18.

781 (Spring 2016): Floor and

arithmetic functions
Darij Grinberg
January 22, 2021

Contents
1. The floor function 2
1.1. Definition and basic properties . . . . . . . . . . . . . . . . . . . . . 2
1.2. Interlude: greatest common divisors . . . . . . . . . . . . . . . . . . 9
1.3. de Polignac’s formula . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2. Arithmetic functions 22
2.1. Arithmetic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2. Multiplicative functions . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3. The Dirichlet convolution . . . . . . . . . . . . . . . . . . . . . . . . 33
2.4. Examples of Dirichlet convolutions . . . . . . . . . . . . . . . . . . . 40
2.5. Möbius inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.6. Dirichlet convolution and multiplicativity . . . . . . . . . . . . . . . 49
2.7. Explicit formulas from multiplicativity . . . . . . . . . . . . . . . . 54
2.8. Pointwise products . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.9. Lowest common multiples . . . . . . . . . . . . . . . . . . . . . . . . 59
2.10. Lcm-convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3. Appendix: a proof of Bézout’s identity 67

***
These are the extended notes for the 18.781 (Introduction to Number Theory)
class on 14 April 2016 (the actual class covered about 1/3 of what is in these
notes). I roughly follow [NiZuMo91, §4.1–§4.3], although not always using the
same notations.
I use the notation N for {0, 1, 2, . . .}, and the notation N+ for {1, 2, 3, . . .}.

1
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

1. The floor function

Let me first recall two basic facts about divisibility of integers:

Proposition 1.0.1. Let a be an integer. Let u and v be two integers that are
both divisible by a. Then, their sum u + v must also be divisible by a.

Proof of Proposition 1.0.1. There exists an integer p such that u = ap (since u is

divisible by a). There exists an integer q such that v = aq (since v is divisible
by a). Consider these p and q. Now, |{z} u + |{z}
v = ap + aq = a ( p + q). Hence,
= ap = aq
u + v is divisible by a. This proves Proposition 1.0.1.

Proposition 1.0.2. Let u and v be two nonnegative integers such that u | v and
v | u. Then, u = v.

Proof of Proposition 1.0.2. We have v | u. In other words, there exists an integer w

such that u = vw. Consider this w. If v = 0, then we have u = |{z} v w = 0 = v.
=0
Hence, if v = 0, then Proposition 1.0.2 holds. Thus, for the rest of this proof, we
can WLOG assume that v 6= 0. Assume this. Thus, v > 0 (since v is nonnegative).
We have u | v. In other words, there exists an integer z such that v = uz.
Consider this z. Since uz = v 6= 0, we have z 6= 0. If u = 0, then we have
u z = 0 = u and thus u = v. Hence, if u = 0, then Proposition 1.0.2 holds.
v = |{z}
=0
Thus, for the rest of this proof, we can WLOG assume that u 6= 0. Assume this.
Thus, u > 0 (since u is nonnegative).
From uz = v > 0, we obtain z > 0 (since u > 0). Hence, z ≥ 1 (since z is
an integer). Hence, v = u |{z}z ≥ u1 (since u > 0) and thus v ≥ u1 = u. The
≥1
same argument (with the roles of u and v swapped) yields u ≥ v. Combining
this with v ≥ u, we obtain u = v. This proves Proposition 1.0.2.

1.1. Definition and basic properties

I shall first discuss the floor function, following [NiZuMo91, §4.1].

Definition 1.1.1. Let x be a real number. Then, b x c is defined to be the unique

integer n satisfying n ≤ x < n + 1. This integer b x c is called the floor of x, or
the integer part of x.

Remark 1.1.2. (a) Why is b x c well-defined? I mean, why does the unique
integer n in Definition 1.1.1 exist, and why is it unique? I will not answer
this question in general (the answer probably depends on how you define real

2
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

numbers anyway). However, in the case when x is rational, the proof is simple
(see Corollary 1.1.4 below).
(b) What we call b x c is typically called [ x ] in older books (such as
[NiZuMo91]). I suggest avoiding the notation [ x ] wherever possible; it has
too many different meanings (whereas b x c almost always means the floor of
x).
(c) The map R → Z, x 7→ b x c is called the floor function or the greatest
integer function. There is also a ceiling function, which sends each x ∈ R to the
unique integer n satisfying n − 1 < x ≤ n; this latter integer is called d x e. The
two functions are connected by the rule d x e = − b− x c (for all x ∈ R).
The floor and the ceiling functions are some of the simplest examples of
discontinuous functions.
(d) Here are some examples of floors:

bnc = n for every n ∈ Z;

b1.32c = 1; bπ c = 3; b0.98c = 0;
b−2.3c = −3; b−0.4c = −1.

(e) You might have the impression that b x c is “what remains from x if the
digits behind the comma are removed”. This impression is highly imprecise.
For one, it is completely broken for negative x (for example, b−2.3c is −3, not
−2). But more importantly, the operation of “removing the digits behind the
comma” from a number is not well-defined; the periodic decimal representa-
tions 0.999 . . . and 1.000 . . . belong to the same real number (1), but removing
their digits behind the comma leaves us with different
  integers.
1
(f) A related map is the map R → Z, x 7→ x + . It sends each real x to
2
the integer that is closest to x, choosing the larger one in the case of a tie. This
is one of the many things that are commonly known as “rounding” a number.

Floors of rational numbers are directly related to division with remainder:

Proposition 1.1.3. Let a and b be integers such that b > 0. Let q and r be the
quotient and the remainder obtained when dividing a by b. Then, q is the
a
unique integer n satisfying n ≤ < n + 1.
b

Proof of Proposition 1.1.3. We know that q and r are the quotient and the re-
mainder obtained when dividing a by b. In other words, we have q ∈ Z,
r ∈ {0, 1, . . . , b − 1} and a = qb + r.
From r ∈ {0, 1, . . . , b − 1}, we obtain 0 ≤ r < b. Now, from 0 ≤ r, we obtain
a
qb ≤ qb + r = a. Dividing this inequality by b, we obtain q ≤ (since b > 0).
b
Also, a = qb + |{z} r < qb + b = (q + 1) b. Dividing this inequality by b, we obtain
<b

3
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

a a
< q + 1 (since b > 0). Thus, q ≤ < q + 1. Hence, q is an integer n satisfying
b b
a
n ≤ < n + 1. It thus remains to show that q is the unique such integer. In
b
a
other words, it remains to show that if n is an integer satisfying n ≤ < n + 1,
b
then n = q.
a
So let n be an integer satisfying n ≤ < n + 1. We must show that n = q.
b
a
We have n ≤ < q + 1. Since n and q are integers, this yields n ≤ (q + 1) −
b
1 = q.
a
We have q ≤ < n + 1. Since q and n are integers, this yields q ≤ (n + 1) −
b
1 = n. Combining this with n ≤ q, we obtain n = q. As we said, this completes
our proof.

Corollary 1.1.4. Let x be a rational number.

(a) The integer b x c is well-defined.
a
(b) Write x in the form x = where a and b are integers such that b > 0.
b
Let q and r be the quotient and the remainder obtained when dividing a by b.
Then, b x c = q.
a
Proof of Corollary 1.1.4. Write x in the form x = where a and b are integers
b
such that b > 0. Let q and r be the quotient and the remainder obtained when
dividing a by b. Then, Proposition 1.1.3 yields that q is the unique integer n
a
satisfying n ≤ < n + 1. In other words, q is the unique integer n satisfying
b
a
n ≤ x < n + 1 (since x = ). Thus, the unique integer n satisfying n ≤ x < n + 1
b
exists. Thus, b x c is well-defined. This proves Corollary 1.1.4 (a).
(b) We have just seen that q is the unique integer n satisfying n ≤ x < n + 1.
But this latter integer has been denoted by b x c. Thus, b x c = q. This proves
Corollary 1.1.4 (b).

Proposition 1.1.5. Let m be an integer, and let x be a real number. Then, m ≤ x

holds if and only if m ≤ b x c holds.

Proof of Proposition 1.1.5. Recall that b x c is the unique integer n satisfying n ≤

x < n + 1. Thus, b x c is an integer satisfying b x c ≤ x < b x c + 1.
If m ≤ x holds, then m ≤ b x c holds as well1 . Conversely, if m ≤ b x c holds,
then m ≤ x (because m ≤ b x c ≤ x). Combining these two implications, we
conclude that m ≤ x holds if and only if m ≤ b x c holds. Proposition 1.1.5 is thus
proven.
1 Proof.
Assume that m ≤ x holds. We need to prove that m ≤ b x c holds.
Indeed, assume the contrary. Thus, m > b x c. Hence, m ≥ b x c + 1 (since m and b x c are
integers). Thus, b x c + 1 ≤ m ≤ x, which contradicts x < b x c + 1. This contradiction proves
that our assumption was wrong. Hence, m ≤ b x c is proven, qed.

4
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Corollary 1.1.6. Let x be a real number. Then, b x c is the greatest integer that
is smaller or equal to x.

Proof of Corollary 1.1.6. Clearly, b x c is the greatest integer that is smaller or equal
to b x c. In other words, b x c is the greatest integer m satisfying m ≤ b x c. Equiv-
alently, b x c is the greatest integer m satisfying m ≤ x (since Proposition 1.1.5
shows that the condition m ≤ b x c for an integer m is equivalent to the condition
m ≤ x). In other words, b x c is the greatest integer that is smaller or equal to x.
This proves Corollary 1.1.6.
Corollary 1.1.6 is often used as a definition of b x c. It is also the reason why
the map R → Z, x 7→ b x c is called the greatest integer function.
Before we come to anything interesting, we shall prove a few more basic prop-
erties of the floor function.

Proposition 1.1.7. Let m be an integer. Then, bmc = m.

Proof of Proposition 1.1.7. Clearly, m ≤ m < m + 1. But bmc is the unique integer
n satisfying n ≤ m < n + 1 (because this is how bmc is defined). Hence, if n is
any integer satisfying n ≤ m < n + 1, then n = bmc. Applying this to n = m, we
obtain m = bmc (since m ≤ m < m + 1). This proves Proposition 1.1.7.
The next fact that we shall prove is [NiZuMo91, Theorem 4.1 (5)]:

Proposition 1.1.8. Let x be a real number. Then,

(
0, if x ∈ Z;
b x c + b− x c = .
/Z
−1, if x ∈

Proof of Proposition 1.1.8. We must be in one of the following two cases:

Case 1: We have x ∈ Z.
Case 2: We have x ∈ / Z.
Let us consider Case 1 first. In this case, we have x ∈ Z. In other words, x
is an integer. Hence, − x is an integer as well. Thus, Proposition 1.1.7 (applied
to m = − x) yields that b− x c = − x. But Proposition 1.1.7 (applied to m =
x) yields b x c = x (since x is an integer). Thus, b x c + b− x c = x + (− x ) =
|{z} | {z }
( =x =− x
0, if x ∈ Z;
0. Comparing this with = 0 (since x ∈ Z), we obtain b x c +
−1, if x ∈/Z
(
0, if x ∈ Z;
b− x c = . Thus, Proposition 1.1.8 is proven in Case 1.
−1, if x ∈ /Z
Let us now consider Case 2. In this case, we have x ∈ / Z. Recall that b x c is
the unique integer n satisfying n ≤ x < n + 1 (since this is how b x c is defined).

5
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Thus, b x c is an integer and satisfies b x c ≤ x < b x c + 1. In particular, b x c ∈ Z

(since b x c is an integer).
We cannot have b x c = x (since otherwise, we would have b x c = x ∈ / Z, which
would contradict b x c ∈ Z). Thus, b x c 6= x. Combining this with b x c ≤ x, we
obtain b x c < x.
Now, x < b x c + 1, so that −1 − b x c < − x. Thus, −1 − b x c ≤ − x. On the
other hand, b x c < x, so that − x < − b x c = (−1 − b x c) + 1.
Thus, −1 − b x c ≤ − x < (−1 − b x c) + 1. But b− x c is the unique integer
n satisfying n ≤ − x < n + 1 (since this is how b− x c is defined). Thus, if
n is any integer satisfying n ≤ − x < n + 1, then n = b− x c. Applying this
to n = −1 − b x c, we obtain −1 − b x c = b− x c (since −1 − b x c is an integer
satisfying −1 − b x c ≤ − x < (−1 − b x c) +(1). Hence, −1 = b x c + b− x c, so that
0, if x ∈ Z;
b x c + b− x c = −1. Comparing this with = −1 (since x ∈ / Z),
−1, if x ∈/Z
(
0, if x ∈ Z;
we obtain b x c + b− x c = . Thus, Proposition 1.1.8 is proven in
−1, if x ∈/Z
Case 2.
We have now proven Proposition 1.1.8 in each of the two Cases 1 and 2. Thus,
Proposition 1.1.8 always holds.
Now, let us prove [NiZuMo91, Theorem 4.1 (2)]:

Proposition 1.1.9. Let x be a nonnegative real number. Then, b x c = ∑ 1.

m ∈N+ ;
m≤ x

Proof of Proposition 1.1.9. First of all, 0 ≤ x (since x is nonnegative). But Proposi-

tion 1.1.5 (applied to m = 0) shows that 0 ≤ x holds if and only if 0 ≤ b x c holds.
Thus, 0 ≤ b x c holds (since 0 ≤ x holds). Thus, b x c ∈ N (since b x c is an integer).
Hence,
bxc
∑ 1= ∑ 1 = bxc .
m ∈N+ ; m =1
m≤b x c

Hence,
bxc = ∑ 1= ∑ 1
m ∈N+ ; m ∈N+ ;
m≤b x c m≤ x

(because Proposition 1.1.5 shows that the condition m ≤ b x c for an integer m is

equivalent to the condition m ≤ x). This proves Proposition 1.1.9.
We shall next prove [NiZuMo91, Theorem 4.1 (3)]:

Proposition 1.1.10. Let x be a real number. Let k be an integer. Then,

b x + kc = b x c + k.

6
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Proof of Proposition 1.1.10. Recall that b x c is the unique integer n satisfying n ≤

x < n + 1. Thus, b x c is an integer satisfying b x c ≤ x < b x c + 1.
Also, b x c + k is an integer (since both b x c and k are integers). Also, b x c +k ≤
|{z}
≤x
x +k < b x c + 1 + k = (b x c + k ) + 1, so that b x c + k ≤ x + k <
x + k and |{z}
<b x c+1
(b x c + k) + 1. Thus, b x c + k is an integer n satisfying n ≤ x + k < n + 1.
But b x + kc is the unique integer n satisfying n ≤ x + k < n + 1 (because this is
how b x + k c is defined). Hence, if n is any integer satisfying n ≤ x + k < n + 1,
then n = b x + k c. We can apply this to n = b x c + k (since b x c + k is an integer
n satisfying n ≤ x + k < n + 1), and thus obtain b x c + k = b x + k c. Proposition
1.1.10 is proven.
jnk
Proposition 1.1.11. Let n ∈ N and b ∈ N+ . Then, ∑ 1= .
k∈{1,2,...,n}; b
b|k

Proof of Proposition 1.1.11. The elements of {1, 2, . . . , n} are precisely the elements
k of N+ satisfying k ≤ n. Hence,

∑ 1= ∑ 1= ∑ 1
k∈{1,2,...,n}; k ∈N+ ; k ∈N+ ;
b|k k≤n; b|k;
b|k k≤n
 
here, we have substituted bm for k in
 the sum (since the k ∈ N+ satisfying 
= ∑ 1 
 b | k are precisely the integers of 

m ∈N+ ;
bm≤n the form bm with m ∈ N+ )
 n
= ∑ 1 since bm ≤ n is equivalent to m ≤
b
m ∈N+ ;
n
m≤
j n kb
=
b
n jnk
(because Proposition 1.1.9 (applied to x = ) yields = ∑ 1). Thus,
b b m ∈N+ ;
n
m≤
b
Proposition 1.1.11 is proven.
The floor function is weakly increasing:

Proposition 1.1.12. Let x and y be real numbers such that x ≤ y. Then,

b x c ≤ b y c.

7
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Proof of Proposition 1.1.12. Recall that b x c is the unique integer n satisfying n ≤

x < n + 1. Thus, b x c is an integer satisfying b x c ≤ x < b x c + 1. Hence,
b x c ≤ x ≤ y.
Proposition 1.1.5 (applied to b x c and y instead of m and x) shows that b x c ≤ y
holds if and only if b x c ≤ byc holds. Hence, b x c ≤ byc holds (since b x c ≤ y
holds). This proves Proposition 1.1.12.
Let us now prove [NiZuMo91, Theorem 4.1 (4)]:

Proposition 1.1.13. Let u ∈ R and v ∈ R. Then, buc + bvc ≤ bu + vc ≤

buc + bvc + 1.

Proof of Proposition 1.1.13. All of buc, bvc and bu + vc are integers (by their defi-
nition).
Recall that buc is the unique integer n satisfying n ≤ u < n + 1. Thus, buc is
an integer satisfying buc ≤ u < buc + 1.
Recall that bvc is the unique integer n satisfying n ≤ v < n + 1. Thus, bvc is
an integer satisfying bvc ≤ v < bvc + 1.
Recall that bu + vc is the unique integer n satisfying n ≤ u + v < n + 1. Thus,
bu + vc is an integer satisfying bu + vc ≤ u + v < bu + vc + 1.
Proposition 1.1.5 (applied to m = buc + bvc and x = u + v) shows that buc +
bvc ≤ u + v holds if and only if buc + bvc ≤ bu + vc holds. Thus, buc + bvc ≤
bu + vc holds (since buc + bvc ≤ u + v holds).
|{z} |{z}
≤u ≤v
We know that bvc + 1 is an integer (since bvc is an integer). Hence, Proposition
1.1.10 (applied to x = u and k = bvc + 1) yields bu + bvc + 1c = buc + bvc + 1.
But u + |{z}v < u + bvc + 1. Hence, Proposition 1.1.12 (applied to x = u + v
<bvc+1
and y = u + bvc + 1) shows that bu + vc ≤ bu + bvc + 1c = buc + bvc + 1.
Combining this with buc + bvc ≤ bu + vc, we obtain buc + bvc ≤ bu + vc ≤
buc + bvc + 1.
This proves Proposition 1.1.13.
Finally, let us prove [NiZuMo91, Theorem 4.1 (6)]:

 j k
bxc x
Proposition 1.1.14. Let x ∈ R and m ∈ N+ . Then, = .
m m

Proof of Proposition 1.1.14. Recall that b x c is the unique integer n satisfying n ≤

j x bkx c is an integer satisfying b x c ≤ x < b x c +x1.
x < n + 1. Thus,
Recall that is the unique integer n satisfying n ≤ < n + 1 (by the
j xmk jxk j x k mx jxk
definition of . Thus, is an integer satisfying ≤ < + 1.
m m m m m
k mx∈ N+ and thus m ≥ 1 >
j xBut j x0.k Hence, we can multiply the inequality
≤ by m. We thus obtain m ≤ x.
m m m

8
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

jxk jxk
But Proposition 1.1.5 (applied to m instead of m) shows that m ≤x
jxk m jxk m
holds if and only if m ≤ b x c holds (since m is an integer). Thus,
jxk m j k m
x
m ≤ b x c holds (since m ≤ x holds). Dividing this inequality by m, we
m j k m
x bxc
obtain ≤ .
m m
We can divide the inequality b x c ≤ x by m (since m > 0). We thus obtain
bxc x bxc x jxk
≤ . Hence, ≤ < + 1.
m m j x km b xmc jmx k jxk
So we have ≤ < + 1. In other words, is an integer n
m m m m
bxc
satisfying n ≤ < n + 1.
  m
bxc bxc
But is the unique integer n satisfying n ≤ < n + 1 (because this is
 m m
bxc bxc
how is defined). Hence, if n is any integer satisfying n ≤ < n + 1,
m   m
bxc jxk jxk
then n = . We can apply this to n = (since is an integer n
m m   m
bxc jxk bxc
satisfying n ≤ < n + 1), and thus obtain = . Proposition 1.1.14
m m m
is proven.
I refer to [NiZuMo91, §4.1] for further properties of the floor function.

1.2. Interlude: greatest common divisors

Before we move on, let me remind you of some basic facts about coprime num-
bers and greatest common divisors. First, we recall how greatest common divi-
sors are defined:

Definition 1.2.1. Let b and c be two integers. If (b, c) 6= (0, 0), then gcd (b, c)
means the greatest of all common divisors of b and c. We also set gcd (0, 0) =
0. Thus, gcd (b, c) is defined for any two integers b and c.
If b and c are two integers, then gcd (b, c) is called the greatest common divisor
of b and c (even if gcd (0, 0) is not literally the greatest of all common divi-
sors of 0 and 0) or, briefly, the gcd of b and c. Clearly, gcd (b, c) = gcd (c, b)
and gcd (b, c) | b and gcd (b, c) | c for any two integers b and c. Notice
that gcd (b, c) is a nonnegative integer (and actually a positive integer unless
(b, c) = (0, 0)).

Older books such as [NiZuMo91] tend to denote the gcd of two integers b
and c by (b, c) (rather than by gcd (b, c) as we do); this is a convention that I
shall decidedly not follow (since it risks confusion with the notation (b, c) for
the ordered pair of b and c).

9
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

The most important property of gcds is Bézout’s theorem ([NiZuMo91, Theorem

1.3]):

Theorem 1.2.2. Let b and c be two integers. Then, there exist integers x and y
such that gcd (b, c) = bx + cy.

See [NiZuMo91, Theorem 1.3] for the proof of Theorem 1.2.2 in the case when
(b, c) 6= (0, 0). In the case when (b, c) = (0, 0), Theorem 1.2.2 obviously holds
(since we can take x = 0 and y = 0).
For another proof of Theorem 1.2.2, see the Appendix (Chapter 3) below.
A basic property of gcds that follows directly from Theorem 1.2.2 is the fol-
lowing:

Proposition 1.2.3. Let a, b and c be three integers such that a | b and a | c.

Then, a | gcd (b, c).

In words, Proposition 1.2.3 says that any common divisor of two integers must
divide the gcd of these two integers.
Proof of Proposition 1.2.3. Theorem 1.2.2 shows that there exist integers x and y
such that gcd (b, c) = bx + cy. Consider these x and y. Now, a | b | bx and
a | c | cy. Hence, both integers bx and cy are divisible by a. Thus, their sum
bx + cy must also be divisible by a (by Proposition 1.0.1, applied to u = bx
and v = cy). In other words, a | bx + cy. In other words, a | gcd (b, c) (since
gcd (b, c) = bx + cy). This proves Proposition 1.2.3.

Corollary 1.2.4. Let a, b, c and d be four integers such that a | c and b | d.

Then, gcd ( a, b) | gcd (c, d).

Proof of Corollary 1.2.4. We have gcd ( a, b) | a | c and gcd ( a, b) | b | d. Thus,

Proposition 1.2.3 (applied to gcd ( a, b), c and d instead of a, b and c) shows that
gcd ( a, b) | gcd (c, d). This proves Corollary 1.2.4.
We shall now use Theorem 1.2.2 to derive a slight generalization of [NiZuMo91,
Theorem 1.8]:

Proposition 1.2.5. Let a, b and m be three integers such that gcd ( a, m) = 1.

Then, gcd (b, m) = gcd ( ab, m).

Proof of Proposition 1.2.5. Theorem 1.2.2 (applied to a and m instead of b and c)

shows that there exist integers x and y such that gcd ( a, m) = ax + my. Consider
these x and y. We have ax + my = gcd ( a, m) = 1.
Let g = gcd (b, m) and h = gcd ( ab, m). Thus, g and h are nonnegative inte-
gers.
We have g = gcd (b, m) | b | ab and g = gcd (b, m) | m. Thus, Proposition 1.2.3
(applied to g, ab and m instead of a, b and c) shows that g | gcd ( ab, m). In other
words, g | h (since h = gcd ( ab, m)).

10
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

On the other hand, h = gcd ( ab, m) | ab | abx and h = gcd ( ab, m) | m | mby.
Thus, both integers abx and mby are divisible by h. Therefore, the sum of these
two integers must also be divisible by h (by Proposition 1.0.1, applied to abx,
mby and h instead of u, v and a). In other words, h | abx + mby. Since

abx + mby = b ( ax + my) = b,

| {z }
=1

this rewrites as h | b. So we have h | b and h | m. Thus, Proposition 1.2.3 (applied

to h, b and m instead of a, b and c) shows that h | gcd (b, m). In other words,
h | g (since g = gcd (b, m)).
Now, we can apply Proposition 1.0.2 to u = g and v = h (since g and h are
nonnegative integers satisfying g | h and h | g), and thus we obtain g = h.
Hence, gcd (b, m) = g = h = gcd ( ab, m). This proves Proposition 1.2.5.
Now, let us recall how coprime integers are defined: We say that an integer a
is coprime to an integer b if gcd ( a, b) = 1. The relation of being coprime is clearly
symmetric (i.e., an integer a is coprime to an integer b if and only if b is coprime
to a), because gcd ( a, b) = gcd (b, a). Thus, instead of saying that “a is coprime
to b”, we can also say “the integers a and b are coprime”.
The following corollary is precisely [NiZuMo91, Theorem 1.8]:

Corollary 1.2.6. Let a, b and m be three integers. Assume that a is coprime to

m, and assume that b is coprime to m. Then, ab is coprime to m.

Proof of Corollary 1.2.6. We know that a is coprime to m. In other words, gcd ( a, m) =

1. Also, we have assumed that b is coprime to m. In other words, gcd (b, m) = 1.
Now, Proposition 1.2.5 yields gcd (b, m) = gcd ( ab, m). Thus, gcd ( ab, m) =
gcd (b, m) = 1. In other words, ab is coprime to m. This proves Corollary
1.2.6.
The following corollary generalizes Corollary 1.2.6 to n integers instead of the
two integers a and b:

Corollary 1.2.7. Let c1 , c2 , . . . , cn be n integers. Let m be an integer. Assume

that cu is coprime to m for every u ∈ {1, 2, . . . , n}. Then, c1 c2 · · · cn is coprime
to m.

Proof of Corollary 1.2.7. We shall prove that

c1 c2 · · · c g is coprime to m for every g ∈ {0, 1, . . . , n} . (1)

Proof of (1): We shall prove (1) by induction on g:

Induction base: We have c1 c2 · · · c0 = (empty product) = 1 (since empty prod-
ucts are 1 by definition). But 1 is clearly coprime to m (since gcd (1, m) | 1
and thus gcd (1, m) = 1). In other words, c1 c2 · · · c0 is coprime to m (since

11
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

c1 c2 · · · c0 = 1). In other words, (1) holds for g = 0. This completes the induction
base.
Induction step: Let G ∈ {0, 1, . . . , n} be positive. Assume that (1) holds for
g = G − 1. We must prove that (1) holds for g = G.
We have G ∈ {0, 1, . . . , n}, and thus G ∈ {1, 2, . . . , n} (since G is positive).
We know that c1 c2 · · · cG−1 is coprime to m (since we assumed that (1) holds for
g = G − 1). Also, we assumed that cu is coprime to m for every u ∈ {1, 2, . . . , n}.
Applying this to u = G, we see that cG is coprime to m. Now, Corollary 1.2.6 (ap-
plied to c1 c2 · · · cG−1 and cG instead of a and b) shows that (c1 c2 · · · cG−1 ) cG is co-
prime to m. In other words, c1 c2 · · · cG is coprime to m (since (c1 c2 · · · cG−1 ) cG =
c1 c2 · · · cG ). In other words, (1) holds for g = G. This completes the induction
step, and thus (1) is proven.
Now, applying (1) to g = n, we conclude that c1 c2 · · · cn is coprime to m. This
proves Corollary 1.2.7.
A further consequence of Proposition 1.2.5 is the following fact ([NiZuMo91,
Theorem 1.10]):

Proposition 1.2.8. Let x, y and z be three integers such that x | yz and

gcd ( x, y) = 1. Then, x | z.

Proof of Proposition 1.2.8. We have x | yz and x | x. Hence, Proposition 1.2.3

(applied to x, yz and x instead of a, b and c) shows that x | gcd (yz, x ).
We have gcd (y, x ) = gcd ( x, y) = 1. Hence, Proposition 1.2.5 (applied to a = y,
b = z and m = x) shows that gcd (z, x ) = gcd (yz, x ). Now, x | gcd (yz, x ) =
gcd (z, x ) | z. This proves Proposition 1.2.8.
Another important result on gcds is the following fact:

Proposition 1.2.9. Let g be a positive integer. Let a and b be two integers.

Then, g gcd ( a, b) = gcd ( ga, gb).

Proof of Proposition 1.2.9. Both gcd ( a, b) and g are nonnegative integers; hence,
g gcd ( a, b) is a nonnegative integer.
We have g gcd ( a, b) | ga and g gcd ( a, b) | gb. Thus, Proposition 1.2.3 (ap-
| {z } | {z }
|a |b
plied to g gcd ( a, b), ga and gb instead of a, b and c) shows that g gcd ( a, b) |
gcd ( ga, gb).
On the other hand, Theorem 1.2.2 (applied to a and b instead of b and c) shows
that there exist integers x and y such that gcd ( a, b) = ax + by. Consider these
x and y. We have gcd ( ga, gb) | ga | gax and gcd ( ga, gb) | gb | gby. Thus, both
integers gax and gby are divisible by gcd ( ga, gb). Therefore, their sum gax + gby
is also divisible by gcd ( ga, gb) (by Proposition 1.0.1, applied to gcd ( ga, gb), gax
and gby instead of a, u and v). In other words, we have gcd ( ga, gb) | gax +

12
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

gby. Since gax + gby = g ( ax + by) = g gcd ( a, b), this rewrites as gcd ( ga, gb) |
| {z }
=gcd( a,b)
g gcd ( a, b).
But we can apply Proposition 1.0.2 to u = g gcd ( a, b) and v = gcd ( ga, gb)
(since g gcd ( a, b) and gcd ( ga, gb) are nonnegative integers satisfying g gcd ( a, b) |
gcd ( ga, gb) and gcd ( ga, gb) | g gcd ( a, b)), and thus we obtain g gcd ( a, b) =
gcd ( ga, gb). This proves Proposition 1.2.9.
Here is another property of gcds, which we will use later:

Proposition 1.2.10. Let m and n be two coprime positive integers. Let u be an

integer. Then, gcd (u, m) · gcd (u, n) = gcd (u, mn).

Proof of Proposition 1.2.10. Set h = gcd (u, mn), v = gcd (u, m) and w = gcd (u, n).
We shall prove that vw = h.
We have v = gcd (u, m) ∈ N+ (since m is positive) and w = gcd (u, n) ∈ N+
(since n is positive). Thus, both v and w are positive integers; hence, vw is a
positive integer. Also, mn is positive (since m and n are positive). Now, h =
gcd (u, mn) ∈ N+ (since mn is positive).
We have v = gcd (u, m) | m and w = gcd (u, n) | n. Hence, Corollary 1.2.4
(applied to v, w, m and n instead of a, b, c and d) yields gcd (v, w) | gcd (m, n) = 1
(since m and n are coprime). Hence, gcd (v, w) = 1.
u u
We have w = gcd (u, n) | u and thus ∈ Z. Now, v = gcd (u, m) | u = w · .
w w
u
Thus, Proposition 1.2.8 (applied to v, w and instead of x, y and z) shows that
w
u u u u
v| (since gcd (v, w) = 1). In other words, /v ∈ Z. Now, = /v ∈ Z,
w w vw w
so that vw | u.
m
But v = gcd (u, m) | m and thus ∈ Z. Also, w = gcd (u, n) | n and thus
v
n mn m n m n
∈ Z. Now, = · is the product of two integers (since and are
w vw v w v w
integers), and thus itself an integer. In other words, vw | mn.
So we have vw | u and vw | mn. Proposition 1.2.3 (applied to vw, u and mn
instead of a, b and c) thus yields vw | gcd (u, mn) = h.
Proposition 1.2.9 (applied to n, u and m instead of g, a and b) shows that
n gcd (u, m) = gcd (nu, nm). Thus, gcd (nu, nm) = n gcd (u, m) = nv = vn.
| {z }
=v
Now, h = gcd (u, mn) | mn = nm and h = gcd (u, mn) | u | nu. Hence,
Proposition 1.2.3 (applied to h, nu and nm instead of a, b and c) shows that
h | gcd (nu, nm). In other words, h | vn (since gcd (nu, nm) = vn).
Proposition 1.2.9 (applied to v, u and n instead of g, a and b) shows that
v gcd (u, n) = gcd (vu, vn). Thus, gcd (vu, vn) = v gcd (u, n) = vw.
| {z }
=w
Now, h | u | vu and h | vn. Thus, Proposition 1.2.3 (applied to h, vu and
vn instead of a, b and c) shows that h | gcd (vu, vn). In other words, h | vw

13
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

(since gcd (vu, vn) = vw). Combining this with vw | h, we obtain h = vw (by
Proposition 1.0.2 (applied to h and vw instead of u and v), since h and vw are
positive integers). Now,

gcd (u, m) · gcd (u, n) = vw = h = gcd (u, mn) .

| {z } | {z }
=v =w

This proves Proposition 1.2.10.

Let us derive an easy corollary from Corollary 1.2.7:

Corollary 1.2.11. Let a and b be two coprime integers. Let n ∈ N.

(a) The integer an is coprime to b.
(b) Let m ∈ N. Then, the integer an is coprime to bm .

Proof of Corollary 1.2.11. (a) For each i ∈ {1, 2, . . . , n}, the integer a is coprime to
b. Thus, Corollary 1.2.7 (applied to ci = a and m = b) shows that |aa {z · · · }a is
n times
coprime to b. In other words, an is coprime to b (since an = |aa {z
· · · }a). This proves
n times
Corollary 1.2.11 (a).
(b) Corollary 1.2.11 (a) shows that the integer an is coprime to b. In other
words, b and an are two coprime integers. Hence, Corollary 1.2.11 (a) (applied
to b, an and m instead of a, b and n) shows that the integer bm is coprime to
an . In other words, the integer an is coprime to bm . This proves Corollary 1.2.11
(b).
Let us next use the above theory of greatest common divisors to prove some
properties of primes. We begin with the probably most basic one:

Proposition 1.2.12. Let p be a prime. Let a be an integer such that p - a. Then,

a is coprime to p.

Proof of Proposition 1.2.12. If we had gcd ( a, p) = p, then we would have p =

gcd ( a, p) | a, which would contradict p - a. Hence, we cannot have gcd ( a, p) =
p.
We have ( a, p) 6= (0, 0) (since p 6= 0). Thus, gcd ( a, p) is a positive integer.
Hence, gcd ( a, p) is a positive divisor of p (since gcd ( a, p) | p). But the only
positive divisors of p are 1 and p (since p is prime). Hence, gcd ( a, p) must be
either 1 or p (since gcd ( a, p) is a positive divisor of p). Thus, we must have
gcd ( a, p) = 1 (since we cannot have gcd ( a, p) = p). In other words, a is coprime
to p. This proves Proposition 1.2.12.

Corollary 1.2.13. Let s and t be two distinct primes.

(a) The integers s and t are coprime.
(b) Let n ∈ N and m ∈ N. Then, the integers sn and tm are coprime.

14
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Proof of Corollary 1.2.13. (a) Assume the contrary. Thus, the integers s and t are
not coprime. In other words, s is not coprime to t.
If we had s - t, then s would be coprime to t (by Proposition 1.2.12, applied
to p = s and a = t), which would contradict the fact that s is not coprime to t.
Thus, we cannot have s - t. Hence, we have s | t. The same argument (with the
roles of s and t switched) yields t | s. Thus, Proposition 1.0.2 (applied to u = s
and v = t) yields s = t. But this contradicts the fact that s and t are distinct. This
contradiction shows that our assumption was wrong. Hence, Corollary 1.2.13
(a) is proven.
(b) Corollary 1.2.13 (a) shows that the integers s and t are coprime. Hence,
Corollary 1.2.11 (b) (applied to a = s and b = t) shows that the integer sn is
coprime to tm . In other words, the integers sn and tm are coprime. This proves
Corollary 1.2.13 (b).2

Proposition 1.2.14. Let p be a prime. Let a1 , a2 , . . . , an be n integers. Assume

that p | a1 a2 · · · an . Then, there exists an i ∈ {1, 2, . . . , n} such that p | ai .

Proof of Proposition 1.2.14. Assume the contrary. Thus, there exists no i ∈ {1, 2, . . . , n}
such that p | ai . In other words, for each u ∈ {1, 2, . . . , n}, we have p - au . Hence,
for each u ∈ {1, 2, . . . , n}, the number au is coprime to p (by Proposition 1.2.12,
applied to a = au ). In other words, au is coprime to p for each u ∈ {1, 2, . . . , n}.
Hence, Corollary 1.2.7 (applied to m = p and ci = ai ) shows that a1 a2 · · · an is
coprime to p. In other words, gcd ( p, a1 a2 · · · an ) = 1.
But p | p and p | a1 a2 · · · an . Hence, Proposition 1.2.3 (applied to a = p, b = p
and c = a1 a2 · · · an ) yields p | gcd ( p, a1 a2 · · · an ) = 1. Combined with 1 | p, this
yields p = 1 (by Proposition 1.0.2, applied to u = p and v = 1). This is absurd,
since p is a prime. This contradiction shows that our assumption was wrong;
hence, Proposition 1.2.14 is proven.

Remark 1.2.15. Proposition 1.2.14 can be restated as follows: If a prime p

divides a product of finitely many integers, then p must divide (at least) one
of these integers.

2 Here is a different way to prove Corollary 1.2.13 (b) using the uniqueness of prime factoriza-
tion:
Assume the contrary. Thus, the integers sn and tm are not coprime. In other words,
gcd (sn , tm ) > 1. Hence, the integer gcd (sn , tm ) has a prime divisor q. Consider this q.
But s is a prime. Thus, the prime factorization of sn is sn = |s · s ·{z · · · · }s. Hence, the only
n times
prime divisor of sn is s.
But q | gcd (sn , tm ) | sn . Thus, q is a prime divisor of sn (since q is a prime and divides sn ).
Since the only prime divisor of sn is s, this shows that q = s. The same argument (applied to
t instead of s) shows that q = t. Hence, s = q = t. This contradicts the fact that s and t are
distinct. This contradiction proves that our assumption was wrong, qed.

15
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

1.3. de Polignac’s formula

One of the most useful applications of the floor function is computing the p-adic
valuation of factorials. Let us first define our notations:
Definition 1.3.1. Let p be a prime. Let n be a nonzero integer. Then, v p (n)
is defined to be the highest nonnegative integer k such that pk | n. This
nonnegative integer v p (n) is called the p-adic valuation of n.

Remark 1.3.2. Some authors use the notation e p (n) instead of v p (n).
Another way to characterize v p (n) in Definition 1.3.1 is by the following
n
statement: The number v (n) is an integer not divisible by p.
p p
Yet another (probably simpler) way to define v p (n) is the following: v p (n)
is the exponent with which p occurs in the prime factorization of n. 3 (This
is clearly equivalent to the definition of v p (n) above.) While I will try to avoid
using prime factorizations wherever I can, there should be nothing stopping
you from using them; in general, the prime factorization of n is probably the
quickest way to get an intuition for v p (n) (although not the quickest way to
compute it!).
Often, the definition of v p (n) is extended to all rational numbers n. Then,
one defines v p (n) to be the unique integer k (not necessarily nonnegative)
n
such that the rational number k can be written as a fraction whose numerator
p
and denominator are both integers coprime to p. This works when n 6= 0. In
the case of n = 0, one commonly defines v p (0) to be −∞; here, −∞ is a
symbol which (when it comes to comparing it with integers) is smaller than
every integer.

Theorem 1.3.3 (de Polignac’s formula). Let p be a prime. Let n ∈ N. Then,

∞  
n
v p (n!) = ∑ i
.
i =1 p

(The sum on the right hand side is infinite, but only finitely many of its terms
are nonzero, and thus it is a well-defined integer.)
Before we prove this theorem, here are two simple lemmas:
Lemma 1.3.4. Let p be a prime. Let n be a nonzero integer. Then,
v p (n) = ∑ 1.
i ∈N+ ;
pi | n

3 This exponent should be understood as 0 if p does not occur in the prime factorization of n at
all.

16
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Proof of Lemma 1.3.4. Recall that v p (n) is defined as the highest nonnegative in-
teger k such that pk | n. Thus, v p (n) is a nonnegative integer satisfying pv p (n) | n.
Every i ∈ N+ that satisfies i ≤ v p (n) must satisfy pi | n (since i ≤ v p (n)
leads to pi | pv p (n) | n). Conversely, every i ∈ N+ satisfying pi | n must satisfy
i ≤ v p (n) (since v p (n) is the highest nonnegative integer k such that pk | n).
Thus, the i ∈ N+ that satisfy i ≤ v p (n) are exactly the i ∈ N+ that satisfy pi | n.
Consequently,
∑ 1 = ∑ 1.
i ∈N+ ; i ∈N+ ;
i ≤v p (n) pi | n

v p (n)
Hence, ∑ 1= ∑ 1 = ∑ 1 = v p (n). This proves Lemma 1.3.4.
i ∈N+ ; i ∈N+ ; i =1
pi | n i ≤v p (n)

Lemma 1.3.5. Let p be a prime. Let a1 , a2 , . . . , an be finitely many nonzero

integers. Then,

v p ( a1 a2 · · · a n ) = v p ( a1 ) + v p ( a2 ) + · · · + v p ( a n ) .

This lemma is fairly obvious if you follow the “exponent in prime factoriza-
tion” interpretation of v p (n). The proof below avoids this interpretation (for the
sake of greater generalizability).
Proof of Lemma 1.3.5. Lemma 1.3.5 can be proven straightforwardly by induction
over n, provided that the following two claims are shown:

Claim 1: We have v p (1) = 0.

Claim 2: We have v p ( ab) = v p ( a) + v p (b) whenever a and b are two

nonzero integers.

(In fact, Claim 1 settles the induction base, while Claim 2 is used in the induc-
tion step.)
Claim 1 is obvious. It thus remains to prove Claim 2.
Proof of Claim 2: Let a and b be two nonzero integers. Recall that v p ( a) is
defined as the highest nonnegative integer k such that pk | a. Thus, v p ( a) is a
nonnegative integer satisfying pv p (a) | a, but pv p (a)+1 - a. We have pv p (a) | a; thus,
we can write a in the form a = pv p (a) g for some g ∈ Z. Consider this g. If we
had p | g, then we would have pv p (a)+1 = pv p (a) p | pv p (a) g = a; this would
|{z}
|g
v p ( a)+1
contradict p - a. Hence, we cannot have p | g. We thus must have p - g.
Hence, Proposition 1.2.12 (applied to g instead of a) shows that g is coprime to
p (since p is a prime).

17
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Thus, we have found a g ∈ Z which is coprime to p and satisfies a = pv p (a) g.

The same argument (but made for b instead of a) shows that there exists an
h ∈ Z which is coprime to p and satisfies b = pv p (b) h. Consider this h.
Both g and h are coprime to p. Hence, gh is also coprime to p (by Corollary
1.2.6, applied to g, h and p instead of a, b and m). Therefore, ( gh)1 is also coprime
to pv p (ab) (by Corollary 1.2.11 (b), applied to gh, p, 1 and v p ( ab) instead of a, b,
n and m). In  other words,
 gh is coprime to pv p (ab) (since ( gh)1 = gh). In other
words, gcd pv p (ab) , gh = 1.
Multiplying the equalities a = pv p (a) g and b = pv p (b) h, we obtain ab =
pv p (a) gpv p (b) h = pv p (a) pv p (b) gh = pv p (a)+v p (b) gh. Hence, pv p (a)+v p (b) | ab.
On the other hand, recall that v p ( ab) is defined as the highest nonnegative
integer k such that pk | ab. Thus, v p ( ab) is a nonnegative  integer and  satisfies
pv p (ab) | ab = pv p (a)+v p (b) gh = ghpv p (a)+v p (b) . Since gcd pv p (ab) , gh = 1, this
v p ( ab) v p ( a)+v p (b)
entails that p | p (by Proposition 1.2.8, applied to x = pv p (ab) ,
y = gh and z = pv p (a)+v p (b) ). Therefore, v p ( ab) ≤ v p ( a) + v p (b).
But v p ( ab) is the highest nonnegative integer k such that pk | ab. Hence, every
nonnegative integer k such that pk | ab must satisfy k ≤ v p ( ab). Applying this
to k = v p ( a) + v p (b), we obtain v p ( a) + v p (b) ≤ v p ( ab) (since v p ( a) + v p (b)
satisfies pv p (a)+v p (b) | ab). Combined with v p ( ab) ≤ v p ( a) + v p (b), this yields
v p ( ab) = v p ( a) + v p (b). This proves Claim 2; and as we said, this completes the
proof of Lemma 1.3.5.

18
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Proof of Theorem 1.3.3. We have

 

n!  = v p (1 · 2 · · · · · n) = v p (1) + v p (2) + · · · + v p (n)

v p  |{z}
=1·2·····n
(by Lemma 1.3.5, applied to ai = i )
n
= ∑ v p (k) = ∑ ∑ 1
k =1 | {z } k ∈N+ ; i ∈N+ ;
|{z} = ∑ 1 k≤n pi | k
= ∑ i ∈N+ ;
k ∈N+ ;
| {z }
pi | k = ∑ ∑
k≤n (by Lemma 1.3.4, applied to k i ∈N+ k ∈N+ ;
instead of n) k≤n;
pi | k
(here, we are interchanging
the order of summation)
∞
= ∑ ∑ 1 = ∑ ∑ 1
i ∈N+ k ∈N+ ; i =1 k∈{1,2,...,n};
|{z} k≤n; pi | k
∞
=∑ pi | k | ${z % }
i =1
| {z } n
= ∑ 1 =
k∈{1,2,...,n}; pi
pi | k (by Proposition 1.1.11,
(since the elements k of N+ applied to b= pi )
satisfying k≤n are precisely
the elements of {1,2,...,n})
∞  
n
= ∑ pi
.
i =1

This proves Theorem 1.3.3.

As an application of Theorem 1.3.3, we can check that binomial coefficients
are integers (as long as the inputs are nonnegative integers):
 
n
Corollary 1.3.6. Let n ∈ N and m ∈ N. Then, ∈ Z.
m

Of course, Corollary 1.3.6 can be proven in various simple ways – for example,
by induction using the recurrence
  relation of the binomial coefficients, or com-
n
binatorially by interpreting as the number of m-element subsets of a given
m
n-element set. But let us prove it using Theorem 1.3.3, just to show how to use
the latter:

Lemma 1.3.7. Let a and b be two nonzero integers. Assume that v p ( a) ≥ v p (b)
for every prime p. Then, b | a.

19
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

First proof of Lemma 1.3.7. Let P be the set of all primes. Every nonzero integer n
satisfies n = ± ∏ pv p (n) 4 . Thus,
p ∈P

a = ± ∏ pv p ( a) and b = ± ∏ pv p (b) . (2)

p ∈P p ∈P

(The two ± signs may and may not be equal.)

But every p ∈ P satisfies pv p (b) | pv p (a) (since v p ( a) ≥ v p (b)). Hence, ∏ pv p (b) |
p ∈P
v p ( a)
∏ p . In light of (2), this becomes b | a (indeed, the ± signs clearly have no
p ∈P
effect on the divisibility). This proves Lemma 1.3.7.
Second proof of Lemma 1.3.7. Let me, again, give a proof which avoids the use of
prime factorizations. As before, this comes at the cost of brevity (but again, it
leads to more generality).
Let g = gcd ( a, b). Then, g = gcd ( a, b) | a. Hence, there exists some a0 ∈ Z
such that a = ga0 . Consider this a0 . Clearly, a0 is nonzero (since ga0 = a is
nonzero).
Also, g = gcd ( a, b) | b. Hence, there exists some b0 ∈ Z such that b = gb0 .
Consider this b0 . Clearly, b0 is nonzero (since gb0 = b is nonzero). Of course, g
is also nonzero (since g = gcd ( a, b) with a and b being nonzero). Moreover, g is
nonnegative (since g = gcd ( a, b)) and therefore positive (since g is nonzero).
0 0 0 0
Proposition
  (applied to a and b instead of a and b) shows that g gcd ( a , b ) =
1.2.9

gcd  ga0 , gb0  = gcd ( a, b) = g. Cancelling g from this equality (since g is

|{z} |{z}
=a =b
nonzero), we obtain gcd ( a0 , b0 ) = 1.
Let p be any prime dividing b0 . We shall derive a contradiction (thus conclud-
ing that no such primes exist).
We have p | b0 (since p is a prime dividing b0 ). But recall that v p (b0 ) is defined
as the highest nonnegative integer k such that pk | b0 . Thus, every nonnegative
integer k such that pk | b0 must satisfy k ≤ v p (b0 ). Applying this to k = 1, we
obtain 1 ≤ v p (b0 ) (since p1 = p | b0 ). Hence, v p (b0 ) ≥ 1.
0 0
Now, Lemma 1.3.5 (applied  ton = 2, a1 = g and a2 = a ) yields v p ( ga ) =

v p ( g) + v p ( a0 ). Thus, v p |{z}
a  = v p ( ga0 ) = v p ( g) + v p ( a0 ). The same ar-
= ga0
gument (used for b and b0
instead of a and a0 ) yields v p (b) = v p ( g) + v p (b0 ).
But by assumption, we have v p ( a) ≥ v p (b). Thus, v p ( g) + v p ( a0 ) = v p ( a) ≥
4 Indeed,for any prime p, we know that v p (n) is the exponent with which the prime p appears
in the prime factorization of n. Hence, the prime factorization of n is ± ∏ pv p (n) . (The ±
p ∈P
sign is due to the fact that n can be negative.)

20
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

v p (b) = v p ( g) + v p (b0 ). Since v p ( g) is an integer, we can cancel v p ( g) from this

inequality, and obtain v p ( a0 ) ≥ v p (b0 ) ≥ 1.
Recall that v p ( a0 ) is defined as the highest nonnegative integer k such that
0 0
pk | a0 . Thus, pv p (a ) | a0 . But v p ( a0 ) ≥ 1, whence p | pv p (a ) | a0 .
Now, p | a0 and p | b0 . Hence, Proposition 1.2.3 (applied to p, a0 and b0 instead
of a, b and c) shows that p | gcd ( a0 , b0 ), so that p | 1 (since gcd ( a0 , b0 ) = 1). This
is absurd (since p is a prime).
Now, forget that we fixed p. Thus, we have obtained a contradiction for every
prime p dividing b0 . Therefore, there exist no such primes p. Therefore, b0 is
either 1 or −1. In either case, b0 | 1. Hence, b = g |{z} b0 | g1 = g | a. This proves
|1
Lemma 1.3.7.
 
n
Proof of Corollary 1.3.6 (sketched). If m > n, then = 0; thus, Corollary 1.3.6
m
is obviously correct in this case. Hence, we WLOG assume that we don’t have
m >
  n. Therefore, m ≤ n. Consequently, a well-known  formula shows that
n n! n
= . Hence, in order to prove that ∈ Z, it suffices to
m m! (n − m)! m
show that m! (n − m)! | n!. In light of Lemma 1.3.7 (applied to a = n! and
b = m! (n − m)!), we can achieve this by showing that

v p (n!) ≥ v p (m! (n − m)!) for every prime p. (3)

Proof of (3): Let p be a prime. Lemma 1.3.5 yields

v p (m! (n − m)!)
= v p (m!) + v p ((n − m)!)
| {z } | $ {z }
∞ m ∞ n−m
$ % %
=∑ =∑
i =1 p i i =1 pi
(by Theorem 1.3.3) (by Theorem 1.3.3)
∞   ∞  ∞
n−m n−m
    
m m
= ∑ p i
+∑
pi
= ∑ pi
+
pi
i =1 i =1 i =1 | {z }
m n−m
$ %
≤ +
pi pi
(by the formula buc+bvc≤bu+vc
from Proposition 1.1.13)
 
 
 
 
 
∞   ∞  
m n − m  = ∑ n = v p (n!)
≤ ∑

 i + (by Theorem 1.3.3) .
i =1  |p pi  pi
 {z  i =1
}
 n 
=
p i

21
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

This proves (3).

As we know, this completes the proof of Corollary 1.3.6.
Note that Corollary 1.3.6 also holds for all n ∈ Z (not just for all n ∈ N); but
this would require a different method of proof5 .
Our proof of Corollary 1.3.6 using Theorem 1.3.3 was a slight overkill (as I
said, there are easier and better ways to achieve the same goal); however, the
method is useful, as it also allows proving other results which are harder to
obtain in other ways. Here are two examples of such results (without proof):
 
a/b
Proposition 1.3.8. Let a ∈ Z, b ∈ Z \ {0} and m ∈ N. Then, is a
m
rational number which can be written as a ratio of
 two integers whose de-
a/b
nominator is a power of b. More precisely, b2m−1 ∈ Z when m > 0
  m
a/b
(and = 1 ∈ Z when m = 0).
m

(2m)! (2n)!
Proposition 1.3.9. Let m ∈ N and n ∈ N. Then, ∈ Z.
m!n! (m + n)!

2. Arithmetic functions
2.1. Arithmetic functions
Next, I will discuss the notion of arithmetic functions, and some examples
thereof; here I will not really follow [NiZuMo91, §4.2] but rather build up the
same theory from my perspective.

Definition 2.1.1. An arithmetic function shall mean a function from N+ to C.

My Definition 2.1.1 appears to be slightly incompatible with the definition in

[NiZuMo91, §4.2]; indeed, the latter defines an arithmetic function to be a func-
tion from N+ to a subset of C. However, Niven, Zuckerman and Montgomery
never specify the target of the arithmetic functions they introduce in [NiZuMo91,
§4.2]; thus, I believe that my Definition 2.1.1 is the definition they have actually
meant. Anyway, most people are cavalier about the target of an arithmetic func-
tion, and prefer to equate any two arithmetic functions which differ only in the
choice of target.

 the n ∈ Z case to the n ∈ N case is by using the upper negation

5 Theeasiest
 way to reduce
n m m−n−1
formula = (−1) .
m m

22
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Let us define a bunch of arithmetic functions:6

Definition 2.1.2. We define the following arithmetic functions:

• The function φ : N+ → C shall send every n ∈ N+ to the number of all
k ∈ {1, 2, . . . , n} coprime to n. This function φ is called the Euler totient
function, or the phi function (and is often denoted by ϕ as well).
• The function d : N+ → C shall send every n ∈ N+ to the number of
positive divisors of n. This function d is called the divisor function.
• The function 1 : N+ → C shall send every n ∈ N+ to 1.
• The function 0 : N+ → C shall send every n ∈ N+ to 0.
• The function ι : N+ → C shall send every n ∈ N+ to n.
• The function σ : N+ → C shall send every n ∈ N+ to the sum of all
positive divisors of n.
• For each k ∈ Z, the function σk : N+ → C shall send every n ∈ N+ to
the sum of the k-th powers of all positive divisors of n. Note that σ0 = d
and σ1 = σ.
• The function ω : N+ → C shall send every n ∈ N+ to the number of all
distinct primes dividing n. (For example, ω (12) = 2, since the primes
dividing 12 are 2 and 3.)
• The function Ω : N+ → C shall send every n ∈ N+ to the number
of all prime factors of n counted with multiplicity. In other words, if
Ω (n) is the k ∈ N such that n can be written as a product of k primes
(not necessarily distinct primes). (For example, Ω (12) = 3, since 12 =
2 · 2 · 3.)
• The
( function µ : N+ → C shall send every n ∈ N+ to
(−1)ω (n) , if n is squarefree;
. This function µ is called the Möbius mu
0, otherwise
function.

• The function λ : N+ → C shall send every n ∈ N+ to (−1)Ω(n) . This

function λ is called Liouville’s lambda function.
(
1, if n = 1;
• The function ε : N+ → C shall send every n ∈ N+ to .
0, if n 6= 1

6 Recall that a positive integer n is said to be squarefree if no perfect square other than 1 divides
n. Equivalently, a positive integer n is squarefree if and only if n is a product of distinct
primes. Equivalently, a positive integer n is squarefree if and only if every prime p satisfies
v p (n) ≤ 1.

23
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Of course, you can come up with more examples easily. Most arithmetic func-
tions that anyone cares about tend to have their images belong to Z, but the
added generality of allowing any complex numbers as images does not hurt, so
I see no point in restricting it.
We introduce one more standard notation:

Definition 2.1.3. Any summation sign of the form “ ∑ ” (where n is a given

d|n
positive integer) will be understood to mean “sum over all positive divisors
d of n”. This similarly applies when there are further conditions under the
summation sign; for instance, “ ∑ ” means “sum over all positive divisors d
d|n;
d ≤3
of n satisfying d ≤ 3”.

Remark 2.1.4. Some of the functions defined in Definition 2.1.2 can easily
be reexpressed using the notation from Definition 2.1.3: Namely, for every
n ∈ N+ , we have

d (n) = ∑ 1; σ (n) = ∑ d;
d|n d|n

σk (n) = ∑d k
(for every k ∈ Z) .
d|n

Some of the arithmetic functions defined above can be written explicitly in

terms of the prime factorization of n. I will first state some of the explicit repre-
sentations before I show a method for proving them.

Definition 2.1.5. If n is an integer, then PF n will denote the set of all prime
factors of n. Note that this set PF n is finite whenever n is nonzero.

Theorem 2.1.6. For every n ∈ N+ , we have

 
φ (n) = ∏ pv p (n)−1 ( p − 1) .
p∈PF n

Theorem 2.1.7. For every n ∈ N+ , we have

∏


d (n) = v p (n) + 1 .
p∈PF n

24
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Theorem 2.1.8. For every n ∈ N+ and every nonzero k ∈ Z, we have

pk(v p (n)+1) − 1
σk (n) = ∏ pk − 1
.
p∈PF n

Theorem 2.1.6 appears in [NiZuMo91, Theorem 2.19] (in a slightly restated

form). Theorem 2.1.7 is [NiZuMo91, Theorem 4.3], and Theorem 2.1.8 is a
straightforward generalization of [NiZuMo91, Theorem 4.5]. There are simple
and elementary ways to prove each of these theorems; I will give a more abstract
approach to highlight the theory.

2.2. Multiplicative functions

Definition 2.2.1. Let f : N+ → C be an arithmetic function.
(a) The function f is said to be multiplicative if and only if it satisfies f (1) =
1 and

f (mn) = f (m) f (n) for any two coprime m ∈ N+ and n ∈ N+ .

(b) The function f is said to be totally multiplicative if and only if it satisfies

f (1) = 1 and

f (mn) = f (m) f (n) for any two m ∈ N+ and n ∈ N+ .

(Another word for “totally multiplicative” is “completely multiplicative”.)

It turns out that totally multiplicative functions are somewhat rare, but multi-
plicative functions abound in number theory. Here are some examples:

Proposition 2.2.2. Consider the functions defined in Definition 2.1.2.

(a) The function φ is multiplicative.
(b) The function d is multiplicative.
(c) The function 1 is totally multiplicative and multiplicative.
(d) The function ι is totally multiplicative and multiplicative.
(e) For every k ∈ Z, the function σk is multiplicative. In particular, the
function σ is multiplicative.
(f) The function µ is multiplicative.
(g) The function λ is totally multiplicative and multiplicative.
(h) The function ε is totally multiplicative and multiplicative.
(i) Every totally multiplicative function is multiplicative.
(j) Let f ∈ Z [ x ] be a polynomial. Let NP : N+ → C be the function
which sends every n ∈ N+ to the number of solutions of the congruence
f ( x ) ≡ 0 mod n. Then, the function NP is multiplicative.

25
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

(k) For every integer u, the function N+ → C, n 7→ gcd (u, n) is multiplica-

tive.

Proof of Proposition 2.2.2 (sketched). (i) This is obvious (since the requirements for
a totally multiplicative function clearly encompass the requirements for a multi-
plicative function).
(a) We know that φ (1) = 1. We thus only need to show that φ (mn) =
φ (m) φ (n) for any two coprime m ∈ N+ and n ∈ N+ . But this is precisely
the statement of [NiZuMo91, first sentence of Theorem 2.19]. (Here is a brief
reminder of the proof: For every N ∈ N+ , let R ( N ) denote the set of all
k ∈ {1, 2, . . . , N } coprime to N. Now, let m ∈ N+ and n ∈ N+ be coprime. Then,
there is a bijection R (mn) → R (m) × R (n) which sends every k ∈ R (mn) to
(k0 , k00 ) ∈ R (m) × R (n), where k0 is the unique element of R (m) congruent
to k modulo m, and where k00 is the unique element of R (n) congruent to k
modulo n. The fact that this map is well-defined and a bijection can be proven
using the Chinese Remainder Theorem. Having this bijection in place, we im-
mediately conclude that |R (mn)| = |R (m) × R (n)| = |R (m)| · |R (n)|. Since
|R ( N )| = φ ( N ) for every N ∈ N+ , this rewrites as φ (mn) = φ (m) φ (n), qed.)
Proposition 2.2.2 (a) is thus proven.
Proposition 2.2.2 (j) is essentially [NiZuMo91, second sentence of Theorem
2.20], and is proven in a similar way as Proposition 2.2.2 (a).
Parts (c), (d) and (h) of Proposition 2.2.2 are completely straightforward.
(g) We claim that the following two assertions hold:

Assertion 1: We have λ (1) = 1.

Assertion 2: We have λ (mn) = λ (m) λ (n) for any two m ∈ N+ and

n ∈ N+ .

Proof of Assertion 1. The number 1 can be written as a product of 0 primes (be-

cause the empty product equals 1). Hence, Ω (1) = 0 (by the definition of Ω).
The definition of λ yields λ (1) = (−1)Ω(1) = (−1)0 (since Ω (1) = 0). Thus,
λ (1) = (−1)0 = 1. This proves Assertion 1.
Proof of Assertion 2. Let m ∈ N+ and n ∈ N+ .
Write m as a product of primes; i.e., write m in the form m = p1 p2 · · · pk for
some primes p1 , p2 , . . . , pk (which may and may not be distinct). Thus, m is a
product of k primes; hence, Ω (m) = k (by the definition of Ω).
Write n as a product of primes; i.e., write n in the form n = q1 q2 · · · q` for some
primes q1 , q2 , . . . , q` (which may and may not be distinct). Thus, n is a product
of ` primes; hence, Ω (n) = ` (by the definition of Ω).
Now, multiplying the equalities m = p1 p2 · · · pk and n = q1 q2 · · · q` , we obtain
mn = ( p1 p2 · · · pk ) (q1 q2 · · · q` ) = p1 p2 · · · pk q1 q2 · · · q` . Hence, mn is a product

26
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

of k + ` primes (since all of p1 , p2 , . . . , pk , q1 , q2 , . . . , q` are primes). Therefore,

Ω (mn) = k + ` (by the definition of Ω). Hence,

Ω (mn) = |{z} ` = Ω (m) + Ω (n) .

k + |{z}
=Ω(m) =Ω(n)

Now, the definition of λ yields λ (m) = (−1)Ω(m) and λ (n) = (−1)Ω(n) and
λ (mn) = (−1)Ω(mn) . Hence,

λ (mn) = (−1)Ω(mn) = (−1)Ω(m)+Ω(n) (since Ω (mn) = Ω (m) + Ω (n))

= (−1)Ω(m) (−1)Ω(n) = λ (m) λ (n) .
| {z } | {z }
=λ(m) =λ(n)

This proves Assertion 2.

Now, the function λ is totally multiplicative if and only if Assertions 1 and 2
hold (by the definition of “totally multiplicative”). Thus, the function λ is totally
multiplicative (since Assertions 1 and 2 hold). Consequently, λ is multiplicative
(since every totally multiplicative function is multiplicative (by Proposition 2.2.2
(i))). This proves Proposition 2.2.2 (g).
(k) We leave the proof of Proposition 2.2.2 (k) to the reader. (It is completely
straightforward using Proposition 1.2.10 and the fact that gcd (u, 1) = 1.)
We defer the proofs of parts (b) and (e) until later. (Actually, we shall also give
a second proof of part (a) later.) It thus remains to prove Proposition 2.2.2 (f).
(f) The definition of ω (1) shows that ω (1) is the number of all distinct primes
dividing 1. But the latter number is clearly 0 (since there are no primes dividing
1). Hence, ω (1) = 0.
The integer 1 is squarefree; hence, the definition of µ yields

µ (1) = (−1)ω (1) = (−1)0 (since ω (1) = 0)

= 1.

Hence, we only need to show that µ (mn) = µ (m) µ (n) for any two coprime
m ∈ N+ and n ∈ N+ . So let us show this.
Let m ∈ N+ and n ∈ N+ be coprime. We must prove the equality µ (mn) =
µ (m) µ (n). If mn is not squarefree, then this equality holds7 . Hence, we WLOG

7 Proof.Assume that mn is not squarefree. Thus, there is some integer g > 1 such that g2 | mn.
Consider this g.
There exists a prime p such that p | g (since g > 1). Consider such a p. The prime p
cannot divide both m and n (since m and n are coprime). Hence, either p - m or p - n (or
both). We WLOG assume that p - m (since otherwise, we can simply switch m with n). Thus,
m is coprime to p (since p is prime). In other words, p is coprime to m. In other words,
gcd ( p, m) = 1.
We have p | g, so that p2 | g2 | mn. Hence, p | p2 | mn. Hence, Proposition 1.2.8 (applied to

27
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

assume that mn is squarefree. Therefore, m is also squarefree (since any perfect

square dividing m would also divide mn). Similarly, n is squarefree.
Since m is squarefree, we have µ (m) = (−1)ω (m) (by the definition of µ). Since
n is squarefree, we have µ (n) = (−1)ω (n) (by the definition of µ). Since mn is
squarefree, we have µ (mn) = (−1)ω (mn) (by the definition of µ).
But ω (mn) is the number of all distinct primes dividing mn (by the definition
of ω). Thus,
ω (mn) = (the number of distinct primes dividing mn)
= (the number of distinct primes dividing m or dividing n) (4)
(since the primes dividing mn are precisely the primes dividing m or dividing
n).
Moreover, there is no overlap between the primes dividing m and the primes
dividing n (since m and n are coprime). Hence,
(the number of distinct primes dividing m or dividing n)
= (the number of distinct primes dividing m)
| {z }
=ω (m)
(since this is how ω (m) is defined)
+ (the number of distinct primes dividing n)
| {z }
=ω (n)
(since this is how ω (n) is defined)
= ω (m) + ω (n) .
Thus, (4) becomes
ω (mn) = (the number of distinct primes dividing m or dividing n)
= ω (m) + ω (n) . (5)
Now,

µ (mn) = (−1)ω (mn) = (−1)ω (m)+ω (n) (by (5))

= (−1)ω (m) (−1)ω (n) = µ (m) µ (n) .
| {z } | {z }
=µ(m) =µ(n)

p, m and n instead of x, y and z) shows that p | n. In other words, n = pn0 for some integer
n0 . Consider this n0 .
n = mpn0 = pmn0 . We can cancel p from this divisibility (since p 6= 0)
Now, pp = p2 | m |{z}
= pn0
and thus conclude p | mn0 . Hence, Proposition 1.2.8 (applied to p, m and n0 instead of x, y
and z) shows that p | n0 . Hence, pp | pn0 = n. In other words, p2 | n (since pp = p2 ). Hence,
n is not squarefree (since p2 is a perfect square other than 1). Therefore, µ (n) = 0 (by the
definition of µ). The definition of µ also shows that µ (mn) = 0 (since mn is not squarefree).
Now, comparing µ (mn) = 0 with µ (m) µ (n) = 0, we obtain µ (mn) = µ (m) µ (n), qed.
| {z }
=0

28
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

This completes the proof of µ (mn) = µ (m) µ (n). Thus, Proposition 2.2.2 (f)
holds.
Note that the function 0 : N+ → C is not multiplicative; in fact, it fails to
satisfy 0 (1) = 1.
The pointwise product of multiplicative functions is multiplicative, and the
same holds for totally multiplicative functions:

Proposition 2.2.3. Let g : N+ → C and h : N+ → C be two arithmetic

functions. Let f : N+ → C be the function defined by

f (n) = g (n) h (n) for every n ∈ N+ . (6)

(a) If the functions g and h are multiplicative, then the function f is multi-
plicative.
(b) If the functions g and h are totally multiplicative, then the function f is
totally multiplicative.

Proof of Proposition 2.2.3. (a) Assume that the functions g and h are multiplica-
tive. We have to prove that the function f is multiplicative.
The function g is multiplicative. In other words, it satisfies g (1) = 1, and
g (mn) = g (m) g (n) for any two coprime m ∈ N+ and n ∈ N+ . (7)
The function h is multiplicative. In other words, it satisfies h (1) = 1, and
h (mn) = h (m) h (n) for any two coprime m ∈ N+ and n ∈ N+ . (8)
Now, we want to prove that f is multiplicative. In order to do so, we shall
prove the following two assertions:
Assertion 1: We have f (1) = 1.
Assertion 2: We have f (mn) = f (m) f (n) for any two coprime m ∈
N+ and n ∈ N+ .
Proof of Assertion 1: Applying (6) to n = 1, we obtain f (1) = g (1) h (1) = 1.
| {z } | {z }
=1 =1
This proves Assertion 1.
Proof of Assertion 2: Let m ∈ N+ and n ∈ N+ be coprime. Then, (6) (applied
to m instead of n) yields f (m) = g (m) h (m). Also, (6) shows that f (n) =
g (n) h (n). But (6) (applied to mn instead of n) yields
f (mn) = g (mn) h (mn) = ( g (m) g (n)) (h (m) h (n))
| {z } | {z }
= g(m) g(n) =h(m)h(n)
(by (7)) (by (8))

= ( g (m) h (m)) ( g (n) h (n)) = f (m) f (n) .

| {z }| {z }
= f (m) = f (n)

29
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

This proves Assertion 2.

Now we know that Assertions 1 and 2 hold. In other words, the function f
is multiplicative (by the definition of “multiplicative”). This proves Proposition
2.2.3 (a).
(b) The proof of Proposition 2.2.3 (b) is completely analogous to our above
proof of Proposition 2.2.3 (a). (More precisely, we have to replace every “mul-
tiplicative” by “totally multiplicative” in our proof of Proposition 2.2.3 (a), and
remove the word “coprime” everywhere it appears; this results in a proof of
Proposition 2.2.3 (b).)
The reader might have already noticed that there is a relation between the two
arithmetic functions ω and Ω and the squarefreeness of a number. Let us state
this explicitly:

Proposition 2.2.4. Let n ∈ N+ .

(a) We have Ω (n) ≥ ω (n).
(b) We have Ω (n) = ω (n) if and only if n is squarefree.
(c) We have µ (n) = 0Ω(n)−ω (n) · λ (n). (In particular, the expression
“0Ω(n)−ω (n) ” is well-defined, since Ω (n) − ω (n) ≥ 0.)

Proof of Proposition 2.2.4. Write n as a product of primes; i.e., write n in the form
n = p1 p2 · · · pk for some primes p1 , p2 , . . . , pk (which may and may not be dis-
tinct). Then, the primes dividing n are precisely p1 , p2 , . . . , pk 8 . In other

words, (the set of all primes dividing n) = { p1 , p2 , . . . , pk }. Now, the definition

of ω yields

ω (n) = (the number of all distinct primes dividing n)

= (the set of all primes dividing n)

| {z }
={ p1 ,p2 ,...,pk }
= |{ p1 , p2 , . . . , pk }| (9)
≤ k.

But n can be written as a product of k primes (because n = p1 p2 · · · pk , and

the factors p1 , p2 , . . . , pk in this product are primes). Hence, the definition of Ω
yields Ω (n) = k. Now, ω (n) ≤ k = Ω (n). This proves Proposition 2.2.4 (a).
(b) We shall prove the following two statements:

Statement 1: If Ω (n) = ω (n), then n is squarefree.

Statement 2: If n is squarefree, then Ω (n) = ω (n).

8 Thisis because if a prime q divides n, then q | n = p1 p2 · · · pk , and therefore q | pi for some

i ∈ {1, 2, . . . , k} (since q is prime); but this entails q = pi for this i.

30
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

[Proof of Statement 1: Assume that Ω (n) = ω (n). We must prove that n is

squarefree.
Assume the contrary. Thus, n is not squarefree. Hence, there exists a perfect
square other than 1 that divides n (by the definition of “squarefree”). In other
words, there exists some integer h > 1 such that h2 | n. Consider this h.
The integer h is larger than 1. Thus, there exists a prime q such that q | h.
Consider this q. From q | h, we obtain q2 | h2 | n.
From (9), we have |{ p1 , p2 , . . . , pk }| = ω (n) = Ω (n) = k. In other words, the
set { p1 , p2 , . . . , pk } has exactly k elements. Hence, the k elements p1 , p2 , . . . , pk
must be distinct (since otherwise, the set { p1 , p2 , . . . , pk } would have fewer than
k elements).
But q | q2 | n = p1 p2 · · · pk . Since q is a prime, this entails that q divides at least
one of the k integers p1 , p2 , . . . , pk (because of Remark 1.2.15, applied to q instead
of p). In other words, at least one i ∈ {1, 2, . . . , k } satisfies q | pi . Consider this i.
Thus, q is a prime divisor of pi (since q is a prime and satisfies q | pi ). But the
only prime divisor of pi is pi itself (since pi is a prime). Thus, q must be pi itself.
In other words, q = pi .
Now,
qq = q2 | n = p1 p2 · · · pk = pi · ( p1 p2 · · · pi−1 pi+1 pi+2 · · · pk )
|{z}
=q
= q · ( p 1 p 2 · · · p i −1 p i +1 p i +2 · · · p k ) .
We can cancel q from this divisibility (since q 6= 0), and thus obtain
q | p 1 p 2 · · · p i −1 p i +1 p i +2 · · · p k .
Since q is a prime, this entails that q divides at least one of the k − 1 integers
p1 , p2 , . . . , pi−1 , pi+1 , pi+2 , . . . , pk (because of Remark 1.2.15, applied to q instead
of p). In other words, at least one j ∈ {1, 2, . . . , k} \ {i } satisfies q | p j . Consider
this j.
We have j 6= i (since j ∈ {1, 2, . . . , k} \ {i }) and thus p j 6= pi (since the k primes
p1 , p2 , . . . , pk are distinct).
But q is a prime divisor of p j (since q is a prime and satisfies q | p j ). But the
only prime divisor of p j is p j itself (since p j is a prime). Thus, q must be p j
itself. In other words, q = p j . Hence, q = p j 6= pi = q. This is absurd. This
contradiction shows that our assumption was false; hence, n is squarefree. This
proves Statement 1.]
[Proof of Statement 2: Assume that n is squarefree. We must prove that Ω (n) =
ω ( n ).
The k primes p1 , p2 , . . . , pk are distinct9 . Hence, |{ p1 , p2 , . . . , pk }| = k. Now, (9)
becomes ω (n) = |{ p1 , p2 , . . . , pk }| = k = Ω (n). In other words, Ω (n) = ω (n).
This proves Statement 2.]
9 Proof. Assume the contrary. Thus, two of these k primes are equal. In other words, there
exist two distinct elements i and j of {1, 2, . . . , k } such that pi = p j . Consider these i and
j. Clearly, pi and p j are two distinct factors in the product p1 p2 · · · pk (since i 6= j). Hence,

31
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Combining Statement 1 with Statement 2, we conclude that we have Ω (n) =

ω (n) if and only if n is squarefree. This proves Proposition 2.2.4 (b).
(c) Proposition 2.2.4 (a) yields Ω (n) ≥ ω (n). Hence, Ω (n) − ω (n) ≥ 0. Thus,
the expression “0Ω(n)−ω (n) ” is well-defined. It remains to prove that µ (n) =
0Ω(n)−ω (n) · λ (n).
We are in one of the following two cases:
Case 1: The positive integer n is squarefree.
Case 2: The positive integer n is not squarefree.
Let us first consider Case 1. In this case, the positive integer n is squarefree.
Hence, Proposition 2.2.4 (b) yields that Ω (n) = ω (n). Hence, Ω (n) − ω (n) = 0
and therefore 0Ω(n)−ω (n) = 00 = 1. But the definition of µ yields
(
(−1)ω (n) , if n is squarefree;
µ (n) = = (−1)ω (n) (since n is squarefree) .
0, otherwise

Comparing this with

Ω(n)
0| Ω(n{z
)−ω (n)
} ·λ (n) = λ (n) = (−1) (by the definition of λ)
=1
= (−1)ω (n) (since Ω (n) = ω (n)) ,

we obtain µ (n) = 0Ω(n)−ω (n) · λ (n). Thus, µ (n) = 0Ω(n)−ω (n) · λ (n) is proven in
Case 1.
Let us next consider Case 2. In this case, the positive integer n is not square-
free.
If we had Ω (n) = ω (n), then n would be squarefree (by Proposition 2.2.4
(b)), which would contradict the fact that n is not squarefree. Hence, we cannot
have Ω (n) = ω (n). Thus, we have Ω (n) 6= ω (n).
But Proposition 2.2.4 (a) yields Ω (n) ≥ ω (n). Combining this with Ω (n) 6=
ω (n), we find Ω (n) > ω (n). Hence, Ω (n) − ω (n) > 0, so that 0Ω(n)−ω (n) = 0.
Now, the definition of µ yields
(
(−1)ω (n) , if n is squarefree;
µ (n) = =0 (since n is not squarefree) .
0, otherwise

pi p j | p1 p2 · · · pk = n.
We have pi > 1 (since pi is a prime). Thus, p2i is a perfect square other than 1. Moreover,

p2i = pi pi = pi p j | n.
|{z}
= pj

Thus, a perfect square other than 1 (namely, p2i ) divides n. In other words, n is not squarefree
(by the definition of “squarefree”). This contradicts our assumption that n is squarefree. This
contradiction shows that our assumption was false. Qed.

32
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Comparing this with

0| Ω(n{z
)−ω (n)
} ·λ (n) = 0,
=0

we obtain µ (n) = 0Ω(n)−ω (n) · λ (n). Thus, µ (n) = 0Ω(n)−ω (n) · λ (n) is proven in
Case 2.
We have now proven µ (n) = 0Ω(n)−ω (n) · λ (n) in both Cases 1 and 2. Hence,
µ (n) = 0Ω(n)−ω (n) · λ (n) always holds. This completes the proof of Proposition
2.2.4 (c).

2.3. The Dirichlet convolution

Let us now define a way to produce a new arithmetic function from two given
ones: the Dirichlet convolution.

Definition 2.3.1. Let f : N+ → C and g : N+ → C be two arithmetic func-

tions. We define a new arithmetic function f ? g : N+ → C by
n
( f ? g) (n) = ∑ f (d) g for every n ∈ N+ .
d|n
d

This new function f ? g is called the Dirichlet convolution of f and g.

Here is a more symmetric way to rewrite the definition of f ? g:

Remark 2.3.2. Let f : N+ → C and g : N+ → C be two arithmetic functions.

Let n ∈ N+ . Then,

( f ? g) (n) = ∑ f (d) g (e) .

d ∈N+ ; e ∈N+ ;
de=n

Proof of Remark 2.3.2. Let us first show two simple claims:

Claim 1: For each d ∈ N+ satisfying d | n, we have

n
∑ g ( e ) = g
d
.
e ∈N+ ;
de=n

Claim 2: For each d ∈ N+ satisfying d - n, we have

∑ g (e) = 0.
e ∈N+ ;
de=n

33
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

n
[Proof of Claim 1: Let d ∈ N+ be such that d | n. Then, ∈ Z (since d | n).
d
n
Thus, ∈ N+ (since n ∈ N+ and d ∈ N+ ). Thus, there is exactly one e ∈ N+
d
n
satisfying de = n, namely e = . Hence, the sum ∑ g (e) contains exactly one
d e ∈N+ ;
de=n
n n
addend, namely the addend for e = . Thus, ∑ g (e) = g . Claim 1 is
d e ∈N+ ; d
de=n
proven.]
[Proof of Claim 2: Let d ∈ N+ be such that d - n. Then, there exists no e ∈ N+
satisfying de = n (since d - n). Therefore, the sum ∑ g (e) is empty and thus
e ∈N+ ;
de=n
equals 0. This proves Claim 2.]
Now,

∑ f (d) g (e)
d ∈N+ ; e ∈N+ ;
de=n
| {z }
= ∑ ∑
d ∈N+ e ∈N+ ;
de=n
= ∑ ∑ f (d) g (e) = ∑ f (d) ∑ g (e)
d ∈N+ e ∈N+ ; d ∈N+ e ∈N+ ;
de=n de=n
= ∑ f (d) ∑ g (e) + ∑ f (d) ∑ g (e)
d ∈N+ ; e ∈N+ ; d ∈N+ ; e ∈N+ ;
d|n
|de=n{z } d-n
|de=n
{z }
n
 
=0
=g (by Claim 2)
d
(by Claim 1)
since each d ∈ N+ satisfies either d | n or d - n
 

(but not both at the same time)

n n n
= ∑ f (d) g
d
+ ∑ f (d) 0 = ∑ f (d) g
d
= ∑ f (d) g
d
d ∈N+ ; d ∈N ; + d ∈N ; d|n+
d|n d-n d|n
| {z }
=0
= ( f ? g) (n) (because this is how ( f ? g) (n) was defined) .

This proves Remark 2.3.2.

Remark 2.3.3. Here is a little digression which might make the Dirichlet con-
volution f ? g appear less mysterious (but might also confuse you). I claim that
Dirichlet convolution of arithmetic functions is “like multiplication of power
series, but with (some) additions replaced by multiplications”. Here is what I
mean by this:

34
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

A power series (say, with complex coefficients) is defined as a sequence

( a0 , a1 , a2 , . . .) of complex numbers. This sequence is usually written in the
form ∑ ai X i . (For us here, “power series” means “formal power series in one
i ∈N
indeterminate X with complex coefficients”; we do not care about any ques-
tions of convergence.) The product of two power series ∑ ai X i and ∑ bi X i
i ∈N i ∈N
is defined by ! !
∑ ai X i ∑ bi X i = ∑ ci X i ,
i ∈N i ∈N i ∈N
where
n
cn = ∑ a m bn − m = ∑ a d be . (10)
m =0 d∈N; e∈N;
d+e=n

To every arithmetic function f : N+ → C, we can assign a power series fb

with constant term 0, defined by fb = ∑ f (i ) X i . This assignment is a 1-to-1
i ∈N+
correspondence between the arithmetic functions and the power series (in one
indeterminate) with constant term 0. In other words, every power series with
constant term 0 can be written as fb for a unique arithmetic function f . Thus,
we can define a “Dirichlet convolution“ on the set of all power series with
constant term 0, by setting fb ? gb = [
f ? g for every two arithmetic functions f
and g. Explicitly, this Dirichlet convolution of power series is given by
! !
∑ ai X i ? ∑ bi X i = ∑ di X i ,
i ∈N+ i ∈N+ i ∈N+

where
dn = ∑ ad bn/d = ∑ a d be . (11)
d|n d ∈N+ ; e ∈N+ ;
de=n
The similarities between the equations (10) and (11) should be palpable.
Roughly speaking, (11) is a “multiplicative” variant of (10): Whereas the sum
∑ ad be in (10) runs over all decompositions of n into a sum of two non-
d∈N; e∈N;
d+e=n
negative integers d and e, the analogous sum ∑ ad be in (11) runs over
d ∈N+ ; e ∈N+ ;
de=n
all decompositions of n into a product of two positive integers d and e. (Yes,
the multiplicative analogue of nonnegative integers in this context are positive
integers.) So, roughly speaking, Dirichlet convolution is like multiplication of
power series, except that two monomials X m and X n are taken to X mn and not
to X m+n .
This analogy has a consequence: It suggests that Dirichlet convolution
should be associative and commutative, and that this should be provable in

35
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

the same way as one proves the associativity and the commutativity of the
multiplication of power series. And, indeed, this is the case: see Theorem
2.3.4 below.
See also [NiZuMo91, §8.2] for the notion of Dirichlet series, which are “for-
a
mal expressions” of the form ∑ si for an “indeterminate” exponent s. If
i ∈N+ i
ai
we replace the term s by ai X i , then these Dirichlet series turn into standard
i
formal power series with constant term 0, but their product turns into the
Dirichlet convolution of power series.

Theorem 2.3.4. (a) We have ε ? f = f ? ε = f for every arithmetic function f .

(b) We have f ? ( g ? h) = ( f ? g) ? h for every three arithmetic functions f ,
g and h.
(c) We have f ? g = g ? f for every two arithmetic functions f and g.

Remark 2.3.5. If you know the notion of a monoid, then you will be able
to restate Theorem 2.3.4 as follows: The set of all arithmetic functions is a
commutative monoid under the operation ? with neutral element ε.
Actually, we can also define an addition operation on arithmetic functions
(namely, pointwise addition: ( f + g) (n) = f (n) + g (n)). The addition oper-
ation + and the Dirichlet convolution ? turn the set of arithmetic functions
into a commutative ring.

Proof of Theorem 2.3.4. (c) Let f and g be two arithmetic functions. Let n ∈ N+ .
Remark 2.3.2 (applied to g and f instead of f and g) yields

( g ? f ) (n) = ∑ g (d) f (e) . (12)

d ∈N+ ; e ∈N+ ;
de=n

Remark 2.3.2 yields

( f ? g) (n) = ∑ f (d) g (e) = ∑ f (e) g (d)

d ∈N+ ; e ∈N+ ; e ∈N+ ; d ∈N+ ;
| {z }
de=n ed=n = g(d) f (e)
| {z }
= ∑
d ∈N+ ; e ∈N+ ;
ed=n
= ∑
d ∈N+ ; e ∈N+ ;
de=n
 
here, we have renamed the summation
indices d and e as e and d
= ∑ g (d) f (e) .
d ∈N+ ; e ∈N+ ;
de=n

Comparing this with (12), we obtain ( f ? g) (n) = ( g ? f ) (n).

36
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Now, forget that we fixed n. We thus have shown that ( f ? g) (n) = ( g ? f ) (n)
for each n ∈ N+ . In other words, f ? g = g ? f . This proves Theorem 2.3.4 (c).
(a) Let f be an arithmetic function. Every n ∈ N+ satisfies
n
(ε ? f ) (n) = ∑ ε (d) f (by the definition of ε ? f )
d|n 
|{z} d
1, if d = 1;
=
0, if d 6 = 1
(by the definition of ε)
(
1, if d = 1;  n 
=∑ f
d|n 0, if d 6= 1 d
( (
1, if 1 = 1;  n  1, if d = 1;  n 
= f +∑ f
0, if 1 6= 1 | {z1 } d|n; 0, if d 6= 1 d
| {z } = f (n) d 6 =1 | {z }
=1 =0
(since d6=1)
 
here, we have split off the addend for d = 1 from the sum
(since 1 is a positive divisor of n)
n
= f (n) + ∑ 0 f = f (n) .
d|n;
d
d 6 =1
| {z }
=0

In other words, ε ? f = f . But Theorem 2.3.4 (c) (applied to g = ε) yields

f ? ε = ε ? f . Thus, f ? ε = ε ? f = f . This proves Theorem 2.3.4 (a).
(b) Let us first make a general observation: If F and G are two arithmetic
functions, and if N ∈ N+ , then

( F ? G) ( N ) = ∑ F ( D ) G ( E) . (13)
D ∈N+ ; E ∈N+ ;
DE= N

(This is simply Remark 2.3.2, with the letters f , g, n, d, e renamed as F, G, N, D, E.

We are playing this renaming game in order to avoid collisions between nota-
tions.)

37
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Now, let f , g and h be three arithmetic functions. Let n ∈ N+ . We have

(( f ? g) ? h) (n)
= ∑ ( f ? g) ( D ) h ( E)
D ∈N+ ; E ∈N+ ;
DE=n
(by (13), applied to F = f ? g, G = h and N = n)
= ∑ ( f ? g) (d) h (e)
d ∈N+ ; e ∈N+ ;
| {z }
de=n = ∑ f ( D ) g( E)
D ∈N+ ; E ∈N+ ;
DE=d
(by (13), applied to F = f , G = g and N =d)

(here, we have renamed the summation indices D and E as d and e)

 

= ∑ ∑ f ( D ) g ( E) h (e)
 

d ∈N+ ; e ∈N+ ; D ∈N+ ; E ∈N+ ;
de=n DE=d
= ∑ ∑ f ( D ) g ( E) h (e)
d ∈N+ ; e ∈N+ ; D ∈N+ ; E ∈N+ ;
de=n DE=d
| {z }
= ∑ ∑ ∑
d ∈N+ D ∈N+ ; E ∈N+ ; e ∈N+ ;
DE=d de=n
(here, we are interchanging the order of summation)

= ∑ ∑ ∑ f ( D ) g ( E) h (e)
d ∈N+ D ∈N+ ; E ∈N+ ; e ∈N+ ;
DE=d de=n
| {z }
= ∑
e ∈N+ ;
DEe=n
(since d= DE)
= ∑ ∑ ∑ f ( D ) g ( E) h (e) = ∑ ∑ f ( D ) g ( E) h (e)
d ∈N+ D ∈N+ ; E ∈N+ ; e ∈N+ ; D ∈N+ ; E ∈N+ e ∈N+ ;
DE=d DEe=n DEe=n
| {z } | {z }
= ∑ = ∑
D ∈N+ ; E ∈N+ D ∈N+ ; E ∈N+ ; e ∈N+ ;
DEe=n
= ∑ f ( D ) g ( E) h (e)
D ∈N+ ; E ∈N+ ; e ∈N+ ;
DEe=n
= ∑ f (c) g (d) h (e) (14)
c ∈N+ ; d ∈N+ ; e ∈N+ ;
cde=n
(here, we have renamed the summation indices D and E as c and d) .

38
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

On the other hand,

( f ? ( g ? h)) (n)
= ∑ f ( D ) ( g ? h) ( E)
D ∈N+ ; E ∈N+ ;
DE=n
(by (13), applied to F = f , G = g ? h and N = n)
= ∑ f (c) ( g ? h) (d)
c ∈N+ ; d ∈N+ ;
| {z }
cd=n = ∑ g( D )h( E)
D ∈N+ ; E ∈N+ ;
DE=d
(by (13), applied to F = g, G =h and N =d)

(here, we have renamed the summation indices D and E as c and d)

 

= ∑ f (c)  ∑ g ( D ) h ( E)
 
c ∈N+ ; d ∈N+ ; D ∈N+ ; E ∈N+ ;
cd=n DE=d
= ∑ ∑ f (c) g ( D ) h ( E)
c ∈N+ ; d ∈N+ ; D ∈N+ ; E ∈N+ ;
cd=n DE=d
| {z }
= ∑ ∑ ∑
d ∈N+ D ∈N+ ; E ∈N+ ; c ∈N+ ;
DE=d cd=n
(here, we are interchanging the order of summation)

= ∑ ∑ ∑ f (c) g ( D ) h ( E)
d ∈N+ D ∈N+ ; E ∈N+ ; c ∈N+ ;
DE=d cd=n
| {z }
= ∑
c ∈N+ ;
cDE=n
(since d= DE)
= ∑ ∑ ∑ f (c) g ( D ) h ( E) = ∑ ∑ f (c) g ( D ) h ( E)
d ∈N+ D ∈N+ ; E ∈N+ ; c ∈N+ ; D ∈N+ ; E ∈N+ c ∈N+ ;
DE=d cDE=n cDE=n
| {z } | {z }
= ∑ = ∑
D ∈N+ ; E ∈N+ c ∈N+ ; D ∈N+ ; E ∈N+ ;
cDE=n
= ∑ f (c) g ( D ) h ( E) = ∑ f (c) g (d) h (e)
c ∈N+ ; D ∈N+ ; E ∈N+ ; c ∈N+ ; d ∈N+ ; e ∈N+ ;
cDE=n cde=n
(here, we have renamed the summation indices D and E as d and e) .

Comparing this with (14), we obtain ( f ? ( g ? h)) (n) = (( f ? g) ? h) (n).

Now, forget that we fixed n. We thus have proven that ( f ? ( g ? h)) (n) =
(( f ? g) ? h) (n) for every n ∈ N+ . In other words, f ? ( g ? h) = ( f ? g) ? h. This
proves Theorem 2.3.4 (b).

39
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

2.4. Examples of Dirichlet convolutions

Let us see what Dirichlet convolution does to the arithmetic functions we know.
We start with some simple observations:

Proposition 2.4.1. We have 1 ? 1 = d. (See Definition 2.1.2 for the definitions

of 1 and d.)

Proof of Proposition 2.4.1. For every n ∈ N+ , we have

n
(1 ? 1) ( n ) = ∑ 1 (d) 1 (by the definition of 1 ? 1)
d|n
|{z} | {zd }
=1
(by the definition of 1) =1
(by the definition of 1)

= ∑ 1 = (the number of positive divisors of n) ·1

d|n
| {z }
=d( n )
(since this is how d(n) was defined)
= d (n) · 1 = d (n) .

In other words, 1 ? 1 = d. This proves Proposition 2.4.1.

Proposition 2.4.2. (a) We have ι ? 1 = σ.

(b) Let k ∈ Z. Let ι k : N+ → C be the function sending each n ∈ N+ to nk .
Then, ι k ? 1 = σk .

We leave the proof of Proposition 2.4.2 to the reader.

Before we go on, let us show an auxiliary fact:

Lemma 2.4.3. Let n ∈ N+ . Let D (n) be the set of all positive divisors of n.
Then, the map
D (n) → D (n) , d 7→ n/d
is well-defined and bijective.

Proof of Lemma 2.4.3. For every d ∈ D (n), we have n/d ∈ D (n) 10 . Thus, we
can define a map ρ : D (n) → D (n) by

ρ (d) = n/d for every d ∈ D (n) .

Consider this map ρ. This map ρ is the map

D (n) → D (n) , d 7→ n/d (15)

10 Proof. Let d ∈ D (n). Thus, d is a positive divisor of n (since D (n) is the set of all positive
divisors of n). Hence, d is a positive integer and satisfies d | n. Now, n/d is an integer (since
d | n) and is positive (since n and d are positive). Hence, n/d is a positive integer. Thus, n/d
is a positive divisor of n (since n/d | n). In other words, n/d ∈ D (n) (since D (n) is the set
of all positive divisors of n). Qed.

40
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

(because ρ (d) = n/d for every d ∈ D (n)). Thus, the map (15) is well-defined.
We have
 

(ρ ◦ ρ) (d) = ρ ρ (d) = ρ (n/d) = n/ (n/d) (by the definition of ρ)

 
| {z }
=n/d
= d = id (d)

for every d ∈ D (n). In other words, ρ ◦ ρ = id. Hence, the maps ρ and ρ
are mutually inverse. In particular, the map ρ is invertible, i.e., bijective. In
other words, the map (15) is bijective (since the map ρ is the map (15)). Thus,
we have shown that the map (15) is well-defined and bijective. Lemma 2.4.3 is
proven.
Here is a more interesting result:

Proposition 2.4.4. We have φ ? 1 = ι.

Before we prove this, let us restate it in a more elementary fashion:

Proposition 2.4.5. We have

∑ φ (d) = n for every n ∈ N+ .

d|n

Proposition 2.4.5 is [NiZuMo91, Theorem 4.6]. The proof we are going to give
for it here is actually the second proof given for it in [NiZuMo91]:
Proof of Proposition 2.4.5. Fix n ∈ N+ . Let me first show that
n
∑ 1 = φ
d
(16)
k∈{1,2,...,n};
gcd(k,n)=d

for every positive divisor d of n.

Proof of (16): Let d be a positive divisor of n. Define a set K by

K = {k ∈ {1, 2, . . . , n} | gcd (k, n) = d} .

Thus, ∑ 1 = ∑ 1 = | K | · 1 = | K |.
k∈{1,2,...,n}; k∈K
gcd(k,n)=d
On the other hand, define a set F by
n n no no
F = k ∈ 1, 2, . . . , | k is coprime to .
d d

41
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

nno n
Then, | F | is the number of all k ∈ 1, 2, . . . , coprime to ; this number is
n n d d n
φ (since this is how φ is defined). In other words, | F | = φ .
d d d
u
But every u ∈ K satisfies ∈ F 11 . Thus, we can define a map
d
u
α : K → F, u 7→ .
d
On the other hand, every v ∈ F satisfies dv ∈ K 12 . Thus, we can define a
map
β : F → K, v 7→ dv.
The two maps α and β that we have now defined are mutually inverse (since one
of them divides its input by d, whereas the other multiplies it by d). Hence, α is
abijection. Thus, there is a bijection K → F (namely, α). Hence, | F | = |K |. Now,
n
φ = | F | = |K | = ∑ 1, and therefore (16) is proven.
d k∈{1,2,...,n};
gcd(k,n)=d
Now, let D (n) be the set of all positive divisors of n. Then, the summation
sign ∑ means the same thing as ∑ (namely, a summation over all positive
d∈D(n) d|n
divisors d of n).
But Lemma 2.4.3 shows that the map
D (n) → D (n) , d 7→ n/d
11 Proof. Let u ∈ K. Thus, u is an element of {1, 2, . . . , n} and satisfies gcd (u, n) = d (by the
u u
definition of K). Now, d = gcd (u, n) | u, so that is an integer. This integer must belong
d d
n no u n
to 1, 2, . . . , (since u belongs to {1, 2, . . . , n}). But Proposition 1.2.9 (applied to d, and
d   d d
u n  u n
instead of g, a and b) yields d gcd , = gcd  d· ,d·  = gcd (u, n) = d. Cancelling
d d 
|{z}d |{z}
d
u n =u =n
u
d from this equality, we obtain gcd , = 1 (since d 6= 0). In other words, is coprime to
d d d
n u n no n
. Thus, we have shown that is an element of 1, 2, . . . , and is coprime to . In other
d d d d
u
words, ∈ F (by the definition of F), qed.
d
12 Proof. Let v ∈ F. Thus, v is an element of 1, 2, . . . , n and is coprime to n (by the definition
n o
 n d d
n
of F). Now, gcd v, = 1 (since v is coprime to ). But Proposition 1.2.9 (applied to d, v
d d  
n  n  n
and instead of g, a and b) shows that d gcd v, = gcd  dv, d ·  = gcd (dv, n), so that
d d  d
|{z}
 n =n
n no
gcd (dv, n) = d gcd v, = d. Also, from v ∈ 1, 2, . . . , , we obtain dv ∈ {1, 2, . . . , n}.
| {z d } d
=1
Hence, we have shown that dv is an element of {1, 2, . . . , n} and satisfies gcd (dv, n) = d. In
other words, dv ∈ K (by the definition of K), qed.

42
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

is well-defined and bijective. Thus, this map is a bijection. Hence, we can sub-
n
stitute for d in the sum ∑ φ (d). We thus obtain
d d∈D(n)
n
∑ φ (d) = ∑ φ = ∑ ∑ 1
d∈D(n) d∈D(n) | {zd } d|n k∈{1,2,...,n};
| {z } = ∑ 1 gcd(k,n)=d
=∑ k∈{1,2,...,n}; | {z }
d|n gcd(k,n)=d = ∑
k ∈{1,2,...,n}
(by (16))
(because for every k∈{1,2,...,n},
the number gcd(k,n) is a positive divisor of n)

= ∑ 1 = n.
k∈{1,2,...,n}

Therefore, n = ∑ φ (d) = ∑ φ (d). This proves Proposition 2.4.5.

d∈D(n) d|n
| {z }
=∑
d|n

Proof of Proposition 2.4.4. For every n ∈ N+ , we have

n
( φ ? 1) ( n ) = ∑ φ ( d ) 1 (by the definition of φ ? 1)
d|n | {zd }
=1
(by the definition of 1)

= ∑ φ (d) = n (by Proposition 2.4.5)

d|n
= ι (n) (since ι (n) is defined to be n) .

In other words, φ ? 1 = ι. This proves Proposition 2.4.4.

Here is another important fact about Dirichlet convolution:

Proposition 2.4.6. We have µ ? 1 = ε.

Again, we shall first restate it in concrete language before proving it:

Proposition 2.4.7. We have

∑ µ (d) = ε (n) for every n ∈ N+ .

d|n

Proposition 2.4.7 is [NiZuMo91, Theorem 4.7], and the book gives two proofs
for it. Let us sketch a third:13

13 See [Grinbe15, proof of Proposition 2.6] for a detailed version of this proof.

43
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Proof of Proposition 2.4.7 (sketched). Fix n ∈ N+ . We must prove the identity

∑ µ (d) = ε (n) . (17)

d|n

First of all, we recall that µ (1) = 1. (This has already been proven in our
proof of Proposition 2.2.2 (f).) Hence, ∑ µ (d) = µ (1) = 1 = ε (1) (because ε (1)
d |1
is defined to be 1). In other words, (17) is proven for the case when n = 1. Thus,
we WLOG assume that n 6= 1 from now on. Hence, ε (n) = 0 (by the definition
of ε). Also, n has at least one prime divisor (since n 6= 1). Pick any prime divisor
q of n.
Let D be the set of all squarefree positive divisors d of n satisfying q - d.
Let E be the set of all squarefree positive divisors d of n satisfying q | d.
The map
D → E, d 7→ qd
is well-defined and a bijection14 . Moreover, every d ∈ D satisfies

µ (qd) = −µ (d) (18)

15 .

14 Check this! (Or see [Grinbe15, proof of Proposition 2.6] for the proof.)
15 Proof of (18): Let d ∈ D. Thus, d is a squarefree positive divisor d of n satisfying q - d (by the
definition of D). From q - d, we conclude that q is coprime to d (since q is prime). Hence,
µ (qd) = µ (q) µ (d) (since the function µ is multiplicative).
But q is a prime; thus, q is squarefree. Hence, the definition of µ yields µ (q) = (−1)ω (q) =
(−1)1 (since ω (q) = 1 (again since q is a prime)). Thus, µ (qd) = µ (q) µ (d) = −µ (d),
| {z }
=(−1)1 =−1
qed.

44
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Now,

∑ µ (d) = ∑ µ (d) + ∑ µ (d)

d|n d|n; d|n;
| {z }
d is squarefree
=0
d is not squarefree (by the definition of µ,
since d is not squarefree)

= ∑ µ (d) + ∑ 0= ∑ µ (d)
d|n; d|n; d|n;
d is squarefree d is not squarefree d is squarefree
| {z }
=0
= ∑ µ (d) + ∑ µ (d)
d|n; d|n;
d is squarefree; d is squarefree;
q|d q-d
| {z } | {z }
=∑ = ∑
d∈ E d∈ D
(by the definition of E) (by the definition of D)

= ∑ µ (d) + ∑ µ (d) = ∑ µ (qd) + ∑ µ (d)

d∈ E d∈ D d∈ D
| {z } d∈ D
=−µ(d)
(by (18))
 
here, we have substituted qd for d in the first sum, since
the map D → E, d 7→ qd is a bijection
= ∑ (−µ (d)) + ∑ µ (d) = − ∑ µ (d) + ∑ µ (d) = 0
d∈ D d∈ D d∈ D d∈ D
= ε (n) (since ε (n) = 0) .

This proves (17). Thus, Proposition 2.4.7 is proven.

Our proof of Proposition 2.4.7 used a standard technique: In order to prove
that a sum is 0, we split the sum into two smaller sums, which cancelled each
other out term by term (i.e., every term of one cancelled a term of the other).
This kind of proof is widespread in combinatorics and other disciplines.
Proof of Proposition 2.4.6. For every n ∈ N+ , we have
n
( µ ? 1) ( n ) = ∑ µ ( d ) 1 (by the definition of µ ? 1)
d|n | d
{z }
=1
(by the definition of 1)

= ∑ µ (d) = ε (n) (by Proposition 2.4.7) .

d|n

In other words, µ ? 1 = ε. This proves Proposition 2.4.6.

The Dirichlet convolutions we have so far computed allow us to compute
other Dirichlet convolutions without actually working with sums, but simply by
applying Theorem 2.3.4. Here is one result we can obtain in this way:

45
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Proposition 2.4.8. We have µ ? ι = φ.

In concrete language, this says the following:

Proposition 2.4.9. We have

n
∑ µ (d) d = φ (n) for every n ∈ N+ .
d|n

Instead of proving the concrete version combinatorially and then deriving the
Dirichlet convolution from it, we will go the opposite way this time:
Proof of Proposition 2.4.8. Proposition 2.4.4 yields ι = φ ? 1 = 1 ? φ (by Theorem
2.3.4 (c), applied to f = φ and g = 1). Now,

µ ? |{z}
ι = µ ? (1 ? φ ) = ( µ ? 1) ?φ
| {z }
=1? φ =ε
(by Proposition 2.4.6)

(by Theorem 2.3.4 (b), applied to f = µ, g = 1 and h = φ)

= ε?φ = φ (by Theorem 2.3.4 (a), applied to f = φ) .

Thus, Proposition 2.4.8 is proven.

Proof of Proposition 2.4.9. Proposition 2.4.8 yields φ = µ ? ι. Thus, every n ∈ N+
satisfies
n
φ (n) = (µ ? ι) (n) = ∑ µ (d) ι (by the definition of µ ? ι)
|{z} d | n
d}
=µ?ι
| {z
n
=
d
(by the definition of ι)
n
= ∑ µ (d) .
d|n
d

This proves Proposition 2.4.9.

Proposition 2.4.9 is [NiZuMo91, (4.1)].

2.5. Möbius inversion

A particularly useful consequence of the “calculus of Dirichlet convolution” we
have established is the so-called Möbius inversion formula:

46
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Theorem 2.5.1 (Möbius inversion formula). Let f : N+ → C and F : N+ → C

be two arithmetic functions. Then, we have the following logical equivalence:
 
 F (n) = ∑ f (d) for all n ∈ N+ 
d|n
 
n
⇐⇒  f (n) = ∑ µ (d) F for all n ∈ N+  .
d|n
d

Theorem 2.5.1 is [NiZuMo91, Theorems 4.8 and 4.9]. It is merely the most well-
known of the many “Möbius inversion formulas” that appear in various parts
of mathematics; see [BenGol75] or [Rota64] or [Stanle11, §3.7] for introductions
into the more general theory of Möbius functions (of partially ordered sets).
We shall prove Theorem 2.5.1 by rewriting it in the following equivalent form:

Proposition 2.5.2. Let f : N+ → C and F : N+ → C be two arithmetic

functions. Then, we have the following logical equivalence:

( F = f ? 1) ⇐⇒ ( f = µ ? F ) .

Proof of Proposition 2.5.2. We have f ? 1 = 1 ? f (by Theorem 2.3.4 (c), applied to

g = 1). Also, µ ? 1 = 1 ? µ (by Theorem 2.3.4 (c), applied to µ and 1 instead of f
and g). But µ ? 1 = ε (by Proposition 2.4.6). Hence, 1 ? µ = µ ? 1 = ε.
We must prove the equivalence ( F = f ? 1) ⇐⇒ ( f = µ ? F ). In other words,
we must prove the two implications ( F = f ? 1) =⇒ ( f = µ ? F ) and ( F = f ? 1) ⇐=
( f = µ ? F ).
Proof of the implication ( F = f ? 1) =⇒ ( f = µ ? F ): Assume that F = f ? 1.
Then, F = f ? 1 = 1 ? f . Hence,

µ ? |{z}
F = µ ? (1 ? f ) = ( µ ? 1) ? f
| {z }
=1? f =ε
 
by Theorem 2.3.4 (b), applied to µ, 1 and f
instead of f , g and h
= ε? f = f (by Theorem 2.3.4 (a)) .

Thus, f = µ ? F. This proves the implication ( F = f ? 1) =⇒ ( f = µ ? F ).

Proof of the implication ( F = f ? 1) ⇐= ( f = µ ? F ): Assume that f = µ ? F.

47
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Hence,
f ?1 = 1? f = 1 ? ( µ ? F ) = (1 ? µ ) ? F
|{z} | {z }
=µ? F =ε
 
by Theorem 2.3.4 (b), applied to 1, µ and F
instead of f , g and h
= ε?F = F (by Theorem 2.3.4 (a), applied to F instead of f ) .
Thus, F = f ? 1. This proves the implication ( F = f ? 1) ⇐= ( f = µ ? F ).
Now, both implications are proven; hence, the proof of Proposition 2.5.2 is
complete.
Proof of Theorem 2.5.1. We have the following chain of equivalences:
( F = f ? 1)
⇐⇒ ( F (n) = ( f ? 1) (n) for all n ∈ N+ )
 

⇐⇒  F (n) = ∑ f (d) for all n ∈ N+ 

d|n

(since every n ∈ N+ satisfies

n
( f ? 1) ( n ) = ∑ f ( d ) 1 (by the definition of f ? 1)
d|n | {zd }
=1
(by the definition of 1)

= ∑ f (d)
d|n

). Hence, we have the following chain of equivalences:

 
 F (n) = ∑ f (d) for all n ∈ N+ 
d|n

⇐⇒ ( F = f ? 1)
⇐⇒ ( f = µ ? F ) (by Proposition 2.5.2)
⇐⇒ ( f (n) = (µ ? F ) (n) for all n ∈ N+ )
 
n
⇐⇒  f (n) = ∑ µ (d) F for all n ∈ N+ 
d|n
d

(since every n ∈ N+ satisfies

n
(µ ? F ) (n) = ∑ µ (d) F (by the definition of µ ? F )
d|n
d

). This proves Theorem 2.5.1.

48
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

2.6. Dirichlet convolution and multiplicativity

We will now connect the concept of multiplicative functions with the Dirichlet
convolution:

Theorem 2.6.1. Let f and g be two multiplicative arithmetic functions. Then,

the arithmetic function f ? g is also multiplicative.

Theorem 2.6.1 is a generalization of [NiZuMo91, Theorem 4.4], and the fol-

lowing proof follows the same ideas as the proof of [NiZuMo91, Theorem 4.4].16
Proof of Theorem 2.6.1. The function f is multiplicative. In other words, it satisfies
f (1) = 1, and

f (mn) = f (m) f (n) for any two coprime m ∈ N+ and n ∈ N+ . (19)

The function g is multiplicative. In other words, it satisfies g (1) = 1, and

g (mn) = g (m) g (n) for any two coprime m ∈ N+ and n ∈ N+ . (20)

The definition of f ? g yields

   
1 1
( f ? g ) (1) = ∑ f (d) g = f (1) g = 1.
d |1
d | {z } 1
=1 | {z }
= g(1)=1

Now, we want to prove that f ? g is multiplicative. In order to do so, we need

to verify that ( f ? g) (1) = 1 and that

( f ? g) (mn) = ( f ? g) (m) · ( f ? g) (n) (21)

for any two coprime m ∈ N+ and n ∈ N+ . Since ( f ? g) (1) = 1 is already

proven, it thus only remains to prove (21).
So let m ∈ N+ and n ∈ N+ be coprime. We need to prove (21).
For any N ∈ N+ , let D ( N ) be the set of all positive divisors of N.
Consider the map

f : D (m) × D (n) → D (mn) , (d, e) 7→ de.

This map f is well-defined (because if d and e are positive divisors of m and n,
respectively, then de is a positive divisor of mn).
Consider the map

g : D (mn) → D (m) × D (n) , u 7→ (gcd (u, m) , gcd (u, n)) .

This map g is well-defined (because if u is a positive divisor of mn, then gcd (u, m)
and gcd (u, n) are positive divisors of m and n, respectively).
16 Note that the claim of Theorem 2.6.1 is also the first part of [NiZuMo91, §8.2, problem 1].

49
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

We have f ◦ g = id 17 and g ◦ f = id 18 . Hence, the maps f and g are

mutually inverse. In particular, this shows that the map f is a bijection. In other
words, the map D (m) × D (n) → D (mn) , (d, e) 7→ de is a bijection (since this
map is precisely f).
We make two more simple observations:

1. We have

f (de) = f (d) f (e) for any d ∈ D (m) and e ∈ D (n) . (22)

17 Proof.
Let u ∈ D (mn). Thus, u is a positive divisor of mn. Therefore, gcd (u, mn) = u.
The definition of g shows that g (u) = (gcd (u, m) , gcd (u, n)). Now,
 

(f ◦ g) (u) = f  g (u)  = f (gcd (u, m) , gcd (u, n))

 
| {z }
=(gcd(u,m),gcd(u,n))
= gcd (u, m) · gcd (u, n) (by the definition of f)
= gcd (u, mn) (by Proposition 1.2.10)
= u.

Now, forget that we fixed u. We thus have proven that (f ◦ g) (u) = u for every u ∈ D (mn).
In other words, f ◦ g = id.
18 Proof. Let ( d, e ) ∈ D ( m ) × D ( n ). Then, f ( d, e ) = de (by the definition of f), and

 

(g ◦ f) (d, e) = g f (d, e) = g (de) = (gcd (de, m) , gcd (de, n))

 
| {z }
=de

(by the definition of g).

We have (d, e) ∈ D (m) × D (n). In other words, d ∈ D (m) and e ∈ D (n). In other words,
d is a positive divisor of m, and e is a positive divisor of n. Thus, d | m and e | n.
Let d0 = gcd (de, m). Then, d0 = gcd (de, m) | m and e | n. Hence, Corollary 1.2.4 (applied
to d0 , e, m and n instead of a, b, c and d) yields gcd (d0 , e) | gcd (m, n) = 1 (since m and n are
coprime). Hence, gcd (d0 , e) = 1.
Note also that d0 = gcd (de, m) | de = ed. Thus, Proposition 1.2.8 (applied to x = d0 , y = e
and z = d) yields d0 | d (since gcd (d0 , e) = 1).
On the other hand, d | de and d | m. Thus, Proposition 1.2.3 (applied to d, de and m instead
of a, b and c) yields d | gcd (de, m). In other words, d | d0 (since d0 = gcd (de, m)).
Now, the integers d and d0 are positive and thus nonnegative. Hence, Proposition 1.0.2
(applied to u = d and v = d0 ) yields d = d0 (since d | d0 and d0 | d). Thus, d = d0 = gcd (de, m).
In other words, gcd (de, m) = d. The same argument (with the roles of m and n interchanged,
and correspondingly also the roles of d and e interchanged) shows that gcd (ed, n) = e. Now,
 

(g ◦ f) (d, e) = gcd (de, m), gcd (de, n)  = (d, e) .

 
| {z } | {z }
=d =gcd(ed,n)=e

Now, forget that we fixed (d, e). We thus have shown that (g ◦ f) (d, e) = (d, e) for each
( e) ∈ D (m) × D (n). In other words, g ◦ f = id.
d,

50
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

19

2. We have
 mn  m n
g =g g for any d ∈ D (m) and e ∈ D (n) . (23)
de d e
20

19 Proof of (22): Let d ∈ D (m) and e ∈ D (n). In other words, d is a positive divisor of m, and e is
a positive divisor of n. Hence, d | m and e | n. Therefore, Corollary 1.2.4 (applied to d, e, m
and n instead of a, b, c and d) yields gcd (d, e) | gcd (m, n) = 1 (since m and n are coprime).
Hence, gcd (d, e) = 1. In other words, d and e are coprime. Hence, (19) (applied to d and e
instead of m and n) yields f (de) = f (d) f (e), qed.
20 Proof of (23): Let d ∈ D ( m ) and e ∈ D ( n ). In other words, d is a positive divisor of m, and
m n
e is a positive divisor of n. Hence, d | m and e | n. This shows that and are integers.
d e
m n m
Furthermore, the numbers and are positive (since m, d, n and e are positive). Hence,
d e d
n
and are positive integers.
e
m n m n
Moreover, | m and | n. Hence, Corollary 1.2.4 (applied to , , m and n instead
d e  d e
m n
of a, b, c and d) yields gcd , | gcd (m, n) = 1 (since m and n are coprime). Hence,
m n d e
m n m n
gcd , = 1. In other words, and are coprime. Hence, (20) (applied to and
d e d e   d e
 
m n m n  
 mn 
 = g m·n =
 
instead of m and n) yields g · = g g . Hence, g 
d e d e  de 
 |{z}  d e
 m n
= ·
m n d e
g g , qed.
d e

51
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Now, the definition of f ? g yields

( f ? g) (mn)
 mn   mn   
here, we have renamed the
= ∑ f (d) g
d
= ∑ f (u) g
u summation index d as u
d|mn u|mn
|{z}
= ∑
u∈D(mn)
 mn   mn 
= ∑ f (u) g
u
= ∑ f (de) g
de
u∈D(mn) (d,e)∈D(m)×D(n)
| {z }
= ∑ ∑
d∈D(m) e∈D(n)
 
here, we have substituted de for u in the sum, since the
map D (m) × D (n) → D (mn) , (d, e) 7→ de is a bijection
 mn 
= ∑ ∑ | {z }
f ( de ) g
d∈D(m) e∈D(n) | {z de }
| {z } | {z } = f (d) f (e)  m   n 
=∑ =∑ (by (22)) = g g
d|m e|n d e
(by (23))
  
m n m n
= ∑ ∑ f (d) f (e) g g =  ∑ f (d) g  ∑ f (e) g 
d|m e|n
d e d|m
d e|n
e
  
m
 ∑ f (d) g n 
 
=  ∑ f (d) g
d|m
d d|n
d

(here, we renamed the summation index e as d in the second sum). Comparing

this with

( f ? g) (m) · ( f ? g) (n)
| {z } | {z }
m
 
n
 
= ∑ f (d) g = ∑ f (d) g
d|m d d|n d
(by the definition of f ? g) (by the definition of f ? g)
  
m n
=  ∑ f (d) g  ∑ f (d) g ,
d|m
d d|n
d

we obtain ( f ? g) (mn) = ( f ? g) (m) · ( f ? g) (n). Thus, (21) is proven. As we

have said, this completes the proof of Theorem 2.6.1.
Notice that Theorem 2.6.1 has no analogue for totally multiplicative functions:
The Dirichlet convolution f ? g of two totally multiplicative functions might not
be totally multiplicative.
We can use Theorem 2.6.1 to prove (and sometimes reprove) parts of Proposi-
tion 2.2.2:

52
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Proof of Proposition 2.2.2 (b). The arithmetic function 1 is clearly multiplicative

(and totally multiplicative). Thus, Theorem 2.6.1 (applied to f = 1 and g = 1)
shows that 1 ? 1 is multiplicative. But since 1 ? 1 = d (by Proposition 2.4.1), this
shows that d is multiplicative. This proves Proposition 2.2.2 (b).
Proof of Proposition 2.2.2 (e). Let k ∈ Z. Define the arithmetic function ι k as in
Proposition 2.4.2 (b). This ι k is clearly multiplicative (and totally multiplicative).
We also know that the arithmetic function 1 is clearly multiplicative. Thus,
Theorem 2.6.1 (applied to f = ι k and g = 1) shows that ι k ? 1 is multiplicative.
But since ι k ? 1 = σk (by Proposition 2.4.2 (b)), this shows that σk is multiplicative.
Applying this to k = 1, we conclude that σ is multiplicative (since σ1 = σ). This
completes the proof of Proposition 2.2.2 (e).
Second proof of Proposition 2.2.2 (a). The arithmetic function µ is multiplicative (by
Proposition 2.2.2 (f)). The arithmetic function ι is clearly multiplicative (and to-
tally multiplicative). Hence, Theorem 2.6.1 (applied to f = µ and g = ι) shows
that µ ? ι is multiplicative. But since µ ? ι = φ (by Proposition 2.4.8), this shows
that φ is multiplicative. This proves Proposition 2.2.2 (a) again.
As an easy consequence of Theorem 2.6.1, we can obtain [NiZuMo91, Theorem
4.4]:

Corollary 2.6.2. Let f : N+ → C be a multiplicative arithmetic function. De-

fine an arithmetic function F : N+ → C by

F (n) = ∑ f (d) for every positive integer n. (24)

d|n

Then, the function F is multiplicative.

Proof of Corollary 2.6.2. The arithmetic function 1 is clearly multiplicative (and

totally multiplicative). Thus, Theorem 2.6.1 (applied to g = 1) shows that f ? 1
is multiplicative. But every positive integer n satisfies
n
( f ? 1) ( n ) = ∑ f ( d ) 1 (by the definition of f ? 1)
d|n | {zd }
=1
(by the definition of 1)

= ∑ f (d) = F (n) (by (24)) .

d|n

Hence, f ? 1 = F. But recall that f ? 1 is multiplicative. In other words, F is

multiplicative (since f ? 1 = F). This proves Corollary 2.6.2.

53
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

2.7. Explicit formulas from multiplicativity

One of the nice things about multiplicative arithmetic functions is that, in order
to compute their values, it suffices to compute their values on prime powers:

Proposition 2.7.1. Let f : N+ → C be a multiplicative function. Let n ∈ N+ .

Then,  
f ( n ) = ∏ f pv p (n) .
p∈PF n

(See Definition 2.1.5 for the definition of PF n.)

Applying this proposition to f = φ, f = d and f = σk , we easily obtain

Theorem 2.1.6,
 Theorem
 2.1.7 and Theorem 2.1.8, respectively (once we compute
the values f pv p (n) , but this is easy in all three cases).
Proposition 2.7.1 follows from the following fact:

Proposition 2.7.2. Let f : N+ → C be a multiplicative function. Let

a1 , a2 , . . . , ak be finitely many pairwise coprime21 positive integers. Then,

f ( a1 a2 · · · a k ) = f ( a1 ) f ( a2 ) · · · f ( a k ) .

Both the proof of Proposition 2.7.2 (by induction over k) and the proof of
Proposition 2.7.1 (using Proposition 2.7.2) are rather straightforward:
Proof of Proposition 2.7.2. The function f is multiplicative. In other words, it sat-
isfies f (1) = 1 and
f (mn) = f (m) f (n) for any two coprime m ∈ N+ and n ∈ N+ (25)
(by the definition of “multiplicative”).
The integers a1 , a2 , . . . , ak are pairwise coprime. In other words,
au is coprime to av (26)
for any integers u and v satisfying 1 ≤ u < v ≤ k.
We shall show that
f ( a1 a2 · · · a i ) = f ( a1 ) f ( a2 ) · · · f ( a i ) (27)
for every i ∈ {0, 1, . . . , k }.
Proof of (27): We shall prove (27) by induction over i:
Induction base: We have a1 a2 · · · a0 = (empty product) = 1. Applying the map
f to both sides of this equation, we obtain
f ( a1 a2 · · · a0 ) = f (1) = 1.
21 We say that k integers a1 , a2 , . . . , ak are pairwise coprime if they have the property that au is
coprime to av for any integers u and v satisfying 1 ≤ u < v ≤ k.

54
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Comparing this with f ( a1 ) f ( a2 ) · · · f ( a0 ) = (empty product) = 1, we obtain

f ( a1 a2 · · · a0 ) = f ( a1 ) f ( a2 ) · · · f ( a0 ). In other words, (27) holds for i = 0. This
completes the induction base.
Induction step: Let j ∈ {0, 1, . . . , k } be positive. Assume that (27) holds for
i = j − 1. We must prove that (27) holds for i = j.
We have assumed that (27) holds for i = j − 1. In other words, we have



f a 1 a 2 · · · a j −1 = f ( a 1 ) f ( a 2 ) · · · f a j −1 .

But au is coprime to a j for every u ∈ {1, 2, . . . , j − 1} 22 . Hence, Corollary

1.2.7 (applied to n = j − 1, cu = au and m = a j ) shows that a1 a2 · · · a j−1 is
coprime to a j . Therefore, (25) (applied to m = a1 a2 · · · a j−1 and n = a j ) yields





f a 1 a 2 · · · a j −1 a j = f a 1 a 2 · · · a j −1 f a j
| {z }
= f ( a1 ) f ( a2 )··· f ( a j−1 )





= f ( a 1 ) f ( a 2 ) · · · f a j −1 f a j = f ( a 1 ) f ( a 2 ) · · · f a j .





Comparing this with f a1 a2 · · · a j−1 a j = f a1 a2 · · · a j , we obtain f a1 a2 · · · a j =


f ( a1 ) f ( a2 ) · · · f a j . In other words, (27) holds for i = j. Thus, the induction
step is complete, and so (27) is proven.
Now, we can apply (27) to i = k. As a result, we obtain f ( a1 a2 · · · ak ) =
f ( a1 ) f ( a2 ) · · · f ( ak ). This proves Proposition 2.7.2.
Let us restate Proposition 2.7.2 in a more convenient form before we come to
the proof of Proposition 2.7.1:

Corollary 2.7.3. Let f : N+ → C be a multiplicative function. Let S be a finite

set. Let ms be a positive integer for each s ∈ S. Assume that the integers ms
and mt are coprime whenever s and t are two distinct elements of S. Then,
!
f ∏ ms = ∏ f (ms ) .
s∈S s∈S

Proof of Corollary 2.7.3. Let (s1 , s2 , . . . , sk ) be a list of all elements of S (with each
element appearing exactly once in the list). Then, ∏ ms = ms1 ms2 · · · msk and
s∈S
∏ f ( m s ) = f ( m s1 ) f ( m s2 ) · · · f ( m s k ).
s∈S
Also, if i and j are two distinct elements of {1, 2, . . . , k }, then the integers msi
and ms j are coprime23 . In other words, ms1 , ms2 , . . . , msk are pairwise coprime
22 Proof. Let u ∈ {1, 2, . . . , j − 1}. Thus, u is an integer satisfying 1 ≤ u ≤ j − 1. Now, 1 ≤ u ≤
j − 1 < j ≤ k (since j ∈ {0, 1, . . . , k}). Therefore, (26) (applied to v = j) shows that au is
coprime to a j . Qed.
23 Proof. Let i and j be two distinct elements of {1, 2, . . . , k }.

55
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

integers. Hence, Proposition 2.7.2 (applied to ai = msi ) yields

f ( m s1 m s2 · · · m s k ) = f ( m s1 ) f ( m s2 ) · · · f ( m s k ) .

Thus,
 
 
∏ ms  = f (ms ms · · · ms ) = f (ms ) f (ms ) · · · f (ms ) = ∏ f (ms ) .
 
f
  1 2 k 1 2 k
 s∈S  s∈S
| {z }
=ms1 ms2 ···msk

This proves Corollary 2.7.3.

Proof of Proposition 2.7.1. The prime factorization of n is n = ∏ p v p (n) = ∏ s vs (n)

p∈PF n s∈PF n
(here, we renamed the index p as s in the product). Clearly, svs (n) is a positive
integer for each s ∈ PF n. Furthermore, the integers svs (n) and tvt (n) are coprime
whenever s and t are two distinct elements of PF n 24 . Hence, Corollary 2.7.3
(applied to S = PF n and ms = svs (n) ) yields
!
   
f ∏ s vs (n)
= ∏ f s vs (n)
= ∏ f p v p (n)
s∈PF n s∈PF n p∈PF n

(here, we renamed the index s as p in the product). Thus,

 
!
   
n
f  |{z} = f ∏ s vs (n)
= ∏ f p v p (n)
.
 
s∈PF n p∈PF n
 
= ∏ svs (n)
s∈PF n

This proves Proposition 2.7.1.

The list (s1 , s2 , . . . , sk ) contains no element more than once (because of its definition). In
other words, the elements s1 , s2 , . . . , sk are pairwise distinct. Hence, si 6= s j (since i 6= j). In
other words, the elements si and s j are distinct.
But the integers ms and mt are coprime whenever s and t are two distinct elements of S.
Applying this to s = si and t = s j , we conclude that the integers msi and ms j are coprime
(since si and s j are distinct). Qed.
24 Proof. Let s and t be two distinct elements of PF n. We must prove that the integers svs (n) and

tvt (n) are coprime.

All elements of PF n are primes. Hence, s is a prime (since s is an element of PF n). Similarly,
t is a prime. Hence, s and t are two distinct primes (since s and t are distinct). Thus, Corollary
1.2.13 (b) (applied to vs (n) and vt (n) instead of n and m) yields that the integers svs (n) and
tvt (n) are coprime. Qed.

56
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

2.8. Pointwise products

Let us briefly discuss a much simpler way to “multiply” two arithmetic functions
than Dirichlet convolution: the pointwise product.

Definition 2.8.1. Let f : N+ → C and g : N+ → C be two arithmetic func-

tions. We define a new arithmetic function f · g : N+ → C by

( f · g) (n) = f (n) g (n) for every n ∈ N+ .

This new function f · g is called the pointwise product of f and g.

We have the following (much simpler) analogue of Theorem 2.3.4:

Theorem 2.8.2. (a) We have 1 · f = f · 1 = f for every arithmetic function f .

(b) We have f · ( g · h) = ( f · g) · h for every three arithmetic functions f , g
and h.
(c) We have f · g = g · f for every two arithmetic functions f and g.

Proof of Theorem 2.8.2. (c) Let f and g be two arithmetic functions. Let n ∈ N+ .
The definition of f · g yields ( f · g) (n) = f (n) g (n) = g (n) f (n). But the def-
inition of g · f yields ( g · f ) (n) = g (n) f (n). Comparing these two equalities,
we obtain ( f · g) (n) = ( g · f ) (n).
Now, forget that we fixed n. We thus have shown that ( f · g) (n) = ( g · f ) (n)
for each n ∈ N+ . In other words, f · g = g · f . This proves Theorem 2.8.2 (c).
(a) Let f be an arithmetic function. Every n ∈ N+ satisfies
(1 · f ) ( n ) = 1 (n) · f (n) (by the definition of 1 · f )
| {z }
=1
(by the definition
of 1)
= f (n)
In other words, 1 · f = f . But Theorem 2.8.2 (c) (applied to g = 1) yields
f · 1 = 1 · f . Thus, f · 1 = 1 · f = f . This proves Theorem 2.8.2 (a).
(b) Let f , g and h be three arithmetic functions. Let n ∈ N+ . The definition of
( f · g) · h yields
(( f · g) · h) (n) = ( f · g) (n) h (n) = f (n) g (n) h (n) .
| {z }
= f (n) g(n)
(by the definition
of f · g)

But the definition of f · ( g · h) yields

( f · ( g · h)) (n) = f (n) · ( g · h) (n) = f (n) g (n) h (n) .
| {z }
= g(n)h(n)
(by the definition
of g·h)

57
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Comparing these two equalities, we obtain ( f · ( g · h)) (n) = (( f · g) · h) (n).

Now, forget that we fixed n. We thus have proven that ( f · ( g · h)) (n) =
(( f · g) · h) (n) for every n ∈ N+ . In other words, f · ( g · h) = ( f · g) · h. This
proves Theorem 2.8.2 (b).
We also get the following essentially obvious analogue of Theorem 2.6.1:

Theorem 2.8.3. Let f and g be two multiplicative arithmetic functions. Then,

the arithmetic function f · g is also multiplicative.

Theorem 2.8.3 is just Proposition 2.2.3 (a) with different notations, but let us
prove it again:
Proof of Theorem 2.8.3. The function f is multiplicative. In other words, it satisfies
f (1) = 1, and
f (mn) = f (m) f (n) for any two coprime m ∈ N+ and n ∈ N+ . (28)
The function g is multiplicative. In other words, it satisfies g (1) = 1, and
g (mn) = g (m) g (n) for any two coprime m ∈ N+ and n ∈ N+ . (29)
The definition of f · g yields
( f · g) (1) = f (1) g (1) = 1.
| {z } | {z }
=1 =1

Now, we want to prove that f · g is multiplicative. In order to do so, we need

to verify that ( f · g) (1) = 1 and that
( f · g) (mn) = ( f · g) (m) · ( f · g) (n) (30)
for any two coprime m ∈ N+ and n ∈ N+ . Since ( f · g) (1) = 1 is already
proven, it thus only remains to prove (30).
So let m ∈ N+ and n ∈ N+ be coprime. We need to prove (30).
The definition of f · g yields
( f · g) (mn) = f (mn) g (mn) = f (m) f (n) g (m) g (n) .
| {z } | {z }
= f (m) f (n) = g(m) g(n)
(by (28)) (by (29))

Comparing this with

( f · g) (m) · ( f · g) (n) = f (m) g (m) f (n) g (n)
| {z } | {z }
= f (m) g(m) = f (n) g(n)
(by the definition of f · g) (by the definition of f · g)
= f (m) f (n) g (m) g (n) ,
we obtain ( f · g) (mn) = ( f · g) (m) · ( f · g) (n). Thus, (30) is proven. As we have
said, this completes the proof of Theorem 2.8.3.

58
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

2.9. Lowest common multiples

Let us next define yet another way of multiplying two arithmetic functions, the
so-called lcm convolution. We begin by introducing lowest common multiples of
two positive integers:

Definition 2.9.1. Let a be an integer. A multiple of a means an integer that is

divisible by a.

For instance, 12 is a multiple of 3, since 3 | 12.

Definition 2.9.2. Let b and c be two integers. A common multiple of b and c

means an integer that is both a multiple of b and a multiple of c.

For example, 12 is a common multiple of 4 and 6, since 4 | 12 and 6 | 12.

Proposition 2.9.3. Let b and c be two positive integers. Then, there exists a
smallest positive common multiple of b and c. (In other words, the set of all
positive common multiples of b and c has a smallest element.)

Proof of Proposition 2.9.3. We know that b and c are positive integers. Hence,
their product bc is a positive integer as well. Moreover, this positive integer bc is
clearly a multiple of b (since b | bc) and a multiple of c (since c | bc); thus, it is a
common multiple of b and c. Hence, bc is a positive common multiple of b and
c. Therefore, the set of all positive common multiples of b and c has at least one
element (namely, the element bc). Thus, this set is nonempty. Therefore, this set
is a nonempty set of positive integers.
But it is well-known that any nonempty set of positive integers has a mini-
mum element. Hence, the set of all positive common multiples of b and c has
a minimum element (because this set is a nonempty set of positive integers). In
other words, there exists a smallest positive common multiple of b and c.

Definition 2.9.4. Let b and c be two positive integers. Then, lcm (b, c) is de-
fined to be the smallest positive common multiple of b and c. (This is well-
defined, because Proposition 2.9.3 shows that there exists a smallest positive
common multiple of b and c.)
This number lcm (b, c) is called the lowest common multiple of b and c or the
least common multiple of b and c or, briefly, the lcm of b and c. Clearly, it satisfies
lcm (b, c) = lcm (c, b) and b | lcm (b, c) and c | lcm (b, c). Notice that lcm (b, c)
is a positive integer (by its definition).

We have been slightly lazy here and only defined the lowest common multiple
of two positive integers. We could extend this definition to two arbitrary inte-
gers, or even to several integers. But we will not need this generality in what
follows.

59
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Older books such as [NiZuMo91] often denote the lcm of two integers b and c
by [b, c] (rather than by lcm (b, c) as we do).
The most important property of lcms is the following fact, which is in a sense
a mirror image of Proposition 1.2.3:
Proposition 2.9.5. Let b and c be two positive integers. Let a be an integer
such that b | a and c | a. Then, lcm (b, c) | a.
In words, Proposition 2.9.5 says that any common multiple of two positive
integers must be divisible by the lcm of these two integers.
Proof of Proposition 2.9.5. Let ` = lcm (b, c). Then, ` is the smallest positive com-
mon multiple of b and c (by the definition of the lcm). Hence, ` is a positive
integer.
Let q and r be the quotient and the remainder obtained when dividing a by
`. Thus, we have q ∈ Z, r ∈ {0, 1, . . . , ` − 1} and a = q` + r (by the definition
of division with remainder). From a = q` + r, we obtain a − q` = r. From
r ∈ {0, 1, . . . , ` − 1}, we obtain r ≥ 0 and r ≤ ` − 1 < `.
We have b | a and b | lcm (b, c) = ` | −q`. In other words, the two integers a
and −q` are both divisible by b. Hence, the sum a + (−q`) of these two integers
must also be divisible by b (by Proposition 1.0.1, applied to b, a and −q` instead
of a, u and v). In other words, b | a + (−q`). In view of a + (−q`) = a − q` = r,
this rewrites as b | r. The same argument (applied to c instead of b) yields c | r
(since c | lcm (b, c) = `).
Now, r is an integer that is both a multiple of b (since b | r) and a multiple of
c (since c | r). In other words, r is a common multiple of b and c.
Recall that ` is the smallest positive common multiple of b and c. Hence, each
positive common multiple of b and c must be ≥ `. Thus, if r was positive, then r
would be ≥ ` (because r is a common multiple of b and c, and therefore would be
a positive common multiple of b and c); but this would contradict r < `. Hence,
r cannot be positive. In other words, we must have r ≤ 0. Combining this with
r ≥ 0, we obtain r = 0. Thus, a = q` + |{z} r = q`. Now, lcm (b, c) = ` | q` = a.
=0
This proves Proposition 2.9.5.
Corollary 2.9.6. Let b and c be two positive integers. Let a be an integer. Then,
we have the logical equivalence

(b | a and c | a) ⇐⇒ (lcm (b, c) | a) .

Proof of Corollary 2.9.6. We have the logical implication

(lcm (b, c) | a) =⇒ (b | a and c | a) (because if lcm (b, c) | a holds, then we
have b | lcm (b, c) | a and c | lcm (b, c) | a, and therefore (b | a and c | a)). But
we also have the logical implication (b | a and c | a) =⇒ (lcm (b, c) | a) (by
Proposition 2.9.5). Combining these two implications, we obtain the equivalence
(b | a and c | a) ⇐⇒ (lcm (b, c) | a). This proves Corollary 2.9.6.

60
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

The gcd and the lcm of two positive integers b and c are connected by the
equality gcd (b, c) · lcm (b, c) = bc (see, for example, [NiZuMo91, Theorem 1.13]
or [Conrad19, Theorem 7]). We will not have any use for this connection, how-
ever. Most texts on number theory study the lcm as an afterthought of the gcd;
however, as Keith Conrad shows in [Conrad19, Theorem 7], it is actually easier
to build up the theory of gcds and lcms (of two positive integers) by starting
with lcms first.
For future use, let us state a trivial property of lcms:

Lemma 2.9.7. Let n ∈ N+ . Then, the set {(d, e) ∈ N+ × N+ | lcm (d, e) = n}

is finite.
Proof of Lemma 2.9.7. If (d, e) ∈ N+ × N+ satisfies lcm (d, e) = n, then (d, e) ∈
{1, 2, . . . , n}2 25 . In other words, the set {(d, e) ∈ N+ × N+ | lcm (d, e) = n} is
a subset of {1, 2, . . . , n}2 . Therefore, this set {(d, e) ∈ N+ × N+ | lcm (d, e) = n}
is finite (since the set {1, 2, . . . , n}2 is finite). This proves Lemma 2.9.7.

2.10. Lcm-convolution
We can now define another way of “multiplying” two arithmetic functions:

Definition 2.10.1. Let f : N+ → C and g : N+ → C be two arithmetic

? g : N+ → C by
functions. We define a new arithmetic function f e

? g) (n) =
(f e ∑ f (d) g (e) for every n ∈ N+ .
d ∈N+ ; e ∈N+ ;
lcm(d,e)=n

(This is well-defined, because the sum on the right hand side of this equality
is finite26 .)
This new function f e? g is called the lcm-convolution of f and g.

25 Proof. Let (d, e) ∈ N+ × N+ be such that lcm (d, e) = n. We must prove that (d, e) ∈
{1, 2, . . . , n}2 .
We have (d, e) ∈ N+ × N+ ; thus, d ∈ N+ and e ∈ N+ . In other words, d and e are positive
integers. Also, n is a positive integer (since n ∈ N+ ); thus, n > 0. Now, d | lcm (d, e) = n. In
other words, there exists an integer z such that n = dz. Consider this z. If we had z ≤ 0, then
we would have n = d |{z} z ≤ d0 (since d is positive), which would contradict n > 0 = d0.
≤0
Thus, we cannot have z ≤ 0. Hence, we have z > 0. Thus, z ≥ 1 (since z is an integer). Now,
z ≥ d1 (since d is positive), so that n ≥ d1 = d. In other words, d ≤ n. Hence, d ∈
n = d |{z}
≥1
{1, 2, . . . , n} (since d is a positive integer). Similarly, e ∈ {1, 2, . . . , n} (since e | lcm (d, e) = n).
Combining d ∈ {1, 2, . . . , n} with e ∈ {1, 2, . . . , n}, we find (d, e) ∈ {1, 2, . . . , n}2 . Qed.
26 Proof. Let n ∈ N . We must prove that the sum
+ ∑ f (d) g (e) is finite. In other words,
d ∈N+ ; e ∈N+ ;
lcm(d,e)=n
we must prove that there are only finitely many pairs (d, e) ∈ N+ × N+ such that lcm (d, e) =

61
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Example 2.10.2. Let f : N+ → C and g : N+ → C be two arithmetic functions.

Then,

? g ) (1) =
(f e ∑ f ( d ) g ( e ) = f (1) g (1) ;
d ∈N+ ; e ∈N+ ;
lcm(d,e)=1

? g) ( p) =
(f e ∑ f (d) g (e)
d ∈N+ ; e ∈N+ ;
lcm(d,e)= p
= f (1) g ( p ) + f ( p ) g (1) + f ( p ) g ( p ) for any prime p;
? g ) (6) =
(f e ∑ f (d) g (e)
d ∈N+ ; e ∈N+ ;
lcm(d,e)=6
= f (1) g (6) + f (2) g (3) + f (2) g (6) + f (3) g (2) + f (3) g (6)
+ f (6) g (1) + f (6) g (2) + f (6) g (3) + f (6) g (6) .

? on the set of arithmetic functions appears in [Lehmer31, The-

The operation e
orem 1] (where f e? g is denoted by h) and in [Toth14, §4.4] (where this operation
? is denoted by ⊕, and generalized to functions of r arguments). It is also known
e
as the Lehmer convolution (or Lehmer product) or von Sterneck convolution. We shall
prove the following of its properties (analogues of Theorems 2.3.4 and 2.6.1,
respectively):

Theorem 2.10.3. (a) We have ε e ?f = fe ? ε = f for every arithmetic function f .

(b) We have f e? (g e
? h) = ( f e
? g) e
? h for every three arithmetic functions f ,
g and h.
?g = ge
(c) We have f e ? f for every two arithmetic functions f and g.

Theorem 2.10.4. Let f and g be two multiplicative arithmetic functions. Then,

? g is also multiplicative.
the arithmetic function f e

The proofs will rely on the following result of von Sterneck and Lehmer
(see, e.g., [Lehmer31, Theorem 1]), which connects the lcm-convolution with
the pointwise product and the Dirichlet convolution:

Theorem 2.10.5. Let f : N+ → C and g : N+ → C be two arithmetic functions.

Then,
? g ) = (1 ? f ) · (1 ? g ) .
1 ? (f e

Proof of Theorem 2.10.5. Define three arithmetic functions F, G and H by

F = 1 ? f, G = 1?g and H = 1 ? (f e
? g) .
n. In other words, we must prove that the set {(d, e) ∈ N+ × N+ | lcm (d, e) = n} is finite.
But this follows immediately from Lemma 2.9.7. Qed.

62
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Let n ∈ N+ . Then, H = 1 ? ( f e
? g) = ( f e
? g) ? 1 (by Theorem 2.3.4 (c), applied
? g instead of f and g). Applying both sides of this equality to n, we
to 1 and f e
obtain

H (n) = (( f e
? g ) ? 1) ( n )
n
= ∑(f e
? g) (d) 1 ? g ) ? 1)
(by the definition of ( f e
d|n | {zd }
=1
(by the definition of 1)

= ∑(f e
? g) (d) = ∑ ? g) (s)
(f e
d|n s|n
| {z }
= ∑ f (d) g(e)
d ∈N+ ; e ∈N+ ;
|{z}
= ∑
s is a positive lcm(d,e)=s
divisor of n ? g)
(by the definition of f e
(by Definition 2.1.3)

(here, we have renamed the summation index d as s)

= ∑ ∑ f (d) g (e)
s is a positive d ∈N+ ; e ∈N+ ;
divisor of n lcm(d,e)=s
| {z }
= ∑
d ∈N+ ; e ∈N+ ;
lcm(d,e) is a positive
divisor of n
= ∑ f (d) g (e) . (31)
d ∈N+ ; e ∈N+ ;
lcm(d,e) is a positive
divisor of n

Now, let d ∈ N+ and e ∈ N+ be arbitrary. Then, d and e are two positive

integers. Hence, Corollary 2.9.6 (applied to d, e and n instead of b, c and a)
shows that we have the logical equivalence

(d | n and e | n) ⇐⇒ (lcm (d, e) | n) . (32)

Also, lcm (d, e) is positive (by the definition of lcm (d, e)). Thus, lcm (d, e) is a
positive divisor of n if and only if lcm (d, e) is a divisor of n. Hence, we have the
following chain of logical equivalences:

(lcm (d, e) is a positive divisor of n)

⇐⇒ (lcm (d, e) is a divisor of n) ⇐⇒ (lcm (d, e) | n)
⇐⇒ (d | n and e | n) (by (32)) .

Now, forget that we fixed d and e. We thus have proven the logical equivalence

(lcm (d, e) is a positive divisor of n) ⇐⇒ (d | n and e | n)

for every d ∈ N+ and e ∈ N+ . Hence, we have the following equality of

63
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

summation signs:

∑ = ∑ = ∑ ∑ =∑ ∑.
d ∈N+ ; e ∈N+ ; d ∈N+ ; e ∈N+ ; d ∈N+ e ∈N+ d|n e|n
lcm(d,e) is a positive d|n and e|n d|n e|n
divisor of n | {z } | {z }
= ∑ = ∑
d is a positive e is a positive
divisor of n divisor of n
=∑ =∑
d|n e|n
(by Definition 2.1.3) (by Definition 2.1.3)

Thus, (31) becomes

H (n) = ∑ f (d) g (e) = ∑ ∑ f (d) g (e) . (33)

d ∈N+ ; e ∈N+ ; d|n e|n
lcm(d,e) is a positive
divisor of n
| {z }
=∑ ∑
d|n e|n

On the other hand, F = 1 ? f = f ? 1 (by Theorem 2.3.4 (c), applied to 1 and f

instead of f and g). Applying both sides of this equality to n, we obtain
n
F ( n ) = ( f ? 1) ( n ) = ∑ f ( d ) 1 (by the definition of f ? 1)
d|n | d
{z }
=1
(by the definition of 1)

= ∑ f (d) . (34)
d|n

The same argument (applied to G and g instead of F and f ) yields G (n) =

∑ g (d). Hence,
d|n
G (n) = ∑ g (d) = ∑ g (e) (35)
d|n e|n

(here, we have renamed the summation index d as e). Now, the definition of
F · G yields
  

( F · G ) (n) = F (n) G (n) = ∑ f (d) ∑ g (e)

d|n e|n
| {z } | {z }
= ∑ f (d) = ∑ g(e)
d|n e|n
(by (34)) (by (35))

=∑ ∑ f (d) g (e) = H (n) (by (33)) .

d|n e|n

Now, forget that we fixed n. We thus have proven that ( F · G ) (n) = H (n) for
each n ∈ N+ . In other words, F · G = H.

64
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

F · |{z}
Comparing this with |{z} G = (1 ? f ) · (1 ? g), we obtain
=1? f =1? g

(1 ? f ) · (1 ? g ) = H = 1 ? ( f e
? g) .

This proves Theorem 2.10.5.

An easy consequence of Theorem 2.10.5 and Proposition 2.5.2, we now obtain
the following:

Corollary 2.10.6. Let f : N+ → C and g : N+ → C be two arithmetic func-

tions. Then,
? g = µ ? ((1 ? f ) · (1 ? g)) .
fe

Proof of Corollary 2.10.6. Define an arithmetic function H by H = 1 ? ( f e

? g ).
Then,

? g ) = (1 ? f ) · (1 ? g )
H = 1 ? (f e (by Theorem 2.10.5) .

Also, H = 1 ? ( f e
? g) = ( f e
? g) ? 1 (by Theorem 2.3.4 (c), applied to 1 and f e
?g
instead of f and g).
But Proposition 2.5.2 (applied to f e ? g and H instead of f and F) yields that
we have the following logical equivalence:

? g) ? 1) ⇐⇒ ( f e
(H = ( f e ? g = µ ? H) .

? g = µ ? H (since we have H = ( f e
Hence, we have f e ? g) ? 1). Thus,

?g = µ?
fe H
|{z} = µ ? ((1 ? f ) · (1 ? g)) .
=(1? f )·(1? g)

This proves Corollary 2.10.6.

We also have the following simple corollary of Proposition 2.5.2:

Corollary 2.10.7. Let f : N+ → C and g : N+ → C be two arithmetic functions

such that 1 ? f = 1 ? g. Then, f = g.

Proof of Corollary 2.10.7. Define an arithmetic function F by F = 1 ? f . Thus,

F = 1 ? f = f ? 1 (by Theorem 2.3.4 (c), applied to 1 and f instead of f and g).
But Proposition 2.5.2 yields that we have the following logical equivalence:

( F = f ? 1) ⇐⇒ ( f = µ ? F ) .

Hence, we have f = µ ? F (since F = f ? 1).

65
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

On the other hand, F = 1 ? f = 1 ? g = g ? 1 (by Theorem 2.3.4 (c), applied

to 1 and g instead of f and g). But Proposition 2.5.2 (applied to g instead of f )
yields that we have the following logical equivalence:
( F = g ? 1) ⇐⇒ ( g = µ ? F ) .
Hence, we have g = µ ? F (since F = g ? 1).
Comparing f = µ ? F with g = µ ? F, we obtain f = g. This proves Corollary
2.10.7.
Now, proving Theorem 2.10.3 and Theorem 2.10.4 is a child’s play:
Proof of Theorem 2.10.3. (c) Let f and g be two arithmetic functions. Then, Theo-
rem 2.8.2 (c) (applied to 1 ? f and 1 ? g instead of f and g) yields (1 ? f ) · (1 ? g) =
(1 ? g) · (1 ? f ). But Corollary 2.10.6 (applied to g and f instead of f and g) yields
? f = µ ? ((1 ? g) · (1 ? f )) .
ge (36)
On the other hand, Corollary 2.10.6 yields
? g = µ ? ((1 ? f ) · (1 ? g)) = µ ? ((1 ? g) · (1 ? f )) = g e
fe ?f (by (36)) .
| {z }
=(1? g)·(1? f )

This proves Theorem 2.10.3 (c).

(a) Let f be an arithmetic function. Theorem 2.3.4 (a) (applied to 1 instead of
f ) yields ε ? 1 = 1 ? ε = 1. Theorem 2.8.2 (a) (applied to 1 ? f instead of f ) yields
1 · (1 ? f ) = (1 ? f ) · 1 = 1 ? f .
Now, Theorem 2.10.5 (applied to g = ε) yields
? ε ) = (1 ? f ) · (1 ? ε ) = (1 ? f ) · 1 = 1 ? f .
1 ? (f e
| {z }
=1

? ε and f instead of f and g) yields f e

Hence, Corollary 2.10.7 (applied to f e ?ε = f.
But Theorem 2.10.3 (c) (applied to g = ε) yields f e ?ε = εe? f . Hence, ε e?f =
? ε = f . This proves Theorem 2.10.3 (a).
fe
(b) Let f , g and h be three arithmetic functions. Theorem 2.10.5 (applied to
? h instead of g) yields
ge
1 ? (f e ? h)) = (1 ? f ) ·
? (g e (1 ? ( g e
? h)) = (1 ? f ) · ((1 ? g) · (1 ? h))
| {z }
=(1? g)·(1?h)
(by Theorem 2.10.5,
applied to g and h instead of f and g)
= ((1 ? f ) · (1 ? g)) · (1 ? h)
(by Theorem 2.8.2 (b) (applied to 1 ? f , 1 ? g and 1 ? h instead of f , g and h)). On
the other hand, Theorem 2.10.5 (applied to f e ? g and h instead of f and g) yields
1 ? (( f e
? g) e
? h) = ? g)) · (1 ? h) = ((1 ? f ) · (1 ? g)) · (1 ? h) .
(1 ? ( f e
| {z }
=(1? f )·(1? g)
(by Theorem 2.10.5)

66
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Comparing these two equalities, we obtain 1 ? ( f e ? (g e

? h)) = 1 ? (( f e
? g) e
? h ).
Hence, Corollary 2.10.7 (applied to f e ? (g e
? h) and ( f e
? g) e
? h instead of f and g)
? (g e
yields f e ? h) = ( f e
? g) e
? h. This proves Theorem 2.10.3 (b).
Proof of Theorem 2.10.4. The function 1 is multiplicative (by Theorem 2.2.2 (c)).
Also, the function µ is multiplicative (by Theorem 2.2.2 (f)). Now, the functions
1 and f are multiplicative. Hence, Theorem 2.6.1 (applied to 1 and f instead of
f and g) yields that the arithmetic function 1 ? f is also multiplicative. The same
argument (applied to g instead of f ) yields that the arithmetic function 1 ? g is
also multiplicative. Hence, Theorem 2.8.3 (applied to 1 ? f and 1 ? g instead of f
and g) yields that the arithmetic function (1 ? f ) · (1 ? g) is also multiplicative.
Now, the arithmetic functions µ and (1 ? f ) · (1 ? g) are multiplicative. Hence,
Theorem 2.6.1 (applied to µ and (1 ? f ) · (1 ? g) instead of f and g) yields that
the arithmetic function µ ? ((1 ? f ) · (1 ? g)) is also multiplicative.
But Corollary 2.10.6 yields that f e ? g = µ ? ((1 ? f ) · (1 ? g)). Hence, the arith-
metic function f e ? g is multiplicative (since we know that the arithmetic function
µ ? ((1 ? f ) · (1 ? g)) is multiplicative). This proves Theorem 2.10.4.

3. Appendix: a proof of Bézout’s identity

Let me finally give a proof of Theorem 1.2.2, which was used several times
above. Proofs of this theorem abound in the literature; yet I have never seen the
following proof written up. I believe that this proof has the advantage of being
constructive (unlike the proof in [NiZuMo91, proof of Theorem 1.3], which starts
out by choosing the least positive integer in a potentially infinite set) and yet not
too messy (unlike some proofs using the extended Euclidean algorithm). Of
course, all the standard proofs of Theorem 1.2.2 are “essentially the same”, in
the sense that they offer different points of view on one and the same idea (viz.,
that of the Euclidean algorithm).
We first prepare for our proof by showing some simple lemmas:

Lemma 3.0.1. Let b and c be two integers. Then, gcd (b, c) = gcd (c, b).

Proof of Lemma 3.0.1. If (b, c) = (0, 0), then Lemma 3.0.1 is obvious. Hence, for
the rest of this proof, we WLOG assume that (b, c) 6= (0, 0). Thus, (c, b) 6=
(0, 0). Hence, gcd (c, b) is the greatest of all common divisors of c and b (by the
definition of gcd (c, b)). In other words, gcd (c, b) is the greatest of all common
divisors of b and c (since the common divisors of c and b are the same as the
common divisors of b and c). On the other hand, gcd (b, c) is the greatest of
all common divisors of b and c (by the definition of gcd (b, c)). Hence, the two
numbers gcd (c, b) and gcd (b, c) have been characterized in precisely the same
way (namely, as the greatest of all common divisors of b and c). Therefore, these
two numbers are equal. In other words, gcd (b, c) = gcd (c, b). This proves
Lemma 3.0.1.

67
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Lemma 3.0.2. Let b and c be two integers. Then:

(a) We have gcd (b, c) = gcd (−b, c).
(b) We have gcd (b, c) = gcd (b, −c).
(c) We have gcd (b, c) = gcd (|b| , |c|).

Proof of Lemma 3.0.2. If (b, c) = (0, 0), then Lemma 3.0.2 is obvious. Hence, for
the rest of this proof, we WLOG assume that (b, c) 6= (0, 0).
(a) We make the following two observations:

Observation 1: Every common divisor of b and c is a common divisor

of −b and c.

Proof of Observation 1. Let d be a common divisor of b and c. We must prove that

d is a common divisor of −b and c.
We know that d is a common divisor of b and c; hence, d | b and d | c. Now,
d | b | (−1) b = −b. So we know that d divides the two integers −b and c (since
d | −b and d | c). Hence, d is a common divisor of −b and c. This completes the
proof of Observation 1.

Observation 2: Every common divisor of −b and c is a common divisor

of b and c.

Proof of Observation 2. Observation 1 (applied to −b instead of b) shows that ev-

ery common divisor of −b and c is a common divisor of − (−b) and c. In other
words, every common divisor of −b and c is a common divisor of b and c (since
− (−b) = b). This proves Observation 2.
Combining Observation 1 with Observation 2, we conclude that the common
divisors of b and c are the same as the common divisors of −b and c.
Now, (−b, c) 6= (0, 0) (since (b, c) 6= (0, 0)). Hence, gcd (−b, c) is the greatest
of all common divisors of −b and c (by the definition of gcd (−b, c)). In other
words, gcd (−b, c) is the greatest of all common divisors of b and c (since the
common divisors of b and c are the same as the common divisors of −b and c).
On the other hand, gcd (b, c) is the greatest of all common divisors of b and c (by
the definition of gcd (b, c)). Hence, the two numbers gcd (−b, c) and gcd (b, c)
have been characterized in precisely the same way (namely, as the greatest of all
common divisors of b and c). Therefore, these two numbers are equal. In other
words, gcd (−b, c) = gcd (b, c). This proves Lemma 3.0.2 (a).
(b) Lemma 3.0.2 (b) can be proven in the same way as Lemma 3.0.2 (a) (but
now we must use the fact that the divisors of c are the same as the divisors of
−c).

68
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

[An alternative proof of Lemma 3.0.2 (b) proceeds as follows: We have

gcd (b, c) = gcd (c, b) (by Lemma 3.0.1)

 
by Lemma 3.0.2 (a), applied to
= gcd (−c, b)
c and b instead of b and c
 
by Lemma 3.0.1, applied to
= gcd (b, −c) ;
−c and b instead of b and c
thus, Lemma 3.0.2 (b) is proven.]
(c) Lemma 3.0.2 (c) can be proven in the same way as Lemma 3.0.2 (a) (but
now we must use the fact that the divisors of b are the same as the divisors of
|b|, and that the divisors of c are the same as the divisors of |c|).
[Here is an alternative proof of Lemma 3.0.2 (c): Using Lemma 3.0.2 (a), we can
find that gcd (|b| , c) = gcd (b, c) 27 . Using Lemma 3.0.2 (b), we can find that
gcd (|b| , |c|) = gcd (|b| , c) 28 . Hence, gcd (|b| , |c|) = gcd (|b| , c) = gcd (b, c).
This proves Lemma 3.0.2 (c).]

Lemma 3.0.3. Let b, c and u be three integers. Then:

(a) We have gcd (b, c) = gcd (b + uc, c).
(b) We have gcd (b, c) = gcd (b, ub + c).

Proof of Lemma 3.0.3. If (b, c) = (0, 0), then Lemma 3.0.3 is obvious (because if
(b, c) = (0, 0), then all four integers b, c, b + uc and ub + c are 0). Hence, for the
rest of this proof, we WLOG assume that (b, c) 6= (0, 0).
(a) We make the following two observations:

Observation 1: Every common divisor of b + uc and c is a common

divisor of b and c.

Proof of Observation 1. Let d be a common divisor of b + uc and c. We must prove

that d is a common divisor of b and c.
We know that d is a common divisor of b + uc and c; hence, d | b + uc and
d | c. Now, d | c | −uc. So we know that d divides the two integers b + uc and
27 Proof.We must prove that gcd (|b| , c) = gcd (b, c). If |b| = b, then this is obvious. Hence, for
the rest of this proof, we WLOG assume that |b| 6= b.  

Clearly, |b| is either b or −b. Thus, |b| = −b (since |b| 6= b). Hence, gcd  |b| , c =
 
|{z}
=−b
gcd (−b, c) = gcd (b, c) (by Lemma 3.0.2 (a)), qed.
28 Proof. We must prove that gcd (| b | , | c |) = gcd (| b | , c ). If | c | = c, then this is obvious. Hence,

for the rest of this proof, we WLOG assume that |c| 6= c.

Clearly, |c| is either c or −c. Thus, |c| = −c (since |c| 6= c).
But Lemma 3.0.2  (b) (applied
 to |b| instead of b) yields gcd (|b| , c) = gcd (|b| , −c). Com-

pared with gcd |b| , |c|  = gcd (|b| , −c), this yields gcd (|b| , |c|) = gcd (|b| , c), qed.
 
|{z}
=−c

69
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

−uc (since d | b + uc and d | −uc). Hence, d must also divide the sum of these
two integers. In other words, we have d | (b + uc) + (−uc) = b. Now, d divides
both b and c (since d | b and d | c). Hence, d is a common divisor of b and c. This
completes the proof of Observation 1.
Observation 2: Every common divisor of b and c is a common divisor
of b + uc and c.
Proof of Observation 2. Let d be a common divisor of b and c. We must prove that
d is a common divisor of b + uc and c.
We know that d is a common divisor of b and c; hence, d | b and d | c. Now,
d | c | uc. So we know that d divides the two integers b and uc (since d | b
and d | uc). Hence, d must also divide the sum of these two integers. In other
words, we have d | b + uc. Now, d divides both b + uc and c (since d | b + uc and
d | c). Hence, d is a common divisor of b + uc and c. This completes the proof of
Observation 2.
Combining Observation 1 with Observation 2, we conclude that the common
divisors of b + uc and c are the same as the common divisors of b and c.
But (b + uc, c) 6= (0, 0) 29 . Hence, gcd (b + uc, c) is the greatest of all com-
mon divisors of b + uc and c (by the definition of gcd (b + uc, c)). In other words,
gcd (b + uc, c) is the greatest of all common divisors of b and c (since the com-
mon divisors of b + uc and c are the same as the common divisors of b and c).
On the other hand, gcd (b, c) is the greatest of all common divisors of b and c (by
the definition of gcd (b, c)). Hence, the two numbers gcd (b + uc, c) and gcd (b, c)
have been characterized in precisely the same way (namely, as the greatest of all
common divisors of b and c). Therefore, these two numbers are equal. In other
words, gcd (b + uc, c) = gcd (b, c). This proves Lemma 3.0.3 (a).
(b) One way to prove Lemma 3.0.3 (b) is by arguing similarly to how we
argued in our proof of Lemma 3.0.3 (a). Let us, however, proceed differently:
We have
gcd (b, c) = gcd (c, b) (by Lemma 3.0.1)
 
 
by Lemma 3.0.3 (a), applied to c and b
= gcd c| +

{zub}, b

instead of b and c
=ub+c
= gcd (ub + c, b)
 
by Lemma 3.0.1, applied to
= gcd (b, ub + c) .
ub + c and b instead of b and c
Thus, Lemma 3.0.3 (b) is proven.
29 Proof.
Assume the contrary (for the sake of contradiction). Thus, (b + uc, c) = (0, 0). Hence,
b + uc = 0 and c = 0. Now, 0 = b + u |{z} c = b, so that b = 0. Combined with c = 0,
=0
this yields (b, c) = (0, 0), which contradicts (b, c) 6= (0, 0). This contradiction shows that our
assumption was false, qed.

70
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Lemma 3.0.4. Let a ∈ N.

(a) We have gcd ( a, 0) = a.
(b) We have gcd (0, a) = a.

Proof of Lemma 3.0.4. (a) We have gcd (0, 0) = 0. In other words, Lemma 3.0.4 (a)
holds for a = 0. Thus, for the rest of the proof of Lemma 3.0.4 (a), we can WLOG
assume that a 6= 0. Assume this. Thus, a is a positive integer (since a ∈ N and
a 6= 0). Hence, every divisor of a is ≤ a. Thus, the greatest of all divisors of a is
a itself (since a itself is a divisor of a).
We make the following two observations:

Observation 1: Every divisor of a is a common divisor of a and 0.

Proof of Observation 1. Let d be a divisor of a. We must prove that d is a common

divisor of a and 0.
We have d | a (since d is a divisor of a) and d | 0 (obviously). Thus, d divides
both a and 0. Hence, d is a common divisor of a and 0. This completes the proof
of Observation 1.

Observation 2: Every common divisor of a and 0 is a divisor of a.

Proof of Observation 2. Observation 2 is obvious.

Combining Observation 1 with Observation 2, we see that the common divi-
sors of a and 0 are the same as the divisors of a.
We have ( a, 0) 6= (0, 0) (since a 6= 0). Thus, gcd ( a, 0) is the greatest of all
common divisors of a and 0 (by the definition of gcd ( a, 0)). In other words,
gcd ( a, 0) is the greatest of all divisors of a (since the common divisors of a and
0 are the same as the divisors of a). In other words, gcd ( a, 0) is a (since the
greatest of all divisors of a is a). This proves Lemma 3.0.4 (a).
(b) We could prove Lemma 3.0.4 (b) similarly how we proved Lemma 3.0.4
(a). But we can just as easily derive Lemma 3.0.4 (b) from Lemma 3.0.4 (a): We
have
 
by Lemma 3.0.1, applied to 0 and a
gcd (0, a) = gcd ( a, 0)
instead of b and c
=a (by Lemma 3.0.4 (a)) .

This proves Lemma 3.0.4 (b).

Now, we prove the (trivial) particular case of Theorem 1.2.2 when b and c are
nonnegative integers one of which is 0:

Lemma 3.0.5. Let b ∈ N and c ∈ N be such that either b = 0 or c = 0 (or

both). Then, there exist integers x and y such that gcd (b, c) = bx + cy.

71
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Proof of Lemma 3.0.5. We have either b = 0 or c = 0. Thus, we are in one of the

following two cases:
Case 1: We have b = 0.
Case 2: We have c = 0.  

Let us first consider Case 1. In this case, we have b = 0. Thus, gcd |{z}
b , c =
=0
gcd (0, c) = c (by Lemma 3.0.4 (b), applied to a = c). Compared with b0 + c1 =
c1 = c, this yields gcd (b, c) = b0 + c1. Hence, there exist integers x and y such
that gcd (b, c) = bx + cy (namely, x = 0 and y = 1). Thus, Lemma 3.0.5 is proven
in Case 1.  

Let us now consider Case 2. In this case, we have c = 0. Thus, gcd b, |{z}
c =
=0
gcd (b, 0) = b (by Lemma 3.0.4 (a), applied to a = b). Compared with b1 + c0 =
b1 = b, this yields gcd (b, c) = b1 + c0. Hence, there exist integers x and y such
that gcd (b, c) = bx + cy (namely, x = 1 and y = 0). Thus, Lemma 3.0.5 is proven
in Case 2.
Hence, Lemma 3.0.5 is proven in each of the two Cases 1 and 2. Thus, Lemma
3.0.5 always holds.
Next, we prove the particular case of Theorem 1.2.2 when b and c are nonneg-
ative:

Lemma 3.0.6. Let b ∈ N and c ∈ N. Then, there exist integers x and y such
that gcd (b, c) = bx + cy.

Proof of Lemma 3.0.6. We shall prove Lemma 3.0.6 by strong induction on b + c:

Let N ∈ N. Assume that Lemma 3.0.6 holds in the case when b + c < N. We
must prove that Lemma 3.0.6 holds in the case when b + c = N.
We have assumed that Lemma 3.0.6 holds in the case when b + c < N. In
other words, the following holds:

Observation 1: If b ∈ N and c ∈ N satisfy b + c < N, then there exist

integers x and y such that gcd (b, c) = bx + cy.

Let now b ∈ N and c ∈ N be such that b + c = N. We are going to show that

there exist integers x and y such that gcd (b, c) = bx + cy. (37)

If we have either b = 0 or c = 0 (or both), then (37) is true (by Lemma 3.0.5).
Thus, for the rest of this proof of (37), we can WLOG assume that we have
neither b = 0 nor c = 0. Assume this.
We have neither b = 0 nor c = 0. In other words, we have b 6= 0 and c 6= 0.
Thus, b > 0 (since b ∈ N and b 6= 0) and c > 0 (since c ∈ N and c 6= 0). Now,
we are in one of the following two cases:

72
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Case 1: We have b < c.

Case 2: We have b ≥ c.
Let us first consider Case 1. In this case, we have b < c. Thus, b ≤ c, so
that c − b ∈ N. Moreover, c = (c − b) + |{z}b > c − b, so that c − b < c and
>0
thus b + (c − b) < b + c = N. Therefore, we can apply Observation 1 to c − b
| {z }
<c
instead of c. As a result, we conclude that there exist integers x and y such that
gcd (b, c − b) = bx + (c − b) y. Denote these x and y by x0 and y0 . Hence, x0 and
y0 are integers satisfying gcd (b, c − b) = bx0 + (c − b) y0 .
Lemma 3.0.3 (b) (applied to u = −1) yields
 

gcd (b, c) = gcd b, (−1) b + c = gcd (b, c − b) = bx0 + (c − b) y0

 
| {z }
=c−b
= bx0 + cy0 − by0 = b ( x0 − y0 ) + cy0 .

Thus, there exist integers x and y such that gcd (b, c) = bx + cy (namely, x =
x0 − y0 and y = y0 ). Therefore, (37) is proven in Case 1.
Let us now consider Case 2. In this case, we have b ≥ c. Thus, b − c ∈ N.
Moreover, b = (b − c) + |{z}c > b − c, so that b − c < b and thus (b − c) +c <
| {z }
>0 <b
b + c = N. Therefore, we can apply Observation 1 to b − c instead of b. As a
result, we conclude that there exist integers x and y such that gcd (b − c, c) =
(b − c) x + cy. Denote these x and y by x0 and y0 . Hence, x0 and y0 are integers
satisfying gcd (b − c, c) = (b − c) x0 + cy0 .
Lemma 3.0.3 (a) (applied to u = −1) yields
 

gcd (b, c) = gcd b + (−1) c, c = gcd (b − c, c) = (b − c) x0 + cy0

 
| {z }
=b−c
= bx0 − cx0 + cy0 = bx0 + c (y0 − x0 ) .

Thus, there exist integers x and y such that gcd (b, c) = bx + cy (namely, x = x0
and y = y0 − x0 ). Therefore, (37) is proven in Case 2.
We have now proven (37) in each of the two Cases 1 and 2. Thus, (37) always
holds (since Cases 1 and 2 cover all possibilities). So we have proven that there
exist integers x and y such that gcd (b, c) = bx + cy.
Now, forget that we fixed b and c. We thus have shown that if b ∈ N and
c ∈ N are such that b + c = N, then there exist integers x and y such that
gcd (b, c) = bx + cy. In other words, Lemma 3.0.6 holds in the case when b + c =
N. This completes the induction step; thus, Lemma 3.0.6 is proven by strong
induction.

73
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

Now, we can finally deliver the coup-de-grâce to Theorem 1.2.2:

Proof of Theorem 1.2.2. We have |b| ∈ N and |c| ∈ N. Hence, Lemma 3.0.6 (ap-
plied to |b| and |c| instead of b and c) yields that there exist integers x and y
such that gcd (|b| , |c|) = |b| x + |c| y. Denote these x and y by x0 and y0 . Hence,
x0 and y0 are integers satisfying gcd (|b| , |c|) = |b| x0 + |c| y0 .
But |b| is either b or −b. In either case, |b| is divisible by b (since both b and
−b are divisible by b). Hence, there exists a β ∈ Z such that |b| = βb. Similarly,
there exists a γ ∈ Z such that |c| = γc. Consider these β and γ. Now, Lemma
3.0.2 (c) yields

gcd (b, c) = gcd (|b| , |c|) = |b| x0 + |c| y0 = βbx0 + γcy0 = b ( βx0 ) + c (γy0 ) .
|{z} |{z} |{z} |{z}
= βb =γc =b( βx0 ) =c(γy0 )

Hence, there exist integers x and y such that gcd (b, c) = bx + cy (namely, x =
βx0 and y = γy0 ). This proves Theorem 1.2.2.

References
[BenGol75] Edward A. Bender, J. R. Goldman, On the applications of Möbius in-
version in combinatorial analysis, American Mathematical Monthly, vol.
82, October 1975, pp. 789–803.

[Conrad19] Keith Conrad, Divisibility without Bezout’s identity.

https://kconrad.math.uconn.edu/blurbs/ugradnumthy/
divnobezout.pdf

[Grinbe15] Darij Grinberg, Mathematical Reflections problem U275, version of May

11, 2018.
http://www.cip.ifi.lmu.de/~grinberg/mrmoeb.pdf

[Lehmer31] D. H. Lehmer, On a theorem of von Sterneck, Bull. Amer. Math. Soc.

37 (1931), pp. 723–726.

[NiZuMo91] Ivan Niven, Herbert S. Zuckerman, Hugh L. Montgomery, An In-

troduction to the Theory of Numbers, 5th edition, Wiley 1991.

[Rota64] Gian-Carlo Rota, On the Foundations of Combinatorial Theory: I. The-

ory of Möbius Functions, Z. Wahrscheinlichkeitstheorie 2, pp. 340–368
(1964).

[Stanle11] Richard P. Stanley, Enumerative Combinatorics, volume 1, Cambridge

University Press, 2011.
http://math.mit.edu/~rstan/ec/ec1/

74
18.781 (Spring 2016): floor and arithmetic functions January 22, 2021

[Toth14] László Tóth, Multiplicative Arithmetic Functions of Several Variables: A

Survey, arXiv:1310.7053v2. Published in: Mathematics Without Bound-
aries, Surveys in Pure Mathematics, T. M. Rassias, P. M. Pardalos
(eds.), Springer, 2014, pp. 483–514.

75