On the Prime Number Theorem

Alice Yang

June 5th 2020

1 Introduction
In this paper, we will discuss the prime number theorem. The proof we will
give is based on a concise version of Newman’s proof by Zagier [7]. It relies on
some properties of the Riemann zeta function, the most essential of which is the
fact that ζ(s) has no zeros on Re(s) > 1. In addition, it relies on an analytic
theorem commonly known as “Tauberian theorem”, which is of interest in its
own right.
Since this is a topic in complex analysis, the reader should expect to be
familiar with the convergence and uniform convergence of infinite series and
infinite products of real and complex numbers, complex integration, and ana-
lyticity.
We should note that there are also “elementary” proofs of the prime number
theorem obtained independently by Erdös [2] and Selberg [5] in the 1940s that
do not use the Riemann zeta function, but they are only elementary in the
sense that they do not involve complex analysis; these “elementary” proofs are
actually much more complicated than those that use complex analysis.

2 Analysis background
In this section, we will give functions, transforms, and theorems that we will be
using in our proofs for the prime number theorem. We will not prove any of the
results. Most results can be found in Gamelin’s Complex Analysis [3].

1
2.1 Riemann Zeta Function
The Riemann Zeta function is defined by
∞
X
ζ(s) = n−s ,
n=1

where the series converges absolutely and uniformly for Re(s) > 1 + ,  > 0 by
Weierstrass M-test, setting Mn = n−(1+) . Thus it defines an analytic function
on this domain since n−s is analytic for Re(s) > 1, n > 1.

2.2 Chebyshev function

X
ϑ(x) = log p.
p6x

This is also called the log-weighted prime counting function. The sum above
range over primes p: the symbol p will always denote a prime number, and
any sum or product over p is understood to be over primes even if it is not
specifically stated.
The function ϑ(x) is an increasing, piecewise continuous function of x.

2.3 Auxiliary function

X log p
Φ(s) = .
p
ps

We can easily verify that the series converges absolutely and locally uni-
formly for Re(s) > 1, thus it defines an analytic function in this domain given
log p
that s is analytic for Re(s) > 1, p prime.
p

2.4 Laplace Transform

Let h(t) be a real-valued piecewise continuous function that is defined for all
t > 0. The Laplace transform L h(s) is the complex function defined by
Z ∞
L h(s) = h(t)e−st dt.
0

The following properties of Laplace transform are easy to verify, and is left
to the reader:

• L (g + h) = L g + L h.

2
 • For any a ∈ R, L (ah) = a(L h).
a
• If h(t) = a ∈ R is constant, then L h(s) = .
s
• L (eat h(t))(s) = L h(s − a).

2.5 Complex analysis background

∞
Y
Definition 2.5.1(Convergence of infinite product). The infinite product pj
j=1
converges if lim Pj = P 6= 0, where Pj is the partial product of pj .
j→∞
If the infinite product converges and one of the pj 0 s is 0, then we define the
value of the product to be 0, otherwise, we define it to be
∞
Y ∞
X
pj = exp ( log pj ),
j=1 j=1

where log(pj ) is the principal value of the logarithm. Thus, any questions about
infinite products can be translated into a question about infinite series by taking
∞
Y X∞
the logarithms, pj converges if log pj converges.
j=1 j=1
Pj+1
Notice that this implies that pj 6= 0 and pj = Pj → 1.

∞
Y ∞
X
Theorem 2.5.1. pj converges if and only if log(1 + aj ) converges, where
j=1 j=1
log(1 + aj ) is the principal value of the logarithm.
∞
Y ∞
X
It is not necessarily true that log( pj ) = log(1 + aj ) = S.
j=1 j=1
For a more complete discussion of the convergence of the infinite product,
along with a proof, see section 2.2 of Ahlfors [1].

Definition 2.5.2 (Logarithmic differentiation). If G(z) = g1 (z)...gm (z) is a

finite product of analytic functions, then by taking the logarithms and differ-
G0 (z) g 0 (z) g 0 (z)
entiating, we obtain = 1 + ... + m . This procedure is called
G(z) g1 (z) gm (z)
logarithmic differentiation.
The logarithmic differentiation also holds for uniformly convergent infinite
products of analytic functions.

Theorem 2.5.2. Let gk (z) (k > 1) be analytic functions on a domain D such

3
 m
Y ∞
Y
that gk (z) converges normally on D to G(z) = gk (z). Then
k=1 k=1

∞
G0 (z) X gk0 (z)
=
G(z) gk (z)
k=1

where the sum converges normally on D.

Proof : We can apply the logarithmic differentiation to finite subproducts and

G0 (z)
passing the limit. Note that the function has poles at zeros of G (z), and
G(z)
0
G (z)
the order of pole of equals to the order of the zero of G(z). Moreover,
G(z)
the hypothesis implies that gk (z) → 1 uniformly on any compact subset of D,
g 0 (z)
so the summands k are analytic on any compact subset for k large.
gk (z)
We know that the uniform convergence of the series is not affected by the
first terms of the series, the poles has no affect on the uniform convergence of
the series. Moreover, we can derive from the theorem that if f is meromorphic
f0
on a set S, then so is its logarithmic derivative .
f
For a more rigorous proof, see section XIII.3 in Gamelin [3].

3 Prime Number Theorem

In this section, we will prove the prime number theorem. The proof is really
similar to Newman’s proof [7], we will follow the outline of Newman’s proof,
while being slightly more expansive.
As with most proofs of the prime number theorem, the proof is divided into
two parts. The first is to show that the zeta function has no zeros on Re(s) > 1.
The second is to establish the ”Tauberian theorem”. The following proofs were
inspired by Sutherland [6] and Gamelin [3] and are essential to prove the prime
number theorem.

The prime counting function π(x) : R → Z>0 is defined as

X
π(x) = 1,
p6x

it counts the number of primes up to x. The prime number theorem says that

4
the number π(x) of prime numbers not exceeding x satisfies
x
π(x) ∼
log(x)
as x → ∞.
Here, the notation f (x) ∼ g(x) (“f and g are asymptotically equal”) means
f (x)
that lim = 1. In other words, the functions f and g grow at the same
x→∞ g(x)
rate, asymptotically.

To prove the prime number theorem, the first step is to show that the zeta
function has no zeros on Re(s) > 1. Unless otherwise stated, log(z), z ∈ C
stands for the principal branch or the principal value of the logarithm.

Theorem 3.1 (Euler Product). For Re(s) > 1, we have

−1
X Y
ζ(s) = n−s = (1 − p−s ) ,
n>1 p

where the product converges absolutely. In particular, ζ(s) 6= 0 for Re(s) > 1.

Proof. Since n > 1, n ∈ Z, by the Fundamental Theorem of Arithmetic, any

positive integer greater than 1 can be factored into a product of prime numbers
X
in a unique way, that is, n = 2e2 3e3 .... Since n−s converges absolutely for
n>1
−s
X X
−s e2 e3
Re(s) > 1, we have n = (2 3 ...) .
n>1 e2 ,e3 ...>0
Now for each m > 1, let Sm be the m-smooth numbers: positive integers
with prime factors p 6 m, and define
X
ζm (s) = n−s ,
n∈Sm

which converges absolutely and uniformly on Re(s) > 1. If p1 ...pk are the primes
up to m, we can rewrite

−s
X X X X
ζm (s) = n−s = (p1 e1 ...pk ek ) =( p1 −e1 s )...( pk −ek s ),
n∈Sm e1 ...ek >0 e1 >0 ek >0

since ζm (s) converge absolutely, rearrangement is justified. For Re(s) > 1, we

X X e 1
have |p−s | < 1, p−es = (p−s ) = (geometric series) for any prime
1 − p−s
e>o e>o

5
p. If we apply this k times, we obtain
−1
X Y X Y
ζm (s) = n−s = ( p−es ) = (1 − p−s ) .
n∈Sm p e>o p6m

Now for any  > 0, Re(s) > 1 + δ, δ > 0, we have

Z ∞
X X X 1 −δ
|ζm (s)−ζ(s)| 6 n−s 6 |n−s | = n−Re(s) 6 x−1−δ dx 6 m <
m δ
n>m n>m n>m

for m sufficiently large, given n−Re(s) is a positive, decreasing function of s on

[m, ∞). Then by definition, ζm (s) converge to ζ(s) uniformly on Re(s) > 1,
−1
Y
hence the sequence of functions Pm (s) = (1 − p−s ) converges to ζ(s)
p6m
uniformly as well. Therefore we have for Re(s) > 1,
−1
X Y
ζ(s) = n−s = (1 − p−s ) .
n>1 p
−1
Y
Then clearly, the sequence log Pm (s) converge uniformly to log (1 − p−s )
p
locally for Re(s) > 1 as m → ∞, since log(s) is a continuous function on that
X zn
domain. Because log(1 − z) = − for |z| < 1, and Re(s) > 1, |p−s | < 1,
n
n>1
we have
−1 −1
X1
log |(1 − p−s ) | = log (1 − p−s ) = p−es .
e
e>1

Given that Pm (s) is never 0, it follows that

∞ m
Y 1 Y 1
|ζ(s)| = −s )
= lim
n=1
(1 − p n m→∞
n=1
(1 − pn −s )
m m
Y 1 X 1
= lim exp(log −s
) = lim exp( log ).
m→∞
n=1
(1 − pn ) m→∞
n=1
(1 − pn −s )
Moreover, we have
X −1
X X1 XX 1 XX
| log (1 − p−s ) | = p−es 6 | p−es | 6 |p−s |e
p p
e p e>1
e p e>1
e>1
−1 s −1
X X X
s −Re(s)
= (|p | − 1) < |p | = p < ∞.
p p p

−s −1 −1
X Y
Therefore | log (1 − p ) | converge absolutely (i.e. finite), thus (1 − p−s )
p p
converge absolutely, hence nonzero because the exponential function is contin-
−1
Y
uous and never 0. Thus we have shown that ζ(s) = (1 − p−s ) 6= 0 for
p
Re(s) > 1.

6
Theorem 3.2. For Re(s) > 1, we have
1
ζ(s) = + φ(s),
s−1
where φ(s) is analytic on Re(s) > 0. Thus we conclude that ζ(s) extends to a
meromorphic function on Re(s) > 0 and has only one simple pole at s = 1 with
residue 1 and no other poles.

Proof. For Re(s) > 1, we have

Z ∞
1 X
ζ(s) − = n−s − x−s dx.
s−1 1
n>1

X Z ∞
−s
Since n converges absolutely and uniformly for Re(s) > 1, and x−s dx =
n>1 1
XZ n+1 
x−s dx converges absolutely for Re(s) > 1, we have
n>1 n

1 X X Z n+1
ζ(s) − = n−s − ( x−s dx)
s−1
n>1 n>1 n
X Z n+1 XZ n+1
−s −s
= (n − x dx) = (n−s − x−s )dx.
n>1 n n>1 n

Z n+1
For each n > 1, define φn (s) = (n−s − x−s )dx. Since (n−s − x−s ) is a
n
continuous complex-valued function for n 6 x 6 n + 1, and for each
Z n+1fixed x,
−s −s
(n − x ) is an analytic function of s on Re(s) > 0, φn (s) = (n−s −
n
x−s )dx is analytic on Re(s) > 0. For each fixed s such that Re(s) > 0, x ∈
[n, n + 1], we have
x x x
|s| |s| |s|
Z Z Z
|n−s − x−s | = st−s−1 dt 6 dt = dt 6 Re(s)+1 .
n n |ts+1 | n tRe(s)+1 n
Since the interval [n, n + 1] is a piecewise smooth curve with length 1, by ML-
estimate [3, page 105], we have
Z n+1 n+1
|s|
Z
−s −s
|φn (s)| = (n − x )dx 6 |n−s − x−s ||dx| 6 .
n n nRe(s)+1
Re(s0 )
For any s0 with Re(s0 ) > 0, if we take  = ,  > 0, and let U be the union
2
of closed disk D centered at s0 with radius less than , then for each n > 1, we

7
have
|s0 | + 
sup |φn (s)| 6 .
s∈U n1+
|s0 | +  X
Let Mn = Mn = (|s0 | + )ζ(1 + ) converges by the p-test.
, then
n1+ n
X
Then by Weierstrass M-test, φn converges uniformly on each closed disk in
n
X 1
Re(s) > 0. Hence φn converges normally to a function φ(s) = ζ(s) − .
n
s−1
1
Since φn (s) is analytic on Re(s) > 0, φ(s) = ζ(s) − is analytic on
s−1
1
Re(s) > 0. Therefore, ζ(s) = + φ(s) extends to a meromorphic function
s−1
on Re(s) > 0 and has only one simple pole at s = 1 with residue 1 and no other
poles.

We have shown from Theorem 3.1 that ζ(s) 6= 0 for Re(s) > 1. We now want to
show that ζ(s) has no zeros on Re(s) = 1, which is the key to prove the prime
number theorem. To prove this, we rely on the following lemma.

3 4
Lemma 3.3 (Mertens). For all x, y ∈ R, x > 1, we have |ζ(x) ζ(x + iy) ζ(x +
2iy)| > 1.
−1
Y
Proof. We have from Euler product ζ(s) = (1 − p−s ) , and we know that
p
X zn
log |z| = Re(log z), log(1 − z) = − for |z| < 1. Since Re(s) > 1, |p−s | < 1,
n
n>1
we have
X X X X Re(p−ns )
log |ζ(s)| = − log |1 − p−s | = − Re(log(1 − p−s )) = .
p p p
n
n>1

Set ζ = x + iy, since sα = eαlogs for s 6= 0, we have

p−ns = p−n(x+iy) = e−n(x+iy) log p = e−nx log p e−i(ny log p)

= p−nx [cos(ny log p) − i sin(ny log p)].

Re(p−ns ) = Re(p−nx [cos(ny log p) − i sin(ny log p)]) = p−nx cos(ny log p).

Then we have
X X cos(ny log p)
log |ζ(x + iy)| = ,
p
npnx
n>1

8
 4
X X 4 cos(ny log p)
log |ζ(x + iy) | = 4 log |ζ(x + iy)| = .
p
npnx
n>1

Similarly, we have
3
XX 3
log |ζ(x) | = ,
p n>1
npnx
X X cos(2ny log p)
log |ζ(x + 2iy)| = .
p
npnx
n>1

Now, recall that

3 4 3 4
log |ζ(x) ζ(x + iy) ζ(x + 2iy)| = log |ζ(x) | + log |ζ(x + iy) | + log |ζ(x + 2iy)|,
−1
| log (1 − p−s )
P
and p | converges absolutely for Re(s) > 1, then we have

3 4
X X 3 + 4 cos(ny log p) + cos(2ny log p)
log |ζ(x) ζ(x + iy) ζ(x + 2iy)| = .
p
npnx
n>1

2
Recall the double angle formula for cosine, cos 2θ = 2(cos θ) −1, let θ = ny log p,
then we have

2 2
3 + 4 cos θ + cos 2θ = 3 + 4 cos θ + 2(cos θ) − 1 = 2(cos θ + 1) > 0.

Then since every term in the series is positive, we have

3 4
X X 3 + 4 cos(ny log p) + cos(2ny log p)
log |ζ(x) ζ(x + iy) ζ(x+2iy)| = > 0.
p
npnx
n>1

3 4
Hence |ζ(x) ζ(x + iy) ζ(x + 2iy)| > 1 as desired.

Corollary 3.4. ζ(s) has no zeros on Re(s) > 1.

Proof. From Theorem 3.1, we know that ζ(s) has no zeros on Re(s) > 1. From
Theorem 3.2, we know that ζ(s) extends to a meromorphic function on Re(s) >
0.
Let s = x + iy, and suppose that ζ(1 + iy) = 0 for some y ∈ R. We know
1
that y 6= 0, since by Theorem 3.2, ζ(s) = + φ(s) has only one simple pole
s−1
at s = 1 with residue 1 and no other poles. Then ζ(s) does not have a pole at
4
(1 + 2iy) 6= 1. Since ζ(1 + iy) = 0 for some y ∈ R, y 6= 0, ζ(1 + iy) = 0, that is,
4
ζ(x + iy) has a zero at (1 + iy) with order at least 4. Suppose (1 + iy) is a
4
zero of ζ(x + iy) with order 4. Given ζ(s) extends to a meromorphic function

9
 4 4
on Re(s) > 0, ζ(s) extends to a meromorphic function on Re(s) > 0. ζ(s) =
4 4 4
ζ(x + iy) = ((x + iy) − (1 + iy)) f (x + iy) = (x − 1) f (x + iy), f (1 + iy) 6= 0.
Since ζ(s) has a simple pole at s = 1 with residue 1, ζ(x) has a simple
3
pole at x = 1. Moreover, ζ(x) has a pole at x = 1 with order 3, that is,
3 g(x) 3
ζ(x) = 3 , g(1) 6= 0, since ζ(x) extends to a meromorphic function on
(x − 1)
Re(s) > 0.
Then we have
4
3 4 g(x)(x − 1) f (x + iy)ζ(x + 2iy)
lim+ |ζ(x) ζ(x + iy) ζ(x + 2iy)| = lim+ | 3 | = 0,
x→1 x→1 (x − 1)
since ζ(s) has a simple pole at s = 1 and a zero at (1 + iy) and no pole at (1
+ 2iy), but this contradicts Lemma 3.3. Therefore ζ(s) doesn’t have a zero at
(1 + iy), hence ζ(s) has no zeros on Re(s) > 1.

Notice that none of the proofs of the prime number theorem treats the function
π(x) directly. They treat the function
X
ϑ(x) = log p
p6x

instead. We will now introduce a useful property of ϑ(x).

Lemma 3.5 (Chebyshev). For x > 1, ϑ(x) = O(x), where O(f ) denotes a
quantity bounded in absolute value by a fixed multiple of f . ϑ(x) = O(x)
means that there is a constant C > 0 such that |ϑ(x)| 6 C|x| for all x > 1.

Proof. For any integer n > 1, the binomial theorem gives us

2n   2n    
2n
X 2n X 2n 2n 2n!
22n = (1 + 1) = (12n )(12n−m ) = > = .
m=0
m m=0
m n n!n!

Since 2n! is divisible by each prime number p between n and 2n and n! is not
divisible by such p, we have
2n! Y
22n > > p = e(ϑ(2n)−ϑ(n)) ,
n!n!
n<p62n
X
since ϑ(2n) − ϑ(n) = log p, e(x+y) = ex ey . Take the logarithm on both
n<p62n
sides, we have

log(22n ) > log(e(ϑ(2n)−ϑ(n)) ) ⇒ ϑ(2n) − ϑ(n) 6 2n log 2.

10
Now for any integer m > 1, we have
X m  X
X X 
m
ϑ(2 ) = log p = log p − log p
p62m n=1 p62n p62n−1
m
X m
X
= (ϑ(2n ) − ϑ(2n−1 )) 6 2n log 2 6 2m+1 log 2,
n=1 n=1

since ϑ(2n ) − ϑ(2n−1 ) 6 2n log 2, and the elephant tea-cup formula gives us
2m+1 − 1
1 + 2 + ... + 2m = ; 2 + ... + 2m = 2m+1 − 2.
2−1
Then for any real x > 1, we can choose m > 1 such that 2m−1 6 x < 2m ,
then we have

ϑ(x) 6 ϑ(2m ) 6 2m+1 log 2 = (4 log 2)2m−1 6 (4 log 2)x.

Hence ϑ(x) = O(x) as desired.

1
Lemma 3.6. The function Φ(s) − extends to a meromorphic function on
s−1
Re(s) > 12 , and is analytic on Re(s) > 1.

Proof. By Theorem 3.2, we know that ζ(s) extends to a meromorephic function

on Re(s) > 0 and has one simple pole at s = 1 with residue 1. Given the
absolute convergence of the Euler Product (Theorem 3.1), if we differentiate
logarithmically the product formula for the zeta function, we obtain

ζ 0 (s) d d Y −1 d X
− = (− log(ζ(s))) = (− log (1 − p−s ) ) = ( log (1 − p−s ))
ζ(s) ds ds p
ds p
X p−s log p X log p X  log p log p 
= = = + s s
p
1 − p−s p
ps − 1 p
ps p (p − 1)
X log p X log p X log p
= + = Φ(s) + .
p
ps p
ps (ps− 1) p
ps (ps − 1)

The last series converges absolutely and locally uniformly to an analytic

ζ 0 (s)
function by Weiersrass M-test on Re(s) > 21 . Since − is meromorphic on
ζ(s)
Re(s) > 0, it follows that Φ(s) extends to be meromorphic for Re(s) > 21 , with
simple poles and zeros of the zeta function.
Since by Corollary 3.4, ζ(s) has no zeros on Re(s) > 1, and by Theorem
3.2, ζ(s) has only one simple pole at s = 1 with residue 1, Φ(s) has only one

11
 1
simple pole at s = 1 with residue 1. Then it follows that Φ(s) − extends
s−1
1
to a meromorphic function on Re(s) > 2 , and it is analytic on Re(s) > 1 as
desired.

The next step is to show that ϑ(x) ∼ x. To prove this, we rely on a general
analytic criterion that is applicable to any non-decreasing real function f (x).

∞
f (t) − t
Z
Lemma 3.7. Let f : R>1 → R be a nondecreasing function. If dt
1 t2
converges, then f (x) ∼ x.
Z ∞
f (t) − t
Proof. Let F (x) = dt. Since the integral converges, lim F (x) ex-
1 t2 x→∞
ists. Then by the definition of limit, for all λ > 1, and all  > 0, we have
|F (λx) − F (x)| <  for all sufficiently large x. Now fix λ > 1, and suppose
there is an arbitrary large x such that f (x) > λx. Then for such x, since f is
nondecreasing, we have
Z λx Z λx Z λ
f (t) − t λx − t λ−t
F (λx) − F (x) = 2
dt > 2
dt = dt = c > 0,
x t x t 1 t2

since the integrand is nonnegative on [1, λ], so we have |F (λx) − F (x)| > c > 0.
Now if we take  < c, x large enough, we have |F (λx) − F (x)| > c > , which is
a contradiction. Thus f (x) < λx for all sufficiently large x.
Similarly, fix λ > 1, and suppose there is an arbitrary large x such that
1
f (x) 6 x. For such x, we have
λ
 1  Z x f (t) − t Z x 1 Z 1 1
λx − t λ −t
F (x) − F x = 2
dt 6 2
dt = dt < 0,
λ 1
λ
t 1
λx
t 1
λ
t2

since the integrand is nonpositive on [ λ1 , 1]. But this again contradicts our hy-
1
pothesis. Thus f (x) > x for sufficiently large x. Since those inequalities hold
λ
f (x)
for all λ > 1, we must have lim = 1, that is f (x) ∼ x as desired.
x→∞ x

In order to show that f (x) = ϑ(x) satisfies the hypothesis of

Z ∞ Lemma 3.7, given
ϑ(t) − t
ϑ(x) is a nondecreasing function, we need to show that dt con-
1 t2
verges. Instead of showing that this integral converges directly, we will work
with the function H(t) = ϑ(et )e−t − 1. We first want to connect Φ(s) and ϑ(x)
via the Laplace transform.

12
 Recall the Laplace transform of a real-valued piecewise continuous function
h(t), that is defined for all t > 0, (L h)(s) is the complex function defined by
Z ∞
(L h)(s) = h(t)e−st dt,
0

which is analytic and converges absolutely on Re(s) > c for any c ∈ R for which
h(t) = O(ect ).

Lemma 3.8. The Laplace transform of ϑ(et ) is analytic and converges ab-
solutely for Re(s) > 1, and

Φ(s)
L (ϑ(et ))(s) = ,
s
Re(s) > 1.

Proof. By Lemma 3.5, ϑ(et ) = O(et ) since et > 1 for all t > 0, then the Laplace
transform of ϑ(et ) is analytic and converges absolutely on Re(s) > 1. Let pn
denote the nth prime and p0 = 0, then ϑ(et ) is constant for t ∈ (log pn , log pn+1 ),
X
that is ϑ(et ) = log p = ϑ(pn ). We then have
pn <p<pn+1

Z log pn+1 Z log pn+1

−st 1
e t
ϑ(e )dt = ϑ(pn ) e−st dt = ϑ(pn )(pn −s − pn+1 −s ).
log pn log pn s

If we sum over the primes and use ϑ(pn ) − ϑ(pn−1 ) = log pn , we obtain for
Re(s) > 1,
Z ∞ ∞
1X
L (ϑ(et ))(s) = e−st ϑ(et )dt = ϑ(pn )(pn −s − pn+1 −s )
0 s n=1
∞ ∞
1X 1 X −s Φ(s)
= (ϑ(pn ) − ϑ(pn−1 ))pn −s = pn log pn = ,
s n=1 s n=1 s

since L (ϑ(et ))(s) converges absolutely for Re(s) > 1, rearrangement is justified.

Now let’s look at the function H(t) = ϑ(et )e−t − 1. It follows from Lemma
3.8 and the standard property of Laplace transform that on Re(s) > 0, we have

1 Φ(s + 1) 1
L (H(t))(s) = L (ϑ(et )e−t )(s)−L (1)(s) = L (ϑ(et ))(s+1)− = − .
s s+1 s

13
 Now we know the effect of multiplying ϑ(e−t ) by e−t is to translate the
variable s of the Laplace transform by 1, and the effect of subtracting 1 is to
eliminate the pole of Φ(s) at s = 1.

1 Φ(s + 1) 1
Corollary 3.9. The function Φ(s + 1) − and L (H(t))(s) = −
s s+1 s
both extends to meromorphic functions on Re(s) > − 12 that are analytic on
Re(s) > 0.

Proof. The first statement follows immediately from Lemma 3.6, since Φ(s) −
1
extends to a meromorphic function on Re(s) > 12 , and is analytic on
s−1
1
Re(s) > 1, Φ(s + 1) − extends to a meromorphic function on Re(s) > − 12 that
s
is analytic on Re(s) > 0. For the second statement,

Φ(s + 1) 1 1 1 1
− = (Φ(s + 1) − ) − .
s+1 s s+1 s s+1
1
Since Φ(s + 1) − extends to a meromorphic function on Re(s) > − 12 , that
s
1 Φ(s + 1) 1
is analytic on Re(s) > 0, is analytic for Re(s) > −1, − is
s+1 s+1 s
1
meromorphic on Re(s) > − 2 and is analytic on Re(s) > 0 as desired.

The final theorem we need for the proof of the prime number theorem is known
as the “Tauberian Theorem”. It contains the link between the analyticity of
Φ(s) and the asymptotic behavior of H(t).

Theorem 3.10 (Tauberian Theorem). Let f (t) (t > 0) be a bounded piece-

wise
Z ∞ continuous function (locally integrable) and suppose its Laplace transform
f (t)e−st dt (Re(s) > 0) extends analytically to g(s) on Re(s) > 0, then
Z0 ∞
f (t)dt converges and equals to g(0).
0

Proof. For fixed T > 0, define

Z T
gT (s) = f (t)e−st dt.
0

Since f (t) is piecewise continuous, Ze−st is analytic on C for each fixed t, gT (s) is
∞
an entire function. By definition, f (t)e−st dt = lim gT (0), thus it suffices
0 T →∞

14
to show that
lim gT (0) = g(0).
T →∞

We will show that |g(0) − gT (0)| <  for large T,  > 0 small. Since f (t) is
bounded, without loss of generality, assume |f (t)| 6 1 for all t > 0. Now choose
1 
R large enough such that < , then choose δ > 0 small enough such that g(s)
R 4
is defined and analytic in and across the boundary of the domain Dδ consisting
of s such that |s| < R and Re(s) > −δ. We know such δ exist since g(s) is
analytic on Re(s) > 0, hence on some open ball B(iy) with radius less than or
equal to 2δy for each y ∈ [−R, R] and we can take δ = inf δy , y ∈ [−R, R].
s2
For each T > 0, h(s) = (g(s) − gT (s))esT (1 + 2 ) is analytic on Dδ since
R
it’s a combination of analytic functions on Dδ . Given Dδ is a bounded domain
with piecewise smooth boundry, we have by Cauchy’s Integral Formula [3, page
113],

s2 ds
Z
1
h(0) = g(0) − gT (0) = (g(s) − gT (s))esT (1 + 2 )
2πi ∂Dδ R s
Z
1 1 s
= (g(s) − gT (s))esT ( + 2 )ds.
2πi ∂Dδ s R

Now we will break ∂Dδ into pieces.

Let γ+ be the semicircle {|s| = R, Re(s) > 0}, let αδ be the vertical interval
along ∂Dδ where Re(s) = −δ, and let βδ be the two short arcs of the circle
{|s| = R} connecting αδ to ±iR all oriented according to the orientation of
∂Dδ .
For s ∈ γ+ , we have since |f (t)| 6 1,
Z ∞
sT 1 s 1 s
(g(s)−gT (s))e ( + 2 ) = f (t)e−st dt · eRe(s)T · + 2
s R T s R
Z ∞ Z ∞
−st s̄ s 2Re(s)
= f (t)e dt · e Re(s)T
· 2 + R2 6 |f (t)e−st |dt · eRe(s)T ·
T |s| T R2
Z ∞
2Re(s)
6 e−Re(s)t dt · eRe(s)T ·
T R2
1 2Re(s) 2
= e−Re(s)T · eRe(s)T · = 2.
Re(s) R2 R

Hence we have on γ+
1 s 2
(g(s) − gT (s))esT ( + 2 ) 6 2 .
s R R

15
Then by ML-estimate, we have
Z
1 1 s 1 2 1 
(g(s) − gT (s))esT ( + 2 )ds 6 · πR · 2 = < .
2πi γ+ s R 2π R R 4
Now for the integral of g(s) − gT (s) over the remainder of ∂Dδ , we will treat the
functions separately. Since gT (s) is entire, the path for the integral involving
gT (s) can be replaced by the semicircle of radius R, Re(s) < 0. Let γ− be the
semicircle {|s| = R, Re(s) < 0}, then
Z Z
1 1 s 1 1 s
gT (s)esT ( + 2 )ds = gT (s)esT ( + 2 )ds.
2πi αδ ∪βδ s R 2πi γ− s R
For s ∈ γ− , we have
Z T
1 s 1 s
gT (s)e sT
( + 2) = f (t)e−st dt · eRe(s)T · + 2
s R 0 s R
Z T
(−2Re(s))
= f (t)e−st dt · eRe(s)T ·
0 R2
Z T
(−2Re(s))
6 |f (t)e−st |dt · eRe(s)T ·
0 R2
Z T
(−2Re(s))
6 e−Re(s)t dt · eRe(s)T ·
0 R2
1 (−2Re(s))
= (1 − e−Re(s)T ) · eRe(s)T ·
Re(s) R2
2
= 2 (1 − Re(s)eRe(s)T ).
R
As T → ∞, since Re(s) < 0, (1 − Re(s)eRe(s)T ) → 1. Hence we have
1 s 2
gT (s)esT ( + 2 ) 6 2
s R R
for T sufficiently large. Then by ML-estimate, we have
Z
1 1 s 1 2 1 
gT (s)esT ( + 2 )ds 6 · πR · 2 = < .
2πi γ− s R 2π R R 4
1 s
Now since g(s)( + 2 ) is independent of T , |esT | 6 1 on βδ and the length of
s R
βδ tends to 0 with δ, we can choose δ > 0 so small such that for any T > 0,
Z
1 1 s 
g(s)esT ( + 2 )ds < .
2πi βδ s R 4
Finally, |esT | = e−δT on αδ , so for T large enough, the integral of g(s) over αδ is
also bounded by 4 . Adding all fourZ estimates, we obtain |g(0) − gT (0)| <  for T
∞
large. Hence we have shown that f (t)dt converges and equals to g(0).
0

16
 x
Theorem 3.11 (Prime Number Theorem). π(x) ∼ .
log(x)
Proof. By Lemma 3.5, since et > 1 for t > 0, we know that ϑ(et ) = O(et ), that
is ϑ(et ) is bounded in absolute value by some fixed multiple of et . Then we have
L (t) = ϑ(et )e−t −1 is also bounded on t > 0. Since H(t) is piecewise continuous,
and by Corollary 3.9, the Laplace transform of H(t), L (H(t))(s)
Z extends to ∞
an analytic function on Re(s) > 0, Theorem 3.10 implies that H(t)dt =
Z ∞ 0

(ϑ(et )e−t − 1)dt converges.

0
1
Let t = log x, then we have e−t = e− log x = , t ∈ [0, ∞), x ∈ [1, ∞), dt =
x
1
dx. Then
x ∞ ∞
ϑ(x) − x
Z Z
(ϑ(et )e−t − 1)dt = dx
0 1 x2
converges. Given ϑ(x) is a nondecreasing function for x > 1, Lemma 3.7 implies
that ϑ(x) ∼ x. Since
X X
ϑ(x) = log p 6 log x = π(x) log x,
p6x p6x

then
ϑ(x) π(x) log x
06 6 (1)
x x
for x > 1.
For any  > 0,
X X
ϑ(x) = log p > log p > (1 − )(log x)(π(x) − π(x1− ))
p6x x1− <p6x

> (1 − )(log x)(π(x) − x1− ),

then we have
ϑ(x)
π(x) 6 + x1− , (2)
(1 − ) log(x)
for 0 <  < 1.
Combine inequality (1) and (2), we obtain, for all 0 <  < 1,

ϑ(x) π(x) log x ϑ(x) log x x1− log x 1 ϑ(x) log x

6 6 + = .
x x (1 − )x log(x) x 1 −  x x

17
 log x
As x → ∞, → 0, so by choosing  sufficiently small, we can make the
x
x
ratios of ϑ(x) to x, and π(x) to arbitrarily close together, and if one of
log x
x
them tends to 1, the other must too. Therefore π(x) ∼ as x → ∞.
log x

4 Conclusion
The prime number theorem allows one to predict how the primes are distributed.
In this paper, we’ve provided one of the many proofs of the prime number
theorem that requires only careful use of inequalities and complex analysis.
One disadvantage of our proof is that it does not give us an error term. How
close is our estimate? We will use x = 10,000,000 as an example to show how
accurate is our estimate.
For x = 10,000,000, we have

π(x) = 664, 579

x
≈ 620, 420.688433.
log x
This estimate is not the best estimate we have for the prime number dis-
tribution. In fact, this formula π(x) for the asymptotic distribution of primes
dates back to Gauss in the late 18th century. Gauss believed that the function
(also called the logarithmic integral) :
Z x
dx
Li(x) = ,
2 log x
x
provides a much more accurate numerical approximation to π(x) than .
log x
Gauss calculated all primes up to about 3,000,000 and compared the number
of primes found with the above integral [4]. The result he found is fascinating.
For example, between 2,600,000 and 2,700,000, Gauss found 6762 primes and
Z 2,700,000
dx
= 6761.332.
2,600,000 log x

However Gauss never published his investigations on the distribution of primes.

Because of the connection of Riemann Zeta function and π(x), many math-
ematicians have shown if the Riemann hypothesis is true, it will yield a better
estimate of the error involved in the prime number theorem compared to any
other estimates that is available today.

18
References
[1] Lars V. Ahlfors. Complex Analysis: An Introduction to the Theory of An-
alytic Functions of one Complex Variable. International series in pure and
applied mathematics. McGraw-Hill, Inc, 1979. isbn: 0070006571.
[2] Paul Erdos. “On a new method in elementary number theory which leads
to an elementary proof of the prime number theorem ”. In: Proceedings of
the National Academy of Science 35.7 (1949), pp. 374–384. doi: https:
//www.renyi.hu/~p_erdos/1949-02.pdf.
[3] Theodore W. Gamelin. Complex analysis. Springer Science+Business Me-
dia, Inc, 2001. isbn: 0387950699.
[4] Carl Friedrich Gauss. “TAFEL DER FREQUENZ DER PRIMZAHLEN”.
In: Werke. Vol. 2. Cambridge Library Collection - Mathematics. Cambridge
University Press, 2011, pp. 435–448. doi: 10.1017/CBO9781139058230.
019.
[5] Atle Selberg. “An Elementary Proof of the Prime-Number Theorem”. In:
Annals of Mathematics 50.2 (1949), pp. 305–313. doi: https : / / www .
jstor.org/stable/1969455.
[6] Andrew Sutherland. Riemann’s zeta function and the prome number theo-
rem. url: https://math.mit.edu/classes/18.785/2016fa/LectureNotes16.
pdf. (accessed: 05.15.2020).
[7] D. Zagier. “Newman’s Short Proof of the Prime Number Theorem”. In: The
American Mathematical Monthly 104.8 (1997), pp. 705–708. doi: http :
//www.jstor.org/stable/2975232..

19