Mordell Weil

Aspects of Mathematics
Jean-Pierre Serre
Lectures on the
Mordell-Weil
Theorem
Third Edition
Jean-Pierre Serre
Lectures on the
Mordell-Weil Theorem
Asped~f
Ivbthematia
Edited by Klas Diederich

Vol. E 3:
G. Hector/U. Hirsch: Introduction to the Geometry of Foliations, Part B
Vol. E 5:
P. Stiller: Automorphic Forms and the Picard Number of on Elliptic Surface
Vol. E 6:
G. Faltings/G. Wustholz et 01.: Rational Points*
Vol. E 9:
A. Howard/P.-M. Wong (Eds.): Contribution to Several Complex Variables
Vol. E 10: A. J. Tromba (Ed.): Seminar of New Results in Nonlinear Partial Differential
Equations *
Vol. E 15: J.-P. Serre: lectures on the Mordell-Weil Theorem
Vol. E 16: K. Iwasaki/H. Kimura/S. Shimomura/M. Yoshida: From Gauss to Painleve
Vol. E 17: K. Diederich lEd.): Complex Analysis
Vol. E 18: W. W. J. Hulsbergen: Conjectures in Arithmetic Algebraic Geometry
Vol. E 19: R. Rocke: lectures on Nonlinear Evolution Equations
Vol. E 20:
F. Hirzebruch, Th. Berger, R. Jung: Manifolds and Modular Forms*
Vol. E 21:
H. Fujimoto: Value Distribution Theory of the Gauss Mop of Minimal

Surfaces in Rm
Vol. E 22:
D.
V. Anosov/A. A. Bolibruch: The Riemann-Hilbert Problem
Vol. E 23: A. P. Fordy/J.

Vol. E 24:
C. Wood (Eds.): Harmonic Mops and Integrable Systems
D: S. Alexander: A History of Complex Dynamics
Vol. E 25: A. Tikhomirov/A. Tyurin (Eds.): Algebraic Geometry and its Applications
Vol. E 26:
H. Skoda/J.-M. Trepreau IEds.): Contributions to Complex Analysis and

AnalytiC Geometry
Vol. E 27:
D. N. Akhiezer: lie Group Actions in Complex Analysis
Vol. E 28:
R. Gerard, H. Tahara: Singular Nonlinear Partial Differential Equations
Vol. E 29:
R.-P. Holzapfel: Boll and Surface Arithmetics
Vol. E 30:
R. Huber: Etale Cohomology of Rigid AnalytiC Varieties and Adic Spaces
* A Publication af the Max-Planck-Institut fUr Mathematik,
Bonn
Jean-Pierre Serre
Lectures on the
Mordell-Weil Theorem
Translated and edited by Martin Brown
from notes by Michel Waldschmidt
3rd edition
Springer Fachmedien Wiesbaden GmbH
II
Vleweg
Prof. lean-Pierre Serre

College de France
Chaire d'Algebre et Geometrie
75005 Paris
AMS Subject Classification: 14 G 13, 14 K 10, 14 K 15
1st edition 1989
2nd edition 1990
3rd edition 1997
All rights reserved

Springer Fachmedien Wiesbaden 1997
Originally published by Friedr. Vieweg & Sohn Verlagsgesellschaft mbH, BraunschweiglWiesbaden, 1997
No part of this publication may be reproduced, stored in a retrieval

system or transmitted, mechanical, photocopying or otherwise, without
prior permission of the copyright holder.
Cover design: Wolfgang Nieger, Wiesbaden

Printed on acid-free paper
ISSN 0179-2156
ISBN 978-3-663-10634-0
DOI 10.1007/978-3-663-10632-6
ISBN 978-3-663-10632-6 (eBook)
Foreword
This is a translation of "Autour du theoreme de Mordell-Well", a course

given by J.-P. Serre at the College de France in 1980 and 1981.
These notes were originally written weekly by Michel Waldschmidt and have
been reproduced by Publications Mathematiques de l'Universite de Paris VI,
by photocopying the handwritten manuscript.
The present translation follows roughly the French text, with many modifications and rearrangements. We have not tried to give a detailed account
of the new results due to Faltings, Raynaud, Gross-Zagier ... ; we have just
mentioned them in notes at the appropriate places, and given bibliographical
references.
Paris, Fall 1988
M.L.Brown
I.-P. Serre
VII
CONTENTS
1. Summary.
1.1. Heights.
1.2. The Mordell-Weil theorem and Mordell's conjecture.
1.3. Integral points on algebraic curves. Siegel's theorem.
1.4. Baker's method.
1.5. Hilbert's irreducibility theorem. Sieves.
1
3
3
4
5
5
2. Heights.
2.1. The product formula.
2.2. Heights on P m(K).
2.3. Properties of heights.
2.4. Northcott's finiteness theorem.
2.5. Quantitative form of Northcott's theorem.
'2.6. Height associated to a morphism <p: X -+ P n
2.7. The group Pic(X).
2.8. Heights and line bundles.
2.9. he = 0(1) {:} c is of finite order (number fields).
2.10. Positivity of the height.
2.11. Divisors algebraically equivalent to zero.
2.12. Example-exercise:, projective plane blo~ up at a point.
7
7
10
13
16
17
19
20
22
24
24
25
26
3. Normalised heights.
3.1. Neron-Tate normalisation.
3.2. Abelian varieties.
3.3. Quadraticity of he on abelian varieties.
3.4. Quality and Poincare divisors.
3.5. Example: elliptic curves.
3.6. Exercises on elliptic curves.
3.7. Applications to properties of heights.
3.8. Non-degeneracy.
3.9. Structure of A(K): a preliminary result.
3.10. Back to 2.11 (c algebraically equivalent to zero).
3.11. Back to 2.9 (torsion c).
29
29
31
35
36
39
40
41
'42
43
44
46
VIII
Contents
4. The Mordell-Well theorem.

4.1. Hermite's finiteness theorem.
4.2. The Chevalley-Weil theorem.
4.3. The Mordell-Weil theorem.
4.4. The classical descent.
4.5. The number of points of bounded height on an
abelian variety.
4.6. Explicit form of the weak Mordell-Wei! theorem.
49
49
50
51
53
53
55
5. Mordell's conjectUre.
5.1. Chabauty's theorem.
5.2. The Manin-Demjanenko theorem.
5.3. First application: Fermat quartics (Demjanenko).
5.4. Second application: modular curves Xo(pn) (Manin).
5.5. The generalised Mordell conjecture.
5.6. Mumford's theorem; preliminaries.
5.7. Application to heights: Mumford's inequality.
58
58
62
66
67
73
74
77
6. Local calculation of normalised heights.

6.1. Bounded sets.
6.2. Local heights.
6.3. Neron's theorem.
6.4. Relation with global heights.
6.5. Elliptic curves.
81
81
83
87
7. Siegel's method.
7.1. Quasi-integral sets.
7.2. Approximation of real numbers.
7.3. The approximation theorem on abelian varieties.
7.4. Application to curves of genus ~ 1.
7.5. Proof of Siegel's theorem.
7.6. Application to P(J(n.
7.7. Effectivity.
94
94
8. Bal(er's method.
8.1. Reduction theorems.
8.2. Lower bounds for 'B{3i log (ri.
8.3. Application to P l - {O, 1,00}.
8.4. Applications to other curves.
8.5. Applications to elliptic curves with good reduction outside
a given finite set of places.
89
90
95
98
101
102
105
106
108
108
110
112
114
118
Contents
IX
9. Hilbert's irreducibility theorem.

9.1. Thin sets.
9.2. Specialisation of Galois groups.
9.3. Examples of degrees 2,3,4,5.
9.4. Further properties of thin sets.
9.5. Hilbertian fields.
9.6. The irreducibility theorem: elementary proof.
9.7. Thin sets in PI: upper bounds.
121
121
122
123
127
129
130
132
10. Construction of Galois extensions.

10.1. The method.
10.2. Extensions with Galois group Sn.
10.3. Extensions with Galois group An.
10.4. Further examples of Galois groups: use of
elliptic curves.
10.5. Noether's method.
10.6. Infinite Galois extensions.
10.7. Recent results.
137
137
138
144
145
147
147
149
11. Construction of elliptic curves of large rank.

11.1. Neron's specialisation theorem.
11.2. Elliptic curves of rank ~ 9 over Q.
152
152
154
158
161
12. The large sieve.

12.1. Statement of the main theorem.
12.2. A lemma on finite groups.
12.3. The Davenport-H8.lberstam theorem.
12.4. Proof of the Davenport-Halberstam theorem.
12.5. End of the proof of the main theorem.
163
163
164
166
167
13. Applications of the large sieve to thin sets.

13.1. Statements of results.
13.2. Proof of theorem 1.
13.4. Proof of theorem 3 from theorem 1.
177
172
177
179
183
186
Contents
Appendix: The class number 1 problem and integral points on

modular curves.
A.I. Historical remarks.
A.2. Equivalent conditions for h( -p) = 1.
A.3. Orders of Rd.
A.4. Elliptic curves with complex multiplication.
A.5. Modular curves associated to normalisers of Cartan
subgroups and their CM integral points.
A.6. Examples.
A.7. The Gel'fond-Linnik-Baker method.
196
197
Bibliography.
200
Index.
210
188
188
190
191
192
194
1
1. SUMMARY
The aim of this course is the study of rational and integral points on
algebraic varieties, especially on curves or abelian varieties. Before the
end of the last century only special cases had been considered. The first
general results are found, around 1890, in the work of Hurwitz and Hilbert
[HH] where they introduced the, nowadays natural, viewpoint of algebraic
geometry: if X and XI are two birationally equivalent algebraic curves
over Q in P 2 , their rational points correspond. Hence the importance of
birational invariants, in particular the genus. They studied especially the
case of genus zero:
Theorem. A curve of genus zero is isomorphic to a conic. If it has a
rational point over Q, then it is isomorphic to PI, and thus has an infinite
number of rational points over Q.
The principle of the proof is the following. As the genus is zero, the
canonical divisor has degree -2. Changing the sign, one obtains a divisor of
degree 2. The correspondence between divisors and morphisms to projective
space provides a morphism X -+ P 2 whose image is of degree 2, hence is a
comc.
Around 1901, Poincare [P] took up this question again, apparently unaware of the work of Hurwitz and Hilbert. He gave another treatment of the
case 9 = o. When 9 = 1 (elliptic case), he made use of the affine structure
of the set of rational points. His paper is not very clear. Still, one can credit
him with descent arguments (3-descents, while his successors used mostly
2-descents), and with the following conjecture (proved by Mordell [M] in
1922):
Theorem. The group of rational points on an elliptic curve is finitely generated.
In the same paper Mordell stated his famous conjecture [now proved
by Faltings]:
Mordell's Conjecture. A curve of genus
tional points.
2 has only finitely many ra-
In his thesis (1928), Weil [WI] considered a curve of genus 9 ~ 1 over a

number field K, with jacobian J. Weil's theorem concerns the group J(K)
of points of J rational over K:
Theorem. (Mordell-Weil.) The group J(K) is finitely generated.

One of the main themes of this course is the proof of this theorem (with
J replaced by an arbitrary abelian variety).
Here is a brief sketch of the contents of each section:
1.1. Heights.
For simplicity, let us assume that K is Q. Let P n be projective n-space

and P E Pn(Q) a rational point of P n . We may write P uniquely (up to
sign) as
P=(Xo, ... ,Xn),

where Xo, ... , Xn are rational integers with no common factor. Define the
height of P to be
H(P) = sup IXil.
The number of points of P n (Q) of height ::; N is asymptotically
2n
(n+1)
N n +1
The logarithmic height of P is h(P) = log H(P) (which is approximately the number of digits required to write the coordinates of P). If
4> : X -+ P n is a morphism, we define on X(Q): H<p(x) = H(4)(x)) ,
h<p(x) = h(4)(x)). If X is a projective variety, one associates to 4>: X -+ P n
the divisor class c and .,p such
that c = c<p - c,p. We put he = h<p - h"" , which is defined up to 0(1). Then
there are properties ofthe type he+el = he+hel +0(1). If moreover X is an
abelian variety, there is the Neron normalisation of the height he: to every
c E Pic(X) is associated a well-defined function he. For simplicity, suppose
that c is symmetric, that is to say [-l]c = c, where [-1] is the morphism
x 1-+ -x on X. The Neron function is characterised by
-
he(nx)
= n 2-he(x),
for
x E X(Q)
and
n E Z.
(In fact, this holds over X(Q).) This function he is a quadratic form. There
is also a decomposition of he into local components.
3
1.2. The Mordell-Weil theorem and Mordell's conjecture.
The proof of the Mordell-Weil theorem divides into two steps.
The weak Mordell- Weil Theorem. Let A be an abelian variety defined
over Q, r = A( Q) the group of rational points on A, and n an integer ~ l.
Then r /nr is a finite group.
The proof goes as follows. Suppose that Y -+ X is a finite etale covering
of a projective variety X, where Y and X are defined over Q. Let m be
the degree of the covering. For each x E X(Q) we choose a lifting y E Y.
Let Ky be the field of rationality of y. Evidently, [Ky : Q] ~ m. Further,
one can show (Chevalley-Weil [CW]) that Ky is unramified outside a finite
set of prime numbers PI, ... , Ph, (depending only on X, Y and the covering
map). But, by a theorem of Hermite, there are up to isomorphism only
finitely many number fields of given discriminant and given degree. Thus
the fields Ky are finite in number, and there is a number field K for which
all rational points of X can be lifted to points of Y (K) (that is to say, the
image of Y (K) contains X (Q. One applies this to Y = X = A and to the
morphism A -+ A of multiplication by n. The finiteness of r / nr follows
easily.
The second part of the proof of the Mordell-Wei! theorem is a descent
argument using heights. Select an integer n ~ 2. Let Xi be representatives
of the classes of r modulo nr. One shows that there is an integer r such
that the points of r satisfying h( x) ~ r generate r: choosing r large, if
hex) > r one has that x = ny+xi with y E rand hey) < h(x)/2 ; the proof
then proceeds by induction.
As for Mordell's conjecture, it has been the subject of work of (a)
Chabauty [Ch], (b) Demjanenko [D], and Manin [Mal], (c) Mumford [Mull.
[and (d) Faltings [F]].
(a) Let X be an algebraic curve of genus ~ I and J its jacobian. Suppose
that the genus of X is strictly larger than the rank of J(Q). Then X(Q) is
finite (Chabauty [Ch]).
(b) Let Xo EX. Let A be an abelian variety over Q. Suppose that
rank Hom(X, A)
> rank A(Q),
where Hom(X, A) denotes the group of morphisms from X to A which send

Xo to the neutral element of A. Then Demjanenko and Manin showed that
X(Q) is finite. This result has applications to modular curves.
(c) Let X be a curve of genus ~ 2 over Q. Suppose that it has an infinite
number of rational points PI. P2 , ordered by increasing height. Then
11UIDfordproved,
[( d) Finally, Faltings proved 11ordell's conjecture: if X is a curve of genus

2:: 2 over Q, then X(Q) is finite.]
1.3. Integral points on algebraic curves. Siegel's theorem.
Let X be an affine curve over Q. and X its projective completion. The
algebra A of regular functions on X is of finite type over Qj let Xl, . ,Xn E
A be a set of generators. A subset S of X(Q) is quasi-integral if there is
.x =1= 0 such that Xi(S) E (1/.x)Z(1 ::; i ::; n) ; in other words, the Xi(S) have
bounded denominators. One checks that this condition is independent of
the choice of generators.
Theorem (Siegel [SiD. If 9 2:: 1, or if 9 = 0 and X - X has at least 3
elements, then all quasi-integral subsets are finite.
The hypotheses of this theorem are necessary: if 9 = 0, the case of just
one point at infinity corresponds to the affine line Ai (which has an infinite
number of Z-integral points), and the case where there are two points at
infinity corresponds to the Pell-Fermat equation X2 - Dy2 = 1.
Siegel's proof uses on the one hand, the weak 11ordell-Weil theorem,
and on the other hand a theorem on Diophantine approximation (ThueSiegel-Roth theorem).
The idea consists of approximating the slope of an asymptote by rational numbers. For example, if the curve X 3 - 3y3 = const. had an infinite
number of integral points, one would immediately obtain better approximations of 31 / 3 than are permitted by Roth's theorem. In the general case,
one uses a theorem of bad approximation on abelian varieties:
Theorem. Let A be an abelian variety defined over Q , and let Xl, X2, ..
be a sequence of distinct points of A(Q) converging to a point X E A(Q).
Then
lim log l/d(x, xn) = o.
n-+oo
h(xn)
The proof uses Roth's theorem combined with the 11ordell-Weil thearem.
The application to curves can be made in the following way:
5
Theorem. Let X be a curve of genus ~ 1 over Q, f a non-constant rational
function on X, and Xl, X2, an infinite number of distinct rational points.
Put f(Xi) = ai/bi , i = 1,2 ... , where, for i ~ 1 , ai and bi are coprime
integers. Then,
as
i-+oo
For the proof, one embeds X in its jacobian and applies the theorem
above.
1.4. Baker's Method.
For this method, the fundamental case is 9 = 0 with 3 points at infinity.

The affine algebra of PI - {O, 1,00} is generated by 4 elements X, y, z, t with
the relations
xy = 1, zt = 1, X + z = 1.
Therefore, associated to an integral point over K is a pair of units x, z of K
with X + z = 1. More generally, the search for quasi-integral points reduces
to the determination of pairs (x, z) of units of K satisfying some equation
Ax+Bz=C.
Expressing x, z in terms of a base of the group of units, one then uses
Baker's theorem on lower bounds for linear forms in logarithms and obtains
a finiteness result.
Other cases can be reduced to this by means of coverings; for example
curves of genus 1 or curves of genus 2 with a pair of points invariant by
the involution removed. Moreover the proof is effective (whereas Siegel's is
not).
1.5. Hilbert's Irreducibility Theorem. Sieves.
Let C be an irreducible algebraic curve over Q, and let C -+ PI be a
morphism of degree n ~ 2. Then "few" rational points of P I lift to rational
points of C, and indeed, "most" points give extensions of degree n.
Equivalently, if X is the variable of PI and F(T, X) = 0 is the equation
of C, of degree n in T and irreducible over Q(X), there are "many" X E Q
for which F(T, x) is irreducible over Q.
The rational points of PI with H(P) ~ N which lift to rational points
on C may be counted. Out of a total number rv cN2 rational points of
height ~ N, the number of those which lift is
~N2/n ~
N,
g=O
~ (logN)P, if g=1 (Neron)
~ loglogN, if g?2 (Mumford)
[~ 1, if g?2 (Faltings).]
if
Out of a total of eN integral points on the affine line Al with H :::; N,

the number of those which lift is
0(1) if g? 1, or if g = 0 with ? 3 points at infinity

O(log N) if g = 0 with 2 points at infinity
O( ~N)
if
=0
with 1 point at infinity
Hilbert's theorem can be stated for several variables: if X -+ P n IS a

morphism of degree? 2, "few" rational points on P n lift to X(Q). For
instance:
Theorem. The number of integral points of An of height:::; N which lift

to X(Q) is ~ Nn-I/2(log N)'Y, for some ,.
This result is due to S.D. Cohen [Co], with, = l.
Here is the principle of the proof. For simplicity, we consider the special
case Z2 = (XI, . . , xn) , where cP E Z[XI,"" xnJ has no square factors of
degree ? 1. One wants to bound the number of points of
As z2 - is absolutely irreducible modulo p, for almost all p, a theorem

of Lang-Weil shows that the number of points on the variety modulo p is
pn + 0(pn-l/2). As z and -z give the same point, this shows that the
reduction of SN modulo p has ~ (1/2 + E)pn number of points. Since we
have then excluded about half of the residue classes modulo p, we may apply
the large sieve. In this particular case, it gives:
as desired.
Hilbert's theorem may be applied to construct field extensions ofQ
with given Galois group, such as Sn or An [or sporadic simple groups].
7
2. HEIGHTS
References: [L2] and [L6].
2.1. The product formula.
Let K be a field and MK a family of absolute values x H lxl v (where
lxl v is a real number ~ 0) satisfying the following three properties:
1. All v E MK, except for a finite number, are ultrametric:
where v is a real valuation on K (always a discrete valuation in practice)

and c> 1. The others (finite in number) are archimedean absolute values
induced by embeddings K -4 C.
2. For all x E K*, one has lxl v = 1 for all except finitely many v.
We suppose given a family Av of numbers> 0 and we put IIxll v = Ixl;".
3. For all x E K*, one has (product formula):
We then say that K is equipped with a product formula.
Examples. a. K = Q. MQ = PU{oo} where P is the set of prime numbers,

Av = 1 for all v E MQ. For x E Q* , we write
and put
IIxli oo = Ixl =
II pJ,lp(x)(usual absolute value of x)

pEP
b. Let V be a projective normal k-variety, K its function field, Mv the set

of irreducible divisors W of V (i.e. irreducible subvarieties of co dimension
1). As V is normal, for all W E Mv the local ring of W is a discrete
8
valuation ring. We write Vw : K* -+ Z , where vw(l) is the order of
W,and
I/lw = c-vw(f),
(I E K*).
on
I E K* has only a finite number of zeros and poles, we have

vw(l) = 0 except for a finite number of WE Mv.
As each
Let C be a curve on V not intersecting the singularities of V (as V

is normal, its singularities have co dimension at least 2; one obtains such
a curve by taking, for example, a generic section by a linear variety or by
a hypersurface of sufficiently large degree). Let C.W be the intersection
multiplicity of C and W (this number is defined first when W does not
contain C, and one checks that it depends only on the class of W; this then
allows C.W to be defined when W :) C).
We assume that C. W > 0 lor all W E Mv. This is the case, for
example, when C is a section of V by a linear Variety of the right dimension.
We then put AW = C.W.
For I E K*, (I) is linearly equivalent to zero, thus C.(I) = 0 (if C is
not a pole of I , then C.(I) is the degree of the divisor on C given by the
restriction of I to C, that is to say the number of zeros minus the number
of poles). As
(I) = Lvw(l)W,
w
we get
L AWVW(l) = 0 for all I

w
K*.
We may thus obtain different product formulae for the same field K
(for example, take a quadric surface in 3-dimensional space with different
choices of C).
c. Finite field extensions.
Let K be a field with a product formula, and let L be a finite extension
of K of degree [L: K] = d. Each v E MK is known to extend to a finite
number ofabsolute values w of L; call ML the set of such extensions. Using
ML, a product formula for L is obtained in the following way:
Write Kv for the completion of Kat v. Then Lv = LK Kv is a semilocal algebra of degree dover Kv; hence it is a product of local algebras
where the residue field Lw of w is the completion of L at w. (If L/ K is

separable - which is the main case for applications - then w = L w.) Let
us write
For x E L, put:
(In the separable case this becomes simply
Then for all x E L * we have
(1)
II IIxll w = IINL/K(X)W/
wlv
If x E K*, we get
II IIxll
(2)
wlv
Formula (1) implies that

L.
nWEML
IIxli v
IIxll w =
1: we have a product formula for
Proof of (1).
Denote by Xw the image of x E L * in w. On taking the norm from

L K Kv to K v , one gets
NL/K(X) =
II NLw/K.,(xw).
wlv
Let Xw be the image of Xw by the projection w ~ Lw. We have a JordanHolder series of w (considered as a module over itself) of length dw/dw ,
with factor modules isomorphic to Lwo We deduce
whence (1). It is clear that (1)
'* (2).
10
Number fields.
Let K be a number field of degree dover Q. Let v be a place of K.
We have defined above the normalised absolute value IIxll v .
Another way to define it is the following: for x E K, the homothety
y 1-+ xy transforms the additive Haar measure on Kv to a multiple of itself
and the multiplier Ixlv,Haar is related to IIxllv by:
If v is associated to a real embedding K
-+
R, then
If v is associated to a complex embedding K
-+
C, then
We have:
if v is ultrametric,
if v is archimedean.
2.2. Heights on Pm(K).
Let K be a field with a product formula. By 2.1 example (c), every
subfield L of an algebraic closure K of K, and or" finite type over K, has a
product formula.
Let x E Pn(K), x = (xo, ... ,x n ), with Xi E K not all zero. We put
Almost all terms in the product are equal to 1. Further
= H(xo, ... , x n ),
for all A E K*j this defines H on Pn(K). Finally, we have H(x) ~ 1 for all
x E Pn(K), for if Xo =I 0, for example, then H(x) ~ TIv IIxoliv ~ 1.
11
We put hex) = 10gH(x). Then, hex)

K and x E Pn(L) we define
HL(X) =
II
O. If L is a finite extension of
s~p IIXiliw.
wEML
Then
HL(X) = HJ(x),
for
x E Pn(K).
Indeed, for each v E M K let iv be such that
Then, IIXi .. llw ~
IIxillw for all j
and all
wlv.
Therefore
II s~p Ilxillw = II IIXi .. llw = IIXi.. llv.

wlv
wlv
We have thus defined a height H : Pn(R") ~ R with values ~ 1, and

a logarithmic height h : P nCR") -+ R+ with values ~ O.
Examples
a. K=Q.
For x E P n(Q), we may write x = (xo, ... , xn) where Xi E Z have
greatest common divisor 1. For all primesp one has Vp(Xi) ~ 0, and Vp(Xi) =
o for some i. Thus, infivp(xi) = 0 and sUPi Ilxillp = 1. Therefore, H(x) =
sUPi IXil
b. Number jields.
For a number field, there is an analogous formula: Let K be a number
field, OK the ring 'of integers of K, and MK,oo the set of archimed~an
places of MK. To x = (xo, ... ,xn) E Pn(K), let us associate the ideal
g, = xoOJ( + ... + XnOK; we have:
H(x) = l/N(g,)l/d
II
s~p IIxillv.
vEMK,oo
c. Function jields.
Let K = key) be the function field of a projective normal k-variety V.
An element x = (xo, ... , xn) E P n{K) can be considered as a rational map
of V to P n
12
(the set .6. of points of V where x is not defined has co dimension ~ 2).
Thus x is a morphism of V - .6. to P n with co dim .6. ~ 2.
Let H be a general hyperplane of P n (not containing the image of x)
and let x*(H) be the inverse image of Hj x*(H) is a divisor on V -.6., thus
its Zariski closure is a divisor on V.
Let C be the curve chosen to define the product formula on K, cf. 2.1,
example b).
Formula.
hex) = C.x*(H)
It is enough to prove the formula in the particular case where Xi i= 0 for

all i. Denote by Hi the coordinate hyperplanes in P n, Di = x*(Hd, and
Di - Dj = (xi/Xj). For W an irreducible divisor on V, denote by VW(Di)
the coefficient of Win Di. We have
Taking j = 0:
As Di ~ 0, we have VW(Di) ~ OJ as the Hi have no common points,

infi VW(Di) = O. Therefore
-(W.C)vw(Do) = (W.C)i~VW(Xi) - (W.C)vw(xo).
l
But
Do
=L
and
C.Do =
W.vw(Do),
L (C.W)vw(Do).
w
On the other hand
L (WC)vw(xo) = C.(xo) = 0,
W
whence
-C.Do =
L (W.C)i~fvw(xi)
W
13
= -hex),
which proves the formula.
2.3. Properties of heights.
1. If x has coordinates Xi and y has coordinates Yj, let x Y be the point
with coordinates Xi.Yj. Then:
H(x y)
= H(x).H(y).
2. For x = (xo, ... , x n ) and m ~ 1, let x(m) be the point whose projective coordinates are all the monomials of degree m in the Xi, and xm (x~, ... , x~). Then:
3.
Formula for changing coordinates.
Let 4Yo, ... , 4Yr be homogeneous

polynomials of degree m in X o, ... , Xn with coefficients in K. If x (xo, ... ,x n ) E Pn(I<), and if the 4Yi(X) are not all zero, define 4Y(x) =
(4Yo(x), ... ,4Yr(x E Pr(K). Then:
H(4Y(x
IINllooH(c)H(x)m
where c and N are defined in the following way. Each 4Yi is a linear combination of monomials in X o, . .. , Xn of degree mj c is the sequence of coefficients
of all the 4Yi considered as a point in projective space; N is the maximum
number of monomials appearing in anyone of the 4Yij finally
IINlloo =
II
IINliv
vEMoo
(if K is a number field with the usual product formula,
IINiloo = N).
Corollary. For fixed 4Y, we have
H(4Y(x
= O((H(xm)
for
x E Pn(K).
In particular if 4Y E PGL n +1(K) is an automorphism of P n , we have
h(4Y(x
hex)
+ 0(1),
i.e. the difference is a bounded function on P n (K). If one neglects bounded

functions, the logarithmic height is independent of the chosen coordinates
on P n .
4. Let F(Xo , ... , Xn) be a homogeneous polynomial of degree m in which
the coefficient of x~ is non-zero, that is to say F(O, ... , 0,1) =1= 0. For
x = (xo, ... ,xn), let x' = (xo, ... ,xn-d.
14
Proposition. There is a constant C depending only on F such that if
x E Pn(K) satisfies F(x) = 0, then
hex) - C ~ hex') ~ hex).
(If F(x)
= 0 then x =1= (0, ... 0, 1) thus x' defines a point of Pn-1(K).)
Proof
Let x" be the point whose coordinates are all the monomials of degree
min Xo, ... ,Xn , with the exception of x~. As F(x) = 0, x~ is a linear
combination of other monomials, thus the map x" 1-+ x( m) is linear, and
h(x(m) ~ h(x") + 0(1).

The point x(m-l) x' has as coordinates the monomials in Xo, ... , Xn
of degree m containing at least one variable different from x n . Up to repetitions of the coordinates this is the point x", hence:
h(x") = h(X(m-l)
+ hex').
As h(x(m = mh(x) and h(x(m-l = (m - l)h(x), we have the result.

An explicit bound for the constant C can be given namely:
C ~ log IINlloo + h(c),
where N is the number of monomials occuring in F apart from
is the sequence of coefficients of F.
x~,
and c
Geometric interpretation.
The map x 1-+ x' defines a morphism of P n - {e} to P n-l, with e =
(0, ... ,0,1); this is the central projection from the point e.
More generally, if A is a linear subvariety of P n of dimension A, one
has (on selecting some coordinates) a projection tr: P n - A -+ P n - A- 1
Proposition. If X
A, then
h(tr(x
P n is a closed projective subvariety not intersecting
= hex) + 0(1),
where the 0(1) depends only on X.
for all x E X(K),
15
Proof.
Induction on A. The case ), = 0 is the previous proposition. It is only
necessary to check that if X is a closed sub-variety of P n, not meeting A,
then the image of X is closed in P n-A-l. This follows from the fact that a
projective variety is complete, thus its image by a morphism is closed.
We may also see this more concretely. It suffices to consider a projection
whose centre is a point (A = 0). Let F, Fa be the equations defining X, with
F(O, . .. ,0,1) i= O. One wishes to find the equations Gf3(xo, ... , xn-d = 0
of the projection, that is to say, to eliminate Xn from the equations Fa. Let
us write
F(x) = x~ + al(xO, ... , Xn_l)X~-l + ... ,
and we factorise it formally as:

m
F(x)
= II(x n -
(i(XO, ... , xn-d)
i=l
We introduce indeterminates ta and put:

m
II L taFa(XO, ... , Xn
i=l
-!,
(i) =
L taGa(xo, . .. ,Xn-l).
a
One checks that the G a give the equations of the projection. (If there is a
single Fa, the resultant of F and Fa is then obtained.)
Proposition. Let 4>0, ... , 4>r be homogeneous polynomials of degree m
which are not simultaneously zero on K. Then
h(4)(x))
= mh(x) + 0(1),
for x E P n(K).
Proof
Let i : P n - t PM be defined by i(x) = x(m), where M +1 is the number
of monomials of degree m. There is a linear projection, defined outside the
closed subvariety of its centre .6., 7r : PM - .6. - t P r such that 7r 0 i = 4>.
i
--t
P M -.6.
Then i(Pn) n.6. = 0. Thus on i(P n ) we have h(7r(Y

y = i(x) = x(m) the result follows.
= hey) + 0(1).
As
16
Exercise. (Alternative proof.) Use Hilbert's Nullstellensatz to show that
there is an integer A for which x A is the image of x(A-m) 0cjJ(x) by a linear
transformation. Hence
mh(x)
S h(cjJ(x + 0(1),
and the reverse inequality is obvious.

2.4. Northcott's finiteness theorem.
Theorem ([NoD. Let n, d, X be integers ~ 1. There are only finitely many
points ofPn(Q) of height S X and of degree S d.
For x = (xo, ... , x n ) E Pn(Q), the degree of x is the degree of the field
i,j S n,xj #- O.
generated by the xi/Xj,
Corollary. If K is a. finite extension of Q, there are only finitely many
points ofPn(K) of height S X.
Proof of the theorem.

a). d = 1. The points of Pn(Q) of bounded height are finite in number.
Indeed, let us write x E Pn(Q) as x = (xo, ... ,xn ), Xj E Z with greatest
common divisor 1. Then H(x) = sup lXii, whence the finiteness.
b). We reduce the case d ~ 1 to the case d = 1.
Let x = (xo, ... , xn) E Pn(K), with [K : Q] = d, Xi E K. Let
(J'o : K -+ Q , (1 sa S d) be the embeddings of Kin Q , and XO = (J'oX =
(x
x~). Thus (xl, ... , x d ) E P n X x P n(Q). The symmetric group
Sd of degree d acts on P n X . X P n . Let En,d = (P n X x Pn)/Sd ,
and let 7r : P n X X P n -+ En,d be the natural morphism. The image of
!f.. = (Xl, ... , xd) by 7r is invariant under conjugation, thus 7r(!f..) E En,d(Q).
There is a natural immersion : En,d -+ P N (consider the homogeneous
x~ for each a, acted on by the symmetric group, and
polynomials in
take invariant polynomials). Then
o,... ,
xo, ... ,
H(7r(!f..
s CIT H(xO)d
S CH(x)D, for some do,C,D.
Using a), this shows that the 7r(!f..) are finite in number. Each fibre has at
most d! elements, whence the finiteness of the x's.
Exercise. Derive part (b) of this proof by introducing variables To, ... , Tn,
and considering the polynomial
d
II .LTi(J'o(xi) = .L ca(x)Ta.
0=1 i=O
17
Then
Ca
(x) E Q and the point c( x) with coordinates Ca (x) is such that:

H(c(x :5 CH(x)D.
Conclude as above.
2.5. Quantitative form of Northcott '8 theorem.
Theorem (Schanuel [Sch]). Let Nx be the number of x E Pn-I(K) with

H(x) :5 X. Then
h
nd {O(XIogX),
Nx = (K(n) cnX + o(xnd-l),
where
cn =(
2rl (21!y2
..jl:l.
if d = 1, n = 2,
otherwise.
n R r
) wn .
(As usual, rl is the number of real places, r2 the number of complex places,
r = rl + r2 - 1 the rank of the group of units, d = [K : Q], w the number of roots of unity in K, R the regulator, l:l. the absolute value of the
discriminant, and h is the class number).
Remark. The error term is probably not best possible for (n, d)
:f:. (2, 1).
Sketch of proof.
Let x = (xo, ... ,Xn-l) E Pn-I(K). Let ~x be the ideal generated by
the Xi. For ~ an ideal of K, let Nx(~) be the number of points of height
:5 X with class(~) = class(~). Schanuel proves the more precise estimate:
Nx(~)
Recall that if ~ =
~,
= 1/(K(n)Cn X nd +O(-).
we have
H(x) = N(~)-l/d
II
sup Ilxiliv = N(~)-l/d Hoo(x).
vEMK,oo
We are thus led to study the n-tuples (x!, . .. ,xn ), with

that a) Hoo(x) :5 XN(~)l/d, and b) the Xi generate~.
We define an equivalence relation on ~ x ... x ~ = ~n :
(Xl, ... , Xn) rv (YI, ... , Yn) if there is a unit
Xi
E ~, such
of K with Yi = Xi for all i.
Let Nx(~) be the number of (Xl, ... ,Xn ) E ~n, modulo this equivalence
relation, with Hoo(x) :5 XN(~)l/d.
18
Theorem. Nx(g)
= cnxnd + OK,n(X nd - l ).
Since N (g) depends only on the class of g, which takes only finitely
many values, the O-constant can be chosen to be independent of g.
a. Computation of Nx(g).
Let KR = K R (product of the archimedian completions of K). In
K R X .. x K R (n times) we count the points of the lattice g x ... x g. For
this we take a fundamental domain E of K R modulo the units. For each
v E MK,oo let i(v) be such that IIxi(v)riv = sUPi IIxdlv. We then have a
point (Xi(v) E IT Kv' Define a domain Dx of Kit by the 3 conditions:
(1) (Xi(v) E E,
(2) Hoo(x) :5 XN(g)I/d,
(3) IIXjllv :5 IIxi(v)lIv for all j and all v.
It remains to count the points of the lattice in Dx.
Notice that
Hoo(>'x) = >"Hoo(x), and Dx = XD I . It is easily shown that the frontier

F = aDI of the compact set DI is contained in a finite number of images of
[0, l]nd-1 by morphisms 4>i of class C l and therefore is (nd-l)-dimensional,
in Minkowski's sense.
Given a compact subset Dl of RN whose frontier has the above property, the number of points of ZN which belongs to XD I is vol(Dt)XN +
O(XN-I). [The idea of the proof essentially derives from Gauss, and consists of counting the number of translates of a fundamental domain for the
lattice which intersect X Dl and then showing that the number of translates which intersect the boundary XF is o(xnd-I). This last estimate is
derived by dissecting [0, l]nd-l into [Xl/nd]nd-l small cubes whose images
under the maps 4>i have diameter bounded independently of X, by the C l
property.] The estimate for N (g) follows.
b. Computation of Nx(g).
Let (Xl, ... , Xn) E g n with H 00 (x) :5 X N (g) 1/d. The ideal generated
by Xl, ... ,Xn is of the form gb with b integral. Hence
Nx(g) = ~NX(Nb)-l/d(gb).
b
Using the Mobius formula, we get
Nx(g)
= L J.t(b)N~(Nb)-l/d(gb),
b
'" ~J.t(b)Cnxnd(Nb)-n
b
19
'" CnXndLJ.L(b)(Nb)-n.
b
Finally, Schanuel's theorem follows from

(K1(s)
= ~ J.L(b)(Nb)-s.
The function field analogue.

Let C be an absolutely irreducible projective curve over F q of genus 9
and let K be its function field. If x = (xo, ... ,xn-d E Pn-1(K), we have
hex) = degree(x*H) ,where x*H is the inverse image by x: C ---+ P n - 1 of
a hyperplane H of P n-l not containing the image of C. One can prove:
Theorem. The number ofx E Pn-1(K) with hex) = d is

A qn(d+1-g)
-:-(q-=--l-:-)(-K-:-(n-:-)
+ 0 (q(n-l)d) ,
where A is the number of F q-rational points on the jacobian of C (=

number of divisor classes of degree 0 of K).
2.6. Height associated to a morphism : X ---+ P n'
Let K be a field equipped with iCl. product formula, K an algebraic
closure of K , and X an algebraic variety over K. Let : X ---+ P n be a
morphism. For x E X (K), we put
H(x) = H((x)),
h(x)
= logH(x) = h((x)).
Theorem. Let : X ---+ P n be an isomorphism of X onto a locally closed

subvariety of P n' Let 'IjJ : X ---+ Pm be any morphism. There are real
numbers Cl > 0 and C2 such that
In other words, h"" ::::; 0(1
+ h).
Proof.
We view X as embedded in P n (by ) with coordinates Xo, ... , x n .
20
a) Special case: there are homogeneous polynomials 'l/Jo, ... , 'l/Jm of the
same degree, not simultaneously zero on X, such that
'I/J(x) = ('l/Jo(x), ... , 'l/Jn(x)).

Let d be the common degree of the 'l/Ji. Then
b) General case. We cover X by a finite number of open subvarieties Xi

such that the pairs (Xi, 'l/Jlx.) are special.
By the quasi-compactness of algebraic varieties it suffices to do this
locally. Let x EX. We write
'I/J(x)
= (AO,""
An),
with, let's say, AO = 1. Then in a neighbourhood of x
'I/J
where the
= (l,h,,fm),
Ii belong to the local ring of x :
where the 'l/Ji are homogeneous of the same degree and 'l/Jo(x) = 1. So
'I/J = ('l/Jo, ... , 'l/Jm) in a neighbourhood of x, and the result follows from a).
Remark. If 'I/J is also an immersion, we have
for some constants
Ci,
hence h", and h.p have roughly the same size.
2.7. The group Pic(X) (reference: [H]).

Let X be a variety over a field K. One defines Pic(X) to be the group
of line bundles (= locally free sheaves of Ox-modules of rank 1) on X, the
multiplication being the tensor product; one has Pic(X) ~ Hl(X, Ox) ([H],
Ch.III, Ex.4.5). If f : X -+ Y is a morphism and C is a line bundle on Y,
then C , which is a line bundle on X, denotes the inverse image of c by
f. This defines a homomorphism
f* : Pic(Y) -+ Pic(X).
21
Suppose now that X is a normal variety (in fact, normality is unnecessary here, [G, Ch.IV, 21.6]). A (Weil) divisor on X is a linear combination,
with integral coefficients, of irreducible sub-varieties of co dimension 1:
D=2:'.:: n w. W .
The support Supp D of D is Unw~O W.

If f is a rational function on X, its divisor is the divisor of zeros minus
the divisor of poles:
(J)
= (J)o -
(J)oo
= 2:'.:: vw(J).W.
w
The quotient of the group of divisors by the subgroup of divisors of

functions is the group Cl(X) of divisor classes. One says that the divisor D
is linearly equivalent to zero, written D rv 0, if it is the divisor of a rational
function, that is, D is in the zero class.
A line bundle on X has rational sections; if S is such a section one
defines its divisor (s) = (s)o - (s)oo, and the class of this divisor is independent of the section chosen. For a normal variety, one then obtains an
injection Pic(X) -+ CI(X).
In the non-singular case (locally factorial suffices, [H, Ch. II, Corollary
6.16]), this gives an isomorphism
Pic(X)
Cl(X).
On P n we have a natural line bundle, written Cl, which corresponds to

the class of a hyperplane. For a morphism 4> : X -+ P n, the inverse image
4>*Cl = c</> defines an element of Pic(X) (if H is a hyperplane not containing
4>(X), X being irreducible and smooth, c</> is the class of 4>* H). Hence one
has a map 4> I-t c</>.
If for any x E X there is a global section of the line bundle E on
X which does not vanish at x, then E is said to be is generated by its
global sections. Let X be projective and E a line bundle on X. Then
rex, E) is finite dimensional. Furthermore, if E is generated by its global
sections, a morphism 4> from X to P n is obtained in the following way.
Let so, ... , Sn be a basis of rex, E). The morphism 4> : X -+ P n, where
4>(x) = (so(x), ... , sn{x for all x E X, is well defined, and we have c'" = E.
Suppose that X is non-singular and projective. A necesssary and sufficient condition for C E Pic(X) to be generated by its global sections is that
22
if the divisor D has class c, then the complete linear system IDI has no base
points: that is to say, for all x EX, there is D' linearly equivalent to D
with D' ~ 0 and x SuppeD').
An element c E Pic(X), which is generated by its global sections, is
very ample if the corresponding morphism ifJ : X -T P n is an immersion.
We say that c is ample if there is an integer m > 0 such that mc is very
ample, in which case mc is very ample for all sufficiently large m.
2.8. Heights and line bundles
Let K be a field with a product formula, V a projective algebraic
variety over K, and Pic(V) the group of classes of line bundles on V.
Let H be the quotient of the vector space of real-valued functions on
V (K) by the space of bounded functions on V (K). It will be convenient to
choose representatives of elements of H and to write that formulae are true
modulo 0(1), that is to say up to the addition of a bounded function.
Theorem. There is a unique map c 1-+ he of Pic(V) to H such that,
1 ) h e+e, = he + h~ + 0(1) for all c,c' E Pic(V).
2 ) If c corresponds to a morphism ifJ : V -T P n, then he = h</> + 0(1).
We recall that the correspondence between ifJ : V -T P n and c</> E

Pic(V) is given by ifJ*Cl = c</> where Cl is the standard line bundle on P n'
On the other hand, h</>(x) = h(ifJ(x where h is the logarithmic height on
Pn .
Proof
Uniqueness. It suffices to prove the following result: every c E Pic(V) is
of the form c</> - c.p where ifJ, 'I/J are immersions.
Let ifJl : V -T P n be an immersion, and let C</>l = ifJi Cl.
a. If m is sufficiently large, c' = c+mc</>l is generated by its global sections.
This is a general result on sheaves twisted by 0(1).
b. As c' is generated by its global sections, it corresponds to a morphism
V -T Pm. We embed Pm X P n in a P N by the tensor product (Segre
embedding). Then, c' + C</>l corresponds to an immersion V -T Pm X P n -T
P N, thus c' + C</>l is very ample.
c. We write c</> = C + (m + l)c</>p c.p = (m + l)c</>p where ifJ and 'I/J are
immersions. Then c = c</> - c.p.
Existence.
a. We check that if C</>l = C</>2' then h</>l = h</>2 + 0(1). We consider for this
the complete linear system, i.e. the vector space r = HO (V, E) of all the
global sections of the line bundle E associated to C</>l' The ifJi correspond
23
to bases of sub-vector spaces of r. We have already seen that a change of
basis or of subspace does not change ht/>.
b. We write C E Pic(V) in the form C = ct/> - c.p, and we would be able
to put he = ht/> - h.p if we could verify that C = ct/>, - 4' => ht/>, - h.p' =
ht/> - h.p + 0(1). But as Ct/> + C.p' = Ct/>, + C.p one has ht/>.p' = ht/>'.p + 0(1),
from a), thus ht/> + h.p' = ht/>' + h.p + 0(1).
Exercise. Give another proof of the existence using the following lemma of
Mumford: Pic(V) is presented by its generators ct/>, for morphisms = Cit/>.
2) ct/>.p = ct/> + c.p.
Functoriality. Let f
Then
:V
-+
V', let d E Pic(V'), and let C = f*c' E Pic(V).
he = he'
f + 0(1).
This is immediate by writing, for example, d = Ct/>, - c.p'.

This property can be applied to a product variety VI X V2 where VI
and V2 are non-empty. By considering the projections VI x V2 -+ Vi, there
is a natural inclusion
(In general this is not an equality, there is an extra factor, as we shall see
later (3.2).) Let CI E Pic(Vt), C2 E Pic(V2), and C = CI +C2 E Pic(Vi X V2)'
For Xi E Vi(K), (i = 1,2), we have
Examples.
1). To any morphism f : P n X Pm -+ P N we associate a pair of integers
(db d2 ) (corresponding to the degrees of f expressed in terms of multihomogeneous polynomials). Here there is no extra factor: Pic(P N) = Z,
Pic(P n x Pm) = Z x Z, and we get
2). Let E be an elliptic curve with origin 0, m an integer ~ 1, Cl = m(O).

(For m ~ 3, Cl is a very ample divisor. For m = 3, it gives the standard
plane cubic immersion; for m = 4, it gives an intersection of two quadrics
in P 3 .)
24
We cannot apply the preceding formula to the composition law S
Ex E --+ E, for S*Cl does not belong to PieCE) EB Pic(E). But one can take
the morphism Ex E --+ E x E, (x, y) I-t (x + y, x - V). The inverse image
of Cl 1 + 1 Cl is 2(Cl 1 + 1 Cl), as one can see, either geometrically or
by analysis, from formulae for p(u + v).p(u - v) , or for the 4 Jacobi theta
functions: Oi(U+V).Oi(U-V) is an homogeneous polynomial of degree (2,2)
in the OJ(u),Oj(v) , i,j E {1,2,3,4} (see e.g. [Wh] p.456, exerc. 8, and p.
488, exerc. 3). For the associated height on E x E we deduce formulae of
the type (cf. 3.3, 3.6)
hex + y,x - y) = 2h(x)
+ 2h(y) + 0(1).
3). Let d be a positive integer, [d] : E --+ E the morphism x

a multiple of (0). Then, [d]*c = d2 c. Whence
h(dx)
2.9. he = 0(1)
{:>
I-t
dx , and C
= d2 h(x) + 0(1).
C is of finite order (number fields).
The function he has real values (positive, negative, or zero). If C is of

finite order in Pic(V), we have he = 0(1). Conversely:
Theorem. If V is non-singular and projective and K is a number field,
then he = 0(1) implies that cis offinite order in Pic(V).
The proof will be given in 3.1l, as an application of Neron's quadraticity theorem.
2.10. Positivity of the height.
We shall show that if c is ample, then he is bounded below (that is, he
is positive up to 0(1. In the non-singular case, if c E Pic(V) corresponds
to a positive divisor ~, then he ~ 0 (up to 0(1 on V - Supp(~).
The general statement is the following:
Theorem. Let c E Pic(V), E a line bundle associated to c, and s a global
section of E. Then, he ~ 0 (up to 0(1) outside the variety of zeros of s.
Proof.
There are two immersions 1J' and 1J" such that c = C</>" - c</>', thus
there are two very ample line bundles E' and E" such that E E' ~ E".
We must show that he" ~ he' + 0(1) outside the zeros of the section s
(c, = c</>', c" = c</>" )
25
Let So, ... , Sn be a basis of the global sections of E'
we have
he' (X) = h(so(x), ... , sn(X.

Complete the set sSo, .. . , ssn of global sections of E" to a basis
(sso, . .. ,SSn, (JI, .. ,) of
E"). Then
rev,
he' (x) = h(sso(x), ... , sSn(x), (Jl(X), .. .).

At a point where sex) =f:. 0, we have
he,,(x) 2: h(sso(x), ... , sSn(x),

and by homogeneity
he' (x) = h(sso(x), ... , sSn(x,

whence the result.
If c corresponds to a positive divisor b., then he is positive modulo 0(1)

outside the base points and base components of the linear system 1b.1, that
is to say outside of the intersection of the b.' 2: 0, b.' '" b.. In particular, if
c is generated by its global sections then he 2: 0 everywhere (this was clear
a priori : he = hef> + 0(1).
2.11. Divisors algebraically equivalent to zero.
Let V be a non-singular projective variety, C an irreducible curve, and
b. a divisor on C x V without a vertical component. For A E C let
VA = b. n ({..\} x V).
In terms of divisor classes, VA is the class of the inverse image of b. by
the map V - t ex v,x 1-+ (A,X). We then obtain a I-parameter family
of divisors. A divisor is said to be algebraically equivalent to zero if its
class in Pic(V) belongs to the subgroup PicO(V) generated by the divisors
VA - V#,, A, J,L E C (for all possible choices of C, b., A, J,L).
The simplest case is that where V is a non-singular projective curve:
Pic(V) is just the group of divisors of degree O. The ample divisors are
those of degree> O.
26
Theorem. Let c and Cl be two elements of Pic(V), with

Cl ample. Then for all >
we have
E pica (V) and
If Cl and C2 are ample, then mC2 - Cl is ample for sufficiently large m,

and we deduce that hc! ~ hC2 + 0(1). Thus if the formula is true for one
choice of Cl it is true for all Cl. We use the following result.
Theorem. If C E PicO(V) and if Cl is ample, then
+ Cl is ample.
In other words, ampleness descends to the Neron-Severi group of V:

NS(V) = Pic(V)/Pic(V).
This can be proved using the Nakai-Moishezon criterion (see [H, Ch. V,
Th. 1.10], for the case of surfaces, and [K], for the general case): a positive
divisor D is ample if and only if for all irreducible sub-varieties .6. one has
D6.6. >
where 6 is the dimension of 6. (in the case of a surface, these
conditions become D2 > 0, and D.C > for all irreducible curves C on the
surface).
We now prove the theorem on Ihcl. As C E Pic(V), we have nc E
Pic(V) for all n E Z ; therefore nc + Cl is ample and nhc + hc! is positive
modulo 0(1). One bounds hc from above by choosing n < -1/ , and one
bounds -hc from above by choosing n > 1/.
We have thus shown hc(x) = o(hc! (x, as hc! (x) -+ 00. Neron's theory
gives a more precise statement:
Theorem (Neron). Ihc(x)1 ~ 0(h~f2 + 1).

The proof will be given in 3.1O.
2.12. Example-exercise: projective plane blown up at a point.

We consider the surface V obtained by blowing up the point 0=(0,0,1)
of the projective plane P 2. That is to say, V C P2 X PI is the closure of
the graph of
P 2 - {(O,O, I)} -+ PI
(x,y,z)
I-t
(x,y).
Taking homogeneous coordinates (x, y, z) on P 2 and (u, v) on P bone

may define V in P 2 X P 1 by the equation
xv - yu
= 0.
27
Let 1r : V -+ P2 be the surjection induced from the projection P 2 x

P 1 -+ P 2. Then 1r is an isomorphism except at the inverse image of the
point 0: we have replaced 0 by the slopes x/y of the lines through O.
The group Pic(V) is Z x Z with basis L, D defined as follows:
L is the blow up of 0, i.e. 1r-l(O).
D is the class of the inverse image in V of a line of P 2 not going through
O.
If Li is a line of P 2 through 0, its inverse image by 1r is L +~, where

~ is the strict transform of Li ([H), Ch5, 3). The class of ~ is thus equal
to D-L.
One checks (e.g. by the Nakai-Moishezon criterion [H, Ch.5, Theorem
1.10)) that the ample divisor classes are the linear combinations,
c = aD + beD - L)
I'V
aD + b~, with a, b > 0.
Intersection numbers are readily computed from the formulae ([H],

Ch.5, Prop. 3.2):
D2 = 1,D.L = 0,L 2 =-1.
Let c = aD + beD - L) be an ample divisor, and let he be the corresponding logarithmic height on V(K). Let He = exp he. Since c is ample,
we know that the number of rational points x E V(Q) with He(x) $ X is
finite for every X. The point of the exercise is to determine the growth of
this number as X -+ 00. The result is:
The number N(X) ofx E V(Q) such that He(x) $ X is:::::: X"Y, where
"I = max(2/a, 2/b).
(The symbol :::::: means that, for large enough X, the ratios of the two
numbers lies between Cl and C2 with Ci positive. A more precise result
cannot be expected since he is defined only up to the addition of a bounded
function.)
(a) Show that the number of x E L(Q) with He(x) $ X is :::::: X2/b.
[Note that c.L = b, then apply Schanuel's theorem to L.)
(b) Show that the number of x E ~(Q) with He(x) $ X is::::::
that N(X) > X"Y.
x 2 /a.
(c) Show that N(X) ~ the number of (x, y, z) E P 2(Q), (x, y, z)

with
Deduce
i- (0,0,1)
28
We normalise, as usual, a point (x,y,z) E P2(Q) so that X,y,z E Z
and gcd( x, y, z) = 1. We may then select the heights in (c) to be
H(x, y, z) = sup(lxl, Iyl, Izl),

H(x, y) = sup( I~I,
I~I),
where d = gcd(x, y).
(d) Let C C V be the strict transform by 1r of the reducible curve xyz = 0

in P 2 Show that there are <t:: X"Y points P E C(Q) with He(P) ::; X.
[Use Schanuel's theorem.]
(e) Show that if P E (V - L - C)(Q) with He(P) < X then H(x,y) (= Y,
say) ::; X1/(a+b), Izi ::; x1/ay-b/a and d ::; xl/ay-l-b/a.
Deduce that the number of P E (V - L - C)(Q) with He(P) ::; X is
I:
y-2b/a.
(f) Show that the sum in (e) has the upper bound
X2/a
""'
L....J
y-2b/a <t:: {X2/a
Y$Xl/(Cl+b)
+ X 3 /(a+b) ,
x 2 /a log X ,
a =/:. 2b} <t:: X"'f

a = 2b
'
and deduce the result stated above.

The interest of this exercise is to give an example where one can evaluate the rate of growth of the number of points on V with He ::; X and
see how it varies with c. It might be interesting to do the same for other
varieties, for instance cubic surfaces in P 3 (cf. [MFT]).
Notice that for all c, we found essentially powers of X. For abelian
varieties Neron's theory (cf. Chapter 3 and 4.5) gives instead powers of
log X. Problem: can one find a growth which is intermediate between
powers of X and powers of log X? Nothing is known on such questions.
29
3. NORMALISED HEIGHTS
3.1. Neron-Tate normalisation.
Lemma. (Tate). Let S be a set and 7r : S --+ S a map. Let f be a realvalued function on S such that f 0 7r = >.f + 0 (1) , with >. > 1. Then there
is a unique function 1 on S such that
(a) l=f+O(l)
(b)
10 7r = >.1
and we have
j(x) = lim (l/>.n)f(7r nx),for every xES.

n-+oo
We shall show (a) in the following more precise form: if c is such that
If(7rx) - >.f(x)1 ~ c , then
Ij(x) - f(x)1 ~ c/(>. - 1).

Proof.
Let c> 0 be such that If(7rx) - >.f(x)1 ~ c. We have
so the condition>. > 1 shows that the sequence (l/>.n)f(7r n x) is convergent.

Let l(x) be its limit. Property (b) is clear, and
I: c/>.n = c/(>. - 1).

00
Ij(x) - f(x)1 ~
n=l
Observe that if f is bounded then
f=9
Uniqueness. If 9 = f
1 = O.
Thus if f
= 9 + 0(1), then
+ 0 (1) with 9 0 7r = >.g , we have from the above that
9 = 1, and 9 = g, thus 9 = j.
Exercise. (Alternative proof) Let E be the Banach space of bounded realvalued functions on S equipped with the sup norm. Show that the operator
A: E --+ E, f 1-+ f 0 7r is such that IIAII ~ 1 and hence that 1- (l/>.)A is
invertible. If 4> satisfies
(1- (l/>.)A)4> = - f
+ (l/>.)f 0 7r,
30
show that} = /
+ > has the properties (a) and
(b) of Tate's lemma.
Functoriality of /.
Let (7 : 8 -+ 8 ' be a map, and let 'If : 8 -+ 8, 'If' : 8' -+ 8' be such that
'If' 0 (7 = (70 'If. Let f' : 8' -+ R have /' 0 'If' = A/' +0(1), and let / = /' 0 (7.
1~
8
8'
,.1'
R
Then /0 'If = A/ + 0(1), and }

Proof: by the uniqueness.
= l' 0
(7.
Commutativity property.
Let 8, 'If, /, A be as in Tate's lemma (in particular, A > 1). Let 'lfa be
endomorphisms of 8 commuting with 'If, and Aa real numbers, such that
Then
} 0
'lfa
= AaT
In particular, if 'Ifa are commuting endomorphisms of a set 8, and f is

a real-valued function on 8 such that /0 'lfa = Aa/ + Oa(l) with Aa E R,
where one of the Aa'S is> 1, then there is a unique} such that }O'lfa = Aa}
for all c.
Proof: this follows from the functoriality applied to
1~
8
1~
~
8.
Normalised heights (Neron [N4], Tate).

Let K be a field equipped with a product formula, X a projective
variety over K, 8 = X(K), 'If a morphism X -+ X, and c E Pic(X).
Suppose that 'If*C = AC with A > 1. By 2.8, we have he ('If x) =
Ahe(x)+O(l) on X(K). Hence by Tate's lemma, there is a unique function
31
he such that he = he + 0(1) and he(-7rx) = Ahe(x). This is the normalised
logarithmic height.
Example. Projective space P n'

Let d E Z and 1rd(XO, , xn) = (xg, ... , x!). Let c be the natural generator of Pic(P n ) = Z (corresponding to the sheaf 0(1) and to the usual
height on projective space). Then 1rdC = dc. We recover the standard height
since
3.2. Abelian varieties.

For any abelian variety A, one has an exact sequence
o -+ Pic(A) -+ Pic(A) -+ NS(A) -+ 0,

where PicO(A) is the Picard variety of A (divisor classes algebraically equivalent to zero) and NS( A) (Neron -Severi group) is a finitely generated group.
[Over C, we may see this by transcendental methods: we take the cohomology of the exact sequence of sheaves
21fi
exp
O---+Z---+Ohol---+hol---+ O
The HO terms provide the exact sequence of groups 0 -+ Z -+ C -+
C* -+ 0, because global holomorphic functions on connected projective
varieties are constant. We have then the exact sequence
0-+ H 1(A, Z)
-+
H1(A, Ohol)
-+
H1(A, 0hol)
-+
H2(A, Z)
-+ ...
This gives an injection of H1(A, Ohol)/H1(A, Z) in Hl(A, 0bol) where

the cokernel (which may be identified with NS(A is a subgroup of
H2(A, Z). Further
Hl(A, 0hod ~ Pic(A),
H1(A,Ohol) is a C-vector space of dimension dimA, and H 1 (A, Z) is a
lattice of rank 2 dim A. We thus have an analytic description of the Picard
variety as a quotient
For d E Z, we write 1rd = [dJ = d for the endomorphism x 1-+ dx. In

particular, [-lJ is the natural symmetry of the variety, inducing an involution on Pic(A) : c 1-+ C' = [-lJ*c. If c is represented by a divisor D, c' is
32
represented by the image of D under x 1-+ -x. We say that c is symmetric
if c = d, antisymmetric if c = -c/o The symmetry [-1] acts trivially on the
Neron-Severi group NS(A) and induces the natural symmetry c 1-+ -c on
PicO(A).
We need the 'theorem of the cube':

Theorem. Let X, Y, Z be three projective varieties; any element
c E Pic(X x Y x Z) which induces 0 on X x Y, X x Z, X x Z is zero.
For the proof, see [Mu], 6 and 1O.
Consider now the following morphisms from A x A x A to A:
8123(X, y, z)
= x + y + z,
= x + y, 823(X, y, z) = y + z, 813(X, y, z) = x + z,
81(X,y,Z) = x, 82(X,y,Z) = y, 83(X,y,Z) = z.
812(X, y, z)
Theorem. Let c E Pic(A). In Pic(A x A x A) we have
Proof
Let
By symmetry and the theorem of the cube, it suffices to show that u is zero
on 0 x A x A. As 8123(0, y, z) = y + z = 823(0, y, z), on 0 x A x A we have
8123 = 823, and also 812 = 82,813 = 83, 81 = O. Whence the theorem.
Functions of degree ~ 2 between abelian groups.
If G 1 and G 2 are two abelian groups and F : G 1 --+- G 2 is a map such
that F(O) = 0, one says that
a) F is of degree ~ 1 if F(x + y) - F(x) - F(y) = O.
b) F is of degree ~ 2 if FI(x, y) = F(x+y) - F(x) - F(y) is additive in x
and y, or equivalently, if the second difference F2(x, y, z) = F(x + y +
z) - F(x + y) - F(x + z) - F(y + z) + F(x) + F(y) + F(z) is zero.
Theorem. Let X be an algebraic variety, and A an abelian variety. Let

c E Pic(A). Then the map of abelian groups,
Hom(X,A)
--+-
Pic(X),
1-+
j*c,
33
has degree
2.
Proof.
We deduce this from the preceding theorem by a simple functoriality
argument. Let, Ii : X -+ A, i = 1,2,3, be three morphisms, and
f = (h,h,fa): X
-+
A x A x A.
Take as before
We have
j*(J
= (h + 12+ fa)*c-(h + h)*c-(h + fa)*c-(h+fa)*c+ gc+ fic+ f;c.
Since (J
= 0, we have r = OJ hence the theorem.

(J
This implies the following formula for [d]*c, where d =
[-l]*c:
Corollary (Weil , [Mu2]). If dE Z and c E Pic(A), we have

[d] *c= d(d 2+ 1) c+ d(d 2- 1) c.I
In particular:
[d]* = { ,pc, if c is symmetric;

c
dc, if c is antisymmetric.
Proof
The map Z -+ Pic(A) , d ...... [d]*c, has degree ~ 2 by the previous
theorem, therefore the result follows from the next lemma.
Lemma. Let G be an abelian group, and F : Z

~ 2 with F(O) = o. Then
-+
G a function of degree
F(n) = n(n 2+ 1) F(l) + n(n 2- 1) F(-l).

Proof of lemma.
Taking the difference between the two sides, we reduce to the case
where F(l) = F( -1) = F(O) = o. As
F1 ( -1,1) = F(O) - F( -1) - F(1) = 0,
34
and Fl is biadditive, we have Fl

therefore F = O.
= 0, therefore F
is additive. But F(l)
= 0,
Exercises: quadratic functors on projective varieties.

Let X -+ F(X) be a contravariant functor from the category C of
pointed (equipped with a base point) smooth connected projective varieties
over K to the category of abelian groups. We assume that if P is a point,
then F(P) = O.
1) Let X, Y, Z be three varieties in C. Let us identify X, Y, Z with subvarieties of the products X x Y, X x Y X Z, by means of the base points. Show
that we have the decompositions
F(X x Y) = F(X) EB F(Y) EB F2(X, Y),
F(XxYxZ) =
F(X) EB F(Y) EB F(Z) EB F2(X, Y) EB F2(X, Z) EB F2(Y, Z) EB F3(X, Y, Z),
where F2(X, Y), F3(X, Y, X) are functors in X, Y and X, Y, Z, respectively.

(We say that F has degree ~ 1, or is additive if F2 = 0; F has degree ~ 2
if F3 = 0, etc.)
2) Show that the contravariant functors Hl(X) (the usual complex cohomology over Z), Hl(X, Ox), PicD(X) are additive.
3) Show that H2(X, Z) and the coherent cohomology H2(X, Ox) are functors of degree ~ 2. (Over C, the extra factor can be described by the
Kiinneth formula:
More generally, show that Hq(X, Z) is a functor of degree

4) Show that the Picard functor Pic(X) has degree ~ 2 by
(a) using the theorem of the cube,
or by
(b) over C, using the complex analytic exact sequence
and that Hl(X, Ox), H2(X,Z) have degree ~ 2.

We may then normalise the heights he,
E Pic(A):
q.
35
Theorem. (Neron-Tate). Let A be an abelian variety over K. There is a
unique function c I-t he on Pic(A) with values in the space of real valued
functions on A(K) such. that,
1) he (X) = he(!) + 0(1).:
_
2) Additivity: hCl +C'l = hCl + hw
3) Functoriality: for all endomorphisms > : A -+ A, we have
h4>*c = he 0 > for c E Pic(A).

Further if B is another abelian variety and t/J : B -+ A is a homomorphism, then
h1/J*e = hc 0 t/J for all c E Pic(A).
Proof
= [2]. From
the corollary above, we know that >* c = 4c. There is then a unique hc
satisfying 1) and hc(2x) = 4hc(x). The uniqueness of hc follows from this
and 2),3).
If > is any endomorphism, then >* c is symmetric, and the commutativity lemma (3.1) applied to the diagram,
1) Suppose that c is symmetric. Consider the homothety >
A ~ A
14>
14>
~
gives hc</> = hc 0 >.

The additivity is clear, by the uniqueness.
2) Suppose that cis antisymmetric. Then >*c = 2c and one uses the same
argument.
32 In the_general case, we have 2c = (c + c') + (c - c') and we put hc =
(hc+c'
+ he-cl )/2.
3.3. Quadraticity of
hc on abelian varieties.
Theorem. Let c E Pic(A).

degree ::; 2 on A( K).
The normalised height hc is a function of
Proof
Let
~2hc(x, y, z)
= hc(x + y + z) - hc(x + y) - hc(x + z) - hc(Y + z)+

+he(x) + he(y) + he(z).
36
We must show that !:lile = O. Let 8123,812, 8}, ... : A x A x A ---. A be
the seven morphisms given in 3.2. By the functorial property of normalised
heights, we have
and similarly for th~ other terms in !:l2he' ~s q =_8i23C- L: 812c+ L: 8ic_= 0
(3.2), we deduce h q = O. By additivity, h q = hSr23e - L: hS~2e + L: hsre,
whence the theorem.
Let Be(x, y) be the bilinear form associated to he:
If cis antisymmetric, the function he(x) is of degree::; 2 and odd: h e( -x)

-he(x)j so Be(x, y) is biadditive and antisyIDmetric:
Be(-x,-y) = -Be(x,V),
therefore Be = 0 and he is additive.
If c is symmetric, he is a quadratic function:
Hence:
Theorem. Let A be an abelian variety, c E Pic(A). Then:
1). If c is symmetric, the normalised height he is quadratic on A(K).
2). If cis antisymmetric, he is Z-linear (additive).
In general, the difference, f(x) = he(x) - !Be(x,x) is additive. It remains to explain how this additive part can be computed from the quadratic
part. For this we recall some properties of the dual abelian variety.
3.4. Duality and Poincare divisors.
Let A be an abelian variety. The dual abelian variety A' has the following properties (which characterise it, up to isomorphism in characteristic
zero, and up to purely inseparable isogeny in characteristic p).
37
The group A' may be identified with Pic(A). This identification is

given in the following way. On A x A' there is a canonical element P E
Pic(A x A') (Poincare divisor class) with the following properties:
a) P induces 0 on A x 0 and on 0 x A'.
b) For a' E A', the divisor class P induces on A x {a'} (identified with A)
the element a', viewed as an element of PicO(A).
The equivalence class of such a P is unique. Exchanging the roles of A
and A' leaves P invariant, i.e. identifies A" with A.
Example. The dual of the elliptic curve E is E itself. The Poincare divisor
is the divisor ~ - {O} x E - Ex {O} on Ex E , where ~ is the diagonal. We
obtain the usual identification of E with Pic(E): x ~ class of (x) - (0).
The uniqueness of the Poincare divisor class shows that it is symmetric
P = P' = [-l]*P.
We associate to P E Pic(A x A') the normalised height
A'(K).
hp
on A(K) x
Proposition. Tbe beigbt hp is a quadratic form wbich is trivial on tbe

two factors:
hp(x,O) = 0, for x E A(K),
hp(O, y) = 0 for y E A'(K).
This follows from property a) of the Poincare divisor class.

Let Bp be the bilinear form associated to hp. As (x, y) = (x, 0)+(0, V),
we have
hp(x, y) = hp(x, 0) + hp(O, y) + Bpx, 0), (y,
= Bpx,O),(y,O,
therefore hp(x, y) is additive in x, y. We then have a canonical bilinear
form on A(K) X A'(K).
We now use pro~erty b). To y E A'(K) (i.e. y E Pic(A corresponds
the canonical height h y
Theorem. For (x, y) E A(K) x A'(K), we bave
We first prove a general lemma on translations.
38
Lemma. Let A be an abelian variety, C E Pic(A), a E A(K), Ta the
translation x I-t x + a, and Ca = Tac = T~ac. Then
where Be is the bilinear form associated to he.

Proof of the lemma. Define f(x) by
Since
he,. (x) = he(x - a)
+ 0(1) =
he(x)
+ h e( -a) + Be(x, -a) + 0(1),
we have f(x) = 0(1). Hence f(x) is a bounded function of degree

f(O) = o. Therefore f = o.
2 and

Consider the map fy : A --t A X A', X I-t (x, y). We have fy = To,y 0 fo.
From b), f;P = y E PicO(A). Hence y = f;P = foT;,yP, and since fo is a
homomorphism, we have
that is to say
hy(x)
from the lemma with a
Whence the theorem.
= hT>0,11 p(x,O) = hp(x, 0) -
= (0, -y).
But hp(x, 0)
B(x, y),
= 0 and B(x, y) = hp(x, y).
Relation between quadratic and linear forms.

Let c E Pic(A). We define <Pea = Tac - c for a E A. As <Pea is in an
algebraic family containing 0, we have <Pea E Pic(A). The theorem of the
cube shows that >c : A --t A' is a homomorphism. Further, if c is ample,
then >e is an isogeny.
Let c E Pic(A). Let Be be the bilinear form attached to the canonical
height he.
39
Theorem. Hx,y E A(K), we have
Proof
As 4>cY = Tyc - c, we have
But the lemma above gives
whence the theorem.
Remark. If one knows hp on A(K) x A'(K), one can reconstruct all the
canonical heights associated to the elements of Pic( A). For if c is symmetric,
we have from the last theorem
In the general case, one writes 2c = (c + c') + (c - d), where c + c' is

symmetric and c- c' E A'(K), since [-1] acts trivially on NS(A); therefore
-
1-
hc(x) = '2Bc(x,x) + '2hc-c'(x)

1= -'2 hp (x, 4>cx)
1-
+ '2hp(x, c -
c').
3.5. Example: elliptic curves.

Let E be an elliptic curve, that is an abelian variety with dim E = 1.
A divisor D = L: n p.P on E is equivalent to zero if and only if deg D =
L: np = 0 and L: npP = 0 in the group E = E(K). Hence PieCE) = Z x E,
the projection PieCE) -+ Z is given by the degree and Pic(E) -+ E by the
sum on the elliptic curve; a generator c of Z is D = (0) (a divisor of degree
1) and the points of E in PieCE) are represented by (a) - (0).
The action of the symmetry [-1] : c t-+ d is the following: on the factor
Z, it is the identity (the degree does not change by any automorphism) and
on the factor A it is -1. A divisor class represented by (n, a) is symmetric if
and only if 2a = 0, and it is antisymmetric if and only if n = O. In particular,
40
every element of Pic( E) is a sum of a symmetric and an antisymmetric
element.
There are three interesting choiees of ample divisors: CI = (0), C2 =
2(0), C3 = 3(0). By the additivity, we have hCi = ihct) (i = 1,2,3).
The ample divisor CI corresponds to a generator of Z in PieCE). The
linear system corresponding to (0) is just (0).
The linear system corresponding to C2 has no fixed points (C2 is generated by its global sections); it consists of pairs of opposite points on the
elliptic curve. In the Weierstrass embedding y2 = 4x 3 - 92X - 93, c2 corresponds to E -+ PI given by (x, y) 1-+ (1, x). Thus H 2(P) = H(I, x).
The divisor C3 has no fixed points and is very ample. In the Weierstrass
embedding, H3(P) = H(l,x,y). We have
H3(1, x, y) ::::::: H2(1, x)3/2

and
1
h C1 (P) = 2 logH2 (1, x) +0(1)
= 31ogH3(I,x,y) +0(1).
Explicit bounds for the 0(1) can be given (see [MZ] for C4 = 4(0); [Z]).
3.6. Exercises on elliptic curves.

Let E be an elliptic curve as above. Let C = CI be the element of
Pic(E) C>! Z x E corresponding to (1,0). Consider the following morphisms
Ex E -+ E : the projections PI(X, y) = x, P2(x, y) = y, the sum sex, y) =
x + y and the difference d(x, y) = x - y. Let
4> = (s,d) : E x E
(x,y)
1-+
-+
E xE
(x + y,x - y).
(1). Show that

4>*(C 1 + 1 c) = 2(c 1 + 1 c),
that is to say,
Deduce that
41
[In the Weierstrass embedding E - t P 2, P ~ (1, x( P), y( P)), the
function F(P,Q) = x(P) - x(Q) on E x E has divisor /}. + /}._ - 2{(0) x
E+E x (O)}, where /}. is the diagonal x = y and /}._ is the divisor x+y = 0.]
(We have already seen this formula (2.8, Example 2) without the
tilde's but with an O(l)-term; for normalised heights there is no 0(1).)
A function on a group which satisfies this property is necessarily
quadratic:
(2). Let E be an abelian group, R an abelian group in which 2 is invertible,
and h : E - t R a map such that hex + y) + hex - y) = 2h(x) + 2h(y).
Show that the function B(x,y) = hex + y) - hex) - hey) is Z-bilinear and
symmetric, and hex) = ~B(x,x).
[The second difference C(x, y, z) (see 3.3) of h equals B(x + y, z) B(x, z) - B(y, z) which is an odd function of x and is symmetric in x, y, z;
but h is even so that C( -x, -y, -z) = C(x, y, z), hence C = 0.]
(3). If ..\ is an endomorphism of E, show that
he(..\x) = (deg "\)he(x).
(4). Let a, x E E(K); identify a with class((a) - (0)) in PicO(E2. Show that
ha(x) = -B(a,x), where B is the bilinear form associated to he.
3.7. Applications to properties of heights.
Torsion. Let V = A( K) j A( K)tors where A( K)tors is the torsion subgroup
of A( K) (as we are over an ~lgebraically closed field, A is divisible). Plainly,
V is a Q-vector space and he factors through V:
he(x + t) = he(x), for c E Pic(A), x E A(K), t E A(K)tors.

[This is a general property of functions of finite degree with values in a
Q-vector space.]
Positivity. If c E Pic(A) is ample and symmetric, tben he (x) ~ 0 for all
x E A(K).
Proof
By 2.10, there is C E R such that
he(x) ~ C, for all x E A(K).

Thus for all n ::: 0, we have he(nx) ~ C, therefore he (x) ~ Cjn 2 Letting
n - t (X), we get he(x) ~ O.
Exercise. Show that the condition "c is symmetric" cannot be dropped.
42
3.8. Non-degeneracy.
Theorem. Suppose that A is defined over Q. If c is ample, the quadratic
part of he is a positive non-degenerate form on V = A(Q)/A(Q)torsWe write he = Qe + L e, where Qe, Le are, respectively, quadratic and
linear on V. We extend these to VR = R V.
Let V be a vector space over Q, F a real valued quadratic form on V .
We shall say that F is non-degenerate if for all finite dimensional sub-spaces
V' of V, F considered as a quadratic form on
is non-degenerate in the
usual sense (non-zero discriminant).
If F is positive, this is equivalent to VR having no non-zero isotropic
vectors (but such a condition for V itself is insufficient; for example, in 2
dimensions with coordinates (x, y), the form (x - 1ry)2 is positive, has no
non-zero isotropic vector over Q but acquires some after tensoring with R).
Va

The quadratic part of he is ~ (he + he' ). As c + c' is symmetric, we may
suppose that c = c'.
Lemma 1. H A is a finitely generated subgroup of V , then, for any MER,
there are only finitely many elements'x of A such that he('x) ::; M.
Proof.
We lift a set of generators of A to points xl, ... , Xn of A(K) with K
finite over Q. As c is ample, one of its multiples is very ample, and we may
suppose that c itself is so; up to 0(1), he is the usual height and we then
apply Northcott's theorem (2.4).
Lemma 2. Let F be a quadratic form in n variables xl, ... , Xn with real
values. Suppose that for all MER there are only finitely many points
(Xl, ... , xn) E zn with F(x) ::; M. Then F is positive and non-degenerate.
Proof.
The assumption on F implies that F is ?: 0 on zn, hence on Qn, hence
on Rn by continuity.
We now show that F is non-degenerate. If not, F would come from a
form on Rn-l by some projection p : Rn -+ Rn-l. Now, plz .. is injective,
by the finiteness hypothesis. Therefore p(zn) is not discrete in Rn-I, thus
there is a sequence of distinct points xi of zn such that p(xi) -+ 0 in Rn-I.
Then F(x i ) -+ 0, whence a contradiction with the hypothesis for M = l.
The theorem follows from lemma 1 and lemma 2.
43
Questions. a) Is there an elliptic curve E over Q and two non-torsion points

x, y of E(Q) such that B(x, y) = 0 (where B is the scalar product provided
by the Poincare divisor on E x E)?
b) What is the transcendence degree of the field generated by the values of
B? Can it be < r(r + 1)/2, where r = rank E(Q)?
3.9. Structure of A(K): a preliminary result.
Theorem. If K is a number field and A an abelian variety defined over
K, tben A(K) = FEB L, wbere F is finite and L is free over Z.
[We shall see in the next chapter that L is finitely generated (MordellWeil theorem).]
Proof
Let F be the torsion subgroup of A(K), L = A(K)/F, and VK = QL.
1) F is finite.
We select c to be symmetric and very ample. We have hc(x) = 0 if
x E F. Northcott's theorem (2.4) implies that the number of such x is
bounded, thus the order of F is finite.
[We may also prove the finiteness of F by a p-adic argument. Let Kv
be the completion of K for v above the prime number p, then F c A(Kv).
Now, A(Kv) is a compact p-adic Lie group, hence has an open subgroup U
which is torsion free. Since F embeds into A(Kv)/U, F is finite.]
Each of these two arguments show a little more: if d is a given integer,
the elements of A( K)tors of degree ::; dover K form a finite set.
2) L is free over Z.
Lemma. If V' is a finite dimensional Q-subspace of V K, tben L'
is a lattice of V' (i.e. is isomorpbic to zdim V' ).
= L n V'
Proof
The positive definite quadratic form h~ = hclv",- on vt is such that for
all M, {A E L', h~()") ::; M} is finite. Therefore L' is discrete in vt. As
V' = Q L', we have the lemma.
We then construct a basis of L in the following way If V has finite
dimension, then the lemma shows that L is free. If not, dim V = ~o (V is
at most countable); let vi, ... , V n , ... be a basis of V. Let Vn be the subspace
with basis (Vi, ... , V n ). From the lemma, L n Vn is free for all n. Suppose
that one has constructed a basis El, ... , En of L n Vn . We shall complete this
to a basis of L n Vn+1' The image of L n Vn+1 -+ Vn+1/Vn ~ Q is finitely
44
generated hence isomorphic to Z; let n+l E LnVn + 1 be such that its image
is a generator. Then I, , n, ... is a basis of L.
Complement.
Let red) be the number of torsion points of A(K) of degree ~ d. Then:
a) By Northcott's theorem, red) is finite and can be effectively bounded.
b) If A is an elliptic curve without complex multiplication, we have red) ~
d3 / 2 , as d -+ 00. Indeed, one knows that the Galois group is "large" and
transforms one torsion point to all other torsion points of the same order
(up to a finite factor). The number of points of order 6 is essentially 62 ,
and the number of points of degree ~ d is
62 ~ d3 / 2
6~dl/2
c) If A is an elliptic curve with complex multiplication, we have r( d) ~ ~. (1)

Indeed, let R be the ring of integers of the imaginary quadratic field of
complex multiplication of A. If ~ is a non-zero ideal of R, let 4>(~) (Euler's
function) be the number of invertible elements of R/~. Then one has
red) ~
4>(~).
It can be shown that the Dirichlet series Ea=F o 4>(~) - 8 can be continued to
an analytic function for Re( s) ;::: 0 with a si~ple pole at s = 1. One deduces
using the Wiener-Ikehara Tauberian theorem that E4>(~9 4>(~) ""' C~.
d) For an abelian variety of arbitrary dimension n, the order of magnitude of

r( d) is not easy to determine, but one can at least prove a bound r( d) < dN ,
where N depends only on n.
3.10. Back to 2.11 (c algebraically equivalent to zero.)

In the next two , we prove two previously stated results about the
elements of the Picard group.
(1) This was stated as a conjecture in the French text; the proof
sketched below is due to R. Odom.
45
Theorem. Let V be a non-singular projective variety over K, c E Pic(V),

and d an element of Pic(V) which is ample. Then
1/2
hc ~ O(hc' ) + 0(1), on V(K).

Proof.
1). The case where V = A is an abelian variety.

We may then identify c with a point of the dual A'. We may take the
normalised heights hc, hc' in place of hc and hc" Further, we may suppose
that d is symmetric: d = [-l]*d.
Associated to c' E Pic (A) is a homomorphism 4>c A -+ A' defined by
4>c,(a) = Tad - d. As d is ample, 4>c' is an isogeny, and in particular is
surjective. There is a E A(K) such that c = 4>c,(a) and we have (cf. 3.4):
l
hc(x) = hp(x,c) = hp(x,4>e'(a = -Bc'(x, a),

where
Bc,(x,a) = hc'(x + a) - hc'(x) - hc,(a).
But hc' (x) is a quadratic form, and we have seen that hc' (x) ~ 0 as c' is
ample and symmetric. The Cauchy-Schwarz inequality implies
hence the result in this case.
2). General case.

Let A be the Albanese variety of V, and let 4> : V -+ A be the canonical
morphism of V into A (relative to a given base point of V). It is known
that if c E Pic(V), then there is CA E Pic(A) = A' with c = 4>*CA. Then
hc = hCA 04>+ 0(1).
Let dAE Pic(A) be ample, and let C" = 4>*c'A E Pic(V). Then
1/2) ,
hCA = 0 ( 1 + hcll
A
therefore
As c' is ample, we have
whence the result.
1/2) .
hc = 0 ( 1 + hcll
Ihell I = 0(1 + he' ),
46
3.11. Back to 2.9 (torsion c).
Theorem. Let V be a non-singular projective variety over a number field
K. Let c E Pic(V). If Ihel = 0(1) on V(K), then c is an element of finite
order in Pic(V).
The proof will show that there is an integer dv = d, depending only
on V, such that if Ihel = 0(1) on the points P E V(K) of degree ~ dover
K, then c is torsion. For an elliptic curve, d = 2 suffices.
Proof
l)Special case when V is a curve.
We have an exact sequence
--+
J(K)
--+
degr~e
Pic(V)
where J is the Jacobian.

We show first that deg c = o. If one had deg c > 0, c would be ample.
We choose d = dv such that V has infinitely many points of degree ~ d
over K (just write the curve as a ramified covering of degree d of Pl).
From Northcott's theorem (2.4), he --+ 00 on this set of points, thus is
unbounded. The same argument applied to -c shows that deg c = o.
Therefore, we have c E J. The Jacobian is principally polarised, and
so is equal to its dual: J = J f , canonically. We select a rational point of
V (by making an extension of I< of bounded degree) j we have then a map
</> : V --+ J and the height he on V can be chosen to be he = he,J 0 </>.
Suppose that c is not of finite order on J (in particular, the genus is ~ 1).
By 3.4, we have
he,J(x) = -B(x, c),
where B is the bilinear form associated to a strictly positive quadratic form
on J(K) modulo torsion. As this bilinear form is non-zero, there is a point
x of the Jacobian (for example x = c) such that B(x, c) =f. o. There is
then a sequence of points Xn of J(I<) (for example, Xn = nc) such that
he,J(xn) --+ 00.
Let 9 be the genus of the curve. It is well known that the morphism
</>g : vg
=V
(Vl, ... ,Vg )
x ... x V
1-+
--+
'L:(Vi)
i
is surjective. We use the following simple lemma.
47
Lemma. Let f : X - t Y be a surjective morphism of algebraic varieties
over K. There is an integer df such that for all fields K' with K :J K' :J K
any point P E Y(K') can be lifted to X over an extension of K' of degree
~ df
[This is clear if X is finite over Y, by taking d to be the degree of
the covering. We reduce to this case by divissage. By induction, Y may
be assumed to be irreducible and of dimension n. We then select in X a
subvariety X' of dimension n for which the morphism X' - t Y is surjective
and finite outside a subvariety Y' of Y of dimension < n. On Y', one uses
the induction hypothesis.]
We apply this lemma to <pg : vg - t J. There is d 1 such that all points
of J(K) lift to points of degree ~ d1 . Let y~l), ... , y~g) be in V(Kn) with
[Kn : K] ~ d 1 such that
L <p(y~
9
= Xn
j=l
As the height is additive (it equals -B(x, c we have
But he,J(x n ) - t
00,
therefore the he(Y~ are unbounded.
2) General case.
Let V be a non-singular projective variety of any dimension. We have
a homomorphism (see 3.2) Pic (V) - t NS(V) ~ ZO EEl (finite group). We
first show that c has trivial image in ZO, i.e. is homologous to 0, up to
torsion.
There are a finite number of curves Gi C V (in fact, 6 curves suffice)
defined over K ,such that if c E Pic(V} induces an element of degree 0 on
each Gi, then c is homologically equivalent to zero, up to torsion. In terms
of divisors, the cohomology class, with coefficients in Q, of a divisor D is
zero if its intersection number with each Gi is zero.
As we have assumed that he is bounded, we may apply the above: on
each curve he is bounded, therefore c is homologically equivalent to zero,
up to torsion. Replacing c by a non-zero multiple, we may then suppose
that c E Pico (V) .
We now select a curve G on V which is a non-singular plane section
(that is, if dim V = n, take a section by a linear variety of co dimension n-l
48
in general position). The restriction map PicO(V) --+ Pic(C) is known to
be injective (it is enough to know that its kernel is finite, which is easy).
From 1), c is of finite order in Pico (C), thus is of finite order in Pico (V).
49
4. THE MORDELL-WEIL THEOREM

4.1. Hermite's finiteness theorem.
Theorem. There are only finitely many number fields K (up to isomorphism) of given discriminant dK (and of given degree nK = [K: Q]).
If nK ~ 2, one has IdKI ~ (J3)nK (there are better results, for example, nK > 20 =? IdKI > lOn K (Odlyzko, thus if the discriminant is
bounded, the degree is also bounded.
Sketch of proof
Assume for simplicity that K is totally real (the general case is similar,
see e.g. [L6]). Let n = n K. The integers OK of the field K form a lattice in
Rn. Applying Minkowski's theorem, there is a constant C (depending only
on the discriminant) and a non-zero element x E OK such that if xl, "',Xn
are the conjugates of x, then one has
The integer x generates the field K and the coefficients of its minimum
polynomial are bounded by a function of the discriminant. Therefore, there
are only finitely many such x, and such fields K, up to isomorphism.
Variant. Let K be a number field, S a finite set of places of K, and nan

integer ~ 1. Then there are only finitely many extensions L of K (up to
isomorphism) of degree n which are unramified outside S.
This variant is deduced from Hermite's theorem in the following way.
We may first reduce to the case f( = Q. It remains to bound the discriminant of L in terms of S and n only. Such a bound has been given by
Hensel:
IdL/QI =
pOP,
II
pES
with
6p
:::;
n -1 + nlog(n)/log(p),
see e.g. l of [Sell.

For the application below, the following particular case of Hermite's
theorem is sufficient (and can be proved directly from the unit theorem):
There are only finitely many cyclic extensions of K of given degree which
are unramified outside S.
50
4.2. The Chevalley-Wei! theorem.
Theorem. (Chevalley-Weil, [CW]). Let K be a number field. Let 'Ir : Y ~

X be a finite etale morphism of degree d of projective varieties over K.
There is a finite set of places S of K such that if K' is a finite extension of
K and if P E X(K'), then there exists K" :) K' and Q E Y(K") such that
'Ir( Q) = P, [K" : K'] ~ d, and the extension K" / K' is unramified outside
the places SKI of K' above S.
There is an analogous statement for integral points without the projectivity hypothesis (a non-projective case: G m ~ G m , x ~ x 2 ; the square
roots of units are unramified except at 2).
Proof.
Let 0 be the ring of integers of K; if S is a finite set of non-archimedean
places of K, one can view S as a subset of Spec 0; let Os be the ring of
S-integers:
Spec Os = Spec 0 - S,
and
UOs = K.
From [G, IV,8], all algebraic constructions over K can be made over some
Os for S sufficiently large. Hence, there are finite etale morphisms 'Irs:
Ys ~ Xs of projective Os-schemes which induce 'Ir by the base change
s ~ K. Select such a suitable S.
Let 0' be the ring of integers of K'. From the viewpoint of schemes,
P E X(K') corresponds to a morphism Spec(K') ~ X which, as X is
projective, provides a morphism P : Spec(O~KI) ~ Xs, where SKI is the
set of places of K' lying over S. Base changing 'Irs by the morphism P,
we obtain a scheme A, finite and etale over Spec(O~K/)' which provides
Q E Y(K"):
SpecO~K I
Ys
p
--+
1~s
Xs
The theorem can be refined and stated for an arbitrary field: the discriminant of the extension K" / K' divides a fixed element of K (see, e.g.
[L6], Chapter 2, Theorem 8.1.). Over a number field, these are equivalent,
because of Hermite's theorem (4.1).
51
4.3. Mordell-Weil theorem.
Theorem. (Weak Mordell-Weil theorem.) Let K be a number field, A an
abelian variety over K, and r = A(K). Then r /nr is finite for all integers
n 2: l.
Proof. (Under the hypothesis that the n-torsion points of A be K-rational.)
Let An be the kernel of the map 'Jr : x f-+ nx on A. From the ChevalleyWeil theorem (4.2), applied to the covering 'Jr, if P E A(K) then there is
Q E A(K'), where [K' : K] ::; d(= n 2dim A) such that nQ = P, and K' /K is
unramified outside a finite number of places. By Hermite's theorem (4.1),
the corresponding fields K' are finite in number; let L be a finite Galois
extension of K containing all the K'. Let G be the Galois group of L / K.
We shall embed r/nr in Hom(G,An):
If PEr, there is Q E A(L), with P = nQ. Let s E G; we have
P= sP= nsQ,
therefore
n(sQ - Q) = 0,
whence
sQ - Q E An.
As Q is unique, up to addition by an element of An, and as the elements
of An are rational over K (by hypothesis), sQ - Q is independent of the
choice of Q. For PEr and s E G we put (P)(s) = sQ - Q. For s, t E G,
we have
stQ - Q = s(tQ - Q) + sQ - Q
= (tQ - Q)
+ (sQ -
Q),
therefore, (P) E Hom(G,An). If P = nP' with P' E r, we have (P) = 0,

thus factors through a homomorphism r / nr -+ Hom( G, An). Finally, if
sQ - Q =
for all s then Q E r thus P = nQ E nr, whence we have
an injective homomorphism, : r/nf -+ Hom(G, An). As Hom(G, An) is
finite, we deduce the finiteness of r / nf.
[Cohomological variant. By applying Galois cohomology to the exact sequence

- + An
- + A(K)
~ A(K) - +
one gets an injection : r/nr -+ Hl(K,An), where

Hl(K,An) = Hl(Gal(K/K), An). If S is a finite set of places of K, let
52
H}(K, An} denote the subgroup of HI(K, An) made up of the elements
which are "unramified outside S". One then proves that:
a) H}(K, An) is finite (by Hermite's theorem);
b) (r /nr) is contained in H1(K, An) if S is large enough (e.g. if S
contains all places dividing n, and all places where A has bad reduction.
This shows that r /nr is finite.
In the proof above, we have assumed that the points of An are rational
over K so that Hl(K, An) can then be identified with Hom(Gal(K/K),An)
and one can dispense with the cohomologicallanguage.]
Non-abelian generalisation. The above theorem could be reformulated by
saying that, if 7r : B ~ A is an isogeny of abelian varieties, then Coker{ 7r :
B(K) ~ A(K)} is finite. There is a similar statement for algebraic groups
(not necessarily commutative), with A(K) and B(K) replaced by suitable
"S-arithmetic" groups. The proof is basically the same. See e.g. Borel
[Bor] and Honda [Ho].
We can now prove:
Mordell-Weil Theorem. Let K be a number field and A an abelian

variety over K. Then the group A(K) is finitely generated.
Proof.
Select an integer n ~ 2. Let K' = K(An); we have just proved that
A(K')/nA(K') is finite. But, by 3.9, A(K') is a direct sum of a free abelian
group and a finite group. Hence A(K') is finitely generated and the same
is true for A(K), since A(K) is a subgroup of A(K').
Effectivity. The group r /nr is embedded effectively in a finite group,
whence one has an effective bound for Ir/nrl, and also for the rank of
r. But there is no known effective method for characterising the elements
of r /nr in the image nor for finding a priori a set of representatives Ii E r
of r /nr. It is this which makes the search for rational points on elliptic
curves an art but not a science.
Remark. The Mordell-Weil theorem has been extended by Neron [Ne] to
the case where K is only assumed to be finitely generated over the prime
field (Q or F p). See e.g. Lang [L6], where this statement is reduced to
the standard one by means of the Lang-Neron theorem (if K is a function
field over a ground field k, A an abelian variety over K, and r : B ~ A a
"K/k-trace" of A, then A(K)/rB(K) is a finitely generated group).
53
4.4 The classical descent.
Let r be a Z-module, and let V be a Q-vector space containing r /r tors .

We suppose given a real valued positive quadratic form on V, written x.x =
Ixf. Let n :2: 2 be an integer, and let E r be representatives of r /nr.
Let C be a real number with C > o. Assume that hil :::; C for all i.
,i
Descent Lemma. The group

that 1,1 :::; 2C.
r is generated by the elements, E r
such
Proof
For each real number D > 0, we denote by r D the subgroup of r
generated by the elements, E r with 1,1 :::; D. We wish to show that
r 2C = r; as Urn?:l r mC = r, it suffices to check by induction on m that
r mC C r(m-l)C,
for
m:2: 3.
Let y E r mC; there is i and x E r with

y
= nx +'i.
Then
thus
Ixl:::; 2(lyl + hd)

< (m+l)C.
-
As m :2: 3, we have, (m + 1)/2C :::; (m - l)C, thus x E
r(m-l)C.
Proof of the Mordell- Weil theorem via the descent lemma.

The lemma is applied to r = A(K), with Ixl 2 being the normalised
height associated to some ample symmetric divisor class. By the weak
Mordell-Weil theorem (4.3), r/nr is finite, hence C exists; since the elements of A(K) with hi :::; C are finite in number (Northcott's theorem,
2.4), this shows that A(K) is finitely generated.
4.5. The number of points of bounded height on an abelian variety.
Assume K is a number field, and choose a projective embedding 4> :

A -)- P n , whence an associated logarithmic height h = log(H 0 4. Let
N(X) be the number of, E r such that he,) :::; X.
54
Theorem. (Neron, [N4]). We have
where
= r tors EB ZP,
w is the order ofrtors ,
p is the rank of r jrtors,

Cp is the volume of the unit ball in RP: cp = 7r p/ 2jr(! + pj2),
D = det(B("Yi,,},;)j2), where '}'!, ... ,'}'p is a basis ofrjrtors , and B is
the bilinear form attached to the normalised height h:
B(x, y)
hex + y) - hex) - hey).
[If the element of the group Pic(A) attached to <p is symmetric, we may
replace O(X(P-l)/2) by O(Xp/2- p/(P+l).]
Corollary. N(X) '" CXp/2, as X
--+ 00,
with
= cpwj..,fi5.
Proof of theorem.
For p = 0 we have N(X) = w for X sufficiently large. We now suppose
that p ~ l.
It suffices to prove the same estimate for the number R(X) of'}' E r for
which he'}') ~ X. Indeed, hand h differ by a bounded quantity: Ih-hl ~ a,
therefore
R(x - a) ~ N(X) ~ R(X + a).
Thus if we prove the estimate for
R, we have
R(x + a) ~ c(X + a)p/2 + O(X(P-l)/2)

= CXp/2
+ O(Xp/2-1) + O(X(p-l)/2)
CXp/2 + O(X(p-l)/2),
whence the estimate for N(X).

It is enough to count the points modulo torsion, provided that one
finally multiplies by w.
Identifying r jrtors with ZP by means of the basis '}'!, .. , '}'P' we are led
to count the (n!, ... , np) E ZP with
P
h(L:: ni"}'i) ~ X.
i=l
55
h is a positive definite quadratic form plus a linear term in the "Ii. Let
Nquad be the number of "I for which the quadratic part hquad of h is ::; X.
But
Then
Nquad(X
- bX 1/2 )::; N(X)
::; Nquad(X
+ bX 1/2 ),
for a certain constant b. It is then enough to study Nquad (X).

This means one has to estimate the number of integral points in the
interior of the ellipsoid hquad ::; X. This number is equal to the volume
cXp/2 of the ellipsoid hquad ::; X up to some O(XQ) error term. Elementary
methods give a = p/2 - 1/2, which is enough for what we want (by [La]
and [CN], one can take a = p/2 - p/(p + 1), and even better results are
known when p = 2).
The constant c is the volume of the ellipsoid for X = 1. If the quadratic
form is l: x~ , we find cpo If not, we change the basis and the volume change
appears as the determinant of a matrix, whence the factor ..JJ5.
4.6. Explicit form of the weak Mordell-Weil theorem.

As in 4.3, let n be an integer ~ 1 and An the kernel of [n] : A -+ A.
Let fLn be the group of roots of unity, and d the dimension of A.
Assume:
(1) An C A(K);
(2) fLn C/(.

[Exercise. Prove that (1)
=?
(2), if d
1.]
The covering [n] : A -+ A is abelian and unramified with Galois group

An, thus of type (n, ... , n) and of degree n 2d Let KA be the function field
of A. Using Kummer theory, one sees that:
The function field extension of /(A corresponding to the covering [n] :

A -+ A is the composite of the extensions /(A( if(f)), where <p runs through
the elements of KA having the following two properties:
(i) The divisor (<p) is divisible by n;
(ii) <p is defined at 0 and <p(O) = 1.
(Condition (ii) expresses the fact that 0 splits in the covering.)
Suppose that one has explicitly the functions
covering [n] : A -+ A.
We defined in 4.3 an embedding
<PI, ... , <P2d describing the
r/nr '"--+ Hom(Gal(K/K),A n ).
56
As An is an abelian group of type (n, ... , n),
Hom(Gal(K / K), An) ~ Hom(Gal(K / K), Z/nZ

~ Hom(Gal(K/K),JLn)
X . X
X X
Z/nZ)
Hom(Gal(K/K),JLn).
Using Kummer theory again, we have Hom(Gal(K/K),JLn) ~ K*/K*n,

hence an injection
r/nr <-+ K* /K*n
X X
K* /K*n,
which can be shown to be:
[Here 4>j(-y) is the value (mod K*n) of 4>j at "I in case the divisor of 4>j
does not go through "I; in case it does, one just replaces 4>j by 4>jf n , with
a suitable f E K A, so that this does not happen.]
From this point of view, the proof of the weak Mordell-Weil theorem
reduced to saying that the image of "I 1-+ 4>j('Y) is a finite subgroup of
K* / K*n for each j.
Example: elliptic curves.

We take n
= 2, d = 1 and write the equation of A
as
with e1, e2, e3 distinct elements of K, which we may assume to be integers.

The divisor D1 = (Pt) - (0) where P1 = (el, 0) and 0 = (00,00), is
such that 2D1 = (4)1) with 4>1 = (x - el) = (x - e2)/(x - e3)x square.
Define similarly 4>2 = (x - e2). The construction above gives an injective
homomorphism
:
r /2r -+ K* / K*2
K* / K*2 ,
(Exercise. Check that is a homomorphism, and that it is injective, using

explicit addition formulae.)
The finiteness of the image of follows from the more precise statement:
57
Lemma. If g:J is a prime ideal of J( not dividing (el - e2)(e2 - e3)(e3 - ed,
then
vp(x - el) == 0 (mod 2) and vp(x - e2) == 0 (mod 2),
for any (x, y) E f.
Proof.
Put mi = vp(x - ei), i = 1,2,3. Since (x - ed(x - e2)(x - e3) is a
square, we have ml + m2 + m3 == 0 (mod 2). To prove that mi == 0 (mod
2) for all i we distinguish two cases:
a) x is not g:J-integral. If vp(x) = -m, then mi = -m for i = 1,2,3, hence
3m == 0 (mod 2) and mi == 0 (mod 2).
b) If x is g:J-integral, at most one of the mi can be f: 0 (otherwise, one of
the ei - ej would be divisible by g:J). Hence the result by using again
ml + m2 + m3 == 0 (mod 2).
Exercise.
Let E be an elliptic curve over C given in standard Weierstrass form
and denote by WI, W2 a pair of fundamental periods, so that
Let Yi(U), i = 1,2,3, be defined by:

Yi (u)
= y'g:J( u) -
ei, Yi
= -U + ....
The field generated by YI, Y2, Y3 over C (g:J( u), g:J' (u)) is the function field of
2-division of E, i.e. C(g:J(u/2) , g:J'(u/2)). Check this by using the relations
(d. [Wh]):
g:J(u/2) = g:J(u) + YIY2 + Y2Y3 + Y3Yl,
1 '( U/2)
2"g:J
= g:J'()
U -
(2
YIY2
2
2
2
2
2) ,
+ YlY3
+ Y2Yl
+ Y2Y3
+ Y3Yl
+ Y3Y2
and
Yi
= _g:J/(U)-I(g:J(u/2) -
for i,j E {I, 2}, if: j.
g:J(wi/4))(g:J(u/2) - g:J(Wj/2 + wi/4)) ,
58
5. MORDELL'S CONJECTURE
The conjecture is the following [it has now been proved by Faltings,
see [F,FW,Sz]].
(M) If G is a curve over a number field K, and if the genus of G is 2: 2,
then the set G(K) of points of G rational over K is finite.
It is no restriction to assume that the curve is projective and nonsingular, for this changes only a finite number of points.
We shall give three partial results on this conjecture. The first two
(Chabauty [Ch] and Demjanenko and Manin [Mal]) prove the finiteness
of G(K) under certain hypotheses, the third (Mumford [Mul]) shows that
G(K) is "sparse".
5.1. Chabauty's theorem.

We select a K-rational point Po on the curve, and map G into its
Jacobian by
</>: G
P
-+
1-+
= Jac(G)
CI((P) - (Po)).
This is an embedding, which allows us to identify G with its image:
G c J,
G(K)
= Gn J(K).
By the Mordell-Weil theorem, J(K) is a finitely generated abelian group.

It is natural to make the following conjecture:
(M') If A is an abelian variety over a field L of characteristic zero, dim A 2: 2,
G a curve contained in A which generates A, and r a finitely generated
subgroup of A(L), then G(L) n r is finite.
One says that G generates A if the differences P - Q (P, Q E G)
generate the group A. Alternatively, one may suppose after translation
that 0 E G, and then ask that G generates A in the usual sense. This
amounts to the homomorphism J -+ A being surjective (in the geometric
sense, i.e. over an algebraic closure).
It is clear that (M') => (M). Conversely,
59
Theorem. (M)
=}
(M').
The proof is the following. As everything is defined over a field finitely

generated over Q, we may suppose that L is finitely generated over Q. A
theorem of Neron (see 11.1, [Nl]) gives a specialisation of L to a number
field K and an injective homomorphism A(L) ~ AK(K), where AK is the
corresponding specialised variety of A. One is then reduced to the case of
a number field. The Jacobian J of G projects onto A, therefore the genus
of G is ~ 2, and from (M), G(L) is finite.
Theorem. (Chabauty [Ch]). As in conjecture (M'), let L be a field of
characteristic zero, G a generating curve in A, and r a finitely generated
subgroup of A(L). Suppose that rank r < dimA. Then, G(L) r is finite.
Consequently, conjecture (M) is true for the pair (G, K) if

rank J(K) < genus of G.
The proof uses the following well known result:
Lemma. If W =f 0 is a differential form of the first kind on A, then the
restriction of W to the curve G is =f O.
Proof of lemma.
The injection of G in A factors through the Jacobian of the curve,
G ~ J
1 ./
A
and by hypothesis, the map J ~ A is surjective; therefore the tangent map
is surjective, and the differential forms of A embed in those of J. As </> gives
an isomorphism between the space H(J, 0) of diffetential forms of J and
the space HO(G, O~) of those on G, the lemma follows.
[This lemma can also be proved by noting that the product map G x
... x G ~ A is surjective, if sufficiently many factors are taken; if We is the
restriction of w to G, then the inverse image of w is We 1 + ... + 1 we.]
Proof of Chabauty's theorem.
1). First reduction. It suffices to prove the theorem when L is finitely
generated over Q. Indeed, the group r is finitely generated and everything
is defined over a finitely generated extension of Q.
2). We now embed L in a finite extension of Qp (this is possible, see the
appendix at the end of this section).
60
3) We may thus assume that L is a finite extension of Qp; hence L is a locally
compact field and A( L) is a compact analytic group over L of dimension
d = dim A. Such a group is locally isomorphic to its Lie algebra: there is
an open subgroup U of A( L) isomorphic, as a Lie group, to 0 LX ... X 0 L
(d factors), where OL is the ring of integers of L.
[This can be proved as in the complex case by taking a basis of the
differential forms of the first kind, and integrating them locally; assuming
that 0 E C, one obtains the coordinates Xi =
Wi which give a local
isomorphism between A(L) and its Lie algebra.
Another method is to use the logarithm map
J:
log: A(L) -+ Lie(A),

which is everywhere defined (since A(L) is compact) and gives a local isomorphism (cf. Bourbaki, Lie III, 7.6).]
4). Let us now show that C n I' (where I' is the closure of r in A(L) is
finite. Suppose that C n I' is infinite; as A(L) is compact, we can extract a
sequence of distinct points Pi of cnI', such that P.. -+ Po where Po E CnI'.
We may assume that Po = 0, by translation, that Pi belongs to the
open subgroup U chosen above, and that r n U is free of rank d' < d.
Changing coordinates, we then have that all, = (Xl, ... , Xd) E unI' satisfy
Xl = O. Thus in d-dimensional space, the analytic curve C intersects the
hyperplane Xl = 0 in infinitely many points; in other words, Xl is a function
on the curve having infinitely many zeros Xl(Pi) = 0, which accumulate at
O. The principle of isolated zeros for p-adic analytic functions, gives Xl = 0
in a neighbourhood of 0 on C.
Let WI, ... , Wd be a base of the differential forms corresponding to the
system of coordinates Xl, ... , Xd: dXi = Wi in a neighbourhood of O. The
differential form Wl, restricted to C has at most 2g - 2 zeros, therefore
wlle = 0, which contradicts the lemma.
Chabauty proved this result in 1941. Previously, Skolem had already
used p-adic methods to study products of multiplicative groups. For an
exposition of Skolem's method, see e.g. [BSh], Chap.lV, 6.
Remark. Chabauty's method can sometimes be used to give an effective

(and small) bound for the number of rational points on a curve, cf. [Col].
61
Appendix. Embedding a field finitely generated over Q in Qp.
Theorem. Let L be a field finitely generated over Q.

a) For every prime P there is an embedding of L in some finite extension
ofQp.
b) For infinitely many primes p there is an embedding of Lin Qp.
Proof. a) Let Qp be an algebraic closure of Qp. As the transcendence
degree of Qp over Q is infinite, L is embedded in Qp. The compositum
L.Qp is then an algebraic extension of Qp (since it is contained in Qp), and
finitely generated; it is therefore a finite extension of Qp.
Let Lo be an algebraic closure of Q in L (that is, Lo = LnQ). We write
L = Q(V) = Lo(V), where V is an absolutely irreducible algebraic variety
over Lo; as V is defined up to birational isomorphism, we may suppose that
V is affine and smooth. We wish to embed L in Qp; we begin by embedding
Lo
Consider the prime ideals galp of the number field Lo where p does not
divide the discriminant of Lo; then, Lo embeds in Qp if and only if one of
the ga's has degree 1. Denote by P(L o} the set of P for which Lo embeds in
Qp. By Chebotarev's theorem, the density of such p's is > 0.
Proposition. For all p E P(L o), except finitely many, the field L embeds
in Qp.
Proof of proposition.
Recall that V is affine. Let A = Lo [V] be its affine coordinate ring. A
point P E V(Qp) defines a homomorphism A - t Qp, and we say that P
is generic if this homomorphism is injective. We have to prove that such
generic points exist.
Lemma. If Po E V(Qp) is a smooth point, then V(Qp) contains generic

points arbitrarily near to Po.
Proof of the lemma.
The elements f E A, f f= 0, form a countable family. The equation
f = 0 cuts out a subvariety Wf of dimension < dim V. These varieties
have no interior points (p-adically). We now use Baire's category theorem
(or even measure theory): the complement in V(Qp) of the union of the
Wf(Qp) is dense. If P belongs to that complement, then P is generic.
Thus to prove the proposition, it suffices to show that V has at least

one smooth Qp-rational point. To do so, one selects a finite subset S of
Spec 0 Lo such that V comes by extension of scalars from an affine scheme
62
Va
over Spec 0 Lo - S. If p E Spec 0 Lo - S, Vo may then be reduced mod

p; this gives a variety Vi" over the residue field K(p), It is known that Vi" is
smooth and absolutely irreducible for almost all p. Moreover:
Lemma. The variety Vi" has a K(p )-rational point for almost all p E
Spec (OLo)'
Proof One may reduce to the case of curves; from a theorem of Weil, the
number of points on a non-singular projective curve is ~ q + 1 - 2gJfj,
where q = N(p) and 9 is the genus (see also the Lang-Weil theorem 13.3).
Thus for q sufficiently large, it has at least one point.
Given a smooth point over K(p), Hensel's lemma provides a smooth
point of V over the completion 6", of the local ring of 0 Lo at p. So the
variety V has a point over the completion L o,,,, of Lo at p. For the p of
degree 1, we have L o,,,, = Qp.
5.2. The Manin-Demjanenko theorem.
Let C be a projective non-singular absolutely irreducible variety over
a number field K. Let A be an abelian variety over K, and lb"" 1m
morphisms from C to A defined over K.
We make two hypotheses:
First hypothesis. The Ii are independent in the following sense. Let Po E

C(K), and let ii(X) = hex) - h(Po). Then A,., im are independent in
the group Morph( C, A) (i.e. E ndi = 0, for integers ni, =} ni = 0 for all
i).
This amounts to saying that if E nili is constant, then ni = 0 for

all i (hence the condition is independent of the choice of Po). If we denote by Alb (C) the Albanese variety of C, Ii defines a homomorphism
4>; : Alb(C) -+ A, and the hypothesis means that 4>1, ... , 4>m are independent in Hom(Alb(C),A).
Second hypothesis. The Neron-Severi group NS(C) has rank: 1.

If C is a curve, NS( C) = Z (NS(C) is the group of divisor classes
modulo algebraic equivalence) and the hypothesis is satisfied.
Theorem 1. For all P E C(K) but a finite number, the points
il(P), ... , im(P) E A(K) are linearly independent over Z.
This has the following direct consequence:
63
Theorem 2.([Mal]). Under these hypotheses, if m
G(K) is finite.
> rank A(K), then
Theorem 2 immediately implies:

Corollary. Let G be a curve. Assume there exists an abelian variety A
over K such that
rank Mor(G, A) > rank A(K).
Then G(K) is finite.

[The proof will be effective.]
Prool 01 theorem 1.
Let H = Hom(Alb(C), A); this is the set of morphisms I : C - t A
such that I(Po ) = 0; H is a free Z-module of finite rank !! ~ Z5; let
HR = R0 H. We shall suppose Ii(Po ) = 0 (that is to say, Ii = Ii); thus
Ii E H C HR. Let VR = R0 A(K).
We select:
- a projective embedding of G in P N; it corresponds to an element
Cc E Pic(G);
- a symmetric and very ample element CA E Pic(A); this corresponds
to an embedding 'Ij; : A - t PM.
For IE H, we have f*cA E Pic(C); as we have a projective embedding
of C, the degree of f*cA is defined.
Lemma 1. The map If-> deg f*cA is a quadratic form with integer values
on H which extends to a positive definite quadratic form on HR.
We already know that the map
H
-t
Pic(C),
is quadratic (3.2).
If I =1= 0, deg f*CA > O. On HQ = Q 0 H, this quadratic form is
rational valued and positive except at 0 hence non-degenerate. When one
extends the scalars to R, it remains positive definite.
[This form can be interpreted as a canonical height. A morphism I :
C - t A is a point of A with values in the function field Kc of G; as we have
chosen a projective embedding of G, we have a product formula (cf. 2.2),
and the natural height on A(Kc) is deg (J*q); it is also the canonical
height since it is quadratic.]
64
On VR, the canonical height attached to CA defines a positive definite

quadratic form. Hence both HR and VR have natural Euclidean space
structures.
We denote the corresponding quadratic forms by:
1/12 =
deg (rCA),
lal 2 =hCA(a),
If
if
E H,
aEA(K).
On the other hand, if P E C( K), P defines a morphism I t-+ I( P) of H to

A(K); by extension of scalars we obtain a morphism 4>p : Ha -I- VR. We
shall show that 4>p is injective lor almost all P, which will prove theorem
1.
Clearly, we may assume that C(K) is infinite; let Pi, ... , Pi, ... be its
elements.
Lemma 2.
1) Let I E H, I ::J. O. Then 4> Pi (f) ::J. 0 if i is sufficiently large.
2) Let 9 E H be another element. Then
~
-I- 00.
Let us use the hypothesis that NS(C) has rank 1. The image of Cc E
Pic(C) is a basis of Q NS(C) ~ Q. We may thus write in Q Pic(C)
(1)
rCA = )..(f)cc +
In particular, if d
C, we have
class in
= deg(cc)
Q PicO(C),
where
)..(f) E Q.
is the degree of the projective embedding of
(2)
For P E C(K), let us compare h(P), the natural height given by the
chosen embedding C -I- PN, and I/(P)1 2 We have
The height on C associated to an element of Q PicO(C) is o(h(P (cf.

2.11, 3.10); hence from (1) and (2) we obtain
I/(P)1 2 =
~1f12h(P) + Of(l) + o(h(P.
65
This then gives:
For each fixed
E H, we have
If(PiW ~
!:.III2 , a s
~
h(Pi )
d
. -+ 00.
Since <pp(f) = I(P), we obtain then the two assertions of lemma 2:

= I(Pi ) i= 0 if i is sufficiently large, because h(f(Pi)) i= 0 for
large i.
2) As i -+ 00, we have
1) <PpJf)
It remains to show that <pp; is injective for large i. This follows from
the following lemma.
Lemma 3. Let W i ,W2 be two Euclidean spaces, A a lattice in Wi, and <Pi
a sequence of linear maps Wi -+ W 2 such that for all I, 9 E A with f i= 0
Then <Pi is injective for a111arge i.

Proof
Let e E A, e i= O. By the hypothesis, we have <Pi (e) =/:. 0 for all large
i. Replacing <Pi by a non-zero multiple, if necessary, we may assume that
l<pi(e)1 = lei Taking I = e, we have l<pi(g)l-+ Igi for all 9 E A. Comparing
this result for g,h, and 9 + h, we deduce that for allg,h E A, then
<Pi(g).<Pi(h) -+ g.h,
as
-+
00.
Let el, ... , e,\ be a basis of the lattice A. We have as i
-+ 00
thus det(<Pi(e,\).<Pi(eJL))'\,JL i= 0 for large i. Therefore the <Pi(ej), 1 ~ j ~ A,

are linearly independent for i large, and so <Pi is injective for large i.
66
In the next two sections, we give two applications one to quartic curves
(due to Demjanenko [D]), the other to modular curves Xo (pn) (due to Manin
[Mal]).
5.3. First application: Fermat quartics, (cf. [D]).

Let a, b, c be non-zero elements of the number field K, and C the plane
quartic
ax 4 + by4 + cz4 = O.
The genus of C is 3. Its Jacobian J( C) is isogenous to a product of the
three elliptic curves EOt x E(3 x By defined by the equations
with
= 4bcja2,/3 = 4acjb2" = 4ab/c2.
[This can be seen as follows. Let C(3 be the curve of genus 1 defined by
ax 4
+ by 2 z 2 + cz 4 =
\
O.
One has a natural map (X,Y,Z) 1-+ (XZ,y2,Z2) from C to C(3, hence a
morphism 4>(3 : J(C) --+ Jac(C(3). By using the formulae in Weil [W4], one
checks that Jac(C(3) ~ E(3. One defines similarly maps 4>01 : J(C) --+ EOt
and 4>.., : J(C) --+ E.., and one checks that they give an isogeny J(C) --+
EOt x E(3 X E.." with kernel of type (2,2,2).]
Assume now that a = b = 1, so that a = /3 = 4c. We have rank
~ 2; hence the Manin-Demjanenko theorem applies with
m = 2 (5.2, Theorem 2) and gives:
Hom(Jac(C),EOt)
Theorem. Let c E K* be such that the elliptic curve
has K -rank
1. Then the quartic
has finitely many rational points over K (which can be found effectively).
67
Remark.
I have not been able to find any concrete example, over K = Q, where
the theorem above applies in a non-trivial way. [The finiteness of C(K) is
trivial if C(Kv) is empty for some completion Kv of K, or if one of the two
elliptic curves Eo" E-p with a = 4c, 'Y = 4/c2 has rank 0.] In all cases I have
tried, either C(Kv) is empty for some v, or rank Ea is even.
[Note that even with the help of Faltings' theorem one still does not
know how to find the rational points on a quartic such as x4 + y4 = 17z4;
are there others than the eight obvious ones (1,2,1), ... ?]
Exercise. Let c be a positive integer not divisible by the 4'th power of a
prime. Show that there are p-adic points on the curve
for all p if and only if

a) c == 1 or 2 (mod 16),
b) every odd prime factor of c is == 1 (mod 8),
c) c;f: 3 or 4 (mod 5),
d) c;f: 7,8 or 11 (mod 13),
e) c;f: 6, 7 (mod 17),
f) c;f: 4,5,6,9,13,22,28 (mod 29).
5.4. Second application: modular curves Xo(pn), (Manin [Mal]).

Elliptic curves E equipped with a cyclic subgroup of order N correspond to points on an affine curve Yo(N) defined over Q, whose function
field is Q(j (z), j (N z It can be described as the normalisation of the plane
curve given by the equation relating j(z) and j(N z); here j is the standard
modular function.
Let K be a field of characteristic zero. A K-rational point of Yo(N)
corresponds to a pair (E, C) where E is an elliptic curve defined over K,
and C is a cyclic group of order N of points of E over the algebraic closure
of K, which is stable under the action of Gal(k/K).
More precisely, there is a function (E, C) 1----+ j (E, C) where j (E, C) is
a point of Yo (N); all points of Yo(N) are obtained thus, and two pairs give
the same point: j(E,C) = j(E',C' ) if (E,C) and (E',C') are isomorphic
over k.
(When N = 1, C = 0, j(E, 0) is just j(E).)
The curve Yo(N) is affine, and Xo(N) is its projective completion. Over
C, Yo(N)(C) may be identified with the quotient of the Poincare upper half
68
plane by
ro(N) = {(:
~)
SL2(Z);C == 0 mod N};
the curve Xo(N) is obtained by adding the "cusps". The genus of this curve
tends to infinity with N.
Theorem 1. (Manin). Let p be a prime number and K a number field.

There is an integer n = n(p, K) such that the curve Xo(pm) has only a
finite number of K -rational points for m ~ n.
It is enough to prove that Xo(pn) has only finitely many K-rational
points for some n. Indeed, if m ~ n, there is an obvious map
which, from the analytic viewpoint, comes from ro(pm) being a subgroup
of ro(pn). This map may be described another way: to a pair (E, Cm),
where C m is a cyclic group of order pm, is associated (E, Cn), where Cn
is the cyclic subgroup of Cm of order pn. Thus if Xo(pn) has only a finite
number of rational points, the same holds for Xo (pm) for m ~ n.
More generally, for 0 ~ i ~ m - n there is an arrow fi : Xo(pm) - t
Xo(pn) given by
(E, Cm) t-+ (E/Ci , Cn+i/Ci ).
From the analytic point of view, fi (z) = pi z.
Let np be the least integer n such that the genus of Xo(pn) is at least
1, and let A be the jacobian of Xo(pnp).
[Valuesofnp : n2 = 5, n3 = 3, ns = 3, n7 = 2, nll = 1, n13 = 2, np = 1
for p ~ 17.]
For each m ~ n p , one then has a collection of morphisms, again denoted
by h:
h : Xo(pm) - t A, 0 ~ i ~ m - np'
Lemma. If w is a non-zero invariant differential form on A, the forms

ft(w), (0 ~ i ~ m - np) are linearly independent over C.
Proof of the lemma.
Expanding with respect to the variable q
= e2i1rz , one has
w = cqN dq + ... , c =1= 0;

as
fi transforms q to qPi, we have

ftw =
cNqNpi+(pi_l)dq
+ ... ;
69
the exponents are distinct for the different values of i; thus the orders of
the zero of the It w are distinct, whence the lemma.
From the lemma, the Ii are linearly independent over Z. Let PK be the
rank of A(K). If n ~ np + Pl(, then the number of Ii is m - np + 1 > PK,
and the Manin-Demjanenko theorem applies, which proves theorem 1.
Theorem 2. For given p and K, choose n = n(p, K) as in theorem 1. Then
there are only finitely many elliptic curves over K (up to K-isomorphism)
which have a K-rational point of order pn (where n = n(p, K).
Proof
Note first that pn ~ 4, since otherwise Xo(pn) would be a projective
line.
Let E be an elliptic curve over K having a K -rational point P of order
pn. Let C be the cyclic subgroup generated by P. From theorem 1, there
are only a finite number of pairs (E, C) up to K-isomorphism. Given two
pairs (E, P) and (E', P'), there is at most one isomorphism from (E, P)
to (E', PI) over K (because pn ~ 4). By Galois, such an isomorphism is
defined over K. Theorem 2 follows easily.
Corollary. There is an integer m(p, K) = m such that no elliptic curve
over K contains a point of order pm.
Proof
From theorem 2, it suffices to take pm strictly larger than the orders
of the p-torsion points of a finite set of elliptic curves.
Conjecture.("Theorem" of Demjanenko). Let K be a number field. The
order of the torsion group of an elliptic curve over K is bounded by a
constant depending only on K (and even only on [K : Q]).
Mazur [Mz2] has proved this for K = Q with an explicit list of possible
orders: 1, 2, 3, ... , 10, 12, 16 (those for which the corresponding modular curve
has genus zero).
[An even more optimistic conjecture is that if A is an abelian variety
over Q, one has
IA(Q)torsl $ C
where C depends only on dim A.]
70
Application to the Tate module of an elliptic curve.

Let E be an elliptic curve over the number field K, let p be a prime
number and
the p-adic Tate module of E. Put
Vp(E) = Qp Tp(E).
The group Gal(i?IK) acts on Vp(E).
Theorem. If the action of Gal( K I K) on Vp is reducible to triangular form:
(~
:), then E has complex multiplication.
[If Q( H) is the field of complex multiplication, one can show that
p splits in Q( H) and that the representation is a direct sum of two
representations of degree 1:
(~ ~), (see [Se3, Ch.
IV]).]
Proof.
Assume Vp(E) has a 1-dimensional subspace stable under Gal(KIK).
Intersecting this space with Tp(E), gives a Zp-subspace r of Tp(E), of rank
1, which is a direct factor as a Zp-module and is stable under Gal(K I K).
The image C n of r in Tp(E)lpnTp(E) = Ep .. (K) is cyclic of order pn,
and stable under Gal(K I K). Let Pn E Xo(pn) be the point of Xo(pn)
corresponding to the pair (E, Cn).
From theorem 1, if m is large then Xo(pm) has only finitely many Krational points. The points corresponding to (EICi , Cm+dCi ) are therefore
finite in number; in particular, there is i < j such that E' = E I Ci is
isomorphic to EICj = E'ICj_i ( where Cj_i is a cyclic subgroup of order
j - i of E'). We consider the projection E' --t E'ICj_i:
E' --t E'ICj_i = EICj ~ EICi = E'.
We then have an endomorphism of E' whose kernel is the cyclic subgroup
Cj-i' Hence E' has complex multiplication.
For another proof of this theorem, see [Se3, Ch.IV] as well as 8.5.
[This is now a special case of Faltings' theorems.]
Appendix. Belyi's theorem.

We have just seen that some special modular curves have few rational
points. A theorem of Belyi shows that general modular curves can be almost
anything. More precisely:
71
Theorem. (Belyi, [Be]). Let X be a non-singular projective curve over C.

The following properties are equivalent.
1) X is definable over Q (that is to say, X arises by extension of scalars
from an algebraic curve over a number field).
2) There is a subgroup r of SL 2 (Z) of finite index such that X is isomorphic to the quotient by r of the Poincare upper half-plane H compactified by adding the cusps.
3) There is a finite covering 1r : X -+ PI unramified outside {O, 1,00}.
[The proof will also show that if X is defined over Q, and S is any
finite subset of X(Q), the isomorphism
X ~ H/rU {cusps}
of 2) can be chosen such that S be contained in {cusps}.)
Proof.
3) => 1). This is a general fact: every covering of a Q-variety can be defined
over Q (see e.g. Grothendieck [SGAl]).
2) => 3). As r c SL 2 (Z), we have a covering H/r -+ H/SL 2 (Z) = PI {oo}; it is known that this is unramified outside 0, 1728, and 00; changing
coordinates gives a covering which is unramified outside 0,1 and 00.
3) => 2). Let r l = SL2(Z) and r 2 = (rI' r l ), the commutator subgroup
of r I . It is known that r 2 has index 12 in rI, that it is a free non-abelian
group of rank 2, and that
H/r 2 ~ PI - {a,b,c},
where a, b, c are three distinct points of PI (which we may choose to be
0,1,00). Hence r 2 is 7r1(PI- {a,b,c}), and every finite (connected, nonempty) covering of PI - {a, b, c} is of the form H/r, with r c r 2 and
[r2 : r] < 00, hence [r 1 : r] < 00.
1) => 3). There are two steps.

a) Every curve is a ramified covering of a projective line. Indeed, if C is a
curve, f any non-constant rational function on C, one takes f : C -+ Pl'
We select f defined over K C Q. Let S c P1(Q) be the finite set
of ramification points of f ("critical values"), i.e. the set of images of the
points of C where the differential is zero.
Theorem B. If S c PI(Q) is a finite set, there is a covering 3), by taking 7r = denote the union of (8) and the set of critical values of
. We begin with,
Special case: 8 C PI(Q). Let n = 181. If n :::; 3, there is an automorphism

of PI, defined over Q, such that (8) C {O, 1,00}. Suppose now that n ~ 4.
We shall construct : PI - t PI, defined over Q such that 84> c P 1 (Q) and
184>1 :::; n - 1; by induction on n, this will prove theorem B in this case.
Construction of . We may suppose that 0,1,00 and a are in 8, a
being a rational number distinct from 0,1,00. We seek in the form
where A and B are rational integers, A =1= 0, B =1= 0, A + B =1= 0. We shall

have then (0),(00),(1) E {O,oo}. The ramification is given by
d -_ A dz
+ B (z dz
-- ,
- 1)
i.e. z = Aj(A+B). We now choose A and B in Z such that a

Then
84> C {O, 00, (a)} U (S'),
= Aj(A+B).
where 8' = 8 - {O, 1, 00, a}, hence ( 8 f ) has at most n - 4 elements.

General case.
We shall deal with the general case by a double induction on the maximum degree over Q of the elements of 8 and the number of such elements
with this degree.
Let a E 8 be a non-rational point, and let an +aIan - 1 + ... +a n = be
its minimal equation over Q. We suppose that n is maximal for the points
of 8 and that the number of such a with this value of n is p. Take
(z) = zn + alz n - 1 + ... + an.

Then is defined over Q, and
84>
(8) U {oo} U (8')
where 8' consists of z E Q such that ' (z) = 0, hence of elements of degree
:::; n - 1. As the points of ( 8) have degree :::; n and the number of those of
73
degree exactly n is ::; p - 1, the induction process works and the theorem
follows.
Remark. Manin and Drinfeld ([Dr], [Ma3]) have proved that for a congruence subgroup of SL2 (Z), the cusps on the corresponding curve generate a
finite group in the jacobian. Belyi's theorem shows that this becomes false
for arbitrary subgroups of finite index.
5.5. The generalised Mordell conjecture.
We consider a group extension of an abelian variety by a group of

multiplicative type (a linear torus; the most interesting case is that of an
abelian variety). In other words, G is a connected commutative algebraic
group not containing Ga. Let X be a subvariety of G and r a subgroup of
G(C). We suppose that rankQr is finite.
Conjecture. ([L4]). There is in X a finite number of translates Xi of
subgroups G i of G such that
Consequence. If X contains no translate of a subgroup of dimension 2: 1,

then X n r is finite.
This hypothesis is satisfied for a curve X of genus 2: 2 embedded in its
Jacobian.
That r has finite rank over Q does not constrain its torsion. In particular, r can contain the full torsion group of G. More generally, if /'1, ... , /'n
are points' of G, we may take r to be all the 9 E G for which there is
m E Z, m =1= 0, with mg in the subgroup generated by /'1, ... , /'n. For n = 0,
r = Gtors . A consequence is then the following conjecture [now proved by
Raynaud [R3]].
Conjecture. (Manin, Mumford). A curve of genus 2: 2 embedded in any
abelian variety contains only finitely many torsion points.
[Lang's conjecture is now (1988) known in the following three special
cases:
(1) G is a linear torus (Laurent, [Lau));
(2) G is an abelian variety, and dim X = 1 (Raynaud [R2, Th. 0.3.2],
combined with Faltings [FJ);
74
(3) G is an abelian variety and
its torsion subgroup (Raynaud, [R3]; for

a different proof, based on Galois action, see Hindry [Hnl]).
In particular, the Manin-Mumford conjecture is proved (Raynaud [Rll,
[R2]).]
5.6. Mumford's theorem [Mul] ; preliminaries.
We begin by recalling standard results on divisor classes on a curve C

of genus 2: 1, and on its jacobian J.
Let C be a curve of genus 9 2: lover an algebraically closed field. Let
a be a divisor class of C of degree 1. We send C to its jacobian J by the
map
<p = <Pa : C - t J, P f-t Cl(P) - a.
Changing the class a alters <P by a translation.
If J d is the principal homogeneous space representing the divisor classes
of degree d, there is a canonical map C - t J 1 , and the identification of h
and J corresponds to a choice of a.
The map
C X .. x C = cg - t J
is surjective of degree g! and gives a birational map
The theta divisor is defined as the image of cg- 1 :

g-l
8 =
<p(cg- 1 )
= (L<P(Pi);Pi
C}.
i=l
This is a divisor of J, depending only on a (it is defined intrinsically in

J g - 1 ).
The divisor 8' = [-1]*8 is given by
(1)
a'
= -K + (2g - 2)a.
where K is the canonical class. [If one regards 8 as inside J g - 1 , then [-1]*8
is just 8 - K.]
The proof of (1) uses the Riemann-Roch theorem, cf. [W3].
It is convenient to choose a such that (2g-2)a rv K. Since the canonical
class has degree 2g - 2, and an abelian variety is divisible, this is possible
75
at the cost of a finite extension of the base field. It is unique up to the

addition of a torsion class, which is unimportant for the associated heights.
With this choice of divisor a, 8 is symmetric and the associated normalised
height is a quadratic form. Embedding C in its jacobian by 4>a then gives a
canonical height on C: on every curve of genus 2:: 2, there is one canonical
height.
We consider 8 as an element of Pic(J). For c E J

4>a-c : x ~ 4>a(x) + c. In Pic(C), we have
= PicO(C),
let
4>:_c(8') = -c + gao
An additivity argument shows that it suffices to prove this for a generic
We look for the x such that 4>a(x) + c E 8', that is to say
C.
g-1
4>a(x) + c = (x) - a + c "" - L(Pi) + (g - l)a.

i=1
Or again
g-1
(x)
+ L(Pi) "" -c + gao

i=1
As a is of degree 1, and c is generic of degree 0, the divisor -c+ga is generic

of degree gj there is then a unique way of writing it as Ef=1(Qi). In order
that 4>a(x) + c E 8' it is necessary and sufficient that x be one of the Q/s,
thus there are 9 possibilities. One checks that the 9 points of intersection
of the curve with 8 have multiplicity 1, thus
9
4>:-cC8') = L(Qi) '" -c + gao

i=1
In the particular case of the normalisation with (2g - 2)a = K, and c = 0,

we obtain
(2)
4>*8 =
gao
This shows that 8 intersects C in a divisor of degree 9 and that the canonical
height he on J induces on C a height equal to 2/=2hK + 0(1).
76
[Recall from 3.4 the following. If A is an abelian variety and X a

divisor, the image of X by the automorphism X r+ X + a of A is written
TaX. We define a homomorphism if>X : A ~ A' by sending a to the class
of TaX - X. Hence the divisor 8 defines a homomorphism if>e of J to its
dual J ' . It is known that the map if>e is an isomorphism which identifies J
and J'. This is equivalent to the self-intersection of 8 with itself 9 times
being gL]
Let us determine explicitly the Poincare class P E Pic( J x J) (3.4).
Consider the homomorphisms Pl, P2 and 8 from J x J to J defined by
Pl(X,y)=X,
P2(X,y)=y,
8(X,y)=x+y.
We have the formula
(3)
P =
p~8
+ p~8 -
8*8.
This shows in particular that P is symmetric.

We check formula (3) by using the characterisation of the Poincare
divisor given in 3.4.
Put P l = pr8 + P28 - 8*8.
As x r+ TxX - X is a homomorphism in x, we obtain
That is to say, Pl/ J x {x} = x E Pic( J), since J is identified with its dual
via if>e. As P l induces 0 on {c} x J, this implies P = Pl.
The morphism if> x if> sends C x C to J x J. The inverse image of P is
given by the formula
(4)
(if> x if* P '" 6. - C x a - a x C,
where 6. is the diagonal, and where we write (by abuse of notation)
Proof of (4). First, if X and Y are non-singular projective irreducible

connected varieties, and if c E Pic(X x Y) induces zero on all the verticals
and on one horizontal, then c = 0 (the analogue of - but distinctly simpler
than - the theorem of the cube; see [Mu2, p.54]).
77
Now, by symmetry it suffices to restrict (4) to a vertical. Let Xo E C,
and let f : C - t J x J:
f:C-tCxC-tJxJ
X l-+ (x,xo)
As P
l-+
(rP(x),rP(xo)).
= pre + P2e - s*e, we have
But rP*8 '" ga, (from (2)), therefore
f* P
'" ga + 0 - (ga - rPa(XO)) '" (xo) - a,
as rPa(XO) '" (xo) - a.

The right hand side of (4) gives (xo) - a on C x {xo}. Whence the
formula (4).
5.7. Application to heights: Mumford's inequality.
Let K be a field equipped with a product formula, i? its algebraic
closure. We select a (by extending the base field) such that (2g - 2)a = K,
where K is the canonical class, and we embed C in J via rP = rPa. As
e = e', he is a quadratic form on J (K) / torsion. On the real vector
space VR = J(i?) R, this gives a positive quadratic form which we write
hex) = Ix12. We write x.y for the associated scalar product:
1
x.y = 2(h(x + y) - hex) - heY))

Mumford's main result is the following.
Theorem. There is a constant Ml (C) depending only on C such that if
x, y E C(k) with x ~ y, then
78
Proof
On J x J, the Poincare divisor P defines a normalised height and we
have
hp(x, y) = -2x.y
from the last theorem of 3.4 with c = e, (with our conventions <Pe(y)
We use (4) of 5.6 : on C x C, we have
= y).
hp(x,y) = h.t:.(x,y) - ha(x) - ha(y) + 0(1).

But h.t:.(x, y) ~ 0(1) outside ~ (2.10), and as e induces ga on C (from
(2) of 5.6), h induces gha on C; therefore if x i= y
1 2
1 2
-Ixl
+ -IYI
-
2x.y = ha(x)
~
+ ha(y) + hp(x,
y)
h.t:.(x,y)+O(I)
~
0(1).
As the statement of the theorem is uninteresting for 9 = 1, we assume

from now on that 9 ~ 2.
Corollary. There is a constant M2 = M 2(C) such that if x and y are
points of C(k) with M2 ~ Ixl ~ IYI ~ 21xl and x i= y then the angle Ox,y
between the vectors x ,y in VR is ~ i.
Proof We write x.y
Take M2
= y'1OM1 .
= Ixl.lyl cos(O); as 9 ~ 2 we have
Then
cos 0
But
7r
cos 6" =
3
- 4
Ml
< - +-
85
.
IxF <
- -4 + -10 = 100
2v'3 = 1.732 ... /2 =
0.86 ...
therefore 0 > i.
[Note that if 9 ~ 3 one even obtains 0 ~
> 85/100 ~ cosO,
f.J
79
We call a point x E C(K) "large" if
Ixi =
h(x)I/2 > M 2.
The space VR has infinite dimension. It is easily seen that the points
x E C(f() where f( is a number field lie in a finite dimensional subspace of
VR.
For an angle 8 with < 8 < 1f/2, denote by N(n,8) the maximum
number of points Xl, ... ,XN in the unit sphere of the Euclidean space Rn
whose mutual angles are ~ (). For e = 1f/3, there is a sphere packing
interpretation: N (n, 1f /3) is the maximum number of mutually disjoint unit
spheres tangential externally to a given unit sphere ("kissing number").
This is an interesting function of n. For small values of n, one has
N(1,1f/3)
= 2;
N(2,1f/3)
= 6;
N(3,1f/3)
= 12,
N(4,1f/3)
= 24 or 25.
Odlyzko and Sloane have shown that N(8,1f/3) = 240, and N(24,1f/3) =
196560. For n = 16, it is only known that
4320 ::; N(16, 1f/3) ::; 8313.
We use the following rough bound.
Theorem. N(n,
i") ::; 5n .
Proof.
Let Xl, ... , XN, with N = N(n, i"), be points of the unit sphere of Rn,
n ~ 1, whose mutual angles are ~ i". We remark that the spheres with
centres Xj and of radius i do not intersect.
For this, it suffices to see that sin ::z > i; indeed, we have
.
SIn -
1f
12
1 sin i"
2 cos 12
---7r-
1 1
4 cos 12
---7r-
1
4
> -.
The unit sphere having a volume v, say, the spheres of radius ~ have
volume (~)nv and are all contained in a sphere of radius 5/4, and of volume
(5/4)N v. Thus N.(i)n v < (5/4)nv and therefore N < 5n .
80
Theorem. Let PK = rank J(K). Then the number of large points x of
C(K) such that Ixl ::; A (with A ~ M 2 ) is
< 5PK (1 + log(A/M2 .

-
log 2
Indeed, let n be such that 2n M2 ::; A < 2n +1 M 2. If x, y are distinct points of C(K), such that Ixl and Iyl belong to the same interval
(2i M 2 , 2i+1 M 2 ), 0 ::; j ::; n, the corollary above shows that the angle Ox,y
between these points is ~ tr/6; consequently, the number of such points is
::; 5PK , by the previous theorem. Since there are n + 1 intervals, the total
number of large points with Ixl ::; A is
Corollary. The number of points x of C (K) with h( x) ::; A is

log A
::; 5PK 2 log 2
+ const., as A
-+ 00.
Therefore the number of points of C( K) of (exponential) height H (x) ::;

A is ::; o (log log A).
Remark. Indeed, the result obtained [although superseded by Faltings for
number fields] is optimal for function fields over finite fields.
Let Y be an algebraic curve over F q, K = F q(Y), and C a curve defined
over F q' The elements of C(K) are morphisms defined over F q of Yin C;
for I E C(K), the height of I is
h(f) =
1/12 =
A deg I
The constant morphisms (which are finite in number) are the f's of degree
zero. If I is a non-constant point, III i= 0, the composites with the Frobenius
are:
endomorphisms
tr;
Then deg In =
~ logA.
qn deg I,
and the number of points
In
of height ::; A is
81
6. LOCAL CALCULATION OF NORMALISED HEIGHTS
Neron [N4] has shown that the canonical height over a number field can
be broken as a sum of local heights. This leads to the question of calculating
these local heights both for the prime numbers and for the infinite primes.
Tate [T2] has made this theory explicit for the case of elliptic curves in
two letters to Serre; the first letter has been reproduced by Lang [L5] and
taken up again by Zimmer [Z].
We only sketch the results; for complete proofs the reader should look
at [N4] or [L5].
6.1. Bounded sets.
Let [( be a complete normed field (K will be R, or C, or a complete

field for a real valuation).
We write
veX) = -log Ixlv.
Let V be an algebraic variety over K. Besides the Zariski topology
on V(K), we have a topology induced by that of K, which is called the
K -topology.
We now define the notion of a bounded subset of V(K), first in the
affine case, then in the general case.
(1) Assume that V is affine. The affine algebra of V is of finite type over
K; we write this as
K[V] = [([II, ... , 1m].
A subset B of V(K) is called bounded if li(B) is bounded in K for all i.
This condition is independent of the choice of generators II, ... , 1m of K[V].
(2) General case. Cover V by a finite number of open affines: V = Ui Ui.
A subset B of V(K) is called bounded if there are Bi C Ui(K) which are
bounded and B = Ui Bi.
Definitions (1) and (2) are readily seen to be compatible.
When K is locally compact, bounded is equivalent to relatively compact for the K -topology.
Projective space is bounded (and so also is any projective variety). To
see this take homogeneous coordinates (xo, ... , x n ), and define
82
The Wi cover P n/ [( and Wi is bounded in the open affine given by Xi i= 0:

in affine coordinates Xj/Xi, (j i= i, 0 :S j :S n), Wi is the unit cube.
For example, the projective plane is a union of 3 bounded subsets.
Here are some elementary properties of bounded sets:(1) Let f : V --+ W be a morphism of K-varieties. If Be V(K) is bounded
then feB) is bounded in W(K).
(2) Let V be a closed subvariety of W and Be V(K). Then B is bounded
in V(K) {:} B is bounded in W(K).
(3) Let f : V --+ W be a proper morphism of quasi-projective varieties. If
Be W(K) is bounded then f-l(B) is bounded in V(K).
(4) Let V be an affine variety, W a closed subvariety of V and V' = V - W.
Let <Po be a finite set of generators of the ideal defining W. Then
Be V'(K) is bounded in V'(K) if and only if
a) B is bounded in V(K),
and
b) info I/I<Pol is bounded on B.
(5) Let V be a closed subvariety of P n and W a closed subvariety of V
defined by homogeneous equations <Po(xo, ... , xn) = 0 of degree do and
V' = V - W. If x E V(K), <Po i= 0, put
bo(x) = sup Ixt<> l<Po(xo, ... , xn)l;
If x E V'(K) put b(x) = info bQ(x). A subset B C V'(K) is bounded

if and only if the function b is bounded on B.
Exercise. Assume that K is a complete field with respect to a discrete

valuation; let 0[( be its ring of integers and 7r a uniformizing element.
Let V be a closed subvariety of affine n-space An and V its Zariski
closure in P n . Suppose V(K) = V(K), i.e. V has no K-point at infinity.
Prove that V(K) is bounded in An. [When K is locally compact, this
83
follows from a compactness argument. In the general case, use a theorem
of Greenberg [Gr] according to which an a K-scheme of finite type has an
a K-point if and only if it has an a K / '!rna K-point for every n ;::: 1.]
6.2. Local heights.
Let V, J< be as above.
Definition. We say that a real valued function f on V(K) is strongly continuous if it is

1) continuous,
2) bounded above and below on all bounded subsets of V(K).
When K is a locally compact field, the first condition implies the second, thus strongly continuous is equivalent to continuous (this is false if J<
is not locally compact).
If a function on V(J<) induces strongly continuous functions on an
affine open covering, then it is strongly continuous.
Assume now that V is quasi-projective and D is a Cartier divisor of
V with support Supp D (if V is non-singular, D = I:nw.W where W has

co dimension 1, cf. 2.7).
It will be convenient to assume that V (J<) - Supp D is dense in V (J<),
which is true if K is algebraically closed (which is a harmless assumption)
or if V is smooth (for example an abelian variety).
The local height is a real valued function
),: V(K) - Supp D
-t
R,
which satisfies the following condition. If U is an open affine of V on which

D is a principal divisor (</J), where </J is a rational function on U, then the
function
X 1-+
),(x) - v(</J(x,
defined on U(J<) - Supp D, extends to a strongly continuous function (in

the above sense) on U (K).
If D is a positive divisor with local equation <p, then v( <p) = -log 1<p1
tends to 00 near the divisor, and ), is the sum of v( <p) and a strongly
continuous function.
We say that), is associated to D. If ), exists, it is unique up to the addition of a strongly continuous function on V (the difference of two such ),'s
is a strongly continuous function on a covering, thus is strongly continuous
on V).
84
Proposition. Local heights exist.
Example. If V = P n , D is the hyperplane t = 0 in P n , where

n
t=
LUiXi,
i=O
(for instance t = xo), then
is a local height associated to D.
Proof of the proposition.

By enlarging K we may assume that it is algebraically closed.
Notice that
a) if D is a principal divisor (4)) we may take AD = v(4)),
b) if D = Dl +D2 ru:,.d AD}) AD2 are associated with Db D 2, then ADl +AD,
is associated to Dl + D 2.
It is known that the line bundle defined by D is a difference of ample
line bundles (see the proof of the theorem in 2.8). Hence using a) and b)
one is reduced to the case where D = j-l(H), for f : V ~ P n , H being a
hyperplane of P n. In that case, one may take AD to be AH 0 f where AH is
a local height attached to H, see above.
Examples.
1) Trivial examples.
a) If D is a principal divisor, D = (4)), take A = v( 4>( x
Even if the variety
is projective, A is not canonical, for 4> is only defined up to multiplication
by a constant.
b) Let D = Dl + D 2. We suppose that ADl and AD, are associated to
Dl and D2, respectively; then ADl + AD, is defined outside Supp Dl U
Supp D2 :) Supp D, and extends by continuity outside of Supp D to a
function A attached to D.
2) Over C. A Cartier divisor is the same as a line bundle together with

a meromorphic section of this bundle (2.7). Put a hermitian structure
on this line bundle (first locally, then use a partition of unity); this gives
a measure of the length of sections of this bundle. The absolute value of
a section gives a real-valued function on V(C) - Supp D (which one can
85
extend by 0 at the zeros of the section and which is not defined at the
poles). A function A satisfying the definition is
A(X)
= -logls(x)l.
3) K locally compact. The above construction can be imitated. We have

a line bundle whose structural group is the group K*; the absolute value
1.1: K* -+ R* then gives a topological line bundle with R+ as its structural
group (in the ultrametric case 1(* is totally disconnected and the bundle is
trivial). Put a hermitian structure on this bundle, as above, and take the
same construction.
4) Affine varieties. Let V be an affine variety. A positive divisor D corresponds to an ideal d of the affine algebra K[V). Let (QI, ... , Qn) be a
system of generators of d. We put
which makes sense except if all the Qi(X) are zero, that is to say outside of
Supp D. To check that A is associated to D, we must show that if D = (<fJ),
then A(X) - v(<fJ(x is regular.
But if D = (<fJ), we have Qi = Pi<fJ, and the ideal generated by the Pi
is 1: L-i RiPi = 1; that the function infi v(~) is regular is then a simple
exerCIse.
5) Projective case. We embed the variety V in P n with homogeneous coordinates (xo, ... , x n ). The positive divisor D is defined by an homogeneous ideal
d, generated by a system of homogeneous elements (Po), with degPo = do.
We then put
One checks that multiplying the Xi by a constant leaves A(X) invariant and
that A(X) is defined outside Supp D.
That A is associated to the divisor D is seen by examining each open
affine Xi 1= 0; the change to affine coordinates recovers the formula of the
affine case.
6) Local heights as intersection multiplicities. Suppose that K has a discrete

valuation and let the ring of integers be OK. Take a scheme V projective
86
and smooth over OK, with absolutely irreducible fibres, and denote by V
the variety over K derived from this scheme (i.e. V is the generic fibre of
V).
~D
generic point
closed point
By taking the Zariski closure, a divisor D on V extends to a divisor D

on V and in the same way a point x E V (K) extends to an integral point
x E V(OK)'
Assume that x tf. Supp D. Define m(x, D) as follows:
a) If x does not intersect D, then m(x, D) = o.
b) If x does intersect D in a point Xo of the special fibre, then m(x, D)
is the intersection multiplicity of x and D and Xo (computed on Vj
note that if V has dimension n, then V has dimension n + 1, D has
dimension n, and x has dimension 1).
[One way to compute m(x,D) in case b) is to write D as () at Xo,
for some , and put m(x, D) = v((x. Another way is to view x as a
morphism Spec( 0 J() -+ V and take the inverse image of D, which is an
integral multiple m(xo) of Xo; then m(x, D) = m(xo).]
We thus get a function x I-t m(x, D), and one checks that it is a local
height associated to D.
Exercise. Let V be a variety, U = {U1 , . , Un} a finite open affine covering
of V, Ai the affine ring of Ui, <l>i a finite subset of Ai containing 1 and generating Ai as a K-algebra. Assume that K is complete for a non-archimedean
absolute value 1.1.
(1) If x E Ui(K), define Ei(X) = sUP4>E IP i 1(x)l. Define Wi C Ui(K) as the
set of x E Ui(K) such that Ei(X) :s: Ej(X) for all j such that x E Uj(K).
Prove that Wi is open and closed in X (K) (use the properties of bounded
sets in 6.1).
(2) If x E V(K), define i(x) to be the smallest i with x E Wi, and let ni =
{xli(x) = i}. Then V(K) is the disjoint union of the ni , which are op~n and
closed in V (K). Show that if i, iI, ... , j k are any indices and B is bounded in
UiI (K) n ... nUjk(K), with respect to Ujl n ... nUjk' then ninB is bounded
87
with respect to Ui n UiI n .. .Ujk
(3) Show that the cohomology groups Hk(U, S) are zero for k ~ 1, where
S is the sheaf of strongly continuous functions.
(If (fio ... ik) is a k-cocycle, show that it is the coboundary of 9i l ...ik
defined by
9i l ... ik(X) = fiil ...ik(X), where i = i(x).)
(4) Use the vanishing of HI(U, S) to give another proof of the existence of
local heights.
In particular, if D is a positive divisor on V such that D = (fi) on each
Ui (Ji E Ai), show that AD can be taken to be
AD(X)
= v(fi(x)),
where i
= i(x).
6.3. Neron's theorem.
We now consider abelian varieties. Here is a first normalisation result.

Theorem. (Neron [N4]). To all pairs (A, D), where A is an abelian variety
over K and D a divisor on A, there is an associated local height
AD: A(K) - Supp D
-t
R,
(defined up to the addition of a constant function), such that:

(1) If D = (), AD(X) = v((x)) up to a constant.
(2) ADl + AD2 = AD l +D2 modulo constants
(3) If 7r : B - t A is a morphism of abelian varieties and if DB = 7r* D is
defined (i.e. if7r(B) ct. Supp D), then A1r*D = AD 0 7r up to a constant (we
do not assume that 7r maps 0 to 0).
Proof (abridged).
We may suppose that the class of Din Pic (A) is symmetric (resp. antisymmetric). Under the map x I-t 2x on A, D becomes (cf.3.2 Corollary)
2* D
= rD + (),
with r = 4 (resp. 2)
where is a rational function. If f.t is a local height associated to D, we

have
f.t(2x)
= rf.t(x) + v((x)) + t(x)
where t(x) is strongly continuous on A(K) (the two local heights f.t(2x) and
rfl(x) + v((x are associated to 2*D and rD + (), respectively). From
88
a general fact on Banach spaces (cf. 3.1, Exercise) there is a strongly
continuous function 'fJ such that
= 'fJ(2x) - {~} 'fJ(x).
Put A = J.L - 'fJ.
This is then a local height associated to D, and

A(2x) = rA(x)
+ v(4)(x)).
(The normalised function to be found had to satisfy this formula.)

One then shows that this is the appropriate local height. The uniqueness up to the addition of a bounded function (this ambiguity arises from
the choice of 4 results from the following elementary lemma.
Lemma. Ifv(2x) = rv(x) and v is bounded, then v = O.

We call the function AD of the theorem the normalised local height
(even so, it is only defined up to an additive constant).
Example 1. Computation over C via theta functions.
Let A be an abelian variety over C. This is a quotient of a C-vector
space V by a lattice r (cf. 3.2). A "theta function on A" is a meromorphic
function 0 on V which satisfies the identity
O(x +,) = O(x)e 2 11"i{L(x,'Y)+H(-y)},
, E r,
where L is C-linear in x (further normalisation is possible).

The divisor of 0 is invariant under translation by r hence may be
identified with a divisor of A.
Theorem. Every divisor of A is a divisor of a theta function, defined up

to multiplication by a trivial theta function eQ(x) (where Q is a quadratic
polynomial in x).
(For the proof, see e.g. [Sw, Theorem 18] or [Mu2, Chapter1].)
Calculating O(x + Ct' + (3) in the two obvious different ways gives some
relations between the exponents. This type of argument shows that the
function L( x, ,) extends to a function L( x, y) linear in y, and that there is
a R-quadratic polynomial R in x such that the function
O(x) = IO(x)leR(x)
satisfies O(x+,)
We then put
= O(x).
Hence 0 is a positive real valued function on A(C).
A(X) = v(O(x.
89
It is defined outside D
div(O), and one checks that it has properties

(1),(2),(3) of Neron's theorem with respect to D; hence it is the normalised
local height relative to D.
Example 2. Good reduction.

Suppose that K has a discrete valuation v and AI K has good reduction
at v. Then the closed fibre of the Neron model A over OJ( is again an abelian
variety and the intersection multiplicity AD(X) = m(x, D) (cf.6.2.(6 is a
normalised local height.
6.4. Relation with global heights.

As in 2.1, suppose K is a field equipped with a product formula with
local exponents nv'
Let D be a divisor on A and hD the corresponding canonical height.
Theorem. There is a choice of AD,v where AD,v is a normalised local height
associated to D over the local field K v, such that for all x t/ Supp D
where for each x all except finitely many terms of the sum are zero.
We only sketch the proof; for a more complete account see [L6, Chapterll].
The abelian variety has good reduction at almost all places. For such
places, extending the divisor so that it has no component on the special
fibre gives a canonical AD,v. We have AD,v(X) = 0 for almost all v. For the
finite number of bad reduction places v, we choose any AD,v, and we must
show that the formula is correct up to the addition of a constant.
Let x E A( K); select cp such that the divisor D' = D + (cp) does not go
through x, and let AD',v be the unique normalised local height such that
AD',v = AD,v
+ v(cp),
outside Supp D U Supp D'.
Put
v
it does not depend on the choice of

can take cp = 1 hence
h'(x)
=L
v
cpo
In particular, if x
nvAD,v(X),
t/ Supp D.
t/
Supp D, one
90
By changing one of the AD,v, it may be assumed that h'(O) = O.
Then one checks:
a) hI is a global height relative to D (i.e. it is a "he" attached to the class
c of D, cf. 2.8),
b) h'(2x) = 4h' (x) if D is symmetric (resp. h'(2x) = 2h' (x) if D is
antisymmetric) .
The theorem follows.
6.S. Elliptic curves.
Let E be an elliptic curve over a local field K and D a divisor on E.
To define a local height A associated to D, by translation and additivity we
may reduce to the case where D = (0). The function A will be continuous
except at 0, and bounded outside every neighbourhood of O. Further, if z
is a local coordinate at 0,
A(P) - v(z(P))
-t
const.,
as P
-t
O.
We shall completely normalise A. There are two ways of doing this, both
suggested by Tate.
First normalisation. Select a differential form of the first kind w =I- 0 on E
and choose z such that w = dz+(terms of higher degree) about O. We then
demand that,
A(P) - v(z(P)) - t 0, as P - t O.
This depends only on w, not on z. The corresponding local height will be
denoted Aw.
Second normalisation. Choose w as above; let .6. be the discriminant of the
corresponding Weierstrass equation. Put
1
A(P) = Aw(P) + 12 v(.6.).

One checks that A is independent of the choice of w (it is equal to Aw when
w is such that .6. is a unit).
There are three types of computations to do.
91
1. Non-archimedean calculations; there are two cases:
- Curves with good reduction. In this case one has A( P) = 0 if the
reduction P of Pis i 6, and A(P) = v(z(P)) if P = 6, where z = x/y
is a local parameter normalised with respect to the differential form.
- Tate curves, with multiplicative bad reduction.
These curves are dealt with by means of theta functions. In the Tate
model, the curve is expressed as E = Gm/qZ with 0 < Iql < 1. The function
O(t)
= (1 -
IT (1 - qnt)(l - qnc
00
t)
1)
n=l
satisfies the identity
O(qt)
= -~O(t).
t
As in the complex case, one corrects the absolute value of 0 by a quadratic

polynomial
a 2 -a
Aw(t) = v(O(t)) + 2 v(q)
where a
= v(t)/v(q), for the natural choice w = dt/t.
we have v(.6..)
As
= v(q), and then the absolute A is

A(t)
= v(O(t)) + i(a2 -
a + 6)v(q).
i)
Note that (X2 - X +

is the second Bernoulli polynomial B 2(X), which
is an interesting feature of this normalisation.
2. Calculation over C.
One possible method is to use a theta function, for example the Weierstrass sigma function (where we use the standard notation)
cr(z)
= z IT (1 w:;tO
Then
Z.
- )e w
W
+1 .2
2"-;;;T.
Aw(Z) = -log Icr(z)1 + iRe(Z1J(z)),
92
where TJ is the R-linear function C - t C which transforms the fundamental

periods Wl, W2 to TJl, TJ2 and W is the standard differential form dz.
Practical computation by this method is inconvenient. But there is a
better one, due to Tate.
With the usual notation:
we take
One has
)"w(P + Q)
+ )"w(P -
Q) = 2)"w(P) + 2)..w(Q)
)..w(2P) = 4)"w(P)
+ v(x(P) -
x(Q,
+ v((2y + alx + a3)(P,
and for the canonical )..
)..(P + Q)
+ )..(P -
)..(2P)
Q)
= 2)"(P) + 2)..(Q) + v(x(P) - x(Q -
= 4)..(P) + v((2y + alx + a3)(P -
6v(~),
4v(~).
These formulae follow from an equivalence of the corresponding divisors,

for example:
shows that the formula for )..w(2P) is true up to a constant; checking the
behaviour of each side near 0 shows that the constant is zero.
Tate's recipe for numerical calculation is the following. Put
so that the equation of the curve becomes
and define Z by
93
that is to say
The formula
>.(P) =
2 log Ix(P)1 + 8 L
1
00
n=O
1
4 n log IZ(2n P)I
has been proved by Tate (see [T2] and [Z]) in the case when K = R and all
real points on the curve satisfy x > O. In this case Z is a rational function
without zeros or poles on the real points, thus its absolute value is bounded
above and below. Then log IZ(2n P)I is bounded in absolute value and the
series converges rapidly.
In the general case, Z has zeros and poles, and 2n P can approach a
singular point. Nevertheless, one conjectures that this formula (*) remains
valid [it has been proved by Masser [Mas] in the case of complex multiplication using elliptic analogues of Baker's theorem].
94
7. SIEGEL'S METHOD
In the next two chapters we study integral points on curves, first with
Siegel's (ineffective) method and second with Baker's effective method of
linear forms in logarithms.
7.1. Quasi-integral sets.
We begin with a particular case. Let F(x, y) = 0 be an irreducible plane

curve X over Q. Siegel's theorem says that the integral points (x, y) E Z x Z
on this curve are finite in number, except when the curve has genus zero
and has either 1 or 2 points at infinity.
Here are examples of such exceptional curves. The curve y = 0 is an
affine line having infinitely many integral points x E Z; it has one point at
infinity.
The conic y2 - Dx 2 = 1, where D is a non-square positive integer, has
two points at infinity. It is well known that this Pell-Fermat equation has
infinitely many integral solutions.
Over an arbitrary number field, another exceptional case arises: xy =
1, where the integral points correspond to units in the number field.
To give the general statement, we take a number field K (of finite
degree over Q), S will be a finite set of places of K containing the infinite
places, and A = As denotes the ring of S-integers of K:
As =
{a E K; lal v
1 for all v S}.
Let X be an affine variety over K, Ax = HO(X,Ox) the affine algebra

of X, and M a set of rational points of X: Me X(k). We shall say that
M is quasi-integral relative to the ring A if for all I E Ax there is a E K*
such that I(M) c aA (that is to say, the 'denominators of I(x), x E Mare
bounded').
It suffices to check this condition for a set of generators of the K -algebra
Ax. In particular, if we select an immersion of the variety X C An, with
some coordinates Zl, ... , Zn, to say that M is quasi-integral means that the
z-coordinates of the points of M have a common denominator.
We can also reduce this to the notion of an integral point on a scheme.
The following properties are equivalent:
(i) The set M is quasi-integral relative to A.
95
(ii) There is an A-scheme X of finite type such that,
(1) X = X XA K,
(2) every point x of M extends to an A-valued point of X.
(iii) There is an affine A-scheme X of finite type satisfying (1) and (2).
To show that (i) :::> (iii), we choose an immersion, multiply the coordinates by a common denominator, and take the A-subalgebra generated by
these.
To check that (ii) :::> (i), we take f E Ax; then there is a E A, a =1= 0,
such that af extends to X; then the values of af at integral points are
integers.
The implication (iii):::>(ii) is trivial.
We now assume that X is an affine absolutely irreducible curve over
K. We complete the normalisation of X and we then obtain the unique
normal complete (projective) model X of X. We denote by Xoo the points

at infinity of X; thus X - Xoo is the normalisation of X. We say that X
is exceptional if X has genus 0 and X 00 has at most two geometric (that is
to say, after passing to an algebraic closure of K) points.
Siegel's Theorem. ([Si]). If the affine curve X is not exceptional, every

subset of X(K) which is quasi-integral relative to As is finite.
Here is the plan of the proof.
1). Review of the approximation of irrational numbers (essentially, the
Thue-Siegel-Roth theorem)
2). Approximation of rational points on abelian varieties and on curves of
genus ~ l.
3). Siegel's theorem for genus ~ 1.
4). The case of genus 0 and at least 3 points at infinity.
7.2. Approximation of real numbers.

1). For all real irrational numbers a, there is a sequence Pn/qn of rational
numbers with
Proof. Let Q be an integer > 1. We consider the numbers qa, 1 S q sQ.

By the box principle, two of them, modulo 1, have difference S 1/Q. By
subtracting, there is q, 1 S q Q, such that IIqall l/Q (where 1111 is the
96
distance to the nearest integer). In other words, there is p E Z such that
Iqa - pi:::; I/Q. Therefore, la - p/ql :::; q~ :::; ~.
As q~ -+ as Q -+ 00 we obtain then infinitely many such Pn/ qn'
One can improve this statement and replace l/q~ by 1/...(5q!, using
continued fractions, and show that ...(5 is the best constant (see [HW, Theorems 193,194]).
We say that a is approachable to within 1/q6 if there is a sequence of

rational numbers Pn/qn, qn ~ 1, with la - Pn/qnl :::; l/q!.
2). H 6 > 2, the set of real numbers a approachable to within l/q6 has
measure zero.
: :;
Proof. We can limit ourselves to the segment [0,1]. For integers P and q
with
P:::; q, q> 0, we denote by Ip,q the interval
Ip
1
[-p
-,
+
-]
n [0,1].
q q6 q
q6
Let Q be an integer
> 1 and let

IQ
q>Q,05p5q
Ip,q'
Let a E [0,1] and a IQ, then a is not approachable to within 1/q6. But
meas(IQ) :::; O(L q6) = O(L 6~1)'

q?Q q
q?Q q
As 6 > 2, the measure of IQ tends to zero as Q tends to infinity; therefore
the intersection of the IQ, Q > 1, has measure zero.
The above result remains true with 1/q6, 6 > 2, replaced by l/q1fJ(q),
where the series E 1/1fJ(q) converges.
3). Roth's Theorem. If a is an irrational algebraic number, then for

every 6> 2, a is not approachable to within 1/q6.
In other words, the set of pairs (p,q), q ~ 1, with la - p/ql :::; 1/q6 is
finite. But one does not know how to bound q as a function of a and 6. In
contrast, the number of solutions can be effectively bounded (cf. [DR]).
97
Here is a more general statement for number fields. Let K be a number

field. We choose a place v of Kj we denote by a 1-+ lal v the corresponding
absolute value. There is a product formula on K compatible with the choice
of Vj let H be the corresponding height on K (cf.2.1,2.2).
Let a E Kv be algebraic over K. Let 8 > 2. Then there is no sequence
Wn of distinct elements of K such that
For K = Q and v the archimedean absolute value, this is Roth's theorem. There is a more general statement using several places v and the
product Ilv la v - wnl v (cf. [L2] ,[L6], [Schml], [Schm2] ,[Mi]). For the proof
of Roth's theorem and its variants, we refer to Lang (loc. cit.).
Application to integral points.

Thue observed the following relation between diophantine equations
and rational approximations. We take a sequence of solutions (x, y) E Zx Z,
tending to infinity, of the equation x 3 - 2y 3 = k (k E Z fixed). Then
xjy -+ {/2. More precisely, (xjy)3 -2 = kjy3, therefore Ixjy- {/21-< kjy3.
Hence, an explicit bound on the approximations to {/2 provides an effective
bound for the solutions of x 3 - 2 y 3 = k. For example, Baker has shown
that for all pjq E Q, q > 0, one has
3
1{12 - ql ~
10-6
q2.955 '
which implies that every solution (x, y) E Z x Z of x 3
2 y3 = k has
The exponent 2.955 of Baker has been improved to 2.43 by Chudnovsky

[Cu]j so for the equation x 3 - 2 y 3 = k, we obtain
where c is effectively computable.
98
7.3. The approximation theorem on abelian varieties.
The argument above is insufficient for equations such as y2 - X 3 = k.
Siegel's method consists of refining the approximation theorem on abelian
varieties.
Let V be an algebraic variety over K, immersed in projective space
P N. Let K v be a completion of K. We are interested in the approximation
of algebraic points 0:' E V(Kv) by Wn E V(K), with Wn --t 0:' in the v-adic
topology.
To define a distance on the variety V at the point 0:', we choose a
neighbourhood U of 0:' with coordinates Zl, ... , ZN taking the value 0 at 0:',
and we define
d (
v 0:',
(3) - {SUP19$N IZ i(3)lv if (3 E U,

1,
if (3 U.
The logarithm of the distance is independent, up to a bounded function, of

the coordinates chosen and the immersion of V in P N. (One could define
in the same way, up to 0(1), the logarithm of the distance to a subvariety.)
Theorem. If 0:' E V(Kv) is algebraic (i.e. 0:' E V(K) and if b
is no sequence of distinct elements Wn E V(K) such that Wn
>
2, there
--t 0:' and
dv(O:',w n )::; l/H(wn )c.

This follows from Roth's theorem applied to the coordinates of W n .
When V is an abelian variety, this can be improved:
Theorem. If V is an abelian variety, 0:' E V(Kv) is algebraic, and b > 0,
then there is no sequence of distinct elements Wn E V(K) with Wn --t 0:' and
1
dv(O:',w n )::; H(wn)c

Remarks. 1). This theorem is also true for any variety embeddable in
an abelian variety, thus in particular for curves of genus 2: 1 as they are
embedded in their jacobians.
2). The theorem can be restated as:
if Wn
--t 0:',
then log d (1
0:', Wn
) = o(logH(wn)) .
One can ask whether o(logH(wn may be replaced by a smaller estimate.

This is indeed true. It has been proved by Masser, Wiistholz, Philippon,
Waldschmidt ... that
1
c
log d (
) = O((1og log H(w n
v 0:', Wn
),
99
for a suitable C (cf. [Ber]; it is not known whether C can be taken equal
to 1). Moreover the constants involved are effective, provided a set of
generators of the Mordell-Weil group of V over K(a) has been found.
Proof of the approximation theorem on abelian varieties.

We select a projective embedding of the abelian variety and we take
the height H relative to this embedding.
Let 8 > 0 and let m be an integer ~ 1 with (m 2 - 1)8 ~ 3. Let
(w n ) be a sequence of distinct elements of r = V(K) approaching a to
within 1/H (w n )6 . From the weak Mordell-Weil theorem, r / rnr is finite.
By taking a subsequence of (w n ) we may assume that they are in the same
class modulo mr: there are Jr E r and w~ E r with
Wn
= mw~ + Jr.
As V(Kv) is a compact group, again taking a subsequence we may assume

that w~ converges to a point a' E V (Kv). Then a = rna' + Jr, therefore a'
is algebraic.
As the morphism W 1-+ rnw + Jr is etale, we have
On the other hand, as n
-t
00,
since logH is quadratic up to a bounded function (cf.3.3); therefore for all

sufficiently large n we have
Then
dv(a',w~) ~ 1/H(w~)6(m2-1) ~ 1/H(w~)3,
which contradicts the previous theorem since 3
> 2.
Remark. 3). The approximation theorem on abelian varieties is ineffective

because its proof uses Roth's theorem and the Mordell-Weil theorem, which
are both ineffective at present. If the Mordell-Weil theorem could be made
100
effective, one could dispense with Roth's theorem by using the abelian analogue of Baker's lower bounds for linear forms in logarithms, see Remark 2)
above.
On the other hand, for the applications to the finiteness of integral
points, work of Robinson-Roquette [Rob], and McIntyre [Mac] allows one
to suppress the ineffectivity due to the Mordell-Weil theorem. They use a
strong form of Roth's theorem, with several places, which allows the passage
to a finite extension field (cf. 7.7). One can also obtain an algorithm
giving the integral points on a curve, by assuming known a function M(a, 6)
(6) 2) such that all solutions wE V(K) of
satisfy H(w) ~ M(a,6). We sketch the idea.

Going back to the above proof, the problem consists now of finding 11"
such that Wn = mw~ + 11". Consider the kernel Vm of multiplication by m,
and write
H = {z E Hl(K, Vm ): the image of z is zero in every Hl(local, V)}.

Then H is finite and effectively constructible. Let Vz be the image of z in
Hl(K, V); Vz is a principal homogeneous space, and it has points in all the
completions, but one has no method for constructing rational points on Vz
(this is the reason why the Mordell-Wei! theorem is ineffective). But one
can effectively construct a morphism </> : Vz -+ V such that </>z(x + a) =
</>z(x) + ma, a E V and </>z(Vz(K = {mw + 11"} for a certain 11". One can
then dispense with 11" by choosing w~ E Vz such that </>(w~) = W n . But
Vz is not embedded in the same projective space as V. On V we have an
ample line bundle L; the class of mL is invariant under the action of the
Galois group. The obstruction to the existence of a sheaf defined over K
on Vz corresponding to mL is given by an element of the Brauer group.
But as there are points in all the completions, this element of the Brauer
group is zero. To give an algorithm it is necessary to write the 2-cocycle
corresponding to the element of the Brauer group as equal to a coboundary
(all the co chains can be enumerated). Once one has the sheaf on Vz it
provides a projective embedding and thus a height, and the proof can be
completed as before.
[The fact that the Mordell-Weil theorem is not really necessary for the
proof of Siegel (and Mahler) was already pointed out by Weil as early as
1935. See his Collected Papers, Vol. 1, p.99, where he writes:
101
"Siegel recourt, il est vrai, au theoreme final de ma these, mais une
analyse detaillee du memoire montre que c'est inutile, et que Ie theoreme
des extensions non ramifiees sufE.t a la demonstration: de sorte que les
resultats de Siegel sont independants de la methode de descente infinie. 11
en est de meme, naturellement, des beaux resultats de Mahler, qui etendent
notablement ceux de Siegel, au moins pour Ie genre 1."]
7.4. Application to curves of genus
1.
We apply the approximation theorem on an abelian variety to a curve

embedded in its jacobian. Let X be a projective non-singular absolutely
irreducible curve over K of genus ~ 1. Let v be a place of K.
We consider a sequence of distinct points Pn E X(K)j in particular
their height tends to infinity. Let </> be a non-constant rational function
on X defined over K. For sufficiently large n, Pn is not a pole of </> and
Zn = </>(Pn ) belongs to K.
Theorem. We have
lim log IZn Iv = O.

n..... oo logH(zn)
Proof Assume this is false. By taking a subsequence and replacing

II</> we may suppose that
log IZnlv
logH(zn)
-+
\
A,
.h
WIt
00
\
A
</>
by
< O.
In particular, Zn -+ 0 in K v and by taking a subsequence we may assume

that Pn converges to a zero Po of </>. As we are on a curve, Po is an algebraic
point of X.
Between H(Pn ), the height corresponding to a morphism X -+ PN,
and H(zn), corresponding to a morphism X -+ PI, we have an inequality
(cf. 2.6)
On the other hand, if e is the multiplicity of Po as a zero of

IZnlv ~ dv(Pn,PoY. Therefore, there is c > 0 such that
which contradicts the approximation theorem.
</>,
we have
102
In the particular case J( = Q, v = the infinite place, the preceding

theorem can be equivalently stated:
- 1,
Iim log IPnl n-+oo log Iqnl
where Zn = Pn/qn = (Pn ), Pn E X(Q). Therefore, the ratio ofthe number
of digits in the numerators and denominators of rational points tends to 1,
which is more precise than the finiteness of the number of integral points.
7.5. Proof of Siegel's theorem.
Let X be a curve of genus 2: 1 defined over a number field K, a nonconstant rational function on X defined over K, Pn a sequence of distinct
points of X(K), v a place of K. We have proved that (7.4)
log IZnlv
(1)
logH(zn)
--t,
where Zn = (Pn ).
We shall deduce the following result: if 8 is a finite set of places of K
containing the archimedian places 8 00 , and if As is the ring of 8-integers
of K, then there are only finitely many n such that (Pn ) = Zn E As.
If not, by taking a subsequence, we have Zn E As for all n, that is
to say, IZnlw ~ 1 for w 8 and for all n. But the height H(z) has been
defined as
H(z) =
sup(l,lzlv),
II
vEMK
where the
Z EAs
1.1 are normalised to satisfy the product formula (2.1).

H(z) =
II sup(l, Izlv),
vES
and
logH(z)
=L
sup(O, log Izlv).
vES
By applying this to
Zn
we get
which contradicts formula (1) as n
--t 00.
Thus for
103
Corollary. If X is an affine curve defined over K of genus 2: 1, every set
of K -rational points of X which is quasi-integral relative to As is finite.
It remains to treat the case of curves of genus 0 with m points at
infinity, m 2: 3. By making an extension of scalars, we can assume that
these points are rational. We shall reduce to the case of 3 points at infinity.
If : X ~ Y is a morphism of affine varieties, over K, and if L: C
X(K) is quasi-integral relative to As, then (L:) is also quasi-integral. In
particular, if we remove one point Q from X, we have a morphism X {Q} --t X. If QI, ... ,Qm are the m points at infinity of X, by considering
the morphism X - {QI, ... , Qm} --t X - {Q1. Q2, Q3} we are reduced to the
case of 3 points at infinity.
When m = 3 we can assume that X = PI - {O, 1, 00 }. The affine ring
of PI - {oo} is K[T], and the affine ring of X is K[T, l/T, l/(T - 1)] =
K[T, l/T(T -1)]. If we write t = T, x = l/T, y = 1- T, z = 1/(1- T) we
obtain another presentation of the affine ring of X, with generators t, x, y, z
and relations
(2)
tx
= 1, yz = 1, y + t = 1.
If Y is an affine variety, a morphism Y --t X amounts to a morphism

of the ring K[T, l/T, l/(T - 1)] of X to the ring Ay of Y, that is to say
elements t, x, y, z of Ay satisfying (2); the relations (2) mean that y and t
are units of sum 1. (The units of a number field of sum 1 have been studied
since Nagell under the name of exceptional units.)
Let L: be a quasi-integral subset of X(K). On L: the functions t, l/t,
l/(t -1) take their values in D- I As, for some D E K, D f. o. We identify
the points of L: with their corresponding values of t.
We first show that t = uto, where to belongs to a finite set, and u E
This amounts to proving that the As-ideals generated by the values of tare
finite in number. But
As.
= D-1a,a E As,
1
and t
= D- 1(3,f3
therefore
E As,
= D2 and (a) divides (D 2).

In the same way y = 1 - t = VYo, where Yo belongs to a finite set and v is
a unit of As. Therefore tou + yov = 1, and clearing the denominators we
a(3
reduce to a finite number of equations
(3)
Au+Bv
= C,
104
where A,B,C are given non-zero integers in As and the unknowns u,v are
units of As.
Siegel's theorem in this particular case states that there are only a
finite number of u, v E
satisfying such an equation.
[In particular the equation u + v = 1 defining exceptional units is of
this type.]
There axe several ways to complete the proof. Here is Siegel's method
which consists of reducing to a curve of higher genus.
We suppose that there axe infinitely many solutions (un, v n) of equation
(3). We know that the group r of S-units is finitely generated. Let m be an
integer ~ 3; then r /r m is finite. By taking a subsequence we can assume
that
As
where Uo, Vo E r are fixed elements (independent of n). Putting a = Auo,

f3 = Bvo, 'Y = C, we obtain infinitely many quasi-integral points (Un' V n)
on the curve
As m ~ 3 the genus of this curve is ~ 1, so we have a contradiction, which
finishes the proof.
(For the equation aX m + f3ym = 'Y, one can use directly the result of
Thue.)
Remarks.
1). On Siegel's proof.
In 1929, Siegel had only a weaker form of the approximation theorem of
Roth (1955): in place of the exponent 2 + , Siegel had the exponent 2yn,
where n is the degree of Q(a) over Q. But in the proof, the field Q(a)
may have to be extended (one writes a = rna' + 7r, where a' is algebraic
over Q (a. In the case of genus 1, enough is gained from the height to
compensate this degree increase. This is not true for genus ~ 2 and Siegel
then had to use simultaneous approximations. But all this part of Siegel's
proof can now be elided, thanks to Roth's theorem.
2). S-integral points.
Siegel proved his theorem for the usual integral points (S = Soo). In
1933, Mahler extended this result to S-integral points, but only in the case
of genus 1 (because of the above difficulties) and over the rationals Q. He
used his p-adic analogue of Siegel's approximation theorem.
105
In 1955, Roth proved his theorem with the exponent 2 + E for an archimedian valuation. In 1958, Ridout treated the case of an arbitrary valuation. Some generalisations were given by Lang [L 7] and LeVeque [Le]
around 1960. Lang showed that Roth's theorem could be extended, under
reasonable hypotheses, to any field equipped with a product formula, in
particular to function fields in characteristic O.
(If <P is a formal power series in one variable T over G, there is a
sequence f n I gn of rational functions, where f n ,gn are polynomials of degree
~ d n , such that 2, the analogue of Roth's theorem
is that there is no sequence with <p - f nl gn = O(Tl<d n ). Such results had
already been obtained by Fenna [Fe] and Uchiyama [U] between 1956 and
1960.)
Siegel's theorem is then valid for all rings (of characteristic 0) of finite
type over Z.
Roth's theorem for function fields does not extend directly to characteristic p (consider <p = T + TP + Tp2 ... which is a root of the Artin-Schreier
polynomial XP - X + T = 0). It would be interesting to know the analogue in characteristic p of Roth's theorem and of Siegel's theorem. The
only counterexamples known arise from the analogue of Mordell's conjecture over functions fields and are obtained by composing a constant map
with the Frobenius endomorphism (d. 5.7). Further, it would be interesting to combine Siegel's arguments with those of Mumford saying that the
rational points are widely spaced (5.7).
7.6. Application to P(f(n)).

For an integer n
pen).
i= 0, 1,
we denote the greatest prime factor of n by
Theorem. (Siegel). Let f E Z[X] be a non-constant polynomial having at

least two distinct zeros over G. Then P(f(n)) - t 00 with n.
For example, for every sufficiently large integer n, n 3
factor 2: 1010 .
+ 2 has
a prime
First proof
If not, there is a sequence (nd of integers tending to infinity, and an
integer M such that If(ni)1 > 1 for all i, and all the prime factors of f(ni)
are bounded by M. If S is the set of prime numbers ~ M, the 1I f (ni) are
S-integers. But Q[T, II f(T)] is the affine algebra of the variety obtained by
removing the zeros of f from the affine line; this variety is an affine curve
106
of genus 0, where the point at infinity and at least two other points (the
zeros of 1) have been removed. From Siegel's theorem, the set of S-integral
points is finite.
Second proof.
As Siegel did not have the finiteness theorem for arbitrary S, he proceeded differently. If the f(ni) involve only the prime numbers PI, "',Pm,
we can write
f(ni) = p~l ...p~m.
Let 1 be an integer> deg f. We reduce the hj modulo 1 and obtain:
for some Yi E Z. But each of the curves
is irreducible, and of genus ~ 1 (because of the choice of I) and has only

finitely many integral points.
Both proofs are ineffective. It is now known by effective methods that
P(f(n
~f
loglogn.
(see [BM, Chapter 3]).
7.7. Effectivity.
The above exposition is based on a version of Roth's theorem for a
single absolute value. Here is a statement for a finite number of places.
Let S be a finite set of places, and let Wn be a sequence of elements
of K, with H(w n ) ---t 00. For each v E S we suppose given Cv E K v , Cv
algebraic over K. If 8 is a real number> 2, then one cannot have
II IW
vES
n -
cvlv ~ H(w )6' for all n.

n
(see [L2]; some previous results are due to Mahler, Parry).

Using this, the approximation theorem on abelian varieties (7.3) can
be proved in the form: if one takes Cv E V(Kv), Cv algebraic, for each
107
v E S, and if 6
V(K) with
> 0, then there is no sequence Wn of distinct elements of
This is true by Roth's theorem if 6> 2 (by projecting onto a line). On the
other hand, to prove this formula J( can be replaced by a finite extension
L, under the condition that S be replaced by the set of places of Labove
those of S: indeed, for A E K v, we have
Let M be an integer such that (m 2 - 1)6 ~ 3. We replace the Wn by

their mth roots: mw~ = W n . From the proof of the Mordell-Weil theorem,
there is an extension L of K in which these Wn are mth roots (L is a finite
extension uruamified outside m and the places of bad reduction). Using
this version of Roth's theorem, the ineffectivity due to the Mordell-Weil
theorem is removed.
108
8. BAKER'S METHOD
Baker's method does not actually prove Siegel's theorem in full generality, but in every case where it applies, it is effective.
8.1. Reduction theorems.

We first study the relations between quasi-integral points under morphisms of affine varieties.
We keep the notation of the previous chapter: K is a number field,
A = As the ring of S-integers of K (most of the following extends to any
integral domain A).
1). If : X -+ Y is a morpbism of affine varieties, and if E c X (K) is
a set of rational points, quasi-integral relative to A, tben (E) in Y(K) is
quasi-integral relative to A.
This is clear.
2). A morphism : X -+ Y is finite if and only if it is proper and has finite
fibres. If X and Y are affine with affine rings Ax and Ay, this means that
Ax is a finite Ay-module, or that Ax is integral over Ay.
If : X -+ Y is a finite morpbism of affine varieties, and ifE is a subset
of X(K), tben E is quasi-integral if and only if (E) is quasi-integral.
Proof
Suppose that (E) is quasi-integral. Let
integral over A y : there is a relation
fm
E Ax. As is finite,
+ 91!m-l + ... + 9m =
f is
0,
where the 9i belong to the image of Ay in Ax. There is D E A, D =j: 0,

such that
9i(E)E
~A,
fori=l, ... ,m.
Then
Thus for u E E, D feu) is an element of K integral over A. As A is integrally
closed, we have D f( u) E A.
Consequences. a). To study quasi-integral sets, we may assume that the
curve is non-singular (use the normalisation map which is finite).
109
b). Let X be a non-singular affine curve, X its projective completion,
and X= = X-X. From the Riemann-Roch theorem, there is a function
1 : X - t PI whose set of poles is Xoo; then 11x gives a finite morphism
X - t PI - {oo}, and a subset 'B of XCI<) is quasi-integral if and only if
its image is quasi-integral. Thus finiteness theorems can be reformulated in
terms of functions on the curve.
Assertion 2) only gives us information on the points of Y which are
liftable to X over I<. We need more.
3). Let 1: X - t Y be a morphism of affine varieties; assume that 1 is finite,

etale, and surjective. If 'By is a quasi-integral set of points of Y, there is a
finite extension I<' of I< such that 1- 1 ('By) is rational over I<'.
We first show that there is b.. E A, b.. i= such that if a E 'By, then
I1-1(a is an extension of I< unramified outside b...
(If for example we take X and Y to be the affine line, and the morphism
1 to be z f-t z2, then 1 is finite and etale outside 0, but is not etale at 0, and
the field I via) , a E A can be ramified anywhere. But if we take X and
Y to be the affine line with removed, and 1 : z f-t Z2, then 1 is finite and
etale, and an integral point is a unit a of I< and then I via) is unramified
outside the divisors of 2.)
The proof of the existence of b.. is the same as that of the Chevalley-Weil
theorem (which assumes that the varieties are projective and the morphisms
are everywhere etale; here the morphisms can be ramified at infinity cf.
4.2).
For b.. E A, b.. i= 0, we consider the ring ALl = A[i]. By elementary
arguments of algebraic geometry (d. [G, IV,8]) one can find affine schemes
X, Y of finite type over a suitable ALl such that
XA
I< = X, Y
XA
I< = Y,
and a finite etale surjective morphism <P : X - t Y corresponding to 1. The

elements of 'By are points of XCI<) having a common denominator, and
we can assume that 'By C Y(A.6.). A point a of 'By can be viewed as a
morphism Spec A.6. - t Y. Then a lifts over an extension of I< unramified
outside the divisors of b...
The proof of 3) is now completed using Hermite's theorem (cf.4.1):
there are only finitely many extensions of bounded degree and unramified
outside a given finite set of places. Here the degree is bounded by the
number of points in the fibres of 1.
110
Exercise. Let > : X ~ Y be a finite etale surjective morphism of nonsingular affine curves over a field of characteristic o. Then X is exceptional
(that is, of genus 0 with at most two points at infinity) if and only if Y is
exceptional.
[Note that the finite unramified coverings of PI - {O, oo} are given by
z 1-+ zn, n E Z.]
4). If X is a variety over K, let (FX,K) denote the property:

(Fx,ld Every set of quasi-integral points of X(K) is finite.
Let (Fx) denote the property:
(Fx) We have (FX,KI) for every finite extension K' of K.
Proposition.
a). Let > : X
Y be a morphism with finite fibres. Then (FY,K) =>
(FX,K).
b). Let > : X ~ Y be a finite etale surjective morphism. Then (Fx) is

equivalent to (Fy).
Proof
a). If (FY,K) is true, and if Ex is a quasi-integral subset of X(K), then

>(Ex) is a quasi-integral subset of Y(K), therefore finite, and as the fibres
are finite Ex is finite.
b). It is clear that (Fy) => (Fx), from a). We show that (Fx) => (Fy). Let
Ey be a quasi-integral subset of Y(K). From reduction 3), we can assume
(after enlarging K) that the points of E y lift: there is Ex C X(K) such
that >(Ex) = Ey. From the reduction 2), Ex is a quasi-integral subset.
Then (Fx) implies that Ex is finite, thus its image by > is finite.
Remarks. 1). All the reduction theorems above are effective.

2). Siegel's trick of reducing the non-exceptional case of genus 0 to the
case of genus 2: 1 can now be explained by considering the Fermat curve
Xl + yl = Zl (which has genus 2: 1 for 1 2: 3) as an etale covering of PI {0,1,00}, via the map (X,Y,Z) 1-+ (X/Z)l. Hence to prove the finiteness
theorem for PI - {O, 1,00} amounts to proving it for the Fermat curve.
8.2. Lower bounds for L:,Bdogai.

Baker's method provides lower bounds for non-zero linear combinations
L:,Bi log ai, where Q'i and ,Bi are algebraic numbers. When the numbers ,Bi
are ordinary integers (we denote these by bi) this is equivalent to bounding
(IT Q'~i) - 1 from below.
l11
Let K be a number field of degree :$ Dover Q, aI, ... , an some elements

of K, bI, ... , bn rational integers, and v an absolute value of K. The bounds
can be written in terms of
and the heights of the ai. Baker uses the following notion of height:
Let a be an algebraic number of degree d( a) = d; we can write a
minimal equation of a in the form
aoa d + ala d - 1 + ... + ad = 0,

where ao, ... , ad are rational integers of greatest common divisor 1. We
define
For example, if a = p/q is a rational number with (p,q) = 1, then A(a)
is the usual height sup(lpl, Iql) = H{P/q). In the general case, it is easily
proved that
A(a) ::::::d(o:) H(a)d(o:),
where H(a) is the normalised height of a. Geometrically, one sees this by

considering (ao, ... , ad) as a point of projective space P d; on the other hand,
a and its conjugates give a point in the d-fold symmetric product of the
projective line. But PI X X PdSd is naturally isomorphic to P d, and the
above equivalence follows from this.
Explicitly, we have
A(a) :$ 2d(o:) H(a)d(o:), and H(a)d(o:) :$ (1 + d(aA(a).

It is possible to replace 2d(o:) by
Put Ai
([d1~);2])
and 1 + d(a) by 1 + d(a)1/2.
= sup(4,A(ai, 1 :$ i:$ n.
Baker's Theorem. (d. [Ba], [BM], [L5]). If n:=l a~i
=f: 1, then
\(ll a~i) -l\v > exp{-cO(logO)(logB)},

i=l
where 0 = n:=llogAi, and c is an effectively computable constant which

depends only on Kv,D and n.
For example, if Kv = C, we can take c to be the number (16nD)2on;
this can be refined to 2I2n+53n2n D n+2 (d. [Wa2]).
We shall use this theorem in the following form.
112
Corollary. If n:=l a~i
f= 1, then
n
I(II a~i) -llv > l/Bc(a),

i=l
where c(a) is a constant depending on
/(v, ab ... , an.
For every algebraic number x f= 0, the product formula implies the

inequality Ixlv ~ I/H(x). Applied to x = (n:=l a~i) -1 this gives
n
I(II a~i) - Ilv ~ l/c B ,

i=l
where c depends only on ab ... , an and Kv. In place of c B , it is essential

for the applications to have a function which increases less quickly than
an exponential of B. Baker's first results, with a lower bound of the form
exp{-c(logB)II:}, amply sufficed. Nevertheless, the box principle shows
that the correct order of magnitude is indeed 1/Bc(a).
For the most general linear combination E .Bdog ai with the .Bi algebraic, one has an analogous assertion with Q(log n)(1og Bn), where B is an
upper bound for the heights of the .Bi (see [BM, Chapters 1 and 2]).
8.3. Application to PI - {O, 1,00}.

As usual, we denote by As the ring of S-integers of K. To show the
finiteness of quasi-integral sets of P l - {O, 1,00} amounts to showing that
if A, B, C are non-zero elements of As, then the equation
Au+Bv = C,
As
has only finitely many solutions u,v in

(cf. 7.5, Equation (3.
As the group
of S-units is finitely generated, it can be written
J.L x L where J.L is a finite group (group of roots of unity in K) and L is a free
abelian group. We select a basis E}, ... , Er of
so that an S-unit u E
can be written uniquely
As
L,
As
with ( a root of unity and bi E Z. We then put b(u) = sUPl<i<r Ibil.

That u is an S-unit means that lul w = 1 if w S. If we-associate to
uE
the family (log lulw)wES of logarithms of its absolute values at the
places of S, we obtain an embedding
As
1: A*s/J.L = L
'-+
Risi.
113
The product formula implies
L log lulw = 0,
wEB
so the image of L is a discrete subgroup contained in a hyperplane of Risi.

On this discrete subgroup there are two equivalent norms one given by
sup Ibil, the other by the coordinates of Risi.
Hence there is a constant c> 0 such that for all u E
we have
As
c- 1 sup 110glulwl ~ b(u) ~ csup 110glulwl.

wEB
wES
It is convenient to remove the absolute values of the logarithms, which can
be done by means of the following remark: if X w , (w E S), are real numbers
with EWES Xw = 0, then sUPwES Ixwl ~ (181- 1) sup XW. We thus obtain:
Lemma. There is a constant c > 0 such that for all u E
As we have
c- 1 sup log lul w ~ b( u) ~ c sup log lul w.

wEB
wES
Suppose now there is an infinite sequence (un, v n ) of solutions of the
equation Au + Bv = C. By taking a subsequence, we may assume (by
symmetry) that b(u n ) ~ b(vn ), and that there is w E S with log lunl w ~
b(un)/c. Writing u = Un, V = V n , we then have
Iwhere
Cl
> 0 and
C2
Bv
1
-b(u)
,
Au - 1 w < CIC2
> 1 are independent of u. We write
and we have
As Au + Bv = C
gives
# 0, the left hand side is non-zero, and Baker's theorem
I1b~ -b
rb~-br O! -
11 w > l/b( u .
)C3
As Cl,C2,C3 are independent of n, we obtain the finiteness of the number of

solutions.
114
This method gives an effective bound for the solutions of Au+Bv = C,
knowing Ell . , Er . On the other hand, one can also find effectively a system
of generators El, ... , Er of L, with an explicit bound for the absolute values
of the Ei ([Gy], [Si2]).
8.4. Applications to other curves.

We have dealt with the crucial case of the projective line minus 3 points.
To go further, we use the following two reduction principles (see 8.1(4:
Let X and Y be two affine curves, f : X -+ Y a morphism with finite fibres.
Then:
Reduction 1. We have (Fy) => (Fx).
Reduction 1J. If f is finite, etale and surjective, then (Fx) {:} (Fy).
If X and Y are the corresponding projective curves, we can write X =
X - X oo , Y = Y - Y00, and f extends to a morphism! : X -+ Y such that
!-I(yoo ) C Xoo. For f to be proper (equivalently, finite), it is necessary
and sufficient that !-I(yoo) = X oo , and then Yoo = /(Xoo). For the second
reduction one further needs that the covering X -+ Y be unramified outside
Yoo'
Case l.Thue's equation F(x,y) = m.
Let F be a homogeneous polynomial in X and Y with coefficients in the
number field K, having at least 3 distinct linear factors over the algebraic
closure. Let m E K, m '1= O. Then the equation F(x, y) = m has only a
finite number of solutions x, y in As.
To see this, we apply the first reduction principle to the morphism
which sends the point (x, y) of the affine curve X (with equation F(x, y) =
m) to the line which joins this point to the origin:
X
(x, y)
-+
1-+
PI - {zeros of the polynomial F(l, t)}
= y/x.
As at least 3 points have been removed from PI, we obtain the result we
want.
[When Baker applies his theorem to the effective approximation of an
algebraic number a by rational numbers p/q, he selects two conjugates a'
and d' , constructs a morphism to PI -{a, d, a"}, and changes the variables
to send the 3 points to 0, 1,00. He thus reduces the problem to an equation
Au+Bv=C.
115
This method gives a bound for the absolute values of x, y. For example,
over Z, one has a bound
max(lxl, Iyl) ~ (m + l)C(F),

for the solutions of F(x, y) = m.]
Case 2. Elliptic curves.

Let X be an affine curve whose compactification X is an elliptic curve.
As X is affine, Xoo = X - X is non-empty. By extending the ground field
it suffices to treat the case where Xoo has exactly one point, rational over
K. We take this point as origin and put on X the corresponding group law.
In classical language, this amounts to saying that we study the Weierstrass
equation y2 = I(x) where I is a polynomial of degree 3.
Multiplication by 2 is etale and the inverse image of consists of 4
points 0, Et, E 2 , Ea (the subgroup of 2-division points). By the second
reduction, proving the theorem for X - {o} reduces to proving it for X {O, El, E 2 , Ea}. We then apply the first reduction to the morphism given
by the coordinate x:
x:
X - {0,EI ,E2 ,E3 }
~ PI - {oo,el,e2,e3},
where I(x) = (x - el)(x - e2)(x - ea). As the finiteness result is true for
PI - {oo,et,e2,ea}, it is then true for X.
Reductions of this type are quite old. They have been used by Siegel,
Chabauty, Kubert, Lang and others.
Case 3. Hyperelliptic case.

Assume that X is hyperelliptic of genus 9 ~ 2. Let u be the hyperelliptic involution of X, so that X / {1, u} is isomorphic to P lover some
extension of K.
Assume further that Xoo n uXoo f:. 0. By the first reduction we may
assume (after some field extension) that Xoo is:
either a rational point P, fixed under u,
or a pair (P, uP), where P is rational and not fixed by u.
This means that X is the affine curve defined by the equation
y2 = I(x),
where
roots.
is a monic polynomial of degree 2g + 1 or 2g + 2 without multiple
116
Let aI, a2, a3 be three distinct roots of

curve E defined by the equation
f. We introduce the elliptic
Then via the map x, E is a covering of the projective line, ramified at

aI, a2, a3 and 00. On the other hand, the map X -+ PI, defined by (x, y) 1-+
x, is ramified at aI, a2, a3 and some other points:
./
Consider the normal projective curve Z, whose function field is the
composite of the function fields of E and X; if x, yare variables for X, and
t, x variables for E, then x, y, t are variables for Z. We denote by Zoo the
points of Z above 00 E PI, and by X oo , Eoo those of X, E:
Z
E
'\.
./
'\.
PI
./
The finiteness theorem is true for E - E oo , therefore by the first reduction it is true for Z - Zoo. To apply the second reduction, it suffices to check
that the arrow Z -+ X is proper and etale. By base change it is clearly
proper. To see that it is etale, we use the fact that E -+ PI is ramified
only at al, a2, a3, and 00, and X -+ PI is ramified at aI, a2, a3 and these
ramifications cancel each other (Abhyankar's lemma). Equivalently, to pass
to E we extract the square root of (x -al)(x -(2)(x- (3); to pass to X we
extract the square root of (x - ad ... (x - am); thus for Z -+ X it suffices to
extract the square root of (x - (4) .. '(X - am), which gives no ramification
at aI,a2,a3. Thus the finiteness theorem is true for X.
The first case not treated by Baker's method is an hyperelliptic curve
of genus ~ 2 with Xoo = {P} and P =1= uP. For example, we take y2 = f(x)
with f of degree 6 and add one of the two points at infinity. Siegel's theorem
gives the finiteness of integral points, but whether it can be made effective
is unclear.
117
Case 4. Superelliptic equations.
We consider the equation ym = f(x), where m is an integer

has at least 2 simple roots:
f(x) = (x - a)(x - f3)g(x),
a =1=
f3,g(a)
=1=
O,g(f3)
=1=
3, and
o.
We introduce the curve E defined by t m = (x - a)(x - f3); E is hyperelliptic

(or elliptic if m = 3 or 4):
We use the same principle:
Z(t,x,y)
(t, x) E
,/
x (x,y)
The finiteness theorem is true for E, thus for Z, and

00 by the same arguments as before.
Z -+ X is etale outside
Remark. A further example of an effective (and completely explicit) result

obtained by Baker's method is the following [L5, Appendix to Chapter 6] :
Theorem. Let f E Z[x, y] be an absolutely irreducible polynomial for

which the curve f(x, y) = has genus 1, and let H be the maximum of
the (usual) absolute values of the coefficients of f. Then for every point
(x, y) E Z x Z on the curve, we have
sup(lxl, Iyl) ~ expexpexp(2H 10
n 10
).
Baker's method applies to other types of equations, for instance to

exponential equations. The most striking example is the result of Tijdeman
on the Catalan equation: every solution (x, y, z, t) E Z4 in integers;::: 2 of
the equation x Y - zt = 1 can be effectively bounded:
xY
See [Til, [Lg].
exp expexp 250.
118
8.5. Applications to elliptic curves with good reduction outside a

given finite set of places.
For many modular curves, one can construct non-constant morphisms
to PI - {O, 1,00} (see Exercise below).
The first example is that of the curve X(2) which classifies elliptic
curves equipped with a basis of points of order 2. The parameter which
is classically called>' gives a morphism>. : X(2) -+ PI where X(2) is the
completed modular curve i.e. with cusps added. Writing an elliptic curve
in Legendre form
y2 = x(x - 1)(x - >.),
>. is connected with the j-invariant by
.
8(>.2_>'+1)3
J = 2 >.2(>. _1)2
The cusps are>. = 0,1,00 so that X(2) = PI - {0,1,00}. In particular,
every quasi-integral set of points of X(2) is finite.
Here is an application. Recall that an elliptic curve has good reduction
at a place v if there is a model of the curve
where the ai are v-integers, and the discriminant is invertible at v. Reduction modulo v then gives an elliptic curve.
Theorem. Let K be a number field and S a finite set of places of K. Then
up to isomorphism, there are only a finite number of elliptic curves over K
with good reduction outside S and these can be found effectively.
First proof.
We can assume that S contains the prime divisors of 2. Let E be an
elliptic curve having good reduction outside S. Let K' be the field obtained
by adjoining to K the points of order 2 of E (if we write the equation of
E as y2 = f(x), K' is obtained by adjoining the roots of f to K). Then
the extension K' / K is unramified outside S. By Hermite's theorem and by
malcing a finite extension of K, we may assume that the points of order 2
are in the field K. In other words, the curve can be written in the Legendre
form
y2 = x(x - 1)(x - >.).
The set of >. so obtained is quasi-integral in X(2), for j is integral in As

and the inverse image of 00 by the morphism PI -+ PI, >. 1-+ j is {O, 1, 00 }.
119
Thus the set of these A is finite and can be found effectively by Baker's
method (cf. 8.3).
Second proof
We may enlarge S so that it contains the divisors of 2 and 3, and that
As is principal. One checks that the equation of the curve can then be
written as
y2 = x 3 + Ax + B,
where the discriminant b.. = -4A3 - 27B2 belongs to As. As b.. is defined
modulo the 12th powers, this leads to a finite number of equations 4x 3 +
27y2 = const (see [Se3, IV.1.4] for more details), to which one applies Case
2 of 8.4.
Problem. What are the elliptic curves over Q with good reduction outside
II? (Those with conductor 11 are known, but not those with conductor
11 2 , as far as I know.)
Corollary. Let E be an elliptic curve defined over K. There are only
finitely many elliptic curves E' over K, up to isomorphism, which are Kisogenous to E.
To see this, we take S so large that E has good reduction outside S.
By a general property of abelian varieties, the same holds for E'. From the
previous theorem, the E' are finite in number.
It would be interesting to have an explicit bound for this number (for
K = Q, M. Kenku [Kel] has shown that 8 is the precise bound).
Remark. The above corollary can be used (see [Se3, Ch.IV, 2.1]) to give
another proof of the irreducibility theorem (5.4) that if E does not have
complex multiplication then the action of Gal( f( / K) on the Tate module
Vi(E) is irreducible.
Exercise. (Kubert-Lang). Let Xl(n) be the modular curve which classifies
pairs (E, P) where E is an elliptic curve and P a point of order n on E.
Assume n ;::: 6. Construct a non-constant morphism
by
<p(E P) = x(P) - x(2P)
x(P) - x(3P) ,
120
where x is the "x-coordinate" on E. Show that <p takes values in PI {O, 1,00} and deduce from that the effective finiteness of quasi-integral sets
on XI{n).
121
9. HILBERT'S IRREDUCIBILITY THEOREM
Hilbert's irreducibility theorem has quite a large number of proofs,
based on different principles; some give precise estimates for the number of
integers that one seeks. There are several applications: the construction of
elliptic curves over Q having rank ~ 9 and the construction of extensions
of Q with Galois groups Sn, An, ...
9.1. Thin sets.

Let k be a field of characteristic O. Let
An(k)
= kn.
n be
a subset of P n (k) or of
Definition. We say that n is "thin" (Fr.: "ensemble mince") if there is an

algebraic variety X over k and a morphism 1f : X - t P n (resp. 1f : X - t An)
with the following properties:
(a) n C 1f(X(k,
(b) The fibre of 1f over the generic point is finite and 1f has no rational
section over k (i.e. in a neighbourhood of the generic point 1f is quasi-finite
and has no section).
A subset of P n(k) is thin if and only if its intersection with An is thin.
Any finite union of thin sets is thin.
There are two important types of thin sets. We shall say that a subset
n has type 1 if it is contained in a Zariski-closed subvariety F, different
from the entire space. To see that such a subset n is thin, take X = F and
1f the injection. The generic fibre is empty.
We shall call a subset type 2 if it has the form 1f(X(k)), where X is an
irreducible variety over k of dimension nand 1f is a dominating morphism
X - t P n or X - t An of degree d ~ 2. This is the most interesting case
of a thin set. If we take tl, ... , tn as affine coordinates, the function field
k(X) of X is an extension of k(tt, ... , t n) of degree d. If X is not absolutely
irreducible, then 1f(X(k)) is of type 1. Indeed, in that case the field k(X)
contains a subfield kl finite over k and =1= k. On the other hand, if x E X (k ),
the local ring of x does not contain kl' hence is not normal; this shows that
X(k) is contained in the non-normal locus of X, and 1f(X(k is of type 1.
[An example is the variety x 2 + y2 = 0 over R, which defines an irreducible,
but not absolutely irreducible, variety; R( i) = C is in the function field.
There is only one point over R; the normalised variety has no point over
R.]
If necessary, we may then assume that X is absolutely irreducible. The

following statement is obvious.
122
Proposition. Every thin subset of P n (k) or An (k) is contained in a finite

union of subsets of types 1 and 2.
Examples of thin sets.
Let d;::: 2 be an integer. In AI(k) = k, the subset k d = {xd,x E k} is
a thin subset (take X = AI, and 7r : X ~ Al defined by 7r(x) = x d ).
A finite union U cikd of classes modulo k d is also a thin set.
Polynomial interpretation.
The definitions above can be reformulated as follows for An, say.
That 0 is thin of type 1 is equivalent to the existence of a non-zero
polynomial G(TI' ... , Tn), with coefficients in k, such that G(tl, ... , tn) = 0
for all (tl' ... , tn) E O.
For type 2, we take a polynomial F(X, T) E k(T)[X] where
T = (TI' ... , Tn). We assume that F is absolutely irreducible, that is to
say, irreducible in k(T)[X] (where k is an algebraic closure of k). We also
assume that F has degree ;::: 2 in X. Then the set
OF = {
t;
tis not a pole of the coefficients of F, }

and F(X, t) has a root in k
is thin of type 2, and we obtain all of them thus (modulo type 1). [If we
take an irreducible variety as in the definition of type 2, its function field
is a finite separable extension of k(T), and we choose for F a generator of
that extension.]
9.2. Specialisation of Galois groups.
We shall see that specialisation outside a thin set preserves the irreducibility of a polynomial and the structure of its Galois group.
For the first statement, we assume that X is irreducible over k, of
dimension n, and 7r : X ~ P n is dominating.
Proposition 1. There is a thin set 0 in Pn(k) such that for all t 0,

(t) consists of a finite number of points which are conjugate over k.
7r- 1
In terms of polynomials, this means that if F(X; T I , ... , Tn) is an irreducible polynomial over k(T), then for all t = (tI, ... , t n ) 0, F(X, t) is
irreducible over k. More precisely, it follows from Proposition 2 below that
F(X, t) has the same Galois group as F(X, T) outside a thin set.
123
For the second statement, start again with 11" : X -+ P n with X irreducible, and consider a finite group G acting faithfully on X such that X/G
is identified by the projection 11" with an open subscheme U of P n . We take
U small enough so that 11" is etale. Let u be a rational point of U and select
a closed point x of the scheme X above u. Up to conjugation, the subgroup
G z of G which stabilises x, depends only on u. The extension k(x)/k is
Galois with Galois group Gz .
Proposition 2. There is a thin set
n of P n (k) outside of which G z = G.
Proof of Proposition 2.
Let {Hi} be the subgroups of G different from G (one can restrict
oneself to maximal subgroups). Define Xi = X/Hi; this is a variety which
projects onto X/G, thus generically on P n, and the degree of the projection
is [G: Hi] > 1. The union n of the images of the (X/Hi)(k) is a thin set.
Proposition 2 follows.
Proof of Proposition 1.
It suffices to embed the field extension corresponding to X
Galois extension, and to use proposition 2.
-+
P n in a
9.3. Examples of degrees 2,3,4,5.

We consider a polynomial F(X; Tb ... , Tn) = F(X; T) irreducible over
k(T), of degree d ;::: 2 in X and we shall describe a thin set n c k n such
that for t 'I n, F(X;t) has the same Galois group as F(X;T).
1). d = 2. One reduces to the equation X 2 = geT), where geT) E k(T) is
not a square in k(T). We take
n = {t;g(t) is a square in k}.

2). d = 3, the Galois group of F(X;T) being S3. The proper maximal
subgroups of 8 3 are A3 and the subgroups of order 2. The group A3 has
index 2 in S3, the corresponding extension is given by the square root of the
discriminant ~(T) of F(X;T). The first condition (relative to A 3 ) is that
~(t) is not a square in k (note that by assumption ~(T) is not a square in
k(T. The second condition (relative to the subgroups of order 2) is that
F(X; t) has no root in k (that is to say, is irreducible over k since d = 3).
Hence, we take
n = {t;~(t) is a square in k} U {t;F(X;t) has a root in k}.
124
3). d = 4, the Galois group of F(X; T) being 8 4 . The proper maximal

subgroups are: the alternating group A 4 , the groups 8 3 fixing one letter,
and the Sylow 2-subgroups D4 dihedral of order 8.
The condition relative to A4 is that the discriminant'is not a square,
that relative to 8 3 is that the equation has no root in k.
The subgroup D4 appears as the stabiliser of the decomposition of a
set of 4 elements in 2 pairs:
2
We introduce the expressions
and write the unique monic equation having these 3 roots (the cubic resolvent). The condition relative to D4 is that the cubic resolvent has no root
in k.
The cubic resolvent has the same discriminant as the initial equation,
because
XIX2
+ X3X4 -
XIX3 -
X2 X 4
= (Xl -
X4)(X2 -
X3).
Therefore if the roots Xi of F are distinct, the roots of the cubic resolvent are
distinct. (We then see that the above thin set is the smallest appropriate.)
4). d = 5, the Galois group of F(X; T) being 8 s .
The maximal subgroups of 8 s are As, 8 4 (the stabiliser of one letter),
8 2 x 8 3 , and H 2o the normaliser of a Sylow 5-subgroup, of index 6.
The conditions on As and 8 4 require that F and the equation t 2 = .6.
have no root in k.
The condition relative to 8 2 X 8 3 means that the polynomial F(X, t)
does not decompose as a product of a quadratic factor and a cubic factor.
If Xl, . ,Xs are the roots (assumed distinct, since .6. =F 0), we write the
equation of degree 10 whose roots are the Xi + xj, (i =1= j). We then take t
outside the thin set for which this equation has a root.
The group H 2o can be described as
{X
1-+
ax + bj bE Zj5Z, a E (Zj5Z)*},
125
if 8 5 is realised as the group of permutations of Z/5Z. A Sylow 5-subgroup
of 8 5 corresponds to a pentagonal arrangement
Given such an arrangement of {I, 2, 3, 4, 5} we introduce the following

expression:
(the signs + correspond to the sides of the convex polygon, the signs correspond to the sides of the pentagonal star). As this expression is antiinvariant under the permutations of H 2o we divide it by
d= II(xi-Xj),
i<j
and write the equation of degree 6 (the sextic resolvent) whose 6 roots are
u/d and the 5 other analogous expressions.
a =
Theorem. Assume that ~ =I 0 and that the Galois group G of F(X, t) is

not contained in one of the groups 8 4 , 8 2 X 8 3 Then G is contained in
some H 2o (i.e. a normaliser of a Sylow 5-subgroup) if and only if one of the
a is in the ground field.
Therefore to check the condition relative to H 2o , it is necessary and
sufficient to check that the sextic resolvent has no rational root.
The hypothesis G rt 8 4 and G rt 82 X 8 3 (up to conjugation) means
that G is transitive, i.e. that the polynomial is irreducible.
Proof of theorem.
If G is contained in some H 2o , then the corresponding a is in the ground
field.
Conversely, suppose that one of the a is in k and that G is not contained
in any H 2o As G is transitive, its order is divisible by 5, and the number
of its Sylow 5-subgroups is == 1 mod 5. Therefore all the Sylow 5-subgroups
126
of 8 5 (there are 6 of them) are contained in G; hence G contains A 5 But
A5 permutes transitively the a's; the a's are thus conjugate over k. By
hypothesis one of them is in k. Therefore
We have
al
= d{XIX2 + X2 X 3 + X3 X 4 + X4 X 5 + X5 X l (XIX3
+ X3 X 5 + X5 X 2 + X2 X 4 + x4xd}.
We permute 1 and 2 and use the fact that d is anti-invariant. This gives
(XIX2
+ XIX3 + X3 X 4 + X4X5 + X5 X 2)}'
The equation ~ (al - (2) = 0 becomes
that is to say
Similarly, we have:
Subtracting the preceding equation, we get
l.e.
(2X4 -
As the
Xi
Xl -
X5)(X2 -
X3)
are distinct this implies
But by permuting the variables we also have
therefore
X4
= X3,
which contradicts
~ =1=
o.
= o.
127
9.4. Further properties of thin sets.
Plane sections.
Theorem. Let nand d be integers, n ~ d ~ 1, and let n be a thin subset of

P n (k). Then there is a non-empty open subscheme U of the Grassmannian
Gn,d of d-dimensionallinear subspaces ofP n , such that nnL is thin for all
L E U(k).
The most interesting case is d = 1. For example, if is a thin subset

of P 2 , then the intersection of n with a line is almost always thin.
This theorem can be used to reduce many questions on thin sets to
dimension 1.
Proof
We may assume n is of type 1 or 2, and type 1 poses no problem.
Let then X be absolutely irreducible with a projection onto P n of
degree ~ 2; let L be a linear variety of dimension d and let XL be the
pullback over L:
XL
Pn
If XL is irreducible, as the degree of X L ~ L is equal to the degree of

X ~ P n we see that n n L is thin.
It remains to tell for which linear varieties L E Gn,d the variety XL is
irreducible. The answer is provided by Bertini's theorem: this is true for a
non-empty open subset of Gn,d (see [H, Chapter II, Theorem 8.18]).
Remark. The linear subspaces L for which n n L is thin can be described

more precisely. As the given covering of P n is non-trivial, it ramifies over a
non-empty hypersurface H. If L is a line which only intersects H at smooth
points of H, and transversally at each of these points, then XL ~ L is
irreducible (thus n n L is thin).
Over C, we can see this by topological arguments: the property is true
for the generic line and if a line L of the preceding type is moved slightly,
one passes from L to its transform by a homeomorphism.
Example. X = "Double plane": T2 = 4>, where 4> is a polynomial of degree
4 defining a non-singular quartic of P 2.
The covering is ramified along the quartic. If L cuts the quartic in 4
distinct points, the intersection has for its equation T2 = polynomial with
4 simple roots, that is to sayan elliptic curve.
128
If L is an ordinary tangent cutting the quartic in 2 other distinct points,

or if L is an inflection tangent, we obtain an equation T2 = polynomial of
degree 2, thus a curve of genus 0, which projects on the line with degree 2.
Finally, if L is one of the 28 bitangents, we obtain an equation T2 = a
square, which decomposes in two curves of genus O.
We can therefore take U to be all the lines minus the 28 bitangents.
Extension of the ground field.
Proposition. Let kl be a tinite extension of k. IfO I C P nCkd is thin over

k1' then 0 = 0 1 n Pn(k) is thin over k.
Proof.
It is enough to prove this with P n replaced by An. We may assume
that 0 1 = 1f1(X1(kt}), where 1fl : Xl -+ An is a k-morphism without a
rational section, Xl being some k-variety, which we may suppose to be
affine. This last assumption allows us to apply the "restriction of scalars"
functor Rkt/kl which transforms it to an affine k-variety X = Rk1/kXI.
[A characteristic property of X is that it represents the functor Y 1---4
Morph k1 (Y, Xl), i.e. for every k-variety Y, we have a natural bijection
Morphk(Y, X) = Morph k1 (Y, Xd. Taking Y to be a point, this gives
X(k) = Xl(kd. For more details, see rOe], App. 2, where Rkt/k is denoted llkt/k and is defined in a more general setting.]
By applying this functor to the krmorphism 1fl : Xl -+ An we get a
k-morphism 1f : X -+ Y, where Y = Rk1/k(An) (which is isomorphic to
Adn, if d = [kl : k)). But since An is a k-variety, there is a "diagonal"
morphism (3 : An -+ Y ([Oe], A.2.6), which on k-points corresponds to the
natural injection
An(k) ---+ Y(k) = An{kt}
Since 0 = An(k) n 0 1, we have 0 = (3-1(Od
define Z by the Cartesian diagram
= (3-1 (1f(X(k).
-+
X = Rkt/kXl
An
-+
Let us now
1
= Rkl/kAn
If cfJ is the projection of Z on An, we have 0 = cfJ( Z (k )). Hence to prove that
o is thin, it is enough to check that the k-morphism cfJ : Z -+ An does not
have any rational section. But if 8 : An -+ Z were such a section, it would
give a rational map a 0 8 : An -+ X, which, by the universal property of
X (extended to rational maps) would give a kl-rational map 81 : An -+ Xl
129
which would be a rational section of 11"1, contrary to the hypotheses. This
concludes the proof.
Example. Take the fields to be Q and Q( i) and the equation z2 = it 1

defining a thin set fh over Q(i)j we seek the tl E Q such that there is a
solution:
that is to say
We eliminate y:
4X4 -
ti = O.
This is an equation with rational coefficients, which factorises as
As 2i is a square in Q(i), these factors over Q(i) are simply z2 - itl and
z2 + it1'
Exercise. Let X be an irreducible algebraic variety over k, and 11" : X -+ P n

a dominating morphism. Show that there is a thin set in P n outside of
which the fibres are irreducible (over k).
(First take the case where X has dimension n, then for the general
case, factorise 11" birationally in order to have first an absolutely irreducible
fibre and then a finite fibre.)
9.5. Hilbertian fields.

Definition. A field k is called hilbertian if for all n 2:: 1, P n (k) is not thin.
Remarks.
1). For k to be hilbertian, it is enough that P1(k) is not thin. This follows
from the theorem of 9.4. One might also replace P1(k) by A1(k).
2). If k is hilbertian, the complement of a thin set n of Pn(k) is Zariski
dense (otherwise P n (k) would be the union of n and a proper closed subset,
hence would be thin).
3). If k is hilbertian, then for every integer d 2:: 2, k* /k*d is infinite.
In particular a local field is not hilbertian (neither is an algebraically
closed field). Thus Qp, R, C are not hilbertian.
4). If k is hilbertian and kl is finite over k, then k1 is hilbertian (this follows
from the last proposition of 9.4).
130
The converse does not always hold ([FJ], Ex. 12.17).
-5). For every field I<, the field IT) is hilbertian. Indeed, let n be a thin
set in IT). Then the intersection of n with the set of polynoinials aT + b
of degree 1 with coefficients in I< is contained in a closed subset i= I< x K
for the Zariski topology on K x I< (this result follows directly from Bertini's
theorem (see [Ro])). Hence n cannot be equal to K(T).
Therefore, if k is hilbertian, every finitely generated extension of k is
hilbertian.
For more examples of hilbertian fields see [FJ, ChapterI2].
9.6. The irreducibility theorem: elementary proof.
Hilbert's Theorem. The field Q is hilbertian.
Hence number fields, and more generally finitely generated extensions
of Q, are hilbertian.
Proof.
Here is a variant of Lang's proof [L2, Ch. VIII] which is itself a variant
of Hilbert's proof. We consider a thin set in the affine line A 1 (Q) = Q,
and we examine its intersection with Z, say n. For B > 0, let WB be the
number of wEn with Iwl $ B. We prove the theorem in the following
refined form.
Theorem. We have
WB
= O(Bl-E), for some
> O.
The proof will give and 0 effectively. (We shall see in 9.7 that can
be taken to be
We may assume that we have a curve X which is absolutely irreducible,
and a finite morphism X -+ Al of degree ~ 2, and that the points of n
lift to rational points of X. We may complete X to a projective curve X
above PI, and we write Xoo = X-X. The points of n lift to rational
points of X which form a quasi-integral set. In the affine ring of X, take
a function which is not in the affine ring of Al (that is to say, which is
not a polynomial). We take a branch of the curve X in a neighbourhood of
infinity and take the corresponding Puiseux expansion of :
!.)
(x)
= Lc~x\).
-t
-oo,c~ E R;
the series converges for sufficiently large real x.

Multiplying by a constant, we may assume that (x) E Z for all
x E n. (This follows from the fact that is integral over Q[X].) We are
thus led to prove the following statement.
131
Theorem. Let
</>(t) =
CAt", (CA E R, A E Q)
A.... -oo
be a Puiseux expansion in the neighbourhood of infinity, convergent for

t ~ R. Let n", be the set oft ~ R such that t E Z and </>(t) E Z. H </> is not
a polynomial, then the number oft EO", with It I ~ B is O(B1-E) for some
> o.
Proof.
There is an integer n ~ 1 such that all the exponents of the series for
the n - l'st derivative </>(n-l)(t) are < 0, and this series is not identically
zero. In particular
</>(n-l)(t) '" clt-I-' , as t -+
00,
with J.t
> 0, Cl
i= o.
The idea of the proof is to show that n points of 0", cannot be too
close; more precisely
Lemma. There are two constants a > 0, C > 0 such that if t is sulIiciently
large, the interval [t, t + eta] contains at most n - 1 points of 0",.
Let tl < ... < tn be sufficiently large elements in 0"" and Yi = </>(ti).
We construct the interpolation polynomial P of degree n - 1 such that
P(ti) = Yi; this is given explicitly by the Lagrange formula
peT) =
Yj nl<i<n,i~j(~
j=l nl~i~n,i~j(tJ
=t~)
t.)
Then </> - P vanishes on tl, ... , tn. By Rolle's theorem, there is

such that
.
</>(n-l)(~)
E [tl' t n ]
= p(n-I)(~).
As P has degree n - 1, p(n-I)(~) is the highest degree term multiplied by

(n - 1)!
p(n-l)(~) = (n - I)!
L nI9~n,i~j(ti
Yj
.
- tj)
n
j=l
This is a rational number whose denominator is at most
II
I~i<j~n
(ti - tj)1
< (tn
- tl)n(n-l)/2.
132
On the other hand, as tl is sufficiently large, we have
If we put 1 = tn - tb we have lP(n-1)(OI 2:
1/z n(n-l)/2
i.e.
whence
1 2: etf, with a = 2j.t/n(n - 1), and e = e;2/n(n-l).
We deduce from this lemma an upper bound for the number of points
of ~l,/> between 1 and B in the following way. We put = 1/(1 + a) > O.
Cut [l,B] into [l,Be] and [B",B], then decompose [Bt:,B] into O(B/BOlt:)
equal intervals of length < eBOlt:; in each of these small intervals there are
at most 0(1) points of 04>, thus in [1, B] there are at most O(Be + B1-0lt:)
points of 04>, which gives O(B") since = 1/(1 + a). As a > 0, we have
< l.
The estimate O(Bl-t:) for WB shows that the complement of a thin set
of Z contains infinitely many prime numbers, infinitely many sums of two
squares etc.
9.7. Thin sets in PI: upper bounds.
We continue the study of thin sets by showing that over a number field
these sets are "small". We consider only the case of 1 variable; for the
general case see chapter 13.
Let K be a number field of degree dover Q. Let X be an irreducible
algebraic curve over K. Consider a morphism 1r : X --t P 1 of finite degree
{) 2: 2. We wish to show that 1r(X(K is small in a suitable sense in
P 1 (K) = K U {oo }. We do this by two different methods.
First method.
We count the rational points of bounded height. For B 2: 2, denote by
N(B) the number of t E Pl(K) with H(t) S; B, and Nx(B) the number of
t E 1r(X(K with H(t) S; B.
With our normalisation of the height, Schanuel's theorem (2.5) gives
us: N(B) rv eB 2d , for some e > O. For example, over Q we have d = 1,
e= 12/1r2.
133
Theorem. We have Nx(B) = O(Bd) as B
--t
00.
Proof.
This is a simple exercise on heights. We may assume that the curve
X is smooth and projective (this only changes a finite number of points).
We then have a covering map 7r : X --t P l which gives a height on X:
H(x) = HpI (7r(x. This is the height on X associated to 7r* D where D is
the ample line bundle 0(1) on Pl. Thus up to a bounded factor, Nx(B)
is the number of x E X(K) with H(x) ~ B.
There are essentially 3 cases to consider.
Case 0: X has genus O.
If X has no rational point, we have Nx(B) = O. If X(K) =f 0 then X
is a projective line. Let 4> : X --t P l be an isomorphism. The divisor 7r* D
has degree equal to the degree 8 of 7r and the associated height is equivalent
to the 8-th power of the height on P l :
H(x) X H(4)(x))O,for x E X(K).

Therefore
since 8
2.
Case 1: X has genus l.

If X(K) =f 0, we take a point of X(K) as an origin, and we have an
elliptic curve. Let p be the rank of X(K). Then
Nx(B):;::: (logB)p/2,
(cf.4.5).
Case 2: X has genus ~ 2.
We have seen by Mumford's theorem (5.7) that
Nx(B) = O(1oglogB).
[Mordell's conjecture gives Nx(B) = 0(1).]
The theorem is therefore proved.
Note that only in the case of genus 0 does Nx(B) increase as a power
of B and in this case the result is effective.
It ought not to be difficult to prove an effective upper bound of the
form O( (log B)A) in the case of genus ~ 1. For elliptic curves, we do not
134
know how to compute p effectively, but we know how to bound it. One uses
then a lower bound for hex) when h is the Neron normalised height and x
is a non-torsion rational point (d. [Ma2]). For genus ~ 2, one can bound
Nx(B) just by working in the jacobian.
Second method.
We count the number of integral points, as in Hilbert's proof (we would
also be able to count the S-integral points).
Let A be the ring of integers of K. For a E A, we write lIall = sup 100ai
where 0" runs over the the embeddings of K in C. An equivalent norm is
obtained by writing A ~ Zd and imposing a norm on Rd. The number
NCB) of a E A with lIali =5 B satisfies
N(B) ~ Bd.
Let 0 be a thin set, 0 C A. We define
Nn(B) = number of a E 0 with

Theorem. We have Nn(B)
lIall =5 B.
= O(B d/ 2).
For example for K = Q, if 0 is a thin set contained in Z, the number of

points of 0 between -B and +B is O(Bl/2). This estimate is best possible,
since {n 2 , n E Z} is a thin set. The estimate obtained above in 9.6 was
O(BI-E) for some > O.
Proof.
The principle is the same as for rational points. Let Xoo be the inverse
image of 00 E Pl. The affine curve Xaff = X - Xoo projects onto Al =
PI - {oo} by a finite, and thus proper, morphism.
Let Ox be the set of rational points of X whose image by 1r is an
integral point of Al and let 0 be the image of Ox. Since 1r is proper, Ox
is a quasi-integral set in Xaff. By Siegel's theorem, Ox is finite (thus 0 is
also) except when the genus of X is zero, and Xoo has just 1 or 2 points.
Let us consider these two exceptional cases.
First case: X has genus 0 and Xoo = {Pl.

As Xoo reduces to a point, this point is rational. We can therefore
identify Xaff with AI. Then 1r is given by a polynomial: 1r(x) = aoxm +
... + am, with ai E K, and ao =I O. We have m = 6 (the degree of the
polynomial is equal to the degree of the morphism). Let c be a positive
135
integer such that the Wi are integers, (0 ~ i ~ m). If 7I"(x) is an integer

with x E K, then CQ'oX is an integer. To count the x E K such that 71"( x) E A
and 117I"(x)1I ~ B, it suffices to count the x E A such that Ilxll ~ rB I / m , for
some fixed r > 0: their number is ;:::: Bd/m. Therefore
Nn(B)
= O(B d / m ),
with m ~ 2.
Second case: X has genus 0 and 2 points at infinity.

After a ground field extension, Xaff becomes isomorphic to PI -{O, oo}.
The set U of units, corresponding to the affine equation xy = 1, is an
integral set, and the quasi-integral sets are finite unions Ui CiU. We are
thus led to count the units of bounded height. We take fundamental units
ll ... , A, and write = W~l ... ~~, where w is a root of unity, and ni E Z.
For C E K, c#-O fixed, we have logH(c) ;:::: sup Inil. Therefore the number
of units with H(c) ~ M is O({logM)A).
Under the morphism X ~ PI -+ PI, any height on X ~ PI is a power
of the standard height on PI:
logH{x) ;:::: logH{7I"{x)) ;:::: log 117I"{x)II.

We take M to be a power of B. Then
Nn{B) = O({logB)A).
The theorem follows from this.
The above relation between integral points and Hilbert's theorem had
already been pointed out by Siegel in his first paper [Sil].
S -integral points.
We take the ground field to be Q. Take
finite set of prime numbers.
We write
Q'
E Q and S = {PI. ... ,PI} a
Qa,S = {t E Qi t -
Q'
is of the form IIpfi, mi E Z}.

i=l
In particular, for Q' = 0, Qo,s is the set of units of

111
Zs = Z[-, ... , -] = Z[--].

PI
PI
PIPI
If S #- 0, Qa,S is an infinite set. It is a quasi-integral set for PI - {Q',oo}

over the ring Z s.
136
Theorem. Let n be a thin set in Q and S a nnite set of primes. For a
outside a nnite set, n n Qo:,s is nnite.
Proof
We may assume that n is contained in 7r(X(Q)), where 7r, X are as
above. We choose a such that 7r- 1 (a) has at least two elements (for example, we take a different from the ramification points, since the degree of 7r
is ~ 2). We take Xoo = 7r- 1(00 U {a}). Then Xoo has at least 3 points.
From the Siegel-Mahler-Lang theorem, every quasi-integral set with respect
to Zs over Xaff = X - Xoo is finite, hence the theorem.
Exercises. 1). Generalise the above theorem to arbitrary number fields.

2). Show that every thin subset of Z is contained in the finite union of sets
of the following type:
a) a finite set,
b) the intersection of Z with the set of values of ( x, y), where is a rational function with coefficients in Q and x, y run through the solutions
of some Pell equation x 2 - Dy2 = 1 (where D > 0 is a non-square).
c) the intersection of Z with the image of a polynomial of degree b ~ 2
with rational coefficients.
If n is of type (b), resp. (c), show that Nn(B) is of order of magnitude
log B, resp. Bl/o.
137
10. CONSTRUCTION OF GALOIS EXTENSIONS
The following conjecture is part of the folklore.

If G is a finite group, there is a Galois extension F of Q whose Galois group
is G.
This applied to G x G x ... X G would imply that there are infinitely
many disjoint extensions F of Q with Gal(FIQ) = G; the same statement
for number fields would then follow.
This conjecture is known when G is solvable, as proved by Shafarevich,
d. [Sh], [Is] (both proofs [Sh] and [Is] contain a mistake relative to the
prime 2, but this is believed to be repairable); for more complete results in
the case of odd order, see [Ne].
As we will see, Hilbert's irreducibility theorem can be used to handle
other cases.
10.1. The method.
Theorem. Let]{ be a hilbertian field and G a finite group. If there exists
a Galois extension FI ]{(Tl , ... , Tn), with Galois group G, there is such an
extension for]{. Moreover, ifF is ]{-regular (i.e. ]{ is algebraically closed
in F), there exist infinitely many such extensions which are linearly disjoint.
Proof. The first assertion follows directly from proposition 2 of 9.2. Recall
that the extension ]{ is obtained by "specialising" T = (TI, ... , Tn) to t.. =
(tl' ... , in) E ]{n, and that the set of 1's which give a Galois extension FJ]{
with Galois group G is the complement of a thin set OK.
Assume now that F is ]{-regular, and that we have constructed m
linearly disjoint Galois extensions F l , ... , F m of ]{ with Galois group G. Let
]{' be the composite of Fl, ... ,Fm Since F is ]{-regular, the extension
]{'FI ]{'(Tl , ... , Tn) still has Galois group G. Applying the proposition in
9.4, we see that }(n n OK' is thin in ]{n. Since]{ is hilbertian, we can
choose f E ]{n, with f OJ(' and the corresponding extension FJ]{ has
Galois group G, and is linearly disjoint from ]{'. By adding it to F l , ... , F m
we get m + 1 extensions of the required type. The theorem follows by
induction on m.
138
Corollary. Assume that G has the property:

(GalT) - There exists a regular Galois extension ofQ(T) with Galois group
G.
Then every hilbertian field of characteristic zero has an infinite number of
Galois extensions with Galois group G.
Proof.
If E/Q(T) is regular with Galois group G, and K is any field of characteristic zero, then F = T0Q(T) K(T) is a regular Galois extension of K(T)
with Galois group G. One then applies the theorem to F.
Remark. An optimistic conjecture is that GalT is true for every finite group
Gj this is stronger than the conjecture above.
Exercises. 1) Prove that GalT is true for all finite abelian groups (whether
this extends to solvable groups is unknown).
2) If GalT is true for two groups G l and G 2 then it is true for G l x G 2
[Use an automorphism of Q(T) to make the ramification sets disjoint.]
10.2. Extensions with Galois group Sn.

A first example is to apply the method to the generic equation of degree
n:
xn + TlX n - l
+ '"
+Tn = 0,
which has Galois group Sn over K(Tb ... , Tn). Hence if K is hilbertian, it
has infinitely many disjoint extensions with Galois group Sn.
A more interesting example is that given in Hilbert's original paper
[Hi], namely:
Theorem. Let I(X) = xn + a1X n - l + ... + an, n ~ 2, be a polynomial

over a field K of characteristic 0, such that
a) the roots f3l, ... ,f3n-l of I' are simple,
b) the values of I at the points f3i are distinct.
Then the equation I(X) - T = 0 is irreducible over K(T) and has
Galois group Sn.
Corollary. There is a thin set n c K such that for every t E K - n, the
equation I(X) - t = 0 has Galois group Sn. Further, if K is hilbertian
139
tben tbere are inJi.nitely many linearly disjoint extensions of K witb Galois
group Sn.
For example over Q, xn - X - t = 0 has Galois group Sn for all t
outside a thin set, in particular for infinitely many prime numbers t.
We shall need a lemma on finite groups.
Lemma 1. Let n be an integer ~ 1, and let G be a subgroup of Sn. Assume

tbat
i) G is transitive
ii) G contains a transposition
iii) G is generated by cycles of prime order.
TbenG=Sn.
Proof
Let E be a finite set of n elements, SE the symmetric group of E, and
G a subgroup of SE with the properties (i),(ii) and (iii). Let us show by
induction on m that if m is an integer:::; n, there is a subset Em of E with
m elements such that G :) S Em (where S Em is identified with a subgroup of
SE with trivial action on E - Em). For m = n this will show that G = Sn.
For m = 1 there is nothing to prove. To make the step from m to
m + 1, where 1 :::; m < n, we consider two cases.
(a) There is a transposition T = (xy) E G with x E Em, y Em.
In this case we select Em+! = EmU{y}, Let H be the group generated
by S Em and T. It is clear that H is a subgroup of G which stabilizes
E m+l, that H fixes E - E m+b that H acts transitively on E m+b and
that H contains S Em' The index of S Em in H is at least m + 1 therefore
H = S E"'+l' and we have shown that G contains S Em+l .
(b) If a transposition T = (xy) belongs to G, then either x and y E Em or
x and y E E - Em.
Let N be the subgroup generated by the transpositions of G. By (b),
N stabilises Em, and there are at least two orbits of Non E. We decompose
E into orbits under the action of N:
E=FI U ... UFk,
with k ~ 2, and the Fi are pairwise disjoint. The subgroup N is normal in

G, consequently the Fi are permuted by G; but G is transitive, therefore the
Fi are permuted transitively by G, and have the same number of elements.
From hypothesis (ii), the Fi are not all reduced to 1 point, therefore
IFII = .. , = IFkl ~ 2.
140
Let u = (Xl X p ) be a cycle of prime order p belonging to G. Let us
show that UFi = Fi for all i. Indeed, if i l is an index such that UFil =1= F il ,
define i 2 , .. , ip by UFik = Fik+l' (1 ~ k < p) and since p is prime, the indices
ib i 2 , , ip are distinct. Let x and y be two distinct elements of Fi l ; the
orbits under {I, u, ... , u p - l } of x and y are distinct, and of order p, which
is impossible for a cycle.
From hypothesis (iii), we have UFi
Fi for all u E G and all i, which
contradicts the fact that the Pi are permuted transitively by G with k ~ 2.
Hence case (b) is impossible and this concludes the proof of the induction
step.
We go back to Hilbert's example: let f be a polynomial of degree n

over K, such that the roots /31, ... ,/3n-l of f' are simple, and the f(/3i) are
pairwise distinct. We wish to show that the equation f(X) - T = 0 is
irreducible with Galois group Sn for generic T, therefore also for T E K - [2
with n thin.
We map the projective line to itself by f which is a morphism of degree
n, and K(X, T) is an extension of K(T) of degree n (the irreducibility of
the polynomial f(X) - T over K(T) is clear). Let F be the Galois closure
of K(X, T) over K(T) and let G C Sn be the Galois group of F over K(T),
and H = G n Sn-l the Galois group of F over K(X, T).
F
G
)H
K(X)
if
K(T)
The homogeneous space G / H may be identified with the set of roots of the
equation f(X) - T = O.
We use a lemma on coverings of the projective line. Over a field of
characteristic 0 (that we may assume is algebraically closed; we may even
assume that it is C), we consider a Galois covering 1f : X ---4 PI, with X
irreducible. Let G be the Galois group.
Lemma 2. Let Q E Pl. The inertia groups of the points x E X for which
1f(x)
=1=
Q generate the group G.
Taking the subgroup generated by these inertia groups, and passing to

the quotient, this reduces to the following lemma.
141
Lemma 3. If a Galois covering X
Q, then it has degree 1.
PI is unramified except at one point
This is clear over C, because 7rl(PI - {Q}) = {1} (the affine line is
simply connected).
[More generally, Riemann's existence theorem allows one to describe
the Galois coverings of P l (over C, or over any algebraically closed field
of characteristic 0) which are unramified outside a given set of points
{QI, ... , Qm}: the Galois group G of such a covering can be generated by m
elements O"I, ... , O"m (which are generators of some inertia subgroups above
the Qi's), with the relation O"I"'O"m = 1.
Conversely, any finite group G with such generators can be obtained
in this way. This follows from the structure of the fundamental group of a
sphere with m punctures.]
We can now complete the proof of the theorem.
The zeros of f' are 131, ... ,13n-l. Hence the covering X ~ PI, corresponding to the field extension FjK(T), is unramified outside Q = 00 and
1(131), ... , 1(13n-l). Using the fact that the 1(13i) are distinct, one sees that
the inertia subgroups G p of G at a point P above one of the 13i is cyclic
of order 2 and generated by a transposition. Since G is generated by such
subgroups, lemma 1 shows that G is equal to Bn.
Complements. Here is some more information on this example. Assume K

is a number field, and let B be a finite set of places of K containing the
archimedean places, and A the ring of S-integers of K. Let 11 / ,A be the set
of tEA such that I(X) - t does not have Bn for its Galois group. It is
clear that if t is a value of I, then t E 11 I,A:
11 / ,A :::> An I(K).
Theorem. If n
5, then 11 I,A is the union of An I(K) and a finite set.
The proof of finiteness will unfortunately be ineffective.

Corollary. Given n
5, the equation
xn -
X - P = 0 has Galois group
Bn over Q for all prime numbers p except a finite number.
Indeed, no prime number p > 2 is of the form xn - X, X E Q, therefore

for I(X) = xn - X and A = Z, 11 I,A contains only a finite number of prime
numbers.
142
Proof of theorem.
Consider the curve X attached to
of PI, with Galois group 8 n
f as before; it is a Galois covering
If t belongs to nf,A' then either t is equal to one of the fUji) (in which
case f(X) - T has a double root) or t can be lifted to a rational point of
some XIH where H is a subgroup of 8 n distinct from 8 n . The subgroup
8 n - 1 (and its conjugates) corresponds to An f(K). We have to show that
the others only give a finite set. This follows (via Siegel's theorem) from
the following lemma.
Lemma 4. If n ~ 5, and if H is a subgroup of 8 n distinct from 8 n , and

not conjugate to 8 n - b then the covering XI H -+ PI is not of exceptional
type, i.e. it is not of genus a with at most two points at infinity.
Proof.
a) If H is the alternating group An, XI H -+ PI is quadratic, and ramified
at n - 1 points or at n points according to whether n is odd or even. As
n ~ 5, this gives an hyperelliptic curve of genus ~ 1.
b) In the other cases, the representation of 8 n in 8 m (where m = [G: H])
attached to H is faithful (use n ~ 5 and the simplicity of An). This implies
in particular m ~ n (for n! must divide m!) and even m > n if n =I- 6 (for
the automorphisms of 8 n are inner for n =I- 6).
Let p : 8 n -+ 8 m be this representation. The image by p of a transposition is an element of order 2, i.e. a product of p transpositions (p ~ 1),
fixing m - 2p elements. On the other hand, a cycle of degree n gives by p a
product of cycles of degrees nIdI, ... , nldh' with m = E nIdi, gcd(di) = 1.
As the inertia at infinity is cyclic of order n, if XI H is of exceptional type
then h ::; 2 since h is the number of points at infinity in XI H.
On the other hand, Hurwitz' formula for the genus gives
2g - 2 = -2m + (n -l)p+
L(;. -1),
i=1
'
where 9 is the genus of XI H. In the exceptional case we have 9 = 0, h = 1

or 2, whence
m = (n - l)p + 2 - h.
143
i) If h = 1, we have m = nidI, and as m 2': n this implies d l = 1, m =
n, whence m = n = 6, and p is a non-inner automorphism S6 --t S6.
This automorphism transforms a transposition (ab) to a product of three
transpositions: (ab)(cd)(ef). We have thus p = 3, and the equation m =
(n - l)p + 1 is violated.
.
ii) If h = 2, we have m = nidI + nld2 , whence m :S 2n. As m = (n - l)p
this gives p :S 2n/(n - 1) < 3, whence p = 1 or 2. If p = 1 we have
m = n -1 < n, impossible. Therefore p = 2 and m = 2(n -1). But then d l
and d2 are not both equal to 1: one of them is 2': 2, and nidI + nld2 :S ~n.
Whence 2(n - 1) = m :S ~n and n :S 4.
Remark. For n = 4 and H = D 4 , the dihedral group of order 8, one finds
that XI H has genus 0 and there are two points at infinity. This corresponds
with the above notation to m = 3, P = 1, h = 2, d 1 = 2, d2 = 4. In this
case Sn --t Sm is not faithful.
Exercises. 1) Give an algebraic proof of lemma 3, by using Hurwitz' formula for the genus of a ramified covering. Show that the results holds in
characteristic p > 0 provided the covering is tamely ramified at Q.
2) Let E be a finite set with n 2': 1 elements and G a subgroup of SE. Define
a relation R on E X E by
R(x, y) = {x = y or the transposition (xy) belongs to G}.

a) Prove that R is an equivalence relation on E (use (xz) = (xy)(yz)(xy)
if x, y, z are distinct elements of E).
b) Let F be an equivalence class mod R. Show that SF is contained in G.
c) Let F I , ... ,Fm be different equivalence classes modulo R. Prove that
SFl x ... X SF", is a normal subgroup of G.
d) Let G be the group of permutations of E = E IR defined by G. Show
that 8 ESE belongs to G if and only if it is compatible with R and its
image in SE belongs to G.
e) Use d) to recover lemma 1.
3) Let F(X) = xn + alXn-1 + ... + an be a polynomial with coefficients in
Z, with distinct roots (in Q). Assume that, for every prime number p, the
reduction j of f mod p has either n distinct roots (i.e. p does not divide
discr(f)), or it has n - 2 distinct simple roots and one double root. Prove
that the Galois group G of f (viewed as a subgroup of Sn) is generated by
transpositions (use the fact that G is generated by the inertia subgroups,
and show that these are either trivial or generated by a transposition).
144
Conclude that, if / is irreducible, then G = Sn and the An extension of
Q( discr(f)) thus obtained is unramified.
4) Let n ~ 2. Using the fact (proved by E. Selmer, Math. Scand. (1956),
287-302) that
X -1 is irreducible over Q, show that its Galois group
is Sn (use exercise 3). Give a direct proof for n = 2,3,4,5.
xn -
10.3. Extensions with Galois group An.

Here is another example of Hilbert.
Let n be an even integer 2 4. Consider a polynomial g(X) of degree
n, divisible by X 2 , such that
where h = n/2 -1. Assume that the f3i E K are distinct and non-zero, and
the g(f3i) are distinct and non-zero.
Theorem. For generic t (thus also for tin K outside a thin set) the Galois
group of the equation g(X) + (_1)n/2t 2 = 0 is An.
For n odd, Hilbert's paper gives a similar equation also leading to the
alternating group An.
Proof
We begin by showing that the Galois group of F(X) = g(X) -T is Sn.
The proof is the same as before; the inertia groups over the f3i are generated
by cycles of order 3, and at 0 by a transposition. Then G is generated by
these elements, and as G is transitive we obtain G = Sn. Hence the Galois
extension OF generated over k(T) by the roots of F(X) has Galois group
Sn.
We now factorise F(X) = g(X) - T in the algebraic closure:
n
F(X) = II (X - O'i),
i=l
where the discriminant is:

discr F = II(O'i - O'j)2.
i<j
Eliminating X between F and its derivative, one finds the classical formula
II Fbj),
n-l
discr F = nn(_lt(n-l)/2
j=l
145
where 11, ... , In-1 are the roots of F'. Here
and the discriminant of F is the product of (_I)1+n/2T with a square.

Hence, the subfield of OF corresponding to An is the field
K(T, VdiscrF) = K(t), where t is a square root of (_I)1+n/2T. This shows
that the Galois group of the equation g(X) + (_I)n/2t 2 is indeed An.
Schur has given some numerical examples of equations with Galois
group Sn or An ("Affektlose Gleichungen"). For example, the Galois group
of 1 + X + ... + xn In! = 0 over Q is Sn if 4 An and An if 41n (see [Schul).
10.4. Further examples of Galois groups: use of elliptic curves.

Let E be an elliptic curve over a field K, and n an integer 2: 1. The
kernel of multiplication by n gives a Galois extension whose Galois group
GE,n embeds in the group GL(2, Z/nZ).
Over Q, Hilbert's theorem shows that G E,n is "usually" equal to
GL(2, Z/nZ), but does not provide any specific example.
Over the field Q(t), where t is an indeterminate, we take an elliptic
curve with modular invariant j equal to t e.g.
+ xy = x
36
- t _ 1728 x - t _ 1728
For all n, the Galois group G E,n over Q(t) is the group GL(2, Z/nZ).
Indeed, we remark first that the image 6 E,n in GL(2, Z/nZ)/(I) is
canonical (independent of the choice of E with j = t). The function field
of the curve which classifies the n-division points is the field of modular
functions of level n with coefficients in the cyclotomic field Q( V'l). The
theory of modular functions implies 6 E,n = GL(2, Z/nZ)/(I). Finally, if
H is a subgroup of GL(2, Z/nZ) such that H.(I) = GL(2, Z/nZ), then
= GL(2, Z/nZ).
(If not, H would have index 2, and as
(1
= (~ ~1)
satisfies (12 = -1 we would have (12 = -1 E H, whence a contradiction.)

Let Kn be the field generated by the n-division points of the generic
146
curve, K n,t that of the corresponding curve E t :
Kn
t
GL2 (Z/nZ)
)SL2 (z/nZ)
Q( v'i)(t)
= (Z/nZ)*
)GL 1 (Z/nZ)
Q(t)
We have
Gal(Kn/Q(t = GL2 (Z/nZ),

Gal(Kn/Q( V'i)(t
= SL 2 (Z/nZ),
Gal(Q( V'i)(t}/Q(t = (Z/nZ)*.

For all n, there is a thin set On C Q (depending on n) such that for
t On, the corresponding curve E t satisfies Gal(Kn,t/Q) = GL2 (Z/nZ).
(When Q is replaced by a number field K one obtains a similar assertion,

assuming that the extensions K/Q and Q( v'i)/Q are disjoint.)
Example. For the curve y2 - Y = x 3 - x of conductor 37, one can prove that
G E,n = GL(2, Z/nZ) for all n not divisible by 74 = 2 X 37. Nevertheless,
there is no elliptic curve for which G E,n = GL(2, Z/nZ) for all n (see [Se6,
5.5, Prop. 22]).
Let us now search for Galois extensions of Q with Galois group the
simple group PSL2 (F p), where p is a prime ~ 5.
If, instead of Q, we were interested in Q(
J'-(_-I)-~-2-1-p), we could use the
GL2 (Z/pZ)-extensions of Q constructed above. This does not work over

Q. Nevertheless, the following result has been proved by K-y. Shih ([Shl],
[Sh2]):
Theorem. Let p be an odd prime number such that (~) = -1, or (~) =
-1, or CJ,) = -1. Then there is a regular Galois extension of Q(T) with
Galois group PS~(F p).
The proof is roughly as follows. Let p* = (-1) 2 P and Lp = Q( #).
Let MN (where N is any positive integer) be the function field Q(j,iN) of
the modular curve Xo(N) (recall that iNeZ) = i(Nz) = i(-l/Nz; let w
.t=.!.
147
be the canonical involution of MN (which permutes i and iN)' Call MN,p
the "twisted form" of MN relative to LpjQ, i.e. the subfield of Lp Q9 MN,p
fixed by a Q9 w, where a is the involution of Lp. (This field has a modular
interpretation in terms of N-isogenies of a curve E onto its a-conjugate
E(T.) Using this modular interpretation, Shih shows:
a) if (~) = -1, there is a regular Galois extension of MN,p with Galois
group PSLz(F p)j
b) if N =2,3, or 7, MN,p is a pure transcendental extension of Q (i.e. it
corresponds to a curve of genus 0 which has a rational point).
For more details, see Shih, loco cit.
This applies in particular to p = 7, where PSLz(F p) has order 168 and
is isomorphic to SL3(Fz). In that case, another construction of a regular
extension of Q(T) with that Galois group has been given by LaMacchia
([LaM], see also [MM])j an especially interesting extension of Q obtained
in this way is the field of roots of x 7 - 7x + 3 = O.
10.5. Noether's method.
Here is a tentative method suggested by Emmy Noether for constructing extensions with given Galois group G. We make G operate faithfully
on a finite set E. We consider the field Q(Tx, x E E) where Tx are indeterminates indexed by E, and make G operate by Tx I-t T(TX for a E G. Let
KG be the field of invariants. If KG is a purely transcendental extension of
Q, the group G can be realised as a Galois group by 10.1.
The simplest try consists of choosing E = G, and making G operate on
itself by translation. But Swan has shown that even for cyclic extensions,
KG may not be pure. The first example of Swan was G = Zj47Z. Lenstra
has analysed this question and his simpler example is Z/8Z [Len].
[It thus appears that Noether's method does not work, at least in the
crude form described above. Nonetheless, one can hope that the field KG,
without being pure, still has enough good properties for some form of the
irreducibility theorem to apply, cf. Ekedahl [E)j see also [Se7).]
If we take G = An operating on n letters, KG is the field generated by
the elementary symmetric functions al,'" ,an and by the square root of
the discriminant. For n = 2,3,4,5, KG is a purely transcendental extension
of Q (see [Mae] for the case n = 5). One would like to know what happens
for other values of n.
10.6. Infinite Galois extensions.
We now extend Hilbert's irreducibility theorem to some infinite Galois
148
extensions. We need first an auxiliary result on profinite groups.
Proposition. Let G be a profinite group. The following properties are
equivalent:
(i) There is an open normal subgroup U of G such that H.U =f. G for every
closed normal subgroup H of G with H =f. G.
(ii) The Frattini subgroup 4>( G) of G is open.
(iii) The prime numbers dividing the order of G are finite in number, and
the Sylow subgroups of G are (topologically) finitely generated.
(iv) There is an open normal subgroup N of G with N = I1 PE P N p, where
P is a finite set of primes, and each Np is a finitely generated pro-p-group.
(Recall that the Frattini subgroup 4>( G) of G is the intersection of the

maximal closed proper subgroups of G, just as in the finite case, cf. [Hup],
Kap. III,3. One has 4>(G) = lim4>(GjV), where V runs through the open
normal subgroups of G.)
...-
Proof
The equivalence (i) {:} (ii) is clear; so is the implication (iv) =?- (iii).
Note that if P is a pro-p-group, 4>(P) = (P.p)cl.pp is the intersection of
the kernels of the continuous homomorphisms P -+ ZjpZ; hence 4>(P) is
open if and only if P is (topologically) finitely generated.
Let us now assume (iii) and prove (i). Let Pl, ... , Ph be Sylow subgroups
of G corresponding to the primes Pi, ... ,Ph dividing the order of G. By
assumption, 4>(Pi ) is open in Pi. Hence there is a normal open subgroup U
of G such that Un p.. C 4>(Pi ) for all i. We show that U has property (i).
Note first that the Sylow subgroups are conjugate and that U is invariant
under conjugation, thus the property un Pi C 4>(P") is satisfied for all the
Sylow Pi-subgroups.
Let H be a closed subgroup such that H.U = G. Let Hb ... , Hh be
Sylow subgroups of H for the different primes Pb ... ,Ph. We can select
the Pi so that Hi C Pi. As H maps onto GjU, Hi maps onto a Sylow
pi-subgroup of GjU, thus the images of Hi and Pi in GjU coincide:
Pi = HdU n Pi).
As Un Pi C 4>(Pi) (by con~truction), we find

Pi = Hi.4>(P..)
therefore Pi = Hi. But a group is generated by a choice of its Sylow psubgroups for each p (this is true by Sylow's theorem for finite groups, and
this extends to profinite groups). Whence H = G.
149
It remains to prove that (ii) :::} (iv). Since N = <p( G) is pro-nilpotent
([Hup], p.270, Satz 3.6), it can be written as a direct product N = TIpEP N p ,
where the Np are pro-p-groups. By assumption N is open in G; if S is a finite
set of elements generating G and N, then S generates G (topologically); this
shows that G, and hence N and the N p , are finitely generated. If p is a
prime which does not divide [G : N], then p does not divide the order of N
(see [Hup], p.270, Satz 3.8); this shows that Np = 1 for almost all p, and
concludes the proof.
Examples. 1) A finite product of compact p-adic Lie groups has properties

(i)-(iv) above (property (iv) is especially easy to check).
2) The group Z = TIp Zp does not have these properties.
Profinite groups G having properties (i)-(iv) behave as well as finite
groups for Hilbert's irreducibility theorem.
Namely, let G be such a group, and FIK(T) a Galois extension with
Galois group G. Assume:
(Ram) There is a finite set S of places of K(T) outside of which F is
unramified.
Then if t E K does not belong to S, the decomposition subgroup G t of
t is the Galois group of the corresonding residue field extension Ft/ K; it is
a subgroup of G (defined up to conjugation).
Theorem. There is a thin set 0 C
J(
such that G t
=G
for all t E K - O.
Proof
Let N = <p(G) be the Frattini subgroup of G and let FN IK(T) be the
corresponding extension of J( (T). The Galois group of FN I K (T) is GIN,
which is finite by assumption. By Proposition 2 of 9.2, there is a thin set
o of K such that, if t 0 US, the decomposition group (G / N)t is equal
to GIN. For such a t we have Gt.N = G, hence G t = G since N is the
Frattini subgroup of G.
Exercise. Use the theorem above to construct extensions of Q with Galois
group GL 2 (Zl) for any given 1.
[10.7. Recent results.
There has been much progress in the 1980's towards realising a given
group as a Galois group over Q, or over Q(T). For a detailed exposition
(resp. a resume), see Matzat's book [Mat] (resp. the Bourbaki report [Se5]).
150
One of the main results concerns groups having a rigid family of rational conjugacy classes. Let us define "rational" and "rigid":
A conjugacy class C of a finite group G is called rational if the following
equivalent properties hold:
(a) every character of G takes rational values on C;
(b) if g E C, one has gm E G for every integer m which is prime to the
order of g.
Let now C l l ... , Ck be conjugacy classes of G. Assume G has trivial
centre. Define P = P( C l , ... , Ck) as the set of all (g1, ... , gk) E C l X ... X Gk
such that gl ... gk = 1i define P' as the subset of P made of those (gl, ... , gk) E
P which generate G. The group G acts by conjugation on P and P': an
element 9 E G transforms (g1, ... , gk) into (gglg-1, ... , ggkg- 1 ). One says
that (Ct, ... , C k) is rigid if P' =1= 0 and G acts transitively on P' (i.e. P' /G
has one element).
Note that G acts freely on P'. Hence the rigidity property is equivalent
to IP'I = IGI, and it implies IPI ~ IGI. (This is useful, because IPI can be
computed from the character table of G:
k
IPI = I~I !!ICil~x(cl) ... X(Ck)/X(I)k-2,
where Ci is a representative element of Gi and X runs through the irreducible

characters of G, d. [Se7], 7.2.)
Theorem. (Belyi-Fried-Matzat-Thompson). Let G be a finite group with
trivial centre, and G1 , , Gk be conjugacy classes of G. Assume that each
C i is rational and that the family (G ll ... , Ck) is rigid. Then there exists a
regular Galois extension of Q(T) with Galois group G.
Hence G has property GalT of 1O.1. In particular, G can be realised
as a Galois group over every hilbertian field of characteristic zero.
The proof gives a more precise result. One selects k distinct rational
points tt, ... , tk of the projective line PI, and one constructs a regular Galois covering X -- PI over Q with Galois group G having the following
properties:
(1) it is unramified outside t1, ... , tk;
(2) for each i, 1 ::5 i ::5 k, the inertia group in G relative to ti (which is well
defined up to conjugation) is generated by an element of Ci.
(Idea of the construction : one does it first over C, using the explicit
description of the fundamental group of a sphere with k punctures; then, one
151
shows that the covering so obtained is unique, up to unique isomorphism;
by descent theory this implies that its field of definition can be lowered from
C to Q.)
The condition that G contains a rigid rational family of conjugacy
classes is rather restrictive. It can be weakend in several ways (see [Mat],
[Se5], [Se7]). Using the variants so obtained, the property GalT has been
proved(l) for the following groups:
An,Sn;
PSL3(Fp ),p == 1(mod4);
PSp4(Fp ),p 2: 3,p == 2,3(mod5)
G 2 (Fp ),p 2: 5;
All the sporadic groups (Mu, M 12, J 1, ... ), except M 23 .
Explicit equations have been given for some groups: PSL2(F7),
PSL2(Fu), PSL2(F13), SL 2(Fg), M 12
As for groups with non-trivial centre, one may mention An = 2.An
(non-trivial extension of An by a group of order 2): It has been proved
by N. Vila (see e. g. [Mat], Kap IV, 6) that An has property GalT when
n == 0, l(mod 8); the general case was handled later by Mestre [Me2]. Mestre
has also shown (unpublished) that 6.A6 and 6.A7 have property GalT.
(1) In some cases, the proof assumes the classification of finite simple
groups.
152
11. CONSTRUCTION OF ELLIPTIC CURVES
OF LARGE RANK
11.1. Neron's specialisation theorem.
Let k be a number field and A an abelian variety over K = k(Tb ... , Tn)
where the Ti are indeterminates over k. By a theorem of Neron ([Nl]), A(K)
is a finitely generated group.
Theorem. (Neron). There is an abelian variety At over k, of the same
dimension as A, such that
rankAt(k) ? rankA(K).
The variety At is obtained by "specialising" A. This is done as follows:
since K is the function field of P n and A is defined over K, A comes from
some abelian scheme Au over a non-empty open subset U of P n' Let
s}, ... , Sn E A(K) generate A(K). One can view the s/s as rational sections
of the abelian scheme 'Jr : Au --+ U. By replacing U by a smaller open set,
one can also assume that the Si are morphisms [in fact, such a replacement
is not necessary: since U is smooth, the Si can be proved to be everywhere
regular on U]. Hence if t is any rational point of U and S E A(K), set)
is a well defined rational point of At = 'Jr-l(t). The map S 1-+ set) is a
homomorphism 4>t : A(K) --+ At(k).
The precise form of Neron's theorem is:
Theorem. The set oft E U(k) for which 4>t is not injective is thin.
Therefore outside a thin set, rank At(k) ? rank A(K).
We shall use the following criterion on abelian groups.
Criterion. Let 4> : M --+ N be a homomorphism of abelian groups, and n

an integer? 2. We assume that,
1) M is finitely generated,
2) M I nM --+ N InN is injective,
3) 4> is injective on the torsion group of M,
4) 4> defines an isomorphism of Mn = {x EM, nx = O} onto
N n = {x E N,nx = O}.
Then 4> is injective.
Proof
The kernel I of 4> is of finite type from (1), and torsion free, from (3).
We shall show that I = nI, which implies I = O.
153
Let x E I. We have (x) = 0 E nN, therefore, from (2), x E nM.
We write x = ny, with y E M. Then n(y) = (x) = 0, that is to say
(y) E N n . From (4), there is z E Mn with (z) = (y). Since nz = 0, we
have x = n(y - z). But y - z E I, therefore x E nI.
Proof of Neron's theorem.
We use the criterion for y : A(K) -+ At(k), with an arbitrary choice
of n ~ 2. Condition (1) is known. Condition (3) follows from a general
fact on abelian schemes: if s is a section of order exactly n, set) is also of
order n provided that the residue characteristic at t does not divide n - but
here we are in characteristic o. (The n-division points give a sub-scheme
An etale over U. A section of this scheme which is non-zero at the generic
point is non-zero elsewhere.)
To check condition (4), we have to prove that there are no more ndivision points in At than in A. Take the subscheme An which is the kernel
of multiplication by n : A -+ A. As it is etale over the base U, we can
decompose it into a disjoint sum of irreducible subschemes
Let di be the degree of the projection Bi -+ U, and let I (resp. J) be the

set of i's with di = 1 (resp. di ~ 2). If t is chosen outside the thin set
UiEJ 1f(Bi(k, we have
as desired.
To check condition (2), we have to show that, for t outside a suitable
thin set, the map
A(K)jnA(K)
-+
At(k)jnAt(k)
is injective. It is enough to prove that, for every a E A(K), a nA(K),

there is a thin set n". such that aCt) nAt(k) for all t n".. (We construct
this for a finite number of a: some set of representatives of the classes
modulo n).
The image of a: U -+ Au is a subvariety of the scheme Au; let Vn,,,. be
the inverse image by the morphism n : Au -+ Au. Then Vn,,,. is a covering
of U, and the hypothesis a nA(K) translates to Vn ,,,. having no rational
section. We then take n". to be the image of the rational points of Vn,,,.(k).
This concludes the proof.
154
Remarks.
1) The set of t E U(k) such that A(K) -+ At(k) is not injective is in fact
finite (and not merely thin) if dim U= 1. This result is due.to Silverman,
see e.g. [L6] , Chap. 12, 2. The proof is a generalisation of that of ManinDemjanenko given in 5.2.
2) Neron's theorem can be extended to algebraic groups which are extensions of abelian varieties by groups of multiplicative type. More precisely,
if G is such a group, defined over K, and r is a finitely generated subgroup
of G(K), then r "specialises injectively" in Gt(k) for all t outside a thin
set. [Let f' be the set of I E G(K) such that there exists m ~ 1 with
m1 E r. An elementary argument shows that f' is finitely generated. One
then applies to f' the same proof as above for A(K).]
This applies in particular to the multiplicative group G m (but not to
the additive group G a ).
3) Very little is known about specialising non-commutative groups (even
matrix groups).
For instance, let
Over Q(T), A and BT generate a free group r of rank 2; if one specialises T

to t E Q one gets a group r t with a surjective homomorphism >t : r -+ r t .
An explicit description of the t's for which >t is injective is not known. For
partial results see e.g. [LU}.
Here now is a construction of an elliptic curve over Q of rank ~ 9.
We consider 9 points (PI, ... ,P9 ) of the projective plane. We suppose that
these points are "generically independent": if Pi has projective coordinates
(Xi,}Ii, 1), we assume that XI, ... ,X9 , YI, ... ,1'9 are independent indeterminates. Then K = Q(P1 , . ,P9 ) is a purely transcendental extension of Q of
degree 18. There is a unique cubic curve C defined over K passing through
PI, ... , P9 , and this cubic is non-singular. There are 9 obvious points on
this cubic, and an obvious divisor class of degree 3 given by a straight line
section. We denote such a section by D. (If P is the 3rd point of C on the
line PI P2 , we have D rv PI + P2 + P.) We now show that
9
(L=ni~+mD;Lni+3m=O}, inPic(C),
i=l
i=1
155
has rank 9, that is to say that the ~ are linearly independent modulo the
section by a line.
To do so, let kc be the field of definition of C in P2 (Le. the extension
of Q generated by the nine ratios of the coefficients of an equation of C);
put m = tr.degQkc and n = tr.degk c Q(Pl, ... , P9 ). We have m ~ 9, n ~ 9
and m + n = 18, hence m = n = 9. This shows that C is "the generic"
cubic, and that the Pi'S are generically independent over k c . Hence, any
linear equivalence 2:;=1 miPi + mD '" 0 implies 2:;=1 mixi + mD '" 0 for
all Xi E C, which is possible only if m1 = ... = m9 = m = O.
Now it is clear from Neron's -theorem (1l.1) that if we specialise
P1 , .. , P9 outside a thin set we obtain a curve of rank ~ 9.
We give two complements.
1) Effective construction of such curves.

This construction can be made in an elementary way, without using
Neron's general theorem.
Let S = {7,1l,19,23,29,31}. For each pES, we select an elliptic
curve Ep over F p, in Weierstrass form and nine distinct rational points
P1 (P), ... , P9(P) on Ep with the following properties:
(a) Pi(P) is not 0 (i.e. is not infinity);
(b) 2:;=1 Pi(P) =1= 0, i.e. Ep is the only cubic of P 2 going through the
Pi (p);
(c) E7 has no rational point of order 2;
(d) P9 (1l) is not divisible by 2 in Ell(Fll)j
(e) for p = 19,23,29,31, Ep(F p) has a subgroup of type (2,2), and
seven of the Pi(P) are divisible by 2, the other two being generators of
the 2-component of Ep(F p), these two being Pt, P2 for p = 19; P3 , P4
for p = 23; P5 ,P6 for p = 29 and P7 ,PS for p = 3l.
(For the actual construction of the Ep's, see below).
One now chooses 9 integral points P1 , . , P9 in affine 2-space with Pi ==
Pi(P) (mod p) for i = 1, ... ,9, pES. Let E be the cubic through the
Pi (which is unique, because of (b, and let us show that P1 , . , P9 are
independent on E. Indeed suppose there is a relation 2: miPi == 0 mod
D and let us show by induction on sup Imil that all the mi are zero. By
reduction mod 19, we see that m1P1 + m Z P2 is divisible by 2 in E 19 (F 19 ),
hence m1 == m2 == 0 mod 2. Using similarly reduction mod p for p =
23,29,31,11, we see that m3 == ... == m9 == 0 mod 2. Hence 22: yPi rv O.
But since E 7 (F 7 ) has no points of order 2, the same is true for PieCE).
Hence 2: TPi rv 0, and the induction hypothesis shows that all mi are o.
It remains to see that (a), ... ,(e) can actually be met.
156
From Hasse's theorem, the number Np of points of Ep over F p satisfies
and conversely for every integer Np satisfying these inequalities there is a

curve Ep over F p with exactly Np points.
a) For p = 7, we take Np = 11. As 11 ::; 1 + 7 + 2.;7. there is an elliptic
curve E7 over F7 having 11 points. We take any 9 of the 10 non-zero points
(their sum is the negative of the 10th, therefore i= 0). As the number of
points is odd, there is no point of order 2.
Explicitly we can choose y2 = x 3 + 2x - 1, the 11 points being
x=
y=
1
3
2
2
3
2
4
1
5
1
00
00
b) We take a cubic modulo 11 having 16 points, whence a subgroup of order

8. We take for PI (11), ... , Pg(11) the 8 distinct points of this subgroup, and
P9(11) a point outside.
For example, y2 = x 3 + 4x + 6 with 16 points
x=
y=
1 2 3
0 0 1
4
3
6
2
7
5
8 9
0 1
10
1
00
00
c) For p = 19,23,29 and 31, we have p + 1 - 2v'P ::; 28 ::; p + 1 + 2v'P

therefore we can choose a cubic modulo p having 28 points. Such a cubic
has a subgroup of order 4. If it is not cyclic, it has type (2,2). If it is
cyclic, it contains a unique subgroup of order 2; dividing by this subgroup
we obtain a cubic whose points of order 2 can all be shown to be rational
over F p' In the 2 cases one obtains a cubic with a subgroup of type (2,2)
and a subgroup r of order 7, with E(F p)/r ~ Z/2Z x Z/2Z. For p = 19
we take PbP2 as a basis of Z/2Z x Z/2Z, and P3 , ... , P9 are the 7 points
of r. In the same way for p = 23 (with P3 , P4 ), p = 29 (with P5 , P6 ) and
p = 31 (with P7 , P g ).
In the literature, there is an entirely explicit construction [Na).
2) In place of 9 independent generic points, we fix 8 points and specialise
only the 9th.
Let PI, ... , Pg be points of P 2 (Q), pairwise distinct, such that no 3 lie
on a straight line and no 6 lie on a conic.
The cubics passing through PI, ... ,Pg form a I-parameter family ("linear pencil"), that is to say the 8 conditions given by PI. ... , P g are independent (see [GH, p.481]). These cubics pass through a 9th fixed point Po. We
assume that Po is distinct from PI, ... , Pg.
157
If we blow up the projective plane P 2 at PI, ... , P6 we obtain a cubic
surface S of P 3 in which the ~ are blown up to lines D i , (i = 1, ... ,6),
and the plane cubics passing through PI, ... , P6 become plane sections of S;
those we consider correspond to planes passing through P7 and Ps , and Po
is the third point of intersection of the line P7 Ps with the surface S.
Let P9 be a generic point of the plane. We shall show that PI, ... , P9 are
independent modulo D on the elliptic curve C going through Po, PI, ... , P9
We denote by E the surface obtained by blowing up the points
Po, ... , Ps . Let L i , i = 0, ... ,8 be the lines of E corresponding to the ~; let M
be a line of P 2 not passing through the Pi and let M be the corresponding
curve of E. The elementary theory of blowing up (cf. [H), Chapter 5 3)
shows that the Picard group of E (which is here equal to the Neron-Severi
group) is the free abelian group with basis {Lo,LI, ... ,Ls,M} (thus of rank
10).
The linear system on P 2 , defined by the cubics passing through
Po, ... ,Ps , defines a morphism 1r : E -I- PI; the fibres of 1r are the strict
transforms of the cubics passing through the Pi; the hypotheses on Po, ... , Ps
insure that these cubics are irreducible. If T is a generic point of PI and
CT = 1r- I (T) is the generic fibre at T, consider the restriction homomorphism
Pic(E) -I- Pic( CT ).
We have an obvious element in its kernel: the class of a special fibre (note
that all the fibres are linearly equivalent). On the basis {L o, ... , Ls, M} this
class is that of 3M - E~=o Li [indeed, this is the strict transform of a cubic
pasing through Po, ... ,Ps; the total transform would be '" 3M, but in the
strict transform, the lines Li due to the blowing up are removed].
As the fibres (even special fibres) are irreducible, one knows that the
kernel of Pic(E) -I- Pic( CT) is generated by the class of a fibre, i.e. here by
s
3M - Ei=oLi.
We conclude from this that the classes of M and the Li (i = 1, .... ,8)
in Pic(CT) are linearly independent over Z. The interpretation of these
classes is clear: that of M is the linear section (denoted by D) of the cubic
CT and that of Li is Pi (1 ~ i ~ 8) modulo the multiples of D.
Now let P9 be generic in P 2 and take the cubic passing through P9 (and
Po, ... , Ps ). This is the same thing as taking a generic CT and a generic point
P9 on CT. If we have
9
LniPi '" mD, (ni E Z,m E Z),

i=l
158
then, as Pg is generic and I:~=I niPi - mD is fixed, we have ng = O. As
PI, ... ,Ps are linearly independent modulo D we have ni = 0 (1 SiS 8)
and m = 0, therefore PI, ... , Pg are linearly independent as claimed.
By Neron's theorem, this shows that Pg can be specialised so that the
rank of the corresponding curve is ~ 9.
Exercise. Show that the construction in this section (resp. in the next)
gives an infinite number of curves over Q, with distinct j invariants, of
rank ~ 9 (resp. ~ 10).
We start again from a pencil L of cubics passing through 8 general
points PI, ... , P s . We select 2 of these points, say PI, P2, and lines D I , D2
passing through PI and P 2 , respectively (and not passing through any other
point Pd. Each cubic C E L cuts DI in PI and 2 other points; this defines
an involution on D I , and similarly for D 2 These involutions will be denoted
R 1-+ R', resp. S 1-+ S'.
D.
Let now r be the (strict) graph of the correspondence on Dl X D2

defined by L: a pair (R, S) E DI X D2 belongs to r if there exists C E L
with C.DI = PI + R + R', C.D2 = P2 + S + Sf.
We have a diagram:
r
,/
,/
PI
and r is the fibre product of DI and D2 over L, hence a (ramified) covering
of L of type (2,2).
Since DI is a genus zero covering of degree 2 of L, it is ramified in
two points ar,/3I of L (which correspond to cubics in L tangent to Dt).
Similarly, D2 is ramified in two points a2, /32 of L.
159
There are now three possibilities:
1) {O'I,/h} and {0'2,,B2} are disjoint. In that case r has genus 1 (and is
non-singular).
2) {O'I,,Bl} and {0'2,J32} have exactly one point in common, say 0'1 = 0'2.
In that case r has genus 0 (and has one double point, above 0'1, which
corresponds to a cubic C E L tangent to both Dl and D 2 ).
3) {ab,Bd = {0'2,,B2}. In that case r decomposes into 2 curves of genus
o (over a suitable extension of the ground field).
The favourable case is case 2). Indeed, r can then be chosen to reparametrise the pencil L: to (R, S) E r we associate the corresponding
cubic C R,S. On the generic member of this family, we now have 10 points:
PI, ... , P s , R, S. If these can be proved to be independent (modulo the section D by a line, as in 11.2), Neron's theorem will give us elliptic curves
of rank ~ 10 over Q.
In resume, there are two conditions to be checked:
a) that case 2) occurs;
b) that PI, ... , P s , R, S are independent on the generic CR,S.
We now show how this can be done. The construction is essentially
that of Neron [N1] except for the use of generic data instead of rational
data.
Details of the construction.

We start with the generic independent points Ql, Q2, P3, ... , P s , X of
the plane P2. This gives a purely transcendental extension of Q of degree 18.
We denote by C l the cubic passing through these 9 points. We have already
seen that the field of definition Q(Cl ) of C1 is a purely transcendental
extension of Q of transcendence degree 9, and that QI, Q2, P3 , , P s , X are
generic independent points on C l with respect to Q(C1 ). We consider the
pencil L of cubics through Qb Q2, P3 , , P s .
We choose a line Dl (among the 4 possessing this property) passing
through Ql and tangent to C l at a point PI with PI =I- Ql. The field
Q(C1 ,Pd is a finite extension (of degree 4) of Q(CbQI)). We choose in
the same way a line D2 passing through Q2 and tangent to C I at P2 with
P2 =I- Q2. The transcendence degree over Q of the field Q(PI, ... , P s , X) is
then 18, therefore PI, ... , Ps , X are generic independent points of P 2, and
we can view C I as the unique cubic passing through PI, ... ,Ps,X.
160
Lemma. The curve r associated to D I , D2 and the pencil L of cubics

passing through Q17 Q2, Pa, "" Pg has type 2,
Proof,
We do not have case 1 since the cubic G l is tangent to both Dl and
D 2 We must check there is no other, to avoid case 3,
We assume that there is a cubic G' different from G l passing through
QI,Q2,PS ,.",Pg tangent to DI at P{ and to D2 at~,
Let K = Q(GI , Pb P2 ), The points Q1 and Q2 are rational over K,
therefore D1 and D2 also, The transcendence degree of K(G') over K is
at most 5, Indeed the point P{ belongs to D 1 , therefore depends on one
parameter; the same is true for P~ and the cubics passing through Q17 Q2,
tangent at Pl., P;' to Db D 2 , respectively, depend on 3 parameters, except
when P{ and P;' are equal to Dl n D2 in which case there are 4 parameters,
On the other hand Q( GI, PI, "" Pg ) = K (P3 , "" Pg ) has transcendence
degree 17 over Q, thus 6 over K. But P3 , .. "Pg are in the intersection of
G l and G', thus are algebraic over K(G'), whence a contradiction,
We have seen that P1> .. " P8 are generic points of P 2 , thus the field
Q(P1> ,." Pg ) is a purely transcendental extension of Q of degree 16, and
E = Q(P1 , ... , P8 , G I ) is a purely transcendental extension of Q of degree
17,
Let r be the curve in DI x D2 defined as above with respect to the

pencil L, Since we have type 2, r is irreducible and its genus is 0, Moreover
r contains the point (R, R), where R is the point of intersection of Dl and
D 2 ; this point is a simple point and is rational over E. Hence r is birational
to PI,
Let (R, S) be a generic point of r with respect to E. The field F =
E(R, S) is the function field of r; it is a purely transcendental extension of
Q of degree 18, Let C be the cubic passing through R, S, Q1> Q2, P3 , ... , P8 .
As C is defined over a purely transcendental extension of Q, it only remains
to prove the following lemma.
Lemma. The 10 points R, S, QI, Q2, Pa, .. " P8 are independent on C modulo a line section.
Consider a relation
mR + nS + qlQl + q2Q2
+ LPiPi
i=S
tv
(mod D),
161
with m,n,ql,q2,Pi E Z.
Let R' be the third point of intersection of Dl with C. The point
(R', S), the image of (R, S) by an involution, is a generic point of V. Therefore
s
mR' + nS + qlQ1 + q2Q2 + LPiPi rv 0 (mod D).
i=3
But R, R' ,Q1 are collinear:
R+R' + Q1
rv
0 (mod D).
Whence
s
2nS + (2ql - m)Ql + 2q2Q2 + 2 LPiPi
i=3
rv
0 (mod D).
But S, Q1, Q2, P3, ... , Ps are generic independent points of Q. [Indeed, the
field K = Q(Qt, Q2,P3, ... , Ps ) has transcendence degree 16 by assumption,
and one easily checks that tr.degKK(D2 ) = 1 and tr.degK(D2)K(D2 ,S) =
1.] This implies that S, Q1, Q2, P3, ... , Ps are independent (mod D), hence:
= 0, m = 2q1, q2 = 0,P3 = ... = Ps = O.
A similar argument, with R and S permuted, gives

m
= 0, n = 2q2, q1 = 0,P3 = ... = Ps = 0,
showing that all the coefficients are O.
11.4. Elliptic curves of rank 2:: 11 over Q.

Neron [N2] has sketched a method for constructing elliptic curves of
rank ~ 11 over Q. One starts with a pencil L of elliptic curves through
points P b ... , Ps and lines Db D 2, D3 through PI, P2, P3. As before, the
correspondence between D 1, D 2, D3 defined by L gives a curve r c Dl x
D2 X D 3. If one has
a) DI,D2,D3 are concurrrent,
b) there is a cubic L tangent to D 1, D2 and D 3,
then r is (in general) of genus 1. If moreover r has infinitely many rational
points (Rt,R 2 ,R3 ), then every such point gives a cubic with 11 "obvious"
points (Rt, R 2 , R 3 , PI, ... , Ps ), which one can hope to be independent (in
162
general). Neron only gives hints about all this, but a detailed account can
be found in [Fr]j see also [To].
Questions.
1) Can one construct elliptic curves over Q of arbitcarily high rank?
Neron's methods give examples of rank ~ 11. Since then, 11 has been
successively improved to 12 (Mestre, 1982), 14 (Mestre [Mel], 1987), 15
(Mestre, 1992), 17 (Nagao, 1992), 18 (Tunnell, 1992), 19 (Fermigier, 1992)
and 21 (Nagao-Kouya, 1994).
2) Is there a bound for the rank of an elliptic curve over Q(T) with non
constant j invariant? Examples are known with rank 2:: 12 (Mestre [Me3],
1991), and with rank 2:: 13 (Nagao, 1994).
3) A different question is the Variation with the ground field of the
rank of a fixed elliptic curve. For instance, Honda has conjectured ([Ho],
misprinted on p. 98) that for an abelian variety A over a number field k
there is a constant C = C(A,k) such that
rankA(k') ::; C[k' : k],
for every finite extension k' of k. This is a rather doubtful conjecture;

it would imply that all the quadratic twists of a given elliptic curve have
bounded rank.
163
12. THE LARGE SIEVE
Let A be a thin set in zn, and AN the intersection of A with the ball
of diameter N centred at the origin. When n = 1 we have seen in 9.7 (as a
consequence of Siegel's theorem) that IANI = O{NI/2) when N ~ 00. To
prove a similar result when n 2: 2 one needs a different method, based on
the large sieve inequality (cf. [Co]).
The present chapter contains a proof of this inequality (after [Bom1]
and [Hun. Applications to thin sets are given in the next chapter.
12.1. Statement of the main theorem.
Let K be a number field, d = [K : Q], OK the ring of integers of K,
and A a torsion free OK-module of finite type with rank n over OK. We
select a norm 1.1 on AR = R A.
[A typical example is K = Q, OK = Z, A = zn, and for x =
(XI, ... , xn) E zn we take Ixi = sup IXil.]
We consider a subset X of A. We impose bounds on the the "size"
of X both from the archimedian and non-archimedian viewpoints. More
precisely, let N and Q be real numbers, N 2: 1, Q > 0; for every prime
ideal p of OK, let wp be a real number in [0,1]. We assume that:
1) The set X is contained in a ball of diameter N: there is Xo E AR such
that Ix - xol < N/2 for all x E X.
2) For every p with Np ~ Q, the image Xp of X in A/Ap by reduction
modulo p is such that
The assumption (2) means that a proportion
wp
of the classes modulo
p are missing from Xp; they have been "sieved out".
Theorem. There is a constant c = c(A,I.1) depending only on K, A and

the norm chosen on AR, such that for all N,Q,w p , and all X satisfying (1)
and (2), we have
where
L(Q) =
"11
L..J
1-w
wp
!! pi!!
164
the sum being over all ideals g of 0 K without square factors and of norm
5: Q.
Ixi
The constant c can be made explicit. For example, in the case A =

= sup lXii, one can take c = 2n.
zn,
Example. Take K = Q, n = 1, and X the set of squares in [1, N]. One

can then take wI'" close to 1/2, and L(Q) ::::: Q; choosing Q ::::: N I / 2 gives
IXI ~ N I / 2 , a not very surprising statement!
The name "large sieve" refers to the size of WI'" (i.e. the size of the
missing subset for X modulo p) e.g. wI'" a constant> 0; this is the relevant
case for thin sets. In the small sieve wI'" may be as small as liN pn.
The above theorem is an assertion in harmonic analysis; it expresses
properties of functions on a torus. The lattice A plays the role of the group
of characters, in the Pontrjagin sense, of a compact torus T = T A If >.. E A,
let X>. be the corresponding character on T A . Associated to a subset X of
A is the function
Ix
L: X>.
>'EX
where the Fourier coefficients are 1 in X and 0 elsewhere. Hypothesis (1)

means that the Fourier coefficients of I x are zero outside a ball of radius
N.
Consider now the reduction of A modulo p . By duality, the quotient
group AI pA is the dual group of T p, where Tp is the subgroup of T of
p-torsion points. Hypothesis (2) means that when the restriction of I to
the p-torsion points is expanded as a finite Fourier series, only a proportion
5: 1 - wI'" of the Fourier coefficients are non-zero.
We shall need two auxiliary results. The first is about functions on
finite abelian groups; the second is about the values of Fourier series on
sets of well spaced points on a torus.
12.2. A lemma on finite groups.
Lemma. Let G I , ... , G m be finite abelian groups written additively. For
15: i 5: m, let G i = Hom(G;) C*) be the dual ofGi , Oi a subset ofG i , and
Wi = IOiI/IGil. Let I be a complex valued function on G I x ... xG m Assume
that I is a linear combination of products XI " ,Xm, with Xi E G i - OJ. Then
165
We say that an element of the product G I X X G m is primitive if no
component is zero; we can then write the latter sum of the lemma in the
form
II(x)12.
zEG, z primitive
Proof.
By induction on m.
a) m
= 1, G = GI .
The Fourier expansion of 1 is:
By the orthogonality formulae, the L2 norm of 1 is
We have
x
and by the Cauchy-Schwarz inequality
11(0)1 2 ~
(L la xl L
2)
xEG-!h
~ IGI(1 - WI)
~ (1- WI)
laxl2
II(x)12.
zEG
Therefore
11(0)1 2 ~ (1- wI)II(0)1 2 + (1- wd
L
zEG,z~O
hence
WIII(OW ~ (1- Wl)
L
zEG,z~O
II(x)12.
II(x)12,
166
We finally divide by Wl to obtain the result (if Wl = 0, the lemma is trivial).
b) General case.
For Xm E Gm, we define Ix ..(xb,xm-l) = I(xl, ... ,xm). Then IX m
is a function on G l X ... X G m - l to which we can apply the induction
hypothesis:
m-l
Ilxm(xb ... ,xm-dI2~(II
xi:;CO,i=l, ... ,m-l
i=l
1:
iw)llxm(0)1 2.
t
Hence we have
m-l
I/(x)12~(II l:iw'>
xEG, x primitive
i=l
1/(0, ... ,0,xm)1 2.
xmEGm,xm:;cO
We then apply part (a) of the proof to the function 1(0, ... ,0, Xm) on Gm:
1/(0, ... , 0, Xm)12 ~ 1 ::m 1/(0)1 2,
x ... EGm,xm:;CO
whence the lemma.

This lemma shows that a function involving few characters does not
oscillate much. An extreme case is that of one character (the absolute value
is constant). At the other extreme, the Dirac b-function (which equals 1 at
the origin and elsewhere) involves all the characters.
12.3. The Davenport-Halberstam theorem.

For the archimedian setting, the OK-module structure of A plays no
role. We consider then a free Z-module A of rank m. Let A' = Hom(A, Z)
be the adjoint (or dual) of Ai AR = R A is a vector space over R of
dimension m, its dual is Ak = R A', and A' is a lattice in Ak. Let TA
be the quotient torus Akl A' i its dual is Ai if ..x E A we denote by X.>. the
corresponding character of TA:
x>.(X) = exp{27ri < ..\,X > },x ETA =
Ak/A'.
167
Theorem. (Davenport-Halberstam, see [Bom1]'[Hu],[Mo]). Let A be a free
abelian group of rank m, and let TA be the dual torus. Select norms on AR
and on Aft. There is a constant c, depending on the chosen norms, such
that if 6 > 0, and Xl, . , Xr E TA are 6-spaced in the sense that
and if 1 is a complex function on TA which is a linear combination of

characters X>. with A E A in a ball of diameter ~ N, with N 2:: 1, then
r
L I/(Xi)1
~ csup(N, 6- 1 )mll/ll~
i=l
Remarks. 1) The norm X

from the norm on Aft:
2) The L2-norm
is:
t-+
IIxil on TA is defined by passage to the quotient

IIxil =
inf
y ...... 31
IIyll
11/112 of 1 relative to the Haar measure dx with volume 1
12.4. Proof of the Davenport-Halberstam theorem.

It suffices to prove the theorem for a particular absolute value. We shall
do this for A = zm, T = Rm /zm and It I = sup Itil for t = (tl, ... , t m ) E Rm.
With this choice, we shall see that c = 2m suffices.
There are proofs of this theorem based on different principles. We
present here the original method, which is well suited to the case of several
variables. The methods which were subsequently invented give finer results
for m = 1.
a) The theorem is true in the case where 6 > 1/2.
Indeed, two distinct rational points Xi,Xj ofRm /zm satisfy !lxi-xjll ~
1/2. If 6 > 1/2, then r ~ 1; but by hypothesis 1 = E>.ex a>. X>. , where X
is contained in a box of diameter ~ N, therefore
I/(xdl 2 ~
(L la>.12)(L 1) ~ (N + l)mll/ll~ ~ (2N)mll/ll~.

>.ex
>.ex
168
b) We now assume that 6 ~ 1/2. Let 4> be a continuous complex-valued
function on Rm which is zero for It I 2: 6/2. Extending 4> by periodicity,
we can consider it as a function on the torus T; the coefficient of X>. in its
Fourier expansion is then
C>.(4))
= f
JRn> 4>(x)X->. dx = JTf 4>(x)X->.(x) dx.
The L2-norm is given by
We now assume:
(t) IC>.(4)) I 2: 1 for all A E X.
Lemma 1. We have E~=1If(xi)12 ~ 114>lgII/II~.
Proof. Consider the function
'"' c>.(J)
g= ~ -(A..)X>'.
>'EX
C>.
0/
Its Fourier coefficients are
c>.(g) =
CA(f)
{ c A (4))
Thus c>.(J) = c>.(4))c>.(g), therefore I =

product
I(x) =
if A E X
if A ~ X
4> * g,
where
* is the convolution
4>(x - t)g(t) dt.
Let Bi be the ball of centre Xi and radius 6/2. As 4>(x - t) = 0 for !Ix - til
8/2, we have
I(Xi)
therefore
= f
JB.
4>(Xi - t)g(t) dt,
>
169
But the Bi are pairwise disjoint, hence we have
r
L 1!{XiW ::; 11<p11~lIgll~

i=l
Finally, as
Ic>.(g)1 ::; Ic>, (f) I by (t), we have

IIgll~ =
>.ex
Ic>.{g)1 2 ::;
>.ex
IC>.(fW = II!II~,
whence the lemma.

We now construct such functions <p. First, we consider the case of one
variable.
Lemma 2. Let 1/2.
Then the Fourier transform ~ of <b satisfies

~(y) ~ 1 if
and
Iyl ::; 1/2,
+00
1I<b1l~ =
Proof.
We have
~(y)
-00
1/2
1<b(xW dx = 2.
2cos(7rx)e- 2 1l"ix y dx.
-1/2
As <b is even, ~ is real:
<b(y) =
A
1 cos 7ry
2 cos( 7rx) cos(27rxy) dx = - 1
2
-1/2
7r 4" - Y
11/2
This shows that ~ is decreasing in the interval [0,1/2]. Other hand, we

have
1/2
~(1/2) =
2( cos 7rX)2 dx = 1/211<b1l~ = 1.
-1/2
170
The lemma follows.
In the general case of m variables, we can then construct the function
> in the following way. Put M = sup(N, 8- 1). Let >'0 E R m be the centre
of a box of diameter ~ N containing X. For x = (x(1), ... , x(m E Rm, put
II ~(M xCi,
m
>(x(1), ... , x(m = MmX>'o (x)
i=1
where ~ is as above .
.) If Ixl > 8/2, we have > = o.
Indeed, Ixl > 8/2 implies that there is i such that Ix Ci ) I > 8/2, hence
Mlx(i)1 > M8/2 > 1/2; therefore ~(Mx(i = 0 and >(x) = o.
.) The L2-norm of > is:
.) The absolute value of the Fourier transform of > is ~ 1 in the box with
centre >'0 and diameter N.
As multiplication by X>'o just translates the Fourier transform, it suffices to see that the absolute value of the Fourier transform of the function
x ~ Mm n~1 (Mx(i is ~ 1 on the box with centre 0 and diameter N.
The value of this Fourier transform at the point (y(1) , ... , y(m is
II ~(y(i) /M).
m
i=1
then Iy(i) /MI ~ 1/2, therefore I n~1 ~(y(i) /M)I ~ l.

The Davenport-Halberstam theorem now follows from lemma 1 and
the formula
If
ly(i)1 ~ N/2,
Commentary on the constants.

We restrict ourselves to the case m = 1 (Fourier series in one variable)
which is the only case that has been studied in detail.
171
We consider two numbers N ~ 1 and f> $ 1, and we seek the best
possible constant ~ = ~(N, f for which
r
L I/(Xi)1
$ ~(N, fI1/I1~
i=l
given that the Xi are f>-spaced and I is a trigonometric polynomial of length

$N:
jo+N
=L
I(x)
Cje- 211'ij z.
j=jo
We have seen that:

~(N,f $ 2sup(N,f>-1).
For arbitrary N and f>, the exact value of ~ is not known, but one has good
bounds for it.
Theorem. (Selberg, see [MoD. We have ~ $ N
+ f>-1.
This can be proved by the preceding method, if Nf> $ 1, but the

interesting case for some applications is Nf> > 1, for which we refer to [Mo].
This Selberg bound is optimal for some values of Nand f>. For example,
take XI, ,Xr to be the M-torsion points on T = R/Z, with r = M. We
choose I(x) = O(Mx), where
Le
L
O(x) =
2 11'inz,
for some L ~ l.
n=O
We may take
f> = 1/M,N = ML.

We have
f(Xj) = 0(0) = L + 1
r
L If(xj)12 = (L + 1)2M
j=l
and
1I/11~ = L
+ 1,
172
therefore
L 1/(Xj)12 = (N + 8-1),,/,,~
r
j=l
in this particular case. Note that N8 = L ~ 1.

Bombieri and Davenport obtained a bound which is better when N 8 is
small:
They also constructed examples with N 8 arbitrarily small and
for some explicit value of c (they give c = 1/12, but this can be improved
to c = 5/8).
Therefore f:1(N,8) is known reasonably well when N8 is small. For
N8 ~ 1, Selberg's bound remains the most useful, see [Boml], [Mo].
Exercise. Show that
for any integer M
1.
12.5. End of the proof of the main theorem.

First, we show that the torsion points on a torus are well spaced. We go
back to the notation of 12.1 and 12.3: K is a number field, [K : Q] = d,
A is an OK-module of rank n, A ~ zm with m = nd. The dual torus
T = Ak/ A' of A also has an OK-module structure. If q is a non-zero ideal
of OK, a point x E T is called a q-division point if
= 0; such a point
has finite order.
-
q.x
Lemma. There is a constant Cl (depending on K and the norm on AkJ

such that the q-division points on T by the ideals q with N q :S Q are
a-spaced with 0-= CIQ-2/d.
- The OK-module A' is known to be isomorphic to a direct sum of ideals
of OK:
A' = l!.1 ffi ... ffi l!.n
Then
Q A' = K ffi ... E9 K,
173
and
Aa = KR EB .,. EB KR,
where K R = R K is the product of the archimedian completions of K.
We define an absolute value on KR by putting
Ixl =
sup 100xi
a
for x E K,
where 0' runs over the embeddings K -+ C, and one extends this to KR by
continuity.
t:'
ror x -- ( X (1) , , x (n E A'R, we take
Ixl =
sup 100(x(iI.
O',i
For this particular norm, we prove the lemma with

= i~f N(fu)l/d.
Cl
We take a q-division point t and a q'-division point t' with t =1= t', Nq ~ Q
and N q' ~ Q. Then x = t - t' is a q7l-division point and N qq' ~ Q2. Write
x = (x(l), .. ,x(n where x(i) E K.The condition
-
qq' x
c A' =
Q1 EB ... EB Qn
means that
qq'X{i) C Qi'
Let i be such that x(i)
=1=
(1 ~ i ~ n).
O. Then
N(qq'x(i ~ N(Qi);
therefore
a
and
whence the lemma.
We return to the hypotheses of the large sieve (12.1): given
A, T, Q, N, {w",} the set X is contained in a ball of diameter N ~ 1, and its
image X", in A/~A satisfies
IX",I ~
(1 - w",)[A : ~A],
~ Q.
for all prime ideals ~ with N ~

Take a function f = EA CAXA on the torus T whose spectrum {A E
A, CA =1= O} equals X.
174
Theorem. There is a constant c = c(A, 1.1) such that for all x E T, we

have
where
L(Q) =
'"
~
II 1-w'
wp
Nq:5Q
pI!
! squ;re-free
It suffices to show this for x = O. Indeed, the translation x t--+ x + xo

has the effect of multiplying c>.x>. by the constant X>.(xo), therefore it does
not change the spectrum.
Let TQ be the set of q-division points of T for all ideals q of OK with
N9: ~ Q. We write T[qJ =-{t,~ = OJ. Thus
TQ =
T[9:]'
N!:5Q
From the above lemma, the set TQ is <5-spaced, with <5 = cdQ 2 / d The
Davenport-Halberstam theorem (12.3) then gives
If(x)12 ~ c2sup(Nnd,<5-nd)lIfll~
zETq
C3 = C2 sup(l, c1nd ).
We say that a q-division point x is primitive if the conditions q'lq and
q' =1= q imply q' x =1= 0: Each point x E T Q is primitive for a unique q,-namely
its allnihilator {a E OK, ax = OJ.
We restrict ourselves to ideals q without square factors: 9: = Pl ... Ph,
where the Pi are pairwise distinct prime ideals. In this case
with
and a point x of T[qJ is primitive if and only if each of its components in

the T[Pi] is different from O.
We have
zETq
N!:5Q
zET[q)
squa.re-free z primitive
175
We shall now apply the lemma on finite groups of 12.2 to show that
I/(x)12?:: 1/(0)12
primitive xET[11
II 1 : :
1"11
I"
First note that the dual of T[ID is AI'1A: the injection

T~] <-+ T
is dual to the projection

A
-t
AI'1A.
As the spectrum of 1 on T is X, it follows that the spectrum of the restriction of 1 to T[ID is contained in the image Xq of X under the homomorphism
A - t AI'1A = TI plq AI pA. Taking account O"f the given upper bounds of the
IXpl, we may then apply 12.2 and obtain the above inequality. By summation over '1 this gives
I/(x)12?:: 1/(0)12
II 1 : :
N1$Q
1"11
1 square-free
xETq
I"
?:: 1/(0)1 2L(Q),

which proves the theorem above.
To obtain the large sieve theorem stated in 12.1, we choose
E>.EX X>.; we have
1(0) = IXI = II/II~,
1-
hence the theorem above gives
dividing by
IXI gives the result (if IXI =
0 the theorem is obviously true).
Example. If K = Q, A = zn, d = 1, with the usual norm sup lXii, the proof
above of the large sieve gives the constant c = 2n :
176
Remark.
We can give a lower bound for L(Q) in which the sum runs over all
ideals (and not only over the square-free ideals ~ of bounded norm); for this
we define
W! =
w~(p) for ~ =
pn(p)
II
II
pI!
and
L'(Q) =
wi
N!~Q
We have
L(Q) 2 L'(Q).
Indeed, the radical ../9,. of ~ is
L'(Q) =
li pl!
p; as
I:
I:
N!~Q
Nq'~Q
! square-free
it suffices to check
-A.' =!
w!',
177
13. APPLICATION OF THE LARGE SIEVE TO THIN SETS

13.1. Statement of results.
1) Affine thin sets.
Let K be a number field of degree d, and OK the ring of integers of K.
For x E OK, we put
!xl = max l(1xl
u
where (1 runs over the embeddings of Kin C.

Let n be an integer ~ 1. Let x = (x(1), ... , x(n) E OK' We put
Finally let X be an integer
1. If M is a subset of OK we define
M(X) = I{x E M;1x1 ~ X}I.

For M = OK this number is xnd, when X - t 00.
Assume now that M is thin in An(I<) n OK cf. 9.1.
IV
Theorem 1. (S. Cohen). There is 1< 1 such that
M(X)
= 0(x(n-l/2)d(logXp).
Cohen only proved this result with 'Y = 1, but made precise the 0 as
a function of the coefficients of equations defining M [CoJ. The proof gives
a uniform bound for the number of points of M in every box of diameter
X. Note that we gain a factor of X-d/2(logX)"l' over the trivial bound
o(xnd): the set of x = (Xl, ... , xn) of OK with Xl a square shows that this
improvement is essentially optimum.
2) Number of integral points on affine varieties.
Let N be an integer ~ 1, and V an irreducible algebraic variety of
dimension n in affine space AN over K. Let VeX) be the number of points
X E V with xCi) E OK, and !xl ~ X. We suppose that V is not linear, i.e.
that deg V ~ 2.
178
Theorem 2. VeX) = o(x<n-l/2)d(log xyr) with 'Y
< 1.
This result is derived from theorem 1 as follows. Select coordinates of

AN such that the projection onto the first n coordinates induces a finite
morphism 7r : V - t An. The image 7r(V) is then a thin set to which we can
apply theorem 1. Each fibre of 7r has a finite bounded number of points,
whence theorem 2.
The same reasoning gives the bound o(xnd) when deg V = 1.
3) Thin projective sets.

On projective space P n over the field K, let H be the standard height
associated to an hyperplane section (normalised so that H is invariant under
finite extension d. 2.2). Consider a thin subset M C Pn(K), and denote
by MK(X) the number of x E M with H(x) ~ X.
Theorem 3. We have MK(X)
= o(x(n+1/2)d(logXyr) with'Y < 1.
By Schanuel's theorem d. 2.5, the number of x E Pn(K) with H(x) ~

X is ,..., cx(n+1)d. We therefore gain a factor of X- d/ 2(log xyr on the trivial
bound.
It would be interesting to know if this bound is the best possible. It
seems more reasonable to expect a bound of the form o (xnd(log X)C).
The proof that theorem 1 implies theorem 3 will be given in 13.4.
4)Number of rational points on projective varieties.

Let V be an irreducible projective variety over K of dimension n, in
projective space P N. We assume that V is not a linear variety. Define
VK(X) = number of x E V(K) with H(x) ~ X.
Theorem 4. There is 'Y
< 1 such that
Question. Can this bound be improved to o(xnd(log X)C) for some c? The
factor (log X)C is inevitable as the quadric xy = zt in P 3 shows.
Exercise. What about quadrics in n-space, n
31
Theorem 4 follows from theorem 3 by projecting just as theorem 2

follows from theorem 1.
179
Properties of thin sets.

While proving theorem 1, we shall obtain an interesting property of
the reduction modulo P of a thin set.
Let P be a prime ideal of OK. The residue field kp = OK / P is finite,
and the map of reduction modulo p is everywhere defined on projective
space:
[This is clear from the point of view of schemes. If OK is principal, take
coordinates (Xl, . ,xn ) of x E Pn(K) with Xi E OK and Xl, ... ,x n coprime,
and reduce them modulo p. In the general case where OK is not principal,
one uses the fact that the local ring at p is principal.]
Hence, a subset Mp ofPn(kp) is associated to any given subset M ofPn(K).
If S = {Pl, ... , Ph} is a finite set of prime ideals, let us denote by Ms the
image of Min
Pn(k p1 ) x ... x Pn(kp,,) = Pn,s.
Theorem. If M is thin, there is a finite set of prime ideals S such that
Ms =1= Pn,s.
Indeed, it is enough to prove this when M is of type 1 or type 2, in the

sense of 9.1. For type 1, the assertion is immediate: we have Mp =1= P n(k p )
for almost all p's. For type 2, it follows from the corollary to theorem 5
below (13.2).
Remarks. 1) When K = Q, n = 1, the theorem implies that the complement

of a thin set contains an arithmetic progression, cf. [Schi].
2) The set S cannot always be taken to have one element, see Exercise.
Exercise. Consider the following thin subsets of PI(K):
MI = {x E P1(K)1
M2 = {x E P1(K)1
M3 = {x E P1(K)1
x is a square}
x + 1 is a square}
x(x + 1) is a square}.
Let M = MI U M2 U M3. Show that, for every prime ideal p, the map
M -+ PI(kp) is surjective.
Let M C OK be a thin set. An easy redution allows us to assume that
M = OK n 1r(Z(K)), where Z is an irreducible affine variety of dimension
180
n, and 7r is a finite morphism of degree 2: 2, 7r : Z --+ An. Furthermore,
one can replace this by a finite morphism of schemes of finite type over OK
such that
We then say that such a thin set M is primitive.
Let p be a prime ideal of OJ(. Let Mp be the image of M under
reduction modulo p in kp x ... x kp = An(kp).
Theorem 5. There is a finite Galois extension K7r:/ K and a constant
with 0 < C7r < 1 such that if p splits completely in K 7r then
Corollary. If Np is large and p splits completely in K 7r , then Mp

An(kp).
C 7r
oF
The proof will be given in 13.3. We now prove theorem 1, using

theorem 5.
We apply the large sieve inequality (12.1) to M:
where
L(Q)
II 1- w
wp
=
N!!SQ
pi!!
!! square-free
and
l-w
= IMpl.
Npn
The constant CK,n depends only on K and n.

We select Q = X d / 2 Therefore
and it only remains to show that
L(Q) A> Q/(logQ)\

We begin with an easier result.
'Y
< 1.
'
181
Lemma 1. We have L(Q)
QjlogQ.
To see this, we only keep the prime ideals in the sum defining L(Q):
This is bounded below by the prime ideals which split (completely) in Kp;
for these by theorem 5 we have wp ~ , with > 0, for Np sufficiently large;
hence:
1.
L(Q) ~
no~Np~Q
p splits in [("
By the prime number theorem, we have
1rv
Np~Q
-.!L
[K7r : K] logQ'
P splits in [("
whence lemma 1.
To improve this estimate and get an exponent 'Y
L(Q) ~
< 1, we write
c!!.'
N.!!~Q
where
if g is squarefree and all pig split completely in K 7r ,
otherwise.
We introduce the function
This is a Dirichlet series, and if we write
then
182
As
Q ~ c!!.
is a multiplicative function, we have
I(s) =
II(1 + cI'>Np-S)
I'>
where
cl'> = { 1-0W p
if p splits completely in K 11',

otherwise.
If p splits completely in K 1'>' we have from theorem 5 that
cl'>
= NIMpn 1-1 = ~ -1 + O(Np-1/2).

I'>
C1I"
Write
c= -
Then there are constants
c~
-1> O.
> 0 such that
1/ 2,
cI'> >
- c'I'> and Ic - c'I'> I ~
.,., NVlIf'
for all p which split completely in K1I"' We introduce the functions
II(s) =
II
I'> splits in
and
h(s) =
II
I'> splits in
for Re s
K"
K"
> 1. By taking logarithms, one finds that

hCs) = 12(s)g(s)
for g( s) = exp( h( s, where h is a Dirichlet series which converges for

Re s > 1/2.
One shows easily that
h(s) = (K" (S,gl(S), with>' = c/[K7r : K],
183
where gl(8) = exp(h 1(8), hI converges for Re 8 > 1/2 and (K,.. (8) is the
zeta function of the number field K 11'. Hence
has no zeros or poles except at

and
JI (8) '"" c' (8 ~ 1V
= 1 in an open set containing Re 8 2:
1,
for 8 in a neighbourhood of 1.
If we had A = 1, Ikehara's Tauberian theorem would give for the coefficients

al,n of JI:
For arbitrary A, an analogous result proved by Delange [De] and Raikov

[Ra] gives
Q
D al,n '"" c /I (10 Q)I-)..'
'"
n<Q
with
/I
C'
= rCA)"
[This method can be refined using a method of Landau (d. [Se4])

which gives an asymptotic expansion for l:n;:;Q al,n.]
This proves theorem 1 with 'Y = 1 - A.
Let 7r : Z - t An be as in 13.2. Let K(Z) be the function field of Z,
a finite extension of degree r > 1 of K(TlJ ... , Tn), and 0 the Galois closure
of K(Z) over K(TI, ... , Tn).
Let K 11' be the algebraic closure of K in OJ this the largest subfield of
o which is algebraic over K.
[A typical example is n = 1, Z = AlJ and 7r(z) = z3.
o
Z
K(O)
111'
T
Al K(T)
184
The corresponding field extension K(Z)/K(T) is K(O)/K(T) with 03 = T;

hence 0 is the compositum of K(O) and K1r where K1r is the, at most
quadratic, extension of K obtained by adjoining the cube roots of unity.]
Let Zn be the affine variety which is the normalisation of Z in O. Then
there is a sufficiently small non-empty open subscheme U of An over which
the corresponding open subsets Uz c Z, Un C Zn are finite and etale. Let
G = Gal(O/K(T and H = Gal(O/K(Z.
H(
Un
!
Uz
U
The field of constants of Un is K 1r ; in other words, Un is absolutely
irreducible over K 1r .
Lemma. (Lang-Weil). If B is an OK-scheme of finite type whose generic

fibre is an absolutely irreducible variety of dimension n over K, then
[Here Bp means B(Kp ).]

One possible proof is to fibre the variety by curves and use the Weil
estimates on each fibre plus induction on the dimension. This is essentially
the method of Lang and Weil, see [LW].
The lemma can also be proved using the formula (Grothendieck Deligne):
where the Ai(P) are the eigenvalues of the Frobenius acting on l-adic cohomology with compact support. Since the reduction of the variety B modulo
P is absolutely irreducible, for almost all p, we have Ai(p) = N pn for
exactly one i, and for the others IAi(P)1 ~ (Np)n-l/2. The number of
eigenvalues appearing in the formula is uniformly bounded (for an explicit
bound in terms of the degrees of defining equations, see [Bom2]).
It remains to show that if p splits completely in K 1r , the image of
U(Kp) has ~ cNpn + O(Npn-1/2) number of
UZ,p = Uz(Kp) in Up
185
elements, with c < 1. We shall even obtain c ~ 1-1/r!, where r = [Uz : U],
as above.
We change nothing by replacing K by K 1r . We may therefore suppose
that K = K 1r ; then Un is absolutely irreducible over K, and the lemma
above gives
H(
Un,p
!
UZ,p
Up
As Un and Uz are etale over U, the group G acts freely on Un,p' We
decompose UZ,p as the image of Un,p and its complement. We count the
number of elements of UZ,p up to O(Ngall-1/2). The image of Un,p in UZ,p
is represented by the H -orbits and hence has cardinality
r1li N gall, and
then the complement has cardinality (1 - r1li)N gall. Therefore the image
IblNgall, and hence the image of UZ,p in
of Un,p in Up has cardinality
Up has cardinality at most
r-.I
r-.I
r-.I
In conclusion, the image of UZ,p in Up has cardinality
with
IGI = riHI.
As r ~ 2 and
r - 1
IGI ~ r!, we have
>
rlHI -
_1_
> ~ > ..!:..
rlHI - IGI -
r!'
This proves theorem 5 and hence theorem 1.
Example. Take the equation 03 = T, with K = Q, K1r = Q( {II), n = 1,

r=3;
if p == 1 mod 3, that is to say p splits completely in K 1r , the image mod
p is formed of the cubes in F P' therefore has (p - 1)/3 + 1 elements;
186
if p == -1 mod 3, all the elements of F p are cubes, and the image mod
p has p elements (such p's are of no use for the sieve).
Exercise. Apply the sieve method of 13.2 to the example above.

(One gets the - disappointing - result that in Z the number of positive
cubes ~ N is O(Nl/2).)
13.4. Proof of theorem 3 from theorem 1.
We shall use an auxiliary result on the representations of points of
Pn(K).
Let K be a number field, and n an integer
1.
Proposition. There is a constant c = c( K, n) such that every point x E

P n (K) is representable by coordinates (xo, ... , x,..) with x j E OK and
IXj I~ cH{x),
for all j.
The statement is trivial over Q with c = 1: we select the Xi E Z

coprime; then
H(x) = sup IXil.
In the general case, note first that if (xo, ... , xn) are coordinates of x E
Pn(K) with Xi E OK, the divisor class of the ideal generated by these
coordinates ~ = (Xo, ... ,xn ) depends only on x (and not on the chosen coordinates). We then take integral ideals f!1' ... ,!lIt representing the different
classes; by applying a homothety, we may assume that xo, ... , Xn generate
one of the ideals f!i' (I ~ i ~ h). We may then write (see 2.2(b
H(x) =
=
with d = [K : Q],
Ci
Ci
Ci
II .sup
v arch J=O, .. ,n
II .
SUp
v arch J=O, ... ,n
IIxjllv
IXjl~,,/d,
= l/N(f!i)l/d, and
N _{I if
v 2 if
Therefore
H(x)
= Ci
V
V
isreal
is complex
II
v arch
hv(x),
187
Lemma. There is a constant CK > 1, depending only on K, such that
for every family hv (v runs over the set E of arcmmedean places of K) of
numbers> 0, there is a a unit of K and a real number t > 0 with
We first prove this lemma. We want

log 11lIv = loghv -logt + 0(1),
that is to say
log t + log 1Ic:llv = log hv + 0(1).
This follows from the compactness of the quotient of R ~ by the subgroup

generated by the logarithms of the units and the diagonal. One can make
explicit cKas a function of the degree and the regulator of K (cf. [Gy]).
We now use this lemma with hv = hv(x): if is the unit obtained in
this way, we multiply Xi by -l j if h is one of the hv(x) we have
H(x) ,...., h<'ENv)/d,...., h,

as
L:Nv = d.
This proves the proposition, since I Xj I:::; sUPv hv(x) ,...., h.
To prove theorem 3, we consider a thin set M C P n(K), and we define

M' = {x E OJ/ljX
- 0, the image of x
in Pn(K) belongs to M}.
Then M' is thin in A n + 1 The proposition above shows that
and theorem 1 gives

M'(cX)
whence theorem 3.
= 0(x(n+l-l/2)d(logXP),
188
APPENDIX: THE CLASS NUMBER 1 PROBLEM AND

INTEGRAL POINTS ON MODULAR CURVES.
[This appendix gives a brief account of the relation between imaginary quadratic fields
of class number 1 and integral points on certain modular curves. Many details have been
left out, especially on the modular side. To do otherwise would have taken us too far afield
and, besides, more complete results are now known, thanks to the work of Goldfeld [Go1)
and Gross-Zagier [GZ).)
A.1. Historical remarks.
There are nine imaginary quadratic fields Q( v'd), d
to 1:
< 0,
having class number equal
-d = 3,4,7,8,11,19,43,67,163.
Denote by d the discriminant of such a quadratic field Kd, d < 0, and h( d) the class
number of Kd, that is to say the number of elements of the group Cl(Rd) = Pic(Rd), where
Rd is the ring of integers of K d.
Tables show that h( d) tends to infinity as d -+ -00. Gauss posed the problem of
proving this (and of proving that only the nine above have h = 1).
In 1913, Gronwall showed that if Ld (the L-function attached to the primitive character
of (Z/dZ)* of order 2) has no zero between 1 and 1 - a/log Idl, (a is a given positive
constant), then
In particular, if GRH is true for Ld, then
h(d) ~ IdlI/2 flog Idl,

(the (log log Idl)1/2 term was eliminated by Landau in 1918). For the reverse inequality, it
is not hard to show that (see Exercise below)
h(d) :::;
7r- I
ldI 1 / 2 log Idl
if d
< -4.
In 1934-35, Mordell and Deuring showed that ifthe Riemann Hypothesis (for the usual
zeta function) is false, then h( d) -+ 00 with Idl [The idea is that if h( d) is small then Ld
has a zero near 1; but the existence of a zero for one Ld function entails the non-existence
of zeros for the others; hence if the Ld have such zeros then ( has none.)
Later, Heilbronn showed the same result without assuming the Riemann Hypothesis.
In 1935, Siegel proved
h(d) ~ Idl t -" for all

therefore
logh(d),....
> 0,
2log Idl
His proof is ineffective (it uses two values of d not satisfying the desired inequality).
Heilbronn and Linfoot refined this and showed that there is at most one value of d
with Idl > 163 and h( d) = 1. Landau showed more: given h, there is an explicit constant
189
c(h) such that h(d) $ h entails Idl $ c(h) except for at most one value of d. One can
choose c( h) = 2100h2 (log 13h)2. In his thesis in 1966, Stark proved that if there is a tenth
imaginary quadratic field having class number 1, its discriminant d lO satisfies
FUrther progress came from modular functions. In 1952, Heegner [He] published
a paper proving h(d) = 1 => -d E {3,4,7,8,1l,19,43,67,163} and hence solving the
problem of the "tenth imaginary quadratic field". But his proof was not detailed enough
to be entirely convincing (and indeed was not accepted at the time). In 1967, Stark gave a
complete proof, apparently different (there are no explicit modular functions) but in fact
substantially equivalent.
Also in 1967, Baker, by means of his explicit lower bounds for linear forms in logarithms, showed that h(d) = 1 => Idl < 10500 (Indeed, Gel 'fond and Linnik had proved in
1948 that, if h(d) 1, then a certain expression of the form I/h log al + fh log a2 +fJ310g a31
is very small cf. [Ba] and A.7 below.)
The problem of h(d) = 2 was resolved between 1972 and 1975 by Baker, Stark,
Montgomery-Weinberger. By a variant of the method of Gel'fond and Linnik, they obtained an effective but very large bound for Idl : Idl < 101100 , and to descend to Idl $ 427
required more hard work.
In 1976, D. Goldfeld [Dol] showed that, if a certain L-function has a zero of order
~ 3 at s
1 (as it should, by the Birch and Swinnerton-Dyer conjecture), then h(d) -+ 00
with Idl, in an effective way.
[The desired property of this L-function has now been proved by Gross-Zagier [GZ],
thus settling the whole problem. For instance,
h(d) = 3,d < 0 => log Idl $ 21000

h(d) = 4,d < 0 => logldl $ 33600
h(d) = 5,d < 0 => log Idl $ 35000,
see [G02], [Oe2].]
Exerci.se. Let Xd be the quadratic character mod Idl associated with the imaginary
quadratic field Kd. One has
Xd(X)
IT Idl
= -Xd(Y)
> 4, it is known ([BSh], V)

h(d) =
if x == -Y (mod Idl).
that
7r- 1 IdI 1 / 2
L(I,Xd), where L(l,Xd) =
f: Xd~n).
n=l
(i) IT j is an integer with 1 $ j < Idl/2, put:

Ij
Show that
"
n=j
Iljl $
(modldl).n~l
Xd(n)
. 1
- n - = Xd(J)(-:- J
Idl- . + Idl + . - 21dl- . + ...).

J
Iii and that L(l,Xd) = El~j<ldI/21j.
190
(ii) Show that 2:1:S i <D/2 Ifj ::; log D for every integer D 2: l.
(iii) Deduce from (i) and (ii) that h(d) ::; 7r- 1 IdI 1 / 2 Iog 14
A.2. Equivalent conditions for h( -p) == l.

The ring of integers Rd of Kd
where
Wd ==
== Q( Vd) has {l,Wd} for a basis (if d is the discriminant)
Vd/2
{ (1 +
Vd)/2
if d
if d
== 0 (mod 4)
== 1 (mod 4)
The elements of order 2 of Pic(Rd) are easily determined. They exist if and only if dis
divisible by two distinct prime factors. Thus h( d) is odd if and only if d == -4, or d == -8,
or d == -p with p prime and p == 3 (mod 4).
The fields Q(i), Q( H) and Q( A) have class numbers equal to 1. We assume
from now on that p is a prime number> 3 with p == 3 (mod 4).
Proposition. The following properties are equivalent:
(i) h( -p) == 1,
(ii) (~)
== -1 for all primes [ such that
[<
p/4.
VPf3.
(ii)' (~) = -1 for all primes I such that 1<

(iii) (Assumingp> 7) p == 3 (mod 8) and Rp-Np = 3, whereRp (resp. N p) is the number
of quadratic residues (resp. non-residues) in the interval [I, (p - 1)/2].
(iv) The integer x 2 + x + ~ is prime for x = 0, I, ... , (p - 7)/4.
(iv)' The same assertion with 0 ::; x ::;
1).
t( VPf3 -
Example3. Let us use (iv)' to find those p, 3 < p < 75, such that h( -p) == 1. Let
m == (p + 1)/4. For p < 75, we have
1) < 2, and condition (ivy becomes that
m and m + 2 are prime. As m < 18, this gives m == 3,5,11,17 whence p == 11,19,43,67.
To prove that h(-163) == I, it suffices to check that x 2+x+41 is prime for x == 0,1,2,3
(which gives 41,43,47, and 53). From (iv), this property remains true for x == 4, ... ,39 (as
noticed by Euler, Mem. Acad. Sci. Berlin, 1772).
We can also use (iii) to check that h(.,.-l1) == 1: in the interval [l,5J, the quadratic
residues are 1,3,4,5 and the only quadratic non-residue is 2; thus Rll == 4, Nll == 1 and
Rll - N n == 3.
t( VPf3) -
Proof of the proposition.

(i)
=}
(ii). For d
==
-p, we have
(1;)
-1 if and only if I is inert in the extension
Kd/Q. Consequently, if (~) == 1, then the ideal (I) decomposes in Kd : (I) == p.p', with
1 = Np == Np'. If further h( -p) = I, there is a E Rd with P == (a), and Na == 1. We
write a = x + Y A with x,Y E
have y i- 0 and I 2: py2 2: p/4.
(ii)
=}
(ii)' is clear.
tZ.
Thus N(a) = x 2
+ py2 == I.
As I is not a square we
VPf3
is generated by an integer (of Z), thus is

(ii)' =} (i). Every ideal Q of Rd of norm <
principal [it suffices to see this for Q == p prime; by (ii)', we have N(p) == [2, thus p == (l)].
But every ideal class can be shown to contain an integral ideal of norm::; Jldl/3. Therefore
h(d) == l.
(iii) -} (i). If p > 7, h( -p) = 1 implies p == 3 (mod 8), see below. Conversely, if p == 3
(mod 8), and p > 3, Jacobi showed that Rp - Np = 3h( -p). (The proof is analytical;
191
one computes the value of the L series E(~)n- at s = 1, see e.g. [BSh], V.4). Hence
h(-p) = 1 {:} Rp - Np 3.
(ii) :::} (iv). We write m = (p + 1)/4, and Jp(X) = X2 + X + m. For 0 :5 x :5 m - 2 we
have 2:5 Jp(X) < m 2 (if p > 3). If, for such an integer x, Jp(x) were not prime, it would
be divisible by a prime number 1 < m. Thus, x 2 + x + m == 0 (mod 1), and the discriminant
1 - 4m
-p would be a square modulo 1; therefore (T) = 1, whence (~) = 1, which
contradicts (ii).
(iv) :::} (iv), is obvious.
(ivy:::} (iiy. If (ii)' were false, there would be a prime number 1<
with (~) 1. We
have I =1= 2 (otherwise p == 7 (mod 8) and the values of Jp(X) are even, which contradicts
(iv)'). Therefore (T) = 1, and the congruence x 2 + x + m == 0 (mod I) has two solutions
whose sum is 1- 1 mod I. We can then choose a solution x with 0 :5 x :5 (I - 1)/2.
Then Jp(x) is divisible by I and Jp(x) ~ m = ~ > I; therefore, Jp(x) is not prime. As
x :5
1), this contradicts (iv)'.
The equivalence between the properties (i) to (iv)' is therefore proved.
-IP13
t< -IP13 -
Suppose that p == 3 (mod 4) and h( -p) = 1. From condition (ii) we deduce that
for p > 7, (;) = -1, that is to say p == 3 (mod 8),
for p > 11, (=!) (~) -1, that is to say p == 1 (mod 3),
for p > 19, (=f) = (~) = -1, that is to say p == 2 or 3 (mod 5).
A.3. Orders of Rd.

Recall that Rd is the ring of integers of Kd = Q(Vd), where d is the discriminant. An
order of Rd is a subring of R 1, Rd,/ is not a normal (i.e. integrally closed) ring (its normalisation is Rd). The
number of elements of Pic( Rd,!) is denoted by h( dj2).
We are interested in the (d,!) with J ~ 2 for which h(dj2) 1. For this we use the
localization exact sequence
For d =1= -3, -4 one has
Rd =
{1} and this makes the computation easy.
Examples.
1) d = -7 and J = 2. In Q(v'-'i), we have (2) = pp, and (Rd/JRd) = F; x F; = (1),
therefore IPk(Rd,!)1 = IPic(Rd)1 = 1.
2) d = -11, J = 2. Then (2) is inert in Q(V-ll), F4: is cyclic of order 3, therefore
h(-1l.22 ) is equal to 3.
Similar arguments give the following list:

for d < 0 and J 2! 2, we ha.ve h(dJ2)
d - -3
= 1 {:} { d ;: -4
d = -7
and
and
and
J = 2 or 3
J= 2
J = 2.
192
A.4. Elliptic curves with complex multiplication.

Let E be an elliptic curve over a field K of characteristic zero, whose ring of endomorphisms End(E) over the algebraic closure of K is not equal to Z. In this case, End(E)
has rank 2 over Z, and is an order Rd,! of an imaginary quadratic field K d.
For a given order Rd,J, there are h = h(dP) curves E up to Q-isomorphism with
End(E) = Rd,J. (One can see this over C by writing E as C/U, w~ere U is a projective
Rd,! module of rank 1.) Thus if h = I, there is a single ~urve up to Q-isomorphism; hence
we have j(E) E Q (indeed, if q is an automorphism of Q, q(E) is an elliptic curve having
complex multiplication by Rd,! therefore j(q(E = j(E. Thus we can define E over Q,
but the complex multiplication is defined only after extension to Kd.
Table oj the 13 values oj j corresponding to h( dP) = 1, d < 0, J ~ 1, i.e. to elliptic curves

over Q with complex multiplication.
Discriminant
and conductor
d= -3,1 = 1
J=2
J=3
d= -4,J = 1
J=2
d=-7,j=1
J=2
d= -8
d=-l1
d= -19
d=-43
d= -67
d = -163
,=P/3
j _ 26 3 3
0
24 3353
_2 15 3.53
2 6 33
233311 3
_3 353
3 35 3173
26 5 3
_2 15
_2 15 33
_2 18 3353
_2153353113
_ 218 3 35323 3 293
_2 6 33
24 33112
-3.2611223 2
0
2336 72
-7.3 6
387.19 2
2772
-11.2 6 72
-19.3626
-43.2 6 3872
-67.2636 72312
-163.2636721121921272
223
2.3.11
-3.5
3.5.17
225
_2 5
-253
-26 3.5
-253.5.11
-263.5.23.29
The most interesting case is that or" Q( r-PJ with p > 11. We take E s:;; C/ Rd where
+ FP)/2, d = -po Then we have
Rd is the lattice with basis {l,wd}, where Wd = (1
and
j =
~ + 744 + 196884q + ...

q
= _e 1fvP + 744 + 8e- 1fvP,

with 105 ~ 8 ~ 3.105
We shall see below that j is a rational integer (and even a cube, d. A.6). Thus e1rvP
is close to an integer. For example (Hermite)
e"'v'!63 = 262537412640768743.99999999999925007...
If one could bound the p for which
193
(where 11.11 is the distance from the nearest integer) one would obtain directly the finiteness
of the p such that h( -p) = 1. The preceding inequality gives
106
Ilognp -1I"ylP15 e2 '1f"/p' where np
but
= 1 + [e'lf../P],
is too large for Baker's inequalities to apply.

To compute j, one may use the fact that ~/96 is an integer close to - exp( 11" ..;p/3log 96), this last number being
np
-1.00028...
-10.0000028...
-55.000000092...
-6670.000000000006...
for
for
for
for
p= 19
p=43
p=67
p= 163
Integrality of j.
Theorem. The j invariant of an elliptic curve with complex multiplication is an algebraic

integer.
In the case h
= 1, as j
E Q, we then have j E Z.
First proof.
We choose r E Rd,! whose norm is a prime number 1. We then have an isogeny
r : E ~ E whose kernel is cyclic of order l. But the invariants j ,j/ of two elliptic curves
related by an isogeny of degree 1 satisfy a relation
F,(j,j')
= 0,
where F, is a symmetric polynomial over Z with highest degree term jlj/I. Here j = j'
and
F,(j,j) = j21 +
21-1
2: aiji = 0,
(ai E Z),
i=O
thus j is an algebraic integer.

Second proof.
That j is integral at a place v is equivalent to E having potential good reduction at
v (that is to say, good reduction after making a finite extension of the ground field). But
if E does not have potential good reduction then it has potential multiplicative reduction
at v, i.e. a reduction to the group G m The endomorphisms of E then operate on G m ,
which is absurd, as the ring of endomorphisms of G m is Z.
(For a generalization of this to abelian varieties, see [ST].)
Remark. H E has complex multiplication by RtJ, then over Q( va), j(E) generates the
Hilbert class field of Q( v'd), cf. e.g. [Shm].
194
A.5. Modular curves associated to normalisers of Cartan subgroups, and their

CM-integral points.
The method.
Let us explain the relation between the class number 1 problem and integral points
on modular curves.
Let p be a sufficiently large prime number (p > 19, say) such that the imaginary
quadratic field Q( A ) has class number h( -p) = 1. Let us consider the elliptic curve
E, unique up to C-isomorphism, which has complex multiplication by the ring of integers
R = R_p of the field Q( A). Its modular invariant j is in Z, thus defines an integral
point of the affine line PI - {oo} which is the simplest modular curve "X(1 )aff". We shall
attach to E some integral points on some other modular curves. After that, there are
essentially three ways to procede:
1). (Heegner, Siegel). A direct method gives all the integral points of the modular curve,
thus giving an explicit list of the possible p's.
2). Baker's method is applied to bound the integral points of the modular curve, hence
bounding p.
3). One obtains a linear form in logarithms for which Baker's lower bound applies. This
method is closely related to the two preceding, and also to the "limit formula" of Kronecker.
Modular curves associated with Cartan subgroups and their normalisers.
Let N be an integer> 1, and X(N) the modular curve which classifies elliptic curves
with a basis of points of order N, see e.g. [DeRa). The field of functions of X(N) is the
field of modular functions of level N rational over Q( VI) cf. [Shm); its field of constants
is Q( VI). The group GN = G~(Z/NZ) acts on X(N), the quotient being the j-line
X(I)
= Pl.
IT H is a subgroup of GN, we write X H for the modular curve X(N)/H; its points
over a field K correspond to elliptic curves E over K such that Gal(K/K) acts on the
N-division points of E via H, see [DeRaJ, IV-3. IT det: H --. (Z/NZ)* is surjective, XH is
absolutely irreducible over Q.
The subgroups H we are interested in are essentially the Carlan subgroups of G N and
some subgroups of their normalisers. More precisely:
Let A be a free commutative etale Z/NZ-algebra of rank 2: if pIN, A/pA is either
Fp x Fp (called the "split" case) or Fp2 (the "non-split" case at p).
Given an arbitrary decomposition of the primes piN into "split" and "non-split" types,
then up to isomorphism there is a unique corresponding such algebra A. In the case where
A is non-split at p for all pIN, we say more briefly that A is "non-split".
Let A * be the multiplicative group of invertible elements of Aj then A * operates on
A, and as A ~ (Z/NZ)2, A* embeds in AutZ/NZ(A) ~ GL2(Z/NZ). The image of A* is
a Cartan subgroup C of GL2(Z/NZ).
As A is an algebra of rank 2, it has a unique automorphism u of order 2, which induces
the natural involution on each A/pA; for all a, a + ua E Z/NZ. Let C = C U uC be the
subgroup of GL 2(Z/NZ) generated by C and u.
Let X Xc and X
Xc be the corresponding curves. As C has index 2 in C, we
have a quadratic covering X --. X.
Let now E be an elliptic curve with complex multiplication by the ring R, and suppose
that N is prime to the discriminant of R. Then the algebra A = R/N R is etale, thus defines
195
a C and a C. Let K be a field over which E is defined, and suppose that Vd 'I K. Let
EN be the kernel of multiplication by N in E(K); this is a free R/NR-module of rank
1. If s E Gal(K/K) is trivial on K(Vd), then the action of s on EN commutes with that
of R/N R, hence is defined by an element of the Cartan subgroup C = (R/ N R)*. If s
is non-trivial on K(Vd), its action on EN comes from an element of C - C. This a.llows
us to attach to E a K-rational point of X, which does not lift to a point of X over K,
but which lifts over K( ..;d). In other words, we obtain two points of X conjugate under
Gal(K( Vd)/ K).
We now take the case which interests us: K = Q, R = R_p and h( -p) = 1. Recall
that every prime number 1< p/4 is inert in Q( FP). In particular if p is large with respect
to N, then C and C are of non-split type.
We then obtain rational points on suitable modular curves X. These points are integral
points in the following sense:
First, by removing the cusps from a modular curve XH, one obtains an affine curve
x;f, which is finite over the j-line Al:
By taking the normalisation of A~, one then obtains an affine scheme X';iffz over Z. A
rational point x of XH is then ca.lled integral if it has the following equiva.le~t properties:
a) j (x) is integral,
b) x is an integral point of the scheme X;f.z.
The points XE E X(Q) constructed above have property a) by A.4: they are indeed
integral.
The case where N is prime.
Assume N = I is a prime number ;?: 5 and consider the curve X corresponding to the
normaliser C of the non-split Cartan subgroup C.
Since C h,as order [2 -1, it has index 1(1-1) in GL2(Z/IZ), thus X -+ P l has degree
1f.1- 1) and X -+ P l has degree I(l - 1)/2. The number of geometric cusps of X (resp.
X) is 1-1 (resp. (1-1)/2). All these cusps have ramification index l. The 1- 1 cusps of
X are conjugate over Gal(Q(~)/Q) and the (1-1)/2 cusps of X are conjugate over the
maximal real subfield of the cyclotomic field Q( ~).
l!y working out the ramification above j = 0,1128 one obtains a formula for the genus
9 of X (in the non-split case):
(l2 - 101 + 33)/24

__ {
(1- 5)2/24
g(1- 3)(1 - 7)/24
(12 - WI + 13)/24
Examples. 1=5,
if I == 1 mod 12
if 1==5 mod 12
if 1== 7 mod 12
if 1== 11 mod 12
9 = 0, 2 cusps; 1= 7, 9 = 0, 3 cusps; 1= 11, 9 = 1, 5 cusps.
Note that, already for I = 7 and I = 11, the number of integral points of X is finite
(and effectively so by Baker's method, once explicit equations for X have been found).
This is enough to give, by method 2, that the set of p's with h( -p) 1 can be effectively
bounded.
196
A.6. Examples.
In what follows, p is a prime number> 3 with h( -p) = 1; we denote by jp the
j-invariant of an elliptic curve with CM by R_p i.e.
(For a model of this curve with good reduction outside p, see [Gr}, p.80.) We have jp E Z
d. A.4. By using the modular curves X associated to N = 2,3, ... we will obtain further
properties of jp (and hence of p).
N = 2. The group GL2 (Z/2Z) is isomorphic to S3. The non-split Carlan subgroup C is
the cyclic subgroup of order 3, and C ~ S3, therefore X = PI and X is a 2-sheeted covering
of Pl' By looking at its ramification, and at the cusps, one can see that X -+ X = PI is
the quadratic extension associated to ..jj - 1728.
Theorem. H h( -p) = 1 and p == 3 (mod 8), tben jp - 17:?.8 = _py2 witb y E Z.
(Indeed, the corresponding point of X is rational over Q(..;=p).) This can also be
checked using the classical formulae
= 3.
Consider the natural homomorphisms
The group C is a 2-Sylow subgroup corresponding to the inverse image of a subgroup of

order 2 of S3. The covering X -+ P l has degree 3, and becomes cyclic over Q( ...;::3). One
shows that it is obtained by extracting the cube root of j. As above this implies:
Theorem. H p is such tbat 3 is inert in Q(..;=p), tben j is a cube: j = 1'3, l' E Z.
The result is again true when 3 splits, for the normaliser of a split Carlan subgroup
is contained in the normaliser of a non-split Carlan subgroup. (Here again an elementary
proof is possible by using an explicit formula for 6. 1/3 see e.g. [Se6}, 5.3(b).)
Remark. If we write jp - 1728 = _py2 and jp
Y = p2y, and k = 2633p 3,
As
l' = _e"vP/3(1
= 1'3, we obtain by putting X = -n,
+ O(e-1I'v'p,
X and Y have order of magnitude the exponential of vp '" P/6. This alone should be
enough to imply a bound for p [it would if even a weak form of the ABC-conjecture" were
true}.
Returning to the case N

that
= 3, if one uses the 2-sheeted covering X
-+
X,
one finds
197
As jp -1728 = _py2, and jp = ..y3, putting tI = y/24z we have
A priori
tI
E Q. In fact, one can show that

p
tI
=
=
19
3
tI
E Z and
43
67
9 21
tI
is divisible by 3:
163
231
N 5. The curve j( has genus 0 and 2 cusps rational over Q( v'5). An explicit parametrisation of it is given in [Si4]. The corresponding integral points can be identified with the
units of Q( v'5).
= 6. The equation"Y =
for N
= 6.
12(1- tl 2 ) given above corresponds in fact to the equation of j(
N = 7. One has 9 = 0, and there are 3 cusps, conjugate over the real cyclotomic subfield
of Q( 0) (cubic field).
One finds that h( -p) = 1, and (V) = -1, gives an exceptional unit of that field. See
[Ke2]. There are analogous results for N = 9.
N = 12. One again finds a curve of genus 0 with 2 cusps which are rational over Q( V3).
One can either use the method of approximating e'lfVP, or pass to level 24.
N = 15. This is the level used by Siegel [Si4]. An elliptic curve is obtained, with a finite
number of integral points.
N
24. An elliptic curve is obtained. This is the level considered in effect by Heegner
[He].
A.7. The Gel'fond-Linnik-Baker method.

_ Let N be the discriminant of a real quadratic field - t~is field is then Q( .Jii). Let
X be the curve attached to the non-split Cartan subgroup C relative to N. There is (see
below) a non-trivial morphism </I over Q of j( onto a curve of genus 0, such that the image
of j(oo is formed of two points P,P' conjugate to each other over the field' Q(.Jii), and
the inverse image of {P,P'} is j(oo. (For N = 5 one can select </I to be an isomorphism.)
Over Q(.Jii) one can interpret </I as a function j( -+ P 1 having neither zeros nor poles
over j(aff and with </1.</1" = cst., where (j is the involution of Q( .Jii).
Here is a first method for showing the existence of such a function, in case N is a prime
1 with 1 == 1 (mod 4). Consider the group of divisor classes generated by the differences
of two cusps. By general results of Drinfeld-Manin and Kubert-Lang, this group is finite.
The cusps are conjugate among themselves under the Galois group (Z/IZ)* /(1). As the
Legendre character is trivial on I, it gives a homomorphism : (Z/IZ)* /(I) -+ (I).
Since has order 2, it gives a decomposition of the cusps into two subsets of (I - 1)/4
elements each. For a suitable integer a ~ 1, one can thus construct a function whose
divisor is a times an element of the first set, minus a times an element of the second.
One obtains thus the desired function </I; that it is defined over Q(Vii) derives from its
uniqueness (up to a constant factor).
198
Classical formulae provide another method for constructing 1/>. For instance, in case
N is a prime 1 == 1 (mod 4), one looks at the action of the Cartan subgroup e on the
projective line PI(F,).
The cyclic group e/F; acts freely and transitively on PI(F,)j its unique subgroup of
index 2 has two orbits A and B. To each x E PI(F,), let us associate the modular form
tJ.", defined by
tJ. oo = tJ.(lz),
tJ.i(Z) = tJ.(z; \
i E Z/IZ,
where tJ. is the standard discriminant. The function I/> given by
can be shown to have the required properties. (One can sometimes use TJ = tJ.I/24 instead
of tJ. e.g. for 1 = 5, cf. [Si4]j this gives a smaller order at infinity.)
From the existence of 1/>, one deduces a bound for the p for which h( -p) = 1 in the
following way. We may assume that p is sufficiently large so that we are in the non-split
case (every prime divisor 1 of N is < p/4). To p is associated an integral point xp of X.
The image I/>(xp) of xp belongs to a quasi-integral subset of PI - {O,oo}. More concretely,
we can write
where Ep is a unit, and c belongs to a fixed set, independent of p. Let EN > 1 be a
fundamental unit of Q( ...IN). It suffices to consider a sequence of p for which I/>(xp) -+ +00,
thus Ep = E';{ where np is a sequence of integers tending to +00. Let us look at the size
of I/>(xp).
As I/> and j are related by a polynomial equation we have a Puiseux expansion
where a and f3 are rational numbers> 0, A =1= o. As
we obtain
where v is a rational number> O. Then
In particular, np has order of magnitude..;p. We thus obtain a linear form in 3 logarithms,

to which one can apply Baker's lower bounds.
199
This had been seen by Gel'fond and Linnik in a different manner. They looked at the
extensions
Q(.../N,~
r ""
r
Q(FNP)
Q(.../N)
Q(A)
""
and considered the L-function attached to the character X of the imaginary quadratic field
Q( A) corresponding to its quadratic extension Q( .../N, FP). We have
(1)
L(s,X)
where XN (resp.
Q( FNP'J/Q.
= L(S,XN)L(s,X-Np),
X-Np) is the Dirichlet character corresponding to Q( .../N)/Q (resp.

In particular:
L(I,X)
= L(l,XN).L(l,X-Np).
The numbers L(I, XN) and L(I, X-Np) are logfN and h( -Np), up to elementary factors.
As for L(I, X), one computes it using Kronecker's limit formula applied to the expansion
L(I, X)
=~
'E XN(Fp(x, yFp(x, y)-I,

x,yEZ
where
Fp ( x, y )
p+l 2
= x 2 + xy + -4Y .
One finds that L(I, X) is essentially log <fo(xp). Using (1), this gives the expression of <fo(xp)
as C';r as above (and gives the values of c and np). For more details see e.g. [Si3].
Baker chose N = 21; then t = tN = H5 + V2i) and one finds
32
Ih(-21p)10gN - -
1r
v'P1 < e-rr-./i/IOO.
21
Note that these linear forms in logarithms have algebraic numbers as coefficients (for
example H-iv'P), so that Baker's method does apply, and gives an effective bound for p.
Remarks. 1) Note the connection between logarithms of units and values of L-functions
at s = 1! It has been the starting point of Stark's conjectures, cf. [StJ and [T3J.
2) It is not known how to solve the class number 2 problem by methods of integral points
on modular curves. But this problem can be attacked via linear forms in logarithms in the
following way.
Suppose h( -d) = 2. The difficult case is that where d = PI]J2, with PI and P2 large
prime numbers, PI == 1 (mod 4), ]J2 == 3 (mod 4). One knows that the corresponding
j-invariant belongs to Q(y'Pl). Then <fo(Xd) is essentially a unit in Q(.../N,y'Pl). Only 2
of the 3 fundamental units of this biquadratic field are involved, namely tN and Npt. By
working this out, Baker obtains [Ba, p.53J:
Ih( -21d) log EN
+ h(21pI)h( -21]J2) log t21pt - ~~ nidi < e- Vd/ 10
As t21pt is a fairly large algebraic number, one must use delicate lower bounds to be able
to conclude.
200
Bibliography.
[A] S. ARAKELOV - Intersection theory of divisors on an arithmetic surface
(in Russian), Izv. Akad. Nauk SSSR 8 (1974), 1167-1180.
[Ba] A. BAKER - Transcendental number theory, Cambridge Univ. Press,
Cambridge, 1975.
[Be] G. V. BELYI - On the Galois extensions of the maximal cyclotomic field
(In Russian), Izv. Akad. Nauk. SSSR 43 (1979), 267-276.
[Ber] D. BERTRAND - Lemmes de zeros et nombres transcendants,
Seminaire Bourbaki 1985-86, expose 652 (Asterisque 145-146, S. M. F. 1987,
21-44).
[BM] A. BAKER and D.W. MASSER (edit.) - Transcendence Theory:
Advances and Applications, Acad. Press, London, 1977.
[Bo] F.A. BOGOMOLOV - Sur l'algebricite des representations l-adiques,
C.R. Acad. Sci. Paris, serie A, 290 (1980), 701-703.
[Boml] E. BOMBIERI - Le grand crible dans La theorie analytique des nombres, Asterisque 18 (1974), 2nd ed. (1987).
[Bom2] E. BOMBIERI - On exponential sums over finite fields II, Invent.
Math. 47 (1978), 20-39.
[Bor] A. BOREL - Some finiteness properties of adele groups over number
fields, 1.H.E.S. Publ. Math. 16 (1963), 5-30 (= Oe., II, no. 60, p.329-354).
[BSD] B.J. BIRCH and H.P.F. SWINNERTON-DYER - Notes on elliptic
curves, I, J. Crelle 212 (1963), 7-25; II, ibid., 218 (1965), 79-108.
[BSh] Z. BOREVICH and 1. SHAFAREVICH - Number Theory, Acad.
Press, New York, 1966.
[Cal J.W.S. CASSELS - Diophantine equations with special reference to
elliptic curves, J. London Math. Soc. 41 (1966), 193-29l.
[Ch] C. CHABAUTY - Sur les points rationnels des courbes algebriques,
C.R.Acad. Sci. Paris 212 (1941), 882-885.
[CN] K. CHANDRASEKHARAN and R. NARASIMHAN - Functional
equations with multiple gamma factors and the average order of-arithmetical
functions, Ann. of Math. 76 (1962), 93-136.
[Co] S.D. COHEN - The distribution of Galois groups and Hilbert's irreducibility theorem, Proc. London Math. Soc. 43 (1981), 227-250.
201
[Col] R. COLEMAN - Effective Chabauty, Duke Math. J. 52 (1985), 765770.
[Cu] G.V. CHUDNOVSKY - Formules d'Hermite pour les approximants de
Pade de logarithmes et des fonctions binomes, et mesures d'irrationalite, C.
R. Acad. Sci. Paris, Ser. A, 288 (1979), 965-967.
[CW] C. CHEVALLEY and A. WElL - Un theoreme d'arithmetique sur
les courbes algebriques, C.R. Acad. Sci. Paris 195 (1932), 570-572. (= A.
Weil,Oe., I [1932c]).
[D] V.A. DEMJANENKO - Rational points of a class of algebraic curves

(in Russian), Izv. Akad. Nauk SSSR 30 (1966), 1373-1396.
[De] H. DELANGE - Generalisation du theoreme de Ikehara, Ann. Sci.
ENS 71 (1954), 213-242.
[DeRa] P. DELIGNE and M. RAPOPORT - Les schemas de modules de
courbes elliptiques, Lect. Notes in Math. 349, Springer-Verlag, 1973, 143316.
[Dr] V.G. DRINFELD - Two theorems on modular curves (in Russian),
Funk. Anal. i Prilozhen. 7 (1973), 83-84.
[DR] H. DAVENPORT and K.F. ROTH - Rational approximation to algebraic numbers, Mathematika 2 (1955), 160-167. (= H. Davenport, Coil.
Works, II, p.689-696).
[E] T. EKEDAHL - An effective version of Hilbert's irreducibility theorem,
preprint, Stockholm, 1987.
[F] G. FALTINGS - Endlichkeitssatze fiir abelsche Varietaten iiber
Zahlkorpem, Invent. Math. 73 (1983), 349-366.
[Fe] D. FENNA - Rational approximation with series, J. Austral. Math.
Soc. 2 (1961/2), 107-119.
[FJ] M.D. FRIED and M. JARDEN - Field Arithmetic, Springer-Verlag,
1986.
[Fr] M. FRIED - Constructions arising from Neron's high rank curves,

Trans. AMS 281 (1984), 615-631.
[FW] G. FALTINGS, G. WUSTHOLZ et al. - Rational Points, Vieweg,
Braunschweig, 1984.
202
[G] A. GROTHENDIECK - Elements de Geometrie Algebrique (written
with the collaboration of J. Dieudonne), I.H.E.S. Publ. Math. 1960-67.
[GH] P. GRIFFITHS and J. HARRIS - Principles of Algebraic Geometry,
Wiley, 1978.
[Go1] D. GOLDFELD - The class numbers of quadratic fields and the conjectures of Birch and Swinnerton-Dyer, Ann. Sci. Norm. Sup. Pis a 3
(1976), 623-663.
[Go2] D. GOLDFELD - Gauss' class number problem for imaginary
quadratic fields, Bull. AMS 13 (1985), 23-37.
[Gr] M. GREENBERG - Rational points in Henselian discrete valuation
rings, I.H.E.S. Pub!. Math. 31 (1966), 56-64.
[Gy) K. GYORY - Resultats effectifs sur la representation des entiers par des
formes decomposables, Queen's Papers in Pure and Applied Math., no.56
(1980).
[GZ) B.H. GROSS and D.B. ZAGIER - Heegner points and derivatives of
L-series, Invent. Math. 84 (1986), 225-320.
[H] R. HARTSHORNE - Algebraic Geometry, GTM 52, Springer-Verlag,
1977.
[Ha] M. HARRIS - Systematic growth of Mordell-Weil groups of abelian
varieties in towers of number fields, Invent. Math. 51 (1979), 123-14l.
[He] K. HEEGNER - Diophantische Analysis und Modulfunktionen, Math.
Zeit. 56 (1952), 227-253.
[HH] D. HILBERT and A. HURWITZ - Uber die diophantischen Gleichungen vom Geschlecht Null, Acta Math. 14 (1891), 217-224. (= D. Hilbert,
'Ges. Abh., Bd. 2,258-263 = A. Hurwitz, Math. Werke, Bd.2, 116-121).
[Hi] D. HILBERT - Uber die Irreducibilitat ganzer rationaler Funktionen
mit ganzzahligen Koeffizienten, J. Crelle 110 (1892),104-129 (= Ges. Abh.,
II, 264-286).
[Hn1] M. HINDRY - Points de torsion sur les sous-varietes abeliennes, C.
R. Acad. Sci. Paris 304 (1987), Serie 1, 311-314.
[Hn2] M. HINDRY - Geometrie et hauteurs dans les groupes algebriques,
Universite de Paris VI, These, 1987.
203
[Ho] T. HONDA - Isogenies, rational points and section points of group
varieties, Jap. J. Math. 30 (1960), 84-101.
[Hu] M.N. HUXLEY - The large sieve inequality for algebraic number fields,
Mathematika 15 (1968), 178-187.
[Hup] B. HUPPERT - Endliche Gruppen I, Springer-Verlag, 1967.
[HW] G.H. HARDY and E.M. WRIGHT - An introduction to the theory of
numbers, 5th ed. Oxford Univ. Press, London 1979.
[Is] V.V. ISHANOV - The semi-direct imbedding problem with a nilpotent
kernel, Izv. Akad. Nauk SSSR 40 (1976), 3-25 (= Math. USSR Izv. 10
(1976), 1-23).
[K] S. KLEIMAN - Toward a numerical theory of ampleness, Amer. J.
Math. 84 (1966), 293-344.
[Kel] M.A. KENKU - On the number of Q-isomorphism classes of elliptic
curves in each Q-isogeny class, J. Number Theory 15 (1982), 199-202.
[Ke2] M.A. KENKU - A note on the integral points of a modular curve of
level 7, Mathematika 32 (1985),45-48.
[KL] D. KUBERT and S. LANG - Modular Units, Springer-Verlag, Heidelberg 1981.
[Ll] S. LANG - Abelian Varieties, Interscience Publ., New York, 1959.
[L2] S. LANG - Diophantine Geometry, Interscience PubL, New York, 1962.
[L3] S. LANG - Les formes bilineaires de Neron et Tate, Sem. Bourbaki
1963/64, expose 274, Benjamin Publ., New York, 1966.
[L4] S. LANG - Division points on curves, Ann. Mat. Pura Appl. IV, 70
(1965), 229-234.
[L5] S. LANG - Elliptic Curves-Diophantine Analysis, Springer-Verlag, Heidelberg, 1978.
[L6J S. LANG - Fundamentals of Diophantine Geometry, Springer-Verlag,
Heidelberg; 1983.
[L7J S. LANG - Integral points on curves, LH.E.S. Publ. Math. 6 (1960),
27-43.
[La] E. LANDAU - Uber die Anzahl der Gitterpunkte in gewisser Bereichen,
Gott. Nach. (1912),687-771, and (1915), 209-243.
204
[LaM] S. LAMACCHIA - Polynomials with Galois group PSL(2, 7), Comm.
in Algebra 8 (1980), 983-992.
[Lau] M. LAURENT - Equations diophantiennes exponentielles, Invent.
Math. 78 (1984), 299-327.
[Len] H.W. LENSTRA, Jr. - Rational functions invariant under a finite
abelian group, Invent. Math. 25 (1974), 299-325.
[Lev] J. LEVEQUE - Rational points on curves of genus 1, J. Crelle 206
(1961), 45-52.
[Lg] M. LANGEVIN - Quelques applications de nouveaux n~sultats de van
der Poorten, Seminaire Delange-Pisot-Poitou 17 (1976), no. G12.
[LN] S. LANG and A. NERON - Rational points of abelian variants over
function fields, Amer. J. Math. 81 (1959), 95-118.
[LU] R.C. LYNDON and J.L. ULLMAN - Groups generated by two
parabolic linear fractional transformations, Canad. J. Math. 21 (1969),
1388-1403.
[LW] S. LANG and A. WElL - Number of points on varieties in finite fields,
Amer. J. Math. 76 (1954), 819-827 (= A. Weil, Oe., II, [1954f]).
[M] L.J. MORDELL - On the rational solutions of the indeterminate equations of the third and fourth degrees, Proc. Cambridge Phil. Soc. 21
(1922), 179-192.
[Mal] Y. MANIN - The p-torsion of elliptic curves is uniformly bounded (in
Russian), Izv. Akad. Nauk SSSR 33 (1969), 459-465 (= Math. USSR Izv.
3 (1969), 433-438).
[Ma2] Y. MANIN - Cyclotomic fields and modular curves (in Russian), Usp.
Mat. Nauk 26 (1971), 7-71 (= Russian Math. Surveys 26 (1971), 7-78).
[Ma3J Y. MANIN - Parabolic points and zeta functions of modular curves
(in Russian), Izv. Akad. Nauk SSSR 36 (1972), 19-66 (= Math. USSR Izv.
6 (1972), 19-64).
[Mae] T. MAEDA - Noether's problem for A5, J. of Algebra 125 (1989),
418-430.
[Masl] D.W. MASSER - Diophantine approximation and lattices with complex multiplication, Invent. Math. 45 (1978), 61-82.
205
[Mas2] D.W. MASSER - Small values of the quadratic part of the NeronTate height, Seminaire Delange Pisot Poitou, 1979-1980, Birkhauser-Verlag,
1981, 213-222.
[Mat] B.H. MATZAT - Konstruktive Galoistheorie, Lect. Notes in Math.
1284, Springer-Verlag, 1987.
[Mel] J-F. MESTRE - Formules explicites et minorations de conducteurs
de varietes algebriqu~s, Compo Math. 58 (1986), 209-232.
[Me2] J-F. MESTRE - Extensions regulieres de Q(T} de groupes de Galois
J. of Algebra 131 (1990), 483-495.
An,
[Me3] J-F. MESTRE - Courbes elliptiques de rang 2': 12 sur Q(T), C. R.

Acad. Sci. Paris 313 (1991), 171-174.
[MFT] Y. MANIN, J. FRANKE, Y. TSCHINKEL - Rational points of
bounded height on Fano varieties, Invent. Math. 95 (1989), 421-435.
[Mi] M. MIGNOTTE - Approximation des nombres algebriques, Pub!.
Math. Orsay, no. 77.74 (1977).
[MM] G. MALLE and B.H. MATZAT - Realisierungvon Gruppen PSL2 (F p)
als Galoisgruppen tiber Q, Math. Ann. 272 (1985),549-565.
[Mo] H.L. MONTGOMERY - The analytic principle of the large sieve, Bull.
AMS 84 (1978), 547-567.
[Mull D. MUMFORD - A remark on Mordell's conjecture, Amer. J. Math.
87 (1965), 1007-1016.
[Mu2] D. MUMFORD - Abelian Varieties, 2nd ed., Tata Inst., Oxford Univ.
Press, Oxford, 1974.
[Mz1] B. MAZUR - Rational points of abelian varieties with values in towers
of number fields, Invent. Math. 18 (1972), 183-266.
[Mz2] B. MAZUR - Rational isogenies of prime degree, Invent. Math. 44
(1978), 129-162. (See also Sem. Bourbaki 1977/78, expose 511.)
[N1] A.NERON - Problemes arithmetiques et geometriques rattaches a. la
notion de rang d'une courbe algebrique dans un corps, Bull. Soc. Math.
France 80 (1952), 101-166.
[N2] A. NERON - Proprietes arithmetiques de certaines familles de courbes
algebriques, Proc. Int. Congo Amsterdam, 1954, voLIII, 481-488.
206
[N3] A. NERON - Arithmetique et classes de diviseurs sur les varietes
algebriques, Proc. Int. Symp. Tokyo-Nikko, 1955, 139-154.
[N4] A. NERON - Quasi-fonctions et hauteurs sur les varietes abCliennes,
Ann. of Math. 32 (1965), 249-33l.
[Na] K. NAKATA - On some elliptic curves defined over Q of free rank 2: 9,
Manus. Math. 29 (1979), 183-194.
[No] D.G. NORTHCOTT - An inequality in the theory of arithmetic on
algebraic varieties, Proc. Cambridge Phil. Soc. 45 (1949), 502-509; A
further inequality ... , ibid., 510-518.
[Ne] J. NEUKIRCH - On solvable number fields, Invent. Math. 53 (1979),
135-164.
[Oe1] J. OESTERLE - Nombres de Tamagawa et groupes unipotents en

caracteristique p, Invent. Math. 78 (1984), 13-88.
[Oe2] J. OESTERLE - Le probleme de Gauss sur Ie nombre de classes,

L'Ens. Math. 34 (1988), 43-67.
[Oe3} J. OESTERLE - Nombres de classes des corps quadratiques imaginaires, Sem. Bourbaki no.631, Asterisque 121-122, SMF, 1985.
[P] H. POINCARE - Sur les propriCtes arithmetiques des courbes
algebriques, J. de Liouville 7 (1901), 166-233 ( = Oe., V, 483-548).
[Pal A.N. PARSHIN - Algebraic curves over function fields I (in Russian),
Izv. Akad. Nauk SSSR 32 (1968), 1191-1219.
[Raj D.A. RAIKOV - A generalization of the ikehara-Landau theorem (in
Russian), Math. Sbornik 45 (1938), 559-568.
[R11 M. RAYNAUD - Courbes sur une variCte abelienne et points de torsion,
Invent. Math. 71 (1983), 207-233.
[R21 M. RAYNAUD - Around the Mordell conjecture for function fields and
a conjecture of Serge Lang, Lect. Notes in Math. 1016, Springer-Verlag,
(1983), 1-19.
[R3J M. RAYNAUD - Sous-varietes d'une variCte abClienne et points de torsion, Arithmetic and Geometry, Progress in Math. 35, Birkhauser-Verlag,
Boston (1983), 327-352. (See also Sem. Bourbaki 1983-84, expose 625.)
207
[Rob] A. ROBINSON and P. ROQUETTE - On the finiteness theorem of
Siegel and Mahler concerning Diophantine equations, J. Number Theory 7
(1975), 121-176.
[Ro] P. ROQUETTE - Non standard aspects of Hilbert's irreducibility theorem, Lect. Notes in Math. 498, Springer-Verlag, 1975, 231-275.
[Sch] S.H. SCHANUEL - Heights in number fields, Bull. Soc. Math. France
107 (1979), 433-449.
[Schi] A. SCHINZEL - On Hilbert's irreducibility theorem, Acta Arith. 16
(1965), 334-340.
[Schml] W.M. SCHMIDT - Approximation to algebraic numbers,
L'Ens. Math. 17 (1971) 187-253. (Monographie no. 19, 1972).
[Schm2] W.M. SCHMIDT - Diophantine Approximation, Lecture Notes in
Math. 785, Springer-Verlag, 1980.
[Schul 1. SCHUR -- Gleichungen ohne Affekt, Sitz. Preuss. Akad. Wiss.
1930, Physik.-Math. Klasse 443-449 (=Ges. Abh. III, 191-197).
[Se1] J-P. SERRE - Quelque applications du theoreme de densite de Chebotarev, Publ. Math LH.E.S. 54 (1981) 123-201 (=Oe. 125).
[Se2] J-P. SERRE - Quelque proprietes des groups algebriques commutatifs,
Asterisque 69-70 (1979), 191-202 (=Oe. 122).
[Se3] J-P. SERRE - Abelian l-adic representations and elliptic curves, Benjamin, New York 1968; 2nd edition, Addision-Wesley, 1989.
[Se4] J-P. SERRE - Divisibilite de certaines fonctions arithmetiques, L'Ens
Math. 22 (1976), 227-260 (=Oe. 108).
[Se5] J-P. SERRE - Groupes de Galois sur Q, Seminaire Bourbaki 1987/88,
expose 689 (Asterisque 161-162, SMF, 1988).
[Se6] J-P. SERRE - Proprietes galoisiennes des points d'ordre fini des courbes
elliptiques, invent. Math. 15 (1972), 259-231 (=Oe. 94).
[Se7] J-P. SERRE - Topics in Galois Theory, notes written by H. Darmon,
Jones and Bartlett Publ., Boston, 1992.
[Sh1] K-y. SHIH - On the construction of Galois extensions of function
fields and number fields, Math. Ann. 207 (1974), 99-120.
[Sh2J K-y. SHIH - p-division points on certain elliptic curves, Compo Math.
36 (1978), 113-129.
208
[Shaj I.R. SHAFAREVICH - Construction of fields of algebraic numbers
with given solvable Galois groups, Izv. Akad Nauk SSSR 19 (1954) 525578; AMS Transl. Series 2,4 (1956), 107-142. (=C. P. 139-191).
[Shm] G. SHIMURA - Introduction to the arithmetic theory of automorphic
functions, Publ. Math. Soc. Japan, voL 11, Princeton Univ. Press, 1971.
[Si1] C.L. SIEGEL - Uber einige Anwendungen diophantischer Approximationen, Abh. Pr. Akad. Wiss. (1929), no.1 (=Ges. Abh., I, 209-266).
[Si2] C.L. SIEGEL - Abschatzung von Einheiten, Gott. Nach. (1969),71-86
(= Ges. Abh., IV, 66-81.)
[Si3] C.L. SIEGEL - Lectures on advanced analytic number theory, Tata
Institute, Bombay 1965.
[Si4] C.L. SIEGEL - Zum Beweise des Starkschen Satzes, Invent. Math. 5
(1968), 180-191 (=Ges. Abh., IV, 41-52).
[Sk] T. SKOLEM - Diophantische Gleichungen, Springer-Verlag, 1938
(Chelsea Publ. Comp., New York, 1950).
[Sp] V.G. SPRINDZUK - Hilbert's irreducibility theorem and rational
points on algebraic curves, Soviet Math. Dokl. 20 (1979), 701-705.
[St] H. STARK - Values of L-functions at s = 1, I,II,III,IV; Adv. in Math.
7 (1971), 301-343; 17 (1975), 60-92; 22 (1976), 64-84; 35 (1980), 197-235.
[ST] J-P. SERRE and J. TATE - Good reduction of abelian varieties, Annals
of Math. 78 (1968), 492-517.
[Sw] H.P.F. SWINNERTON-DYER - Analytic theory of abelian varieties,
London Math. Soc. Lecture Notes 14, C.U.P. (1974).
[Sz] L. SZPIRO - Seminaire sur les pinceaux arithmetiques: La conjecture
de Mordell, Asterisque 127 (1985), S.M.F.
[T1] J. TATE - The arithmetic of elliptic curves, Invent. Math. 23 (1974),
179-206.
[T2] J. TATE - Quasifonctions locales, letters to J-P. Serre, 21 June 1968,
and 1st October 1979 (see [L5], [Z]).
[T3] J. TATE - Les conjectures de Stark sur les fonctions L d'Artin en
s = 0, Progress in Math. 47, Birkhauser-Verlag, Boston, 1984.
209
[Ta] T. TATUZAWA - On a theorem of Siegel, Jap. J. of Math. 21 (1951),
163-178.
[Ti] R. TIJDEMAN - Exponential diophantine equations, Proc.
Congress Math. Helsinki, Vol. 1 (1978), 381-387.
Inter.
[To] J . TOP - Neron's proof of the existence of elliptic curves over Q with
rank at least 11, preprint, Utrecht 1987.
[U] S. UCHIYAMA - On the Thue-Siegel-Roth theorem, III, Proc. Japan
Acad. 36 (1960), 1-2.
[Wa1] M. WALDSCHMIDT - Nombres transcendants et groupes algebriques,
Asterisque 69-70 (1979).
[Wa2] M. WALDSCHMIDT - A lower bound for linear forms in logarithms,
Acta. Arith. 37 (1980), 257-283.
[WI] A. WElL - L'arithmetique sur les courbes algebriques, Acta Math. 52
(1928), 281-315 ( = Oe., I, [1928]).
[W2] A. WElL - Sur un theoreme de Mordell, Bull. Sc. Math. 54 (1929),
182-191 ( =Oe., I, [1929]).
[W3] A. WElL - Varietis abeliennes et courbes algebriques, Hermann, Paris,
1948.
[W4] A. WElL - Remarques sur un memoire d'Hermite, Arch. d. Math. 5
(1954),197-202 (= Oe., II, [1954a]).
[Wh] E.T. WHITTAKER and G.N. WATSON - A course of modern analysis, 4th edition, C.U.P. (1969).
[Z] H.G. ZIMMER - Quasifunctions on elliptic curves over local fields,
J. Crelle 307/308 (1979), 221-246; Corrections, ibid. 343 (1983), 203-21l.
[Za] D. ZAGIER - L-series of elliptic curves, the Birch - Swinnerton- Dyer
conjecture, and the class number problem of Gauss, Notices AMS 31 (1984);
739-743.
210
INDEX
ABC conjecture
196
Abelian variety
2,31
- dual abelian variety
36-39,45
- elliptic curve (see Elliptic curve)
- approximation theorem on abelian varieties
4,95,98-101
- Mordell-Weil theorem (see Mordell-Weil theorem)
- normalised heights (see Height, normalised)
. . . . . . . . . . . . . . . . . . . . . 43-44,53-54,69
- torsion points
- points of bounded height . . . . . . . . . . . . . . . . .
53-55
- Manin-Demjanenko theorem (see Manin-Demjanenko theorem)
- generalised Mordell conjecture
73-74
- Poincare divisor
37,76,78
- Manin-Mumford conjecture
73-74
Abhyankar's lemma . . . . . .
116
Absolute values, satisfying a product formula
7-10
Albanese variety Alb(X) . .
45,62
Antisymmetric divisor class
.32
Approachable to within l/q6
.96
Approximation of real numbers
95-97
Approximation theorem on an abelian variety
98-101
- effectivity . . . . . . . . . . . . .
99-101, 106-107
- Thue-Siegel-Roth theorem (see Thue-Siegel-Roth theorem)
105
Artin-Schreier polynomial . . . . . . . . . . . . . . . . .
Baire's category theorem . . . . . . . . . . . . . . . . . . . . . . . . . 61
Baker's lower bounds for linear forms in logarithms 5,93,100,110-112,189,194,198-199
5,94,97,108-120
Baker's method (see also Integral points) .
- on pI - {O, 1, oo} . .
112-114
115,117
- on elliptic curves
115-116
- on hyperelliptic curves
- on superelliptic curves
117
. . 114
- on Thue curves
119-120
- on XI(n) . . . . . .
- on X(2) . . . . . .
. . 118
- elliptic curves with good reduction outside a given set of places
118-120
- effectivity.
5,111,114,115,117
Banach space
29,88
Belyi's theorem
70-73
Belyi-Fried-Matzat-Thompson theorem
. 150
Bertini's theorem . . . . . . . .
127,130
. . 189
Birch Swinnerton-Dyer conjecture. . .
. 26-27,157
Blowing up . . . . . . . . . . . .
Bombieri-Davenport large sieve constant
. 172
81-83
Bounded sets of points
Brauer group
. 100
211
Cartan subgroup . . .
- split and non-split
- normaliser of
Cartier divisor . .
Catalan equation .
Chabauty's theorem
Chebotarev density theorem
Chevalley-Weil theorem . .
Classification of finite simple groups
Class number of imaginary quadratic fields
- class number 1
- class number 2
Cohen's theorem
Complex multiplication, see Elliptic curve
Cube, theorem of the,
Cubic resolvent
. .. . .. . ... .
Curves
- elliptic, see Elliptic curve
- exceptional (Siegel's theorem)
- hyperelliptic . . . . . . .
- modular (see Modular curves)
- superelliptic
- Tate . . . . . . . . . . .
- Thue . . . . . . " . . . .
- Mordell's conjecture (see Mordell's conjecture)
- Chabauty's theorem (see Chabauty's theorem)
- Mumford's inequality (see Mumford's inequality)
- Mumford's theorem (see Mumford's theorem)
Cusps, on modular curves
Davenport-Halberstam theorem
Demjanenko- Manin theorem
Descent lemma (see also Mordell-Weil theorem)
Differential form
Dirac b"-function
Dirichlet series .
Divisor, on a variety
- support of
- algebraically equivalent to zero
- ample . . . . . . . . . . .
- class group . . . . . . . . .
- height associated to a (see also Height)
- Poincare divisor class . . . . . . .
- symmetric and antisymmetric divisor classes
- a -divisor , on a jacobian . . . . . . . . .
Effectivity
- of the Mordell-Weil theorem
194
194
194
. 83
117
3,58-62
. . 61
3,50,51,109
151
188-199
188-199
199
6,177
32,34,38,77
124
. . 94-95,110
. 115-116,142
117
. 91
114
68,71,73,118
166-172
3,58,62-67
. 3,53
59,68,90
166
44,181,182
. . . 21
. . . 21
25-26,44-45
. . . 22
. . 21
2,22-24
37,76
32
. .74
. . . . . . . . . . . . . . . . 52,99-100,107
212
- of the approximation theorem on an abelian variety
. . . . . . . 99-101
- of Siegel's theorem
5,100-101,106-107,116
- of Baker's method . . . . . . . .
5,111,114,115,117
- of Chabauty's theorem . . . . . .
. . . 60
- of the Manin-Demjanenko theorem
. . . 63
- effective construction of an elliptic curve EjQ of rank ~ 9
155-156
Embedding a field finitely generated over Q in Qp
61-62
Elliptic curve
............ .
- torsion points
. 69
........ .
- complex multiplication
44,70,93,119,191-194,196
- with good reduction outside a finite set of places
. . . . . . . 118-120
67,118,145-147,158,162,191-193
- j-invariant . . . . . . . . . . .
- local heights . . . . . . . . . . . .
90-93
- normalised heights (see also Height) . .
40-41,90-93
- integral points (see also Integral points)
97,115,117
- oflarge rank over Q . . . . . . . .
121,154-162
- explicit form of Mordell-Weil theorem
56-57
- Tate module . . . . . . . .
. 70,119
- conductor . . . . . . . . .
. . 119
Exceptional curve (Siegel's theorem)
94-95,110
104
Exceptional units
Fermat curve . .
Fermat quartics (Demjanenko's theorem)
Function fields . . . . . . . . . . . .
- number of points of bounded height in pn
- heights
.............. .
Functions of degree ::; 2 between abelian groups
Fundamental group
ro(N)
GalT
Galois cohomology
Galois group, construction of field extensions of Q with given -,
- Sn
..................... .
- An . . . . . . . . . . . . . . . . . . . . . .
- using elliptic curves: GL2(ZjnZ), PSL 2 (F p), SL 3 (F 2 )
- infinite
............... .
- finite simple
. . . . . . . . . . .
- with rigid family of rational conjugacy classes
- Belyi-Fried-Matzat-Thompson theorem
- Shih's theorem
- Noether's method . . .
- Schur's examples
Gel 'fond-Linnik-Baker method
Generalised Riemann Hypothesis (GRH)
Goldfeld's theorem
Grassmannian Gn,d . . . . . . . . .
110
66-67
7-8,19
. 19
11-13
32-34,38
141,150
. 68
138
51-52
6,121
138-144,145
144-145
145-147
147-149
151
149-150
150
146-147
147
145
197-199
188
189
127
213
Greenberg's theorem . . . . . . .
Gross-Zagier theorem . . . . . . .
Grothendieck-Deligne (Weil conjectures)
Hasse's theorem . . . . . .
Height
- of a rational point H(x)
- logarithmic height h( x )
- associated to a morphism h,,( x )
- associated to a line bundle h c ( x )
- elementary properties . . . . .
- functoriality of . . . . . . . .
- behaviour under change of coordinates
- change of height under projection . .
- normalised height h(x), hc(x) . . . .
- bilinear form Bc(x,y) associated to hc(x)
- non-degeneracy of normalised heights
- quadraticity of normalised heights on abelian varieties
- relation between quadratic and linear parts of hc ( x )
- local height (see Local heights)
- associated to a Poincare divisor . . . . . . . . .
- associated to a divisor algebraically equivalent to zero
- associated to a torsion divisor
- for number fields
- for function fields
- positivity
_ on pn . . . . .
- Northcott's finiteness theorem
- Schanuel's theorem
. . 83
188,189
184
156
2-6
2,11
19-20
2,22-28
10-16
. . 23
. . 13
13-16, 19-20
2,30-31,35-43,63,134
36,41
41-43
35-41
38-39
36-39
25-26,44-45
24,46-48
. . 11
11-13
24-25
10-13
16-17
17-19
Hensel's lemma
. . 62
Hermite's finiteness theorem
3,49-52,109
Hermitian structure on a line bundle
84-85
Hilbert's irreducibility theorem . . .
5-6,121,130,149
- hilbertian field (see also Hilbertian field)
129-130,137-138
- Q is hilbertian . . . . . . . . . . .
130-132
- thin sets (see Thin sets)
- specialisation of Galois groups (see also Galois groups)
122-126,137-138
- relation with integral points. . . . . . . . . . . .
. . . . . 135
- Neron's specialisation theorem, see Neron's specialisation theorem
129-130,137-138
Hilbertian field . . . . . . . . . . . . . . . . . . . . . . .
- finitely generated extensions of hilbertian fields are hilbertian
. . 130
Honda's conjecture . .
. . 162
Hurwitz' genus formula
142-143
Hyperelliptic curve
115-116
Integral points . .
- behaviour under morphisms .
4-6,94-95,97,100,102-106
. . . . . . 108-110,114
214
- quasi-integral sets . . . . . . . . .
94-95
94,104-105
- S-integral points . . . . . . . . .
- Siegel's theorem (see also Siegel's theorem)
95,102-104
- exceptional curves . . . . . . . . . . .
. 94-95,110
5,94,97,108-120
- Baker's method (see also Baker's method)
5,102-104,112-114
- on PI - {O, 1,00}
97,115,117
- on hyperelliptic curves
. 5,115-116
117
- on superelliptic curves
114
- on Thue curves
119-120
- on XI(n)
- on X(2)
118
- of bounded height
6,115,117,177-178
118-120
- elliptic curves with good reduction outside a finite set of places
194-199
- on modular curves and the class number 1 problem
Intersection multiplicities
85-86
Inertia subgroup
140,143,150
Jacobian of a curve
- is principally polarised
- e divisor . . . . . .
- Poincare divisor class
1-2,5,46,58-59,66,73,98,101,134
46,76
74
76
Kissing number
Kronecker's limit formula
Kubert-Lang . .
Kummer theory
79
194,199
115,119-120,197
55-56
L-function . . .
Lang-Weil theorem
Large point
Large sieve inequality
- Bombieri-Davenport bound
- Selberg's bound . . . . .
- Davenport-Halberstam theorem
- improved inequalities
Legendre's equation of an elliptic curve
Lenstra's example
Lie group
Linear torus . .
Line bundle, generalities
- ample . . . . .
- generated by its global sections
- the Picard group Pic(X) .
Local heights
....... .
- as intersection multipicities
- normalisation on abelian varieties
- in terms of theta functions
188,189,198-199
. 6,62,184
79,80
. 6,163-164
172
171-172
166-172
170-172
118
147
60,149
. . 73
20-22
. . 22
21-22
20-22
83-93
85-86
87-89
88-89
215
- case of an abelian variety with good reduction
- relation with global heights . . . .
- Tate normalisation on elliptic curves
Locally compact field . . . . . . . . .
Logarithmic height (see also Height)
Lower bounds for linear forms in logarithms (see also Baker's method)
. 89
89-90
90-93
. 81
2,11
5,94,110-112
Manin-Demjanenko theorem . . . . . . . . . . . . .
........ .
- application to modular curves
- application to Fermat quartics ax4 + by4 + cz 4 = 0
- application to the Tate module of an elliptic curve
Manin-Mumford conjecture
Manin-Drinfeld theorem
Mazur's theorem
Modular curves
- cusps
- Belyi's theorem .
- Manin's theorem
- Manin-Drinfeld theorem
- associated to Cartan subgroups and their normalisers
58,62-66,154
67-69
66-67
73-74
. 76
73,197
. 69
X(N)
Xo(N)
X1(N)
Yo(N)
69,71,73,118,195-197
70-73
67-69
. 73
194-197
118,193
67-70,146
119-120
. . . 67
1,3-4,58-59,105,133
73-74
. . . . . . 80,105
Mordell's conjecture
- generalised .
- in characteristic p > 0
- Chabauty's theorem (see Chabauty's theorem)
- Mumford's theorem (see Mumford's theorem)
- Manin-Demjanenko theorem (see Manin-Demjanenko theorem)
- Manin-Mumford conjecture (see Manin-Mumford conjecture)
Mordell-Weil theorem
1,3-4,52,58,99,100
- classical descent
. . . . . 53
- explicit form
55-57
- effectivity
52,99-101,107
- for finitely generated ground fields
. . . 52
- weak Mordell-Weil theorem (see also Weak Mordell-Weil theorem)
3,51-52,99
Mordell-Weil group
43-44,51-53
- generators for
. . . 99
- rank . . . .
3,58-59,69
- torsion subgroup
43-44,54,69
Multiplicative function
182
Mumford's inequality
. . . 77
Mumford's lemma
. . . 23
Mumford's theorem .
3-4,58,74-80,105,133
- optimal for function fields
. . . . . . . . 80
NS(X), Neron-Severi group
26,31-32,47,62,64,157
216
Nakai-Moishezon criterion . . . .
26,27
. . 89
Neron model of an abelian variety
Neron's estimate . . . . . . .
53-55
Neron's specialisation theorem for abelian varieties
59,152-154
154
- Silverman's theorem . . . . . . . . . . .
154
- specialisation of extensions of abelian varieties
- specialisation of non-commutative groups
154
Neron's normalisation of local heights . . . . . . .
87-89
...........
25,31-32,47,62,64,157
Neron-Severi group NS(X)
Neron-Tate normalisation (see also Height - normalised)
. . . . . 2,29-31,35
................
. . . . .
147
Noether's method
Non-degeneracy of normalised heights (see also Height, Normalised height)
42-43
Normalised heights
2,30-31,34-48
- non-degeneracy
42-44
- positivity
. . 41
35-36
- quadrati city
39-41
- normalised local heights
87-89,90-93
Northcott's finiteness theorem
16-17,44,46,53
- quantitative form (Schanuel's theorem)
. 17-19,132
Number of integral points of bounded height
- on an affine variety . . . . . . . .
177-178
6,134-136,163,177
- on a thin set . . . . . . . . . . .
Number of rational points of bounded height
- on an algebraic variety
178
.2,17-19
- on P n
- on a curve
5-6,80
53-55
- on an abelian variety
26-28
- on P 2 blown up at a point
5-6,132-134,178
- on a thin set . . . . . .
.44
Number of torsion points of bounded degree on an abelian variety.
Order of a ring of integers .
- conductor of an order
p-adic methods . . . . . .
- Chabauty's theorem .
P(n), the greatest prime factor of n
Partition of unity
...... .
Pell-Fermat equation x 2 - Dy2 = 1
Picard group Pic(X)
- PicO(X)
Poincare divisor class

Poincare upper half plane
Pontrjagin dual
Prime number theorem
Principal homogeneous space
Product formula, for a family of absolute values
191
191
60,67
58-60
105-106
. . . 84
4,94,136
2,20-22,34-36,157
25,31
37,76,78
67-68,71
164
181
100
. 7-10,22
217
- function fields
- number fields
Profinite group . .
- Frattini subgroup
Puiseux expansion
Pure transcendental extension
Quadratic functors on projective varieties
Quadraticity of heights on abelian varieties (see also Height)
- on elliptic curves . . . . . . . . . . . .
Quasi-integral set of points (see also Integral point) . . . .
7-8
7-10,113
147-149
148-149
130-131,198
147,154,159
. . . 34
35-39,42
40-41
. 4-6,94
Rank of Mordell-Weil group (see also Mordell-Weil theorem) 3,52,63,66,69,80,152,154-162

- is effectively bounded . . . . . .
. . . 52
- Honda's conjecture . . . . . . .
. . 162
- elliptic curves over Q of large rank
154-162
Regular field extension
137
Restriction of scalars functor
. 128
Riemann Hypothesis
188
Riemann-Roch theorem . .
74,109
Riemann's existence theorem
. 141
S-integers . . . .
102-105,112,134-136,141
S-units . . . . .
. . . . 103,112,
Schanuel's theorem
17-19,27-28,132,178
Selberg's bound, for the large sieve
171-172
Sextic resolvent . . . . . . . .
. . . . . . 125
Shih's theorem . . . . . . . . .
. . . . 146-147
Siegel's theorem (see also Integral points)
4-5,94-95,101-105,116,134,136,163
105-106
- application to P(J(n . . . . . .
- exceptional curve (see Exceptional curve)
- application to curves of genus ~ 1
101-102
- the case of curves of genus 0 . . . . .
102-104
- effectivity (see also Effectivity)
5,100-101,106-107,116
- approximation theorem on an abelian variety
. . . . . 98-101
- Thue-Siegel-Roth theorem . . . . . . .
4,95-100,104-105,107
Sieve, the large - (see also Large sieve inequality)
5-6,163-176
Sigma function of Weierstrass . . . . . . . .
. . . . . . 91
Silverman's theorem . . . . . . . . . . . .
. . . . . 154
Specialisation of Galois groups (see also Galois groups)
122-126,137-138
Specialisation of Mordell-Weil groups, Neron's theorem
59,152-154
Superelliptic curve . . . . .
117
Stark's conjectures . . . . .
199
Strongly continuous functions
. 83
Swan's example
147
Sylow subgroup
124,147-148
e, theta divisor
. . . . 74
218
T", translation on an abelian variety

Tate module of an abelian variety
Tate curve . . . . . . . . .
Tate's recipe for local heights
Tate's normalisation lemma
Tauberian theorem . . . . .
Theta functions (and local heights)
Thue-Siegel-Roth theorem . . . .
- approximation of real numbers
- for function fields . . . . . .
Thue's equation . . . . . . . . .
Thin set of points (see also Hilbert's irreducibility theorem)
- oftype 1 . . . . . . .
- oftype 2 . . . . . . . . . .
- extension of ground field . . .
- intersected with plane sections
- polynomial interpretation .
- in Ph upper bounds
- in P n, upper bounds . . .
- Cohen's theorem
- reduction modulo p of a thin set
Torsion subgroup, of the Mordell-Wei! group
- number of torsion points . . . . . .
Torsion points of bounded degree on an abelian variety
. 38
70,119
. 91
91-93
29-30
44,183
88-89,91
4,95-100,104-105,107
95-97
105
114
121
121
121
128-129
127-128
122
5-6,132-136
.6,177-178
.6,177-178
179-180,183-186
43-44,53-54,68
44,69
. . . . 44
Weak Mordell-Weil theorem (see also Mordell-Weil theorem)

- via Galois cohomology . .
- explicit form . . . . . . . . . .
- non-abelian generalisation
Weierstrass embedding of an elliptic curve
3,51-52,99
51-52
55-57
. 52
40,57,90,115,119,155
X(N) .
Xo(N) .
X}(N) .
118,193
67-70
119-120
Yo(N) .
. . . 67
eK( s), zeta function of a number field K
182,188

Mordell Weil

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Mordell Weil

Enviado por

Direitos autorais:

Formatos disponíveis

Aspects of Mathematics

Edited by Klas Diederich

G. Hector/U. Hirsch: Introduction to the Geometry of Foliations, Part B

P. Stiller: Automorphic Forms and the Picard Number of on Elliptic Surface

G. Faltings/G. Wustholz et 01.: Rational Points*

A. Howard/P.-M. Wong (Eds.): Contribution to Several Complex Variables

F. Hirzebruch, Th. Berger, R. Jung: Manifolds and Modular Forms*

H. Fujimoto: Value Distribution Theory of the Gauss Mop of Minimal

V. Anosov/A. A. Bolibruch: The Riemann-Hilbert Problem

Vol. E 23: A. P. Fordy/J.

C. Wood (Eds.): Harmonic Mops and Integrable Systems

D: S. Alexander: A History of Complex Dynamics

H. Skoda/J.-M. Trepreau IEds.): Contributions to Complex Analysis and

D. N. Akhiezer: lie Group Actions in Complex Analysis

R. Gerard, H. Tahara: Singular Nonlinear Partial Differential Equations

R.-P. Holzapfel: Boll and Surface Arithmetics

R. Huber: Etale Cohomology of Rigid AnalytiC Varieties and Adic Spaces

* A Publication af the Max-Planck-Institut fUr Mathematik,

Springer Fachmedien Wiesbaden GmbH

Prof. lean-Pierre Serre

All rights reserved

No part of this publication may be reproduced, stored in a retrieval

Cover design: Wolfgang Nieger, Wiesbaden

ISBN 978-3-663-10632-6 (eBook)

This is a translation of "Autour du theoreme de Mordell-Well", a course

4. The Mordell-Well theorem.

6. Local calculation of normalised heights.

9. Hilbert's irreducibility theorem.

10. Construction of Galois extensions.

11. Construction of elliptic curves of large rank.

12. The large sieve.

13. Applications of the large sieve to thin sets.

Appendix: The class number 1 problem and integral points on

2 has only finitely many ra-

In his thesis (1928), Weil [WI] considered a curve of genus 9 ~ 1 over a

Theorem. (Mordell-Weil.) The group J(K) is finitely generated.

For simplicity, let us assume that K is Q. Let P n be projective n-space

P=(Xo, ... ,Xn),

> rank A(Q),

where Hom(X, A) denotes the group of morphisms from X to A which send

[( d) Finally, Faltings proved 11ordell's conjecture: if X is a curve of genus

For this method, the fundamental case is 9 = 0 with 3 points at infinity.

Out of a total of eN integral points on the affine line Al with H :::; N,

0(1) if g? 1, or if g = 0 with ? 3 points at infinity

with 1 point at infinity

Hilbert's theorem can be stated for several variables: if X -+ P n IS a

Theorem. The number of integral points of An of height:::; N which lift

As z2 - <I> is absolutely irreducible modulo p, for almost all p, a theorem

where v is a real valuation on K (always a discrete valuation in practice)

We then say that K is equipped with a product formula.

Examples. a. K = Q. MQ = PU{oo} where P is the set of prime numbers,

II pJ,lp(x)(usual absolute value of x)

b. Let V be a projective normal k-variety, K its function field, Mv the set

I E K* has only a finite number of zeros and poles, we have

Let C be a curve on V not intersecting the singularities of V (as V

L AWVW(l) = 0 for all I

where the residue field Lw of w is the completion of L at w. (If L/ K is

(In the separable case this becomes simply

Then for all x E L * we have

Formula (1) implies that

1: we have a product formula for

Denote by Xw the image of x E L * in w. On taking the norm from

whence (1). It is clear that (1)

If v is associated to a real embedding K