Math 782: Analytic Number Theory

(Instructor’s Notes)*

Analytic Versus Elementary:

• Terminology (Analytic Number Theory makes use of Complex Analysis and Elemen-
tary Number Theory does not; but it isn’t so simple to distinguish.)
• Writing an integer as a sum of two squares. This is the first of a few examples of
how Complex Analysis can be used to answer a question seemingly unrelated to it: if a
and b are sums of 2 squares, why must ab be also?
• Coverings. A covering of the integers is a system of congruences x ≡ aj (mod mj )
such that every integer satisfies at least one of these congruences. A perfect covering is one
in which each integer satisfies exactly one of the congruences. Suppose we have a finite
perfect covering of the integers with each modulus mj > 1. Show that the largest modulus
occurs twice. Use that
1 Xr
z aj
= for |z| < 1,
1−z j=1
1 − z mj

where 1 < m1 ≤ m2 ≤ · · · ≤ mr and 0 ≤ aj < mj . Suppose that mr−1 6= mr . Then the

right-hand side, and not the left-hand side, gets large (in absolute value) as z approaches
exp(2πi/mr ). The contradiction implies mr−1 = mr .
• Preliminaries on exponentials. The following are useful formulas:

X
n−1 
(i2πk/n)j 0 if k 6= 0
e =
j=0
n if k = 0,
Z Z 
1
i2πkθ 1 2π
ikθ 0 if k 6= 0
e dθ = e dθ =
0 2π 0 1 if k = 0.

• Putnam Problem A-6, 1985. If p(x) = a0 + a1 x + · · · + am xm is a polynomial with

real coefficients aj , then set Γ(p(x)) = a20 + a21 + · · · + a2m . Let f (x) = 3x2 + 7x + 2.
Find, with proof, a polynomial g(x) with real coefficients such that (i) g(0) = 1, and (ii)
Γ(f (x)n ) = Γ(g(x)n ) for all positive integers n.
There are perhaps better ways to do this problem, but we use what we just learned to
establish Z 1



Γ(p(x)) = p e2πiθ p e−2πiθ dθ.
0

Note that f (x) = (3x + 1)(x + 2). That g(x) = 6x2 + 5x + 1 = (3x + 1)(2x + 1) provides
a solution follows from
Z 1

n
n
n
Γ (f (x) ) = f e2πiθ f e−2πiθ dθ
0

*These notes are from a course taught by Michael Filaseta in the Fall of 1996.

1
2

Z 1
n
n
n −2πiθ
n
= 3e2πiθ + 1 e2πiθ + 2 3e−2πiθ + 1 e + 2 dθ
0

Z 1
n
n
n
n
= 3e2πiθ + 1 2e2πiθ + 1 3e−2πiθ + 1 2e−2πiθ + 1 dθ
0

Z 1
n
n
= g e2πiθ g e−2πiθ dθ
0

= Γ (g(x)n ) .

• A Geometric Problem. Let R be a rectangle, and suppose R can be expressed as a

union of rectangles Rj with edges parallel to R and common points only along these edges.
Suppose each Rj has at least one edge of integer length. Then R itself must have an edge
of integer length.
Let R be positioned so that it is in the first quadrant of the xy-plane, with an edge on
each of the x and y axes. Let Rj have bottom left-hand corner (uj , vj ), and let αj and βj
be the horizontal and vertical dimensions of Rj respectively. Observe that
Z w+γ
1 2πiγ
e2πit dt = e −1 =0 ⇐⇒ γ ∈ Z.
w 2π

One uses that one of αj and βj is an integer for each j and that
Z Z r Z Z
X
2πi(x+y)
e dx dy = e2πi(x+y) dx dy
R j=1 R
j

to obtain the desired result.

Homework:
(1) (a) If a and b are integers which can each be expressed in the form x2 + 5y 2 for some
integers x and y, explain why it is possible to express ab in this form as well.
(b) Determine whether the following is true: if a and b are integers and ab can be expressed
in the form x2 + 5y 2 with x and y integers, then either a or b can be as well. Either prove
the result or give a counterexample.
(2) (Putnam Problem A-5, 1985) For m a positive integer, define
Z 2π
Im = cos x cos(2x) cos(3x) · · · cos(mx) dx.
0

(a) Use one of the preliminary results above (yes, you must to get credit for the problem)
to determine with proof for which integers m with 1 ≤ m ≤ 10 we have Im 6= 0. (Hint:
Use that cos x = (eix + e−ix )/2.)
 3

(b) Generalize part (a). In other words, determine which positive integers m are such
that Im 6= 0. The answer should be in the form: Im 6= 0 if and only if m ≡ u (mod v)
where v is an integer and u is a list of integers. Justify your answer.

Elementary Aspects:
• The Fundamental Theorem of Arithmetic (the atoms of which the matter called
integers is made are “primes”)
n
• There are infinitely many primes Q (Give three proofs: (i) Euclid’s, (ii) 22 + 1 is
2 −1
divisible by a prime > 2 n+1
, and (iii) p (1 − 1/p ) = π 2 /6, an irrational number.)
• Explain the notion of rearranging the terms of a series as well as the following lemmas
about absolute convergence.
P∞
Lemma 1. If the terms of an absolutely converging series n=1 an with an ∈ C are
rearranged, P∞ then the value of the series remains unchanged. On the other hand, if the
series n=1 an with an ∈ R converges but not absolutely, then any real value can be
obtained from an appropriate rearrangement of the series.
P∞
Use the example of the alternating harmonic series n=1 (−1)n+1 /n to illustrate the
second Phalf of the lemma
P∞ (and to give the first half more meaning). For the first half, let
∞ 0
S
P= n=1 an and let n=1 an be a rearrangement of it. Let ε > 0. Fix n0 such that
n>n0 |an | < ε. Let N be sufficiently large so that the terms a1 , . . . , an0 appear among
a1 , . . . , a0N . Then
0

X
N X
a0n − S ≤ |an | < ε.
n=1 n>n0

The first part of the lemma follows.

Lemma 2. The product of two absolutely converging series converges absolutely.
P∞ P∞ P∞ Pn−1
The product of n=1 an and n=1 bn is n=2 cn where cn = k=1 ak bn−k . The result
PN
follows since n=2 |cn | is an increasing function of N bounded above by

X
N  X
N  X
∞  X
∞ 
|an | |bn | ≤ |an | |bn | .
n=1 n=1 n=1 n=1

P∞ P∞
In fact, k=1 `=1 |ak b` | converges. Lemma 2 is of interest but is not really needed in
what follows.

Homework:
(1) Let f : (Z+ )2 7→ C. We say that the iterated series

∞ X
X ∞ ∞  X
X ∞ 
f (m, n) = f (m, n)
n=1 m=1 n=1 m=1
4

P∞
converges if for each positive integer n, the series m=1 f (m, n) converges, and if

N  X
X ∞ 
lim f (m, n)
N →∞
n=1 m=1

exists. The value of this limit is called the value of the iterated series. We say that the
double series X
f (m, n)
1≤n<∞
1≤m<∞

converges if there is an S such that for every ε > 0, there exists a K = K(ε) such that if
N and M are positive integers ≥ K, then
X
N X
M 
f (m, n) − S < ε.
n=1 m=1

The number S is called the value of the doubleXseries. Suppose the double series is ab-
solutely convergent, in other words that |f (m, n)| converges. Prove that
1≤n<∞, 1≤m<∞
each of
X ∞ X
X ∞ ∞ X
X ∞
f (m, n), f (m, n), and f (m, n)
1≤n<∞ n=1 m=1 m=1 n=1
1≤m<∞

converges and that their three values are equal.

• Euler’s product identity, namely

Y 1
−1 X ∞
1
1− s = for s > 1.
p
p n=1
ns

Let P (x) denote the product above restricted to primes p ≤ x. Use Lemma 1 to show

X[x]
X∞
1 1
s
≤ P (x) ≤ s
.
n=1
n n=1
n

The identity follows.

P
• First proof that 1/p diverges.
p
P
Assume not. Fix N such that p>N 1/p < 1/2. Then
 
Y  1
−1 X 1  X 1 2 XM
1
1− 1 + + 
+ ··· ≥
p p p n=1
n
p≤N p>N p>N
 5

for any M > 0. This is a contradiction (the left-hand side is finite).

Q∞
• Logarithms and n=1 (1 − an ).
Suppose that 0 ≤ an ≤ 1/2. This condition implies
∞
X ∞
X
ak
(∗) n
≤ akn ≤ 2a2n ≤ an .
k
k=2 k=2

We use
∞
X xk
log(1 − x) = − for |x| < 1 (actually for − 1 ≤ x < 1).
k
k=1

Observe that
Y
N X
N X
N
log (1 − an ) = log(1 − an ) ≥ −2 an .
n=1 n=1 n=1
P∞ Q∞
The left-hand side is decreasing so that if n=1 an converges, then so does n=1 (1 − an ).
QN PN Q∞
The inequality
P∞ log n=1 (1 − an ) ≤ − n=1 an implies that if n=1 (1 − an ) converges,
then so does n=1 an .
 
Q 1 1
• The estimate p≤x 1 − ≤ .
p log x
Use
Y 1
−1 X [x]
1
Z x
1
1− ≥ ≥ dt = log x.
p n=1
n 1 t
p≤x

P
• Second proof that p 1/p diverges.
Q P
Let P (x) = p≤x (1 − (1/p))−1 and S(x) = p≤x 1/p. Then by (∗)

X 1
log P (x) = S(x) + E where 0≤E≤2 ≤ 2.
p2
p≤x

The result follows since P (x) ≥ log x. (Note: we have found a “good” lower bound for
P
p≤x 1/p.)

• A noteworthy
Q example. P∞
The product p (1 − (1/p)) and n=1 µ(n)/n are related (if you expand the product
and rearrange the terms, you get the sum). The value of the product is 0 and the value of
the sum is 0; however, it is considerably harder to show the latter. (It is equivalent to the
Prime Number Theorem).
• The notation π(x) and a proof that limx→∞ π(x)/x = 0.
Show that in fact π(x) < Cx/ log log x for some constant C by using the sieve of Er-
atosthenes. (Discuss sieves in general.)
6

Homework:
(1) (a) Modify the previous argument to show that almost all positive integers n have a
prime factor ≤ log n. More precisely (and moreover) show that

|{n ≤ x : n does not have a prime factor ≤ log n}|

is bounded above by Cx/ log log x for some constant C. (Hint: Most n ≤ x which do not
have a prime factor ≤ log n also do not have a prime factor ≤ (log x)/2.)
(b) Suppose w(n) is any function defined for n ∈ Z+ which tends to infinity with n. Prove
that
|{n ≤ x : n has a prime factor ≤ w(n)}|
lim = 1.
x→∞ x

(2) Give the analysis details to the second half of the proof of Lemma 1.

The Functions π(x), ϑ(x), and ψ(x), and Chebyshev’s Estimate:

X X X  log x 
• Define ϑ(x) = log p and ψ(x) = log p = log p.
m
log p
p≤x p ≤x p≤x

• The limit connection (with possibly infinite limits):

Theorem. The following hold:

ϑ(x) ψ(x) π(x) log x

lim inf = lim inf = lim inf
x→∞ x x→∞ x x→∞ x

and
ϑ(x) ψ(x) π(x) log x
lim sup = lim sup = lim sup .
x→∞ x x→∞ x x→∞ x
More generally, if xn is such that limn→∞ xn = ∞ and any of the limits limn→∞ ϑ(xn )/xn ,
limn→∞ ψ(xn )/xn , limn→∞ π(xn ) log(xn )/xn exists, then they all exist and are equal.
Comment: This theorem follows fairly quickly from the fact that limx→∞ π(x)/x = 0 and
Abel summation (to be discussed momentarily).
Proof of Theorem. We show that for every ε > 0 and for x sufficiently large (x ≥ x0 (ε)),

π(x) log x ϑ(x) ψ(x) π(x) log x

(1 − ε) + o(1) ≤ ≤ ≤ ,
x x x x

from which the theorem follows. The second inequality is obvious. The last inequality is
a consequence of

X  log x  X
ψ(x) = log p ≤ log x = π(x) log x.
log p
p≤x p≤x
 7

For the first inequality, suppose as we may that ε < 1 and take w = 1 − ε. From
X X
ϑ(x) = log p ≥ log p ≥ (w log x)(π(x) − π(xw )),
p≤x xw <p≤x

we deduce that
ϑ(x) π(x) log x w log x
≥w − 1−w ,
x x x
and the first inequality follows.

• Chebyshev’s Theorem (in a weaker form)

Theorem. If x is sufficiently large, then
x x
0.69 < π(x) < 2.78 .
log x log x

Lemma. Let n be a positive integer, and let p be a prime. Then the non-negative integer
k for which pk exactly divides n! is
     
n n n
k= + 2 + 3 + ··· .
p p p

Proof of Theorem. Let N = 2n k
n . Let kp denote the integer k such that p exactly divides
N . From the Lemma, we deduce that
∞ 
X   
2n n
kp = −2 j .
j=1
pj p

The upper limit of summation could be replaced by [(log 2n)/(log p)]. Observe that [2x] −
2[x] is 0 or 1 depending respectively on whether x = [x] +P {x} is such that {x} < 1/2 or
not. It follows that kp ≤ [(log 2n)/(log p)]. Using ψ(x) = p≤x (log x)/(log p) log p and
Q
that N = p≤2n pkp , we deduce that log N ≤ ψ(2n).

m−k+1 m

Since N is the largest coefficient in (x + 1)2n (to see this use that m k = k k−1 ),
2n 2n
we deduce that (i) N < 2 and (ii) (2n + 1)N > 2 . We are interested in (ii) for the
moment which implies a lower bound on log N . We deduce from the above that

ψ(2n) ≥ 2n log 2 − log(2n + 1).

For x > 2 and n = [x/2], we obtain

ψ(x) ≥ ψ(2n) > (x − 2) log 2 − log(x + 1).

Dividing by x, we deduce that lim inf x→∞ ψ(x)/x ≥ log 2 > 0.69. The previous theorem
now implies the lower bound in Chebyshev’s theorem.
8

Q
The upper bound can be obtained by using n<p≤2n p ≤ N and (i) to deduce that
X
ϑ(2n) − ϑ(n) = log p ≤ log N < 2n log 2.
n<p≤2n

Let x > 1 and denote by m the non-negative integer for which 2m < x ≤ 2m+1 . Letting
n = 1, 2, 4, . . . , 2m above and summing, we obtain
m 
X 



ϑ(x) ≤ ϑ(2 m+1
)= ϑ 2 j+1
−ϑ 2j
< 2m+2 log 2 < (4 log 2)x.
j=0

It follows that lim supx→∞ ϑ(x)/x ≤ 4 log 2 < 2.78. The upper bound in Chebyshev’s
theorem now follows from the previous theorem.

• Why did Chebyshev care?

Chebyshev was interested in proving Bertrand’s hypothesis that for every positive inte-
ger n, there is a prime in the interval (n, 2n]. He did this by obtaining a stronger version
of his theorem than we have stated here. He showed that
x x
0.92 < π(x) < 1.11
log x log x

for x sufficiently large. It follows that for n sufficiently large

2n n
π(2n) − π(n) > 0.92 − 1.11 >0
log(2n) log n

which establishes Bertrand’s hypothesis (when n is sufficiently large). To complete the

proof of Bertrand’s hypothesis, one can determine what “sufficiently large” means here
and then check the hypothesis directly for the n which are not sufficiently large.

Homework:
(1) (a) Using the ideas above and Chebyshev’s theorem as we have established it, find
a constant c as small as you can such that for every ε > 0 and for x sufficiently large
(depending on ε), there is a prime in the interval (x, (c + ε)x].
(b) Find a constant c0 as small as you can such that for every ε > 0 and for x sufficiently
large (depending on ε), there are at least 3 primes in the interval (x, (c0 + ε)x].
(2) (a) Show that for n sufficiently large, the nth prime is ≤ 2n log n.
(b) Show that for n sufficiently large, the nth prime is ≥ n(log n)/3.
(3) Explain why the Chebyshev estimates established here imply that there are positive
constants C1 and C2 such that
x x
C1 < π(x) < C2 for all x ≥ 2.
log x log x
 9

More Background Material:

• Abel’s Summation Formula
X
Theorem. For a function a with domain Z+ , set A(x) = a(k). Suppose f is a
k≤x
function with a continuous derivative on the interval [u, v] where 0 < u < v. Then

X Z v
a(n)f (n) = A(v)f (v) − A(u)f (u) − A(x)f 0 (x) dx.
u<n≤v u

Proof. Clearly, the left-hand side is not changed by replacing u with [u] and v with [v].
One checks that the same is true of the right-hand side. It therefore suffices to consider u
and v to be integers. Observe that a(n) = A(n) − A(n − 1). The theorem follows from

X X
v
a(n)f (n) = (A(n) − A(n − 1))f (n)
u<n≤v n=u+1

X
v−1
= A(v)f (v) − A(u)f (u + 1) − (f (n + 1) − f (n))A(n)
n=u+1

X
v−1 Z n+1
= A(v)f (v) − A(u)f (u) − A(n) f 0 (x) dx
n=u n
Z v
= A(v)f (v) − A(u)f (u) − A(x)f 0 (x) dx.
u

Homework:
(1) Let A be a set of positive integers, and let

A(x) = |{a ∈ A : a ≤ x}|.

Suppose that there is a constant c > 0 such that

X1
> c log log x for all x sufficiently large.
a
a∈A
a≤x

Prove that given any t > 0, there is an x ≥ t for which

cx
A(x) > .
2 log x

(Note that I did not say that for all x ≥ t the last inequality holds.)
10

(2) Let n denote an arbitrary positive integer. Recall that

x2 x3
(∗) ex = 1 + x + + + ··· for all x ∈ R.
2! 3!
(a) Using (∗), show that n! > (n/e)n .
(b) Let x = n + 1 in (∗) and show that the first 2n + 2 terms are each ≤ (n + 1)n /n!.
Also, show that the remaining terms sum to a number < (n + 1)n /n!.
(n + 1)n
(c) Use (b) to show that n! < (2n + 3).
en+1  n+1
n+1
(d) Modify the above to show that n! < 2 .
e
(3) (a) Let n be a positive integer. Using Abel summation, prove that
X
n Z n
[x]
log k = n log n − dx.
1 x
k=1

(b) Using (a) and the inequality [x] ≤ x, prove that n! > (n/e)n .
nn+1
(c) Do a similar argument to show that n! < n .
e
(d) Which upper bound on n! is better when n is large: the one given in (2)(d) or the
one given in (3)(c)? Explain.

• Derivatives of Sums and Sums of Derivatives

Theorem. Let {fn (x)}∞
P n=1 be a sequence of Pdifferentiable functions such that F (x) =
∞ ∞ 0
f
n=1 n (x) converges for x > x0 and G(x) = f
n=1 n (x) converges uniformly for x > x0 .
0 0
Then F (x) exists and F (x) = G(x) for x > x0 .
PN PN 0
Proof. Let ε > 0. Let FN (x) = n=1 fn (x) and GN (x) = n=1 fn (x). There is an
N0 = N0 (ε) such that if M ≥ N0 and x is any real number > x0 , then
ε
|GM (x) − G(x)| < .
4
Let N and M be ≥ N0 . Then
ε
|GN (t) − GM (t)| < for all t > x0 .
2
Fix x > x0 . Then there is a δ = δ(M, x) ∈ (0, x − x0 ) such that if |h| < δ, then

FM (x + h) − FM (x) ε
− GM (x) < .
h 4

Fix |h| < δ. Since FN − FM is differentiable on (x0 , ∞), we deduce from the Mean Value
Theorem that there is a t between x and x + h such that

(FN (x + h) − FM (x + h)) − (FN (x) − FM (x)) = h(FN0 (t) − FM

0
(t)) = h(GN (t) − GM (t)).
 11

We deduce that
FN (x + h) − FN (x) FM (x + h) − FM (x) ε
− < .
h h 2

This is true independent of the choice of N ≥ N0 so by letting N tend to infinity, we obtain

F (x + h) − F (x) FM (x + h) − FM (x) ε
− ≤ .
h h 2

Combining the first, third, and fifth of the above displayed inequalities, we obtain that

F (x + h) − F (x)
− G(x) < ε.
h

It follows that F 0 (x) exists and F 0 (x) = G(x) as desired.

Further Elementary Results:

• Comment on this material.
Following classical lines, we show shortly that if limx→∞ π(x)(log x)/x exists, then it is 1.
The approach below is of some interest in itself, but we note if the goal is simply to establish
this information about limx→∞ π(x)(log x)/x, there is an easier way. One can skip the
next two sections, and go to Merten’s formulas. Then this result on limx→∞ π(x)(log x)/x
follows as a consequence of Merten’s formulas and an application of Abel summation (see
the third homework problem below).
• The function ζ(s) and its derivative
We use Euler’s identity discussed earlier and define the Riemann zeta function to be

X∞
1 Y 1
−1
ζ(s) = = 1− s for s > 1.
n=1
ns p
p

We will also use the von Mangoldt function Λ(n) which is defined to be log p if n = pk for
some
P prime p and some positive integer k and to be 0 otherwise. In particular, ψ(x) =
n≤x Λ(n). Observe that (for s > 1)

X   ∞
XX
1 1
log ζ(s) = − log 1 − s = ms
.
p
p p m=1
mp

By taking derivatives on both sides of this first equation (and recalling the preliminary
results above), we deduce that
∞
ζ 0 (s) X p−s log p X log p X Λ(n)
− = −s
= = .
ζ(s) p
1 − p p,m
pms
n=1
n s
12

By Abel summation, we now deduce that

Z ∞
ζ 0 (s) ψ(x)
− =s dx for s > 1.
ζ(s) 1 xs+1

• If limx→∞ π(x) log x/x exists, it is 1.

π(x) log x π(x) log x
Theorem. lim inf ≤ 1 ≤ lim sup .
x→∞ x x→∞ x
Lemma. lim (s − 1)ζ(s) = 1 and lim+ (s − 1)2 ζ 0 (s) = −1.
s→1+ s→1

Proof of Lemma. For the first limit, use that

Z ∞ X∞ Z ∞
1 1 1 1 s
= dx < < 1 + dx = .
s−1 1 xs n=1
n s
1 xs s − 1

For the second use an integration by parts to obtain

X∞ Z ∞
0 log n log x 1
ζ (s) = − =− dx + E = − + E,
n=1
n s
1 x s (s − 1)2

where |E| ≤ 1.

Proof of Theorem. Let f (s) = −ζ 0 (s)/ζ(s). The lemma implies that lims→1+ (s − 1)f (s)
exists and is 1. It suffices to show

ψ(x) ψ(x)
lim inf ≤ 1 ≤ lim sup .
x→∞ x x→∞ x

Assume the first inequality is incorrect. Then there is a K > 1 and an x0 such that if
x ≥ x0 , then ψ(x)/x > K. Hence, for s > 1,
Z ∞ Z Z ∞ Z
ψ(x) x0
ψ(x) − Kx K x0
ψ(x) − Kx sK
f (s) = s dx > s dx + s dx = s dx + .
1 xs+1 1 xs+1 1 xs 1 xs+1 s−1

Multiplying through by s − 1 and letting s → 1+ , we obtain a contradiction. In the same

manner, one handles the possibility that ψ(x)/x < L < 1 for x ≥ x00 .

• Merten’s Formulas
The following results are due to Merten:
X Λ(n) X log p Z x
ψ(t)
= log x + O(1), = log x + O(1), dt = log x + O(1),
n p 1 t2
n≤x p≤x
 13

X1 Y 1

B
= log log x + A + O(1/ log x), and 1− ∼ ,
p p log x
p≤x p≤x

where A and B are constants. Discuss what these results mean (the notation).
Observe that for any positive integer m,
∞
X
m mj mm
e = > .
j=0
j! m!

For m ≥ 1, we easily deduce that

m log m ≥ log(m!) ≥ m log m − m.

Also,

[log m/ log p]  
X X m X m X mΛ(n)
log(m!) = log p = Λ(n) = + O(m)
j=0
pj n n
p≤m n≤m n≤m

P
(where we have used that ψ(m) = n≤m Λ(n) = O(m)). Dividing through by m, we
obtain
X Λ(n)
= log m + O(1).
n
n≤m

The first of Merten’s formulas now follows (consider replacing x with [x]). The second
formula follows from the first by observing
∞
XX X
log p log p
=
p j=2
pj p
p(p − 1)

is a convergent series. The third formula follows from the first by using Abel summation
(take a(n) = Λ(n) and f (t) = 1/t) and using that ψ(x) = O(x) (by Chebyshev’s estimates).
The fourth formula follows from the second by Abel summation (take a(p) = (log p)/p,
a(n) = 0 otherwise, and f (t) = 1/ log t). The fifth formula is left as a homework problem.

Homework:
(1) We showed that if {xn }∞n=1 is a sequence tending to infinity and limn→∞ ϑ(xn )/xn or
limn→∞ π(xn ) log xn /xn exists, then they both do and are equal. Show that this follows
from Abel summation (you may also use limx→∞ π(x)/x = 0, but this is not necessary).
Y 1

B
(2) Prove the formula 1− ∼ above. (Hints: Use the Maclaurin series for
p log x
p≤x
log(1 − x) as discussed earlier. You will want to take advantage of another of Merten’s
formulas.)
14

(3) Use the fourth formula of Merten above and Abel summation to give an alternative
proof of the result
π(x) log x π(x) log x
lim inf ≤ 1 ≤ lim sup .
x→∞ x x→∞ x

Complex Preliminaries:
• Analytic functions on a region Ω (non-empty, open, and connected)
The derivative of f (z) exists on Ω.
All the derivatives of f (z) exist on Ω.
The function f (x + iy) = u(x, y) + iv(x, y) is such that u and v satisfy the Cauchy-
∂u ∂v ∂u ∂v
Riemann equations in Ω: = and =− .
∂x ∂y ∂y ∂x
The function f (z) may be represented as a power series at each point in Ω.
• The Identity Theorem
If S ⊆ Ω has an accumulation point in Ω and f and g are analytic functions for which
f (z) = g(z) for all z ∈ S, then f ≡ g on Ω.
• Weierstrass’ Theorem (a version of it)
Let fx (s) denote an analytic function in Ω for each x ≥ 1. Suppose, as x approaches
infinity, fx (s) converges uniformly to f (s) on every compact subset of Ω. Then f (s) is
analytic in Ω and fx0 (s) converges uniformly to f 0 (s) on every compact subset of Ω.

Homework:
(1) Fix σ ∈ R. Let C be a compact set in the region Re(s) > σ. Prove that there is a
σ 0 > σ such that C is in the region Re(s) > σ 0 .

The Riemann Zeta Function in the Complex Plane:

• The function ζ(s) in the right-half plane
P∞
The definition ζ(s) = n=1 1/ns is well-defined for s = σ + it (σ and t real) with σ > 1
and defines an analytic function in this region (here we interpret ns as es log n where “log”
refers to the principal branch of the logarithm). It converges uniformly for σ ≥ σ0 > 1.
By Abel summation (with a(n) = 1 and f (t) = 1/ts ), we deduce that

X∞ Z ∞
1 1 {u}
=1+ −s du for σ > 1.
n=1
ns s−1 1 us+1

Z
{u}
x
By considering Weierstrass’ Theorem with fx (s) = s+1
du, we deduce that the right-
1 u
hand side is analytic for σ > 0 except for a pole with residue 1 at s = 1. By the Identity
Theorem, the right-hand side is the only possible continuation of ζ(s) to the right-half plane
(as a meromorphic function). The Riemann zeta function ζ(s) refers to the right-hand side
above when σ > 0.
 15

Y 1
−1
Euler’s identity ζ(s) = 1− s holds for σ > 1. Observe that for σ > 1,
p
p

Y 1
 Y 1

1− s ≤ 1+ σ ≤ ζ(σ).
p p
p≤x p≤x

It follows that the absolute value of the product in Euler’s identity is bounded below by
1/ζ(σ). In particular, ζ(s) is non-zero for σ > 1.

Homework:
(1) Look at our argument for Euler’s identity given for real s > 1. Explain how to modify
it to show the identity holds for σ > 1 as stated above.

• The Riemann Hypothesis (with ζ(s) defined as above, ζ(s) = 0 =⇒ σ = 1/2;

discuss some partial results that are known and implications such as on the gap problem
for primes)
• The line σ = 1
Theorem. If t 6= 0, then ζ(1 + it) 6= 0.
Lemma 1. log |z| = Re(log z).
Lemma 2. 3 + 4 cos θ + cos(2θ) = 2(1 + cos θ)2 ≥ 0.
Proof of Theorem. From Lemma 1, for σ > 1,
X X
∞  X
∞  X∞
1 an an
log |ζ(s)| = Re(log ζ(s)) = Re = Re = cos(t log n),
p m=1
mpms n=1
ns n=1
nσ

where an = 1/m if n = pm for some prime p and an = 0 otherwise. Hence,

X∞
an
log |ζ (σ)ζ (σ + it)ζ(σ + i2t)| =
3 4
(3 + 4 cos(t log n) + cos(2t log n)).
n=1
nσ

The definition of an and Lemma 2 imply that the right-hand side is ≥ 0. It follows that,
for σ > 1,
4
ζ(σ + it) 1 1
|(σ − 1)ζ(σ)|3 |ζ(σ + i2t)| = |ζ 3 (σ)ζ 4 (σ + it)ζ(σ + i2t)| ≥ .
σ−1 |σ − 1| |σ − 1|

Assume ζ(1+it) = 0 with t 6= 0. We let σ approach 1 from the right above. We have already
shown (σ − 1)ζ(σ) approaches 1 (also clear from the analytic continuation formulation of
ζ(s)). Observe that ζ(σ + it)/(σ − 1) approaches ζ 0 (1 + it). Thus, taking a limit of the
left-hand side above, we obtain |ζ 0 (1 + it)|4 |ζ(1 + i2t)|. On the right-hand side, the limit
approaches infinity. Thus, we obtain a contradiction, establishing the theorem.
16

Proof of the Prime Number Theorem:

• The Prime Number Theorem was first established independently by Hadamard and
de la Vallée Poussin in 1896. It is the following:
π(x) log x
Theorem. lim = 1.
x→∞ x

Homework:
(1) Let pn denote the nth prime. Using the Prime Number Theorem, prove that

pn
lim = 1.
n→∞ n log n

(2) Let ε > 0. Using the Prime Number Theorem, prove that there is an x0 = x0 (ε) such
that Y
e(1−ε)x < p < e(1+ε)x for all x ≥ x0 .
p≤x

(3) Let pj denote the jth prime, and suppose that a1 , a2 , . . . , am are positive integers. If
pr is the largest prime factor of a1 a2 · · · am , then we can factor each ak as

Y
r
e (k)
ak = pj j where each ej (k) ≥ 0.
j=1

The least common multiple of the integers a1 , a2 , . . . , am , denoted lcm(a1 , a2 , . . . , am ),

satisfies

Y
r
f
lcm(a1 , a2 , . . . , am ) = pj j where fj = max{ej (k) : 1 ≤ k ≤ m}.
j=1

Let c denote a constant. Using the Prime Number Theorem, prove that

lcm(1, 2, . . . , n) 0 if c > 1
lim =
n→∞ ecn ∞ if c < 1.

• For the proof of the Prime Number Theorem, we will make use of previous material
together with the following result (the Wiener-Ikehara Theorem).
Theorem. Let A(x) be a non-negative,
Z ∞ non-decreasing function of x, defined for x ∈
[0, ∞). Suppose that the integral A(x)e−xs dx converges for σ > 1 to a function f (s)
0
which is analytic for σ ≥ 1 except for a simple pole at s = 1 with residue 1. Then
lim e−x A(x) = 1.
x→∞
 17

• The connection. We show here how the Wiener-Ikahara Theorem implies the Prime
Number Theorem. Recall that we showed
Z ∞
ζ 0 (s) ψ(x)
− =s dx for s > 1.
ζ(s) 1 xs+1
For
Z xσ > 1, the right-hand side is analytic by Weierstrass’ Theorem (consider fx (s) =
ψ(t)
s s+1
dt). Since ζ(s) is a non-zero analytic function for σ > 1, we deduce that the
1 t
left-hand side is analytic for σ > 1. The above equation for real s > 1 now implies by the
Identity Theorem that the same equation holds for all complex s = σ + it with σ > 1.
Replacing x with ex in the integral, we deduce that
Z ∞
ζ 0 (s)
(∗) − = ψ(ex )e−xs dx for σ > 1.
sζ(s) 0

By our knowledge of ζ(s) on the line s = 1 + it, the left-hand side of (∗) is analytic
for σ ≥ 1 except for at s = 1. We have already seen lims→1 (s − 1)ζ(s) = 1 and that
lims→1 −(s − 1)ζ 0 (s)/ζ(s) = 1. It follows from the Wiener-Ikahara Theorem with A(x) =
ψ(ex ) that limx→∞ ψ(x)/x = 1 which as we have seen implies the Prime Number Theorem.
• Some preliminary analysis results:
Lemma 1. Let f : [a, b] × [0, ∞) → C be a continuous function. Suppose that there are
positive numbers C and ε such that |f (t, x)| ≤ Ce−εx for every pair (t, x) ∈ [a, b] × [0, ∞).
Then Z Z Z Z
b ∞ ∞ b
f (t, x) dx dt = f (t, x) dt dx.
a 0 0 a

Comment: To prove the lemma, one can consider the double integral of |f (t, x)|. This
integral is finite, and this justifies such an interchange.
Lemma 2. Let f (s) be analytic on σ = Re(s) ≥ 1 (hence, in an open region containing
such s). Let a and b be real numbers with a < b. Then as w → 0+ , f (1 + w + iy) converges
uniformly to f (1 + iy) for y ∈ [a, b].
Proof. Write f (x + iy) = u(x, y) + iv(x, y). Since f is anaytic on σ ≥ 1, each of u and v
has continuous partial derivatives for (x, y) with x ≥ 1. Let M be a bound on |ux (x, y)|
and |vx (x, y)| in the set [1, 2] × [a, b]. Then with z = 1 + iy and 0 ≤ w ≤ 1, the Mean Value
Theorem gives

|f (z + w) − f (z)| ≤ |u(1 + w, y) − u(1, y)| + |v(1 + w, y) − v(1, y)| ≤ 2wM.

The lemma follows.

Lemma 3. Let a and b be real numbers with a < b. Let h(t, w) : [a, b] × [0, ∞) → C be a
continuous function such that h(t, w) converges to h(t, 0) uniformly for t ∈ [a, b]. Then
Z b Z b
lim h(t, w) dt = h(t, 0) dt.
+ w→0 a a
18

Proof. Fix ε > 0. Choose δ > 0 such that if 0 ≤ w < δ and t ∈ [a, b], then |h(t, w) −
h(t, 0)| < ε/(b − a). Then
Z b Z b Z b
h(t, w) dt − h(t, 0) dt ≤ |h(t, w) − h(t, 0)| dt < ε.
a a a

The result follows.

Z ∞
Lemma 4. Let f : R → R be a non-negative function for which f (x) dx exists. Then
Z ∞ Z ∞ 0
−wx
lim f (x)e dx = f (x) dx.
w→0+ 0 0
Z ∞ Z t
Proof. Let ε > 0. Fix t > 0 such that f (x) dx < ε/4. Let I = f (x) dx. Fix δ > 0
t 0
ε
such that if 0 ≤ w < δ, then 1 − < e−wt ≤ 1. Then
2I + 1
Z ∞ Z ∞ Z t Z ∞
−wx ε
f (x)e dx − f (x) dx ≤ f (x) dx + 2 f (x) dx < ε.
0 0 2I + 1 0 t

The result follows.

Z ∞
Lemma 5. Let f : R → R be a non-negative function for which lim f (x)e−wx dx
w→0+
Z ∞ Z ∞ 0

exists. Then lim+ f (x)e−wx dx = f (x) dx.

w→0 0 0

Homework:
Z ∞
(1) Prove Lemma 5. (Do not assume f (x) dx exists.)
0

We will use the following weak version of the Riemann-Lebesgue Lemma.

Lemma 6. Let a and b be real numbers with a < b. Let f : [a, b] → C be such that:
(i) f 0 exists everywhere in [a, b], and
Z b
(ii) |f 0 (t)| dt is finite.
a
Z b
Then lim f (t)eiyt dt = 0.
y→∞ a
Comment: Observe that (ii) holds if f 0 is continuous on [a, b]. This will be the case for
our use of the lemma.
Proof. Integration by parts gives
Z b b Z b
eiyt 1
f (t)eiyt dt = f (t) − f 0 (t)eiyt dt.
a iy a iy a
 19

Thus,
Z Z
b
2(|f (a)| + |f (b)|) 1 b
f (t)e iyt
dt ≤ + |f 0 (t)| dt.
a y y a

The result follows.

We will also make use of
Z ∞
sin2 x
Lemma 7. 2
dx = π.
−∞ x
There is no real need to discuss the proof as the value of the integral will not come into
play; only its existence will (which is clear).
• A Proof of the Wiener-Ikehara Theorem
Set B(x) = A(x)e−x . It follows that
Z ∞
1
f (s) − = (B(x) − 1)e−(s−1)x dx for σ > 1.
s−1 0

Define g(s) to be the left-hand side. By the conditions on f , we deduce that g is analytic
for σ ≥ 1. Fix ε > 0, and let h(t) = g(1 + ε + it). Fix λ > 0. Then
Z   Z 2λ   Z ∞ 
2λ
|t| iyt |t| iyt −(ε+it)x
h(t) 1 − e dt = 1− e (B(x) − 1)e dx dt.
−2λ 2λ −2λ 2λ 0

We justify that for σ > 1 we can interchange the order of integration by using Lemma 1.
We consider s = 1 + (ε/2) and x > 0. Observe that
Z ∞ Z ∞
−us A(x)e−xs
f (s) = A(u)e du ≥ A(x) e−us du = .
0 x s

Thus, B(x) = A(x)e−x ≤ sf (s)e(ε/2)x . It follows that

 
|t| iyt
1− e (B(x) − 1)e−(ε+it)x ≤ Ce−(ε/2)x ,
2λ

where C = sf (s) + 1. Lemma 1 now justifies the interchange of the order of integration.
A direct calculation gives
Z  
1 2λ
|t| i(y−x)t sin2 (λ(y − x))
1− e dt = .
2 −2λ 2λ λ(y − x)2

We deduce that
Z   Z ∞
1 2λ
|t| iyt sin2 (λ(y − x))
(∗) h(t) 1 − e dt = (B(x) − 1)e−εx dx.
2 −2λ 2λ 0 λ(y − x)2
20

Since g(s) is analytic for σ ≥ 1, we deduce from Lemma 2 that h(t) (see its definition)
converges uniformly to g(1 + it) for t ∈ [−2λ, 2λ] as ε approaches 0 from the right. It
follows from Lemma 3 that
Z 2λ   Z 2λ  
|t| iyt |t| iyt
lim h(t) 1 − e dt = g(1 + it) 1 − e dt.
ε→0+ −2λ 2λ −2λ 2λ

By Lemma 4,
Z ∞ Z ∞ Z
2
−εx sin (λ(y − x)) sin2 (λ(y − x)) λy
sin2 v
lim e dx = dx = dv.
ε→0+ 0 λ(y − x)2 0 λ(y − x)2 −∞ v2

Since B(x) is non-negative, Lemma 5 and (∗) now give

Z ∞ Z ∞
2
−εx sin (λ(y− x)) sin2 (λ(y − x))
lim+ B(x)e dx = B(x) dx
ε→0 0 λ(y − x)2 0 λ(y − x)2
Z λy  
v sin2 v
= B y− dv.
−∞ λ v2

We let ε → 0+ in (∗). Next, we let y tend to infinity. Observe that

Z   Z 0   Z 2λ  
2λ
|t| iyt t iyt t
g(1+it) 1− e dt = g(1+it) 1+ e dt+ g(1+it) 1− eiyt dt
−2λ 2λ −2λ 2λ 0 2λ

and we can apply Lemma 6 to each of the integrals on the right. From this and Lemma 7,
we obtain
Z λy  
v sin2 v
(∗∗) lim B y− dv = π.
y→∞ −∞ λ v2

Explain why intuitively (∗∗) implies we are through. Let a be such that 0 < a < λy.
Since A(x) = B(x)ex is non-decreasing, we get for −a ≤ v ≤ a that
     
a v a
e y−(a/λ)
B y− ≤e y−(v/λ)
B y− ≤e y+(a/λ)
B y+
λ λ λ

so that      
a −2a/λ v a 2a/λ
B y− e ≤B y− ≤B y+ e .
λ λ λ
We now deduce from (∗∗) and the fact that B(x) is non-negative that
Z   Z
−2a/λ

a sin2 v a −2a/λ a sin2 v
e lim sup B(y) 2
dv = lim sup B y − e 2
dv
y→∞ −a v y→∞ λ −a v
Z a  
v sin2 v
≤ lim sup B y− dv ≤ π.
y→∞ −a λ v2
 21
√
The above holds for any positive a and λ. Letting a = λ and letting λ tend to infinity
gives that lim supx→∞ B(x) ≤ 1. To finish the proof, we show lim inf x→∞ B(x) ≥ 1.
Observe that√lim supx→∞ B(x) ≤ 1 implies B(x) ≤ c for some constant c and all x ≥ 0.
We take a = λ again and use that
Z λy  
v sin2 v
π = lim inf B y− dv
y→∞ −∞ λ v2
 Z −a Z ∞  Z a  
sin2 v sin2 v v sin2 v
≤c dv + dv + lim inf B y− dv
−∞ v2 a v2 y→∞ −a λ v2
 Z −a Z ∞  Z
sin2 v sin2 v
a sin2 v
≤c dv + dv + e 2a/λ
lim inf B(y) dv.
−∞ v2 a v2 y→∞ −a v
2

The theorem follows now by letting λ tend to infinity.

Intermission:
• Question: Are there infinitely many primes whose decimal representation contains
the digit 9? Answer: Yes.
Theorem. Given any block of digits d1 d2 . . . dn base b ≥ 2, there exist infinitely many
primes whose base b representation contains the block of digits.
We will specialize our arguments to the original question concerning the digit 9. We give
three proofs that infinitely many such primes exist. One proof will use the main result of
what is to come, one will use the main result of where we have been, and the other won’t
use much. Each of the arguments easily generalizes to give the above theorem. Before
continuing, we note that it is unknown whether or not there exist infinitely many primes
whose decimal representation does not contain the digit 9. (Actually, it is known; but no
proof exists.)
• What is to come. The next main goal for the course is to establish Dirichlet’s
theorem that if a and b are relatively prime integers with a > 0, then there are infinitely
many primes of the form an + b. Observe that with a = 10 and b = 9, we can deduce
that there are infinitely many primes whose decimal representation contains (in fact, ends
with) 9. (Note that the more general theorem stated above can be done the same way by
considering b = d1 d2 . . . dn × 10 + 1.)
• Where we have been. We could instead use the Prime Number Theorem as follows.
Lemma. Let  > 0. Then there is an x0 = x0 (ε) such that if x ≥ x0 , then the interval
(x, x + εx] contains a prime.
Before proving the lemma, observe that it implies what we want by taking ε = 1/9 and
x = 9 × 10k ≥ x0 . (In fact, a similar argument shows there are infinitely many primes
whose decimal representation begins with any given block of digits.)
Proof. We may suppose that ε ≤ 1 and do so. From the Prime Number Theorem, there is
an x00 = x00 (ε) such that if x ≥ x00 , then
   
ε x ε x
1− < π(x) < 1 + .
10 log x 10 log x
22

Thus, x ≥ x00 implies

   
ε x + εx ε x εx x log 2 ε 2x
π(x+εx)−π(x) ≥ 1− − 1+ ≥ − −
10 log(2x) 10 log x log(2x) (log x)(log(2x)) 5 log x

(where the last term on the right is a bound on the contribution from the parts involving
ε/10). We deduce that if x is sufficiently large, the interval (x, x + εx] contains a prime.

• An elementary argument. Let n1 , n2 , . . . be the positive integers in increasing order

whose decimal representations do not contain the digit 9. Then for N ≥ 1

X 1 X
N X 1 X 9j
N
= ≤ < 90.
nk nk 10j−1
nk <10N j=1 10j−1 ≤nk <10j j=1
P∞
It follows that the partial sums of k=1 1/nk form a bounded increasing sequence, and
hence the series converges. Since the sum of the reciprocals of the primes diverges, it
follows that there are infinitely many primes not among the numbers nk . Hence, there
exist infinitely many primes whose decimal representation contains the digit 9.

Homework:
(1) An open problem of Erdős is to show that if {aj }∞ j=1 is an arbitrary infinite sequence
P∞
of integers such that (i) 1 ≤ a1 < a2 < a3 < · · · and (ii) j=1 1/aj diverges, then there
exist arbitrarily long arithmetic progressions among the aj . More precisely, given a positive
integer N , one can find N distinct aj in an arithmetic progression. If true, this would imply
there are arbitrarily long arithmetic progressions among the primes, something which as
yet is unknown. One approach to resolving the problem might be to consider a sequence
{a }∞ satisfying (i) and having no N term arithmetic progression and to show that
Pj∞ j=1
j=1 1/aj must then converge. If one can show this is true regardless of the value of
N ≥ 1, then the problem of Erdős would be resolved. The case N = 1 and N = 2 are
not interesting. Deal with the following special case with N = 3. Begin with a1 = 1 and
a2 = 2. Let a3 be as small as possible so that no 3 term arithmetic progression occurs
among the aj then selected. We get a3 = 4. Choose a4 now as small as possible avoiding
again 3 term arithmetic progressions. ThenP∞ a4 = 5. Continue in this way. The next few
aj ’s are 10, 11, 13, and 14. Prove that j=1 1/aj converges. (Hint: Think base 3.)

Algebraic Preliminaries:
• A group is a set G of objects together with a binary operation ∗ satisfying:
(i) If a and b are in G, then so is a ∗ b.
(ii) If a, b, and c are in G, then (a ∗ b) ∗ c = a ∗ (b ∗ c).
(iii) There is an element e of G such that a ∗ e = e ∗ a = a for every a ∈ G.
(iv) For every a ∈ G, there is a b ∈ G such that a ∗ b = b ∗ a = e.
An abelian group is a group G such that a ∗ b = b ∗ a for all a and b in G. A group is finite
if G is finite. We will simply use ab to denote a ∗ b and refer to the binary operation as
multiplication (unless noted otherwise).
 23

• Examples. The set of integers under addtion, the set of non-zero rational numbers
under multiplication, and the reduced residue system modulo n under multiplication are
all abelian groups. Another example, is given by {ζ j : 0 ≤ j ≤ n − 1} where ζ = e2πi/n and
the binary operation is multiplication. Yet another example is given by {φ1 , φ2 , φ3 } under
composition where the φj are defined multiplicatively on the elements of {ζ, ζ 2 , ζ 4 } where
ζ = e2πi/7 with φ1 (ζ) = ζ, φ2 (ζ) = ζ 2 , and φ3 (ζ) = ζ 4 . Finally, consider the multiplicative
mappings from the reduced residue system modulo 7 to {ζ j : 0 ≤ j ≤ 5} where ζ = e2πi/6 ;
show that these mappings form a finite abelian group under multiplication.
• A simple theorem on finite groups. We will use the fact that if a is an element of a
finite abelian group G, then a|G| = e where e is the identity element in G. Give a proof.
Note that the same is true for any finite group.
• A theorem on finite abelian groups
Theorem. Let H be a subgroup of a finite abelian group G. Let a ∈ G, and let k be the
minimal positive integer for which ak ∈ H. Let

H 0 = {aj b : b ∈ H, 0 ≤ j < k}.

Then H 0 is a subgroup of G and |H 0 | = k|H|.

Proof. Let ai b and aj b0 be elements of H 0 with 0 ≤ i, j < k and b and b0 in H. If
0 ≤ i+j < k, then bb0 ∈ H implies ai aj bb0 = ai+j bb0 ∈ H 0 . If i+j ≥ k, then 0 ≤ i+j−k < k.
In this case, ak bb0 ∈ H implies ai aj bb0 = ai+j−k ak bb0 ∈ H 0 . Thus, H 0 is closed under
multiplication. One also checks that the inverse of ai b is ak−i (a−k b−1 ) ∈ H 0 (note that if
i = 0, then ai b ∈ H ⊆ H 0 ). The other group properties of H 0 are easily determined to
hold, and thus H 0 is a subgroup of G.
To prove |H 0 | = k|H|, it suffices to show that if ai b = aj b0 , then i = j. Assume
otherwise; we suppose as we may that i > j. Then ai−j = b0 b−1 ∈ H contradicts the
minimality condition on k. The contradiction completes the proof.

Characters:
• The definition. A character χ of a finite abelian group G is a multiplicative complex
valued function, not identically zero, defined on the group. Thus, if a and b are elements
of the group, then χ(ab) = χ(a)χ(b).
• Properties of characters:
(i) If e is the identity element of G and χ any character, then χ(e) = 1. This follows
from two observations. First, χ(e) 6= 0; otherwise, χ(a) = χ(a)χ(e) = 0 for all a ∈ G
contradicting the requirement in the definition that χ is not identically zero. Next, χ(e) =
χ(e)χ(e), and we can deduce from our first observation that χ(e) = 1.
(ii) If |G| = n, then χ(a) is an nth root of unity for every a ∈ G. This follows from
χ(a)n = χ(a|G| ) = χ(e) = 1.
(iii) For every a ∈ G, χ(a) 6= 0. (See (ii).)
24

(iv) If |G| = n, then there are exactly n distinct characters defined on G. Let H1 = {e}
where e is the identity element of G. Thus, H1 is a subgroup of G. If G 6= H1 , we apply
the theorem of the previous section (on groups) to construct a new subgroup H2 of G.
Note that H2 will be a cyclic subgroup of G generated by an element a different from e
in G. How a 6= e is chosen doesn’t matter. If G 6= H2 , then we again apply the theorem
to obtain a new subgroup H3 of G. We continue defining for i ≥ 1, as long as G 6= Hi ,
the subgroup Hi+1 = {aji b : b ∈ Hi , 0 ≤ j < ki } where ai 6∈ Hi and ki is the minimal
positive integer for which aki i ∈ Hi . We show by induction on i that the number of distinct
characters on Hi is |Hi |. The result will then follow.
We deduce from (i) that there is precisely one character defined on H1 = {e}. Suppose
there are precisely |Hi | different characters defined on Hi for some i ≥ 1. If G = Hi , then
we are done (as there are no more subgroups to consider). Otherwise, we wish to show
that there are precisely |Hi+1 | = ki |Hi | different characters defined on Hi+1 . Observe
that any character on Hi+1 is necessarily a character when restricted to Hi (make use of
(i) or (iii)). We finish the argument by showing that each character χ of Hi extends to
exactly ki different characters on Hi+1 . Fix χ. Observe first that the definition of Hi+1
and the multiplicativity of χ imply that χ will be defined on Hi+1 once the value of χ(ai )
is determined. On the other hand, the value of χ(ai )ki = χ(aki i ) is known (since aki i ∈ Hi ).
Call this number γ. Then χ(ai ) must be one of the ki distinct ki th roots of γ. This shows
that there are at most ki different extensions of χ to Hi+1 . On the other hand, defining
χ(ai ) to be such a ki th root of γ is easily shown to produce an extension of χ to Hi+1 .
This completes the proof.
(v) If a ∈ G and a 6= e, then there is a character χ defined on G such that χ(a) 6= 1. In
our construction of the Hi above, we take H2 = hai. The order of a given in the argument
there is k1 . The construction gives a character χ of H2 such that χ(a) is an arbitrary k1 th
root of unity. We choose a k1 th root of unity other than 1. This character extends to a
character of G with the desired property.
(vi) The characters of G form a finite abelian group under multiplication. If χ1 and
χ2 are two characters defined on G, then χ = χ1 χ2 means χ(a) = χ1 (a)χ2 (a) for every a
in G. One checks directly that χ is a multiplicative, complex valued function which is not
identically zero; hence, the product of two characters is itself a character. Associativity
and commutitivity follow from the same properties for multiplication in the set of complex
numbers. The identity element, called the principal character, is defined by χ(a) = 1 for
every a ∈ G; and this is easily checked. One also checks that if χ is a character, then so is
χ0 defined by χ0 (a) = (χ(a))−1 and that χ0 is the inverse of χ. From (iv), we now deduce
that the characters form a finite abelian group. We note that the group G and the group
of characters defined on G are isomorphic; we do not need this fact and so do not bother
with the details of an explanation.
(vii) The following holds:
X 
0 if χ is not principal
χ(a) =
a∈G
|G| if χ is principal.

Observe that in the case that χ is principal, the result is trivial. If χ is not the principal
 25

character, then there is a b ∈ G such that χ(b) 6= 1. The proof follows by multiplying the
sum, say S, by χ(b) and using that {ab : a ∈ G} = G. Thus, χ(b)S = S so that S = 0.
(viii) The following holds:
X 
0 if a 6= e
χ(a) =
χ
|G| if a = e.
If a = e, the result follows from (iv). If a 6= e, then by (v) there is a character χ such that
χ(a) 6= 1. If the sum above is S, we obtain χ(a)S = S and the result follows along the
same lines as (vii).
• Dirichlet characters. A Dirichlet character χ is a character on the reduced residue
system modulo n extended to the complete system modulo n by defining χ(a) = 0 whenever
a and n are not relatively prime.
• Examples of Dirichlet characters. Discuss the two Dirichlet characters modulo 6,
the four Dirichlet characters modulo 8, and the twelve Dirichlet characters modulo 36 by
making Dirichlet character tables in the first two cases and explaining how to go about
doing the same in the latter (by defining χ(5) to be a sixth root of unity and χ(35) to be
±1).

Homework:
(1) (a) Make a Dirichlet character table of the eight characters modulo 15.
(b) Make a Dirichlet character table of the eight characters modulo 20.

Dirichlet Series:
P
• The definition. A Dirichlet series is a series of the form ∞ s
n=1 an /n where an and
s s log n
s denote complex numbers (we interpret n as e where “log” refers to the principal
branch of the logarithm).
• Preliminary results.
P∞
Theorem 1. Fix θ ∈ (0, π/2). If the series n=1 an /ns converges for s = s0 ∈ C, then it
converges uniformly in {s : | arg(s − s0 )| ≤ θ}.
P∞ P∞
Proof. By considering bn = an n−s0 so that n=1 an /ns = n=1 P∞bn /n
s−s0
, we see that it
suffices to prove the theorem in the case that s0 = 0. Thus, n=1 an converges and its
partial
P sums form a Cauchy sequence. Fix ε > 0. Consider a positive integer M such that
| M <n≤x an | < ε for all x ≥ M . We definie a0n = an if n > M and a0n = 0 if n ≤ M . Let
P
N > M . We apply Abel summation with A(x) = n≤x a0n and f (x) = x−s to obtain
X an Z N
A(x)
(∗) = A(N )f (N ) − A(M )f (M ) + s dx.
ns M x
s+1
M <n≤N

The definition of A(x) implies that |A(x)| < ε for all x ≥ M . We write s = σ + it and
obtain
Z N Z N Z N  
A(x) A(x) 1 ε 1 1
s+1
dx ≤ dx ≤ ε dx = − σ .
M x M xs+1 M x
σ+1 σ Mσ N
26

Note that
|s| 1
1< ≤ .
σ cos θ
From (∗), we deduce easily that

X an 4|s|ε 4|s|ε 4ε
≤ ≤ ≤ .
ns σM σ σ cos θ
M <n≤N

P∞
Therefore, the partial sums of n=1 an /ns form a Cauchy sequence and hence the series
converges. The above inequality in fact implies uniform convergence.
P∞
Theorem 2. If n=1 an /ns converges for s = s0 = σ0 + it0 , then it converges in the
half-plane s = σ + it with σ > σ0 . Furthermore, in this case, the convergence is uniform
on every compact subset of the half-plane with σ > σ0 .

The proof of Theorem 2 is clear (given Theorem 1). The smallest value of σ0 is called
the abscissa of convergence for the
P∞Dirichlet series. Observe that if σ0 is the abscissa of
convergence for a Dirichlet series n=1 an /ns and if σ0 is finite, then the series converges
if σ = Re(s) > σ0 and diverges if σ < σ0 . What happens on the line σ = σ0 is unclear. It
should be noted that σ0 could be +∞ or −∞ (with, for example, an = n! and an = 1/n!,
respectively).
P∞
Theorem 3. Let σ0 be the abscissa of convergence for the Dirichlet series n=1 an /ns .
Then the series represents an analytic function in the half-plane σ > σ0 and its derivatives
can be computed by termwise differentiation.

In the complex preliminaries

P section of the notes, we introduced a theorem of Weierstrass.
By considering fx (s) = n≤x an /ns and applying Theorem 2, the above result follows.
P∞ P∞
Theorem 4. Let σ 0 ∈ R. Suppose n=1 an /ns and n=1 bn /ns both converge for σ > σ 0
and that they are equal on some non-empty open set in this half-plane. Then an = bn for
all n ≥ 1.
Proof. Let cn = an −Pbn . The conditions in the theorem, Theorem 3, and the Identity
∞
Theorem imply that n=1 cn /ns is identically 0 in the half-plane σ > σ 0 . We wish to
prove that cn = 0 for all n ≥ 1. Assume otherwise,
P∞ and let M be minimal with cM 6= 0.
0
Fix for the moment σ > σ . Observe that n=1 cn /nσ converges so that |cn |/nσ < 1 for
P∞say n ≥ σ+u
n sufficiently large, N ≥ M + 2. Now, for u > 0, |cn |/nσ+u < 1/nu for n ≥ N .
We obtain from n=M cn /n = 0 that

∞
X  NX
−1  ∞
X
|cM | |cn | 1 1 C
≤ ≤ |cn | + ≤ ,
M σ+u n σ+u (M + 1)σ+u n u (M + 1)u
n=M +1 n=M +1 n=N

for some constant C. Multiplying through by M σ+u and letting u tend to infinity gives a
contradiction and establishes the theorem.
 27

Part I of the Proof of Dirichlet’s Theorem:

X∞
χ(n)
• The Dirichlet series L(s, χ) = s
where χ denotes a character modulo a
n=1
n
positive integer m converges for σ > 0 if χ is not the P principal character. To see this,
observe that property (vii) of characters easily implies n≤x χ(n) is bounded as x goes
to infinity. An application of Abel summation now shows that L(s, χ) converges for σ > 0.
It is easy to see that L(s, χ) diverges for σ < 0. It follows that 0 is the abscissa of
convergence. In particular, L(s, χ) is analytic in the region σ > 0.
• The connection between L(s, χ) and ζ(s) when χ is the principal character.
First, for an arbitrary Dirichlet character χ, a proof similar to that done before gives

Y χ(p)
−1
L(s, χ) = 1− s for σ > 1.
p
p

For the principal character χ0 modulo m, we obtain

Y 1
Y
1
−1 Y 1

L(s, χ0 ) = 1− s 1− s = 1 − s ζ(s).
p p
p p
p|m p|m

It
Q follows that L(s, χ0 ) is analytic for σ > 0 except for a simple pole at s = 1 with residue
p|m (1 − (1/p)).
• For every Dirichlet character χ, L(s, χ) 6= 0 for σ > 1. A proof similar to that given
for the analogous result for ζ(s) can be used here.
• The logarithm of L(s, χ).
For an arbitrary Dirichlet character χ modulo m, we define
∞
XX χ(pk )
w(L(s, χ)) = for σ > 1.
p k=1
kpks

The right-hand side is what we would obtain from using the Maclaurin series for log(1 − x)
with the product formulation of L(s, χ) if we (incorrectly) assume log(z1 z2 ) = log z1 +log z2
for complex numbers z1 and z2 . Locally, w behaves like the logarithm of L(s, χ) but globally
it may not correspond to any branch of the logarithm. In particular, since for the principal
branch of the logarithm, log(1 − z) = −z − z 2 /2 − z 3 /3 − · · · for |z| < 1, it follows that
exp(−z − z 2 /2 − z 3 /3 − · · · ) = 1 − z for |z| < 1 so that
XX
∞  Y −1
χ(pk ) χ(p)
exp = 1− s for σ > 1.
kpks p
p≤x k=1 p≤x

Thus,

exp w(L(s, χ)) = L(s, χ) for σ > 1.
28

Observe that w is defined as a Dirichlet series which is analytic for σ > 1. In particular,
it is differentiable, and we have by taking derivatives above that
d L0 (s, χ)
(L(s, χ)) = for σ > 1.
ds L(s, χ)
Hence, for fixed s = σ > 1 and c > σ, we obtain
Z c 0
L (u, χ)
(∗) w(L(s, χ)) = − du + log L(c, χ).
s L(u, χ)

Homework:
(1) (a) Let χ be a character on a group G. Let χ be the function defined by χ(g) = χ(g)
(the complex conjugate of χ(g)). Prove that χ is a character on the group G.
(b) Let L(s, χ) be the Dirichlet L-series corresponding to a Dirichlet character χ modulo
a positive integer m. Let χ0 denote the principal character modulo m. Recall that L(s, χ)
is analytic for σ > 0 if χ 6= χ0 and that L(s, χ0 ) is analytic for σ > 0 except for a simple
pole at s = 1. Suppose Y
L(1, χ) 6= 0,
χ

where the product is over all Dirichlet characters χ modulo m. Also, suppose χ is a
character for which χ(a) 6= ±1 for some integer a. Explain why L(1, χ) 6= 0.

Part II of the Proof of Dirichlet’s Theorem:

• We will show in the next section that L(1, χ) 6= 0 if χ is a non-principal Dirichlet
character. We use this fact for the remainder of this section to show how Dirichlet’s
theorem on primes in an arithmetic progression follows. Observe that from (∗) above, for
χ a non-principal character, we deduce from the fact that L(s, χ) and L0 (s, χ) are analytic
for σ > 0 and L(1, χ) 6= 0 that w(L(s, χ)) remains bounded as s → 1+ .
• The proof of Dirichlet’s Theorem assuming L(1, χ) 6= 0 if χ is a non-principal Dirich-
let character.
Let a and m be integers with m > 1 (the case m = 1 is trivial) and gcd(a, m) = 1. We
prove there are infinitely many primes p ≡ a (mod m) by showing that the sum of the
reciprocals of such primes diverges. Let b ∈ Z with ab ≡ 1 (mod m). Observe that bx ≡ 1
(mod m) if and only if x ≡ a (mod m). Consider the Dirichlet characters modulo m, and
let r ≥ 1 denote the number of them. By property (viii) of characters, it follows that
X 
r if p ≡ a (mod m)
χ(pb) =
χ
0 otherwise.

We consider s = σ > 1 and define E by

X χ(p)
(∗∗) w(L(s, χ)) = +E
p
ps
 29

so that
∞
XX χ(pk )
E= .
p k=2
kpks

A simple estimate gives that |E| ≤ 1. From (∗∗), we obtain for some E 0 bounded in
absolute value by r that

X X X χ(p) X X χ(pb) X 1
χ(b)w(L(s, χ)) = χ(b) + E0 = + E0 = r + E0.
χ χ p
ps p χ
ps ps
p≡a (mod m)

We now let s → 1+ and observe that each term in the sum on the left remains bounded
except for the summand involving the principal character and this summand increases in
absolute value to infinity. This implies that the right-hand side must increase in absolute
value to infinity so that the sum of the reciprocals of the primes p ≡ a (mod m) diverges.

Part III of the Proof of Dirichlet’s Theorem:

• We begin with some preliminaries:

Lemma 1. Suppose f (z) is analytic in a region Ω containing a point z0 . Suppose the disk
D = {z : |z − z0 | < r} where r > 0 is in Ω. Then the series

f 0 (z0 ) f 00 (z0 ) f 000 (z0 )

f (z0 ) + (z − z0 ) + (z − z0 )2 + (z − z0 )3 + · · ·
1! 2! 3!

converges in D to f (z).

The lemma asserts that f (z) is equal to its Taylor series representation about z0 in any
open disk centered at z0 that is contained in Ω. We omit the proof of this lemma but
use it to establish the following result of Landau (discuss the connection with ζ(s) as an
example to help clarify the theorem).

Theorem 1. Let f (s) be analytic for σ = Re(s) > 0. Suppose that there is a Dirichlet
∞
X
series an /ns with abscissa of convergence σ0 that equals f (s) for σ > max{0, σ0 }. If
n=1
an ≥ 0 for every n ≥ 1, then σ0 ≤ 0.

Proof. Assume σ0 > 0. Since f (s) is analytic for σ > 0, f (s) can be represented as a
Taylor series about σ0 + 1 that (by Lemma 1) converges in a disk of radius > 1 + 12 σ0
centered at σ0 + 1. For k a positive integer, the Dirichlet series representation of f (s) gives
us
X∞
(k) an (− log n)k
f (σ0 + 1) = .
n=1
n1+σ0
30

Thus, inside this disk,

∞
X f (k) (σ0 + 1)
f (s) = (s − σ0 − 1)k
k!
k=0
X∞ ∞
(s − σ0 − 1)k X an (− log n)k
=
k! n=1
n1+σ0
k=0
X∞ ∞
(1 + σ0 − s)k X an (log n)k
= 1+σ0
.
k! n=1
n
k=0

Fix s = u ∈ (σ0 /2, σ0 ) (in the disk). Then the double sum converges and, since the terms
are non-negative, it converges absolutely. We may therefore rearrange the order of the
summations to obtain that
∞
X ∞
X ∞
X X∞
an (1 + σ0 − u)k (log n)k an (1+σ0 −u)(log n) an
f (u) = = e = .
n=1
n1+σ0 k! n=1
n1+σ0 n=1
n u
k=0
P∞
This is impossible as u is to the left of the abscissa of convergence for n=1 an /ns . The
theorem follows.
X∞
Lemma 2. If a Dirichlet series an /ns converges for s = s0 , then the series converges
n=1
absolutely for Re(s) > Re(s0 ) + 1.
Proof. Let σ0 = Re(s0 ), and consider s = σ +it with σ > σ0 +1. Convergence at s0 implies
that limn→∞ |an /nσ0 | = 0. Thus, P∞ |an /nσ0 | < 1 for nPsufficiently large.
P∞Since σ >σ0σ0s−σ
+ 1,
∞
it follows by comparison with n=1 1/nσ−σ0 that a
n=1 n /ns
= a
n=1 n /(n n 0
)
converges absolutely.
P∞ P∞
The product of P∞ two Dirichlet series n=1 P an /ns and n=1 bn /ns is defined to be the
s
Dirichlet series n=1 cn /n where cn = k`=n ak b` (here, k and ` represent positive
integers). If the Dirichlet series involving an and bn both converge for s = σ + it with
σ > σ0 , then they converge absolutely for σ > σ0 + 1 by Lemma 2. It follows that
the Dirichlet series representing their product will converge absolutely for σ > σ0 + 1.
The product will, therefore, converge in this same region (we view the product then as
converging in a possibly smaller region then the initial two Dirichlet series). Observe also
P∞

s k
that if we consider n=1 an /n for k any positive integer (and expanding this product
in the obvious way), it is easy to see that independent of k the resulting series converges (in
fact, absolutely) for σ > σ0 + 1. Note that the values of a series representing the product
of two Dirichlet series is in fact equal to the product of the values of the two Dirichlet
series if the series involved converge absolutely.

Homework:
P∞ d(n)
(1) Prove that ζ(s)2 = n=1 for σ > 1 where d(n) denotes the number of positive
ns
integer divisors of n.
 31

X Λ(n)∞
ζ 0 (s)
(2) Recall that we showed − = for s > 1. Explain why this implies that
ζ(s) ns
P n=1
0
d|n Λ(d) = log n. (Hint: Write down the series representation for ζ (s).)

(3) Prove that

X∞
(−1)n
s
= (21−s − 1)ζ(s) for s = σ + it with σ > 0 (and s 6= 1).
n=1
n

(Hint: It may help to consider the case σ > 1 first, but don’t forget to address the case
σ > 0.)

• We are now ready to complete the proof of Dirichlet’s Theorem. As we saw before,
we need only justify the following.
Theorem 2. If χ is not the principal character modulo an integer m > 1, then L(1, χ) 6= 0.
Proof. Recalling the definition of w(L(s, χ)) from the previous section (Part II of the Proof
of Dirichlet’s Theorem), we define for σ > 1,

X ∞
XXX χ(pk )
Q(s) = w(L(s, χ)) = ,
χ χ p k=1
kpks

where the sum on χ is over all characters modulo m. The right-hand side converges
absolutely for σ > 1 so that we can rearrange the order of summation. Let r denote the
number of characters modulo m. Then by making the inner sum be over χ and using
property (viii) of Dirichlet characters, we deduce that

X X X∞
r an
Q(s) = ks
= s
,
p
kp n=1
n
1≤k<∞
pk ≡1 (mod m)

where an = r/k if n = pk ≡ 1 (mod m) (k ∈ Z+ ) and an = 0 otherwise. This last series

converges absolutely for σ > 1. Let d be such that ud ≡ 1 (mod m) for every integer u
with gcd(u, m) = 1 (so we may take d = r = φ(m), but we needn’t be this precise). Then
P∞
an = r/d whenever n = pd and p - m. It follows that the Dirichlet series n=1 an /ns does
not converge for s = 1/d (use that the sum of the reciprocals
P∞ of the primes diverges and
that each an ≥ 0). Hence the abscissa of convergence for n=1 an /ns is in [1/d, 1].
Now, consider

Q(s)2 Q(s)3
f (s) = eQ(s) = 1 + Q(s) + + + ··· .
2! 3!
The powers of Q(s) are themselves Dirichlet series, and each power converges absolutely
for σ > 1 since each an ≥ 0. Also, a1 = 0 implies that any given 1/ns appears as a
32

term with a non-zero coefficient for only finitely many of the powers of Q(s). Since the
expression on the right-hand side above converges for σ > 1, we obtain that the terms
on the right-hand side can be rearranged to form a Dirichlet series representation for f (s)
that converges for σ > 1. This Dirichlet series will contain non-negative
P∞ coefficients and
the coefficient of 1/ns will be at least an . Since the Dirichlet series n=1 an /ns does not
converge for s = 1/d, it follows that neither does the Dirichlet series representing f (s).
Thus, this Dirichlet series has abscissa of convergence σ0 ≥ 1/d.
Recall that ew(L(s,χ)) = L(s, χ). The definition of Q(s) implies that for σ > 1,
Y
f (s) = eQ(s) = L(s, χ).
χ

Assume there is a non-principal character χ0 such that L(1, χ0 ) = 0. Then (by considering
the Taylor series for L(s, χ0 ) about 1) we deduce that L(s, χ0 ) = (s − 1)g(s) for some
analytic function g(s). Recall that if χ0 is the principal character, L(s, χ0 ) is analytic for
σ > 0 except for a simple pole at 1. It follows that L(s, χ0 )L(s, χ0 ) can be viewed as
an analytic function in the region σ > 0. Therefore, we can view f (s) as being analytic
for σ > 0. But then f (s) is an analytic function for σ > 0 which, for σ > σ0 , can be
represented by a Dirichlet series with non-negative coefficients and with a positive abscissa
of convergence. This contradicts Theorem 1, completing the proof.

Homework:
(1) A stronger version of Dirichlet’s Theorem and the Prime Number Theorem is as
follows:
Theorem. Let a and m be integers with m ≥ 1 and gcd(a, m) = 1. Let

A(x) = |{p ≤ x : p ≡ a (mod m)}|.

Then there are positive constants C1 and C2 such that for all x ≥ 2
Z x √
1 dt
A(x) − < C1 xe−C2 log x .
φ(m) 2 log t

In the theorem, φ(m) denotes the number of positive integers k ≤ m with gcd(k, m) = 1.
Z x
dt x 5x
(a) Show that − ≤ for x sufficiently large. (Suggestion: Integrate
2 log t log x log2 x
by parts and express
√ the resulting integral as a sum of two integrals, one with limits of
integration from x to x.)
√ 1
(b) Show that e−C2 log x < for x sufficiently large. (Hint: Compare the loga-
log2 x
rithms of both sides.)
(c) Using the theorem above, show that there is a constant C 0 for which

x x
A(x) − < C0 2 for x sufficiently large.
φ(m) log x log x
 33

(d) Using the theorem above or (c), show that there is a constant C 00 for which

x x
π(x) − < C 00 2 for x sufficiently large.
log x log x

(2) Using (1)(c) and Abel summation, prove that

X 1 1
= log log x + E,
p φ(m)
p≤x
p≡a (mod m)

where |E| ≤ C for some constant C.

The Pure Brun Sieve:

• A quick review. Recall that we showed that π(x) is small compared to x by estimating
the size of
A(z, x) = |{n ≤ x : p|n =⇒ p > z}|.
We began with the identity
X x X 
x

A(z, x) = [x] − + − ··· .
p p1 p2
p≤z p1 <p2 ≤z

We then obtained the estimate

Y 1

A(z, x) ≤ 1− x + 2π(z)
p
p≤z

by removing the brackets of the greatest integer function and estimating the resulting error.
This is the basic idea of the Sieve of Eratosthenes. In this section, we give an alternative
approach to bounding A(z, x) called the Pure Brun Sieve. Unlike most of the material in
this course, the approach here is elementary.
• A first estimate. To find an upper bound on A(z, x), we first prove that
X x X  x  X 
x

(∗) A(z, x) ≤ [x] − + − ··· + ,
p p1 p2 p1 p2 · · · p2k
p≤z p1 <p2 ≤z p1 <p2 <···<p2k ≤z

where k is any positive integer. For this purpose, define

X X X
αn = 1 − 1+ 1 − ··· + 1.
p≤z p1 <p2 ≤z p1 <p2 <···<p2k ≤z
p|n p1 p2 |n p1 p2 ···p2k |n

To prove (∗), it suffices to show that


= 1 if p|n =⇒ p > z
αn is
≥ 0 otherwise
34

P
(use that A(z, x) ≤ n≤x αn and interchange summations). The first part is obvious for
if every prime factor of n is > z, then all the sums in the definition of αn are empty and
only the term 1 is non-zero in this definition. Now, suppose n = pe11 pe22 · · · perr m where the
pj are distinct primes ≤ z, each of r, e1 , . . . , er are positive integers, and every prime factor
of m is > z. It follows that
     
r r r
(∗∗) αn = 1 − + − ··· + ,
1 2 2k


where we interpret ab as 0 is b > a. To show that αn ≥ 0, consider three cases: (i) r ≤ 2k,
(ii) 2k < r ≤ 4k, and (iii) r > 4k. Case (i) is dealt with by using (1 − 1)r = 0 to show
αn = 0. For Case (ii), use (1 − 1)r = 0 to obtain
     
r r r
αn = − + ··· ±
2k + 1 2k + 2 r
       
r r r r
≥ − + − + · · · ≥ 0.
2k + 1 2k + 2 2k + 3 2k + 4

For Case (iii), use (∗∗) directly to show that αn ≥ 1 (by again grouping the binomial
coefficients in pairs).
• Modifying the above. The above can be used to obtain an upper bound for π(x)
that is smaller than the estimate we made with the Sieve of Eratosthenes. However, our
real goal is to find an upper bound for

πa (x) = |{n ≤ x : n and n + a are primes }|,

where a is any fixed positive integer (which we will take to be even for obvious reasons).
We define
A0 (z, x) = |{n ≤ x : p|n(n + a) =⇒ p > z}|.
Observe that for any z ≥ 1, πa (x) ≤ A0 (z, x) + z. Thus, we seek a good estimate for
A0 (z, x). We use that
X
A0 (z, x) ≤ αn(n+a)
n≤x
X X X X X
= 1− 1+ 1
n≤x p≤z n≤x p1 <p2 ≤z n≤x
p|n(n+a) p1 p2 |n(n+a)
X X
− ··· + 1.
p1 <p2 <···<p2k ≤z n≤x
p1 p2 ···p2k |n(n+a)

We fix momentarily z ≥ a so that if p|a, then p ≤ z. For a given p ≤ z, we consider two

possibilities, p|a and p - a. If p|a, then the number of n ≤ x for which p|n(n + a) is [x/p],
which is within 1 of x/p. If p - a, then the number of n ≤ x for which p|n(n + a) is within
 35

2 of 2x/p. In general, if p1 , . . . , pu are distinct primes dividing a and pu+1 , . . . , pu+v are
distinct primes not dividing a, then the number of
n ≤ x for which n(n + a) is divisible
by p1 p2 · · · pu+v is within 2v of 2v x/ p1 p2 · · · pu+v (this can be seen by using the Chinese
Remainder Theorem and considering the number of such n in a complete system of residues
modulo p1 p2 · · · pu+v ). It follows that
Xx X 2x
A0 (z, x) ≤ x − −
p p
p≤z p≤z
p|a p-a
X x X 2x X 4x
+ + +
p1 p2 p1 p2 p1 p2
p1 <p2 ≤z p1 <p2 ≤z p1 <p2 ≤z
p1 p2 |a p1 |a,p2 -a p1 -a,p2 -a
X 22k x
+ ··· + + E1
p1 · · · p2k
p1 <p2 <···<p2k ≤z
p1 -a,...,p2k -a
Y 1
Y
2

= 1− 1− x + E 1 + E2 ,
p p
p|a p≤z
p-a

where
     
π(z) π(z) 2k π(z)
E1 ≤ 1 + 2 +4 + ··· + 2
1 2 2k
 2 2k

2 2
≤ π(z)2k 1 + 2 + + ··· + ≤ e2 π(z)2k
2! (2k)!
and
X 22k+1 x X 22k+2 x
E2 ≤ + + ··· .
p1 p2 · · · p2k+1 p1 p2 · · · p2k+2
p1 <p2 <···<p2k+1 ≤z p1 <p2 <···<p2k+2 ≤z

We now find a bound for this last expression on the right to bound E2 as follows:
X∞  u ∞
1 X2 X 1
u
E2 ≤ x ≤x 2 log log z + 2C1 ,
u! p u!
u=2k+1 p≤z u=2k+1
P∞
where C1 is some appropriate constant. Using eu = j=0 uj /j! ≥ uu /u! and choosing
k = [6 log log z], we obtain
∞
X  u ∞
X  u
2e log log z + 2eC1 1 x x
E2 ≤ x ≤x = 2k <
u 2 2 (log z)6
u=2k+1 u=2k+1

for z sufficiently large. We also have

E1 ≤ e2 π(z)2k ≤ z 12 log log z

36

for z sufficiently large. We now choose z = x1/(24 log log x) and consider x sufficiently large
Y   
to deduce that
0 1 Y 2
A (z, x) ≤ 1− 1− x + E,
p p
p|a p≤z
p-a

where |E| ≤ x/(log x)5 . Observe that, for some constants C2 and C3 depending on a,

Y Y  Y  !2
1 2 1 x
1− 1− x ≤ C2 x 1− ≤ C3 2
(log log x)2 .
p p p (log x)
p|a p≤z p≤z
p-a

Hence, we deduce the

Theorem. Let a be any fixed positive integer. Then for x sufficiently large,
x
πa (x) ≤ C (log log x)2
(log x)2
for some constant C depending on a.

• Twin primes. A twin prime is a prime which differs from another prime by 2. Thus,
the twin primes are 3, 5, 7, 11, 13, 17, 19, 29, 31, . . . . It is unknown whether or not there are
infinitely many twin primes. Brun introduced what is now called the pure Brun sieve to
establish that the sum of the reciprocals of the twin primes converges. Since π2 (x) can be
used to bound the number of twin primes up to x (π2 (x) is not precisely the number of
these twin primes, but clearly this number is ≤ 2π2 (x) for x ≥ 1), one can use the previous
theorem and Abel summation to estimate the sum of the reciprocals of the twin primes.
This shows the sum converges.

Homework:
(1) Let pn denote the nth prime.
(a) Explain why the Prime Number Theorem implies that lim sup(pn+1 − pn ) = ∞.
n→∞
(b) Recall that we showed that
x
πa (x) ≤ C (log log x)2
(log x)2
for some constant C, where

πa (x) = |{n ≤ x : n and n + a are prime}|.

Use this to prove that for every positive integer k,

lim sup min{pn+1 − pn , pn+2 − pn+1 , . . . , pn+k − pn+k−1 } = ∞.
n→∞

(Note that this would follow from part (a) if “min” were replaced by “max”; the problem
is to figure out how to handle the “min” situation.)