Quadratic Diophantine Equations

Arkadii Slinko
After Diophantus, an equation is said to be a Diophantine equation if only
integer solutions to this equation are sought. In what follows we consider only
Diophantine equations.
1 Pythagoras triples
Let us consider the equation
x
2
+ y
2
= z
2
. (1)
Geometrically, any solution (x, y, z) to the equation (1) corresponds to a
right-angled triangle with the hypothenuse of length z and with the legs
of lengths x, y. That is why such a solution is called a Pythagoras triple.
For example, the triples (3, 4, 5) and (5, 12, 13) are Pythagoras triples since
3
2
+ 4
2
= 5
2
and 5
2
+ 12
2
= 13
2
.
Let (x, y, z) be a Pythagoras triple and d = gcd (x, y). Then x = x
1
d
and y = y
1
d, and z is also divisible by d, i.e., z = z
1
d. Thus the equation (1)
takes form x
2
1
d
2
+ y
2
1
d
2
= z
2
1
d
2
. We cancel d
2
and get
x
2
1
+ y
2
1
= z
2
1
, (2)
where gcd (x
1
, y
1
) = 1.
Let (x, y, z) be a Pythagoras triple. It is called primitive if gcd (x, y) = 1.
We showed above that any Pythagoras triple (x, y, z) can be obtained as
(x
1
d, y
1
d, z
1
d), where (x
1
, y
1
, z
1
) is a primitive Pythagoras triple. Now we
are going to characterize all primitive Pythagoras triples. Let (x
1
, y
1
, z
1
) be
one of them. Then at least one of the two numbers x
1
or y
1
must be odd.
Thus without loss of generality we assume that x
1
is odd. Moving y
2
to the
right we get
x
2
1
= z
2
1
y
2
1
or x
2
1
= (z
1
+ y
1
)(z
1
y
1
), (3)
1
Let d
1
= gcd (z
1
+ y
1
, z
1
y
1
). Then
z
1
+ y
1
= ad
1
and z
1
y
1
= bd
1
, (4)
where a and b are coprime. Substituting (4) in (3) we get
x
2
1
= abd
2
1
, (5)
which means that the product ab is a square of an integer. Since a and b
are coprime they must be squares of integers themselves: a = u
2
, b = v
2
for
some coprime integers u and v. In this case x
2
1
= u
2
v
2
d
2
1
and
x
1
= uvd
1
. (6)
Since x
1
is odd, the numbers u, v, d
1
are also odd. Adding and subtracting
the two equations in (4) we nd
2z
1
= ad
1
+ bd
1
= u
2
d
1
+ v
2
d
1
, 2y
1
= ad
1
bd
1
= u
2
d
1
v
2
d
1
,
or
y
1
=
u
2
v
2
2
d
1
, z
1
=
u
2
+ v
2
2
d
1
. (7)
Since x
1
and y
1
are coprime and d
1
is their common divisor we get d
1
= 1.
Substituting this into (6) and (7) we get
x
1
= uv, y
1
=
u
2
v
2
2
, z
1
=
u
2
+ v
2
2
, (8)
where u and v are odd and coprime. These formulae give us all primitive
Pythagorass triples (x, y, z). All Pythagorass triples will be given by for-
mulae
x = uvd, y =
u
2
v
2
2
d, z =
u
2
+ v
2
2
d, (9)
where u and v are odd and coprime. Thus we obtain
Theorem 1. For an arbitrary d and for arbitrary odd coprime u and v the
equation (9) gives us a Pythagorass triple and all Pythagorass triples can
be obtained in this way.
Exercise 1. Suppose that a pair of integers (p, q) satises the equation x
2

Ay
2
= 1, where A is not the square of an integer. Let us consider the
numbers u = p + q

A and u
2
= z + t

A. Then a pair of integers (z, t) is a

solution to Pells equation x
2
Ay
2
= 1.
2
Diophantus himself in his book takes a slightly more geometric approach.
Let (x, y, z) be a primitive Pythagorass triple. Then the pair of rational
numbers
_
x
z
,
y
z
_
is a solution to the equation
p
2
+ q
2
= 1 (10)
or a point with rational coordinates on a unit circle centered at the origin.
Clearly any Pythagoras triple (xd, yd, zd) will give us the same point on the
circle. The set of points with rational coordinates in the rst quadrant of the
unit circle can be described as follows.
Let us draw the straight line q = p 1, > 0. Then it will intersect
the unit circle twice: at the point (0, 1) and at some point (p

, q

). If
the numbers p

, q

are rational, then =

q

1
p

is also rational. If is
rational, then the coordinate p

satises the equation p

2

+ (p

1)
2
= 1,
i.e., p

=
2

2
+ 1
and
p

=
2

2
+ 1
, q

=

2
1

2
+ 1
. (11)
and they are both rational. Thus the formula (11) gives us all rational points
in the rst quadrant of the unit circle. Substituting =
a
b
we get all primitive
Pythagoras triples (8) but in a slightly dierent form:
x = a
2
b
2
, y = 2ab, z = a
2
+ b
2
, (12)
where a, b are two comprime integers of dierent parities. To see that (12) is
equivalent to (8) we note that (12) can be obtained from (8) by the substi-
tution u = a + b, v = a b.
As an application we will prove the following theorem which is due to
Fermat. The method of descent used in it is very powerful.
Theorem 2. The Fermat equation
x
4
+ y
4
= z
4
does not have integer solutions (x, y, z) such that xyz = 0.
3
We will prove a stronger statement that the equation
x
4
+ y
4
= z
2
(13)
does not have integer solutions (x, y, z) such that xyz = 0. Suppose that this
equation has a solution. Then it has a solution (x
0
, y
0
, z
0
) with positive com-
ponents. We will prove that it is always possible to nd a solution (x
1
, y
1
, z
1
)
to (13) with positive components and such that z
1
< z
0
. This will give us a
contradiction because if at least one solution existed it would be possible to
choose the solution (x
0
, y
0
, z
0
) with minimal z
0
.
Let us prove this statement and assume that (x
0
, y
0
, z
0
) is a solution to
(13) with positive components. Firstly we suppose that gcd (x
0
, y
0
) = d > 1.
Then
x = x
1
d, y = y
1
d,
where gcd (x
1
, y
1
) = 1. Then z is also divisible by d and z = z
1
d. We get
x
4
1
d
4
+ y
4
1
d
4
= z
2
1
d
2
. It follows that z
1
is divisible by d
2
and
x
4
1
+ y
4
1
=
_
z
1
d
2
_
2
= (z

1
)
2
,
where z

1
= z
1
/d
2
. Thus (x
1
, y
1
, z

1
) is a solution to (13) with z

1
< z and
we are done. It means that we may assume that gcd (x
0
, y
0
) = 1 with x
0
being odd. Then (x
2
0
, y
2
0
, z
0
) is a primitive Pythagoras triple with an odd
rst component. Using the characterization (12) we have
x
2
0
= a
2
b
2
, y
2
0
= 2ab, z
0
= a
2
+ b
2
(14)
with a and b being coprime and of dierent parities. As x
0
is odd, we have
x
2
0
1 (mod 4). If a is even and b is odd, then a
2
b
2
0 1 (mod 4)
3 (mod 4). Hence the only possibility is a being odd and b being even. Since
gcd (a, b) = 1 and a is odd, we have gcd (a, 2b) = 1. As according to (14)
y
2
0
= a 2b we see that a and 2b are perfect squares, i.e
a = t
2
and 2b = s
2
. (15)
Applying (14) again we have x
2
0
+ b
2
= a
2
and (x
0
, b, a) is a Pythagoras
triple. Since a and b are coprime this triple is primitive. Therefore
x
0
= m
2
n
2
, b = 2mn, a = m
2
+ n
2
(16)
4
with m and n being coprime and of dierent parities. Due to (15) 2b =
4mn = s
2
and it follows that m = p
2
and n = q
2
. Thus using (15) again we
get
p
4
+ q
4
= t
2
,
i.e., the triple (p, q, t) is a solution to (13). All we need to show is that t < z
0
.
This is true since t =

a <
4

a
2
+ b
2
=
4

z
0
< z
0
.
This proves the theorem.
2 Pells equation
Let A be a positive integer which is not a perfect square. The equation
x
2
Ay
2
= 1 (17)
is called Pells equation.
The requirement that A is not the square of a whole number is equivalent
to the fact that the number

A is irrational. It is very important! In this

case the set of numbers Q(

A) consisting of
p + q

A, p, q Q, (18)
are quadratic irrationalities with the following properties:
(p + q

A) + (r + s

A) = (p + r) + (q + s)

A, (19)
(p + q

A)(r + s

A) = (pr + qsA) + (ps + qr)

A, (20)
(p + q

A)
1
=
1
p
2
q
2
A
(p q

A). (21)
Note that if p
2
q
2
A = 0, then either p = q = 0 or q = 0. In the latter
case A =
p
2
q
2
or

A =
p
q
which is a contradiction. Thus p
2
q
2
A = 0 unless
p +q

A = 0 which is equivalent to p = q = 0, and the inverse (21) exists for

every non-zero quadratic irrationality (18). This number deserves a special
notation and for u = p + q

A we denote N(u) = p
2
q
2
A.
In relation to the solutions of (17) we will be especially interested in
quadratic irrationalities
z
1
+ z
2

A, z
1
, z
2
Z.
This set of numbers we will denote Z(

A).
5
Proposition 1. Let u, v Z(

A), then u + v, uv Z(

A). If u Z(

A)
and N(u) = 1, then u
1
Z(

A).
Proof. Let us denote u = p q

A, then the formula (21) can be written as

u u = N(u). Thus we have
u
1
=
u
N(u)
= u Z(

A).
Inspecting (20) and replacing q by q and s by s we also get
u v = uv. (22)
(Note the analogy with the complex numbers!) Formulae (21) and (22) also
imply a very important formula
N(uv) = N(u)N(v). (23)
Indeed, we have N(uv) = uvuv = uv u v = u uv v = N(u)N(v).
It is time now to relate these properties of the new function N(x) to the
solutions of (17) and also to the solutions of the equation
x
2
Ay
2
= k, k Z (24).
Proposition 2. A pair of integers (x, y) is a solution to Pells equation (24)
if and only if N(u) = k for u = x + y

A. In particular, a pair of integers

(x, y) is a solution to Pells equation (17) if and only if N(u) = 1.
Proof. As N(u) = x
2
Ay
2
we see that the statement N(u) = 1 is simply
a reformulation of the statement that the pair (x, y) is a solution to the
equation (17).
Theorem 3. Suppose that a pair of integers (a, b) is a solution to Pells
equation (17) and (x, y) is an arbitrary solution to the Diophantine equation
(24). Let us denote u = x + y

A, v = a + b

A, and
uv = (xa + ybA) + (xb + ya)

A = x

+ y

A, (25)
where x

= xa + ybA and y

= xb + ya. Then this pair of integers (x

, y

) is
also a solution to the equation (24).
6
Proof. This follows from the multiplicative property of the norm. Indeed,
N(uv) = N(u)N(v) = 1 k = k.
This theorem gives us a very important tool to obtain a number of so-
lutions of (24) if we know at least one solution of (17) dierent from the
trivial solution (1, 0). We reformulate Theorem 3 now in terms of geometric
transformations of the plane.
Theorem 4. Suppose that a pair of integers (a, b) is a solution to Pells
equation (17). Let us consider a linear transformation of the plane (x, y)
(x

, y

), where
x

= ax + bAy,
y

= bx + ay.
Then this transformation maps the solutions of (24) again onto the solutions
of (24).
It is clear now why the solution (1, 0) of (17) is called trivial. It is because
of the fact that the corresponding linear transformation for a = 1 and b = 0
is simply the identity transformation.
Example 1. Let us consider the equation
x
2
2y
2
= 1. (26)
It has a nontrivial solution (x, y) = (3, 2). Then the following linear trans-
formation
x

= 3x + 4y,
y

= 2x + 3y
will produce more solutions of
x
2
2y
2
= k.
if we know one. For example, we can get some more solutions of (26). Apply-
ing twice the linear transformation to the pair (3, 2) we get two more solutions
of (26):
(3, 2) (17, 12) (99, 70).
Or else the solution (5, 3) to the equation
x
2
2y
2
= 7
gives us another solution of this equation, namely: (5, 3) (27, 19).
7
It is clear now that it is important to prove that the equation (17) always
has a nontrivial solution for every positive integer A which is not the square
of a whole number. We shall start proving this with the following
Lemma 1. Let be an irrational number. Then for every positive integer t
the inequality

p
q

<
1
tq
. (27)
has an integer solution (p, q) such that 1 q t.
Proof. Let [] be the integer part and {} be the fractional part of a real
number , i.e., [] Z and 0 {} < 1. (For example,
_
3
2

= 1, {
3
2
} =
1
2
,
and
_
5

= 2, {

5} =

5 2.) It is always true that = [] +{}.

Let us divide the unit interval [0, 1) (where 0 is included and 1 is not)
into t intervals
_
0,
1
t
_
,
_
1
t
,
2
t
_
, . . . ,
_
t 1
t
, 1
_
(28)
of equal length 1/t. Let us consider t + 1 numbers {k}, k = 1, 2, . . . , t + 1.
By Pigeonhole Principle at least two of them, say {k
1
} and {k
2
}, will be
situated in the same interval of the partition (28) of the unit interval. Thus
|{k
1
} {k
2
}| <
1
t
.
Replacing here {k
i
} by k
i
[k
i
] we get
|k
1
[k
1
] (k
2
[k
2
])| <
1
t
.
or |q p| <
1
t
, where q = k
1
k
2
and p = [k
1
] [k
2
]. Dividing by q we
get (27). It is clear that q (t + 1) 1 = t.
Corollary 1 (Dirichlet). Let be an irrational number. Then the inequal-
ity

p
q

<
1
q
2
. (29)
has innitely many integer solutions (p, q).
8
Proof. As

p
q

<
1
tq

1
q
2
we see that every solution to (27) is also a solution to (29). It is also clear
that as t grows more and more new solutions of (29) will emerge.
Lemma 2. For some integer k such that |k| < 2

A + 1 the equation
x
2
Ay
2
= k
has innitely many integer solutions.
Proof. Let (p, q) be a solution to the inequality (29) with =

A. Then
p
q
<

A + 1 and

p
2
Aq
2

= q
2

A
p
q

A +
p
q

< q
2

1
q
2

A +
p
q

< 2

A + 1.
Therefore |p
2
Aq
2
| can take only nitely many integer values k satisfying
(2

A + 1) < k < 2

A + 1.
As the inequality (29) has innitely many solutions by Pigeonhole Principle
for some k in this range the equation x
2
Ay
2
= k also has innitely many
integer solutions.
Lemma 3. There exist a nonzero integer k and two positive integers 0
a, b < |k| such that the equation (24) has innitely many integer solutions
(x, y) such that x a (mod |k|) and y b (mod |k|).
Proof. Let k be such that the equation (24) has an innite number of solu-
tions. Such k exists according to Lemma 2. We assume that k = 1 and we
consider this case later.
We need Pigeonhole Principle again. For an arbitrary solution (x, y) of
(24) we have x i (mod |k|) and y j (mod |k|) for some 0 i, j |k|.
As we have k possibilities for i and k possibilities for j, in total, we have k
2
possibilities for the pair (i, j). Again we have a nite number of boxes and
innite number of solutions to (24) to go into them. Therefore there will be
an innite number of solutions at least in one of them.
9
Theorem 5. For every positive integer A which is not the square of a whole
number Pells equation (17) has a nontrivial integer solution (a, b) = (1, 0).
Proof. The idea of constructing a solution to Pells equation is as follows.
Let (x
1
, y
1
) and (x
2
, y
2
) be two distinct solutions to (24). This means that
for u
1
= x
1
+y
1

A and u
2
= x
2
+y
2

A we have N(u
1
) = N(u
2
) = k. Since
N(u
1
u
1
2
) = N(u
1
)N(u
2
)
1
= 1, the idea is to consider v = u
1
u
1
2
= a+b

A.
Then N(v) = 1, hence (a, b) is a solution of (17). We now have to take care
of two things: to secure that a and b are integers and to check that this
solution is nontrivial. To deal with the rst problem let us calculate a and b:
v =
_
x
1
+ y
1

A
_
_
x
2
y
2

A
_
k
=
(x
1
x
2
Ay
1
y
2
)
k
+
(x
1
y
2
x
2
y
1
)
k

A,
whence
a =
(x
1
x
2
Ay
1
y
2
)
k
, b =
(x
1
y
2
x
2
y
1
)
k
. (30)
Let (x
1
, y
1
) and (x
2
, y
2
) be two distinct solutions to (24) such that
x
1
x
2
(mod |k|) y
1
y
2
(mod |k|),
which existence is guaranteed by Lemma 3. Then
x
1
x
2
Ay
1
y
2
x
2
1
Ay
2
1
0 (mod |k|),
and
x
1
y
2
x
2
y
1
x
1
y
1
x
1
y
1
0 (mod |k|).
Therefore in (30) a and b are integers.
We need to prove now that the solution obtained is a nontrivial one.
Suppose that (a, b) = (1, 0). In this case v = u
1
u
1
2
= 1 + 0

A = 1 and
u
1
= u
2
. This contradiction completes the proof.
Theorem 6. Let k = 1 be an integer and A be a positive integer which is
not the square of a whole number. Suppose that the equation (24) has at least
one solution. Then it has innitely many solutions.
Proof. Suppose that (x
1
, y
1
) be a solution of (24) and (a, b) be a nontrivial
solution of Pells equation. Let u = x
1
+ y
1

A and v = a + b

A. Then
N(u) = k and N(v) = 1 and by (23)
N(uv
m
) = N(u)N(v)
m
= N(u) = k,
10
i.e., if we write uv
m
= x
m
+ y
m

A, then (x
m
, y
m
) is a solution of (24) for
all m Z. Let us prove that they are all dierent. Suppose that (x
m
, y
m
) =
(x
n
, y
n
) for m < n. Then uv
m
= uv
n
and v
nm
= 1. It follows that v =
1 or a = 1 and b = 0. But the solution (a, b) is nontrivial which is a
contradiction.
We can say much more about the solutions to Pells equation. We need
the following comment.
Lemma 4. Let (x, y) be an integer solution to Pells equation (17) and u =
x + y

A.
1. If x > 0 and y > 0, then u > 1;
2. If x > 0 and y < 0, then 0 < u < 1;
3. If x < 0 and y > 0, then 1 < u < 0;
4. If x < 0 and y < 0, then u < 1;
Proof. The rst statement is clear. If x > 0 and y < 0, then u = x +y

A =
x (y)

A = (x + (y)

A)
1
= v
1
, where v = x + (y)

A > 1 since
x > 0 and y > 0. Hence 0 < u < 1. The third and the forth statements
follow from the rst two.
Denition 1. Let (a, b) be a nontrivial solution to Pells equation (17) with
positive integer components a > 0, b > 0. We say that this solution is minimal
if the number u = a + b

A takes the minimal possible value.

Note that the number u is uniquely determined since a+b

A = a

+b

A
implies (b b

A = a

a and

A is rational unless b = b

and a = a

. Let
us also note that u > 1 by Lemma 4.
Theorem 7. Let (x
1
, y
1
) be the minimal solution to Pells equation (17) and
u = x
1
+ y
1

A. Let
u
n
= x
n
+ y
n

A, n = 0, 1, 2, . . . (31)
Then (x
n
, y
n
), n = 0, 1, 2, . . . , is the complete set of solutions to Pells
equation.
11
Proof. The trivial solution (1, 0) is in this set and we get it for n = 0. Let
(x, y) be an arbitrary nontrivial solution to Pells equation. We may assume
that x > 0. Since (x +y

A)
1
= x y

A, we may also assume that y > 0.

All we need to show is that v = x + y

A can be represented as u
n
for some
positive integer n. Let us assume the contrary. As x > 0 and y > 0, we
know that v > 1. Since u > 1 the terms of the sequence 1, u, u
2
, . . . , u
n
,
. . . get arbitrary large, thus there exists n such that u
n
< v < u
n+1
. Let us
multilply this inequality by (u
n
)
1
. We get
1 < v(u
n
)
1
< u, (32)
where v(u
n
)
1
= x + y

A for some x, y Q. Let us make a number of

observations. Firstly, (u
n
)
1
= (u
1
)
n
= (x
1
y
1

A)
n
. This means that
(u
n
)
1
Z(

A) and hence v(u

n
)
1
Z(

A), i.e., x, y are integers. Sec-

ondly, N(v(u
n
)
1
) = N(v)N(u)
n
= 1 and ( x, y) is a solution to Pells
equation. Thirdly, by Lemma 4 and (32) we get x > 0, y > 0 because of
the inequality 1 < v(u
n
)
1
. Finally, this contradicts to (32), namely to
v(u
n
)
1
< u, since u was minimal. The theorem is proved.
Exercise 2. Suppose that a pair of integers (x
1
, y
1
), x
1
> 0, y
1
> 0, is a
solution to Pells equation x
2
Ay
2
= 1. Then this solution is minimal if and
only if y
1
is minimal among all integers solutions with positive components.
This Exercise gives us an algorithm how to nd the minimal solution. We
have to try subsequently y
1
= 1, 2, . . . until a matching x
1
is found. This
algorithm is not an ecient one. For example, for the equation x
2
109y
2
= 1
the minimal solution (x
1
, y
1
) will have y
1
= 15140424455100. A more ecient
method will be given in exercises below. A better algorithm is beyond the
scope of this lecture.
Corollary 2. Suppose that a pair of integers (x
1
, y
1
) is the minimal solution
to Pells equation (17). Then
x
n
=
1
2
__
x
1
+ y
1

A
_
n
+
_
x
1
y
1

A
_
n
_
,
y
n
=
1
2

A
__
x
1
+ y
1

A
_
n

_
x
1
y
1

A
_
n
_
.
Proof. We note that (22) implies that (u
n
) = u
n
. Thus
x
n
+ y
n

A = u
n
,
x
n
y
n

A = u
n
.
12
Therefore
x
n
=
1
2
(u
n
+ u
n
),
y
n
=
1
2

A
(u
n
u
n
).
and this is exactly what we wanted to prove.
Note that as 0 < u < 1 by Lemma 4 the second term u
n
0 as n gets
large.
My last word on this topic is to show how Pells equation relates to the
general solution of quadratic Diophantine equations. As an example, I will
analize the Diophantine equation
(y
2
x
2
) + xy = 1. (33)
At rst let us have a look at this equation using the common sense. We see
that (1, 1) is clearly a solution to it. If you are lucky you may well notice a
linear transformation which produces new solutions to (33) out of existing
ones. One of them is:
x

= 2x + y
y

= x + y.
Indeed, it is easy to check that
(2x + y)
2
+ (x + y)
2
+ (2x + y)(x + y) = x
2
+ y
2
+ xy.
Starting from (1, 1) and applying this linear transformation many times we
may nd innitely many solutions to (33):
(1, 1) (3, 2) (8, 5) (21, 13) . . .
By induction we may prove that (f
2n
, f
2n1
), where f
i
is the ith Fibonacci
number is a solution to (33).
What if we are not so lucky? Then we may write (33) as a quadratic
equation in x
x
2
xy y
2
+ 1 = 0
13
and then solve it assuming that y is known. We will get
x
1,2
=
1
2
_
y
_
5y
2
4
_
.
Now we need a sucient supply of ys with 5y
2
4 being a square of an
integer. Thus y must be a solution to Pells equation
z
2
5y
2
= 4. (34)
Then x will be computed as x = 1/2(yz). Of course, we should keep an eye
on x because we need it to be an integer. The pair (z
0
, y
0
) = (1, 1) is clearly
a solution to (34) and we know now how to get innitely many solutions to
it. As (9, 4) is the minimal solution to z
2
5y
2
= 1 the linear transformation
x

= 9x + 20y
y

= 4x + 9y
will produce new solutions to (34) out of existing ones. We get
(1, 1) (29, 13) (521, 233) . . .
Since we started with two odd integers, by induction it is easy to see that all
pairs will contain two odd integers. This gives us the integer solutions
(1, 1) (21, 13) (377, 233) . . .
to the equation (33). Note that if we started with (z
0
, y
0
) = (1, 1) we would
get another sequence of solutions.
3 Problems to solve
1. Find all solutions to the Diophantine equation x
2
+ 2y
2
= z
2
.
2. Prove that the equation 2x
2
+3y
2
= z
2
does not have integer solutions.
3. Prove that for any Pythagorass triple (x, y, z) one of the numbers x or
y is divisible by 3 and one of them is divisible by 4.
14
4. Prove that for any Pythagorass triple (x, y, z) one of the numbers x, y
or z is divisible by 5.
5. (Croatia, 1997) Let x, y, z, a, b, c be integers such that
x
2
+ y
2
= a
2
x
2
+ z
2
= b
2
y
2
+ z
2
= c
2
.
Prove that the number xyz is divisible by 55.
6. (Diophantus) A given number, which is a sum of two squares, say
13 = 2
2
+ 3
2
, split into a sum of two other squares. (They must bee
rational as irrational numbers were not known to Diophantus.)
7. Let
F(x, y) = ax
2
+ bxy + cy
2
+ dx + ey + f, a, b, c, d, e, f Z.
Suppose that the equation F(x, y) = 0 has at least one rational solu-
tion. Then this equation has an innite number of rational solutions.
8. Given that the Diophantine equation
x
2
+ y
2
az
2
au
2
= 0,
where a is an integer, has a solution (x
0
, y
0
, z
0
, u
0
) such that x
2
0
+y
2
0
+
z
2
0
+ u
2
0
= 0, prove that the Diophantine equation
x
2
+ y
2
az
2
= 0
has a solution (x
0
, y
0
, z
0
) such that x
2
0
+ y
2
0
+ z
2
0
= 0.
9. (GDR, 81) Prove that the equation
x
2
1
+ x
2
2
+ + x
2
n
= y
2
has an integer solution for every positive integer n.
10. (UK, 70) For each positive integer n, let us denote by a
n
the number
of integer solutions (x, y) to the equation
n
2
+ x
2
= y
2
such that x n, y n.
15
(a) Prove that, for every number M, the inequality a
n
> M is satised
for at least one positive integer n.
(b) Is it true that lim
n
a
n
= ?
11. (Turkey, 1997) Prove that, for each prime number p 7, there exist a
positive integer n and integers x
1
, x
2
, . . . , x
n
, y
1
, y
2
, . . . , y
n
which are
not divisible by p and such that
x
2
1
+ y
2
1
x
2
2
(mod p)
x
2
2
+ y
2
2
x
2
3
(mod p)
.
.
.
x
2
n1
+ y
2
n1
x
2
n
(mod p)
x
2
n
+ y
2
n
x
2
1
(mod p).
12. (USA, 76) Prove that the equation
x
2
+ y
2
+ z
2
= x
2
y
2
has the only integer solution x = y = z = 0.
13. (USA, 1979) Find all integer solutions to the equation
x
4
1
+ x
4
2
+ . . . + x
4
14
= 1599.
14. (Hungary, 83) Prove that the equation
x
3
+ 3y
3
+ 9z
3
9xyz = 0
has the only rational solution x = y = z = 0.
15. Prove that for every positive integer n > 1 the equation
x
n
+ y
n
= z
n
does not have integer solutions (x, y, z) such that x n and y n.
16. (IMO, 81) Determine the maximum value of m
2
+n
2
, where m 1981
and n 1981 are positive integers satisfying
(n
2
mn m
2
)
2
= 1.
16
17. Show that each positive integer n can be represented in the form
n =
a
2
+ b
a + b
2
,
where a, b are positive integers.
18. (IMO, 84) Prove that if n is a positive integer such that the equation
x
3
3xy
2
+ y
3
= n
has a solution in integers (x, y), then it has at least three such solutions.
Show that the equation has no solutions in integers when n = 2891.
19. For each positive integer k, let M(k) denote the set of numbers repre-
sentable as
x
2
+ y
2
+ k
xy
,
where x, y are positive integers. Prove that each of the sets M(k)
contains only nitely many integers.
Solutions
16. The solution of the Diophantine equation
(n
2
mn m
2
)
2
= 1 (1)
is the union of the solutions of the two quadratic Diophantine equations
n
2
mn m
2
= 1, n
2
mn m
2
= 1, (2)
thus we can expect that some linear transformation exists which pro-
duces new solutions out of existing ones. Such a transformation is easy
to nd, e.g.,
T
_
x
y
_
=
_
y
x y
_
=
_
0 1
1 1
__
x
y
_
. (3)
Indeed, let (n, m) be a solution to (1) in positive integers. Then n(n
m) = m
2
1 0 and n > m unless n = m = 1. If (n, m) = (1, 1)
17
we set n
1
= m, and m
1
= n m. Then n
1
and m
1
are again positive
integers and
n
2
1
m
1
n
1
m
2
1
= m
2
(nm)m(nm)
2
= n
2
+ mn + m
2
= (n
2
mn m
2
).
Thus, with the exception of the trivial solution (1, 1), the linear trans-
formation (3) converts the solutions of the rst equation (2) into the
solutions of the second equation (2) and vice versa. Since for every
solution (n, m) to (1) we have gcd nm = 1, and T being applied to
(n, m) makes the maximal number of the pair smaller, eventually for
some k we will get
T
k
_
n
m
_
=
_
1
1
_
. (4)
The linear transformation T has the inverse transformation
S
_
x
y
_
=
_
x + y
x
_
=
_
1 1
1 0
__
x
y
_
. (5)
Let us check how the linear transformation S transforms a solution
(n, m) to the equation (1). We set n
1
= n + m, and m
1
= n. Then n
1
and m
1
are again positive integers and
n
2
1
m
1
n
1
m
2
1
= (n+m)
2
(n+m)nn
2
= n
2
+mn+m
2
= (n
2
mnm
2
).
We see that the linear transformation (5) converts the solutions of the
rst equation (2) into the solutions of the second equation (2) and vice
versa with no exceptions this time. And equation (4) shows that an
arbitrary solution (n, m) to (1) can be obtained as
_
n
m
_
= S
k
_
1
1
_
. (6)
An easy induction shows that
_
n
m
_
= S
k
_
1
1
_
=
_
f
n+2
f
n+1
_
,
where f
n
is the nth Fibonacci number.
The largest pair of Fibonacci numbers not exceeding 1981 is (1597, 987);
so the maximal value of n
2
+ m
2
is 1597
2
+ 987
2
.
18
17. The number n = 1 can be represented as required by taking a = b. We
assume that n > 1. Let us note that
n =
a + b
2
a
2
+ b
, (1)
is equivalent to
a(a n) = b(bn 1). (2)
Suppose gcd ab = c with a = xc and b = yc, where gcd xy = 1. Then
(2) is equivalent to
x(xc n) = y(nyc 1), (3)
which, in turn, is equivalent to
xc n = yd and xd = nyc 1. (4)
or
xc yd = n
nyc xd = 1
Solving this system for c and d we have
c =
nx y
x
2
ny
2
, d =
n
y
x
x
2
ny
2
. (5)
If relatively prime integers x and y can be chosen so that c and d in
(5) are positive integers, then for a = xc and b = yc the equation (1)
will be satised.
(1) If n is not a square of an integer, then the Pells equation x
2
ny
2
=
1 has a solution (x, y) in positive integers. It is clear that for these
integers we have gcd xy = 1. Now c = nx y and d = n
2
y x. We
note that
c = nx y = n
_
ny
2
+ 1 y > n

ny y = (n

n 1)y > 0,
and
d = n
2
y x =
_
n
4
y
2

_
ny
2
+ 1
_
n
4
y
2

_
(n + 1)y
2
> 0
19
as n
4
> n + 1 for n > 1. Thus c and d are positive integers.
(2) Let n = m
2
be the square of an integer. Then a and b can be
constructed as polynomials in m. If m is even, the numbers
(a, b) =
_
m
5
4
+
m
2
2
,
m
4
4
_
satisfy (1). If m is odd, then the numbers must be
(a, b) =
_
(m
2
m + 2)(m
2
+ 1)
4
,
(m1)(m
2
+ 1)
4
_
.
18. The equation
x
3
3xy
2
+ y
3
= n. (1)
is not a quadratic Diophantine equation and we cannot expect to nd
a linear transformation which produces an innite number of solutions.
But linear transformation still exists! We may write
x
3
3xy
2
+ y
3
= (y x)
3
3(y x)x
2
+ (x)
3
.
Hence, if pair of integers (x, y) is a solution to (1), then so is pair
(y x, x). Thus the linear transformation is
T
_
x
y
_
=
_
y x
x
_
=
_
1 1
1 0
__
x
y
_
.
We may try to produce more solutions by repeated application of T:
T
2
_
x
y
_
=
_
1 1
1 0
_
2
_
x
y
_
=
_
0 1
1 1
__
x
y
_
=
_
y
x y
_
.
Solutions (x, y), (y x, x), (y, x y) are clearly distinct. Hence
the rst statement is proved. Those who are tempted to get even more
solutions, will be disappointed. As T
3
is the identity matrix, applying
T for the third time we will get back to solution (x, y).
Let us consider now equation (1) mod 3
x
3
+ y
3
= 2 (mod 3). (3)
20
There are three possibilities to satisfy this congruence: (a) x 2 (mod 3),
y 0 (mod 3); (b) x 0 (mod 3), y 2 (mod 3); (c) x 1 (mod 3),
y 1 (mod 3). All these cases do not hold mod 9. For example, if
x = 3m + 2 and y = 3n, then
x
3
3xy
2
+y
3
= (3m+2)
3
3(3m+2)(3n)
2
+(3n)
3
(2)
3
8 (mod 9)
while 2891 = 2 (mod 9) and (a) does not hold.
19. One of the integers with the property is clearly n = 1. Let us assume
that n > 1 satises the property, that is
2
n
+ 1
n
2
is an integer. Since
2
n
+1 is odd and n
2
| 2
n
+1, it follows that n is odd. Let p be the smallest
prime factor of n. The p 3 and p | 2
n
+ 1, that is 2
n
1 (mod p).
Let m be the smallest positive integer such that 2
m
1 (mod p). By
Little Fermat theorem 2
p1
1 (mod p), whence m < p1. Applying
the division algorithm we write n = km+r, where 0 r < m. The we
have
2
n
= 2
km
2
r
(1)
k
2
r
(mod p).
Suppose that k is even. Then 2
r
2
n
1 (mod p) and because of
minimality m we have r = 0 and 2 0 (mod p). This contradicts
to p 3. Thus k is odd and 2
r
1 (mod p). We claim that r = 0
anyway. Suppose the contrary, i.e., r > 0. Then 2
mr
2
mr
2
r

2
m
1 (mod p), which contradicts to the choice of m. Hence n = km
and m| n. Since m < p and p is the smallest prime factor, it follows
that m = 1. Thus 2 1 (mod p), whence p = 3.
Let n = 3
k
d, where k 1 and gcd d3 = 1. We shall prove that k = 1.
Suppose k 2. Since n
2
| 2
n
+ 1, we have 3
2k
| (3 1)
n
+ 1. Since
k + 2 2k,
3
k+2
| 3n
n

i=2
(1)
i
_
n
i
_
3
i
. (1)
The exponent of 3 in the decomposition of 3n = 3
k+1
d is exactly k +1.
The exponent of 3 in the decomposition of i! is
_
i
3
_
+
_
i
3
2
_
+
_
i
3
3
_
+
i
3
+
i
3
2
+
i
3
3
+ <
i
2
.
21
Thus the exponent of 3 in the decomposition of
_
n
i
_
3
i
=
(n i + 1)(n i + 2) . . . n
i!
is greater than k+
h
2
, i.e., greater than or equal to k+2. This contradicts
to (1). Hence n = 3d.
We claim that n = 3. Suppose that d > 1. Let q be the smallest prime
factor of d. Since d is odd and gcd d3 = 1, we have q 5. Also q | 2
n
+1
and 2
n
1 (mod q). Let j be the smallest positive integer such that
2
j
1 (mod q). An argument similar to that in the rst paragraph
gives j | n. By Little Fermats theorem j < q1. Since q is the smallest
prime factor of d, it follows that j = 1 or j = 3. Thus the comgruence
2
j
1 (mod q) yields that q | 3 or q | 9. But this contradicts the fact
that q 5.
Summarizing, we have proved that for n 3 the only possible case
when n
2
| 2
n
+ 1 is n = 3.
Finally, it is easy to verify that n
2
| 2
n
+ 1 for n = 1 and n = 3.
22