Universitat Autònoma de Barcelona

Bachelor’s Thesis

On the critical strip of the

Riemann zeta-function

Author: Supervisor:
Niels Gleinig Dr. Francesc Bars

June 18, 2014

Contents
1 Preface 2

2 Introduction 3
2.1 The Euler product . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Riemann’s functional equation for ζ(s) . . . . . . . . . . . . . . . 5
2.2.1 The Poisson Summation Formula . . . . . . . . . . . . . . 5
2.2.2 The Γ-function and ψ-function . . . . . . . . . . . . . . . 6
2.2.3 The Mellin Transform . . . . . . . . . . . . . . . . . . . . 8
2.2.4 Riemann’s functional equation for ζ(s) . . . . . . . . . . . 9
2.3 Extending ζ(s) via η(s) . . . . . . . . . . . . . . . . . . . . . . . 12

3 Zeros of ζ(s) 17
3.1 Weierstrass infinite products . . . . . . . . . . . . . . . . . . . . . 17
3.2 The Riemann hypothesis . . . . . . . . . . . . . . . . . . . . . . . 18
3.3 Some results about the distribution of the zeros of ζ(s) . . . . . . 19
3.4 Numerical computation of zeros of ζ(s) . . . . . . . . . . . . . . . 21
3.4.1 Using Euler-Maclaurin summation to compute values of
ζ(s) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.4.2 Finding zeros on the critical line . . . . . . . . . . . . . . 23
3.4.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4.4 The number of zeros in a given set . . . . . . . . . . . . . 25
3.4.5 A simple program in C to compute ζ(s) . . . . . . . . . . 26

4 Universality of ζ(s) 31
4.1 The universality theorem . . . . . . . . . . . . . . . . . . . . . . 31
4.2 Self-similarity and the Riemann hypothesis . . . . . . . . . . . . 32

5 Appendix 33

1
1 Preface
The Riemann zeta-function ζ(s) is a meromorphic function. Many of the results
about ζ(s) that can be found with the usual tools of complex analysis have very
important consequences on number theory (see [4, p.43]). The growth of π(x)
for example can be related to the zeros of ζ(s), where π(x) is defined as the
amount of prime numbers less or equal to x (see [5, p.183]).
The first chapter will be devoted to basic results about ζ(s). In the second and
third chapter we will see some results about the zeros of ζ(s) and the behavior
of ζ(s) on the part of the complex plane where 0 < Re(s) < 1, which are both
of special interest to number theorists (see [5, p.133]).

I would like to thank my supervisor Dr.Francesc Bars for supporting me ex-

tensively.

2
2 Introduction
Definition. For s ∈ C we define
k
X
ζk (s) := n−s . (1)
n=1

Definition. If Re(s) > 1, then the Riemann zeta-function is defined as

∞
X
ζ(s) = lim ζn (s) = n−s . (2)
n→∞
n=1

We note that for Re(s) > 1 the sum on the right hand side converges uni-
formly and absolutely, since
 
P∞  ∞ ∞
−s 
|n−s | = n−Re(s) .
P P
n ≤


n=1 n=1 n=1

Proposition. ζ(s) is holomorphic in Re(s) > 1.

Proof. Apply theorem A.1 to ζn (s), with Ω being the half-plane Re(s) > 1.
Definition 2. Let U1 and U be regions of C such that U contains U1 . If f :
U1 → C and F : U → C are analytic functions such that F (s) = f (s), ∀s ∈ U1 ,
then F is called an analytic continuation of f .
The next proposition shows that analytic continuations are unique in a cer-
tain way.

Proposition. Let F1 , F2 : U → C be analytic continuations of f : U1 → C.

Then F1 (s) = F2 (s), ∀s ∈ U .
Proof. Since F1 and F2 are analytic on U, F1 − F2 is analytic too and we can
develop it as a Taylor-series around any point of U . Since F1 − F2 = 0 in
U1 the coefficients of the Taylor-series around a point on the border of U1 are
identically 0 and therefore F1 − F2 = 0 on an open ball around any point of
the border. On the border of these open balls we can develop F1 − F2 again as
a Taylor series and again the coefficients have to be 0. Repeating this process
until we have covered the whole of U with open balls, we see that F1 − F2 = 0
on U .

2.1 The Euler product

Now we introduce the Euler Product representation of ζ(s).

3
Theorem (Euler Product). If Re(s) > 1, then
Y 1
ζ(s) = , (3)
1 − p−s
p∈P

where P is the set of all prime numbers.

Proof. We will prove the identity for real s with Re(s) > 1 and conclude by
analytic continuation that it holds for all s with Re(s) > 1.
Obviously for every natural number n ≤ N , there is one and only one way to
mk −s
write n−s = (pm1 · · · pk )
1
, with p1 , ..., pk being different prime numbers and
in that representation m1 , ..., mk , p1 , ..., pk ≤ N .
Therefore the finite product
Y
(1 + p−s + p−2s + ... + p−N s ), (4)
p∈P,p≤N

is a finite sum of terms of the form m−s , which are all positive, since m are
natural and s real numbers. Furthermore, for every n ≤ N, n−s is ’contained’
in this sum and therefore

Y
ζN (s) = 1 + 2−s + ... + N −s ≤ (1 + p−s + p−2s + ... + p−N s ), (5)
p∈P,p≤N

and on the other hand

Y Y ∞
X
−s −2s −N s
(1 + p +p + ... + p )≤ (p−s )i
p∈P,p≤N p∈P,p≤N i=0
(6)
Y 1 Y 1
= ( )≤ ( ).
1 − p−s 1 − p−s
p∈P,p≤N p∈P

Hence
Y 1
ζ(s) = lim ζN (s) ≤ ( ). (7)
N →∞ 1 − p−s
p∈P

Now we have to proof ζ(s) ≥ p∈P ( 1−p1 −s ). Therefore we consider the finite
Q
product Y
(1 + p−s + p−2s + ... + p−M s ), M ≥ N (8)
p∈P,p≤N

and observe that this is a finite sum of terms of the form n−s in which no term is
contained more than once, because for any n−s there is only one way to write it
as a product of p−m
1
1s
, ..., p−m
k
ks
, with p1 , ..., pk being different prime numbers.
Hence

4
 Y ∞
X
−s −2s −M s
(1 + p +p + ... + p )≤ n−s = ζ(s), (9)
p∈P,p≤N n=1

and letting M and N tend to infinity

Y 1 Y 1
( ) = lim ( ) (10)
1 − p−s N →∞ 1 − p−s
p∈P p∈P,p≤N
Y
= lim lim (1 + p−s + p−2s + ... + p−M s ) (11)
N →∞ M →∞
p∈P,p≤N
∞
X
≤ n−s = ζ(s), (12)
n=1

which completes the proof.

2.2 Riemann’s functional equation for ζ(s)

In order to prove Riemann’s functional equation we will introduce the functions
ψ(s) and Γ(s) and discuss some of their properties. To prove a certain prop-
erty of ψ(s) that has an important role in the proof of Riemann’s functional
equation we will introduce Poisson summation formula. Furthermore we will
introduce the Mellin transform and its inversion formula, since in the proof of
Riemann’s functional equation there will be several integrals that are precisely
Mellin transforms. The function Γ(s), for example, turns out to be the Mellin
transform of e−s .

2.2.1 The Poisson Summation Formula

Definition. If it exists, the Fourier transform of f is defined as
Z ∞
f (t) =
b f (x)e2πitx dx, ∀t ∈ R. (13)
−∞

Definition. RA differentiable function f : R → C is said to be of bounded total

∞
variation, if −∞ |f 0 (x)|dx converges.
Theorem (Poisson Summation Formula). If f : R → R is an integrable
contiuous function of bounded total variation, such that there exists a bounded
function g : R → R+ that satisfies:
1)g(x) = g(−x)
2)g(x)
R ∞ is decreasing in R+
3) −∞ g(x)dx converges
4)|f (x)| < g(x), ∀x ∈ R,
then: X X
f (x + n) = lim fb(m)e2πimx (14)
N →∞
n∈Z m∈Z,|m|<N

5
and in particular, with x = 0, we obtain the formula which is usually called the
Poisson summation formula:
X X
f (n) = lim fb(m) (15)
N →∞
n∈Z m∈Z,|m|<N
R∞ P
Proof. As −∞ g(x)dx converges and g(x) is decreasing in R+ , Pn∈Z g(x + n)
converges. Since |f (x)| < g(x), ∀x ∈ R, by Weierstrass criterionP n∈Z f (x + n)
converges
P to a continuous function F (x). Furthermore F (x) = n∈Z f (x+n) =
n∈Z f (x + 1 + n) = F (x + 1), ∀x ∈ R and therfore F (x) is periodic. The Four-
rier coefficients are:

Z Z X
Fb(m) = F (x)e2πimx dx = f (x + n)e2πimx dx
[0,1] [0,1] n∈Z
XZ XZ
= f (x + n)e2πimx dx = f (x)e2πimx dx (16)
n∈Z [0,1] n∈Z [n−1,n]
Z ∞
= f (x + n)e2πimx dx = fb(m)
−∞

And finally:
X X X
f (x+n) = F (x) = lim Fb(m)e2πimx = lim fb(m)e2πimx
N →∞ N →∞
n∈Z m∈Z,|m|<N m∈Z,|m|<N
(17)

Corollary.
R∞ Let f R: R → C be a function of total bounded variation such that
∞
−∞
|f (x)|dx and −∞ |fb(s)|ds converge and fb is also of total bounded variation.
For u 6= 0 the following holds:
X X
f (nu) = |u|−1 lim fb(mu) (18)
N →∞
n∈Z m∈Z,|m|<N

−1 b x
u (x) = |u|
Proof. If we define fu (x) = f (ux), then fc f ( u ). Substituting this
in the proof of the previous theorem we get the result.

2.2.2 The Γ-function and ψ-function

R ∞ s−1 −x R1 R∞
Since 0 x e dx = 0 xs−1 e−x dx + 1 xs−1 e−x dx = A + B and A ∼
R 1 s−1 R∞
0
x dx and B ∼ 1 e−x dx, the integral in the following definition converges
whenever Re(s) > 0.
R∞
Definition. For s ∈ C with Re(s) > 0, we define Γ(s) = 0 xs−1 e−x dx.

6
Theorem. The function Γ(s) initially defined for Re(s) > 0 has an analytic
continuation to a meromorphic function on C whose only singularities are simple
poles at the negative integers s = 0, −1, ....
A proof of this can be found in [7, p.161].
Proposition. Γ(s) has no zeros.
1
Proof. According to [7, p.165], Γ(s) is an entire function.
P∞ 2
Proposition. The sum n=−∞ e−πn s
converges absolutely when Re(s) > 0.
Proof.
∞   ∞ ∞
X  −πn2 s  X
−πn2 Re(s)
X 2
e = e <2 e−πn Re(s)
n=−∞ n=−∞ n=0
∞
(19)
X
−πnRe(s) 1
<2 e =2 .
n=0
1 − e−πRe(s)

P∞ 2
Definition. For s ∈ C, Re(s) > 0 we define the function ψ(s) = n=−∞ e−πn s
=
P∞ −πn2 s
−∞ fs (n), where fs (n) = e .
Theorem.
1 1
ψ(s) = √ ψ( ), ∀s ∈ C, Re(s) > 0 (20)
s s
√ √
where s is taken over the branch that makes s a real positive number when
s is a real positive number.
P∞
Proof. Due to Poisson summation formula ψ(s) = −∞ fbs (n). So in order to
apply Poisson summation formula we calculate the Fourier coefficients

Z ∞ Z ∞
2 πn2 in 2
fbs (n) = e−πx s e−2πinx dx = e− s e−πs(x+ s ) dx
−∞ −∞
∞ πn2 ∞
(21)
e−
2
Z Z
s
− πn −πsz 2 −t2
=e s e dz = √ e dt.
−∞ πs −∞
√
It is well known that the value of the last integral is π and therefore
πn2
e− s
fbs (n) = √ . (22)
s
So now we are in position to apply Poisson summation formula and obtain

2
∞ ∞ ∞ πn ∞
X X X e− s 1 X − πn2 1 1
ψ(s) = fs (n) = fbs (n) = √ =√ e s = √ ψ( ). (23)
−∞ −∞ −∞
s s −∞ s s

7
2.2.3 The Mellin Transform
Definition. If it exists, the Mellin transform of a function f : U → C, R ⊂ U
is defined as Z ∞
M (f, s) = f (x)xs−1 dx, (24)
0
where the integral is taken over the positive real line.
Observation. By the way we defined Γ(s) in the previous section it is obvious
that it is just the Mellin transform of e−x .
Definition. A function f : U → C, R ⊂ U is said to be of type (α, β), if for all
s = σ + it with α < σ < β the integral
Z ∞
M (f, s) = f (x)xs−1 dx (25)
0

converges.
Let us study of what type the function e−x is. Since
Z ∞
e−x xσ+it−1 dx = Γ(σ + it) (26)
0

converges whenever σ > 0, e−x is of type (0, ∞).

Mellin inversion theorem. If f is of type (α, β), continuous, and for α <
σ < β the function x → f (x)xσ−1 is of bounded total variation, then the Mellin
transform can be inverted in the following way
Z σ+i∞
1
f (x) = M (f, s)x−s ds. (27)
2πi σ−i∞

See [6, p.122].

Notation: In this paper the path of integration of integrals as the one in

Mellin inversion theorem have to be understood as
Z σ+i∞ Z σ+it
f (s)ds = lim f (s)ds. (28)
σ−i∞ t→∞ σ−it

We note that since f is of type (α, β), the function M (f, s)x−s is analytic
on the strip α < σ < β and by Cauchy’s theorem the value of the integral does
not depend on σ.
In particular, since Γ(s) is the Mellin transform of e−x and e−x is of type (0, ∞)
we have Z σ+i∞
1
e−x = Γ(s)x−s ds, (29)
2πi σ−i∞
for any σ > 0. We will use identity 29 in the proof of Riemann’s functional
equation.

8
2.2.4 Riemann’s functional equation for ζ(s)
In this subsection we will find an analytic continuation of ζ(s) and present and
prove Riemann’s functional equation.
Theorem. For s ∈ C with Re(s) > 1 the following equality holds

1 ∞
Z
s − 2s 1 1 1−s s
Γ( )ζ(s)π =− − + (ψ(v) − 1)(v 2 −1 + v 2 −1 )dv. (30)
2 s 1−s 2 1

Proof. We have seen, that Γ(s) is the Mellin transform of e−x . Inverting the
Mellin transform, which we can do with any σ > 0, and making the change of
variable x = πn2 u, we obtain
Z σ+i∞
−πn2 u 1
e = Γ(s)(πn2 u)−s ds, ∀σ > 0. (31)
2πi σ−i∞

2 2
Since e−πn u
= e−π(−n) u ,
∞ −1 ∞
X 2 X 2 X 2
ψ(u) = e−πn u
= e−πn u
+1+ e−πn u

n=−∞ n=−∞ n=1

∞ ∞ Z σ+i∞ (32)
X 2 X 1
=1+2 e−πn u
=1+2 Γ(s)(πn2 u)−s ds, ∀σ > 0.
n=1 n=1
2πi σ−i∞

But as the sum and the integral converge absolutely, we can change the order.
With the restriction σ > 1 we have
Z σ+i∞ ∞
2 X
ψ(u) = 1 + Γ(s)( n−2s )(πu)−s ds
2πi σ−i∞ n=1
Z σ+i∞ (33)
2 −s
=1+ Γ(s)ζ(2s)(πu) ds
2πi σ−i∞

and Z σ+i∞
1
ψ(u) − 1 = 2Γ(s)ζ(2s)π −s u−s ds, (34)
2πi σ−i∞

which tells us that 2Γ(s)ζ(2s)π −s is the Mellin transform of ψ(u) − 1 (ap-

plying Mellin inversion theorem).
So by the definition of the Mellin transform
Z ∞
2Γ(s)ζ(2s)π −s = (ψ(u) − 1)us−1 du. (35)
0

With the simple change of variable s = z/2 we obtain

9
 Z ∞
z z z
2Γ( )ζ(z)π − 2 = (ψ(u) − 1)u 2 −1 du
2 0
Z 1 Z 1 Z ∞
z z z
= ψ(u)u 2 −1 du − u 2 −1 du + (ψ(u) − 1)u 2 −1 du
0 0 1
Z 1 Z ∞
z 2 z
= ψ(u)u 2 −1 du − + (ψ(u) − 1)u 2 −1 du.
0 z 1
(36)

Substituting ψ(u) = √1u ψ( u1 ) and taking the change of variable u = v1 we

have
Z 1 Z 1
z 1 z 3
ψ(u)u 2 −1 du = ψ( )u 2 − 2 du
0 u
Z0 ∞
z 1
= ψ(v)v − 2 − 2 dv
Z1 ∞ Z ∞ (37)
− z2 − 12 − z2 − 12
= (ψ(v) − 1)v dv + v dv
Z1 ∞ 1
1−z 2
= (ψ(v) − 1)v 2 −1 dv −
1 1 − z

and consequently

1 1 12 1 ∞
Z Z
z z z z
Γ( )ζ(z)π − 2 = ψ(u)u 2 −1 du − + (ψ(u) − 1)u 2 −1 du
2 2 0 2z 2 1
1 ∞ 1 1 ∞
Z Z
1−z 1 2 z
= (ψ(v) − 1)v 2 −1 dv − − + (ψ(u) − 1)u 2 −1 du (38)
2 1 21−z z 2 1
1 ∞
Z
1 1 1−z z
=− − + (ψ(v) − 1)(v 2 −1 + v 2 −1 )dv.
z 1−z 2 1

Theorem. There exists an analytic continuation of ζ(s) to the whole complex

plane.
Proof. Since ψ(v) − 1 = O(e−πv ) as v → +∞,1 the integral
Z ∞
1−s s
(ψ(v) − 1)(v 2 −1 + v 2 −1 )dv (39)
1

converges absolutely and uniformly on the half-plane Re(s) > x for any x.
1 We write f (x) = O(g(x)) as x → ∞, if there exist positive real constants C and x , such
0
that |f (x)| < C |g(x)| , ∀x > x0 .

10
Therefore we can express the integral as
Z ∞ Z ∞ N
1−s s
X 2 1−s s
(ψ(v) − 1)(v 2 −1 + v 2 −1 )dv = ( lim 2 e−πn v )(v 2 −1 + v 2 −1 )dv
1 1 N →∞
n=1
N Z ∞
X 2 1−s s
= lim 2 e−πn v (v 2 −1 + v 2 −1 )dv
N →∞ 1
n=1
= lim fN (s),
N →∞
(40)
with each fN (s) being holomorphic. Now applying theorem A.1 we see that
Z ∞
1−s s
(ψ(v) − 1)(v 2 −1 + v 2 −1 )dv (41)
1

is holomorphic on C. Consequently
π s/2 1 ∞
 Z 
1 1 1−s
−1 s
−1
− − + (ψ(v) − 1)(v 2 +v 2 )dv (42)
Γ(s/2) s 1−s 2 1
is a meromorphic function on the whole complex plane. Since for s with Re(s) >
1 it takes the same values as ζ(s), it provides an analytic continuation of ζ(s).

An important point about 30 is, that its right hand side has the same value
in s and 1 − s. That implies
Corollary (Riemann’s functional equation). For all s ∈ C
s s 1−s 1−s
Γ( )ζ(s)π − 2 = Γ( )ζ(1 − s)π − 2 . (43)
2 2
Proof.
1 ∞
Z
s − 2s 1 1 1−s s
Γ( )ζ(s)π =− − + (ψ(v) − 1)(v 2 −1 + v 2 −1 )dv
2 s 1−s 2 1
1 1 ∞
Z
1 s 1−s
=− − + (ψ(v) − 1)(v 2 −1 + v 2 −1 )dv (44)
1−s s 2 1
1−s 1−s
= Γ( )ζ(1 − s)π − 2
2
.
Definition. We define Riemann’s ξ-function as
1 s s
ξ(s) := s(s − 1)Γ( )ζ(s)π − 2 , ∀s ∈ C. (45)
2 2
Note. Using ξ(s), Riemann’s functional equation can be expressed as
ξ(s) = ξ(1 − s), ∀s ∈ C. (46)

11
2.3 Extending ζ(s) via η(s)
In this section we are going to introduce the Dirichlet eta-function η(s) and
discuss its relation with ζ(s).

Proposition. The sum

∞
X
n−s (−1)n+1 (47)
n=1

is uniformly convergent in any closed subset of Re(s) > 0.

Proof. When s0 is a positive real number, n−sP0
is a strictly decreasing and
∞
real sequence that tends to zero and therefore n=1 n−s0 (−1)n+1 converges.
Applying theorem A.4 we conclude the proof.
Definition. For s ∈ C with Re(s) > 0 we define the Dirichlet η-function as
∞
X
η(s) = n−s (−1)n+1 . (48)
n=1

Proposition. ζ(s) can be extended to Re(s) > 0 via

η(s)
ζ(s) = . (49)
1 − 21−s
Proof. For s with Re(s) > 1 we have

(1 − 21−s )ζ(s) = ζ(s) − 21−s ζ(s)

X∞ ∞
X
−s
= n −2 1−s
n−s
n=1 n=1
X∞ ∞
X
= n−s − 2 (2n)−s (50)
n=1 n=1
= 1 − 2−s + 3 − 4−s + ...
−s

X∞
= n−s (−1)n+1 := η(s).
n=1

However η(s) converges for every s with Re(s) > 0 and consequently
P∞
n−s (−1)n+1 η(s)
ζ(s) = n=1 1−s
= (51)
1−2 1 − 21−s
gives an analytic continuation of ζ(s) to Re(s) > 0.
Proposition. Besides a simple pole in s = 1, ζ(s) has no other poles.

12
 P∞
Proof. Since η(1) = n=1 n−1 (−1)n+1 = log(2) 6= 0 and 1−211−s has a simple
pole in s = 1 we deduce that ζ(s) = η(s) 1−211−s has a simple pole in s = 1.
P∞
Furthermore as 1−211−s has no other poles in Re(s) > 0 and η(s) = n=1 n−s (−1)n+1
converges for any s with Re(s) > 0, we deduce that ζ(s) has no other poles in
Re(s) > 0 and due to Riemann’s functional equation it has no other poles on
the whole complex plane.
As this expression of ζ(s) converges on the critical strip (0 < Re(s) < 1)
and it is ’only’ an infinite sum, which is in a certain way easier to approximate
numerically than the infinite products or improper integrals which arise in other
analytic continuations of ζ(s), it could be a candidate to approximate values of
ζ(s) on the critical strip.
Another
PNgood thing about this expression is that for real x with 0 < x < 1, the
series n=1 n−x (−1)n+1 alternates and therefore it complies
2N
X 2N
X +1
n−x (−1)n+1 < η(x) < n−x (−1)n+1 (52)
n=1 n=1

which gives a bound for the error of approximating η(x) (and therefore ζ(x))
P2N +1
by n=1 n−x (−1)n+1 .

Lemma. Suppose Re(s) > 0. Then

X 1 1

− =a<∞ (53)
ns (n + 1)s
n≡1(mod2)

if and only if
X 1 
n+1
(−1) = b < ∞. (54)
ns
n∈N

In that case a = b.
Proof. If
X 1 1
− =a<∞ (55)
ns (n + 1)s
n≡1(mod2)

for any  there exists a natural N ( 2 ) such that

 
 

a −
X 1 1 < 

s
− s
(56)
 n (n + 1)  2
 n≡1(mod2),n≤K 

for any natural K ≥ N ( 2 ). Now since Re(s) > 0, 1

tends to 0 as n
 s 
(n+1)

goes to infinity. Therefore we can find N 0 ( 2 ) such that  (n+1)s  < 2 for any
 1 

n > N 0 ( 2 ).

13
 So if K ≥ N ( 2 ), N 0 ( 2 ) we have:
case 1: If K = 2k + 1

    
 X 1     
1 X 1 1
(−1)n+1  = a − 
   
a − + − 
 n s (2k + 1) s n s (n + 1) s 
 n≤2k+1   n≡1(mod2),n≤2k 
 
    
 X 1 1   1  ≤  + .

≤ a −

s
− s
 +
s
n (n + 1)   (2k + 1)  2 2
 n≡1(mod2),n≤2k
(57)

case 2: If K=2k
   
 X 1    
X 1 1  ≤ .
(−1)n+1  = a −
   
a − − (58)
 n s n s (n + 1) s  2
 n≤2k   n≡1(mod2),n≤2k 

Now we will see a bound for the error, when estimating η(s) for any complex
number s with Re(s) > 0 by the sum of its first 2N terms:
2N  
X 1 X 1 X 1 1
η(s)− s
(−1)n+1 = (−1)n+1 = − .
n=1
n ns ns (n + 1)s
n≥2N +1 n≥2N +1,n≡1(mod2)

  (59)
So we will search an upper bound for  n1s − 1
 
(n+1)s .

n
Lemma. If s = a + bi, a > 0 and −1 < log( n+1 ) |b| < 0, then
    √
1
 − 1 < 1 −
 1 2 |b|
 ns + 1+a . (60)
(n + 1)s  na (n + 1)a n

14
Proof.
 2   
1 1  = 1 − 1 1 1

 − −
 ns (n + 1)s  ns (n + 1)s ns (n + 1)s
s s
(n + 1)s − ns
   
(n + 1) − n
= s
(n(n + 1)) (n(n + 1))s
(n + 1)a+bi − na+bi (n + 1)a−bi − na−bi
  
=
(n(n + 1))a+bi (n(n + 1))a−bi
(n + 1)a+bi (n + 1)a−bi + na+bi na+bi − na+bi (n + 1)a−bi − na−bi (n + 1)a+bi
 
=
(n(n + 1))2a
 h i
n bi n −bi
(n + 1)2a + n2a − (n(n + 1))a ( n+1 ) + ( n+1 )
= 
(n(n + 1))2a
 h i
n bi
(n + 1)2a + n2a − (n(n + 1))a 2Re(( n+1 ) ))
= 
(n(n + 1))2a
 h i
n
(n + 1)2a + n2a − (n(n + 1))a 2cos(log( n+1 )b)
=  = (∗)
(n(n + 1))2a
(61)

Now we are going to use Taylor series in order to find a lower bound of
n
cos(log( n+1 )b) and therefore an upper bound of (*).
For −1 < x < 1 we have

cos(x) = 1 − x2 + x4 − x6 ... > 1 − x2 . (62)

Furthermore for n ≥ 1
 2  3 !
n 1 1 1 1 1 1 1
log( ) = −log(1 + ) = − − + ... > − . (63)
n+1 n n 2 n 3 n n

n
Since cos(x) is strictly increasing in (−1, 0) and by hypothesis −1 < log( n+1 ) |b| <
0, we have
n b b2
cos(log( )b) > cos(− ) > 1 − 2 . (64)
n+1 n n

15
If we plug this inequality into (*) we obtain
h  i 
b2


1 1
2
 (n + 1)2a + n2a − (n(n + 1))a 2 1 − n2
 −  < 
 ns (n + 1)s  (n(n + 1))2a
 2

2
((n + 1)a − na ) + 2(n(n + 1))a nb 2
=
(n(n + 1))2a (65)
2
(n + 1)a − na 2b2

= + a
(n(n + 1))a (n(n + 1)) n2
2
2b2

1 1
≤ a
− a
+ a
n (n + 1) (n(n)) n2
 2
2b2 √
Since ≤ n1a − (n+1)
1
a and (n(n)) a 2 are both positive and
n x is an increasing
and concave function for real positive x
  s 2
1
 − 1  1 1 2b2
 ns s
<
a
− a
+ a
(n + 1)  n (n + 1) (n(n)) n2
s 2 s
1 1 2b2 (66)
< a
− a
+ a
n (n + 1) (n(n)) n2
  √
1 1 2 |b|
= a
− a
+ 1+a
n (n + 1) n

PN 1 n+1
PN 1
We define ηN (s) := n=1 ns (−1) and ζN (s) = 1 ns

Proposition. If s = a + bi, a > 0 and −1 < log( NN+1 ) |b| ≤ 0, then

√
|η(s) − η2N (s)| < η(a) − η2N (a) + 2 |b| (ζ(1 + a) − ζ2N (1 + a)) (67)

16
Proof.
 
 
 X 1 1 
|η(s) − η2N (s)| = 

s
− s

n≥2N +1,n≡1(mod2) n (n + 1) 
 
X 1 1 
≤  −
 ns s

(n + 1) 
n≥2N +1,n≡1(mod2)
  √
X 1 1 2 |b|
< a
− a
+ 1+a
n (n + 1) n
n≥2N +1,n≡1(mod2)
  √
X 1 1 X 2 |b|
= − +
na (n + 1)a n1+a
n≥2N +1,n≡1(mod2) n≥2N +1,n≡1(mod2)
  √
X 1 1 X 2 |b|
< − +
na (n + 1)a n1+a
n≥2N +1,n≡1(mod2) n≥2N +1
√
= η(a) − η2N (a) + 2 |b| (ζ(1 + a) − ζ2N (1 + a)) .
(68)
The inequality between the second and the third line comes from the previous
Lemma.
We note that since log( NN+1 ) is negative and tends to zero when N goes
to infinity, given b we can always chose N sufficiently large in order to have
−1 < log( NN+1 ) |b| ≤ 0.

3 Zeros of ζ(s)
3.1 Weierstrass infinite products
Notation: Let f : R → R and F : R2 → R. We write
f (R) << F (R, ) (69)
if for any  there exists C such that
f (R) < C F (R, ), ∀R. (70)
Theorem. Suppose g(s) is an entire function and not identically 0. Let Ω
denote the set of zeros of g(s) and vα the multiplicity of the zero α (let v0 = 0
if g has no zero in s = 0). Suppose also that
 
log max |g(s)| << (R + 1)1+ . (71)
|s|=R

Then:
Y  s  s  vα
g(s) = sv0 eA+Bs 1− eα (72)
α
α∈Ω,α6=0

17
 A proof of this theorem is given in [4, p.328].
The previous theorem does not apply directly to ζ(s) since ζ(s) is not an entire
function. However we know that ζ(s) has only one pole (s = 1) and we have
also seen that this pole has order 1. Hence (s − 1)ζ(s) is an entire function. In
[4, p.20] it is shown that the second condition of the previous theorem is also
met and therefore we can write
sv0 A+Bs Y  s  s  vα
ζ(s) = e 1− eα , (73)
s−1 α
α∈Ω,α6=0

where Ω denotes the set of zeros of ζ(s). Since ζ(0) = − 12 6= 0 we can specify A
and v0 from the previous expression to obtain
1 Y  s  s  vα
ζ(s) = eBs 1− eα . (74)
2(s − 1) α
α∈Ω,α6=0

3.2 The Riemann hypothesis

Proposition. If s ∈ R then ζ(s) ∈ R.
η(s)
Proof. For real s with Re(s) > 0 we have ζ(s) = 1−2 1−s and since η(s) is a sum

of real numbers, ζ(s) is real. Due to Riemann’s functional equation ζ(s) is real
whenever s is real.
Proposition. For all s ∈ C we have ζ(s) = ζ(s).
Proof. Apply Schwarz reflection principle (see theorem A.4) to ζ(s)(s−1) = p(s)
(which is an entire function, since the only pole of ζ(s) is a simple one in
s = 1).
Proposition. ζ(s) has simple zeros at the negative even integers and nowhere
else on the real line. These zeros are called the trivial zeros of ζ(s).
Proof. From Riemann’s functional equation we know that
z z 1−z 1−z
Γ( )ζ(z)π − 2 = Γ( )ζ(1 − z)π − 2 . (75)
2 2
If z is a negative even integer then the right hand side of the previous equation
takes a finite value different from 0, since each of the terms Γ( 1−z
2 ), ζ(1 − z) and
− 1−z
π 2 takes a finite value different from 0. But since Γ has simple poles at each
negative integer (but nowhere else on the real line), Γ( z2 ) has simple poles at
the negative even integers and consequently ζ has simple zeros at the negative
even integers (and nowhere else on the real line).
In the next theorem we will see that besides the trivial zeros ζ(s) has more
zeros. They are called the non-trivial zeros.

18
Theorem. ζ has only zeros at the negative even integers and in a set of complex
numbers αn that lie in the critical strip (0 < Re(s) < 1). Furthermore the zeros
on the critical strip are situated symmetrically with respect to the real line and
to Re(s) = 21 .
Proof. Define the Möbius function µ(s) by

 1 if n = 1,
µ(n) = (−1)k if n = p1 ...pk , and p1 , ..., pk are distinct primes (76)
0 otherwise.


From the Euler-product we know that

∞
1 Y X µ(n)
= (1 − p−s ) = (77)
ζ(s) n=1
ns
p∈P

and therefore
  ∞  ∞ ∞ Z n+1 Z ∞
 1   X µ(n)  X 1 X
−σ σ
= < < 1+ x dx = 1+ x−σ dx =
 
 ζ(s)   n s  n σ σ − 1
n=1 n=1 n=1 n 1
(78)
From the previous inequality we deduce that ζ(s) has no zeros with Re(s) > 1
and due to Riemann’s functional equation it has no zeros with Re(s) < 0, except
for the zeros at the negative even integers.
Now we need to proof the symmetries of the zeros.
Since ζ(s) = ζ(s), s is a zero if and only if s is a zero. Finally, due to Riemann’s
functional equation 1 − s (the reflection of s on Re(z) = 12 ) is a zero too.
The Riemann hypothesis. If s is a non-trivial zero of ζ(s), then Re(s) = 21 .
At the present, the Riemann hypothesis is considered one of the most im-
portant unsolved problems in mathematics, as its verification would imply a lot
of other results in mathematics, especially in number theory.

3.3 Some results about the distribution of the zeros of ζ(s)

From now on log(s) is always taken in such a way, that it is a real number, when
s is a real positive number.
Theorem. Let Ω be the set of zeros of ζ(s). Then the following equation holds
∞ 
ζ 0 (s)
  
1 X 1 1 X 1 1
= +B+ + + − (79)
ζ(s) 1−s s−α α n=1
s + 2n 2n
α∈Ω,Im(α)6=0

Proof. We have seen that

1 Y  s  s  vα
ζ(s) = eBs 1− eα (80)
2(s − 1) α
α∈Ω

19
and therefore
  0
ζ 0 (s) 0 1 Y  s  s 
= (log(ζ(s))) = log  eBs 1− eα
ζ(s) 2(s − 1) α
α∈Ω,α6=0
!0
0
X  s s
= c − log(s − 1) + Bs + log 1 − +
α α
α∈Ω
1 X 1 1
= +B+ +
1−s s−α α
α∈Ω
   
1 X 1 1 X 1 1
= +B+ + + +
1−s s−α α s−α α
α∈Ω,Im(α)6=0 α∈Ω,Im(α)=0
  X ∞  
1 X 1 1 1 1
= +B+ + + − .
1−s s−α α n=1
s + 2n 2n
α∈Ω,Im(α)6=0
(81)
The last step is justified, because we know that the real zeros of ζ(s) are precisely
the negative, even integers.
Theorem. Let Ω be the set of zeros of ζ(s) and T ≥ 2. Then
X 1
< c log(T ). (82)
1 + (T − Im(α))2
α∈Ω,Im(α)6=0

For a proof see [4, p.23].

Definition. We define the function N(T) as
N (T ) = |{α ∈ Ω : 0 ≤ Re(α) ≤ 1, 0 ≤ Im(α) ≤ T }| . (83)
Corollary. There exists an absolute constant c such that N (T + 1) − N (T ) ≤
c log(T ) when T ≥ 2.
Proof. For α ∈ Ω with T ≤ Im(α) ≤ T + 1 we have
1 1 1
= ≤ . (84)
2 1+1 1 + (T − Im(α))2
Hence
X 1
(N (T + 1) − N (T )) = 2
2
α∈Ω,T ≤Im(α)≤T +1
X 1
≤2
1 + (T − Im(α))2 (85)
α∈Ω,T ≤Im(α)≤T +1
X 1
≤2
1 + (T − Im(α))2
α∈Ω,Im(α)6=0

≤ 2c log(T ).

20
The last inequality is due to the previous theorem.
The theorem of De la Vallée-Poussin bounding the zeros of ζ(s). There
exists a positive real constant c such that there are no zeros in the region:
c
Re(s) > 1 − . (86)
log(|Im(s)| + 2)

For a proof see [4, p.25].

This theorem does not tell us a lot about zero-free regions in the interior of
the critical strip (0 < Re(s) < 1) since c could be arbitrary small. However it
tells us that there are no zeros with Re(s) = 1, which is essential in the proof
of the prime-number theorem.

3.4 Numerical computation of zeros of ζ(s)

3.4.1 Using Euler-Maclaurin summation to compute values of ζ(s)
Definition. We define the Bernoulli functions, Pn (x), recursively.
1
P0 (x) = 1, P1 (x) = x − [x] − ,
2
1 0
Pn−1 (x) = Pn (x), (87)
n
Z k+1
Pn (x)dx = 0, ∀k ∈ Z,
k

where [x] denotes the greatest integer that is smaller or equal to x.

Note that the condition Pn−1 (x) = n1 Pn0 (x) determines Pn (x) up to a con-
R k+1
stant c, which is fixed by the condition k Pn (x)dx = 0.
Since P1 (x) is periodic (the value in x and x + 1 are obviously the same), the
two previous conditions let Pn (x) be periodic too.
Definition. We define the Bernoulli numbers as Bn := Pn (0).
P∞ n
The previous definition is equivalent to exx−1 = n=0 Bn · xn! . Furthermore
It can be shown that B3 = B5 = B7 = ... = 0.
The Euler-McLaurin summation formula. Let f (x) be a differentiable
function and k ∈ Z, q ∈ N. The following equation holds
Z k+1
1 X B2m h (2m−1) i
f (x)dx = (f (k+1)+f (k))− f (k + 1) − f (2m−1) (k) +Rq
k 2 (q+1)
(2m)!
m≤ 2
(88)
where
k+1
(−1)q+1
Z
Rq = f (q+1) (x)Pq+1 (x)dx. (89)
(q + 1)! k

21
 See [3, p.119].

Theorem. For s ∈ C, n
b ∈ N the following holds

n ∞
b1−s n
n b−s X B2m
b
X
−s
ζ(s) = k + − + n−s−2m+1
s(s + 1)(s + 2)...(s + 2m − 2)b
s−1 2 m=1
(2m)!
k=1
(90)
Proof. Applying Euler-McLaurin summation formula to the function f (x) =
x−s we obtain
    
Z k+1 1 B2m  d(2m−1) d(2m−1)
 
−s −s −s −s  −s 
X 
x dx = ((k+1) +k )−  (x ) − (x ) +Rq
k 2 (2m)!  dx(2m−1)
 dx (2m−1) 
(q+1)

m≤ x=k+1 x=k
2
(91)

where
k+1
(−1)q+1 d(2m−1) −s
Z  
Rq = (x ) Pq+1 (x)dx. (92)
(q + 1)! k dx(2m−1)
Since
k+1
(−1)q+1 d(2m−1) −s
Z  
lim Rq = lim (x ) Pq+1 (x)dx = 0 (93)
q→∞ q→∞ (q + 1)! k dx(2m−1)

we have
    
Z k+1 1 ∞ B2m  d(2m−1) d(2m−1)
 
−s −s −s −s  −s 
X 
x dx = ((k + 1) +k )−  (x ) − (x ) .
k 2 (2m)!  dx(2m−1) dx (2m−1) 
m=1 
x=k+1 x=k

(94)

Summing this expression over all k ≥ n

b we obtain
∞ ∞ Z k+1
b1−s
Z
n −s
X
= x dx = x−s dx
1−s n k
b k=b
n
∞ ∞  (2m−1) 
X 1 −s X B2m d
= k −s + nb + (x −s
) (95)
2 m=1
(2m)! dx(2m−1) x=b
n
k=b
n+1
n ∞  (2m−1) 
1 −s X B2m d
b
X
= ζ(s) − k −s + nb + (x −s
) .
2 m=1
(2m)! dx(2m−1) x=b
n
k=1

Substituting
 (2m−1) 
d −s
(x ) n−s−2m+1
= s(s + 1)(s + 2)...(s + 2m − 2)b (96)
dx(2m−1) nb

and rearanging the terms we obtain the result.

22
3.4.2 Finding zeros on the critical line
Proposition. ξ(s) takes real values on the critical line.
Proof. Because of Riemann’s functional equation and Schwarz reflection princi-
ple we have

1 1 1 1 1
ξ( + it) = ξ(1 − ( + it)) = ξ( − it) = ξ( + it) = ξ( + it). (97)
2 2 2 2 2

Observation. Since ξ(s) takes real values on the critical line, we can use much
more simple methods to find its zeros.
1
Definition. For t ∈ R and s = 2 + it we define

s t
ϑ(t) := Im(log(Γ( ))) − log(π) (98)
2 2
and
s t 1 1
Z(t) := eiIm(log(Γ( 2 ))) π −i 2 ζ( + it) = eiϑ(t) ζ( + it), (99)
2 2
where s = 21 + it and the log(s) is taken in such a way that it is real, when s is
a real positive number.
Proposition. For t ∈ R, the sign of ξ( 21 + it) is always opposite to the sign of
Z(t).
1
Proof. If we set and s = 2 + it we have

1 1 s s s t s(s − 1) 1
ξ( + it) := s(s − 1)Γ( )ζ(s)π − 2 = e(log(Γ( 2 ))) π −i 2 ζ( + it)
2 2 2 2 2
2 1 (100)
1 −t −

s
Re(log(Γ( 2 ))) − 4 4
= e π Z(t),
2

where the term in the brackets is always negative.

So in order to see if there is a zero between two points on the critical line,
we have to check the sign of Z(t) = eiϑ(t) ζ( 12 + it) at these points, but we do not
need to know exact values of Z(t). For that purpose the following approximation
of ϑ(t) will be good enough.

Proposition. For t ∈ R
t t t π 1
ϑ(t) = log( ) − − + + e(t) (101)
2 2π 2 8 48t
7 2
where |e(t)| < 5760t3 + t5 .See [2, page.120].

23
3.4.3 Example
ζ(s) has a zero at p ≈ 0.5 + 14.156875i.

To prove that result we will search a rough estimate of Z(15) = eiϑ(15) ζ( 12 + 15i)
and check its sign. Therefore we will use the approximation of ϑ(t) of the previ-
ous proposition. To approximate ζ( 12 +15i) we will use the representation of ζ(s)
that we derived from the Euler-Maclaurin summation formula (90), truncating
the infinite series at j such that the difference between the estimates truncating
at j and j − 1 has absolute value less than 0.001. Clearly the greater we choose
n
b the faster the infinite series converges. We will set n
b = 30 and obtain

ϑ(15) ≈ −1.36501
1
j = 1 : ζ( + 15i) ≈ 0.147082 + 0.704736i (102)
2
1
j = 2 : ζ( + 15i) ≈ 0.14711 + 0.704752i.
2
Since the absolute value of the difference of the estimates that we obtained
with j = 1 and j = 2 is less than 0.001, we will use the approximation of
ζ( 12 + 15i) that we obtained with j = 2. With these approximations we obtain
an approximation of Z(15)
1
Z(15) = ζ( +15i)eϑ(15)i ≈ (0.14711+0.704752i)e−1.36501i ≈ 0.719942−3.99976·10−7 i,
2
(103)
which can be interpreted as a positive real number.

Now we will do the same for Z(14).

ϑ(14) ≈ −1.78294
1
j = 1 : ζ( + 14i) ≈ 0.0222604 − 0.10324i (104)
2
1
j = 2 : ζ( + 14i) ≈ 0.0222413 − 0.103258i.
2
Again we can use the approximation that we obtained with j = 2, since it is
close enough to the one with j = 1. So we have
1
Z(14) = ζ( +14i)eϑ(14)i ≈ (0.02224−0.103258i)e−1.78294i ≈ −0.105626−1.32961·10−7 i,
2
(105)
which can be interpreted as a negative real number.
Since the sign of Z(14) is opposite to the sign of Z(15), there is at least one
zero between t = 14 and t = 15.
Now we will search a smaller interval for that zero. Therefore we have to com-
pute Z(14.5) ≈ 0.297351 − 1.18712 · 10−7 i, which has an opposite sign to Z(14).
We conclude that there has to be a zero between t = 14 and t = 14.5. We repeat

24
this process of dividing intervals until the intervals are smaller than 0.1.

Z(14.25) ≈ 0.0922651 − 3.88103 · 10−8 i ⇒ I = (14, 14.25)

Z(14.125) ≈ −0.00770742 + 3.32897 · 10−9 i ⇒ I = (14.125, 14.25) (106)
−8
Z(14.18875) ≈ 0.0430373 − 1.83389 · 10 i ⇒ I = (14.125, 14.18875)

We could repeat this process until we have the accuracy that we want. For now
we have the estimation
(14.125 + 14.18875)
p ≈ 0.5 + i = 0.5 + 14.156875i. (107)
2

The 10 first zeros that we find with this method with an accuracy of 0.00001 are

p1 ≈ 0.5 + 14, 134725i

p2 ≈ 0.5 + 21.022040i
p3 ≈ 0.5 + 25.010858i
p4 ≈ 0.5 + 30.424876i
p5 ≈ 0.5 + 32.935061i
p6 ≈ 0.5 + 37.586178i
p7 ≈ 0.5 + 40.918719i
p8 ≈ 0.5 + 43.327073i
p9 ≈ 0.5 + 48.005151i
p10 ≈ 0.5 + 49.773832i.
Actually we know the first 20 zeros now, since the conjugate of each zero is a
zero too.

3.4.4 The number of zeros in a given set

Considering a set St = {s ∈ C, 0 < Re(s) < 1, 0 < Im(s) < t} such that there
are no zeros on its boundary, we will discuss the following question: If we have
found n zeros in St using the same method as in the previous example, how do
we know if that are all the zeros in St ?
The answer is theoretically simple: Compare n with the number given by
the argument principle
ζ 0 (s)
Z
m= ds. (108)
∂{St \B(1,)} ζ(s)

The interpretation of m is not that simple. We have to distinguish different

possibilities:

1.) If n = m, then we have found all the zeros in St , since ζ(s) has no poles in
the critical strip and therefore m is equal to the number of zeros in St .

25
2.) There are three possible reasons why n < m,
2.1) In one of the intervals there was a even number of zeros and therefore we
did not detect a change of sign. This can be avoided by taking other intervals.
2.2) One of the zeros that we detected has multiplicity greater than one.
2.3) The Riemann-hypothesis is false and there is a zero beyond the critical line.

3.4.5 A simple program in C to compute ζ(s)

Using the formula 90 we can make a program in C which computes ζ(s) over a
grid:

#i n c l u d e <s t d i o . h>
#i n c l u d e <math . h>
#i n c l u d e <complex . h>

l o n g d o u b l e B[ 5 ] =
{0.16666666666666666666666666666666666666 ,
−0.03333333333333333333333333333333333333 ,
0.02380952380952380952380952380952380952 ,
−0.03333333333333333333333333333333333333 ,
0.07575757575757575757575757575757575757};

void zeta ( double a , double b){

i n t i , N=150;
l o n g d o u b l e complex q=0+0∗ I ;
l o n g d o u b l e complex c=a+b∗ I ;
l o n g d o u b l e complex S1=1+0∗I , S2=0+0∗ I ;
i n t F=2, m=1;
l o n g d o u b l e complex s f ;
w h i l e ( c a b s ( S1−S2 ) >0.0000000000001){
s f =1+0∗ I ;
f o r ( i =0; i <2∗m−1; i ++){
s f=s f ∗ ( c+i ) ;
}
S1=S2 ;
S2=S2+B [m] ∗ s f ∗cpow (N,−c −2∗m+1)/F ;
m=m+1;
F=F∗2∗m∗ ( 2 ∗m−1);
}

f o r ( i =1; i <N+1 ; i ++){

q=q+cpow ( i ,− c ) ;
}
q=q+cpow (N,1− c ) / ( c−1)+cpow (N,− c )/2+S2 ;
p r i n t f (”%g %g %g %g \n ” , a , b , c r e a l ( q ) , cimag ( q ) ) ;

26
}

v o i d p r i n t ( i n t nx , i n t ny )
{

d o u b l e dx =0.01 , dy = 0 . 0 1 ;
int i , j ;
f o r ( j =0; j <ny +1; j ++){
f o r ( i =0; i <nx +1; i ++){
z e t a ( dx∗ i ,1+ dy∗ j ) ;
}
p r i n t f (”\n ” ) ;
}

main ( )
{

print (100 ,2000);

return 0;
}.
With the data that we obtain from that program we can generate interesting
plots in GnuPlot.

Figure 1: Re(ζ(s))

27
-2cm (a) Im(ζ(s)) (b) Re(ζ(s))

Figure 3: (c) |ζ(s)|

28
4 Universality of ζ(s)
4.1 The universality theorem
In this section we will discuss a very interesting theorem, that states that any
analytic function different from 0 can be approximated in a certain way by ζ(s).
This property is called universality.

Theorem (Universality of ζ(s)). Let 0 < r < 41 . Suppose that f (s) is a

function which is analytic for |s| < r and continuous for |s| ≤ r. If f (s) 6= 0
for |s| < r, then for any  > 0 there exists T () such that
3
max |f (s) − ζ(s + + iT ())| < . (109)
|s|≤r 4

For a proof of this theorem see [4, p.241].

Example 1 If we take the constant function f (s) = 1 6= 0, for any 2 > 0

we can find a T (2 ) such that
3
max |1 − ζ(s + + iT (2 ))| < 2 , (110)
|s|≤r 4

which implies
3
max |ζ(s + + iT (2 ))| < 1 + 2 := . (111)
|s|≤r 4
Since 1 and 2 can be chosen arbitrary small,  can be chosen arbitrary small
1
too. That means that for an arbitrary small number, for example  = 1000000 ,
and an arbitrary r < 0.25, somewhere on the critical strip there exists a closed
1
ball of radius r such that |ζ(s)| < 1000000 on this closed ball.

Example 2 For any function f (s), analytic and zero-free on a closed ball of
s
f ( 5R )
radius R, 5R (which can be seen as a miniature-version of f ) is analytic too,
and therefore there exists a closed ball of radius 15 , such that ζ is very close to
s
f ( 5R )
5R on this ball. So for any analytic function f there exists an approximated
miniature-version of this function somewhere on the critical strip.

Example 3 Consider any continuous, zero-free function f : [−r, r] → C, where

0 < r < 41 . Since f does not need to be analytic, we can also consider functions
like f (s) = 1 + |s| or f (s) = s cos( 1s ) + 1.
It is well known that there exists a sequence of polynomials pn (s) that converges
uniformly to f (s) on [−r, r]. Therfore given a 1 > 0 there is a pn (s) such that
maxs∈[−r,r] |f (s) − pn (s)| < 1 . Since pn (s) is a polynomial, it is analytic and
we can apply the universality theorem on it: For any 2 > 0 there exists T (2 )
such that maxs∈[−r,r] |pn (s) − ζ(s + 34 + iT (2 ))| < 2 . Consequently, for any

29
 := 1 + 2 there exists T () = T (2 ) such that
3
maxs∈[−r,r] |f (s) − ζ(s + + iT (2 ))|
4
3
< maxs∈[−r,r] |f (s) − pn (s)| + maxs∈[−r,r] |pn (s) − ζ(s + + iT (2 ))| < 1 + 2 .
4
(112)

Now we are going to discuss some theorems that can be derived from the
universality theorem.

Theorem. Let fσ : R 7−→ CN be defined as

fσ (t) = (log(ζ(σ + it)), (log(ζ(σ + it)))0 , ..., (log(ζ(σ + it)))(N −1) ), (113)

where log(s) is taken in such a way, that it is a real number, when s is a real
positive number. Then for σ with 21 < σ < 1, fσ (R) is dense in CN .

For a proof see [4, p.252].

Theorem. Let fσ : R 7−→ CN be defined as

fσ (t) = (ζ(σ + it), (ζ(σ + it))0 , ..., (ζ(σ + it))(N −1) ). (114)
1
Then for σ with 2 < σ < 1, fσ (R) is dense in CN .
For a proof see [4, p.253].
Theorem. Let F : CN 7−→ C be a continuous function. If

F (ζ(s), (ζ(s))0 , ..., (ζ(s))(N −1) ) = 0, ∀s ∈ C, (115)

then F ≡ 0.
Proof. If F was not identically 0, then there would exist a point s ∈ CN such
that F (s) 6= 0. Since F is continuous by hypothesis, there would also exist an
open neighborhood A of s, such that F is different from 0 for each point of A.
However from the previous theorem we know that (ζ(s), (ζ(s))0 , ..., (ζ(s))(N −1) )
is dense in CN and therefore there would exist s ∈ C such that (ζ(s), (ζ(s))0 , ...,
(ζ(s))(N −1) ) ∈ A and consequently F (ζ(s), (ζ(s))0 , ..., (ζ(s))(N −1) ) 6= 0. But
that is a contradiction!

4.2 Self-similarity and the Riemann hypothesis

In order to link the universality theorem to the topics of the previous chapters
we may ask if this theorem could be used to find zeros of ζ(s) or to prove the
non-existence of zeros beyond the critical line. In this section we will see a
theorem that shows that indeed there is a surprising connection.

30
Consider an open ball B = B(s0 , r) where r < 41 and 1 ∈ / B. Then ζ(s) it-
self is an analytic function on B and ζ(s − s0 ) is analytic on B(0, r). Due to the
universality theorem we know that for any  > 0 there exists T () such that
3
max |ζ(s − s0 ) − ζ(s + + iT ())| < . (116)
|s|≤r 4

So any shape that ζ(s) takes is repeated somewhere else on the critical strip,
which makes it self-similar.
In the following theorem we will see, that for a given compact set K with
Re(K) > 1 there are not only some places where ζ(s) repeats the behavior that
it had at K, but there are actually quite a lot places where it does this.
First we need a
Definition. The Lebesgue measure of the set A is defined as
(∞ ∞
)
X [
meas(A) = inf (bi − ai ) : [ai , bi ] ⊃ A . (117)
1 1

Theorem 4. Let U be a compact set such that Re(U ) > 1 and  > 0. Then
 
1
lim inf meas t ∈ [0, T ] , max |ζ(u + it) − ζ(u)| <  > 0. (118)
T →∞ T u∈U

For a proof see [8, p.57].

It turns out that the Riemann hypothesis is true if and only if the previous
theorem can be extended to certain sets U of the critical strip.
Theorem 4. The Riemann hypothesis is true if and only if for any compact
subset U with connected complement, such that 21 < Re(U ) < 1 and for any
 > 0 the following holds:
 
1
lim inf meas t ∈ [0, T ] , max |ζ(u + it) − ζ(u)| <  > 0. (119)
T →∞ T u∈U

For a proof see [8, p.54].

5 Appendix
Theorem A.1. If {fn } is a sequence of holomorphic functions that converges
uniformely to a function f in every compact subset of U , then f is holomorphic
on U .
For a proof see [1, p.53].

Let Ω denote an open subset Tof C such that s ∈ Ω ⇔ s ∈ Ω. Let Ω+ =

{s ∈ Ω : Re(s) > 0} and I = Ω R. Then

31
Theorem A.2 (Symmetry principle). If f + (s) and f − (s) are holomorphic
functions on Ω+ and Ω− respectively, that extend continuously to I and f + (s) =
f − (s), ∀s ∈ I, then the function
 +
 f (s) if s ∈ Ω+
+ −
f (s) = f (s) = f (s) if s ∈ I (120)
 −
f (s) if s ∈ Ω− .

is holomorphic on Ω.
For a proof see [7, p.58].
Theorem A.3 (Schwarz reflection principle). Suppose that f is a holo-
morphic function on Ω+ that extends continuously to I and such that f is real-
valued on I. Then there exists a function F (s) holomorphic in all of Ω such that
F (s) = f (s), ∀s ∈ Ω+ . Furthermore the function F (s) is given by

F (s) = f (s), s ∈ Ω− . (121)

For a proof see [7, p.60].

Since analytic continuations are unique we conclude that if F : Ω → C is an

analytic function and it is real-valued on I, then

F (s) = F (s), ∀s ∈ Ω. (122)

P∞
Theorem
P∞ A.4. Suppose the series n=1 an n−s0 is convergent. Then the se-
−s
ries n=1 an n is uniformly convergent in everyPclosed region contained in
∞
the region Re(s) > Re(s0 ). The function f (s) = n=1 an n−s is analytic for
Re(s) > Re(s0 ).
For a proof see [4, p.354]

References
[1] B.Bagchi, The statistical behavior and universality properties of
the Riemann zeta-function and other allied Dirichlet series, PHD
Thesis, Calcutta, Indian Statistical Institute, 1981
[2] H.M. Edwards, Riemann’s Zeta Function, Academic Press, 1974.
[3] Samuel w. Gilbert, The Riemann Hypothesis and The Roots Of The
Riemann Zeta Function, BookSurge Publishing, 2009

[4] A.A. Karatsuba, S.M. Voronin,The Riemann zeta-function, de Gruyter

1992.

32
[5] W. Narkievicz, The Development of Prime Number Theory, Springer,
2000.
[6] S.J.Patterson, An introduction to the theory of the Riemann Zeta-Function,
Cambridge University Press, 1988

[7] Elias M. Stein, Rami Shakarchi, Complex Analysis, Princeton lectures

in Analysis, 2003
[8] Jörn Steuding, On the Universality of the Riemann zeta-function,
Frankfurt, Johann Wolfgang Goethe-Universität Frankfurt, 2002

33