Artin L–functions, Artin primitive roots Conjecture and applications

CIMPA-ICTP Research School, Nesin Mathematics Village 2017

An introduction to the Riemann zeta function

Michel Waldschmidt

Content of this file (29 pages).

• p.2–3: a list of references.

• p. 4–22: from the book Arithmetics by M. Hindry: p. 127–131, 134–139,

148–155.

• p. 23–29: from Éléments d’analyse et d’algèbre (et de théorie des nom-

bres) by P. Colmez : p. 40–43 and 51–53.

Further main references (not included in this file):

• Lectures on the Riemann zeta function by H. Iwaniec, Chap. 1–12, p.

3–43.

• An introduction to the theory of the Riemann zeta-function by S. J. Pat-

terson, Chap. 1–11, p. 1–66.

• An introduction to zeta functions by P. Cartier, p 1–63.

1
References

LMFDB is the database of L-functions, modular forms, and related objects. It

is an extensive database of mathematical objects arising in Number Theory.
http://www.lmfdb.org/
A basic course in number theory, containing the basic elements of analytic num-
ber theory, is

• M. Hindry, Arithmetics, Universitext, Springer, London, 2011. Trans-

lated from the 2008 French original.

Perhaps the simplest and more direct introduction to the Riemann zeta function
is contained in the recent small (and a bit synthetic) book :

• H. Iwaniec, Lectures on the Riemann zeta function, vol. 62 of University

Lecture Series, American Mathematical Society, Providence, RI, 2014.

Chapters 5–7 (p.21–26) present the theory from the beginning to the functional
equation; chapters 8–11 (p.27–42) contain the rest of the basic theory, ending
with the proof of PNT.

A discussion of the special values of the Riemann zeta function is included in

• P. Cartier, An introduction to zeta functions, in From number theory

to physics (Les Houches, 1989), Springer, Berlin, 1992, pp. 1–63.

One of the most classical reference is

• G. H. Hardy and E. M. Wright, An introduction to the theory of

numbers, Oxford University Press, Oxford, sixth ed., 2008. Revised by D.
R. Heath-Brown and J. H. Silverman, With a foreword by Andrew Wiles.

Other classical books, all containing the basic theory of ⇣(s), are:

• A. E. Ingham, The distribution of prime numbers, Cambridge Mathe-

matical Library, Cambridge University Press, Cambridge, 1990. Reprint
of the 1932 original, With a foreword by R. C. Vaughan.

2
• E. C. Titchmarsh, The theory of the Riemann zeta-function, The Claren-
don Press, Oxford University Press, New York, second ed., 1986. Edited
and with a preface by D. R. Heath-Brown.

• T. M. Apostol, Introduction to analytic number theory, Springer-Verlag,

New York-Heidelberg, 1976. Undergraduate Texts in Mathematics.

• A. A. Karatsuba, Basic analytic number theory, Springer-Verlag, Berlin,

1993. Translated from the second (1983) Russian edition and with a
preface by Melvyn B. Nathanson.

• S. J. Patterson, An introduction to the theory of the Riemann zeta-

function, vol. 14 of Cambridge Studies in Advanced Mathematics, Cam-
bridge University Press, Cambridge, 1988.

• H. Davenport, Multiplicative number theory, vol. 74 of Graduate Texts

in Mathematics, Springer-Verlag, New York, third ed., 2000. Revised and
with a preface by Hugh L. Montgomery.

• H. L. Montgomery and R. C. Vaughan, Multiplicative number theory.

I. Classical theory, vol. 97 of Cambridge Studies in Advanced Mathemat-
ics, Cambridge University Press, Cambridge, 2007.

• G. Tenenbaum, Introduction to analytic and probabilistic number theory,

vol. 163 of Graduate Studies in Mathematics, American Mathematical
Society, Providence, RI, third ed., 2015. Translated from the 2008 French
edition by Patrick D. F. Ion.

• P. Colmez, Éléments d’analyse et d’algèbre (et de théorie des nombres),

les éditions de l’école polytechnique, 2011, 678 pages.

3
§1. Elementary Statements and Estimates 127

1.6. Theorem. As x tends to infinity, we have the following asymptotic

behavior:

π(x; a, b) := card{p prime , p ! x, p ≡ a mod b} ∼ x · (4.4)

φ(b) log x

In this section, we will expand on some so-called “elementary” methods

(which in this context means that they do not involve complex variables)
which can be used to prove the previous assertions, except for Dirichlet’s
theorem on arithmetic progressions and the prime number theorem. They
will however allow us to prove a partial version: there exist two constants,
c1 , c2 > 0, such that c1 x/ log x ! π(x) ! c2 x/ log x.
! "
1.7. Lemma. The following estimate holds: n log 2 ! log 2n n ! n log 4.
! " #2n !2n"
Proof. From the binomial theorem, we know that 2n n $! % k=0 k =
2n (2n)!
(1+1)2n = 4n . Next, we have the following lower bound: = =
n (n!)2
2n(2n − 1) · · · (n + 1)
" 2n . #
n(n − 1) · · · 1
& '
# n ;
1.8. Lemma. The following formula holds: ordp (n!) = m!1 pm
furthermore, the sum can be restricted to m ! log n/ log p.
(n
Proof. Write n! = 1 ·2 ·3 · · · n = k=1 k. The number of integers ! n which
are divisible by p is ⌊n/p⌋, and the number of integers ! n divisible by p2
is ⌊n/p2 ⌋, etc. Thus ordp (n!) is the sum of the ⌊n/pm ⌋. Finally, pm ! n is
equivalent to m ! log n/ log p, hence the first statement is proved. #

We can therefore write

⎛ ⎞
$ % ) $ % &
) ) 2n ' & '
2n 2n
log = ordp log p = ⎝ m − 2 nm ⎠ log p.
n n p p
p"2n p"2n m!1
(4.5)
To find a lower bound, we only !2nkeep
" the terms that satisfy n < p ! 2n. In
fact, such a p clearly divides n = (2n)!/(n!)2 , and thus we obtain
$ % )
2n
n log 4 " log " log p = θ(2n) − θ(n).
n
n<p"2n

From this, we obtain an upper bound of the form θ(x) ! Cx. This is true
because
128 4. Analytic Number Theory

m−1
) m−1
)
θ(2 ) =
m
θ(2 k+1
) − θ(2 ) !
k
2k log 4 = (2m − 1) log 4.
k=0 k=0

Therefore, if 2m ! x < 2m+1 , then

θ(x) ! θ(2m+1 ) ! 2m+1 log 4 ! (2 log 4)x. (4.6)
To obtain an upper bound, we could notice that ⌊2u⌋ − 2⌊u⌋ always equals
0 or 1 and equals 0 whenever u < 1/2. Thus
⎛ ⎞
$ % ) ) & ' & '
2n 2n − 2 n ⎠ log p
n log 2 ! log = ⎝
n pm pm
p"2n m!1

) $ log(2n) %
! log p = log(2n)π(2n).
log p
p"2n

From this, we have a lower bound of the form π(x) " Cx/ log x. This is
true because if 2n ! x < 2(n + 1), then
n log 2 . / log 2
π(x) " π(2n) " " x −1 · (4.7)
log(2n) 2 log x
Furthermore, we can easily see that
) )
θ(x) = log p ! log x 1 = π(x) log x. (4.8)
p"x p"x

Next, notice that for 2 ! y < x,

) )
π(x) − π(y) = 1! 1 log p = 1 (θ(x) − θ(y)) .
log y log y
y<p"x y<p"x

It follows that
θ(x) θ(x)
π(x) ! + π(y) ! + y.
log y log y
By choosing y = x/(log x)2 and by recalling the previous inequality (4.8),
we have
θ(x) θ(x) x
! π(x) ! + · (4.9)
log x log x + 2 log log x (log x)2
To summarize, it is easy to see from inequalities (4.6), (4.7), (4.8) and (4.9)
that (θ(x) ∼ x) is equivalent to (π(x) ∼ x/ log x) and that
C1 x ! θ(x) ! C2 x and C3 x/ log x ! π(x) ! C4 / log x. (4.10)

Furthermore, the following comparison of the function θ(x) to the function

ψ(x) is not difficult to see:
§1. Elementary Statements and Estimates 129

) √ √
θ(x) ! ψ(x) := log p = θ(x) + θ( x) + θ( 3 x) + . . .
pm "x

log(x) √ √
! θ(x) + θ( x) ! θ(x) + C log x x.
log 2
Finally, if we denote by pn the nth prime number, we have π(pn ) = n by def-
inition. The prime number theorem therefore implies that n ∼ pn / log(pn )
and that pn ∼ n log n. We can check that the latter statement is in fact
equivalent to the prime number theorem.

#
1.9. Lemma. (Abel’s formula) Let A(x) := n"x an and f be a function
of class C 1 . Then,
) 0 x
an f (n) = A(x)f (x) − A(y)f (y) − A(t)f ′ (t)dt. (4.11)
y<n"x y

1 n+1 1 n+1
Proof. We first point out that n A(t)f ′ (t)dt = A(n) n f ′ (t)dt =
A(n) (f (n + 1) − f (n)). Therefore, setting N = ⌊x⌋ and M = ⌊y⌋ yields
0 N N
) −1 0 n+1 N
) −1
A(t)f (t)dt =
′
A(t)f (t)dt =
′
A(n + 1) (f (n) − f (n))
M n=M n n=M
N
)
= f (n)(A(n − 1) − A(n)) + f (N )A(N ) − A(M )f (M )
n=M +1
N
)
=− f (n)an + f (N )A(N ) − A(M )f (M ).
n=M +1

This proves the formula when x and y are integers. For the general formula,
observe that
0 x
A(t)f ′ (t)dt = A(⌊x⌋) (f (x) − f (⌊x⌋)) = A(x)f (x) − A(⌊x⌋)f (⌊x⌋).#
⌊x⌋

Applications. 1) The formula gives a fairly precise comparison between

the “sum” and the “integral” (see Exercise 4-6.10 for some refinements). To
be more precise, if we take an = 1 and integrate by parts, we have:
N
) 0 N 0 N
f (n) = f (t)dt + (t − ⌊t⌋)f ′ (t)dt. (4.12)
n=M +1 M M

If we choose f (t) = 1/t, we obtain

130 4. Analytic Number Theory

)N 0 N 0 N
1 =1+ dt − (t − ⌊t⌋) dt2
n 1 t 1 t
n=1
$ 0 ∞ % 0 ∞
= log N + 1 − dt
(t − ⌊t⌋) 2 + (t − ⌊t⌋) dt2
1 t N t
. /
= log N + γ + O 1 ,
N
1∞
where γ := 1 − 1 (t − ⌊t⌋) dt2 is Euler’s constant.
t
2) Take an = 1, so A(t) = ⌊t⌋, y = 1 and f (t) = log t. We therefore have
0 x
⌊t⌋dt
log (⌊x⌋!) = ⌊x⌋ log(x) −
1 t
0 x 0 x
⌊t⌋ − t
= x log x − dt + (⌊x⌋ − x) log x − dt
1 1 t
= x log x − x + O(log x).

We should point out that Stirling’s

√ formula gives a slightly more precise
statement, namely n! ∼ n en −n
2πn, and hence log(n!) = n log n − n +
1 log n + 1 log(2π) + ϵ(n) where lim
n→∞ ϵ(n) = 0.
2 2
Furthermore, we see that
)
log (⌊x⌋!) = ordp (⌊x⌋!) log p
p"x
))& x '
= m log p
p"x m!1
p
) log p ) .2 3 / ))& '
=x x x x
p + log p p − p + pm log p
p"x p"x p"x m!2
) log p
=x p + O(x),
p"x
#
where the last estimate comes from the upper bound θ(x) = p"x log p =
O(x), from (4.6) and the estimate
))& x ' ) ) log p ) log p
log p ! x = x = O(x).
pm pm p(p − 1)
p"x m!2 p"x m!2 p"x

From this, we can deduce the first formula in Proposition 4-1.2,

) log p
p = log x + O(1). (4.13)
p"x
§2. Holomorphic Functions (Summary/Reminders) 131

To get the second, we apply Abel’s formula (Lemma 4-1.9) letting f (t) =
1/ log t and an = log p/p if n = p is prime and an = 0 otherwise. By setting
# log p
A(x) = p"x p , we have that
) 1 )
p = an f (n)
p"x n"x
0 x
A(x) A(t)dt
= +
log x t(log t)2
2
0 x 0 x
dt + (A(t) − log t)
= 1 + O(1/ log x) + dt
2 t log t 2 t(log t)2
0 ∞
(A(t) − log t)
= log log x − log log 2 + 1 + dt + O(1/ log x).
2 t(log t)2

2. Holomorphic Functions
(Summary/Reminders)
This section, without proofs, is a summary of some of the fundamental
properties of functions of a complex variable that we will be using. It could
be helpful to use [74] as a reference.
Concerning series, we will use the product rule for calculating the product
of two absolutely convergent series:
4∞ 54 ∞ 5 ∞
4 n 5
) ) ) )
an bn = ak bn−k ,
n=0 n=0 n=0 k=0

as well as rearrangement of the order of summation in a series with positive

terms am,n : 4 ∞ 5 4∞ 5
)∞ ) )∞ )
am,n = am,n .
n=0 m=0 m=0 n=0
#∞
A power series S(z) = n=0 an z n is said to have a radius of convergence
R " 0 (possibly R = 0 or R = +∞) if the series converges for all |z| < R
and diverges for all |z| > R; furthermore, the convergence is absolute in
the interior#of the disc of convergence and the function is of class C ∞ with
∞
S (k) (z) = n=k n(n − 1) · · · (n − k + 1)an z n−k . In fact, the function S can
be expanded as a power series around every point z0 ∈ # D(0, R), in other
∞
words, for every z ∈ D(z0 , r) ⊂ D(0, R), we have S(z) = n=0 bn (z − z0 )n
(with bn = S (n) (z0 )/n!). Such a function only has a finite number of
zeros in every closed disc (or compact set) which is contained in D(0, R).
We define the multiplicity of a zero, z0 , as the integer k such that S(z) =
134 4. Analytic Number Theory

The series converges normally at every point in the open disk D(1, 1) =
{z ∈ C | |1 − z| < 1}, and the convergence is uniform (and also normal)
on every closed disk with center 1 and radius r < 1. Therefore, F (z) is
holomorphic on D(1, 1). If z is a real number in the interval ]0, 2[, we can
see that F (z) = log z (ordinary logarithm), and in particular,
exp (F (z)) = z.
The previous formula indicates that the two functions, the identity and
exp ◦F , which are analytic on the disk D(1, 1), coincide on the segment
]0, 2[ and hence on the whole disk. Thus F defines a complex logarithm on
the disk |z − 1| < 1.

2.8. Definition. Let f (z) be a holomorphic function on U . We say

that F (z) is a branch of the logarithm of f on U (and we write, with a
slight abuse of notation, F (z) = log f (z)) if F (z) is holomorphic and if
exp (F (z)) = f (z).

2.9. Remark. If F (z) is a branch of the logarithm of f , then f is never

0 on U , we have |exp (F (z))| = exp (Re F (z)) = |f (z)|, and hence
Re log f (z) = log |f (z)|.
Likewise, f ′ (z)/f (z) = F ′ (z) exp (F (z)) / exp (F (z)) = F ′ (z), and also

d log f (z) = f (z) ·

′

dz f (z)
Finally, if F1 and F2 are two logarithms, then F2 (z) = F1 (z) + 2kπi on any
connected set U .

This remark suggests that we should construct the logarithm of f (z) as an

antiderivative f ′ (z)/f (z), with the condition that f is not zero. We have
seen that this is possible if U is simply connected.

2.10. Proposition. Let U be a simply connected open subset of the

complex plane and f (s) a holomorphic function without any zeros in U .
Then there exists a holomorphic branch F (s) = log f (s) on U . Two such
branches differ by an integer multiple of 2πi.

We will finish this summary by explaining the notion of an infinite prod-

uct. The first idea consists of saying that a product is convergent if
!N
limN n=0 pn exists. This could be confusing because it is not true that
such a product is zero if and only if one of the factors is zero. For example,
it can be easily checked that
§3. Dirichlet Series and the Function ζ(s) 135

N #
" $
lim 1− 1 = 0.
N →∞ n+1
n=1

We can overcome this inconvenience by defining infinite products a little

differently. Observe first
! that a necessary condition for the convergence
of a non-zero product n pn is to have limn pn = 1; it therefore does not
hurt to assume% that pn = 1 + un where un tends to zero. In particular,
∞
log(1 + un ) = k=1 (−1)k ukn /k is well-defined when |un | < 1 and hence
when n ! n0 , which justifies the following definition.
!∞
2.11. Definition. A product n=0 (1 + un ) is convergent (resp. absolutely
%∞ if there exists n0 such that |un | < 1 for all n ! n0 and the
convergent)
series n=n0 log(1 + un ) is convergent (resp. absolutely convergent). A
!∞
product of functions n=0 (1 + un (z)) is uniformly convergent on K if there
%∞ n0 such that |un (z)| < 1 for all n ! n0 and z ∈ K and the series
exists
n=n0 log(1 + un (z)) is uniformly convergent (on K).

!∞
2.12. Lemma. A product P := n=0 (1 + un ) is absolutely convergent
%∞
if and only if the series n=0 |un | is convergent. If P is convergent, it is
zero if and only if one of the factors 1 + un is zero.

2.13. Proposition. Let (un (z)) be% a sequence of holomorphic functions

on an open set U such that the series n log(1+un (z)) converges uniformly
on every compact subset of U .
i ) Then the function defined by the infinite product
∞
"
P (z) := (1 + un (z))
n=0

is holomorphic on U .
ii ) For every z0 ∈ U , only a finite number of pn (z) := 1 + un (z) are zero
at z0 , and hence
∞
&
ordz0 P (z) = ordz0 pn (z).
n=0

3. Dirichlet Series and the Function ζ(s)

%∞ an
We call a Dirichlet series a series of the form F (s) = · We will
n=1 ns
now state its first important property.
136 4. Analytic Number Theory

%∞ a
3.1. Proposition. Let F (s) = n=1 ns be a Dirichlet series that we
n
will assume to be convergent at s0 . Then it converges uniformly on the sets
EC,s0 = {s ∈ C | Re(s − s0 ) ! 0, |s − s0 | " C Re(s − s0 )}.
%
Proof. For M ! 1, set AM (x) := M <n!x an n−s0 . By the hypothesis,
we then have |AM (x)| " ϵ(M ) where ϵ(M ) tends to zero as M tends to
infinity. Abel’s formula gives
& &
an n−s = an n−s0 n−(s−s0 )
M <n!N M <n!N
' N
= AM (N )N −(s−s0 )
+ (s − s0 ) AM (t)t−(s−s0 +1) dt.
M

We can find an upper bound for the integral as follows:

(' ( ' N
( N (
( −(s−s0 +1) (
( AM (t)t dt( " ϵ(M ) t−(σ−σ0 +1) dt
( M ( M

M −(σ−σ0 ) − N −(σ−σ0 )
= ϵ(M ) ·
(σ − σ0 )
By restricting to an angular sector EC,s0 bounded by σ − σ0 = Re(s) −
Re(s0 ) ! 0 and |s − s0 | = C(σ − σ0 ), we obtain
( (
( (
( & (
( −s (
an n ( " ϵ(M )(1 + C),
(
(M <n!N (

which suffices to show the uniform convergence on this sector (cf. the figure
below). #

σ − σ0
the domain |s − s0 | "
cos θ

The following corollary is a result of the general theorems recalled in the

previous section (in particular, Proposition 4 -2.6).

%∞
an
3.2. Corollary. Every Dirichlet series F (s) = has an abscissa
ns n=1
of convergence, say σ0 , such that the series converges for Re(s) > σ0 and
§3. Dirichlet Series and the Function ζ(s) 137

diverges for Re(s) < σ0 . Furthermore, the function F defined by the se-
ries is holomorphic in the half-plane of convergence Re(s) > σ0 , and its
%∞
derivatives are given by F (k) (s) = n=1 an (− log n)k n−s .
Proof. It suffices to let σ0 = inf{σ ∈ R | the series converges at σ}, then
to observe that every compact set in the (open) half-plane of convergence
is contained in a sector, as above, where the convergence is uniform. #

3.3. Remarks. 1) If a Dirichlet series converges at s0 = σ0 + it0 to

the number S, then the proof of Proposition 4 -3.1 (above) shows that
S = limϵ→0+ F (σ0 + ϵ + it0 ).
%
2) Set A(t) := n!t an . The previous proof allows us to establish the
formula
⎛ ⎞
&∞ ' ∞ & ' ∞
an n = s
−s ⎝ ⎠
an t −s−1
dt = s A(t)t−s−1 dt. (4.14)
n=1 1 n!t 1
%
In particular, if A(t) = n!t an is bounded, then the series converges
whenever Re(s) > 0.

The most famous

%∞ Dirichlet series is the Riemann zeta function, defined
by the series n=1 n . It is well-known, at least for real values and the
−s

general case follows from the real case, that the abscissa of convergence is
+1.

3.4. Theorem. (Euler Product) If Re(s) > 1, then the following formula
holds
&∞
1 "- 1
.−1
ζ(s) = = 1− s . (4.15)
n=1
ns p
p
% % %
Proof. Notice that p |p−s | = p p−σ " n n−σ , hence the product is
absolutely convergent. If Re(s) > %∞0, the convergence of geometric series
allows us to write (1 − p−s )−1 = m=0 p−ms . By taking the product over
the prime numbers p1 , . . . , pr which are " T , we obtain
/ ∞ 0
" " &
(1 − p−s )−1 = p−ms
p!T p!T m=0
&
= (pm
1 · · · pr )
1 mr −s

m1 , . . . , mr ! 1
p1 , . . . , pr " T
&
= n−s ,
n∈N (T )
138 4. Analytic Number Theory

where we denote by N (T ) the set of integers all of whose prime factors are
" T . Thus whenever Re(s) > 1, we have
( ( ( (
(∞ .−1 ( ( & (
(& 1 "- 1 ( ( 1 ( & (( 1 (( & 1
( ( ( (
( s − 1 − s =
( ( s (" ( s (= σ ·
(n=1 n p ( ( n ( n>T
n n>T
n
p!T n∈N
/ (T )

The last sum is the tail-end of a real convergent series (when σ := Re(s) >
1) and thus tends to zero, which proves both the convergence of the product
and Euler’s formula. #

3.5. Corollary. The function ζ(s) does not have any zeros in the open
half-plane Re(s) > 1. A holomorphic branch of log ζ(s) for Re(s) > 1 can
be constructed by setting
& & p−ms
log ζ(s) = m · (4.16)
p m"1

Furthermore, if we define the von Mangoldt function by

1
log p if n = pm ,
Λ(n) =
0 if not,
then
& & log p &∞
ζ ′ (s) Λ(n)
− = ms = s · (4.17)
ζ(s) p m"1
p n=1
n

Proof. We know 1 − p−s ̸= 0 and that the product is convergent. The first
assertion is therefore obvious. The second formula can be deduced from
Euler’s formula by taking the series expansion of the logarithm (valid for
|x| < 1) and summing:
∞
&
2 3 xm ·
log (1 − x) −1
= m
m=1

The second formula is therefore gotten by differentiating the first. #

Interlude (I). These formulas can be generalized by replacing Z and Q by

OK and K, and the uniqueness of the decomposition into prime factors by
the uniqueness of the decomposition into prime ideals (Theorem 3-4.18).
We denote by IK the set of non-zero ideals and PK the set of non-zero
(maximal) prime ideals in OK . Now we can introduce the Dedekind zeta
function and prove that
& " 2 3−1
ζK (s) := N(I) =
−s
1 − N(p)−s for Re(s) > 1.
I∈IK p∈PK
§4. Characters and Dirichlet’s Theorem 139

3.6. Proposition. The function ζ(s) can be analytically continued to a

meromorphic function to the half-plane Re(s) > 0, with a unique pole at
s = 1 with residue equal to +1.
Proof. The statement says that ζ(s) − 1/(s − 1), originally defined for
Re(s) > 1, has a holomorphic continuation to the half-plane% Re(s) > 0. To
prove this, we can write ζ(s), using the expression ⌊t⌋ = n!t 1, as
&∞ ' ∞
ζ(s) = 1
s =s ⌊t⌋t−s−1 dt
n=1
n 1
' ∞ ' ∞
=s t dt + s
−s
(⌊t⌋ − t) t−s−1 dt
1 1
' ∞
= 1 +1+s (⌊t⌋ − t) t−s−1 dt.
s−1 1

We know that |⌊t⌋ − t| " 1. The last integral is hence convergent and
defines a holomorphic function for Re(s) > 0. #

3.7. Remark. Actually, the function ζ(s) − 1/(s − 1) can be extended to

the whole complex plane, and moreover, ζ(s) satisfies a functional equation
(see Theorem 4-5.6 further down).

4. Characters and Dirichlet’s Theorem

4.1. Definition. If G is a finite abelian group, a homomorphism from
G to C∗ is called a character. The set of characters of G forms a group
denoted by Ĝ.

4.2. Proposition. The group Ĝ is isomorphic to the group G (but not

canonically).
Proof. If G = Z/nZ, a character satisfies χ(1) ∈ µn (where µn denotes as
usual the group of nth roots of unity), and the map χ '→ χ(1) provides
an isomorphism between Ĝ and µn . The latter is isomorphic to Z/nZ
and hence to G. We will now show that G! ∼
1 × G2 = Ĝ1 × Ĝ2 ; to see
why this is true, a character χ of G1 × G2 can be written as χ(g1 , g2 ) =
χ(g1 , e2 )χ(e1 , g2 ), and, by setting χ1 = χ(· , e2 ) and χ2 = χ(e1 , ·), we obtain
an isomorphism χ '→ (χ1 , χ2 ) from G! 1 × G2 to Ĝ1 × Ĝ2 . The general case
∼
is now easy: we have G = Z/n1 Z × · · · × Z/nr Z, hence

Ĝ ∼
b
= (Z/n1 Z × · · · × Z/nr Z)
∼ !
= Z/n ! ∼ ∼
1 Z × · · · × Z/nr Z = Z/n1 Z × · · · × Z/nr Z = G. #
148 4. Analytic Number Theory

Dedekind zeta function of the field Q(exp(2πi/m)), which explains in a

more conceptual manner why the coefficients are positive.

5. The Prime Number Theorem

We will prove the following form of the prime number theorem.
!∞
5.1. Theorem. The integral 1
(θ(t) − t)t−2 dt is convergent.

We will show that the convergence of this integral implies θ(x) ∼ x, and
hence π(x) ∼ x/ log x. To see this, suppose that lim sup θ(x)x−1 > 1, then
n ! 1 + ϵ.
there exists ϵ > 0 and xn tending to infinity such that θ(xn )x−1
For t ∈ [xn , (1 + ϵ/2)xn ], we therefore have
θ(t) − t θ(xn ) − (1 + ϵ/2)xn ϵxn /2 ϵxn
! ! ! ,
t 2
t 2
t 2
2(1 + ϵ/2)2 x2n
and consequently
" (1+ϵ/2)xn
θ(t) − t ϵ2
dt ! ,
xn t2 4(1 + ϵ/2)2
which contradicts the convergence of the integral. We can therefore con-
clude that lim sup θ(x)x−1 " 1. A symmetric argument shows that
lim inf θ(x)x−1 ! 1. It follows that lim θ(x)x−1 = 1.
To prove the theorem, we will use the following result from complex analysis
(due to Newman, see [55, 81]) concerning the Laplace transform.

5.2. Theorem. (“The analytic theorem”) Let h(t) be a bounded, piecewise

continuous function. Then the integral
" +∞
F (s) = h(u)e−su du
0

is convergent and defines a holomorphic function on the half-plane Re(s) >

0. Suppose that this function can be analytically continued to a holomorphic
function on the closed half-plane Re(s) ! 0. Then the integral converges
for s = 0 and
" +∞
F (0) = h(u)du.
0

Let us provisionally admit that this result is true and see how to apply it
§5. The Prime Number Theorem 149

to the function
" +∞
θ(t) − t
F (s) = dt
1 ts+2
" "
+∞
θ(eu ) − eu +∞ # u −u $
= e du =
u
θ(e )e − 1 e−us du.
0 eu(s+2) 0

The function h(u) := θ(eu )e−u − 1 is indeed bounded and piecewise con-
tinuous. If the analytic continuation hypothesis is satisfied, then we know
! +∞ ! +∞ θ(t) − t
that F (0) = 0 (θ(eu )e−u −1)du = 1 dt is indeed convergent.
t2
We could transform the integral which defines F (s) (for Re(s) > 0) as
follows: " +∞
θ(t) − t
F (s) = dt
1 ts+2
%∞ " n+1 " +∞
= θ(t)t −s−2
dt − t−s−1 dt
n=1 n 1

%∞
n−s−1 − (n + 1)−s−1
= θ(n) − 1s
s+1
n=1
%∞
= 1 n−s−1 (θ(n) − θ(n − 1)) − 1s
s+1
n=1
1 %
= p−s−1 log(p) − 1s ·
s+1
p

We have also seen that

ζ ′ (s) % % %
− = log(p)p−ms = log(p)p−s + log(p)p−ms .
ζ(s) p,m!1 p p,m!2

The second term in the last expression is a convergent series and hence holo-
&
morphic for Re(s) > 1/2. From this, we can deduce that p log(p)p−s =
ζ ′ (s)
− + a holomorphic function on Re(s) > 1/2 and finally that
ζ(s)
ζ ′ (s + 1)
F (s) = − − 1s + a holomorphic function on Re(s) > −1/2.
(s + 1)ζ(s + 1)
The key point in the proof is therefore the following result.

5.3. Theorem. (Hadamard, de la Vallée-Poussin) The function ζ(s) does

not have any zeros on the line Re(s) = 1.
150 4. Analytic Number Theory

Proof. We start with the formula

4 cos(x) + cos(2x) + 3 = 2(1 + cos(x))2 ! 0.
Recall that
% p−mσ−mit
log ζ(σ + it) = m , and
p,m
% p−mσ
log |ζ(σ + it)| = m cos(mt log p).
p,m

This implies that

' (
log |ζ(σ + it)|4 |ζ(σ + 2it)|ζ(σ)3
% p−mσ
= m (4 cos(mt log p) + cos(2mt log p) + 3) ! 0.
p,m!1

We can conclude from this, assuming σ > 1, that

|ζ(σ + it)|4 |ζ(σ + 2it)|ζ(σ)3 ! 1. (4.22)
Now, if ζ(s) had a zero of order k at 1 + it and with order ℓ at 1 + 2it,
then |ζ(σ + it)| ∼ a(σ − 1)k , |ζ(σ + 2it)| ∼ b(σ − 1)ℓ and ζ(σ) ∼ (σ − 1)−1
(where σ tends to 1 from above). The left-hand side of inequality (4.22) is
therefore (asymptotically) equivalent to c(σ − 1)4k+ℓ−3 , which implies that
4k + ℓ − 3 " 0, and hence k = 0. #

5.4. Corollary. The function defined on Re(s) > 1 by

ζ ′ (s)
G(s) := − − 1
sζ(s) s−1
extends to a holomorphic function on Re(s) ! 1.
Proof. The previous theorem shows that the function ζ ′ (s)/ζ(s) is holo-
morphic on the line Re(s) = 1, except for s = 1. Consequently, the
function G(s) also is. To study G(s) in a neighborhood of s = 1, we
use the fact that ζ(s) has a simple pole at s = 1, and consequently
ζ ′ (s)/ζ(s) = −1/(s − 1) + g(s), where g(s) is holomorphic in a neigh-
borhood of 1. Thus G(s) is indeed holomorphic in a neighborhood of 1 and
hence on the line Re(s) = 1. #

Appendix. Proof of the “analytic theorem”

Recall the statement of the analytic result used in the proof of the prime
number theorem.
§5. The Prime Number Theorem 151

5.5. Theorem. If h(t) is a bounded piecewise continuous function, then

the integral (the Laplace transform of h)
" +∞
F (s) = h(u)e−su du
0

is convergent and defines a function which is holomorphic on the half-plane

Re(s) > 0. Suppose that this function can be analytically continued to a
holomorphic function on the closed half-plane Re(s) ! 0. Then the integral
for s = 0 converges and
" +∞
F (0) = h(u)du.
0

Proof. The first part is analogous to the theorem of convergence for Dirich-
let series (see Exercise 4-6.2). We will therefore prove the second state-
!T
ment. For a (large) real number T , let FT (s) := 0 h(t)e−st dt; these
are functions which are holomorphic for all s ∈ C. We now need to
show that limT →∞ FT (0) exists and equals F (0). To do this, we con-
sider for some large R the contour γ = γ(R, δ) which bounds the region
S := {s ∈ C | Re(z) ! −δ and |s| " R}. Once we have fixed R, we can
choose δ > 0 sufficiently small so that F (s) is analytic on this region.

The trick lies in introducing the function

) *
2
GT (s) := (F (s) − FT (s)) e sT s
1+ 2 ,
R
so that GT (0) = F (0)−FT (0). Therefore, everything comes back to proving
152 4. Analytic Number Theory

that limT →∞ GT (0) = 0. To do this, we will use the residue theorem a first
time, noticing that
" ) *
GT (0) = F (0) − FT (0) = 1 (F (s) − FT (s)) e sT
1+ 2s2
ds ·
2πi γ R s

To find an upper bound on this integral, we cut the contour into two pieces:
γ1 , which is the piece of γ which lives in the half-plane Re(s) > 0, and γ2 ,
which lives in the half-plane Re(s) < 0. We then carry out the following
computation.
Let s be a number such that |s| = R or s = Reiθ . Then we have
+ ) * +
+ sT
+e 1 + s2
1 ++ = eRe(s)T ++e−iθ + eiθ ++ 1 = eRe(s)T 2 Re(s) ·
+ R2 s+ R R2
We also have the upper bound
+" ∞ + " ∞
+ + + −st + − Re(s)T
+
|F (s) − FT (s)| = + −st +
h(t)e dt+ " M +e + dt = M e ·
T T Re(s)
This gives us
+ " ) * +
+ 1 s2
ds +
+ (F (s) − F (s)) e sT
1 + +" M ·
+ 2πi T
R 2 s + R
γ1

Thus assuming that R is very large, this part of the integral will be ar-
bitrarily small. Now, cut the integral over γ2 into two pieces, I1 and I2 ,
where " ) *
I1 := 1 F (s)e sT s
1+ 2
2
ds ,
2πi γ2 R s
" ) *
I2 := 1
2
FT (s)esT 1 + s 2 ds ·
s
2πi γ2 R
To find an upper bound on I2 , observe first that FT (s) is entire. The
residue theorem (or actually the Cauchy formula in this case) allows us
to then replace the contour γ2 by the arc of a circle of radius R which
lives in the half-plane Re(s) < 0 and, by using the same upper bounds, to
conclude that |I2 | " M/R. To find an upper bound of I1 , simply notice
) *
that the function F (s)e sT
1+ 2 s2
1 converges to 0 when T tends to
R s
+∞ and converges uniformly on every compact set contained in Re(s) < 0.
Consequently,
" ) *
lim 1 F (s)e sT s
1+ 2
2
ds = 0.
T →∞ 2πi γ2 R s
§5. The Prime Number Theorem 153

By putting the three upper bounds together, we see that

|FT (0) − F (0)| " 2M + ϵ(T )

R
where ϵ(T ) tends to zero (in a way dependent on R). We needed to show
that lim FT (0) = F (0), which is now accomplished. #

Supplement. Analytic continuation and the functional equation

We will now outline the main steps of the proof of the following theorem
due to Riemann.

5.6. Theorem. (The functional equation of the Riemann zeta function)

The function ζ(s)−1/(s−1) can be analytically continued to the whole com-
plex plane. Furthermore, the function ζ(s) satisfies the functional equation
given by
ξ(s) = ξ(1 − s), (4.23)
where ξ(s) := π −s/2 Γ(s/2)ζ(s).

As a preliminary, we will recall the construction of the function Γ(s) and

the Poisson formula which gives the functional equation for the theta series.
!∞
5.7. Lemma. The integral Γ(s) := 0 e−t ts−1 dt defines a holomorphic
function for Re(s) > 0, which satisfies the functional equation Γ(s + 1) =
sΓ(s). It can be continued to all of C as a meromorphic function with
simple poles at 0, −1, −2, −3, . . . .
Proof. Showing that the integral is convergent does not pose any problems.
The functional equation can be obtained by integrating by parts. The
functional equation also allows us to analytically continue by induction from
Re(s) > −n to Re(s) > −n − 1 by using the fact that Γ(s) = s−1 Γ(s + 1).
Finally, the expression

Γ(s) = 1 Γ(s + n + 1)
s(s + 1) · · · (s + n)
makes it clear where the poles are. #

We can also prove that for all s, Γ(s) ̸= 0 (see Exercise 4-6.19).

5.8. Lemma. (Poisson formula) Let f (x) be an integrable function over

R (i.e., in L1 (R)). We define its Fourier transform by
" +∞
ˆ
f (y) := f (x) exp(2πixy)dx
−∞
154 4. Analytic Number Theory

&
and assume that the function n∈Z f (x + n) is of bounded variation on
[0, 1] and continuous. Then the following formula holds:
% %
f (n) = fˆ(m). (4.24)
n∈Z m∈Z
&
Proof. We introduce the function G(x) := n∈Z f (x + n) (the hypotheses
guarantee the existence and continuity of such a function), which is clearly
a periodic function. Dirichlet’s theorem on Fourier series allows us to write
its Fourier expansion as
%
G(x) = Ĝ(m) exp(2πimx),
m∈Z

where the Fourier coefficients can be calculated as follows:

" 1 %" 1
Ĝ(m) := G(t) exp(−2πimt)dt = f (t + n) exp(−2πimt)dt
0 n∈Z 0
" +∞
= f (x) exp(−2πixm)dx = fˆ(−m).
−∞

This gives
% %
f (x + n) = fˆ(m) exp(−2πimx).
n∈Z m∈Z

The Poisson formula follows from that by taking x = 0. #

This formula is most often applied to a function f which is continuously

differentiable and fast decreasing (i.e., f (x) = O(|x|−M ) for all M ), and
therefore the function G is itself continuously differentiable. This is the
case when applying the formula to the following “theta” function.

& ' (
5.9. Corollary. The function2 θ(u) := n∈Z exp −πun2 satisfies the
functional equation for all u ∈ R∗+ given by:
√
θ(1/u) = u θ(u). (4.25)

Proof. It suffices to apply the Poisson formula to the function f (x) =

√
exp(−πux2 ) and to verify that fˆ(y) = exp(−πy 2 /u)/ u. #

Proof. (of Theorem 4 -5.6) We start with the following computation (where
2 Wehope that the context will allow the reader to distinguish this function from the
P
Tchebychev function θ(x) = p"x log p.
§5. The Prime Number Theorem 155

we introduce t = πn2 u) which is valid for Re(s) > 1.

%" ∞
ξ(s) = π −s/2
Γ(s/2)ζ(s) = e−t ts/2 π −s/2 n−s dt
t
n!1 0
⎧ ⎫
" ∞ ⎨% ⎬
= exp(−πun ) us/2 du
2
u
0 ⎩ ⎭
n!1
" ∞ s/2
= θ̃(u) u u du
0

where
% ' ( θ(u) − 1
θ̃(u) := exp −πun2 = ·
2
n!1

Let us point out that θ̃(u) = O(exp(−πu)) when u tends to infinity and
that the functional equation of the function θ can be translated into
2 3 √
θ̃ u 1 = u θ̃(u) + 1 '√u − 1( . (4.26)
2
!∞
By using the simple computation 1 t−s = 1/(s − 1) and the functional
equation of the theta function (4.25), we obtain
" 1 s/2
" ∞ s/2
ξ(s) = θ̃(u) uu du + θ̃(u) u u du
0 1
" ∞ −s/2
" ∞
= θ̃(1/u) u du + θ̃(u) us/2 du
1
u 1
u
" ∞4 " ∞
√ 1 '√ (5 u−s/2 du us/2 du
= uθ̃(u) + u−1 + θ̃(u)
1 2 u 1
u
" ∞ 6 s 1−s 7
= θ̃(u) u + u 2
2 du + 1 − 1 ·
1
u s−1 s

We have a priori obtained the desired expression only for Re(s) > 1,
but we can easily see that the integral defines an entire function since
θ̃(u) = O(exp(−πu)) and since it is symmetric under the transformation
s '→ 1 − s. #

Supplement without proofs

1) To establish the prime number theorem, we could, in the place of the “an-
alytic theorem”, use Ikehara’s theorem [40] (sometimes called the Ikehara-
Wiener theorem), which is more powerful but also more tricky to prove. We
will settle with stating the theorem. Its extension to the case of a multiple
pole was proven by Delange [25].
40 P. COLMEZ

seulement si son terme général tend vers 0), nous avons consacré un chapitre à
leur construction (la construction du corps Cp est un peu plus longue que celle
de C, mais pas beaucoup) et un chapitre à l’analyse sur Zp . Le dernier chapitre
est, quant à lui, consacré à la construction de la fonction zêta p-adique. Le
second chapitre contient beaucoup plus de choses que ce qui est nécessaire pour
cette construction (les mesures suffiraient, mais les distributions deviennent
indispensables pour construire des fonction L p-adiques plus générales, par
exemple celles attachées aux formes modulaires). Pour d’autres points de vue
sur les nombres p-adiques, le lecteur pourra consulter les livres p-adic numbers,
p-adic analysis, and zeta-functions et p-adic analysis : a short course on recent
work de Koblitz ou An introduction to G-functions de Dwork, Gerotto et
Sullivan. L’énoncé précis du théorème de Mazur et Wiles n’est pas donné dans
le texte, pas plus que l’application au théorème de Kummer et nous renvoyons
le lecteur intéressé aux livres Introduction to cyclotomic fields de Washington
ou Cyclotomic fields I and II de Lang (pour le théorème de Kummer, on peut
aussi lire la démonstration dans l’ouvrage de Borevich et Chafarevitch déjà
cité).

CHAPITRE I
LA FONCTION ZÊTA DE RIEMANN

I.1. Prolongement analytique et valeurs aux entiers négatifs

! "
Soit ζ(s) = +∞n=1 n
−s = −s −1 la fonction zêta de Riemann. Soit
p (1 − p )
# +∞
dt
Γ(s) = e−t ts la fonction Γ d’Euler. Cette fonction est holomorphe
t=0 t
pour Re(s) > 0 et satisfait l’équation fonctionnelle Γ(s + 1) = sΓ(s), ce qui
permet de la prolonger en une fonction méromorphe sur C tout entier.

Lemme I.1.1. Si Re(s) > 1, alors

# +∞
1 1 s dt
ζ(s) = t .
Γ(s) 0 et − 1 t

1 !
Démonstration. Il suffit d’écrire t sous la forme +∞
n=1 e
−nt et d’utiliser
# +∞ e − 1
dt Γ(s)
la formule e−nt ts = s .
0 t n
 ARITHMÉTIQUE DE LA FONCTION ZÊTA 41

Proposition I.1.2. Si f est une fonction C ∞ sur R+ à décroissance rapide à

l’infini, alors la fonction
# +∞
1 dt
L(f, s) = f (t)ts
Γ(s) 0 t
définie pour Re(s) > 0 admet un prolongement holomorphe à C tout entier et,
si n ∈ N, alors L(f, −n) = (−1)n f (n) (0).

Démonstration. Soit ϕ une fonction C ∞ sur R+ , valant 1 sur [0, 1] et 0 sur

[2, +∞[. On peut écrire f sous la forme ϕf + (1 − ϕ)f et L(f, s) sous la
forme L(ϕf, s) + L((1 − ϕ)f, s) et comme (1 − ϕ)f est nulle dans un voisinage
# +∞
dt
de 0 et à décroissance rapide à l’infini, l’intégrale f (t)ts définit une
0 t
fonction holomorphe sur C tout entier. Comme de plus, 1/Γ(s) s’annule aux
entiers négatifs, on a L((1 − ϕ)f, −n) = 0 si n ∈ N. On voit donc que, quitte
à remplacer f par (1 − ϕ)f , on peut supposer f à support compact. Une
intégration par partie nous fournit alors la formule L(f, s) = −L(f ′ , s + 1) si
Re(s) > 1, ce qui permet de prolonger L(f, s) en une fonction holomorphe sur
C tout entier. D’autre part, on a
# +∞
L(f, −n) = (−1)n+1 L(f (n+1) , 1) = (−1)n+1 f (n+1) (t)dt = (−1)n f (n) (0),
0
ce qui termine la démonstration.
t
On peut en particulier appliquer cette proposition à f0 (t) = . Soit
!+∞ −1 et
n
n=0 Bn t /n! le développement de Taylor de f0 en 0. Les Bn sont des nombres
rationnels appelés nombres de Bernoulli et qu’on retrouve dans toutes les
branches des mathématiques. On a en particulier
1 1 −1 −691
B0 = 1, B1 = − , B2 = , B4 = , . . . , B12 = ,
2 6 30 2730
et comme f0 (t) − f0 (−t) = −t, la fonction f0 est presque paire et B2k+1 = 0
si k ! 1. Un test presque infaillible pour savoir si une suite de nombres a un
rapport avec les nombres de Bernoulli est de regarder si 691 apparaı̂t dans les
premiers termes de cette suite.

Théorème I.1.3
(i) La fonction ζ a un prolongement méromorphe à C tout entier, holo-
morphe en dehors d’un pôle simple en s = 1 de résidu 1.
Bn+1
(ii) Si n ∈ Q, alors ζ(−n) = (−1)n ; en particulier, ζ(−n) ∈ Q.
n+1
1
Démonstration. On a ζ(s) = s−1 L(f0 , s − 1) comme on le constate en utilisant
la formule Γ(s) = (s − 1)Γ(s − 1) ; on en déduit le résultat.
42 P. COLMEZ

I.2. Valeurs aux entiers positifs pairs

Proposition I.2.1. Si z ∈ C − Z, alors

1 $% 1 1 &
+∞
+ + = πcotg πz.
z z+n z−n
n=1

Démonstration. Notons F(z) la fonction

1 $% 1 1 & 1 $ 2z
+∞ +∞
F(z) = + + = + .
z z+n z−n z z 2 − n2
n=1 n=1

La convergence absolue de la série dans le second membre montre que F(z)

est une fonction méromorphe sur C, holomorphe sur C − Z avec des pôles
simples de résidu 1 en les entiers, que F est impaire et périodique de période 1.
La fonction G(z) = F(z) − πcotg πz est donc holomorphe sur C, impaire et
périodique de période 1.
Si −1/2 " x " 1/2, on a |z 2 − n2 | ! y 2 + n2 − 1/4 et |z| " y + 1/2. On
obtient donc
'$+∞
2z '' $ '' 2z '' $ 2y + 1
+∞ +∞
'
' ' " ' 2 ' "
n=1
z 2 − n2
n=1
z − n2
n=1
y 2 − 41 + n2
# +∞
2y + 1 π 2y + 1
" 2 1 2
dx = · ( .
0 y − 4 +x 2 y2 − 1
4

Comme la fonction πcotg πz est bornée sur |Im(z)| ! 1, il existe c > 0 tel
que |G(z)| " c si z = x + iy et −1/2 " x " 1/2, y ! 1. De plus, G étant
holomorphe, il existe c′ > 0 tel que |G(z)| " c′ si z = x + iy et −1/2 " x "
1/2, 0 " y " 1 et G étant impaire et périodique de période 1, on a alors
|G(z)| " sup(c, c′ ) quel que soit z ∈ C. La fonction G est donc bornée sur C
tout entier, donc constante et nulle car impaire. Ceci permet de conclure.

Maintenant, on a d’une part

+∞ +∞ +∞ +∞
1 $ 2z 1 $ $ z 2k 1 $
+ = − 2z = − 2 ζ(2k)z 2k−1 ,
z z 2 − n2 z n2k+2 z
n=1 n=1 k=0 k=1

et d’autre part,
+∞
e2iπz + 1 2iπ 1$ (2iπz)n
πcotg πz = iπ 2iπz = iπ + 2iπz = iπ + Bn .
e −1 e −1 z n!
n=0

On en déduit le résultat suivant.

 ARITHMÉTIQUE DE LA FONCTION ZÊTA 43

Théorème I.2.2. Si k est un entier ! 1, alors

1 (2iπ)2k
ζ(2k) = − B2k .
2 (2k)!

En particulier, π −2k ζ(2k) est un nombre rationnel.

I.3. Polylogarithmes et valeurs aux entiers positifs

de la fonction zêta
1. Polylogarithmes
Si k est un entier ! 1, on note Lik (z) la fonction définie, pour |z| < 1, par
la formule
+∞ n
$ z
Lik (z) = .
n=1
nk

Ces fonctions Lik , appelées polylogarithmes (Li2 est le dilogarithme, Li3 le

trilogarithme. . .), ont été introduites par Leibnitz, mais n’ont commencé à
jouer un rôle important que depuis une vingtaine d’années ; on s’est aperçu
récemment qu’elles apparaissaient naturellement dans de multiples questions
(volumes de variétés hyperboliques, valeurs aux entiers des fonctions zêtas,
cohomologie de GLn (C)...).
d
On a Li1 (z) = − log(1 − z), et Lik (z) = z −1 Lik−1 (z), ce qui permet
dz
de prolonger analytiquement Lik (z), par récurrence sur k, en une fonction
holomorphe multivaluée sur C − {0, 1}. (On est obligé de supprimer 0 bien
!
que la série +∞ n k
n=1 z /n n’ait pas de singularité en 0 car, après un tour autour
de 1, la fonction Li1 (z) = − ) log(1 − z) augmente de 2iπ et donc vaut 2iπ en 0,
ce qui fait que Li2 (z) = z −1 Li1 (z) a une singularité logarithmique en 0.)
Pour prolonger analytiquement log z et les Lik (z), k ! 1, il faut choisir un
chemin γz dans C − {0, 1} reliant un point fixe (nous prendrons 1/2 comme
point de base de tous nos chemins) à z et poser
#
dt
log z = − log 2 + ,
γz t
#
dt
Li1 (z) = log 2 +
γz 1 −t
+∞
$ 2−n #
dt
et Lik (z) = k
+ Lik−1 (t) .
n=1
n γz t
 ARITHMÉTIQUE DE LA FONCTION ZÊTA 51

Cette formule étant valable quel que soit ϕ, la formule d’inversion de Fourier
!!
ϕ(0) = ϕ(0) nous permet de d’obtenir l’égalité

cn = ζ(n)cn−1

qui permet de conclure.

Remarque I.3.9. En voyant la formule du théorème, on se dit qu’il doit exister

une « décomposition » de SLn en morceaux reflétant la factorisation naturelle
du volume de SLn (R)/SLn (Z). C’est cette vague intuition qui mène à la
K-théorie et aux régulateurs de Borel (cf. démonstration du théorème I.3.7).

I.4. Équation fonctionnelle de la fonction zêta

Soit ξ(s) = π −s/2 Γ(s/2)ζ(s).

Théorème I.4.1. La fonction ξ(s) admet un prolongement méromorphe à C tout

entier, holomorphe en dehors de pôles simples de résidus respectifs −1 et 1 en
s = 0 et s = 1, et vérifie l’équation fonctionnelle

ξ(s) = ξ(1 − s).

Démonstration. Il y a un nombre assez conséquent de méthodes pour arri-

ver au résultat. Nous en donnerons deux dans ce no et une autre au § III.4
(cf. th. III.4.5).

1. Première méthode : la fonction thêta

Si t > 0, on a
" +∞ " +∞
2 −1 2 −1 y)2 −1 y 2
e−πtx e−2iπxy dx = e−πt y e−πt(x+it dx = t−1/2 e−πt .
−∞ −∞
√
(C’est classique ; faire le changement de variables u = t(x + it−1 y), uti-
liser
# +∞ le−πu
théorème des résidus pour revenir sur la droite réelle et la formule
2
−∞ e du = 1 pour conclure.)
$ 2
Si t > 0, soit θ(t) = n∈Z e−πtn . La formule de Poisson
% % " +∞
ϕ(n) = ϕ(x)e−2iπnx dx,
n∈Z n∈Z −∞
2
utilisée pour ϕ(x) = e−πtx , nous fournit l’équation fonctionnelle

θ(t) = t−1/2 θ(t−1 ).

52 P. COLMEZ

On a alors
"
1 +∞ dt
ξ(s) = (θ(t) − 1)ts/2
2 0 t
" "
1 1 −1/2 −1 dt 1 +∞ dt
= (t θ(t ) − 1)ts/2 + (θ(t) − 1)ts/2
2 0 t 2 1 t
et on peut changer t en t−1 dans la première intégrale pour obtenir
"
1 1 1 +∞ dt
ξ(s) = − + + (θ(t) − 1)(ts/2 + t(1−s)/2 ) .
s s−1 2 1 t
On en déduit le résultat car θ(t) − 1 est à décroissance rapide à l’infini, ce qui
implique que l’intégrale est une fonction holomorphe de s sur C tout entier,
et le membre de droite est évidemment invariant par s #→ 1 − s.

2. Deuxième méthode : intégrale sur un contour

Si c > 0, soit Cc le contour obtenu en suivant la droite réelle de +∞ à cπ,
puis en parcourant le carré de sommets cπ(±1±i) dans le sens trigonométrique,
et en retournant en +∞ le long de l’axe réel. Soit
"
1 dz
Fc (s) = z
(−z)s ,
Cc e − 1 z
où (−z)s = exp(s log(−z)) et la branche du logarithme choisie est celle dont
la partie imaginaire est comprise entre −π et π ; en particulier, on a (−z)s =
e−iπs z s de +∞ à cπ et (−z)s = eiπs z s de cπ à +∞ (après avoir parcouru le
carré). Comme ez1−1 est à décroissance rapide à l’infini, la fonction Fc (s) est
holomorphe sur C pour tout c qui n’est pas un entier pair (pour éviter les pôles
de ez1−1 ). D’autre part, le théorème des résidus montre que Fc (s) ne dépend
pas de c si c reste dans un intervalle du type ]2N, 2N + 2[, avec N ∈ N. En
particulier, on a F1 (s) = Fc (s) quel que soit c ∈ ]0, 2[. Si Re(s) > 1, quand c
tend vers 0, l’intégrale sur le carré de sommets cπ(±1 ± i) tend vers 0, et on
obtient, en passant à la limite
" 0 " +∞
1 s dz 1 dz
F1 (s) = e−iπs z −1
z + eiπs
z −1
zs = 2i · sin πs · Γ(s) · ζ(s).
+∞ e z 0 e z
Maintenant, quand N tend vers +∞, la fonction F2N+1 (s) tend vers 0 quand
Re(s) < 0 car ez1−1 est majorée, indépendamment de N, sur C2N+1 . La
différence entre F1 (s) et F2N+1 (s) peut se calculer grâce au théorème des
résidus. La fonction ez1−1 (−z)s−1 a des pôles en z = ±2iπ, ±4iπ, . . . , ±2Niπ
dans le contour délimité par la différence entre C2N+1 et C1 . Si k ∈ {1, . . . , N},
le résidu de ez1−1 (−z)s−1 est (2kπ)s−1 e−iπ(s−1)/2 en 2ikπ et (2kπ)s−1 eiπ(s−1)/2
 ARITHMÉTIQUE DE LA FONCTION ZÊTA 53

en −2ikπ, ce qui nous donne

N
1 s−1 %
(F2N+1 (s) − F1 (s)) = 2 cos π · (2kπ)s−1 .
2iπ 2
k=1

En passant à la limite, on obtient donc F1 (s) = 4iπ·cos(π s−1

2 )·(2π)
s−1 ·ζ(1−s),

si Re(s) < 0. On en déduit l’équation fonctionnelle

s−1
sin πs · Γ(s) · ζ(s) = cos π · (2π)s · ζ(1 − s).
2
Pour passer de cette équation fonctionnelle à celle de ξ, il faut utiliser les
formules classiques
π &s' &s + 1' &1' &1' √
Γ(s)Γ(1 − s) = , Γ Γ = 21−s Γ Γ(s), et Γ = π.
sin πs 2 2 2 2

I.5. Fonctions L de Dirichlet

1. Caractères de Dirichlet et sommes de Gauss
Si d est un entier, on appelle caractère de Dirichlet modulo d un morphisme
de groupes de (Z/dZ)∗ dans C∗. L’image d’un caractère de Dirichlet est bien
évidemment incluse dans le groupe des racines de l’unité.
Si d′ est un diviseur de d et χ est un caractère de Dirichlet modulo d′ , on
peut aussi voir χ comme un caractère de Dirichlet modulo d en composant χ
avec la projection (Z/dZ)∗ → (Z/d′ Z)∗. On dit que χ est de conducteur d si
on ne peut pas trouver de diviseur d′ de d distinct de d, tel que χ provienne
d’un caractère modulo d′ . De manière équivalente, χ est de conducteur d si
quel que soit d′ diviseur de d distinct de d, la restriction de χ au noyau de la
projection (Z/dZ)∗ → (Z/d′ Z)∗ n’est pas triviale.
Si χ est un caractère de Dirichlet modulo d, on note χ−1 le caractère de
Dirichlet modulo d défini par χ−1 (n) = (χ(n))−1 si n ∈ (Z/dZ)∗.
Si χ est un caractère de Dirichlet modulo d, on considère aussi souvent χ
comme une fonction périodique sur Z de période d en composant χ avec la
projection naturelle de Z sur Z/dZ et en étendant χ par 0 sur les entiers non
premiers à d. On a donc χ−1 (n) = (χ(n))−1 si (n, d) = 1, mais χ−1 (n) = 0 si
(n, d) ̸= 1.
Si d est un entier, χ un caractère de Dirichlet de conducteur d et si n ∈ Z,
on définit la somme de Gauss tordue G(χ, n) par la formule
%
G(χ, n) = χ(a)e2iπna/d
a mod d

et on pose G(χ) = G(χ, 1).