LECTURE 1: DIVISIBILITY

1. Introduction
Number theory concerns itself with studying the multiplicative and additive
structure of the natural numbers
N = {1, 2, 3, . . . }.
Frequently, number theoretic questions are better asked in the set of all integers
Z = {0, ±1, ±2, ±3, . . . },
and better answered by making use of the rational numbers
 
p
Q= : p ∈ Z, q ∈ N ,
q
the real numbers R, and the complex numbers C, where more structure may
become apparent.
Some form of number theory was developed by the ancient Babylonians,
Egyptians and Greeks, and many modern problems are motivated by this work.
When studying other long-standing areas of mathematics, such as Euclidean
geometry, calculus or linear algebra, it is easy to come away with the conclusion
that everything was worked out long ago. Number theory is not like that, and
for many problems, some of them ancient, we have more conjectures than
theorems. Nevertheless, new methods and results emerge in fits and starts,
and the subject has seen many great advances in just the last few decades.
Here are a few examples of number-theoretic problems that have been solved
only recently or still remain open.
Problem 1.1 (Fermat’s Last Theorem). For any integer n > 3, the equation
xn + y n = z n has no solutions with x, y, z ∈ N.
This conjecture was stated by Fermat around 1637, and was motivated by much
earlier work of Diophantus (c. 200-300AD). It was finally proven by Wiles in 1995.
Problem 1.2 (Congruent Number Problem). For which n ∈ N is there a right
triangle with rational sides and area n (2n = xy with x2 +y 2 = z 2 , x, y, z ∈ Q)?
Such an n is called a congruent number. For instance, 6 is congruent because it
is the area of the 3–4–5 right triangle, but it can be shown that 3 is not a congruent
number. The question can be very subtle, as illustrated by the example 53, which is
congruent, but for which the simplest suitable right triangle has legs 1472112483
202332130 and
21447205780
1472112483 .
This question appeared in 10th century Arab manuscripts, but is possibly even
older. In 1983, Tunnell gave a simple numerical criterion for determining whether a
given n is congruent or not, but its correctness depends on the unproven Birch and
1
2 LECTURE 1

Swinnerton-Dyer conjecture (one of the $1 million Clay Millennium Prize problems).

Nevertheless, there is an algorithm that is guaranteed to work if the BSD conjecture
is true (even if we cannot prove it) and efficient in practice.
Problem 1.3 (Catalan’s Conjecture). The only consecutive powers of natural
numbers are 8 and 9 (xn − y m = 1).
This was conjectured by Catalan in 1844 (though again it is no doubt much older)
and proven by Mihăilescu in 2002.
Problem 1.4 (Twin Prime Conjecture). There are infinitely many pairs of
prime numbers that differ by 2.
This conjecture was first stated in print by de Polignac in 1849, but its origins are
probably much older, perhaps as far back as Euclid, who recorded a proof that there
are infinitely many primes in the Elements, c. 300BC; we will study Euclid’s proof
early on in the course. In July 2014, the mathematics consortium D. H. J. Polymath,
led by Terry Tao, has built on the pivotal work of Yitang Zhang and James Maynard
to prove that there are infinitely many pairs of primes that differ by at most 246.
Problem 1.5 (Goldbach Conjecture). Every integer n > 1 can be expressed
as the sum of at most three prime numbers.
Goldbach stated this conjecture in a letter to Euler in June 1742. Euler replied
that it is equivalent to the statement “every even integer n > 2 is the sum of two
prime numbers”, and this is often taken as the statement of the problem. In 2013,
Helfgott claimed a proof of Goldbach’s conjecture for odd numbers n. The problem
for even n, including Euler’s reformulation, remains open. However, it is known that
“almost all” even natural numbers can indeed be written as the sum of two primes.
Problem 1.6 (ABC Conjecture). For each ε > 0, there exists Cε > 0 (de-
pending only on ε) such that whenever abc 6= 0 and a + b + c = 0, then
 Y 1+ε
max{|a|, |b|, |c|} 6 Cε p
p|abc

where the product is taken over distinct prime divisors of a, b and c.

The ABC Conjecture was stated by Oesterlé and Masser in 1985. It has many
profound implications, but until very recently seemed far beyond reach. In 2012
Shinichi Mochizuki has recently claimed to have proved this conjecture, however,
and there is considerable activity attempting to verify his proof.
Problem 1.7. Is the number
∞
X 1
ζ(k) =
n=1
nk
irrational for every integer k > 1?
This question has its roots in the Basel problem from 1644, which asked for the
value of ζ(2). Euler became famous when he solved the Basel problem in 1734,
 LECTURE 1 3
2
proving that ζ(2) = π6 . More generally, ζ(k) is a rational multiple of π k whenever
k is even. Lindemann proved that π is transcendental in 1844, and it follows that
ζ(k) is irrational for every even k. Apéry stunned the number theory community
in 1978 by proving that ζ(3) is irrational. Since then, Apéry’s methods have been
broadened to prove that ζ(k) is irrational for infinitely many odd values of k, but
the full question remains open.

2. Divisibility
Definition 2.1. (i) Suppose that a, b ∈ Z. We say that b divides a (written
b | a) when there exists c ∈ Z such that a = bc. In such circumstances, we say
that a is divisible by b, or that b is a divisor of a;
(ii) When a is not divisible by b, we write b - a;
(iii) When b | a and 1 6 b < a, we say that b is a proper divisor of a;
(iv) We write ak k b when ak | b but ak+1 - b.
It is understood that b | a makes sense only when b is non-zero.
The next theorem records the basic properties of divisibility that are intu-
itively clear, but easily established from the definition.
Theorem 2.2. (i) a | a for every a ∈ Z \ {0};
(ii) a | 0 for every a ∈ Z \ {0};
(iii) if a | b and b | c, then a | c;
(iv) if a | b and a | c, then for all x, y ∈ Z, one has a | (bx + cy);
(v) if a | b and b | a, then a = ±b;
(vi) if a | b and a > 0 and b > 0, then a 6 b;
(vii) when m 6= 0, one has a | b ⇔ ma | mb.
Proof. We will leave these assertions as exercises, though in order to illustrate
ideas, we will give a formal proof of part (vii). Suppose that m 6= 0 and a | b.
Then there exists c ∈ Z with the property that b = ac, whence mb = m(ac).
So there exists c ∈ Z with the property that (mb) = (ma)c, whence by the
definition of divisibility (ma) | (mb). Conversely, if m 6= 0 and ma | mb, then
there exists c ∈ Z with mb = (ma)c. But since m 6= 0, the latter implies that
b = ac. So there exists c ∈ Z with the property that b = ac, so from the
definition of divisibility, one has a | b. 
The next theorem lays the groundwork for the development of the theory of
congruences.
Theorem 2.3 (The Division Algorithm). For any a, b ∈ Z with a > 0, there
exist unique integers q and r with b = qa + r and 0 6 r < a. If, further, a - b,
then the stronger inequality 0 < r < a holds.
Proof. Let aq be the greatest multiple of a not exceeding b. Then if we put
r = b − aq, one has r > 0. Moreover, by hypothesis one has a(q + 1) > b,
and thus r = b − aq < a. This establishes the existence of the integers q
and r as stated. In order to establish uniqueness, suppose that another pair
4 LECTURE 1

q 0 , r0 satisfy analogous conditions. If r 6= r0 , there is no loss of generality in

supposing that r < r0 . Then since aq 0 +r0 = b = aq+r, one has a(q−q 0 ) = r0 −r,
whence a | (r0 − r) and 0 < r0 − r < a. But the latter contradicts case (vi) of
Theorem 2.2 (which would imply that r0 − r > a). Thus we find that r = r0 ,
and this now leads to the equation qa = q 0 a. But a is non-zero, so q = q 0 .
Thus we find that (q, r) = (q 0 , r0 ), and this establishes uniqueness.
Finally, if r = 0 then b = qa, whence a | b. The final assertion of the theorem
is now immediate. 
Definition 2.4. (i) Suppose that a ∈ Z \ {0} and b, c ∈ Z. We say that a is a
common divisor of b and c when a | b and a | c;
(ii) When b and c are not both zero, the number of common divisors of b and c
is finite (see Theorem 2.2(vi)), and thus we may define the greatest common
divisor (or highest common factor) of b and c to be the largest (positive)
common divisor. The greatest common divisor of b and c is written (b, c) (or
gcd(b, c) or hcf(b, c));
(iii) When g1 , . . . , gn are integers, not all zero, we similarly write (g1 , . . . , gn )
for the greatest integer d satisfying the condition that d | gi (1 6 i 6 n).
Example 2.5. One has (0, 2) = 2, (1, 3) = 1 and (1729, 182) = 91 (at
this point one can use trial and error, observing that (a, b) must be at most
min{|a|, |b|}).

The next theorem provides a useful tool to establish simple properties of

greatest common divisors.
Theorem 2.6. If g = (b, c), then there exist integers x and y with g = bx + cy.
Proof. Define the integer d by setting
d = min{bu + cv : u, v ∈ Z and bu + cv > 0}.
Also, let x and y be the values of u and v corresponding to this minimum, so
that d = bx + cy.
We first prove that d | b. If to the contrary d - b, then by the Division
Algorithm (Theorem 2.3), there exist integers r and q with b = dq + r and
0 < r < d. Then
r = b − dq = b − q(bx + cy) = b(1 − qx) + c(−qy),
whence
r > min{bu + cv : u, v ∈ Z and bu + cv > 0} = d.
This gives a contradiction, since r < d, and thus we find that d | b.
A similar argument shows that d | c, and thus d is indeed a common divisor
of b and c, which is to say that d 6 (b, c). But g = (b, c), and so there
exist integers B and C with b = gB and c = gC. Consequently, one has
d = g(Bx + Cy), and hence g | d. Thus g > 0, d > 0 and g | d, so by
Theorem 2.2(vi) one has g 6 d. Then one has d > (b, c) in addition to the
relation d 6 (b, c) which we derived above, so that necessarily d = (b, c). But
then (b, c) = bx + cy, and this completes the proof of the theorem. 
 LECTURE 1 5

Theorem 2.7. The greatest common divisor of b and c is:

(i) the least positive value of bx + cy, as x and y range over Z;
(ii) the positive common divisor of b and c that is divisible by all other such
divisors.

Proof. The assertion (i) is plain from Theorem 2.6. For part (ii), observe that
there exist integers x and y with (b, c) = bx + cy. Then if d | b and d | c, say
b = dB and c = dC, one finds that (b, c) = d(Bx + Cy), whence d | (b, c). So
(b, c) is divisible by all other positive common divisors of b and c. 
Remark 2.8. If g1 , . . . , gn are not all zero, then it follows as in the proof of
Theorem 2.6 that there exist integers x1 , . . . , xn with (g1 , . . . , gn ) = g1 x1 +
· · · + gn xn .
The criterion for determining the greatest common divisor recorded in The-
orem 2.6, and (in modified form) in Theorem 2.7, provides a simple and di-
rect approach to establishing simple properties of the greatest common divisor
function.
Theorem 2.9. Whenever m ∈ N, one has (ma, mb) = m(a, b).

Proof. Making use of Theorem 2.7(i) (twice), one has

(ma, mb) = min{max + mby : x, y ∈ Z and max + mby > 0}
= m min{ax + by : x, y ∈ Z and ax + by > 0}
= m(a, b).

Remark 2.10. Similiarly, when d ∈ N, and d | a and d | b, one has (a/d, b/d) =
(a, b)/d. In particular, if g = (a, b), then (a/g, b/g) = 1.

Proof. The first assertion follows from Theorem 2.9 by means of the relation
(d(a/d), d(b/d)) = d(a/d, b/d), and the second is immediate from the first. 
Theorem 2.11. Whenever a, b, m are integers with (a, m) = (b, m) = 1, one
has (ab, m) = 1.

Proof. By Theorem 2.6, there exist integers x, y, u, v with 1 = ax + my =

bu + mv. Thus we obtain
(ax)(bu) = (1 − my)(1 − mv) = 1 − mw,
say, with w = y + v − mvy. Consequently, one has (ab)(xu) + mw = 1. But
then by Theorem 2.2(iv), any common divisor of ab and m divides 1. We
therefore conclude that (ab, m) = 1. 
Theorem 2.12. For any integer x, and for any integers a and b, not both
zero, one has
(a, b) = (b, a) = (a, −b) = (a, b + ax).
6 LECTURE 1

Proof. The first assertions of the theorem are plain from Theorem 2.7(i). In
order to prove that (a, b) = (a, b+ax), observe that by Theorem 2.6, there exist
integers u and v with (a, b) = au+bv, whence (a, b) = a(u−xv)+(b+ax)v. We
therefore have (a, b + ax) | (a, b). But (a, b) | a and (a, b) | b, so (a, b) | (b + ax).
But now we have (a, b + ax) | (a, b) | (a, b + ax), and so by virtue of positivity,
Theorem 2.2(v) establishes the desired conclusion. 
Theorem 2.13. Suppose that c | ab and (b, c) = 1. Then c | a.

Proof. By Theorem 2.9, the hypotheses of the theorem imply that (ab, ac) =
|a|(b, c) = |a|. But by hypothesis, one has c | ab, which implies that c | (ab, ac).
We thus conclude that c | a. 

At last we are positioned to describe an algorithm for calculating greatest

common divisors. Of course, by exhaustive checking one could determine the
greatest common divisor of two integers b and c with O(min{|b|, |c|}) appli-
cations of the Division Algorithm, but the Euclidean Algorithm requires only
O(log min{|b|, |c|}) such divisions.
Theorem 2.14 (Euclidean Algorithm). Suppose that b ∈ Z and c ∈ N. Define
the integers ri and qi for i > 1 by repeated application of the Division Algorithm
thus:
b = cq1 + r1 , with 0 < r1 < c,
c = r1 q2 + r2 , with 0 < r2 < r1 ,
r1 = r2 q3 + r3 , with 0 < r3 < r2 ,
..
.
rj−2 = rj−1 qj + rj , with 0 < rj < rj−1 ,
rj−1 = rj qj+1 .
(Here we adopt obvious conventions if the process terminates prematurely.)
Then (b, c) = rj , the last non-zero remainder in the division process.

Proof. Repeated application of Theorem 2.12 yields

(b, c) = (b − cq1 , c) = (r1 , c)
= (c − r1 q2 , r1 ) = (r2 , r1 )
= (r1 − r2 q3 , r2 ) = (r3 , r2 )
= · · · = (rj , rj−1 ) = (rj , 0) = rj .
This conclusion of the theorem follows at once. 
Observation 2.15. One can apply the Euclidean Algorithm to obtain integral
solutions (x, y) to linear equations of the shape bx + cy = (b, c) by “reversing”
the application of the algorithm. In general, one can apply this method to
solve the equation bx + cy = k whenever (b, c) | k. (Why? Convince yourself
that this is the case.)
 LECTURE 1 7

Proof. Using the notation employed in the statement of the Euclidean Algo-
rithm, one finds that r1 is a linear combination of b and c, and then that r2
is a linear combination of c and r1 , and hence of b and c, and that r3 is a
linear combination of r1 and r2 , and hence of b and c, and so on. In this way,
we see that every remainder ri that occurs in the algorithm is itself a linear
combination of b and c, and the desired conclusion follows. 
Example 2.16. Determine the greatest common divisor of 2016 and 323, and
find integers x and y with 2016x + 323y = (2016, 323).

Proof. Applying the Euclidean Algorithm, we obtain

2016 = 323 · 6 + 78
323 = 78 · 4 + 11
78 = 11 · 7 + 1
11 = 1 · 11,
and so (2016, 323) = 1. Reversing this application of the Euclidean Algorithm,
we find that
1 = 78 − 11 · 7
= 78 − (323 − 78 · 4) · 7 = 78 · 29 − 323 · 7
= (2016 − 323 · 6) · 29 − 323 · 7 = 2016 · 29 − 323 · 181.
Thus we see that the equation 2016x + 323y = 1 has the solution (x, y) =
(29, −181). 
Note 2.17. One can obtain integral solutions to linear equations in more
variables by breaking the equation down into subequations of two variables
each. In order to illustrate the strategy, consider the equation 18x+39y+77z =
1. One can verify easily that (18, 39) = 3, and so the equation 18x + 39y = 3
possesses an integral solution, say 18x0 +39y0 = 3, which may be found via the
Euclidean Algorithm. Now substitute this solution into the original equation
with an additional parameter, and solve the resulting equation. We obtain the
equation 3l + 77z = 1. Since (3, 77) = 1, the latter equation has an integral
solution (l, z) = (l0 , z0 ), say, which may be found via the Euclidean Algorithm.
A solution of the original equation is then given by (x, y, z) = (l0 x0 , l0 y0 , z0 ).

We finish this section by introducing the concept of least common multiples.

Definition 2.18. (i) Integers a1 , . . . , an are said to have a common multiple b
when ai | b for 1 6 i 6 n.
(ii) The least common multiple of the integers a1 , . . . , an is the smallest pos-
itive common multiple of these integers, which we denote by [a1 , . . . , an ] or
lcm(a1 , . . . , an ).
Theorem 2.19. (i) If m is a positive integer, then [ma, mb] = m[a, b].
(ii) One has [a, b](a, b) = |ab|.
8 LECTURE 1

Proof. First consider the assertion of part (i) of the theorem. Let D = [ma, mb]
and d = [a, b]. Then md is a multiple of both ma and mb, so that md > D.
Also, D is a multiple of both ma and mb, so that D/m is a multiple of both a
and b. Then D/m > d. We have therefore shown that md 6 D 6 md, whence
D = md. This establishes part (i) of the theorem.
Now consider part (ii). Suppose first that (a, b) = 1. There is no loss of
generality in supposing that a > 0 and b > 0. Write [a, b] = ma, with b | ma.
Since (a, b) = 1, it follows from Theorem 2.13 that b | m, whence b 6 m. Then
ba 6 ma. But ba > [a, b] = ma. We therefore conclude that ab = [a, b], and
since (a, b) = 1, this yields the desired conclusion (a, b)[a, b] = |ab|.
Turning to the general case, put g = (a, b) and a0 = a/g, b0 = b/g. Then
(a , b ) = (a, b)/g = 1 and [a0 , b0 ] = [a, b]/g, so by the above,
0 0

(a, b)[a, b] |ab|

2
= (a0 , b0 )[a0 , b0 ] = |a0 b0 | = 2 .
g g
2
The desired identity follows on multiplying by g . 
Theorem 2.20. Suppose that b1 , . . . , bn are integers and that k = [b1 , . . . , bn ].
Then the set of all common multiples of b1 , . . . , bn is given by {km : m ∈ Z}.
Proof. Exercise.