70 The Geometry of Numbers

x
0
C´

Figure 5.4
The M -set C contracted to C0 .

Proof. We consider the M -set C with center at (0, 0) and area A > 4. We
contract C by mapping every point (x, y) of C into the point (x/2, y/2);
see Figure 5.4. This contraction yields an M -set C 0 similar to C and such
that lengths of segments in C 0 are just half of the lengths of corresponding
Copyright © 1962. American Mathematical Society. All rights reserved.

segments in C. Hence the Area A0 of C 0 satisfies

 2
0 1 1
A = A = A,
2 4

and since A > 4, A0 > 1.

Next we place a replica of C0 on every lattice point of Λ. In other
words, we translate the M -set C 0 from the origin to every point (p, q) with
integer coordinates. Thus, if (x0 , y0 ) is a point of C 0 , then (x0 + p, y0 + q)
is the corresponding point of the translate of C 0 centered at (p, q).
Examine these translates in Figure 5.5.
Do they appear to overlap? Do some points in the plane seem to
belong to more than one translate?
We shall show that these translates indeed overlap and that, as a
consequence, our original M -set C must have contained a lattice point
other than the origin.

Olds, C., et al. Geometry of Numbers, American Mathematical Society, 1962. ProQuest Ebook Central,
http://ebookcentral.proquest.com/lib/princeton/detail.action?docID=3330416.
Created from princeton on 2024-04-18 00:02:58.
 Minkowski's Fundamental Theorem 71

(–s, n + s) (0, n + s) (n + s, n + s)

(n, n)
(0,n)

(0, 2)

(0, 1)

(n, 0)
s
(–s, 0) (n + s, 0)
0 (1, 0) (2, 0)

Figure 5.5
The translates of C 0 overlap.
Copyright © 1962. American Mathematical Society. All rights reserved.

We begin by considering the square in Figure 5.5 with vertices (0, 0),
(n, 0), (n, n), (0, n), where n is some positive integer. The square has
(n + 1)2 lattice points in its interior and on its boundary. The sum of the
areas of the (n + 1)2 translates of C 0 centered on these lattice points is
(n + 1)2 A0 .
Let s be the maximum distance of any point of C0 from the center
(0, 0). Then all these (n + 1)2 M -sets are contained in a square of side
n + 2s and of area (n + 2s)2 . Again, see Figure 5.5. As we shall show,
the sum of the areas of the (n + 1)2 translates of C 0 exceeds the area of
the square containing them; that is, we shall show that

(n + 1)2 A0 > (n + 2s)2 . (5.2)

From this, we will conclude that the translates must overlap.

Olds, C., et al. Geometry of Numbers, American Mathematical Society, 1962. ProQuest Ebook Central,
http://ebookcentral.proquest.com/lib/princeton/detail.action?docID=3330416.
Created from princeton on 2024-04-18 00:02:58.
 72 The Geometry of Numbers

( p2 , q2 )

( p0 + p2 - p1, q0 + q2 - q1 ) ( p1, q1 )
Figure 5.6
Translates of overlapping sets.

( p0 , q0 )

To prove inequality (5.2), we subtract (n + 2s)2 from both sides and

prove the equivalent inequality

(n+1)2 A0 −(n+2s)2 = (A0 −1)n2 +2n(A0 −2s)+A0 −4s2 > 0. (5.3)

Since A0 > 1, the leading coefficient of the quadratic expression in n is
positive, and the entire expression (5.3) is therefore positive for sufficiently
large n. If we choose n so large that (5.3) is positive, then (n + 1)2 A0 >
Copyright © 1962. American Mathematical Society. All rights reserved.

(n + 2s)2 and the translates of C 0 overlap in the square.

Do you see why every translate of C0 has points in common with
some other translate? Consider the two overlapping sets centered at (p1 , q1)
and (p2 , q2) as in Figure 5.6. A third translate centered at any lattice

(p, q) C²

(a ², b ² ) Figure 5.7
Common points between C0 and its
translate C 00 .
(0, 0)
C¢
(a¢, b¢)

Olds, C., et al. Geometry of Numbers, American Mathematical Society, 1962. ProQuest Ebook Central,
http://ebookcentral.proquest.com/lib/princeton/detail.action?docID=3330416.
Created from princeton on 2024-04-18 00:02:58.
 Minkowski's Fundamental Theorem 73

point (p0 , q0 ) must then have points in common with a fourth centered at
(p0 + p2 − p1 , q0 + q2 − q1 ). In particular, C 0 itself, centered at (0, 0),
has points in common with the translate C 00 centered at (p, q), where
p = p2 − p1 , q = q2 − q1 ; see Figure 5.7. Thus every point (x00 , y00 ) of C 00
may be written

x00 = x0 + p, y00 = y0 + q,
where (x0 , y0 ) is the corresponding point of C 0 . Hence, if two translates
overlap, then every translate must overlap at least one other.
Now, consider any point (a00 , b00 ) that is in C 0 and also in C 00 . A point
(a , b ) must exist in C 0 such that
0 0

a00 = a0 + p, b00 = b0 + q. (5.4)

Because C 0 is symmetric, C 0 also contains the point (−a0 , −b0 ). And be-
cause C 0 is convex, it contains the midpoint of the segment connecting any
two of its points. The midpoint connecting points (a00 , b00) and (−a0 , −b0 )
has coordinates
 00
a − a0 b00 − b0

, ,
2 2
which by (5.4) is the point (p/2, q/2). Therefore C 0 contains the point
(p/2, q/2), where (p, q) are the integers associated with the center of C00 .
It follows that the original M -set C contains the lattice point (p, q).
Is our proof complete? Not quite, because we assumed that the area
Copyright © 1962. American Mathematical Society. All rights reserved.

of C was greater than 4, whereas Theorem 5.1 asserts that C contains a

lattice point in its interior or on its boundary even if A = 4.
So, letting A = 4, suppose that C has no lattice points other than
(0, 0) either inside or on its boundary. Then every lattice point is farther
than some positive distance δ from any point of C. We now expand C
so slightly that all the points of the resulting enlarged M -set C ∗ are still
at least δ/2 away from the nearest lattice point. Now C∗ has area greater
than 4, but it still contains no lattice points except (0, 0). However, this
contradicts the results just obtained. Thus C must have had a lattice point
other than the origin either in its interior or on its boundary. This completes
the proof of Minkowski's Fundamental Theorem.

What we have presented is essentially one of Minkowski's own orig-

inal proofs. Other proofs exist; see Koksma's list [4] and, for one of the
most interesting proofs, by Hajos [2], see Hardy and Wright [3]. The proof
by Mordell [8] is also recommended.

Olds, C., et al. Geometry of Numbers, American Mathematical Society, 1962. ProQuest Ebook Central,
http://ebookcentral.proquest.com/lib/princeton/detail.action?docID=3330416.
Created from princeton on 2024-04-18 00:02:58.
 74 The Geometry of Numbers

Problem Set for Section 5.3

1. Consider an M -set C with area 6, where the maximum distance from

(0, 0) to a point of C is 5. Find the smallest positive integer n such
that (n + 1)2 A0 > (n + 2s)2 , where A0 and s are the quantities defined
in the text.
2. Using straight lines, draw a nonconvex figure with symmetry about
(0, 0), area A > 4, and no lattice points except (0, 0) inside it or on its
boundary.
3. Here is another proof of Minkowski's Fundamental Theorem, due to
Blichfeldt. It is like ours, except for our argument establishing a pair of
points in C 0 whose coordinates differ by integers; see relations (5.4).
Instead, we will consider the M -set C 0 of area A0 > 1, which is cut
into pieces by the lattice L as shown in Figure 5.8.
We fit each piece into a unit square U in such a way that it occupies
the same position in U as it occupied in the lattice square from which
it came. Since A0 > 1, while the area of U = 1, some pieces of C0
overlap in U . Let D1 and D2 be two pieces of C 0 that overlap in U ,
Copyright © 1962. American Mathematical Society. All rights reserved.

P2

P1 P

Figure 5.8
Lattice square and unit square.

Olds, C., et al. Geometry of Numbers, American Mathematical Society, 1962. ProQuest Ebook Central,
http://ebookcentral.proquest.com/lib/princeton/detail.action?docID=3330416.
Created from princeton on 2024-04-18 00:02:58.