REGULAR CLASS DIVISION OF INTEGERS (mod r)

Pentti Haukkanen

Department of Mathematics, Statistics and Philosophy,

FIN-33014 University of Tampere, Finland

1 Introduction
Let r be an integer > 1. Let t1 (= 1), t2 , . . . , ts (= r) be the divisors of r, where s
is the number of divisors of r. The integers x with 1 ≤ x ≤ r fall into s mutually
disjoint classes
C1 , C2 , . . . , Cs ,
where
Ci = {x : 1 ≤ x ≤ r, (x, r) = ti }.
The number of elements in Ci is equal to φ(r/ti ), where φ is the Euler totient
function.
The sum of Ci and Cj , denoted by Ci ⊕ Cj , is defined as the multiset given by

Ci ⊕ Cj = {x + y (mod r) : x ∈ Ci , y ∈ Cj }.

It is known that in Ci ⊕ Cj elements of a class Ck occur the same number M (i, j, k)

of times, see R. Vaidyanathaswamy [11]. For a concrete example, see [10, Chapter
XV] or [11]. The coefficients M (i, j, k) can be evaluated in terms of Ramanujan
sums c(n, r), see K. G. Ramanathan [9]. To be more precise,


s
Ci ⊕ Cj = M (i, j, k)Ck , (1.1)
k=1

where 
M (i, j, k) = r−1 c(r/d, r/ti )c(r/d, r/tj )c(tk , d). (1.2)
d|r

The product M (i, j, k)Ck means the multiset, where each of the elements of the set
Ck occurs M (i, j, k) times.
A divisor d of r is said to be a unitary divisor of r and is denoted by dr if
(d, r/d) = 1. K. Nageswara Rao [7] gives class division of integers mod r with
respect to unitary divisors of r. In this paper we give a class division of integers
mod r with respect to A-divisors of r, where A is Narkiewicz’s regular convolution
[8]. This class division contains as special cases the usual class division of integers
mod r and class division of integers mod r with respect to unitary divisors of r.

1
2 Regular convolution
In this section we introduce the concept of Narkiewicz’s regular convolution. Back-
ground material on regular convolutions can be found e.g. in [6, Chapter 4] and [8].
We here review the concepts and notations needed in this paper.
For each n let A(n) be a subset of the set of positive divisors of n. The elements
of A(n) are said to be the A-divisors of n. The A-convolution of two arithmetical
functions f and g is defined by

(f ∗A g)(n) = f (d)g(n/d).
d∈A(n)

Narkiewicz [8] defines an A-convolution to be regular if

(a) the set of arithmetical functions forms a commutative ring with unity with
respect to the ordinary addition and the A-convolution,
(b) the A-convolution of multiplicative functions is multiplicative,
(c) the constant function 1 has an inverse µA with respect to the A-convolution,
and µA (n) = 0 or −1 whenever n is a prime power.
It can be proved [8] that an A-convolution is regular if and only if
(i) A(mn) = {de : d ∈ A(m), e ∈ A(n)} whenever (m, n) = 1,
(ii) for each prime power pa (> 1) there exists a divisor t = tA (pa ) of a such that

A(pa ) = {1, pt , p2t , . . . , prt },

where rt = a, and
A(pit ) = {1, pt , p2t , . . . , pit }, 0 ≤ i < r.
The positive integer t = tA (pa ) in part (ii) is said to be the A-type of pa . A positive
integer n is said to be A-primitive if A(n) = {1, n}. The A-primitive numbers are
1 and pt , where p runs through the primes and t runs through the A-types of the
prime powers pa with a ≥ 1.
For all n let D(n) denote the set of all positive divisors of n and let U (n) denote
the set of all unitary divisors of n, that is,

U (n) = {d > 0 : d | n, (d, n/d) = 1} = {d > 0 : d  n}.

The D-convolution is the classical Dirichlet convolution and the U -convolution is

the unitary convolution [3]. These convolutions are regular with tD (pa ) = 1 and
tU (pa ) = a for all prime powers pa (> 1).
Throughout the rest of the paper A will be an arbitaray but fixed regular con-
volution.
The A-analogue of the Möbius function µA is the multiplicative function given
by 
−1 if pa (> 1) is A-primitive,
µA (pa ) =
0 if pa is non-A-primitive.

2
In particular, µD = µ, the classical Möbius function, and µU = µ∗ , the unitary
analogue of the Möbius function [3].
The symbol (m, n)A denotes the greatest divisor of m, which belongs to A(n).
In particular, (m, n)D is the usual greatest common divisor (m, n) of m and n.
The generalized Ramanujan’s sum cA (n; r) is defined [5] by

cA (n; r) = exp(2πinx/r).
x (mod r)
(x,r)A =1

An arithmetical evaluation is given [5] by


cA (n; r) = dµA (r/d).
d∈A((n,r)A )

In particular, denote
φA (r) = cA (0; r).
Then φA (r) is the number of integers x (mod r) such that (x, r)A = 1. Also

φA (r) = dµA (r/d).
d∈A(r)

Cohen [1] defined an arithmetical function f (n; r) to be even (mod r) if

f (n; r) = f ((n, r); r)

for all n. An arithmetical function f (n; r) is said to be A-even (mod r) [5] if

f (n; r) = f ((n, r)A ; r)

for all integers n. Alternatively, we may say that f (n; r) is A-even (mod r) if

f ((n, r)A ; r) = f ((m, r)A ; r)

whenever (n, r)A = (m, r)A . In particular, D-even functions (mod r) are even
functions (mod r).
The Cauchy product of two A-even functions f (n; r) and g(n; r) (mod r) is de-
fined by 
(f ◦ g)(n; r) = f (a; r)g(b; r).
n≡a+b (mod r)

The concept of the Cauchy product of even functions (mod r) originates in Cohen
[2].
It is known [5] that an arithmetical function f (n; r) is A-even (mod r) if and
only if it has a unique representation of the form

f (n; r) = α(d; r)cA (n; d),
d∈A(r)

3
where 
α(d; r) = r−1 f (r/δ; r)cA (r/d; δ)
δ∈A(r)

The numbers α(d; r) are called the Fourier coefficients of f . If f (n; r) and g(n; r) are
A-even functions (mod r) with Fourier coefficients α(d; r) and β(d; r), respectively,
then their Cauchy product is given as

(f ◦ g)(n; r) = r α(d; r)β(d; r)cA (n; d)
d∈A(r)

and so the Cauchy product (f ◦ g)(n; r) is also an A-even function (mod r), cf. [4].

3 Regular class division of integers (mod r)

Let r be an integer > 1 and let A be a regular convolution. Let t1 (= 1), t2 , . . . , ts (= r)
be the A-divisors of r, where s is the number of A-divisors of r. The integers x with
1 ≤ x ≤ r fall into s mutually disjoint classes

C1 , C2 , . . . , Cs ,

where
Ci = {x : 1 ≤ x ≤ r, (x, r)A = ti }. (3.1)
The number of elements in Ci is equal to φA (r/ti ).
The sum of Ci and Cj , denoted by Ci ⊕ Cj , is defined as the multiset given by

Ci ⊕ Cj = {x + y (mod r) : x ∈ Ci , y ∈ Cj }. (3.2)

We show that in Ci ⊕ Cj elements of a class Ck occur the same number of times and
give a generalization for the formula (1.2), see Theorem 3.1.
We begin with a concrete example. Let r = 12 and let A = U , the unitary con-
volution. The unitary divisors of 12 are 1, 3, 4, 12. Further, C1 = {1, 2, 5, 7, 10, 11},
C2 = {3, 6, 9}, C3 = {4, 8} and C4 = {12}. It is easy to verify that

C1 ⊕ C1 = 2C1 ∪ 4C2 ∪ 3C3 ∪ 6C4 ,

C1 ⊕ C2 = 2C1 ∪ 3C3 ,
C1 ⊕ C3 = C1 ∪ 2C2 ,
C1 ⊕ C4 = C1 ,
C2 ⊕ C2 = 2C2 ∪ 3C4 ,
C2 ⊕ C3 = C1 ,
C2 ⊕ C4 = C2 ,
C3 ⊕ C3 = C3 ∪ 2C4 ,
C3 ⊕ C4 = C3 ,
C4 ⊕ C4 = C4 .

4
Theorem 3.1 Let Ci and Ci ⊕ Cj be given as in (3.1) and (3.2). Then

s
Ci ⊕ Cj = M (i, j, k)Ck , (3.3)
k=1

where 
M (i, j, k) = r−1 cA (r/d, r/ti )cA (r/d, r/tj )cA (tk , d). (3.4)
d∈A(r)

Proof Let 
(i) 1 if (n, r)A = ti (or n ∈ Ci ),
ρ (n; r) =
0 otherwise.

The function ρ(i) (n; r) is A-even (mod r) with Fourier coefficients given as

α(d; r) = r−1 ρ(i) (r/δ; r)cA (r/d; δ) = r−1 cA (r/d; r/ti ).
δ∈A(r)

Thus ρ(i) ◦ ρ(j) (n; r) is A-even (mod r) and


ρ(i) ◦ ρ(j) (n; r) = r−1 cA (r/d; r/ti )cA (r/d; r/tj )cA (n; d). (3.5)
d∈A(r)

Let m ∈ Ck . The number of the element m in Ci ⊕ Cj is equal to


1
m≡a+b (mod r)
a∈Ci ,b∈Cj

or
ρ(i) ◦ ρ(j) (m; r).
Since the function ρ(i) ◦ ρ(j) (n; r) is A-even (mod r), we have ρ(i) ◦ ρ(j) (n; r) =
ρ(i) ◦ ρ(j) (m; r) whenever (n, r)A = (m, r)A or whenever n ∈ Ck . This proves (3.3).
Further,
ρ(i) ◦ ρ(j) (m; r) = ρ(i) ◦ ρ(j) ((m, r)A ; r) = ρ(i) ◦ ρ(j) (tk ; r).
Thus, by (3.5), we obtain (3.4). 2

References
[1] E. Cohen: A class of arithmetical functions, Proc. Nat. Acad. Sci. (USA) 41
(1955), 939–944.

[2] E. Cohen: Representations of even functions (mod r), II. Cauchy products,
Duke Math. J. 26 (1959), 165–182.

[3] E. Cohen: Arithmetical functions associated with the unitary divisors of an

integer, Math. Z. 74 (1960), 66–80.

5
 [4] P. Haukkanen and P. J. McCarthy: Sums of values of even functions, Portugal.
Math. 48 (1991), 53–66.

[5] P. J. McCarthy: Regular arithmetical convolutions, Portugal. Math. 27 (1968),

1-13.

[6] P. J. McCarthy: Introduction to Arithmetical Functions, Springer-Verlag Uni-

versitext (1986).

[7] K. Nageswara Rao: Unitary class division of integers mod n and related arith-
metical identities, J. Indian Math. Soc., n. Ser. 30 (1966), 195–205.

[8] W. Narkiewicz: On a class of arithmetical convolutions, Colloq. Math. 10

(1963), 81–94.

[9] K. G. Ramanathan: Some applications of Ramanujan’s trigonometrical sum

Cm (n), Proc. Ind. Acad. Sci. (A) 20 (1944), 62–69.

[10] R. Sivaramakrishnan: Classical Theory of Arithmetic Functions, Marcel

Dekker: Monographs and Text Books in Pure and Applied Mathematics No.
126 (1989).

[11] R. Vaidyanathaswamy: A remarkable property of integers (mod N ) and its

bearing on group theory, Proc. Ind. Acad. Sci. Section A (1937), 63–75.