Preprint
Article

This version is not peer-reviewed.

A Note on the Mean Square of Riemann Zeta-Function

Submitted:

14 April 2025

Posted:

16 April 2025

You are already at the latest version

Abstract
In this paper, we will give a new proof of a known result of the mean square of Riemann zeta-function
Keywords: 
;  

1. Introduction

Let
I = 0 T | ζ ( 1 / 2 + i t ) | 2 | A ( 1 / 2 + i t ) | 2 d t
where ζ ( s ) is the Riemann zeta-function, and
A ( s ) = 1 m M a ( m ) m s
is a finite Dirichlet series. For this type of mean square of ζ ( s ) , there are a number of researches, which is related to estimate the number of the zeros of ζ ( s ) on the critical line Re ( s ) = 1 / 2 , see [1,2,3,4,5,6]As before, for two positive integers h , k , denote h * = h / ( h , k ) , k * = k / ( h , k ) , and h * ¯ is the least positive integer such that h * ¯ h * 1 mod k * . For s = c + i t , c = 1 + η , 0 < η < 1 , denote by
M ( s ) = n = 1 d ( n ) n s h , k M a ( h ) a ( k ) ¯ ( h k ) 1 s ( h , k ) ( 1 2 s ) e n h * ¯ k *
and V = sup | t | M { M ( c + i t ) } .
Balasubramanian, Conrey, and Heath-Brown proved that
For a ( m ) m ϵ , and log ( m ) log T , there is
I = T h , k M a ( h ) h a ( k ) ¯ k ( h , k ) log T ( h , k ) 2 2 π h k + 2 γ 1 + E .
where E V T ϵ + o ( T ) . In the paper [1], it is applied an auxiliary function w ( t , T 1 , T 2 ) , which plays an important role of "ferrying", Instead, in this paper, we will make use of a new one, i.e. ω ( t , T 1 , T 2 ) defined in Lemma 2.1. Our result is that
Theorem 1.1.
Suppose that a ( m ) m ϵ , and log ( m ) log T , then
I = T h , k M a ( h ) h a ( k ) ¯ k ( h , k ) log T ( h , k ) 2 2 π h k + 2 γ 1 + E ˜ .
where E ˜ V T η + ϵ + o ( T ) .
As it is easy to know that V M 2 + 2 ϵ + 2 η , then with a properly re-define ϵ , η , it has
E ˜ , E M 2 T ϵ .
The main arguments in this paper are most similar to the ones in [1].

2. The Proof of Theorem 1.1

Lemma 2.1.
Suppose that λ = T 2 ϵ , define
ω ( t , T 1 , T 2 ) = e λ 2 π T 1 T 2 Γ ( λ + ( u t ) i ) λ ( λ + ( u t ) i ) d u .
Let α 10 , denote Δ = ( 2 α λ log T ) 1 / 2 , if t [ T 1 + Δ , T 2 Δ ] , then
ω ( t , T 1 , T 2 ) 1 T α .
And if t T 1 Δ , or t T 2 + Δ , then
ω ( t , T 1 , T 2 ) T α .
Proof. 
It is familiar that
e λ = 1 2 π i ( c ) Γ ( s ) λ s d s .
Hence,
1 ω ( t , T 1 , T 2 ) = R 1 + R 2 ,
where
R 1 = e λ 2 π i T 2 Γ ( λ + ( u t ) i ) λ ( λ + ( u t ) i ) d u ,
R 2 = e λ 2 π i T 1 Γ ( λ + ( u t ) i ) λ ( λ + ( u t ) i ) d u ,
By Stirling’s formula, it has
Re ( log Γ ( λ + ( u t ) i ) ) = λ 1 2 log ( λ 2 + ( u t ) 2 ) 2 λ + 1 2 log ( 2 π ) ( u t ) arctan u t λ + O ( 1 / λ )
And
Re ( log ( Γ ( λ + ( u t ) i ) ) ) + Re ( log ( λ ( λ + ( u t ) i ) ) ) + λ = ( u t ) 2 4 λ 2 ( u t ) 2 2 λ 1 2 log λ + 1 2 log ( 2 π ) + O ( 1 / λ )
Hence, if t [ T 1 + Δ , T 2 Δ ] , then
R 1 T 2 e λ Γ ( λ + ( u t ) i ) λ ( λ + ( u t ) i ) d u λ 1 / 2 T 2 exp ( ( u t ) 2 / 2 λ ) d u T α .
and similarly
R 2 λ 1 / 2 T 1 exp ( ( u t ) 2 / 2 λ ) d u T α .
if t T 1 Δ , or t T 2 + Δ , then
ω ( t , T 1 , T 2 ) λ 1 / 2 T 1 T 2 exp ( ( u t ) 2 / 2 λ ) d u T α .
Lemma 2.2
( Estermann [ 2 ] ). Let
S x , h k = n = 1 d ( n ) e n h k e 2 π i n x
Denote by
D s , h k = n = 1 d ( n ) e n h k n s
write z = 2 π i x , then for Im x > 0 , Re s > 1 , k 1 , then
S x , h k = 1 z k * ( γ log z 2 log k * ) + D 0 , h * k * i ( c ) ( 2 π ) s Γ ( s ) k * 2 s 1 sin π s D s , h * ¯ k * + ( c o s π s ) D s , h * ¯ k * z s 1 d s
where 1 < c < 2 , and it has
D 0 , h * k * k * ( log 2 k * ) 2
As usual, denote
χ ( 1 s ) = 2 ( 2 π ) s Γ ( s ) cos ( π s / 2 ) .
Lemma 2.3.
Suppose that 1 < c < 2 , let
J ( y ) = 1 2 π i ( c ) Γ ( s 1 s ) λ ( s 1 s ) χ ( 1 s ) y s d s ,
then
J ( y ) = 0 v s 1 e λ v ( e 2 π i y v + e 2 π i y v ) d v v .
Proof. 
J ( y ) = 1 2 π i ( c ) Γ ( s 1 s ) λ ( s 1 s ) Γ ( s ) ( ( 2 π i y ) s + ( 2 π i y ) s ) d s = J 1 + J 2
where
J 1 = 1 2 π i ( c ) Γ ( s 1 s ) λ ( s 1 s ) Γ ( s ) ( 2 π i y ) s d s , J 2 = 1 2 π i ( c ) Γ ( s 1 s ) λ ( s 1 s ) Γ ( s ) ( 2 π i y ) s d s .
By the theory of Mellin Transforms(refer to [7])
1 2 π i ( c ) G ( s ) F ( s ) x s d s = 0 f 1 v g ( x v ) d v v
where
f ( v ) = 1 2 π i ( c ) F ( x ) x s d s , g ( v ) = 1 2 π i ( c ) G ( x ) x s d s .
We apply this formula to J 1 and J 2 respectively.
For J 1 ,
f ( v ) = λ s 1 2 π i ( c ) Γ ( s 1 s ) ( v / λ ) s d s = λ s 1 2 π i ( λ ) Γ ( w ) ( v / λ ) ( s 1 w ) d w = λ s 1 ( v / λ ) s 1 e λ / v = v s 1 e λ / v ,
i.e.
f ( 1 / v ) = v s 1 e λ v .
And
g ( v ) = 1 2 π i ( c ) Γ ( s ) v s d s = e v ,
i.e.
g ( x v ) = e x v .
So
J 1 = 0 v s 1 e ( λ v + 2 π i y v ) d v v
Similarly,
J 2 = 0 v s 1 e ( λ v 2 π i y v ) d v v
Denote by L δ the straight line from 0 to e i δ .
Lemma 2.4.
Let s 1 = λ + 1 / 2 + i u , 0 δ π / 2 , then
L δ v s 1 e λ v v ( v 1 ) d v L δ v s 1 e λ v v ( v 1 ) d v = e λ 2 π i ,
And let
K = L δ v s 1 e λ v log ( i ( v 1 ) ) v ( v 1 ) d v L δ v s 1 e λ v log ( i ( v 1 ) ) v ( v 1 ) d v ,
then
K = e λ 2 π i ( log u + c 0 + c 2 u 2 ) .
where c 0 and c 2 are two constants.
Proof. 
Equality (2.6) is followed by the residue theorem for the two integral paths form a contour with a pole at v = 1 .
To prove (2.7), we change the integral paths L δ and L δ to the positive real axis but with an indentation around v = 1 with Im v > 0 and Im v < 0 respectively.
And let v = e x , then
K = L + exp ( λ x + i x u λ e x ) log ( i ( e x 1 ) ) 2 sinh ( x / 2 ) d x L exp ( λ x + i x u λ e x ) log ( i ( e x 1 ) ) 2 sinh ( x / 2 ) d x
where the integral path L + is from to ϵ then along a upper semicircle C ϵ + to + ϵ and tend to + . The integral path L is same but with a lower semicircle C ϵ . Let C ϵ be the union of C ϵ and the reversal of C ϵ + .
Then
K = π i ϵ exp ( λ x + i x u λ e x ) 2 sinh ( x / 2 ) d x π i + ϵ + exp ( λ x + i x u λ e x ) 2 sinh ( x / 2 ) d x C ϵ exp ( λ x + i x u λ e x ) log ( i ( e x 1 ) ) 2 sinh ( x / 2 ) d x π i 2 C ϵ + C ϵ exp ( λ x + i x u λ e x ) 2 sinh ( x / 2 ) d x
And then
K = C ϵ e λ log x x d x R ϵ + π i ϵ e λ e i u x x d x + π i P π i + ϵ + e λ e i u x x d x π i P + π i 2 R ϵ
where
R ϵ = C ϵ + C ϵ exp ( λ x + i x u λ e x ) 2 sinh ( x / 2 ) d x R ϵ = C ϵ exp ( λ x + i x u λ e x ) log ( i ( e x 1 ) ) 2 sinh ( x / 2 ) e λ log x x d x P = ϵ e i u x exp ( λ x λ e x ) 2 sinh ( x / 2 ) e λ e i u x x d x P + = + ϵ + e i u x exp ( λ x λ e x ) 2 sinh ( x / 2 ) e λ e i u x x d x
It is easy to know that R ϵ and R ϵ 0 as ϵ 0 . Let P = lim ϵ 0 ( P P + ) , then
K = e λ 2 π i ( log u + c 0 + P )
where
c 0 = 0 1 1 cos x x d x 1 c o s x x d x , P = 1 2 0 cos ( u x ) 1 x exp ( λ x λ ( e x 1 ) ) 2 sinh ( x / 2 ) d x + 1 2 0 cos ( u x ) 1 x exp ( λ x λ ( e x 1 ) ) 2 sinh ( x / 2 ) d x + i 2 0 sin ( u x ) exp ( λ x λ ( e x 1 ) ) 2 sinh ( x / 2 ) 1 x d x i 2 0 sin ( u x ) exp ( λ x λ ( e x 1 ) ) 2 sinh ( x / 2 ) 1 x d x
By the partial integration, P can be expressed as
P = 2 n < N c n u n + O ( u N ) .
where N is a any positive integer.
The following Lemma is direct.
Lemma 2.5.
0 v s 1 1 e λ x d x = λ s 1 0 x s 1 1 e x d x = λ s 1 Γ ( s 1 )
Lemma 2.6.
Let c = 1 + η , 0 < η < 1 , s 1 = λ + 1 / 2 + i u , T u 2 T , δ = 1 / T , and let
W = L δ ( c ) e λ v s 1 e λ v ( 1 + | s | ) Γ ( s ) sin π s ( cos π s ) e π i s / 2 ( v 1 ) s 1 d s d v v .
Then
W ϵ , η T ϵ ( log T / T ) c .
and the estimation is also holds in the cases the term c o s π s is removed, or L δ is replaced by L δ and e π i s / 2 by e π i s / 2 .
Proof. 
Let v = x e i δ , then
| Γ ( s ) | ( 1 + | t | ) c 1 / 2 e π | t | / 2 ,
cos π s sin π s 1 ,
| e π s / 2 | = e π t / 2 ,
| v s 1 | x λ + 1 / 2 ,
1 / | v | 1 / x ,
| e λ v | = e λ x ,
where λ = λ cos δ , and clearly, λ T ϵ / 2 < λ < λ .
And
| ( v 1 ) s 1 | = | v 1 | c 1 e t arg ( v 1 ) = a ( x , δ ) exp ( t b ( x , δ ) )
where
a ( x , δ ) = ( ( x a ) 2 + 2 x ( 1 cos δ ) ) ( c 1 ) / 2
b ( x , δ ) = arctan x sin δ x cos δ 1
Suppose that β 10 , let θ = ( 2 β log T / λ ) 1 / 2 , there are the following estimates
a ( x , δ ) 1 , i f x 1 θ , θ c 1 , i f | x 1 | θ , x c 1 , i f x 1 + θ .
and
b ( x , δ ) > δ , x > 0 , θ 1 δ , | x 1 | θ , π C θ 1 δ , | x 1 | θ , π C x δ , 0 < x 1 θ , δ θ 1 , x 1 + θ .
where C is a constant. Hence
( 1 + | t | ) c + 1 / 2 e π ( t | t | ) / 2 e t b ( x , δ ) d t δ c 3 / 2 , x 1 + θ , ( θ δ 1 ) c + 3 / 2 , | x 1 | θ , ( δ x ) c 3 / 2 , 0 < x 1 θ .
Divide the x integral into three pieces W 1 , W 2 and W 3 with 0 x 1 θ , 1 θ x 1 + θ and 1 + θ x respectively, then there are
W = W 1 + W 2 + W 3 . W 1 δ 5 / 2 η e λ 0 1 θ x λ c 2 e λ x d x δ 5 / 2 η T β W 2 T ϵ θ c W 3 δ 5 / 2 η e λ 1 + θ x λ + c 3 / 2 e λ x d x δ 5 / 2 η T β
Lemma 2.7.
Let s 1 = λ + 1 / 2 + i u , define
g ( u ) = e λ 2 π i ( 1 / 2 ) Γ ( s 1 s ) λ s 1 + s ζ ( s ) χ ( 1 s ) A ¯ ( s ) A ( 1 s ) d s .
Then for T u 2 T , there is
g ( u ) = 1 h , k M a ( h ) a ( k ) ¯ h k ( h , k ) log u ( h , k ) 2 2 π h k + b 0 + O ( u 2 ) + E 0
where E 0 V T c + ϵ , and b 0 is a constant.
Proof. 
We move the integral path from ( 1 / 2 ) to ( c ) , c = 1 + η , 0 < η < 1 , the residue at s = 1 is
R = e λ Γ ( λ 1 / 2 i u ) λ ( λ 1 / 2 i u ) ζ ( 0 ) A ( 1 ) A ¯ ( 0 ) M 1 + ϵ T α T α + 1
Hence,
g ( u ) = g ^ ( u ) R
where g ^ ( u ) is the one of g ( u ) moved in the new integral path.
g ^ ( u ) = 1 h , k M a ( h ) a ( k ) ¯ k e λ n = 1 d ( n ) J ( n h k ) = 1 h , k M a ( h ) a ( k ) ¯ k n = 1 d ( n ) e λ 0 v s 1 e λ v ( e 2 π i n h / k + e 2 π i n h / k ) d v v = 1 h , k M a ( h ) a ( k ) ¯ k ( I 1 + I 2 )
where
I 1 = M 1 + R 1 i E 1 , I 2 = M 2 + R 2 i E 2 .
M 1 = e λ L δ v s 1 e λ v γ log ( 2 π i ( v 1 ) h / k ) 2 log k * 2 π i h * d v v R 1 = D 0 , h * k * e λ L δ v s 1 e λ v d v v E 1 = e λ L δ v s 1 e λ v F 1 ( v ) d v v F 1 ( v ) = ( c ) ( 2 π ) s Γ ( s ) sin π s D s , h * k * + ( cos π s ) D s , h * ¯ k * · ( 2 π i h * k * ( v 1 ) ) s 1 d s
And
M 2 = e λ L δ v s 1 e λ v γ log ( 2 π i ( v 1 ) h / k ) 2 log k * 2 π i h * d v v R 2 = D 0 , h * k * e λ L δ v s 1 e λ v d v v E 2 = e λ L δ v s 1 e λ v F 2 ( v ) d v v F 2 ( v ) = ( c ) ( 2 π ) s Γ ( s ) sin π s D s , h * k * + ( cos π s ) D s , h * ¯ k * · ( 2 π i h * k * ( v 1 ) ) s 1 d s
By Lemma 2.4,
M 1 + M 2 = 1 h * log u ( h , k ) 2 2 π h k + γ + c 0 + O ( u 2 )
and by Lemma 2.5
R 1 + R 2 Re D 0 , h * k * T α M ( log M ) 2 T α T α + 1
So
1 h , k M a ( h ) a ( k ) ¯ k ( R 1 + R 2 ) M 1 + 2 ϵ log M T α + 1 T α + 2
The other error term is
1 h , k M a ( h ) a ( k ) ¯ k ( E 1 + E 2 ) ,
which may be written as a sum of four terms of a typical one is that
Z = L δ ( c ) G ( v , s 1 , s ) M ( s ) d s d v
where
G ( v , s 1 , s ) = e λ v s 1 e λ v Γ ( s ) sin π s ( 2 π ) 2 s ( 2 π i ( v 1 ) ) s 1 v 1 ,
M ( s ) = n = 1 d ( n ) n s h , k M a ( h ) a ( k ) ¯ ( h k ) 1 s ( h , k ) ( 1 2 s ) e n h * ¯ k * .
By Lemma 2.6,
Z V T ϵ θ c .
The Proof of Theorem 1.1.
Proof. 
By Lemma 2.1, it has
I ( T , 2 T ) = T 2 T ω ( t , T Δ , 2 T + Δ ) | ζ A ( 1 / 2 + i t ) | 2 d t + o ( 1 ) = e λ 2 π T 2 T T Δ 2 T + Δ Γ ( λ + ( u t ) i ) λ ( λ + ( u t ) i ) d u | ζ A ( 1 / 2 + i t ) | 2 d t + o ( 1 ) = e λ 2 π T Δ 2 T + Δ T 2 T Γ ( λ + ( u t ) i ) λ ( λ + ( u t ) i ) | ζ A ( 1 / 2 + i t ) | 2 d t d u + o ( 1 ) T Δ 2 T + Δ g ( u ) d u + o ( 1 ) = T Δ 2 T + Δ 1 h , k M a ( h ) a ( k ) ¯ h k ( h , k ) log u ( h , k ) 2 2 π h k + b 0 + O ( u 2 ) + O ( E 0 T )
= T 1 h , k M a ( h ) a ( k ) ¯ h k ( h , k ) log T ( h , k ) 2 2 π h k + b 0 1 + 2 log 2 + O ( E 0 T ) + O Δ log T 1 h , k M a ( h ) a ( k ) ¯ h k ( h , k )
and
1 h , k M ( h , k ) h k 1 t M t h M t | h h 1 2 log 3 M
that is, the last term is O ( Δ log 4 M M 2 ϵ ) .
Then replacing T by T / 2 k , 1 k log T , and summing, it follows
I T h , k M a ( h ) h a ( k ) ¯ k ( h , k ) log T ( h , k ) 2 2 π h k + b 0 1 + E ˜
On the other hand, similarly it has
I ( T , 2 T ) = T 2 T ω ( t , T + Δ , 2 T Δ ) | ζ A ( 1 / 2 + i t ) | 2 d t + o ( 1 ) = e λ 2 π T 2 T T + Δ 2 T Δ Γ ( λ + ( u t ) i ) λ ( λ + ( u t ) i ) d u | ζ A ( 1 / 2 + i t ) | 2 d t + o ( 1 ) = e λ 2 π T + Δ 2 T Δ T 2 T Γ ( λ + ( u t ) i ) λ ( λ + ( u t ) i ) | ζ A ( 1 / 2 + i t ) | 2 d t d u + o ( 1 ) T + Δ 2 T Δ g ( u ) d u + o ( 1 )
Hence, it will lead same bound as the upper bound above. As for the fact b 0 = 2 γ , it may be followed from the known result of Ingham [5]. □
It should be mentioned that the integrant of ω ( t , T 1 , T 2 ) in general is not as the one of w ( t , T 1 , T 2 ) a positive real number, nevertheless, its argument is about t u 2 λ ( t u ) 3 6 λ 2 , which is very small in the context, so can be approximately viewed as a positive real number, and the deduction above is valid as the one in [1].

Acknowledgments

The article was communicated with Prof. Heath-Brown and Prof. Conrey, and they gave me some good advisements.

References

  1. R. Balasubramanian, J.B. Conrey and D.R. Heath-Brown, Asymptotic mean square of the product of the Riemann zeta-function and a Dirichlet polynomial, J. reine angew. Math. 357 (1985), 161-181.
  2. J.-M. Deshouillers and H. Iwaniec, Kloosterman sums and Fourier coefficients of cusp forms, Invent. Math. 70 (1982), 219-288.
  3. T. Estermann, On the representation of a number as the sum of two products. Proc. London Math. Soc. (2) 31 (1930), 123-133.
  4. D.R. Heath-Brown, The fourth power moment of the Riemann zeta-function, Proc. London Math. Soc. 3 (38) (1979), 385-422.
  5. A.E. Ingham, Mean-Value theorems in the theory of the Riemann zeta-function, Proc. London Math. Soc. (2) 27 (1926) 273-300.
  6. H. Iwaniec, On mean values for Dirichlet’s polynomials and Riemann zeta-function, J. London Math. Soc. (2) 22, (1980), 39-45.
  7. E.C. Titchmarsh, Introduction to the Theory of Fourier Integrals, Oxford, 1937.
  8. E.C. Titchmarsh, The Theory of the Riemann Zeta-Function, Oxford, 1986.
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.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.
Prerpints.org logo

Preprints.org is a free preprint server supported by MDPI in Basel, Switzerland.

Subscribe

Disclaimer

Terms of Use

Privacy Policy

Privacy Settings

© 2025 MDPI (Basel, Switzerland) unless otherwise stated