04 Absolute Value of Dirichlet Eta Function

4.1 Dirichlet Eta Function

4.1.1 Definition
Dirichlet Eta Function (z ) is defined in the half plane Re(z) > 0 as follows.

 (-1)r-1
(z) = Σ (1.0)
r=1 rz
This series is analytically continued to the whole complex plane by applying some kind of acceleration method.
The easiest of these is the Euler transformation as follows.
r-1

 r
 1 k k (-1)
(z) = Σ k+1 Σ
(1.1)
k=1 2 r=1 rz
(1.0) and (1.1) are the same in Re(z) > 0 . Although (1.0) can not express the left side of line the of

convergence, (1.1) can express also the left side of this. Therefore, we can define Dirichlet Eta Function (z)
by (1.1) .

4.1.2 Overview
The 3D figures of the real part and the imaginary part of Dirichlet Eta Function (x +i y ) are as follows.

Further, the 3D figure of the absolute value is as follows. In the left figure, trivial zeros of (z) are observed
along the x axis. The right figure is a view of the left figure from the bottom. We can see that zeros of (z)
are located in two lines along x =1/2 and x =1 .

-1-
As seen in these figures, Dirichlet Eta Function (z ) has three kinds of zeros as follows.
(1) Trivial zeros -2, -4, -6, -8, 
(2) Non-trivial zeros 1/2 i 14.1347 , 1/2 i 21.0220 , 1/2 i 25.0108 , 
(3) (z)specific zeros 1 2 i /log 2 , 1 4 i /log 2 , 1 6 i /log 2 , 

Among these, (1) and (2) are common to the zeros of Riemann zeta function. Further, it is well known that
0 < x < 1 called Critical Strip . Moreover, it is proved that they have to exist
non-trivial zeros (2) exist in
symmetrically with respect to x =1/2 . And, fortunately, this critical strip is included within the convergence
range of the series (1.0) .

4.1.3  specific zeros

The figures of the real part and the imaginary part of Dirichlet Eta Function (z) at x =1 are as follows.

The purple points are  specific zeros. We can see the followings from the figure.
(1) The real part resembles a negative cosine curve and the imaginary part resembles a sine curve.
(2) The extrema points of the real part are close to the zeros of the imaginary part but not zeros.
(3)  specific zeros are close to the local minimum points of the real part, but are not so.
(4)  specific zeros are the uphill zeros (end of cycles) of the imaginary part. However, the reverse is not true.

4.1.4 Non-trivial zeros

The figures of the real part and the imaginary part of Dirichlet Eta Function (z) at x =1/2 are as follows.

-2-
 The red points are non-trivial zeros. We can see the followings from the figure.
(1) The real part resembles a negative cosine curve and the imaginary part resembles a sine curve.
(2) The extrema points of the real part are close to the zeros of the imaginary part but not zeros.
(3) Non-trivial zeros are close to the local minimum points of the real part, but are not so.
(4) Non-trivial zeros are the uphill zeros (end of cycles) of the imaginary part. However, the reverse is not true.
(5) The purple points which were  specific zeros, have moved below the y -axis.

Downhill zeros
Exceptions exist in (3) and (4). For example, the figure in the vicinity of y =72 is as follows. Non-trivial zero
point 72.0671 (red point) is near the local maximum point of the real part and is the downhill zero point
of the imaginary part.

4.1.5 Feature of Dirichlet Eta Series

Observing both 4.1.3 and 4.1.4 , we can see that this Dirichlet series contains two periodic functions. Among
these, the periodic function giving  specific zeros is constant in period and amplitude, but the periodic function
giving non-trivial zeros is not constant in period and amplitude.

-3-
4.2 Squared Absolute Value of Dirichlet Eta
Squared absolute value of Dirichlet eta function is

g(x, y) = (x, y)

2
(2.0)

This is a real-valued function with two variables. And it is shown in the figure as follows.

In the left figure, dents are observed alongx =1/2 and x =1 . The right figure is a view of the left figure from
the bottom. We can see that zeros of (z) are located in two lines along x =1/2 and x =1 . The zeros on
the x =1/2 correspond to the zeros of  (z) function and the zeros on the x =1 are  specific zeros.
On the other hand, there is no zero on the x =0 .

Features in 0  x  1/2
Let us focus on space 0  x  1/2 . The figures of section in x =0 , 1/4 , 1/2 are drawn as follows.

Looking at this, it looks like (0, y ) > (1/4 , y) > (1/2 , y)
2
2.6222  y  25 .
2 2
in
It is the same also in 1200  y  1225 . Below, we observe this in more detail.

(1) 0  y  2.5841
The front view of 3D in this interval is the left figure. The cutaway view at y = 0, 1.7227, 2.5841 of this
is the right figure. In this interval, (x,y )
2
seems to be monotonically increasing with respect to x.

-4-
(2) 2.5841 < y < 2.6222
The front view of 3D in this interval is the left figure. The cutaway view at y = 2.5842, 2.6055, 2.6221
2
of this is the right figure. In this interval, (x , y ) is not monotonic with respect to x . In the right figure,
although the curve of y = 2.5842 looks like monotonically increasing, it is decreasing at the left end when
it is seen enlarged. Although the curve of y = 2.6221 looks like monotonically decreasing, it is increasing
at the right end when it is seen enlarged.

(3) y  2.6222
The front view of 3D in this interval is the left figure. The cutaway view at y = 2.6222 , 8 , 14 of this is
the right figure. In this interval, (x,y )
2
seems to be monotonically decreasing with respect to x.

Based on the observations above, I present the next hypothesis equivalent to the Riemannian hypothesis.

Hypothesis 4.2.1
When (x , y) is the Dirichlet eta function on the complex plane, the squared absolute value (x,y )
2

is a monotonically decreasing function in the region 0 < x < 1/2 , y  3 .

Note1
The zeros common to the Riemann zeta function exist in 0< x <1 called critical strip . Moreover, it is

-5-
 x =1/2 . So, if (x,y) is monotonically
2
proved that they have to exist symmetrically with respect to
decreasing with respect to x in the region 0 < x < 1/2 , y  3 , zeros do not exist in the region and
the opposite region 1/2 < x < 1 , y  3 . This is equivalent to the Riemann hypothesis.

Note2
1/2 < x < 1 , y  3 , (x,y)
2
Incidentally, in the opposite region is not necessarily a monotone

function with respect to x . For example, (x,72.21)2 is as follows.

-6-
4.3 Expression of Squared Absolute Value by Series

4.3.1 Expression of Dirichlet Eta Function by Series

As seen in 4.1.1 , Dirichlet Eta Function (z) was defined as follows.
 (-1)r-1
(z) = Σ Re(z) > 0
r=1 rz
When Re(z) > 0 , let z = x + i y . Then,

(x, y) = Σ(-1)r-1 r -x- i y x >0
r=1
If this is represented by an exponential function,
 
(x, y) = Σ(-1)r-1 e -(x+ i y)log r = Σ(-1)r-1 e -xlog r-i ylog r
r=1 r=1
If this is represented by a trigonometric function,
 cos(y log r)  sin(y log r)
(x , y) = Σ(-1)r-1 x
- iΣ(-1)r-1 (3.1)
r=1 r r=1 rx

4.3.2 Expression of   by Double Series

2

Squared absolute value of Dirichlet eta function (x,y )

2
is expressed using (3.1) as follows.
2 2
 

Σ(-1)  + Σ(-1) 
cos(y log r) sin(y log r)
(x , y )
2 r-1 r-1
=
r=1 rx r=1 rx
Although it looks like a very complicated, it becomes an unexpectedly simple expression when it is expanded
and organized.

Formula 4.3.2
When (x , y) is the Dirichlet Eta Function,
  (-1)r+s
 
s
(x , y ) = ΣΣ
2
x
cos y log  := g (x , y)  (3.2)
r=1 s=1 (rs) r

Proof
Let (-1)r-1cos(y log r) = Cr . Then,
 (-1)r-1cos(y log r) C1 C2 C3 C4
Σ
r=1 rx
=
1x
+
2x
+
3x
+
4x
+

This square is
C1 C2 C3 C4 C5 C6
+ + + + + +
1x 2x 3x 4x 5x 6x
C1 C2 C3 C4 C5 C6
 + x + x + x + x + x +
1x 2 3 4 5 6
C1 C1 C1 C2 C1 C3 C1 C4
= x x
+ x x + x x + x x +
1 1 1 2 1 3 1 4
C2 C1 C2 C2 C2 C3 C2 C4
+ x x + x x + x x + x x +
2 1 2 2 2 3 2 4

-7-
 C3 C1 C3 C2 C3 C3 C3 C4
+ x x
+ x x
+ x x
+ +
3 1 3 2 3 3 3x 4x
C4 C1 C4 C2 C4 C3 C4 C4
+ + + + +
4x 1x 4x 2x 4x 3x 4x 4x

  Cr Cs
= ΣΣ
r=1 s=1 rxsx
i.e.
2 2
(-1)r-1cos(y log r)
Σ  Σ r 
  Cr   Cr Cs
= = ΣΣ
r=1 rx r=1
x
r=1 s=1 (rs)x
Let (-1)r-1sin(y log r) = S r . Then, in a similar way, we obtain
2 2
(-1)r-1sin(y log r)
  Σ
r=1 r 
  Sr   Sr Ss
Σ
r=1 rx
= x
= ΣΣ
r=1 s=1 (rs)x
Then,
2 2
(-1)r-1cos(y log r) (-1)r-1sin(y log r)
Σ  + -Σ 
 
(x , y )
2
=
r=1 rx r=1 rx
  Cr Cs + S r Ss
= ΣΣ
r=1 s=1 (rs)x
Returning to the original symbol ,
  1
(x , y ) = ΣΣ
2
x
(-1)r-1cos(y log r) (-1)s-1cos(y log s)
r=1 s=1 (rs)
+ (-1)r-1sin(y log r) (-1)s-1sin(y log s)
i.e.
  cos(y log r) cos(y log s) + sin(y log r) sin(y log s)
(x , y ) = ΣΣ(-1)r+s
2
r=1 s=1 (rs)x
Here,

 
s
cos(y log r) cos(y log s) + sin(y log r) sin(y log s) = cos y log
r
Using this,
  (-1)r+s
 
s
(x , y ) = ΣΣ
2
cos y log (3.2)
x r
r=1 s=1 (rs)

The first few lines are as follows.

     
2 1 1 1 2 1 3
x,y = x
- cos y log
x
cos y log + x
cos y log -+
(11) 1 (12) 1 (13) 1

     
1 1 1 2 1 3
- x
cos y log + x
cos y log - x
cos y log +-
(21) 2 (22) 2 (23) 2

     
1 1 1 2 1 3
+ x
cos y log - x
cos y log + x
cos y log -+
(31) 3 (32) 3 (33) 3


-8-
 When both sides of (3.2) are illustrated together, it is as follows. Here, the upper limit of  is 300 x 300.
Both sides almost coincide near x =1 , but do not coincide near x =0 .

Then, we attach the parallel accelerator ( See " 13 Convergence Acceleration of Multiple Series " (A la carte) )
to the right side of (3.2) .

q k-r-s (-1)r +s
 r +s 
 k
 
k k s
g(x,y,q ) = ΣΣΣ cos y log (3.2')
k=1 r=1 s=1 (q +1) k+1 ( r s) x r
When this is illustrated at q =1/2 and m =40 , it is as follows. ( m is the upper limit of  . Same as below.)
Both sides overlap exactly and look like spots.

(3.2) and (3.2') are the same. (3.2) may be used where y is large, but (3.2') is required where y is small. ..
Hereinafter, (3.2) is used for theoretical explanation, and (3.2') is used for drawing and calculation. Further,
2
x,y is described as gx,y .

-9-
4.3.3 2D figures of g(x,y)

(1) Cutting figure at y

The left is a cut figure at y = 14.1347 , 9.5 , 9.0647 gx ,y , and the right is a cut figure at
of
y = 72.21 . The red points are zeros on the critical line and the purple points are  specific zeros.

When observing both figures, it looks like gx ,y > 0 at x < 1/2 .

(2) Cutting figure at x

The left is a cut figure at x = 0 , 1/4 , 1/2 gx ,y , and the right is a cut figure at x = 1/2 ,
of
3/4 , 1 . The red points are zeros on the critical line and the purple points are  specific zeros.

In x  1/2 , y  3 ( left figure ) , gx,y is inversely proportional to x , maximum at x =0 and minimum

( =0 ) at x =1/2 . That is, gx,y has no zero at x < 1/2 , y  3 .

If the right figures of (1) and (2) are expressed analytically, the following hypothesis equivalent to the Riemann
hypothesis is obtained.

Hypothesis 4.3.3
When (x , y) is the Dirichlet eta function on the complex plane, the following inequality holds.
  (-1)r+s
 >0
s 0 < x < 1/2
gx,y = ΣΣ cos y log for (3.3)
r=1 s=1 (rs) x r y3

Note
Since gx,y  0 , this proof only excludes the equal sign. This may be the easiest in this chapter.

- 10 -
4.4 Theorems at Zeros
As seen in the previous section , squared absolute value of Dirichlet eta function was expressed as follows.
  (-1)r+s
 
s
(x , y ) = ΣΣ
2
x
cos y log  := g (x , y)  (3.2)
r=1 s=1 (rs) r

The following theorems hold for the zeros of this double series.

Theorem 4.4.0
When (x , y) is Dirichlet Eta Function, if (a , b) = 0 ,
  (-1)r+s
 =0
s
Σ Σ
r=1 s=1 (rs) a
cos b log
r
(4.0c)

  (-1)r+s
 =0
s
Σ Σ
r=1 s=1 (rs) a
sin b log
r
(4.0s)

Proof
Since the left side of (4.0c) is the absolute value of a ,b , (4.0c) is natural. (4.0s) is proved at the end
of this section.

Interestingly, at a zero point (a , b ) of (x ,y) , each of these rows have to be all 0 . Below, we state
this as a theorem.

Theorem 4.4.1
When (x , y) is Dirichlet Eta Function, if (a , b) = 0 ,
 (-1)r+s
 =0
s
Σ
s=1 (r s) a
cos b log
r
for r =1, 2, 3,  (4.1c)

 (-1)r+s
 =0
s
Σ sin b log for r =1, 2, 3,  (4.1s)
s=1 (r s) a r

Proof
Let Cr be the r th row of the double series (4.1c) . Then,
 (-1)r+s
 
s
Cr = Σ a
cos b log
s=1 (r s) r
(-1)r  (-1)s
 
s
=
ra
Σ
s=1 sa
cos b log
r
Here,

  = cos(b log r)cos(b log s) + sin(b log r)sin(b log s)

s
cos b log
r
Using this,
(-1)r  (-1)s
Cr =
ra
Σ
s=1 sa
cos(b log r ) cos(b log s) + sin(b log r) sin(b log s )

i.e.
(-1)r (-1)s (-1)s
cos(b log r)Σ 
 
Cr = cos(b log s) + sin(b log r)Σ sin(b log s)
ra s=1 sa s=1 sa

- 11 -
At a zero point (a , b ) of ,
s
 (-1)  (-1)s
Σ
s=1 sa
cos(b log s) = 0 , Σ
s=1 sa
sin(b log s) = 0

Therefore, Cra ,b = 0 for r =1, 2, 3,  .

In a similar way, let Sr be the r th row of the double series (4.1s) . Then,
 (-1)r+s
 
s
Sr = Σ a
sin b log
s=1 (r s) r
(-1)r  (-1)s
 
s
=
r a Σ
s=1 s a
sin b log
r
Here,

  = cos(b log r)sin(b log s) - sin(b log r)cos(b log s)

s
sin b log
r
Using this,
(-1)r (-1)s (-1)s
 
 
Sr = cos(b log r)Σ sin(b log s) - sin(b log r)Σ cos(b log s)
ra s=1 sa s=1 sa
For the same reason as the above, S r(a ,b ) = 0 for r =1, 2, 3,  at a zero point (a , b ) of (x ,y) .
From this, the following corollary follows.

Corollary 4.4.1'
When (x , y) is Dirichlet Eta Function, if (a , b) = 0 ,
 s

 =0
(-1) s
Σ
s=1 s a
cos b log
r
for r =1, 2, 3,  (4.1c')

 (-1)s
 =0
s
Σ
s=1 sa
sin b log
r
for r =1, 2, 3,  (4.1s')

Putting  = -b log r in this corollary, we obtain the following.

Corollary 4.4.1"
When (x , y) is Dirichlet Eta Function, if (a , b) = 0 , the following expressions hold for arbitrary real
number .
 (-1)s
Σ cos(b log s +  ) = 0 (4.1c")
s=1 sa
 (-1)s
Σ sin(b log s +  ) = 0 (4.1s")
s=1 sa

Example
We illustrate (4.1c"). Since the convergence of this left side is slow and it is difficult to draw an accurate figure,
we apply Knopp transformation ( See " 10 Convergence Acceleration & Summation Method " (A la carte) ) to
this as follows.

q k-s (-1)s
s
m k k
hc(x, y, , q, m) = ΣΣ cos(y log s + )
k =1 s=1 (q +1) k+1 sx

- 12 -
Here, this is illustrated at q =1/3 and m =20 .

The left figure is a cutaway view at y =14.1347 when x =1/2 . We can see that h c = 0 for any 
in this cutting surface.
The right figure is a cutaway view at y =14.1347 and y =21.0220 when x =1/2 . We can see
that h c = 0 for any  in these cutting surfaces. It is surprising that the contour lines appear innumerably in
such a twisted figure.

Note
However, Corollary 4.4.1" can be obtained directly and easily using the angle sum identities of trigonometric
functions. When R is a set of real numbers, the following equation holds for any   R .
cos( + b log s) = cos cos(b log s) - sin  sin(b log s)
Multiplying (-1)s-1/s
a
on both sides and summing up from 1 to  for s ,
 (-1)s-1  (-1)s
Σ cos( + b log s) = Σ cos  cos(b log s) - sin  sin(b log s)
s=1 sa s=1 sa
 (-1)s-1  (-1)s-1
= cos Σ cos(b log s) - sin  Σ sin(b log s)
s=1 sa s=1 sa
Since, (a , b) is a zero of (x ,y) ,
 (-1)s-1  (-1)s-1
Σ
s=1 sa
cos(b log s) = 0 & Σ
s=1 sa
sin(b log s) = 0

Substituting this for the rght side

 (-1)s-1
Σ cos( + b log s) = 0   R (4.1c")
s=1 sa
Replacing  with  + /2 ,
 (-1)s-1
Σ sin( + b log s) = 0   R (4.1s")
s=1 sa

Using Corollary 4.4.1' , we obtain the following important theorem.

Theorem 4.4.2
When (x , y) is Dirichlet Eta Function and c(r) is arbitrary real valued function,

- 13 -
if (a , b) = 0 , the following expressions hold.
 
Σ(-1)r+s (rs)a cos b log r 
c(r) s
Σ
r=1 s=1
=0 (4.2c)

 
sin b log 
c(r) s
Σ Σ
r=1 s=1
(-1)r+s
(rs)a r
=0 (4.2s)

Proof
From Corollary 4.4.1' (4.1c') ,
 (-1)s
 =0
s
Σ
s=1 sa
cos b log
r
for r =1, 2, 3, 
a
Multiplying both sides by (-1)rc(r)/r ,


 
c(r) s
Σ
s=1
(-1)r+s
(rs)a
cos b log
r
=0 for r =1, 2, 3, 

summing up for r , we obtain (4.2c) . In a similar way, (4.2s) is obtained.

Proof of Theorem 4.4.0 (4.0s)

Particularly placed c(r) = 1 in Theorem 4.4.2 (4.2s) , (4.0s) is obtained.

- 14 -
4.5 Partial Derivatives of Squared Absolute Value

4.5.1 First order Partial Derivatives

Formula 4.5.1
When squared absolute value of Dirichlet eta function is
  (-1)r+s
 
s
g(x , y) = ΣΣ = (x , y) 
2
cos y log (3.2)
x r
r=1 s=1 (rs)
The 1st order partial derivatives are givern as follows.
 

 
log r s
g x = -2ΣΣ(-1)r+s x
cos y log (5.1x)
r=1 s=1 (rs) r
 

 
log r s
g y = 2ΣΣ(-1)r+s sin y log (5.1y)
r=1 s=1 (rs)x r

Proof
Differentiating (3.2) with respect to x
 

 
log(rs) s
g x = -ΣΣ(-1)r+s x
cos y log (5.0x)
r=1 s=1 (rs) r
   

   
log r s log s s
= -ΣΣ(-1)r+s cos y log - ΣΣ(-1)r+s cos y log
r=1 s=1 (rs)x r r=1 s=1 (rs)x r
Swapping r and s in the 2nd term on the right side,
   

   
log s s log r r
-ΣΣ(-1)r+s x
cos y log = -ΣΣ(-1)s+r x
cos y log
r=1 s=1 (rs) r s=1 r=1 (sr) s
 

 
log r s
= -ΣΣ(-1)r+s cos y log
r=1 s=1 (rs)x r
 cos(-z) = cos z
Substituting this for the 2nd term on the right side, we obtain (5.1x) .
Next, differentiating (3.2) with respect to y
  log(s /r)
 
s
g y = -ΣΣ(-1)r+s x
sin y log (5.0y)
r=1 s=1 (rs) r
   

   
log s s log r s
= -ΣΣ(-1)r+s x
sin y log + ΣΣ(-1)r+s x
sin y log
r=1 s=1 (rs) r r=1 s=1 (rs) r
Swapping r and s in the 1st term on the right side,
   

   
log s s log r r
-ΣΣ(-1)r+s x
sin y log = -ΣΣ(-1)s+r x
sin y log
r=1 s=1 (rs) r s=1 r=1 (sr) s
 
sin y log 
log r s
= Σ Σ
r=1 s=1
(-1)r+s
(rs)x r
 sin(-z) = -sin z
Substituting this for the 1st term on the right side, we obtain (5.1y) .

- 15 -
4.5.2 Second order Partial Derivatives

Formula 4.5.2
When squared absolute value of Dirichlet eta function is
  (-1)r+s
 
s
g(x , y) = ΣΣ = (x , y) 
2
cos y log (3.2)
x r
r=1 s=1 (rs)
The 2nd order partial derivatives are givern as follows.
2
  log r log s + log r
 
s
g xx = 2ΣΣ(-1) r+s
x
cos y log (5.xx)
r=1 s=1 (rs) r
  2

 
log r s
g xy = -2ΣΣ(-1) r+s
x
sin y log  = gyx (5.xy)
r=1 s=1 (rs) r
2
  log r log s - log r
 
s
g yy = 2ΣΣ(-1) r+s
x
cos y log (5.yy)
r=1 s=1 (rs) r

Proof
The first-order partial derivatives in the proof of Formula 4.5.1 are
 

 
log(rs) s
g x = -ΣΣ(-1)r+s x
cos y log (5.0x)
r=1 s=1 (rs) r
  log(s /r)
 
s
g y = -ΣΣ(-1)r+s x
sin y log (5.0y)
r=1 s=1 (rs) r
Differentiating these with respect to x , y ,

 

  2 2

 
log (rs) s
g xx = ΣΣ(-1)r+s cos y log = ||2
r=1 s=1 (rs)x r x 2

2 2
  log r + 2log r log s + log s
 
s
= ΣΣ(-1)r+s cos y log
r=1 s=1 (rs)x r
2
  log r log s + log r
 
s
= 2ΣΣ(-1) 
r+s
cos y log cos(-z) = cos z
r=1 s=1 (rs)x r

 

  2
log(rs) log(s/r)
 
s
g xy = ΣΣ(-1)r+s sin y log = ||2
r=1 s=1 (rs)x r x y
  2   2

   
log s s log r s
= ΣΣ(-1)r+s sin y log - ΣΣ(-1)r+s sin y log
r=1 s=1 (rs)x r r=1 s=1 (rs)x r
  2   2

   
log r r log r s
= ΣΣ(-1)s+r sin y log - ΣΣ(-1)r+s sin y log
s=1 r=1 (sr)x s r=1 s=1 (rs)x r
  2

 
log r s
= -2ΣΣ(-1)r+s sin y log  sin(-z) = -sin z
r=1 s=1 (rs)x r
2

 

  2
log (s /r)
 
s
g yy = -ΣΣ(-1)r+s cos y log = ||2
r=1 s=1 (rs)x r y 2

- 16 -
 2 2
  log r - 2log r log s + log s
 
s
= -ΣΣ(-1) r+s
x
cos y log
r=1 s=1 (rs) r
2
  log r log s - log r
 
s
= 2ΣΣ(-1) 
r+s
x
cos y log cos(-z) = cos z
r=1 s=1 (rs) r

- 17 -
4.6 Partial Differential Coefficients at Zeros
In this section, we consider what the value of the partial derivative obtained in the previous section is at the
zeros of Dirichlet Eta Function (x , y ) .

4.6.1 1st-order Partial Differential Coefficients

Theorem 4.6.1
When a ,b is a zero of Dirichlet eta function (x , y ) , the following equations hold for the partial
derivatives in Formula 4.5.1 .
 

 =0
log r s
g xa ,b = -2ΣΣ(-1)r+s cos b log (6.1x)
r=1 s=1 (rs)a r
 

 =0
log r s
g ya ,b = 2ΣΣ(-1)r+s a
sin b log (6.1y)
r=1 s=1 (rs) r

Proof
Putting c(r) = log r in Theorem 14.4.2 , we obtain the desired expressions immediately.

Note
This theorem is called "stationary condition" in the extreme value problem.

4.6.2 2nd-order Partial Differential Coefficients

Theorem 4.6.2
When a ,b is a zero of Dirichlet eta function (x , y ) , the following equations hold for the partial
derivatives in Formula 4.5.2 .
  log r log s
 >0
s
g xx(a , b) = g yy(a , b ) = 2ΣΣ(-1)r+s a
cos b log (6.xx)
r=1 s=1 (rs) r
g xy(a , b) = g yx(a , b ) = 0 (6.xy)

Proof
Substituting zero point a ,b for (5.xx) , (5.xy) , (5.yy) in Formula 4.5.2 ,
2
  log r log s + log r
 
s
g xx(a , b) = 2ΣΣ(-1) r+s
a
cos b log
r=1 s=1 (rs) r
  2

 
log r s
g xy(a , b) = -2ΣΣ(-1) r+s
a
sin b log  = gyx(a , b ) 
r=1 s=1 (rs) r
2
  log r log s - log r
 
s
g yy(a , b) = 2ΣΣ(-1)r+s cos b log
r=1 s=1 (rs)a r
2
Since log r is a real valued function with respect to r , from Theorem 4.4.2 ,
  2

 =0
log r s
Σ Σ(-1)r+s
r=1 s=1 (rs)a
cos b log
r
Substituting this for the aboves,
g xy(a , b) = 0  = g yx(a , b)  (6.xy)
  log r log s
 
s
g xx(a , b) = g yy(a , b ) = 2ΣΣ(-1)r+s cos b log
r=1 s=1 (rs)a r

- 18 -
Here, zero a ,b of (x ,y) is the minimum value ( = 0 ) of the following expression.
  (-1)r+s
 
s
g(x , y) = ΣΣ x
cos y log (3.2)
r=1 s=1 (rs) r
So, gx(x , b) changes from negative to positive before and after a , and gy(a, y) changes from negative
to positive before and after b . Thus, it has to be gxx(a , b) > 0 , gyy(a , b ) > 0 .

Extreme Value Judgment by Hessian Matrix

Generally, in the extreme value problem, the extreme value is determined by the Hessian matrix as follows.
g xx(a ,b) >< 0
2
g xx(a ,b) g yy(a ,b) - gxy(a ,b ) >< 0
In the case of (3.2), if
g(a , b ) > 0 this judgment is necessary. However, if g(a , b) = 0 this judgment
is unnecessary. Because, g(a , b ) = 0 is the minimum value of g(x , y )  0 .
Even so, if we dare to make a judgment, these are as follows from Theorem 4.6.2 .
g xx(a ,b) > 0
2 2
g xx(a ,b) g yy(a ,b) - gxy(a ,b ) = gxx(a ,b) > 0
That is, g(a , b ) is determined to be the local minimum of g(x , y) .

Example a = 1/2 , b = 14.134725

g xx(a ,b) = g yy(a ,b) = 7.089093

Note
The two theorems in this section generally hold for holomorphic complex functions with zeros. The proof can
be easily done using Cauchy-Riemann equation and Laplace'equation.

- 19 -
4.7 Geometric Relationships between Functions
In this section, we consider the geometrical relationship between squared absolute value of Dirichlet eta
function and the partial derivatives a little analytically.

4.7.1 g(x,y) and gx(x,y)

Theorem 4.6.1
Let real valued function with two variables g(x , y) , g x(x , y) are as follows respectively.
  (-1)r+s
 
s
g(x , y) = ΣΣ  = (x , y) 
2
cos y log (3.2)
x r
r=1 s=1 (rs)
 

 
log r s
g x(x , y) = -2ΣΣ(-1)r+s cos y log (5.1x)
r=1 s=1 (rs)x r
Then, when a ,b is a zero of Dirichlet eta function (x , y ) ,

(1) b is the common root of g(a , y) = 0 , g x(a , y) = 0 .

(2) b is at least a multiple root in both g(a , y ) = 0 , g x(a , y ) = 0 .

(3) g(a , y) and g x(a , y ) are almost symmetric with respect to the y -axis.

g and gx on the critical line ( Red point is b )

Proof
(1) g(a , b) = (a , b) = 0 . On the other hand, g x(a , b ) = 0 according to Theorem 4.6.1 . So,
2

(a , b ) is the common root of g(x , y) and g x(x , y) ,

(2) Since g(x , y) = (x , y ) , the root of g(a , y) = 0 is at least multiple root. On the other hand,
2

since gxy(a , b ) = 0 from Theorem 4.6.2 , g x(a , b) has to be an extremum with respect to y .

Therefore, the root b of g x(a , y) = 0 is also at least a multiple root.

(3) Expanding cosy log(s/r) to the Maclaurin series,

2t

log ( s/r) 2t
 
s
cos y log = Σ(-1) yt
r t=0 (2t) !
Using this, g(x , y) , gx(x , y) are expressed as follows, respectively.

- 20 -
 ΣΣ(-1) y
2t
   1 log ( s/r)
g(x , y) = Σ r+s+t 2t
(7.1)
x (2t) !
t=0 r=1 s=1 (rs)

 
2t
  
log r log ( s/r)
g (x , y) = -2Σ ΣΣ(-1) r+s+t 2t
x y (7.1x)
x (2t) !
t=0 r=1 s=1 (rs)
Both (7.1) and (7.1x) are even functions with respect to y , and have similar shapes except for the sign.
However, the two have different signs. So, g x(a , b ) has to be a local maximum with respect to the y .
If (1) , (2) are added to this, g(a , y) and g x(a , y) are almost symmetric with respect to the y -axis.

Dirichlet eta function has  specific zeros at x =1 . The 2D figure of this is as follows. The difference from
the zeros on the critical line is that b is not the maximum point of gx(a , y) but the local maximum point.

g and gx on x =1 ( Purple point is b )

4.7.2 g(x,y) and gy(x,y)

These are the functions expressed by the following equations.
  (-1)r+s
 
s
g(x , y) = ΣΣ  = (x , y) 
2
cos y log (3.2)
x r
r=1 s=1 (rs)
 

 
log r s
g y(x , y) = 2ΣΣ(-1)r+s x
sin y log (5.1y)
r=1 s=1 (rs) r
These cutting figure at x = 1/2 & x = 1 are as follows. The left is a cut figure at x = 1/2 and the left is
a cut figure at x = 1 . The red points are zeros on the critical line and the purple points are  specific zeros.

- 21 -
The importance of this pair is that gy is a derivative of g with respect to y . That is, on the y -axis, the
extremum of g corresponds to the zero of g y . The black vertical lines show that. Therefore, they have to

match at zero (a , b ) of (x , y) . It is according to Theorem 4.6.1 . Further, the local minimum points of
g are the uphill zeros of gy . Theorem 4.6.2 g yy(a , b ) > 0 supports this.

4.7.3 gx(x,y) and gy(x,y)

 

 
log r s
g x(x , y) = -2ΣΣ(-1)r+s x
cos y log (5.1x)
r=1 s=1 (rs) r
 

 
log r s
g y(x , y) = 2ΣΣ(-1)r+s x
sin y log (5.1y)
r=1 s=1 (rs) r
These cutting figure at x = 1/2 & x = 1 are as follows. The left is a cut figure at x = 1/2 and the left is
a cut figure at x = 1 . The red points are zeros on the critical line and the purple points are  specific zeros.

As seen in Theorem 4.7.1 , gx shares zeros with g on the y -axis, and they are local maximum. And

gx = gy = 0 from Theorem 4.6.1 .Thus, the red and purple dots are their zeros. And these shared zeros

of gx and gy are always the uphill zeros of gy . ( i.e. g yy(a , b ) > 0 )

4.7.4 g(x,y) and gxy(x,y)

 

 
log r s
g x(x , y) = -2ΣΣ(-1)r+s cos y log (5.1x)
r=1 s=1 (rs)x r
  2

 
log r s
g xy(x , y) = -2ΣΣ(-1) x
r+s
sin y log = g yx(x , y) (5.xy)
r=1 s=1 (rs) r
These cutting figure at x = 1/2 & x = 1 are as follows. The left is a cut figure at x = 1/2 and the left is
a cut figure at x = 1 . The red points are zeros on the critical line and the purple points are  specific zeros.

gxy is a derivative of gx with respect to y . So, the extremums of gx correspond to the zeros of g xy .

- 22 -
The black vertical lines show that. Further, the local maximum points of gx coincide with the downhill zeros
of g xy .

4.7.4 2D figures of g x(x , y )

Of the first-order partial derivatives, the one of particular interest is gx . So let us observe this in a little
more detail.

(1) Cutting figure at y

The left is a cut figure at y = 14.1347 , 10 , 9.0647 of gxx ,y , and the right is a cut figure at
y = 72.21 . The red points are zeros on the critical line and the purple points are  specific zeros.

Observing both figures, it seems that g xx ,y < 0 at x < 1/2 .

(2) Cutting figure at x

The left is a cut figure at x = 0 , 1/4 , 1/2 of gxx ,y , and the right is a cut figure at x = 1/2 ,
3/4 , 1 . The red points are zeros on the critical line and the purple points are  specific zeros.

Inx  1/2 , y  3 ( left figure ), g xx,y is inversely proportional to x , minimum at x =1/2

and maximum at x = 0 . That is, g x(x , y ) has no zero in x  1/2 , y  3 .

If the left figures of (1) and (2) are expressed analytically, the following hypothesis is obtained, which is
equivalent to the Riemann hypothesis.

Hypothesis 4.7.5
When (x , y) is the Dirichlet eta function on the complex plane, the following inequality holds.
 

 <0
log r s 0 < x < 1/2
g xx,y = -2ΣΣ(-1)r+s cos y log for (7.5)
r=1 s=1 (rs) x r y3

- 23 -
Note
This hypothesis excludes the possibility of gxx,y  0 in the critical strip. Unlike Hypothesis 4.3.3 ,

not only " = " but " > " must be excluded. It may be more difficult than the proof of Hypothesis 4.3.3 .

2018.05.01
2019.07.07 Renewed
2022.01.22 Renewed
Kano Kono
Alien's Mathematics

- 24 -