1

2 Journal of Integer Sequences, Vol. 15 (2012),

3
47 6
Article 12.3.8
23 11

The Arithmetic Derivative and Antiderivative

Jurij Kovič
Institute of Mathematics, Physics, and Mechanics
University of Ljubljana, Slovenia
amadea.kovic@siol.net

Abstract
The notion of the arithmetic derivative, a function sending each prime to 1 and
satisfying the Leibnitz rule, is extended to the case of complex numbers with ratio-
nal real and imaginary parts. Some constraints on the solutions to some arithmetic
differential equations are found. The homogeneous arithmetic differential equation of
the k-th order is studied. The factorization structure of the antiderivatives of natural
numbers is presented. Arithmetic partial derivatives are defined and some arithmetic
partial differential equations are solved.

1 Introduction: Basic concepts

The goal of this paper is: (i) to review what is known about the arithmetic derivative [1, 2, 3],
(ii) to define some related new concepts, and (iii) to prove some new results, mostly related
to the conjectures formulated in [1, 3].

1.1 Definition of the arithmetic derivative

Barbeau [1] defined the arithmetic derivative as the function D : Z → Z, defined by the
rules:
D(1) = D(0) = 0
D(p) = 1 for any prime p ∈ P := {2, 3, 5, 7, . . . , pi , . . .}.
D(ab) = D(a)b + aD(b) for any a, b ∈ N (the Leibnitz rule)
D(−n) = −D(n) for any n ∈ N.
Ufnarovski and Åhlander [3] extended D to rational numbers by the rule:

a a a′ b − ab′
D( ) = ( )′ =
b b b2

1
 Qk
They also defined D for real numbers of the form x = i=1 pixi , where pi are different
primes and xi ∈ Q, by the additional rule:
k
X xi
D(x) = x′ = x .
i=1
pi

They showed that the definition of D can be extended to all real numbers and to the
arbitrary unique factorization domain.
I extend the definition of the arithmetic derivative to the numbers from the unit circle in
the complex plane E = {eiϕ , ϕ ∈ [0, 2π]} and to the Gaussian integers Z[i] = {a+bi, a, b ∈ Z}.
It is not possible to extend the definition of the arithmetic derivative to the Gaussian
integers by the demand D(q) = q ′ = 0 for all irreducible Gaussian integers q (the analog of
primes). If one wants the Leibnitz rule to still remain valid, one easily gets a contradiction:
2 = (1 + i)(1 − i), hence D(2) = 1 · (1 − i) + (1 + i) · 1 = 2, but D(2) = 1 because 2 is a
prime.
However, one can extend D to Q[i] = {a + bi, a, b ∈ Q} using the polar decomposition of
complex numbers as follows:

Proposition 1. i) For any ϕ = (m/n)π, where m, n ∈ Q and n 6= 0, the equation D(eiϕ ) = 0

holds.
ii) Let E = {z ∈ C; |z| = 1} be the unit circle in the complex plane. There are uncountably
many functions D : E → E satisfying the condition D(zw) = D(z)w+zD(w) for all z, w ∈ E.
iii) If one defines the arithmetic derivative on E as follows: D(eiϕ ) = 0 for all eiϕ ∈ E,
then the definition of the arithmetic derivative can be uniquely extended to Q[i] so that the
Leibnitz rule and the quotient rule remain valid.

Proof. i) Take any ϕ = (m/n)π, where m, n ∈ Z and n 6= 0. Then einϕ = ±1, hence
D(einϕ ) = 0. The Leibnitz rule implies D(einϕ ) = D(eiϕ )(eiϕ )n−1 and because (eiϕ )n−1 6= 0
it must be that D(eiϕ ) = 0.
ii) Sketch of the proof: One can easily see that there is a 1-1 correspondence between the
functions D : E → C satisfying the Leibnitz rule D(zw) = D(z)W + zD(w) and the additive
functions L : R → C with a period of 2π (the correspondence is given by the formulas
is )
L(s) := D(e eis
and D(eis ) := L(s)eis ).
It is also easy to see that for any additive function (satisfying Cauchy’s functional equation
L(s + t) = L(s) + L(t)) it is: L(qx) = q · L(x) for all q ∈ Q and x ∈ R. Because R and C
are linear spaces over Q, this means that every additive function L : R → R is a Q-linear
transformation. Additive functions from R into C with a period of 2π are therefore in a 1-1
correspondence with Q-linear transformations L : R → C such that L(2π) = 0.
A linear transformation is uniquely determined with its values on any of the bases of its
domain. Let B be a base of the space R such that π ∈ B (its existence is guaranteed by
Zorn’s lemma). Thus the Q-linear transformations L : R → C such that L(2π) = L(π) = 0
are in a 1-1 correspondence with the set of all functions L : B → C such that L(π) = 0.
The base B is obviously not countable, hence the same also holds for the set of all
functions L : B \ {π} → C. By the chain of proven 1-1 correspondences, this also holds for
the set of all functions D : E → E satisfying the Leibnitz rule.

2
 iii) If one writes such numbers in their polar form a+bi = reiϕ , then D(reiϕ ) = D(r)eiϕ +
rD(eiϕ ) = D(r)eiϕ = D((a2 +b2 )1/2 )eiϕ because D(eiϕ ) = 0 and r = (a2 +b2 )1/2 . The Leibnitz
rule is still valid because for any z = reiϕ and w = seiψ it is: D(zw) = D((reiϕ )(seiψ )) =
D(rs)eiϕ+ψ = (D(r)s + r(D(s))eiϕ+ψ = D(r)eiϕ seiψ + reiϕ D(s)eiψ = D(z)w + zD(w). The
D(r)eiϕ seiψ −reiϕ D(s)eiψ (D(r)s−rD(s))eiϕ
quotient rule is also still valid because D(z)w−zD(w)
w 2 = 2
s e i2ϕ = s2 eiψ
=
r i(ϕ−ψ) r i(ϕ−ψ) reiϕ z
D( s )e = D( s e ) = D( seiψ ) = D( w ).
Stay [2] generalized the concept of the arithmetic derivative still further, practically for
any number, using advanced techniques such as exponential quantum calculus. In this paper
I focus on the arithmetic derivative as a function defined for natural numbers and rational
numbers.

1.2 Higher derivatives and the logarithmic derivative

Higher derivatives n(k) are defined inductively: n(2) = D2 (n) = D(D(n)) = n′′ = (n′ )′ ,
n(k+1) = Dk+1 (n) = D(Dk (n)). Many conjectures on the arithmetic derivative focus on the
behavior of sequences (n, n′ , n′′ , . . .).
Ufnarovski and Åhlander [3] conjectured that for each n ∈ N exactly one of the following
can happen: either n(k) = 0 for sufficiently large k, or limk→∞ = ∞, or n = pp for some
prime p (in this case n(k) = n for each k ∈ N). They introduced the function L, called the
logarithmic derivative, satisfying the condition

x′ D(x)
L(x) = = .
x x
Qk
For any x = i=1 pixi , where pi are different primes and xi ∈ Q, L satisfies the condition
k
X xi
L(x) = .
i=1
pi
m
m
For every prime p and every m, n ∈ N the following formulas hold: L(p) = p1 , L(p n ) = np ,
L(1) = 0. From the definition of the logarithmic derivative also follow the formulas L(−x) =
L(x), L(0) = ∞ and D(x) = L(x) · x [3], hence D2 (x) = D(L(x))x + L(x)D(x). I also use
the notation L(x) = x∗ .
The logarithmic derivative is an additive function: L(xy) = L(x) + L(y) for any x, y ∈ Q.
Consequently, using a table of values L(p) = p1 (computed to sufficient decimal places!) and
the formula D(x) = L(x) · x, it is easy to find D(n) for n ∈ N having all its prime factors in
the table. For example, D(5·113 ) = L(5·113 )·5·113 and because L(5·113 ) = L(5)+3L(11) =
0.2000 + 3 · 0.0909 = 0.4727, one can calculate D(5 · 113 ) = ⌈0.4727 · 6655⌉ = ⌈3145.8185⌉ =
3146.

PN 2 3 5 7 11 13 17
1
PN
0.5000 0.3333 0.2000 0.1429 0.090 0.0769 0.0588

Table 1: Values of the logarithmic derivative for the first seven primes

3
2 Brief review of known results and conjectures
√
Barbeau [1] proved that if n is not a prime or unity then D(n) ≥ 2 n with equality only if
n = p2 , p ∈ P. He showed that, for integers possessing a proper divisor of the form pp , p ∈ P,
limk→∞ Dk (n) = ∞.
Ufnarovski and Åhlander [3] translated some famous conjectures in number theory (e.g.,
the Goldbach conjecture, the Prime twins conjecture) into conjectures about the arithmetic
derivative. They formulated many other conjectures, mostly related to (arithmetic) differ-
ential equations; for example that the equation x′ = 1 has only primes as positive rational
solutions (but it has a negative rational solution x = − 45 ).
They also conjectured that the equation n′′ = n has no other solutions than n = pp , p ∈ P
in natural numbers and that there are some rational numbers without antiderivatives (or
”integrals”). I study these conjectures in Section 3.

2.1 Arithmetic differential equations and integrals

Most of the known results by Ufnarovski and Åhlander [3] focus on arithmetic differential
equations of the first and second order. For example:
The only solutions to the equation n′ = n in natural numbers are n = pp , where p is any
prime.
The nonzero solutions to x′ = 0 are the rational numbers of the form: x = ± pαi i pi ,
Q
where p1 , . . . , pk are different primes and {α1 , . . . , αn } is a set of integers such that Σki=1 αi = 0.
The rational solutions to differential equations x′ = xα for all rational numbers α = a/b,
where gcd(a, b) = 1 and b > 0, are all of the form x = x0 y, where x0 is a nonzero particular
solution and y is any rational solution of the equation y ′ = 0.
Let I(a) denote all the solutions of differential equation n′ = a for n ∈ N√ and let i(a)
denote the number of such solutions, called the integrals of n. Because n′ ≥ 2 n if n is not
2
a prime or unity [1], the solutions satisfy n ≤ a4 , hence i(a) < ∞ for any a ∈ N. Because
gcd(n, n′ ) = 1 if and only if n is square-free, all integrals of primes are products of different
primes: p1 · · · pk . It will be seen (Corollary 21) that there are primes without integrals (e.g.,
primes 2, 3, 17). If D(n) were known for each n ∈ N, one would know which natural numbers
are primes (because the equation n′ = 1 has only primes as solutions in natural numbers).
Ufnarovski and Åhlander [3] gave a list of all a ≤ 1000 having no integral, a list of those
numbers a ≤ 100 having more than one integral, and a list of those a ≤ 100 for which
i(a) = 1. It will be seen (Corollary 25) that something can also be said about the possible
factorization structure of the integrals of a given natural number.

3 New results
Ufnarovski and Åhlander [3] solved the equation x′ = αx in rational numbers for every
rational number α. Nonetheless it is interesting to know which natural numbers solve this
equation when α = m is a natural number.
Proposition 2. Let m ∈ N. The solutions to the equation x′ = mx in natural numbers x
are exactly the numbers of the form x = pp11 n1 · · · pkpk nk , where n1 + · · · + nk = m.

4
Proof. Let x = pe11 · · · pekk be the factorization of x. The condition x′ = L(x)x = mx implies
L(x)(p1 · · · pk ) = ( pe11 + · · · + pekk )(p1 · · · pk ) = m(p1 · · · pk ). Because the right side of this
equation is divisible by any of the primes p1 , . . . , pk , the left side must also be. This is
possible only if for every prime pi the corresponding exponent ei is the multiple of this prime
ei = ni pi . The derivative of such a number x = pp11 n1 · · · pkpk nk is D(x) = D(pp11 n1 · · · pkpk nk ) =
(n1 + · · · + nk )x (by the Leibnitz rule and because D(ppi ni ) = ni ppi ni ) and this is equal to
mx if and only if n1 + · · · + nk = m.

3.1 The homogeneous differential equation of the k ′ th order

What are the solutions x of the differential equation
ak x(k) + ak−1 x(k−1) + · · · + a2 x(2) + a1 x′ + a0 x = 0
with rational coefficients ai ? In order to answer this question I introduce the concept of a
logarithmic class.
Definition 3. Let A be any chosen subring of the ring of complex numbers for which the
arithmetic derivative is defined. The logarithmic class Nr,A of the number r consists of all
numbers x ∈ A with the same logarithmic derivative: Nr,A = {x ∈ A; x∗ = L(x) = r}.
Remark 4. If it is made perfectly clear which A is being worked with the shorter notation
Nr can be used. For the purpose of this article, let A be the set of all complex numbers with
rational real and imaginary parts: A = Q[i] = {a + bi, a, b ∈ Q}.
Example 5. Besides the number 0 and the complex numbers z = eiϕ ∈ A onQ the unit circle
in the Gaussian plane the class N0 also contains rational solutions x = ± pαi i pi , where
p1 , . . . , pk are different primes and {α1 , . . . , αn } is a set of integers such that Σki=1 αi = 0; for
4
instance, x = 27 [3]. All x ∈ N0 solve the equation x′ = 0.
The class N1,Q contains all numbers pp , where p is any prime because D(pp ) = pp . The
class N−1,Q contains all numbers p−p , where p is any prime because D(p−p ) = −p−p . Do N1,Q
and N−1,Q also contain other rational solutions? Yes:
Proposition 6. If ( ab )′ = ± ab ∈ Q and if a and b have no common factors greater than 1,
then a = pp11 n1 · · · pkpk nk , b = q1q1 s1 · · · qkqk ns , and a∗ −b∗ = ±1, where ni , mj , k, l are nonnegative
integers and a∗ , b∗ logarithmic derivatives. All such numbers are solutions because: ( ab )′ =
( ab )∗ ( ab ) = ±( ab ).
′ ′
Proof. If ( ab )′ = ± a b−b b2
a
= ± ab then a′ b − b′ a = ±ab, hence gcd(a, b) = 1 implies a′ = ma
and b = nb, where m = a∗ and n = b∗ are natural numbers. Hence (by Proposition 2)
′

a = pp11 n1 · · · pkpk nk , where n1 + · · · + nk = m, and b = q1q1 s1 · · · qkqk ns , where s1 + · · · + sk = n.

Because ( ab )′ = ± ab implies ( ab )∗ = ±1, it must be a∗ − b∗ = ±1.
Proposition 7. i) The derivative D sends logarithmic classes into logarithmic classes as
follows: D(Nr ) ⊆ Nr∗ +r . Consequently D(Nr ) ⊆ Nr if and only if r∗ = 0 and therefore
r′ = 0.
ii) If r∗ + r = 0 and r 6= 0 then D(Nr ) = N0 − {0} 6= N0 . So in this case D(Nr ) ⊇ Nr∗ +r
is not true.
iii) If r∗ + r 6= 0 then D(Nr ) = Nr∗ +r .

5
Proof. If r = 0 then Nr = N0 = Nr∗ +r . So let us now assume that r 6= 0.
i) D(Nr ) ⊆ Nr∗ +r is true because x ∈ Nr implies x′ = rx, hence (x′ )′ = r′ x + rx′ =
′
rx ′ ′
r
+ rx′ = ( rr + r)x′ = (r∗ + r)x′ .
ii) It is possible indeed that r∗ + r = 0 while r 6= 0 (an example is r = 15 , r′ = − 512 , r∗ =
− 51 ). If r∗ + r = 0 then 0 ∈ Nr∗ +r because 0′ = 0. Now suppose there is an x ∈ Nr such that
D(x) = 0. Wowever, this implies x ∈ N0 , hence r = 0 and there is a contradiction with the
assumption r 6= 0. This means that in this case it is not true D(Nr ) ⊇ Nr∗ +r = N0 because
there is no x ∈ Nr such that D(x) = 0.
Now let it be proved that for each nonzero y ∈ Nr∗ +r there is an x ∈ Nr such that
D(x) = y. If there is any such x ∈ Nr , then it must be y = x′ = rx, hence there is at most
one such x and it is defined by the formula x = yr . Now suppose the derivative of this x
is not equal to y. This assumption leads to a contradiction because D( yr ) = D(y)r−yD(r) r2
=
(r ∗ +r)yr−yr ∗ r yr 2
r2
= r2 6= y implies 1 6= 1. Because D(x) = y = rx, this x is indeed a member of
Nr . Thus it has been proved that D(Nr ) ⊇ N0 − {0}.
Because it is already known that i) is true and because it has been shown that there is
no x ∈ Nr such that D(x) = 0, the equation D(Nr ) = N0 − {0} holds. Moreover, because
0 ∈ N0 , it is also true that N0 is a proper subset of N0 − {0}.
iii) If r∗ + r 6= 0 then r 6= 0, and Nr∗ +r contains only nonzero elements because 0 ∈ N0 .
Now it is possible to repeat the reasoning as in ii) and for any y ∈ Nr∗ +r one finds an
x ∈ Nr such that y = D(x). Hence D(Nr ) ⊇ Nr∗ +r and this together with i) implies
D(Nr ) = Nr∗ +r .
Remark 8. It is already known that there are many rational solutions to the equation r′ = 0;
for example, r = 1, −1, 0, hence: D(N1 ) ⊆ N1 , D(N−1 ) ⊆ N−1 , D(N0 ) ⊆ N0 .

Proposition 9. All the derivatives x(k) of any number x can be expressed as functions of
the logarithmic derivative L(x) = x∗ as follows: x(k) = fk (x∗ )x where f1 (x∗ ) = x∗ and
fk+1 (x∗ ) = ((fk (x∗ ))′ + fk (x∗ ))x∗ . Thus x′ = x∗ x, x(2) = ((x∗ )′ + (x∗ )2 )x etc.

Proof. This is true for k = 1. Suppose x(k) = fk (x∗ )x. Then by the Leibnitz rule: x(k+1) =
(fk (x∗ ))′ x + fk (x∗ )x′ = ((fk (x∗ ))′ + fk (x∗ )x∗ )x.

Proposition 10. Any homogeneous differential equation f (x) ≡ ak x(k) + ak−1 x(k−1) + · · · +
a2 x(2) +a1 x′ +a0 x = 0 reduces to an equation g(x∗ ) = 0. Consequently, if the set of nontrivial
solutions to any homogeneous differential equation f (x) = 0 is not empty, then it consists of
some classes Nr .
′
Proof. Let r = x∗ = xx . Because x(k) = fk (x∗ )x, one can divide the differential equation by
x and get an equation of the form g(r) = 0. Thus f (x) = 0 if and only if g(x∗ ) = 0. Hence
whether f (x) = 0 or not depends only on the logarithmic class Nr = Nx∗ . Note also that
the degree of the polynomial g is k − 1, one less than the degree of the polynomial f .

Example 11. The nonzero solutions to the equation x′′ −x = 0 implying ((x∗ )′ +(x∗ )2 )x−x =
0 satisfy the equation ((x∗ )′ + (x∗ )2 − 1 = 0 or (x∗ )′ = 1 − (x∗ )2 . Thus the nonzero solutions
to x′′ − x = 0 exist if and only if the nonhomogeneous equation r′ = 1 − r2 can be solved.

6
Proposition 12. For every r ∈ Q the class Nr is not empty because it contains the numbers
ppr , where p is any prime.
Proof. L(ppr ) = r for any prime p, hence ppr ∈ Nr .
Thus in solving the homogenous differential equation f (x) = 0 one can always search for
the solutions of the form x = ppr . Of course, other solutions are also possible.
Remark 13. Numbers ppr behave in the first derivative just like the exponential function exr
whose derivative is rexr . However, for the higher derivatives the analogy is no longer valid
because (pp r)′′ = (r′ + r2 )ppr , while (exr )′′ = r2 exr . This is one of the reasons why solving an
arithmetic differential equation is harder than solving the analogous problem for functions.
Proposition 14. i) In addition to the trivial solutions x ∈ N0 the equation x′′ = 0 also has
solutions x ∈ N1/p , where p is any prime.
ii) All solutions x ∈ Nr of x′′ = 0 satisfy the condition: r∗ + r = 0 (hence r ∈ N−r and
r′ = r∗ r = −r2 ) and all such x are solutions to x′′ = 0. If r = ab ∈ Q then ab = b∗ −a∗ = ( ab )∗ ,
hence ( ab )′ = 1.
Proof. i) If x′ = p1 then x′′ = − p12 x + p1 p1 x = 0 = (x′ )∗ x′ , hence (x′ )∗ = 0.
ii) Any solutions x ∈ Nr of x′′ = 0 satisfy the condition D2 (Nr ) ⊆ D(Nr∗ +r ) = D0 , hence
r + r = 0. Conversely, r∗ + r = 0 implies D2 (Nr ) ⊆ D(Nr∗ +r ) = D(N0 ) = {0}, hence
∗
′ ′ ′ ′ a ′ ′
x′′ = 0. If r = ab then r′ = a b−abb2
= a b−ab
ab b
, hence ( ab )∗ = a b−ab
ab
= −r = − ab , therefore
a ′ ′ ∗
a
− bb = − ab and ab = b∗ − a∗ = ( ab )∗ . Hence ( ab )′ = ab ab = ab ab = 1.
Example 15. Ufnarovski and Åhlander [3] observed that if x = − 54 then x′ = 1 (hence
x′′ = 0). Then r = x∗ = − 54 , r′ = −4·5−(−4)·1
52
= − 16
25
= 54 −4
5
, hence r∗ = 45 and r∗ + r =
4
5
+ (− 45 ) = 0. Hence all x ∈ N− 4 solve x′′ = 0.
5

3.2 The graph of derivatives of natural numbers

Of interest is the structure of the infinite directed graph GD , whose vertices correspond to
natural numbers n and whose arcs n → D(n) connect the number and its derivative. The
corresponding dynamic system: n → D(n) has two obvious attractors: 0 and ∞. There are
numbers n with an increasing sequence of derivatives n < n′ < n′′ < · · · < n(k) < n(k+1) < · · ·
(e.g., n = ppk , p ∈ P), so there are paths of infinite length in the graph GD .
Ufnarovski and Åhlander [3] conjectured that the equation n(k) = n has only trivial
solutions pp , where p ∈ P, satisfying n′ = n. If this is true then the only cycles in GD are
′ ′
the loops in these fixed points. They Qkhave shown that Qifl m = n and n = m then m and n
must be square-free numbers: n = i=1 pi and m = j=1 qj , where all pi are distinct from
all qi . I present further constraints (Propositions 16, 17, 18, 19) on the structure of possible
solutions of the equation n′′ = n in natural numbers.
However, the equation x′′ = x has non-trivial rational solutions of the form x = p−p
where p ∈ P and they also satisfy the equation x′ = −x.
Let us first show that any eventual nontrivial solutions m, n (different from m = n = pp
where p ∈ P) of the system n′ = m, m′ = n in natural numbers cannot be just a product of
two primes; at least one of the numbers m, n ∈ N solving such a system must have at least
three different prime factors.

7
Proposition 16. The system n′ = m, m′ = n has no solutions in natural numbers of the
form m = p1 p2 and n = q1 q2 , where p1 , p2 , q1 , q2 ∈ P.

Proof. Because p1 , p2 , q1 , q2 must be different primes, it is not possible for both m and n
to have the factor 2. Therefore one can assume that p1 and p2 are both odd. Then m′ =
p1 + p2 = n = 2q2 and n′ = 2 + q2 = m = p1 p2 , hence 2q2 = 2p1 p2 − 4 = p1 + p2 and
p2 +4 5−p2
p1 = 2p 2 −1
= 1 + 2p 2 −1
, and this implies p2 ≤ 5 because p1 must be a natural number.
Moreover, it was assumed that p2 is odd. However, for p2 = 3 one would get p1 = 1 + 52 = 75
and for p2 = 5 one would get p1 = 1.
A computer search showed that there are no natural solutions to x′′ = x such that
x < 10000. This result can be improved, at least if the smaller of the numbers m and n is
odd.

Proposition 17. Let n′ = m, m′ = n and let n = 2j − 1 < m. Then there are at least nine
primes in the factorization of n and n > 3 · 5 · 7 · 11 · 13 · 17 · 19 · 23 · 29.

Proof. It is known that m and n must be products of different primes n = ki=1 pi and
Q

m = lj=1 qj , where all pi are distinct from all qi . Because n < m = n′ = L(n)n we have
Q

L(n) = ki=1 p1i > 1. Because n is odd, all pi are greater than 2. The sum of the reciprocals
P

of the first eight odd primes is: 13 + 51 + 71 + 11 1 1

+ 13 1
+ 17 1
+ 19 1
+ 23 = 0.9987 · · · < 1. Thus
k ≥ 9 and n > 3 · 5 · 7 · 11 · 13 · 17 · 19 · 23 · 29.
If n = 2j < m then the sum 12 + 13 + 15 is already greater than 1 and one does not get any
such estimate.
Another possible approach to the system n′ = m, m′ = n in natural numbers is based on
the comparison of a square-free number and its derivative; this comparison will imply some
inequalities for the smallest primes and biggest primes in the factorizations of n and m.

Proposition 18. Let n = p1 · · · pr , where p1 < · · · < pr , and let m = q1 · · · qs where

q1 < · · · < qs be square-free numbers such that n′ = m, m′ = n. Then: p1 qs < rs ≤
( r+s
2
)2 , q1 pr < rs ≤ ( r+s 2
)2 , p1 < r, q1 < s. As a consequence n and m must together
have at least 34 prime factors pi and qj , thus: r + s ≥ 34. Hence at least one of m and
n has at least 17 prime factors and is not smaller than the product of the first 17 primes:
2 · 3 · 5 · 7 · 11 · 13 · 17 · 19 · 23 · 27 · 31 · 37 · 41 · 43 · 47 · 53 · 59. If min{p1 , q1 } ≥ 3, then
r + s ≥ 57, hence at least one of m and n has at least 29 prime factors. If min{p1 , q1 } ≥ 5,
then r + s ≥ 110, hence at least one of m and n has at least 55 prime factors.

Proof. Let n = p1 · · · pr and m = q1 · · · qs be square-free numbers such that n′ = m, m′ = n.

Then 0 < r pnr < n′ < r pn1 and 0 < s qms < m′ < s qm1 . Hence: p1 qs < rs and q1 pr < rs. Because
r > 1, s > 1 we have pr > r and qs > s, hence p1 < r and q1 < s. Let r + s = N . Then rs ≤
2
( r+s
2
)2 = ( N2 )2 . Thus 2qs ≤ p1 qs < ( N2 )2 and 2pr ≤ q1 pr < ( N2 )2 , hence 2 max{pr , qs } < N4 .
Let Pi denote the i-th prime (thus P1 = 2, P2 = 3, P3 = 5, etc.) Because all the primes
p1 , . . . , pr , q1 , . . . , qs are distinct, we have max{pr , qs } ≥ Pr+s = PN . Thus 2PN = 2Pr+s ≤
2
2 max{pr , qs } < rs ≤ N4 .
2
It is possible to directly check that PN ≥ N8 if N ≤ 33.

8
 N 1 2 3 4 5 6 7 8 9 10 11 12
PN 2 3 5 7 11 13 17 19 23 27 31 37
N2 1 4 9 16 25 36 49 64 81 100 121 144

N 13 14 15 16 17 18 19 20 21 22 23
PN 41 43 47 53 59 61 67 71 73 79 83
N2 169 196 225 256 289 324 361 400 441 484 529

N 24 25 26 27 28 29 30 31 32 33 34
PN 89 97 101 103 107 109 113 127 131 137 139
N2 576 625 676 729 784 841 900 961 1024 1089 1156

Table 2: The first 34 primes pN and squares N 2

Hence it must be that N ≥ 34. If min{p1 , q1 } ≥ 3, one can get a much better estimate
2
from the condition: pN < N12 , which is first fulfilled when N ≥ 57. If min{p1 , q1 } ≥ 5, then
2
pN < N20 , which is first fulfilled when N ≥ 110.

Proposition 19. Suppose n′ = m and n′ = m and n, m are both odd. Let n have r1 primes
pi ≡ 1 (mod 4) and s1 primes qj ≡ −1 (mod 4) and let m have r2 primes pi ≡ 1 (mod 4)
and s2 primes qj ≡ −1 (mod 4). Then (r1 − s1 )(r2 − s2 ) ≡ 1 (mod 4).

Proof. Barbeau [1, pp. 121–122] proved that if n = p1 · p2 . . . pr · q1 · p2 . . . qs , where pi ≡ 1

(mod 4), qj ≡ −1 (mod 4) are primes, not necessarily distinct, then D(n) ≡ (−1)s (r − s)
(mod 4). Hence m = n′ = (−1)s1 (r1 − s1 ) and n = m′ = (−1)s2 (r2 − s2 ). Because m and n
are odd, r1 − s1 and r2 − s2 are not even. However, n ≡ (−1)s1 (mod 4) and m ≡ (−1)s2
(mod 4). Hence mn ≡ (−1)s1 (−1)s2 ≡ (−1)s1 (−1)s2 (r1 − s1 )(r2 − s2 ) (mod 4). Consequently
(r1 − s1 )(r2 − s2 ) ≡ 1 (mod 4).

3.3 Integrals of natural and rational numbers

Any a such that a′ = b is called an integral of b. The set of all such integrals is denoted
as I(b). The same number can have different integrals: 25′ = (52 )′ = 2 · 5 = 10 and
21′ = 3 · 7 = 3 + 7 = 10. Because p′ = 1 for any prime, I(1) = P. It is shown that 1 is the
only natural number with infinitely many integrals among the natural numbers.

Proposition 20. i) Let b < 2 · 3 · 5 · · · Pn , where Pn is the n-th consecutive prime and let
a′ = b, where a ∈ N. Then a = pn1 1 · · · pnmm , where m ≤ n (hence a is divisible by at most n
primes pi ).
ii) If a′ = b > 1, where a ∈ N, then a ≤ max(2 · 3 · 5 · · · Pn , bbn ). Consequently every
b > 1 has at most a finite number of integrals a ∈ N.
Qm ni
Proof. For each natural number b there is a n ∈ N such that b < 2·3·5 · · · PP n . If a = i=1 pi
has more than n different prime factors pi , then each summand of a′ = ( m ni
i=1 pi ) · b has at
least n prime factors, and the smallest of them is not smaller than 2 · 3 · 5 · · · Pn , therefore
in that case a′ > 2 · 3 · 5 · · · Pn > b, so it cannot be a′ = b. Thus a = pn1 1 · · · pnmm and m ≤ n.

9
 ii) If a contains a factor pp , then b = a′ ≥ a, hence a ≤ b < 2 · 3 · 5 · · · Pn . The other
possibility is that all exponents of a = pn1 1 · · · pnmm are smaller than their primes: ni < pi . It
is necessary to consider two cases:
If m = 1 then a = pn1 1 and a′ = n1 p1n1 −1 = nq11a = b. Now b > 1 implies n1 ≥ 2 thus p1
divides b, hence p1 ≤ b and a = bp n1
1
≤ bq1 ≤ b2 .
If m ≥ 2 then a = p1 · · · pnmm < pp11 · · · ppmm < bp1 +···+pm < bbn because pi < a′ = b for each
n1

pi .
Hence a ≤ max(2 · 3 · 5 · · · Pn , bbn ) and I(b) N is finite for any b > 1.
T

Corollary 21. If a′ = p and p is a prime, then a = p1 · · · pm and all pi < p. There are some
primes without integrals a ∈ N; for example 2,3,17.

Proof. If a = p2 c, then a′ = p · (2 + pc′ ) is not a prime. If a = p1 · · · pm , then p = a′ > pi .

Because 2 < 2 · 3 and 3 < 2 · 3, the only candidates for integrals of 2 and 3 are numbers
of the form a = p · q, where p, q ∈ P. However, then (p · q)′ = p + q > 5, hence 2 and 3
can have no integrals a ∈ N. Ufnarovski and Åhlander [3] found their integrals in rational
21 ′
numbers: (− 16 ) = 2, (− 134
)′ = 3. Any integral of 17 has at most 3 different prime factors
because 17 < 2 · 3 · 5. It cannot have only two different prime factors because if a = pq
then a′ = p + q and the sum of any two primes is not 17. However, if a = pqr, then
(pqr)′ = pq + pr + qr ≥ 2 · 3 + 2 · 5 + 3 · 5 = 31 > 17. Thus there is no integral of 17.
The numbers s ∈ N not divisible by any square c2 where c > 1 are called square-free
numbers. They can be either products of different primes or equal to 1. The set of square-free
numbers is denoted S.

Definition 22. Let a = pn1 1 · · · pnmm , where pi ∈ P. The number ni is called the exponent of
a prime pi in a. Let us define the following functions of a:
a
s(a) is the greatest square-free divisor of a, such that gcd(s(a), s(a) ) = 1,
p(a) = p1 · · · pm is the product of all prime factors of a,
a a
f (a) = p(a) = p1 ···p m
, r(a) = a · p(a) = a · (p1 · · · pm ),
hmax (a) = max{n1 , . . . , nm }, hmin (a) = min{n1 , . . . , nm }.
For a = 1 we define s(a) = p(a) = f (a) = r(a) = hmax (a) = hmin (a) = 1.
a
From this definition it follows that if a 6= s(a) then hmin ( p(a) ) ≥ 2.

Proposition 23. i) Let a ∈ N. Then a′ = f (a)p(a)a∗ , p(a)a∗ ∈ N and a = r(f (a))s(a).

ii) If a ∈ P then p(a)a∗ = 1 and f (a) = a′ . If a ∈ N is not a prime then p(a)a∗ > 1 and
f (a) < a′ .
iii) If a = pn1 1 · · · pnmm and ni is not divisible by pi for all i ∈ {1, 2, . . . , m} then gcd(f (a), p(a)a∗ ) =
1 and gcd(a, a′ ) = f (a).
iv) If hmin (a) ≥ 2, then s(a) = 1, hence a = r(f (a)). If hmax (a) = 1 then p(a) = a and
f (a) = 1, hence r(f (a)) = r(1) = 1.

Proof. i) If a = 1 then a′ = a∗ = 0 implies a′ = f (a)p(a)a∗ . If a = pn1 1 · · · pnmm then a′ =

a
aa∗ = ( p(a) a
)p(a)a∗ and p(a) = f (a) = p1n1 −1 · · · pm
nm −1
. Obviously p(a)a∗ = (p1 · · · pm )( np11 +

10
· · · + npmm
) ∈ N. Because f (a) is divisible by exactly those primes pi for which p2i divides a, it
a
is r(f (a)) = s(a) . Hence a = r(f (a))s(a).
ii) If a ∈ P then p(a) = a, a′ = 1, a∗ = a1 , p(a)a∗ = a a1 = 1 and f (a) = 1. Now suppose
a ∈ N is not a prime. Then either a = pn1 1 , where n1 > 1, or a = pn1 1 · · · pnmm , where m ≥ 2. In
the first case p(a) = p1 , f (a) = p1n1 −1 , a′ = n1 p1n1 −1 , a∗ = np11 , hence p(a)a∗ = p1 np11 = n1 > 1.
In the second case, p(a)a∗ = (p1 · · · pm )( np11 + · · · + npm m
) is a sum of at least two natural
∗ a′ ′
numbers, hence p(a)a > 1. Therefore in both cases f (a) = p(a)a ∗ < a .

iii) Now p(a)a∗ = (p1 · · · pm )( np11 +· · ·+ npm

m
) is not divisible by any of the primes pi dividing
a because ni is not divisible by pi . Hence p(a)a∗ = q1u1 · · · qtut and a′ = p1n1 −1 · · · pnmm −1 q1u1 · · · qtut ,
where all the primes p1 , · · · , pm , q1 , · · · , qt are distinct. Hence gcd(f (a), p(a)a∗ ) = 1 and
gcd(a, a′ ) = f (a).
iv) If hmin (a) ≥ 2, then p(a) = p(f (a)), hence r(f (a)) = a. If hmax (a) = 1 then p(a) = a
and f (a) = 1, hence r(f (a)) = 1.

Definition 24. If a = pn1 1 · · · pnmm where all ni < pi then a is in the set L whose elements are
called ”low” numbers.

Now it is possible to describe the factorization and consequently obtain some bounds of
the integrals of ”low” numbers.
T T
Corollary 25. i) If b ∈ L then I(b) N ⊂ L. Moreover, every a ∈ I(b) N is of the
form a = r(c)s(a), where b = cd, c < b, gcd(c, d) = 1, and b = a′ = r′ (c)s(a) + r(c)s′ (a).
Therefore:
i.a) If s(a) = a then r(c) = 1 hence c = 1. Conversely, if c = 1 then s(a) = a.
i.b) If s(a) = 1 then a = r(c).
i.c) If r(c) 6= 1 then s(a) ≤ b−r(c)
r ′ (c)
≤ cb , hence a ≤ r(b) − r(r(c)) ≤ r(b) − 1.
b−r(c)
i.d) If r(c) 6= 1 and s(a) is a prime then s(a) = r ′ (c)
.
i.e) If r(c) 6= 1 and s(a) = p1 p2 then r′ b−r(c)
(c)+r(c)
≤ s(a) ≤ b−5r(c)
r ′ (c)
.
n1 nk k
ii) For every b = p1 · · · pk ∈ L there are at most 2 − 1 different divisors c of b such
that there is an integral a of b of the form a = r(c)s(a).
iii) If b = pn1 1 · · · pnk k and ni = pi − 1 for all i, then any integral a ∈ I(b) N is a
T
square-free number: a = s(a).

Proof. i) If a′ = b and a contains a factor ppi i then a′ = b is also divisible by ppi i , too, hence
b is not in L. Thus, because b ∈ L, it must be a ∈ L. Hence f (a) divides a′ = b, f (a) < a′ ,
a
and r(f (a)) = s(a) . Thus one writes c = f (a), one really can get all the integrals of b ∈ L in
the described form.
i.a) and i.b) are obvious.
′ (a)
i.c) If r(c) 6= 1 then r′ (c) 6= 0 and s(a) = b − r(c)s
r ′ (c)
≤ b−r(c)
r ′ (c)
b
= (cp(c)) b b
′ = c′ p(c)+cp′ (c) ≤ c .

Hence a = r(c)s(a) ≤ c · p(c) b−r(c)

c
≤ p(b)b − p(c)r(c) = r(b) − r(r(c)) ≤ r(b) − 1 because
p(c) = p(r(c)).
i.d) If r(c) 6= 1 and s(a) is a prime, then s′ (a) = 1 and s(a) = b−r(c)
r ′ (c)
.

11
 b−5r(c)
i.e) If r(c) 6= 1 and s(a) = p1 p2 then s′ (a) = p1 + p2 ≤ 2 + 3 = 5 hence s(a) = r ′ (c)
.
The inequality r′ b−r(c)
(c)+r(c)
≤ s(a) follows from the fact that p1 + p2 ≤ p1 p2 for any two primes
p1 and p2 implying a = (r(c)p1 p2 )′ = r′ (c)p1 p2 + r(c)(p1 + p2 ) ≤ (r′ (c) + r(c))p1 p2 .
′

ii) Because each factor pni i of b = cd = pn1 1 · · · pnk k belongs either to c = f (a) or d = f (a)
b
,
′ k ′
and because c < a , there are at most 2 − 1 possible factors c dividing b = a such that
a = r(c)s(a). However, different integrals of b may correspond to the same c, as in the
example (3 · 13)′ = (5 · 11)′ = 16, in which both integrals 3 · 13 and 5 · 11 correspond to c = 1.
iii) If pi = ni − 1 for all i, then r(c) = 1 or r(c) contains at least one factor ppi i , but
then this is a contradiction because in that case a = b′ would also be divisible by ppi i . Hence
r(c) = 1 and a = s(a).

Example 26. Let b = 52 11. There are three possible factorizations b = cd such that
2
gcd(c, d) = 1 and 1 ≤ c < b, T corresponding to numbers c1 = 1, c2 = 5 and c3 = 11. Thus
the candidates for a ∈ I(b) N are:
square-free numbers a = r(1)s(a) = s(a)2 · 52 11, not divisible by 5 or 11,
2
numbers of the form: a = r(52 )s(a) = 53 s(a), where s(a) ≤ r′ b(c) = 53·5112 = 11 3
≤ 4, hence
s(a) ∈ {1, 2, 3}, and
2 11 2
numbers of the form a = r(11)s(a) = 112 s(a), where s(a) ≤ r′ b(c) = 52·11 = 52 , hence
s(a) ∈ {1, 2, 3, 5, 6, 7, 10, 11}.
It can easily be seen that s(a) > 1 because s(a) = 1 implies r′ (c) = b, but it is 1′ = 0 6= b,
(5 ) = 3 · 52 6= b and (112 )′ = 2 · 11 6= b. Thus in the case c2 = 52 it must be s(a) ∈ {2, 3}
3 ′

and in the case of c3 = 11 it is seen that s(a) ∈ {2, 3, 5, 6, 7, 10, 11} is either a prime or a
product of two primes.
Hence in the case a = 53 s(a) one can use the formula for the prime s(a) = b−r(c) r ′ (c)
=
52 11−53
3·52
= 11−5
3
= 2 and one can verify directly that (2 · 53 )′ = 53 + 2 · 3 · 52 = 52 (5 + 6) = b.
Now let a = 112 s(a). If s(a) is a prime one gets s(a) = b−r(c)
2 2
r ′ (c)
= 5 11−11
2·11)
= 25−11
2
= 7 and
2 ′ 2
one can check directly that (11 · 7) = 2 · 11 · 7 + 11 = 11(14 + 11) = b. It can easily be
seen that s(a) cannot be a product of two primes because that would imply a contradiction:
0 ≤ s(a) ≤ b−5r(c)
2 2
r ′ (c)
= 5 11−5·11
2·11
< 0.

Square-free integrals corresponding to c1 = 1 are more difficult to find. It is necessary to

2
check the derivatives of all square-free numbers a ≤ b4 and compare them with b.
A similar estimate as in Proposition 17 can be made about positive solutions to the
equation ( ab )′ = p2 , where p ∈ P.

Definition 27. Let Pi+1 denote the i-th odd prime. For any d ∈ R let Od = 1 · m
Q
i=1 Pi+1
denote the product of the first m odd primes such that the sum of their reciprocals, denoted
R(d), is not smaller than d.

Thus R(d) = m
P 1
P∞ 1
i=1 Pi+1 ≥ d. Because the series i=1 Pi+1 diverges, Od is well defined for
any real d.

Example 28. If d ≤ 13 then Od = P2 = 3. If 31 ≤ d ≤ 31 + 15 then Od = P2 P3 = 3 · 5 = 15. If

1
3
+ 15 ≤ d ≤ 13 + 51 + 71 then Od = P2 P3 P4 = 3 · 5 · 7 = 105.

12
Proposition 29. Let ( ab )′ = p2 , where p ∈ P, p > 2, gcd(a, b) = 1, ab > 0. Then a =
p1 · · · pm is an odd square-free number with m ≥ 9 prime factors pi ∈ P and b is of the form
b = q1q1 n1 . . . qsqs ns where qi ∈ P and ni ∈ N. Moreover, L(b) ∈ N, L(a) > L(b) ≥ 1 and
a′ > a > O1 ≥ 3 · 5 · 7 · 11 · 13 · 17 · 19 · 23 · 29.
′ ′
Proof. ( ab )′ = p2 implies a b−ab
b2
= p2 , hence b = pc, where c ∈ N. Therefore a′ pc − ab′ = 2c2 p
and because a and b have no common prime factors, we have b′ = kc. Hence a′ pc−akc = 2c2 p
and by dividing this equation by c one gets a′ p − ak = 2cp, hence k = pd because p divides
b so it cannot divide a. This implies a′ p − apd = 2cp, hence a′ − ad = 2c, and b′ = pdc = bd.
Thus d = L(b) = b∗ . By Proposition 2, b = q1q1 n1 · · · qsqs ns , where n1 + · · · + ns = d.
Because ab > 0, it can be assumed a > 0 and b > 0. It must be a > 1 because a = 1
implies a′ = 0 and −d = 2c, implying a contradiction: 0 = 2c + d > 0.
From the equation a′ − ad = 2c follows gcd(a, a′ ) = 1, because any such common prime
factor different from 2 would also divide c and b and this would contradict gcd(a, b) = 1, and
if 2 divides a and a′ then it must be a = 4e, hence a′ = 4(e + e′ ) and 4(e′ + e − 4ed) = 2c
would imply that 2 also divides b.
Now it can be seen that a cannot be divisible by 2. In that case 2 would also divide a′
or p. However, this is impossible because gcd(a′ , a) = 1 and p > 2.
Because gcd(a, a′ ) = 1, a must be a square-free number. Because a > 1, it must be:
′
a = p1 · · · pm , and because a′ − ad = 2c > 0 it is a′ > ad ≥ a, hence L(a) = aa =
P m 1
i=1 pi > d ≥ 1. Therefore a > Od ≥ O1 ≥ 3 · 5 · 7 · 11 · 13 · 17 · 19 · 23 · 29 and
m ≥ 9, because the sum of the reciprocals of the first eight odd primes is less than 1:
1
3
+ 15 + 71 + 11
1 1
+ 13 1
+ 17 + 19 1 1
+ 23 = 0.9987 · · · < 1.

Ufnarovski and Åhlander [3] conjectured that some rational numbers have no integrals.
The above proposition shows that the integrals of 23 are not easy to find, if they exist at all.
The same holds for any p2 , p ∈ P.
In Proposition 7.iii) I showed that if y ∈ Ny∗ = Nr+r∗ and if r + r∗ 6= 0 then there is
exactly one x ∈ Nr such that x′ = y and this is x = yr . Hence to find a rational integral x
of any nonzero rational number y one just has to find a rational number r = x∗ such that
r + r∗ = y ∗ . In other words, one has to find a logarithmic class Nr such that D(Nr ) = Ny∗ .
The equation r + r∗ = y ∗ translates into r2 − y ∗ r + r′ = 0. Perhaps for some rational numbers
y this equation cannot be satisfied by any rational number r, hence such y cannot have a
′
rational integral x. Expressing√ r as a function of its derivative r , one gets at most two
y∗ ± (y ∗ )2 −4r ′ ∗ 2
different solutions r1,2 = 2
. Now r ∈ Q implies (y ∗ )2 − 4r′ ≥ 0 and r′ ≤ (y4) .
Because x ∈ Q implies r = x∗ ∈ Q the expression (y ∗ )2 − 4r′ = q 2 must be a square of a
rational number q.

Example 30. If y = 32 then y ∗ = 21 − 31 = 16 . If there is a positive rational number x = ab ∈ Nr

such that x′ = y = 32 then r = x∗ = ( ab )∗ = xy = 3x 2
> 0 satisfies the equation r + r∗ = y ∗ = 61 .
2 1
This equation translates into r − 6 r + r = 0. For each r′ one gets at most two different
′
√ 1
√1 ′ q
y ∗ ± (y ∗ )2 −4r ′ ± 36 −4r ∗ 2
solutions r1,2 = 2
= 6
2
1
= 12 ± 144 1
− r′ . Hence r′ ≤ (y4) = 144
1
.
∗ ∗
P m 1
It is known from Proposition 29 that r = a − b = i=1 pi − d > 0, where d ∈ N. Thus
x = yr > 0. Now r′ = ( 3x 2 ′
) = 3x−6
9x2
= x−23x2
= x−2 3x 2
3x2 2 3x
, hence r∗ = x−2
2x
. If x ≤ 2 then r∗ ≤ 0

13
 2
1 y y
and r = y ∗ − r∗ ≤ y ∗ = 6
leads to a contradiction: x = r
≥ y∗
= 3
1 = 4. Thus it must be
6
2
1 y y
x > 2, hence r∗ > 0. Now 0 < r = y ∗ − r∗ < y ∗ = 6
implies x = r
> y∗
= 3
1 = 4, hence
6
′ 2
a > 4b. Thus there is no positive rational number x such that x ≤ 4 and x = 3
.

3.4 Partial derivatives and partial differential equations

Definition 31. Let a = ki=1 pixi be the factorization of a ∈ N into primes. The partial
Q
∂a
derivative ∂pi ∂a
= Dpi (a) is defined as ∂p i
= a xpii . If p ∈ P is not in the factorization of a, then
∂a
∂pi
= 0.
Pk ∂a
From this definition it immediately follows that D(a) = a i=1 ∂pi . One can also define
∂(∂a/∂pj )
= Dp22 = ∂(∂a/∂p i)
2 2a
higher partial derivatives; for example: ∂p∂i ∂p
a
j
= Dp2i pj = ∂pi
, ∂p∂i ∂p j ∂p i
,
i
etc., and study partial differential equations.
Example 32. D2 (23 · 54 ) = 3 · 22 · 54 , D5 (3 · 22 54 ) = 3 · 22 · 4 · 53 = 3 · 24 · 53
D5 (23 · 54 ) = 23 · 4 · 53 = 25 · 53 and D2 (25 · 53 ) = 5 · 24 · 53 . Thus the order of applying Dp
and Dq is important: Dp Dq is not always equal to Dq Dp .
Proposition 33. If a = pe q f and gcd(e, q) = 1 and gcd(f, p) = 1, then Dp Dq (a) = Dq Dp (a).
Proof. In that case it is Dp Dq (pe q f ) = Dp (pe f q f −1 ) = epe−1 f q f −1 and Dq Dp (pe q f ) =
Dq (epe−1 q f ) = epe−1 f q f −1 .
Proposition 34. Let n = pe q f c, where p, q are primes not dividing c. Then Dp (pe q f c) =
Dq (pe q f c) if and only if e = kp and f = kq, where k ∈ N.
Proof. From Dp (pe q f c) = epe−1 q f c = f pe q f −1 c = Dq (pe q f c) follows eq = bp, hence e = kp
and f = kq. Then Dp (pkp q kq c) = k(pkp q kq c) = Dq (pe q f c).
Proposition 35. i) The only solutions to the partial differential equation Dp22 (n) = n in
natural numbers are n = ppk c, where gcd(c, p) = 1.
ii) If n = pe d where gcd(p, d) = 1 and e ≥ p then Dp22 (n) ≥ n.
iii) The only solutions to the partial differential equation (Dp22 + Dq22 )(n) = n in natural
numbers are n = pp c and n = q q c, where gcd(pq, c) = 1.
Proof. i) Let n = pe c, where e ∈ N {0} and c ∈ N is not divisible by p.
S
If e = 0 then Dp (n) = 0, hence Dp22 (n) = 0. If e = kp and k ∈ N then Dp (n) = Dp (ppk c) =
pkppk−1 c = ppk c = n hence Dp22 (n) = n. If e = 1 then Dp (n) = c, hence Dp22 (n) = Dp (c) = 0.
ii) If e = kp and k ∈ N then Dp22 (n) = n, as is already known from i). If gcd(e, p) = 1
and e > p then Dp (n) = Dp (pe c) = epe−1 c, hence Dp22 (n) = Dp (epe−1 c) = e(e − 1)pe−2 c ≥
(p + 1)ppe−2 c > pe c = n.
iii). One can write n = pe q f c, where {e, f } ⊆ N {0} and c ∈ N is not divisible by p
S
or q. So it is necessary to study the equation (Dp22 + Dq22 )(pe q f c) = pe q f c. If f = 1 then
Dq22 (pe c) = 0 and one gets the equation Dp2 (n) = n whose only solution is n = pp c, as is
known from i). Likewise if e = 1 one gets n = q q c.

14
 In the case of e ≥ 2 and f ≥ 2, one can use the following argument:
Obviously gcd(qc, p) = 1 implies that Dp22 (n) is divisible by q f c. Likewise gcd(pc, q) = 1
implies that Dq22 (n) is divisible by pe c. So one can write Dp22 (n) = aq f c, Dq22 (n) = bpe c, where
a, b ∈ N {0}. Suppose a 6= 0 and b 6= 0. Hence one gets the equation aq f c + bpe c = pe q f c.
S
This equation can be solved only if a is divisible by pe and if b is divisible by q f . Suppose
a 6= 0 and b 6= 0. Then Dp22 (n) ≥ n and Dq22 (n) ≥ n, hence Dp22 (n) + Dq22 (n) ≥ 2n and the
equality is not possible. Hence it must be either a = 0 or b = 0. However, this is not possible
if e ≥ 2 and f ≥ 2.

4 Concluding remarks
I have proved some new results about the arithmetic derivative and integral. I have defined
arithmetic partial derivatives and solved some arithmetic partial differential equations. I
have shown that, for any solution to the system m′ = n, n′ = m in natural numbers, at
least one of the numbers m and n is not smaller than the number 2 · 3 · 5 · 7 · 11 · 13 ·
17 · 19 · 23 · 31 · 37 · 41 · 43 · 47 · 53 · 59. The arithmetic derivative can be defined on se-
quences of numbers as follows: D(a1 , a2 , . . . , an , . . .) = (D(a1 ), D(a2 ), . . . , D(an ), . . .)). Thus
taking any integer sequence (a) = (a1 , a2 , . . . , an , . . .) one can get an infinite family of de-
rived sequences D(a), D2 (a), . . . , Dk (a), . . . . I believe many other useful applications of the
arithmetic derivative will be discovered.

5 Acknowledgements
I thank the referee for many suggestions that helped me to improve this paper. I am also
grateful to Marko Petkovšek for his explanation of why there are uncountably many functions
on the unit circle in the complex plane satisfying the Leibnitz rule (Proposition 1.ii).

References
[1] E. J. Barbeau, Remarks on arithmetic derivative, Canad. Math. Bull, 4 (1961), 117–122.

[2] M. Stay, Generalized number derivatives, J. Integer Seq., 8 (2005), Article 05.1.4.

[3] V. Ufnarovski and B. Åhlander, How to differentiate a number, J. Integer Seq., 6 (2003),
Article 03.3.4.

2010 Mathematics Subject Classification: Primary 11A25; Secondary 11A41, 11N05, 11N56,
11Y55.
Keywords: arithmetic derivative, arithmetic antiderivative, arithmetic partial derivative,
Leibnitz rule.

(Concerned with sequences A000040, A000290, A051674.)

15
Received May 19 2011; revised version received March 18 2012. Published in Journal of
Integer Sequences, March 25 2012.

Return to Journal of Integer Sequences home page.

16