Analytic Number Theory

Winter 2005
Professor C. Stewart
CHRIS ALMOST
Contents
1 Introduction 2
2 Elementary Approximations of (x) 3
3 Bertrands Postulate 4
4 Asymtotic Analysis 6
5 Riemanns Zeta Function 11
6 Proof of the Prime Number Theorem 14
7 Generalizing (x) 17
8 Law of Quadratic Reciprocity 20
9 Dirichlets Theorem 24
9.1 Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1
2 ANALYTIC NUMBER THEORY
Grades: Assignments 10%, Midterm 25%, Final 65%
Midterm: Friday, February 18, 2005; in class
References:
An Introduction to the Theory of Numbers by Hardy and Wright (Cambridge Press)
Introduction to Analytic Number Theory by Apostel (Springer-Verlag)
1 Introduction
1.1 Denition. A prime number is a positive integer greater than 1 such that the only positive factors of it are 1
and itself. Let (x) denote the number of primes less than or equal to x.
Basic Question: How are the primes distributed amoung the integers? How does (x) grow with x?
Let p
n
denote the n
th
prime. Is there a polynomial f [x] such that f (n) = p
n
? Clearly no, as if q is
prime and q| f (n) then q| f (n +kq) for all k . This observation shows further that any polynomial that takes
only prime values on the integers must be constant. There are examples of polynomials whose initial values are
surprisingly often prime. For example, Euler noticed that n
2
+n+41 is prime for n = 0, . . . , 39, and by translation,
(n 40)
2
+(n 40) +41 is prime for n = 0, . . . , 79. This is related to the fact that (

163) has class number

1. In the 70s Matijasevic proved Hilberts 10
th
problem, and in the process was able to prove that there is a
polynomial f [a, b, c, . . . , z] such that the positive values in f (
26
) is exactly the set of primes. In 1977 he
showed that 10 variables sufces.
Can we nd a non-constant polynomial f [x] such that f (n) is prime innitely often? Clearly yes,
f (x) = x + k works for any k . Dirichlet showed that for coprime k, there are innitely many primes
of the form kn +. Is x
2
+1 prime innitely often? Almost surely yes, but the best result known to date is that
n
2
+1 is a product of two primes for innitely many n. There is no polynomial of degree greater than one in one
variable known to take prime values innitely often. If instead we consider polynomials of two variables we can
go further. Friedlander and Iwaniec (1998) proved that there are innitely many primes of the form n
2
+m
4
. In
(2001) Heath Brown proved there are innitely many primes of the form n
3
+2m
3
.
Let the Mbius function : {1, 0, 1} be dened by
(n) =
_
_
_
1 if n = 1
(1)
k
if n is squarefree and k is the number of distinct prime factors of n
0 otherwise
In 1971, Gandhi proved that if Q
n
:= p
1
. . . p
n
, where p
i
is the i
th
prime, then p
n+1
is the unique integer which
satises the inequality
1 < 2
p
n+1
_
_
_
1
2
+

d|Q
n
(d)
2
d
1
_
_
_< 2
(S. Golmab, American Mathematical Monthly, 1974, p. 752-754)
Recall Wilsons Theorem, that an integer n > 1 is prime if and only if n|(n 1)! +1. Thus, for x > 1,
(x) =

2jx
__
( j 1)! +1
j
_

_
( j 1)!
j
__
where [] is the greatest integer function. This estimate is not particularily useful, as n! is extremely large
compared to n.
ELEMENTARY APPROXIMATIONS OF (x) 3
1.2 Theorem (Euclid). There are innitely many primes.
PROOF: Assume not. If p
1
, . . . , p
n
are all of the primes, then p
1
. . . p
n
+1 has no prime factors. This is a contradic-
tion.
Notice that we can extract a bound for (x) from Euclids proof. By an easy induction we can prove that
p
n
2
2
n
for all n. Indeed, p
1
= 2 2
2
1
, and p
n+1
p
1
. . . p
n
2
2++2
n
+1 = 2
2
n+1
2
+1 2
2
n+1
. Therefore for
x > 1, (x) satises 2
2
(x)
x < 2
2
(x)+1
. Taking logarithms, log
2
(log
2
x) 1 < (x).
2 Elementary Approximations of (x)
Fermat numbers are integers of the form 2
2
n
+1, n 0. He observed that F
n
:= 2
2
n
+1 is prime for n = 0, 1, 2, 3, 4.
He conjectured that this would always be the case. Almost certainly he was absolutely wrong. 641 divides F
5
,
and F
n
is known to be composite for n = 5, 6, . . . , 32.
2.1 Theorem (Plya). For any integers n, m with 1 n < m, gcd(F
n
, F
m
) = 1.
PROOF: Let us put m= n +k. Observe that the polynomial x
2
k
1 is divisible by x +1 in [x]. In particular,
x
2
k
1
x +1
= x
2
k
1
x
2
k
2
+ 1
If we take x = 2
2
n
then we get F
n
|F
m
2. The result follows since no Fermat number is even.
As an immediate consequence of this, p
n
2
2
n
+1. Also note that for x > 1,
2
(x)

px
_
1
1
p
_
1
=

px
_
1 +
1
p
+
1
p
2
+
_

nx
1
n

_
x
1
du
u
= log x
Thus (x)
loglog x
log2
.
2.2 Theorem. Let x , x 2. Then
(x)
1
2log2
log x and p
n
4
n
PROOF: Let p
1
, . . . , p
k
be the primes less than or equal to x. For any n with 1 n x we may write n = n
2
1
m
where n
1
and m is square-free. Then m = p

1
1
. . . p

k
k
where
i
{0, 1}. Thus there are at most 2
k
possible
choices for m. Since 1 n x, we see that there are at most

x choices for n
1
. Thus there are at most 2
k

x
numbers between 1 and x, which implies that x 2
k

x and hence

x 2
k
. Whence
1
2log2
log x k = (x).
Taking x = p
n
gives k = n and

p
n
2
n
, so p
n
4
n
.
Let p be a prime and n . What is the power of p that divides n!? Clearly it is

i=1
_
n
p
i
_
=
[log
p
n]

i=1
_
n
p
i
_
4 ANALYTIC NUMBER THEORY
2.3 Theorem. Let x 2. Then
3log2
8
x
log x
< (x) < 6log2
x
log x
PROOF: For each prime p, let r
p
denote the unique integer for which p
r
p
2n < p
r
p
+1
. Then the power of p
dividing
_
2n
n
_
is
r
p

k=1
_
2n
p
k
_
2
_
n
p
k
_
which is less that or equal to r
p
, since each term in the sum is either 0 or 1. Therefore
_
2n
n
_
divides

p2n
p
r
p
.
Hence 2
n

_
2n
n
_
(2n)
(2n)
, so nlog2 (2n) log(2n) and we get that
log2
2
2n
log(2n)
(2n)
Given x, choose n so that 2n x < 2n +2, so (x)
log2
2
2n
log(2n)
. Suppose x > 6. Then 2n >
3
4
x and, since
y
log y
is increasing for y > 6, we have
3log2
8
x
log x
< (x). The result follows for 2 x 6 as may be checked.
For the upper bound, observe that

n<p2n
p divides
_
2n
n
_
. Hence n
(2n)(n)

n<p2n
p
_
2n
n
_
4
n
, so
(2n) log n (n) log n 2nlog2. Thus
(2n) log n (n) log
n
2
(2log2)n +(n) log2 (3log2)n
If n = 2
k
, 2
k1
, . . . 4 then we get a telescoping collection of inequalities
(2
k+1
) log2
k
(2
k
) log2
k1
(3log2)2
k
(2
k
) log2
k1
(2
k1
) log2
k2
(3log2)2
k1
.
.
.
(8) log4 (4) log2 (3log2)4
Adding gives (2
k+1
) log2
k
< (3log2)2
k+1
, hence (2
k+1
) < 6log2
_
2
k
log2
k
_
. Therefore given x we choose k so
that 2
k
x < 2
k+1
. Whence
(x) (2
k+1
) < 6log2
_
2
k
log2
k
_
< 6log2
_
x
log x
_
for x > 4. Checking the result for 2 x 4 completes the proof.
3 Bertrands Postulate
In 1845 Bertrand found that for 1 n 10
6
there was always a prime between n and 2n. He postulated that
this always occurs. In 1850 Chebyshev proved it true for all n 1. Note that this is not trivial, in that it doesnt
occur for free just because (x)
x
log x
. Take S to be the set containing [2
3n
, 2
3n+1
] for each n. Then S does not
have this property, and it is quite a bit more dense than the set of primes.
BERTRANDS POSTULATE 5
3.1 Theorem. For all n ,

pn
p < 4
n
PROOF: By induction on n. Clearly the theorem holds for n = 1, 2. Suppose that the theorem holds for k =
1, . . . , n1, with n 3. Observe that we may restrict our attention to odd n, since if n is even then it is not prime
and

pn
p
n
=

pn1
p
n
. Take n = 2m+1 where m . Note that every prime p with m+2 p 2m+1
divides
_
2m+1
m
_
, so

m+2p2m+1
p divides
_
2m+1
m
_
. It follows that

p2m+1
p
_
2m+1
m
_

pm+1
p
_
2m+1
m
_
4
m+1
4
m
4
m+1
= 4
2m+1
since
_
2m+1
m
_
=
_
2m+1
m+1
_
and they both occur in the binomial expansion of (1 +1)
2m+1
.
Notation. If p is a prime and n is an integer and p
a
| n but p
a+1
n then we abbreviate this by p
a
n.
3.2 Lemma. If n 3 and p is a prime with
2
3
n < p n then p
_
2n
n
_
.
PROOF: Since n 3 we see that p is odd. The only multiples of p with p 2n are p and 2p. Thus pn! and
p
2
(2n)!. Hence p
_
2n
n
_
.
3.3 Theorem (Bertrands Postulate). For any n there is a prime p with n < p 2n.
PROOF: The result holds for n = 1, 2, 3. We argue by contradiction. Suppose that result is false for some integer
n 4. By Lemma 3.2 there is no prime larger than
2
3
n which divides
_
2n
n
_
. Let p be a prime with p
2
3
n and let
a
p
be the number such that p
a
p

_
2n
n
_
. As in the proof of Theorem 2.3 we see that a
p
r
p
, where r
p
is that integer
for which p
r
p
2n < p
r
p
+1
. Thus
_
2n
n
_

p
2
3
n
p
a
p
and so
_
2n
n
_
(2n)
t

p
2
3
n
p
where t is number of primes p
2
3
n for which a
p
2. Since p
a
p
2n we see that t

2n. Therefore
_
2n
n
_
(2n)

2n

p
2
3
n
p (2n)

2n
4
2
3
n
by Theorem 3.1
But
_
2n
n
_
>
4
n
2n+1
. Thus
4
n
2n+1
< (2n)

2n
4
2
3
n
, so
4
n
3
< (2n)

2n
(2n +1) < (2n)

2n+2
We can now check that the result holds when 4 n 16, so assume that n > 16. Taking logarithms,
log4
3
n < (

2n +2) log2n < 2

nlog2n < 2

nlog n
5
4
<
5
2

nlog n
Hence

n
log n
<
15
2log4
, and

n
log n
is increasing for n > e
2
. Furthermore,

1600
log1600
= 5.421. . . >
15
2log4
, so n < 1600. But
2, 3, 5, 7, 13, 23, 43, 83, 163, 317, 557, 1109, 2207 are all prime and so the result follows.
6 ANALYTIC NUMBER THEORY
The obvious question to ask now is whether we can do better. Baker and Harman proved that there is a
positive number C such that for x > C there is a prime in the interval [x, x + x
0.535
]. This is the best result
known so far. Assuming the Riemann Hypothesis we can replace 0.535 with
1
2
+. In 1930, Cramer conjectured
that
limsup
n
p
n+1
p
n
(log p
n
)
2
= 1
He arrived at this conjecture by probabilistic reasoning. We expect that p
n+1
p
n
= 2 for innitely many n. This
is known as the twin primes conjecture. The best result known on small gaps is due to Maier in 1985. He proved
that p
n+1
p
n
< (0.248. . .) log p
n
for innitely many positive integers n.
Can one prove that there are large gaps innitely often? In 1935, Erds proved that there is a positive number
c such that, for innitely many positive integers n, p
n+1
p
n
> c log n
loglog p
n
(logloglog p
n
)
2
. In 1938, Rankin added a factor
of loglogloglog p
n
. Erds offered $10,000 USD for any proof which showed one could replace the constant c by
any function tending to innity with n.
4 Asymtotic Analysis
Notation. Let f , g : and suppose that g > 0. We write f = O(g) whenever there are c
1
, c
2
> 0 such that if
x > c
1
then | f (x)| < c
2
g(x). We write f = o(g) if lim
x
f (x)
g(x)
= 0 We write f g, pronouced f is asymtotic to
g if lim
x
f (x)
g(x)
= 1
Recall that one of the aims of this course is to prove the Prime Number Theorem, which states that
(x)
x
log x
or, equivalently (x) =
x
log x
+o(
x
log x
). Let > 0. Notice that
(1 +)x
log x +log(1 +)
=
(1 +)x
log x
_
1 +
log(1+)
log x
_ = (1 +)
x
log x
+O
_
x
(log x)
2
_
= (1 +)
x
log x
+o
_
x
log x
_
since 1 y + y
2
>
1
1+y
> 1 y for all | y| < 1. Then
((1 +)x) (x) =
(1 +)x
log(1 +)x

x
log x
+o
_
x
log x
_
=
x
log x
+o
_
x
log x
_
This says that the number of primes between x and (1 +)x is about
x
log x
, which is a much stronger result than
Bertrands Postulate. Taking = 1 we have (2x) (x) =
x
log x
+o(
x
log x
). This seems to suggest that is linear,
but this is not the case.
For any integer n, let
(n) :=
_
log p if n is a positive power of p
0 otherwise
We dene
(x) :=

px
log p
and
(x) :=

p
m
x
log p =

nx
(x)
ASYMTOTIC ANALYSIS 7
Observe that (x) =

px
_
log x
log p
_
log p. Notice that (x) =

px
log p x log x and thus
(x) = (x) +(

x) +(
3

x) + +(
k

x) where k =
_
log x
log2
_
(x) +

x log x + +
k

x log x
(x) +(

x log x)
log x
log2
= (x) +O(

x(log x)
2
)
By Theorem 2.3, there is a c
1
> 0 such that (x) < c
1
x for x 2, since (x) =

px
log x (x) log x.
Therefore there is a c
2
> 0 such that (x) < c
2
x for x 2. Furthermore, the proof of Theorem 2.3 shows
nlog2 = log2
n
log
_
2n
n
_

p2n
_
log2n
log p
_
log p = (2n)
for n . Therefore there is a c
3
> 0 such that (x) > c
3
x for x 2. As a consequence of this there is a c
4
> 0
such that (x) > c
4
x for x 2.
4.1 Theorem.
(x) log x (x) (x)
PROOF: (x) (x) since (x) c
4
x for x 2 and (x) = (x) +O(

x(log x)
2
). It remains to show that
(x) log x (x). Clearly (x)
(x)
log x
. Let 1 > > 0. Then
(x)

x
1
<px
log p

x
1
<px
(1 ) log x
((x) (x
1
))(1 ) log x
((x) x
1
)(1 ) log x
(x) + x
1
(1 ) log x (1 )(x) log x
Therefore
1
(x) log x
(x)

1
1
_
1 +
(1 )x
1
log x
(x)
_

1
1
+
x
1
log x
(x)
Since (x) > c
4
x for x 2 we see that
1
(x) log x
(x)

1
1
+
log x
c
4
x

Given > 0, choose > 0 so that

1
1
< 1 +

2
and choose x
0
(, ) so that for x > x
0
(, ),
log x
c
4
x
<

2
. Then
1
(x) log x
(x)
1 +
and the result follows.
8 ANALYTIC NUMBER THEORY
4.2 Lemma (Abels summation Formula). Let (a
n
)

n=1
be a sequence of complex numbers and let f be a con-
tinuously differentiable function from {x | x 1} to . For each x we dene A(x) =

nx
a
n
. Then

nx
a
n
f (n) = A(x) f (x)
_
x
1
A(u) f

(u)du
PROOF: Put N = [x]. Then

nx
a
n
f (n) =
N

i=1
a
i
f (i)
= A(1) f (1) +
N

i=2
(A(i) A(i 1)) f (i)
=
N1

i=1
A(i)( f (i) f (i +1)) +A(N) f (N)
= A(N) f (N)
N1

i=1
_
i+1
i
A(u) f

(u)du
Therefore

nx
a
n
f (n) = A(N) f (N)
_
N
1
A(u) f

(u)du
but
_
x
N
A(u) f

(u)du = A(x) f (x) A(N) f (N), and the result follows.
4.3 Denition. Eulers constant is dened by
:= 1
_

1
u [u]
u
2
du 0.5772. . .
Open Question: Is irrational?
4.4 Theorem.

nx
1
n
= log x + +O
_
1
x
_
PROOF: By Lemma 4.2. We take f (x) =
1
x
and a
n
= 1 for each n. Then A(x) = [x], and thus

nx
1
n
= [x]
1
x
+
_
x
1
[u]
u
2
du
=
x (x [x])
x
+
_
x
1
u (u [u])
u
2
du
= 1 +O
_
1
x
_
+
_
x
1
du
u

_
x
1
u [u]
u
2
du
= log x + +
_

x
u [u]
u
2
du +O
_
1
x
_
= log x + +O
_
1
x
_

ASYMTOTIC ANALYSIS 9
4.5 Theorem.

nx
(n)
n
= log x +O(1)
PROOF: Take a
n
= 1 and f (x) = log x in Lemma 4.2. Then

nx
log n = [x] log x
_
x
1
[u]
u
du
= (x (x [x])) log x
_
x
1
u ([u] u)
u
du
= x log x +O(log x) (x 1) +
_
x
1
[u] u
u
du
= x log x x +O(log x) +O
__
x
1
1
u
du
_
= x log x x +O(log x)
Note that log[x]! =

nx
log n, so this is a weak form of Stirlings formula. But
log[x]! =

px
__
x
p
_
+
_
x
p
2
_
+
_
log p
=

px
_
_
log x
log p
_

m=1
_
x
p
m
_
_
log p
=

px
_
x
n
_
(n)
=

nx
_
x
n

_
x
n

_
x
n
___
(n)
= x

nx
x
(n)
n
O((x))
= x

nx
x
(n)
n
O(x)
Thus from the rst equation,
x log x x +O(log x) = x

nx
(n)
n
O(x)
x log x +O(x) = x

nx
(n)
n
log x +O(1) =

nx
(n)
n

10 ANALYTIC NUMBER THEORY
4.6 Theorem.

px
log p
p
= log x +O(1)
PROOF:

nx
(n)
n

px
log p
p
=

p
m
x
m2
log p
p
m
Thus by Theorem 4.5 it sufces to show that the right hand side is O(1). But
1
p
2
+
1
p
3
+ =
1
p(p1)
and so

p
m
x
m2
log p
p
m

2nx
log n
n(n 1)

n=2
log n
n(n 1)
<
and the result follows.
4.7 Theorem. There is a number B
1
such that

px
1
p
= loglog x +B
1
+o(1)
PROOF: Take
a
n
=
_
log p
p
if n = p
0 otherwise
and f (x) =
1
log x
. By Lemma 4.2

px
1
p
=
_

px
log p
p
_
1
log x
+
_
x
1
_

pu
log p
p
_
1
u(logu)
2
du
= 1 +O
_
1
log x
_
+
_
x
2
A(u)
u(logu)
2
du by Theorem 4.6
= 1 +O
_
1
log x
_
+
_
x
2
logu +(u)
u(logu)
2
du where (u) = O(1), by Lemma 4.2
= 1 +O
_
1
log x
_
+
_
x
2
1
ulogu
du +
_
x
2
(u)
u(logu)
2
du
= loglog x +1 loglog2 +O
_
1
log x
_
+
_
x
2
(u)
u(logu)
2
du
= loglog x +B
1

_

x
(u)
u(logu)
2
du +O
_
1
log x
_
B
1
:= 1 loglog2 +
_

2
(u)
u(logu)
2
du
= loglog x +B
1
+o(1)
B
1
can be calculated to about 0.261447. . . It can be shown that B
1
= +

p
_
log
_
1
1
p
_
+
1
p
_
.
RIEMANNS ZETA FUNCTION 11
5 Riemanns Zeta Function
The Riemann zeta function, (s) is a function of a complex variable s. It is dened for (s) > 1 by
(s) =

n=1
1
n
s
Since

n=1
1
n
s
is uniformly convergent on compact subsets of {z | (z) > 1} we deduce that (s) is an
analytic function on (s) > 1. We also note that (s) may be represented by an Euler product in the same region.
Observe that

p
_
1
1
p
s
_
1
=

p
_
1 +
1
p
s
+
1
p
2s
+
1
p
3s
+
_
=

n=1
1
n
s
by the Fundemental Theorem of Arithmetic. Since (s) is represented by a convergent innite product for
(s) > 1, we conclude that (s) = 0 for (s) > 1.
Notation. For the remainder of this section, s denotes a complex number and we put s = +i t, with , t .
5.1 Theorem. (s) can be analytically continued to (s) > 0, with the exception of a simple pole at s = 1 with
residue 1.
PROOF: Let s with (s) > 1. We apply Abel summation with a
n
= 1 and f (x) =
1
x
s
. Then

ny
1
n
s
=
[y]
y
s
+s
_
y
1
[u]
u
s+1
du
Letting y we nd that
(s) = s
_

1
[u]
u
s+1
du
= s
_

1
u (u [u])
u
s+1
du
= s
_

1
1
u
s
du s
_

1
u [u]
u
s+1
du
=
s
s 1
s
_

1
u [u]
u
s+1
du
The rst term in the sum is analytic on , except at the point s = 1 where it has a simple pole of residue 1. The
integral in the second term is uniformly convergent on compact subsets of (s) > 0. Furthermore, it represents
an analytic function on (s) > 0. Thus (s) may be analytically continued to (s) > 0, with the exception of the
simple pole at s = 1.
(s) can be analytically continued to all of , except for s = 1. There is a functional eqution which relates the
behavior of (s) with the behavior of (1 s). The region {s | 0 (s) 1} is known as the critical strip,
and information on the zeroes of (s) in the critical strip can be translated into information on the distribution
of prime numbers. We shall deduce the prime number theorem from the fact that (s) is not zero for (s) = 1.
5.2 Conjecture (Riemann Hypothesis). All of the zeroes of (s) in the critical strip satisfy (s) =
1
2
.
12 ANALYTIC NUMBER THEORY
5.3 Theorem. (s) is non-zero for (s) 1.
PROOF: We have already seen from the Euler product representation of (s) that (s) = 0 for (s) > 1. Thus we
may restrict our attention to (s) = 1. If (s) > 1 then we have
(s) =

p
_
1
1
p
s
_
1
Recall s = +i t, and for > 1 we have
log

(+i t) =

p
log
_
1
1
p
+i t
_
where log indicates the principal branch and log

indicates some branch of the logarithm. By the power series

expansion for log,
log

(+i t) =

n=1
p
n(+i t)
n
Recall for z \ {0} we can write z = |z|e
i
for some 0 < 2 and logz = log|z| +i +2ki for some k .
So (logz) = log|z|. Observe that p
int
= e
int log p
= cos(nt log p) +i sin(nt log p). Therefore
log|(+i t)| =

n=1
p
n
n
cos(nt log p)
Note that 0 2(1 +cos )
2
= 2 +4cos +2cos
2
= 3 +4cos +cos 2. Thus

n=1
p
n
n
(3 +4cos(nt log p) +cos 2(nt log p)) 0
and so
3log|()| +4log|(+i t)| +log|(+i2t)| 0
which implies
|()|
3
|(+i t)|
4
|(+i2t)| 1
Suppose that 1 +i t
0
is a zero of (s). Since (s) has a pole at s = 1, we see that t
0
= 0. Then there exist positive
numbers c
1
, c
2
and c
3
such that:
1. |()(1)| c
1
for (1, 2], since the pole at s = 1 is simple.
2. |(+i t
0
)(1)
1
| c
2
for (1, 2], since 1 +i t
0
is a zero of (s).
3. |(+i2t
0
)| c
3
for (1, 2], since the only pole of (s) for (s) > 0 is s = 1.
Thus
|()(1)|
3
|(+i t
0
)(1)
1
|
4
|(+i2t
0
)| c
3
1
c
4
2
c
3
and hence
1 |()|
3
|(+i t)|
4
|(+i2t)| c
3
1
c
4
2
c
3
|1|
for every (1, 2], a contradiction. Therefore (s) has no zeros for (s) 1.
5.4 Theorem (D.J. Newman). Suppose that a
n
with |a
n
| 1 for n . The series

n=1
a
n
n
s
converges to an
analytic function F(s) for (s) > 1. If F(s) can be analytically continued to (s) 1 then the series converges to
F(s) for (s) 1,
RIEMANNS ZETA FUNCTION 13
PROOF: Let with () = 1. Then F(z +) is analytic for (z) 0. Let R > 1 and choose = (R) 1 so
that F(z +) is analytic on the region {x | (z) , |z| R}. This region is compact, so let M = M(R, )
be the maximum value of F(z +) on this region. Let be the contour obtained by traversing the boundry of
the region in a counterclockwise direction. Let A be the part of in the right half plane and let B be the rest of
. By Cauchys residue theorem,
2i F() =
_

F(z +)
1
z
dz =
_

F(z +)N
z
_
1
z
+
z
R
2
_
dz
Note that on A, F(z +) is represented by the series

n=1
a
n
n
z+
and we split the series into two parts. Let N
and put S
N
(z +) =

N
n=1
a
n
n
z+
and R
N
(z +) = F(z +) S
N
(z +). Again by Cauchys residue theorem,
2iS
N
() =
_
|z|=R
S
N
(z +)
1
z
dz =
_
|z|=R
S
N
(z +)N
z
_
1
z
+
z
R
2
_
dz
Let C be the contour |z| = R traversed in the counterclockwise direction and let A be the open right semicircle of
radius R, so that C = AA{iR} {iR}.
2iS
N
() =
_
A
S
N
(z +)N
z
_
1
z
+
z
R
2
_
dz +
_
A
S
N
(z +)N
z
_
1
z
+
z
R
2
_
dz
=
_
A
(S
N
(z +)N
z
+S
N
(z)N
z
)
_
1
z
+
z
R
2
_
dz
Thus 2i(F() S
N
()) =
_
B
F(z +)N
z
_
1
z
+
z
R
2
_
dz +
_
A
(R
N
(z +)N
z
S
N
(z)N
z
)
_
1
z
+
z
R
2
_
dz.
We will now prove that S
N
() F() as N . Note that for z = x +i y,
1. If z A we have
1
z
+
z
R
2
=
z
zz
+
z
R
2
=
z+z
R
2
=
2x
R
2
.
2. |R(z +)|

n=N+1
1
n
x+1

_

N
1
u
x+1
du =
1
xN
x
.
3. |S
N
(z)|

N
n=1
n
x1
N
x1
+
_
N
0
u
x1
du = N
x1
+
N
x
x
= N
x
_
1
N
+
1
x
_
.
Therefore
_
_
_
_
_
_
A
(R
N
(z +)N
z
S
N
(z)N
z
)
_
1
z
+
z
R
2
_
dz
_
_
_
_
_

_
A
__
_
_
_
1
x
_
_
_
_
+
_
_
_
_
1
N
+
1
x
_
_
_
_
_
_
_
_
2x
R
2
_
_
_
_
_
dz

_
A
4
R
2
+
2
NR
dz
R
_
4
R
2
+
2
NR
_
=
4
R
+
2
N
14 ANALYTIC NUMBER THEORY
And
_
_
_
_
_
_
B
F(z +)N
z
_
1
z
+
z
R
2
_
dz
_
_
_
_
_

_
B
|F(z +)||N
z
|
_
_
_
_
1
z
+
z
R
2
_
_
_
_
dz

_
B
MN
x
_
_
_
_
1
z
+
z
R
2
_
_
_
_
dz

_
R
R
MN

d y +2
_
0

MN
x
2x
R
2
(2dx)

4RM
N

+
8M
R
2
_
0

|x|N
x
dx

4RM
N

+
8M
R
2
1
log N
Since, on the line segment (z) = , |z| R, we have |
1
z
+
z
R
2
|
1

(1 +
1
R
)
2

. Therefore
|2i(F() S
N
())|
4
R
+
2
N
+
4RM
N

+
8M
R
2
log N
which goes to zero for any xed R as N .
6 Proof of the Prime Number Theorem
Recall the Mbius function dened by
(n) =
_
_
_
1 if n = 1
(1)
k
if n is squarefree and k is the number of distinct prime factors of n
0 otherwise
6.1 Lemma.

k|n
(k) =
_
1 if n = 1
0 if n > 1
PROOF: Plainly,

k|1
(k) = (1) = 1. Suppose that n has r distinct prime factors p
1
, . . . , p
r
. Then

k|n
(k) =

k|p
1
...p
r
= 1 r +
_
r
2
_

r
3
+ +(1)
r
= (1 1)
r
= 0

6.2 Theorem (Mbius Inversion).

1. Let f :
+
and dene F :
+
by F(x) =

nx
f (
x
n
). Then
f (x) =

nx
(n)F
_
x
n
_
PROOF OF THE PRIME NUMBER THEOREM 15
2. Let f :
+
and dene F :
+
by F(n) =

d|n
f (d) =

d|n
f (
n
d
). Then
f (n) =

d|n
(d)F
_
n
d
_
PROOF: By Lemma 6.1,
f (x) =

nx
_

k|n
(k)
_
f
_
x
n
_

klx
(k) f
_
x
kl
_
=

kx
(k)
_

l
x
k
f
_
x
kl
_
_
=

kx
(k)F
_
x
k
_
For the second part,
f (n) =

c|n
_

d|
n
c
(d)
_
f (c) =

cd|n
(d) f (c) =

d|n
(d)

c|
n
d
f (c) =

d|n
(d)F
_
n
d
_

6.3 Theorem.

n=1
(n)
n
= 0
PROOF: For (s) > 1,

n=1
(n)
n
s
represents an analytic function. But note that for (s) > 1,

n=1
(n)
n
s
=

p
_
1
1
p
s
_
=
1
(s)
By Theorems 5.1 and 5.3, (s 1)(s) is a non-zero analytic function for (s) 1. Hence
1
(s1)(s)
is analytic and
non-zero for (s) 1, so
1
(s)
is analytic and non-zero for (s) 1, s = 1. Thus by Newmans theorem,
1
(s)
is
represented by

n=1
(n)
n
s
for (s) 1. In particular, since lim
s1
1
(s)
= 0 we see that

n=1
(n)
n
s
= 0.
6.4 Theorem.

nx
(n) = o(x)
PROOF: We apply Abel summation with a
n
=
(n)
n
and f (x) = x. Then A(u) =

nu
(n)
n
= o(1) by Theorem 6.3.
Thus

nx
(n) = A(x)x
_
x
1
A(u)du = o(x) o(x) = o(x)

6.5 Denition. For any n let d(n) denote the number of postive divisors of n.
6.6 Theorem.
n

m=1
d(m) =
n

m=1
_
n
m
_
= nlog n +(2 1)n +O(

n)
where is Eulers constant.
16 ANALYTIC NUMBER THEORY
PROOF: Consider the hyperbola x y = n, and let D
n
be the region with x 0, y 0, and x y n. The integer
points (a, b) in D
n
correspond to the divisors of ab n. Thus

n
m=1
d(m) =

n
m=1
_
n
m
_
. It remains to evaluate

n
m=1
_
n
m
_
. Notice that the number of integer points in the region D
n
that lie above the line y = x is the same
as the number of points below. Thus
n

m=1
d(n) = [

n]+2
[

n]

m=1
_
n
m
_
[m] = O(

n)+2
[

n]

m=1
_
n
m
m
_
+O(

n) = 2n
[

n]

m=1
1
m
[

n]([

n]+1)+O(

n)
By Theorem 4.4,

n]
m=1
1
m
= log[

n] + +O(
1

n
), so
2n
[

n]

m=1
1
m
[

n]([

n] +1) +O(

n) = 2nlog(

n {

n}) +2n +O(

n) (

n {

n})(

n +1 {

n})
= 2nlog(

n {

n}) +(2 1)n +O(

n)
= nlog n +(2 1)n +O(

n)
Fill in the detail.
6.7 Theorem (Prime Number Theorem).
(x)
x
log x
PROOF: By Theorem 4.1 it sufces to show that (x) x. Put
F(x) =

nx
_

_
x
n
_

_
x
n
_
+2
_
By Mbius inversion, Theorem6.2, (x)[x]+2 =

nx
(n)F(
x
n
), and so it sufces to showthat

nx
(n)F(
x
n
) =
o(x). Now

nx

_
x
n
_
=

nx

m
x
n
(m)
=

mx
(m)

m
x
m
1
=

mx
(m)
_
x
m
_
=

px
(log p)
__
x
p
_
+
_
x
p
2
_
+
_
= log([x]!) = x log x x +O(log x)
By Theorem 6.6,

nx
_
[x]
n
_
= [x] log[x] +(2 1)[x] +O(

x)
Notice that

nx
_
[x]
n
_

nx
_
x
n
_

nx+1
_
[x] +1
n
_
GENERALIZING (x) 17
Therefore

nx
_
x
n
_
= x log x +(2 1)x +O(

x), and so F(x) =

nx
_

_
x
n
_

_
x
n
_
+2
_
= O(

x). Let
c > 0 be such that |F(x)| < c

x for all x 1. Let t 2 be a real number. Then

_
_
_
_

n
x
t
(n)F
_
x
n
_
_
_
_
_

n
x
t
_
_
_
_
F
_
x
n
_
_
_
_
_

n
x
t
c
_
x
n
c

n
x
t
1

n
x

x
_
x
t
= c
x

t
Since F is a step function with jumps only at integer points, F(x) = F([x]) for all x 1. We see that

x
t
<nx
(n)F
_
x
n
_
= F(1)

x
2
<nx
(n) + F(2)

x
3
<n
x
2
(n) + + F([t])

x
t
<n
x
[t]
(n)
Therefore
_
_
_
_

x
t
<nx
(n)F
_
x
n
_
_
_
_
_
(|F(1) + +|F([t])|)
__
_
_
_

x
2
<nx
(n)
_
_
_
_
+ +
_
_
_
_

x
t
<n
x
[t]
(n)
_
_
_
_
_
(|F(1) + +|F([t])|)o(x)
since

nx
(n) = o(x). Given > 0 choose t so large that
c

t
<

2
and then take x sufciently large that
|

x
t
<nx
(n)F
_
x
n
_
| <

2
x, so that |

nx
(n)F
_
x
n
_
| < x. The proof is complete.
Li(x) =
_
x
2
du
logu
is a better approximation to (x) than
x
log x
. In fact, (x) = Li(x) +O(x exp(c(log x)
3
5
)) has
been proved. Littlewood proved that there are c
1
, c
2
> 0 such that for innitely many integers x, (x) Li(x) >
c
1

x
log x
logloglog x and Li(x) (x) > c
2

x
log x
logloglog x. Initial calculation suggests that Li(x) > (x) for all
x. Skewes in 1955 showed that there exists x
0
for which (x
0
) > Li(x
0
) with x
0
< 10
10
10
964
. In 1966 Lehmann
lowered the bound to 10
1166
. Probably Li(x) > (x) for x < 10
20
.
7 Generalizing (x)
7.1 Denition. For any positive integer n, let (n) denote the number of prime factors of n, counted with
multiplicity. Let (n) denote the number of distinct prime factors of n. For each k let
k
(x) denote the
number of positive integers n x with (n) = k. Let
k
(x) denote the number of positive integers n x for
which (n) = (n) = k.
Note that (x) =
1
(x) =
1
(x).
7.2 Theorem (Landau, 1900). Let k . Then

k
(x)
k
(x)
1
(k 1)!
x
log x
(loglog x)
k1
PROOF: Well prove the result by induction on k. The case k = 1 is the Prime Number Theorem. We introduce
the functions L
k
(x),
k
(x), and
k
(x), which are dened by
L
k
(x) =

p
1
p
k
x
1
p
1
p
k

k
(x) =

p
1
p
k
x
1
k
(x) =

p
1
p
k
x
log(p
1
p
k
)
18 ANALYTIC NUMBER THEORY
The star indicates that the sum is taken over all k-tuples (p
1
, , p
k
) of primes with p
1
p
k
x. Observe that
more than one such k-tuple may be associated with the same product. For each integer n let c
n
= c
n
(k) denote
the number of k-tuples whose product is n. Thus if (n) = (n) = k then c
n
= k! since n is a product of k distinct
primes. Similarily, if (n) = k then c
n
k!, and if (n) = k then c
n
= 0. It follows that
k
(x) =

nx
c
n
and

k
(x) =

nx
c
n
log n. Further observe that
k!
k
(x)
k
(x) k!
k
(x)
For k 2, the number of integers n x for which (n) = k and (n) = k is
k
(x)
k
(x). In particular,

k
(x)
k
(x)

p
1
p
k
x
p
i
=p
j
for some
(i, j) with i=j
1
_
k
2
_

q
1
q
k1
x
1 =
_
k
2
_

k1
(x)
Thus it sufces to show that
k
(x) k
x
log x
(loglog x)
k1
. For xed k 2, apply Abel summation with a
n
= c
n
and f (x) = log x. Then

k
(x) =

nx
c
n
log n =
_

nx
c
n
_
log x
_
x
1

nu
c
n
u
du
=
k
(x) log x
_
x
1

k
(u)
u
du
=
k
(x) log x +O(x) since
k
(u) k!u
Observe that
1
(x) = (x), and so by Theorem 4.1 and the Prime Number Theorem,
1
(x) x. We shall now
assume that k 2 and that
j
(x) j x(loglog x)
j1
for 1 j < k. This is our induction hypothesis.
By Theorem 4.7,

px
1
p
loglog x. Since
_

px
1
k
1
p
_
k
L
k
(x)
_

px
1
p
_
k
we get that L
k
(x) (loglog x)
k
. We can write
k
and L
k
in terms of
k1
and L
k1
, respectively:
(k 1)
k
(x) =

p
1
p
k
x
(k 1) log(p
1
p
k
)
=

p
1
p
k
x
log(p
1
p
k1
) +log(p
1
p
k2
p
k
) + +log(p
2
p
k
)
= k

p
1
p
k
x
log(p
2
p
k
)
= k

p
1
x

_
x
p
1
_
and
L
k
(x) =

p
1
p
k
x
1
p
1
p
k
=

p
1
x
1
p
1
L
k1
_
x
p
1
_
GENERALIZING (x) 19
By our induction hypothesis,
k1
(x) (k 1)x L
k2
(x) = o(x(loglog x)
k2
). Given any > 0, there is x
0
=
x
0
(k, ) such that |
k1
(x) (k 1)x L
k2
(x)| < x(loglog x)
k2
for all x > x
0
. Let c be such that |
k1
(x)
(k 1)x L
k2
(x)| < c for all x x
0
. Then
|
k
(x) kx L
k1
(x)| =
k
k 1
_
_
_
_

px

k1
_
x
p
_
(k 1)
x
p
L
k2
_
x
p
_
_
_
_
_
2

px
_
_
_
_

k1
_
x
p
_
(k 1)
x
p
L
k2
_
x
p
_
_
_
_
_
2

x
x
0
<px
c +2

p
x
x
0

x
p
_
loglog
x
p
_
k2
2xc +2x(loglog x)
k2

p
x
x
0
1
p
2xc +4x(loglog x)
k1
by Theorem 4.7
5x(loglog x)
k1
for x sufciently large
Since > 0 was arbitrary, this completes the proof by induction.
7.3 Theorem.

nx
(n) = x loglog x +B
1
x +O
_
x
log x
_

nx
(n) = x loglog x +B
2
x +o(x)
where B
2
= B
1
+

p
1
p(p1)
.
PROOF: See online notes.
Let N and let A {1, . . . , N}. Plainly,
|A|
N
is a measure of the density or thickness of A in {1, . . . , N}. We
extend this notion to subsets A of . For each integer N we denote by A(N) the set A{1, . . . , N}. We dene the
upper density of A, denoted d(A), by limsup
N
|A(N)|
N
. The lower density of A, d(A), is dened as liminf
N
|A(N)|
N
.
If d(A) = d(A) then we put d(A) to be this quantity and say that A has asymtotic density d(A).
7.4 Example. 1. Note that the even integers have density
1
2
.
2. Put A = {n | 10
2k1
n < 10
2k
, k = 1, 2, . . .}. Note that
A(10
2k1
)
10
2k1

10
2k2
10
2k1
=
1
10
, while
A(10
2k
)
10
2k

10
2k
10
2k1
10
2k
=
9
10
. Hence A does not have an asymtotic density.
Let f , F : . We say that f has normal order F if for each > 0 the set A() = {n | (1 )F(n) <
f (n) < (1 +)F(n)} has density 1. Note that if we put B() = \ A() then d(B()) =0.
7.5 Example. 1. Let f (x) = (x) and F(x) =
x
log x
. By the Prime Number Theorem, (x) has normal order
x
log x
.
2. For any function f , it has normal order of itself.
20 ANALYTIC NUMBER THEORY
7.6 Theorem (Turan). Let > 0. The number of integers n x for which
|(n) loglog n| > (loglog n)
1
2
+
is o(x) and the number of integers n x for which
|(n) loglog n| > (loglog n)
1
2
+
is o(x).
PROOF: This proof is incomplete.
7.7 Theorem. Let > 0. Then
2
(1) loglog n
< d(n) < 2
(1+) loglog n
on a set of postive integers with asymtotic density one.
PROOF: Exercise.
8 Law of Quadratic Reciprocity
8.1 Denition. For any n , let (n) denote the number of invertible equivalence classes in the ring /n.
is known as Eulers totient function.
8.2 Theorem (Euler). Let a and n be positvie integers with gcd(a, n) = 1. Then
a
(n)
1 (mod n)
PROOF: Let c
1
, . . . , c
(n)
be all of the intvertible elements modulo n. Then since a is invertible, the collection
ac
1
, . . . , ac
(n)
is also all of the invertible elements modulo n. Therefore c
1
c
(n)
(ac
1
) (ac
(n)
(mod n), so
a
(n)
1 (mod n).
The special case of Eulers Theorem where n is prime is known as Fermats Little Theorem.
8.3 Theorem (Wilson). If p is prime then (p 1)! 1 (mod p).
PROOF: By Fermats Little Theorem x
p1
1 factors as (x 1) (x (p 1)) in (/p)[x]. Therefore, by
comparing constant coefcients, 1 (1)
p1
(p 1)! (mod p). The result follows.
Wilsons Theorem was conjectured by Wilson (1741-1793). He communicated the conjecture to Waring
(1734-1798), who pbulished it in 1770. Shortly afterwards, Lagrange gave the rst proof. In fact, Leibniz had
conjectured the result in 1682. Here is a proof due to Stern in 1860. For |x| < 1,
log
_
1
1 x
_
= log(1 x) = x +
x
2
2
+
x
3
3
+
and so
exp
_
x +
x
2
2
+
x
3
3
+
_
=
1
1 x
= 1 + x + x
2
+
LAW OF QUADRATIC RECIPROCITY 21
But
exp
_
x +
x
2
2
+
x
3
3
+
_
= exp(x) exp
_
x
2
2
_
exp
_
x
3
3
_

= 1 + x +
_
1
2!
+
1
2
_
x
2
+
_
1
3!
+
1
2
+
1
3
_
x
3
+ +
_
1
p!
+ +
1
p
_
x
p
+
In particular, the coefcient of x
p
can be written
1
p!
+
r
s
+
1
p
, where
r
s
is in lowest terms and s is coprime with p.
Comparing coefcients in the power series shows that
1 =
1
p!
+
r
s
+
1
p
s r =
s
p!
+
s
p
(s r)(p 1)! =
s((p 1)! +1)
p
Now (s r)(p 1)! is an integer, so p | s((p 1)! +1). But gcd(s, p) = 1, so p | (p 1)! +1, as required.
8.4 Denition. Let p be a prime and a an integer coprime with p. The Legendre symbol is dened to be
_
a
p
_
=
_
1 if x
2
a (mod p) has a solution
1 otherwise
If (
a
p
) = 1 then we say that a is a quadratic residue modulo p, otherwise we say that a is a quadratic nonresidue
modulo p.
8.5 Theorem (Eulers Criterion). Let p be an odd prime and let a be an integer coprime with p. Then
a
p1
2
(
a
p
) (mod p)
PROOF: The congruence x
2
a (mod p) has a most 2 solutions modulo p since /p is a eld. Suppose that it
has a solution x = b. Then
a
p1
2
b
p1
2
1 (
a
p
) (mod p)
On the other hand, suppose that there is no such solution. We may partition /p into pairs (r, s) such that
rs a (mod p). Then by Wilsons Theorem, 1 (p 1)! = a
p1
2
(mod p), as required.
Let us extend the denition of the Legendre symbol (
a
p
) to include the case where p | a. In this case, dene
(
a
p
) = 0.
8.6 Theorem. Let p be an odd prime and let a and b be integers. Then
_
a
p
__
b
p
_
=
_
ab
p
_
and (
1
p
) = (1)
p(p1)
2
.
22 ANALYTIC NUMBER THEORY
PROOF: The second part holds by Eulers Criterion since (1)
p
= 1. If p | ab then the rst part clearly holds.
Suppose that p ab. By Eulers Criterion
_
ab
p
_
(ab)
p1
2
a
p1
2
b
p1
2

_
a
p
__
b
p
_
(mod p)
Since p is an odd prime and the Legendre symbols take on values in {1, 1}, the result follows.
8.7 Theorem (Gau Lemma). Let p be an odd prime and a an integer coprime with p. Let be the number of
integers from a, 2a, . . . ,
p1
2
a whose residue modulo p of least absolute value is negative. Then (
a
p
) = (1)

.
PROOF: Replace the numbers a, 2a, . . . ,
p1
2
a be their residues of least absolute value, say by r
1
, . . . , r p1
2

and
s
1
, . . . , s

, where the r
i
s and s
j
s are positive. Plainly, the r
i
s are all distinct and the s
j
s are all distinct.
Suppose that r
i
= s
j
for some i and j. Then m
1
a r
i
(mod p) and m
2
s
j
(mod p) for distinct integers
1 m
1
, m
2

p1
2
. But then (m
1
+m
2
)a 0 (mod p), and since p a, p | m
1
+m
2
. But 2 m
1
+m
2
p 1, a
contradiction. Therefore the r
i
s and s
j
s are all distinct, so they are a rearrangement of the numbers 1, . . . ,
p1
2
.
Accordingly, a(2a) (
p1
2
a) (
p1
2
)!(1)

(mod p), so a
p1
2
(1)

(mod p). Then (

a
p
) = (1)

by Eulers
Criterion.
8.8 Corollary. Let p be an odd prime. Then (
2
p
) = (1)
p
2
1
8
.
PROOF: By Gau Lemma, (
2
p
) = (1)

, where is the number of the rst

p1
2
multiples of 2 which lie in
the range [
p
2
, p 1]. We now check what happens when p 1, 3, 5, 7 (mod 8) in turn. If p = 8k + 1 then
=
p1
2
[
p
4
] = 2k. If p = 8k +3 or 8k +5 then by the same formula = 2k +1. Finally, if p = 8k +7 then
= 2k +2. The result follows.
8.9 Proposition. Let p be an odd prime and a an integer coprime with 2p. Then (
a
p
) = (1)
t
where t =

p1
2
j=1
_
ja
p
_
.
PROOF: We let r
1
, . . . , r p1
2

and s
1
, . . . , s

, as before, be the residues of least absolute value modulo p of the

integers a, 2a, . . . ,
p1
2
a, where the r
i
s and s
j
s are positive. Notice that if 1 j
p1
2
then ja = p[
ja
p
] +
j
,
where 0
j
< p. So
j
is either r
k
for some 1 k
p1
2
or is it p s
k
for some 1 k . Thus
p1
2

j=1
ja =
p1
2

j=1
p
_
ja
p
_
+
j
= p
p1
2

j=1
_
ja
p
_
+
p1
2

j=1

j
But
p1
2

j=1

j
= r
1
+ + r p1
2

+ p(s
1
+ +s

)
= (r
1
+ + r p1
2

+s
1
+ +s

) + p2(s
1
+ +s

)
= (1 +2 + +
p 1
2
) + p2(s
1
+ +s

)
LAW OF QUADRATIC RECIPROCITY 23
Therefore
p1
2

j=1
ja
p1
2

j=1
j = p
p1
2

j=1
_
ja
p
_
+ p2(s
1
+ +s

)
(a 1)
p1
2
p+1
2
2
= p
p1
2

j=1
_
ja
p
_
+ p2(s
1
+ +s

)
Since a is odd, a 1 is even, so 0 p+ p

p1
2
j=1
[
ja
p
] (mod 2), so

p1
2
j=1
[
ja
p
] (mod 2), as required.
8.10 Theorem (Law of Quadratic Reciprocity). If p and q are distinct primes then
_
p
q
__
q
p
_
= (1)
(
p1
2
)(
q1
2
)
Euler stated the law. Legendre attempted to prove it. Gau gave 8 proofs.
PROOF: By Gau Lemma, (
q
p
) = (1)

and (
p
q
) = (1)

, where and are the number of integers from

{q, 2q, . . . ,
p1
2
q} and {p, 2p, . . . ,
q1
2
p}, respectively, whose residue modulo p and q, respectively, of least absolute
value is negative. It sufces to show that + (
p1
2
)(
q1
2
) (mod 2). Given x with 1 x
p1
2
, let y be such
that
p
2
< qx py <
p
2
. Notice that
1
2

q
p
x < y <
1
2

q
p
x, so y is uniquely determined. Then qx py is the
residue of qx modulo p of least absolute value. y is non-negative, and if y = 0 then there is no contribution to
since qx 0. Further, if x =
p1
2
then
y <
q
p
x +
1
2
=
q
2
_
p 1
p
_
+
1
2
Therefore y
q1
2
since it is an integer. It follows that corresponds to the number of combinations of x and y
from the sequences 1, 2, . . . ,
p1
2
and 1, 2, . . . ,
q1
2
, repectively, such that
p
2
< qx py < 0. Similarily, is the
number of combinations such that
q
2
< py qx < 0. For any pair (x, y) with 1 x
p1
2
and 1 y
q1
2
,
either py qx <
q
2
or py qx >
p
2
. Let be the number of pairs for which the former holds and be the
number of pairs for which the latter holds. Then
_
p 1
2
__
q 1
2
_
= + + +
As x and y run through their respective domains, x

=
p+1
2
x and y

=
q+1
2
y fun through the same domains,
but in reverse order. Notice that py qx >
p
2
if and only if py

qx

=
pq
2
(py qx) <
q
2
. By symmetry,
py qx <
q
2
if and only if py

qx

>
p
2
. Therefore = and so + (
p1
2
)(
q1
2
) (mod 2), and the result
is proven.
8.11 Example. Let k . The equation y
2
= x
3
+ k is known as Mordells equation. Here we are looking for
solutions in integers x and y. There are only nitely many solutions in the integers for any xed k. In general,
it is not easy to nd all solutions However, in about 1970, Harold Stark proved that for each > 0 there exists a
positive number c() > 0, such that if (x, y) is a solution in the integers, then |x|, | y| < e
c()k
1+
.
For some k, all solutions can be found by congruence considerations. For example, consider the equation
y
2
= x
3
+ 45. Note that if x is even then y
2
45 (mod 8), so y
2
5 (mod 8), which is not possible. Thus
24 ANALYTIC NUMBER THEORY
it y
2
= x
3
+ 45 has a solution in the integers then x is odd. We consider the four possiblilties x 1, 3, 5, 7
(mod 8). Suppose that x 1 (mod 8) or x 5 (mod 8). Then x
3
1 (mod 4) and so y
2
2 (mod 4), which
is impossible. Suppose now that x 7 (mod 8). We have y
2
18 = x
3
+27 = (x +3)(x
2
3x +9). We claim
that there is a prime p 3 (mod 8) such that p | x
2
3x +9. This is so since if all primes dividing x
2
3x +9
were congruent to 1 modulo 8 then x
2
3x +9 would also be equivalent to 1 modulo 8, which it is not. Now
consider the equation modulo p. We nd that y
2
18 (mod p), or equivalently, (
18
p
) = 1. But (
18
p
) = (
23
2
p
) =
(
2
p
) = 1, a contradiction. Therefore x 3 (mod 8). Note that y
2
2 6
2
= x
3
27 = (x 3)(x
2
+3x +9).
Since x 3 (mod 8), x
2
+3x +9 3 (mod 8). As before, we see that x
2
+3x +9 is divisible by a prime p 3
(mod 8). It follows that y
2
= x
3
+45 has no integer solutions.
8.12 Example. How does 9997 factor? We could just factor it, but were mad keen to use the Law of Quadratic
Reciprocity. Notice that 9997 = 100
2
3. By the Law of Quadratic Reciprocity, if p is odd and p | 9997 then
100
2
3 (mod p), so
1 =
_
3
p
_
=
_
p
3
_
(1)
p1
2
If p 1 (mod 12) then (
p
3
)(1)
p1
2
= (
1
3
) = 1. If p 5 (mod 12) then (
p
3
)(1)
p1
2
= 1, so this case is
impossible. If p 7 (mod 12) then (
p
3
)(1)
p1
2
= 1, and if p 11 (mod 1)2 then (
p
3
)(1)
p1
2
= 1. Therefore
if p | 9997 then p 1 (mod 1)2. We now test 11, 13,. . . In fact, 9997 = 13 769. The same argument shows
that the primes dividing 769 are 1 modulo 12. Clearly

769 < 30, so we need only check 11, 13, 23, none of
which work. Therefore 769 is prime.
9 Dirichlets Theorem
For any pair of integers a and b not both zero, we can nd, by means of the Euclidean algorithm, integers x and
y for which ax + by = gcd(a, b).
9.1 Theorem (Chinese Remainder Theorem). Let m
1
, . . . , m
t
be pairwise coprime positive integers. Let m =
m
1
m
t
and b
1
, . . . , b
t
be any integers. The simlutaneous congruences
x b
1
(mod m
1
)
.
.
.
x b
t
(mod m
t
)
have a unique solution modulo m.
PROOF: Let n
i
=
m
m
i
for 1 i t. Then note that gcd(n
i
, m
i
) = 1, so there are integers r
i
, s
i
such that r
i
m
i
+s
i
n
i
=
1. Thus s
i
n
i
1 (mod m
i
). Put e
i
= s
i
n
i
and notice that b
i
e
i
b
i
(mod m
i
). But n
i
0 (mod m
j
) for j = i, so
b
i
e
i
0 (mod m
j
) for j = i. Let x = b
1
e
1
+ + b
t
e
t
, a solution to the simultaneous congruences.
Suppose that x
0
and x
1
are solutions to the simultaneous congruences. Then x
0
x
1
(mod m
i
) for i =
1, . . . , t. Since the m
i
s are coprime, m
1
m
t
| x
0
x
1
. In particular, x
0
x
1
(mod m).
9.2 Theorem. Let m
1
, . . . , m
t
be pairwise coprime positive integers. Let m = m
1
m
t
. The ring /m is
isomorphic to /m
1
/m
t
and the group (/m)

is isomorphic to (/m
1
)

(/m
t
)

.
PROOF: Let : /m
1
/m
t
: n (n + m
1
, . . . , n + m
t
). is a ring homomorphism. By the
Chinese Remainder Theorem, is onto. ker = {n : m
1
m
t
| n} = m. The First Isomorphism Theorem
gives us the rst result.
DIRICHLETS THEOREM 25
Let : (/m)

(/m
1
)

(/m
t
)

: (n+m) (n+m
1
, . . . , n+m
t
). Then is a well dened
group homomorphism. By the Chinese Remainder Theorem is an isomorphism.
9.3 Corollary. Let m
1
, . . . , m
t
be pairwise coprime positive integers. Let m= m
1
m
t
. Then
(m) = (m
1
) (m
t
)
PROOF: (m) = |(/m)

|.
9.4 Corollary. Let m . Then
(m) = m

p|m
_
1
1
p
_
9.5 Theorem.
n

j=1
( j) =
3

2
n
2
+O(nlog n)
PROOF:
n

j=1
( j) =
n

j=1
j

p| j
_
1
1
p
_
=
n

j=1
j

d| j
(d)
d
=

dn
d

(d)
=
n

d=1
(d)
_
n
d
_

=1
d

=
n

d=1
(d)
_
n
d
_
(
_
n
d
_
+1)
2
=
1
2
n

d=1
(d)
_
n
d
_
2
+(d)
_
n
d
_
=
1
2
n

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

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

d=1
(d)
1
d
2
+O(nlog n)
=
n
2
2

d=1
(d)
1
d
2

n
2
2

d=n+1
(d)
1
d
2
+O(nlog n)
=
n
2
2
(2)
1
+O(nlog n)
=
3

2
n
2
+O(nlog n)
26 ANALYTIC NUMBER THEORY
If p is an odd prime and then (/p

is cyclic. To prove this well need some preliminary results.

Notice that the result does not hold if p = 2, as (/8)

has order four but all of its elements have order 1 or 2.

9.6 Proposition. Let p be a prime and a positive integer. If a b (mod p

) then a
p
b
p
(mod p
+1
).
PROOF: Since a b (mod p

) there is c such that a = b +cp

. Then
a
p
= (b +cp

)
p
= b
p
+
_
p
1
_
b
p1
cp

+
_
p
2
_
b
p1
(cp

)
2
+ +(cp

)
p
This implies that a
b
b
p
(mod p
+1
).
9.7 Proposition. If p is an odd prime and 2 is an integer then for any integer a,
(1 +ap)
p
2
1 +ap
1
(mod p

)
PROOF: By induction on . The result holds for = 2, so suppose it holds for some 2. We have
(1 +ap)
p
2
1 +ap
1
(mod p

)
and so by Proposition 9.6,
(1 +ap)
p
1
(1 +ap
1
)
p
(mod p
+1
)
But
(1 +ap
1
)
p
= 1 +
_
p
1
_
ap
1
+
_
p
2
_
(ap
1
)
2
+ +(ap
1
)
p
1 +ap

(mod p
+1
)
and the result is proved since p
21
divides each term in the sum except for the rst two.
9.8 Proposition. Let . If p is an odd prime and a is an integer coprime with p then the order of 1+ap+p

in (/p

is p
1
.
PROOF: The result is immediate if = 1, so suppose that 2. By Proposition 9.7,
(1 +ap)
p
2
1 +ap
1
0 (mod p

)
and by Proposition 9.6
(1 +ap)
p
1
(1 +ap
1
)
p
(mod p
+1
)
so (1 +ap)
p
1
1 (mod p

). Therefore 1 +ap has order p

1
.
9.9 Theorem. Let be a positive integer and let p be an odd prime. Then (/p

is a cyclic group.
PROOF: The cardinality of (/p

is (p

) = p
1
(p 1), so it sufces to nd an integer of order p
1
(p 1)
modulo p

. Let g be a primitive root modulo p. We have g

p1
1 (mod p), so either g
p1
1 + ap (mod p
2
)
for some a coprime with p, or g
p1
1 (mod p
2
). In that latter case,
(g + p)
p1
= g
p1
+
p 1
1
g
p2
p + + p
p1
so (g +p)
p1
1+(p 1)g
p2
p (mod p
2
). Note that (p 1)g
p2
is copirme with p, so either g
p1
or (g +p)
p1
is congruent to 1 +ap (mod p
2
) with a coprime with p. Without lose of generality, we may suppose that g
p1

1 + ap (mod p
2
). We claim that g generates (/p

. Suppose that g has order m. Then m | (p 1)p

1
, so
m= dp
s
with d | p1 and 0 s 1. Thus g
dp
s
1 (mod p

), which implies that g

dp
s
1 (mod p), so g
d
1
(mod p) by Fermats Little Theorem. Since g is a primitive root, p 1 | d, so d = p 1. Thus m= (p 1)p
s
. But
g
p1
1 +ap (mod p
2
), thus (g
p1
)
p
t
1 +ap
t+1
(mod p
t+2
) by Proposition 9.8. Therefore s = 1 and the
result follows.
DIRICHLETS THEOREM 27
9.10 Theorem. If 2 then (/2

is cyclic. If 3 then (/2

)

= /2/2
2
. Finally, if 3 then
(/2

= {(1)
a
5
b
+2

| 0 a 1, 0 b < 2
2
}.
PROOF: Plainly, (/2)

is cyclic if = 1 or 2. Suppose that 3. We claim that for 3,

5
2
3
1 +2
1
(mod 2

)
The result holds for = 3 by inspection. Suppose that the equation holds for some 3. Then there is an integer
k so that 5
2
3
= 1 +2
1
+k2

Squaring both sides,

5
2
2
= (1 +2
1
+k2

)
2
= 1 +2

+k2
+1
+
Therefore [5
2
2
1+2

(mod 2
+1
), and the claim follows by indution. It follows that the order of 5 in (/2

is 2
2
. Next we claim that the elements (1)
a
5
b
, with 0 a 1, 0 b < 2
2
are all distinct modulo 2

.
Suppose that (1)
a
1
5
b
1
(1)
a
2
5
b
2
(mod 2

). Since 3, (1)
a
1
5
b
1
(1)
a
2
5
b
2
(mod 4), so (1)
a
1
(1)
a
2
(mod 4), so a
1
= a
2
. It follows that b
1
= b
2
since 5
b
1
b
2
1 (mod 2

) and |b
1
b
2
| < 2
2
. These 2
1
elements
are all distinct in (/2

, which has order 2

1
. Therefore (/2

= {(1)
a
5
b
+2

| 0 a 1, 0 b <
2
2
}. It is now clear that (/2

)

= /2/2
2
.
9.11 Theorem. The only positive integers m for which primitive roots exist modulo m are 1, 2, 4 and those
integers off the form p

or 2p

, with p an odd prime and a postive integer.

PROOF: Let m > 1 be an integer. Then let m = 2

0
p

1
1
p

k
k
, where
i
0 and the p
i
are odd primes. m has a
primitive root if and only if (/m)

is cyclic. By the Chinese Remainder Theorem,

(/m)

= (/2

0
)

(/p

1
1
)

(/p

k
k
)

Now (/p

i
i
)

is cyclic of order (p
i
1)p

i
1
i
and (/2

0
)

is cyclic if
0
= 1 or 2, and is ismorphic to /2
/2

0
2
for
0
> 2. Put (m) = lcm(b, (p

1
1
), . . . , (p

k
k
)), where b = (2

0
) if
0
= 1 or 2 and b =
(2

0
)
2
) if

0
> 2. The order of an element of (/m)

divides (m). The order of (/m)

is (m). Since 2 | p
i
1, we
see that (m) < (m) whenever m is divisible by more than one odd prime or by a power of 2 larger than 4.
Further, (m) < (m) if 2
2
| m and m is divisible by an odd prime. The remaining cases are m= 1, 2, 4, p

, and
2p

. In each case the corresponding (/m)

is cyclic.
The function (m) given in the proof is known as the universal exponent of m. The proof above gives us the
proof of the following theorem as well. Note that since (m) | (m), Theorem 9.12 is a strengthening of Eulers
Theorem.
9.12 Theorem. Let m be a positive integer and let a be coprime with m. Then a
(m)
1 (mod m).
Question: What is the smallest postive integer a such that a is a primitive root modulo p? Burgess proved
that a < c()p
1
4
+
, where c() is a positive number which depends on .
9.13 Theorem. If p is a prime of the form 4q +1, where q is an odd prime, then 2 is a primitive root modulo p.
PROOF: First notice that p 5 (mod 8). Let t be the order of 2 in (/p)

. Then t | p 1 = 4q, so t = 1, 2, or
4, or q, 2q, or 4q. But p = 13 or p 29, so t = 1, 2, or 4. It is enough to show that 2
2q
1 (mod p) to conclude
that t has order 4q and hence that 2 is a primitive root modulo p. Note that
2
2q
= 2
p1
2

_
2
p
_
(mod p)
by Eulers Criterion. Since p 5 (mod 8), (
2
p
) = 1 and the result follows.
28 ANALYTIC NUMBER THEORY
9.14 Theorem. Let n be a positive integer. There are innitely many primes p with p 1 (mod n).
PROOF: Let a > 2 be an integer. We dene the n
th
cyclotomic polynomial by

n
(x) :=
n

j=1
( j,n)=1
(x
j
n
)
where
n
= e
2i
n
. Then
n
(x) [x] and
n
has degree (n). Further, x
n
1 =

d|n

d
(x). If p is a prime that
divides
n
(a) then p 1 (mod n) or p | n. To see this, note that if p |
n
(a) then p | a
n
1. If p a
d
1 for any
proper divisor d of n then n is the order of a modulo p. Hence n | p 1, so p 1 (mod n). Suppose now that
p | a
d
1 for some proper divisor d of n. Since p |
n
(a), we see that p |
a
n
1
a
d
1
. Observe that
a
n
= (1 +(a
d
1))
n
d
= 1 +
n
d
(a
d
1) +
_
n
d
2
_
(a
d
1)
2
+
so
a
n
1
a
d
1
=
n
d
+
_
n
d
2
_
(a
d
1) ++
_
n
d
3
_
(a
d
1)
2
+
But this implies that p |
n
d
, so p | n.
Observe that if p |
n
(na) then p n, so p 1 (mod n). Suppose there are only nitely many primes
congruent to 1 modulo n, say p
1
, . . . , p
k
. Then
n
(np
1
p
k
a) is only composed of primes congruent to 1 modulo
n and is coprime with p
1
, . . . , p
k
. Thus |
n
(np
1
p
k
a)| = 1 for all a, which is a contradiction.
9.1 Characters
In order to prove that for each pair of coprime integers a and b with b > 0 that there are innitely many primes
congruent to a modulo b we need to introduce characters.
9.15 Denition. Let G be a nite Abelian group. A character of G is a homomorphism : G

. The set of
characters of G is called the dual group of G, denoted

G.
The dual group truely is a group under pointwise multiplication. The identity element is the trivial homomor-
phism. Observe that (G) , and in fact (G) is a collection of |G|
th
roots of unity, since (g)
|G|
= (g
|G|
) =
(1) = 1, for all g G.
9.16 Theorem. Let G be a nite Abelian group. Then
1. |G| = |

G|.
2. G and

G are isomorphic.
3.

G
(g) =
_
|G| if g = e
0 otherwise
4.

gG
(g) =
_
|G| if = 1
0 otherwise
PROOF: Recall that since G is a nite Abelian group it is a direct product of cyclic groups. In particular, there are
elements g
1
, . . . , g
r
G and positive integers h
1
, . . . , h
r
such that every element g G has a unique representation
of the form g = g
t
1
1
g
t
r
r
, where 0 t
i
< h
i
for i = 1, . . . , r. Note that |G| = h
1
h
r
. Here g
0
1
g
0
r
is the identity
of the group and g
i
has order h
i
for i = 1, . . . , r. A character

G is completely determined by its action on
DIRICHLETS THEOREM 29
g
i
, for i = 1, . . . , r. But ((g
i
))
h
i
= (g
h
i
i
) = (e) = 1, so (g
i
) is an h
th
i
root of unity. Accordingly, there are
at most h
1
h
r
different characters. But there are at least that many different characters since if
i
is an h
th
i
root of unity then the map (g) = (g
t
1
1
g
t
r
r
) = ((g
1
))
t
1
((g
r
))
t
r
=
t
1
1

t
r
r
is a character of G. Thus
|G| = |

G|. Dene : G

G by (g) = (g
t
1
1
g
t
r
r
) =
t
1
1

t
r
r
, where
i
: G

: g
i
e
2i
h
i
and
i
(g
j
) = 1
for i = j. Then is a group isomorphism.
Clear

G
(e) = |

G| = |G| =

gG
1(g). Suppose that g = e. Then there is
1


G such that
1
(g) = 1.
Futher, the map
1
is a bijection. Therefore

G
(g) =

1
(g)(g) =
1
(g)

G
(g)
so

G
(g) = 0 since
1
(g) = 1. The proof for the last part is analogous.
We shall be interested in characters associated with the group (/k)

, for k . Suppose that is a

character of (/k)

. We associate to a map : , dened by

(n) =
_
([n]) if gcd(k, n) = 1
0 otherwise
The map is known as a character modulo k. For any character of (/k)

, we can dene the character of

(/k)

by ([n]) = ([n]). Notice that = 1, so is the inverse of in the group

(/k)

.
9.17 Theorem. Let be a character modulo k.
1. If gcd(k, n) = 1 then (n) is a (k)
th
root of unity.
2. (nm) = (n)(m) for all n, m , so is completely multiplicative.
3. is periodic, with smallest period k.
4.

k
n=1
(k) =
_
(k) if = 1
0 otherwise
5.

k
(n) =
_
(k) if n 1 (mod k)
0 otherwise
6. Let

be a character modulo k. Then

k
n=1

(n) =
_
(k) if

=
0 otherwise
7.

k
(m)(n) =
_
(k) if m n (mod k) and gcd(m, k) = 1
0 otherwise
PROOF: Trivial, given Theorem 9.16.
We have seen by the Chinese Remainder Theorem that the study of characters modulo k reduces to the study
of characters modulo p
a
for p prime and a . First suppose that p is odd. Let g be a primitive root modulo p
a
.
Suppose that gcd(n, p) = 1. Then there is a unique integer 1 v (p
a
) such that g
v
n (mod p
a
). For each
integer 1 b (p
a
) we dene the character

b
(n) =
_
exp
_
2i vb
(p
a
)
_
if gcd(b, p) = 1
0 otherwise
30 ANALYTIC NUMBER THEORY
We thus have (p
a
) such characters and so we have the complete collection. It remains to consider the characters
modulo 2
a
with a . If a = 1 then the only character is the principal character
0
, where
0
(n) = (n mod 2).
If a = 2 then we have the additional character

1
(n) =
_
_
_
1 if n 1 (mod 4)
1 if n 3 (mod 4)
0 if n 0 (mod 2)
If a 3 then (/2
a
)

is not cyclic. Weve shown that if n 1 (mod 2) then there are unique integers x and y
such that n (1)
x
5
y
(mod 2
a
) with 0 x 1 and 0 y < 2
a2
. For each pair (b
1
, b
2
), with 0 b
1
1 and
0 b
2
< 2
a2
, we dene the character

(b
1
,b
2
)
=
_
exp
_
i x b
1
+
i y b
2
2
a3
_
if n 1 (mod 2)
0 if n 0 (mod 2)
This gives all of the (2
a
) = 2
a1
characters modulo 2
a
. To get an explicit description of the group of characters
modulo k for k composite, we just factor k into prime powers and take the product of the associated characters
for each prime power.
9.18 Denition. Let k and let be a character modulo k. We dene the function L(s, ) for s with
(s) > 1 by
L(s, ) =

n=1
(n)
n
s
L(s, ) is known as a Dirichlet L function.
The series

n=1
(n)
n
s
is uniformly convergent on compact subsets of (s) > 1 and so it denes an analytic
function for (s) > 1. Since is completely multiplicative, L(s, ) has an Euler product representation for
(s) > 1 given by
L(s, ) =

p
_
1
(p)
p
s
_
1
9.19 Theorem. Let k be a postive integer and let be a character modulo k. The function L(s, ) can be
analytically continued to (s) > 0, with the exception of the case where is the principal character, where there
is a simple pole at s = 1 of residue
(k)
k
.
PROOF: Let A

(x) =

nx
(n) and let f (x) =
1
x
s
. Note that
A

(x) =
_
(k)
_
x
k
_
+R
0

0
(x) if =
0
R

(x) otherwise
with |R
0

0
(x)| (k) and |R

(x)| (k) by Theorem 9.17. In the principal case, A

0
(x) = (k)
x
k
+R

0
(x) with
|R

0
(x)| 2(k). By Abel summation,

nx

0
(n)
n
s
=
(k)
k
x
1s
+
R

0
(x)
x
s
+s
_
x
1
A

0
(u)
u
s+1
du
=
(k)
k
x
1s
+
R

0
(x)
x
s
+s
(k)
k
u
1s
1 s
_
x
1
+s
_
x
1
A

0
(u)
u
s+1
du
=
(k)
k
_
x
1s
+
s
s 1
x
1s

s
1 s
_
+
R

0
(x)
x
s
+s
_
x
1
A

0
(u)
u
s+1
du
DIRICHLETS THEOREM 31
If =
0
then again by Abel summation,

nx
(n)
n
s
=
R

(x)
x
s
+s
_
x
1
R

(u)
u
s+1
du
Letting x we nd that for (s) > 1,

n=1

0
(n)
n
s
=
(k)
k
s
s 1
+s
_

1
R

0
(u)
u
s+1
du and

n=1
(n)
n
s
= s
_

1
R

(u)
u
s+1
du
for =
0
. Note that the right hand sides of both of these expressions converge to an analytic function for
(s) > 0. The result follows.
9.20 Denition. Let (
n
)

n=1
be a strictly increasing sequence of positive real numbers. Let (a
n
)

n=1
be a sequence
of complex numbers. The Dirichlet series associated to (
n
)

n=1
with coefcient sequence (a
n
)

n=1
is the series

n=1
a
n
e

n
z
for z .
9.21 Theorem. If the Dirichlet series f (z) =

n=1
a
n
e

n
z
converges at z = z
0
then it converges uniformly in the
region (z z
0
) > 0 and | arg(z z
0
)| < , for any < 2.
PROOF: By replacing z by z z
0
and modifying the a
n
s, we may assume without loss of generality that z
0
= 0.
Therefore

n=1
a
n
converges. In particular, for each > 0 there is N = N() such that if , m > N and we put
A
,m
=

m
n=
a
n
then |A
,m
| < . Observe that
m

n=
a
n
e

n
z
=
m

n=
(A
,n
A
,n1
)e

n
z
= A
,m
e

m
z
+
m1

n=
A
,n
(e

n
z
e

n+1
z
)
Thus for , m> N,
_
_
_
_
m

n=
a
n
e

n
z
_
_
_
_

_
|e

m
z
| +
m1

n=
|e

n
z
e

n+1
z
|
_
We may suppose that (z) > 0. Since
m

+
we see that |e

m
z
| 1. Further,
|e

n
z
e

n+1
z
| =
_
_
_
_
z
_

n+1

n
e
tz
dt
_
_
_
_
|z|
_

n+1

n
e
t x
dt = |z|
_

e
t x
x
_

n+1

n
_

|z|
x
(e

n
x
e

n+1
x
)
where z = x +i y. Thus
_
_
_
_
m

n=
a
n
e

n
z
_
_
_
_

_
1 +
|z|
x
m1

n=
(e

n
x
e

n+1
x
)
_
=
_
1 +
|z|
x
(e

x
e

m
x
)
_
2
_
1 +
|z|
x
_
If
|z|
x
< k then the series converges uniformly. But
|z|
x
< k implies that | arg(z)| < for some depending upon k.
The result now follows.
If the Dirichlet series converges for z = z
0
then it converges uniformly on compact subsets of (z) > (z
0
)
and so it denes an analytic funciton in this halfplane.
9.22 Theorem. Let f (z) =

n=1
a

n
z
n
be a Dirichlet series with a
n
0 for all n. Let
0
and suppose that
the series converges for z =
0
. If f is analytic in a neighbourhood of
0
then there is > 0 such that the series
converges at
0
.
32 ANALYTIC NUMBER THEORY
Note by Theorem 9.21 that f converges for (z) >
0
, and we conclude that it converges for (z) >
0
.
PROOF: By translating by
0
we may assume without loss of generality that
0
= 0. Since f is analytic ina
neighbourhood of 0 and by Theorem 9.21 is analytic for (z) > 0, there is an > 0 such that f is a analytic in
the disc |z 1| 1 +2. Consider the Taylor expansion of f in this disc. For (z) > 0,
f
(m)
(z) =

n=1
(
n
)
m
a
n
e

n
z
for m 0. Hence f
(m)
(1) =

n=1
(
n
)
m
a
n
e

n
. The Taylor series of f around 1 on |z 1| 1 +2 is given by
f (z) =

m=0
f
(m)
(1)
m!
(z 1)
m
=

m=0

n=1
(
n
)
m
a
n
e

n
(z 1)
m
m!
Thus, if we take z = then
f () =

m=0

n=1

m
n
a
n
e

n
(1 +)
m
m!
Observe that since a
n
0 we may change the order of summation. Thus
f () =

n=1
a
n
e

m=0

m
n
(1 +)
m
m!
=

n=1
a
n
e

n
e

n
(1+)
=

n=1
a
n
e

n
()
So the series representation holds in this disc.
If we have a Dirichlet series

n=1
a
n
e

n
z
dening a function f (z) with a
n
0 then the only obstruction to
the series representing is a pole
1
of f on the real axis. The series will represent the function for all z with
(z) >
1
.
9.23 Theorem. Let k be a positive integer and let be a character modulo k. Then L(s, ) is non-zero for
(s) > 1. Further, if is not the priniciple character then L(1, ) = 0.
PROOF: L(s, ) has a Euler product representation for (s) > 1 given by L(s, ) =

p
(1
(p)
p
s
)
1
, and so is
non-zero for (s) > 1.
Suppose rst that is a complex character. Then is a character modulo k that is different from . From
the Euler product, for any character modulo k, we have for (s) > 1,
log

L(s, ) =

p
log
_
1
(p)
p
s
_
where log denotes the principal branch of the logarithm and log

denotes some branch of the logarithm. Then

log

L(s, ) =

a=1
(p)
a
ap
as
Let be an integer comprime with k. By Theorem 9.17

k
() log

L(s, ) =

a=1

k
()(p)
a
ap
as
= (k)

p,a
p
a
(k)
1
ap
as
(1)
DIRICHLETS THEOREM 33
In particular, we may take = 1 in (1) and exponentiate to conclude that

k
L(s, ) 1, for s and s > 1.
Suppose that is a non-prinicipal character modulo k. Then = if is not a real character. If L(1, ) = 0
then L(1, ) = L(1, ) = 0. Thus if is a complex character modulo k then there exist c
1
, c
2
, c
3
> 0 such that for
s with 1 < s 2, we have
|L(s,
0
)|
c
1
s 1
L(s, )L(s, ) = |L(s, )|
2
c
s
(s 1)
2
|L(s, )| < c
3
for =
0
. Thus, for s with 1 < s 2,
1
_
_
_
_

k
L(s, )
_
_
_
_

c
1
s 1
c
2
(s 1)
2
c
3
c
1
c
2
c
3
(s 1)
Letting s 1 we obtain a contradiction. Therefore if is a complex character then L(1, ) = 0
Suppose now that is a real character with =
0
. We introduce the function g(s) dened for (s) > 1 by
g(s) =
(s)L(s,)
(2s)
. By the Euler product representation for and L, we see that for (s) > 1,
g(s) =

p
_
1
1
p
2s
_
_
1
1
p
s
__
1
(p)
p
s
_
=

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

p
_
1 +
1
p
s
_

a=0
((p))
a
p
as
=

p
_

a=0

a
(p)
p
as

a=0

a
(p)
p
(a+1)s
_
=

p
_
1 +

a=1

a
(p) +
a1
(p)
p
as
_
Since is a real character, b(a, p) :=
a
(p)+
a1
(p) is either 0 or 2. Accordingly, g(s) =

n=1
a
n
n
s
, where the a
n
s
are non-negative real numbers and where a
1
= 1. Recall that is non-principal, so g(s) is analytic for (s) > 1
and if L(1, ) = 0 then (s)L(s, ) is analytic for (s) > 0 since the simple pole of at s = 1 is cancelled by the
zero. Therefore if L(1, ) = 0 then g(s) is anaytic for (s) >
1
2
, since (2s) is non-zero for (s) >
1
2
. We now
apply Theorems 9.21 and 9.22 to conclude that the series

n=1
a
n
n
s
converges to g(s) for (s) >
1
2
. Since (2s)
has a simple pole at s =
1
2
, we see that g(s) tends to 0 as s tends to
1
2
from above on the real line. But a
1
= 1 and
so g(s) does not tend to 0 as s tends to
1
2
from above. Therefore L(1, ) = 0.
Proving that L(1, ) = 0 is what is needed to prove that whenever is an integer coprime with k that there
are innitely many primes p with p (mod k).
9.24 Theorem (Dirichlets Theorem). Let and k be coprime integers with k 2. The series

p (k)
1
p
is
divergent, and so in particular, there are innitely many primes p with p (mod k).
PROOF: Recall from the proof of Theorem 9.23, that
1
(k)

k
() log L(s, ) =

a=1

p
a
(k)
1
ap
as
34 ANALYTIC NUMBER THEORY
We now dene E() by
E() =
_
1 if =
0
0 otherwise
Note that as s tends to 1 from above on the real axis, (s 1)
E()
L(s, ) is bounded. Therefore, on the interval
(1, 2), E() log(s 1) +log L(s, ) is also bounded since L(1, ) = 0. Therefore there is a positive number c
1
,
which depends on k, such that
_
_
_
_
1
(k)

k
() log L(s, ) +
1
(k)
log(s 1)
_
_
_
_
< c
1
for s (1, 2). Accordingly,
_
_
_
_

a=1

p
a
(k)
1
ap
as
+
1
(k)
log(s 1)
_
_
_
_
< c
1
for s (1, 2). We have

a=1

p
a
(k)
1
ap
as
=

p (k)
1
p
s
+

a=2

p
a
(k)
1
ap
as
Note that for s (1, 2)

a=2

p
a
(k)
1
ap
as

a=2

n=2
1
an
as

a=2

n=2
1
2n
as

1
2

n=2

a=2
1
n
as

1
2

n=2

a=2
1
n
2s
_
1
1
1
n
_

n=2
1
n
2
<

2
6
Thus
_
_
_
_

p (k)
1
p
s
+
1
(k)
log(s 1)
_
_
_
_
< c
1
+

2
6
for s (1, 2). But as s tends to 1 from above on (1, 2) we see that
1
(k)
log(s 1) tends to . Therefore

p (k)
1
p
diverges.
Suppose that and k are coprime integers with k 2. Let (x, k, ) denote the number of primes p with
p x for which p (mod k). Then it can be proved that
(x, k, )
1
(k)
x
log x

1
(k)
Li(x)
DIRICHLETS THEOREM 35
Let H be a positive real number. It can be proved that if k (log x)
H
then
(x, k, ) =
Li(x)
(k)
+O(x exp(a
_
log x))
for a a positive real number. On the other hand, with no constraint it can be shown that
(x, k, ) =
Li
(k)
+O(
x
(log x)
H
)
However, the big-O constants depend on H in an ineffective way. In other words, one cannot compute then in
general.
Given
1
and
2
coprime with k 2, with
1

2
(mod k), we have (x, k,
1
) (x, k,
2
). Chebyshev
noted that for small x, (x, 3, 1) < (x, 3, 2) and (x, 4, 1) < (x, 4, 3). In 1957 Leech found the smallest x for
which (x, 4, 1) exceeds (x, 4, 3), and it is 26861. Bays and Hudson found the smallest x such that (x, 3, 1)
exceeds (x, 3, 2), and it is 608981813029.