PRIME NUMBER THEOREM

RYAN LIU

Abstract. Prime numbers have always been seen as the building blocks of
all integers, but their behavior and distribution are often puzzling. The prime
number theorem gives an estimate for how many prime numbers there are
under any given positive number. By using complex analysis, we are able to
find a function π(x) that for any input will give us approximately the number
of prime numbers less than the input.

Contents
1. The Prime Number Theorem 1
2. The Zeta Function 2
3. The Main Lemma and its Application 5
4. Proof of the Main Lemma 8
5. Acknowledgements 10
6. References 10

1. The Prime Number Theorem

A prime number is an interger = 2 which is divisible only by itself and 1.

Thus the prime numbers start with the sequence 2,3,5,7,11,13,17,19, . . . Since these
numbers are indivisible but anything other than itself and 1, we can see them as
the building blocks of all other numbers. In fact, the Fundamental Theorem of
Arithmetic tells us that every natural number can be factored into a unique set of
primes. It would therefore be beneficial for us to find out these prime numbers, or
how they are distributed amongst the natural numbers.
By looking at the first few prime numbers, it is hard to believe that there could
be any pattern among the prime numbers, but using complex analysis, we are able
to come out with a very simple representation of the distribution of prime numbers.
In order to do so, we must first define a function π(x) which will output the
number of primes 5 x. The prime number theorem then describes how π(x) behaves
for real numbers x.
Theorem 1.1. (Prime Number Theorem) We have
x
π(x) ∼ .
log x

Date: July 20, 2012.

1
2 RYAN LIU

Somehow, we are able to relate the distribution of prime numbers to a simple

x
fraction, logx . To understand how this makes sense, we will need to start from a
more basic function, the zeta function.

2. The Zeta Function

Let s be a complex variable. For Re(s) > 1 the series

∞
X 1
s
,
n=1
n
converges absolutely, and uniformly for Re(s) = 1 + δ, with any δ > 0. This series
is called the zeta function. One sees this convergence by estimating
1 1
5 1+δ ,
ns n
1/n1+δ , which has positive
P
and by using the integral test on the real series
terms.
The Reimann Zeta Function has been used in algebraic number theory because
of its usefulness in dealing with prime numbers. We will see its power in the next
proposition.
Proposition 2.1. The product
Y 1

1− ,
p
ps

converges absolutely for Re(s) > 1, and uniformly for Re(s) = 1 + δ with δ > 0,
and we have −1
Y 1
ζ(s) = 1− s .
p
p

Proof. The product converges since each term is less than 1 and thus
In the same region Re(s) = 1 + δ, we can use the geometric series estimate to
conclude that
 −1
1 1 1 1
1− s = 1 + s + 2s + 3s + . . . = Ep (s).
p p p p

Using a basic fact from elementary number theory that every positive integer has
unique factorization into primes, up to the order of the factors, we conclude that
in the product of the terms Ep (s) for all primes p, the expression 1/(ns ) will occur
exactly once, thus giving the series for the zeta functino in the region Re(s) > 1.
This concludes (Euler’s) proof. 
The product
Y −1
1
1− s ,
p
p

is called the Euler product. The representation of the zeta function as such a
product shows that ζ(s) 6= 0 for Re(s) > 1. Since this product relates all primes
 PRIME NUMBER THEOREM 3

to all natural numbers, if we are able to study this function, we can conclude the
distribution of primes amongst the natural numbers. In order to make this function
more usable, we want to extend it to all Re(s) > 0, however we also have to make
sure that the resulting function behaves in a way we want it to so we can analyze
it further.
Theorem 2.2. The function
1
ζ(s) −
,
s−1
extends to a holomorphic function on the region Re(s) > 0
Proof. For Re(s) > 1, we have
∞ Z ∞
1 X 1 1
ζ(s) − = s
− dx
s − 1 n=1 n 1 xs
∞ Z n+1  
X 1 1
= − dx.
n=1 n
ns xs
We estimate each term in the sum by using the relations
Z b
f (b) − f (a) = f 0 (t)dt and so |f (b) − f (a)| 5 max |f 0 (t)||b − a|,
a a5t5b

therefore each term is estimated as follows:

Z n+1  
1 1 1 1 s
s
− s dx 5 max s
− s 5 max s+1
n n x n5x5n+1 n x x
|s|
5 .
nRe(s)+1
Thus the sum of the terms coverge absolutely and uniformly for Re(s) > δ. This
concludes the proof of the theorem

We note that the zeta function has a certain symmetry about the x-axis, namely
ζ(s) = ζ(s).
This is immediate from the Euler product and the series expansion of Theorem
2.2. It follows that if s0 is a complex number where ζ has a zero of order m (which
may be a pole, in which case m is negative), then the complex conjugate s¯0 is a
complex number where ζ has a zero of the same order m.

We now define
X X log p
ϕ(x) = log p and Φ(s) = for Re(s) > 1.
p
ps
p5x

We will now try to use log because it allows us to turn the Euler product into
sums which may be more manageable.
The sum defining Φ(s) converges uniformly and absolutely for
Re(s) = 1 + δ,
4 RYAN LIU

by the same argument as for the sum defining the zeta function. We merely use
the fact that given  > 0
log n 5 n for all n = n0 ()
Theorem 2.3. The function Φ is meromorphic for Re(s) > 21 . Furthermore, for
Re(s) = 1, we have ζ(s) 6= 0 and
1
Φ(s) − ,
s−1
has no poles for Re(s) = 1
Proof. For Re(s) > 1, The Euler product shows that ζ(s) 6= 0. By taking the
derivative of log ζ(s) we get
ζ0 X log p
− = .
ζ(s) p
ps − 1

Using the geometric series we get the expansion

 
1 1 1 1 1 1
= s = s 1 + s + 2s + · · ·
ps − 1 p 1 − 1/ps p p p
1 1
= s + 2s + · · · ,
p p
so
ζ0 X log p
(2.4) − = Φ(s) + hp (s) where |hp (s)| 5 C 2s ,
ζ(s) p
|p |

for some constant C. But the series (log n)/n2s converges absolutely and
P
uniformly for Re(s) = 12 + δ with δ > 0, so Theorem 2.2 and Equation 2.4 imply
that Φ is meromorphic for Re(s) > 21 , and has a pole at s = 1 and at the zeros of
ζ, but no other poles in this region.

There remains only to be proved that ζ has no zero on the line Re(s) = 1.
Suppose ζ has a zero of order m at s = 1 + ib with b 6= 0. Let n be the order of
the zero at s = 1 + 2ib. Then m, n = 0 by Theorem 2.2.

From Equation 2.4 we get

lim Φ(1 + i) = 1, lim Φ(1 +  ± ib) = −m,
→0 →0
lim Φ(1 +  ± 2ib) = −n.
→0

Indeed, if a meromorphic function f (z) has the factorization

f (z) = (z − z0 )k h(z) with h(z0 ) 6= 0, ∞

then
f0 k
= + holomorphic terms at z0 ,
f (z) z − z0
whence our three limits follow immediatley from the definitions and the pole of the
zeta function at 1, with residue 1.
 PRIME NUMBER THEOREM 5

Now the following identity is trivially verified:

Φ(s + 2ib) + Φ(s − 2ib) + 4Φ(s + ib) + 4Φ(s − ib) + 6Φ(s)
X log p
= 2s
(pib/2 + p−ib/2 )4
p
p

For s = 1 +  > 1 we see that the above expression is = 0. using our limits, we
conclude that
−2n − 8m + 6 = 0,

whence m = 0. This concludes the proof of Theorem 1.4. 

Proposition 2.5. For Re(s) > 1 we have

Z ∞
ϕ(x)
Φ(s) = s dx
1 xs+1
Proof. To prove this, compute the integral on the right between successive prime
numbers, where ϕ is constant. Then sum by parts. 

3. The Main Lemma and its Application

Theorem 3.1. (Chebyshev) We have

ϕ(x) = O(x).
Proof. Let n be a positive integer. Then
X 2n 2n Y
2n 2n
2 = (1 + 1) = = = p = eϕ(2n)−ϕ(n) ,
j
j n
n<p52n

hence we get the equality

ϕ(2n) − ϕ(n) 5 2n log 2.

But if x increases by 1, then ϕ(x) increases by at most log(x + 1), which is
O(log x). Hence there is a constant C > log 2 such that for all x = x0 (C) we have

ϕ(x) − ϕ(x/2) 5 Cx.

We apply this inequality in succession to x, x/2, x/22 , · · · , x/2r and sum.
This yields

ϕ(x) 5 2Cx + O(1),

which proves the theorem. 
6 RYAN LIU

We shall now state the main lemma, which constitutes the delicate part of the
proof. Let f be a function defined on the real numbers = 0, and assume for
simplicity that f is bounded, and piecewise continuous. What we prove will hold
under much less restrictive condition: instead of piecewise continuous, it would
suffice to assume that the integral
Z b
|f (t)|dt,
a
exists for every pair of numbers a, b = 0. We shall associate to f the Laplace
transform g defined by
Z ∞
g(z) = f (t)e−zt dt for Re(z) > 0.
0

Lemma 3.2. (Main Lemma) Let f be bounded, piecewise continuous on the reals
= 0. Let g(z) be defined by
Z ∞
g(z) = f (t)e−zt dt for Re(z) > 0
0
If g extends to an analytic function for Re(z) = 0, then
Z ∞
f (t) dt
0
exists and is equal to g(0)
The main lemma will be proved in the next section, for now we will apply the
main lemma to prove Lemma 3.3
Lemma 3.3. The integral
∞
ϕ(x) − x)
Z
dx,
1 x2
converges
Proof. Let
ϕ(et ) − et
f (t) = ϕ(et )e−t − 1 = .
et
f is certainly piecewise continuous, and is bounded by Theorem 2.1. Making the
substitution x = et in the desired integral, dx = et dt, we see that
Z ∞ Z ∞
ϕ(x) − x
dx = f (t) dt.
1 x2 0
It suffices to prove that the integral on the right converges. By the main lemma,
it suffices to prove that the Laplace transform of f is analytic for Re(z) = 0, so we
have to compute this Laplace transform. We claim that in this case,

Φ(z + 1) 1
g(z) = − .
z+1 z
Once we have proved this formula, we can apply Theorem 1.3 to conclude that g
is analytic for Re(z) = 0, thus concluding the proof of Lemma 2.3. Now to compute
g(z), we use the integral formula of Proposition 1.4. By this formula, we obtain
Z ∞
Φ(s) 1 ϕ(x) − x
− = dx.
s s−1 1 xs+1
 PRIME NUMBER THEOREM 7

By substituting z + 1 for s we get

∞
ϕ(x) − x
Z
Φ(z + q) 1
− = dx
z+1 z 1 xs+1
∞
ϕ(et ) − et −zt t
Z
= e e dt
e2t
Z0 ∞
= f (t)e−zt dt.
0
This gives us the Laplace transform of f , and concludes the proof of Lemma 2.3.

Let f1 and f2 be functions defined for all x = x0 , for some x0 . We say that f1
is asymptotic to f2 , and write
f1 ∼ f2 if and only if lim f1 (x)/f2 (x) = 1
x→∞

Theorem 3.4. We have ϕ(x) ∼ x

Proof. The assertion of the theorem is logically equivalent to the comibnation of
teh following two assertions:
Given λ > 1, the set of x such that ϕ(x)/x = λ x is bounded;
Given 0 < λ < 1, the set of x such that ϕ(x) 5 λ x is bounded.
Let us prove the first. Suppose the first assertion is false. Then there is some
λ > 1 such that for arbitrarily large x we have ϕ(x)/x = λ. Since ϕ is monotone
and increasing, we get for such x:
Z λx Z λx Z λ
ϕ(t) − t λx − t λ−t
2
dt = 2
dt = dt > 0.
x t x t 1 t2
The number on the far right is independent of x. Since there are arbitrarily large
x satisfying the above inequality, it follows that the integral of Lemma 2.3 does
not converge, a contradiction. So the first assertion is proved. The second
assertion is proved in the same way. 
Theorem 3.5. (Prime Number Theorem) We have
x
π(x) ∼ .
log x
Proof. We have X X
ϕ(x) = log p 5 log x = π(x) log x;
p5x p5x
and given  > 0,
X X
ϕ(x) = log p = (1 − ) log x
x1− 5p5x x1− 5p5x

= (1 − ) log x π(x) + O(x1− ).

Using Theorem 2.4 that ϕ(x) ∼ x concludes the proof of the prime number theorem

8 RYAN LIU

4. Proof of the Main Lemma

We recall the main lemma

Let f be bounded, piecewise continuous on the reals = 0. Let

Z ∞
g(z) = f (t)e−zt dt f or Re(z) > 0
0
If g extends to an analytic function for Re(z) = 0, then
Z ∞
f (t) dt exists and is equal to g(0)
0

Proof. For T > 0 define

Z T
gT (z) = f (t)e−zt dt.
0
gT is an entire function, as follows at once by differentiating under the integral
sign. We have to show that
lim gT (0) = g(0).
T →∞

Let δ > 0 and let C be the path consisting of the line segment Re(z) = −δ and
the arc of circle |z| = R and Re(z) = −δ,

By our assumption that g extends to an analytic function for Re(z) = 0, we can

take δ small enough so that g is analytic on the region bounded by C, and on its
boundary. Then

z 2 dz
Z Z
1 Tz 1
g(0) − gT (0) = (g(z) − gT (z))e (1 + 2 ) = HT (z) dz,
2πi C R z 2πi C

where HT (z) abbreviates the expression under the integral sign. Let B be a
bound for f , that is |f (t)| 5 B for all t = 0.

In order to prove |g(0) − gT (0)| < , we will split up the path C into C − and C +

Let C + be the semicircle |z| = R and Re(z) = 0. Then we can prove

Z
1 2B
(4.1) HT (z) dz 5
2πi C+ R
by first noting that for Re(z) > 0 we have
Z ∞ Z ∞
|g(z) − gT (z)| = f (t)e−zt dt 5 B |e−zT | dt
T T
B −Re(z)T
= e
Re(z)
and for |z| = R,
z2
 
Tz 1 R z 1 2|Re(z)|
e 1+ 2 = eRe(z)T + = eRe(z)T .
R z z R R R2
 PRIME NUMBER THEOREM 9

Taking the product of the last two estimates and multiplying by the length of the
semicircles gives the bound for the integral over the semicircle, and proves the
claim

Now let C − be the part of the path C with Re(z) < 0. We wish to estimate
Z −
z 2 dz
 
1 Tz
(g(z) − gT (z)) e 1+ 2
2πi C R z
Now we estimate separately the expression under the integral with g and gT .
We want to show that
z 2 dz
Z  
1 Tz B
(4.2) gT (z)e 1+ 2 5 .
2πi C − R z R
we do this by letting S − be the semicircle with |z| = R and Re(z) < 0. Since gT is
entire, we can replace C − by S − in the integral without changing the value of the
integral, because the integrand has no pole to the left of the y-axis. Now we
estimate the expression under the integral sign on S − . We have

Z T Z T
−zT
|gT (z)| = f (t)e dt 5 B e−Re(z)T dt
0 0

B e−Re(z)T
5
−Re(z)

For the other factor we use the same estimate as previously. We take the product
of the two estimates, and multiply by the length of the semicircle to give the
desired bound in (3.2).

Third, we claim that

z 2 dz
Z  
(4.3) g(z)eT z 1 + 2 →0 as T → ∞
C− R z
We can write the expression under the integral sign as

z2 1
 
g(z)eT z 1 + 2 = h(z)eT z where h(z) is independent of T .
R z
Given any compact subset K of the region defined by Re(z) < 0, we note that

eT z → 0 rapidly uniformly for z ∈ K, as T → ∞

The word “rapidly” means that the expression divided y any power T N also tends
to 0 uniformly for z in K, as T → ∞. From this our claim (3.3) follows easily
We may now prove the main lemma. We have
Z ∞
f (t) dt = lim gT (0) if this limit exists.
0 T →∞

But given , pick R so large that 2B/R < . Then by (3.3), pick T so large that
10 RYAN LIU

z 2 dz
Z  
Tz
g(z)e 1+ 2 <
C− R z
Then by (3.1), (3.2), and (3.3) we get |g(0) − gT (0)| < 3. This means that the
equation converges, proving the main lemma. This concludes the proof of the
prime number theorem.


5. Acknowledgements
I would like to thank my mentor, Mohammad Abbas Rezaei, for introducing me
to and helping me read through Lang’s book, as well providing comments and
suggestions on the initial drafts of this manuscript. I would like to thank Peter
May for reading through the manuscript, as well as for his role in putting on the
VIGRE program at the University of Chicago, which allowed me to pursue my
interest in number theory.

6. References
[1] Serge Lang, Complex Analysis, Springer-Verlag, New York, 1993