Accepted Manuscript International Journal of Number Theory

Accepted manuscript to appear in IJNT
Accepted Manuscript
International Journal of Number Theory
by UNIVERSITY OF WARWICK on 07/05/19. Re-use and distribution is strictly not permitted, except for Open Access articles.
Article Title: On certain sums concerning the gcd’s and lcm’s of k Positive Integers
Author(s): Titus Hilberdink, Florian Luca, Laszlo Toth
DOI: 10.1142/S1793042120500049
Received: 28 May 2018
Accepted: 19 June 2019

Int. J. Number Theory Downloaded from www.worldscientific.com
To be cited as: Titus Hilberdink, Florian Luca, Laszlo Toth, On certain sums concerning
the gcd’s and lcm’s of k Positive Integers, International Journal of Number
Theory, doi: 10.1142/S1793042120500049
Link to final version: https://doi.org/10.1142/S1793042120500049
This is an unedited version of the accepted manuscript scheduled for publication. It has been uploaded
in advance for the benefit of our customers. The manuscript will be copyedited, typeset and proofread
before it is released in the final form. As a result, the published copy may differ from the unedited
version. Readers should obtain the final version from the above link when it is published. The authors
are responsible for the content of this Accepted Article.
T
IP
June 17, 2019 10:50 WSPC/INSTRUCTION FILE
Sums˙gcd˙lcm˙2019˙June
CR
International Journal of Number Theory
c World Scientific Publishing Company
US
On certain sums concerning the gcd’s and lcm’s of k positive integers
Titus Hilberdink
Department of Mathematics, University of Reading, Whiteknights
PO Box 220, Reading RG6 6AX, UK
AN
t.w.hilberdink@reading.ac.uk
Florian Luca
School of Mathematics, University of the Witwatersrand
Private Bag X3, WITS 2050, Johannesburg, South Africa
and
Research Group in Algebraic Structures and Applications
King Abdulaziz University, Jeddah, Saudi Arabia
and
DM
Department of Mathematics, Faculty of Sciences, University of Ostrava

30 dubna 22, 701 03 Ostrava 1, Czech Republic
florian.luca@wits.ac.za
László Tóth
Department of Mathematics, University of Pécs
Ifjúság útja 6, H-7624 Pécs, Hungary
ltoth@gamma.ttk.pte.hu
Received (Day Month Year)

Accepted (Day Month Year)
TE
We use elementary arguments to prove results on the order of magnitude of certain sums
concerning the gcd’s and lcm’s of k positive integers, where k ≥ 2 is fixed. We refine and
generalize an asymptotic formula of Bordellès (2007), and extend certain related results
of Hilberdink and Tóth (2016). We also formulate some conjectures and open problems.
Keywords: greatest common divisor; least common multiple; gcd-sum function; lcm-sum
function; asymptotic formula; order of magnitude
EP
Mathematics Subject Classification 2010: 11A25, 11N37
1. Introduction
Consider the gcd-sum function
C
n
X X
G(n) := (k, n) = dϕ(n/d) (n ∈ N),
k=1 d|n
AC
1
T
IP
CR
2 T. Hilberdink, F. Luca, L. Tóth
where ϕ(n) is Euler’s totient function. The function G(n) is multiplicative and the
asymptotic formula
X
x2 1 ζ 0 (2)
+ O(x1+θ+ε ),
US
G(n) = log x + 2γ − − (1.1)
2ζ(2) 2 ζ(2)
n≤x
holds for every ε > 0, where γ is Euler’s constant, and θ is the exponent appearing
in Dirichlet’s divisor problem. See the survey paper [8] by the third author.
The function
Xn X ϕ(n/d)
1
G(−1) (n) := = (n ∈ N),
(k, n) d
AN
k=1 d|n
is also multiplicative. Bordellès [1, Th. 5.1] deduced that

X ζ(3) 2
G(−1) (n) = x + O x(log x)2/3 (log log x)4/3 . (1.2)
2ζ(2)
n≤x
The error term of estimate (1.2) comes from the classical result of Walfisz [9,
Satz 1, p. 144],
X 1
DM
R(x) := ϕ(n) − x2 = O x(log x)2/3 (log log x)4/3 . (1.3)

2ζ(2)
n≤x
We remark that recently (1.3) was improved by Liu [4] into

R(x) = O x(log x)2/3 (log log x)1/3 , (1.4)
therefore, this serves as the remainder of (1.2). Also see the preprint by Suzuki [7].
The lcm-sum function
 
X n X
n
L(n) := [k, n] = 1+ dϕ(d) (n ∈ N).
2
TE
k=1 d|n
was investigated by Bordellès [1], Ikeda and Matsuoka [3], and others. The function
L(n) is not multiplicative and one has, see [1, Th. 6.3],
X ζ(3) 4
L(n) = x + O x3 (log x)2/3 (log log x)4/3 . (1.5)
8ζ(2)
n≤x
By using (1.4), the exponent of the log log x factor in the error of (1.5) can be
EP
improved into 1/3.

Now let
n
X 1
L(−1) (n) := (n ∈ N).
[k, n]
k=1
Bordellès [1, Th. 7.1] proved that

C
X 1
L(−1) (n) = 2 (log x)3 + A(log x)2 + O(log x), (1.6)
π
n≤x
AC
T
IP
CR
On certain sums concerning the gcd’s and lcm’s of k positive integers 3
with an explicitly given constant A.

By the general identity
X n
XX X
US
ψ(m, n) = 2 ψ(m, n) − ψ(n, n),
m,n≤x n≤x m=1 n≤x
valid for any function ψ : N2 → C , which is symmetric in the variables, (1.1), (1.2),
(1.5) and (1.6), together with the remark on (1.4) lead to the asymptotic formulas
X
x2 1 ζ(2) ζ 0 (2)
(m, n) = log x + 2γ − − − + O x1+θ+ε , (1.7)
ζ(2) 2 2 ζ(2)
m,n≤x
AN
X 1 ζ(3) 2
= x + O x(log x)2/3 (log log x)1/3 , (1.8)
(m, n) ζ(2)
m,n≤x
X ζ(3) 4
[m, n] = x + O x3 (log x)2/3 (log log x)1/3 , (1.9)
4ζ(2)
m,n≤x
and
DM
X 1 2
= 2 (log x)3 + A1 (log x)2 + O(log x), (1.10)
[m, n] π
m,n≤x
respectively, where A1 = 2A.

It is easy to generalize (1.7) and (1.8) for sums with k variables by using the
general identity
X X
f ((n1 , . . . , nk )) = (µ ∗ f )(d)bx/dck ,
n1 ,...,nk ≤x d≤x
where f is an arbitrary arithmetic function, µ is the Möbius function and ∗ stands

TE
for the Dirichlet convolution of arithmetic functions. For example, we have the next
result: For any k ≥ 3,
X 1 ζ(k + 1) k
= x + O xk−1 .
(n1 , . . . , nk ) ζ(k)
n1 ,...,nk ≤x
However, it is more difficult to derive asymptotic formulas for similar sums

involving the lcm [n1 , . . . , nk ]. As corollaries of more general results concerning a
EP
large class of functions f , the first and third authors [2, Cor 1] proved that for any
k ≥ 3 and any real number r > −1,
X 1

[n1 , . . . , nk ]r = Ar,k xk(r+1) + O xk(r+1)− 2 min(r+1,1)+ε (1.11)
n1 ,...,nk ≤x
and
C
X r
[n1 , . . . , nk ] 1
= Ar,k xk + O xk− 2 min(r+1,1)+ε ,
n1 · · · nk
n1 ,...,nk ≤x
AC
T
IP
CR
where Ak,r are explicitly given constants. Here, (1.11) is the k dimensional gener-
alization of (1.9). Furthermore, [2, Cor 2] shows that for any k ≥ 3 and any real
number r > 0,
X [n1 , . . . , nk ] r
US
1

= Br,k xk(r+1) + O xk(r+1)− 2 +ε ,
(n1 , . . . , nk )
n1 ,...,nk ≤x
with explicitly given constants Bk,r . The proofs use the fact that (n1 , . . . , nk ) and
[n1 , . . . , nk ] are multiplicative functions of k variables and the associated multiple
Dirichlet series factor over the primes into Euler products. The proofs given in [2]
cannot be applied in the case r = −1.
AN
It is the goal of the present paper to investigate the order of magnitude of the
sums
X 1
Sk (x) := , (1.12)
[n1 , . . . , nk ]
n1 ,...,nk ≤x
X (n1 , . . . , nk )
Tk (x) := , (1.13)
[n1 , . . . , nk ]
n1 ,...,nk ≤x
DM
X 1
Uk (x) := , (1.14)
[n1 , . . . , nk ]
n1 ,...,nk ≤x
(n1 ,...,nk )=1
X n1 · · · nk
Vk (x) := , (1.15)
[n1 , . . . , nk ]
n1 ,...,nk ≤x
where k ≥ 2 is fixed, by using elementary arguments. Theorem 2.1, concerning

the sum S2 (x), refines formulas (1.6) and (1.10) of Bordellès [1]. Theorems 2.3 and
3.1 give the exact order of magnitude of the sums Sk (x) and Uk (x), respectively,
for k ≥ 3. Theorem 4.1 concerns the sums Vk (x), while Theorem 5.2 provides an
TE
asymptotic formula with remainder term for Tk (x), for any fixed k ≥ 2. Some
conjectures and open problems are formulated as well.
2. The sums Sk (x)

First consider the sums Sk (x) defined by (1.12). In the case k = 2 we use Dirichlet’s
hyperbola method to prove the next result, which improves formulas (1.6) and
EP
(1.10).
Theorem 2.1.
X 1
L(−1) (n) = 2 (log x)3 + A(log x)2 + B log x + C + O x−1/2 (log x)2 , (2.1)
π
n≤x
that is,
C
X 1 2
= 2 (log x)3 + A1 (log x)2 + B1 log x + C1 + O x−1/2 (log x)2 ,
[m, n] π
m,n≤x
AC
T
IP
CR
where the constants A, B, C can be explicitly computed, and A1 = 2A, B1 = 2B − 1,

C1 = C − γ.
US
Proof. We have
n n n/d
X (k, n) 1X X 1 1X X 1 1X
L(−1) (n) = = d = = h(d), (2.2)
kn n k n t=1
t n
k=1 d|n k=1 d|n d|n
(k,n)=d (t,n/d)=1
where
n n n/d
X 1 X 1 X X µ(d) X 1
h(n) :=
AN = µ(d) =
m=1
m m=1 m d j=1 j
d|(m,n) d|n
(m,n)=1
X µ(d) n
X
d µ(d) n ϕ(n)
ω(n)
2
= log + γ + O = log + γ +O .
d d n d d n n
d|n d|n
Hence,
 
X X µ(d) X X ϕ(n) X 2ω(n)
DM
H(x) := h(n) = log m + γ +O .

d n n
n≤x d≤x m≤x/d n≤x n≤x
By using the known estimates

X
log n = x log x − x + O(log x),
n≤x
X ϕ(n) 6
= 2 x + O(log x),
n π
n≤x
X 2ω(n)
= O((log x)2 ),
TE
n
n≤x
we deduce that
X µ(d) X µ(d) log d 6
H(x) = (x log x − x) 2
−x 2
+ 2 γx + O((log x)2 )
d d π
d≤x d≤x
6
(x log x + cx) + O((log x)2 ),
EP
= (2.3)
π2
with a certain constant c. Let 1(n) = 1 (n ∈ N), and let ∗ denote the Dirichlet
convolution. By Dirichlet’s hyperbola method,
X X √ √
(1 ∗ h)(n) = (H(x/n) + h(n)bx/nc) − b xcH( x)
√
n≤x n≤ x
C
X X h(n) √ √ √
= H(x/n) + x − xH( x) + O(H( x)).
√ √ n
n≤ x n≤ x
AC
T
IP
CR
By partial summation,
Z √
X h(n) √ √ x
H(t)
x = xH( x) + x dt,
n t2
US
√ 1
n≤ x
and using (2.3) we deduce

x 6x Z x log t
√
X 6 X x x dt √
(1 ∗ h)(n) = 2 log +c + 2 +c + O x(log x)2
π √ n n n π 1 t t
n≤x n≤ x

3 √
=x (log x)2 + a log x + b + O( x(log x)2 ),
AN
π2
for some constants a, b, which can be explicitly calculated.
Here (1 ∗ h)(n) = nL(−1) (n), according to (2.2), and we obtain (2.1) by partial
summation.
It is more difficult to handle the sums Sk (x) in the case k ≥ 3. We will apply the
following general result proved by the second and third authors [5], using elementary
DM
arguments.
Theorem 2.2. ([5]) Let k be a positive integer and let f : N → C be a multiplicative

function satisfying the following properties:
(i) f (p) = k for every prime p,
(ii) f (pν ) = ν O(1) for every prime p and every integer ν ≥ 2, where the constant
implied by the O symbol is uniform in p.
Then
X f (n) 1
= Cf (log x)k + Df (log x)k−1 + O (log x)k−2 ,
n k!
TE
n≤x
where Cf and Df are constants,

!
Y 1
k ∞
X f (pν )
Cf = 1− .
p
p ν=0
pν
We have the following result.

EP
Theorem 2.3. Let k ≥ 3 be a fixed integer. Then

k
−1
Sk (x) (log x)2 as x → ∞.
Proof. Since [n1 , . . . , nk ] ≤ n1 · · · nk ≤ xk , we can write

X 1 X
C
Sk (x) = 1 (2.4)
n
k
n≤x n1 ,...,nk ≤x
[n1 ,...,nk ]=n
AC
T
IP
CR
Let
X
ak (n) := 1.
n1 ,...,nk ∈N
US
[n1 ,...,nk ]=n
Now if n ≤ x, then the inner sum in (2.4) is just ak (n) (since n ≤ x forces
n1 , . . . , nk ≤ x), while in any case it is at most ak (n). Thus
X ak (n) X ak (n)
≤ Sk (x) ≤ . (2.5)
n n
n≤x k n≤x
AN
To see the properties of the function ak (n) write

X X X X X
ak (d) = 1= 1= 1 = τ (n)k .
d|n d|n [n1 ,...,nk ]=d [n1 ,...,nk ]|n n1 |n,...,nk |n
Therefore, by Möbius inversion, we have ak = µ ∗ τ k . This shows that ak (n) is

multiplicative and its values at the prime powers pν are given by ak (pν ) = (ν +
1)k − ν k (ν ≥ 1). In particular, ak (p) = 2k − 1.
Applying Theorem 2.2 for the function f (n) = ak (n), with 2k − 1 instead of k,
DM
we get that
X ak (n) k
∼ αk (log x)2 −1 as x → ∞, (2.6)
n
n≤x
for some constant αk . Now, from (2.5) and (2.6) the result follows.
k
Remark 2.4. It is natural to expect that Sk (x) ∼ ck (log x)2 −1 as x → ∞, with a
certain constant ck . In fact, in view of Theorem 2.1, the plausible conjecture is that
TE
Sk (x) = P2k −1 (log x) + O(x−r ), (2.7)
where P2k −1 (t) is a polynomial in t of degree 2k − 1 and r is a positive real number.

We pose as an open problem to find the constants ck and to prove (2.7).
3. The sums Uk (x)

EP
Next consider the sums Uk (x) defined by (1.14). In the case k = 2,

6
U2 (x) ∼ (log x)2 as x → ∞,
π2
and it is not difficult to deduce a more precise asymptotic formula.
We have the following general result.
C

k
−2
Uk (x) (log x)2 as x → ∞.
AC
T
IP
CR
Proof. Similar to the proof of Theorem 2.3. We have

X 1 X 1 X
Uk (x) = = 1. (3.1)
[n1 , . . . , nk ] n
US
n1 ,...,nk ≤x k n≤x n1 ,...,nk ≤x
(n1 ,...,nk )=1 [n1 ,...,nk ]=n
(n1 ,...,nk )=1
Let
X
bk (n) = 1.
n1 ,...,nk ∈N
[n1 ,...,nk ]=n
(n1 ,...,nk )=1
AN
Now if n ≤ x, then the inner sum in (3.1) is exactly bk (n), while in any case it
is at most bk (n). Thus
X bk (n) X bk (n)
≤ Uk (x) ≤ . (3.2)
n n
n≤x k n≤x
Write
X X X X
DM
bk (d) = 1= 1
d|n d|n [n1 ,...,nk ]=d [n1 ,...,nk ]|n
(n1 ,...,nk )=1 (n1 ,...,nk )=1
X X X
= µ(δ) = µ(δ)
n1 |n,...,nk |n δ|(n1 ,...,nk ) δa1 b1 =n,...,δak bk =n
X X X X
= µ(δ) 1··· 1= µ(δ)τ (t)k .
δt=n a1 b1 =t ak bk =t δt=n
Therefore, by Möbius inversion bk = µ ∗ µ ∗ τ k . This shows that bk (n) is

TE
multiplicative and its values at the prime powers pν are given by bk (pν ) =
(ν + 1)k − 2ν k + (ν − 1)k (ν ≥ 1). In particular, bk (p) = 2k − 2.
Applying now Theorem 2.2 for the function f (n) = bk (n), with 2k − 2 instead
of k, we deduce that
X bk (n) k
∼ αk0 (log x)2 −2 as x → ∞ (3.3)
n
EP
n≤x
k
for some constant αk0 . Now, from (3.2) and (3.3) we have Uk (x) (log x)2 −2
.
k
Remark 3.2. We conjecture that Uk (x) ∼ dk (log x)2 −2 as x → ∞, with a certain
constant dk . The sums Sk (x) and Uk (x) are strongly related. Namely, by grouping
the terms according to the values (n1 , . . . , nk ) = d one obtains
C
X1
Sk (x) = Uk (x/d), (3.4)
d
d≤x
AC
T
IP
CR
and conversely,
X µ(d)
Uk (x) = Sk (x/d). (3.5)
US
d≤x
k
If Uk (x) ∼ dk (log x)2 −2 holds, then by (3.4) it follows that Sk (x) ∼
dk k
2k −1
(log x)2 −1 . Conversely, assume that the asymptotic formula (2.7) is true, where
ck is the leading coefficient of the polynomial P2k −1 (t). Then (3.5), together with
the well known results
X µ(n) X∞
µ(n) log n
= O((log x)−1 ), = −1,
n
AN n
n≤x n=1
and Shapiro’s estimates [6, Th. 4.1]

X µ(n) x m m−2
X (m)
log = m(log x)m−1 + cj (log x)j + O(1),
n n i=1
n≤x
(m)
valid for any integer m ≥ 2, where ci are constants, imply that
DM
k k
−2 −3
Uk (x) = (2k − 1)ck (log x)2 + b2k −3 (log x)2 + · · · + b1 log x + O(1),
with some constants bi .
4. The sums Vk (x)

The sums Vk (x) defined by (1.15) are sums of integers. In the case k = 2 we have,
according to (1.7),
X 6
V2 (x) = (m, n) ∼ 2 x2 log x. (4.1)
π
m,n≤x
TE

k
−2
xk Vk (x) xk (log x)2 as x → ∞.
Proof. The lower bound is trivial by n1 · · · nk ≥ [n1 , . . . , nk ]. Also, by group-

ing the terms according to the values (n1 , . . . , nk ) = d, and by denoting M =
max(m1 , . . . , mk ) we have
EP
X dm1 · · · dmk X m1 · · · mk X
Vk (x) = = dk−1
[dm1 , . . . , dmk ] [m1 , . . . , mk ]
dm1 ,...,dmk ≤x m1 ,...,mk ≤x d≤x/M
(m1 ,...,mk )=1 (m1 ,...,mk )=1
X m1 · · · mk X 1
xk ≤ xk = xk Uk (x),
[m1 , . . . , mk ]M k [m1 , . . . , mk ]
C
m1 ,...,mk ≤x m1 ,...,mk ≤x
(m1 ,...,mk )=1 (m1 ,...,mk )=1
and the upper bound follows from Theorem 3.1.

AC
T
IP
CR
k
Remark 4.2. We conjecture that Vk (x) ∼ λk xk (log x)2 −k−1 as x → ∞, with a
certain constant λk , in accordance with (4.1) for the case k = 2. We pose as another
open problem to prove this and to find the constants λk .
US
5. The sums Tk (x)
Finally, we investigate the sums Tk (x) defined by (1.13) and establish an asymptotic
formula with remainder term for it. We give a short direct proof in the case k = 2.
Then for any fixed k ≥ 2 we use multiple Dirichlet series to get the result.
Let
AN
n
X (k, n)
F (n) := (n ∈ N). (5.1)
[k, n]
k=1
Theorem 5.1.
X
F (n) = 2x + O (log x)2 , (5.2)
n≤x
that is,
DM
X (m, n)
= 3x + O (log x)2 .
[m, n]
m,n≤x
P
Proof. Let φ2 (n) = d|n d2 µ(n/d) be the Jordan function of order 2. We have
n
X n n
(k, n)2 1X1 X 1X X 1
F (n) = = φ2 (d) = φ2 (d)
kn n k n k
k=1 k=1 d|(k,n) d|n k=1
d|k
TE
n/d
1 X φ2 (d) X 1 1 X φ2 (d)
= = Hn/d ,
n d j=1 j n d
d|n d|n
Pm
where Hm = j=1 1/j is the harmonic sum. Therefore, using that
X φ2 (n) x
2
= + O(1),
n ζ(3)
EP
n≤x
we deduce
X X φ2 (d) X Hm X φ2 (d)
F (n) = 2
Hm =
d m m d2
n≤x dm≤x m≤x d≤x/m
 
C
X Hm x
x X Hm X Hm
= + O(1) = +O 
m ζ(3)m ζ(3) m2 m
m≤x m≤x m≤x
AC
T
IP
CR
 
X∞ X X
x Hm Hm Hm 
= + O(x )+O
ζ(3) m=1 m2 m>x
m 2 m
m≤x
US
!  
x X log m X log m
= · 2ζ(3) + O x +O  = 2x + O((log x)2 ),
ζ(3) m>x
m2 m
m≤x
by using that
X∞
Hn
= 2ζ(3), (5.3)
n2
AN n=1
which is Euler’s result.
Theorem 5.2. If k ≥ 2, then

k

Tk (x) = βk x + O (log x)2 −2 ,
where
∞
X X∞
1 1 1
βk := = .
DM
n1 ,...,nk =1
[n1 , . . . , nk ] max(n1 , . . . , nk ) ζ(2) n ,...,n =1 [n1 , . . . , nk ] max(n1 , . . . , nk )
1 k
(n1 ,...,nk )=1
Proof. By grouping the terms according to (n1 , . . . , nk ) = d, where nj = dmj

(1 ≤ j ≤ k), (m1 , . . . , mk ) = 1, we have
X d X 1
Tk (x) = =
[dm1 , . . . , dmk ] [m1 , . . . , mk ]
dm1 ,...,dmk ≤x dm1 ,...,dmk ≤x
(m1 ,...,mk )=1 (m1 ,...,mk )=1
X 1 X X bx/M c
= 1= ,
[m1 , . . . , mk ] [m1 , . . . , mk ]
TE
m1 ,...,mk ≤x d≤x/M m1 ,...,mk ≤x

(m1 ,...,mk )=1 (m1 ,...,mk )=1
where M = max(m1 , . . . , mk ). Let

(
1
[n1 ,...,nk ] , if (n1 , . . . , nk ) = 1,
h(n1 , . . . , nk ) :=
0, otherwise.
Hence,
 
EP
X h(n1 , . . . , nk ) X
Tk (x) = x +O h(n1 , . . . , nk ) (5.4)
max(n1 , . . . , nk )
n1 ,...,nk ≤x n1 ,...,nk ≤x
and we estimate the right-hand sums in turn. Here h(n1 , . . . , nk ) is a symmetric and
multiplicative function of k variables and for prime powers pν1 , . . . , pνk (ν1 , . . . , νk ≥
0) one has
(
C
1
ν1 νk max(ν1 ,...,νk ) , if min(ν1 , . . . , νk ) = 0,
h(p , . . . , p ) = p
0, otherwise.
AC
T
IP
CR
Consider its Dirichlet series

X∞ ∞
h(n1 , . . . , nk ) Y X 1
H(s1 , . . . , sk ) := = .
ns11 · · · nskk
p
pmax(ν1 ,...,νk )+ν1 s1 +···+νk sk
US
n ,...,n =1
1 k ν1 ,...,νk =0
min(ν1 ,...,νk )=0
By grouping the terms according to the values of r = max(ν1 , . . . , νk ) we deduce

Y 1 X ∞ ∞
X 1
H(s1 , . . . , sk ) = ,
p
pr r=0 ν1 ,...,νk =0
pν1 s1 +···+νk sk
max(ν1 ,...,νk )=r
min(ν1 ,...,νk )=0
AN
which converges absolutely for <sj > 0 (1 ≤ j ≤ k).

We shall need an estimate for Hk (ε, . . . , ε) for ε > 0 (small). We have
 
Y 1 X k 1
k−1
1
H(ε, . . . , ε) = 1 + +O .
p
p j=1 j pjε p2
Therefore,
X k 1
X 1 k−1 X k X 1
k−1
log H(ε, . . . , ε) = + O(1) = + O(1).
DM
p
p j=1 j pjε j=1
j p
p1+jε
P
But p p−1−ε = log 1ε + O(1) as ε → 0. Thus,
 
X k
k−1
1
2k −2
1
H(ε, . . . , ε) = exp  log + O(1)  . (5.5)
j=1
j ε ε
Furthermore, for any ε > 0, we have

X X h(n1 , . . . , nk )
h(n1 , . . . , nk ) = ε/k
(n1 · · · nk )ε/k
n1 ,...nk ≤x n1 ,...nk ≤x
(n 1 · · · n k )
TE
X h(n1 , . . . , nk )
≤ xε ε/k
≤ xε H(ε/k, . . . , ε/k). (5.6)
n1 ,...nk ≤x
(n 1 · · · n k )
Next, note that max(n1 , . . . , nk ) ≥ (n1 · · · nk )1/k , so that

X h(n1 , . . . , nk ) X h(n1 , . . . , nk )
≤ 1/k
≤ H(ε/k, . . . , ε/k),
max(n1 , . . . , nk ) (n 1 · · · nk )
EP
n ,...,n ≤x
1 k n ,...,n ≤x
1 k
which converges. Hence,

∞
X h(n1 , . . . , nk )
βk =
n1 ,...,nk =1
max(n 1 , . . . , nk )
is finite and βk ≤ H(ε/k, . . . , ε/k). Also,

C
X h(n1 , . . . , nk ) X h(n1 , . . . , nk )
βk − =
max(n1 , . . . , nk ) max(n1 , . . . , nk )
n1 ,...,nk ≤x n1 ,...,nk ∈N
some ni >x
AC
T
IP
CR
X h(n1 , . . . , nk ) X h(n1 , . . . , nk )
≤k ≤k 1−ε
n1 n1 (n1 n2 · · · nk )ε/k
n1 ≥n2 ,...,nk n1 ≥n2 ,...,nk
n1 >x n1 >x
US
∞
X
k h(n1 , . . . , nk )
≤ = kxε−1 H(ε/k, . . . , ε/k). (5.7)
x1−ε n1 ,...,nk =1
(n 1 · · · n k ) ε/k
Hence, (5.4) and the estimates (5.6), (5.7) give

Tk (x) = βk x + O (xε H(ε/k, . . . , ε/k)) .
Now we choose ε = 1/ log x and use the bound (5.5). The proof is complete.
AN
Remark 5.3. For k = 2, Theorem 5.2 recovers Theorem 5.1. Note that
∞ ∞ m
1 X 1 2 X 1 X1
β2 = = − 1 = 3,
ζ(3) m,n=1 mn max(m, n) ζ(3) m=1 m2 n=1 n
by Euler’s result (5.3). Is it possible to evaluate the constants βk for any k ≥ 2?

The sums Tk (x) and Uk (x) are related by the formulas
X X
DM
Tk (x) = Uk (x/d), Uk (x) = µ(d)Tk (x/d).

d≤x d≤x
Acknowledgments
Part of the work in this paper was done when both the second and third authors
visited the Max Plank Institute for Mathematics Bonn, in February 2017. They
thank this institution for hospitality and a fruitful working environment. F. L. was
also supported by grant CPRR160325161141 and an A-rated scientist award both
from the NRF of South Africa and by grant no. 17-02804S of the Czech Grant-
TE
ing Agency. L. T. was also supported by the European Union, co-financed by the
European Social Fund EFOP-3.6.1.-16-2016-00004.
References
[1] O. Bordellès, Mean values of generalized gcd-sum and lcm-sum functions, J. Integer
Seq. 10 (2007), Article 07.9.2, 13 pp.
EP
[2] T. Hilberdink and L. Tóth, On the average value of the least common multiple of k
positive integers, J. Number Theory 169 (2016), 327–341.
[3] S. Ikeda and K. Matsuoka, On the lcm-sum function, J. Integer Seq. 17 (2014), Article
14.1.7, 11 pp.
[4] H.-Q. Liu, On Euler’s function, Proc. Roy. Soc. Edinburgh Sect. A 146 (2016), 769–
775.
[5] F. Luca and L. Tóth, The rth moment of the divisor function: an elementary ap-
C
proach, J. Integer Seq. 20 (2017), Article 17.7.4, 8 pp.

[6] H. N. Shapiro, On a theorem of Selberg and generalizations, Ann. Math. 51 (1950),
485–497.
AC
T
IP
CR
[7] Y. Suzuki, On error term estimates á la Walfisz for mean values of arithmetic func-
tions, Preprint, 2018, 32 pp., arXiv:1811.02556 [mathNT].
[8] L. Tóth, A survey of gcd-sum functions, J. Integer Seq. 13 (2010), Article 10.8.1, 23
pp.
US
[9] A. Walfisz, Weylsche Exponentialsummen in der neueren Zahlentheorie, Mathema-
tische Forschungsberichte, XV, VEB Deutscher Verlag der Wissenschaften, 1963.
AN
DM
TE
C EP
AC

Accepted Manuscript International Journal of Number Theory

Uploaded by

Document Information

Original Description:

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

Accepted Manuscript International Journal of Number Theory

Uploaded by

Copyright:

Available Formats

Accepted manuscript to appear in IJNT

Author(s): Titus Hilberdink, Florian Luca, Laszlo Toth

Received: 28 May 2018

Accepted: 19 June 2019

Link to final version: https://doi.org/10.1142/S1793042120500049

Department of Mathematics, Faculty of Sciences, University of Ostrava

Received (Day Month Year)

Mathematics Subject Classification 2010: 11A25, 11N37

is also multiplicative. Bordellès [1, Th. 5.1] deduced that

R(x) := ϕ(n) − x2 = O x(log x)2/3 (log log x)4/3 . (1.3)

We remark that recently (1.3) was improved by Liu [4] into

improved into 1/3.

Bordellès [1, Th. 7.1] proved that

with an explicitly given constant A.

respectively, where A1 = 2A.

where f is an arbitrary arithmetic function, µ is the Möbius function and ∗ stands

However, it is more difficult to derive asymptotic formulas for similar sums

where k ≥ 2 is fixed, by using elementary arguments. Theorem 2.1, concerning

2. The sums Sk (x)

where the constants A, B, C can be explicitly computed, and A1 = 2A, B1 = 2B − 1,

H(x) := h(n) = log m + γ +O .

By using the known estimates

and using (2.3) we deduce

Theorem 2.2. ([5]) Let k be a positive integer and let f : N → C be a multiplicative

where Cf and Df are constants,

We have the following result.

Theorem 2.3. Let k ≥ 3 be a fixed integer. Then

Proof. Since [n1 , . . . , nk ] ≤ n1 · · · nk ≤ xk , we can write

To see the properties of the function ak (n) write

Therefore, by Möbius inversion, we have ak = µ ∗ τ k . This shows that ak (n) is

Sk (x) = P2k −1 (log x) + O(x−r ), (2.7)

where P2k −1 (t) is a polynomial in t of degree 2k − 1 and r is a positive real number.

3. The sums Uk (x)

Next consider the sums Uk (x) defined by (1.14). In the case k = 2,

Theorem 3.1. Let k ≥ 3 be a fixed integer. Then

Proof. Similar to the proof of Theorem 2.3. We have

Therefore, by Möbius inversion bk = µ ∗ µ ∗ τ k . This shows that bk (n) is

and Shapiro’s estimates [6, Th. 4.1]

4. The sums Vk (x)

Theorem 4.1. Let k ≥ 3 be a fixed integer. Then

Proof. The lower bound is trivial by n1 · · · nk ≥ [n1 , . . . , nk ]. Also, by group-

and the upper bound follows from Theorem 3.1.

open problem to prove this and to find the constants λk .

which is Euler’s result.

Theorem 5.2. If k ≥ 2, then

Proof. By grouping the terms according to (n1 , . . . , nk ) = d, where nj = dmj

m1 ,...,mk ≤x d≤x/M m1 ,...,mk ≤x

where M = max(m1 , . . . , mk ). Let

Consider its Dirichlet series

By grouping the terms according to the values of r = max(ν1 , . . . , νk ) we deduce

which converges absolutely for <sj > 0 (1 ≤ j ≤ k).

Furthermore, for any ε > 0, we have

Next, note that max(n1 , . . . , nk ) ≥ (n1 · · · nk )1/k , so that

which converges. Hence,

is finite and βk ≤ H(ε/k, . . . , ε/k). Also,

Hence, (5.4) and the estimates (5.6), (5.7) give

by Euler’s result (5.3). Is it possible to evaluate the constants βk for any k ≥ 2?

Tk (x) = Uk (x/d), Uk (x) = µ(d)Tk (x/d).

proach, J. Integer Seq. 20 (2017), Article 17.7.4, 8 pp.

You might also like