INTRODUCTION TO ZETA AND L-FUNCTIONS FROM

ARITHMETIC GEOMETRY AND SOME APPLICATIONS

[INTRODUC

AO
`
AS FUNC

OES L E ZETA DA GEOMETRIA
ARITM

ETICA E ALGUMAS APLICAC

OES]
MARC HINDRY (UNIVERSIT

E DENIS DIDEROT PARIS 7)

Abstract. Zeta or L-functions are modelled on the Riemanns zeta function
originally dened by the series (s) =
P
n1
n
s
and then extended to the
whole complex plane. The zeta function has an Euler product, a functional
equation and though very much studied still keeps secret many of its proper-
ties, the greatest mystery being the so-called Riemann Hypothesis. Many
similar (or thought to be similar) series
P
n1
a
n
n
s
have been introduced in
arithmetic, algebraic geometry and even topology, dynamics (we wont discuss
the latter). We plan basically to discuss zeta functions attached to algebraic
varieties over nite elds and global elds.
The rst applications of zeta functions have been the arithmetic progession
theorem (Dirichlet, 1837) there exists one (hence innitely) prime congruent
to a modulo b, whenever a and b are coprime and the prime number the-
orem (Riemann 1859, with an incomplete proof; Hadamard and de la Vallee
Poussin, 1896) the number of primes less than x is asymptotic to x/ log x.
But further applications were not restricted to the study of prime numbers,
they include the study of the ring of algebraic integers, class eld theory,
the estimation of the size of solutions of (some) diophantine equations, etc.
Moreover L-functions have provided or suggested fundamental links between
algebraic varieties (motives over Q), Galois representations, modular or au-
tomorphic forms; for example, though they do not appear explicitly in Wiles
work, it seems fair to say they played an important r ole in the theory that
nally led to the solution of Shimura-Taniyama-Weil conjecture and thus of
Fermats Last Theorem.
The rst four lectures develop results and denitions which though all clas-
sical are perhaps not too often gathered together. The rst lecture intro-
duces Riemanns zeta function, Dirichlet L-function associated to a character,
Dedekind zeta functions and describes some applications of zeta functions; the
second introduces the Hasse-Weil zeta functions associated to algebraic vari-
eties dened over a nite eld, a number eld or a function eld as well as
L-functions associated to Galois representations and modular forms; the third
reviews techniques from complex analysis and estimates for zeta functions; the
fourth touches the theory of special values of zeta functions, some known like
the class number formula and some conjectured like the Birch and Swinnerton-
Dyer formula. The fth and nal lecture is an exposition of recent work of
Boris Kunyavski, Micha Tsfasman, Alexei Zykin, Amlcar Pacheco and the
author around versions and analogues of the Brauer-Siegel theorem.
Prerequisite will be kept minimal whenever possible : a course in complex
variable and algebraic number theory, a bit of Galois theory plus some exposure
to algebraic geometry should suce.
1
2 M. HINDRY
Mini-curso, XXI Escola de

Algebra, Braslia, julho 2010
Nota bene : Notes are given in English but the mini-curso will be given in Por-
tuguese. Also these sketchy notes will be completed later and posted on my web
page. [Added August 2010 : Hopefully this is done in the present version]
Nota bene : Estas notas estao escritas em ingles mas o mini-curso sera dado em
portugues. Ela contem as ideias essenciais e seu esboco sera completado mais tarde
e colocado na pagina web deste autor. [Esperamos que esteja feito nesta presente
versao]
Contents
1. Lecture I : Introduction and examples 3
1.1. Riemanns zeta function (and the prime number theorem) 3
1.2. Dirichlets L-functions (and the arithmetic progression theorem) 6
1.3. Some problems from arithmetic 8
1.4. Analogies between number elds and function elds 10
2. Lecture II : The zeta functions from algebraic geometry 12
2.1. The L-function associated to a Galois representation 12
2.2. The zeta function of a scheme over Z 15
2.3. The Weil zeta function of a variety over a nite eld 16
2.4. The Hasse-Weil L-functions 18
2.5. The L-function associated to a modular form 20
3. Lecture III : Some techniques and estimates from complex analysis 24
3.1. Classical lemmas 24
3.2. Tauberian theorems 26
3.3. Estimates for zeta functions 27
4. Lecture IV : Special values of zeta functions 30
4.1. The residue of Dedekind zeta function 32
4.2. Class number formulas 32
4.3. The Birch and Swinnerton-Dyer conjecture 33
5. Lecture V : Brauer-Siegel type theorems (and conjectures) 36
5.1. Brauer-Siegel theorem for number elds 36
5.2. Brauer-Siegel theorem for abelian varieties 39
6. Commented bibliography 42
References 42
AND L-FUNCTIONS 3
1. Lecture I : Introduction and examples
1.1. Riemanns zeta function (and the prime number theorem). The cen-
tral object of the theory is the Riemann zeta function, dened for 's > 1 by the
series
(1.1) (s) =

n=1
n
s
.
A fundamental property is its factorisation, known as the Euler product formula,
which can be viewed as an analytic presentation of the unique factorisation theorem.
Theorem 1.1. (Euler product formula) When 's > 1, we have :
(1.2) (s) =

n=1
n
s
=

p
(1 p
s
)
1
.
This can be used to dene the main branch of log (s) or compute

(s) and 1/(s)

as follows
(1.3)
log (s) =

m1,p
p
ms
m
and

(s) =

m1,p
(log p)p
ms
=

n1
(n)n
s
.
Here the Mangoldt function is implicitely dened as
(n) =
_
log p if n = p
m
0 else.
One can also compute
1
(s)
=

n=1
(n)n
s
=

p
(1 p
s
),
where the Moebius function is dened as
(n) =
_

_
1 for n = 1
(1)
k
if n = p
1
. . . p
k
0 else.
It is easy to extend a bit this function say to 's > 0, for example via the formula
(1.4) (s) =

n=1
n
s
=
1
s 1
+ 1 +s
_

1
([t] t)t
s1
dt,
which clearly displays a (simple) pole at s = 1 with residue 1. This formula is a
special case of the following general (and easy) lemma.
Lemma 1.2. Let a
n
be a sequence of complex numbers with a
n
= O(n
c
) for some
constant c. Put A(t) :=

nt
a
n
then, for 's > c + 1:
(1.5)

n=1
a
n
n
s
= s
_

1
A(t)t
s1
dt.
But actually much more is true; if you have never seen the denition of the
function see the beginning of lecture III.
4 M. HINDRY
Theorem 1.3. (Functional Equation) The function (s) extends to a function on
the whole complex plane, holomorphic except for a simple pole at s = 1 with residue
1. Further it satises the following functional equation. Let (s) :=
s/2
(s/2)(s)
then, away from 0 and 1, the function (s) is bounded in any vertical strip and sat-
ises :
(1.6) (s) = (1 s).
Corollary 1.4. The function (s) does not vanish in the half-plane 's > 1;
inside the half-plane 's < 0 it has only simple zeroes at even negative integers
2, 4, 6, . . . . All the other zeroes are inside the critical strip 0 's 1.
Proof. (Sketch) The proof relies on harmonic analysis or Fourier analysis. Denote

f(x) :=
_
R
f(x) exp(2ixy)dx, the Fourier transform of an integrable function f.
Start with the Poisson formula:
(1.7)

nZ
f(n) =

nZ

f(n).
Dene (u) :=

nZ
exp
_
un
2
_
, then, applying Poisson formula to f(x) =
exp(ux
2
) whose Fourier transform is

f(y) = exp(y
2
/u)/

u, one gets the

functional equation for the theta function (this can be interpreted as being a
modular form of half integer weight, see section 2.5).
(1.8) (1/u) =

u(u).
One then computes (using the change of variables t = n
2
u) for '(s) > 1.
(s) =
s/2
(s/2)(s) =

n1
_

0
e
t
t
s/2

s/2
n
s
dt
t
=
_

0
_
_
_

n1
exp(un
2
)
_
_
_
u
s/2
du
u
=
_

0

(u)
u
s/2
du
u
where

(u) :=

n1
exp
_
un
2
_
=
(u) 1
2

Notice

(u) = O(exp(u), when u goes to innity; plugging in (1.8) one gets
(1.9)

_
1
u
_
=

u

(u) +
1
2
_
u 1
_
.
Since
_

1
t
s
= 1/(s 1) and using (1.9), we get
(1.10)
(s) =
_
1
0

(u)
u
s/2
du
u
+
_

1

(u)
u
s/2
du
u
=
_

1

(1/u)
u
s/2
du
u
+
_

1

(u)
u
s/2
du
u
=
_

1
_

(u) +
1
2
_
u 1
_
_
u
s/2
du
u
+
_

1

(u)
u
s/2
du
u
=
_

1

(u)
_
u
s
2
+u
1s
2
_
du
u
+
1
s 1

1
s

AND L-FUNCTIONS 5
The latter expression is a priori valid for '(s) > 1, but it is easy to see that, since

(u) = O(exp(u), the function is entire, it is also clearly symmetric with respect
to the transformation s 1 s. Finally the function dened by the integral in the
last line is bounded in every vertical strip.
The zeroes inside the critical strip have two symmetries s s and s 1 s
this perhaps may suggest the following.
Conjecture 1.5. (Riemann Hypothesis) All the zeroes inside the critical strip lie
on the line 's =
1
2
.
In this direction we have the following weak but essential and sucient to
prove the prime number theorem!
Theorem 1.6. (Hadamard de la Vallee Poussin) The function (s) does not
vanish on the line '(s) = 1.
Proof. Start with the trivial but useful inequality 3 + 4 cos(t) + cos(2t) = 2(1 +
cos t)
2
0. Using equation 1.3 we may write (for > 1):
log () +4 log [( +it)[ +log [( +2it)[ =

p,m
p
m
m
'(3 +4p
mit
+p
2mit
) 0.
Taking exponentials on both side, we obtain :
(1.11) ()
3
[( +it)[
4
[( +i2t)[ 1.
Call h (resp. k) the order of (s) at s = +it (resp. at +2it), then the inequality
(1.11) reads, as tends to 1
1 ()
3
[( +it)[
4
[( +i2t)[ c( 1)
k+4h3
with c > 0, hence k +4h 3 0. Since h, k are non negative integers we conclude
that h = 0.
Thus the non trivial zeroes lie inside the critical strip; one can extend the HdVP
argument to show that ( + i) ,= 0 in regions like > 1
c
1
log
and this is
essentially the best know result.
Theorem 1.7. (Prime Number Theorem) Let (x) := #p prime [ p x then
(1.12) (x)
x
log x
.
Equivalently if (x) =

nx
(n), then (x) x.
The basic idea is to use a formula of Perron (see proposition 3.10), picking c > 1
and denoting
_
(c)
for the integral on the vertical line 's = c :
(1.13) (x) =

nx
(n) =
1
2i
_
(c)

(s)x
s
ds.
The function inside the integral has a pole at s = 1 with residue x, it has no other
pole in an open neighbourhood of 's 1 (thanks to theorem 1.6!), thus one can
move the line of integration to a contour slightly to the left of 's = 1, picking a
residu x at s = 1 and making it plausible that the contour integral is o(x) though
there remains work to be done before fully proving that (see lecture III, section 3.2,
for other approaches, each requiring the use of theorem 1.6).
6 M. HINDRY
Notice, that [(s)[ (c) for 's c > 1; using the functional equation and the
fact that (s) is bounded in vertical strips one gets easily a bound for (s) or (s)
of the shape O(exp(c[s[
1+
)) (away from the poles) which means that the function
is of order one in Hadamards terminology and thus we deduce :
Theorem 1.8. (Hadamard product) The function (s) :=
s/2
(s/2)(s) can be
written as a product
(1.14) s(1 s)(s) = e
a+bs

_
1
s

_
e
s

,
where runs over the zeroes in the critical strip.
We summarise the properties that all zeta functions L(s) will (or should) enjoy:
(1) be dened by a Dirichlet series L(s) =

n=1
a
n
n
s
(convergent in some
half-plane).
(2) be equal to an Euler product L(s) =

p
L
p
(s) with L
p
(s) = 1+

m1
a
p
mp
ms
;
in fact they will be of the shape
L
p
(s) =
d

j=1
(1
p,j
p
s
)
1
,
where d is called the degree of the Euler product and [
p,j
[ p
w/2
with
equality for almost all p and w is called the weight (the Riemann zeta
function has degree one and weight zero).
(3) have analytic continuation to the whole complex plane with a functional
equation (s) = (w + 1 s) where (s) = Q
s
L

(s)L(s) is bounded
in every vertical strip; here Q is a positive real number and L

(s) is the
product of Gamma functions (as +b).
(4) be a function of order one and therefore have Hadamard factorisation.
(5) satisfy the analogue of the Riemann hypothesis.
1.2. Dirichlets L-functions (and the arithmetic progression theorem).
Theorem 1.9. Let a, b be coprime integers (i.e. gcd(a, b) = 1) then there exists
(innitely many) primes p congruent to a modulo b. Moreover primes are equidis-
tributed in congruence classes in the sense that
(1.15) (x; a, b) := #p x [ p a mod b
1
(b)
(x).
The proof is based on the use of the following L-functions. First dene Dirichlet
characters as follows. Let : (Z/NZ)

be a homomorphism from the group

of invertible elements mod N to C

, we extend it to a map denoted again from

Z to C by
(n) :=
_
(n mod N) if gcd(n, N) = 1
0 if gcd(n, N) > 1.
It is convenient to introduce primitive characters as those who do not come from
a smaller level M dividing N, i.e. for all M dividing strictly N, there is an n 1
mod M such that (n) ,= 0, 1. The trivial character will be denoted
0
.
Then dene Dirichlets L-function :
(1.16) L(, s) :=

n=1
(n)n
s
.
AND L-FUNCTIONS 7
When is non trivial, the series is convergent for 's > 0 (absolutely convergent
for 's > 1); it has an Euler product
(1.17) L(, s) =

p
_
1 (p)p
s
_
1
.
When =
0
the trivial character modulo N, we get L(
0
, s) =

p | N
(1p
s
)(s)
and thus a pole at s = 1 but when is non trivial we notice that [

nx
(n)[
N (because

x<nx+N
(n) = 0) hence the series

(n)n
s
is convergent for
's > 0 (using lemma 1.2). From the Euler product, one may infer formulas for the
logarithm and logarithmic derivative :
(1.18)
log L(, s) =

p,m
(p
m
)
p
ms
m
and
L

(, s)
L(, s)
=

p,m
(p
m
)p
ms
log p.
The function L(, s) has analytic continuation to the whole complex plane with
a functional equation such that if is primitive modulo N and
(, s) =
_
N

_
s/2

_
s +
2
_
L(, s),
with = 0 if (1) = 1 and = 1 if (1) = 1 then
(1.19) (, s) = w()( , 1 s),
for a complex number w() with modulus 1 which can be computed as
w() =
G()
i

N
,
where G() =

x mod N
(x) exp(2ix/N) is the classical Gauss sum.
The key result is the following non vanishing statement.
Theorem 1.10. Let be a non trivial character, then L(, 1) ,= 0.
The proof will be sketched later (section 3.1, using lemma 3.3).
To deduce the Arithmetic Progression Theorem from this, we notice the easy
algebraic formula, where the sum is over characters modulo N

(a)(x) =
_
(N) if x a mod N
0 else
and proceed to write for 's > 1 and s approaching 1:

pa mod N
p
s
=
1
(N)

p
p
s
+
1
(N)

=
0
(a)

p
(p)p
s
=
1
(N)
log(s 1)
1
+
1
(N)

=
0
(a) log L(, s) +O(1).
The non vanishing of L(, 1) ensures that the second term remains bounded as s
goes to 1 and thus the sum on the left has to be innite.
8 M. HINDRY
1.3. Some problems from arithmetic. We already discussed prime number dis-
tribution. Lets us now consider other type of problems where zeta and L-function
may be used to advantage.
Problem 1. The classical approach to Fermats Last Theorem led Kummer to
introduce the cyclotomic elds and rings
K = Q(exp(2i/N) and O
K
= Z[exp(2i/N)].
It quickly became obvious that it was important to settle the question:
(Q1) For which value of N is the ring O
K
= Z[exp(2i/N)] a unique factorisa-
tion domain?
Later, realising the answer to the rst question is too often negative, Kummer
introduced the following renement: Let C
K
be the quotient of non zero ideals by
principal ideals; this turns out to be a nite group whose cardinality is called the
class number of K and denoted h
K
.
(Q2) For which primes p, is it true that p does not divide h
K
for K=Z
_
exp
_
2i
p
__
?
A (partial) answer can be given using zeta and L-functions.
Problem 2. Consider rings of quadratic integers if d ,= 0, 1 is squarefee put
K = K
d
= Q(

d) and O
K
= O
d
= Z[
1+

d
2
] or Z[

d] according to wether d 1
mod 4 or not. One would like to know when O
d
is a unique factorisation domain,
study units, etc. [When d < 0 the group of units is nite; when d > 0, modulo 1
the group of units is innite cyclic and we want to study the size of the generator
> 1 called the fundamental unit].
A (partial) answer can be given using zeta and L-functions.
Problem 3. Let us introduce a bit more of algebraic number theory. Let K/Q
be a nite extension, every rational prime p decomposes in K as a product of prime
ideals pO
K
= P
e
1
1
. . . P
e
g
g
; the prime p is said to be ramied in K/Q if some e
i
2
and if f
i
= [O
K
/P
i
: F
p
] then

g
i=1
e
i
f
i
= [K : Q]. Suppose now that K/Q is a
Galois extension with Galois group G.
Denition 1.11. The decomposition group of P/p is the subgroup
D(P/p) := G [ (P) = P.
When is in the decomposition group one may dene : O
K
/P O
K
/P by
the diagram
O
K

O
K

O
K
/P

O
K
/P.
Denition 1.12. The kernel of the map from G to Gal ((O
K
/P)/F
p
) is
called the inertia group of P/p and denoted I(P/p).
Notice that if #D(P/p) = f and #I(P/p) = e then #G = [K : Q] = efg. The
inertia group I(P/p) is trivial unless P/p is ramied. The map from G to
Gal ((O
K
/P)/F
p
) is known to be surjective and a canonical generator of the cyclic
group on the right is given by the map x x
p
.
AND L-FUNCTIONS 9
Denition 1.13. The Frobenius at Pis the element Frob
P
such that for all x O
K
we have
Frob
P
(x) x
p
mod P.
Notice that, in the ramied case, the Frobenius is actually a coset modulo the
inertia group. Further if we pick another prime above p, say P

= (P) then
Frob
P
= Frob
P

1
thus Frob
P
depends only on p up to conjugation; we will
denote Frob
p
this conjugacy class.
Let L/Q be a nite Galois extension with Galois group G, let C be a conjugacy
class in G, does there exist (innitely many) primes such that Frob
p
C?
The answer can be given using zeta and L-functions.
Theorem 1.14. (Tchebotarev) Let L/Q be a nite Galois extension with Galois
group G, let C be a conjugacy class in G, then
(1.20) #p x [ Frob
p
C
[C[
[G[
(x).
Notice that this theorem is a vast generalisation of Dirichlets theorem. Indeed
if we choose K = Q() with = exp(2i/N) then O
K
= Z[] and the Galois group
is abelian G = Gal(K/Q) = (Z/NZ)

. The Frobenius at p is simply given by

Frob
p
() =
p
(for p not dividing N), thus Tchebotarev theorem in this case is
equivalent to Dirichlet theorem.
Problem 4. Consider the following diophantine equations :
(1) (Pell-Fermat equation) x
2
dy
2
= 1 where d is a squarefree integer and
(x, y) Z
2
.
(2) (units in radical cubic eld) x
3
+ dy
3
+ d
2
z
3
3dxyz = 1 where d is a
squarefree integer and (x, y, z) Z
3
.
(3) (elliptic curves) y
2
= x
3
+ax+b where a, b are integers such that 4a
3
+27b
2
,=
0 and (x, y) Q
2
.
Algebro-geometric consideration will reveal that in these three cases the set of
solutions form a nitely generated group (of rank r = 1 in the rst two cases and
rank r 0 in the third case); the main question is what can we say about the
(minimal) size of generators of this group?
Again a (partial) answer can be given via zeta functions (wait for the last lecture
though!).
A rst natural generalisation of Riemanns zeta function is Dedekind zeta func-
tion associated to a number eld K (i.e. [K : Q] = d < ). Recall that the norm
of a non zero ideal 1 of O
K
is dened as
N(1) := #O
K
/1.
Recall also that the embeddings of elds : K C can be divided into r
1
real
embeddings : K R and r
2
pairs of complex embeddings , : K C.
If
1
, . . . ,
d
is a Z-basis of O
K
and Tr = Tr
K
Q
denotes the sum of the d
conjugates of , we dene the (absolute value of the) discriminant:
(1.21)
K
:= [det(Tr(
i

j
)[ .
The sign of the discriminant may be an interesting issue but since well have no
use for it, well neglect it. An important property is that p is ramied in K/Q if
10 M. HINDRY
and only if p divides
K
. This permits the following important denition of the
Dedekind zeta function of a number eld K:
(1.22)
K
(s) :=

I
N(1)
s
.
where the sum runs over non zero ideals of O
K
.
This zeta function has an Euler product (it has degree d and weight zero):
(1.23)
K
(s) :=

p
_
1 N(p)
s
_
1
,
where the product runs over non zero prime ideals of O
K
(i.e. maximal ideals).
This formula is an analytic version of the theorem of unique factorisation of ideals
into product of prime ideals for Dedekind domains. One easily infers that
log
K
(s) =

p,m
N(p)
ms
m
and

K
(s) =

p,m
N(p)
ms
log N(p).
Theorem 1.15. (Hecke) The Dedekind zeta function has analytic continuation to
the whole complex plane (except for a single simple pole at s = 1) and satises a
functional equation. Let

K
(s) :=
_

K
2
r
1

d/2
_
s

_
s
2
_
r
1
(s)
r
2

K
(s),
then

K
(s) =
K
(1 s).
It will be convenient to introduce the following notation.
Notation 1.16. We introduce the modied Gamma functions as follows.
(1.24)
R
(s) :=
s/2

_
s
2
_
et
C
(s) := (2)
s
(s) .
Notice that we may rewrite the function
K
(s) from the functional equation as :

K
(s) :=
s/2
K

R
(s)
r
1

C
(s)
r
2

K
(s).
1.4. Analogies between number elds and function elds. The rst idea is
to compare R = Z and K[T] (with K a eld); both rings are principal domains
The analogy becomes more arithmetic if K = F
q
because then quotients R/I and
K[T]/I are nite (for a non zero ideal).
Absolute values on Z are of two types : the absolute values attached to a prime
ideal (or prime number) p and normalised for example by [x[
p
= p
ord
p
x
and the
archimedean or usual absolute value [x[

= [x[. On K[T] the absolute value are

apparently of two types, those associated to a prime ideal (or irreducible polyno-
mial) P and normalised by say [x[
P
= e
ord
P
x
, and the one given by the degree
say [x[

= e
deg x
; in both cases we have the product formula, for all non zero x in
the eld of fractions:
(1.25)

v
[x[
v
= 1.
AND L-FUNCTIONS 11
For a function eld, this is a restatement of the fact that the sum of multiplicities of
zeroes minus the sum of multiplicities of poles of a function on a smooth complete
curve is zero. For K = Q the formula follows from an essentially trivial verication.
But notice that, in the case R = K[T], the innite absolute value can be
interpreted as associated to a point at innity, introducing the projective line P
1
=
A
1
and viewing K[T] as the ring of regular functions on the ane line A
1
,
then deg x = log [x[

is equal to ord

x. We will denote M
K
the set of places of
K (i.e. the set of closed points of C or, if one prefers, the set of Galois conjugacy
classes of points in C(

F
p
)).
When K = F
q
(C) and v M
K
, we put q
v
:= q
deg v
and dene :

K
(s) =

v
(1 q
s
v
)
1
.
Then we have the following theorem
Theorem 1.17. (Artin, Schmidt, Weil) Let K = F
q
(C) with a curve of genus g.
The function
K
(s) satises the following.
(1) The function is a rational fraction in q
s
ie.
K
(s) = Z(C, q
s
) with
Z(C, T) =
L(C, T)
(1 T)(1 qT)
=

2g
j=1
(1
j
T)
(1 T)(1 qT)
.
(2) It satises the functional equation

K
(1 s) = q
(2g2)(
1
2
s)

K
(s).
(3) (Riemanns hypothesis) [
j
[ =

q.
The name Riemann hypothesis is coined because
K
(s) = 0 implies 1
j
p
s
=
0 and hence 's =
1
2
. Therefore
K
(s) has a simple pole at s = 1 with residue
lim
s1
(s 1)
K
(s) =
h
K
p
g
(1 1/p) log p
,
where h
K
= L(C, 1) =

2g
1
(1
j
) is the class number #Pic
0
(F
p
), the number
of F
p
points on the Jacobian of C. Notice that h
K
= p
g
+ O(p
g1/2
) thus 1
lim
s1
(s 1)
K
(s) 1.
Perhaps one of the most useful tools when studying an algebraic variety X dened
over K = F
p
(C) is the possibility of spreading X into a variety A dened over F
p
and bered over the complete curve : : A C such that the generic ber is
(isomorphic to) X. When the variety is dened over Q, one can spread it out over
Z but the curve Z is not complete and one should try to add bres at innity, this
is one of the main purposes of Arakelov theory.
12 M. HINDRY
2. Lecture II : The zeta functions from algebraic geometry
2.1. The L-function associated to a Galois representation. A central object
in arithmetic or arithmetic geometry is the absolute Galois group
(2.1) G
Q
:= Gal(

Q/Q).
It is natural to study this huge pronite group via its representations. Let V
be a (complex) vector space of dimension n and a Galois representation (all
representations are assumed to be continuous):
: G
Q
GL
n
(C) = GL(V ).
Remark 2.1. The following observations will be applied to the Frobenius elements
and the inertia subgroup (whose denition is recalled in 1.13). If f = hgh
1
G
then the characteristic polynomials of (f) and (g) acting on V are equal. If H is
a subgroup of G, denote
V
H
:= v V [ h H, (h)(v) = v,
the subspace of xed vectors. If f gH and g centralises H (i.e. for every h H,
we have gh = hg) then g and f leave V
H
stable and the characteristic polynomials
of (f) and (g) acting on V
H
are equal.
This allows the denition of Artin L-function associated to a represention as :
(2.2) L(, s) =

p
det
_
1 (Frob
p
)p
s
[ V
I
p
_
1
.
Denote =

= Tr the character of the representation; it is well known that

the character determines the representation thus we may write
L(

, s) := L(, s).
When the representation is one dimensional, it is essentially abelian (it factors via
the abelian group G/ Ker ) and it is then known via Artin reciprocity law that the
L-function is a generalised Dirichlet L-function (also called Hecke L-function) and
satises analytic continuation plus functional equation. The general case is more
dicult but can be attacked through Brauers theorem.
We need to slightly extends denitions and introduce induction
1
of representa-
tions.
Denition 2.2. Let G = Gal(K/k) be the Galois group of an extension of number
elds and let : G GL(V ) be a complex representation. The Artin L-function
associated to (or equivalently to its character

) is
(2.3) L(s, ) = L(s, , K/k) =

p
det
_
1 (Frob
p
)N(p)
s
[ V
I
p
_
1
,
(where the product runs over maximal ideals of O
k
).
The basic functoriality properties are
(1) L(s,
1

2
) = L(s,
1
)L(s,
2
).
1
The quickest way to dene induction of nite dimensional representations of nite groups is
via the tensor product of group algebras : if H G and : H GL(V ) is a representation
dened over K, it corresponds to a K[H]-module structure on V ; the representation Ind
G
H
is
associated to the K[G]-module W = V
K[H]
K[G].
AND L-FUNCTIONS 13
(2) If H G is a subgroup and : H GL(V ) a representation, by Galois
theory H = Gal(K/K
H
), and we have :
L(s, , K/K
H
) = L(s, Ind
G
H
, K/k).
The following is a purely group theoretical statement.
Theorem 2.3. (Brauer) Let G be a nite group and the character of a complex
representation of G then there are subgroups H
i
, one dimensional representations
of H
i
with characters
i
and integers n
i
Z such that
(2.4) =

i
n
i
Ind
G
H
i

i
.
Putting together this algebraic result and the known properties of abelian L-
functions one obtains
Theorem 2.4. Let G = Gal(K/k) be the Galois group of an extension of number
elds and let : G GL(V ) be a complex representation. The Artin L-function
associated to extends to a meromorphic function on the complex plane and sat-
ises a functional equation. Let n = dimV and c Gal(

Q/Q) be the complex

conjugation, denote n
+
= dimV
+
and n

= dimV

, where V
+
(resp. V

) is the
eigen space corresponding to +1 (resp. 1) for (c).
(, s) := N
s/2

R
(s)
n
+

R
(s + 1)
n

L(, s).
Then
(2.5) (, s) = w

( , 1 s),
where [w

[ = 1 and is the contragredient representation

2
.
Artin conjectured that L(, s) is entire except for a pole at s = 1 with order the
multiplicity of the trivial representation inside .
Notice that for topological reasons any continuous representation : Gal(

Q/Q)
GL
n
(C) factors through a nite group G = Gal(K/Q); there are very interesting
Galois representations for which this is no longer true; the -adic representations
are such examples. Well study now the concrete example of the Tate module of an
elliptic curve.
Recall (or learn) that the eld of -adic numbers Q

can be constructed as the

completion of the eld Q with respect to the absolute value [x[

:=
ord

x
. The
ring of integers Z

= x Q

[ [x[

1 is then the completion of Z. Alternatively

one may dene Z

as the inverse limit of the nite groups Z/

n
Z (with obvious
homomorphisms):
Z

= lim

n
Z/
n
Z.
The eld Q

can then be viewed as the eld of fractions of Z

.
To simplify the introduction we assume in the following ad hoc denition that
the characteristic of the ground eld is not 2 or 3.
Denition 2.5. An elliptic curve over a eld K is a plane projective curve which
can be dened in P
2
by an equation
(2.6) Y
2
Z = X
3
+aXZ
2
+bZ
3
2
If : G GL(V ) and V

is the dual of V , then : G GL(V

) is given by (g)(v), v

=
v, (g
1
)(v

); we have

(g) =

(g
1
) =

(g).
14 M. HINDRY
where a, b are in K and 4a
3
+ 27b
2
,= 0 (setting x = X/Z and y = Y/Z, one often
writes the equation in the ane plane Z ,= 0 as y
2
= x
3
+ax +b). The group law
is dened as follows:
(1) The origin is the point at innity (0 : 1 : 0).
(2) The inverse (or symmetrical) of (X : Y : Z) is (X : Y, Z).
(3) To add two points P, Q draw the line joining them (if P = Q this means
draw the tangent at P to the curve) pick the third intersection point of the
line with the curve, then P +Q is the symmetrical point.
Remark 2.6. Changing (ane) coordinates (x

, y

) = (
2
x,
3
y) will give a new
equation y
2
= x
3
+a

+b

with (a

, b

) = (
4
a,
6
b) and

=
12
. Thus if a, b
are rational, after a rescaling, we may assume that a, b are integers and minimal
in the sense that for all primes either p
4
does not divide a or p
6
does not divide b.
We will often tacitly assume the equation is chosen to be minimal and will call the
corresponding discriminant minimal. (Notice the denition should be amended to
incorporate the primes 2 and 3).
Such an elliptic curve is a projective variety and an algebraic group (a group
whose addition law can be expressed in terms of polynomials) and is the rst ex-
ample of abelian variety. Like for any algebraic group, there are natural maps such
as translations, dened by t
Q
(P) = P + Q, and multiplication by n, dened (for
positive n) by [n](P) = P + +P.
Denition 2.7. Let X be an elliptic curve dened over K of characteristic 0 or p
dierent from . The kernel of multiplication by
n
is denoted
X[
n
] := Ker
_
[
n
] : X(

K) X(

K)
_
,
and is isomorphic as a group to (Z/
n
Z)
2
. The Tate module is the inverse limit
T

(X) := lim

X[
n
]

=
_
lim

Z/
n
Z
_
2

= Z
2

.
The Galois group G
K
:= Gal(

K/K) acts on each X[
n
] and hence on T

(X),
providing a representation:

X,
: G
K
GL(T

(X))

= GL
2
(Z

).
Let K be a number eld. Though the representation is dened over Q

it is
known that the characteristic polynomial of Frob
p
has rational integer coecients
which further do not depend on ; in fact det
X,
(Frob
p
) = p and Tr
X,
(Frob
p
) =
a
p
= p + 1 #X(F
p
). This important fact makes the following denition possible,
where we write V for the Q

vector space T

(X)
Z

.
Denition 2.8. The L-function associated to the set of representations
X,
is:
(2.7) L(
X
, s) =

p
det
_
1
X,
(Frob
p
)N(p)
s
[ V
I
p
_
1
.
Let us specialise a bit further to an elliptic curve X dened over Q. The elliptic
curve has good reduction modulo p whenever p does not divide the discriminant
=
X
and then we dene a
p
= a
p
(X) = p + 1 #X(F
p
). When p divides the
discriminant, the curve X modulo p has either a cusp (in which case dimV
I
= 0)
we set a
p
= 0 or a node with a pair of tangents (in which case dimV
I
= 1) which
AND L-FUNCTIONS 15
can be dened over F
p
, in which case we set a
p
= 1, or over a quadratic extension
of F
p
, in which case we set a
p
= 1. We have then the more explicit expression:
(2.8) L(
X
, s) =

p |
(1 a
p
p
s
)
1

p |
(1 a
p
p
s
+p
12s
)
1
.
There is a ner invariant called the conductor of the elliptic curve (or the asso-
ciated representation) which can be computed up to a factor 2
a
3
b
by the following
formula.
N
X
=

p
p
m
X,p
,
where, if we dene
n
X,p
= codim
V
V
I
=
_

_
0 if p does not divide
X
1 if p divides
X
and a
p
= 1
2 if p divides
X
and a
p
= 0,
we have m
X,p
n
X,p
with equality except when p divides the discriminant, a
p
= 0
and p = 2 or 3. Notice that the discriminant and conductor have the same prime
factors but with dierent exponents.
2.2. The zeta function of a scheme over Z. Let R be a nitely generated ring,
that is R = Z[t
1
, . . . , t
n
]/I then for all maximal ideal m the quotient R/m is a
nite eld of cardinality say N(m); in view of the denition of the Dedekind zeta
function, it is natural to introduce the zeta function of the ring R:
(2.9)
R
(s) =

m
_
1 N(m)
s
_
1
,
where the product is taken over all maximal ideals of R. In view of the local deni-
tion of schemes, this denition immediately extends to schemes of nite presentation
over Z, replacing maximal ideals by closed points.
Denition 2.9. Let A be a scheme
3
of nite presentation over Z and denote [A[
its set of closed points, then
(2.10) (A, s) :=

x|X|
_
1 N(x)
s
_
1
,
where N(x) is the cardinality of the residual eld at x.
If dimA = d, one can prove easily that the product and associated series con-
verges for 's > d. Formally (A
1
. A
2
, s) = (A
1
, s)(A
2
, s). Thus decomposing
the set of closed points according to their residual characteristic we get the Euler
product decomposition where we denote A
p
the bre above p:
(2.11) (A, s) =

p
(A
p
, s).
Examples 2.10. If A = Spec(Z) (resp. A = Spec(O
K
)), then (A, s) is just
Riemanns zeta function (resp. Dedekind zeta function for the eld K).
If A = A
1
Z
= Spec(Z[T]), we can identify closed points (maximal ideals) with
residual characteristic p with monic irreducible polynomials in F
p
[T], whose set
3
A scheme can be taken as a patching of ane schemes Spec(R
i
) and for each R
i
we have

R
i
(s) = (Spec(R
i
), s).
16 M. HINDRY
we denote Irr
p
; we also denote M
p
the set of monic polynomials. The following
computation is then straightforward :
(A
1
Z
, s) =

QIrr
p
(1 p
s deg Q
)
1
=

QIrr
p

m=0
p
smdeg Q
=

PM
p
p
s deg P
=

d=0
p
dds
=

p
(1 p
1s
)
1
= (s 1).
Decomposing the closed points of P
1
Z
as A
1
Z
. A
0
Z
we obtain
(P
1
Z
, s) = (s)(s 1).
It is clear that we should start by studying the zeta function of a variety dened
over F
p
.
2.3. The Weil zeta function of a variety over a nite eld. Let X be a
smooth projective variety dened over F
p
; for a closed point x we denote d
x
=
[F
p
(x) : F
p
]. We compute
log
X
(s) =

x,m
N(x)
ms
m
=

x,m
p
d
x
ms
m
=

n1
p
ns
_

md
x
=n
1
m
_
(say)
=

n1
u
n
p
ns
.
We complete the computation by observing that, for a closed point x [X[ the
fact of having residual degree d
x
dividing n is equivalent to corresponding to a
(conjugacy class of) point in X(F
p
n), thus :
u
n
=

x|X|, d
x
=
n
m
1
m
=
1
n

x|X|, d
x
|n
d
x
=
1
n
#X(F
p
n).
We have thus shown the following.
Proposition 2.11. Let X be a variety over F
p
, then we have the formula
(2.12) (X, s) = Z(X, p
s
),
where
(2.13) Z(X, T) = exp
_

m=1
#X(F
p
m)
m
T
m
_
.
The latter formal series is known as Weils zeta function attached to X/F
p
; it
has been much studied and is the subject of the famous Weil conjectures (solved
by Grothendieck and Deligne).
AND L-FUNCTIONS 17
Example 2.12. Let us compute the (easy) example of X = P
n
. In this case
#X(F
p
m) =
p
m(n+1)
1
p
m
1
= p
mn
+p
m(n1)
+ +p
m
+ 1, thus

m=1
#X(F
p
m)
m
T
m
=
n

j=0

m=1
p
mj
m
T
m
=
n

j=0
log(1 p
j
T),
thus:
(2.14) Z(P
n
, T) =
1
(1 T)(1 pT) . . . (1 p
n1
T)(1 p
n
T)
.
Notice that a simple verication will reveal that :
Z(P
n
, T) = (1)
n+1
p
n(n+1)
2
T
n+1
Z
_
P
n
,
1
p
n
T
_
.
The pattern is actually quite general :
Theorem 2.13. (Weils conjectures) Let X be a smooth projective variety over F
p
of dimension n.
(1) (Rationality) There are polynomial P
j
(X, T) =

b
j
i=1
(1
j,i
T) Z[T] for
j = 0, . . . , 2n such that
(2.15) Z(X, T) =
P
1
(X, T) . . . P
2n1
(X, T)
P
0
(X, T) . . . P
2n
(X, T)
=
2n

j=0
P
j
(X, T)
(1)
j+1
.
Further b
0
= b
2n
= 1 and P
0
(X, T) = 1 T and P
2n
(X, T) = 1 p
n
T.
(2) (Functional equation) Let (X) =

2n
j=0
(1)
j
b
j
be the Euler-Poincare
characteristic then
(2.16) Z(X, T) = p
n(X)
2
T
(X)
Z
_
X,
1
p
n
T
_
.
(3) (Riemann hypothesis) The algebraic integers
j,i
satisfy [
j,i
[ = p
j/2
.
(4) The numbers b
j
= b
j
(X) satisfy continuity in smooth families and in par-
ticular if X is the reduction modulo a prime ideal of a variety Y dened
over a number eld K, then the b
j
(X) are equal to the Betti numbers of the
complex variety Y
K
C.
Example. Let X be an elliptic curve over F
p
. We have b
0
= b
2
= 1, b
1
= 2 and
P
0
(T) = 1 T, P
2
(T) = 1 pT and P
1
(T) = 1 aT +pT
2
. There is an algebraic
integer with = p such that
Z(X, T) =
(1 T)(1 T)
(1 T)(1 pT)
.
This is equivalent to Hasses theorem which says that
#X(F
p
m) = p + 1
m

m
.
The proof of this theorem is way beyond the scope of these notes. Suces to
quote the existence of a cohomology theory
4
, which well denote H
j
(X) which is
functorial, hence the Frobenius x x
p
induced a map f = f
j
: H
j
(X) H
j
(X),
and is such that the following Lefschetz trace formula holds:
4
For the expert we are writing H
j
(X) = H
j
et
(X

F
p
, Q

), the -adic etale cohomology.

18 M. HINDRY
(2.17)
#X(F
p
m) = #x xed by Frobenius power m =
2n

j=0
(1)
j
Tr(f
m
[ H
j
(X)).
A formal computation, based on the elementary formula
exp
_

m=1
Tr(f
m
[ V )
T
m
m
_
= det(1 fT [ V )
1
,
then gives
(2.18) Z(X, T) =
2n

j=0
det
_
1 fT [ H
j
(X)
_
(1)
j+1
,
and we just have to dene P
j
(X, T) = det(1fT [ H
j
(X)) to obtain the rst part.
The functional equation follows formally from the Poincare duality: a canonical
non degenerate pairing
(2.19) H
j
(X) H
2nj
(X) H
2n
(X) = Q

,
such that f
j
x, f
2nj
y) = p
n
x, y). Indeed an elementary argument shows that the
existence of such a pairing implies
det(1 f
j
T [ H
j
(X)) = (1)
B
j
p
nB
j
T
B
j
det(f
j
[ H
j
(X))
det
_
1
f
2nj
p
n
T
[ H
2nj
(X)
_
.
The Riemann Hypothesis lies deeper; again the name comes from the easy con-
clusion that it implies that zeroes (resp. poles) of
X
(s) lies on vertical lines
's = m +
1
2
, with 0 m dimX 1 (resp. on vertical lines 's = m, with
0 m dimX).
2.4. The Hasse-Weil L-functions. Let X be again a smooth projective variety of
dimension n dened over Q. When M = H
i
(X) one uses the Hodge decomposition:
H
i
(X(C), C) =
p+q=i
H
p,q
,
together with the Frobenius at innity, i.e. the map F

induced by the complex

conjugation on X(C) to dene numbers:
(2.20) h
p,q
= dim
C
H
p,q
and h
p,
= dim
C
H
p,
where H
p,
:= c H
p,p
[ F

(c) = (1)
p
c. We can nally dene the Gamma
factor as
(2.21)
L

(M, s):=
_

p<q,p+q=i

C
(s p)
h
p,q
if i is odd

p<q,p+q=i

C
(s p)
h
p,q

R
(s
i
2
)
h
i
2
,+

R
(s + 1
i
2
)
h
i
2
,
if i is even.
When X is smooth modulo p, it is clear (or should be . . . ) what the correct
Euler factor for the L and zeta function of X must be. For a model A of X over Z
AND L-FUNCTIONS 19
we have
X
(s) =

X
p
(s); to ease notation we write X
p
for the reduction modulo
p; we have
X
p
(s) = Z(X
p
, p
s
) and from the Weil conjecture we know that
Z(X
p
, p
s
) =
2n

j=0
det(1 Frob
p
p
s
[ H
j
(X
p
))
(1)
j+1
=
2n

j=0
det(1 Frob
p
p
s
[ H
j
(X))
(1)
j+1
,
the last equality being a property of continuity (or smooth base change) of the
cohomology.
We thus naturally dene

p
(X, s) := (X
p
, p
s
) and L
p
(H
j
(X), s) := det(1 Frob
p
p
s
[ H
j
(X))
1
.
The cohomological interpretation is needed to provide the correct factor at the bad
primes, namely
L
p
(H
j
(X), s) := det(1 Frob
p
p
s
[ H
j
(X)
I
p
)
1
.
Denition 2.14. Let X be a smooth projective variety of dimension n dened
over Q; the Hasse-Weil zeta and L-functions are dened by:
(2.22) L(H
j
(X), s) :=

p
L
p
(H
j
(X), s) :=

p
det(1 Frob
p
p
s
[ H
j
(X)
I
p
)
1
,
and
(2.23) (X, s) =
2n

j=0
L(H
j
(X), s)
(1)
j
.
One may now state the following conjecture (known in many interesting cases
but widely open).
Conjecture 2.15. Let X be a smooth projective variety dened over Q. The
function L(H
j
(X), s) =

p
L
p
(H
j
(X), s) extends analytically to the complex plane
except, when j is even, for a pole at s = 1 + j/2. Further there exists an integer
N, the conductor of the Galois representation on H
j
(X), such that if we dene:
(2.24) (H
j
(X), s) := N
s/2
L

(H
j
(X), s)L(H
j
(X), s),
then we have the functional equation
(2.25) (H
j
(X), s) = (H
j
(X), j + 1 s)
For example, if X is an elliptic curve dened over Q, we write L(X, s) instead
of L(H
1
(X), s); then we have
(X, s) = N
s/2
X

C
(s)L(X, s) and (X, s) = (X, 2 s),
and the conjecture is in this case a theorem thanks to the work of Wiles et al.
Notice that the Hasse-Weil zeta function can be written

X
(s) =
(s)(s 1)
L(X, s)
.
20 M. HINDRY
Remark 2.16. All this can be formulated, mutadis mutandis, over function elds,
replacing the number eld K by F
q
(C) and prime ideals by places of the function
eld. For example, if X is a (smooth projective) variety over K = F
q
(C), the
L-functions associated can be dened as
L(H
j
(X), s) =

v
det(1 Frob
v
q
s
v
[ H
j
(X)
I
v
)
1
,
where q
v
= q
deg v
. The theory is far more advanced over function elds than
over number elds; for example analytic continuation and functional equation are
known, in fact Grothendieck theory shows that the L-function is actually a ra-
tional fraction in q
s
and the functional equation reads L(H
j
(X), j + 1 s) =
q
d
j
(s
j+1
2
)
L(H
j
(X), s), where d
j
is an integer which can be interpreted as the de-
gree of a conductor. Finally the (analogue of) Riemann hypothesis is a theorem in
the function eld case : the zeroes of L(H
j
(X), s) all have real part equal to
j+1
2
;
via Deligne-Grothendieck this is equivalent to the statement that the numerator
has the form

j
(1
j
q
s
) with [
j
[ = q
j+1
2
.
2.5. The L-function associated to a modular form.
Denition 2.17. Poincares half-plane H := z C [ (z) > 0 ; its compacti-
cation H

:= H P
1
(Q).
The group GL
+
2
(R) of matrices 22 with positive determinant or SL
2
(R) acts on
H via
__
a b
c d
_
, z
_
(az +b)/(cz +d). The groups GL
+
2
(Q) and SL(2, Z) act on
H

. The latter action is discrete thus one may form the quotients Y := SL(2, Z)H
and X := SL(2, Z)H

, which are Riemann surfaces. In fact, Y

= A
1
(C) and
X

= P
1
(C).
Among subgroups of nite index in SL(2, Z) the most important are
Denition 2.18. A subgroup SL(2, Z) is a congruence subgroup if it contains
(N) for some N, where
(N) :=
_
A =
_
a b
c d
_
SL(2, Z) [ A I mod N
_
.
We denote Y (N) := (N)H and X(N) := (N)H

.
Apart from (N) itself, other examples include :
(1) The congruence subgroup

1
(N) :=
_
A =
_
a b
c d
_
SL(2, Z) [ A
_
1
0 1
_
mod N
_
.
We denote Y
1
(N) :=
1
(N)H and X
1
(N) :=
1
(N)H

;
(2) The congruence subgroup

0
(N) :=
_
A =
_
a b
c d
_
SL(2, Z) [ c 0 mod N
_
.
We denote Y
0
(N) :=
0
(N)H et X
0
(N) :=
0
(N)H

.
The subgroup
1
(N) is normal in
0
(N) and
0
(N)/
1
(N)

= (Z/NZ)

, via the
map
_
a b
c d
_
d mod N.
AND L-FUNCTIONS 21
One knows Y
0
(N) (resp. Y
1
(N), Y (N)) are algebraic ane curves whereas
X
0
(N) (resp. X
1
(N), X(N)) are projective algebraic curves. Further, X
0
(N)
and X
1
(N) are dened over Q, whereas X(N) is dened over Q(exp(2i/N)).
Denition 2.19. Let be a congruence subgroup. A modular form of weight k
with respect to , is a holomorphic function f : H C such that :
(1) For
_
a b
c d
_
and z H, we have
(2.26) f
_
az +b
cz +d
_
= (cz +d)
k
f(z);
(2) The function f is holomorphic on H

, i.e. for all

_
a b
c d
_
SL(2, Z), the
limit of f
_
az+b
cz+d
_
(cz + d)
k
, when z goes to innity, exists. If this limit
is zero f is said to be parabolic.
Write M
k
() for the space of modular forms for with weight k and S
k
() for the
subspace of parabolic forms.
Any congruence subgroup contains some T
h
:=
_
1 h
0 1
_
, with h non zero
minimal : for example, T
1

1
(N). Hence if f M
k
() then f(z + h) = f(z), so
we can write the Fourier expansion :
(2.27) f(z) =

nZ
a
n
q
n
h
, where q
h
:= exp
_
2iz
h
_
.
If f is holomorphic on H

then a
n
= 0 for n < 0, while vanishing at reads a
n
= 0
for n 0 (nota bene : the full condition for f to be a modular form is holomorphy
at all points in P
1
(Q) = H

H).
Remark 2.20. If =
_
a b
c d
_
GL
2
(R)
+
, then := ad bc > 0 and, if we put

=
1/2
, well have

SL
2
(R) and

z = z. If f is modular with weight

k for

, well have
(2.28) f
_
az +b
cz +d
_
=
k/2
(cz +d)
k
f(z).
Denition 2.21. Let f S
k
(
0
(N)) and f(z) =

n=1
a
n
exp(2inz) =

n=1
a
n
q
n
its Fourier expansion. The Dirichlet series associated to f is :
(2.29) L(s, f) :=

n=1
a
n
n
s
.
Notice the relation :
(2.30)
C
(s)L(f, s) = (2)
s
(s)L(f, s) =
_

0
f(it)t
s1
dt.
Denition 2.22. Let f =

n
a
n
(f)q
n
M
k
(
0
(N)). The Hecke operators are
dened as follows.
22 M. HINDRY
(1) if p does not divide N, the operator f T
p
f is dened by :
a
n
(T
p
f) := a
np
(f) +p
k1
a
n/p
(f),
where, by convention, a
n/p
= 0 if p does not divide n.
(2) If p divides N, the operator f U
p
f is dened by :
a
n
(U
p
f) := a
np
(f).
Theorem 2.23. (Hecke, see [DiSh08]) The Hecke operators commute. If f =

n
a
n
(f)q
n
S
k
(
0
(N)) is an eigenvector for each operator, i.e T
p
f =
p
f,
U
p
f =
p
f, then a
p
(f) =
p
a
1
(f), if further f is normalised by the condition
a
1
(f) = 1, the L-function L(s, f) factorises into an Euler product of the shape :
(2.31)
L(s, f) =

n=1
a
n
(f)n
s
=

p | N
(1 a
p
(f)p
s
)
1

p | N
(1 a
p
(f)p
s
+p
k12s
)
1
.
Proof. (partial sketch) The commutation follows from a direct computation. We
do now computations for p not dividing N, leaving the (easier) computation for
p dividing N to the reader. Suppose that T
p
(f) =
p
f then we have
p
a
n
(f) =
a
np
(f) + p
k1
a
n/p
(f) hence a
p
(f) =
p
(f)a
1
(f) and, if a
1
(f) = 1 (which we now
suppose) then a
p
(f) =
p
(f). When p is coprime with n we infer a
np
= a
n
a
p
and
more generally a
np
r = a
n
a
p
r from which an Euler product decomposition follows.
Finally the recurrence relations
p
a
p
r = a
p
a
p
r = a
p
r+1 + p
k1
a
p
r1 are equivalent
to the formula

r
a
p
r T
r
= (1 a
p
T +p
k1
T
2
)
1
.

When k = 2, this L-function looks like the L-function associated to an elliptic

curve. Let us check that, under a further condition, L(f, s) satises the functional
equation expected for an elliptic curve. Notice the matrix W
N
:=
_
0 1
N 0
_
(which
does not belong to SL(2, Z) but to GL
+
2
(Q)) normalises the subgroup
0
(N), since
W
N
_
a b
c d
_
W
1
N
=
_
d c/N
bN a
_
.
Therefore W
N
acts on M
k
(
0
(N)) (resp. S
k
(
0
(N))), and since W
2
N
= NId, it
acts as an involution and the spaces M
2
(
0
(N)) (resp. S
2
(
0
(N))) decompose into
the sum of two eigenspaces such that :
(2.32) f
_

1
Nz
_
= f(W
N
z) = N
k/2
z
k
f(z)
Theorem 2.24. (Hecke) Let = 1 and f() =

n1
a
n
exp(2in) a parabolic
modular form for
0
(N) of weight k such that
(2.33) f
_

1
N
_
= N
k/2

k
f().
Put (s, f) := N
s/2
(2)
s
(s)L(s, f), where L(s, f) :=

n=1
a
n
n
s
. The function
(s, f) has analytic continuation to C and satises a functional equation
(2.34) (s, f) = i
k
(k s, f).
Further, (s, f) is bounded in every vertical strip.
AND L-FUNCTIONS 23
Proof. Notice that for = it (with t R
+
) the equation (2.33) reads
f
_
i
Nt
_
= (i)
k
N
k/2
t
k
f(it).
The analogy with equation (1.8) used to prove the functional equation of the Rie-
mann zeta function is now evident. We may compute, invoking (2.30) and using a
change of variable t 1/Nt :
(2.35)
(s, f) = N
s/2
_

0
f(it)t
s1
dt
= N
s/2
_ 1

N
0
f(it)t
s1
dt +N
s/2
_

1

N
f(it)t
s1
dt
= N
s/2
_

1

N
f(i/Nt)t
s1
dt +N
s/2
_

1

N
f(it)t
s1
dt
= i
k
N
1
2
(ks)
_

1

N
f(it)t
k1s
dt +N
s/2
_

1

N
f(it)t
s1
dt
=
_

1

N
f(it)
_
i
k
N
1
2
(ks)
t
ks
dt +N
s/2
t
s
_
dt
t
.
The latter expression denes an entire function (one needs the elementary fact
that [a
n
[ = O(n
c
), and hence [f(it)[ = O(exp(2t)) as t goes to innity). The
(i
k
)-symmetry when s is replaced by ks is now obvious, as the property of being
bounded in any vertical strip.
We close this chapter stressing the marvelous result of Wiles which identies
L-functions associated to Galois representations (the Tate module), Hasse-Weil L-
function of elliptic curves over Q and L-functions of weight two modular forms with
rational coecients. More precisely, if X is an elliptic curve dened over Q, with
conductor N = N
X
and L(X, s) =

n1
a
n
n
s
is its Hasse-Weil function as dened
in (2.8), then Wiles theorem asserts that the function f(z) :=

n1
a
n
exp(2inz)
is a modular form with respect to the group
0
(N); in fact this result gives a
correspondence between (isogeny classses of) elliptic curves of conductor N dened
over Q and parabolic modular forms with respect to
0
(N) which are eigen forms
for all Hecke operators, have rational Fourier coecients and do not come from
modular forms with respect to some
0
(M) with M strictly dividing N. This
actually can be viewed as one instance of the very general program of Langlands,
part of which says that Hasse-Weil functions should all be automorphic.
24 M. HINDRY
3. Lecture III : Some techniques and estimates from complex analysis
3.1. Classical lemmas. We begin with the Stirling formula for the Gamma func-
tion.
The Gamma function. It is rst dened for 's > 0 by the integral
(3.1) (s) :=
_

0
e
t
t
s
dt
t
.
Integration par part yields the formula s(s) = (s + 1) and hence analytic con-
tinuation to C with simple poles at non positive integers 0, 1, 2, 3, . . . .
Lemma 3.1. (Stirling formula) When s stays away from negative reals, we have:
log (s) =
_
s
1
2
_
log s s +
1
2
log(2) +O
_
1
[s[
_
.
The following corollary of Stirlings formula will be useful: away from poles and
uniformly in any vertical strip we have, as [[ tends to innity:
(3.2) [( +i)[ c() exp(

2
[[)[[

1
2
.
Dirichlet series.
Lemma 3.2. Suppose that the series L(s) =

n=1
a
n
n
s
converges for s = s
0
,
then it converges uniformly in domains of the shape s C [ 's c > 's
0

or s C [ [s s
0
[ C'(s s
0
). Consequently any Dirichlet series L(s) has
an abscissa of convergence
c
such it diverges for 's <
c
and converges on the
half-plane 's >
c
to a holomorphic function.
Lemma 3.3. (Landau) Suppose a
n
0 and that the series L(s) =

n=1
a
n
n
s
converges for 's >
0
and the function has analytic continuation to a neighbourhood
of
0
, then the abscissa of convergence is strictly smaller than
0
.
Application. We can now give the promised proof of the non vanishing L(, 1) ,= 0
(theorem 1.10) which is used in the proof of the arithmetic progression theorem
(theorem 1.9). In fact the key is Landaus lemma plus the observation that, if
K = Q(exp(2i/N)) then we have the factorisation
(3.3)
K
(s) = (s)

=
0
L(, s),
where the product is over primitive non trivial characters modulo N. Now if some
L(, 1) = 0, the product on the right is an entire function, thus, by Landaus
lemma, the Dirichlet series dening
K
(s) would be convergent for all s, but this is
clearly false.
Holomorphic functions
Lemma 3.4. (Cauchys formula) Let f(s) be holomorphic inside the circle ( of
center z and radius R, then

f
(n)
(z)
n!

1
2i
_
C
f(s)
(s z)
n+1
dz

max
|sz|=R
[f(s)[/R
n
.
AND L-FUNCTIONS 25
This can be viewed as stating that a bound valid for f remains essentially valid
for its derivatives.
Lemma 3.5. (Phragmen-Lindelof ) Let f(s) be holomorphic in a strip a '(s) b
with reasonable growth (i.e. for example [f(s)[ exp([s[
c
)).
(1) Suppose sup
tR
([f(a +it)[, [f(b +it)[) = M then [f(s)[ M in the whole
strip.
(2) Suppose [f(a + it)[ M
a
(1 + [t[)

and [f(b + it)[ M

b
(1 + [t[)

then
[f( +it)[ M

(1 +[t[)
v()
, where M

= M
u()
a
M
1u()
b
, v() = u() +
(1 u()) and u is the ane function with values u(a) = 1 and u(b) = 0.
Lemma 3.6. (Borel-Caratheodory) Let f(s) be holomorphic for [s[ R; assume
max
|s|=R
'f(s) = A then, for 0 < r < R we have
max
|s|r
[f(s)[
2Ar
R r
+
R +r
R r
[f(0)[.
Lemma 3.7. (Hadamards three circles lemma) Let f(s) be holomorphic in an
annulus R
1
[s[ R
2
then the function M(r) := sup
|s|=r
[f(s)[ is log convex in
log r, that is
log M(r)
log(R
2
/r)
log(R
2
/R
1
)
log M(R
1
) +
log(r/R
1
)
log(R
2
/R
1
)
log M(R
2
).
Lemma 3.8. (Jensens formula) Let f be holomorphic in a disk D(0, R) with no
zero at s = 0 or on the circle [s[ = R, denote z
1
, . . . , z
n
its zeroes and n(r) :=
#[z[ < r [ f(z) = 0, then :
1
2
_
2
0
log

f(Re
i

d log [f(0[ = log

R
n
[z
1
[ . . . [z
n
[
=
_
R
0
n(r)
dr
r
.
We recall the notation
_
(c)
f(s)ds = i
_
+

f(c +it)dt.
Lemma 3.9. (Perrons integral)
(3.4)
1
2i
_
(c)
y
s
ds
s(s + 1) . . . (s +h)
=
_
_
_
1
h!
_
1
1
y
_
h
if y 1
0 else.
This follows by moving the line of integration to the left (resp. to the right) if
y 1 (resp. if y < 1) picking up residues at s = k equal to
1
h!
_
h
k
_
(1)
k
x
k
(resp.
no residue).
Proposition 3.10. (Perrons Formula) Let L(s) =

n=1
a
n
n
s
be absolutely con-
vergent for 's = c then
(3.5)
1
2i
_
(c)
L(s)x
s+h
ds
s(s + 1) . . . (s +h)
=
1
h!

nx
a
n
(x n)
h
.
(For h = 0 one has to slightly modify the formulas so that when y = 1 (or x = n)
one gets a contribution 1/2 instead of 1).
26 M. HINDRY
3.2. Tauberian theorems. We give two examples of so-called tauberian theorems,
each one leading to a proof of the prime number theorem. It is relatively easy to
show that if

nx
a
n
= x +o(x) then lim
1
( 1)

n
a
n
n

= , this type of
summation results are called abelian theorems; the converse is not always true and
typically theorems asserting (under additional hypothesis) the reciprocal are called
tauberian theorems.
Theorem 3.11. (Ikehara) Let Z(s) =

n1
a
n
n
s
be a Dirichlet series convergent
for 's > 1 with a
n
0; suppose that Z(s)

s1
extends to a holomorphic function
on 's 1 then we have:
(3.6)

nx
a
n
= x +o(x).
Notice that the theorem applied to Z(s) =

(s)/(s) gives
(x) :=

nx
(n) x,
and thus immediately implies the prime number theorem, once we have Hadamard
de la Vallee Poussin result (theorem 1.6).
The next result, due to Newmann, has a similar but distinct avour.
Theorem 3.12. Let h(t) be a bounded locally integrable function then the integral
F(s) =
_
+
0
h(u)e
su
du
is convergent and holomorphic on the half-plane '(s) > 0. Suppose this function
has analytic continuation to an open neighbourhood of '(s) 0, then the integral
for s = 0 converges and
F(0) =
_
+
0
h(u)du.
Denote (t) :=

px
log p. We may apply the latter theorem to the function
F(s) =
_
+
1
(t) t
t
s+2
dt
=
_
+
0
(e
u
) e
u
e
u(s+2)
e
u
du =
_
+
0
_
(e
u
)e
u
1

e
us
du.
Indeed the function h(u) := (e
u
)e
u
1 is piecewise continuous. Checking holo-
morphic continuation well conclude that the integral
F(0) =
_
+
0
((e
u
)e
u
1)du =
_
+
1
(t) t
t
2
dt
is convergent. It is then elementary to conclude (x) x which is a form of the
prime number theorem.
AND L-FUNCTIONS 27
To check analytic continuation, apply the following transformations to F(s) (for
'(s) > 0) :
F(s) =
_
+
1
(t) t
t
s+2
dt
=

n=1
_
n+1
n
(t)t
s2
dt
_
+
1
t
s1
dt
=

n=1
(n)
n
s1
(n + 1)
s1
s + 1

1
s
=
1
s + 1

n=1
n
s1
((n) (n 1))
1
s
=
1
s + 1

p
p
s1
log(p)
1
s

On the other hand

(s)
(s)
=

p,m1
log(p)p
ms
=

p
log(p)p
s
+

p,m2
log(p)p
ms
.
The second term is holomorphic for '(s) > 1/2, hence

p
log(p)p
s
=

(s)
(s)
+
holomorphic function on '(s) > 1/2, and nally
F(s) =

(s + 1)
(s + 1)(s + 1)

1
s
+ holomorphic function on '(s) > 1/2.
Using Hadamard de la Vallee Poussin result (theorem 1.6) we may conclude.
3.3. Estimates for zeta functions. We now use the previous theory to give
estimates for zeta functions. Let us axiomatise the properties we will use (we
think of M as being H
w
(X) for some smooth projective variety): there is a weight
w = w(M), a Betti number b = b(M), Hodge numbers h
i
, h

i
, a conductor
N = N(M).
(1) (Euler product) We have L(M, s) =

p
L
p
(M, s) with
L
p
(M, s) =
b(M)

j=1
(1
p,j
p
s
)
1
,
such that [
p,j
[ p
w/2
with equality whenever p does not divide N =
N(M). Therefore the Euler product is absolutely convergent for 's >
1 +w/2.
(2) (Functional equation) Dene the completed L-function as:
(M, s) = N
s/2

R
(s +a
i
)
h
i

C
(s +b
i
)
h

i
L(M, s),
then (M, s) = (M, w + 1 s).
We will also study some consequences of the (generalised) Riemann Hypothesis
which states that the zeroes of (M, s) are located on the line 's = (w + 1)/2.
28 M. HINDRY
Lemma 3.13. Let > 1 +w/2 then
(3.7)
[L(M, +it)[ ( w/2)
b(M)
and [ log L(M, +it)[ b(M) log ( w/2).
Proof. We simply use the inequality [1 z[
1
(1 [z[)
1
(when [z[ < 1) to write
[L(M, +it)[ =

p
b(M)

j=1
[1
p,j
p
s
[
1

p
(1 p
w
2

)
b(M)
= (
w
2
)
b(M)
.
The second inequality is proven in the same fashion.
Using the functional equation and PhragmenLindelof, one gets
Corollary 3.14. (Convexity bound) Dene H =

i
h
i
+ 2

i
h

i
then
(3.8) [L(M, +it)[
_
N(1 +[t[)
H
_
()+
where is a convex function such that
() =
_
0 when 1 +
w
2
w+1
2
when
w
2
.
Proof. Indeed from the functional equation we get that
L(M, s) = N
w+1
2
s
L

(M, w + 1 s)
L

(M, s)
L(M, w + 1 s).
Now Stirling formula (3.2) implies that, as t goes to innity,
L

(M, w + 1 it)
L

(M, +it)
C[t[
H(
w+1
2
)
.
This gives the formula for () outside the critical strip; the convexity of () is a
restatement of PhragmenLindelof estimate.
Notice that when [
w
2
, 1 +
w
2
] then min0,
w+1
2
()
1
2
(1 +
w
2
)
so that for example
L
_
w + 1
2
+it
_
(N(1 +[t[)
H
)
1
4
+
and L
_
w + 1
2
_
N
1
4
+
.
If we allow the generalised Riemann hypothesis then using Borel-Caratheodory and
the three circles lemma one can show
Corollary 3.15. Assume further the L function satises the generalised Riemann
hypothesis then we may choose () = min0,
w+1
2
and in particular
(3.9) L
_
w + 1
2
+it
_
(N(1 +[t[)
H
)

and L
_
w + 1
2
_
N

.
Proof. Assuming the Riemann Hypothesis, the function g(s) = log L(M, s) is holo-
morphic on 's >
w+1
2
and 'g(s) = log [L(M, s) log(N(1 + [t[
H
). Using
Borel-Caratheodory lemma we obtain a bound of the shape [g(
w+1
2
+ + it)[

1
log(N(1 + [t[
H
). Applying Hadamards three circles lemma to the circles of
center
1
+it (where
1
is chosen very large) and radii
1

w
2
1 ,
1
and

w+1
2
, we will obtain a bound [g(
w+1
2
+2+it)[
1
_
log(N(1 +[t[
H
)
_
1
from which it follows that L(M, s) is O
_
(N(1 +[t[
H
))

_
when 's
w+1
2
.
AND L-FUNCTIONS 29
Applying Cauchys bounds, one gets similar estimates for the derivatives, i.e. for
example
1
r!
L
(r)
_
w+1
2
_

r,
N

. As explained in the last lecture, it would be very

interesting to show a similar lower bound for the rst non vanishing derivative, i.e.
if we let r be the order of vanishing of L(s) at s =
w+1
2
then does there exist a
lower bound of the shape:
1
r!
L
(r)
_
w + 1
2
_
N

?
This question is intimately connected with the distribution of zeroes near s =
w+1
2
as one can see for example on the Hadamard factorisation.
L(M, s) = e
As+B

_
1
s

_
e
s

.
In fact, if such an estimate is true, it implies, via Jensen formula (lemma 3.8)
applied to f(s) = L

(
w+1
2
+s) := L(
w+1
2
+s)s
r
with a small radius , that, if we
denote n(r) the number of zeroes in the disk of center
w+1
2
and radius r, we have
_

0
n(r)
dr
r
= o(log N). In particular this would imply that
0
the zero closest to
w+1
2
(but dierent) satises [
0

w+1
2
[ N

.
30 M. HINDRY
4. Lecture IV : Special values of zeta functions
Convention 4.1. By convention we will call the special value at s = k (in
practice k will be an integer or half an integer) of a meromorphic function L(s) the
leading term of its Taylor expansion; that is if L has a zero of order r at s = k
(resp. a pole of order r) we dene:
L

(k) := lim
sk
(s k)
r
L(s).
There is a wealth of fascinating formulas, the oldest being
(2) =

2
6
, (4) =

4
90
, more generally
2n
(2n) Q

.
In general the special value will have the shape:

(M, k) = (rational number) (period) (regulator).

In the previous example the period is
2n
and the regulator is trivial (i.e. equal to
1). The interest of special values was noted explicitely by Hecke whom we quote :
the precise knowledge of the behaviour of an analytic function in the neighbourhood
of its singular points is a source of number-theoretic theorems.
We will mainly be concerned by the size of regulators; the idea is simple : suppose
L

= Z
n
is a lattice in E = R
n
, then, in order to give bounds for a set of generators
of L it suces to have two estimates:
(1) An upper bound for the covolume vol(E/L);
(2) A lower bound for the norm of the smallest non zero vector in L.
Indeed one of Minkowski s theorem in geometry of numbers states that there is
a basis e
1
, . . . , e
n
of L such that
n

i=1
[e
i
[ c
n
vol(E/L)
(Notice that for any basis we have vol(E/L)

n
i=1
[e
i
[.)
The two examples of lattices of interest will be the following.
Lattice of units of a number eld. Let F be a number eld with r
1
real
embeddings
1
, . . . ,
r
1
and r
2
pairs of complex embeddings
j
,
j
(for 1 j
r
2
). Dene the logarithmic map
LOG : O

F
R
r
1
+r
2
,
by LOG(x) = (log [
1
(x)[, . . . , log [
r
1
(x)[, 2 log [
1
(x)[, . . . , 2 log [
r
2
(x)[). Then the
kernel is given by the roots of unity and the image is a discrete subgroup contained
in the hyperplane H with equation x
1
+ + x
r
1
+ y
1
+ + y
r
2
= 0. In fact
L := LOG(O

F
) is a lattice in H which means that the rank of O

F
is r := r
1
+r
2
1
(the fact that the rank is at most r
1
+r
2
1 is clear, the equality is a theorem due
to Dirichlet). The regulator is the covolume of this lattice which can be computed
by the explicit formula, where
1
, . . . ,
r
is basis of O

F
/ roots of unity :
(4.1) R
F
= [det (
i
log [
i
(
j
)[)[
AND L-FUNCTIONS 31
where
i
runs among all pairwise non conjugate embeddings minus one
5
and
i
= 1
(resp. = 2) for
i
real (resp. for
i
complex).
The question of bounding the size of generators of the group of units is thus
tantamount to bounding R
F
.
Mordell-Weil Lattice. Let A be an elliptic curve (or more generally an abelian
variety) dened over K = Q (or more generally over a global eld K). The Mordell-
Weil theorem states that A(K) is a nitely generated group. Further there is a
canonical height (also called Neron-Tate height
6
)

h : A(K) R, which provides an
intrinsic notion of size of a rational point and satises the following properties :
(1) It is a counting function, which means that the set x A(K) [

h(x) C
is nite for any C.
(2) It is a positive denite quadratic function, which means that
x, y) :=
1
2
(

h(x +y)

h(x)

h(y))
is bilinear and the induced quadratic map

h
R
: A(K) R R is positive
in the usual sense.
One can give a quick construction of the Neron-Tate height as follows.
Dene rst the (logarithmic) nave height of a point P P
n
(K) with projective
coordinates P = (x
0
, . . . , x
n
) by the formula
(4.2) h(P) :=

vM
K
log max [x
i
[
v
.
Notice that the denition is independent of the choice of projective coordinates,
thanks to the product formula (1.25). Choosing an embedding : A P
n
we
obtain a height h := h

: A(K) R dened by h

(P) := h((P)). The Neron-

Tate height is then equal to the limit :
(4.3)

h(P) =

h

(P) = lim
m
h(2
m
P)
4
m
.
Denition 4.2. Let P
1
, . . . , P
r
denote a Z-basis of A(K)/A(K)
tor
, the Neron-Tate
regulator is the determinant :
(4.4) Reg(A/K) := det (P
i
, P
j
))
1i,jr
.
The regulator is the square of the covolume of the lattice A(K)/A(K)
tor
in
A(K) R. The question of bounding the size of generators of the Mordell-Weil
group A(K) is thus tantamount to bounding Reg(A/K).
5
The denition seems to depend on the omitted embedding but the formula
P
r
1
i=1
log |
i
()| +
2
P
r
2
j=1
log |
j
()| = 0 is true for all unit and shows that the absolute value of the determinant
does not depend on this choice.
6
To be precise, the canonical height is in the general case a bilinear form (, ) : A(K)A

(K)
R where A

denotes the dual abelian variety; one then needs a polarisation : A A

(or
alternatively an embedding A P
n
) to dene the height

h

(x) =
1
2
(x, (x)). Since well deal
essentially with the example of elliptic curves this subtlety disappears.
32 M. HINDRY
4.1. The residue of Dedekind zeta function. The rst deep theorem about
special values we will mention is:
Theorem 4.3. The special value of the Dedekind zeta function at s = 1 is:
(4.5)

F
(1) = lim
s1
(s 1)
F
(s) =
h
F
R
F

2
r
1
(2)
r
2
w
F
.
Via the functional equation, this translates into the following slightly more ele-
gant formula : the function
F
(s) has a zero of order r
1
+ r
2
1, the rank of the
group of units, and :

F
(0) := lim
s0
s
r
1
r
2
+1

F
(s) =
h
F
R
F
w
F
.
Here all quantities have been dened except w
F
which is the number of roots of
unity inside F

; it can be viewed as the cardinality of O

F,torsion
.
4.2. Class number formulas. Let denote a non trivial Dirichlet character. It
is possible to give a nite expression for the series L(, 1) :=

n1
(n)
n
using the
following elementary formula. Put () :=

n1
exp(in)n
1
. Then, for ]0, 2[,
we have
() = log(2 sin(/2)) +i
_

2


2
_
.
For a primitive character ,=
0
one may deduce
(4.6) L(, 1) =
_
i
G()
N
2

N
a=1
(a)a if (1) = 1

G()
N

N
a=1
(a) log [1 exp(2a/N)[ if (1) = +1.
On the other hand, when K = Q(exp(2i/N), we have

K
(s) = (s)

L(, s),
where the product is over non trivial primitive Dirichlet characters modulo N. We
may deduce the rst form of the class number formula :
(4.7)
2
r
1
(2)
r
2
w
hR

=
0
L(, 1).
Let us specialise a bit to N = p a prime number. By writing the corresponding
formulas for K = Q(exp(2i/p) and K
+
:= KR = Q(cos(2i/p)) we then obtain
formulas
(4.8) h
K
= w
K
R
K
+
R
K

odd
_

1
2
B

_
where we set B

=
1
p

p
a=1
(a)a and recall that w
K
= 2p is the number of roots
of unity in K.
Notice that r
1
(K) = 0, r
2
(K) = [K : Q]/2 = (p 1)/2 whereas r
1
(K
+
) =
[K
+
: Q] = (p 1)/2 and r
2
(K
+
) = 0 thus O

K
and O

K
+
have the same rank and
the latter is a subgroup of nite index in the former. One can nally identify the
quotient
R
K
+
R
K
with the class number of K
+
and obtain the following class number
formula.
AND L-FUNCTIONS 33
Theorem 4.4. Let K = Q(exp(2i/p)) and K
+
:= K R = Q(cos(2i/p)) and
let h
p
and h
+
p
be their respective class number then
(1) One may factor h
p
= h
+
p
h

p
where h

p
is an integer.
(2) One has the formula
(4.9) h

p
= w
K

odd
_

1
2
B

_
.
One may in fact prove similar class number formulas for any abelian extensions
of Q. For a quadratic eld K with discriminant the corresponding basic formula
is
(4.10) Res(
K
, 1) =
_
2h
K
w
K

if K imaginary
2h
K
log

if K real,
where is the fundamental unit. On the other hand if K = Q(

D) we have a
decomposition
K
(s) = (s)L(
D
, s) and hence
(4.11) Res(
K
, 1) = L(
D
, 1),
where the character
D
can be described as follows. Any prime number p is either
decomposed or inert or ramied in K/Q, which means that pO
K
= p
1
p
1
or p or p
2
;
if we denote T, 1, the set of decomposed, inert or ramied primes we have
(4.12)

K
(s) =

pD
(1 p
s
)
2

pI
(1 p
2s
)
1

pR
(1 p
s
)
1
= (s)

pD
(1 p
s
)
1

pI
(1 +p
s
)
1
= (s)

p
(1
D
(p)p
s
)
1
.
Thus for p ,= 2 we have
D
(p) =
_
D
p
_
(the Legendre symbol); the fact that the
function
D
thus dened is a Dirichlet character is essentially equivalent to the
quadratic reciprocity law, the ancestor of Artin reciprocity law.
These formulas are basic for theoretical study and explicit computations of class
numbers.
4.3. The Birch and Swinnerton-Dyer conjecture. There are many results and
conjectures about special values of Hasse-Weil zeta and L-functions, let us mention
there are general conjectures by Lichtenbaum, Deligne, Beilinson and Bloch-Kato.
Ill only discuss here the mother of them all : the Birch and Swinnerton-Dyer
conjecture.
We rst need to dene more quantities; well do it for elliptic curves over K = Q
or F
p
(C) though it would not be much more dicult to do it for abelian varieties
over global elds. Let R := Spec(Z) or C, we have a so-called Neron model / R
with neutral section e : R /.
(1) We already introduced the Neron-Tate regulator which we denote Reg(A/K).
(2) The Tate-Shafarevich group is dened as a measure of a local-global coho-
mological obstruction (where v runs over all places of K):
X(A/K) := ker
_
H
1
(Gal(K
sep
/K), A
K
)

v
H
1
(Gal(K
sep
v
/K
v
), A
K
v
)
_
.
34 M. HINDRY
(3) TheTamagawa numbers is the product T (A/K) =

v
c
v
(A/K) where the
quantity c
v
(A/K) is the number of components dened over the residual
eld of the special ber above v of the Neron model of A/K.
(4) (When K = Q) The period is
A
=
_
A(R)
where is the Neron dierential.
(5) The height h(A/K) is the Faltings height in the case of number elds and
simply deg e

1
A/C
in the function eld case; we also introduce the expo-
nential height as H(A/K) = exp(h(A/K)) in the number eld case and
H(A/K) = p
h(A/K)
in the function eld case.
For convenience dene
J(A/K) :=
_

A/K
inthe number eldcase
H(A/K)
1
p
d(1g)
inthe functioneldcase,
then we may formulate
Conjecture 4.5 (The Birch & Swinnerton-Dyer conjecture). The following state-
ments holds true.
(1) The L-function L(A/K, s) is analytic at s = 1 and
rank A(K) = ord
s=1
L(A/K, s).
(2) The group X(A/K) is nite.
(3) The special value at s = 1 is given by
(4.13)
L

(A/K, 1) = lim
s1
L(A/K, s)
(s 1)
r
=
#X(A/K) Reg(A/K)
#A(K)
tor
#A

(K)
tor
T (A/K) J(A/K).
The conjecture can be viewed as a sophisticated version of the local global princi-
ple. One collects local informations (for elliptic curves this involves only the number
of points modulo p) and builds with them a global object (the L-function), using
analytical continuation beyond the half-plane of convergence, one hopes to gather
important global information encoded in the L-function such as the rank of the
Mordell-Weil group, the Neron-Tate regulator. Notice that for example the BSD
conjecture asserts that:
(1) The Mordell-Weil group is nite if and only if L(A/K, 1) ,= 0.
(2) (Parity conjecture) Let r = rank A(K) then the sign of the functional equa-
tion is (1)
r
.
Both of these corollaries have been checked on thousands of examples and it is
known, for modular elliptic curves, that if L(A/K, 1) ,= 0 (resp. if L(A/K, 1) = 0
and L

(A/K, 1) ,= 0) then r = 0 (resp. r = 1).

This marvelous conjecture is far from settled; the Tate-Shafarevich group is not
even known to be nite in general and analytic continuation until s = 1 of the
L-function is only known for elliptic curves over Q (Wiles) and modular or CM
abelian varieties (Shimura). Over Q if the order of the zero of L(A/Q, s) is 1,
the rst two items are true and the third is true up to a rational number which, in
a nite number of cases has been computed to be 1!
The situation is better over function elds where we know that rank A(K)
ord
s=1
L(A/K, s), and niteness of some -primary component of X(A/K) is equiv-
alent to the equality rank A(K) = ord
s=1
L(A/K, s). Further, when true, these
AND L-FUNCTIONS 35
equivalent statement imply the full conjecture; thus there is a large supply of abelian
varieties over function elds for which the full BSD conjecture is settled.
36 M. HINDRY
5. Lecture V : Brauer-Siegel type theorems (and conjectures)
5.1. Brauer-Siegel theorem for number elds.
Theorem 5.1 (Brauer-Siegel Theorem). Suppose that F varies over a family of
number elds of xed degree over Q. Denote by
F
the absolute value of the dis-
criminant of F/Q and suppose that
F
tends to innity, then :
(5.1) log(h
F
R
F
) log
_

1/2
F
_
,
where h
F
denotes the class number of F and R
F
the regulator of the group of units
of F.
Applications.
(1) For imaginary quadratic elds K = Q(

D) the theorem yields

D
1
2

h
D
D
1
2
+
,
displaying in an explicit way that the class number goes to innity quite
rapidly.
(2) For real quadratic elds K = Q(

D) the theorem yields

D
1
2

h
D
log
D
D
1
2
+
,
displaying in an explicit way that at least one of the two quantities (the
class number and the size of the fundamental unit) goes to innity quite
rapidly. It also provides a bound
(5.2) log
D
D
1
2
+
,
which is best possible for the Ds with say bounded h
D
(conjectured to be
innitely many).
(3) For any degree d, it is possible to construct sequences of elds F of degree
d with small regulator hence large class number, say h
F

1
2

F
. For
example take F = Q() with a root of P(T) =

d
j=1
(T ja) + b (for
well chosen a, b).
(4) In general we get bounds for both the regulator and class number
h
F

1
2
+
F
and R
F

1
2
+
F
.
The proof of the Brauer-Siegel starts with the formula for the residue of
F
(s)
and essentially boils down to show that

F
(1)

F
for elds of xed
degree [F : Q] = d. The upper bound is relatively easy and follows from an integral
representation similar to (1.10) of the shape:

F
(s) =

F
s(s 1)
+f
F
(s),
where
F
=

F
(1) is the residue of
F
(s) at s = 1 and f
F
(s) is such that for real s
we have f
F
(s) 0 and even
s/2
F
. Picking > 1 we obtain

F
( 1)
F
().
Elementary estimates give
F
() ()
d

_
1 +
1
1
_
d
and thus choosing =
1 + 1/ log
F
well get a bound of the shape
F

1
2
F
(log
F
)
d
, hence

F
(1) =
(2)
r
2

1
2
F

F
(log
F
)
d
.
AND L-FUNCTIONS 37
The lower bound is much harder. Granting the Riemann hypothesis or more
modestly granting the non existence of Siegel zeroes, i.e. zero of
F
(s) located on a
segment [1
c
log
, 1[ we can give a quick proof. Indeed selecting a point 0 < < 1
where
F
() < 0 (or equivalently
F
() < 0) we get

F
> (1 )f
F
() (1 )
1
2
F
.
Thus if we are allowed to choose = 11/ log
F
we get the required estimate. In
fact this proof requires non existence of zeroes of the zeta function on the segment
[1 c

, 1]; this is exactly what Siegel proved for quadratic elds (and was
extended by Brauer to all number elds, using theorem 2.3) with, unfortunately,
ineective (not computable) constants c

; we give below a sketch of proof of Siegels

result.
One can ask what happens if we relax the conditions and allow say [K : Q] to
go to innity; the question has been thoroughly explored by Tsfasman and Vladut .
To describe one of their main results we need a bit of notation.
Denition 5.2. For a number eld K and q = p
m
we denote
q
(K) the number
of places of K with norm equal to q. For a sequence of number elds K
i
we dene
the limits (when they exist):

q
:= lim
i

q
(K
i
)
log
_

K
i
;
R
:= lim
i
r
1
(K
i
)
log
_

K
i
and
C
:= lim
i
r
2
(K
i
)
log
_

K
i
.
We call asymptotically exact a family for which these limits all exists (this can
always be achieved by extracting a subsequence).
In the following a sequence K
i
of elds is almost normal if each K
i
is contained
or contains a normal eld K

i
with [K

i
: K
i
] or [K
i
: K

i
] bounded (the denition
given in [TsVl02] is actually more general) .
Theorem 5.3. (Tsfasman Vladut ) Let K
i
be an asymptotically exact family of
number elds and assume that either all K
i
/Q are almost normal or that GRH
holds, then
(5.3) lim
i
log (h
K
i
R
K
i
)
log
_

K
i
= 1 +

q
log
_
q
q 1
_

R
log 2
C
log 2.
Remark 5.4. If [K : Q] is bounded or even if [K : Q]/ log
K
goes to zero, we see
easily that
q
=
R
=
C
= 0 and we recover the original Brauer-Siegel theorem,
but there exists many interesting families of elds for which the limit in theorem
5.3 is not equal to 1.
5.1.1. Appendix: Siegels theorem according to Goldfeld. The proof of BrauerSiegel
theorem without using GRH is complicated. We give here Golfelds proof [Go74]
in the (essential) case of quadratic elds. It is based on the relatively elementary
(see below) fact that for a quadratic (real) character of conductor D, we have the
bounds D
1/2
[L(1, )[ log D; to obtain a better lower bound is more delicate
and is the true content of Siegels theorem.
38 M. HINDRY
Theorem 5.5. For all > 0 there exists C() > 0 such that for all real characters
of conductor D we have
(5.4) L(1, )
C()
D

.
As mentionned before, the proof requires the following two lemmas.
Lemma 5.6. Let be a Dirichlet character of modulus D then
[L(1, )[ log D.
Proof. We split the converging series at Y :
L(1, ) =

mY
(m)
m
+

m>Y
(m)
m
.
The rst sum is bounded by

mY
1
m
log Y +c
1
; for the second sum we introduce
M

:= max [

A<m<B
(m)[; clearly M

D. Abels summation then gives:

m>Y
(m)
m

mY

nm
(n)

(m
1
(m+ 1)
1
) M

Y
1
The choice Y = M

provides L(1, ) log M

log D.
Lemma 5.7. Let be a quadratic Dirichlet character of modulus D then
[L(1, )[ D
1/2
.
Proof. Indeed if corresponds to an imaginary quadratic eld K (in particular
K
is equal to D up to a small factor) then we have
L(, 1) = Res
s=1

K
(s) =
2h
K
w

K
D
1/2
,
whereas if corresponds to a real quadratic eld K then we have
L(, 1) = Res
s=1

K
(s) =
2h
K
log

K
D
1/2
.

Proof. (sketch of proof of theorem 5.5) Introduce the product of zeta functions:
f(s) := (s)L(s,
1
)L(s,
2
)L(s,
1

2
).
The crucial property is that the Dirichlet coecients of f(s) =

a
n
n
s
are posi-
tive
7
; indeed the Euler factor is
f
p
(s) =
1
(1 p
s
)(1
1
(p)p
s
)(1
2
(p)p
s
)(1
1

2
(p)p
s
)
and thus
log f
p
(s) =

m1
(1 +
1
(p
m
))(1 +
2
(p
m
))
p
ms
m
.
We denote := L(1,
1
)L(1,
2
)L(1,
1

2
) the residue of f(s) at s = 1.
7
The true reason for this is that if the characters
i
correspond to quadratic elds K
i
then,
up to nitely many Euler factors, the function f(s) is the Dedekind zeta function of the eld
K
1
K
2
.
AND L-FUNCTIONS 39
The following claim is easy but its innocent and trivial proof is responsible for
the ineectivity in Siegels theorem.
Lemma 5.8. For all > 0, there exists
1
mod D
1
and ]1 , 1[ such that for
any
2
we have f() 0.
Indeed if there are no Siegel zeroes, any ]1 , 1[ will do; otherwise let
]1 , 1[ be a zero for some
1
.
By shifting in the following positive integral the line of integration from = 2
to = we get, using Perrons formula
1
1
4!

nx
a
n
n

(1
n
x
)
4
=
1
2i
_
2+i
2i
f(s+)
x
s
s(s + 1)(s + 2)(s + 3)(s + 4)
ds
=
x
1
(1 )(2 )(3 )(4 )(5 )
+
f()
4!
+O
_
(D
1
D
2
)
1+
x

(1 )
_
.
Here the upper bound for
_
()
f(s + )
x
s
s(s+1)(s+2)(s+3)(s+4)
ds relies on the func-
tional equation and Phragmen-Lindelof as in section 3.3.
Since by lemma 1 we have f() 0 we infer that 1
x
1
1
if (D
1
D
2
)
2+
x
because
1
D
1
D
2
. Consequently, since L(1,
1
) log(D
1
D
2
) log D
1
we get
upon picking x = c(D
1
D
2
)
2+
= c

D
2+
2
:
L(1,
2
) > cD
(2+)(1)
2
(log D
2
)
1
,
which gives what we want provided (say) (2 + )(1 ) < /2 and D
2
is large
enough.
5.2. Brauer-Siegel theorem for abelian varieties. A look at the residue for-
mula for the Dedekind zeta function and the Birch and Swinnerton formula suggests
the following correspondence:

F
(s) L(A/K, s)
h
F
[X(A/K)[
R
F
Reg(A/K)
O

F,torsion
(A(K) A

) (K)
torsion

F

A
or H(A/K),
and, being a bit more audacious:
Conjecture 5.9. (Analogue of Brauer-Siegel theorem for abelian varieties) Con-
sider the family of abelian varieties A of xed dimension d, dened either over
a xed number eld K or over the function eld K of a smooth, projective, geo-
metrically connected algebraic curve ( dened over the nite eld F
q
with q := p
n
elements. Denote by H(A/K) its exponential height and suppose that it tends to
innity, then :
(5.5) log (#X(A/K) Reg(A/K)) log H(A/K),
40 M. HINDRY
where #X(A/K), resp. Reg(A/K), denotes the order of the Tate-Shafarevich
group, resp. the Neron-Tate regulator of A/K.
If true this conjecture implies the following remarks.
(1) It provides an upper bound Reg(A/K) H(A/K)
1+
and thus a bound
for the size of generators of the Mordell-Weil group.
(2) Similarly, since the regulator cannot be too small, it implies a bound for
the size of the Tate-Shafarevich group, namely #X(A/K) H(A/K)
1+
.
(3) When r = 0 (or equivalently under the BSD conjecture when L(A/K, 1) ,=
0) we get a lower bound #X(A/K) H(A/K)
1
.
Just like in the number eld case, it is interesting to look at the orthogonal
case when one xes the abelian variety and makes the eld K grow in towers.
There is the following result
8
over function elds.
Theorem 5.10. (Tsfasman-Kunyavski [KuTs08]; Zykin [Zy09]) Let A
0
be an el-
liptic curve dened over F
p
and let K
i
= F
p
(C
i
) be a tower of elds with the genus
of C
i
going to innity. then
(5.6) lim
i
log
p
(#X(A
0
/K) Reg(A/K))
g(C
i
)
= 1

m=1

m
log
p
[A
0
(F
p
m)[
p
m
.
where
m
is dened as the limit (assumed to exist which is always possible after
extracting a subsequence) of [C
i
(F
p
m)[/g(C
i
).
Notice that niteness of the Tate-Shafarevich group is, in the case of a constant
abelian variety, a theorem due to Milne.
The situation over number elds is still very conjectural (we formulate here
everything over Q but there is no diculty in extending at least the statements to
arbitrary abelian variety and number eld) :
Theorem 5.11. (H[Hi07]) Assuming BSD conjectures for elliptic curves A/Q and
the generalised Riemann hypothesis for L(A/Q, s) then
limsup
log (#X(A/Q) Reg(A/Q))
log H(A/Q)
1.
Assuming further an analytic estimates of the type L

(A/Q, 1)[ H(A/Q)

then
log (#X(A/Q) Reg(A/Q)) log H(A/Q).
The situation over function elds is better.
Theorem 5.12. (HPacheco [HiPa10]) Consider the family of elliptic curves over
a function eld K = F
q
(C). Assume niteness of some -primary component of
X(A/K) (or equivalently that the order of the zero of L(A/K, s) at s = 1 is the
expected one) then
5liminf
log (#X(A/K)Reg(A/K))
log H(A/K)
limsup
log (#X(A/K)Reg(A/K))
log H(A/K)
1.
Assuming further an analytic estimates of the type L

(A/K, 1)[ H(A/K)

then
log (#X(A/K) Reg(A/K)) log H(A/K).
Granting the BSD conjecture, the proof requires the following three estimates :
8
Tsfasman indicated to me there is a gap in the paper [KuTs08] hence, though the result is
most probably true, the proof should be regarded as incomplete at the present time.
AND L-FUNCTIONS 41
(1) Show that 1 #A(K) A

(K)
tor
H(A)

.
(2) Show that 1 T (A/K)
tor
H(A)

.
(3) Show that H(A)

(A/K, s) H(A)

.
Each is somewhat delicate but the lower bound in the third is the most dicult and
a general argument is still missing. Nevertheless for some families the conjecture
may be fully proven :
Theorem 5.13. (HPacheco [HiPa10]) Consider the family of elliptic curves E
d
over the function eld K = F
p
(t) = F
p
(P
1
) dened by their Weierstrass equation:
y
2
+xy = x
3
t
d
.
The Tate-Shafarevich group of E
d
/K is nite and, as d goes to innity, we have:
log (#X(E
d
/K) Reg(E
d
/K)) log H(E
d
/K)
d
6
log q.
This family was studied by Ulmer who showed that is contains curves with very
large rank.
42 M. HINDRY
6. Commented bibliography
We give here indications about where and what to read to go further than this
brief and sketchy introduction.
Lecture I. Among excellent general references on analytic number theory Ill
recommend [Da80, IwKo04, Te95]. For algebraic number theory, including an in-
troduction to Artin and Dedekind zeta functions read [La70].
Lecture II. A concise introduction to Weil conjectures may be found in [Ha83]
(appendix C.); for the construction of an adequate cohomology a comprehensive
reference is [Mi80], the presentation of Hasse-Weil zeta functions is inspired by
[Se70, Se65], for a nice introduction to elliptic curves see [Sil86], for modular forms
see [DiSh08]; the theory of automorphic representation has its roots in [Ta50], see
[Bu97] for a friendly introduction.
Lecture III. The classical complex variable lemmas, Tauberian theorems and
analytic estimates may be found in [Da80, IwKo04, Te95]. To see the use of
tauberian theorems in the context of Hasse-Weil L-functions, read for example
[HiPa05, HiPaWa05].
Lecture IV. For a description of many formulas (some conjectural) describing
special values of zeta and L-functions see [RaScSc88, Be85, BlKa90, De79, GrZa86,
Li83, Wa82], the BSD formula over function elds was formulated in [Ta66], for
related conjectures see also [Ta65],
Lecture V. The classical Brauer-Siegel theorem is proven in [Br47, Sie35]; a later
and simpler proof appeared in [Go74]; for further study see [St74, Ts01, TsVl02].
The various analogues for elliptic curves and abelian varieties are studied in [Hi07,
HiPa10, KuTs08, Zy09], the example is studied in [Ul02]. For the lower bound
conjecture and its link with small zeroes [IwLuSa00].
References
[Be85] A. Beilinson, Higher regulators and values of L-functions J. Soviet Math. 30, 2036
2070, 1985.
[BlKa90] S. Bloch, K. Kato, L-Functions and Tamagawa Numbers of Motives, in The
Grothendieck Festschrift, Vol. 1 (P. Cartier et al. eds.), Birkhauser, 1990.
[Br47] R. Brauer. On zeta-functions of algebraic number elds, Amer. J. Math. 69, 243250,
1947.
[Bu97] D. Bump, Automorphic forms and representations, Cambridge University Press,
1997.
[Da80] H. Davenport, Mutiplicative Number Theory, GTM 74, Springer, 1980.
[De79] P. Deligne, Valeurs de fonctions L et periodes dintegrales. Symp. Pure Math. A.M.S.
33, 313346, 1979.
[DiSh08] F. Diamond, J. Shurman, Introduction to modular forms, GTM 228, Springer, 2008.
[Go74] D. Golfeld, A simple proof of Siegels theorem, Proc. Nat. Acad Sci. US 71, 1974,
page 1055.
[GrZa86] B. Gross, D. Zagier, Heegner points and derivatives of L-series, Inventiones Mat.
84, 225320, 1986.
[Ha83] R. Hartshone, Algebraic geometry, GTM 52, Springer-Verlag, 1983.
[Hi07] M. Hindry, Why is it dicult to compute the Mordell-Weil group?, In Diophantine
Geometry Proceedings (ed. Zannier), Scuola Norm. Pisa 2007, 197219.
[HiPa05] M. Hindry, A. Pacheco, Sur le rang des jacobiennes sur un corps de fonctions. Bull.
Soc. Math. France 133, 275295, 2005.
[HiPaWa05] M. Hindry, A. Pacheco, R. Wazir, Fibrations et conjecture de Tate, J. Number Th.
112, 345368, 2005.
AND L-FUNCTIONS 43
[HiPa10] M. Hindry, A. Pacheco, Analogue of the Brauer-Siegel theorem for abelian varieties
in positive characteristic. Preprint, 2010.
[IwKo04] H. Iwaniec, E. Kowalski, Analytic number theory. Colloquium Publications. Ameri-
can Mathematical Society 53, 2004.
[IwLuSa00] H. Iwaniec, W. Luo, P. Sarnak, Low lying zeroes of families of L-functions. Publ.
Math. I.H.

E.S. 91, 55131, 2000.

[KuTs08] B. Kunyavski, M. Tsfasman, Brauer-Siegel theorem for elliptic surfaces, Int. Math.
Res. Not. IMRN 8 , 9 pp, 2008.
[La70] S. Lang, Algebraic number theory, Addison-Wesley, 1970.
[Li83] S. Lichtenbaum, Zeta functions of varieties over nite elds at s = 1. Arithmetic
and geometry, Vol. I, 173194, Progr. Math., 35, Birkhauser, 1983.
[Mi80] J. S. Milne,

Etale cohomology, Princeton University Press, 1980.
[RaScSc88] M. Rapoport, N. Schappacher, P. Schneider, Beilinsons conjectures and Special Val-
ues of L-Functions, Perspectives in Mathematics, Academic Press, 1988.
[Se70] J-P. Serre, Facteurs locaux des fonctions zeta des varietes algebriques (denitions et
conjectures), Seminaire Delange-Pisot-Poitou 1969/70, expose 19.
[Se65] J-P. Serre, Zeta and L-functions, in Arithmetic Algebraic Geometry (ed. Schilling),
8292, New York: Harper and Row, 1965.
[Sie35] C. L. Siegel.

Uber die Classenzahl quadratischer Zahlk orper, Acta Arith. 1, 8386,
1935.
[Sil86] J. H. Silverman, The arithmetic of elliptic curves. GTM 106, Springer-Verlag, 1986.
[St74] H. M. Stark, Some eective cases of the Brauer-Siegel theorem, Inv. Math. 23, 135
152, 1974.
[Ta50] J. Tate, Fourier Analysis in Number elds and Hexkes Zeta-Functions, Ph.D Prince-
ton 1950 [reproduced in Algebraic Number Theory, ed. Cassels, Frolich, Academic
Press, 1967].
[Ta65] J. Tate, Algebraic cycles and poles of zeta functions, in Arithmetic Algebraic Geom-
etry (ed. Schilling), 93110, New York: Harper and Row, 1965.
[Ta66] J. Tate, On the Birch and Swinnerton-Dyer conjecture and a geometric analog, Sem.
Bourbaki exp. 306, 1965/66.
[Te95] G. Tenenbaum, Introduction `a la theorie analytique et probabilistique des nombres,
Cours Specialises. Societe Mathematique de France, Paris, seconde edition, 1995.
[Ts01] M. Tsfasman, Asymptotic properties of global elds. Mullen, Gary L. (ed.) et al.,
Finite elds with applications to coding theory, cryptography and related areas. Pro-
ceedings of the 6th international conference on nite elds and applications, Oaxaca,
Mexico, May 2001.
[TsVl02] M. Tsfasman, S. Vladut , Innite global elds and the generalized Brauer-Siegel the-
orem. Mosc. Math. J. 2, 329402, 2002.
[Ul02] D. Ulmer, Elliptic curves with high rank over function elds, Annals of Math. 155,
295315, 2002.
[Wa82] L. C. Washington, Introduction to cyclotomic elds, GTM Springer, 1982 .
[Zy09] A. Zykin, On the Brauer-Siegel theorem for families of elliptic surfaces over nite
elds (in Russian), Mat. Zametki, 86, 148-150. English translation: Math. Notes 86,
140142, 2009.
Universit e Denis Diderot Paris VII, Institut de Math ematiques de Jussieu, U.F.R.
Math ematiques, case 7012, 175 rue de Chevaleret, 75013 Paris, France
E-mail address: [email protected]