Next Article in Journal
Characterization of Monotone Sequences of Positive Numbers Prescribed by Means
Previous Article in Journal
Mathematical Models of Critical Soft Error in Synchronous and Self-Timed Pipeline
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Average Size of Ramanujan Sum Associated with Divisor Function

Department of Mathematics, China University of Mining and Technology, Beijing 100083, China
*
Author to whom correspondence should be addressed.
These authors contributed equally to this work.
Mathematics 2025, 13(5), 697; https://doi.org/10.3390/math13050697
Submission received: 17 December 2024 / Revised: 9 February 2025 / Accepted: 17 February 2025 / Published: 21 February 2025

Abstract

:
Let m and n be positive integers, and let c ( m , n ) denote the Ramanujan sum. In this paper, we establish an asymptotic formula for the mean value of c ( m , n ) , which is associated with the divisor function, over both m and n.
MSC:
11H60; 11A25

1. Introduction

The Ramanujan sum function holds significant importance in number theory due to its numerous properties. For any positive integers m and n, the classical Ramanujan sum c ( m , n ) is defined as
c ( m , n ) : = 1 j m gcd ( j , m ) = 1 e j n m = d | gcd ( m , n ) d μ m d ,
where e ( z ) : = e 2 π i z , μ ( x ) is the Möbius function, and gcd ( m , n ) denotes the greatest common factor of m and n.
Many authors have conducted a series of mean value studies on the arithmetic function c ( m , n ) . Suppose that Y X 3 are real numbers, k 1 is a fixed integer, and ζ ( s ) is the Riemann zeta function. Define
C k ( X , Y ) : = 1 n Y 1 m X c ( m , n ) k .
Chan and Kumchev [1] proved that
C 1 ( X , Y ) = Y 1 ζ ( 2 ) X 2 + O ( X Y 1 / 3 log X ) + O ( X 3 Y 1 ) .
For k = 2 , if Y X 2 ( log X ) B , for some fixed B > 0 , it holds that
C 2 ( X , Y ) = Y X 2 2 ζ ( 2 ) + O ( X 4 + X Y log X ) ;
if X Y X 2 ( log X ) B , then
C 2 ( X , Y ) = X 2 Y 2 ζ ( 2 ) 1 + κ ( u ) + O X 2 Y ( log X ) 10 ( X 1 / 2 ) + ( Y / X ) 1 / 2 ,
where u = log X 2 Y and
κ ( u ) = 1 2 π e i t u ζ ( 1 i t ) ( 1 + i t ) 2 ( 1 i t ) ζ ( 1 + i t ) d t .
Hereafter, this question has been studied in number fields F , where F is a finite-degree field extension of the rational number field Q . Let Y > X 3 be sufficiently large real numbers. Suppose that k 1 is a fixed integer. Define
C F , k ( X , Y ) : = 1 N ( n ) Y 1 N ( m ) X c ( m , n ) k .
W. G. Nowak [2] established the main term of C F , 1 ( X , Y ) for quadratic number fields under the condition Y > X δ , where δ > 1973 / 820 . W. Zhai [3] obtained a more precise result by deriving an asymptotic formula for C F , 1 ( X , Y ) in the case of quadratic number fields. Furthermore, W. Zhai [4] provided the asymptotic formula of C F , 2 ( X , Y ) in quadratic number fields. The most recent development is due to C. Sneha and G. Shivani [5] establishing more precise results for C F , 2 ( X , Y ) on the quadratic and cubic fields, as well as results for general number fields. For related works involving Ramanujan sums in certain number fields, one may refer to [6,7].
“Sums of sums” is a highly interesting topic that has attracted the attention of numerous researchers. Various functions have been studied in this context, including generalized Ramanujan sums (see [8,9,10]), sums involving the von Mangoldt function (see [11]), sums related to the Liouville function (see [12]), and sums with weighting functions (see [13]).
As is well known, the divisor function plays a fundamental and important role in number theory. For k 2 , let d k ( n ) denote the number of ways that n can be expressed as a product of k factors. When k = 2 , d 2 ( n ) is the familiar Dirichlet divisor function. There have been a series of advancements in the study of d 2 ( n ) . Using the hyperbola method, Dirichlet proved that
n x d 2 ( n ) = x log x + ( 2 γ 1 ) x + O ( x 1 2 ) ,
for x 1 , where γ is Euler’s constant. We denote
Δ ( x ) = n x d 2 ( n ) x log x + ( 2 γ 1 ) x .
It follows that Δ ( x ) = O ( x 1 2 ) . The exponent 1 2 has been improved by many authors (for example, [14,15,16,17]). Currently, the best result, due to Huxley [18], is
Δ ( x ) x 131 416 ( log x ) 26497 8320 .
Furthermore, many other properties have been extensively studied, including the omega results ([19,20]), sign changes ([21]) and moments related to Δ ( x ) ([22,23,24]). Results concerning d k ( n ) for k 3 are presented in [25]. In this paper, we combine the Ramanujan sum with the divisor function. We rewrite (2) in the form of
C k ( X , Y ) = 1 n Y j = 1 k 1 m j X c ( m j , n ) .
We multiply the variables m 1 , m 2 , , m k together, then we denote, for k 2 ,
D k ( X , Y ) : = 1 n Y 1 m 1 m 2 m k X j = 1 k c ( m j , n ) .
For the case k = 2 , we obtain the following theorem.
Theorem 1.
Suppose Y > X 3 , and let L 1 : = log X and L 2 : = log Y . For X L 1 6 L 2 2 < Y < X 3 2 L 1 1 , we have
D 2 ( X , Y ) = X Y 2 ζ ( 2 ) ζ ( 0 ) X 2 L 1 2 ζ ( 2 ) + ζ ( 0 ) X 2 L 2 ζ 2 ( 2 ) + X 2 ζ 2 ( 2 ) ζ ( 0 ) + ζ ( 0 ) + ζ ( 2 ) ζ ( 0 ) ζ ( 2 ) + O X 1 2 Y L 1 + X 5 2 Y 1 2 L 1 3 L 2 .
For Y X 3 2 L 1 1 , we have
D 2 ( X , Y ) = X Y 2 ζ ( 2 ) + ζ ( 0 ) X 2 L 1 2 ζ ( 2 ) ζ ( 0 ) 1 + ζ ( 2 ) X 2 L 2 2 ζ 2 ( 2 ) + O ( X 1 2 Y L 1 ) .
Notations. We use N , Q , R , and C to denote the sets of positive integers, rational numbers, real numbers, and complex numbers, respectively. We use expressions f = O ( g ) or f g to mean | f | C g for some constant C > 0 . We write f g to indicate that f g and g f . The greatest common divisor of m and n is denoted by ( m , n ) and their least common multiple is denoted by [ m , n ] . Let s = σ + i t C , where σ , t R . Throughout this paper, ε denotes a sufficiently small real positive number, not necessarily the same at each occurrence.

2. Analytic Continuation

The following lemmas form the foundation for the proof of Theorem 1.
Lemma 1.
Let k 2 be a fixed integer. Suppose that f ( n 1 , , n k ) is a multivariable arithmetic function whose Dirichlet series is given by
F ( s 1 , , s k ) = n 1 = 1 n k = 1 f ( n 1 , , n k ) n 1 s 1 n k s k
which is absolutely convergent for σ j > μ j > 0   ( j = 1 , , k ) . Assume that x j , T j 5   ( j = 1 , , k ) are parameters satisfying x j N + 1 2   ( j = 1 , , k ) and define
b j = μ j + 1 log x j , ( j = 1 , , k ) .
Then, we have
n 1 x 1 n k x k f ( n 1 , , n k ) = 1 ( 2 π i ) k b 1 i T 1 b 1 + i T 1 b k i T k b k + i T k F ( s 1 , , s k ) x 1 s 1 x k s k s 1 s k d s k d s 1 + O ( x 1 b 1 x k b k E ) ,
where
E : = j = 1 k n 1 = 1 n k = 1 | f ( n 1 , , n k ) | n 1 b 1 n k b k × 1 T j | log x j n j | + 1 .
Proof. 
See Lemma 2.6 in [4]. □
Taking
f ( n , m 1 , m 2 , , m k ) = i = 1 k 1 m 1 m 2 m k X c ( m i , n ) ,
its Dirichlet series is given by
F k ( ω , s ) = n = 1 m 1 = 1 m k = 1 f ( n , m 1 , m 2 , , m k ) n ω ( m 1 m 2 m k ) s ,
where s, ω C .
For any s C , we define
σ s ( n ) = d | n d s .
For s > 1 , we derive
m = 1 c ( m , n ) m s = m = 1 1 m s d | m , d | n d μ m d = d | n d t = 1 1 ( d t ) s μ ( t ) = σ 1 s ( n ) ζ ( s ) ,
which implies
F k ( ω , s ) = 1 ζ k ( s ) n = 1 σ 1 s k ( n ) n ω .
Proposition 1.
Suppose s, ω C . If ω > max { 1 , 2 2 s , 3 2 s } and s > 1 , then
F 2 ( ω , s ) = ζ ( ω ) ζ 2 ( ω + s 1 ) ζ ( ω + 2 s 2 ) ζ 2 ( s ) ζ ( 2 ω + 2 s 2 ) .
Proof. 
Let s 1 , s 2 C , such that s 1 > 1 , s 2 > 1 . Also, assume that ω > max { 1 , 2 s 1 , 2 s 2 , 3 s 1 s 2 } . Then, we define the function F 2 ( ω , s 1 , s 2 ) as follows
F 2 ( ω , s 1 , s 2 ) = 1 ζ ( s 1 ) ζ ( s 2 ) n = 1 σ 1 s 1 ( n ) σ 1 s 2 ( n ) n ω .
By applying Ramanujan’s identity (refer to [26] for details), we can transform the series on the right-hand side. Specifically, we obtain
n = 1 σ 1 s 1 ( n ) σ 1 s 2 ( n ) n ω = ζ ( ω ) ζ ( ω + s 1 1 ) ζ ( ω + s 2 1 ) ζ ( ω + s 1 + s 2 2 ) ζ ( 2 ω + s 1 + s 2 2 ) .
Finally, when we set s 1 = s 2 = s , we arrive at the result that concludes the proof. □
We also require some upper bound estimations of the Riemann zeta function.
Lemma 2.
For σ, t R , the following estimates hold:
If σ 0 ,
ζ ( σ + i t ) ( | t | + 2 ) 1 2 σ log ( | t | + 2 ) .
If 0 σ 1 2 ,
ζ ( σ + i t ) ( | t | + 2 ) 1 2 2 3 σ log ( | t | + 2 ) .
If 1 2 σ 1 ,
ζ ( σ + i t ) ( | t | + 2 ) 1 σ 3 log ( | t | + 2 ) .
If σ 1 ,
ζ ( σ + i t ) min 1 σ 1 , log ( | t | + 2 ) ,
1 ζ ( σ + i t ) min 1 σ 1 , log ( | t | + 2 ) ,
ζ ( k ) ( σ + i t ) min σ ( 1 σ ) k , log k + 1 ( | t | + 2 ) .
Proof. 
The detailed proofs can be found in Chapter 3 of the work by E. C. Titchmarsh [26] and Chapter 7 of the work by Pan and Pan [27]. □
Lemma 3.
Suppose V > 10 is a large parameter and | t 1 / 2 | log 1 V . Then,
V V ζ ( σ + i t ) 4 d t V log 4 V ,
V V ζ ( σ + i t ) 2 d t V log V .
Proof. 
See Chapter 25 of Pan and Pan [27]. □
Lemma 4.
Suppose that T and U are two real large numbers, where U is a half integer and g ( n ) is an arithmetic function satisfying g ( n ) n ε for any ε > 0 . Define
G ( s ) : = n = 1 g ( n ) n s , s C ,
which is absolutely convergent for s > 1 . For 1 < σ < 2 , define
E ( U , T ; σ ) = n = 1 g ( n ) n σ × 1 T | log U n | + 1 .
Then, we have
E ( U , T ; σ ) G ( σ ) U ε T .
Proof. 
We express the function E ( U , T ; σ ) as the sum of three components:
E ( U , T ; σ ) = E 1 ( U , T ; σ ) + E 2 ( U , T ; σ ) + E 3 ( U , T ; σ ) ,
where
E 1 ( U , T ; σ ) = n U / 2 g ( n ) n σ × 1 T | log U n | + 1 , E 2 ( U , T ; σ ) = U / 2 < n 2 U g ( n ) n σ × 1 T | log U n | + 1 , E 3 ( U , T ; σ ) = n > 2 U g ( n ) n σ × 1 T | log U n | + 1 .
Trivially, we can obtain the following inequality
E 1 ( U , T ; σ ) + E 3 ( U , T ; σ ) G ( σ ) T .
For the component E 2 ( U , T ; σ ) , we first note that
E 2 ( U , T ; σ ) U ε σ U / 2 < n 2 U 1 T log U n + 1 U ε T ,
By applying a well-known estimate
U / 2 < n 2 U 1 T | log U n | + 1 U log U T ,
we further simplify the inequality for
E 2 ( U , T ; σ ) U ε T .
Combining the results of (13)–(15), we derive the inequality (12). □

3. Proof of Theorem 1

Suppose X and Y are half integers. Recall that
F 2 ( ω , s ) = ζ ( ω ) ζ 2 ( ω + s 1 ) ζ ( ω + 2 s 2 ) ζ 2 ( s ) ζ ( 2 ω + 2 s 2 ) .
Let s = σ 0 + i t 0 , ω = σ 1 + i t 1 , and b 0 = b 1 = b = 1 + 1 log Y . For Y > X 3 , assume T 0 = X 2 , T 1 = 4 Y 2 . Applying Lemma 1, we obtain
D 2 ( X , Y ) = M + O ( X Y E ) ,
where
E = n = 1 m 1 = 1 m 2 = 1 c ( m 1 , n ) c ( m 2 , n ) n b ( m 1 m 2 ) b 1 X 2 log X m 1 m 2 + 1 + 1 4 Y 2 log Y n + 1
and
M = 1 ( 2 π i ) 2 b i T 0 b + i T 0 b i T 1 b + i T 1 g ( ω , s ) d ω d s
with
g ( ω , s ) : = ζ ( ω ) ζ 2 ( ω + s 1 ) ζ ( ω + 2 s 2 ) X s Y ω ζ 2 ( s ) ζ ( 2 ω + 2 s 2 ) ω s .

3.1. Estimation of E

First, we handle E. Let
E = E 1 + E 2 ,
where
E 1 = n = 1 m 1 = 1 m 2 = 1 c ( m 1 , n ) c ( m 2 , n ) n b ( m 1 m 2 ) b 1 X 2 log X m 1 m 2 + 1
and
E 2 = n = 1 m 1 = 1 m 2 = 1 c ( m 1 , n ) c ( m 2 , n ) n b ( m 1 m 2 ) b 1 4 Y 2 log Y n + 1 .
By writing m 1 = d 1 k 1 and m 2 = d 2 k 2 , we obtain
E 1 = n = 1 d 1 , d 2 , k 1 , k 2 = 1 d 1 | n d 1 μ ( k 1 ) d 2 | n d 2 μ ( k 2 ) n b ( d 1 k 1 d 2 k 2 ) b 1 X 2 log X d 1 k 1 d 2 k 2 + 1 .
This can be rewritten as
E 1 = d 1 , d 2 , k 1 , k 2 = 1 μ ( k 1 ) μ ( k 2 ) d 1 d 2 ( d 1 k 1 d 2 k 2 ) b d 1 | n , d 2 | n n = 1 1 n b 1 X 2 log X d 1 k 1 d 2 k 2 + 1 .
Since
d 1 | n , d 2 | n n = 1 1 n b = k = 1 1 ( k [ d 1 , d 2 ] ) b
and we know the well-known relation
d 1 d 2 = [ d 1 , d 2 ] ( d 1 , d 2 ) ,
we can conclude the following inequality:
E 1 ζ ( b ) d 1 , d 2 , k 1 , k 2 = 1 | μ ( k 1 ) μ ( k 2 ) | ( d 1 , d 2 ) ( d 1 k 1 d 2 k 2 ) b 1 X 2 log X d 1 k 1 d 2 k 2 + 1 .
Now, we make a substitution. Let d = ( d 1 , d 2 ) , and write d 1 = l 1 d , d 2 = l 2 d , where ( l 1 , l 2 ) = 1 . Therefore, we have
E 1 ζ ( b ) d , l 1 , l 2 , k 1 , k 2 = 1 ( l 1 , l 2 ) = 1 1 ( k 1 k 2 l 1 l 2 ) b d 2 b 1 1 X 2 log X k 1 k 2 l 1 l 2 d 2 + 1 ζ ( b ) d , l 1 , l 2 , k 1 , k 2 = 1 1 ( k 1 k 2 l 1 l 2 ) b d b 1 X 2 log X k 1 k 2 l 1 l 2 d 2 + 1 ζ ( b ) d , l 1 , l 2 , k 1 , k 2 = 1 1 ( k 1 k 2 l 1 l 2 ) b d b 1 X 2 + E 1 = X 2 L 2 6 + E 1 ,
where
E 1 = ζ ( b ) X / 2 k 1 k 2 l 1 l 2 d 2 X 1 ( k 1 k 2 l 1 l 2 ) b d b 1 X 2 log X k 1 k 2 l 1 l 2 d 2 + 1 .
Next, we turn our attention to handling the term E 1 .
E 1 = ζ ( b ) X / 2 k 1 k 2 l 1 l 2 d 2 X d b ( k 1 k 2 l 1 l 2 ) b d 2 b 1 X 2 log X k 1 k 2 l 1 l 2 d 2 + 1 X 1 2 ζ ( b ) X / 2 k 1 k 2 l 1 l 2 d 2 X 1 ( k 1 k 2 l 1 l 2 ) b d 2 b 1 X 2 log X k 1 k 2 l 1 l 2 d 2 + 1 .
Applying Lemma 4, we have
E 1 X 3 2 + ε L 2 6 .
Further from (16) and (17), we conclude that
E 1 X 3 2 + ε L 2 6 .
Now, we deal with E 2 . Recalling (3), from the estimate (8), we have
m 1 = 1 m 2 = 1 c ( m 1 , n ) c ( m 2 , n ) ( m 1 m 2 ) b σ 1 b 2 ( n ) ζ 2 ( b ) d 2 2 ( n ) L 2 2 .
Then, from the estimate (9), we obtain
E 2 L 2 2 n = 1 d 2 2 ( n ) n b 1 Y 2 log Y n + 1 .
Applying Lemma 4, we conclude that
E 2 Y ε 2 .
From (18) and (19), finally, we obtain
O ( X Y E ) = O ( X ε 1 2 Y L 2 6 ) .

3.2. Treatment for M

In the following, we deal with M. Consider a rectangular domain with vertices at the points ω = b ± i 4 Y 2 and ω = 1 2 ± i 4 Y 2 . Within this domain, the integrand g ( ω , s ) has a second-order pole ω = 2 s and two simple poles ω = 1 and ω = 3 2 s . By applying the residue theorem, we can deduce that
M = 1 2 π i b i X 2 b + i X 2 R 1 + R 2 + R 3 d s + 1 ( 2 π i ) 2 b i X 2 b + i X 2 1 2 i 4 Y 2 1 2 + i 4 Y 2 + 1 2 + i 4 Y 2 b + i 4 Y 2 1 2 i 4 Y 2 b i 4 Y 2 g ( ω , s ) d ω d s = 1 2 π i b i X 2 b + i X 2 R 1 + R 2 + R 3 d s + E 01 + E 02 E 03 ,
where
R 1 : = Res ω = 2 s g ( ω , s ) = ζ ( 2 s ) X s Y 2 s s ( 2 s ) ζ ( 2 ) ζ ( s ) 1 2 + L 2 + 1 2 s , R 2 : = Res ω = 1 g ( ω , s ) = ζ ( 2 s 1 ) X s Y s ζ ( 2 s ) , R 3 : = Res ω = 3 2 s g ( ω , s ) = ζ ( 3 2 s ) ζ 2 ( 2 s ) X s Y 3 2 s s ( 3 2 s ) ζ 2 ( s ) ζ ( 4 2 s ) ,
and
E 01 : = 1 ( 2 π i ) 2 b i X 2 b + i X 2 1 2 i 4 Y 2 1 2 + i 4 Y 2 g ( ω , s ) d ω d s ,
E 02 : = 1 ( 2 π i ) 2 b i X 2 b + i X 2 1 2 + i 4 Y 2 b + i 4 Y 2 g ( ω , s ) d ω d s ,
E 03 : = 1 ( 2 π i ) 2 b i X 2 b + i X 2 1 2 i 4 Y 2 b i 4 Y 2 g ( ω , s ) d ω d s .
We start by estimating the quantity E 01 to analyze. For | t 0 | X 2 , | t 1 | 4 Y 2 , according to Lemma 2, we have
g 1 2 + i t 1 , b + i t 0 X Y 1 2 ζ 1 2 + i t 1 ζ 1 2 + 1 log Y + i t 1 + i t 0 2 ζ 1 2 + 2 log Y + i t 1 + i 2 t 0 ζ 1 + 1 log Y + i t 0 2 ζ 1 + 2 log Y + i t 1 + i 2 t 0 ( | t 1 | + 1 ) ( | t 0 | + 1 ) X Y 1 2 L 2 3 × ζ 1 2 + i t 1 ζ 1 2 + 1 log Y + i t 1 + i t 0 2 ζ 1 2 + 2 log Y + i t 1 + i 2 t 0 ( | t 1 | + 1 ) ( | t 0 | + 1 ) .
Applying Lemmas 2 and 3, we can deduce
E 01 X Y 1 2 L 2 3 X 2 X 2 4 Y 2 4 Y 2 ζ 1 2 + i t 1 ζ 1 2 + 1 log Y + i t 1 + i t 0 2 ζ 1 2 + 2 log Y + i t 1 + i 2 t 0 ( | t 1 | + 1 ) ( | t 0 | + 1 ) d t 1 d t 0 X Y 1 2 L 2 3 1 + 2 ,
where
1 = X 2 X 2 4 Y 2 4 Y 2 | t 1 | | t 0 | ζ 1 2 + i t 1 ζ 1 2 + 1 log Y + i t 1 + i t 0 2 ζ 1 2 + 2 log Y + i t 1 + i 2 t 0 ( | t 1 | + 1 ) ( | t 0 | + 1 ) d t 1 d t 0
and
2 = X 2 X 2 4 Y 2 4 Y 2 | t 0 | | t 1 | ζ 1 2 + i t 1 ζ 1 2 + 1 log Y + i t 1 + i t 0 2 ζ 1 2 + 2 log Y + i t 1 + i 2 t 0 ( | t 1 | + 1 ) ( | t 0 | + 1 ) d t 1 d t 0 .
Now, we estimate 1 . Let L 1 ( V ) : = 0 V | ζ ( σ i t ) | 4 d t for σ 1 2 1 log V . By Lemma 3 and partial summation, we obtain
T T ζ ( σ + i t ) 4 | t | + 1 d t 10 10 ζ ( σ + i t ) 4 | t | + 1 d t + 10 < | t | < T ζ ( σ + i t ) 4 t d t 1 + 10 T ζ ( σ + i t ) 4 t d t 1 + 10 T L 1 ( t ) t d t 1 + L 1 ( T ) T + 10 T L 1 ( t ) t 2 d t log 4 T 2 T log 4 t t d t log 5 T .
Let L 2 ( V ) : = 0 V ζ ( σ i t ) 2 d t for σ 1 2 1 log V . Similarly, we have
V V ζ ( σ + i t ) 2 | t | + 1 d t 1 + L 2 ( V ) V + 10 V L 2 ( t ) t 2 d t log 2 V .
From (11) and Cauchy’s inequality, we obtain
0 T ζ 1 2 + i t d t T log 1 2 T ,
which, by partial summation, yields
T T ζ 1 2 + i t | t | + 1 d t log 1.5 T .
Note that for the integral 1 , we have
| t 1 + t 0 | + 1 | t 1 | + | t 0 | + 1 2 ( | t 0 | + 1 )
and
| t 1 + 2 t 0 | + 1 | t 1 | + 2 | t | 0 | + 1 3 ( | t 0 | + 1 ) ,
so that
( | t 1 | + 1 ) ( | t 1 + t 0 | + 1 ) 1 2 ( | t 1 + 2 t 0 | + 1 ) 1 2 ( | t 1 | + 1 ) ( | t 0 | + 1 ) 6 .
Thus, from (23) and Cauchy’s inequality, we obtain
1 6 | t 1 | | t 0 | ζ 1 2 + i t 1 ζ 1 2 + 1 log Y + i t 1 + i t 0 2 ζ 1 2 + 2 log Y + i t 1 + i 2 t 0 ( | t 1 | + 1 ) ( | t 1 + t 0 | + 1 ) 1 2 ( | t 1 + 2 t 0 | + 1 ) 1 2 d t 1 d t 0 X 2 X 2 ζ 1 2 + i t 1 ( | t 1 | + 1 ) d t 1 | t 1 | | t 0 | ζ 1 2 + 1 log Y + i t 1 + i t 0 4 ( | t 1 + t 0 | + 1 ) d t 0 1 2 × | t 1 | | t 0 | ζ 1 2 + 2 log Y + i t 1 + i 2 t 0 2 ( | t 1 + 2 t 0 | + 1 ) d t 0 1 2 X 2 X 2 ζ 1 2 + i t 1 ( | t 1 | + 1 ) d t 1 2 X 2 2 X 2 ζ 1 2 + 1 log Y + i y 1 4 ( | y 1 | + 1 ) d y 1 1 2 × 3 X 2 3 X 2 ζ 1 2 + 2 log Y + i y 2 2 ( | y 2 | + 1 ) d y 2 1 2 L 1 1.5 2 Y 2 2 Y 2 ζ 1 2 + 1 log Y + i y 1 4 ( | y 1 | + 1 ) d y 1 1 2 3 Y 2 3 Y 2 ζ 1 2 + 2 log Y + i y 2 2 ( | y 2 | + 1 ) d y 2 1 2 L 1 1.5 L 2 3.5 ,
where we substitute the variables y 1 = t 1 + t 0 and y 2 = t 1 + 2 t 0 in the fourth line.
At present, we estimate 2 . Note that for the integral 2 , we have
| t 1 + t 0 | + 1 | t 1 | + | t 0 | + 1 2 ( | t 1 | + 1 ) , | t 1 + 2 t 0 | + 1 | t 1 | + 2 | t 0 | + 1 3 ( | t 1 | + 1 ) ,
which imply
1 | t 1 | + 1 2 | t 1 + t 0 | + 1 , 1 | t 1 | + 1 3 | t 1 + 2 t 0 | + 1 .
Therefore, through Hölder’s inequality, we have
2 = | t 0 | < | t 1 | ζ 1 2 + i t 1 ζ 1 2 + 1 log Y + i t 1 + i t 0 2 ζ 1 2 + 2 log Y + i t 1 + i 2 t 0 ( | t 1 | + 1 ) ( | t 0 | + 1 ) d t 1 d t 0 2 3 4 X 2 X 2 d t 0 | t 0 | + 1 × | t 0 | < | t 1 | 4 Y 2 ζ 1 2 + i t 1 ζ 1 2 + 1 log Y + i t 1 + i t 0 2 ζ 1 2 + 2 log Y + i t 1 + i 2 t 0 ( | t 1 | + 1 ) 1 4 ( | t 1 + t 0 | + 1 ) 1 2 ( | t 1 + 2 t 0 | + 1 ) 1 4 d t 1 X 2 X 2 d t 0 | t 0 | + 1 4 Y 2 4 Y 2 ζ 1 2 + i t 1 4 ( | t 1 | + 1 ) d t 1 1 4 8 Y 2 8 Y 2 ζ 1 2 + 1 log Y + i y 1 4 ( | y 1 | + 1 ) d y 1 1 2 × 12 Y 2 12 Y 2 ζ 1 2 + 2 log Y + i y 2 4 | y 2 | + 1 d y 2 1 4 L 1 L 2 5 .
Consequently, from (22), (25), and (26), it can be deduced that
E 01 X Y 1 2 L 1 L 2 8 .
Next, we estimate E 02 . In this case, by Lemma 2, for | t 0 | X 2 and 1 2 σ b , we have
g ( σ + i Y 2 , b + i t 0 ) X Y σ ζ σ + i Y 2 ζ 2 σ + 1 log Y + i Y 2 + i t 0 ζ σ + 2 log Y + i Y 2 + i 2 t 0 ζ 2 1 + 1 log Y + i t 0 ζ 2 σ + 2 log Y + i Y 2 + i 2 t 0 Y 2 ( | t 0 | + 2 ) X Y σ Y 2 ( 1 σ ) 3 log Y × ( | Y 2 + t 0 | + 2 ) 2 ( 1 σ ) 3 log 2 ( | Y 2 + t 0 | + 2 ) Y 2 ( | t 0 | + 1 ) × ( | Y 2 + 2 t 0 | + 2 ) 1 σ 3 log ( | Y 2 + 2 t 0 | + 2 ) X Y 2 3 5 3 σ ( | t 0 | + 1 ) 1 L 2 5 ,
so that we obtain
E 02 X Y 2 3 L 2 5 X 2 X 2 1 | t 0 | + 1 d t 0 1 2 b Y 5 3 σ d σ X Y 1 6 L 1 L 2 5 .
Similarly, we have
E 03 X Y 1 6 L 1 L 2 5 .
From (27)–(29), we deduce that
M = 1 2 π i b i X 2 b + i X 2 ( R 1 + R 2 + R 3 ) d s + O ( X Y 1 2 L 1 L 2 8 ) .
Then, we calculate the main terms, respectively. Let
M j : = 1 2 π i b i X 2 b + i X 2 R j d s ( j = 1 , 2 , 3 ) .

3.3. Treatment for M 1

Firstly, we deal with M 1 . Recall that
R 1 = ζ ( 2 s ) X s Y 2 s s ( 2 s ) ζ ( 2 ) ζ ( s ) 1 2 + L 2 + 1 2 s .
We consider the rectangle domain formed by the four points s = b ± i X 2 , s = 5 2 ± i X 2 . In this domain, the integrand R 1 has a 2-order pole s = 2 . By the residue theorem, we have
M 1 = Res s = 2 R 1 1 2 π i 5 2 i X 2 5 2 + i X 2 b i X 2 5 2 i X 2 + b + i X 2 5 2 + i X 2 R 1 d s = Res s = 2 R 1 E 11 + E 12 E 13 ,
where
E 11 : = 1 2 π i 5 2 i X 2 5 2 + i X 2 R 1 d s , E 12 : = 1 2 π i b i X 2 5 2 i X 2 R 1 d s , E 13 : = 1 2 π i b + i X 2 5 2 + i X 2 R 1 d s .
We calculate that
Res s = 2 R 1 = ζ ( 0 ) X 2 L 1 2 ζ ( 2 ) ζ ( 0 ) X 2 L 2 ζ 2 ( 2 ) X 2 ζ 2 ( 2 ) ζ ( 0 ) + ζ ( 0 ) + ζ ( 2 ) ζ ( 0 ) ζ ( 2 ) .
From the estimates (5) and (8), for | t | X 2 , we have
R 1 X 5 2 Y 1 2 L 2 ζ 1 2 i t ( | t | + 1 ) 2 X 5 2 Y 1 2 L 2 log 2 ( | t | + 2 ) ( | t | + 1 ) .
Then we have
E 11 X 5 2 Y 1 2 L 2 X 2 X 2 log 2 ( | t | + 2 ) ( | t | + 1 ) d t X 5 2 Y 1 2 L 1 3 L 2 .
For b σ 5 2 , we have
R 1 ζ ( 2 σ i X 2 ) X σ Y 2 σ X 4 ζ ( σ + i X 2 ) 1 2 + log Y + 1 X 2 X 4 + σ Y 2 σ | ζ ( 2 σ i X 2 ) | L 1 L 2 .
Through Lemma 2, we have
R 1 X 14 3 + 5 σ 3 Y 2 σ L 1 2 L 2 , b < σ 3 2 ; X 17 3 + 7 σ 3 Y 2 σ L 1 2 L 2 , 3 2 < σ 2 ; X 7 + 3 σ Y 2 σ L 1 2 L 2 , 2 < σ 5 2 .
It follows that
E 12 b 3 2 + 3 2 2 + 2 5 2 R 1 d σ L 1 2 L 2 X 2 Y + X 13 6 Y 1 2 + X 1 + X 1 2 Y 1 2 X 2 Y L 1 2 L 2 + X 1 2 Y 1 2 L 1 2 L 2 .
Similarly, we obtain
E 13 X 2 Y L 1 2 L 2 + X 1 2 Y 1 2 L 1 2 L 2 .
Then, from (32)–(34), we obtain
M 1 = ζ ( 0 ) X 2 L 1 2 ζ ( 2 ) + ζ ( 0 ) X 2 L 2 ζ 2 ( 2 ) + X 2 ζ ( 2 ) ζ ( 0 ) + ζ ( 0 ) + ζ ( 2 ) ζ ( 0 ) ζ ( 2 ) + O ( X 2 Y L 1 2 L 2 + X 5 2 Y 1 2 L 1 3 L 2 ) .

3.4. Treatment for M 2

Let
h ( n ) = n = n 1 2 n 2 2 n 1 μ ( n 2 ) ,
then we have
n X h ( n ) = n 2 X 1 2 μ ( n 2 ) n 1 X 1 2 n 2 1 n 1 = n 2 X 1 2 μ ( n 2 ) X 2 n 2 2 + O X 1 2 n 2 = X 2 n 2 X 1 2 μ ( n 2 ) n 2 2 + O X 1 2 n 2 X 1 2 1 n 2 = X 2 ζ ( 2 ) + O ( X 1 2 L 1 ) .
Since
n = 1 h ( n ) n s = n 1 = 1 n 2 = 1 n 1 μ ( n 2 ) ( n 1 n 2 ) 2 s = ζ ( 2 s 1 ) ζ ( 2 s ) ,
by Perron’s formula, we obtain
n X h ( n ) = 1 2 π i b i X 2 b + i X 2 ζ ( 2 s 1 ) X s ζ ( 2 s ) s d s + O X 1 + ε .
Consequently, recalling that
R 2 = ζ ( 2 s 1 ) X s Y s ζ ( 2 s ) ,
collecting (36) and (37), it can be derived that
M 2 = X Y 2 ζ ( 2 ) + O ( X 1 2 Y L 1 ) .

3.5. Treatment for M 3

We write
M 3 = 1 2 π i b i b + i b i b i X 2 b + i X 2 b + i R 3 d s = 1 2 π i b i b + i R 3 d s + O | t | > X 2 | R 3 | d t .
Recalling that
R 3 = ζ ( 3 2 s ) ζ 2 ( 2 s ) X s Y 3 2 s s ( 3 2 s ) ζ 2 ( s ) ζ ( 4 2 s ) ,
when s = b + i t for | t | > X 2 , we obtain by Lemma 2
R 3 X Y | t | 4 3 log Y 2 log 6 | t | .
Then, we have
O | t | > X 2 | R 3 | d t = O X Y | t | > X 2 | t | 4 3 log Y 2 log 6 | t | d t = O ( X 1 Y L 1 6 ) .
Since R 3 is holomorphic in 1 s 3 2 , we obtain
1 2 π i b i b + i R 3 d s = 1 2 π i 3 2 i 3 2 + i R 3 d s .
When s = 3 2 + i t , by Lemma 2, we have
R 3 = ζ ( 2 i t ) ζ 2 1 2 i t X 3 2 3 2 + i t ( 2 i t ) ζ 3 2 + i t ζ ( 1 2 i t ) X 3 2 | t | + 2 7 6 log 6 ( | t | + 2 ) .
It follows that
1 2 π i 3 2 i 3 2 + i R 3 d s X 3 2 .
Consequently, from (39) and (40), we obtain
M 3 = O ( X 3 2 + X 1 Y L 1 6 ) .

4. Conclusions

Collecting the above computations for (20), (30), (35), (38), and (41), we have
D 2 ( X , Y ) = X Y 2 ζ ( 2 ) ζ ( 0 ) X 2 L 1 2 ζ ( 2 ) + ζ ( 0 ) X 2 L 2 ζ 2 ( 2 ) + X 2 ζ 2 ( 2 ) ζ ( 0 ) + ζ ( 0 ) + ζ ( 2 ) ζ ( 0 ) ζ ( 2 ) + O ( X ε 1 2 Y L 2 6 + X Y 1 2 L 1 L 2 8 + X 2 Y L 1 2 L 2 + X 5 2 Y 1 2 L 1 3 L 2 + X 1 2 Y L 1 + X 3 2 + X 1 Y L 1 6 ) = X Y 2 ζ ( 2 ) ζ ( 0 ) X 2 L 1 2 ζ ( 2 ) + ζ ( 0 ) X 2 L 2 ζ 2 ( 2 ) + X 2 ζ 2 ( 2 ) ζ ( 0 ) + ζ ( 0 ) + ζ ( 2 ) ζ ( 0 ) ζ ( 2 ) + O ( X 1 2 Y L 1 + X 5 2 Y 1 2 L 1 3 L 2 ) .
It is evident that the main term must dominate the error term. Thus, the proof of Theorem 1 is complete.

Author Contributions

Conceptualization, W.Z.; methodology, X.L. and W.Z.; writing and editing, X.L.; review, X.L. and W.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Natural Science Foundations of China (Grant numbers 12471009, 12301006) and partially supported by the Beijing Natural Science Foundation (Grant Number 1242003).

Data Availability Statement

No data were used in this study, so there is no issue regarding data availability.

Conflicts of Interest

The authors declare no conflicts of interest. Our second author is the funder.

References

  1. Chan, T.H.; Kumchev, A.V. On sums of Ramanujan sums. Acta Arithm. 2012, 152, 1–10. [Google Scholar] [CrossRef]
  2. Nowak, W.G. The average size of Ramanujan sums over quadratic number fields. Arch. Math. 2012, 99, 433–442. [Google Scholar] [CrossRef]
  3. Zhai, W. The average size of Ramanujan sums over quadratic number fields. Ramanujan J. 2021, 56, 953–969. [Google Scholar] [CrossRef]
  4. Zhai, W. The average size of Ramanujan sums over quadratic number fields (II). arXiv 2019, arXiv:2019.11111v1. [Google Scholar]
  5. Chaubey, S.; Goel, S. On the distribution of Ramanujan sums over number fields. Ramanujan J. 2003, 61, 813–837. [Google Scholar] [CrossRef]
  6. Grytczuk, A. On Ramanujan sums on arithmetical semigroups. Tsukuba J. Math. 1992, 16, 315–319. [Google Scholar] [CrossRef]
  7. Ma, J.; Sun, H.; Zhai, W. The average size of Ramanujan sums over cubic number fields. Period. Math. Hung. 2023, 87, 215–231. [Google Scholar] [CrossRef]
  8. Kühn, P.; Robles, N. Explicit formulas of a generalized Ramanujan sum. Int. J. Number Theory 2016, 12, 383–408. [Google Scholar] [CrossRef]
  9. Robles, N.; Roy, A. Moments of averages of generalized Ramanujan sums. Monatsh. Math. 2017, 182, 433–461. [Google Scholar] [CrossRef]
  10. Robles, N. Twisted Second Moments and Explicit Formulae of the Riemann Zeta-Function. Ph.D. Thesis, University of Zurich, Zurich, Switzerland, 2015. [Google Scholar]
  11. Kiuchi, I.; Takeda, W. On sums of sums involving the von Mangoldt function. Results Math. 2024, 79, 247. [Google Scholar] [CrossRef]
  12. Kiuchi, I. On sums of sums involving the Liouville function. Funct. Approx. Comment. Math. 2024, 70, 245–262. [Google Scholar] [CrossRef]
  13. Kiuchi, I. Sums of logarithmic weights involving r-full numbers. Ramanujan J. 2024, 64, 1045–1059. [Google Scholar] [CrossRef]
  14. Huxley, M.N. Exponential sums and lattice points. II. Proc. Lond. Math. Soc. 1993, 66, 279–301. [Google Scholar] [CrossRef]
  15. Kolesnik, G. On the order of ζ( 1 2 +it) and Δ(R). Pac. J. Math. 1982, 98, 107–122. [Google Scholar] [CrossRef]
  16. van der Corput, J.G. Zum Teilerproblem. Math. Ann. 1928, 98, 679–716. [Google Scholar] [CrossRef]
  17. Voronoi, G. Sur une fonction transcendante et ses applications à la sommation de quelques séries. Ann. Sci. E´cole Norm. Sup. 1904, 21, 207–267. [Google Scholar] [CrossRef]
  18. Huxley, M.N. Exponential sums and lattice points. III. Proc. Lond. Math. Soc. 2003, 87, 591–609. [Google Scholar] [CrossRef]
  19. Harfner, J.L. New omega theorems for two classical lattice point problem. Inve. Math. 1981, 63, 181–186. [Google Scholar] [CrossRef]
  20. Soundararajan, K. Omega results for the divisor and circle problems. Int. Math. Res. Not. 2003, 2003, 1987–1998. [Google Scholar] [CrossRef]
  21. Heath-Brown, D.R.; Tsang, K.M. Sign changes of E(T), Δ(x), P(x). J. Number Theory 1994, 49, 73–83. [Google Scholar] [CrossRef]
  22. Tong, K.C. On divisor problem III. Acta Math. Sin. 1956, 6, 515–541. [Google Scholar]
  23. Tsang, K.M. Higher-power moments of Δ(x), E(t) and P(x). Proc. Lond. Math. Soc. 1992, 65, 65–84. [Google Scholar] [CrossRef]
  24. Zhai, W. On higher-power moments of Δ(x). Acta Arith. 2004, 112, 367–395. [Google Scholar] [CrossRef]
  25. Fouvry, É.; Iwaniec, H. The divisor function over arithmetic progressions. Acta Arith. 1992, 61, 271–287. [Google Scholar] [CrossRef]
  26. Titchmarsh, E.C. The Theory of the Riemann Zeta-Function; Oxford University Press: London, UK, 1986. [Google Scholar]
  27. Pan, C.; Pan, C. Foundations of Analytic Number Theory; Science Press: Bejing, China, 1990; pp. 149–151. [Google Scholar]
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

Li, X.; Zhai, W. Average Size of Ramanujan Sum Associated with Divisor Function. Mathematics 2025, 13, 697. https://doi.org/10.3390/math13050697

AMA Style

Li X, Zhai W. Average Size of Ramanujan Sum Associated with Divisor Function. Mathematics. 2025; 13(5):697. https://doi.org/10.3390/math13050697

Chicago/Turabian Style

Li, Xin, and Wenguang Zhai. 2025. "Average Size of Ramanujan Sum Associated with Divisor Function" Mathematics 13, no. 5: 697. https://doi.org/10.3390/math13050697

APA Style

Li, X., & Zhai, W. (2025). Average Size of Ramanujan Sum Associated with Divisor Function. Mathematics, 13(5), 697. https://doi.org/10.3390/math13050697

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