16 views

Uploaded by scribeknorr

some notes on Analytic number thoery

- W11_16C1_lec_2_28&3_2_11
- unitplan ch11
- mit6 003f11 lec14
- Harold Davenport - The Higher Arithmetic
- unitplan ch11
- 16_2
- Convergin Series
- Bsc Maths & Statistics
- Infinite Series.pdf
- Stewart, I.a. - Galois Theory, 3rd Ed
- LECTURE_8_Sequences and Series_S2_2015-2016.pdf
- 4 Dispersion Representations
- Advance Engineering Mathematics for Ie
- Taylor Series - Wikipedia
- Chap01 Metric and Normed Spaces
- A FUNCTIONAL DETERMINANT EXPRESSION FOR THE RIEMANN XI FUNCTION
- 1st Grading Summary of Lessons
- Formulas Differential Eqn From StudentPStuder
- Functional Analysis
- Woon1998_FractalZeta_9812031

You are on page 1of 25

Lecture Notes

Taught Course Centre 2007

Tim D. Browning

Typeset by Sean Prendiville

Part A

Basic Topics in Analytic Number

Theory

1 Introduction

1.1 What is Analytic Number Theory?

A basic human instinct is to count interesting objects. For example:

Prime numbers less than some bound

(x) := #p N : p prime and p x

What is

R

s,d

(N) := #(x

1

, . . . , x

s

) N

s

: x

d

1

+ +x

d

s

= N ?

For which s, d do we have

R

s,d

(N) > 0 ?

Lagrange (1770):

R

1,2

(N) +R

2,2

(N) +R

3,2

(N) +R

4,2

(N) > 0 for all N N

Often non-exact answers are sucient, which is where analysis enters the picture.

It is not always clear why or how analysis should help with theorems like:

Every odd number greater than 10

1347

is a sum of three primes (Vinogradov).

The sequence of primes contains arbitrarily long arithmetic progressions (Green

and Tao).

All errors are the responsibility of the typesetter. In particular there are some arguments

which, as an exercise for the typesetter, have been eshed out or re-interpreted, possibly incor-

recly. Tims lectures were neater and more concise. Corrections would be gratefully received at

sean.prendiville@bristol.ac.uk

1

1.2 Notation

Let f, g : R C with g non-negative. Then we say:

(1) f = O(g), or f g, if there exists x

0

and C > 0 such that for all x x

0

[f(x)[ Cg(x)

(2) f g if f is also non-negative and g f

(3) f g if

lim

x

f(x)

g(x)

= 1

(4) f g if f g f.

(5) f = o(g) if for any > 0 there exists X R such that for all x X

[f(x)[ g(x)

So f = o(g) certainly when

lim

x

f(x)

g(x)

= 0

2 Arithmetic Functions

An arithmetic funciton is any function f : N C. It is said to be multiplicative

if it is not identically zero and for any m, n N with (m, n) = 1

f(mn) = f(m)f(n) (1)

f is said to be completely multiplicative if (1) holds even when m and n are not

coprime.

Lemma 1. If f, g are multiplicative, then f g : N C dened by

f g(n) :=

d|n

f(d)g(n/d)

is also multiplicative.

Proof. Let (m, n) = 1. Then every divisor d[mn can be written as a product d = ab

with a[m and b[n. In particular

(a, b) =

_

m

a

,

n

b

_

= 1

It can be checked that this induces a bijection

d N : d[mn a N : a[m b N : b[n

2

given by d = ab (a, b). Hence

f g(mn) =

a|m

b|n

f(ab)g

_

mn

ab

_

=

a|m

b|n

f(a)g

_

m

a

_

f(b)g

_

n

b

_

=

_

f g(m)

__

f g(n)

_

Remark. The operation is called Dirichlet convolution.

In fact the set of all multiplicative arithmetic functions, equipped with , forms

an abelian group with identity I dened by

I(n) =

_

1

n

_

=

_

1 if n = 1

0 otherwise

Note. If f is multiplicative then

f(1) = 1

and if p

1

, . . . , p

r

are distinct primes, then for any non-negative integers e

i

f (p

e

1

1

p

e

r

r

) = f (p

e

1

1

) f (p

e

r

r

)

Lets have some examples:

2.0.1 The Mobius Function

(n) =

_

_

1 if n = 1

(1)

r

if n = p

1

p

r

where p

1

, . . . , p

r

are distinct primes.

0 otherwise

Clearly is multiplicative. A key property of is:

Lemma 2. We have

d|n

(d) =

_

1 if n = 1

0 otherwise

Proof. Suppose n > 1. Then there exist distinct primes p

1

, . . . , p

r

and positive

integers e

1

, . . . , e

r

such that n = p

e

1

1

p

e

r

r

.

d|n

(d) = 1 +(p

1

) +(p

2

) + +(p

r

) +(p

1

p

2

) +. . .

= 1 +

_

r

1

_

(1)

1

+

_

r

2

_

(1)

2

+ +

_

r

r

_

(1)

r

= (1 1)

r

= 0

3

Lemma 3 (Mobius Inversion). Let f be an arithmetic function and dene

F(n) :=

d|n

f(d)

Then

f = F

Proof.

F (n) =

ab=n

F(a)(b)

=

ab=n

cd=a

f(c)

_

n

cd

_

=

cd|n

f(c)

_

n

cd

_

=

c|n

f(c)

d|

n

c

_

n/c

d

_

= f(n) by Lemma 2.

2.0.2 The Euler Totient Function

(n) :=

dn

I

_

(d, n)

_

= #d n : d and n coprime

Lemma 2 implies that

(n) =

dn

c|(d,n)

(c)

=

c|n

(c) #d n : c[d

=

c|n

(c)

n

c

= h(n)

4

where h(x) := x for all x N. Hence, by Lemma 1, is a multiplicative arithmetic

function. In fact:

(n) = (p

e

1

1

p

e

r

r

)

=

i

(p

e

i

i

)

=

i

_

_

d|p

e

i

i

(d)

p

e

i

i

d

_

_

=

i

_

p

e

i

i

p

e

i

1

i

_

= n

p|n

_

1

1

p

_

2.0.3 The Divisor Function

d(n) :=

d|n

1 = u u(n)

where u(x) := 1 for all x N. Hence d is another example of a multiplicative

function. More generally:

d

k

(n) :=

a

1

a

k

=n

1

=

a

k

|n

a

1

a

k1

=

n

a

k

1

= (u d

k1

)(n)

It follows by a simple induction and the fact that d

1

= u (or d

2

= d) that d

k

is

multiplicative for all k N.

Lemma 4.

d

k

(n) d(n)

k1

Proof. Notice that when c[n then d

k

(c) d

k

(n). Hence for all k 2:

d

k

(n) =

c|n

d

k1

(c)

d

k1

(n)

c|n

1

= d

k1

(n)d(n)

The result now follows by induction.

3 Partial Summation

The behaviour of arithmetic functions can be quite erratic. For example, d(n) takes

the value 2 innitely often, but can also be extremely large (see the exercises).

However, one still thinks of d(n) as a small function:

5

Lemma 5. For all > 0

d(n)

(2)

Proof. If n =

p

p

e

then

d(n)

n

p

e + 1

p

e

=

p

f

p

(e)

where each f

p

(e) :=

e+1

p

e

is a non-negative function which tends to 0 as e .

In particular, f

p

has a maximum e

p

N 0. Then

f

p

(e

p

) f

p

(e

p

1)

Re-arranging, this implies

1

p

1

e

p

1

p

1

1 (3)

Hence e

p

=

_

1

p

1

_

, unless e

p

=

1

p

1

1. In the latter case, reversing our previous

re-arrangement gives f

p

(e

p

) = f

p

(e

p

+1), so

_

1

p

1

_

= e

p

+1 is again a maximum

of f

p

. We may therefore assume e

p

=

_

1

p

1

_

for all p. For all n N we thus have

that

d(n)

n

p

f

p

(e

p

) := C()

Notice that e

p

= 0 if p > 2

1/

, so C() is certainly nite.

Note. C() is in fact the best possible implied constant in (2) (equality can be

achieved by taking n =

p

p

e

p

).

It is often more revealing to study the mean value of an arithmetic function f:

1

x

nx

f(n) as x

An alternative reason for studying these averages is when f is an indicator function

for an interesting set, e.g.

(x) :=

nx

n

where

n

=

_

1 if n is prime

0 otherwise

Part A of this course will culminate in a proof of the Prime Number Theorem (de

la Vallee-Poussin, Hadamard; 1896), which states:

(x)

x

log x

as x

6

Lemma 6 (Partial Summation). Suppose a

n

C (n N) and that f is continu-

ously dierentiable on [x, y]. Dene

A(t) :=

nt

a

n

Then

x<ny

a

n

f(n) = A(y)f(y) A(x)f(x)

_

y

x

A(t)f

(t)dt

Proof. We can clearly extend f to a continuously dierentiable function on [1, y]

by dening it linearly on [1, x). Then

_

y

x

A(t)f

(t)dt =

_

y

1

A(t)f

(t)dt

_

x

1

A(t)f

(t)dt

Well partition [1, y] into [y] sub-intervals by dening y

i

:= i for i = 1, . . . , [y] and

y

i

:= y for i = [y] + 1. Then

_

y

1

A(t)f

(t)dt =

iy

_

y

i+1

y

i

A(t)f

(t)dt

=

iy

ni

a

n

_

y

i+1

y

i

f

(t)dt

=

ny

in

a

n

_

y

i+1

y

i

f

(t)dt

=

ny

a

n

_

y

n

f

(t)dt

Similarly

_

x

1

A(t)f

(t)dt =

nx

a

n

_

x

n

f

(t)dt

Therefore

_

y

x

A(t)f

(t)dt =

ny

a

n

(f(y) f(n))

nx

a

n

(f(x) f(n))

= A(y)f(y) A(x)f(x)

x<ny

a

n

f(n)

Lemma 7.

nx

1

n

= log x + +O

_

1

x

_

for some constant > 0.

7

Proof. Let a

n

= u(n) and f(x) = x

1

in Lemma 6. Then

A(x) = [x] = x x

= x +O(1)

Hence

nx

1

n

= 1 +

1<nx

1

n

= 1 +

[x]

x

[1]

1

+

_

x

1

[t]

t

2

dt

= 1 +O

_

1

x

_

+

_

x

1

1

t

dt

_

x

1

t

t

2

dt

= log x + +O(x

1

)

where

= 1

_

1

t

t

2

dt

Here = lim

x

_

_

nx

1

n

log x

_

_

0.5772 is Eulers constant.

Were now ready to study the mean value for the divisor function. Our rst

attempt proceeds as follows:

D(X) :=

nX

d(n) =

nX

d|n

1 (4)

=

dX

nX

d|n

1

_

_

_

_

=

d,e

deX

1

_

_

_

_

(5)

=

dX

_

X

d

_

(6)

= X log X +O(X) (7)

The rightmost sum in (5) counts the number of points in the lattice N N which

lie on or below the hyperbola xy = X. Dirichlet signicantly improved the error

term above by utilising the symmetry in the inequality ab X:

abX

1 =

abX

a

X

1 +

abX

b

X

1

abX

a,b

X

1

= 2

abX

a

X

1

_

X

_

2

= 2

X

_

X

a

_

[

X]

2

8

Here we have fewer terms of summation than in our rst attempt. This trick is

called the Dirichlet hyperbola method. By Lemma 7

2

X

_

X

a

_

[

X]

2

= 2X

_

1

2

log X + +O

_

X

1/2

__

+O

_

X

1/2

_

X

Dividing the latter by X we obtain the following lemma.

Lemma 8.

1

x

nx

d(n) = log x + (2 1) +O

_

x

1/2

_

Remark. Let (x) := D(x)(xlog x + (2 1)x). The problem of understanding

the behaviour of (x) is called the Dirichlet divisor problem and is still actively

investigated. We saw in Lemma 8

(x) x

1/2

and in fact we know that

(x) =

_

x

1/4

_

(this means there are innitely many values of x for which (x) grows faster than

x

1/4

). The current record is

(x) x

131

416

+

(Huxley)

We now dene three functions which play an important role in prime number

theory. The von Mangoldt function:

(n) =

_

log p if n = p

k

for some prime p and some k N

0 otherwise

and the Chebyshev functions:

(x) =

px

log p, (x) =

nx

(n)

How do these relate to (x)? Well show that

(x)

x

log x

(x) x (8)

(x) x (9)

In order to establish the above equivalences we utilise the following: f g if and

only if f(x) = g(x)

_

1 +o

_

1(x)

_

_

, where 1(x) = 1 for all x.

Applying partial summation we have

(x) = (x) log x

_

x

2

(t)

t

dt

Supposing (x)

x

log x

_

x

2

(t)

t

dt =

_

x

2

1 +o

_

1(t)

_

log t

dt

= O

__

x

2

1

log t

dt

_

9

Therefore

lim

x

(x)

x

= 1 + lim

x

O

_

1

x

_

x

2

1

log t

dt

_

The right-hand side above is equal to one, since

_

x

2

dt

log t

=

_

x

2

dt

log t

+

_

x

x

dt

log t

x +

x

log x

For the converse, we once again begin by using partial summation to deduce:

(x) =

(x)

log x

+

_

x

2

(t)

t(log t)

2

So if (x) x, then

lim

x

(x) log x

x

= 1 + lim

x

1

x

_

x

2

t +o(t)

t(log t)

2

dt

= 1 + lim

x

O

_

1

x

_

x

2

1

(log t)

2

dt

_

The last expression is equal to one by a similar argument to that given before. Thus

(8) holds. To prove (9), note that

0 (x) (x) =

2m

log x

log 2

p

p

m

x

log p

=

2m

log x

log 2

_

x

1/m

_

2m

log x

log 2

x

1/m

log

_

x

1/m

_

log x

log 2

x

1/2

log

_

x

1/2

_

x(log x)

2

Hence it follows that

(x)

x

=

(x)

x

+O

_

(log x)

2

x

_

which establishes (9).

To prove the Prime Number Theorem (PNT), we begin by showing that (x)

has the right order of magnitude.

Lemma 9. We have x (x) x.

Proof. In view of (x) = (x) + O

_

x(log x)

2

_

it suces to establish the result

for . An easy application of partial summation implies that

T(x) :=

nx

log n

= xlog x x +O(log x)

10

Taking logs in the prime factorisation of n, its easy to see that

log n =

d|n

(d)

Thus it follows that

T(x) =

nx

d|n

(d) =

dx

(d) [x/d]

and so

T(x) 2T(x/2) =

dx

(d)

__

x

d

_

2

_

x

2d

__

Now

_

x

d

2

_

x

2d

x

2

< d x. Hence

(x) (x/2) T(x) 2T(x/2) (x)

whilst T(x) 2T(x/2) = xlog 2 +O(log x). In particular, (x) x and further-

more

(x) (x/2) +xlog 2 +O(log x)

(x/4) + (1 + 1/2)xlog 2 +O(2 log x)

.

.

.

(x/2

r

) + (1 + 1/2 + + 1/2

r1

)xlog 2 +O(r log x)

for any r 1. Choose r such that 2

r

x < 2

r+1

. Then

(x) (2 log 2)x +O

_

(log x)

2

_

Euler observed that the sum of the reciprocals of the primes is like the logarithm

of the harmonic series. He proved

px

1

p

Theorem A 1 (Mertens). There exists a constant B such that

px

1

p

= log log x +B +O

_

1

log x

_

We begin the proof with:

Lemma 10.

px

log p

p

= log x +O(1)

11

Proof.

px

log p

p

=

nx

(n)

n

px

log p

2k

log x

log p

1

p

k

We have

px

log p

2k

log x

log p

1

p

k

px

log p

p

2

1 p

1

px

log p

p

2

n=1

log n

n

2

1

Therefore

px

log p

p

=

nx

(n)

n

+O(1)

By Lemma 9

nx

(n)

n

= x

1

nx

(n)

_

x

n

_

+x

1

nx

(n)

_

x

n

_

= x

1

nx

log n +x

1

O((x))

= log x 1 +x

1

O(log x) +x

1

O(x)

= log x +O(1)

Proof of A 1. Let

a

n

=

_

log n

n

if n is prime

0 otherwise

By partial summation (Lemma 6):

px

1

p

=

1

2

+

2<nx

a

n

log n

=

1

2

+

_

_

nx

a

n

_

_

1

log x

1

2

+

_

x

2

_

_

nt

a

n

_

_

1

t(log t)

2

dt

=

_

log x +O(1)

_

1

log x

+

_

x

2

_

log t +O(1)

_

1

t(log t)

2

dt

= 1 +O

_

1

log x

_

+ log log x log log 2 +O

_

1

log x

1

log 2

_

= log log x +B +O

_

1

log x

_

for some B

12

4 Dirichlet Series

A Dirichlet series is an innite series of the form

F(s) :=

n=1

a

n

n

s

where a

n

will be an arithmetic function and s = +it is a complex variable. They

are ubiquitous in analytic number theory, forming an important class of generating

functions.

Lemma 11. Suppose F(s

0

) converges for some value s

0

=

0

+ it

0

. Then F(s)

converges uniformly on every compact subset of the half-plane >

0

.

Proof. Let K be a compact subset of the half-plane >

0

. Then K is bounded,

so there is some B such that [s[ B for all s K and it can be checked that

there exists > 0 such that for all s K,

0

+. We apply partial summation

(Lemma 6), taking

a

n

n

s

0

(n N) as our sequence and f(t) = t

(ss

0

)

as our

continuously dierentiable function, to get that

a<nb

a

n

n

s

1

b

ss

0

nb

a

n

n

s

0

1

a

ss

0

na

a

n

n

s

0

+ (s s

0

)

_

b

a

_

_

nt

a

n

n

s

0

_

_

dt

t

ss

0

1

S

_

1

b

+

1

a

+

[s s

0

[

0

_

1

b

1

a

0

__

2S( +B +[s

0

[)

a

Where S = sup

NN

n=1

a

n

n

s

0

completes the proof.

Dene the abscissa of convergence of a Dirichlet series F(s) to be

c

:= inf R : F( +it) converges

Similarly the abscissa of absolute convergence is given by

a

:= inf

_

R :

n=1

[a

n

[

n

converges

_

Example. We dene the Riemann zeta function to be

(s) :=

n=1

1

n

s

(for > 1)

Its clear that (s) is absolutely convergent for > 1 and diverges for s = 1. Hence

(in this case)

c

=

a

= 1.

In general we have

c

a

c

+1 for any Dirichlet series.

c

a

is obvious.

To see the second inequality suppose F(s

0

) converges for some s

0

. It suces then

13

to show F(s) converges absolutely for >

0

+ 1. Since we must have

a

n

n

s

0

0,

there exists A such that for all n N

a

n

n

s

0

A

Hence for >

0

+ 1

n=1

a

n

n

s

n=1

1

n

0

<

Lemma 12. Let

F(s) =

n=1

a(n)

n

s

, G(s) =

n=1

b(n)

n

s

be Dirichlet series with abscissae of absolute convergence

a

(F),

a

(G). Provided

> max

a

(F),

a

(G), we have

F(s)G(s) =

n=1

(a b)(n)

n

s

Proof.

F(s)G(s) =

m,n=1

a(m)b(n)

(mn)

s

=

k=1

mn=k

a(m)b(n)

k

s

Example. Recall that u(n) = 1 and I(n) = [1/n] (n N). Then we know u = I.

Clearly the Dirichlet series

n=1

(n)

n

s

has abscissa of absolute convergence 1. So for > 1

(s)

n=1

(n)

n

s

=

n=1

u (n)

n

s

n=1

I(n)

n

s

= 1

Thus we conclude that when > 1

n=1

(n)

n

s

= (s)

1

(10)

14

Example. Since d(n) = u u(n)

(s)

2

=

n=1

d(n)

n

s

( > 1)

The following is a key property of Dirichlet series (sometimes called the analytic

fundamental theorem of arithmetic):

When a(n) is a multiplicative arithmetic function, F(s) has an Euler

product.

Lemma 13 (Euler Product Formula). Suppose f is a multiplicative function and

F(s) :=

n=1

f(n)

n

s

has abscissa of absolute convergence

a

. Then provided >

a

F(s) =

p

_

f(1)

1

+

f(p)

p

s

+

f(p

2

)

p

2s

+

_

In particular, if f is completely multiplicative, then

F(s) =

p

_

1

f(p)

p

s

_

1

Proof. Let >

a

. Since F(s) is absolutely convergent, so is

m=0

f(p

m

)

p

ms

for all primes p. We can therefore dene

P(s; x) :=

px

m=0

f(p

m

)

p

ms

(x 2)

Let r = (x) . Absolute convergence ensures the following identity is valid

P(s; x) =

m

1

=0

m

r

=0

f (p

m

1

1

) f (p

m

r

r

)

(p

m

1

1

p

m

r

r

)

s

=

n

f(n)

n

s

#(m

1

, . . . , m

r

) : n = p

m

1

1

p

m

r

r

By the fundamental theorem of arithmetic

n

f(n)

n

s

#(m

1

, . . . , m

r

) : n = p

m

1

1

p

m

r

r

=

n

p|npx

f(n)

n

s

Therefore

[F(s) P(s; x)[

n>x

[f(n)[

n

15

Example. For > 1

(s) =

n=1

u(n)

n

s

Hence

(s) =

p

_

1

1

p

s

_

1

The above example is the rst piece of evidence for the following correspondence

principal:

Analytic properties of (s) Properties of primes

Exercise. By using Lemma 13, show that

n=1

(n)

n

s

= (s)

1

( > 1)

One of the key properties of Dirichlet series lies in the following result.

Lemma 14 (Perron). Let F(s) =

n=1

f(n)

n

s

be a Dirichlet series with abscissa of

absolute convergence

a

. Provided x / Z and > max

a

, 0 we have

nx

f(n) =

1

2i

_

+iT

iT

F(s)x

s

s

ds +O

_

x

n=1

[f(n)[n

log

_

x

n

_

_

Proof. Its an easy exercise in contour integration to show that for any > 0

1

2i

_

+iT

iT

u

s

s

ds =

_

_

1 +O

_

u

T log u

_

if u > 1

1

2

+O

_

T

_

if u = 1

O

_

u

T log(

1

u

)

_

if u (0, 1)

By Lemma 11

nN

f(n)

n

s

converges uniformly to F(s) on the set [iT, +iT] :=

+it : t [T, T]. We are therefore justied in swapping the order of integra-

tion and summation in the following:

_

+iT

iT

F(s)x

s

s

ds =

n=1

f(n)

_

+iT

iT

(x/n)

s

s

ds

=

nx

_

2if(n) +O

_

[f(n)[(x/n)

T log(x/n)

__

+

n>x

O

_

[f(n)[(x/n)

T log(n/x)

_

We end this section by establishing the analyticity of a Dirichlet series in an

appropriate half-plane.

Lemma 15. F(s) =

n=1

f(n)

n

s

is holomorphic for >

c

with derivative

F

(s) =

n=1

f(n) log n

n

s

16

Proof. Let H := s C : >

c

and let be a closed contour in H. Since the

partial sums F

x

(s) :=

nx

f(n)n

s

are entire functions, Cauchys Theorem tells

us that for all x

_

F

x

(z)dz = 0

By uniform convergence (Lemma 11) we can pass to a limit under the integral sign,

giving

_

F(z)dz = 0

It follows from Moreras Theorem that F is holomorphic on H. The treatment of

derivatives is similar

1

.

5 The Riemann Zeta Function

We now explore some basic properties of (s) as needed to establish the Prime

Number Theorem.

It follows from Lemma 15 that (s) is holomorphic on > 1. It is clearly

non-zero on > 1 (see (10)). Assume > 1. Then

(s) =

n=1

log n

n

s

Hence by Lemma 12

(s)

n=1

(n)

n

s

=

n=1

u (n)

n

s

=

n=1

log(n)

n

s

=

(s)

This establishes:

Lemma 16. For > 1 we have

n=1

(n)

n

s

=

(s)

(s)

Recall (x) =

nx

(n). Letting T in Perrons Formula (Lemma 14)

we get

(x) =

1

2i

_

+i

i

_

(s)

(s)

_

x

s

s

ds ( > 1)

This is one of the most popular approaches to PNT and naturally leads to the

study of analytic properties of and

1

Calculate the derivative of the partial sum, then use Cauchys integral formula for the rst

derivative, passing to the limit under the integral sign.

17

to Newman) but we will still need to make sense of (s) for 1. An application

of Lemma 6 gives

nx

n

s

=

[x]

x

s

+s

_

x

1

[t]t

s1

dt (11)

= x

1s

x x

s

+

s

s 1

_

1 x

1s

_

s

_

x

1

t t

s1

dt (12)

for all s ,= 1. Letting x when > 1 we obtain

Lemma 17. For > 1

(s) =

s

s 1

s

_

1

t

t

s+1

dt (13)

Note. For values of > 0, we have that

_

1

t

t

s+1

dt

_

1

1

t

+1

dt

1

Lemma 17 therefore gives a formula for (s) which continues to make sense for

(0, 1].

Consider the sequence of functions

f

N

(s) :=

s

s 1

s

_

N

1

t

t

s+1

dt (N N)

dened on the punctured half-plane := s C : s ,= 1 and > 0. Since t

t is periodic modulo 1, a quick calculation establishes that each f

N

is a nite sum

of functions holomorphic on , and therefore f

N

is itself holomorphic on . An

argument similar to that given in Lemma 11 shows that the f

N

converge uniformly

to (as dened by (13)) on all compact subsets of . Using Moreras Theorem (as

in Lemma 15) we can show that is holomorphic on every open ball contained in ,

and hence holomorphic in . Clearly has a simple pole at s = 1. To summarise:

Formula (13) gives an analytic continuation of (s) to > 0, whose

only pole is a simple one at s = 1.

Since (s) is not identically zero in the open connected set , it follows that the

zeroes of (s) form a discrete subset of . Moreover, we now also know that

(s)

is holomorphic in with a pole of order 2 at s = 1. Thus

(s)

(s)

is meromorphic in

the half-plane > 0 with a simple pole at s = 1.

Lemma 18. (i) For all 1 and t 2

[(s)[ = O(log t)

(ii) For all (0, 1), if > and t 1 then

[(s)[ = O

_

t

1

_

18

Proof. Let > 0, t 1, x 1. Combining (13) and (12) gives

(s)

nx

1

n

s

=

1

(s 1)x

s1

+

x

x

s

s

_

x

u

u

s+1

du

Therefore

[(s)[

nx

1

n

+

1

tx

1

+

1

x

+[s[

_

x

du

u

+1

(14)

First suppose 1 and t 2. Take x = t in (14) to deduce that

[(s)[ log t +

1

t

+

[ +it[

t

log t +

1

t

+

1

t

1

log t + 1 log t

Next suppose t 1 and where (0, 1). Again taking x = t in (14) we

have

[(s)[ log t +

_

1

t

+

1

_

t

1

t

1

Taking = 1/2 in Lemma 18.(ii) gives

(

1

2

+it)

= O

_

t

1/2

_

(t 1) (15)

The Lindelof Hypothesis predicts that for all > 0

(

1

2

+it)

= O

(t

)

The record is [(

1

2

+ it)[ = O

_

t

_

for = 0.156 . . . (M. Huxley). In fact the

Riemann Hypothesis implies the Lindelof Hypothesis.

Dene

(s) :=

p

log p

p

s

Then:

Lemma 19. For 1 we have that (s) ,= 0 and (s)

1

s1

is holomorphic.

Remark. We say f is holomorphic on the set X (where X is not necessarily open) if

there exists an open set C which contains X and on which f is holomorphic. So

in Lemma 19 we claim that (s)

1

s1

is holomorphic on some open set containing

1.

Proof. Weve already seen that (s) ,= 0 for > 1. For > 1, Lemma 16 implies

that

(s)

(s)

=

n=1

(n)

n

s

=

k=1

log p

p

ks

=

p

log p

p

s

1

19

Since

1

x 1

=

1

x

+

1

x(x 1)

we have

(s)

(s)

= (s) +

p

log p

p

s

(p

s

1)

(16)

Notice that

p

log p

p

s

(p

s

1)

converges absolutely for >

1

2

. It therefore follows from

Lemma 17 and the remarks after it, that (s) has a meromorphic continuation to

>

1

2

with poles only at s = 1 and the zeros of . To complete the proof of the

lemma, it suces to show that

(1 +it) ,= 0 for all real values of t. (17)

Since if this holds then for each t R there exists

t

(0, 1) such that is non-zero

on B

t

(1 +it). The set

z C : '(z) > 1

_

t

B

t

(1 +it)

is then open, contains '(z) 1 and contains no zeros of . It follows that (s)

1

s1

is holomorphic on this set.

To prove (17), suppose has a zero of order Z at s

0

:= 1 + i, where

R 0. By which we mean, there is a function g which is non-zero at

s

0

, holomorphic in a neighbourhood of s

0

and with (s) = (s s

0

)

g(s) in this

neighbourhood. So a simple pole is a zero of order 1. Similarly suppose (s) has

a zero of order at s = 1 +i2. Then certainly , 0 by Lemma 17 (the only

pole of in the half-plane > 0 is at s = 1). By (16) we have

lim

0

(1 + ri) =

_

_

1 r = 0

r = 1

r = 2

Note that

2

r=2

_

4

2 +r

_

(1 + +ir) =

p

log p

p

1+

_

p

i/2

+p

i/2

_

4

0

Multiplying through by and taking the limit as 0 we get

lim

0

((1+i2) + 4(1+i) + 6(1+) + 4(1++i) +(1++i2))

= 6 8 2 0

Therefore = 0 and so there does not exist a zero of (s) with = 1.

Remark. The above proof relies on a trick which cannot be used for general L-

functions: is special.

20

6 The Prime Number Theorem

We now have everything in place to prove:

Theorem A 2. (x) x as x .

We saw prior to Lemma 9 that this is equivalent to

(x)

x

log x

The main analytic tool is the following:

Lemma 20 (Newmans Analytic Lemma). For t 0 let f(t) be a bounded function

which is locally integrable, i.e. integrable on any compact subset of [0, ). Suppose

g(z) :=

_

0

f(t)e

zt

dt ('(z) > 0)

extends holomorphically to '(z) 0. Then

_

0

f(t)dt

exists and is equal to g(0).

Proof. When we say g extends holomorphically to '(z) 0 we mean that there

exists an open set E C which contains the closed half-plane '(z) 0 and on

which g has an analytic extension. We will rst show that for each R > 0 there

exists = (R) > 0 such that the set

E

R

:= z C : [z[ < R + 1 and '(z) > 2

is contained in E. Suppose otherwise. Then for each n N there exists z

n

CE

with

[z

n

[ < R + 1 and '(z) >

1

n

Since (z

n

)

n

is a sequence in compact set B

R+1

(0), it has a convergent subsequence

(z

k(n)

)

n

with limit z

0

. Then

z

0

B

R+1

(0) z C : '(z) 0 E

But C E is closed, so z

0

C E (a contradiction).

Next we dene the closed contour C

R

to be the boundary of the closed set

F

R

:= z C : [z[ R and '(z) E

R

We also dene:

C

R,+

:= z C

R

: '(z) > 0

C

R,

:= z C

R

: '(z) < 0

For each T > 0 let

g

T

(z) :=

_

T

0

f(t)e

tz

dt (z C)

21

Clearly each g

T

is entire.

We wish to prove that

lim

T

_

g(0) g

T

(0)

_

= 0

By Cauchys Integral Formula, we have that for all T > 0

g(0) g

T

(0) =

1

2i

_

C

R

g(z) g

T

(z)

z

_

1 +

z

2

R

2

_

e

Tz

dz (18)

Since the left-hand side of (18) is independent of R, it suces to show that for all

large R

limsup

T

_

C

R

I(z; R, T)dz R

1

(19)

where

I(z; R, T) :=

g(z) g

T

(z)

z

_

1 +

z

2

R

2

_

e

Tz

To establish this we will estimate the integral over C

R,+

and C

R,

separately. Let

B := sup[f(t)[. For z C

R,+

[I(z; R, T)[ =

1

R

_

R

z

+

z

R

_

e

Tz

_

T

f(t)e

tz

dt

=

2['(z/R)[

R

e

T(z)

_

T

f(t)e

tz

dt

2'(z)

R

2

e

T(z)

B

_

T

e

t(z)

dt

=

2'(z)

R

2

e

T(z)

B

e

T(z)

'(z)

=

2B

R

2

Hence

_

C

R,+

I(z; R, T)dz R

1

Next, dene

C

R,

:= z C : [z[ = R and '(z) < 0

Since

J(z; R, T) :=

g

T

(z)

z

_

1 +

z

2

R

2

_

e

Tz

is analytic on simply connected domain C [0, ), by Cauchys Theorem

_

C

R,

J(z; R, T)dz =

_

C

R,

J(z; R, T)dz

22

For z C

R,

[J(z; R, T)[ =

1

R

R

z

+

z

R

e

Tz

_

T

0

f(t)e

tz

dt

2['(z)[

R

2

e

T(z)

B

['(z)[

_

e

T(z)

1

_

2B

R

2

Finally we estimate

K(z; R, T) :=

g(z)

z

_

1 +

z

2

R

2

_

e

Tz

on C

R,

. K is analytic on the intersection of E

R

with the half-plane '(z) < 0.

This intersection is in fact an open convex set, so by Cauchys Theorem for convex

domains, for each (0, (R)]

_

K(z; R, T)dz =

_

C

R,

K(z; R, T)dz

where

z C : [z[ R and '(z) F

R

which intersects '(z) < 0.

Let

S

R

:= sup

zF

R

[f(z)[

and

:= z

: '(z) =

If z

then

[K(z; R, T)[

2S

R

R

Using elementary trigonometry, one can verify that the contour

has length

at most

2

2

R

, hence

K(z; R, T)dz

4

2

S

R

R

2

So choosing := min

_

(R),

1

S

R

_

we have that when R 1

_

K(z; R, T)dz

1

R

2

1

R

Now for z

[K(z; R, T)[ S

R

e

T(z)

[z[

S

R

e

T

So K(z; R, T) 0 uniformly on

as T . (19) follows.

23

We apply Newmans Lemma to study the integral

_

1

(x) x

x

2

For > 1, Lemma 6 tells us

(s) =

p

log p

p

s

= lim

x

px

log p

p

s

= lim

x

_

(x)

x

s

+s

_

x

1

(t)

t

s+1

dt

_

= s

_

1

(t)

t

s+1

dt

= s

_

0

(e

u

)e

su

du

The last equality is obtained by making the change-of-variables t = e

u

. Hence for

> 1

(s) = s

_

0

e

su

(e

u

)du

Let

f(t) := (e

t

)e

t

1

g(z) :=

(z + 1)

z + 1

1

z

Then

_

0

f(t)e

zt

dt =

_

0

(e

t

)e

t(z+1)

dt

_

0

e

zt

dt

= g(z)

By Lemma 9, f(t) is bounded. By Lemma 19, g(t) = (z + 1)(z + 1)

1

z

1

extends holomorphically to '(z) 1, i.e. '(z) 0.

By Newmans Lemma

_

0

_

(e

t

)e

t

1

_

is a convergent integral and equal to

_

1

(t) t

t

2

dt

(by a simple change of variables).

Assume that for some > 1, there exists arbitrarily large values of x such that

(x) x

Since is non-decreasing, we have

_

x

x

(t) t

t

2

_

x

x

x t

t

2

dt

=

_

1

t

t

2

dt

= ( 1 log ) > 0

24

This contradicts the convergence of

_

1

(t) t

t

2

dt

Similarly, for each < 1 there cant exist arbitrarily large values of x for which

(x) x

Therefore (x) x.

25

- W11_16C1_lec_2_28&3_2_11Uploaded byRuchi Sharma
- unitplan ch11Uploaded byapi-214017049
- mit6 003f11 lec14Uploaded byapi-246008426
- Harold Davenport - The Higher ArithmeticUploaded bymahmood1972
- unitplan ch11Uploaded byapi-214017049
- 16_2Uploaded byGuddu Balasubramaniyan
- Convergin SeriesUploaded bylostkin
- Bsc Maths & StatisticsUploaded byAkash Malhotra
- Infinite Series.pdfUploaded byAgus Leonardi
- Stewart, I.a. - Galois Theory, 3rd EdUploaded bysMgLeu
- LECTURE_8_Sequences and Series_S2_2015-2016.pdfUploaded byFaIz Fauzi
- 4 Dispersion RepresentationsUploaded bymaurizio.desio4992
- Advance Engineering Mathematics for IeUploaded byKristine Joy Mendoza Inciong
- Taylor Series - WikipediaUploaded byMadhav Sethia
- Chap01 Metric and Normed SpacesUploaded bygod40
- A FUNCTIONAL DETERMINANT EXPRESSION FOR THE RIEMANN XI FUNCTIONUploaded byJose Javier Garcia Moreta
- 1st Grading Summary of LessonsUploaded byRecil Marie Boragay
- Formulas Differential Eqn From StudentPStuderUploaded byNedel Labile
- Functional AnalysisUploaded byAniket Joshi
- Woon1998_FractalZeta_9812031Uploaded bymathimagics
- Lecture Notes to Transition to Advanced Mathematics - MauchUploaded byKennedy Bright
- A Course in Mathematical AnalysisUploaded byBerkics Péter

- Nature of EconomicsUploaded bydinesanv
- cloudsUploaded byapi-359749049
- paperUploaded byapi-403908859
- Noor Al Suwaidi London exhibition: Like Coral I Create CloudsUploaded bysultansq
- EOC Biology Practice Assessment Final[1]Uploaded byehshonorsbio
- SAP Catch Weight ManagmentUploaded bydcm0907
- Lx en ISO27002Uploaded byBouslmame Tariq
- Kamal - Present SimpleUploaded byFarhan Hakimi
- Bj Mba ResumeUploaded byLisa Wilson
- recount learning sequenceUploaded byapi-198669584
- master thesisUploaded byapi-231165167
- JAKAB a. (Obuda 2014) (1)_EN_teljesUploaded byCivitas Europica Centralis
- childcare professional norma rollins-last updated resumeUploaded byapi-316727865
- Mullins v. Salem Arkansas, City of et al - Document No. 21Uploaded byJustia.com
- KTPeilUploaded byKatherine Peil
- Marking Code Rg 73Uploaded byJosé Adelino
- augmentative and alternative communication reflectionUploaded byapi-389764408
- 8 Guide HinduismUploaded byGirish Kumar
- DEIF PPM3Uploaded byΝεκταριος Μαστορακης
- CV - Fabio Madeira en - 082018Uploaded byinspetorfabiomadeira
- EC303 Applied Electromagnetic Theory Syllabus(1)Uploaded byShabeeb Ali Oruvangara
- msrit 7-8 sem syllabus bookUploaded byWendy Bass
- Simulation of Switching ConvertersUploaded byL30N1
- Team Time 12_28Uploaded bypathwayskz
- NWN2 Animations ListUploaded bywitcher2
- C5962_Lab02Uploaded bysagravat
- QCAmapIntroUploaded byReem Hassan
- Exam RevisionUploaded byRita West
- Approach to Arthritis PatientUploaded byapi-3716867
- Notes 01Uploaded byaymanieee