Vous êtes sur la page 1sur 5

MINKOWSKIS THEOREM AND REPRESENTATIONS OF INTEGERS AS

BINARY QUADRATIC FORMS


JOSEPH STAHL
1. Introduction
The well known theorem of Minkowski in the geometry of numbers states that given a lattice /
with determinant d(/) and a point set o that is convex and symmetric about the origin with volume
V (o) (both living in R
n
), if o satises
V (o) > 2
n
d(/)
then o contains a point in the lattice other than the origin. This theorem relates geometric properties
of the set (convexity, symmetry, and volume) to an arithemetical property of o (the existence of
a point belonging to /). In this paper, we aim to prove Minkowskis theorem and discuss some of
its applications to representations of integers as quadratic forms. For example, we shall show the
result that an odd prime p can be represented in the form x
2
+ y
2
, x and y integers, if and only if
p 1 mod 4.
2. Proof of Minkowskis theorem
To begin, we shall dene some notation and terminology.
Denition 1. Let o R
n
be a point set in n-dimensional euclidean space. Then o is convex if
x
1
, x
2
o = tx
1
+ (1 t)x
2
o, (0 t 1)
That is to say, a set o is convex if when it contains 2 points it contains the entire line segment
between them.
Denition 2. Let a
1
, . . ., a
n
be linearly independent vectors in R
n
(that is, the only set of numbers
t
1
, . . ., t
n
, for which t
1
a
1
+ . . . + t
n
a
n
= 0 is t
1
= . . . = t
n
= 0). The set
/ =
_
t
1
a
1
+ . . . + t
n
a
n

t
k
Z k
_
is called the lattice with basis a
1
, . . ., a
n
.
It is useful to note that a lattice can also be characterized in the following way:
Denition 3. A lattice / is a set of points in n-dimensional space which
contains n linearly independent vectors
is a group under addition
has only the origin in some sphere x
2
1
+ . . . + x
2
n
< r
2
, where r > 0.
Obviously, every lattice satises these properties, but it is harder to see that every set satisfying
these properties is a lattice; however, we shall not prove that the two denitions are equivalent here.
Denition 4. Let / be a lattice with basis a
1
, . . ., a
n
. The determinant of /, denoted d(/), is
given by
d(/) = [det(a
1
, . . . , a
n
)[
where det(a
1
, . . . , a
n
) is the determinant of the n n array whose j-th row is the vector a
j
.
1
2 JOSEPH STAHL
Notice that d(/) represents the n-dimensional volume taken up by the set
T
A
=
_
t
1
a
1
+ . . . + t
n
a
n

0 t
k
1
_
We shall call T
A
the fundamental parallelopiped of the lattice /. From the denition of d(/), one
can see that if we dene
C/ :=
_
t
1
Ca
1
+ . . . + t
n
Ca
n

t
k
Z k
_
for some xed C R, then
d(C/) = C
n
d(/)
Now, we shall prove Minkowskis Theorem in 2 dimensions for clarity, and then generalize this
proof to n dimensions. To do so, we shall restrict the set o modulo the lattice 2/. We note that
this restriction does not preserve area, because V (o) > d(2/) = 4d(/) by the hypothesis. From this
point, we use symmetry and convexity to construct a point that is a member of / that belongs to
o.
Theorem 1. Minkowskis Theorem in two dimensions: Given a lattice / R
2
, with determinant
d(/) and a point set o R
2
that is convex and symmetric about the origin and has area A(o), if o
satises
A(o) > 4d(/)
then o contains a point of / other than the origin.
Proof. Let / be a lattice with determinant d(/) and basis l
1
, l
2
, and let o R
2
be convex and
symmetric about the origin, and suppose that A(o) > 4d(/). Dene 2/ to be the lattice with basis
2l
1
, 2l
2
, and notice that d(2/) = 4d(/). Dene f : o R
2
/2/ by f : s s mod 2/. That is,
f takes points in o and restricts them modulo 2/ (i.e., to the fundamental parallelopiped of 2/,
T
2L
). By this restriction, we see that A(f(o)) A(T
2L
) = d(2/) = 4d(/). But, by hypothesis,
A(o) > 4d(/), so f is not injective. Thus, there exist p
1
, p
2
o, p
1
,= p
2
, such that f(p
1
) = f(p
2
).
This means that p
1
p
2
mod 2/, or equivalently, p
2
= p
1
+ 2nl
1
+ 2ml
2
, for some n, m Z.
Furthermore, we note that n and m are not both 0, because p
1
and p
2
are distinct. Since o is
symmetric about the origin, if x o, then x o. So, we can say that p
1
o. Convexity tells
us that if two points are in o, then the line segment between them is contained within o as well.
Then we have
1
2
(p
1
+ p
2
) =
1
2
(p
1
+ p
1
+ 2nl
1
+ 2ml
2
)
= (nl
1
+ ml
2
) o
This is clearly a point of the lattice /, because l
1
, l
2
is a basis for /. Since n and m are not both
0, the point nl
1
+ ml
2
,= 0, and thus the theorem is proved.
We may note in fact, that this theorem proves that under the hypothesis, o contains at least
3 distinct points of /: the origin, the point nl
1
+ ml
2
, and the point nl
1
ml
2
, because o is
symmetric about the origin. To prove the theorem in n dimensions, we will use a generalization of
the proof presented for 2 dimensions.
Theorem 2. Minkowskis Theorem in n dimensions: Given a lattice / R
n
with determinant d(/)
and a point set o R
n
that is convex and symmetric about the origin and has volume V (o), if o
satises
V (o) > 2
n
d(/)
then o contains a point of / other than the origin.
MINKOWSKIS THEOREM AND REPRESENTATIONS OF INTEGERS AS BINARY QUADRATIC FORMS 3
Proof. Let / be a lattice with determinant d(/) and basis l
1
, . . . , l
n
, and let o R
n
be convex
and symmetric about the origin, and suppose that V (o) > 2
n
d(/). Dene 2/ to be the lattice with
basis 2l
1
, . . . , 2l
n
, and notice that d(2/) = 2
n
d(/). Dene f : o R
n
/2/ by f : s s mod 2/.
That is, f takes points in o and restricts them modulo 2/ (i.e., to the fundamental parallelopiped
of 2/, T
2L
). By this restriction, we see that V (f(o)) V (T
2L
) = d(2/) = 2
n
d(/). But, by
hypothesis, V (o) > 2
n
d(/), so f is not injective. Thus, there exist p
1
, p
2
o, p
1
,= p
2
, such that
f(p
1
) = f(p
2
). This means that p
1
p
2
mod 2/, or equivalently, p
2
= p
1
+ 2m
1
l
1
+ . . . + 2m
n
l
n
,
for some m
1
, . . . , m
n
Z. Furthermore, we note that not all the m
k
are 0, because p
1
and p
2
are
distinct. Since o is symmetric about the origin, if x o, then x o. So, we can say that p
1
o.
Convexity tells us that if two points are in o, then the line segment between them is contained within
o as well. Then we have
1
2
(p
1
+ p
2
) =
1
2
(p
1
+ p
1
+ 2m
1
l
1
+ . . . + 2m
n
l
n
)
= (m
1
l
1
+ . . . + m
n
l
n
) o
This is clearly a point of the lattice /, because l
1
, . . . , l
n
is a basis for /. Since not all the m
k
are
0, the point m
1
l
1
+ . . . + m
n
l
n
,= 0, and thus the theorem is proved.
3. Quadratic Forms
Next, let us introduce some denitions that will be used in the following sections.
Denition 5. Let R be a commutative ring with multiplicative identity 1. Then a homogeneous
polynomial of degree 2 in R[X
1
, . . . , X
n
] is a quadratic form over R in n indeterminates.
Specically, we will look at binary quadratic forms, meaning quadratic forms over Z in 2 indetermi-
nates. It is from here that problems such as which integers can be represented as the sum of two
squares? stem.
Denition 6. Let Q(X, Y ) = aX
2
+bXY +cY
2
be a binary quadratic form. Then the discriminant
of Q is dened to be (Q) = b
2
4ac
Denition 7. A quadratic form Q = aX
2
+ bXY + cY
2
is said to be positive denite if (Q) < 0
and a > 0.
In this discussion, we shall only be concerned with positive denite binary quadratic forms, as these
are the easiest to understand.
Denition 8. A ring R is called a Euclidean domain if there exists : R 0 Z
0
such that for
all a, b R with b ,= 0, there exist q, r R such that a = bq + r and either r = 0 or (r) < (b)
We will use these denitions to prove that an odd prime p can be written as the sum of two squares
if and only if p is equivalent to 1 mod 4.
4. Sums of Two Squares
lemma 1. Let p be an odd prime. Then p 1 mod 4 i 1 x
2
mod p
Proof. Suppose 1 u
2
(p), so 1 = u
2
in U
p
. Then u
4
= 1. We can note that the powers of u are
as follows: 1, u, u
2
, u
3
, 1, . . ., where u, u
2
, and u
3
are not 1. It is easily seen that these powers are
not 1, because if u were 1, then u
n
would be 1 for all n. But, u
2
1. The only element in U
p
that
would make u
3
1 is 1, and u , 1, so u, u
2
, and u
3
are all distinct from 1. So, u
p1
= 1 implies
that 4[p 1. To prove the reverse implication, we assume p 1 mod 4. By Wilsons theorem,
(p 1)! 1 (p). But
(p 1)! = 1 2 3 . . .
_
p 1
2
_

_
p + 1
2
_
. . . (p 1) = 1 (p)
The second half of this product can also be written
_
p + 1
2
_
. . . (p 1) =
_
p 1
2
_

_
p 3
2
_
. . . (1)
4 JOSEPH STAHL
Combining the original representation, we have
1 2 3 . . .
_
p 1
2
_

_
p + 1
2
_
. . .
_
p 1
2
_

_
p 3
2
_
. . . (1) = (1)
p1
2
__
p 1
2
_
!
_
2
= 1
However, the only time (1)
p1
2
__
p1
2
_
!
_
2
is a square is when p 1 mod 4, which proves the
lemma.
Using this lemma, we shall provide two proofs, one geometric, one algebraic, of the following
theorem.
Theorem 3. If p 1 (4) is an odd prime, then there exist x, y Z such that p = x
2
+ y
2
.
Proof. Let p 1 (4). By the lemma, there exists a Z such that a
2
1 (p). Thus, we see that
p[(a
2
+ 1) in Z. So, p[(a + i)(a i) in Z[i]. Since p does not divide a + i or a i, p is not prime
in Z[i]. Since Z[i] is a Euclidean domain, p is not irreducible in Z[i]. Then we have p = for
some non unit , in Z[i]. Since p is real, we have p
2
= N()N(). Since N() and N() are
greater than one and the norms must divide p
2
, we have N() = N() = p. But if = a +bi, then
N() = a
2
+ b
2
= p, and we are done.
Proof 2
Proof. Let p 1 (4). By the lemma, there exists a Z such that a
2
1 (p). Consider the set
/ =
_
(x, y) Z
2

y ax (p)
_
. It is not dicult to verify that / is a lattice. Now we shall make a
few observations about the lattice which will help us apply Minkowskis theorem.
1. For all (x, y) /, x
2
+ y
2
0 (p). 2. (1, a) /, (0, p) / forms a basis for the lattice. To
see this, we consider a point (x, y) /. (x, y) = x(1, a) +
_
yax
p
_
(0, p), so (x, y) is an integral
combination of the basis vectors. Now, notice that d(/) = p. Let =
_
(x, y)

x
2
+ y
2
< 2p
_
.
Then A() = 2p > 4p = 4d(/). is obviously convex and symmetric about the origin, as it is
the interior of a circle. Thus, by Minkowskis theorem, there is a point in that belongs to /. But
points in / satisfy x
2
+ y
2
0 (p) and 0 < x
2
+ y
2
< 2p. Since the only number between 0 and 2p
that divides p is p, x
2
+ y
2
= p.
It is not hard to see that this theorem can be extended to an if and only if statement. Simply note
the squares mod 4 are 0 and 1, and the sums of any two of these will only give 0, 1, and 2. However,
odd primes can only be equivalent to 1 mod 4. From this point, one can see that the only way a
positive prime p can be represented as the sum of two squares is if p 1 mod 4 or p = 1
2
+ 1
2
= 2.
5. Binary Quadratic Forms of the Type X
2
+ nY
2
However, the previous methods do not apply to all quadratic forms. For example, we can look at
the form X
2
+ 5Y
2
.
We are looking for primes such that X
2
+5Y
2
= p. To apply the previous methods, we shall assume
there exists a such that a
2
5 (p). Proceeding in the spirit of Minkowski, we consider the set
/ =
_
(x, y)

y ax (p)
_
. Again, / is a lattice and d(/) = p. Dene =
_
(x, y)

x
2
+ 5y
2
< 2p
_
.
Then A() =
2p

5
. However,

5
< 2, so Minkowskis theorem does not apply.
Perhaps we can apply the method of Euclidean domains to show that if there is an a such that
a
2
5 (p), then there exist X, Y Z such that X
2
+ 5Y
2
= p. Consider , Z[

5]. We
must check whether or not Z[

5] is a Euclidean domain. Visualizing multiples of Z[

5] as
a lattice, we can see that the basis is given by ,

5. Then in any parallelogram of this lattice,


the farthest a point in the plane can be from a lattice point is half the diagonal of the parallelogram,
which is

6||
2
=

2
[[. So
[r[

2
[[
N(r)
3
2
N()
MINKOWSKIS THEOREM AND REPRESENTATIONS OF INTEGERS AS BINARY QUADRATIC FORMS 5
So if Z[

5] is a Euclidean domain, the norm function cannot be used as . However, we can


show that Z[

5] is not a Euclidean domain. Consider 2, which is irreducible (one may check


this by inspection of norms of possible factors of 2). For if 2 = for some , Z[

5], then
4 = N()N() = N() = 2 But = a + b

5, so N() = a
2
+ 5b
2
= 2, which cannot happen
for any a, b Z. Thus, 2 is not irreducible. But 2[(1 +

5)(1

5) and 2 does not divide either


of the two terms, so 2 is not prime. In a Euclidean domain, however, irreducible implies prime, so
Z[

5] cannot be a Euclidean domain, and thus there is no that will x our second method.
In general, we are now interested in quadratic forms of the type X
2
+nY
2
, where n is a positive
integer. These types of quadratic forms have discriminant given by = 4n, and are thus positive
denite. Specically, we are interested in whether or not we can apply the method of Minkowski to
a form of this type for a given n and also whether or not we can apply the Euclidean domain method
for that n. Aside from the assumptions, the key ingredient required for the Minkowski method to
hold is that the area of the set =
_
(x, y)

x
2
+ ny
2
< 2p
_
be greater than 4p, the determinant
of the lattice / =
_
(x, y)

y ax (p)
_
, where a
2
n (p). But A() =
2p

n
. If n = 2, then the
method will work, because

2
> 2. However, if n = 3, the method fails, since

3
< 2.
It makes sense to consider the key requirement for the other method as well. In that case, we require
that Z[

n] be a Euclidean domain. Using the norm as our , we can determine when it fails to be
a Euclidean domain. For Z[

n] to be a Euclidean domain, we require that given , Z[

n]
with ,= 0, there exist q, r Z[

n] such that = q + r, where N(r) < N(). As we have


seen, Z[

n] can be represented as a lattice with a fundamental parallelogram that has one side
of length [[ and one side of length

n[[. Then the farthest r can be from a point of the lattice
is half the diagonal of the parallelogram, which is

n+1
2
[[. So, N(r)
n+1
4
N(). For the norm
to be a Euclidean function for Z[

n],
n+1
4
must satisfy
n+1
4
< 1, from which we can see that the
n Z
+
that satisfy this condition are precisely 1 and 2. For these n, the discriminant of the form is
(X
2
+ nY
2
) = 4n. Applying Minkowskis theorem or using the method of Euclidean domains,
we can prove that X
2
+ 2Y
2
= p for those p such that 2 a
2
(p). As it turns out, Z[

3] is not
a Euclidean domain, for 4 = 2 2 = (1 +

3)(1

3), but 2 and 1

3 are non-associate
primes in Z[

3]. For the forms X


2
+ nY
2
with n 3, a dierent method will be needed to nd
out whether or not it represents a given prime p. Now, we can generalize and begin to look at forms
beyond X
2
+nY
2
, and more specically attempt to solve this related problem: nd all rings R C
that are also lattices.

Vous aimerez peut-être aussi