Next Article in Journal
Vector-Valued Fuzzy Metric Spaces and Fixed Point Theorems
Previous Article in Journal
The Enumeration of (⊙,∨)-Multiderivations on a Finite MV-Chain
Previous Article in Special Issue
3F4 Hypergeometric Functions as a Sum of a Product of 2F3 Functions
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Generalized Limit Theorem for Mellin Transform of the Riemann Zeta-Function

by
Antanas Laurinčikas
1,*,† and
Darius Šiaučiūnas
2,†
1
Institute of Mathematics, Faculty of Mathematics and Informatics, Vilnius University, Naugarduko Str. 24, LT-03225 Vilnius, Lithuania
2
Institute of Regional Development, Šiauliai Academy, Vilnius University, Vytauto Str. 84, LT-76352 Šiauliai, Lithuania
*
Author to whom correspondence should be addressed.
These authors contributed equally to this work.
Axioms 2024, 13(4), 251; https://doi.org/10.3390/axioms13040251
Submission received: 13 March 2024 / Revised: 6 April 2024 / Accepted: 8 April 2024 / Published: 10 April 2024

Abstract

:
In the paper, we prove a limit theorem in the sense of the weak convergence of probability measures for the modified Mellin transform Z ( s ) , s = σ + i t , with fixed 1 / 2 < σ < 1 , of the square | ζ ( 1 / 2 + i t ) | 2 of the Riemann zeta-function. We consider probability measures defined by means of Z ( σ + i φ ( t ) ) , where φ ( t ) , t t 0 > 0 , is an increasing to + differentiable function with monotonically decreasing derivative φ ( t ) satisfying a certain normalizing estimate related to the mean square of the function Z ( σ + i φ ( t ) ) . This allows us to extend the distribution laws for Z ( s ) .

1. Introduction

Let s = σ + i t be a complex variable. One of the most important objects of the classical analytic number theory is the Riemann zeta-function ζ ( s ) , which is defined, for σ > 1 , by the Dirichlet series
ζ ( s ) = m = 1 1 m s .
Moreover, the function ζ ( s ) has analytic continuation to the region C \ { 1 } , and the point s = 1 is its simple pole with residue 1. The first value distribution results for ζ ( s ) with real s were obtained by Euler. Riemann was the first mathematician who began to study [1] ζ ( s ) with complex variables, proved the functional equation for ζ ( s ) , obtained its analytic continuation, proposed a means of using ζ ( s ) for the investigation of the asymptotic prime number distribution law
π ( x ) = p x 1 , x ,
and stated some hypotheses on ζ ( s ) . The most important hypothesis, now called the Riemann hypothesis, states that all zeros of ζ ( s ) in the region σ 0 are located on the line σ = 1 / 2 . Riemann’s ideas concerning π ( x ) were correct, and Hadamard [2] and de la Vallée Poussin [3], using them, independently proved that
lim x π ( x ) 2 x d u log u 1 = 1 .
However, the Riemann hypothesis remains open at present; it is among the seven Millennium Problems of mathematics [4]. In the theory of ζ ( s ) , there are other important problems. One of them is connected to the asymptotics of moments
M k ( σ , T ) = def 0 T ζ ( σ + i t ) 2 k d t , k > 0 , σ 1 2 ,
as T . For example, at the moment the asymptotics of M k ( σ , T ) , σ = 1 / 2 is known only for k = 1 and k = 2 ; see [5]. For the investigation of M k ( σ , T ) , Motohashi proposed (see [6,7]) to use the modified Mellin transforms
Z k ( s ) = 1 ζ 1 2 + i x 2 k x s d x , k N .
Let g ( x ) be a certain function, e.g., g ( x ) x σ 1 L ( 0 , ) , and
G ( s ) = 0 g ( x ) x s 1 d x .
Then, using the Mellin inverse formula leads to the following equality (see [8]):
1 g x T ζ 1 2 + i x 2 k d x = 1 2 π i c i c + i G ( s ) T s Z k ( s ) d s
with a certain c > 1 . This shows that a suitable choice of the function g ( x ) reduces investigations of M k ( 1 / 2 , T ) to those of properties of Z k ( s ) . The latter assertion inspired the creation of the analytic theory of the functions Z k ( s ) .
In this paper, we limit ourselves to the probabilistic value distribution of the function Z ( s ) = def Z 1 ( s ) only. Before this, we recall some known results of the function Z ( s ) .
Let γ = 0.577 denote the Euler constant and E ( T ) be defined by
0 T ζ ( 1 2 + i t 2 d t = T log T 2 π + ( 2 γ 1 ) T + E ( T ) .
Moreover, let
F ( t ) = 1 T E ( t ) d t π T and F 1 ( T ) = 1 T F ( t ) d t .
The analytic behavior of the function Z ( s ) was described in [9] and forms the following theorem.
Theorem 1 
([9]). The function Z ( s ) is analytically continuable to the region σ > 3 / 4 , except the point s = 1 , which is a double pole, and
Z ( s ) = 1 ( s 1 ) 2 + 2 γ log 2 π s 1 E ( 1 ) + π ( s + 1 ) + s ( s + 1 ) ( s + 2 ) 1 F 1 ( x ) x s 3 d x .
Moreover, the estimates
Z ( σ + i t ) ε t 1 σ + ε , 0 σ 1 , t t 0 > 0 ,
and
1 T Z ( σ + i t ) 2 d t ε T 3 4 σ + ε if 0 σ 1 / 2 , T 2 2 σ + ε if 1 / 2 σ 1 ,
are valid.
Here and in what follows, ε is an arbitrary fixed positive number that is not always the same, and the notation x ε y , x C , y > 0 , means that there is a constant c = c ( ε ) > 0 such that | x | c y .
In [10], Bohr proposed to characterize the asymptotic behavior of the Riemann zeta-function by using a probabilistic approach. This idea is acceptable because the value distribution of ζ ( s ) is quite chaotic. Denote by J A the Jordan measure of the set A R . Then, Bohr, jointly with Jessen, roughly speaking, obtained in [11,12] that, for σ > 1 / 2 and every rectangle R C with edges parallel to the axes, there exists a limit
lim T J { t [ 0 , T ] : ζ ( σ + i t ) R } .
In modern terminology, the Bohr–Jessen theorem is stated as a limit theorem on weakly convergent probability measures. Let B ( X ) stand for the Borel σ -field of the space X (in general, topological), and let P n , n N , and P be probability measures defined on ( X , B ( X ) ) . By this definition, P n converges weakly to P as n ( P n n w P ) if
lim n X g d P n = X g d P
for every real continuous bounded function g on X . Let L A stand for the Lebesgue measure of a measurable set A R . Then, the modern version of the Bohr–Jessen theorem is of the following form: for every fixed σ > 1 / 2 , there exists a probability measure P σ on ( C , B ( C ) ) such that
1 T L { t [ 0 , T ] : ζ ( σ + i t ) A } , A B ( C ) ,
converges weakly to P σ as T .
The first probabilistic limit theorems for the function Z ( s ) were discussed in [13]. For A B ( C ) , set
Q T , σ ( A ) = 1 T L { t [ 0 , T ] : Z ( σ + i t ) A } .
Assuming that σ > 1 / 2 , it was obtained that there is a probability measure Q σ on ( C , B ( C ) ) such that Q T , σ T w Q σ . On the other hand, for every κ > 0 , we have
1 T L { t [ 0 , T ] : | Z ( σ + i t ) | κ } 1 κ T 0 T | Z ( σ + i t ) | d t 1 κ 1 T 0 T | Z ( σ + i t ) | 2 d t 1 / 2 .
This, together with Theorem 1, implies that, for 1 / 2 < σ < 1 ,
lim T 1 T L { t [ 0 , T ] : | Z ( σ + i t ) | κ } = 0 .
The latter equality remains valid also for σ > 1 . Thus, the limit measure Q σ is degenerated at the point s = 0 . In order to avoid this situation, we propose to consider Z ( σ + i φ ( t ) ) with a certain function φ ( t ) . Moreover, it is more convenient to deal with t [ T , 2 T ] because, in this case, additional restrictions for φ ( t ) with t = 0 are not needed.
Denote
I σ ( T ) = 1 T | Z ( σ + i t ) | 2 d t .
We suppose that φ ( t ) is a positive increasing to + differentiable function with a monotonically decreasing derivative, such that
I σ ε ( φ ( T ) ) φ ( T ) T , T .
The class of such functions φ ( t ) is denoted by W σ . Consider the weak convergence for
P T , σ ( A ) = 1 T L { t [ T , 2 T ] : Z ( σ + i φ ( τ ) ) A } , A B ( C ) .
In this case, we have, by ε 0 , that
I σ ( φ ( T ) ) φ ( T ) T ,
and
1 T T 2 T | Z ( σ + i φ ( t ) ) | 2 d t = 1 T φ ( T ) φ ( 2 T ) 1 φ ( t ) | Z ( σ + i u ) | 2 d u 1 T φ ( 2 T ) I σ ( φ ( 2 T ) ) 1
for φ ( t ) W σ . Thus, we cannot claim that the limit measure for P T , σ is degenerated at zero.
Now, we state a limit theorem for P T , σ .
Theorem 2. 
Assume that σ ( 1 / 2 , 1 ) is a given fixed number, and φ ( t ) W σ . Then, on ( C , B ( C ) ) , there exists a probability measure P σ such that P T , σ T w P σ .
In virtue of Theorem 1, we see that
I σ ε ( T ) T α σ
with certain 0 < α σ < 1 . Take φ ( t ) = ( log t ) β σ , t 2 , β σ > 0 . Then, φ ( t ) is decreasing, and
I σ ε ( φ ( T ) ) φ ( T ) T ( log T ) α σ β σ + 1 β σ T
if we choose
β σ = ( 1 α σ ) 1 .
This shows that ( log T ) β σ is an element of the class W σ .
Theorem 2 shows that the asymptotic behavior of the function Z ( s ) on vertical lines is governed by a certain probabilistic law, and this confirms the chaos in its value distribution. Moreover, Theorem 2 is an example of the application of probability methods in analysis. Thus, it continues a tradition initiated in works [11,12] and developed by Selberg [14], Joyner [15], Bagchi [16], Korolev [17,18], Kowalski [19], Lamzouri, Lester and Radziwill [20,21], Steuding [22], and others; see also a survey paper [23]. We note that a generalization of Theorem 2 for the functional spaces can be applied for approximation problems of some classes of functions.
We divide the proof of Theorem 2 into several parts. First, we discuss weak convergence on a certain group. The second part is devoted to the case related to a integral. Further, we consider a measure defined by an absolutely convergent improper integral. In the last part, Theorem 2 is proven. For proofs of all assertions on weak convergence, the notions of relative compactness as well as of tightness and convergence in distribution are employed.

2. Fourier Transform Method

Let b > 1 be a fixed finite number, and
I b = x [ 1 , b ] { s C : | s | = 1 } .
The Cartesian product I b consists of all functions i : [ 1 , b ] { s C : | s | = 1 } . On I b , the product topology and operation of pointwise multiplication can be defined. This reduces I b to a compact topological group. We will give a limit lemma for probability measures on ( I b , B ( I b ) ) .
For A B ( I b ) , put
V T , b ( A ) = 1 T L t [ T , 2 T ] : x i φ ( t ) : x [ 1 , b ] A .
Lemma 1. 
Suppose that the function φ ( t ) has a monotonically decreasing derivative φ ( t ) such that
( φ ( T ) ) 1 = o ( T ) , T .
Then V T , b converges weakly to a certain probability measure V b as T .
Proof. 
We use the Fourier transform approach. Denote the elements of I b by i = { i x : x [ 1 , b ] } . Then, the Fourier transform f T , b ( k ̲ ) , k ̲ = ( k x : k x Z , x [ 1 , b ] ) of the measure V T , b is the integral
f T , b ( k ̲ ) = I b x [ 1 , b ] i x k x d V T , b ,
where only a finite number of integers k x are not zeros. Therefore, the definition of V T , b yields
f T , b ( k ̲ ) = 1 T T 2 T x [ 1 , b ] x i k x φ ( t ) d t = 1 T T 2 T exp i φ ( t ) x [ 1 , b ] k x log x d t .
For brevity, let A b ( k ̲ ) = k [ 1 , b ] k x log x . Then, the second mean value theorem, (4), and (3) give
Re f T , b ( k ̲ ) = 1 T T 2 T cos φ ( t ) A b ( k ̲ ) d t = 1 A b ( k ̲ ) T T 2 T 1 φ ( t ) d sin φ ( t ) A b ( k ̲ ) 1 | A b ( k ̲ ) | 1 φ ( 2 T ) T = o ( 1 ) , T ,
provided that A b ( k ̲ ) 0 . Clearly, the same estimate holds for Im f T , b ( k ̲ ) . Hence, for A b ( k ̲ ) 0 ,
lim T f T , b ( k ̲ ) = 0 .
Obviously,
f T , b ( k ̲ ) = 1
if A b ( k ̲ ) = 0 . This and (5) show that
V T , b T w V b ,
where V b is a probability measure on ( I b , B ( I b ) ) defined by the Fourier transform
f b ( k ̲ ) = 1 if A b ( k ̲ ) = 0 , 0 if A b ( k ̲ ) 0 .
Now, we will apply Lemma 1 for the measure defined by means of a certain finite sum.
Let θ > 1 / 2 be a fixed number, and, for x , y [ 1 , ) ,
u ( x , y ) = exp x y θ .
Moreover, we use the notation ζ ^ ( t ) = | ζ ( 1 / 2 + i t ) | 2 . Consider the nth integral sum
U n , b , y ( σ + i φ ( t ) ) = b 1 n l = 1 n ζ ^ ( a l ) u ( a l , y ) a l σ i φ ( t ) , n N ,
where a l [ x l 1 , x l ] and x l = 1 + ( ( b 1 ) / n ) l .
For A B ( C ) , set
P T , n , b , y ( A ) = 1 T L t [ T , 2 T ] : U n , b , y ( σ + i φ ( t ) ) A .
For simplicity, here and in the following, we omit the dependence on σ of some objects. Before the statement of the limit lemma for P T , n , b , y , we will present some lower estimates for the mean square I σ ( T ) . For this, we will apply the following general lemma from [8]. Let F ( s ) be the modified Mellin transform of f ( x ) , i.e.,
F ( s ) = 1 f ( x ) x s d x .
Lemma 2 
([8], Lemma 5). Let f ( x ) C [ 2 , ] be a real-valued function such that
1
1 X f ( k ) ( x ) d x ε , k X 1 + ε , k N 0 ;
2 F ( s ) has analytic continuation to the half-plane σ > 1 / 2 , except for a pole of order l at the point s = 1 ;
3 For σ > 1 / 2 , F ( s ) is of polynomial growth in | t | .
Then, for 1 / 2 < σ < 1 and any fixed ε > 0 ,
T 2 T f 2 ( x ) d x ε log l 1 T T / 2 5 T / 2 | f ( x ) | d x + T 2 σ 1 0 T 1 + ε | F ( σ + i t ) | 2 d t .
Lemma 3. 
For 1 / 2 < σ < 1 , and any ε > 0 , the estimate
I σ ( T ) ε T 2 2 σ ε
holds.
Proof. 
As usual, denote by Z ( t ) , t R , the Hardy function, i.e.,
Z ( t ) = ζ 1 2 + i t χ 1 / 2 1 2 + i t ,
where
χ ( s ) = ζ ( s ) ζ ( 1 s ) .
It is well known that Z ( t ) is a real-valued function satisfying | Z ( t ) | = | ζ ( 1 / 2 + i t ) | . Moreover, the estimate [8]
Z ( k ) ( t ) k t 1 / 4 ( log T ) k + 1 + m t / ( 2 π ) m 1 / 2 log t / ( 2 π ) m k
holds. Take f ( x ) = Z 2 ( x ) . Then, we have
F ( s ) = 1 Z 2 ( x ) x s d x = 1 ζ 1 2 + i x 2 x s d x = Z ( s ) .
In view of Theorem 1 and (6), the function satisfies the hypotheses of Lemma 1 with l = 2 . Thus, for 1 / 2 < σ < 1 ,
T 2 T f 2 ( t ) d t = T 2 T ζ 1 2 + i t 4 d t ε log T T / 2 5 T / 2 ζ 1 2 + i t 2 d t + T 2 σ 1 0 T 1 + ε | Z ( σ + i t ) | 2 d t .
Since [5]
0 T ζ 1 2 + i t 4 d t = 1 2 π 2 T log 4 T + O ( T log 3 T )
and
0 T ζ 1 2 + i t 2 d t T log T ,
this implies
T log 4 T ε T 2 σ 1 0 T 1 + ε | Z ( σ + i t ) | 2 d t .
Consequently,
I σ ( T ) ε T ( 2 2 σ ) / ( 1 + ε ) ε T 2 2 σ ε .
Lemma 4. 
Assume that σ ( 1 / 2 , 1 ) is a given fixed number, and φ ( t ) W σ . Then, on ( C , B ( C ) ) , there exists a probability measure P n , b , y such that P T , n , b , y T w P n , b , y .
Proof. 
Lemma 3 implies that, for σ ( 1 / 2 , 1 ) , I σ ( T ) as T . Therefore, if φ ( t ) W σ , then
1 φ ( T ) T I σ 1 ( φ ( T ) ) = o ( T )
as T . Thus, the application of Lemma 1 is possible.
Consider the mapping v n , b : I b C defined by
v n , b ( i ) = b 1 n l = 1 n ζ ^ ( a l ) u ( a l , y ) a l σ i a l .
Since the latter sum is finite, and I b is equipped with the product topology, the mapping v n , b is continuous. Moreover, in view of (7),
v n , b x i φ ( t ) : x [ 1 , b ] = b 1 n l = 1 n ζ ^ ( a l ) u ( a l , y ) a l σ i φ ( t ) = U n , b , y ( σ + i φ ( t ) ) .
Hence, for A B ( C ) ,
P T , n , b , y ( A ) = 1 T L t [ T , 2 T ] : v n , b x i φ ( t ) : x [ 1 , b ] A = 1 T L t [ T , 2 T ] : x i φ ( t ) : x [ 1 , b ] v n , b 1 A = V T , b v n , b 1 A ,
where V T , b is from Lemma 1. The continuity of the mapping v n , b implies its ( B ( I b ) , B ( C ) ) -measurability. Therefore, the mapping v n , b and any probability measure P on ( I b , B ( I b ) ) define the unique probability measure P v n , b 1 on ( C , B ( C ) ) given by
P v n , b 1 ( A ) = P ( v n , b 1 A ) , A B ( C ) .
See Section 2 of [24]. Thus, by (8), we have P T , n , b , y = V T , b v n , b 1 . Therefore, Lemma 1, the continuity of v n , b , and the principle of the preservation of week convergence under continuity mappings (Theorem 5.1 of [24]) show that
P T , n , b , y T w P n , b , y ,
where P n , b , y = V b v n , b 1 , and V b is the limit measure in Lemma 1. □

3. Limit Lemma for Integral

Denote
Z b , y ( σ + i φ ( t ) ) = 1 b ζ ^ ( x ) u ( x , y ) x σ i φ ( t ) d x ,
and, for A B ( C ) , set
P T , b , y ( A ) = 1 T L t [ T , 2 T ] : Z b , y ( σ + i φ ( t ) ) A .
In this section, we will prove the weak convergence for P T , b , y as T . Before this, we recall some known probabilistic results. Let { Q } be a certain family of probability measures on ( X , B ( X ) ) . The family { Q } is called tight if, for every δ > 0 , there is a compact set K X such that
Q ( K ) > 1 δ
for all Q { Q } . The family { Q } is said to be relatively compact if every sequence contains a subsequence weakly convergent to a certain probability measure on ( X , B ( X ) ) . The Prokhorov theorem connects two above notions, and, for convenience, we state it as the following lemma.
Lemma 5. 
If a family of probability measures is tight, then it is relatively compact.
The proof of the lemma is given in [24], Theorem 5.1.
Moreover, we recall one useful property on convergence in distribution. Let ξ n and ξ be X -valued random elements defined on the probability space ( Ω , F , μ ) with distributions P n and P, respectively. Then, ξ n converges in distribution to ξ as n ( n D ) if
P n n w P .
Now, we state a lemma on convergence in distribution.
Lemma 6. 
Assume that the metric space ( X , d ) is separable, and ξ n k , ξ n are X -valued random elements defined on the same probability space ( Ω , F , μ ) . Let
ξ n k n D ξ k
and
ξ k k D ξ .
If, for every δ > 0 ,
lim k lim sup n μ { d ( ξ n k , η k ) δ } = 0 ,
then
η n n D ξ .
The lemma is proven in [24], Theorem 3.2.
Lemma 7. 
Assume that σ ( 1 / 2 , 1 ) is a given fixed number, and φ ( t ) W σ . Then, on ( C , B ( C ) ) , there exists a probability measure P b , y such that P T , b , y T w P b , y .
Proof. 
First, we will show that Z b , y ( σ + i φ ( t ) ) is close in a certain sense to U n , b , y ( σ + i φ ( t ) ) .
Let
J T , n = def 1 T T 2 T Z b , y ( σ + i φ ( t ) ) U n , b , y ( σ + φ ( t ) ) d t .
Clearly,
J T , n 2 1 T T 2 T Z b , y ( σ + i φ ( t ) ) U n , b , y ( σ + φ ( t ) ) 2 d t .
We have
T 2 T Z b , y ( σ + i φ ( t ) ) 2 d t = T 2 T 1 b ζ ^ ( x ) u ( x , y ) x σ i φ ( t ) d x 1 b ζ ^ ( x ) u ( x , y ) x σ + i φ ( t ) d x d t = T 1 b 1 b x 1 = x 2 ζ ^ ( x 1 ) ζ ^ ( x 2 ) u ( x 1 , y ) u ( x 2 , y ) x 1 σ x 2 σ d x 1 d x 2 + 1 b 1 b x 1 x 2 ζ ^ ( x 1 ) ζ ^ ( x 2 ) u ( x 1 , y ) u ( x 2 , y ) x 1 σ x 2 σ T 2 T x 1 x 2 i φ ( t ) d t d x 1 d x 2 .
Since
Re T 2 T x 1 x 2 i φ ( t ) d t = log x 1 x 2 1 T 2 T 1 φ ( t ) d sin φ ( t ) log x 1 x 2 log x 1 x 2 1 1 φ ( 2 T ) ,
and the same bound is true for the imaginary part of the latter integral, we obtain by (10) that
T 2 T Z b , y ( σ + i φ ( t ) ) 2 d t = o ( T ) , T .
Reasoning similarly, we find
T 2 T U n , b , y ( σ + i φ ( t ) ) 2 d t = T b 1 n 2 l = 1 n ζ ^ 2 ( a l ) u 2 ( a l , y ) a l 2 + O b 1 n 2 l 1 = 1 n l 2 = 1 n l 1 l 2 ζ ^ ( a l 1 ) ζ ^ ( a l 2 ) u ( a l 1 , y ) u ( a l 2 , y ) a l 1 σ a l 2 σ log a l 1 a l 2 1 .
Thus,
lim n lim sup T 1 T T 2 T U n , b , y ( σ + i φ ( t ) ) 2 d t = 0 .
By (9),
J T , n 2 1 T T 2 T Z b , y ( σ + i φ ( t ) ) 2 d t + T 2 T Z b , y ( σ + i φ ( t ) ) 2 d t T 2 T U n , b , y ( σ + i φ ( t ) ) 2 d t 1 / 2 + T 2 T U n , b , y ( σ + i φ ( t ) ) 2 d t .
Therefore, (11) and (13) yield
lim n lim sup T J T , n = 0 .
Now, we will deal with the sequence { P n , b , y : n N } . By (12), we have
sup s N lim sup T 1 T T 2 T U n , b , y ( σ + i φ ( t ) ) d t sup s N lim sup T 1 T T 2 T U n , b , y ( σ + i φ ( t ) ) 2 d t 1 / 2 sup n N b 1 n l = 1 n ζ ^ 2 ( a l ) u 2 ( a l , y ) a l 2 σ 1 / 2 C b , y , σ <
because
lim n b 1 n l = 1 n ζ ^ 2 ( a l ) u 2 ( a l , y ) a l 2 σ = 1 b ζ ^ 2 ( x ) u 2 ( x , y ) x 2 σ d x .
Take a random variable θ T given on the probability space ( Ω , F , μ ) that is uniformly distributed on [ T , 2 T ] . Consider the complex-valued random variables
x T , n . , b , y = x T , n , b , y ( σ ) = U n , b , y ( σ + i φ ( θ T ) ) ,
and x n , b , y ( σ ) with the distribution P n , b , y , σ . Then, rewrite the assertion of Lemma 4 in the form
x T , n , b , y T D x n , b , y .
Fix δ > 0 . Then, in view of (15) and (16),
μ x n , b , y ( σ ) > δ 1 C b , y , σ sup n N lim sup T μ x T , n , b , y ( σ ) > δ 1 C b , y , σ sup n N lim sup T δ C b , y , σ T 2 T U n , b , y ( σ + i φ ( t ) ) d t δ .
The set K = { s C : | s | δ 1 C b , y , σ } is compact in C . Moreover, by (17),
μ x n , b , y K = 1 μ x n , b , y K > 1 δ
for all n N . This and the definition of x n , b , y show that, for all n N ,
P n , b , y , σ ( K ) > 1 δ .
This means that the sequence { P n , b , y , σ : n N } is tight. Therefore, by Lemma 5, it is relatively compact. Hence, there exists a subsequence { P n l , b , y , σ } { P n , b , y , σ } and a probability measure P b , y , σ on ( C , B ( C ) ) such that P n l , b , y , σ l w P b , y , σ . In other words,
x n l , b , y l D P b , y , σ .
This, (16), and (14) show that all hypotheses of Lemma 6 for x T , n , b , y , x n l , b , y and
y T , b , y = y T , b , y ( σ ) = Z b , y ( σ + i φ ( θ T ) )
are satisfied. Thus, we have
y T , b , y T D P b , y , σ ,
which proves the lemma. □

4. Case of Improper Integral

This section is devoted to a limit lemma for the integral
Z y ( σ + i φ ( t ) ) = 1 ζ ^ ( x ) u ( x , y ) x σ i φ ( t ) d x .
It is well known that ζ ( 1 / 2 + i x ) ( 1 + | x | ) 1 / 6 . Therefore, the integral for Z ( σ + i φ ( t ) ) converges absolutely for σ > σ ^ with every finite σ ^ .
For A B ( C ) , let
P T , y , σ ( A ) = 1 T L t [ T , 2 T ] : Z y ( σ + i φ ( t ) ) A .
Lemma 8. 
Assume that σ ( 1 / 2 , 1 ) is a given fixed number, and φ ( t ) W σ . Then, there is a probability measure P y , σ on ( C ( B ( C ) ) such that P T , y , σ T w P y , σ .
Proof. 
We use a similar method as in the proof of Lemma 7. We begin with a mean value
J T , y = def 1 T 0 T Z y ( σ + i φ ( t ) ) Z b , y ( σ + i φ ( t ) ) d t .
Clearly, the absolute convergence of the integral for Z y ( σ + i φ ( t ) ) shows that, for every fixed y > 0 ,
Z y ( σ + i φ ( t ) ) Z b , y ( σ + i φ ( t ) ) = b ζ ^ ( x ) u ( x , y ) x σ i φ ( t ) d x b ζ ^ ( x ) u ( x , y ) x σ d x = o y ( 1 )
as b . Hence, we obtain
lim b lim sup T J T , y = 0 .
Let y b , y ( σ ) be the complex-valued random variable with distribution P b , y , σ , and, in the notation of Lemma 7,
y T , b , y = y T , b , y ( σ ) = Z b , y ( σ + i φ ( θ T ) ) .
Then, by Lemma 7,
y T , b , y T D y b , y .
Moreover, in virtue of (11),
sup b 1 lim sup T 1 T T 2 T Z b , y ( σ + i φ ( t ) ) d t C y , σ < .
This together with (19) gives, for δ > 0 ,
μ y b , y > δ 1 C y , σ sup b 1 lim sup T μ y b , y > δ 1 C y , σ sup b 1 lim sup T δ C y , σ T 2 T Z b , y ( σ + i φ ( t ) ) d t δ .
Taking a set K = { s C : | s | δ 1 C y , σ } , from this, we deduce that
μ y b , y K > 1 δ .
This implies that the family { P b , y , σ : b 1 } is tight. Therefore, in view of Lemma 5, it is relatively compact. Thus, there is a sequence { P b l , y , σ } and a probability measure P y , σ on ( C , B ( C ) ) such that
y b l , y , σ l D P y , σ .
This, (19), (18), and the application of Lemma 6 complete the proof of the lemma. □

5. Proof of Theorem 2

We recall that
u ( x , y ) = exp x y θ , x , y [ 1 , ) ,
with a fixed θ > 1 / 2 . For brevity, set
f ( s , y ) = 1 θ Γ s θ y s ,
where Γ ( s ) is the Euler gamma-function. For the approximation of Z ( σ + i φ ( t ) ) by Z y ( σ + φ ( t ) ) , we use the representation
Z y ( s ) = 1 2 π i θ i θ + i Z ( s + z ) f ( z , y ) d z , 1 2 < σ < 1 ,
obtained in [25], Lemma 9.
Lemma 9. 
Under the hypotheses of Theorem 2,
lim y lim sup T 1 T T 2 T Z ( σ + i φ ( t ) ) Z y ( σ + i φ ( t ) ) d t = 0 .
Proof. 
Let θ 1 = ε and θ = 1 / 2 + ε . The integrand in (20) has a double pole z = 1 s and a simple pole z = 0 lying in θ 1 < Re z < θ . Therefore, by the residue theorem and (20), we have
Z y ( s ) Z ( s ) = 1 2 π i θ 1 i θ 1 + i Z ( s + z ) f ( z , y ) d z + r y ( s ) ,
where
r y ( s ) = Res z = 1 s Z ( s ) f ( s , y ) .
From this, we obtain
Z y ( σ + i φ ( t ) ) Z ( σ + i φ ( t ) ) = 1 2 π Z σ ε + i φ ( t ) + i τ f ε + i τ , y d τ + r y ( σ + i φ ( t ) ) Z σ ε + i φ ( t ) + i τ f ε + i τ , y d τ + r y ( σ + i φ ( t ) ) .
Thus,
1 T T 2 T Z ( σ + i φ ( t ) ) Z y ( σ + i φ ( t ) ) d t I T , y ,
where
I T , y = def 1 T T 2 T Z σ ε + i φ ( t ) + i τ d t f ε + i τ , y d τ + 1 T T 2 T r y ( σ + i φ ( t ) ) d t = I T , y ( 1 ) + I T , y ( 2 ) .
To estimate I T , y ( 1 ) , we observe that
1 T T 2 T Z σ ε + i φ ( t ) + i τ d t 1 T T 2 T Z σ ε + i φ ( t ) + i τ 2 d t 1 / 2 = 1 T T 2 T Z σ ε + i φ ( t ) + i τ 2 φ ( t ) d t φ ( t ) 1 / 2 1 T φ ( 2 T + | τ | ) 0 φ ( 2 T + | τ | ) Z σ ε + i u 2 d u I σ ε φ ( 2 T + | τ | ) T φ ( 2 T + | τ | ) 1 / 2 2 T + | τ | T 1 / 2 ( 1 + | τ | ) 1 / 2 .
For the gamma-function, the estimate
Γ ( σ + i t ) exp { c | t | } , c > 0 ,
is valid. Therefore,
f ε + i τ , y y ε exp { c 1 | τ | } , c 1 > 0 .
This together with (23) leads to the bound
I T , y ( 1 ) y ε ( 1 + | τ | ) 1 / 2 exp { c 1 | τ | } d τ y ε .
Let a = 2 γ log 2 π . In view of the formula for Z ( s ) in Theorem 1,
r y ( s ) = f ( 1 s , y ) + a f ( 1 s , y ) = 1 θ 2 Γ 1 s θ y 1 s + 1 θ Γ 1 s θ y 1 s log y + a θ Γ 1 s θ y 1 s = y 1 s θ Γ 1 s θ 1 θ Γ Γ 1 s θ + log y + a .
Hence, the estimates (24) and
Γ Γ ( σ + i t ) log ( | t | + 2 )
yield
I T , y ( 2 ) θ y 1 σ log y 1 T T 2 T exp c θ φ ( t ) log φ ( t ) d t θ y 1 σ log y exp c 2 θ φ ( T ) .
This, (25), and (22) show that
I T , y δ y ε + y 1 σ log y exp c 2 θ φ ( T ) .
Therefore,
lim y lim sup T 1 T T 2 T Z ( σ + i φ ( t ) ) Z y ( σ + i φ ( t ) ) d t = 0
because φ ( T ) as T . □
Now, we return to the limit measure P y , σ of Lemma 8.
Lemma 10. 
Under the hypotheses of Theorem 2, the family { P y , σ : y [ 1 , ) } is tight.
Proof. 
We have
1 T T 2 T Z y ( σ + i φ ( t ) ) d t 1 T T 2 T Z ( σ + i φ ( t ) ) Z y ( σ + i φ ( t ) ) d t + 1 T T 2 T Z ( σ + i φ ( t ) ) d t .
Therefore, by (2) and (26),
sup y 1 lim sup T 1 T T 2 T Z y ( σ + i φ ( t ) ) d t C < .
Let
z T , y = z T , y ( σ ) = Z y ( σ + i φ ( θ T ) ) ,
and z y = z y ( σ ) be the complex-valued random variable with the distribution P y , σ . Then, the statement of Lemma 8 can be written as
z T , y T D z y .
From this and (27), we obtain that, for every δ > 0 ,
μ z y > δ 1 C sup y 1 lim sup T μ z T , y > δ 1 C δ T C T 2 T Z y ( σ + i φ ( t ) ) d t δ .
This shows that, for K = { s C : | s | δ 1 C } ,
P y , σ ( K ) 1 δ ,
and the lemma is proven. □
Proof of Theorem 2. 
Lemma 10 together with Lemma 5 implies that the family { P y , σ } is relatively compact. Therefore, there is a sequence { P y k , σ } { P y , σ } weakly convergent to a certain probability measure P σ on ( C , B ( C ) as k . This also can be written as
z y k , σ k D P σ .
Define one more random variable,
z T = z T ( σ ) = Z ( σ + i φ ( θ T ) ) .
Then, Lemma 9 implies, for every δ > 0 ,
lim k lim sup T μ z T z T , y k > δ lim k lim sup T 1 δ T T 2 T Z ( σ + i φ ( t ) ) Z y k ( σ + i φ ( t ) ) d t = 0 .
This, (28), and (29) together with Lemma 6 prove that
z T T D P σ .
The theorem is proven. □

6. Conclusions

In the paper, we considered the asymptotic behavior of the modified Mellin transform of the square of the Riemann zeta-function by using a probabilistic approach. We proved a limit theorem on the weak convergence of probability measures defined by shifts Z ( σ + i φ ( t ) ) , 1 / 2 < σ < 1 , where φ ( t ) is a differentiable increasing to infinity function with a monotonically decreasing derivative φ ( t ) satisfying a certain estimate connected to the mean square of the function Z ( s ) . We expect that such normalization of the function Z ( s ) extends the class of limit distributions for Z ( s ) . Our future plans are related to a similar theorem in the space of analytic functions.

Author Contributions

Conceptualization, A.L. and D.Š.; methodology, A.L. and D.Š.; investigation, A.L. and D.Š.; writing—original draft preparation, A.L. and D.Š. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

Data are contained within the article.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Riemann, B. Über die Anzahl der Primzahlen unterhalb einer gegebenen Grösse. Monatsber. Preuss. Akad. Wiss. Berlin 1859, 671–680. [Google Scholar]
  2. Hadamard, J. Sur les zéros de la fonction ζ(s) de Riemann. Comptes Rendus Acad. Sci. Paris 1896, 122, 1470–1473. [Google Scholar]
  3. de la Vallée-Poussin, C.J. Recherches analytiques sur la théorie des nombres premiers, I–III. Ann. Soc. Sci. Brux. 1896, 20, 183–256, 281–362, 363–397. [Google Scholar]
  4. The Millennium Prize Problems. Available online: https://www.claymath.org/millennium-problems/ (accessed on 1 March 2024).
  5. Ivič, A. The Riemann Zeta-Function; John Wiley & Sons: New York, NY, USA, 1985. [Google Scholar]
  6. Motohashi, Y. A relation between the Riemann zeta-function and the hyperbolic Laplacian. Ann. Sc. Norm. Super. Pisa Cl. Sci. IV. Ser. 1995, 22, 299–313. [Google Scholar]
  7. Motohashi, Y. Spectral Theory of the Riemann Zeta-Function; Cambridge University Press: Cambridge, UK, 1997. [Google Scholar]
  8. Ivič, A. On some conjectures and results for the Riemann zeta-function and Hecke series. Acta Arith. 2001, 99, 115–145. [Google Scholar] [CrossRef]
  9. Ivič, A.; Jutila, M.; Motohashi, Y. The Mellin transform of powers of the zeta-function. Acta Arith. 2000, 95, 305–342. [Google Scholar] [CrossRef]
  10. Bohr, H. Über das Verhalten von ζ(s) in der Halbebene σ > 1. Nachr. Akad. Wiss. Göttingen II Math. Phys. Kl. 1911, 409–428. [Google Scholar]
  11. Bohr, H.; Jessen, B. Über die Wertwerteiling der Riemmanshen Zetafunktion, erste Mitteilung. Acta Math. 1930, 54, 1–35. [Google Scholar] [CrossRef]
  12. Bohr, H.; Jessen, B. Über die Wertwerteiling der Riemmanshen Zetafunktion, zweite Mitteilung. Acta Math. 1932, 58, 1–55. [Google Scholar] [CrossRef]
  13. Laurinčikas, A. Limit theorems for the Mellin transform of the square of the Riemann zeta-function. I. Acta Arith. 2006, 122, 173–184. [Google Scholar] [CrossRef]
  14. Selberg, A. Old and New Conjectures and Results about a Class of Dirichlet Series. In Proceedings of the Amalfi Conference on Analytic Number Theory, Maiori, Italy, 25–29 September 1989; University of Salerno: Salerno, Italy, 1992; pp. 367–385. [Google Scholar]
  15. Joyner, D. Distribution Theorems of L-Functions; Ditman Research Notes in Math; Longman Scientific: Harlow, UK, 1986. [Google Scholar]
  16. Bagchi, B. The Statistical Behaviour and Universality Properties of the Riemann Zeta-Function and Other Allied Dirichlet Series. Ph.D. Thesis, Indian Statistical Institute, Calcutta, India, 1981. [Google Scholar]
  17. Korolev, M.A. Gram’s law in the theory of the Riemann zeta-function. Part 1. Proc. Steklov Inst. Math. 2016, 292, 1–146. [Google Scholar] [CrossRef]
  18. Korolev, M.A. Gram’s law in the theory of the Riemann zeta-function. Part 2. Proc. Steklov Inst. Math. 2016, 294, 1–78. [Google Scholar] [CrossRef]
  19. Kowalski, E. An Introduction to Probabilistic Number Theory; Cambridge Studies in Advanced Mathematics (192); Cambridge University Press: Cambridge, UK, 2021. [Google Scholar]
  20. Lamzouri, Y.; Lester, S.; Radziwill, M. An effective universality theorem for the Riemann zeta function. Comment. Math. Helv. 2018, 93, 709–736. [Google Scholar] [CrossRef]
  21. Lamzouri, Y.; Lester, S.; Radziwill, M. Discrepancy bounds for the distribution of the Riemann zeta-function and applications. J. Anal. Math. 2019, 139, 453–494. [Google Scholar] [CrossRef]
  22. Steuding, J. Value-Distribution of L-Functions; Lecture Notes Math; Springer: Berlin/Heidelberg, Germany; New York, NY, USA, 2007; Volume 1877. [Google Scholar]
  23. Matsumoto, K. Probabilistic value-distribution theory of zeta-functions. Sugaku Expo. 2004, 17, 51–71. [Google Scholar]
  24. Billingsley, P. Convergence of Probability Measures, 2nd ed.; John Wiley & Sons: New York, NY, USA, 1999. [Google Scholar]
  25. Korolev, M.; Laurinčikas, A. On the approximation by Mellin transform of the Riemann zeta-function. Axioms 2023, 12, 520. [Google Scholar] [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Laurinčikas, A.; Šiaučiūnas, D. Generalized Limit Theorem for Mellin Transform of the Riemann Zeta-Function. Axioms 2024, 13, 251. https://doi.org/10.3390/axioms13040251

AMA Style

Laurinčikas A, Šiaučiūnas D. Generalized Limit Theorem for Mellin Transform of the Riemann Zeta-Function. Axioms. 2024; 13(4):251. https://doi.org/10.3390/axioms13040251

Chicago/Turabian Style

Laurinčikas, Antanas, and Darius Šiaučiūnas. 2024. "Generalized Limit Theorem for Mellin Transform of the Riemann Zeta-Function" Axioms 13, no. 4: 251. https://doi.org/10.3390/axioms13040251

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop