Research Article: An Integral Equation For Riemann's Zeta Function and Its Approximate Solution

Hindawi
Abstract and Applied Analysis

Volume 2020, Article ID 1832982, 29 pages
https://doi.org/10.1155/2020/1832982
Research Article
An Integral Equation for Riemann’s Zeta Function and Its
Approximate Solution
Michael Milgram
Geometrics Unlimited, Ltd., Box 1484, Deep River, Ontario, Canada K0J 1P0
Correspondence should be addressed to Michael Milgram; [email protected]
Received 23 October 2019; Accepted 6 January 2020; Published 15 May 2020
Academic Editor: Ying Hu
Copyright © 2020 Michael Milgram. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Two identities extracted from the literature are coupled to obtain an integral equation for Riemann’s ξ(s) function and thus ζ(s)
indirectly. The equation has a number of simple properties from which useful derivations flow, the most notable of which relates
ζ(s) anywhere in the critical strip to its values on a line anywhere else in the complex plane. From this, both an analytic expression
for ζ(σ + it), everywhere inside the asymptotic (t ⟶ ∞) critical strip, as well as an approximate solution can be obtained, within
the confines of which the Riemann Hypothesis is shown to be true. The approximate solution predicts a simple, but strong
correlation between the real and imaginary components of ζ(σ + it) for different values of σ and equal values of t; this is illustrated
in a number of figures.
1. Introduction plane can be determined with respect to the

properties of ζ(s2 ) (the “master”) on a line (R(s2 ) �
The Riemann zeta function ζ(s) is well known to satisfy a constant) anywhere else in the punctured (s1 ≠ s2 )
functional equation, and many representations, both in- complex plane
tegral and series, have been developed over the years for
(ii) The transfer function that mediates the aforemen-
both ζ(s) and its avatar ξ(s). Additionally, and more
tioned action is a simple rational polynomial
significantly, at least two independent contour integral
function of s1 and is therefore quite amenable to
representations are known ([1], Section (1.4)) and ([2],
analysis
equation (7)) either of which could be utilized as the
primary definition, from which many of the properties of (iii) The only s1 dependence resides in the transfer
ζ(s) can be derived. On the contrary, it has been long-ago function
proven that ζ(s) does not satisfy a differential equation of
In this work, after initially listing some definitions and
quite general form ([3] (and citations therein)); also known
lemmas (Section c), a derivation of the integral equation over
are three integral equations which ζ(s)does satisfy: [4],
a bounded region of the s2 -plane (Section 3) is given, and its
equation (1.6), [5], equation (3.29), and [6], equation (1.5)
analytic continuation over the remainder of the punctured
with α � 0. In this work, a series representation of ξ(s) in
(s1 ≠ s2 ) complex plane (Section 4) is obtained. In Section 5,
terms of generalized integro-exponential functions (Es (z))
various representations are then used to obtain a few simple
is coupled with a contour integral representation of these
integrals involving ζ(s2 ), corresponding to special values of
functions, to obtain a new integral equation satisfied by
s1 ; it is also demonstrated that the simple integrals under
ξ(s) and, equivalently, ζ(s). This equation has several re-
consideration are convergent.
markable properties:
Following these preliminaries, the general form of the
(i) Through the action of an integral operator, the value transfer function is carefully presented in a series of Ap-
of ζ(s1 ) (the “dependent”) anywhere in the complex pendices. Based on a particularly useful property of the
2 Abstract and Applied Analysis
transfer function presented in Appendix C—it closely Definition. Polar form means that a complex function f(s)
mimics a Dirac delta function on the s2 line— a reasonably is written as exp(iθ(s))|f(s)|, where θ(s) ≡ arg(f(s)).
accurate model for ζ(s1 ) is established, which yields rea-
(i) The following asymptotic limits ([8], equation (5.6.9);
sonably accurate estimates of the asymptotic nature of ζ(s1 )
see also [7], equation (4.12.2)) will be required:
inside the critical strip (Section 6). In consequence, it is
shown that, asymptotically
􏼌􏼌 􏼌 (σ− 1)/4
|ζ(σ + it)| ∼ t1/2− σ/2 log(log(t)) 􏼌 σ it 􏼌􏼌 √�� σ 2 t2 − πt
×􏼠
log(t)
+ · · ·􏼡, (1) lim 􏼌􏼌􏼌􏼌Γ􏼒 + 􏼓􏼌􏼌􏼌􏼌 ≈ 2π􏼠 + 􏼡 exp􏼒 􏼓, (6)
t⟶∞ 2 2 4 4 4
and, within the confines of the model, it is both proven that
ζ(s1 ) ≠ 0 if R(s1 ) ≠ 1/2, and explained, in transparent terms and the special case
(Section 7), why this happens (does not happen?). Graphical
examples and comparisons are given, demonstrating the 􏼌􏼌 􏼌
􏼌 1 it 􏼌􏼌 √�� − πt
accuracy of a predicted universality between the real and lim 􏼌􏼌􏼌􏼌Γ􏼒 ± 􏼓􏼌􏼌􏼌􏼌 ≈ 2π exp􏼒 􏼓. (7)
imaginary components of ξ(σ + it) at equal values of t but
t⟶∞ 2 2 4
different σ. In Section 8, two well-accepted properties of
ζ(s), notably Universality ([7], Chapter XI) and Lindelöf’s
hypothesis ([1], Section 9.2) are discussed in the context of Remark. Approximation (6) was tested numerically and
the results reported here, and suggestions are made that found to be remarkably accurate for even modest values of t
merit future consideration. Section 9 discusses requirements when 0 < σ < 1 ≪ t.
needed to improve the rigour of the model developed here. (ii) From the functional equation of ζ(s) with (6), for
large values of t,
2. Preamble
2.1. Notation. Throughout, I use s � σ + it to denote the 􏽲��
t
independent variable defining ζ(s). In those instances, |ζ(it)| ∼ |ζ(1 − it)| . (8)
2π
where dependence on, for example, t is the focus, I will
sometimes just shorten, for example, M(c, s, v) ⇒ M(t, v),
for typographical brevity and clarity. Throughout, subscripts
‘R’ and ‘I’ refer to the real and imaginary components of Lemma 1
whatever they are attached to. Much use is made of Rie- 1 1 it t log(π)
arg􏼒ζ 􏼒 + it􏼓􏼓 + arg􏼒Γ􏼒 + 􏼓􏼓 − � ± kπ. (9)
mann’s ξ function, defined by 2 4 2 2
s
ξ(s) ≡ (s − 1)π− s/2 ζ(s)Γ􏼒1 + 􏼓, (2) Well-known ([7], equation (4.17.2)) with k � 0, this is
2 sometimes referred to as the Riemann–Siegel identity. A
as well as simple derivation with k ≠ 0 follows.
s
Υ(s) ≡ ζ(s)Γ􏼒 􏼓π− s/2
, (3)
2 Proof. Expand ξ(σ + it) about σ � 1/2 at constant t giving
􏼌􏼌
both of which satisfy 1 z 􏼌
lim ξ(σ + it) � ξ 􏼒 + it􏼓 + 􏼠 ξ(σ + it)􏼌􏼌􏼌􏼌 􏼡
ξ(s) � ξ(1 − s), (4)
σ⟶1/2 2 zσ σ�1/2
(10)
1
Υ(s) � Υ(1 − s). (5) · 􏼒σ − 􏼓 + · · · ,
2
Throughout, k, N � 0, 1, 2, . . .. and find the imaginary part from (2), after rewriting ζ(s) and
Γ(s/2) in polar form:
2.2. Definitions and Lemmas
1
I􏼒ξ 􏼒 + it􏼓􏼓 ≡ I􏼒 lim ξ(σ + it)􏼓
2 σ⟶1/2
(11)
4t2 + 1􏼁sin[arg(ζ(1/2 + it)) + arg(Γ(1/4 + it/2)) − t/2ln(π)]|Γ(1/4 + it/2)ζ(1/2 + it)|
�− .
8π1/4
Abstract and Applied Analysis 3
Because ξ((1/2) + it) is known to be real, (11) vanishes, the critical line on that basis. Here, the sums are treated as
and (9) follows immediately. QED. □ an infinite series representation of ξ(s), and hence an
identity because each series is easily shown to be con-
Remark. Apply the identity vergent due to the asymptotic property of Es (z) (see [12],
equation (2.25)). Similar, but inequivalent series repre-
1 sentations will be found in Paris ([14], equation (1)),
arg􏼒ζ 􏼒 + it􏼓􏼓 � I(log Γ(1/2 + it)) + N. (12)
2 Patkowski ([4], equation (1.20)), and elsewhere. From [12],
some useful integral and contour integral representations
to (9) to obtain Backlund’s formula which counts the dis-
of the function Es (z) are
continuities (not zeros) of arg(ζ((1/2) + it)) (see [9], Section
9). This is consistent with Theorem 1 in [10].
Es (z) � zs− 1 Γ(1 − s, z), (21)
(iii) Define
∞
� 􏽚 v− s exp(− zv) dv, (22)
1
s 1
ϕ(s) ≡ arg(ζ(s)) + arg􏼒Γ􏼒 􏼓􏼓 − 􏼒 􏼓t log(π) � arg(Υ(s)),
2 2 1 c+i∞ Γ(− v)zv
Es (z) � 􏽚 dv. (23)
(13) 2πi c− i∞ s − 1 − v
3 In (21), Γ(1 − s, z) is the (upper) incomplete gamma
θ(v) � arg􏼒Γ􏼒 + iv􏼓􏼓 + arg(ζ(1 + 2iv)) − v ln(π), (14) function, and (22) provides the fundamental definition of
2
Es (z). The result (23) is given in [12] (equation (2.6a)). Here,
it t the integration contour, originally defined to enclose the real
Φ(t) ≡ arg(ζ(it)) + arg􏼒Γ􏼒 􏼓􏼓 − 􏼒 􏼓log(π), (15)
2 2 axis v ≥ 0 as well as the singularity at v � s − 1 in a clockwise
direction, has been converted into the line c < R(s) − 1
1 because the integrand vanishes as v ⟶ ± i∞. This paper
α(t) ≡ argζ 􏼒 + it􏼓, (16)
2 investigates the application of (23) to (19).
1
β(t) ≡ argζ ′ 􏼒 + it􏼓. (17)
2 3.1. A Novel Integral Representation. The following, focuses
on the limited range 0 ≤ σ ≤ 1, where s � σ + it, in which
However, for brevity, expressions can be abbreviated as
case, applying (23) to the first term on the right-hand side of
follows:
(19) yields
cos(Φ(t)) ≔ cos(Φ). (18)
∞
π(s − 1) 􏽘 n2 E− s/2 􏼐πn2 􏼑
n�1
3. LeClair’s Representation (24)
In a recent work, LeClair ([11], equation (15)) has obtained (s − 1) c1 +i∞ Γ(− v)π v 􏽐∞
n�1 n
2+2v
� 􏽚 dv,
the following series representation of Riemann’s ξ 2i c1 − i∞ − s/2 − 1 − v
function:
∞ ∞ (s − 1) c1 +i∞ Γ(− v)π v ζ(− 2v − 2)
2 2 2 2 � 􏽚 dv, (25)
ξ(s) � (s − 1)π 􏽘 n E− s/2 􏼐πn 􏼑 − πs 􏽘 n E(s− 1)/2 􏼐πn 􏼑 2i c1 − i∞ − s/2 − 1 − v
n�1 n�1
∞ where the interchange of integration and summation in
πn2
+ 4π 􏽘 n2 e− , (24) is justified if c1 < − 3/2 since the sum converges under
n�1 this condition. The general requirement that c1 < − σ/2 − 1
(19) imposes the effective constraint c1 < − 3/2 for 0 ≤ σ ≤ 1: the
contour may be shifted leftwards with impunity since
where generically, Es (z) is the (generalized) “Exponential there are no singularities in that direction. Similarly,
Integral,” a limiting case of what is elsewhere [12] re- under the transformation s ⟶ 1 − s, we find the fol-
ferred to as the “Generalized Integro-Exponential lowing expression for the second right-hand side term of
Function.” Furthermore, it has been shown ([13], (19):
equation (3)) that the infinite sum in (19) can be readily
evaluated as follows: ∞
s c2 +i∞ Γ(− v)π v ζ(− 2v − 2)
∞ 1/4
− πs 􏽘 n2 E(s− 1)/2 􏼐πn2 􏼑 � − 􏽚 dv,
πn2 π 2i c2 − i∞ − 3/2 + s/2 − v
4π 􏽘 n2 e− � . (20) n�1
n�1 2Γ(3/4) (26)
In his work, LeClair truncated the sum(s) at N terms, again valid for c2 < − 3/2 if σ > 0. In the case that c1 � c2 � c,
referred to the result as an “approximation” and proceeded where c < − 3/2, and focussing on 0 ≤ σ ≤ 1, we find the in-
to obtain approximations to the location of the zeros on tegral equation:
is(s − 1/2) c+i∞ ξ(v + 3) Notice that with a simple application of the recursion
ξ(s) � 􏽚 dv formula for Γ(− v), the numerator in (28) can be written in
π c− i∞ (v + 3)(3 − s + v)(s + 2 + v)
terms of Υ(− 2v − 2) (see (3)). Alternatively, under a simple
i c+i∞ ξ(v + 3) π1/4 change of variables, with the same condition c < − 3/2 and the
+ 􏽚 dv + , same range of σ, (28) becomes
2π c− i∞ (v + 3)(s + 2 + v) 2Γ(3/4)
(27) π1/4
ξ(s) � − J(s, c), (29)
2Γ(3/4)
which, in terms of ζ(s), can be rewritten as
s 1 where
(s − 1)ζ(s)Γ􏼒 + 1􏼓π− s/2 � 2is􏼒s − 􏼓
2 2
c+i∞ Γ(− v)π v ζ(− 2v − 2)
·􏽚 dv (28)
c− i∞ (− 3 + s − 2v)(s + 2 + 2v)
i c+i∞ Γ(− v)π v ζ(− 2v − 2) π1/4

− 􏽚 dv + .
2 c− i∞ s/2 + 1 + v 2Γ(3/4)
∞
J(s, c) � 􏽚 πc− iv ζ(− 2c + 2iv − 2)Γ(− c + iv)M(c, s, v) dv, (30)
−∞
􏼐− (2s − 1)2 /2 + 2iv − 2(c + 5/4)􏼑

M(c, s, v) � . (31)
− (2s − 1)2 /4 + 2i(2iv − 4(c + 5/4))v + 4(c + 5/4)2
This result is worthy of a few comments: 4. Analytic Continuations

(i) Its form is almost (exception: see [4], equation In the following, let c1 � c2 � c. By shifting the contour (28)
(1.18)) unique among representations of ζ(s). into the region c > − 3/2 (see Figure 1(a)), various repre-
Usually s-dependence that is formally embedded sentations are obtained for each of the regions labelled in
inside an integral or series representation of ζ(s) Figure 1(b), by subtracting (2πi×) the residues of the sin-
appears either as an exponent or buried inside the gularities that are transited in the complex v-plane. To enter
argument of a transcendental function; here, s-de- the region labelled “II” in Figure 1(b), the contour in
pendence exists only in the form of a coefficient in a Figure 1(a) must pass to the right of the fixed pole at
simple rational (polynomial) function. This augers v � − 3/2, whose residue is given by
well for further analysis and leads to some sur-
prising predictions; 1
residueblue � , (32)
(ii) (27) and (28) present a prescription, in the form of (2πi)
an integral transform, for the value of ξ(s) or ζ(s)
anywhere in the complex strip 0 ≤ R(s) ≤ 1, that giving, for region II,
depends only on its values on a vertical line in the
π1/4
complex plane corresponding to c < − 3/2. The re- ξ(s) � − 1 − J(s, c). (33)
gion in which (29) is valid is labelled “I” in 2Γ(3/4)
Figure 1(b) and delineated as everything to the left To enter the region labelled III from region II (crossing
of the point at v � − 1.5 in Figure 1(a); the red line in Figure 1(b)), the corresponding residue (red
(iii) In (29) and (30), the function M(c, s, v) acts as a pole in Figure 1(a))
transfer function between ζ(v) on the line R(v) �
− c − 1 and ζ(s) elsewhere in the complex plane, iξ(s)
residuered � , (34)
through the medium of the integral operator (30). (2π)
Its properties will be of interest in studying (29) for
varying values of the parameters c and s; must be subtracted. The result, valid for region III is
(iv) Based on (4), it is possible to show ([15], Theorem 1) π1/4
0� − 1 − J(s, c). (35)
that (28) can be cast into a simple closed (Cauchy) 2Γ(3/4)
contour integral for ξ(s).
To cross the green line, either from region II, entering
In the following sections, simple choices of s and c, c1 , region V, or from region III, entering region IV, the
and c2 will be applied to (29). residue
–1
i∞
3
–1.1 IV
2
1 –1.2 V III
(v)
0
c –1.3
–2 –1.5 –1
(v)
–1
–1.4 II
–2
–1.5
–3 I
–i ∞
–1.6
0 0.2 0.4 0.6 0.8 1
σ
(a) (b)
Figure 1: The various regions of validity of the analytic continuations developed in Section 4 both (a) in the complex v-plane and (b) in the
cross section of a complex plane cut by the real (σ, c) plane at t � 0. (a) The location of the contour of integration (black arrow at
R(v) < − 3/2) extending from − i∞ to i∞ as well as the various poles. The fixed pole at v � − 3/2 corresponds to the singularity belonging to
ς(− 2v − 2). The green and red poles correspond to singularities of the integrand that depend on the value of s acting as a parameter. The
coloured arrows show the motion of corresponding poles as R(s) increases from zero to one if t > 0. The colours correspond to colours used
in Figure 1(b). (b) The various regions where different continuations of (29) apply, projected onto the t � 0 plane. Five distinct regions are
bounded by (coded and coloured) lines c � − 3/2 (blue, solid), c � (− σ/2) − 1 (red, dash-dot), and c � σ/2 − 3/2 (green, dash-dot) when
0 ≤ σ ≤ 1. The bounding regions extend vertically out of the plane of the figure when t ≠ 0 because the contour used in (23) is chosen to be a
straight line, vertical in the complex v-plane (c is a constant). The dotted line corresponds to the interesting case c � − 5/4.
isπ(s− 3)/2 ζ(1 − s)Γ(− s/2 + 1/2)(s − 1) 5. Crossing the Line at c � − 3/2
residuegreen � ,
4
(36) Remark. Several of the identities in this section are (re-
dundantly) declared to be real and marked R. This is done as
corresponding to the green pole in Figure 1(a) must be a reminder that unless so-declared, an attempted numerical
subtracted. A similar argument applies for a transition from computer evaluation of that expression will likely fail be-
region V to region IV, the residue (34) must be subtracted. cause the imaginary part of the identity contains a singu-
This gives, for region IV, larity (e.g., (42)).
π1/4
− ξ(s) � − 1 − J(s, c). (37)
2Γ(3/4)
5.1. The General Case c � − 3/2. Disregarding the contour
In particular, this analysis relates the results applicable to integral representations from which they are obtained, (29),
region IV to that for region I, including regions where both (33), and (35) simply involve several complex functions of two
representations are unconstrained with respect to 0 ≤ σ ≤ 1 real variables (if t � 0 ) and ought to be amenable to analysis
and equates the result for region V to that for region III. It is from that viewpoint. Compare (29) and (33), respectively,
emphasized that the regional definitions in Figure 1(b) exclude below and above the horizontal line demarking regions I and
the lines, thereby defining the punctured regions (s1 ≠ s2 ) II of Figure 1(b) by setting c � − 3/2 − η in (29) and c �
alluded in the Introduction. Of particular interest, is the value − 3/2 + η in (33), where 0 < η ≈ 0. Note that η is a real variable
at the point (s � 1/2, c � − 5/4), as well as the values along the (there is no cut); since none of the terms in either equation
horizontal (dotted) line s � σ + it|c�− 5/4 and a vertical line (other than J(s, c)) depend on c, we are free to choose a
c|s�1/2+it . To reiterate, Figure 1(b) is intended to illustrate the convenient value for c, subject to the (regional) conditions
projection of a (complex) space (s � σ + it) defined by a third under which each equation is valid. We have in region I:
axis extending out of the plane of the Figure, corresponding to π1/4 −3
t ≠ 0, onto the (real) plane (σ, c) where t � 0. ξ(s) � − J􏼒s, − η􏼓, (38)
2Γ(3/4) 2
and in region II where c is Euler’s constant. Therefore, the real part of the
π1/4 −3 integrand converges at v � 0 and the imaginary part, sin-
ξ(s) � − 1 − J􏼒s, + η􏼓. (39) gular at the origin, integrates to zero by antisymmetry.
2Γ(3/4) 2
In general, when s � 1/2 + it, the left-hand side of both
Simple comparison between (38) and (39) suggests that (38) and (39) are real, and so the equivalent of (42) applies.
−3 −3 Written in terms of t, (42) generalizes to
J􏼒s, − η􏼓 − J􏼒s, + η􏼓 � 1. (40)
2 2 1
ξ 􏼒 + it􏼓
This can be independently verified by a change of var- 2
iables v ⟶ ηv in both integrals, followed by a series ex- 1 ∞ π− iv ζ(1 + 2iv) 4t2 + 4iv + 1􏼁Γ(3/2 + iv)
pansion about η � 0, leading to �− 􏽚 R􏼠 􏼡dv
2π3/2 − ∞ t2 + 2iv − 4v2 + 1/4
−3 −3 1 ∞ 1
J􏼒s, − η􏼓 − J􏼒s, + η􏼓 ⟶ 2 􏽚 dv � 1. 1 π1/4
2 2 π − ∞ 1 + v2 − + .
(41) 2 2Γ(3/4)
(44)
As is often done where discontinuities arise in the theory
of functions of a real variable, when η � 0 it is conventional
to assign half the discontinuity at that point (c � − 3/2). This
is equivalent to deforming the contour in (28) to avoid the 5.2. The Case c � − 5/4. In regions III and V, writing (35) in
fixed pole belonging to ζ(− 2v − 2) at v � − 3/2 by including full gives
only half the residue at that point. ∞ π− iv ζ(− 2c + 2iv − 2)Γ(− c + iv) 2iv − 2s2 − 2c + 2s − 3􏼁
πc 􏽚 dv
−∞ (− 2c + 2iv + s − 3)(− 2c + 2iv − s − 2)
Remark. This cannot be done for the other poles since they
are located by s (complex), rather than v � − 3/2 (fixed). π1/4
� − 1,
2Γ(3/4)
(45)
5.1.1. The Case σ � 1/2. The result from (38) and (39) after or, equivalently
setting η � 0 (see Section 5.1) when s � 1/2 is ∞ 2iv − 2s2 − 2c + 2s − 3􏼁ξ(− 2c + 2iv − 2)
1 ∞ ξ(1 + 2iv) 􏽚 dv
􏽚 I􏼠 􏼡dv − ∞ (− 2c + 2iv + s − 3)(− 2c + 2iv − s − 2)(− 2c + 2iv − 3)
2π − ∞ (2iv + 1/2)v
π1/4
� − 1.
∞ π− 3/2− iv
ζ(1 + 2iv)Γ(3/2 + iv) 2Γ(3/4)
�􏽚 R􏼠 􏼡dv (42) (46)
−∞ 2iv + 1/2
Setting c � − 5/4 in (45) leads to
√�
3/4
ζ(1/2)π 2 π1/4
� + − 1/2. ∞ ζ(1/2 + 2iv)π − 5/4− iv
Γ(5/4 + iv)􏼐iv − (2s − 1)2 /4􏼑
8Γ(3/4) 2Γ(3/4) 􏽚 dv
−∞ (2s − 1)2 /8 + 2v2
This result (compared with ([7], page 204, Lemma β))
can be verified numerically. The need to specify that only the π1/4
real part of the integrand in (42) is to be used is twofold: �1− ,
2Γ(3/4)
(i) The right-hand side is real and so must be the left- (47)
hand side(s).
(ii) Ostensibly, the integrand of (42) appears to be a result that is valid for all s (see the dotted line in
singular at v � 0 unless one notes that Figure 1(b)). Since the left-hand side is a function of s and
the right-hand side is not, it must be true that the derivative
of the left-hand side with respect to s vanishes. Performing
i this calculation (the integral is convergent, so the derivative
ζ(1 ± 2iv) ≈ ∓ + c + ···, (43)
(2v) and integral operator can be interchanged) yields
∞ π− 5/4− iv
􏼐− 2 − iv +(2s − 1)2 /4􏼑ζ(1/2 + 2iv)Γ(5/4 + iv)
(s − 1/2) 􏽚 2 dv � 0. (48)
−∞ 􏼐(2s − 1)2 /8 + 2v2 􏼑
4 ∞ ξ(1/2 + 2iv) π1/4

Compare (47) with Patkowski ([4], Theorem 1). 􏽚 dv � 1 − , (52)
π − ∞ 16v2 + 1 2Γ(3/4)
5.2.1. σ � 1/2. Consider (47) in the case s � 1/2, which gives and, for the imaginary part,
∞ vξ(1/2 + 2iv)
5/4− iv
∞ iπ − 1 5 π1/4 􏽚 dv � 0. (53)
􏽚 ζ 􏼒 + 2iv􏼓Γ􏼒 + iv􏼓dv � 1 − , 2 2 2
− ∞ (4v − t )(16v + 1)
−∞ 2v 2 4 2Γ(3/4)
(49) Notice that in the former case, there is no t dependence,
so in (51), the variable t is in reality a free parameter,
and the integrand apparently has a singularity at v � 0. consistent with the argument applied to obtain (48). In the
However, a simple expansion of the integrand about that latter case, although a singularity exists, the integrand is
point shows that antisymmetric about v � 0, and so the singularities at ± v
5/4− iv
iπ− 1 5 ζ(1/2) cancel.
R􏼠 ζ 􏼒 + 2iv􏼓Γ􏼒 + iv􏼓􏼡 � − √� 1/4 + O􏼐v2 􏼑,
2v 2 4 2π Γ(3/4)
(50) 6. An Asymptotic Approximation to ζ(σ + it)
demonstrating that the real part of the integrand in (49) is 6.1. Integral Equations for the Real and Imaginary Parts of
nonsingular at v � 0. However, near v � 0, it is similarly ξ(s). Consider (37)—region IV—in the general case
shown that the imaginary part of the integrand diverges like s � σ + it, written in terms of the transfer function
v− 1 and the imaginary part of the integral vanishes by M(c, s � σ + it, v). The case c � − 1 is of interest since the
antisymmetry about v � 0. integral will span the 0-line, which connects to the 1-line by
In the case s � 1/2 + it, (47) can be written in terms of reflection (see (8)), and, acting in its capacity as a master
ξ(1/2 + 2iv) as function, ζ(1 ± it) has many well-known properties (e.g.,
1 ∞ ξ(1/2 + 2iv) iv + t2 􏼁 π1/4 [16]), the most relevant one being that it has no complex
􏽚 dv � 1 − , (51) zeros. In addition, according to Figure 1(b), (37) is valid for
π − ∞ (− t2 /2 + 2v2 )(− 1/2 + 2iv) 2Γ(3/4)
all values of σ spanning the critical strip 0 < σ < 1. In this
whose integrand appears to become singular at v � ± t/2. case, the transfer function Mx (− 1, σ + it, v) and its real and
Writing the integrand in terms of its real and imaginary imaginary components, shortened, except where necessary,
parts, and noting that ξ(1/2 + 2iv) is real, we find, for the real to MR (t, v) and MI (t, v), are, respectively,
part
(2σ + 2it − 1)(σ + it)

Mx (− 1, σ + it, v) � − +(2iv − σ − it)− 1 , (54)
(2iv − 1 + σ + it)(2iv − σ − it)
8v2 t2 − σ 2 + σ 􏼁 + 2tv(1 − 2σ) + 2σ 3 (2 − σ) − 4t2 + 3􏼁σ 2 + 4t2 + 1􏼁σ − 2t4 − t2

MR (t, v) � − , (55)
(t2 + 4vt + σ 2 + 4v2 − 2σ + 1)(t2 − 4vt + σ 2 + 4v2 )
8v3 +(− 16σ + 8)tv2 + − 6t2 + 6σ 2 − 6σ + 2􏼁v +(2σ − 1)t

MI (t, v) � − . (56)
(t2 + 4vt + σ 2 + 4v2 − 2σ + 1)(t2 − 4vt + σ 2 + 4v2 )
π1/4 i ∞
Furthermore, these functions possess the following 1− − ξ(σ + it) � − 􏽚 MR (t, v) + iMI (t, v)􏼁
2Γ(3/4) π −∞
symmetry:
· Υ(2iv)v dv.
MR (σ + it, v) � MR (1 − σ + it, − v), (57)
(59)
MI (σ + it, v) � − MI (1 − σ + it, − v), (58) It is now convenient to rewrite (59) as an integral over
the range [0 · · · ∞) by first splitting the integral, setting
which symmetry also holds true for all values of c. A v ⟶ − v and combining the two halves. The expression
complete description of these functions is given in Ap- obtained is straightforward, although rather lengthy. A
pendixes A and B. From (37), the basic equation in region IV second lengthy equation can be obtained by first replacing
using c � − 1 is σ ⟶ 1 − σ in (59) and similarly reducing the range to
[0 · · · ∞]. After splitting Υ(2iv) into its real and imaginary 2σv − t 2(1 − σ)v − t
components, these two equations can be added and sub- T1 (s, v) � +
(t − 2v)2 +(σ − 1)2 (t − 2v)2 + σ 2
tracted, and with the help of (57) and (58), two new fun- (68)
damental equations emerge: 2σv + t 2(1 − σ)v + t
+ 2+ ,
(t + 2v) +(σ − 1) (t + 2v)2 + σ 2
2
π1/4 1 ∞
− 1 � − ξ R (σ + it) + 􏽚 T1 (s, v)ΥR (2iv)
2Γ(3/4) π 0 (60) − t(t − 2v) +(1 − σ)σ − t(t − 2v) +(1 − σ)σ
T2 (s, v) � +
+ T2 (s, v) − 4􏼁ΥI (2iv)􏼁v dv, (t − 2v)2 + σ 2 (t − 2v)2 +(1 − σ)2
− t(t + 2v) +(1 − σ)σ
+
1 ∞ (t + 2v)2 + σ 2
ξ I (s) � 􏽚 T3 (s, v)ΥR (2iv) + T4 (s, v)ΥI (2iv)􏼁v dv,
π 0 − t(t + 2v) +(1 − σ)σ
+ + 4,
(61) (t + 2v)2 +(1 − σ)2
(69)
where
− t(t − 2v) − σ(σ − 1) t(t − 2v) + σ(σ − 1)
T1 (s, v) ≡ − MI (s, v) + MI (s, − v), (62) T3 (s, v) � +
(t − 2v)2 +(σ − 1)2 (t − 2v)2 + σ 2
t(t + 2v) + σ(σ − 1) − t(t + 2v) − σ(σ − 1)
T2 (s, v) ≡ − MR (s, v) − MR (s, − v) + 4, (63) + + ,
(t + 2v)2 +(σ − 1)2 (t + 2v)2 + σ 2
(70)
T3 (s, v) ≡ MR (s, v) − MR (s, − v), (64)
− 2v(σ − 1) − t − 2σv + t
T4 (s, v) ≡ − MI (s, v) − MI (s, − v). (65) T4 (s, v) � +
(t − 2v)2 + σ 2 (t − 2v)2 +(σ − 1)2
(71)
2σv + t 2v(σ − 1) − t
In this manner, because ξ(s) is self-conjugate, the real + + .
(t + 2v)2 +(σ − 1)2 (t + 2v)2 + σ 2
and imaginary components of ξ(s) are isolated,
and expressed in terms of convergent integrals. Further-
more, subtracting (D.2) from (60) yields a simpler form of
(60): Remark. Although not evident when written in the
decomposed form, when expressed in factored form, both
1 ∞ (70) and (71) contain an overall factor (σ − 1/2), a re-
ξ R (s) � 􏽚 T1 (s, v)ΥR (2iv) + T2 (s, v)ΥI (2iv)􏼁v dv. quirement that ξ I (s) vanishes when σ � 1/2. Furthermore, it
π 0
(66) is easily seen that T1,2 (σ + it, v) � T1,2 (1 − σ + it, v),
T3,4 (σ + it, v) � − T3,4 (1 − σ + it, v), T1,2 (σ + it, v) �
T1,2 (σ − it, v), and T3,4 (σ + it, v) � − T3,4 (σ − it, v).
For the case t � 0, 0 < σ < 1, (66) reduces to a relation
As written, the first two terms of each of the above ex-
between ξ(σ) on a section of the real line, and Υ(2iv) on the
plicitly demonstrate the existence of a pole in the complex
complex line R(v) � 0, specifically
v-plane at v � ( ± t/2 ± iσ), along with invariance under the
1 ∞ 1 2
symmetry σ ⟶ 1 − σ. Since we are in general interested in
ξ R (σ) � 􏽚 􏼐4v (3σ(σ − 1) the case t ⋙ {σ, 1 − σ } > 0, such poles lie relatively close to the
π 0 (σ + 4v )􏼐(σ − 1)2 + 4v2 􏼑
2 2
positive real v-axis as demonstrated by the nature of the
peak(s) in Figure 14 (see Appendix A). By reducing the in-
+ 4v2 + 1􏼑ΥR (2iv) + v􏼐64v4 + 16􏼐σ 2 − σ + 1􏼑v2 tegral to the range [0 · · · ∞), the influence of poles corre-
sponding to negative values of v has been effectively removed
from the integrand (v > 0). Drawing on these observations, it
− 2σ(σ − 1)ΥI (2iv)􏼁v dv.
is suggestive that the functions T1,2,3,4 (s, v) can be separated
(67) into two components—pole terms associated with the first
two terms of each, plus background terms associated with all
four terms, in each of the above; it is further suggestive that
only the pole terms will contribute to the integrals (see
6.2. The Background Terms. Although the functions Figures 16 and 18), thereby allowing an approximate solution
T1,2,3,4 (s, v) look formidably complicated when written in to be obtained. How can this be done?
full, each can be written in a more transparent form when Consider (66). From (68) and (69), it would appear su-
decomposed using partial fractions. Specifically, with perficially that both T1 (s, v) and T2 (s, v) vanish as 1/t for
c � − 1, large values of t, which immediately demonstrates a potential
inconsistency—the left-hand side of (66) vanishes exponen- 1 2σv − 4v

tially and so must the right-hand side. Since the remaining T1,4 (s, v) ⟶ +
t t2
integrand factors ΥR (2iv) and ΥI (2iv) contain no t depen- (77)
− 4(2σ − 3)v2 − (σ − 1)2
dence, the only possibility is that 1/t dependence must vanish + + O􏼐t− 4 􏼑.
from the (highly oscillatory) integral, and somewhere buried t3
in the higher order asymptotic terms will be found a term with Clearly, the leading asymptotic dependence (1/t) cancels
exponentially decreasing t dependence. Furthermore, the between all of “pole” and “background” terms in pairs,
possibility also exists that cancellations will occur between the demonstrating that the pole terms also contribute signifi-
two integrand terms containing T1 (s, v) and T2 (s, v) in (66). cantly to the asymptotic background. This verifies that a
Any attempt at a numerical evaluation of (66) for reasonably numerical evaluation of the T1 (s, v) integral will be chal-
large values of t will immediately demonstrate the truth of this lenging as t increases. More importantly, it is apparent that
prediction in the form of a severe cancellation of significant the overall asymptotic dependence of the T1 (s, v) part of the
digits. An example follows. integral will have the leading dependence 1/t2 because the
Choose σ � 1/3 and t � 50, chosen sufficiently large to multiplicative factor ΥR (2iv) lacks t dependence as noted.
distinguish inverse powers of t from exponentially de- But, as also noted, this is inconsistent with (66), where the
creasing terms of the form exp(− bt), but not so large that left-hand side has an exponentially decreasing asymptotic
multidigit computer arithmetic must be used. For this case, behaviour in t. One explanation for this inconsistency could
calculated with 15 digits, we find be that there exists a cancellation between the two terms in
1 ∞ 1 the integrand labelled by T1 and T2 to at least order 1/t2 . This
􏽚 vT1 􏼒 + 50i, v􏼓ΥR (2iv)dv would explain the cancellation observed in the numerical
π 0 3 (72) results (72) and (73). The following gives the asymptotic
� 0.000810442386190651, expansion of each of the corresponding elements of T2 (s, v)
in analogy to (74)–(77):
1 ∞ 1 v − 4v2 + σ
􏽚 vT2 􏼒 + 50i, v􏼓ΥI (2iv)dv T2,1 (s, v) ⟶ − 1 − 2 +
π 0 3 (73) t t2
� − 0.0008104423861872670, (78)
− 8v3 + 2σ(σ + 2)v
and the sum 3.3840 × 10− 15 is comparable with + + O􏼐t− 4 􏼑,
t3
ξ(1/3 + 50i) � 3.38361 × 10− 15 . Clearly, each of the inte-
grals has an absolute value much greater than the sum of v − 4v2 − σ + 1
the two, and it is only because of a cancellation of digits that T2,2 (s, v) ⟶ − 1 − 2 +
t t2 (79)
the final result can be found. Here, we see that the can- − 8v3 + 2σ 2 − 8σ + 6􏼁v
cellation of digits has resulted in the loss of 10 digits from + + O􏼐t− 4 􏼑,
t3
the sum; for larger values of t, the effect will be more
significant, and one might despair of utilizing (66) for v − 4v2 + σ
anything. It will now be shown that these (and other) T2,3 (s, v) ⟶ − 1 + 2 +
t t2 (80)
cancellations can be analysed analytically, and an expo-
8v3 − 2σ(σ + 2)v
nentially decreasing term can be extracted, which is the + + O􏼐t− 4 􏼑,
basis for the approximate model solution. t3
First of all, for v ≤ t/2, consider the asymptotic expansion
(t ⟶ ∞) of each of the individual terms composing T1 v − 4v2 − σ + 1
T2,4 (s, v) ⟶ − 1 + 2 +
labelled, respectively, by a second subscript: t t2 (81)
3 2
1 2σv − 4v 8v + − 2σ + 8σ − 6􏼁v
T1,1 (s, v) ⟶ − + + + O􏼐t− 4 􏼑.
t t2 t3
(74) Again, we see a cancellation between each of the four
4(2σ − 3)v2 +(σ − 1)2 − 4 terms to order 1/t as well as a cancellation of the terms of
+ + O􏼐t 􏼑,
t3 order t0 with the fifth term in (69). And, again we find an
inconsistency with (66) because the leading term of T2 (s, v)
1 2(1 − σ)v − 4v
T1,2 (s, v) ⟶ − + is also of order 1/t2 . Define each term corresponding to t− k
t t2 in the asymptotic series of T1 (s, v) by
(75)
− 4(2σ + 1)v + σ 2
2
+ + O􏼐t− 4 􏼑, T􏽥1 (k) ≡ coefficient of t− 2k in the series lim T1 (s, v),
t3 t⟶∞
v<t
1 2(1 − σ)v − 4v (82)

T1,3 (s, v) ⟶ +
t t2 and similarly for T2 (s, v). When the first two asymptotic
(76)
4(2σ + 1)v2 − σ 2 terms of the factors in the integrand on the right-hand side
+ + O􏼐t− 4 􏼑,
t3 of (66) are written in full, we find
12v 4v 9σ 2 − 20v2 − 9σ + 4􏼁
T1 (s, v) ∼ − + + O􏼐t− 6 􏼑, (83)
t2 t4
− 16v2 + 2 − 64v4 + 48σ 2 − 48σ + 48􏼁v2 − 6σ 2 + 6σ − 2

T2 (s, v) ∼ + + O􏼐t− 6 􏼑 , (84)
t2 t4
∞ √�
⎝ 2 􏽘 e− n2 x π 1⎠
so that π⎛ − √� + ⎞
n�1 x 2
􏽥 1 (1) � − 12v,
T (85)
t2 ∞
(91)
� 􏽚 􏼐πiv x− iv
+ π− iv xiv 􏼑ΥR (2iv)dv
2 2 0
􏽥 1 (2) � 4v 9σ − 20v − 9σ + 4􏼁,
T (86) ∞
t 4
+ i 􏽚 􏼐πiv x− iv
− π− iv xiv 􏼑ΥI (2iv)dv,
0
2
􏽥 2 (1) � − 16v + 2,
T (87) if c � 0, after converting the integration limits to (0, ∞) and
t2 writing Υ(2iv) as the sum of its real and imaginary parts. By an
4 2 2 2 increasing sequence of higher order derivatives, evaluated at
􏽥 2 (2) � − 64v + 48σ − 48σ + 48􏼁v − 6σ + 6σ − 2,
T x � π, it is possible to obtain relevant sums using identities
t4 given by Romik [13]. The first seven such sums are listed in
(88) Appendix D, from which it is possible to obtain the first seven
and clearly there can be no cancellation of terms of equal even moments of ΥR (2iv) and odd moments of ΥI (2iv), listed
order in 1/t2 , or 1/t4 between terms containing T1 (s, v) and also in Appendix D. Substituting the moments (D.2)–(D.4)
T2 (s, v). The only remaining possibility is that the integral into (89) verifies that the prediction (89) is true. QED.
containing each of these asymptotic forms itself vanishes Employing similar logic, a further prediction arises re-
identically, at least to the orders discussed above. In fact, this garding the next term of order 1/t4 . It too must vanish for all
σ. From (83), (84) and (89) we predict
must be true to all asymptotic orders in t− 2k , k ≥ 0 because it
is impossible for a series of the form 􏽐∞ k�1 ak t
− 2k
to add up to 1 ∞ 4 v 96v4 − 40v2 − 1􏼁ΥI (2iv)
􏽚 􏼠 ΥR (2iv)v + 􏼡dv � 0.
an exponential of the form exp(− bt). Furthermore, if the t4 0 120
integrals vanish as predicted, they must vanish for all σ.
(92)
Otherwise, (66), which is an exact result cannot be as-
ymptotically and universally true. And somewhere, ac- Note that σ dependence has disappeared from (92), as
companying all these dominant terms that vanish expected. By substituting (D.2)–(D.6) into (92) the truth of
asymptotically to inverse polynomial order, exist other terms this prediction can be verified. This sequence of forecasts can
associated with the pole terms, that carry the exponentially be continued to the next recursive level, producing the
vanishing result corresponding to the left hand-side of (66) forecast identity:
in the asymptotic limit t ⟶ ∞. Numerically, these lesser 1 ∞
terms equate to the sum of (72) and (73). It will now be 􏽚 􏼠224ΥR (2iv)v6 − 􏼠− 128v7 + 112v5
t6 0
proven that the integrals do in fact vanish, as predicted, to at (93)
least the first few orders in t− 2k , k � 1, 2, 3, following which a 28v3 v
model will be constructed to extract the terms that instead + + 􏼡ΥI (2iv)􏼡dv � 0.
vanish exponentially. 3 3
Based on the first terms of (83) and (84), it is predicted
that the sum of the integrals involving t− 2 must vanish, that is, Again, σ dependence has cancelled of its own volition.
The truth of the prediction (93) can similarly be verified by
1 ∞
􏽚 − 12v2 ΥR (2iv) + v􏼐− 16v2 + 2􏼑ΥI (2iv)dv � 0. (89) substituting the set of moments (D.2)–(D.8) into (93).
t2 0 To summarize, we have the following proposition.
Proof. Consider the classical result ([7], equation (2.15.6)) Proposition 1 (deduced and partially demonstrated). The
relating the inverse Mellin transform of Υ(c + iv) to the above sequence of predictions can be proven in general for all
Jacobi Θ3 function: terms on the right-hand side of (66) that asymptotically
∞
vanish to inverse polynomial order t− 2k , k > 3 independently
∞
􏽚 Υ(c + iv)dv � 4π 􏽘 e− n2 x
� 2π Θ3 (0, exp(− πx)) − 1􏼁, of σ.
−∞ n�1
(90) Whatever remains must then be of exponentially de-
creasing order and must therefore equate to the left-hand side
valid for c > 1. We are interested in the case c � 0, so shift the of (66), and the challenge is to somehow isolate such term(s).
contour by subtracting the residue at v � (1 − c)/i and half The cancellation of integrals of asymptotically vanishing in-
the residue at v � ic resulting in the identity verse polynomial order explains the loss of significant digits
observed between (72) and (73) and suggests that such loss of Remark. Because absolute values have been used, the cor-
significant digits will only increase for larger values of t, as the responding Figures for the function T2 (s, v) are qualitatively
interested reader can verify for herself. identical to Figure 3.
Expressed another way, the crucial point is that although the To understand how all this affects the integral itself, consider
peaks shown in Figure 14 originate from the pole terms, these Figure 4 which shows both integrands J1 (s, v) and J2 (s, v),
pole terms also contribute to the background terms of inverse again using s � 1/3 + it with t � 104 , in close proximity to the
polynomial order (t− 2k , k > 1) that cancel. Nonetheless, the peak region v � t/2. Since a logarithmic scale is used, in both
signal associated with the pole terms naturally separates from the figures, the alternating function(s) J1|2 (s, v) are shown (dotted)
background. The separation of signal from background is the colour coded, positive (red), and negative (blue). The green
essence of the approximate solution to be presented here. (dash-dot) line shows the leading asymptotic behaviour of
To gain further insight into these issues, consider J1|2 (s, v) originating from the factor Γ(iv) embedded in the
Figure 2(a) which shows (the absolute value of ) the factor Υ(2iv), scaled to fit the Figure. This is the background.
integrand: Focusing on Figure 4(a), the peak of the factor T1 (s, v) is
J1 (s � σ + it, v) ≡ T1 (s, v)ΥR (2iv), (94) indicated by a magenta (dashed) line, also marked by the
vertical arrow. The effect of the peak of T1 (s, v) on the in-
as a function of v for several values of t at σ � 1/3. Similarly, tegrand J1 (s, v) can be seen as an enhancement of the inte-
Figure 2(b) shows the same result for the second term in the grand in the immediate vicinity of the peak. This is indicated by
integrand defined by the superimposed solid curve representing J1 (s, v), again
colour coded red and blue, but only for those values of
J2 (s � σ + it, v) ≡ T2 (s, v)ΥI (2iv). (95) t/2 − 2 < v < t/2 + 2. The top horizontal dashed line equates to
Three properties are immediately evident: most of the the ordinate of the point J1 (s, t/2) and the lower line delineates
absolute value of both integrands occurs near v � 0, both points lying one order of magnitude smaller that the ordinates
|J1 (s, v)| and |J2 (s, v)| decrease exponentially with increasing v at the peak. Figure 4(b) is similar to Figure 4(a), with the
and are of similar magnitude—a necessity if they are to cancel exception that instead of a peak at v � t/2, the curve repre-
numerically (see (72) and (73)). A fourth significant property is senting J2 (s, v) vanishes (see Appendix C).
that the pole term, embedded in T1 (s, v) at v � t/2 is clearly
visible above the exponentially decreasing background in each
curve of Figure 2(a), although it lies many orders of magnitude 6.4. The Approximate Solution. Figures 4(a) and 4(b) hold
below the numerically significant portion of the integrand. In the key to the approximate solution of (66). Up to this point,
fact, it appears to be enhanced as t increases, as might be ex- all analysis has consisted of observation, (generalized) de-
pected because the magnitude of its peak varies as t—see (C.1)— duction, and a small amount of proof. The crucial point to be
whereas the background varies as t− 2k , k > 1. However, in extracted from Figure 4(a) is that the peak of T1 (s, v)
Figure 2(b) the zero of T2 (s, v) is not readily apparent because “extrudes” the integrand of J1 (s, v) above background in a
of the resolution of each curve; moreover, each of the curves in small region about v � t/2. Although the background
Figure 2(b) is consistent with what appears to be the back- (dotted part of the curve) represents the vast majority of the
ground as discussed previously. It will now be shown that these magnitude of the integrand over many orders of magnitude,
observations can be used to extract a signal from the back- when this background is decomposed into terms that are of
ground, and in so doing obtain an approximate solution to (66). inverse polynomial order (asymptotically in t), the inte-
gration of these terms cancels between the two terms labelled
by T1 (s, v) and T2 (s, v) and illustrated in Figures 4(a) and
6.3. The Signal. Consider Figure 3(a) which illustrates the peak 4(b) (see Section 6.2), and only that part of the curve near
of the function |T1 (s, v)| over a wide range of v using v � t/2 (shown solid) contributes to the integral. Conversely,
s � 1/3 + 104 i. Also shown are the cumulative background with reference to Figure 4(b), the strong zero of the function
terms 􏽐N 􏽥 T2 (s, v) acts to depress the function J2 (s, v) near v � t/2,
k�1 |T1 (k)|. It can be seen that as N increases, the sum
of the background terms give a better and better approximation although the background terms still assume their role of
to the background part of the function |T1 (s, v)| when v < t/2. acting to cancel similar background terms shown in
In fact, when the each of the cumulative background terms are Figure 4(a). All these suggest an approximation that should
subtracted from |T1 (s, v)|, that part of the function associated give a reasonable representation of a solution to (66).
with small values of v decreases (Figure 3(b)) as the number of
background terms increases. But it is precisely small values of v
that contribute most to the integral itself (see Figure 2). And as 6.4.1. Approximation 1. Let
we have seen from (89), (92) and (93), when coupled with the o
full part of the integrand associated with T2 (s, v) the integral 􏽥 j (s, v) + Tj (s, v),
Tj (s, v) � T (96)
associated with each of these background terms has either been o
where T 􏽥 j (s, v) represents the background, Tj (s, v) repre-
proven, or predicted, to vanish. What is left? Only the pole term
remains, wherein resides the sought after “signal.” sents the nonbackground associated with the pole and
j � 1, 2. From (82), we have
1042 1042
1021 1021
|J1 (s, v)|
|J2 (s, v)|

100 100
10–21 10–21
10–42 10–42
0 20 40 60 80 100 120 0 20 40 60 80 100 120

v v
t = 12.5 t = 100 t = 12.5 t = 100
t = 25 t = 200 t = 25 t = 200
t = 50 t = 50
(a) (b)
Figure 2: The individual terms composing the integrand of (66) as a function of the integration variable v, using σ � 1/3 for several values of
t. Each curve has been scaled by a (smooth) factor exp(πt/4)/t2 to separate them for easier viewing; otherwise, all of the curves coincide to a
good degree of approximation, except near v ≈ 0. (a) An illustration of the integrand |J1 (1/3 + it, v)| at several values of
t � {12.5, 25, 50, 100, 200}; the arrows mark the points v � t/2 that being the location of the peak of the pole term contained in T1 (s, v) for
each of the curves at each individual value of t. (b) An illustration of the integrand J2 (1/3 + it, v) at several values of
t � {12.5, 25, 50, 100, 200}; the arrows mark the points v � t/2 that being the location of the zero of the pole term contained in T2 (s, v) for
each of the curves at each individual value of t.
105 106
104 105
|T1(1/3 + 104 i, v)|
i, v)|
103 104
104
102 103
|T1(1/3 +
101 102
100 101
10–1 100
4000 5000 6000 7000 4500 5000 5500 6000
v v
4 |T~ 4 ~
|T1| k=1 1 (k)| |T1| |T1| – k=1|T |(k)
2 |T~ 5 |T~ 2 ~ 5 ~
k=1 1 (k)| k=1 1 (k)|
|T1| – k=1|T |(k)
|T1| – k=1|T |(k)
~ 3 ~
3 |T |T1| –
k=1 1 (k)| k=1|T |(k)
(a) (b)
Figure 3: The function |T1 (1/3 + 104 i, v)| in comparison with the first few cumulative background terms 􏽐N 􏽥
k�1 |T1 (k)| for wide range of v
near t/2. In both figures, the vertical arrow marks the location of the peak at v � t/2. (a) An illustration of the function |T1 ((1/3) + 104 i, v)| as
well as the cumulative first few background terms 􏽐N 􏽥
k�1 |T1 (k)| for the first 5 values of N for a wide range of v about t/2. (b) An illustration of
the function |T1 ((1/3) + 104 i, v)| as well as the effect of subtracting the first few cumulative background terms 􏽐N 􏽥
k�1 |T1 (k)| for the first 5
values of N for a wide range of v about t/2.
s = (1/3) + 10000 i s = (1/3) + 10000 i

1010 1010
109 109
108
108
107
107
106
105 106
104 105
103 104
2
10 103
10
102
4992 4994 4996 4998 5000 5002 5004 5006 5008 4992 4994 4996 4998 5000 5002 5004 5006 5008
v v
0 < J1(s, v) –T1(s, v)
0 < J2(s, v) e–(1/2)πv/ v
J1(s, v) < 0 e–(1/2)πv/ v J2(s, v) < 0 0 < J2(s, v)
–T1(s, v) J2(s, v) < 0
(a) (b)
Figure 4: J1 (s, v) and J2 (s, v) in close proximity to the peak of T1 (s, v) or zero of T2 (s, v), respectively, at v � t/2, as well as the shape of the
background of both. (a) An illustration of the function J1 ((1/3) + 104 i, v) accompanied by T1 (1/3 + 104 i, v) and the scaled asymptotic shape
of R(Υ1 (1/3 + 104 i, v)) very close to v � t/2. (b) An illustration of the function J2 ((1/3) + 104 i, v) accompanied by T2 (1/3 + 104 i, v) and the
scaled asymptotic shape of R(Υ1 (1/3 + 104 i, v)) very close to v � t/2.
∞
􏽥 j (s, v) � 􏽘 T
􏽥 j (k). 1 t/2+δ1 (σ,t) o
T (97) ξ R (s) � 􏽚 T (s, v)ΥR (s, v)v dv
k�1
π t/2− δ1 (σ,t) 1
(101)
1 t/2+δ2 (σ,t) o
When applied to (66), this leads to + 􏽚 T (s, v)ΥI (s, v)v dv,
π t/2− δ2 (σ,t) 2
o
􏽥 1 (s, v) + T1 (s, v)􏼓ΥR (s, v)
J(s, v) ≡ J1 (s, v) + J2 (s, v) � 􏼒T where δj (σ, t) is ao (small) function of O(σ 0 ), O(t0 ) asso-
ciated with each of Tj (s, v)—see (C.4); that is, only that part of
o
􏽥 2 (s, v) + T2 (s, v)􏼓ΥI (s, v) the integrand in close proximity to v � t/2 contributes to the
+ 􏼒T overall value of the integral, which integral is approximately
symmetric about the point v � t/2. Everything else cancels.
� 􏼐T 􏽥 2 (s, v)ΥI (s, v)􏼑
􏽥 1 (s, v)ΥR (s, v) + T Equivalently, this recognizes that only that section of the curve
coloured solid in Figures 4(a) and 4(b) will contribute to the
o o
+ 􏼒T1 (s, v)ΥR (s, v) + T2 (s, v)ΥI (s, v)􏼓. integral, notwithstanding the fact that the dotted sections
ostensibly contribute to the integrand by many more orders of
(98) magnitude. This is demonstrated numerically below (see
(105)).
But, according to Section 6.2 (i.e., Proposition 1),
∞
Remark. For clarity, I write δj (σ, t) ≡ δj , recognizing that
􏽥 1 (s, v)ΥR (s, v) + T
􏽚 􏼐T 􏽥 2 (s, v)ΥI (s, v)􏼑vdv � 0; (99) δj (σ, t) is a simple and low-order function of both σ and t;
0 the functional dependence is investigated later.
thus, (66) becomes
1 ∞ o 1 ∞ o 6.4.3. Approximation 3. Since each of the δj are much less
ξ R (s) � 􏽚 T1 (s, v)ΥR (s, v)v dv + 􏽚 T2 (s, v)ΥI (s, v)v dv. than any value of t in any range of interest, first rewrite (101)
π 0 π 0
as
(100)
δ 1 o t
o ξ R (s) � 1 􏽚 T1 􏼒s, + δ1 y􏼓ΥR i 2δ1 y + t􏼁􏼁 2δ1 y + t􏼁dy
Since Tj (s, v) represents the nonbackground (pole) 2π − 1 2
terms this leads to Approximation 2.
δ2 1 o t
+ 􏽚 T2 􏼒s, + δ2 y􏼓ΥI i 2δ2 y + t􏼁􏼁 2δ2 y + t􏼁dy.
2π − 1 2
6.4.2. Approximation 2. Let (102)
By applying an obvious change of variables, expand (102) integration (see Figure 18(a)). Note that the second term
about δj � 0, and consider only those terms of leading as- associated with δ2 is numerically smaller than the term
ymptotic order in t, giving the following to first order in δj associated with δ1 as predicted. Significantly, the order of
and asymptotic t: magnitude of (105) is correct, demonstrating that the signal
t2 δ1 ΥR (it) tδ2 σ 2 − σ − 1􏼁ΥI (it) has been extracted, although the numerical estimate is not as
ξ R (s) ≈ + . (103) good as the approximate one because the functional de-
σ(σ − 1)π σ(σ − 1)π
pendence of δ1,2 has not been generally selected for this
Notice that the second term in (103) is smaller than
o the purpose. In fact, resetting the limits of (105) using δ1 � δ2 �
first term by one power of t, reflecting the fact that T2 (s, v) 0.2983 yields the correct result to four significant figures. Of
vanishes at v � t/2. This means that asymptotically, we ex- course, there is no reason to expect that δ2 � δ1 ; this be-
pect the second term in (103) to vanish, relative to the first comes less of an issue as t increases since the relative
term, except if ΥR (it) ≈ 0. o
contribution from the T2 (s, v) term asymptotically vanishes
Remark. By expanding about δj � 0, we are selecting for except at a zero.
values of the integrand at v � t/2. This could be interpreted as
the introduction of a parameter w1 to model the width of 6.4.4. Approximation 4. To obtain an asymptotic approxi-
o o
T1 (s, v) near its peak—effectively replacing T1 (s, v) by mation of ζ R (σ + it) as t ⟶ ∞, first write the left-hand side
inserting a box of equal area around the peak and thereby of (103) in polar form and employ (7), to obtain
√��
approximating parts of the integrand near v � t/2 which ξ R (s) ⟶ − e− πt/4 π− σ/2 |ζ(σ + it)| 2π(t/2)σ/2− 1/2
o
dominate T1 (s, v) and thereby, the integral itself. According (106)
to the Mean Value Theorem for integrals, this can always be 1 2 (cos ϕ)
· 􏼠􏼒σ − 􏼓t sin(ϕ) + t 􏼡.
done and w1 ≠ 0. The parameter w1 is generalized by δ1 which 2 2
o
captures the area of the product term T1 (s, v)ΥR (iv) and Now, consider the right-hand side of (103) by identifying
reasonable models for this parameter should be considered in ΥR (it) in terms of its components (see (3)), apply the same
future studies, recognizing that it is a (weak) function of both approximations employed above, reunite both sides of the
o
σ and t (but see Section 8). Similarly, since T2 (s, v) vanishes equation, cancelling common nonzero factors on both sides
antisymmetrically about v � t/2, it is reasonable to expect that and eventually arrive at
it will integrate somewhat close to zero. Thus, we expect that 1 t
terms involving δ2 will be much smaller than those involving |ζ(s)|􏼒􏼒σ − 􏼓sin(ϕ) + cos(ϕ)􏼓
2 2
δ1 which appears to be the case for large values of t. Here, let
􏽰��
δ1 � σ(1 − σ)/2 based on numerical experimentation with t cos(Φ)δ1 + σ 2 − σ − 1􏼁sin(Φ)δ2 /t􏼁 t − σ/2
the above approximation, and later, δ4 � δ1 . Since the relative ≈ |ζ(it)|􏼒 􏼓 .
o
πσ(1 − σ) 2π
width of the peak of T1 (s, v) decreases for large values of t, (107)
this approximation should improve as t increases (see Starting from (61), a similar set of approximations can be
Figure 16). applied. As in the model for ξ R (s) and with reference to
With reference to the example presented in (72) and Figure 18, the background terms obtained from the de-
(73), the two terms on
√�the right-hand side of (103) with δ1 � composition of functions T3 (s, v) and T4 (s, v) integrate to
δ2 � 0.2357 (� 1/(3 2 )) give zero exactly as in Section 6.2 with the change that each term
1 contains an additional overall multiplicative factor of
ξ R 􏼒 + 50i􏼓 ≈ 3.63105 × 10− 15 − 1.01201 × 10− 16
3 (104) (2σ − 1)/t. Again, the dominant contribution to ξ I (s) arises
from the pole term associated with T4 (s, v). Thus, the
� 3.52985 × 10− 15 , equivalent of (103) is
compared with the correct result ξ R (1/3+
50i) � 3.3836 × 10− 15 , a reasonable approximation. Further- 2(2σ − 1)t2 δ4 ΥI (it) + δ3 /tΥR (it)􏼁
ξ I (s) ≈ , (108)
more, setting δ1 (1/3, 50) � δ2 (1/3, 50) � 0.2357 in (101) gives πσ(1 − σ)
1 25.2357 o and the equivalent of (106) is
􏽚 T (s, v)ΥR (s, v)v dv
π 24.76429 1
πt/4 − σ/2
√�� t σ/2− 1/2
1 25.2357 o ξ I (s) ⟶ e− π |ζ(σ + it)| 2π􏼒 􏼓
+ 􏽚 T (s, v)ΥI (s, v)v dv 2
π 24.76429 2 (109)
� 2.3675 × 10 − 15
+ 5.0984 × 10 − 16
� 2.8773 × 10 − 15
. 1 t2
· 􏼠􏼒σ − 􏼓t cos(ϕ) − sin(ϕ)􏼡.
2 2
(105)
o Again, because 0 < σ < 1 ≪ t, follow the exact steps
where I have used T1,2 (s, v) � T1,2 (s, v) since the latter is used previously and ultimately arrive at the equivalent of
dominated by the pole term over the narrow range of (107)
1 arg(ζ(σ + it)) is embedded within the definition of ϕ (see

|ζ(s)|􏼒􏼒 − σ 􏼓cos(ϕ) + t sin(ϕ)􏼓 (13)), by noting that cos(arg(ζ(s)))|ζ(s)| � ζ R (s) (and
2
similarly for ζ I (s)). Asymptotically (because both cos(Φ)
4t(1/2 − σ) sin(Φ)δ4 + cos(Φ)δ3 /t􏼁 t − σ/2 and sin(Φ) cannot both vanish simultaneously), this solu-
≈ |ζ(it)|􏼒 􏼓 .
πσ(1 − σ) 2π tion yields a simple representation after applying the next
(110) approximation.
Remark. If σ � 1/2 in (110), sin(ϕ) on the left-hand side 6.4.5. Approximation 5. Let
vanishes (see (9) and Section 6.5). δ2 δ 3
Furthermore, it is possible to extract approximations for � � 0, (111)
t t
ζ R (σ + it) and ζ I (σ + it) by solving (107) and (110) simul-
taneously, using a simple trigonometric expansion because giving
− cos(ω)cos(Φ)δ1 − 2(σ − 1/2)sin(Φ)sin(ω)δ4 􏼁 t − σ/2

ζ R (σ + it) ≈ 2|ζ(it)| 􏼒 􏼓 , (112)
σ(σ − 1)π 2π
− sin(ω)cos(Φ)δ1 + 2(σ − 1/2)cos(ω)sin(Φ)δ4 􏼁 t − σ/2

ζ I (σ + it) ≈ 2|ζ(it)| 􏼒 􏼓 , (113)
σ(σ − 1)π 2π
where consistency, this imposes a constraint on the parameter δ1 ,

σ it t specifically
ω � arg􏼒Γ􏼒 + 􏼓􏼓 − log(π). (114)
2 2 2
δ1 (σ) � δ1 (1 − σ). (116)
Notice that the right-hand sides of both (112) and (113)
do not contain any components of ζ(σ + it), and therefore With this constraint in mind, and consistent with the
the left-hand sides are expressed in terms of independent approximations specified previously, it is possible to obtain
functions (compare with [17]). some insight into a reasonable model for δ1 (σ, t) by solving
(103) using the approximation δ2 /t � 0 (see Approximation
5). Introduce (σ, t) dependence by letting the solution be
6.4.6. Numerical Examples. Figure 5 illustrates both approx- denoted by Δ1 (σ, t) and consider Δ1 (σ, t) over the range
imations (112) and (113) for an arbitrary choice of σ. From (103), 0 ≤ σ ≤ 1, for several choices of t. The result is given in
it is seen that any structure in ξ R (σ + it) must originate from the Figure 7(a), providing evidence that it is likely that a
function Υ(it) on the right-hand side, and this function lacks σ smooth function exists to capture the σ dependence of
dependence. Thus, (103) predicts that the structure of ξ R (σ + it), δ1 (σ, t), although the normalization factor varies consid-
particularly the location of the zeros, must be relatively inde- erably as a function of t. The somewhat arbitrary and simple
pendent of σ since, as was noted in Section 6.1, σ dependence model
only resides in the transfer functions and to a lesser extent any 􏽰��
model developed to approximate δ1,2 as a function of σ and/or t. σ(1 − σ)
δ1 (σ, t) � , (117)
Thus, in (66), if ΥR (iv) happens to vanish at v � t/2, then the 2
peak associated with T1 (t, v) is multiplied by a vanishing factor
and contributions to the integral from both functions T1 (t, v � used throughout this paper is also shown; it is clear that this
t/2) and T2 (t, v � t/2) vanish, independently of σ. So, the model, although a reasonable first approximation, requires
location of the zeros is approximately governed by cos(Φ) � 0 future refinement. A similar result can be found for Δ4 (σ, t),
(see the discussion surrounding (124) in Section 7). This pre- the solution of (108) (see Figure 7(b)). (Remark: to obtain the
diction can be checked (see Figure 6(a)) which demonstrates that normalization of the latter figure, expand the function ξ(σ +
this property appears to be approximately true, for two ran- it) about the point σ � 1/2 and cancel the numerator and
domly chosen, disparate values of σ. denominator factors (σ − 1/2) that appear in the imaginary
From this observation, arises a constraint on the ap- part).
proximation. Consider two instances of (103) with different All this suggests that it is of interest to investigate ξ(σ +
choices σ � σ 1 and σ � σ 2 . The approximation predicts that it) for two different choices of σ. (115) and (117) predict that
􏽱�� 􏽱��
ξ R σ 1 + it􏼁 δ1 σ 1 􏼁σ 2 1 − σ 2 􏼁 σ 1 1 − σ 1 􏼁ξ R σ 1 + it􏼁 ≈ ξ R σ 2 + it􏼁 σ 2 1 − σ 2 􏼁, (118)
� , (115)
ξ R σ 2 + it􏼁 δ1 σ 2 􏼁σ 1 1 − σ 1 􏼁
an example of which is shown in Figure 6(b). Finally, it is of
since the model suggests that δ1 is at least a (weak) interest to consider approximations to |ζ(σ + it)| as given in
function of σ. Now, suppose that σ 2 � 1 − σ 1 , in which case (107) (see Figure 8).
the left-hand side of (115) becomes unity (see (2)); for
3
1
ζR(s)
ζI(s)
2
0
505 510 515 520
t
1
–1
0 –2
505 510 515 520
lhs (σ = 0.7)
t
rhs (σ = 0.7)
lhs (σ = 0.7)
rhs (σ = 0.7)
(a) (b)
􏽰��
Figure 5: The approximations (112) and (113) to ζ R (σ + it) and ζ I (σ + it), respectively, near t � 500 using δ1 � δ4 � σ(1 − σ)/2. (a) A
comparison of the left- and right-hand sides of (112) for σ � 0.7 near t � 500. (b) A comparison of the left- and right-hand sides of (113) for
σ � 0.7 near t � 500.
1.5
0.3
0.2
1
0.1
0
505 510 515 520
δ (σ) ξR(σ + it)
0.5 –0.1 t
ξR(s)
–0.2
0 –0.3
705 710 715 720
t –0.4
–0.5 –0.5
–0.6
σ = 0.05 lhs (σ = 0.12)

σ = 0.53 rhs (σ = 0.63)
(a) (b)
Figure 6: For several values of σ and t, this is a comparison of ξ R (σ 1 + it) and ξ R (σ 2 + it) as they exist (left) and as they are predicted to exist by the
approximation (right) (see (118)). In both cases, the functions have been scaled by a factor exp(πt/4)/t2 . (a) A comparison of ξ R (σ + it) for two
very different values of σ near t � 700. (b) A comparison of the left- and right-hand sides of (118) for two very different values of σ near t � 500.
1 1
0.8 0.8
0.6 0.6
Δ1 (σ, t) Δ4 (σ, t)
1 1
Δ1 ,t Δ4 ,t
2 2
0.4 0.4
0.2 0.2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
σ σ
1 σ (1 – σ) Δ1 (σ, t = 10) 1 σ (1 – σ) Δ4 (σ, t = 10)

–
2 0.25 0.42 2 0.25 0.15
Δ1 (σ, t = 100) Δ1 (σ, t = 500) Δ4 (σ, t = 100) Δ4 (σ, t = 500)
– –
0.35 1.75 0.16 1.18
(a) (b)
Figure 7: A simple model for both δ1 (σ, t) and δ4 (σ, t) exists as a function of σ, although the dependence on t may not be simple (see
Figure 10(a)). (a) The solution function Δ1 (σ, t) for several values of t over a range of σ. Each of the curves has been normalized by Δ1 (1/2, t)
with values shown. (b) The solution function Δ4 (σ, t) for several values of t over a range of σ. Each of the curves has been normalized by
Δ4 (1/2, t) with values shown.
8
6
4
2
0
905 910 915 920
–2
t
–4
–6
–8
–10
lhs (σ = 0.1) lhs (σ = 0.52)

rhs (σ = 0.1) rhs (σ = 0.52)
Figure
􏽰 8: A comparison of the left- and right-hand sides of (107) for two very different values of σ near t � 900. The parameter δ1 �
��
σ(1 − σ)/2 in both cases, and all curves have been normalized by the factor 1/t.
6.5. The Case σ � 1/2. From the real part of ξ(s), (107) is of factor cos(ϕ) � (− 1)k because of (13), where k approximates
particular interest when σ � 1/2. By straight substitution, the number of zeros of ζ(1/2 + it) relative to t � 0. For
using (117) and (8), we obtain further discussion, see Section 8.2.
􏼌􏼌 􏼌􏼌 From the imaginary part of ξ(s), in the case of (108) with
􏼌􏼌 1 􏼌 1 cos(Φ)|ζ(1 − it)|t1/4 23/4
􏼌􏼌ζ 􏼒 + it􏼓􏼌􏼌􏼌cos(ϕ) ≈ 4δ1 􏼒 , t􏼓 . σ � 1/2, both sides vanish, as they must because it is known
􏼌 2 􏼌 2 π5/4
that ξ I (1/2 + it) � 0. This is also true of (110) because in the
(119) limit σ � 1/2, we have ϕ � kπ due to (9) and (13). From (10),
This is consistent with equation (5.1.8) in [7]; for a by straightforward differentiation of (2) applied to (108), we
numerical example, see Figure 9. In (119), note that the obtain
􏼌􏼌 coincide, and, in fact, the two should be proportional to one

ξ I (1/2 + it) z 􏼌
lim � I􏼠 ξ(σ + it) 􏼌􏼌􏼌􏼌 σ�1/2 􏼡 another if σ ≠ 0 and σ ≠ 1 where the model (c � − 1) breaks
σ⟶1/2 (σ − 1/2) zσ
down anyway. The second of these relates ξ I (s) to ΥI (it)
and predicts a further coincidence between the zeros of
(− 1)k |Γ(1/4 +it/2)|
� ξ I (σ + it) and ΥI (it) unless σ � 1/2, in which case ξ I (1/2 +
π1/4 it) is always zero ([1], Section 6.5); ΥI (it) is therefore in-
1 2 1 1 it dependent if σ � 1/2. But, as noted, because |ζ(1 − it)| ≠ 0
· 􏼒􏼒t − 􏼒t + 􏼓I􏼒ψ 􏼒 + 􏼓􏼓􏼓 then no zeros of ΥR (it) and ΥI (it) coincide (see also (8)).
4 4 4 2
Consequently, neither do the zeros of ξ R (σ + it) and
􏼌􏼌 􏼌􏼌
􏼌 1 􏼌 ξ I (σ + it). Furthermore, if ξ(σ + it) � 0 and σ ≠ 1/2, this
· 􏼌􏼌􏼌􏼌ζ 􏼒 + it􏼓􏼌􏼌􏼌􏼌 would require the right-hand sides of (103) and (108) to
2
􏼌􏼌 􏼌􏼌 vanish simultaneously, a contradiction with the known
1 1 􏼌 1 􏼌 properties of |ζ(1 − it)| because neither of the two pa-
+ 􏼒t2 + 􏼓sin(α − β)􏼌􏼌􏼌􏼌ζ ′ 􏼒 + it􏼓􏼌􏼌􏼌􏼌􏼓
2 4 2 rameters δ1 and δ4 vanish.
Therefore, if σ ≠ 1/2, it is impossible for ξ(σ + it) to
4t2 δ4 ΥI (it) + δ3 /tΥR (it)􏼁 vanish because it is overly constrained by the properties of
≈ ,
πσ(1 − σ) ζ(1 − it); if σ � 1/2, one of the constraints vanishes, and
(120) ξ(1/2 + it) � 0 becomes possible. Thus, within the confines
where ψ(s) is the digamma function and α and β are defined of this approximation, the Riemann hypothesis (RH) is true.
in (16) and (17), respectively. Asymptotically this becomes However, none of the above predicts that ζ(1/2 + it) � 0
􏼌 􏼌􏼌 􏼌􏼌 􏼌􏼌 anywhere. That is a separate issue. A comparison between
1 1 it 􏼌􏼌 1 􏼌 􏼌 1 􏼌
(2π)3/4 􏼒 ψ I 􏼒 + 􏼓􏼌􏼌􏼌􏼌ζ 􏼒 + it􏼓􏼌􏼌􏼌􏼌 − sin(α − β)􏼌􏼌􏼌􏼌ζ ′ 􏼒 + it􏼓􏼌􏼌􏼌􏼌􏼓 the right- and left-hand sides of (103) and (108) is given in
2 4 2 2 2 Figure 11, demonstrating that the discussion here closely
1/4 approximates reality.
≈ − 64δ4 |ζ(it)|sin(Φ)t− . It is also worthwhile to consider how (112) and (113)
(121) allow us to reach the same conclusion. Suppose that ζ R (σ +
it) � 0 on the left-hand side of (112) and simultaneously
For this fixed value of σ � 1/2, it is possible to obtain an ζ I (σ + it) � 0 on the left-hand side of (113). Since it is known
estimate of δ4 (1/2, t) by solving (121) for δ4 . Denote the
that |ζ(1 − it)| ≠ 0 (and therefore, |ζ(it)| ≠ 0) it must be that
solution by Δ4 (1/2, t) and consider Figure 10(a) which
each of the terms in parenthesis in those two equations
shows Δ4 (1/2, t) for a range of t. As a first approximation,
vanish simultaneously. That is, after removing all nonzero
choose δ4 (1/2, t) � 0.16 as suggested in the figure. It should
common factors, and noting that the equations are only valid
be noted that the denominator of the solution can vanish,
for 0 < σ < 1 we have
but the numerator does not vanish simultaneously; these √� √�
singular points of the solution are apparent in the figure; 2(− 1 + 2σ)sin(ω)δ4 sin(Φ) 2 cos(Φ)cos(ω)δ1
√�� + � 0,
however, at such points, the parameter δ3 (1/2, t) must be 1− σ 2(1 − σ)
incorporated into the model; this will prevent the singularity √� √�
from occurring. Thus, the approximate solution in 2(− 1 + 2σ)cos(ω)δ4 sin(Φ) 2 sin(ω)δ1 cos(Φ)
√�� − � 0.
Figure 10(a) is valid in between the singular points; the 1− σ 2(1 − σ)
horizontal line Δ4 (1/2, t) ≈ 0.16 used in this example has (122)
been chosen on that basis. Figure 10(b) uses δ4 = .16 and δ3 = 0
to compare the right- and left-hand sides of (120). If σ ≠ 1/2, if cos(Φ) ≠ 0, the only solution to (122) re-
quires that
7. Zeros of ζ(σ + it) cos2 (ω) + sin2 (ω) � 0, (123)
As noted earlier, c � − 1 was chosen in (59) because it would which is obviously impossible. If cos(Φ) � 0, a second formal
lead to an expression involving a master function ζ(1 − it) solution exists; it requires that either sin(Φ) � 0 also or that
(see (8)) whose properties are fairly well known and whose sin(ω) and cos(ω) vanishes simultaneously. Both of these
validity spans the critical strip. In particular, it is known conditions are incompatible. Therefore, ζ(s) ≠ 0 if σ ≠ 1/2.
that |ζ(1 − it)| ≠ 0 (see [18], Theorem (6.1)), although both However, in the case that σ � 1/2, the only solution to (122) is
ζ R (1 − it) and ζ I (1 − it) do have many (noncoincident) cos(Φ(t)) � 0. (124)
zeros, and their location is the subject of some interest (e.g.,
[19]). Furthermore, it is well known that ξ(s) shares the which is entirely possible and consistent with everything else
zeros of ζ(s) and Υ(1 − it) shares the zeros of ζ(1 − it), so that has been deduced so far (see Figure 12).
any properties of the zeros belonging to the first of each Notice that if t � t0 defines a zero on the critical line,
pair, apply to the second (of each pair). the term |ζ(1/2 + it0 )| on the left-hand side of (121)
Therefore, consider (103) and (108) in the company of vanishes, and since it is known ([20], equation (4.29) and
Approximation 5. The first of these relates ξ R (s) to ΥR (it) Figure 3) that at a zero defined by t � t0 , the argument
and predicts that the zeros of these functions should functions
0
905 910 915 920
–1 t
–2
–3
–4
lhs (σ = (1/2))
rhs (σ = (1/2))
􏽰��
Figure 9: A comparison of the left- and right-hand sides of (119) near t � 900. The parameter δ1 � σ(1 − σ)/2 and σ � 1/2.
0.5 60
0.4 50
40
0.3
Δ4 30
0.2
20
0.1 10
0
705 710 715 720 3000 3005 3010 3015 3020
t t
|lhs|
|rhs|
(a) (b)
Figure 10: A test (a) of the solution to (121) over the range 700 ≤ t ≤ 720 applied (b) to that same equation over a completely different range of t.
(a) The curve in this Figure shows the solution Δ4 (1/2, t) of (121) over a range of t. The horizontal line Δ4 (1/2, t) ≈ 0.16 illustrates a first
approximation. (b) This is a comparison of the absolute value of the right and left-hand sides of (120) using Δ4 (1/2, t) � 0.16.
π of δ1 and δ4 . Both these parameters were introduced to

α t0 􏼁 − β t0 􏼁 � ± , (125)
2 model the width of the peaks (i.e., extrusions) in the products
T1 (s, v)ΥR (iv) and T4 (s, v)ΥI (iv), and it has been shown in
then the accuracy of the approximation δ4 (1/2, t0 ) � 0.16
the Appendices that the underlying widths of T1 (s, v) and
can be tested, because (121) becomes
􏼌􏼌 􏼌􏼌 T4 (s, v) never vanish. Therefore, neither do δ1 and δ4 , and so
􏼌􏼌 1 􏼌 􏼌􏼌
􏼌􏼌ζ ′ 􏼒 + it􏼓􏼌􏼌􏼌 � 2.17|ζ(it)|sin(ϕ(it))(2/t)1/4 􏼌􏼌􏼌 . (126) neither affects the discussion regarding the solution of (122)
􏼌 2 􏼌t�t0 t�t0 and the existence of the zeros.
A numerical exploration of (126) is given in Figure 13
where the absolute ratio of the left- and right-hand sides of 8. Caveats
(126) is shown. The vertical lines mark each of the zeros
8.1. Universality
t � t0 for the range of t under consideration, and if (126) is
true, the solid (blue) curve should cross each vertical line 8.1.1. A Correspondent Has Written. Unfortunately the
with the value of unity, indicated by the dotted horizontal fundamental flaw of the paper is the premise that (approx-
line. imately) the zeta function at s � σ + it is related in an ele-
Finally, it needs to be mentioned that the above dis- mentary way to the same function at z � σ 0 + it (i.e., for the
cussion did not consider any detailed functional dependence same t but different σ). This contradicts Voronin’s
1.5 1
1
0.5
0.5
0
805 810 815 820
t
0
805 810 815 820
t –0.5
–0.5
–1
–1
lhs (σ = (1/2)) lhs (σ = 0.22)
rhs rhs
(a) (b)
Figure 11: A comparison of the predicted relationship between ξR (σ + it) (left) and ξ I (σ + it) (right) versus the corresponding real and
near t � 800. The figures are scaled by eπt/4 /t2 . (a) A comparison of the left- and
imaginary components of Υ(it) at two different choices of􏽰σ��
right-hand sides of (103) near t � 800. The parameter
􏽰�� 1δ � σ(1 − σ)/2, σ � 1/2 and δ2 � 0. (b) A comparison of the left- and right-hand sides
of (108) near t � 800. The parameter δ4 � σ(1 − σ)/2, σ � 0.22 and δ3 � 0.
0.5
cos (Ф(t))
0
905 910 915 920
t
–0.5
–1
Figure 12: A test of (124) when σ � 1/2 near t � 900. The vertical lines mark the location of the points t0 , where ζ(1/2 + it0 ) � 0, theorized to
coincide with cos(Φ(t0 )) � 0.
Universality Theorem, which basically says the zeta func- approximates that function to within ϵ for some (probably
tion, sufficiently far up the critical strip, mimics (to within extremely large) fixed t and 1/2 < σ < 1. Therefore, ap-
arbitrary accuracy) all non-vanishing analytic functions in proximations of the form (107) and (119) cannot hold
simply connected compact subsets of the strip 1/2 < σ < 1. uniformly for large t.
So, while the behaviour of ζ(σ + it) may appear to be quite
regular for 1/2 < σ < 1 and fixed t of moderate size (indeed
monotonic for real, imaginary and absolute values), this 8.1.2. The Author Comments. The fundamental result of this
cannot be true as one moves further up the strip. Indeed, paper is a proof that ζ(s1 ) is related to the same function
one can choose an arbitrary “wild” nonvanishing analytic ζ(s2 ) integrated over a contour through the intermediary of
function, and an arbitrary ϵ > 0 , such that the zeta function a specific integral operator. This is a rigorous result, which,
1.5
(lhs/|rhs|)
0.5
0
300 305 310 315 320
t
Figure 13: Ratio of the left and absolute value of the right-hand sides of (126) for general values of t > 0. Vertical lines mark the location of the zeros.
as noted earlier, has been shown ([15], Theorem 1) to be bounded by known limits on |ζ(1 − 2iv)| through
equivalent to the Cauchy Contour integral theorem. In this well-established theorems. Crudely |ζ(1 − 2iv)| ≪
paper, a specific example has been chosen such that s2 covers log2/3 (2v), as well as other more strict limits given
the 1-line through the intermediary of the functional in [16] and [21] (see (131) and (132)).
equation (8). Subject to Proposition 1, the representation
(iv) The transfer functions embedded in (61) and (66)
(101) introduces arbitrary functions δ1,2,3,4 (σ, t), and the
are simple polynomials whose maximum t− value
remainder of the paper investigates the observation that if
anywhere on the contour is limited to O(t1 ), and
these functions are chosen in an elementary manner, a
these functions are the only elements in the inte-
reasonable representation of the function ζ(σ + it) can be
grand(s) that carry t dependence.
found with reference to the function ζ(it) (and, by reflection,
ζ(1 − it)) for 0 < σ < 1. Surprisingly, the numerical examples (v) The only other significant function appearing in the
presented suggest that the connection between the values of integrand of (61) and (66) is the product Γ(iv)π − iv
ζ(σ + it) and ζ(it) are remarkably insensitive to the par- whose smoothness (continuity) and magnitude
ticular model used for δ1,4 (σ, t). This was investigated in a properties are well-known.
superficial fashion (see Figures 7 and 10(a)); a detailed (vi) None of the elements of the integrand in (61) and
analysis is beyond the scope of this paper. Certainly, if at an (66) contain real poles or divergences and the in-
extremely large value of t the zeta function assumes a “wild” tegrals themselves are obviously convergent at both
value, the functions δ1,2,3,4 (σ, t) could be chosen appropri- endpoints.
ately to satisfy whatever wildness arises. However, the
(vii) Consequently, the integrals in (61) and (66) are all
purpose of an approximate solution to a complex problem is
limited in magnitude and constrained in their
to convert chaos into order, not the reverse.
t− dependence and so cannot assume an arbitrarily
Therefore, notice that the correspondent, perhaps in-
“wild” value or approximate an arbitrarily “wild”
advertently, raises an interesting point: in addition to the
function. This is a rigorous result contingent on the
Universality Theorem, (see ([7], Chapter XI)), the function
latter four results cited above, all of which are
ξ(s) (and hence, ζ(s)) must also obey restraints embedded in
rigorous and true both in their own right as well as
the newly derived integral equations (61) and (66). Consider
more general instances corresponding to any c > −
the consequence(s), taking note that the following results are
1 (see Section 4).
all rigorously proven and widely accepted:
All of this suggests that an incompatibility exists among
(i) The Universality Theorem recognizes that some-
one or more of the above-cited rigorous results. Specifically, it
where “further up the strip,” the function ζ(s) will
is not apparent how an arbitrarily “wild” function, whose
assume a “wild” value that approximates an arbi-
magnitude exceeds the known maximum value of the product
trary function within ϵ.
of all the components of the integrands of (61) and (66) in-
(ii) Equations (61) and (66) impose a strict relationship tegrated over the range [0, ∞), could be approximated “within
between ξ(s) and ξ(2iv) on the s � 0 + 2iv line, and ϵ” by ζ(σ + it), which is subject to the integral equation
therefore a relationship exists between ζ(s) and constraint(s) specified in the above. Furthermore, the corre-
ζ(1 − 2iv) through these equations (2) and the spondent’s observation that the approximate solutions (107)
functional equation. and (119) do not satisfy the Universality Theorem suggests that
(iii) The function ζ(1 − 2iv) appearing in (61) and (66) their precursor—the (rigorously proven) integral equation
via the functional equation and its polar-form is itself—does not satisfy the Universality Theorem. In summary,
the reconciliation of the noted apparent incompatibility this point, the functions δ1,2,3,4 (σ, t) have been simply
suggests a number of fruitful subjects for future inves- modelled as a function of σ, although clearly t dependence
tigation—again, beyond the scope of this work. exists, e.g., see Figure 10(a). So, considering that δ1 (1/2, t) is
expected to be a weak function of t as noted previously, it is
reasonable to anticipate that t dependence residing in
8.2. Lindelöf’s Hypothesis
δ1 (σ, t) and to a lesser extent cos(Φ(t)) could affect any (dis)
8.2.1. A Second Correspondent Has Written. “(1) seems to agreement between (127), (129), (133), and Lindelöf’s
indicate that t− 1/4 ζ(1/2 + it) is unbounded as t ⟶ ∞ hypothesis.
which we know is not true; see [22]. In fact, the Lindelöf In contrast to Section 8.1, where an arbitrary “wildness”
hypothesis states that ζ(1/2 + it) � O(tϵ ) as t ⟶ ∞ for all was needed to generate agreement, here it is noted that the
ϵ > 0, and if this is false, so is the Riemann hypothesis.” functions δ1,2 (1/2, t) capture the value of the integrand of
the integral equation only in close proximity to the point
v � t/2 in (105). Thus t-dependence of δ1,3 and δ2,4 is
8.2.2. The Author Comments. According to Edwards ([1], sensitive to the density of the real and imaginary zeros
Section 9.2), Lindelöf’s theorem is based on Lindelöf’s ob- (respectively) of the function ζ(1 − 2iv) near a particular
servation that “there is a constant K such that point on the 1-line. And, as previously noted, such
t− dependence must persist for any c > − 1, in which case it
|ζ(σ + it)| < Kt1/2− σ/2
log t, (127) is useful to recall [19] that R(ζ(σ + it)) > 0 if σ > 1.192...
corresponding to c > − 0.904.... No arbitrariness here. In
throughout the half-strip 0 ≤ R(s) ≤ 1.” Assuming that
any case, it is evident that investigations into these ob-
|ζ(1/2 + it)| asymptotically obeys a power law, Lindelöf
servations might bear fruit. Again, this is beyond the scope
denotes by μ(σ) the least upper bound of the numbers A
of this work.
such that |ζ(σ + it)|t− A is bounded as t ⟶ ∞, shows that
1 1
0 ≤ μ􏼒 􏼓 ≤ , (128) 9. Summary
2 4
and hypothesizes that μ(1/2) � 0. This basically posits that as Remark. throughout this work, it should be considered that
t ⟶ ∞, |ζ(1/2 + it)| increases at least faster than O(t0 ). the figures presented were all obtained from the computer
Several proven refinements of (128) are known, over the program Maple [23] the accuracy of whose numerical
years gradually reducing the upper inequality limit from 1/4 evaluation of functions that do not lie in its mainstream is
to the most recent [22] value 13/84 � 0.1548, yielding the open to question. Spot check comparisons were performed
least upper asymptotic bound between Figures from Maple and Mathematica [24], and no
􏼌􏼌 􏼌􏼌 grounds for concern were found.
􏼌􏼌 1 􏼌
􏼌􏼌ζ 􏼒 + it􏼓􏼌􏼌􏼌 ≪ O􏼐t13/84 􏼑. (129) Although it is known in the literature that ξ(s) sat-
􏼌 2 􏼌
isfies integral equations as cited in Section 1, only the one
By comparison, the equivalent result obtained here (see presented by Patkowski ([4], Theorem (1.1)) comes close
(119)) is to sharing the desirable properties uncovered here. In
􏼌􏼌 􏼌􏼌 particular, because the s− dependence is isolated within
􏼌􏼌 1 􏼌 1 cos(Φ(t))|ζ(1 − it)|t1/4 23/4
􏼌􏼌ζ 􏼒 + it􏼓􏼌􏼌􏼌cos(ϕ) ≈ 4δ1 􏼒 , t􏼓 , the transfer function, it became possible to associate
􏼌 2 􏼌 2 π5/4
properties of ξ(s) anywhere in the critical strip with its
(130) properties on the 1-line, which are relatively well known.
where cos(ϕ) � (− 1) k
counts the zeros and One surprising prediction is that the zeros of ξ(σ + it) and
− 1 ≤ cos(Φ(t)) ≤ 1. For emphasis, Φ has been written fully as those of Υ(1 − it) (and Υ(it)) are closely correlated, as
Φ(t) (see (15)). With ([21], Conjecture A) demonstrated by a number of Figures. Although the
correlation shown in the figures is not exact, the pre-
max |ζ(1 + it)| ∼ exp(c)log log T, (131)
t∈[1,T] diction is based on a model that leaves several questions
unanswered:
or ([16], Theorem 1)
max |ζ(1 + it)| ≥ exp(c)(log log T + log log log T − C), (i) What is a more reasonable model to employ for the
√�
t∈[ T ,T] functions δ1 (σ, t) and δ4 (σ, t)?
(132) (ii) How will the two terms δ2 (σ, t) and δ3 (σ, t) affect
(103) and (108) once they are incorporated into the
both of which are valid for sufficiently large T, (130) becomes model?
􏼌􏼌 􏼌􏼌
􏼌􏼌 1 􏼌 1 cos(Φ(t))t1/4 23/4
􏼌􏼌ζ 􏼒 + it􏼓􏼌􏼌􏼌cos(ϕ) ≈ 4δ1 􏼒 , t􏼓 (iii) What will the numerical estimates look like at values
􏼌 2 􏼌 2 π5/4 of t significantly greater than those explored here,
(133)
which are limited by restrictions on simple com-
· (log log(t) + · · ·). putational arithmetic?
Notice the similarity between (107) (with (8)) and (127) These questions as well as those raised in Section 8,
for general σ, as well as (127) and (133) when σ � 1/2. Up to implore further investigation.
Appendix does not appear to be amenable to direct analysis except

in the case σ � 1/2. To solve the general case σ ≠ 1/2, we
A. Properties of MI (t, v) expect that any extremum of the function MR (σ + it, v)
will approximately coincide with a zero of the derivative
This appendix gives a detailed description of the properties of the denominator, consistent with the presence of a
of MI (c, s, v) in the case c � − 1 and t > 0. For simplicity, c nearby pole in the complex v− plane (see (54)). Equate the
dependence is omitted from the notation. Because the in- derivative of this denominator to zero in the asymptotic
terest is in the properties as a function of t, in this Appendix, limit t ⟶ ∞, solve, and then substitute the result into
MI (t, v) ≡ MI (σ + it, v) and MR (t, v) ≡ MR (σ + it, v), ex- the expression for (z/zv)MR (σ + it, v). With a bit of ex-
cept in those cases where σ dependence is considered. Be- perimentation using the result so-obtained as a guide, the
cause of symmetry properties about the line c � − 5/4 (see location of three extrema can be found, those being to
Figure 1(b)), the cases c � − 1 and c � − 3/2 (see Section 5.1) first order in t− 1 , in order of decreasing v:
are related by the functional equation of ζ(s). With reference
to its definition (31), equating (z/zv)MI (t, v) � 0 and t σ σ(1 − σ)
V1 � + − , (B.1)
solving for v, identifies, as t ⟶ ∞, the fact that MI (t, v) has 2 2 2t
only three extrema, those being
t σ σ(1 − σ)
(i) An extremum near v � 0, consistent with the illus- V2 � − − , (B.2)
2 2 2t
trations in Figures 14 and 15.
(ii) Two maxima, (1/2 − σ)
V3 � , (B.3)
t 4t
v± � ± . (A.1)
2 t (1 − σ) σ(1 − σ)
At the positive maximum, we find V4 � − + + , (B.4)
2 2 2t
t
MI t ⟶ ∞, v+ 􏼁 � , (A.2) t (1 − σ) σ(1 − σ)
σ V5 � − − + . (B.5)
and for v < 0, we find 2 2 2t
t
MI t ⟶ ∞, v− 􏼁 � − . (A.3) Notice that the sets 􏼈V1 , V5 􏼉 and 􏼈V2 , V4 􏼉 are related by the
(1 − σ)
interchange σ ⇔ (1 − σ) together with an overall change of
These are the characteristics of two sharp “pulses”—see sign. The result V3 corresponds to an inflection near the origin.
Figure 14—in which case, it is of further interest to deter- The extreme values at each of these extremum locations
mine the width. For v > 0, solving can be found by substitution into MR (σ + it, v) in the limit
MI (t ⟶ ∞, v) � MI (t ⟶ ∞, v+ )/2 identifies the t ⟶ ∞. For the case v > 0, we obtain
abscissa of v at half-height, that being
t σ t (1/2 − σ) 2σ 2 − 2σ + 1􏼁
vh � ± . (A.4) MR t ⟶ ∞, 􏼈V1 , V2 􏼉􏼁 ≈ ∓ − ∓ ,
2 2 2σ σ 4σt
Thus, the full-width at half-height of the function (B.6)
MI (t ⟶ ∞, v) is σ, independent of t in the asymptotic
limit and thereby mimics, but does not approach, a repre- and for the case, v < 0 the extreme values are
sentation of a Dirac delta function (whose width vanishes). t (1/2 − σ)
Figure 15, which illustrates these estimates in specific nu- MR t ⟶ ∞, 􏼈V4 , V5 􏼉􏼁 ≈ ± +
2(1 − σ) (1 − σ)
merical form, suggests that the function values of any in- (B.7)
tegrand containing MI (e.g., see (59)) are best evaluated at 2σ 2 − 2σ + 1􏼁
v � ± t/2 corresponding to the sign of v. Embedded in this ± ,
4(1 − σ)t
model are the factors (t/σ) and t/(1 − σ) reflecting the
magnitudes of the respective peaks that depend on the sign all again in order, and, as in the previous case, the extrema
of v. Included also is a width factor w± intended to capture are separated by a constant distance Δv � σ or Δv � 1 − σ
the fact that nonzero width exists in the asymptotic limit. according as v > 0 or v < 0. Using the same methods, the two
midpoints can be more accurately obtained; in the as-
B. Properties of MR (σ + it, v) ymptotic limit for v > 0 and v < 0, respectively, they are
The function MR (s, v), slightly more complicated, can be t 3(σ − 1)2
analysed in a similar fashion, again for t > 0. First of all, ig- V+mid � + ,
2 (4t)
noring any extrema, in the asymptotic limit t ⟶ ∞, it is (B.8)
easily shown that MR ≈ 2 if |v| < t/2. An overview of MR (σ + t 3σ 2
it, v) can be seen in Figures 16 and 17. The details follow. V−mid �− − .
2 (4t)
To locate the extrema, a solution of (z/zv)MR (σ + Figure 16 illustrates the above in general. MR (t, v)
it, v) � 0 leads to a sixth order polynomial equation which fundamentally differs from MI (t, v) because, in the
800
600
M1 (σ + it, v)
400
200
–60 –40 –20 0 20 40 60

v
Figure 14: The properties of the transfer function MI (σ + it, v) using σ � 0.1 over a range of v with t � 100.
(a) (b)
Figure 15: The properties of MI (σ + it, v) when t � 100, σ � 0.1 and v < 0 (a) and v > 0 (b). Shown is the value of the extremum (dotted) as
well as the location of one of its half-width intercepts (dashed) on the v− axis to first (asymptotic) order in t− 1 . Note the difference in vertical
scales for the same values of t and σ between these two Figures, as illustrated in Figure 14 where both are shown at the same scale.
asymptotic limit, it vanishes with slope − 2t/(σ − 1)2 near Remark. In the physics lexicon, sharp peaks of the form
either of the midpoints V+mid of the extrema rather than shown in Figure 14 are labelled “Breit-Wigner” or Lor-
rising to an extreme value near that point, although it spans entzian. The shape is a reflection of the presence of a nearby
2t between maximum and minimum ordinates as does pole in the complex v− plane, the closer, the sharper the peak.
MI (t, v). It can also be shown that, to a reasonable degree of Choosing c � − 1 with 0 < σ < 1 guarantees that the peak
approximation, MR (t, v) is nearly symmetric (in the as- shapes will be sharp, optimizing the approximation dis-
ymptotic limit) about the midpoint. cussed in Section 6. As can be seen from Figure 1(a), when
4
109/0.9
2
MR 0
102 104 106 108 1010
–2 v
–4 –108
–6
–8
Figure 16: The properties of the transfer function MR (σ + it, v) using σ � 0.9 over a range of v with t � 109 .
(a) (b)
Figure 17: The properties of the transfer function MR (σ + it, v) when t � 100, σ � 0.1 and v < 0 (a) and v > 0 (b). Shown are values of the
extrema (dotted) as well as the location of their abscissa (dashed) on the v− axis to first (asymptotic) order in 1/t (see (B.1) to (B.5)). Note the
difference in vertical scales for the same value of t and σ between these two Figures.
the two parametrized moving poles coalesce, they create a improved in future; they are listed below, employing to the
pole of one higher order, and this occurs only when following notation:
σ � 1/2, t � 0. The locations of the poles can alternatively be
(i) Vj (min/max) refers to the abscissa of an extremum
read from the denominator parameters of (68) and (69).
projected onto the v− axis.
C. Properties of the Functions T1,2,3,4 (s, v) (ii) Tj gives the magnitude (amplitude) of the
extremum.
The functions T1,2,3,4 (s, v) were defined previously (see (62) (iii) Zj locates any points where the function crosses the
through (65)). Since they are combinations of MR,I they v− axis.
share similar properties which can be visualized by exam-
ining Figure 18. As before, the formal equations defining (iv) Hj locates points corresponding to the half-height
these functions, available from (55), (56), (62)–(65) involve and
6th order rational polynomials which do not lend themselves (v) Wj gives the full-width at half-height, all with
to analytic solution in order to determine the explicit values reference to function Tj (s, v), j � 1, 2, 3, 4. Note
of the extrema and their locations. However, in the limit that in some cases, the minimum and maximum
t ⟶ ∞, it is possible to obtain reasonable approximations reverse or vanish according to whether σ > 1/2,
by experimentation. It is likely that these estimates could be σ < 1/2 or σ � 1/2:
t 3
T1 (Min) � + , (C.1)
σ(σ − 1) 4t
􏽱��
􏽰��
t 1
H1 � ± − 2 − 8σ 2 + 2 4σ(σ − 1)(5σ 2 − 5σ + 2) + 1 + 8σ , (C.2)
2 4
􏽱��
􏽰��
1
W1 � − 2 − 8σ 2 + 2 4σ(σ − 1)(5σ 2 − 5σ + 2) + 1 + 8σ , (C.3)
2
2
Min t 1 1
V2 􏼒 􏼓 � ∓ 􏼠 − 􏼒 − σ 􏼓 􏼡, (C.4)
Max 2 4 2
Min t􏼐− 4σ(1 − σ)(2σ − 1)2 + 2􏼑

T2 􏼒 􏼓�∓ , (C.5)
Max σ(1 − σ)(4σ 2 + 1)􏼐4(1 − σ)2 + 1􏼑
Min 2t(σ − 1/2)

T3 􏼒 􏼓�∓ , (C.6)
Max σ (σ − 2σ + 2)(1 − σ)(σ 2 + 1)
2
Min t σ(1 − σ)
V3 􏼒 􏼓� ∓ , (C.7)
Max 2 2
(σ − 1/2)
T4 (Min) � 2t , (C.8)
σ(1 − σ)
(σ − 1/2)
T4 (Max) � − 2t 􏽰��, (C.9)
1 + 2 σ(1 − σ)
􏽱��
􏽰��
Min t 1
V4 􏼒 􏼓� ∓ σ(1 − σ) + σ(1 − σ), (C.10)
Max 2 2
Min t 1 􏽰��
Z4 􏼒 􏼓� ∓ σ(1 − σ), (C.11)
Max 2 2
􏽲��
􏽱��
t 1 (C.12)
H4 � ± − 2 + 2 4(σ − 1)2 σ 2 + 1,
2 4
􏽲��
􏽱��
1 (C.13)
W4 � − 2 + 2 4(σ − 1)2 σ 2 + 1.
2
There are several important points to note about these (iv) The magnitude estimates given here are all expected
functions: to be accurate to O(t); the locations (and therefore
widths) less so; in Section 6.1, both are used to
(i) To at least first order in t each is more-or-less
obtain an asymptotic estimate of ζ(s) which is
symmetric about the point v � t/2
therefore expected to be reasonably accurate, but
(ii) Other than near v � t/2 the functions otherwise not exact because the widths are only estimated. It is
vanish as O(t− 2 ) or O(t− 3 ), for an overview see the magnitude(s) of the pole terms embedded in
Figures 14 and 16. Therefore, the properties of ξ(s) T1 (s) and T4 (s) that govern the leading order in t;
are determined to a large extent by the properties of since these are exact to leading order in t, so will be
the integrand of (61) and (66) near the point v � t/2, the accuracy of ξ(s) (and hence ζ(s)) to the leading
although other nonvanishing, but ultimately can- order in t. The widths govern the accuracy of the
celling terms obfuscate this result numerically approximation details shown in various figures in
(iii) The function extrema approximate both a unit delta the main text, but do not appear to affect the
function and its derivative in the limit t ⟶ ∞ derivation of the location of the zeros
4000
2000 1000
0
999.5 1000 1,000.5 1001 0
v 999.5 1000 1,000.5 1001
–2000 v
–1000
–4000
–6000 –2000
–8000
–3000
T1 T2 (Min) T3 T4
T1 (Min) T2 (Max) T3 (Min) T4 (Min)
T2 T3 (Max) T4 (Max)
(a) (b)
Figure 18: The nature of the functions T1 (s, v) to T4 (s, v) using σ � 1/3, t � 2000. The extremum values are shown to demonstrate the
accuracy of the estimates (C.1), (C.5), (C.6), (C.8), and (C.9). The location points (C.4), (C.7), and (C.10) are also shown (dashdot). The
arrows correspond to (C.11) and the vertical dotted lines show the location ((C.2) and (C.12)) of the half-height line.
D. Listing of Sums and Moments at x � π. They are needed to obtain the moments of Υ(2iv)
below, for use in Section 6.2. Note: Γ(3/4)n means (Γ(3/4))n .
The following sums were calculated by applying higher order
differentiation (with respect to x) to Romik’s Theorem 1 [13]
∞
πn2 1
􏽘 − n2 e− �− ,
n�1 8π3/4 Γ(3/4)
∞
πn2 3 + π4 /2Γ(3/4)8
􏽘 n4 e − � ,
n�1 32π7/4 Γ(3/4)
∞
πn2 15 15π 5/4
􏽘 − n6 e − �− − ,
n�1 128π11/4 Γ(3/4) 256Γ(3/4)9
∞ √�
πn2 105 105 4 π π17/4
􏽘 n8 e − � + − ,
n�1 512π Γ(3/4) 512Γ(3/4) 2048Γ(3/4)17
15/4 9
∞
πn2 945 1575
􏽘 − n10 e− �− −
n�1 2048π19/4 Γ(3/4) 2048π3/4 Γ(3/4)9
(D.1)
45π13/4
+ ,
8192Γ(3/4)17
∞
πn2 10395 51975
􏽘 n12 e− � +
n�1 8192π23/4 Γ(3/4) 16384π 7/4 Γ(3/4)9
1485π9/4 51π25/4
− + ,
32768Γ(3/4)17 65536Γ(3/4)25
∞
πn2 135135 945945
􏽘 − n14 e− �− −
n�1 32768π27/4 Γ(3/4) 65536π11/4 Γ(3/4)9
45045π5/4 4641π21/4
+ − .
131072Γ(3/4)17 262144Γ(3/4)25
From the above sums, it is possible to obtain the moments

of Υ(2iv) by evaluating the derivatives of (91) at x � π as
follows:
∞ π π1/4
􏽚 vΥI (2iv)dv � − 􏼠 − 1􏼡,
0 4 2Γ(3/4)
(D.2)
∞ π π5/4 π21/4
􏽚 v2 ΥR (2iv)dv � + − ,
0 8 32Γ(3/4) 64Γ(3/4)9
(D.3)
∞ 3π21/4 − 10π5/4 Γ(3/4)8 − 16πΓ(3/4)9

􏽚 v3 ΥI (2iv)dv � ,
0 256Γ(3/4)9
(D.4)
∞ − π37/4 − 76π21/4 Γ(3/4)8 + 68π5/4 Γ(3/4)16 − 64πΓ(3/4)17

􏽚 v4 ΥR (2iv)dv � ,
0 2048Γ(3/4)17
(D.5)
∞ 5π37/4 + 420π21/4 Γ(3/4)8 − 484π5/4 Γ(3/4)16 + 128πΓ(3/4)17

􏽚 v5 ΥI (2iv)dv � ,
0 8192Γ(3/4)17
(D.6)
∞ 1
􏽚 v6 ΥR (2iv)dv � 􏼐− 51π
53/4
− 350π37/4 Γ(3/4)8 − 11644π 21/4 Γ(3/4)16 + 5768π5/4 Γ(3/4)24 + 512πΓ(3/4)25 􏼑,
0 65536Γ(3/4)25
(D.7)
∞ 1 3 8 3 16 3 24 3 25
􏽚 v7 ΥI (2iv)dv � 􏼠 357π 53/4
+ 2590π 37/4
Γ 􏼒 􏼓 + 93492π 21/4
Γ􏼒 􏼓 − 54760π 5/4
Γ􏼒 􏼓 − 1024πΓ 􏼒 􏼓 􏼡.
0 262144Γ(3/4)25 4 4 4 4
(D.8)
Data Availability References

No data were used to support this study. [1] H. M. Edwards, Riemann’s Zeta Function, Dover, IL, USA, 2001.
[2] M. S. Milgram, “Integral and series representations of Rie-
mann’s Zeta function, Dirichlet’s Eta function and a medley of
Conflicts of Interest related results,” Journal of Mathematics, vol. 2013, Article ID
181724, 17 pages, 2013.
The author declares that he has no conflicts of interest. [3] Kačinskaitė, “A note on functional independence of some
Zeta-functions,” Annales Universitatis Scientiarium Buda-
Acknowledgments pestinensis de Rolando Eötvös Nominatae Sectio Computa-
torica, vol. 41, pp. 5–13, 2013.
I am grateful to Larry Glasser, who early on provided me [4] A. E. Patkowski, “A new integral equation and some integrals
with a bespoke derivation of (20); shortly afterwards, he associated with number theory,” 2018, https://arxiv.org/abs/
outdid himself by pointing out Romik’s paper [13], thereby 1407.2983.
negating his invitation to pen a guest Appendix. I also thank [5] K. Sugiyama, “Derivation of the reflection integral equation of
Larry, who identified the value of the Mellin transforms, and the zeta function by the complex analysis,” 2016, http://vixra.
org/abs/1404.0281.
Vini Anghel for commenting on a preliminary version of
[6] A. C. L. Ashton and A. S. Fokas, “Relations among the
this manuscript. All expenses associated with the work Riemann zeta and Hurwitz zeta functions as well as their
presented here have been borne by myself. I am therefore products,” 2018, https://arxiv.org/abs/1809.09440.
grateful to the Board of Director of Geometrics Unlimited, [7] E. C. Titchmarsh and D. R. Heath-Brown, The Theory of the
Ltd. who have agreed to finance any further (i.e., publica- Riemann Zeta-Function, Oxford Science Publications, Ox-
tion) costs. ford, UK, 2nd edition, 1986.
[8] F. W. J. Olver, D. W. Lozier, R. F. Boisvert, and C. W. Clark,

NIST Handbook of Mathematical Functions, Cambridge
University Press, New York, NY, USA, 2010.
[9] M. Milgram, “Exploring Riemann’s functional equation,”
Cogent Mathematics, vol. 3, no. 1, p. 1179246, 2016.
[10] A. Ivic, “A note on the zeros of the zeta-function of Riemann,”
2018, https://arxiv.org/abs/1801.04711.
[11] A. Leclair, “An electrostatic depiction of the validity of the
Riemann Hypothesis and a formula for the N-th zero at large
N,” International Journal of Modern Physics, vol. 28, no. 30,
https://arxiv.org/abs/1305.2613v3, 2013.
[12] M. S. Milgram, “The generalized integro-exponential func-
tion,” Mathematics of Computation, vol. 44, no. 170,
pp. 443–458, 1985.
[13] D. Romik, “The Taylor coefficients of the jacobi theta constant
θ3 .,” 2018, https://arxiv.org/abs/1807.06130.
[14] R. B. Paris, “A generalization of an expansion for the Riemann
Zeta function involving incomplete Gamma functions,” Ap-
plied Mathematical Sciences, vol. 3, no. 60, pp. 2973–2984,
2009.
[15] M. L. Glasser, “A note on the Riemann ξ-function,” 2019,
https://arxiv.org/abs/1901.07011.
[16] C. Aistleitner, K. Mahatab, and M. Munsch, “Extreme values
of the Riemann zeta function on the 1-line,” International
Mathematics Research Notices, vol. 2019, no. 22, pp. 6924–
6932, 2019, https://arxiv.org/abs/1703.08315v2.
[17] A. S. Fokas and J. Lenells, “On the asymptotics to all orders of
the Riemann zeta function and of a two-parameter general-
ization of the Riemann zeta function,” 2015, https://arxiv.org/
abs/1201.2633.
[18] A. Ivić, The Riemann Zeta-Function: Theory and Applications,
Dover, Mineola, NY, USA, 1985.
[19] J. Arias de Reyna, R. P. Brent, and J. van de Lune, “A note on
the real part of the Riemann zeta-function,” 2011, https://
arxiv.org/abs/1112.4910.
[20] M. Milgram, “Further exploration of Riemann’s functional
equation,” Journal of Classical Analysis, vol. 11, no. 1,
pp. 23–44, 2017.
[21] W. Heap, “A note on the maximum of the Riemann zeta
function on the 1-line,” 2018, https://arxiv.org/abs/1812.
01415.
[22] J. Bourgain, “Decoupling, exponential sums and the Riemann
zeta function,” 2016, https://arxiv.org/abs/1408.5794.
[23] Maplesoft, a division of Waterloo Maple Inc. Maple.
[24] Wolfram Research, Champagne, IL, Mathematica, 2014.

Research Article: An Integral Equation For Riemann's Zeta Function and Its Approximate Solution

Uploaded by

Copyright:

Available Formats

Research Article: An Integral Equation For Riemann's Zeta Function and Its Approximate Solution

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Research Article: An Integral Equation For Riemann's Zeta Function and Its Approximate Solution

Uploaded by

Copyright:

Available Formats

Hindawi

Abstract and Applied Analysis

Correspondence should be addressed to Michael Milgram; [email protected]

Received 23 October 2019; Accepted 6 January 2020; Published 15 May 2020

Academic Editor: Ying Hu

1. Introduction plane can be determined with respect to the

i c+i∞ Γ(− v)π v ζ(− 2v − 2) π1/4

􏼐− (2s − 1)2 /2 + 2iv − 2(c + 5/4)􏼑

This result is worthy of a few comments: 4. Analytic Continuations

4 ∞ ξ(1/2 + 2iv) π1/4

(2σ + 2it − 1)(σ + it)

8v2 t2 − σ 2 + σ 􏼁 + 2tv(1 − 2σ) + 2σ 3 (2 − σ) − 4t2 + 3􏼁σ 2 + 4t2 + 1􏼁σ − 2t4 − t2

8v3 +(− 16σ + 8)tv2 + − 6t2 + 6σ 2 − 6σ + 2􏼁v +(2σ − 1)t

inconsistency—the left-hand side of (66) vanishes exponen- 1 2σv − 4v

1 2(1 − σ)v − 4v (82)

− 16v2 + 2 − 64v4 + 48σ 2 − 48σ + 48􏼁v2 − 6σ 2 + 6σ − 2

|J2 (s, v)|

0 20 40 60 80 100 120 0 20 40 60 80 100 120

s = (1/3) + 10000 i s = (1/3) + 10000 i

1 arg(ζ(σ + it)) is embedded within the deﬁnition of ϕ (see

− cos(ω)cos(Φ)δ1 − 2(σ − 1/2)sin(Φ)sin(ω)δ4 􏼁 t − σ/2

− sin(ω)cos(Φ)δ1 + 2(σ − 1/2)cos(ω)sin(Φ)δ4 􏼁 t − σ/2

where consistency, this imposes a constraint on the parameter δ1 ,

σ = 0.05 lhs (σ = 0.12)

1 σ (1 – σ) Δ1 (σ, t = 10) 1 σ (1 – σ) Δ4 (σ, t = 10)

lhs (σ = 0.1) lhs (σ = 0.52)

􏼌􏼌 coincide, and, in fact, the two should be proportional to one

7. Zeros of ζ(σ + it) cos2 (ω) + sin2 (ω) � 0, (123)

π of δ1 and δ4 . Both these parameters were introduced to

Appendix does not appear to be amenable to direct analysis except

–60 –40 –20 0 20 40 60

Min t􏼐− 4σ(1 − σ)(2σ − 1)2 + 2􏼑

Min 2t(σ − 1/2)

From the above sums, it is possible to obtain the moments

∞ 3π21/4 − 10π5/4 Γ(3/4)8 − 16πΓ(3/4)9

∞ − π37/4 − 76π21/4 Γ(3/4)8 + 68π5/4 Γ(3/4)16 − 64πΓ(3/4)17

∞ 5π37/4 + 420π21/4 Γ(3/4)8 − 484π5/4 Γ(3/4)16 + 128πΓ(3/4)17

Data Availability References

[8] F. W. J. Olver, D. W. Lozier, R. F. Boisvert, and C. W. Clark,

You might also like