E Z for all n.
PROOF. (a) Using (7.3.2) we compute
We must show that every coefficient of E1- El is divisible by 26 33 = 123.

From (7.3.1) we have
E4(T) = 1 + 240 L 0"3 (n)qn and E6(T) = 1- 504 L0"5(n)qn.

n21 n21
To ease notation, let us write
and
Then
E4(T)3 - E6(T)2 = (1 + 240A)3 -(1 - 504B)2
= 122(5A + 7B) + 12 3(100A 2 -147B2 + 8000A3 ).
It remains to show that every coefficient of 5A + 7B is divisible by 12.
We have
and for any integer d,
Hence 5d 3 + 7d 5 == 0 (mod 12). This proves that

~(T) = (27r)12 LT(n)qn
n21
for integers T(n).

Finally, the coefficient of q is
(2 )12
1~3 ·122. (50"3(1) + 70"5(1») = (271")12,
so T(l) = 1.
(b) We use (a), (7.3.1), (7.3.2), and the definition of j(7) to compute
(271")12E ( )3
( )3 - - 47
j(7) = 1728~ = 1728_-,,1=23_,.....-_ _
6(7) (271")12 LT(n)qn
Since the 0"3(n)'s and the T(n)'s are all integers, this last expression gives a
Laurent series of the form q-1 + E c( n )qn with integer coefficients. (Note
the reciprocal of a power series with integer coefficient and leading term 1
will again have integer coefficients.) 0
Remark 7.4.1. Using the formulas developed in the proof of (7.4), it is

easy to compute the first few values of T(n) and c(n). Thus
(271")- 12 6(7) =q- 24q2 + 252q3 - 1472q4 + 4830q5 + ... ,
j(7) = q-1 + 744 + 196884q + 21493760q2 + ....
For a more extensive list, see (A §2).
Remark 7.4.2. In the next section we will prove that 6(7) has the prod-
uct expansion
6(7) = (271")12q (1- qn)24. II
n~1
This gives an alternative (but less elementary) proof of (7.4a)

Remark 7.4.3. The c(n) coefficients of j(7) have many interesting arith-
metical properties. For example, Lehner [1,2] proved that they satisfy the
following divisibility conditions. (See also Apostol [1, Ch. 4].)
n == 0 (mod 2 e) ==> c(n) == 0 (mod 23e +B),

n == 0 (mod 3e ) ==> c(n) == 0 (mod 32e +3),
n == 0 (mod 5e ) ==> c(n) == 0 (mod 5e+1),
n == 0 (mod 7e ) ==> c(n) == 0 (mod 7e),
n == 0 (mod ne) ==> c(n) == 0 (mod ne).
Remark 7.4.4. The values in (7.4.1) and (A §2) suggest that the c(n)'s
grow quite rapidly. This is indeed the case, as is clear from the following
asymptotic formula proven by Petersson [1] using the circle method of
Hardy, Ramanujan, and Littlewood:
e 47ffo
c( n) '" -vI2=2c- as n
n -3 /-4 --> 00.
It also turns out that the c(n)'s are intimately connected with represen-
tations of the largest sporadic groups, in particular with the Fischer-Griess
monster group. See Conway [1] and Conway-Norton [1] for an interesting
account of this surprising connection.
Remark 7.4.5. Ramanujan's T-function also has many interesting prop-
erties. For example, we will later prove (10.7) that it satisfies the identities
r(mn) = r(m)r(n) if (m, n) = 1,

r(pe+l) = r(p)r(pe) - pllr(pe-l) for p prime and e 2: 1.
These identities were conjectured by Ramanujan; the first proof was given
by Mordell.
We will also prove (11.2) that the r(n)'s grow much more slowly
than the c(n)'s. Precisely, we will show that there is a constant c such
that Ir(n)1 :s: cn 6 for all n 2: 1. Another conjecture of Ramanujan, proven
by Deligne as a consequence of his proof of the Riemann hypothesis for
varieties over finite fields, says that one can do better.
Theorem 7.5. (Deligne [1,2])
for all n 2: 1.
(Here ao(n) is the number of divisors of n. For example, if n is prime,

then ao(n) = 2.)
In the other direction, there is the following open conjecture of Lehmer.
Conjecture 7.6. (Lehmer [1])
r(n) 1= 0 for all n 2: 1.

§8. Jacobi's Product Formula for 6.(7)
In this section we will prove Jacobi's beautiful product expansion for the
mod ular discriminant 6. (7 ) .
Theorem 8.1. (Jacobi)
6.(7) = (211")12q II (1 _ qn)24.

n~l
Remark 8.2. We will derive the product (8.1) directly from the definition
of 6.(7) and the product representation (6.4) for the Weierstrass IT-function.
There are other methods which can be used to prove (8.1). For example, see
Serre [3, Ch. 7, Thm. 6] for a proof based on rearrangement of conditionally
convergent double series, and Apostol [1, Ch. 3, §2] or Siegel [1] for an
exposition of Siegel's clever proof using residue calculations. The heart of
both of these proofs lies in first proving that the function
F(7) = q II (1 - qn)24
n~1
satisfies
Since F visibly satisfies
F(7 + 1) = F(7) and lim F(7) = 0,

T~1.cx)
and since Sand T generate the modular group r( 1), it follows that F is
a cusp form of weight 12. Hence F(7)/6.(7) is a holomorphic modular
function of weight 0, so it is constant. Finally, letting 7 --+ ioo, one easily
checks that this constant is (211")-12.
PROOF (of Theorem 8.1). By definition,
is the discriminant of the cubic polynomial
But we know the roots of this polynomial from [AEC, VI.3.6], namely
§8. Jacobi's Product Formula for ~(7) 63
Thus
The idea of our proof is to express ~(T) in terms of special values of the
Weierstrass O"-function and then use the product expansion (6.4) for 0".
If we differentiate the equation
and divide by 2g:/, we find
. . 1 T T+1 .
Now If we evaluate successIvely at Z = 2' Z = 2' and Z = -2-' we see III
each case that only one of the three terms survives:
p" (~'T) = 2(el - e2)(el - e3),
p" (~, T) = 2(e2 - el)(e2 - e3),
p" (7; 1, T) = 2(e3 - el)(e3 - e2).
Comparing these formulas with the expression for ~(T), we write ~(T) in
terms of values of p",
Recall (5.6b) that we have expressed p' in terms of the Weierstrass 0"-
function,
'( ) __ 0"(2Z,T)
P Z, T -
0" ( Z,T )4·
Taking derivatives gives
"( ) _ _ 20"'(2z, T) 40"(2z, T)o"'(Z, T)

P z, T -
Z,T +
0" ()4 0" ( )5·
Z,T
. 1 T 7+1
If we evaluate p" successIvely at = -, -, - - , the second term will
Z
2 2 2
vanish, since O"(z, T) has zeros at points in the lattice ZT + Z. We obtain
,,(~ T) = -2 O"'(w, T) for w = 1, T, T + 1.

p 2' 0" (w2,7 )4
Combining this with the above formula expressing ~ in terms of pI!, we

find
In order to compute the values of the derivatives in this expression, we

take the transformation formula (5.4c) for o-(z, T) and differentiate it with
respect to z. This gives
o-'(z + w, T) = 1/'(w)ry(w)e1)(w)(z+!w)o-(z) + 1/'(w)e1)(w)(z+!w)o-'(z)

for all w E ZT + Z.
Now put z = 0 and use the fact that 0-(0) = 0 and 0-'(0) = 1 to get
o-'(w) = 1/'(w)e w1)(w)/2. Taking w = 1, w = T, and w = T + 1 in succession
yields
Next we use Legendre's relation (5.2d), which in our situation reads 7'1](1)-
'I](T) = 27Ti, to eliminate 'I](T). After some algebra we obtain
where to ease notation we write 'I] = '1](1).

The next step is to use the product expansion (6.4) for 0- to compute
0- at the half periods. Thus
§8. Jacobi's Product Formula for ~(T) 65
o
In view of the exponent appearing in the product expansion for .6.(T),
it is natural to study the function obtained by taking 24 th -roots.
Definition. The Dedekind TJ-fl1nction TJ( T) is defined by the product
TJ( T) = e 21rir /24 II (1 _ qn) for T E H, q = e21rir .

n21
Warning. Do not confuse the Dedekind TJ-function with the quasi-period

map (5.2b) TJ : A ---> C. This may be especially confusing when A = AT>
since then the symbol TJ( T) has two meanings, and it is quite possible for
both to appear in a single formula. For example,
(27r) 12 (Dedekind 1]( T)) 24 = .6.( T) = product of values of a-( T),
and using the product expansion (6.4) for a will give a formula involving the
quasi-period TJ(l). Why, you may ask, do we continue to use this confusing
notation? Tradition!
Proposition 8.3. (a) The Dedekind TJ-function satisfies the identities
and TJ ( -~) = V-iTTJ(T).
Here we take the branch of V which is positive on the positive real axis.
(b)
PROOF. Note first that (b) is immediate from the definition of TJ(T) and Ja-
cobi's product formula (8.1) for .6.(T). Next, since the transformation T ~
T + 1 does not change q, we see from the definition of TJ(T) that
TJ(T + 1) = e 21ri (r+1)/24 IT (1 - qn) = e21ri/24TJ(T).

n21
Finally, we know that A( T) is a modular form of weight 12, so
Using (b) and taking 24th-roots shows that
for some 24 th _root of unity c. Now evaluate at T = i. Since -Iii = i, we

find that c = 1. D
Remark 8.4. More generally, let
with c:::: o.
Taking the 24th-root of (8.3b) and using the known transformation property
of A(T) shows that
for some integer h) depending on 'Y. For example, (8.3a) says that (S) =
o and (T) = 1. Note that although h) is only defined modulo 24, we
can pin down a particular value for h) by fixing a branch of log 7](T),
setting
if c:::: 0
and requiring (-'Y) = h) if c < O.

For many purposes it is important to know precisely how 7] transforms.
The following theorem of Dedekind supplies the answer. First we need one
definition.
Definition. Let x and y be relatively prime integers with y > O. The

Dedekind sum s(x, y) is defined to be
s(x, y) =
y-l -j (jX
L - - [jX]
- -"21) .
j=l y Y Y
(The square brackets denote the greatest integer function.)

§9. Heeke Operators 67
Theorem 8.5. (Dedekind) Let 'Y = (~ ~) E SL 2 (Z) with c > O. The

Dedekind 1]-function satisfies the transformation formula
where V is the branch of the square root which is positive on the positive
real axis, <Ph) is given by the formula
1 d
<Ph) = -
c
+ -c -12s(d,c),
and s(x, y) is the Dedekind sum defined above.
PROOF. Since we will not need this result, we omit the lengthy proof. The
interested reader might consult Apostol [1, Thm. 3.4J or Lang [2, Ch. IX].
o
Remark 8.6. Dedekind sums s(x, y) satisfy many interesting relations.
Of particular importance is Dedekind's reciprocity law: Let x, y > 0 be
integers with gcd(x, y) = 1. Then
x 1
12s(x,y) + 12s(y,x) = -
y
+ -Yx + -xy - 3.
See Apostol [1, Thm. 3.7J or exercise 1.17. A good source for information
about Dedekind sums is Grosswald-Rademacher [1].
§9. Heeke Operators
Let E/C be an elliptic curve. We have seen amply demonstrated in [AEC]

the importance of studying isogenies connecting our given elliptic curve E
with other elliptic curves. If E(C) ~ C/ A for some lattice A E £." then an
isogeny E' -> E of degree n corresponds to a sublattice A' c A of index n
by the natural map
C/ A' ----> C/ A, Z r--t Z.
In keeping with our general philosophy in this chapter, rather than

focusing on a single isogeny, we instead consider the set of all isogenies
to E of degree n. Equivalently, we look at all sublattices of A of index n.
This is the same as studying degree n maps from E to other elliptic curves,
since we can always take the dual isogeny. In our situation, the dual isogeny
C/ A -> C/ A' is induced by the map z f--t nz. This leads to the notion of a
Hecke operator.
Definition. For any set S, let Div(S) denote the divisor group of the set S,
that is, the free abelian group generated by the elements of S,
Div(S) = EB z· s.
sES
A homomorphism T : Div(S) --+ Div(S) is called a correspondence on S.

Notice that a correspondence is determined by linearity once its values are
known on the elements of S.
Definition. Let n 2: 1 be an integer. The nth Hecke operator T(n) is the
correspondence on the set of lattices £., whose value at a lattice A E £., is
T(n)A = L (N).
A'CA
[A:N]=n
If two lattices are homothetic, then they give the same elliptic curve.
This suggests that we should also look at the following homothety operator.
Definition. Let A E ([*. The homothety operator' R)., is the correspondence
on £., whose value at a lattice A E £., is
R).,A = AA.
Since the T(n)'s and the R).,'s are homomorphisms which map the
group Div(,C) to itself, they can be composed with one another. The fol-
lowing fundamental calculation describes the algebra that they generate.
Theorem 9.1.
(a) R).,Rt" = R)"t" for all A, ME C*.
(b) R).,T(n) = T(n)R)., for all A E C*, n 2: 1.
(c) T(mn) = T(m)T(n) for all m, n 2: 1 with gcd(m, n) = 1.
for p prime, e 2: 1.
PROOF. (a)
(b) This follows immediately from the definitions and from the fact that N
is a sublattice of A of index n if and only if AN is a sublattice of AA of
index n.
(c) Let A" en
A, where the superscript mn denotes the index. Since m
and n are relatively prime, the quotient AI A" has a unique decomposition
with I<pml = m and I<pnl = n.

It follows that there is a unique intermediate lattice A' satisfying
A" cA' CA,

namely
A' = {x E A : mx E A"}.
Using this fact, it is now easy to verify (c).
T(mn)A = L (A") = L L (A")
= L T(m)(A') = T(m)( L (A')) = T(m)T(n)A.
pe+1
(d) Let A E £.,. For a given sublattice A' c A, let a(A') and b(A') be the
integers defined by
a(A') = #{r: A' err:. A} and b(A') = {Io ifif A'A' cpA,
rt. pA.
Then
T(pe)T(p)A = L L (A') = L a(A')(A'),
e e+1
p p p
reA A'e r A'e A
p
e+1
A'e A
T(pe-l)RpA = L (A") = L b(A')(A').

e-1
p
Aile pA
(Note that pA has index p2 in A.) The identity (d) we are trying to prove
is thus reduced to verifying
pe+l
a(A') = 1 + pb(A') for all A' c A.
We consider two cases.

lease 1. A' cpA, b(A') = 11
p
Let rcA. Then r :l pA :l A', so a(A') is increased by one for each
such r. Hence
p
a(A') = #{r : rcA} = p + 1 = 1 + pb(A').
[Quick proof of the middle equality: The set of r A corresponds to t

subgroups of AjpA ~ (ZjpZ)2 of index p (or equivalently of order p).
These are the lines in A,2 (ZjpZ), so there are #1P'1 (ZjpZ) = p+ 1 of them.
For a proof of a more general result, see (9.3) below.]
lease 2. A' rt. pA, b(A') = 01
Let r satisfy A' crt A. Note that pA c r. We have inclusions
0 c A'
c r c
p A
A' npA pA pA'
I I
Not equal, Index p,
since A' rt. pA. since r t A.
But AjpA has order p2, so we conclude that the middle inclusion must be
an equality. Therefore
r = A' + pA.
Thus for a given A' rt. pA there is exactly one r satisfying A' crt A.
Hence
a(A') = 1 = 1 + pb(A').
D
Corollary 9.1.1. Every T(n) is a polynomial in the T(p) 's and Rp's for
primes p. lvIore precisely, the rings
Z[T(n), Rn : nEZ, n ~ 1] and Z[T(p), RI' : p prime]
are the same. This ring is called the Heeke algebra (afr(l)). (Notice that
the Hecke algebra is a subring of the ring of correspondences
End(Div(,C)) = {homomorphisms Div('c) -7 Div('c)}.)
PROOF. Factor n = p11 ... P;". From (9.1a) and (9.1c) we find
r
II T (p:i).
r
Rn = II R~: and T(n) =

i=l i=l
Finally, (9.1d) and an easy induction on e shows that T(pe) is a polynomial

in T(p) and RI" D
Corollary 9.1.2. The Hecke algebra Z[T(n), Rn : nEZ, n ~ 1] is com-

mutative. In particular,
T(m)T(n) = T(n)T(m) for all m, n ~ 1.

(Note that End(Div(£)) is definitely not commutative.)

PROOF. From (9.1a,b,c), we are reduced to showing that T(pe) commutes
with T(pf). This follows from (9.1.1), since both T(pe) and T(pf) are
polynomials in T(p) and R p , which commute from (9.1b). D
Example 9.2. Using (9.1d), it is easy to illustrate (9.1.1) for small pow-
ers T(pe). For example,
T(p2) = T(p)2 - pRp,

T(p3) = T(p)3 - 2pRpT(p),
T(p4) = T(p)4 _ 3pRpT(p)Z + p2 R;.
For a general recursion, see exercise 1.19.
The Hecke operator T(n) sends a lattice A to the sum of its sublattices
of index n. We now describe these sublattices more precisely. Let A E £."
n
and fix an oriented basis ZWI + ZW2 for A. For any A' c A, we choose an
oriented basis w~, w~ for A' and write
with integers a, b, c, d. Then one easily checks that
n = [A : A'] = det (~ :) = ad - bc.
Here's a quick geometric proof of this fact. The linear transformation a =

(~ ~) acting on the vector space ffi.2 ~ ffi.wl +ffi.wz = C sends a fundamental
parallelogram D for C / A to a fundamental parallelogram for C / A'. Hence
[A .. A'] = Area of aD
Area 0 f D
= d ( )
et a .
Conversely, if ad - bc = n, then
is a sublattice of A of index n. We thus obtain a map

n
{a E Mz(Z): det(a) = n} ---> {A' : A' c A}
a=(~~) f---+ a(A) = Z(awl + bw 2 ) + Z(CWI + dw 2 ).
(Here M2 (Z) is the ring of 2 x 2 matrices with integral coefficients.) Note
that a(A) depends on the choice of basis for A, although our notation does
not reflect this dependence.
It is possible, of course, for different a's to give the same sublattice.

According to (1.2), we have a(A) = a'(A) if and only if a = ,a' for
some, E SL 2(Z). Note that the basis for a(A) will be oriented if the basis
for A is, since (1.1) gives
1m (aWl + bW2) = (deta)lm(wl/w;),

CWl + dW2 IC(Wl/W2) + dl
and det(a) = n 2:: 1. This proves the first half of the next lemma, which
we state after setting some notation.
Notation. Let n 2:: 1 be an integer. We define
TIn = { (~ ~) E M 2 (Z) : ad - bc = n} ,
Sn = { (~ ~) E M2(Z) : ad = n, a, d > 0, 0 :s; b < d} .
Note that Sn is a finite subset of TIn having order

d-l
#Sn = LL 1 = al(n).
din b=O
Note also that SL 2(Z) acts on TIn via multiplication: if, E SL 2(Z) and a E
TIn, then detha) = n, so ,a E TIn.
Lemma 9.3. Let A E £.., be a lattice given with a fixed oriented basis A =
ZWl + ZW2·
(a) There is a one-to-one correspondence
SL 2 (Z)\TI n ~ {A' : A' CA}

a = (~ ~) f-------+ a(A) = Z(awl + bw2) + Z(CWl + dw 2).
(b) The natural inclusion Sn C TIn induces a one-to-one correspondence
PROOF. (a) This was proven during the discussion above.

(b) Let a = (~ ~) E TIn. We construct a , E SL2(Z) such that ,a E
Sn. Suppose first that c i- O. Write the fraction -alc in lowest terms,
say -alc = sir. Since rand s are relatively prime, we can find integers p
and q so that ps - qr = 1. Then
(~~)(~ ~)=(~:) and

so we are reduced to the case that c = O. Replacing 0: by -0: if necessary, we

may also assume that a, d > O. Finally, for an appropriate choice of t E Z,
the matrix
satisfies 0 ::; b + td < d, so it is in Sn. This proves that Sn surjects

onto SL 2 (Z)\1J n .
Suppose now that 0:,0:' E Sn have the same image. Thus there is a
'Y = (~ ~) in SL 2 (Z) such that
b') - (a' P b'p + d' q )

( aO db) = (Pr qs) (ao' d' - a'r b'r + d's .
Since a' -I- 0, the lower left-hand entry gives r = O. Next, comparing
diagonal entries, we find
a'd'
a = a'p, d= d's, ps ad = 1, a,d,a',d' > O.
It follows that p = s = 1, and so a = a' and d = d'. Finally, we have
b = b' + d'q and (by assumption) o ::; b, b' < d' = d.
Hence Id'ql = Ib - b'l < d', from which we conclude that q = 0 and b = b'.
Therefore 0: = 0:'. 0
Proposition 9.4. Let A E £., be a lattice, and let A = Zwl + Zw2 be an

oriented basis for A. Then the Hecke operator T(n) is given explicitly by
the formulas
ad=n, a2l
O::;b<d
(The notation o:(A) is as in (9.3a).)

PROOF. Immediate from (9.3), which says that the sublattices of A of
index n are precisely the lattices o:(A) with 0: E Sn. 0
Example 9.4.1. For primes p, (9.4) gives the formula
p-l
T(p)A = (ZPWI + ZW2) + E(Z(Wl + bW2) + Zpw2)'

b=O
§10. Hecke Operators Acting on Modular Forms
In the last section we described Hecke operators T(n) which assign to a

lattice A E L a formal sum of lattices
T(n)A = L (A'),
A'CA
and we also gave homothety operators RA defined by RA(A) = )"A. Letting

F : L --> C be any function on the space of lattices, we define new
functions T(n)F and RAF on L in the natural way,
(T(n)F) (A) = L F(A') and
We would like to define an action of T( n) on the space of modular

functions f of weight 2k. Unfortunately, a modular function f is not a
well-defined function on the space of lattices L; it is only a function on the
space of lattices with given bases:
A= ZWI + ZW2 ~ f (~~) .

However, we can use the fact that f is modular to construct a function
on L having a certain homogeneity property, as described in the following
proposition.
Proposition 10.1. There is a one-to-one correspondence
lattice functions F : L ----> C }
{ weakly modular functions}
f; H ----> C of weight 2k
1-1
+----+
{ satisfying F()"A) = ).. -2k F(A)
for all ).. E C*
f --> Ff(ZWl + ZW2) = W 22k f(wI/ W2),
F.
PROOF. First we check that Ff(A) depends only on A, and not on the
choice of an (oriented) basis for A. From (1.2a), any other oriented basis
has the form
for some (~ ~) E SL2 (Z)i

§10. Heeke Operators Acting on Modular Forms 75
Next, it is clear that

Ff(AA) = Ff(ZAwl + ZAW2) = A-2k Ff(A).
Similarly, if'Y = (~ ~) E SL 2 (Z), then
k'(T = (cr + d)-I (Z(ar + b) + Z(cr + d)) = (cr + d)-I AT'

which implies that
fFhr) = F(A'YT) = F((cr+d)-lA T) = (cr+d)2kF(AT) = (cr+d)2kfF(r).
This shows that the indicated maps are well defined.
Finally, we check that they are inverse to one another, which will prove
that they give one-to-one correspondences between the indicated sets.
Ffp (A) = wi 2k fF ( WI ) = ~F (A wdw2 )

w, ,:w," F ( Z ~: +Z) ~ F(wd w,) d'(A).
D
Using (10.1), we can define Hecke operators on the space of modular
functions of weight 2k. It turns out to be convenient to multiply by the
scalar factor n 2k - 1 , which will prevent the appearance of denominators
in (10.3) below.
Definition. The nth Heeke operator T 2k (n) on the space of (weakly) mod-
ular functions of weight 2k is defined by the formula
(T2k (n)f)(r)=n 2k - 1 L Ff(A')=n 2k - 1 L d- 2kf (ar;b).

A'eAT ad=n,a~l
O::;b<d
Here Ff(ZWl + ZW2) = wi2kf(wI/W2) is as in (10.1). The equality of the

last two expressions is immediate from (9.3), which says that
n } = {Z(ar + b) + Zd : ad = n, a 2: 1, 0 ::; b < d}.
{ A' : A,CAT
Theorem 10.2. Let f be a modular function (respectively modular form,
respectively cusp form) of weight 2k. Then so is T2k(n)f.
PROOF. First, we verify that T 2k (n)f has weight 2k. By definition, T 2k (n)f
is associated as in (10.1) to the lattice function n 2k - 1 T(n)Ff. The scalar
factor n 2k - 1 is immaterial, and we have
(T(n)F)(AA) = L F(A') = L F(AA')
n
A'cA
= r2k L F(A') = A- 2k (T(n)F) (A).

A'eA
Again invoking (10.1), it follows that T(n)Ff corresponds to a weakly mod-

ular function of weight 2k, and so T 2k (n)J is weakly modular of weight 2k.
Next we observe that if J is meromorphic (respectively holomorphic)
on H, then the formula
(T2k (n)J)(T) = n 2k - 1 L d- 2k J (aT: b)

ad=n, a:2:1
O':;b<d
shows that the same is true of T(n)J.

It remains to check the behavior ofT2k (n)J at 00. The next proposition
gives an explicit formula for the Fourier coefficients ofT2k (n)J, from which
it follows by inspection (10.3.2) that T 2k (n)J is meromorphic (respectively
holomorphic, respectively zero) at 00 if J is. This completes the proof
of (10.2). 0
Proposition 10.3. Let J(T) = 2: c(m)qm be a modular function of

weight 2k. Then the Fourier series for T 2k (n)J is
(T2k (n)J)(T) = L ,(m)qm, where ,(m) = L a2k - 1c (:~) .

mEl': al gcd(m,n)
As a special case of (10.3), we list the values of ,(0), ,(I), and ,(p) for
primes p. Notice in particular that ,(I) = c(n). Thus in some sense T2k(n)
acts as a shifting operator on the Fourier coefficients of J.
Corollary 10.3.1. With notation as in (10.3),

(a)
and ,(I) = c(n).
(b) For primes p,
( ) = {c(pn) + p2k-1 c(n/p) ifpin,

, p c(pn) ifp t n.
Remark 10.3.2. Notice that if c(m) = 0 for m ::; -mo ::; 0, then ,(m) = 0
for m ::; -mono This is clear because ,(m) is a sum of terms of the
form c(mn/a 2) with ai gcd(m, n), so mn/a 2 ::; -mo. Thus T2k(n)J will be
meromorphic (respectively holomorphic, respectively zero) at 00 if J is.
§1O. Hecke Operators Acting on Modular Forms 77
PROOF (of Proposition 10.3). We use the formula defining T 2k (n)f and
compute
(T2k (n)f)(T) = n 2k - 1 L d- 2k f (aT: b)

ad=n,a~l
O~b<d
= n 2k - 1 L d- 2k L c(m)e 211"im( aT+b)/d

ad=n, a~ 1 mEZ
O~b<d
= n 2k - 1 L L c(m)d- 2k e211"imaT/d L e27rimb/d.

mEZ ad=n, a~ 1 O~b<d
The innermost sum is

~ e27rimb/d = {do if dim,
L...J if d f m.
O~b<d
C:
Replacing m by md = mnja and using njd = a, we find
(T2k(n)f)(T) = L L a2k-le27rimaT c n) ,
mEZ ad=n, a~ 1
and collecting equal powers of q = e 211"iT (let M = ma) yields
L L a 2k - 1c ( ~2n ) e 27riMT .
MEZ al gcd(M,n) 0
Suppose that a modular function f is an eigenfunction for the Heeke
operator T2k(n). This means that there is a constant oXen) E C so that
(T2k(n)f)(T) = oX(n)f(T) for all T E H.
Using (10.3) to compare the Fourier coefficients of oX(n)f and T 2k(n)f, it

is clear that the eigenvalue oXen) is related in some way to the Fourier
coefficients of f.
Of particular importance are those modular forms which are simul-
taneous eigenfunctions for every T 2k(n). Although it may seem unlikely,
a priori, that there are any such functions, we will later observe (10.9) that
in fact M~k has a basis of such functions. In any case, we can already
construct the following examples.
Example 10.4. The modular discriminant fl.(T) is an eigenfunction for
every Heeke operator T 12 (n), n ~ 1. To see this, note that (10.2) says
that T12(n)fl. is also a cusp form of weight 12. But from (3.10.2) the space
of weight 12 cusp forms MP2 has dimension 1. It follows that T 12 (n)fl. is a
constant multiple of fl..
Similarly, G4(T) and G6(T) are eigenfunctions for T4(n) and T6(n)
respectively, since the spaces M4 and M6 have dimension 1 (3.10.1). In
fact, it is not hard to show that G 2k(T) is an eigenfunction for T2k(n) for
all k ~ 2 and n ~ 1. See exercise 1.25.
We now describe the relationship between the eigenvalues and the

Fourier coefficients of simultaneous eigenfunctions.
Theorem 10.5. Let f(7) = 2: c(m)qm i- 0 be a cusp form of weight 2k,
and suppose that f is an eigenfunction for all Hecke operators T 2k (n), say
T 2k (n)f = A(n)f·
Then
c(l)i-0, and c(n) = A(n)c(l) for all n ~ 1.
PROOF. Comparing the leading coefficient in
A(n)f = A(n)c(l)q + ... and T 2k (n)f = c(n)q + ... ,
(see (10.3.1b)), we find that c(n) = A(n)c(l). This proves the second part
of the theorem.
Suppose now that c(l) = O. Then what we have proven implies
that c(n) = A(n)c(l) = 0 for all n ~ 1, so f = O. This contradicts
our original assumption that f i- O. Hence c(l) i- O. 0
Definition. A simultaneous eigenfunction as in (10.5) is called normalized

if c(l) = 1. In view of (10.5), every simultaneous eigenfunction is a constant
multiple of a normalized eigenfunction.
In the last section we proved several identities (9.1) for Heeke oper-
ators T(n) acting on the space of lattices'c. These give us the following
identities for the action of Heeke operators on modular functions, which in
turn give us relations on the Fourier coefficients of simultaneous eigenfunc-
tions.
Proposition 10.6. Let f be a (weakly) modular function of weight 2k.
(a) for all m, n E Z with gcd(m, n) = 1.
(b) T 2k (pe)T2k (p)f = T 2k (pe+l)f + p2k- 1T 2k(pe-l)f

for all primes p and all e ~ 1.
PROOF. (a) This is immediate from (9.1c).

(b) We apply the identity (9.1d) to the lattice function F f described
in (10.1). Since
we find
§10. Hecke Operators Acting on Modular Forms 79
By definition, Tzk(n)J = n 2k - 1 T(n)Fj , so multiplying by p(e+l)(2k-l) gives

the desired result. 0
Corollary 10.6.1. Let J(T) = L c(n)qn f. 0 be a cusp form of weight 2k

that is a normalized eigenfunction for every Hecke operator TZk (n).
(a) c(mn) = c(m)e(n) for all m, n E Z with ged(m, n) = 1.
for all primes p and e ;::: 1.
PROOF. (a) We combine (10.5) and (10.6a) to find (note that e(l) = 1)
c(mn)J = ),.(mn)J = T2k (mn)J
= T zdm)T2k (n)J = ),.(m),.(n)J = e(m)e(n)f.
(b) This follows similarly from (10.5) and (1O.6b). o

Example 10.7. Let T(n) be Ramanujan's T-function defined by
(See (7.4a) and (8.1).) Then (10.4) says that L T(n)qn is a normalized
simultaneous eigenfunction, so (10.6.1) gives the relations
T(mn) = T(m)T(n) for all m, n E Z with gcd(m, n) = 1.
for all primes p and all e ;::: 1.

These identities, conjectured by Ramanujan, were first proven by Mordell.
Remark 10.8. It is clear from (10.6.1) that modular forms that are si-
multaneous eigenfunctions have many interesting arithmetical properties.
(We will see some additional ones in the next section.) We have given
examples of such functions (e.g., b.(T)), but so far we only know finitely
many such examples. The following theorem of Petersson shows that there
are many functions to which (10.6.1) applies. We will not give the proof,
which requires additional machinery involving subgroups of SL2(Z),
Theorem 10.9. (Petersson [2]) The set
{J E MYk : J is a normalized eigenfunction for all T2k (n), n ;::: I}

is a basis for the space MJk of CllSp forms of weight 2k.
PROOF. See Lang [2, Ch. III §4]' Ogg [1], Shimura [1, Ch. 3 §§4,5]' or
exercise 1.22. 0
§11. L-Series Attached to Modular Forms
Let I (T) = L: c( n )qn i- 0 be a cusp form of weight 2k which is a normalized

eigenfunction for all Hecke operators T2k (n). Then the Fourier coefficients
of I satisfy the identities given in (10.6.1). We now show that these identi-
ties are equivalent to an Euler product decomposition for a certain Dirichlet
series attached to I.
Definition. For any power series
1= L c(n)qn E C[q],
n~l
the L-series attached to I is the (formal) Dirichlet series
L(f, s) = L c(n)n- s .
n~l
Proposition 11.1. Let I = L: c(n)qn be a power series with c(l) = 1.

n~l
Then the coefficients of I satisfy the identities
(i) c(mn) = c(m)c(n) for all m, n with gcd(m, n) = 1,
(ii) c(pe)c(p) = c(pdl) + p2k-l c(pe-l) for primes p and e ~ 1,
if and only if the associated L-series L(f, s) has the Euler product expansion
(iii) L(/)
,s = IT 1 _ c(p)p-s1+ p2k-1-2s·
p
(Note that this is an equality of formal Dirichlet series. We have said

nothing yet about convergence properties.)
PROOF. Suppose first that I satisfies (i) and (ii). The multiplicativity
relation (i) implies that we can decompose L(f, s) into a product over
primes,
If we multiply the inner sum by 1 - c(p)p-S + p2k-1-2s, we find
(1 - c(p)p-S + p2k-1-2s) (2: c(pe)p-es)

e~O
= {c(l)+c(p)p-S} - {c(p)c(l)p-S}
+ 2: (c(pe) - c(p)c(pe-l) + c(pe-2)p2k-l)p-eS
e~2
= 1 using (ii) and c(l) = 1.

Hence
" ( e) -es 1
L...Jc P P = 1 _ c(p)p-S + p2k-1-2s'
e~O
which proves that LU,s) has the Euler product expansion (iii).
We leave the converse as an exercise, since it will not be needed in the
sequel. See exercise 1.23. 0
In order to prove that the formal Dirichlet series LU, s) converges in

some half-plane Re( 8) > 80, we need an estimate for the size of the Fourier
coefficients of I.
Theorem 11.2. (Hecke) Let 1(7) be a cusp form of weight 2k with
Fourier expansion ~ c( n )qn. There is a constant /'i" depending only on I,
such that
for all n ~ 1.
Remark 11.2.1. Let 1(7) = ~c(n)qn be a normalized cusp form of

weight 2k which is a simultaneous eigenfunction for all Hecke operators
T2k(n). Then the Fourier coefficients of 1 actually satisfy the stronger
estimate
Ic(n)1 ao(n)n k -!, s
where ao(n) is the number of positive divisors of n. This is the generalized
Ramanujan conjecture (for r(1)), which was proven by Deligne [1,2] as a
consequence of his proof of the Riemann hypothesis for varieties over finite
fields.
Remark 11.2.2. If 1(7) is a modular form of weight 2k which is not a
cusp form, then the Fourier coefficients of 1 grow at the faster rate
See exercise 1.24.
PROOF (of Theorem 11.2). For any y > 0, we can extract the nth Fourier
coefficient of 1 by integrating
c(n) = 11 e- 27rin (x+iy) I(x + iy) dx.
Hence
Ic(n)1 s e 27rny sup
0~x~1
I/(x + iy)l·
Next consider the (non-negative) real-valued function
Using (1.1) and the fact that f is modular of weight 2k, we see that
for all 'Y E r(l).
Hence
SUp¢(T) = SUP¢(T),
TEH TE:F
where J'is the usual fundamental domain (1.5) for r(l)\H. Further, ¢ is
continuous on J', and
Note the importance of knowing that f is a cusp form, since if c(o) =F 0,

the limit would not exist. It follows that ¢ is bounded on J', and so it is
bounded on all of H.
Let
C = SUp¢(T).
TEH
Then
If(x + iy)1 = ¢(x + iy)y-k S Cy-k for all x + iy E H.
Substituting this estimate for f into the above inequality for Ic( n) I yields
This inequality is valid for all y > 0. In particular, putting y = lin gives
the desired result. 0
Corollary 11.3. Let f be a cusp form of weight 2k. Then the as-
sociated L-series L(f, s) converges to give a holomorphic function in the
half-plane
Re(s) > k + 1.
PROOF. From (11.2) we have
Hence I: c(n)n- S is absolutely convergent provided Re(s) > k + 1. 0
Our next goal is to show that the L-series L(f, s) attached to a cusp
form f has an analytic continuation to all of C and that it satisfies a
functional equation similar to the functional equation satisfied by the Rie-
mann (-function.
Theorem 11.4. (Hecke) Let J (7) be a cusp form of weight 2k.

(a) L(j, s) has an analytic continuation to all ofe.
(b) Let
R(j, s) = (27r)-Sf(s)L(j, s),
where f(s) is the usual f-function. Then
R(j, 2k - s) = (_l)k R(j, s) for all sEe.
PROOF. The f-function is given by the integral
for Re(s) > O.
(For basic facts about f, see Ahlfors [1].) Replacing t by 27rnt in the
integral, we obtain the useful formula
Write J(7) = L-c(n)qn. Multiplying our formula for n- S by c(n) and

summing over all n 2: 1 gives
= (2JrYf(S)-l 1=
o
ts- 1 L c(n)e- 27rnt dt
1=
n2'l
= (27r)Sf(s)-1 t s - 1J(it) dt.
Note that since Ic(n)1 ::; tcn k from (11.2), the quantity
is absolutely convergent for Re( s) > k + 1, so it is permissible for us to

reverse the order of the sum and the integral.
We split the above integral for L(j, s) into two parts. For large t the
integral will converge for all sEe. For small t we replace t by lit and use
the fact that J satisfies
J C) = J(S(it)) = (it)2k J(it).

Thus
from above,
= 11 t s- 1 f(it) dt + /00 t s- 1 f(it) dt
= J~ (~r-l f (D d (~) + /00 t S- 1 f(it)dt
= /00 (_1)k t 2k-S-l f(it) dt + /00 t s- 1 f(it) dt.
This gives us the integral representation
valid a priori for Re( s) > k + 1.

But r(S)-l is holomorphic on C, and by inspection the integral is
absolutely and uniformly convergent for s in any compact subset of C.
(Note that since f is a cusp form, if(it)i goes to 0 like a multiple of e- 21rt
as t --> 00.) Hence this integral gives the analytic continuation of L(J, s)
to C. Finally, we observe that the expression
satisfies c(2k - s, t) = (_1)kc(s, t).

It follows immediately that
R(J,s) = (271rSr(s)L(J,s) = 100 c(s,t)f(it)dt
has the same functional equation, R(J, 2k - s) = (_1)k R(J, s). 0
We record as a corollary the useful integral expression for L(J, s) de-

rived during the course of proving (11.4).
Corollary 11.4.1. Let f(7) be a cusp form of weight 2k. Then
L(J,s) = (271")sr(s)-11°O {t S- 1 + (-1)k t2k-s-l}f(it)dt.
EXERCISES
1.1. Prove that the modular group rei) is the free product of its subgroups (S)
and (ST) of orders 2 and 3.
Exercises 85
1.2. Let TO E Q, and let "( E r(1) satisfy "(00 = TO. Prove that "( sends the set
{T E H : Im(T) > "'}

to the interior of a circle in H which is tangent to the real axis at TO. Prove
that the radius of the circle goes to 0 as '" -> 00.
1.3. Give an example of a Hausdorff space X and a topological group r acting
continuously on X such that the quotient space r\X, taken with the quo-
tient topology, is not Hausdorff. (By definition, the action of r on X is
continuous if the map r x X -> X, ("(, x) I---> "(x, is continuous.)
1.4. For any a E <C, let ga(T) = (T - a)/(T - a).
(a) Prove directly that
and
(As usual, i = e 7ri / 2 and p = e 27ri / 3 .)

(b) Find the largest disk U C H centered at i such that the map
{1, S} \U '---> <C,

is injective. Compute its image and its inverse.
(c) Same as (b) for U centered at p with
{l,ST, (ST)2} \U '---> <C,
1.5. Let T E H be a point satisfying a quadratic equation

T2 - aT + b, a,b E Z, a2 - 4b < O.
Suppose further that Z[T] is the ring of integers of the quadratic imaginary
field Q(T).
(a) Prove that the fractional ideals of Q( T) are in one-to-one correspon-
dence with the lattices L contained in Q(T) which satisfy TL c; L. (In this
context, a lattice is a free Z-module of rank 2.)
(b) Prove that every ideal class is represented by a fractional ideal of the
form
satisfying the following conditions:

(i) x, y E Z with y > 0,
(ii) 4y2 - (4b - a 2 ):<:; (2x + a? :<:; y2,
(iii) Ylx 2 + ax + b.
Conclude that the class number of Q( T) is finite.
(c) Prove that the ideal classes in (b) are distinct provided that we discard
all pairs (x, y) satisfying either of the following conditions:
(iv) 2x + a = -y,
(v) x 2 + ax + b = y2 with 2x + a < O.
(d) Use the above algorithm to compute the class number of the following
quadratic fields:
Q(H), Q(v'-5), Q(V-23), Q(V-29), Q(V-47).

1.6. (a) Prove that the natural reduction map
is surjective.
(b) Define r(N) to be the subgroup of reI) consisting of matrices congru-
ent to 1 (mod N), that is,
r(N)={(ac db) Er(I): a==d==I(mOdN)}.

b == c == 0 (mod N)
Prove that
if N = 2,
ifN ~ 3.
(c) Prove that r(N) is a normal subgroup of r(l) and that
r(1)/r(N) ~ SL 2 (Z/NZ) /{±1}.
1.7. Define subgroups ro(N) and r1(N) of reI) by
ro(N) = { (~ ~) E reI) : c == 0 (mod N) } ,
r 1 (N) = { (~ ~) E r(I) : a == d == 1 (mod N), c == 0 (mod N) } .
(a) Prove that rl(N) is a normal subgroup of ro(N), and show that
rO(N)/rl(N) ~ (Z/NZ)* /{±l}.
(b) Prove that
r 1 (N)/r(N) ~ Z/NZ (taken additively).
(c) Prove the following two formulas:
[reI) : ro(N)] =N I1(1 + p-l).

piN
if N = 2,
if N ~ 3.
Exercises 87
1.8. Let X/C be a smooth projective curve of genus g. For any divisor D =
L nx(x) with real coefficients nx E R, let
[DJ = 2)nx](x) E Div(X)

xEX
be the integer part of D, where [nxJ denotes the greatest integer in n x .

Also let
.c(D) = {J E qX)* : div(f) ~ -D} U {O}.
(a) Prove that £"'(D) ~ £..,([DJ).

(b) Let k ~ 1 be an integer, and let Kx be a canonical divisor on X.
Prove that
{W E O~ : div(w) ~ -D} ~ £..,(kKx + [DJ).
1.9. Let tjJ: H* -> X(l) be the usual projection, and let
Do = HtjJ(i)) + HtjJ(p)) + (tjJ(oo)) E Div(X(l)) 0 Q.
(a) Prove that the map
M2k ------> {w E O~(l) : div(w) ~ -kDo}, f 1-----+ wf
is an isomorphism. Here wf is the k-form described in (3.7a) having the

property tjJ*wf = f(T) (dT)k.
(b) Conclude that M2k ~ £..,(kKx(l) + [kDoJ). Use the Riemann-Roch
theorem [AEC II.5.4J to calculate the dimension of M 2k , thereby giving an
alternative proof of (3.lOc).
1.10. (a) Prove that the set
{G~G~ : a, bE z., a, b ~ 0, 2a + 3b = k}
is a basis for M 2 k.
(b) Conclude that the map
is an isomorphism of graded C-algebras, where we grade qx, YJ by assign-

ing weights wt(X) = 2 and wt(Y) = 3. In particular, the functions G4(T)
and G6(T) are algebraically independent over Co
1.11. This exercise outlines an elementary proof that the modular j-invariant
defines a bijectivemap
j : r(1)\H --+ C.
Fix some jo E C, let H be a large real number, and let ~(H) c H be the
region bounded by the curves
ITI = 1, Re(T) =~, Re(T) = -~, Im(T) = H.
Let a~(H) be the boundary of ~(H), which we take with a counter-
clockwise orientation. Assume for now that j(T) # jo for all T E a~(H).
(a) Prove that
#{T E ~(H) : j(T) = jo} = f= ( /t).

7l't i8:J'(H) J T - Jo
dr.
1
(b) Prove that
·
11m _1
2 . . j'
((T)
) . dT -- 1•
H_oo 7l't 8:J'(H) J T - Jo
(Hint. Use j(T) = j(T+1) = j( -l/T) to cancel out most of the line integral,
and use j(T) = q-l + (power series in q) to evaluate the remaining piece.)
(c) Conclude that j(~) = C and that j is injective on the interior of~.
(d) If j(TO) = jo for some T E a~, use a slightly modified region to show
that j is still injective. Conclude that j maps the quotient r(l)\H bijec-
tively to C.
(e) Use the bijectivity from (d) to prove the Uniformization Theorem (4.3).
1.12. Let A, A' c C be lattices satisfying

G4 (A) = G4 (A') and G 6 (A) = G 6 (A').
Prove that A = A'.
1.13. *Let A = IEwI + 1Ew2 be a lattice given with an oriented basis, and let 'f/ :
A --+ C be the associated quasi-period map. Prove that
'f/(wI) "" 1 'f/(W2) " " 1

~ = L-, L-, (mwl
nEZ mEZ
+ nw2)2' w;- = L-, L-, (mwl
mEZ nEZ
+ nw2)2'
n¥O m¥O m¥O n¥O
N.B. These double series are not absolutely convergent; the order of sum-
mation really does matter.
1.14. (a) Provethat
II
u,vEj,A/A
(p(z + U; A) - p(z + V; A»
2 2 (2N 2 _3)(N 2 _1)
u¥v = ±NN p'(NZ;A)N -l,6,(A) 12 .
(b) Prove that

2 (N 2 _1)(N 2 _3)
II
u,vE.t,A/A
(p(U;A) - p(v;A» = ±N- 2 (N -3),6,(A) 6 .
uiE±v (mod A)
u,viEO (mod A)
Exercises 89
1.15. Let EIC be the elliptic curve associated to the oriented lattice A = ZWl +
Zw2. Recall the Weil pairing
em : E[m] X E[m] --> J-tm
defined in [AEC III §8]. Prove that on
E[m] = m-lA/A C CIA,
the Weil pairing is given by the formula
em (
awl + bw2 , CWl + riMJ2) _
- e
27ri(ad-bc)/m
.
m m
(Hint. Use (5.5) to write the elliptic functions appearing in the definition
of em as products of (J" functions.)
1.16. Let sex, y) be the Dedekind sum defined in §8.
(a) Prove that
s(l,y) = (y - ~;~ - 2).
(b) Derive a similar formula for s(2, y). (The answer will depend on the
parity of y.)
(c) Prove that S(y2 + 1, y) = 0 for all integers y > o.
1.17. Let 'Y = (~ ~) E SL 2 (Z) with c> 0, and let be as in (8.4) and (8.5).
(a) Prove that ("(T) = ("() + 1.
(b) • Prove that ("(S) = ("() - 3 provided that d > O. (Hint. Use the
definition of (8.5) and (8.3a) to show that
2ri ri
2(("(S) - ("()) "24 = log(cSr + d) + log(r) -log(dr - c) - "2.
Now evaluate at r = i.)

(c) Use (b) and (8.5) to deduce Dedekind's reciprocity law (8.6),
12s(x,y) + 12s(y,x) = -xy + -xy + -xy

1
- 3.
1.18. Let
Po = II (1 _ qn), Pl = II (1- qn-~),

n;:::l n~l
P2 = II (1 + qn), P3 = II (1 + qn-~).
n2:1 n2:1
(a) Prove that
{ ~ (~) - ~ (~) } ! = PoP;,
{~G) _~C;l)}! =PoP{,

{~C;l)_~G)}! =2 qkpo p;'
(Hint. Use (5.6a) and the product expansion (6.4) of G.)

(b) Prove that
P 1 P 2 P3 = 1.
(Hint. It's easier to show that POP 1 P2 P3 = Po.)
(c) Use (a) and (b) to prove Jacobi's formula
~(T) = (27r)12q II (1- qn)24.

n2:1
1.19. Verify the following identities for the Hecke and homothety operators acting
as correspondences on the space of lattices £'. (Note that there are similar
identities for the operators T 2k(n) acting on the space of modular forms of
weight 2k which will differ from these identities by various scalar factors.)
(a)
(b) T(pr)T(pS) = LpiR~T(pr+s-2i) for 0:::; T :::; S.
i=O
(c) T(m)T(n) = L dRdT (:~) .

dl gcd(rn,n)
1.20. (a) Let f(T) be a modular function of weight 2k. Prove that
df )2 d2f
9 = (2k + 1) ( -dT - 2k· f ·dr2
-
is a modular function of weight 4k + 4.

(b) If f is a modular form, prove that 9 is a cusp form.
(c) If f is the Eisenstein series G4(T), prove that
Similarly, if f = G 6 (T), prove that 9 = CG4(T)~(T), and find the value of

the constant c.
Exercises 91
1.21. For any matrix a = (~ ~) with real coefficients and det(a) > 0, define
/1(00, r) = cr + d.
For any function f : H ----> C, define a new function fl[abk by
(a) Prove that f is weakly modular of weight 2k if and only if
(b) Prove that
where wf is the differential form described in (3.7).

(c) Verify the identities
(d) Prove that the action of T 2k (n) on a weight 2k modular function f is

given by the formula
(See §9 for the definition of'D n .)

1.22. Let f, 9 E Mfk' The Petersson inner product of f and 9 is defined by the
integral
( -- k dr /\dr
(f,g) = I/(r)g(r)(Imr) -2i(Imr)2'
(Here J" is the usual fundamental domain for r(l)\H. See (1.5).)
(a) Prove that the integral converges. (Note that f and 9 are assumed to
be cusp forms.)
(b) Prove that ( , ) is a positive definite Hermitian inner product on the
complex vector space Mfk'
(c) Let w(j, g) be the integrand
-- k dr/\dr
w(j,g) = f(r)g(r)(Imr) -2i(Imr)2
Prove that for any matrix a with real coefficients and det(a) > 0, and any
functions f, 9 on H,
In particular, if I, g E M2k and "( E SL 2 (Z), then

w(j,g) o"{ = w(j,g).
(See exercise 1.21 for the notation II[ahk.)
(d) • Prove that T 2k(n) is self-adjoint with respect to the Petersson inner
product:
for all I,g E M~k' n ~ 1.
(e) If I,g E M~k are normalized eigenfunctions for every T 2k(n), prove
that either
(j,g) =0 or I=g·
(f) Prove that
{j E M~k : I is a normalized eigenfunction for all T2k (n), n ~ I}

is a basis for M~k'
1.23. Prove that if L(j,s) has an Euler product expansion as in (iii) of (11.1),
then the coefficients of I satisfy the identities (i) and (ii) of (11.1).
1.24. (a) Let G 2k(r) = Ec(n)qn be the Fourier expansion of the Eisenstein
series G 2 k. Prove that there are constants 11:1,11:2 > 0, depending only on k,
such that
for all n ~ 1.
(b) Let I(T) = Ec(n)qn be a modular form of weight 2k which is not

a cusp form (i.e. c(O) =f 0). Prove that there are constants 11:1,11:2 > 0,
depending on I, such that
for all n ~ 1.
1.25. (a) Prove that the normalized Eisenstein series E2k is a normalized eigen-
function for every Hecke operator T 2 k(n). See (7.3.1) for the definition
of E2k.
(b) Let I E M2k be a modular form of weight 2k ~ 4 which is not a cusp
form, and suppose that I is a normalized eigenfunction for every Hecke
operator T2k(n). Prove that 1= E 2k.
1.26. Let f E M2k be a modular form of weight 2k ~ 4 which is not a cusp
form, say f has the Fourier expansion I = c(O) + c(I)q + ... with c(O) =f O.
Let L(j,s) be the L-series attached to I as described in §11.
(a) Prove that L(j, s) can be analytically continued to C" {2k} and that
it has a simple pole at s = 2k with residue
( _I)kc(0)(271')2k
reSB=2k L(j,s) = r(2k)
(b) Let R(j, s) = (271rBr(s)L(j, s). Prove that L(j, s) satisfies the func-
tional equation R(j, 2k - s) = (_l)k R(j, s).
Exercises 93
1.27. Let f(T) be a cusp form of weight 2k with k an even integer.

(a) Prove that
2 ~ c(n)
L(J, k) = (k _ I)! ~ ~r(k, 2·rrn) ,
n~l
where r(s,x) is the incomplete r-function
r(s,x) = l=e- J e- t dt forsEC,x>O.
(b) Prove that

k
L(J, k) = 2(27rl L c(n)e- L 2

11"n (27l"n)m~k _ m)!·
n~J m=J
(Note that this series converges quite rapidly, since from (11.2), Jc(n) I grows
no faster than n k.)
1.28. Let f( T) = 2: c( n )qn be a cusp form of weight 2k, let p be a prime, and let
X : (Z!pZ)* ---> C*
be a primitive Dirichlet character and extend X to Z be setting X(p) = O.
The Gauss sum g(X) associated to X is given by the formula
L X(b)e
p-l
g(X) = 2 11"ib/ p .
b=O
We define the the twist of f by X to be the function
n=l
(As usual, we set x(n) = 0 if gcd(p, n) > 1.) We will denote the associated
twisted L-series by
L(J, x, s) = L(j(X, . ), s) = L c(n)x(n)n- s.
( a) Prove that
L
p-J
x(n) = !g(x) X( -a) e 2 11"ian/ p •
p a=O
(b) Let R(J, X, s) be the function
P)8r(S)
R(J,X,s) = ( 27l" g(x)L(J,x,s).
R(j, X, s) = 1:
Prove that R has the integral representation
{(Pt)S f;(~)t) + (_1)k(pt)2k-S f;~~)t) } ~t.

(c) Prove that L(J, X, s) has an analytic continuation to all of C and that
it satisfies the functional equation
R(J, X, s) = (-l)k X ( -l)R(J, X, 2k - s).
1.29. Let aI, a2, ... be a sequence of complex numbers, and suppose that there
is a constant c > 0 such that Ian I ::; n C for all n. Let ,\ > 0 be a constant
and k > 0 an integer, and define functions
f(7) = Lane27rinT/>'.
n21
(a) Prove that ¢(s) is absolutely convergent provided Re(s) is sufficiently

large.
(b) Prove that f(7) is holomorphic on H.
(c) 'Prove that the following two facts are equivalent:
(I) (s) has an analytic continuation to all ofC, is bounded on every
vertical strip, and satisfies the functional equation
(k - s) = ±(s).
(A vertical strip is a region of the form CI ::; Re( s) ::; C2')

(II) f (7)
satisfies the functional equation
CHAPTER II
Complex Multiplication
Most elliptic curves over <C have only the multiplication-by-m endomor-
phisms. An elliptic curve that possesses extra endomorphisms is said to
have complex multiplication, or CM for short. Such curves have many spe-
cial properties. For example, the endomorphism ring of a CM curve E is
an order in a quadratic imaginary field K, and the j-invariant and tor-
sion points of E generate abelian extensions of K. This is analogous to the
way in which the torsion points of G m (<C) = <C* generate abelian extensions
of Ql. An important result in the cyclotomic theory is the Kronecker-Weber
Theorem, which says that every abelian extension of Ql is contained in a
cyclotomic extension. We will prove corresponding results for a quadratic
imaginary field K. For example, we will show how to construct an elliptic
curve E such that K (j (E)) is the Hilbert class field of K, and we will
explain how to use the torsion points of E to generate the maximal abelian
extension of K.
We have generally not tried to assign credit for the results described
in this chapter but will content ourselves with mentioning Kronecker, We-
ber, Fricke, Hasse, Deuring, and Shimura, who are largely responsible for
that part of the theory of complex multiplication that we will cover. In
particular, the algebraic proofs in §§4 and 5 are essentially due to Deuring,
and the idelic description of complex multiplication in §§8 and 9 is mainly
due to Shimura.
The material included in this chapter barely scratches the surface of
the theory of complex multiplication; a complete treatment of even the
basics would fill (at least) an entire volume. The reader desiring further
information might profitably consult the following sources, as well as the
references they contain. We must especially acknowledge Lang [1 J, Serre [6J,
and Shimura [1], whose expositions strongly influenced our organization of
this chapter.
Borel et al. [1 J: A development of the basic theory of CM using an analytic
approach, together with some useful computational methods.
Cassou-Nogues-Taylor [lJ: The basic theory of CM is developed in the first
few chapters, followed by the use of CM to generate rings of integers.
96 II. Complex Multiplication
Coates [1]: The basic theory of CM is described, followed by an introduc-

tion to the Iwasawa theory of CM elliptic curves.
Lang [1]: Part II develops the theory of CM much as we do, with addi-
tional material on the arithmetic properties of special values of elliptic
and modular functions.
Perrin-Riou [1]: Iwasawa theory for CM elliptic curves.
Serre [6]: A very brief, but beautifully written, summary of the main
theorems of CM for elliptic curves.
Shimura [1]: The idelic formulation of CM for elliptic curves is covered in
Chapter 5 and is extended to abelian varieties in §§5.5 and 7.8. For a
more complete treatment of the theory of complex multiplication on
abelian varieties, see Shimura-Taniyama [1].
Vladut [1]: A nice historical account of Kronecker's Jugendtraum, the
theory of complex multiplication, and the relationship with the theory
of modular forms.
The main prerequisite for this chapter is some familiarity with the basic
theorems of class field theory. We have provided in §3 a resume (without
proof) of the results we will need. We also assume that the reader is familiar
with basic properties of elliptic curves over the complex numbers.
§l. Complex Multiplication over C
In this section we are going to discuss elliptic curves with complex multi-
plication from the viewpoint of complex analysis. Although interesting in
its own right, this should be viewed mainly as the preparation needed to
study arithmetic questions.
Let E IC be an elliptic curve with complex multiplication. We know
from [AEC VI.5.5] that End(E) Q9Q1 is isomorphic to a quadratic imaginary
field and that End(E) is an order in that field. If End(E) ~ R c C
and K = R Q9 QI, then we will say that that "E has complex multiplication
by R" or that "E has complex multiplication by K." We also let
RK = ring of integers (maximal order) of K.
Much of the theory becomes easier if one restricts attention to elliptic curves
with complex multiplication by R K , so we will usually take this course. For
the general theory, see Lang [1] or Shimura [1, Ch. 5].
The uniformization theorem for elliptic curves [AEC VI. 5 .1] says that
for every elliptic curve E IC there is a lattice A C C and an isomorphism
f: CIA ---? E(C)

z f-----+ (SJ(z,A),SJ'(z,A)).
§l. Complex Multiplication over IC 97
We will denote the elliptic curve corresponding to a lattice A by EA; it is

given by the usual Weierstrass equation
If E has complex multiplication, then there are two ways to embed the
order End( E) into C. It is important to pin down one of these embeddings.
This is done by the following proposition, which also provides an important
tool for studying arithmetic properties of various analytically defined maps.
The reader might compare Proposition 1.1 with [AEC 111.5.3], which gives
the case that a E Z. We will use Proposition 1.1 to make deductions in a
manner similar to the way we used [AEC 111.5.3] to deduce [AEC 111.5.4]
and [AEC 111.5.5].
Proposition 1.1. Let E/C be an elliptic curve with complex multipli-
cation by the ring R C C. There is a unique isomorphism
[ .] : R --=::.. End(E)
such that for any invariant differential wE OE on E (see [AEC III §5]),
[a]*w = aw for all a E R.
We say in this case that the pair (E, [ . ]) is normalized.

PROOF. Choosing a lattice A and an isomorphism E ~ EA, it suffices to
prove the proposition for EA. (Note that [AEC III §1, Table 1.2] says
an isomorphism has the effect of multiplying an invariant differential by a
constant.)
Next we recall [AEC V1.5.3] that the endomorphism ring of EA is
isomorphic to
{a E C : aA C A} = R c C.
More precisely, each a E R gives an endomorphism [a] : EA -+ EA deter-
mined by the commutativity of the following diagram:
z ........ az
[aJ
We claim that this map [.] : R --=::.. End(E) satisfies [a]*w = aw.
To verify our claim, we first note that any two non-zero invariant
differentials on EA are scalar multiples of one another. This follows trivially
from the fact that their quotient would be a translation invariant function,
hence would be constant. So if we take any invariant differential w E OE
and pull back via the isomorphism f : C/ A -+ EA(C) , we obtain a multiple

of the invariant differential dz on C/ A, say
f*w = cdz.
Now tracing around the commutative diagram shown above gives the de-
sired result:
[al*w = (1-1)* 0 ¢: 0 j*(w)

= (1-1)* 0 ¢:(cdz) = (1-1)*(cadz) = aw.
o
Corollary 1.1.1. Let (E1' [ . 1E,) and (E2' [ . 1E2 ) be normalized elliptic
curves with complex multiplication by R, and let ¢ : E1 -+ E2 be an
isogeny. Then
for all a E R.
PROOF. Let 0 -=f. w E OE2 be an invariant differential. Then
(¢ 0 [alE,)* w = [alEl (¢*w)

= a¢*w since ¢*w is an invariant differential on E1
= ¢*aw
= ¢* ([alE 2 w)
= ([alE2 0 ¢)* w.
Every non-zero isogeny E1 -+ E2 is separable (we're working in character-

istic 0), so [AEC II.4.2cl says that the map
Hom(E1' E2) ---+ Hom(OE 2 , OE,),
is injective. Therefore ¢ 0 [alEl = [alE2 0 ¢. o

We have seen in Chapter I that in order to understand particular ellip-
tic curves, it is often useful to study the set of all elliptic curves. Similarly,
in order to study a particular elliptic curve with complex multiplication,
it turns out that one should look at the set of all elliptic curves with the
same endomorphism ring. Of course, by "elliptic curves" we really mean
isomorphism classes of elliptic curves, which leads us to define the following
set:
e.G.G(R) = {elliptic curves E /C with End(E) ~ R}
isomorphism over C
{lattices A with End(EA) ~ R}
homothety
§1. Complex Multiplication over iC 99
If we start with a quadratic imaginary field K, how might we construct

an elliptic curve with complex multiplication by RK? If a is a non-zero ideal
of R K , or more generally if it is a non-zero fractional ideal of K, then using
the embedding a eKe C we see that a is a lattice in C. (This is clear
from the definition of fractional ideal, which for quadratic imaginary fields
implies that a is a Z-module of rank 2 which is not contained in JR.) Hence
we can form an elliptic curve Ea whose endomorphism ring is
End(Ea) ~ {a E C : aa C a}
= {a E K : aa C a} since a C K
since a is a fractional ideal.
Thus each non-zero fractional ideal a of K will give an elliptic curve

with complex multiplication by R K . On the other hand, since homothetic
lattices give isomorphic elliptic curves, we see that a and ca give the same
elliptic curve in U:... .G(RK). This suggests that we look at the group of
fractional ideals modulo principal ideals, which the reader will recognize as
one of the fundamental objects of study in algebraic number theory:
e.G(R K ) = ideal class group of RK

{non-zero fractional ideals of K}
{non-zero principal ideals of K} .
If a is a fractional ideal of K, we denote by a its ideal class in e.G(RK).

We have seen that there is a map
More generally, if A is any lattice with EA E £.G.G(RK) and a is any non-

zero fractional ideal of K, we can form the product
We will now prove the elementary, but crucial, fact that this induces a
simply transitive action of the ideal class group e.G(RK) on the set of
elliptic curves £.G.G (R K ). This proposition forms the basis for all of our
subsequent work on complex multiplication.
Proposition 1.2. (a) Let A be a lattice with EA E £.G.G(RK)' and let a
and b be non-zero fractional ideals of K.
(i) aA is a lattice in C.
(ii) The elliptic curve EaA satisfies End(E"A) ~ RK.
(iii) EaA ~ EbA if and only if a = b in e.G(RK).
Hence there is a well-defined action of e.G(RK ) on £.G.G(RK) determined
by
(The reason for using ll-l instead of II will become apparent below.)
(b) The action ofe,qR K ) on C£.,,qRK) described in (a) is simply transi-
tive. In particular,
PROOF. (a) (i) By assumption, End(EA) = R K , so RKA = A. Choose a

non-zero integer d E Z so that da c RK, which is possible by the definition
1
of fractional ideal. Then llA C dA, so llA is a discrete subgroup of C. Simi-
larly, choosing a non-zero integer d so that dRK C ll, we find that dA ellA,
hence llA spans C. This proves that llA is a lattice.
(ii) For any a E C and any fractional ideal a =f. 0, we have
Hence
End(EIlA) = {a E C aaA C aA}

= {a E C : aA C A} = End(EA) = RK.
(iii) From [AEC VI.4.1.l], the isomorphism class of EllA is exactly deter-
mined by the homothety class of aA. In other words, EllA ~ EbA if and
only if there is acE C' such that llA = cbA. Multiplying by a- 1 and using
the fact that RK A = A, we see that
Similarly, multiplying by c- 1 b- 1 gives
Hence if EllA ~ EbA, then both Cll- 1 b and c- 1 ab- 1 take A to itself, so they
are both contained in RK, and hence are equal to RK. Therefore
II = cb,
from which we see immediately that c E K and a = b. This completes the

proof of (iii).
Finally, the trivial observation
shows that the definition a * EA = EIl-l A gives a group action of e£.,(RK)

on C£.,,qRK).
§l. Complex Multiplication over C 101
(b) Let EAl and EA2 be two elliptic curves in £'cL(RK). To show that
the class group e,C(R K ) acts transitively on £,C,C(R K ), we must find a
fractional ideal a with the property a * EAl = EA 2 • Choose any non-zero
1
element >'1 E AI, and consider the lattice al = Al AI. From [AEC VI.5.5]
we see that al is contained in K, and by assumption it is a finitely gen-
erated RK-module, hence it is a fractional ideal of K. Similarly, choosing
1
a non-zero A2 E A2, we obtain a second fractional ideal a2 = A2 A2 of K.
Then
Note the last equality follows from the fact that homothetic lattices give iso-
morphic elliptic curves. This shows that the action of eL(RK ) on £'cL(RK)
is transitive.
To prove that the action is simply transitive, we must show that if
a * EA = b * EA, then a = b. But this is immediate from part (ii) of (a).
o
We have already seen two sorts of elliptic curves which have complex
multiplication, namely the curves with j = 0 and j = 1728 whose auto-
morphism groups are strictly larger than {±1}. (See [AEC lILlO.l].) Now
we'll look at these curves from a complex analytic viewpoint.
Example 1.3.1. Let A = Z[i] be the lattice of Gaussian integers. Then
the endomorphism ring of EA is Z[i]. In particular, Aut(EA) ~ {±1, ±i},
so our general theory [AEC IILlO.I] tells us that j(EA) = 1728. But we
can see this directly in the following way. The lattice A satisfies iA = A.
Hence
g3(A) = g3(iA) = i 6 g3(A) = -g3(A),
so g3(A) = o. Therefore EA is given by the Weierstrass equation
from which we see immediately that j(EA ) = 1728.

Since j(EA) is rational, we know that EA is isomorphic over C to an
elliptic curve defined over Ql; for example, it is isomorphic to the curve y2 =
x 3 + x. But it does not follow that g2(A) itself is in Ql. In fact, a theorem
of Hurwitz [1] says that
g2 (Z[i]) = 64 (11
o
~
dt
1 - t4
)4
Example 1.3.2. Similarly, let p = e27ri / 3 be a primitive cube root of unity,

and let A = Z[p] be the associated lattice. Then pA = A, so
g2(A) = g2(pA) = p4g2(A) = pg2(A),
and hence g2(A) = o. Thus EA is given by the equation
EA : y2 = 4x 3 - g3(A),
so j(EA ) = O. This confirms [AEC III.lO.l], since Aut(EA) = Z[p]* =
{±l, ±p, ±p2}. Further, we see that EA is C-isomorphic to the curve
y2 = x 3 + 1, which is defined over {Q.
If E has complex multiplication by K, we will eventually use torsion
points of E to generate abelian extensions of K. We could restrict ourselves
to studying points of order m for various integers m, but because E has
complex multiplication, there are other natural finite subgroups to look at.
In general, if a is any integral ideal of R K , we define
E[a] = {p E E : [alP = 0 for all a E a}.
We call E[a] the group of a-torsion points of E. For example, if a = mRK,
then E[a] is just E[m]. Notice that the definition of E[a] depends on
choosing a particular isomorphism [.] : RK .::. End(E); we always choose
the normalized isomorphism described in (1.1).
If a is an integral ideal of RK, then A c a-I A. This means that there
is a natural homomorphism
Z f-----+ Z,
which in turn induces a natural isogeny

EA ---+ a * EA·
The following useful proposition gives a precise description of this isogeny
and of E[a].
Proposition 1.4. Let E E £.GqRK), and let a be an integral ideal
ofRK·
(a) E[a] is the kernel of the natural map E ~ a * E.
(b) E[a] is a free RK/a-module of rank 1.
PROOF. Let A be a lattice corresponding to E. Fixing an analytic isomor-
phism C/ A ~ E(C), we find that
E[a]~{zEC/A: az=OforallaEa}
= {z E C : az E A for all a E a} / A
= {z E C : za C A} / A
= a-IA/A
= ker (C/A ~ C/a-IA)

= ker(E ~ a * E).
§1. Complex Multiplication over ([ 103
(b) Continuing with the notation from (a), we choose a non-zero lattice
element'x E A. Then [AEC VI.5.5] says that the lattice (1/ 'x)A is contained
in K, and it is a finitely generated RK-module, so it is a fractional ideal
of K. Since homothetic lattices give isomorphic elliptic curves, we may
assume that A is a fractional ideal of K.
From (a) we know that E[a] ~ a-lA/A as RK/a-modules. Note that
if q is any integral ideal dividing a, then the fact that RK A = A implies
(a-lA/A) Q9RK (RK/q) ~ a-lA/(A+ qa-lA) = a- l A/qa- 1 A.
Hence if we use the Chinese Remainder Theorem to write
RK/a ~ II RK/pe(p), then E[a]~ II a-lA/pe(p)a-lA.
p prime p prime
So it suffices to prove that if b is a fractional ideal of RK (such as b = a-I A)

and if pe is a power of a prime ideal, then b/peb is a free RK /pe-module of
rank one.
To ease notation, we momentarily write
and
Notice that R' is a local ring with maximal ideal p'. (In fact, the only
ideals in R' are (0), p,e-l, ... , p', (1).) Consider the quotient
b'/p'b' ~ b/pb as a vector space over the field R'/p' ~ RK /p.
We claim that it is a one-dimensional vector space.
First we observe that any two elements of bare RK-linearly dependent,
so the dimension of b/pb over RK/P is at most one. On the other hand,
if the dimension were zero, then we would have b = pb, which is absurd.
Hence the dimension is one. By Nakayama's lemma (Atiyah-MacDonald [1,
Prop 2.8]) applied to the local ring R' and the R'-module b', it follows
that b' is a free R'-module of rank one. This completes the proof of Propo-
sition 1.4.
o
We can use (1.4) to compute the degree of the isogeny E ---> a * E, as
well as the degree of an endomorphism [0:] : E ---> E.
Corollary 1.5. Let E E £.c.qR K ).
(a) For all integral ideals a C R K, the natural map E ---> a * E has de-
gree N{§ a.
(b) For all 0: E RK, the endomorphism [0:] : E ---> E defined in (1.1) has
degree IN{§ 0:1·
PROOF. Both parts are immediate from (1.4). For example,
deg(E ---> a * E) = #E[a] from (1.4a)
K
= N,Qa from (l.4b).
Similarly,
deg[a] = #ker[a] = #E[aRK] = N{§(aRK) = IN{§al.

D
Remark 1.6. Before going on to arithmetic questions, we want to make

one brief remark about terminology. The classical name for the j-invariant
of an elliptic curve with complex multiplication is a singular j -invariant.
This terminology, meant to single out such j-invariants as being unusual,
is somewhat unfortunate, since it suggests that the elliptic curve itself has
singularities. We will not use the word "singular" in this sense, but the
reader should be aware of this usage, since it is still fairly common.
Notice that an elliptic curve defined over a finite field always has a
"singular" j-invariant, since its endomorphism ring is always larger than Z
[AEC, V.3.1]. In those rare cases that the endomorphism ring is a quater-
nion algebra, the singularity is especially exceptional, which explains the
origin of the term "supersingular" to describe such curves.
§2. Rationality Questions
In this section we will study the field of definition for complex multiplica-
tion elliptic curves and their endomorphisms. We begin by showing that
every elliptic curve with complex multiplication is defined over an algebraic
extension of Q.
Proposition 2.1. (a) Let EIC be an elliptic curve, and let (j : C ----+ C be
any field automorphism of C. Then
End(EO") ~ End(E).
(b) Let E IC be an elliptic curve with complex multiplication by the ring

of integers RK of a quadratic imaginary field K. Then j(E) E Q. (Later
we will show that j(E) is an algebraic integer. See (II §6) and (V. 6. 3).)
(c)
U.J£.,(R K ) ~ {elliptic curves EIQ with End(E) ~ RK}.
isomorphism over Q
(Note that the original definition of £££.,(R K ) is in terms of isomorphism
classes of elliptic curves over C, not over Q.)
PROOF. (a) This is clear, since if ¢ : E ----+ E is an endomorphism of E,

then ¢O" : EO" ----+ EO" is an endomorphism of EO".
(b) Let a E Aut(C) be as in (a). Now E" is obtained from E by letting a

act on the coefficients of a Weierstrass equation for E, and j(E) is a rational
combination of those coefficients, so it is clear that
On the other hand, (a) implies that End(E") ~ RK, so (1.2b) implies
that E" is in one of only finitely many ([:-isomorphism classes of elliptic
curves. Since the isomorphism class of an elliptic curve is determined by
its j-invariant [AEC III.1.4b], it follows that j(E)" takes on only finitely
many values as a ranges over Aut(C). Therefore [Q(j(E)) : Q] is finite,
so j(E) is an algebraic number.
(c) For any subfield F of C, let us momentarily denote by e,CJ:.JF(RK) the
set
e["[,,F(RK) ~ {elliptic cu~ves E/~ with End(E) ~ RK}.
IsomorphIsm over F
If we fix an embedding Q C C, then there is a natural map
We need to show that this map is a bijection.

Let E/C represent an element of e["[,,rc(RK). Then we have:
(i) j(E) E Q, from (b);
(ii) there is an elliptic curve E' /Q(J(E)) with j(E') = j(E), from
[AEC III.1.4c];
(iii) E' is isomorphic to E over C, from [AEC III.1.4b].
These three facts imply that seE') = E, which proves that s is surjective.
Next let EdQ and E2/Q represent elements of c["[,,iQ(R K ), and sup-
pose that s(E 1 ) = S(E2)' Then j(El) = j(E2) from [AEC III.1.4b]' and
another application of [AEC III.1.4b] says that E1 and E2 are isomorphic
over Q. Hence El and E2 represent the same element of e["[,,iQ (RK), which
shows that s is also injective. 0
Next we study the effect that field automorphisms have on the maps
[a] : E ----t E described in (1.1). In particular, we will find a field of defini-
tion for these maps. Note that if is an endomorphism of E and a is any
automorphism of C, then <1>" will be an endomorphism of E".
Theorem 2.2. (a) Let E/C be an elliptic curve with complex multiplica-
tion by the ring R C C. Then
for all a E R and all a E Aut(C),
where the isomorphisms [']E : R ~ End(E) and [']E~ : R ~ End(E")

are normalized as in (1.1).
(b) Let E be an elliptic curve defined over a field L c C and with com-
plex multiplication by the quadratic imaginary field K c C. Then every
endomorphism of E is defined over the compositum LK.
(c) Let Ed Land E2 / L be elliptic curves defined over a field L c C. Then
there is a finite extension L' / L such that every isogeny from E1 to E2 is
defined over L'.
PROOF. Let w E OEo" be a nOll-zero invariant differential on E. Then the

normalization described in (1.1) says that
[a]~w = ow for all 0 E R.
Further, wO" is an invariant differential on EO", so again from (1.1) we get
for all ;3 E R.
Now for any a E R and any (J" E Aut(C), we compute
Thus [a] EO" and [aCT] E~ have the same effect on the invariant differential WCf •
Now we use [AEC II.4.2c]' which says that the natural map
End(E") ---> End(nEa),
is injective. (Note we are working in characteristic 0, so all finite maps are

separable.) This proves that [a]E<T = [aIT]w.
(b) Let (J" E Aut(C) be an automorphism of C that fixes L. Since E is
defined over L, we can take a ~Weierstrass equation for E with coefficients
in L, so E Cf = E. Then (a) says that for all 0 E R,
If in addition (J" fixes K, then aa = a. This proves that

[OlE IT = [alE for all (J" E Aut(C) such that (J" fixes LK.
Hence the endomorphism [a] is defined over LK.

(c) As in (b), we take Weierstrass equations for E1 and E2 with coefficients
in L. Let ¢ E Hom(E1' E 2) be an isogeny. Then for any (J" E Aut(C) such
that (J" fixes L, we have ¢CT E Hom(E1,E2). Note that deg¢a = deg¢.
From [AEC III.4.11], we see that an isogeny ¢ E Hom(E 1, E 2 ) is determined
by its kernel, at least up to an automorphism of E1 and E 2. Since E1 has
only finitely many subgroups of any given finite order, and since Aut(E1 )
and Aut(E2 ) are finite, it follows that Hom(E 1 , E 2 ) contains only finitely
many isogenies of a given degree. Therefore the set
{¢CT : (J" E Aut(C), (J" fixes L}

is finite, which implies that ¢ is defined over a finite extension of L. Fi-

nally, we observe from [AEC III.7.5] that Hom(El' E 2 ) is a finitely gener-
ated group, so it suffices to take a field of definition for some finite set of
generators. 0
Remark 2.2.1. Notice that the proof of (2.1b) together with the estimate
in (1.2b) shows that if End(E) ~ RK, then
where hK = # e£(RK) is the class number of K. We will prove later (4.3)

that this is an equality. In particular, j(E) is in Q if and only if K has
class number 1. For a complete list of these Q-rational j-invariants, see
Appendix A §3.
Remark 2.2.2. In view of (2.2.1), we see that if RK has class number 1,
then E has a model defined over Q. We have already seen examples of this
in (1.3.1) and (1.3.2), where we looked at curves with complex multiplica-
tion by Z[i] and Z[p]. (Here p = e 271"i/3.) We can also illustrate (2.2) for
these curves. For example, to normalize the curve
we use the isomorphism [.j : Z[ij -> End(E) determined by
[i](x,y) = (-x,iy).
To see that this is the correct normalization, we compute
[ij*dx = d(~x) =i dx .
y zy y
If a E Aut(CC) is complex conjugation, then
([i](x,y))"" = (-x,iy)O" = (_xO",iO"yO")

= (_xO",_iyO") = [_i](xO",yO") = [iO"](xO",yO").
Hence [W equals [iO"], as it should by (2.2).

Similarly, for the curve
we take the isomorphism [.] : .z[p] -> End(E) determined by
[p](x, y) = (px, y).

Remark 2.2.3. There is an interesting converse to (2.1b). Suppose that

A = Zwl + Zw2 is a lattice satisfying
that is, WI/W2 is an algebraic number of degree at least 3 over 1Ql. Then one
can show that j(EA ) is a transcendental number. Notice the analogy with
the Gel'fond-Schneider theorem, which says that if a E Q with a i=- 0,1
and if f3 satisfies 2 ::; [1Ql(f3) : 1Ql) < 00, then a{J is transcendental. The
transcendence of j(EA) was first proven by Schneider. The interested reader
will find a proof of this fact in Schneider [1, Thm. 17] or Waldschmidt [1,
Cor. 3.2.4]. For a general account of the transcendence properties of elliptic
and modular functions, see for example Waldschmidt [1, Ch. 3].
It is an immediate consequence of (l.4b) and (2.2b) that the torsion
points of E generate abelian extensions of K (j (E) ). Before giving the
proof, we remind the reader of the analogous result for cyclotomic fields.
Thus let (E C* be a primitive Nth-root of unity and let (j E Gal(IQl«()/IQl).
Then (17 is another primitive Nth_root of unity, say C = (PC 17 ), and it is
an easy matter to check that the map
p: Gal(IQl«()/IQl) ---> Aut(ILN) ~ (7!../N7!..)*
is an injective homomorphism. (Here ILN = e

is the group of Nth_roots
of unity.) Hence 1Ql«()/1Ql is an abelian extension. We now prove the same
thing for elliptic curves. (For another proof, see exercise 2.6.)
Theorem 2.3. Let E Ie be an elliptic curve with complex multiplication
by the ring of integers RK of the quadratic imaginary field K, and let
L = K(j(E), E tars )
be the field generated by the j-invariant of E and the coordinates of all

of the torsion points of E. Then L is an abelian extension of K(j(E»).
(N.B. In general, L will not be an abelian extension of K.)
PROOF. To ease notation, let H = K(j(E)). Further let
Lm = K(j(E), E[m]) = H(E[m])
be the extension of H generated by the m-torsion points of E. Since L is

the compositum of all of the Lm's, it suffices to show that Lm is an abelian
extension of H.
As usual, there is a representation
p: Gal(K / H) ---> Aut(E[m])

determined by the condition
p(a)(T) = T a for all (J E Gal(R/H) and all T E E[m].
(See [AEC III §7].) For an arbitrary elliptic curve, all we would be able to
deduce from this is that Gal(L m/ H) injects into the automorphism group
of the abelian group E[m], so we would find that Gal(L m/ H) is isomorphic
to a subgroup of GL 2 (Z/mZ).
But the fact that our elliptic curve has complex multiplication gives us
additional information. We take a model for E defined over H = K(j(E)),
and then (2.2b) says that every endomorphism of E is also defined over H.
So elements of Gal(Lm/ H) will commute with elements of RK in their
action on E[m]:
for all (J E Gal(LTn/ H), T E E[m], and a E RK.
In other words, p is actually a homomorphism from Gal( R / H) to the group

of RK /mRK-module automorphisms of E[m]. Hence p induces an injection
Now we use (l.4b), which says that E[m] is a free RK/mRwmodule

of rank one. This implies that
and hence Gal(Lm/ H) is abelian. o
Before proceeding with the general theory, we will pause to construct a

few more examples of elliptic curves having complex multiplication. More
precisely, we will find all elliptic curves that possess an endomorphism of
degree 2. We already know one such curve, namely the curve y2 = x 3 + x
with complex multiplication by Z[i], since the map [1 + i] has degree 2.
From (1.5b), we need to find all quadratic imaginary fields K that have an
element a E RK satisfying IN[fal = 2. This is an easy exercise (which we
leave to the reader), the answer being that there are three such fields:
K =Q(R), RK = Z [R], a = 1 + R;
K = Q (vC2), RK = Z [vC2], a= vC2;
K = Q (V-'l), RK =Z[l+P]
2' a= l+P
2
Since all three of these rings RK have class number 1, we know from (2.2.1)
that the corresponding elliptic curves have j-invariants in Q.
How can we find equations for these curves and their endomorphisms?
It is possible to proceed analytically, but we will take another approach.
If ¢ : E -> E has degree 2, then its kernel E[¢] consists of two points, 0
and a point of order 2. If we move the point of order 2 to (0,0), then E
will have a Weierstrass equation of the form
For this elliptic curve E, we have already determined an elliptic curve E'
and an isogeny ¢ : E -> E' whose kernel is {O, (0, O)}, namely
(See [AEC III.4.5]' although note we have taken the negative of the isogeny
defined there and have substituted x 3 + ax 2 + bx for y2.) Hence E will
possess an endomorphism of degree 2 if and only if this E' is isomorphic
to E.
To see when E and E' are isomorphic, we set their j-invariants to be
equal and solve for a and b, or more precisely for the ratio a 2 lb. Now
'(E) = 256(a 2 - 3b)3 and '(E') = 16(a2 + 12b)3

J b2(a2 - 4b) J b(a2 _ 4b)2 .
Setting j(E) = j(E'), we find after some calculation that
The first two cases, b = 0 and a2 - 4b = 0, give singular curves, so

we discard them. The third case, a = 0, gives the curve y2 = x 3 + bx
with j(E) = 1728 and complex multiplication by Z[i].
Next consider the case a 2 - 8b = O. Taking b = 2 and a = 4 gives the
curve
E : y2 = x 3 + 4x 2 + 2x
with j (E) = 8000 = 26 53 . Similarly, E' is given by the equation
E' : y2 = X3 - 8X 2 + 8X
with j-invariant j(E') = 8000. Hence E and E' are isomorphic, and we
easily find an isomorphism
E' --+ E, (X Y)
,
f-----+ (-~X _ _
2' 2H l_y) .
Composing the isogeny E -+ E' with this isomorphism, we obtain the

desired endomorphism of E of degree 2:
E : y2 = x 3 + 4x2 + 2x E
(x,y) ( -~2 (X+4+~) , - - y (1- ~)).
x 2yC2 x 2
There remains the case 16a4 - 81a 2 b + 324b2 , which (taking a = 36)
leads to the elliptic curves
E: y2 = x 3 + 36x 2 + 18 (9 + 50) x,
E': y2 = X 3 -72X2 + 72 (9 - 50) X
with j(E) = j(E') = -3375 = -3 3 53 . We will leave it to the reader to

make the appropriate variable changes which lead to models for E and E'
over Q and to an explicit formula for the corresponding endomorphism
of degree 2. The final answer is given in the following summary of our
calculations.
Proposition 2.3.1. There are exactly three isomorphism classes of el-
liptic curves over <.C which possess an endomorphism of degree 2. The
following are representatives for these curves and endomorphisms.
(i) E: y2 = x3 + x, j = 1728, a = 1+O,
(ii) E: y2 = x 3 + 4x 2 + 2x, j = 8000, a =.;=2,
(iii) E : y2 = x3 - 35x + 98, j = - 3375, a=

l+H
2
We now resume our development of the general theory of complex

multiplication. From here on we will use (2.1) to identify ££..,qRK) with
the Q-isomorphism classes of elliptic curves having complex multiplication
by RK. Then there is a natural action of Gal(K/K) on ££"'£"'(RK) defined
by the property that a E Gal( K / K) sends the isomorphism class of E to
the isomorphism class of El7. On the other hand, (1.2b) says that the action
of the class group e£"'(RK) on ££"'£"'(RK) is simply transitive, so there is
a unique a E e£.,(RK ) , depending on u, such that a* E = ElI. In other

words, there is a well defined map
characterized by the property
ElI=F(u)*E for all u E Gal(R / K).
It is by studying this map F that we will be able to precisely describe

the field K(j(E)). An easy property of F, which we will prove below, is
that F is a homomorphism. A much deeper property, which we will also
prove, is that F is independent of the choice of the curve E E ££'-£.,(RK).
The astute reader will have noticed that F is actually well defined on the
larger group Gal(Q/Q). However, it is only on the smaller group Gal(R / K)
that F will be independent of E.
Before proving these basic properties about F, we want to stress that
the definition of F has an essential analytic component, since F(u) depends
on the way in which the lattice of an elliptic curves changes when the lattice
is multiplied by an ideal. Thus if we denote by j(A) the j-invariant of
the elliptic curve EA, then as described in Chapter I, j(A) is an analytic
function of A. The map F is then characterized by the formula
so F converts the algebraic action of u into the analytic action of multipli-

cation by F (u ) -1 .
Proposition 2.4. Let K /Q be a quadratic imaginary field. There exists
a homomorphism
uniquely characterized by the condition
ElI=F(u)*E for all u E Gal(R/K) and all E E e£'-£'-(RK)'
PROOF. As described above, (2.1) and (1.2b) ensure that for any element
u E Gal(R / K) and any E E ££.-£.,(RK), there is a unique a E e£.,(RK) with
ElI = a * E. So for a fixed E, we get a well-defined map
F: Gal(R/K) --> e£.,(RK)
determined by the property ElI = F(u) * E for all u E Gal(R/K). It is

easy to check that F is a homomorphism, since
F(UT) * E = ElIT = (ET)lI = (F(T) * Et

= F(u) * (F(T) * E) = (F(u)F(T») * E.
(Note that Gal(KjK) acts on the left.)

It remains to show that the definition of F is independent of the choice
of a particular elliptic curve in c£..J:,,(RK). So let E l , E2 E Cf..,qRK) ,
let a E Gal(KjK), and write Ef = ill * El and E2 = il2 * E 2 . We need to
show that ill = il2' Since ef..,(RK) acts transitively on Cf..,f..,(RK), we can
find some b with E2 = b * E l . Then
So if we can prove that (b * EdO' is equal to b * Ef, then we can cancel b

from both sides to conclude that Ef = (il2illl) * Ef; and then (1.2(iii))
will give 01 = 02. Hence the following proposition completes the proof of
Proposition 2.4. (Note that bO' = b, since be K and a E Gal(K j K).)
Proposition 2.5. Let E jQ be an elliptic curve representing an element
of cf..,f..,(R K ), let 0 E ef..,(R K ), and let a E Gal(QjlQ). Then
Although the statement of Proposition 2.5 looks relatively innocuous,

it is giving a relationship between the algebraic action of a and the ana-
lytic action of multiplication by o. This suggests that the proof may not
be entirely straightforward. The main idea is to find an algebraic descrip-
tion of 0 * E. One of the tools we will need is the following lemma from
commutative algebra, whose proof we leave as an exercise.
Lemma 2.5.1. Let R be a Dedekind domain, let a be a fractional ideal
of R, and let M be a torsion-free R-module. Then the natural map
¢ : a- 1 M -----+ HomR(a, M)
x f---t (¢x: a ax) f---)
is an isomorphism.
PROOF (of Proposition 2.5). Choose a lattice A so that E 9! EA. Also fix
a resolution (i.e., an exact sequence)
R'K ~ RK -----+ a -----+ 0, (i)
where A is an m x n matrix with coefficients in RK. The idea underlying

the proof of Proposition 2.5 is that we should have
Cja- l A 9! il *E 9! Hom(a, E),
where we want to describe Hom( a, E) as an algebraic variety and not just

as an RK-module. (Here and in the following, Hom means homomorphisms
of RK-modules.) \Ve begin by applying Horn to the "product" of the exact

sequence (i) and the exact sequence of RK-modules
o -----> A -----> e -----> E -----> O. (ii)
This gives us the following commutative diagram:
0 0 0
1 1 1
0 -----> Hom(a, A) -----> Hom(a,q -----> Hom(a, E)
1 1 1 (iii)
0 -----> Hom(R7(,A) -----> Hom(R7(,q -----> Hom(R7(, E)
lA lA lA
0 -----> Hom(R K , A) -----> Hom(RK,q -----> Hom(RK' E)
For any RK-module M, we have Hom(RK,AJ) ~ AI", and applying

Lemma 2.5.1, first with M = A and then with AI = C, we get
and Hom(a, q = a- 1 e = C,
Using these isomorphisms, we can rewrite the diagram (iii) as
0 0 0
1 1 1
0 -----> a- 1 A -----> e -----> Hom(a,E)
1 1
en En
1 (iv)
0 -----> A" -----+ -----+ -----+ 0
l'A ltA ltA

0 -----+ Am -----+ em -----+ Em -----+ 0
Here tA is the transpose of the matrix A, and the bottom two rows are
clearly exact on the right, since they are just a number of copies of the
exact sequence (ii).
Applying the snake lemma to the bottom two rows of (iv) gives the
exact sequence
o -----> a-1 A -----> e -----+ (ker ~on ~ Em) -----+ An jtAA m . (v)
Notice that En ~ Ern is an algebraic map of algebraic varieties, since tA is

an m x n matrix whose coefficients are elements of End(E) = RK. Hence
the inverse image of the point (0,0, ... ,0) E Em is an algebraic subvariety
of En. Of course, En and Em are group varieties, so what we are saying is
that the kernel of En ~ Ern is an algebraic group variety. Further, (2.2a)
says that for any CT E Aut(IC), the corresponding map from E an ----> Earn
is obtained by applying CT to the entries of tA, treating those entries as
elements of RK C IC.
On the other hand, looking at the complex topology for one more mo-
ment, we note that An;tAAm is discrete and lC/a- 1 A is connected. Hence
the exact sequence (v) gives
(a * E)(iC) = lC/a- 1 A ~ identity component of ker(En ~ Ern).

We have thus described a *E algebraically in terms of the algebraic
map En ~ Em, and it now easy to finish the proof of Proposition 2.5.
For any CT E Gal(rQ/Q), we apply our characterization first to E and then
r
to E(T to deduce that
(a * E)(T = (identity component of ker(E n ~ E'n)
= identity component of ker ( (Ea) 11 t~ (Ea) Tn)

= aCT * Ea.
This completes the proof of Proposition 2.5, and with it the proof of Propo-
sition 2.4.
o
§3. Class Field Theory - A Brief Review
Class field theory describes the abelian extensions of a number field K

in terms of the arithmetic of K. The theory of complex multiplication
provides an analytic realization of class field theory for quadratic imaginary
fields, much as cyclotomic theory gives a realization of class field theory
for Q. In this section we will briefly review, without proof, the basic facts
from class field theory which will be used in the sequel. We will begin with
the classical version using ideals and ideal class groups. Afterwards we will
present the more modern idelic version. For proofs of the theorems stated
in this section and for additional material on (global) class field theory, the
reader might consult Lang [5], Tate [7], or Neukirch [lJ. We will mostly
restrict attention to totally imaginary fields, that is, fields with no real
embeddings, since (except for §7) that is the only case we will use in the
sequel.
Let K be a totally imaginary number field and let L be a finite abelian
extension of Kj that is, L/ K is Galois with abelian Galois group. As usual,
we write RK and RL for the rings of integers of K and L respectively. Let p
be a prime of K which does not ramify in L, and let q:J be a prime of L
lying over p. Thus the picture is
L q:J RL/q:J
I finite abelian
extension
I unra~ified
prIme
I extension of
finite fields
K p RK/P
By restriction, we get a homomorphism from the decomposition group of q:J
to the Galois group of the residue fields,
{a E Gal(L/ K) : q:J" = q:J} --- (~:7~ ~~~~~~/p) .

The right-hand Galois group is cyclic, generated by the Frobenius auto-
morphism
Further, since p is unramified, there is a unique element ap E Gal(L/ K)

which maps to Frobenius. Our notation reflects the fact that ap is deter-
mined by the prime ideal p in K. For a general Galois extension L/K, p
will only determine the conjugacy class of a p , and making a new choice
for q:J will change ap by conjugation. But in our situation a p will not
change, since we have assumed that L/K is abelian. Thus ap E Gal(L/K)
is uniquely determined by the condition
for all x E R L .
Let c be an integral ideal of K that is divisible by all primes that
ramify in L / K, and let
J(c) = group of fractional ideals of K which are relatively prime to c.
Then the Arlin map is defined using the ap's and linearity:
(. ,L/K): J(c) --- Gal(L/K),
(a,L/K) = (I]pnp,L/K) ~f I]a;p.

Notice that the Artin map is defined by piecing together local information,
one prime at a time. The following theorem, which is a weak version of
Artin's reciprocity law, provides important global information.
Proposition 3.1. (Artin Reciprocity) Let L/K be a finite abelian ex-

tension of number fields. There exists an integral ideal c c RK, divisible
by precisely the primes of K that ramify in L, such that
((a),L/K) =1 for all a E K* satisfying a == 1 (mod c).
If (3.1) is true for the ideals C1 and C2, then it also true for C1 + C2.
There is thus a largest ideal for which (3.1) is true. We call this ideal the
conductor of L/K and denote it by CL/K.
In view of (3.1), it is natural to define the group of principal ideals
congruent to 1 modulo c:
Pee) = {(a) : 0 E K*, a == 1 (mod en.

Artin reciprocity says that the kernel of the Artin map contains pee) for
an appropriate choice of Co More precisely,
a E P(CL/K) ===} (a,L/K) = 1.
It is important to observe that a principal ideal (0) may be in P( e) even

if 0 ¢. 1 (mod e); all that is necessary is that there exist a unit ~ E R'K such
that ~o == 1 (mod e).
Let p be a prime of K which is unramified in L. Then p splits com-
pletely in L if and only if the extension of residue fields has degree 1, or
equivalently if and only if (p, L / K) = 1. Thus the unramified prime ideals
in the kernel of the Artin map are precisely the primes of K that split
completely in L.
Definition. Let e be an integral ideal of K. A my class field of K (mod-
ulo e) is a finite abelian extension Ke/ K with the property that for any
finite abelian extension L / K,
Intuitively, one can think of the ray class field as the "largest" field
with a given conductor. However, it is important to note that the conductor
of Ke need not actually equal e. For example, the ray class field of Q(i)
modulo the ideal (2) is just Q(i) itself, so Q(i)(2) has conductor (1).
Theorem 3.2. (Class Field Theory) Let L/ K be a finite abelian exten-
sion of number fields, and let e be an integral ideal of K.
(a) The Artin map
(·,L/K): I(eL/K) -> Gal(L/K)
is a surjective homomorphism.
(b) The kernel of the Artin map is (Nj{h)P(cL/ K), where h is the group
of non-zero fractional ideals of L.
(c) There exists a unique ray class field Ke of K (modulo c). The conductor
of Ke/ K divides c.
(d) The ray class field Ke is characterized by the property that it is an
abelian extension of K and satisfies
primes of K that } { .. . }
{ . Ke
splI·t camp1e t e1y In = prIme Ideals In P(c) .
Example 3.3. Consider the ray class field of K modulo the unit ideal c =
(1). It is the maximal abelian extension of K which is unramified at all
primes. We call K(1) the Hilbert class field of K and denote it by H or H K .
Notice that
I(CH/K) = 1((1)) = {all non-zero fractional ideals of K},

P(CH/K) = P((I)) = {all non-zero principal ideals of K},
so the Artin map induces an isomorphism between the ideal class group
of K and the Galois group of the Hilbert class field of K:
(. ,HjK) : e,c(RK) ~ Gal(H/K).
We will also need the following version of Dirichlet's theorem on primes

in arithmetic progressions.
Theorem 3.4. Let K be a number field and c an integral ideal of K.
Then every ideal class in I (c) j P( c) contains infinitely many degree 1 primes
of K.
The Idelic Formulation of Class Field Theory
We will now briefly recall how class field theory is formulated using ideles.
This material will not be used until §7, so the reader may wish to omit the
rest of this section until arriving at that point.
Let K be a number field, and for each absolute value v on K, let Kv
be the completion of K at v. Further, let Rv be the ring of integers of Kv
if v is non-archimedean, and let Rv = Kv otherwise. The idele group of K
is the group
where prime indicates that the product is restricted relative to the Rv's.
This means that an element SEll K; in the unrestricted product is in A'K
if and only if Xv E R~ for all but finitely many v. In particular, we can

embed K* into Ai< by using the natural diagonal embedding
Q 1----+ ( " ' , Q, Q, Q," .),
since any Q E K* is in R~ for all but finitely many K. Similarly, for any
given v we embed K~ as a subgroup of Ai< via
t 1----+ ( ••• ,1,1, t, 1, 1, ... ).

i
v-component
If v is a non-archimedean absolute value corresponding to a prime

ideal 13, we will often write Kp and Rp in place of Kv and Rv' We will also
write ord p for the corresponding normalized valuation.
Let s E Ai< be an idele. We define the ideal of s to be the fractional
ideal of K given by
(s) = II
pord p sp,
p
where the product is over all prime ideal of K. Note that (s) is well defined,
since sp is a p-adic unit for all but finitely many p.
One makes Ai< into a topological group in the usual way; we will not
need the precise definition of the topology. For any integral ideal c of K,
let U, be the subgroup of Ai< defined by
U, = {s E Ai< : sp E R~ and sp == 1 (mod cRp) for all primes p}.

Then U, is an open subgroup of Ai<, and one proves that K*U, is a subgroup
of finite index in Ai<.
If Lj K is a finite extension, then there is a natural norm map from Ai'.
to Ai<. This is a continuous homomorphism
Nf< : Ai'. ----> Ai<
defined by the prescription that the v-component of Nf<x is
IIN~:Xw.
wlv
The idelic formulation of class field theory is given in terms of the

reciprocity map described in the following theorem.
Theorem 3.5. Let K be a number field, and let Kab be the maximal
abelian extension of K. There exists a unique continuous homomorphism
S 1----+ [s, K],

with the following property:

Let L/ K be a finite abelian extension, and let s E Ai.
be an idele whose ideal (s) is not divisible by any primes
that ramify in L. Then
[s,KliL = ((s),L/K).
Here (., L/ K) is the Artin map, and Gal(Kab / K) is given the usual pro-
finite topology. The homomorphism [., K] is called the reciprocity map
forK.
The reciprocity map has the following additional properties:
(a) The reciprocity map is surjective, and K* is contained in its kernel.
(b) The reciprocity map is compatible with the norm map,
for all x E A L.
(c) Let p be a prime ideal of K, let I;b C Gal(Kab/K) be the inertia

group of p for the extension Kab / K, let 7rp E K; be a uniformizer at p,
and let L/ K be any abelian extension that is unramified at p. Then
[7r p ,K]IL = (p,L/K) = Frobenius for L/K at p,

and
There is, of course, much more to class field theory that we have not
mentioned. For example, one often wants to know the exact kernel of
the reciprocity map and the correspondence between subgroups of Ai. and
subfields of Kab. However, the only additional fact that we will need in
this chapter is the following idelic characterization of ray class fields.
Theorem 3.6. Let K be a number field, let c be an integral ideal of K,
let Kc be the ray class field of K modulo c, and let Uc be the subgroup
of Ai. described above. Then the reciprocity map induces an isomorphism
[. ,K] : Ai./K*Uc ~ Gal(Kc/K).
In other words, [s, K] acts trivially on the ray class field Kc if and only if s
can be written as s = au with a E K* and u E Uc.
§4. The Hilbert Class Field
Our goal in this section is to prove the following theorem.

Theorem 4.1. Let KIQ be a quadratic imaginary field with ring of
integers RK, and let Elf:- be an elliptic curve with End(E) ~ RK. Then
K (j (E») is the Hilbert class field H of K.
Remark 4.1.1. Note that it is easy to produce an elliptic curve with
endomorphism ring equal to RK. For example, we could take E to be the
curve corresponding to the lattice RK. Then
is given in terms of series g2(RK) and g3(RK ) involving the elements of RK.
Alternatively, if we write RK = ZT + Z, then
1 00 .
j(E) = j(RK ) = -e2 - ' + '"'

?ITT L c(n)e
21UnT ,
n=O
where the c( n) E Z are the coefficients in the q-series expansion of j (1. 7.4b).
So Theorem 4.1 says that the Hilbert class field of a quadratic imaginary
field K is generated by the value of a certain holomorphic function jeT)
evaluated at a generator for the ring of integers of K.
We will actually prove much more than the mere statement of The-
orem 4.1. We will give an explicit description of how the Galois group
of HI K acts on j (E). To do this, we recall the homomorphism
from §2 characterized by the condition
ElY = F(a) * E for all a E Gal( K I K) and all E E ££.,£,( RK ).
Note that the kernel of F is actually a finite quotient of Gal(K I K), since
any E will be defined over some finite extension LIK, and then F(a) = 1
for a E Gal(KIL). Since e£,(RK) is an abelian group, F factors through
where K ab is the maximal abelian extension of K. Recall also the Frobenius

element a p E Gal(Kab I K) corresponding to a prime pin K. The following
proposition, together with basic class field theory, will serve to completely
determine F.
Proposition 4.2. There is a finite set of rational primes S C Z such that

ifp 1. S is a prime which splits in K, say as pRK = pp', then
Proposition 4.2 does not look very strong, since it determines F on

fewer than half of all Frobenius elements. But we will be able to use it
to get complete information about F. Before proceeding with the proof of
Proposition 4.2, we will derive some of its consequences, including a proof
of Theorem 4.1.
Theorem 4.3. Let E be an elliptic curve representing an isomorphism
class in ELL(RK).
(a) K (j (E)) is the Hilbert class field H of K.
(b) [Q(j(E)) : Q] = [K(j(E)) : K] = hK,
where hK = # eJ:.,(R K ) = # Gal(H/ K) is the class number of K.
(c) Let E 1 , •.• ,Eh be a complete set of representatives for ELL(RK).
Then j(Ed, ... ,j(Eh) is a complete set ofGal(K / K) conjugates for j(E).
(d) For every prime ideal p of K,
j(E)l7 p = j(jJ * E).
More generally, for every non-zero fractional ideal a of K,
j(E)(a,H/K) = j(o. * E).
Remark 4.3.1. It is now clear why we took the inverse when we defined
the action of an ideal class 0. on an elliptic curve EA. If we had used the
more natural definition 0. * EA = EllA, then the action of the Artin symbol
on j(E) in (4.3d) would instead have been j(E)(Il,H/K) = j(o.-l * E). Thus
we put the inverse into the action of eJ:.,(R K ) on ELJ:.,(RK) so that the
Artin symbol would act without an inverse.
PROOF (of Theorem 4.3). Let L/ K be the finite extension corresponding to
the homomorphism F: Gal(K/K) -+ eJ:.,(RK), by which we mean that L
is the fixed field of the kernel of F. Then
Gal(Kj L) = ker F
= {a E Gal(KjK) : F(a) = I}
= {a E Gal(K / K) : F(a) * E = E} since by (1.2), eL(RK )
acts simply transitively on ELJ:.,(RK)

= {a E Gal(K / K) : EI7 = E} from the definition of F
= {a E Gal(K / K) : j(E 17 ) = j(E)}
= {a E Gal(K / K) : j(E)17 = j(E)}
= Gal(K j K(j(E))).
Hence L = K(j(E»). Further, since F maps Gal(LjK) injectively into

eL(R K ), we see that Lj K is an abelian extension. So we have shown
that L = K (j (E») is an abelian extension of K.
Let CLjK be the conductor of LjK, and consider the composition of
the Artin map with F,
We claim that this composition is just the natural projection of I(cLjK)

onto eL(RK)' In other words, we wish to establish the
Claim: F(o, LjK») = Ii for all 0 E I(cLjK)'
Let 0 E I( CLj K), and let S be the finite set of primes described in (4.2).
From Dirichlet's theorem (3.4) there exists a degree 1 prime p E I(cLjK)
in the same P(cLjK)-ideal class as 0 and not lying over a prime in S. In
other words, there is an a E K* satisfying
a == 1 (mod CLjK) and 0= (a)p.
We compute
F(o, LjK») = F((a)p,LjK») since 0 = (a)p

= F(p, LjK») since a == 1 (mod CLjK)
=p from (4.2), since Nlfp tJ. S
=0 since 0 = (a)p.
This completes the proof of the claim.

Notice that as an immediate consequence we find that
F ( (a), L j K») = 1 for all principal ideals (a) E I(cLjK),
and not just for those that are congruent to 1 modulo CLjK' We also know
that the map F : Gal(L/ K) ---+ eqR K ) is injective, so this implies that
(a),LjK) = 1 for all (a) E I ( CL j K ).
But the conductor of Lj K is the smallest integral ideal c with the property
that
a == 1 (mod c) ~ (a),L/K) = l.
(See §3.) It follows that CL/K = (1). The conductor is divisible by every
prime that ramifies (3.1), from which we conclude that the extension Lj K
is everywhere unramified. Therefore L is contained in the Hilbert class
field H of K.
On the other hand, the natural map 1(cL/K) = 1((1)) --+ eqRK) is
clearly surjective, so the claim implies that F : Gal(LI K) --+ eqRK ) is
surjective, hence an isomorphism. Therefore
[L: K] = # Gal(LIK) = #eqRK ) = # Gal(HIK) = [H: K].
This combined with the inclusion L c H proves that L = H. Since L =

K(j(E)), this completes the proof of (a), as well as the second equality
in (b).
To prove the first equality in (b), we use the observation (2.2.1) that
This inequality combined with [K(j(E)) : K] = hK and [K : Q] = 2

implies that [Q(j(E)) : Q] = hK, which completes the proof of (b).
Next, from (1.2b) we know that eqRK) acts transitively on the set
of j-invariants
a= {j(Ed,··· ,j(Eh )},
since by [AEC III.1.4b] the set e.c.c(RK) may be identified with the j-
invariants of its elements. The map F : Gal(K I K) --+ e.c(RK) is defined
a
by identifying the action of Gal( K I K) on with the action of eq RK )
on a, so Gal( K I K) also acts ttansitively on a. Therefore a is a complete
set of Gal( K I K) conjugates of j (E), which proves (c).
Finally, we see that the claim proven above gives (d) for all ideals
in 1(cL/K). But CL/K = (1), so 1(cL/K) is the set of all non-zero fractional
ideals of K. 0
It remains to prove Proposition 4.2. For that purpose we will need the
following result which says that isogenies behave nicely under reduction.
Proposition 4.4. Let L be a number field, ~ a maximal ideal of L, Ell L
and E21 L e11iptic curves with good reduction at ~, and El and E2 their
reductions modulo~. Then the natural reduction map
¢~¢,
is injective. Further, it preserves degrees,
deg(¢) = deg(¢).
PROOF. Since the degree of a non-zero isogeny is non-zero, the injectivity

follows from the preservation of the degree. However, the proof of injectiv-
ity is more elementary, so we will give a separate proof.
Let ¢ : E1 --+ E2 be an isogeny satisfying ¢ = [0]. For any integer m

prime to 13, [AEC VII.3.lb] says that E2[m] injects into E2. On the other
hand, if T E Edm], then by assumption
¢(T) = ¢(T) = O.
Since ¢(T) E E2[m], it follows that ¢(T) = O. Therefore Edm] C ker(¢).
This holds for arbitrarily large m, so we must have ¢ = [0].
Now we begin the proof that deg(¢) = deg(¢). Choose a rational
prime e relatively prime to 13. Our idea is to use the Weil pairing and
calculate everything on the Tate modules. (See [AEC 111.8.3] for the prop-
erties of the Weil pairing eE : T£(E) x 1£(E) --+ Te(/L) that we will need.)
For any x, y E Tp(Ed we have
eE, (x, y)de g ¢ = eEl ((deg cp)x, y) = eE, (¢cpx, y) = eE 2 (¢x, cpy), (i)
and a similar calculation on E1 gives
(ii)
Next we observe that if E / L is any elliptic curve with good reduc-

tion at 13, then T£ (E) ~ Tp (E). This crucial equality is a consequence
of [AEC VII.3.lb]' which says that E[e n ] ~ E[e n ] for all n. Looking at the
definition of the Weil pairing [AEC III §8]' we see that
for all x, y E Te(E). (iii)
We now take x, y E T£(Ed and compute
~ dcg¢
- -)deg ¢ = eE, (
eEL (x,y )
x,y from (iii)
= eE 2 (cpx, cpy) from (i)

= e E2 (cpx,cpy) from (iii)
= e E2 (¢x, ¢y)
- -)deg¢
= eE, ( x,y from (ii).
This equality holds for all x, y E Te(EJ), hence for all X, y E T£(EJ). The
non-degeneracy of the Weil pairing on Te(EJ) now implies that deg ¢ =
degcp. 0
PROOF (of Proposition 4.2). We know that ££-.qRK) is finite from (1.2b)
and that every curve in ££-£-(RK) can be defined over Q from (2.lc), so
we can choose a finite extension field L / K and representatives E 1 , ... , En
defined over L for the distinct k isomorphism classes in c£.c(R). Further,

using (2.2c) we may replace L by a finite extension so that every isogeny
connecting every pair of Ei'S is defined over L. We now let S be the finite
set of rational primes satisfying anyone of the following three conditions:
(i) p ramifies in L,
(ii) some Ei has bad reduction at some prime of L lying over p,
(iii) p divides either the numerator or the denominator of one of the
numbers N~(j(Ei) - j(E k )) for some i -=f- k.
Notice that condition (iii) means that if p rt- S and if q3 is a prime of L
dividing p, then Ei ~ Ek (mod q3), since their j-invariants are not the same
modulo q3.
Now let p rt- S be a prime which splits as pRK = pp' in K, and let q3
be a prime of L lying over p. Also let A be a lattice for E, so E(C) ~ CIA.
Choose some integral ideal a C RK relatively prime to p such that ap is
principal, say
ap = (a).
From [AEC VI.4.1b] there are isogenies connecting E, p * E, and a * p * E
corresponding to the natural analytic maps as indicated in the following
diagram:
CIA ----> C/p-1A ----> Cla-1p-l A C/(a-1)A ----> CIA

Z'-'Z ZI->Z z~az
11 11 11 11
¢ 1/1 A
E ----> P* E ----> a*p*E (a)*E ----> E
Next we choose a Weierstrass equation for ElL which is minimal

at q3 (see [AEC VII §§1,2]) and let
dx
w=-----
2y + alx + a2
be the associated invariant differential on E. The pull-back of w to CI A will
be some multiple of dz. Since the map along the top row of our diagram is
simply z --- az, we see that dz pulls back to d(az) = adz. Tracing around
the commutative diagram, we conclude that
(>. 0 '¢ 0 ¢)*w = aw.

As usual, we will use a tilde to denote reduction modulo q3. Since the
equation for ElLis minimal at q3, we obtain an equation for E by reducing
the coefficients modulo q3, and so the reduced differential
dx
w--=:-----
- 2y + ihx + ih
is a non-zero invariant differential on E. Further, since (a) ap and

since IfJ divides p, we find
(,\ 0 -If; 0 J)*w = (>. 0 1jJ 0 ¢)*w = aw = O.

It follows from [AEC II.4.2c] that
,\ 0 -If; 0 J is inseparable.
On the other hand, using (4.4) and (1.5a), we see that
-
deg ¢ = deg ¢ = NQIK P = p,
- K
deg 1jJ = deg 1jJ = NQI a,
deg ,\ = deg >. = 1.
Since N{I a is prime to p by assumption, both -If; and ,\ are separable, so we
conclude that
¢:E----p*E
must be inseparable. Now any map (such as J) factors as a qth_power
Frobenius map followed by a separable map [AEC II.2.12], so the fact
that J has degree p and is inseparable implies that J must "be" the pth_
po~Frobenius map. More precisely, there is an isomorphism from E(p)
to p * E so that the composition
pth_power ~
E -----+) E(p) ~ P* E
Frobenius
equals J.
In particular, we find that
so
But from the original choice of excluded primes S, we have
if and only if
Hence p * E ~ F(al» * E, and the simplicity of the action of e£(R K )

on e££.,(RK) (1.2b) gives the desired conclusion
D
We also record for later use the following fact which we proved during
the course of proving Proposition 4.2.
Lemma 4.5. Let E be an elliptic curve with complex multiplication

by R K , and suppose that E is defined over a number field L. Then for all
but finitely many degree 1 primes p of K, the natural map
has degree p, and its reduction
is purely inseparable. (Here we reduce modulo some prime $ of L lying

above p.)
§5. The Maximal Abelian Extension
Let E be an elliptic curve with complex multiplication by a quadratic imag-

inary field K. In this section we are going to describe the field generated
by the points in E(C)tors, much as we described the field generated by j(E)
in the last section. Our goal is to use the torsion points of E to generate
abelian extensions of K.
Before beginning, we briefly recall the analogous (but simpler) case of
cyclotomic extensions. In this case the elliptic curve E(C) is replaced by
the multiplicative group Gm(C) = C*. Let
be the group of N-torsion points of G'Tn as usual; that is, /LN is the group
of Nth-roots of unity. As is well known, the extension Q(/LN )/Q is an
abelian extension that is ramified only at primes dividing N. Let p be a
prime with p t N, choose a generator ( for /LN, and let (Tp E Gal(Q(()/Q)
be the Frobenius element associated to p. Also let $ be a prime of Q( ()
lying above p. Then by the definition of (Tp we have
But 1, (, (2, ... , (N-l are distinct modulo $, since our assumption $ t N
implies that X N - 1 is separable in characteristic p. Hence the congruence
is an equality,
and so we conclude that
(Tp =1 ¢==> p == 1 (mod N).

Therefore Q«() = Q(J.LN) is the ray class field of Q of conductor N. (Ac-

tually, it is the ray class field of conductor N 00. Since the base field Q is
not totally imaginary, we have to also consider ramification of the infinite
place.)
Now let L/Q be any abelian extension, and let N be the conductor
of L. Then class field theory (3.2) says that L is contained in the ray class
field of conductor N, so we recover the following famous result.
Theorem 5.1. (Kronecker-Weber Theorem) Every abelian extension of
Q is contained in a cyclotomic extension; that is, given any finite abelian
extension L/Q, there is a root of unity ( such that L C Q«().
Thus the ray class fields of Q are generated by the values of the analytic
function
evaluated at points of finite order in the circle group lR/Zj that is, they
are generated by numbers of the form e27ria/N with a, NEZ. Further, the
action of a Frobenius element up on the value e27ria/N is given explicitly by
the formula
provided p f N.
Thus the Galois action of up is transformed into a multiplication action on

the circle group. The reader who has a good understanding of this cyclo-
tomic theory will have no trouble seeing how all of its main elements are
reproduced in the theory of complex multiplication as described in this sec-
tion and in §8, albeit with a number of additional technical complications.
As usual, we let RK be the ring of integers in a quadratic imaginary
field K, and let E be an elliptic curve with complex multiplication by RK.
We will always assume that the isomorphism [. J : RK ..':::', End(E) is nor-
malized as in (1.1).
We begin with an important lemma which tells us when an endomor-
phism of the reduced curve E (mod!.p) actually comes from an endomor-
phism of E.
Lemma 5.2. Suppose that E is defined over the number field L, let !.p
be a prime of L at which E has good reduction, and let E be the reduction
of E modulo!.p. Let
(): End(E) - > End(E)
be the natural map which takes an endomorphism to its reduction mod-
ulo!.p. Then for any'Y E End(E),
'Y E Image«()) {=:::> 'Y commutes with every element in Image«()).

In other words, Image(B) is its own commutator inside End(E).

PROOF. One direction is trivial, since if'Y E Image(B), then, certainly
commutes with elements of Image(B). This is immediate from the fact
that End(E) ~ R K , which implies that Image(B) is a commutative ring.
For the other direction, we first note that B is injective from (4.4).
From [AEC 111.9.4]' End(E) is an order in either a quadratic imaginary
field or in a quaternion algebra. If it is an order in a quadratic imaginary
field, then B is an isomorphism, since, by assumption, End(E) ~ RK is the
maximal order in K. So in this case we are done.
Next we consider the case that End(E) is an order in a quaternion
algebra JC. Then Image(B) 129 Q is a quadratic subfield of ~, call it X.
(Note X ~ K, but it is possible for ~ to contain several distinct sub fields
each isomorphic to K.) We start by choosing a Q-basis {I, a} for X such
that 00 2 E Q; and then we extend it to a Q-basis for ~ of the form
~ = Q + Qa + Q(3 + Qa(3
satisfying
and 00(3 = -(300.
(See the proof [AEC 111.9.3].) Now it is easy to find the commutator of X
in~. For any, E ~, we write 'Y = d + aa + b(3 + ca(3 with a, b, c, dE Q
and compute:
'Y commutes with X
{==? ,a = a'Y
{==? (d+ aa + b(3 + ca(3)a = a(d + aa + b(3 + ca(3)
{==? da + aa 2 + b(3a + ca(3a = da + aa 2 + ba(3 + ca 2 (3
{==? -ba(3 - ca 2 (3 = ba(3 + ca 2 (3 since 00(3 = -(300
{==? b=c=O
since 00 2 E Q and {I, a, (3, a(3} is a Q-basis for ~
{==? 'Y = d + aa E Q + Qa = X.
Finally, let {j E End(E) commute with Image(B). Then {j commutes
with X, so from what we have just done, {j is in X. But we also know that {j
is integral over Z and that Image(O) ~ RK is the maximal order in X ~ K,
hence {j E Image(B). This completes the proof of Lemma 5.2. 0
As usual, let E be an elliptic curve with complex multiplication by RK.

In the last section (4.3a) we proved that
H = K(j(E))
is the Hilbert class field of K. Since j(E) E H, this means we can find an
equation for E with coefficients in H, so we may as well assume that E is
defined over H. The next proposition says that we can lift the pth_power
Frobenius map E --> E(p) to a map in characteristic o.
Proposition 5.3. Let K be a quadratic imaginary field, H the Hilbert

class field of K, and E/ H an elliptic curve with complex multiplication
by R K . Let ITp E G H / K be the Frobenius element associated to a prime p
of RK, and let \13 be a prime of H lying over p. Assume that p has degree 1
and is not in the finite set of primes specified in (4.5), so in particular E
has good reduction at \13. Then there exists an isogeny
whose reduction modulo \13,

5. : jj; ----+ jj;(p) ,
is the pth-power Frobenius map.

Remark 5.3.1. In general, there is no reason to expect an elliptic curve
to be isogenous to one of its Galois conjugates. Of course, there are always
maps
E E"p
1 pth power
1
E jj;(p)
Frobenius
where the vertical maps are "reduction modulo \13." The content of Propo-
sition 5.3 is that there is an isogeny .x : E ---> E"p which makes this picture
into a commutative square. Thus .x lifts the Frobenius map from charac-
teristic p to characteristic O.
PROOF (of Proposition 5.3). To ease notation, we will write IT in place

~. From (4.5) there is an isogeny E ---> p * E whose reduction jj; --->
p * E is purely inseparable of degree p. Composing this isogeny with the
isomorphism p * E ~ E" provided by (4.3), we get an isogeny 5. : jj; ---> E"
which is purely inseparable of degree p. It follows from [AEC 111.4.6] that 5.
factors as
jj; .-.'!..... jj;(p) --=--. E",
where ¢ is the pth_power Frobenius map and deg E = 1. But, by definition,
the reduction of E" is precisely jj;(p) , so E is an automorphism of E". If we
can show that E is the reduction modulo \13 of some EO E Aut(E"), then we
can replace .x by EOl o.x and be done. So we need to prove the
Claim: E lies in the image of Aut(E") inside Aut(E").
From (5.2), it suffices to show that E commutes with the image of End(E")
inside End(E"). (This will allow us to lift E to an Eo in End(E"), and
then (4.4) will imply that Eo has degree 1, so it is in Aut(E").) Recall that
we have normalized isomorphisms
[ ·]E : RK ....::-.... End(E) and

and that from (1.1.1) these isomorphisms satisfy
A 0 [alE = [alE" 0 A for all a E RK.
Next we look at the reduction of [al modulo \p. In general, suppose

that f : V ----> W is any rational map of algebraic varieties over a field k
of characteristic p, let CPv : V ----> V(p) and CPw : W ----> W(p) be the pth_
power Frobenius maps, and let a E Aut(k) be the pth-power Frobenius
automorphism of k. Then ffI : V(p) ----> W(p) is a rational map and
CPwof =r ocpv.
To see that this is true, write f = [fa, ... , fnl (locally) as a map given
by homogeneous polynomials. The desired result then follows from the
observation that for a polynomial f(x) = f(Xl, ... ,xm ) = L: aixi, we have
We now apply this general fact to the map [alE: E ----> E. Note that
[alE = [alE"
from (2.2a), since a E G H / K fixes a E K. Thus we get
cp 0 [alE = [alE 0 cp = [alE" 0 cp.
Using this, we compute
[alE" 0 c 0 cp = ~ 0 >- since c 0 cp = >-

= >- 0 [alE from above,
= cO cpo [alE
= cO [alE" 0 cp from above.
Therefore [alE" 0 c = cO [alE'" which completes the proof of our claim and
with it the proof of Proposition 5.3. 0
An important special case of Proposition 5.3 occurs when the ideal p

is principal, in which case ap = (p, HI K) = 1. Then A is an endomorphism
of E. We can identify that endomorphism quite precisely as follows.
Corollary 5.4. Let K be a quadratic imaginary field, H the Hilbert class

field of K, and E / H an elliptic curve with complex multiplication by RK.
For all but finitely many degree 1 prime ideals l' of K that satisfy
(p,H/K) = 1,
there is a unique 'iT = 'iTp E RK such that

[IT I
E E
l' ='iTRK, and

1 pth power
1
E E
Frobenius
is a commutative diagram. (Note that the condition (1', H / K) 1 1S
equivalent to l' being a principal idea1.)
PROOF. Let ',p be a prime of H lying over p. Having excluded finitely

many p's, including those for which E (mod ',p) is singular, we may use (5.3)
to obtain a commutative diagram
1 1
E
Here O"p = (p,H/K), A is an isogeny, ¢ is the pth_power Frobenius map,

and the vertical maps are reduction modulo ',p.
Our assumption that (1', H / K) = 1 means that EeYp = E, so A is really
an endomorphism of E, say A = ['iT]. It also implies that E(p) = E. Thus
we have a commutative diagram
[IT I
E --+ E
1 1
E ~ E.
Now we compute
K
NQJp = p since l' has degree 1
= deg¢ since ¢ is pth power Frobenius
= deg['iT] from (4.4), since ['iT] = ¢
= INS''iT I from (l.4b).
Since l' is a prime ideal in the quadratic field K, this means that either
or
where 7r' is the Gal(K/Ql)-conjugate of 1r. We can use the fact that (E, [.])
is normalized to check which one it is.
We take an equation for E / H with good reduction at \JJ and let W E OE
be a non-zero invariant differential whose reduction wis a non-zero invariant
differential on E. Then the normalization (1.1) says that [7r]*w = 7rW, so
irw = irW = [7r]*w = [7r]*w = ¢*w = o.

The last equality follows from [AEC II.4.2c], since the Frobenius map ¢ is
inseparable. Now OF; is a one-dimensional vector space generated by W,
so ir = o. In other words,
7r == 0 (mod \JJ),
so 7r E \JJ n K = p. Since we saw above that p equals either 7r RK or 7r' R K ,

we conclude that p = 7rRK. This finishes the existence half of (5.4).
To see that 7r is uniquely determined, we need merely observe that the
composition
[.J -
RK ------+ End(E) ------+ End(E)
is injective. Since 7r is required to satisfy [7r] = ¢ E End(E), there is at

most one such 7r. D
Our goal is to show that the torsion points of an elliptic curve E with
complex multiplication by RK can be used to generate abelian extensions
of K. It would be nice if the torsion points themselves should generate
abelian extensions of K, but unfortunately it turns out that they only
generate abelian extensions of the Hilbert class field H of K. In order to
pick out the correct subfield, we take a model for E defined over H and fix
a (finite) map
h : E ------+ E/ Aut(E) ~ pI
also defined over H. Such a map h is called a Weber function for E / H.
Example 5.5.1. If we take a Weierstrass equation for E of the form
with A,B E H,
then the following is a Weber function for E / H:
if AB i= 0,
h(P) = hex, y) = {~2 if B = 0,
x3 if A = O.
So in essence, except for the two exceptional cases j = 0 and j = 1728, a

Weber function is just an x-coordinate for the curve.
Example 5.5.2. It is also possible to define a Weber function analytically

in such a way that we don't have to worry about fields of definition. For
example, if we choose a lattice A and an isomorphism
f:C/A~E(q, z""'-' (p(z,A), p'(z,A)),
then the following is a Weber function for E:
g2(A)g3(A) ( A) if j(E) -10,1728,
~(A) P Z,
g2(A)2 )2
h(J(z)) = ~(A) p(z,A if j (E) = 1728,
g3(A) ( )3 if j(E) = o.
~(A) p z,A
Here ~(A) = g2(A)2_27g3(A)3 -lOis the usual modular discriminant. The

reader may easily verify that this Weber function is model independent;
that is, it does not change if we take a new lattice for E, or equivalently
a new Weierstrass equation for E. Since we know from (4.3a) that it is
possible to find an equation for E defined over H, it follows from the model
independence that this Weber function h : E _pI is defined over H.
To generate abelian extensions of K, we will use the values of a Weber
function on torsion points, which essentially means we will take the x-
coordinates of the torsion points. Recall from §1 that for any integral
ideal c of RK we defined the group of c-torsion points of E to be
E[cl={PEE: blP=Oforall"(Ec}.
The reader is advised to compare the following theorem with the cyclotomic
theory discussed at the beginning of this section.
Theorem 5.6. Let K be a quadratic imaginary field, let E be an elliptic
curve with complex multiplication by RK, and let h : E - pI be a Weber
function for E / H as described above. Let c be an integral ideal of R K .
Then the field
K(j(E), h(E[c]))
is the ray class field of K modulo c.
Corollary 5.7. With notation as in Theorem 5.6,
K ab = K(j(E),h(Etors)).
In particular, if j(E) -10,1728 and if we take an equation for E with coef-
ficients in K (j (E) ), then the maximal abelian extension of K is generated
by j(E) and the x-coordinates of the torsion points of E.
PROOF (of Theorem 5.6). Let
L = K(j(E),h(E[c])).
Then L ::J K(j(E)), and from (4.3a) we know that K(j(E)) = H is the
Hilbert class field of K. In order to show that L is the ray class field of K
modulo e, we need to prove that
(p,L/K)=l ¢=} pEP(e).
As usual, it suffices to prove this for all but finitely many degree 1 primes
in K.
Suppose first that p is a degree 1 prime of K with P E P( e). This
means that
for some p, E RK with p, == 1 (mod e).

In particular, p is principal, so (p,H/K) = 1. Hence we can apply (5.4)
(after excluding finitely many p's) to get some 7r E RK such that
EnE
p = 7rRK and
1 1 commutes.
E ~ E
Since 7r RK = P = P,RK, there is a unit ~ E R'K such that 7r = ~p,. Notice
that [~] E Aut(E), so [7r] and [p,] differ by an automorphism of E.
We already know that (p,L/K) fixes H = K(j(E)), so in order to
show that it fixes all of L, we must show that it fixes h(E[eJ). Let T E E[e]
be any e-torsion point. Then the commutative diagram gives
T(p,L/ K) = 4>(1') = [7r]T.
On the other hand, [AEC VII.3.1b] tells us that the reduction map E -+ E
is injective on torsion points whose order is prime to p. So if we exclude from
consideration the finitely many p's which divide #E[e], then the reduction
map
E[e] ------> E[e]
is injective. Therefore
T(p,L/ K) = [7r]T.
Now we compute
h(T)(p,L/K) = h (T(P,L/K)) since (p,H/K) = 1 and

h:E -+ pI is defined over H
= h([7rJT) from above
= h([~] 0 [p,]T) since 7r = ~p,
= h([p,]T) since h is Aut(E)-invariant and
[~] E Aut(E)
= h(T) since T E E[e] and p, == 1 (mod e).
This completes the proof that
P E pee) => (p,LjK) = 1.
In order to prove the converse, we take a prime of degree 1 satisfy-

ing (p, Lj K) = 1. Then
(p,HjK) = (p,LjK)1 = 1,
H
so (excluding finitely many p's) we can apply (5.5) as usual to get a 1r E RK

such that
E kL E
and
1 1 commutes.
E ~ E
We also choose some a E G K I K whose restriction to K ab is (p, Kab j K).
Then in particular aiL = (p, Lj K) = 1, and also alH = 1 since He L.
Now let T E E[el be any e-torsion point. We compute
h([1rlT) = h([1rlT)
= h(¢(T)) from the commutative diagram
= h (Ta) since a reduces to pth power Frobenius
= h(Ta)
= h(T)a since a IH = 1 and h is defined over H
aiL
~
= h(T) since h(T) ELand = 1

= h(T).
Next we observe that the reduction of h modulo IiJ is the map
h: E -+ Ej AutE ~ EjA.;;t"E.
(N.B. The image is not Ej Aut E, since Aut E may be larger than Aut(E).)
It follows from this and the equality h([1rlT) = h(T) proven above that
there is an automorphism [~l E Aut(E) such that
Again using the injectivity of the torsion E[e] '"--> E[e] from [AEC VII.3.1b]'
we find that [1r - ~]T = o.
A priori, the particular ~ for which [1r - ~]T = 0 might depend on T.
But from (l.4b) we know that E[e] is a free RKje-module of rank one.
Hence there is a single ~ E R'K such that [7f - ~] annihilates all of E[c],
which implies that 7f == ~ (mod c). Therefore
C l 7f == 1 (mod c),
and of course we have p = 7fRK = (~-l7f)RK since ~ is a unit. This proves
that p E P(c), which completes the proof of Theorem 5.6. 0
PROOF (of Corollary 5.7). Let L / K be any finite abelian extension and
let CL/K be the conductor of L/ K. By class field theory (3.2c), L is con-
tained in the ray class field of K modulo CL/K' Using (5.6), this means
that
L C K(j(E), h(E[CL/K]))'
Taking the compositum over all conductors gives L c K(j(E),h(Etors )),
and then taking the union over all L's gives Kab C K(j(E),h(Etors))'
But (5.6) says that K(j(E), h(Etors )) is a compositum of abelian exten-
sions, hence it is abelian, hence it equals Kab. This completes the first part
of (5.7).
The second part of (5.7) is then immediate from (5.5.1), which says
that if j(E) f= 0,1728, then the x-coordinate on a Weierstrass equation
for E/Q(j(E)) is a Weber function for E. 0
Example 5.8. Corollary 5.7 raises the obvious question of what happens if
we adjoin all of E tors to K, rather than just the values of a Weber function.
In general one does not get an abelian extension of K, although we have
seen (2.3) that E tors generates an abelian extension of H. (The reader
might try to use (5.4) to construct another proof of this fact.) Suppose
now we look at the special case that K has class number 1, so H = K.
Then we have inclusions
Thus
K has class number 1 ==} K ab = K(h(Etors )) = K(Etors).

The j-invariants of these curves will be in Q. For a complete list of all CM j-
invariants in Q, together with representative Weierstrass equations, see
Appendix A §3.
Example 5.8.1. We will illustrate (5.6) and (5.8) with the curve
E: y2 = x3 + x
which has complex multiplication by the ring of Gaussian integers Z[i] in
the field K = Q(i). Clearly
E[2] = {O, (0,0), (±i,O)},

so K (E[2]) = K. One can easily check that the ray class field of K mod-
ulo 2 is K, so this confirms (5.6).
Next we look at points of order 3. Letting T = (x, y) E E, the dupli-
1)
cation formula reads
_ (X4 - 2X2 + 1 x 6+ 5x 4- 5x 2 -
2T - 4y2' 8y3 .
So setting x(2T) = x(T), we find (after some algebra) that

3T = ° {::=} 3x 4 + 6x 2 - 1 = O.
The four roots of this equation are
1 1 J2v'3 - 3
a, -a, v'3a' - v'3a' where a= 3'
Since the Weber function on E is h(x,y) = x 2, this gives K(h(E[3]))

K ( v'3), which the reader may verify is indeed the ray class field of K
modulo 3.
Substituting these four values for x into y2 = x 3 + x and solving for y,
we find the y-coordinates of the points in E[3]. If we let
(3 = «8v'3 - 12 = ~
9 yJ3'
then the nine points in E[3] are
E[3] = {o, (a,±(3), (-a, ±i(3) , (~a' ~(3)' (~~, ;;(3)}'

Since K = Q(i) has class number 1, (5.8) says that the field K((3) is an
abelian extension of K, but it is not necessarily a ray class field. We leave
it for the reader to check directly that K K is abelian.((3) /
Next, T = (x, y) is a point of exact order 4 if and only if y(2P) = O.
Using the duplication formula given above, if we let "( = (J2 - 1) i, then
the x-coordinates of the points of order 4 satisfy
0= x 6 + 5x 4 - 5x 2 - 1 = (x - l)(x + l)(x - "()(x + "()(x - "(-I)(X + "(-1).
Hence K(h(E[4])) = K("(2) = K(J2), which is the ray class field of K
modulo 4. Finally, if we let 8 = (1 + i) (J2 - 1), then we find that
E[4] = {O, (0,0), (±i,O), (1, ±J2), (,,(, ±8),
(-,,(, ±i8), (,,(-l, ±,,(-28), (-,,(-I, ±h-28)}.
So in this case K(E[4]) = K(,,(,8) = K (J2) is equal to K(h(E[4])).

§6. Integrality of j
We have seen (2.1b) that the j-invariant of an elliptic curve E with complex
multiplication is an algebraic number. In this section we are going to
prove that j(E) is in fact an algebraic integer or, equivalently, that E has
everywhere potential good reduction. The results of this section will not
be needed until §1O, so the reader primarily interested in the relationship
between complex multiplication and class field theory may wish to skip
directly to §7.
Theorem 6.1. Let EIC be an elliptic curve with complex multiplication.
Then j (E) is an algebraic integer.
We are going to give three proofs of this important fact, two in this section
and a third in (V.6.3). In order to help the reader understand the different
approaches used in these three proofs, we will start with a brief description
of each.
The Complex Analytic Proof
Let Al and A2 be lattices corresponding to elliptic curves EriC and E2/C,
and suppose that El and E2 are isogenous. Then we will show that j(E 1 )
and j(E2 ) are algebraically dependent over IQl by explicitly constructing
a polynomial F(X, Y) E Z[X, Y] with F(j(E 1 ),j(E2)) = O. If E has
complex multiplication, then by taking El = E2 = E we will obtain a
monic polynomial with j(E) as a root. Thus we show that j(E) is integral
over Z by explicitly constructing a monic polynomial with j(E) as a root.
This proof has the advantage of being very explicit, and the disadvantage
that it does not generalize to higher dimensions.
The f-adic (Good Reduction) Proof
This proof, which is due to Serre and Tate [1], readily generalizes to abelian
varieties of arbitrary dimension. The idea is to use the criterion of Neron-
Ogg-Shafarevich [AEC VII.7.3] to prove directly that E has potential good
reduction at all primes, which implies by [AEC VII.5.5] that j(E) is in-
tegral at all primes. Thus let L be a local field and ElL an elliptic
curve with complex multiplication. We have seen (2.3) that the action
of Gal(LI L) on the Tate module Ti(E) is abelian. (For another proof ofthis
fact that uses nothing more than a little linear algebra, see exercise 2.6.)
In other words, Gal(Lab I L) acts on Ti(E). Next we use the description
of Gal( Lab I L) provided by local class field theory to show that the action
must factor through a finite quotient of Gal(Lab I L), which allows us to
apply [AEC VII.7.3J.
The p-adic (Bad Reduction) Proof
For this proof, which is due to Serre, we assume that j(E) is not integral at
some prime p and prove that E has no non-trivial endomorphisms. Let L
be a complete local field with maximal ideal p, and let E/ L be an elliptic

curve whose j-invariant is non-integral at p. We will show (V §§3,5) that
after replacing L by a quadratic extension, there is an element q E L * and
a p-adic analytic isomorphism of groups
L* /qZ ~ E(L).
Using this isomorphism, we construct (V.6.1) an element of Gal(L/ L) which
acts on the Tate module of E via the matrix (6 I) (relative to a suitable
basis). Then the fact that endomorphisms commute with the action of Ga-
lois will allow us to conclude that there are no non-trivial endomorphisms,
so E does not have complex multiplication.
It is worth remarking that the second and third proofs are local; one
shows that j(E) is integral by working one prime at a time. The first proof,
on the other hand, is more global in nature. We are going to give the first
two proofs in this section. For proof number three, see (V.6.3).
Example 6.2.1. Note that the three elliptic curves in (2.3.1) possessing
an endomorphism of degree 2 all have j-invariants in Z, as they should
from (6.1), since the corresponding quadratic imaginary fields have class
number 1. More generally, if K has class number 1, then j(RK) will be a
rational integer.
As is well known, there are only nine quadratic imaginary fields of
class number 1, a fact conjectured by Gauss and proven by Heegner [1].
(See also Baker [1] and Stark [1].) These fields are
Q(V-I) , Q(v=2) , Q(vC3) , Q(P), Q(v'-ll),

Q (v'-19), Q (v'-43), Q(v'-67), Q (v'-163) .
A list of the corresponding j-invariants is given in Appendix A §3. It follows

for example that
.(1 + vCI63)
J 2 '11
E~.
Recall (I.7.4b) that j(7) has the q-expansion
j(q) = ~ + 744 + 196884q + 21493760q2 + ... ,

q
where q = e 2rriT . If we substitute 7 = (1 + v'-163) /2, then
q = _e- rr v'163 "" -3.809. 10- 18
is very small. Thus the main term in j(q) will be l/q, which means that l/q
should be "almost" an integer. Computing l/q to 40 significant digits we
find that
e rr v'163 =262537412640768743.999999999999250072597 ... ,

so e7rv'T63 is an integer to 12 decimal places. Of course, we know a priori

that e7rv'T63 is not an integer and in fact is not even an algebraic number,
since the Gel'fond-Schneider theorem says that e7r <> = (_1)-i<> is transcen-
dental whenever ia is algebraic of degree at least 2 over Q.
Example 6.2.2. Now let's look at an example with class number larger
than 1. For example, consider the field K = Q ( y' -15) and its ring of
integers RK = Z[a], where to ease notation we will write a = 1 + fI5. It

is not hard to check that RK has class number 2 and that a non-trivial ideal
class is given by a = 2Z + aZ. Further, One can check that the field H =
K ( v'5 , R ) is everywhere unramified over K, so it is the Hilbert class
field of K. (See exercise 2.11.) It follows from (4.3) that H = K(j(RK))
and that Q (j (RK )) is a quadratic extension of Q contained in Hand
disjoint from K. Hence Q(j(R K )) is either Q(v'5) or Q(R). We will
see in a moment that j(RK) E JR, so we must have Q(j(RK)) = Q(v'5).
(This also follows from exercise 2.9, which says in general that j(c) E JR if
and only ifC 2 = 1 in e.c(RK ).)
It remains to compute j(R K ) explicitly as an element ofQ (v'5). Let A
and B be rational numbers so that
From (4.3c) we see that j(a) is the Gal(K/K)-conjugate of j(R K ), so
j(a)=A-BV5.
Solving these two equations for A and B gives
A = j(R K ) + j(a) and

B = j(RK) - j(a).
2 2v'5
In order to compute the two values of j numerically, we can use the q-

series (I.7.4b), where q = e27rir is the parameter for the normalized lat-
tice Z + ZT:
-
j(q) = ~ + 744 + 196884q + 21493760q2 + 864299970q3 + 20245856256q4

q
+ 333202640600q5 + 4252023300096q6 + 44656994071935q7 + ....
Thus for RK we find that
j(RK) = j(Z + aZ) = j (e 27ri <» = j( _e-v'I57r )

;:::; j (-5.19748331238· 10- 6 ) ;:::; -191657.832863.
(Notice in this case that q = _e- V15 71' E JR, so j(RK) E JR.) Similarly for a
we calculate
j(a) = j(2Z + aZ) = j ( Z + ~aZ ) = j(e- V1571'/2i)
~ j (2.27979896315 . 1O- 3 i) ~ 632.83286254.
Using these values gives (to 12 significant digits) A = -95512.5000002

and B = -42997.5000001, so
j(R K ) +2-
~ -95512.5 - 42997.5V5 = -52515 - 85995- 1 v'5 [1 + v'5]
E Z -2- .
Thus j(R K ) is (at least approximately) integral over Z, which gives a nu-
merical verification of Theorem 6.1 for this example.
The A nalytie Proof of Theorem 6.1
Before beginning, we give a few words of motivation. It is not hard to see

that an elliptic curve E has complex multiplication if and only if there is
an endomorphism E ~ E whose degree is not a square. This suggests that
we take an arbitrary elliptic curve E and a positive integer n and study
the set of all elliptic curves E' for which there is an isogeny E ~ E' of
degree n. We took this point of view in (I §9,1O) when we studied Heeke
operators. What we are going to do is show that in this situation j(E') is
integral over Z [j (E)]. We will do this by explicitly constructing a monic
polynomial Fn (j(E), X) with coefficients in Z[J(E)] having j(E') as a root.
Finally, if E has complex multiplication, then for an appropriate choice of n
we can take E' = E. This means that Fn(j(E),j(E)) = 0, which we will
show implies that j(E) is integral over Z.
We now begin the analytic proof of Theorem 6.l. We fix a positive
integer n and recall the sets of matrices 'Dn and Sn defined in (I §9):
'Dn = { (~ ~) E M 2 (Z) ad - be = n} ,
Sn = { (~ ~) E M 2 (Z) ad = n, d> 0, 0:::; b < d}.
We also recall (1.9.2), which says that Sn = SL 2 (Z)\'D n . For any ma-
trix a = (~ ~) E M2 (JR) with det a > 0, we define the function j 0 a as
usual by the formula
. () =J.(aT
Joa T -+ -b) .
cT+d
Of course, if a E SL 2(1:), then j 0 a = j. We are going to study the

polynomial
whose coefficients 8 m = 8 m (T) are holomorphic functions on the upper half-

plane H. More precisely, 8 m is the m th elementary symmetric function in
the j 0 a's. We are going to prove several claims concerning the 8 m 'S.
Claim 1: 8m ('yT) = 8m (T) for all 'Y E SL2(1:) and all T E H.
Let I E SL2(1:). For any a E Sn we have al E :D n , so (1.9.2) says that
there is a (unique) 8", E SL 2(1:) such that baal is back in Sn. Further,
if 8aal = 8f3!h for some (3 E Sn, then (3 = (bi318a)a, so (1.9.2) implies
that a = (3. In other words, the map
is one-to-one, hence is a bijection since Sn is a finite set.

Now we observe that
{j 0 (a l ) : a E Sn} = {j 08;;10 (8aal) : a E Sn}

= {j 0 (8aal) : a E Sn} since j is SL 2(1:)-invariant
= {j 0 a : a E Sn} since Sn = {baa l : a E Sn}.
Hence any symmetric function on the set {j 0 a : a E Sn} will be invariant
under T f-+ IT for I E SL2(1:). In particular, this applies to the 8 m (T)'S,
which completes the proof of Claim l.
Claim 2: 8m E qj].
In other words, we are claiming that there is a polynomial fm(X) E qX]
such that 8m (T) = fm(j(T)) for all T E H. From Claim 1 we know that 8 m is
holomorphic on H and is SL 2 (1:)-invariant. In particular, 8m (T+1) = 8m (T),
so 8 m has a Fourier expansion in q = e2 -n:iT. We want to study what happens
as T --+ ioo, or equivalently as q --+ O. Recall (1.7.4b) that j has the Fourier
expansion j = q-1 + ~k>O Ckqk, so j has a pole of order 1 at q = o. Now
if a = (3 ~) E Sn, then -
so in particular qn+l (j 0 a) (T) --+ 0 as q --+ O. It follows from the definition

of the 8 m 'S that there is an integer N such that qN 8m (T) --+ 0 as q --+ o.
This means that each 8m (T) is meromorphic at 00 (see I §3), so 8 m is a
modular function of weight 0 which is holomorphic on H. Now (1.4.2b) says
that 8 m E qj], which completes the proof of Claim 2.
Claim 3: The Fourier expansion of 8 m has coefficients in Z.

To ease notation, let ( = e21filn and Q = ql/n = e 27riT In. For any 0
(3 ~) E 3 n we have
2rr'; ar+b
qOO(T)=e d = (abQa 2 •
(Note that ad = n.) Using the q-expansion of j(T) (I.7.4b), we find as

above that j 0 0 has the Q-expansion
j 0 a(t) = (-abQ-a 2 +L
"" Ck(abkQa 2 k,
k=O
where CO, Cl, ... are integers. In particular, the Fourier coefficients of j 0 0
lie in Z[(], and so the same is true of the 3 m 'S.
Let a E Gal(Q(()/Q), and write ('" = (r(O") for some integer r(a)
relatively prime to n. If we apply a to the Q-Fourier coefficients of j 0 0,
we get the series
(j 0 or = (-r(u)abQ-a 2 +L
"" Ckc(u)abkQa 2 k.
k=O
Comparing the series for j 0 0 and (j 0 0)"', we see that
r(a)b)
d .
In general, the value of j 0 (3 ~) only depends on b (mod d), since
and j is SL 2 (Z)-invariant. Further, if r is any integer prime to n = ad,

then the set {rb : 0 S; b < d} is a complete set of residue classes modulo d.
It follows that for any integer r relatively prime to n we have
{. (a
)0 0
rb)
d :
(a 0
Applying this with r = r(a) for a E Gal(Q(()/Q), it follows that

{(joa)"": OESn}={joo: OES n }.
Now consider the Q-Fourier coefficients of the 3m (T)'S, which we know
from above lie in Z[(]. Since 3m (T) is a symmetric polynomial in the
functions {j 0 0 : 0 E 3 n }, we see that its Q-Fourier coefficients are fixed
by Gal(Q(()/Q) and so lie in Q. Hence the Fourier coefficients of Sm(T)
are in Z[(]nQ = Z. Finally, we note that arn(T+l) = am(T) from Claim 1,
so am is in fact represented by a Fourier series in q = Qn. This completes
the proof of Claim 3.
Claim 4: 8m (T) E Z[j].

\Ve already know from Claim 2 that Sm E C[j] and from Claim 3 that 8m E
Z[q, q~l]. \Ve will show that
which will give the desired result. Let f(j) E C[j] n Z[q, q~l] be a polyno-
mial of degree d, and wriLe f(j) = ao1'l + add~l + ... + ad with ai E C.
Substituting in the q-expansion of j (1. 7.4b) gives
f = ao + a 1 + 744dao + ...
qd qd~l '
so the fact that f E Z[q, q~l] implies that ao E Z. Now
f - aoll = add~l + ... + ad E C[j] n Z[q, q~l].

so repeating the above argument gives (11 E Z. Continuing in this way, we
find that every coefficient of f in Z, which completes the proof of Claim 4.
Combining Claims 1, 2. 3, and 4, we have completed the proof of the
first half of the following important result.
Theorem 6.3. (a) There is a polynomial Fn(Y X) E Z[Y, X] so that
II (X - j 00) = Fn(j,X).
nES n
(b) Let ,3 E J\h(Z) be a matrix with integer coefficients and det;3 > O.
Then the function j 0 .3 is integral over the ring Z [j].
(c) Ifn is not a perfect square, then the polynomial Hn(X) = F,,(X,X)
is non-constant and has leading coefficient ± 1.
PROOF. (a) The four claims proven above say that
II (X - j 0 0) = L 8m X m with 8m E Z[j].
nES n
(b) Let Tl = det;3, so ;3 E 'Dn. Using (1.9.2), we can find a matrix 1 E

SL 2 (Z) such that 1;3 E Sn. The SL2(Z)-invariance of j says that j 0;3 =
j 0 (r(:I), while the definition of Fn shows that X = j 0 (r,3) is a root
of Fn(j,X). Since Fn is monic by definition and has coefficients in Z[j]
from (a), it follows that j 0 ,3 is integral over Z[j].
(c) Let ct = (3 ~)
Sn. Then using the Q-expansion of j 0 a described
E
above during the proof of Claim 3, we sec that the Q-expansion of j - j 0 a
is
(Here we are again writing ( = e 21ri / n and Q = ql/n.) Since n is not a

square, the leading terms cannot cancel, so j - j 0 a has a pole as Q - 7
o and the coefficient of the leading term is necessarily a root of unity.
(Precisely, the coefficient is 1 if n > a 2 , and it is _(-ab otherwise.) It
follows that Fn(j,j) has a pole as Q - 7 0 and that the leading Q-coefficient
is a root of unity. But the Q-expansion of Fn(j, j) has integer coefficients,
so the leading coefficient is a root of unity in Z; hence it must be ±1.
Further, Fn (j, j) is actually a series in q = Qn, so we have proven that
for some m ~ 1. But we also know that Fn(j,j) E ZU] and that j has a
simple pole at q = O. Hence Fn(j,j) = ±j'm + ... E Z[j], which proves
that Fn(X, X) is a non-constant polynomial with leading coefficient ±l.
o
It is now a simple matter to complete the proof of Theorem 6.l.

Corollary 6.3.1. (Theorem 6.1). Let EIC be an elliptic curve with
complex multiplication. Then j(E) is an algebraic integer.
PROOF. Let R ~ End(E) be an order in a quadratic imaginary field K. We

consider first the case that R = RK is the ring of integers of K. Choose
pi
some element pER such that n = IN[f is not a perfect square. For
example, if K = Q(i), take p = 1 + i, and if K = Q (V-D) with square-
free D ~ 2, take p = V-D. Then (l.5b) says that the isogeny [p] : E -7 E
has degree n. Fix aTE H with jeT) = j(E). Then multiplication by p
sends the lattice ZT + Z to a sublattice of index n, say
pT = aT +b
for some a, b, c, d E Z with ad - bc = n.
p=cT+d
So if we let a = (~ ~) E TIn, then
.
J(aT) = J
. (aT +
CT + d
b).J(T) = J(E).
=
.
By definition, j 0 a is a root of Fn(j, X), so if we substitute X = j 0 a and
evaluate at T, we get
0= Fn(j(T),j(aT)) = Fn(j(E),j(E)) = Hn(j(E)).
From (6.3c), the polynomial Hn(X) has integer coefficients and leading
coefficient ±1. This proves that j(E) is integral over Z.
Now we deal with the case that R is an arbitrary order in K. Let A =

ZWI + ZW2 be a lattice for E. From [AEC VI.5.5] we know that K =
Q(WI/W2). Hence replacing A by )'A for an appropriate). E C*, we may
assume that A c RK. We also choose a 7 E H so that RK = Z7 + Z. Then
we can write
for some a, b, c, d E Z.
W2=C7+d
Let n = ad - bc. Switching WI and W2 if necessary, we may assume
that n ~ 1. The matrix a = (~ ~) is in nn, so (6.3b) says that the func-
tion j 0 a is integral over the ring Z[j]. Taking the equation Fn(j, X) = °
which gives that integrality and evaluating it at 7, we find that j(a7) is
integral over Z [j (7)]. But j (a7) = j (E), and we already know that j (7) is
integral over Z because it is the j-invariant of an elliptic curve with complex
multiplication by R K . Therefore j(E) is integral over Z. 0
Example 6.3.2. The polynomials Fn(Y, X) E Z[Y, X] and Hn(X) E Z[X]

described in (6.3) can be extremely complicated. For example,
F2(Y, X) = _(Xy)2 + X 3 + y3 + 24 .3.31. XY(X + Y)
+ 34 .53 .4027. XY - 24 .34 . 53(X2 + y2)
+2 8 .37 . 56 (X + Y) - 212 . 39 .5 9 ,
H 2 (X) = _X4 + 2.1489. X 3 + 34 .5 4 .17.47. X2
+2 9 .3 7 .5 6 . X - 212 . 39 . 59.
According to [AEC III.4.5], the elliptic curves E : y2 = x 3 + ax 2 + bx
and E' : y2 = x 3 - 2ax 2 + (a 2 - 4b)x are connected by an isogeny E --> E'
of degree 2. It follows from exercise 2.19 that
. ., (256(a 2 -3b)3 16(a 2 +12b)3)
F2(J(E),J(E)) = F2 b2(a 2 _ 4b) , b(a2 _ 4b)2 = 0,
a fact that the interested reader can check by a direct computation (prefer-
ably with the assistance of a symbolic calculator).
The £-adic Proof of Theorem 6.1
We now begin the £-adic proof of Theorem 6.1. If E / L is an elliptic curve

with complex multiplication, then we know from (2.3) that the action
of Gal(L/ L) on E tors is abelian, so in particular the action on the Tate
module T£(E) is abelian. (For another proof of this result which uses only
a little linear algebra and is valid even if End(E) is not a maximal order,
see exercise 2.6.) We now use Proposition 2.3, local class field theory, and
the criterion of Neron-Ogg-Shafarevich to prove that E has everywhere po-
tential good reduction. Although somewhat involved, this proof has the
advantage that it generalizes to abelian varieties of arbitrary dimension
(see Serre-Tate [1]).
Theorem 6.4. Let L be a number field and E / L an elliptic curve with

complex multiplication. Then E has potential good reduction at every
prime of L.
PROOF. (Serre-Tate) Every endomorphism of R is defined over a finite

extension of L (2.2b), so replacing L by a finite extension, we may assume
that EndL(E) is strictly larger than Z. Fix a prime v of L. We set the
following notation:
Lv = the completion of L at v,
Rv = the ring of integers of Lv,
rotv = the maximal ideal of R v ,
p = char Rv/rotv = the residue characteristic of R v ,
f = a rational prime not equal to 2 or p,
Iv = the inertia subgroup of Gal(Lv/Lv),
L~b = the maximal abelian extension of Lv,
I~b = the inertia subgroup ofGal(L~b/Lv).
By assumption, EndL(E) -=I Z, so certainly EndL v (E) is strictly larger

than Z. Applying (2.3), we see that the action of Gal(Lv/ Lv) on Tt(E) is
abelian. In particular, Iv acts through the quotient I~b.
Local class field theory says that there is an isomorphism
ab!:>! R*
lv -v·
(See, e.g., Lang [5], Serre [4, XIV §6, Cor. 2(ii) to Thm. 1], Serre [5].) This
gives us a very good picture of I~b, since we can decompose R~ using the
exact sequence
(Rv/rotv )*
--------
1 ~
R~,l ~ R*v ~ ~ 1.
--------
pro-p group
III
lab
finite
Here R~,l is the group of I-units,
R~,l = {u E R~ : u == 1 (mod rotv )}.

There is an isomorphism from the formal multiplicative group Gm (rotv )
to R~,l given by
t f-----+ 1 + t.
Hence R~,l is a pro-p group; that is, it is the inverse limit of finite groups
of p-power order (see [AEC IV.3.l.2] and [AEC IV.3.2]).
Similarly, if we fix an isomorphism Aut T£(E) ~ GL 2 (Z£) correspond-
ing to some basis for T£ (E), then there is an exact sequence
1 --> GL 2 (Zdl --> GL 2 (Z£) --> GL 2 (ZIi!.Z) --> l.

'"--..---'
pro-£ group
III
Aut T£(E) --> AutE[i!.]
'---v--'
finite
Here GL 2 (Z£h is the group of matrices congruent to the identity matrix
modulo i!., and it is not hard to see that this is a pro-i!. group. More precisely,
the logarithm map gives an isomorphism
where M 2 (Zl) is the group of 2 x 2 matrices with coefficients in i!.Zl under

addition. This isomorphism is the GL 2 analogue of [AEC IV.6.4b]. (See
also exercises 2.22 and 2.23.)
It follows from the above discussion that the map
Iv --> Aut Tl(E)
fits into the following diagram:
Iv
lab
1
v
III
1 --> R~,l R*v --> (Rvlmv)* --> 1
1
1 --> GL2(Z£h --> Aut Tl(E) --> GL 2 (ZI CZ) --> 1
Next we observe that since i!. i= p, then there can be no non-trivial ho-
momorphisms from a pro-p group to a pro-i!. group, so the images of R~,l
and GL2(Zlh in Aut T£(E) have trivial intersection. Therefore there is an
injection
Image(R~,l --> Aut T£(E)) <---> GL 2 (ZIi!.Z).
Since also (Rvlmv)* is finite, it follows that
Image(R~ --> AutT£(E)) is finite,

since it consists of finitely many cosets ofImage(R~,l ~ Aut Tf(E)).

This proves that the image of Iv in Aut Tf(E) is finite. Now the crite-
rion of Neron-Ogg-Shafarevich (specifically [AEC VII.7.3]) says that E has
potential good reduction at v, which concludes the proof of Theorem 6.4.
o
It is now a simple matter to deduce (6.1) from (6.4).
Corollary 6.4.1. (Theorem 6.1). Let EIC be an elliptic curve with
complex multiplication. Then j(E) is an algebraic integer.
PROOF. The elementary result (2.1b) says that j(E) is an algebraic num-
ber, so we may take an equation for E with coefficients in the number
field L = Q(j(E)). Then (6.4) says that E has potential good reduction
at every prime of L, so [AEC VII.5.5j implies that j(E) is integral at every
prime of L. 0
§7. Cyclotomic Class Field Theory
In this section we are going to formulate the class field theory of Q in

terms of special values of analytic functions, specifically special values of
the exponential function. This is analogous to the way we will later be
describing the class field theory of quadratic imaginary fields via the theory
of complex multiplication. We hope that studying the simpler cyclotomic
case first will aid the reader in understanding the more intricate proofs
required in the complex multiplication case. However, the results in this
section will not be used later, so the reader who already feels comfortable
with class field theory may wish to skip directly to §8.
We begin with the multiplicative group
The exponential map provides a complex analytic parametrization of the

multiplicative group,
f: C/Z
t
Sitting inside of Gm(C) is its torsion subgroup
The elements of Gm(Chors are roots of unity, so they generate abelian

extensions of Q. Our aim is to give an analytic description of the action
of Gal(Qab IQ) on Gm(Chors.
From class field theory (see §3) we have the reciprocity map, which is
a surjective homomorphism
AQ -+ Gal(Qab /Q), 8 t---+ [8, Q].
Each idele 8 thus defines an isomorphism
Gm(Chors -+ IGm(C}tors,
( t---+ ([s,lQll.
The algebraic action of Gal(Qab /Q) on IGm(C)tors is determined by these

isomorphisms.
In general, if x E AQ is any idele, we want to define a subgroup xZ c Q
and a multiplication-by-x map
Q/Z~Q/xZ.
The definition of xZ is easy; it is just the ideal of x, which we recall is the

fractional ideal of Q given by
xZ = (x) = IIpordpx p . Z = NxZ.

p
For convenience, we will write N x as indicated for a rational number gen-

erating the ideal xZ. Later we will pin down N x precisely by requiring
that sign(Nx ) = sign(x oo ).
In order to define the multiplication-by-x map, we decompose Q/Z
into its p-primary components and multiply the p-component by xp. Note
that
(p-primary component of Q/Z) = Z(P-l]/Z 9! Qp/Zp.
The first equality is immediate, since if t = a/n E Q/Z has p-power order,
then n must be a power of p. For the second equality, we clearly have an
injection
Z(P-l]/Z ~ Qp/Zp.
To check surjectivity, let ~ E Qp/Zp. We can write ~ = a/pe for some a E
Zp and some integer e ::::: O. Choose an integer a E Z with a == a (mod pezp).
Then
and
Similarly, we observe that for any N E Q*, the p-primary part of Q/ NZ is

isomorphic to Qp/NZp.
It is a general fact that an abelian group whose elements all have finite
order is the direct sum of its p-primary components. (See (8.1) and (8.1.1)
for something stronger.) Hence
and for any idele x E AQ,

p p
The last equality follows from the fact that ordp(Nx ) = ordp(xp), so the
ideals NxZ p and xpZp are the same. Now we can define the multiplication-
by-x map to be multiplication of the p-component by xp; in other words,
multiplication-by-x is defined by the commutativity of the following dia-
gram:
p p
We are now ready for the main theorem of this section. The reader
should compare this cyclotomic result (7.1) with the corresponding complex
multiplication theorem (8.2).
Theorem 7.1. Fix the following quantities:
u E Aut(C), an automorphism of the complex numbers,
8 E AQ, an idele ofQl satisfying [8, QlJ = UllQlab.
Further, fix the complex analytic isomorphism

J: C/Z ~ IGm(C), J(t) = e 27fit .
Then there exists a unique complex analytic isomorphism
l' : C/8- 1 Z ~ IGrn(C)
so that the following diagram commutes:
~ Ql/8- 1 z
-1
a
---'t
Remark 7.1.1. Theorem 7.1 says that

J(t) [s,lQl] = !'(s-lt) for all t E Ql/Z.
Of course, l' depends on s. We will see during the proof of (7.1) that
!,(t) = e27riNst,
where Ns is a certain non-zero rational number. Thus written out explic-

itly, (7.1) says that
for all t E Ql/Z.
This formula makes it very clear how the algebraic (Galois) action of [s, QlJ
is transformed into the analytic (multiplication) action t -+ Nss-1t. Later
we will have more to say about N s , see (7.2).
PROOF (of Theorem 7.1). Let t E Q/Z, say t = a/n(mod Z) as a frac-

tion in lowest terms. To ease notation, let ( = f(t) be the corresponding
primitive nth-root of unity. Suppose first that our idele s has the property
sp == 1 (mod nZp ) for all primes p, and further that s= > o. (*)
In particular, sp is a unit for all primes dividing n, and we know that Q( ()

is ramified only at these primes, so by (3.5) the action of [05, Q] on Q(() is
given by the Artin symbol
[s,Q]IQl(() = ((s),Q(()/Q).
For any idele s, we will write Ns E Q* for the unique rational number
satisfying
and
Then ((s),Q(()/Q) = (NsZ,Q(()/Q), from which it follows that ([;;.QlJ =

(Ns, or equivalently
Next we decompose t (mod Z) into p-primary components,
Then
by definition of multiplication by S-l
== t.
This suggests that we should take f' to be the map is defined by
Then for any idele s satisfying (*), we have
To recapitulate, we have proven that
provided t == 5!. E Q/Z and s satisfies (*).

n
Now let s E AQ and t E IQ/Z be arbitrary, and as usual write t =

a/n (mod Z). Using the weak approximation theorem, we can find a ratio-
nal number r E IQ* so that r s satisfies (*). From the definitions, it is easy
to check that
and
Notice that the requirement sign(Ns ) = sign(soo) ensures that N rs = rNs,

even when r is negative. Using these equalities, we compute
f(t) [s,lQl] = f(t)[rs,lQl] since [rs, IQJ = [s, IQJ
= frs((rs)-St) from above, since rs satisfies (*)
= e27riNrs-{rs)-1t from the definition of frs
= e27riNs·s-1t since N rs = rNs and (rs)-1t = r-1(s-1t)
= fs(S-lt) from the definition of fs.
This completes the proof of the existence half of (7.1), with the additional
information that f' is given by the map f' (t) = fs (t) = e27riNs t. As for
uniqueness, we need merely observe that the commutative diagram deter-
mines f' on IQ/ s -1 Z, which is a dense subset of G mUC), so there is at most
one possibility for f'. 0
As an alternative version of (7.1), we could use only the single analytic

parametrization f and replace the multiplication-by-s-1 map so as to make
the following diagram commute:
We can (try) to do this because every a E Aut(C) maps Gm(C) to itself.

In the case of an elliptic curve E, this will only be possible for those a's
such that EO- ~ E. For the elliptic analogues of our next two results (7.2)
and (7.3), see (9.1) and (9.2).
Theorem 7.2. Let s E AQ be an idele. With notation as in (7.1), there
is a unique rational number Ns E IQ* such that the following diagram
commutes: N sS -1
--->
Gm(C) ~ GmUC).
More precisely, Ns is the unique rational number satisfying
and
PROOF. During the proof of (7.1) we showed that f' is the map f'(t) =
where Ns is as specified in (*). The commutative square in (7.1)
e27riNst,
then says that
This proves that the square in (7.2) is commutative with Ns chosen to

satisfy (*). It remains to prove that this commutative diagram uniquely
determines N s . But if N~ also makes the diagram commute, then we
find that multiplication by N s- 1 N~ induces the identity map on rQjZ.
Hence Ns = N~. 0
From (7.2) we have a well-defined map
AQ ----> Q* c C,
and it is clear that this map is a homomorphism. Further, the explicit
description of Ns given by (*) in (7.2) shows that the map is continuous.
Recall that for any number field L, a homomorphism
X: AI, ----> <C*

is called a Grossencharacter of L if it is continuous and satisfies X(L*) = 1;
that is, the kernel of X must contain the image of L * in A I,. It is easy to
see that our map S I---> Ns does not have this property. In fact, if s is the
image of some a E L *, then clearly Ns = a. We can get a Grossencharacter
by making a small modification to N s .
Theorem 7.3. For any idele s E AQ, let Soo be the archimedean compo-
nent of s. Define a map
X : AQ ----> lR. * c C*,
where Ns E Q* is the unique rational number satisfying
NsZ=(s)=sZ and
Then X is a Grossencharacter of Q.
PROOF. It is clear that both of the maps
and Sf------> Soo
are continuous homomorphisms from AQ to C*. Further, they clearly take

the same value on the image of Q* in AQ. Hence X is a continuous homo-
morphism that is trivial on Q*; that is, it is a Grossencharacter.
o
§8. The Main Theorem of Complex Multiplication
Let K be a quadratic imaginary field with ring of integers RK as usual.

For each prime ideal p of K, let Kp be the completion of Kat p and let Rp
be the ring of integers of Kp. Similarly, if (1 is any fractional ideal of K,
let (1p = (1Rp be the fractional ideal of Kp generated by (1.
Let M be an RK-module. The p-primary component of M, which by
definition is that part of M annihilated by some power of p, is denoted by
M[pOO] = {m EM: pe m = (0) for some e 2: o}.

We begin with an elementary lemma about p-primary decompositions.
Lemma 8.1. (a) Let M be a torsion RK-module; that is, for everym E M
there is a non-zero a E RK such that am = O. Then the natural summation
map
S : EB M[pOO] ~ M,
p
is an isomorphism. Here the sums are over all prime ideals of R K , and /-Lp
denotes the p-component of /-L.
(b) Let (1 be a fractional ideal of K. Then for each prime ideal p of K, the
inclusion K '----+ Kp induces an isomorphism
(c) Again let (1 be a fractional ideal of K. Then there is an isomorphism
Remark 8.1.1. As our proof will show, Lemma 8.1 is true more gen-
erally for any Dedekind domain R with fraction field K. For example,
taking R = Z gives the decomposition of a torsion abelian group into p-
primary components as discussed in §7.
PROOF (of Lemma 8.1). (a) Suppose first that /-L E kereS). For each
prime p, let e(p) 2: 0 be the smallest integer such that pe(p) /-Lp = (0).
Note that e(p) exists, since /-Lp E M[poo], and that all but finitely many
of the e(p) 's are zero since /-L has only finitely many non-zero components.
Now fix a prime ideal q and let
By construction, we have ilJ1p (0) for all p =I q. On the other hand,

since 8(J1) = 0 by assumption, we have
(0) = il8(J-t) = il L J-tp = L ilJ-tp = ilJ-tq.

p p
But il is relatively prime to q, so () + qe(q) = (1). Hence
which proves that J-tq = O. Since q was arbitrary, we have proven that J-t = 0
and hence that 8 is injective.
Next we check surjectivity. Take any element m E M, and choose a
non-zero a E RK with am = O. Factor the ideal aRK as
Then we can find 101, ... ,lOr E RK satisfying
101 + ... + lOr =1 and
[Proof: Let e; = ap-e i . Then e1 + .. ·+e r = (1), so it suffices to take 10; E e;

with 101 + .. ·+cr = 1.] Notice that pe i c ; C aRK, so peic;m C amR K = (0).
Hence Cim E M[Pi], so if we set
= {cim if p = Pi for some 1 :::; i :::; r,

J-tp 0 otherwise,
then J-t E EBM[poo] and 8(J-t) = L,J-tp = 101m + ... + Crm = m. This
completes the proof that 8 is surjective.
(b) First, suppose that a E (Kia) [pOOl is in the kernel of T. Choosing a
representative a E K for a, this means that pea C a for some integer e ~ 0
and that a E ap = aR p • These two inclusions imply respectively that
Therefore aRK C a, so a E a, which means that a = O. This proves that T

is injective.
Next, let fJ E Kp/ap and choose a representative f3 E Kp for fJ. By the
weak approximation theorem (essentially the Chinese Remainder Theorem)
we can find an a E K satisfying
and ordq(a) ~ ordq(a) for all q =I p.

The first inequality says that a == (3 (mod ap), so T(a) = 13. Let e be a non-
negative integer greater than ordp(a) - ordp(a). Then ordq(pe a ) ~ ordq(a)
for all primes q, including q = p, so pea Ea. Hence a E (K/a) [1'00], which
completes the proof that T is surjective.
(c) This is immediate from (a) and (b). 0
Let x E A'K be an idele. Recall that the ideal of x is the fractional

ideal
(x) = IIpordp(x p).
p
If a is any fractional ideal of K, we define xa to be the product (x)a. Using

the equality (x)p = (x)Rp = xpRp, we see that
Now (8.1c) gives natural isomorphisms
and K/xa ~ EBKp/xpap.

p
We define the multiplication-by-x map on K / a to be multiplication

of the p-primary component by xp. In other words, multiplication-by-x is
defined by the commutativity of the following diagram:
x
K/a ~ K/xa
11 11
EBKp/ap ~ EB Kp/xpap
p p
(tp) f---+ (Xptp)
We also recall from §3 that the reciprocity map for K,
S f---+ [s, K],
is surjective and its kernel contains K*.

We are now ready to state and prove the main theorem of complex
multiplication in its adelic formulation. At the risk of making the statement
overly long, we include a summary of our notation and assumptions.
Theorem 8.2. (The Main Theorem of Complex Multiplication) Fix the
following quantities:
K /Q a quadratic imaginary field with ring of integers RK,
E/f:- an elliptic curve with End(E) ~ RK,
E Aut(C), an automorphism of the complex numbers,

s E A~, an idele of K satisfying [s, K] = O"IKab.
Further, fix a complex analytic isomorphism
I: Cia ~ E(<C),
where a is a fractional ideal of K. Then there exists a unique complex

(depending on I and 0") so that the following diagram commutes:

Kia ~ KIs-1a
if if'
E(<C) ~ Ea(<C).
Remark 8.2.1. The statement of Theorem 8.2 remains true for elliptic
curves whose endomorphism ring is a non-maximal order of K. Of course,
one first must explain how to mUltiply Kia by an idele x when a is an
arbitrary lattice in K. For details, see Shimura [1] or Lang [1, Ch. 8, 10].
Remark 8.2.2. Notice how Theorem 8.2 transforms the algebraic action
of 0" on the torsion subgroup I (Kia) = E tors into the analytic action of
multiplication by S-l:
for t E Kia and s E A~.
Compare with (7.1.1).

PROOF (of Theorem 8.2). Clearly, there is at most one 1', since the
commutative diagram determines I' on K Is-la, which is a dense subset
ofCls- 1a.
Suppose that EdC is an elliptic curve that is isomorphic to E and
that h : Cia 1 ---+ E1 (<C) is an analytic isomorphism. We are going to
begin by proving that if Theorem 8.2 is true for (£1. h), then it is also
true for (E, f). This will allow us to reduce to the case that E is defined
over Q (j (E)) and a is an integral ideal.
So we are assuming that there is an analytic isomorphism
and a commutative diagram

-1
Kl a1 "----+ K/s- 1a1

Since E and El are isomorphic, we may fix an isomorphism i : El -::+ E.

Further, the lattices for E and El are homothetic [AEC VI.4.l.1], so we
have al = ')'a for some,), E K*. Then each of the squares in the following
diagram is commutative:
1/;
Ef(e)
1i~
EC7(e).
Hence (8.2) is true for (E, f) if we take for f' the map
We are now reduced to proving (8.2) under the assumptions that E

is defined over Q(j(E)) and that a C RK is an integral ideal. Fix an
integer m ~ 3 and let Lj K be a finite Galois extension satisfying
j(E) E L and E[m] C E(L).
We note that (5.6) implies that L contains K(m), the ray class field of K
modulo m. We are going to begin by proving that (8.2) is true on the m-
torsion points of E. As usual, our main tool will be reduction modulo a
suitable prime.
Let l,}} be a prime ideal of L satisfying the following five conditions:
(i) aiL = (l,}}, Lj K); that is, the restriction of a to L is a Frobeniuse-
lement for l,}}.
(ii) P = l,}} n K is a prime of degree 1; that is, p = Nffp is a rational
prime.
(iii) p is unramified in L.
(iv) p is not one of the finitely many primes excluded in (4.5).
(v) l,}} does not divide m.
Such an ideal always exists, since the Tchebotarev Density Theorem (Lang
[5, Ch. VIII, Thm. 10]) says that there are infinitely many primes satis-
fying (i) and (ii), whereas each of (iii), (iv), and (v) excludes only finitely
many primes.
Using the fact that L contains K(rn)' we find
[s,K]IK (=) =al K (=) since [s, K] = alKab by assumption
= (p,K(rn)/K) from (i) and (ii).
This last map is the Frobenius element associated to p. Let 1r E A'k be an

idele with a uniformizer at the p-component and l's elsewhere. Then (3.5c)
says that [1r,K] also equals (p,K(m)/K). Hence [s1r- 1 ,K] acts trivially
on K(m)' so the idelic characterization (3.6) of the ray class field K(rn) says
that the idele S1r- 1 factors as
Here 0: E K*, and for each prime q, U E A'k satisfies
Uq E R~ and Uq == 1 (mod mR q ).
Next we use (5.3) to find an isogeny
whose reduction modulo IlJ is the pth_power Frobenius map. Note that
since L contains K(j(E)), (i) implies that
aIK(j(E)) = (IlJ,L/K)IK(j(E)) = (p,K(j(E))/K).
Since E is defined over Q(j(E)), we see that the isogeny described in (5.3)
is indeed from E to Er7.
We claim that on m-torsion points, ). acts like a. To see this, let T E
E[m] and use tilde's to denote reduction modulo 1lJ. Then
)'(T) = 5..(T) = Tr7,
since both), and aiL = (1lJ, L/ K) act on the residue field modulo IlJ as
the pth_power map. Now IlJ f m from (v), so [AEC VII.3.1b] says that
on m-torsion points the reduction map ElT[m] --+ ElT[m] is injective. Hence
for all T E E[m].
In other words, we have produced a commutative diagram
E[m] ~ ElT[m]
1 1
E(C) ~ E""(C).
(N.B. This diagram describes a deep mathematical relationship, because it

transforms the algebraic action of (J on m-torsion points into the geometric
action of >.. Notice that for general points P E E(k), >.(P) and pO' will
not be equal.)
We also note from (4.3d) that EO' is isomorphic to p * E, where p * E
is an elliptic curve associated to the lattice p -1 a. Using the given analytic
isomorphism f : Cia ----+ E(C), this means there is an analytic isomor-
phism ff!: C/p-1 a ----+ EO'(C) so that we have a commutative diagram
id
------> C
1 1
C/p-1 a (*)
if"
EO'(C).
Recall that we factored the idele 8 as 8 = O:7rU. Since every component

of U is a unit, and every component of 7r is 1 except for a uniformizer in
the p-component, we have
(8) = (0:)(7r) = (o:)p, and so
Thus multiplication by 0:- 1 gives an isomorphism
so we can extend (*) to form the larger commutative diagram

id a- 1
C ------> C ------> C
1 1 11
Cia ------> C/p-1 a ------> C18- a (** )
if A
if"
id
if'
E(C) ------> EO'(C) ------> EO'(C).
Here f' : C I 8 -1 a -=:-. EO' (C) is the unique analytic isomorphism making (**)
commute. We claim that f' satisfies
To verify this claim, we note from above that
f(t)O' = >.(J(t)).
Combining this with the commutativity of (**), we must check that
for all t E m- 1 a/a;
and since l' is bijective, this is equivalent to showing that

?
a-It - S-l t E s-la
Recalling that s-l t E K/s- 1 a is defined by multiplying the q-primary

component by Sq, we must prove for each prime q of K that
Now multiplying through by Sq and using the decomposition Sq

a7r qu qfrom above, we must check that
or equivalently that
?
(7r q U q - l)a q C ma q .
By construction, u q is in R~ and satisfies u q 1 (mod mRq), so we are
red uced to proving
?
(7r q -l)a q C ma q •
There are two cases to consider. First, if q # p, then 7r q = 1, so we are
done. Second, if q = p, we know that 7rp is a uniformizer, so (7r p -1) ap = ap.
Further, we know from (v) that p f m, so m is a p-adic unit and hence map =
ap . This proves the desired inclusion for all q, thereby completing the proof
of our claim.
To recapitulate, for each integer m ::::: 3 we have produced an analytic
isomorphism
and a commutative diagram

s-1
m- 1a/a ---+ m- 1 S-1 a/ s-1a
1/ 1/:..
u
E(C) ---+ Ei7(C).
To complete the proof of (8.2) it suffices to show that all of the maps f:n
are the same, since then these commutative diagrams will fit together to
give the desired result on all of K/a.
§9. The Associated Grossencharacter 165
So let n 2: 1 be an integer, and let f:nn : Cis-Ia ~ EO"(C) be the

corresponding analytic isomorphism. Note that the composition f:nn 0 f:n -1
is an automorphism of EO", say f:nn 0 f:n -1 = [~] E Aut(EO"). Then for
any t E m-Iala we have
[~] 0 f:n(S-I t ) = f:nn(s-I t ) by definition of ~
= f(tt from construction of f:nn
= f:n(s-I t ) from construction of f:n.
This holds for all t E m-Ia, so we conclude that
[~]T=T for all T E EO"[m].
Since m 2: 3, this can only happen if [~] = [1], since for [~] -I [1] the kernel
of [1 - ~] contains at most six points. (If j(E) -I 0,1728, then ~ = ±1,
and in the two exceptional cases, ~4 = 1 or ~6 = 1. See [AEC IILlO.l].)
Therefore f:nn = f:n, which concludes the proof of (8.2). 0
§9. The Associated Grossencharacter
In this section we will use the main theorem of complex multiplication

to define a Grossencharacter associated to an elliptic curve with complex
multiplication. Recall that a Grossencharacter on a number field L is a
continuous homomorphism
1jJ : Ai ---+ C*
with the property that 1jJ(L*) = 1. Our first result describes a map Ai ~
C* which, with some small modifications, will be the desired Grossencha-
racter.
Theorem 9.1. Let ElL be an elliptic curve with complex multiplication
by the ring of integers RK of K, and assume that L ::) K. Let x E Ai
be an idele of L, and let s = Nf{x E Ak. Then there exists a unique a =
aE/dx) E K* with the following two properties:
(i) aRK = (s), where (s) c K is the ideal of s.
(ii) For any fractional ideal a C K and any analytic isomorphism
f: Cia ---+ E(C),
the following diagram commutes:
Kia Kia
Before beginning the proof of (9.1), we should make a few remarks.

First, from (2.3) we know that K(j(E), E tors ) is an abelian extension
of K(j(E)). Further, since E is defined over L, we know that j(E) E L,
and so we see that L(Etors) C Lab. This shows that the images of the verti-
cal maps in (ii) do lie in E(L ab ). Second, since K(j(E)) is the Hilbert class
field of K from (4.3), we know that N~2t is principal for any ideal 2t of L.
Since (s) = (N~x) = N~((x)), we see that there always exists an a E K*
satisfying (i), and (i) determines a up to a unit of K. It then remains
for (ii) to pin a down precisely. Third, we note that (i) gives
so the top row of the diagram in (ii) is well-defined.

L' = L(Etors ).
Since j(E) E L, it follows from (5.7) and (2.3) respectively that there are
inclusions
Choose an automorphism a E Aut(C) such that
alLab = [x,L].
A standard property of the reciprocity map (3.5b) says that
so applying the main theorem of complex multiplication (8.2), we find an

analytic isomorphism f' : Cia ----+ E(C) and a commutative diagram
8- 1
Kia -----+ Kis-la
11
CT
1/'
E(C) -----+ ECT(C).
Now E CT = E, since a fixes L. Hence a and s-la must be homothetic, so

there is a (3 E K* such that (3s-1a = a. Our commutative diagram then
becomes
Kia Kia
E(C) ~ E(C).
Note that 1" 0 f- 1 is an automorphism of E, say 1" = [~] 0 f. If we

set 0: = ~!3 and use the facts that alLab = [x, L] and E tors C E(Lab), we
get
as -1
Kia --'> Kia
E(Ub ) ~ E(Lab),
which is exactly (ii). Further, we have an equality of ideals
so o:R K = (s).
This proves that 0: satisfies both (i) and (ii), which completes the proof of
the existence of part of (9.1).
Next we check that 0: is unique. Suppose that 0:' E K* also has prop-
erties (i) and (ii). From (ii) and the fact that f and [x, L] are isomorphisms,
we get a commutative triangle
Kia
~ 0",.,--1
Kia Kia
Hence multiplication by 0:'0:- 1 is the identity map on Kia, so 0:' = 0:.
Finally we must show that 0: is independent of the choice of f. Suppose
that!, : Cia' -> E(C) is another analytic isomorphism. Then a = ,a'
for some, E K*, and I' 0 f- 1 is an automorphism of E, so there is a
unit ~ E R'K such that I'(z) = f(~,z). Then (ii) for f gives
l'(t)[x,L] = f(~,t)[x.L] = f(O:S-lbt) = !'(o:s-l t ) for all t E Kia,
so (ii) remains true if f is replaced by 1'. Hence 0: is independent of f.

o
Theorem 9.1 gives us a well-defined map
O:£/L : A~ --'> K* c C*,

and it is clear that O:£/L is a homomorphism. However, it is easy to see
that 0:£/ d L *) i= 1, so 0: £ / L is not a G rossencharacter. More precisely,
if!3 E L* and xj3 E A~ is the corresponding idele, then [Xi', L] = 1. So (9.1)
says that 0: = o:£/dxj3) is the unique element of K* such that o:R K =
N~((x(,))RK = N~(!3)RK and such that multiplication by GNJ(XiJl in-
duces the identity map on Kia. Clearly, the required 0: is just N~!3. In
other words, we have proven that
for all !3 E L * ,
which is the first step in proving the following important result.
Theorem 9.2. Let ElL be an elliptic curve with complex multiplication

by the ring of integers RK of K, assume that L :J K, and let aE/ L : Aj> --->
K* be the map described in (9.1). For any idele 8 E Ax, let 8 00 E C*
be the component of 8 corresponding to the unique archimedean absolute
value on K. Define a map
'l/JE/L : AI ---> C*,
(a ) 'l/JE / L is a Grossencharacter of L.
(b) Let S-lJ be a prime of L. Then 'l/J E /L is unramified at;:p if and only if E
has good reduction at S-lJ. (Recall that a Grossencharacter 'l/J : At ---> C* is
said to be unramified at S-lJ if 'l/J(R~) = 1.)
PROOF. (a) It is clear that 'l/JE/L is a homomorphism. We saw above that

if (3 E L *, then a E / d x (3) = N i (3. On the other hand, untwisting the
definitions we find
Ni(x{3)oo = II fr = Ni{3·
r:L<-+C
rlK=l
Therefore 'l/JE/dx(3) = 1. This holds for all (3, so 'l/JE/dL*) = 1.

Next we are going to verify that a E / L : At ---> C is continuous. Fix an
integer m 2': 3. We know from (2.3) that L(E[mJ) is a finite abelian exten-
sion of L. Let Em C At be the open subgroup corresponding to L(E[mJ);
that is, Em is the subgroup so that the reciprocity map induces an isomor-
phism
AjjBm ~ Gal(L(E[mJ)/L)
x f----+ [x, LlIL(E[TnJ).
Let
Wm = {8 E x
A 8p E R~ and 8p == 1 (mod mRp) for all p},
and let
Urn = Em n {x E At : Nix E W rn }.
We note that Urn is an open subgroup At. We are going to prove the
Claim: aE/L(x) = 1 for all x E Urn.
Let x E UTn, and to ease notation let a = a E / d x). Also fix an analytic
isomorphism
f: Cia ~ E(C)
as in (9.1). Then for any t Em-la/a we have J(t) E E[m], so

f(t) = f(t)[x,L] since x E B m , so [x, L] fixes L(E[mJ)
= f(o:Nix-1 t) from (9.1ii)
= J(o:t) since t Em-la/a and
(Nix)p E (1 + mRp) n R; for all p.
Hence multiplication by 0: fixes m-1a/a or, equivalently,

(0: - l)m- 1 a C a.
This inclusion of fractional ideals means that (0: - I)RK C mRK' so
and
On the other hand, for any prime 13 of K we have
ord p0: = ord p(Nix)p from (9.1i)
=1 since the p-component of Nix E Wm is a unit.
This holds for all 13, so 0: must be a unit, 0: E R'K. But 0: == 1 (mod mRK)
from above, so the only possibility is 0: = 1. This proves our claim.
It follows from the claim and the definition of '¢E/L that
for all x E Um .
From this formula it is clear that '¢E/L is continuous on Um. But Um is
an open subgroup of At. Therefore '¢E/L is continuous on all of Ai" which
completes the proof that '¢E/L is a Gr6ssencharacter.
(b) Let I~r c Gal(Lab / L) be the inertia group for~. The reciprocity map
sends % to I~b,
[R~, L] = I~b,
where we embed R~ into At in the usual way,
R~ ~At, U f----> [ ••• , 1, l,u, 1, 1, ... ].
i
~ - component
Let m be an integer with ~ t m. We know from (2.3) that E[m] C
E(Lab), so I~b will act on E[m]. We want to characterize when this action
is trivial in terms of values of the Gr6ssencharacter '¢E/L. Thus
I~b :~t~t[~iallY ~ f(t)U = f(t) for all (J' E I~b and all t Em-la/a
~ f(t)[x,L] = f(t) for all x E R~ and all t Em-la/a

~ f(O:E/dx)(Nix-l)t) = f(t)
for all x E % and all t Em-la/a,
where for the last equivalence we have used (9.1). We make two observa-
tions. First,
for all x E R~,
since the archimedean components of x E R~ are all 1. Second, multipli-
cation by NJ<x- 1 induces the identity map on m-1a/a. This follows from
Lemma 9.3 (see below) and the assumption that Ifj t m. Hence we find
I<$b acts trivially -¢==} f('l/JE/dx)t) = f(t)

onE[m]
for all x E R~ and all t E m-la/a
-¢==} 'l/JE/dx) == 1 (mod mR K)
for all x E R~, since f : m-1a/a -.:::::. E[m].
Next we apply the criterion of Neron-Ogg-Shafarevich, which relates

the action of I<$b on E[m] to the reduction of E modulo Ifj. More pre-
cisely, [AEC VII.7.1] says that
I<$b acts trivially on E[m] for -¢==} E has good reduction at Ifj.
infinitely many m prime to Ifj
Combining this with the equivalence proved above, we obtain the desired
result:
E has good there are infinitely many m with Ifj t m such that
-¢==}
reduction at Ifj 'l/JE/L(X) == 1 (mod mR K) for all x E R~
-¢==} 'l/JE/L(X) = 1 for all x E R~
-¢==} 'l/J E / L is unramified at Ifj.

o
It remains to prove the elementary result used in the proof of (9.2b).
Lemma 9.3. Let a be a fractional ideal and b an integral ideal of K.
Let s E A'K be an idele with the property that
sp = 1 for all primes p dividing b.
Then the multiplication-by-s map s : K / a -+ K / a induces the identity map

on b-la/a. In other words,
st = t for all t E b-1a/a.
PROOF. From (8.1), the decomposition of b-la/a into p-primary compo-

nents is
b-1a/a ~ EB (b-la/a)p ~ EB bp-lap/ap = EB bp-lap/ap.

p p plb
§1O. The L-Series Attached to a CM Elliptic Curve 171
Here the last equality follows from the fact that bp = Rp for all p t b,
so the only non-zero terms in the direct sum are those with plb. The
multiplication-by-s map on b-1aja is now defined by the commutative
diagram s
-----+
But by assumption, sp 1 for all plb, so multiplication-by-s is just the

identity map. o
§10. The L-Series Attached to a CM Elliptic Curve
The L-series attached to an elliptic curve is an analytic function that is

used to encode arithmetic information about the curve. One then hopes
to deduce further arithmetic properties of the elliptic curve by studying
the analytic properties of its L-series, much as one uses the Riemann zeta
function to study the set of rational primes. In this section we will define
the L-series of an elliptic curve E and show that if E has complex multipli-
cation, then its L-series can be expressed in terms of Hecke L-series with
Gr6ssencharacter.
Let L jQ be a number field, and let E j L be an elliptic curve. For each
prime >,p of L, let
IF\j3 = residue field of L at >,p,
q\j3 = N~>,p = #IF \j3.
If E has good reduction at >,p, we define

a\j3 = q\j3 + 1 - #E(lF\j3),
L\j3(Ej L, T) = 1 - a\j3T + q\j3T2.
The polynomial L\j3( E j L, T) is called the local L-series of E at >,p. If E has
bad reduction at >,p, we define the local L-series according to the following
rules:
1-T if E has split multiplicative reduction at >,p,
{ if E has non-split multiplicative
L\j3(Ej L, T) = : +T
reduction at >,p,
if E has additive reduction at >,po
Remark 10.1. In the case that E has good reduction at Sfj, we can give
a more intrinsic definition of the local L-factor in terms of the action of
Frobenius on the Tate module. Thus let ¢<p : E - t E be the q<p-power
Frobenius map, and let
be the associated map on the Tate module of E (see [AEC III. §7]), where
we take some P relatively prime to the characteristic of IF'<p. If we choose a
basis for the Tate module, so Tp(E) ~ Zp x Zt, then ¢<p,£ is represented by
a 2 x 2 matrix with coefficients in Z£. The characteristic polynomial of the
linear transformation cP<p,£ is
and this polynomial is independent of the chosen basis for Tp(E).

In fact, [AEC V.2.3] says that the characteristic polynomial of ¢<p,p
has coefficients in Z and is independent of P. More precisely, we find that
tr¢<p.f = 1 + deg¢ - deg(l- ¢) from [AEC V.2.3]

= 1 + qv - #E(lF'<p) from [AEC II.2.l1c] and [AEC V §1].
Similarly,
det ¢<P'£ = deg ¢ from [AEC V.2.3]
= qu from [AEC 1I.2.l1c].
Hence
For a general discussion of this material in terms of P-adic cohomology, see

Hartshorne [1, App. CJ.
We now piece together the local L-factors to form the global L-series
of E.
Definition. The (qlobal) L-series of E / L is defined by the Euler product
L(E/L,s) = II L<p(E/L,q;p8)-1,
<p
where the product is over all primes of L.
Using the estimate la<p1 :S 2VQ\i3 from [AEC V.2.4], it is not hard to
show that the product converges and gives an analytic function for all s
satisfying Re( s) > ~. Conjecturally, far more is true.
§1O. The L-Series Attached to a CM Elliptic Curve 173
Conjecture 10.2. Let ElL be an elliptic curve defined over a num-

ber field. The L-series of ElL has an analytic continuation to the entire
complex plane and satisfies a functional equation relating its values at s
and 2 - s.
We are going to verify this conjecture for elliptic curves with complex
multiplication by showing that L( ElL, s) is a product of Hecke L-series
with Grossencharacter. In general, suppose that
'IjJ: Ai - t C*
is a Grossencharacter on Lj that is, 'IjJ is a continuous homomorphism which
is trivial on L * . Let ~ be a prime of L at which 'IjJ is unramified, so
'IjJ(R;JJ) = 1. We then define 'IjJ(~) to be
'IjJ(~) = 'IjJ( ... , 1, 1,71", 1, 1, ... ),

i
~ - component
where 71" is a uniformizer at ~. Note that since 'IjJ is unramified at ~, 'IjJ(~)

is well-defined independent of the choice of 71". For convenience, we also set
'IjJ(~) =0 if'IjJ is ramified at ~.
Definition. The Hecke L-series attached to the Grossenchamcter
'IjJ: Ai - t C*
is defined by the Euler prod uct
L(s,'IjJ) = II(1- 'IjJ(~)q;PSrl,

'.J.l
where the product is over all primes of L.
Heeke L-series with Grossencharacter have the following important
properties, whose proof we will omit.
Theorem 10.3. (Heeke) Let L(s, 'IjJ) be the Hecke L-series attached to
the Grossencharacter 'IjJ. Then L( s, 'IjJ) has an analytic continuation to the
entire complex plane. Further, there is a functional equation relating the
values of L(s, 'IjJ) and L(N - s,;j;) for some real number N = N('IjJ).
PROOF. This was originally proven by Hecke. It was reformulated and

reproven by Tate [8] using Fourier analysis on the adele ring A L . 0
The key to expressing L( ElL, s) in terms of Heeke L-series is to express

the number of points in E(IF'.J.l) in terms of the Grossencharacter attached
to ElL.
Proposition 10.4. Let ElL be an elliptic curve with complex multipli-

cation by the ring of integers RK of K, and assume that L ::J K. Let \13
be a prime of L at which E has good reduction, let E be the reduction
of E modulo \13, and let ¢'lJ : E -> E be the associated q'lJ-power Frobenius
map. Finally, let 'l/JE/L : At -> C* be the Grossencharacter (9.2) attached
to ElL. Then the following diagram commutes:
[..pE/d'lJ)]
E • E
1 <P'P
1
------+. E,
where the vertical maps are reduction modulo \13.
PROOF. Before we begin the proof, two remarks are in order. First, 'l/JE/L
is unramified at \13 from (9.2b), so 'l/JEI d\13) is well-defined. Second, since
'l/JE/d\13) is the value of 'l/JE/L at an idele with l's in its archimedean com-
ponents, we have 'l/JE/d\13) = D:E/d\13) E R K , so it makes sense to talk
about ['l/JE/d\13)] as an endomorphism of E.
Let x E At be an idele with a uniformizer in its \13-component and l's
elsewhere. Then as we just remarked,
The commutative diagram (9.1) used to define D:E/L tells us that
for all t E Kia.
Fix some integer m with \13 f m. Then (9.3) says that Nix-It = t for
all t E m-In/n, so we get
f(t)[x,L] = ['l/JE/dx)]f(t) for all t E m-1ala.
Now consider what happens when we reduce modulo \13. We have [x, L] =
(\13,p b IL) from (3.5), so [x,L] reduces to the q'lJ-power Frobenius map.
Hence
for all t E m-1nla.
Since this is true for all m prime to \13, and since an endomorphism of E
is determined by its effect on torsion (or even on e-primary torsion for a
fixed prime e [AEC 111.7.4]), we conclude that
o
Corollary 10.4.1. With notation as in (10.4), we have
(a) q'll = N~q3 = N{f (1/JE/dq3)) ,
(b) #E(F'll) = N~1.l3 + 1 -1/JE/L(I.l3) -1/JE/dq3),
(c) a'll = 1/JE/dq3) + 1/JE/dq3)·

(The bar indicates complex conjugation of elements of K.)
PROOF. (a) We compute
N~q3 = deg</>'ll from [AEC II.2.11c]
=
-------
deg [1/JE/dl.l3)] from (10.4)
= deg[1/JE/dq3)] from (4.4)
= N{f (1/JE/L(I.l3)) from (1.5).
(b) Similarly, we compute
#E(F'll) = # ker(l - </>'ll)

= deg(l - </>'ll) from [AEC III.5.5]
and [AEC 1Il.4.IOc]
= deg [1 -1/JE/dq3)] from (10.4)

= deg[l -1/JE/dl.l3)] from (4.4)
= N{f (1 -1/JE/L(q3)) from (1.5)
= (I-1/JE/dq3))(1 -1/JE/dl.l3))
= 1 -1/JE/dq3) -1/JE/dl.l3) + N~1.l3 from (a).
(c) Immediate from (a), (b), and the definition of a'll' 0
We now have all of the tools needed to relate the L-series of E to

the L-series attached to its Grossencharacter, at least in the case that the
field of definition of E contains the CM field. We will leave the proof in
the other case to the reader.
Theorem 10.5. (Deuring) Let E / L be an elliptic curve with complex
multiplication by the ring of integers RK of K.
(a) Assume that K is contained in L. Let 1/JE/L : AI, ---+ C* be the Gros-
sencharacter (9.2) attached to E / L. Then
L(E/L,s) = L(S,1/JE/dL(s,1/JE/d.
(b) Suppose that K is not contained in L, and let L' = LK. Further
let'l/JEILI : ALI - t C* be the Gr8ssencharacter attached to E/L'. Then
L(E/L,s) = L(S,'l/JElu).
Using Heeke's theorem (10.3), we immediately deduce that the L-series

of a CM elliptic curve has an analytic continuation and satisfies a functional
equation. A more careful analysis yields the following result. We will leave
the proof to the reader.
Corollary 10.5.1. Let E/L be an elliptic curve with complex multiplica-
tion by the ring of integers RK of K. The L-series of E admits an analytic
continuation to the entire complex plane and satisfies a functional equation
relating its values at s and 2 - s.
More precisely, define a function A( E / L, s) as follows:
(i) If K c L, let
A(E/L,s) = (N~(1)LIQC",)r(27r)-sr(s))[L:Q]L(E/L,s),
where c", is the conductor of the Grossencharacter 'l/JEIL, 1)L/Q is the dif-
10
ferent of L/Q, and res) = 00 tS-1e- t dt is the usual r-function.
(ii) If K ct
L, let L' = LK and
A(E/ L, s) = (N~' (1) U IQC~))sI2 (27r)-Sr(s)) [L:Q] L(E/ L, s),
where c~ is the conductor of the Gr8ssencharacter 'l/J E I U·

Then A satisfies the functional equation
A(E/L,s) = wA(E/L,2 - s),
where the quantity w = WElL E {±1} is called the sign of the functional
equation of ElL.
PROOF (of Theorem 10.5). We know from (6.1) and [AEC VII.5.5] that E
has potential good reduction at every prime of L, so [AEC VII.5.4(b)] tells
us that E has no multiplicative reduction. Hence
L (E/L T) = {1- a<.pT + q<.pT2 if E has good reduction at \13,

<.p , 1 if E has bad reduction at \13.
Now suppose E has good reduction at \13. Then
L<.p(E I L, T) = 1- a<.pT + q<.pT2 by definition of L<.p,

= 1 - ('l/JE/d\13) + 'l/JE/L(\13))T + (N[f'l/JEILC\13))T2
from (10.4.1)
On the other hand, (9.2b) says that 'l/JE/L is unramified at $ if and only
if E has good reduction at $, and the same is true for 'l/JE/L. Thus
if E has bad reduction at $,
so the formula given above for L'i3(E / L, T) is also true for primes of bad
reduction, since it reduces to L'i3(E / L, T) = l. Therefore
L(E/L,s) = IT L'i3(E/L,q;i/)-l
'i3
= IT(1- 'l/JE/d$)q$S)-l(l- 'l/JE/d$)q$S)-l
'i3
= L(s, 'l/JE/dL(s, 'l/JE/d.
(b) See exercises 2.30, 2.31, and 2.32. D
Example 10.6. Let D E Z be a non-zero integer, and let E be the elliptic

curve
E: y2 = x 3 + D
having complex multiplication by the ring of integers RK of the field K =

Q (A). Let p be a prime of RK with P t 6D. Since RK is a PID, we
can write p = (7r), and one can check that there is a unique 7r generating p
which satisfies 7r == 2 (mod 3). It is then a moderately difficult exercise
using Jacobi sums (see Ireland-Rosen [1, 18 §§5,7]) to show that
where (:;)6 is the 6th -power residue symbol; that is, (:;)6 is the 6th -root of
unity satisfying
a(N{iP-l)/6 == (;)6 (mod 7r).
Using (10.4.1), we see that the Grossencharacter attached to E is given
either by
or else by
To determine which one it is, we use (5.4) to find a root of unity ~ E

Rk such that the reduction of [~7rl modulo p is Nlj p-power Frobenius.
Note that (5.4) says this is possible for almost all degree 1 primes of K.
But (10.4) says that [1PEjK(P)] also reduces to Frobenius. We conclude

that
where P = (7r) and 7r == 2 (mod 3),
at least for almost all degree 1 primes p of K. By the continuity of 1P and

the reciprocity law for (;i:-)6' we see that this formula holds for all p.
Using (10.5) and Ntj7r = 7rn, we find that the L-series of E over K
and over Ql can be written explicitly using residue symbols as
L(E/K,s) = II
7rE RK prime
(1+ C:),K'-"-')-'
,,=2 (mod 3)
X (1 + (4;)6 7r- S n1 - S ) -1 ,
L(E/Ql, s) = II
7rERK prime
(1+ (4:)/,._,)-'
7r=2 (mod 3)
EXERCISES
2.1. Let K/Q be a quadratic field with ring of integers RK, and let R c K be
an order in K. Prove that there is a unique integer f 2: 1 such that
R = Z+ f· RK.
The integer f is called conductor of the order R.

2.2. 'Let A = Z[i] be the lattice of Gaussian integers, and let A be the elliptic
1
integral
1 dt
A- ---
- o~·
(a) Prove that
(b) More generally, prove that for all integers n 2: 1, G 4n (A) is a rational
number multiplied by A4n .
2.3. Let K/Q be a quadratic imaginary field, and let E/C be an elliptic curve
with End(E) @ Q ~ K. Let E' /C be another elliptic curve. Prove that E'
is isogenous to E if and only if End(E') @ Q ~ K.
Exercises 179
2.4. Let E be an elliptic curve defined over a number field L with complex
multiplication by K, and let '+l be a prime of L of characteristic p at which E
has good ordinary reduction. Prove that Qp contains a subfield isomorphic
to K.
2.5. Let EjQ be an elliptic curve with complex multiplication by the ring of
integers in Q (~). 'Without using an explicit Weierstrass equation for E,
prove the following two facts:
(a) If cjJ : E -+ E is an endomorphism of degree 2 and P is the non-zero
point in the kernel of ¢, then P rt E(Q) and PEE (Q (~)).
(b) R(Q) contains exactly one point of order 2.
(Hint. Use (2.2a).)
2.6. (a) Let F be a field, let G be a subgroup of GL 2 (F), and let C(G) be the
centralizer of G; that is,
C(G) = {ex E GL 2 (F) : ex, = ,cx for all, E G}.
Prove that one of the following two conditions is true.
(i) C(G)={(~~) :CEF*}.

(ii) G is abelian.
(Hint. If C(G) contains a non-scalar matrix ex, make a change of basis to
put cx into Jordan normal form and then calculate C(cx).)
(b) Let L be a perfect field, let E j L be an elliptic curve, and let £ be a
prime with £ # char(L). Suppose that EnddE) is strictly larger than Z.
Use (a) to prove that the action of Gal(Lj L) on the Tate module T£(E) is
abelian.
2.7. The following fact (2.5.1) from commutative algebra was used in the proof
of Proposition 2.4. Let R be a Dedekind domain, let a be a fractional ideal
of R, and let it! be a torsion-free R-module. Prove that the natural map
q; : a-lit! HomR(a, it!)
x t---------+ (¢;.r:OI------4Qx)
is an isomorphism. (Hint. Prove that it is an isomorphism after you localize
at any prime p of R. Note that the localization Rp is a principal ideal
domain.)
2.8. Let Lj K be a finite abelian extension of number fields.
(a) Let Q{ be a non-zero fractional ideal of L. Prove that
(N~Q{, Lj K) = 1.
(b) Prove that an unramified prime p of K splits completely in L if and

only if (p, Lj K) = 1.
2.9. Let a be a fractional ideal of K, and let E be an elliptic curve corresponding
to the lattice a.
(a) Prove that j(E) E lR. if and only if (j2 = 1 in eqRK).
(b) Prove that the following are equivalent:
(i) Q(j(E)) is Galois over Q.
(ii) Q(j(E)) is totally real.
(iii) Every element of eqR K ) has order 2.
2.10. Let K be a number field, RK its ring of integers, and c an integral ideal
of RK. Prove that there is an exact sequence
R'K --+ (RK/C)* -> I(c)/P(c) -> e.c(RK) -> 1.

o(modc)....... (0)
2.11. Let K= Q (vI-15), and let RK be the ring of integers of K.

(a) Prove that e.c(RK) = Z/2Z.
(b) Let L = Q (A, v'5). Prove that L/K is everywhere unramified, and
deduce that L is the Hilbert class field H of K.
(c) Let K = Q ( vi - 23). Prove that the Hilbert class field of K is given
by H = K(o), where 0 satisfies 0 3 - 0 - 1 = O.
2.12. Let E/f:- be an elliptic curve such that End(E) is an order in the quadratic
imaginary field K.
(a) Show that there exists an elliptic curve E'If:- and an isogeny 4> : E -> E'
such that End(E') = RK.
(b) "If E is defined over the field L, prove that it is possible to choose E'
and 4> in (a) so that both are defined over L.
2.13. Let K = Q(i) and let KN be the ray class field of K modulo N.
(a) Prove that
(b) Verify directly that the field
is an abelian extension of K, and compute its Galois group over K.

(This exercise verifies some of the statements made in (4.9.1).)
2.14. Let E be the elliptic curve y2 = x 3 + 1, and let K = Q (A). For each
integer N 2': 1, let KN = K(h(E[NJ») and LN = K(E[nJ).
(a) Calculate K2, K3, and K4 explicitly and in each case verify that KN
is the ray class field of K modulo N.
(b) Calculate L2, L 3 , and L4 explicitly and show that they are abelian
extensions of K.
2.15. Let E be the elliptic curve E : y2 = x 3 + 4x 2 + 2x, and let K = Q (A).
From (2.3), this curve has complex multiplication by K. Redo the previous
exercise for this curve E and field K.
'.
Exercises 181
2.16. For each of the quadratic imaginary fields in the following table, verify that
the given a generates the Hilbert class field of K, and calculate the value
of j (RK) explicitly as an element of K (a). (We have filled in the first row
for you, see Example 6.2.2.)
Disc K/rQ hK a a root of j(RK)
(a) -15 2 x2 - 5 -52515 - 85995¥

(b) --20 2 x2 +1
(c) -23 3 x3 - x-I
(d) -24 2 x 2 +3
(e) -31 3 x3 + x-I
2.17. Let 'D~ and S~ be the sets of matrices defined by
'D~ = { (~ ~) ENh(Z) : ad - bc = n, gcd(a, b, c, d) = I} ,

S~ = { (~ ~) E 'D~ : d> 0, 0::; b< d} .
(a) Prove that the natural map S~ --> SL2(Z)\'D~ is a bijection.
(b) Prove that
#S~ = n II (1 + ~ ) .
pin
(Notice that if n is squarefree, then 'D~ and S~ are just the sets 'Dn and Sn
considered in §6.)
2.18. Let S~ be as in the previous exercise, and define
<pn(X) = II (X - j 0 a).
<Pn is called the modular polynomial of order n.

(a) Prove that <Pn E Z[jJ[X]. We will write <Pn(j, X) to indicate that <Pn
is a polynomial in two variables.
(b) Prove that <Pn is irreducible over <C(j).
(c) Prove that <pn(Y, X) = <Pn(X, V).
(d) Prove that if n is not a perfect square, then <pn(X, X) is a non-constant
polynomial with leading coefficient ±l.
(e) Let Fn(j,X) be the polynomial from (6.3). Prove that
Fn(j,X) = II <P nld 2 (j,X).
d 2 1n,d21
In particular, if n is square-free, then Fn = <Pn.

(f) 'Let I<pnl denote the magnitude of the largest coefficient of <pn(Y, X).
Prove that
lim log l<Pn I = 6
n_= (deg<pn)(logn) .
2.19. Let Fn(Y,X) be the polynomial from (6.3), and let <l>n(Y,X) be the poly-
nomial from the previous exercise. Let El/C and E 2 /C be elliptic curves.
(a) Prove that F n (j(E 1 ),j(E2 )) = 0 if and only if there is an isogeny
El --> E2 of degree n.
(b) Prove that <l>n (j (Ed, j (E2)) = 0 if and only if there is an isogeny
El --> E2 whose kernel is cyclic of degree n.
2.20. Let p be a prime. Prove Kronecker's congruence relation
Fp(Y, X) == (X - yP)(X P - Y) (mod pZ[X, Yj),
where Fp is the polynomial defined in (6.3).

2.21. Let J(T) be a modular function of weight 0 that is holomorphic on Hand
that has the q-expansion J = L: anqn. Let R be a ring containing all of
the an's. Prove that J E R[j], where j = jeT) is the modular j-function.
This strengthens (I.4.2b), which says that J E C[j).
2.22. Let f 2': 3 be a prime, let
and let M2 (fZe) be the additive group of 2 x 2 matrices with fZe-coefficients.

Prove that the map
1 + fA f---7 log(1 + fA) = L - l)n+lf

<Xl (
n
nA n
'
n=l
is a well-defined isomorphism.
2.23. This exercise generalizes the previous one. Let L be a finite extension of Qe,
let R be the ring of integers of L, and let wt be the maximal ideal of R.
For each integer r 2': 1 define a subgroup G r of GLn(R) by
G r = {A E GLn(R) : A == In (mod wt r )},
where In is the n x n identity matrix.

(a) Prove that for every r 2': 1, the quotient Gr/Gr+ 1 is a finite group
whose order is a power of f.
(b) Prove that
and deduce that G 1 is a pro-f group.

(c) Prove that if r is sufficiently large, then G r is isomorphic to the additive
group of n x n matrices with coefficients in R. (Hint. For the case n = 1,
see [AEC IV.6.4b].)
Exercises 183
2.24. Let E and E' be elliptic curves given by Weierstrass equations
and let cp : E --t E' be a non-zero isogeny.

(a) Prove that there is a rational function R(x) and a constant c such
that cp has the form
cp(x, y) = ( R(x), cy d~~X)) .

(Hint. Look at the invariant differentials.)
(b) Prove that there is a commutative diagram
CIA ----> CIA'

Zl-+c- 1 z
1 1
E(C)
'"
----> E'(C)
where the vertical maps are complex analytic isomorphisms and c is the
constant from part (a).
2.25. Let ElL be an elliptic curve defined over a number field L with com-
plex multiplication by the ring of integers of K, and assume that K c L.
Let X E H1(GL/L,Aut(E)), and let EXIL be the corresponding twist
of E. (See [AEC X §5] for basic facts about twists.) If we identify Aut(E)
with p,n [AEC III.1O.2], then X gives a homomorphism
and we can extend X to a homomorphism on the ideles by the rule
X : Ai- ----> C' , x(x) = X([x, L]).

Prove that
2.26. Let ElL be an elliptic curve defined over a number field (not necessar-
ily with complex multiplication). Prove that the infinite product defin-
ing the L-series L( ElL, s) converges absolutely and uniformly in the half-
plane Re(s) > ~.
2.27. Let ElL be an elliptic curve defined over a number field (not necessarily
with complex multiplication), let \P be a prime of L at which E has good
reduction, and let IF<,p be the residue field at \p. For each integer n ~ 1,
let IF<,p,n be the extension of IF<,p of degree n. Recall [AEC V §2] that the
zeta function of EIlF<,p is the formal power series
Z(E/lF<,p,T) = exp(f:#E(lF<,p,n)~).
n=l
(a) Prove that
Z(EIF T) = £.:p(EIF'i), T)
'i), (1 - T)(l - q'i)T) ,
where L'i) is the local L-series described in §1O and q'i) = N~\p.
(b) The zeta function of the field L is given by the usual Euler product
(L(S) = 11(1- q-il )-I, 8
'i)
and the (global) zeta function of ElL is defined by the product
((ElL,s) = 11 Z(EIF'i),q-il 8
).
'i)
Find the "correct" definition for the factor Z(EIF'i) , T) in the case that E
has bad reduction at \p, and prove that
((ElL,s) = (ds)(ds - l)L(EIL,s)-I.
2.28. With notation as in §1O, prove that
ll''i),q'i)-1) -_ #En.(F'll) .
L 'i) (E-/""
q'i)
Here En. is the non-singular part of E. Note that we do not assume E has
good reduction at \p. (See [AEC III §2] and [AEC exer. 3.5].)
2.29. Prove the functional equation (10.5.1) for the L-series of an elliptic curve
with complex multiplication. (Hint. Use the functional equation for Hecke
L-series with Grossencharacter as described, for example, in Tate [8].)
In the next three exercises we sketch the proof of Theorem 1O.5(b). We
set the following notation: Let ElL be an elliptic curve with complex
multiplication by the ring of integers RK of K, and assume that L does
not contain K. Let L' = LK, so L' is a quadratic extension of L, and let \p
be a prime of L. From (9.2) there is a Grossencharacter 'l/JE/LI : ALI --+ C'.
Let q'i), a'i),'" be the quantities described in §1O.
2.30. Assume that E has good reduction at \p.
(a) Prove that \p is unramified in L'.
(b) Suppose \p splits in L' as \pRu = \p'\p". Prove that
and
(c) Suppose \p remains inert in L', say \pRu = \p'. Prove that
a'll = 0,
(d) Let E be the reduction of E modulo \p, and let p be the residue
characteristic of \p. Prove that
-. {ordinary if \p splits in L' and p splits in K,
E IS
supersingular if \p is inert in L' and p does not split in K.
Exercises 185
2.31. We continue with the notation and assumptions from above. Let \13' be a
prime of L' lying over \13.
(a) If \13 ramifies in L', prove that E has bad reduction at \13.
(b) If \13 is unramified in L', prove that
E has good reduction at \13 ¢=> E has good reduction at \13'.

2.32. We continue with the notation and assumptions from above.
(a) Prove that the local L-series of Eat \13 is given by
(1- V;E/LI(\13')T)(l - V;E/u(\13")T)

{ if \13Ru = \13'\13" splits in L',
Lrv(E / L, T) = 1 - ./. (m')T'f
'l'E/L' .,.,
mR
1.,., L' =.,.,m'"IS mert m. L' ,
1 if \13Ru = \13,2 ramifies in L'.

(b) Prove that the global L-series of E / L is given by
L(E/L,s) = L(S,V;E/L')'
2.33. Fix a non-zero integer D E IZ and let E be the elliptic curve
E : y2 = x 3 - Dx.
Let p E IZ be a prime with p f 2D.

(a) If p == 3 (mod 4), prove that
and
(b) If p == 1 (mod 4), factor p in lZ[i] as
with 7r == 1 (mod 2 + 2i).

Prove that
where (;:) 4 is the 4th -power residue symbol.

2.34. Continuing with the notation from the previous exercise, let p C lZ[i] be a
prime ideal with P f 2D. Write
P= (7r) for an element 7r E lZ[i] satisfying 7r == 1 (mod 2 + 2i).

Prove that the Grossencharacter associated to E /Q( i) is given explicitly
by the formula
V;E/Q(i)(P) = (~) 4 7r ·
Here V;E/Q(i)(P) equals the value of V; at an idele with a uniformizer at

the p-component and 1 '6 elsewhere.
2.35. Let E / L be an elliptic curve with complex multiplication by the ring of

integers RK of K, and assume that K C L. Let I' be a degree 1 prime
of K, let p = N[P"p, and let Rp be the completion of RK at p. Notice
that Rp ~ Zp.
Fix an integer h ~ 1 so that ph is principal, say ph = 7r RK. We make
the collection of groups E[pn], n = 1,2, ... , into an inverse system using
the maps
n = 1,2, ....
The p-adic Tate module of E is then defined to be the inverse limit
(a) Prove that Tp(E) is a free Rp-module of rank 1. Deduce that Aut Tp(E)
is isomorphic to R; ~ Z;.
(b) Let Lp be the compositum of the fields L(E[pnJ) for all n ~ lor,
equivalently, the field defined by
Gal(L/Lp) = ker{Gal(L/L) ---> Aut Tp(E)}.
Prove that there is a finite extension L' / L contained in Lp such that
(c) Let r = Zp, and define a ring Zp[q to be the inverse limit
(Note that Zp[q is not the same as the group ring Zp[r].) Prove that
and hence that Tp(E) is a Zp[q-module.
Let , E r be a topological generator, send ,n

(d) Prove that Zp[q is isomorphic to the power series ring Zp[T]. (Hint.
to (1 + T) n, and show that
this extends to an isomorphism.) Zp[q-modules are often called Iwasawa
modules.
(e) Let Lp be the field of definition of Tp(E) = lim E[pn]. Prove that there
is a finite extension L" / L contained in Lp such that
(Hint. Write pRK = PI". Show that Lp = LpLp/, and that Lp n Lpl is a
finite extension of L.)
CHAPTER III
Elliptic Surfaces
Elliptic surfaces appear in many guises. They are one-parameter algebraic

families of elliptic curves, they are algebraic surfaces containing a pencil of
elliptic curves, and they are elliptic curves over one-dimensional function
fields. In this chapter we will see elliptic surfaces arising in all of these
ways. Since our emphasis in this book is primarily on arithmetic questions,
we will concentrate on those properties of elliptic surfaces which resemble
the arithmetic properties of elliptic curves defined over number fields. This
means we will neglect many of the fascinating geometric questions raised by
the study of elliptic surfaces over algebraically closed fields, especially the
classical theory of elliptic surfaces defined over C. The interested reader will
find a nice introduction to this material in Beauville [1], Griffiths-Harris [1,
Ch. 4, §5] and Miranda [1].
We will also find it necessary to restrict attention to fields of character-
istic zero. We do this in order to apply the results from [AEC] , especially
Chapters I, II, and III, to elliptic curves defined over the function field k( C)
of a projective curve C /k. All of the main theorems in [AEC] were proven
for elliptic curves E / K under the assumption that the field K is perfect; if k
has characteristic p > 0, then the field k(C) will certainly not be perfect.
However, we must in all honesty point out that elliptic curves defined
over non-perfect fields such as IFq(T) are also extensively studied. (Equiv-
alently, one would study elliptic surfaces £ ...... pI defined over IF q.) In fact,
since the rings IF q [T] and Z share the property that all of their residue
fields are finite, elliptic curves defined over IFq (T) behave arithmetically
very much like elliptic curves defined over Q. Thus conjectures about el-
liptic curves defined over Q are often first tested and proven in the easier
setting of elliptic curves over IFq(T). Notice that k(T) does not have this
property if char(k) = 0, so we will find that the theory of elliptic surfaces
in characteristic zero differs in some respects from the theory of elliptic
curves over number fields.
The main results proven in this chapter are the Mordell-Weil theo-
rem for elliptic surfaces (6.1), two constructions of the canonical height
pairing (4.3, 9.3), and specialization theorems for the canonical height
188 III. Elliptic Surfaces
(11.1, 11.3.1) and for the homomorphism from sections to points on fibers
(11.4). Unfortunately, it will not be possible to prove all of the background
results we need from algebraic geometry. However, we will give a pre-
cise statement of the results we use and give at least some indication of
the proofs. Briefly, we will use abelian varieties and Jacobian varieties in
§2, rational maps between varieties in §3, intersection theory and minimal
models of surfaces in §§7 and 8, and divisors on varieties in §10. Much of
the material we need is contained in Hartshorne [1], but in any case, we
will give references for all assertions that we do not prove.
§1. Elliptic Curves over Function Fields
One way to define an elliptic surface is as a one-parameter algebraic family

of elliptic curves. In this guise we have already seen numerous examples
of elliptic surfaces. For example, during the proof of [AEC III.1.4(c)] we
wrote down the elliptic curve
E' 2 X _ 3 _ 36 x _ 1
.Y + Y- x jo - 1728 jo - 1728
with j-invariant jo. In reality, E is a family of elliptic curves, one for
each choice of the parameter jo (except that E is singular or non-existent
for jo = 0 and jo = 1728). Similarly, in [AEC IX §7] and [AEC X §6] we
looked at the elliptic curves
and
for varying values of D. Again these are families of elliptic curves, in this
case parametrized by D, and each value of D other than D = 0 gives an
elliptic curve.
More generally, if k is any field (of characteristic not equal to 2) and
if A(T), B(T) E k(T) are rational functions of the parameter T, then we
can look at the family of elliptic curves
ET : y2 = x3 + A(T)x + B(T).
For most values of t E k we can substitute T =t to get an elliptic curve
Et : y2 = x3 + A(t)x + B(t).
Precisely, E t will be an elliptic curve provided
A(t) =I 00, B(t) =I 00, and ~(t) = -16(4A(t)3 + 27B(t)2) =I O.
Later in this chapter we will pursue further this idea of an algebraic
family of elliptic curves. But for now we want to alter our perspective a
bit. Rather than considering the equation
E : y2 = x 3 + A(T)x + B(T)
§l. Elliptic Curves over Function Fields 189
as defining a family of elliptic curves, one for each value of T, we will

instead view E as a single elliptic curve defined over the field k(T). As
long as the discriminant
f}.(T) = -16(4A(T)3 + 27B(T)2) =1= 0 in k(T),
E will be an elliptic curve defined over the field k(T). So we will be able
to apply much of the general theory developed in [AEC] to the elliptic
curve E/k(T), at least provided that the field k(T) is perfect. For this
reason we will henceforth make the assumption:
Ik is a field of characteristic zero. [
(For further comments about this assumption, see the introduction to this
chapter.)
Example 1.1.1. Consider the elliptic curve E/Q(T) given by the Weier-
strass equation
with discriminant
f}. = 16T4 (4T 2 - 27).
This curve has the rational point
P= (T,T) E E(Q(T)),
and one can easily use the addition law to compute
2P = (T2 - 2T, _T3 + 3T 2 - T),
3P = (T 3 - 2T2 - 3T +
(T - 3 ) 2 '
3T4 4 - 15T3 + 21T2 - 9T +
(T - 3)3
8) .
If we substitute T = t for some t E Q, then we will obtain an elliptic
curve E t unless t = 0 or t = ±~V3.
Example 1.1.2. The elliptic curve
E : y2 = x 3 + (T2 - l)x + T2
has many rational points defined over Q(T), such as the point (-1,0) of
order 2 and the point (0, T) of infinite order. However, if we replace Q
by Q ( V2), we find a new point of infinite order:
(1, V2T) E E(Q(vI2) (T)).
In general, if E/Q(T) is a non-constant elliptic curve (i.e., j(E) (j. Q), then
there exists a finite extension k/Q so that
E(k(T)) = E(CCT)).
In practice it is often difficult to find k. See Kuwata [1,2J and exercise 3.17.
Shioda [lJ has constructed an interesting example for which the action
of Gal(k/Q) on E(k(T)) is a representation of type E 8 , so [k : QJ may be
quite large.
We have been taking the coefficients A and B of the elliptic curve
to lie in the field ofrational functions k(T). This is unnecessarily restrictive.

We observe that k(T) is the function field of the projective line pI, so we
might consider choosing A and B from the function field of some other
curve. Thus we can take a non-singular projective curve C / k and look at
elliptic curves E defined over the field k(C).
Example 1.1.3. Let C/Q be the (elliptic) curve
that is, C is the projective curve corresponding to this affine equation.

Then the equation
E : y2 + (st + t - s2)xy + s(s - 1)(s - t)t2y = x 3 + s(s - 1)(s - t)tx 2
defines an elliptic curve E over the function field Q( C) of C. Notice that E

contains the rational point
P = (0,0) E E(Q(C)).
It is not hard to verify (at least if you have access to a computer with a
symbolic processor) that
[l1lP = 0,
so P is a point of order 11.

In fact, E is in some sense the universal family of elliptic curves con-
taining a point of order 11. This means the following: Let A be any
elliptic curve and Q E A a point of order 11. Then there is a unique
point (so, to) E C such that if we substitute (s, t) = (so, to) into the equa-
tions for E and P, we will obtain an elliptic curve Eo and a point Po E Eo
of order 11 such that there is an isomorphism ¢ : A ---> Eo with ¢(Q) = Po·
In the literature, the curve C is called the modular curve Xl (11), and E is
usually denoted EI(11). For further details, see exercise 3.2.
§2. The Weak Mordell-Weil Theorem 191
§2. The Weak Mordell-Weil Theorem
Our task in this section is to prove the following weak Mordell-Weil theorem
for elliptic curves defined over function fields. We emphasize again our
assumption that the constant field k always has characteristic O.
Theorem 2.1. (Weak Mordell-Weil Theorem) Let k be an algebraically

closed field, let K = k (C) be the function field of a curve, and let E / K be
an elliptic curve. Then the quotient group E(K)/2E(K) is finite.
The theory of function fields of curves is analogous to the theory of

number fields, and the proof of the weak Mordell-Weil theorem is similar
in both cases. In other words, we could save space here by merely quoting
the proof given in [AEC, VIII §1] with the words "number field" replaced
by the words "function field." However, there are enough differences that
we feel it is worthwhile giving the proof. We will place our main emphasis
on highlighting the differences between the two cases.
Recall that the proof for number fields has two main steps. The first
step [AEC, VIII.1.5] depends on properties of the elliptic curve E / K. It
says that the extension field L = K([m]-lE(K)) is an abelian extension
of K, has exponent m, and is unramified outside a certain finite set of
primes S. This step carries over word-for-word to the function field case
once one has developed the theory of valuations. We will give a slight
variant of the argument in [AEC, VIII §1]' using divisors supported on a
finite set of points, but the reader will have no trouble seeing that this is
merely a matter of using geometric language to deal with the same ideas.
We will also consider only the case m = 2, since this allows us to work
more directly with the equation for E/ K.
The second step [AEC, VIII. 1.6] has nothing to do with elliptic curves.
Instead, one uses Kummer theory to show that the maximal abelian ex-
tension of K of exponent m unramified outside of S is a finite extension.
The proof of this proposition is not hard, but it ultimately relies upon the
two fundamental finiteness theorems of algebraic number theory, namely
the finiteness of the class group and the finite generation of the unit group.
In general, neither of these last two results is true for function fields.
For example, if K = k( C) is a function field, then the "unit group"
in K* is the constant field k*. To see that this is the right analogy, note
that for a number field K, the unit group can be described as the set of
elements 0: E K* satisfying v(o:) = 0 for all discrete valuations on K*.
But the discrete valuations on a function field K = k( C) correspond to the
points of C(k) (at least if k is algebraically closed). Thus if a function f E K
has valuation 0 for all valuations, then it has no zeros or poles on C, so
by [AEC, 11.1.2] it is a constant. Hence the "unit group" of K will be k*,
and in general k* will not be finitely generated.
Similarly, we will see during the proof of the weak Mordell-Weil theo-
rem that the "ideal class group" of a function field K = k( C) is the Picard
group Pic( C), that is, the group of divisors modulo linear equivalence. The
Picard group need not be finitely generated; see (2.6) below.
However, all is not lost. A closer examination of [AEC, VIII.l.6] shows
that it does not require the full strength of the finiteness theorems. In-
stead we used the facts that the ideal class group has only finitely many
elements of order m and the unit group R* has the property that the quo-
tient R* / R*m is finite. These weaker results are true for function fields
under appropriate assumptions on the constant field k of K. For exam-
ple, if k is algebraically closed, then k* / k*m is certainly finite, since it is
actually trivial.
Similarly, the Picard group Pic( C) has only finitely many elements of
order m. Unfortunately, the proof of this last statement requires results
from the theory of Jacobians and abelian varieties which we will not be
able to develop in full. So we will just state here the proposition that we
need and postpone until the end of the section a sketch of the proof. For
the proof of the weak Mordell-Weil theorem (2.1), we will only need to use
the m = 2 case of the following proposition.
Proposition 2.2. Let C be a non-singular projective curve defined over
an algebraically closed field k. Then for any integer m 2 1, the Picard
group Pic( C) has only finitely many elements of order m.
PROOF. See (2.7) at the end of this section for a complete description of
the torsion subgroup of Pic(C). In particular, if C has genus g, then (2.7)
implies that Pic(C)[m] is isomorphic to (Z/mZ)29. 0
PROOF (of Theorem 2.1). Our first observation is that if L / K is a fi-

nite Galois extension and if we can prove that E(L)/2E(L) is finite, then
it will follow that E(K)/2E(K) is also finite. This is the content of
[AEC, VIII.l.l.1], and we gave another proof using Galois cohomology
in [AEC, VIII §2]. We leave it to the reader to verify that these proofs
made no use of the assumption that the field K is a number field, so they
are also valid in our case. Note that [AEC, II.2.5] ensures that any such L
will be the function field of a curve over k.
Replacing K by a finite extension and C with the corresponding curve,
it thus suffices to prove (2.1) under the assumption that E(K) contains
all of the points of order 2. Equivalently, we may assume that E has a
Weierstrass equation of the form
Consider the map
¢ : E(K)/2E(K) --> (K* / K*2) x (K* / K*2)

defined by
ifx:/=el,e2,
ifx=el,
if x = e2,
if x = 00 (P = 0).
In the case that K is a number field, we proved in [AEC, X.1.4] that ¢

is an injective homomorphism, and the same proof works for an arbitrary
field K.
The map ¢ can also be defined using group cohomology. We briefly
sketch the proof. Taking G K / K-cohomology of the exact sequence
- [2J -
o ----+ E[2] ----+ E(K) ----+ E(K) ----+ 0
gives an exact sequence
Now our assumption that E[2] c E(K) implies that there are isomorphisms
HI (G K/ K, E[2]) ~ HI (G K/ K, (7l/27l)2) ~ Hom( G K/ K, (7l/27l)2)

~ Hom(G K / K ,7l/27l)2 ~ Hom(G K / K ,JL2)2 ~ HI(G K / K ,JL2)2.
Finally, the Kummer sequence for fields and Hilbert's theorem 90 give an
isomorphism [AEC, VIII.2.2]
Combining the above maps, we obtain an injective homomorphism
E(K)/2E(K) ~ HI(G K / K ,E[2]) ~ HI(G K / K ,JL2)2

8-~-1 (K* / K*2) x (K* / K*2).
Of course, one still needs to unscramble the connecting homomorphisms

and check that this map is the same as the map ¢ defined above. See
[AEC, X §1] for the details.
We will prove that E(K)/2E(K) is finite by proving that the im-
age ¢(E(K)/2E(K)) is finite. The basic idea is to show that for any P E
E(K), the two coordinates of ¢(P) are almost squares in K. The following
lemma quantifies this assertion.
Lemma 2.3.1. Let k be an algebraically closed field, let K = k( C) be

the function field of a curve, let E / K be an elliptic curve, and suppose
that E has a \Veierstrass equation of the form
E : y2 = (.r - ed(x - e2)(x - e3) with el, e2, e3 E K.
Let SeC be the set of points where anyone of e l, e2, e3 has a pole,
together with the points where.6. = (el - e2)2(eJ - e3)2(e2 - (3)2 vanishes.
Then for any point P = (x, y) E E(K) with x of. el,
ordt(:r - ed == 0 (mod 2) for all t E C wiih t ¢:. S.
Here ord t : k( C) * ---> Z is the normalized valuation on k( 0) which measures
the order of vanishing of a function at t [AEC II §1].
PROOF. Let t E C with t ¢:. 8, and let n = ordt(x - ed. Our choice of S
implies that ord t (ei) ~ O. We consider three cases. First, if n = 0, then
clearly ordt(x - ed = n == 0 (mod 2).
Second, if n < 0, t.hen t must be a pole of .J: and n = ord t (x). It follows
that
n = ordt(x - ed = ordt(x - e2) = ordt(:r - e3)'
Using the vVeierstrass equation for E, we find that
2ordt (y) = ord t (y 2) = ordt((x - eJ(::r - e2)(x - e3)) = 3n,
which proves that ordt(x - eJ) = n == 0 (mod 2).
Third, suppose that n > O. This means that x - el vanishes at t. vVe
claim that x - e2 and J; - e3 do not vanish at t. To see this, we let i = 2
or 3 and use the triangle inequality to compute
min{ordt(x-eJ),ordt(x-ei)} :::; ordt((.J:-eJ)-(x-ei)) = ordt(e;-cJ) = O.
The last equality follows from the assumption that C1, C2, e3 do not. have
poles at. t and .6. does not vanish at t. But ordt(x - C1) = n ~ 1, so we
get ordt(x - e2) = ordt(x - e:,) = O. Therefore
= ord t (y2) = ordt((x - eJ)(.r - e2)(::r - C3)) = ordt(x - C1).
2ord t (y)
Hence ordt(x - cJ) == 0 (mod 2), which completes the proof of (2.3.1).
o
vVe now resume the proof of (2.1). Let S be as in the statement
of (2.3.1), and define a subgroup of K* / K*2 by
K(S,2) = {J E K* / K*2 : ordt(J) == 0 (mod 2) for all t ¢:. S}.
(Compare with [AEC, X.1.4].) Then (2.3.1) tells us that the image of ¢
lies in K(S, 2) x K(S, 2), so we have an injective homomorphism
¢ : E(K)/2E(K) --> K(S, 2) x K(S, 2).
In order to complete the proof of (2.1), it suffices to prove that the group
K(S, 2) is finite. Note that the finiteness of K(S, 2) is an assertion about
the curve C; it has nothing to do with the elliptic curve E. We record this
statement in the following lemma, whose proof also completes the proof of
the weak l\lordell-Wei! theorem (2.1).
Lemma 2.3.2. Let k be an algebraically closed field, let K = k( C) be the

function field of a curve, let SeC be a finite set of points, and let m ~ 1
be an integer. Then the group
K(S, m) = {J E K* jK*m : ordt(f) == 0 (mod m) for all t t/:- S}
is a finite subgroup of K* j K*m.
PROOF. Let s = #S. Then there is an exact sequence

o ----> K(0,m) ----> K(S,m) ----> (ZjmZ)S.
f f---> (ordt(f))tES
It thus suffices to prove that
K(0, m) = {J E K* j K*m : ordt(f) == 0 (mod m) for all t E C}
is finite.
Let f (mod K*m) E K(0, m). Then div(f) has the form
div(f) = mDf for some Df E Div(C).
Notice that if we take some other representative f gm for the coset of f

in K* j K*m, then Dfg~ = Df + div(g) changes by a principal divisor.
Thus the divisor class of D f in Pic( C) is independent of the choice of
representative. Further, mDf = div(f) is itself principal, so we get a well-
defined homomorphism
K(0, m) ----> Pic(C)[m],
where Pic(C)[m] denotes the elements of Pic(C) of order dividing m.

Now suppose that f (mod K*m) is in the kernel of this homomorphism.
Then D f is principal, say D f = div( F f ) for some function Ff E K*, which
means that
div(f Fim) = div(f) - m div( P f ) = div(f) - mD f = O.

Thus fFim has no zeros or poles, so [AEC, 11.3.1] tells us that it is con-
stant. Using the assumption that k is algebraically closed, we can write this
constant as cm , so f = (cFf)m. In other words, f == 0 (mod K*m). This
proves that the homomorphism K(0, m) ----> Pic(C)[m] is injective. Finally,
we use the fact (2.2) that Pic(C)[m] is finite to conclude that K(0,m) is
finite. D
In the remainder of this section we briefly discuss abelian varieties and

Jacobians, including a sketch of the proof that the Picard group of a curve
has only finitely many elements of a given order.
Definition. An abelian variety consists of a non-singular projective vari-

ety A, a point 0 E A, and two morphisms
p, : A x A -----> A, i :A -----> A,
which make the points of A into an abelian group. In other words,

(i) p,( 0, P) = p,(P, 0) = P for all PEA,
(ii) p,(P,i(P)) =0 for all PEA,
(iii) p,(p,(P,Q),R) = p,(P,p,(Q,R)) for all P,Q,R E A,
(iv) p,(P, Q) = p,(Q, P) for all P, Q E A.
If A, p" and i are defined over a field k, and 0 E A(k), then we say that A
is defined over k. Basic references for the theory of abelian varieties are
Griffiths-Harris [1, Ch. 2, §6], Milne [2], and Mumford [1], but we will need
very little of the general theory.
Example 2.4.1. An elliptic curve is an abelian variety of dimension one.
The fact that the addition and negation operations on an elliptic curve
satisfy (i)-(iv) is [AEC, III.2.2]' and the fact that they are morphisms
is [AEC, III.3.6]. Conversely, every abelian variety of dimension one is an
elliptic curve, see exercise 3.5.
Example 2.4.2. Let A be an abelian variety of dimension d defined over C.
Then one can show that there is a lattice A C Cd and a complex analytic iso-
morphism A(C) ~ Cd / A. By lattice we mean a full sublattice, that is, a free
subgroup of Cd of rank 2d which contains an ffi.-basis for Cd. The isomor-
phism A(C) ~ Cd / A is an isomorphism both as complex manifolds and as
abelian groups. For elliptic curves, the existence of this isomorphism is the
essential content of the uniformization theorem [AEC, VI.5.1.1]. Further,
in the one dimensional case every lattice A c C corresponds to an elliptic
curve [AEC, VI.3.6]. But in higher dimensions, a complex torus Cd / A will
only be isomorphic to an abelian variety if the lattice A satisfies the Rie-
mann conditions (see Griffiths-Harris [1, Ch. 2, §6]). In other words, there
are certain restrictions on the lattice A in order for there to be a complex-
analytic embedding of Cd / A into projective space ]P'n(C). Just as in the
case of elliptic curves, this complex uniformization of abelian varieties is
very useful in analyzing the group structure of A(C).
Remark 2.5. A more succinct way to define abelian varieties is to say that
an abelian variety is a group variety in the category of projective varieties.
It turns out that the group law forces A to be non-singular. Further, the
completeness of the variety A forces the group law to be commutative, so
we actually did not need to include property (iv) in our definition. See
Mumford [1, pp. 1, 41, 44] for three proofs of this fact.
The next proposition tells us that the Picard group of a curve is es-
sentially an abelian variety, called the Jacobian variety of the curve. Its
dimension is equal to the genus of the curve. Some general references for
Jacobian varieties include Griffiths-Harris [1, Ch. 2, §§2, 3, 7], Milne [3],
and Mumford [2].
Proposition 2.6. Let C be a non-singular projective curve of genus 9
defined over an algebraically closed field k.
(a) The degree map deg : Div(C) ----> Z induces an exact sequence
where Pico( C) is the group of degree 0 divisor classes on C.

(b) There exists an abelian variety Jac(C) of dimension 9 and a natural
isomorphism of groups
Pico(C) ~ Jac(C).
(For the meaning of "natural", see (2.6.1) below.) Jac(C) is called the
Jacobian variety of C.
PROOF. (a) The exact sequence merely defines the subgroup Pico(C), so
all we need to do is verify that the degree map is well-defined on Pic( C).
This follows immediately from the fact that every principal divisor has
degree 0 [AEC, II.3.l].
(b) (Proof Sketch) We start with the two easy cases. First, if 9 = 0, then
every divisor of degree 0 on C ~ pI is principal [AEC, II.3.2]. It follows
that Pico (C) = 0, which is the desired result in this case.
Next suppose that 9 = 1. Fixing a point 0 E C(k), we turn Clk into
an elliptic curve, so C is an abelian variety (2.4.1). Then there is a natural
group isomorphism Pico(C) ..::-. C as described in [AEC, III.3.4]. Hence C
is its own Jacobian variety.
For curves of higher genus, one can construct the Jacobian variety
analytically if k = <C or algebraically in general. We briefly describe both
approaches. For the algebraic method, we fix a basepoint Po E C and
consider the map
Let the symmetric group Sg act on cg by permuting the coordinates. The

map <pg is clearly invariant under this action, so it induces a map on the
quotient C(g) d~f cg ISg . One checks that C(g) is a non-singular variety,
that cPg : C(g) ----> Pic°(C) is surjective, and that <pg is injective off of a
Zariski closed subset of C(g). Further, the group law on PicO(C) induces
an algebraic (Le., rational) map C(g) x C(g) ----> C(g). Unfortunately, this
rational map is not defined everywhere, so one takes certain "group chunks"
and glues together enough translates to form a group variety. This idea
of gluing together group chunks is due to Andre Weil [3] and works in all
characteristics. We refer the reader to the discussion in (IV, §6) for further
details. See also exercise 4.29.
The analytic approach to constructing the Jacobian for curves over «:::
is much older. The Riemann-Roch theorem [AEC, 11.5.3, 1I.5.5a] says that
on a curve of genus g, the space of holomorphic differential forms has di-
mension g. Let l<h, ... , Wg be a basis for this space. Next, a curve C of
genus 9 over «::: is a Riemann surface with 9 holes, so there are 2g inde-
pendent cycles on C. Let r 1 , ... , r 2g be a basis for the space of cycles;
that is, r 1 , ... ,r2g is a basis for the first homology Hl(C,Z). We fix a
basepoint Po E C and consider the map
Here the integrals are to be computed along some path from Po to P.

Unfortunately, the value of the integrals is not path-independent! (See
[AEC, VI, §1] for a discussion when 9 = 1.)
In order to salvage this idea, we consider the subgroup A c «:::g defined
to be the image of the map
Then the integrals give a well-defined map cp : C -+ «:::g / A, since A elim-

inates the path dependence of the integrals. Next one proves that A is
a lattice which satisfies Riemann's conditions (2.4.2), so «:::g / A is complex-
analytically isomorphic to an abelian variety. Denoting this abelian variety
by Jac(C), one verifies that the map cp : C -+ Jac(C) is a morphism of al-
gebraic varieties.
Extending cp by linearity gives a map
cp : Divo(C) --+ Jac(C),
In other words, use d> to map the points in a divisor to points of J ac( C), and
then use the group law on Jac(C) to add them up. Finally, the theorems
of Abel and Jacobi say that the map cp : Divo(C) -+ Jac(C) is a surjective
homomorphism whose kernel consists of precisely the principal divisors.
Hence d> induces the desired isomorphism Pico(C) ..::::. Jac(C). For further
details, see Griffiths-Harris [1, Ch. 2, §§2,7]. 0
Remark 2.6.1. What do we mean in (2.6b) when we say that the isomor-
phism Pico(C) -+ Jac(C) is "natural"? Recall [AEC, 11.3.7] that a mor-
phism d> : C 1 -+ C 2 of curves induces a homomorphism cp* : Pico(C2 ) -+
PicO(C1 ) of their Picard groups. Then one can prove that the corresponding
map cp* : Jac(C2 ) -+ Jac(C1 ) is a morphism of varieties. In fancy language,
the association C f---> Jac(C) is a functor from the category of (non-singular
projective) curves to the category of abelian varieties.
Remark 2.6.2. If C is defined over an arbitrary field k, then its Jacobian

variety Jac( C) will be an abelian variety defined over k. Further, the group
isomorphism PicD (C) ---> J ac (C) will commute with the action of the Galois
group G k / k . This is another way in which the Jacobian is a natural object.
For example, if C is a curve of genus 1 with C(k) = 0, then C cannot be its
own Jacobian variety. However, we can always find an elliptic curve Elk
so that Clk is a homogeneous space for E [AEC, exercise 10.3]. Then
[AEC, X.3.8] says that there is a group isomorphism PicD(C) ~ E, so E is
the Jacobian of C.
Remark 2.6.3. For hyperelliptic curves, it is possible to construct the
Jacobian variety quite explicitly, see Mumford [3, Ch. IlIa, §§2,3]. In this
case one can also precisely describe all of the elements of order 2 in Pic( C),
see exercise 3.38.
The following corollary of (2.6) and (2.4.2) is a strengthened version
of (2.2).
Corollary 2.7. Let Clk be a non-singular projective curve of genus 9
defined over an algebraically closed field k. Then
In particular, for any integer m ::::: 1, Pic(C)[m] ~ (ZjmZ)2g, so Pic (C) has
only finitely many elements of order dividing m.
PROOF. The field k has characteristic 0 by assumption, so the Lefschetz

principle [AEC, VI §6] says that we may take k to be a subfield of C.
Let J = Jac(C) be the Jacobian variety of C. Then (2.6) tells us that
Pic(Chors ~ PicD(C)tors ~ J tors . On the other hand, (2.4.2) implies
that there is a lattice A c reg and a complex-analytic group isomor-
phism J(C) ~ C9 I A. (See also the proof of (2.6) for a concrete description
of the isomorphism PicD (C) ~ C9 I A.)
As abstract groups, reg ~ 1R2g and A ~ Z29 , and if we use a Z-basis
for A as our IR-basis for re 9 , then we obtain a group isomorphism of the
quotients reg I A ~ (IR/Z) 29 . Hence
This proves the first assertion of (2.7), and the other assertions are an
immediate consequence. 0
§3. Elliptic Surfaces
We now return to the idea that an elliptic surface should be a one-parameter

family of elliptic curves. For example, we might consider a family
ET : y2 = x 3 + A(T)x + B(T)
with rational functions A(T), B(T) E k(T). Or, more generally, we could
fix a non-singular projective curve C / k and take
E : y2 = x 3 + Ax + B
for some A, B E k( C) with 4A 3 + 27 B2 =I O. Then for almost all points t E
C(k) we can evaluate A and B at t to get an elliptic curve
Et : y2 = x 3 + A(t)x + B(t).
Suppose now that we do not evaluate A and B at particular points
of C, but instead we treat t as a variable just like x and y. In other words,
we look at the subset of]P>2 x C defined by
Note that c is a subvariety of ]p>2 x C of dimension two; it is a surface

formed from a family of elliptic curves.
Remark 3.1. Actually, in defining c we need to take a little more care
with the points t E C where A or B has a pole. Here's one way to handle
this problem. Consider the set
2 A or B has a pole at t, or }
{ ([X, Y, z], t) E]P> xC: y2 Z = X 3 + A(t)X Z2 + B(t)Z3 .
This set will consist of a number of irreducible components, all but one of
which will look like
]p>2 X {to}
for a pole to of A or B. We take c to be the one component not of this
form. Equivalently, c is the Zariski closure in JP'2 x C of the set
2 t is not a pole of A or B, and }

{
([X, Y, Z], t) E]P> xC: y2 Z = X3 + A(t)X Z2 + B(t)Z3 .
Since c is a subvariety of ]p>2 x C, projection onto the second factor

defines a morphism
7r: c -----> C,
([X, Y, Z], t) f------7 t.
And for almost every t point t E C, the fiber
is the curve E t that we considered earlier. Further, since we have assumed

that
~ = -16(4A3 + 27B2) i- 0 in k(C),
it will be true that almost every fiber £t is an elliptic curve. We just need
to choose points t E C such that A(t) i- 00, B(t) i- 00, and ~(t) i- O.
However, our family of elliptic curves £ has one other important prop-
erty. Recall [AEC III §3] that an elliptic curve is really a pair (E,O),
where E is a curve of genus 1, and 0 is a point of E. The equation we
used to define £ gives a one-parameter family of elliptic curves. This means
that for almost all values of t we get an elliptic curve £t, which we should
really write as (£t, Ot) to emphasize that each £t comes equipped with a
zero element Ot E £t.
The family £ is an algebraic family, which is a fancy way of saying
that it is given by an equation whose coefficients A and B are algebraic
functions, in our case functions on the curve C. The additional property
that £ possesses is that the collection of zero elements Ot is an algebraic
family of points. In other words, we claim that the coordinates of Ot are
algebraic functions of t; the coordinates of Ot are in the function field of C.
Using the definition of £ given above, we see that
Ot = ([0,1,0], t) E £t C jp'2 x C.
So the coordinates [0,1,0] of Ot are actually constant functions.

We can describe this property in another way. Since each fiber £t is
an elliptic curve with zero element Ot, we get a map
0'0: C -----+ £,
t ~ Ot.
Clearly, this map has the property that
for all t E C(k).
Further, the fact that Ot is an algebraic family of points is equivalent to

the fact that the map 0'0 is a rational map of varieties. (In fact, since C is
a non-singular curve, 0'0 will be a morphism [AEC 11.2.1].) This prompts
the following definition.
t The phrases "for almost every" and "for almost all" are contractions of the
expression "for all but finitely many."
Definition. Let Jr : V -+ W be a morphism of algebraic varieties. A

section to Jr is a morphism
such that the composition

Jroa: W ----+ W
is the identity map on W.
Example 3.2. Consider the surface given by the equation
£ : y2 Z = X 3 _ T2 X Z2 + T2 Z3.
(See (1.1) for our original discussion of this equation.) More precisely, let £
be the projective surface in p2 x pI corresponding to this equation. The
projection Jr is the map
Jr : £ ----+ pI, ([X, Y, Z], T) f-t T.
This map has a section
a : pI ----+ £, T f-t ([T, T, 1], T).
To avoid excessive notation, one often says that Jr : £ -+ pI has the sec-
tion a = [T, T, 1]; one can even use the inhomogeneous equation
y2 = x 3 _ T2 X + T2
for £ and say that a = (T, T) is a section to Jr.
We are now ready for the formal definition of an elliptic surface.

Definition. Let C be a non-singular projective curve. An elliptic surface
over C consists of the following data:
(i) a surface £, by which we mean a two dimensional projective vari-
ety,
(ii) a morphism
Jr : £ ----+ C
such that for all but finitely many points t E C(k), the fiber
£t = Jr-I(t)
is a non-singular curve of genus 1,
(iii) a section to Jr,
ao : C ----+ £.
Let £ -+ C be an elliptic surface. The group of sections of £ over C is
denoted by
£(C) = {sections a : C -+ £}.
Note that any rational map C -+ £ is automatically a morphism, since C
is a non-singular curve and £ is a projective variety [AEC, II.2.1], so every
section is a morphism. We will see later (3.10) that fCC) is a group with
zero element ao.
Remark 3.3.1. Since all but finitely many fibers of an elliptic surface
have genus 1, one can show that any fiber which is a non-singular curve
will automatically have genus 1. See Hartshorne [1, III.9.13]. These non-
singular fibers are often called the good fibers. The fibers Gt which are
not non-singular curves will be called the singular fibers or the bad fibers.
Of course, when we refer to the non-singular fiber Gt, we really mean the
elliptic curve consisting of the pair (Gt, ao (t)) .
Remark 3.3.2. Our definition of elliptic surface is non-standard in two
ways. First, most books require that G be a non-singular surface. In such
a case we will call G a non-singular elliptic surface. Second, most algebraic
geometers would define an elliptic surface to be a (non-singular) surface
satisfying properties (i) and (ii) of our definition; they would not require
that there be a section. This leads to many interesting geometric questions,
such as the possible existence of multiple fibers. (See Griffiths-Harris [1,
p. 564].) It is only our emphasis on questions with an arithmetic flavor
which prompts us to require the existence of at least one section.
Remark 3.3.3. The classical theory of elliptic surfaces deals with surfaces
defined over the field k = <C, or more generally over an algebraically closed
field. We will also want to look at other fields, such as k = IQ. We will say
that an elliptic surface E over C is defined over k if the curve C is defined
over k, the surface G is defined over k, and both of the maps
7r : E ------> C and ao : C ------> E
are defined over k. In this case we write
c(C / k) = {sections a : C ----> E such that a is defined over k}
for the group of sections defined over k. For example, the elliptic sur-
face (3.2) is defined over IQ, and the section a = (T, T) is in G(C /IQ).
Let E be an elliptic surface over C defined over k. We would like
to associate to E an elliptic curve E / k( C). Conversely, to each elliptic
curve E / k( C) we will assign a birational equivalence class of elliptic sur-
faces. In this way we will be able to apply our earlier results to study
elliptic surfaces.
We begin by recalling some general definitions and basic facts about
rational maps. For more details, see Hartshorne [1, I §4]' Harris [1, Lec-
ture 7], or Griffiths-Harris [1, 4 §2J.
Definition. Let V and W be projective varieties. A rational map from V
to W is an equivalence class of pairs (U, <Pu), where U is a non-empty Zariski
open subset of V and <Pu : U ----> W is a morphism. Two pairs (U, <Pu)
and (U', <Pu are deemed equivalent if <Pu = <Pu on un U'. If (V) , is defined as follows:

Let U c V be an open subset on which ¢ is defined, take the Zariski closure
in V x W of the graph
{(u,¢(u)) E V x W : u E U},
and take the projection to W. (N.B. Since ¢ is not actually defined at

all points of V, this "image" of ¢ may not agree with your intuition. For
example, there may be points w E ¢(V) which are not equal to ¢( v) for
any v E V.)
A rational map ¢ : V ~ W is dominant if ¢(V) = W. Equivalently, ¢
is dominant if for one (hence every) open set U on which it is defined, the
image ¢(U) is Zariski dense in W.
The domain of definition of a rational map ¢ : V ~ W, which we de-
note by Dom(¢), is the largest open subset of Von which ¢ is a morphism.
(Such a largest set exists, see Hartshorne [1, I exercise 4.2].)
A rational map ¢ : V ~ W is a birational isomorphism if it has a
rational inverse 'lj; : W ~ V; that is, ¢ and 'lj; are dominant and
¢o'lj;:W~W and
are the identity maps at all points for which they are defined. If there
is a birational isomorphism from V to W, then V and Ware said to be
birationally equivalent. If V, W, ¢, and 'lj; are all defined over a field k,
then we say that V and Ware birationally equivalent over k.
Remark 3.4. In [AEC, I §3] we defined rational maps more naively using
coordinates on lpm. The reader will easily check that the two definitions
are equivalent. For examples of rational maps that are not morphisms,
see [AEC, 1.3.6 and 1.3.7]. Notice that [AEC, 1.3.7] gives an example of non-
isomorphic varieties that are birationally equivalent. Another important
example of this phenomenon is provided by the process of "blowing-up";
see Hartshorne [1, I §4] or Harris [1, Lecture 7].
Proposition 3.5. Let ¢ : V ~ W be a rational map of projective
varieties.
(a) The image ¢(V) is an algebraic subset ofW. If V is irreducible, then
so is ¢(V).
(b) Suppose that V is non-singular. Then ¢ is defined except on a set of
codimension at least two. In other words, every component of the comple-
ment of Dom( ¢) in V has codimension at least two.
PROOF. (a) See Harris [1, Lecture 7, p. 75] or Griffiths-Harris [1, 4 §2,
p. 493] for the first part. The second part is immediate from the definition
of irreducibility; see exercise 3.7.
(b) See Griffiths-Harris [1, 4 §2, p. 491]. 0
Proposition 3.6. Let V/k and W/k be projective varieties. The follow-
ing are equivalent.
(i) V and Ware birationally equivalent over k.
(ii) The function fields k(V) and k(W) are isomorphic as k-algebras.
(iii) There are non-empty Zariski open sets U 1 C V and U2 C W defined
over k such that U 1 and U2 are isomorphic over k.
\
PROOF. This is a standard (and elementary) result in algebraic geometry.
See, for example, Hartshorne [1,1.4.5] or Harris [1, exercise 7.10]. 0
Let ¢ : V ----> W be a dominant rational map, and let f E k(W)

be a rational function on W. Then f is defined (i.e., regular) on a non-
empty open subset of W, and ¢(Dom(qI)) is Zariski dense in W, so the
composition f 0 ¢ is a regular function on a non-empty open subset of V. In
other words, fo¢ is a rational function on V, so we obtain a homomorphism
of k-algebras
k(W) ~ k(V),
f f-------> fo¢.
The next result says that the theory of varieties up to birational equiv-
alence is essentially the same as the theory of their function fields. (See
Hartshorne [1,1.4.4] for a more precise categorical statement.)
Proposition 3.7. Let V/k and W/k be projective varieties. The associ-
ation
{ dominant rational maps} ~ { k-algebra homomorPhisms}

V ----> W defined over k k(W) ----> k(V) ,
¢ f-------> (fr-tfo¢)
is a bijection.
PROOF. See Hartshorne [1, 1.4.4]. o
We are now ready to apply the theory of rational maps to study elliptic
surfaces. Recall that according to our definition, an elliptic surface consists
of three pieces of data:
(i) a projective surface c,
(ii) a projection map 7r : C ----> C,
(iii) a zero section 0'0 : C ----> c.
Given two elliptic surfaces c and C' over the same base curve C, it thus
makes sense to consider the rational maps c ----> C' which commute with
projections and/or zero sections. This prompts us to make the following
definitions.
Definition. Let 7r : £ --+ C and 7r' : £' --+ C be elliptic surfaces over C.
A rational map from £ to £' over C is a rational map ¢ : £ --+ £' which
commutes with the projection maps, 7r' 0 ¢ = 7r. The elliptic surfaces £
and £' are birationally equivalent over C if there is a birational isomor-
phism ¢ : £ --+ £' which commutes with the projection maps. If the elliptic
surfaces and rational maps are defined over a field k, we will say that £
and £' are k-birationally equivalent over C.
The next two propositions explain precisely how the theory of elliptic
curves over k( C) is the same as the birational theory of elliptic surfaces
over C.
Proposition 3.8. (a) Fix an elliptic curve E/k(C). To each Weierstrass
equation for E,
E : y2 = x 3 + Ax + B, A, BE k(C),
we associate an elliptic surface
as described in (3.1). Then all of the £(A, B) associated to E are k-bira-

tionally equivalent over C.
(b) Let £ be an elliptic surface over C defined over k. Then £ is k-bira-
tionally equivalent over C to £(A, B) for some A, B E k(C). Further, the
elliptic curve
E : y2 = x 3 + Ax + B
over k(C) is uniquely determined (up to k(C)-isomorphism) by £.
(c) Let E / k( C) be an elliptic curve and £ --+ C an elliptic surface associ-
ated to E as in (a). Then
k(£) ~ k(C)(E) as k(C)-algebras.
Here the projection map 7r : £ --+ C induces an inclusion of fields k( C) <---+

k(£) which makes k(£) into a k(C)-algebra.
We say that E/k(C) is the generic fiber of £ --+ C.
PROOF. (a) Suppose we take another Weierstrass equation for E/k(C), say
E : y,2 = x,3 + A' x' + B', A', B' E k(C).
Then there is au E k(C)* such that u 4 A' = A and u 6 B' = B [AEC III.1.3].
Now the map
£(A', B') ---+ £(A, B), ([X',Y',Z'],t) t---+ ([u 2 X',u 3 Y',Z'],t)
shows that £(A, B) and £(A', B') are k-birationally equivalent over C.
(b,c) The projection map 7r : £ --> G induces an inclusion of function

fields k( G) '----t k( £) in the usual way, I f-+ I 0 7r. Further, £ is a surface
over k, and G is a curve over k, so k(£)jk has transcendence degree 2
and k( G) j k has transcendence degree 1. It follows that k( £) j k( G) has
transcendence degree I, so there exists a curve Ejk(G), unique up to k(G)-
isomorphism, whose function field k( G) (E) is isomorphic to k( £) as k( G)-
algebras [AEC 11.2.5].
We claim that E is a curve of genus 1. To see this, we write E as a
subvariety of 1pm, so E is the set of zeros of a collection of homogeneous
polynomials {hex) : 1::; i::; r} with coefficients in keG). Note that for
almost all t E G, we can evaluate the coefficients of the I;'s to get polyno-
mials with coefficients in k. To indicate the dependence of the h's on t we
will write h(t, x). Then we can consider the algebraic variety in 1pm x G
defined by
V ~f {(x,t) E]p>n x G : h(t,x) = 0 for 1::; i::; r}.
Projection onto the second factor gives a map V --> G which makes k(V)
into a k(G)-algebra, and by construction we see that k(V) is isomorphic
to k(G)(E) as k(G)-algebras. Hence k(V) is isomorphic to k(£) as k(G)-
algebras, so (3.6) tells us that V and £ are birationally equivalent over G.
In particular, for almost all t E G the fibers Vi and £t are isomorphic, so
almost all of the Vi's are curves of genus 1.
Now suppose that W E OEjk(C) is a differential form on E. (See
[AEC, II §4] for general properties of differential forms on curves.) Any such
differential can be written as a sum W = 2: Uj dVj with Uj, Vj E k( G) (E).
For almost all t E G we can evaluate the Uj'S and Vj'S at t to get a differ-
ential form Wt = wet, x) on Vi. Further, if w is a holomorphic differential
form on E, then Wt will be a holomorphic differential form on Vi for almost
all t E G.
Let WI,W2 E OE/k(C) be non-zero holomorphic differential forms. We
claim that they are k(G)-linearly dependent. To prove this, we observe
that for almost all t E G, the forms WI (t, x) and W2(t, x) are holomor-
phic differentials on the curve Vi of genus I, so they are k-linearly de-
pendent from [AEC, II.5.3,II.5.5a]. In other words, there are non-zero
constants at, bt E k such that
But this means that the function WI(t,X)jW2(t,X) E k(Vi) is constant;

that is, it is in k. It follows that the function WdW2 E k(V) = k(G)(E) is
actually in k(G), which proves that WI and W2 are k(G)-linearly dependent.
To recapitulate, we have proven that the vector space of holomorphic
differential forms in OEjk(C) has k(C)-dimension at most one. It follows
from [AEC, II.5.3,II.5.5a] that E has genus at most 1. Suppose that Ejk(C)
has genus O. We will see below that E(k(G)) is non-empty, so we can

assume that E = pI and V = pI X G. But then the fibers vt ~ pI are
all curves of genus 0, contradicting the fact that vt ~ Ct has genus 1 for
almost all t. This contradiction shows that E j k( G) does not have genus 0,
so the only possibility is that Ejk(G) has genus l.
Further, the section ao : G --; C corresponds to a point Po E E(k(G)).
To see this we use the fact that V and care birationally equivalent over G
to get a section ab : G --; V. This section is a map of the form ab =
[h o, ... , hnl for certain functions ho, ... , hn E keG), which is the same as
saying that Po = [ho, ... ,hnl E pn (k( G)) is a point in E (k( G)). Taking Po
to be the identity element, E becomes an elliptic curve defined over keG).
We have now proven that Ejk(G) is an elliptic curve, so we can take
a Weierstrass equation for it, say
with A, B E keG).
Then the corresponding surface V is precisely the elliptic surface c(A, B)

described in (a), and we have already observed above that V is birationally
equivalent to £ over G. This completes the proof of the first part of (b)
which asserts that every elliptic surface is birationally equivalent over G
to some £(A, B). Further, we showed above that there are isomorphisms
of k(G)-algebras
k(c) ~ k(V) ~ k(G)(E),
which completes the proof of (c).
It remains to prove that E is determined up to k(G)-isomorphism by C.
Suppose that c is also birationally equivalent to c(A',B'). Then c(A,B)
and £(A', B') are birationally equivalent over G. This means that for al-
most all t E G, there is an isomorphism on the fibers
From general principles [AEC, III.3.1b]' we know that this isomorphism is

given by a map of the form (x, y) f-t (atx, f3tY) for some at, f3t E k. But the
birational equivalence c(A, B) --; £(A', B') is an algebraic map, so we see
that a and f3 are functions on c(A, B) which depend only on t. In other
words, a, f3 E k( G), which proves that the corresponding elliptic curves
and
are isomorphic over keG). o

Proposition 3.9. There is a natural bijection
{ dominant rational maps} {non-constant maps E --; E'}

E --; E' over G r---t defined over k( G) ,
where we identify the set of elliptic curves defined over k( G) with the set
of (birational equivalence classes) of elliptic surfaces over G as described in
Proposition 3.8.
PROOF. Let <p : £ ----+ £' be a dominant rational map over G. Choose
Weierstrass equations for £ and £', or equivalently for E and E',
£ : y2 = x 3 + Ax + B, C' : y,2 = X,2 + A'x' + B'.
Then the map -> ((J(t,x,y),g(t,x,y)),t),
where f,g E k(£) ~ k(G)(E) = k(G)(x,y). Hence F = (f,g) defines a
mapF:E----+E'.
Suppose that F is a constant map from E to E' over keG). This
means that f, 9 E k( G), so <p( (x, y), t) = ((J(t), g(t)), t) depends only on t,
independent of x and y. It follows that the image of <p has dimension at
most one, since it is the image of a map G ----+ £', so in particular <p is not
dominant. This proves that if <p : £ ----+ £' is dominant, then the associated
map F : E ----+ E' is non-constant.
The proof going the other direction is similar. Fix Weierstrass equa-
tions for E and E' as above, and let F : E ----+ E' be a map defined over k( G).
Then F has the form F = (f,g) for some f,g E k(G)(E) ~ k(£), and we
can define a map <p: £ ----+ C' over G by
<p: ((x,y),t) >-> ((J(t,x,y),g(t,x,y)),t).
If <p is not dominant, then its image must consist of a curve, since 1f
maps the image <p(c) onto G. But this means that if we fix (almost any
point) t E G and vary x, yon the fiber Ct, then <p((x, y), t) can assume only
finitely many values. In other words, the map <p(', t) : Ct ----+ p1 takes on
only finitely many values, so it is constant [AEC 11.2.3]. Hence f and g do
not depend on x,y, so f,g E keG) and F = (f,g) : E ----+ E' is a constant
map. This proves that if F : E ----+ E' is non-constant, then the associated
map <p : C ----+ £' is dominant, which completes the proof of Proposition 3.9.
o
Our final task is to explain how the set of sections £ (G) has a natural
group structure. Recall that for almost all points t E G, the fiber Ct is an
elliptic curve (3.3.1), so given any two points on Ct, we can add them or
take their inverses. Let 0'1,0'2 E C(G) be two sections to £. We define new
sections 0'1 + 0'2 and -0'1 by the rules
and
valid for all t E G such that the fiber Ct is non-singular. We will verify
below (3.10) that 0'1 + 0'2 and -0'1 define rational maps G ----+ C, so in fact
they define morphisms since G is a non-singular curve [AEC, 11.2.2.1]. The
next proposition says that this "fiber-by-fiber" addition makes £(G) into a
group.
Proposition 3.10. Let c ---> C be an elliptic surface defined over k.

(a) Let CTI,CT2 E c(Cjk) be sections defined over k. Then the maps CTI +CT2
and -CT2 described above are in c(Cjk).
(b) The operations (CTI' CT2) ---> CTI + CT2 and CT ---> -CT make c(C j k) into an
abelian group.
(c) Let Ejk(C) be the elliptic curve associated to c as described in (3.8).
Then there is a natural group isomorphism
E(k(C)) c(Cjk),
P=(xp,yp) (CTP : t ---> ((xp(t), yp(t)), t)).
PROOF. (a) Take a Weierstrass equation for c ---> C as described in (3.8),
A,B E k(C).
Then a section CTi : C ---> c is given by a pair of functions

CTi : t f----> (Xi(t), Yi(t))
which satisfy the given Weierstrass equation for (almost all) t E C. Equiva-
lently Xi, Yi E k(C) are functions satisfying Y; = xy + AXi + B as elements
of k (C). By definition, (CTI + CT2) (t) is the sum of the two points CTI (t)
and CT2(t) using the addition law on the elliptic curve
Ct : y2 = x3 + A(t)x + B(t).
So the usual addition formula [AEC III.2.3] says that if XI(t) f= X2(t), then
(We leave it to you to fill in the y-coordinate.) In other words, if Xl f= X2

in k(C), then the map CTI + CT2 is given by the formula
which shows that CTI +CT2 is a rational map from C to c defined over k. Simi-
larly, if Xl = X2 and YI f= Y2, the duplication formula [AEC III.2.3(d)] yields
the same conclusion. Finally, the map -CTI is given by -CTI = (Xl, -yd,
so -CTI also gives a rational map C ---> c defined over k. But C is a non-
singular curve, so all of these rational maps are morphisms [AEC 11.2.2.1],
which completes the proof that CTI + CT2 and -CTI are in c( C j k).
(b) This is clear from the fact that the points on almost every fiber form a
group. For example, for any three sections 0'1,0'2,0'3 E [( C I k) and almost
all points t E C, we have
+ 0'2) + a3)(t) = (0'1 + a2)(t) + a3(t) = (al(t) + a2(t)) + a3(t)

(0'1
= al(t) + (a2(t) + a3(t)) = al(t) + (0'2 + a3)(t) = (0'1 + (0'2 + a3))(t).
Hence (0'1 + 0'2) + 0'3) = (0'1 + (0'2 + 0'3)) as sections, which verifies
the associative law. The other group axioms can be verified in a similar
fashion.
(c) Fix a Weierstrass equation for [ ---+ C as in (a). If P = (xp,yp) E
E(k(C)), then Xp and yp satisfy the given Weierstrass equation as ele-
ments of the function field k(C), so (xp(t),yp(t)) E [t for almost all t E
C. This shows that ap is a well-defined element of [(Clk). Similarly,
any a E [(Clk) has the form a(t) = (xo-(t),yo-(t)),t) for some rational
functions xo-, Yo- E k( C) satisfying
for almost all t E C.
It follows that Po- = (xo-, Yo-) satisfies the given Weierstrass equation,
so Po- E E(k(C)). The identifications P f--+ ap and a f--+ Po- are clearly
inverse to one another, so they define bijections E(k(C)) +-+ [(Clk). Fi-
nally, for any PI, P2 E E(k(C)) we have
(a P, + ap2) (t) = (xP, (t), YP, (t)) + (XP2 (t), YP2(t)) = ap, +P2 (t).
Similarly, -ap = a_p, which shows that the map E(k(C)) ---+ [(Clk) is a
homomorphism, hence an isomorphism. 0
Remark 3.11. It is important to observe that we can only add points

on [ if they lie on the same (non-singular) fiber. This enables us to add
two sections together, but it does not make the surface [ itself into a
group. Another way to say this is to use the notion of fiber product (Hart-
shorne [1, II §3]). The fiber product [ Xc [ of [ with itself relative to the
map 7f : [ ---+ C is the set of pairs (ZI' Z2) in the ordinary product [ X [
with the property that 7f(zr) = 7f(Z2); that is, [ Xc [ consists of all pairs
of points on [ which lie on the same fiber over C. It is a variety, and the
"group operation" is then the rational map
[ Xc [ ---t [
defined by addition on each (non-singular) fiber. One might say that c

is a group relative to the projection map 7f : [ ---+ C. We will see this
construction appearing in a much more general setting when we discuss
group schemes in the next chapter.
§4. Heights on Elliptic Curves over Function Fields
Let ElK be an elliptic curve defined over a function field. We have

proven (2.1) that the quotient E(K)/2E(K) is finite. We would like to
use the Descent Lemma [AEC VIII.3.1] to prove that E(K) is a finitely
generated group. This means we need a height function on E(K) that sat-
isfies certain properties. We begin by defining a height function on K and
then use it to define a height on E(K).
Definition. Let K = keG) be the function field of a non-singular algebraic

curve Glk. The height of an element f E K is defined to be the degree of
the associated map from G to pI,
h(f) = deg(f : G --+ pI).
In particular, if f E k, then the map is constant and we set h(f) = O.

Let ElK be an elliptic curve given by a Weierstrass equation
The height of a point P E E(K) is defined to be
h(P) _ { 0 if P = 0,
- hex) if P = (x, y).
(Note that h(f) is really the height relative to the field K, and h(P) depends
on the choice of a Weierstrass equation for E, although our notation does
not reflect this. See [AEC VIII §5].)
Remark 4.1. For each t E G, let
ord t : k(G)* ~ Z
be the normalized valuation on keG) [AEC II §1]; that is, ordt(f) is the
order of vanishing of the function f at the point t. Then [AEC II.2.6a]
implies that
h(f) = deg f = L max{ ordt(f), O} =L max{ - ordt(f), O}.

tEC tEC
This definition of the height as a sum of local values is analogous to the

definition of the height for number fields described in [AEC VIII §5]. Notice
that we can count either the total number of zeros of f or the total number
of poles.
The properties that we want the height to possess are of two very
different sorts. First, we want the height to satisfy certain transformation
properties relative to the group law on E. We will be able to prove this
below by a straightforward calculation that is very similar to the proof in
the number field case. Indeed, there is a theory of height functions for
a wide class of fields which includes number fields and function fields as
special cases. (See Lang [4] for details.)
The second property we require of the height is a finiteness property,
namely that a set of bounded height should contain only finitely many
points. In the case of a number field K, we first showed that this was true
of K itself; that is, a number field contains only finitely many elements of
bounded height [AEC VIII.5.11]. This immediately implied the same result
for E(K). However, matters are more complicated for function fields, since
a function field may have infinitely many elements of bounded height. For
example, the elements of height 0 in k(T) are precisely the elements of the
field k. More generally, the elements of height at most d in k(T) are the
rational functions of the form
ao + a1T + a2T2 + ... + ad Td
bo + bIT + b2T2 + ... + bdTd .
Thus it is not clear whether E( K) might possess infinite subsets of bounded

height. We will postpone further discussion of this question until the next
section.
The following proposition summarizes the principal geometric trans-
formation properties of the height.
Theorem 4.2. Let ElK be an elliptic curve defined over a function
field K.
(a) h(2P) = 4h(P) + 0(1) for all P E E(K).
(b) h(P + Q) + h(P - Q) = 2h(P) + 2h(Q) + 0(1) for all P, Q E E(K).
(The 0(1) bounds depend on the curve E and can be given explicitly; see
exercise 3.11.)
Remark 4.2.1. We will prove Theorem 4.2 using the triangle inequality
and elementary polynomial computations. Later we will give another proof
using intersection theory on a non-singular model for the elliptic surface c;
see (9.3).
PROOF (of Theorem 4.2). Fix a Weierstrass equation for E of the form
E : y2 = x 3 + Ax + B.
For any t E C, we write ordt(f) as usual for the order of vanishing of f E

k( C) at t. To ease notation and avoid the inevitable confusion caused by
repeated minus signs, we will also write
7rt(f) = (order of the pole of f at t) = - ordt(f).

Note that the (non-archimedean) triangle inequality for 7ft has the form
7ft(f + 1') :::; max{ 7ft(f), 7ft(f')}, with equality unless 7ft(f) = 7ft(f').
(a) Let P = (x, y) E E(K). The duplication formula [AEC III.2.3d] says
that x(2P) = ¢j1j;, where
¢=¢(x)=x4 -2Ax 2 -8Bx+A 2 and 1/J=1/J(x)=4x 3 +4Ax+4B.

We compute
h(2P) = h(x(2P)) definition of height

= 2....: max{ 7ft (¢/1/J) , o} from (4.4)
tEG
tEG tEG
So we need to show that max{7ft(¢),7ft(1/J)} and 4max{7ft(x),0} are ap-

proximately equal. If x has a large order pole at t, this is fairly clear,
since ¢ = X4 + . .. will then have a pole four times larger than x. On
the other hand, if x does not have a large pole, we will be in good shape
provided ¢ and 1/J don't both have large order zeros at t. In order to make
these vague comments precise, we define a quantity
and consider the following two cases.

17f t (x) > ILt (x has a large pole at t) I
The definition of JLt shows that in this case we have strict inequalities
7ft(x4) > max{ 7ft(2Ax2), 7ft(8Bx), 7ft (A2) },

7ft(x3) > max{ 7ft(Ax), 7ft (B) }.
It follows from the triangle inequality that 7ft(¢) = 7ft(x4) and 7ft(1/J)
7ft(x3). We also have 7ft(x) > JLt .:2: 0, which proves that
max{7ft(¢),7ft(1/J)} = 7ft(x4) = 4max{ 7ft(x),O}.
l7ft(x) :::; ILt (x has a small pole at t)1

The triangle inequality gives us trivial upper bounds
7ft(¢) :::; max{ 7ft(x4), 7ft(2Ax2), 7ft (8B:c) , 7ft(A2)}

:::; max{ 4JLt, 7ft (A) + 2JLt, 7ft (B) + (tt, 27ft (A)}
and
7rt('l/J) :::; max { 7rt(X3), 7rt(Ax), 7rt(B)}

:::; max{3J.tt, 7rt(A) + J.tt, 7rt(B)}
:::; 3J.tt·
Hence
max{ 7rt( 4», trt (1,b)} :::; 4J.tt :::; 4J.tt + 4 max { 7rt(x), O}.
In order to get a lower bound, we need to know that 4> and 1,b cannot
both vanish to high order at t. We define functions
q> = 12x2 + 16A, w= 3x 3 - 5Ax - 27 B, 6. = 4A 3 + 27B 2 ,

and observe that there is an identity
q> . 4> - w. 'l/J = 46..

(We've used this identity many times before, for example [AEC VIII.4.3].)
Note that the discriminant 6. = 4A 3 + 27 B2 E k( C) is not identically zero,
since E / k( C) is assumed to be non-singular. Intuitively, our assumption
that 7rt(x) :::; J.tt implies that q> and W have bounded poles at t, and then
the identity says that 4> and 1,b cannot both have high order zeros at t. More
precisely, we start with the upper bounds
7rt(q» :::; max{ 7rt(12x 2), 7rt(16A)} :::; 2J.tt,

7rt(w) :::; max{ 7rt(3x 3), 7rt(5Ax), 7rt(27B)} :::; 3J.tt.
Next, using the above identity, we find that
ord t (6.) = ord t (q>4> - w1,b) ~ min{ ord t (4)), ord t (1,b)} - max{ 7rt(q», 7rt(w)}
~ min {ordt ( 4», ord t (1,b)} - 3J.tt·
Now multiplying by -1 and using - ord t = trt yields
max{ 7rt(4)), 7rt('l/J)} ~ -3J.tt - ord t (6.)

~ -3J.tt - ord t (6.) - 4 (J.tt - max{ 7rt(x), O})
since J.tt ~ max{ 7r t(x),O}
= -7J.tt - ord t (6.) + 4 max{ 7rt(x), O}.
Combining the upper and lower bounds in this case, we obtain
-7J.tt - ord t (6.) :::; max{ 7rt(4)), 7rt('l/J)} - 4 max { 7rt(x), O} :::; 4J.tt.
In both cases we have now proven bounds of the form

where the quantities CI (t) and C2(t) have the property that they are inde-
pendent of the point P = (x, y) and are equal to zero for all but finitely
many t E C. Summing this inequality over t E C gives the desired estimate
CI ::; h(2P) - 4h(P) ::; C2
for certain constants Ci = ci(E) which do not depend on P.

(b) If P = 0 or Q = 0, the assertion is trivial, and if P = ±Q, then (b)
reduces to (a). So we will assume that P, Q -I 0 and P -I ±Q. We write
The condition that P -I ±Q ensures that the coordinates are all finite. The
addition formula [AEC III.2.3d] on the elliptic curve gives
_ (Y2 - YI)2 _ (A + XIX2)(XI + X2) + 2B - 2YIY2

X3 - - Xl - X2 - ,
X2 - Xl (X2 - Xt}2
X4= (Y2+YI)2 -XI-X2= (A+XIX2)(XI+X2)+2B+2YIY2.
X2 - Xl (X2 - XI)2
Next we compute
h(P + Q) + h(P - Q) = h(x3) + h(x4)

= Lmax{7rt(x3),O} +max{ 7rt(x4),O}
tEG
::; Lmax{7rt(x3X4),7rt(X3 + X4)'O},

tEG
where the last inequality needs some justification. In fact, for any func-
tions a, b, c, d E k( C)* we have
max { 7rt(a), 7rt(b)} + max{ 7rt(c), 7rt(d)} = max {7rt(ac), 7rt(ad + bc), 7rt(bd)}.
This is easily verified using the triangle inequality and checking the various
cases. For reasons which will become apparent in a moment, we will add
o = E 7rt ((Xl - X2)2) to both sides of this inequality, which yields
h(P + Q) + h(P - Q) ::; Lmax{7rt((XI - X2)2 X3X4 ),
tEG
7rt ((Xl - X2)2(X3 + X4)), 7rt ((Xl - X2)2)}.
Next we use a little algebra and the fact that the points P3 and P4 lie
on E to compute
(Xl - X2)2 x3X4 = (XIX2 - A)2 - 4B(XI + X2),

(Xl - X2)2(X3 + X4) = 2(XI + x2)(A + XIX2) + 4B.
Substituting in above and using the triangle inequality yields
h(P + Q) + h(P - Q)
~ L max { trt ((XIX2 - A)2 - 4B(XI + X2)),
tEe 7rt (2(Xl + X2) (A + XIX2) + 4B), 7rt ((Xl - X2)2) }
~ L max{ 7rt(xix~), 7rt(AXIX2), 7rt(A2), 7rt(BXl), 7rt(BX2),
tEe 7rt(AXl), 7rt(AX2), 7rt(XrX2), 7rt(XIX~),
7rt(B), 7rt(xi), 7rt(XIX2), 7rt(X~)}
~ L(2max{7rt(Xl),O} +2max{ 7rt(X2),O}
tEe +2max{7rt(A),o} +max{7rt(B),O})
= 2h(XI) + 2h(X2) + 2h(A) + h(B)
= 2h(P) + 2h(Q) + 0(1).
It remains to prove an inequality in the opposite direction. It is possi-
ble to do this directly, as is done, for example, in [AEC VIII.6.2]. But we
will instead use the following clever trick which is due to Don Zagier. We
have proven that the inequality
2h(P) + 2h(Q) ~ h(P + Q) + h(P - Q) + 0(1)
holds for all P, Q E E(K). Given two points P', Q' E E(K), we apply this
identity with P = pI + Q' and Q = P' - Q' and then use (a) to obtain
2h(P' + Q') + 2h(P' - Q') ~ h(2P') + h(2Q') + 0(1)

= 4h(P') + 4h(Q') + 0(1).
Dividing by 2 gives the opposite inequality, which completes the proof
of (b). D
Just as in the number field case, we can construct a canonical height

which is a quadratic form on the group E(K). However, in order to prove
that the canonical height is non-degenerate, we need to know that sets of
bounded height are finite. The exact conditions under which this occurs
will be described in the next section. We have included the construction of
the canonical height here, since it seems to fit in better with the material
in this section. We hope the reader will excuse this textual non-linearity.
Theorem 4.3. Let E / K be an elliptic curve defined over a function
field K = k(C).
(a) For every point P E E(K), the limit
exists. The quantity h( P) is called the canonical (or N eron- Tate) height
of P.
(b) The canonical height has the following properties:
(i) h(P) = ~h(P) + 0(1) for all P E E(K).
(ii) h(mP) = m 2 h(P) for all P E E(K) and all mE Z.
(iii) h(P + Q) + h(P - Q) = 2il.(p) + 2h.(Q) for all P, Q E E(K).
(c) The canonical height is a quadratic form on E(K). In other words,
h(-P) = h(P), and the pairing
( . , . ) : E(K) x E(K) -+ lR
(P, Q) = h(P + Q) - i~(p) - h(Q)
is bilinear. (N.B. The pairing is normalized so that h(P) = ~ (P, P).)

(d) Assume that E does not split over k. (This means that E is not K-
isomorphic to an elliptic curve defined over k. See §5 for more details.)
Then h(P) 2:: 0, and
h(P) = 0 if and only if P is a point of finite order.
(e) Any function E( K) -+ lR which satisfies (b )(i) and (b )(ii) for some
integer m 2:: 2 is equal to the canonical height.
PROOF. We will just briefly sketch the proof, since it is exactly the same
as in the number field case [AEC VIII.9.1, VIII.9.3]. For any integers n 2::
m 2:: 0 we have
L
T/'-1
14- nh (2 n p) - 4-mh(2rnp) I =I 4- i - 1 h(2 i + 1 P) - 4- i h(2 i P) I
n-I
i=m
n-I
::; L 4- i - 1 0(1) from (4.2a)
i=rrt
00
i=rn
This shows that the sequence 4-7) h(2n P) is Cauchy, hence converges, which
proves (a). Further, taking m = 0 and letting n -+ oc gives
12h(P) - h(p)1 ::; 0(1),
which is (b)(i).
N ext we apply (4.2b) to the points 2nP and 2nQ to obtain
Dividing by 4n and letting n ---> 00 gives (b)(iii). Evaluating (b) (iii) at P =

mQ yields the identity
h((m + l)Q) + h((m - 1)Q) = 2h(mQ) + 2h(Q).
Taking m = 0 gives h( -Q) = h(Q), and then an easy induction (up and
down) on m gives (b)(ii). This completes the proof of (b).
It is a standard computation to show that a function satisfying the
parallelogram law (b)(iii) is a quadratic form; see for example the proof
of [AEC VIII.9.3c]. This gives (c).
If P E E(K) has finite order, then 2n P takes on only finitely many
values, so it is obvious from the definition that h(P) = O. Conversely,
suppose that h(P) = o. Then for all m E Z we use (b)(ii) and (b)(i) to
compute
h(mP) = 2h(mP) + 0(1) = 2m2h(P) + 0(1) = 0(1).

It follows that {mP : m E Z} is a set of bounded height. We will prove in
the next section (5.1) that if E does not split over k, then sets of bounded
height are finite. Hence P is a point of finite order, which completes the
proof of (d).
Finally, suppose that 9 : E(K) ---> lR satisfies (b) (i) and (b)(ii) for
some m 2: 2. Then we compute for every P E E(K) and every i 2: 1,
2h(P) - 2g(P) = m- 2i (2h(m i p) - 2g(mip))

= m- 2i ((h(m i p) + 0(1)) - (h(mip) + 0(1)))
= 0(m- 2i ).
Letting i ---> 00 shows that h(P) = g(P). D
Remark 4.3.1. It is also possible to construct the canonical height using

intersection theory on the minimal elliptic surface associated to E. See §9
for details, especially (9.3). One consequence of the geometric construction,
which is not at all evident from the definition, is that for function fields
the canonical height h(P) is always a rational number. This is (probably)
false for number fields, where it is conjectured that h(P) is transcendental
for all non-torsion points.
§5. Split Elliptic Surfaces and Sets of Bounded Height
We proved in the last section that the height function h : E(K) ----+ lR.
behaves nicely with respect to the group law on E. In order to prove
that E(K) is finitely generated, it remains to show that sets of bounded
height in E(K) are necessarily finite. The reader will recall that in the
case of number fields, this was comparatively easy to do. Unfortunately,
for function fields it is easy to construct a counterexample to this assertion!
For example, let Eo I k be an elliptic curve, let E = Eo x C be the elliptic
surface with E ----+ C being projection onto the second factor, and let ElK
be the corresponding elliptic curve over K. Then every point, E Eo(k)
gives a section
(J', : C ---+ E = Eo X Co
and this section corresponds to a point P, E E(K). Clearly, distinct ,'s

give distinct P, 's, and just as clearly the map
Eo(k) ---+ E(K),
is a homomorphism. It follows that E(K) cannot possibly be finitely gen-

erated, since the fact that k is algebraically closed means that Eo (k) is not
finitely generated. (For example, if k = re, then Eo(k) ~ iLl A for some
lattice A c IC.)
It will turn out that this is the only way in which E(K) can fail to be
finitely generated, which suggests that we make the following definition.
Definition. An elliptic surface E ----+ C splits (over k) if there is an elliptic
curve Eo I k and a birational isomorphism
i: E ~ Eo x C
such that the following diagram commutes:
Eo xC
71' " " ,/ proiz
There are several other ways of characterizing split elliptic surfaces.

The following one will be used later in this section. For others, see exer-
cises 3.9 and 3.10.
Proposition 5.1. Let E ----+ C be an elliptic surface over k, and let E / K
be the corresponding elliptic curve over the function field K = k( C). The
following are equivalent:
(i) The elliptic surface c ---+ C splits over k.

(ii) There is an elliptic curve Eo/k and an isomorphism E ...:::..... Eo defined
over K.
PROOF. Suppose first that 7r : e ---+ C splits. This means that there is a
birational isomorphism i : e ---+ Eo x C so that proh 0 i = 7r. A dominant
rational map induces a corresponding map on function fields (3.7), so we
obtain an isomorphism k(c) ~ k(Eo x C) which is compatible with the
inclusions
keG) '--> k(c) and keG) '--> k(Eo x G).
In other words, if we let K = keG) as usual, then the fields k(c) = K(E)
and k(Eo x G) = K(Eo) are isomorphic as K-algebras. Each of them is
a field of transcendence lover K, so each corresponds to a unique non-
singular curve defined over K (see [AEC, II.2.5] or Hartshorne [1, 1.6.12]).
In other words, there is an isomorphism E ~ Eo defined over K. This
completes the proof that (i) implies (ii).
Conversely, suppose that we are given an elliptic curve Eo/ k and an
isomorphism E ...:::..... Eo defined over K. Then K(E) ~ K(Eo) as K-
algebras, which is the same as saying that k(e) ~ k(Eo x C) as k(C)-
algebras. Again using (3.7), this isomorphism of fields induces a birational
isomorphism of varieties c ---+ Eo x G commuting with the maps to G, which
shows that e ---+ G is split over k. Hence (ii) implies (i), which completes
the proof of (5.1). 0
Example 5.2. Note that the isomorphism in (5.lii) is not required to be

defined over the constant field k. In fact, since E is only defined over K,
it really only makes sense to talk about maps being defined over K . For
example, take C = pi and K = k(T), and consider the elliptic surfaces
Cl : y2 = x3 + 1, C2 : y2 = x3 + T 6 ,
C3 : y2 = x 3 + T, e4 : y2 = x 3 + X + T.
Also let Eo/k be the elliptic curve
Eo : y2 = x 3 + 1.
Then Cl is clearly split over k, since it is precisely Eo x G. The surface e2
also splits over k, as can be seen from the isomorphism
The elliptic surface e3 does not split over k, although it will split if we
replace the base field k(T) by the larger field k(T 1 / 6 ). Finally, C4 does not
split over k; and since its j-invariant is non-constant, it will still not split
even if we replace k(T) by a finite extension. See exercises 3.9 and 3.10 for
general statements.
Remark 5.3. For a number field K, it was not hard to show that there are
only finitely many elements of K having bounded height [AEC VIII.5.11],
which immediately gave the same result for E(K). Unfortunately, this
assertion is clearly false for function fields, since there may be infinitely
many maps C -> 1P'1 of any given degree. In other words, for number fields
we proved that a set of bounded height
{p E E(K) : h(P):::; d}
is finite by reducing to an assertion about elements of bounded height in K.

But for function fields we will need a new sort of argument which makes
use of the fact that the coordinates of P = (x, y) satisfy the equation of
an elliptic curve. Further, we need to rule out split elliptic curves, since
they will have infinitely many points of bounded height. All of this will
help to explain why the proof of the following result is far from trivial and
requires techniques different from those used when studying number fields.
The proof will take us the rest of this section.
Theorem 5.4. Let c -> C be an elliptic surface over an algebraically
closed field k, let E / K be the corresponding elliptic curve over the function
field K = k( C), and let d be a constant. If the set
{p E E(K) : h(P) :::; d}
contains infinitely many points, then c splits over k.

PROOF (of Theorem 5.4). We will divide the proof of Theorem 5.4 into
two steps. The first step says that if c has infinitely many sections of
bounded degree (i.e., E(K) has infinitely many points of bounded height),
then there is a one-parameter family of such sections. The second step says
that if there is a one-parameter family, then c splits. We begin with the
existence of the family.
Proposition 5.5. Under the assumptions of Theorem 5.4, there is a (non-
singular projective) curve r / k and a dominant rational map ¢ : r x C -> c
such that the following diagram commutes:
1>
fxC
PROOF. We fix a Weierstrass equation for E / K of the form
with A, BE K = k(C),
and we define a set

E(K, d) ~f {p E E(K) : h(P) ::; d}.
Our assumption is that E(K, d) is infinite, and we wish to use this fact to
find an appropriate curve r and map r x C --. e
The first step is to parametrize the set of maps from C to lP'2. Recall
that if D is a divisor on C, then L(D) is defined to be the space ofrational
functions
L(D) = {J E k(C) : div(f) + D 2: a}.
This is a finite dimensional vector space whose dimension is denoted by
C(D). (For basic facts about L(D), see [AEC, II §5].) Taking three functions
from L(D) will define a map from C to lP'2, so we get a natural map
L(D)3 ,,{O} -----. Map( C, lP'2),
(Fa, F I , F2 ) f--+ (t f-> [Fo(t), FI(t), F2 (t)]).
Multiplying (Fa, F I , F2 ) by a scalar clearly gives the same map C --. lP'2, so
we actually have an association
lP'3f(D)-I ~ L(D):." {O} -----. Map(C,lP'2).
The key here is that we have taken a collection of maps in Map( C, lP'2)
and have parametrized this collection using the points of the algebraic
variety lP'3R-l, where to ease notation we will write Cfor C(D). Some of these
maps C --. lP'2 will actually correspond to elements of E(K). The next step
is to show that the maps corresponding to E(K) form an algebraic subset
of lP'3R-l .
For simplicity, we will assume henceforth that D 2: 0, and we fix a
basis h, ... , Ie for L(D). Further, we choose a divisor D' 2: 3D large
enough so that 1, A, B E L(D' - 3D), and we let hI, .. " hr be a basis
for L(D').
Every element in L(D)3 can be written uniquely in the form
f f R
F = (Fa,Fb,Fc) = (I:adi,I:bdi,I:cdi)'
i=1 i=1 i=1
Such an F will give an element of E(K) if and only if Fa, Fb, Fe satisfy the
homogeneous equation of E,
F;Fc = F~ + AFaF; + BF:.

In other words, F will give an element of E(K) if
(I: bdi) 2(I:Cdi) = (I: adi) 3+ A (I:adi ) (I: Cdi) 2

+ B (I: cdi )3.
Multiplying this out gives a sum involving monomials of the form fdjik
and Afdjik and Bfdjfk. Our choice of D' ensures that each of these
monomials is in L(D'), hence can be written uniquely as a linear com-
bination of h 1 , ... , hr. So finally we end up with an equation that looks
like
r
2..: <Pi (a, b, c)h i = 0,

;=1
where each <Pi is a homogeneous polynomial in the coordinates
Now the maps C ---+ jp'2 from above which correspond to elements of E(K)
are associated to the points of the variety
VD ~ {[a, b, c] E jp'3f(D)-1 : <Pi(a, b, c) = 0 for all 1 :::; i :::; r}.

Example 5.5.1. We briefly interrupt the proof of Proposition 5.5 to pre-
sent an example. If we take C = jp'1 and K = k(T), then E / K has a
Weierstrass equation of the form
y2 = x 3 + A(T)x + B(T) with A, BE k[T].
We let D = n(=), which means that L(D) is the set of polynomials in k[T]
of degree at most n. (Here £(D) = n + l.) The corresponding family of
maps jp'1 ---+ jp'2 is parametrized by jp'3n+2 as described above,
Map(jp'1, jp'2),
[ao, ... , an, bo,·· ., bn , Co,···, cn ] [t

;=0
ai Ti , t
i,=O
biTi , t
i=O
ci Ti ].
Writing A(T) = I: Ai Ti and B(T) = I: BiTi , this map jp'l ---+jp'2 will give
an element of E(K) if it satisfies the equation
(2..:biTif (2..:ciTi) = (2..: a ;Ti f
+ (2..: Ai Ti ) (2..:aiTi) (2..:ciTif + (2..: Bi Ti ) (2..:ciTif·
Multiplying this out and writing it as a polynomial in T gives a formula

that looks like r
2..: <Pi (a, b, c)Ti = 0,
,i=O
and the system of homogeneous equations <Po = <PI = ... = <PI' = 0 defines
the variety VD.
To illustrate how this procedure works in practice, we will consider the

elliptic curve
E : y2 = X 3 + T 2 x - 1.
We will find all points of degree at most 1 by setting
and substituting P into the (homogenized) equation for E. Multiplying

everything out, we obtain the equation
a1ciT5 + (2a1cOc1 + aoci)T4 + (ai + a1c~ - bic1 + 2aOCOC1 - cr)T3

+ (3aoai - bico + aoc~ - 2bob1C1 - 3eocDT2
+ (3a~a1 - 2bob1Co - b~C1 - 3c~cdT + (a~ - b~co - c~) = 0.
°
Setting the coefficients equal to gives six homogeneous equations for the
six variables ao, ... , C1' These six equations define the variety VD C ]P'5.
After some work one finds that VD consists of the 3 lines
{[O, 0, u, v, 0, OJ} U {[O, 0, iu, iv, u, vJ} U {[O, 0, -iu, -iv, u, vJ}
and the 22 isolated points
{[0,(b(2,0,1,OJ: a=(~=-1}u{[p2,0,0,p,1,OJ: p6=1}

U{[-2 p2,(,3(p3,2p,1,OJ: (2=_1, p6=1}.
Note that although VD itself is not zero-dimensional, its image in E(K)

consists of a finite set of points, namely
where (, (1, (2, and p satisfy (2 = (f = (~ = -1 and p6 = 1. For

example, all of the points on the line [0,0, iu, iv, u, vJ in VD are mapped
to the single point [0, iu + ivT, u + vTJ = [0, i, 1] E E(K). We also observe
that ofthese 24 points in E(K), only the 3 points [0,1, OJ and [1, ±T, 1] are
in E(Q(T)).
We now resume the (regularly scheduled) proof of Proposition 5.5.
Recall that we have constructed a variety VD and a map VD ---- E(K). The
following lemma shows that if deg(D) is sufficiently large, then the image
of VD in E(K) will contain E(K, d).
Lemma 5.5.2. Let g be the genus of the curve C. If
5 1
degD ~ g + "2d + "2 (deg A + degB),
then the image of VD in E(K) contains E(K, d). (For a more accurate
estimate, see exercise 3.13')
PROOF (of Lemma 5.5.2). This is an exercise using the Riemann-Roch

theorem. Let P = (xp, yp) E E(K, d), so, by definition, deg(xp) = h(P) :::::
d. Using the Weierstrass equation for E, we can check that deg(yp) is also
bounded,
2deg(yp) = deg(x~ + Axp + B) ::::: 3deg(xp) + deg(A) + deg(B) ,

31'
deg(yp)::::: "2d+ "2 (deg(A) +deg(B)).
In order to prove the lemma, we need to find functions F o, F l , F2 E L(D)

such that (xp,yp) = (Fo/F2,FdF2)'
Recall that any function f E K defines a map f : C ___ jp'l, and the
divisor of f has the form
div(f) = divo(f) - divoo(f) = 1*((0)) + 1*((00)),
where divo (f) and div 00 (f) are the divisors of zeros and poles of f respec-
tively. (See [AEC 11.3.5].) We also note from [AEC, I1.3.6(a)] that
deg(f) = deg(divo(f)) = deg(divoo(f)).
We are going to apply the Riemann-Roch theorem to the divisor
whose degree we estimate as
deg(D") = deg(D) - deg(xp) - deg(yp)

;::: deg(D) - d - (~d + ~(deg(A) + deg(B))) ;::: g.
The Riemann-Roch theorem [AEC, 11.5.4] then tells us that
£(D") ;::: deg(D") - 9 + 1 ;::: 1,
so there exists a non-zero function F E L(D"). We claim that the three

functions
Fo = Fxp,
are all in L(D), which will complete the proof of the lemma.
To check this last assertion, we use the fact that div(F) + D" ;::: 0 and
compute
div(Fxp) +D = div(Fxp) + D" + divoo(xp) + divoo(Yp)

= div(F) + D" + divo(xp) + divoo(Yp) ;::: 0,
div(Fyp) +D = div(Fyp) + D" + divoo(xp) + divoo(Yp)
= div(F) + D" + divoo(xp) +divo(Yp) ;::: 0,
div(F) +D = div(F) + D" + divoo(xp) + divoo(Yp) ;::: o.
This completes the proof of Lemma 5.5.2. o

Continuing with the proof of Proposition 5.5, we fix a divisor D ;::: 0
whose degree is large enough so that we can apply (5.5.2). Then (5.5.2)
and our assumption that E(K, d) is infinite tell us that the image of VD
in E(K) is infinite.
We now change perspective a bit and consider the associated elliptic
surface e ----> C. We have assigned to each point 'Y E VD a point P"{ E E(K),
and (3.10c) says that the point P"{ corresponds to a section u"{ : C ----> e. In
this way we get a natural rational map
¢ : VD x C -----> e, (,,(, t) ~ u"{(t).
It is clear that ¢ is an algebraic map, since using notation from above, we

see that ¢ can be written as
If there exists an irreducible curve f C VD such that the map
¢ :f x C -----> e,
is dominant, then the proof of Proposition 5.5 will be complete. So we
assume that ¢ : f x C ----> e is not dominant for every irreducible curve f c
VD and derive a contradiction.
Let f C VD be an irreducible curve. We are assuming that ¢ : f x C ---->
e is not dominant, so the image ¢(f x C) has dimension at most one.
However, we know that 7f(u"{(t)) = t, which shows that 7f maps ¢(f x C)
onto C. It follows that the image ¢(f x C) must have dimension exactly
one. Further, the product f x C is irreducible, so ¢(f x C) is also irreducible
by (3.5a). Therefore ¢(f x C) must consist of a single irreducible curve. On
the other hand, for any given 'Y E f the map u"{ is a section to 7f : e ----> C,
so we have
is already an irreducible curve contained in ¢(r X C). Hence ¢( b} x C)

must be equal to ¢(r x C) for every "Y E r. Equivalently, every "Y E r gives
the same section a'Y' which means that the image of r in E(K) consists of
a single point.
We have now shown that every irreducible curve r c VD maps to a
single point in E(K). But any two points in any connected component
of VD can be linked by a connected chain of irreducible curves. (In fact,
on any irreducible component of VD , any two points can be connected by
a single irreducible curve; see exercise 3.14.) Since VD has only finitely
many connected components, it follows that the image of VD in E(K)
is finite. This contradicts the fact shown above that the image of VD
contains E(K, d). Hence there exists an irreducible curve r c VD with
the property that ¢ : r x C -> C is dominant. Replacing r with a non-
singular model for r (see Hartshorne [1, 1.6.11]) completes the proof of
Proposition 5.5. 0
The following proposition, taken together with (5.5), completes the

proof of Theorem 5.4.
Proposition 5.6. Let 7r: C-> C be an elliptic surface over k, let r/k be
a non-singular projective curve, and suppose that there exists a dominant
rational map ¢ : r x C -> C so that the following diagram commutes:
¢
rxc
C
Then c splits.
PROOF. The fact that ¢ is a dominant map of varieties of the same di-
mension means that there is a non-empty Zariski open subset cO of cover
which ¢ is a finite map, say of degree m. Let to E C be a point such that
the fiber Cto is non-singular and such that ¢ is well-defined at every point
of r x {to}. Note that the set of such to's is a non-empty Zariski open
subset of C, since Ct is non-singular for all but finitely many t E C, and ¢
is well-defined except at finitely many points of r x C by (3.5b). To ease
notation, we let Eo = Ct o ' Note that Eo/ k is an elliptic curve.
We define a map
'lj;: Eo x C,
m
((x,y),t) f----+ (L¢("Yi,tO),t)

i=l
where the points "Yi are determined by the formula

m
¢* ((x, y), t) = L("Yi, t).
i=l
In other words, 'ljJ takes a point on eO, pulls it back by ¢ to get a collection
of m points on r x C (counted with multiplicity), changes the t-coordinate
to to to get a collection of points on r x {to}, uses ¢ to push them forward
to a collection of points on Eo, and finally uses the group law on Eo to add
them up.
Note that the map 'ljJ is a well-defined rational map on eO. This is
true despite the fact that the definition of 'ljJ involves applying ¢-I, since
ultimately we take a symmetric expression of the points in ¢-l((X,y),t),
so the resulting point can be expressed as a rational combination of x, y, t.
This is clearest for those points ((x,y),t) E eO for which ¢-l((x,y),t)
consists of m distinct points, which suffices for our purposes since we only
need to define 'ljJ on an open subset of eO.
(Aside: An alternative description of 'ljJ is as follows. Let r(m) be
the m-fold symmetric product of r. Then the map ¢ defines in a natural
way a morphism from eO to r(m) x C which sends a point in eO to the
collection of points in its inverse image. Next we use the map ¢( . ,to) : r --+
Eo to map r(m) xC --+ Egm ) x C. Finally, the summation map Egm ) --+ Eo
using the group law on Eo gets us to Eo x C, and the composition of all
these maps is 'ljJ : eO --+ Eo x C. For information about the symmetric
product, see Harris [1, Lecture 10, especially 10.23].)
Note that if 'ljJ were a (birational) isomorphism, we would be done.
Unfortunately, there is no reason that this should be true. We begin our
analysis of the map 'ljJ by computing it on the fiber over to.
'ljJ((x,y),to) =( L ¢(!"to),to) = (m(x,y),to).

h,t')E.p* «x,y),to)
Thus 'ljJ : eto --+ Eo x {to} is just the multiplication-by-m map on Eo.
In particular, since the multiplication-by-m map is surjective, we see that
'ljJ(e O) contains Eo x {to}. This implies that the rational map e --+ Eo x C
is dominant, since otherwise the irreducibility of 'ljJ( eO) would imply that
'ljJ(e O) = Eo x {to}, contradicting the fact that 'ljJ(e O) maps onto C (Le.,
'ljJ(e O) must contain at least one point on each fiber on r x C --+ C).
We now consider the elliptic curve E / K associated to the elliptic sur-
face e. We also take the elliptic curve Eo/k and think of it as the el-
liptic curve Eo/ K associated to the split elliptic surface Eo x C. Then
the dominant rational map 'ljJ : e --+ Eo x C defined above corresponds
to a non-constant map E --+ Eo of elliptic curves over K (3.9). Any such
map can be written as the composition of a translation followed by an
isogeny [AEC, 11104.7], so we obtain a non-zero isogeny
A:E -----+ Eo
defined over K. Taking the dual isogeny [AEC, III §4] gives a map 5. : Eo --+
E defined over K, and this induces an isomorphism [AEC, 111.4.11,111.4.12]
5. : Eo/ ker(5.) ~ E.
Now ker('x) is a finite subgroup of Eo(k), so the fact that Eo is defined

over k implies that ker('x) C Eo(k). To see why this is true, take a Weier-
strass equation for Eo with coefficients in k. Then the n-torsion points of E
have coordinates which are roots of certain polynomials having coefficients
in k, so E[n] C E(k). (For explicit formulas, see [AEC, exercise 3.7]. Note
we are assuming that k is algebraically closed.) It follows that the elliptic
def
curve El = Eo/ ker(>.) is defined over k.
A
We have now produced an elliptic curve Ed k and an isomorphism of

elliptic curves El ----t E defined over K. It follows from (5.1) that E/ K splits
over k. This completes the proof of Proposition 5.6 and, in conjunction with
Proposition 5.5, also completes the proof of Theorem 5.4. 0
§6. The Mordell-Weil Theorem for Function Fields
We have now assembled all of the tools needed to prove the following im-
portant result.
Theorem 6.1. (Mordell-Weil Theorem for Function Fields) Let e ----t C
be an elliptic surface defined over a field k, and let E / K be the correspond-
ing elliptic curve over the function field K = k(C). If e ----t C does not split,
then E(K) is a finitely generated group.
PROOF. Suppose first that k is algebraically closed. The weak Mordell-

Weil theorem (2.1) tells us that the quotient group E(K)/2E(K) is finite.
Next let h : E(K) ----t Z be the height function defined in §4. This height
function satisfies
(i) h(P + Q) = 2h(P) + 2h(Q) - h(P - Q) + 0(1)

~ 2h(P) + OQ(l) for all P, Q E E(K),
(ii) h(2P) = 4h(P) + 0(1) for all P E E(K),
(iii) {P E E(K) : h(P) ~ C} is finite.
The first two statements are (4.2a,b), whereas the third statement is (5.4)
and uses the assumption that e ----t C does not split. We now have all of the
hypotheses needed to apply the Descent theorem [AEC, VIII.3.1], which
completes the proof that E(K) is finitely generated under the assumption
that k is algebraically closed. Finally, for arbitrary constant fields k, it suf-
fices to observe that E(K) = E(k(C)) is a subgroup of E(k(C)), so E(K)
is finitely generated. 0
Remark 6.2.1. If C ----t C splits over k, then the group E(K) need not be
finitely generated. More precisely, if C ~ Eo x C, then each point Z E Eo(k)
is also a point in Eo(K) = E(K). Thus there is an inclusion Eo(k) '-+

E(K). So for example, if k = C, then Eo(k) will certainly not be finitely
generated, and the same is then true of E(K). However, the quotient
group E(K)j Eo(k) will be finitely generated. This relative version of the
Mordell-Weil theorem is due to Lang and Neron; see exercise 3.15.
Remark 6.2.2. If k is a number field, or more generally if k is a finitely
generated extension of Q, then the Mordell-Weil theorem (6.1) is true re-
gardless of whether or not c -; C splits. This generalization of the original
Mordell-Weil theorem for number fields is due to N eron; see exercise 3.4.
§7. The Geometry of Algebraic Surfaces
All of our previous work in this chapter has dealt with the birational ge-
ometry of elliptic surfaces. In order to investigate the finer structure of
elliptic surfaces, we will need to study them up to isomorphism. This sec-
tion reviews the basic theory of non-singular algebraic surfaces, including
especially intersection theory and minimal models. Our main reference will
be Chapter 5 of Hartshorne [1], specifically §1 for intersection theory and
section 5 for the theory of minimal models, although we will also need a
few additional facts from other sources concerning minimal models. For
more information about surfaces, the reader might consult Beauville [1]
and Griffiths-Harris [1, Ch. 4].
Let Sjk be a non-singular surface defined over an algebraically closed
field k of characteristic O. A divisor on S is a formal sum
n
D = Laifi,
i=1
where ai E Z and the fi C S are irreducible curves lying on the surface S.

The f i 's that appear in the sum are called the components of the divisor D.
The group of divisors on S is denoted Div(S).
Recall that the local ring of S at a point PES is defined to be
Os,P = {J E k(S) : f is defined at p}.

Similarly, for any irreducible curve f C S, the local ring of S at f is
Os,r = {J E k(S) : f is defined at some point PEr} = U Os,p.

PEr
Our assumption that S is non-singular implies that each Os,r is a discrete

valuation ring. We denote its valuation by ordr, since intuitively ordr(J)
is the order of vanishing of f along f. We extend this as usual to a homo-

morphism
ord r : k(S)* ----; Z,
and then use this to define a homomorphism
div : k(S)* ----; Div(S),
f f---> L
ordr(J)f.
reS
A divisor is principal if it is the divisor of a function dive!). Two divi-
sors D 1, D2 E Div(S) are linearly equivalent if their difference D1 - D2 is
principal, in which case we write D1 ~ D 2. Linear equivalence is an equiv-
alence relation on the divisor group Div(S), and the Picard group of S is
the corresponding quotient group,
Pic(S) = Div(S)/ ~.
Example 7.1. Consider a divisor D = L aifi E Div(]P2) in the projective
plane. To each irreducible curve fi C ]p2 we can associate its degree, and
by extending linearly we obtain a homomorphism
deg : Div(]P2) ----; Z,
L aifi f---> L ai deg(fi).
If D = div(J) is principal, then it is easy to check that deg(D) = 0, so the
degree map induces a homomorphism
deg: Pic(]P2) ----; Z.
We will leave it to you (exercise 3.18) to verify that this last map is an
isomorphism, so Pic(]P2) ~ Z. This is the analogue for surfaces of [AEC,
II.3.1bj. It says that a rational function on ]p2 has the "same number" of
zeros and poles.
In order to study the geometry of a surface, we will look at the curves it
contains and how those curves intersect. For example, a curve of degree m
and a curve of degree n in ]p2 will "usually" intersect in mn distinct points,
and they will always intersect in mn points if we count tangencies and
singularities with the correct multiplicities. This famous result is known as
Bezout's theorem; see (7.3) below. Our next step is to define an intersection
index for curves and divisors on arbitrary non-singular surfaces.
Let fl and f2 be irreducible curves on S, and let P E f1 n f 2. Fix
local equations iI, 12 E k(S)* for f 1 , f2 around P; that is, choose li E Os,P
so that ord ri (Ji) = 1 and ordr(Ji) = 0 for every other irreducible curve f
containing P. We say that f1 and f2 intersect transversally at P if iI
and 12 generate the maximal ideal of the local ring Os,p. (See exercise 3.19
for the intuition behind this definition.)
If f 1 and f 2 are irreducible curves that meet everywhere transversally,
then it is natural to define the intersection f1 . f2 to be the number of
intersection points. The next theorem says that this definition can be
extended in a natural way to all divisors.
Theorem 7.2. There is a unique symmetric bilinear pairing

Div(X) x Div(X) --+ Z,
with the following two properties:
(i) If r 1 and r 2 are irreducible curves on S that meet everywhere transver-
sally, then r l . r 2 = #(r 1 n r2)'
(ii) If D, D I , D2 E Div(S) are divisors with DI '" D 2, then D·D I = D·D 2·
PROOF. (Sketch) Given divisors D I , D2 E Div(S), one uses ampleness and
a Bertini theorem to find divisors D~, D~ E Div( S) with D~ '" D I , D~ '"
D 2, and such that D~ and D~ are sums of irreducible curves that meet each
other transversally. Then D~ . D~ is defined using linearity and (i). One
then checks that the answer is independent of the choice of D~ and D~.
For details, see Hartshorne [1, V.l.1]. 0
Example 7.3. We have seen ((7.1) and exercise 3.18) that the degree
map defines an isomorphism Pic(]P2) ~ Z. Let r l ,r 2 C ]p2 be curves of
degrees nl,n2 respectively, and let HI,H2 C]p2 be (distinct) lines. Then
deg(r i ) = ni = deg(niHi)' which implies that r i '" niHi.
Further, HI . H2 = 1, since distinct lines in ]p2 intersect transversally in a

single point, so we can compute
r l . r2 = (nIH I ) . (n2H2) = nIn2(HI . H 2) = nIn2 = deg(rI) deg(r2)'
The equality r l . r2 = deg(rI) deg(r2) is called Bezout's theorem for the
plane.
Theorem 7.2 is a powerful existence theorem, but it does not give a
very practical method for computing the intersection DI . D 2 . In principle,
one can find divisors D~ '" DI and D~ '" D2 which intersect transversally
and then count the number of points in D~ n D~, but in practice it is
better to assign multiplicities to the points in DI n D 2 . This is done in the
following way.
Let D E Div(S) be a divisor, and let PES. A local equation for D
at P is a function f E k(S)* with the property that
P ~ D - div(f).
Notice that if D = r is an irreducible curve, then this is equivalent to the
condition that ord r (f) = 1 and that ord r , (f) = 0 for all other irreducible
curves r' containing P.
Now let D 1 , D2 E Div(S) be divisors, and let PES be a point which
does not lie on a common component of Dl and D 2 . Choose local equa-
tions II, h E k(S)* for D I , D2 respectively. The (local) intersection index
of Dl and D2 at P is defined to be the quantity
(DI . D 2)p = dimk r:J s ,P/(IIJ2)'
Notice that (D1 . D 2 )p = 0 if P ~ D1 n D 2 , since if P ~ D i , then Ii = 1

will be a local equation for Di at P. The next result explains how the
local intersection indices can be used to calculate the global intersection
number D1 . D 2 .
Proposition 7.4. Let DI, D2 E Div(S) be divisors with no common
components. Then the local intersection index (D1 . D 2)p is finite for
all PES, and
D1 . D2 = L
(D 1 · D 2 )p.
PED,nD 2
PROOF. See Hartshorne [1, V.1.4l. D
Example 7.5. The local intersection indices are comparatively easy to

calculate. As illustration, we will compute the intersection index of
f 1 :y 2Z=X 3 and f 2 :YZ=X 2
in ]p>2 at the point P = [0,0, ll. (See Figure 3.1.) We dehomogenize x =
XjZ, Y = YjZ, so the local ring at Pis
OIP'2,P = k[x,yl(o,o) = {i E k(x,y) : g(O,O) i= O}.

Then
where the middle equality follows from the fact that x -1 is a unit in k[xlo.
Hence
k[x,yl(o,o) (k k
(f l ' f) d·
2 P = Imk (2
y -x ,y
d'
3 ) = Imk + x + k x 2 + k x 3) = 4.
Notice that f1 is singular at P, but it has a unique tangent line there which
is the same as the tangent line to f2 at P. This explains why (fl' f2)P is
so large.
Remark 7.6. Proposition 7.4 gives a method for computing the inter-
section number D1 . D2 when the divisors D1 and D2 have no common
components. However, one frequently wants to compute the intersection of
divisors with components in common. An important example is the self-
intersection D2 = D . D of a divisor D. This cannot be calculated directly
using (7.4). One approach to calculating D2 is to find a D' '" D such
that D and D' have no common components, and then compute D . D'.
For example, if f c ]p>2 is a curve of degree n, then the computation in (7.3)
shows that r 2 = n 2 . The argument in (7.3) works for self-intersections be-
cause r '" nH for any line H C ]p>2. In general, it may be difficult to find
an appropriate D'. Another approach to computing self-intersections is to
use the adjunction formula; see Hartshorne [1, V.1.5l.
P = (0,0)
The Curves r1: y2 = x3 and r2 : Y = x2 Near P = (0,0)

Figure 3.1
If C is a (possibly singular) curve, then there is a unique non-singular

projective curve C' such that C is birationally equivalent to C'. (See
[AEC, II.2.5] or Hartshorne [1, 1.6.11,1.6.12].) The situation for surfaces
is more complicated. It is true that every surface is birationally equivalent
to a non-singular projective surface. However, if 5 is a non-singular pro-
jective surface, we can always blow up a point P E 5 to produce a new
non-singular surface 5' that is birationally equivalent to 5 but not isomor-
phic to 5. The blown-up surface has the property that there is a birational
morphism 5' --> 5; that is, the map 5' --> 5 is a morphism, and it has an
inverse 5 --> 5' that is a rational map. This leads us to make the following
definition.
Definition. A surface 5 is relatively minimal if it has the following two

properties:
(i) 5 is a non-singular projective surface.
(ii) If 5' is another non-singular projective surface, and if ¢ : 5 --> 5'
is a birational morphism, then ¢ is an isomorphism.
Theorem 7.7. Every surface 50 is birationally equivalent to a relatively

minimal surface 5. If the original surface 50 is non-singular, then there is
a birational morphism So --> 5.
PROOF. This theorem is really an amalgamation of two important results

in the theory of algebraic surfaces. First, resolution of singularities tells
us that every surface is birationally equivalent to a non-singular projective
surface. See Hartshorne [1, V.3.8.1] for a discussion of resolution and a
history of its proof. Second, every non-singular projective surface is bira-

tionally equivalent to a relatively minimal surface (Hartshorne [1, V.5.S]).
We will not prove either of these results but will content ourselves with
a few brief remarks. In order to prove resolution of singularities, one starts
with the surface So and continually blows up points and curves until all of
the singularities disappear. Of course, the hard part is to show that after
each blow-up, the singularities have become quantitatively better.
A curve C C So is called exceptional if C ~ ]!Dl and C 2 = -1. Castel-
nuovo's criterion (Hartshorne [1, V.5.7]) says that if C is an exceptional
curve on So, then there is a non-singular surface S\ and a birational mor-
phism ¢ : So ---7 SI with the property that ¢ is an isomorphism away from C
and ¢ sends C to a point PI E S 1. We say that ¢ is the blow-rlown of the
curve C, since ¢ is the blow-up of S1 at Pl' If 51 has any exceptional
curves, we choose one and blow it down. Continuing in this fashion, we
obtain a sequence of surfaces So, S 1, 52, .... The hard part is to show that
this process terminates with a non-singular surface 5 which contains no ex-
ceptional curves. Then one shows that such a surface is relatively minimal.
See Hartshorne [1, V.5.S] for details. This also proves the last part of (7.7),
since the blow-down maps 50 ---7 51 ---7 5",2 ---7 • . . are all morphisms. 0
§8. The Geometry of Fibered Surfaces
An elliptic surface Jr : [ ---7 C is an example of a fibered surface: that is, a

surface that is described as a collection of fibers [t = Jr-l(t) parametrized
by the points t of a curve C. Other examples of fibered surfaces include
ruled surfaces, which are surfaces whose fibers are all isomorphic to ]!D1
(Hartshorne [1, V §2]), and products C 1 x C 2 , which can be made into
fibered surfaces in two ways by using the projections onto either the first
or second factor. In this section, we will prove some geometric properties
that are true for all fibered surfaces. In subsequent sections, we will apply
these results to our study of elliptic surfaces.
Definition. A fiber'ed surface is a non-singular projective surface S, a 1l01l-
singular curve C, and a surjective morphism Jr : 5 ---7 C. For any t E C,
the fiber of 5 lying over t is the curve St = Jr- 1(t). Note that 5 t will be a
non-singular curve for all but finitely many t E S.
The irreducible divisors on a fibered surface naturally divide into two
different sorts, those that lie in a single fiber and those that cover C. More
precisely, let r c 5 be an irreducible curve lying on a fibered surface Jr :
5 ---7 C. Then 1r induces a map of curves Jr : r --+ C that is either constant
or surjective [AEC, II.2.3]. If it is constant, say Jr(f) = {t}, then r lies
c
A Fibered Surface with Horizontal and Fibral Curves
Figure 3.2
entirely in the fiber St, and we call r fibral. If not, then 7f : r --> C is a
finite map of positive degree, and we call r horizontal. See Figure 3.2.
Definition. A divisor D E Div(S) on a fibered surface S is called fibral if
all of its components are fibral. D is called horizontal if all of its components
are horizontal. Note that every divisor can be uniquely written as the sum
of a horizontal divisor and a fibral divisor, since every irreducible curve is
either horizontal or fibral.
Let 7f : S --> C be a fibered surface. If t E C, then the components
of 7f- 1 (t) are irreducible fibral divisors. We assign multiplicities to these
components in the following way. Let Ut E k(C) be a uniformizer at t, that
is, ordt(u) = 1. Then Ut 07f is a function on S, so we can take its divisor,
or more precisely that part of its divisor lying in the fiber St. Extending
linearly, this gives us a homomorphism from Div(C) to Div(S).
Definition. Let 7f : S --> C be a fibered surface, and for each t E C, fix a
uniformizer Ut E k( C) at t. We define a homomorphism
7f* : Div(C) ---7 Div(S),
L nt(t) t---> L nt L ordr(ut 07f)r,
tEG tEG reS,
where the inner sum on the right is over all irreducible curves r contained in
the fiber St = 7f- 1 (t). It is easy to see that 7f* is independent of the choice
of uniformizers Ut, since if u~ is another uniformizer, then (ut/u;)(t) =1= 0,00
at t. Hence (ut/uD 07f is not identically 0 or 00 on any component of St,
so ordr(ut 0 7f) = ordr(u; 07f).
It is clear from the definition of Jr* that the divisors in Jr* (Div( C))
are fibral. We begin by showing that they have trivial intersection with all
other fibral divisors.
Lemma 8.1. Let Jr : S --+ C be a fibered surface, let 8 E Div(C), and
let D E Div(S) be a fibral divisor. Then D . Jr*8 = O.
PROOF. Using the linearity of the intersection pairing and the fact that Jr*
is a homomorphism, we may assume that D is an irreducible fibral divisor
and that 8 = (t) consists of a single point. Then Jr(D) consists of one point.
If that one point is not t, then D and Jr* (t) have no points in common, so
clearly D . Jr* 8 = O.
We have reduced to the case that Jr(D) = {t}. To complete the
proof, we will move 8 = (t) by a linear equivalence. We can choose a
non-constant function f E k(C)* by applying the Riemann-Roch theorem
for curves [AEC, II.5.5c] to the divisor (29 + l)(t) E Div(C), where 9 is
the genus of C. Riemann-Roch then says that £ ( (29 + 1) (t») = 9 + 2, so in
particular there exists a non-constant function f whose only poles are at t.
Let ordt(f) = -n, and consider the divisor
Jr* (n8 + div(f)) = nJr* 8 + div(f 0 Jr).

The point t does not appear in the divisor n8 + div(f), so the left-hand
side has no points in common with St, and hence it has trivial intersection
with D. On the other hand, div(f 0 Jr) is linearly equivalent to 0, so it has
trivial intersection with every divisor on S. Intersecting both sides with D,
we find that
o = D . Jr* (n8 + div(f)) = D . (nJr* 8 + div(f 0 Jr)) = nD . Jr* 8.

Hence D . Jr*8 = O. o
We now show that the intersection pairing is negative semi-definite
when it is restricted to fibral divisors, and we calculate its null space.
Proposition 8.2. Let Jr : S --+ C be a fibered surface, and let D E Div(S)
be a fibral divisor on S.
(a) D2 ::; O.
(b) D2 = 0 if and only if D E Jr* (Div( C) @ Ql). In other words, D2 = 0
if and only if there is a divisor 8 E Div( C) such that aD = bJr* 8 for some
non-zero integers a, b E Z.
PROOF. (a) Write D = Dl + ... + D n , where each Di is contained in a

different fiber. Then Di . D j = 0 for i of j, since they have no points in
common, which implies that D2 = Dr
+ ... + D;. It thus suffices to prove
the proposition for each D i , so we may assume that D is contained in a
single fiber, say D cSt.
Write
L
T
F ~f ]f*(t) = nifi
i=()
as a sum of irreducible divisors. It is clear from the definition of ]f" that
the ni's are all positive. Our assumption that D c St means that D has
the form
L
T
D = aif i for some integers ai·

i=()
We rewrite D and define another divisor D' by the formulas
= L -(n.if;),
r 2
L a; (nifd.
r
ai
D and D' =
i=O
n "t i=O n,
Proposition 8.1 tells us that D' . F F·D' O. We use this to

compute
-2D2 = D' . F - 2D2 + F . D'
The terms with i = j in this last sum are zero, so we find that
D
2
= -2
1
LT ( ai
~ - -;;: )2 (nifi)' (njf
aj
j ).
i.)=O L J
ioJ)
For i I j, the divisors fi and fj are distinct irreducible divisors,

so fi . f J ~ O. Further, as noted above, the multiplicities no, ... ,n T are all
positive, so
for all i I j.
This immediately implies that D2 So 0, which completes the proof of (a).
(b) Suppose now that D2 = O. Then the formula for D2 shows that
ai
for all i,j such that fi . fj > O.
n 1.
In other words, the ratios ad ni and aj / n J will be the same if the divi-
sors fi and f j have a point in common. On the other hand, it is a general
fact that the fibers of a fibered surface are connected. This is a special
case of Hartshorne [I, nI.l1.3j, or see exercise 3.21. So given any two
components f i and f j, we can find a sequence of components
with fik . f ik + 1 > 0 for all k = 0, 1, ... , m - 1. Hence ai/ni = aj/nj for
all i and j. Let a = ao/no E Q be this common ratio. Then
r r r
D = Laifi = L ~inifi = a Lnifi = aF E 7r*(Div(C) ® Q),

i=O i=O' i=O
which completes the proof of Proposition 8.2. D
Remark 8.2.3. With notation as in the proof of (8.2), consider the inci-
dence matrix
1= (fi . fj)O~i,j~r
which describes how the components of the fiber St intersect one another.
Then (8.2) may be restated as follows: The quadratic form
is negative semi-definite, with one dimensional null space spanned by the

vector (no, ... , nr)' In particular, det(I) = 0, but every det(1;;) i=- 0,
where 1;; is the minor obtained by deleting the ith row and ith column of I.
Next we show that for a large class of divisors it is possible to add on
a fibral divisor so that the sum will have trivial intersection with all fibral
divisors. We will use this construction in the next section to describe the
canonical height in terms of intersection theory.
Proposition 8.3. Let 7r : S -'> C be a fibered surface, and let D E Div(S)
be a divisor on S with the property that
D· 7r*(t) = 0 for some (every) t E C.
(The quantity D· 7r*(t) is independent oft; see exercises 3.22 and 3.23 or
Hartshorne [1, exercise V.1.7}.) Then there exists a fibral divisor 1> D E
Div(S) ® Q such that
(D + 1>D) . F = 0 for all fibral divisors F E Div(S).
If 1>~ is another divisor with this property, then

In other words, <P D is uniquely determined by D up to divisors that come

from C.
PROOF. We are going to try to write <PD in the form E arf and solve for
the coefficients ar. More precisely, for every point t E C, write
r,
7r*(t) = L ntifti
i=O
as a sum of irreducible components. We set atO = 0 for all t. Further,

when Tt ~ 1 we consider the following system of linear equations:
r,
L atifti . f tj = -D· f tj , l:Sj:STt·
i=l
Note we are discarding the oth-component ftO, so this is a system of Tt

equations in the Tt variables ati. Proposition 8.2 says that the incidence
matrix
(f ti . f tj h::;i,j::;r,
has non-zero determinant (see also (8.2.3)), so this system of equations has
a unique solution in rational numbers ati E Q.
We claim that the divisor
r,
<PD = LLatifti
tEG i=O
has the desired property. Note that this is a finite sum, since Tt = 0 for all
but finitely many t, and atO = O. To check that <PD works, it suffices by
linearity to show that (D+<p D)·F = 0 for every irreducible fibral divisor F.
Each irreducible fibral divisor has the form F = f tj for some tEe and
some 0 :s j :s Tt. We consider three cases.
First, ifTt = 0, then F = ftO = 7r*(t). Using (8.1) and the assumption
that D· 7r*(t) = 0, we find that
(D + <PD) . F = D· 7r*(t) + <PD' 7r*(t) = O.
Second, suppose that rt 2 1 and F = f tj with j 2 1. Then the fact

that the ati's give a solution to the system of linear equations allows us to
compute
r,
(D + <PD) • F = D· f tj +L L atifti . f tj = O.
tEG i=l
• c
t
A Reducible Fiber on a Fibered Surface
Figure 3.3
Finally, we consider that case that rt 2: 1 and F = rIO. Then
0= (D + if>D) . Jr*(t) from (8.1)

T"t
= Lnti(D+if>D)·fti since Jr*(t) =2:ntifti

i=O
= ntO (D + if> D) . f to from the previous case.
This completes the proof that (D + if> D) . F = 0 for all tibral divisors F.
It remains to show that if> D is unique up to addition of a divisor from C.
Suppose if>~ is another divisor with the same property. Then for every tibral
divisor F we have
But 1>D -1>~ is itself fibraL so (1)D -1>~j2 = o. It follows from (8.2b)
that if> D - 1>~ is in Jr* (Div( C) G-0 Q). 0
Example 8.3.1. Let Jr : 5 ---> C be a fibered surface, and suppose that

the fiber 5 t consists of four components arranged in the shape of a square
with transversal intersections, as illustrated in Figure 3.3. In other words,
if i - j := ± 1 (mod 4),
Jr* (t) = r 0 + r 1 + r 2 + r :0 with r i · rj = { ~ if i - j := 2 (mod 4).
We can use (8.1) to compute the self-intersections of the components.

For example,
so f6 = -2, and similarly f; = -2 for the other i's. Thus the incidence
matrix for this fiber is
~ ).
-2 1 0
( -2 1
1= (fi · fJ')O<i - =
- ,J'<3 1
0 1 -2
1 o 1 -2
Suppose now that D E Div(8) is a (horizontal) divisor with
D· fo = -1, D· r l = 1, D· r 2 = 0, D· f3 = O.
For example, D might consist of two curves D = DI - Do each of which

maps isomorphically 71' : Di --; C, with DI going through fl and Do going
through fo. To find the part of D lying over t, call it D,t, we take
 D,t = aIr l + a2r2 + a3r3, set (D + D,t) . r i = 0 for i = 1,2,3, and solve
for the ai's. Doing this gives
311
D ,t = -rl
4
+ -f2
2
+ -f3'
4
The reader can check that (D + D,t)' r i = 0 for 0 :S i :S 4. Similarly, if D

were to satisfy
D· fo = -1, D· r l = 0, D· r 2 = 1, D· r3 = 0,
then
See exercise 3.24 for a generalization to the case that the fiber is an n-gon
with transversal intersections.
The final topic for this section is minimal models of fibered surfaces.
These will be minimal models which respect the fact that the surface is
fibered. More precisely, we might say that a fibered surface 8 --; C is
relatively minimal if for every fibered surface 8' --; C, every birational
map 8 --; 8' commuting with the maps to C is a morphism. In the case
that the non-singular fibers of 8 --; C have genus at least 1, then it turns
out that there is a unique relatively minimal model. Further, this model
will have the stronger minimality property described in the next theorem.
Theorem 8.4. Let 8 ~ C be a fibered surface with the property that

its non-singular fibers are curves of genus at least 1. Then there exists
a fibered surface 8 min ~ C and a birational morphism ¢ : 8 ~ 8m in
commuting with the maps to C with the following property:
Let 8' ~ C be a fibered surface, and let ¢' : 8' ~ 8 be a birational
map commuting with the maps to C. Then the rational map ¢ 0 ¢' extends
to a morphism. In other words, the top line of the commutative diagram
q,' q,
8' ---+ 8 --+ 8 min
'\, 1 /
C
extends to a morphism.
PROOF. The basic idea is as follows. For any given 8, let 8 min be obtained
from 8 by blowing down all of the exceptional curves on the reducible
fibers. Next, given an 8' birational to 8, take the resulting birational
map 8,min ~ 8 min and factor it into the smallest number of quadratic
transformations
8 ,min = 8 0 --+ 8 1 --+ 8 2 --+ ... --+ 8 n = 8 min .
Then by studying the behavior of the exceptional curves on these quadratic
transformations, one shows that it is possible eliminate one of the "blow-
up-blow-down pairs." In other words, if n ;:::: 1, then one shows that 8,min
and 8 min are connected by a smaller chain of quadratic transformations.
Hence 8,min and 8 min are isomorphic, which gives the desired result.
Unfortunately, we do not have at our disposal the tools needed to turn
this brief sketch into a rigorous proof. We refer the reader to Lichten-
baum [1, Thm. 4.4] or Shafarevich [2, p. 131] for the complete proof of
Theorem 8.4. 0
Definition. It is clear that the surface 8 min described in (8.4) is uniquely

determined up to a unique isomorphism commuting with the maps to C.
A fibered surface 8 ~ C is called a minimal fibered surface (over C) if it
is equal to 8 min .
Corollary 8.4.1. Let 1[" : 8 ~ C be a minimal fibered surface over C,
and let T : 8 ~ 8 be a birational map commuting with the map to C
(i.e., 1[" 0 T = 1["). Then T is a morphism.
PROOF. By assumption, 8 is minimal, so the map ¢ : 8 ~ 8 min in (8.4) is
an isomorphism. Now applying (8.4) with 8 = 8' and ¢' = T, we deduce
that the composition
is a morphism. Hence the same is true of T = ¢-l 0 (¢ 0 T). o

§9. The Geometry of Elliptic Surfaces
Let 7f : C ---+ C be a minimal elliptic surface, and let E j K be the associated

elliptic curve over the function field K = k(C) of C (3.8). Recall (3.lOc)
that each point P E E(K) corresponds to a section (J'p : C ---+ c. We define
a translation-by-P map
Tp: C ----+ C
on c by using the translation-bY-(J'p(t) map on each non-singular fiber Ct.

It is clear that Tp is a birational map, since it is certainly given by rational
functions and it has the rational inverse Lp. The minimality of c then
implies that Tp extends to a morphism. We record this important fact in
the following proposition.
Proposition 9.1. Let 7f : C ---+ C be a minimal elliptic surface with
associated elliptic curve E j K.
(a) For any point P E E(K), the translation-by-P map
Tp: C ----+ C
extends to an automorphism of c.
(b) Let
Aut( ej C) = {automorphisms 'I : C ---+ C satisfying 7f 0 'I = 7f}.
Then the map

E(K) ----+ Aut(cjC), P~Tp,
is a homomorphism.
PROOF. (a) This a special case of Corollary 8.4.1, which says that any bi-
rational map of a minimal fibered surface to itself extends to a morphism.
(b) If the fiber Ct is non-singular, then Tp maps Ct to itself by definition.
It follows that 7f 0 Tp = 7f on all non-singular fibers. But the non-singular
fibers are Zariski dense in C, and a morphism is determined by its values
on any Zariski dense set (Hartshorne [1, I.4.1]), so 7f 0 Tp = 7f on all of c.
This proves that Tp E Aut(cjC). Similarly, the identity Tp+Q = TpOTQ is
clearly true on all non-singular fibers, so it is true everywhere. Finally, TO
is the identity map, which completes the proof that E(K) ---+ Aut(cjC) is
a homomorphism. 0
Let P E E(K) with corresponding section (J'p : C ---+ c. The im-

age (J'p(C) of (J'p is a curve on the surface C, which we can think of as a
divisor on £. We will write
(P) E Div(c)
for this divisor. It is important to note that the divisors
(P) + (Q) and (P+Q)
are very different. The former is the sum of the two divisors (P) and (Q)
in Div(£), whereas the latter is the image of the section O"P+Q = O"p + O"Q
which is defined using the group law on E. The following proposition shows
how they are related.
Proposition 9.2. With notation as above, let PI"'" Pr E E(K) be
points, and let nl, ... ,n r E Z be integers such that
Let n = nl + ... + n r . Then the divisor
is linearly equivalent to a fibral divisor.

In particular, for all P, Q E E(K), the divisor
(P+Q) - (P) - (Q) + (0)

is linearly equivalent to a fibral divisor.
PROOF. If P E E(K), our notation (P) is potentially ambiguous, since we

could mean either the divisor on the curve E/ K consisting of the point P,
or the divisor on the surface £ consisting of the curve O"p(C). To resolve
this difficulty, we will denote the former by (P)E and the latter by (P)e.
Fix a Weierstrass equation for E / K, say
E : y2 = x 3 + Ax + B, A,BEK,
and consider the divisor
By assumption, D has degree 0 and sums to the zero element of E(K).

Applying [AEC, III.3.5] to the divisor D on the elliptic curve E / K, we find
that D is linearly equivalent to O. Thus there is a function f E K(E) such
that
D = (div f)E E Div(E).
The relationship (3.8) between E and £ says that K(E) ~ k(£), so
we can consider f as an algebraic function on the surface £. When we
compute its divisor on £, we find that
(div f)E and

are almost the same. To see this, note that f E K(E) = k(C)(x,y) is a
rational function in x and y with coefficients in k( C). Hence for all but
finitely many t E C, we can evaluate those coefficients to get a function it E
k(c.t) whose divisor will be precisely
This proves that the difference
is contained in finitely many fibers, hence it is fibral. o
For any point P E E(K), the divisor (P) - (0) E Div(c.) satisfies
((P) - (0)) . 7r*(t) = 0 for all t E C.
This is true because the image of a section will intersect a fiber 7r*(t)
exactly once. (See exercise 3.22.) This shows that we can apply (8.3) to
the divisor (P) - (0), as in the following definition.
Definition. For each point P E E(K), let <Pp E Div(c.) ® Ql be a fibral
divisor so that the divisor
Dp def (P) - (0) + <Pp

satisfies
for all fibral divisors F E Div(c.).
Such a divisor exists by (8.3) and the remarks made above. Then we define
a pairing on E(K) by the formula
( " .) : E(K) x E(K) ---+ Ql,

(F,Q) = -Dp· D q .
The next result shows that this geometrically defined pairing is equal
to the canonical height pairing (4.3), which justifies our use of the same
notation for the two pairings! This geometric construction of the canonical
height is due to 11anin [1]. See also Shioda [2] for a more detailed analysis
of the induced Euclidean structure on the lattice E(K)j E(K)tors.
Theorem 9.3. (l\lanin [1]) Let 7r : C. -+ C be a minimal elliptic surface
with associated elliptic curve E j K. The pairing
(.,.): E(K) x E(K) ---+Ql,
defined above has the following two properties:

(a) (.,.) is bilinear.
(b) (P, P; = h(P) + 0(1) for all P E E(K), where we recall from §4
that h(P) = h(xp) is the degree of the map Xp : C ~ pl.
Hence this pairing agrees with the canonical height pairing defined
in (4.3). In particular, h(P) = ~(P, P; E Q for all P E E(K).
PROOF. (a) Let P, Q, R E E(K) be any three points. Applying (9.2), we

choose a fibral divisor F such that
(Q + R) - (Q) - (R) + (0) "-' F.
Then using standard properties of the intersection pairing (7.2), we com-

pute
(P,Q + R; - (P,Q; - (P,R;

= -Dp' DQ+R + Dp' DQ + Dp' DR
= -Dp . ((Q + R) - (0) + <PQ+R) + Dp· ((Q) - (0) + <PQ)
+ Dp· ((R) - (0) + <PR)
= -Dp . ((Q + R) - (Q) - (R) + (0) + <PQ+R - <PQ - <PR)
= -Dp· (F + <PQ+R - <PQ - <PR)
= o.
The last line follows from the fact that D p has trivial intersection with all
fibral divisors. Hence
(P, Q + R; - (P, Q; - (P, R; = O.
It is also easy to check that the pairing is symmetric,
(P, Q; = -Dp . DQ = -DQ . Dp = (Q, Pl'
This completes the proof that the pairing is bilinear.

(b) Directly from the definition we find that
(P,P; = -Dp· Dp
= -((P) - (0) + <pp). Dp
= -((P) - (0») . Dp since Dp· (fibral) = 0
= 2(P) . (0) - (P) 2 - (0) 2 + ( (P) - (0) ) . <Pp.
Our first claim is that (P)2 does not depend on P. To see this, consider
the translation-by- P map
7p: e - - 7 e.
We know from (9.1a) that 7p extends to an automorphism of e. It follows
that 7pDl . 7pD2 = Dl . D2 for any two divisors D 1 , D2 E Div(e). Hence
(P) . (P) = 7p(P) . 7p(P) = (0) . (0)

is independent of Tp. Alternative approach: first show that the canoni-

cal divisor on c has the form 7[*8, and then use the adjunction formula
(Hartshorne [1, V.l.5]) to compute
(p)2 = 2g(C) - 2 - (P) ·7[*0 = 2g(C) - 2 - dego.
Our second observation is that although <P p depends on P, there are

essentially only finitely many choices for <P p. More precisely, for each t E C,
write
Tt
7[*(t) = L ntifti
i=O
as a sum of irreducible components. Note that rt = 0 for all but finitely

many t E C. Looking back at the proof of (8.3), we see that <pp can be
written in the form
Tt
<Pp = LLatifti +7[*(0)

tEG i=l
for some 0 E Div(C), where the integers ati are uniquely determined by
the finitely many intersection indices
(P) - (0)) . ftj, t E C, 1 ~ j ~ rt.
But every (P)· f tj is either 0 or 1, so as we take different points P E E(K),

there will be only finitely many possibilities for the ati's. Further,
(P) - (0)) . 7[*(0) = (P) ·7[*(0) - (0) ·7[*(0) = deg(o) - deg(o) = 0

from exercise 3.22(b), so we find that
Tt
(P) - (0)) . <Pp = (P) - (0)) . (LLatifti)

tEG i=l
can take on only finitely many values as we vary P E E(K).

Combining these two observations with the calculation from above
yields
(P,P) = 2(P)· (0) + 0(1) for P E E(K).
It remains to calculate the intersection index (P)· (0). Adjusting the 0(1)
if necessary, we may assume that [2]P i- O. Fix a Weierstrass equation
forE,
E : y2 = x 3 + Ax + B,
and write P = (xp, yp). Changing coordinates if necessary, we may as-
sume xp and yp have no poles in common with the poles of A and B.
Let t E C. We will compute the local intersection index of (P)

and (0) at the point O'o(t). We denote this local intersection by (p. O)t.
If ordt(xp) :2" 0, that is, if xp(t) -I 00, then (P) and (0) do not intersect
on the fiber Et , so (P . O)t = 0. Suppose now that ordt(xp) < 0, so the
equation for E tells us that
We make a change of coordinates w = x/y, z = l/y, so E now has the

equation
and P = (wp, zp) = (xp/yp, l/yp). Also let u E k(C) c k(E) be a

uniformizer at t, so we are looking at the intersection of (P) and (0) at
the point (w,z,u) = (0,0,0). The local ring of E at this point is
k[w, z, u](O,o.O)
(z - w 3 - Awz 2 - B z3)'
Further, in this ring, (P) has the local equation w - wp = and (0) has °
the local equation w = 0, so by definition the intersection index (P,O)t is
equal to the dimension over k of the vector space
k[w, z, n](o.o.O) "" k[z, u](O,O) ~ k[u]o

(z-w 3 -Awz 2 -Bz3,w-wp,w) (Z_B z3,Wp) = (wp)'
Note the last equality follows from the fact that z - Bz:l = z(l - Bz2)
°
and 1 - Bz2 is a unit in k[z, uj(o.O)' If we write Wp = uew'p for some
function w~, that is neither nor ex:; at t, then we have
. k[u]o . k[u]o
dank -(- )
wp = dlmk ~(
u e~) = e,
and also
1
e = ord t Wp = ordt(xp/yp) = -2 ordt(xp).
This proves that
( p·O) _
t -
{o_~ ordt(xp)
if ordt(;Y;p) :2" 0,
if ordt(xp) < 0.
Adding over t E C gives

Hence
(P, P) = 2(P) . (0) + 0(1) = h(P) + 0(1),
which completes the proof of (b).
Let g(P) = ~ (P, P) be the quadratic form associated to our pairing.
Then g(P) = ~h(P) + 0(1) from (b), whereas the bilinearity in (a) tells
us that g(2P) = 4g(P). This shows that g satisfies properties (i) and (ii)
of (4.3b), so by the uniqueness (4.3e) of the canonical height, we have g = h.
o
Theorem 9.3 shows that the canonical height on E(K) can be com-
puted using intersection theory. Our next goal is to define a natural pair-
ing on a certain subgroup E(K)o of E(K). This pairing takes its values
in Pic(C), and the composition .
E(K)o x E(K)o ---> Pic(C) ~ Z
will be the canonical height pairing. We begin by describing E(K)o.

Let P E E(K) be a point and Tp : C -> C the translation-by-P auto-
morphism. We know from (9.1b) that Tp gives an automorphism of each
fiber Ct. In particular, it must permute each of the components of Ct.
Definition. Define a subset E(K)o of E(K) by
E(K)o = {p E E(K) : Tp(r) = r for all fibral curves r c £}.
Lemma 9.4. E(K)o is a subgroup of finite index in E(K).
PROOF. Let P, Q E E(K)o. From (9.1b) we know that TP+Q = Tp 0 TQ,

so for any fibral curve r we have TP+Q(r) = Tp(TQ(r)) = Tp(r) = r.
Therefore P + Q E E(K)o. Similarly, r = TO(r) = LP+P(r) = Lp(r),
so -P E E(K)o. This proves that E(K)o is a subgroup of E(K).
For the second part, we observe that if Ct is an irreducible fiber, then
clearly Tp(Ct) = Ct. Let {r 1 , ... , rr} be the set of all components of the
reducible fibers of c. It is a finite set, since C has finitely many reducible
fibers, and each reducible fiber has finitely many components. Then E(K)
acts on this set by
E(K) ---> Aut{r 1 , ... ,rr} ~ Sr,

P I---> (ri f-+ TP(r i )).
In other words, there is a homomorphism from E(K) into the symmet-

ric group Sr on r letters. From the definition of E(K)o, the quotient
group E(K)j E(K)o injects into STl which proves that E(K)o is a sub-
group of finite index.
o
Remark 9.4.1. There is another way to characterize E(K)o in terms of

the sections ap : C -+ E associated to points P E E(K). Let ao : C -+ E
be the zero-section. Then a point P E E(K) is in E(K)o if and only if
the curves ap(C) and ao(C) hit the same component of every fiber of E.
We will study the group E(K)o and the quotient E(K)j E(K)o in greater
generality and detail in the next chapter; see (IV.6.12), (IV.9.1), (IV.9.2)
and exercise 4.25.
For any two points P, Q E E(K), Proposition 9.2 tells us that there is
a fibral divisor <P P,Q E Div( E) satisfying
(P + Q) - (P) - (Q) + (0) '" <PP,Q.
Clearly, <P P,Q is determined by P and Q up to principal divisors, so its

class in Pic(E) is well-defined. This gives a pairing on E(K) with val-
ues in Pic(E). The next result shows that this pairing is quite nice when
restricted to the subgroup E(K)o.
Theorem 9.5. Let 7[" : E -+ C be a minimal elliptic surface with associ-
ated elliptic curve E j K.
(a) Let P, Q E E(K)o. Then there exists a divisor [P, QJ E Div( C) such
that
(P + Q) - (P) - (Q) + (0) '" 7["* ([P, QJ).
The divisor [P, QJ is determined by P and Q up to linear equivalence.
(b) The pairing
E(K)o x E(K)o ---+ Pic(C), (P, Q) f-----> class [P, Q],
is a well-defined symmetric bilinear pairing. (See also exercise 3.26')

(c)
(P, Q) = deg[P, QJ for all P, Q E E(K)o.
In particular, h(P) = ~ deg[P, PJ for all P E E(K)o.
PROOF. (a) For any two points P, Q E E(K), let <PP,Q E Div(E) be a fibral
divisor satisfying
(P+ Q) - (P) - (Q) + (0) '" <PP,Q
as described in (9.2). Then for any fibral divisor F E Div(E),
<PP,Q . F = (P + Q) . F - (P) . F - (Q) . F + (0) . F

= TQ(P) . F - (P) . F - TQ(O) . F + (0) . F
= TQ(P) - (0)) . F - «P) - (0)) . F
= (P) - (0)) . LQ(F) - (P) - (0)) . F
= 0 since Q E E(K)o implies T_Q(F) = F.
But P,Q itself is fibral, so we deduce that ~,Q = O. It follows from (8.2b)
that there is a divisor [P, Q] E Div( C) ® Q such that P,Q = 71'* ([p, QJ).
This will suffice for our purposes in this chapter, so we will leave it for the
reader (exercise 3.28c) to show that [P, Q] is actually in Div(C).
The divisor P,Q is clearly determined by P and Q up to linear equiv-
alence on E'... In order to show that [P, Q] is determined up to linear equiv-
alence on C, we will prove that if 8 E Div( C) satisfies 71'* 8 '" 0, then 8 '" O.
Write 71'*8 = div(f) for some I E k(E'..). For all but finitely many t E C
we can restrict I to the fiber E'..t to get a rational function It E k(E'..t). By
assumption, the poles and zero of I lie on finitely many fibers, so for almost
all t E C we see that It E k(E'..t} has no zeros or poles. It follows that It
is constant. Let (J : C ----> E'. be any section, for example the zero section.
Then the fact that I is constant on almost all fibers means that the func-
tion I - I 0 (J 0 71' is identically 0 on those fibers. But a rational function
is determined by its values on any non-empty open set, so I = f 0 (J 0 71'.
Therefore
71'* 8 = div(f) = div(f 0 (J 0 71') = 71'* (div(f 0 (J)),
so 8 = div(f 0 (J) is a principal divisor on C. Notice that what this result

really says is that the natural map 71'* : Pic( C) ----> Pic( E'..) is injective.
(b) The pairing is well-defined from (a), and it is clearly symmetric. To
see that it is bilinear, we let P, Q, R E E(K)o and compute
P,Q+R '" (P + Q + R) - (P) - (Q + R) + (0)

= (P + Q + R) - (P + R) - (Q + R) + (R)
+ (P + R) -
+ (0) (P) - (R)
= TR((P + Q) - (P) - (Q) + (0)) + ((P + R) - (P) - (R) + (0))
'" TR ( P,Q) + P,R

= P,Q + P,R.
Note that the last equality is true because R E E(K)o, so TR fixes the fibral
divisor P,Q' Now write each X,Y as 71'* ([X, YJ) and use the fact proven
above that 71'* : Pic( C) ----> Pic( E'..) is injective. This yields the desired result,
[P,Q+R] '" [P,Q] + [P,R].
(c) Let P E E(K)o. Then for every fibral divisor FE Div(E'..) we have
(P) . F = Tp(O) . F = (0) . Lp(F) = (0) . F.

In other words,
((P) - (0)) . F = 0 for all fibral divisors F E Div(E'..),

so in the notation of (9.2), Dp = (P) - (0). Note that this is only valid
for points in E(K)o. Let P, Q E E(K)o. We compute
-(P, Q) = Dp . DQ by definition of (', .)
= (P) - (0)) . (Q) - (0)) from above
= (P) . (Q) - (P) . (0) - (Q) . (0) + (0) . (0)
= LP(P) . Lp(Q) - Lp(P) . Lp(O) - (Q). (0) + (0) . (0)
= (0) . (-P + Q) - (0) . (-P) - (Q) . (0) + (0) . (0)
= (-P + Q) - (-P) - (Q) + (0))· (0)
= 7[* ([-p, Q]) . (0) by definition of [', .] in (a)
= deg[-P, Q] from exercise 3.22(b)
= - deg[P, Q] by linearity of [', .] from (b).
This proves that (P, Q) = deg[P, Q] for all P, Q E E(K)o. Putting P = Q
gives h(P) = ~ (P, P) = ~ deg[P, P], which completes the proof of Theo-
rem 9.5.
o
Remark 9.6. Theorem 9.5 says that the canonical height pairing (', .) en-
dows E(K)o with the structure of a Euclidean lattice whose inner product
takes integer values. Similarly, the height pairing gives E(K) a Euclidean
structure with an inner product taking rational values having severely lim-
ited denominators. It is an interesting problem to classify the possible lat-
tice structures on E(K)o and E(K). In a series of papers, T. Shioda [1,2,4-
7] has investigated these Mordell-Wei 1 lattices and proven many interesting
results, including the construction of examples for which E(K)o is isomor-
phic to a root lattice of type E 6 , E 7 , and Es.
Remark 9.7. Let c ----> C be a non-split minimal elliptic surface, and
let E / K be the associated elliptic curve. The Neron-Severi group of C, de-
noted by NS(c), is the group of divisors modulo algebraic equivalence. (For
the definition of algebraic equivalence, see Hartshorne [1, exercise V.1.7].)
One can prove that NS(c) is a finitely generated group and that the inter-
section pairing on Div( c) gives a well-defined pairing on NS (c). It is thus
an interesting question to relate NS(c) and its intersection pairing to E(K)
and its height pairing. Shioda [3, Thm. 1.1] has shown how to find genera-
tors for NS(c) by using generators for E(K) and fibral components of C. In
particular, he proves the fundamental rank relation (Shioda [3, Cor. 1.5])
rankNS(c) = rankE(K) + 2 + I)rt - 1),

tEG
where rt is the number of irreducible components in the fiber Ct. He also

gives a formula relating the intersection regulator of NS( c) to the canonical
height regulator of E(K).
§10. Heights and Divisors on Varieties
Let C -+ C be an elliptic surface defined over a number field k. For each

point t E CCk) such that Ct is non-singular, there is a canonical height
function h : ctCk) -+ IR on the elliptic curve Ct. In the next section we
will investigate how the canonical height varies from fiber to fiber, espe-
cially for points lying on the image of a section (J : C -+ C. To carry out
this investigation, we will need to develop more fully the theory of height
functions on varieties.
For our purposes in this chapter, it would suffice to consider only
curves and surfaces, but the theory is hardly more difficult for general
varieties. We will, however, need to assume that the reader is familiar
with standard properties of divisors on varieties, as covered for example
in Hartshorne [1, II §§6,7]. Some of the proofs in this section are fairly
technical, so some readers may want to read the definitions and statements
of the main results (10.1, 10.2, 10.3) and then proceed directly to the next
section.
For this section, we set the following notation:
k a number field, with algebraic closure k,

V/ k a non-singular projective variety defined over k,
Div(V) the group of divisors on V,
hp the (absolute logarithmic) height function h p : ]p>r (k) -+ IR on
projective space as defined in [AEC, VIII §5].
A morphism ¢ : V -+ W between non-singular varieties induces a

homomorphism of their divisor groups ¢* : Div(W) -+ Div(V) in the fol-
lowing way. Let r E Div(W) be an irreducible divisor and fix a function ur
which vanishes to order 1 along r. Equivalently, ur is a generator for the
maximal ideal in the discrete valuation ring ()w,r. Then
¢*r = L ord,6.(ur 0 ¢)tJ.,

,6.EDiv(V)
where the sum is over all irreducible divisors tJ. E Div(V) and we are writing
ord,6. : k(V)* -+ Z for the normalized valuation on the local ring eJ v ,,6.. Of
course, we have cheated a little bit. The divisor ¢*r will only be defined if
the image ¢(V) is not contained in r, since otherwise Ur 0 ¢ is identically
zero. However, ¢* sends principal divisors to principal divisors, so it induces
a map ¢* : Pic(W) -+ Pic (V) which is well-defined on all of Pic(W), since
we can always move r by a linear equivalence so that it intersects ¢(V)
properly.
The following theorem is often called the "Height Machine." It is the

main result of this section. The Height Machine associates a height function
to each divisor on V, or, more precisely, it associates an equivalence class
of height functions to each divisor class on V. The power of the height
machine is that it takes geometric relations involving divisor classes on V
and translates them into height relations between points on V. It is thus
a tool for transforming geometric information into arithmetic information.
We have already seen this machine in action in [AEC, VIII §6], where
the geometric group law on an elliptic curve was transformed into the
arithmetic statement h(P + Q) + h(P - Q) = 2h(P) + 2h(Q) + 0(1).
The general formulation of the Height Machine is due to Andre Weil. For
further details and additional properties of heights, see Lang [4], Hindry-
Silverman [1], and exercises 3.31 and 3.32.
Theorem 10.1. (Weil's Height Machine, Weil [2]) Let V be a non-
singular projective variety defined over a number field k. There is a map
h: Div(V) ---> {functions V(k) ----+ lR},
uniquely determined up to bounded functions on V(k), with the following
two properties:
(a) (Normalization) Let ¢ : V ----+ ]P'T be a morphism, and let H E Div(]P'T)
be a hyperplane with the property that ¢(V) ct H. Then
hq,*H(P) = hJf'(¢(P)) + 0(1) for all P E V(k).
(b) (Additivity) Let D, D' E Div(V). Then

hD+D,(P) = hD(P) + hD,(P) + 0(1) for all P E V(k).
The height mapping has the following additional properties:
(c) (Equivalence) Let D, D' E Div(V) be linearly equivalent divisors. Then
hD(P) = hD,(P) + 0(1) for all P E V(k).
(d) (Functoriality) Let 1j; : V ----+ W be a morphism of non-singular projec-

tive varieties over k, and let D E Div(W). Then
hV,,p*D(P) = hW,D(1j;(P)) + 0(1) for all P E V(k).
Remark 10.1.1. Another way to formulate Theorem 10.1 is as follows.

For every V/Q there is a unique homomorphism
. {functions V(Q) ----+ lR}
hv : PIC(V) ---> ----='---------'-"-'-~,-:----,--
{bounded functions V(IQ) ----+ lR}
such that hJf'r is the usual height on projective space [AEC, VIII §5] and
such that hv,,p* D = hW,D 0 1j; for every morphism 1j; : V ----+ W.
§1O. Heights and Divisors on Varieties 257
Example 10.1.2. Let E / k be an elliptic curve. Consider the divisor re-

lation [2]*(0) '" 4(0), which follows from [AEC, I1L3.5] and the fact that
the four points in E[2] sum to zero. Then (10.1) gives the height relation
h(o) ([2]P) = h[2]*(O)(P) + 0(1) = h 4 (o) + 0(1) = 4h(o)(P) + 0(1).
This is one of the properties of the height that was used in the proof of the
Mordell-Weil theorem [AEC, VIIL6.7].
Example 10.1.3. Two divisors D, D' E Div(lP'l) on the projective line are
linearly equivalent if and only if they have the same degree. It follows from
the additivity and equivalence properties (1O.1b,c) that
deg(D')hD(P) = + 0(1)
deg(D)hD1(P) for all P E 1P'1(k).
In particular, (1O.1a) implies that hD(P) = deg(D)hp(P) + 0(1).
For curves of higher genus, the identity (10.1.3) will not be true, since
divisors of the same degree need not be linearly equivalent. However, a
slightly weaker result is valid, as described in the following result. For a
generalization to varieties of arbitrary dimension, see exercise 3.32.
Theorem 10.2. Let C be a curve, let D, D' E Div(C) be divisors
with deg(D) ¥ 0, and let hD, hD' be associated height functions. Then
1. hDI(P) deg(D')
1m
PEC(k), hD(P) ..... OO hD(P) deg ( D ).
The last theorem that we will be proving in this section is a finiteness

result. Recall [AEC, VIII. 5. 11] that in projective space IP'T(k), there are only
finitely many points of bounded height. Of course, here height means hp,
the standard height on projective space. It is clear that this result cannot
be true for every height hD on every variety. For example, if V(k) is infinite,
then it cannot be true for both hD and h-D, since h-D = -hD + 0(1).
In order to give the correct statement, we need one definition.
Definition. A divisor D E Div(V) is called very ample if there is an
embedding 4> : V --+ IP'T and a hyperplane H E DiV(IP'T) not containing 4>(V)
so that D = 4>* H. (To say that 4> : V --+ IP'T is an embedding means that 4>
°
maps V isomorphically onto its image.) The divisor D is called ample if
there is an integer n > so that nD is very ample.
Theorem 10.3. Let DE Div(V) be an ample divisor on V, and let hD :
V(k) --+ lR be an associated height function. Then for all a, b > 0, the set
{p E V(k) : hD(P) ::; a and [k(P): k] ::; b}
is finite. In particular, the set {p E V(k') : h(P) ::; a} is finite for any
finite extension k' / k.
Example 10.3.1. Let C be a non-singular curve of genus g, and let D E
Div(C) be a divisor. Then D is ample if deg(D) > 0, and D is very
ample if deg(D) 2: 2g + 1. See Hartshorne [1, IV.3.2, IV.3.3] or [AEC,
exercise I1I.3.6].
Example 10.3.2. Let S be a non-singular surface. The criterion of Nakai-

Moishezon (Hartshorne [1, V.1.lO]) says that a divisor DE Div(S) is ample
if and only if D2 > 0 and D· r > 0 for all irreducible curves reS.
As mentioned above, the reader may at this point wish to proceed
directly to the next section, where we will apply the Height Machine to
study the specialization map on elliptic surfaces. The remainder of this
section is devoted to proving Theorems 1O.1~1O.3.
Definition. Let ¢ : V ~ JPlr be a morphism of V into projective space. A
divisor DE Div(V) is said to be associated to ¢ if there is a hyperplane HE
Div(pr), not containing ¢(V), such that D = ¢* H. Note that the divisor
class of D is uniquely determined by ¢, since any two hyperplanes in pr
are linearly equivalent.
The height on V associated to ¢ is the height function
hq, : V(k) ---> JR, hq,(P) = hll'(¢(P)).
Our ultimate goal is to associate to every divisor D on V a height

function hD with the properties described in (10.1). In particular, we will
want the heights attached to linearly equivalent divisors to be essentially
the same. The following important proposition will be crucial for this
construction.
Lemma 10.4. Let ¢ : V ~ pr and 'l/J : V ~ JPlS be morphisms which are
associated to the same divisor class. Then
h",(P) = h",(P) + 0(1) for all P E V(k).
Here the 0(1) depends on ¢ and 'l/J but is independent of P.
PROOF. Let D be any positive divisor in the divisor class associated to ¢

and'l/J. This means that on the complement of D we can write ¢ and 'l/J in the
form ¢ = [1o, ... , Ir] and'l/J = [go, ... , gs] with rational functions Ii, gj E
k(V) satisfying
div(fi) = Di - D and div(gj) = Dj - D for divisors D i , Dj 2 o.

Further, the fact that ¢ is a morphism means that the D/s have no points in
common, and similarly for the Dj's. (For general facts about the relation~
ship between morphisms ¢ : V ~ pr and divisors, see Hartshorne [1, II §7],
especially the section on linear systems.)
Now fix some j, let Vj = V '- Dj be the complement of Dj, and
let Fi = Idgj for 0 ~ i ~ r. Notice that
SO Fi is a regular function on Vj. Taken together, the Fi's define a morphism
We also observe that the Fi's have no common zeros on Vj, since any
common zero would lie on all of the Di's.
We need to recall how the maximal ideals in the ring 9t = k[Fo, ... , Frl
correspond to the points of Vj. If 9J1 c 9t is a maximal ideal, then 9t/9J1
is a finitely generated k-algebra which is also a field. It follows from
the weak Nullstellensatz (Atiyah-MacDonald [1, 5.24, 7.10]' Lang [7, X §2
Cor. 2.2]) that 9t/9J1 is isomorphic to k. More precisely, the natural in-
clusion k - t 9t/9J1 is an isomorphism. This means that there are unique
elements ao, ... , a r E k so that Fi == ai (mod 9J1), and then there is a
unique point P!Vl E Vj(k) with F(P!Vl) = (ao, ... , a r ). Equivalently, the
point P!Vl is determined by the congruences
for 0 ::; i ::; r.
Now consider the ideal J = (Fo, ... , Fr) C 9t generated by the Fi's.
We claim that J must be the unit ideal. To prove this, we assume that J
is not the unit ideal and derive a contradiction. Every non-unit ideal is
contained in at least one maximal ideal, so we take a maximal ideal 9J1
with J c 9J1. Then 9J1 corresponds to a point P!Vl E Vj as described above.
On the other hand, we have Fi E J c 9J1 from the definition of J, so
Hence Fi(P!Vl) E k n 9J1, so Fi(P!Vl) = O. In other words, P!Vl is a common

zero of Fo, ... , Fr , which is a contradiction. This completes the proof that J
is the unit ideal.
We can rephrase this last argument in slightly fancier language. The
scheme Spec(9t) is isomorphic to Vj, and by the weak Nullstellentsatz,
maximal ideals in Spec(9t) correspond to k-valued points in Vj. But then
any maximal ideal 9J1 containing J would correspond to a point P in the
zero set of J, contradicting the fact that the F/s have no common zero.
The fact that J = (Fo, ... , Fr) is the unit ideal in 9t means that we
can find a polynomial Aj(To , ... , Tr) E k[To, ... , Trl with no constant term
such that
For any finite extension k' /k, any point P E Vj(k'), and any absolute
value v on k', we evaluate this identity at P, take the v-adic absolute
value, and use the triangle inequality to get an estimate of the form
Here CI = cI(v,¢,1/J,D,A j ) > 0 is a constant that does not depend on P.

Further, for all but finitely many absolute values on k', we can take CI = 1.
(See the proof of [AEC, VIII.5.6J for a similar calculation.)
Recall that Pi = fd gj, so if we multiply both sides by Igj (P) Iv we
obtain the estimate
Notice that this bound is still valid if gj (P) = 0, so it holds for all points at
which the Ii's and gj'S are defined, that is, at all points on the complement
of D. Taking the maximum for 0 :s: j :s: s gives
where C2 = C2(V, ¢, 1/J, D) is again positive and is equal to 1 for all but
finitely many v. Now we take the logarithm of both sides, multiply by the
local degrees [k~ : QvJ/[k' : Q], and sum over all absolute values on k' to
obtain
h( [go(P), ... , gs(P)j) :s: h( [fo(P), ... , fr(p)]) + C3,
where C3 = C3(¢, 1/J, D) is independent of P. In other words, we have shown
that
h(1/J(P)) :s: h((¢(P)) + C3 for all P E (V" D)(k).
The divisor D was chosen to be any positive divisor in the divisor

class associated to ¢ and 1/J. In other words, we can take D to be ¢* H
for any hyperplane H c JlDr not containing ¢(V). Let HI"'" Hm C JlDr be
hyperplanes not containing ¢(V) with the property that HI n ... n Hm =
0. Then the corresponding divisors ¢* HI"'" ¢* Hm have no points in
common, so their complements cover V. Applying the above estimate to
each of these D's and letting C4 be the maximum of the C3'S gives
h(1/J(P)) :s: h((¢(P)) + C4 for all P E V(k).
This is one of the inequalities we are trying to prove, and the opposite
inequality follows if we interchange the role of ¢ and 1/J. D
If D E Div(V) is a very ample divisor, then we can choose an em-

bedding ¢ : V ---> JlDr associated to D and attach to D the height func-
tion hD = h¢. For arbitrary divisors D, we will write D = DI - D2 as a
difference of very ample divisors and define hD linearly by hD = hDl - hD2 .
The following lemma shows that every divisor can be decomposed in this
way.
Lemma 10.5. Every divisor on V can be written as a difference of two

very ample divisors.
PROOF. This is a basic result from algebraic geometry. Before giving the
general proof, we consider a special case. Suppose V is a curve of genus g.
Then a divisor on V is very ample if it has degree at least 2g + 1 (10.3.1).
So for an arbitrary divisor D E Div(V), say of degree d, we can write D as
a difference of very ample divisors
D = (D + n(P)) - n(P)
by choosing n = 2g + 1 + Idl. Similarly, let V be a surface, D E Div(V)
an arbitrary divisor, and H E Div(V) an ample divisor. Then one can use
the Nakai-Moishezon criterion (10.2.2) to show that nH + D is ample for
all sufficiently large n, after which it is easy to write D as a difference of
very ample divisors. We will leave the details to the reader (exercise 3.30)
and go on to the general case.
Let D E Div(V) be an arbitrary divisor, and fix a very ample divi-
sor H E Div(V). Serre's theorem (Hartshorne [1, II.5.17, II.7.4.3]) says
that there is an integer n 2': 1 so that D + nH is ample. (Note we have
translated from the language of invertible sheaves into the language of divi-
sors, as explained in the last part of Hartshorne [1, II §6].) It follows from
Hartshorne [1, II.7.6] that m(D+nH) is very ample for all sufficiently large
integers m. Further, nH is very ample, so nH + m( D + nH) is very ample,
since it is the sum of two very amples. Hence
D = (m + l)(D + nH) - (nH + m(D + nH))

is a difference of very ample divisors. o
Lemma 10.5 lets us decompose a divisor D into a difference Dl - D2 of
very ample divisors. In particular, Dl and D2 are associated to morphisms
from V into projective space. The next result gives some basic properties
of height functions associated to such morphisms.
Lemma 10.6. Let (Pl : V - IP'r and <P2 : V - IP'S be morphisms, and
let Dl and D2 be divisors associated to <Pl and <P2 respectively.
(a) There exists a morphism <P3 : V - IP'rs+r+s associated to Dl + D 2 .
(b) If <P : V - IP'n is any morphism associated to Dl + D 2, then
hq,(P) = hq,l (P) + hq,2 (P) + 0(1) for all P E V(k).
PROOF. (a) The Segre embedding (Hartshorne [1, exercise I.2.14] or Har-
ris [1, 2.11-2.29]) is the map
IP'r x IP's ------> IP'rs+r+s
([xo, ... , Xr], [Yo, ... , Ys]) f------+ [xoYo, ... , XiYj,···, xrYs].
It is clear from this definition that if we pull a hyperplane back by the Segre
embedding, we will get H x IP's + IP',. x H, where the H's are hyperplanes
in the appropriate projective spaces. Hence Dl + D2 is associated to the
morphism
(b) Write
and
with rational functions fo, ... , 9s E k(V). Lemma 10.4 says that it suffices
to prove (b) for anyone morphism ¢ associated to Dl + D 2, so we will
take ¢ to be the map using the Segre embedding described in (a). In other
words,
¢ = [f090, ... , fi9j, ... , f1'9s] : V --+ lP'1's+1'+s.
Let P E V(k) be any point. Replacing k be a finite extension, we may
assume that P E V(k). Then directly from the definition of the height on
projective space we have
hq,(P) = h(¢(P))
= h([fogo (P) , ... , J;gj(P), ... , f,.gs(P)])
= LM [~~ ~ ~l]lOg (0~i~~~~j~.I(fi9j)(P)I)

vE k
= L [~~:: ~l]lOg (max

vEMk
0<,<,.
- -
max 19j(P)I)
Ifi(P)I· O<)<s
- -
[~~:
= "~: ~l] (log (max
0<,<1' IJ;(P)I) + log (max
O<J<S Igj(P)I))
vEMk - - - -
= h([fo(P), ... , f1'(p)]) + h([90(P), ... ,gs(p)]).

o
After these lengthy preliminaries, we are finally ready to tackle the
proof of the Height Machine.
PROOF (of the Height Machine (10.1)). Take each divisor D E Div(V)
and write it as a difference D = Dl - D2 of divisors with the property
that there are morphisms ¢l : V --+ 1P'1' and ¢2 : V --+ IP'S associated to Dl
and D2 respectively. Note that (10.5) assures us that this is possible; in
fact, (10.5) says that we can even choose Dl and D2 so that ¢l and ¢2 are
embeddings. In any case, having fixed D 1 , D 2 , ¢l, ¢2, we define h D to be
for all P E V(k).

Our first observation is that if the Height Machine exists, then up to a

bounded function, this is the only choice for h D . This follows from (1O.1a)
and (10.lb), which let us compute
This gives the uniqueness assertion in Theorem 10.1.

Next we show that up to bounded functions, hD is independent of the
choice of D 1 , D 2 , (PI, Ih. Once we have proven this independence, the rest
of (10.1) will follow very easily. So suppose that D = Di - D; is another
decomposition, and let ¢~ : V ---" ]p>r' and ¢; : V ---" ]p>" be morphisms asso-
ciated to Di and D; respectively. Then Di + D2 = Dl + D;. Lemma 10.6
says that there exists a morphism ¢ : V ---" ]p>n associated to this divisor,
and then two applications of (10.6b) yields
Therefore
which proves that up to bounded functions, the definition of hD is inde-

pendent of the choice of D 1 , D 2 , ¢l, ¢2.
(a) The divisor D = ¢* H is already associated to a morphism, so we can
write D as D = D - O. Note that the divisor 0 is associated to the trivial
morphism 7/J : V ---" ]p>o which maps V to a point. Then
hD = h¢ - h,p + 0(1) = h¢ + 0(1),

since h,p(P) = h p (7/J(P)) is a constant.
(b) We decompose each of the given divisors into a difference of divisors
that are associated to morphisms, say D = Dl - D2 and D' = D~ - D;.
Then Dl + Di and D2 + D; are also associated to morphisms (1O.6a), and
their difference is D + D', so we can use (a) and (1O.6b) to compute
hD+ D, = hD 1 +D'1 - hD 2 +D'2 + 0(1) = + hD' - hD -

hD 1 1 2
hD'2 + 0(1)
= hD + hD' + 0(1).
(c) Write D - D' as a difference of divisors associated to morphisms,
say D - D' = Dl - D 2, with Dl associated to ¢l and D2 associated to ¢2'
Note that Dl and D2 are linearly equivalent by assumption, so (10.4) tells
us that hCPl = h¢2 + 0(1). Now using (b) we obtain the desired result,
hD - hD' = hD- D, + 0(1) = hCPl - h¢2 + 0(1) = 0(1).

(d) By the linearity proven in (b), it suffices to prove (d) for a divisor D
associated to a morphism ¢ : W ---4 IP'T. Then the divisor 'Ij;* D is associated
to the morphism ¢ 0 'Ij; : V ---4 W ---4 IP'T. Using (a) twice, we find
hv,,p* D = hv,q,o,p + 0(1) = hw,q, o'lj; + 0(1) = hW,D o'lj; + 0(1).

Note that the middle equality is trivial, since hq,o,p = hpo¢o'lj; by definition.
D
PROOF (of Theorem 10.2). Let d = deg(D) and d' = deg(D'). Replacing D
by - D if necessary, we may assume that d :2: 1. For any integer n we
consider the divisor
Hn = (2g + I)D + n(d'D - dD').

Notice that deg(Hn) = d(2g + 1), so (10.3.1) says that Hn is a very ample
divisor on C. In particular, there is an embedding ¢n : C ---4 IP'T associated
to H n , so (1O.la) gives
for all P E C(k).
Now using the definition of H n , the linearity property of height func-

tions (10.lb), and the fact that the height on projective space is non-
negative [AEC, VIII.5.4b], we obtain the estimate
Note that the constant en depends on n, but it is independent of P. As-

suming hD(P) > 0, a little algebra then gives the inequality
Taking the liminf as hD(P) ---400, we obtain
..
hmmf
PEC(k), hD(P)-+OO
n (d'- - - - -
d
hD1(P)) >
hD(P) -
-2g+1
--.
d
This is true for every value of n (positive and negative), which gives the
desired result,
D
PROOF (of Theorem 10.3). Replacing D by nD and using the fact (1O.lb)
that hnD = nhD + 0(1), we may assume that D is very ample. Let ¢ :
V ----t lP'r be an embedding and H E Div(lP'r) a hyperplane with ¢* H = D.

Taking a finite extension of k if necessary, we may assume that ¢ is defined
over k. Then (lO.la) implies that there is a constant c so that
for all P E V(k).
It follows that ¢ maps the set
{p E V(k) : hD(P) :::; a and [k(P): k] :::; b}
injectively into the set
{Q E lP'r(k) : hJf'(Q) :::; a + c and [k(Q): k] :::; b}.
This last set is finite from [AEC, VIII.5.n], which proves the first part
of (10.3). The second part follows by setting b = 1. 0
§11. Specialization Theorems for Elliptic Surfaces
In this section we will prove a theorem of Tate which describes how the
canonical height h(O"P(t)) varies as one moves along a section of an elliptic
surface. As a corollary we obtain a theorem of Silverman, strengthening
earlier results of Neron, Dem'janenko, and Manin, which says that the
specialization homomorphism E(K) ----t Ct(k) is injective for all but finitely
many t E C(k).
Let 7r : C ----t C be a minimal elliptic surface with corresponding elliptic
curve Ej K, and let P E E(K). To ease notation, we will write
Pt = O"p(t)
for the image of a point t E C by the section O"p : C ----t C associated to P.
Theorem 11.1. (Tate [4]) Assume that the elliptic surface C ----t C is
defined over a number field k. For each t E C(k) such that the fiber Ct is
non-singular, let
(', ·)t : ct(k) x ct(k) ----+ lR
be the canonical height pairing on the elliptic curve Ct [AEC, VIII §9].
Fix two points P,Q E E(K)o, let [P,Q] E Div(C) be the divisor
described in (9.5), and let h[p,Q] : C(k) ----t lR be an associated height
function on C (10.1). Then
(Pt , Qt)t = h[p,Q] (t) + 0(1) for all t E C(k) such that Ct is non-singular.
Note that the 0(1) bound depends on P and Q but is independent oft.
Remark 11.1.1. Putting P = Q in (11.1) gives hE,(Pt ) = h(p,p](t)+O(l),
where
hE, : et(k) ----> ~
is the canonical height on e [AEC, III §9]. In other words, for any point P E
E(K)o, the map
C(k) ----- ~,
is a height function on C(k) corresponding to the divisor [P, Pl. Silver-
man [6] shows that it is possible to choose the height h(p,p] in such a way
that the difference hE,(Pt ) - h(p,p] (t) varies quite regularly as a function
of t. For example, consider the elliptic surface and section
Then there is a power series fez) E ~[zn with f(O) = 0 so that for all
sufficiently large integers t E Z,
For details, see Silverman [5,6].

Before beginning the proof of Tate's theorem (11.1), we need to de-
scribe how to use a height function on the surface G to compute canonical
heights on the individual fibers. For any integer n and any non-singular
fiber Gt, we will write
[n]t : et ----> Gt
for the multiplication-by-n map on et. These maps clearly fit together to
give a rational map on the surface G,
[n]: e ----> e, [n](x, y, t) = ([n]t(x, y), t).
(N.B. Even if the surface G is minimal, the rational map [n] : e ----> 8- will
generally not extend to amorphism.) With these preliminaries completed,
we are ready for the following lemma.
Lemma 11.2. Let 7r : e ----> C be an elliptic surface defined over a number
field k, and let hE,(o) : e(k) ----> ~ be a heigh! function on 8- associated to the
divisor (0) E Div(G). Then for all t E C(k) such that et is non-singular,
and all points (x, y) E et(k),
hE-,(x,y) = lim ~hE-

n '
(O)([n](x,y,t)).
n-----)(X)
§1l. Specialization Theorems for Elliptic Surfaces 267
PROOF. Fix a Weierstrass equation for C,

C:y2=x 3 +Ax+B, A,BEk(C).
For the moment we will restrict attention to points t E C(k) such that A
and B are defined at t and ~(t) i- O. Then et is obtained by evaluating A
and B at t.
For each such t we let Pt be the map
Pt(X, y) = x.
This x-coordinate function on Ct has a double pole at Ot, which means
that
We also write cPt : et --> e for the inclusion of the fiber et into the surface e.
Now let (x, y) E et(li:) be any point on a non-singular fiber. We use
standard properties of the Height Machine (10.1) to compute
h£,(o)(x, y, t)
= h£.(o)(cPt(x, y)) definition of cPt
= h£"q,;(O)(x, y) + 0(1) functoriality of height (lO.ld)
= h£,,(o,)(x, y) + 0(1) since cP;(O) = (Od
1
= "2h£,,2(O,J(X, y) + 0(1) additivity of height (10.lb)
1
= "2h£"p;(oo)(x, y) + 0(1)
1
= "2hIP1,(oo) (Pt(x,y)) +0(1) functoriality of height (10.ld)
1
= "2h(x) + 0(1) definition of Pt.
It is important to note that in this computation the 0(1) bounds will

depend on t. This dependence arises because we have used the mor-
phism cPt : et --> e, and the 0(1) in the functoriality property (lO.ld)
depends on the morphism. However, for a given t, the O(l)'s are indepen-
dent of the point (x, y) E et(k). To make the dependence visible, we will
write
for all (x, y) E ct(k).
For any point (x, y) E et(7;;) and integer n, write [nJt(x, y) = (xn' Yn).
Then standard properties of the canonical height [AEC, VIII.9.3b,ej allow
us to compute
1 1 1 A
-2 lim 2h(x n ) = lim 2{hc,([n]t(x,y)) +Ot(1)}

n---+oo n n--oo n
1 { 2
lim 2 n hc,(x, y) + Ot(l)}
= n---+oo A
n
= hc,(x, y).
Now combining these two formulas gives the desired result,
lim
n~CX)
~he
n '
(o)([n](x,y,t)) = lim
n---+oo
~he
n '
(O)(xn,Yn,t)
lim z
= n--<Xl n
1 (1-2 h (xn) + Ot(l) ) = -1 lim zh(x
2 n __ n
1 '
n) = hc,(x,
DC
y).
It is instructive to note that the Ot(l) disappears in the limit because the
"t-coordinate" of [n](x, y, t) is independent of n.
This completes the proof of (11.2) for all points t E Gek) such that A
and B are defined at t and ~(t) =I O. But by choosing different Weierstrass
equations for e, we can cover {t E G(k) Ct is non-singular} by finitely
many such sets. D
PROOF (of Theorem 11.1). The divisor [P, Q] is determined up to linear

equivalence by the relation (9.5a),
(P + Q) - (P) - (Q) + (0) '" 7r* ([p, Q]).
For any point Z E c(k) lying on a non-singular fiber et, we use standard
properties of the Height Machine (10.1) to compute
h[p,Q](t) = h[p,Q] (7r(z)) since Z E et

= h.".* [P,Q] (z) + 0(1) functoriality of height (lO.ld)
= h(p+Q)_(P)_(Q)+(O)(z) + 0(1) equivalence of heights (lO.lc)
= h(p+Q)(z) - h(p)(z) - h(Q)(z) + h(o)(z) + 0(1)
additivity of height (lO.lb)
= hr:p_Q(O)(z) - hr:p(o)(z) - hr:Q(o)(z) + h(o)(z) + 0(1)
where TR : e ....... c is translation-by-R
= h(o) (LP_Q(Z)) - h(o) (LP(Z)) - h(o) (LQ(Z))
+ h(o)(z) + 0(1) functoriality of height (lO.ld)
= h(o)( -Pt - Qt + z) - h(o)( -Pt + z) - h(o)( -Qt + z)
+ h(o)(z) + 0(1) since Z E et(k), so TR(Z) = R t + z.
Note that the 0(1) constants appearing in this calculation depend on P
and Q, but they are independent of Z and t = 7r(z). To indicate this
dependence, we will write Op,Q(l). For each pair of integers 1 :S i,j :S n,
we evaluate at the point z = iPt + jQt E e(k) to obtain
h[p,Q](t) = h(o)((i - l)Pt + (j -l)Qt) - h(Ol ((i -l)Pt + jQt)

-h(o) (iPt + (j - l)Qt) + h(o) (iPt + jQt) + Op,Q(l).
§1l. Specialization Theorems for Elliptic Surfaces 269
Summing this identity over 1 ::; i, j ::; n, we find that most of the terms
telescope, leaving
n 2 h[p,Q](t) = h(o) (nPt + nQt) - h(o) (nPt ) - h(o)(nQt)

+h(o)(Ot) + Op,Q(n 2 ).
Now dividing by n 2 , letting n -+ 00, and using (11.2) yields
. 1 . 1
h[pQ](t)
,
= hm zh(o)(nPt
n
n~O()
+ nQd - n--+O()
hm zh(o)(nPt
n )
- lim
n--+oo
~h(o)(nQt)
n
+ Op,Q(l)
= hc,(Pt + Qt) - hc,(Pt ) - hc,(Qt) + Op,Q(l)
= (Pt, Qt)t + Op,Q(l).
This completes the proof of Theorem 11.1. o
Taking the limit of (11.1) as the height of t goes to infinity, we can re-
cover the following result of Silverman which will be used below to prove the
injectivity of the specialization map. In the special case that the elliptic sur-
face e -+ G is split, this result had earlier been proven by Dem'janenko [1]
and Manin [2].
Corollary 11.3.1. (Silverman [1], [7]) Let e -+ G be an elliptic surface
defined over a number field k, fix two points P, Q E E(K), and let (P, Q)
denote the canonical height pairing (4.3,9.3) of P and Q on E(K). Further,
let ho : GCk) -+ ~ be a height function on G corresponding to a divisor 8 E
Div(G) of degree 1, and for each t E G(k) such that et is non-singular
let (', ')t be the canonical height pairing on et(k). Then
Notice that (11.3.1) applies to all points in E(K), not just those in the
subgroup E(K)o. It is possible to improve (11.3.1) as described in our next
result, but we will only give the proof in the case that the base curve G
is pl.
Corollary 11.3.2. (Tate [4]) Let e -+ G, P, Q E E(K), ho, and (., ')t
be as in (11.3.1).
(a) Suppose that the base curve G is isomorphic to pl. Then
(Pt , Qt)t = (P, Q)ho(t) + 0(1) for t E pI (k) with et non-singular.

Notice that in this result we can take ho to be the usual height function
on pl.
(b) For an arbitrary base curve C, we have
(Pt, Qt)t = (P, Q)h6(t) + O( y'h 6(t) ) for t E C(k) with Ct non-singular.
(See exercise 3.34 for the case that C has genus 1.)
We will prove (11.3.1) and (11.3.2) simultaneously.
PROOF (of Corollaries 11.3.1 and 11.3.2). The subgroup E(K)o has finite
index in E(K) from (9.4), so given any two points P, Q E E(K), we can
find an integer N such that N P, NQ E E(K)o. Note that N depends only
on P and Q. Further, the canonical height pairings on E(K) and Ct(k) are
bilinear, so
and (NP,NQ) = N 2 (P,Q).
Replacing P, Q by N P, NQ and dividing each of the formulas in (11.3.1)

and (11.3.2) by N 2 , we see that it suffices to prove the two corollaries for
points P, Q E E(K)o.
Assuming now that P and Q are in E(K)o, (9.5c) tells us that
(P, Q) = deg[P, Q].
Hence the divisor
(3 = [P, Q] - (P, Q)8 E Div(C)
is a divisor of degree o. Using (11.1) and the additivity of heights (lO.lb)

yields the estimate
We consider three cases.

First, to prove (11.3.1), we divide by h6(t) and take the limit
Now (10.2) and the fact that deg((3) = 0 imply that
which completes the proof of (11.3.1).

Next, suppose that C = ]p>l. Then (3 ,....., 0, since two divisors in DiV(]P>l)
are linearly equivalent if and only if they have the same degree [AEC, 11.3.2].
The equivalence property of heights (lO.lc) implies that hf3(t) is bounded,

so the above estimate becomes
This finishes the proof of (l1.3.2a).

We will not give the proof of (l1.3.2b), other than to say that one uses
an estimate
hf3(t) = O( y'hti(t) ) + 0(1),
valid for divisors (3 of degree o. This in turn follows from properties of the
canonical height on the Jacobian variety of the curve C. For the complete
proof, see Lang [4, 12 corollary 5.4] or Tate [4]. The special case that C
has genus 1 is discussed in exercises 3.33 and 3.34. 0
Let e -+ C be an elliptic surface. Each point P E E(K) defines a

morphism Up : C -+ e which we have been denoting by t ~ Pt. Turning
this around, we can also say that each point t E C(k) determines a map
called the specialization map of e at t. If the fiber et is non-singular, then

it is clear that the specialization map is a homomorphism,
This follows from the fact that on a non-singular fiber et, the section ap+Q
is defined by the relation ap+Q(t) = ap(t) + aQ(t). We now show that for
"most" values of t the specialization homomorphism is injective.
Theorem 11.4. (Silverman [1], [7]) Let e -+ C be a non-split elliptic

surface defined over a number field k, and let {) E Div( C) be a divisor of
positive degree. Then there is a constant c > 0 so that
at : E(K) -+ et(l~) is injective for all t E C(k) satisfying h6(t) ~ c.
(One says that the set of points where at fails to be injective is a set of
bounded height.) In particular, the specialization map at : E(K) -+ et(k)
is injective for all but finitely many points t E C(k).
Remark 11.4.1. In the case that the elliptic surface is split, there is a
version of (11.4) due to Dem'janenko [1] and Manin [2]. Both Dem'janenko
and Manin used their results to prove the Mordell conjecture (now Faltings'
theorem) for certain curves. See exercise 3.16.
Remark 11.4.2. There is an earlier result, due to Neron [3] using a Hilbert
irreducibility argument, which says that if C(k) is infinite, then there are in-
finitely many t E C(k) for which the specialization map at : E(K) ---> Gt(k)
is injective. This is sufficient for one of the main applications of (11.4),
namely the construction of elliptic curves of elevated rank over Q or over
number fields k. The idea is to find an elliptic surface G ---> Cover k for
which C(k) is infinite and such that E(K) has high rank. Then specializ-
ing t E C(k) gives elliptic curves over k of high rank. Neron [3] used this
procedure to construct infinitely many elliptic curves over Q with rank at
least 10. More recently, Mestre [2] constructed an elliptic surface G ---> ]P'l
defined over Q so that E(Q(t)) has rank at least 12, and Nagao [2] extended
this result to get rank at least 13. By taking particular values for t E Q,
it is possible to find specific elliptic curves over Q with even higher ranks.
See Fermigier [1], Nagao [1], and Nagao-Kouya [1] for examples with ranks
at least 19, 20, and 21. And the quest continues!
PROOF (of Theorem 11.4). Our assumption that G ---> C is non-split means
that the Mordell-Weil theorem (6.1) is valid, so the group E(K) is finitely
generated. In particular, the torsion subgroup E(K)tors is finite.
Let P E E(K) be any non-zero point. Then there are only finitely
many t E C(k) for which Pt = Ot, since the two divisors (P) and (0)
intersect in only finitely many points. This holds for each of the finitely
many points in E(K)tors, so we see that on torsion points, the specialization
map
at : E(Khors ---> Gt(k)
is injective for all but finitely many t E C(k). (In fact, the specialization
map is injective on torsion whenever Gt is non-singular, since the residue
field k has characteristic O. This follows from the identification of the
kernel of the specialization map with the formal group of the elliptic curve;
see [AEC, IV.3.2b, VII.2.2].)
Next let pl, ... ,pr E E(K) be generators for the free part of E(K);
that is, pl, ... , pr give a basis for the free group E(K)/ E(Khors. Then
the non-degeneracy of the canonical height pairing on E(K) described
in (4.3cd) implies that
(This is an elementary property of non-degenerate bilinear forms; see Lem-

ma 11.5 below.)
Next we specialize the pi'S to the fiber Gt, take the height regulator
of the resulting pts, and use (11.3.1) to compute
· det ((pit, pj) )

t t l <i,j<r
((
l'1mp it, pj) )
11m = d et t t
h6(t)->OO hlj(t)r h6(t)->OO hlj(t) ..
l::;',J::;r
= det((pi,pj))l::;i,j::;r
#0.
Hence there is a constant c so that
for all t E C(k) with hlj(t) > c.
It follows from (11.5) and the non-degeneracy of the height pairing on Ct(k)
[AEC, VIII.9.3 or VIII.9.6j that the points Pl, ... ,P; are linearly indepen-
dent provided that hlj (t) > c.
Adjusting c if necessary to account for the finitely many points in
the torsion subgroup E(K)tors, we have now proven that for all t E C(k)
with hlj(t) > c, both of the specialization maps
and
are injective. Now a simple diagram chase using the commutative diagram
o ---> E(K)tors ---> E(K) ---> E(K)jE(Khors ---> 0
1 1 1
---> 0
shows that E(K) ---t Ct(k) is injective, which completes the proof of the
first part of (11.4).
The second part is then an immediate consequence of the first part and
of (10.3) once we observe (10.3.1) that on a curve, any divisor of positive
degree is ample. 0
It remains to prove the elementary property of non-degenerate bilinear

forms used in the proof of (11.4).
Lemma 11.5. Let r be a free abelian group, let (', .) be a positive
definite bilinear form on r with values in Q, and let Xl,"" Xr E r. Then
Xl,' .. , Xr are linearly independent <==} det ((Xi, Xj)) l::;i,j::;r # O.
PROOF. Suppose first that the determinant is O. This means that there
are integers aI, ... , ar, not all zero, so that
r
Lai(Xi,Xj) =0 for all 1 ::::: j ::::: r.

i=l
Multiplying by aj, summing over j, and using the bilinearity gives
r r r r
0= 2..:aj 2..:ai(Xi,Xj) = (2..:aixi,2..:ajXj).

j=1 i=1 i=1 j=1
In other words, y = L: aiXi E r satisfies (y, y) = 0, so the positivity of the

bilinear form implies that y = o. Hence the Xi'S are linearly dependent.
Conversely, if the Xi'S are linearly dependent, say L: aixi = 0, then
the linearity of the pairing implies that the rows of the matrix ((Xi, Xj))
are linearly dependent, so the determinant is zero. 0
§12. Integral Points on Elliptic Curves over Function Fields
There is a theory of S-integral points on elliptic curves over function fields

which is completely analogous to the theory over number fields as described
in [AEC, Ch. IX]. However, for function fields it is possible to prove much
stronger results using relatively elementary techniques. In this section
we will give a short and elegant proof of the analogue of Siegel's theo-
rem [AEC, IX.3.2.1] which asserts that an elliptic curve has only finitely
many S-integral points. We will also state and briefly sketch the proof of
an effective version of this result.
The simplest function field analogue of integral points are "polynomial
points." Thus let E / k(T) be an elliptic curve over a rational function field,
say given by a Weierstrass equation
E : y2 = X3 + A(T)x + B(T) with A(T), B(T) E k[T].
The the set of polynomial points of E is the set
{P = (x, y) E E(k(T)) : x, y E k[TJ}.
For example, if
and P=(T,T),
then (1.1.1) says that P and 2P are polynomial points, but 3P is not.
One way to characterize the polynomial ring k[T] is to observe that it
is the sub ring of k(T) consisting of functions with no (finite) poles. More
generally, we define the ring of S-integers of an arbitrary function field in
the following way.
Definition. Let K = k( C) be the function field of a curve, and let 8 c C

be a non-empty finite set of points of C. The ring of 8-integers of K is the
ring
°
Rs = {J E K : ordt(J) ~ for all t tf. 8}.
Here ordt(J) is the order of vanishing of f at t; see [AEC, II §1].
The following function field analogue of Siegel's theorem is a special
case of a result of Lang [8], who proved a general finiteness theorem for
integral points on curves of arbitrary genus over function fields.
Theorem 12.1. Let K = k(C) be the function field of a curve, let 8 c C
be a non-empty finite set of points of C, let E / K be an elliptic curve that
does not split over k, and let F E K(E) be a non-constant function on E.
Then
{p E E(K) : F(P) E Rs}
is a finite set.
PROOF. Our first observation is that it suffices to prove (12.1) for the
special case that F is taken to be the x-coordinate on some Weierstrass
equation for E / K. The reduction from the general case to this special case
is given for number fields in [AEC, IX.3.2.2]' but the proof is the same for
function fields. So we are reduced to showing that
{p E E(K) : x(P) E Rs}
is a finite set.
H P is a point in this set, then the height of P is a sum of local
contributions coming from the points in 8. More precisely, we have
h(P) = h(x(P)) definition of h(P)

=L max{ - ord t (x(P)), o} from (4.1)
tEe
= Lmax{ -ordt(x(P)),O} since ord t (x(P)) ~ 0 for t tf. 8

tES
:s: #8· max{

tES
- ord t (x(P))}.
Further, our assumption that E does not split combined with (5.4) tells
us that E(K) has only finitely many points of bounded height. Hence the
following result (12.2) completes the proof of (12.1). 0
Theorem 12.2. (Manin [3]) Let K = k( C) be the function field of a

curve, let t E C be any point of C, and let E / K be an elliptic curve that
does not split over k, say given by a Weierstrass equation
E : y2 = x 3 + Ax + B.
Then the function ordt(x(P)) is bounded below as P ranges over E(K).

Before giving the proof of (12.2), we want to observe just how strong
a statement it is. For example, the number field analogue of (12.2) is
certainly false. Thus, if we let p E Z be a prime and E /Q be an elliptic
curve with a rational point
P E E(Q) satisfying ordp{x(P)) < 0,

then consideration of the formal group of E(Qp) shows immediately that
ordp{x(pnp)) -+ -00 as n -+ 00.
The following short proof of (12.2), which is due to Voloch [1], uses
the formal group and depends crucially on the fact that the base field k
has characteristic O.
PROOF. (of Theorem 12.2, Voloch [1]) We may replace the constant field k
by its algebraic closure, since this will only have the effect of making E(K)
larger. Further, replacing x by u 2 x for some u E K*, we may assume that
the given Weierstrass equation is a minimal equation for the valuation ord t .
For each integer n 2: 1, let
En(K) = {p E E(K) : ordt{x(P)) ::; -2n}.

This is the standard filtration on the formal group of E, see [AEC, Ch. IV].
The crucial facts to note here are that each En(K) is a subgroup of E(K),
and each quotient group En(K)/ E n+l (K) is isomorphic to a subgroup of k.
To see this, let K t be the completion of the field K for the valuation ord t ,
let R t be the ring of integers of K t , let M t be the maximal ideal of R t , and
let E be the formal group of E/Kt . It follows from [AEC, IV.3.1.3] that
En(Kt) ~ E(Mr), and then [AEC, IV.3.2(a)] tells us that
En(Kd/En+I(Kt) ~ E(Jvq)/E(M~+1) ~ M~/M~+1 ~ k.
(The last isomorphism uses Rt/M t ~ k.) Now the fact that En(K)
E(K) n En(Kt) immediately implies our two assertions that En(K) is a
subgroup of E(K) and En(K)/En+1(K) is isomorphic to a subgroup of k.
In particular, our assumption that k has characteristic zero means that the
quotient groups En(K)/ En +1 (K) have no elements of finite order.
Suppose now that ordt{x(P)) is not bounded below on E(K). Then
we can choose a sequence of points PI, P2 , ... with
and
We claim that the points PI, P 2 , ... are linearly independent, which will
contradict the Mordell-Weil theorem (6.1) and thus complete the proof
of (12.2).
Suppose to the contrary that Pl , P2 , ... are linearly dependent. Dis-

carding the first few Pi'S if necessary and relabeling, we may assume that
there is a relation
with mI =1= O.
Using the fact that nI < n2 < ... < nr and that the En(K)'s are nested
subgroups of E(K), we see that
But as noted above, the quotient Enl (K) / Enl +1 (K) is isomorphic
to a subgroup of k, and k is a field of characteristic zero. So the fact
that mI PI is zero in this quotient implies that PI itself is zero in the
quotient. In other words, PI E En1+I(K), which contradicts the fact that
ordt{x(Pd) = -2nI. Hence the Pi'S are independent. 0
The proofs of (12.1) and (12.2) are ineffective because they depend on
the Mordell-Weil theorem (6.1). Notice that (12.2) is analogous to Siegel's
theorem [AEC, IX.3.1]' although (12.2) is both stronger and considerably
easier to prove.
Similarly, one can prove effective bounds for S-integral points in the
function field case which are analogous to the bounds provided by linear-
forms-in-Iogarithms methods for number fields [AEC, §5J. Again the func-
tion field estimates are stronger and much easier to prove. A number of
people have given such bounds, including for example Schmidt [1], Ma-
son [1], and Hindry-Silverman [2]. We will briefly sketch the proof of the
following version. The argument is the same for elliptic or hyperelliptic
curves, so we give the more general case.
Theorem 12.3. Let K = keG) be the function field of a curve G of
genus g, let S c G be a non-empty finite set of points of G, let Rs be the
ring of S-integers of K, and let f(x) E Rs[x] be a monic polynomial with
discriminant ~ satisfying ~ E R'S. Suppose that x, y E Rs satisfy
y2 = f(x).
Then
h{y4 /~) :s: 4n(n - 1) max{2g - 2 + #S, a}.
(Recall from §4 that the height of an element f E K is defined to be the
degree of the map f : G - t ]P'I. Also note that the set S has to be chosen
large enough to contain all of the zeros and poles of ~.)
Remark 12.3.1. The bound in (12.3) is stated for y4 / ~ because this
quantity is invariant under linear change of variables. However, it is easy
to use (12.3), the relation y2 = f(x), and elementary properties of height
functions to give a bound for hex) in terms of the coefficients of f. See
exercise 3.39 for the particular case of an elliptic curve.
PROOF. (Sketch of Theorem 12.3) The first step is to reduce the problem
of S-integral points on elliptic curves to the problem of solving the S-unit
equation
u +v = 1, u,v E R~.
This reduction procedure is due to Siegel and is described in [AEC, IX.4.3].

Next, one proves that if u, v E R~ satisfy u + v = 1, then
h(u) ::; max{2g - 2 + #S, a}.
This bound is the function field analogue of the abc-conjecture of Masser

and Osterle. There are several elementary proofs available; see for exam-
ple Mason [1], Silverman [8], or Vojta [lJ. Tracing back through Siegel's
argument gives the estimate described in (12.3). We leave the details to
the reader, or see Hindry-Silverman [2, Prop. 8.2]. 0
EXERCISES
3.1. For any pair (a, b), let E a •b be the curve
Ea,b : y2 + axy + by = x 3 + bx 2 .
Notice that the point (0,0) is on each of these curves.

(a) Let Elk be an elliptic curve, let P E E(k) be a point, and assume that
P, 2P, 3P =1= O. Prove that E has a Weierstrass equation of the form Ea,b
with P corresponding to (0,0). (Hint. Move P to (0,0), rotate so that the
tangent line at (0,0) is the x-axis, and make a dilation to get a2 = a3.)
(b) Prove that 5P = 0 if and only if a = b + 1. Conclude that every
elliptic curve Elk with a point P of exact order 5 is isomorphic to some
fiber of the elliptic surface
C : y2 + (t + 1 )xy + ty = x 3 + tx 2
by an isomorphism taking P to the point (0,0) on that fiber.

(c) More precisely, if E is defined over k and if P E E(k) is a point
of exact order 5, prove that there is a unique point to E ]Pl(k) and an
isomorphism Cta defined over k such that <P(P) = (0,0).
(d) Using a similar construction, find an elliptic surface C -t ]pI which
classifies elliptic curves with a given point of order 7.
Exercises 279
3.2. Let G IQ be the (elliptic) curve
let EIQ(G) be the elliptic curve
E: y2 + (st + t - S2)xy + s(s - l)(s - t)t 2y = x 3 + s(s -l)(s - t)tx 2,

and let P = (0,0) E E(Q(G))[ll] be the point of order 11 as described
in (1.1.3).
(a) Let A be an elliptic curve and Q E A a point of order 11. Prove that
there is a unique point (so, to) E G such that if we substitute (s, t) = (so, to)
into the equations for E and P, then we obtain an elliptic curve Eo and a
point Po E Eo of order 11 such that there is an isomorphism ¢ : A --> Eo
satisfying ¢(Q) = Po.
(b) If A is defined over k and Q E E(k), prove that the point (so, to)
obtained in (a) will lie in G(k).
3.3. This exercise gives a function field analogue of [AEC, exercise 8.1]. Let Glk
be a curve of genus 9 with function field K = k( G), and let ElK be an
elliptic curve which does not split over K.
(a) Suppose that ElK has a Weierstrass equation of the form
Let 8 c G be the set of points where anyone of e1, e2, e3 has a pole together
with the points where the product (e1 -e2) (e1 -e3)( e2 -e3) vanishes. Prove
that
rank E(K) s:
4g + 2#8 - 2.
(b) Suppose that ElK has a Weierstrass equation of the form
with A,B E K.
Let 8 c G be the set of points where A or B has a pole together with the
points where ~ = 4A 3 + 27B2 vanishes. Find an explicit bound for the
rank of E(K) in terms of 9 and #8.
3.4. Let klQ be a finitely generated field extension, that is, k = Q( 01, ... ,Or)
for some 01, ... ,Or E C. Let G Ik be a curve with function field K = keG),
and let ElK be an elliptic curve. Prove that E(K) is a finitely generated
group. In particular, this is true if k is a number field. (Hint. You may
find exercise 3.15 below useful for doing this problem.)
3.5. Let A be an abelian variety of dimension one, so in particular A is a non-
singular projective curve. (See §2 for the definition of abelian variety.)
Prove that A has genus 1, so A is an elliptic curve.
3.6. Let ¢ : V --> W be a rational map of projective varieties.
(a) Prove that the image ¢(V) is an algebraic subset of W.
(b) Suppose that V is non-singular. Prove that the set of points where ¢
is not defined has codimension at least two in V.
3.7. Recall that a topological space X is irreducible if it cannot be written as a

union X = Xl U X 2 of non-empty closed subsets of X.
(a) Let X be a topological space, and let Z C X be a subset taken with
the induced topology. If Z is irreducible, prove that the closure of Z in X
is also irreducible.
(b) Let <p : X -; Y be a continuous map of topological spaces. If X is
irreducible, prove that its image <b(X) is irreducible.
(c) Use (a) and (b) to deduce the second part of Proposition 3.5(a).
3.8. Let G -; C be an elliptic surface over k. Define a map
More precisely, define jc(t) to be the j-invariant of the elliptic curve Gt

provided that the fiber Gt is non-singular, and for the moment leave it
undefined for the remaining points of C. Prove that j[ is an algebraic
map, and conclude that it extends to a morphism from C to pl.
3.9. Let G -; C be an elliptic surface over k, and let jc : C -; pI be the
morphism defined in exercise 3.8.
(a) If G -; C splits over k, prove that jE. is a constant map.
(b) Give an example (with proof) of an elliptic surface G -; C that does
not split over k for which j[ is a constant map.
3.10. Let G -> C be an elliptic surface over k, and let ]r : C -> pI be the
morphism defined in exercise 3.8. Choose a \Veierstrass equation
G: y2 + alXY + a3Y = X3 + a2x 2 + a4X + a6
for G, where aI, ... ,a6 E k( C), and let C4, C6 be the usual associated quan-
tities [AEC III §1]. Prove that G -; C splits over k if and only if one of the
following three conditions is true.
(i) jl:(C) = {O} and C6 is a sixth power in k(C).
(ii) jE(C) = {1728} and C4 is a fourth power in k(C).
(iii) jdC) = {a} with a f= 0, 1728, and C6/C4 is a square in k(C).
(Keep in mind we are assuming that char(k) = 0, although this exercise
remains true if char(k) o?: 5.)
3.11. Let E / K be an elliptic curve defined over a function field K = k( C) by an
equation of the form
E : y2 = x:; + A.x + B.
Further define the "height of E" to be h(E) = h(A. 3 ) + h(B2).
(a) Prove that for all P E E(K),
4h(P) - ::1h(E) ::; h(2P) ::; 4h(P) + h(E).
(b) Prove that for all P,Q E E(K),
4h(E) ::; h(P + Q) + h(P - Q) - 2h(P) - 2h(Q) ::; h(E).
(c) Prove that for all P E E(K),
liL(p) - ~h(p)1 ::; ~h(E).

(In all three parts of this problem, the constants in front of the h( E)'s are
far from best possible. See how much you can improve them.)
Exercises 281
3.12. Let EjCCT) be the elliptic curve
Find all points in E( CCT») of the form
by substituting P into the equation for E and solving for ao, . .. ,C1. How
many of these points are defined over Q(T)?
3.13. With notation as in the statement of Lemma 5.5.2, prove that if
then the image of Vv in E(K) contains E(K, d). This provides a strength-
ened version of (5.5.2).
3.14. Let V be an irreducible projective variety defined over an algebraically
closed field, and let ,/,1, '/'2 E V be distinct points. Prove that there exists
an irreducible curve reV with ,/,1, '/'2 E r.
3.15. This exercise describes the Mordell-Weil theorem (6.1) for split elliptic
surfaces. Let Cjk be a curve, let Eojk be an elliptic curve, and let c =
Eo x C be the corresponding split elliptic surface. Further, let Map k ( C, Eo)
be the set of morphisms from C to Eo defined over k. We use the group
structure on Eo to make MaPk(C, Eo) into a group in the usual way, (c/J +
'IjJ)(t) = c/J(t) + 'IjJ(t). Notice that the constant maps in MaPk(C, Eo) form a
subgroup isomorphic to Eo(k).
(a) Prove that there is a natural isomorphism c(Cjk) ~ Mapk(C, Eo). In
particular, the group of sections c(Cjk) contains a subgroup isomorphic
to Eo(k).
(b) Prove that the quotient group c(Cjk)jEo(k) is finitely generated.
(c) If k is a number field, prove that c(Cjk) is finitely generated.
3.16. Let k be a number field, and let Cjk, Eojk, c = Eo x C, and MaPk(C, Eo)
be as in the previous exercise.
(a) Let 0 E Div(C) be a divisor of degree 1, and fix a height function h6 :
C(k) -+ IR associated to o. Prove that for any map c/J E MaPk(C, Eo),
lim hEo(c/J(t») = deg(c/J).

tEC(k). h6(t) ..... oo h6(t)
(b) Fix a basepoint to E C(k). Prove that there is a constant c so that

if t E C(k) satisfies h6(t) 2: c, then the map
{c/J E MaPk(C,Eo) : c/J(to) = O} ---> Eo(k)

c/J ---> c/J(t)
is an injective homomorphism.
(c) Suppose that the quotient group MapdC, Eo)/Eo(k) has free rank r,
say generated by the maps (PI, ... ,cPT: C ----> Eo. Suppose further that the
group Eo(k) has rank strictly less than r. Prove that C(k) is finite.
(d) Fix an element b E k*, and let C / k and E o/ k be the curves
2
Eo : y z = x
3
+ bz 3 .
Prove that the group Map k (C, Eo) / Eo (k) has rank at least two by showing
that the maps
2 3 3 2 3 3
cPI ( [X, Y, Zl ) = [-X Z,Y ,Z], cP2 ( [X, y, Zl ) = [- Y Z, X ,Z 1
give independent elements. Use this to prove that if rank Eo(k) :S 1, then
C(k) is finite.
3.17. Let E be an elliptic curve defined over Q1(T) that does not split over C(T).
(a) Suppose that kdQl and k2/Q1 are fields with the property that
E(kl(T)) = E(C(T)) and
Let k = kl n k 2 . Prove that E(k(T)) = E(C(T)). Deduce that there is a

smallest field with this property. We will call this field the field of definition
for E(C(T)) and will denote it by k E .
(b) Prove that kE is a finite extension of Q1.
(c) More precisely, find an explicit bound for the degree [k E Q1l that
depends only on the rank of E(C(T)).
(d) *Fix a Weierstrass equation for E of the form
E : y2 = x 3 + A(T)x + B(T)
with A(T), B(T) E Z[T], and let .6..(T) = 4A(T)3 + 27 B(T)2. Prove that
the extension kE /Q1 is unramified except possibly at 2, 3, and the primes
dividing the discriminant of the polynomial .6..(T) E Z[T].
3.18. (a) Let f E k(JP'2). Prove that deg(div(f)) = 0, and deduce that the
degree map deg : Pic(JP'2) ----> Z described in (7.1) is a well-defined homo-
morphism.
(b) Prove that the degree map deg : Pic(JP'2) ----> Z is an isomorphism.
(c) Generalize ( a) and (b) to JP'H.
3.19. Let P = (0,0) E A 2 , and let h, h E k(JP'2) be rational functions satisfying
h(P) = h(P) = 0. Let f] and f2 be the curves h = and h = ° °
respectively. Prove that f 1 and f 2 intersect transversally at P if and only
if the following three conditions are true:
(i) f 1 is non-singular at P;
(ii) f2 is non-singular at P:
(iii) the tangent line to flat P is distinct from the tangent line to f2 at P.
3.20. For each of the following curves f 1, f2 C JP'2, calculate the local intersection
index (fl' f2)P at the point P = [0,0,1]. (Hint. First de homogenize by
setting Z = 1.)
(a) f 1 :y 2 Z=X 3 +X 2 Z, f 2 :Y=0.
(b) f 1 :y 2 Z=X 3 +X 2 Z, f 2 :Y=X.
(c) fl :y 2 Z=X:l , f 2 :Y=X.
(d) f l :y 2 Z=X:l +X 2 Z, f 2 :y 2 Z=X 3 .
Exercises 283
3.21. *Let 7r : S -> C be a fibered surface. Prove that every fiber St is connected.
(You may want to take k = IC.)
3.22. Let 7r : S -> C be a fibered surface, let reS be a non-singular irreducible
horizontal curve, and let ¢ : r -> C be the restriction of 7r to r.
(a) Let PEr and t = ¢(P) E C. Prove that
(r· 7r*(t»)p = ep(¢),
where ep(¢) is the ramification index of ¢: r -> Cat P (see [AEC, II §2]).
(b) Prove that r . 7r* (8) = deg( ¢) deg( 8) for all divisors 8 E Div( C).
(c) 'Prove that r . 7r*(8) = deg(¢) deg(8) remains true even if the irre-
ducible horizontal curve r is allowed to be singular.
3.23. Let 7r : S -> C be a fibered surface, and let 8 E Div( C) be a divisor of
degree O. Prove that D . 7r*8 = 0 for every divisor D E Div(S). (Hint.
Use (8.1) and the previous exercise.)
3.24. Generalize (8.3.1) as follows. Let 7r : S -> C be a fibered surface, and
suppose that the fiber St consists of n components arranged in the shape
of an n-gon with transversal intersections. In other words,
. { 1 ~f ~ == ~ ± 1 (mod n),
7r*(t) = ro+r 1 + .. ·+rn- 1 with ri·rj = -2 If l =),
o otherwise.
(a) Draw a picture illustrating this fiber, and show that the self-intersec-
tion values r; = -2 follow from the values of r i . r j for i i j.
(b) Let I = (ri . rj)oS;i,jS;n-l be the incidence matrix of the fiber, and
let 100 be the minor obtained by deleting the first row and column from I.
Find the value of det(Ioo) in terms of n.
(c) Let k be an integer between 0 and n - 1, and let D E Div(S) be a
divisor satisfying
D· ro = -1, D· r k = 1, D· r i = 0 for i i 0, k.
Find a fibral divisor
n-l
v = L airi such that (D + v) . r i = 0 for all 0:::; i :::; n - 1;
i=l
that is, find an explicit formula for ai in terms of i, k, and n.

3.25. Let 7r : S -> C be a fibered surface.
(a) Let r E Div(S) be an irreducible fibral divisor. Prove that there exists
an irreducible horizontal divisor D E Div(S) satisfying D· r > O.
(b) Prove that there exists a horizontal divisor D E Div(S) with the prop-
erty that D . r > 0 for every irreducible fibral divisor r E Div(S).
(c) Let DE Div(S) be a divisor as in (b), and let t E C. Use the Nakai-
Moishezon criterion (10.3.2) to prove that the divisor D + n7r*(t) is ample
for all sufficiently large integers n.
3.26. Let 7r : £ -+ C be a minimal elliptic surface with associated elliptic curve

ElK. Define an action of E(K) on Div(£) by having P E E(K) send a
divisor D to the divisor TP(D).
(a) Prove that the action of E(K) on Div(£) descends to give a well-defined
action of E(K) on Pic(£).
(b) For P, Q E E(K), let <1>P,Q E Div(£) be a fibral divisor with the
property (P + Q) - (P) - (Q) + (0) '" <1>P,Q' (See (9.5).) Prove that for a
fixed Q E E(K), the map
E(K) --+ Pic(£),
is a one-cocycle from E(K) to Pic(£), where Pic(£) is an E(K)-module as

described in (a). (See (9.5) for a stronger result for E(K)o.)
3.27. Let P E E(K). Prove that P E E(K)o if and only if (P) . F = (0) . F for
every fibral divisor FE Div(£).
3.28. Let 7r : S -+ C be a fibered surface.
(a) Let reS be a curve with the property that 7r : r -+ C is an iso-
morphism. Prove that r . 7r*(t) = 1 for all t E C. Use this to deduce
that
r· 7r*(o) = deg(o) for all divisors 0 E Div(C).
(b) Fix a point t E C. Prove that the image of a section (J" : C -+ S

intersects exactly one component of the fiber St.
(c) Let F c Div(S) be a fibral divisor, and suppose that F = 7r*o for
some 0 E Div(C)0Q. Suppose further that there exists a section (J" : C -+ S.
Prove that 0 E Div(C).
(d) Let 0 E Div(C) be a divisor such that 7r*o is a principal divisor on £.
Prove that 0 is a principal divisor on C. Deduce that 7r induces an injective
homomorphism 7r* : Pic (C) -+ Pic(S). (This is easier if you assume that
there exists a section (J" : C -+ S.)
3.29. Let 7r : £ -+ C be an elliptic surface, let ElK be the associated elliptic
curve over the function field K = k( C) of C, and let PI, .. . ,Pr E E(K).
Prove that if
is linearly equivalent to a fibral divisor, then
This gives the converse to Proposition 9.2.

3.30. Let S be a non-singular surface, let D E Div(S) be a divisor, and let H E
Div(S) be an ample divisor.
(a) Use the Nakai-Moishezon criterion (10.3.2) to prove that the divi-
sor nH + D is ample for all sufficiently large integers n.
(b) Use (a) to prove (10.5) for surfaces; that is, prove that D can be written
as the difference of two very ample divisors on the surface S.
Exercises 285
3.31. Let Vjk be a non-singular projective variety defined over a number field,
let D E Div(V) be a positive divisor, and let hD : V(k) ---> lR be an
associated height function. (A divisor is positive if it can be written as a
sum 2: niDi, where the Di'S are irreducible subvarieties of V and the ni's
are positive.)
(a) Prove that there is a constant c = c(V, D, hD) such that
hD(P) ~ c for all P E V(k), P rt D.

(Note that c need not be positive.)
(b) Give an example to show that (a) need not true for all P E V(k).
(Hint. Take D to be the exceptional curve on jp'2 blown up at one point. A
harder example is to let D be the diagonal in C x C, where C is a curve of
genus 9 ~ 2.)
3.32. Let V j k be a non-singular projective variety defined over a number field,
and let D, H E Div(V) be divisors with H ample and D algebraically
equivalent to zero (see Hartshorne [1, exercise 1. 7]). Prove that
(This generalization of (10.3) is due to Lang.)

3.33. Let E j k be an elliptic curve defined over a number field k. For any divi-
sor f3 = 2: bi(Pi ) E Div(E), we define the canonical height h{3 associated
to f3 by the formula
Thus the usual canonical height h is the height associated to the divi-
sor (0).
(a) If the divisor f3 is symmetric, that is, [-1]*f3 = f3, prove that
(b) If the divisor f3 is anti-symmetric, that is, [-1]*f3 = -f3, prove that
Also prove in this case that the map h{3 : E(K) ---> lR is a homomorphism.
(c) Let f3 E Div(E) be a divisor of degree O. Prove that there is a con-
stant c = c(E, (3) so that
for all P E E(k).

3.34. Let c ---> C be an elliptic surface defined over a number field k, and fix two
points P, Q E E(K). Suppose that the base curve C is an elliptic curve, and
let he be the canonical height on C. With notation as in (11.1) and (11.3),
prove that
(Pt,Qt)t = (P, Q)hc(t) + o( Vhc(t) )

for all t E C(k) with Ct non-singular.
This is a special case of (11.3.2b).

3.35. Let K = k(C) be the function field of a curve over an algebraically closed
field k, let ElK be an elliptic curve, and choose a Weierstrass equation
for ElK of the form
for some A, BE K.
Let ~ = 4A 3 + 27 B2, and for each point t E C let
(Here [r] is the greatest integer in r.) Define the minimal discriminant
divisor of ElK to be the divisor
'D E/ K = L nt(t) E Div(C).

tEC
(This is the analogue of the minimal discriminant ideal for an elliptic curve
defined over a number field [AEC, VIII §8].)
(a) Prove that'D E / K is a positive divisor and that it is independent of the
choice of the Weierstrass equation for ElK.
(b) Let C ---> C be a minimal elliptic surface associated to E. Prove that
the fiber et is non-singular if and only if ordt('D E / K ) = o.
(c) Prove that if 'D E / K = 0, then the j-invariant j(cd is constant.
(d) Prove that if 'D E / K = 0 and C = jp'l, then £ splits over C.
(e) Let Clk and ElK be given by
Prove that 'DE / K = 0 and that the associated elliptic surface e ---> C does
not split.
(f) For the example in (e), show that P = (u 2 ,uv) E E(K) is a point of
infinite order.
Exercises 287
3.36. We continue with the notation from the previous exercise, with the addi-
tional assumption that E does not split over K. Let 9 be the genus of C.
For each point t E C, we take a Weierstrass equation for E that is minimal
for the valuation ord t and we let E t be curve obtained by evaluating the
coefficients at t. We define integers it by
0 if E t is non-singular (good reduction),
it = { 1 if E t has a node (multiplicative reduction),
2 if E t has a cusp (additive reduction).
Then the conductor divisor of E j K is defined to be the quantity
fE/K = L ft(t) E Div(C).

tEG
(For another description of the conductor, see (IV §1O).)

(a) Prove that ordt(fE/ K) ::; ord t CD E/ K) for all t E C, and deduce that
deg(fE/K) ::; deg(DE/K).
(b) Prove that

deg(DE/K)::; 6deg(fE/K) + 12(g - 1).
This inequality is a precise function field analogue of Szpiro's conjecture
(IV.1O.6). It was originally discovered by Kodaira (see Shioda [3, Proposi-
tion 2.8]) long before Szpiro formulated his conjecture.
3.37. -This exercise is the function field analogue of Lang's conjectural lower
bound for the canonical height [AEC, VII1.9.9]. We continue with the
notation from the previous two exercises. Let 9 be the genus of C, and
let h: E(K) -> lR. be the canonical height on E (4.3, 9.3).
(a) Prove that there is a constant Cl (g) > 0, depending only on g, such that
if P E E(K) is a point of infinite order, then h(P) :2: Cl(g) degD E / K .
(b) Prove that there is an absolute constant C2 > 0 so that if degD E / K :2:
2g- 2 and if P E E( K) is a point of infinite order, then h( P) :2: C2 deg DE / K.
3.38. Let C be a non-singular hyperelliptic curve of genus two given by an equa-
tion
C : y2 = ax 5 + bx 4 + cx 3 + dx 3 + ex + f,
let i : C -> C be the involution i(x, y) = (x, -y), and let Po E C be the
point at infinity on C. (See [AEC, 11.2.5.1 and exercise 2.14] for basic facts
about hyperelliptic curves.)
(a) Prove that i(Po) = Po. Find all other points satisfying i(P) = P.
(b) Let P, Q E C be any two points. Prove that the divisors (P) + (i(P))
and (Q) + (i(Q)) are linearly equivalent.
(c) Let D E Div o (C). Prove that there exist points P, Q E C such that
D ~ (P) + (Q) - 2(Po ).
(d) Prove that the points P and Q in (c) are uniquely determined by D
unless P = i(Q).
(e) Prove that Pic( C) [2] is finite. More precisely, prove that it is isomor-
phic to (Zj2Z)4.
(f) Generalize (a)-(e) to the case of a hyperelliptic curve C: y2 = lex) of
genus g, where lex) is a polynomial of degree 2g + 1 with distinct roots.
3.39. Let K = k(C) be the function field of a curve C of genus g, let SeC be a
non-empty finite set of points of C, let Rs be the ring of S-integers of K,
and let E / K be an elliptic curve given by a Weierstrass equation
with A, BE Rs and 6. = 4A 3 + 27B2 E Rs.
Let P E E(K) be a point satisfying x(P), y(P) E Rs. Prove that
h(P) = h(x(P)) ::; 4max{2g - 2 + #S,O} + ~(h(6.) + h(A 3 ) + h(B2)).
(Hint. Use (12.3) and elementary properties of height functions.)

3.40. Let c --.> C be a minimal elliptic surface, let E / K be the associated elliptic
curve, and let E(K)o be the subgroup of E(K) described in §9.
(a) Suppose that c has non-constant j-invariant; that is, the function
jc: : C --.> pI defined in exercise 3.8 is non-constant. Prove that E(K)o has
no non-trivial torsion.
(b) For each int.eger m E {2, 3, 6}, give an example of a non-split elliptic
surface (necessarily with const.ant j-invariant) such that E(K)o contains a
point of exact order m. Prove that these are the only orders possible.
CHAPTER IV
The N eron Model
Let R be a discrete valuation ring with maximal ideal p and fraction field K,
and let ElK be an elliptic curve given by a Weierstrass equation
E : y2 + alxy + a3Y = X3 + a2x2 + a4X + a6,

say with coefficients aI, a2, a3, a4, a6 E R. This equation can be used to
define a closed subscheme W C Ph. An elementary property of closed
subschemes of projective space says that every point of E(K) extends to
give a point of W(R), that is, a section Spec(R) ....... W.
An important property of the elliptic curve E is that it has the struc-
ture of a group variety, which means that there is a group law given by
a morphism E x E ....... E. This group law will extend to a rational map
W XR W ....... W, but in general it will not be a morphism, so W will not be
a group scheme over R. However, if we discard all of the singular points on
the special fiber of W (i.e., the singular points on the reduction of E mod-
ulo p) and call the resulting scheme WO, then we will prove that the group
law on E does extend to a morphism WO x R WO ....... Wo. This makes WO
into a group scheme over R, but, unfortunately, we may have lost the point
extension property. In other words, not every point of E(K) will extend
to give a point in WO(R).
A Neron model for ElK is a scheme clR which has both of these
desirable properties. Thus every point in E(K) extends to a point in c(R),
and further the group law on E extends to a morphism c x R C ....... c which
makes c into a group scheme over R. It is by no means clear that such
a scheme exists. Our main goals in this chapter are to construct Neron
models, describe what they look like, and give some applications.
The material in this chapter is of a more technical nature than most of
the rest of this book. We will assume that the reader has some familiarity
with basic scheme theory, as described for example in Hartshorne [1, Ch. II]
or Eisenbud-Harris [1]. When we need more advanced material, we will give
at least a brief explanation together with a reference for further reading.
We begin in §1 with a brief discussion of group varieties. This ma-
terial is not strictly necessary for the remainder of the chapter but may
290 IV. The Neron Model
prove helpful for those readers who have not studied group schemes pre-
viously. Section 2 contains some basic material on abstract schemes and
schemes over a base S, including material on fiber products, special fibers,
regularity, properness, and smoothness. In §3 we define group schemes
and describe some of their elementary properties. Section 4 is devoted to
the theory of arithmetic surfaces. An arithmetic surface e over a discrete
valuation ring R is a "nice" scheme whose generic fiber is a non-singular
curve C / K. We give several examples and prove that the smooth part eO
of a regular proper arithmetic surface e has the point extension prop-
erty C(K) = eO(R). We also state the fundamental existence theorems
concerning minimal regular models of arithmetic surfaces.
In §5 we define Neron models and show that the smooth part WO
of a Weierstrass equation is a group scheme. In some cases, for example
when E/ K has good reduction, this will imply that WO is a Neron model
for E / K. The general construction of Neron models is given in §6, where
we prove that the smooth part c / R of a minimal proper regular model e/ R
for E / K is a Neron model. The proof is quite technical and may be omitted
on first reading. This is especially true for those readers who are mainly
interested in applications of the theory of Neron models. Frequently, it is
enough to know that a Neron model for E / K is a group scheme cover R
whose generic fiber is K and which has the property that c(R) = E(K).
We next take up the question of what the special fibers E. (mod p) and
e (mod p) look like. Section 7 contains a discussion of intersection theory
on general arithmetic surfaces, and then in §8 we apply this theory to give
the Kodaira-Neron classification of the special fibers of an elliptic fibration.
Section 9 contains a description and verification of an algorithm of Tate
e
which gives an efficient method of computing the special fiber (mod p)
from a given Weierstrass equation. In §10 we define the conductor of an
elliptic curve and give some of its properties. Finally, in §11, we state
and mostly verify an important formula of Ogg which gives a relation be-
tween the conductor, the minimal discriminant, and the special fiber of the
minimal proper regular model of an elliptic curve.
In order to simplify the discussion in this chapter, we will make the
following convention:
All Dedekind domains and all discrete
valuation rings have perfect residue fields.
Notice this includes Dedekind domains and discrete valuation rings whose
residue fields are finite, which is the case that will mainly interest us.
§1. Group Varieties
A group variety is an algebraic variety that is also a group. In slightly

fancier language, a group variety is a group in the category of algebraic
varieties. This means that the group law is given by algebraic functions.
Definition. A group variety (or algebraic group) is an algebraic variety G

and two morphisms
/-1: GxG ---+ G and i:G---+G
satisfying the following group axioms:
(i) There is a point 0 E G such that /-1(P, 0) = /-1(0, P) = P

for all PEG.
(ii) /-1(p, i(P)) = /-1(i(P), P) = 0 for all PEG.
(iii) /-1(p, /-1 ( Q, R)) = /-1 (/-1(P, Q), R) for all P, Q, REG.
G is called a commutative group variety if it further satisfies
(iv) /-1(P, Q) = /-1(Q, P) for all P, Q E G.
The group variety G is defined over K if G is defined over K, the mor-

phisms /-1 and i are defined over K, and 0 E G(K).
Example 1.1.1. An elliptic curve E / K is a commutative group variety
defined over K. This follows from [AEC, 111.2.2] and [AEC, III.3.6].
Example 1.1.2. The additive group G a and the multiplicative group Grn
are the commutative group varieties
and Grn ~ {x E AI : x -I- o}.

The group laws on G a and Grn are defined by the formulas
/-1 : G a x G a ---+ G a and /-1 : Grn x Grn ---+ Grn

(x,y) f--+x+y (x, y) f--+ xy.
The additive group is clearly an affine variety. The multiplicative group is
also an affine variety, since there is an isomorphism
Grn ---+ {(x, y) E A 2 : xy = I}, X f--+ (x, l/x).
Example 1.1.3. The general linear group GL n is defined by
Xll
GL n = {
M= (
:
Xnl
It is a group variety with group law given by matrix multiplication. Note

that the inverse map i(M) = M- 1 is a morphism on GL n , since the func-
tion 1/ det(M) is a regular function on GL n . Just as with the multiplica-
tive group, we observe that GL n is an affine variety, since it is the com-
plement of a hypersurface in A n 2 • (In general, the complement of f = 0
in Am is affine, since it is isomorphic to {(x, y) E A",+l : yf(x) = 1 }.) We
have GL 1 = G rn , and GL n is non-commutative for n ::::: 2.
Definition. A homomorphism of group varieties ¢ : G ~ H is a mor-

phism of varieties that is also a homomorphism of groups; that is, ¢ is a
morphism, ¢( Oc) = 0 H, and
¢(JLG(P, Q)) = JLH(¢(P), ¢(Q)) for all P, Q E G.
Example 1.2.1. An isogeny El ~ E2 of elliptic curves is a homomor-

phism of group varieties. This follows from [AEC, 1I1.4.8J, which says that
any morphism ¢ : El ~ E2 satisfying ¢(OEJ = OE2 is automatically a
homomorphism.
Example 1.2.2. The determinant map defines a homomorphism of group
varieties
det : GL n ---+ G m •
The kernel of the determinant map is another affine group variety called
the special linear group,
SLn = {M E GL n : det(M) = I}.
In general, an algebraic subgroup of GL n is called a linear group. It turns

out that every connected affine group variety is a linear group (Water-
house [1, §3.4J). For some other examples of linear groups, see exercises 4.1
and 4.2.
Proposition 1.3. Let G be a group variety defined over a field K. Then
the set of K-rational points G(K) is a subgroup ofG.
PROOF. The identity element 0 of G is in G(K) by definition. Further, the
morphisms JL : G x G ~ G and i : G ~ G are defined over K, so G(K) is
closed under the group operations. Hence G(K) is a subgroup of G. 0
Example 1.4.1. Let ElK be an elliptic curve. Then E(K) is the group
of K-rational points of E. If K is a number field, then E(K) is a finitely
generated group [AEC, VIII.6.7J.
Example 1.4.2. For any field K we have Ga(K) = K and Gm(K) = K*.
Similarly, GLn(K) is the group of n x n invertible matrices with coefficients
in K.
Proposition 1.5. Let G be a group variety.
(a) G is a non-singular variety.
(b) Every connected component of G is irreducible.
(c) The connected component of G which contains the identity element is
a normal subgroup of G of finite index.
Definition. Let G be a group variety. The connnected component of G
containing the identity element is denoted by GO and is called the identity
component ofG. The quotient group GIGO called the group of components
ofG.
PROOF (of Proposition 1.5). (a) There is a Zariski open subset U c G

which is non-singular (Hartshorne [1, 1.5.3]). For any PEG, let Tp : G ---->
G be the translation-by-P map, Tp(Q) = f-L(P, Q). Note that Tp is an
isomorphism from G to itself. Now G is covered by the non-singular open
sets Tp(U), PEG, so every point of G is non-singular.
(b) Suppose that G has a connected component that consists of more
than one irreducible component. Then that connected component would
contain distinct irreducible components that have a point in common, and
the common point would be singular. This contradicts (a). Hence every
connected component of G is irreducible.
(c) A variety has only finitely many connected components, since it actu-
ally has only finitely many irreducible components (Hartshorne [1, 1.1.6]).
We label the connected components of G as GO, G 1 , ... , Gn, where GO is the
connected component of G containing the identity element. Let P E GO.
The translation-by-P map Tp permutes the connected components of G,
so Tp(G O) = Gj for some j. But
P = Tp(O) E Tp(G °) = Gl,.
so the connected components GO and Gj have the common point P. Hence

GO = Gj. This means that f-L(P,Q) = Tp(Q) E GO for all P,Q E GO. Sim-
ilarly, 0 E GO n i(G O), so i(G O) = GO. This proves that GO is a subgroup
ofG.
Next, fix a point Q E G and consider the conjugation-by-Q map
¢: G ----t G, ¢(P) = f-L(i(Q), f-L(P, Q)).

¢ is an automorphism of G, so it permutes the components of G. Fur-
ther, ¢(O) = 0, so as above we conclude that ¢(GO) = GO. Therefore GO
is a normal subgroup of G.
Finally, for each 0 ::::: j ::::: n we fix a point Pj E Gj. Then the maps
¢j : G ----t G,
permute the components of G and satisfy ¢j(Pj ) = 0, from which we

conclude that ¢j (Gj) = GO. Hence Po, ... ,Pn includes a complete set of
coset representatives for GIGO, so GO has finite index in G. 0
We have seen above (1.1.1, 1.1.2) that the additive group, the multi-
plicative group, and elliptic curves are group varieties. We will now prove
that these are the only connected group varieties of dimension one.
Theorem 1.6. Let G be a connected group variety of dimension one de-
fined over an algebraically closed field. Then either G ~ eGa, G ~ eG m, or G
is an elliptic curve. (For non-algebraically closed fields, see exercise 4.13')
Before beginning the proof, we prove a lemma that gives conditions
under which a curve has only finitely many automorphisms.
Lemma 1.7. Let C be a non-singular projective curve of genus g, let S c

C be a finite set of points, and suppose that S satisfies one of the following
conditions:
(i) #S ~ 3 if 9 = O. (ii) #S ~ 1 if 9 = 1. (iii) S is arbitrary if 9 ~ 2.
Then
Aut(C; S) def {4> E Aut(C) 4>(S) c S}
is a finite set.
PROOF. Suppose first that 9 = 0, so we can take C = pl. Fix three

distinct points PI, P2 , P3 E S. Every automorphism of pI is given by a
linear fractional transformation (Hartshorne [1, II.7.1.1])
4> ([x, yJ) = [ax + by, cx + dyj.

An automorphism 4> will thus be determined by the images of PI, P2 , P3 ,
which proves that the map (of sets)
is injective. But S is finite by assumption, so Aut(C; S) is finite.

Next suppose that 9 = 1. We make C into an elliptic curve by taking
the origin 0 to be a point in S. Then every isomorphism C --+ C is a
translation followed by an isomorphism fixing 0 [AEC, 111.4.7]. But there
are only finitely many isomorphisms C --+ C that fix 0 [AEC, lIL10.1], so
the map (of sets)
Aut( C; S) --+ C,
is finite-to-one. Since 4>(0) E Sand S is finite, this proves that Aut( C; S)

is finite.
Finally, we recall a theorem of Hurwitz which says that if a curve
has genus 9 ~ 2, then it has at most 84(g - 1) automorphisms. (See
Hartshorne [1, exercises IV.2.5, IV.5.2, V.1.l1j.) This completes the proof
of Lemma 1.7. D
PROOF (of Theorem 1.6). We know that the variety G is non-singular, irre-
ducible (1.5), and has dimension one, so we can embed it as a Zariski open
subset of a non-singular projective curve, say G C C (Hartshorne [1, I §6]).
Let S = C '-. G be the complement of Gin C.
For every point PEG, the translation-by-P map Tp : G --+ G is an
automorphism of G as a variety. Then Tp induces a rational map Tp : C --+
C which extends to an isomorphism since C is non-singular [AEC, II.2.1j.
Clearly, we have Tp(S) C S, since TP(G) = G. In this way, we obtain an
inclusion G '----+ Aut(C; 5) given by P 1--7 Tp, where Aut(C; 5) is as in (1.7).

But G is a variety of dimension one, so it has infinitely many points. It
follows from (1.7) that C has genus less than 2, that 5 = 0 if C has genus 1,
and #5 ::; 2 if C has genus O. There are thus four cases to consider.
Suppose first that C has genus 0 and 5 = 0. Then G = C = pI, and
the group law on G is a morphism
Such a map has the form p,(x,y) = [f(x,y),g(x,y)], where f and g are
bihomogeneous polynomials. The fact that p, is a morphism means that f
and g can have no common roots in pI x pI, which implies that p, must
look like either
p,(x,y) = [f(x),g(x)] or p,(x,y) = [f(y),g(y)].

But then either p,( 0, P) or p,(P, 0) is constant, so p, cannot define a group
law. This proves that it is not possible to have G = pl.
Next suppose that C has genus 0 and that #5 = 1. Then we can
identify C with pI in such a way that the point in 5 is the point at infinity
and the identity element of G is the point O. In other words, we have C =
pI, G = A I, and 0 E A I is the identity element of G. The group law on G
is a morphism
p,; Al X Al ----? AI,
so p, is a polynomial map, p,(x, y) E K[x, yj. For every fixed value of x,

we know that the map y 1--7 p,( x, y) is an automorphism of A!, so p,( x, y)
must be linear in y. Similarly for x 1--7 p,(x, y), so p,(x, y) is also linear in x.
Further, p,(x,O) = x and p,(0, y) = y, so we conclude that p, has the form
p,(x, y) = x + y + cxy for some c E K.
Finally, we observe that if c f. 0, then c- I would not have an inverse,

since p,( -c- I , y) = _c- I is constant. Hence c = 0 and p,(x, y) = x + y,
which proves that G ~ Ga.
The next case to consider is a curve C of genus 0 and #5 = 2. This
time we identify C with pI so that the two points in 5 are 0 and 00 and
so that the identity element of G is the point 1. Then the group law
on G = Al " {O} is a morphism
so p, is a Laurent polynomial, p,(x, y) E K[x, X-I, y, y-Ij. As above, the

map y 1--7 p,(x, y) is an automorphism for every fixed x, which means it
must have the form
p,(x, y) = a(x) + b(x)y or p,(x, y) = a(x) + b(X)y-I.

We further know that p,(1, y) = y, which rules out the second possibility
and tells us that a(l) = 0 and b(l) = 1. In particular, b(x) i= o.
If a(x) i= 0, then we can find an 0: E K* so that a(o:) i= 0 and {3(0:) i= O.
Then
M(o:,-~~:D =0,
which contradicts the fact that p,( G x G) c G. Therefore a( x) = O. We
have now shown that p,(x, y) = b(x)y. Reversing the roles of x and y
and using the fact that p,(1, 1) = 1, we conclude that p,(x, y) = xy, which
completes the proof that G ~ G m .
It remains to consider the case that C = G is a curve of genus 1.
The group variety G has an identity element 0, and we use this point to
give (C,O) the structure of an elliptic curve. It remains to show that the
identification G and C as curves is also an isomorphism of groups. In other
words, we need to prove that
P,e(P, Q) = P +Q and ie(P) = -P,

where P,e : G x G ----+ G is the given group law on G and + is the group law
on the elliptic curve (C,O). Note that we do not assume, a priori, that G
is commutative.
Consider the map
¢ :C x C -----> C, ¢(P, Q) = Me(P, Q) - P - Q.
°
The point
that ¢(P,O) = °and ¢( 0, Q) = °
E C is the identity element for both group laws, so we find
for all P, Q E C. It follows from
an elementary rigidity lemma (1.8) which we will prove below that ¢ is
constant. Hence
¢(P, Q) = ¢(O, 0) = 0, and so P,e(P, Q) = P + Q for all P, Q E C.
Finally, we observe that
P+ie(P) =p,e(P,ie(P)) =0,
which proves that ie(P) = -P. This completes the proof of (1.6), subject
to our proving the following lemma. 0
Lemma 1.8. (Rigidity Lemma) Let C 1 , C 2 , C 3 be irreducible projective

curves, and let
¢ : C 1 x C2 -----> C3
be a morphism. Suppose that there are points PI E C 1 and P2 E C 2 with
the property that each of ¢( PI x C 2 ) and ¢( C 1 X P 2 ) consists of a single
point. Then ¢ is a constant map.
PROOF. We are given that ¢(P1 x C2 ) consists of a single point, say
Choose a point R' E C3 , R' R, and consider the set

=1=
U 1 = {Q E C 1 : R' tJ. ¢(Q x C 2 )}.

Notice that the complement of U 1 in C 1 is the set
C 1 " U1 = proh (¢-l(R')),
where proh : C 1 x C 2 ---* C I is projection onto the first factor. The pro-
jection map sends closed sets to closed sets. This follows from the fact
that C 2 is projective, hence proper (Hartshorne [1, IIA.9]) , and the defi-
nition of properness implies that any projection V x C 2 ---* V is a closed
morphism.
Now the set ¢-l(R') is closed, so the same is true of projl (¢-l(R')),
which shows that U1 is an open subset of C I . Further, it is clear that
that PI E U I , so U I is non-empty. For any Q E U I , we consider the
morphism
C 2 - + C3 , S ~ ¢(Q, S).
The fact that Q E UI tells us that R' is not in the image of this map.
In other words, this map is not surjective, so it follows from [AEC, 11.2.3]
that it is constant. In other words, if Q E U I , then ¢(Q, S) is independent
of S E C 2 • Equivalently, ¢(Q x C2 ) consists of a single point.
We now repeat the above argument using the fact that ¢(C1 x P2 )
consists of one point. Doing so yields a non-empty open set U2 C C 2 with
the property that for all S E U2, the set ¢(CI x S) consists of a single point.
Combining these two facts, we find that ¢(UI x U 2 ) consists of one
point. But U I x U 2 is Zariski dense in C 1 x C 2 , and a morphism is deter-
mined by its values on a Zariski dense set. Therefore ¢ is constant.
o
§2. Schemes and S-Schemes
In this section we are going to review some basic notions about schemes,
especially schemes over a fixed base scheme. We assume that the reader
has some familiarity with this material, as covered for example in Hart-
shorne [1, II §§2,3] or Eisenbud-Harris [1].
Definition. Let S be a fixed scheme. An S -scheme is a scheme X equipped
with a morphism X ---* S. A morphism of S-schemes (or S-morphism) is a
morphism X ---* Y so that the diagram
X Y
S
'" /
is commutative. If S = Spec(R), we will often refer to R-schemes and R-
Tnorphisms instead of Spec(R)-schemes and Spec(R)-morphisms.
Intuitively, an 5-scheme X --+ 5 can be viewed as an algebraic family

of schemes, namely the family of fibers Xs parametrized by the points
s E 5. We have seen an example of this in Chapter III, where the elliptic
surface E --+ C is a C-scheme whose fibers Et form a family of elliptic
curves.
Two other important examples are provided by affine and projective
space over a ring R. These are defined to be
and 1P'7{ = Proj R[.TO . ... ,:1'n].
See Hartshorne [1, II.2.5.1] for details.

In Chapter III we studied the group of sections C --+ E to the elliptic
surfacc. Similarly. we can look at the set of sections 5 --+ X of an 5-scheme.
These are precisely the 5-morphisms from 5 to X. More generally, for
any 5-scheme T, we can consider the set of 5-morphisms from T to X.
Definition. Let X and T be 5-schemes. The set of T -valued points 01 X
is the set
X(T) = Homs(T. Xl = {5-morphisms T --+ X}.
IfT = 5, we will sometimes call X(5) the set of sections of the 5-8cherne X.
Similarly, if 5 = T = Spec(R), we will refer to the R-valued points 01 X
and write X(R).
Example 2.1.1. Let K be a field, let 5 = Spec(K), and let XI K be an
affine scheme, say given by equat.ions
h = h = ... = fT' = 0 wit.h fl,"" If' E K[XI,"" Xn].
Then
X(5) = {K-morphisms Spec(K) --+ X}

. K[XI,."'X n]
~ { K-algebra homomorphIsms (h, ... , fr.)
={PEK n : h(P)=···=f,.(P)=O}.
Thus X(5) agrees with our intuition of what X(K) should be. l\Iore gen-
erally, if R is any ring and X c AR is an affine scheme given by equations
h = ... = IT = 0 with fi E R[XI"'" xn], then X(R) is naturally identified
with the set of n-tuples (Xl, ... ,X n ) E R n satisfying the equations.
Remark 2.1.2. Each 5-scherne X defines a functor Fx on the category
of 5-schemes by assigning to an 5-scheme T the set of T-valued points
of X. Thus
Fx : (5-schernes) ---> (Sets), T f---> X(T).

If cp : T ~ T' is an S-morphism of S-schemes, the associated map Fx(cp)

is given by composition,
Fx(cp) : Fx(T') ---> Fx(T), a 1--+ a 0 cp.
Notice that Fx is a contravariant (i.e., arrow reversing) functor. It is

a basic categorical fact (called Yoneda's lemma, see Eisenbud-Harris [1,
Lemma IV.ID that the functor Fx determines the scheme X. Similarly,
morphisms of functors Fx ~ Fy correspond bijectively with S-morphisms
X ~ Y. We will not make use of this functorial approach, but the reader
should be aware that it is a convenient language which is in common usage.
Next we describe One of the most important constructions of algebraic
geometry.
Definition. Let X and Y be S-schemes. The fiber product of X and Y
over S is an S-scheme, denoted X x s Y, together with projection mor-
phisms
Pi : X x S Y ---> X and P2 : X Xs Y ---> Y
over S with the following universal property:
Let Z be an S-scheme, and let f : Z ~ X and 9 : Z ~ Y
be S-morphisms. Then there exists a unique S-mor-
phism Z ~ X x s Y so that the following diagram com-
mutes:
Z
1
f
./
X X xsY ~ Y
The fiber product exists and is unique up to unique isomorphism; see Hart-
shorne [1, II.3.3] or Eisenbud-Harris [1, LC.i, IV.BD. If S = SpeeR, we will
often write X XR Y.
The fiber product is the smallest scheme that fits into the commutative
diagram
PI
X xsY ---> X
1
Y ---> S.
In some sense, X x s Y should "look like" the set of ordered pairs (x, y)
having the property that x and y have the same image in S. This is
literally true in the category of sets, but care must be taken when applying
this intuition in the category of schemes. In fact, X x s Y will generally
be quite large, even when S consists of a single point; see for example
Hartshorne [1, II, exercise 3.1].
Example 2.2.1. Let s E 5, and let Y = {s} ~ 5 be the subscheme of 5

consisting of the point s. l\,Jore precisely, if we write k(s) = Os,s/Ms.s for
the residue field of the local ring at s, then Y is the scheme Spec k(s). The
fiber of X over s is defined to be the scheme
X, clef
= }
Xxs{s.
It is a scheme over k(s). In this case the underlying topological space
of X x s {s} actually equals the set of points x E X such that the image
of x in 5 is s, see Hartshorne [1, II exercise 3.10]. Thus at the level of
points, this definition of the fiber X, agrees with our intuition of what the
fiber should be.
Example 2.2.2. Let R be a ring, let p be a maximal ideal of R, and let X
be an R-scheme. Then the fiber
Xp=XXRP
is the reduction of X modulo p. It is a scheme over the residue field Rip.
This agrees with our intuition, since an (affine) scheme X over R is defined
by a system of polynomial equations with coefficients in R, and Xp is the
scheme defined by reducing the coefficients of the polynomials modulo p.
Example 2.2.3. Let R be an integral domain, and let TJ = (0) E Spec R
be the generic point of Spec R. If X is an R-scheme, then the fiber
XY] = X XR TJ
is called the generic fiber- of X. It is a scheme over the fraction field K
of R. In particular, if R is a discrete valuation ring with maximal ideal p,
then X has two fibers, its gener-ic fiber- XY]I K and its special (or closed)
fiber Xplk, where k = Rip is the residue field of R.
°
For example, suppose that X C lP'7t is given by a single homogeneous
equation f(x, y, z) = with coefficients in R. Then the generic fiber XI) C
lP'~ is the variety defined by the same equation f(x, y, z) = 0, and the
special fiber Xp C lP'~ is the variety defined by the equation j(x, y, z) = 0,
where j is obtained by reducing the coefficients of f modulo p.
Example 2.3. Let 7r : X -+ S be an S-scheme. In the definition of the
fiber product, if we take Z = X and f and 9 to be the identity map X -+ X,
then we obtain the diagonal morphism
Ox : X ---+ X Xs X:
that is, Ox is the unique map to the fiber product with the property that P10
Ox and P2 0 Ox are each the identity map on X.
More generally, let ¢ : X -+ Y be an 5-morphism. Then the graph
of ¢ is the unique morphism
fir/>: X ---+ X X8 Y
such that Pl 0 o¢ is the identity map on X and P2 0 o¢ = (p. Notice the
diagonal morphism is the graph of the identity map X -+ X.
Example 2.4. Let '!r : £ ~ C be an elliptic surface, say defined over

an algebraically closed field k. Then the fiber product £ Xc £, or more
precisely, the set of k-valued points on the fiber product, is the set
(£ Xc £)(k) = {(P,Q) : P,Q E £(k) and '!rep) = '!r(Q)}.
Thus (e Xc e)(k) consists of pairs of points which lie on the same fiber.
In particular, if P and Q lie on a non-singular fiber £t, then we can add
them using the group law on et. In this way (most of) the fiber prod-
uct (e Xc e)(k) becomes a group.
Example 2.5. Recall that every scheme S admits a unique morphism S ~
SpecZ (Hartshorne [1, exercise 11.2.4]). Affine and projective space over S
are defined to be
and
Projection onto the second factor makes A~ and IFs into S-schemes. Notice
that if S = Spec R is an affine scheme, then A~ ~ A'R and IFs ~ IPR, so
these definitions are compatible with the definitions of affine and projective
space over a ring.
We are now faced with the task of discussing three important proper-
ties of schemes and morphisms, namely regularity, properness, and smooth-
ness. The definition of each of these properties is somewhat technical, and
in truth we will make very little use of the formal definitions of properness
and smoothness in subsequent sections. On the other hand, the intuitions
underlying all of these properties are quite easy to understand, especially
if one works in a "nice" setting. So we are going to begin with an informal
discussion, including examples and basic material which we will give with-
out proof. This discussion should suffice for reading the remainder of this
chapter, except possibly for parts of §6. Then, at the end of this section,
we will give precise definitions and provide references for further reading.
Intuitive "Definitions." A scheme X is regular if it is non-singular, by
which we mean that every point of X has a tangent space of the correct
dimension.
A morphism of schemes X ~ S is proper if all of its fibers are complete
and separated. (These are algebraic analogues of compact and Hausdorff.)
Essentially, this means that the fibers of X ~ S are not missing any points
and do not have too many points. We also say that X is a proper S -scheme.
A morphism of schemes X ~ S is smooth if all of its fibers are non-
singular, or, to put it another way, if X is a family of regular schemes. We
also say that X is a smooth S -scheme.
In order to define regularity, we recall that the Kroll dimension of a
ring A is the largest integer d such that there is a chain of distinct prime
ideals of A,
A local ring A with maximal ideal 9)1 is called regular if the dimension
of9)1/9)12 as an A/9)1-vector space is equal to the Krull dimension of A. (See
Matsumura [1, Ch. 7] or Atiyah-MacDonald [1, Ch. 11] for basic material on
regular local rings.) Intuitively, 9)1/9)12 is the cotangent space of Spec(A)
at the point 9)1, and the regularity of A is an assertion that 9)1 is a non-
singular point of Spec(A). For arbitrary schemes one defines dimension
and regularity in terms of the local rings as follows:
Definition. The dimension of a point P of a scheme X is the Krull di-
mension of the local ring () p at P. If every closed point of X has the same
dimension, we call this the dimension of X.
Definition. A point P of a scheme X is said to be regular (or non-
singular) if the local ring (') p is a regular local ring. The scheme X is
regular (or non-singular) if every point of X is regular. In fact, it suffices
to check that every closed point is regular; see exercise 4.5.
Example 2.6.1. Let R be a Dedekind domain. Then Spec(R) is a regular
scheme of dimension one. To see this, note that in a Dedekind domain,
every non-zero prime ideal is maximal by definition. Hence the longest
chain of prime ideals is (0) C p, so R has dimension one. Further, each
localization Rp is a discrete valuation ring, so its maximal ideal Mp is
principal. It follows that Mp/M'i has dimension one as an Rp/Mp-vector
space, so Rp is regular.
Example 2.6.2. If R is a regular local ring, or more generally if Spec(R)
is regular, then both A'R and IP''R are regular schemes.
Example 2.6.3. Let R be a discrete valuation ring, let 7[ be a uniformizer
for R, and assume that 2,3 E R*. Let a E R, and define a scheme X c IP'h
by the equation
X : y2 z = x 3 + az 3 .
°
Then X is a regular scheme if and only if a ¢ (mod 7[2). To see this, one
first checks that the only possible singular point is the point "( = [0,0,1]
on the special fiber Xp and that this can only occur if a == 0 (mod 7[).
Dehomogenizing the equation by setting z = 1, we find that the maximal
ideal M"( of the local ring (')"( is generated by x, y, and 7[, and that these
quantities are related by the equation
If a ¢ °(mod 7[2), then a is itself a uniformizer for R. Hence
°
so x and y generate M"(/M~, which shows that ()"( is regular. Conversely,
if a == (mod 7[2), then M"( / M~ cannot be generated by fewer than three
elements, so (')"( is not regular.
Next we look at proper morphisms, which, recall, are supposed to be

morphisms X -+ S whose fibers are separated and complete. For example,
suppose that we are given a "curve" C, a point 'Y E C, and a commutative
diagram of morphisms F
---> x
1 1
C ~ S.
If X -+ S is a proper morphism, then the fiber of X over f ('Y) is separated
and complete, so there should be a unique way to extend F to all of C. This
statement is essentially the following valuative criterion for properness.
Theorem 2.7. (Valuative Criterion of Properness) Let ¢ : X -+ S be
a morphism of finite type of Noetherian schemes. The map ¢ is proper if
and only if for every (discrete) valuation ring R with fraction field K and
every commutative square of morphisms
Spec(K) ---> X
1
Spec(R) ---> S,
there is a unique morphism Spec(R) -+ X fitting into the diagram. (In
other words, there is a unique morphism Spec(R) -+ X so that the compo-
sition Spec(R) -+ X -+ S agrees with the bottom line of the square.)
PROOF. Hartshorne [I, II.4.7]. See also Hartshorne [I, exercise H.4.11] for
the assertion that it suffices to consider only discrete valuation rings. D
In order to better understand what the valuative criterion is saying, we

note that if R is a discrete valuation ring with fraction field K, then Spec( R)
is a regular one-dimensional scheme (2.6.1), and Spec(K) is Spec(R) with
its closed point removed. Thus Spec(K) looks like a curve with one point
removed.
An important collection of proper S-schemes is the set of projective
schemes over S, as described in the following result.
Theorem 2.8. Let S be a Noetherian scheme, and let X c lP's be a
closed subscheme of projective space over S. Then X is proper over S. In
particular, lP's itself is proper over S.
PROOF. See Hartshorne [1, II.4.9]. D
We continue our informal discussion by looking at smooth morphisms.

For our purposes, the most important examples of smooth morphisms will
be schemes which are smooth over a discrete valuation ring or Dedekind
domain R. In this situation, the condition that X be smooth over R is
essentially equivalent to the assertion that all of its fibers are non-singular
and have the same dimension.
Proposition 2.9. Let R be a discrete valuation ring with fraction field K,

residue field k, and maximal ideal p. Let X be an integral (i.e., reduced
and irreducible) R-scheme of finite type over R whose generic fiber XTJ/ K is
non-empty. Then X is a smooth R-scheme if and only if XTJ(K) and Xp(k)
contain no singular points.
PROOF. The scheme X is irreducible, so it has a unique generic point. Our

assumption that the generic fiber XTJ is non-empty shows that the generic
point of X maps to the generic point of Spec R, so X is flat over Spec R from
Hartshorne [1, III.9.7j. If R contains its residue field k (the so-called func-
tion field case), the desired result then follows from Hartshorne [1, III.lO.2j.
The general case is Milne [4, I Prop. 3.24]' see also Bosch-Liitkebohmert-
Raynaud [1, §2.4, Prop. 8j.
o
There are many theorems in algebraic geometry which say that some
property of morphisms, such as properness, smoothness, separability, finite-
ness, etc., is preserved under composition, base extension, and products.
The only result of this sort that we will need is the following assertion that
the composition of smooth morphisms is again smooth.
Proposition 2.10. If ¢ : X ---> Y and 'ljJ : Y ---> Z are smooth morphisms,
then the composition 'ljJ 0 ¢ : X ---> Z is a smooth morphism.
PROOF. See Hartshorne [1, III.lO.1c] or Altman-Kleiman [1, VII.1.7iij.

o
We now look at some examples of regular schemes and proper and
smooth morphisms.
Example 2.11.1. Let R be a discrete valuation ring with uniformizer 1r.
We assume that 2 E R*. Let X C IP'~ be the scheme given by the equation
X : x2 + 7fy2 = Z2.
Then X is proper over R from (2.8), since it is a closed subscheme of IP'~. It
is also easy to check that the scheme X is irreducible and regular. However,
the special fiber of X is given by the equation x 2 = z2, so the special fiber
is reducible and singular. Hence X is not smooth over R.
Example 2.11.2. We continue analyzing example (2.6.3), so R is a dis-
crete valuation ring with uniformizer 7f and 2,3 E R*, and X C IP'~ is the
scheme defined by the equation
for some a E R.
We also let K be the fraction field of R, k the residue field of R, and p the
maximal ideal of R. Notice that X is proper over R by (2.8), since it is a
closed subscheme of IP'~.
Suppose first that a E R*. Then the special fiber Xp/ k is a non-
singular curve, so X is smooth over R from (2.9).
Next, suppose that a == 0 (mod p). Then the special fiber Xp/k is
given by the equation y2 z = x 3 , so the special fiber is singular and X
is not smooth over R. Let 'Y E Xp C X be the singular point on the
special fiber, and let XO = X " 'Y be the scheme obtained by removing 'Y
from X. This makes the special fiber xg / k non-singular, so X D is smooth
over R. However, removing the point 'Y has destroyed the completeness of
the special fiber, so XO is not proper over R.
Finally, we observe that if a is a uniformizer for R, then (2.6.3) says
that X is regular. This is true despite the fact that its special fiber Xp
is singular and so X is not smooth over R. We will prove later (4.4) that
since X is regular, every R-valued point of X lies in the smooth part XO.
In other words, the natural inclusion XO(R) C X(R) is an equality, so in
this situation XO retains a sort of properness property over R.
This last example (2.11.2) illustrates an important general phenome-
non. Let X be a (nice) scheme which is proper over a discrete valuation
ring R and which has a smooth generic fiber. Then X need not be smooth
over R, since its special fiber Xp may have singularities. We can create
a smooth R-scheme XO by removing from X the singular points on its
special fiber, but then XO will not be proper over R. Thus the attributes
of properness and smoothness are somewhat antithetical to one another.
However, if the original scheme X is regular, then it turns out that
every R-valued point in X(R) actually lies in XO(R). So for regular
schemes, XO still behaves to some extent as if it were proper over R. We
will prove this later (4.4) when X has relative dimension one over R. The
general case we leave as an exercise.
We are now ready to define properness and smoothness, but we want to
stress that the most important thing is for the reader to understand the un-
derlying intuitions and the examples described above. For further reading
on this material, see Hartshorne [1, III §§9,10], Altman-Kleiman [1, V,VI],
and Bosch-Liitkebohmert-Raynaud [1, 2.1-2.4].
Definition. Let ¢ : X --- S be a morphism of finite type. The map ¢
is separated if the diagonal morphism Ox : X --- X Xs X (2.3) is a closed
immersion. The map ¢ is universally closed if for any base extension S' ---
S, the map X Xs S' --- S' sends closed set to closed sets. The map ¢ is
proper if it is separated and universally closed. We also say that X is
proper over S, or that X is a proper S -scheme.
Definition. Let ¢ : X --- S be a morphism of finite type, let x EX, and
let s = ¢(x) E S. The map ¢ is smooth (of relative dimension r) at a
point x E X if there are affine open neighborhoods
s E Spec ReS and x E Spec A C X
with
306 IV. The N eron Model
A = R[t 1 , ... , tn+r]/(h, ... , In) for some h,···, In E R[tl, ... , t n+r ]
so that the n x n minors of the Jacobian matrix (aId atj) generate the unit
ideal in A.
We say that <p is smooth (or that X is smooth over S) if <p is smooth
at all points of X. A morphism that is smooth of relative dimension zero
is called an elale morphism.
Remark 2.12. The Jacobian condition in the definition of smoothness is
similar to the criterion we used to define non-singular points on varieties
in [AEC, I §1]. In particular, it is clear that if X --> SpecR is a smooth
morphism and p E Spec R is a maximal ideal, then the fiber Xp is a non-
singular variety over the residue field Rip, which is a special case of (2.9).
There are many other ways to define smoothness. One of the most useful
is in terms of the sheaf of relative differentials of XIS, see for example
Hartshorne [1, II §8, III §1O], Altman-Kleiman [1, VI,VII], Milne [4, 1§3]
or Bosch-Ltitkebohmert-Raynaud [1, §2.1,2.2]. For a fancy functorial defi-
nition, see Milne [4, 1.3.22].
§3. Group Schemes
A group scheme over S is an S-scheme G whose fibers form an algebraic

family of groups, similar to the example described in (2.4). This means we
should be able to multiply two points provided they lie on the same fiber,
and the multiplications should fit together to give a group law on the fiber
product G Xs G. More formally, a group scheme over S is a group in the
category of S-schemes. This leads to the following definition. Note that we
must be careful to define everything in terms of maps, rather than in terms
of points. (An alternative approach is to define a group scheme in terms of
its associated functor of points; see (2.1.2) or Eisenbud-Harris [1, IV.A.v].)
Definition. Let S be a scheme. A group scheme over S is an S-scheme
7r : G --> Sand S-morphisms
ao : S ---+ G, i :G ---+ G, M : G Xs G ---+ G,
such that the following diagrams commute:

(i) (identity element)
G xsG G xsG
0"0 X 1 1x 0"0
/ /
S Xs G G G Xs S G
(ii) (inverse)
lxi ixl
G xsG GxsG G xsG G xsG
loa 1p
G ~
7r
S
0'0
~ G G ~
7r
s ~ G
(Here G ~ G xs G is the diagonal map (2.3).)

(iii) (associativity)
pxl
GxsGxsG ~ GxsG
11xP
GxsG G
Example 3.1.1. Let G be a group variety defined over a field K as dis-

cussed in §l. Then G is a group scheme over the one point scheme S =
Spec(K). This is clear from the definitions. Note that the identity mor-
phism 0'0 : S --; G sends the one point in S to the identity element of G.
Example 3.1.2. The additive group scheme G a over Z is the scheme G a =
SpecZ[T]. The group law on G a is given by
~ SpecZ[T],
where the morphism Spec Z[Tl' T 2 ] --; Spec Z[T] is induced by the ring
homomorphism
For any ring R, we have Ga(R) = R with group law given by addition
on R. The additive group scheme Gals over an arbitrary scheme S is the
group scheme G a Xz S obtained by base extension. In particular, G alR =
SpecR[T].
Example 3.1.3. The multiplicative group scheme G m over Z is the scheme

G m = SpecZ[T, T-1j. The group law on G m is given by
--> SpecZ[T, T- 1 ],
where the morphism Spec Z[Tb T1- 1, T 2 , T2- 1j --+ Spec Z[T, T-1j is induced
by the ring homomorphism
For any ring R, we have Gm(R) = R* with group law given by multi-
plication on R*. The multiplicative group scheme G m / s over an arbitrary
scheme S is the group scheme G m Xz S obtained by base extension. In
particular, G mlR = Spec R[T, T-1j.
Example 3.1.4. Let R be a discrete valuation ring with maximal ideal p
and fraction field K, and let ElK be an elliptic curve with good reduction
at p. Fix a minimal Weierstrass equation for E,
The coefficients of this equation are in R, so we can use the equation to

define an R-scheme c c IP'~. (Of course, we need to homogenize the equa-
tion first.) The fact that E has good reduction implies that the scheme c
is smooth over R, since good reduction is equivalent to the fact that the
special fiber cp of c --+ Spec R is a smooth elliptic curve over the residue
field Rip.
The addition law on E is given by rational functions with coefficients
in R, so it induces a rational map
We know from [AEC, III.3.6j that the addition law E x E --+ E on the
generic fiber is a morphism. We will later give two proofs that M itself is
an R-morphism. The first proof (5.3) uses explicit equations and is similar
to the argument in [AEC, III.3.6, III.3.6.1j. The second proof (6.1) uses
fancier machinery to prove a much stronger result.
The next proposition shows that the set of T-valued points of a group
scheme form a group.
Proposition 3.2. Let G be a group scheme over S, let T be an arbi-

trary S-scheme, and let G(T) be the set of T-valued points of G, which
recall is the set of S-morphisms T ~ G. For any two elements ¢>, 'l/J E G(T),
define a new element ¢> * 'l/J E G(T) by the commutativity of the diagram
TXsT GxsG
T G,
where bT is the diagonal map (2.3). In other words,
¢> * 'l/J = J-lO (¢> x 'l/J) 0 bT E G(T).
This operation gives G(T) the structure of a group. The identity element
is 0"0 o 7fT, where 7fT : T ~ S is the map making T into an S-scheme. The
inverse of ¢> is i 0 ¢>.
More precisely, the association T ~ G(T) is a contravariant functor
from the category of S-schemes to the category of groups.
PROOF. All of this follows from the definitions and elementary diagram
chases. For example, to verify that 0"0 o 7fT is the identity element of G(T),
we observe that the following diagram is commutative:
G xsT G XsS
11 XC70
T xsT GxsG
T G
But the definition of 0"0 tells us that the map J-lO (1 x 0"0) : G Xs S ~ G
down the right-hand side of this diagram is projection onto the first factor.
Hence tracing around the boundary of the diagram yields
We will leave it to the reader to perform the similar computations

needed to check that i 0 ¢> is the inverse of ¢> and that the associative law
holds, which completes the proof that G(T) is a group.
Finally, the functoriality statement means that if j : T' ~ T is an S-
morphism of S-schemes, then the map
G(T) ---> G(T'), ¢> f---> ¢> 0 j,

is a homomorphism of groups. It is clear that the identity element is

mapped to the identity element, so we must verify that
(4; * 'ljJ) 0 f = (4; 0 f) * ('ljJ 0 f) for all 4;, 'ljJ E G(T).
The definition of 4; * 'ljJ says that the right-hand square of the following
diagram is commutative, and the left-hand square is clearly commutative:
T' x s T' ~ T xs T ~ Gx G
T' T G
The map (4; x 'ljJ) 0 (f x f) along the top row of this diagram is equal to
(4; 0 f) x ('ljJ 0 f), so by definition the map along the bottom row equals
(4; 0 f) * ('ljJ 0 f). This is the desired result, which completes the proof of
Proposition 3.2. 0
Remark 3.3. In our study of elliptic curves and group varieties, the
translation-by-P maps provided an important tool. A group scheme GIS
is not a group, so we cannot translate G by a point of the scheme G. In-
stead, G is a family of groups parametrized by the points of S. So in order
to translate G, we need to start with a family of points on G parametrized
by S. Then we can translate each group in the family by the appropriate
point. We formalize this idea in the following manner.
Let GIS be a group scheme, and let 0" E G(S) be an S-valued point
of G. Then the (right) translation-by-O" morphism is the S-morphism
Tu : G -> G defined by the composition
Tu : G C:;! G Xs S GxsG G.
To understand Tu further, we note that the S-valued point 0" is a map
0" : S -> G. In particular, for each point 8 E S, we get a point 0"(8) on
the fiber G., where G s is a group variety over the residue field at s. The
restriction of T u to the fiber G s is precisely translation by the point 0"( s) E
G s . Thus Tu can be viewed as a family of translations of the fibers of G.
Remark 3.4. Another important tool in our study of elliptic curves was
the collection of multiplication-by-m maps. These maps can be defined
inductively on every group scheme in the following way. Let 7r : G -> S be
a group group scheme over S with identity element 0"0 : S -> G, inverse map
i : G -> G, and group multiplication f.L : G x s G -> G. Also let ide: G -> G
be the identity map on G. For each integer m, the multiplication-by-m map
on G is the morphism
[mJ: G --> G
defined inductively by the rules
[IJ = ide, [m+ 1] = f.L0 ([m] x [1]), [m - 1] = f.L 0 ([m] x i).
§4. Arithmetic Surfaces
Let R be a Dedekind domain. An arithmetic surface over Spec(R) is the

arithmetic analogue of the fibered surfaces studied in III §8. Here Spec(R)
plays the role of the base curve, and an arithmetic surface is an R-scheme
e ~ Spec(R) whose fibers are curves. For example, if R is a discrete valu-
ation ring, then there will be two fibers. The generic fiber will be a curve
over the fraction field of R and the special fiber will be a curve over the
residue field of R. Just as in the case of fibered surfaces, an arithmetic
surface e may be regular (non-singular) even if it has singular fibers.
Definition. Let R be a Dedekind domain with fraction field K. For exam-
ple, R could be a discrete valuation ring. Intuitively, an arithmetic surface
(over R) is a "nice" R-scheme e whose generic fiber is a non-singular con-
nected projective curve C / K and whose special fibers are unions of curves
over the appropriate residue fields. Note that the special fibers may be
reducible or singular or even non-reduced. This intuitive definition will
suffice for our purposes, but for the technically inclined, we indicate that
the word "nice" is an abbreviation which means that e is an integral, nor-
mal, excellent scheme which is flat and of finite type over R. t
Remark 4.1.1. An arithmetic surface e is a one-dimensional family of
one-dimensional varieties, so it is a scheme of dimension two. One might
instead call e a curve over R, since it has relative dimension one over R
(i.e., the fibers are one-dimensional). We will frequently be interested in
arithmetic surfaces which are regular, or proper over R, or smooth over R.
We recall the intuitions from section 2. An arithmetic surface e is regular if
it is non-singular as a surface, e is proper over R if its fibers are complete,
and e is smooth over R if its fibers are non-singular. If e is smooth over R,
then it is automatically regular, but in general the converse is not true.
Remark 4.1.2. The definition of an arithmetic surface e ensures that even
if e is not regular, its set of singular points is a finite set of closed points.
In other words, an arithmetic surface is regular in codimension one. This
means that there is a theory of Wei I divisors on e. In particular, for any
irreducible curve Fee (equivalently, any point FEe of codimension
one), the local ring () F of e at F is a discrete valuation ring. We denote
the corresponding normalized valuation by
ordF: K(e)* -----+ Z,
Integral is equivalent to reduced and irreducible (Hartshorne [1, I1.3.1]),
normal means that all local rings are integrally closed (Hartshorne [1, II ex-
ercise 3.8]), flat means that the fibers vary "nicely" (Hartshorne [1, III §9]),
finite type means the extensions of local rings are finitely generated algebras
(Hartshorne [1, II §3]), and excellent is a somewhat technical condition which
won't concern us, but see for example Matsumura [1, Ch. 13].
The Arithmetic Surface e : y2 = x 3 + 2x2 + 6 over Spec(Z)

Figure 4.1
where K(e) is the function field of e. For the basic thoery of Weil divisors,
principal divisors, and the divisor class group, see Hartshorne [1, II §6]. We
will continue our discussion of divisors on arithmetic surfaces in section 7.
Example 4.2.1. The projective line lP'k over R is an arithmetic surface
over R. For any maximal ideal p of R, the fiber over p is lP'~., the projective
line over the residue field k = R/p. Notice that lP'k is both proper and
smooth over R.
Example 4.2.2. Let e C lP'i be the closed subscheme of lP'i given by the
equation
e: y2 = x 3 + 2X2 + 6.
The generic fiber of e is an elliptic curve E /Q with discriminant ~ =
-2 6 ·3· 97, so for all primes p i= 2,3,97, the fiber ep is a (non-singular)
elliptic curve over lFp . The fibers over the "bad" primes are
e97 : Y
2
= (x + 66) 2 (x + 64).
The arithmetic surface e/z is illustrated in Figure 4.l.
The arithmetic surface e is proper over Z, since it is a closed subscheme
of lP'~ (2.8). It is clear that e is not smooth over Z. since it has singular
fibers. \Ve claim that e is a regular scheme. To see this, it suffices to
check that e is regular at the singular points on the fibers. We will check
that the point PEe corresponding to the cusp x = y = 2 = 0 on the
fiber e2 is non-singular. The maximal ideal Mp of the local ring (') p at P
is generated by x, y, and 2, and the residue field at P is (')p/Mp ~ IF2 • By
definition, e is regular at P if
This dimension cannot be less than two, so we must show that Mp /M~
can be generated by two of x, y, 2. Using the equation for e, we see that
so x and yare generators. This proves that e is regular at P. We will

leave for the reader the analogous calculations at x = y = 3 = 0 and
x + 66 = y = 97 = 0 (exercise 4.14).
The scheme e is thus regular and proper over Z. If we discard the
three singular points on the three singular fibers, we obtain an open sub-
scheme eO c e with the property that eO is smooth over Z. Of course, eO
will not be proper over Z, since some of its fibers are missing points.
Example 4.2.3. Let ec IP~ be the closed subscheme of IP~ given by the
equation
e : y2 = x 3 + 2X2 + 4.
The singular fibers of e are e2, e5, and e7. The scheme e is not regular,
since one easily checks that the point x = y = 2 = 0 is a singular point
of e.
Let 7r : e -> Spec(R) be an arithmetic surface and let p E Spec(R) be
a point with residue field kp = Rip. The fiber
ep = e XR p = e xSpec(R) Spec(kp)
is a curve, but it may be reducible or singular or even non-reduced. More
precisely, we can write the fiber as a union
r
ep = 2: n F i i
i=l
for certain irreducible curves Fl, ... ,Frj kp and multiplicities nl, ... ,nr 2: 1
in the following manner. Fix a uniformizer U E R for p, that is, ordp(u) = 1.
Then 7r*(u) = U 0 7r is a rational function on e, and the fiber of e over p is
given by
ep = 2: ordF(7r*u)F.
FC7r-'(p)
The Special Fiber e5 : (y2 - x 3 - 3x 2)(y - 2f(2y - x - 3) =a

Figure 4.2
Here the sum is over the irreducible components of the fiber over p, and
ordF is the normalized valuation on K(e) corresponding to F (4.1.2).
There are several ways in which a point x E ep can be a singular point
of the fiber ep . It may lie on a component F with multiplicity n ~ 2, it
may be a point where two or more components intersect, or it may be a
singular point of a particular component. The following example illustrates
these ideas.
Example 4.2.4. Consider the arithmetic surface ec A~ defined by the
equation
e : 2y5 - (x + 1)y4 - (2 x 3 + x 2 + X)y3

+ (x 4 - x 3 + 3x 2 + X - 2)y2 + (x 4 + 3x 3)y - x4 - x 3 + x 2 = 5.
We are going to look at the special fiber e5 of e over the point (5) E Spec Z.
This special fiber is the curve in A~5 defined by reducing the equation of e
modulo 5, so after some algebra we find
Thus e5 consists of three irreducible components, which we label as
We have illustrated the special fiber e5 in Figure 4.2. Such illus-

trations can be very useful for visualizing components, multiplicities, and
intersections, as long as one keeps in mind that one is looking at a drawing
in ]R2 which purports to represent a curve in characteristic p! In particu-
lar, there may be intersection points which are "hidden" because]R is not
algebraically closed.
The component F2 of e5 appears with multiplicity 2, and each of the
other components has multiplicity 1, so as a scheme the special fiber has
the form
In particular, the scheme e 5 is neither irreducible nor reduced. Every point

on F2 is a singular point of e 5 , since F2 itself appears with multiplicity
greater than 1. The other singular points on the special fiber are the
node (0,0) on Fl and the points where the various F/s intersect, such as
the point (3,2) on Fl n F2 and the points (2,0) and (3,3) on Fl n F3·
(Remember that the fiber e 5 lives in characteristic 5.)
The next proposition says that if an arithmetic surface e is regular
and if a point x E e p on its special fiber lies in the image of an R-valued
point P E e(R) (Le., if x = P(p)), then x is automatically a non-singular
point of ep.
Proposition 4.3. Let n : e -> Spec(R) be a regular arithmetic surface
over a Dedekind domain R, and let p E Spec(R).
(a) Let x E e p c e be a closed point on the fiber ofe over p. Then
ep is non-singular at x ~ n*(p) ct. M~ x'

Here n* is the natural map n* : R -> Oe,x induced by n, and Me,x is the
maximal ideal of the local ring Oe,x of e at x.
(b) Let P E e(R). Then ep is non-singular at P(p).
PROOF. To ease notation, we will write
for the fiber of e over p, and we will let llJ = n* (p )(')e,x' Notice that llJ C
Me,x, since x lies on the special fiber over p.
(a) We first assume that llJ ct. M~,x and prove that x is a non-singular point
of e. We are given that e is regular, so, by definition, Oe,x is a regular local
ring of dimension two. This means that we can find elements II, hEMe ,x
so that
Me,x = II (')e,x + h(')e,x + M~,x'
If we write p = tR, then n*(t) E llJ c Me,x, so
Our assumption is that n*(t)(')e,x = llJ ct. M~,x' which means that at least
one of al and a2 is not in Me,x, and hence at least one of them is a unit
in (')e,x' Switching II and h if necessary, this means that
Me,x = n*(t)Oe,x + h(')e,x + M~,x = llJ + h(')e,x + M~,x'
The fiber e(as a scheme, which includes multiplicities associated to

e
non-reduced components) is = e x R (Rip), so its local ring at x is ob-
tained from the local ring of e by reduction modulo p. In other words,
and Me,x = Me,xlllJ·

Therefore
Hence Me ,x jM~ is generated by the single element 12, which shows

L.!X
that C) e,x is a regular local ring of dimension one, and so x is a non-singular

point of e.
This proves the implication that we will need for part (b). We will
leave the proof of the opposite implication as an exercise for the reader
(exercise 4.17).
(b) We assume that 7r*(p) C M~ x and derive a contradiction. Using the
fact that 7r 0 P is the identity map on Spec(R), we compute
p = (7r 0 P)*(p) = P* 0 7r*(p) c P*(M~,x) = (P*M e ,x)2 = p2.
The last equality follows from the fact that P : Spec(R) -> e is a mor-
phism of schemes, so by definition (Hartshorne [1, II §2]) the induced map
P* : C)e,x -> Rp is a local homomorphism of local rings. This means in
particular that P*Me,x = p.
But P is a maximal ideal of the Dedekind domain R, so the inclu-
sion p C p2 is impossible. Therefore 7r*(p) 1:. M~ x. Applying (a), we
conclude that x is a non-singular point of the fiber e~, which concludes the
proof of (b). 0
The following important corollary says that the smooth part of a

proper regular arithmetic surface is large enough to contain all of the ra-
tional points on the generic fiber. For an example which shows that the
regularity condition is necessary, see (5.4.4) in the next section.
Corollary 4.4. Let R be a Dedekind domain with fraction field K,
let ej R be an arithmetic surface, and let C j K be the generic fiber of e.
(a) If e is proper over R, then
C(K) = e(R).
(b) Suppose that the scheme e is regular, and let eO c e be the largest
subscheme of e such that the map eo -> Spec(R) is a smooth morphism.
Then
(c) In particular, if e is regular and proper over R, then
C(K) = e(R) = eO(R).

PROOF. (a) This is really just a special case of the valuative criterion of
properness. Any point in e(R) can be specialized to the generic fiber to
give a point in C(K), so there is a natural map e(R) ~ C(K). This
map is clearly one-to-one, since two morphisms Spec(R) ~ e which agree
generically (i.e., on a dense open set) are the same. Thus e(R) '--+ C(K).
Let P E C(K) be a point. We are given that e is proper over R, so the
valuative criterion (2.7) says that there is a morphism O'p : Spec(R) ~ e
making the following diagram commute:
Spec(K) ----> Spec(R).
This proves that every point in C(K) comes from a point in e(R), so
e(R) = C(K).
(b) Proposition 4.3 says that every point in e(R) intersects each fiber
at a non-singular point of the fiber. But, by definition, eO is the com-
plement in e of the singular points on the fibers. Therefore the natural
inclusion eO(R) '--+ e(R) is a bijection.
(c) This is immediate from (a) and (b). D
The previous corollary (4.4) says that if e is a regular arithmetic sur-

face that is proper over R, then the smooth part eO of e is large enough
so that all of the K-valued points on the generic fiber extend to R-valued
points of eo. This raises two questions. First, given a (non-singular projec-
tive) curve C defined over K, does there exist a regular arithmetic surface e
proper over R whose generic fiber is C I K? Second, assuming such proper
regular models exist, to what extent is there a minimal such model? The
following theorem gives the answer to these questions. It is the arithmetic
analogue of the geometric results described in (nI. 7. 7) and (nI.8.4). We
will discuss the construction of these minimal models further in §7.
Theorem 4.5. Let R be a Dedekind domain with fraction field K, and
let CIK be a non-singular projective curve of genus g.
(a) (Resolution of Singularities for Arithmetic Surfaces, Abhyankar [1,2]'
Lipman [1,2]) There exists a regular arithmetic surface el R, proper over R,
whose generic fiber is isomorphic to C I K. We call e/ R a proper regular
model for C I K.
(b) (Minimal Models Theorem, Lichtenbaum [1], Shafarevich [2]) Assume
that 9 ::::: 1. Then there exists a proper regular model emin I R for C I K
with the following minimality property:
Let eI R be any other proper regular model for C I K. Fix an isomor-
phism from the generic fiber of e to the generic fiber of emin . Then the
induced R-birational map
e -_+ emin
is an R-isomorphism. We call emin / R the minimal proper r'egular model

for C / K. It is unique up to unique R-isomorphism.
PROOF. (a) See Abhyankar [1,2] and Lipman [1,2]. There is also a nice
exposition of Lipman's proof in Artin [2]. In the case that C has genus 1,
we will explicitly construct a proper regular model for C / K in §9.
(b) See Lichtenbaum [1, Thm. 4.4] and Shafarevich [2, lectures 6,7,8].
There is a nice summary of the main results with sketches of the proofs in
Chinburg [1]. See also §7 for a further discussion. D
Just as in (1II.8.4.1), the importance of the minimal regular model

lies in the fact that every automorphism of its generic fiber extends to a
morphism of the entire scheme.
Proposition 4.6. Let R be a Dedekind domain with fraction field K,

and let C / K be a non-singular projective curve of genus 9 :::: 1. Let e/ R
be a minimal proper regular model for C / K, and let eO c e be the largest
subscheme of e which is smooth over R. Then every K -automorphism
7 : C / K --) C / K of the generic fiber of e extends to give R-automorphisms
and
PROOF. The fact that 7 extends to an R-automorphism e --) e is exactly

the definition of minimality given in (4.5b). Next take any point .1: E eO
and choose some neighborhood U c eO of 1:. Then U is smooth over R.
Further, U is an open subset of e, since eO is open in e. We know that
7: e --) e is an R-isomorphism, so 7(U) is an open neighborhood of 7(X)
and is smooth over R. Therefore 7(X) E eO, which proves that 7(eO) Ceo.
Applying the same argument to 7- 1 gives 7- 1(eO) c eO, which completes
the proof that 7 gives an R-automorphism of eO. D
§5. Neron Models
Let K be the fraction field of a discrete valuation ring R. The Neron model
of an elliptic curve E / K is an arithmetic surface G/ R whose generic fiber is
the given elliptic curve. The scheme G/ R is characterized by the fact that
it is large enough so that every point of E gives a point of G, but small
enough so that the group law on E extends to make G into a group scheme
over R. Of course, when we talk of "points of E," we mean more than just
the points of the underlying scheme. This leads to the following definition.
Definition. Let R be a Dedekind domain with fraction field K, and let

ElK be an elliptic curve. A Neron model for ElK is a (smooth) group
scheme £1 R whose generic fiber is ElK and which satisfies the following
universal property:
Let XI R be a smooth R-scheme (i.e., X is

smooth over R) with generic fiber XI K, and Neron )
let 1>K : X IK -+ ElK be a rational map de- ( Mapping
fined over K. Then there exists a unique R- Property
morphism 1> R : X I R -+ £ I R extending 1> K .
Remark 5.1.1. A Neron model £1 R is a smooth R-scheme. This means

that for every point P E Spec(R), the fiber £p of £ -+ Spec(R) is a non-
singular variety defined over the residue field k(p); see (2.9). However,
as we will soon see, the fiber £p over a closed point p may have several
components and may not be complete, so in general £ will not be proper
over R.
Remark 5.1.2. In the Neron mapping property we have only required
that the map 1>K : XIK -+ ElK on the generic fiber be a rational map. It
turns out that any rational map from a non-singular variety to an elliptic
curve is a morphism. See (6.2b) below for an even more general statement.
Remark 5.1.3. The most important instance of the Neron mapping prop-
erty is the case that X = Spec(R) and X = Spec(K). Then the set of K-
maps X IK -+ ElK is precisely the group of K-rational points E(K), and
the set of R- morphisms XI R -+ £ IRis the group of sections £ (R). So in
this situation the Neron mapping property says that the natural inclusion
£(R) <---4 E(K)
is a bijection. If R is a complete discrete valuation ring with algebraically

closed residue field, then one can show that the equality £(R) = E(K)
suffices to ensure that the group scheme £IR is a Neron model for ElK.
See exercise 4.30.
We begin our study of Neron models by proving that they are unique
and behave well under unramified base extension.
Proposition 5.2. Let R be a Dedekind domain with fraction field K,
and let ElK be an elliptic curve.
(a) Suppose that cdR and c21R are Neron models for ElK. Then there
exists a unique R-isomorphism 'IjJ : £d R -+ £2/ R whose restriction to the
generic fiber is the identity map on ElK. In other words, the Neron model
of ElK is unique up to unique isomorphism.
(b) Let K' / K be a finite unramified extension, and let R' be the integral
closure of R in K'. Let £ I R be a Neron model for E / K. Then £ x R R' is a
Neron model for ElK'. (N.B. If K' I K is ramified, this result will generally
not be true.)
PROOF. (a) The identity map ElK -> ElK is a rational map from the
generic fiber of £1 to the generic fiber of £2, and £1 is smooth over R,
so the Neron mapping property for £2 says that the identity map extends
uniquely to an R-morphism 'lj; : £1/ R -> £21 R. In a similar fashion we
obtain a unique R-morphism cP : £21 R -> £1/ R which is the identity map
on the generic fiber. But then cP 0 'lj; : £1/ R -> £1/ R and the identity map
£1/ R -> £1/ Rare R-morphisms which are the same on the generic fiber, so
the uniqueness part of the N eron mapping property says that cP 0 'lj; equals
the identity map. This proves that cP and 'lj; are isomorphisms.
(b) Let X' I R' be a smooth R'-scheme with generic fiber X' I K', and let
cPK' : X/ K , -> ElK' be a rational map. The composition
X' -----> Spec(R') -----> Spec(R)
makes X' into an R-scheme. Further, our assumptions on K' imply that
the map
Spec(R') -----> Spec(R)
is a smooth morphism. (See exercise 4.19.) Hence the composition is a
smooth morphism (2.10), so X' is a smooth R-scheme.
Now the Neron mapping property for £1 R tells us that there is an R-
morphism
X'.!!.!!... £
whose restriction to the generic fiber is the composition
X'~ExKK'~E.
The two R-morphisms cPR : X' -> £ and X' -> Spec(R') determine an R-
morphism (and thus an R'-morphism) to the fiber product,
cPR' : X' -----> £ xR R'.
Further, the restriction of cPR' to the generic fiber is cPK'. This gives the
existence part of the N eron mapping property. We will leave it to the reader
to prove the uniqueness part, which completes the proof that £ x R R' is a
Neron model for ElK'. 0
Let R be a discrete valuation ring with fraction field K. We are going

to use the Weierstrass equation of an elliptic curve ElK directly to con-
struct an R-group scheme WO c IP'~ whose generic fiber is ElK. If the clo-
sure W ofWo in IP'~ is regular, then we will also prove that WO(R) = E(K),
so WO satisfies the most important instance of the Neron mapping prop-
erty (5.1.3). In particular, if E has good reduction, then we will see (6.3)
that a minimal Weierstrass equation for ElK already defines a Neron
model.
Theorem 5.3. Let R be a discrete valuation ring with fraction field K,

let E / K be an elliptic curve, and choose a Weierstrass equation for E / K
with coefficients in R,
This Weierstrass equation defines a scheme W c IP'~. Let WO c W be the

largest subscheme of W which is smooth over R.
(a) Both W / Rand WO / R have generic fiber E / K.
(b) The natural map W(R) ---+ E(K) is a bijection. lfW is regular, then
the natural map WO(R) ---+ W(R) is also a bijection, so in this case there is
a natural identification WO(R) = E(K).
(c) The addition and negation maps on E extend to R-morphisms
and
which make WO into a group scheme over R. The addition map further
extends to an R-morphism
W O xRW --+ W
giving a group scheme action ofW o on W.

Remark 5.4.1. If E / K has good reduction and if we take a minimal
Weierstrass equation for E/ K, then W itself is smooth over R. So in this
case (5.3) says that W = WO is a group scheme over R.
Remark 5.4.2. If E / K has bad reduction, then there is exactly one sin-
gular point on the reduction E (mod p). In other words, the special fiber W
of W contains exactly one singular point, say r EWe W, and then WO is
obtained by discarding that point,
WO = W" {r}.
Remark 5.4.3. The scheme W / R in (5.3) is proper over R, since it is a

closed subscheme of IP'~ (2.8). It follows from the valuative criterion (4.4a)
that W(R) = E(K). However, in general, the scheme W will not be regular,
since a singular point on the special fiber will often be a singular point
of W. So, in general, we cannot use (4.4b) to deduce that WO(R) = E(K).
Intuitively, if W is singular, then there will be points P E E(K) = W(R)
which go through the singular point of W. Thus, in general, WO will not
be large enough to be a Neron model, because WO(R) =F E(K), whereas W
itself will be too large to be a Neron model, because the group law on E
will not extend to all of W.
Example 5.4.4. We illustrate the previous remark by looking at the curve
If a6 E R*, then W is smooth over R (exercise 4.20(b)), so W is a Neron

model for E. If v(a6) 2 1, then the special fiber
has the singular point (0,0), so
We consider two cases.

First, if v(a6) = 1 (i.e., a6 is a uniformizer in R), then W is a regular
scheme from exercise 4.20(a). It follows from (4.4c) that WO(R) = W(R) =
E(K). We can also see this directly as follows. If P E W(R) were to
go through the singular point on the special fiber, then we would have
p = (x,y) == (O,O)(modp), which means that x,y E p. But then the
equation for E would give
contradicting the assumption that v(a6) = 1. Hence WO(R) = E(K).

Second, if v(a6) 2 2, then W is not a regular scheme and WO will
not be a Neron model for E, despite the fact that WO is a group scheme
with generic fiber E. For example, suppose that a6 = a 2 with v(a) 2 1.
Then the point P = (0, a) E E(K) = W(R) is not in WO(R), since P ==
(0, O)(mod p).
PROOF (of Theorem 5.3). (a) W is projective over R, since it is the closed
subscheme of IP'~ = Proj R[X, Y, Zl defined by the single homogeneous
equation
Its generic fiber is the variety in IP'k defined by this same equation. Thus
the generic fiber of W is precisely E / K.
IfWo is not equal to W, then as described above in (5.4.2), WO consists
of W with one point on the special fiber removed. In particular, WO and W
have the same generic fiber, so the generic fiber of WO is also E / K.
(b) The scheme W is a closed subscheme of IP'~, so it is proper over R (2.8).
Further, its generic fiber is E / K from (a), so (4.4a) tells us that E(K) =
W(R). This proves the first part of (b). If in addition W is regular,
then (4.4b) says that W(R) = WO(R), which gives the second equal-
ity E(K) = WO(R).
(c) Let
J-l : W xR W ----t W and i: W ----t W
be the rational maps on W induced by the addition and negation laws on
the generic fiber E / K of W. The fact that the generic fiber is a group
variety means that J-l and i satisfy all of the group axioms on a non-empty
open subscheme of W, so they will satisfy the group axioms on the largest
open subscheme on which they are defined. In other words, if we can show
that J-l is a morphism on WO XR Wand that i is a morphism on W, then
the group axioms are automatically true.
If W is smooth over R, we are going to prove that J-l is a morphism on
all of W XR W. If W is not smooth over R, then as explained above (5.4.2),
the special fiber Wcont(l,ins a unique singular point 'Y and WO = W " b}.
In this situation we will show that J-l is a morphism except at the single
point
In particular, it is a morphism on WO x R W.
In order to simplify our calculations, we will assume that the residue
field k does not have characteristic 2 or 3 (equivalently, 2 and 3 are units
in R). The general case is similar, but the formulas are considerably longer.
This assumption allows us to make a change of variables in IP'~ and put our
Weierstrass equation in the form
W : y2 Z = X 3 + AX Z2 + BZ3.
In other words, W is the closed subscheme of IP'~ defined by this homoge-
neous equation.
Let Waft be the affine open subscheme of W defined by
Waft = {Z I- O} c W,
and let x = X/Z and Y = Y/Z be affine coordinates on Waft. The addition
map J-l is then given by the usual formula [AEC, III.2.3]
J-l = [(X2 - Xd((Y2 - yd 2 - (X2 - xd 2(X2 + xd),

(Y2 - yd 3 + (X2 - XI)2(XIYI - X2Y2 + 2X2YI - 2XIY2), (X2 - XI)3].
More precisely, this formula gives the restriction of J-l to Waft XR Waft,
and on this affine scheme J-l will be a morphism except possibly along the
closed subscheme where its three coordinate functions vanish. Looking
at the third coordinate and then at the second coordinate, we see that J-l
is a morphism on Waft XR Waft except possibly on the closed subscheme
defined by the equations
X2 - Xl = Y2 - YI = O.
In other words, /-L is a morphism off of the diagonal.

To deal with points on the diagonal, we use the relations
and Y§ = x~ + AX2 + B
which hold identically on Waif XR Waif to rewrite the addition map /-L as
/-L = [(YI+ Y2)((XI + X2)(YI + Y2)2 + (xi + XIX2 + x~ + An,

(xi + XIX2 + x~ + A)3
- (YI + Y2)2 ((Xl + X2)3 + A(XI + X2) + B - YIY2), (YI + Y2)3].
(See [AEC, III.3.6.1] for a similar calculation.) Just as above, we see that /-L
is a morphism on Waif XR Waif except possibly on the closed subscheme
YI + Y2 = xi + XIX2 + x~ + A = 0.
We have now proven that /-L is a morphism on Waif XR Waif except on
the sub scheme defined by the four equations
X2 - Xl = Y2 - YI = YI + Y2 = xi + XIX2 + x§ + A = 0.
A little algebra and the fact that 2 E R* shows that this subscheme is
YI = Y2 = 0, 3xi + A = 0.
In particular, it is contained in the diagonal of Waif X R Waif, so if we iden-
tify Waif with this diagonal, then /-L is a morphism except on the subscheme
Y = 3x2 +A = 0.
Using the relation y2 = x3 + Ax + B and the fact that 3 E R*, we see that
the discriminant 4A 3 + 27 B2 is contained in the ideal generated by Y and
3x2 + A. Hence if W is smooth over R, which implies that its discriminant
is a unit in R, then /-L is a morphism on all of Waif X R Waif. Similarly, if W
is not smooth over R, then /-L will be a morphism on Waif X R Waif away
from the subscheme
YI = Y2 = 0, 3xi + A = 0,
which is precisely the singular point on the special fiber of the diagonal.
Next let W~1f be the affine open subscheme of W defined by
W~1f = {Y t O} C W.
Notice that W XR W is covered by the four affine subschemes
Waff XR Waff,
since W does not intersect the scheme Y = Z = O. We have already

dealt with the restriction of J.1 to the first set, so it remains to show that J.1
is a morphism on each of the last three. We will leave this task for the
reader (exercise 4.22), since the proof is similar to the argument given
above.
Finally, we observe that the negation map
i :W XR W ---> W, [X, Y, Z] f---+ [X, - Y, Z],
is a morphism on W, since it is actually the restriction of a morphism on

1P''k
XR 1P''k. This completes the proof of Theorem 5.3. 0
§6. Existence of Neron Models
In this section we are going to prove the existence of Neron models for ellip-
tic curves. The proof, which closely follows the exposition of Artin [1, §1]'
is largely scheme-theoretic and is at a more advanced level than the other
material in this chapter. The reader who is willing to accept the statement
of Theorem 6.1 should read Remarks 6.1.1-6.1.3 and can then proceed to
the next section with no loss of continuity.
Theorem 6.1. Let R be a Dedekind domain with fraction field K,

let ElK be an elliptic curve, let el R be a minimal proper regular model
for ElK (4.5), and let £/R be the largest subscheme of elR which is
smooth over R. Then £ / R is a Neron model for E / K.
Remark 6.1.1. The generic fiber of e is the non-singular curve E, so e

has only finitely many singular fibers. Each fiber of e consists of one or
more irreducible components, possibly with multiplicities (see §4), say
Tp
ep = l:.: npiFpi.
i=1
Then £ is formed by discarding from e all Fp;'s with npi 2': 2, all singular
points on each F pi , and all points where the FPi'S intersect one another.
Remark 6.1.2. Continuing with the notation from (6.1.1), the group
scheme el R comes equipped with an identity element, which is an R-valued
point 0'0 E e(R). The image O'o(R) of the identity element is a curve on e
which will intersect the fiber ep at the point O'o(p). Proposition 4.3 tells us
that 0'0 (p) will be a non-singular point of ep , so it will lie on an FPi hav-
ing multiplicity npi = 1. The component of ep containing O'o(p) is called
the identity component of ep . The scheme obtained by removing all non-
identity components from the fibers of e is called the connected component
(of the identity) ofe. The image O'o(R) of the identity element lies in e, so
we can define the identity component of ep and the connected component
(of the identity) of e in an analogous manner. The connected component
of e is a subgroup scheme of e; see exercise 4.25. We will see later (9.1)
that it is isomorphic to the smooth part of the scheme defined by a minimal
Weierstrass equation for E.
Remark 6.1.3. With notation as in (6.1), one can prove that the group
law e x R e ---+ e extends to give a group scheme action
See exercise 4.23. We proved a special case of this in (5.3c).

Before beginning the proof of (6.1), we want to say a few words explain-
ing why the smooth part of a minimal proper regular model for ElK turns
out to be a Neron model. In other words, how will we use the four proper-
ties "smooth," "minimal," "proper," and "regular"? First, the properness
of e over R ensures that E(K) = e(R) (4.4a). Next, the regularity of e tells
us that every R-point lies in the smooth part of e (4.4b), so e(R) = e(R).
This gives E(K) = C(R), which is an important case of the Neron mapping
property (5.1.3). Thus the properness and regularity of e are mainly used
to obtain the Neron mapping property. On the other hand, the smoothness
and minimality of e are used to prove that e is a group scheme over R.
In particular, the minimality implies that any K-automorphism of E, such
as a translation-by-P map for some point P E E(K), will extend to give
an R-automorphism of e. These translation maps on e will be essential for
showing that the group law on ElK extends to give e the structure of a
group scheme over R.
With these preliminary comments completed, we begin the proof of
Theorem 6.1. The first step in the proof is the following generalization of
a theorem of Weil [3]. Weil's theorem asserts that a rational map from a
smooth variety to a complete group variety is automatically a morphism,
and Artin [1, Prop. 1.3] has extended Weil's theorem to a scheme-theoretic
setting.
Proposition 6.2. (Weil [3], Artin [1]) Let R be a Dedekind domain,
let G IR be a group scheme over R, let XI R be a smooth R-scheme, and
let ¢ : X --.. G be a rational map over R. Write Dom(¢) for the domain
of ¢, and suppose that Dom( ¢) is dense in every fiber of XI R.
(a) The complement X" Dom(¢) is a subscheme of X of pure codimension
one.
(b) If G is proper over R, then Dom(¢) = X. In other words, ¢ is a
morphism.
PROOF. We will write }L(g,h) = gh and i(g) = g-l. Further, to simplify
our exposition and help reveal the underlying ideas, we will phrase our
argument in terms of points. But the reader should be aware that in
order to be completely rigorous, our "points" should be T-valued points for
arbitrary R-schemes T. As an alternative, the proof can be given purely
scheme-theoretically, a task which we will leave for the reader.
Having made this disclaimer, we begin by considering the rational map
F: X XR X --- .. G, F(x, y) = ¢(x)¢(y)-l.
We claim that there is a natural identification
Dom(¢) ......::..-. .6. n Dom(F), X f---+ (x,x),
where .6. is the diagonal in X XR X. To see this, take a point x E Dom(¢).

Then F(x, x) = ¢(x)¢(X)-l is defined, so (x, x) E Dom(F). Conversely,
if (x, x) E Dom(F), we can use the fact that Dom(F) is open to find a non-
empty open set U C X so that x xR U c Dom(F). Next, since Dom(¢) is
open, we can find a point y E Un Dom(¢). It then follows from
¢(x) = F(x, y)¢(y)
that ¢ is defined at x, so x E Dom(¢). This completes the proof of the

claim.
Let K(X x X) be the function field of the scheme X x R X, and let (')c,o
be the local ring of G along the identity section; that is, (')c,o is the ring of
rational functions on G which are well-defined at some point on the image
of the map 0"0: Spec(R) ---+ G, where 0"0 is the identity element of the group
scheme G.
The rational map F defines a ring homomorphism
F* : (')c,o ----> K(X x X),

Let f E (')c,o. If x E Dom(¢), then (x, x) E Dom(F) from the claim
proven above, and further F(x,x) = ¢(x)¢(X)-l is the identity element
of G, so F*(f) is defined at (x, x). Conversely, if F*(f) = f 0 F is defined
at (x,x) for all functions f E (')c,o, then F must be defined at (x,x). This
proves that
x E Dom(¢) <¢=> (x, x) E Dom(F* f) for all f E (')c,o

<¢=> F*((')c,o) C (')xxX,(x,x),
where tJxxx.(x,x) C K(X x X) is the local ring of X XR X at (x, x).

The scheme X x R X is smooth over R, so in particular it is normal.
This implies that a function f E K(X x X) will be defined at (x, x) unless
its polar divisor divoo(f) goes through (x,x). (This is a standard property
of normal schemes. It follows, for example, from Hartshorne [1, II.6.3A].)
In other words, the local ring tJXxX,(x,x) can be characterized as
tJXXX,(x,x) = {g E K(X x X)* ; (x, x) tJ- divoo(g) }U{O}.
Combining this with our description of the domain of ¢ from above yields
X" Dom(¢) = {x EX; F*(tJe,o) rt tJXXX,(x,x)}

= {x EX; (x,x) E divoo(F* f) for some f E tJe,o}
~~n U divoo(F* f)
!Er:Jc.o
U (~n divoo(F* f)).

!Er:Jc,o
The diagonal ~ is a complete intersection in X x R X, and each divisor

divoo(F* f) has pure co dimension one in X XR X, so each of the intersec-
tions ~ n divoo(F* f) has pure co dimension one in ~. It follows that the
union over f E tJe,o also has pure co dimension one in ~, since we know
a priori that it is a proper closed subset of~. This completes the proof
of (a).
(b) The following lemma (6.2.1) says that a rational map from a smooth
scheme to a proper scheme is defined off of a subset of codimension at least
two. Then (6.2.1) and (a) imply (b).
Lemma 6.2.1. Let R be a Dedekind domain, let XI R be a smooth R-
scheme, let 11 I R be a proper R-scheme, and let ¢ ; X - - + ~ be a dominant
rational map defined over R. Then every component of X " Dom( ¢) has
codimension at least two in X.
PROOF. Let Z C X be an irreducible subscheme of codimension one in X.

We need to show that ¢ is defined at the generic point of Z (i.e., ¢ is defined
on a non-empty open subset of Z). Consider the local ring tJx,z of X at Z.
It is a local ring of dimension one, and it is regular since XI R is smooth,
so it is a discrete valuation ring.
The dominant rational map ¢ ; X --+ ~ induces a morphism
Spec K(X) ----> Spec K(~)
from the generic point of X to the generic point of~. In other words,
composition with ¢ induces an inclusion of function fields K(~) ~ K(X).
§6. Existence of Neron l\1odels 329
This gives us the commutative diagram
x
T
Spec (')x,z
T
Spec K(X) -+ SpecK(13).
The discrete valuation ring (')x.z has fraction field K(X), and we are given
that 13 is proper over R, so the valuative criterion of properness (2.7) implies
that the rational map
l'
Spec (')x.z -+ X ---~ 13
extends to a morphism Spec (') x, z ----+ 13. This says precisely that cP is defined
at the generic point of Z, which completes the proof of (6.2.1), and with it
also the proof of (6.2b). 0
We can use (6.2) to find the Nerem model of an elliptic curve with
good reduction.
Corollary 6.3. Let R be a Dedekind domain with fraction field K,
let ElK be an elliptic curve given by a Weierstrass equation
having coefficients in R, and let W C lP'k

be the closed subscheme of lP'k
defined by this VVeierstrass equation. Suppose that W is smooth over R
or, equivalently, that the Weierstrass equation has good reduction at every
prime of R. Then WI R is a Neron model for ElK.
PROOF. Theorem 5.3 says that the addition law on ElK extends to make
W into a group scheme over the localization of R at each of its prime ideals.
These group laws are given by the same equations, so they fit together to
make W into a group scheme over R. It remains to verify that W has the
Neron mapping property.
Let XI R be a smooth R-scheme with generic fiber XI K, take any
rational map cPK : X IK --~ ElK defined over K, and let cP : X --~ W be
the associated rational map over R. The fact that W is a closed subscheme
of lP'~ implies that it is proper over R (2.8), so we can use (6.2b) to deduce
that the rational map c/J extends to a morphism. This proves that WI R
has the Neron mapping property, so it is a Neron model for ElK. 0
Our next step is to prove that if the scheme £ in (6.1) is a group

scheme, then it will be a Neron model for E, at least provided that the
ring R is large enough. The precise property we will require R to have is
described in the following definition.
Definition. A discrete valuation ring R is called Henselian if it satisfies

Hensel's lemma; that is, R is Henselian if for any monic polynomial f(x) E
R[x] and any element a E R satisfying
f(a) == 0 (mod p) and f'(a) =I'- 0 (mod p),
there exists a unique element 0: E R satisfying
0: == a (mod p) and f(o:) = o.

The ring R is called strictly Henselian if it is Henselian and if its residue
field k = Rip is algebraically closed. (Remember our residue fields are
perfect. The usual definition of strictly Henselian requires k to be separably
closed.)
For example, if R is a discrete valuation ring, then the completion of R
with respect to its maximal ideal p is Henselian. We have seen many in-
stances in which it is helpful to work with complete discrete valuation rings,
for example in our study of formal groups [AEC, IV §6] and the reduction
theory of elliptic curves [AEC, VII §2]. In particular, we used Hensel's
lemma for complete discrete valuation rings to prove the surjectivity of the
reduction map Eo(K) --4 Ens(k) in [AEC, VII.2.1].
However, for many purposes the completion is too large, since the
completion of R will generally not be flat over R. The following general-
ization of [AEC, VII.2.1] says that the reduction map is surjective for any
Henselian ring.
Proposition 6.4. Let R be a discrete valuation ring with maximal ideal p
and residue field k, let XI R be a smooth R-scheme, and let ilk be its
special fiber. Consider the reduction map
X(R) ---+ i(k).
(a) If R is Henselian, then the reduction map is surjective.

(b) If R is strictly Henselian, then the image of the reduction map is dense
in 5::.
PROOF. (a) Replacing X by an affine neighborhood, we can assume that
X = Spec A with A = R[iI, ... , tn+rl/(h,···, fn)

for certain polynomials h, ... , fn E R[h, ... , tn+r]' Further, the assump-
tion that X is smooth over R means that the n x n minors of the Jacobian
matrix
generate the unit ideal in A.

Choose any point b = (b 1 , ... , bn ) E X(k); that is, let b E p".n(R) satisfy
h(b) == ... == fn(b) == 0 (mod p).
We need to construct a point (3 E p". n (R) with the property that
(3 == b(mod p) and h((3) = ... = fn((3) = o.

For example, suppose that X is locally a hypersurface (i.e., m = I),
given by the single equation f(t1' ... ,tn+ r ) = O. Then the Jacobian con-
a
dition says that the partial derivatives f / Oti generate A, so in particular
one of the values (af /ati)(b) must be a unit in A, say (at /atd(b) E A*.
Thus the polynomial F(t) = j(t,b 2 , .•• ,bn+r ) satisfies the hypotheses of
Hensel's lemma, so it has a root (31 E R with (31 == b1 (mod p). The
point (3 = ((31, b2 , ... ,bn ) E X(R) reduces modulo p to the original point b
in X(k), which completes the proof in this case. We will leave the general
case, which is somewhat more difficult, for the reader to complete, or see
the references for exercise 4.27.
(b) The residue field k is algebraically closed, so it is clear that Xp (k) is
dense in Xp. Now (b) follows from (a). 0
As the next proposition explains, every discrete valuation ring R can

be embedded in a minimal (strictly) Henselian ring.
Proposition 6.5. Let R be a discrete valuation ring with maximal
ideal p, residue field k, and fraction field K. Let K S be a separable closure
for K, let RS be the integral closure of R in KS, and choose an ideal pS
of KS lying above p. Let
D = {O" E G K' I K pS) = ps},

: 0" (
1= {O" ED: O"(x) - x E pS for all x E RS}
be the associated decomposition and inertia groups.

(a) Let RS(D) denote the subring of R S fixed by D, and define Rh to be
the localization of RS(D) at the maximal ideal pS n RS(D). Then Rh is
Henselian. It is called the Henselization of R.
(b) Let RS(I) denote the subring of R S fixed by I, and define R sh to be the
localization of RS (I) at the maximal ideal pS n RS (I). Then R sh is strictly
Henselian. It is called the strict Henselization of R.
(c) With the obvious notation, we have
ph = pR h , kh = k, psh = pRsh , k sh = k.
Further the natural map
of Galois groups is an isomorphism.
PROOF. (a) We are going to verify that Rh has the Henselian property.
Let f(x) E Rh[x] be a monic polynomial, which we may assume to be
irreducible and separable over Kh. (If f is inseparable, then f' (x) is iden-
tically 0, so the Henselian property is vacuously true!) Factor f over K S
as
f(x) = (x - al)(x - (2)··· (x - ad).
Suppose that a E Rh satisfies f(a) E ph and f'(a) t/: ph. This implies that
there is exactly one root, say a = ai, with the property that a - a E pS.
Further, if a E D is any element of the decomposition group of pS, then we
have
a - a(a) = a(a - a) E a(pS) = pS.
But there is exactly one root of f which is congruent to a, so a(a) must
equal a. This shows that a is fixed by D, and hence a E Rh.
(b) The proof that R sh is Henselian is similar to the proof of (a), and it
is a standard fact that the residue field is an algebraic closure of k; see for
example Serre [4, I §7]. (Remember we are assuming that k is perfect.)
(c) Again these are standard properties of Galois extensions of local fields.
o
Remark 6.6.1. It is clear from the construction (6.5) that the fraction
fields Kh and K sh are separable algebraic extensions of K. Note that in
general the fraction field of the completion of R will be transcendental
over K, in fact, of infinite transcendence degree. This is one reason why it
is often better to work with the Henselization. The moral is that one should
work with Henselizations if one only needs to solve polynomial equations,
but one has to go to the completion in order to use convergent power series.
We also mention that strictly Henselian rings play the same role for the
etale topology that local rings play for the Zariski topology.
Remark 6.6.2. The Henselization can also be described in terms of a
universal mapping property, which essentially says that it is the smallest
Henselian extension of R, and similarly for the strict Henselization. See
exercise 6.28.
Proposition 6.7. Let R be a strictly Henselian discrete valuation ring
with fraction field K, let ElK be an elliptic curve, let e I R be a minimal
proper regular model for ElK (4.5), and let cI R be the largest su bscheme
ofel R which is smooth over R (6.1.1). If the group law on ElK extends to
make c into a group scheme over R, then cIRis a Neron model for ElK.
PROOF. Let XI R be a smooth R-scheme with generic fiber XI K, and

let cPK : X ---t E be a rational map. In order to verify the Neron mapping
property, we must show that cPK extends to a morphism X --+ c.
The elliptic curve E is a proper group scheme over K, and X is smooth

over K, so applying (6.2b) with R = K to the map cPK : X --~ E, we see
that cPK extends to a morphism. This means that the rational map
cP: 1: ---~ E
induced by cPK is a morphism on the generic fiber.

We suppose that cP is not a morphism and derive a contradiction.
By assumption, E is a group scheme, so (6.2a) tells us that the set of
points where cP is not defined is a set of pure co dimension one in 1:. Hence
there is an irreducible closed subscheme Z c 1: such that cP is not defined
at the generic point 'f/z of Z. Note that the generic point of Z is given
by 'f/z = Spec Ox,z, and that the local ring Ox,z is a discrete valuation
ring because 1: is regular and Z has co dimension one. We now have the
following picture:
q,
x ---~ E c e
l'
'f/z = Spec Ox,Z
l'
q,K
1 1
Spec K(1:) ----7 Spec K(e) SpecK(e).
The scheme e is proper over R, and Ox,z is a discrete valuation ring,

so the valuative criterion of properness (2.7) says that cP extends to a mor-
phism cP : 'f/z ----+ e. In other words, if we map to e rather than to the smaller
scheme e, then cP is defined generically on Z.
We are assuming that cP : 1: --~ E does not extend generically to Z,
or equivalently that cP('f/z) E e is not contained in E. In particular, if
we let k be the residue field of R and take any point Xo E Z(k) so that
cP: 1: --~ e is defined at xo, then cP(xo) ~ E. (This is another way of saying
that cP : 1: - - ~ E is not defined generically on Z.)
The set of R-valued points 1:(R) maps to a dense set of points in the
special fiber of 1: by (6.4b). Note that this is where we use our assumption
that R is strictly Henselian. In particular we can find a point x E 1:(R)
which intersects Z at a point, call it Xo E Z(k), at which the map cP : 1: --~ e
is defined. Composing x with cP gives a rational map
which by the valuative criterion of properness (2.7) extends to a morphism

Spec(R) ----+ e. In other words cP 0 x E e(R), and by our construction it
is clear that cP 0 x ~ e(R). However, (4.4b) says that e(R) = e(R). This
contradiction completes the proof that cP extends to a morphism 1: --+ E,
and hence that E has the Neron mapping property. 0
In order to complete the proof of Theorem 6.1 for strictly Henselian

rings, it remains to show that the scheme c. is a group scheme over R. This
is done in two steps. First, we prove that there exists some group scheme A
over R such that c. and A have isomorphic dense open subsets. Second,
using the group operation on A, we show that c. and A must be isomorphic.
The proof of the first part uses an argument of Weil to construct a group
variety (or scheme) by pasting together group chunks. We will not give
the full proof of Weil's result but will be content to give a brief sketch and
refer the reader to Artin [1, §2] for the details.
Definition. Let R be a Dedekind domain, let V/ R be a smooth R-scheme
with non-empty fibers, and let
be a rational map defined over R. The map p, is called a normal law on V

if it satisfies the following two conditions:
(i) The map p, is associative; that is,
p,(p,(x,y),z) = p,(x,p,(y,z)) whenver both sides are defined.
(ii) Define rational maps
4> :V XR V---+ V XR V 'ljJ :V XR V---+ V XR V

(x,y) f---+ (x,p,(x,y)), (x,y) f---+ (y,p,(x,y)).
Then the domains of definition of 4> and 'ljJ contain a dense subset of
each fiber of V XR V, and the restriction of 4> and 'ljJ to each fiber is a
birational isomorphism.
Remark 6.8. If G is a group scheme over a Dedekind domain R, then its
group law is a normal law. Condition (i) is true because the group law
on G is associative by definition, and condition (ii) is immediate since 4>
and 'ljJ are isomorphisms. For example, the inverse of 'ljJ is the map
(x,y) f---+ ((p,(y,i(x)),x).
For a general normal law p" the requirement that the rational maps 4>
and 'ljJ satisfy condition (ii) provides a sort of inverse for the hidden group
law that p, is trying to emulate.
The following theorem says that a normal group law on V makes a
large chunk of V into a large chunk of a group scheme.
Theorem 6.9. (Weil [3]) Let R be a Dedekind domain, and let V/ R be
a smooth R-scheme of finite type over R. Suppose that every fiber of V is
non-empty and that VCR) is dense in each fiber.
Let J-l be a normal law on V. Then there exists a group scheme G / R

of finite type over R, an open subscheme U / ReV/ R, and an open sub-
scheme U' / ReG / R with the following two properties:
(i) U and U' are dense in every fiber.
(ii) There is an R-isomorphism U ~ U' so that the normal law J-l restricted
to U coincides with the group law of G restricted to U'.
PROOF. (Sketch) The underlying idea is to start with a good open sub-
scheme U of V and construct G as a union of translates of U. More pre-
cisely, for each x E U(R), let Ux be a copy of U. Then one treats Ux as if
it were "U translated by x" and uses J-l to provide gluing data to attach Ux
to U. For further details, see Artin [I, Thm. 1.12]. 0
The next result, combined with (6.7), will complete the proof that c/R
is a N eron model for E / K, at least over strictly Henselian rings.
Proposition 6.10. Let R be a strictly Henselian discrete valuation ring
with fraction field K, let E / K be an elliptic curve, let e/ R be a minimal
proper regular model for E/ K (4.5), and let E/ R be the largest subscheme
of e/ R which is smooth over R (6.1.1). Then the group law on E extends
to make E into a group scheme over R.
PROOF. The proof consists of two steps. First, we verify that the group
law on E defines a normal law on E. This allows us to apply (6.9), which
yields a group scheme G / R that is birational to E. The second step is to
show that the resulting birational map E -> G is actually an isomorphism.
Notice that the proof is somewhat indirect. Rather than proving that the
group law on E extends to E, we instead construct an auxiliary group
scheme G which extends the group law on E, and then we show that G
must equal E.
We begin with the assertion that the group law on E defines a normal
law
J-l: E XR E ---~ E.
The associativity of J-l is clear, since {1 is associative on the generic fiber

of E, and a rational map is determined by its restriction to any dense open
subset. So it remains to verify that the maps
¢:EXRE---~ EXRE 1jJ:£XR£---~ EXRE

(x,y) I---> (x,{1(x,y)), (x,y) I---> (Y,J-l(x,y)).
are defined on a dense subset of the special fiber and that their restrictions
to the special fiber are birational isomorphisms.
Suppose that R' is any discrete valuation ring which is the localization
of a smooth R-scheme, and let K' be the fraction field of R'. Then the
minimal proper regular model of E over the field K' is e XR Spec(R'). We

apply this fact, taking R' to be the ring
R' = () = (the local ring of £ at a gen~ric )

E,~ point ~ of its special fiber £ .
Note that the special fiber £ need not be irreducible, so it may have several
generic points, one for each component. We can take R' to be the local
ring at anyone of these generic points. Note also that here is where we
use the fact that e is smooth over R, since this fact implies that R' is the
localization of a smooth R-scheme.
There is a natural map Spec(R') = ~ ---> e, in other words an R'-valued
point of e. We let T be the corresponding translation map,
T : Spec(R') XR e --+ Spec(R') XR e.

This translation map is an automorphism on the generic fiber, so it fol-
lows from (4.6) and the minimality of e x R Spec(R') that T is actually a
morphism.
The map T on Spec(R') XR £ is translation on the second factor by
the generic point of the first factor, so we obtain a commutative diagram
Spec(R') XR e ~ Spec(R') XR e
1 1
This proves that 1> is defined at every generic point of the special fiber
of £ XR e lying over~. But ~ is an arbitrary generic point of the special
fiber £, which implies that 1> is defined at every generic point of the special
fiber of £ x R e. A similar argument can be applied to 1/;, which proves that
the domains of definition of 1> and 1/; contain a dense subset of the special
fiber of e XR e. This verifies the first part of property (ii) in the definition
of a normal law.
Now take any point P E e(R), and let Tp : £ ---> £ be the automor-
phism of e extending the translation-by-P morphism on the generic fiber
of E (4.6). Then the map
Spec(R) XR e Pxl
--+
is precisely the map P x Tp, so it is one-to-one. It follows that the fibers of 1>
are not all positive dimensional. Therefore 1> must be generically surjective,
and hence a birational isomorphism on the closed fiber of e x R e. Similarly
for 1/;, which completes the verification that J.L is a normal law on e.
We can now apply (6.9) to deduce the existence of a group scheme
G / R, open subschemes U / R c £ and U' / ReG / R which are dense in
every fiber, and an R-isomorphism U ~ U' so that the restriction of tL to U

coincides with the group law of G restricted to U'. The isomorphism U ~
U' extends to give an R-birational map A : G -+ G, and the proof of (6.10)
will be complete if we can show that A is an isomorphism.
As described in (3.3), any point P E G(R) defines a translation mor-
phism Tp : G -+ G. The generic fiber of G is ElK, so P gives a point
in E(K). Translation-by-P on ElK is an automorphism of the generic
fiber of G. It follows from (4.6) and the minimality of G that this induces
a translation map on G, which we will also denote by Tp : G -+ G. Now
if 9 EGis any point, then we can find aPE G(R) so that Tp(g) E U'.
(Note R is strictly Henselian, so G(R) maps to a dense subset of the spe-
cial fiber of G from (6.4b).) Then we can extend the definition of A to a
neighborhood of 9 by using the fact that A = L p 0 A 0 Tp at every point
where it is defined, since the right-hand side is dearly defined at g. This
proves that A is a morphism.
On the other hand, the map A-I : G -+ G in the opposite direction is
a rational map from a scheme smooth over R to a group scheme over R.
Suppose that A-I is not a morphism. Then Weil's theorem (6.2) tells us
that there is an irreducible curve Z C G such that A-I is undefined at the
generic point Tf z of Z. But we know that A -1 is defined on U, and U is
dense in the special fiber, so Z cannot be a component of the special fiber.
It follows that Tfz is contained in the generic fiber E of G. In other words,
there is a point of E at which A-I is not defined. But on the generic fiber, A
is the identity map E -+ E. This contradiction shows that A-I is defined
everywhere on G. Therefore A is an isomorphism. and hence G is a group
scheme over R. 0
\iVe now have all of the tools needed to prove the existence of Neron
models for elliptic curves over Dedekind domains.
PROOF (of Theorem 6.1). If the ring R is strictly Henselian, then combin-
ing (6.7) and (6.10) shows that G is a N eron model for ElK. There are
two more steps needed to complete the proof of (6.1). First, we have to
descend from the strict Henselization of a discrete valuation ring down to
the ring itself. Second, we have to glue together Neron models over discrete
valuation rings to create a Neron model over a Dedekind domain.
So suppose first that R is a discrete valuation ring, and let R sh be the
strict Henselization of R (6.5). Then esh = e XR R sh will be a minimal
proper regular model for E I Rsh. To see this, we note first that proper
morphisms are stable under base extension (Hartshorne [1, II.4.8c]), so esh
e
is proper over R sh . Next, the regularity of esh follows, since is regular
and R sh is fiat and unramified over R. Finally, the minimality of esh is
a consequence of the construction of the minimal proper regular model
in terms of a regular model with all exceptional curves blown down. See
Lichtenbaum [1], Shafarevich [1], Chinburg [1], and the discussion in §7 of
this chapter.
Letting Es h = E x R Rs h , it is clear that Es h is the largest open sub-

scheme of esh which is smooth over R sh , so our previous work implies
that Es hI Rs h is a Neron model for E I Ks h . We are going to use this fact
to verify that E has the Neron mapping property over R.
Let XI R be a smooth R-scheme with generic fiber XI K, and let ¢K :
X jK --> EjK be a rational map defined over K. Consider the extension
of X and ¢K to R sh , say xs h = X XR R sh and ¢"J( : X/~Sh --> EjKsh. The
scheme xs h is smooth over R sh from Hartshorne [1, III.10.1b] or Altman-
Kleiman [1, VII. I. 7], so the Neron mapping property for Es h tells us that ¢"J(
extends to a unique morphism ¢W : X/~Sh --> E/~Sh. This gives us the
commutative diagram
xs h
1 <PK
1
X ---~ E,
where the top row is obtained from the bottom row using the base extension
Spec R sh --> Spec R. The strict Henselization Rs h is faithfully flat over R
(Bosch-Liitkebohmert-Raynaud [1, 2.4, corollary 9]), so this is exactly the
situation in which we can apply faithfully flat descent (see, e.g., Bosch-
Liitkebohmert-Raynaud [1, Chap. 6] or Milne [4, I §2]) to conclude that
the rational map on the bottom row is a morphism. Therefore EI R has the
Neron mapping property, which concludes the proof that EI R is a Neron
model for ElK in the case that R is a discrete valuation ring.
Finally, suppose that R is a Dedekind domain. From what we have
already done, we know that for each prime P E Spec R, the localization
E XR Rp is a Neron model for E over Kp. Further, (6.3) tells us that if we
fix a Weierstrass equation WI R for ElK, and if we let S c Spec R be the
set of primes for which W has bad reduction, then W XR Rs (i.e., the part
of W lying over Rs) is a Neron model for E over Rs. This gives the Neron
model over a dense open subset of Spec R, and gluing this large piece to
the finitely many bad fibers produces a Neron model over all of Spec R.
This completes the proof of (6.1). 0
§7. Intersection Theory, Minimal Models, and Blowing-Up
In Chapter III we saw amply demonstrated the power of intersection theory

as a tool for studying the geometry of surfaces. In this section we will
describe, without proof, the analogous theory on arithmetic surfaces due
to Lichtenbaum [1] and Shafarevich [1]. Unfortunately, the fact that an
arithmetic surface is not complete means that it is not possible to define
an intersection theory on the full divisor group, but we will be able to
compute intersections with divisors which lie on the special fiber. In the
next section we will use intersection theory to completely describe all of
the possible special fibers for a minimal proper regular model of an elliptic
curve. We will work over a discrete valuation ring, rather than a Dedekind
domain, since everything we do in this section can be done fiber-by-fiber.
Let R be a discrete valuation ring with maximal ideal p and residue
field k = RIp, and let el R be an arithmetic surface over R. The scheme e
is normal by definition, so there is a good theory of Wei! divisors on e
as described in Hartshorne [1, II §6]. An irreducible divisor r on e is a
closed integral subscheme of dimension one, in other words a curve, and
the divisor group Div(e) of e is the free abelian group generated by the
irreducible divisors. Further, each non-zero function J E K(e) defines a
principal divisor
div(f) = L ordr(f)r E Div(e),
r
and as usual we say that two divisors are linearly equivalent if the differ-
ence is principal. Here ord r is the normalized valuation associated to the
irreducible divisor r; see (4.1.2).
Let r E Div(e) be an irreducible divisor and let x E e p be a point on
the special fiber of e. Informally, a uniJormizer Jar r at x is a function
which vanishes to order 1 along r and has no other zeros or poles in a
neighborhood of x. More precisely, a uniformizer for r at x is a function J E
(')e,x in the local ring of e at x with the property that
ordr(f) = 1, and ordr,(f) = 0 for all irreducible r l -=1= r with x E r/.

To see that such a function exists, we need merely note that if x E r,
then (')e,r is a discrete valuation ring containing'the integrally closed local
ring (')e,x.
Definition. Let r 1, r 2 E Div( e) be distinct irreducible divisors and let
x E e be a closed point on the special fiber e p of e. Choose uniformiz-
ers It, h E (')e,x for r 1, r 2 respectively. The (local) intersection index
oj r 1 and r 2 at x is the quantity
Notice that this is the same definition that we gave in (III §7) for
the local intersection index on geometric surfaces. Just as in Chapter III,
we would like to add up these local intersection indices to get a global
theory. Further, we would like our intersection theory to have the important
functorial properties described in (III.7.2). In particular, linearly equivalent
divisors should give the same intersection index. Unfortunately, it is not
possible to define an intersection theory on Div(e) with this property. The
problem is that e is not complete. This is true even if e is proper over R,
which will ensure that the fibers are complete, because the base Spec(R)
itself is not complete. This non-completeness means that it is possible to
use a linear equivalence to move an intersection point "out to infinity,"
where it then disappears. The following simple example illustrates this
difficulty.
Example 7.1. Let e = IP'k = Proj R[X, Y], and consider the two divisors
r1 = {X = O} and
Here 7r E R is a uniformizer for the maximal ideal p of R, and n ::::: 1 is an

integer. These two divisors intersect at the point
x = {X = 7r = O} E ep = lP'i
on the special fiber. To compute the local intersection index we dehomog-
enize by setting Y = 1 and then compute
The divisor r 2is linearly equivalent to the divisor r3 defined by

. (X +7r y) =
r3 = r 2+ dlv Y
n
{Y = O}.
Notice that r 1 and r3 have no points in common. So the linear equiva-

lence r 2 '" r 3 has caused the intersection point of r 1and r 2to disappear.
The preceding example gives us two options. Either we can drop the
requirement that intersections be invariant under linear equivalence, or we
can restrict the allowable divisors. As we will see in the next section, it
is extremely important to be able to compute the intersection of a divisor
with itself, and to do this we need to be able to move the divisor in some
way while not changing total the intersection index. So we will adopt the
second alternative and restrict the set of divisors.
The irreducible divisors on an arithmetic surface e/ R come in two
flavors. First, there are the components of the special fiber, as described
in §4. Second, if r c e is an irreducible divisor which does not lie in the
special fiber, then the map r ~ Spec(R) will be surjective. An irreducible
divisor which is a component of the special fiber is called a fibml divisor,
and an irreducible divisor which maps onto Spec(R) is called a horizontal
divisor. For example, the image <T(Spec R) of a section <T E e(R) is a
horizontal divisor.
Definition. A divisor D = L: niri is called fibml if every component r i
of D is a fibral divisor. The group of fibml divisors on e is denoted
Divp(e) = {D E Div(e) : Dis fibral}.
Notice that Divp(e) is a subgroup of Div(e).

Definition. A divisor D = E nir i is called positive if every ni 2: 1. The

set of positive divisors clearly does not form a group.
The fibral divisors are fairly rigid, at least in the sense that their
intersections generally cannot be moved off of the special fiber by a linear
equivalence. This makes them suitable for intersection theory as described
in the following result.
Theorem 7.2. Let R be a discrete valuation ring, and let e/ R be a
regular arithmetic surface which is proper over R. There is a unique bilinear
pairing
(D,F) ~D·F,
with the following properties.
(i) Ifr E Div(e) and F E Divp(e) are distinct irreducible divisors, then
r·F= L (r·F)x.
(ii) If D I , D2 E Div(e) and F E Divp(e) are divisors with DI linearly

equivalent to D 2 , then DI . F = D2 . F. In particular,
div(f)· F = 0 for all f E K(e)* and all F E Divp(e).
The intersection pairing also has the following symmetry property.
(iii) If FI, F2 E Divp(e) are fibra1 divisors, then FI . F2 = F2 · Fl.
PROOF. Just as in the geometric case (III.7.2), the main idea is to use
linearity and the linear equivalence property (ii) to reduce the computation
of D . F to the case of distinct irreducible divisors, and then apply (i). The
principal difficulty, as always, is to show that the result is independent of
the various choices made. For details, see Lichtenbaum [1], Shafarevich [1],
Lang [6, III §§2,3], or Chinburg [2, §4]. 0
Remark 7.2.1. Let e = IPk = Proj R[X, Y], let a E R be an element

which is not a square in R, and consider the two irreducible divisors
and
Then r n F consists of either one or two points, depending on whether or
not a is a square in the residue field k. If Va r:J. k, then rand F intersect
at the one point
x = {7r = X2 - ay2 = O} E e p,
where 7r is a uniformizer for R. We compute the intersection index at x by
dehomogenizing Y = 1,
(r· F)x = dimk R[X]ex2-a)/(7r, X2 - a) = dimk k[X]/(X2 - a) = 2.
Similarly, if Va E k, then rand F intersect at the two points
and
We will leave it to the reader to check that (r· F)y = (r . F)z = 1.
We proved (III.8.2) that the intersection pairing on a fibered surface

is negative semi-definite on fibral divisors, with kernel the entire fiber. The
argument given in Chapter III carries over almost verbatim to give the
same result for arithmetic surfaces.
ideal 13, and let e/ R be a regular arithmetic surface proper over R.
(a) The special fiber ep is connected.
(b) Let F E Divp(e) be a fibral divisor. Then F2 ~ 0, and the following
three conditions are equivalent:
(i) F2 = O.
(ii) F· F' = 0 for every F' E Divp(e).
(iii) F = ae p for some a E Q, where e p = e XR 13 is the special fiber
of e with appropriate multiplicities; see §4.
PROOF. (a) This is a special case of Hartshorne [1, III.l1.3J.

(b) Clearly, (ii) implies (i). Further, the divisor e p is principal, since
it is equal to div(7r) for a uniformizer 7r E R, so (7.2ii) shows that (iii)
implies (ii). The fact that F2 ~ 0 and the remaining implication (i) =?
(iii) are proven in exactly the same way as the geometric case (III.8.2).
For further details, see Lichtenbaum [1], Shafarevich [1], or Lang [6, III
Prop.3.5J. 0
To simplify our discussion for the remainder of this section, we are

going to assume that our discrete valuation ring R has an algebraically
closed residue field k. In practice, most of what we say will remain true
with some slight modifications.
With this assumption, an irreducible fibral divisor F on e is an irre-
ducible curve defined over k. If we further assume that e is proper over R,
then F will be proper (hence projective) over k. Recall that the arithmetic
genus of such a curve F / k is defined to be
More generally, if we write the special fiber as e p = 2:. niFi, then any
positive fibral divisor F = 2:. aiFi is a one-dimensional scheme over k.
If F is connected, the arithmetic genus of F is defined by exactly the
same formula. See Hartshorne [1, exercise III.5.3J for a discussion of the
arithmetic genus. The next proposition describes the few facts that we will
need to know about the arithmetic genus.
ideal 13, fraction field K, and algebraically closed residue field k. Let e/ R
be a regular arithmetic surface proper over R.
(a) (Adjunction Formula) There is a divisor Ke E Div(e) with the property
that
F2 + Ke' F = 2Pa(F) - 2 for every FE Divp(e).
The divisor Ke is called a canonical divisor on e. (N.B. The adjunction

formula is only valid for fibml divisors.)
(b) Let G / K be the generic fiber of e. Then
where g(G) is the usual genus ofG/K [AEC, II.5.4].

(c) Let F E Divp(e) be an irreducible fibral divisor. Then Pa(F) > 0,
and Pa (F) = 0 if and only if F is isomorphic to lP'~.
PROOF. (a) The classical adjunction formula for a non-singular curve on
a non-singular surface is proven in Hartshorne [1, V.L5], and the case of
singular curves is described in Hartshorne [1, exercise V.L3]. The adjunc-
tion formula for arithmetic surfaces is due to Lichtenbaum [1, Thm. 3.2].
For further information, see also the discussion in Lang [6, remark 1 on
p.117].
(b) The first equality follows from the general fact that in a flat family, the
arithmetic genus of the fibers remains constant (Hartshorne [1, III.9.10]).
The second equality is Hartshorne [1, IV.L1], since G is non-singular.
(c) The inequality Pa(F) 2: 0 is clear, since by definition the arithmetic
genus is the dimension of a certain cohomology group. For the second
assertion, see Hartshorne [1, IV exercise LS(b)]. 0
Remark 7.4.1. The description of the canonical divisor Ke in (7.4a) is

not, of course, the usual definition. A canonical divisor is normally defined
to be the divisor of a differential form of top dimension, so in the case of
an arithmetic surface e, the divisor of a differential 2-form on e. But for
our main application (S.l) in the next section, we will only need to know
that there exists some divisor satisfying the adjunction formula (7.4a). In
fact, it would suffice to know that the map
is a homomorphism; see Lichtenbaum [1, Thm. 3.2].

Earlier (4.5) we stated the existence of a minimal proper regular model
for a curve G / K. We now want to briefly describe how such models are
constructed and give Castelnuovo's criterion for minimality. If e/ R is any
regular model for G / K, and if x E ep is a point on the special fiber of e,
then we can blow up x to get another regular model e' / R and a birational
morphism
¢:e'~e.
(See Hartshorne [1, I §4, II §7, V §3] and the discussion (7.7, 7.7.1) at the
end of this section.) The map ¢ is an isomorphism away from x, and the
inverse image of x is a divisor D = ¢-l(x) with the property that
and
Thus blowing up x has the effect of replacing x by a projective line whose

self-intersection is -1.
In general, an irreducible fibral divisor D E Divp(e) which satis-
fies D ~ lP'1 and D2 = -1 is called an exceptional divisor or an exceptional
curve. Castelnuovo showed (in the geometric setting) that such curves can
always be blown back down.
Proposition 7.5. (Castelnuovo's criterion) Let R be a discrete valua-
tion ring with fraction field K and algebraically closed residue field k, and
let C / K be a non-singular projective curve of genus 9 2'> 1.
(a) Let e' / R be a proper regular model for C / K {4.5a), and let D E
Divp(e') be an exceptional divisor on e'. Then there exists a proper regular
model e/ R for C / K and a birational morphism ¢ : e' -> e so that x = ¢(D)
is a single point and ¢ is the blow-up of e at x.
(b) Let e be a minimal proper regular model for C / K. Then e contains
no exceptional divisors.
PROOF. (a) See Hartshorne [1, V.5.7] for a proof in the geometric situation.
The arithmetic version is due to Lichtenbaum [1, Thm. 3.9] and Shafare-
vich [1, p. 102], see also Chinburg [2, Thm. 3.1].
(b) If e contains an exceptional curve, then (a) says that we can blow
it down to get a smaller regular model for C / K. But a blow-down map
is clearly not an isomorphism, which contradicts the assumed minimality
of e. 0
Remark 7.5.1. Continuing with the notation from (7.5b), it can be shown
that a proper regular model e/ R for C / K is minimal if and only if it con-
tains no exceptional divisors. Further, starting with any proper regular
model for e/ R, one can produce a minimal model by blowing down ex-
ceptional curves until none are left. The geometric case is described in
Hartshorne [1, V §§3,5]. The arithmetic case is due to Lichtenbaum [1,
Thm. 4.4] and Shafarevich [1, p. 126]; see also Chinburg [2, Thm. 1.2].
Remark 7.6. Let R be a Dedekind domain, say the ring of integers of a
number field K, and let e/ R be an arithmetic surface. Then we can define
an intersection pairing
where Divfib(e) denotes the group of divisors generated by the components

of the special fibers and the pairing is defined linearly using (7.2). Unfor-
tunately, if we want to retain the linear equivalence property (7.2ii), then
it still is not possible to extend this pairing to all of Div(e). As before,
the underlying problem is that Spec(R) is not complete, so intersection
points can move out to infinity and disappear. Arakelov [1] had the bril-
liant idea of adding in some extra fibers "at infinity." More precisely, he
adds one fiber for each archimedean absolute value of K, and then uses
tools from differential geometry to define real-valued local intersection in-

dices on these archimedean fibers. Arakelov's intersection theory extends
to the full divisor group while retaining the linear equivalence property,
and many of the most important theorems from the classical geometry of
surfaces, such as the Riemann-Roch and adjunction formulas, extend to
the Arakelov setting. For more information about Arakelov intersection
theory on arithmetic surfaces, see for example Chinburg [1], Faltings [1],
or Lang [6].
The final topic we want to discuss in this section is the blowing-up
process. This is described with varying degrees of generality in Hart-
shorne [1, I §4, II §7, V §3]. With an eye towards the explicit computations
we will be doing in §9, we offer the following brief primer on blowing-up
surface singularities.
Remark 7.7. (Blowing-Up Singularities on Arithmetic Surfaces) Let R be
a discrete valuation ring with uniformizing element 7r and residue field k.
Let e c A~ be an arithmetic surface defined by a single equation
f(x,y) =0 for some polynomial f(x, y) E R[x, y].
In other words, e = Spec R[ x, y] / (f). In order to ensure that e is a two-

dimensional scheme whose special fiber has dimension one, we will assume
that f is not a constant polynomial and that at least one coefficient of f
is a unit in R. In fancier terminology, this means that e is flat over R.
Keep in mind that e is a "surface" (i.e., a two-dimensional scheme)
sitting inside the three-dimensional scheme A~. Intuitively, the three "co-
ordinate functions" on A~ are 7r, x, and y, and in order to calculate the
special fiber we always set 7r = O.
We are going to assume that e has a singularity at the point 7r = x =
y = 0 on the special fiber. In other words, we assume that
f(O,O) =of
ox (0, 0) =of
oy (0, 0) =0 (mod 7r).
Let m = (7r, x, y) E e be the singular point on the special tiber of e.

Then the blow-up of e at m is formed by taking the following three schemes
and gluing them together as explained below.
Chart 1. Define new variables
and y = 7rYl,
and let v be the largest integer so that
In other words, factor out a power of 7r so that the coefficients of hare

in R and at least one coefficient is a unit. Then the first coordinate chart
for the blow-up of eat m is the scheme e l C A'h = SpecR[xl,YI] defined

by
Chart 2. The second chart is formed using new variables 7f', x' ,Y' defined
by
7f = 7f'y', x = x'y', y = y'.
We substitute these into the polynomial f(x, y). This means we do two
things. First, we replace x and y by x'y' and y'. Second, we take each
coefficient a of f(x, y) and replace the largest power of 7f dividing a by that
power of 7f'y'. For example, if 7f21a and 7f3 f a, then we would replace a
by (1f'y')27f-2 a. We factor out the largest possible power of y' to get
f(x'y',y') = (y't'J'(x',y') with f'(x', y') E R[7f', x', y']'
and then the second coordinate chart of the blow-up is the scheme
e' : Spec R[7f', x', y'l/ (7f - 7f' y', J' (x', y')).
Note that 7f' is a new variable, just like the variables x' and y'. The
scheme e' is the closed subscheme of A~ = Spec R[7f', x',y'] defined by the
two equations 7f = 7f'y' and J'(x',y') = o.
Chart 3. The third chart is formed similarly to the second chart using the
variables 7f", x", y" defined by
7f = 7f" x", x=x", y = y"x".

Substituting these into f(x, y) as explained above and pulling out the
largest power of x" gives
f(x", y"x") = (x")V" f"(x", y") with f"(x",y") E R[x",y"].
Then the third coordinate chart of the blow-up is the scheme
e" : SpecR[7f", x", y"]/(7f - 7f"X", f" (x" , y")).
It is easy to see how to glue the three coordinate charts together. For
example, in order to map e l to e', we just need to solve for (7f', x', y') in
terms of (7f, Xl, yd. Thus
, 7f 7f 1 , x x
7f =-=-=- X=- y' = y = 7fYI·
y' Y YI' y' y YI
These equations define a birational map e l -+ e' which is defined every-
where except at the points of e l with YI = O. Similarly, we get a birational
map e l -> e" by using the equations
7f 7f 1 " y y
x" = x = 7fXl,
y = x"
x" X x
and a birational map e' -+ e" using

linn n'
n = x" = ;; = x" x" = x = x' y',
These maps are used to glue the three coordinate charts together, and the
resulting scheme is the blow-up of e at m.
In order to find the special fiber of the blow-up, we take the special
fibers of each of the coordinate charts and then glue them together. The
special fiber of a coordinate chart is calculated by setting n = 0 and looking
at the resulting curve defined over k. The first coordinate chart is easiest,
and we find that
e
1 = Spec k[Xl' Yl]/(il(Xl, yd)·
e
In other words, 1 is the curve in A% defined by the single equation = o. A
Similarly, the special fiber of e' is obtained by setting n = 0, which
means that
e'
= Spec k[n', x', y']/(n'y', l' (x', y')).
Here n' is to be treated as a variable, so e'

consists of two pieces, one
obtained by setting n' = 0 and the other obtained by setting y' = o. Of
course, each piece may consist of several components, or a piece could be
empty. Finally, e"
is given by
e" = Spec k[n", x", y"] / (nil x", j" (x", y")),
so e" also consists of two pieces, one with nil = 0 and the other with x" = o.
Example 7.7.1. We are going to illustrate (7.7) by blowing-up the arith-
metic surface
e : x 2 + y5 = n 4
at its singular point n = x = y = O. For simplicity, we will assume that
the residue field does not have characteristic 2, 3, or 5.
To find the first coordinate chart of the blow-up, we substitute x = nXl
and y = nYl into the equation for e and cancel n 2 , which yields
The second coordinate chart is obtained by substituting n = n' y', x = x' y',
and y = y', and then canceling y,2 to obtain
r.J'
L.:X
,2 +,3 ,4,2
y =n y, n = n'y'.
Finally, to get the third chart we substitute n = nil x", x = x", and y =
y" x" and cancel X" 2 to get
II
r.J
L.: 1 + x 11 3 y 115 =n114 x 112 , n = nil x" .
Next we compute the special fibers by setting 1r = o. The special fiber

of e1 is
e- 1 : Xl2 = 0,
so e is a non-singular rational curve appearing with multiplicity 2.
1
To find the special fiber of e', we must set 1r = 1r'y' = O. This gives us
two pieces, one with 1r' = 0 and one with y' = O. In this way we find two
fibral components, which we will denote by F{ and F~:
e'= { F'·
1·
1r' - 0
-,
x,2 + y,3 -
- 0
,
2F~ : y' = 0, X,2 = o.
Keep in mind that F{ and F~ are curves in A~ = Spec k[1r', x', y']. Thus F{
is a rational curve with a cusp, whereas F~ is a non-singular rational curve
which appears with multiplicity 2 in the fiber. In other words, as a divisor
we have e'
= F{ + 2F~.
Similarly, the special fiber of elf is obtained by setting 1r = 1r"X" = o.
However, when we set x" = 0 we obtain the equation 1 = 0, so x" = 0 does
not give any components of e".
Hence elf consists of the single rational
curve
e" : 1r" = 0, 1 = x,,3 y1l5.
We claim that when we glue e1 , e', and elf together, the special
fiber e" e
is identified with F{, and the special fiber 1 is identified with 2F~.
To verify the first statement, we observe that the special fiber elf is defined
by the equations 1r" = 0 and 1 = X,,3 y ,,5. According to (7.7), the gluing
map e' -+ e" is given by the substitutions
1r" = 1r'lx', x" = x'y', y" = l/x'.

Substituting these into the equations for elf yields
1r'lx' = 0 and
which are exactly the equations of F{. We leave it to the reader to verify
e
the assertion that 1 is glued to 2F~.
Thus the special fiber of e' contains all of the components of the special
fiber of the blow-up. If e' were regular, we would have completed our
construction of a regular model, but unfortunately it is not regular. In
fact, every point on the component F~ of the special fiber is a singular
point, so next we want to blow up e' along the entire curve y' = o. We
didn't discuss blowing up a surface along a curve in (7.7), but the procedure
is very similar.
Recall that e' c 1\1
= Spec R[1r', x', y'] is given by the equations
e' :x ,2 + y,3 =1r,4,2

Y , 1r = 1r'y'.
The Special Fiber S of the Scheme S : X 2 + Y = T 4, 7r = TY

Figure 4.3
To blow up e' along the curve y' = 0, we make the substitutions

x' = XY, y' = Y, 7r' = T,
and cancel y2 from the first equation. This yields the arithmetic sur-
face S c A'k = Spec R[T, X, YJ given by the equations
S : X2 + Y = T 4 , 7r = TY.
Note that S is regular at X = Y = T = 7r = 0, since the equations
for S show that the maximal ideal at that point is generated by the two
variables X and T. One can similarly verify that S is regular at all other
points, so it is a regular model for e.
To compute the special fiber S of S, we set 7r = TY = 0. The part
with T = ° is the non-singular rational curve X 2 + Y = 0, whereas the
part with Y = 0 factors as
so it consists of two non-singular rational curves which intersect tangen-

tially. All three of the components of S intersect at the point X = Y =
T = 0, so the special fiber S looks ~ illustrated in Figure 4.3.
Label the three components of S as indicated in Figure 4.3,
Fl : Y = X - T2 = 0, F2 : Y = X + T2 = 0, F3: T = X2 + Y = o.
Looking at Figure 4.3 or directly from the equations for the components,
we can compute the pairwise intersections
Next, using the fact (7.3) that the intersection of a component with the
entire fiber S = Fl + F2 + F3 is zero, we compute the self-intersections,
F'f = -(Fl' F2 + Fl . F3) = -3,

Fi = -(F2 . Fl + F2 . F3) = -3,
F~ = -(F3 . Fl + F3 . F2) = -2.
This shows that F I , F 2 , and F3 are not exceptional curves, so S is a minimal

regular model.
We have now computed the incidence matrix of the special fiber,
The incidence matrix of the special fiber of a minimal proper regular model
of a curve can be used to compute the group of components of the Neron
model of its Jacobian variety, see Raynaud [1] and exercises 4.32 and 4.33.
In this example, the 2 x 2 minors of the incidence matrix have determi-
nant 5. Raynaud's theorem then implies that the group of components of
the Neron model of the Jacobian variety is a cyclic group of order 5.
§8. The Special Fiber of a Neron Model
In this section we are going to describe the Kodaira-Neron classification

of special fibers on minimal proper regular models of elliptic curves. Our
main tool will be the intersection theory described in the previous section.
We will work over a discrete valuation ring with algebraically closed residue
field. In the next section we will give an algorithm of Tate which computes
the special fiber and also provides some additional information, including
a description of what happens when the residue field is not algebraically
closed.
We begin with a proposition which describes the intersection properties
of the components of the special fiber. The most important part of this
proposition is the last formula in (d), since it is this formula which puts
severe constraints on the possible configurations of the components.
ideal p, fraction field K, and algebraically closed residue field k. Let ElK
be an elliptic curve, and let el R be a minimal proper regular model
for ElK. Suppose that the special fiber of e contains r irreducible compo-
nents, say F I , ... , Fr , and write the special fiber as
r
ep = LniFi'
i=I
(a) At least one of the ni 's is equal to 1.

(b) Let Ke be a canonical divisor on e (7.4a). Then
Ke . F =0 for all fibral divisors F E Divp(e).

(c) Ifr = 1, then F'.f = 0 and Pa(FI ) = l.

(d) Suppose that r :::: 2. Then for each 1 ::; i ::; r,
Fl = -2, Fi ~ lP'~, and L njFj · Fi = 2ni.

l:5,j:5,r,#i
PROOF. The scheme e is proper over R, so e(R) ~ E(K) from (4.4a). By

definition, an elliptic curve E / K has at least one K-rational point, namely
its identity element, so we can find a point P E e(R). Let P(p) E ep be
the image of P on the special fiber of e, and let Fi be a component of ep
containing P(p). The scheme e is regular, so (4.3b) says that P(p) is a
non-singular point of ep . It follows that ni = 1, since if ni :::: 2, then every
point of Fi would be a singular point of ep • This completes the proof of (a).
Next we consider the special fiber ep as a divisor on e. It has the
following three properties:
e~ = 0 from (7.3b)
Pa(e p ) = g(E) = 1 from (7.4b)
e~ + Ke· ep = 2Pa(e p ) - 2 adjunction formula (7.4a).
Substituting the first two equations into the third gives
Ke· ep = o.
We next apply the adjunction formula (7.4a) to an irreducible fibral
component Fi to get
The arithmetic genus of an irreducible divisor is non-negative (7.4c), since

by definition it is the dimension of a certain cohomology group. Thus the
right-hand side of the adjunction formula is at least -2. On the other hand,
we know that Fl ::; 0 from (7.3b). This leads to the following possibilities:
(i) Fl = 0 for some 1 ::; i ::; r.

(ii) Fl < 0 and Ke· Fi < 0 for some 1 ::; i ::; r.
(iii) Fl < 0 and Ke· Fi :::: 0 for all 1 ::; i ::; r.
In case (i) we know from (7.3b) that Fi must be a multiple of the entire
special fiber. More precisely, we must have r = 1 and ep = nIFI . Then (a)
tells us that nl = 1, so FI = ep . Now the equality Ke· ep = 0 proven
above gives us (b), and the facts e~ = 0 and Pa(e p ) = 1 noted above give
us (c). This completes the proof of Proposition 8.1 in case (i).
Next consider case (ii). Each term in the left-hand side of the adjunc-
tion formula is negative, whereas the right-hand side is at least -2. The
only way for this to happen is if
and
Now (7.4c) says that Fi ~ lP'L and hence Fi is an exceptional divisor. But
Castelnuovo's criterion (7.5b) says that a minimal proper regular model
contains no exceptional divisors, so case (ii) cannot occur.
It remains to consider case (iii). The strict inequality F? < 0 implies
in particular that r ?: 2. We take the equality Ke' ep = 0 proven above
and write it out in terms of the fibral components as
r
LniKe' Fi = O.
i=l
Each ni ?: 1, and since we are in case (iii), each Ke' Fi ?: 0, so the only
way that this can be true is if we have
Ke' Fi =0 for all 1 ::::; i ::::; r.
This proves (b).

Next we substitute Ke . Fi = 0 into the adjunction formula for F i ,
which yields
F? = 2Pa(Fi ) - 2.
We are in case (iii), so F? < 0, whereas Pa(Fi) ?: 0 from (7.4c). It follows
that
and
and then (7.4c) tells us that Fi ~ lP'l;,. Finally, we note that ep . Fi
Fi . ep = 0 from (7.3b), which allows us to compute
r
o = ep . Fi = L njFj . Fi = -2ni + L njFj . Fi ·

j=l l<::;j<::;r,#i
This completes the proof of (d). o
We are now going to use (8.1) and a combinatorial argument to give

the Kodaira-Neron classification of the fibers of minimal proper regular
model of elliptic curves.
Theorem 8.2. (Kodaira [1], Neron [1]) Let R be a discrete valuation
ring with maximal ideal p, fraction field K, and algebraically closed residue
field k. Let ElK be an elliptic curve, and let el R be a minimal proper
regular model for ElK. Then the special fiber ep of e has one of the
following forms. (See Figure 4.4. Note that the small numbers in Figure 4.4
indicate the multiplicities of the components.)
Type 10. ep is a non-singular curve of genus l.
Type Ii. ep is a rational curve with a node.
Type In. ep consists of n non-singular rational curves arranged in the
shape of an n-gon, where n 2: 2.
Type II. ep is a rational curve with a cusp.
Type III. ep consists of two non-singular rational curves which intersect
tangentially at a single point.
Type IV. ep consists of three non-singular rational curves intersecting at
a single point.
Type 10. Cp is a non-singular rational curve of multiplicity 2 with four
non-singular rational curves of multiplicity 1 attached.
Type I~. Cp consists of a chain of n + 1 non-singular rational curves of
multiplicity 2, with two non-singular rational curves of multi-
plicity 1 attached at either end.
Type IV*. Cp consists of seven non-singular rational curves arranged as
pictured in Figure 4.4.
Type III*. Cp consists of eight non-singular rational curves arranged as
Type II*. Cp consists of nine non-singular rational curves arranged as
Remark 8.2.1. If the residue field k of R is not algebraically closed,

then Cp may have some components which are irreducible over k but be-
come reducible over a finite extension of k. In other words, the Galois
group G k / k may act non-trivially on the k-irreducible components of Cp ,
and then the k-irreducible components of Cp are the orbits. We will discuss
this situation further in the next section when we describe Tate's algorithm.
Remark 8.2.2. The dual graphs of the pictures in Figure 4.4 turn out
to be extended Dynkin diagrams. There is a discussion of this in Mi-
randa [1, I §6]' as well as a proof of (8.2) based on the negative semi-definite
quadratic forms attached to the extended Dynkin diagrams. The proof re-
lies only on the facts proven in (7.3), namely that Cp is connected and that
the intersection pairing on Cp is negative semi-definite with kernel equal to
the entire fiber.
Remark 8.2.3. \Ve will use Kodaira's [1] notation In, II, ... , III*, II*
to describe the various types of special fibers (8.1). There is a second
notational system, due to Neron [1], which is also in common use. For the
convenience of the reader, we briefly list the equivalences.
Kodaira 10 In II III IV 1*a 1*n IV* III * II*

Neron a bn c1 c2 c3 c4 c5 n c6 c7 c8
Reduction Number of Configuration

Type Components (with multiplicity)
10 1 C :=:>1
II 1
0<1
In n
01 1
I
I
II
III
1
2
K
-<
*
IV 3
10 5 111 I 111 I
lU----U
2
1*n n+5
1 1
2 2
IV* 7
21 21' 21 3
4
111* 8
1 I 31 21 13 11
2 2
4 2
II* 9
1 I 31 51 31 41
2 6
The Kodaira-Neron Classification of Special Fibers

Figure 4.4
Remark 8.2.4. Ogg [3J and Namikawa and Ueno [IJ have given a classi-
fication, similar to (8.2), for the special fibers of proper regular minimal
models of curves of genus 2. It turns out that there are more than 100
configurations!
PROOF (of Theorem 8.2). We will write the special fiber of ep as usual as
r
ep = LniFi.
i=1
We are going to consider a number of cases, which we will box for clarity.
Ie p has r = 1 component I
Proposition 8.1(a,c) tells us that nl = 1 and Pa(e p ) = 1, so ep = Fl is an
irreducible curve of arithmetic genus 1. If ep is non-singular, then it is a
non-singular curve of genus 1 (Hartshorne [1, IV.l.l]), so we have Type 10 .
If ep is singular, then Hartshorne [1, V.3.7] and the fact that Pa(e p ) = 1
means that a single blow-up of a singular point on ep will produce a non-
singular rational curve. Hence ep is a rational curve with exactly one
singular point of multiplicity 2, from which it is not hard to show that the
singular point is either an ordinary node or an ordinary cusp. This gives
Types II and II.
(Alternative proof for r = 1. We will see in the next section that
if r = 1, then the scheme W / R defined by a minimal Weierstrass equation
for E is already a regular scheme. It will follow that e = w, so the
e
special fiber p is obtained by reducing the minimal Weierstrass equation
modulo p. But we already know from [AEC, VII §5] that the reduction Wp
is either a non-singular curve of genus 1, a rational curve with a node, or
a rational curve with a cusp.)
We assume henceforth that r ~ 2, which means that we can apply the
formula given in (8.1d),
L njFj . Fi = 2ni·
l~j~r,j#i
Note that each nj ~ 1, so every term in the sum is non-negative. We

will be making frequent use of this important formula (*). We also note
from (8.1d) that every component is a non-singular rational curve; that is,
Fi ~ Pi, for every 1 :::; i :::; r.
Proposition 8.1(a) says that one of the ni's equals 1, so relabeling
the Fi'S if necessary, we may assume that nl = 1. We further know that ep
is connected (7.3a), and r ~ 2 by assumption, so after further relabeling
we may also assume that Fl . F2 ~ 1.
Iep has r = 2 components I
This means that ep = Fl + n2F2, so applying (*) for i = 1 and i = 2 gives
and
Further. FI . F2 = F2 . FI from (7.2iii), so we deduce that
and
This means either that FI and F2 intersect tangentially in a single point,

which gives Type III, or else they intersect transversally at two distinct
points, which gives Type 12 . This completes the analysis of the special
fiber in the case that ep has exactly two components.
Iep has I' 2 3 components I
vVe claim in this case that intersecting components always intersect trans-
versally; that is, we claim that
Fi ·Fi,:::; 1 for all 1 :::; i,i':::; T, i i= i'.

To see this, we use the fact that ep is connected and contains at least three
components to find a third component, say F k , so that Fk intersects at
least one of Fi or Fi" say Fk . Fi 2 1. Applying (*) to Fi and to Fi' gives
the inequalities
vVe now multiply these two estimates to obtain the strict inequality
Therefore Fi . Fi' < 2, and since the intersection index is an integer, we

find that Fi . Fi' :::; 1 as desired.
In particular, we have FI . F2 = 1. Using this and the fact that n I = 1,
we can apply (*) to FI to obtain the bound
Thus n2 equals either 1 or 2, which leads to two further subcases.

IT 23, n2 = 11
Applying (*) to F2 gives
r r
n l F1 . F2 +L njFj · F2 = 2n2, and hence L njFj . F2 = 1,

j=3 j=3
since FI ·F2 = 1 and nl = n2 = 1. This means that there is exactly one more
component intersecting F 2 , call it F 3 , and n;, = 1. If F3 also intersects F I ,
then we get two possible configurations depending on whether or not F1 nF3
is the same point as F2 n Fc,. If they are the same point, then we get a fiber
of Type IV, and if they are not, then we get three rational curves arranged
in a triangle, which is Type 13 .
Suppose now that F3 does not intersect Fl. Then applying (*) to F3
gives
+L L
T T
n2 F2 . F3 njFj . F3 = 2n3, and hence njFj . F3 = 1,

j=4 j=4
since F2 . F3 = 1 and n2 = n3 = 1. Therefore there is exactly one more
component intersecting F 3 , call it F 4 , and n4 = 1. If F4 also intersects F l ,
then we have four rational curves arranged in a square, which is Type I 4 .
If F4 does not intersect H, then applying (*) to F4 gives in the same way
one more component F5 intersecting F 4 , and n5 = 1. The fiber ep has only
finitely many components, so this process must eventually terminate. More
precisely, since ep has r components, the process will terminate with FT
intersecting Fl. At this point we will have r rational curves, each of multi-
plicity 1, arranged in the shape of a polygon, which means that the fiber ep
is of Type IT'
Ir:::: 3, n2 = 21
Applying (*) to Fl gives
T T
n2 F2' Fl +L njFj . Fl = 2nl, and hence L njFj . Fl = 0,

j=3 j=3
since Fl . F2 = 1, nl = 1, and n2 = 2. This means that there are no more
components intersecting Fl. Next applying (*) to F2 gives
+L L
T T
nlFl . F2 njFj . F2 = 2n2, and hence njFj · F2 = 3.

j=3 j=3
Thus F2 intersects either one, two, or three additional components.
Suppose first that F2 intersects three additional components, which
we label F3, F 4 , and F 5. Then n3 = n4 = n5 = 1, which gives a fiber of
Type I~, and applying (*) to F 3 , F 4 , and F5 shows that ep contains no
other components.
Next suppose that F2 intersects exactly two other components, say F3
and F 4 . Switching these two components if necessary, we have n3 = 1
and n4 = 2. Thus ep contains the configuration illustrated in Figure 4.5(a),
where the small 2's next to F2 and F4 indicate that they are components
of multiplicity 2. Of course, ep may contain some additional components.
Let s :::: 4 be the largest integer so that ep contains the configuration
illustrated in Figure 4.5(b). Applying (*) to Fi for any 4 :S i < s gives
L L
T T
ni-1Fi-l·Fi+nH1Fi+l·Fi+ njFj"Fi = 2ni, so njFj"Fi = 0,

j=8+l j=8+l
(b)
Building a Fiber of Type I;,
Figure 4.5
since F i- 1 . Pi = P i +1 . Fi = 1 and ni-1 = ni = nH1 = 2. Thus there are

no more components intersecting any of F 41 ... ,P,-l. On the other hand,
applying (*) to p, gives
L L
T T
n 8-1P,-1 . p, + nJFj · p, = 2n,,, and hence nj F J . F8 = 2.

j=-,+l j=8+1
If Fs intersects exactly one additional component, then that compo-

nent will have multiplicity 2, which means that Cp contains the config-
uration in Figure 4.5(b) with one more multiplicity-2 component. This
contradicts our choice of s, so F2 must intersect two additional compo-
ncnts, each of which has multiplicity 1. This gives us a fiber of Type I~
(with n = s - 3), and it is thcn easy to check using (*) that there are no
more components. This completes the proof in the case that F2 intersects
exactly two components in addition to Fl.
Finally we suppose that F2 intersects exactly one additional compo-
nent, say F';J, with multiplicity n3 = 3. This means that Cp contains the
configuration illustrated in Figure 4.6(a). Let t ~ 3 be the largest integer
so that Cp contains the configuration illustrated in Figure 4.6(h). Apply-
ing (*) to Fi for any 3 ::; i < t gives
L L
T
ni-1Fi-1·Fi+nH1Fi+1·Fi+ njFj"Pi = 2ni, so njFj·Fi = 0,

j=t+1 j=t+1
since
F i - 1 . Fi = FH1 . Fi = 1, ni-1 = i-I, ni = i, and nH1 = i + 1.

Thus there are no more components intersecting any of F 3 , ... ,Ft - 1. On
the other hand, applying (*) to F t gives
nt-1Ft-1 . F t + L njFj · F L =:= 2nt, and hence L njFj · F t = t + 1.

J=t+1 j=t+1
§S. The Special Fiber of a Neron Model 359
(b)
~
Fi I F3 3
2 ' ...
.... " F
' .... , t-l t-l
-
Ft F 1+1
t+I2
F
t+22
1+1
-+_ __
F2 Ft+3 Ft+4
(C)
F 1+2
t+22
(d)
Building Fibers of Type IV*, III*, II*
Figure 4.6
If F t were to intersect exactly one more component, then that component

would have multiplicity t + 1, contradicting the fact that we chose the
largest t so that Cp contains the configuration in Figure 4.6(b). Thus F t
intersects at least two additional components, say F t + 1 and F t + 2 .
Let Fi be any component intersecting Ft. Applying (*) to F; gives
ntFt . Fi :s: 2ni, and hence ni:2: t/2.
It follows that F t intersects only the two additional components F t + 1

and Ft+2' and we have the estimates
nt+l + nH2 = t + 1, nt+2 :2: t/2.

Switching Ft+l and Ft+2 if necessary, we may assume that nHl :s: nt+2.
Then there are only two possibilities, depending on the parity of t.
It == 1 (mod 2)1
In this case we must have
t+l
nt+l = nt+2 = -2-'
Applying (*) to Ft+l then gives

r r
ntFt ·FH1 + 2..: njFj ·FH1 = 2nHl, and hence 2..: njFj ·Ft+1 = 1,
j=t+3 j=t+3
since F t . Ft+l = 1, nt = t, and nHl = (t + 1)/2. Therefore Ft+l intersects

exactly one more component, say F t +3 , with multiplicity nH3 = 1. Ap-
plying the exact same argument to FH2 gives the same conclusion, so ep
contains the configuration illustrated in Figure 4.6(c).
Finally, applying (*) to F t +3 gives
r
nt+l F t+l· F t+3+ 2..: n j Fj"Ft+3 = 2nt+3, so

j=t+5
since Ft+l . F H3 = 1, nt+l = (t + 1)/2, and nt+3 = 1. The sum is non-

negative, and t :::: 3 by assumption, so we must have t = 3. Therefore ep
looks like Figure 4.6( c) with t = 3, which is precisely Type IV*. This
completes the proof when t is odd.
It == 0 (mod 2) I
In this case we must have
t+2
and nt+2 = -2-'
Applying (*) to Ft+l gives

r r
ntFt ·Ft+l + 2..: njFj ·FH1 = 2nt+l, and hence 2..: njFj ·Ft+l = 0,
j=t+3 j=H3
since F t . Ft+l = 1, nt = t, and nt+l = t/2. Thus there are no additional

components intersecting Ft+ 1.
Next we apply (*) to F t +2 . This gives
r r
nt Ft· FH2+ 2..: njFj·Ft+2=2nt+2, and hence 2..: n j F j ·FH2 =2,

j=t+3 j=H3
since F t . Ft+2 = 1, nt = t, and nt+2 = (t + 2)/2. Hence Ft+2 intersects

at least one additional component, say F t +3 , whose multiplicity nt+3 is
either 1 or 2. So we now know that ep contains the configuration illustrated
in Figure 4.6(d).
Don't despair, we're almost done! Applying (*) to F H3 gives
r
nt+2 Ft+2 . Ft+3 + 2..: njFj · FH3 = 2nt+3,

j=t+4
and hence
= 1,
.L njFj · Ft+3 = 2nt+3 -
r
]=t+4
t+2
-2- =
{(2-t)/2
(6 - t)/2
if nt+3
if nt+3 = 2,
since Ft+2 . Ft+3 = 1 and nt+2 = (t + 2)/2. But the sum is non-negative,
and t is even and 2': 3 by assumption, so we must have
t = 4 or 6 and nt+3 = 2.
If t = 6, then there are no additional components, and Figure 4.6(d) is
exactly Type II*. Finally, if t = 4, then there is one more component FtH
hitting F t +3 , and its multiplicity is ntH = 1. This gives Type III*, which
completes the proof of Theorem 8.2. 0
§9. Tate's Algorithm to Compute the Special Fiber
In this section we are going to describe an algorithm of Tate which com-

putes, among other things, the reduction type of an elliptic curve given by
a Weierstrass equation. We set the following notation, which will be used
throughout this section.
R a discrete valuation ring with maximal ideal p, uniformizing
element Jr, fraction field K, perfect residue field k of character-
istic p, and normalized valuation v.
ElK an elliptic curve given by a Weierstrass equation
E : y2 + alxy + a3Y = X3 + a2x2 + a4X + a6.
elR a minimal proper regular model of E over R (4.5b).
clR the largest subscheme of eI R which is smooth over R (6.1.1).
Note that clR is a Neron model for ElK (6.1).
= e XR k, the special fiber of e.
= c x R k, the special fiber of c. It is a group variety over k.
the identity component of c (6.1.2); that is, cO is the open
subset of c obtained by discarding the non-identity components
of the special fiber. It is a subgroup scheme of G (exercise 4.25).
the identity component of the group variety elk (1.5c).
Tate's algorithm is essentially a set of instructions for computing e
and c from a given Weierstrass equation. For this reason its statement
as a formal theorem has the unsatisfying form: "The following 11 step
procedure leads to the stated results." So before we describe the algorithm
itself, we want to give two corollaries. This will serve to explain (if not
to excuse) why the following results are called "corollaries," when in fact
they are really conclusions which can be deduced from the description and
validity of Tate's algorithm.
Corollary 9.1. Take a minimal Weierstrass equation for E j K, and

let W C lP'h be the closed subscheme defined by this equation. Fur-
ther, let WO j R be the largest subscheme of W which is smooth over R.
Then WO ~ co; that is, WO is the identity component of a Neron model
for Ej K.
Corollary 9.2. Take a minimal Weierstrass equation for E j K. We recall

from [AEC, VII §2] the following notation:
Ejk the reduction of the given Weierstrass equation modulo p.
Ens(k) the set of non-singular points of E(k).
Eo(K) = {p E E(K) : P E E(k)ns}, the set of points of E(K) with
non-singular reduction.
El(K) = {p E E(K) : P = 6}, the set of points of E(K) which
reduce to the identity element.
Further, let
c1(R) = {a E c(R) : a(p) = 6 E £(k)}.
The isomorphism E(K) ~ c(R) described in (5.1.3) induces the following
identifications:
(a)
E(K) ::) Eo(K) ::) El(K)
II
c(R) ::) cO(R) ::) c1(R).
(b)
E(K)jEo(K) ~
c(R)jcO(R) '---+ £(k)j£O(k)
r
E(K)j El(K) ~
r
c(R)jc1(R) '---+
r
£(k).
If K is complete, or even merely Henselian, then both inclusions in (b) are

isomorphisms.
(c) Ens(k) = £O(k).
(d) The group E(K)jEo(K) is finite. More precisely, if E has split multi-
plicative reduction, then E(K)j Eo(K) is a cyclic group of order -v(j(E));
otherwise, E(K)j Eo(K) has order 1, 2, 3, or 4.
Remark 9.2.1. Note that Corollary 9.2(d) is exactly [AEC, VII.5.1], a
result which was left unproven in [AEC]. The fact that E(K)j Eo(K) is
finite, even if the residue field k is infinite, played an important role in the
proof of the criterion of Neron-Ogg-Shafarevich [AEC, VII.7.1], and we will
use it again in the next section (10.2) when we prove a generalization of
this criterion.
Remark 9.2.2. Corollary 9.2 can be used to bound the torsion subgroup
of an elliptic curve defined over a local field or a number field. Con-
tinuing with the notation set at the beginning of this section, we recall
from [AEC, VII.3.1(a)] that the subgroup EI(K) contains no prime-to-p
torsion, since it is isomorphic to the formal group of E. Suppose now
that E has additive reduction. Then £O(k) = Ens(k) = k+ is a p-group
from (9.2c), whereas the quotient E(K)jEo(K) ~ £(k)j£O(k) is a group of
order 1, 2, 3, or 4 from (9.2d). Using these facts and the exact sequences
o ~ EI (K) ~ Eo(K) ~ £O(k),

o ~ Eo(K) ~ E(K) ~ E(K)jEo(K) ~ 0,
we conclude that if E j K has additive reduction, then E(K)tors has or-

der ape for some a E {I, 2,3, 4} and some e ?: O.
Now consider an elliptic curve E defined over a number field K. The
torsion subgroup of E(K) injects into the torsion subgroup of E(Kp) for
each completion of K, so the local estimate we just proved can often be used
to obtain strong global estimates. For example, suppose that E has additive
reduction at primes PI, P2 of K with distinct residue characteristics PI, P2·
Then E(K)tors has order dividing 12, and if PI ?: 5, then E(Khors has
order at most 4.
Tate's algorithm, which we are now going to describe, computes the
following quantities associated to the elliptic curve E j K:
Type e
the reduction type of the special fiber over the algebraic clo-
sure k of k. We will use the Kodaira symbols (8.2) to describe
the reduction type.
m( E j K) the number of components, defin,=:d over k and counted without
multiplicity, on the special fiber e.
v('D E/ K) the valuation of the minimal discriminant of E j K.
f(E j K) the exponent of the conductor of E j K. This quantity will be
defined in §1O, but for now we note that it can be computed
using Ogg's formula (11.1),
f(EjK) = V('D E/ K ) - m(EjK) + 1.
c(EjK) the order of the group of components £(k)j£O(k). Equiva-

lently, c( E j K) is the number of components of the special
e
fiber which have multiplicity 1 and are defined over k.
To ease notation, we will sometimes write
m = m(EjK), f = f(EjK), c = c(EjK),

when the curve E and field K are clear from the context.
Remark 9.3. If K is a complete local field, then (9.2b) says that c(E/ K)
equals the order of E(K)/ Eo(K). This quantity is the so-called Birch-
Swinnerton-Dyer "fudge factor." It is a sort of p-adic period and appears
in the conjectural formula for the leading coefficient of L( E, s) around
s = 1 [AEC, C.16.5]. See Tate [2, §5] for details.
Tate's Algorithm 9.4. (Tate [2]) The following algorithm computes the
reduction type, the values ofm(E/K), v('D E / X ), !(E/K), and c(E/K),
and the various other quantities described during the course of the algo-
rithm.
Remark 9.4.1. At the conclusion of Tate's algorithm, one obtains a min-
imal Weierstrass equation for the given elliptic curve. In practice, however,
it is considerably easier to implement Tate's algorithm if one knows, a pri-
ori, that the initial Weierstrass equation is minimal. Further, if one only
wants a minimal equation and is not interested in computing other quan-
tities, such as the reduction type, then there are easier methods available.
For example, if the characteristic p of k satisfies p 2': 5, then a given Weier-
strass equation is minimal if and only if either V(C4) < 4 or V(C6) < 6.
In general, one can use a short algorithm of Laska [1] to find a minimal
Weierstrass equation. See also exercise 4.36 for another method.
Remark 9.4.2. In the case that the residue field k is algebraically closed,
we have assembled information about the various reduction types in Ta-
ble 4.1. This table is taken, with minor modifications, from Tate [2, §6].
Notice that if char(k) i= 2,3, then everything about E (reduction type,
exponent of conductor, group of components E(K)/Eo(K)) can be read
off from Table 4.1 once one has a minimal Weierstrass equation for E / K.
Our description of Tate's algorithm follows very closely Tate's exposi-
tion [2]. The idea is to begin with an arbitrary Weierstrass equation
for E / K and manipulate it to produce a minimal proper regular model e.

Once we have this model, we will be able to read off all of the information
we want. As we go along we will be making various assumptions. These
assumptions are cumulative, and will be Iboxed I for clarity. We will
delay the proofs of the various steps until after describing the complete
algorithm.
Making a change of variables, we may assume that the Weierstrass
equation has coefficients lal,a2,a3,a4,a6 E RI.
We let
b2 = ai + 4a2, b4 = ala3 + 2a4, b6= a~ + 4a6,

bs = aia6 + 4a2a6 - ala3a4 + a2a~ - a~ = (b 2 b6 - b~)/4,
A = -b~b8 - 8b~ - 27b~ + 9b 2 b4b6
CC>
~
Kodaira 10 In II III IV 1*0 I"n IV" m" II"
symbol (n 2: 1) (n 2: 1)
~
<1l
Ul-
Special fiber e >

(The numbers f,1\1
1~ 4 3 •
,I , , 5 3 I
~;:t:
indicate multi-
1 2 3 6 4' ::r
plicities) 1 1 1 1 2 2 i 8
M-
0W ~ X :~ l~ 18 IE ~ o
m = number of I
1 n 1 2 3 5 5+n 7 8 9 (1
irred. components o
* J 8
'0
IE IE I
-x- >=
M-
I
21E 21E <1l
E(K)/Eo(K) IE IE n even Il Il I M-
IE IE IE ::r
(0) (0) (0) i <1l
~ t(k)/tO(k) nlE 21l 31l 21E x 21l Il 31l 21E (f)
41E '0
115
n odd [
taCk) E(k) k" k+ k+ k+ k+ k+ k+ k+ k+ 'Tj
6'
<1l
Entries below this line only valid for char(k) = p as indicated
..,
char(k) = p p f. 2,3 pf.2 pf.3 pf.2 pf.2 pf.3 pf.2 p f. 2,3
V('D E/ K)
0 n 2 3 4 6 6+n 8 9 10
(discriminant)
f(E/K)
0 1 2 2 2 2 2 2 2 2
(conductor)
behavior of j v(j) 2: 0 v(j) =-n 3=0 3 = 1728 3=0 v(j) 2: 0 v(j) =-n 3=0 3 = 1728 3=0
~- --_. -_.- - ~
-
Table 4.1: A Table of Reduction Types <:..:>

Ol
Ql
be the usual quantities [AEC, III §1] associated to the given Weierstrass
equation.
It is not necessary to assume that the original Weierstrass equation is
minimal. When the algorithm terminates, the resulting Weierstrass equa-
tion will be minimal, so its discriminant will equal V('D E/ K ). Further, we
will see that the smooth part of the final Weierstrass equation, considered
as a scheme over R, is the identity component of e, which will prove (9.1).
Step 1. If Jr f 6., then the special fiber E/ k is an elliptic curve, and we
have
Type la, v(6.) = 0, m = 1, f = 0, c = 1.
Step 2. Assume iJr/6.i. This means that f;; has a singular point.
Make a change of variables to move the singular point to (0,0). Then
iJr/a3, a4 and aGIo If Jr f b2, then we have Type In with n = v(6.). More pre-
cisely, let k' be the splitting field over k of the polynomial T2 + a1T - a2.
Then we have
Type In' v(6.) = n ~ 1, m = n, f = 1.
Further, if k' = k, then E has split multiplicative reduction,

EO(k)~k* and c=n;
whereas if k' 1= k, then E has non-split multiplicative reduction,

if n is odd,
and c = {~ if n is even.
From now on, f;; has a cusp and 1EO(k) ~ k+ I. We are going transform
the Weierstrass equation so as to make the ai's more and more divisible
by Jr. To keep track, we introduce the convenient notation
a':.T = Jr -1'
ai.
Step 3. Assume now that iJr/b2i. If Jr2 faG, then
Type II, m = 1, f = v(6.), c = 1.
Step 4. Assume that iJr2/a6i (which implies that Jr2/b 6 and Jr2/b s ). If
Jr3 f bs , then
Type III, m = 2, f = v(6.) - 1, c = 2.
Step 5.Assume that iJr3/bs i (which implies that Jr 2/b 4 ). If Jr3 f b6, then
Type IV. Let k' be the splitting field over k of T2 + a3,lT - a6.2 = 0. Then
= k,
Type IV, m=3, f = v(6.) - 2, c = {~ if k'
if k' 1= k.
Step 6. Assume that 17r 3 1b6 1. Then we can change coordinates to get
More precisely, the boxed assumptions up to this point show that we can
factor
y2 + a1Y - a2 == (Y - a)2 (mod 7r),

y2 + a3,lY - a6,2 == (Y - (3)2 (mod 7r),
and then the substitution y' = y + ax + (37r will have the desired effect.
Having done this, we consider the factorization over k of the polynomial
To assist in explicit computations, we note that P has discriminant
If P(T) has distinct roots in k (i.e., if 7r f Disc( P)), then
Type 1(;, m = 5, f = v(.6.) - 4, c = 1 + #{a E k : P(a) = a}.
Step 7. If P(T) has one simple root and one double root in k, then
Type I~, m = n + 5, f = v(.6.) - 4 - n, C = 2 or 4.
If p oJ 2, then n = v(.6.) - 6, so m = v(.6.) - 1 and f = 2. For arbitrary p,

one can calculate the values of nand c using the following subprocedure
to Step 7.
Step 7. (Subprocedure) Translate x so that the double root of P(T) is T =
o. Then 7r 2 f a2, 7r3Ia4, and 7r4Ia6' If the polynomial y2 + a3,2Y - a6,4 has
distinct roots in k, let k' be its splitting field. Then
Type Ii, m=6, f = v(.6.) - 5, c = {~ if k' = k,

if k' oJ k.
If y2 + a3,2Y - a6,4 has a double root in k, translate y so that the root

is Y = O. Then 7r31a3 and 7r5Ia6' If the polynomial a2,lX 2 + a4,3X + a6,5
has distinct roots in k, let k' be its splitting field. Then
4 if k' = k,
Type!:}, m=7, f=v(.6.)-6, c= { 2 ifk'oJk.
If a2.1X2 + a4,3X + aG.5 haH a double root in k, translate x HO that the

root iH X = 0. Then 7r41a4 and 7r6Ia6. If the polynomial y2 + a3.3 Y - aG.G
has diHtinct roots in k, let k' be itH Hplitting field. Then
Type I;, m=8, f = v(6.) - 7, c = {i if k' = k,

if k' =1= k.
If y2 + a3.3 Y - a6.G has a double root in k, etc. Continue this proce-

dure until the quadratic polynomial which appears has distinct roOtH in k.
The proceHH will terminate because after each two steps we have will have
forced a3, a4, and a6 to each be diviHible by at leaHt one additional power
of 7r. This meanH that b4 , bu, and bs are also divisible by at least one addi-
tional power of 7r, and hence the Harne is true of 6.. But the discriminant 6.
is invariant under all of the translations involved, HO the procesH will stop.
Step 8. Suppose now that P(T) has a triple root in k. l\Iaking a trans-
lation on x, we rna' assume that the root is T = 0, which means that
7r2Ia2, 7r:3Ia4, and 7r41a6 . If the polynomial y2 + a3,2 Y - a6,4 has distinct
roots in k, let k' be its splitting field. Then
Type IV*, m,=7, f=v(6.)-6, c = {i if k' = k,

if k' =1= k.
Step 9. Suppose now that y2+a:;.2Y -a6.4 has a double root in k. Making
a translation on we may aSHume that the root is Y = 0, which means
that 7r 3 1a3 and 7r 5 1a6 . If 7r 4 t a4, then
Type 111*, m = 8, f = v(6.) -7, c = 2.
Step 10. SuppoHe that 17r4Ia41. If 7r 6 t a6, then
Type 11*, Tn = 9, f = v(6.) - 8, c = l.
Step 11. Finally, suppose that I7rDla6i. Then the original Weierstrass
equation was not minimal. The substitution (x, y) = (7r 2 :r', 7r 3 y') leads to
the equation
with coefficients in R and discriminant 6.' = 7r- 12 6.. Go back to Step 1

and begin the algorithm again with this new equation. Note that we can
only get to Step 11 a finite number of times, since each time we get here,
the discriminant of the original \VeierstraHs equation must be divisible by
an additional factor of 7r 12 . Therefore the algorithm will terminate.
This concludes our description of Tate's algorithm. We are now going

to give some indication of why the various steps in Tate's algorithm yield
the stated conclusions. The idea is to start with the given Weierstrass
equation and perform a sequence of blow-ups to produce a minimal regular
model for E. In practice, we will really only need to carry out the blowing-
up process until we are able to recognize which type of fiber is emerging.
Let W C lP'k be the scheme defined by the given Weierstrass equation,
and let WO j R be the largest subscheme of W that is smooth over R. In
other words, WO is formed by removing from W all singular points (if any)
on its special fiber W. Just as in the description of Tate's algorithm, we
will put a box around cumulative assumptions as we make them.
Proof of Step 1. The condition n f b.. means that the special fiber W is
non-singular, so W itself is smooth over R. Hence e = e = W, which shows
that the special fiber is of Type 10.
We assume now that lnlb..l, which means that the reduction E has a
singular point. (Equivalently, the special fiber W is singular.) Making a
linear change of variables, we may assume that the singular point is (0,0) E
E. This means that if we write
then /(0, 0) == 0 (mod n), and further both partial derivatives (8f j8x)(0, 0)
and (8fj8y)(0,0) vanish modulo n. Hence Ha3,a4,a61.
Proof of Step 2. This is the case that E has multiplicative reduction.
We are going to leave it to the reader (exercise 4.37) to perform the blow-
ups necessary to resolve the singularity in this case. At the end of this
section (9.6) we will briefly explain another approach to analyzing mul-
tiplicative reduction using Tate's analytic models for elliptic curves over
complete local fields. We will also prove the following lemma which covers
Types II and II.
Lemma 9.5. Let R be a discrete valuation ring with fraction field K,
let E j K be an elliptic curve given by a Weierstrass equation
with coefficients in R, let W C lP'k be the R-scheme defined by this equation,

and let WO j R be the largest subscheme of W that is smooth over R.
(a) Ifv(b..) = 1, then W is regular, e = W, and £ = Woo The curve E has
Type h reduction.
(b) Ifnla3, a4, a6 and n 2 f a6, then W is regular, e = W, and e = Woo The
curve E has Type h reduction if n f b2 , and Type II reduction if nlb2 •
PROOF. (a) As described above, the fact that nib.. means that we can make
a linear change of variables to get nla3, a4, a6. This implies that nlb4' b6 , bs .
If we make the assumption that v(D.) = 1, then we must have v(b 8) = 1,

since all of the other terms in the formula for D. are divisible by at least 7r 2 .
Writing bs in the form
we find that if v(D.) = 1, then v(a6) = 1 and 7r f h

We now drop the assumption that v(D.) = 1 and prove that W is
regular assuming only that 7ria3, a4, a6 and 7r 2 f a6. This will verify the
first statements in both (a) and (b). We need to prove that W is regular
at the singular point (0,0) E W on its special fiber. In other words, if we
let m = (7r, x, y) be the maximal ideal corresponding to the singular point
on the special fiber, then we must show that the local ring of W at m,
is a regular local ring. By assumption, v(a6) = 1, so a6 is a uniformizer

for R. On the other hand, a6 is in the ideal of (')W,m generated by x and y,
since
Therefore the maximal ideal (7r, x, y) of (')W,m is generated by the two el-
ements x and y, so (')W,m is a regular local ring. This proves that W is
regular, and since it is clearly also proper over R, we find that e = W
and c = Woo This pro~es t~e first part of (a) and (b).
The special fiber e = W is the curve
in A~. It will have a node (respectively cusp) at (0,0) if the quadratic

form y2 + iilxy - ii2X2 has distinct roots (respectively a double root) in k.
The discriminant of this quadratic form is iii + 4ii2 = b2, so Whas a node
if 7r f b2 and a cusp if 7rib2. By definition, the special fiber is of Type 11
if it has a node, and of Type II if it has a cusp, and we saw above that
if v(D.) = 1, then 7r f b2. This completes the proof of (9.5). 0
Continuing on past Step 2, we now assume that l7rib21. Notice that

b2 is the discriminant of the quadratic form y2 + al xy - a2x2, so this form
has a double root in k, say
Y2 + alxy - a2x 2 =_ ( Y - ax )2 (mod 7r).
The substitution y ~ y + ax allows us to assume that l7rial, a21. Notice

that this substitution leaves the other ai's and all of the bi's unchanged.
Proof of Step 3. We are given that 7rla3, a4, a6, b2 and that 7r 2 t a6' This
is exactly the situation in (9.5b), which shows that the special fiber is of
Type II.
Proof of Steps 4 and 5. We now add the assumption that 17r2Ia61, and
recall that our model satisfies 7rlal, a2, a3, a4. We are going to blow up the
singular point 7r = x = y = 0 of W using the procedure described in (7.7).
Thus the blow-up consists of the following three coordinate charts glued
together in an appropriate fashion:
7r'y' = 7r,
7r" x" = 7r.
Note that we are using the notation ai,r = 7r- r ai introduced earlier.
Looking at the special fibers of each of WI, W', and W", it is easy to
verify that the special fiber of W" contains all of the components and all of
the singular points of the special fiber of the blow-up. Thus all of the action
will be happening on W". Further, it is not hard to see that the projec-
tion A~ -+ Ah induced by the natural inclusion R[y", 7r"l '--+ R[x", y", 7r"l
maps W" isomorphically to the subscheme of Ah = Spec R[y", 7r"l given
by the single equation
(The map in the other direction is x" -+ y,,2 + alY" + a3.17r"y" - a2 -

a4 17r " - a6 27r ,,2 .)
, We ne;t take the closure of this scheme in lP'h. This means we homog-
enize
y" = YjX and 7r" = Zj X,
which yields the scheme V c lP'h = Proj R[X, Y, Zl given by the equation
Notice that V is a model for E j K, since its generic fiber is isomorphic

over K to the original Weierstrass equation defining E.
To find the special fiber of V, we set 7r = 0, so V is the curve in A%
given by the equation
Thus the special fiber consists of the line Z = 0 and the (possibly degen-
erate) conic
This line and conic intersect at the point Y = Z = 0 with multiplicity 2.

Suppose first that 71"3 f bs , which is the condition for Step 4. The
formula defining bs is
and our cumulative assumptions imply that every term except the last one
is divisible by 71"3. Hence 71"3 f bs if and only if 71"2 f a4, and this in turn
is equivalent to the assertion that (*) is a non-singular conic (as opposed
to being two lines). So 71"3 f bs implies that V consists of two non-singular
rational curves intersecting at a single point with multiplicity 2, which is
exactly Type III.
l,
Next assume that 17I"31bs or equivalently that 7I"2Ia4' This means that
the special fiber V is given by the equation
(**)
so over k it consists of three lines. These lines will be distinct if and only
if the quadratic form y2 + a3,l y Z - a6,2Z2 has distinct roots, which is
equivalent to the condition that its discriminant a~,l + 4a6,2 = b6,2 does
not vanish. So if we assume that 71"3 f b6 , which is exactly the condition for
Step 5, then Vconsists of three non-singular lines intersecting transversally
at a single point, which is a fiber of Type IV. Further, the number c of
components defined over k will be 3 if y2+ a3 ,l y Z -a6,2Z2 splits into linear
factors over k and will be 1 otherwise. This completes our consideration of
Steps 4 and 5 of Tate's algorithm.
We now assume that 17I"31b61. This means that the quadratic form
in (**) has a double root in k. Making a translation Y ---> Y + (3Z moves
the double root to y = O. We now have 17I"Ial,a2, 7I"2Ia3,a4, and 7I"3Ia61.
The equation for V can be written as
and its special fiber y2 Z = 0 consists of the line Y = 0 with multiplicity 2

and the line Z = 0 with multiplicity 1.
The next step is to blow-up the double line 71" = Y = O. To ease
notation, we are going to dehomogenize at the same time, so we set
and divide the equation for V by 71". (The reason for the subscripts on Xl
and Y2 is that they are related to our original Weierstrass coordinates by
the formulas X = 7I"Xl and Y = 7I"2 y2 . For the rest of this proof we will use
the notation x = 7I"TXT and Y = 7I"TYT')
We now have the scheme
and the total blow-up consists of V and Vo glued together in the natural
way. The special fiber of the total blow-up is thus formed by gluing together
the two pieces
V: y 2 Z = 0,
170 : 0 = xf + a2,lxi + a4,2xl + a6,3'
There are now three cases to consider, depending on the number of distinct
roots (in k) of the polynomial
Proof of Step 6. For Step 6 we assume that P(T) has distinct roots
in k. Then 170 consists of three distinct lines, so the blow-up is composed
of the double line y2 = 0 together with four lines of multiplicity 1 in-
tersecting it. This means we have a fiber of Type 10, Further, there is
always one component Z = 0 of multiplicity 1 defined over k, and the
other multiplicity-l components correspond to the roots of P(T). Hence
the number c of multiplicity-l components is one more than the number of
roots of P(T) in k.
Proof of Step 7. For Step 7 we assume that P(T) has one simple root and
one double root. Making a translation of the form Xl --> Xl + '"Y, we may
assume that the double root is T = 0, which implies that 7r 2 f a2, 7r 3 !a4,
and 7r 4 !a6. The special fiber of Vo is now
so we need to blow-up Vo along the double line 7r = Xl = O. To do this, we

make the substitution Xl = 7rX2 and divide by 7r to obtain the scheme
Our total special fiber is now composed of the following components: the
simple lines Z = 0 and Xl + a2,1 = 0, the double lines Y = 0 and Xl = 0,
and the special fiber of VI' Notice how a fiber of Type I~ is emerging.
The special fiber of VI is
V- 1 : Y22 + a3,2Y2
- -
- a6,4 =
0
.
If this quadratic equation has distinct roots in k, then VI consists of two

distinct lines, and we have a fiber of Type Ii. Further, there are already
two multiplicity-l components defined over k, namely the lines Z = 0
and Xl + a2,1 = 0, so c = 4 if the polynomial Y§ + a3,2Y2 - a6,4 has its roots

in k, and c = 2 otherwise.
If the polynomial Y§ + a3,2Y2 - a6,4 has a double root, then making
a translation on Y2 allows us to take the double root to be Y2 = O. This
means that 7l'3ta3 and 7l'5ta6, and the special fiber of Vl is Y§ = O. We
blow-up V l along this double line by making the substitution Y2 = 7l'Y3 and
dividing by 7l', which gives the scheme
The special fiber of V2 is
If this quadratic equation has distinct roots in k, then V2 consists of two

distinct lines, we have a fiber of Type 12, and we're done. Otherwise the
quadratic equation has a double root and V2 is a double line, so we translate
to make the double line x~ = 0, blow it up using X2 = 7l'X3, and continue
on our merry way.
As explained during the description of the Step 7 subprocedure, this
process will eventually terminate. The point is that the special fiber at
each stage looks like
if n = 2v - 3 is odd,
if n = 2v - 2 is even.
So each two steps of the algorithm force a3, a4, and a6 to be divisible by
an additional power of 7l'. This implies the same for b4 , b6 , and bs , and
hence also for ~. But ~ is invariant under the various translations we are
using, which shows that eventually we must get a quadratic polynomial
with distinct roots. We will leave for the reader the easy verification that
if the residue characteristic p i= 2, then the fiber Vn consists of two distinct
lines precisely when n = v(~) - 6. This concludes our discussion of Step 7
of Tate's algorithm.
Proof of Step 8. We now assume that the polynomial P(T) has a triple root
in k, which after a translation we can take to be T = O. This means that
17l'2ta2, 7l'3ta4, and 7l'4ta61, so the special fiber Vo is the triple line x~ = O.
Our total special fiber now consists of the simple line Z = 0, the double
line y2 = 0, and the triple line xr = O. The scheme Vo is regular except at
the points on the special fiber satisfying
Making a translation Y2 ~ Y2 + 'Y allows us to assume that the polynomial

y§ + a3,2Y2 - a6,4 has Y2 = 0 as a root in k. This may require making
a quadratic extension of k, in which case the components of the special

fiber corresponding to the two roots of the quadratic polynomial will not
be defined over k.
We now have 17r5Ia61, and we blow-up Vo at the point 7r = Xl = Y2 = 0
by making the change of variables
7r = 7r'y', Xl = X
,y,, y =y'.
(This is chart 2 of the blow-up as described in (7.7).) This yields the
scheme 11' C A1. = Spec R[7r', x', y'] given by the equations
7r = 7r'y'.
The special fiber of 11' consists of three components, which we label as
F{ : 7r' = x' = 0, F~: 7r' = y' = 0, F~: y' = ib,2 - a4,3x' - a6,57r' = o.
Notice that when we glue 11' to Vo, F{ is identified with the mul-
tiplicity-3 component x~ = 0 of 170 . Our next step is to compute the
multiplicities of the new components F~ and F~. To do this, we rewrite the
equation for 11' as
7r Y
'{ , + al,lx ,Y, + a3,2 ,2 ,
- a2,2x Y - a4,3x - a6,57r
, ,},3 ,
= X y.
The function x' does not vanish identically on F~, so it is a unit in the local
ring (') F~. Similarly, since we are making the Step 8 assumption that 7r 3 f a3,
the quantity in braces is also a unit in (') pl.
2
It follows that both 7r' and y'
are uniformizers for (') F~; that is, they each vanish to order 1 on F~, so
ordFJU') = ordF~(7r) = ordF~(7r'Y') = 2.
We leave for the reader the analogous verification that ordF~ (11') = l.
But we're not done with Step 8, because we have to perform an iden-
tical blow-up of Vo at the singular point 7r = Xl = Y2 + a3,2 = o. This gives
another pair of components, one of multiplicity 2 and one of multiplicity l.
The resulting configuration is of Type IV· , which completes the verification
of Step 8 of Tate's algorithm.
Proof of Step 9. For this step we have 1~3Ia31, so the scheme 11' given
above is singular at the point 7r' = x' = y = o. We blow it up at that
point by making the substitution
7r' = 7r" x", X' = x", y' = y"x".
This gives the scheme 11" C A1. = Spec R[7r", x", y"] defined by the equa-
tions
11" : y" 7r" + a 1,1 x"y" 7r" + a3,3 x"y"7r,,2
= X,,2 Y " + a2,2x ,,2 Y" 7r " + a4,37r " + a6,57r ,,2 ,
7r = X,,2 y" 7r".
Under our Step 9 assumption that 11.4 f a4, we find that U" consists of the
following four components:
F{' : x" = 1':" = 0, F~' : y" = 1':" = 0,

F~' : x" = y" - a4,3 - a6,51':" = 0,
F~' : y" = a4,3 + a6,51':" = O.
We compute the multiplicity of F{' in the fiber by writing
1':
II {
Y II + aI,Ix II Y" + a3,3 x II YII 1':II - a2,2x
112
Y II - a4,3 - a6,51': " } = x 112 y.
II
The function in braces does not vanish identically on F{" so it is a unit

in C) F".
1
This means that x" is a uniformizer for F{' and ordF"1 (1':") =
2 ord F 1" (x") = 2. Hence
So we now have a chain of components of multiplicities 1, 2, 3, and 4.

Notice how the fibers of Type III* and II* are emerging.
A similar calculation shows that ordp 2" (U") = 2. Further, our Step 9
assumption that 1':4 f a4 implies that U" is regular at the point 1':" =
x" = y" = 0 where F{' and F~' intersect. Hence there is a multiplicity-2
component attached to the multiplicity-4 component of the regular minimal
model. This means that the fiber is of Type III*, which completes our
analysis of Step 9. For those who wish to recover the full Type III* fiber,
we mention that U" is singular at the intersection of FI and F 3 , that is, at
the point 1':" = x" = y" - a4,3 = O.
Proof of Step 10. We now assume that 11':4Ia41. Then U" is singular at
the point 1':" = x" = y" = 0, so we blow it up using the substitution
x" = x"' y"', y" = y"'.

This gives the scheme
U '" : 1': ", + aI,Ix ", Y"'1':", + a3,3x ", Y",2 1': ",2
= X ,"2 Y", + a2,2x ",2 Y",2 1':", + a4,4x ",2 Y",3 1':",2 + a6,51':,
",2
1': = X",2 y",4 1':"'.
The special fiber U'" consists of four components,
F{" : x'" = 1':'" = 0, F~" : y'" = 1':'" = 0,

F~" : x", = 1 - a6,51':'" = 0, F~" : y"' = 1 - a6,51':"' = O.
We compute the multiplicity of Ft in the special fiber by writing
71"
", {I + a1,lx", Y", + a3,3 x ",y",2",
71" -
a
2,2
X",2 y ",2
-a4,4x
",2 ",3
Y 71"
",
- a6,571"
"'}
= x",2 y '" .
Both x'" and the quantity in braces are units in (') Fill,
2
which shows that 71"'"
and y'" each vanish to order 1 on F~". This allows us to compute
ord Fill - III) = or d Fill ()

(U 71" = or d Fill (",2",4",)
x Y 71" 5
=.
2 2 2
Further, the Step 10 assumption that 71"6 t a6 implies that U'" is regular
at the intersection point 71"'" = x'" = y'" = 0 of F{" and F~". Hence the
appearance of the multiplicity-5 component F~" tells us that the fiber is of
Type II*. As usual, we leave for the reader the enthralling task of perform-
ing the additional blow-ups necessary to find the other II* components.
Proof of Step 11. Finally, suppose that i71"6Ia6i. Our cumulative assump-
tions to this point are that 71"la1, 71"2Ia2, 7r3Ia3, 7r4Ia4, and 7r6Ia6' In U'" we
make the substitutions
7r'" = 1 j 7ry~ , x'" = xV 7ry~

which leads to the R-scheme
defined by a Weierstrass equation whose discriminant is 7r~ 12~. We can

now begin again at Step 1 using this "smaller" Weierstrass equation. Note
that each time we pass through Step 11, we will have shown that the original
discriminant is divisible by an additional 7r 12 . Therefore the algorithm will
terminate. This concludes the proof of Tate's algorithm (9.4). 0
Remark 9.6. During our verification of Tate's algorithm (9.4), we left

the case of multiplicative reduction (Type In) for the reader to analyze.
There is another approach to multiplicative reduction using Tate's p-adic
analytic uniformization. We will describe Tate's uniformization in the next
chapter (V.3.1, V.5.3), but briefly, if E has split multiplicative reduction
and K is a complete local field, then there is a q E K* with v(q) = v(~) > 0
and an isomorphism of groups
K* jqZ ~ E(K).
This isomorphism is given by v-adically convergent power series. Further,

the isomorphism identifies the subgroups R* 2;! Eo(K), so we get isomor-
phisms
Here the second map is induced by the valuation v : K* -+ Z, and
n = v(q) = v(~).
We know that E(K) = C(R), and it is clear from the definitions

that Eo(K) = WO(R). Further, one can show that WO(R) = £O(R), ei-
ther by a direct calculation or using the argument in (V §4). It follows
that £(R)/£O(R) ~ Z/nZ, and then (9.2b) gives c = n.
Similarly, if E has non-split reduction, then (V.5.4) says that we can
find an unramified quadratic extension K' / K with residue field k' such
that E has split reduction over K'. Then
E(K') ~ K'* /l',

E(K)~{UEK'*/q'll.: Nf(u) Eq'll./q27!.},
Eo (K) ~ { K' (u) = 1 } ~ { u E k 1* : Nkk' (u) = 1 } .
u E R 1* : NK
Note the last isomorphism depends on the fact that K' / K is unramified.
Finally, we have
The fact that K' / K is unramified means that the norm map is surjective on
units, Nk : R'* ---» R*, from which one easily deduces that this last group
is trivial if n is odd, and has order 2 if n is even.
PROOF (of Corollary 9.1). If we start with a minimal Weierstrass equation

for E / K, then we never get to Step 11 of Tate's algorithm, so the original
equation defining W never changes. Tracing through the various stages
of Tate's algorithm, we see that the non-singular part WO of the Weier-
strass equation ends up as an open subset of the minimal regular model e.
Since WO £learly contains the image of 0Ie zero section, and since the spe-
cial fiber W is irreducible, we see that W is the identity component of the
special fiber of e. Equivalently, WO = £0. (For an alternative proof of (9.1)
which uses a bit more algebro-geometric machinery and does not rely on a
case-by-case analysis, see Liu [1].) 0
PROOF (of Corollary 9.2). (a) First, the equality E(K) C(R) follows
from the definition of the Neron model (5.1.3). Next we observe that the
definitions of Eo and WO are both given in terms of the reduction of the
given Weierstrass equation, so Eo(K) = WO(R) is automatic. Now (9.1)
says that WO = £0, so we get the middle equality Eo(K) = £O(R). Fi-
nally, El (K) and £ 1 (R) each consists of the points which reduce to the
identity on the special fiber, and these reductions are compatible since
we already know that Eo(K) = £O(R). This proves the third equal-
ity El(K) = £l(R).
(b) The isomorphisms E(K)jEo(K) ~ £(R)j£O(R) and E(K)jEl(K) ~

£(R)j£l(R) are immediate from the identifications proven in (a). Further,
the reduction map £(R) ---> £(k) has kernel £l(R), and the inverse image
of £O(k) is £O(R) by definition, which gives the injectivity of the right-hand
maps. Finally, if R is complete (or merely Henselian), then the reduction
map £(R) ---> £(k) is surjective (6.4a), so in this case the right-hand maps
are isomorphisms.
(c) We have Ens(k) = \\JO(k) directly from the definitions. Now (9.1)
implies that \\JO(k) = £O(k), which gives the desired result.
(d) From (b) we have an injection E(K)jEo(K) '-' £(k)j£O(k). The
e
group £(k)j£O(k) is formed by looking at the special fiber of the minimal
proper regular model and taking the components that have multiplicity 1
and are defined over k. A quick perusal of the list of reduction types shows
that only Type In has more than four multiplicity-1 components. Further,
Tate's algorithm (Step 2, see also (9.6)) says that if a fiber of Type In has
non-split reduction, then it has at most two components defined over k.
This proves that E(K)jEo(K) has order at most 4 unless EjK has split
multiplicative reduction. Finally, if E j K has split multiplicative reduction,
say with a Type In fiber, then Step 2 of Tate's algorithm says that n equals
the valuation of the minimal discriminant, which is also equal to -v (j (E)).
D
§10. The Conductor of an Elliptic Curve
The conductor of an elliptic curve E j K is a quantity which measures the

arithmetic complexity of E j K, similar in some ways to the minimal dis-
criminant. Just like the discriminant, the conductor is a product over the
primes p at which E has bad reduction, but the exponent of p is defined
in terms of the representation of the inertia group on the torsion subgroup
of E. The conductor is an important quantity which appears in the func-
tional equation of the L-series of E, in the modular parametrization of
elliptic curves over Q, and in various questions concerning the cohomology
of E.
Before defining the conductor, we briefly recall some standard facts
about local fields. For more details, see Serre [4J. Let K be a local field of
residue characteristic p, let Lj K be a finite Galois extension with normal-
ized valuation VL and ring of integers RL, and let G(Lj K) be the Galois
group of Lj K. Then for each integer i ::::: -1, the ith-higher ramification
group of Lj K is the subgroup of G(Lj K) defined by
Gi(LjK) = {a E G(LjK) : vLCa II - a) ::::: i + 1 for all a E RL}.

We write
gi(L/ K) = #Gi(L/ K)
for the order of the ith-ramification group. One should think of the higher
ramification groups as measuring the extent to which the extension L/ K
is wildly ramified. The following lemma records some basic facts about the
higher ramification groups.
Lemma 10.1. Let L/ K be a finite Galois extension of local fields.
(a) The higher ramification groups Gi(L/ K) are normal subgroups of
G(L/K).
(b) G_1(L/K) = G(L/K).
(c) Go(L/K) is the inertia group of L/K.
(c) [Go(L/K): G1(L/K)] is relatively prime to p.
(d) G1(L/K) is a p-group. Thus L/ K is wildly ramified if and only
if G 1 (L/ K) i- 1.
PROOF. See Serre [4], especially Chapter IV, Proposition 1 and Corollar-
ies 1 and 3 to Proposition 7. 0
The conductor of an elliptic curve consists of two pieces, a tame part

and a wild part. It turns out that if the residue characteristic p is at
least 5, then the wild part will be zero. So if one is willing to ignore residue
characteristics 2 and 3, then 8(E/ K) can just be set equal to 0 in the
following definition.
Definition. Let E / K be an elliptic curve defined over a local field of
residue characteristic p, and let I(K / K) be the absolute inertia group
of K. Fix a prime £ different from p, let Vi(E) = T£(E) Q9Zt Q£ be the £-
adic Tate module of E, and write Ve(E)ICK/K) for the subspace of Ve(E)
that is fixed by I(K/K). The tame part of the conductor of E/K is the
quantity
Next let L = K(E[£]). Then the wild part of the conductor of E/K is the
quantity
8(E/K) =~ gi(L/K) dim (E[£j/E[£jGiCL/K))

~ go(L/ K) 1Ft •
The exponent of the conductor of E / K is the sum of the tame and wild
parts,
f(E/K) = c(E/K)+8(E/K).
The conductor is a representation-theoretic quantity, since it is defined
in terms of the action of the Galois group G(K / K) on the torsion subgroup
of E. The following generalization of the criterion of Neron-Ogg-Shafare-

vich [AEC, VII.7.1] provides a geometric interpretation for the tame part
of the conductor. We note that in many books (including [AEC, C §16])
one finds this geometric description (1O.2b) used as the "definition" of the
conductor.
Theorem 10.2. Let K be a local field of residue characteristic p, and
let E / K be an elliptic curve.
(a) The tame part of the conductor of E/ K is given by
0 if E has good reduction,

c( E / K) = { 1 if E has multiplicative reduction,
2 if E has additive reduction.
(b) If E / K has good or multiplicative reduction, or if p ::::: 5, then
0 if E has good reduction,

8(E/K) = 0 and f(E/K) = { 1 if E has multiplicative reduction,
2 if E has additive reduction.
(c) In all cases, the exponent of the conductor f (E / K) is an integer which

is independent of the choice of C.
PROOF. (a) Notice that
c(E/K) = 0 ~ Ve(E)I(KIK) = Ve(E)

~ I (k / K) acts trivially on T£ (E)
~ Tc(E) is unramified.
So the assertion that
c( E / K) = 0 ~ E/ K has good reduction
is precisely the criterion of Neron-Ogg-Shafarevich [AEC, VIL7.1]. We are

going to mimic the proof of [AEC, VIL7.1] to obtain a somewhat stronger
result. This proof is taken from Serre-Tate [1].
Let K nr be the maximal unramified extension of K, and consider the
two exact sequences
o -- Eo (Knr) __ E(Knr) __ E(Knr)/ Eo(Knr) __ 0,
o -- E1(Knr) -- Eo (Knr) -- EnsCk) -- o.

Here k, the residue field of Knr, is the algebraic closure of the residue field
of K. We note that E(Knr)/Eo(Knr) is a finite group from (9.2d), and
that El(Knr) has no £-torsion from [AEC, VII.3.1]' since it is isomorphic
to the formal group of E.
For any abelian group A, we let Te(A) denote the Tate module of A,
and we set
We observe that Vc(A) will be 0 if A has no f-torsion, or if A is a finite

group. In particular, -
and
Hence the two exact sequences given above yield isomorphisms
On the other hand, we clearly have
which proves the fundamental isomorphism
Now we compute
c(EjK) = 2 - dimIQf (Vc(E)I(k/K))
= 2 - dimIQf (Vc (Ens (k) ) )

dimIQ, (Vc(E)) if E has good reduction,
= 2 - { dimIQ£ (Ve. (§*)) if E has multiplicative reduction,
dimIQ,(Vc(k+)) if E has additive reduction,
where the last line follows from the standard description of the various
reduction types [AEC, VII.5.1]. Using the fact that f i p, we find that
which completes the proof of (1O.2a).

(b) Fix a prime f i p, and let L = K (E[f]). If E / K has good reduction,
then Lj K is unramified from [AEC, VII.4.1], so the inertia group Go(L/ K)
is trivial. It is then clear from the definition that 8(E / K) = O.
Next suppose that E / K has non-integral j-invariant, that is, VK(jE) <
O. Let f i p be a prime. We will prove later (see (V.5.3) and exercise 5.11)
that there is an extension K' / K with the following properties:
§1O. The Conductor of an Elliptic Curve 383
(1) [K': K] = 1 or 2.
(2) K'/ K is unramified (respectively ramified) if E/ K has multiplica-
tive (respectively additive) reduction.
(3) There is a q E K' such that K'(E[l]) = K'(J.t£,ql/£).
Here J.tl denotes the group of fth_roots of unity.
Thus the extension K'(E[lJ)/K' is composed of an unramified cyclo-
tomic extension K'(J.tl)/K' and a Kummer extension K'(J.t£,ql/£)/K'(J.te)
whose order divides f. This shows that the extension K'(E[lJ)/K' is
at worst tamely ramified. It follows from properties (1) and (2) that
K (E[l]) / K is tamely ramified if either E / K has multiplicative reduction
or if p ~ 3, which completes the proof of (1O.2b) in the case that jE is
non-integral.
Finally we consider the case that E / K has integral j-invariant, or
equivalently from [AEC, VII.5.5], E has potential good reduction. A key
tool in proving (10.2) in this case is the following strengthening of the
criterion of Neron-Ogg-Shafarevich.
Proposition 10.3. Let K be a local field of residue characteristic p, and
let E / K be an elliptic curve with integral j-invariant.
(a) The following are equivalent.
(i) E has good reduction over K.
(ii) E[m] is unramified for every integer m ~ 1 relatively prime to p.
(iii) E[m] is unramified for at least one integer m ~ 3 relatively prime
to p.
(b) Let m ~ 3 be an integer relatively prime to p. Then E has good
reduction over K(E[mJ).
PROOF. (a) The equivalence of (i) and (ii) is [AEC, VII.7.1], and the im-
plication (ii) =? (iii) is trivial. So it suffices to prove that (iii) implies (i).
We are given that E / K has potential good reduction, so we can find
a finite Galois extension L / K such that E has good reduction over L. We
are also given an integer m ~ 3 such that E[m] is unramified over K. Let f
be the largest prime dividing m, and let
£' = f if f -1= 2 and f' = 4 if f = 2.
Notice that f'lm since m ~ 3, so E[f'] C E[m]. Thus E[t"] is unramified

over K.
The fact that E has good reduction over L means that the inertia
group !t/L acts trivially on the Tate module Te(E) [AEC, VII.4.1b], so
the inertia group h/K of L/K acts on Te(E). This action gives us a
homomorphism
p: h/K --> Aut(Te(E»).
Further, we are given that E[f'] is unramified over K, so h/K acts trivially
on E[f']. In other words, the image P(h/K) is contained in the kernel of
the natural map

Aut(Tt(E)) ---> Aut(E[£']).
If we choose bases Tc(E) ~ Z; and E[l"] ~ (Z/£'Z)2, then this last
map becomes
It is an elementary exercise to verify that the kernel of this map, namely

{M E GL2(Zc) : M == 1 (mod £')},
has no elements of finite order. (See exercise 4.38. This is the point at which
we need £' = 4 if £ = 2, since if we took £' = 2, then the matrix (-6 _~)
would be in the kernel.)
vVe saw above that the image p( h / K) is contained in this kernel. But
the group h/K is finite, so it follows that its image P(h/K) is trivial. In
other words, the inertia group h/K acts trivially on T£(E), which proves
that Tc(E) is unramified over K. Now [AEC, VII.7.1] tells us that E has
good reduction over K, which completes the proof that (iii) implies (i).
(b) This follows immediately from (a), since E[m] is clearly unramified
over the field K(E[m]). D
We now resume the proof of Theorem 10.2(b), where, recall, we are

assuming that E has integral j-invariant and that p 2: 5, and we are trying
to verify that 6(E / K) = O. Without loss of generality, we may replace K
by its maximal unramified extension.
For each integer m 2: 3 relatively prime to p, let Lm = K(E[m]). Now
Proposition 10.3(a) tells us that E has good reduction over L m , and then
another application of (10.3a) says that for any other m', the set E[m'] is
unramified over L m , which means that the cornpositurn Lrn' Lm is an un-
ramified extemlion of Lm. But we took K = Knr, so Lm has no unramified
extensions, and hence L m , eLm. Reversing the role of m and m' gives
the opposite inclusion, which proves that all of the Lrn's are the same. We
write L for this common field.
Now let £ 2: 3 be a prime with £ =I p. (We will deal with the case £ = 2
later.) The action of the Galois group G(Ld K) on E[£] gives an injection
It follows that
#G(Ld K) 1# GL2(Z/£Z),
But we showed above that the field Lc = L is independent of £, so we find
that #G(L/ K) divides # GL2(Z/£Z) for all £ =l2,p. The group GL 2 (Z/£Z)
has order £(£ - 1)2(£ + 1), and it is easy to see using Dirichlet's theorem
on primes in arithmetic progressions that
gcd {£(£ - 1?(£ + 1)} = 48.
£#2.p
Hence G(L/ K) has order dividing 48.

In particular, [L : K] is not divisible by p, since p :::: 5 by assumption.
This proves that the extension L / K is at worst tamely ramified, so the
higher ramification groups Gi(L/K) are trivial for i:::: 1 (lO.ld). It follows
directly from the definition of 8(E/K) that 8(E/K) = O. This completes
the proof of (lO.2b) for C :::: 3.
Finally, if C = 2, we use the fact that L = L4 and that E has good
reduction over L to conclude that
G(L/K) '----> Aut (E[4]) ~ GL2(Z/4Z).
This last group has order # GL 2(Z/4Z) = 96 = 25 .3, so L/ K is not wildly

ramified at p, since P:::: 5. Hence 8(E/K) = O.
(c) If P::::: 5, then (b) says that f(E/K) is an integer which depends only
on the reduction type of E/ K, hence is independent of C. The general case,
which is due to Ogg [2], uses more machinery than we want to develop
here. We have sketched the proof in exercise 4.46. For further details, see
Ogg [2], Serre [7, chapter 19], and Serre-Tate [1, §3]. 0
If the residue characteristic of K is not equal to 2 or 3, then we

have seen that 8(E/K) = 0, and so the exponent of the conductor sat-
isfies f(E/K) :::; 2 from (10.2b). When the residue characteristic is 2 or 3,
the exponent of the conductor is still bounded as described in the following
result.
Theorem lOA. (Lockhart-Rosen-Silverman [1], Brumer-Kramer [1]) Let
K /Qp be a local field with normalized valuation v K, and let E / K be an
elliptic curve. Then the exponent of the conductor of E / K is bounded by
(Here VK(p) is the ramification index of K/Qp.) Further, this bound is

best possible in the sense that for every finite extension K/Qp there is an
elliptic curve ElK whose conductor attains this bound.
PROOF. We are going to prove the slightly weaker bound
since the proof is easier and the weaker estimate suffices for most appli-
cations. For an elementary, but involved, proof of the stronger inequality
in certain cases, see Lockhart-Rosen-Silverman [1]. The proof for gen-
eral K requires heavier machinery from representation theory; see Brumer-
Kramer [1].
We begin with the observation that if p :::: 5, then 8(E/K) = 0
from (10.2b) and c(E/K) :::; 2 directly from its definition, so f(E/K) :::; 2.
It remains to deal with the cases p = 2 and p = 3.
Let £ be any prime other than p, let L = K(E[£]), let 'IJ L1K be the
different of L/ K, and let r be the smallest integer such that Gr(L/ K) = l.
We will need the following elementary properties of local fields:
00
(i) vd'IJLIK) = 'L)gi(L/K) -1),

i=O
(ii) vd'IJ LI K) :S go(L/ K) - 1 + vL(go(L/ K)),
(iii) r < vdp) + l.

- p-l
See Serre [4, IV, §1, Prop. 4J for (i), Serre [4, III, § 7, remark following
Prop. 13J for (ii), and Serre [4, IV, §2, exercise 3(c)J or Lockhart-Rosen-
Silverman [1, Lemma l.2(b)] for (iii).
We are now ready to compute.
f(E/K) = c(E/K)+8(E/K)
:S 2+8(E/K) since clearly c(E/K):S 2
= 2+ ~ gi(L/K) dim (E[£]/E[£]C,(LIK))

~ go(L/ K) IFe
since Gi(L/ K) = 1 for i 2': r
2 r-l
:S 2 + go(L/ K) ~ gi(L/ K) since dimIFe (E[£]) =2

2 r-l
= go(L/ K) ~ gi(L/ K)
2 00
= go(L/ K) (r + ~(9i(L/ K) - 1))

2
= go(L/K) (r + vd'IJLIK)) from property (i)
:S go(:/K) (;L~p1 + 1 + go(L/K) -1 + VL(go(L/K)))

from properties (ii) and (iii)
= 2VK(P) +2+2vK(go(L/K)) sincego(L/K) is the

P- 1 ramification index of L / K.
Now suppose that p = 3, and take (say) £ = 5. Then

G(L/K) "-----> Aut (E[5]) ~ GL 2 (Z/5Z),
so in particular go(L/ K) divides # GL 2 (Z/5Z) = 480 = 25 . 3 . 5. Hence
vK(go(L/K)) :S vK(480) = vK(3),

and substituting this in above with p = 3 gives the desired estimate,
Similarly, if P = 2, then we can take (say) f = 3 and use the injection

G(L/ K) '-------+ Aut (E[3]) ~ GL 2 (Z/3Z)
to conclude that 9o(L/ K) divides # GL 2 (Z/3Z) = 48 = 24 . 3. However,

we can easily save a little bit. We are allowed to make an unramified
extension of K, so we may adjoin to K a primitive cube root of unity.
Then basic properties of the Weil pairing [AEC, III.8.1] imply that the
image of G(L/ K) lies in 8L2(Z/3Z). (We sketch the proof below.) Hence
the ramification index 90(L/ K) divides # 8L 2 (Z/3Z) = 24 = 23 ·3. It
follows that
VK(9o(To/K)) ::; 1!K(24) = vK(2:l ),
and then substituting this in above with p = 2 yields
Let 'if be a uniformizer for K. If p ?: 5, then (1O.2b) says that any

elliptic curve E / K with additive reduction will hit the maximum conductor
exponent f(E / K) = 2. If p = 3, then we claim that the elliptic curve
E: y2 = x3 + 'if
satisfies f (E / K) = 2 + 3v K (3). 8imilarly, if p = 2, we claim that the elliptic
curve
E :1/ + 2xy = .];:3 - x2 + 'if X
satisfies f(E/K) = 2 + 6VK(2). We could verify these claims by a lengthy
direct calculation, but instead we will leave them for the reader to check
(exercises 4.52 and 4.53) using Tate's algorithm (9.4) and a formula of
Ogg (11.1) to be proven in the next section.
It remains to prove thf' assertion from above that the image of G(L/ K)
lies in 8L 2 (Z/3Z). Fix a basis S, T E E[3]. Then e3(S, T) is a primitivf'
cube root of unity, so it is in K. Let cr E G(L/ K), and let p( cr) = (~ ~) E
GL 2 (Z/3Z) be the matrix giving the action of cr on E[3] relative to the
chosen basis. Using [AEC, III.8.1(a,b,c,d)], we compute
e3(S, T) = e3(S, T)" = e3(S", T cr ) = e3(aS + cT, bS + dT)

= e;; (S, T)(ui-bc = C;; (S, T)dct p(u).
Therefore det p(cr) = 1, so the image of G(L/ K) lies in 8L2(Z/3Z). 0
The conductor of an elliptic curve over a number field is defined by

combining all of the local conductor exponents, just as the minimal dis-
criminant was defined as a product of the local discriminants.
Definition. Let ElK be an elliptic curve defined over a number field K,

and for each prime 13 of K, let f (E I K p) be the exponent of the conductor
of E consider as an elliptic curve over the local field Kp. The conductor
of ElK is the ideal
f(EIK) = IIp!(E/Kp ).
p
Example 10.5. Let ElK be an elliptic curve defined over a number field,
and suppose that E has everywhere semi-stable reduction, by which we
mean that E has either good or multiplicative reduction at every prime.
Then the conductor of E / K is the product of the primes of bad reduction,
f(EIK) = II 13,
P!"D E / K
where TJ E / K is the minimal discriminant of ElK [AEC, VII §8]. Con-

versely, if the conductor f(E I K) is square-free, then ElK has everywhere
semi-stable reduction. This follows from (10.2), which says that the con-
ductor exponent satisfies f(EIKp) ;::: 2 if and only if EIKp has additive
reduction.
Both the minimal discriminant and the conductor measure the extent
to which an elliptic curve has bad reduction. We will see in the next
section (11.2) that the exponent of the minimal discriminant is always
greater than the exponent of the conductor, so we always have an inequality
of the form
N{f(fE/K):=:; N{f(TJ E/ K ).
Szpiro has conjectured that there should be an inequality in the other
direction.
Szpiro's Conjecture 10.6. Fix a number field K and an c > O. There
is a constant c( K, c) so that for every elliptic curve ElK,
N{f(TJ E/ K ):=:; c(K,c)N{f(fE/K)6+ E •
This conjecture, if true, lies very deep. Its validity would imply the
solution to many other Diophantine problems, including for example the
assertion that if a, b E IQ* are fixed, if n ;::: 2, and if m is sufficiently large,
then the equation ax n + bym = 1 has no non-trivial solutions x, y E IQ.
Surprisingly, it is quite easy to prove a function field analogue of
Szpiro's conjecture; see exercise 3.36. One can even prove such a result
with c = 0 and with an explicit constant c. The function field version
of Szpiro's conjecture was originally discovered by Kodaira (see Shioda [3,
Prop. 2.8]) long before Szpiro formulated his conjecture, and it has been
frequently rediscovered since that time; see for example Hindry-Silverman
[2, Thm. 5.1] and Szpiro [1].
§11. Ogg's Formula
Let ElK be an elliptic curve defined over a local field. Ogg's formula relates
the minimal discriminant of ElK, the exponent of the conductor of ElK,
and the number of components on the special fiber of the minimal model
°
of E over the ring of integers of K. This formula was originally proven
by Ogg [2] in all cases except when K is a field of characteristic with
residue field of characteristic 2. Ogg's proof relies on a lengthy case-by-
case analysis. A more conceptual proof using scheme-theoretic techniques
and working in all residue characteristics has been given by Saito [1], who
proves Ogg's formula as a special case of a general result for curves of
arbitrary positive genus. An expanded exposition of Saito's proof just in
the case of elliptic curves can be found in Liu [1].
Ogg's Formula 11.1. (Ogg [2], Saito [1]) Let KIQp be a local field,
let ElK be an elliptic curve, and let
v K ('D E j K) = the valuation of the minimal discriminant of ElK,

f(EIK) = the exponent of the conductor of ElK,
m(EIK) = the number of components on the special fiber of ElK.
Then
VK('D EjK ) = f(EI K) + m(EI K) - 1.
Remark 11.1.1. The number m(EIK) in (10.1) is the number of irre-

ducible components defined over k on the special fiber of the minimal proper
regular model of ElK. This includes all of the components, not just the
multiplicity-l components which make up the special fiber of the Neron
model. Further, each component is counted once, regardless of its multi-
plicity. For example, if ElK has Type In reduction, then m( ElK) = n, and
if ElK has Type 111* reduction, then m( ElK) = 8. The value of m( ElK)
for these and the other reduction types can be found in Table 4.1.
Remark 11.1.2. The minimal discriminant and special fiber of ElK can
be computed in a straightforward manner using Tate's algorithm (9.4). For
this reason, Ogg's formula (11.1) is frequently used to compute the expo-
nent of the conductor of ElK for residue characteristics 2 and 3. See for
example exercises 4.52 and 4.53, as well as the conductor tables contained
in Birch-Kuyk [1] and Cremona [1].
PROOF (of Ogg's Formula 11.1). If ElK has good reduction, then
VK('D EjK ) = 0, f(EIK) = 0, and m(EIK) = 1.

These three equalities follow from [AEC, VII.5.1a], (1O.2b), and (6.3) re-
spectively. (Notice that (6.3) says that the minimal Weierstrass equation
for ElK is already the Neron model, so its special fiber is irreducible.)
This verifies Ogg's formula when ElK has good reduction.
Next suppose that ElK has multiplicative reduction. Then f(EI K) =
1 from (1O.2b). Further, Step 2 of Tate's algorithm (9.4) tells us that the
special fiber of the minimal model of ElK is an m( E I K)-sided polygon,
with m(EIK) = VK('D E / K ). This verifies Ogg's formula when ElK has
multiplicative reduction.
Finally, suppose that ElK has additive reduction. Consider first the
case that p 25. Then (10.2b) tells us that f(EI K) = 2, so Ogg's formula
becomes
VK('D E / K ) = m(EIK) + l.
It is now a simple matter using Table 4.1 to verify Ogg's formula case-by-
case, checking each of the reduction types II, III, ... , II*.
It remains to consider p = 3 and p = 2 when ElK has additive
reduction. We will give a direct case-by-case verification for p = 3, since
it only takes a few pages. A similar proof for p = 2 would be extremely
lengthy, so for this last case we refer the reader to Saito's proof [1] which
works in all residue characteristics and does not rely on a case-by-case
analysis. (See also Liu [1].) Unfortunately, the papers of Saito [1] and
Liu [1] use techniques which are beyond the scope of this book.
So we now assume that p = 3 and that ElK has additive reduc-
tion. In particular, (9.2a) tells us that the tame part of the conductor
is c(E I K) = 2. Further, the fact that p = 3 means that we can find a
minimal Weierstrass equation for ElK of the form
E : y2 = x 3 + a2x2 + a4X + a6.

The discriminant of this equation is
b. = -16(4a~a6 - a~a~ + 4a~ + 27a~ - 18a2a4a6).

Using this simplified form for E will make all of our calculations easier.
Let L = K(E[2]) be the field generated by the 2-torsion points of E,
so L is the splitting field over K of the cubic polynomial
f(x) = x3 + a2x2 + a4x + a6·

Further, let M = K ( VK ). Notice that the discriminant of the polyno-
mial f(x) satisfies b. = -16 Disc(f). Thus we see that K c MeL and
that
[M' K] = {2 if [L: K] = 2 or 6,
. 1 if [L : K] = 1 or 3.
Suppose first that ElK has Type III reduction, so m(E I K) = 2.
A quick perusal of Step 4 of Tate's algorithm (9.4) shows that E has a
Weierstrass equation satisfying
We claim that L/ K is at worst tamely ramified. To see this, we let 7rK

be a uniformizer for K and use the fact that v K (~) = 3 to observe that
7rM = 7r J/v'X is a uniformizer for M. Notice in particular that M/K is a
ramified extension of degree 2. Now consider the polynomial
The coefficients of 9 are in the ring of integers of M, and the discriminant

of 9 has valuation
Hence the splitting field of 9 over M, which is L, is unramified over M.

This proves that L/ K is not wildly ramified, so b(E I K) = O. We have now
computed all of the pieces in
VK('D E/ K ) - f(EIK) - m(EIK) + 1 = 3 - (2 + 0) - 2 + 1 = 0,
which completes the proof of Ogg's formula for p = 3 and Type III reduc-
tion.
If ElK has Type III* reduction, then a similar calculation shows
that VK(~) = 9 and that K(E[2]) is tamely ramified over K. So again we
find that
VK('D E/ K ) - f(E/K) - m(EIK) +1= 9- (2 + 0) - 8 + 1 = O.

We leave the details to the reader (exercise 4.54a).
For the remaining reduction types (II, III, IV, I;" IV*, III*, II*), we
are first going to show that if Ogg's formula is true for ElM, then it is also
true for E/ K. More precisely, if we write Ogg(EI K) for the quantity
Ogg(E/K) = VK('D E/ K ) - f(E/K) - m(EIK) + 1,

then we will show that
Ogg(EIM) = e(MI K) Ogg(E/ K),
where e( M / K) is the ramification index of M / K. It is clear that this equal-

ity holds if M / K is unramified, since none of the quantities in Ogg( E / K)
will change, so we only need to consider the case that M I K is ramified.
Assuming now that M/K is ramified, we have e(M/K) = 2, so the
ramification is tame. It follows that the higher ramification groups for LI K
and LIM are the same,
for all i 2: 1.
Further, M/ K is a ramified extension of degree 2, so go(L/ K) = 2go(L/M).

Using these two facts and the definition of wild part of the conductor, we
compute
6(E / K) = f=
i=l
gi(L/ K) dimlF (E[2j/ E[2jG i (L/ K»)
go(L/ K) 2
=
8~ 2go(L/M)
gi(L/M) dim
1F2
(E[2j/E[2jGi(L/M»)
= ~6(E/M).
Notice that the full conductor does not satisfy such a simple relation,
since E(E/K) = 2, and we will see that E(E/M) may be any of 0, 1,
or 2.
The next step in our proof that Ogg( L / M) = 2 Ogg( L / K) is to verify
the following table describing how various quantities change when we make
the ramified quadratic extension M/ K. (For more extensive tables, see
exercises 4.48, 4.49, and 4.50.)
Type(E/K) II IV 1*0 1*n IV* 11*

(n> 1)
Type(E/M) IV IV* 10 I2n IV IV*
2VK('D E / K ) -VM('D E / M ) 0 0 12 12 12 12
2f(E/K) - f(E/M) 2 2 4 3 2 2
2m( E / K) - m( E / M) -1 -1 9 10 11 11
Notice that the value of 2m(E/K) - m(E/M) in the last line is easy to
compute by using the first two lines and reading off the number of com-
ponents for each reduction type from Table 4.1. Similarly, the identity
6(E/K) = 6(E/M)/2 that we proved above implies that
2f(E/K) - f(E/M) = 2E(E/K) - E(E/M),
so the penultimate line of the table follows immediately from the first two
lines and the fact (9.2a) that E = 0 for good reduction, E = 1 for multi-
plicative reduction, and E = 2 for additive reduction. The verification of
the remainder of the table is now simply a matter of tracing the various
reduction types through Tate's algorithm. We will do Type IV* reduction
to illustrate the idea, and leave the other cases for the reader.
So suppose that E / K has Type IV* reduction. Then Step 8 of Tate's
algorithm gives a minimal Weierstrass for E / K satisfying
(Remember we can assume that a1 = a3 = 0.) Fix a uniformizer 7rM

for M and make the change of variables x = 7rXrx', y = 7r'ity'. This gives
a Weierstrass equation for ElM of the form
VM(a;) = 2vK(a2) - 2 ~ 2, VM(a~) = 2vK(a4) - 4 ~ 2,

vM(a~) = 2vK(a6) - 6 = 2, VM(~') = 2VK(~) - 12.
Now a quick check of Step 5 of Tate's algorithm shows that this is a minimal
equation for ElM and that ElM has Type IV reduction. Further, ~' =
M12 ~ , so
7r-
Finally, a fiber of Type IV* has seven components, and a fiber of Type IV
has three components, so 2m( ElK) - m( ElM) = 11. This completes the
verification of the Type IV* column in the above table. The other columns
may be verified similarly.
We now use this table to compute
20gg(EIK) - Ogg(EIM)
= {2VKCD E / K ) - vMCD E / M )} - {2f(EIK) - f(EIM)}
- {2m(EIK) - m(EIM)} + {2 -I}
0 - 2 - (-1) + 1 = 0 if Type(EIK)=II or IV,
_ { 12-4-9+1=0 if Type(EIK)=I(j,
- 12 - 3 -10 + 1 = 0 if Type(EIK)=I~, n ~ 1,
12 - 2 - 11 + 1 = 0 if Type(EIK)=IV* or II*.
This completes the proof that Ogg(EIM) = 20gg(EIK), so it now suffices
to prove Ogg's formula for ElM. We note from the table that ElM is of
Type IV, IV*, 10, or In, so it suffices to consider these four cases.
If ElM has Type 10 reduction, which is to say ElM has good reduc-
tion, or if ElM has Type In reduction with n ~ 1, which means ElM has
multiplicative reduction, then we are done, since we have already verified
Ogg's formula for good and multiplicative reduction.
Suppose now that ElM has Type IV reduction. Using Step 5 of Tate's
algorithm, we see that our Weierstrass equation satisfies
Let a E L be a root of the polynomial f(x) = x 3 + a2x2 + a4X + a6. Note

that the degree of LIM is either 1 or 3, and L is the splitting field of f, so
we have L = M(a). We want to use the fact that a satisfies the equation
f(a) = a 3 + a2a2 + a4a + a6 = 0
to compute its valuation. First we observe that
3VM(a) = vM(a2a2 + a4a + ac) :::: min{ vM(a2a2), vM(a4a), vAI(ac)}

:::: min {I + 2v AI (a). 2 + V 111 (n), 2} ,
so vM(a) :::: 2/3. Similarly,
which gives the opposite inequality 2/3 :::: vA[(a). Hence vA[(a) = 2/3. In
particular, this proves that L /!vI is totally ramified of degree 3, so L / A1 is
wildly ramified. Further, if we choose a uniformizer 7rA{ for AI, then
def /
7rL = 7rA[ n E L
will be a uniformizer for L, since VA[(7rL) = 1/3.

We can use 7rL to determine the higher ramification groups for L/!vI.
The Galois group G(L/M) is a cyclic group of order 3, so there is an
integer r :::: 1 such that
G(L/M) = G 1 (L/M) = ... = G r - 1 (L/M),

and
Gr(L/M) = Gr+ 1 (L/M) = ... = 1.
We want to compute this integer r. Writing G(L/M)

definition of the higher ramification groups says that
We substitute 7rL = 7rM /0' and use the fact that 7rA{ E AI to get
The extension L / 111 is totally ramified, so v L = 3v A{. This means that

vL(7rl\1) = 3 and vL(a) = vL(aO") = 2, and hence
r=vda- n O")-1.
2
Further, the exact same calculation gives r = vL(a - n a ) - 1, and since
2
the valuation is Galois invariant, we also find that r = udn a - nO" ) - 1.
Adding these three expressions for r yields
3r = vda - na) + vL(a - 00"2) + vda a - 0',,2) - 3

= VL ((a - n")(a - 0,0-2)(0'0" - aCT 2)) - 3.
Notice that a, a lT , and 0:',.2 are the three roots of f (x), so the discrim-
inant of f(x) is
2
(a - aO")(a - alT )(a IT - aO" )
2 2 )
Disc(f) = - (
We observed above that 6. = -16 Disc(f), which gives us the formula

1 1 1
r = -vdDisc(f)) - 1 = -vd6.) -1 = -VM(6.) - 1.
6 6 2
This relation is the key to proving Ogg's formula, since it relates the conduc-
tor, via the higher ramification groups, to the discriminant of the minimal
Weierstrass equation.
The non-trivial elements of G (L / AI) ad on the non-zero elements
of E[2] via a permutation of order 3, so the only element of E[2] fixed
by G(L/M) is O. This means that E[2]G i (L/AI) equals E[2] for i :::: rand
is trivial for i < r, and hence
dimIF2 (E[2]/ E[2f,(L/AJ)) = { ~ if i < r,

if i :::: T".
Using this and the value for T" computed above, we can determine the wild
part of the conductor directly from the definition:
6(E/M) = f.
9i(L/M) dimF 2 (E[2]/ E[2]G,(L/M))
9o(L(M)
1=1
=2(r-1)
= VM(6.) - 4
= vM('D E / M ) - 4.
Adding this to the tame part c(E / Ai) = 2 of the conductor gives the
relation
f(E/M) = c(E/M) + 6(E/M) = VM('D E / M ) - 2.
It only remains to recall that we are working with a curve E / AI having
Type IV reduction. This means that m( E / A1) = 3, so the last relation can
be rewritten as
VM('D E/IV!) = f(E/Ai) + 2 = f(E/Ai) +;) - 1 = f(E/1vf) + m(E/M) - 1.
This completes the proof of Ogg's formula when E /!vi has Type IV reduc-
tion. The proof for Type IV' reduction is similar, so we leave it for the
reader. 0
Corollary 11.2. Let KIQp be a local field, let ElK be an elliptic curve,
let VK('D E/ K) be the valuation of the minimal discriminant of ElK, and
let f (E I K) be the exponent of the conductor of ElK. Then
with equality if and only if ElK has reduction type la, h, or II.
PROOF. When p 2: 5, the stated inequality is quite easy to verify by a
direct calculation, since it essentially comes down to showing that if the
discriminant satisfies VK('D E/ K ) = 1, then ElK has multiplicative reduc-
tion. We proved this fact earlier (9.5a). In general, Ogg's formula (11.1)
tells us that
VK('D E/ K ) - f(EIK) = m(EIK) -12: 0,
since the number of components certainly satisfies m( ElK) 2: 1. Further,

there is equality if and only if m(EI K) = 1, which occurs exactly for
reduction types la, h, and II. 0
EXERCISES
4.1. (a) The special orthogonal group SOn is defined to be
SOn = {M E SLn : t M M = I},
where t M denotes the transpose of the matrix M. Prove that SOn is an

affine group variety.
(b) The orthogonal group On is defined to be
Prove that there is an isomorphism On ~ SOn xZ/2Z of group varieties.

4.2. Let A be the matrix A = (-~n 10 ), where In is the n x n identity matrix.
The symplectic group SP2n is defined to be
SP2n = {M E SL2n : tMAM = A}.
Prove that SP2n is an affine group variety.

4.3. Let E be an elliptic curve, and let 4> : E -> E be a morphism satisfy-
ing 4>(0) = O. Use the Rigidity Lemma 1.8 to prove that 4> is a group
homomorphism. (This provides an alternative proof of [AEC, III.4.8] not
requiring the theory of Picard groups. This proof readily generalizes to
abelian varieties of arbitrary dimension.)
Exercises 397
4.4. Let E be an elliptic curve, and let /1 : E x E ---> E be a morphism satisfying

/1(P,O) = /1(O,P) = P for all PEE.
Prove that /1(P, Q) = P + Q for all P, Q E E.
4.5. (a) Let A be a regular local ring, and let ~ be a prime ideal of A. Prove
that the localization of A at ~ is a regular local ring.
(b) Let X be a scheme, let x, y E X be points, and suppose that x is in
the closure of y. (N.B. Points of a scheme need not be closed.) If x is a
regular point of X, prove that y is also a regular point. Hence X is regular
if and only if all of its closed points are regular.
(c) With notation as in (b), give an example to show that it is possible to
have x singular and y non-singular.
4.6. Let R be a discrete valuation ring with fraction field K, residue field k,
and maximal ideal p, and let 7r be a uniformizer for R. Let XI R be the R-
scheme defined by X = SpecR[t]/(7rt).
(a) Prove that the generic and special fibers of X are given by X'1 ~ Spec K
and Xp ~ Al. Note that X'1 is smooth over K and that Xp is smooth
over k.
(b) Prove that X is not smooth over R. This shows that something like
the irreducibility condition in (2.9) is necessary.
4.7. *Let cP : X ---> S be a morphism of finite type of Noetherian schemes,
let x EX, and let s = cP( x). Prove that cP is smooth at x if and only if cP is
flat at x and the fiber X. is smooth over the residue field of S at s. (This
shows that what is really going wrong in the previous exercise is the fact
that X is not flat over R, since its fibers have different dimensions.)
4.8. Complete the proof that G(T) is a group by verifying the following two
facts, where we use the notation from (3.2).
(a) (i 0 cP) * cP = lTD 07rT for all cP E G(T).
(b) cP * ('1j; d.) = (cP * '1j;) d. for all cP, '1j;,). E G(T).
4.9. Let S be a scheme, and let G a and G m be the additive and multiplicative
group schemes respectively (3.1.2, 3.1.3).
(a) Prove that Ga(S) = r(S,C9s).
(b) Prove that Gm(S) = r(S, (9 8).
Here C9s is the structure sheaf on S, and r(S,:f) denotes the global sections
of the sheaf :f.
4.10. (a) For each integer r ;::: 1, let I-'r be the scheme
I-'r = SpecZ[T]/(Tr - 1).
Prove that there is a natural inclusion I-'r '--> G m so that I-'r is a (closed)
subgroup scheme of G m . The group scheme I-'r is called the scheme of
rth_roots of unity.
(b) Let R be a ring of characteristic p > 0, and for each integer r ;::: 1
let Up' be the R-scheme
up" = SpecR[TJ/(TP').
Prove that there is a natural inclusion up" '--> Gal R so that up' is a (closed)
subgroup scheme of GaI R '
(c) Let R be as in (b). Prove that up" and I-'p'l R are isomorphic as schemes
over R, but that they are not isomorphic as group schemes over R.
4.11. Let G be a group scheme over S.

(a) Prove that for each integer m, the multiplication-by-m map [m] : G ->
G described in (3.4) is a morphism.
(b) Prove that [-m] = [-1] 0 [m].
(c) More generally, prove that [mn] = [m] 0 [n].
(d) Prove that [m + n] = /l 0 ([m] x [n]).
(e) With notation as in (3.4), prove that [0] = 0'0 07r and that [-1] = i.
4.12. Let R be a ring, let dE R, and let Gd C A~ be the affine scheme given by
the equation
(a) Prove that the composition law
Gd XR Gd ---+ Gd,
(Xl,yI), (X2,Y2») I----> (XIX2 + dYIY2,XIY2 + X2Yl)
gives Gd the structure of a group scheme over R.
(b) Prove that Go fits into the following exact sequence of group schemes
over R:
o ----> G a / R ----> Go ----> 1'2/ R ----> 1.
(Here 1J.2jR = SpecR[T]/(T2 - 1) is the scheme of square roots of unity;
see exercise 4.1O(a).)
(c) Prove that Go is isomorphic to G a / R x Rp,2/R as group schemes over R.
(d) Let d 1 , d2 E R. Prove that Gdl is isomorphic to Gd2 as group schemes
over R if and only if there is a unit U E R* such that d 1 = u 2d2'
(e) If 2 E R*, prove that G 1 and G mjR are isomorphic as group schemes
over R.
4.13. Let K be a field of characteristic 0, and for each d E K, let Gd/ K C Ak
be the group variety
described in the previous exercise. Prove that every connected group va-
riety of dimension one defined over K is isomorphic over K to one of the
other following group varieties:
(i) The additive group G ajK .
(H) The group G d for some d E K*. (Note that G 1 is isomorphic to the
multiplicative group G m / K .)
(Hi) An elliptic curve defined over K.
4.14. Let e C 1P'~ be the arithmetic surface given by the equation
Complete the proof (4.2.2) that e is a regular scheme by verifying that e

is regular at the points x = Y = 3 = 0 and x + 66 = Y = 97 = O.
Exercises 399
4.15. Let R be a discrete valuation ring with normalized valuation v, and let W c
lP'~ be the arithmetic surface given by the Weierstrass equation
W : y2 + alxy + a3Y = X3 + a2x 2 + a4X + a6.
Let ~ and j be the associated discriminant and j-invariant.
(a) If v(~) = 1, prove that W is a regular scheme.
(b) If v(~) = 2 and v(j) ?: 0, prove that W is a regular scheme.
(The computations are simpler if you assume that 2 and 3 are units in R,
but the results are true in general.)
4.16. Let elzbe the arithmetic surface in A~ defined by the equation
e: (x3 + 4X2 + 3x - 1)y3 - (2X4 + x 3 - x 2 - 2x)l
_ (x 6 _ 3x 5 + 3x4 _ x 3 )y + 2X7 + x6 - x 5 - 2X4 = 7.
Describe the special fiber e7 of e over the point (7) E Spec Z; that is,
e
describe the components of 7 , their multiplicities, and their intersection
e
points. Draw a sketch (in ]R2) illustrating 7 . (See (4.2.4) and Figure 4.2
for a similar calculation.)
4.17. Let 7r : e -+ Spec(R) be a regular arithmetic surface over a Dedekind
domain R, let p E Spec(R), and let x E ep c e be a non-singular closed
point on the fiber of e over p. Complete the proof of Proposition 4.3 by
proving that 7r*(p) rt M~,x'
4.18. This exercise generalizes (4.4). Let R be a Dedekind domain with fraction
field K, and let X be a "nice" scheme over R whose generic fiber XI K is
a smooth, projective variety. (Here "nice" has the same meaning as in the
definition of arithmetic surface; see §4.)
(a) If X is proper over R, prove that X(K) = X(R).
(b) Suppose that X is a regular scheme, and let XO c X be the largest
subscheme of X with the property that XO is smooth over R. Prove
that X(R) = XO(R).
4.19. Let R' I R be an extension of discrete valuation rings with maximal ide-
als p, p', fraction fields K, K', and residue fields k, k' respectively. Sup-
pose that K' I K is a finite extension. Prove that R' is the localization of a
smooth R-scheme if and only if p' = pR' (Le., R' IRis unramified).
4.20. Let R be a discrete valuation ring with fraction field K, and let ElK be
an elliptic curve given by a Weierstrass equation
E: y2 + ajxy = x 3 + a6 with al E R* and a6 E R.
Let W c lP'~ be the R-scheme defined by this Weierstrass equation.
(a) Prove that W is a regular scheme if and only if v(a6) ::::: 1.
(b) Prove that W is smooth over R if and only if v(a6) = O.
4.21. Let ElK, WIR and WOIR be as in the statement of Theorem 5.3, where
we assume that we start with a minimal Weierstrass equation for ElK.
(a) If ElK has split multiplicative reduction, prove that the special fibers
of WO and G m / R are isomorphic as group schemes over the residue field
of R.
(b) If ElK has additive reduction, prove that the special fibers of WO
and G a / R are isomorphic as group schemes over the residue field of R.
(c) Give a similar description of the special fiber of WO in the case that
the curve ElK has non-split multiplicative reduction.
4.22. Let R be a discrete valuation ring with fraction field K, let ElK be an
elliptic curve, and choose a Weierstrass equation for ElK with coefficients
in R,
Let W c lP'h be the R-scheme defined by this Weierstrass equation, and let
f1 : W x W --- .. W
be the rational map induced by the addition law on the generic fiber. Define
affine subsets of W by
Waif = {Z 1= O} and W~1f = {Y 1= O}.

Prove that f1 is a morphism when restricted to each of the following sets:
(a) Waif XR W~If.
(b) W~1f XR Waif.
(c) W~1f XR W~If.
(The formulas will be easier if you assume that 2 and 3 are units in Rand
take a Weierstrass equation of the form y2 Z = X 3 + AX Z2 + B Z3. We
described the behavior of f1 on Waif XR Waif during the proof of (5.3). This
exercise is asking you to complete the proof of (5.3).)
4.23. Let R be a Dedekind domain with fraction field K, let ElK be an elliptic
curve, let eI R be a minimal proper regular model for ElK, and let cI R be
the largest subscheme of el R which is smooth over R. Note that cl R is a
Neron model for ElK, so in particular c is a group scheme over R (6.1).
(a) Let P E E(K) ~ c(R), and let Tp : C --t C be translation-by-P (3.3).
Prove that Tp extends to an R-morphism e --t e.
(b) Prove that every automorphism a: : ElK --t ElK extends to an R-
morphism e --t e.
(c) 'Prove that the group law c XR C --t c extends to give a group scheme
action c x R e --t e.
(d) Prove that in general the group law C XR C --t C does not extend to
give an R-morphism e XR e --t e.
4.24. Let R be a discrete valuation ring with fraction field K, residue field k, and
residue characteristic p. Let ElK and E' I K be elliptic curves, and let cI R
and C'IR be Neron models for ElK and E'IK respectively. Let CPK: E--t
E' be an isogeny of degree m 2 1 defined over K. Assume that either p
does not divide m, or else that ElK does not have additive reduction.
(a) 'Prove that CPK extends to an R-morphism CPR : C --t c'.
(b) Prove that CPR is a homomorphism of R-group schemes.
(c) Prove that the restriction of CPR to the special fiber is a finite mor-
phism CPk : elk --t e'lk which maps the identity component of e to the
identity component of e'.
(d) Prove that there is an R-morphism ¢R : C' --t C with the property that
the composition ¢ROCPR : C --t C is the multiplication-by-m map on C (3.4).
This generalizes the construction of the dual isogeny for elliptic curves over
fields [AEC, III §6J.
Exercises 401
4.25. Let R be a Dedekind domain with fraction field K, let E / K be an elliptic

curve, and let £ / R be a N eron model for E / K. Prove that the connected
component of £/ R as described in (6.1.2) is a subgroup scheme of £/ R.
4.26. Let R be a Henselian discrete valuation ring with valuation v. Let I(x) E
R[x) be a monic polynomial, and let a E R be an element with the property
that
v(f(a)) > 2v(f'(a)).
Prove that there is a unique element a E R satisfying
v(a - a) > v(J'(a)) and I(a) = O.

(Note the strict inequalities.)
4.27. Let R be a discrete valuation ring with maximal ideal p and residue field k,
let 11, ... , 1m E R[Xl, ... , Xn], let
x= Spec R[Xl, ... , x n )/(/1, ... , 1m)

be the scheme defined by the Ii'S, and let
J = J(Xl, ... ,xn ) = (81;/oXjh$.i$.m,1$.j$.n

be the associated Jacobian matrix.
(a) Prove that the generic fiber of X is empty if and only if some power
of p is contained in the ideal (/1, ... , 1m).
(b) Let a = (al, ... , an) E :X(k) be a point on the special fiber of X. Prove
that Xp is smooth over k at a if and only if the matrix J(a) satisfies
rankJ(a) = n - dimXp.
(c) Assume that X is irreducible, reduced, and has non-empty generic fiber.
Suppose further that the ring R is Henselian, and let a = (al, ... , an) E
An(R) be a point satisfying
/1 (a) == ... == 1m ( a) == 0 (mod p) and rank J(a) = n - dimXp.
Prove that there exists a (unique) point a E An(R) such that
/1 (a) = ... = Im(a) = 0 and a == a (mod p).
This result is a multi-variable version of Hensel's lemma. It implies the

surjectivity of the reduction map for smooth schemes over Henselian rings.
(Hint. First prove (c) under the assumption that R is complete.)
4.28. Let R be a discrete valuation ring with residue field k, and let Rh and R sh
be the Henselization and strict Henselization of R respectively (6.5).
(a) Let R' be a Henselian discrete valuation ring, and let i : R -> R'
be a local homomorphism. Prove that there is a unique local homomor-
phism Rh --+ R' extending i.
(b) Let Rlf be a strictly Henselian discrete valuation ring with residue
field kif, let i : R --+ Rlf be a local homomorphism, and let u : kif --+ k
be a k-homomorphism. Prove that there exists a unique local homomor-
phism R sh -> Rlf which extends i and which induces the map u on the
residue fields.
4.29. 'Let K be an algebraically closed field, let CIK be a curve of genus 9 ~ 1,

and let V be the g-fold symmetric product of C. Note that the points of V
can be naturally identified with the positive divisors of degree 9 on C.
(a) Fix a basepoint Po E C. Prove that there is a rational map J-t : V x V -+
V determined by the property
J-t(x, y) ~ x +y - g(Po).
(Here ~ denotes linear equivalence of divisors.)

(b) Prove that J-t is a normal law, and hence from (6.9) that there is a
group variety J I K associated to J-t.
(c) Prove that the map V -+ J is a morphism, and deduce that J is proper
over K.
(d) Prove that the map of sets
X f---+ class[x - g(Po )],
induces an isomorphism of groups J(K) -+ Pic°(C).

The group variety J is an abelian variety called the Jacobian variety
of C. This construction of the Jacobian variety is due to Weil [3]. For a
further discussion, see the proof sketch of Proposition III.2.6(b).
4.30. Let R be a strictly Henselian discrete valuation ring with fraction field K.
Let eI R be a group scheme over R whose generic fiber ElK is an ellip-
tic curve. Prove that el R is a Neron model for ElK if and only if the
inclusion e(R) -+ E(K) is a bijection.
4.31. Let R be a discrete valuation ring with uniformizing element 1[" and alge-
braically closed residue field k. Assume that char(k) i- 2,3,5. Let e c A~
be the affine scheme defined by
that is, e = SpecR[x,yl/(y2 - x 5 - 1["2).

(a) Show that e is regular except at the one point 1[" = x = y = 0 on the
special fiber.
(b) Compute the blow-up of e at the singular point 1[" = x = y = 0 as
explained in (7.7). Show that the resulting scheme is still not regular.
(c) Continue blowing up until you get a regular scheme. Draw a picture
of the special fiber similar to the diagrams in Figures 4.3 and 4.4.
(d) Repeat (a), (b), and (c) for the arithmetic surface y2 = x 5 + 1[" 7 •
4.32. (a) For each of the Kodaira-Neron reduction types (8.2), compute the
intersection incidence matrix of the special fiber.
(b) Show that each of the incidence matrices in (a) has determinant o.
(c) Let !vI be the matrix obtained by taking anyone of the incidence matri-
ces in (a) and deleting a row and column corresponding to a multiplicity-l
component. Show that det(M) is equal to plus or minus the number of
multiplicity-l components on the special fiber. Equivalently, Idet(M)I is
the order of the group of components of the Neron model.
Exercises 403
4.33. Let R be a discrete valuation ring, and let e/ R be a proper regular model
of a curve of positive genus. Generalize part (b) of the previous exercise
by showing that the incidence matrix of its special fiber has determinant O.
(For a generalization of part (c), see Raynaud [1].)
4.34. Let R be a discrete valuation ring with maximal ideal p, fraction field K,
and algebraically closed residue field k. Let C / K be a non-singular projec-
tive curve of genus 9 2: 1, and let e/ R be a minimal proper regular model
for C / K. Suppose that the special fiber ep of e contains a configuration
with t components of the form shown in Figure 4.6(b), where each of the
illustrated components satisfies Pa(F;) = 0 and F? = -2.
(a) Prove that t :s:
4g + 2.
e
(b) If t = 4g + 2, complete the picture of p • In particular, show that ep
has exactly 4g + 5 components. (For 9 = 1, you'll get a fiber of Type 11*.)
4.35. Let R be a complete discrete valuation ring with fraction field K and al-
gebraically closed residue field k of characteristic P of- 2,3. Let E / K be
an elliptic curve with additive reduction, and let Eo(K) be the subgroup
of E(K) consisting of points with non-singular reduction (9.2).
(a) Prove that Eo(K) is uniquely divisible by 2 and 3.
(b) Prove that E(K)/ Eo(K) is killed by 12.
(c) Prove that the natural map E(K)[12] -> E(K)/Eo(K) is an isomor-
phism.
4.36. (a) Let C 4 , C6 E Z be integers with C] - cl of- O. Prove that there exists
a Weierstrass equation
with coefficients al, a2, a3, a4, a6 E Z satisfying C4 = C4 and C6 = C6 if and

only if one of the following two conditions is true:
(i) ord 3 (C6) of- 2 and C6 == -1 (mod 4).
(ii) ord 3 (C6) of- 2, ord2(C4) 2: 4, and C 6 == 0 or 8 (mod 32).
(b) Use the criteria in (a) to devise a quick algorithm to check whether a
given Weierstrass equation with integer coefficients is a minimal Weierstrass
equation.
(c) Generalize the criteria in (a) to an arbitrary field K/Q.
4.37. Let R be a discrete valuation ring with fraction field K, let ElK be an
elliptic curve given by a Weierstrass equation
with coefficients in R, and assume that 7r!a3, a4, 7r 2!a6, and 7r f b2. Resolve
the singularity on the special fiber by a sequence of blow-ups and show that
the special fiber is of Type In with n = v(.6.).
4.38. Let £ be a prime, let £' = £ if £ i- 2, and let £' = 4 if C = 2. Prove that the
group
{M E GL2(Ztl : M == 1 (mod £')}
contains no elements of finite order.
4.39. This exercise generalizes the previous exercise. Let K/Qp be a p-adic field
with ring of integers R, maximal ideal p, and normalized valuation VK.
(a) Suppose that there is a matrix M E GLn(R) of exact order m ~ 2
satisfying
Prove that
and
(b) If r > VK(p)/(p - 1), prove that the group

{M E GLn(R) : M == 1 (mod pr)}
contains no elements of finite order other than the identity element.
4.40. Let E/ K and E' / K be elliptic curves defined over a local field, and let
cp : E --> E' be a non-constant isogeny defined over K. Prove that
e(E/K) =e(E'/K), 6(E/K) = 6(E'/K), f(E/K) = f(E'/K).
Notice that this generalizes the assertion [AEC, VII.7.2) that E and E'
either both have good reduction or both have bad reduction, since (10.1)
says that good reduction is equivalent to e = O.
4.41. Let K/Q3 be a 3-adic field, let E/K be an elliptic curve, let l =I 2 be a
prime, and let L = K(E[l]).
(a) Prove that the first higher ramification group G 1 (L/ K) is either trivial
or a cyclic group of order 3. (Hint. Show that GI(L/ K) is independent of l,
and then take l = 2.)
(b) Prove that G1(L/ K) is a cyclic group of order 3 if and only if E/ K
has reduction type II, IV, IV·, or n·.
4.42. Let K /Q2 be a 2-adic field, let E / K be an elliptic curve with potential
good reduction, let l ~ 3 be a prime, let L = K(E[lJ) , and let G1(L/K)
be the first higher ramification group of L/ K. Prove that
G1(L/K) e! {1} or Z/2Z or Z/4Z or Hs ,
where Hs is the quaternion group of order 8. (Hint. Show that GI(L/K)

is independent of l, and then take l = 3.)
4.43. For each of the following elliptic curves E/Q2' let L = Q2(E[3]) and
compute the first higher ramification group G 1 (L/Q2).
(a) E: y2 + 2y = x 3 .
(b) E: y2 + 2xy + 8y = x 3 •
(c) E: y2 + 2y = x 3 + 2x.
4.44. Let K be a number field. Prove that for any constant B there are only
finitely many elliptic curves E/ K whose conductor fE/K satisfies
Exercises 405
4.45. Let Lj K be a finite extension of local fields as described at the beginning

of §1O, let 1[L be a uniformizer for L, and define an index function
iL/K : G(LjK) --+ Z U {oo}, i L/ K (17) = vL(7rL - 7rL).
(a) Prove that i L/K(17) ~ i + 1 if and only if 17 E Gi(LjK).

(b) Prove that i L/ K (7177- 1) = i L/ K (17) for all 17,T E G(LjK).
(c) Prove that iL/K(177) ~ min{i L/ K (17),iL/K(T)} for all 17,7 E G(LjK).
4.46. We continue with the notation from the previous exercise, with the addi-
tional assumption that the extension Lj K is totally ramified. For basic
material on the representation theory used in this exercise, see for example
Serre [7]. The Arlin character ArL/K and the Swan character SWL/K are
defined to be the functions
ArL/K : G L/ K --+ Z, SWL/K : G L/ K --+ Z,
ArL/K(17) = -i L/ K (17) if 17 oF 1, SWL/K(17) = 1 - i L/ K (17) if 17 oF 1,
ArL/K(l) = L i L/ K (17), SWL/K(l) = L(iL/K(17) - 1).
u#1 u#1
(a) A function 'I/J on G(L j K) is called a class function if 'I/J( T17T- 1) = 'I/J( (7)
for all 17,7 E G(Lj K). Prove that ArL/K and SWL/K are class functions.
(b) *For any pair of functions 'l/J1,'l/J2 on G(LjK), define
('l/J1,'l/J2) = go(LK) L 'l/Jl(17)'l/J2(17- 1).

uEG(L/K)
If X is the character of an irreducible representation of G(Lj K), prove that

(X, ArL/K) and (X,SWL/K) are non-negative integers.
(c) Replace K by its maximal unramified extension, and let E j K be an
elliptic curve with integral j-invariant. Let Lj K be a finite Galois extension
such that E has good reduction over L. Further let XE be the character of
the representation of G(Lj K) on Te(E). Prove that XE takes values in ;;Z
and is independent of R.
(d) *Continuing with the assumptions from (c), prove that
6(EjK) = (SWL/K,XE) and f(EjK) = (ArL/K,XE).
Deduce that 6(EjK) and f(EjK) are integers that are independent of R.
This proves (1O.2c) in the case that E has potential good reduction.
(e) If EjK has non-integral j-invariant, prove that 6(EjK) and f(EjK)
are integers and are independent of.e, thus completing the proof of (1O.2c).
(Hint. For (e), use the isomorphism E(K) ~ K* jqZ described in (V.5.3)
and exercise 5.11.)
4.47. Let KjQp be a p-adic field with p ~ 5, let VK be the normalized valuation
on K, and let E j K be an elliptic curve. Prove that there exists a minimal
Weierstrass equation for Ej K with at = a2 = a3 = 0, and with a4, a6,
and ~ as described in the following table:
Type 10 In II III IV 10 1*n IV* III* II'
vK(a4) =0 >1 =1 >2 =2 =2 >3 =3 >4
vK(a6) =0 =1 >2 =2 =3 =3 =4 >5 =5
VK(~) =0 =n =2 =3 =4 =6 =n+6 =8 =9 =10
406 IY. The N eron Model
4.48. Let KIQ3 be a 3-adic field with normalized valuation VK, and let ElK be
an elliptic curve. Prove that there exists a minimal Weierstrass equation
for ElK with al = a3 = 0, and with az, a4, a6, and .6. as described in the
following table:
Type 10 In II III IY 1(; 1*n IY' III' II'

vK(a2) =0 >1 >1 >1 >1 =1 >2 >2 >2
vK(a4) 21 21 =1 22 22 2 [nt5] 23 =3 24
vK(a6) >1 =1 >2 =2 >3 >n+3 =4 >5 =5
VK(.6.) =0 =n >3 =3 >5 =6 =n+6 >9 =9 >11
4.49. Let KIQp be a p-adic field with p 2 3, let VK be the normalized valuation
on K, and let ElK be an elliptic curve with Type I~ reduction.
(a) If n = 0, prove that vdj(E)) 2 0 and VK('])EjK) = 6.
(b) If n 21, prove that vK(j(E)) = -n and VK('])EjK) = n + 6.
(c) Let LI K be a tamely ramified extension with ramification degree e =
e(LI K). Prove that E / L has Type Ine reduction if e == 0 (mod 2), and ElL
has Type I~e reduction if e == 1 (mod 2).
(d) Give an example of an elliptic curve EIQ2 with Type I~ reduction
satisfying n 21 and v2(j(E)) 2 O. This shows that (b) is not true for p = 2.
What is the largest possible value of n in this situation?
4.50. Let KIQ3 be a 3-adic field with normalized valuation VK, and let LIK be
a tamely ramified extension with normalized valuation VL. Let e = e(LIK)
be the relative ramification degree, so v L = ev K. Let E / K be an elliptic
curve whose reduction type is one of II, IY, IY', or II'.
(a) Prove that the reduction type of ElL is given by the following table:
Type(EIK) e == 1 (6) e == 2 (6) e == 4 (6) e == 5 (6)

II II IY IY' II'
IY IY IY' IY IY'
IY' IY' IY IY' IY
II' II' IY' IY II
(b) Let ']) E j K and ']) E j L be the minimal discriminants of ElK and ElL
respectively. Prove that the value of the difference
is given by the following table:
Type(EIK) e == 1 (6) e== 2 (6) e== 4 (6) e == 5 (6)

II 2e - 2 2e - 4 2e - 8 2e -10
IY 4e -4 4e- 8 4e -4 4e- 8
IY' 8e - 8 8e-4 8e - 8 8e - 4
II' 10e - 10 10e - 8 lOe-4 lOe - 2
Exercises 407
4.51. This exercise illustrates how wild ramification can cause the reduction type
of an elliptic curve to change in an irregular fashion. It may be compared
with the previous exercise, which dealt with the tamely ramified case.
Let KIQ3 be a 3-adic field with normalized valuation VK, and let ElK
be an elliptic curve given by a Weierstrass equation
Let LI K be a ramified extension of degree 3, so LI K is wildly ramified.

(a) Prove that ElK has Type II reduction.
(b) If vK(3) = 1 (i.e., if KIQ3 is unramified), prove that ElL has Type III-
reduction.
(c) If VK (3) = 2, prove that ElL has Type 10 (i.e., good) reduction.
(d) IfvK(3) = 3, prove that ElL has Type III reduction.
(e) Try to find a general formula for the reduction type of ElL. Does the
reduction type of ElL depend only on vK(3)?
4.52. We continue with the notation from the previous exercise, so KIQ3 is a
3-adic field and ElK is an elliptic curve
Prove that the conductor exponent of ElK is f(EI K) = 2+3vK(3). Notice

that (10.4) says that this is the largest allowable conductor exponent for
an elliptic curve defined over a 3-adic field.
4.53. Let KIQ2 be a 2-adic field with normalized valuation VK, and let ElK be
an elliptic curve given by a Weierstrass equation
(a) Prove that the equation given for E is a minimal Weierstrass equation
and that VK('D ElK) = 6VK(2) + 3.
(b) Prove that E has Type III reduction and that the special fiber of E
has two components.
(c) Prove that the conductor exponent of ElK is f(EIK) = 2 + 6VK(2).
Notice that (10.4) says that this is the largest allowable conductor exponent
for an elliptic curve defined over a 2-adic field.
4.54. This exercise asks you to verify two cases of Ogg's formula (11.1) that were
not completed in §11. Let KIQ3 be a 3-adic field, let ElK be an elliptic
curve, and let L = K(E[2]).
(a) If ElK has Type III- reduction, prove directly that LIK is a tamely
ramified extension. Use this to verify Ogg's formula for ElK.
(b) Let M = K(VK), and suppose that ElM has Type IV- reduction.
Give a direct proof of Ogg's formula in this situation. (Hint. Mimic the
proof for Type IV reduction given in §11.)
CHAPTER V
Elliptic Curves over Complete Fields
Every elliptic curve E IC admits an isomorphism C* I qZ ~ E( q by complex

analytic functions, and we have seen amply demonstrated in Chapters I
and II the importance of such uniformizations. In this chapter we are
going to study uniformizations over other complete fields such as lR and
finite extensions KIQp. We begin in §1 with a brief review of the relevant
formulas over C, and then in §2 we use the complex uniformization to
investigate elliptic curves over R
We next turn to elliptic curves defined over p-adic fields KIQp. Tate [9]
has shown that for every q E K* with Iql < 1 there is an elliptic curve Eql K
and a p-adic analytic isomorphism K* IqZ ~ Eq(K). In §3 we will describe
the Tate curve Eq and prove all of its main properties except for the sur-
jectivity of the map K* ---> Eq(K), which we reserve for §4. Tate has also
shown that every elliptic curve ElK with non-integral j-invariant is iso-
morphic, possibly over a quadratic extension of K, to some E q . We will
prove this result and describe the necessary twisting in §5, and then in §6
we will give some applications, including Serre's proof that an elliptic curve
with complex multiplication has integral j-invariant.
§1. Elliptic Curves over C
We have already discussed elliptic curves and elliptic functions over C in

some detail; see [AEC VI] and (I §§5~8). The purpose of this section is to
gather and rewrite in a convenient form the formulas we will use later in
this chapter when we study elliptic curves over lR and over p-adic fields.
Let EIC be the elliptic curve corresponding to the normalized lattice
AT = ZT + Z for some T E H.
We know that E(C) ~ CI AT) the isomorphism being given in terms of the
Weierstrass gJ-function and its derivative. As in (I §6), it is convenient to
let
§l. Elliptic Curves over C 409
Note that since Im(T) > 0, we have iqi < 1. There is a complex analytic
isomorphism
Z t-----+ U = e 27riz ,
and we can use the q-expansions from (I §§6,7) to explicitly describe the
isomorphism E(C) ~ C* fl--.
The elliptic curve E has the Weierstrass equation
From (1.7.3.2) we have
1 1
(271"i)492(T) = 12 [1 + 240S3(q)] ,
1 1
(271"i)693(T) = 216 [-1 + 504s 5 (q)] ,
where in general
(For the second equality, see exercise 5.1. Here O"k (n) = :Ldln d k as usual.)
We've collected the powers of 271"i as indicated to make it easier to eliminate
them.
Next, the isomorphism
C* fl' ---+ E(C)

U t-----+ (p(u, q), S:/(u, q))
is given by the power series described in (1.6.2) and (1.6.2.1):
(Note that p' is the derivative of p with respect to z, where u = e 27riz .)

Jacobi's formula (1.8.1) says that the discriminant of the Weierstrass
equation for E has the product expansion
6.(T) = 92(T)3 - 2793(T)2 = (271"i)12q II (1- qn)24,

n~l
410 V. Elliptic Curves over Complete Fields
and the j-invariant of E is given by the series (1.7.4)
j(q)
1
= - + 744 + 196884q + ... = - +
1
L c(n)qn with c(n) E Z.
q q n?-O
It is convenient to make a change of variables, partially to remove the

powers of 27ri and partially to eliminate the powers of 2 and 3 appearing
in the denominators of the series for g2, g3, and 8J. Thus we let
1 ,1
(27ri)2X = X + 12'
1 "
(27ri)3 Y = 2y + x ,
which gives the new Weierstrass equation
with
1 1 1
a4 = -4: . (27ri)4 g2 (T) + 48'
1 1 1 1 1
a6 = -4: . (27ri)6 g3 (T) - 48 . (27ri)4 g2 (T) + 1728'
Now using the seriefl for g2, g3, 8J, 8J' and doing a little algebra, we find we
have proven virtually all of the following result.
Theorem 1.1. For U, qE C with Iql < 1, define quantities
Eq : y 2 + xy = x3 + a4(q)x + ao(q).
(a) Eq is an elliptic curve, and X and Y define a complex analytic isomor-
phism
if U tf. q'L,
u
if U E q'L.
§1. Elliptic Curves over C 411
(b) Written as power series in q, both a4(q) and a6(q) have integer coeffi-
cients; that is, a4(q), a6(q) E Z[q]'
(c) The discriminant and j-invariant of Eq are given by the formulas
6.(q) = -a6 + a~ + 72a4a6 - 64a~ - 432a~

= q II (1 - qn)24 E Z[q],
n;:,l
. 1
J(q) = - + 744 + 196884q + ...
q
= - +
1
L 1
c(n)ql1 E - + Z[q].
q n;:,O q
(d) For every elliptic curve EIC there is a q E C* with Iql < 1 such that E
is isomorphic to E q •
PROOF. The discussion given above has proven all of (a), (b), and (c)
except for the minor point that the power series for a6(q) has integer coef-
ficients. Since
it suffices to observe that
5d a + 7d 5 == 0 (mod 12) for all d E Z.

(Notice that we used this same fact in the proof of (1.7.4a).) This completes
the proof of (c).
Finally, to prove (d), we note that the uniformization theorem (1.4.4)
says that every elliptic curve E IC is isomorphic to CI A for some lattice A.
Then the change of variables used above transforms the \;Veierstrass equa-
tion for E into an E q .
o
Remark 1.2. It is sometimes convenient to rewrite 81 (q) in the alternative
form
To check that these two expressions for 81 (q) are the same, we substi-
tute T = qn into
T
(1 - T)2 = T dT
d(l) d
1 _ T = T dT L T
m
= L mT
m
m;:,O m;:,l
and sum over n ?: 1, which yields
For later reference, it will be helpful to rewrite the formulas
( ) _ () __ (]'(z + a)(]'(z - a) I (]'(2z)

f9 z f9 a - (]'(z)2(]'(a)2 and f9 (z) = - (],(z)4
from (1.5.6) in terms of X(u, q) and Y(u, q). For this purpose, we introduce
a normalized theta function.
Proposition 1.3. Define a normalized theta function 8(u, q) by the for-
mula
8(
U, q
) = (1 _
U
) II (1 - qn u)(l - qnu- l )
(1 _ n)2 .
n~l q
(a) 8(u,q) converges for all u,q E C* with Iql < 1 and satisfies the func-
tional equation
1
8(qu,q) = --8(u,q).
u
(b) 8 is related to the Weierstrass (]' function by the formula
1 1 (1) 2 .
(]'(u,q) = --2.e 2 '7 z e- 7nz 8(u,q),
'In
where u = e271'iz, q = e2 71'iT, and 1](1) is the quasi-period associated to the

period 1 in the lattice ZT + Z.
(c) 8 is related to the functions X(u,q) and Y(u,q) described in (1.1) by
the formulas
(i)
(ii)
PROOF. (a) Since Iqnul < 1 and Iqnu-ll < 1 for all sufficiently large n, it
is clear that the product defining 8 converges. Then replacing u by qu and
renumbering the factors in the product gives
§2. Elliptic Curves over IR 413
(b) This is immediate from the definition of () and the product formula
for a given in (1.6.4).
(c) From above, X and Yare related to g;J and g;J' by the formulas
and
Hence writing Ul = e27fizl and U2 = e27fiz2, we find (dropping the q from

our notation)
1
X(ud - X(U2) = (27ri)2 (g;J(Ul) - g;J(U2))
1 a(ulu2)a(ului 1)
from (1.5.6a)
- (27ri)2 a(ul)2a(u2)2
= _e!1/(l)«(Zl +Z2)2+(Zl-Z2)2_2z~ -2z~)
. e-7fi((Zl+Z2)+(Zl-Z2)-2z1-2z2). 8(UIU2)e(UIUi 1 )
()( ud 2()( U2)2
from (b)
This proves (i), and (ii) is proven by the similar calculation
2Y(u) + X(u) = (2:i)3 g;J'(u)

a(u 2 )
1
from (1.5.6b)
- (27ri)3 a(u)4
= _e!1/(1)«(2z)2- 4z2) . e-7ri(2z-4z) ()(u 2 ) from (b)
()( U)4
8(u 2)
= -u 8(u)4'
o
§2. Elliptic Curves over ffi.
The uniformization theorem (1.4.4) says that every elliptic curve defined
over C is analytically isomorphic to C* / qZ for some q = e 27fir • Since an
elliptic curve defined over ffi. is automatically defined over C, it has such a
model. We begin by describing a set of r's which classifies elliptic curves
over ffi. up to C-isomorphism.
Proposition 2.1. Let E/'&. be an elliptic curve. Then there exists a

unique T in the set
e= {it : t :::: 1 } U { eiO -I3-

T -
- 2
v'3}
such that
j(T) = j(E).
(The set e is illustrated in Figure 5.1.)
PROOF. First we check that j(e) c RIfT = it or T = ~ + it with t E '&.,
then q E '&., so the q-expansion (1. 7.4b)
c(n) E Z,
shows that j(T) E R Next, for any 7 we have
so in general
J-:--(
) _ . ( -1712)
T -J - .
T
Hence for T = eiO we have
This proves that j (e) c R

Next we observe that
lim j(it)
t-->oo
= lim q-l +
q-->O+
L c(n)qn = +00,
lim j (-21 + it) = lim q-l + L c(n)qn = -00.
t-->oo q-->O-
By continuity, we conclude that j(e) = R (Note that j : e -+ '&. is

continuous, since j : H -+ C is holomorphic.) Finally, (1.1.5b) and (1.4.1)
imply that j : e -+ '&. is injective, which concludes the proof that the
map j : e -+ '&. is a bijection. 0
Proposition 2.1 completely describes all C-isomorphism classes of ellip-

tic curves defined over R However, since'&. is not algebraically closed, it is
possible to have more than one '&.-isomorphism class in each C-isomorphism
class. For a given E /'&., these other curves are called the twists of E
(see [AEC X §5]). Our next result classifies these twists.
The Set e for Which j(7) E IR

Figure 5.1
Proposition 2.2. (a) Let E/IR be an elliptic curve. Then the C-iso-
morphism class of E contains exactly two IR-isomorphism classes. (In the
notation of [AEC X §5], Twist((E,O)/IR) ~ {±1}.)
(b) More precisely, define an invariant I'(E/IR) E {±1} by the rule
sign(c6), if j =J 1728 (i.e., if C6 =J 0),

{
')'(E/IR) = . ()
sign c4,
'f'
1 J =
1728 (.J.e., J'f C6 = 0) .
(Here C4 and C6 are the usual quantities associated to some Weierstrass

equation for E /IR.) Let E /IR and E' /IR be elliptic curves. Then
E ~ E' over IR {==} j(E) = j(E') and I'(E/IR) = ')'(E' /IR).
PROOF. From [AEC X.5.4], the twists of E are in one-to-one correspon-

dence with the elements of the group
where n = # Aut(E) E {2, 4, 6}.
Since n is even, IR* /IR*n ~ {±l} has two elements. This completes the
proof of (a).
For (b) we use the more precise description of the twist provided by
the second part of [AEC X.5.4]. If E/ffi. is given, and if E' /ffi. is the twist
of E/ffi. corresponding to some D E ffi.* /ffi.*" , then [AEC X.5.4] says that
c6(E') = D3C6(E) if j(E) '" 0, 1728,

c6(E') = DC6(E) if j(E) = 0,
c4(E') = DC4(E) if j(E) = 1728.
Hence in all cases
. ,(E/ffi.)
slgn(D) = ,(E' /ffi.)'
so the ratio ,(E /ffi.)h(E' /ffi.) determines whether E' is isomorphic to E or
to its non-trivial twist. D
Remark 2.2.1. For an analogous result over p-adic fields, see (5.2).
By combining Propositions 2.1 and 2.2, we can now give a convenient
set of q's which completely classifies all ffi.-isomorphism classes of elliptic
curves. For any q = e 2 -rr:iT, we let Eq be the elliptic curve
where a4(q) and a6(q) are the power series described in (1.1), and we let
¢(u) = (X(u,q),Y(u,q)),
be the C-analytic isomorphism from (1.1).

Theorem 2.3. Let E /ffi. be an elliptic curve.
(a) There is a unique q E ffi. with 0 < iqi < 1 such that
(i.e., E is ffi.-isomorphic to Eq).

(b) Composing the isomorphism from (a) with the map dJ described above,
we obtain an isomorphism
which commutes with complex conjugation, that is, 7./J is defined over R In
particular,
is an ffi.-analytic isomorphism.
PROOF. Note first that if q E ffi., then the series (1.1) imply that a4(q)
and a6(q) are in ffi., so Eq is defined over R We want to start by using (2.1)
to find a T with j(T) = j(E). As noted during the proof of (2.1), the
points T = it and T = ~ + it give real values of q. However, for T = eiO we
do not have q E 1Ft So we use the transformation
-1 iO 1 i ()
a(T) = - - which satisfies a( e ) = - + - cot -.
T-1 2 2 2
Thus a E r(l) yields a bijection
a: { eiO : -7r < () < -7r }

3- -2
~
--> {I + . 1<
-
2
zt : - t < -V3 } .
2- - 2
Since j(aT) = j(T), we conclude from (2.1) that there is a unique T in the
set
{it : t ~ 1} U { ~ + it :t > ~}
such that j ( T) = j (E). Note that in the second set we do not allow t = ~,
since this would give the duplicate value
j (~+ ~i) = j(a(i)) = j(i).
The set of q = e 27riT E 1R. with 0 < Iql < 1 corresponds bijectively with
the T'S in the set
'J = {it : t > 0} U { ~ + it t > 0} .
Further, the transformations
1 T-1 -1
ST = --, {3T = 2T - l' aT = - -
T T-1
give identifications
S:{it: t>l}--=::"'{it: t<l},
{3: {2"I+ it : t > 2" I} ~ {I2" +

--> it :
I} '
t < 2"
. 1 1.
a: z ~ - +-2-
2 2
(See Figure 5.2.)

418 v. Elliptic Curves over Complete Fields
o 1
2
A Set of 7 Giving All q E lR with 0 < Iql < 1

Figure 5.2
Since j({7) = j(7) for any, E SL2(Z), we see from above that the
map
j:'J~lR
is exactly two-to-one. From (2.2a) there are exactly two lR-isomorphism

classes of elliptic curves with a given j-invariant. Hence to complete the
proof of (a) we must check that if 7,7' E 'J are distinct points with j(7) =
j(7'), then Eq and Eq, are non-trivial twists of one another. To do this we
will use (2.2b).
The change-of-variable formulas for Weierstrass equations [AEC III §1]
imply that
and
for some u E C*. (In fact, our explicit formulas imply that u = (27ri)-1,
but for our purposes it suffices to know that u does not depend on 7.) Now
suppose that
a7 +b
7
,
=--
cr+d
for some (~ : ) E SL2(Z).
Then using the l'-invariant defined in (2.2b) and the fact that g2 and g3
are modular forms, we find that
I'(Eql/JR) = sign
I'(Eq/JR)
(cC6(q)
6(q')) = sign (93( T')) = sign{ (CT + d)6}
g3(T)
if jeT) =1= 1728.
Similarly,
I'(Eq)JR) = sign (C4(q')) = sign
I'(Eq/JR) C4(q)
(92( T'))
g2(T)
= sign{ (CT + d)4}
if jeT) = 1728.
We must show that for the T'S and T"S described above, all of these signs
are -1. This requires checking several cases.
lease I: T = it, t > 1. T' = ST = -~.I

In this case j (T) =1= 1728, and
(CT + d)6 = (it)6 = _t 6 < O.
T -1 I
lease II: T = -1 + it, t > 1. T' = (3T = - -.
2 2T-1
Again we are in a case in which j (T) =1= 1728, and
(CT + d)6 = (2T - 1)6 = (2it)6 = -64t6 < O.
lease III: T=
.,
2. T = 007
-1
= T _ 1 = 2 + 22.
1 1. I
This is the case that j (T) = 1728, and
(CT + d)4 = (T - 1)4 = (i - 1)4 = -4 < O.
Hence in all cases Eq and Eql are distinct twists, which completes the
proof of (a).
(b) The series for X(u,q) and Y(u,q) show that if q E JR, then the map
¢: C* /q'lt ~ Eq(C), ¢(u) = (X(u, q), Y(u, q))
commutes with complex conjugation. Since the isomorphism E ~ Eq in (a)
is defined over JR, it follows that the composition 'IjJ : C* /q'lt ----> E(C) is also
defined over lR. This proves the first half of (b), and the second follows by
taking GlCjlR-invariants of the exact sequence
1 --> q'lt --> C* --> E(C) --> 0,
so we get JR* /qZ ~ E(JR). (Note that HI (Gc / lR , qZ) = Hom(Gc / lR , Z) =
0.) Finally, it is clear that this isomorphism is lR.-analytic, since we know
that 'IjJ is C-analytic and is given by power series with coefficients in lR.
D
An elliptic curve over lR. has either one or two components. We can
use (2.3) to give a criterion to determine which case holds.
y2 = x 3 _ 2x2 - x+2 y2 =X3 -x + 1

L\ = 576 L\ =-368
Elliptic Curves over lR. with One and Two Components
Figure 5.3
Corollary 2.3.1. Let E(lR.) be an elliptic curve, and let tl.(E) be the
discriminant of some Weierstrass equation for E /R Then there is an iso-
morphism of real Lie groups
lR./IZ, if tl.(E) < 0,

E(lR.) ~ {
- (lR./IZ) x (1Z/21Z) , if tl.(E) > O.
PROOF. Fix an isomorphism E ~/lR Eq as in (2.3a). Then
u 12 tl.(E) = tl.(Eq) = q II (1- qn)24

n~l
for some u E lR., so
signtl.(E) = signtl.(Eq) = signq.

Now (2.3b) says that E(lR.) ~ Eq(lR.) ~ lR.* /qZ, so the following isomor-
phisms complete the proof of Corollary 2.3.1:
lR.* /qZ ~ lR./IZ, u ~~ C:: :~I- sign(u) + 1) (modlZ), if q < 0,
lR.* /qZ ~ (lR./IZ) x {±1}, u ~ C~:gl~1 (modZ), Sign(u)) , if q > o.

o
We conclude our discussion of elliptic curves over ffi. by describing the

Weil-Chatelet group WC(E/ffi.). (See [AEC X §3] for basic facts about the
Weil-Chatelet group.) As in (2.2), the fact that GC;'R is so small leads to
a very simple answer.
Theorem 2.4. Let E/ffi. be an elliptic curve, and let tl(E) be the dis-
criminant of some Weierstrass equation for E /ffi.. Then
WC(E/ffi.) C:<
-
{OZ/2Z if tl(E) <
if tl(E) > O.
0,
PROOF. From [AEC X.3.6] there is an isomorphism
Choose a q E ffi. and an ffi.-isomorphism E ~/1R Eq as in (2.3). Then
as Gc/lR-modules,
so we have an exact sequence
o ----+ qZ ----+ C* ----+ E (q ----+ 0
of Gc/lR-modules. Since
from Hilbert's Theorem 90 (or by an easy direct calculation), the long exact
sequence in GC/'R-cohomology gives
Now GC;IR = {I, a} is cyclic of order 2, so for any GC/'R-module M

(written multiplicatively) we have
(This is a special case of a general formula for the cohomology of cyclic

groups. See exercise 5.2.) Since GC/ IR acts trivially on qZ, we find
Hence we finally obtain an exact sequence
where the right-hand map is induced by the natural inclusion qZ '--> ~*.
From this exact sequence we immediately conclude that
WC(E/~) ~ {O
if q < 0,
Z/2Z if q > O.
But as we observed during the proof of (2.3.1),
U12 fl(E) = fl(q) = q II (1- qn)24 for some u E ~,

n2:1
so signfl(E) = signq. This completes the proof of Theorem 2.4. D
§3. The Tate Curve
We have seen that every elliptic curve defined over the complex numbers
has a parametrization C/ A for some lattice A c C. Suppose we replace C
by rQp and endeavor to parametrize an elliptic curve E /rQp by a group of the
form rQp/ A. Unfortunately, this approach immediately fails, because <Q!p
has no non-trivial lattices. Indeed, if A c rQp is any non-zero subgroup
:md 0 i- tEA, then
pUt E A for all n ~ 0 and lim pnt

n--->(X)
= 0,
so 0 is an accumulation point of A. Hence rQp contains no discrete subgroups

other than O.
Tate's idea is to first exponentiate, which leads to the alternative de-
scription C* / qZ for elliptic curves over C. Now the analogous situation
over <Q!p is much more promising, since rQ; has lots of discrete subgroups.
For example, any q E <Q!; with Iql < 1 defines the discrete subgroup
Further, the series described in (1.1) will converge in rQp and give a p-adic
analytic isomorphism of the quotient rQ;/ qZ with a certain elliptic curve E q .
The situation is nicely summarized by the following picture (taken from

Robert [1, II §5]).
Complex case p-adic case

CIA no p-adic analogue
1 exponential
map e 27riz
no exponential available
Q;/ qZ : p-adic elliptic curve.
More generally, we can work over any p-adic field K, by which we mean
a finite extension K/Qp. All of these facts (and more) are contained in
the next theorem, the proof of which will keep us busy for the next two
sections.
Theorem 3.1. (Tate) Let K be a p-adic field with absolute value 1 . I,
let q E K* satisfy Iql < 1, and let
be the series described in (1.1).

(a) The series a4(q) and a6(q) converge in K. Define the Tate curve Eq by
the equation
(b) The Tate curve is an elliptic curve defined over K with discriminant
f)" =q II (1 - qn)24
n2':l
and j-invariant
. 1
J(Eq ) = - + 744 + 196884q + ... = -1 + L c(n)qn,
q q n2':O
where the c(n) 's are the integers described in (1.1).

(c) The series
converge for all u E K, u ¢ qZ. They define a surjective homomorphism
¢: K* Eq(K)
{bX(u,q), Y(u,q») ifu ¢ qZ,
u
ifu E qZ.
The kernel of ¢ is qZ.

(d) The map ¢ in (c) is compatible with the action of the Galois group
G k / K in the sense that
In particular, for any algebraic extension Lj K, ¢ induces an isomorphism
Remark 3.1.1. The p-adic uniformization described in Theorem 3.1 is es-

pecially useful for arithmetic applications because it is compatible with the
action of Galois as described in (3.1d). Note that a complex uniformization
Cj A ---> E(C) or C* jqZ ---> E(C) does not have this compatibility (except
relative to GC/IR), since in general one cannot apply an element of Galois
to the value of a convergent series by applying it to each term of the series.
(See exercise 5.8.)
Remark 3.1.2. Theorem 3.1 is actually true for any field K that is com-
plete with respect to a non-archimedean absolute value. The only time we
will use the fact that K is a finite extension of Qp will be in the proof that
the map ¢ in (3.1c) is surjective. (In fact, we will really only need the fact
that the absolute value is discrete, so our proof actually is valid somewhat
more generally, for example over the completion of Q~r.) For a proof of
Theorem 3.1 in the most general setting, using p-adic analytic methods,
see Roquette [1].
PROOF. (a) From (LIb), the series defining a4 and a6 are in Z[q], so they
will converge in K for any value of q E K satisfying Iql < 1.
(b) The discriminant of Eq is
Substituting in the series for a4(q) and a6(q), we find the usual power series
.6.(q) = q - 24q2 + 252q3 + ... == q (mod q2).

Hence /.6.(q)/ = Iql =I- 0, so Eq is a non-singular elliptic curve.
Next we observe from (1.1c) that the identity
~(q) = q II (1 - qn)24
n~l
holds for all q E C with Iqloo < 1, where for the moment we write I . I
for the usual absolute value on C. It follows that this identity is true as
an identity of formal power series in Z[q]. Hence it remains true when we
take q to be an element of absolute value less than 1 in any field that is
complete with respect to a non-archimedean absolute value.
Finally, the formula
.( ) _ (1 + 48a4(q)) 3
J q - ~(q)
1 + 240q + 2160q2 + .. .
q - 24q2 + 252q3 + .. .
1
= -(1 + 744q + 196884q2 + ... )
q
holds in the non-archimedean case, since it is obtained formally by taking

the quotient of the appropriate power series.
(c) We begin by rewriting the series for X and Y as follows, where we've
used the alternative expression (1.2) for 81(q):
These expressions show immediately that X(u, q) and Y(u, q) converge for
all u E K*" qZ. (Note that although K itself is not complete, every term in
these series is in the field K(u) = Qp(u, q), which is a finite extension of Qp;
so we are really working in the complete field K(u). Similar comments will
apply below whenever we speak of substituting elements of K into a series.)
The functional equations
X(qu,q) = X(u,q) = X(u-1,q)

are now obvious, the first equality from the original series for X and the
second from the rearranged series we just gave. A little algebra gives similar
functional equations for Y,
Y(qu,q) = Y(u,q) and Y(U~I,q) = -Y(u,q) - X(u,q).
If we restrict u to the range Iql < lui < Iql~l, we have Iqnul < 1
and Iqnu~ 11 < 1 for all positive integers n. So we can use the expansion
with T equal successively to u, U~I, and 1 to rewrite the series for X as
X(u, q) = (1:: U)2 + :L(:L m(um + u~m - 2))qd E Q(u)[q],

d~1 mid
A similar calculation allows us to write Y as a power series in q with

coefficients in Q( u):
Y( )= u2 m_m(m+1) ~m
U, q (1 _ U)3 + ""(,,,,{(m-1)m
L L 2 u 2 u + m } ) qd
d~l mid
E Q(u)[q], valid for Iql < lui < Iql~l.
We begin our proof of (c) by showing that the image of the map ¢ is
contained in the curve Eq given by the Weierstrass equation
This amounts to showing that when we substitute the series X(u,q) and
Y(u, q) for x and y in this equation, we get an identity valid for all u E
K* "qz. By the periodicity of X and Y, it is enough to consider values
of u such that Iql < lui :S: 1 and u oJ 1. In this range we can use the above
formulas which express X and Y as power series in q with coefficients that
are rational functions of u. Thus we will be done if we can show that the
equation
Y(u, q)2 + X(u, q)Y(u,q) = X(u, q)3 + a4(q)X(U, q) + a6(q)

is valid as a formal identity in the ring of formal power series in q with
coefficients which are rational functions of the indeterminate u. In other
words, we want to verify that this identity holds in the ring Q(u)[q].
From (1.1), we know that this equation is true numerically if we sub-

stitute any pair of complex numbers u, q E C in the domain of conver-
gence \q\oo < \u\oo < \q\~}, u i= 1. If we fix some u with \q\oo < \u\oo < 1
and let q vary, we conclude that the resulting power series in q with com-
plex coefficients are equal coefficient-wise. Then letting u vary, we deduce
that the coefficients are formally equal as rational functions of u. Hence
we have an equality of formal power series in lQl(u)[q].
Next we prove that cP is a homomorphism. Given UI, U2 E K*, we
put U3 = UI U2 and must prove that
where Pi = cP(Ui), i = 1,2,3.

In view of the periodicity cP( qu) = cP( u), we may restrict consideration to
values of UI and U2 in the ranges
Then all three Ui are within the domain of convergence of the power series
expressions for X, Y E lQl(u)[q] described above.
Since cP(I) = 0 by definition, the relation P 3 = PI + P2 holds triv-
ially if UI = 1 or U2 = 1. Using the functional equations for X(u-l,q)
and Y(u-l,q) and the fact that PI + P2 = 0 if and only if Xl = X2
anel YI + Y2 = -Xl, it is also not hard to verify that P3 = PI + P 2 in the
case that u I U2 = 1. So we are reel uced to the case that PI, P2 , and P3 are
all different from O. We write Pi = (Xi, Yi); that is, we set Xi = X(Ui, q)
and Yi = Y(Ui, q) for i = 1,2,3.
Suppose first that Xl i= X2. Then writing out the addition law on E q ,
we see from [AEC, III.2.3] that the relation PI + P2 = P 3 is equivalent to
the two identities
(X2 - xd2X3 = (Y2 - YI)2 + (Y2 - YI)(X2 - Xl) - (X2 - Xd(XI + X2),
(X2 - XdY3 = -((Y2 - yd + (X2 - Xl))X3 - (YIX2 - Y2xd·
Now we can argue as above that (1.1) implies that these identities hold for
all complex numbers UI,U2,q in the specified ranges. Hence they are iden-
tities in the ring 1Ql( UI, U2) [q] of formal power series in q with coefficients
that are rational functions of UI and U2, and so are true for UI,U2,q E K.
To deal with the remaining case Xl = X2, we could use the duplication
formula, or we could invoke a p-adic continuity argument, but perhaps the
simplest solution is to observe that Xl = X2 if and only if PI = ±P2, and
then use the following lemma.
Lemma 3.1.2. Let cP be a map of a (multiplicative) group into an (addi-
tive) group which takes on an infinite number of distinct values and satisfies
the identity
Then ¢ is a homomorphism.
PROOF. Given any Ul and U2, the fact that ¢ takes on infinitely many
distinct values means that we can choose a U such that
¢(u) i- ±¢(ut},
Then ¢(uut} = ¢(u) + ¢(Ul) i- ±¢(U2), so
¢(u) + ¢(ud + ¢(U2) = ¢(uud + ¢(U2) = ¢(UUIU2) = ¢(u) + ¢(UIU2).
Canceling ¢(u) gives ¢(Ul)+¢(U2) = ¢(UIU2), valid for all Ul and U2, which
shows that ¢ is a homomorphism. 0
To finish the proof that our ¢ : K* --; Eq(K) is a homomorphism,

we need merely observe that ¢ certainly takes on infinitely many distinct
values. For example, the series for X (u, q) shows that for any t E K
with It I < 1, we have IX(1 + t,q)1 = Itl- 2 . Hence (3.1.2) applies in our
case.
So we now know that ¢ is a homomorphism of K* into Eq(K). That
the kernel of ¢ is qZ is apparent from its very definition. It remains to
prove that ¢ is surjective. This is the hardest part of the proof, which we
will leave to the next section.
(d) As noted above, the series for X (u, q) and Y (u, q) converge in the
complete field K(u), so it really suffices to prove (d) for 0" E G L/K, where L
is any finite Galois extension of K containing K(u). Any such 0" maps the
maximal ideal of the ring of integers of L to itself, so 0" will preserve the
absolute value on L:
laul = lal for all 0" E G L / K and all a E L.

It follows easily from this that if E O:i is a convergent series with ai E L,
then (E ait= E af· (See exercise 5.8.) Applying this to the series
for X(u,q) and Y(u,q), we deduce that ¢(u)U = ¢(U U). This proves the
first part of (d).
For the second part, we use (c) to produce the exact sequence
We now know that the maps in this exact sequence commute with the
action of G K / K. Hence for any algebraic extension L / K, we can take G K / L
invariants of this short exact sequence to obtain the exact sequence
To obtain surjectivity on the right, we observe that it suffices to prove

surjectivity in the case that L is a finite extension of !Qlp; we will prove
exactly this fact in the next section. Alternatively, we may observe that
the next term in this last sequence is the cohomology group HI (G K / L, qZ).
Since q E K, the action of G K / L on qZ is trivial, so this is just the group
of continuous homomorphisms from the profinite group G K / L to the dis-
crete group qZ ~ z. The only such homomorphism is the trivial one,
so HI(G K / L , qZ) = 0, which proves that L* - t Eq(L) is surjective. This
completes the proof of (d). 0
Before resuming the proof of Theorem 3.1 in the next section, we

briefly pause to repeat Proposition 1.3 in the context of p-adic theta func-
tions. These formulas will be used in our study of local height functions in
Chapter 6.
Proposition 3.2. Define a function O(u, q) by the formula
= (1 _ ) II (1 - qn u)(1 - qnu- I )
O(
u, q ) u (1 _ n)2 .
n~l q
(a) O(u,q) converges for all u,q E Q; with iqi < 1 and satisfies the func-
tional equation
1
O(qu,q) = --O(u,q).
u
(b) 0 is related to the functions X(u, q) and Y(u, q) described in (3.1c) by

the formulas
(i)
(ii)
PROOF. The convergence of the infinite product defining 0 is clear, and the
functional equation follows formally by substituting qu for u and renum-
bering. This proves (a). Next we observe from (1.3) that the two for-
mulas in (b) are valid over C. Now an argument similar to that used to
prove (3.1c) shows that they are valid over K. We will leave the details to
the reader. 0
§4. The Tate Map Is Surjective
The final step in the proof of (3.1), which we postponed from the previous
section, is to show that the map
¢(u) = (X(u,q), Y(u,q)),
is surjective. One approach is to reprove in the p-adic case some classical

results from complex analysis concerning Laurent series. In particular, one
proves Schnirelmann's Theorem that a Laurent series f(X) which converges
for all X =I- 0 can be written as a convergent product
where the product is over all roots a of f. Then one constructs the field
of p-adic meromorphic functions on K* /qZ and shows using Riemann-Roch
that it is a field of genus lover K.This leads to an isomorphism with some
elliptic curve, and after some work with the classical power series for p
and p', one deduces that the elliptic curve is indeed the curve we have
denoted E q . For details of this line of proof, see Robert [1], Roquette [1],
and Tate [9]. We will take a more computational, geometrically inspired
approach. However, we should note that the theory of p-adic analytic
functions has many important applications in modern arithmetic geome-
try. The reader might consult Bosch-Giinter-Remmert [1] for a thorough
introduction to this subject which is called rigid analysis.
In order to prove that ¢ : K* -> Eq(K) is surjective, we need to show
that for any given point P E Eq(K) there is some u E K* with ¢(u) = P.
But the point P will be defined over some finite extension of K, so it suffices
to prove that ¢ : L* -> Eq(L) is surjective for all finite extensions L/ K. In
fact, this is even stronger than the original statement of (3.1c), although
it is precisely the result we needed to complete the proof of (3.1d). For
notational simplicity, we will write K in place of L, so we are reduced to
showing that for any finite extension K/Qp and any q E K with iqi < 1,
the map ¢ : K* -> Eq(K) described in (3.1) is surjective.
We also set the following notation which we will use for the remainder
of this section:
R the ring of integers of K,
9J1 the maximal ideal of R,
1f a uniformizer for R, 9J1 = 1f R,
k the residue field of R, k = Rj9J1,
ord v the normalized valuation ord v : K* --» Z on K.
The group Eq(K) admits the usual filtration (see [AEC, VII §2])
Eq(K) :J Eq,o(K) :J Eq,l(K),
where
Eq,o(K) = {p E Eq(K) : FE Eq,ns(k)},
Eq,l(K) = {P E Eq(K) : F = 6}.
Here Eq/k is the reduction of Eq modulo VJt, and Eq,ns are the non-singular
points on Eq • From [AEC, VII.2.1] and [AEC, VII.2.2] we have isomor-
phisms
Eq,o(K)/Eq,l(K) ~ Eq,ns(k) and

Pt-->F
where E is the formal group of E [AEC, IV §1].

Similarly, the quotient group K* / qZ has a natural filtration
where
Ri = {u E R : u == 1 (mod VJt) }
is the group of I-units in R. There are also isomorphisms
R* /Ri ~ k* and Ri = Gm(VJt)

a t--> a u t - - > 1 - u,
where Gm is the formal multiplicative group [AEC, VI.2.2.2]. We are going

to prove not only that the map ¢ : K* /qZ -> Eq(K) is an isomorphism but
that it respects the filtrations we just described.
We begin with the formal groups. First, from the formula for X(u,q),
it is clear that
u == 1 (mod VJt) ==} ordv(X(u,q») < 0,

u
since only the term (1 _ u)2 will be non-integral. This proves that
Next we show that this inclusion is an equality.

Using the isomorphisms described above, we look at the map
Gm(VJt) ~ Ri ~ Eq,t{K) ~ Eq(VJt)

X(I+t,q)
t
Y(l + t, q)"
Note that as sets, Gm(001) and Eq(001) are just the set 001; they merely have
different group structures attached to them. If we substitute u = 1 + t into
the series for X(u, q) and Y(u, q) and expand as Laurent series in t, we find
that
X(l+t,q) = r2(1+ L amtm) and Y(1+t,q)=r 3 (1+ L,Bmtm),
m2::1 m2::1
with coefficients am,,Bm E R. By taking the ratio of X and Y, we are

reduced to showing that if /1,)'2, ... E R, then the map
t I---t t (1 + L /mtm)
m2::1
is surjective. This follows immediately from [AEC, IV.2.4], which asserts

the existence of a power series 'x(T) E R[T~ satisfying 1/1 ('x(T) ) = T (Le., 1/1
is surjective, since for any w E 001 we have ,x(w) E 001 and 1/I('x(w)) = w.)
Next we look at the behavior of ¢ on R*. If we take the series
for X (u, q) and reduce it modulo 001, we see that
u
X(u, q) =- (1 _ U)2 ¢ a (mod 001) for all u E R*,
so ¢(R*) C Eq,o(K). Since ¢(Ri) = Eq,l(K) from above, we get a well-

defined injective homomorphism on the quotient groups
k* ~ R*/Ri ~ Eq,o(K)/Eq,l(K) ~ Eq,ns(k)
UI C1 ~U)2' (1 ~2U)3).
This map k* ----> Eq,ns(k) is clearly surjective, the inverse being
y2
(x,y) I---t 3 '
x
so the map on the quotient groups is an isomorphism. (Note that Eq has
the equation y2 + xy = x 3.) Then the commutative diagram
1 ~ R*1 ~ R* ~ k* ~ 1
1/ 1q; 1/
a ~ Eq,l(K) Eq,o ~ Eq,ns(k) ~ ~ a
implies that the map ¢ : R* ~ Eq,o(K) is an isomorphism.
We are left to show that the injective homomorphism
¢: K* /R*qZ ~ Eq(K)/Eq,o(K)
is surjective. The group on the left is easy to describe, since the map
K* / R*qZ ~ Z/ordv(q)Z
u
is clearly an isomorphism. So the following proposition will complete the
proof of (3.1).
Proposition 4.1.
To prove this estimate, we will use geometry to divide Eq(K) into

several subsets, and then we will show that these subsets actually corre-
spond to the cosets of Eq,o(K) inside E(K). We start with an elementary
characterization of the points in Eq,o(K).
Lemma 4.1.1. Let P = (x, y) E Eq(K). The following are equivalent:
(i) P E Eq,o(K),
(ii) Ixl 2': 1,
(iii) Iyl 2': 1.
PROOF. Taking partial derivatives of the equation for E q , we observe that
P E Eq,o(K) ~ Iy - 3x 2- a41 2': 1 or 12y + xl 2': 1

~ max{ly - 3x 21, 12y + xl} 2': 1 since la41 = Iql < 1
~ max{lxl, Iyl} 2': 1.
Suppose first that Ixl 2': 1 > Iyl. Then
Ixl 3 = ly2 + xy - a4X - a61 ~ max{lyI2, Ixyl, la4xl, la61} < max{l, Ixl}.
This strict inequality is a contradiction, so Ixl 2': 1 implies Iyl 2': 1. Similarly,
the assumption Iyl 2': 1 > Ixl gives the contradiction
so Iyl 2': 1 implies Ixl 2': 1. This proves that
max{lxl, Iyl} 2': 1 ~ Ixl 2': 1 and Iyl 2': 1,

so (i) implies (ii) and (iii).
Conversely, if P fJ- Eq,o(K), then P = (x, y) reduces to the singular
point (0,0) of Eq(k), so Ixl < 1 and Iyl < 1. Hence either of (ii) or (iii)
implies (i), which completes the proof of the lemma. 0
Next we use similar criteria to partition the points of Eq(K) that do

not lie in Eq,o(K).
Lemma 4.1.2. Let P = (x, y) E Eq(K) "Eq,o(K). Then exactly one of
the following three conditions is true:
(i) 1> Iyl > Ix + yl, in which case Iyl > Iql!,
(ii) 1 > Ix + yl > IYI, in which case Ix + yl > Iql!,
(iii) jyj = jx + yj = jqj!.

(Note that (iii) can only occur if ordv(q) is even.)
PROOF. Let
n = min{ordvx,ordvy} and N = ordvq.
Dividing the equation for Eq by 7r 2n , we obtain the equation
Yn2 + XnYn = 7r n Xn3 + 7r -n a4 ()

q Xn + 7r -2n a6 ()
q ,
where
and
(In fancy terminology, we've blown up the scheme E q / Spec(R) to find
the affine subscheme on which P lies.) Since ja4(q)j = ja6(q)j = jqj, it is
immediate from this equation that 7r-2na6(q) E R, so
(The fact that 1 :::; n comes from (4.1.1).)

We ;:"W consider two cases. First, if n < !N, then reducing the above
equation modulo 7r gives
y~ + XnYn == 0 (mod 7r).

This means that either
Yn == 0 (mod 7r), or Yn + Xn == 0 (mod 7r), or both.
But they cannot both be zero, since otherwise Xn == Yn == 0 (mod 7r), which
would contradict the definition of n. Hence one of the following two asser-
tions is true:
(i) Yn"¢ 0 (mod 7r) and Yn + Xn == 0 (mod 7r),
(ii) Yn == 0 (mod 7r) and Yn + Xn "¢ 0 (mod 7r).
These correspond to (i) and (ii) in the statement of the lemma. For exam-
ple, for (i) we find that
and
It remains to deal with the case n = ~N. Since a6(q) = -q + .. " our
equation becomes
Hence IYnl = IYn + xnl = 1, which implies that
o
Lemmas (4.1.1) and (4.1.2) allow us to divide Eq(K) into the following
subsets:
Eq,o(K) = {(x,y) E Eq(K) Ixl 2: 1 or Iyl 2: 1},
Un = {(x, y) E Eq(K) 17rl n = Iyl > Ix + yl},
Vn = {(x, y) E Eq(K) 17rl n = Ix + yl > Iyl},
W = {(x,y) E Eq(K) Iyl = Ix+yl = Iql!}.
Notice that (4.1.2) says that Un and Vn are empty unless n < ! ordv q,
so Eq(K) can be written as the union
Eq(K) = Eq,o(K) UW U
Further, if ordv q is odd, then W = 0. So we have partitioned Eq(K) into

(at most) ordv(q) pieces. The final step in the proof of Proposition 4.1,
which will also complete the proof of Theorem 3.1, is to show that these
subsets are the cosets of Eq,o(K) in Eq(K). More precisely, it will suffice
to show that two points in the same subset are in the same coset, since
this will imply that the number of cosets is no larger than ordv(q). This is
exactly what we do in the following lemma.
Remark 4.1.3. A more intrinsic explanation for the above decomposition
of Eq(K) is that the subsets Un, Vn , and Ware neighborhoods of the
non-identity components of the special fiber of the Neron model of Eq
over Spec(R). This decomposition may be compared with the description
of special fibers of Type In in (IV §8), Tate's algorithm (IV.9.1), and the
discussion in (IV.9.6).
Lemma 4.1.4. Let P, pI E Eq(K) be points satisfying anyone of the
following conditions:
(i) P, pI E Un; (ii) P, pI E Vn ; (iii) P, pI E W.
Then P - pI E Eq,o(K).
PROOF. The proof of this lemma is completely elementary, although some-

what computationally involved. We merely have to combine the geometric
description of Un, Vn , and W with the algebraic formulas giving the group
law on E q .
If P = pI, there is nothing to prove. Assume for now that also P i=
_P'. (We'll deal with P = _pI at the end.) Then writing P = (x,y)
and P' = (x', y'), the addition law on Eq and a little algebra yield the
formula
(P _ P') = (y + y' + x')(y + x + y') _ _ '
x (x-x
')2 X x.
In all three cases, (i), (ii), and (iii), we have Ixl < 1 and Ix'i < 1, so (4.1.1)
and the formula for x(P - P') gives
P - P' E Eq,o(K) ~ Ix(p - p')1 ::::: 1

~ Iy + y' + x'I·ly + x + y'l : : : Ix - x'1 2 •
Suppose first that P, P' E Un. Then
On the other hand, since Iyl = 11l"In and Iy' + x'i < 11l"1n, we get
Therefore
Iy + y' + x'i . Iy + x + y'l = 11l"1 2n : : : Ix - x'1 2 , so P - P' E Eq,o(K).
This proves (i).

The proof of (ii) can be done in a similar fashion, but it is even easier
to observe that
P E Un {::::::} -P E Vn ·
This follows from the formula -(x, y) = (x, -y - x). Hence (i) implies (ii).
Further, since Un and Vn are disjoint, we see that P =1= -P' in cases (i)
and (ii).
We turn now to case (iii), which is the most difficult. We claim that
in this case we have
Iy + x + y'l = Iy + y' + x'l·

To see this, we note that
then we compute
I(y + x + y')y - (y + y' + x')y'l

= l(y2 + xy) - (y,2 + x'y') I
= l(x 3 + a4x + a6) - (X,3 + a4x' + a6)1
= Ix - x'l·lx 2 - xx' + x,2 + a41

:s: Ix - x'llql since lxi, Ix'i :s: Iql!, la41 = Iql
= I(y + x + y') - (y + y' + x') I . Iql
:s: max{ly + x + y'l, Iy + y' + x'l} 'Iql

= max{l(y + x + y')yl, I(y + y' + x')y'l} 'Iql!
since Iyl = Iy'l = Iql!,

< max{l(y + x + y')yl, I(y + y' + x')y'l}
since Iql < 1.
The only way that this strict inequality can possibly be true is if
I(y + x + y')yl = I(y + y' + x')y'l·

Further, we know that Iyl = ly'l, so we have proven our claim
Iy+x+y'l = Iy+y' +x'l·
Using this equality, we compute
Ix - X'I2 = I(y + x + y') - (y + y' + x')1 2

:s: max{ly + x + y'l, Iy + y' + x'l} 2
= Iy + x + y'l'ly + y' + x'l·
From above, this inequality implies that P - P' E Eq,o(K).

Finally, we must deal with the case P' = -P E W, so P - P' = 2P.
We could argue by continuity, but here is a direct argument using the
duplication formula, which on Eq reads
Here f(x) and g(x) are the indicated polynomials. From general principles,
one knows that f(x) and g(x) are relatively prime in K[x]. More precisely,
if we let
then a little algebra suffices to verify the relation
f(x)F(x) - g(x)G(x) =~,
where ~ is the discriminant of the Weierstrass equation for E q . Substitut-

ing f(x) = x(2P)g(x) into this relation gives
g(x){x(2P)F(x) - G(x)} =~.
We are assuming that PEW, so
Ig(x)1 12y + Xl2 :::; max{lyl, Iy + xl} 2 = Iql,

=
IG(x)1 = 112x 3 - x 2 - 20a4x + 2a4 - 108a61 :::; max{l x I2, la41, la61} = Iql,
IF(x)1 = 148x 2 + 8x + 64a4 -11 = 1,
I~I = Iq - 24q2 + 252q3 - .. ·1 = Iql·
Hence
1:::; Igt) 1 = Ix(2P)F(x) - G(x)1
:::; max{lx(2P)I'IF(x)l, IG(x)1}
:::; max{lx(2P)I, Iql}·
Since Iql < 1, it follows that Ix(2P)1 ~ 1, so 2P E Eq,o(K) from (4.1.1).
This completes the proof of Lemma 4.1.4, and with it the proofs of Propo-
sition 4.1 and Theorem 3.1.
o
§5. Elliptic Curves over p-adic Fields
In the previous two sections we have shown that for any p-adic field K /Qp
and any q E K* with Iql < 1, the quotient group K* /qZ is (analytically)
isomorphic to an elliptic curve Eq(K). In the analogous situation over
the complex numbers, we know (LId) that every elliptic curve E/C is
isomorphic to Eq for some q E C*. However, in the p-adic case we have
Ij(Eq)1 = I~ +744 +196884q +···1 = I~I > 1,
so it is clear that not every elliptic curve over K can be isomorphic (over K)
to an Eq. A necessary condition is Ij(E)1 > 1. We begin by showing that
this condition is also sufficient.
Lemma 5.1. Let a E Qp be an element with lal > 1. Then there is a

unique q E Q; with Iql < 1 such that j(Eq) = a. This value oE q lies in
Qp(a).
PROOF. The j-invariant of Eq is given by the series (3.1), which we write

as
"() 1 + 744q + 196884q2 +.".
J q = "
q
The reciprocal of this series, which we will call I(q), is given by the formula
1 q
I(q) = j(q) = 1 + 744q + 196884q2 +.""
= q- 744q2 + 356652q3 - .. " E Z[q].
Applying [AEC, IV.2.4] to the series I, we get a series g(q) = q+ . .. E Z[q~

such that g(J(q») = q as formal power series in Z[q]. Since 9 has integer
coefficients and leading term q, it will converge if we evaluate it at any
element 13 E Qp of absolute value less than 1 and will satisfy Ig(13) = 1131. I
In particular, since lal > 1, we find that
satisfies
and jtq) =/(q)=/(g(~)) =~.

Hence j(q) = a as desired. This proves the existence part of (5.1).
To prove uniqueness, suppose that j(q) = j(q') with Iql < 1 and Iq'l <
1. Then I(q) = I(q'), so
0= I/(q) - l(q')1
= Iq - q'l .11 - 744(q + q') + 356652(q2 + qq' + q'2) +···1
= Iq-q'l·
Therefore q = q'. o
Before proving the p-adic uniformization theorem, we describe an in-
variant which is useful for studying the twists of a curve.
Lemma 5.2. Let Ej K be an elliptic curve defined over a field oE char-
acteristic not equal to 2 or 3, and choose a Weierstrass equation
for ElK. Let C4 and C6 be the usual quantities [AEC, III §1} associated to
this equation. Assuming that j (E) f:- 0, 1728, we define
(The reason for the negative sign will become apparent later when we prove
that ,(EqIK) = 1.)
(a) ,( ElK) is well-defined as an element of K* I K* 2 , independent of the
choice of Weierstrass equation for ElK.
(b) Let E'I K be another elliptic curve with j (E') f:- 0, 1728. Then E
and E' are isomorphic over K if and only if
j(E) = j(E') and ,(ElK) = ,(E'IK).
(c) Let ElK and E'IK be elliptic curves with j(E') = j(E) f:- 0,1728,
and suppose that ,(ElK) f:- ,(E'IK), so
,(ElK) )
,(E'I K)
is a quadratic extension of K. Let
x: GR/K ---> GL / K ---> {±1}
be the quadratic character associated to L I K. Then there is an isomor-

phism
1jJ: E-tE'
with the property that
1jJ(PU) = x(a)1jJ(P) for all a E GR/K and all P E E(K).
PROOF. (a) The condition j(E) f:- 0,1728 is equivalent to C4 f:- 0 and C6 f:-
0, so ,(ElK) exists. If we choose a new Weierstrass equation for EI K, then
the new C4 and C6 are related to the old ones by the formulas u4c~ = C4
and u6c~ = C6 for some u E K*. (See [AEC, III Table 1.2]. The fact
that u E K follows from [AEC, III.3.1].) Hence
which proves that ,(ElK) is independent of the chosen Weierstrass equa-

tion.
(b) If E and E' are isomorphic over K, then [AEC, III.1.4b] asserts that
j(E) = j(E'). Further, since the Weierstrass equations for E and E' are
Weierstrass equations for the same elliptic curve over K, it follows from (a)
that '"'((ElK) = '"'((E'IK).
Conversely, suppose that j(E) = j(E') and '"'((EI K) = '"'((E'I K). Since
the characteristic of K is not 2 or 3, we can find Weierstrass equations for E
and E' over K of the form
E : y2 = x3 + Ax + B, E' : y2 = x 3 + A' x + B',

with A, B, A', B' E K. The fact that j (E) = j (E') 01 0, 1728 implies that
4A 3
since j(E) = 1728 4A3 + 27B2
Similarly, since C4 = -48A and C6 = -864B, our assumption '"'(( ElK) =

'"'((E'I K) means that
2A == C4 == -,",((ElK) == -,",((E'IK) == c~ == 2A' (mod K*2),

B C6 c~ B'
so there is some t E K* such that AB' = t 2 A' B. Using these relations

between A, B and A', B', it is now easy to check that the map
E ---> E',
is a K-isomorphism.
(c) We take models for ElK and E'I K as in (b), and again the assump-
tion j (E) = j (E') 01 0, 1728 implies that A 3 B,2 = A,2 B3. Next we let
t=
Since '"'((EI K) 01 '"'((E' I K), we know that L = K(t) is a quadratic extension

of K; and as in (b), the map
'l/J: E ---> E',
is easily seen to be an isomorphism. Finally, for any a E G k / K, we know

that t<7 = x(a)t. So for P = (x, y) E E(K) we have
'l/J(P)<7 = 'l/J(x, y)<7 = (t 2x, t 3y)<7(x(a)2t2 x O", x(a)3t 3yO")

=
= (t 2x<7, x(a)t3 y<7) = x(a)(t 2x O", t 3yO") = x(a)'l/J(P'd.

We are now ready to prove Tate's p-adic uniformization theorem, which
applies to all curves whose j-invariant has absolute value greater than 1.
Theorem 5.3. (Tate) Let K be a p-adic field, let ElK be an elliptic

curve with Ij(E)1 > 1, and let ,(ElK) E K* I K*2 be the invariant defined
in (5.2).
(a) There is a unique q E K* with Iql < 1 such that E is isomorphic over K
to the Tate curve E q . Further, this value of q lies in K.
(b) Let q be chosen as in (a). Then the following three conditions are
equivalent:
(i) E is isomorphic to Eq over K.
(ii) ,(EIK)=1.
(iii) E has split multiplicative reduction.
PROOF. (a) From (5.1) there is a unique q E K* with Iql < 1 such that
j(Eq) = j(E). This implies [AEC, III.1.4b] that Eq is isomorphic to E
over K, which completes the proof of (a).
(b) From (5.2) we know that E is isomorphic to Eq over K if and only
if j(E) = j(Eq) and ,(ElK) = ,(EqIK). So in order to prove that (i)
and (ii) are equivalent, we must show that ,(Eq / K) = 1. Using (3.1) we
find that the C4 and C6 values associated to the Tate curve
are
C4(q) = 1 - 48a4(q) = 1 + 240S3(q),
C6(q) = -1 + 72a4(q) - 864a6(q) = -1 + 504s5(q).
So the ,-invariant of Eq / K equals
To see that ,(Eql K) is a square, we use the following elementary calcula-

tion which implies that C4(q) and -C6(q) are themselves squares in K.
Lemma 5.3.1. Let a E K with lal < 1. Then 1 + 4a is a square in K.
PROOF. We first observe that the binomial coefficient
is an integer divided by 4n. Hence the coefficients of the series

are integers, so the series converges in K. Therefore (1 +40:)-1 is a square

in K, so the same is true of 1 + 40:. This completes the proof of the lemma,
and with it the fact that (i) and (ii) are equivalent. 0
Next we note that since la4(q)1 = la6(q)1 = Iql < 1, the equation of
the reduced curve Eq is
which clearly has split multiplicative reduction. This shows that (i) im-
plies (iii).
Conversely, suppose that E has split multiplicative reduction. We will
show that 'Y(E/K) = 1, which will prove that (iii) implies (ii). Take a
minimal Weierstrass equation for E,
Making a linear change of variables, we may assume that the singular point
modulo wt is the point (0,0), where as usual we write wt for the maximal
ideal of the ring of integers of K. Then the fact that (0,0) is on the curve
and singular modulo wt implies that
and hence that

b4 = a1a3 + 2a4 == 0 (mod wt) and C4 = b~ - 24b4 == b~ (mod wt).
From [AEC, VII.5.1b], the fact that E has multiplicative reduction

implies C4 t- 0 (mod wt), so we see that b2 t- 0 (mod wt). It follows that b2
is a unit (Le., Ib2 1 = 1). Hence
1- 24b4 )
'Y( E / K) = _ C4 = ~ . ( b~
C6 b2 1- 36 b4 +216 b6
b~ b~
Applying (5.3.1) to the numerator and denominator of the bracketed frac-

tion on the right-hand side of this equation, we find that
It remains to show that if the mUltiplicative reduction of E is split, then b2

is a square in K*.
Note that the reduction of E is
We factor the polynomial
The fact that E has multiplicative reduction means that E has a node,
so Q =f. i3i and the fact that the reduction is split means that Q and i3 are
actually in the residue field of K, rather than in a quadratic extension.
(See [AEC, III §1, VII §5].) It follows from Hensel's lemma applied to the
polynomial T2 + alT - a2 that Q and {3 lift uniquely to elements a, f3 E K
such that
Hence
b2 = ai + 4a2 = (-a - (3)2 + 4( -a(3) = (a - (3)2 E K*2,
so "((ElK) == b2 == 1 (mod K*2).

We have now proven (ii) {:==} (i) ==? (iii) ==? (ii), which completes
the proof of Theorem 5.3. 0
Suppose that we have an elliptic curve ElK as in Theorem 5.3 with

invariant "((EI K) =f. 1. If we let L = K ( v,,((EI K) ), which is well-defined,
since "((ElK) is defined up to squares in K, then it is clear that "((ElL) =
1. Applying (5.3) to ElL, we find that E is isomorphic to Eq over L, so
We will now describe E(K) in terms of this identification.

Corollary 5.4. With notation as in the preceding paragraph,
PROOF. First we observe that the norm map N~ is a homomorphism
N~ : L* IqZ --> K* Iq2Z,
so N~(u) is well-defined modulo q2Z. Applying (5.2c) to E and E q, there

is an isomorphism
7jJ: Eq(K) --> E(K)
satisfying 7jJ(PU) = x(a)7jJ(P)U for all a E GR/K, where X : GR/K ->
G L / K -> {±1} is the quadratic character associated to LIK. On the other
hand, the isomorphism ¢ : K* IqZ -> Eq(K) is defined over K, which means
that ¢(PU) = ¢(P)u. We look at the composition
which we know from above is an isomorphism of groups. Let rEG R I K

be an element with x(r) = -1, so r represents the non-trivial element
in G L I K· Then for any u E L * ,
(1jJ 0 ¢)(u) E E(K) {::=? 1jJ(¢(u)r= 1jJ(¢(u»)

{::=? _1jJ(¢(u = 1jJ(¢(u»)
T )) since x(r) = -1
{::=? 1jJ(¢(u- T )) = 1jJ(¢(u»)
since -1jJ(P) = 1jJ(-P) and -¢(u) = ¢(u- 1 )
{::=? u- T == u (mod qZ)
since ¢ and 1jJ are isomorphisms
Since U1+T = N}((u), this completes the proof of the corollary. o
§6. Some Applications of p-adic U niformization
As we have seen amply demonstrated, the arithmetic properties of the

torsion points on an elliptic curve are of fundamental importance. In the
case that the curve has a p-adic uniformization, E(K) ~ K* IqZ, it is easy to
describe the torsion subgroup of E. Further, since the p-adic uniformization
commutes with the action of GRIK, it is similarly easy to describe the
action of GRIK on the torsion subgroup of E. We will not prove the most
general theorem in this direction but will be content with the following
fundamental result. (See also exercise 5.13.)
Proposition 6.1. Let K be a p-adic field with normalized valuation ord v ,
let ElK be an elliptic curve with ij(E)i > 1, and let e ~ 3 be a prime
not dividing ordv j(E). Then there is an element a in the inertia subgroup
ofGRIK which acts on the e-torsion subgroup E[e] of E via a matrix of
the form ("6 1). In other words, there is a basis PI, P2 E E[e] sllch that
and
(One sometimes says that a acts as a transvection on E[e].)

Remark 6.1.1. Recall that there is an e-adic representation [AEC, III §7]
Pi: GRIK ---> Aut(Te(E».

Since E[f] ~ Te(E)/fTe(E), another way to state (6.1) is that relative to

an appropriate basis, there is a a E G K / K satisfying
pe(a) == (~ ~) (mod f).
Remark 6.1.2. It is also worth pointing out that the proof of Proposi-
tion 6.1 does not need the full strength of Theorem 3.1. Specifically, we
only need to know that the map ¢ : K* /qZ ---t Eq(K) is an injective ho-
momorphism; we do not need to know that it is surjective. The reason
injectivity suffices is that we are really only interested in the torsion sub-
group of E q , and a simple count shows that the there are m 2 points of
order m in K* /qZ, so we get (essentially for free) that ¢ is an isomorphism
on torsion.
PROOF. First we observe that if L/ K is a finite extension of degree prime
to f, and if (6.1) is true for E / L, then it is true for E / K. This follows from
the equality ord w j(E) = ew / v ordv j(E), where w is the extension of v
to L, and the ramification index ew / v is prime to f, since it divides [L : K].
Hence f will not divide ord w j (E), so there is a a E G K / LeG K / K that
acts as a transvection on E[f].
From (5.3b) we know that E is isomorphic to a (unique) Tate curve Eq
over an (at most) quadratic extension of K. So replacing K by this ex-
tension, it suffices to prove (6.1) for E q , where q E K*. Similarly, we may
assume that K contains a primitive fth_root of unity (, since the degree
of K «() / K divides £ - 1, so the degree is prime to e.
Let Q = qi E K be a fixed fth_root of q. Since ordvj(Eq) = -ordvq
is not divisible by £, the Kummer extension K(Q)/K is totally ramified
of degree f. Hence there exists a a in the inertia subgroup of G K / K such
that QU = (Q. We claim that this is the desired a; it remains to pick the
right basis for Eq[f].
To do this, we use the p-adic uniformization (3.1)
¢: K* /qZ ~ Eq(K).
With this identification we clearly have
¢: (e. QZ)/qZ ~ Eq[f].
Further, the p-adic uniformization map ¢ commutes with the action of
Galois (3.1d) (Le., ¢(PU) = ¢(P)U), so the action of G K / K on Eq[f] is
the same as its action on the quotient group (e· QZ)/qz. As our basis
for Eq[f], we take the elements PI = ¢«() and P2 = ¢(Q). Then
Pf = ¢«()U = ¢«(U) = ¢«() = P1 ,
Pf = ¢(Q)U = ¢(Qr7) = ¢«(Q) = ¢«() + ¢(Q) = PI + P2 . o
We next observe that Proposition 6.1 remains true for certain elliptic
curves over number fields.
Corollary 6.2. Let K /Q be a number field, let E / K be an elliptic

curve, and assume that the j-invariant of E is not in the ring of integers
of K. Then for all but finitely many primes f, the image of the f-adic
representation Pe : GklK - Aut(Te(E») contains an element satisfying
pe(cr) == (~ ~) (mod f)
relative to a suitable basis for Te(E)/fTe(E) = E[f].
PROOF. Let v be a (finite) place of K for which j(E) is non-integral,

so Ij(E)lv > 1. Let Gf<vlKv C GklK be the decomposition group of v
for the extension of v to K corresponding to some embedding K ~ Kv.
Now (6.1) gives an element cr E GkvlKv which acts like on E[f]. (6 l)
But with our identifications, cr E GRIK and E[f] C E(K) c E(Kv ), which
gives the desired result. 0
It is a legitimate question to ask why one should care that Gal(K / K)

contains an element that acts on E[f] as a transvection. One answer is that
this puts severe constraints on the allowable maps between such elliptic
curves. For example, we will now give Serre's p-adic proof that an elliptic
curve with complex multiplication has integral j-invariant. (For alternative
proofs of this important fact, see [AEC, exercise 7.10] and (II §6).)
Theorem 6.3. Let K /Q be a number field, and let E / K be an el-
liptic curve whose j-invariant j(E) is not in the ring of integers of K.
Then End(E) = Z.
PROOF. (Serre) We begin by recalling that there is a representation of the

endomorphism ring of E [AEC, III §7j,
End(E) -+ End(Te(E») ,
Further, we proved in [AEC, V.2.3] that for any'ljJ E End(E), this repre-
sentation can be used to compute the degree of 'ljJ via the formula
deg('ljJ) = det('ljJe).
(This result appears in [AEC] in the chapter on elliptic curves over finite
fields, but the proof depends only on the non-degeneracy of the Wei! pair-
ing, which is valid in general.)
Let 'ljJ E End(E) be an isogeny. Taking a finite extension of K if
necessary, we may assume that 'ljJ is defined over K. This means that
for all 'ljJ E GRIK and all P E E(K).

We need to show that 'IjJ E Z.

Let
m = deg(l + 'IjJ) - deg('IjJ) - 1.
Notice that if we knew that 'IjJ was in Z, then m would equal 2'IjJ. So we
will try to show that m = 2'IjJ by showing that the degree of m - 2'IjJ is O.
Using (6.2), choose a "large" prime f, an element a E Gk/K' and an
ordered basis {Pl , P2 } for E[f] so that relative to this basis,
pt(a) == (~ ~) (mod f).
(We will see below that any f larger than deg( m - 2'IjJ) will suffice.) Looking
at the action of'IjJ on E[f], we find that 'IjJ is represented by a matrix
'ljJt == (~ ~) (mod f)
for some a, b, c, dE Z/fZ. In other words, 'IjJ(Pl ) = aPl + cP2 and 'IjJ(P2 ) =
bPl +dP2 . Now, since'IjJ and a commute in their action on E(K), it follows
that their matrices commute in End(E[f]) ~ GL 2 (Z/fZ):
(~ ~) (~ ~) == (~ ~) (~ ~ ) (mod f).
Multiplying this out, we find that a = d and c = 0, so
'ljJt == (~ !) (mod f).

Next we determine the relationship between a and m:
m = deg(l + 'IjJ) - deg('IjJ) - 1 by definition of m
= det(l + 'ljJt) - det('ljJt) - 1 from [AEC, V.2.3]
== det ( 1 ba I! a) - det (~ !) - 1 (mod f)
== 2a (mod f).
This now allows us to compute the degree of m - 2'IjJ, at least modulo e.
deg(m - 2'IjJ) = det(m - 2'IjJt) from [AEC, V.2.3] again
== det [ (~ ~) - 2 (~ !)] (mod f)

== 0 (mod f) since m == 2a (mod f) from above.
We have now proven that deg(m - 2'IjJ) == 0 (mod f) for all primes f
such that (6.2) is true, which means that deg(m - 2'IjJ) == 0 (mod f) for
all but finitely many f's. Hence deg(m - 2'IjJ) = 0, so m = 2'IjJ. But
every endomorphism is integral over Z [AEC, 111.9.4], so m must be even
and 'IjJ E Z. This completes the proof that End(E) = Z. 0
Exercises 449
EXERCISES
5.1. Let q E IC, Iql < 1, and let k E JR. Prove that
where as usual t7k(n) = Ldk.

din
5.2. Let G be a cyclic group of order n, let t7 be a generator for G, and let M
be a G-module.
(a) Prove that
H2 (G M) C>< {x EM: x - t7X = o}

, - {x+t7x+···+t7 n - 1x: xEMr
(This piece of elementary group cohomology is used in the proof of (2.4).)

(b) Prove that
HI(G, M) ~ {x EM: x + t7X + ... + t7n - 1X = O}.

{x - t7X : x E M}
Use this directly to show that for q E JR., the Weil-Chatelet group
has order 1 (respectively 2) if q < 0 (respectively q > 0.)

5.3. Let E/JR be an elliptic curve, let 6(E) be the discriminant of a Weierstrass
equation for E/JR, and let m be an even integer.
(a) Prove that
7../rnZ, if 6(E) < 0,

{
E(R)[m] ~ (7../27..) x (7../m7..), if 6(E) > o.
(b) Consider the Kummer sequence
0--+ E(JR)/mE(R) --+ HI(GC / R , E[m]) --+ WC(E/R) --+ O.
If 6(E) < 0, prove that all three terms are O. If 6(E) > 0, prove that the
sequence is
o --+ 7../27.. --+ 7../27.. x 7../27.. --> 7../27.. --> O.

5.4. (a) Let a, b E JR, and let E /JR be the elliptic curve
E : y2 = x 3 + ax 2 + bx.
If t:.(E) > 0, so in particular if b < 0, prove that WC(E/JR) has order 2,
and its non-trivial element is represented by the homogeneous space
(Hint. See [AEC X.3.7].)

(b) Let E/JR be an elliptic curve with t:.(E) > 0. Exercise 5.3(a) says
that E[2] C E(JR), so we can factor
4x 3 + b2x2 + 2b4 x + b6 = 4(x - eI)(x - e2)(x - e3)

with el < e2 < e3.
Prove that the non-trivial element of WC(E/JR) is represented by the ho-
mogeneous space
5.5. °
Let E/JR be an elliptic curve, and choose q E JR, < Iql < 1, so that j(E) =
j(Eq). Suppose, however, that E is not JR-isomorphic to E q. Then if we
consider the isomorphisms
the second map will not be defined over R Prove that with this identifica-
tion,
5.6. For this problem we will write E(r) for Eq with q = e 27ri -r, and for a
given E/JR we write EX for the non-trivial twist of E. (See Proposi-
tion 2.2(a).)
(a) Prove that
for all t > 0, t =1= 1,
E (-21+ .:..2't)X ~/R E (1'

- + ~)
2 2t '
for all t > 0, t =1= 1,
E(i)X ~/R EG +~).

(b) Fix E/JR with j(E) =1= 1728. From (2.3) there is a unique t > 0 so that
E (y { E(it),. if t:.(E) > 0,
-/R E(~+~), ift:.(E) <0.
Let 'Y(E/JR) = signC6(E) be the invariant defined in (2.2b). Prove that
'Y(E/JR) = sign(I - t).

Exercises 451
5.7. Let KIQp be a finite extension with ring of integers R and normalized
valuation ord v • Let q E K* satisfy Iql < 1, let Eq be the corresponding
Tate curve, and let Eq(K) be the homomorphism described
in (3.1). Prove that for every r ::::: 1, Eq.r(K),
where
R; = {u E K* : ord v (u - 1) ::::: r},
Eq,r(K) = {(x,y) E Eq(K) : ordv(x) S -2r} U {O}.
5.8. (a) Let L I K IQp be a finite tower of fields with L Galois over K, let ai E L
be a sequence of elements such that the series L: ai converges, and let 0' E
G L / K . Prove that
(f>~ir =
i=l
faT.
i=l
(b) Show that (a) is true if we replace Qp by lR and take K = lR and L =

C.
(c) Find a sequence of elements ai E Q such that L: ai = y'2, and deduce
that there is an element 0' E G Q/Q such that
(faJ' # faf.
i=l i=l
Thus (a) is not true if we replace Qp by Q.

5.9. Fill in the details needed to rigorously prove the formulas (i) and (ii) in
Proposition 3.2(b).
5.10. Let K be a p-adic field, let q,q' E K* satisfy Iql < 1 and III < 1, and
let Eq and Eql be the corresponding Tate curves.
(a) If Eq and Eql are isogenous, prove that there are positive integers m, n
such that qTn = qtn .
(b) * Conversely, if qrn = qln for some integers m, n ::::: 1, prove th~t E!j,
and Eq' are isogenous. (Clearly, there are homomorphisms from K* Iq
to j{* Iq'Z, for example u -> urn. What is unclear is that the corresponding
homomorphisms Eq -> E q, are given by rational functions, rather than by
power series.)
5.11. Let K be a p-adic field, and let ElK be an elliptic curve such that Ij (E) I >
1. Let ,(ElK) be the invariant described in (5.2), and consider the field
L = K (V,(EIK»).
(a) Prove ElK has split multiplicative reduction if and only if L = K.
(b) Prove ElK has non-split multiplicative reduction if and only if L I K
is unramified of degree 2.
(c) Prove ElK has additive reduction if and only if LIK is ramified of
degree 2.
(d) Let Eo(K) be the group of points whose reduction is non-singular.
Describe the quotient group E(K)I Eo(K) in each of the three cases (a), (b),
and (c).
5.12. (a) Prove that Lemma 5.2 is still true if K has characteristic 3.
(b) Show that if K has characteristic 2 and E / K is an elliptic curve
with j (E) f 0, then ,( E / K) is always equal to 1. Hence Lemma 5.2
is not true in characteristic 2.
5.13. Let K be a p-adic field, and let E / K be an elliptic curve with split multi-
plicative reduction.
(a) Prove that for each prime l f p there is an exact sequence of GkIK-
modules
1 ~ Ti(p,) ~ Tt(E) ~ Zi ~ 0,
where T£(p,) is the Tate module of K (see [AEC, III.7.3]) and GklK acts
trivially on Z£.
(b) Prove that there is a basis for Ti(E) so that the image of the inertia
group of f< / Kin Aut(Tt(E)) ~ GL 2 (Zt) is equal to
{ (~ ~) E GL 2 (Zt) : ord£(b) ~ ordt(VK(jE))}.

(c) Prove that the exact sequence in (a), considered as a sequence ofGklK-
modules, does not split.
5.14. Let E q / K be a Tate curve over a p-adic field K.
(a) Prove that there is an exact sequence
(b) Prove that
(Hint. For the second isomorphism, use local class field theory.)
(c) It is well known from local class field theory Serre [4J that there is an
isomorphism H2(GkIK' f<-) = Br(K) ~ Q/Z. Prove that the map
obtained by composing this isomorphism with the maps from (a) and (b)
is given by the rule f 1--+ f(q).
(d) Deduce that
WC(Eq/ K) ~ Hom (Eq(K), Q/Z).
In other words, the Weil-Chiitelet group of E q / K is dual to the group of

rational points Eq(K).
(e) -More generally, prove that WC(E/K) ~ Hom (E(K),Q/Z) for any
elliptic curve E/ K satisfying Ij(E)1 > 1.
(In fact, it is true that WC(E/ K) ~ Hom(E(K) , Q/Z) for all elliptic curves
over p-adic fields, not just those with non-integral j-invariant. The proof
of this result, which is due to Tate [5,6J, requires different methods than
those used in this chapter. See also Milne [1])
Exercises 453
5.15. Let KIQp be a p-adic field, let ElK be an elliptic curve, let N 2:' 5 be a
prime not equal to p, and suppose that there is a point P E E(K) of exact
order N.
(a) Prove that E has either good or multiplicative reduction.
(b) Let E -+ E' be an isogeny of elliptic curves whose kernel is the cyclic
subgroup generated by P (i.e., E' = EIZP). Prove that
Here VK : K* -+ Z is the normalized valuation on K, and 'D E / K and'D E I / K

are the minimal discriminants of ElK and E'I K respectively. (Hint. Take
Tate models Eq and Eql and look at the isogeny Eq -+ Eql.)
5.16. Let ElK be an elliptic curve defined over a number field, let N be a
prime, and suppose that there is a point P E E(K) of exact order N. Use
the previous exercise and Szpiro's conjecture (IV.1O.6) to prove that N is
bounded by a constant that depends only on the field K. This approach
to proving the boundedness conjecture [AEC, VIII.7.7j is due to Frey.
CHAPTER VI
Local Height Functions
The canonical height function

h: E(K) ------> [0,00)
is a quadratic form whose value at a point P measures the arithmetic
complexity of P. The importance of the canonical height stems from the
fact that it relates the geometrically defined group law to the arithmetic
properties of the algebraic points on E. See [AEC VIII, §9] for details.
Recall that the ordinary height of a non-zero point P E E(K) (relative
to the function x) is defined as a sum of local terms, one for each absolute
value. Thus
h(P) =
1
2[K: Q] E
vEMK
nv max{ -v(x(P)), o}.
The canonical height is then the limiting value of these ordinary heights,
, 1
h(P) = lim 2h([n]P).
n-+oo n
It is natural to ask whether the canonical height itself can be natu-
rally decomposed as a sum of quadratic forms, one for each absolute value
in K. The answer is "no," but Neron and Tate have shown that there is a
decomposition into local functions which are almost quadratic. Precisely,
they show that for each v E MK there is an almost quadratic function
Av : E(Kv) ,,{O} ------> lR
such that
,
h(P) 1 '"
= [K : Q] ~ nvAv(P) for all P E E(K) " {O}.
vEMK
In §1 of this chapter we will use an averaging argument to prove the
existence of local height functions for all absolute values v E MK, and in
§2 we will prove that the canonical height is equal to the sum of the local
heights. It is also of interest to have explicit formulas for the local height
functions, so we will give such formulas for archimedean absolute values in
§3 and for non-archimedean absolute values in §4. For further information
about local height functions, see for example Lang [3] and Zimmer [2].
§1. Existence of Local Height Functions 455
§1. Existence of Local Height Functions
Let K be a field and let I . Iv be an absolute value on K. The absolute

value can be used to define a topology on K in the usual way: a basis of
open neighborhoods around an element 0: E K is the collection of (open)
balls
all E > O.
Let E / K be an elliptic curve. In a similar way we can define a topology
on E(K).
Definition. The v-adic topology on E(K) is defined as follows. For a
point Po = (xo, Yo) E E(K), a basis of open neighborhoods of Po consists
of the sets
U€={(X,y)EE(K): Ix-xolv<EandIY-Yol<E}, all E > O.

For the point 0 E E(K) at infinity we take as a basis the open neighbor-
hoods
u€ = {(x, y) E E(K) : Ixl v > E- 1 } U {O}, all E > O.

(Notice that for the neighborhoods of 0 there is no need to require both Ixl v
and IYlv to be large. The Weierstrass equation ensures that they simulta-
neously go to 00. For an alternative definition of the v-adic topology, see
exercise 6.1.)
In this section we will prove the existence of almost quadratic local
height functions. These will be certain continuous functions
E(K) " {O} ----> lR,
where E( K) " {O} is given the v-adic topology induced from E( K), and lR
is given its usual topology. The following formulation is due to Tate.
Theorem 1.1. (N eron, Tate) Let K be a field which is complete with
respect to an absolute value I . Iv, and let
v( . ) = -log I . Iv
be the corresponding additive absolute value. Let E / K be an elliptic curve.

Choose a Weierstrass equation for E / K,
E : y2 + alXY + a3Y = x 3 + a2x2 + a4X + a6,

and let ~ be the discriminant of this equation.
(a) There exists a unique function
A: E(K) " {O} ----> lR

456 VI. Local Height Functions
with the following three properties:

(i) A is continuous on E(K)" {O} and is bounded on the complement
of any v-adic neighborhood of O.
(ii) The limit
lim {A(P) + ~v(x(P))}
p v~ic 0
exists.
(iii) For all P E E(K) with [2]P =1= 0,
A([2]P) = 4A(P) + v((2y + alx + a3)(P)) - iv(~).
(b) A is independent of the choice of Weierstrass equation for E / K.

(c) Let L / K be a finite extension and w the extension of v to L. Then
(with the obvious notation)
for all P E E(K) " {O}.
Definition. The function A described in Theorem 1.1 is called the (local)

Neron height function on E associated to v.
Remark 1.1.1. For other properties of A which are equivalent to (iii), see
exercises 6.3 and 6.4.
PROOF (of Theorem 1.1). (a) We begin with uniqueness. Let
A,A': E(K) " {O} ----> lR.
be two functions satisfying (i), (ii), and(iii); and let A = A - X be their

difference. From (ii) we see that the limit
lim A(P)
P--->O
exists; so if we define A(O) to be this limiting value, then (i) implies that
A: E(K) ----> lR.
is a continuous bounded function on all of E( K).

Next we observe from (iii) that A([2]P) = 4A(P) provided [2]P =1= O.
But the points satisfying [2]P = 0 form a discrete subset of E(K) (notice
there are at most four such points), so by continuity A([2]P) = 4A(P)
holds for all P. Iterating this relation N times and dividing by 4N gives
A(P) = 4~A(2N P), valid for all P E E(K) and all N :::: 1.
Since A is bounded, we may let N --+ 00 to deduce that A(P) = O.

Hence A = A', which proves uniqueness.
Before proving the existence of A, which is more complicated, we will
prove (b) and (c).
(b) It is clear that conditions (i) and (ii) are independent of the choice of
Weierstrass equation. Since the quantity
(2y + alx + a3)4
.6.
is invariant under change of coordinates (see [AEC, III §1]), we see that (iii)
is likewise independent. Hence A, if it exists at all, does not depend on the
Weierstrass equation.
(c) Since Aw satisfies conditions (i), (ii), and (iii) for E(L) and w, and
since W restricted to K equals v, we see that Aw satisfies (i), (ii), and (iii)
for E(K) and v. By the uniqueness already proven, the restriction of Aw
to E(K) equals Av.
We turn now to the proof of existence. Property (ii) says that A should
look like ~v 0 X-I, at least close to O. Of course, this is no good for points
with x(P) close to 0, since away from 0, A is supposed to be bounded. So
as a first guess for A we might try the function
1
AI(P) = "2 max { v(X(p)-l), o}.
It turns out that Al almost satisfies property (iii). We first need to make
this precise, and then we will modify Al to produce A.
Let
¢(x) = x4 - b4x 2 - 2b6 x - bs ,
'!f;(x) = 4x 3 + b2x2 + 2b4x + b6 = (2y + alX + a3)2
be the usual functions on E, so the duplication formula [AEC III.2.3(d)]
reads
¢(P)
x(2P) = '!f;(P)'
Define a function
1 1
f(P) = Al ([2]P) - 4AI (P) - "2v( '!f;(p)) + 4v(.6.)
for all P E E(K) with [2]P =1= O.
Notice that if Al were to satisfy (iii), then f would be identically O. We
are going to show that f is bounded.
Using the definition of Al and the duplication formula, we can rewrite f
as
f(P) = ~log (max{I¢(p)/v' I~(P)IJ) + ~v(.6.).
2 max{ Ix(p) lv' 1} 4
(Recall that vet) = -log Itl v .) A priori, f is defined at all points P E E(K)
with [2]P =1= O. The following crucial result shows that more is true.
Lemma 1.2. 'Vith notation as above, 1 extends to a bounded continuous

function on all of E(K).
PROOF. Let
F(P) = max{I¢(P)lv' I~(P)IJ,
max{ Ix(p) Iv' I}
s01= ~log(F)+~v(~). Clearly,FiscontinuousonE(K),,{O}. Further,
as P --; 0, we find
lim F(P) = lim max{I¢(P)lv' ~1f0(P)IJ = l.

P~O )x(Pl)v->= Ix(p)lv
Hence F extends to a continuous, bounded function on E(K).

Since the limit equals 1, we also see that there is a constant Cl > 0,
depending on the chosen Weierstrass equation, such that
So in order to prove that 10g(F) is continuous and bounded on E(K), we

are reduced to showing that F is bounded away from 0 on the set Ix IIi ::; Cl·
From the definition of F, we must show that there is a constant C2 > 0 so
that
for all P E E(K) with Ix(P)lv ::; Cl.
The polynomials ¢(x) and 1jJ(x) are relatively prime in K[x]. We give
two quick proofs of this fact.
Proof 1 (theoretical): The map [2] : E --; E has degree 4 [AEC III.6.2(d)].
From the commutative diagram
[21
E ------> E
1" 11
pI ¢/1/J pI
------>
the rational map ¢/1f0 has degree 4, so ¢(x) and 1f0(x) have no common
roots.
Proof 2 (computational): An explicit computation shows that
Resultant(¢(x),1f0(x)) = ~2,
where ~ is the discriminant of the given Weierstrass equation. Since ~ i= 0,

we see that ¢(x) and 'lj!(x) have no common roots.
Therefore we can find polynomials , 1)1 E K[x] satisfying
¢(x)(x) + 'l/J(x) 1)1 (x) = 1.
Evaluating this identity at x = x(P) and using the triangle inequality yields
the desired result:
1:::; 1¢(x)(x)lv + 11b(x)l)1(x) Iv

:::; 2max{I¢(x)lv' 1'l/J(x)IJ· max{I(x)lv' II)1(x)IJ
:::; c;-l max{ I¢(x) lv' 11/i(x) IJ for Ixl" :::; C1·
D
According to Lemma l.2, the "naive" local height function ),1 satisfies
condition (iii) up to a bounded function. The next proposition shows how
to decompose such a bounded function into a difference of two functions.
Then these new functions will be used to modify ),1 so as to make (iii) hold
exactly.
Proposition 1.3. (Tate) Let
f: E(K) -t JR
be any bounded continuous function. Then there exists it unique hounded

continuous function
Jl : E(K) - t JR
such that
f(P) = 4Jl(P) - Jl([2]P) for all P E E(K).
PROOF. If the function Jl exists, then we can use its defining relation N
times to compute
1 1
Jl(P) = 4: f(P) + 4:Jl([2]P)
I I I
= 4: f (P) + 16 f ([2]P) + 16 Jl ([4jP)
Since Jl is supposed to be bounded, if we let N -+ 00, then the last term

should disappear. So we define Jl by the formula
Jl(P) = f :+
n=O
47 1 f([2 n jP)
and verify that it has the required properties. (Notice that if /-L exists, it
must be given by this formula, so we get uniqueness for free.)
First, since I is a bounded function, it is clear that the series is ab-
solutely convergent, so /-L is well-defined. But more is true. Each of the
functions
is bounded and continuous. (It is easy to check that the multiplication

maps [mJ ; E(K) --> E(K) are continuous for the v-adic topology.) It
follows that the series defining /-L gives a bounded continuous function
on E(K). Finally, using Tate's telescoping series trick,
4/-L(P) - /-L([2JP) = f: 4~/([2nJP) f: 4n~1/([2n+lJP) =

n=O
-
n=O
I(P).
o
We now have all the tools needed to prove the existence of A and so
complete the proof of Theorem 1.1. As above, let
1
Al(P) = "2 max{ v(X(p)-l ),O},
1
I(P) = Al ([2JP) - 4Al(P) - v((2y + alx + a3)(P)) + 4v(.6.).
From Lemma 1.2, I extends to a bounded continuous function (also de-

noted f) on all of E(K). Then Proposition 1.3 gives a bounded continuous
function /-L ; E(K) --> ~ satisfying
f(P) = 4/-L(P) - /-L([2JP) for all P E E(K).
Define
A(P) = Al(P) + /-L(P).
We now verify that A satisfies properties (i), (ii), and (iii) of Theorem 1.1.
(i) By inspection, Al is continuous on E(K) " {O} and is bounded on the
complement of any v-adic neighborhood of O. Since /-L is continuous and
bounded on all of E(K), A satisfies (i).
(ii) If Pis v-adically close to 0, then Al(P) equals -~v(x(P)). We com-
pute
lim {>.(P)
p~o
+ ~v(x(P))}
2
= lim {Al(P)
p~o
+ /-L(P) + -21v(x(P))}
= lim /-L(P) = /-L(O).
p~o
Hence>. satisfies (ii).

§2. Local Decomposition of the Canonical Height 461
(iii) Using the formulas defining and relating A, AI, /-l, and j, we find
A([2]P) = + /-l ([2] p)

Al ([2]P)
= Al ([2]P) - j(P) + 4f-l(P)
= 4Al(P) + v((2y + alx + a3)(P)) - iv(~) + 4f-l(P)
= 4A(P) + v((2y + alx + a3)(P)) - iv(~).
This proves that A verifies (iii) and completes the proof of Theorem 1.1.
o
§2. Local Decomposition of the Canonical Height
The canonical height [AEC VIII §9]
h : E(K) -+ [0, (0)
is a quadratic form defined in terms of the arithmetic of E(K). We now

show that h can be decomposed as a sum of local height functions.
Theorem 2.1. Let K be a number field, MK the standard set of ab-
solute values on K, and nv = [Kv : IQv] the local degree of v E MK.
(See [AEC VIII §5] for a description of MK.) Let E/ K be an elliptic
curve, and for each v E M K , let Av : E(Kv) " {O} ~ lR be the local Neron
height function associated to v as described in (1.1). Then
for all P E E(K) " {O}.
In order to prove Theorem 2.1, we will use the defining properties of A

(especially Theorem 1.1(iii)), together with the following fact, which we
will prove later.
Lemma 2.2. There is a finite set of absolute values S C MK so that for
all v ~ S,
for all P E E(Kv) " {O}.
PROOF. This lemma says that for almost all absolute values, the naive
local height ~ max { v(x- 1 ), a} actually satisfies the quadratic property (iii)
of Theorem 1.1. We will postpone the proof of (2.2) until §4, where we
will prove the more precise result (4.1) that Av = ~ max{ v(x- 1 ), o} for all
finite places v such that the given Weierstrass equation has good reduction.
o
PROOF (of Theorem 2.1). Let S be the set described in (2.2). Define a
function
L : E(K) " {O} ----+ JR,
For any given P E E(K) " {O}, (2.2) implies that
if v rt. Sand v(x(P)) :::: o.

Hence the sum L: nvAv(P) has only finitely many non-zero terms, so L(P)
is well-defined.
Next we compare L(P) with h(x(P)). From Theorem 1.1(i),(ii), for
each v E M K there is a constant Cv so that
for all P E E(Kv) " {O},
and (2.2) allows us to take Cv = 0 for all v rt. S. Now multiply by n v , sum
over v E M K , and divide by [K : Q]. This gives
1
-c:::; L(P) - "2h(x(P)) :::; c for all P E E(K) " {O},
where
is finite and independent of P. In other words, if we set L(O) = 0, then
L(P) = ~h(X(P)) + 0(1) for all P E E(K).
Finally, we verify the quadratic nature of L. Let P E E(K) be a point

with [2]P i- O. Then
L([2]P) = [K: Q]
1
L nv Av([2]P)
vEMK
= [K ~ QJ L nv{ 4Av(P) + v(2y + alX + a3)(P)) + v (.6.) }

vEMK
by Theorem 1.1 (iii)
4
= [K : QJ L nvAv(P) product formula [AEC VIII.5.3]
vEMK
= 4L(P).
Since we have defined L(O) = 0, the relation L([2JP) = 4L(P) holds

for P = 0, too. We must also verify it for P E E[2J, P 1= O. The quickest
way to do this is to use the triplication formula for A (exercise 6.4e) ,
2
Av([3]Q) = 9>'v(Q) + v((3x4 + b2 x 3 + 3b4x 2 + 3b6 x + bs)(Q)) - 3v(~)
for all Q E E(Kv) with [3]Q 1= O.
Summing over v E MK as above gives
L([3]Q) = 9L(Q) for all Q E E(K) with [3]Q 1= O.

In particular, if P 1= 0 and [2]P = 0, then
L(P) = L([3]P) = 9L(P), so L(P) = o.

Hence
L([2]P) = L(O) =0= L(P) = 4L(P).
We have now proven the two relations
L(P) = ~h(x(P)) + 0(1) and L([2]P) = 4L(P) for all P E E(K).
The canonical height h also satisfies these relations [AEC VIII.9.3]. It

follows that the difference F = L - h is bounded and satisfies F([2]P) =
4F(P). Hence
~ 0, so F(P) = 0 for all P E E(K).

N-+oo
Therefore L = h. o
§3. Archimedean Absolute Values - Explicit Formulas
Let K be a field which is complete with respect to an archimedean absolute

value I . Iv, and let E / K be an elliptic curve. Then K is isomorphic to
either lR or C, and I . Iv corresponds to some power of the usual absolute
value. Thus in order to compute the local height function>. over E(K), it
suffices to consider the case that K = C, so in this section we will derive
explicit formulas for the local height for elliptic curves over the complex
numbers.
Recall that an elliptic curve E IC has an analytic parametrization
CIA -> E(C), Z f---> (p(z; A),g;/(z; A)),

with Weierstrass equation
having discriminant
(See [AEC VI.3.6] and (1.4.4).) We put this in standard Weierstrass form
by the substitution
1 ,
x = p(z), y = 2P (z),
yielding the equation
with discriminant ~ = ~(A).
Recall also the Weierstrass a-function (1.5.4)
a(z)=a(z;A) = IT (1_~)e~+~(;~)2,
wEA
w#O
which has a simple zero at each lattice point and satisfies the transformation
formula
a(z + w) = 'Ij;(w)e 17 (W J(z+!w)a(z) for all z E C, w E A.
Here 'Ij; : A --> {±l} is the map with 'Ij;(w) = 1 if and only if w E 2A,
and TJ: A --> C is the quasi-period homomorphism.
We have seen (1.5.6b) that there is a factorization
'( ) __ a(2z)
p z - a(z)4.
Applying log I . I yields
logla(2z) 1 = 4logla(z) 1 + loglp'(z) I·

Since p'(z) = 2y, comparison of this equation with Theorem l.l(iii) (and
the fact that the local height has a pole at 0) suggests that the local height
function on E(C) = CIA should look like -logja(z)l. Unfortunately, the
transformation formula for a(z) shows that la(z)1 is not invariant under
translation by A, so -logla(z) 1 is not well-defined on CIA. The next
proposition explains how to modify a(z) to obtain an invariant function.
Proposition 3.1. Let Ace be a lattice. Extend the quasi-period

map 17 : A ~ C linearly to obtain an JR.-linear homomorphism (also de-
noted 17)
(a) For all z, wEe, the quantity
Z17(W) - W17(z)
is purely imaginary.
(b) Define a function
Then
F(z + w) = 'Ij!(w)e!(Z1J(w)-W1J(Z» F(z) for all z E C, w E A.
(Note that F(z) is not holomorphic, because 17(z) is only JR.-linear.}

(c ) The function IF I
(z ) is a well-defined function on C/ A and is real-
analytic and non-vanishing away from o.
PROOF. (a) Choose a basis Wl,W2 for A with Im(wt/w2) > o. Legendre's
relation (1.5.2d) says that
Write
with a, b, c, d E JR..
Then the JR.-linearity of 17 and Legendre's relation give
Z17(W) - W17(z) = (ad - bC)(Wl17(W2) - W217(Wt})

= (ad - bc)27ri.
(b) Using the transformation formula (1.5.4) for a(z) stated above, we
compute
F(z + w) = e-!(z+w)1J(z+w)a(z + w)
= e-!(z+w)1J(z+w)'Ij!(w )e1J(w)(z+!w) a(z)
= 'Ij!(w)e!(Z1J(w)-W1J(z» F(z).
(c) Since 'Ij!(w) = ±1, and since (a) implies that
it follows from (b) that IF(z)1 is well-defined on CIA. Further, F(z) is

clearly real-analytic on C and vanishes only at points of A, so IF(z)1 is
real-analytic and non-vanishing on C\A. 0
Theorem 3.2. Let EIC be an elliptic curve with period lattice A. Then
the Neron local height function
A: E(C)" {O} ->lR
is given by the formula
A(Z) = -log le-!Z1)(z)a(z)~(A)f2l

1 1
= 2" Re(z1](z)) -logla(z)I-1210gl~(A)I,
where 1] : C ----> C is the extension of the quasi-period map described above

in (3.1).
PROOF. Let A(Z) be the indicated function. We must verify that A satisfies
properties (i), (ii), and (iii) of Theorem 1.1. First, (3.1c) ensures that A
is well-defined on E(C) ,,{O}. Further, since a(z) is holomorphic on C
and non-vanishing on C" A, it is clear that A(Z) is actually a real-analytic
function on E(C) " {O}. Hence it satisfies property (i) of Theorem 1.1.
Next we observe that the limit
. 1
l~A(Z) + 2"V(p(z))
= lim
Z-'O
{I
Re(z1](z)) l
-
2
- l} -logla(z)2p(z)l- -logl~(A)1
2 12
exists, since a(z) has a simple zero and p(z) has a double pole at z = o.
(In fact, the limit equals 112 10gl~(A) I.) This verifies property (ii).
Finally, we must check property (iii). From (I.5.6b) we have
logla(2z)1 = 410gla(z) 1+ loglp'(z) I,
and the linearity of 1](z) gives
Re(2z1](2z)) = 4 Re(z1](z)).
Subtracting the first equation from half the second, and then subtract-
ing 11210gl~(A)1 from both sides, we obtain the desired relation
1
A(2z) = 4A(Z) -loglp'(z)1 + 4Iogl~(A)I·
(Note that p'(z) = 2y and ~(A) = ~ for the Weierstrass equation y2 =

X3 - 4g2x
1 1)
- 4g3 .
This proves that A has properties (i), (ii) ,and (iii) of Theorem 1.1, so A
is the Neron local height function on E(C). 0
Remark 3.2.1. In proving Theorem 3.2, we verified directly that the func-
tion
-log le-!Z'1(Z)O"(z)~(A)i21
has properties (i), (ii), and (iii). This gives an alternative proof of existence
for Theorem 1.1 in the case of archimedean absolute values.
Corollary 3.3. The local height function
.\: E(C) '- {O} ---+ lR
satisfies the quasi-parallelogram law
1
'\(P + Q) + '\(P - Q) = 2A(P) + 2A(Q) + v(x(P) - x(Q)) - 6v(~)
for all P,Q E E(C) with P,Q,P±Q =I O.
(Note that the quantity (x(P) - x(Q))6 / ~ is well-defined, independent of
the choice of a particular lVeierstrass model for E.)
PROOF. The Weierstrass p-function has the factorization (I.5.6a)
( ) _ ( ) __ O"(z+w)O"(z-w) for all z, w E C.

p Z p W - 0"(Z)20"(W)2
Hence
-logIO"(Z+w) I-IogIO"(z-w) 1 = -2IogI0"(z) 1-210gI0"( w) I-Ioglp(z)-p(w) I.
Next, the linearity of TJ(z) immediately implies
(z + w)TJ(z + w) + (z - w)TJ(z - w) = 2zTJ(z) + 2wTJ(w).
Apply ~ Re( . ) to this last equation and add it to the previous one. Com-
parison with the formula (3.2) for A(Z) yields
1
A(Z + w) + A(Z - w) = 2A(Z) + 2A(W) -loglp(z) - p(w)1 + -logl~(A)I,
6
which is exactly the desired identity. o
It is often convenient to use the Fourier expansions for O"(z) and ~(T)
to rewrite the formula for the local height A(Z).
Theorem 3.4. Let E Ie be an elliptic curve with lattice Zr + Z normal-

ized so that Im(r) > O. As usual, let
and identify
z u.
(See (1 §6) and (V §1).) Then the local height function
>. : E(C) " {O} --> IR
where
2 1
B 2 (T)=T -T+ 6
is the second Bernoulli polynomial.
Remark 3.4.1. The formula (3.4) for the local height >.(z) is sometimes
rewritten using the equivalent quantities
Imz log lui v(u)

Imr log Iql v(q)'
PROOF. The Weierstrass l7-function has the product expansion (1.6.4)
The modular discriminant function has the product expansion (1.8.1)
~(T) = (27r)12q II (1 _ qrL)24.

n:2:1
Hence
le-~Z1](Z)l7(z)~(r)+'-1
= le~Z(rl(1)z-1](Z)-211"i) q+'- (1 - u) II (1 - qnu) (1 - qnu- 1 ) \.
n:2:1
To simplify the exponential, we use Legendre's relation (1.5.2d), which

in the case of a normalized lattice ZT + Z says
Writing
z = aT +b with a, b E JR.,
we find
7](l)z - 7](z) - 27ri = a(T7](l) - TI(T)) - hi = 27ri(a - 1).
Hence
e ~ z(1)(1)z-T/(z)-27ri) q-f2 = e ~ (ar+b)·27ri(a-l) . e-f2 ·27rir
= e(a 2-a+i )7rir . e(a-l)b7ri,
so
Substituting this in above yields the formula
le-~ZT/(Z)(}(Z)b..(T)-f2 I = Iq~(a2-a+i)(1_ u) II (1 - qn u)(l- qnU-l)l·

n2:1
Theorem 3.2 says that applying -log(· ) to the left-hand side gives the
local height 'x(z). Since Im(z) = Im(aT + b) = aIm(T), this completes the
proof of Theorem 3.4. 0
§4. Non-Archimedean Absolute Values ~ Explicit Formulas
Let K be a field with absolute value v, and let E / K be an elliptic curve. If

the absolute value v on K is non-archimedean, then we can talk about the
reduction of E modulo v. More precisely, fix a Weierstrass equation for E
with v-integral coefficients (i.e. v(a;) 20). We consider the reduction E of
that Weierstrass equation modulo the maximal ideal of the local ring R =
{o: E K : v(o:) 2 o}. The reduced curve E may be singular. Define a
subset Eo(K) of E(K) by
Eo(K) = {p E E(K) : P is a smooth point of E}.

In particular, if the reduced equation is smooth (E has good reduction),
then Eo(K) = E(K). (See [AEC VII], which discusses in some detail the
case that v is a discrete valuation.)
We now show that for points in Eo(K) the local height is given by a
simple formula.
Theorem 4.1. Let K be a field complete with respect to a non-ar-

chimedean absolute value v, let E j K be an elliptic curve, and choose a
Weierstrass equation for E with v-integral coefficients,
Let .t. be the discriminant of this equation. Then the Neron local height
function A : E(K) " {O} ----+ lR is given by the formula
1 1
A(P) = "2 max{ v(X(p)-l), o} + 12 v(.t.) for all P E Eo(K).
Remark 4.1.1. If E has good reduction, then we can find a Weierstrass

equation for Ej K with Eo(K) = E(K) and v(.t.) = O. In this situation,
the proof of (4.1) will show that the function
1
"2 max{v(x(P)-l), O}
has properties (i), (ii), and (iii) of Theorem 1.1. This provides an alterna-
tive proof of the existence of A in this case. Further, since A is invariant
under finite extension of the field K (LIe) and is independent of the choice
of Weierstrass equation (LIb), we actually obtain an existence proof when-
ever E has potential good reduction.
Remark 4.1.2. The local height A is independent of the choice of Weier-
strass equation. But the formula for A given in Theorem 4.1 does not
appear to be independent of this choice. For example, the change of co-
ordinates x u- 2 x', y = u- 3 y' will alter the formula in Theorem 4.1
to
However, this new formula for A(P) is valid only for points in Eb(K),
where Eb(K) is defined using the equation with coordinates (x', y'). One
can verify that the two formulas agree on the intersection Eo(K) n Eb(K),
as they should.
By inspection, Al satisfies conditions (i) and (ii) of Theorem 1.1. We next

verify condition (iii) for points in Eo(K) ...... {O}.
As usual, let
¢(x) = X4 - b4 x 2 - 2b 6 x - bs,
1jJ(x) = 4x 3 + b2 x2 + 2b 4 x + b6 = (2y + alx + a3)2
be the functions appearing in the duplication formula
¢(P)
x([2]P) = 1jJ(P)'
Then the equation
to be verified is equivalent (after some algebra) to
min{ v(¢(P)), v( 1jJ(P))} = min{ 4v(x(P)), o}.
First suppose that v(x(P)) < O. Then the (non-archimedean) triangle

inequality yields
v(¢(P)) = 4v(x(P)) and v(1jJ(P)) :::: v(4x(p)3) > 4v(x(P)),
so the desired relation is true. We are left to prove
P E Eo(K) and v(x(P)):::: 0 ===? min{v(¢(P)),v(1jJ(P))} = O.
To prove this, we must express the condition P E Eo(K) in terms

of ¢(P) and 1jJ(P). Let
be the polynomial defining E. Recall that a point (xo, YO) E E is singular

if and only if
Fx(xo, YO) = Fy(xo, Yo) = O.
(See [AEC 1.1.5]. The subscripts denote partial derivatives.) Now Eo(K)
consists of all points whose reduction modulo v is non-singular on the
reduced curve E. So we see that
P E Eo(K) ===? v(Fx(P)) ~ 0 or v(Fy(P)) ~ O.
We also recall the addition formula [AEC IlI.2.3(c)], which says that
where m is the slope of the tangent line to E at P. (Note that m i=- 00

since [2]P i=- 0.) Thus
Fx(P)
m = Fy(P) ,
and so we find
( [2]P) = F X (P)2 + G(P)Fy(P)
x F y (P)2
for the polynomial
G = alFx - (a2 + 2X)Fy E R[X, Y] c K(E).
(N.B. G has coefficients in the valuation ring R, since by assumption the
coefficients of F are v-integral.) Thus
¢>=Fi+GFy and 'IjJ=F$.
Now let P E Eo(K) satisfy v(x(P)) 2: O. Then
o ::; min{ v( ¢>(P)), v('IjJ(P)) } since v(x(P)) 2: 0
= min{ v(FX (p)2 + G(P)Fy(P)), v(Fy (p)2)}

= 0 since either v(Fx(P)) ::; 0 or v(Fy(P)) ::; O.
This completes the proof that
1
Al ([2]P) = 4Al (P) + v( (2y + alX + a3)(P)) - 4v(.6.)
for all P E Eo(K) " {O}.
We have now shown that Al satisfies conditions (i), (ii), and (iii) of
Theorem 1.1 for all points in Eo(K) " {O}. If Eo(K) = E(K), that is, if E
has good reduction and we take a minimal Weierstrass equation for E, then
the uniqueness assertion of Theorem 1.1 implies that A = Al. However,
even if the Weierstrass equation for E has singular reduction, the proof of
uniqueness in Theorem 1.1 works for the subgroup Eo(K). A brief sketch
follows.
From (i) and (ii), the difference A = A - Al extends to a bounded,
continuous function on all of Eo(K) (in fact, on all of E(K)). Further,
from (iii) it satisfies A([2]P) = 4A(P) for all P E Eo(K). (By continuity,
this holds even when [2]P = 0.) Then
A(P) = IN A ([2 N ]P) ---t 0, so A(P) = 0 for all P E Eo(K).

4 N_= 0
It remains to find an explicit formula for the local height in the case
that E has bad reduction at v and P is not in Eo(K). Since the local
height is invariant under finite extension of K (LIe), it suffices in principle
to consider the case that E has split multiplicative reduction at v. For
computational purposes, however, it is often more convenient to work di-
rectly over K. Explicit formulas for A in the case of additive reduction are
given in exercises 6.7 and 6.8.
Theorem 4.2. Let K be a p-adic field (i.e., a finite extension ofQp) with
absolute value v = -log I . Iv, let q E K* satisfy Iqlv < 1, and let E q / K be
the Tate curve (V.3.4) with its parametrization
(a) The Neron local height function
(b) If we choose u (by periodicity) to satisfy
O::S: v(u) < v(q),
then
I B (v(u)) () if 0 < v(u) < v(q),
.>.(¢(u)) = { "2 2 v(q) 1 v q ,
v(l - u) + 12 v(q), ifv(u) =0.
(Note that for a Tate curve, v(q) = -v(j(Eq)) = v(~(q)).)

PROOF. (a) The Tate parametrization (V.3.4) is a v-adic analytic map
from K* / qZ to the elliptic curve with Weierstrass equation
defined by
¢(u) = (X(u), Y(u)).

(For the series defining a4(q), a6(q), X(u), and Y(u), see (V.3.1).) The
discriminant of the Weierstrass equation for Eq has the product expan-
sion (V.3.1b)
~(q) = q II
(1- qn)24.
n~l
Recall also the v-adic B-function (V.3.2)

and the factorization (V.3.2b)
uO( U 2 )
2Y(u) + X(u) = - O(U)4 .
Applying v to this relation and doing a little algebra, we find that
{v(O(U 2») - ~V(U2)} = 4 {V(O(U») - ~V(U)} + v(2Y(u) + X(u»).

This suggests that v(O(u») - ~v(u) - 112V(~) would be a good candidate
for A, since it has property (iii), but unfortunately it is not invariant under
the transformation u !---t quo So we make a slight alteration to obtain an
invariant function.
Define
'\(u) = ~B2 (~~~~) v(q) + v(O(u»).
We will show that ,\ is the Neron local height function by verifying that it
satisfies properties (i), (ii), and (iii) of Theorem 1.1.
First, using the identities
v(qu) v(u) 1
----+1 O(qu) = --O(u)
v(q) - v(q) , u
(the last is (V.3.2a», it is easy to check that ,\(qu) = '\(u); so'\ is well-
defined on
Next, the product defining 0 is absolutely convergent and non-zero

away from qZ. Hence A is continuous (in fact, v-adically analytic) on
Eq(K) " {O} and bounded on the complement of any neighborhood of O.
This verifies (i).
To check (ii) we compute
lim
u~l
I
2
} -B
{'\(u) + -v(X(u» 12 2(O)v(q) + 1
= -2 lim V(O(U)2 X(u»
u~l
1
= -B2(O)V(q)
2
+ -12 u~l
.
hm v«l - U)2 X(u»).
The series for X(u) given in (V.3.1) is

so we see that the pole at u = 1 comes only from the n = 0 term. Hence
lim(l- U)2X(U)
u~l
= 1,
which proves that the above limit exists. (In fact, the limit is 112v(q).)
Finally, we verify the duplication formula (iii). Note first that for
all n ::::: 1 we have v(l - qn) = 0, so
v(~(q)) = v(q II (1- qn)24) = v(q).

n;::':l
N ext we add the formula
obtained above to the identity
( -1 (V(U2))2
-- + -1 ) v(q) = 4 (1- (V(U))2
- + -1 ) v(q) - -v(q).
1
2 v(q) 12 2 v(q) 12 4
Since v(q) = v(~(q)), this gives property (iii):
1
A(U 2) = 4A(U) +v(2Y(u) +X(u)) - 4v(.6.(q)).
We have now shown that A(U) satisfies properties (i), (ii), and (iii) of
Theorem 1.1, so it is the Neron local height function.
(b) Since 0 S v(u) < v(q), we have
for all n ::::: 1.
So the formula in (a) becomes
1
A(U) = 'iB2 (v(u))
v(q) v(q) + v(l - u).
If in addition v(u) > 0, then v(l - u) = 0, which gives the first expres-
sion. Similarly, if v(u) = 0, then the second expression is a consequence
of B 2 (0) = 1/6. 0
Remark 4.2.1. The proof of (4.2) shows directly that the function de-
scribed in (4.2a) satisfies conditions (i), (ii), and (iii) of Theorem 1.1, so
we obtain an independent proof of the existence of the Neron local height
for Tate curves. We have now given proofs of the existence of the Neron
local height, independent of the proof in §1, in the following three cases:
(A) K = C, see (3.2.1);

(B) KIQp and ElK has good reduction, see (4.1.1);
(C) KIQp and E = Eq is a Tate curve, see (4.2).
But if K is the completion of a number field with respect to some absolute
value, and ElK is an elliptic curve, then we can find a finite extension L I K
so that ElL falls into one of the three cases (A), (B), or (C). This follows
from [AEC, VII.5.5] and (V.5.3). Now using the elementary fact (1.1c) that
the local height is invariant under field extension, we obtain an independent
proof of the existence of A for all ElK.
EXERCISES
6.1. Let K be a field, I . Ivan absolute value on K, and E/ K an elliptic curve.

For any rational function f E K(E), let
Uj(K) = {P E E(K) : f is defined at Pl.
(a) Prove that the map
is continuous. (Here Uf(K) inherits the v-adic topology from E(K), and JR
is given the usual topology.)
(b) Prove that the topology on E(K) described in §1 is the weakest topol-
ogy (i.e., the topology containing the fewest open sets) such that the maps
in (a) are continuous for every rational function f E K(E).
6.2. Let K be a field, I . Ivan absolute value on K, and E/ K an elliptic curve.
If K is locally compact, prove that E(K) is compact. In particular, E(K)
is compact if K = JR, K = IC, or K is a finite extension of Qp.
6.3. Let K be the completion of a number field with respect to some abso-
lute value, and let E / K be an elliptic curve. Prove that for all P, Q E
E(K) with P, Q, P ± Q =1= 0, the Neron local height>. satisfies the quasi-
parallelogram law
>.(P + Q) + >.(P - Q) = 2>.(P) + 2>.(Q) + v(x(P) - x(Q» - ~v(.6.).
(Hint. We already proved this for K = IC in Corollary 3.3. Going to an

extension field, it suffices to prove the result when K is a finite extension
of Qp and E has either good or split multiplicative reduction. For the
former, the formulas for X3 + X4 and X3X4 in the proof of [AEC, VIII.6.2]
may prove useful, at least if p =1= 2,3; whereas for the latter, you can use
the Tate curve together with (V.3.2b) and (4.2) to mimic the proof of
Corollary 3.3.)
Exercises 477
6.4. Let K and E be as in the previous exercise, and fix a Weierstrass equation
E : y2 + alXY + a3Y = x 3 + azx2 + a4X + a6.
For each integer m define a function
Frn(x) = m 2 IT (x - x(T») E K(E).

TEE[m]
TolD
(a) Prove that
div(Frn) = 2( L (T») - 2mZ(0).

TEE[m]
(b) Prove that

Fz = 4x 3 + bzx 2 + 2b4X + b6 = (2y + alX + a3)z,
F3 = (3X4 + b2x 3 + 3b4x 2 + 3b6x + bS )2,
F4 = Fz(2x 6 + bzx 5 + 5b4x 4 + lOb6x3
+ lObsx z + (b2bs - b4b6)x + b4 bs - b~)2.
Generally, show that there exist functions 1jJm E K (x, y) = K (E) satisfying
Fm = 1jJ;'. (1jJm is the mth-division polynomial; see [AEC, exercise 3.7).)
(c) Let ~ be the discriminant of the given equation. Prove that the func-
tion
~m2-1
is independent of the Weierstrass equation.

(d) Prove the recurrence formula
Fm+1Fm- 1 = (x 0 [m) - x)F;', for all m 2: 2.
By convention, we set H(x) = 1. (Hint. Compare divisors. Then to find
the constant, let P ---> 0.)
(e) Prove that
1 ( ) m - 1 2
>. ( [mJP ) = m 2 >.(P) + 2"v Fm(P) - -1-2-v(~),
for all P E E(K) with [mJP =I=- O.
6.5. Let EjC be an elliptic curve with normalized lattice ZT+Z, and let >.(z) =
>.(x + iy) be the local height function on E(C) " {O}.
(a) Prove that
r
JE(e)
>'(z)dxdy=O.
(Note that this is an improper integral, since >.(z) blows up at z = O. Be

sure to check that the integral converges.)
(b) Prove that >.(z) is a solution of the differential equation
(~:2 + ~:2 ) I:T'

>.(z) =
(This exercise, together with Theorem 1.1(ii), says that >.(z) is the Green's
function on E(C) for the divisor (0).)
6.6. Let K be a field with absolute value v and let E / K be an elliptic curve.
Fix a Weierstrass equation for E,
E : y2 = X3 + Ax + B,
with discriminant and j-invariant
and
(a) If v is non-archimedean, and A and B are v-integral, prove that for

all P E E(Kv) ,,{O},
- 2~ max{v(j(E)-l),O}::::: .>.(P) - ~max{v(x(P)-l),O}::::: 112v (D.).
(Hint. Use the explicit formulas (4.1) and (4.2).)

(b) If v is archimedean, prove that for all P E E(Kv) " {O},
I.>.(p) - ~ max { v (~~~) ,0 } I : : : ~ max{ V(j(E)-l), O} + v(3).

(c) Now suppose that K is a number field, and A and B are in the ring of
integers of K. Prove that for all P E E(K),
Ih(P) - ~h(x(p))1 ::::: ~h(j) + 112h (.6.) + log 3.

(The constants in (b) and (c) are certainly not best possible. See if you
can improve them.)
6.7. The next three exercises give formulas for the local height which are es-
pecially well suited for numerical computations. We begin with the non-
archimedean case.
Let K be a field complete with respect to a discrete valuation v,
let E / K be an elliptic curve, and fix a minimal Weierstrass equation for E,
y2 + alXY + a3Y = x 3 + a2x 2 + a4X + a6.
Let D. be the discriminant of this equation, and let P E E(K).
(a) Prove that P E Eo(K) if and only if either
v((3x 2 + 2a2x + a4 - alY)(P)) ::::: 0, or v((2y + alX + a3)(P)) ::::: o.
Note that if P E Eo(K), then (4.1) says that
.>.(P) = ~ max{ v(X(p)-l), O} + 112 v(.6.).
(b) Assume that v(D.) > 0 and V(C4) = O. (This means that E has multi-
plicative reduction at v; see [AEC VII.5.1bJ.) Let
(P) _ . {v((2 y +al x+a3)(p)) ~}

a - mm v(D.) '2'
Prove that if P rt Eo(K), then the local height of P is given by the formula
1
.>.(P) = "2B2(a(P))v(D.).
Exercises 479
6.8. 'Let K, v, E and D. be as in the previous exercise. Suppose that E has

additive reduction (Le., v(D.) > 0 and V(C4) > 0.) Let P E E(K) with
P fJ- Eo(K). Let F2 and F3 be the polynomials defined in exercise 6.4b.
Prove that
-~V(F2(P» + 112v (D.) ifv(F3(P» 2: 3v(F2(P»,

A(P) ={ 1 1
-16 V(F3(P» + 12 v(D.) otherwise.
6.9. Let EjR be an elliptic curve given by the usual Weierstrass equation
and suppose that x(P) # 0 for all P E E(R). (Note that one can always
achieve this condition by making a shift x = x' + r for sufficiently large T.)
Then the functions
are well-defined for all points in E(R).

(a) Prove that
w(P)
t([2]P) = z(P)'
This gives a convenient recursive formula for computing z([2n]p).

(b) Prove that there are constants Cl, C2 > 0, depending on the Weierstrass
equation, so that
for all P E E(R).
Conclude that the series
L ;n logiz([2n]p)i
00
n=O
is absolutely convergent for all P E E(R).

(c) Prove that the local height function on E(R) " {O} is given by the
formula
A(P)
1
= 2Iogix(P)i-
1
12 log ID.I
1
+8 L4
00 1
n logiz([2 n ]P)i·
n=O
(d) Give a modification of the series in (c) which converges to give the
local height function on E(C) ,,{O}. (Note that in this case there will
always be points with x(P) = 0.)
6.10. Using the previous three exercises, compute the canonical height h(P) of
the indicated point on each of the following curves.
(a) y2+ y =X 3 _x, P=(O,O),
(b) y2+y=X 3 _x 2 , P=(O,O),
(c) y2 + xy + y = x 3 - x 2 - 48x+ 147, P = (13,33),
(d) y2 + xy + y = x 3 + x 2 - 100lx + 12375,
P = (45,224).
6.11. Define the periodic second Bernoulli polynomial B2(t) by
B2(t) = (t - [t]/ - (t - [tD +~

(i.e., B2 equals B2 on the interval 0 :::; t < 1, and is extended periodically
modulo 1 to all of JR).
(a) Prove that B2(t) has the Fourier expansion
(b) Let tl, ... , tN E JR. Prove that

N
B2(ti - tj) :::: -6'
l~i,j'5:N
i#j
(c) Let K be a complete field with discrete valuation v, let ElK be an
elliptic curve, and let ~ be the discriminant of a minimal Weierstrass equa-
tion for E at v. Prove that for any collection of distinct points PI, . .. , P N E
E(K),
L >'(Pi - Pj) :::: - ~ v(~).
l~i,j~N
i#j
(Hint. Since>. is invariant under finite extension of K, it suffices to consider

the two cases of good and split multiplicative reduction.)
(d) Let K be complete with respect to a discrete valuation v, and sup-
pose that E has split multiplicative reduction. Choose a parametriza-
tion E(K) ~ K* Il, and let P E E[N + 1] correspond to ql/(N+I). (Take
any root in k.) For each i, 1:::; i :::; N, let Pi = [i]P. Prove that
L
l~i,j~N
>'(Pi-Pj)=-~V(~)(l- N~I)'
i#j
Thus the estimate in (c) is essentially best possible.

(e) 'Let EIC be an elliptic curve. Prove that there is a constant c = c(E)
so that for any set of distinct points PI, ... , PN E E(C),
l~i,j~N
i"ej
(This is the archimedean analogue of (c). It is quite difficult.)

APPENDIX A
Some Useful Tables
§1. Bernoulli Numbers and «(2k)
Values of the Riemann (-Function at Even Integers, «(8) =~ ~.

~ns
n2':1
1r2 1r4
«(2) = «(4) = 2.3 2 .5
2·3
1r6 1r s
«(6) = 33 .5.7
«(8) = 2.3 .5 2 .7
3
1r1O 6917r 12
«(10) = 35 .5.7.11
«(12) = 36 . 53 . 72 . 11 . 13
21r14
«(14) = 36 . 52 . 7 . 11 . 13
x 00 Xk
Bernoulli Numbers - -
eX -1
= ~ Bk-
~ kl.
k=O
1 1 1
B2 = 6
- B4 = -- B6 =
30 42
1 5 691
Bs = BlO = BI2 =
30 66 2730
7 3617 43867
BI4 = B16 = B I8 =
6 510 798
283·617 11 ·131·593 103·2294797
B 20 = - B22 = B24 = -
330 138 2730
13·657931 7 . 9349 . 362903
B 26 = B 28 = -
6 870
482 A. Some Useful Tables
§2. Fourier Coefficients of ~(T) and jeT)
Fourier Coefficients of (21T)-12~(T) = l: T(n)qn = q IT (1 _ qn)24.

n20 n21
T(l) = 1 T(2) = -24 T(3) = 252

T(4) = -1472 T(5) = 4830 T(6) = -6048
T(7) = -16744 T(8) = 84480 T(9) = -113643
T(10) = -115920 T(l1) = 534612 T(12) = -370944
The function T(n) is called the Ramanujan T function. For values of T(n)
with n ::::: 300, see Lehmer [lJ.
1
Fourier Coefficients of jeT) = - + l:
c(n)qn.
q n20
c(O) = 744 c(1) = 196884

c(2) = 21493760 c(3) = 864299970
c(4) = 20245856256 c(5) = 333202640600
c(6) = 4252023300096 c(7) = 44656994071935
c(8) = 401490886656000
Inversion of Series for j Function, q = l: d( n) -;-;;:-.

1
n21 J
d(l) = 1 d(2) = 744

d(3) = 750420 d(4) = 872769632
d(5) = 1102652742882 d(6) = 1470561136292880
d(7) = 2037518752496883080 d(8) = 2904264865530359889600
§3. Elliptic Curves over IQl with Complex Multiplication 483
§3. Elliptic Curves over Q with Complex Multiplication
In this section we describe all elliptic curves defined over Q with complex
multiplication by an order R = Z + f RK of conductor f in a quadratic
imaginary field K = Q ( J - D) of discriminant - D. The first table gives
the j-invariant for each such order. The second table gives a representative
elliptic curve E over Q with the specified j, together with the minimal
discriminant llE and conductor N E of E. Those curves possessing endo-
morphisms of degree 2 are discussed in (II.2.3.1).
Discriminant Conductor j-invariant

-DofK f of R of E
-3 1 0
2 24 3353
3 _2 15 3.53
-4 1 26 33
2 233311 3
-7 1 -3 353
2 3353173
-8 1 26 53
-11 1 _2 15
-19 1 _2 15 33
-43 1 _2 18 3353
-67 1 _2153353113
-163 1 -2183353233293
Minimal Weierstrass
-D f equation of E over Q
llE NE
-3 1 y2 +y= x3 33 33
2 y2 = x 3 - 15x + 22 2833 2233
3 y2 + y = x 3 - 30x + 63 35 33
-4 1 y2 = x 3 + X 26 26
2 y2 = x 3 - 11x + 14 29 25
-7 1 y2 + xy = x3 - x 2 - 2x - 1 73 72
2 y2 = x3 - 595x + 5586 212 73 24 72
-8 1 y2 = x3 + 4x 2 + 2x 29 28
-11 1 y2 +y = x3 - x 2 - 7x + 10 11 3 112
-19 1 y2 +y = x3 - 38x + 90 193 19 2
-43 1 y2 +y = x3 - 860x + 9707 43 3 43 2
-67 1 y2 +y = x3 - 7370x + 243528 673 67 2
-163 1 y2 +y = x3 - 2174420x + 1234136692 1633 163 2
Notes on Exercises
Many of the exercises in this book are standard results which were not included
in the text due to lack of space, whereas others are special cases of results which
appear in the literature. The following list thus serves two purposes. First, it
is an attempt by the author to give credit for the theorems which appear in the
exercises, and second, it will aid the reader who wishes to delve more deeply into
some aspect of the theory. However, since any attempt to assign credit is bound
to be incomplete in some respects, the author herewith tenders his apologies to
anyone who feels that they have been slighted.
Except for an occasional computational problem, we have not included so-
lutions (nor even hints). Indeed, since it is hoped that this book will lead the
student on into the realm of active mathematics, the benefits of working without
aid clearly outweigh any advantage that might be gained by having solutions
readily available.
CHAPTER I
(1.1) For an elementary proof, see Alperin [1].
(1.5) (e) Let h( -D) denote the class number ofQ (y'-D). Then h( -3) = 1,
h( -5) = 2, h( -23) = 3, h( -29) = 6, h( -47) = 5.
(1.10) See Serre [3, VII, §3.2, Cor. 2].
(1.11) A similar argument is given in Serre [3, VII, Thm. 3] and Apostol [1,
Thm.2.4].
(1.13) See de Shalit [1, II §2, equation (4)]. See also Weil [1, Ch. III, IV].
(1.14) Proven in Stark [2] using Kronecker's limit formula. Alternatively, one
can compare poles and zeros to see that the ratio is constant, and then
let z ----> 0 to find the constant.
(1.16) Answer: s(2,y) = (y-1)(y-5)/24 ifyisodd,and s(2,y) = (y2+5)/24y
if Y is even.
(1.19) (a) See Shimura [1, exercise 3.27]. (b,c) See Shimura [1, Thm. 3.24].
(1.20) (a,b) See Shimura [1, exercise 2.8].
(1.22) See Lang [2, Ch. III §4], Shimura [1, Ch. 3, §§4,5], or Ogg [1].
(1.24) See Serre [3, Ch. VII §4.3].
(1.25) See Apostol [1, Thm. 6.16].
(1.26) This is due to Hecke. See Apostol [1, Thm. 7.20].
(1.29) (c) This is due to Hecke. See Ogg [1, Ch. I, Thm. 1, p. 1-5].
CHAPTER II
(2.1) Write RK = Z + ZT, and for any a E R, write a = a", + b",T with
a", E Z and b", E Z. Then f = min{b", : a E R,b", > O}.
(2.2) This is due to Hurwitz [1]. A nice exposition of the proof is given in
the appendix to Rosen [1].
(2.9) (a) See Shimura [1, (5.4.3), p. 124]. (b) See Shimura [1, exercise 5.8,
p. 124].
(2.12) (b) This exercise was suggested by David Rohrlich.
(2.13) The minimal polynomial of j3 is 27x 8 + 72x 4 - 16.
(2.14) K2 = L2 = K, K3 = K, L3 = K( if.4), K4 = K()3), and
J
L4 = K( -9 + 6)3). Gal(L4/ K) is cyclic of order 4. In computing
L4, the identity -10 - 6)3 = (-1- )3)3 is useful.
(2.17) See Lang [1, Ch. 5, §1].
(2.18) (a,b,c,d) See Lang [1, Ch. 5, §2]' Shimura [1, Ch. 4.6]. (f) See P. Cohen
[1] and Silverman [4].
(2.19) See Lang [1, Ch. 5, §3].
(2.20) See Lang [1, Ch. 5, §2].
(2.21) See Lang [1, Ch. 5, Thm. 2].
(2.25) See Gross [1, Lemma 9.2.5].
(2.30) See Lang [1, Ch. 10, §4, Thm. 10]. For the general case of abelian
varieties, see Shimura [1, Thm. 7.46 and Prop. 7.47].
(2.33) See Ireland-Rosen [1, Ch. 18, §4].
CHAPTER III
(3.4) The Mordell-Weil theorem for abelian varieties over finitely generated
fields is due to Neron. See, for example, Lang [4, Ch. 6, Thm. 1].
(3.11) (c) See also Chapter VI text and exercises for an approach using local
height functions which give a better estimate in (c). One can then use
(c) to prove (a) and (b).
(3.12) [0,1,0], [0, ±T, 1], [T, 0,1]' [1 - T, ±(1 - 2T), 1],
[v'2, ±(2T - 1), 2v'2].
(3.14) See Mumford [1, §6, Lemma on p. 56].
(3.15) In general, the Mordell-Wei! group of an abelian variety is finitely gen-
erated in the following two cases: (i) (Neron) if the field of definition is
finitely generated over Q. (ii) (Lang-Neron) if one takes the quotient
by the subgroup of points defined over the constant field. For details,
see Lang [4, Ch. 6, theorems 1 and 2].
(3.16) This version of (III.11.3.1) and (III.l1.4) for split elliptic surfaces is
due to Dem'janenko [1], with a generalization to abelian varieties due
to Manin [1]. The example in (d) is due to Dem'janenko.
(3.17) See Kuwata [1,2].
(3.20) (a) 2. (b) 3. (c) 2. (d) 4.
486 Notes on Exercises
(3.21) This is a special case of Zariski's Main Theorem, see Hartshorne [1,
V.5.2, III.11.3, exercise III. 11.4).
(3.23) This is a special case of the general fact that algebraically equivalent
divisors are numerically equivalent, see Hartshorne [1, exercise V.1.7].
(3.24) (b) det(Ioo) = n. (c) ai = i(n ~ k)!n if 1::; i::; k, and ai = ken ~ i)!n
if k < i < n. See, for example, Cox and Zucker [1, Table 1.14).
(3.31) See Lang [4, Ch. 4, Prop. 5.2).
(3.32) See Lang [4, Ch. 4, Prop. 3.3 and Cor. 3.4).
(3.34) This estimate is due to Tate [4), see also Lang [4, 12 Cor. 5.4).
(3.36) (b) This is due to Kodaira, see Shioda [3, Prop. 2.8). It has been
frequently rediscovered, see for example Hindry-Silverman [2, Thm. 5.1)
and Szpiro [1).
(3.37) This result is due to Hindry and Silverman [2).
(3.38) For an explicit construction of the Jacobian variety of a hyperelliptic
curve, see Mumford [3, Ch. IlIa). In particular, Pic(C)[2) is described
in Mumford [3, Ch. lIla, Lemma 2.4 and Cor. 2.11).
(3.40) See Shioda [3, Prop. 1.6).
CHAPTER IV
(4.5) (a) See Matsumura [1, corollary to Thm. 45 (18.G)).
(4.7) See Bosch-Liitkebohmert-Raynaud [1, §2.4, Prop. 8).
(4.10) See Shatz [1, §2) or Waterhouse [1, Ch. 1 and 2].
(4.15) (a) This is a restatement of Lemma IV.9.5.
(4.16) e7 : (y2 ~ x 3)(x ~ 1)3(y ~ 2x) = O.
(4.19) See Artin [1, §O).
(4.24) See Bosch-Liitkebohmert-Raynaud [1, Ch. 7, Prop. 6).
(4.25) Combine (IV.5.3) and (IV.9.1). For a more intrinsic proof for general
group schemes, see Artin [1, Lemma 1.16).
(4.26) See Greenberg [1, §3, Lemma 2) for a multi-variable version.
(4.27) (b) See Bosch-Liitkebohmert-Raynaud [1, 2.2, Prop. 7]. (c) See Bosch-
Liitkebohmert-Raynaud [1, 2.3, Prop. 5).
(4.29) See Milne [3, §7) or Wei! [3).
(4.30) See Artin [1, Cor. 1.6).
(4.31) (c) The special fiber has five components, three of multiplicity 1 and
two of multiplicity 2. This is a fiber of Type IX-1 in the classification
of Namikawa and Ueno [1). (d) The special fiber has four components,
one of multiplicity 1, two of multiplicity 2, and one of multiplicity 3.
This is a fiber of Type VIII-3 in the classification of Namikawa and
Ueno [1).
(4.32) (c) Raynaud [1) has proven a general result which allows one to compute
the group of components on the Neron model of the Jacobian of a curve
in terms of the incidence matrix of the special fiber of a minimal proper
regular model of the curve.
(4.35) This exercise is taken from Tate [2, end of §6).
(4.36) (a,b) This is an unpublished result of Mestre; see Kraus [1]. (c) This
is due to Kraus [IJ.
(4.37) See Neron [1, §IIIJ.
(4.39) The proof is essentially the same as the proof of [AEC, IV.6.1]. See
also exercises 2.22 and 2.23.
(4.43) (a) {I}. (b) 7l./27l.. (c) Hs.
(4.44) This follows easily from [AEC, IX.6.1].
(4.45) See Serre [4, Ch. IV, Section 1].
(4.46) (b) See Serre [4, Ch. VI, Thm. 1']. (d) See Ogg [2], Serre-Tate [1, §3],
and Serre [7, Ch. 19]. (e) This is due to Ogg [2].
CHAPTER V
(5.2) See Serre [4, VIII §4].

(5.13) This is due to Serre [1, (A.1.2), pp. IV-31,2.].
(5.14) This approach to Tate's theorem for elliptic curves with non-integral j
is due to Shatz [1].
CHAPTER VI
(6.5) See Lang [6, II §5].

(6.6) (a) This is due to Tate; see Lang [3, III, Thm. 4.5]. (b,c) See Silverman
[3].
(6.7) See Silverman [2, Lemma 5.1].
(6.8) See Silverman [2, Thm. 5.2].
(6.9) The series (c) is due to Tate, and the extension to E(C) is due to
Silverman. For proofs with error estimates, see Silverman [2].
(6.10) Solutions: (a) 0.02555 ... (b) j~(r) = 0, r is a 5-torsion point, (c)
0.01028 ... , this point has very small height, (d) 0.01049 ... , this point
has very small ratio h(?)/ log(6.).
(6.11) (b) This is due to Blanksby and Montgomery; see Hindry-Silverman
[1]. (c) Hindry-Silverman [1]. (e) This is due to Elkies, see Lang [6, VI
§5]. One can give the explicit lower bound
1 1
-2N log N - 12Nmax{log Ij(E)I, O} - 3.64N,
where the 3.64 is not best possible.

References
[AEC] The Arithmetic of Elliptic Curves, GTM 106, J.H. Silverman, Springer-
Verlag, New York, 1986.
Abhyankar, 8.8.
[1] Resolution of singularit.ies of arithmetical surfaces. In Arithmetical
Algebraic Geometry, Harper and Row, New York, 1965.
[2] Resolution of singularities of algebraic surfaces. In Algebraic Geometry,
Oxford University Press, London, 1969, I-II.
Ahlfors, L.
[1] Complex Analysis: An Introduction to the Theor'y of Analytic Functions
of One Complex Variable, 3rd ed., McGraw-Hill, New York, 1979.
Alling, N.
[1] Real Elliptic Curves, North-Holland, New York, 1981.
Alperin, R.
[1] PSL 2(£:) = £:2 * £:3. Amer. Math. Monthly 100 (1993), 385-386.
Altman, A. and Kleiman, 8.
[1] Introduction to Grothedieck Duality Theory, Lect. Notes in Math. 146,
Springer-Verlag, Berlin, 1970.
Apostol, T.
[1] 1110dular Functions and Dirichlet Series in NumbeT Theor'y, GTM 41,
Springer-Verlag, New York, 1976.
Arakelov, 8.
[1] Intersection theory of divisors on an arithmetic surface. Izv. Akad. Nauk
SSR Ser. Mat. 38 (1974), AMS Transl. 8 (1974), 1167-1180.
Artin, M.
[1] Neron models. In Arithmetic Geometr'y, G. Cornell and J. Silverman,
eds., Springer-Verlag, New York, 1986, 213-230.
[2J Lipman's proof of resolution of singularities for surfaces. In Arithmetic
Geometry, G. Cornell and J. Silverman, eds., Springer-Verlag, New
York. 1986, 268-287.
Atiyah, M.F. and MacDonald, I.G.
[1] Introd'uction to Commutative Algebra, Addison-Wesley, Reading, lvfA,
1969.
References 489
Baker, A.
[1] Transcendental Number Theory, Cambridge Univ. Press, Cambridge,
1975.
Beauville, A.
[1] Complex Algebraic Surfaces, London Math. Soc. Lect. Note 68, Cam-
bridge Univ. Pres::;, Cambridge, 1983.
Birch, B.J. and Kuyk, W., eds.
[1] Modular Functions of One Variable IV, Lect. Notes in Math. 476, Sprin-
ger-Verlag, Berlin, 1975.
Birch, B.J. and Swinnerton-Dyer, H.P.F.
[1] Elliptic curves and modular functions. In Modular Functions of One
Variable IV, Lect. Notes in Math. 476, B.J. Birch and W. Kuyk, eds.,
Springer-Verlag, Berlin, 1975, 2-32.
Borel, A., Chowla, S., Herz, C.S., Iwasawa, K. and Serre, J.-P.
[1] Seminar on Complex Multiplication, Lect. Notes in Math. 21,
Bosch, S., Gunter, U. and Remmert, R.
[1] Non-Archimedean Analysis: A Systematic Approach to Rigid Analytic
Geometry, Springer-Verlag, Berlin, 1984.
Bosch, S., Lutkebohmert, W. and Raynaud, M.
[lJ Neron Models, Springer-Verlag, Berlin, 1990.
Brumer, A. and Kramer, K.
[lJ The conductor of an abelian variety. Compositio Math. (1994), to
appear.
Cassou-Nogues, Ph. and Taylor M.J.
[lJ Elliptic Functions and Rings of Integers, Birkhiiu::;er, Boston, 1987.
Chinburg, T.
[IJ An introduction to Arakelov intersection theory. In Arithmetic Geom-
etry, G. Cornell and J. Silverman, eds., Springer-Verlag, New York,
1986, 291-307.
[2J Minimal models for curves over Dedekind rings. In Arithmetic Geometry,
C. Cornell and J. Silverman, eds., Springer-Verlag, New York, 1986,
309-326.
Coates, J.
[1] Elliptic curves and Iwasawa theory. In Modular Forms, R.A. Rankin,
ed., Ellis Horwood Ltd., Chichester, 1984, 51-74.
Cohen, H.
[1] A Course in Computational Algebraic Number Theory, Springer-Verlag,
Berlin, 1993.
Cohen, P.
[lJ On the coefficients of the transformation polynomials for the elliptic
modular function. Math. Proc. Camb. Philos. Soc. 95 (1984), 389-402.
Conway, J .H.
[IJ Monsters and moonshine. Math. Intelligencer 2 (1979/80), 165-171.
Conway, J.H. and Norton, S.
[1] Monstrous moonshine. Bull. London Math. Soc. 11 (1979), 308-339.
490 References
Cox, D. and Zucker, S.

[1] Intersection numbers of sections of elliptic surfaces. Invent. Math. 53
(1979), 1-44.
Crernona, J .E.
[1] Algorithms for Modular Elliptic Curves, Cambridge Univ. Press, Cam-
bridge, 1992.
Deligne, P.
[1] Formes modulaires et representations l-adic. In Sem. Bourbaki, 21e
annee, 1968/69, no. 355, Lect. Notes in Math. 179, Springer-Verlag,
Berlin, 1971, 139-172.
[2] La conjecture de Weil I. Publ. Math. IHES 43 (1974), 273-307.
Dern'janenko, V.A.
[1] Rational points of a class of algebraic curves. Izv. Akad. Nauk SSSR
Ser. Math. 30 (1966), 1373-1396. AMS Transl. 66 (1968), 246-272.
[2] An estimate of the remainder term in Tate's formula. Mat. Zametki 3
(1968), 271-278. Math. Notes 3 (1968), 173-177.
de Shalit, E.
[1] Iwasawa theory of elliptic curves with complex multiplication, Academic
Press, Boston, 1987.
Eisenbud, D. and Harris, J.
[1] Schemes: The language of modem algebraic geometry, Wadsworth &
Brooks/Cole, Pacific Grove, CA, 1992.
Elkies, N.
[1] The existence of infinitely many supersingular primes for every elliptic
curve over Q. Invent. Math. 89 (1987), 561-567.
Faltings, G.
[1] Calculus on arithmetic surfaces. Ann. Math. 119 (1984), 387-424.
Ferrnigier, S.
[1] Un exemple de courbe elliptique definie sur Q de rang > 19. CRAS
Serie 1 315 (1992), 719-722.
Greenberg, M.
[1] Rational points in Henselian discrete valuation rings. Publ. Math. IHES
31 (1966), 563-567.
Griffiths, P. and Harris, J.
[1] Principles of Algebraic Geometry, Wiley, New York, 1978.
Gross, B.
[1] Arithmetic on Elliptic Curves, Lect. Notes in Math. 776, Springer-Ver-
lag, Berlin, 1980.
Gross, B., Kohnen, W. and Zagier, D.
[1] Heegner points and derivatives of L-series. Math. Ann. 278 (1987),
497-562.
Gross, B.H. and Zagier, D.B.
[1] Heegner points and derivatives of L-series. Invent. Math. 84 (1986),
225-320.
Grosswald, E. and Rachrnacher, H.
[1] Dedekind Sums, Carus Math. Monograph, Math. Assoc. of America,
Providence, RI, 1972.
References 491
Harris, J.
[1] Algebraic Geometry: A First Course, Springer-Verlag, New York, 1992.
Hartshorne, R.
[1] Algebraic Geometry, Springer-Verlag, New York, 1977.
Heegner, K.
[1] Diophantische Analysis und Modulfunktionen. Math. Zeit. 56 (1952),
227-253.
Hindry, M. and Silverman, J.H.
[1] Introduction to Diophantine Geometry, in preparation.
[2] The canonical height and integral points on elliptic curves. Invent.
Math. 93 (1988), 419-450.
Hurwitz, A.
[1] Uber die Entwicklungskoeffizienten der lemniscatischen Funktionen. In
Mathematische Werke, vol. 2, Birkhiiuser, Basel, 1962, 342-373.
Ireland, K. and Rosen, M.
[1] A Classical Introduction to Modern Number Theory, GTM 84, Sprin-
ger-Verlag, New York, 1982.
Katz, N. and Mazur, B.
[1] Arithmetic Moduli of Elliptic Curves, Princeton Univ. Press, Princeton,
NJ,1985.
Kenku, M.A.
[1] On the number of Ql-isomorphism classes of elliptic curves in each Ql-
isogeny class. J. Number Theory 15 (1982), 199-202.
Knapp, A.
[1] Elliptic Curves, Math. Notes 40, Princeton Univ. Press, Princeton, NJ,
1992.
Koblitz, N.
[1] Introduction to Elliptic Curves and Modular Forms, Springer-Verlag,
New York, 1984.
Kodaira, K.
[1] On the structure of compact complex analytic surfaces I, II. Amer. J.
Math. 86 (1964), 751-798; 88 (1966), 682-721.
Kolyvagin, V.A.
[1] Finiteness of E(Ql) and IlI(Ql) for a class of Wei! curves. Math. USSR
Izv. 32 (1989), 523-542.
Kraus, A.
[1] Quelques remarques a propos des invariants C4, C6, et b. d'une courbe
elliptique. Acta Arith. 54 (1989), 75-80.
Kubert, D. and Lang, S.
[1] Modular Units, Springer-Verlag, New York, 1981.
Kuwata, M.
[1] The field of definition of the Mordell-Wei I group of an elliptic curve over
a function field. Compos. Math. 76 (1990), 399-406.
[2] Ramified primes in the field of definition for the Mordell-Wei! group of
an elliptic surface. Proc. AMS 116 (1992), 955-959.
492 References
Lang, S.
[1] Elliptic Functions, GTM 112, Springer-Verlag, New York, 1987.
[2] Introduction to Modular Forms, Springer-Verlag, Berlin, 1976.
[3] Elliptic Curves: Diophantine Analysis, Springer-Verlag, Berlin, 1978.
[4] Fundamentals of Diophantine Geometry, Springer-Verlag, New York,
1983.
[5] Algebraic Number Theory, GTM 111, Springer-Verlag, New York, 1986.
[6] Introduction to Arakelov Theory, Springer-Verlag, New York, 1988.
[7] Algebra, 2nd edition, Addison-Wesley, Menlo Park, CA, 1984.
[8] Integral points on curves. Publ. Math. IHES 6 (1960), 27-43.
Laska, M.
[1] An algorithm for finding a minimal Weierstrass equation for an elliptic
curve. Math. Compo 38 (1982), 257-260.
Lehmer, D.H.
[1] Ramanujan's function T(n). Duke Math. J. 10 (1943), 483-492.
Lehner, J.
[1] Divisibility properties of the Fourier coefficients of the modular invari-
ant j(T). Amer. J. Math. 71 (1949), 136-148.
[2] Further congruence properties for the Fourier coefficients of the modular
invariant j(T). Amer. J. Math. 71 (1949), 373-386.
Lenstra, H.W.
[1] Factoring integers with elliptic curves. Ann. of Math. 126 (1987),
649-673.
Lichtenbaum, S.
[1] Curves over discrete valuation rings. Amer. J. Math. 90 (1968),
380-403.
Lipman, J.
[1] Rational singularities with applications to algebraic surfaces and unique
factorization. Publ. Math. IHES 38 (1970), 195-279.
[2] Desingularization of two-dimensional schemes. Ann. Math. 101 (1978),
151-207.
Liu, Q.
[1] Formule d'Ogg d'apres Saito. preprint.
Lockhart, P., Rosen, M. and Silverman, J.
[1] An upper bound for the conductor of an abelian variety. J. Alg. Geo. 2
(1993), 569-60l.
Manin, Ju.
[1] The Tate height of points on an abelian variety. Its variants and appli-
cations. Transl. AMS 59 (1966), 82-110.
[2] The p-torsion of elliptic curves is uniformly bounded. Izv. Akad. Nauk
SSSR 33 (1969), 433-438.
[3] Rational points on an algebraic curve over function fields. Transl. AMS
50 (1966), 189-234.
Mason, R.C.
[1] The hyperelliptic equation over function fields. Math. Proc. Camb.
Philos. Soc. 93 (1983), 219-230.
Matsumura, H.
[1] Commutative Algebra 2nd ed., Benjamin/Cummings, Reading, MA,
1980.
References 493
Mazur, B.
[1] Modular curves and the Eisenstein ideal. Publ. Math. IHES 47 (1977),
33-186.
[2] Rational isogenies of prime degree. Invent. Math. 44 (1978), 129-162.
Mestre, J.-F.
[1] Formules explicites et minorations de conducteurs de varieties alge-
briques. Compos. Math. 58 (1986), 209-232.
[2] Courbes elliptiques de rang 2 11 sur Q(T), Courbes elliptiques de rang
2 12 sur Q(T), Un example de courbes elliptiques sur Q de rang 2 15.
C.R. Acad. Sci. Paris 313 (1991), 139-142; 313 (1991), 171-174;
314 (1992), 453-455.
Milne, J.8.
[1] Arithmetic Duality Theorems, Academic Press, Boston, 1986.
[2] Abelian varieties. In Arithmetic Geometry, G. Cornell and J. Silverman,
[3] Jacobian varieties. In Arithmetic Geometry, G. Cornell and J. Silverman,
[4] Etale Cohomology, Princeton University Press, Princeton, NJ, 1980.
Miranda, R.
[1] The basic theory of elliptic surfaces, Dottorato di Ricerca in Matematica,
ETS Editrice, Pisa, 1989.
Mumford, D.
[1] Abelian Varieties, Oxford Univ. Press, Oxford, 1974.
[2] Curves and their Jacobians, The University of Michigan Press, Ann
Arbor, MI, 1975.
[3] Tata Lectures on Theta II, Prog. in Math. 43, Birkhauser, Basel, 1984.
Mumford, D. and 8uominen, K.
[1] Introduction to the theory of moduli. In Algebraic Geometry, Oslo 1970,
F. Oort, ed., Wolters-Noordhoff, Groningen, 1972,
Nagao, K.
[1] An example of an elliptic curve over Q with rank 2 20. Pmc. Japan
Acad.69 (1993), 291-293.
[2] An example of an elliptic curve over Q(T) with rank 2 13. Pmc.Japan
Acad. 70 (1994), 152-153.
Nagao, K. and Kouya, T.
[1] An example of an elliptic curve over Q with rank 2 21. Proc. Japan
Acad. 70 (1994), 104-105.
Namikawa, Y. and Veno, K.
[1] The complete classification of fibres in pencils of curves of genus two.
Manuscripta Math. 9 (1973), 143-186.
Neron, A.
[1] Modeles minimaux des varieMs abeliennes sur les corps locaux et glo-
baux. Publ. Math. IHES 21 (1964), 361-482.
[2] Quasi-fonctions et hauteurs sur les varieMs abeliennes. Ann. of Math.
82 (1965), 249-331.
[3] Problemes arithmetiques et geometriques rat taches a la notion de rang
d'une courbe algebrique dans un corps. Bull. Soc. Math. France 80
(1952), 101-166.
494 References
Neukirch, J.
[lJ Class Field Theory, Grund. der Math. Wiss. 280, Springer-Verlag,
Berlin, 1986.
Ogg, A.
[lJ Modular Forms and Dirichlet Series, Benjamin, New York, 1969.
[2J Elliptic curves and wild ramification. Amer. J. Math. 89 (1967), 1-21.
[3J On pencils of curves of genus two. Topology 5 (1966), 355-362.
Perrin-Riou, B.
[lJ Arithmetique des courbes elliptiques et theorie d'Iwasawa. Mem. Soc.
Math. France 17 (1984), 1-130.
Petersson, H.
[lJ Uber die Entwicklungskoeffizienten der automorphen formen. Acta
Math. 58 (1932), 169-215.
[2J Uber eine Metrisierung der ganzen Modulformen. Jber. Deutsche Math.
49 (1939), 49-75.
Raynaud, M.
[lJ Specialisation du foncteur de Picard. Publ. Math. IHES 38 (1970),
27-76.
Robert, A.
[lJ Elliptic Curves, Lect. Notes in Math. 326, Springer-Verlag, Berlin, 1973.
Roquette, P.
[lJ Analytic Theory of Elliptic Functions Over Local Fields, Vandenhoeck
& Ruprecht, Gottingen, 1970.
Rosen, M.
[lJ Abel's theorem on the lemniscate. Amer. Math. Monthly 88 (1981),
387-395.
Rubin, K.
[lJ Tate-Shafarevich groups and L-functions of elliptic curves with complex
multiplication. Invent. Math. 89 (1987), 527-560.
Saito, T.
[lJ Conductor, discriminant, and the Noether formula of arithmetic sur-
faces. Duke Math. J.57 (1988), 151-173.
Schmidt, W.
[IJ Thue's equation over function fields. Aust.Math. Soc. Gaz. 25 (1978),
385-422.
Schneider, Th.
[IJ Introduction aux Nombres Transcendents, Grund. der Math. Wiss. 81,
Serre, J .-P.
[1 J Abelian C-adic Representations and Elliptic Curves, Benjamin, New
York,1968.
[2J Proprietes galoisiennes des points d'ordre fini des courbes elliptiques.
Invent. Math. 15 (1972), 259-331.
[3J A Course in Arithmetic, GTM 7, Springer-Verlag, New York, 1973.
[4J Local Fields, trans!. by M. Greenberg, Springer-Verlag, New York, 1979.
[5J Local class field theory. In Algebraic Number Theory, J.W.S. Cassels
and A. Frohlich, eds., Academic Press, London, 1967, 129-162.
[6J Complex multiplication. In Algebraic Number Theory, J.W.S. Cassels
References 495
Serre, J .-P. (continued)

[7J Linear Representations of Finite Groups, GTM 42, Springer-Verlag,
New York, 1977.
Serre, J .-P. and Tate, J.
[IJ Good reduction of abelian varieties. Ann. Math. 68 (1968), 492-517.
Shafarevich, I.R.
[IJ Basic Algebraic Geometry, Springer-Verlag, New York, 1977.
[2J Lectures on Minimal Models, Tata Institute, Bombay, 1966.
Shatz, S.
[IJ The cohomology of certain elliptic curves over local and quasi-local
fields. Illinois J. Math. 11 (1967), 234-241.
[2J Group schemes, formal groups, and p-divisible groups. In Arithmetic
Geometry, G. Cornell and J. Silverman, eds., Springer-Verlag, New
York, 1986, 29-78.
Shimura, G.
[IJ Introduction to the Arithmetic Theory of Automorphic Forms, Princeton
Univ. Press, Princeton, NJ, 1971.
Shimura, G. and Taniyama, Y.
[IJ Complex Multiplication of Abelian Varieties and its Application to Num-
ber Theory, Publ. Math. Soc. Japan 6, 1961.
Shioda, T.
[IJ The Galois representation of type Es arising from certain Mordell-Weil
groups. Pmc. Japan. Acad. 65 (1989), 195-197.
[2J On Mordell-Weillattices. Univ. Sancti Pauli 39 (1990), 211-240.
[3J On elliptic modular surfaces. J. Math. Soc. Japan 24 (1972), 20-59.
[4J Mordell-Weillattices and Galois representations I, II, III. Pmc. Japan
Acad. 65 (1989), 268-271, 296-299, 300-303.
[5J Mordell-Weillattices and sphere packings. Amer. J. Math. 113 (1991),
931-948.
[6J Mordell-Weil lattices of type Es and deformation of singularities. Lect.
Notes Math. 1468 (1991), 177-202.
[7J Construction of elliptic curves with high rank via the invariants of the
Weyl groups. J. Math. Soc. Japan 43 (1991), 673-719.
Siegel, C.L.
[IJ A simple proof that 7J( -1/T) = 7J(T)F/i. Mathematika 1 (1954), 4.
Silverman, J.H.
[AEC] The Arithmetic of Elliptic Curves, GTM 106, Springer-Verlag, New
York, 1986.
[IJ Heights and the specialization map for families of abelian varieties. J.
Reine Angew. Math. 342 (1983), 197-211.
[2J Computing heights on elliptic curves. Math. Comp.51 (1988), 339-358.
[3J The difference between the Weil height and the canonical height on
elliptic curves. Math. Compo 55 (1990), 723-743.
[4] Hecke points on modular curves. Duke Math. J. 60 (1990), 401-423.
[5J Variation of the canonical height on elliptic surfaces I. J. Reine Angew.
Math. 426 (1992), 151-178.
[6] Variation of the canonical height on elliptic surfaces II, III. J. Number
Theory (1994), to appear.
496 References
Silverman, J .H. (continued)

[7J The Neron-Tate Height on Elliptic Curves. Ph.D. thesis, Harvard, 198!.
[8J The S-unit equation over function fields. Math. Pmc. Camb. Philos.
Soc. 95 (1984), 3-4.
Stark, H.
[IJ A complete determination of the complex quadratic fields of class-num-
ber one. Michigan Math. J. 14 (1967), 1-27.
[2J The Coates-Wiles theorem revisited. In Number Theory Related to
Fermat's Last Theorem, N. Koblitz, ed., Birkhiiuser, Boston, 1982,
349-362.
Szpiro, L.
[1 J Seminaire sur les pinceaux de courbes de genre au moins deux. Aste-
risque 86 (1981), 44-78.
Tate, J.
[IJ The arithmetic of elliptic curves. Invent. Math. 23 (1974), 171-206.
[2J Algorithm for determining the type of a singular fiber in an elliptic
pencil. In Modular Functions of One Variable IV, Lect. Notes in Math.
476, B.J. Birch and W. Kuyk, eds., Springer-Verlag, Berlin, 1975,
33-52.
[3J Letter to J.-P. Serre (1979), unpublished.
[4J Variation of the canonical height of a point depending on a parameter.
Amer. J. Math. 105 (1983), 287-294.
[5J WC-groups over p-adic fields. Sem. Bourb. Expose 156 (1957/58),
13pp.
[6J Duality thoerems in Galois cohomology. Pmc. Intern. Congress Math.,
Stockholm, 1962, 234-24l.
[7J Global class field theory. In Algebmic Number Theory, J.W.S. Cassels
[8J Fourier analysis in number fields and Heeke's zeta-functions. In Algebraic
Number Theory, J.W.S. Cassels and A. Frohlich, eds., Academic Press,
London, 1967, 305-347.
[9J A review of non-archimedean elliptic functions. In Elliptic Curves, Mod-
ular Forms, & Fermat's Last Theorem, J. Coates and S.T. Yau, eds.,
International Press, Boston, 1995, 162-184.
Vladu1;, S.G.
[IJ Kronecker's Jugendtraum and modular functions, Gordon and Breach,
New York, 199!.
Vojta, P.
[IJ Diophantine approximations and value distribution theory, Lect. Notes
in Math. 1239, Springer-Verlag, Berlin, 1987.
Voloch, J.F.
[IJ Siegel's theorem for complex function fields. Proc. AMS 121 (1994),
1307-1308.
Waldschmidt, M.
[IJ Nombres transcendent et groupes algebriques, Asterisque 69-70, 1979.
Waterhouse, W.
[IJ Introduction to Affine Group Schemes, GTM 66, Springer-Verlag, New
York,1979.
References 497
Weil, A.
[1] Elliptic Functions According to Eisenstein and Kronecker, Springer-Ver-
lag, Berlin, 1976.
[2] Arithmetic on algebraic varieties. Ann. Math. 53 (1951), 412-444.
[3] Varietes Abeliennes et Courbes Algebriques, Hermann, Paris, 1948.
Zimmer, H.
[1] On the difference of the Weil height and the Neron-Tate height. Math.
Z. 147 (1976), 35-51.
[2] Quasifunctions on elliptic curves over local fields. J. reine angew. Math.
307/308 (1979), 221-246; Corrections and remarks concerning quasi-
functions on elliptic curves. J. reine angew. Math. 343 (1983), 203-211.
List of Notation
EA the elliptic curve y2 = 4x 3 - 92(A)x - 93(A), 6

p(z; A) Weierstrass p function relative to the lattice A, 6
,c the set of lattices in 1[, 6
H the upper half plane {T E 1[: Im(T) > O}, 7
SL 2(7.) special linear group over 7., 8
AT the lattice 7.T + Z, 9
r(1) the modular group, 10
S the element r/) E r(1), 10
(~
T the element (6 i) E r(l), 10
:r fundamental domain for action of SL2(7.) on H, 10
I(T) the stabilizer of T, 11
X(I) modular curve, 14
H* extended upper half plane, 14
Y(I) affine modular curve, 14
X(I) projective modular curve, 14
I(Tl, T2) transformations sending Tl to T2, 17
I(Ul, U2) transformations "( with "(Ul n U2 =I- 0, 17
'ljJx local parameter for X (1), 20
j j-invariant map X(I) ---> jp'1(1C), 23
ord=(f) order at 00 of a modular function, 24
f(oo) value at 00 of a modular function, 24
G 2 k(A) Eisenstein series, 25
G2k(T) Eisenstein series, 25
92 equals 60G4 , 26
93 equals 140G6 , 26
~(T) the modular discriminant, 26
12x space of differential I-forms on X, 27
121: space of meromorphic k-forms on X, 27
ordx(w) order of a differential form, 27
Wj differential form on X(I) attached to modular form f, 28
M2k space of modular forms of weight 2k, 31
Mg k space of cusp forms of weight 2k, 31
c'c'cc isomorphism classes of elliptic curves over 1[, 36
«(z; A) Weierstrass ( function for the lattice A, 39
1] quasi-period homomorphism, 41
o-(z; A) Weierstrass 0- function for the lattice A, 44
1j; character describing periodicity of Weierstrass 0- function, 44
AT normalized lattice Z7 + Z, 47
8J(Z;7) Weierstrass 8J function for the normalized lattice AT, 47
«(Z;7) Weierstrass ( function for the normalized lattice AT, 47
0-(Z;7) Weierstrass 0- function for the normalized lattice AT, 47
U = Uz = e 27riz , 47
q = qT = e 27riT , 47
qZ the cyclic group generated by q, 47
F(u;q) an elliptic function, 49
o-k (n) sum of kth powers of divisors of n, 55
Bk Bernoulli number, 57
E2k(7) normalized Eisenstein series, 58
T(n) Ramanujan 7 function, 59
c(n) Fourier coefficients of the modular j function, 59
1](7) Dedekind 1] function, 65
.ph) integer in Dedekind 1] function transformation formula, 66
SeX, y) Dedekind sum, 66
Div(S) divisor group of a set, 68
T(n) nth Hecke operator, 68
R>. homothety operator, 68
A'CA A' is a sublattice of A of index n, 69
M2(Z) ring of 2 x 2 matrices with integral coefficients, 72
Q(A) the lattice Z(awl + bW2) + Z(CWI + dW2), 72
TIn integer matrices of determinant n, 72
Sn special integer matrices of determinant n, 72
Fj lattice function associated to the modular function I, 74
IF modular function associated to the lattice function F, 74
T2k(n) nth Hecke operator on space of modular functions, 76
"((m) mth Fourier coefficient of T 2k (n)I, 76
L(f, s) formal Dirichlet series attached to the power series I, 80
R(f, s) normalized L-series attached to I, 83
r(s) gamma function, 83
r(N) a congruence subgroup of rei), 86
ro(N) a congruence subgroup of r(I), 87
rl(N) a congruence subgroup of r(I), 87
qD) linear series attached to the divisor D, 87
Weil pairing, 89
/1>(Q,7) = C7 + d, where = ( ~ ~), 91
Q
II[Qhk = (detQ)k/1>(Q,7)-2kIoQ, 91
(j,g) Petersson inner prod uct of I and g, 92
r(s,x) incomplete gamma function, 93
g(X) Gauss sum associated to X, 93
I(x,7) twist of the cusp form I by the character X, 93
L(j, x, s) twist of the L-series L(j, s) by the character X, 93
RK ring of integers (maximal order) of K, 96
EA the elliptic curve isomorphic to iC/ A, 97
500 List of Notation
[.] normalized isomorphism R --> End(E), 97

eJ:.,qR) elliptic curve, with endomorphism ring R, 98
eqRK) ideal class group of RK, 99
n the ideal class of a, 99
aA the product of the ideal a and lattice A, 99
n*EA action of ideal class n on elliptic curve E A , 99
E[a] group of a-torsion points, 102
hK class number of K, 107
F complex multiplication map Gal(R / K) --> eJ:.,(RK), 112
Up Frobenius element in Gal(LI K), 116
I( c) fractional ideal prime to c, 116
P(c) principal ideals congruent to 1 modulo c, 117
K, ray class field of K modulo c, 117
H,HK Hilbert class field of K, 118
Kv completion of K at v, 119
Rv ring of integers of K v , or Kv if v is archimedean, 119
A'K idele group of K, 119
Kp completion of Kat p, 119
Rp ring of integers of K p, 119
ord p normalized valuation on Kp and R p, 119
(s) ideal of the idele s, 119
u, an open subgroup of the idele group A'K, 119
Nf( norm map on idele groups, 119
K ab maximal abelian extension of K, 120
[·,K] the reciprocity map for K, 120
h Weber function E --> E/ Aut(E) == 11"1, 134
Fn(X) the polynomial IT<:>ESn (X - j 0 a), 144
Sm(T) the coefficients of the polynomial Fn(X), 144
Fn(Y,X) the modular polynomial Fn(j, X) = ITaESn (X - j 0 a), 146
Hn(X) the modular polynomial Fn(X, X), 146
M[pOO] p-primary component of an RK-module M, 157
t/JEIL the Grossencharacter of a eM elliptic curve ElL, 168
L'iJ(EIL,T) local L-series of E at p, 171
L(s, t/J) Heeke L-series attached to the Grossencharacter t/J, 173
A(EIL),s) modified L-series of ElL, 176
wElL sign of the functional equation of ElL, 176
1>;;' primitive integer matrices of determinant n, 181
S;;. primitive special integer matrices of determinant n, 181
<Pn(X) the modular polynomial of order n, 181
Z(E/JF'iJ' T) zeta function of the elliptic curve E/JF'iJ, 183
(£(s) zeta function of the field L, 183
(EI L, s) zeta function of the elliptic curve EI L, 183
Tp(E) p-adic Tate module, 186
K(8,2) subgroup of K* / K*2, 194
K(8,m) subgroup of K*IK*m, 195
Pic°(C) group of divisor classes of degree zero, 197
Jac(C) Jacobian variety of C, 197
C(g) symmetric product of a curve, = cg 18g , 197
G(C) group of sections to an elliptic surface, 202

G(Cjk) group of sections defined over k to an elliptic surface, 203
¢(V) the image of a rational map ¢: V --> W, 204
Dom(¢) domain of definition of a rational map, 204
7ft (f) order of the pole of f at t, 213
it canonical height on an elliptic surface, 217
(- , . ) canonical height pairing on an elliptic surface, 218
E(K,d) points in E(K) with height at most d, 223
Div(S) group of divisors on a surface, 231
(ls.p local ring of a surface S at a point P, 231
(ls,r local ring of a surface S at a curve r, 231
ordr order of vanishing along a curve r, 232
div homomorphism k(S)* --> Div(S), 232
linear equivalence of divisors, 232
Pic(S) Picard group of a surface S, 232
D l · D2 intersection pairing of divisors Dl and D 2 , 233
(Dl . D 2 )p local intersection index at P, 233
7f* homomorphism Div(C) --> Div(S) on a fibered surface, 237
<PD divisor on S with (D + <PD)' F = 0 for all fibral F, 240
Tp translation-by-P map on an elliptic surface, 245
(P) divisor associated to a section to an elliptic surface, 245
Dp divisor on elliptic surface G for a point P E E(K), 247
( . , .) canonical height pairing on an elliptic surface, 247
Eo(K) subgroup of an elliptic surface, 251
<PP,Q divisor with (P + Q) - (P) - (Q) + (0) '" <PP,Q, 252
[P,Q] canonical height pairing with values in Pic(C), 252
Div the group of divisors on a variety V, 255
hp absolute logarithmic height on projective space, 255
hD height on a variety associated to a divisor D, 256
h", height on a variety associated to a morphism, 258
Pt the image of a section a P to an elliptic surface, 265
at the specialization map on an elliptic surface, 271
Rs the ring of S-integers of a function field, 275
Mapk(C, E) the set of morphisms from C to Eo defined over k, 281
kE field of definition of sections to an elliptic surface, 282
fE/K the conductor divisor of an elliptic surface, 287
lG a the additive group variety, 291
IG m the multiplicative group variety, 291
GL n the general linear group, 291
SLn special linear group, 292
C(K) the group of K-rational points on a group variety, 292
CO identity component of a group variety, 292
AR affine space over the ring R, 298
lP"R projective space over the ring R, 298
X(T) the set of T-valued points of a scheme, 298
X(R) the set of R-valued points of an R-scheme, 298
Fx functor defined by an S-scheme, 298
X xsY the fiber product of two S-schemes, 299
Xs the fiber of an S-scheme X over a point s E S, 300
502 List of Notation
reduction of the scheme X modulo p, 300

the generic fiber of an R-scheme X, 300
affine space over the scheme S, 301
projective space over the scheme S, 301
identity section 170 : S --+ G of a group scheme, 306
inverse map i : G -> G of a group scheme, 306
{t group law {t : G x s G ...... G of a group scheme, 306
Ga the additive group scheme over Z, 307
Ga / s the additive group scheme over S, 307
Ga / R the additive group scheme over R, 307
Gm the multiplicative group scheme over Z, 308
Gm / s the multiplicative group scheme over S, 308
Gm / R the multiplicative group scheme over R, 308
translation map Ta : G --> G on a group scheme., 310
[m] multiplication-by-m map on a group scheme, 310
OF the local ring of a curve on an arithmetic surface, 311
ord F valuation attached to a curve on an arithmetic surface, 311
eO largest subscheme of e smooth over R, 316
w the subscheme of lP'~ defined by a Weierstrass equation, 321
WU largest subscheme of W smooth over R, 321
elR minimal proper regular model of an elliptic curve ElK, 325
clR largest subscheme of el R which is smooth over R, 325
Dom(¢) domain of a morphism ¢, 327
:X special fiber of a scheme over a discrete valuation ring, 330
Rh the Henselization of a discrete valuation ring, 331
R sh the strict Henselization of a discrete valuation ring, 331
Div(e) the divisor group of an arithmetic surface e, 339
div(f) the divisor of f on an arithmetic surface, 339
(r 1 . r 2 )x local intersection index on an arithmetic surface, 339
Divp(e) the group of fibral divisors on the arithmetic surface e, 340
D·F intersection index of divisors on an arithmetic surface, 341
Pa(F) the arithmetic genus of a curve on an arithmetic surface, 342
Ke a canonical divisor on the arithmetic surface e, 342
Divfib(e) group of fibral divisors on the arithmetic surface e, 344
Elk the reduction of a Weierstrass equation modulo p, 362
Ens(k) set of non-singular points of E(k), 362
Eo(K) set of points of E(K) with non-singular reduction, 362
E1(K) set of points of E(K) which reduce to the identity element, 362
m(EIK) number of components over k on special fiber e, 363
equals 7r- r ai, 366
Gi(LIK) the ith higher ramification group of L I K, 379
gi(LI K) the order of the ith higher ramification group of LI K, 380
E(EIK) the tame part of the conductor of ElK, 380
8(EIK) the wild part of the conductor of ElK, 380
f(EIK) the exponent of the conductor of ElK, 380
f(EIK) the conductor of an elliptic curve over a number field, 388
/Lr the scheme of Tth roots of unity, 397
Ctpr a finite group scheme in characteristic P, 397
Gd a one-dimensional group scheme, 398
iL/K index function on the Galois group of local fields, 405

Ar the Artin character, 405
Sw the Swan character, 405
O"k(n) the sum Edln d k , 409
Eq elliptic curve over C with j(Eq ) = j(q), 410
Sk(q) the series En>l O"k(n)qn, 410, 423
a4(q), a6(q) Weierstrass coefficients of E q , 410, 423
X(u, q), Y(u, q) series giving parametrization of E q , 410, 423
O( u, q) theta function, 412
"(EjJR.) ,,(-invariant for an elliptic curve defined over JR., 414
Eq Tate curve over p-adic field, 423
E formal group of an elliptic curve, 431
Gm formal multiplicative group, 431
"(EjK) ,,(-invariant for an elliptic curve defined over K, 439
!?q,r filtration of Tate curve E q , 450
h the canonical height on an elliptic curve, 454
I . Iv absolute value on the field K, 455
,\ the local Neron height function on E associated to v, 455
lvh the standard set of absolute values on K, 461
= [Kv : Qv], the local degree of v E MK, 461
local height associated to v, 461
the second Bernoulli polynomial T2 - T + ~, 468
the reduction of E modulo v, 469
points in E(K) with non-singular reduction, 469
division polynomial, 477
Index
abc-conjecture, 278 on a modular function, 76, 79, 91

Abel-Jacobi theorem, 198 Action, group scheme, 321, 326, 400
Abelian extension Action of SL2(Z) on 'Dn, 72
See also Class field theory Action of SL2(Z) on H, 9
Artin map, 117 fundamental domain, 10, 92
conductor, 117, 118, 123 stabilizer of a point, 11
Frobenius element "'p, 116, 121, 128, Action of SL2(Z) on H*, 14
129, 130, 132 Addition formula on an elliptic curve,
generated by j(E), 121 210, 216, 323, 471
generated by CM torsion points, 108, Additive group, 398, 399
128, 135, 148, 149, 166, 168 S-valued points, 397
generated by roots of unity, 108, 128, over a field, 291, 293, 398
153 over R, 307
inertia group, 120, 149 over S, 307
maximal, 120, 121 over Z, 307
of 1Ql, 129, 151 rational points of, 292
of Hilbert elMS field, 134 Tate module of, 382
ray class field, 117, 118, 120, 129, 135, Additive reduction, 171, 287, 388, 399,
180 403
unramified conductor, 381
See Hilbert class field torsion limited, 453
Abelian group Adjunction formula, 234, 249, 345
bilinear form on, 273 on arithmetic surface, 342, 351
is sum of p-primary components, 152, Affine group variety, 396
157 Affine scheme, set of R-valued points,
Abelian variety, 148, 196 298
automatically commutative, 196 Affine space
field of definition, 196 over a ring, 298
Jacobian, 197, 402 over a scheme, 301
Mordell-Weil theorem, 485 Algebraic equivalence, 486
of dimension one, 196, 279 of divisors, 254, 285
uniformization theorem over C, 196, Algebraic family
199 of elliptic curves, 201
with complex multiplication, 96 of points, 201
Abhyankar, S.S., 317 Algebraic group variety, 115
Absolute value Algebraic group
defines a topology, 455 See Group variety
on a field, 455 Algebraic integer, j-invariant is, 140,
on a number field, 461 147, 151,447
Action of a Heeke operator Algorithm
on a cusp form, 76, 78, 79 Laska's, 364
on Fourier coefficients, 76, 78 minimal Weierstrass equation, 364, 403
on a lattice function, 74 Tate's, 353, 361, 364, 387, 389, 435
on a modular form, 76 to compute local heights, 478, 479
Index 505
Almost every, 201 singular point on fiber, 314

Almost quadratic function, 454, 455 smooth,311
See also Local height smooth part, 316, 318, 321, 325, 332,
Ample divisor, 233, 257 335, 361, 362, 369, 378, 399, 400
height associated to, 257 special fiber, 311, 314, 350, 352, 361,
on a curve, 257, 264 371,399
on a surface, 258 of blow-up, 347, 371
Serre's theorem, 261 Wei! divisor on, 311
Analytic continuation of L-series at- Artin character, 405
tached to cusp form, 83, 94 Artin map, 117, 118, 120, 174
Approximation theorem, weak, 155, 158 for Q, 154
Arakelov intersection theory, 344 for Hilbert class field, 118
adjunction formula, 345 is surjective, 118
Riemann-Roch theorem, 345 kernel of, 117, 118, 179
Arakelov, S., 344 prime splits completely, 117, 179
Archimedean fiber, 345 trivial on norms, 179
Archimedean local height, 464 Artin reciprocity, 11 7
Arithmetic genus, 342, 351 Artin, M., 325, 327, 334
in fiat family, 343 Associative law, of a group scheme, 307
one, 355 Associativity, normal law, 334, 335
zero, 343, 352 Automorphism
Arithmetic surface, 311 of arithmetic surface, 318, 336
adjunction formula, 342, 351 of a curve, 294
Arakelov intersection theory, 344 of Neron model, 400
arithmetic genus of curve on, 342, 351 Automorphism group
blow-up, 343, 345, 371 of an elliptic curve, 183
example, 347, 402 of an elliptic surface, 245
of a curve, 348, 372
canonical divisor, 342, 350, 351 Bad fiber, 203
closed fiber, 313 Bad reduction, 185, 321
connected fiber, 342, 353 Baker, A., 141
dual graph of special fiber, 353 Basis for a lattice
exceptional divisor, 344 normalized, 7
extension of automorphisms, 318, 336 oriented, 6, 9, 71, 72, 73, 89
fiat, 345 Bernoulli numbers, 57
Galois action on special fiber, 353 Bernoulli polynomial
generic fiber, 311 Fourier expansion, 480
incidence matrix, 350, 402, 403, 486 second, 468, 473, 478, 480
intersection pairing, 341, 342, 353 Bertini's theorem, 233
intuitive definition, 311 Bezout's theorem, 232, 233
is curve over R, 311 Bilinear form, non-degenerate, 273
local intersection index, 339 Birational equivalence, 204, 205
local ring category of elliptic surfaces up to, 206
of a closed point, 315 category of varieties up to, 205
of a curve, 311, 313, 339 of elliptic surfaces, 206
of a point, 370 over k, 204
minimal proper regular model, 317, Birational isomorphism, 204
318, 344, 361 Birational map of fibered surface, 244
model of curve of genus two, 355 Birational morphism, 235
negative semi-definite intersection pair- arithmetic surface, 343
ing, 342, 353 blow-up, 343
non-singular, 311 Birch, B.J., 389
non-singular point on fiber, 315, 351, Birch-Swinnerton-Dyer fudge factor, 364
399 Blanksby, P., 487
projective linelP'h, 312 Blow-down, 236, 244
proper, 311, 316, 317, 351 Castelnuovo's criterion, 344
proper regular model, 317 Blow-up, 204, 235
regular, 311, 315, 316, 317, 351, 370, of an arithmetic surface, 343, 345
399 Castelnuovo's criterion, 344
regular in co dimension one, 311 coordinate chart, 345, 347, 371
resolution of singularities, 317 example, 347, 402
506 Index
Blow-up (continued) Class field theory 115-120

exceptional divisor, 344 See also Complex mUltiplication
gluing of charts, 346, 348, 371 Artin map, 117, 118, 154
height associated to, 285 idele group, 119
of multiplicative reduction curve, 403 idelic formulation 118-120
primer on, 345 local
of a scheme, 434 See Local class field theory
special fiber, 347, 348, 371 of <QI, 151, 153
of a Weierstrass equation, 371 ray class field, 117, 118, 120, 129, 135,
Bounded conductor, 404 180
Boundedness conjecture, 453 reciprocity map, 120, 152, 159, 165,
Brauer group, of a local field, 452 166, 168, 169, 174
Brumer, A., 385 Class function, 405
Class number, 107
is fini te, 86
Calculus, freshman, 50 on~ 107, 109, 138, 141
Canonical divisor, 27, 87 two, 142, 181
on an arithmetic surface, 342, 350, 351 Closed fiber over a DVR, 300
on an elliptic surface, 249 CM
Canonical height, 252, 454 See Complex multiplication
as intersection index, 247 Co dimension one
as limit, 454 generic point of subscheme, 328
as sum of local heights, 454, 461 regular in, 311
associated to a divisor, 285 Cohomology, C-adic, 172
comparison with naive height, 478 Commutator subring of endomorphism
explicit 0(1) estimate, 280 ring, 129
for divisor of degree zero, 285 Complete intersection, 328
is a quadratic form, 218, 454, 461 Complete ring is Henselian, 330
lattice structure associated to, 254 Complex Lie group, 48
lower bound, 287 Complex multiplication, 95
on an elliptic surface, 217, 247, 265, a-torsion points are free module, 102,
266, 269, 281, 286 109, 138
on Jacobian variety, 271 action of ideal class group, 99, 122
over function fields is rational, 219, additional references, 95
247 associated Grossencharacter, 165, 168,
parallelogram law, 218 174, 175, 184, 185
positive definite, 218 by
regulator, 273 K,96
standard properties, 267 R,96
Canonical height pairing, 218, 247, 252, Z[(l + v'=7)/2], Ill, 179
265, 269, 286 Z[i], 101, 107, 109, 138, 185
non-degeneracy, 272 Z[yC2], 110, 180
with values in Pic( £), 284 Zip], 102, 107, 177, 180
with values in Pic(C), 252, 265 non-maximal order, 160, 180
Castelnuovo's criterion, 236, 344, 352 computation of j, 142, 181
Category degree not square, 143
of S-schemes, 298 degree of an isogeny, 103, 124
of elliptic surfaces up to birational elliptic curve with CM by R K , 99
equivalence, 206 endomorphism ring, 96, 129
of varieties up to birational equiva- field of definition
lence,205 of endomorphisms, 106
Centralizer, 179 of isogenies, 105
Character for abelian varieties, 96, 148
Artin,405 Galois action
Swan, 405 on E, 112, 113, 122, 131
Characteristic polynomial of Frobenius, on j, 112, 122, 166
172 generate ring of integers, 95
Characteristic zero, 187, 189 good reduction, 184
Chinese remainder theorem, 103, 158 group of a-torsion points, 102, 135
Circle group, 129 isogenous curves, 178, 180
Circle method, 61 Iwasawa theory, 96
Index 507
j-invariant, 121, 122, 166 Connected component of Neron model,

in 02, 138 326, 361, 401
is algebraic, 104 Connected fibers, 240, 283, 342, 353
is integral, 140, 147, 151, 176,447 Contravariant functor. 309
j = 0,102,107,177,180 Correspondence, 68 '
J = 1728,101,107,138,185 Heeke, 68
j = -3375, III See also Heeke operator
j = 8000, 110, 180 hornothety, 68
lift of Frobenius map, 130, 132, 162 on set of lattices, 68, 90
L-series, 171, 175, 185 ring of, 70
main theorem, 157, 159, 166 Cox, D., 486
map F : Gal(K / K) ~ t:;L-(RK), 112, Cremona, J.E., 389
121 Criterion of Neron-Ogg-Shafarevich, :362
number of points modulo 'P, 175 (~urvc
See also Riemann surface
of degree two, 109, 141, 179
ample divisor on, 257, 264
ordinary reduction, 179
automorphism group, 294
potential good reduction, 140, 447
canonical divisor, 27, 87
reduction mod 'P, 121, 127, 129, 131
differential form on, 207
Tate module, 179, 186
exceptional, 236, :344
torsion generates abelian extensions, height on, 257, 264, 265
108, 135, 148, 149, 166, 168
holomorphic differential on, 198
transitive action homology of, 198
of Galois group, 122 hyperelliptic, 287, 486
of ideal class group, 100, 113 Jacobian variety of, 197, 199, 402
twist, 183 minimal proper regular model of, 317,
Weber function, 1:34, 135 :318, 344, 361
with j E JR, 142, 179 normal law on symmetric product, 402
Complex structure, 19 of genus two, 287
Components of fiber Picard group of, 192, 195, 199, 402
group of, 363, 364 projective, 27, 87
multiplicity of, :313 proper regular model of, 317
number on special fiber, 363, 364, 389 symmetric product, 229
non-reduced, 31:3 uniformizer for, 339
of a divisor, 231 Cusp, 287, 370
of multiplicity one, 102 of X(1), 14
reducible, 313 ofH*,14
singular, 313 stabilizer of, 15
Conductor Cusp form, 25, 91
bounded, 404 See also l\lodular function, modular
constant under isogeny, 404 form
divisor of an elliptic surface, 287 action of Hecke operator, 76, 78, 79, 92
exponent of, 363, :364, 380, 389 basis of eigenfunctions, 78, 80, 92
bound for, 385 dimension of space of, 31
computation of, 389, 407 discriminant is a, 26. 31
less than discriminant, 396 Fourier coefficients o'f, 79, 80, 81
maximum value, 387, 407 L-series attached to, 83, 85, 93
over 2-adic field, 407 simultaneous eigenfunction, 79, 80, 92
over 3-adic field, 407 size of Fourier coefficients, 81
independent of e, 381, 405 space of (M~k)' 31, 78
of a Grossencharacter, 176 spaces with dimension one, 34
of an abelian extension, 117, 118, 123 twisted L-series attached to, 94
of an elliptic curve, 379-388 twisted by x, 93
Cyclotomic extension, unramified, 383
of an order, 178
Cyclotomic field, 9.5, 128, 151
of ray class field, 1l7, 118, 129, 135
is abelian, 108, 128, 151
square-free, 388
Szpiro's conjecture, 287, 388
tame part, 380, :381, 405 TIn, 72, 91,143
wild part, 380, 381, 405 action of SL2(Z), 72
Conductor-discriminant formula for ellip- TI.~, 181
tic curves, 389 Decomposition group, 116, 331
508 Index
Dedekind, R., 67 half-plane of convergence, 83, 94

Dedekind domain, 113, 179 relations satisfied by coefficients, 80,
arithmetic surface over, 311 92
is regular scheme, 302 Dirichlet's theorem on primes in progres-
Dedekind 'T/ function, 65 sions, 118, 123
transformation formula for, 67, 90 Discrete subgroup
Dedekind reciprocity, 67, 90 of Qp, 422
Dedekind sum s(x, y), 66, 89 of Q;, 422
Degree map on divisors, 197 Discrete valuation ring, 328
Degree one prime, in ideal class, 118, arithmetic surface over, 311
123, 161 closed fiber over, 300
Degree generic fiber over, 300
non-square is CM, 143 Henselian, 330, 401
of a CM isogeny, 103, 124 Henselization of, 331
of a divisor, 232, 282 smooth scheme over, 304
of isogeny is preserved under reduc- special fiber over, 300
tion, 124 strict Henselization of, 331
Deligne, P., 61, 81 strictly Henselian, 330, 402
Dem'janenko, V.A., 265, 269, 271, 485 unramified extension, 399
Descent theorem, 230 Discriminant function, Fourier expansion,
Descent, faithfully flat, 338 59, 60, 467, 468
Determinant Discriminant
homomorphism of group varieties, 292 Fourier coefficients, 482
kernel of, 292 See also Ramanujan 'T function; Ja-
Deuring, M., 95, 175 cobi's formula
Diagonal is eigenfunction for Hecke operators,
height associated to, 285 78,79
is complete intersection, 328 minimal, 363, 364, 388, 389, 396
morphism, 300, 309, 327 for elliptic surface, 286
Different, 176, 386 over 2-adic field, 407
Differential form, 207 over 3-adic field, 406
associated to a modular function, 28, modular, 26, 31, 32, 62, 89, 135
31, 87, 91 of a cubic polynomial, 62
divisor of, 27 of a Weierstrass equation, 62, 183, 455,
holomorphic, 27 458, 470, 478
invariant, 43, 183 of isogenous curves, 453
of the first kind, 43 of Tate curve E q , 377, 423, 424, 473
of the second kind, 43 product expansion, 60, 62, 90, 409, 468
of the third kind, 43 vanishes only at 00, 32
on a curve, 198 weight 12 cusp form, 26, 31, 32, 34, 66
on a Riemann surface, 27 Div
on elliptic curve, 43 See Divisor group
on H, 26 Division polynomial ,pm, 477
on X(l), 23 Divisor class group
order of, 27 of 11"1, 271
regular, 27 of a curve,
sheaf of relative, 306 See Picard group
Dimension of a function field,
of a point on a scheme, 302 See Picard group
of a scheme, 302 Divisor group, 231
of Jacobian variety, 197 induced homomorphism, 255
regular scheme of dimension one, 302 of a fibered surface, 284
Dirichlet character, primitive, 93 of a set, 68
Dirichlet series of a variety, 255
See also L-series of an arithmetic surface, 339
analytic continuation, 83, 94 Divisor
attached to a cusp form, 83, 85 algebraic equivalence, 254, 285, 486
attached to a power series, 80 ample, 233, 257, 258
Euler product expansion, 80, 92, 172, associated to a morphism, 258
173 canonical height, 285
functional equation, 83, 94 components of, 231
Index 509
degree of, 232, 282 G 2 k, 25, 31, 32, 55

exceptional, 344 is a modular form, 25
fibral, 246, 247, 284, 340 is eigenfunction for Hecke operators,
group of, 231 78,93
height associated to, 256, 265 normalized, 58
horizontal, 340 q-expansion of, 55
integer part of, 87 special value of, 178
intersection pairing, 233, 238 value at 00, 25
is difference of very amples, 261, 284 Elkies, N., 487
linear equivalence, 232, 339 Elliptic curve
local equation, 232, 233 addition formula, 323
minimal discriminant, 286 automorphism group, 183
numerical equivalence, 486 bad reduction, 185
of a differential form, 27 classification of special fibers, 350, 352
of a section, 245 conductor, 363, 364, 388
of an elliptic function, 45 endomorphism of degree two, 109, 141
on a surface, 231 endomorphism ring R, 98
on an arithmetic surface, 311, 339 endomorphisms are integral, 448
positive, 285, 341 existence of Neron model, 325, 332,
principal, 232, 339 335,337
self-intersection, 234, 238, 243, 283, field of definition, 37, 38
342, 349, 351 field of moduli, 37, 38
Serre's theorem on ample, 261 formal group, 276
transversal intersection, 232, 282 Galois conjugate, 131
very ample, 257, 261, 284 good reduction, 184
Wei!, 339 group of a-torsion points, 102, 135
Domain Grossencharacter of, 165, 168, 174, 175
of a rational map, 204, 279 height over function field, 212, 213
of morphism to group scheme, 327, homogeneous space of, 199
333, 337 invariant differential, 43, 97, 134
Dominant rational map, 204, 222, 228 is abelian variety, 196
from smooth scheme to proper scheme, is group variety, 291
328 isogeny, 67, 182, 183
induces map on function fields, 205 j-invariant
of elliptic surfaces, 208 j = 0, 102, 107, 177, 180, 280
Dual graph of special fiber, 353 j = 1728, 101, 107, 138, 185, 280
Dual isogeny, 67, 125, 229
j = -3375, 111
of Neron model, 400
j = 8000, 110, 180
Duplication formula, 139, 427, 437
non-integral has End(E) = Z, 447
for height, 257
f-adic representation, 405, 445
for local height, 456, 471, 475
L-series, 171, 172, 175, 183, 185, 364
has degree four, 458
map to pI, 134
on an elliptic curve, 210, 214, 457, 471
minimal discriminant, 363, 364
Dynkin diagram, 353
minimal proper regular model, 325,
332, 335, 337, 350, 352, 361, 400
Effective bound for S-integer points, multiplication map, 310
277, 288 multiplicative reduction, 403
Eigenfunction negation map, 325
basis of Mg k , 80, 92 Neron model, 319
for Hecke operators, 77 normalized, 97, 131, 134
normalized, 78, 79, 80, 92, 93 of high rank, 272
simultaneous for all T 2k 'S, 78, 79, 80, ordinary reduction, 184
92,93 over IC
Eisenstein series, 25, 91 See Elliptic curves over IC
algebra generated by, 88 over function field
E 2k ,58 See Elliptic surface
Fourier coefficients of, 58 over locally compact field, 476
92 and 93, 26, 35, 58 over p-adic fields
G4 and G6 algebraically independent, See Elliptic curves over p-adic fields
88 over lR
G4 and G6 determine A, 35, 88 See Elliptic curves over lR
510 Index
Elliptic curve (continued) uniformization over ~, 416

potential good reduction, 140, 447 Weil-Chiitelet group, 421, 449
reduction modulo ~, 131 with CM, 142, 179
reduction type, 352, 353, 363, 364 Elliptic function, 39
representation on m-torsion, 108, 383 See also Weierstrass S" function,
supersingular reduction, 184 Weierstrass (Y function, Weierstrass
translation map, 310 ( function
twists of, 283, 414, 439 divisor of, 45
v-adic topology factorization as product of (Y's, 45
See v-adic topology, on an elliptic Laurent series of F(u; q), 49
curve Elliptic integral, 178
Weber function, 134, 135 Elliptic surface
with complex multiplication See also Fibered surface, surface
See Complex multiplication algebraic equivalence, 254
Elliptic curves over C, 408-413 as subset of 1l"2 x C, 200
analytic parametrization, 6, 34, 408, associated elliptic curve, 206
411, 464, 468 automorphism, 245
discriminant, 26, 409, 411, 464, 467 bad fiber, 203
See also Discriminant birational equivalence, 206
homology group, 37, 43 canonical divisor, 249
isomorphism class of, 36, 37 canonical height on, 217, 247, 252, 265,
j-invariant, 34, 36, 411 266, 269, 281, 284, 286
See also j-invariant classical theory, 187
lattice of, 408, 468, 477 conductor divisor, 287
local height, 464, 466, 468, 477 definition of, 202, 203, 205
periods, 37, 43 divisor Dp, 247
Uniformization Theorem, 6, 34, 35, 88, divisor of a section, 245
96, 196,411 dominant rational map, 208, 222, 228
Weierstrass equation, 34, 409, 464 fiber, 201, 202
Elliptic curves over non-archimedean fiber product, 211, 301
fields, 424 field of definition, 203
Elliptic curves over p-adic fields, 438-444 of section, 189, 282
See also Tate curve Eq generic fiber, 206
additive reduction, 451 good fiber, 203
applications, 445-448 group of sections, 202, 210, 298
l'-invariant, 439, 441, 451 defined over k, 203, 210
not equal one, 444 height on, 212, 213, 265
local height, 469, 472, 480 infinitely many points of bounded
multiplicative reduction, 377,478 height, 220, 222
non-split multiplicative reduction, 378, is a C-scheme, 298
451 j-invariant
quadratic character, 440, 444 is algebraic function, 280
split multiplicative reduction, 377, 441, constant, 286
451, 452, 472, 480 Kummer sequence for, 193
Tate module of, 452 lattice structure, 254
torsion points, 383, 445 lower bound for height, 287
uniformization, 141, 423, 441, 444, 473, map on Pic is injective, 253
480 minimal discriminant divisor, 286
Weil-Chiitelet group, 452 Mordell-Weil theorem, 230, 276, 279
Elliptic curves over ~, 413--422 multiple fibers, 203
C isomorphism classes, 414 multiplication-by-n map, 266
l'-invariant, 414, 419, 450 Neron-Severi group, 254
not equal one, 450 non-singular fiber, 286
homogeneous spaces, 449 non-split, 218, 247
isomorphic to real Lie group, 420 with zero discriminant, 286
Kummer sequence, 449 one-cocycle on sections, 284
number of components, 419 one-parameter family, 188
~ isomorphism classes, 414, 416 ordt(x) is bounded, 275
sign of the discriminant, 420, 421, 449 over non-perfect field, 187
torsion subgroup, 449 points not a group, 211
twists, 414, 450 rational map between, 206
Index 511
S-integral points on, 275, 277, 288 Fermigier, T., 272

section, 202 Fiber
of order eleven, 190, 279 archimedean, 345
of order five, 278 closed, 300
of order seven, 278 generic, 300
singular fiber, 203 multiplicity of components, 313
specialization map, 271 of a scheme over a point, 300
is injective, 271, 281 of an arithmetic surface, 313
of height, 265, 266, 269, 281, 286 of an elliptic surface, 201
split, 220, 222, 228, 231, 271, 280, 281 singular point, 314
over 11"1,286 smooth,397
subgroup Eo(K), 251, 284, 288 special, 300
torsion subgroup, 288 Fiber product, 299
translation map, 245, 248, 251 diagonal, 300, 309, 327
weak Mordell-Weil theorem, 191-195, graph in, 300
230 group operation on, 211
zero section, 202 of elliptic surface, 211, 301
Endomorphism universal property, 299
field of definition of, 106 Fibered surface, 236
of degree two, 109, 141, 179 See also Surface, elliptic surface
of an elliptic curve, 448 connected fibers, 240, 283, 342
Endomorphism ring fiber of, 236
commutator subring, 129 genus at least one, 244
elliptic curve with given, 121 fibral curve, 237
normalized isomorphism, 97, 129, 131, fibral divisor, 237
134 fibral homomorphism, 237
quaternion algebra, 130 fibral intersections, 238
reduction modulo ~, 129 horizontal curve, 237
E q, See Tate curve Eq horizontal divisor, 237
Etale morphism, 306 map on Picard group, 284
Etale topology, 332 is injective, 253
Eta function (1/) minimal, 244
See Quasi-period homomorphism, multiplicity of fibers, 237
Dedekind 1/-function negative semi-definite intersection pair-
Euclidean structure on Mordell-Wei! ing, 238
group, 247 relatively minimal, 243
Euler product, 80, 92, 172, 173 Fibral curve, 237
Excellent scheme, 311 Fibral divisor, 237, 246, 247, 284
Exceptional curve, 236, 244 group of, 340, 344
Exceptional divisor, 344 intersects fiber trivially, 238
Castelnuovo's criterion, 344, 352 on an arithmetic surface, 340
on an arithmetic surface, 344 self-intersection, 238, 342, 351
Exponent of the conductor, 380, 389 Field
bound for, 385 locally compact, 476
computation of, 389, 407 non-perfect, 187
independent of £, 381, 405 of characteristic zero, 187, 189
less than discriminant, 396 perfect, 330, 332
Exponential function, special values, 151 totally imaginary, 116
Exponential map on multiplicative Field of definition
group, 151 for E(qT)), 282
Extended upper half plane H*, 14 of an abelian variety, 196
action of r(1), 14, 85 of an elliptic curve, 37, 38
cusps of, 14, 85 of an elliptic surface, 203
is a Hausdorff space, 16 of Jacobian variety, 199
topology on, 16, 85 of section to elliptic surface, 189, 282
Field of moduli of an elliptic curve, 37,
38
Faithfully fiat Finite group scheme, 397
descent, 338 Finite order elements in Picard group,
strict Henselization is, 338 192, 195, 199
Faltings, G., 271 Finite type scheme, 311
512 Index
First kind differential form, 43 sign of, 176

Fischer-Griess monster group, 61 for £I(u, q), 412, 429, 474
Flat family, arithmetic genus in, 343 for X(u, q), 425
Flat morphism, 304, 397 for Y(u,q), 425
Flat scheme, 311, 345 Functor
Formal Dirichlet series, 80 contravariant, 299, 309
Formal group defined by a group scheme, 309
of an elliptic curve, 276, 381 defined by an S-scheme, 298, 306
of Tate curve E q , 431 Jacobian, 198
Formal logarithm, 150, 182 Functoriality of height function, 256
Formal multiplicative group, 149, 431 Fundamental domain (3'") for action of
Fourier analysis, 173 SL2(2:) on H, 10, 14, 82, 92
Fourier coefficients Fundamental domain, for C/ A, 41, 71
See also q-expansion
action of Hecke operator, 76, 78, 79
G 2 k> See Eisenstein series
integer, 145
"1- invariant
of a simultaneous eigenfunction, 79, 80
in characteristic two and three, 451
relations satisfied by, 79, 80
of an elliptic curve, 439
size of, 81, 82, 92
defined over JR, 414, 419
Fourier expansion f(l), See Modular group f(l)
See q-expansion fo(N),87
Fractional ideal
fl(N),87
is a lattice, 99 f(N),86
of quadratic imaginary field, 85 res), See Gamma function
Free module of rank one, 102, 138, 186 Galois action
Frey, G., 453 on series, 424, 428, 451
Fricke, M., 95 on special fiber, 353
Frobenius map Galois cohomology
action on Tate module, 172 of a cyclic group, 421
automorphism, 116, 128, 129, 130, 132, of a discrete group, 428
161 Galois conjugate of a eM elliptic curve,
characteristic polynomial, 172 131
lift to characteristic zero, 130, 132, 162 Galois group
of varieties, 132 decomposition group, 331
Fudge factor, 364 higher ramification group, 379
Function field inertia group, 120, 149, 331
birational equivalence, 205 of Hilbert class field, 118
canonical height is rational, 219, 247 Galois theory, 37
elliptic curve over Gamma function, 83, 176
See Elliptic surface incomplete, 93
finite subgroup of, 195 Gauss, K.F., 141
height on, 212 Gauss sum, 93
elliptic curve over, 212, 213 Gaussian integers, 178
ideal class group, 192 Gel'fond-Schneider theorem, 108, 142
infinitely many elements of bounded General linear group, 291
height, 213 centralizer of a subgroup, 179
Kummer theory, 191 determinant, 292
local heights, 212 elements of finite order, 403, 404
map induced by dominant rational formal logarithm, 150, 182
map, 205 rational points of, 292
Mordell-Weil theorem, 230, 279 Generic fiber, 300
S-integers, 275 empty, 401
Szpiro's conjecture, 287, 388 of an arithmetic surface, 311
unit group of, 191 of an elliptic surface, 206
valuation on, 194 over a DVR, 300
Functional equation Generic point, 304, 328
of Dirichlet series attached to cusp of a scheme, 300
form, 94 of Spec(R), 300
of L-series, 176, 184 Genus
of L-series attached to cusp form, 83, arithmetic, 342, 351
94 in flat family, 343
Index 513
of a smooth curve, 27, 87 translation map, 310, 336, 337, 400

of the modular curve X(l), 20, 21 Group variety, 291
two, special fiber of curve of, 355 abelian, 196
zero, 343, 352 additive group, 291, 293, 398
GL n , See General linear group algebraic, 115
Global sections, 397 defined over K, 291
Good fiber, 203 determinant homomorphism, 292
Good reduction, 184, 287, 321, 329, 383, elliptic curve, 291, 293, 398
388 general linear group, 291
conductor, 381 homomorphism, 292
Grossencharacter is unramified, 168 identity component, 115, 292
potential, 140, 148, 151, 176, 383 irreducible components, 292
torsion for, 453 is a group scheme, 307
Graded algebra, 27, 88 is non-singular, 292
Graph of a morphism, 300 Jacobian, 402
Green's function, 477 linear group, 292, 396
Group multiplicative group, 291, 293, 398
algebraic of dimension one, 293, 398
See Group variety orthogonal group, 396
pro-p, 150, 182 rational points of, 292
sporadic, 61 special fiber of Neron model, 361
topological, 85 special linear group, 292
Group chunk, 198, 334 special orthogonal group, 396
Group cohomology, 193 symplectic group, 396
of a cyclic group, 421, 449 Grossencharacter, 156, 165, 168, 173, 184
Group law conductor of, 176
induced by normal law, 334, 337 for j = 0 curve, 177
is normal law, 334 L-series, 171, 173, 175, 185
Group of components of twisted curve, 183
of a group variety, 292 on Q, 156
of Neron model, 350, 362, 363, 364, reduction modulo 'l!, 174
379, 402, 486 unramified, 168, 173, 174
of special fiber, 363, 364
Group of sections to an elliptic surface, Half-plane of convergence
202, 203, 210, 298 of L-series attached to cusp form, 83
Group scheme, 306, 398 of L-series attached to elliptic curve,
action, 321, 326, 400 183
additive group, 307, 397, 398, 399 Hardy, G.H., 61
associated functor, 309 Hasse, H., 95
associated to normal law, 334, 337 Hausdorff space, 16, 18, 19, 85
associativity, 307 Hecke, E., 81, 173, 484
diagonal morphism, 309 Hecke algebra, 70
extension of group law, 332, 335 is commutative, 71
finite, 397 of reI), 70
identity component of special fiber, Hecke L-series, 173
326,361 Hecke operator, 67-74
identity section, 306, 326 See also Hecke algebra
in characteristic p, 397 action on a cusp form, 76, 78, 79, 92
inverse map, 306 action on a lattice function, 74
local ring along identity, 327 action on a modular form, 76
morphisms to, 327, 333, 337 action on a modular function, 76, 79,
multiplication map, 310, 398 91
multiplicative group, 308, 397, 398, action on Fourier coefficients, 76, 78,
399 79
Neron model, 319 as shifting operator, 76
of roots of unity, 397, 398 commutativity of, 71
proper, 327 definition of, 68
set of T-valued points, 309, 397 discriminant is eigenfunction, 78
smooth part of Weierstrass equation, eigenfunction for, 77
321, 362, 378, 399, 400 explicit formula for, 71, 73, 74, 90
smooth Weierstrass equation, 329 recursive formula for, 90
514 Index
Hecke operator (continued) Hermitian inner product, 92

relations satisfied by, 68, 79, 90 Higher ramification group, 379
self-adjoint, 92 index function for, 405
simultaneous eigenfunction for all, 78, inertia group, 380
79, 80, 92, 93 is normal, 380
T(p), 74, 77 over 2-adic field, 404
Heegner, 141 over 3-adic field, 404
Height Hilbert class field, 95, 118, 130, 132, 142,
additivity, 256, 261 181
and linear equivalence, 256 abelian extension of, 134
associated to a divisor, 256, 265 Artin map for, 118
associated to a morphism, 258 Galois group of, 118
associated to a positive divisor, 285 generated by j(E), 121, 122, 166
associated to an ample divisor, 257 of !QJ(v'-15), 180
associated to diagonal of curves, 285 of!QJ( v'=i3), 180
associated to exceptional curve, 285 of quadratic imaginary field, 121, 122
canonical, 217, 247, 265, 266, 269, 281, Hilbert irreducibility theorem, 272
286 Hilbert theorem 90, 193, 421
canonical Hindry, M., 277, 278, 486, 487
See Canonical height Holomorphic differential form, 27, 207
duplication formula, 213, 257, 280 on a curve, 198
explicit 0(1) estimate, 280 Homogeneous space of an elliptic curve,
finitely many points of bounded, 257, 199
265 Homology group, 37, 43
for algebraically equivalent divisors, of a curve, 198
285 Homomorphism
functoriality, 256 of groups, 427
geometric transformation properties, of group varieties, 292
213 Homothetic lattices, 6, 9, 14, 37, 101,
infinitely many elements of bounded, 103, 161
213, 220 Homothety operator, 68, 90
Neron-Tate, 217, 247, 265, 266, 269, relations satisfied by, 68
281, 286 Horizontal curve, 237
See also Canonical height intersection with, 283
normalization, 256 Horizontal divisor, 237
on a function field, 212 on an arithmetic surface, 340
on a variety, 256 Hurwitz, A. 101, 294, 485
on an elliptic surface, 212, 213, 265 Hyperelliptic curve
on curves, 257, 264, 265 integer points on, 277
on projective space, 255 Jacobian variety, 199, 287, 486
parallelogram law, 213 of genus two, 287
quasi-parallelogram law, 280 over function field, 277
regulator, 273 Picard group, 287
specialization of, 265, 266, 269, 281,
286
sum of local, 212 Ideal
Height Machine, 256 of an idele, 119, 152, 159
verification of, 262 principal, 132
standard properties, 267, 268 Ideal class group, 99, 180
Hensel's lemma, 330, 331, 401, 443 See also Hilbert class field
multi-variable version, 401 acts on e,c£1..RK), 99
Henselian ring, 330, 401 acts on e,c£1..RK) transitively, 100,
complete ring is, 330 113
strictly, 330, 402 algorithm to compute, 85
surjectivity of reduction, 330, 333, 337, class contains degree one primes, 118,
401 123
Henselization is finite, 86
fraction field, 332 is Picard group, 192
of a DVR, 331 of a function field, 192
strict, 337 of quadratic imaginary field, 85
universal mapping property, 332, 401 Idele
Index 515
ideal of, 119, 152, 159 Irreducible topological space, 280

multiplication by, 159, 170 Isogeny, 67, 182
Idele group, 119 between CM curves, 178, 180
characterization of ray class fields, 120, between Tate curves, 453
162 comparison of discriminants, 453
class field theory using, 118-120 conductors are equal, 404
contains K*, 119 degree preserved under reduction, 124
contains K~, 119 dual, 67, 125, 229, 400
norm map, 119 equations for, 183
topology on, 119 field of definition of, 105
Identity component is a homomorphism, 396
of a group variety, 115, 292 lift of Frobenius map, 130, 132, 162
of Neron model is Weierstrass equa- of Neron model, 400
tion, 362, 378 purely inseparable, 127
Identity element Neron model, 326 reduction mod <:p, 124, 127, 129
Identity section of a group scheme, 306 Isomorphism birational, 204
Image of a rational map, 204, 279 Iwasawa module, 186
Incidence matrix, 240, 241, 243, 283, 350, Iwasawa theory, 96
402, 403, 486
Incomplete gamma function, 93
j-invariant
Index function for higher ramification
classifies elliptic curves, 36
groups, 405
CM in IQI, 107, 138
Inert prime, 184 computation of, 142, 181
Inertia group, 120, 149, 169, 331, 380, divisibility properties of Fourier coeffi-
445 cients, 60
absolute, 380 elliptic surface with constant, 286
Inner product Galois conjugate of, 112
Hermitian, 92 generates Hilbert class field, 121, 122,
Petersson, 92 166
positive definite, 92 growth properties of Fourier coeffi-
Inseparable isogeny, 127 cients, 61
Integer points in Z, 141
effective methods, 277, 288 integral, 383, 405
on an elliptic surface, 275 integrality properties, 140, 151, 447
Integers of a function field, 275 j = 0, 102, 107, 180, 280
Integral scheme, 311 j = 1728, 101, 107, 138, 280
Integral Grossencharacter, 185
elliptic, 178 number of points modulo p, 185
not path independent, 198 j = -3375, 111
Intersection j = 8000, 110, 180
self, 234, 238, 243, 283, 342, 349, 351 map X(l) -+ 1P'1(C), 23, 34, 36, 88
transversal, 232, 282 modular, 34, 88, 121, 143
Intersection index modular function of weight zero, 34,
Arakelov, 344 112, 144
computation of, 283 non-integral, 382, 405
example, 349 of an elliptic surface, 280, 288
linear equivalence, 341 of CM curve is algebraic, 104
local, 233, 283 of CM curve is integral, 140, 147, 151,
on IP'k, 340, 341 447
on an arithmetic surface, 339, 341 q-series, 59, 60, 121, 141, 142, 144,
sum of local indices, 234 145, 146, 410
symmetric, 341 rationality properties, 37, 104
Intersection pairing, 233 singular, 104
Arakelov,344 supersingular, 104
incidence matrix, 240, 241, 243, 283 Tate curve E q , 423, 425, 438, 473
on an arithmetic surface, 341, 342, 353 transcendental values, 108
on fibered surface, 238 valuation of, 362, 379
symmetric, 341 Jacobi, C., 62
Invariant differential, 43, 97, 134, 183 Jacobi sum, 177
Inverse on a group scheme, 306 Jacobi's formula, 60, 62, 65, 90, 409, 468
Irreducibility theorem of Hilbert, 272 Jacobian matrix, 330, 401
516 Index
Jacobian variety, 197, 199, 402 of an elliptic curve, 171, 172, 175, 183,
canonical height on, 271 185
construction of, 197 relations satisfied by coefficients, 80,
curve of genus two, 287 92
dimension of, 197 residue of, 93
field of definition, 199 special values of, 93
is a functor, 198 £-adic cohomology, 172
Neron model, 350 Lang, S., 231, 275, 285, 485
of a homogeneous space, 199 Lang's conjecture, height lower bound,
of a hyperelliptic curve, 199 287
of genus one curve, 197 Laska's algorithm, 364
of genus zero curve, 197 Laska, M., 364
of hyperelliptic curve, 486 Lattice
Jordan normal form, 179 associated to a Weierstrass equation,
Jugendtraum, 96 35,88,89
Euclidean, 254
Heeke correspondence on, 68, 90
Kernel of Artin map, ll7, ll8, 179 homothetic, 6, 9, 14, 37, 101, 103, 161
Kodaira, K., 287, 352, 486 homothety correspondence on, 68, 90
Kodaira-Neron classification of fibers, normalized, 47
350, 352 normalized basis, 7
Kodaira notation for special fiber, 353 of an abelian variety, 196, 198
Kodaira reduction type, 352, 353, 363, of an elliptic curve, 408
364 oriented basis, 6, 9, 71, 72, 73, 89
Kouya, T., 272 set of all (,C), 6, 36, 68, 74
Kramer, K., 385 sublattice, 67
Kraus, A., 486 sublattice of index n, 71, 72, 73
Kronecker, L., 95 Lattice function, 74
Kronecker congruence relation, 182 associated to a modular function, 74,
Kronecker Jugendtraum, 96 79
Kronecker limit formula, 484 Laurent series
Kronecker-Weber theorem, 95, 129 for F(u; q), 49, 51
Krull dimension, 301, 302 for 1", 39, 51
Kummer extension, 383, 446 for (, 39
Kummer sequence p-adic, 429
for an elliptic surface, 193 Lefschetz principle, 199, 463
for fields, 193 Legendre relation, 41, 465, 469
Kummer theory for function fields, 191 Lehmer, D.H., 61
Kuwata, M., 189,485 Lehner, J., 60
Kuyk, W., 389 L'Hopital's rule, 45
Lichtenbaum, S., 317, 338, 344
Lie group
L-series complex, 48
See also Dirichlet series real, 420
analytic continuation, 83, 93, 94, 173, Limit formula of Kronecker, 484
176 Linear equivalence
attached to a cusp form, 83, 85, 93 and height functions, 256
attached to a modular form, 93 intersection index, 341
attached to a power series, 80 of divisors on a surface, 232
attached to a twisted cusp form, 94 Linear fractional transformation, 294
elliptic curve, 364 Linear group, 292, 396
Euler product expansion, 80, 92, 172, Linearly equivalence, 339
173 Lipman, J., 317
for j = 0 curve, 178 Littlewood, J., 61
functional equation, 83, 93, 94, 173, Liu, Q., 390
176, 184 Local L-series of E, 171, 184, 185
half-plane of convergence, 83, 172, 183 Local class field theory, 140, 148, 149,
Heeke, 171, 173, 175, 185 452
integral representation, 84, 85, 94 Local degree n v , 461
leading coefficient, 364 Local equation for a divisor, 232, 233
local of B at p, 171, 184, 185 Local field
Index 517
Artin character, 405 l\lanin, Yu., 247, 265, 269, 271, 275, 485
different, 386 Mason. R.e., 277, 278
higher ramification group, 379 Masser, D .. 278
index function, 405 Matrix
inertia group, 380 incidence, 350, 402, 40:3, 486
maximal unramified extension. 381 Jacobian, 330, 401
Swan character, 405 Maximal abelian extension
Local height function, 429 of exponent m unramified outside S,
algorithm to compute, 478, 479 191
associated to v, 461 of quadratic imaginary field, 135
comparison with naive height, 478 Maximal unramified extension
differential equation for, 477 abelian
duplication formula, 456, 475 See Hilbert class field
existence of, 455-461, 476 of a local field, 381
archimedean absolute values, 467 Merumorphic function
non-archimedean absolute values, j is a, 23, 35
470.475 on X(I), 23, 35
integral over E(C), 477 on pI, 35
is a Green's function, 477 on H, 24
is continuous, 456 Mestre, J.-F., 272, 486
is invariant under field extension, 456, Minimal discriminant, 363, 361, 388, 389
476 greater than conductor, 396
is well-defined, 456, 470 of an elliptic surface, 286
is zero for almost all v. ,161 of isogenous curves, 4-53
lower bound for. 480 over 2-adic field, 407
multiplication formula. 477 over 3-adic field, 406
on Eo(K), 469, 478 Szpiro's conjecture, 287, 388
on Tate curve E q . 473, 475 valuation one, :369, 399
over IC, 464, 466, 477, 479, 480 valuation two, :399
over non-archimedean fields, 469-476 Minimal fibered surface, 244
over p-adic fields. 469-476, 478, 479, birational map is morphism, 244
480 Minimal proper regular model
over nt 479 components of special fiber, 389
quasi-parallelogram law, 467, 476 connected component, 326, 361, 401
triplication formula, 463 identity component of special fiber,
Local intersection index, 233
on pk, 340, 341
326, 361
number of components of special fiber,
on an arithmetic surface. 3:39
363, :~64
Local Neron height function
of a curve, 317, :318, 344
See Local height function
of an elliptic curve, 325, 332, :335, 3:37,
Local parameter, 19. 20, 29, 30, 85
350, 361, 400
Local ring
reduction type, 352, 353, 363, 364
is discrete valuation ring, 232
special fiber, 350, 352, 403
of a curve on an arithmetic surface,
Minimal surface, relatively, 235
:311, 313, 3:39
Minimal Weierstrass equation, 321. 403,
of a surface at a curve. 231
478, 480
of a surface at a point, 231
algorithm. 364, 403
of arithmetic surface, 315
for p ?: :3, 406
of group scheme along identity, 327
for p ?: 5, 405
regular, 302, 370, 397
Modular j-invariant
regular of dimension two, 315
Localization, 179 See j-invariant
Modular curve X(I), 14
Locally compact field, 476
Lockhart, P., 385 affine part Y(l), 14
Logarithm complex structure on, 16, 20, 26, 28,
formal, 150. 182 35
principal branch, 44, 54, 56, 66 cusp, 14
differential forms on, 23, 28, 87
has genus zero, 20, 21
Magic, 48 is a Hausdorff space, 18, 19
Main theorem of complex multiplication, is a Riemann surface, 20, 32
157, 159, 166 is compact, 19
518 Index
lvlodular curve X (1) (continued) for elliptic surfaces, 276

is connected, 21 for function fields, 230, 279
j-function on, 23, 34, :36 for split elliptic surfaces, 231, 281
local parameter, 20, 21, 29, :{O, 85 ineffective, 277
meromorphic functions on, 23, 35 over finitely generated fields, 279
open cover of, 20 relative, 231, 281
order of a differential form, 28 weak. 191-195, 230
projection is open map, 19 Morphism
topology on, 16, 18, 19 birational, 235
I\Iodular curve XJ(ll), 190, 279 diagonal, 300, 309, 327
Modular discriminant divisor associated to, 258
See Discriminant etale, 306
Modular form, 2,5, 91See also Modular fiat, :301, :397
function, cusp form graph of, 300
action of Hecke operator, 76 of S-schemes, 297
dimension of space of, :n, 87 proper, :{05
Eisenstein series is a, 25, 31, 32 separated, 305
Fourier coefficients of, 82, 92 smooth, 304, 305, 306, 320
L-series attached to. 93 universally closed. 305
size of Fourier coefficients, 82, 92 Multiple fibers, 203
space of (M2h)' 31, 87 Multiplication map
space of is generated by 04, 0 6 , 88 continuous for v-adic topology, 460
spaces with dimension one, :n on elliptic curve, 310
Modular function. 24, 91 on group scheme. 310, 398
See also j-invariant; modular form: Multiplication by an idele, 159, 170
cusp form Multiplication-by-n map on an elliptic
j is a, 34 surface, 266
action of Heeke operator, 76, 79, 91 Multiplication-by-~: map on Qj£:, 152.
associated differential form, 28. 31, 87, 1.53
91 Multiplicative group, 128, 151, 398, 399
associated lattice function. 74 exponential map. 151
eigenfunction for Heeke operator, 77 formal, 149, 4:n
formula for order, 30 over a field, 291, 293, 398
Fourier series, 24, 76, 145 over H, :~08
holomorphic at CXl, 24 over S, :308
meromorphic at CXl, 24 over £:, 308
of weight zero, 34, :35, 182 rational points of, 292
order at CXl, 24 S-valued points, 397
value at oc, 24, 32 Tate> module of. :382
weight of, 24 torsion subgroup, 151
why even weight, 24 Multiplicative reduction, 287, 369, 3t;8,
Modular group [(1), 6-14 399
See also Action of SL2 (:2:) on H conductor, 381
generated by Sand T, 14, 24, 62, 85 Neron model, 403
is a free product, 14, 85 non-split, 171, 378
Modular polynomial split, 171. 362, 377, 379
examples, 148 torsion for, 453
F H , 144, U6. 181, 182 Multiplicity of fibral components, 313
<Pn, 181, 182
H n , 144, 146
size of coefficients, 181 Nagao, K., 272
Monster group, 61 Nakai-Moishezon criterion, 25R, 261, 283,
Montgomery, H., 487 284
Mordell, L.J., 61, 79 Nakayama's lemma, 103
Mordell conjecture. 271 Namikawa, Y.. :~55, 486
l\Iordcll-\'Veil group Natural map, 197, 198
Euclidean structure, 247 Negatioll map, on an elliptic curve, 325
lattice structure, 25,1 Neron, A., 231, 265, 272, 352, 4,,4, 455,
Mordell-\'Veil theorem 485
hound for rank. 279 Ncron height fUllction
for abelian varieties, ,lS5 See Local height function
Index 519
Neron-Kodaira classification of fibers, compatibility with reciprocity map,

350, 352 120
Neron mapping property, 319, 329, 333, of fields, 444
338 on ideles, 119
Neron model, 319, 361 Normal law, 334, 335
addition formula on Weierstrass equa- associativity, 334, 335
tion, 323 induces group law, 334, 337
automorphism of, 400 on symmetric product of a curve, 402
connected component, 326, :361, 401 Normal scheme, 311
dual isogeny, 400 principal divisor on, 328
existence, 325, 332, 33.5, 337 Normalized basis, 7
extension of group law, :3:32, 3:35 Normalized eigenfunction, 78, 79, 80, 92,
generic fiber, 319 93
group of components, 350. 362, :363, Sere al.so Eigenfunction
364, 379, 402, 486 Normalized elliptic curve (E, [.]), 97,
group of sections, 319 131, 134
group scheme action, 326, 400 Normalized lattice, 47
identity component is \"Ieierstrass Nullstellensatz, 259
equation, 362, :378 Number field
identity component of special fiber, abelian extension, 128
326, 361 See also Class field theory
identity section, 326 Artin map, 117, 118, 154
isogeny, 400 decomposition group, 116
multiplicative reduction, 403 ideal class group, 99, 180
negation map on Weierstrass equation, idele group, 119
325 maximal abelian extension, 120, 121
Neron mapping property, 319, 320 of class number one, 138. 141
not proper, :n 9 Q(i), 138
of Jacobian variety, 350, 486 unit group of, 191
of Tate curve E q , 435 :Lv-extension, 186
over a strictly Henselian ring, 402 Numerical equivalence, 486
over strictly Henselian ring, 332, 335
special fiber, 350, 361, 400, 435
Ogg, A., 355, 389, 487
subgroup scheme, 401
Ogg's formula, 363, 364, 387, :389, 396,
translation map, 400
407
uniqueness, 319
One-cocycle on elliptic surface, 284
unramified base extension, 320
One-parameter family of elliptic curves,
Weierstrass equation, 321, 329, 362,
188, 200
369, 378, :399, 400
Operator
Neron notation for special fiber, 353
Hecke
Neron-Ogg-Shafarevich criterion, 140,
See Hecke operator
151, 148, 170, 362, 381, 383
homothety, 68, 90
Neron-Severi group Order
of an elliptic surface, 254
at 00 of a modular function, 24
rank, 254
cond uctor of, 178
Nice scheme, 311, 399 non-maximal, 160, 178, 180
Node, 287, :370 of a differential form, 27
Non-linearity, textual, 217 Ordinary reduction, 179, 184
Non-perfect field, 187 Oriented basis, 6, 9, 71, 72, 73, 89
Non-singular arithmetic surface, 311 Orthogonal group, 396
Non-singular point .
speciaL 396
on fiber of arithmetic surface, 315, 351, Osterle, J., 278
399
on reduction, 362, 379
on SChenl€ p-adic analysis, 424, 429
See Regular point p-adic field, 423
Non-singular scheme squares in, 442
See Regular scheme p-adic period, 364
Non-split multiplicative reduction, 378, p-adic Tate module, 186
399 p-primary component. 157
Norm map p, See \Veierstrass p function
520 Index
Parallelogram law, 218 arithmetic genus, 343, 352

for height, 213 arithmetic surface, 312
Path independent integral, 198 proper over R, 312
Perfect field, 330, 332 smooth over R, 312
Period map, 43 Projective plane, Picard group of, 232,
Period, p-adic, 364 282
Periodic second Bernoulli polynomial, Projective scheme, is proper, 303, 322
480 Projective space
Periods, 37 height on, 255
Petersson, H., 61, 80 over a ring, 298
Petersson inner product, 92 over a scheme, 301, 303, 322
Hecke operators are self-adjoint, 92 Picard group of, 282
Picard group Proper arithmetic surface, 311, 316, 317,
See also Jacobian variety 351
divisor class of degree zero, 197 Proper group scheme, 327
elements of finite order, 192, 195, 199 Proper morphism, 297, 305
induced homomorphism, 255 intuitive definition, 301
is ideal class group, 192 valuative criterion, 303, 317, 329, 333
of 1P'2, 232, 282 Proper regular model
of IP'n, 282 minimal, 325, 332, 335, 337, 389, 400
of a curve, 402 of a curve, 317
of a fibered surface, 284 Proper scheme, 312, 321, 329, 399
of a surface, 232 group, 327
of curve of genus two, 287 map from smooth scheme, 328
of hyperelliptic curve, 287 projective scheme is, 303, 322
Points Proper S-scheme, 301, 305
of a group scheme, 309, 397
of a scheme, 298, 309, 397 q-expansion
Polar divisor on normal scheme, 328 See also Fourier coefficients
Positive divisor, 341 of Bernoulli polynomial, 480
associated height, 285 of discriminant ~, 59, 409, 467, 468
Potential good reduction, 140, 148, 151, of 92 and 93, 409
176, 383, 447 of G 2b 55
Power divisor function O'k of j-invariant, 59, 144, 145, 146, 410
See O'k of a modular function, 24, 55, 182
Prime ideal of p and pi, 47, 50, 409
in arithmetic progression, 118, 123 of Weierstrass 0' function, 53, 467
in ideal class, 118, 123 of Weierstrass ( function, 52
inert, 184 Quadratic character, 440, 444
split, 184 Quadratic field, non-maximal order, 178,
splits completely, 117, 118, 179 180
Primitive Dirichlet character, 93 Quadratic form, canonical height, 454
Primitive root of unity, 21, 23, 29, 154 Quadratic imaginary field
Principal branch Q(i), 138
of logarithm, 44, 54, 56, 66 abelian extension of, 128, 135
of square root, 65, 67 algorithm to compute ideal class
Principal divisor, 232 group, 85
on an arithmetic surface, 339 Hilbert class field of, 95, 121, 122, 142,
Principal ideal, 132 166, 181
congruent to 1 modulo c, 117 ideal class group, 85
Pro-finite topology, 120, 182 maximal abelian extension, 135
Pro-p group, 150, 182 of class number one, 138, 141
Product ring of integers, 85
fiber, 299 Quadratic transformation, 244
restricted, 119 Quasi-parallelogram law, 213, 467, 476
Product expansion of sine function, 56, explicit 0(1) estimate, 280
58 Quasi-period homomorphism, 41, 43, 44,
Projection morphism on fiber product, 52, 53, 65, 89, 412, 464, 465, 466
299 See also Legendre relation
Projective closure of a scheme, 371 formula for, 41
Projective line Quasi-periodicity of (, 40
Index 521
Quaternion algebra, 130 split multiplicative, 399

Quaternion group, 404 Reduction type, 352, 3.53, :363, 364
Quotient topology, 8.5 over 3-adic field, 406, 407
on X(l), 18, 19 Regular
arithmetic surface, 311
differential form, 27
Ramanujan , S., 61, 79 in codimension one, 311
Ramanujan conjecture, 61, 81 Regular arithmetic surface, 316, 317,
Ramanujan "T function, .59, 60, 482 3,51,370,399
growth properties, 61 non-singular point on fiber, 315, 351,
multiplicative identities, 61, 79 399
special values, 60 smooth part, 316, 318, 321, 32.5, 332,
Ramification index, 283 335, 361, 362, 369, :378, 399, 400
Ramification, wild, 380 Weierstrass equation, 399
Rank .
Regular local ring, 302, 370
high over Q, 272 localization is regular, 397
of Neron-Severi group, 2.54
of dimension one, 328
upper bound, 279
of dimension two. 31.5
Rank one module, 102, 138, 186
Rational map, 203 . Regular model, minimal proper, 325,
332, 33.5, 337, 400
birational isomorphism, 204
Regular point, 302
blowing-up, 204, 23.5
domain of definition, 204, 279 on a scheme, 397
dominant, 204, 222, 228 Regular scheme, :302, 398, 399
from smooth scheme to proper scheme, example, 313
328 intuitive definition, 301
image, 204 of dimension one, 302
is algebraic, 204, 279 smooth part, 399
is irreducible, 204, 280 \Veierstrass equation, 399
naive definition, 204 Regulator, height, 273
of elliptic surfaces, 206 Reimann conditions, 196, 198
of varieties, 132 Relative differentials. 306
Ray class field, 117, 118, 129, 13.5, 161 Relative dimension ~f a smooth mor-
conductor of, 117, 1l8, 129, 13,5 phism, 305
idelic characterization. 120. 162 Relatively minimal fibered surface, 243
modulo unit ideal, 118 . Relatively minimal surface, 235
of Q( i), 138, 180 Representation
of Q( V"=3), 180 irreducible, 405
primes splitting completely, 118 of type E s , 189
Raynaud, I'vI., 3.50, 486 Residue symbol, 6th-power, 177
Real Lie group, 420 Resolution of singularities, 2:3.5
Reciprocity Restricted product, 119
Artin.117 Resultant, 458
Dedekind, 67, 90 Riemann ( function, 25, 5.5, 171
map, 120, 1.52, 1.59, 165, 166, 168, 169, functional equation, 83
174 special values, 26, 33, .57
Reduction map Riemann hypothesis for varieties over
injective on torsion, 162 finite fields, 61, 81
surjectivity, 330, 333, 337, 401 Riemann surface. 14, 19, 23, 27
Reduction See also Curve
additive, 171,287,381,388, :399, 403 differential I-forms on, 27, 43
bad, 321 meromorphic k-forms on, 27
good, 287, 321, 329, 381, 383, 388 Riemann-Roch theorem, 28, 30, 32, 87,
multiplicative, 171, 287, 369, 381, 388, 198, 226, 238, 34.5, 430
403 Rigid analysis, 430
non-singular points, 362, :379 Rigidity lemma, 296, 396
non-split multiplicative, 399 Ring of integers of a quadratic imaginary
of a scheme. 300 field, 8.5
of an isogeny, 124 Ring
of Weierstrass equation, 362, 378 Krull dimension of, 301
ordinary, 179 of correspondences, 70
semi-stable, 388 regular local, 302
522 Index
R-morphism, 297 Segre embedding, 261

Rohrlich, D .. 485 Self-intersection, 234, 238, 24:3, 28:3, :342.
Root lattice of type En, 254 349, .3;)1 .
Root of unity equals minus one, .344, 3S2
primitive, 21, 23, 29, 154 Semi-stable reduction, :388
group scheme of. 397, 398 Separated morphism, .305
Rosen, lvI., 38;), 485 Serre, J.-P., 140, 149, 261, 381, 447, 487
R-scherne, 297 Serre's theorem. 261
generic fiber. 300 Shafarevich. I.R, 317, :338, 341
reduction modulo p. 300 Shatz, S., ,187
set of R-valued points, 298 Sheaf
Ruled surface. 2:36 global sections, 397
R-valued points. 298 of relative differentials. 306
of an affine scheme, 298 Shimura, G .. 9.5 .
Shioda, T., 189, 247, 254, ·186
Siegel, C.L., 62, 275, 278
sk(q),410
alternative form. 448 Sign of functional equation, 176
81 (q),111. 425
Silverm;;n, .l.R .. 265. 266, 269, 271, 277,
278, .385, 486, 487
Sn, 72, 1·13
order of, 72 Sine function, 56, 58
S.;, , 181 Singular fib{'r, 203
order of, 181 Singular j-invariant, 104
a See Vveierstrass a function Singular point, 471
ax-. 55. 61 on fiber. 314
Saito. T., :389, 390 on special fiber, 32S
Scheme, 434 S-integers
dimension, 302 effective methods, 277, 28S
excpllent, 311 of a function field, 275
fiber produd, 299 points on an elliptic surface. 275
finite type, 311 SL2
flat, 311 See Special linear group
generic fiber, 300 Smooth arithmetic snrfacc, 311 . .316
generic point, 304 Smooth morphism, 305, 306
group, 306 composition of, 304. ;)20
integral. :311 intuitive definition, 301
nice. 311, :399 of relative dimension zero, 30(j
non-singular point, :302 Smooth scheme, :312, :)97, :399
normal, 311, 328 map to proper scheme, 328
of Dedekind domain, 302 morphism to group scheme. 327, .333,
over a base ring, 297 3:37
over a base S, 297 over a DVR, :304
projectivc closure, 371 surjectivity of reduction, 3:~0, .33:3, 3.37,
proper, 305, 321, :,22, 329, :399 401
reduction modulo p. 300 Weierst.rass equation, 321. 329, 362,
regular, :301, 398, 399 369, .378, 399, 400
regular of dimension one, 302 Smooth special fiber, 401
regular point, 302, 397 S-morphism, 297, 299
smooth, 328, :197, 40} Special fiber
smooth part, 399 dual to Dynkin diagram, 353
structure sheaf on. 397 Galois action on, 353
Schmidt, \V .. 277 . group of components, .36:3, 364
Schncider, T. 108 identity component, 326, .361
SchnirPlrnann's Theorem, 429 incidence matrix, 350. 102, 403, 486
Second Bernoulli polynomial. 468, 473, Kodaira notation, :353
478, 480 lIlultiplicity one components, ,102
Second kind differential form, 43 Neron notation, :{53
Section number of components, 363, 364
divisor of a, 245 of an arithmetic surface, :Ul, 311, 361,
to a morphism, 202 .399
to an elliptic surface, 202 of blow-up, 317, 371
to an S-scheme, 298. 309, 397 of curve of genus two. 355
Index 523
of minimal proper regular modeL 350, exceptional curve, 2:36

352, 40:3 fibered, 2:36
of Neron modeL 350, 361 intersection pairing, 233, 238
over a DVR, 300 linear equivalence of divisors, 232
oingular points, 325 local ring at a curve, 231
smooth, 401 local ring at a point, 231
Special linear group, S, 292 Picard group of, 232
See also Action of SL2 (Z) on H principal divisor, 232
action on 'D n, 72 relatively minimal, 235
reduction mod N. 86 resolution of singularities, 235
Special orthogonal group, :~96 ruled, 236
Specialization map, 271 Swan character, 405
constructing high rank curves. 272 Symmetric product, 229
for height, 265. 266. 269, 2S1, 2S6 of a curve, 197. 402
is a homomorphism, 271 Symplectic group. 396
is injective, 271, 2S1 Szpiro, L., 388,486
on split elliptic surface, 271 Szpiro's conjecture, 388
Spectrum of a ring, 259 function field analogue, 287, 388
Split elliptic ourface, 220, 2S6 used to bound torsion, 453
canonical height on, 281
characterization of. 220, 228, 280
Tame part of the conductor, 380, 381,
has constant j, 221, 280
405
if infinitely many bounded height invariant under isogeny, 404
points, 222 Tangent line, 122-429, 472
Mordell-,Veil theorem, 2:31. 281
Tate. J" 140. 149, 173, 265, 269, 361,
specialization map is injective, 281 381,42:3,441, 4.54, 455, 459, 486, 487
specialization of height. 281 Tate curve' E q , 423, 441, 444
Split multiplicative reduction, 362. 377, See also Elliptic curves over p-adic
379. :399 fields
Split prime, 184 diocriminant of, 377, 423, 424, 47:~
Sporadic groups. 61 filtration of rational points, 430, 450
Square-free conductor. 38S formal group of, 431
S'-ocheme, 297 formula for j, 111
fiber product, 299 formula for discriminant, 411
morphism of, 297 ,-invariant, 4·11
proper. :301 has split multiplicative reduction, ::l77,
set of T-valued points, 298, 309, 397 441
set of sections, 29S, 309, 397 isogeny of, 451, 453
smooth, :~01 j-invariant of, 123, 425, 438, 473
Stabilizer, 11 local height on, 473
Stark, I-I., 141 Neron modeL 377, ·135
Strict I-Icnseliza,tion, 337 reduction m;dulo 9)1, :377, ·130, ,142
faithfully fiat, 3:38 scheme associated to, 377. 4:H
fraction field. :332 surjectivity of p-adic uniformization.
of a DVR, 331 429 43S
universal mapping property, 332. 401 torsion subgroup, 445
Strictly I-Ienselian ring, 330. 402 uniformization compatible with Galois.
density of reduction, :3:30, 33:3, 337 42,1
etale topology, 332 uniformization over p-adic fields, 141.
Neron model over, 332, 335 377. 423, 450, 473, 480
Structure ohear, ::l97 \Veierstrass equation, 423, 426, 412,
Sum. telescoping, 269 473
S'-unit equation over function fields, 278 Weil-Chatelet group, 152
Supersingular j-invariant, 104 Tate module. 380
Supersingular reduction, 184 action of Frobenius, 172
Surface of a Cl\I elliptic curve, 148, 149, 179
See also Fibered surface; Elliptic of a field. 452
ourface: Arithmetic surface of an abelian group, :~82
adjunction formula, 2:34 of an elliptic curve, 152
ample divisor on, 258 of the aclditive group, 382
divisor, 231 of the multiplicative group, 382
524 Index
Tate module (continued) of an elliptic curve, 414, 439

p-adic, 186 of an L series, 94
Weil pairing, 125
Tate's p-adic uniformization theorem,
377, 441 Veno, K., 355, 486
Tate's algorithm, 353, 361, 364-368, 387, Uniformization Theorem, 6, 34, 35, 88,
389,435 96, 196, 411, 413
minimal Weierstrass equation, 364 of elliptic curves over p-adic fields, 423,
multiplicative reduction, 377 441
verification, 369-377 of elliptic curves over JR, 416
T(n), See Ramanujan r function Uniformizer, 27
Tchebotarev density theorem, 161 for a curve on an arithmetic surface,
Telescoping sum, 269, 460 339
Theorem 90 of Hilbert, 193 Unit disk, holomorphic automorphism of,
Theta function B( u, q), 412, 429 22
p-adic, 429, 473 Unit group
functional equation, 412, 429, 474 of a function field, 191
relation to X(u,q) and Y(u,q), 412, of a number field, 191
429, 474 Universal family, point of order eleven,
relation to Weierstrass u function, 412 190, 279
Universal property of fiber product, 299
Third kind differential form, 43
Universally closed morphism, 305
Topological group, 85, 119
Unramified extension
Topological space, irreducible, 280
cyclotomic, 383
Topology
generated by torsion, 383
etale,332
of discrete valuation ring, 399
on a valued field, 455
Unramified Grossencharacter, 168, 173,
on an elliptic curve, 455
174
pro-finite, 120, 182
Upper half plane H, 7
Torsion module is sum of p-primary com-
action of SL2(Z), 9
ponents, 157, 170
differential form on, 26
Torsion point
extended (H*)
boundedness conjecture, 453
See Extended upper half plane
field generated by, 128, 135, 383
fundamental domain for SL2(Z) ac-
forces good/multiplicative reduction,
tion, 10, 14, 25, 92
453
stablizer of a point, 11
of multiplicative group, 151
of order eleven, 190, 279
of order five, 278 v-adic topology
of order seven, 278 multiplication map is continuous, 460
on CM curve form free module, 102, on a field, 455
138 on an elliptic curve, 455, 476
on CM elliptic curves, 102, 135 Valuation
on elliptic curves over p-adic fields, of a curve on an arithmetic surface,
383,445 311, 313, 339
reduction is injective on, 162 on a function field, 194
Torsion subgroup of Picard group, 192, Valuative criterion of properness, 303,
195, 199 317, 329, 333
Totally imaginary field, 116 Variety
Transcendence of j, 108 abelian, 196
Translation map, 336 birational equivalence of, 204, 205
on a group scheme, 336 category up to birational equivalence,
on an elliptic surface, 245, 248, 251 205
on elliptic curve, 310 curve through two points, 228, 281
on group scheme, 310, 337, 400 Frobenius map of, 132
on Neron model, 400 group, 291
Transvection, 445, 447 group of divisors, 255
Transversal intersection, 232, 282 height associated to a morphism, 258
Triangle inequality, 459, 471 height on, 256
Triplication formula, for local height, 463 rational map of, 132
Twist Vertical strip, 94
of a cusp form, 94 Very ample divisor, 257, 261, 284
Index 525
Vojta, P., 278 Weierstrass <7 function, 44, 62, 412, 464
Voloch, J.F., 276 factorization of elliptic functions, 45
factorization of p, 45, 63, 90
Fourier expansion, 53, 467, 468
Weak approximation theorem, 155, 158
Weak Mordell-Weil theorem, 191-195,
relation to e, 412
special values of, 63
230 \Veierstrass ( function, 39, 43
\Veakly modular function, 24, 74, 91 Fourier expansion, 52
See also Modular function Laurent series, 39, 52
Weber, R., 95 quasi-periodicity property, 40
Weber function, 134, 135 \Veight of a modular function, 24
analytic definition, 135 Weight 12 cusp form is unique, 34
field of definition, 135 Wei!, A., 198, 256, 327, 334,402
Weierstrass equation Weil-Chatelet group
addition is morphism, 323 of an elliptic curve over 1Ft, 421, 449
algorithm to find minimal, 364, 403 of Tate curve E q , 452
associated lattice, 35, 88, 89 Weil divisor on an arithmetic surface,
bad reduction, 321 311, 339
blow-up of, 371 Weil height
good reduction, 321, 329 See Height
group scheme, 321, :{29, 362, 369, 378, Weil pairing, 89, 125, 387
399,400 Wild part of the conductor, :{80, 381, 405
identity component of Neron model, invariant under isogeny, 404
362. 378 \Vild ramification, 380
mini~al, 321, 403, 478, 480
for p :::: 3, 406
for p :::: 5, 405 X(I), See rvlodular curve X(I)
negation map is morphism, 325
over IQ, 403
proper ticheme, 321, 329 Y(I), See Modular curve X(I), affine
reduction modulo p, 362, 378 part
scheme defined by, 321, :{29, 362, 369, Yoneda's lemma, 299
378, 399, 400
smooth => Neron model, 329 Zp-extension, 186
smooth part over R, 321, 362, 369, ( function
378, 399, 400 See also Weierstrass ( function; Rie-
with good reduction, 462, 469 mann ( function
with rational two-torsion, 192, 194 of a number field, 18:{
Weierstrasti p function, 6, 34, 39, 408, of an elliptic curve over a finite field,
464 183
at ha.lf periods, 62, 90 of an elliptic curve over a number field,
at N-torsion points, 89 183
factorization by <7, 45, 63, 90, 412, 464, Zagier, D., 217
467 Zariski's l\lain Theorem, 283, 486
Fourier expansion, 47, 50, 409 Zero section to an elliptic surface, 202
Laurent series, 39 Zucker, S., 486
Graduate Texts in Mathematics
(continued from page ii)
62 KARGAPOLOv/MERLZJAKOV. Fundamentals 93 DUBROVIN/FoMENKo/NOVIKOV. Modern

of the Theory of Groups. Geometry-Methods and Applications.
63 BOLLOBAS. Graph Theory. Part I. 2nd ed.
64 EDWARDS. Fourier Series. Vol. I 2nd ed. 94 WARNER. Foundations of Differentiable
65 WELLS. Differential Analysis on Complex Manifolds and Lie Groups.
Manifolds. 2nd ed. 95 SHIRYAEV. Probability. 2nd ed.
66 WATERHOUSE. Introduction to Affine 96 CONWAY. A Course in Functional
Group Schemes. Analysis. 2nd ed.
67 SERRE. Local Fields. 97 KOBUTZ. Introduction to Elliptic Curves
68 WEIDMANN. Linear Operators in Hilbert and Modular Forms. 2nd ed.
Spaces. 98 BROCKERlToM DIECK. Representations of
69 LANG. Cyclotomic Fields II. Compact Lie Groups.
70 MASSEY. Singular Homology Theory. 99 GRoVE/BENSON. Finite Reflection Groups.
71 FARKAS/KRA. Riemann Surfaces. 2nd ed. 2nd ed.
72 STILLWELL. Classical Topology and 100 BERGICHRISTENSEN/RESSEL. Harmonic
Combinatorial Group Theory. 2nd ed. Analysis on Semigroups: Theory of
73 HUNGERFORD. Algebra. Positive Definite and Related Functions.
74 DAVENPORT. Multiplicative Number 101 EDWARDS. Galois Theory.
Theory. 2nd ed. 102 VARADARAJAN. Lie Groups, Lie Algebras
75 HOCHSCHILD. Basic Theory of Algebraic and Their Representations.
Groups and Lie Algebras. 103 LANG. Complex Analysis. 3rd ed.
76 lITAKA. Algebraic Geometry. 104 DUBROVIN/FoMENKo/NOVIKOV. Modern
77 HECKE. Lectures on the Theory of Geometry-Methods and Applications.
Algebraic Numbers. Part II.
78 BURRIslSANKAPPANAVAR. A Course in 105 LANG. SLz(R).
Universal Algebra. 106 SILVERMAN. The Arithmetic of Elliptic
79 WALTERS. An Introduction to Ergodic Curves.
Theory. 107 OLVER. Applications of Lie Groups to
80 ROBINSON. A Course in the Theory of Differential Equations. 2nd ed.
Groups. 2nd ed. 108 RANGE. Holomorphic Functions and
81 FORSTER. Lectures on Riemann Surfaces. Integral Representations in Several
82 BorrlTu. Differential Forms in Algebraic Complex Variables.
Topology. 109 LEHTO. Univalent Functions and
83 WASHINGTON. Introduction to Cyclotomic Teichmiiller Spaces.
Fields. 2nd ed. 110 LANG. Algebraic Number Theory.
84 IRELAND/ROSEN. A Classical Introduction 111 HUSEM6LLER. Elliptic Curves.
to Modern Number Theory. 2nd ed. 112 LANG. Elliptic Functions.
85 EDWARDS. Fourier Series. Vol. II. 2nd ed. 113 KARATZAslSHREVE. Brownian Motion and
86 VAN LINT. Introduction to Coding Theory. Stochastic Calculus. 2nd ed.
2nd ed. 114 KOBLITZ. A Course in Number Theory and
87 BROWN. Cohomology of Groups. Cryptography. 2nd ed.
88 PIERCE. Associative Algebras. 115 BERGERlGoSTlAUX. Differential Geometry:
89 LANG. Introduction to Algebraic and Manifolds, Curves, and Surfaces.
Abelian Functions. 2nd ed. 116 KELLEy/SRINIVASAN. Measure and
90 BR0NDSTED. An Introduction to Convex Integral. Vol. I.
Polytopes. 117 SERRE. Algebraic Groups and Class Fields.
91 BEARDON. On the Geometry of Discrete 118 PEDERSEN. Analysis Now.
Groups. 119 ROTMAN. An Introduction to Algebraic
92 DrnSTEL. Sequences and Series in Banach Topology.
Spaces.
120 ZIEMER. Weakly Differentiable Functions: 149 RATCLIFFE. Foundations of
Sobolev Spaces and Functions of Bounded Hyperbolic Manifolds.
Variation. ISO EISENBUD. Commutative Algebra
121 LANG. Cyclotomic Fields I and II. with a View Toward Algebraic
Combined 2nd ed. Geometry.
122 REMMERT. Theory of Complex Functions. 151 SILVERMAN. Advanced Topics in
Readings in Mathematics the Arithmetic of Elliptic Curves.
123 EBBINGHAUS/HERMES et aI. Numbers. 152 ZIEGLER. Lectures on Polytopes.
Readings in Mathematics 153 FuLTON. Algebraic Topology: A
124 DUBROYIN/FoMENKO/NoVlKOY. Modern First Course.
Geometry-Methods and Applications. 154 BRowNIPEARCY. An Introduction to
Part III. Analysis.
125 BERENSTEIN/GAY. Complex Variables: An 155 KASSEL. Quantum Groups.
Introduction. 156 KECHRIS. Classical Descriptive Set
126 BOREL. Linear Algebraic Groups. 2nd ed. Theory.
127 MASSEY. A Basic Course in Algebraic 157 MALLIA YIN. Integration and
Topology. Probability.
128 RAUCH. Partial Differential Equations. 158 ROMAN. Field Theory.
129 fuLTON/HARRIS. Representation Theory: A 159 CONWAY. Functions of One
First Course. Complex Variable II.
Readings in Mathematics 160 LANG. Differential and Riemannian
130 DODSON/POSTON. Tensor Geometry. Manifolds.
131 LAM. A First Course in Noncommutative 161 BORWEIN/ERDELYI. Polynomials and
Rings. Polynomial Inequalities.
132 BEARDON. Iteration of Rational Functions. 162 ALPERINIBELL. Groups and
133 HARRIS. Algebraic Geometry: A First Representations.
Course. 163 DIXONIMORTIMER. Permutation
134 ROMAN. Coding and Information Theory. Groups.
135 ROMAN. Advanced Linear Algebra. 164 NATHANSON. Additive Number Theory:
136 ADKINS/WEINTRAUB. Algebra: An The Classical Bases.
Approach via Module Theory. 165 NATHANSON. Additive Number Theory:
137 AxLERlBoURDONIRAMEY. Harmonic Inverse Problems and the Geometry of
Function Theory. Sumsets.
138 COHEN. A Course in Computational 166 SHARPE. Differential Geometry: Cartan's
Algebraic Number Theory. Generalization of Klein's Erlangen
139 BREDON. Topology and Geometry. Program.
140 AUBIN. Optima and Equilibria. An 167 MORANDI. Field and Galois Theory.
Introduction to Nonlinear Analysis. 168 EWALD. Combinatorial Convexity and
141 BECKER/WEISPFENNING/KREDEL. Grabner Algebraic Geometry.
Bases. A Computational Approach to 169 BHATIA. Matrix Analysis.
Commutative Algebra. 170 BREDON. Sheaf Theory. 2nd ed.
142 LANG. Real and Functional Analysis. 171 PETERSEN. Riemannian Geometry.
3rd ed. 172 REMMERT. Classical Topics in Complex
143 DOOB. Measure Theory. Function Theory.
144 DENNISIFARB. Noncommutative 173 DIESTEL. Graph Theory.
Algebra. 174 BRIDGES. Foundations of Real and
145 VICK. Homology Theory. An Abstract Analysis.
Introduction to Algebraic Topology. 175 LrCKORISH. An Introduction to Knot
2nd ed. Theory.
146 BRIDGES. Computability: A 176 LEE. Riemannian Manifolds.
Mathematical Sketchbook. 177 NEWMAN. Analytic Number Theory.
147 ROSENBERG. Algebraic K-Theory 178 CLARKEILEDYAEY/STERN/WOLENSKI.
and Its Applications. Nonsmooth Analysis and Control
148 ROTMAN. An Introduction to the Theory.
Theory of Groups. 4th ed. 179 DOUGLAS. Banach Algebra Techniques in
Operator Theory. 2nd ed.
180 SRlV AST AVA. A Course on Borel Sets. 189 LAM. Lectures on Modules and Rings.
181 KREss. Numerical Analysis. 190 ESMONDEIMuRTY. Problems in Algebraic
182 WALTER. Ordinary Differential Number Theory.
Equations. 191 LANG. Fundamentals of Differential
183 MEGGINSON. An Introduction to Banach Geometry.
Space Theory. 192 HIRscH/LACOMBE. Elements of
184 BOLLOBAS. Modem Graph Theory. Functional Analysis.
185 CoXiLITILElO'SHEA. Using Algebraic 193 COHEN. Advanced Topics in
Geometry. Computational Number Theory.
186 RAMAKRISHNANIV ALENZA. Fourier 194 ENGEliNAGEL. One-Parameter Semigroups
Analysis on Number Fields. for Linear Evolution Equations.
187 HARRIS/MORRISON. Moduli of Curves.
188 GoLDBLATT. Lectures on the Hyperreals:
An Introduction to Nonstandard Analysis.

(Graduate Texts in Mathematics 151) Joseph H. Silverman (Auth.) - Advanced Topics in The Arithmetic of Elliptic Curves-Springer-Verlag New York (1994) PDF

Uploaded by

Copyright:

Available Formats

(Graduate Texts in Mathematics 151) Joseph H. Silverman (Auth.) - Advanced Topics in The Arithmetic of Elliptic Curves-Springer-Verlag New York (1994) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Graduate Texts in Mathematics 151) Joseph H. Silverman (Auth.) - Advanced Topics in The Arithmetic of Elliptic Curves-Springer-Verlag New York (1994) PDF

Uploaded by

Copyright:

Available Formats

What topics are covered in the texts?

What topics are covered in the texts?

What mathematical areas do the texts focus on?

What mathematical areas do the texts focus on?

Graduate Texts in Mathematics 151

Springer Science+Business Media, LLC

T AKEUTIIZARING. Introduction to 33 HIRSCH. Differential Topology.

Mathematics Subject Classifications (1991): 14-01, llGxx, 14Gxx, 14H52

Library of Congress Cataloging-in-Publication Data

Printed on acid-free paper.

© 1994 Springer Science+Business Media New York

Production managed by Hal Henglein; manufacturing supervised by Vincent Scelta.

9 8 7 6 5 4 3 2 (Corrected second printing, 1999)

ISBN 978-0-387-94328-2 SPIN 10727882

In the introduction to the first volume of The Arithmetic of Elliptic Curves

the analytic theory of p-adic functions (rigid analysis) in Chapter V, and

Acknowledgments for the Second Printing

Notes on Exercises 484

List of Notation 498

In the first volume of The Arithmetic of Elliptic Curves, we pre-

{maps (J : C ---- £ such that 7r 0 a(t) = t}

to an elliptic surface forms a group, and we prove an analogue of the

to represent the integers, rational numbers, real numbers, complex num-

Elliptic and Modular Functions

product expansion and analytic continuation and that it satisfies a func-

§1. The Modular Group

C/ A ---. EA(C) : y2 = 4x 3 - g2X - g3

Z f----> (p(z; A), p'(z; A)).

is the Weierstrass p-function relative to the lattice A. (See [AEC VI,§3].)

if and only if Al and A2 are homothetic.

We let C* act on £. by multiplication,

An Oriented Basis for the Lattice A

Since we only care about A up to homothety, we can normalize our

which suggests looking at the upper half-plane

H = {T E <C : Im( T) > O}.

We have just shown that the natural map

H --> f., j<C* ,

Lemma 1.1. Let a, b, c, dE lR, T E lC, T tJ- lR. Then

1m (aT + b) = (ad - bc) Im(T).

PROOF. Let T = s+it. MUltiplying numerator and denominator by cf+d,

aT+ b {aclTl2 + (ad + bc)s + bd} + {(ad - bc)t}i

Then there are integers a, b, c, d, a', b', c', d' so that

w~ = aWl + bW2, Wl = a'w~ + b'w;,

b) c(a'' d'b') = (10 0

In other words, the matrix (~ ~) is in the special linear group over Z,

This proves the first half of the following lemma.

(c) Let A c C be a lattice. Then there is aTE H such that A is homothetic

~ {T2 = aO'Tl + bO' for some (ac db) E SL 2(Z),

,T = ~;:! for, = (::) E SL2(Z) and T E H.

Definition. The modular group, denoted r(l), is the quotient group

r(l) = SL2 (Z)/{±1}.

Although r(l) is the quotient SL 2 (Z)/{±1}, we will generally just

from which we conclude that c = b = 0 and a = d. Hence, = ±1.

Their action on H is given by

so r(l) contains finite subgroups of order 2 and 3.

~ and Some of Its r(l)-nanslates

(i) Re(T) = -4 and iT = T + 1;

(c) Let T E ~, and let

Write T = S + it. Since t > 0, it is clear that

Then, for all Tl, T2 E H, there exist open neighborhoods U l , U2 C H