Chebyshev’s theorem on the distribution of prime numbers

Nikolaos Stamatopoulos, Zhiang Wu

25 November 2021

1 The Chebyshev functions

Denote by π(x) the number of primes not exceeding x > 0. It is well known that there is infinitely many prime
numbers, i.e., limx→∞ π(x) → ∞. The famous prime number theorem tells us more, namely π(x) ∼ x/ log x.

In this paper, we are going to prove the Chebyshev’s theorem, which is an intermediate result of the prime
number theory, and use similar methodology to derive a few other interesting results.

(1 − 1/p)−1 are both divergent, as p runs through all

P Q
Theorem 1 (Euler). The sum 1/p and the product
the prime numbers.

Proof. We shall first show that the product diverges, and then deduce that the series also does. Let

Y −1
1 X1
P (x) := 1− and S(x) := for x ∈ Z⩾2 .
p p
p⩽x p⩽x

Note that for u ∈ (0, 1) and m ∈ Z⩾1 , we have:

1 1 − um+1
> = 1 + u + · · · + um > 0.
1−u 1−u
Choose m ∈ Z⩾1 , such that 2m ⩾ x. Note that the above inequality holds for all primes p ⩽ x with u = 1/p.
By multiplying all of the resulting inequalities, we get

Y  (a) Xx Z x+1
1 1 1 dy
P (x) > 1 + + ··· + m ⩾ > = log(x + 1),
p p n=1
n 1 y
p⩽x
mk
where (a) follows from that fact that, if n = pm1 m2
1 p2 ...pk is the prime factorization of n ∈ {1, 2, ..., x},
then p1 , p2 , ...pk are all ⩽ x; and m1 , m2 , ...mk are all ⩽ m (since x ⩾ 2m ). Thus every term on the right
hand side is a product of terms on the left hand side.

Hence the product (1 − 1/p)−1 diverges.

Q

To prove the divergence of the series, we consider the expansion:

∞
un
 
1 X
log = for u ∈ [−1, 1).
1−u n=1
n

If u ∈ (0, 1), we have:

∞ ∞
un un u2
 
1 X X
log −u= < = .
1−u n=2
n n=2
2 2(1 − u)

Note that the above equation holds for all primes p ⩽ x with u = 1/p.
By adding all of the resulting inequalities, we obtain:

1
 ∞
!
p−2
 
X 1 1 X 1X 1 1X 1 1
log P (x) − S(x) = log − < = < = .
1 − 1/p p 2(1 − 1/p) 2 p(p − 1) 2 n=2 n(n − 1) 2
p⩽x p⩽x p⩽x

P
It follows that S(x) > log P (x) − 1/2 > log log x − 1/2, and hence the sum 1/p also diverges.

Definition. The Chebyshev functions ϑ and ψ are defined as follows:

X X
ϑ(x) := log p = log p, x > 0; (1)
p⩽x p∈{q prime | q⩽x}

X X
ψ(x) := log p = log p, x > 0. (2)
pm ⩽x (p,m)∈{(q,l)∈N×N | q prime, l⩾1 and q l ⩽x}

Mp log p, where Mp is the highest exponent such that pMp ⩽ x holds,

P
Alternatively, we can define ψ(x) := p⩽x
for example ψ(10) = 3 log 2 + 2 log 3 + log 5 + log 7.
Note that Mp = ⌊log x/ log p⌋, and hence

X log x 
ψ(x) = · log p. (3)
log p
p⩽x

Further, it follows from (1) and (2) that eϑ(x) equals the product of all primes p ⩽ x and, for x ⩾ 1, eψ(x) is the
least common multiple of all positive integers n ⩽ x. If pm ⩽ x, then p ⩽ x1/m , and conversely, hence (2) leads
to the relation
   
ψ(x) = ϑ(x) + ϑ x1/2 + ϑ x1/3 + · · · (4)

Recall the von Mangoldt function

(
log p if ∃p prime, m ∈ Z⩾1 : n = pm
Λ(n) = .
0 otherwise

From (2) it is immediate that

X
ψ(x) = Λ(n) (5)
n⩽x

where the sum is finite because ∀x < 2 : ϑ(x) = 0.

We shall now establish a connection between the functions

π(x) ϑ(x) ψ(x)

, , .
x/ log x x x
Theorem 2. Let

π(x) π(x)
l1 = lim inf , L1 = lim sup ,
x→∞ x/ log x x→∞ x/ log x
ϑ(x) ϑ(x)
l2 = lim inf , L2 = lim sup ,
x→∞ x x→∞ x
ψ(x) ψ(x)
l3 = lim inf , L3 = lim sup .
x→∞ x x→∞ x

Then l1 = l2 = l3 , and L1 = L2 = L3 .

2
Proof. It follows from (3) that, for all x > 0,

X (3) X log x  X log x X

ϑ(x) = log p ⩽ ψ(x) = · log p ⩽ · log p = 1 · log x = π(x) log x
log p log p
p⩽x p⩽x p⩽x p⩽x

and hence

ϑ(x) ψ(x) π(x) log x

⩽ ⩽ .
x x x
For x → ∞, we get

ϑ(x) ψ(x) π(x)

lim sup ⩽ lim sup ⩽ lim sup , i.e., L2 ⩽ L3 ⩽ L1 . (6)
x→∞ x x→∞ x x→∞ x/ log x
Now fix α ∈ (0, 1) and x > 1. Then

X X X
α log x = α log x π(x) − π(xα ) > α log x π(x) − xα ,



ϑ(x) = log p ⩾ log p ⩾
p⩽x xα <p⩽x xα <p⩽x

Dividing by x on both sides, we get

ϑ(x) π(x) log x

>α· − αxα−1 log x.
x x
Since α ∈ (0, 1), it follows that xα−1 log x → 0, as x → ∞, and thus

ϑ(x) π(x)
lim sup ⩾ α · lim sup , i.e., L2 ⩾ αL1 .
x→∞ x x→∞ x/ log x

Because the above inequality holds for all α ∈ (0, 1), we deduce that L2 ⩾ L1 .
Combining this with (6), we get L1 = L2 = L3 .

The proof of l1 = l2 = l3 follows analogously.

It follows from Theorem 2 that, if one of the three functions

π(x) ϑ(x) ψ(x)

, ,
x/ log x x x
tends to a limit as x → ∞, then so do the others, and all three limits are the same. Thus in order to prove the
prime number theorem, it is sufficient to show that limx→∞ ψ(x)/x = 1.

3
2 Chebyshev’s theorem
Theorem 3 (Chebyshev). There exist constants a and A, 0 < a < A, such that

x x
a· < π(x) < A · , for sufficiently large x.
log x log x

Proof. Let
π(x) π(x)
l = lim inf , L = lim sup
x→∞ x/ log x x→∞ x/ log x

We shall prove Theorem 3 by showing that L ⩽ 4 log 2, and l ⩾ log 2.

By Theorem 2 these two inequalities are equivalent to

ϑ(x)
L = lim sup ⩽ 4 log 2; (7)
x→∞ x
ψ(x)
l = lim inf ⩾ log 2. (8)
x→∞ x

Proof of (7). The binomial coefficient

 
2n (n + 1)(n + 2) · · · (2n)
N= =
n 1 · 2 · 3···n
has the following properties:

(a) (b)
(i) N is an integer, and N < 22n < (2n + 1)N (9)

where the first inequality follows as N occurs in the binomial expansion of (1 + 1)2n , which has (2n + 1)
positive terms; while the second from the fact that N is the largest among these (2n + 1) terms.

(ii) N is divisible by the product of all primes p, such that n < p ⩽ 2n, because every such prime appears
in the numerator of N , and not in its denominator.
Q
Because of (ii), we have N ⩾ n<p⩽2n p, hence

(9) X
2n log 2 > log N ⩾ log p = ϑ(2n) − ϑ(n). (10)
n<p⩽2n

If we set n = 1, 2, 22 , · · · , 2m−1 in (10), and add the resulting inequalities, we get

m
X
ϑ(2m ) = ϑ(2m ) − ϑ(1) < 2r log 2 < 2m+1 log 2,
r=1

since ϑ(1) = 0.
Now fix x > 1. Choose m ∈ Z such that 2m−1 ⩽ x < 2m . Since ϑ is non-decreasing, (11) gives

ϑ(x)
ϑ(x) ⩽ ϑ(2m ) < 2m+1 log 2 ⩽ 4x log 2, and hence < 4 log 2,
x
from which we can deduce

ϑ(x)
L = lim sup ⩽ 4 log 2,
x→∞ x
as claimed in (7).

4
Proof of (8). The second part of theorem is proved differently, for which we need the following lemma.

We say that a prime p divides the integer n exactly k times, if pk | n, and pk+1 ∤ n.

Lemma. The number of times a prime p exactly divides m! is equal to

     
m m m
+ 2 + 3 + ··· ,
p p p
where the sum above is finite since ⌊x⌋ = 0 for 0 < x < 1.

Proof. Among the integers 1, 2, . . . , m, there are exactly ⌊m/p⌋ which are divisible by p, namely
 
m
p, 2p, . . . , p.
p
The integers between 1 and m which are divisible by p2 are
 
2 2 m 2
p , 2p , . . . , 2 p ,
p
which are ⌊m/p2 ⌋ in number, and so on.

The number of integers between 1 and m which are divisible by p exactly r times, is therefore
⌊m/pr ⌋ − ⌊m/pr+1 ⌋. Hence p divides m! exactly

X  m   m  X  m  X   X 
m m
r − r+1 = r r − (r − 1) r =
pr p p p pr
r⩾1 r⩾1 r⩾2 r⩾1

times, which proves the lemma.

In order to prove (8), we fix n ∈ Z⩾1 , and consider the integer

 
2n (2n)!
N= = .
n (n!)2
Let p be any prime, such that p ⩽ 2n. By virtue of the previous lemma, the numerator of N is divisible by
p exactly ⌊2n/p⌋ + ⌊2n/p2 ⌋ + · · · times, and n! is divisible by p exactly ⌊n/p⌋ + ⌊n/p2 ⌋ + · · · times, so that


the denominator of N is divisible by p exactly 2 ⌊n/p⌋ + ⌊n/p2 ⌋ + · · · times.
Hence N is divisible by p exactly vp times, where

X  2n   
n
vp = −2 r
pr p
r⩾1

Therefore, since N cannot be divisible by any prime larger than 2n,

Y
N= p vp .
p⩽2n
 2n  n
Since pr = pr = 0 when pr > 2n, i.e., when r > ⌊log 2n/ log p⌋, we have

Mp      
X 2n n log 2n
vp = −2 , where Mp = . (11)
i=1
pi pi log p

Note that, for any y ∈ R, we have 2⌊y⌋ ⩽ 2y < 2⌊y⌋+2 and ⌊2y⌋ ⩽ 2y < ⌊2y⌋+1, hence −1 < ⌊2y⌋−2⌊y⌋ < 2.
Since ⌊2y⌋ − 2⌊y⌋ ∈ Z, we can deduce that

⌊2y⌋ − 2⌊y⌋ ∈ {0, 1}. (12)

5
 Combining (11) and (12), we get vp ⩽ Mp , and hence

Y Y X
N= pvp ⩽ pMp , i.e., log N ⩽ Mp log p. (13)
p⩽2n p⩽2n p⩽2n

Recall from (9) that

22n < (2n + 1)N, i.e., 2n log 2 < log(2n + 1) + log N.

Combining this result with (3), (11) and (13) yields

(3) X  log 2n  (11) X (13)

ψ(2n) = log p = Mp log p ⩾ log N > 2n log 2 − log(2n + 1). (14)
log p
p⩽2n p⩽2n

Now fix x > 2, and set n = ⌊x/2⌋ ∈ Z⩾1 . Then n > (x/2) − 1 and x ⩾ 2⌊x/2⌋ = 2n.
Since ψ is non-decreasing, (14) gives

(14)
ψ(x) ⩾ ψ(2n) > 2n log 2 − log(2n + 1) > (x − 2) log 2 − log(x + 1).

Dividing by x on both sides, we get

ψ(x) x−2 log(x + 1)

> · log 2 − .
x x x

Since (x − 2)/x → 1 and x−1 log(x + 1) → 0, as x → ∞, we obtain the desired inequality:

ψ(x)
l = lim inf ⩾ log 2.
x→∞ x

Corollary. Let pn be the n-th prime. There exist constants c and C, 0 < c < C, such that:

x
X 1
c · log log x < < C · log log x, for sufficiently large x.
p
n=1 n

Proof. Theorem 3 tells us that there exist constants a and A, 0 < a < A, such that

x x
a· < π(x) < A · , for sufficiently large x.
log x log x

Let n0 ∈ Z⩾3 be large enough, such that ∀n ⩾ n0 : (i) a · pn / log pn < π(pn ) < A · pn / log pn ,
√
(ii) log pn < a pn .

Fix n ⩾ n0 . Note that n < pn and π(pn ) = n, hence

(i)
n log n < π(pn ) log pn < A · pn , (15)
√ (ii) pn (i)
pn < a · < π(pn ) = n. (16)
log pn

Thus log pn < 2 log n, and hence

(16)
apn < n log pn < 2n log n, (17)
a (17) 1 (15) A
< < . (18)
2n log n pn n log n

From (18) we can deduce the corollary, because of the following calculation:

6
 x ⌊x⌋ 0n
X 1 X 1 X 1
∀x ⩾ en0 : = +
p
n=1 n n=n0 +1
pn p
n=1 n
⌊x⌋ pn
(18) X 1 X1
< A· +
n=n0 +1
n log n n=1 n
⌊x⌋ 2
0 n
(16) X 1 X 1
< A· +
n=n0 +1
n log n n=1 n
Z ⌊x⌋ Z n0 2
dy dy
⩽ A· +1+
n0 y log y 1 y

= A · log log⌊x⌋ − A · log log n0 + 1 + log n0 2

⩽ A · log log x + 1 + 2 log n0

(b)
⩽ A · log log x + log log x + 2 log log x

= (A + 3) · log log x,

where (b) follows from x ⩾ en0 , i.e., log log x ⩾ log log en0 = log n0 ;

x x
X 1 X 1
∀x ⩾ n0 log n0 : ⩾
p
n=1 n n=n
pn
0
x
a X
(18) 1
> ·
2 n=n n log n
0
Z ⌊x⌋+1
a dy
⩾ ·
2 n0 y log y

a a
= · log log(⌊x⌋ + 1) − · log log n0
2 2

(c) a a
⩾ · log log x − · log log x
2 4

a
= · log log x,
4

where (c) follows from x ⩾ n0 log n0 , i.e., log log x > log log n0 log n0 = log(log n0 · log n0 ) = 2 log log n0 .

7
3 Bertrand’s postulate
Theorem 4. Let n ∈ N. Then there is some prime p such that n < p ≤ 2n.

The first proof is due to Chebyschev, the proof presented here is due to S.S.Pillai.
22n
< N < 22n for N = 2n


Instead of the estimate 2n+1 n employed in the proof of Chebyschev’s theorem we
will use instead the sharper estimate
22n 22n
√ < N < √ (n ≥ 2), (19)
2 n 2n
in order to show that

ϑ(n) < 2n log 2(n ≥ 1). (20)

We will first prove the inequalities in (19).

Define

1 · 3 · 5 · . . . · (2n − 1) 1 · 3 · 5 · . . . · (2n − 1) 2 · 4 · 6 · . . . · (2n) (2n)!

P = = · = 2n ,
2 · 4 · 6 · . . . · (2n) 2 · 4 · 6 · . . . · (2n) 2 · 4 · 6 · . . . · (2n) 2 (n!)2

and so N = 22n P
Now     
1 1 1
1− 2
1> 1 − 2 ... 1 −
2 4 (2n)2
     
1·3 3·5 5·7 (2n − 1)(2n + 1)
=⇒ 1 > ...
22 42 62 (2n)2

2n 2
=⇒ 1 > (2n + 1)P 2 > 2nP 2 = N .
24n
From this we directly obtain the second inequality in (19).
For the first one we have
    
1 1 1
1> 1− 2 1 − 2 ... 1 −
3 5 (2n − 1)2
    
2·4 4·6 (2n − 2)2n
=⇒ 1 > ...
32 52 (2n − 1)2
1 24n
=⇒ 1 > 2
= .
4nP 4nN 2
This then gives us the first inequality, and (19) is proved.
We then prove (20). This trivially holds for n = 1, 2. Now we assume that it holds for some n ≥ 2, we shall
show that ϑ(2n − 1) < 2(2n − 1) log 2, which would imply ϑ(2n) = ϑ(2n − 1) < 4n log 2.
Consider the integer
   
N 1 2n (2n)! n (2n − 1)! 2n − 1
= = = = .
2 2 n (n!)2 2n n!(n − 1)! n−1
This is divisible by every prime p with n < p ≤ 2n − 1, and so also by their product, which gives:

N Y N
≥ p =⇒ log ≥ ϑ(2n − 1) − ϑ(n).
2 2
n<p≤2n−1

(19) gives us on the other hand

1
log N < 2n log 2 − log 2n.
2

8
 Hence, by combining the two, we have

1
ϑ(2n − 1) − ϑ(n) < (2n − 1) log 2 − log 2n.
2

By our assumption ϑ(n) < 2n log 2 and so

1
ϑ(2n − 1) < 2n log 2 + (2n − 1) log 2 − log 2n.
2
This implies when n ≥ 2
ϑ(2n − 1) < 2(2n − 1) log 2,

which is exactly the sought inequality. Hence, if (20) holds for n, it holds for 2n − 1 and by the previous
remark also for 2n. This means that, if it holds for every n ∈ (2r−1 , 2r ], r ≥ 1, then it holds for every
n ∈ (2r , 2r+1 ], and so the claim follows by the induction since it holds for 1 and 2.
Now using these results we shall prove the theorem in two parts, one part for the case n ≥ 26 and one for
n < 26 . Starting with the first,recall from a previous proof that we have
 
2n (2n)! Y
N= = 2
= pvp
n (n!)
p≤2n

with
X  2n  
n

vp = −2 .
pr pr
r≥1

Then
X
log N = vp log p.
p≤2n
P P P P
We partition this sum into the following value ranges for p, calling them 1, 2, 3 4 respectively:

1. n < p ≤ 2n
2n
2. 3 <p≤n
√ 2n
3. 2n < p ≤ 3 with n ≥ 5
√
4. p ≤ 2n.
n
< 1, so ⌊ np ⌋ = 0 and 1 ≤ 2n
< 2 so that ⌊ 2n 2n
P
For 1 we have p p p ⌋ = 1 and ⌊ p2 ⌋ = 0. Then vp = 1 and we get

X X X
= vp log p = log p = ϑ(2n) − ϑ(n).
1 n<p≤2n n<p≤2n

n
≤ 32 , so ⌊ np ⌋ = 1 and ⌊ 2n 2n
P
For 2 we have that 1 ≤ p p ⌋ = 2; for n ≥ 3 we get ⌊ p2 ⌋ = 0, giving us in total
P
2 = 0 for n ≥ 3.
n 2n
< 1, so vp = ⌊ 2n n
P
For 3 we have n ≥ 5 and p2 < p2 p ⌋ − 2⌊ p ⌋ = 0 or 1. Therefore

√
 
X X 2n
≤ log p = ϑ − ϑ( 2n).
3
√ 3
2n<p≤2n/3

Now
√ X X √
ϑ( 2n) = log p ≥ log 2 1 = π( 2n) log 2,
√ √
p≤ 2n p≤ 2n

and so
√
 
X 2n
≤ϑ − π( 2n) log 2.
3
3

9
 P
Finally for 4 we use (13) and obtain

log 2n
vp ≤ M p = ,
log p

so that
X X X log 2n X
≤ Mp log p ≤ · log p = log 2n 1.
4
√ log p √
p≤2n p≤ 2n p≤ 2n

Hence

X √
≤ π( 2n) log 2n.
4

We add up all the sums together and get, for n ≥ 5,

√
 
2n
log N ≤ ϑ(2n) − ϑ(n) + ϑ − π( 2n)(log 2 − log 2n). (21)
3

We will use this to show that ϑ(2n) − ϑ(n) > 0 for large enough n. For this purpose we use the following
three inequalities:
√
1. log N > 2n log 2 − log(2 n) which follows from first inequality of (19);
   
2n 2n
< 2 2n


2. ϑ 3 = ϑ 3 3 log 2 for n ≥ 2 which follows from (20);

n
3. π(n) ≤ 2 for n ≥ 8, since all even numbers greater than 2 are composite.

Combining these three inequalities with (21) we obtain, for n ≥ 32,

√
√ 4n 2n
ϑ(2n) − ϑ(n) > 2n log 2 − log(2 n) − log 2 − log n
3 2

  √ !
2n 2n + 1
=⇒ ϑ(2n) − ϑ(n) > − 1 log 2 − log n.
3 2

It remains to show that this is strictly positive.

This can be manually checked and verified for n = 26 , now we check it for n > 26 . For this purpose we will
rewrite the inequality as
√ √
√ 3 log n 3 2 log 4n
2n − − √ > 0.
2 log 2 log 2 4n
Treating this as two real functions of the form
√ √
√ 3 log x 3 2 log 4x
2x − and − √ ,
2 log 2 log 2 4x
notice that both functions have positive derivatives for x ≥ 26 and their sum is positive for x = 26 , hence
the sum is positive for all x > 26 and so in particular for every natural number in that range.
Then it follows that ϑ(2n) − ϑ(n) > 0 for n ≥ 26 ,and so Bertrand’s postulate follows for n ≥ 64.
Now for the smaller naturals, consider the primes 2, 3, 5, 7, 13, 23, 43, 67. Each one of these is smaller than
twice the previous one, in particular for any n < 64 we may pick one of these primes, which will satisfy Bertrand’s
postulate.
Thus we have proven that the postulate holds for every natural number n.

10
 π(n)
4 Asymptotic bounds for n/ log n
In this section we will prove the following theorem:

Theorem 5.
π(x) π(x)
lim inf ≤ 1 ≤ lim sup .
x→∞ x/ log x x→∞ x/ log x

π(n)
Theorem 5 implies that, if the limit limn→∞ n/ log n exists, then it is equal to 1.
We will prove instead that

ψ(x) ψ(x)
lim inf ≤ 1 ≤ lim sup ,
x→∞ x x→∞ x
which by Theorem 2 is equivalent to the assertion in Theorem 5.
′
Let f (s) = − ζζ(s)
(s)
for every real s > 1, and let

ψ(x) ψ(x)
l = lim inf , L = lim sup
x→∞ x x→∞ x
l′ = lim inf
+
(s − 1)f (s) , L′ = lim sup(s − 1)f (s)
s→1 s→1+

Obviously l ≤ L and l′ ≤ L′ .
We will show that l ≤ l′ ≤ L′ ≤ L and l′ = L′ = 1; together these two give us Theorem 5.
ψ(x)
Choose some arbitrary real B > L, then x < B for all x ≥ x0 = x0 (B), and we can assume without loss
of generality x0 > 1.
Here we will need a statement of Abel’s Theorem, whose proof we will omit:

Theorem 6. Let 0 ≤ λ1 ≤ λ2 ≤ . . . be a sequence of real numbers, such that λn → ∞ as n → ∞ and let

P
(an ) be a sequence of complex numbers. Let A(x) = λn ≤x an , and ϕ(x) a complex-valued function defined for
x ≥ 0. Then
k
X k−1
X
an ϕ(λn ) = A(λk )ϕ(λk ) − A(λn )(ϕ(λn+1 ) − ϕ(λn )). (22)
n=1 n=1

If ϕ has a continuous derivative in (0, ∞), and x ≥ λ1 , then equation 22 can be written as

X Z x
an ϕ(λn ) = A(x)ϕ(x) − A(t)ϕ′ (t)dt. (23)
λn ≤x λ1

If, in addition, A(x)ϕ(x) → 0 as x → ∞, then

∞
X Z ∞
an ϕ(λn ) = − A(t)ϕ′ (t)dt, (24)
n=1 λ1

provided either side is convergent.

Now for s > 1, set λn = n, an = Λ(n), and ϕ(x) = x−s , then A(x) = ψ(x), and A(x)ϕ(x) → ∞ as x → ∞,
since ψ(x) ≤ π(x) log x ≤ x log x, so that A(x)ϕ(x) = O(x1−s log x) = o(1). We will also use without proof that

∞
ζ ′ (s) X Λ(n)
− = , (s real, s > 1).
ζ(s) n=1
ns

Combining this with equation 24 gives

Z ∞ Z x0 Z ∞
ψ(x) ψ(x) B
f (s) = s dx < s dx + dx,
1 xs+1 1 xs+1 x0 xs
so that

11
 Z x0 Z ∞ Z x0
ψ(x) B ψ(x) sB
f (s) < s dx + s dx < s dx + .
1 xs+1 1 xs 1 x 2 s−1
We set Z x0
ψ(x)
dx = K = K(x0 ) = K(x0 , B),
1 x2
so that we can rewrite the above equation in the form

(s − 1)f (s) < s(s − 1)K + sB.

We let s → 1+ , obtaining

L′ = lim sup(s − 1)f (s) ≤ B,

s→1+0

and since B > L was arbitrary we have necessarily L′ ≤ L. An analogous argument gives l ≤ l′ , and so
l ≤ l′ ≤ L′ ≤ L
Now we prove l′ = L′ = 1 by showing

lim −(s − 1)2 ζ ′ (s) = 1 and lim+ (s − 1)ζ(s) = 1.

s→1+ s→1
+
This implies (s − 1)f (s) → 1 as s → 1 .
For s > 1, we have that x−s is decreasing as a function of x, and so
Z ∞ ∞ Z ∞
dx X 1 dx
< <1+ ;
1 xs n=1
n s
1 xs

hence

1 s
< ζ(s) < ,
s−1 s−1
implying (s − 1)ζ(s) → 1 as s → 1+ .
Now for the second part we have that, for s > 1 and x ≥ e, the function x−s log x is decreasing, so

∞ Z ∞
X log n log x
−ζ ′ (s) = s
= dx + O(1).
n=1
n 1 xs

By substituting xs−1 = ey we get

Z ∞
′ 1 1
−ζ (s) = ye−y dy + O(1) = + O(1),
(s − 1)2 0 (s − 1)2
so that

(s − 1)2 ζ ′ (s)
(s − 1)f (s) = − → 1 as s → 1+ .
(s − 1)ζ(s)
.
This means that l′ = L′ = 1 and thus l ≤ 1 ≤ L, concluding the proof of the Theorem.

12