1

Elementary Theory of L-Functions I

E. Kowalski

1 Introduction
In this first chapter we will define and describe, in a roughly chronological order
from the time of Euler to that of Heeke, some interesting classes of holomorphic
functions with strange links to many aspects of number theory. Later chapters will
explain how at least some of the mysterious aspects are understood today. But it
should be emphasized that there are still many points that are not fully explained,
even in a very sketchy, philosophical way.
I will particularly try to mention some of the more peculiar features of the theory
of L-functions (and of automorphic forms) which arise from the point of view of
analytic number theory. I will also give indications at the places where future
chapters after mine will bring new perspectives.
The next chapter will develop the points presented here, and in particular will
sketch proofs of some of the most important ones, especially when such a proof
yields new insights into the theory.
The mathematicians whose name are most important for us now are Euler, Gauss,
Dirichlet, Riemann, Dedekind, Kronecker, Heeke, Artin, and Hasse.

2 The Riemann zeta function

The first L- function has been given Riemann's name. This fact is convenient for us:
it seems to call for some explanation since no one denies that other mathematicians,
most notably Euler, considered this function before, and these explanations are the
best entrance to our subject.
The function in question is defined by the series

1
s(s) = " -.
~ns
n~I

For integers s ~ 1, this was studied even before Euler, and even for s ~ 2,
it is well known that Euler first found an exact formula (see (5.1)). However the
starting point for the theory of L-functions is Euler's discovery that the existence
and uniqueness of factorization of an integer as a product of prime powers im-

D. Bump et al., An Introduction to the Langlands Program

© Springer Science+Business Media New York 2004
2 E. Kowalski

plies that

~(s) = n--1'
p 1--
1
(2.1)
pS

a product over all prime numbers. From the divergence of the harmonic series,
Euler deduced from this a new proof of Euclid's theorem that there are infinitely
many primes, and with some care be obtained the more precise formulation that

L-1 =log log X+ 0(1)

p~X p

as X --+- +oo. 1 Thus (2.1) clearly contained new information about the distribution
of prime numbers.
Riemann's insight [Rie] was to remark that the function ~(s) thus defined, if s
is taken to be a complex variable (s = a + it in his notation), is holomorphic in
its region of convergence. This justifies looking for its (maximal) analytic contin-
uation, but that too had been done before (see [We]). It is the combination of the
Euler product expansion (2.1) and the analytic continuation given by the functional
equation described below, which is the cause for all our rejoicing, as it reveals the
strange "duality" between the complex zeros of ~(s) and prime numbers.
To be more specific, Riemann stated that ~ (s) has a meromorphic continuation
to the whole complex plane, the only singularity being a simple pole with residue
1 at s = 1, and that moreover this analytic continuation satisfied the following
property, aptly named the functional equation: the function

A(s) = rr-sf 2 r(s/2)~(s) (2.2)

is meromorphic except for simples poles at s = 0 and s = 1 and satisfies for all
s E C the relation

A(l - s) = A(s). (2.3)

From the simple poles of ~(s) at s = 1 and of f(s/2) at s = 0 one deduces

in particular that ~(0) = -rr- 112 r(1/2)/2 = -1/2. Moreover, the other poles at
s = -2n, n ~ 1 integer, of r(s/2) show that (2.3) implies that~( -2n) = 0, for
n ~ 1: those zeros are called the trivial zeros of ~(s).
This, and in fact all of (2.3) for integers s ~ 1, was already known to Euler in
the language of divergent series!

1To avoid any controversy, here are the definitions of Landau's 0(· ··)and Vinogradov's « sym-
bols: f = O(g) as x ~ xo means that there exists some (unspecified) neighborhood U of xo. and
a constant C ~ 0 such that lf(x)l ~ Cg(x) for x E U; this is equivalent to f « g for x E U
where now U is specified beforehand, and in this latter case one can speak of the "implicit constant"
C in«. However we sometimes also speak of estimates involving 0(· ··)being "uniform" in some
parameters: this means that the U and C above can be chosen to be independent of those parameters.
 I. Elementary Theory of L-Functions I 3

The presence of the "gamma factor" in the functional equation was not well
understood: indeed (2.3) was often written in completely different ways (see, e.g.,
[Ti, Chapter 2]). The more transparent and conceptual proofs of (2.3), and its
various generalizations, by the Poisson summation formula (see the next chapter)
and theta functions gave the gamma factor a clear de facto standing, but it is only
by the time of Tate's thesis [Ta] that its role was made clear as the Euler factor at
the archimedean place of Q. In general I should mention that an Euler product is
a product over primes of the form

n p
Lp(s) (2.4)

and that Lp(s) is called the Euler factor at p. 2

The interplay of the zeta function with primes was revealed by Riemann in the
form of the so-called "explicit formula." His version (and that proved later by van
Mangoldt and others) obscured somewhat the essential point in purely analytic
difficulties; going around them by means of smooth test functions, one can state
it in the following straightforward manner:
Proposition 2.1. Let rp : ]0, +oo[-+ C be a C 00 function with compact support,
and let

cP(s) = Jo
r+oo rp(x)xs-ldx,
be its Mellin transform, which is entire and decays rapidly in vertical strips. Let

Then we have

A + -.-
L rp(p) 1
2zrr
J (Ir'(s)
--
r
1r'(l-s))A
- - --
r
2
- 2 2 2
rp(s)ds. (2.5)
l;(p )= 0 (-1 /2)
O<Re(p)<i

A more general case of this will be sketched in the next chapter (Proposition 3 .5).
In this general formula, which expresses sums over prime numbers in terms of
sums over zeros of ~(s), the first term on the right-hand side is really the contri-
bution of the pole at s = 1 (very often, giving the main term in some asymptotic

2Be careful that when an Euler product is defined over prime ideals p in a number field i' Q, as
will be described below, the data of the Euler factors Lp (s) is stronger than the product even if the
latter defines a holomorphic function.
4 E. Kowalski

formula). We implicitly make use of the fact that {(s) =F 0 for Re(s) > 1. It is
very important to realize that this is by no means obvious from the series repre-
sentation: it follows immediately from the Euler product expansion (an absolutely
convergent infinite product is nonzero). This must not be underestimated: it is for
instance the key analytical input in Deligne's first proof of the Riemann Hypoth-
esis for varieties over finite fields [De, 3.8]. The functional equation then shows
that {(s) =F 0 for Re(s) < 0, except for the trivial zeros already identified.
Riemann immediately expresses that it is "likely" that all the (nontrivial) zeros
of {(s) satisfy Re(s) = 1/2. This is indeed the best possible situation if one is
interested in prime numbers: if one takes for <p test functions which converge, say,
in the sense of distributions to the characteristic function of [0, X], one finds rather
easily that one has an estimate for the error term in the prime number theorem of
the form

l{I(X) = L (log p) =X+ ocxP*) as X~ +oo (2.6)

pk~X

if and only if {3* > sup{Re(p)}. Since the functional equation implies that
sup{Re(p)} ~ 1/2, the Riemann Hypothesis is seen to be simply the statement
that primes are distributed "in the best possible way."
Following the standard terminology, the strip 0 ~ Re(s) ~ 1, which must
contain the nontrivial zeros is called the critical strip and the line Re(s) = 1/2
is called the critical line. This will apply to all the L-functions in the first two
chapters, but for automorphic L-functions, the critical strip may (depending also
on normalization) be translated to the right by some amount.

3 Dirichlet L-functions
Although it is fundamental, the case of the Riemann zeta function does not exhibit
some very important features of the general theory. Those, notably the notions of
"conductor" and of "primitivity," and the link with class-field theory and algebraic
number theory more generally, appear first in the case of Dirichlet L-functions.
Dirichlet defined those functions [Di] to prove his famous theorem:
Theorem 3.1. Let q ~ 1 and a ~ 1 such that (a, q) = 1. Then there are infinity
many primes p = a (mod q) and more precisely

-p1 = -loglogX
1
<p(q)
+ 0(1)
p~X
p=a (modq)

as X~ +oo.
He proved this result by detecting invertible congruence classes modulo q by
means of harmonic analysis in (Z/qZ) x, which lead him to Dirichlet characters:
 I. Elementary Theory of L-Functions I 5

an arithmetic function x : Z --+ C is a Dirichlet character modulo q );: 1 if there

exists a group homomorphism

X : (ZjqZ)x --+ex

such that x(x) = x(x (modq)) for (x, q) = 1 and x(x) = 0 if (x, q) =f. 1. To
such a character, extending Euler's definition, one associates the L-function

the last equation, the Euler product, being a consequence of unique factorization
and of the complete multiplicativity of x:

x(mn) =x(m)x(n)forallm);: 1, n);: 1.

The orthogonality relations for characters of a finite group imply immediately

the relation

I: ~=- 1- I:~ + - 1- I: x<a) I: x<P)

p~X p cp(q) p~X p cp(q) x;61 p~X p
p:=a (modq}

and Dirichlet's Theorem is easily seen to be equivalent with the assertion that if
x =f. 1 is a Dirichlet character, then L(x, 1) =f. 0. For Dirichlet (coming before
Riemann) this is not in the sense of analytic continuation, but rather a statement
about the sum of the (conditionally) convergent series which "is" L(x, 1). 3 For
the nontrivial character X4 modulo 4, for instance, this series is simply
1 1 1T
L(x4. 1) = 1- 3 + 5 - · · · = 4 .
However, since x is a periodic function of n ~ 1, it is not difficult to extend the
s
proof of the analytic continuation of (s) and its functional equation based on the
Poisson summation formula and theta functions (explained in the second chapter)
to establish analogue statements for Dirichlet characters. This almost forces us
to introduce the notion of primitivity: a Dirichlet character x modulo q is called
primitive, and q is called its conductor if there does not exist ij I q, ij < q and a
x
character of (ZjijZ)x such that

X (n) = X(n (mod ij)) for (n, ij) = 1.

If q is prime, then any nontrivial character is primitive modulo q. Any Dirichlet
character is induced in the way described above by a unique primitive character
X, and one has the relation
L(x,s) = L(x,s) n (1- x(p)p-s) (3.1)
plqfij

3For x ;6 I, the partial sums are bounded so the series converges.

6 E. Kowalski

which shows that the analytic properties of Dirichlet L-functions can be reduced
immediately to that associated to primitive characters, the second factor being a
finite product (which is an entire function).
For X primitive, the case q = 1 corresponds to the zeta function and is special;
if q > 1, the character xis nontrivial and one shows that L(x, s) has an extension
to C as an entire function. The functional equation requires more disctinctions: if
x (-1) = 1, X is called even, and one defines

whereas for x(-1) = -1 (X is odd), one defines

A(x, s) = n-<s+I)f2r((s + 1)/2)L(x, s).

Moreover, let r (X) be the Gauss sum attached to X,

•<x> = I: x<x)e(xjq) (3.2)

x (modq)

where e(z) = e2i:n:z. Then the functional equation is

A(x, s) = e(x)q 112-s A(X, 1- s) (3.3)

where
r(x)
if x(-1) =1
{ (3.4)
e(x) = _1(x)
if x(-1) = -1.
..fii
Thus a number of features appear which are ubiquitous in the more general context:
• The functional equation (3.3) involves certain global invariants of the Dirich-
let character. First, its conductor q; notice that even when starting with an
imprimitive character, the functional equation will only be possible for the
associated ("inducing") primitive character, and thus the a priori unknown
conductor q will appear in this way. This remark, in other contexts, is very
fruitful (see Section 4).

• The second invariant is the argument of the functional equation e(x), also
called the root number.lt is a complex number of absolute value 1, as a simple
calculation (or a second application ofthe functional equation) reveals (in
the case of primitive character), and is related to the Gauss sum r(x). Notice
that the latter is also a kind of finite field analogue of the gamma function
(multiplicative Fourier transform of an additive character). In general, the
argument of the functional equation is a very delicate invariant.

• The functional equation relates L (X, s) with the value of at 1 - s of the dual
character X·
 1. Elementary Theory of L-Functions I 7

An important special case is that of real-valued characters, i.e., characters of order

2 (or quadratic characters). In such a case, we have x
= x and the argument
e(x) becomes a sign ±1. Actually, Gauss had computed the exact value of r(x)
(Dirichlet gave a simple analytic proof) which implies that e (X) = +1 for any
primitive quadratic character x. i.e., r(x) = .fii is x( -I) = 1 and r(x) = i.fii
if X ( -1) = -1.

We now come back to Dirichlet's proof of Theorem 3.1. It is enough to show that
L(x, 1) :F 0 for x :F 1, and because of (3.1) one may assume that x is primitive.
Dirichlet easily showed that if x is not real, then L (X, 1) :F 0 (if not, 1 would also
be zero of L(X, s) :F L(x, s) andthisiseasytoexclude). The difficult case, as it has
remained to this day, is that of a quadratic x. Dirichlet's proofthat L(x, 1) :F 0 is
a direct application of his analytic class number formula (see (4.5) below) and thus
introduces another recurrent and all-important theme, that of linking £-functions
defined by analytic means (here the Dirichlet L-function) to others coming from
algebraic number theory, or algebraic geometry (here the Dedekind zeta function
of a quadratic field). 4 We explain this in the next section in the more general
context of number fields, and will simply conclude by saying that this proof of
nonvanishing of L (X, 1) is spectacular, but somewhat misleading. One can indeed
give very natural elementary proofs of the fact that
1
L(x, 1) » .fii'

for q > 1, which is the more precise outcome of the class number formula (at
least in the imaginary case.) And more importantly, all progress concerning the
problem (going back to Gauss) of understanding the order of magnitude of the
class number of quadratic fields has come from the opposite direction, by using
the class number formula to relate it to the special value of the L-function and
estimating (more often, failing to estimate) the latter quantity by analytic means
(see, e.g., Landau [L], Siegel [Si], Goldfeld [Go]).

4 Dedekind zeta functions

Let K be a number field, so that K /Q is a finite field extension, of degree d =
[K : Q]. We denote 0, or OK, the ring of integers inK. Dedekind defined the
zeta function of K by the series

sK(s) = L (N1a)S
a

(over nonzero integral ideals a c 0) which is easily seen to be absolutely conver-

gent for Re(s) > 1. So, forK = Q, one recovers s(s). Since in 0 there is unique

4 Today one would say of "motivic" origin.

8 E. Kowalski

factorization of ideals into products of prime ideals, the same argument as for ~ (s)
proves that there is an Euler product expansion

~K(s) = np
(1- (Np)-s)-1'

and indeed ~K(s) carries much the same information about the distribution of
prime ideals inK as ~(s) does about ordinary prime numbers.
However, there is more to it: if we write ~K(s) as an ordinary Dirichlet series
with some coefficients rK(n),

~K(s) = L rK(n)n-s, (4.1)

n~l

we immediately see that knowing ~K (s) implies knowing how prime numbers split
in 0: indeed rK(Pk) is the number of ideals with norm= pk, hence TK(P) = d if
and only if p splits completely in K, r K (p) = 1 means that p is totally ramified, etc.
This information is one of the most relevant to the study of the arithmetic of K,
and this brings a new urgency to the exploration ofthe properties of ~K (s), which
concentrates really on the zeta function itself instead of its logarithmic derivative
(which is the key to the analytic distribution of prime ideals) and considers ~(s)
almost as a "trivial case."
Because of the definition by a series expansion, it is possible again to extend the
method used for ~(s) (and for L(x, s)) to derive analogous analytic properties of
~K (s). This is not completely straightforward however, and contains or uses most
of the "basic" notions and results of algebraic number theory, which we now recall
(see, e.g., [La], [CF]):

• The field K has d = r1 + 2rz embeddings a : K ~ C, r1 real embeddings

and rz pairs of complex embeddings. To each is associated an absolute value
I · lu. lzlu = la(z)l if a is real and lzlu = la(z)l 2 if a is complex. Those
absolute values are also called the archimedean places of K; note a and
a give rise to the same place. We use usually v to denote a place. We let
dv = 1 if v is real and dv = 2 otherwise. Note that the norm of an element
z E K is given by
Nz = nv
lzlv (product over the places).

• The ring 0 is of rank d = [K : Q]. Let (wi) be a Z-basis, 1 ~ i ~ d. The

discriminant D of K is
D = det(wj)f.u·
For every prime ideal p in 0, the quotient F p = 0 KIP is a finite field and
the (absolute) norm of p is the cardinality ofF p.
 1. Elementary Theory of L-Functions I 9

• The multiplicative group U = oxof units of 0 is a finitely generated

abelian group of rank rt + rz - 1, the free part of which admits an injection
into Rd through the "logarithmic" map

l: u ~ (logla(u)i)a

the image of which is a lattice in the subspace V = {(xa) I L Xa =

0 and Xa =xu}. The volume R > 0 of a fundamental domain for l(U) C V
(with respect to the Lebesgue measure on V) is called the regulator of K.

• One denotes by w = w K the order of the torsion subgroup of U (the number

of roots of unity in K).
• The group of ideal classes inK, denoted Pic(O) or H (K), is a finite abelian
group of order denoted h(K) = h(O), called the class number of K. The
ring 0 is principal (i.e., every ideal is principal) if and only if h(K) = 1.
If E I K is a finite Galois extension of number fields with Galois group G, for
every prime ideal p in K and prime ideal 1iJ I p in E above p, one defines the
decomposition group of 1iJ

and the inertia subgroup

lq1 ={a E Gq3 I a(x) =x (modliJ) forallx E OE} <l Gq3,

and 1iJ is unramified if lq1 = 1.

The Frobenius conjugacy class aq1, also denoted [liJ, ElK], is the conjugacy
class in Gq3 I lq1 (i.e., in the decomposition group ifliJ is unramified) of the elements
a such that
a(x) = xNI' (mod 'iJ) for all x E 0£,
i.e., the "reduction modulo 'iJ" of aq1 is the canonical generator of the finite
field extension F q3 IF I'. The conjugacy class of a in G only depends on p and is
denoted ap.
If G is an abelian group, the Frobenius conjugacy class is reduced to a single
element, the Frobenius automorphism, which only depends on p and is denoted a I'.
The precise statement about analytic continuation and functional equation for
{K(s) takes the following form: let

AK(s) = Jl"-rJS/2 (2Jr)-'28 f(sl2r 1 rcsr2 {K(s).

Then A K (s) admits analytic continuation to a meromorphic function on C with
simple poles at s = 1 and at s = 0. Thus {K (s) is meromorphic with a simple pole
at s = 1. Moreover the residue of {K(s) at s = 1 is given by the analytic class
number formula

(4.2)
10 E. Kowalski

The functional equation satisfied by A K (s) is

AK(s) = IDI 112-s AKO- s). (4.3)

Compared to the previous functional equations, one notices that the discriminant
appears as the conductor did for Dirichlet L-functions, and that the corresponding
root number is always +1.
The formula (4.2) is a newcomer here, the prototypical example of a large class
of formulae (proved or conjectured to hold) which express the special values of
L-functions in terms of global invariants ofthe "motivic" objects they are related
to. The best known generalization is the conjecture of Birch and Swinnerton-Dyer
but there are many others.
The notion of a "special value" can be put in a rigorous conceptual framework
(Deligne, etc.) explaining which values should have those properties for a given
L-function. One should maybe mention that analytic number theorists often have
reasons to consider as "special" points that escape this algebraic context: a typical
example is the deformation theory of Phillips and Sarnak [PS], where the point in
question is (probably) transcendental.
Let me now explain how this general formula relates to Dirichlet's proof of
Theorem 3.1. Let x be a nontrivial primitive quadratic character of conductor q.
The essence of the argument is that there is a relation

~(s)L(x,s) = ~K(s) (4.4)

between the Dirichlet L-function and the Dedekind zeta function of the quadratic
field K = Q(-./x( -1)q) (so K is imaginary if x( -1) = -1 and real if x (-1) =
1). There is a lot of mathematics behind this seemingly simple statement, namely it
is a reformulation of the quadratic reciprocity law of Gauss, and as such it has had
considerable influence on the development of the whole theory of L-functions.
Before discussing this further, observe that since ~(s) and ~K(s) on both sides
of (4.4) have a simple pole at s = 1, it follows that L (X , 1) i= 0, and more precisely
that
27rh(K)
if K is imaginary
{
L(x. 1) = 2;:({floge
if K is real, e > 1 being the fundamental unit of K
.;q
(4.5)

which implies L(x, 1) > 0 (even L(x, 1) ~ q- 112 if K is imaginary); see also
the discussion in Section 3.
We come back to (4.4). A prime number pis either ramified, inert, or split in
the quadratic field K. Expliciting this in the Euler product expression of ~K (s),
we have

p split p inert pramified

 1. Elementary Theory of L-Functions I 11

and comparing with (4.4) it follows that the latter is equivalent with the following
characterization of the splitting of primes:

p is ramified if and only if X (p) = 0, if and only if p Iq

{ pis split in Kif and only if x(p) = 1 (4.7)
p is inert in K if and only if X(p) = -1.

The point is that this characterization is in terms of the Dirichlet character x which
is a finite amount of data related to Q and not to K (it is after all only a particular
periodic arithmetic function of period q ).
Even more concretely, basic algebraic number theory shows that another char-

I
acterization, involving Legendre symbols, exists:

p is ramified if and only if p Iq

p is split in K if and only if ( ~) = 1 (4.8)

p is inert in K if and only if ( ~) = -1

(where D is the discriminant of K), but there is no reason a priori to expect that
the map

(defined only on the rather chaotic set of prime numbers!) has any particularly good
property. Now the primitive quadratic characters are very easy to characterize using
the structure of (Z/qZ)x. In particular, if q is squarefree and odd, there is a unique
primitive character modulo q, given by

x(n) = n (!:)
pin p

(this is clearly a quadratic character modulo q, and since for any p I q there are
quadratic residues and nonresidues modulo p > 2, x cannot be induced from
qfp).
=
In this case, one has x (-1) q (mod 4), so the quadratic field is K = Q(.fij)
if q = 1 (mod4) and K = Q(R'J if q =
3 (mod4), with discriminant D =
x(q)q. Take q to be a prime> 2: then x(-1) = (-l)<q-l)/ 2 so comparison
of (4.7) and (4.8) gives (after some easy checking)

the original form of the quadratic reciprocity law. Similarly, for even conductors
of quadratic characters. In fact (4.4) for all quadratic fields is equivalent with
quadratic reciprocity.
12 E. Kowalski

This explains why it may seem natural and desirable to seek generalizations of
the factorization (4.4) for other L-functions, specifically at first for Dedekind zeta
functions, and why one would see them as some form of "reciprocity law," if the
factors appearing are (like L (X, s)) defined in terms of objects of the base field Q,
since this would amount to describing the splitting of primes in K solely in terms
of data belonging to Q.
Of course, it is also natural to expect similar situations to exist for relative
extensions E I K of number fields.
The general (still very much conjectural) form of this idea belongs to the the-
ory of Artin L-functions and to some of the conjectures of Langlands, and will
be covered later. There is however one important special case where a (mostly)
satisfactory theory exists and with which I will finish this first chapter. It is the case
when one has an abelian extension Ll K and the results are known as class-field
theory.

5 Heeke £-functions, class-field theory

Let E I K be an extension of number fields and assume that it is a Galois extension
with abelian Galois group G. It follows in particular that for every prime ideal p
of K unrarnified in E, the Frobenius element a p e G is well defined.
A formal computation, although nontrivial (a special case of the in variance by
induction of Artin L-functions), shows that there is a factorization of ~E(s) as

~E(s) = n L(p, s) (5.1)

peG

where G is the (dual) character group of G and

L(p, s) = n (1 - p(ap)(Np)-s)- 1,
pinK

where p(ap) refers to the image of the Frobenius element by the Galois repre-
sentation (either 0 or 1-dimensional) induced on C 1P, to take care of the ramifi-
cation. Of course, if pis unrarnified in E, this is literally p(ap). (For instance, if
pis totally split in E with Np = q, the p-factor on the left of (5.1) is made of
d = [E : K] factors (1 - q-s)- 1, one for each prime above p; on the right there
are IGI = IGI = d characters p, and for each p(ap) = p(l) = 1, hence the
p-factor is indeed the same).
This new Euler product is, for trivial reasons, still absolutely convergent for
Re(s) > 1 since lp(a)l = 1 for any a E G, but its analytic continuation, for
individual p, is by no means obvious: one can see that a simple definition as a series
instead of a product does not appear immediately. There is no clear reason that the
coefficients of this Dirichlet series should have any good property (compare with
the discussion of the quadratic reciprocity law above). 5 In particular, at first sight,

5For instance, are the partial sums of the coefficients bounded, as for Dirichlet characters?
 1. Elementary Theory of L-Functions I l3

computing the coefficient of (N a)-s seems to require knowing the factorization

of a in E.
Example 5.1. For a quadratic field K /Q, there are two characters, the trivial one
with L-function equal to ~(s) and a quadratic character xz with

p split pinert

since G ::::: {±1} and xz(a p) = + 1 (resp., -1) if and only if pis split (resp., if p
is inert).
Of course one has L(xz. s) = L(x, s) in the factorization (4.4) involving the
Dirichlet character (compare with (4.8)).
The L-function part of class-field theory can be thought of as the identification
of the L- functions of Galois characters in terms of generalizations of the quadratic
Dirichlet character appearing in (4.4). For a general base field K, those were
defined by Heeke. However, in the case of Q it suffices to consider the original
Dirichlet characters (not necessarily quadratic), which we state, without striving
for the utmost precision:
Theorem 5.2 (Kronecker-Weber theorem). Let K /Q be an abelian extension
with Galois group G, let p : G --+ ex be a Galois character and L(p, s) its L-
function as above. There exists a unique primitive Dirichlet character x modulo
q for some q ;;:::: 1 such that

L (p, s) = L (X , s).

In particular, the analytic continuation, functional equation and other properties

of L(x, s) are thus inherited by all those L-functions. This gives as a consequence
"reciprocity laws" describing the splitting of prime numbers in abelian extensions
of Q, and also the following corollary (the more usual form of the Kronecker-
Weber Theorem).
Corollary 5.3. Let K /Q be an abelian extension. Then K is contained in some
cyclotomic field, i.e., there exists some m ;;:::: 1 such that K C Q(JLm) where JLm is
the group of mth roots of unity in C.

Proof of the corollary. Let

~K(S) = nX
L(x. s) (5.2)

be the factorization obtained from (5.1) and Theorem 5.2 in terms of (some) Dirich-
let characters. Let m be the l.c.m. of the conductors q x of the occuring characters.
Then in fact we have K C Q(JLm).
To prove this, we recall a simple general fact about number fields (other proofs
are possible): we have E c K for Galois extensions E /Q and K /Q if and only if all
but finitely many of the prime numbers p which are totally split inK are totally split
14 E. Kowalski

in E. Thus let p be a prime number totally split in Q(/Lm). By elementary theory

of cyclotomic fields [CF, Chapter 3] this means that p = 1 (modm). Thus for
any character x modulo m, we have x(p) = 1, and in particular for all characters
occuring in (5.2) we have x(p) = 1; i.e., by the Kronecker-Weber Theorem,
p(a p) = 1 for all p E G. This means that a P = 1 E G, hence that p is totally
split inK. 0

For more about the relationships between Dirichlet characters and cyclotomic
fields, see, e.g., [Wa, Chapter 3].
Remark 5.4. The Kronecker-Weber Theorem, as stated here, bears a striking
resemblance to the L-function form of the modularity conjecture for elliptic curves
(explained in de Shalit's chapters). One can prove Theorem 5.2 by following the
general principles of Wiles's argument [Tu] (deformation of Galois representations,
and computation of numerical invariants in a commutative algebra criterion for
isomorphism between two rings).
We come to the definition of Heeke L- functions of a number field K. As one
may easily imagine, they are L-functions of the form

L(x, s) = L x(a)(Na)-s =
a
np
(1- x(P)(Np)-s)- 1 (5.3)

for some "arithmetic" function x defined on integral ideals in a in K. The proper

definition, in the language of ideals, requires some care however; this becomes
much clearer in the idele-theoretic description. See, e.g., [Co] for further details in
the "classical" language. The difficulties arise because of the various archimedean
places and the class number being possibly > 1.
Let m be a nonzero integral ideal of K, which will play the role of modulus.
One defines the subgroups Im (resp., Pm) of the group I of fractional ideals in K
(resp., of the subgroup of principal ideals) by

Im ={a E I I (a, m) = 1}
Pm ={a= (a) E Im n P I a= 1 (mod m)}.

The finite abelian group Hm = Imf Pm is called the ray-class group modulo m.
Let also Um be the group of units in Pm.
Let ~00 be a (unitary) character

~oo : Kx /Qx -+ ex

(these can be easily described using the various places at infinity, see the example
below) such that Um C ker ~00 • Hence ~00 induces a homomorphism

~oo : Pm-+ ex.

Definition. A Heeke character of weight ~00 for the modulus m is a homomorphism

 1. Elementary Theory of L-Functions I 15

(unitary) such that

x((a)) = ~oo(a)
if a = (a) E Pm. The character x is extended to I by putting x(a) = 0 if
(a, m) ::fo I.
One can define primitive Heeke characters in much the same way as for Dirichlet
characters, with the same basic properties: any character is induced by a unique
primitive one, and their L-functions are the same up to a finite Euler product, as
in (3.1).
For a Heeke character x the L-function L(x, s) has the Euler product expan-
sion (5.3) by multiplicativity and converges absolutely for Re(s) > I.
Example 5.5. If we take m = 1 and ~00 = 1, then the corresponding Heeke char-
acters are just the ideal class-group characters, i.e., the (finitely many) characters
H(K)-+ ex of the ideal class-group of K. More generally, ifm is arbitrary but
~00 = 1, the resulting Heeke characters are called ray-class characters modulo m.
For the trivial Heeke character xo. one has L(xo. s) = ~K(s).
Example 5.6. Let K = Q. Then the only possibility is ~00 = 1, m = (m) for
some unique m ;::: 1; Heeke characters modulo m are the same thing as Dirichlet
characters modulo m. Primitivity also corresponds.
This extends very similarly for any field with class number 1.
Example 5.7. Let K be a quadratic field and a the nontrivial element in the Galois
group of K. Then Kx /Qx ~ {a E Kx I Na = 1} by a r+ a faa. Assume first
that K is imaginary. Then one checks easily that ~00 must be of the form

~oo(a) = (,:,r
(i.e., arg(a)") for some integer u E Z, to which the weight can be identified.
If, on the other hand, K is real, then one finds that

a
~oo(a) = ( ~
)"I ( jaal
aa )"2

with Ul, U2 E {0, 1}.

Heeke managed to prove the fundamental analytic properties of his L-series,
in complete analogy with the case of Dirichlet L- functions. The next chapter will
sketch the proof, which becomes more transparent in the language of adeles: this
translation is the subject of Tate's famous thesis [Ta].
Theorem 5.8 (Heeke). Let x ::fo 1 be a primitive, nontrivial Heeke character of
K. Then L (X, s) admits analytic continuation as an entire function and there is
a gamma factor f(~00 , s), depending only on the weight, which is a product of
gamma functions, such that

A(x,s) = f(~oo.s)L(x,s)
16 E. Kowalski

satisfies
A(x, s) = e(x)(IDINm) 112-s A(x, 1- s) (5.4)
for some complex number e(x) of absolute value 1. Here D is the (absolute)
discriminant of KIQ and Nm the norm from K to Q.
We are a little vague: one can indeed write a formula for e(x) (in terms of a
Gauss sum for x) as well as for f(~e<" s). The point is that they appear "naturally"
in the course of the proof. Writing the gamma factor "as a function of the weight"
is actually reminiscent of adelic arguments. Of course, one should compare this
to (3.3) and (4.3).
Now one can state the analogue of Theorem 5.2 for abelian extensions of a
number field K, which is extremely similar and encompasses much of class-field
theory.
Theorem 5.9 (Artin). Let K be a number field, E I K a finite abelian extension
with Galois group G, let p : G --+ ex be a Galois character and L(p, s) the
associated L-function. Then there exists a unique primitive Heeke character x of
K, of modulus m such that
L(p, s) = L(x, s).
Actually, this identity holds locally, i.e., the p-component of the Euler products
are equal for all prime ideal p of K, namely6
p(ap) = x(P)forall p coprime tom.
Remark 5.10. Actually, as in the more general case of Artin L-functions, one can
define beforehand, in terms simply of p, a conductor f(p) and a weight ~00 (p).
The proof of the theorem shows that f = m and the weight of x is ~00 • This added
precision is very important in applications, if only because it makes this statement
theoretically verifiable for any given p by brute force search (this is similar to
Weil's stipulation that the level of the modular form associated to an elliptic curve
over Q should be its conductor).
Remark 5.11. Comparing the functional equations on both sides of (5.1) after
applying Theorem 5.9, one obtains a relation between the conductors of the Galois
(or Heeke) characters related to E I K and the discriminant of E. In the simple case
K = Q this takes the form
IDI = nqx
X
where Dis the discriminant of E IQ, x are the Dirichlet characters in (5.2) and qx
their conductors. For E = Q(/Le) for .e > 2 prime, the factorization is

~E(s) = n
x (modl)
L(X, s).

6As often remarked by Serre, this last statement is stronger than the first one if K f. Q; compare
the footnote after (2.4).
 1. Elementary Theory of L-Functions I 17

There are e- 1 Dirichlet characters, of which one has conductor 1 and e-

2 have
e,
conductor hence we recover the well-known discriminant IDI = ee-
2.
Similarly, since the root number for ~E(s) is 1, one deduces a relation between
the root numbers e(x). One can also fruitfully compare the residues at s = 1 on
both sides.

Just as in the case of Dirichlet characters, the Heeke characters of a number field
K have (independently of class-field theory) many applications to the equidistri-
bution of ideals in various classes. Actually, those split naturally into two kinds:
one concerns the distribution of prime ideals p in K, and is based on the same
methods using the logarithmic derivatives of Heeke L-functions. The other con-
cerns the distribution (inN) of the norms of integral ideals inK, i.e., the (average)
properties of the arithmetic function TK(n) in (4.1). Here one uses ~K(s) and the
Heeke L-functions themselves, and it is somewhat simpler since their singularities
are completely known. One deduces for instance that

l{a C 0 I Na ~ X}J = (Ress=l ~K(s))X + O(Xi-ifd)

as X -+ +oo, and that the ideals are equidistributed in ideal classes. For example,
taking a quadratic field K, one gets the asymptotic formula for the number of
integers n ~ X represented by a quadratic form (with multiplicity). Standard
methods of analytic number theory can also be used to deduce an asymptotic
formula for the number of integers that are norm of an ideal in K (excluding
multiplicity).

Example 5.12. Take K = Q(i). Then 0 = Z[i] has class number one and the
number of ideals with norm ~ X is equal to the number of lattice points in C inside
a disc of radius ../R (the Gauss circle problem). One gets in this case trivially

L 1 = rr X + 0 ( -JX) as X -+ +oo
n2+m2:=:;;x

(r(n) is the number of representations of n as sum of two squares). More subtle

arguments yield the same formula with X 113 as error term. It is conjectured that
any exponent > 1I 4 will do, which is the best possible. In suitably smoothed form,
even better results are known, for instance the automorphy of theta functions (see
Chapter 3) easily implies that if qJ is a smooth function with compact support in
[0, +oo[, we have

~ ({) (
n2 + m2)
X = rr X Jo
r+oo ({J(x)dx + coqJ(O) + O(X-l/ 2)

as X -+ +oo, where co is some constant. (Use the Mellin transform and move the
line of integration to Re(s) = -1/2.)
18 E. Kowalski

6 Function fields
I will only say a very few words about the "geometric" analogue (or "function
field case") of the various theories described before, mostly by lack of proper
competence. More detailed explanations will come in future chapters.
The analogy between the arithmetic of Z and that of the ring of polynomials
over a field k[X] is a very old one. If one takes fork a finite field Fq with q = pd
elements, the analogy deepens. For instance, if P is an irreducible polynomial, the
residue field Fq[X]/(P) is a finite field, as happens with number fields.
This leads to a theory that is very similar to that described previously in many
respects, although geometric intuition and some other intrinsic characteristics tend
to make it simpler (e.g., one can differentiate polynomials, but not integers.) It
provides both a different set of problems, and a way to get evidence for conjectures
which remain intractable over number fields, for instance, those concerning the
zeros of L-functions (for example, the work of Katz and Sarnak [KS] on the fine
distribution of zeros, or the recent proof by Lafforgue on the global Langlands
correspondence for G L (n) over function fields).
We will describe here briefly the case of curves. Let C /Fq be an algebraic curve
(are there any others?) over Fq given as the set of zeros in P2 of a homogeneous
polynomial F E Fq[X, Y, Z]. Let D/Fq be "the" smooth projective model of C.
In the language of Hasse andArtin, this would be identified with thefunctionfield
K = Fq(C) ofC, which is a finite extension of the fieldFq(X) of rational functions
in one variable (geometrically, the latter corresponds to pi and the extension to a
ramified covering C --+ pi).
The nonarchimedean valuations of K correspond in one-to-one fashion with the
set of closed points of C (in scheme-theoretic language), in other words with the
orbits of the action of Gal(Fq /Fq) on the points of C defined over the algebraic
closure ofFq. If C is affine, they also correspond to the maximal ideals in the ring
of functions regular on C. Define the zeta function of C by the Euler product

Z(C; s) = n X
(1- (Nx)-s)-i,

over the set of closed points, with N x equal to the cardinality of the residue field
at x. This is clearly an analogue of the Dedekind zeta function for the field K, and
it turns out to satisfy many of the same properties.
Since the norm of x is always of the form qa for some a ~ 1, the degree
deg(x) of x, it is convenient to putT = q-s and consider Z(C) as a formal power
series in T

Z(C) = n
X
(1 - ydeg(xl)-i.

This turns out to have another expression, peculiar to the geometric case, which
gives a direct diophantine interpretation of this zeta function.
 1. Elementary Theory of L-Functions I 19

Lemma 6.1. We have

Z(C) = exp(L IC(:qn)l rn ).

n~J

as formal power series.

Sketch of proof On applies the operator T d log to both sides. On the right the
result is
L iC(Fqn)ITn,
n~J

while on the right after expanding a geometric series one gets

X k~J n

with
Nn = L dl{x I deg(x) = d}l.
din

The degree d is the cardinality of the Galois orbit of a closed point x, so we see
that Nn = IC(Fqn)l, as desired. D

The basic properties of the zeta function are as follows:

Theorem 6.2. Suppose C is smooth and projective. Then the zeta function has an-
alytic continuation and functional equation. More precisely (! ), Z (C) is a rational
function ofT of the form

Z(C) = Pzg
(1 - T)(l - qT)

where Pzg E Z[T] is a monic polynomial of even degree 2g, g being equal to the
genus of C. It satisfies

Z(C) = e(C)q-x!Zr-x Z(1/(qT))

where e(C) = ±1 and X = 2- 2g is the Euler-Poincare characteristic of C.

The polynomial Pzg can be expressed over C as

n
2g
Pzg = (1- aiT)
i=l

with lai I = y'q for all i.

The first part was proved by Schmidt in general; it is based on the Riemann-
Roch theorem. The second follows also from this argument. The last part is the
Riemann Hypothesis for Z(C): it was first proved by A. Weil. Weil also made con-
jectures generalizing those statements to higher dimensional varieties, and they
20 E. Kowalski

were proved in even stronger and much more powerful form through the work of
Grothendieck's school and particularly, of course, by P. Deligne.lt seems particu-
larly significant that these results are based on expressions for the polynomial P2g
as characteristic polynomials of a certain operator (the Frobenius operator) acting
on some cohomology group: a (natural) analogue of such a phenomenon is eagerly
sought in the number field case.
Remark 6.3. In terms of s, the functional equation becomes

Z(C; s) = e(C)qx<s- 112>Z(C; 1- s),

and the Riemann Hypothesis says that the zeros of Z(C; s) satisfy

q -a = ITI = ICXi 1- 1 = q - 112 ,

i.e., Re(s) = 1/2. So the analogy is indeed very clear.