Professional Documents
Culture Documents
Advisors:
Jury:
Master in Mathematics
Department of Mathematics
Colombia
APROBADO:
Jhon Jairo Sutachan Rubio, Ph.D. Alba Alicia Trespalacios Rangel, Ph.D.
Contents
1 Preliminaries 12
1.1 Some Theorems from Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Bernoulli Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4 Introduction to Asymptotic Analysis . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.1 Origin of Asymptotic Expansions . . . . . . . . . . . . . . . . . . . . . . 24
1.4.2 Asymptotic Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4.3 Asymptotic Expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.5 The Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.5.1 The Mellin Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2 Special Functions 33
2.1 The Gamma Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.1.1 Euler’s Limit Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.1.2 Infinite Product Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.1.3 The Reflection Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.1.4 Gauss’ Multiplication Formula . . . . . . . . . . . . . . . . . . . . . . . . 37
2.1.5 Hankel’s Loop Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.6 The Bohr-Mollerup Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2 The Digamma Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3 The Exponential, Logarithmic, Sine, and Cosine Integrals . . . . . . . . . . . . . 44
2.3.1 The Exponential and Logarithmic Integral . . . . . . . . . . . . . . . . . 44
2.3.2 The Sine and Cosine Integrals . . . . . . . . . . . . . . . . . . . . . . . . 46
2.4 Arithmetic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
1
3 The Riemann Zeta Function 53
3.1 Analytic Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 The Functional Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3 The Euler Product Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4 The Hadamard Product Formula over the Zeros . . . . . . . . . . . . . . . . . . 59
3.5 The Asymptotic Formula for N (T ) . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.6 Hardy’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.7 The Explicit Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.7.1 The Zero-free Region and the PNT . . . . . . . . . . . . . . . . . . . . . 78
4 Pontryagin Duality 81
4.1 The Haar Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.1.1 Topological Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.1.2 A Quick Tour of Measure Theory . . . . . . . . . . . . . . . . . . . . . . 83
4.1.3 Existence of a Haar Measure . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.1.4 Uniqueness of a Haar measure . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2 Banach Algebras and The Spectral Theorem . . . . . . . . . . . . . . . . . . . . 95
4.2.1 Banach Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.2.2 Quotient Algebras and the Gelfand Transform . . . . . . . . . . . . . . . 108
4.2.3 C ⇤ -algebras and the Spectral Theorem . . . . . . . . . . . . . . . . . . . 111
4.2.4 Some Representation Theory . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.3 The Pontryagin Dual . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.3.1 Positive Definite Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.3.2 The Fourier Inversion Formula . . . . . . . . . . . . . . . . . . . . . . . . 128
4.3.3 Pontryagin Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
2
5.2 Local Zeta-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
5.2.1 Analysis on Local Additive Groups . . . . . . . . . . . . . . . . . . . . . 157
5.2.2 Analysis on Local Multiplicative Groups . . . . . . . . . . . . . . . . . . 164
5.2.3 The Local Functional Equation . . . . . . . . . . . . . . . . . . . . . . . 167
5.3 The Functional Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
5.3.1 The Riemann–Roch Theorem . . . . . . . . . . . . . . . . . . . . . . . . 175
5.3.2 The Global Functional Equation . . . . . . . . . . . . . . . . . . . . . . . 180
5.3.3 Hecke L -Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.4 What’s Next? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
5.4.1 More General L-functions . . . . . . . . . . . . . . . . . . . . . . . . . . 195
3
Acknowledgments
The completion of this thesis could not have been possible without the participation and as-
sistance of so many people whose names may not all be enumerated. Their contributions are
sincerely appreciated and gratefully acknowledged. However, I would like to express my deep
appreciation particularly to the following:
To my thesis advisor Johan Bogoya, for his constant aid with my work during all stages, thank
you for teaching me so many things, you taught me most of my courses during the career, you
are one of the best teachers I’ve ever known.
To my thesis advisor Jorge Plazas, for guiding me throughout this work, you always knew what
to do next and gave my motivation to do everything. Thank you for providing me with your
valuable experience and wisdom.
To my parents and my sister, for their constant support, I cannot thank you enough for all the
things they have done, with tender little things such as making me breakfast every morning
during the week, picking me up from the bus stop, inviting me to eat after a hard day of work,
and so on.
To my friend Diana Gutiérrez for being there with me whenever I needed her and for encouraging
me when times got rough, thank you for helping me to understand so many things and see the
life in a di↵erent way, I cannot thank you enough for all the things you’ve done for me.
To all relatives, friends and others who in one way or another shared their support, either
morally, financially and physically, thank you.
4
Introduction
In the eighteen century, the swiss mathematician Leonhard Euler (1707–1783) introduced the
zeta function defined for s 2 R as
X1
1
⇣(s) = s
.
n=1
n
In elementary courses of calculus, one of the first examples of an infinite series is the one given
by ⇣(s). Using the integral test the student learns that
X1
1
n=1
ns
converges for s > 1 and diverges if s 6 1. Some enthusiastic teachers will point out the fact
that 1
X 1 ⇡2
⇣(2) = 2
= .
n=1
n 6
This is known as the Basel problem, and it was solved by Euler in 1734. Euler’s proof depended
on some assumptions that are rather difficult to justify. For example, at a key point in the
solution, Euler observed that the function
sin x
x
and the infinite product
⇣ x ⌘⇣ x ⌘⇣ x ⌘⇣ x⌘
1+ 1 1+ 1 ··· (0.0.1)
⇡ ⇡ 2⇡ 2⇡
have exactly the same roots and have the same value at x = 0, so Euler asserted that they
describe the same function. As we shall see (Corollary 3.4.1.1) Euler was right but the reasons
just mentioned are insufficient to guarantee it. For example, the function
sin x
ex
x
also has the same roots and the same value at x = 0, but it is a di↵erent function. It was
not until 100 years later, that Karl Weirstrass proved that Euler’s representation of the sine
function as an infinite product was valid, by the Weirstrass factorization theorem. Euler’s
approach to the Basel problem goes as follows: If we formally multiply the product in (0.0.1)
we find out that the coefficient of x2 is
⇣1 1 1 ⌘ 1 X 1
1
+ + + ··· = .
⇡ 2 4⇡ 2 9⇡ 2 ⇡ 2 n=1 n2
5
But from the Taylor expansion of sin x around x = 0 we know that
sin x x2 x4 x6
=1 + + ··· ,
x 3! 5! 7!
1
and hence the coefficient of x2 is also . This shows that
6
X1
1 ⇡2
2
= .
n=1
n 6
Euler used the same approach to find an expression for ⇣(2m) for m 2 N, namely
(2⇡)2m ( 1)n+1 B2m
⇣(2m) = , m = 1, 2, . . . ,
2(2m)!
where B2m are the Bernoulli numbers which will be introduced in Section 1.3. This result not
only provides an elegant formula for evaluating ⇣(2m), but it also tells us the arithmetic nature
of ⇣(2m). In contrast, we know very little about the odd zeta values ⇣(2m + 1). It is known
that ⇣(3), the Apéry’s constant, is irrational (Apéry 1978) and that there are infinitely many
m 2 N such that ⇣(2m + 1) is irrational (Rivoal 2000); for a proof of these facts see [9]. Even so,
we don’t know the nature of ⇣(2m + 1); for instance we don’t know whether ⇣(5) is irrational
or not.
After Euler achieved his objective of evaluating ⇣(2), he then turned to the arithmetic properties
of ⇣(s). In 1737 he published a paper entitled Variae observationes circe series infinitas (Various
observations about infinite series). Here for the first time he proved the famous Euler product
formula in the form Y 1
⇣(s) = s
(s > 1), (0.0.2)
p2P
1 p
where P denotes the set of prime numbers. This result is fascinating because it shows that the
zeta function is related to the prime numbers. Euler then used this result to prove that the
series X1
p2P
p
diverges. Euler’s proof goes as follows: by taking logarithms to each side of (0.0.2) we obtain
X
log ⇣(s) = log(1 p s ).
p2P
Since 1
X xn
log(1 x) = (|x| < 1),
n=1
n
and |p s | < 1, then
X 1 X 1 ⇣1 1 1 ⌘
log ⇣(s) = + + + + · · ·
p2P
ps p2P p2s 2 3ps 4p2s
X 1 X 1
< s
+ s s
p2P
p p2P
p (p 1)
X 1 X 1
1
< s
+ 2s
.
p2P
p n=1
n
6
Since the harmonic series diverges, letting s ! 1 we conclude that
X1
p2P
p
⇣(n) = N ⇡ n ,
and said that if n is even, then N is rational, while if n is odd he conjectures that N is a
function of log 2. In the middle of his paper, Euler states that
1 3 + 5 7 + · · · = 0,
1 3 + 53 73 + · · · = 0,
3
1 22 + 32 42 + · · · = 0,
1 24 + 34 44 + · · · = 0,
1 26 + 36 46 + · · · = 0.
d 1
x f (x) = x + 2x2 + 3x3 + · · · = ;
dx (1 x)2
1 22 + 32 42 + · · · = 0.
7
Since the series converges at each stage of this process for |x| < 1, we see that Euler anticipated
Abel summability1 by some 75 years.
In 1749 he gave a paper to the Berlin Academy entitled Remarques sur un beau rapport entre
les séries des puissances tant directes que réciproques(Remarks on a beautiful relation between
direct as well as reciprocal power series). In his papers he considers the function
X1
( 1)n+1
⌘(s) = ,
n=1
ns
and using the methods described before, and the formulas for ⇣(2m) he notes that
12 22 + 3 2 42 + 52 62 + · · ·
1 = 0,
1 23
+ 313 1
4 3 + 5
1
3
1
6 3 + · · ·
14 23 + 3 3 43 + 53 63 + · · · 1 · 2 · 3 · (24 1)
1 = ,
1 24
+ 314 1
44
+ 514 614 + · · · (23 1)⇡ 4
14 24 + 3 4 44 + 54 64 + · · ·
1 = 0,
1 25
+ 315 1
45
+ 515 615 + · · ·
or if m > 2, 8 m
is true for all s”. Here is the Gamma function which will be discussed in Section 2.1. Then,
Euler continue saying: “The validity of our conjecture for s = 1 (which appeared to be deviated
from the others) is already a strong justification to our conjecture, since it appears unlikely
that a false assumption could support this case. Therefore, we can regard our conjecture as
being solidly based but I shall give other justifications which are equally convincing”. Euler
2k + 1
then proceeds to check the formula for s = . Note that
2
⌘(s) = (1 21 s )⇣(s).
1
P
Let P(an )n2N be a sequence of real or complex numbers. We say that n2N an is Abel summable if
n
limx!1 a
n2N n x exists and is finite. For example 1 2+3 4+· · · is Abel summable since x+2x2 +3x3 +· · ·
converges for every |x| < 1.
8
Therefore, using (0.0.4) we can write
⇣ s⇡ ⌘
⇣(1 s) = 21 s ⇡ s
cos (s)⇣(s),
2
and this is the unsymmetrical form of the famous functional equation proved by Riemann in
1859. The function ⇣(s) defined by Euler is nowadays called the Riemann zeta function because
the german mathematician Bernhard Riemann (1826–1866) was the first to study extensively
its properties. In 1859, he wrote a short paper (8 pages) called Ueber die Anzahl der Primzahlen
unter einer gegebenen Grösse (On the number of primes less than a given magnitude) in which
he expressed fundamental properties of ⇣(s) in the complex variable s = + it. We state these
in the modern nomenclature.
1. The Riemann zeta function ⇣(s) extends to a meromorphic function on the whole complex
plane with a simple pole at s = 1, and the function
1 s
⇣s⌘
⇠(s) := s(s 1)⇡ 2 ⇣(s)
2 2
satisfies the functional equation
⇠(s) = ⇠(1 s)
for every s 2 C.
2. ⇣(s) has simple real zeros at s = 2, 4, 6, . . . , which are called the trivial zeros, and
infinitely many non-trivial zeros of the form
⇢= +i , 06 6 1, 2 R.
T T T
N (T ) = log + O(log T ).
2⇡ 2⇡ 2⇡
This was proved by von Mangoldt in 1905.
where A and B are constants and ⇢ runs through the non trivial zeros of ⇣ in the critical
strip. This was proved by Hadamard in 1893. It played an important role in the proofs
of the prime number theorem by Hadamard and de la Vallée Poussin.
9
Then for x > 1 not a prime power we have
X x⇢ 1 ⇣ 1⌘
(x) = x log 2⇡ log 1 ,
⇢
⇢ 2 x2
5. The Riemann hypothesis, one of the most important unsolved problems in mathematics:
Conjecture (The Riemann Hypothesis). Every non trivial zero of ⇣(s) is on the line
1
Re(s) = .
2
We can give an sketch of the proof of the functional equation (see Chapter 3 for details): by
using Fourier analysis, more specifically, Poisson summation formula one can show that the
Jacobi theta function ✓ : (0, 1) ! R defined by
X 2
✓(x) = e ⇡n x ,
n2Z
By the properties of the theta function and the Gamma function this equation not only gives
analytic continuation of ⇣(s) to C with the exception of a simple pole at s = 1 but also by
observing that the term in brackets is invariant under the transformation s 7! 1 s, we also
have the functional equation
s
⇣s⌘ s 1
⇣1 s⌘
⇡ 2 ⇣(s) = ⇡ 2 ⇣(1 s).
2 2
The Riemann zeta function can be generalized: let K be a number field with ring of integers
OK (for notation see Chapter 5). The Dedekind zeta function is defined as
X 1
⇣K (s) = ,
a✓OK
(N a)s
where the sum runs over all the nonzero ideals of OK . In 1918-1920 the German mathematician
Erich Hecke was able to establish the analytic continuation and functional equation of ⇣K (s)
by using a complicated generalized theta function and the higher analogues of (0.0.6). More
specifically, he showed that the Dedekind zeta funcion has a meromorphic continuation on C
and satisfies the functional equation (Theorem 5.4.2)
1
r1 r2 s
R (s) C (s) ⇣K (s) = |dK | 2 R (1 s)r1 C (1 s)r2 ⇣K (1 s),
10
where r1 is the number of real embeddings of K, r2 the number of conjugate pairs of complex
embeddings of K (so that [K : Q] = r1 + 2r2 ), dK is the discriminant of K,
s s
R (s) =⇡ 2 and C (s) = (2⇡)1 s
(s).
2
Hecke then quickly realized that his method applied also to Dirichlet L functions and even
more general L functions.
A few years later, Emil Artin generalized the functional equation to Artin L functions, and
his student Margaret Matchett, in 1946 wrote her Ph.D thesis on interpreting zeta functions as
a product over idèles. Afterwards, in 1950, John Tate in his Ph.D thesis [33], completed under
the supervision of Emil Artin, building on Matchett’s thesis was able to reprove the analytic
continuation and functional equation for the Hecke L functions using adèlic methods. More
specifically, by using an analogue of Poisson Summation Formula in which Z ✓ R is replaced by
K ✓ AK , where K is a global field and AK its adèle ring, and doing analysis on the local fields
K⌫ , where ⌫ is a place on K, Tate was able to prove that Hecke L functions can be extended
to meromorphic functions and they satisfy some functional equation (Theorem 5.3.6).
The first three chapters of this document cover the classical theory of the Riemann zeta function
with all the prerequisites necessary to understand Riemann’s memoir. Afterwards, we develop
the necessary tools to understand Tate’s thesis such as integration and Fourier analysis on
locally compact abelian groups. The last chapter is devoted to Tate’s thesis and the last
section covers a brief introduction to the analytic theory of more general L functions.
11
Chapter 1
Preliminaries
When manipulating series and integrals we often encounter problems related with interchanging
summation and integration, and di↵erentiation of integrals with respect to a parameter. In this
section we quote some tools that are frequently used in analysis.
The first theorem gives the condition to justify what is usually named interchanging summation
and integration. It is called the theorem of dominated convergence of Lebesgue in the setting of
Riemann integrals. A proof can be found for example in [3].
Theorem 1.1.1 (Dominated convergence theorem of Lebesgue). Let {fn (t)}n2N be a sequence
of complex-valued functions which are continuous in (a, b) ✓ R and have the properties
P1
(i) n=1 fn (t) converges uniformly in any compact interval in (a, b).
Then Zb X
1 1 Zb
X
fn (t) dt = fn (t) dt.
a n=1 n=1 a
The second theorem is the general version of partial integration, the proof runs by induction
on m and the details are left to the reader.
Theorem 1.1.2 (Integration by parts). Let h 2 C(↵, ) and g 2 C m (↵, ). Then, for every
m2N
Z m
X1 Z
k k (k+1) m
g(x)h(x) dx = ( 1) g (x)h (x) + ( 1) g m (x)h m
(x) dx,
↵ k=0 ↵ ↵
m
where h denotes the mth integral of h.
12
The third theorem is an extension to complex variables of a standard theorem concerning
di↵erentiation of an integral over an infinite contour with respect to a parameter; for a proof
see for example [7].
Theorem 1.1.3. Let t be a real variable ranging over a finite or infinite interval (a, b) and
z a complex variable ranging over a domain ⌦. Assume that the function f : ⌦ ⇥ (a, b) ! C
satisfies the following conditions:
(ii) For each fixed value of t, f (·, t) is a holomorphic function of the first variable.
Then F is holomorphic in ⌦, and its derivatives of all orders may be found by di↵erentiating
under the integral sign.
This section is based on [23]. We know that an analytic function f can be represented by a
power series
X1
f (n) (a)
f (x) = cn (x a)n where cn = (1.2.1)
n=0
n!
for all values of x within the radius of convergence of the series. In this section we shall be
interested in functions which may not be smooth, so that f may not be written in the form
(1.2.1). To obtain representations of non smooth functions we need expansions in terms of
trigonometric functions.
and Z⇡
cos mx sin nx dx = 0 for m, n = 1, 2, . . . . (1.2.4)
⇡
13
Theorem 1.2.1. Let f 2 C[ ⇡, ⇡]. Suppose that
X1
1
a0 + (an cos nx + bn sin nx)
2 n=1
Proof. Let
Xk
1
sk (x) = a0 + (am cos mx + bm sin mx).
2 m=1
Since sk converges uniformly to f as k ! 1, then for each fixed n 2 N, sk (x) cos nx converges
uniformly to f (x) cos nx as k ! 1 . Note that
Since the series converges uniformly we use Theorem 1.1.1 to integrate term by term between
⇡ and ⇡. Then, using (1.2.3) and (1.2.4) we get
Z⇡
f (t) cos nt dt = ⇡an .
⇡
The numbers an and bn are called the Fourier coefficients of f . The series (1.2.2) is called
the Fourier series of f . Let f be any integrable function defined on [ ⇡, ⇡], then we can
always define an and bn by (1.2.5). However, this doesn’t guarantee that the Fourier series of
f converges pointwise to f . We shall see later a condition that implies the convergence of the
Fourier series.
(i) There exists a partition P = {a = x0 < x1 < · · · < xn = b} such that f is continuous on
each subinterval (xk 1 , xk ).
We denote the set of piecewise continuous functions on [a, b] by CPW [a, b].
14
Thus a function f 2 CPW [a, b] has a finite number of discontinuities at x0 , x1 , . . . , xn . In each
of this points the lateral limits
lim f (x) and lim f (x)
x!xk x!x+
k
Returning to the variable x and the function f we get the formulas for the coefficients an , bn of
the modified series:
Z Z
1 L n⇡x 1 L n⇡x
an = f (x) cos dx and bn = f (x) sin dx.
L L L L L L
Therefore 1 ✓ ◆
1 X n⇡x n⇡x
a0 + an cos + bn sin
2 n=1
L L
is a Fourier series for f on the interval [ L, L].
15
1
Example 1.2.1. Let f (x) = x be defined on the interval [0, 1]. We extend f as an odd
2
function, as shown in Figure 1.1. Then f is standardized so that fe( 1) = fe(0) = fe(1) = 0.
Since fe is odd, we have an = 0. Also,
Z1 ✓ ◆
1 1 + cos ⇡n
bn = 2 x sin n⇡x dx = .
0 2 ⇡n
Thus 1 1
X 1 + cos ⇡n X sin 2⇡nx
sin ⇡nx =
n=1
⇡n n=1
⇡n
1
is a Fourier series for x . We shall see later that this series converges to f on the interval
2
[0, 1].
0.5
fe(x)
0.5
1 0 1 2
x
It is important to establish a simple criteria determining when a Fourier series converges point-
wise, we will show how to obtain large classes of functions with the property that for each value
of x in the domain of a function f , the Fourier series converges to f (x). We begin with the
following theorem.
X1
1
a0 + (an cos nx + bn sin nx)
2 n=1
Proof. Let
Xn
1
sn (t) = a0 + (ak cos kt + bk sin kt).
2 k=1
16
Now write
Z⇡ Z⇡ Z⇡ Z⇡
2 2
(f (t) sn (t)) dt = f (t) dt 2 f (t)sn (t) dt + s2n (t) dt.
⇡ ⇡ ⇡ ⇡
Also, by multiplying out the terms of s2n (t) and using (1.2.3) and (1.2.4) we have
Z⇡ Z⇡ Z⇡ X
n
1 2
s2n (t) dt = a dt + (a2k cos2 kt + b2k sin2 kt) dt
⇡ ⇡ 4 0 ⇡ k=1
n
X
1
= a20 ⇡ + (a2k ⇡ + b2k ⇡)
2 k=1
Z⇡
= f (t)sn (t) dt.
⇡
Therefore Z⇡ Z⇡ n
1 X
06 (f (t) sn (t)) dt = 2 2
f (t) dt ⇡ a0 + (a2k + b2k ) .
⇡ ⇡ 2 k=1
Bessel’s inequality shows that an and bn tend to zero as n ! 1 for any function that is square
integrable on [ ⇡, ⇡].
For each n 2 N, we define the Dirichlet kernel Dn as
sin (n + 12 )x
Dn (x) = .
2 sin x2
17
Lemma 1.2.1. Let f 2 CPW [ ⇡, ⇡] be a periodic function with period 2⇡, and let
Xn
1
sn (x) = a0 + (ak cos kx + bk sin kx).
2 k=1
Then
Z⇡
1 1
sn (x) [f (x+ ) + f (x )] = [f (x + u) f (x+ )] Dn (u) du
2 ⇡ 0
Z
1 ⇡
+ [f (x u) f (x )] Dn (u) du.
⇡ 0
Proof. We have
X n
1
sn (x) = a0 + (ak cos kx + bk sin kx)
2 k=1
Z⇡ n
1 1 X
= f (t) + (cos kt cos kx + sin kt sin kx) dt
⇡ ⇡ 2 k=1
Z n
1 ⇡ 1 X
= f (t) + cos k(t x) dt.
⇡ ⇡ 2 k=1
Since f and Dn are periodic functions with period 2⇡, we may change the interval of integration
to ⇡ 6 u 6 ⇡. Then
Z Z
1 0 1 ⇡
sn (x) = f (x + v) Dn (v) dv + f (x + u) Dn (u) dv.
⇡ ⇡ ⇡ 0
to get
Z⇡
1 1
sn (x) [f (x+ ) + f (x )] = [f (x + u) f (x+ )] Dn (u) du
2 ⇡ 0
Z
1 ⇡
+ [f (x u) f (x )] Dn (u) du. ⌅
⇡ 0
Now we are ready to give a simple criteria determining when a Fourier series converges, first
we need the following definition.
18
Definition 1.2.3. Let f : [a, b] ! R be a function. We say that f is piecewise smooth on [a, b]
if and only if
(ii) With the notation of Definition 1.2.2, f 0 exists in (xk 1 , xk ) and f 0 2 C(xk 1 , xk ).
The following theorem gives a sufficient condition for the convergence of a Fourier series.
Theorem 1.2.3. Let f be a piecewise smooth function on [ ⇡, ⇡], standardized, and periodic
with period 2⇡. Then the Fourier series of f converges point-wise for all x 2 [ ⇡, ⇡].
where
Z⇡ Z⇡
1 + 1
In (x) = [f (x + u) f (x )] Dn (u) du and Jn (x) = [f (x u) f (x )] Dn (u) du.
⇡ 0 ⇡ 0
Then
Z
1 ⇡ f (x u) f (x )
Jn (x) = u sin (n + 12 )u du
⇡ 0 2 sin 2
Z⇡
1 f (x u) f (x ) u u
= u (sin nu cos + cos nu sin ) du.
⇡ 0 2 sin 2 2 2
Let
f (x u) f (x ) u 1
g1 (x, u) = u cos and g2 (x, u) = [f (x u) f (x )].
2 sin 2 2 2
By L’Hôpital’s rule,
g1 (x, 0+ ) = f 0 (x ).
Since f is piecewise smooth, g1 and g2 are piecewise continuous functions, they are integrable.
Now write Jn in the form
Z
1 ⇡
Jn (x) = (g1 (x, u) sin nu + g2 (x, u) cos nu) du = ân + b̂n .
⇡ 0
1 1
where ân and b̂n are the nth Fourier coefficients of g and g respectively. By Bessel’s
2 1 2 2
inequality, ân , b̂n ! 0 as n ! 1. Thus Jn (x) ! 0 as n ! 1. Similarly Jn (x) ! 0 as
n ! 1. Therefore
sn (x) f (x) ! 0. ⌅
n!1
19
So far we’ve been working with the Fourier series of real-valued functions, but we can extend
the results of this section to complex-valued functions as follows: Let f : [ ⇡, ⇡] ! C be a
function. Then f (t) = u(t) + iv(t), where u, v are real-valued functions. Let
X1 X1
1 1
a0 + (an cos nx + bn sin nx) and ↵0 + (↵n cos nx + n sin nx)
2 n=1
2 n=1
be the Fourier series of u and v respectively. Naturally we define the Fourier series of f to be
X1
1
(a0 + i↵0 ) + [(an + i↵n ) cos nx + (bn + i n ) sin nx].
2 n=1
For each n 2 Z define Z
1 ⇡
cn = f (t) e int dt.
2⇡ ⇡
Using the definition of an , bn , ↵n and n , it is easy to see that
1 1
cn = (an + i↵n ) + ( n ibn ).
2 2
Therefore
X 1 X1 1 1 1
cn einx = (a0 + i↵0 ) + an cos nx + i↵n cos nx + n cos nx ibn cos nx
n2Z
2 n2Z
2 2 2 2
n6=0
1 1 1 1
+ ian sin nx ↵n sin nx + i n sin nx + bn sin nx.
2 2 2 2
Using the fact that if g is an even function and h is an odd function then
X X1 X
g(n) = 2 g(n) and h(n) = 0,
n2Z n=1 n2Z
n6=0 n6=0
we obtain
X X1
inx 1
cn e = (a0 + i↵0 ) + [(an + i↵n ) cos nx + (bn + i n ) sin nx].
n2Z
2 n=1
20
1 1 1
and so on. Bernoulli had a particular interest in the coefficients of n, that is , , 0, , . . ..
2 6 30
Euler called this numbers the Bernoulli numbers B̂1 , B̂2 , B̂3 , B̂4 , . . . . The modern definition of
the Bernoulli numbers uses a generating function.
1
For example, a generating function for the constant sequence 1, 1, 1, . . . is F (x) = , since
1 x
1
X
1
= xn , |x| < 1.
1 x n=0
Definition 1.3.2. The Bernoulli numbers are the sequence of real numbers {Bn }n2N defined
by the generating function
X1
z zn
= B n , |z| < 2⇡.
ez 1 n=0 n!
We will show (Theorem 1.3.1) that the modern definition agrees with Euler’s definition, that is
B̂n = Bn , n > 1.
Proof. We have
z z
f (z) =
coth 1,
2 2
since coth is an odd function the result follows. ⌅
z
By expanding the function as a power series, we obtain the first Bernoulli numbers:
ez 1
1 1
B0 = 1, B1 = , B2 = .
2 6
Therefore, by Lemma 1.3.1 all Bernoulli numbers with odd index greater than 2 vanishes, that
is
B2n+1 = 0, n > 1.
Definition 1.3.3. The Bernoulli polynomials are defined by the generating function
X 1
z exz zn
= B n (x) , |z| < 2⇡. (1.3.1)
ez 1 n=0 n!
21
The first Bernoulli polynomials are
B0 (x) = 1,
1
B1 (x) = x ,
2
1
B2 (x) = x2
x+ .
6
By definition of the Bernoulli polynomials, note that
Bn (0) = Bn . (1.3.2)
P
Now we will show that the Bernoulli numbers appear in the sum ni=01 ip , for this purpose we
need some properties of the Bernoulli numbers and the Bernoulli polynomials.
We have Z1
z exz ez 1
z
dx = = 1.
0 e 1 e z 1 ez 1
Now we use (1.3.1) and Theorem 1.1.1 to interchange summation and integration. Thus
X 1 ✓ Z1 ◆ n
z
Bn (x) dx = 1,
n=0 0 n!
and so Z1
Bn (x) dx = 0, n > 1. (1.3.3)
0
Since 1
X n+1
xz nz
ze = x ,
n=0
n!
we obtain
Bn (x + 1) Bn (x) = nxn 1 . (1.3.5)
For the next lemma, we recall how to multiply two series:
✓X1 ◆✓ X
1 ◆ X1 n
X
n n n
an z bn z = cn z where cn = aj b n j . (1.3.6)
n=0 n=0 n=0 j=0
22
Lemma 1.3.2. Let x, y 2 R, then
n ✓ ◆
X n
Bn (x + y) = Bk (x)y n k .
k=0
k
Proof. We have
1
X ✓ ◆ ✓X
1 ◆✓ X
1 ◆
zn z exz yz Bn (x) n yn n
Bn (x + y) = e = z z .
n=0
n! ez 1 n=0
n! n=0
n!
Note the similarity between Lemma 1.3.2 and the Binomial theorem:
Xn ✓ ◆
n n k n k
(x + y) = x y .
k=0
k
Now we use the identity Bp+1 (0) = Bp+1 and (1.3.7) to obtain
n 1
X X
p+1 ✓ ◆ p ✓ ◆
1 p+1 1 X p+1
p
i = Bk xp+1 k
Bp+1 = Bk np+1 k . ⌅
i=0
p + 1 k=0 k p + 1 k=0 k
23
Property (1.3.8) can be used to prove the following theorem
Theorem 1.3.2. Let n 2 N, then for every x 2 [0, 1] we have
X1
n+1 sin(2⇡mx)
B2n 1 (x) = 2( 1) (2n 1)! 2n 1
. (1.3.10)
m=1
(2⇡m)
Proof. The proof runs by induction on n. For n = 1, we use Example 1.2.1 to obtain
1
1 X sin 2⇡nx
B1 (x) = x = , 0 6 x 6 1.
2 n=1 ⇡n
Integrating twice the above equation and using Theorem 1.1.1 to interchange summation and
integration we get
X1
sin(2⇡mx)
n+2
B2n+1 (x) = 2( 1) (2n + 1)! 2n+1
. ⌅
m=1
(2⇡m)
The special case x = 0 in (1.3.11) gives an interesting result for the Bernoulli numbers:
1
X
n+1 1
B2n = 2( 1) (2n)! 2n
. (1.3.12)
m=1
(2⇡m)
It is of interest, since with this result we can express the Riemann zeta function (which will be
discussed in Chapter 3) of positive even integers in terms of Bernoulli numbers, that is
24
We can try to evaluate this integral by expanding cos t using it’s Taylor series around t = 0
and then integrating the resulting series term by term. We obtain
Z1 ✓ ◆
xt t2 t4 1 1 1
F (x) = e 1 + · · · dt = + ··· .
0 2! 4! x x3 x5
We obtain Z1
xt 1 1! 2! 3!
G(x) = e (1 t + t2 · · · ) dt = 2
+ 3 + ··· . (1.4.1)
0 x x x x4
This series diverges for all x 2 R, and therefore appears to be meaningless. One might ask
why did the procedure succeed in the first case but not in the second, and the answer is given
by Theorem 1.1.1. In the first case, the expansion of cos t converges uniformly in any compact
interval in (0, 1), whereas in the second case the expansion of (1 + t) 1 diverges when t > 1.
However, not everything is lost. Suppose we try to sum the series (1.4.1) numerically for a
particular value of x, say x = 15. The first four terms are given by
1 1! 2! 3!
2
+ 3 = 0.0626963.
15 15 15 154
Surprisingly enough, this is very close to the correct value G(15) = 0.0627203 . . . . To investigate
this unexpected result, we consider the di↵erence Rm (x) between G(x) and the mth partial sum
of (1.4.1):
Z1 X m
e xt 1 (n 1)!
Rm (x) = dt ( 1)n
0 1 + t n=1
xn
Z 1 m xt
t e
= ( 1)m dt,
0 1+t
where the last equality follows from the fact that for all m 2 N,
1 tm
=1 t + t2 · · · ( 1)m 1 tm 1
+ ( 1)m .
1+t 1+t
Therefore, Z1
m!
|Rm (x)| < tm e xt
dt = .
0 xm+1
This means that the partial sums of (1.4.1) approximate G(x) with an error that is smaller
than the first neglected term of the series. The series (1.4.1) is an example of an asymptotic
expansion, the definition is due to Poincaré and will be discussed in the next section.
25
1.4.2 Asymptotic Notation
f (x)
If lim = 1, we write f (x) ⇠ g(x) (x ! a).
x!a g(x)
f (x)
If lim = 0, we write f (x) = o(g(x)) (x ! a).
x!a g(x)
If there exists C > 0 such that |f (x)| 6 C|g(x)| for all x in a neighborhood of a, we write
f (x) = O(g(x)) (x ! a). |
So, for example, if we want to say that f vanishes as x ! a we shall write f (x) = o(1) (x ! a).
If we write f (x) = O(1) (x ! a) this simply means that f is bounded in a neighborhood of
a. As simple examples,
We can extend Definition 1.4.1 to complex-valued functions as follows: Let S be any set, let f
and ' be complex-valued functions defined on S. Suppose there exists K > 0 such that
|f (s)| 6 K|'(s)| 8s 2 S.
Sometimes, it is useful to write f ⌧ g instead of f (x) = O(g(x))1 . We shall use this notation
throughout this document.
1
This is Vinogradov’s notation and it will prove to be quite useful.
26
Example 1.4.1. Let " > 0. Then, by L’Hôpital’s rule we have
log x 1
lim = lim = 0.
x!1 x" x!1 "x"
Now we are ready to give the definition of an asymptotic expansion and mention a very useful
result known as Watson’s Lemma.
Definition
P1 1.4.2 (Poincaré (1886)). Let F be a function of a real or complex variable z; let
n
n=0 an z denote a (convergent or divergent) formal power series, of which the sum of the
first n terms is denoted by Sn (z); let
Rn (z) = F (z) Sn (z).
That is,
a1 an 1
F (z) = a0 + + · · · + n 1 + Rn (z), (n 2 N),
z z
where we assume that when n = 0 we have F (z) = R0 (z). Now, suppose that for each n 2 N
the following relation holds
Rn (z) = O(z n ) (z ! 1)
P1 n
in some unbounded region . Then n=0 an z is called an asymptotic expansion of the
2
function F and we denote this by
1
X an
F (z) ⇠ (z ! 1), z2 .
n=0
zn
Example 1.4.2. The classical example is the exponential integral E1 (x) (which will be dis-
cussed in Chapter 3):
Z1 x t
e
F (x) = x dt = x ex E1 (x) (x > 0).
x t
1
Repeated integration by parts with g(t) = and h(t) = ex t
as in Theorem 1.1.2 yields
t
n 1
X Z1 x t
( 1)k k! n e
F (x) = + ( 1) n!x dt.
k=0
xk x t
n+1
Therefore, Z1 Z1
ex t n! n!
|Rn (x)| = n!x n+1
dt 6 n ex t
dt = .
x t x x xn
This means that for each n 2 N we have
n
Rn (x) = O(x ) (x ! 1).
Hence 1
X n!
F (x) ⇠ ( 1)n (x ! 1).
n=0
xn
2
Note that symbol ⇠ is now being used in two di↵erent ways. It should be clear from the context when ⇠
refers to an asymptotic expansion and when it refers to an asymptotic approximation
27
In the above example we can find the asymptotic expansion in a di↵erent way. Using the
substitution t = x(1 + u) we can write F (x) as a Laplace integral 3 :
Z1
1
F (x) = x e xu f (u) du, f (u) = .
0 1+u
(i) f is a real or complex-valued function of the positive real variable t with a finite number
of discontinuities and infinities.
Then 1
X an
F (z) ⇠ (n + ) (z ! 1),
n=0
z n+
1 ⇡
in the sector | arg z| 6 ⇡ , for some 0 < < and z n+ takes its principal value. Here
2 2
Z1
t z 1
(z) = e t dt (Re(z) > 0),
0
A larger sector for arg z can be obtained when we know that f is analytic in a certain domain of
⇡
the complex plane. For example, when f is analytic in the sector | arg z| < and f (t) = O(e t )
2
in that sector, for some number , then the asymptotic expansion in Watson’s lemma holds in
the sector | arg z| 6 ⇡ for some 0 < < ⇡.
3
Rb st
A Laplace integral is one of the form a
f (t) e dt, where s is a real or complex number.
28
1.5 The Fourier Transform
We denote by L1 the set of functions whose absolute value has finite integral, that is
n Z1 o
L (R) ··= f : R ! C such that
1
|f (x)| dx < 1 .
1
Definition 1.5.1. Let f 2 L1 (R). The Fourier transform of f is the function fˆ: R ! C defined
by Z1
ˆ
f (⇠) = e 2⇡i⇠t f (t) dt.
1
Let f 2 C 1 (R, C). We say that f is a Schwartz function if for every m, n 2 N we have
sup |tm f (n) (t)| < 1.
t2R
Intuitively, we are saying that f and all its derivatives decrease faster than any function of the
form
1
f (t) = ,
P (t)
whereP is a polynomial. Note that if f is a Schwartz function, then
Z1 Z1
1
|f (t)| dt ⌧ 2
dt = ⇡ < 1.
1 1 t +1
P
Proof. Consider the function F : R ! C defined by F (x) = n2Z f (x + n). Since F is a
smooth periodic function with period 1, its Fourier series expansion converges pointwise to F ,
so X
F (x) = cn e2⇡inx ,
n2Z
where Z1 Z1 X
2⇡inx 2⇡inx
cn = F (x) e dx = f (x + m) e dx.
0 0 m2Z
Since f is a Schwartz function, with the aid of Theorem 1.1.1 we can interchange summation
and integration to get
X Z1
cn = f (x + m) e 2⇡inx dx.
m2Z 0
Letting t = x + m we obtain
X Z m+1 Z1
cn = f (t) e 2⇡int
dt = f (t) e 2⇡int
dt = fˆ(n).
m2Z m 1
Therefore X X
f (x + n) = F (x) = fˆ(n) e2⇡inx ,
n2Z n2Z
and the result follows by evaluating at x = 0. ⌅
29
We shall need the following theorem from complex analysis, for a proof see [10].
Theorem 1.5.2 (Uniqueness principle). Let D be a domain and let f, g : D ✓ C ! C be
analytic functions on D. If f (z) = g(z) for z belonging to a set that has a nonisolated point,
then f (z) = g(z) for all z 2 D.
1
For the next Lemma we recall how to evaluate the Gaussian integral. Take z = in the Euler’s
2
reflection formula (2.1.5) to obtain
⇣1⌘ Z1 e t p
= p dt = ⇡,
2 0 t
or equivalently Z1 p
u2 ⇡
e du = . (1.5.1)
0 2
Lemma 1.5.1. For every z, w 2 C with Re(z) > 0, we have
Z1 r
zt2 +2wt ⇡ w2
e dt = ez ,
1 z
where we take the principal branch of the square root.
p y
Proof. Let x > 0, y 2 R and let u = xt p . Then
x
Z1 Z1 p
y2 py )2
xt2 +2yt ( xt
e dt = e x e x dt
1 1
y2 Z
ex 1 u2
=p e du.
x 1
r
⇡ y2
= ex,
x
u2
where the last equality follows from (1.5.1) and the symmetry of e . Now let z, w 2 C with
Re(z) > 0. By the uniqueness principle we obtain
Z1 r
zt2 +2wt ⇡ w2
e dt = ez ,
1 z
where we take the principal branch of the square root. This completes the proof ⌅
⇡t2
Corollary 1.5.2.1. The Schwartz function g : R ! [1, 1) defined by g(t) = e is its own
Fourier transform.
[ 1 ˆ⇣ ⇠ ⌘
f ( t)(⇠) = f . (1.5.2)
30
Theorem 1.5.3. The Jacobi theta function ✓ : (0, 1) ! R defined by
X 2
✓(x) = e ⇡n x ,
n2Z
⇡x2 t
p
Proof. Let ft (x) = e . Note that ft (x) = f1 ( t x). Then, by (1.5.2) we have
p 1 ⇣ ⇠ ⌘
fˆt (⇠) = f1\
( tx)(⇠) = p fˆ1 p .
t t
Now we prove a theorem that essentially says that for many types of functions it is possible to
recover them from its Fourier transform.
Theorem 1.5.4 (Fourier inversion theorem). Let f, fˆ 2 L1 (R), with f continuous. Then
Z1
f (x) = e2⇡i⇠x fˆ(⇠)d⇠.
1
It is very tempting to exchange the order of integration in the above expression, but we can’t
do this because the function f (t) e2⇡i⇠(x t) is not L1 (R ⇥ R). We use the following trick: Given
" > 0, let Z Z 1 1
"2 ⇠ 2
I" (x) ··= f (t) e2⇡i⇠(x t)
e dtd⇠.
1 1
Since
"2 ⇠ 2
|fˆ(⇠) e2⇡i⇠x e | 6 |fˆ(⇠)|,
and fˆ 2 L1 (R), then Z1
I" (x) ! e2⇡i⇠x fˆ(⇠)d⇠. (1.5.3)
"!0 1
31
"2 ⇠ 2
Now, since f (t) e2⇡i⇠(x t) e is L1 (R ⇥ R) we can apply Fubini’s theorem to interchange the
order of integration. Thus
Z1 Z1
"2 ⇠ 2
I" (x) = e2⇡i⇠(x t)
e f (t)d⇠ dt.
1 1
Using the Gaussian integral (1.5.1) and the fact that f is continuous, we obtain
32
Chapter 2
Special Functions
This section is based on the book [26]. The Gamma function originated as a solution of an
interpolation problem for the factorial function: Find a smooth curve that connects the points
(x, y) given by y = (x 1)! at the positive integer values for x. This problem can be solved by
considering the so called Euler’s integral of the second kind
Z1
t z 1
(z) = e t dt (Re(z) > 0),
0
33
Let ↵, 2 R such that ↵ < R(z) < . Then
|tz 1 | 6 t↵ 1
(0 < t 6 1), |tz 1 | 6 t 1
(t > 1).
Therefore, by the Weierstrass’ M -test, the integral converges uniformly with respect to z. That
(z) is holomorphic in the half-plane Re(z) > 0 is a consequence of this result and Theorem
1.1.3. A single partial integration of the Gamma function yields the fundamental recurrence
formula
(z + 1) = z (z).
Hence if z = n 2 N we obtain
(n + 1) = n!.
The recurrence formula enables to be continued analytically into the left half-plane, except
at the points 0, 1, 2, . . .. Now we use Prym’s decomposition to answer the question about
the nature of the non-positive integer points:
Z1 Z1
z 1 t
(z) = t e dt + tz 1 e t dt.
0 1
The second integral represents an entire function of z. In the first integral we expand e t using
it’s Taylor series around t = 0 and then we use Theorem 1.1.1 to interchange summation and
integration. This gives us the following expansion due to Mittag-Le✏er
X1 Z1
( 1)n
(z) = + tz 1
e t
dt,
n=0
n!(n + z) 1
which holds for all z 2 C \ {0, 1, 2, · · · }. It follows that has simple poles at each non-
positive integer value. Moreover, for every n 2 N we have
( 1)n
lim (z + n) (z) = .
z! n n!
( 1)n
This means that the residue at the pole n equals .
n!
An alternative definition of , which is not restricted to the half-plane R(z) > 0, can be derived
in the following way. We have
✓ ◆n
t t
lim 1 =e .
n!1 n
t n
Repeated integration by parts of n (z) with g(t) = 1 n
and h(t) = tz 1
as in Theorem
1.1.2 yields
34
n 1
X n k z+k n Zn
(n 1) · · · (n k + 1) 1 nt t (n 1) · · · 2 · 1 · tz+n 1
n (z) = + dt
k=0
nk 1 z(z + 1) · · · (z + k) 0 0 nn 1 z(z + 1) · · · (z + n 1)
nz n!
= .
z(z + 1) · · · (z + n)
(z) n (z) = I1 + I2 ,
where
Z1 Zn ✓ ✓ ◆n ◆
t z 1 t t
I1 = e t dt and I2 = e 1 tz 1
dt.
n 0 n
Hence
✓ ◆n ⇢ ✓ ◆
t t2 t3 t4
1 = exp t + 2 + 3 + ··· =e t A
6 e t, (2.1.1)
n 2n 3n 4n
where
t2 t3 t4
A= + 2 + 3 + ··· .
2n 3n 4n
n Bt2
Note that if t 6 then A 6 where
2 n
1 1 1
B= + + + ···
2 3 · 2 4 · 22
is a finite number. So,
✓ ◆n
t Bt2
e t
1 = e t (1 e A
)6e t
A6e t
. (2.1.2)
n n
Zn Z n/2
B
6 2e t Re(z) 1
t dt + e t Re(z)+1
t dt
n/2 n 0
" "
< + = ".
2 2
This gives us the Euler’s limit formula
n!nz
(z) = lim . (2.1.3)
n!1 z(z + 1) · · · (z + n)
35
2.1.2 Infinite Product Formula
Now we are going to use Euler’s limit formula to write (z) into the canonical form of an infinite
product, we need the following lemma.
Lemma 2.1.1. The sequence {sn } defined by
n
X 1
sn = log n
k=1
k
converges as n ! 1.
Proof. Since 1/t is a decreasing function for t > 0, then for n > 2
Zn
1 1 1 dt 1 1 1
+ + ··· + < < 1 + + + ··· + .
2 3 n 1 t 2 3 n 1
Therefore 1/n < sn < 1, so sn is a sequence of positive numbers. Now,
✓ ◆
1 n
sn+1 sn = + log < 0.
n+1 n+1
This tells us that sn is a decreasing sequence, thus we conclude that sn converges as n ! 1. ⌅
The limiting value of sn is called the Euler-Mascheroni constant and is usually denoted by .
Numerical computations give
⇡ 0.5772156649 . . . .
Let z 2 C \ {0, 1, 2, · · · }. Note that
⇢ ✓ ◆ Y
n ✓ ◆
1 1 1 z+k z/k
= z exp z 1 + + · · · + log n e .
n (z) 2 n k
k=1
The left hand side function of (2.1.4) is called the reciprocal Gamma function.
Therefore
(x ⇠1 )(x ⇠2 ) · · · (x ⇠ m 1 ) = 1 + x + · · · xm 1 .
In particular, for x = 1 we obtain
m
Y1
(⇠k 1) = ei⇡(m 1)
m.
k=1
Thus
1 m 1
f (z) = m 2 (2⇡) 2 .
37
This gives us the Gauss’ multiplication formula
Y1 ✓
m
k
◆
m 1 1
mz
z+ = (2⇡) 2 m 2 (mz). (2.1.6)
k=0
m
The final formula that we will need for the Gamma function is an integral representation that
is constructed by using a loop contour in the complex plane. The idea is due to Hankel (1864).
Consider Z
I(z) = et t z
dt,
where the path begins at t = 1, encircles t = 0 once in the positive orientation, and returns
to its starting point;
We suppose that the branch of t z takes its principal value at the point where the contour
crosses the positive real axis, and is continuous elsewhere. Let " > 0. Then by Cauchy’s integral
theorem the path can be deformed into the two sides of the interval ( 1, "], together with
the circle |t| = "; see Figure 2.2.
3
1 0
2
Thus Z Z Z
t z t z
I(z) = e t dt + e t dt + et t z
dt,
1 2 3
where 1 is the lower side of the negative real axis (here arg t = ⇡), 2 is the circle |t| = "
and 3 is the upper side of the negative real axis (here arg t = ⇡). Assume temporarily that
Re(z) < 1. Then
Z Z 2⇡
t z
e t dt = "i e"(cos ✓+i sin ✓) (" ei✓ ) z ei✓ d✓ ! 0.
0 "!0
2
38
Therefore Z
1 1
= et t z
dt. (2.1.8)
(z) 2⇡i
This is Hankel’s loop integral. Analytic continuation removes the temporary restriction on
Re(z), provided that the branch of t z is chosen as discussed before. Now, with the aid of
Theorem 1.1.3 we see that the right hand side of (2.1.8) represents an entire function. This
shows that the reciprocal Gamma function is an entire function. As a corollary, the Gamma
1
function has no zeros. The graph of (x) and , for x 2 R is shown in Figure 2.3
(x)
-4 -2 2 4
-5
-10
-15
-20
Figure 2.3: The Gamma function in blue and the reciprocal Gamma function in red.
If f : (a, b) ! R is a positive function such that log f is convex, then we say that f is logarith-
mically convex. Note that if f : (a, b) ! R is convex and a < x < y < z < b, then
f (y) f (x) f (z) f (x)
6 . (2.1.9)
y x z x
Theorem 2.1.1. Suppose that f : (0, 1) ! R is a function such that
(i) f (1) = 1
(ii) f (x + 1) = xf (x)
39
(iii) f is logarithmically convex.
Proof. Let n 2 N and let x 2 (0, 1). Note that conditions (i) and (ii) imply that it is sufficient to
prove the theorem for such x. Since f is logarithmically convex and n < n+1 < n+1+x < n+2,
we use (2.1.9) to obtain
f (n + 1) 1 f (n + 1 + x) f (n + 2)
log 6 log 6 log .
f (n) x f (n + 1) f (n + 1)
Using conditions (i) and (ii) we may simplify this inequalities to get
(x + n)(x + n 1) · · · xf (x)
x log n 6 log 6 x log(n + 1).
n!
Substracting x log n we may rearrange the inequalities as follows:
x(x + 1) · · · (x + n) ⇣ 1⌘
0 6 log + log f (x) 6 x log 1 + .
n!nx n
Therefore,
n!nx
f (x) = lim = (x). ⌅
n!1 x(x + 1) · · · (x + n)
This section is based on [34]. The digamma function is defined as the logarithmic derivative of
the Gamma function and is usually denoted by
0
(z)
(z) = .
(z)
A graph of (x) for x 2 R is shown in Figure 2.4. By using the infinite product (2.1.4) we
obtain
10
-4 -2 2 4
-5
- 10
40
X1 ✓ ◆
z z
log (z) = log z z+ log 1 + , z 6= 0, 1, 2, . . . .
n=1
n n
Hence 1 ⇣
X 1 1 ⌘
(z) = + , z 6= 0, 1, 2, . . . . (2.2.1)
n=0
n+1 n+z
It follows that the digamma function possesses simple poles at all non positive integers. Using
(2.2.1) we obtain the recurrence formula for the function:
1
(z + 1) = (z) + . (2.2.2)
z
Now we are going to find an integral representation for that will be used to find an asymptotic
expansion for log . We first establish two lemmas.
Lemma 2.2.1. The Euler-Mascheroni constant has the following integral representations:
Z1 Z1 Z1 t Z1
t 1 e t e e t e t
e log t dt, dt dt, dt.
0 0 t 1 t 0 1 e t t
Hence Z1 Z1 Z1 Z1
t t t
t e 1 e e
= e log t dt dt = dt dt.
0 1 t 0 t 1 t
u
For the last one, write t = 1 e to get
Z1 Z1 t Z1 Z1 t
1 e t e dt e
= dt dt = lim+ dt
0 t 1 t "!0 " t " t
Z1 Z1 u
e u e
= lim+ du du
"!0 " 1 e u " u
Z1
e t e t
= dt. ⌅
0 1 e t t
41
Lemma 2.2.2. For every z 2 C \ {0, 1, 2, . . .}, we have
Z1 t
e e zt
(z) = dt (Re(z) > 0).
0 t 1 e t
When Re(z) > 0, we use Theorem 1.1.1 to interchange summation and integration. This gives
Z1 ✓ t t 1
X ◆
e e t(n+1) t(n+z)
(z) = t
+ e e dt
0 t 1 e n=0
Z1 t zt
e e
= t
dt. ⌅
0 t 1 e
Now we are ready to give an integral representation for due to Binet (1839).
where ✓ ◆ 1
1 1 1 1 X B2n 2n 2
g(t) = t
+ = t (|t| < 2⇡). (2.2.6)
t e 1 t 2 n=1
(2n)!
42
Proof. We use Lemma 2.2.2 and (2.2.4) to get
Z1 t
e e t(z+1)
(z + 1) = dt
0 t 1 e t
Z1 t
e e zt
= dt
0 t et 1
Z1 ✓ ◆
zt 1 1
= log z + e dt
0 t et 1
Z1 ✓ ◆
1 zt 1 1 1
= log z + e + dt,
2z 0 et 1 t 2
which is equivalent to (2.2.5). The series expansion for g follows from the fact that
1
t t X B2n 2n
1+ = t .
et 1 2 n=1 (2n)!
Now we are ready to find an asymptotic expansion for log . Integrating (2.2.5) we obtain
⇣ ⌘ Z1
1
log (z + 1) = z + log z z + g(t) e zt dt + C, (2.2.7)
2 0
then
1
lim g(t) = 0 and lim g(t) = .
t!1 t!0 12
Moreover, g is decreasing for t > 0, since
2 et (t2 + 4) e2t (t 4) + t + 4
g 0 (t) = < 0 (t > 0).
2t3 (et 1)2
and ⇣ ⌘ Z1
1 2zt
log (2z + 1) = 2z + log 2z 2z + g(t) e dt + C. (2.2.10)
2 0
Now define the following functions:
⇣ 1
⌘ ⇣ 1
⌘
h(z) = z + log z z and H(z) = h(z) + h z h(2z).
2 2
43
It is easy to see that
2z 1 1 1
H(z) = z log
+ log 2.
8z 2 2
1
Expanding H in power series about infinity we get
1 ⇣1⌘
H(z) = 2z log 2 log 2 + O (z ! 1).
2 z
Combining (2.2.7), (2.2.9), (2.2.10), and writing the integrals as in (2.2.8) we get
(z + 1) (z + 12 )
log = H(z) + o(1) + C. (z ! 1) (2.2.11)
(2z + 1)
The Legendre’s duplication formula (2.1.7) can be written in the form
22z (z + 1) (z + 12 ) 1
log = log ⇡.
(2z + 1) 2
Thus, (2.2.11) becomes
1 1 ⇣1⌘
2z
log ⇡ log 2 = 2z log 2 log 2 + O + o(1) + C (z ! 1).
2 2 z
1
Letting z ! 1 we obtain C = log 2⇡. Since
2
log (z + 1) = log z + log (z),
we can write (2.2.7) in the form
⇣p ⌘ Z1
1
z z zt
log (z) = log 2⇡z 2 e + g(t) e dt.
0
Using the series expansion for g (2.2.6), we find via Watson’s lemma (Theorem 1.4.2) an asymp-
totic expansion for the logarithm of the gamma function (Stirling’s series):
⇣p 1
⌘ 1
X B2n 1
z z
log (z) ⇠ log 2⇡z 2 e + , (z ! 1). (2.2.12)
n=1
2n(2n 1) z 1
2n
Since the singularities of g are located on the imaginary axis, the above expansion holds for
| arg z| < ⇡. Taking the exponential of this result, we get the generalization of Stirling’s formula:
✓ ◆
p z 12 z 1 1 139 571
(z) ⇠ 2⇡z e 1+ + + · · · , (z ! 1). (2.2.13)
12z 288z 2 51840z 3 2488320z 4
This section is based on [26]. The exponential integral is defined for z 2 C by the formula
Z1 t
e
E1 (z) = dt.
z t
1
We say that f (z) is analytic at z = 1 if the function g(w) = f (1/w) is analytic at w = 0. Thus we let
w = 1/z, and consider the behavior of f (z) at z = 1 by studying the behavior of g(w) at w = 0.
44
Since t = 0 is a pole of the integrand, z = 0 is a branch point of E1 (z). The principal branch
is obtained by introducing a cut along the negative real axis. Now suppose z > 0, letting
t = (1 + u)z we obtain Z1
z e uz
E1 (z) = e du.
0 1+u
⇡
Since the above integral converges if | arg z| < , then, by analytic continuation we extend the
2
result for complex z to obtain
Z 1 tz
z e ⇡
E1 (z) = e dt, | arg z| < .
0 1+t 2
The complementary exponential integral is defined by
Zz
1 e t
Ein(z) = dt,
0 t
and is entire. By expanding the integrand using it’s Taylor series around t = 0 and then
integrating the resulting series term by term with the aid of Theorem 1.1.1 we get
Zz X1 X1
( 1)n+1 tn ( 1)n+1 z n
Ein(z) = dt = .
0 n=0 (n + 1)! n=1
n n!
and again analytic continuation extends this result for complex z. When z = x 2 R, another
notation used for the exponential integral is given by
Zx t
e
Ei(x) = dt (x 6= 0), (2.3.2)
1 t
R
where means that the integral takes its Cauchy principal value when x is positive2 . The
connection with the previous notation is
The first identity follows from from the substitution t = u. For the second one, use (2.3.1) to
get
45
A related function is the logarithmic integral defined for x > 0 as
Zx
dt
li(x) = (x =6 1). (2.3.3)
0 log t
To avoid the singularity of the integrand at t = 1 in (2.3.3) we define the o↵set logarithmic
integral or Eulerian logarithmic integral as
Zx
dt
Li(x) = .
2 log t
C
CR
C"
R " " R x
46
On the small semicircle t = ✏ ei✓ , ⇡ > ✓ > 0. Therefore
Z Z0
eit i✓
dt = iei" e d✓ ! i⇡.
C" t ⇡ "!0
sin t
Taking imaginary part and using the fact that is an even function the result follows. ⌅
t
Using the definition of the sine integrals and the previous lemma we get
⇡
Si(z) = + si(z).
2
The cosine integrals are defined by
Z1 Zz
cos t 1 cos t
Ci(z) = dt, Cin(z) = dt.
z t 0 t
Ci has a branch point at z = 0 and the principal branch is obtained by introducing a cut along
the negative real axis. Cin is entire. By a change of variables we can express the sine and cosine
integrals in terms of the exponential integrals, the result is
and
2 Ci(z) = E1 (iz) E1 ( iz), 2 Cin(z) = Ein(iz) + Ein( iz).
This result and (2.3.1) connects the two cosine integrals:
47
7
Proof. The proof runs by induction on n. For n = 1, clearly log 1 = ⇤(1). Now suppose n > 1
and write
Yk
↵
n= pj j ,
j=1
where p1 < p2 < · · · < pk are primes and the ↵j are positive integers. Therefore
↵j
k X k
X X X
⇤(d) = ⇤(pm
j ) = ↵j log pj = log n. ⌅
d|n j=1 m=1 j=1
For x > 0 we define the prime counting function ⇡ as the number of primes less than or equal
to x, that is X
⇡(x) = 1.
p6x
3
This inequality easily follows by comparing the graph of the functions sin x and ⇡2 x on the interval (0, ⇡2 ).
4
Do not confuse with the digamma function which is also denoted by . To avoid all possible confusions,
some writers use (0) for the digamma function and confine for the Chebyshev function.
48
A graph of the Chebyshev’s function is shown in Figure 2.7. Now,
X 1 X
X 1 X
X
(x) = ⇤(n) = ⇤(pm ) = log p.
n6x m=1 pm 6x m=1 1
p6x m
1
Note that the sum over m is finite because the sum on p is empty if x m < 2 or equivalently if
m > log2 x. Therefore,
X X X 1
(x) = log p = #(x m ). (2.4.2)
m6log2 x 1 m6log2 x
p6x m
Thus X 1
(x) #(x) = #(x m ) > 0.
26m6log2 x
Therefore
X p
1 1 p p x log2 x
0 6 (x) #(x) 6 x log x
m m 6 log2 x x log x= ,
26m6log2 x
log 4
1000
800
600
400
200
49
Lemma 2.4.1 (Abel’s identity). Given an arithmetic function a, let A : R ! C be defined as
8X
>
< a(n), if t > 1;
A(t) = n6t
>
:0, otherwise.
Proof. Note that the sum in (2.4.3) can be expressed as a Riemann-Stieltjes integral:
X Zx
a(n)'(n) = '(t)dA(t).
y<n6x y
Integrating by parts and using the fact that ' 2 C 1 [y, x] we obtain
X Zx
a(n)'(n) = A(x)'(x) A(y)'(y) A(t)d'(t)
y<n6x y
Zx
= A(x)'(x) A(y)'(y) A(t)'0 (t) dt. ⌅
y
Then X X X X
⇡(x) = 1= a(n), and #(x) = log p = a(n) log n.
p6x 1<n6x p6x 1<n6x
50
1 p
Again, we use Lemma 2.4.1 with '(t) = and y = 2 to obtain
log t
p Zx
#(x) #( 2) #(t)
⇡(x) = p + p 2 dt
x log 2 2 t log t
Zx
#(x) #(t)
= + 2 dt.
log x 2 t log t
x
(1) ⇡(x) ⇠ (x ! 1)
log x
(2) #(x) ⇠ x (x ! 1)
(3) (x) ⇠ x (x ! 1)
Proof. (2) () (3) follows from Theorem 2.4.2, so it suffices to show (1) () (2). First
suppose that
x
⇡(x) ⇠ (x ! 1).
log x
This implies that Zx Zx
1 ⇡(t) ⇣1 dt ⌘
dt = O .
x 2 t x 2 log t
Now,
Zx Z px Zx p p
dt dt dt x 2 x x
= + p
6 + p .
2 log t 2 log t x log t log 2 log x
Thus Zx
1 dt
= o(1) (x ! 1).
x 2 log t
Using Theorem 2.4.3 we get
Z
#(x) ⇡(x) log x 1 x ⇡(t)
= dt
x x x 2 t
⇡(x) log x ⇣ 1 Z x dt ⌘
= +O
x x 2 log t
⇡(x) log x
= + o(1).
x
Therefore
#(x)
⇠ 1.
x
This means that (1) =) (2). Now suppose that #(x) ⇠ x (x ! 1). Then,
Z ⇣ log x Z x dt ⌘
log x x #(t)
dt = O .
x 2 t log2 t x 2 log2 t
Now,
Zx Z px Zx p p
dt dt dt x 2 x x
= + 2 6 + 2p .
2 log2 t 2 log2 t p
x log t 2
log 2 log x
51
Thus Zx
log x dt
= o(1) (x ! 1).
x 2 log2 t
Using Theorem 2.4.3 we obtain
Z
⇡(x) log x #(x) log x x #(t)
= + dt
x x x 2 t log2 t
#(x) ⇣ log x Z x dt ⌘
= +O
x x 2 log2 t
#(x)
= + o(1).
x
This shows that
⇡(x) log x
⇠ 1,
x
that is (2) =) (1). This completes the proof. ⌅
52
Chapter 3
This Chapter is based on [15] and [27]. The Riemann Zeta function is defined by the series
X1
1
⇣(s) = (Re(s) > 1). (3.0.1)
n=1
ns
By the integral test, the series converges absolutely and uniformly in any compact domain
within R(s) > 1, hence ⇣ is holomorphic in this half-plane.
By the alternating series test, the series converges absolutely and uniformly in any compact
domain within Re(s) > 0, hence ⌘ is holomorphic in this half-plane. Note that
⌘(s) = (1 21 s )⇣(s).
This yields analytic continuation of ⇣ to the half plane Re(s) > 0, except for a pole at s = 1.
Note that if Re(s) > 1, then
X1
1 X 1 ⇣1 1 ⌘
⇣(s) = = n
n=1
ns n=1
ns (n + 1)s
X1 Z n+1
dx
=s n s+1
.
n=1 n x
X1 Z n+1 Z1
bxc bxc
⇣(s) = s s+1
dx = s s+1
dx.
n=1 n
x 1 x
53
1 1 1
Let ' : [1, 1) ! [ , ]
2 2
be the periodic function defined by '(x) = x bxc 2
. Then
Z1
s 1 '(x)
⇣(s) = s s+1
dx (3.1.1)
s 1 2 1 x
Z1 ⇣ Zx ⌘ dx
1 1
= + s(s + 1) '(y) dy s+2 ,
s 1 2 1 1 x
where the last equality follows by partial integration. Since the integral of ' is bounded (it
is an odd periodic function), then the last integral in x converges absolutely if Re(s) > 1,
giving the analytic continuation of ⇣ to the half-plane Re(s) > 1. It also shows that s = 1 is
a simple pole of ⇣ with residue 1, that is
lim(s 1)⇣(s) = 1.
s!1
Note that
1
⇣(0) = .
2
Moreover, using (3.1.1) we get
Z1 ZN 1
1 1 '(x) 1 x bxc 2
lim ⇣(s) = dx = lim dx.
s!1 s 1 2 1 x2 2 N !1 1 x2
We evaluate the integral as follows:
ZN 1 X1 Z n+1
N
x bxc 2 1 1 n
dx = log N + dx
1 x2 2N 2 n=1 n x2
XN
1 1 1
= log N + + .
2N 2 n=1
n
54
Therefore
XN
1 1
lim ⇣(s) = lim log N = ,
s!1 s 1 N !1 n=1 n
where is the Euler-Mascheroni constant. Hence
1
⇣(s) = + + O(|s 1|) (s ! 1).
s 1
1
This shows that the function ⇣(s) has order |s 1| near 1. For an illustration of
s 1
this behaviour see the Figure below
1
Figure 3.2: Absolute value of the function ⇣(s) .
s 1
Letting t = n2 ⇡x we get Z1
s
⇣s⌘ s
s 1 n2 ⇡x
⇡ 2 n = x2 e dx.
2 0
Now we sum over all n 2 N⇤ and use Theorem 1.1.1 to interchange summation and integration
to obtain
⇣s⌘ Z1 ⇣X 1 ⌘
s s 2
⇡ 2 ⇣(s) = x2 1 e n ⇡x dx
2 0 n=1
Z1 ⇣
s
1 ✓(x) 1⌘
= x 2 dx,
0 2
55
where X
n2 ⇡x
✓(x) = e
n2Z
is the Jacobi theta function. Thus
Z1 ⇣ ✓(x) 1⌘
s
⇡2 s
1
⇣(s) = s x2 dx.
(2) 0 2
Now we write the integral in the form
Z1 ⇣ Z1 Z1
s
1 ✓(x) 1⌘ s
1
⇣ ✓(x) 1⌘ s
1
⇣ ✓(x) 1⌘
x 2 dx = x 2 dx + x2 dx.
0 2 0 2 1 2
1
In the first integral, we let u = to get
x
Z1 ⇣ Z1
s
1 ✓(x) 1 ⌘ s
1
⇣ ✓(u 1 ) 1⌘
x2 dx = u 2 du.
0 2 1 2
By Theorem 1.5.3 we know that
1
✓(u 1 ) = u 2 ✓(u).
Thus
Z1 ⇣ ✓(x) Z Z
s
1 1⌘ 1 1 s 1 1 1 s
1
x 2 dx = u 2 2 ✓(u) du u 2 du
0 2 2 1 2 1
Z1 Z1
1 s 1 1 s 1 1
= x 2 2 ✓(x) dx x 2 2 dx + .
2 1 2 1 s(s 1)
Z1 ⇣ ✓(x)
s 1 1⌘ 1
= x 2 2 dx + .
1 2 s(s 1)
This shows that
" Z1 #
⇡2
s
1 s 1 s
⇣ ✓(x) 1⌘
1
⇣(s) = s + (x 2 2 +x 2 ) dx . (3.2.1)
( 2 ) s(s 1) 1 2
Thanks to the exponential decay of ✓, the above integral converges for all s 2 C and hence
1
defines an entire function. Since is an entire function and
1
⇣(0) =
,
2
then (3.2.1) gives the analytic continuation of ⇣ to the whole complex plane with the exception
of s = 1. Riemann noticed that formula (3.2.1) not only gives the analytic continuation of ⇣,
1
but can also be used to derive a functional equation for ⇣. He observed that the term
s(s 1)
and the integral in (3.2.1) are invariant under the transformation s ! 1 s. Therefore
s
⇣s⌘ s 1
⇣1 s⌘
⇡ 2 ⇣(s) = ⇡ 2 ⇣(1 s). (3.2.2)
2 2
Using the Legendre’s duplication formula (2.1.7) and the Euler’s reflection formula (2.1.5) it is
easy to see that
( 2s ) 1
⇣ s⇡ ⌘
1 s
= ⇡ 22 cos (s),
(12 s) 2
and if this is used in the functional equation it gives the unsymmetrical form of the functional
equation: ⇣ s⇡ ⌘
⇣(1 s) = 21 s ⇡ s cos (s)⇣(s). (3.2.3)
2
We can summarize the results in the following theorem:
56
Theorem 3.2.1 (The functional equation). The Riemann zeta function ⇣(s) extends to a
meromorphic function with a simple pole at s = 1, and the function
1 s
⇣s⌘
⇠(s) ··= s(s 1)⇡ 2 ⇣(s)
2 2
satisfies the functional equation
⇠(s) = ⇠(1 s)
for every s 2 C.
The function ⇠ defined in Theorem 3.2.1 is called the Riemann xi function and is entire. As we
know from Chapter 1, ⇣(2m) can be expressed in terms of Bernoulli numbers (see (1.3.12)):
a relation known to Euler in 1737. Using the functional equation it is easy to see that in general
we can write
Bm+1
⇣( m) = , m = 1, 2, . . . .
m+1
To begin this section we give an infinite product representation for the Riemann zeta function,
due to Euler. Suppose that Re(s) > 1. Subtracting the series 2 s ⇣(s) from the one in (3.0.1)
we obtain
1 1 1
(1 2 s )⇣(s) = 1 + s + s + s + · · · .
3 5 7
Similarly, we obtain
X 1
(1 2 s )(1 3 s )⇣(s) = ,
ns
where now the sum runs over n > 1, except for multiples of 2 and 3. Now, let pm denote the
m-th prime number. By repeating the above procedure we obtain
m
Y X 1
⇣(s) (1 pn s ) = 1 + ,
n=1
ns
where the sum runs over n > 1, except for the multiples of p1 , p2 , . . . , pm . Note that the sum of
this series vanishes as m ! 1 (since pm ! 1). Therefore, letting m ! 1 we obtain Euler’s
product formula
Y 1
⇣(s) = (Re(s) > 1). (3.3.1)
p2P
1 p s
This shows that ⇣ has no zeros in the half plane Re(s) > 1. Note that (3.2.1) shows that ⇣ has
simple zeros at s = 2, 4, . . .. These zeros, arising from the poles of the gamma function, are
called the trivial zeros. From the functional equation (3.2.2) and using the fact that ⇣(s) 6= 0
for Re(s) > 1, all other zeros, the nontrivial zeros, lie in the vertical strip 0 6 Re(s) 6 1. Now,
let s 2 C with s = + it. Suppose that > 1. By taking logarithms of each side of (3.3.1) we
obtain X
log ⇣(s) = log(1 p s ).
p2P
57
Since 1
X xn
log(1 x) = (|x| < 1),
n=1
n
and |p s | < 1, then
1
XX
log ⇣(s) = n 1p ns
. (3.3.2)
p2P n=1
where ⇤ is the Mangoldt function defined by (3.7.1). Now we will show that ⇣ has no zeros in
Re(s) = 1.
Proof. Let s 2 C with s = + it. Note that (3.3.2) can be written in the form
1
XX
log ⇣(s) = n 1p n
e int log p
.
p2P n=1
Now, since Re log z = log |z| for every z 2 C, using (3.3.4) we obtain
1
XX
3 log |⇣( )|+4 log |⇣( +it)|+log |⇣( +2it)| = n 1p n
3+4 cos(nt log p)+cos(2nt log p) .
p2P n=1
or equivalently
|⇣( )|3 |⇣( + it)|4 |⇣( + 2it)| > 1.
Therefore
⇣( + it) 4 1
|( 1)⇣( )|3 |⇣( + 2it)| > . (3.3.5)
1 1
Now we will show that ⇣(1 + it) 6= 0 using inequality (3.3.5). Suppose by contradiction that
⇣(1 + it) = 0 for some t 2 R \ {0}. Then, by L’Hôpital’s rule we have
⇣( + it)
lim = ⇣ 0 (1 + it).
!1 1
58
Since ⇣ has a simple pole at s = 1, then
lim ( 1)⇣( ) = 1.
!1
The goal of this section is to prove that the ⇠ function has the following product representation:
Y⇣ s ⌘ ⇢s
⇠(s) = eA+Bs 1 e ,
⇢
⇢
where ⇢ = + i runs over the nontrivial zeros of ⇣. This result was conjectured by Riemann
and proved by Hadamard in 1893.
Definition 3.4.1. Let f : C ! C be an entire function. We say that f has finite order if there
exists positive constants ↵, , such that
|f (z)| 6 ↵ e |z|
(3.4.1)
1 s
s(s 1)⇡ 2 ⌧ exp( 1 |s|) (|s| ! 1).
2
Since | arg z| < ⇡, we may apply Stirling’s formula (2.2.13) to get
⇣s⌘
⌧ exp( 2 |s| log |s|) (|s| ! 1),
2
59
for some for some 2 > 0. Using the integral representation for ⇣ obtained in (3.1.1) we get
|⇣(s)| ⌧ |s| = exp(log |s|).
This shows that there exists C > 0 such that
1 s
⇣s⌘
|⇠(s)| = s(s 1)⇡ 2 ⇣(s) ⌧ exp(C|s| log |s|) (|s| ! 1).
2 2
Now, we recall from Example 1.4.1 that for all " > 0 we have
log |s| ⌧ |s|" (s ! 1).
This shows that there is some > 0 such that for all " > 0, we have
|⇠(s)| ⌧ exp |s|1+" (|s| ! 1).
Therefore ⇠ has order at most 1. ⌅
Lemma 3.4.2. An entire function f : C ! C of order which does not have zeros has the
form
f (z) = eg(z) ,
where g is a polynomial of degree .
Proof. Since f is entire and f (z) 6= 0 for all z 2 C, then the function log f (z) exists and is
entire, so
1
X
g(z) = log f (z) = an z n ,
n=0
where the power series converges absolutely. This shows that
f (z) = eg(z) .
Now we show that g is a polynomial of degree . Since f has finite order, then using (3.4.1),
we get
Reg(z) = log |f (z)| 6 |z| + log ↵. (3.4.2)
Let am = bm + icm . In the circle |z| = R e2⇡i✓ , 0 6 ✓ < 1, we have
1
X
Reg(z) = b0 + (bm cos 2⇡m✓ cm sin 2⇡m✓)Rm .
m=1
Using the orthogonality relations of the trigonometric functions (1.2.3) and (1.2.4) we get
Z1
bn R = 2 Reg(R e2⇡i✓ ) cos 2⇡n✓d✓.
n
0
Z01
⇥ ⇤
=2 |Reg(R e2⇡i✓ )| + Reg(R e2⇡i✓ ) d✓ 2bo
Z01
= 4 max{0, Reg(R e2⇡i✓ )}d✓ 2b0
0
6 4( R + log ↵) 2b0 .
Since R can be arbitrarily large, then bn = 0 for n > . Similarly cn = 0 for n > . This
shows that g is a polynomial of degree less than or equal to . Since is the infimum of the
exponents , then = deg g. ⌅
60
The next result gives a connection between the zeros of an analytic function f and log |f |.
Lemma 3.4.3 (Jensen’s Formula). Let f : C ! C be analytic on the disk |z| 6 R. If f (0) 6= 0
and f (z) 6= 0 for |z| = R, then
Z1 ZR
2⇡i✓ |f (0)|Rn n(r)
log |f (R e )|d✓ = log = dr + log |f (0)|,
0 |z1 z2 · · · zn | 0 r
where z1 , z2 , . . . , zn are all the zeros of f (counted with multiplicities) and n(r) is the number
of zeros in |z| 6 r.
Proof. If g is analytic in |z| 6 R and g(z) 6= 0 then, by Cauchy’s integral theorem we have
Z
1 log g(z)
log g(0) = dz.
2⇡i |z|=R z
Let n
Y R 2 z zk
g(z) ··= f (z) .
k=1
R(z z k)
Now suppose without loss of generality that |z1 | 6 |z2 | 6 · · · 6 |zn | 6 |zn+1 | ··= R. Then
ZR Xn Z |zk+1 |
n(r) k
dr = dr
0 r k=1 |z k | r
n
X
= k(log |zk+1 | log |zk |)
k=1
Rn
= log . ⌅
|z1 z2 · · · zn |
Corollary 3.4.0.1. If f : C ! C is entire and satisfies (3.4.1), then the number of zeros in
the disk |z| 6 R satisfies
n(R) ⌧ R (R ! 1).
61
Proof. Let be the order of f . If > , then using (3.4.1) we get
Proof. Let ⌫ > . Let z1 , z2 . . . be the zeros of f (finite or countable zeros). Let rk ··= |zk | and
r0 ··= 0. Then
X X1
⌫
|⇢| = n(rj ) n(rj 1 ) rj ⌫ .
⇢6=0 j=1
Therefore 1 Z rj+1 Z1
X X
⌫ ⌫ 1 ⌫ 1
|⇢| = n(rj ) ⌫r dr = ⌫ n(r)r dr.
⇢6=0 j=1 rj 0
From now on we restrict our analysis to entire functions f with order at most one, since this is
the only case which we shall be concerned later. Since the series
X
|⇢| 1 "
⇢6=0
converges, then ⇣
X z⌘ z X
2
log 1 + ⌧ |⇢| < 1.
⇢6=0
⇢ ⇢ ⇢6=0
62
Therefore we can define the product
Y⇣ z ⌘ ⇢z
P (z) = 1 e . (3.4.3)
⇢6=0
⇢
Since the total length of all the intervals (|⇢| |⇢| 2 , |⇢| + |⇢| 2 ) on the real line is finite, there
are arbitrarily large numbers R such that
2
R |⇢| > |⇢| 8⇢ 6= 0. (3.4.4)
Let P (z) = P1 (z)P2 (z)P3 (z), where these are the subproducts extended over the following sets:
1
P1 : |⇢| < R,
2
1
P2 : R 6 |⇢| 6 2R,
2
P3 : |⇢| > 2R.
it follows that X
1
log |P1 (z)| > R |⇢| R1+" .
0<|⇢|< 12 R
Thus
|P1 (z)| exp( R1+" ).
For the factors of P2 , using (3.4.4) we get
⇣ z ⌘ ⇢z |z ⇢|
1 e >e 2
R 3.
⇢ 2R
"
By Corollary 3.4.0.1 the numbers of factors is less than R1+ 2 . Hence
"
1+ 2
|P2 (z)| (R 3 )R exp( R1+" ).
Finally, for the factors of P3 , note that there exists some C > 0 such that
⇣ z ⌘ ⇢z ⇣ z⌘ z ⇣ R ⌘2
log 1 e > log 1 + > C .
⇢ ⇢ ⇢ |⇢|
Hence ⇣ z ⌘ ⇢z R 2
1 e >e C( |⇢| )
.
⇢
We also have X X
2 1+" 1 "
|⇢| < (2R) |⇢| ,
|⇢|>2R ⇢6=0
63
and therefore
|P3 (z)| exp( R1+" ).
It follows that , on |z| = R, we have
Now, since ⇠ has order at most one (Lemma 3.4.1) we conclude that
Y⇣ s ⌘ ⇢s
⇠(s) = eA+Bs 1 e .
⇢
⇢
1
Let h be a complex-valued analytic function on a bounded domain D such that h extends continuously
to the boundary @D of D. The maximum modulus principle states that if |h(z)| 6 M for all z 2 @D, then
|h(z)| 6 M for all z 2 D. In other words, h attains its maximum modulus on the boundary.
64
Taking the logarithmic derivative, we get
⇠ 0 (s) X⇣ 1 1⌘
=B+ + .
⇠(s) ⇢
s ⇢ ⇢
Since ⇣s⌘
1 s
⇠(s) = s(s 1)⇡ 2 ⇣(s),
2 2
then
⇣ 0 (s) 1 1 1 ⇣s ⌘ X⇣ 1 1⌘
=B + log ⇡ +1 + + ,
⇣(s) s 1 2 2 2 ⇢
s ⇢ ⇢
If ⇢ = +i 2
with 0 6 6 1, then taking the real part of (3.4.7) we get
X 2
+ ⌧ log T.
⇢
(2 )2 + (T )2 2 + 2
Since
2 1 1
> ,
(2 )2 + (T )2 4 + (T )2 1 + (T )2
and > 0, then
X 1
⌧ log T.
⇢
1 + (T )2
Therefore, the number of zeros ⇢ = + i of ⇣ in the box 0 < < 1, T 1< 6 T + 1 satisfies
X
N (T + 1) N (T 1) = 1
T 1< 6T +1
X 1
62
T 1< 6T +1
1 + (T )2
X 1
62
⇢
1 + (T )2
⌧ log T. (3.4.8)
For future references we write the previous result in the following Lemma:
Lemma 3.4.4. The number of zeros ⇢ = + i of ⇣ in the box 0 < < 1, T < 6 T +1
satisfies
N (T + 1) N (T ) ⌧ log T
2
Do not confuse with the Euler-Mascheroni constant which is also denoted by .
65
Proof. We have
N (T + 1) N (T ) 6 N (T + 1) N (T 1) ⌧ log T. ⌅
Theorem 3.4.2. For s = + it in the strip 16 6 2, we have
⇣ 0 (s) X 1
= + O(log t) (t ! 1),
⇣(s) s ⇢
|t |<1
where the sum runs over the nontrivial zeros ⇢ = + i of ⇣ for which |t | < 1.
⇣ 0 (2 + it)
Since is bounded for large t by (3.3.3), then
⇣(2 + it)
⇣ 0 (s) X ⇣ 1 1 ⌘
= + O(log t).
⇣(s) ⇢
s ⇢ 2 + it ⇢
For the terms with |t | < 1, we have |2 + it ⇢| > 1. Thus, using (3.4.8) we obtain
X 1 X
6 1 ⌧ N (t + 1) N (t 1) ⌧ log t.
2 + it ⇢ t 1< <t+1
|t |<1
The purpose of this section is to count the number of zeros of ⇣ in the critical strip with
imaginary part |Im(s)| < T for any positive real number T . Once we know how many zeros
should lie in a region, we can verify the Riemann hypothesis in that region computationally.
For this purpose we need the argument principle, for a proof see [5].
Theorem 3.5.1 (The argument principle). Let f be meromorphic in a domain interior to a
positively oriented simple contour L such that f is analytic and nonzero on L. Then
where Z and P are the number of zeros and the number of poles of f inside L, counting
multiplicities, and L arg f (z) counts the changes in the argument of f along the contour L.
66
Using the argument principle we are ready to prove the following theorem which was conjectured
by Riemann and proved by von Mangoldt.
Theorem 3.5.2. Let N (T ) be the number of zeros ⇢ = + i of ⇣ in the rectangle
Then
T T T
N (T ) = log + O(log T ) (T ! 1). (3.5.1)
2⇡ 2⇡ 2⇡
Proof. Since ⇣ and ⇠ have the same zeros (counted with multiplicity), we shall work with the
entire function ⇠ rather than with ⇣. Let L be the positively oriented rectangle with vertices
1, 2, 2 + iT, 1 + iT , as in Figure 5.1.
1
1 + iT 2
+ iT 2 + iT
1 1 1 2
2
Hence
1
N (T ) =M arg ⇠(s).
⇡
Now we compute the variation of the argument of
1 s
⇣s⌘
⇠(s) = s(s 1)⇡ 2 ⇣(s)
2 2
for each factor separately along every segment. There is no variation from the line segment
1
from to 2, because every factor is real. Now,
2
1 ⇣1 ⌘⇣ 1 ⌘ ⇣1 ⌘
M arg s(s 1) = arg + iT + iT = arg + T 2 = ⇡,
2 2 2 4
67
and
s T 1
i T2
M arg ⇡ 2 = arg ⇡
log ⇡. 4 =
2
To compute the argument variation of we use Stirling’s series (2.2.12) in the form
⇣ 1⌘ 1 ⇣1⌘
log (s) = s log s s + log 2⇡ + O ,
2 2 |s|
to get
⇣s⌘ ⇣1 T⌘
M arg = Im log +i
2 4 2
n⇣ 1 T⌘ ⇣ 1 T⌘ ⇣ 1 T⌘ 1 ⇣ 1 ⌘o
= Im +i log +i +i + log 2⇡ + O
4 2 4 2 4 2 2 T
⇡ T T T ⇣1⌘
= + log +O .
8 2 2 2 T
Adding up the above results we obtain
T T T 7 ⇣1⌘
N (T ) = log + + S(T ) + O , (3.5.2)
2⇡ 2⇡ 2⇡ 8 T
where ⇣1 ⌘
1 1
S(T ) ··=M arg ⇣(s) = arg ⇣ + iT ,
⇡ ⇡ 2
provided this argument is defined by continuous variation along L, or, equivalently, by contin-
1
uous variation along the horizontal segment from 1 + iT to + iT starting at 1 + iT with
2
the value 0. The formula (3.5.1) will follow from (3.5.2) if we show that
S(T ) ⌧ log T.
Therefore, using Theorem 1.1.1 to interchange summation and integration we conclude that
the integral (3.5.3) along the segment from 1 + iT to 2 + iT is bounded by the constant
Z1 ⇣ X
⇤(n) ⌘
1 X ⇤(n) 1
d = .
2 n=1
n n=1
n2 log n
Now,
Z 1 +iT ⇣ 1 ⌘
2
Im ds = | arg(s ⇢)| 6 ⇡.
2+iT s ⇢
Therefore, using Theorem 3.4.2 we get
Z 1 +iT X Z 12 +iT ⇣ 1 ⌘ X
2 ⇣ 0 (s)
Im ds ⌧ Im ds + log T ⌧ 1 + log T.
2+iT ⇣(s) 2+iT s ⇢
|t |<1 | T |<1
68
By (3.4.8) the number of zeros with | T | < 1 is O(log T ). Thus
S(T ) ⌧ log T,
1
Letting y = ex , s = + it and using the fact that ⌅ is an even function we get
2
Z1
1
(x) = g(it)g( it)⌅(t)y it dt
2 1
Z 1 +i1 ⇣
1 2 1⌘ ⇣1 ⌘
= p g s g s ⇠(s)y s ds
2 yi 1
2
i1 2 2
Z ⇣ 1⌘ ⇣1 ⌘ ⇣ s⌘
1
+i1
1 2 s
= p g s g s (s 1) 1+ ⇡ 2 ⇣(s)y s ds.
2 yi 1
2
i1 2 2 2
1
Taking g(s) = 1 we obtain
s+ 2
Z 1 +i1
1 2 1 ⇣ s⌘ s
(x) = p 1+ ⇡ 2 ⇣(s)y s ds
2 yi 1
2
i1 s 2
Z 1
+i1 ⇣s⌘
1 2 s
= p ⇡ 2 ⇣(s)y s ds.
4 yi 1
2
i1 2
69
1 1
Consider the positively oriented rectangle with vertices at iR, + iR, 2 iR, 2 iR, as
2 2
shown the Figure 3.4.
1
2
+ iR 2 + iR
iR
1 2
2
iR 1
2
iR 2 iR
Since n 1 ⇣s⌘ o ⇡p
s
2⇡i Res p ⇡ 2 ⇣(s)y s = y,
s=1 4 yi 2 2
then letting R ! 1, by the Residue Theorem it follows that
Z 2+i1 ⇣ ⌘
1 s s ⇡p
(x) = p ⇡ 2 ⇣(s)y s ds + y
4 yi 2 i1 2 2
1 Z p
1 X 2+i1 ⇣ s ⌘⇣ ⇡n ⌘ s ⇡p
= p ds + y,
4 yi n=1 2 i1 2 y 2
where the last equality follows by the definition of ⇣(s) for Re(s) > 1, and we used Theorem 1.1.1
to interchange summation and integration. Now, by Mellin inversion (Theorem 1.5.5) we know
that for any c is Z
t 1 c+i1 s
e = t (s) ds.
2⇡i c i1
Therefore
⇡ ⇣ 1 ⌘ ⇡p
(x) = p ' 2 + y,
y y 2
where 1
X
n2 ⇡t
'(t) = e .
n=1
This shows that Z1
⌅(t) ⇡ x x
2x
1 cos xt dt = e2 2e 2 '(e ) .
0 t2 + 4 2
Letting x = i↵ we get
Z1 n1 o
2 ⌅(t) ↵ i↵
2i↵
1 cosh ↵t dt = 2 cos 2e + '(e ) . (3.6.1)
2
⇡ 2
t +4 2 2
0
70
Note that if z = x + iy, x > 0, then
1 1
x |y| arg z
|z z 2 e z
| = |z|x 2 e
1 ⇡
n ⇣ ⇡ ⌘o
|y|
= |z|x 2 e 2 exp x |y| arg z
2
1 ⇡
6 |z|x 2 e 2
|y|
.
Hence, by Stirling’s formula (2.2.13) we have
⇣1 1 ⌘ ⇣1 1 ⌘ 1
2
+ 12 it 1 1 1 ⇡t
it
+ it ⌧ + it e 4 2 ⌧t 2 e 4 . (3.6.2)
4 2 4 2
Now, since
1 ⇣ 2 1 ⌘ 1 1 it ⇣ 1 1 ⌘ ⇣ 1 ⌘
⌅(t) = t + ⇡ 4 2 + it ⇣ + it ,
2 4 4 2 2
then using the integral representation for ⇣ (3.1.1) and (3.6.2) we get
⇡t
⌅(t) ⌧ t e 4 ,
for some > 0. This shows that the integral in (3.6.1) may be di↵erentiated with respect to ↵
⇡
any number of times provided that ↵ < . Thus
4
Z1
2 ⌅(t) 2n ( 1)n cos ↵2 d 2n i↵ n 1 2i↵
o
t cosh ↵t dt = 2 2n e 2 + '(e ) . (3.6.3)
⇡ 0 t2 + 14 22n 1 d↵ 2
Now we shall prove that for all n 2 N, we have
d 2n n 1 2i↵
o
+ '(e ) ! 0.
d↵2n 2 ↵! ⇡4
71
1
From this it is easy to see that + '(x) and all its derivatives tend to zero as x ! i along any
2
⇡ ⇡
path in the cone | arg(x i)| < . This proves the claim. Therefore, letting ↵ ! in (3.6.3),
2 4
we obtain Z1
⌅(t) 2n ( 1)n ⇡ cos ⇡8
lim t cosh(↵t) dt = . (3.6.4)
↵! ⇡4 0 t2 + 14 22n
To prove that ⌅ has infinite real zeros we will show that ⌅ changes sign an infinite number of
times. Assume for contradiction that ⌅(t) doesn’t change sign for t > T . By (3.6.4) we know
that Z1
⌅(t) 2n
lim⇡ 2 1 t cosh(↵t) dt = L,
↵! 4 T t +
4
for some L 2 R. Without loss of generality suppose that ⌅(t) > 0 for all t > T . Then
ZT0
⌅(t) 2n
t cosh ↵t dt 6 L
T t2 + 14
⇡ ⇡
for all ↵ < and T 0 > T . Therefore, letting ↵ ! we get
4 4
ZT0
⌅(t) 2n ⇡t
1 t cosh dt 6 L.
T t2+4 4
for every n 2 N. This, however, is impossible; for, taking n odd, the right-hand side is negative,
hence
Z1 ZT
⌅(t) 2n ⇡t ⌅(t) 2n ⇡t
2 1 t cosh dt < 2 1 t cosh dt
T t + 4 4 0 t + 4 4
< KT 2n ,
where K is independent of n. But by hypothesis there exists m = m(T ) > 0 such that
⌅(t)
> m,
t2+ 14
for 2T 6 t 6 2T + 1. Therefore
Z1 Z 2T +1
⌅(t) 2n ⇡t
2+ 1
t cosh dt > mt2n dt > m(2T )2n ,
T t 4
4 2T
72
3.7 The Explicit Formula
Note that has jump discontinuities for prime powers. This leads us to consider the function
0 , that is the same at except that at its jump discontinuities it takes the value halfway
between the values to the left and the right, that is
(
(x), if x 6= pm for some p 2 P and for some m 2 N⇤ ;
0 (x) ·
· = .
(x) 12 ⇤(x), otherwise.
The purpose of this section is to prove an explicit formula for 0 that is related to the zeros of
the Riemann zeta function, namely
X x⇢ 1 ⇣ 1⌘
0 (x) = x log 2⇡ log 1 ,
⇢
⇢ 2 x2
where the sum over the nontrivial zeros ⇢ = + i of ⇣ is to be understood in the symmetric
sense as X x⇢
lim .
T !1 ⇢
| |6T
This formula was conjectured by Riemann and proved by von Mangoldt in 1895. For y > 0, let
8
>
>
1
Z ↵+i1 s
y <0, if 0 < y < 1;
(y) = ds = 12 , if y = 1 , (3.7.2)
2⇡i ↵ i1 s >
>
:1, otherwise.
x
Letting y = in (3.7.2) and summing over n > 1 we get
n
Z ↵+i1 h
1 ⇣ 0 (s) i xs
0 (x) = ds.
2⇡i ↵ i1 ⇣(s) s
The idea of the proof is to move the vertical line of integration away to infinity on the left, and
then express 0 (x) as the sum of the residues of the function
h ⇣ 0 (s) i xs
⇣(s) s
73
1
at its poles. The pole of ⇣(s) at s = 1 contributes x; the pole of at s = 0 contributes
s
⇣ 0 (0)
= log 2⇡;
⇣(0)
To carry out this proof, we begin by proving a technical lemma which is known as Perron’s
formula.
Lemma 3.7.1. For y > 0 let (y) be defined as in (3.7.2), and for T > x > 0 let
Z ↵+iT
1 ys
(y, T ) ··= ds.
2⇡i ↵ iT s
If y 6= 1 then n o
1
| (y) (y, T )| 6 y ↵ min 1, (3.7.3)
T | log y|
and for y = 1
↵
| (1) (1, T )| 6 .
T
Proof. First suppose that 0 < y < 1. Consider the positively oriented rectangle with vertices
at R it, R + it, ↵ + iT, ↵ iT . Since
ys
! 0,
s !1
Now, Z 1+iT Z1
ys 1 y↵
ds 6 y d = ,
↵+iT s T ↵ T | log y|
and similarly for the other integral. This shows that
y↵
| (y, T )| 6 .
T | log y|
Next we use Cauchy’s theorem to move the vertical line of integration to the right arc of the
circle |s| = |↵ + iT | (see the figure below).
74
↵ + iT
↵ iT
where the term containing ⇤(x) is present only if x is a prime power. Now we proceed to
estimate the series on the right of (3.7.4). Since ↵ > 1 is arbitrarily we may choose
1
↵=1+ .
log x
75
3 5
Note that x↵ = e x. We first take the terms with n 6 x or n > x. For these terms,
4 4
x
log ⌧ 1.
n
Note that as ↵ ! 1, x ! 1. Thus
⇣ 0 (↵) 1
⌧ = log x (x ! 1)
⇣(↵) ↵ 1
Hence, the contribution of these terms to the sum is
x h ⇣ 0 (↵) i
1
x X ⇤(n) x
⌧ ↵
= ⌧ log x.
T n=1 n T ⇣(↵) T
3
Now we consider the terms for which x < n < x. Let q be the largest prime power less than
4
3
x; we can suppose that x < n < x, since otherwise the terms under consideration vanish. For
4
the term n = q, since
log(1 + t) 6 t (t > 0)
we have ⇣
x x q⌘ x q
log = log 1 > .
n x x
Therefore the contribution of this term is
⇣ x ⌘↵ n x o n x o
⌧ ⇤(q) min 1, ⌧ log x min 1, .
n T (x q) T (x q)
x
For the other terms, we can put n = q ⌧ , where 0 < ⌧ < . Then
4
x q ⇣ ⌧⌘ ⌧
log > log = log 1 > .
n n q q
Hence the contributions of this terms is
X q x
⌧ ⇤(q ⌧ ) ⌧ log2 x.
x ⌧T T
0<⌧ < 4
5
The terms with x < n < x are dealt similarly, except that q is replaced by q̃, the least prime
4
power greater than x. Let hxi denote the distance from x to the nearest prime power, other
than x itself. Then combining all the obtained estimates we get
x log2 x n x o
| 0 (x) (x, T )| ⌧ + log x min 1, . (3.7.5)
T T hxi
M + iT ↵ + iT
8 6 4 2 0 1 ↵
M iT 1 ↵ iT
2
76
Now consider the positively oriented rectangle with vertices at ↵ iT, ↵+iT, M +iT, M iT ,
where M is a large negative odd integer, as shown in Figure 3.6. As discussed previously the
sum of the residues of the integrand inside the rectangle is
X x⇢ X x 2n
x log 2⇡ .
⇢ 0<2n<M
2n
| |<T
N (T + 1) N (T ) ⌧ log T.
Hence, we can choose T so that no zero ⇢ of ⇣ has height near T . Specifically we may require
1
|T | .
log T
⇣ 0 (s) X 1
= + O(log T ) (T ! 1),
⇣(s) s ⇢
|T |<1
where ⇢ = + i runs over the nontrivial zeros of ⇣. With our choice of T , each term of the
sum is O(log t) and by Lemma 3.4.4 the number of terms is also O(log t). Thus
⇣ 0 (s)
⌧ log2 |s| (3.7.6)
⇣(s)
for s on such segment. To extend the previous bound for 6 1 we use the functional equation
(3.2.3) in the form ⇣ s⇡ ⌘
⇣(1 s) = 21 s ⇡ s cos (s)⇣(s).
2
We shall use this formula since, if 1 6 1 the functions on the right have to be considered
only for > 2. Therefore by Stirling’s formula (2.2.13) it follows that
⇣ 0 (1 s)
⌧ log 2|1 s| ( > 2).
⇣(1 s)
⇣ 0 (s)
⌧ log2 |s|. (3.7.7)
⇣(s)
77
Theorem 3.7.1. For x > 2 and for T > 0 we have
X x⇢ 1 ⇣ 1⌘
0 (x) = x log 2⇡ log 1 + R(x, T ), (3.7.9)
⇢ 2 x2
| |<T
x log2 (xT ) n x o
R(x, T ) ⌧ + log x min 1, . (3.7.10)
T T hxi
where the series over the nontrivial zeros ⇢ = + i of ⇣ is evaluated as the limit
X x⇢
lim .
T !1 ⇢
| |<T
Note that if x 2 Z then hxi > 1 and hence (3.7.10) takes the simpler form
x
|R(x, T )| ⌧ log2 (xT ). (3.7.11)
T
In 1896 Jacqued Hadamard and de la Vallée Poussin proved independently that if ⇡(x) denotes
the number of primes less than or equal to x then
x
⇡(x) ⇠ (x ! 1).
log x
This result is known as the prime number theorem (PNT). Recall that in Theorem 2.4.4 we
showed that the PNT is equivalent to
(x) ⇠ x (x ! 1).
Using the results of the previous section we shall now deduce that there exists c > 0 such that
p
(x) = x + O x exp( c log x)
Lemma 3.7.2. Let s = + it. Then, there exists c > 0 such that
c
⇣(s) 6= 0 for >1 .
log t
78
Proof. As usual, let s = + it and ⇢ = + i be a zero of ⇣. For > 1 we have
1
⇣ 0 (s) X ⇤(n)
= s
.
⇣(s) n=1
n
and following the same line of argument as in the proof of Theorem 3.3.1 it is easy to see that
h ⇣ 0( ) i h ⇣ 0 ( + it) i h ⇣ 0 ( + 2it) i
3 +4 Re +2 Re > 0. (3.7.12)
⇣( ) ⇣( + it) ⇣( + 2it)
⇣ 0( ) 1
= + O(1).
⇣( ) 1
⇣ 0 ( + it) 1
Re < A log t Re ,
⇣( + it) + it ⇢
and
⇣ 0 ( + 2it)
Re ⌧ log t.
⇣( + 2it)
Taking t = we have
1 1
Re = = .
+ it ⇢ ( )2 + (t )2
79
Proof. By the previous lemma there exists c such that if | | < T then
c
<1 .
log T
Therefore c
|x⇢ | = |x | 6 x1 log T
Therefore X x⇢ c
⌧ x(log2 T )x log T .
⇢
| |<T
Without loss of generality we can assume that x 2 Z. Hence using Theorem 3.7.1 and the form
of R(x, T ) as in (3.7.11) we obtain
X x⇢ ⇣x ⌘
(x) = x +O log2 (xT )
⇢
⇢ T
Thereby we get n
2 1o x
| (x) x| ⌧ x log (xT ) x + log T
T
p
Taking T = exp c log x we get (3.7.13) with a di↵erent constant. ⌅
Corollary 3.7.2.1 (The Prime Number Theorem). Let ⇡(x) denote the number of primes less
than or equal to x. Then
x
⇡(x) ⇠
log x
80
Chapter 4
Pontryagin Duality
From this definition it follows at once that translation (on either side) by any given group
element is a homeomorphism (biyective and continuous with continuous inverse) from G onto
itself. To be more precise, for all g 2 G the functions
Lg : G ! G
h 7! gh
and
Rg : G ! G
h 7! hg
are homeomorphisms. Therefore the topology is translation invariant in the sense that for all
g 2 G and U ✓ G the following three assertions are equivalent:
81
(i) U is open.
(ii) gU is open.
(iii) U g is open.
1 1
Moreover, since inversion is a homeomorphism , U is open if and only if U = {h : h 2 U}
is open. Now let X be any topological space and let
Let S ✓ X. We say that X is a homogeneous space under S if for all x, y 2 X, there exists
f 2 S such that f (x) = y. If S = Homeo(X), then we say that X is a homogeneous space.
Clearly any topological group G is homogeneous, since for any g, h 2 G, Lhg 1 2 Homeo(G) and
Lhg 1 (g) = h. From this it follows at once that a local base at the identity e 2 G determines
a local base at any point in G, and therefore the entire topology. This is one of the most
important facts about topological groups.
The unit circle T ··= {z 2 C | |z| = 1} with the usual complex product.
Thus we explicitly distinguish between X (n) and X n , the n-fold Cartesian product of X with
itself. A subset S of G is called symmetric if S = S 1 . The following proposition gives some
basic results about topological groups:
(i) Every neighborhood U of the identity contains a neighborhood V of the identity such that
V (n) ✓ U .
(ii) Every neighborhood U of the identity contains a symmetric neighborhood V of the identity.
82
Proof. (i) Without loss of generality assume that U is open. Consider the continuous map
m: Un ! G
n
Y
(g1 , . . . , gn ) 7! gj .
j=1
Then m 1 (U ) is open and contains the point (e, . . . , e). By definition of the topology on
U n , there exists open sets V1 , . . . , Vn of U such that (e, . . . , e) 2 V1 , . . . Vn . Let
n
\
V ··= Vj .
j=1
µ(?) = 0;
If A1 , A2 , . . . are disjoint sets in M, then
⇣[
1 ⌘ 1
X
µ Ak = µ(Ak ). |
k=1 k=1
83
Think of µ(S) as the “volume” of S. A set N ✓ X is called a null set if N is contained in a
measure 0 set. It is convenient to enlarge M so that all null sets are measurable. We say that
f : X ! C is a null function if {x 2 X | f (x) 6= 0} is a null set.
Example 4.1.3. Let X be any nonempty set, and let M = 2X . Let x0 2 X and define
(
1, if x0 2 A,
µ(A) =
0, otherwise.
A measure space is a triple (X, M, µ), where X is a set, M is a -algebra on the set X and µ
is a measure on (X, M).
Example 4.1.4. (Rn , L, ), where L is the class of all Lebesgue measurable sets A ✓ Rn and
is the Lebesgue measure on Rn .
Definition 4.1.5 (Integration). Let (X, M, µ) be a measure space. Given an element S 2 M
with µ(S) < 1, let 1S be the characteristic function on S, that is
(
1, if x 2 S,
1S (x) = .
0, otherwise.
Define Z
1S dµ = µ(S).
this leads to a notion of distance and Cauchy sequence1 . We say that f : X ! C is integrable if
outside of a null set it equals the pointwise limit of some L1 -Cauchy sequence of step functions
{fj }j2N ; then define Z Z
f dµ = lim fj dµ.
j!1
agrees with the previous one if f is integrable, and yields 1 if f is not integrable.
1
A sequence {fn } is said to be L1 -Cauchy if for every " > 0 there is a N = N (") such that kfm f n k1 < "
for every m, n > N .
2
As usual, f > g means that f (x) > g(x) for every x 2 X.
84
For a measurable f : X ! C, it turns out that f is integrable if and only if |f | is integrable.
Now we mention two important theorems for interchanging limits and integrals; for a proof
see [6].
Theorem 4.1.1 (Monotone Convergence Theorem). Suppose that {fn }n2N is a sequence of
measurable functions X ! [0, 1] such that 0 6 f1 6 f2 6 · · · . Let f be the pointwise limit of
the fn . Then Z Z
fn dµ ! f dµ
n!1
Theorem 4.1.2 (Dominated Convergence Theorem). Suppose that {fn }n2N is a sequence of
measurable functions X ! C such that f is the pointwise limit of the fn . If there is an integrable
function g : X ! C such that |fn | 6 |g| for all n 2 N, then for every n, fn and f are integrable,
and Z Z
fn dµ ! f dµ
n!1
µ⇤ (?) = 0,
µ⇤ (A) = µ⇤ (A \ B) + µ⇤ (A \ B C ).
Note that the subadditivity of the outer measure µ⇤ implies that for all A ✓ X we have
µ⇤ (A) 6 µ⇤ (A \ B) + µ⇤ (A \ B C ).
µ⇤ (A) > µ⇤ (A \ B) + µ⇤ (A \ B C )
holds for all A ✓ X such that µ⇤ (A) < 1. The following well known theorem is the fundamental
fact about outer measures; for a proof see [6].
Theorem 4.1.3. Let X be a set, let µ⇤ be an outer measure on X, and let Mµ⇤ be the collection
of all µ⇤ -measurable subsets of X. Then
Mµ⇤ is a -algebra,
Let X be a set and let M ✓ 2X . Then, we shall write [M] to denote the -algebra generated
by M. We need some definitions.
85
Definition 4.1.6 (Topological Measure Space). A topological measure space is a measure space
(X, M, µ) such that X is a topological space and M is the collection of Borel subsets of X.
Definition 4.1.8 (Radon Measure). Let (X, M, µ) be a Borel measure space. We say that
µ 6= 0 is a Radon measure if
Definition 4.1.9 (Locally Compact Group). A locally compact group is a topological group
G that is locally compact and Hausdor↵.
Definition 4.1.10 (Haar Measure). Let G be a topological group. A left Haar measure (resp.
right Haar measure) on G is a Radon measure µ on G such that µ(gA) = µ(A)(resp. µ(Ag) =
µ(A)) for all g 2 G and all measurable subsets A of G.
Consider the locally compact group G ··= GL(n, R). Then, a Haar measure on G is given
by Z
1
µ(S) = n
d (X),
S | det(X)|
2
where is the Lebesgue measure on Rn identified with the set of all n ⇥ n matrices.
Now we are ready to prove some results that will be used to prove the existence and uniqueness
of a left Haar measure on a locally compact group.
⇥ ⇤
Lemma 4.1.1. Let f : X ! Y and let E ✓ 2Y . Then, f 1 (E) = f 1 [E] .
1 1
⇥ 1
⇤
Proof. (✓) It is easy to verify that f [E] is a -algebra containing f (E). Hence f (E) ✓
f 1 [E] .
⇥ ⇤
(◆) Let M ··= A ✓ Y | f 1 (A) 2 f 1 (E) . Then, M is a -algebra containing E, and
therefore [E] ✓⇥M. Now,
⇤ let A 2 f 1 [E]⇥ . Then,
⇤ A = f 1 (B) for some B 2 [E] ✓ M, so
f 1 (B) = A 2 f 1 (E) . It follows that f 1 (E) = f 1 [E] . ⌅
(1) A 2 M.
86
(2) f (A) 2 M.
1
(3) f (A) 2 M.
Lemma 4.1.2. Let X be a Hausdor↵ space, let K ✓ X compact, and let U1 , U2 be open subsets
of X such that K ✓ U1 [ U2 . Then, there exists compact sets K1 ✓ U1 and K2 ✓ U2 such that
K = K1 [ K2 .
Proof. Let L1 ··= K \ U1 and L2 ··= K \ U2 . Since X is Hausdor↵ and K is compact, it follows
that L1 and L2 are compact. Since K ✓ U1 [ U2 , then L1 \ L2 = ?. Hence, since L1 and L2 are
disjoint compact subsets of a Hausdor↵ space, we can separate them with disjoint open sets,
say V1 and V2 respectively. Now, let K1 = K \ V1 and K2 = K \ V2 . Then, similarly as before,
K1 and K2 are compact. Now,
K1 = K \ V1 ✓ K \ L1 = K \ (K \ U1 ) = K \ (K C [ U1 ) ✓ U1 .
Lemma 4.1.3. Let (X, M, µ) be a measure space, let f : X ! R be measurable, and let A =
{x 2 X | f (x) > 0}. If µ(A) > 0, then, there is some a > 0 such that µ {x 2 X | f (x) >
a} > 0.
Proof. Suppose for the sake of contradiction that µ {x 2 X | f (x) > a} = 0 for all a > 0. Let
Sk ··= {x 2 A | f (x) > k 1 }. Then,
1
[
A= Sk ,
k=1
so 1
X
µ(A) 6 µ(Sk ) = 0,
k=1
which is impossible. ⌅
Now, we shall see how to obtain a right Haar measure from a left Haar measure, and viceversa.
Theorem 4.1.4. Let G be a topological group, let µ be a Haar measure on G, and define
µ̃(A) = µ(A 1 ). Then, µ is a left (resp. right) Haar measure if and only if µ̃ is a right (resp.
left) Haar measure on G.
Proof. It is easy to check that if µ is a Haar measure, then µ̃ is a regular Borel measure on G (By
Corollary 4.1.1.1, µ̃ is defined exactly on the Borel subsets of G). The identity (Ax) 1 = x 1 A 1
then implies that for each x 2 G and each A 2 B(G), we have
This shows that µ is a left Haar measure if and only if µ̃ is a right Haar measure. ⌅
This theorem shows that we may as well simply concern ourselves with the study of left Haar
measures.
Lemma 4.1.4. Let G be a topological group, let K ✓ G compact, and let U be an open subset
of G such that K ✓ U . Then, there is an open set V containing the identity such that KV ✓ U .
87
Proof. For each x 2 K, let Wx ··= x 1 U . Then, Wx is an open neighbourhood of the identity.
Now, let Vx be an open neighborhood of the identity such that Vx Vx ✓ Wx . Then, the collection
{xVx | x 2 K} is an open cover of K, so there is a finite collection of points x1 , . . . , xn such
that n
[
K✓ x k V xk .
k=1
Let n
\
V ··= V xk .
k=1
In case X is a locally compact Hausdor↵ space, we shall denote by K (X) the set of those
continuous functions f : X ! R for which supp(f ) is compact. Now we prove a couple of
lemmas about topological groups that will be used in the proof of uniqueness.
Lemma 4.1.5. Let G be a locally compact group and let f 2 K (G). Then, for every " > 0,
there is an open neighborhood U of the identity such that if y 2 xU , then |f (x) f (y)| < ".
Proof. Let K ··= supp(f ) and let " > 0. Since f is continuous, then for each x 2 K there exists
an open neighborhood Ux of the identity such that if y 2 xUx , then |f (y) f (x)| < ". Now,
for each x 2 K, choose another neighborhood of the identity Vx such that Vx Vx ✓ Ux . Since K
is compact, there is a finite number of elements x1 , . . . , xn such that
n
[
K✓ x k V xk .
k=1
Let n
\
V ··= V xk ,
k=1
and let U ··== V \ V . Clearly U is an open neighborhood of the identity. Now, let y 2 xU .
1
If x, y 2
/ K, then |(f (x) f (y)| = 0, so there’s nothing to prove. Without loss of generality,
assume that x 2 K. Then x 2 xk Vxk for some 1 6 k 6 n, and hence x 2 xk Uxk . On the other
hand, since x 2 xk Vxk and V ✓ Vxk , then y 2 xk Uxk . Therefore,
Lemma 4.1.6. Let G be a topological group and let µ be a left Haar measure on G. Then, for
every x 2 G and for every f 2 L1 (G) we have
Z Z
f (xg)dµ = f (g)dµ.
G G
88
Proof. Step 1: Prove for characteristic functions.
Let A be measurable. Then
Z Z
1
A (xg)dµ = µ(x A) = µ(A) = A (g)dµ.
G G
Before we prove the existence and uniqueness of a left Haar measure, we shall need the following
version of the Riesz representation theorem and a more general version of Fubini’s theorem; for
a proof see [6].
Theorem 4.1.6 (Fubini’s Theorem). Let X and Y be loaclly compact Hausdor↵ spaces, let µ
and ⌫ be regular Borel measures on X and Y respectively, and let f 2 K (X ⇥ Y ). Then the
iterated integrals Z Z Z Z
f (x, y)dµd⌫ and f (x, y)d⌫dµ
Y X X Y
exist and are equal.
Theorem 4.1.7 (Existence of a Left Haar Measure). Let G be a locally compact group. Then,
there exists a left Haar measure on G.
89
Proof. Step 1: Define (K : V )).
Let K ✓ G with K compact, and let V ✓ G such that V̊ 6= ?. Then, {g V̊ | g 2 G} is an open
cover of K, so there exists g1 , . . . , gn 2 G such that
n
[
K✓ gk V̊ .
k=1
Let (K : V ) denote the smallest nonnegative integer for which such a sequence exists.
Step 2: Define µU .
Let K denote the collection of compact subsets of G and let U denote the collection of open
subsets of G containing the identity. Since G is locally compact, there exists K0 ✓ G such that
K0 is compact and K̊0 6= ?. Now, for each U 2 U, let µU : K ! R be defined as
(K : U )
µU (K) = .
(K0 : U )
Then,
m [
[ n
K✓ gi hj U.
i=1 j=1
By step 3, each µU may be thought as a point in X. Doing so, for each V 2 U, let
90
This map is continuous and nonnegative on each C(V ). From this it follows that this map is
also nonnegative at µ, so that µ(K1 ) 6 µ(K2 ).
Step 6: Show that µ(K1 [ K2 ) 6 µ(K1 ) + µ(K2 )).
Let K1 , K2 2 K. Note that µU (K1 [ K2 ) 6 µU (K1 ) + µU (K2 ) for each U 2 U. Proceeding
similarly as in step 5, the map that from X to R defined by
is continuous and nonnegative on each C(V ), and hence nonnegative for µ 2 X. Therefore
µ(K1 [ K2 ) 6 µ(K1 ) + µ(K2 ).
Step 7: Show that µU (K1 [ K2 ) = µU (K1 ) + µU (K2 ), whenever K1 U 1
\ K2 U 1
= ?.
Let K1 , K2 2 K such that K1 U 1
+ K2 U 1
= ?. Let n = (K1 [ K2 : U ) and let g1 , . . . , gn 2 G
such that n
[
K1 [ K 2 ✓ gk U.
k=1
Note that since K1 U 1 + K2 U 1 = ?, then each gk U intersects either K1 or K2 , but not both.
Therefore, there exists a natural number m 6 n such that
m
[ n
[
K1 ✓ gk U and K2 ✓ gk U
k=1 k=m+1
Thus
(K1 : U ) + (K2 : U ) 6 (K1 [ K2 : U ).
Combining this result with the previous step, it follows that µU (K1 [ K2 ) = µU (K1 ) + µU (K2 ).
Step 8: Show that µ(K1 [ K2 ) = µ(K1 ) + µ(K2 ), whenever K1 \ K2 = ?.
Let K1 , K2 2 K such that K1 \ K2 = ?. Then, we may find disjoint open sets U1 and U2
such that K1 ✓ U1 and K2 ✓ U2 . By Lemma 4.1.4 there exists open neighborhoods of the
identity V1 and V2 such that K1 V1 ✓ U1 and K2 V2 ✓ U2 . Let V ··= V1 \ V2 . Since U1 and U2
are disjoint, then K1 V and K2 V are disjoint. Thus, for any U 2 U with U ✓ V 1 , we have
K1 U 1 \ K2 U 1 = ?. Hence, by the previous step, the map that from X to R defined by
(4.1.1) vanishes at each element of C(V 1 ). In particular µ(K1 [ K2 ) = µ(K1 ) + µ(K2 ).
Step 9: Extend µ to all subsets of G .
For U ✓ G open, let
µ(U ) ··= sup{µ(K) | K ✓ U, K 2 K}.
Now we show that if K 2 K and K is open, then these two definitions of µ(K) agree: Trivially,
sup{µ(K 0 ) | K 0 ✓ K, K 0 2 K} 6 µ(K).
This shows that this definition agrees with the previous one. Now, for every A ✓ G, let
91
Similarly as above, this indeed is an extension of our previous definition of µ to all subsets of G.
It also follows that this extension still satisfies the property µ(A1 ) 6 µ(A2 ) whenever A1 ✓ A2 .
Step 10: Show that µ is an outer measure on G .
It is clear that µ is non-negative, that it is monotone, and that µ(?) = 0. To show countable
S {Un }
subadditivity, we first show the result for each countable collection of open set in G. Let
be a countable Sm collection of open subsets of G and let K be a compact subset of n2N Un .
Then, K ✓ j=1 Uj for some m 2SN. Applying Lemma 4.1.2 inductively, we may find compact
sets K1 , . . . , Km such that K = m j=1 Kj and Kj ✓ Uj for 1 6 j 6 m. Then, applying step 6
inductively,
Xm Xm X
µ(K) 6 µ(Kj ) 6 µ(Uj ) 6 µ(Un ).
j=1 j=1 n2N
It follows that
⇣[ ⌘ n [ o X
µ Un = sup µ(K) | K ✓ Un , K 2 K 6 µ(Un ).
n2N n2N n2N
P
Now, let {An } be an arbitrary collection of subsets of G such that n2N µ(An ) < 1. Let " > 0,
then, using (4.1.2), for each n 2 N we may choose an open set Un such that An ✓ Un and
"
µ(Un ) 6 µ(An ) + .
2n
Then, ⇣[ ⌘ X
µ An 6 µ(An ) + ".
n2N n2N
Now,
µ(K [ L) 6 µ (V \ U ) [ (V \ K C ) 6 µ(V ) 6 µ(A) + ".
It follows that
µ(A \ U ) + µ(A \ U C ) 6 µ(A) + 3".
Letting " ! 0 the result follows. Consequently, if ⌧ is the topology of G, then [⌧ ] is included
in the -algebra of µ-measurable sets, and by Theorem 4.1.3 the restriction of µ to [⌧ ] is a
Borel measure.
Step 12: Show that µ is a left Haar measure.
92
By construction, it is clear that µ is regular. Note that for each U 2 U, we have µU (K0 ) = 1,
and the continuous function from X to R defined by
f 7! f (K0 )
is the constant 1 on each C(U ). In particular µ(K0 ) = 1. This shows that µ is nonzero. It
remains to show that µ is left invariant. For this purpose, fix g 2 G. Then, the elements
x1 . . . , xn generate a cover for K if and only if the elements gx1 , . . . , gxn generate a cover of
gK, so that for each U 2 U we have (K : U ) = (gK : U ). Therefore, for each U 2 U ,
µU (K) = µU (gK) for each U 2 U. It follows that the continuous function from X to R defined
by
f 7! f (K) f (gK)
vanishes on each C(U ). In particular µ(K) = µ(gK). This completes the proof. ⌅
The following Lemma gives a basic property of Haar measures and we shall need it in our proof
of Theorem 4.1.8.
Lemma 4.1.7. Let G be a locally compact group, and let µ be a left Haar measure on G. Then,
for every f 2 K (G) nonnegative and not identically 0 we have
Z
f dµ > 0.
G
Proof. Since µ is nonzero, there exists K ✓ G compact such that µ(K) 6= 0. Now, let f 2 K (G)
be nonnegative and not identically 0. Let U ··= f 1 (R+ ). By continuity, U is open and since
K is compact and U 6= ?, there’s a finite number of elements g1 , . . . , gn such that
n
[
K✓ gk U.
k=1
From this it follows that µ(U ) > 0, and hence by Lemma 4.1.3, there is some a > 0 such that
V ··= {g 2 G | f (g) > a} is of positive measure. Therefore
Z Z
f dµ > f dµ > aµ(V ) > 0.
G V
Theorem 4.1.8 (Uniqueness of a left Haar measure). Let G be a locally compact group, and
let µ and ⌫ be two left Haar measures on G. Then, there exists c > 0 such that µ = c⌫.
Proof. Let g 2 K (G) be nonnegative and not identically 0, and let f 2 K (G) be arbitrary.
By Lemma 4.1.7 the function h : G ⇥ G ! R defined by
f (x)g(yx)
h(x, y) ··= Z
g(tx)d⌫(t)
G
93
is well-defined. Since f and g are compactly supported, then h is compactly supported. Now let
us show that h is continuous: For this purpose, it suffices to show that the function ' : G ! R
defined by Z
'(x) ··= g(tx)d⌫(t)
G
is continuous. Let K ··= supp(g), let x0 2 G. Since G is locally compact there exists U ✓ G
1
an open neighborhood of x0 , whose closure is compact. By Tychono↵’s theorem, K ⇥ U
1
is compact, so that KU is compact. It is easy to check that for each x 2 U the function
1
t 7! g(tx) is continuous and vanishes outside KU . Now, let " > 0, and choose > 0 such
1 1
that ⌫(KU ) < " (this is possible since KU is compact, and hence of finite measure).
By Lemma 4.1.5, there is an open neighborhood V of the identity such that if y 2 xV , then
|g(x) g(y)| < . Thus, for each c 2 U \ x0 V and each t 2 G we have tx 2 tx0 V , and so
Z
1
|'(x) '(x0 )| 6 |g(tx) g(tx0 )|d⌫(t) 6 ⌫ KU < ".
G
This shows that ' is continuous, and hence h 2 K (G⇥G). Now, by Fubini’s theorem (Theorem
4.1.6), since h 2 K (G ⇥ G), then the iterated integrals
ZZ ZZ
h(x, y)dµ(x)d⌫(y) and h(x, y)d⌫(y)dµ(x)
we reverse the order, use the translation invariance of the Haar measure µ to replace x with
y 1 x (Lemma 4.1.6), again reverse the order of integration, and finally replace y with xy, we
find that ZZ ZZ
h(x, y)d⌫(y)dµ(x) = h(y 1 , xy)d⌫(y)dµ(x).
Thus Z
f dµ
Z
gdµ
depends on f and g, but not on µ. From this it follows that the Haar measure ⌫ satisfies
Z Z
f d⌫ f dµ
Z = Z ,
gd⌫ gdµ
94
where Z
gd⌫
c= Z > 0.
gdµ
Since this holds for an arbitrary f 2 K (G), the Riesz representation theorem (Theorem 4.1.5)
implies that ⌫ = cµ. ⌅
To begin this section, we shall recall some basic definitions and theorems about functional
analysis. For a complete introduction on this subject see [19].
Definition 4.2.1 (Metric Space). A metric space is a pair (X, d), where X is a set and d is a
metric, i.e., a function d : X ⇥ X ! [0, 1) such that for all x, y, z 2 X we have
A sequence {xn } in a metric space (X, d) is said to be Cauchy if for all " > 0 there is some
N = N (") such that d(xm , xn ) < " for every m, n > N . We say that a metric space (X, d) is
complete with respect to d if every cauchy sequence in X converges.
Definition 4.2.2 (Banach Space). Let K = R or C. A normed space is a pair (X,k·k), where
X is a vector space over K and k·k is a norm, i.e., a function k·k : X ! [0, 1) such that for all
x 2 X and 2 K,
k xk = | |kxk
d(x, y) ··= kx yk .
A Banach space is a complete normed space (complete in the metric defined by the norm).
Example 4.2.1 (Lp spaces). Let (X, M, µ) be a measure space. For 1 6 p 6 1, define
where ✓Z ◆ p1
p
kf kp ··= |f | dµ
if p 6= 0 and
kf k1 ··= ess sup |f | = inf{a 2 R | µ(|f | 1 (a, 1)) = 0)}.
95
Note that strictly speaking, k·kp is not a norm on Lp (X), since there exist nonzero functions f
such that kf kp = 0, namely the null functions. Nevertheless, k·kp induces a norm on
p
Lp (X) ··= L (X) {null functions},
making Lp (X) a Banach space. We will often regard the elements of Lp (X) as functions; this
is well defined as far as integration is concerned.
Definition 4.2.3 (Hilbert Space). Let K = R or C. An inner product space is a pair (X, h·, ·i),
where X is a vector space over K and h·, ·i is an inner product, i.e., a function h·, ·i : X ! K
such that for all x, y, z 2 X and 2 K we have
h x, yi = hx, yi.
hx, xi > 0.
A Hilbert space is a complete inner product space (complete in the metric defined by the inner
product).
Example 4.2.2. Let (X, M, µ) be a measure space. The space L2 (X) is a Hilbert space, under
the inner product defined by Z
hf, gi = f gdµ.
Now we prove one of the most useful inequalities when dealing with inner products:
Lemma 4.2.1 (Cauchy-Schwarz Inequality). Let (X, h·, ·i) be an inner product space over
K = R or C. Then, for all x, y 2 X, we have
Proof. If y = 0, then the inequality holds since hx, 0i = 0. Let y 6= 0. For every 2 K we have
Letting
hy, xi
= ,
hy, yi
we obtain
hy, xi
0 6 hx, xi hx, yi. ⌅
hy, yi
96
An element x of an inner product space X is said to be orthogonal to an element y 2 X if
hx, yi = 0.
Y ? = {z 2 X | z ? Y },
x = y + z,
where y 2 Y and z 2 Z.
Hilbert spaces enjoy the following representation as a direct sum; for a proof see [19].
Theorem 4.2.1. Let Y be any closed subspace of a Hilbert space H. Then
H=Y Y ?.
P: H !Y
x 7! y
is well defined. P is called the orthogonal projection of H onto Y . Note that P is idempotent,
that is,
P 2 = P.
Therefore P Y is the identity operator on Y . The following lemma is particularly useful to
prove density of some subspace Y of H:
Lemma 4.2.2. Suppose M is a subspace of a Hilbert space H and suppose that M has the
following property:
hx, yi = 0 for any y 2 M implies that x = 0.
Then M is dense in H.
The property
hx, yi = 0 for any y 2 M implies that x = 0
simply means that M ? = {0}. Hence M is dense in H. ⌅
97
Definition 4.2.5. Let X, Y be two normed spaces over the same field K = R or C. A linear
operator T (or simply an operator) is a map T : dom(T ) ✓ X ! Y such that for all x, y 2
dom(T ) and ↵, 2 K, we have
T (↵x + y) = ↵T x + T y.
The operator T is said to be bounded if there is some c 2 R such that for all x 2 dom(T ),
kT xk 6 ckxk .
kT xk
= sup
x2dom(T ) kxk
kxk6=0
= sup kT xk .
x2dom(T )
kxk=1
The reader can verify that all the previous definitions agree and that k·kop defines a norm on
B(X, Y ), the space of bounded operators between X and Y . In the case that Y = K, the linear
operator T is called a linear functional and B(X, K) is denoted by X 0 .
The following theorem is easy to prove but it shows that for linear operators, being bounded
is the same as being continuous:
Theorem 4.2.2. Let T : dom(T ) ✓ X ! Y be a linear operator between two normed spaces
X and Y , where dom(T ) is a linear subspace of X. Then T is continuous if and only if T is
bounded.
kxk <
we have
kT xk < ".
Now, let y 6= 0 in dom(T ). Letting
x= y,
kyk
we obtain
kT yk = kT xk < ",
kyk
"
so that kT yk < ckyk, where c = .
98
(() Now suppose that T is bounded and let x0 2 dom(T ). Let " > 0 and suppose that
"
kx x0 k < .
kT kop
Taking all of this into account we are ready to prove one of the most important theorems about
Hilbert Spaces:
Theorem 4.2.3 (Riesz Representation). Let H be a Hilbert space and f 2 H 0 . Then, there is
a unique z 2 H such that for all x 2 H,
Moreover,
kf kop = kzk . (4.2.3)
Applying f , we get
f (v) = 0,
so that v 2 M . Since z0 ? M , we have
f (z0 )
f (x) = hx, z0 i.
hz0 , z0 i
Hence letting
f (z0 )
z= z0 ,
hz0 , z0 i
we obtain f (x) = hx, zi.
Step 2: Prove that z in (4.2.2) is unique.
Suppose that for all x 2 H,
f (x) = hx, z1 i = hx, z2 i.
Then hx, z1 z2 i = 0 for all x 2 H. Letting x = z1 z2 we obtain
hz1 z 2 , z1 z2 i = 0,
99
whence z1 z2 = 0.
Step 3: Formula (4.2.3) holds.
If f = 0 then z = 0 and (4.2.3) holds. Suppose that f 6= 0, so that z 6= 0. Using (4.2.2) with
x = z we obtain
kzk2 = hz, zi = f (z) 6 kf kop kzk ,
whence kzk 6 kf kop . From (4.2.2) and Cauchy-Schwarz inequality (Lemma 4.2.1) we have
Definition 4.2.6 (Sesquilinear form). Let X and Y be vector spaces over the same field K = R
or C. A sesquilinear form ' on X ⇥ Y is a mapping ' : X ⇥ Y ! K such that for all x, z 2 X,
y, w 2 Y and ↵, 2 K, we have
If moreover '(x, y) = '(y, x), then ' is called a Hermitian form. If X and Y are normed spaces,
then we say that ' is bounded if there is some c 2 R such that
Note that the inner product h·, ·i on an inner product space X is a sesquilinear form on X ⇥ X.
Using this definition we can generalize the Riesz representation theorem for the case of bounded
sesquilinear forms:
Theorem 4.2.4 (Riesz Representation for Sesquilinear Forms). Let H1 , H2 be Hilbert spaces
and ' : H1 ⇥ H2 ! K a bounded sesquilinear form. Then for all x 2 H1 and y 2 H2 , we have
'(x, y) = hSx, yi
kSkop = k'k .
100
Therefore
'(x, y) = hz, yi.
From this it follows that with variable x we can define an operator
S : H1 ! H2
x ! z.
It is easy to see that S is a linear operator such that '(x, y) = hSx, yi. Since
k'k = sup |hSx, yi| > sup |hSx, Sxi| = sup kSxk = kSkop ,
kxk=1 kxk=1 kxk=1
kyk=1 kSxk=1
Then, by Lemma ??, Sx = T x for all x 2 H1 , i.e., S = T . This completes the proof. ⌅
Now we mention one of the most important theorems in functional analysis, the Hahn-Banach
Theorem. This is an extension theorem for linear functionals on vector spaces; for a proof
see [19].
Theorem 4.2.5 (The Hahn-Banach Theorem). Let f be a bounded linear functional on a
subspace Z of a normed space X. Then, there exists a bounded linear functional g on X which
is an extension of f to X and has the same norm,
kgkop = kf kop .
To see how powerful the Hahn Banach theorem can be, we mention two useful Corollaries:
Corollary 4.2.5.1. Let X be a normed space and let x0 6= 0 be any element of X. Then there
exists a bounded linear functional g on X such that
Proof. Let
Z ··= { x0 | 2 K}.
Then, Z is a subspace of X. Now consider the mapping f : Z ! K defined by
f (x) = f ( x0 ) = kx0 k .
Then, f is a bounded linear functional with norm kf kop = 1. Therefore by The Hahn-Banach
Theorem, there exists a bounded linear functional g on X such that
101
Proof. From Corollary 4.2.5.1 we see that
We follow [28] closely. An associative algebra (or simply an algebra) A over a field K is a
vector space A over K such that for each a, b 2 A a unique product ab 2 A is defined with the
properties
(a + b)c = ac + bc,
c(a + b) = ca + cb,
(↵a)( b) = ↵ (ab),
(ab)c = a(bc).
ab = ba.
1A a = a1A = a.
Definition 4.2.7. Let A be a complex algebra that also admits the structure of a complex
Banach space. We say that A is a Banach algebra if k·k is submultiplicative, i.e., for all a, b 2 A
we have
kabk 6 kakkbk .
Example 4.2.3. The complex numbers C with the usual norm | · | form a complex Banach
space, since
|zw| = |z||w|
for all z, w 2 C.
The following lemma allows us to assume that our Banach algebras are unitary:
Lemma 4.2.3. A non-unitary Banach algebra A can be embedded isometrically3 into a unitary
Banach Algebra AI .
3
A map f : X ! Y between two metric spaces (X, dX ) and (Y, dY ) is said to be an isometry if
dY (f (a), f (b)) = dX (a, b) for all a, b 2 X.
102
Proof. Let
AI = A ⇥ C
as a vector space, and define multiplication in AI by
It is easy to verify that AI is a complex algebra with this product. Moreover, the element
(0, 1) 2 AI satisfies
(a, ↵)(0, 1) = (a, ↵) = (0, 1)(a, ↵),
so that AI is unitary. Now define
Therefore AI is a Banach algebra with unit and we may identify A with the ideal
{(a, 0) | a 2 A}
Now let us show that we can renorm A without disturbing its topology to force k1A k = 1.
Lemma 4.2.4. Let A be a unitary Banach algebra. Then we may replace the given norm k·k
on A with another norm k·k0 that yields the identical metric topology, with the further property
that k1A k0 = 1.
La : A ! A
b 7! ab,
and define
kak0 = kLa kop = sup kabk .
kbk=1
kak
6 kak0 6 kak ,
k1A k
k1A k0 = 1.
103
Therefore, by the previous lemmas, throughout this section we may assume that our Banach
algebras are unitary and k1A k = 1. As usual, the group of units of A will be denoted by A⇥ .
Note that if x 2 A satisfies kxk 6 1, then 1 x 2 A⇥ , since the series
1
X
xj
j=0
converges and
1
X
1
(1 x) = xj .
j=0
f : A⇥ ! A⇥
a 7! a 1
is a homeomorphism.
1
Proof. Let a 2 A⇥ and let " ··= a 1
. Let b 2 A with ka bk < ". Then
a 1 (a b) < 1,
b = a(1 a 1 (a b)) 2 A⇥ ,
open
whence A⇥ ✓ A. Clearly f is continuous and coincides with its own inverse. ⌅
Definition 4.2.8 (Spectrum). Let A be a Banach algebra and let a 2 A. The spectrum of a
is defined as
sp(a) = { 2 C | 1A a 2 / A⇥ }.
The spectral radius of a is defined as
r(a) = sup | |.
2sp(a)
We will see briefly that for all a 2 A, sp(a) 6= ?, so that the spectral radius is well defined. For
the moment, we can take the spectral radius to be 0 is sp(a) is empty.
be a complex power series with convergence radius ⇢. Let a 2 A with kak < ⇢. Then
P1
(i) f (a) ··= j=0 ↵j aj converges in A.
104
P
Proof. (i) Since A is complete, it suffices to show that sn ··= nj=0 ↵j aj is Cauchy. Let " > 0
and let N = N (") 2 N such that for all |z| < ⇢ and n, m > N ,
n
X m
X
j
↵j z ↵j z j < ". (4.2.4)
j=0 j=0
P
This is possible, since nj=0 ↵j z j converges in the complete space C. Since kak < ⇢, then
kak 6 ⇢ for some > 0. Now, let m, n 2 N such that m, n > N and without loss of
generality, assume that n = m + k for some integer k > 0. Then
k
X
ksn sm k = ↵m+j am+j
j=1
k
X
6 |↵m+j | am+j
j=1
k
X
6 |↵m+j |(⇢ )m+j
j=1
< ",
converges. Since
1
X
1 j 1 j
( 1A a) = a,
j=0
⇥
then 1A a 2 A , a contradiction. Hence
converges. Therefore,
1
X
f ( )1A f (a) = ↵ j ( j 1A aj )
j=0
X1
= ↵ j ( j 1A aj )
j=1
1 j 1
X X
k j 1 k
= ↵ j ( 1A a) a
j=1 k=0
= b( 1A a),
105
where
1 j 1
X X
k j 1 k
b= ↵j a 2A
j=1 k=0
1
conmutes with a. Suppose for the sake of contradiction that (f ( )1A f (a)) = c for
some c 2 A. Then, ( 1A a) 1 = bc, a contradiction. ⌅
Now we are ready to prove the first major result about the spectrum of an element.
compact
(i) sp(a) ✓ C.
(ii) sp(a) 6= ?.
1
(iii) kan k n ! r(a).
n!1
open
Proof. (i) Let g : C ! A be defined as g( ) = 1A a. Since g is continuous and A⇥ ✓ A
by Theorem 4.2.6, then
open
C \ sp(a) = g 1 (A⇥ ) ✓ C.
Moreover, by Lemma 4.2.5, sp(a) is bounded. Therefore sp(a) is compact.
f ( ) = '(( 1A a) 1 ).
Since ' is linear and continuous, then for " sufficiently close to 0,
1
X
f( ") = "n '(( 1A a) n 1
),
n=0
so that f has a valid power series expansion at every point of its domain, i.e., f is analytic.
Moreover, if | | > kak, then
X1
n 1
f( ) = '(an )
n=0
and therefore
k'kop
f( ) 6 . (4.2.6)
| | kak
Now suppose for the sake of contradiction that sp(a) = ?, so that a 6= 0. Then
106
By (4.2.6), f is bounded in { 2 C | | | > 2kak}.
Therefore, by Liouville’s Theorem4 , f is constant. Moreover, by (4.2.6),
f( ) ! 0,
| |!1
so that f = 0. Since this happens for ' 2 A0 arbitrary, by a Corollary of the Han Banach
Theorem (Corollary 4.2.5.2),
1
( 1A a) = sup |'(( 1A a) 1 )| = 0,
'2A0
k'kop =1
i.e., ( 1A a) 1
= 0, a contradiction. Hence sp(a) 6= ?.
(iii) Recall that for ' 2 A0 , the function f : C \ sp(a) ! C defined by
f ( ) = '(( 1A a) 1 )
is analytic on C \ sp(a), and so has a Laurent series expansion on { 2 C | | | > r(a)}.
But for | | > kak, we know that f has the expansion
1
X
n 1
f( ) = '(an ).
n=0
Therefore, this must be absolutely convergent in the region { 2 C | | | > r(a)}. Now, let
= rei✓ , where r > r(a) and ✓ 2 [0, 2⇡). Then, by the Dominated convergence theorem
of Lebesgue we may interchange summation and integration to obtain
Z 2⇡
rn+1 ei(n+1)✓ f (rei✓ )d✓ = 2⇡'(an ).
0
Therefore
|'(an )| 6 rn+1 M (r)k'kop ,
where
M (r) = sup rei✓ 1A a .
✓2[0,2⇡)
Therefore 1 1 1
lim supkan k n 6 r(a) 6 infkan k n 6 supkan k n .
This completes the proof.
⌅
Corollary 4.2.7.1 (Gelfand-Mazur). Let A be a Banach algebra. If A is a division ring, then
A⇠
= C.
Proof. Given a 2 A, by the previous theorem we know that there is some 2 sp(a) such that
1A a 2/ A⇥ . But since A is a division ring, 1A a = 0, i.e., every element of A takes the
form 1A for some 2 C. Therefore A ⇠ = C. ⌅
4
Liouville’s Theorem states that every bounded entire function must be constant.
107
4.2.2 Quotient Algebras and the Gelfand Transform
We consider A J , where A is a Banach algebra and J is a two sided ideal. Recall that A J
consists of elements of the form a + J, where a 2 A. Now, let
Note that k·k is not a norm, since it’s possible that ka + Jk = 0 but a + J 6= J. Nevertheless,
if J is closed in A, we have:
closed
Proposition 4.2.2. Let A be a Banach algebra and let J ✓ A be a two sided ideal. Then,
k·k as defined by (4.2.7) defines a norm on A J and A J is a Banach algebra with respect to
this norm.
Proof. Let a 2 A and suppose that ka + Jk = 0. Then, there exists a sequence of points on J
closed
that converge to a. But since J ✓ A, then a 2 J, whence a + J = J. Now let us show that
k·k is submultiplicative. Since J is a vector subspace of A, then
ka + Jk = inf ka + xk .
x2J
f : Â ! {Maximal ideals of A}
7! ker
is biyective.
Proof. (i) Let M ✓A be a two sided ideal. Since the norm on A is submultiplicative, then
M is also a two sided ideal of A. Thus, to show that M = M , it suffices to show that
M is proper ideal, i.e., show that M does not contain any units. Suppose for the sake
open
of contradiction that M = A. Since A⇥ ✓ A, then any unit in M must be the limit of
units on M , a contradiction. Hence M = M .
108
(ii) Note that every character 2 Â is surjective (if 2 Â and z 2 C, then (z1A ) = z).
Therefore, by the first isomorphism theorem, A ker is a field, so that ker is a maximal
ideal. Thus, f is well defined. Now, let M = ker . Then, the following diagram
commutes:
A ⇡ AM
C,
open
where ⇡ is the canonical projection and (z1A + M ) = z. Let U ✓ C. Then,
1 open
(U ) = U 1A + M = ⇡(U 1A ) ✓ A.
M
A ⇠
M = C.
Let M ··= M ⇡. From this, it is easy to verify that
ker =
ker M = M.
g : {Maximal ideals of A} ! Â
M 7! M
(iv) Let a 2 A. Then 2 sp(a) if and only if 1A a 2 / A⇥ and by Zorn’s lemma (Every ring
with identity contains a maximal ideal), this happens if and only if 1A a 2 M for some
maximal ideal M . By (ii) of this proposition, this happens if and only if 1A a 2 ker
for some 2 Â, i.e., (a) = for some 2 Â.
⌅
Let A be a commutative Banach algebra. Note that for every a 2 A, we have a map
fa : A0 ! C
' 7! '(a).
The weak ⇤-topology on A0 is defined as the weakest/coarsest topology such that fa is continuous
fo every a 2 A. Under this topology, A0 is locally convex. In particular, A0 is Hausdor↵ with
respect to this topology. By (iii) of Proposition 4.2.3, Â ✓ A0 . The subspace topology on
 induced by the weak ⇤-topology on A0 is called the Gelfand Topology. Before proving the
following lemma, we introduce the concept of net. It will prove to be really useful because it
generalizes the concept of sequence.
109
Definition 4.2.10 (Directed Set). A directed set is a nonempty set I together with a reflexive
and transitive relation 6 with the additional property that for all a, b 2 I there is some c 2 I
such that a 6 c and b 6 c.
Definition 4.2.11 (Nets and Convergence). Let I be a directed set and let X be a topological
space. A net is a function ' : I ! X. For each ↵ 2 I, let x↵ ··= '(↵). The nets are denoted by
{x↵ }↵2I . We say that {x↵ } is eventually on Y ✓ X if there is some 2 I such that x↵ 2 Y for
all ↵ > . We say that {x↵ } converges to x 2 X if {x↵ } is eventually on every neighborhood
of x. If {x↵ } converges to x, then we write
x↵ ! x
or
x = lim x↵ .
Example 4.2.4. Every sequence {xn }n2N is a net.
The following lemmas show why the concept of net is so important; for a proof and a complete
introduction on this subject see [17].
closed
Lemma 4.2.6. Let X be a topological space. Then Y ✓ X if and only if the limit points of
all convergent nets in Y are again in Y .
Lemma 4.2.7. Let X and Y be topological spaces. Then a function f : X ! Y is continuous at
x 2 X if and only if for every convergent net {x↵ } such that x↵ ! x we have {f (x↵ )} ! f (x).
Lemma 4.2.8. Let A be a commutative Banach algebra. Then
(ii) Â is Hausdor↵ and weakly compact with respect to the Gelfand topology.
Proof. (i) For each a 2 A and 2 Â, by Lemma 4.2.5 and (iv) of Proposition 4.2.3 we have
| (a)| 6 r(a) 6 kak . (4.2.8)
Therefore k kop 6 1.
Let A be a commutative Banach algebra. For each a 2 A and 2 Â, let â : Â ! C be the
function defined by
â( ) = (a).
By construction, each function â is continuous with respect to the Gelfand topology. Let C(Â)
denote the algebra of complex-valued functions in  that are continuous with respect to the
Gelfand topology, and endow C(Â) with the sup norm (kf k1 = sup |f (x)|).
5
The Banach–Alaoglu theorem states that the closed unit ball of the dual space of a normed vector space
is compact in the weak ⇤-topology.
110
Definition 4.2.12 (Gelfand Transform). The Gelfand transform is the map
: A ! C(Â)
a 7! â.
The fundamental properties of the Gelfand transform can be described in the following propo-
sition:
Proposition 4.2.4. Let A be a commutative Banach algebra. Then, the Gelfand transform
: A ! C(Â) has the following properties:
(ii) (A) separates point is Â, i.e., for all 1, 2 2 Â with 1 6= 2 there is some f 2 C(Â)
such that f ( 1 ) 6= f ( 2 ).
(ii) If 1 and 2 are two distinct characters in Â, then 1 (a) 6= 2 (a) for some a 2 A. Therefore
â 2 C(Â) separates 1 and 2 .
(iii) This follows inmediately by (iv) of Proposition 4.2.3 and the definition of r(a).
(iv) Let a 2 ker , so that â( ) = (a) = 0 for all 2 Â. Therefore a 2 ker for all 2 Â.
By (ii) of Proposition 4.2.3, a 2 rad(A). Similarly, rad(A) ✓ ker .
C0 (X) ··= {f 2 C(X) | 8" > 0, the set {x 2 X | |f (x)| > "} is compact}.
The set C0 (X) is often referred as the continuous functions that vanish at infinity. To begin
this section we need the following version of the Stone-Weirstrass theorem; for a proof see [29].
6
The radical rad(A) of an algebra A is the intersection of all maximal ideals of A.
7
An algebra A is semisimple if rad(A) is trivial.
111
Theorem 4.2.8 (Stone–Weierstrass Theorem). Let A be a self-adjoint subalgebra of C(X) that
separates points. Then A is uniformly dense in (C(X), k·k1 ).
Corollary 4.2.8.1. Let X be a locally compact Hausdor↵ space and let A be a self-adjoint
subalgebra of C0 (X) that separates points with the additional property that for every x 2 X
there exists f 2 A such that f (x) 6= 0. Then A is uniformly dense in (C0 (X), k·k1 ).
Proof. Let X̃ = X [ {1} denote the Alexandro↵ compactification of X. Then, it is easy to see
that f 2 C0 (X) if and only if f extends to a function f˜ 2 C(X̃) such that f˜(1) = 0. Therefore
we may identify A with a subalgebra of C(X̃) by extending each element to a function that
vanishes at 1. Let à be the subalgebra of C(X̃) generated by A and the complex constant
functions. Clearly à is self-adjoint. Moreover, since A separates points and for every x 2 X
there is a function f 2 A that does not vanish at x, then à separates points. Therefore, by
the Stone–Weierstrass Theorem, Ã is uniformly dense in C(X̃). Hence, for each g 2 C0 (X)
(identified with an element of C(X̃)) and for every " > 0 there is some f 2 A (again identidied
with an element of C(X̃)) and ↵ 2 C such that
"
|g(x) f (x) + ↵| <
2
"
for all x 2 X̃. Since f (1) = g(1) = 0, then |↵| < , so that
2
⇤: A ! A
a 7! a⇤
is called an involution if
A C ⇤ -algebra is a Banach algebra A together with an involution ⇤ : A ! A such that for all
a 2 A,
kaa⇤ k = kak2 .
The following example is of vital importance and the reader should always recall this example
when thinking of C ⇤ -algebras.
Example 4.2.5 (The Hilbert Adjoint Operator). Let H be a Hilbert space and T : H ! H a
bounded linear operator. We shall prove that there exists an operator T ⇤ : H ! H such that
for all x, y 2 H,
hT x, yi = hx, T ⇤ yi. (4.2.9)
The operator T ⇤ defined by (4.2.9) is called the Hilbert adjoint operator. To prove its existence,
let h : H ⇥ H ! C be defined as
h(y, x) = hy, T xi,
112
so that h is a sesquilinear form on H ⇥ H. By the Cauchy-Schwarz inequality,
|h(x, y)| 6 kT kop kxkkyk .
This shows that h is bounded and that khk 6 kT k. Moreover, since
|hy, T xi| |hT x, T xi|
khk = sup > sup = kT kop ,
kxk6=0 kxkkyk kxk6=0 kT xkkxk
kyk6=0 kT xk6=0
then
khk = kT kop .
By the Riesz representation theorem (Theorem 4.2.4) there is some bounded linear operator
T ⇤ : H ! H such that
h(y, x) = hT ⇤ y, xi
and
kT ⇤ kop = khk = kT kop .
This completes the proof of existence, and from the definition of T ⇤ it is easy to see that ⇤ is
an involution. Moreover, by the Cauchy-Schwarz inequality, for all x 2 H,
kT xk2 = hT x, T xi = hT ⇤ T x, xi 6 kT ⇤ T xkkxk 6 kT ⇤ T kop kxk2 .
i.e., r(a) = kak. Taking all of this into account we are ready to prove two lemmas that shall be
used in the proof of the Spectral theorem.
113
Lemma 4.2.9. Let A be a C ⇤ -algebra and let a 2 A. Then
If a is unitary, aa⇤ = 1A , so that r(a) = kak = 1. Since (a⇤ ) 1 = a, then by our previous
observation,
1
2 sp(a) () 2 sp(a).
Therefore, if 2 sp(a) then | | 6 1 and | 1
| 6 1, i.e., | | = 1, so that 2 T.
Therefore exp(ia) is unitary. Now, if 2 sp(a), then again using Proposition 4.2.1,
exp(i ) 2 sp(ia), whence | exp(ia)| = 1. Therefore Re(i ) = 0, i.e., 2 R.
⌅
Lemma 4.2.10. Let B be a C ⇤ -algebra and let A be a self-adjoint, closed, commutative subalge-
bra of B. Then, the Gelfand transform : A ! C(Â) is an isometric isomorphism. Moreover,
it is a ⇤-isomorphism in the sense that (a⇤ ) = (a).
for all a 2 A. This shows that is an isometry onto its image. Note that if b 2 A is self-adjoint,
then
b̂( ) = (b) 2 sp(a) ✓ R,
i.e., (a) is a real-valued function. Now, for a 2 A arbitrary, write
a = c + id,
where
a + a⇤ a a⇤
c= and d =
2 2i
114
are self-adjoint. Since (c) and (d) are real valued functions, then
(a⇤ ) = (c id)
= (c) i (d)
= (c) + i (d)
= (c + id)
= (a).
It reamins to show that is surjective. For this, we recall the following properties of :
(i) By Proposition 4.2.4, (A) separates points and since A is unitary, then (A) contains
the constant functions.
(ii) Since A is self-adjoint and (a⇤ ) = (a), then (A) is a self-adjoint subalgebra of C(Â).
closed
(iii) Since A is isometric isomorphic to (A), then (A) ✓ C(Â).
By (i) and (ii), using the Stone-Weirstrass theorem (Theorem 4.2.8), (A) is dense on C(Â).
Combining these facts with (iii) we conclude that (A) = C(Â). ⌅
Before diving into the Spectral theorem, we need one last preparation. Let A be a C ⇤ -algebra
and let a 2 A be normal. We will denote by C ⇤ (a) the smallest self-adjoint, closed subalgebra
of A containing a, i.e.,
C ⇤ (a) = Ch1A , a, a⇤ i.
Since a is normal, then C ⇤ (a) is commutative. Let
Clearly sp(a) ✓ spA (a). Finally, if W is a nonempty subset of C, let ◆W denote the inclusion of
W in C. Taking all of this into account we are ready to state and prove the spectral theorem:
Theorem 4.2.9 (The Spectral Theorem). Let A be a C ⇤ -algebra and let a 2 A be normal.
Then, there exists an isometric ⇤-isomorphism : C(sp(a)) ! C ⇤ (a) such that (◆sp(a) ) = a.
Proof. Consider the Gelfand transform defined on the space of characters of C ⇤ (a):
\
â : C ⇤ (a) ! C
7! (a).
\
â : C ⇤ (a) ! sp (a).
A
8
The open mapping theorem states that if X and Y ara Banach spaces and T : X ! Y is a surjective
bounded linear operator, then T is an open map.
115
Now consider the map
\
⇥ : C(spA (a)) ! C(C ⇤ (a))
f 7! f â.
Then, by the properties of the Gelfand transform, ⇥ is a isometric ⇤-isomorphism. Now, let
··= 1
⇥, so that the following diagram commutes:
\
C(C ⇤ (a)),
From this, it is easy to see that is a isometric ⇤-isomorphism. Let f 2 C(spA (a)). By
definition,
⇥(f )( ) = f ( (a)),
for all \
2C ⇤ (a), but since \
is an isomorphism, every map in C(C ⇤ (a)) is of the form
7! (b)
f ( (a)) = ( (f ))
\
for all 2 C ⇤ (a). From this it is clear that (◆spA (a) ) = a. Thus to complete the proof it suffices
to show that spA (a) ✓ sp(a). For this, let 2 spA (a) and consider a function f 2 C(spA (a))
such that f has a maximum absolute value 1, f ( ) = 1 and f (µ) = 0 whenever | µ| > ".
Let b = (f ). Then, since is an isometry and f is zero away from , we have
1
(a 1A )b = ((a 1A )b) 1
= (◆spA (a) )f
< ".
Hence, if a 1A 2 A⇥ , then
1 = kf k1
= kbk
= (a 1A ) 1 (a 1A )b
1
< (a 1A ) " ! 0;
"!0
116
Corollary 4.2.9.2. Let T be a normal operator in a Hilbert space H. Then, the following
assertions are equivalent
(ii) C ⇤ (T ) = C.
Proof. (i))(ii) by the spectral theorem. (ii))(iii) and (iii))(i) follows by definition of C ⇤ (T )
and sp(T ). ⌅
Definition 4.2.15 (Representation). Let V be a topological complex vector space and let G
be a locally compact group. An abstract representation (⇢, V ) is a homomorphism
⇢ : G ! GL(V ).
G⇥V !V
(g, v) 7! ⇢g (v) ··= ⇢(g)v
117
Definition 4.2.18 (Irreducible). A representation (⇢, V ) is said to be topologically irreducible
(or simply an irreducible representation) if the only closed G-invariant subspaces of V are ?
and V .
With all of these definitions we are ready to prove the Shur’s lemma for unitary representations:
Lemma 4.2.11 (Shur’s Lemma). Let (⇢, H) be a representation of the locally compact group
G in a Hilbert space H and let T 2 B(H) be a normal operator such that
⇢(g) T = T ⇢(g)
Proof. Suppose that 2 sp(T ). Then, there exists a function f : sp(T ) ! C such that f 6= 0
and f (x) = 0 in some open neighborhood of . Let : C(sp(T )) ! C ⇤ (T ) be the isometry of
the spectral theorem. Since for all g 2 G, T commutes with ⇢(g) and ⇢ is unitary, then T ⇤ also
commutes with ⇢(g) for all g 2 G. Therefore, W ··= (f )H is invariant under ⇢(G). Now, since
⇢ is irreducible and f 6= 0, then W = H. Now suppose for the sake of contradiction that sp(T )
is not a singleton. Then, we can find a continuous function h : sp(T ) ! C with complementary
support with respect to f , but
{0} = (h) (f )H,
so that W 6= H; a contradiction. Hence sp(T ) is a singleton and by Corollary 4.2.9.2, T is a
scalar multiple of the identity. ⌅
Corollary 4.2.11.1. Let G be a locally compact abelian group and let (⇢, H) be an irreducible
unitary representation of G on a Hilbert space H. Then dimC (H) = 1.
Proof. For every g 2 G, since G is abelian, then ⇢(g) is G-linear. Therefore, by Shur’s lemma,
⇢(g) acts by scalar. Therefore, every nonzero x 2 H generates the G-invariant subspace Cx
and since (⇢, H) is irreducible, Cx = H, so that dimC (H) = 1. ⌅
Definition 4.3.1. Let G be a LCA group (a locally compact abelian group). The Pontryagin
dual of G is the set defined by
Ĝ = { : G ! T | is a continuous homomorphism}.
118
compact
Now, let K ✓ G and let V be an open neighborhood of 1 2 T. Let
The sets W (K, V ) constitute a neighborhood base for the trivial character, so that {W (K, V )}
and their translates form a base for the topology on Ĝ. This topology is called the compact-open
topology. Now, consider the map
': R ! T
x 7! e2⇡ix .
Then, ' is a continuous homomorphism with ker ' = Z. Let 0 < " 6 1 and define N (") ✓ T
by ⇣⇣ " " ⌘⌘
N (") ··= ' , .
3 3
Lemma 4.3.1. Let m 2 N and suppose that x 2 C satisfies x, x2 , . . . , xm 2 N (1). Then
x lies in N (1/m). Therefore, if U is an open neighborhood of e 2 G and : G ! T is a
homomorphism such that (U (m) ) ✓ N (1), then (U ) ✓ N (1/m).
Proof. Let r 2 N and suppose that xr 2 N (1). Then there is some y 2 N (1/r) such that
x
xr = y r , and thus is a r-th root of unit. Therefore x 2 N (1/r)'(q/r) for some integer
y
0 6 q < r. Now, note that
⇣ 1 ⌘ n 2⇡it ⇣ 1 1 ⌘o
N = e 3 |t2 ,
r r r
and ⇣ 1 ⌘ ⇣ q ⌘ n 2⇡it ⇣ 3q 1 3q + 1 ⌘o
N ' = e 3 |t2 , .
r+1 r+1 r+1 r+1
Therefore, if ⇣1⌘ ⇣ 1 ⌘ ⇣ q ⌘
N \N ' = ?,
r r+1 r+1
then
1 3q 1
> ,
r r
so that 2r + 1 > 3qr, whence q = 0. Now suppose that x 2 N (1/r) and xr+1 2 N (1). Then,
x 2 N (1/(r + 1))'(q/(r + 1)) and by our previous observation, x 2 N (1/(r + 1)). Therefore,
it follows by induction that if x, x2 , . . . , xm 2 N (1), then x 2 N (1/m). This completes the
proof. ⌅
With this lemma we are ready to prove some important properties about the Pontryagin dual:
(ii) The family {W (K, N (1))} indexed over all compact subsets of G is a neighborhood base
of the trivial character for the compact-open topology of Ĝ.
119
(iv) If G is compact, then Ĝ is discrete.
(iii) If G is discrete, then the compact-open topology coincides with the topology of the
pointwise convergence9 . Under this topology,
closed
Ĝ = Hom(G, T) ✓ TG .
(iv) Given 2 Ĝ, (G) is a subgroup of T. Since any subgroup of T can’t be contained in
any set of the form N ("), 0 < " 6 1, then W (G, N (1)) must contain only the trivial
character, which therefore is an open subset of Ĝ and since Ĝ is a topological group, then
Ĝ is discrete.
9
The topology of the pointwise convergence can be generated by taking as subbase the collection of sets of
the form S(x, U ) = {f : G ! T | f (x) 2 U }, where x 2 G and U is open on T.
120
Let us show that W ( ) is an open neighborhood of with respect to ⌧0 , so that ⌧ has a
neighborhood base in contained in ⌧0 . Let V be an open neighborhood of e 2 G such
that V (2m) ✓ K. Since K1 is compact, there exists a finite set F such that K1 ✓ F V . Let
⇣ ⇣ 1 ⌘⌘
·
W0 ( ) ·= W0 F, N \ W.
2m
To complete the proof we have to show that W0 ( ) is a ⌧0 -neighborhood of contained
open
in W ( ). Since W0 (F, N (1/(2m)) ✓ Ĝ0 , we only need to verify that W0 ( ) ✓ W ( ).
Let µ 2 W0 ( ). Then, µ = µ0 2 W for some µ0 2 Ĝ0 such that µ0 (F ) ✓ N (1/(2m)).
Since µ0 = 1 µ 2 W (2) , then
⇣1⌘
µ0 (K) ✓ N ✓ N (1).
2
From this we draw two conclusions:
Thus, we have
⇣ 1 ⌘ ⇣ 1 ⌘ ⇣1⌘
µ0 (K1 ) ✓ µ0 (F )µ0 (V ) ✓ N N =N .
2m 2m m
This shows that µ0 2 W (K1 , N (1/m)), and therefore µ 2 W ( ). ⌅
A(i,j) = '(sj 1 si ).
h⇢(sj 1 si )(x), xi = h⇢(si )(x), ⇢(sj )(x)i = h⇢(sj )(x), ⇢(si )(x)i = h⇢(si 1 sj )(x), xi
hA~z , ~z i > 0.
This shows that A is positive definite Hermitian matrix. This observation is the motivation for
the following definition:
121
compact
If supp(f ) ✓ K ✓ G, then
ZZ
'(s 1 t)f (s) dsf (t) dt 6 k'k1 (µ(K) sup |f |)2 < 1,
where µ is the Haar measure. This shows that the integral in (4.3.1) is bounded. Let ' be a
positive definite function. Then, we can define a sesquilinear form on K (G)⇥K (G) as follows:
ZZ
hf1 , f2 i' ··= '(s 1 t)f1 (s) dsf2 (t) dt.
Ls f : G ! C
t 7! f (s 1 t).
In the particular case f 2 K (G), it is easy to see that Ls f 2 K (G). Thus, the map
⇢ : G ! B(K (G))
s 7! Ls
This shows that L induces an abstract unitary representation of G on V' . To show that ⇢ is a
topological representation we have to show that for all f 2 K (G), the map
G ! K (G)
s 7! Ls f
is continuous. For this, by Lemma 4.2.7 it suffices to show that if {s↵ } is a convergent net in
G with limit s, then
Ls f = lim Ls↵
in K (G) with respect to the '-norm. But this happens if and only if the pointwise convergence
f (s↵ 1 t) ! f (s 1 t)
is uniform for all t 2 G, and this is true by Lemma 4.1.5. Therefore, we conclude the following
theorem:
Theorem 4.3.2. Let G be a locally compact group and ' a positive definite function on G.
Then, the map s 7! Ls induces a unitary representation of G on the Hilbert space V' .
122
Definition 4.3.3 (Convolution). Let f, g : G ! C be Borel functions on a locally compact
group G. We define their convolution by
Z Z
(f ⇤ g)(t) ··= g(s t)f (s) ds = g(s 1 )f (ts) ds.
1
For now, if f 2 K (G) and ' 2 L1 (G), then clearly f ⇤ ' exists. Moreover, if {t↵ } is a
convergent net on G such that t↵ ! t, then
Z
'(s 1 )(f (ts) f (t↵ s)) ds ! 0.
Therefore, by Lemma 4.2.7, f ⇤ ' is continuous. Before proving a theorem that resembles the
Riesz representation theorem for linear functionals, we need the following version of Uryshon’s
lemma; for a proof see [30].
open
Lemma 4.3.2 (Uryshon’s Lemma). Suppose X is a locally compact Hausdor↵ space, V ✓ X
and K ✓ V , where K is compact in X. Then, there exists a function f 2 K (X) such that
(i) 0 6 f 6 1,
(ii) f (x) = 1 for all x 2 K,
(iii) supp(f ) ✓ V .
Theorem 4.3.3. Let ' be a positive definite function on a locally compact group G. Then,
there exists some x' 2 V' such that
Proof. Let {↵} be a set of indices for the collection of open neighborhoods V↵ of e 2 G. Since
G is Hausdor↵, then \
{e} = V↵
and if we define the relation ↵ 6 by V ✓ V↵ , then {↵} is a directed set. By Uryshon’s
lemma, for each ↵ we may construct a function g↵ 2 K (G) such that
(i) supp(g↵ ) ✓ V↵ .
Z
(ii) g↵ (s) ds = 1.
This defines a net {g↵ (s) ds} of positive linear functionals on K (G):
Z
f 7! f (s)g↵ (s) ds.
Note that g↵ (s) ds converges weakly10 to the dirac measure . Let f 2 K (G) and consider the
following integral: ZZ Z
1
'(s t)f (s) ds g↵ (t) dt = (f ⇤ ')(t)g↵ (t) dt,
10
Let X beR a topological
R space. A net {µ↵ } of measures on (X, B) converges weakly to some measure µ on
(X, B) if lim f dµ↵ = f dµ for all f 2 C(X).
123
which exists by the continuity of f ⇤ ' and the fact that g↵ has compact support. Now, we can
define on V' as follows: For each f 2 K (G), let
ZZ Z
1
(f ) = limhf, g↵ i' = lim '(s t)f (s)g↵ (t) dt = lim (f ⇤ ')(t)g↵ (t) dt.
it suffices to note that in the integral we may replace f ⇤ ' by (f ⇤ ') · h, where h 2 K (G)
is fixed and has value 1 in a neighborhood of e that contains the eventual support of the g↵ ’s.
Thus (f ⇤ ') · h 2 K (G) and since g↵ (s) ds converges weakly to , then the limit exists and
(4.3.2) holds. Now, by the Riesz representation theorem (Theorem 4.2.3), there is some x' 2 V'
such that
(⇠) = h⇠, x' i'
for all ⇠ 2 V' , i.e., g↵ converges weakly11 to x' . Note that
Similarly, Z
hLs x' , ⇠i' = '(s 1 u)⇠(u) du.
In particular, for s = e, Z
h⇠, x' i' = '(t)⇠(t) dt. (4.3.4)
and by continuity this equality extends for all h 2 V' . In the special case ⇠ = x' , the previous
equation together with (4.3.4) show that for all 2 V' ,
Z Z
'(s) (s) ds = hx' , i' = hx' , Ls x' i' (s) ds,
whence
'(s) = hx' , Ls x' i
almost everywhere for s 2 G. ⌅
11
Let H be a Hilbert space. A net {x↵ } on H converges weakly to some x 2 H if limhx↵ , yi = hx, yi for all
y 2 H.
124
Corollary 4.3.3.1. Let ' be a positive definite function on a locally compact group G. Then,
' is equal to a positive definite function almost everywhere. If moreover ' is continuous, then
Proof. The main result follows at once from the continuity of h·, ·i' .
'(s)2 = hx' , Ls x' i 6 hx' , x' i' hLs x' , Ls x' i' = hx' , x' i2' = '(e)2 .
(iii) Finally,
'(s 1 ) = hx' , Ls 1 x' i' = hLs x' , x' i' = hx' , Ls x' i = '(s). ⌅
Let
P (G) ··= {' 2 C(G) \ L1 (G) | ' is positive definite and k'k1 6 1}.
Note that by the previous Corollary, the condition k'k1 6 1 is equivalent to '(e) 6 1.
Definition 4.3.4 (Elementary Functions). We define the set E(G) ✓ L1 (G) as follows: A
function ' 2 E(G) if ' = 0 or if ' satisfies the following conditions:
(ii) '(e) = 1.
(iii) If ' = '1 + '2 , where '1 , '2 2 P (G), then there exists some 1, 2 > 0 such that
Note that condition (iii) says that the elements of E(G) are extreme points12 of P (G).
Lemma 4.3.3. The sets P (G) and E(G) have the following properties:
(i) P (G) is a convex and bounded subset of L1 (G). Moreover, P (G) is weakly closed and
therefore weakly compact as a subset of L1 (G)0
(ii) Any closed and convex subset of P (G) that contains its extreme points is all of P (G).
(iii) The extreme points of P (G) are precisely the points of E(G).
12
Let S be a convex set in a real vector space. We say that x 2 S is a extreme point if x = 1y + 2 z, where
y, z 2 S and 1 + 2 = 1 implies x = y or x = z.
125
Proof. (i) Clearly P (G) is bounded and convex. Recall that L1 (G)0 identifies with L1 (G)
via the isometric isomorphism
: L1 (G) ! L1 (G)0
Z
f 7! (f )(g) ··= f g ds.
w
Now, let '↵ ! ' in L1 (G), where '↵ 2 P (G), i.e., for all f 2 L1 (G), we have
Z Z
f ' ds = lim f '↵ ds
so that ' is positive definite and by Corollary 4.3.3.1 ' is continuous. Therefore, as a
subset of L1 (X)0 , P (G) corresponds to a closed subset of the unit ball, and is compact
under the weak ⇤-topology by the Banach–Alaoglu theorem.
(iii) It suffices to show that if ' 2 P (G) is a nonzero extreme point, then '(e) = 1. If '(e) < 1,
then
1
' 2 P (G)
'(e)
and ' would not be an extreme point. ⌅
With this lemma we are ready to prove two important theorems that make a connection between
the elementary functions, the unitary representation associated to them, and the characters of
a LCA group.
Theorem 4.3.4. Let ' a positive definite continuous function on G such that '(e) = 1. Then
' 2 E(G) if and only if the unitary representation s 7! Ls in V' is irreducible.
Proof. ()) Suppose that ' is an elementary function. Let W be a closed G-invariant subspace
of V' . Since each operator Ls is unitary, then the following diagram commutes:
⇡W
W W? W
Ls Ls
⇡W
W W? W
Therefore it suffices to show that if P is any orthogonal projection that commutes with each
Ls , then P = 0 or P = idV' . Since hP x, yi' = hP x, P yi' for any projection, then for all s 2 G,
126
This expresses ' as the sum of two positive definite functions. Thus, since ' is extreme,
hP x' , Ls x' i' = hx' , Ls x' i'
for all s 2 G. By (4.3.5), V' is generated as a CG-module by x' , so that P = idV' . But since
P is idempotent (P 2 = P ), then = 1 or = 0.
(() Suppose that the representation s 7! Ls in V' is irreducible and suppose that ' = '1 + '2 ,
with '1 , '2 2 P (G). Then for all f 2 K (G),
hf, f i'1 6 hf, f i' . (4.3.6)
From this it follows that any element of K (G) that is degenerate with respect to h·, ·i' is also
degenerate with respect to h·, ·i'1 . Therefore h·, ·i'1 defines a Hermitian form on V' and by the
Riesz representation theorem (Theorem 4.2.4) there is some positive definite A 2 B(V' ) such
that
hA⇠, i' = h⇠, i'1
for all ⇠, 2 V' . In particular,
hAx' , Ls x' i' = hx' , Ls x' i'1
for all s 2 G. But note that in the proof of Theorem 4.3.3, the convergence of the net {g↵ }
with respect to the h·, ·i' -norm will also hold with respect to the h·, ·i'1 -norm by (4.3.6), so
that
'1 (s) = hx' , Ls x' i'1 ,
and therefore
'1 (s) = hAx' , Ls x' i' .
Now, let us show that A commutes with each Ls :
hALs f, gi' = hLs f, gi'1
= hf, Ls 1 gi'1
= hAf, Ls 1 gi'
= hLs Af, gi'
for all s 2 G, f, g 2 K (G). Hence A commutes with each Ls . Similarly, it is easy to show that
A is normal and therefore by Shur’s lemma (Lemma 4.2.11),
A= idV'
for some > 0, whence
'1 (s) = h x' , Ls x' i' = '(s). ⌅
Theorem 4.3.5. Let G be a LCA group. Then, the elementary functions on G are precisely
the characters of G.
127
for all f 2 K (G). Since (e) = 1, then by the previous theorem, it suffices to show that
given a positive definite function ' 2 K (G) such that '(e) = 1, the following assertions are
equivalent:
(ii) =) (i) Suppose first that ' 2 Ĝ. Then, given g 2 K (G),
Z 2
hf, f i' = '(s)f (s) ds .
Ls (⇠) = (s)⇠
Definition 4.3.5. Let G be a LCA group and let f 2 L1 (G). The Fourier transform of f is
the function fˆ: Ĝ ! C defined by
Z
fˆ( ) = f (y) (y) dy,
where as usual the integral is taken over the group G and dy is the Haar measure on G.
Now, let
V 1 (G) ··= V (G) \ L1 (G).
128
The objective of this section is to prove a generalization of Theorem 1.5.4; namely that the
inversion formula holds for functions f 2 V 1 (G), i.e., we will show that there is some Haar
measure d on Ĝ such that for all f 2 V 1 (G), we have
Z
f (y) = fˆ( ) (y)d .
Ĝ
In order to do so, we need some preparations and we begin with the following technical lemma
about convolution:
Lemma 4.3.4. Let G be a LCA group and let f, g : G ! C be Borel functions. Then
(i) If (f ⇤ g)(x) exists for some x 2 G, then (g ⇤ f )(x) exists and (f ⇤ g)(x) = (g ⇤ f )(x).
(ii) If f, g 2 L1 (G), then (f ⇤ g)(x) exists a.e. for x 2 G. Moreover, f ⇤ g 2 L1 (G) and
kf ⇤ gk1 6 kf k1 kgk1 .
1
(ii) Let : G ⇥ G ! G ⇥ G be defined as (x, y) = (yx, y). Since (x, y) = (y 1 x, y), then
open
is a homeomorphism. Now, let U ✓ C. Then, (f 1 (U ) ⇥ G) is a Borel set and, by
construction,
(x, y) 2 (f 1 (U ) ⇥ G) () y 1 x 2 f 1 (U ).
This shows that the map
G⇥G!C
(x, y) 7! f (y 1 x)
G⇥G!C
(x, y) 7! f (y 1 x)g(y)
From this, it is clear that |f ⇤ g| 2 L1 (G), so that f ⇤ g 2 L1 (G). This shows that f ⇤ g
exists almost everywhere. Finally, it is clear that
kf ⇤ gk1 6 kf k1 kgk1 .
129
(iii) This is a customary application of Fubini’s theorem and the details are left to the reader.
⌅
Corollary 4.3.5.1. If G is a LCA group, then (L1 (G), ⇤) in an associative and commutative
Banach algebra.
With this observation we are ready to show a connection between the Fourier transform and
the Gelfand transform:
Theorem 4.3.6. Let G be a LCA group. For each 2 Ĝ and f 2 L1 (G), let
Z
⌫ˆ (f ) ··= fˆ( ) = f (y) (y) dy.
: Ĝ ! L \ 1 (G)
7! ⌫ˆ
Proof. Since each character of G has norm 1, then ⌫ˆ is non zero and linear. Moreover, a
customary application of Fubini’s theorem shows that
⌫ˆ (f ⇤ g) = fˆ( )ĝ( ).
whence
(f )'(f ) = (Ly f ) (4.3.7)
almost everywhere for y 2 G. Since K (G) is dense on L1 (G)14 , then by Lemma 4.1.5, the map
y 7! Ly f is continuous and since 6= 0, then ' is continuous. Now, applying (4.3.7) three
times, we get
(f )'(xy) = (f )'(x)'(y),
so that ' is multiplicative. In particular,
'(y 1 ) = '(y) 1 ,
whence |'(y)| = 1 for all y 2 G. This shows that ' 2 Ĝ and that (y) = .
0
is injective: If ⌫ˆ (f ) = ⌫ˆ 0 (f ) for all f 2 L1 (G), then (y) = (y) almost everywhere for
y 2 G, but since and 0 are continuous, then = 0 . ⌅
14
For a proof of this fact, see [13]
130
Note that the previous Theorem implies that
fˆ(ˆ
⌫ ) = fˆ( ),
where the left-hand of the equality represents the Gelfand transform operating on the space of
characters of L1 (G), and the right-hand side of the equality represents the Fourier transform
operating on Ĝ.
Lemma 4.3.5. Let G be a LCA group. Then, for each 2 Ĝ, there is some f 2 L1 (G) such
that fˆ( ) 6= 0.
Proof. Let V be a neighborhood of the identity e 2 G such that 0 < µ(V ) < 1, where µ is
the Haar measure on G. Since G is locally compact, there is some K ✓ V compact. Now, by
Uryshon’s Lemma there is some g 2 K (G) ✓ L1 (G) such that
(i) 0 6 g 6 1,
(iii) supp(g) ✓ V .
Then Â(G) defines a topology on Ĝ: the weakest topology such that each fˆ 2 Â(G) is contin-
uous. This topology is called the transform topology on Ĝ.
Lemma 4.3.6. Let G be a LCA group and endow Ĝ with the transform topology induced by
Â(G). Then, the ring Â(G) is a separating, dense and self-adjoint subalgebra of C0 (Ĝ).
Proof. By Gelfand theory (Lemma 4.2.8) we know that if L1 (G) is unitary, then L \1 (G) is a
weakly compact subset of L1 (G)0 and given f 2 L1 (G), its Gelfand transform fˆ is in C(L \1 (G)).
\
In other case, L 1 (G) may not be closed, since the weak limit of non-trivial homomorphism may
in fact be trivial. Nevertheless, B = L1 (G) [ {0} is closed. Therefore, for each f 2 L1 (G),
fˆ 2 C(B) and thus fˆ 2 C0 (L\1 (G)).
Now, we identify C(Ĝ) with C(L \ 1 (G)) with the topological isomorphism induced by 7! ⌫ˆ .
Then, by Proposition 4.2.4, Â(G) ✓ C0 (Ĝ) and separates points. The density of  follows as
a consequence of the Stone–Weirstrass theorem (Corollary 4.2.8.1 and Lemma 4.3.5). Now, let
us show that Â(G) is self-adjoint: Let f 2 L1 (G). Then, for all 2 Ĝ,
Z Z Z
f (y 1) (y) dy = f (y) (y 1 ) dy = f (y) (y) dy = fˆ( ),
131
Lemma 4.3.7. Let G be a LCA group and endow G ⇥ Ĝ with the product topology induced by
the topology of G and the transform topology on Ĝ.
G ⇥ Ĝ ! C
d
(y, ) 7! L yf ( )
is continuous.
G ⇥ Ĝ ! C
(y, ) 7! (y)
is continuous.
Proof. (i) Let (y0 , 0) 2 G ⇥ Ĝ. Since K (G) is dense on L1 (G), then by Lemma 4.1.5, the
map
L1 (G) ! L1 (G)
f 7! Ly f
is uniformly continuous for all y 2 G. Hence, for all " > 0 there is some neighborhood
U ✓ G of y0 2 G such that
"
Ly0 f L y f 1 <
2
for all f 2 L (G). Now, since |ĝ( )| 6 kgk1 , then in particular,
1
d
|L yf ( )
[
L y0 f ( )| 6 Ly f Ly0 f .
1
So,
d
|L yf ( )
[
L y0 f ( 0 )|
d
6 |L yf ( )
[
L [
y0 f ( )| + |Ly0 f ( )
[
L y0 f ( 0 )| < ".
d
fˆ( ) (y) = L y f ( ).
d
L yf ( )
(y) =
fˆ( )
With this preliminary results, we are ready to prove that the transform topology coincides with
the compact-open topology:
compact
Theorem 4.3.7. Let G and Ĝ be as in the previous lemma. Let K ✓ G and let V ✓ T be
an open neighborhood of 1 2 T.
132
open
(i) W (K, V ) ✓ Ĝ.
(ii) The collection {W (K, V )} is a neighborhood base for the trivial character.
(iii) The compact-open topology coincides with the transform topology on Ĝ.
Proof. Since (ii) =) (iii), it suffices to show (i) and (ii). Moreover, note that it suffices to
consider sets of the form W (K, N (")).
compact
(i) Let K ✓ G and let " > 0. Let 0 2 W (K, N (")). Then, by the previous lemma, for
all y0 2 K there exists an open neighborhood U ✓ G of y0 and an open neighborhood
V ✓ Ĝ of 0 such that (y) 2 N (") for all 2 V and y 2 U . Since K is compact, then
K is covered by finitely many U1 , . . . , Ur . Let V1 , . . . , Vr be the corresponding characters
sets, i.e., Vj = W (K, Uj ). Note that \
Vj
open
is an open neighborhood of 0 contained in W (K, N (")), i.e., W (K, V ) ✓ Ĝ.
open
(ii) Let V1 ✓ Ĝ. By definition of the transform topology, for some "1 > 0 there exists
f1 , . . . fr such that \
{ | |fˆj ( ) fˆj (1)| < "1 } ✓ V.
Since K (G) is dense on L1 (G), we may assume (at the cost of decreasing "1 ) that each
fj has compact support Kj . Let \
K ··= Kj
and let " > 0 such that
3"1 .
"<
2⇡ max fj 1
6 max fj 1
sup | e 2⇡it
1|
|t|< 3"
⇡"
6 max fj 1
· 2 sin
3
2⇡"
6 max fj 1
·
3
< "1 ,
We continue working with G and Ĝ as in the previous theorem. Let µ̂ be a complex Radon
measure15 of total finite mass, i.e, |µ̂|(Ĝ) < 1. We define the Fourier transform of µ̂ as the
function Tµ̂ : G ! C defined by Z
Tµ̂ (y) = (y)dµ̂( ).
15
Complex Radon measure means that µ̂(G) 2 C, not necessarily in [0, 1]. By the Jordan decomposition
P3
theorem, µ̂ = k=0 ik µ̂k , where each µ̂k is a positive and finite Radon measure.
133
Since µ̂ has total finite mass, then Tµ̂ is continuous and bounded by µ̂(Ĝ). By Lemma 4.3.7,
we may use Fubini’s theorem to verify that for all f 2 L1 (G),
Z Z
fˆ( )dµ̂( ) = f (y)Tµ̂ (y) dy. (4.3.8)
Lemma 4.3.8. If Tµ̂ (y) = 0 for all y 2 G, then µ̂ = 0. Therefore µ̂ is completely determined
by its Fourier transform.
for all f 2 L1 (G). Since Â(G) is dense on C0 (Ĝ) (Lemma 4.3.6), then in particular
Z
g( )dµ̂( ) = 0
Before proving a theorem that connects the positive definite functions and the Fourier transform
of Radon Measures, we need yet another version of the Riesz–Markov representation theorem;
for a proof see [30].
Theorem 4.3.8. Let X be a locally compact Hausdor↵ space. Then, for any 2 C0 (X)0 there
is a unique complex Radon measure µ on X such that
Z
(f ) = f dµ. (4.3.9)
With this theorem we are ready to prove the following major result:
Theorem 4.3.9 (Bochner’s Theorem). The set P (G) is precisely the set of Fourier transforms
of positive Radon measures on Ĝ with total mass 6 1.
Proof. Let
M̂ ··= {Tµ̂ | µ̂ is a positive Radon measure and µ̂(G) 6 1}.
We already mentioned that Tµ̂ is a continuous function, with Tµ̂ 1 = µ̂(G) 6 1. Moreover,
an application of Fubini’s Theorem shows that for all f 2 K (G) we have
ZZ Z
Tµ̂ (s t)f (s)f (t) ds dt = hf, f i dµ̂( ) > 0.
1
Hence Tµ̂ is a positive definite function. This shows that M̂ ✓ P (G). Now, if µ̂ is the Dirac
measure ⌘ supported at a character ⌘, then
T ⌘ (s) = ⌘(s),
so that M̂ contains the characters and by Theorem 4.3.5 and (iii) of Lemma 4.3.3 this means
that M̂ contains the extreme points of E(G). Since the set of positive Radon measures with
total mass 6 1 is a convex set and T is linear in the measure, then M̂ is a convex subset of
134
P (G). Therefore, by (ii) of Lemma 4.3.3 to complete the proof it suffices to show that M̂ is
weakly closed. In this regard, let {Tµ̂↵ }↵ be a weakly convergent net with limit ' 2 L1 (G),
i.e., Z Z
lim Tµ̂↵ (s)f (s) ds = '(s)f (s) ds
for all f 2 L1 (G). Since Â(G) is dense in C0 (Ĝ) (Lemma 4.3.6), then
Z
lim h( )dµ̂↵ ( )
exists for every h 2 C0 (Ĝ). Therefore, by Theorem 4.3.8 there exists a unique complex Radon
measure µ̂ such that Z Z
lim h( )dµ̂↵ ( ) = h( )dµ̂( ) (4.3.10)
for every h 2 C0 (Ĝ). Since each µ̂↵ is a positive measure of total mass 6 1, then µ̂ is also a
positive measure of total mass 6 1. Therefore, Tµ̂ is well-defined, and when h is the Fourier
transform of any f 2 L1 (G), by (4.3.8), the equality (4.3.10) becomes
Z Z
lim Tµ̂↵ (s)f (s) ds = Tµ̂ (s)f (s) ds.
This shows that Tµ̂↵ converges weakly to Tµ̂ , so that M is weakly closed. This completes the
proof. ⌅
Proof. Since the measures are completely determined by their Fourier transforms, it suffices to
show the equality for their corresponding Fourier transforms:
Z
Tĝ( )dµ̂f ( ) (y) = (y)ĝ( )dµ̂f ( )
Z Z
= (y) g(z) (z) dzdµ̂f ( )
ZZ
= (z 1 y)dµ̂f ( )g(z) dz
Z
= f (z 1 y)g(z) dz
= (f ⇤ g)(y)
= (g ⇤ f )(y)
= Tfˆ( )dµ̂g ( ) (y). ⌅
135
Lemma 4.3.10. There exists a net of functions in V 1 (G) whose Fourier transform converges
to the constant function 1 on compact subsets of Ĝ.
Proof. Let {↵} be a set of indices for the collection of compact neighborhoods K↵ of e 2 G.
Since G is Hausdor↵, then \
{e} = K↵
and if we define the relation ↵ 6 by K ✓ K↵ , then {↵} is a directed set. By Uryshon’s
Lemma, for each ↵ we may construct a function g↵ 2 K (G) such that
(i) supp(g↵ ) ✓ K↵ .
Z
(ii) g↵ (s) ds = 1.
then f↵ is a positive-definite function, so that each f↵ 2 V 1 (G). Moreover, since the net {gˆ↵ }
converges to the constant function 1, then it is easy to verify that the net {fˆ↵ } also converges
to the constant function 1. ⌅
Definition 4.3.7. Let CB (Ĝ) denote the set of complex-valued, bounded, and continuous
functions ' in Ĝ and let F be the set of function ' 2 CB (Ĝ) with the additional property that
there exists some complex Radon measure ⌫ˆ' in Ĝ of finite total mass such that
By Lemma 4.3.9, the Fourier transform of elements of V 1 (G) is in F : If ' = ĝ for some
ĝ 2 V 1 (G), then ⌫ˆ' is just µ̂g .
Lemma 4.3.11. The set F has the following properties:
(iv) The set F is a module over CB (Ĝ) and the map ' 7! ⌫ˆ' is a morphism of modules over
CB (Ĝ). In particular,
⌫ˆ'+ = ⌫ˆ' + ⌫ˆ and ⌫ˆa' = aˆ ⌫'
for all a 2 CB (Ĝ) and ', 2 F.
136
(v) If ' 2 F , then every translation ↵ of ' also lies in F and to obtain ⌫ˆ↵ from ⌫ˆ' one
applies the same translation. This means that if 0 2 Ĝ is fixed and
L 0 : CB (Ĝ) ! CB (Ĝ)
1
' 7! L 0 '( ) ··= '( 0 )
⌫ˆL 0'
= ⌫ˆ' 0 ,
where for any measure µ on a group and for any element z in the group, the measure µz
is defined as µz (E) ··= µ(z 1 E) for any measurable set E. In particular, if f 2 L1 (G)
and g is obtained by multiplication by a character 0 , then the associated functions ĝ and
µ̂g are obtained by fˆ and µ̂f respectively by multiplication via translation by 0 .
Proof. (i) By Lemma 4.3.10 there exists a net of functions f 2 V 1 (G) such that fˆ converges
to the constant function 1 on compact subsets of Ĝ. This means that
(ii) Since ⌫ˆ' is unique, then by Lemma 4.3.9 if ' = fˆ for some f 2 V 1 (G), then ⌫ˆ' = µ̂f .
(iii) By Bochner’s Theorem, the measure µ̂f that arise from the net in (4.3.12) are positive.
Therefore, since ⌫ˆ' is unique, then it is positive.
(iv) The additivity of ' 7! ⌫ˆ' follows immediately from the uniqueness of ⌫ˆ' . Now, if we
consider any a 2 CB (Ĝ), then the equality
so that
L 0 '( )dµ̂f ( ) = fˆ( )dˆ
⌫' 0 ( ).
This completes the proof. ⌅
137
Now we are ready to prove the main theorem of this section:
Theorem 4.3.10 (Fourier Inversion Formula). Let G be a LCA group. There exists a measure
d in Ĝ such that for all f 2 V 1 (G),
Z
f (y) = fˆ(y) (y)d .
Ĝ
Proof. The identification of V 1 (G) with V 1 (Ĝ) will be proved in the following section. For now,
we only prove the inversion formula in three steps.
Step 1: Show that if ↵ 2 K (G)
(G), then ↵ 2 F .
compact
Let K̂ ✓ Ĝ such that supp(↵) ✓ K̂. By Lemma 4.3.10 there exists a funtion f 2 V 1 (G)
such that fˆ is bounded away from zero on K. Hence,
↵
a ··=
fˆ
is bounded and continuous on K̂ and it can be extended to a continuous and bounded function
on Ĝ (it suffices to define it to be zero on Ĝ \ K̂). Since fˆ 2 F , then by (iv) of the previous
lemma, ↵ = afˆ 2 F .
Step 2: Choose the Haar measure on Ĝ
Ĝ.
Consider the map
⌘ : K (Ĝ) ! C
Z
↵ 7! 1dˆ
⌫↵ ( ).
By step 1, this map is well-defined. Now, if f 2 V 1 (G) is nonzero, then µ̂f 6= 0, i.e., there
exists a continuous and bounded function a : Ĝ ! C such that a( )dµ̂f ( ) 6= 0. If ↵ = afˆ,
then dˆ⌫↵ ( ) = a( )dµ̂f ( ), so that ⌘ is nonzero. Therefore, by (v) of the previous lemma,
Z Z Z Z
1dˆ⌫L 0 ↵ ( ) = 1dˆ ⌫↵ 0 ( ) = L 0 1 1dˆ
⌫↵ ( ) = 1dˆ⌫↵ ( ),
so that ⌘ is a Haar measure d on Ĝ. Therefore, with respect to this measure we may write
Z Z
↵( )d = dˆ ⌫↵ ( ). (4.3.13)
i.e.,
'( )d = dˆ
⌫' ( ),
and by construction, Z Z
f (y) = (y)dµ̂f ( ) = fˆ(y) (y)d . ⌅
138
Corollary 4.3.10.1. Let f : G ! C be an integrable function with respect to the Haar measure
on G.
Therefore, the Fourier transform defines an injection between V 1 (G) and V 1 (Ĝ).
Proof. (i) By the Fourier Inversion Formula, f is precisely the Fourier transform of the
measure fˆ( )d . By Bochner’s Theorem, fˆ( )d is a positive Radon measure of total
finite mass. Therefore, since fˆ is continuous, it is positive.
Z
(ii) f (x) dx = fˆ(1) > 0.
Therefore, since every element of V 1 (G) can be written as a linear combination of positive
integrable functions16 , then by part (iii) of this Corollary, the map
V 1 (G) ! V 1 (G)
f 7! fˆ
Throughout this section G will always denote a LCA group. The aim of this section is to show
that the canonical map
ˆ
↵ : G ! Ĝ
y 7! ↵(y)( ) = (y)
is an isomorphism. This is known as Pontryagin duality. In order to do so, we begin with the
following lemma:
P3
16
This is a consequence of the fact that every z 2 C can be written as z = k=0 ik max{0, Re(ik z)}.
139
Proof. Suppose that z 2 G is not the identity. Since G is a group it suffices to show that there
exists some 2 Ĝ such that (z) = 6 1. Suppose for the sake of contradiction that does not
exists. Then,
fˆ = Ldzf
for all f 2 L1 (G). Therefore, by the Fourier Inversion Formula, f = Lz f for all f 2 V 1 (G).
Now, since G is Hausdor↵, there is some open neighborhood of the identity such that
U \ z 1 U = ?.
Now, we know that there exists some g 2 K (G) such that supp(g) ✓ V and g(e) = 117 . But
for this function g it is impossible that g = Lz g; a contradiction. ⌅
Proposition 4.3.1. The subsets WG (K̂, V ) and their translates constitute a base for the topol-
ogy of G.
Proof. Let U be an open neighborhood of e 2 G. Then, there exists a positive definite con-
tinuous function g such that supp(g) ✓ U and g(e) = 1. By Corollary 4.3.10.1, ĝ is positive.
Moreover, by the Fourier Inversion Formula,
Z
ĝ( )d = 1.
Therefore, we may identify ĝ( )d with a finite Radon measure on Ĝ. Therefore, given " > 0
compact
there is K̂ ✓ Ĝ such that Z
ĝ( )d > 1 ",
K̂
or equivalently Z
ĝ( )d < "
Ĝ\K̂
17
R
By Uryshon’s Lemma there is some f 2 K (G) such that V (2) ✓ U , f (e) = 1 and f (s) ds = 1. Let
g ··= f ⇤ f˜, then g is positive-definite, supp(g) ✓ U and g(e) = 1.
140
so that
V = {z 2 C | |z 1| < "}
and taking y 2 WG (K̂, V ). Therefore, |g(y)| > 1 3". By construction, supp(g) ✓ U , so that
WG (K̂, V ) ✓ U . ⌅
Proof. By a standard result in topology, a dense and locally compact subset of a Hausdor↵
space is open. Since G is locally compact and ↵ is a homeomorphism, then ↵(G) is locally
open
compact. Clearly ↵(G) is dense on ↵(G). Thus ↵(G) ✓ ↵(G). But ↵(G) is a subgroup of
closed
↵(G) and thus by (iv) of Proposition 4.1.1 ↵(G) ✓ ↵(G). Hence ↵(G) = ↵(G). ⌅
ˆ
Therefore, to complete the proof it suffices to show that ↵(G) is dense on Ĝ. To prove this
we still need some sophisticated tools. We begin by extending the Fourier transform to L2 (G):
Let f 2 L1 (G) and as usual, let f˜ be the function defined by f˜(x) = f (x 1 ). It is easy to see
that
fˆ˜( ) = fˆ( ).
Let g = f ⇤ f˜. Then, g is positive definite. If moreover f 2 L2 (G), then by the Fourier Inversion
Formula
Z
|f (x)|2 dx = g(e)
Z
= ĝ( )d
Z
= |fˆ( )|2 d .
141
which is an isometry onto its image. Let
Â1 (G) ··= {fˆ | f 2 L1 (G) \ L2 (G)}.
Note that Â1 (G) is stable under multiplication by elements of ↵(G):
\
(↵(y \
0 )f )( ) = (Ly0 f )( ).
Proof. By Lemma 4.2.2, it suffices to show that if g 2 L2 (Ĝ) is orthogonal to every f 2 Â1 (G),
then g = 0. Since Â1 (G) is stable under multiplication, then for all y 2 G,
Z
g( )f ( ) (y)d = 0,
Proof. This result follows immediately from the previous lemma and the fact that L1 (G)\L2 (G)
is dense on L2 (G). ⌅
Corollary 4.3.11.1 (Parseval’s Identity). For all f, g 2 L2 (G) we have
Z Z
f (x)g(x) dx = fˆ( )ĝ( )d .
= (fˆ ⇤ ĝ)( 0 ). ⌅
Corollary 4.3.11.3. Â(G) consists precisely of convolutions of function in L2 (G).
This shows that h = f · g, where f, g 2 L2 (G) and by the previous Corollary, ĥ = fˆ ⇤ ĝ. Now,
by Plancherel’s Theorem, every convolution in L2 (G) is of the form fˆ ⇤ ĝ, where f, g 2 L2 (G).
Let h = f · g 2 L1 (G), then again by the previous Corollary, fˆ ⇤ ĝ = ĥ. ⌅
142
open
Lemma 4.3.14. Let Û ✓ Ĝ non empty. Then, there exists a non-zero function fˆ 2 Â(G)
such that supp(fˆ) ✓ Û .
compact
Proof. By the regularity of the Haar measure, there is some K̂ ✓ Û with positive measure.
We also know that there is some open neighborhood of the identity V̂ such that V̂ K̂ ✓ Û . We
may assume that V̂ has finite measure. Now, let fˆ = 1K̂ ⇤ 1V̂ , where 1E is the characteristic
function of E. By the previous corollary, fˆ 2 Â(G) and by construction,
supp(fˆ) = V̂ K̂ ✓ Û .
Moreover, Z Z
fˆ( )d = (1K̂ ⇤ 1V̂ )( )d = µ̂(K̂)µ̂(V̂ ) > 0.
is an isomorphism.
ˆ
Proof. Recall that it suffices to show that ↵(G) is dense on Ĝ. If this is not the case, then by
ˆ
the previous lemma, there is some ' 2 L1 (G) such that 'ˆ 6= 0 but 'ˆ = 0 on ↵(G). Let ˆ0 2 Ĝ.
Then, by definition, Z
ˆ( ˆ0 ) = ( ) ˆ0 ( 1 )d .
for all y 2 G. Hence ' = 0 almost everywhere and therefore 'ˆ = 0; a contradiction. ⌅
To end this chapter we complete the proof of the Fourier inversion formula (recall that we need
to show that the Fourier transform identifies V 1 (G) with V 1 (Ĝ)). In Corollary 4.3.10.1 we
showed that the map
V 1 (G) ! V 1 (Ĝ)
f 7! fˆ
ˆ
Identifying G with Ĝ, this equality is written as
f (y) = F̂ (y 1 ),
143
so that f 2 V 1 (G) by Corollary 4.3.10.1. By the Fourier Inversion Formula,
Z
F ( ) = F̂ (y) (y) dy
Z
= f (y 1 ) (y) dy
Z
= f (y) (y) dy
=fˆ( ).
This completes the proof of the identification between V 1 (G) with V 1 (Ĝ).
144
Chapter 5
5.1 Preliminaries
This chapter is based on [4], [18], [21], [22], [28] and [33]. Before exploring the beauty of
Tate’s thesis we need some algebraic background. For this purpose we will state some algebraic
definitions and results. Most of the theorems in this section will be provided without proof but
they can be found in the books mentioned above.
(ii) A principal ideal domain is an integral domain in which every ideal is principal.1
(iii) A discrete valuation ring is a principal ideal domain with exactly one nonzero prime
ideal.2
Example 5.1.1. Let p 2 P. Consider the ring
na o
Z(p) ··= 2 Q | (a, b) = 1 and p - b .
b
a
Clearly Z(p) is an integral domain. Now, let I be any nontrivial ideal in Z(p) . If 2 I, with
b
a 6= 0 and p - b, let m be the maximum exponent such that pm |a, so pm+1 - a. Thus a = pk c for
b
some c 2 Z such that p - c. Therefore 2 Z(p) and
c
ab
pm = 2 I.
bc
1
A principal ideal I in a ring R is an ideal that is generated by a single element a 2 R through multiplication
by every element of R.
2
An ideal P of a commutative ring R is prime if P 6= R and if a, b 2 R are elements such that ab 2 P , then
a 2 P or b 2 P . An element p of R is said to be prime if the principal ideal (p) generated by (p) is a nonzero
prime ideal.
145
This shows that there is some m 2 N such that pm 2 I. Let n be the least natural number
x
such that pn 2 I. Clearly the ideal generated by pn is contained in I. Now suppose that 2 I.
y
Similarly as before, x = pm z, with p - z. By construction, m > n, so that
x pm n z
= pn
y y
belongs to the ideal generated by pn . Therefore I = pn Z(p) for some n 2 N, whence every ideal
of Z(p) is principal. Therefore Z(p) is a principal ideal domain and since (p) is in fact the only
nonzero prime ideal, Z(p) is a discrete valuation ring.
Definition 5.1.1. The unique nonzero prime ideal of a discrete valuation ring R is of the form
R⇡, where ⇡ is a prime element of R. We call such an element an uniformizing parameter of
R.
Note that in a discrete valuation ring R the unique nonzero prime ideal p is also maximal.
Definition 5.1.2. Let R be a discrete valuation ring with unique maximal ideal p. The residue
field is defined as the quotient ring R p.
It is easy to verify that ⇠ is an equivalence relation. Hence we may define the field of fractions
of R as
Frac(R) ··= R ⇠.
Example 5.1.2. Note that Frac(Z) = Q (this is the usual definition of Q).
Definition 5.1.3. Let R be an integral domain with field of fractions K = Frac(R). A fractional
ideal of K is a nonzero R submodule J of K such that there exists a non-zero r 2 R such that
rJ ✓ R. We say that J is invertible if there exists some fractional ideal I such that IJ = R.
Definition 5.1.4. A discrete valuation on a field K is a nontrivial group homomorphism
⌫ : K ⇥ ! Z such that ⌫(a + b) > min{⌫(a), ⌫(b)}.
Since ⌫ is not the zero homomorphism, its image is a nonzero subgroup of Z, and therefore is
of the form mZ for some m 2 Z.
Example 5.1.3. Let M be the field of meromorphic functions on a connected open subset U
of C. Let f 2 M ⇥ and for each z 2 U , let
8
>
>
< m, if f has a pole of order m at z;
⌫z (f ) ··= m, if f has a zero of order m at z;
>
>
:0, otherwise.
146
The previous example shows that every discrete valuation ring gives rise to a discrete valuation
on its field of fractions. The following theorem is a converse to this statement and it justifies
the name “discrete valuation ring”.
n=N
This field can also be constructed as the field of fractions of the integral domain k[[t]] of formal
power series over a field k.
| · |: K ! R
x 7! |x|
147
The following example shows that any discrete valuation induces a nonarchimedean absolute
value:
Example 5.1.6. Let K be a field and let ⌫ : K ⇥ ! Z be a discrete valuation. Let a 2 R with
a > 1. For any x 2 K, let (
a ⌫(x) , if x 6= 0,
|x| ··=
0, otherwise.
Then | · | is a nonarchimedean absolute value on K. For example, for K = Q, given any prime
p 2 P, we have the p-adic absolute value | · |p on Q: Taking a = p, for any q 2 Q, we have
(
p ⌫p (q) , if q 6= 0,
|q|p ··=
0, otherwise,
The completion of Q with respect to the p-adic absolute value is a field denoted by Qp , called
the p-adic rational numbers.
Theorem 5.1.2. Let | · | be a nontrivial nonarchimedean absolute value, and let ⌫ : K ⇥ ! R
be defined as ⌫(x) ··= log |x|. Then for all x, y 2 K ⇥ , we have
! : K ⇥ ! Z.
A ··= {x 2 K | |x| 6 1}
is a subring of K, with
U ··= {x 2 K | |x| = 1}
as its group of units, and
m ··= {x 2 K | |x| < 1}
is a maximal ideal. Moreover, the absolute value | · | is discrete if and only if m is principal, in
which case A is a discrete valuation ring.
Now we need some Galois theory; for a complete introduction see [20] and [32].
Definition 5.1.6. A field extension is a monomorphism4 ◆ : K ! L, where K and L are fields.
4
An injective homomorphism is called a monomorphism and a surjective homomorphism is called a epimor-
phism.
148
Let K and L be fields and let ◆ : K ! L be a field extension. We usually identify K with its
image ◆(K), and in this case K becomes a subfield of L.
We denote by L : K an extension where K is a subfield of L. In this case, ◆ is the inclusion
map. We define the degree [L : K] of L : K as dimK L.
A field extension L : K is called algebraic if every element of L is algebraic over K, i.e., if every
element of L is a root of some nonzero polynomial with coefficients in K.
Definition 5.1.7. Let K be a field. A nonzero polynomial f 2 K[t] splits over K if it can be
expressed as a product of linear factors
f (t) = a(t ↵1 ) · · · (t ↵n ),
where a, ↵1 , . . . , ↵n 2 K.
Definition 5.1.8. Let K be a field and let ⌃ be an extension of K. Then ⌃ is a splitting field
for the polynomial f over K if
(i) A field extension L : K is normal if every irreducible polynomial f over K that has at
least one zero in L splits in L.
(ii) An irreducible polynomial f over a field K is separable over K if it has no multiple zeros
in a splitting field. An algebraic extension L : K is a separable extension if every ↵ 2 L
is separable over K.
(iii) A field extension L : K is called Galois extension if it is both separable and normal. A
K-automorphism is defined to be an automorphism of L that fixes K pointwise, i.e., an
isomorphism ↵ : L ! L such that ↵(x) = x for each x 2 K. If L : K is a Galois extension,
then the set of all K-automorphisms forms a group under composition; this is called the
Galois group of L over K and is denoted by Gal(L : K).
Let L : K be a finite field extension. Some important tools in algebraic number theory are the
trace and the norm of an element ↵ 2 L with respect to L : K. They are defined as follows:
Let ↵ 2 L and let m↵ : L ! L be the map defined by m↵ (x) = ↵x. Then, m↵ is a K-linear
transformation. The trace trL:K (↵) is defined as the trace of this linear transformation. The
norm NL:K (↵) is defined as the determinant of this linear transformation. In the particular case
that the field extension L : K is a Galois extension, the trace and the norm may be computed
as follows: X Y
trL:K (↵) = (↵) and NL:K (↵) = (↵).
2Gal(L:K) 2Gal(L:K)
Definition 5.1.9. Let L : K be a field extension and suppose that ↵ 2 L is algebraic over
K. The minimal polynomial of ↵ over K is the unique monic polynomial P over K of minimal
degree such that P (↵) = 0. The degree of ↵ is the degree of the minimal polynomial of ↵.
149
5.1.3 Local Fields
The simplest examples of algebraic number fields di↵erent from Q are given quadratic fields:
Definition 5.1.11. A quadratic field is a number field of degree 2.
If K is a quadratic field, then any x 2 K \ Q is of degree 2 over K and thus K = Q[x] and
(1, x) is a basis for K over Q. Let F (X) = X 2 + bX + c be the minimal polynomial of x. Then,
p
2x = b ± b2 4c,
p
so that K = Q( b2 4c). Now,
u uv
b2 4c = = 2 ,
p v v p
where u, v 2 Z. From this, K = Q( uv) and in fact K = Q( m) where m is a square-free
integer and m 6= 1.
Definition 5.1.12. A local field is a field F satisfying one of the following equivalent conditions:
(i) F is R or C, or else F is the field of fractions of a complete discrete valuation ring with
finite residue field.
(ii) F is the completion of a global field with respect to a nontrivial absolute value.
(iii) F is a nondiscrete locally compact topological field. N
Note that by (ii) of the previous definition, a local field F is endowed with a nontrivial absolute
value | · |.
If F is R or C, then F is called archimedean; other fields are called nonarchimedean. Therefore,
with the notation of Theorem 5.1.3 if K is a nonarchimedean local field with discrete absolute
value | · |, then the discrete valuation ring A is called the ring of integers of K and is denoted
by OK or simply by O when there is no confusion. If a is a nonzero ideal of O, we define the
absolute norm of a as
N (a) ··= [O : a] = |O a|.
Moreover, by (iii) of the previous definition, a local field F gives rise to two locally compact
groups, namely (F, +), (F ⇥ , ·), with their respective Haar measures. The following classification
theorem will prove to be really useful:
Theorem 5.1.5. Let F be a local field. Then
Another important fact about local fields is that they enjoy the Heine–Borel property:
Theorem 5.1.6 (Heine–Borel Theorem). Let F be a local field and let S ✓ F . Then, S is
compact if and only if S is closed and bounded. The same holds for S ✓ F n for any n > 0.
150
5.1.4 Places
Definition 5.1.13. Let K be a field. Two absolute values | · | and | · |0 on K are equivalent if
there is some ↵ > 0 such that
|a|0 = |a|↵
for all a 2 K. A place on K is an equivalence class of nontrivial absolute values.
Definition 5.1.14. A prime global field is a field that is either the rational numbers Q or a
function field Fq (t).
The purpose of this section is to classify the places of prime global fields. To begin, we need
the following lemma:
for all m, n 2 N0 . Then either F (m) 6 1 for all m 2 N0 or F (m) = m↵ for all m 2 N0 and for
some ↵ > 0.
for all m0 , . . . , mr 2 N0 . Suppose that n, m > 2 and let b = max{f (0), . . . , f (n 1)}. Express
m in base n as follows: r
X
m= d l nl (0 6 dl < n).
l=0
151
W.l.o.g suppose that dr 6= 0, so that nr 6 m. Then,
f (m) a + f (n) b
6 + .
log m log n log m
f (m) = ↵ log m
for some ↵ > 0. If ↵ = 0, then F (m) 6 1 for all m > 2. If ↵ > 0, then F (m) = m↵ for all
m > 2. For the cases m = 0, 1, since F is strictly multiplicative, then
Hence, if F is not constant, F (0) = 0 and F (1) = 1, so that F (m) 6 1 for m = 0, 1. This
completes the proof. ⌅
Theorem 5.1.7. Let K be a prime global field.
(i) (Ostrowski) Suppose that K = Q. Then every nontrivial place of K is given either by the
equivalence class of the usual absolute value denoted by | · |1 or by the equivalence class
of a p-adic absolute value | · |p for some prime p.
(ii) Suppose that K = Fq (t). Then every nontrivial place of K is given either by the ‘infinite
place’ defined as the equivalence class of the absolute value | · |1 given by
f
= q deg(f ) deg(g)
g 1
|x + y| 6 max{|x|, |y|}
for all x, y 2 Q. This inequality shows that |n| 6 1 for all n 2 N. Since | · | is nontrivial,
then |n| < 1 for some n 2 N. Let p be the smallest positive integer with this property.
Let us show that p is prime: suppose on the contrary that p = m1 m2 with m1 , m2 > 1.
Then, the inequality 1 > |p| = |m1 ||m2 | implies that |m1 | < 1 or |m2 | < 1, and this
contradicts the minimality of p.
Now let us show that if gcd(a, p) = 1, then |a| = 1: by the division algorithm, a = dp + r
for integers d, r and 0 < r < p. Since p is the smallest positive integer such that |p| < 1,
then |r| = 1. On the other hand,
152
so that |a| = 1. Therefore,
|apm | = |p|m
for all m 2 Z where gcd(a, p) = 1. The p-adic norm is of the form
1
|p|p = <1
p
and from this it is clear that | · | and | · |p are equivalent. Now suppose that | · | is
archimedean. When | · | is restricted to N, by the triangle inequality we have
|n + m| 6 2 max{|n|, |m|}
| · | = | · |↵1
for some ↵ > 0 and therefore | · | and | · |1 represent the same place.
Lemma 5.1.2. Let K be a field. Then, any nontrivial absolute value on K(t) that is
trivial on K is equivalent to | · |1 or | · |P for some monic irreducible polynomial P 2 K[t].
Xn ✓ ◆
n
|f + g| 6
n
|f |j |g|n j
j=0
j
6(n + 1) max{|f |, |g|}n .
|f + g| 6 max{|f |, |g|},
f (t) = a0 + a1 t + . . . + an tn .
Suppose first that |t| > 1. Since | · | is trivial on K, then |aj | = 0, 1 for all j and |an | = 1
since an 6= 0. Therefore,
|an tn | = |t|n
and for j = 0, . . . , n 1,
|aj tj | 6 |t|j < |t|n
Hence, since | · | is nonarchimedean,
Let
1
c= < 1,
|t|
deg(f )
so that |f (t)| = c . This shows that | · | is equivalent to | · |1 .
153
Now suppose that |t| 6 1. Then,
Thus, since | · | is nontrivial there is some polynomial with absolute value less than 1.
Moreover, since | · | is trivial on K, then such polynomial is nonconstant. Therefore, let
P be a nonconstant polynomial of minimal degree such that |P (t)| < 1. We show that P
is irreducible: if P (t) = a(t)b(t), where deg(a), deg(b) < deg(P ), then
|P (t)| = |a(t)||b(t)| = 1;
where h is not divisible by P . A similar argument to the one on part (i) of this Theorem
shows that |h(t)| = 1. Therefore,
whence | · | is equivalent to | · |P . ⌅
Let K be a global field. If K is an algebraic number field, then the nonarchimedean places of
K are also called finite places; the archimedean ones are also called infinite places. If K is a
function field with p = char(K) > 0, we fix an element t 2 K such that K is a finite separable
extension of Fq (t) with Fq = K \ Fq . The infinite places of K are the ones lying over the place
of Fq (t) represented by
f
= q deg(f ) deg(g) .
g 1
The remaining places of K are called finite. The set of all places of a global field K will be
denoted by V (K), the set of finite places will be denoted by Vf (K) and the set of infinite places
will be denoted by V1 (K).
Definition 5.1.15. Let K be a global field. Then we define the ring of integers of K as the
set \
OK = {x 2 K | |x|⌫ 6 1},
⌫2Vf (K)
i.e., OK is the intersection of the local rings of integers at all finite places of K and is therefore
itself a ring.
b n + an 1 b n 1
+ · · · + a1 b + a0 = 0.
The integral closure of A on B is the set of all elements b 2 B that are integral over A.
154
Definition 5.1.16. Let R be an integral domain. We say that R is a Dedekind domain if R
satisfies the following three properties:
It can be shown that if R is a Dedekind domain, then every fractional ideal is invertible. More-
over this property characterizes Dedekind domains among integral domains (see for example [1]).
We shall need the following theorems from [25]:
Theorem 5.1.8. Let R a Dedekind domain with field of fractions K = Frac(R). Then, every
nonzero ideal in R has a unique factorization as a product of prime ideals. In particular, the
group of nonzero fractional ideals is isomorphic to the free abelian group generated by nonzero
prime ideals, i.e., if a is a nonzero fractional ideal, then
Y
a= prp
p
where the product runs over all the nonzero prime ideals, rp 2 Z and rp = 0 for all but finitely
many p.
Theorem 5.1.9. Let K be a number field with ring of integers OK . Then, the field of fractions
of OK is K and OK is the integral closure of Z on K if char(K) = 0 and the integral closure
of Fq [t] on K if char(K) > 0. In particular, OK is a Dedekind domain.
p
Let K = Q( m) be a quadratic number field and let : K ! K be the function defined by
p p
(a + b m) = a b m.
If x 2 OK , then (x) is a root of the same equation of integral dependence as x ( is an
automorphism), and thus (x) 2 OK . Hence,
x + (x) 2 OK and x (x) 2 OK .
p
If x = a + b m, then
x + (x) = 2a 2 Q and x (x) = a2 mb2 2 Q.
Therefore,
2a 2 Z and a2 mb2 2 Z,
whence (2a)2 m(2b)2 2 Z and since 2a 2 Z, then m(2b)2 2 Z but since m is square-free, then
2b 2 Z. Hence, we may write
u v
a= and b = ,
2 2
where u, v 2 Z and the condition a 2
mb 2 Z becomes
2
u2 mv 2 2 4Z.
If v is even, then this condition implies that u is also even. In this case, a, b 2 Z. If v is odd,
then v 2 ⌘ 1 (mod 4). Now, u2 ⌘ 0, 1 (mod 4), but since m is square-free, then m cannot be a
multiple of 4 and thus u2 ⌘ 1 (mod 4) and m ⌘ 1 (mod 4). We may summarize what we have
done in the following theorem:
155
p
Theorem 5.1.10. Let K = Q( m) be a quadratic field where m is a square-free integer distinct
from 1.
p
(i) If m ⌘ 2, 3 (mod 4) , then OK consists of elements of the form a + b m, where a, b 2 Z.
p
(ii) If m ⌘ 1 (mod 4), then OK consists of elements of the form 12 (u + v m), where u, v 2 Z
have the same parity.
Definition 5.1.17. Let {Xj } be a family of topological spaces indexed by j 2 J and let {Uj }
be a family of open sets such that Uj ✓ Xj for all j 2 J. The restricted direct product is the
topological space
Y0 Y
(Xj , Uj ) ··= {(xj ) 2 Xj | xj 2 Uj for all but finitely many j 2 J}
As sets, Y Y0 Y
Uj ✓ (Xj , Uj ) ✓ Xj ,
but in general the topology of Q
the restricted direct Q
product is not the
Q0 same as the one inherited
Q
from the product topology of Xj . For example, Uj is open in (Xj , Uj ) but not on Xj
unless Uj = Xj for all but finitely many j.
Theorem 5.1.11. Let {Xj }j2J be a family of locally compact and Hausdor↵ topological spaces
and let {Uj }j2J be its corresponding family
Q of open sets Uj ✓ Xj such that Uj is compact for
all but finitely many j 2 J. Then X = 0 (Xj , Uj ) is locally compact.
is locally compact and since Uj is compact for all but finitely many j and moreover Uj is open
in a locally compact Hausdor↵ space, then
Y
Uj
j 2I
/
is locally compact on the product topology. Now, the key point is the following fact: the
product topology on XI agrees with the topology on the restricted direct product. Since XI
covers X we complete the proof. ⌅
156
Definition 5.1.18. Let K be a global field and let K⌫ denote the completion of K with respect
to a place ⌫. The adèle ring of K is the restricted product
Y 0
AK ··= (K⌫ , O⌫ ),
⌫2V (K)
Let F be a local field with Haar measure dx on the additive group (F, +). If the field F is
nonarchimedean, we shall always use the following notation:
O : the ring of integers (or valuation ring) of F,
p : the unique prime ideal of O,
Fq : the finite residue field O p with q elements,
⇡ : a uniformizing parameter of F, i.e. ⇡O = p.
By Theorem 5.1.3 we know that O is the closed unit disk of F centered around 0 and p is the
open unit disk centered around 0. Explicitly,
O = {x 2 F | |x| 6 1} and p ··= {x 2 K | |x| < 1}.
Now, we shall need the following function on F called the normalized valuation or normalized
absolute value:
157
Definition 5.2.1. Let F be a local field and for each a 2 F ⇥ , define the function |·| : F ⇥ ! R⇥
+
as follows:
Note that the normalized valuation is not an absolute value in the case F = C because the
function |·| does not satisfy the triangle inequality. We shall interpret this normalized valuation.
In order to do so, we first need to extend the concept of Schwarz functions:
Recall that a function f : R ! C is a Schwarz function if f 2 C 1 (R, C) and f with all
its derivatives decrease faster than any polynomial. We can generalize this idea to functions
f : Rn ! C. For this purpose we introduce the following nomenclature:
and
D↵ f ··= @1↵1 · · · @n↵n f,
@ ↵i
where @i↵i = .
@x↵i
i
Intuitively, we are saying that f and all its derivatives decrease faster than any polynomial.
158
m(x) 2 N. We may assume w.l.o.g that Ux = x + pm(x) . Now, since supp(f ) is compact, then
we may choose a finite cover {Uxi | i = 1, . . . , n}, where
Uxi = xi + pm(xi ) .
Now, w.l.o.g we may assume that m(xi ) = m is a constant natural number. By the Heine–
Borel Theorem, supp(f ) is bounded and therefore is contained in a set of the form p h for some
h 2 N. With this we have proved the following proposition:
Proposition 5.2.1. Let F be a nonarchimedean local field and let f 2 S (F ). Then, there
exists integers h, m with h 6 m such that f (x) = 0 for x 2
/ p h and for x 2 p h , we have
f (y) = f (x) for all y 2 x + pm .
Now, let f 2 S (F ) with supp(f ) = p h . Then, with the notation of the previous proposition,
Z XZ X
f (x) dx = 1y+pm (x)f (x) dx = N (p m )f (y)µ(O), (5.2.2)
y2R y2R
h
where now R is a system of representatives for p pm in p
h
. In particular, for all ↵ 2 F ⇥ ,
letting f = 1↵O , by (5.2.1) we have
µ(↵O) 1
= N (↵O) = |↵|.
µ(O)
In the archimedean case, if we let
and
OC ··= {↵ 2 C | |↵| 6 1},
then it is easy to verify that for all x 2 R (or x 2 C),
(
µ(xOF ) |x|, if F = R;
=
µ(OF ) |x|2 , if F = C.
159
Theorem 5.2.1. Let F be a local field. Then, for every ↵ 2 F ⇥ ,
µ(↵OF ) = |↵|µ(OF ),
By the previous theorem, the normalized valuation | · | may be interpreted as follows: Let F
be a local field and let ↵ 2 F ⇥ . Then, multiplication by ↵ is an isomorphism, i.e., the map
F !F
x 7! ↵x.
is an isomorphism, under which dx pulls back5 to another Haar measure, and by the uniqueness
of the Haar measure it must be a positive multiple of dx:
F !F
'(↵) dx dx.
: F ! F̂
a 7! a
where K is compact and U is open and their translates constitute a basis for the topology of
F̂ . Let us show that is a homeomorphism onto its image:
compact
(i) Let K ✓ F and let U ✓ F be an open neighborhood of 1 2 U . Since K is bounded
1
and U contains an open disk around 0, then there is some > 0 such that
1
|a| < =) aK ✓ U,
whence is continuous.
5
This is just a fancy word for precomposition, i.e., if f : B ! C and g : A ! B are functions, then the
function h = f g is the pullback of f by the function g.
160
(ii) Now, let " > 0 and let b 2 F such that (b) 6= 1. Let U be an open neighborhood of
1
1 2 U such that (b) 2
/ U , so that b 2
/ U . Let
n |b| o
K ·= x 2 F | |x| 6
· .
"
1
If aK ✓ U , then b 2
/ aK, so that
|b|
|b| > |a| ,
"
1
or equivalently, |a| < ". This shows that is continuous.
(F )? = {0},
where
(F )? ··= {b 2 F | (b) = 1}.
To prove this, let b 2 (F )? . Then, since is nontrivial, a (b) = 1 for all a 2 F implies that
b = 0. ⌅
Example 5.2.1 (Standard Characters). There’s a standard character for each local field F :
161
(iv) Now, let F0 = Fp ((t)) and define F ! T by
0:
⇣X ⌘ 2⇡ia 1
j ·
0 a j t ·= e p .
Now, if we fix a character 2 F , we consider the Fourier transform that is obtained from the
identification of F and F̂ :
Definition 5.2.6. Let F be a local field and fix a character 2 F . Given a Schwartz–Bruhat
function f 2 S (F ), we define the Fourier transform of f as the function fˆ: F ! C defined by
Z
ˆ
f (y) = f (x) (xy) dx.
F
ˆ
fˆ(x) = rf ( x).
for all x 2 F ⇥ . We may choose dx in such a way that r = 1. Then, the Fourier inversion
formula Z
f (x) = fˆ(y) (xy) dy
F
holds for all x 2 F . This measure dx is called the self-dual measure relative to . In order to
describe explicitly this self-dual measure in any local field, we first need the following definition:
Definition 5.2.7. Let F be a noncarchimedean local field and let 2 F̂ . Then, there is some
m 2 Z such that |pm = 1. Let m be the smallest integer with that property. The fractional
ideal pm is called the conductor of .
Therefore, Z
(yx) dx = 0. ⌅
a
162
With this lemma, we are ready to give an explicit description of the self-dual measure in any
local field F :
Theorem 5.2.3. Let F be a local field. Then, the unique Haar measure µ on F that is self-dual
(relative to the standard ) can be described as follows:
Proof. It suffices to show in this case that the Fourier inversion formula holds for a single
nonzero f 2 S (F ).
(i) For F = R, recall that (x) = e 2⇡ix . In this case, we take the function f : R ! [1, 1)
defined by f (x) = e ⇡x . Then, by Corollary 1.5.2.1, fˆ = f .
2
Moreover, (
Z
c µ(O)µ(pm ), if y 2 O
1c
O (y) = µ(O) (yx) dx =
pm 0, otherwise.
Therefore, the self-dual measure is the one that satisfies
µ(O)µ(pm ) = 1.
But (5.2.1) tells us that
1
µ(pm ) = µ(O).
N pm
Thus one obtains the self-dual measure for
p
µ(O) = N pm . ⌅
163
5.2.2 Analysis on Local Multiplicative Groups
Let F be a local field with normalized valuation | · |. We define the following subsets of F ⇥ :
8
>
<{±1}, if F = R;
>
⇥
UF ··= {x 2 F | |x| = 1} = T, if F = C;
>
>
:O ⇥ , otherwise.
and (
R⇥+ , if F is archimedean;
VF ··= |F ⇥ | = {|x| | x 2 F ⇥ } = Z
q , otherwise.
This sets are called the unit group and value group respectively. Note that
F⇥ ⇠
= UF ⇥ V F .
Definition 5.2.8. Let X(F ⇥ ) be the set of all continuous homomorphisms from F ⇥ to C⇥ .
The elements 2 X(F ⇥ ) are called quasicharacters.
Now we would like to relate the quasicharacters of F ⇥ to the (unit) characters of F ⇥ . For this
purpose we need the following lemma:
Lemma 5.2.2. If f : R⇥ ⇥
+ ! R+ is a continuous homomorphism, then f (x) = x for some
s
s 2 R.
Proof. Let x 2 R⇥
+ . Since f is continuous, there is some y 2 R such that
Zy
dt
f (t) 6= 0.
1 t
Now, Zy Zy Z yx
dt dt dt
f (x) f (t) = f (xt) = f (t) ,
1 t 1 t x t
so that Z yx
dt
f (t)
t
f (x) = Zxy .
dt
f (t)
1 t
This shows that f is di↵erentiable at x. Now,
⇣ ⌘
h
f (x + h) f (x) f 1+ x
1 f (x)
f 0 (x) = lim = f (x) lim =s ,
h!0 h h!0 h x
where s = f 0 (1). Therefore,
log f (x) = s log x + C,
for some C 2 R, but since f is a homomorphism, f (1) = 1, whence,
f (x) = xs . ⌅
164
With this lemma we are ready for the following proposition:
Proposition 5.2.2. For every unramified quasicharacter 2 X(F ⇥ ) there is some s 2 C such
that (x) = |x|s for all x 2 F ⇥ .
F⇥ ⇠
= UF ⇥ V F
˜ : V F ! C⇥
|x| 7! ˜(|x|).
R⇥
+ ! R+
⇥
x ! | ˜(x)|
R!T
y 7! ˜(ey ) e y
+it
˜(x) = x .
(ii) Now, we consider the case in which F is nonarchimedean, so that VF = q Z and thus
˜ : q Z ! C⇥
log ˜(q)
s= ,
log q
2⇡i
and s is unique modulo a multiple of . Therefore
log q
for all x 2 F ⇥ . ⌅
165
Corollary 5.2.2.1. Every quasicharacter 2 X(F ⇥ ) factors into the product
= ˜ | · |s ,
= | |
| |
Note that while s 2 C may not be necessarily unique, nevertheless Re(s) is unique. Accordingly,
we call ··= Re(s) the exponent of .
Definition 5.2.10. Let F be a nonarchimedean local field and let 2 X(F ⇥ ) be a quasichar-
acter of F ⇥ . Then, |1+pn = 1 for some n 2 N0 (we interpret 1 + p0 as UF = O⇥ ). Let n be
the smallest integer with this property. The fractional ideal pn is called the conductor of .
Note that is unramified if and only if the conductor of is the unit ideal. Therefore, the
conductor measures the extent to which is ramified.
Now, we want to associate to each quasicharacter of the multiplicative group F ⇥ a local
L-factor. For example, each factor in the product
Y
⇣(s) = (1 p s ) 1
p2P
s,n : C ! C⇥
r ei✓ 7! rs ein✓
166
(iii) Finally, if F = R, then UF = {±1} and thus = ˜| · |s , where ˜ and s are uniquely
determined. Let
sgn : R⇥ ! {±1}
x
x 7!
|x|
where ⇣s⌘
s
R (s) ··= ⇡ 2 .
2
Definition 5.2.12. If 2 X(F ⇥ ), we define the twisted dual or shifted dual of as the
character defined by
_ ··= 1
| · |.
L(( | · |s )_ ) = L(1 s, 1
).
Now, let dx be the self-dual Haar measure on the additive group F . Let d⇥ x be a Haar measure
for the multiplicative group F ⇥ . By Theorem 5.2.1, and the uniqueness of Haar measure,
dx
d⇥ x = c
|x|
for some c 2 R⇥
+ . If F is archimedean we always normalize to c = 1. In the nonarchimedean
case, we let
q
c= .
q 1
Definition 5.2.13. Let d⇥ x be the Haar measure on F ⇥ as described above. Given f 2 S (F )
and 2 X(F ⇥ ), we define the local zeta function as
Z
Z(f, ) ··= f (x) (x)d⇥ x.
F⇥
Theorem 5.2.4. Let f 2 S (F ) and let = ˜| · |s , with ˜ unitary of exponent = Res. Then,
Z(f, ) is absolutely convergent if > 0.
167
Proof. Since = ˜| · |s and ˜ is unitary, it suffices to show that
Z
I(f, ) ··= |f (x)||x| 1 dx < 1.
F⇥
(i) If F is archimedean, then f is a Schwartz function and thus the integral decays rapidly as
x ! 1. Moreover, since > 0, then |x| 1 is integrable around 0, so that I(f, ) < 1.
(ii) Now suppose that F is non archimedean. Then, with the notation of Proposition 5.2.1
and using (5.2.2) we have
Z X
|f (x)||x| 1 dx = N (p m )|f (y)|µ(O)|y| 1 < 1. ⌅
F⇥ y2R
Now, our objective is to show that Z(f, ˜| · |s ) can be continued analytically as a function of s
to the whole complex plane. In order to do so, we need the following lemma:
Lemma 5.2.3. Let f, g 2 S (F ) and 2 X(F ⇥ ) of exponent 2 (0, 1). Then,
Z(f, )Z(ĝ, _
) = Z(fˆ, _
)Z(g, ).
Proof. We have
Z Z ✓Z ◆
_ ⇥
Z(f, )Z(ĝ, )= f (x) (x)d x g(z) (yz) dz (y) 1 |y|d⇥ y.
F⇥ F⇥ F
Now, letting y = xt and using the invariance of the Haar measure, we get
ZZZ
_
Z(f, )Z(ĝ, ) = f (x)g(z) (t 1 ) (txz)|txz|d⇥ xd⇥ yd⇥ z,
(F ⇥ )3
With this we are ready to prove a functional equation that resembles the functional equation
of the Riemann zeta function:
Theorem 5.2.5. Let f 2 S (F ). If = ˜| · |s 2 X(F ⇥ ), with ˜ unitary of exponent 2 (0, 1),
then there is a functional equation of the form
L( )Z(fˆ, _
) = "( )L( _
)Z(f, )
for some "( ) independent of f that lies in C⇥ for all s and is in fact meromorphic as a function
of s.
168
(i) First, if F = R, recall that dx is the Lebesgue measure, (x) = e2⇡ix is the standard
dx
additive character and d⇥ x = . Now, is of the form
|x|
(x) = |x|s
or
(x) = sgn(x)|x|s
for all x 2 R⇥ . In the first case, if (x) = |x|s , we let f (x) = e ⇡x2
= fˆ(x). Then,
Z
Z(f, ) = f (x)|x|s d⇥ x
Z1
2
= e ⇡x |x|s 1 dx
Z1
1
2
=2 e ⇡x xs 1 dx
0
s ⇣s⌘
=⇡ 2
2
= R (s).
=L( ).
(ii) Now suppose that F = C. Recall that |z| = zz is the square of the ordinary absolute
value, dz = 2 dx dy is twice the Lebesque measure, (z) = e 2⇡i(z+z) is the standard
dz
additive character and d⇥ z = . First, we choose the following family of functions: for
|z|
each n 2 Z, let fn : C ! C be the function defined by
8
> n
<⇠ e 2⇡|⇠| , if n > 0;
fn (⇠) =
>
:⇠ n e 2⇡|⇠| , otherwise.
169
We claim that
fˆn (⇠) = i|n| f n (⇠).
First we prove by induction this identity for n > 0: For n = 0, we already proved this in
Example 5.2.1. Now, suppose that
Z
fn (z) e 4⇡iRe(z⇠) dz = in f n (⇠).
Using cartesian coordinates, i.e., letting z = x+iy and ⇠ = ⇠1 +i⇠2 , this equation becomes
Z1 Z1
2 2 2 2
2 (x iy)n e 2⇡(x +y ) 4⇡i(⇠1 x ⇠2 y) dx dy = in (⇠1 + i⇠2 )n e 2⇡(⇠1 +⇠2 ) . (5.2.3)
1 1
Now, if n < 0, then using what we just proved and the fact that dz is the self-dual
measure relative to , we have
ˆ
fˆn (⇠) = in fˆ n (⇠) = in f n( ⇠) = i|n| f n (⇠).
s,n : C ! C⇥
r ei✓ 7! r2s ein✓ ,
where s 2 C and n 2 Z are uniquely determined. For each s,n 2 X(C⇥ ) we consider the
function fn defined above. In polar coordinates,
2⇡r2 in✓
fn (z) = r|n| e
and
2rdrd✓
d⇥ z = .
r2
Therefore,
Z
Z(fn , s,n ) = fn (z) s,n (z)d⇥ z
Z 2⇡ Z 1
2
= r2(s 1)+|n| e 2⇡r 2r drd✓
0
Z 10
|n|
=2⇡ t(s 1)+ 2 e 2⇡t dt
0
|n|
⇣ |n| ⌘
(1 s)
=(2⇡) 2 s+
2
=L( s,n ).
170
This implies that
|n|
⇣ |n| ⌘
Z(fˆn , _
s,n )
|n|
= Z(i f n, 1 s, n )
|n|
= i (2⇡) s 2 1 s+ = i|n| L( _
s,n ),
2
and thus _
|n|
L( s,n )
↵( s,n ) = i ,
L( s,n )
(iii) Finally, let us treat the case where F is non archimedean. In this case, first we recall some
notation: |x| = [O : xO] 1 is the normalized absolute value, is the standard additive
character of F with conductor pm , ⇡ is the uniformizing parameter of F and satisfies
|⇡| = q 1 , where q = N p = [O : p]. The self-dual measure dx is the one that satisfies
Z
p
µ(O) = 1O dx = q m ,
Moreover, the Haar measure d⇥ x on the multiplicative group with respect to the self-dual
measure dx is defined by
q dx
d⇥ x =
q 1 |x|
Let n = ˜n | · |s be a quasicharacter of F ⇥ with conductor pn , where ˜n 2 Fc⇥ . Let
fn : F ! C be the function defined by
(
(x), if x 2 pm n ;
fn (x) =
0, otherwise.
Now let us treat two cases: first, if n = 0, then w.l.o.g we may assume that ˜0 = 1. In
this case, f0 = 1pm , so that
Z
Z(f0 , 0 ) = |x|s d⇥ x.
pm \{0}
171
then
1 Z
X
q
Z(f0 , 0) = |x|s 1
dx
q 1 l=m ⇡l O⇥
1
X
q
= q l(s 1)
µ(⇡ l O⇥ )
q 1 l=m
1
X
q l(s 1)
= q |⇡ l |(µ(O) µ(p))
q 1 l=m
1
X
q ls m m
1
= q (q 2 q2 )
q 1 l=m
1
X
m
ls
=q 2 q
l=m
1
q m( 2 s)
=
1 q s
1
=q m( 2 s)
L( 0 ).
Moreover, by (5.2.4),
m
fˆ0 (y) = q 2 1O (y),
so that Z
m 1
Z(fˆ0 , | · | 1 s
)=q 2 |x|1 s d⇥ x = = L( _
0 ).
O 1 qs 1
Z X1 Z
s ⇥
Z(fn , n ) = (x) ˜n (x)|x| d x = q ls
(x) ˜n (x)d⇥ x. (5.2.5)
pm n
l=m n ⇡l O⇥
The last integral is an example of a so-called Gauss sum, i.e., given an additive character
: O ! T and a character : O⇥ ! T, we define the Gauss sum by
Z
g( , ) = (x) (x)d⇥ x.
O⇥
Note the similarity between the Gamma function and a Gauss sum: For Re(s) > 0, the
Gamma function Z1
(s) = e t ts 1 dt
0
involves an additive character e R ! C , a multiplicative character ts : R⇥ ! C⇥ and
t: ⇥
a Haar measure d⇥ t = dtt on the multiplicative group R⇥ . Let us evaluate the Gauss sum
in (5.2.5) for l > m n: First, if l > m, then ⇡ l O⇥ ✓ pm , so that (x) = 1 for x 2 ⇡ l O⇥ .
Then Z Z Z
⇥ l ⇥
˜n (x)d x = l
˜n (x⇡ )d x = ˜n (⇡ ) ˜n (x)d⇥ x = 0,
⇡l O⇥ O⇥ O⇥
where the last equality follow from the fact that ˜n is a nontrivial character of O⇥ . For
m > l > m n, [
⇡ l O⇥ = · (1 + pm l )⇡ l ,
2R
172
⇥
where R is a system of representatives for O 1 + pm l in O⇥ . On (1 + pm l )⇡ l , we have
(x) = ( ⇡ l ),
and thus Z Z
⇥ l ⇥
(x) ˜n (x)d = ( ⇡ ) n (x)d x.
(1+pm l )⇡ l (1+pm l )⇡ l
Moreover,
Z Z Z
⇥ ⇥
n (x)d x= n (x
l
⇡ )d x = ˜n ( ⇡ ) l
˜n (x)d⇥ x = 0,
(1+pm l )⇡ l 1+pm l 1+pm l
where the last equality follow from the fact that ˜n is a nontrivial character of 1 + pm l
It remains to compute Z
(x) ˜n (x)d⇥ x.
⇡m n O⇥
⇥
To this end, let R be a system of representatives of O pn 1 in O⇥ . Then,
[
⇡ m n O⇥ = · ⇡ m n (pn 1),
2R
and Z X Z
⇥
(x) ˜n (x)d x = ˜n ( ⇡ m n
) ( ⇡ m n
) d⇥ x.
⇡m n O⇥ 2R pn 1
Moreover,
Z
m
Z(fˆn , _
n) = qn 2 1pn 1 (x) ˜n 1 (x)|x|1 s d⇥ x
Z
m
=q n 2 ˜n 1 (x)|x|1 s d⇥ x
pn 1
Z
m
=q n 2 ˜n 1 (x)|x|1 s d⇥ x
n
Z1+p
m
=q n 2 d⇥ x,
1+pn
173
where the last equality follow from the fact that the conductor of n is pn . Now, since
_
n is ramified, then by definition L( n ) = L( n ) = 1. This shows that
Z
n m
q 2 d⇥ x
1+pn
↵( n ) = X Z
q (n m)s
˜n ( ⇡ m n
) ( ⇡ m n
) d⇥ x
2R pn 1
1
_
q (m n)(s 2 ) L( n)
= ,
W ( ˜n ) L( n)
( 1)L( )Z(fˆ, _
) = "( )"( _
)L( )Z(fˆ, _
).
Hence,
( 1)
"( ) = . (5.2.7)
"( _ )
Similarly, we have
"( ) = ( 1)"( ). (5.2.8)
Now, suppose that is a quasicharacter of the form
1
= ˜| · | 2
Then,
_
(x) (x) = | (x)|2 = |x| = (x) (x),
so that
_
(x) = (x). (5.2.9)
Hence, combining (5.2.7), (5.2.8) and (5.2.9) we have
_
( 1) ( 1) 1
"( ) = = = .
"( _) _( _
1)"( ) "( )
|W ( ˜)| = 1. (5.2.10)
174
5.3 The Functional Equation
K : a global field,
⌫ : a place on K,
K⌫ : the completion of K with respect to ⌫ (which is a local field ),
O⌫ : the ring of integers of K⌫ (set O⌫ = K⌫ if ⌫ is archimedean),
p⌫ : the maximal ideal of O⌫ ,
k⌫ : the finite residue field of O⌫ p⌫ with q⌫ elements ,
| · |⌫ : the normalized absolute value on K⌫ ,
dx⌫ : the self-dual measure on K⌫ ,
Y0
AK = (K⌫ , O⌫ ),
Y0
IK = (K⌫⇥ , O⌫⇥ ).
The objective of this section is to provide an analogue formula for the Poisson summation
formula: If f : R ! C is a Schwarz function, then
X X
f (n) = fˆ(n).
n2Z n2Z
Definition 5.3.1. We define the space of adèlic Schwartz-Bruhat functions as the set
O0
S (AK ) = S (K⌫ ) ··= {f = ⌦f⌫ | f⌫ 2 S (K⌫ ) for all ⌫ and f⌫ O⌫ = 1 for all but finitely many ⌫}.
(i) (K is an algebraic number field) First suppose that K = Q. For each K⌫ , let ⌫ be the
associated standard character defined in Example 5.2.1. Since ⌫ O = 1 for all but
K⌫
finitely many ⌫, then
K ··= ⌦ ⌫
175
and ap = 0 for all but finitely many p. Then, by Ostrowski’s theorem,
Y 2⇡iap
2⇡ia pn p 2⇡ib
K (a) =e e =e = 1.
p2P
If K is a finite extension of Q, then for x = (x⌫ ) 2 AK we define the adèlic trace map
tr : AK ! AQ as X
tr(x) = trK⌫ :Qp (x⌫ ),
⌫|p
where p ranges over all places of Q and ⌫|p means that | · |⌫ is an extension of | · |p (i.e.,
|x|⌫ = |x|p for all x 2 Qp ). Then, we define
K (x) = Q (tr(x))
for all x 2 AQ . From this it follows that for every ↵ 2 K we have K (↵) = 1.
(ii) (K is a function field) Similarly as before, first suppose that K = Fpm (t). Since we are in
positive characteristic, then Kv ⇠= Fq ((t)) by Theorem 5.1.5. For each K⌫ , let ⌫ be the
associated standard character defined in Example 5.2.1 and let
K ··= ⌦ ⌫.
Then, for a 2 K,
a = a0 + a1 t + · · · + an t n with ai 2 Fpm ,
so that
trK⌫ :K⌫,0 (a) = b0 + b1 t + · · · + bn tn with bi 2 Fpm ,
whence Y Y Y 2⇡ib 1
where tr is the adèlic trace map from K to Fpm (t). This shows that K (↵) = 1 for all
↵ 2 K.
176
Definition 5.3.3. Given f 2 L1 (AK K ), let fˆ: K ! C be the function defined by
Z
ˆ 1
f () = f (x) (x) dx,
vol(D) D
Proof. Let C be a compact subset of AK . By enlarging C we may assume that C is of the form
Y Y Y
C= C⌫ ⇥ pn⌫ ⌫ ⇥ O⌫ ,
⌫2V1 (K) ⌫2S ⌫ 2S
/
We may also enlarge S so that V1 (K) ✓ S and by Proposition 5.2.1 we may assume that
f ⌫ = 1pm
⌫
⌫ for all ⌫ 2 S \ V1 (K). Let r⌫ ··= min{n⌫ , m⌫ } and let I be the ideal
Y
I ··= pr⌫⌫ .
⌫2S\V1 (K)
where O
f1 = f⌫ and x1 = (x⌫ )⌫2V1 (K) .
⌫2V1 (K)
Q
Since K is discrete on AK , then I is discrete on ⌫2V1 (K) K⌫ . Let us consider two cases:
is a finite sum.
If K is an algebraic number field, then for all ⌫ 2 V1 (K), K⌫ is archimedean, so that
each f⌫ is a Schwartz function on R or C. Therefore, f1 is a rapidly decay function
with
Q the further property that f1 has a uniform absolute bound over the compact set
⌫2V1 (K) C⌫ . This shows that
X
|f1 (x1 + )|
2I
converges uniformly on K.
177
In any case, X
f (x + )
2K
P
converges absolutely and uniformly on C. Finally, translating the function 2K f (x + ) by
an element of K just permutes the terms. ⌅
With this Lemma we are ready to prove Poisson summation formula for the case of a global
field:
Proof. We follow the same ideas as in the proof of Theorem 1.5.1. Let F : AK K ! C be the
function defined by X
F (x) = f (x + ).
2K
For 2 K, we have
Z X
F̂ () = f (x + ) (x) dx
D 2K
XZ
= f (x + ) (x) dx
2K D
XZ
= f (z) ((z )) dz
2K D+
Z
= f (z) (z) dz
AK
=fˆ().
By the previous lemma, we may apply the Fourier Inversion Formula to get
1 X
F (x) = F̂ () (x),
vol(D) 2K
which is equivalent to
X 1 X ˆ
f (x + ) = f () (x).
2K
vol(D) 2K
Letting x = 0, we obtain
X 1 X ˆ
f () = f (),
2K
vol(D) 2K
but then applying Fourier transform again, we get
X 1 X
fˆ() = f ().
2K
vol(D) 2K
Together, these prove that vol(D)2 = 1, so that vol(D) = 1. These completes the proof. ⌅
178
Corollary 5.3.1.1. The volume of AK K is 1.
One of the most important results about the normalized valuation on AK is the following:
and this property may be extended to the quotient. Now, multiplication by a is an isomorphism
AK AK
K! K and by our previous observation it scales the volume by |a|, i.e.,
The Poisson summation formula can be used to prove the following stronger result:
Theorem 5.3.3 (Harmonic Version of the Riemann–Roch Theorem). Let x 2 IK and let
f 2 S (AK ). Then,
X 1 X ˆ
f (x) = f (x 1 ).
2K
|x| 2K
Proof. Fix x 2 AK and consider the function g : AK ! C defined by g(y) = f (xy). Clearly
g 2 S (AK ) and moreover
Z
ĝ() = f (x) (y) dy
AK
Z
1
= f (u) (ux 1 ) du
|x| AK
1
= fˆ(x 1 ).
|x|
179
Now we mention without proof the geometric version of the Riemann–Roch theorem. Let K
be a global function field (i.e., a finite extension of Fq (t)). A divisor on K is a formal linear
combination X
D= n⌫ ⌫,
⌫2V (K)
where n⌫ 2 Z and n⌫ = 0 for all but finitely many places P⌫. The divisors of K form an additive
group denoted by Div(K). The degree of a divisor D = n⌫ ⌫ is
X
deg(D) ··= n⌫ [k⌫ : Fq ].
where ord⌫ (f ) is the valuation of f at ⌫ (Example 5.1.5). Now we define a partial ordering on
Div(K) as follows:
X X
D= n⌫ ⌫ > D 0 = n0⌫ ⌫ if n⌫ > n0⌫ 8⌫ 2 V (K).
Definition 5.3.5. Let |IK | denote the group {|x| | x 2 IK }. For t 2 |IK |, let
ItK ··= {x 2 IK | |x| = t}
denote the set of idèles of norm t.
Definition 5.3.6. A Hecke character is a continuous homomorphism : IK ! C⇥ such that
|K ⇥ = 1.
Example 5.3.2. If s 2 C, then | · |s is a Hecke character.
Definition 5.3.7. For f 2 S (AK ) and a Hecke character, we define the global zeta function
by the equation Z
Z(f, ) = f (x) (x)d⇥ x,
IK
⇥
Q ⇥
where d x is the product measure d x⌫ and
8
>
> dx⌫
< , if K⌫ is archimedean;
|x⌫ |⌫
d ⇥ x⌫ = q⌫ dx⌫ .
>
> , otherwise.
:
q⌫ 1 |x⌫ |⌫
180
Similar to the local case, |IK | = R⇥
+ if K is an algebraic number field and we endow this group
dt
with the measure . If K is a global function field, then |IK | = q Z for some q > 1 and we
t
endow this group with the counting measure multiplied by log q and by convenience we will
dt
also denote this measure by .
t
Theorem 5.2.4 can be generalized to show that Z(f, ) is normally convergent in = Re(s) > 1,
where = ˜| · |s with ˜ unitary. As in the local case, define _ ··= 1
| · |. The objective of
this section is to generalize the analytic continuation of ⇣(s) and the functional equation for
the case of a global field.
Fix f 2 S (AK ) and = ˜| · |s a Hecke character. The idea is to “slice” IK in the following
way: For = Re(s) > 1, we have
Z
dt
Z(f, ) = Zt (f, ) , (5.3.1)
|IK | t
where Z
Zt (f, ) ··= f (x) (x)d⇥ x
ItK
Proof. We have
Z X Z ⇣ X ⌘
⇥
Zt (f, ) = f (ax) (ax)d x = f (ax) (x)d⇥ x,
ItK ItK
K ⇥ a2K ⇥ K⇥ a2K ⇥
where the last equality follows from the fact that is trivial on K ⇥ . Now, the idea is to use
the Riemann–Roch theorem, therefore we want to change the sum over K ⇥ to a sum over K,
so we add an a = 0 term to both sides:
Z Z ⇣X ⌘
⇥
Zt (f, ) + f (0) t (x)d x = t f (ax) (x)d⇥ x
IK IK
K⇥ K⇥ a2K
Z ⇣ 1 X ⌘
= t ˆ
f (ax ) (x)d⇥ x
1
IK ⇥ |x|
Z K ⇣ a2K X ⌘
= t 1 |y| fˆ(ay) (y 1 )d⇥ y
IK
K⇥ a2K
Z ⇣X ⌘
= 1 fˆ(ay) _
(y)d⇥ y
ItK
K⇥ a2K
Z
= Zt 1 (fˆ, _
) + fˆ(0) 1
_
(x)d⇥ x,
ItK
K⇥
where the last equality follows by the same argument in reverse order. ⌅
181
It is possible to show that the quotient IK K ⇥ is compact (see for example [28]). Therefore,
1
ters, we have Z
(x)d⇥ x = 0. ⌅
t IK
K⇥
With these lemmas we are ready to generalize Theorem 3.2.1 for the case of global zeta functions:
Theorem 5.3.5 (The Global Functional Equation). The function Z(f, ) extends to a mero-
morphic function on C with the possible exception of two simple poles at s = 0 and s = 1 with
residues V f (0) and V fˆ(0) respectively. Moreover, Z(f, ) satisfies the functional equation
Z(f, ) = Z(fˆ, _
).
Proof. First suppose that K is an algebraic number field. Following the same ideas used on
Section 3.2, we divide the range of integration in half:
Z1 Z1
dt dt
Z(f, ) = Zt (f, ) + Zt (f, ) .
0 t 1 t
If = | · |s for some s 2 C, combining the previous lemmas and then letting u = t 1 , we have
Z1 Z1 Z1 ⇣ ⇣ 1 ⌘1 s ⌘ dt
dt ˆ _ dt
Zt (f, ) = Zt 1 (f , ) + V fˆ(0) V f (0)ts
0 t 0 t 0 t t
Z1
du V fˆ(0) V f (0)
= Zu (fˆ, _ ) + .
1 u s 1 s
Therefore Z1 Z1
dt dt V fˆ(0) V f (0)
Z(f, ) = Zt (f, ) + Zt (fˆ, _
) + . (5.3.2)
1 t 1 t s 1 s
Since f 2 S (AK ), then both integrals converge to an entire function. This shows that Z(f, )
has simple poles at s = 0 and s = 1 with residues V f (0) and V fˆ(0) respectively. Moreover,
the right hand side of (5.3.2) is invariant under the transformation (f, ) 7! (fˆ, _ ). In the
case that 6= | · |s for some s 2 C, (5.3.2) becomes
Z1 Z1
dt dt
Z(f, ) = Zt (f, ) + Zt (fˆ, _ ) ,
1 t 1 t
182
Now suppose that K is a global function field. Then,
Z
dt
Z(f, ) = Zt (f, )
qZ t
X
= log q Zn (f, )
n2q Z
1
X 1
X
= log qZ1 (f, ) + log q Z (f, ) + log q
qn Zq n (f, ).
n=1 n=1
Z1 (f, ) = Z1 (fˆ, _
) + fˆ(0)V f (0)V,
so that
1 V fˆ(0) V f (0)
Z1 (f, ) = (Z1 (f, ) + Z1 (fˆ, _ )) + . (5.3.3)
2 2 2
Now, again combining the previous lemmas we have
1
X 1
X 1
X 1
X
Zq n (f, ) = Zqn (f, ) + V fˆ(0) (q n s 1
) V f (0) (q n s
). (5.3.4)
n=1 n=1 n=1 n=1
If we let (
1
, if n = 0
n ··= 2
,
1, otherwise.
then (5.3.3) together with (5.3.4) show that
1
X h fˆ(0)q 1 s
f (0)q s
fˆ(0) f (0) i
Z(f, ) = log q ˆ
n [Zq n (f, ) + Zq n (f ,
_
)] + V log q + ,
n=0
1 q1 s 1 q s 2
or equivalently
1
X h 1 + q1 s
1 + qs i
Z(f, ) = log q ˆ
n [Zq n (f, ) + Zq n (f ,
_
)] + V log q fˆ(0) + f (0) . (5.3.5)
n=0
1 q1 s 1 qs
This shows that Z(f, ) extends to a meromorphic function with two simple poles at s = 0 and
s = 1 with residues V f (0) and V fˆ(0) respectively (note the importance of multiplying the
counting measure by log q). Moreover, the right hand side of (5.3.5) is unchanged under the
transformation (f, ) 7! (fˆ, _ ). In case that 6= | · |s for some s 2 C, (5.3.5) becomes
1
X
Z(f, ) = log q n [Zq n (f, ) + Zqn (fˆ, _
)],
n=0
183
5.3.3 Hecke L -Functions
Let be a Hecke character. For each place ⌫ 2 V (K) we define a local character as follows:
K⌫⇥ ! C⇥ ⌫:
t 7! (1, . . . , 1, t, 1, . . . , 1),
Q
where t is located at the ⌫ th component of . Then (y) = ⌫ (y). The global L function
of is defined as Y
L( ) ··= L( ⌫ ),
⌫
Lemma 5.3.4. Let be a Hecke character with exponent . Then, L( ) is absolutely convergent
whenever > 1.
Proof. Let = ˜|·|s . We may ignore the set of finite places where ⌫ is ramified since ⌫ O⌫ =1
for all but finitely many ⌫. Therefore
Y Y 1
|L( ⌫ )| = .
⌫ ⌫
1 ˜⌫ (⇡⌫ )q⌫ s
where the last equality follows from the fact that for |z| < 1, we have
X1
1 zm
log + i arg(z 1) = log(1 z) = .
1 z m=1
m
where ⌫|p means that | · |⌫ is an extension of | · |p . It is possible to show (see [28]) that if
n = [K : Q], then X
[K⌫ : Qp ] = n.
⌫|p
Therefore,
#{⌫ | ⌫|p} 6 n
184
and since q⌫ = pr for some r when ⌫|p, then
XXX 1 XX 1
q⌫ m p m
6n
p2P m=1
m p2P m=1
m
⌫|p
Y 1
=n log
p2P
1 p
X1
1
=n log
m=1
m
<1. ⌅
Before proving our next theorem we need to mention some facts about the discriminant of a
global field and the di↵erent of a local field. Let F be a nonarchimedean local field (i.e. a finite
extension of F0 = Qp or F0 = Fp ((t))) with standard character of conductor pm . The inverse
di↵erent of F is the set
DF 1 ··= {x 2 F | trF :F0 (xOF ) ✓ OF0 }.
The di↵erent of a local field F is the inverse of DF 1 . Therefore, by the construction of
(Example 5.2.1),
DF 1 = pm ,
so that p 1
vol(OF , dx) = N pm = (N DF ) 2 .
Let L be a field of characteristic 0. Recall that a real embedding of L is a field homomorphism
: L ,! R. A complex embedding of L is a field homomorphism ⌧ : L ,! C such that ⌧ (L) ( R.
Note that if is a real embedding, then = , while if ⌧ is a complex embedding, then ⌧ is
another complex embedding, therefore complex embeddings always come in conjugate pairs.
If K is a number field and n = [K : Q], then it is possible to show that K has n = r1 + 2r2
embeddings : K ,! C, r1 real embeddings and r2 pairs of complex embeddings; for a proof
see for example [25].
Example 5.3.3. Let K = Q[x] x3 2. There are 3 embeddings, one for each root of x3 2.
Explicitly, p
3 2⇡i p
3 2⇡i p
3
1 (x) = 2, 2 (x) = e 3 2 and 3 (x) = e 3 2.
Hence K has 1 real embedding and 1 pair of complex embeddings, so that [K : Q] = 1 + 2 = 3.
and we denote it by dK .
185
p
Example 5.3.4. Let K = Q( m) be a quadratic field where m is square free and m 6= 1. By
Theorem 5.1.10 a Z basis for OK is
n 1 + pm o
1, if m ⌘ 1 (mod 4)
2
and p
{1,
m} if m ⌘ 2, 3 (mod 4).
p p p p
The embeddings of K are 1 (a + b m) = a + b m and 2 (a + b m) = a b m. Therefore,
if m ⌘ 1 (mod 4), then
" p #2 p p
1 1+ m ⇣1 m 1 + m ⌘2
dK = det 2p = = m.
1 m
1 2
2 2
It is possible to show that if K is a number field, then (see for example [21])
Y
|dK | = N DK⌫ .
⌫2Vf (K)
Therefore, Y Y 1 1
vol(O⌫ , dx⌫ ) = (N D⌫ ) 2 = |dK | 2 .
⌫2Vf (K) ⌫2Vf (K)
Proof. During the proof of Theorem 5.2.5 we showed that for each place ⌫ there is some function
f⌫ 2 S (K⌫ ) such that
(ii) If K⌫ is non archimedean, ⌫ is the standard character with conductor pm⌫ and ⌫ = ˜⌫ |·|s
has conductor pn⌫ , then
( 1
q m⌫ ( 2 s) L( ⌫ ), if n⌫ = 0;
Z(f⌫ , ⌫ ) = (n⌫ m⌫ )s+ n2⌫ n⌫ ⇥
q W ( ˜⌫ ) vol(p 1, d x⌫ )L( ⌫ ), otherwise,
186
In any case,
Z(f⌫ , ⌫)
= h⌫ ( ⌫ )L( ⌫ ),
where h⌫ is a nonzero entire function. If we let f ··= ⌦f⌫ , then
Z(f, ) = h( )L( ),
Q
where h( ) = ⌫ h⌫ ( ⌫ ) is a nonzero entire function. By Theorem 5.3.5, Z(f, ) is a meromor-
phic function except possibly at s = 0 and s = 1. This shows that L( ) can be extended to a
meromorphic function on all C except possibly at s = 0 and s = 1.
Note that if = | · |s , then n⌫ = 0, so that
1 1
h⌫ ( ⌫) = q m⌫ ( 2 s)
= (N DK⌫ )s 2
and therefore
1
h( ) = |dK |s 2
by our previous observations. Note that f⌫ (0) = 1 for all places ⌫ 2 V (K). Therefore f (0) = 1
and since Z(f, ) has residue f (0)V when = | · |0 , then
sZ(f, ) p
Res L( ) = lim s
= |dK |V.
s=0 s!0 h(| · | )
Now, recall that Z(f, ) has residue fˆ(0)V when = | · |1 . In this case, since n⌫ = 0 for all ⌫,
then f⌫ (x) = 1pm⌫ (x) (pm⌫ is precisely the conductor of ⌫ ). Therefore,
(
1, if K⌫ is archimedean;
fˆ⌫ (0) =
vol(O⌫ , dx⌫ ), otherwise.
1
This shows that fˆ(0) = |dK | 2 and therefore
(s 1)Z(f, ) 1
Res L( ) = lim = |dK | 2 fˆ(0)V = V.
s=1 s!1 h(| · |s )
Moreover, combining the local functional equation and the global functional equation (Theorem
5.2.5 and 5.3.5 respectively), we have
Y
L( ) = L( ⌫ )
⌫
Y "( _
⌫ )L( ⌫ )Z(f⌫ , ⌫)
=
⌫ Z(fˆ⌫ , _⌫ )
_
"( )L( )Z(f, )
=
Z(fˆ, _ )
="( )L( _ ). ⌅
Definition 5.3.8. Let be a Hecke character. For s 2 C we define the Hecke L function
L(s, ) as
L(s, ) = L( | · |s ).
Similarly, we define
"(s, ) ··= "( | · |s ).
It is also convenient to define the finite and infinite versions of the Hecke L functions:
Y
L(s, f ) = L(s, ⌫ )
⌫2Vf (K)
Y
L(s, 1) = L(s, ⌫ ).
⌫2V1 (K)
187
Therefore, the functional equation for the Hecke L functions is
1
L(s, ) = "(s, )L(1 s, ). (5.3.6)
and L(s, 1f ) is called the Dedekind zeta function of K and we denote it by ⇣K (s). Following
the same ideas as in the proof of the Euler’s product formula it is possible to show that
X 1
⇣K (s) = ,
a6=0
(N a)s
Since the volume of IK K ⇥ appears in the functional equation of the Hecke L-functions we
1
mention how to compute it. Since the proof involves some facts about algebraic number theory
we will just mention some of this results and then state the theorem without proof. For a
complete introduction on these topics see [21]. We begin with the following theorem:
⇥
Theorem 5.3.7 (Dirichlet’s Unit Theorem). The group of units OK is a finitely generated
abelian group of rank r = r1 + r2 1.
⇥
Therefore, let {u1 , . . . , ur } be a set of generators for OK . Let
(
1, if j is real
Nj =
2, otherwise.
The group of ideal classes in K is defined as the quotient group JK PK , where JK is the group
of fractional ideals of OK and PK is the subgroup of principal ideals. The order of this group
is called the class number of K and we denote it by hK . The ring OK is principal if and only if
188
p
hK = 1. For example (see [25]) we know that the only imaginary quadratic fields K = Q( m)
that have class number hK = 1 are those with
It is more common for real quadratic fields to have hK = 1. In the range 2 6 m < 24, it occurs
for
m = 2, 3, 5, 6, 7, 11, 13, 14, 17, 19, 21, 22, 23.
Taking all of this into account we can state the following theorem:
Theorem 5.3.8. Let K be a number field with r1 real embeddings, r2 complex embeddings, class
number hK , regulator RK , number of roots of unity !K and discriminant dK . Then
2r1 (2⇡)r2 hK RK
vol IK K ⇥ =
1
p .
!K |dK |
In the function field case, we denote by Div0 (K) the group of divisors of degree 0 in K, i.e.,
formal sums of the form X
D= n⌫ ⌫,
⌫
P
with n⌫ 2 Z and deg D = ⌫ n⌫ deg ⌫ = 0. The Picard group of degree zero is the quotient
group
0
Pic0 (K) ··= Div (K) K ⇥ .
Then we have the following theorem:
189
5.4 What’s Next?
In this section we will study more general L functions. We begin by stating the generalized
Riemann hypothesis:
Conjecture (The Generalized Riemann Hypothesis). Let K be a global field. Then, every
zero of the Dedekind zeta function ⇣K (s) that lies on the strip 0 6 Re(s) 6 1 is on the line
1
Re(s) = .
2
We start with the function field case. During the 1940s André Weil proved the following beatiful
theorem:
Theorem 5.4.1 (Weil). Let K be a function field (i.e., a finite extension of Fq (t)). Then,
the Dedekind zeta function ⇣K (s) is a rational function of q s . More precisely, there exists a
polynomial of degree 2g, where g is the genus of K, with P (0) = 1 such that
P (q s )
⇣K (s) = .
(1 q s )(1 q 1 s )
Moreover, P is a polynomial with integer coefficients and admits a factorization over C of the
form
2g
Y
s
P (q ) = (1 ↵j q s ),
j=1
1
where each ↵j is an algebraic integer with |↵j | = q 2 . Therefore, every zero of ⇣K (s) lies on the
line Re(s) = 21 . In other words, the Riemann hypothesis is true on K!
Since the Riemann hypothesis has been proved for global fields of positive characteristic now
we turn our studies to the number field case. We begin by fixing some notation:
K: A number field
n: [K : Q]
r1 : The number of real embeddings of K
r2 : The number of complex embeddings of K
dK : The discriminant of K
RK : The regulator of K
hK : The class number of K
!K : The number of roots of unity of K
What we denote by L(s, ) is classically denoted by ⇤(s, ). The notation L(s, ) classically
denotes what we called L(s, f ). Thus, from now on we will use the classical notation, that is,
given a Hecke character , we have
Y
L(s, ) ··= L( ⌫ | · |s ),
⌫2Vf (K)
and Y
⇤(s, ) ··= L( ⌫| · |s ).
⌫2V (K)
190
Therefore
r1 r2
⇤(s, 1) = R (s) C (s) ⇣K (s),
where
s s
1 s
R (s) =⇡ and2
C (s) = (2⇡) (s).
2
Let = 1. Since the conductor of ⌫ is pn⌫ with n⌫ = 0, then the epsilon-factor is given by
Y
"(s, 1) = "(| · |s⌫ ),
⌫2V (K)
where (
1, if K⌫ is archimedean;
"(| · |s⌫ ) = 1
q m⌫ (s 2
)
, otherwise.
But q m⌫ = (N DK⌫ ) 1 , so that
Y 1 1
s s
"(s, 1) = (N DK⌫ ) 2 = |dK | 2 .
⌫2Vf (K)
Hence, with the aid of Theorem 5.3.6 and Theorem 5.3.8 we can summarize what we have done
in the following theorem:
2r1 (2⇡)r2 hK RK
V = p .
!K |dK |
1 1
Note that R (1) = ⇡ 2
2
= 1 and C (1) = (2⇡)1 1
(1) = 1, whence by the previous
theorem,
2r1 (2⇡)r2 hK RK
Res ⇣K (s) = p .
s=1 !K |dK |
This equation is known as the class number formula.
21 · (2⇡)0 · 1 · 1
Res ⇣(s) = p = 1.
s=1 2 1
191
Example 5.4.2. If K = Q(i) are the Gaussian rationals, then hK = 1 (The Gaussian integers
Z(i) are a principal ideal domain). There is one pair of complex embeddings:
20 · (2⇡)1 · 1 · 1 ⇡
Res ⇣Q(i) (s) = p = . (5.4.1)
s=1 4 4 4
p
It is possible to show that if K = Q( m) is a quadratic field with discriminant dK and is
the Dirichlet character modulo dK defined for odd p 2 P by6
⇣d ⌘
K
(p) =
p
then
⇣K (s) = ⇣(s)L( , s). (5.4.2)
Actually, it can be shown that (5.4.2) is equivalent to the quadratic reciprocity law: if p, q are
two distinct odd primes, then ⇣ p ⌘⇣ q ⌘ p 1 q 1
= ( 1) 2 2 .
q p
For a proof of this fact see [18] (quadratic reciprocity implies (5.4.2)) and [2]((5.4.2) implies
quadratic reciprocity).
In the particular case K = Q(i), we know that dK = 4 and thus
so that 1
X ( 1)n
L( , s) = s
.
n=0
(2n + 1)
Threfore, combining (5.4.1) and (5.4.2) we obtain the classical formula
X1
( 1)n Res ⇣Q(i) (s) ⇡
L( , 1) = = s=1 = .
n=0
2n + 1 Res ⇣(s) 4
s=1
6
Let p be an odd prime. Given a 2 Z, the Legendre symbol of a with respect to p is defined by
8
>
>0, if a ⌘ 0 (mod p)
⇣a⌘ <
= 1, if a 6⌘ 0 (mod p) and a ⌘ b2 (mod p) for some b 2 Z
p >
>
: 1, otherwise.
192
Now, let us write explicitly the functional equation of a Hecke L function. In order to do
so, we need to associate a “modulus” to a Hecke character. Let K be a number field and
: IK ! T a Hecke character.Q For eachn⌫finite place ⌫ 2 Vf (K), let U⌫ ··= O⌫⇥ be the group of
units of O⌫ = OK⌫ . Let m = ⌫2V (K) p⌫ be an integral ideal of K, i.e., n⌫ > 0 and n⌫ = 0 for
all ⌫ 2 V1 (K). Now let Y
Im
f
·
· = U⌫(n⌫ ) ,
⌫2Vf (K)
where
U⌫(n⌫ ) = 1 + pn⌫ ⌫ .
We say that m is a module of definition for the Hecke character if
(Im
f ) = 1.
Let
Q us show that every Hecke character has a 7module of definition: Since Q the subgroup
⌫2Vf (K) U⌫ is compact and totally disconnected , then the image of : ⌫2Vf (K) U⌫ ! T
Q
is a compact and totally disconnected subgroup of the circle T and therefore ( ⌫2Vf (K) U⌫ ) is
Q
finite. This shows that ker has finite index in the profinite group ⌫2Vf (K) U⌫ , so that ker
Q
is open and since ⌫2Vf (K) U⌫ is also a topological group, then ker is also closed. This shows
Q (n )
that ker contains a subgroup of the form ⌫2Vf (K) U⌫ ⌫ , where n⌫ = 0 for all but finitely
Q
many ⌫. Finally, it suffices to take the ideal m = ⌫2Vf (K) pn⌫ ⌫ as module of definition. Now let
n⌫
⌫ be the local character of , i.e., ⌫ = |K⌫⇥ and let pQ ⌫ be its conductor. By our previous
observations, n⌫ = 0 for all but finitely many ⌫ and m = ⌫2Vf (K) pn⌫ ⌫ is a module of definition
for .
Definition 5.4.2. Let m be an integral ideal of the number field K, and let J m be the group
of all ideals of K which are relatively prime to m. A Grö encharaktere mod m is a character
: J m ! T for which there exists a pair of characters
f : (O m)⇥ ! T, 1: R⇥ ! T,
such that
((a)) = f (a) 1 (a)
The following theorem is one of the most important facts about Grö encharakteres; for a proof
we refer the reader to [25].
Theorem 5.4.3. There is a one-to-one correspondence between Grö encharakteres mod m and
Hecke characters with module of definition m.
Let be a unitary Hecke character with local character ⌫ and module of definition
Y
m= pn⌫ ⌫ ,
⌫2Vf (K)
where pn⌫ ⌫ is the conductor of ⌫. Let ⌫ be the standard character of K⌫ with conductor pm
⌫ .
⌫
193
where (see Theorem 5.2.5)
8
>
> 1, if K⌫ = R and ⌫| · |s⌫ = | · |s⌫ ;
>
>
>
<i, if K⌫ = R and s s⌫
⌫ | · |⌫ = sgn(·)| · | ;
"( ⌫ | · |s⌫ ) =
>
> i|N⌫ | , if K⌫ = C and i✓ i✓ s
⌫ (r e )|r e |⌫ = r
2s⌫ iN⌫ ✓
e ;
>
> 1)
>
: q⌫
(m⌫ n⌫ )(s⌫ 2
W( ⌫)
, otherwise
with n⌫ X
W( ⌫) = q⌫ 2
⌫( ⇡⌫m⌫ n⌫
) ⌫( ⇡⌫m⌫ n⌫
)
2R
⇥
and R is a system of representatives for O⌫ pn⌫ ⇥
1 in O⌫ . For ⌫ archimedean, let
⌫
1
W( ⌫) ··= .
"( ⌫ | · |s⌫ )
By (5.2.10), we have
|W ( )| = 1. (5.4.3)
Q
Since q m⌫ = (N DK⌫ ) 1
and N m = n⌫
⌫2Vf (K) q⌫ , then
1 1
s s
|dK | 2 (N m) 2
"(s, ) = .
W( )
194
5.4.1 More General L-functions
In this section we follow [16]. We will introduce L-functions using some abstract definitions
and then give specific examples concerning this L-functions.
We denote L functions by L(f, s), although the symbol f doesn’t carry any specific mean-
ing until we give some concrete examples. The reason for this notation is that usually this
L functions arise from some arithmetic object.
Definition 5.4.3. We say that L(f, s) is an L function of the complex variable s provided we
have the following data and conditions:
(i) L(f, s) is given by a Dirichlet series with Euler product of degree d > 1,
1
X Y
af (n)
L(f, s) = = (1 ↵1 (p)p s ) 1
· · · (1 ↵d (p)p s ) 1 ,
n=1
ns p2P
where af (1) = 1, af (n) 2 C, ↵j (p) 2 C. The series and Euler products must be absolutely
convergent in the half-plane Re(s) > 1. The ↵j (p), 1 6 j 6 d, are called the local
parameters of L(f, s) at p, and satisfy
where the numbers j are called the local parameters of L(f, s) at infinity. We assume
that these numbers are either real or come in conjugate pairs. Moreover, we also assume
that Re( j ) > 1. Note that this condition implies that (f, s) has no zero in C and no
pole on the half-plane Re(s) > 1
(iii) An integer q(f ) > 1, called the conductor of L(f, s), such that ↵j (p) 6= 0 for p - q(f ) and
1 6 j 6 d. We say that a prime p is unramified if p - q(f ).
By definition, ⇤(f, s) is analytic on Re(s) > 1, yet it must admit analytic continuation to a
meromorphic function for s 2 C of order at most 1, with at most poles at s = 0 and s = 1.
Moreover, there’s a functional equation
and "(f ) is a complex number of absolute value 1, called the root number of L(f, s).
195
We let r(f ) denote the order of the pole (if positive) or zero (if negative) of ⇤(f, s) at s = 0
and s = 1 (equal because of 5.4.5). Since (f, 1) 6= 0, 1, then r(f ) is also the order of the pole
or zero of L(f, s) at s = 1.
The analytic conductor of L(f, s) is
d
Y
q(f, s) = q(f )q1 (s) = q(f ) (|s + j | + 3). N
j=1
Example 5.4.3. Let K be a number field of degree d = [K : Q]. Then, the Dedekind zeta
funcion X 1
⇣K (s) =
(N a)s
(0)6=a✓OK
p
YY
= (1 (N p) s ) 1
p2P p|p
YY
rp s 1
= (1 p )
p2P p|p
d
YY
= (1 ↵j (p)p s ) 1 ,
p2P j=1
2⇡ij
where the last equality follows from the fact that if ⇠j = e rp
, then
rp
Y
rp ⇡i(r 1)
1 x =e (⇠ 1 x 1)(⇠ 2 x 1) · · · (⇠1 r x 1)(1 x) = (1 ⌘j,p (p)p s ) 1 .
j=1
8
Let K be a number field and let p 2 P. Then (p)OK may not be a prime ideal on OK but nevertheless
where pj are prime ideals on OK . We say that p ramifies on K and denote it by p|p if rj >1 for some j. It is
possible to show that there’s a correspondence between prime ideals p|p and places such that ⌫|p.
196
(ii) By the Legendre duplication formula (2.1.7) the gamma factor is
r2 ds
⇣ s ⌘r1 +r2 ⇣ s + 1 ⌘r2
(s) = ⇡ 2 R (s)r1 C (s)r2 = ⇡ 2 ,
2 2
where r1 is the number of real embeddings of K and r2 the number of pairs of complex
embeddings so that d = r1 + 2r2 .
(iii) The conductor is the absolute value of the discriminant of K, D ··= |dK | and ↵j (p) 6= 0
for unramified p follows from the fact that a prime p ramifies on K if and only if p|dK
(see for example [25]).
XXX 1
q⌫ m
=
p2P m=1
m
⌫|p
Y 1
6n log .
p2P
1 p
197
Example 5.4.4. Let K be a number field and let be a Hecke character with module of
definition m. Then, following the same ideas of the previous example, we have:
X1
a (n)
L( , s) = ,
n=1
ns
P
where a (n) = N a=n (a) and
d
YY
L( , s) = (1 ↵j (p)p s ) 1
p2P j=1
(ii) Let ⌦ denote the set of real infinite places where is ramified. By the Legendre duplica-
tion formula (2.1.7) the gamma factor is
ds
⇣ s ⌘r1 +r2 |⌦| ⇣ s + 1 ⌘r2 +|⌦|
( , s) = ⇡ 2 .
2 2
admits a meromorphic continuation for s 2 C with the possible exception of two simple poles
at s = 0 and s = 1. Moreover it satisfies the functional equation
_
⇤( , s) = "( )⇤( ,1 s),
where the root number can be written in terms of Gauss sums and has norm 1 by (5.4.3).
Following the same ideas of the previous example it can be shown that L( , s) is a function of
order at most 1, so that L( , s) defines an L function.
Now we are ready to generalize some results from Chapter 3. We begin with some observations:
since ⇤(f, s) = (f, s)L(f, s) is analytic on C with the possible exception of poles at s = 0 and
s = 1, then the poles of (f, s) on s 6= 0 are zeros of L(f, s). This zeros are called the trivial
zeros and with our notation they are located at the points 2m j 6= 0 where 1 6 j 6 d and
m is a nonnegative integer. All the other zeros of L(f, s) are called the trivial zeros and they
are located in the critical strip 0 6 Re(s) 6 1. It is possible that some of the trivial zeros are
located on the critical strip, nevertheless a conjecture states that they are never in the interior.
Note that if ⇢ = + i , is a zero of ⇤(f, s), then ⇢ is a zero of ⇤(f , s). Therefore, by the
functional equation, 1 ⇢ = 1 + i is also a zero of ⇤(f, s).
Since the Euler product of an L function is absolutely convergent and (f, s) doesn’t vanish for
Re(s) > 1, then ⇤(f, s) has no zeros on this half-plane. Therefore, by the functional equation,
⇤(f, s) 6= 0 for Re(s) < 0, i.e., all the zeros of ⇤(f, s) are on 0 6 Re(s) 6 1. Hence, by the
Hadamard Factorization Theorem we have
198
Theorem 5.4.4. Let L(f, s) be an L function. Then, there exists constants A = A(f ) and
B = B(f ) such that
Y⇣ s ⌘ ⇢s
r A+Bs
(s(s 1)) ⇤(f, s) = e 1 e ,
⇢
⇢
where ⇢ runs over all the zeros of L(f, s) on 0 6 Re(s) 6 1 distinct from 0 and 1. Hence,
L0 1 0
r r X 1 1
(f, s) = log q (f, s) + B + + (5.4.9)
L 2 s s 1 ⇢
s ⇢ ⇢
Let us denote by
X ⇤f (n) 1
L0
(f, s) =
L n=1
ns
the expansion in Dirichlet series of minus the logarithmic derivative of L(f, s). In terms of the
local parameters, if p 2 P and k 2 N, then
d
X
⇤f (pk ) = ↵j (p)k log p. (5.4.10)
j=1
On the other hand, by the functional equation ⇤(f, s) = "(f )⇤(f , 1 s),
L0 L0 0 0
(f, s) + (f , s) = log q (f, s) (f, 1 s).
L L
199
Hence,
X 1 1 1 1
2ReB = + + + . (5.4.11)
⇢
s ⇢ 1 s ⇢ ⇢ ⇢
For s 6= ⇢, we have
1 1 1 1 1 1
+ ⌧ and + ⌧ 2.
s ⇢ 1 s ⇢ |⇢|2 ⇢ ⇢ |⇢|
Therefore, by Corollary 3.4.0.2, the series
X 1 1 X1 1
+ and +
⇢
s ⇢ 1 s ⇢ ⇢
⇢ ⇢
converge absolutely, whence we can separate the series in (5.4.11). Note that since ⇢ and
1 ⇢ are zeros of ⇤(f, s), then
X 1 1
+ = 0.
⇢
s ⇢ 1 s ⇢
This shows that X ⇣1⌘
Re(B) = Re .
⇢
⇢
200
(iii) Let s = + it with 1
2
6 6 2. Note that
L0 L0 0 0
r r r r
(f, s) (f, 3 + it) = (f, s) + (f, 3 + it) + +
L L s s 1 3 + it 2 + it
X 1 1
+ .
⇢
s ⇢ 3 + it ⇢
On the other hand, using the definition of (f, s) and (2.2.2), we get
0
d Xd ⇣s + ⌘
j
(f, s) = log ⇡ +
2 j=1
2
d
d
X 1
d
X ⇣s + ⌘
j
= log ⇡ + +1 .
2 j=1
s + j j=1 2
Hence
L0 r r X 1 X 1 L0 d
(f, s)+ + = (f, 3 + it) + log ⇡
L s s 1 s + j s ⇢ L 2
|s+j |<1 |t |<1
X 1
d
X ⇣s + ⌘ r r
j
+ + +1 + +
s + j j=1 2 3 + it 2 + it
|s+j |>1
X 1 1 X 1
+ .
s ⇢ 3 + it ⇢ 3 + it ⇢
|t |>1 |t |<1
We have to show that each term on the right-hand side is O(log q(f, s)). By (5.4.12),
L0 d X 1 r r
(f, 3 + it) + log ⇡ + + + ⌧ log q(f, s)
L 2 s + j 3 + it 2 + it
|s+j |>1
201
Theorem 5.4.6. Let L(f, s) be an L function of degree d and let N (T, f ) be the number of
zeros ⇢ = + i of L(f, s) with 0 6 6 1 and 0 6 6 T . Then
T qT d
N (T, f ) = log + O(log q(f, iT )) (T > 1)
2⇡ (2⇡ e)d
Proof. Let
⇠(f, s) ··= [s(s 1)]r ⇤(f, s).
Since the zeros of L(f, s) on 0 6 Re(s) 6 1 are the zeros of ⇠(f, s) we shall work with the
entire function ⇠(f, s) rather than with L(f, s). Let L be the positively oriented rectangle with
vertices 1, 2, 2 + iT, 1 + iT , as in Figure 5.1.
1
1 + iT 2
+ iT 2 + iT
1 1 1 2
2
Hence
1
N (T, f ) =
M arg ⇠(f, s).
⇡
Now we compute the variation of the argument of
s ds
Yd ⇣s + ⌘
r j
⇠(f, s) = [s(s 1)] q ⇡2 2 L(f, s)
j=1
2
202
and ⇣q⌘
T s dT dsT
M arg q ⇡ = log q 2 log ⇡ = log d .
2
2 2 2 ⇡
To compute the variation of the argument of the gamma factors we use Stirling’s series (2.2.12)
in the form ⇣ 1⌘ 1 ⇣1⌘
log (s) = s log s s + log 2⇡ + O .
2 2 |s|
Note that this formula implies that
X1
⇤f (n)
log L(f, s) = ⌧ log q(f, iT ).
n=1
ns log n
Now,
Z 1 +iT ⇣ 1 ⌘
2
Im ds = | arg(s ⇢)| 6 ⇡ ⌧ 1
2+iT s ⇢
and similarly,
Z 1 +iT ⇣
2 1 ⌘
Im ds ⌧ 1.
2+iT s + j
By (iii) of the previous theorem,
L0 r r X 1 X 1
(f, s) = + + + O(log q(f, s)).
L s s 1 s + j s ⇢
|s+j |<1 |t |<1
Therefore,
Z 1 +iT X
2 L0
Im (f, s) ds ⌧ 1 + log q(f, s) ⌧ m(T, f ) + log q(f, iT ) ⌧ log q(f, iT ). ⌅
2+iT L
|t |<1
203
Now we can generalize the explicit formula in the following way:
Theorem 5.4.7. Let ' 2 C 1 (R>0 , C) be a function of compact support and let
Z1
M'(s) = '(x)xs 1 dx
0
be its Mellin transform. Let(x) ··= x 1 '(x 1 ), so that M (s) = M'(1 s). Then
1
X Z1
[⇤f (n)'(n) + ⇤f (n) (n)] ='(1) log q + r '(x) dx
n=1 0
Z 1 +i1 h 0 0 i X
1 2
+ (f, s) + (f , 1 s) M'(s) ds M'(⇢),
2⇡i 1
2
i1 ⇢
where ⇢ runs over all the zeros of L(f, s) (counted with multiplicity) on the strip 0 6 6 1.
where ↵ > 1. Let c be a small positive number such that L(f, s) has no zeros on c 6 Re(s) < 0
(this is possible because m(T, f ) ⌧ log q(f, iT )). The idea of the proof is similar to the proof of
the Riemann-Von Mangoldt explicit formula: we move the vertical line of integration Re(s) = ↵
away to the line of integration Re(s) = c and then express the left-hand side of (5.4.14) as
the sum of the residues of the function
h L0 i
(f, s) M'(s)
L
at its poles. We use (5.4.9) to compute the residue of the poles. There’s a simple pole at s = 1
with residue Z1
rM'(1) = r dx.
0
The residue at s = ⇢ = + i with 0 6 6 1 is
M'(⇢).
Therefore,
1
X Z1 X Z c+i1 h i
1 L0
⇤f (n)'(n) = r dx M'(⇢) + (f, s) M'(s) ds. (5.4.15)
n=1 0 ⇢
2⇡i c i1 L
204
by the Mellin inversion theorem. Now,
Z c+i1 h 0 i X1 Z c+i1
1 L 1
(f , 1 s) M'(s) ds = ⇤f (n) · M'(s)ns 1
ds
2⇡i c i1 L n=1
2⇡i c i1
1
X
= ⇤f (n) (n) (5.4.17)
n=1
again by Mellin inversion. Hence, combining (5.4.15), (5.4.16) and (5.4.17) we obtain
1
X Z1
[⇤f (n)'(n) + ⇤f (n) (n)] ='(1) log q + r '(x) dx
n=1 0
Z c+i1 h 0 0 i X
1
+ (f, s) + (f , 1 s) M'(s) ds M'(⇢).
2⇡i c i1 ⇢
Finally, to complete the proof we can move the line of integration for the gamma factors to
0 0
Re(s) = 12 without residues (the poles of (f, s) cancel out with those of (f , 1 s)). ⌅
When we know the nature of the trivial zeros of an L function we can get better results; for
example we can prove the following result for the Riemann zeta function:
1
"
Theorem 5.4.8. Let F 2 C 1 [1, 1] and suppose that for some " > 0, F (u) ⌧ u 2 . Let
Z1
(s) = F (u)us 1 du
1
205
where is the digamma function. Therefore,
Z ↵+i1 h
1 ⇣ s ⌘i
1
X 1 1 1 1
J= n 2 ⇤(n)F (n) + log ⇡F (1) + (s ) ds.
n=1
2 2⇡i ↵ i1 2 2 2
On the other hand, if we move the vertical line of integration away to infinity on the left and
express J as the sum of the residues of the function
h g0 i 1
(s) (s )
g 2
206
at its poles we get
1 1 X 1
J = ( )+ ( ) (⇢ ),
2 2 ⇢
2
where the factors come from the residue at s = 1, s = 0 and s = ⇢ respectively. This completes
the proof. ⌅
207
Bibliography
[3] T. J. I. Bromwich. An Introduction to the Theory of Infinite Series. Macmillan and Co.,
Limited, London, 1908.
[7] E. Copson. Introduction to the Theory of Functions of a Complex Variable. Oxford Uni-
versity Press, 1970.
[8] H. Davenport. Multiplicative number theory. Graduate Texts in Mathematics, Vol. 74.
Springer, 2nd edition, 1982.
[13] G. Grubb. Distributions and Operators. Graduate Texts in Mathematics 252. Springer-
Verlag New York, 1 edition, 2009.
[14] E. Hewitt and K. A. Ross. Abstract harmonic analysis. Vol. I, volume 115 of Grundlehren
der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences].
Springer-Verlag, Berlin-New York, second edition, 1979. Structure of topological groups,
integration theory, group representations.
[15] H. Iwaniec. Lectures on the Riemann Zeta Function. University Lecture Series. American
Mathematical Society, 2014.
208
[16] H. Iwaniec and E. Kowalski. Analytic number theory, volume 53. American Mathematical
Soc., 2004.
[17] J. L. Kelley. General topology. Springer-Verlag, New York-Berlin, 1975. Reprint of the
1955 edition [Van Nostrand, Toronto, Ont.], Graduate Texts in Mathematics, No. 27.
[18] H. Koch. Number Theory: Algebraic numbers and functions. Number 24. American Math-
ematical Soc., 2000.
[20] S. Lang. Algebra, volume 211 of Graduate Texts in Mathematics. Springer-Verlag, New
York, third edition, 2002.
[21] S. Lang. Algebraic number theory, volume 110. Springer Science & Business Media, 2013.
[23] C. B. M. Murray H. Protter. A First Course in Real Analysis, Second Edition. Under-
graduate Texts in Mathematics. Springer, 2nd edition, 1991.
[25] J. Neukirch. Algebraic number theory, volume 322. Springer Science & Business Media,
2013.
[26] F. Olver. Asymptotics and special functions. AKP Classics. A K Peters, Ltd., Wellesley,
MA, 1997.
[27] B. R. A. W. Peter Borwein, Stephen Choi. The Riemann Hypothesis: A Resource for the
Afficionado and Virtuoso Alike. CMS Books in Mathematics. Springer, 1 edition, 2007.
[28] D. Ramakrishnan and R. J. Valenza. Fourier analysis on number fields, volume 186 of
Graduate Texts in Mathematics. Springer-Verlag, New York, 1999.
[29] W. Rudin. Principles of mathematical analysis. International series in pure and applied
mathematics. McGraw-Hill, 3d ed edition, 1976.
[31] P. Samuel. Algebraic Theory of Numbers: Translated from the French by Allan J. Silberger.
Courier Corporation, 2013.
[32] I. Stewart. Galois theory. CRC Press, Boca Raton, FL, fourth edition, 2015.
[33] J. Tate. Fourier analysis in number fields and hecke’s zeta functions. Thesis Princeton,
1950.
209
LA FUNCIÓN ZETA DE RIEMANN Y LA TESIS DE TATE
APROBADO:
Jhon Jairo Sutachan Rubio, Ph.D. Alba Alicia Trespalacios Rangel, Ph.D.
Director de Posgrado Decana
Facultad de Ciencias Facultad de Ciencias