On the nontrivial zeros of the Dirichlet eta

function
arXiv:2007.04317v1 [math.GM] 9 Jul 2020

Vladimir Garcı́a-Morales

Departament de Fı́sica de la Terra i Termodinàmica

Universitat de València,
E-46100 Burjassot, Spain
garmovla@uv.es

We construct a two-parameter complex function ηκν : C → C, κ ∈ (0, ∞), ν ∈

(0, ∞) that we call a holomorphic nonlinear embedding and that is given by a
double series which is absolutely and uniformly convergent on compact sets in the
entire complex plane. The function ηκν converges to the Dirichlet eta function η(s)
as κ → ∞. We prove the crucial property that, for sufficiently P large κ, the function
ηκν (s) can be expressed as a linear combination ηκν (s) = ∞ n=0 an (κ)η(s + 2νn)
of horizontal shifts of the eta function (where
P∞ n a (κ) ∈ R and a0 = 1) and that,
indeed, we have the inverse formula η(s) = n=0 bn (κ)ηκν (s + 2νn) as well (where
the coefficients bn (κ) ∈ R are obtained from the an ’s recursively). By using these
results and the functional relationship of the eta function, η(s) = λ(s)η(1 − s), we
sketch a proof of the Riemann hypothesis which, in our setting, is equivalent to
the fact that the nontrivial zeros s∗ = σ ∗ + it∗ of η(s) (i.e. those points for which
η(s∗ ) = η(1 − s∗ ) = 0) are all located on the critical line σ ∗ = 12 .

1
1 Introduction
Let s := σ + it be a complex number. The Dirichlet eta function η(s), also called
alternating zeta function, is given in the half plane σ > 0 by the conditionally
convergent series [1]
∞
X (−1)m−1
η(s) = (1)
ms
m=1

which is absolutely convergent for σ > 1. Hardy gave a simple proof of the fact
that the eta function satisfies the functional equation [1]

η(s) = λ(s)η(1 − s) (2)

where
1 − 21−s s s−1  πs 
λ(s) = 2 π sin Γ(1 − s) (3)
1 − 2s 2
From this, one immediately obtains the means to extend the definition of the eta
function to the whole complex plane. Indeed, Euler’s acceleration of the condi-
tionally convergent series in Eq. (1) yields a double series that is absolutely and
uniformly convergent on compact sets everywhere [2, 3, 4]
∞ k  
X 1 X k (−1)m
η(s) = (4)
2k+1 m (m + 1)s
k=0 m=0

The Dirichlet eta function is closely related to the Riemann zeta function by

η(s) = 1 − 21−s ζ(s)

(5)

However, while the zeta function is meromorphic, with a pole at s = 1, the eta
function is an entire function. We note, from Eq. (1), that at s = 1, the eta
function becomes the alternating harmonic series and, therefore,
∞
X (−1)m−1 1 1
η(1) = = 1 − + − . . . = ln 2 (6)
m 2 3
m=1

Let s∗ denote a zero of the eta function, η(s∗ ) = 0. There are two kinds of
zeros: the trivial zeros for which, from the functional equation, we have λ(s∗ ) = 0;
and the nontrivial zeros, for which η(1 − s∗ ) = 0. From Eq. (3) the trivial zeros
are the negative even integers and zeros of the form σ ∗ = 1 + i 2nπ
ln 2 where n is a
nonzero integer (see [5] for a derivation that does not make use of the functional
relation).

2
 Since there are no zeros in the half-plane σ > 1, the functional equation implies
that nontrivial zeros of η are to be found in the critical strip 0 ≤ σ ≤ 1. By the
prime number theorem of Hadamard and de la Vallée Poussin [1] it is known that
for σ = 1 (and, therefore, σ = 0) there are no nontrivial zeros of the Riemann zeta
function and, therefore, from Eq. (5), there are nontrivial zeros of the Dirichlet
eta function neither. Thus, the nontrivial zeros are found in the strip 0 < σ < 1.
Furthermore, since, ∀s ∈ C
η(s) = η(s) (7)
where the line denotes complex conjugation, we have that s∗ and 1 − s∗ are also
nontrivial zeros of η. In brief, nontrivial zeros come in quartets, s∗ , 1 − s∗ , s∗ and
1 − s∗ forming the vertices of a rectangle within the critical strip. The statement
that, for the η function, the nontrivial zeros have all real part σ ∗ = 1/2, so that
s∗ = 1 − s∗ and s∗ = 1 − s∗ (the rectangle degenerating in a line segment) is
equivalent to the Riemann hypothesis for the Riemann zeta function [6].
In this article we investigate the position of the nontrivial zeros of the eta func-
tion with help of nonlinear embeddings, a novel kind of mathematical structures
introduced in our previous works [7, 8]. We construct here a nonlinear embedding
with the form of a double series that is absolutely and uniformly convergent on
compact sets in the whole complex plane. We call this embedding a holomorphic
nonlinear embedding. It depends on a scale parameter κ ∈ (0, ∞) and a horizontal
shift parameter ν ∈ (0, ∞), both in R, and converges asymptotically to η(s) every-
where as κ tends to infinity. With help of Möbius inversion, and taking advantage
of absolute convergence of the series concerned, we then show the crucial property
that η(s) itself can be expressed as a linear combination of horizontal shifts of
the holomorphic nonlinear embedding ηκµ (s) and we study the implications of this
linear combination on the position of the nontrivial zeros of the eta function giving
a proof of the Riemann hypothesis.
The outline of this article is as follows. In Section 2 we construct the holo-
morphic nonlinear embedding ηκν (s) for the Dirichlet eta function η(s). We prove
the global absolute and uniform convergence on compact sets of the series defin-
ing ηκν (s) and establish the asymptotic limits of the embedding. In Section 3 we
take advantage of these properties (specifically, we make heavy use of the absolute
convergence of this series) to derive the crucial properties: 1) the holomorphic non-
linear embedding can be expressed as a linear combination of horizontally shifted
eta functions; 2) the eta function itself can be expressed as a linear combination
of horizontally shifted holomorphic nonlinear embeddings. These results are then
exploited in Section 4 to derive a functional relationship for the embedding and to
prove the Riemann hypothesis, a result that emerges from the shift independence
in the limit κ → ∞ of the construction.

3
2 Holomorphic nonlinear embedding for the
Dirichlet eta function
We first introduce some notations and the basic functions on which our approach
is based.

Definition 2.1. Let x ∈ R. We define the Bκ -function [9] as

" ! !#
1 x + 12 x − 12
Bκ (x) := tanh − tanh (8)
2 κ κ

where κ ∈ (0, ∞) is a real parameter.

By noting that

e1/κ − e−1/κ
Bκ (x) = (9)
e1/κ + e2x/κ + e−2x/κ + e−1/κ
Bκ (x) e1/κ + 2 + e−1/κ
= (10)
Bκ (0) e1/κ + e2x/κ + e−2x/κ + e−1/κ
the following properties are easily verified:

0 ≤ Bκ (x) ≤ 1 ∀κ ∈ (0, ∞) (11)

Bκ (−x) = Bκ (x) (12)
lim Bκ (x) = 0 (13)
κ→∞
Bκ (x)
0≤ ≤ 1 ∀κ ∈ (0, ∞) (14)
Bκ (0)
Bκ (x)
lim = 1 (15)
κ→∞ Bκ (0)

Definition 2.2. (Holomorphic nonlinear embedding.) Let η(s) be the Dirich-

let eta function, given by Eq. (4). Then, we define the holomorphic nonlinear
embedding ηκν (s) of η(s) as the series
∞ k  
X 1 X k (−1)m Bκ (1/(m + 1)ν )
ηκν (s) := (16)
2k+1 m (m + 1)s Bκ (0)
k=0 m=0

with the real parameters κ ∈ (0, ∞) and ν ∈ (0, ∞).

Theorem 2.1. The double series in Eq. (16) converges absolutely and uniformly
on compact sets to the entire function ηκν (s).

4
Proof. We build on Sondow [2], who proved that the double series defining the
eta function in Eq. (4) converge absolutely and uniformly on compact sets in the
entire complex plane. In particular, if we define,
k  
1 X k (−1)m
fk (s) := k+1 (17)
2 m (m + 1)s
m=0
P∞
so that η(s) = k=0 fk (s), Sondow proved that there is a sequence of positive real
numbers {Mk } satisfying
k  
(−1)m 

1 X  k
|fk (s)| ≤ k+1  m (m + 1)s  ≤ Mk (18)
2 m=0

and which
∞
X
Mk < ∞ (19)
k=0
so that Weierstrass M-test is satisfied. We now note that we can write ηκν (s) as
∞
X
ηκν (s) = fk,κν (s) (20)
k=0

where
k  
1 X k (−1)m Bκ (1/(m + 1)ν )
fk,κν (s) := (21)
2k+1 m (m + 1)s Bκ (0)
m=0
Now, by the triangle inequality,
k  
 
 1 X k (−1)m Bκ (1/(m + 1)ν ) 
|fk,κν (s)| =  k+1

2 m (m + 1)s Bκ (0)


m=0
k  
(−1)m Bκ (1/(m + 1)ν ) 

1 X  k
≤ k+1  m (m + 1)s
2 Bκ (0) 
m=0
k  
(−1)m   Bκ (1/(m + 1)ν ) 
 
1 X  k
= k+1  m (m + 1)s  
2 m=0
Bκ (0) 
k  
(−1)m 

1 X  k
≤ k+1  m (m + 1)s  (22)
2 m=0

where Eq. (14) has been used. Therefore, from this last expression and Eq. (18)
k  
(−1)m 

1 X  k
|fk,µν (s)| ≤ k+1  m (m + 1)s  ≤ Mk (23)
2
m=0

5
and, thus, the sequence of positive real numbers {Mk } found by Sondow majorizes
the sequence {|fk,µν (s)|} as well, and the result follows.
Theorem 2.2. We have, ∀s ∈ C

lim ηκν (s) = η(s) ∀ν ∈ (0, ∞) (24)

κ→∞
lim ηκν (s) = η(s) − 1 ∀ν ∈ (0, ∞) (25)
κ→0
Bκ (1)
lim ηκν (s) = η(s) + −1 ∀κ ∈ (0, ∞) (26)
ν→∞ Bκ (0)
Bκ (1)
lim ηκν (s) = η(s) ∀κ ∈ (0, ∞) (27)
ν→0 Bκ (0)
3 1 1
where Bκ (1)/Bκ (0) = (tanh 2κ − tanh 2κ )/(2 tanh 2κ ), as given by Eq. (8).
Proof. We first observe that, from Eq. (10),
Bκ (1/(m + 1)ν )
lim =1
κ→∞ Bκ (0)
Bκ (1/(m + 1)ν )

1 if m ≥ 1
lim =
κ→0 Bκ (0) 0 if m = 0
(
Bκ (1/(m + 1)ν ) 1 if m ≥ 1
lim = Bκ (1)
Bκ (0) Bκ (0) if m = 0
ν→∞

Bκ (1/(m + 1)ν ) Bκ (1)

lim =
ν→0 Bκ (0) Bκ (0)
By using Eq. (16) and the above expressions, we get
k  
"∞ #
X 1 X k (−1)m Bκ (1/(m + 1)ν )
lim ηκν (s) = lim
κ→∞ κ→0 2k+1 m (m + 1)s Bκ (0)
k=0 m=0
∞ k  
X 1 X k (−1)m
= = η(s)
2k+1 m=0 m (m + 1)s
k=0
∞ k  
X 1 X k (−1)m
lim ηκν (s) = = η(s) − 1
κ→0 2k+1 m (m + 1)s
k=0 m=1
∞ k  
Bκ (1) X 1 X k (−1)m Bκ (1)
lim ηκν (s) = + k+1 s
= η(s) + −1
ν→∞ Bκ (0) 2 m (m + 1) Bκ (0)
k=0 m=1
∞ k  
Bκ (1) X 1 X k (−1)m Bκ (1)
lim ηκν (s) = = η(s).
ν→0 Bκ (0) 2k+1 m=0 m (m + 1)s Bκ (0)
k=0

6
3 Functional expansions of ηκν (s) and η(s)
We now derive an equivalent expression for ηκν (s) valid for any κ sufficiently large
(specifically, ∀κ > 3/π) and prove that this expression can be inverted to express
η(s) as a function of horizontal shifts of ηκν (s).
Theorem 3.1. If κ > 3/π, ∀ν ∈ (0, ∞) the holomorphic embedding ηκν (s) has the
absolutely convergent series expansion
∞
X
ηκν (s) = η(s) + an (κ)η(s + 2νn) (28)
n=1
where, for n a non-negative integer
∞
1 X 22n (22j − 1)B2j
an (κ) = (29)
tanh (1/2κ) j(2j − 2n − 1)!(2n)!κ2j−1
j=n+1

and B2m denote the even Bernoulli numbers: B0 = 1, B2 = 16 , B4 = − 30

1
, etc.
Proof. For κ > π3 , the hyperbolic tangents in the definition of the B-function,
Eq. (8) with x = 1/(m + 1)ν , can be expanded in their absolutely convergent
MacLaurin series for all ∀m ≥ 0 (note that ∀ν > 0, 1/(m + 1)ν ≤ 1)
 
1
Bκ =
(m + 1)ν
∞
" #
1 X 22j (22j − 1)B2j 1 2j−1 1 2j−1
  
1 1
= + − −
2 (2j)!κ2j−1 (m + 1)ν 2 (m + 1)ν 2
j=1
∞ 2j−1 
1 X 22j (22j − 1)B2j X 2j − 1

1
= 2j−1
(1 − (−1)h )
2
j=1
(2j)!κ h
h=0
2 (m + 1)ν(2j−1−h)
h

∞ j 
2(22j − 1)B2j 22(j−h)

X X 2j − 1
=
(2j)!κ2j−1 2h − 1 (m + 1)2(j−h)ν
j=1 h=1
∞ j−1 
2(22j − 1)B2j X 22n

X 2j − 1
=
(2j)!κ2j−1 2j − 2n − 1 (m + 1)2nν
j=1 n=0
∞ ∞ 
2j − 1 22n+1 (22j − 1)B2j

X 1 X
= (30)
(m + 1)2nν 2n (2j)!κ2j−1
n=0 j=n+1

where we have used the absolute convergence of the series to change the order of
the sums. We also have
∞
1 X 2(22j − 1)B2j
Bκ (0) = tanh = (31)
2κ (2j)!κ2j−1
j=1

7
 If we then replace these expansions in the definition of the embedding, Eq.
(16), we find, by exploiting the absolute convergence of the series
∞ X
∞ k   ∞ 
(−1)m 2j − 1 22n+1 (22j − 1)B2j

X 1 X k X
ηκν (s) =
2k+1 m (m + 1)s+2nν 2n (2j)!κ2j−1 Bκ (0)
n=0 k=0 m=0 j=n+1
∞ ∞ 
2j − 1 22n+1 (22j − 1)B2j
X X 
= η(s + 2νn)
2n (2j)!κ2j−1 Bκ (0)
n=0 j=n+1
X∞ ∞
X
= an (κ)η(s + 2νn) = η(s) + an (κ)η(s + 2νn) (32)
n=0 n=1

where, for all non-negative integer n, we have defined

∞
2(22j − 1)B2j 2n 2j − 1
X  
an (κ) := 2 (33)
(2j)!κ2j−1 Bκ (0) 2n
j=n+1
∞
1 X 22n (22j − 1)B2j
=
tanh (1/2κ) j(2j − 2n − 1)!(2n)!κ2j−1
j=n+1

and we have also used that, from Eq. (31)

∞
2(22j − 1)B2j
 
X 2j − 1
a0 = = 1.
(2j)!κ2j−1 Bκ (0) 2j − 1
j=1

Corollary 1. Asymptotically, for κ large, we have,

 
η(s + 2ν) 1
ηκν (s) = η(s) − +O (34)
κ2 κ4

Proof. From the theorem, we have ∀n ∈ Z+ ∪ {0}

∞
1 X 22n (22j − 1)B2j
an (κ) =
tanh (1/2κ) j(2j − 2n − 1)!(2n)!κ2j−1
j=n+1
∞
1 X 22n (22j − 1)B2j
= 1
− O κ13 j=n+1 j(2j − 2n − 1)!(2n)!κ2j−1


( 2κ
∞
22n+1 (22j − 1)B2j
   X
1
= 1+O
κ2 j(2j − 2n − 1)!(2n)!κ2j−2
j=n+1
22n+1 (22n+2 − 1)B2n+2
   
1 1
= + O = O (35)
(n + 1)(2n)!κ2n κ2n+2 κ2n

8
Therefore,  
η(s + 2ν) 1
ηκν (s) = η(s) + 30B4 +O (36)
κ2 κ4
and the result follows from noting that B4 = −1/30.

Theorem 3.2. (Möbius inversion formula.) For any κ > 3/π and ηκν (s)
given by Eq.(28),
∞
X
ηκν (s) = η(s) + an (κ)η(s + 2νn) (37)
n=1
we have,
∞
X
η(s) = ηκν (s) + bn (κ)ηκν (s + 2νn) (38)
n=1
where the coefficients bn are recursively obtained from
k
X
bk (κ)an−k (κ) = δk0 (39)
n=0

with δk0 being the Kronecker delta (δk0 = 1 if k = 0 and δk0 = 0 otherwise).

Proof. We have
∞
X ∞
X ∞
X
bn ηκν (s + 2νn) = bn am (κ)η(s + 2νn + 2νm)
n=0 n=0 m=0
∞ X
" k #
X
= bn (κ)ak−n (κ) η(s + 2νk)
k=0 n=0
∞
X
= δk0 η(s + 2νk)
k=0
= η(s)

The coefficients bn can be recursively obtained from the known an by solving the
equations

a0 b0 = 1 (40)
a0 b1 + a1 b0 = 0 (41)
a0 b2 + a1 b1 + a2 b0 = 0 (42)
...

Where, since a0 = 1, b0 = 1, and, therefore, b1 = −a1 , b2 = a21 − a2 , etc.

9
Remark 1. Eq. (38), with ηκν given by Eq. (16) is the main result of this work,
since it expresses the Dirichlet eta function in terms of a globally convergent series
(absolutely and uniformly on compact sets) of horizontal shifts of ηκν . These shifts
are weighted by powers of 1/κ.
Remark 2. Theorem 3.2 is, indeed, a specific case of Theorem 3.3 on p. 82 in [11]
particularized to the functions η and ηκν considered here.
Proposition 3.1. For κ > 3/π we have
∞ 3 1
X tanh 2κ − tanh 2κ
an (κ) = 1 −1 (43)
n=1
2 tanh 2κ
∞ 1
X 2 tanh 2κ
bn (κ) = 3 1 −1 (44)
n=1
tanh 2κ − tanh 2κ

Proof. From Eq. (26)

Bκ (1)
lim ηκν (s) = η(s) + −1 (45)
ν→∞ Bκ (0)
and, for κ > 3/π, from Eqs. (37) and (38)
∞ ∞
X Bκ (1) X
lim ηκν (s) = η(s) + an (κ) = η(s) − bn (κ) (46)
ν→∞ Bκ (0)
n=1 n=1

because, from Eqs. (4) and (16), limν→∞ η(s+2νn) = 1 and limν→∞ ηκν (s+2νn) =
Bκ (1) /Bκ (0) for every integer n ≥ 1. By equating Eqs. (45) and (46) and by using
Eq. (8), the result follows.

4 On the nontrivial zeros of η

In this section we prove the Riemann hypothesis. We first introduce three lem-
mas, that establish a functional equation for the embedding and its asymptotic
properties.
Lemma 4.1. (Functional relationship for the embedding.) We have ∀ν ∈
(0, ∞) and ∀κ > π/3
∞
X
ηκν (s) − λ(s)ηκν (1 − s) = bn (κ) [λ(s)ηκν (1 − s + 2νn) − ηκν (s + 2νn)] (47)
n=1

where
1 − 21−s s s−1  πs 
λ(s) = 2 π sin Γ(1 − s) (48)
1 − 2s 2

10
Proof. From Eq. (38) we have
∞
X
η(s) = ηκν (s) + bn (κ)ηκν (s + 2νn)
n=1
∞
X
λ(s)η(1 − s) = λ(s)ηκν (1 − s) + λ(s) bn (κ)ηκν (1 − s + 2νn)
n=1

whence, by subtracting both equations and applying Eq. (2) the result follows.
Lemma 4.2. If s∗ = σ ∗ + it∗ is a non-trivial zero of η(s) then ηκν (s∗ ) 6= 0 and
ηκν (1 − s∗ ) 6= 0 for finite asymptotically large κ and ν > 1/2. Furthermore, we
have
ηκν (s∗ ) η(s∗ + 2ν)
lim = (49)
κ→∞ ηκν (1 − s∗ ) η(1 − s∗ + 2ν)
Proof. The real part σ ∗ of the nontrivial zero s∗ satisfies 0 < σ ∗ < 1. Now, since
η(s∗ ) = η(1 − s∗ ) = 0, we have, from Eq. (34)
η(s∗ + 2ν)
 
1
ηκν (s∗ ) = − + O (50)
κ2 κ4
η(1 − s∗ + 2ν)
 
∗ 1
ηκν (1 − s ) = − 2
+O (51)
κ κ4
We have that ∀ν > 1/2, η(s∗ + 2ν) 6= 0 and η(1 − s∗ + 2ν) 6= 0 because values
s∗ + 2ν and 1 − s∗ + 2ν both lie in the half-plane σ > 1 and η(s) has no zeros there.
Therefore ηκν (s∗ ) 6= 0 and ηκν (1 − s∗ ) 6= 0 are both nonzero for sufficiently large
κ (κ > 3/π being a lower bound). Eq. (49) follows then as a trivial consequence
of Eqs. (50) and (51) and the absolute convergence of all the series involved.
Lemma 4.3. Let sγ ∈ C, sγ ∈ γ be such that ηκν (1 − sγ ) 6= 0 along a path γ in
the complex plane and let s′ be an endpoint of γ. Then,
ηκν (sγ ) ηκν (sγ )
lim lim ′ = lim ′ lim = λ(s′ ) (52)
κ→∞ sγ −
→ s ηκν (1 − sγ ) sγ −
→ s κ→∞ ηκν (1 − sγ )
γ γ

Furthermore, if s′ = s∗ is a nontrivial zero of the Dirichlet eta function, ∀ν > 1/2

η(s∗ + 2ν)
= λ(s∗ ) (53)
η(1 − s∗ + 2ν)
Proof. From Eq. (54) we have, by dividing by ηκν (1 − sγ ) at any point sγ of γ
(since ηκν (1 − sγ ) 6= 0)
∞
ηκν (sγ ) X λ(sγ )ηκν (1 − sγ + 2νn) − ηκν (sγ + 2νn)
= λ(sγ ) + bn (κ) (54)
ηκν (1 − sγ ) ηκν (1 − sγ )
n=1

11
where b1 (κ) = O(κ−2 ). We now have, on one hand,
ηκν (sγ ) ηκν (s′ )
lim lim ′ = lim (55)
κ→∞ sγ −
→ s ηκν (1 − sγ ) κ→∞ ηκν (1 − s′ )
γ

∞
" #
′
X λ(s′ )ηκν (1 − s′ + 2νn) − ηκν (s′ + 2νn)
= lim λ(s ) + bn (κ)
κ→∞
n=1
ηκν (1 − s′ )
∞
" #
X λ(s ′ )η(1 − s′ + 2νn) − η(s′ + 2νn)
= lim λ(s′ ) + bn (κ)
κ→∞ η(1 − s′ )
n=1
= λ(s′ )
and, on the other hand,
ηκν (sγ ) η(sγ )
lim ′ lim = lim ′ (56)
sγ −
→ s κ→∞ ηκν (1 − sγ ) sγ →s η(1 − sγ )
γ

= lim λ(sγ ) = λ(s′ )

sγ →s′

whence the result follows.

Let us now assume that s′ = s∗ is a nontrivial zero of the Dirichlet eta function.
Then, we have that limκ→∞ ηκν (1 − s∗ ) = η(1 − s∗ ) = 0 = η(s∗ ) = limκ→∞ ηκν (s∗ )
and the function in Eq. (55)
ηκν (s′ ) η(s′ )
Φ(s′ ) := lim = (57)
κ→∞ ηκν (1 − s′ ) η(1 − s′ )
is undefined at s′ = s∗ . However, s′ = s∗ is a removable singularity and we can
take Φ(s∗ ) = λ(s∗ ). To see this, note that Φ(s) = λ(s) for all s in the critical strip
0 < σ < 1 save at the nontrivial zeros s∗ . But the function λ(s) is holomorphic
for all s in the critical strip including the nontrivial zeros s∗ of η. Therefore, by
Riemann’s theorem on extendable singularities, Φ(s) is holomorphically extendable
over s∗ and we can have, in consistency with Eq. (56)
 
∗ ∗ η(s)
Φ(s ) = λ(s ) = lim∗ (58)
s→s η(1 − s)

This proves Eq. (52). We then note that, on one hand

ηκν (s∗ )
Φ(s∗ ) = lim = λ(s∗ ) (59)
κ→∞ ηκν (1 − s∗ )

and, on the other, from Eq. (49)

ηκν (s∗ ) η(s∗ + 2ν)
Φ(s∗ ) = lim = (60)
κ→∞ ηκν (1 − s∗ ) η(1 − s∗ + 2ν)

12
Both expressions must be equal at s∗ because: 1) Eq. (59) is a consequence of
Φ(s) being equal to the holomorphic λ(s) in the punctured critical strip (save,
exactly at the zeros s∗ ) and, therefore, holomorphically extendable over s∗ and 2)
Eq. (60) is a consequence of the asymptotic behavior of the embedding close to a
nontrivial zero of the Dirichlet eta function. Thus, Eq. (53) follows.

Alternative proof of Eq. (53). An equivalent way of obtaining Eq. (53) is, directly,
from Eq. (55), applying it to s′ = s∗ . We have,

ηκν (sγ ) ηκν (s∗ )

lim lim ∗ = lim (61)
κ→∞ sγ −
→ s ηκν (1 − sγ ) κ→∞ ηκν (1 − s∗ )
γ

∞
" #
∗
X λ(s∗ )ηκν (1 − s∗ + 2νn) − ηκν (s∗ + 2νn)
= lim λ(s ) + bn (κ)
κ→∞ ηκν (1 − s∗ )
n=1
λ(s∗ )ηκν (1 − s∗ + 2ν) − ηκν (s∗ + 2ν)
= λ(s∗ ) + lim
κ→∞ κ2 ηκν (1 − s∗ )
λ(s )η(1 − s + 2ν) − η(s∗ + 2ν)
∗ ∗
= λ(s∗ ) − (62)
η(1 − s∗ + 2ν)

and since η(1 − s∗ + 2ν) 6= 0 ∀ν ∈ (1/2, ∞) and we have for any ε ∈ C in a disk of
sufficiently small radius
ηκν (s∗ + ε)
Φ(s∗ + ε) = lim = λ(s∗ + ε) (63)
κ→∞ ηκν (1 − s∗ + ε)

by taking the limit ε → 0 and observing that Φ(s) is holomorphically extendable

to s∗ where it has then the value λ(s∗ ) we obtain, from Eq. (62)

λ(s∗ )η(1 − s∗ + 2ν) − η(s∗ + 2ν) = 0 (64)

which is Eq. (53). 

Theorem 4.1. (Riemann hypothesis.) All nontrivial zeros s∗ = σ ∗ + it∗ of the

Dirichlet eta function η(s) have real part σ ∗ = 1/2.

Proof. Let σ ∗ = 12 + ǫ be the real part of a nontrivial zero s∗ = σ ∗ + it∗ of η in the

critical strip. From Eq. (53),
 
 η 1 + 2ν + it∗ + ǫ
   1 
 2
 = λ ∗

+ it + ǫ  , (65)
 
 1 ∗
 η 2 + 2ν − it − ǫ  2

and we note that the right hand side of this expression is shift-invariant (it does not
depend on ν) but the left hand side is not: the horizontal shift parameter ν can be

13
arbitrarily varied in the interval (0, ∞) and, in particular, it can be selected so that
the point s∗ +2ν lies anywhere on the half-plane σ ≥ 1 at height t∗ . The modulus of
η varies on horizontal lines [10]. The only possibility for equation Eq. (65) to have
1
solution for a nontrivial zero s∗ forces ǫ = 0. To see
 this, put x = 2 + 2ν − ǫ >> 1
in Eq. (65). Now, since |η (x − it∗ )| = η (x + it∗ ) = |η (x + it∗ )|, we have
 



1
η + 2ν + it∗ + ǫ   η (x + it∗ + 2ǫ) 
 
2
 =  . (66)
 
1 η (x + it) 
η + 2ν − it∗ − ǫ 

2

Since x can be increased arbitrarily by increasing ν, we can take x so large that,

asymptotically

∗ ∗ ∂η 
η (x + it + 2ǫ) ∼ η (x + it ) + 2ǫ . (67)
∂x x+it∗

In this asymptotic regime, we can truncate Eq. (1) to the first two terms and its
derivative becomes 
∂η  ln 2
∼ x+it∗ . (68)
∂x x+it∗
 2
∗
Furthermore, η (x + it∗ ) ∼ 1 − 2−x−it and thus
 η (x + it∗ + 2ǫ)  
   
 ∼ 1 + 2ǫ ln∗2  .

 (69)
 η (x + it)   2x+it − 1 

Therefore, for ν large,

 
 η 1 + 2ν + it∗ + ǫ
  ln 2

 2
 ∼ 1 + 2ǫ 1

, (70)

 1
 η 2 + 2ν − it∗ − ǫ  2 2 +2ν−ǫ+it − 1
∗ 

and the l.h.s. of Eq. (65) depends explicitly on the free parameter ν. Thus, the
r.h.s. of Eq. (65) can have an infinite number of different values for its modulus,
which is absurd. The only possibility of cancelling the ν dependence, forced by
the consistency of the equation, is to have ǫ = 0. In this way, we obtain,
 
 η 1 + 2ν + it∗
  
1

 2 

 = 1 = λ ∗

+ it  , (71)
 
 1 ∗
 η 2 + 2ν − it  2

an equation that is known to have infinitely many solutions for t∗ . Therefore,

ǫ = 0, s∗ = 12 + it∗ and the result follows.

14
5 Conclusions
In this article a complex entire function called a holomorphic nonlinear embedding
ηκν has been constructed and some of its properties (mainly asymptotic ones) have
been investigated. It has been shown that the function ηκν can be expressed as
a series expansion in terms of horizontal shifts of the Dirichlet eta function. The
holomorphic character of ηκν has been established by proving the global absolute
and uniform convergence on compact sets of its defining double series. The co-
efficients of the expansion and their sum have been explicitly calculated. It has
also been shown that this expansion can be inverted to yield the eta function as a
linear expansion of horizontal shifts of ηκν . This is the central result of this article,
since it allows the Dirichlet eta function to be understood as a linear superposition
of different layers governed by shifts of ηκν and weighted by powers of 1/κ. This
result shows that, although one can envisage, in principle, infinitely many ways
of smoothly embedding the Dirichlet eta function in a more general structure, the
one presented here (ηκν ) is not gratuitous because it is itself embedded within the
structure of the eta function, revealing some of its secrets thanks to its scale and
horizontal shift parameters κ and ν. In particular, the truth of the Riemann hy-
pothesis emerges naturally as a consequence of the functional relationship of the
Dirichlet eta function and its uniform attainment everywhere by a hierarchy of
functional equations of the holomorphic nonlinear embedding in the limit κ → ∞.
Operators yielding vertical shifts of the Riemann zeta function ζ(s) arise nat-
urally in approaches to the Riemann zeros using ideas from supersymmetry [12].
These vertical shifts can, indeed, be expressed more compactly in terms of the
Dirichlet eta function η (see e.g. Eq.(18) in [12]). Whether there exists any rela-
tionship of these vertical shifts induced by lowering and raising operators in [12]
with the shifts obtained here as a result of an explicit series construction is an in-
teresting open question. We point out that ν in this article can be made a complex
number with positive real part and arbitrary imaginary part and all main results
of this article apply without any modification: Eqs. (37) and (38) are indeed valid,
for ν = νr + iνi , with νr > 0 and any νi ∈ (−∞, ∞) and this does not affect the
proof of the Riemann hypothesis here presented in any way. The condition νr > 0
is, however, necessary, for Eqs. (37) and (38) to be valid.
The methods used here may be adapted to other Dirichlet series for which
the Riemann hypothesis is conjectured to hold [13]. All steps may be retraced
for these latter series if: 1) they can be closely related to entire functions with
series expansions that are absolutely convergent in the whole complex plane; 2)
the latter series depend on the complex variable s only through summands of the
form 1/ms ; 3) there exists a functional relation like Eq. (2).
We believe that the methods described in this paper might be useful to get

15
insight in recent intriguing experimental phenomena connecting the coefficients of
truncated Dirichlet series to the Erathostenes sieve [14].

References
[1] E. C. Titchmarsh and D. R. Heath-Brown, The theory of the Riemann zeta-
function, Clarendon Press, Oxford (1986).

[2] J. Sondow, “Analytic continuation of Riemann’s zeta function and values at

negative integers via Euler’s transformation of series”. Proc. Amer. Math.
Soc. 120 (1994) 421-424. doi:10.1090/S0002-9939-1994-1172954-7.

[3] H. Hasse, “Ein Summierungsverfahren für die Riemannsche ζ-Reihe”. Math-

ematische Zeitschrift. 32 (1930) 458-464. doi:10.1007/BF01194645

[4] I. V. Blagouchine, “Three Notes on Ser’s and Hasse’s Representations for

the Zeta-functions”. Integers (Electronic Journal of Combinatorial Number
Theory). 18A (2018) 1-45. arXiv:1606.02044 [math.NT]

[5] J. Sondow, “Zeros of the alternating zeta function on the line R(s) = 1”.
Amer. Math. Monthly 110 (2003) 435-437. arXiv:math/0209393v2 [math.NT].

[6] K. Broughan, Equivalents of the Riemann hypothesis. Volume One: Arith-

metic Equivalents, Cambridge University Press, Cambridge UK (2017). p.
252.

[7] V. Garcı́a-Morales, “Unifying vectors and matrices of different dimensions

through nonlinear embeddings”, J. Phys. Complex. 1 (2020) 025008. doi:
10.1088/2632-072X/ab7d87

[8] V. Garcı́a-Morales, “Nonlinear embeddings: Applications to analysis, frac-

tals and polynomial root finding”, Chaos Sol. Fract. 99 (2017) 312-324.
doi:10.1016/j.chaos.2017.04.021

[9] V. Garcı́a-Morales, “From deterministic cellular automata to coupled map

lattices”, J. Phys. A: Math. Theor. 49 (2016) 295101. doi:10.1088/1751-
8113/49/29/295101

[10] Y. Matiyasevich, F. Saidak and P. Zvengrowski, “Horizontal Monotonic-

ity of the Modulus of the Riemann Zeta Function and Related Functions”,
arXiv:1205.2773 [math.NT]

[11] C. Nanxian, Möbius inversion in Physics, World Scientific, Singapore (2010).

16
[12] A. Das and P. Kalauni, “Supersymmetry and the Riemann zeros on the critical
line”, Phys. Lett. B 791 (2019) 265-269. doi:10.1016/j.physletb.2019.02.040

[13] A. A. Karatsuba and S. M. Voronin, The Riemann Zeta-Function, Walter de

Gruyter, Berlin (1992).

[14] G. Beliakov and Y. Matiyasevich, “Approximation of Riemann’s

Zeta Function by Finite Dirichlet Series: A Multiprecision Nu-
merical Approach”, Experimental Mathematics 24 (2015) 150-161.
doi:10.1080/10586458.2014.976801, arXiv:1402.5295 [math.NT]

17