Algebra and Tiling

Algebra and Tiling
Homomorphisms in the Service of Geometry

The Cams Mathematical Monographs
Number Twenty-five
Algebra and Tiling

Homomorphisms in the Service of Geometry
S h e r m a n K. S t e i n
University of California, Davis
and
S ä n d o r Szabo
University of Bahrain
Published and Distributed by

THE MATHEMATICAL ASSOCIATION OF AMERICA
© 1994 by
The Mathematical Association of America (Incorporated)
Library of Congress Catalog Card Number 2006925584
Hardcover (out of print) ISBN 978-0-88385-028-2

Paperback ISBN 978-0-88385-041-1
eISBN 978-1-61444-024-6
Printed in the United States of America
Current Printing (last digit):

10 9 8 7 6 5 4 3
THE
CARUS MATHEMATICAL MONOGRAPHS
Published by
THE MATHEMATICAL ASSOCIATION OF AMERICA
Committee on Publications
James W. Daniel, Chair
Cams Mathematical Monographs Editorial Board

Marjorie L. Senechal, Editor
Katalin A. Bencsäth
Robert Burckel
Giuliana P. Davidoff
Joseph A. Gallian
The following Monographs have been published:
1. Calculus of Variations, by G. A. Bliss (out of print)

2. Analytic Functions of a Complex Variable, by D. R. Curtiss (out of print)
3. Mathematical Statistics, by H. L. Rietz (out of print)
4. Projective Geometry, by J. W. Young (out of print)
5. A History of Mathematics in America before 1900, by D. E. Smith and
Jekuthiel Ginsburg (out of print)
6. Fourier Series and Orthogonal Polynomials, by Dunham Jackson (out
of print)
7. Vectors and Matrices, by C. C. MacDuffee (out of print)
8. Rings and Ideals, by Ν. H. McCoy (out of print)
9. The Theory of Algebraic Numbers, second edition, by Harry Pollard and
Harold G. Diamond
10. The Arithmetic Theory of Quadratic Forms, by B. W. Jones (out of print)
11. Irrational Numbers, by Ivan Niven
12. Statistical Independence in Probability, Analysis and Number Theory,
by Mark Kac
13. A Primer of Real Functions, third edition, by Ralph P. Boas, Jr.
14. Combinatorial Mathematics, by Herbert J. Ryser
15. Noncommutative Rings, by I. N. Herstein
16. Dedekind Sums, by Hans Rademacher and Emil Grosswald
17. The Schwarz Function and its Applications, by Philip J. Davis
18. Celestial Mechanics, by Harry Pollard
19. Field Theory and its Classical Problems, by Charles Robert Hadlock
20. The Generalized Riemann Integral, by Robert M. McLeod
21. From Error-Correcting Codes through Sphere Packings to Simple Groups,
by Thomas M. Thompson
22. Random Walks and Electric Networks, by Peter G. Doyle and J. Laurie
Snell
23. Complex Analysis: The Geometric Viewpoint, by Steven G. Krantz
24. Knot Theory, by Charles Livingston
25. Algebra and Tiling: Homomorphisms in the Service of Geometry, by
Sherman Stein and Sändor Szabo
26. The Sensual (Quadratic) Form, by John H. Conway assisted by Francis
Y. C. Fung
27. A Panorama of Harmonic Analysis, by Steven G. Krantz
28. Inequalities from Complex Analysis, by John P. D'Angelo
29. Ergodic Theory of Numbers, by Karma Dajani and Cor Kraaikamp
30. A Tour through Mathematical Logic, by Robert S. Wolf
MAA Service Center

P. O. Box 91112
Washington, DC 20090-1112
1-800-331-1622 fax: 1-301-206-9789
Preface
If n-dimensional space is tiled by a lattice of parallel unit cubes,

must some pair of them share a complete ( η - 1 ) - d i m e n s i o n a l face?
Is it possible to tile a square with an odd n u m b e r of triangles,
all of which have the same area?
Is it possible to tile a square with 30°-60°-90° triangles?
For positive integers k and η a (k, n)-semicross consists of
kn + l parallel η-dimensional unit cubes arranged as a corner cube
with η arms of length k glued on to η non-opposite faces of the
corner cube. (If η is 2, it resembles the letter L, and, if η is 3, a
tripod.) For which values of k and η does the (k, n)-semicross tile
space by translates?
T h e resolution of each of these questions quickly takes us away
from geometry and places us in the world of algebra.
T h e first one, which grew out of Minkowski's work on dio-
phantine approximation, ends u p as a question about finite abelian
groups, which is settled with the aid of the group ring, characters of
abelian groups, factor groups, and cyclotomic fields.
Tiling by triangles of equal areas leads us to call on valuation
theory and Sperner's lemma, while tiling by similar triangles turns
out to involve isomorphisms of subfields of the complex numbers.
T h e semicross forces us to look at homomorphisms, cosets,
factor groups, n u m b e r theory, and combinatorics.
Of course t h e r e is a long tradition of geometric questions re-
quiring algebra for their answers. T h e oldest go back to the Greeks:
vii
vüi ALGEBRA AND TILING
" C a n we trisect every angle with straightedge and c o m p a s s ? " "Can

we construct a cube with twice the volume of a given c u b e ? " "Can
we construct a square with the same area as that of a given disk?"
These were not resolved until we had the notion of the dimension
of a field extension and also knew that π is transcendental.
We consider only the algebra that has b e e n used to solve tiling
and related problems. Even so, we d o not cover all such problems.
For instance, we d o not describe Conway's application of finitely
presented groups to tiling by copies of a given figure. See [19] in
the Bibliography o n pp. 2 0 0 - 201. N o r d o we treat Barnes' use of al-
gebraic geometry [1]. Thurston [22] has written a nice exposition of
Conway's work, and providing the algebraic background for Barnes'
work would take too many pages. T h e group with generators a and b
and relations a = e = b plays a key role in obtaining the B a n a c h -
2 3
Tarski paradox, which asserts that a pea can b e divided into a finite
n u m b e r of pieces that can b e reassembled to form the sun. A clear
exposition of the argument was given by Meschkowski [16].
We had two types of readers in mind as we wrote, the under-
graduate or graduate student who has had at least a semester of
algebra, and the experienced mathematician. To m a k e the exposi-
tion accessible to the beginner we have a d d e d a few appendices that
cover some special topics not usually found in a typical introduc-
tory algebra course and also included exercises to serve as a study
guide. For both the beginner and the expert we include questions
that have not yet been answered, which we call "Problems," to dis-
tinguish t h e m from the exercises.
Now a word about the organization of this b o o k and the order
in which the chapters may b e read.
Chapter 1 describes the history leading u p to Minkowski's con-
jecture on tiling by cubes. We give t h e solution of that problem in
the form of Redei's broad generalization of Hajos's original solu-
tion. Its proof, which is much longer than the proofs in the other
chapters, is delayed until Chapter 7. (However, the proof uses only
such basic notions as finite abelian groups, factor groups, h o m o -
morphisms from abelian groups into the complex n u m b e r s , finite
fields, and polynomials over those fields.)
Preface ix
T h e beginner might start with Chapter 1, go to de Bruijn's har-

monic bricks in Chapter 2, and then move on to C h a p t e r s 3 and 4,
which concern the semicross and its centrally symmetric compan-
ion, the cross. After that, Chapters 5 and 6, which concern tiling by
triangles, can be taken in either order. With the experience of study-
ing these chapters, the beginner would then be ready for t h e rest of
Chapter 2 and the proof of Redei's theorem. T h e advanced reader
may examine t h e chapters in any order, since they are essentially
independent.
We h o p e that instructors will draw on these chapters in their
algebra courses, in order to bring abstract algebra ideas down t o
earth by applying them in geometric settings.
T h e little that we include in the seven chapters is only the tip
of the iceberg. T h e references at the end of each chapter and the
bibliography at the end of the book will enable the reader to pursue
the topics much further.
Several of these describe quite recent work. In [14] Laczkovich
and Szekeres obtain the following result. Let r b e a positive real
number. T h e n a square can be tiled by rectangles whose width and
length a r e in the ratio r if and only if r is algebraic and the real
parts of its conjugates are positive. This was done in 1990. I n d e p e n -
dently, Freiling and Rinne obtained the same result in 1994 by simi-
lar means. Kenyon [12] considers the question: Which polygons can
b e tiled by a finite number of squares? Gale [8] obtains a short proof
using matrices of Dehn's theorem, which asserts that in a tiling of
a unit square by a finite number of squares all the squares have ra-
tional sides.
We wish to acknowledge the contributions of several people
to this book. Victor Klee, Miklos Laczkovich, Lajos Posa, Fred
Richman, and J o h n T h o m a s , in response to our requests, described
the background of their work. Mark Chrisman and Dean
Hickerson graciously permitted us to include some of their unpub-
lished proofs. A a r o n Abrams, while a senior at t h e University of
California at Davis, read two earlier versions of t h e manuscript and
m a d e many suggestions that should help the book serve a b r o a d e r
χ ALGEBRA AND TILING
audience. T h e exposition also benefited from the advice of G.

Donald Chakerian and Marjorie Senechal.
Budapest Davis, California

August, 1994 August, 1994
Contents
Preface vii
Chapter 1
Minkowski's Conjecture 1
Chapter 2
Cubical Clusters 35
Chapter 3
Tiling by the Semicross and Cross 57
Chapter 4
Packing and Covering by the Semicross and Cross 85
Chapter 5
Tiling by Triangles of Equal Areas 107
Chapter 6
Tiling by Similar Triangles 135
Chapter 7
Rddei's Theorem 155
Epilog 187
Appendix A
Lattices 189
Appendix Β
The Character Group and Exact Sequences 193
xi
xii ALGEBRA AND TILING
Appendix C
Formal Sums 197
Appendix D
Cyclotomic Polynomials 199
Bibliography for Preface 201
Supplement to the Bibliography 203
Name Index 205
Subject Index 207
Symbol Index 209
Chapter 1
Minkowski's Conjecture
T h e origins of most of the tiling problems we will explore go back

just a short time. But Minkowski's problem, like many ideas in
mathematics, can trace its roots to the Pythagorean t h e o r e m ,
a + b = <?. We shall trace these roots, and then g o on to describe
2 2
some of the problems that Minkowski's question, in turn, suggested.

We do this in order to place the problem in its p r o p e r position in
the intricate web that constitutes mathematics.
1. Introduction
T h e G r e e k s discovered all integer solutions of the equation
x + y = z : pick three integers d, τη, and n , and let o n e of χ
2 2 2
and y be d ( 2 m n ) and the other be d(m -n ),

2 2
a n d ζ = d(m + n ). 2 2
Inspired by this, Diophantos, who lived some time between A.D. 150
and A.D. 350, described techniques for producing rational solutions
of such equations as x + y = z , x + y — z , and x + y =
2 2 3 3 3 2 4 4
z . H e does not refer to the equation x + y = z , whose first

3 3 3 3
recorded mention is in a letter written about the year A.D. 1000 by

the mathematician A b u Djafar M o h a m m e d Ben Alhocain [22]:
I have already explained that the arguments that A b u

M o h a m m e d Alkhodjandi, may G o d have mercy on him,
has proposed in this demonstration that on the addition
1
2 ALGEBRA AND TILING
of two cube numbers there does not result a cube, are de-
fective and inexact.
Alkhodjandi is better known for his observation of the obliqueness

of the earth's orbit in 992.
Fermat, who was familiar with Diophantos through Bachet's

edition of 1621, added a fresh twist, investigating the equation
x + y = p, where χ and y are integers and ρ is a prime. In a letter
2 2
to Mersenne, in 1640, he wrote, "Each prime n u m b e r that is one

m o r e than a multiple of four is uniquely the sum of two squares
and uniquely the hypotenuse of a right triangle." Six m o n t h s later,
he confessed to Frenicle, "Finding those two squares is as difficult
as groping in the dark."
But Fermat did not stop with the form x + y . In a letter to
2 2
Pascal in 1654 he stated, " E a c h prime that is o n e m o r e than a mul-

tiple of 3 is composed of a square and the triple of another square,
such as 7 , 1 3 , 1 9 , 3 1 , 37, etc.
Each prime that exceeds a multiple of 8 by 1 or 3 is composed
of a square and twice another square, such as 11, 17, 19, 4 1 , 43,
etc This will b e followed by the invention of many propositions
which d o not a p p e a r in Diophantos."
In short, Fermat introduced the study of integers represented
by the quadratic forms x + y , x + 3y , and x + 2y . Euler went
2 2 2 2 2 2
further, examining forms of the type Ax +Cy ,

2 2
w h e r e A a n d C are
fixed integers. Lagrange in 1775 studied the most general quadratic
form Ax + Bxy + Cy , where A, B, and C are fixed integers. If the
2 2
integers A, B, and C have the property that Ax + Bxy + Cy is

2 2
positive whenever χ and y are not both 0, the form is called positive
definite. Lagrange investigated the minimum nonzero value of such
a form when χ and y are integers. A n estimate of this minimum
would help in his classification of quadratic forms and, as we will
show in the next section, it turns out that this minimum is related to
the geometry of certain regularly spaced points in the plane, called
"lattices."
2. Quadratic forms and lattices

Informally speaking, a lattice L in the plane, R , is a "homoge-
2
n e o u s " set of isolated points. It is an example of a "regular" point

set in that it looks the same n o matter from which of its points you
observe it. Figure 1 shows part of a lattice. ( T h e whole lattice is
unbounded.) W e also assume that the origin is o n e of the lattice
points.
Say that you are at a point Ρ € L and a friend is at point
Q e L. You see a point R e L. T h e n your friend at Q will be able
to see a point at Q + (R - P). ( H e r e we treat points like vectors.)
This is the algebraic translation of the homogeneity of the lattice,
as shown in Figure 2.
T h u s the lattice is closed u n d e r the operation "Q + (R - P)."
Now let Q = (0,0). Then we see that the lattice is closed under the
operation of subtraction, R - P. This tells us that L is a subgroup
of the additive group of the vector space R . T h e other property
2
we want of a lattice is that its points are not arbitrarily close to each
other. This brings us to the formal definition of a lattice.
A set of points L in R is a lattice if it is a group u n d e r vector
n
addition and each of its points is the center of a ball that contains
n o other points of L.
FIGURE 1 FIGURE 2
In Appendix A it is shown that if L is a lattice in η-space, then

t h e r e are linearly independent vectors vi, v?,..., Vk, k < n, such
that L consists of all the points of t h e form
miVi + m U2 Η
2 Η "ifcffc,
where mi e Ζ, 1 < i < k. Such a set of vectors is called a basis

for L, and k is called the dimension of the lattice. Figure 3 shows a
basis for the lattice of Figure 1.
W h e n you have a basis it is tempting t o draw t h e dissection of
t h e plane by the parallelograms that they suggest, as in Figure 4.
A lattice has an infinite n u m b e r of bases (as Exercise 1 sug-
gests). Figure S shows a n o t h e r basis for t h e lattice of Figure 1 and
t h e associated parallelograms.
FIGURE 5
Let v\, V2, • • •, v b e a basis for an η-dimensional lattice L in τι-

n
space. T h e parallelepiped whose edges a r e v i , V 2 , . . . , v „ is the set

of points
{ x i V i + X2V2 Η h i t ) : 0 < Xi < 1,1 < i <

n n n}.
It is called a fundamental parallelepiped of the lattice. T h e volume

of this parallelepiped is the absolute value of t h e determinant of
t h e matrix whose rows are the vectors νχ, v j , . . . , v . In Appendix
n
A it is shown that this volume is independent of the particular basis

chosen for L. It is called t h e determinant of L and is d e n o t e d d.
T h e most famous lattice in η-space is the set of points all of
whose coordinates relative to the usual cartesian axes are integers,
t h e group Z , called t h e standard lattice. T h e most conspicuous
n
basis for this lattice consists of the η unit vectors
(1,0,...,0),(0 1,0,...,0),... (0,...,1).

> >
T h e determinant of Z n
is 1.
Exercise 1.
(a) D r a w three different bases for t h e lattice Z and their associ-
2
ated parallelograms.
(b) Check that in each case the determinant of t h e matrix whose
rows are the basis is 1 or - 1 .
Exercise 2. Let a and & b e integers. W h e n is the vector (a, b) p a r t

of a basis for Z1
2
In a book review in 1831 Gauss [4] showed that positive defi-

nite quadratic forms in two variables are intimately connected with
plane lattices. Gauss assumed that the coefficient of xy in t h e qua-
dratic form is even. This is no loss of generality, since we could, after
all, double all the coefficients of a quadratic form without losing any
information about its minimum value. We first present his discovery
as h e described it. T h e n we will express it in m o d e r n terms.
T h e positive definite binary form axx + 2bxy + cyy repre-

sents in general the square of the distance between arbi-
trary points in a plane whose coordinates along axes that

meet at an angle whose cosine is b/^/ac are x^fä and ysfc.
In so far as χ and y refer only to whole numbers, this form
is a system of points that lie in the intersection of two sys-
tems of parallel lines. T h e lines of each system lie at equal
distances from each other, and this c o m m o n distance is
s/a, if measured parallel to the lines of the second sys-
tem; the c o m m o n distance of the other system, measured
parallel to the lines of the first system, is y/c. T h e angle
between the two systems is the one specified above.
In this way the plane is divided into congruent parallel-

ograms whose vertices form the system of points w h e r e
n o n e of the points can fall within the parallelograms. T h e
determinant taken with a positive sign, that is, ac — bb,
gives the square of the area of an elementary parallelo-
gram.
O n e and the same system of such points can b e divided
into the vertices of a family of parallelograms in infinitely
many different ways and thus referred back to equally
many different forms, but all these forms are what is tech-
nically called "equivalent" and the area of those elemen-
tary parallelograms remains the same.
Figure 6 shows what Gauss had in mind.
FIGURE 6
T h e square of the length of xv\ + yv 2 is the value of the dot

product
(xvi + yv )- 2 (χνχ + yv ),
2
which expands to
x (i>i · vi) + 2xy(vi

2
• v ) + y {v
2
2
2 • v ).
2
Since
v\ • vi = a, v • v = c,
2 2
and
vi · v = I 1 1 cosö =
2 \/äy/c{b/\/a~c),
the form ax + 2bxy + cy indeed does represent t h e square of t h e

2 2
length of a typical vector in the lattice.
Exercise 3 .
(a) Show that if ax + 2bxy + cy is positive definite, then α > 0,
2 2
c > 0 a n d b < ac. 2
(b) Is the converse true?

So, instead of looking for the minimum nonzero value of a
positive definite quadratic form, we may examine the square of the
length of the shortest nonzero vector in a lattice.
T h e form ax + 2bxy + cy in matrix notation reads
2 2
«-"GOß-
M o r e generally, a quadratic form in η variables,
η η
»=1 j = l
can be written as
( an ai2 · (xi\
ai2
a22 . • ß2n
Xn)
\ani a n 2 . • Ann ) Un/

Denoting the row vector (xi, x , . . . , x„) by x, its transpose (a col-

2
u m n vector) by x , and the matrix (aij) by A, we see that a quadratic

T
form is denoted by the matrix product xAx . T h e determinant of T
A is called the determinant of the quadratic form, which we d e n o t e

D.
T h e matrix A is symmetric. T h u s there is an orthogonal matrix
Ο such that OAO = E, a diagonal matrix. Moreover, if xAx
T T
is positive definite, then the diagonal entries in Ε are positive.

From this it follows that there is a nonsingular matrix Β such that
A = BB .T
Exercise 4.
(a) Fill in the details of the previous argument.
(b) Prove that if Β is a nonsingular η by η matrix, then the qua-
dratic form xBB x T
is positive definite.
T
T h e quadratic form x^4x = xBB x T

can b e rewritten as
T T
(xB)(xB) ,
T
which is simply the square of the length of the vec-
tor xB. If we restrict the coordinates of the vector χ to b e integers,
t h e n the set of vectors of the form xB forms a lattice. To b e specific,
l e t ß = (6y).Then
612 bln\
621 622 t>2n
xB = (xj, x , . . . , x„)
2
\ί>η1 K2 Kn)
= ( X l & l l + X2&21 Η Η X b l,
n n
Xlhn + Χ2&2η Η 1" ΧΚ

η 0
= £ι(&ιι, · · · ,bin) Η V x (b i,-

n n • -,b ).
nn
Denoting the j t h row of Β by Vj, we see that xB — χχνχΛ \-χ ν · η η
H e n c e as χ varies, χ Β sweeps out the lattice with basis υχ, u , . . . , v . 2 n
Exercise 5. What is the relation between the determinant D of a

quadratic form, that is, the determinant of A, and the volume of the
parallelepiped spanned by vi, v ,...,

2 v , that is, the absolute value
n
of the determinant of J5?
Exercise 6. Consider the lattice Z that consists of the points in

2
R both of whose coordinates are integers.

2
(a) O n e basis for this lattice is v\ = (1,0) and v = (0,1). W h a t is

2
the associated quadratic form?

(b) A n o t h e r basis is v\ = (2,3) and v = (1,1). What is the asso-
2
ciated quadratic form in this case?
Exercise 7. Figure 7 indicates the plane tiled by equilateral trian-

gles of side length 1.
(a) Give a simple basis for the set of vertices of the triangles.
(b) W h a t is its determinant?
(c) W h a t is the positive definite quadratic form associated with
this lattice and the basis chosen in (a)?
H e r m i t e in 1845 proved that a positive definite quadratic form
in η variables with determinant 1 assumes a nonzero value less than
or equal to ( 4 / 3 ) ( ^ for some integer vector. His proof was an
n _ 1 2
X2
FIGURE 7
induction o n n. For η — 2, this b o u n d is τ/ϊ/Ζ « 1.15. In particular,

the quadratic form 2x + Wxy + 13y must assume the value 1 for
2 2
some integers χ and y.
Exercise 8. Find integers χ and y such that 2 x + lOxy+I3y

2 2
= 1.
Exercise 9. What does Hermite's theorem imply when the con-

dition "determinant 1" is replaced by the condition "determinant
D"?
Exercise 10. W h a t does Hermite's t h e o r e m imply about a lattice

of determinant 1 in the plane?
Exercise 11. W h a t is the relation between the length of the short-

est nonzero vector in a plane lattice and the largest radius of any set
of congruent nonoverlapping disks whose centers are at the points
of the lattice? ( T h e problem of finding t h e length of the shortest
vector in an η-dimensional lattice is thus intimately connected with
the problem of finding how densely o n e may place nonoverlapping
balls in n-space.)
3. Minkowski's theorem
Minkowski, introducing a new approach, improved on H e r m i t e ' s
bound. As h e wrote his friend Hilbert, in 1889, [14, p.38]
I have now progressed much further into the theory of

positive definite quadratic forms, and in the case of forms
with a greater number of variables it is indeed true that
the situation is very d i f f e r e n t . . .
Perhaps the following theorem would interest you or
Hurwitz (which I can prove in half a page):
In a positive definite quadratic form of d e t e r m i n a n t
D and η ( > 2) variables one can always give the variables
such integer values that the form is less than n D ' " . 1
H e r m i t e has here instead of the coefficient η only

( 4 / 3 ) ( ' / , which obviously in general is a much higher
n - 1 2
bound...
Please give my best greetings to Hurwitz as well as
my best regards to our other colleagues, and I would ask
you, yourself, to r e m e m b e r — a n d now and then to write
to....
Your faithful
H. Minkowski
Exercise 12. Using a calculator find the first η for which

Minkowski's b o u n d is smaller than Hermite's.
Exercise 13. Derive Minkowski's bound for the general case from
the special case when D = 1.
Minkowski's approach is based on properties of translates of

a set Κ by the vectors in a lattice L. Consider a lattice L that has
determinant d and a set Κ of area A. If υ is a vector, the set of points
{v+k:k € K} is called a translate of Κ and denoted ν+Κ. Assume
that the translates of Κ by the vectors in L are disjoint. W h a t can we
then say about the area of A? Figure 8 shows the typical situation,
where Κ is a worm-shaped set.
In Figure 8 we have shaded the part of one parallelogram that
is occuped by translates of Κ. Call this set S. Every part of the worm
appears in S. After all, each part appears in some parallelogram,
and whatever appears in o n e parallelogram appears in all of them.
T h u s the area of S is A. Since S lies in a parallelogram of area d we
conclude that A < d. L e m m a 1 puts this observation in a form that
will be m o r e useful. N o t e that our reasoning applies in all dimen-
sions.
Lemma 1. Let L be an η-dimensional lattice of determinant d Let

Κ be a set in η-space of volume greater than d Then there are distinct
elements vi and v in L such that (v\ + Κ) Π (v + K)is not empty.
2 2
Exercise 14. Show that L e m m a 1 does not hold if the volume of

Κ is only d.
FIGURE 8
T h e theorem that Minkowski d e d u c e d from this l e m m a con-

cerns centrally symmetric convex sets. For convenience we will re-
strict our attention to b o u n d e d sets. A b o u n d e d set C in η-space is
convex if, whenever points Ρ and Q are in C so is the entire line seg-
m e n t PQ. It is centrally symmetric if t h e r e is a point Μ such that for
all points Ρ in C, the point Q such that Μ is the midpoint of PQ
is also in C. A n ellipse or parallelogram is a centrally symmetric
convex set, but a triangle is not.
Theorem 1. Let C be a centrally symmetric convex set in n-space.

Assume that its volume is greater than 2 d and the origin is its center
n
of symmetry. Let L be an η-dimensional lattice of determinant d Then

C contains a point of L other than the origin.
Proof. For convenience suppose η = 2. Let Κ = { x / 2 : x e C } ,

which is similar to C but shrunk by a factor of 2 in all directions,
hence it has area greater than d. It is shown in Figure 9.
Since Κ has area greater than d, we conclude that two trans-

lates of Κ meet, that is, there are vectors x j , x e L and ki, k € K,
2 2
χ ι Φ x , such that
2
Xi + fci = X2 4- k .2
Hence
Xi — x 2 = k 2 — k\.
Since Κ is centrally symmetric, -k\ is an element K, which we de-

note k$. H e n c e
xi - x = k + k
2 2 3 = 2^ f c a
2 fc3
) •
Since i f is convex (fc + fa) / 2 G Ä" and therefore 2 ( k + k ) /2 G C,

2 2 3
that is, k + k € C. H e n c e C contains x i — x , a nonzero vector in

2 3 2
L and the proof is complete. •

Exercise 15. Show that the theorem is not true if either of the as-
sumptions "centrally symmetric" or "convex" is removed.
If we assume that the volume of C is 2 d, then T h e o r e m 1 no
n
longer holds. A counterexample in 2-space is provided by the lattice

Z and C is the square whose vertices are (1,1), ( - 1 , 1 ) , ( - 1 , - 1 ) ,
2
and (1, - 1 ) , but without its border. However, if we add the assump-
tion that C is a "closed" set, then the conclusion of T h e o r e m 1 still
holds, as we will show.
A set C in R is closed if whenever a sequence of points in C
n
converges to a point P, then Ρ must be in C. A polygon without its

boundary is not closed; a polygon with its boundary is closed.
T h e next theorem is the one that led to Minkowski's conjec-
ture.
Theorem 2. Let C be a bounded, centrally symmetric, closed set in

η-space. Assume that its volume is 2 d and the origin is its center of
n
symmetry. Let L be an η-dimensional lattice of determinant d Then

C contains a point ofL other than the origin.
Proof. Assume that the t h e o r e m is false. Enclose C in some very

large box, much larger than C. This box contains only a finite num-
ber of points of L, and let Ρ be o n e of t h e m other than the origin.
We first show that the distance between Ρ and any point in C is
greater than some fixed positive number.
Let d(P, Q) denote the distance between Ρ and point Q. As-
sume that for each positive integer m there is a point Q in C such
m
that d(P, Q ) is less than 1/m. T h e sequence Qi, Q2,

m Q, • • •
m
converges t o P. Since C is closed, Ρ lies in C. This contradicts the

assumption that Ρ is not in C.
Therefore there is a positive n u m b e r ρ such that d(P, Q) > ρ
for all points Q in C. It follows that there is a positive n u m b e r p*
such that the distance between any point of C and any point of L
other than the origin is greater than p*.
Let s b e a n u m b e r slightly larger than 1. Let sC = {sQ:Q €
C } , which is a slight magnification of C. Indeed, choose a so that
d(sQ, Q) < p* for all Q in C. T h e n sC has volume s"d, which is
greater than d, and contains n o point of L o t h e r than the origin.

This contradiction of T h e o r e m 1 completes the proof. •
Minkowski first developed these theorems for the special cases

when C is a box o r ball. Later he showed that they hold for any cen-
trally symmetric convex set. T h e reason h e wanted these t h e o r e m s
is illustrated by t h e following consequence.
Theorem 3. A plane lattice L of determinant 1 contains an element

whose length is not greater than s/2.
Proof. Let C be the square whose vertices are ( 1 , 1 ) , ( - 1 , 1 ) ,

( - 1 , - 1 ) , and ( 1 , - 1 ) . By T h e o r e m 1 it contains an element ν e L,
ν φ (0,0). But every point of C is within a distance y/2 of t h e origin.
Thus the length of υ is less than o r equal to %/2. •
Exercise 16. Justify the claim Minkowski m a d e in his letter.
Minkowski improved the b o u n d in T h e o r e m 3 by using balls

instead of cubes.
Exercise 17. Using a disk instead of a square, prove that a plane

lattice of determinant 1 contains an element whose length is not
greater than 2/,/π. (Note that 2/y/n « 1.13, which is less than \p2,
t h e bound in T h e o r e m 2. This is not the best possible result.)
Exercise 18. W h a t does the previous exercise imply a b o u t positive

definite quadratic forms in two variables?
In Minkowski's obituary, Hilbert [13, p p . v-xxi] c o m m e n t e d on

T h e o r e m 2,
This proof of a d e e p n u m b e r theoretical result without

analytical tools, essentially o n the basis of a geometrically
intuitive consideration, is a pearl of Minkowski's discov-
eries. In t h e generalization to forms of η variables
Minkowski's proofs led to a m o r e natural and far lower
upper bound than that which Hermite had found. Still

more important however than this was that the essential
idea of Minkowski's reasoning used only this property of
the ellipsoid, namely that it is a convex figure and pos-
sesses a center, which could therefore b e applied to arbi-
trary convex figures with a center. This situation led
Minkowski to recognize for the first time that the concept
of a convex body is a fundamental idea in our science.
4. Minkowski's conjecture
Minkowski and others used T h e o r e m 2, the "lattice point t h e o r e m , "
to obtain several fundamental results in algebraic n u m b e r theory.
However to pursue this path would take us far from the r o u t e we
want to follow. It is Minkowski's use of his t h e o r e m in examining
the simultaneous approximation of several real numbers by rational
numbers that will lead us t o consider tilings of space by parallel
congruent cubes.
Rational numbers with a given denominator, x , can approxi-
2
m a t e any n u m b e r α fairly well. A glance at the n u m b e r line in Figure

10 shows that there is an integer x such that
x
Xl
a-
X2
< 2x2
I J L
2_
X2 I 2
X2
FIGURE 10
T h e next exercise obtains a stronger result by using Minkowski's

theorem.
Exercise 19. Let α be a real n u m b e r and η a positive integer. Prove

that there are integers xi and x such that 2
a < —— and x2 < n.

nx 2
Suggestion: Let L b e the lattice in R with basis (α, 1) a n d ( - 1 , 0 ) .

2
Let C be a rectangle of width 2/n and height 2n.
Minkowski considered the problem of approximating two real

numbers, a and b, by rational numbers with the same denomina-
tor, but n o t prescribing the denominator, and proved the following
theorem.
Theorem 4. Let tbea real number greater than 1 and a and b real
numbers. Then there are integers x\, x and x , with x positive,
2 3 3 such
that
Xl X2
α < b- and X3 < t,
2
X3 a? t'
3 X3 xt
3
hence
Xl
a and 6 - 2
~ 3/2' ~ 3/2'
Z3 X3
Proof. L e t Β = (b^) b e a 3 by 3 real matrix of determinant 1. T h e

cube with vertices ( ± 1 , ± 1 , ± 1 ) contains a nonzero vector of t h e
lattice spanned by the rows of B. T h u s the simultaneous inequalities
|i>nxi + 612^2 + 613X31 < 1
|&2lXl + &22Z2 + i>23^3| < 1 (!)
1^31^1 + 632^2 + 633^31 < 1
have an integer solution other than the trivial one, ( 0 , 0 , 0 ) . Now let
t > 1 b e a real n u m b e r a n d α a n d 6 be real numbers. As a special
case of (1) t h e inequalities
I i i i + 0x2 — atxz\ < 1

|0xi + tx2 — 6ix | < 1 3
|0x + 0ΐ2 + ( l / i ) x | < 1

x
2
3
have a nontrivial integer solution. T h u s there are integers xi, x , 2
and £ 3 , with X3 > 0, such that

xi 1 X2
ο <—,, X3< t ,
2
X3 X3^' X3 Χ ί 3
and consequently
xi £2
X3
- 3/2 >
X3 - S/2 -
• (2)
Exercise 20. State a n d prove the analog of (2) for three real num-
bers a, b, a n d c.
Minkowski showed that the inequalities in (1) could not always
be strict. H i s counterexample was
|xi + 612X2 + &13X3I < 1
|0xi + x 2 + 623x31 < 1 (3)
|0χι -I- 0x + 2 x | < 1.

3
( N o t e that t h e determinant of Β is 1.)

Any integer solution of (3) h a s x 3 = 0, hence X2 = 0, and
finally x i = 0.
T h e assertion that t h e inequalities
l&nxi + 612X2 + 613X31 < 1
|6 ixi2 - I - 622X2 + 623X3I < 1 ( )

4
I631X1 + 632X2 + 633X31 < 1
have only t h e trivial solution is equivalent t o saying that t h e interior

of t h e cube with vertices ( ± 1, ± 1, ± 1) contains only o n e point of the
lattice spanned by
«1 = ( 6 1 1 , 6 2 1 , 6 3 1 ) , V2 = ( 6 1 2 , 6 2 2 , 6 3 2 ) , t>3 = (613,623,633),
namely, the origin. This, in turn, is equivalent to the assertion that

the translates of the interior of the closed cube Κ with vertices
( ± 1 / 2 , ± 1 / 2 , ± 1 / 2 ) by the vectors of the lattice spanned by v , υ
x 2
and v are disjoint. Now, Κ has volume 1. Thus the union of the
3
translates of Κ is all of 3-space.
Exercise 21. Why is the union of the cubes all of 3-space?
This is where the concept of tiling space enters the picture.

We shall define a "tiling" in sufficient generality to cover all the
cases m e t in the following chapters.
Let Si, S ,...,
2 Sfc,... be a family of subsets of a set S in τι-
space such that each Sk is an η-dimensional polyhedron or the finite
union of such polyhedra. In Chapters 1 to 4 S is an n-dimensional
fc
cube or the finite union of such cubes. In Chapters 5 and 6 Sk is

a triangle. T h e family is a packing of S if the interiors of the Sk's
are disjoint. T h e family is a covering of S if the union of the Sk's
is S. A family that is both a packing and a covering of S is called
a tiling of S. If each Sk is congruent to fixed set Τ and the family
of Sk's packs (covers, tiles) S, we say that Τ packs (covers, tiles) S.
M o r e specifically, if Sk is a translate of a fixed set T, we say that Τ
packs (covers, tiles) S by translates. If the context makes it clear that
we are considering only translates, we will usually omit the phrase
"by translates." Chapters 1, 2 , 3 , 5 , and 6 concern only tiling, while
C h a p t e r 4 treats packing and covering. In Chapters 1, 2 , 3 , and 4 S
is η-space, while in Chapters 5 and 6 it is a polygon.
As we have just seen, Minkowski was led by an algebraic ques-
tion to consider tilings of space by translates of a unit cube.
W h a t does a lattice of translates of a unit cube in 3-space look
like? l b answer this question, first consider a lattice of translates of
a unit square Κ that tiles 2-space.
For convenience, let Κ be the square whose vertices are (0,0),
(1,0), ( 1 , 1 ) , and (0,1). If n o square shares a complete edge with K,
then the squares that lie along the right-hand edge or top edge of
Κ are arranged as in Figure 11 or Figure 12.
x3
κ
(ha)
(0,0)
FIGURE 11
FIGURE 12
In the first case two squares share a complete c o m m o n hor-

izontal edge. In other words, the lattice contains the unit vector
(0,1). In the second case, the lattice contains the vector ( 1 , 0 ) .
In the first case, the tiling is m a d e u p of vertical strips stag-
gered uniformly, as in Figure 13. In the second case it is m a d e u p
of horizontal strips.
T h e lattice indicated in Figure 13 has a basis of the form (1, a)
and (0,1), for some n u m b e r a. T h u s a typical vector is xi(l,a) +
x ( 0 , 1 ) . Since n o such vector lies in the square whose vertices are
2
( ± 1 , ± 1 ) , t h e inequalities
|xi + 0 x | < 1
2
(5)
|αχχ + x | < 1
2
have only o n e integer solution, ( 0 , 0 ) .

x3
(0,1)
/. X\
FIGURE 13
Exercise 22. Assume that Β = (&<,·) has determinant 1 and t h e

inequalities
\b xi
n + bi x \
2 2 < 1
\b iXi
2 + b x\
22 2 < 1
have only o n e integer solution, ( 0 , 0 ) . Show that some row of Β

consists of integers and they are relatively prime. (Hint: Translate
the problem into o n e on squares.)
With the aid of a few sugar cubes or dice it is not hard to show
that if translates of a unit cube tile 3-space, then some pair share a
complete 2-dimensional face. In fact, by the time that you surround
a cube with other cubes, you will find that two share a complete face.
T h u s a lattice tiling of 3-space by parallel congruent cubes consists
of infinite tubes. These tubes, in turn, form slabs, a n d the slabs fill
u p 3-space.
Exercise 23. Consider a tiling of 3-space by a lattice of unit cubes.

Show that the lattice has a basis of the form
(1,0,0), ( ΐ ι , Ι , Ο ) , ( x , x , l ) ,
2 3
or a basis obtained from this by a permutation of t h e coordinates.

This was Minkowski's conjecture, m a d e in algebraic form in
1896 and in geometric form in 1907.
Conjecture. In a lattice tiling of η-space by unit cubes there must be

a pair of cubes that share a complete (n — l)-dimensional face.
Actually, he never m a d e this a specific conjecture. In his book,

Geometrie der Zahlen, published in 1896, Minkowski [11] wrote on
p. 105, "I plan to give a proof of this theorem in a special article in
connection with detailed arithmetic investigations of η linear forms,
exactly like I will carry out h e r e [in Section 45] for the case η = 2."
T h e argument in Section 45 did not easily generalize to all di-
mensions, and n o "special article" appeared. In his Diophantische
Approximationen [12], published in 1907, we find on page 28, "I
assume and would like to pose it as a problem." H e settled the case
η = 3 on pages 67-74, translated this case into a property of a lat-
tice tiling of 3-space by cubes, and included the diagram shown in
Figure 14 that suggests the general appearance of a lattice of cubes.
However, his argument for the case η = 3 again failed to generalize
to higher dimensions.
In spite of its seeming simplicity, Minkowski's conjecture
reached the venerable age of 45 years before it was settled by Hajos
in 1941 [5].
We will call two η-dimensional cubes whose intersection is a
complete ( η - 1)-dimensional face twins. Schmidt in 1933 [19] ob-
served that to prove Minkowski's conjecture it is enough to prove it
for the case of a rational lattice cube tiling, that is, for a lattice cube
tiling in which the lattice vectors have only rational coordinates.
H e r e we tacitly assume that the cubes are unit cubes and they are
parallel to the coordinate system, since the rationality depends on
the choice of coordinate system. We will justify this reduction in t h e

next chapter.
5. The group theoretic version

of Minkowski's conjecture
It was a milestone when Hajos in 1938 translated Minkowski's con-
jecture into an equivalent conjecture about finite abelian groups.
T h r e e years later this led him to his proof of t h e conjecture. We
will illustrate the connection between the geometry of lattice cube
tilings and abelian groups by describing it in 2-space.
Consider a lattice family of translates of a unit square, whose
vertices a r e (0,0), (1,0), (1,1) a n d ( 0 , 1 ) , which n e e d n o t form a
tiling. Let L b e t h e vectors of the lattice. (They are the bottom left
corners of the squares.) We restrict our attention only to rational
lattices. T h u s if k and l are basis vectors of L, then we may assume
2
that
^ _ /Oil £i2\ ^ _ /Q21 Q 2 2 \
Voll' 612/ \&2l'&22/'
where ay, by are integers and by > 0. Let τ ί b e a positive c o m m o n
multiple of 6 n and 621 and similarly let r b e a positive c o m m o n
2
X2
FIGURE 15
multiple of 612 and 622. T h e straight lines perpendicular to the first

axis going through the points whose first coordinates are integer
multiples of 1/r-j and the straight lines perpendicular to the second
axis going through the points whose second coordinates are integer
multiples of 1 / Γ cut the plane into rectangles of dimensions 1/rj
2
by 1/Γ2, as shown in Figure 15.

Each square of the family is cut into Γ χ Γ rectangles whose di-
2
mensions are l/r\ by Ι / Γ 2 . T h e small rectangles form a lattice tiling

of 2-space by the lattice L' consisting of their b o t t o m left corners.
This lattice has a basis
l[ = (l/ ,0),
ri Z = (0,l/r ).
2 2
Now, L is a sublattice of L', or in other words L is a subgroup of

V. The main algebraic tool we need is the factor group G = L'/L,
whose elements are the cosets modulo L , or, geometrically speak-
ing, the translates of L . T h e coset L is the identity element e of G.
Let οχ and a be the cosets containing the basis vectors l[ and 1' of
2 2
V. If the interiors of two members of the family of squares overlap,

that is, if one of the small rectangles is covered by two squares, then
an element V of V can b e represented in two distinct ways in the
form
I' = I + Χχ1[ + x l' ,

2 2
where I € L , 0 < χχ < τχ - 1,0 < x < r - 1. O n the other hand,

2 2
if there is a rectangle that is not covered by any square of the family

then there is an element V of V which cannot be represented in the
above form. Using the factor group G written multiplicatively we
may say that if the elements
a xa \
x l x
2 0<xi<ri-l, 0 < x 2 < r - 1
2 (6)
are distinct then the lattice of squares is a packing of 2-space. If

these elements represent all the elements of G then the lattice of
squares is a covering of 2-space. If these elements represent all el-
ements of G without any repetition then the lattice of squares is a
tiling. Thus by means of G we can state whether the corresponding
lattice of squares is a tiling.
Next we translate into the language of the group G the asser-
tion that the lattice of squares has no twins.
Two squares form a pair of twins if the lattice elements that
describe their position differ by (1,0) or (0,1). Thus, if twins are
present, (1,0) or (0,1) belongs to the lattice L . In other words,
rxl[ € LOT r l' € L . In the factor group G we then have
2 2
a\l
= e or a? = e. (7)
T h a t twins are present in the lattice square tiling can b e formu-

lated in terms of G as follows: If every element in the abelian group
G can be represented uniquely in the form (6), then (7) follows.
Similar reasoning in η-space gives us Hajos's translation of

Minkowski's conjecture into an equivalent conjecture about finite
abelian groups.
Hajos's version of Minkowski's conjecture. Let Gbea finite abe-

lian group. If αχ, a , . . . , a are elements of G and Γχ, r , . . . , r„ are
2 n 2
positive integers such that each element ofGis uniquely expressible in

the form
a* · · · < " ,
1
0 < χχ < η - 1 , . . . , 0 < x„ < r - 1,
n
then a[ = e for some i, 1 <i <ru

(
Since we have verified Minkowski's conjecture for 1-space and

2-space, a n d sugar cubes quickly settle it in 3-space, we know that
Hajos's version holds for η < 3.
Exercise 24. Show that if Minkowski's conjecture is true when

Γχ = • · · = r „ , then it is true in general. (Hint: G o back to t h e
geometry.)
6. More about the algebraic

version of Minkowski's conjecture
Let G b e a n abelian group and let Αχ,... ,A„ be subsets of G. If
each element g of G is uniquely expressible in the form
9 = αχ·α , η αχ £ Αχ,... ,an e A,

n
then we say that G is factored by t h e subsets Αχ,..., A a n d that n
the product Αχ·· A is & factorization of G. (Note that if G is finite

n
then \G\ = \Αχ\··• \A \.) This concept is a natural extension of the

n
concept of direct product of subgroups a n d we will write G =

ΑχΑ · • • A . We will assume throughout that each Ai is normalized
2 n
to contain t h e identity element of G.

Let a b e an element of the abelian group G a n d r a positive
integer. T h e subset consisting of the elements
will be called a cyclic subset. To avoid trivial cases we suppose that

α Φ e, r > 2 and that the order of α is at least r, so that the ele-
ments e, a, a ,...,
2
α are distinct. If the order of α is r then this
Γ _ 1
cyclic subset coincides with the cyclic group generated by a. A cyclic

subset can be thought of as a "front e n d " of a cyclic group.
Hajos's version now reads: In any factorization of a finite abelian
group by cyclic subsets at least one of the factors is a subgroup.
T h e conjecture in this form could b e thought of as t h e dual of
the fundamental theorem of finite abelian groups. Indeed, the latter
states that there is a factorization of G by cyclic subsets in which
all the factors are subgroups. T h e algebraic version of Minkowski's
conjecture, on t h e other hand, claims that in each factorization of
G by cyclic subsets there is at least one subgroup among the factors.
T h e next two exercises show that in Hajos's version we may
assume that the cyclic subsets have prime orders.
Exercise 25.
(a) Verify that if r > 2 and s > 2, then the cyclic subset
A = {e,a, a ,...,
2
a™' } 1
of cardinality rs can b e factored into the smaller cyclic subsets
B = {e,a,a ,...,a - }
2 r 1
and C = {e,a ,a ,...
r 2r
,α^ ^}
1
of cardinalities r and s respectively.

(b) Show that A can also be factored into the cyclic subsets
B' = {e,a,a , 2
...,α" } 1
and C' - {e, a ,a ,...s 2a
, a ^ s
}
and so the factorization is not always unique.
Exercise 26. (This continues Exercise 25.)

(a) Prove that Β cannot b e a subgroup.
(b) Prove that A is a subgroup of G if and only if C is a subgroup
ofG.
By Exercise 25 every cyclic subset can be factored into cyclic

subsets of prime cardinalities. By Exercise 26 if the original cyclic
subset is not a subgroup t h e n neither are the new ones; and if none
of the new ones is a subgroup then neither is the original. Thus
we may suppose that in Hajos's version all the cyclic subsets are
of prime orders. But now the n u m b e r of the factors need n o longer
correspond to the dimension of the original cube tiling. As we men-
tioned earlier, Hajos settled Minkowski's conjecture in 1941 using
this approach.
7. Related work
In 1930 Keller [6] suggested that the "lattice" condition in
Minkowski's problem is irrelevant, conjecturing that in a tiling of
η-space by parallel unit cubes there must be twins. Perron in 1940
[16] verified this conjecture for η < 6.
If this conjecture were true it would imply Hajos's theorem,
and show that Minkowski's t h e o r e m was really geometric, not alge-
braic. However, in 1992 Lagarias and Shor [9], using an approach
of Corrädi and Szabo [2], showed that Keller's conjecture is false in
all dimensions greater than or equal to 10. (Dimensions 7 , 8 , and 9
are not settled.)
Similarly, Redei wondered whether the sets of prime orders in
Hajos's theorem had to be cyclic. In 1965 he showed [17] that the
assumption could b e removed, proving the following theorem.
Redei's Theorem. Let G be a finite abelian group and Αχ, A ,...,

2 A n
be normalized subsets of G of prime orders. Assume that G ~

A1A2 • • A is a factorization. Then at least one of the sets A, is a
n
subgroup.
We present a simplified version of Redei's proof in Chapter 7.

We delay it till then, even though it is elemertary, since it involves
a lengthy sequence of lemmas. However, we will m a k e use of it in
the next chapter to obtain a surprising property of certain sets that
are m a d e u p of a prime n u m b e r of cubes.
Exercise 27. How would Redei's theorem read if we remove the

assumption that each Ai is normalized?
In 1936 Furtwängler [3] generalized Minkowski's twin problem

to a problem about multiple tilings, which we describe.
Consider a lattice family of cubes that are translates of each
other. Suppose that every point of space belongs to finitely many
cubes of the family. Further, suppose that any point that is not on
the boundary of any cube lies in precisely fc cubes of the system.
This cube system is called a k-fold tiling and its multiplicity is A;.
Furtwängler conjectured that in such a fc-fold lattice tiling there
would b e twins.
Exercise 28. Show that Furtwängler's conjecture is equivalent to

this variant of Hajos's version: Let G be a finite abelian group and
A\, A ,...,
2 A„ cyclic subsets of G such that each element g of G
is representable exactly fc ways in the form g = a\a •• • a , where
2 n
cii is in Ai. (This is called a k-factorization of G by subsets

Ai,A ,...,
2 A .) T h e n at least o n e of the Ai is a subgroup of G.
n
Furtwängler proved his conjecture for η < 3. In 1938 Hajos,

using the algebraic version of the conjecture, exhibited a 9-fold
cube tiling in 4-space that has n o twins. In 1979 R. M. Robinson
[18] characterized all pairs (fc, n) for which there exists a fc-fold τι-
dimensional lattice cube tiling without twins. His result is:
If η < 3, there is no such fc.
If η = 4, fc is any integer divisible by the square of an
o d d prime.
If η = 5, then fc = 3 or fc > 5.
If η > 6, then each fc > 2 is possible.
Exercise 29. Show that Robinson's result implies Hajos's t h e o -

rem.
T h e next exercise refutes Furtwängler's conjecture in 6-space.
Exercise 30. Let G b e the direct product of two cyclic groups of

order 4 and let χ and y be a basis for G, hence x = e =4
y.
4
(a) Show that none of the cyclic subsets
Ai = {e,y}, A 2 = {e,x y}, 2

A3 = {e,x},
A t = {e, xy}, A 5 = {e, xy }, 2
A 6 = {e, xy }3
is a subgroup of G.
(b) Verify that the product ΑιΑ Α3Α Α Αβ 2 4 5 is a 4-factorization of
G.
(c) What does (b) tell us about multiple cube tilings?
T h e next exercise gives another counterexample.
Exercise 3 1 . Let G b e the direct product of cyclic groups of orders

2 , 2 , and 3 and suppose that u, v, and w form a basis for G such that
u = v = w = e.
2 2 3
(a) Show that n o n e of the following cyclic subsets is a subgroup of

G:
A\ = {e, vw, (vw) }, 2

A 2 - {e, uw, (uw) },
2
A3 = {e,uvw, (uvw) },
2
A4 — {e,vw}, A5 = {e, uw}.
(b) Verify that the product ^4x^42-43^4^5 is a 9-factorization of G.

(c) W h a t does (b) tells us about multiple cube tilings?
A n o t h e r path leads off from quadratic forms. Consider a posi-

tive definite quadratic form x Ax that assumes only integer values
T
when the vector χ has only integer coordinates. Represent A in the

form B B. Let the rows of Β be v , v ,...,
T
v . T h e n the square of
x 2 n
the length of every vector in the lattice whose basis is vi, v , • • •, v 2 n
is an integer. Such a lattice is called an integral lattice. This should

be distinguished from the "integer lattice" Z , which is a special n
case of an integral lattice.
Exercise 32. Show that the vectors v\,v ,...,v 2 n generate an inte-
gral lattice if and only if υ* · Vi and 2vi • Vj are integers for all 1 < i,
3 < n.
For an integral lattice L of determinant 1, consider the mini-
m u m value of the square of the lengths of the nonzero vectors of L.
D e n o t e this by μ(Ι/). Let μ b e the maximum value of μ(Ζ,) for all

η
η-dimensional integral lattices L of determinant 1.
Exercise 33. Using Minkowski's bounds on the shortest vector in

any lattice, show that μι = 1 and μ = 1. 2
Conway and Sloane [1] proved that for all sufficiently large n,
Mn < (n + 6 ) / 1 0 and included this table of values:
η 1-7 8 9 10 11 12 13 14-22 23 24 25 2&-31 32 33
Μη 1 2 1 1 1 2 1 2 3 4 2 3 4 3
In 1982, Lenstra, Lenstra, and Loväsz [10] developed an effi-

cient algorithm for finding short vectors in a lattice and applied it
t o the factoring of polynomials with integer coefficients.
Lagarias and Odlyzko [8] in 1985 applied it to public-key knap-
sack cryptography. Odlyzko and te Riele in 1985 [15] used that al-
gorithm to show that a conjecture m a d e by Mertens in 1897 is false.
( H a d the conjecture b e e n true, it would have implied the R i e m a n n
hypothesis.) M e r t e n s ' conjecture concerns the prime factorizations
of those integers which are not divisible by a square larger than 1,
the "square-free" integers. T h e prime factorizations of such inte-
gers have n o repeated factor. For a positive n u m b e r χ let
( n u m b e r of square-free integers less than or \

equal to χ that have an even n u m b e r of prime J
factors /
( n u m b e r of square-free integers less than or \

equal to χ that have an o d d n u m b e r of prime J.
factors /
For instance, the square-free integers less than or equal to 10 are

1, 2, 3, 5, 6, 7, and 10. Of these, 1, 6, and 10 have an even n u m b e r
of prime factors and 2, 3, 5, and 7 have an odd n u m b e r of prime
factors. T h u s M(10) = 3 - 4 = - 1 .
Exercise 34. C o m p u t e M ( x ) for χ = 25 and a; = 36 and c o m p a r e

| M ( x ) | t o y/x.
O n the basis of a tabulation of M(x) for χ u p to 10,000,

Mertens conjectured that for χ > 1
\M(x)\ < yfx.
Odlyzko and te Riele showed that Μ (χ) > 1.06y/x and M ( x ) <
- 1 . 0 0 9 γ / χ for an infinite n u m b e r of integers x, but confessed that,
" O u r proof is indirect and does not produce any single value of χ
for which | M ( x ) | > y/x. In fact, we suspect that t h e r e are n o coun-
terexamples to M e r t e n s ' conjecture for χ < 1 0 or even 1 0 , " and
2 0 3 0
gave reasons for this belief. For all large values of χ for which Μ (χ)
has been evaluated, specifically, u p to 7.8 χ 1 0 , | M ( x ) | < 0.6y/x,
9
illustrating the saying, "Any integer you can test is t o o small."

This chapter showed how even a question about tiling space
with cubes takes us quickly from the realm of geometry into algebra
and n u m b e r theory. Hajos used not only the group G but leaned
heavily on the group ring formed on G. As we will see in C h a p t e r 7,
the proof of Redei's generalization also employs the group ring, but
only briefly, and brings in the character group of G and cyclotomic
fields as well.
This chapter also illustrates how intricate is the web of m a t h e -
matics, for we have seen such varied areas as quadratic forms, lat-
tices, convex sets, approximation by rationale, tiling by cubes, pack-
ing spheres, factorization of abelian groups by subsets, factoring
polynomials, and n u m b e r theory overlapping each other. Clearly
any barrier inserted between one part of mathematics and a n o t h e r
would b e quite artificial. Subsequent chapters will provide far m o r e
evidence for this assertion.
References
1. J. H. Conway and N. J. A. Sloane, A new upper bound for the min-
imum of integral lattice of determinant 1, Bull. Amer. Math. Soc. 23
(1990), 383-387.
2. Κ. Corrädi and S. Szabo, A combinatorial approach for Keller's con-

jecture, Periodica Mat. Hung. 21 (1990), 95-100.
3. Ph. Furtwängler, Über Gitter konstanter Dichte, Monatsh. Math. Phys.
43 (1936), 281-288.
4. C. F. Gauss, Werke Vol. 2, König. Gesellschaft der Wiss. Göttingen,
1876.
5. G. Hajos, Über einfache und mehrfache Bedeckung des n-dimen-
sionalen Raumes mit einem Würfelgitter, Math. Zeit. 47 (1942), 427-
467.
6. Ο. H. Keller, Über die lückenlose Einfüllung des Raumes mit
Würfeln,/. ReineAngew. Math. 163 (1930), 231-248.
7. , Ein Satz über die lückenlose Erfüllung des 5- und 6-dimensional
Raumes mit Würfeln,/. Reine Angew. Math. 177 (1937), 61-64.
8. J. Ε Lagarias and Α. M. Odlyzko, Solving low-density subset sum
problems, JACM 32 (1985), 229-246.
9. J. F. Lagarias and P. W. Shor, Keller's cube-tiling conjecture is false in
high dimensions, Bull. Amer. Math. Soc. 27 (1992), 279-283.
10. A. K. Lenstra, H. W. Lenstra, and L. Loväsz, Factoring polynomials
with rational coefficients, Math. Ann. 261 (1982), 515-534.
11. H. Minkowski, Geometrie der Zahlen, Teubner, Leipzig, 1896.
12. ,DiophantischeApproximationen, Teubner, Leipzig, 1907 (reprint:
Physica- Verlag, Würzberg, 1961).
13. , Gesammelte Abhandlungen von Hermann Minkowski, Chelsea,
New York, 1967.
14. , Briefe an David Hilbert, Springer-Verlag, New York 1973.
15. Α. M. Odlyzko and H. J. J. te Riele, Disproof of Mertens' conjecture,
/. Reine Angew. Math. 357 (1985), 138-160.
16. O. Perron, Modulartige lückenlose Ausfüllung des R mit kongru-
n
enten Würfeln I., IL, Math. Ann. 117 (1940), 415-447, (1941), 609-
658.
17. L. Redei, Die neue Theorie der endlichen abelschen Gruppen und
Verallgemeinerung des Hauptsatzes von Hajos, Acta Math. Acad. Sei.
Hung. 16 (1965), 329-373.
18. R. M. Robinson, Multiple tilings of η-dimensional space by unit cubes,
Math. Zeit. 166 (1979), 225-264.
19. T. Schmidt, Uber die Zerlegung des n-dimensionalen Raumes in git-

terformig angeordnete Würfeln, Sehr. math. Semin. u. Inst, angew.
Math. Univ. Berlin 1 (1933), 186-212.
20. S. K. Stein, Algebraic tiling, Amer. Math. Monthly 81 (1974), 445-462.
21. S. Szabo, A reduction of Keller's conjecture, Periodica Math. Hung. 17
(1986), 265-277.
22. M. F. Woepcke, Recherches sur plusieurs ouvrages de Leonard de Pise
decouverts et publies par Μ. le prince Balthasar Boncompagni et sur les
rapports que existent entre les ouvrages et les travaux mathematiques
des Arabes, Atti della Academia dei Lincei 14 (1861), 301-324.
Chapter 2
Cubical Clusters
In Chapter 1 we were concerned with the way translates of a single

cube fit together to tile space. In this chapter we examine tilings by
translates of a finite collection of cubes, which we will call "clus-
ters." Chapters 3 and 4 will treat a special family of clusters that
exists in all dimensions. Before we can state the main results of this
chapter, we need some definitions.
As in Chapter 1, we assume a fixed coordinate system. We con-
tinue to identify each unit cube whose edges are parallel to the axes
with its vertex that has the smallest coordinates. A n n-dimensional
cluster C is the finite union of unit cubes whose edges are paral-
lel to the axes and which have integer coordinates. A cluster is not
necessarily connected
Let C be a fixed cluster in η-space and assume that L is a set of
vectors in η-space such that the set of translates {v + C: ν e L] tile
η-space. (For a given cluster there may be n o such lattice.) If all the
coordinates of all the vectors in L are integers (rational numbers),
we speak of an integer tiling (rational tiling) by C , or simply a Ζ-tiling
(Q-tiling). If L is a lattice we speak of lattice tiling by C. Combin-
ing the two notions, we speak of a Z-lattice tiling and a Q-lattice
tiling.
We will prove the following theorems, all of which concern
tilings by translates of a cluster. T h e main technique we will use
goes back to Schmidt in 1933 [7].
35
Theorem 1. If there is a lattice tiling by cluster C, then there is a Q-

lattice tiling by C.
Theorem 2. //there is a lattice tiling by cluster C that is not a Z-tiling

then there is a Q-lattice tiling by C that is not a Z-tiling.
Theorem 3. Let C be the cluster consisting of a single cube. If there

is a lattice tiling by C without twins, then there is a Q-lattice tiling by
C without twins.
Theorem 4. If there is a tiling by cluster C, then there is a Z-tiling by

C.
T h e o r e m 3 is the key that makes Minkowski's conjecture ac-

cessible by algebraic techniques, as we saw in C h a p t e r 1. We will
combine T h e o r e m 2 and R i d e i ' s t h e o r e m to prove the following
t h e o r e m about clusters that consist of a prime n u m b e r of cubes.
Theorem 5. If cluster C has a prime number of cubes and contains

the η + 1 cubes with centers at
(Ο,.,.,Ο), ( 1 , 0 , . . . , 0 ) , . . . , ( 0 , . . . , 0 , 1 ) ,
then any lattice tiling ofn-space by C is a Ζ-lattice tiling.
This last assertion is quite surprising. If we delete either one of

the assumptions, it is false. For instance, the cluster consisting of the
two unit squares whose centers are (0,0) a n d (2,0) Q-lattice tiles
the plane but does not Z-lattice tile it. Also the cluster consisting
of the four unit squares corresponding to (0,0), (1,0), (1,1) and
(0,1) lattice tiles the plane in such a way that not all the vectors
have integer coordinates.
Exercise 1. Check that the two clusters mentioned tile as claimed.
After proving these theorems we turn o u r attention to the prob-

lem of filling a box with bricks. This is a tiling problem in which we
wish to tile a particular b o u n d e d region with congruent copies of a
very simple cluster. In this case we allow rotations of the clusters.
T h e solution illustrates one of the simpler algebraic techniques for
analyzing tiling problems.
Cubical Clusters 37
1. Reductions
T h e m e t h o d for altering a tiling to one with simpler translating vec-
tors rests on a certain equivalence relation defined on t h e set of
translation vectors. It turns out that the clusters that correspond
to an equivalence class form a cylinder. This cylinder consists of a
union of lines parallel to an axis, and can therefore b e slid back a n d
forth freely parallel to t h e axis without disturbing t h e o t h e r equiv-
alence classes.
Consider two η dimensional unit cubes in η-space correspond-
ing to t h e coordinate vectors ( α ι , . . . , α ) and ( 6 1 , . . . , 6 ) . If
η n
|α ι - bi\ = 1, | a - 621 < 1,· · •, |a„ - 6 „ | < 1 we call the two

2
cubes adjacent. Geometrically speaking, t h e two cubes are adjacent

if their intersection is an (τι - l)-dimensional subset that is part of
a face of b o t h cubes perpendicular to the first coordinate axis. Fig-
ures 1 (a) and (b) exhibit nonadjacent cubes in 2-space and Figures
1 (c) and (d) exhibit adjacent cubes in 2-space.
Exercise 2. Verify that t h e geometrical and arithmetical descrip-

tions of adjacency are equivalent.
(a)
(b)
(c)
FIGURE 1
(a)
(b)
X2
(c) (d)
FIGURE 2
Consider a tiling of η-space by a cluster C. For convenience as-

s u m e that C contains the unit cube with coordinates ( 0 , . . . , 0). Call
two clusters in this tiling adjacent if there are two adjacent cubes
such that o n e of the cubes belongs to one cluster and the o t h e r cube
belongs to t h e other cluster. Figures 2 (a) and (b) show nonadjacent
and Figures 2 (c) and (d) show adjacent 2-dimensional clusters.
Adjacency between clusters induces an equivalence relation on
t h e m , as follows. Let C and C be clusters. If there is a sequence of
clusters Ci, C ,..., C , where C = d , C = C a n d d is adjacent
2 m m
toCj+i, 1 < i < m-l.wesaythatCiseguiVa/eTjf toC'.Thoughwe

have defined the equivalence relation relative to the χ χ-axis, clearly
it could b e defined relative to each of the axes. Figure 3 illustrates
equivalent clusters in 2-space.
If L is t h e set of translating vectors for a tiling by C, we call two
vectors Ιχ and l in L equivalent if their associated sets h + C and
2
l + C are equivalent.
2
Cubical Clusters 39
12
>xi
FIGURE 3
Exercise 3 . Let C be the cluster consisting of just one square in R . 2
Sketch t h e equivalence classes relative to each axis for t h e tiling by

C when t h e translating vectors are
(a) {(m,n):m,n € Z}.
(b) { m ( l , 1/2) + n ( 0 , 1 ) : m , η e Z}.
Exercise 4. Let C b e t h e cluster consisting of the two squares cor-

responding to (0,0) and ( 0 , 2 ) . Sketch the equivalence classes rel-
ative to each axis for the tiling by C when the translating vectors
are
(a) { m ( l / 2 , l ) + n ( 2 , 0 ) : m , n e Z}
(b) { t n ( l , 0 ) + n ( 0 , 4 ) : m , n e Z } U
{ m ( l , 0) + n ( 0 , 4 ) + (0,1): m, η € Ζ).
Tb discover the geometric meaning of an equivalence class

imagine that adjacent clusters are glued to each other along their
shared surface perpendicular to the first coordinate axis. Then if
we shift o n e of t h e m parallel to the first axis, the entire set of clus-
ters equivalent to it is forced to move parallel to that axis the same
amount. T h e different equivalence classes in the tiling would slide
independently of each other, for each is a cylinder whose generator
is parallel to the first axis. N o t e that any such shift parallel to the
first axis produces another tiling by C.
Now assume that the set of translating vectors, L, is a lattice,
hence an abelian group, whose identity element ( 0 , 0 , . . . , 0) we de-
note 0. If h and l are in L and are equivalent, write ii ~ l . N o t e
2 2
that if i is in L, then (Zi -I- Z ) ~ (l + h). It follows that the set

3 3 2
Κ — {I: I ~ 0} is a sublattice of L and the equivalence classes are

translates of K. This is a special case of the following lemma.
Lemma 1. Let ~ be an equivalence relation on the abelian group G.

Assume that ifx, y, and ζ are in G and χ ~ y, then (χ + ζ) ~ (y + z).
Then Κ = {x:x ~ 0} is a subgroup ofG and the equivalence classes
are cosets of K.
Exercise 5. Prove L e m m a 1.
As a consequence of this lemma, the equivalence class Κ we
obtained from the lattice tiling is not merely a cylinder, but is also
a subgroup of L. N o t e that t h e first coordinate of any vector in Κ is
an integer. We use the symbol Κ to denote both the set of vectors
equivalent t o ( 0 , . . . , 0) and the union of the clusters ν + C, ν e K.
T h e context will show whether Κ denotes a subset of L, an algebraic
object, or a union of clusters, a geometric object. Similarly, each
coset of Κ is viewed as both an element of the factor group L/K
and as a cylinder.
We shall exploit our ability to slide cylinders freely parallel to
the axis relative to which they were defined to prove the reduction
theorems. In our arguments βχ denotes the unit vector along the
positive ith axis.
Theorem 1. If there is a lattice tiling by cluster C, then there is a Q-

lattice tiling by C.
Proof. Let L b e t h e lattice of the given tiling. Assume that some

vector in L has an irrational coordinate. For convenience, assume
the vector has an irrational first coordinate, that is, has the form
αχβι Η h a e , where οχ is irrational.
n n
Cubical Clusters 41
Introduce the equivalence relation relative t o the first axis just

described and let Κ b e the equivalence class containing ( 0 , . . . , 0).
Introduce a second equivalence relation on L, setting two of
its vectors equivalent if their first coordinates differ by a rational
number. Let Μ be the equivalence class containing ( 0 , . . . , 0). N o t e
that Κ c Μ and Μ is a sublattice of L b u t is n o t all of L.
It is important that Μ is n o t merely a sublattice of L but a
direct summand of L, that is, there is another sublattice, T , in L
such that Τ φ Μ = L. Tb show that Μ is a direct summand, we
check that if J g L and η is a nonzero integer, a n d nl e M, then
I € M. (This criterion is established in Appendix A.) But if nl is in
M, its first coordinate is rational and therefore the first coordinate
of / itself is rational, hence I is in M.
Since Μ is a direct summand of L there is a sublattice Τ of L
such that L = Τ θ Μ . Let h,..., t b e a basis for the lattice Τ and
r
t i , . . . , t b e a basis for M . Thus 11,..., t is a basis for L and the

r + n n
first coordinate of each of the vectors i i , . . . , t is irrational, while r
the first coordinate of each of the vectors < i , . . . , < „ is rational. r+
Replace i i , . . . , t by r
t'l — t\ + biC\, • • •, t' = t r r + b e\,

r
where b\,..., b are real numbers to b e chosen in a m o m e n t .

r
Let t't = U for each i, r + 1 < i < n. Then replace t h e vector

xiî -\ + x t n n in L by xit[ + h x t' .
n n Let V b e t h e lattice
generated by the vectors t\,...,t' . n
Since
xit'x Η h x„t' n = xi(ti + b ei)
x Η τ- x (t
r r + b ei)
r
+ ^r+lir+l Η f" ^n*n

= χι*ι Η 1- x t n n + (xibi -\ h x b )ei,
r r
each vector in V is obtained from a vector in L by adding a vector

parallel t o the first axis. N o t e that since i f is a subset of M , its vec-
tors remain unchanged. M o r e generally, if νχ = xiti Η h xt n n
and v = 2/iti + · · + y <n are in t h e same coset of K, they a r e

2 n
changed by the same amount. Indeed, if vi - v € K, then υχ - v € 2 2

Μ, hence χ, - & = 0, or x = yt for each i, 1 < i < r. Thus, since

t
all vectors in a cylinder ν + Κ are changed by the same amount, the

cylinder, as a geometric object, is simply slid onto itself by a motion
parallel t o the first axis. From this it follows that the cluster C tiles
η-space by the vectors of L'.
Next we show that b . . . ,b can be chosen so that t h e first
i t r
coordinates of t ' , . . . , t' a r e rational. Let an b e the first coordinate

x r
of U, 1 < i < r. Indeed, choose 6» to be any n u m b e r of t h e form

- a u + qu qi € Q.
T h e same argument can then b e repeated for t h e second coor-
dinate, a n d so on, through t h e n t h coordinate. Since the alteration
each time affects only o n e coordinate, the sequence of η steps re-
places L by a lattice, with only rational coordinates, which still pro-
vides a tiling by C. •
T h e r e is a wide range in which t o choose the bi in t h e preceding
proof; bi could be any n u m b e r in -an + Q. In particular, w e could
choose bi such that bi+an is not an integer. This proves T h e o r e m 2.
T h e proof of T h e o r e m 3, which concerns t h e absence of twin cubes,
is m o r e involved. F o r convenience, we restate the theorem.
Theorem 3 . If there is a lattice tiling by a cube without twins, then

there is a Q-lattice tiling by a cube without twins.
Proof. (In this case the cluster C has just o n e cube.) We wish to
show that if there are n o twins in the tiling by L, then we can choose
the bi, 1 < i < r, as defined in the proof of T h e o r e m 1, t o avoid
twins in V a n d still have bi + an rational.
Since t h e lattice V is homogeneous, if it has twins it has twins in
which o n e of the cubes is at t h e origin. Since w e move cubes parallel
to t h e first axis to produce V, we d o not introduce twins that share
a face perpendicular to that axis. We must show how to avoid the
creation of twins perpendicular to any of t h e other η - 1 axes.
Consider, for instance, the second axis. In this case we wish to
avoid the presence of the vector ( 0 , 1 , 0 , . . . , 0) in V. Since we alter
vectors in L only by a change in the first coordinate, ( 0 , 1 , 0 , . . . , 0)
can a p p e a r in V only if a vector of the form (ζ?, 1 , 0 , . . . , 0) is in
Cubical Clusters 43
L. (The subscript 2 reminds us that we a r e concerned with twins

forming along a surface perpendicular to the second axis.)
Now
( z , 1, 0, . . . , 0) = X 1 * 1 + · • · + Χ 2 η ί η ,
2 2
where at least o n e of t h e x , 1 < i < r, is not 0 since t h e vector

2 i
( z , 1 , 0 , . . . , 0) is moved in our construction. T h u s

2
2 2 = X 2 1 0 1 1 "I 1" X n O n l ,
2
where at least o n e of t h e x i , 1 < i < r, is not 0. We must choose

2
b\,..., 6 so that bi is in - α ; ι + Q and

r
22 + Η 1- X2rb φ 0. r
But before we make such a choice, we must k e e p in mind that

there may be other vectors in L of t h e form (z, 1 , 0 , . . . , 0). Con-
sider another such vector in L, say (y, 1 , 0 , . . . , 0 ) . T h e vec-
tor (y - z ,0,...,
2 0), being t h e difference of vectors in L, is also
in L. In fact, it is in Κ since it is in t h e same cylinder as ( 0 , . . . , 0 ) .
T h u s y - z is an integer a n d therefore ( z , 1 , 0 , . . . , 0) differs from
2 2
(y, 1 , 0 , . . . , 0) by an element of M. Consequently their components

in Τ are identical. Moreover y = z + u, where u is an integer. T h e
2
first coordinate of the translation of (y, 1 , 0 , . . . , 0) is then
z + u + X2161 -I
2 V x b, 2r r (1)
a fact that will b e used in a m o m e n t . In o r d e r that (1) is n o t 0, we

must choose t h e ftj's so that n o n e of t h e n u m b e r s
X21&1 + Η x iv
2 r
is in the set -z + Z. 2
Before we show that such a choice is possible, recall that twins

m a y b e formed that share a face perpendicular to t h e third axis, a n d
so on. To avoid such twins forming with a face shared perpendicular
t o the ith axis we must also d e m a n d that t h e choice of 6 1 , . . . , b r
does not cause a number of t h e form
Xtl^l + · · · + Xi b r r
to lie in t h e set of the form -z + Z, 2 < i < n. (The n u m b e r s

t
χα,..., Xir are analogous to the numbers x i , . . . , Χ 2 · ) 2 Γ
Consider again the second axis. Let u be an integer. T h e set of

vectors ( 6 1 , . . . , b ) such that
r
X21Ö1 Η h X2rK = - 2 2 + «
is an (r - 1)-dimensional hyperplane in R . A s u varies through Z,
r
we obtain a set of parallel hyperplanes in R at the fixed distance

r
1
y/x^l + • ·· + x\ r
from each other. A s long we choose ( 6 1 , . . . , 6 ) not o n any of these

r
hyperplanes, we avoid forming twins that share a face perpendicular

to the second axis. Similar reasoning holds for each of the o t h e r axes
of interest.
To be sure that V has n o twins we choose ( 6 1 , . . . , b ) not to lie r
in any of t h e η - 1 families of hyperplanes just mentioned.

T h a t this is possible is a consequence of the fact that b can t
be chosen anywhere in -an + Q, while consecutive hyperplanes

in the η - 1 families are at a fixed distance from each other. This
construction then gives us a lattice V without twins and such that
the first coordinate of each of its vectors is rational.
T h e n we repeat the process, starting with V, to m a k e all the
second coordinates rational. After η successive applications of the
same process we obtain a rational lattice without twins. •
Exercise 6. Write out a careful proof that the union of t h e η - 1

families of hyperplanes in t h e preceding proof is not R . r
Exercise 7. T h e proof of T h e o r e m 3 could be simplified by using

the following theorem: R is not a denumerable union of subsets of
n
the form ν + W, where ν € R and W is a subspace of dimension

n
at most η - 1. Prove this theorem. (Suggestion: start with t h e case

η = 2.)
(We should mention that the t h e o r e m in Exercise 7 is a special

case of a topological theorem: Let S\, S , • • •, Sk, • • • be a denumer-
2
Cubical Clusters 45
able family of closed subsets of R . Assume that no set Sk contains a

n
ball. Then the union of the family contains no ball. For a proof see
[4, p. 87].)
T h e proof of T h e o r e m 4, which asserts that if there is a tiling
by C then there is a Z-tiling by C, is much easier. First translate
the cylindrical equivalence classes so that their vectors all have an
integer first coordinate. Starting with this tiling, carry out the same
procedure for the second coordinate. After η successive applica-
tions of this procedure we obtain a Z-tiling by C.
2. Clusters with a prime number of cubes

We now turn to the theorem concerning clusters with a prime num-
ber of cubes.
In addition t o T h e o r e m 2, we will need a way to tell whether a
Q-lattice tiling is a Z-lattice tiling. We first develop such a method.
Consider a cluster C in R , which consists of cubes described
n
by the vectors c i , . . . , c . We also use the same symbol, C, to d e n o t e

q
the set of vectors c i , c , . . . , c . We may assume that c = ( 0 , . . . , 0).

2 q x
Let L be the lattice of translating vectors of a Q-lattice tiling of R n
byC.
Since L is a rational lattice there are positive integers r , . . . , r„
x
such that L is contained in the lattice V generated by
e'i = ( V i ) e i , . . . , < = (l/r„)e„.

r
Moreover, we choose the rVs to b e the smallest positive integers so

that L c U. If Ti = 1, then e\ = ej and the ith coordinate of every
vector in L is an integer.
That translates of C by the vectors in L tile R is equivalent t o
n
the assertion that each vector V e V is uniquely expressible in the

form
I' = 1 + Ci+ x\e\ Η 1- x e' ,

n n
where I e L, ^ 6 C, and x< is an integer 0 < χ, < rj - 1 , 1 < i < n.

See Figure 4, which shows a cluster composed of two squares, where
r\ is 1 a n d r is 3.
2
12
l'
0 ] :2 ;} <1
e'i =ei I' = l + ca + Oe[ + 2e' 2
FIGURE 4
Let Di = {0, e'i,..., (r< - l)e<}, 1 „.
This factorization of V induces a factorization of the quotient group,

G = L'/L,
G = C* + £>ί + · · · + £>;, (2)
where C* = ( C + L)/L and = (D< + L ) / L .
Exercise 8. Let t h e abelian group Η b e factored as Η = A+B+C,

where A is a subgroup of Η a n d i? a n d C a r e subsets of H. Prove
that this induces a factorization of the g r o u p G = Η/A, namely
G = (B + + (C + A)/A.
For convenience we rewrite (2) in multiplicative notation. Let-
ting üi = e\ + L, we have
Dl = {e,ai,al..., a?' },
1
1 ; , (3)

Cubical Clusters 47
where t h e D* are cyclic subsets of G or just {e}, where e is the iden-

tity element of G.
Now we are ready to develop the test created by Szabo [8] for
telling whether a Q-lattice tiling by C is a Z-tiling.
Lemma 2. Let C be a cluster in R that contains the η + 1 vectors

n
( 0 , . . . , 0), e i , . . . , e . / / C* in (2) is a subgroup of G = L'/L, then

n
L c Z , that is, the Q-lattice tiling by translates ofCby the vectors in

n
Lisa Z-latticetiling.
Proof. Each vector I e L can be written uniquely in the form
I = zi{l/ri)ei + · · · + z (l/r„)e ,
n n
where t h e Zj's are integers. We wish to prove that η divides Zj.

In any case, Ζχ = UiTi + v,, where t t j and Vi are integers a n d
0 < Vi < Γχ - 1. We wish to show that Vi = 0.
We have
I = uiei + · · · + u en n + (wi/n)ei + · • · + (v /r )e . n n n (4)
Let e = (1/τ·*)βί and let / be the natural homomorphism / : V —•

t
L'/L. D e n o t e / ( e j ) by Oj. Applying / to (4) yields this equation in

G:
β = (ο; Γ···(α;-)-«α ··χ-.

ι
1
ι
(5)
Since C contains t h e cubes whose centers are ( 0 , . . . , 0),
e χ , . . . , e , C* contains t h e (distinct) elements e, a j , . . . , a £ . T h e r e -
n
1 n
fore, since C* is assumed to be a group, C* contains the element
Κ>Γ···ΚΤ",
which appears in (5). In view of the factorization G = C*D{ •• • D *
and the fact that e e C*, we conclude from (5) that a?' = e. (Other-
wise there would b e two factorizations of e, the other being e · · · e.)
T h u s Vi = 0 and t h e tiling is a Z-tiling. •
We are now ready to prove T h e o r e m 5 in which the cluster has

a prime n u m b e r of cubes.
Theorem 5. If cluster C has a prime number of cubes and contains

the cubes corresponding to ( 0 , . . . , 0), e ,...,
x e , then any lattice tiling
n
by C is a Z-tiling.
Proof. If there is a lattice tiling by C that is not a Z-tiling, then,

by T h e o r e m 2, there is a Q-lattice tiling by C that is not a Z-tiling.
Therefore we may restrict o u r attention to Q-lattice tilings by C.
Let L, L', G, C*, D* denote the same sets as in the preced-
ing proof. Delete any that consists only of the identity element,
£>,* = {e}. Relabeling the remaining cyclic sets D\*,..., D^, m<n,
we have the factorization
G = C*Dl.-D' . m (6)
Since D* - {e, at,a ,...2

, α [ ' ~ } and
χ
and e are distinct ele-
ments of C*, £)* is not a subgroup of G. As we saw in Chapter 1,
D* can be expressed as the product of cyclic sets of prime orders
that are not subgroups. Rewriting (6) in terms of these sets we now
have
G = C'B ---B ,
i k (7)
where each factor has prime order and none of the cyclic sets Bi is
a subgroup. Redei's t h e o r e m implies that C* is a subgroup. L e m m a
2 then tells us that the tiling has only integer coordinates. •
Incidentally, in the next chapter we will exhibit clusters C of

arbitrarily high dimension η that contain ( 0 , . . . , 0), e\,..., e , have
n
a prime n u m b e r of cubes, and lattice tile R . This assures us that

n
the t h e o r e m we just obtained is far from vacuous.
3. Tiling a box by congruent bricks

In the next chapter we return to the study of tiling η-space by trans-
lates of some particular clusters. But now we will illustrate some of
the many results concerning tiling some bounded region by congru-
ent (not necessarily parallel) copies of a given cluster, by consider-
ing the following question that was treated by de Bruijn [1].
Cubical Clusters 49
For which positive integers a, b, and η can we tile the ο χ b

rectangle by congruent copies of the 1 χ η "brick"? By counting
squares we see that if the brick does tile, then η divides ab. Also,
if η divides at least one of the integers α and b, then the translates
of the brick tile the rectangle. T h e condition " n divides ab" is not
sufficient to assure the existence of a tiling: For instance, the 1 x 4
brick does not tile the 2 χ 2 rectangle.
We will analyze the problem of determining when the 1 χ η
brick tiles an α χ 6 rectangle first by a counting argument. T h e n we
will express the argument algebraically.
To illustrate the idea consider whether the 1 χ 4 brick tiles the
6 χ 10 rectangle. Label the 60 squares 0, 1, 2, or 3, labeling the
square (i, j) with the remainder of i+j modulo 4, as in Figure 5. N o
matter where the brick is placed it covers each of the four symbols
the same n u m b e r of times, namely once. So if there is a tiling, each
of the symbols 0 , 1 , 2 , 3 must appear the same n u m b e r of times in the
6 χ 10 rectangle. However, as may be checked, the four symbols d o
not a p p e a r equally often. Therefore the 1 χ 4 brick does not tile the
6 χ 10 rectangle. This argument could b e used to prove the following
1 2 3 0 1 2 3 0 1 2
0 1 2 3 0 1 2 3 0 1
3 0 1 2 3 0 1 2 3 0
2 3 0 1 2 3 0 1 2 3
1 2 3 0 1 2 3 0 1 2
1 2 3 0 1 2 3 0 1
(0,0)
FIGURE 5
theorem. However, it will b e easier to prove it algebraically, with

polynomials and complex numbers doing the bookkeeping for us.
These tools will also come in handy when proving Redei's t h e o r e m
in Chapter 7.
Theorem 6. Let n, a, and b be positive integers. A necessary and suf-

ficient condition that the l χ η brick tiles the a χ b rectangle is that η
divides at least one of a and b.
Proof. If η divides α or 6, then it is possible t o tile the ο χ b rectangle

with parallel bricks. Let us show the converse, that if the 1 χ η brick
tiles the α χ b rectangle, then η must divide a or b.
Introduce a coordinate system whose axes lie on two of the
edges of the a χ b rectangle such that the coordinates of the four
vertices of the rectangle are (0,0), (a, 0), (0, b), and (a, b). Record
each of the ab unit squares that m a k e u p the rectangle by the coor-
dinates of its lower left corner, as in Figure 6. For each such corner
(i, j), 0 < i < a - 1 , 0 < j < b - 1, introduce the monomial xy. i j
T h e n record the ab squares in the rectangle by the polynomial
o-lb-l
i=0 j=0
which equals (1 + χ + χ Η 2
+ χ α - 1
) ( 1 + y + y + ••• +
2
y ~ ).
b l
χ
FIGURE 6
Cubical Clusters 51
W h e n the brick is placed with its long side parallel to the χ axis
its η cells are recorded by a polynomial of the form
xiyi + a-w-y + xi+2 yj + ... + a;*+»-y,
which equals
χν(1-Γ-χ + χ + ···+χ 2 η-1

).
Similarly, when placed parallel to the y axis it is recorded by a poly-

nomial of the form
xV(l + 2/ + 2 / + 2
---+2/ ~ )- n 1
Therefore if the brick tiles t h e a χ b rectangle, then there are

polynomials P(x, y) and Q(x, y) such that
P ( x , y ) ( l + x + --- + x - ) + Q ( x , y ) ( l + y + --- + y - )
n 1 n 1
(8).
= (1 + χ + · · · + x e _ 1
) ( l + y + •••+ y 6 _ 1
)
Now replace χ and y throughout (8) by a primitive n t h root of unity,

say ω = e . Since
2 w i / n
^jn j
1+ ω + ω + ·• · + ω""
2 1
= = 0,
ω - 1
w e have
P( W ) W )(0)+g(a; ( ;)(0)=(^£y)
> 1 (ί=τ) "
Therefore at least one of the n u m b e r s ω - 1 or uj — 1 is 0. In the α b
first case η divides a; in t h e second, η divides b. This concludes t h e

proof. •
Exercise 9. W h e n does a 1 χ 1 χ η brick tile an a χ b χ c box, where

η , a, 6, and c are integers?
Exercise 10. (Extracted from [5]) Let rn, n, a, a n d b b e positive

integers. Show that if the τη χ η rectangle tiles t h e α χ b rectangle
then
(a) each side of t h e α χ & rectangle can b e expressed in t h e form

xm + yn, for nonnegative integers χ a n d y;
(b) m divides a or b,
(c) η divides a or b.
Exercise 11. Establish t h e converse of Exercise 12, in case

(a) m divides a a n d η divides b (or m divides b a n d η divides a ) .
(b) m a n d η both divide α (or both divide b).
Exercises 10 a n d 11 answer the question, " W h e n does an m χ η

rectangle tile an α χ & rectangle?" T h e analogous question in higher
dimensions has not been settled.
Problem 1. W h e n does an m i χ m j χ • · · χ m „ brick tile an

οι χ α χ · · · x α η-dimensional box?
2 η
Exercise 12. Show that if the m i χ m χ · · · χ m brick tiles the

2 n
αϊ χ α χ · · · χ α box, then each nii divides at least o n e of the α*.

2 η
T h e problem just mentioned is completely solved in t h e case

of a special type of brick, called "harmonic." Let m i < m < · · · < 2
m be the dimensions of an η-dimensional brick. If

n divides m j i +
for 1 < i < η - 1 , we call the brick harmonic. A n οι χ α χ · · · χ α

2 η
box is called a multiple of t h e m i χ m χ · · · χ m „ brick if there is a

2
permutation φ of the indices 1 , 2 , . . . , η such that mi divides a ^ j .
Theorem 7. If a harmonic brick tiles a box, the box is a multiple of

the brick.
Proof. Let the dimensions of the brick be m , m , . . . , m „ , where

x 2
mi divides m i , 1 < i < η -1. Using the technique in t h e proof of

i +
T h e o r e m 6, w e see that m „ divides at least o n e of t h e dimensions

of the box, which w e label a . n
Consider an ( η - 1)-dimensional face of the box, of dimen-

sions αϊ χ α χ · · · χ α _ ι . This face is entirely filled with
2 η
(η — 1)-dimensional bricks of the form m i χ m χ · • · χ m _ i , . . . ,

2 n
m χ J7i3 χ • · · χ m . Each of these η types of (η - 1)-dimensional

2 n
Cubical Clusters 53
bricks is a multiple of the m i χ m χ · · • χ τη -ι brick, since the n-

2 η
dimensional brick is harmonic. By induction, the αχ χ α χ · • · χ α _ ι

2 η
box is a multiple of the m i χ m x · • · x m _ i brick. Since, in

2 n
addition, m divides o , the theorem is proved.

n n •
Exercise 13. Show that if the m i χ m brick is not harmonic there

2
is a rectangle that it tiles that is not a multiple of the brick. (Hint:

Consider the ( m i + m ) χ m i m rectangle. Try it first for m i = 2,
2 2
m 2 = 3.)
Exercise 14. W h a t boxes does the 1 χ 2 χ 4 brick tile?
Exercise 15.
(a) O n e square is deleted from an 8 χ 8 chessboard. Assume that
the remaining 63 squares can be tiled by the 1 χ 3 brick. Using
complex numbers, show that the deleted square must be o n e
of the four squares that lie o n the two diagonals and share a
vertex with o n e of the four squares at the center of the board.
(This proof is due to Mackinnon [6]. A proof using a coloring
argument is t o b e found in G o l o m b [3].)
(b) Show that if the deleted square is o n e of the four described in
(a) then the 1 χ 3 brick does tile the remaining 63 squares.
A s the problem of tiling a box with congruent bricks suggests,

deciding whether some b o u n d e d or u n b o u n d e d region can b e tiled
by congruent copies of some given figures can easily be quite diffi-
cult. A s another example, consider the triangular array in Figure 7,
which is part of the tiling of the plane by regular hexagons.
D e n o t e such a triangle with η hexagons on a side, T . Conway n
and Lagarias [2] considered the question "Which T can be tiled n
by congruent copies of T ? " Clearly, 3 must divide n(n + l ) / 2 , the

2
total n u m b e r of hexagons in T . So η = 0 or 2 (mod 3). Using

n
the free group o n two generators they obtained a much stronger

result: η = 0, 2 , 9 , or l l ( m o d 12). Using the same technique, they
also proved that T„ can never be tiled by copies of the set of three
<
FIGURE 7
hexagons in a line (the middle one being adjacent to the o t h e r two).

Conway [9] later showed that for 2 < m < n, T does not tile T„.
m
Exercise 16. For which integers η can T n be tiled by copies of a

set of two adjacent hexagons?
Deciding whether congruent copies of a given set of clusters

tile η-space can also be difficult. In fact, Berger in 1966 and Robin-
son in 1971 proved that there is n o general algorithm for settling the
question. In the next chapter we will consider tilings by two partic-
ular families of clusters in η-space. Though we construct tilings for
many of these clusters, the general question of which ones tile is far
from settled.
References
1. N. G. de Bruijn, Filling boxes with bricks, Amer. Math. Monthly 76
(1969), 37-40.
2. J. H. Conway and J. C. Lagarias, Tiling with polyominoes and combi-
natorial group theory, / Comb. Theory, Series A 53 (1990), 183-208.
3. S. Golomb, Polyominoes, Princeton Univ. Press, 1994.
Cubical Clusters 55
4. J. C. Hocking and G. S. Young, Topology, Addison-Wesley, Reading,

1961.
5. D. A. Klarner, Packing a rectangle with congruent ΛΓ-ominoes, /.
Comb. Theory 7 (1969), 107-115.
6. N. Mackinnon, An algebraic tiling proof, Math. Gazette 465 (1989),
210-211.
7. T. Schmidt, Über die Zerlegung des n-dimensionalen Raumes in git-
terförmig angeordnete Würfeln, Sehr. math. Semin. u. Inst, angew.
Math. Univ. Berlin 1 (1933), 186-212.
8. S. Szabo, On mosaics consisting of multidimensional crosses, Acta
Math. Acad. Sei. Hung. 38 (1981), 191-203.
9. W. P. Thurston, Conway's tiling groups, Amen Math. Monthly 97
(1990), 757-773.
Chapter 3
Tiling by the Semicross and Cross
W h e n we hear the expression "convex body" we probably visualize

such famous sets as the ball, the cube, or the tetrahedron. However,
the expression "non-convex body" triggers n o specific image, just
the general sense of an object with dents. In this chapter and the
next we explore two families of non-convex sets that may well b e
viewed as the prototypes of non-convex bodies.
These two particular families of clusters in η-space have drawn
the attention of mathematicians for several reasons. First, their
tiling, packing, and covering properties can be analyzed with t h e
aid of existing algebraic and combinatorial tools. Second, they raise
many new questions, even about structures as simple as unite cyclic
groups. Third, they are a convenient source of examples a n d coun-
terexamples for questions concerning bodies that are not convex.
Finally, they also appear naturally in such a real-world application
as coding theory. In this chapter we define these two families and
examine the way they tile η-space. In the next chapter we look at
their packings and coverings. At the end of this chapter we sketch
their history.
1. Definitions
In this chapter we will restrict our attention to Z-tilings. Recall that
if a cluster tiles R then it tiles Z ( T h e o r e m 4 in C h a p t e r 2). How-
n n
ever a cluster may lattice tile R but not lattice tile Z , as was shown
n n
57
in Chapter 2 by the cluster consisting of two squares separated by a

square. Throughout, the point ( c i , . . . , c„) 6 Z will represent the
n
cube ( c i , . . . , c ) + Q, where Q is the cube { ( χ χ , . . . , x ) : 0 < x <

n n {
1}. In short, we will use Z to represent R .

n n
For a positive integer k consider the kn + 1 points
(0,0,. ..,0), (i,0,...,0), (0,i,...,0),...,(0,0,...,i),
1 < i < k. Any translate of this set by an element of Z is called n
a (k, n)-semicross. Similarly, any translate by elements of Z of the n
set of 2kn + 1 points
( 0 , 0 , . . . , 0 ) , (ΐ,Ο,.,.,0), (0,v..,0),...,(0,0,...,i),
1 < \iI < k, is called a (k, n)-cross. Figure 1 shows a (3,2)-semicross

and a (2,3)-cross.
Though these clusters are not convex, they are at least "star
bodies." A star body is a subset of R that contains a point Ρ such
n
that for every point Q in it, the entire chord PQ lies in the body, as
shown in Figure 2.
FIGURE 1
FIGURE 2
T h e cross and semicross are clusters and so we may apply the

definitions of Chapter 2.
We therefore may speak of a tiling or lattice tiling by the (fc, n)-
semicross or say that the (k, n)-semicross tiles or lattice tiles Z . n
Similar statements hold for the (k, n)-cross.

We immediately face a basic question.
Problem 1. For which positive integers k and η does the (fc, n)-
semicross or the (fc, n)-cross tile Z ? Lattice tile Ζ Ί
n η
Even the lattice tiling part of this problem is far from being
resolved. To suggest the complexity of the problem, we mention that
the (fc, 10)-semicross lattice tiles Z w
only for fc = 1, but that the
(fc, 12)-semicross lattice tiles Z for fc = 1,2,3, and 10.
12
This chapter presents some of the results motivated by Prob-

lem 1.
A (fc, l)-semicross is an interval of fc + 1 points, which tiles Z ,1
and only as a lattice. A (fc, 2)-semicross resembles the letter L and

tiles Z , again only as a lattice, as shown in Figure 3.
2
Exercise 1. Verify the last statement.

A (fc, l)-cross is an interval of 2fc -I-1 points, which tiles Z , and
1
only as a lattice. T h e (1,2)-cross tiles Z , again only as a lattice.

2
FIGURE 3
Exercise 3. Show that for fc > 2, the {k, 2)-cross does not tile Z .
2
It is not hard to show for η > 2 that if the a r m of a cross is

too long, then the cross cannot tile Z " . T h a t is the content of the
following theorem [10].
Theorem 1. Ifn>2 and k > 2n - 1, the (fc, n)-cross does not tile
Z".
Proof. Suppose that η > 2 and that the (fc, n)-cross tiles Z " . Con-
sider any translate in Z of the (fc + l ) points in the square array
n 2
{(*,j,0,...,0) : n - 2 0's; 0 < i, j < fc}.
Figure 4 illustrates the case fc = 3.

Call such a translate a "tray." Two (fc, n)-crosses whose centers
lie in a tray overlap. Consider a tiling of Z " by trays. Since each
tray contains at most one center of a cross in the alleged tiling, the
density of centers is at most l/(fc + l ) . However, the density of
2
centers is l/(2fcn + 1). We have therefore
2fcn + 1 > (fc + l ) 2
or fc < 2n - 2. This proves the theorem. •
FIGURE 4
As we will see, the (2,3)-cross lattice tiles Z and for an infinite

3
number of even dimensions n, the ((n - 2 ) / 2 , n)-cross lattice tiles

Z . These facts suggest the following problem.
n
Problem 2. If t h e (fc, n)-cross tiles Z , η > 3, is fc less than n ?

n
For lattice tilings by the (fc, n)-cross, η > 3, it is known that

fc < η - 1, and we present a proof in the next section. It may not b e
the best possible result, and we raise the following problem.
Problem 3 . If t h e (fc, n)-cross lattice tiles Z , η > 4, is fc less than

n
n / 2 ? If it tiles Z " ?
T h e situation for t h e semicross is quite different. In this case

we d o not know whether there is a bound on the length of the arm
of a semicross that tiles Z , η > 4. (For η = 3 t h e bound is 1.) O n
n
t h e other hand it is known that for η > 3 in the case of lattice tilings
of Z by the (fc, n)-semicross, we must have fc < η - 2, and this is
n
t h e best possible general result, though for specific η the bound can
b e much lower. This is proved in t h e next section.
Problem 4. For fixed η > 4 is there an upper b o u n d o n the set

of integers fc such that the (fc, n)-semicross tiles Z " ? (For η = 1 or
η = 2 there is n o such bound; for η = 3, the b o u n d is 1.)
Assume that the (fc, n)-semicross lattice tiles Z through trans-

n
lations by the vectors in t h e subgroup Η of Z " . This condition is

equivalent to the fact that the fcn + 1 elements of the semicross
whose corner is at the origin represent t h e distinct cosets of H.
Let ej = ( 0 , . . . , 0 , 1 , 0 , . . . , 0), where the 1 is in t h e jth place. Let
G = Z /H n
and let / : Z —• G be the natural homomorphism.
n
Let Sj = f(ej). T h e n the elements {isj : 1 < i < fc, 1 < j < n } ,
together with 0 6 G, represent each element of G exactly once. In
a sense, we may view G schematically as a semicross, as suggested
in Figure 5, which corresponds to the case fc = 3 and η = 5.
Conversely, let G be an abelian group of order fcn + 1 . Assume
that t h e r e are η elements, s\, s ,...,
2 n s , in G such that each ele-
m e n t of G \ {0} is uniquely expressible in the form isj, 1 < i < fc,
3äi
2s ι
FIGURE 5
1 < j < n. (G \ {0} denotes the set of nonzero elements of G.)

Define / : Z —• G by setting f ( e j ) = Sj. T h e n translates of a
n
semicross by the kernel of / lattice tile Z . n
If G is an abelian g r o u p of order fcn + 1 in which we

can find a set of η elements with the property described,
we say that { 1 , 2 , . . . , k} splits G, with splitting set { s i , s , . . . , s „ } .
2
Extensive computations, together with some theorems, suggest that

splittings are quite rare.
Similar reasoning shows that the (k, n)-cross lattice tiles Z if n
and only if there is an abelian group G of order 2kn + 1 in which

there are η elements s i , s , . . . ,s , such that each element in G \ { 0 }
2 n
is uniquely expressible in the form isj, 1 < |i| < fc, 1 < j < n. If
such a set exists we say that { ± 1 , ± 2 , . . . , ±fc} splits G with splitting
set { s i , s , . . . ,s }.
2 n
For convenience, let 5(fc) = {1,2, . . . , f c } and F(k) =

{ ± 1 , ± 2 , . . . , ±fc}. (S stands for "semicross" and F stands for "full
cross".) So instead of treating lattice tilings of Z by the (fc, n ) -
n
semicross or (fc, n)-cross, we examine splittings by S(k) or F(fc) of

abelian groups of o r d e r fcn + 1 or 2fcn + 1 respectively. T h e next
section uses this approach.
Exercise 5. Show that if the (fc,n)-cross Z-lattice tiles R , n

then
the (fc, 2n)-semicross Z-lattice tiles R . 2n
This exercise raises t h e question, "If the (fc, n)-cross tiles Z , n
does the (fc, 2n)-semicross tile Z ? " , which is answered in the next
2 n
exercise.
Exercise 6. Let S be a finite set of points in Z " such that Sn(—S)

is the origin of Z " . ( - S denotes t h e set {-s : s e 5 } . ) Assume that
SU(-S) tiles Z by translates by the set H. Let Τ = (S, 0) U (0, S),
n
where 0 is the origin of R . T h e n Γ is a subset of Z . Prove that

n 2 n
translates of Τ tile Z . (Suggestion: Use the set of elements of the

2 n
form (χ, χ - h), χ 6 Z , h € H, as translating vectors.)

n
2. Bounds on the lengths of the arms

We are now in position t o put u p p e r b o u n d s on t h e length of t h e
arm, fc, in case a (fc, n)-semicross o r (fc, n)-cross lattice tiles Z " . We
treat the cross first since t h e proof, to be found in [16], is shorter.
Theorem 2. Ifn>2 and the (fc, n)-cross lattice tiles Z " , then fc <
η - 1.
Proof. It is easy to check that t h e theorem is true for η = 2, so

we consider only η > 3. Assume that F(k) splits t h e finite abelian
g r o u p G with splitting set {βι, s , . . . , s }. 2 n
We first show that for each integer i, 2 < i < n, there a r e

integers x y such that fc + 1 < Xi < 2n - 1,
it t < fc a n d x^si +
ViSi = 0.
For simplicity, take i = 2. Consider the 2n(fc + 1) elements
aiSi+a232,0 <a\< 2n—1,0 < <z < fc.Since2n(fc+l) > 2 f c n + l , 2
t h e order of G, there are distinct couples ( 6 1 , 6 2 ) and (ci, c ) , 0 < 2
&i) ci < 2n - 1,0 < 6 > c < fc such that &1S1 + 6 s = c\Si + c s .
2 2 2 2 2 2
It is n o loss of generality to assume that bi > c\. Let d\ =b\ - c\

and d = b - c . T h e n disi + d s = 0, where ( d i , d ) φ (0,0)
2 2 2 2 2 2
and 0 < di < 2n - 1, | d | < fc. If 0 < d < fc, then d i s i = - d s ,

2 x 2 2
which violates t h e fact that s j and s are part of a splitting set. 2

Thus for each integer i, 2 < i < n, there is a pair of integers

(xi, yi) such that fc +1< Xj < 2n- 1, < fc, and XiSi + j / j S j = 0.
Assume that there are distinct integers i and j such that x< =
Xj. Since X j S i + j/jSj = 0 and XjSi + yjSj — 0, it follows that yiai =
yjSj. This violates the splitting condition unless y, = 0 = yj. It
follows that XiSi = 0.
N o t e that the 2/c + 1 elements
-fcSi,...,-Si,0,Si,...,fcei
are distinct since s i is an element of a splitting set. T h u s t h e order

of s i in G is at least 2k + 1. Since X j S i = 0, the o r d e r of si divides
x<; hence x > 2k +1. But x < 2n - 1. Therefore 2k +1 < 2n - 1 ,
t {
and k < η — 1.
If, on t h e o t h e r hand, t h e η - 1 Xi's a r e distinct, we must have
η - 1 < 2n - 1 - (fc + 1) + 1,
since they all lie in the interval [fc + 1 , 2 n - 1]. It follows that fc < n.
Next, using the fact that fc < n, we will show that fc < η - 1.
Consider t h e (2n - l)(fc + 1 ) elements 6iSi + 62^1,0 < 61 < 2 n - 2,
0 2 f c n + n - l > 2fcn+2 > 2 f c n + l .
Reasoning as before, we conclude that for each integer i, 2 < i < η

there are integers Xi and yi such that fc + 1 < X{ < 2n - 2, |y< | < fc,
and X j S i + j/jSj = 0.
If all η - 1 of the Xj's a r e distinct, then from t h e fact that they
are in the interval [ f c + l , 2 n - 2 ] , w e h a v e n - l < 2 n - 2 - ( f c + l ) + l;
that is, fc < η - 1.
If t h e r e is duplication a m o n g t h e Xi's we argue as before, start-
ing with t/jSj = yjSj. This time we conclude that 2fc + 1 < 2n - 2,
hence fc < η - 2.
This concludes the proof. •
Exercise 7. Fill in any omitted steps in the preceding proof.

For semicrosses, the first argument for the corresponding

bound on the arm length was totally algebraic [13]. T h e following
geometric lemma, due to Hickerson, simplifies the proof.
Lemma 1. Let η and k be integers, η > 3, fc > η — 1. Assume that

S(k) splits an abelian group G of order nk+1. Let s and s' be elements
of a splitting set. Then one of these two conditions holds:
(a) There are integers χ and y, l<x<n-2,l<y<k, such that
xs + ys' - 0.
(b) s' = (1 - n)s and G is cyclic with generator s.
Proof. Define f:Z φ Ζ -* G by f(i,j) = is + js'. Let A =

{(*, j)-0<i< η -2,0 < j <k}.
If / : A —* G is not one-to-one, then there are integers χ and y
that sastisfy (a). If it is one-to-one, note that it is also one-to-one o n
B = Au{(n-l,0), (n, 0 ) , . . . , (k, 0 ) } . Now, |J3| = ( n - l)(fc + 1 ) +
fc - (n - 2) — nk + 1. T h u s / , considered only on B, is one-to-one
from Β onto G. H e n c e Β tiles Ζ φ Ζ by the vectors in the kernel
of / . T h e nk + 1 points in Β form an L-shaped set, as indicated in
Figure 6, where fc = 5 and η = 4.
(0,fc)
V
(0,0)*
V
(η-2,0)
Λ
(fc,0)
FIGURE 6
Since Β tiles Z@Z, inspection of Figure 6 shows that (η - 1 , 1 )

is o n e of the vectors in the lattice of translating vectors, that is,
(n - 1,1) lies in the kernel of / . This means that (n - l)a + s' = 0.
Since s and s' generate G and s' = - ( n - l)s, it follows that s
generates G, and G is therefore cyclic. •
With the aid of the lemma, the next t h e o r e m follows quickly.
Theorem 3 . Let η and k be integers, η > 3. If S(k) splits a group G

of order nk + 1, then k < η — 2.
Proof Suppose k > η - 1 . Let s i , s , •. •, s be a splitting set of G,

2 n
a group of order nk + 1. For each index j , 2 < j < n, consider the

pair of elements si and Sj. Assume that for each such j , condition
(a) in L e m m a 1 holds, that is, there are Xj and j/j, 1 < Xj < η - 2,
1 < Vj < k, such that XjSi + yjSj = 0. T h e r e would then b e η - 1
values of Xj in the interval [1, η - 2]. By the pigeon-hole principle,
two of these would be equal, say x = x . It follows that y s =
u v u u
y 8 , violating the assumption that s and s are in a splitting set.

v v u v
Thus there must be an index j such that condition (b) in

L e m m a 1 holds. T h a t means that G is cyclic, s i is a generator of G,
and (1 - n)si is in the splitting set.
T h e same reasoning, with ( l - n ) s i playing the role of s i , shows
t h a t ( l - n ) ( l - n ) s i is also in the splitting set. Since s i i s a g e n e r a t o r
of G, and the order of G is larger than (1 - n ) , the elements s\ and 2
(1 - n) 8\ are distinct. However, since

2
jfc(l - n ) = k + 2 - η
2
(mod fcn + 1),
fc(l - n ) s i = (fc + 2 -
2
n)si,
violating the fact that (1 - n) si and s i are in the splitting set.

2
•
As Exercise 12 will show, the general b o u n d fc < η - 2 in T h e o -

rem 3 cannot be lowered. However, for particular n, it may be much
lower.
3. The search for splittings

Lattice tilings of Z by the semicross or cross led us to introduce
n
the notion of splitting an abelian group by certain sets of integers

S(k) = { l , 2 , . . . , f c } and F(k) = { ± 1 , ± 2 , . . . , ±fc}. T h e ques-
tion,"Which groups do S(k) or F(k) split?," is far from being an-
swered.
For example, S(k) splits C(k + 1) with splitting set {1}, and
splits C(2k +1) with splitting set { - 1 , 1 } . ( C ( m ) denotes the cyclic
group of order m , which we will usually take to be { 0 , 1 , . . . , m - 1 } ,
the additive integers modulo m.) However, it is not known whether
S(k) always splits some other group. In particular, the answer is not
known for S(195) [1], which we will discuss later.
Exercise 8. Show that if F(k) splits G, then S(k) also splits G.
In order to make general statements about which groups S(k)

or F(k) splits, we develop some theorems that relate splittings of
groups A and Β t o splittings of their direct sum ΑφΒ. A n d , rather
than restricting ourselves to the sets S(k) and F(k), we consider
any set of nonzero integers Μ = { m j , m , . . . , m * } , which we call
2
a multiplier set. We say that Μ splits the finite abelian group G if

there is a set of elements s\, 3 2 , . . . , a in G such that each nonzero
n
element of G is uniquely representable in the form rriiSj, 1 < i < k,

1 < j < n, and 0 is not representable in that form. It follows that
\G\ = kn + 1. S is called a splitting set for G.
Let A, B, and G b e groups such that there is an isomorphism α
from A into G, a homomorphism β from G onto B, and the kernel
of β coincides with the image of a . We may think of A as a subgroup
of G and Β as the quotient group G/A. (This includes the case when
G is a direct sum of A and B.) We record this by the diagram
{0} — • A -2-+ G Β —> {0},
where each arrow denotes a homomorphism. T h e groups {0}, A, G,

B, {0} and the four homomorphisms are said to form an "exact se-
quence." (See Appendix Β for an introduction to exact sequences.)
Exercise 9. Show that if G — Α φ Β, then there is an exact se-

quence
{0} — . A -2U G - Λ Β —. {0}.
For any finite abelian group G two questions immediately

come into mind: If a multiplier set Μ splits G, must it split A and
J3? If Μ splits A and B, must it split G ? T h e answers to both ques-
tions are " n o , " as we show by examples. However, with an extra
assumption, the answers turn to "yes."
First of all, Μ = S(5) splits (7(6) and though there is an exact
sequence
{ 0 } _ » C(2) — . C(6) — . C(3) — {0},
it does not split either C(2) or C ( 3 ) . So the first question is settled.
Exercise 10. Show that if Μ splits the cyclic group C ( m ) , then

there is a splitting set that contains 1.
To show that the answer to the second question is also " n o , "
consider Μ = S(3) and the exact sequence
{0} — (7(4) — (7(16) — C(4) — {0}.
Note that Μ splits the two " o u t e r " groups. ( T h e first (7(4)
has the elements 0, 4, 8, and 12. T h e homomorphism from it to
C(16) is inclusion. T h e homomorphism from C(16) to C ( 4 ) is the
remainder modulo 4.) We show that it does not split C ( 1 6 ) . Con-
sider D = { 1 , 3 , 5 , 7 , 9 , 1 1 , 1 3 , 1 5 } C C(16). T h e mapping d -» 3d
of £> consists of two cycles, ( 1 , 3 , 9 , 1 1 ) and ( 5 , 1 5 , 1 3 , 7 ) . T h e r e is n o
loss of generality to assume that if there is a splitting set for C ( 1 6 ) ,
then there is one that contains 1. We call it S. It follows that 3 £ S.
But, in order that 9 b e covered, we must have 9 Ε S. However,
2 1 = 2 - 9 , which shows that S is not a splitting set.
Fortunately, there are some relations between splittings of a
group, a subgroup, and a quotient group. T h e s e relations, devel-
o p e d in [2, 5, 6], are expressed in the following theorems, which
greatly simplify the search for splittings.
Theorem 4. Let Μ split the finite abelian group G. Assume that there
is an exact sequence
{0} — • A G -2-* Β — • {0}. (1)
If each element of Mis relatively prime to the order of B, then Μ splits

A
Proof Let Μ split G, with splitting set S. We show that Μ splits

a(A) with splitting set a(A) Π S.
Let α e A T h e n α (a) = ms, m e M, s e 5 . Applying β,
we have 0 = ßa(a) = mß(s). Since (m, \B\) = 1, ß(a) = 0. By
exactness, s e a(A). T h u s Μ splits a(A), hence A. •
If G is a finite abelian group and there is an exact sequence (1),

then t h e r e is also an exact sequence
{0} — • Β G A —> {0},
that is, the roles of image and kernel can be switched. ( A proof is
outlined in Appendix B.) Consequently, the next theorem follows
immediately from T h e o r e m 4.
Theorem 5. Let Μ split the finite abelian group G. Assume that there
is an exact sequence
{0} —> A -^-> G -^-> Β — • {0}.
If each element of Μ is relatively prime to the order of A then Μ splits

B.
While the proof of T h e o r e m 4 shows how to find the promised

splitting set of A, the proof of T h e o r e m 5 does not give a direct way
t o derive the splitting set of Β from the splitting set of G.
Problem 5. Find a direct proof of T h e o r e m 5.
T h e next theorem shows when splittings of A and Β in (1) im-

ply the existence of a splitting of G.
Theorem 6. Let Μ split the finite abelian groups A and Β and as-
sume there is an exact sequence
{0} — • A -2-+ G -^-> Β — • {0}.
If each element of Μ is relatively prime to the order of A or each ele-

ment of Μ is relatively prime to the order of B, then Μ splits G.
Proof. In the first case the proof is constructive. Let SA —

{ s i , s , . . . ,s } be a splitting set for A and SB = {ti,t ,...
2 a 2 ,tb} be
a splitting set for B. Pick Uj e G, 1 < j < b such that ß(uj) = tj.
T h e n we assert that S = {{u\,u , 2 • • - ,Ub} + a(A)} U a ( S U ) is a
splitting set for G.
To see this, consider g e G, g φ 0. If ß(g) = 0, then g G a(A),
and is of the form m j a ( s j ) , Sj e SA- If ß(g) φ 0, it is of t h e form
mitj, for some tj e S ß . T h u s ß(g-niiUj) = O a n d g - m j U j = α(α)
for some element a 6 A. Since (rrij, |j4|) = 1, α can b e written
as m<a', a' € A. H e n c e g = rmuj + a ( m < a ' ) = m < ( u j 4- ct(a')).
T h u s every </ e G \ {0} has a representation in the form g = m^s,
m , e M , s e 5 , and it is easy to check that the representation is
unique.
T h e second case follows from the first by reversing t h e exact
sequence. •
Exercise 11. Show that t h e representation defined in the proof of

T h e o r e m 6 is unique.
T h e o r e m s 4 , 5 , and 6 suggest that we distinguish between two

types of splittings. A splitting of a g r o u p G in which each element
m in the multiplier set Μ is relatively p r i m e to \G\ is a nonsingular
splitting. Splittings that are not nonsingular a r e singular. It is conve-
nient to distinguish a m o n g t h e singular splittings t h e purely singular,
splittings in which each prime that divides \G\ divides at least o n e
element in t h e multiplier set.
T h e o r e m s 4, S, and 6 tell us that if there is an exact sequence
{0} — • A -^-> G Β — • {0},

where G is a finite abelian group, then Μ splits G nonsingularly if

and only if Μ splits A and Β nonsingularly.
For example, if ρ is a prime that divides no element of M , then
Μ splits C ( p ) if and only if Μ splits C ( p ) , since there is an exact
2
sequence
{0} —> C(p) —. C ( p ) — 2

C(p) —> {0}.
In view of this, the exact sequence
{0} — C(p) — C(p )

3
—> C ( p ) —
2
{0},
then tells us that Μ splits C ( p ) if and only if Μ splits C ( p ) . Induc-

3
tion shows that Μ splits C ( p ) nonsingularly if and only if it splits

n
C(p).
Exercise 12. Prove that if ρ is a prime, fc = ρ - 1, and η = ρ + 1,

then the (fc, n)-semicross lattice tiles n-space.
An argument similar to that before Exercise 12 establishes a

m o r e general theorem.
Theorem 7. Let Gbea finite abelian group and letpi,p2,--,p be r
the prime divisors of the order ofG. Then Μ splits G nonsingularly if

and only if Μ splits C(j>i), for each i, 1 < i < r.
Exercise 13. Prove T h e o r e m 7.
According to T h e o r e m 7, if each element of Μ is relatively

prime to \G\, then to determine whether Μ splits G, it suffices to
determine whether Μ splits C(p) for each prime ρ that divides \G\.
Exercise 14. W h a t can we say about \M\ if Μ splits a group of

o r d e r 5 7 nonsingularly?
3 1 0
In order to examine splittings of a finite abelian group G for

any multiplier set M , we express G as the sum of two special sub-
groups A and Β: Β is the sum of t h e Sylow subgroups corresponding
to primes that divide no element in M ; A is the sum of t h e Sylow
subgroups corresponding t o primes that divide at least o n e element

in M . By T h e o r e m 6, if Μ splits A and B, then it splits G. T h e o -
r e m 4 is almost t h e converse, showing that if Μ splits G it splits A.
T h a t the full converse holds, and Μ also splits B, is a consequence
of the following theorem. T h e proof, which uses a counting argu-
ment, depends on the finiteness of the group G. Without the as-
sumption of finiteness, the t h e o r e m does not hold. To see this, note
that Μ = { - 1 , 1 } splits the infinite cyclic groups Ζ and 4 Z but not
their quotient group C ( 4 ) . (Splitting of infinite abelian groups is
defined like that of finite groups.)
Theorem 8. Let Gbea finite abelian group and
{0} — • A -Û G -^-> Β — • {0}
an exact sequence. If Μ splits G and B, it splits A (By the interchange-

ability of A and B, it follows that if Μ splits G and A it splits B.)
We omit the rather involved counting argument d u e to Hick-

erson [6], which, in a nutshell, is this: Let G b e a finite abelian group
and Η a subgroup of G. Assuming that Μ splits G with splitting set
S a n d also splits t h e quotient group G/H, o n e shows that S Π Η is
a splitting set for H.
Problem 6. Obtain a direct proof of the second part of T h e o r e m

8.
Exercise 15. (If t h e proof of T h e o r e m 8 has b e e n read.) Let G

be a finite group, n o t necessarily abelian, a n d let Η be a normal
subgroup of G. Show that if Μ splits both G and G/H, then Μ
splits H. ( H o w would you define the splitting of any group, abelian
or not?)
Problem 7. Let G be a finite group, not necessarily abelian, and

let Η b e a normal subgroup of G. If Μ splits Η and G, must Μ split
G/H?
Exercise 16. Let G be an infinite abelian group and Η a subgroup

of G. Show that if Μ splits G and G/H, it need not split H. (Hint:
Let G b e the dyadic rationale modulo 1, Η = { 0 , 1 / 2 } , a n d Μ =
{1,2}·)
Exercise 17. Show that 5 (1) splits every finite abelian group. W h a t
does this imply about the (1, n) -semicross?
Exercise 18. Show that F(l) splits every finite abelian group of
odd order. What does this imply about the (1, n)-cross?
Exercise 19. Decide whether 5 ( 2 ) splits C{p), ρ = 3,5,7,11,13,

17,19,23, and 29.
Exercise 20. Show that the (2,3)-cross lattice tiles Z . 3
Exercise 21. D o e s F(2) split C ( 5 ) , C(17), C(85)? What does this

say about certain crosses?
Exercise 22. Prove that 5 ( 8 ) does not split C(105).
Exercise 23. Prove that 5(11) does not split C(210).
The next section uses some of the theorems proved in this sec-
tion to examine splittings by S(k) and F(k).
4. Splitting by S(k) and F(k)

To find out whether the (k, n)-semicross lattice tiles Z , we must
n
find out whether S(k) splits an abelian group of o r d e r kn + 1. A n d

similarly, to find out whether the (k, n)-cross lattice tiles Z™ we
must find out whether F(k) splits an abelian group of order 2kn+1.
In view of the theorems of the preceding section, we face two ques-
tions:
W h a t are the purely singular splittings for 5(fc)?
For which primes ρ does S(k) split C ( p ) ?
These questions and the analogous ones for F(k) are far from
being answered, and we present some of the results obtained so far.
We first obtain a theorem o n purely singular splittings by S(k) or
F(k).
Let G be a finite abelian group and ρ be a prime. A p-group
is a group such that the o r d e r of each element in it is a power of p .
T h e p-dimension of G (written " d i m G " ) is the n u m b e r of p-groups
p
in the factorization of G as a direct sum of cyclic groups of prime

power orders. This direct sum is called the p-component of G. It is
the unique p-Sylow subgroup of G.
T h e next exercise is the key to L e m m a 2, which can b e found
in [6].
Exercise 24. Let ρ b e a prime divisor of the o r d e r of the finite

abelian group G. Assume that Μ splits G with splitting set S. Show
that for m e Μ and a 6 S, ma is not of the form pg,geG if and
only if ρ does not divide m and a is not of the form ph, h e G.
L e m m a 2. Let Gbea finite abelian group such that Μ splits G with

splitting set S. Let ρ be a prime divisor of\G\ and 6 (M)be the number
p
of elements of Μ that are divisible by p. Then
δ (Μ)
ρ < \M\.
Proof. Let pG = {pg : g £ G } a n d £ = G \ pt7. T h e n
|£| \G\-\pG\
\SnB\ =
\Μ\-δ (Μ)
ρ \Μ\-δ (Μ)'
ρ
A straightforward counting argument shows that
|G|
Thus
| G | > | G | - l = |Af||S|>|M||SnB|
_ [M|lG|(l-p- " ) d i m G
\M\ - 6 (M) P
or
\Μ\-δ (Μ)
ρ > |M|(l-p- d i m
" ),
G
from which the lemma quickly follows.

•
Exercise 25. Verify the claim in the preceding proof that \pG\ =
\G\/p " .
dim G
Corollary 1. Let Gbea finite abelian group with a splitting G\{0} =

MS. Let ρ be a prime divisor of\G\ such that the p-component of G
is not cyclic. Then
6 (M)
P < \M\/p\
Corollary 2. Let Μ = S(k) or F(k) and MS be a purely singular

splitting of the finite abelian group G. Then G is cyclic.
Proof. W h e n Μ is S(k) or F(k) and ρ < fc, it is not hard to show

that
MM) > \M\/p .

2
Corollary 1 then implies that G is cyclic.

•
Exercise 26. Carry out t h e details of the proof of Corollary 2.
Corollary 2, together with T h e o r e m s 5, 6, 7, and 8, yields t h e

following theorem, again found in [6].
Theorem 9. If Μ is 5(fc) or F(fc) and splits the finite abelian group

G, then Μ splits t h e cyclic group of the same o r d e r as G, G ( | G | ) .
Proof. Let Ρ be t h e set of prime divisors of \G\ that divide at least

o n e element of M . Let Q be the set of prime divisors of |G| that
divide n o element of M .
Let
tf = £ S y l ( G )
p and AT = £ Syl,(G).
p€P q€Q
T h e n G = Η Θ Κ. By T h e o r e m 5, Μ splits # . Corollary 2 shows

that Η is cyclic. By T h e o r e m 8, Μ splits T h e o r e m 7 then shows
that Μ splits the cyclic group C(\K\). Since Μ splits i / and C(\K\)
we conclude from T h e o r e m 6 that Μ splits Η ® C(\K\), which is
the cyclic group C(\G\). •
Exercise 27. Is the converse of T h e o r e m 9 true? T h a t is, if S(fc)

or F(k) splits C(ro) does it split every abelian group of o r d e r m ?
Corollary 2 asserts that when S(k) splits G purely singularly,

then G is cyclic. W h a t are these purely singular splittings? For in-
stance, if fc + 1 is composite, then S(k) splits C(fc + 1) purely sin-
gularly and if 2fc + 1 is composite S(k) splits C(2fc + 1) purely sin-
gularly. Hickerson has shown, in manuscript, that for fc < 3000, the
only purely singular splittings by S(k) are of C(fc + 1 ) or C(2fc -I-1)
as just described.
Problem 8. Find all the purely singular splittings by S(k).
Exercise 28. Show that 5 ( 2 ) does not have any purely singular
splittings.
Exercise 29. If the result cited for fc < 3000 holds for all fc, for
which values of fc does S(k) have no purely singular splittings?
Exercise 30. Prove that the only purely singular splitting by S(4)
is of C(9).
Now consider the nonsingular splittings by S(k). In this case

we ask, "Which groups of prime o r d e r does S(k) split?" In geomet-
ric terms this question is equivalent to "If a (fc, n)-semicross has a
prime n u m b e r of cubes, when does it lattice tile Ζ Ί" In view of
η
T h e o r e m 5 of Chapter 2, this in turn is equivalent to, "If a (fc, n ) -

semicross has a prime n u m b e r of cubes, when does it lattice tile
n-space?"
For instance, if k + 1 or 2k +1 is prime, S(k) splits at least o n e

such group. In the first case the splitting set is {1} and in the second,
{ - 1 , 1 } . In both cases these splitting sets are subgroups of the mul-
tiplicative group C(p)*, so 5(fc) can be viewed as coset representa-
tives of a subgroup of C(p)*. Strictly speaking, S(k), which consists
of integers, is not a subset of C(p)*, which consists of the residue
classes (mod p) relatively prime to p. However, we may identify
S(k) with subsets of C(p)* without causing any confusion. Indeed
we then have a factorization of the group C(p)*, C(p)* — S(k)H,
where Η is a subgroup of C(p)*, namely, the unique subgroup of
order (p - l)/k.
These splittings suggest that we first look for cases where S(k)
splits C(p)* by being the coset representatives of the subgroup Η
of order (p - \)/k, the splitting set then being Η itself. If there is
such a subgroup H, we say that S(k) coset splits C(p).
Consider, for instance 5(6) and C(103). W h e n the elements of
C(103)* are expressed as powers of the generator 5, we have
1 = 5°, 2 = 5 , 3 = 5
4 4 3 9
, 4= 5 8 8
, 5= 5 , 6 = 5 .
1 8 3
(The exponents, which behave like logarithms, are called indices.)

T h e subgroup Η of order (103 - 1 ) / 6 = 17 consists of the n u m b e r s
5 , 0 < j < 16. Since the indices of 1, 2, 3 , 4 , 5 , 6 are incongruent
e j
modulo 6 , 5 ( 6 ) represents the six cosets of Η in (7(103)*, and 5 ( 6 )

coset splits C(103). (It follows that the (6,17)-semicross lattice tiles
Z .) T h e only primes ρ < 3584 such that 5 ( 6 ) coset splits (7(p) are
17
7,13,103,487,547,823,967,1063, and 3187.
A s we will see in a moment, 5(6) coset splits (7(p) for an infinity of

primes p.
Exercise 3 1 . For which primes ρ does 5 ( 2 ) coset split (7(p)? Ex-

press the answer in terms of the order of 2 modulo p.
We will show that if S(k) coset splits at least o n e group of prime

order, it splits an infinite n u m b e r of them [10]. In order to d o this
we need t o express the notion of coset splitting in terms of h o m o -
morphisms.
Assume that S(k) coset splits C(p), with splitting group H.

There is then a natural homomorphism from C(p)* onto the quo-
tient group C(p)*/H, which is isomorphic to C(k). So there is an
epimorphism
/ : C(py -+ C(k), (2)

that is one-to-one on the subset { 1 , 2 , . . . , k}. (In C(p)* the operation
is multiplication; in C(k) it is addition.)
Exercise 32. Prove that if there is a homomorphism from C(p)*

onto C(k) that is one-to-one on S(k), then S(k) coset splits C(p).
Let φ be the restriction to S(k) of the homomorphism (2). This

function has two properties:
(i) φ is one-to-one from 5(fc) onto C(k)
(it) If x, y, and xy are in S(k), then φ^) = φ(χ) + φ^).
Any function from S(k) to C(k) that satisfies (i) and (ii) we call
a logarithm on A logarithm on S(k) is determined by its values
on the primes in S(k).
For example the following table describes a logarithm on 5(7).
χ 1 2 3 4 5 6 7
φ{χ) 0 1 3 2 6 4 5
The assigment 0(2) = 1 determines φ(4) = φ(2) + φ(2) = 1 +
1 = 2. The choice 0(3) = 3 then forces 0(6) = φ(2 • 3) = 1 + 3 = 4.
Since 5 and 7 are primes in the range (7/2,7], the values, 0(5) and
0(7) may be assigned freely, as long as φ is one-to-one.
Exercise 3 3 . Construct logarithms on S(k) for fc = 13 and 19.
Exercise 34. Show that there is a logarithm on S(k) for fc < 24.
Recall that there is if fc + 1 or 2fc + 1 is prime.
Extensive computer calculations by Forcade and Pollington [1],

show that for fc < 194, S(k) has a logarithm, but 5(195) does not.
Exercise 3 5 . Prove that if there is a logarithm φ on S(k) in which

0(2) is relatively prime to fc, then there is a logarithm φ on S(k) such
that φ(2) = 1.
Since there is n o logarithm on 5(195) it follows that 5(195)

does not coset split any group of prime order. This raises a sequence
of questions.
Problem 9. D o e s 5(195) split some group of prime o r d e r ? (This

is equivalent to " D o e s the (195, n)-semicross, η > 3, ever lattice
tile Z ? " )
n
Problem 10. D o e s the (195,n)-semicross, η > 3, ever lattice tile

Ä " ? (This is equivalent to Problem 9.)
Problem 11. D o e s the (195, n)-semicross, η > 3, ever tile Rl

n
We have shown that if 5(fc) coset splits C(p) for some p, then
there is a logarithm on S(k). If k is odd, the converse holds. How-
ever, when k is even, the logarithm must satisfy extra conditions in
order for the converse to hold, as shown by Mills [8]. These condi-
tions are consequences of quadratic reciprocity.
Theorem 10. Let p i , . . . ,p be distinct primes and let bi,..., b e

r T
C(k). There is an infinite number of primes ρ and homomorphisms

φ : C(p)* —• C(k) such that φ(ρ ) = bi, for 1 < i < r if and only if
χ
one of these conditions holds:

(1) k is odd
(2) fc = 2m where m is odd and
(a) for each Pi = 1 (mod 4) that divides m, biis even, and
(b) for each pi = 3 (mod 4) that divides rn, the corresponding
bi's all have the same parity.
(3) fc = 4 m and for each pt that divides rn, bi is even.
Moreover, if there is one such prime ρ for which there is a homo-
morphism φ : C(p)* - • C(fc) with the prescribed values atpi,...,p , r
then there is an infinite number of such primes.
Exercise 36. Show that if 5(fc) coset splits some group of prime
order it splits an infinite number of groups of prime order.
Exercise 37. Using the quadratic reciprocity theorem, show that

conditions (2) and (3) are necessary when fc is even.
Exercise 38. Prove that the (32, n)-semicross tiles Z n

for an infi-
nite n u m b e r of n.
Exercise 39. Let 0 2 , 0 3 , . . . ,α*, b e distinct integers greater than

or equal to 2. Prove that { 1 , ± o , ± 0 3 , . . . , ±α&} splits n o abelian
2
group (finite or infinite). (Suggestion: I f s i , s , . . . is an alleged split-

2
ting set, consider the representation of - s i . )
A special case of Exercise 39 is the set {1, ± 2 , . . . , ±fc}, which

almost coincides with F(k). This shows that just a slight alteration
of a multiplier set can drastically change the groups it splits.
5. History and applications

T h e origins of the study of the cross and semicross are simple,
though they can b e traced back to several independent sources:
Ulrich in 1957 [18], Kärteszi in 1966 [7], Stein in 1967 [10], and
G o l o m b and Welch in 1968 [3].
Ulrich constructed single-error correcting codes for alphabets
of more than two symbols. H e utilized a packing of { - 1 , 1 } in
C(10), presented the equivalent of a splitting of C(5) Θ C ( 5 ) by
{ - 1 , 1 } . However, this p a p e r did not lead t o subsequent investiga-
tions of crosses or semicrosses in Euclidean space.
Kärteszi asked whether the (1,3)-cross tiles space. This was an-
swered by Freller in 1970; Korchmäros about the same time treated
η > 3. Molnär [9] in 1971 related the n u m b e r of Z-lattice tilings of
R by the ( l , n ) - c r o s s to the n u m b e r of abelian groups of order
n
2n + 1. Medyanik, apparently unaware of Molnär's work, showed

in 1977 that the (1, n)-cross tiles R . n
A r o u n d 1963 Stein posed the following problem. Consider the

standard lattice of unit squares that tile the plane. W h a t is the small-
est density of a set S of such squares with the property that every
square from the lattice has at least o n e edge on the b o r d e r of a
square in 5 ? T h a t the answer is 1/5 follows immediately from the
fact that the (1,2)-cross tiles the plane. (Each such cross must con-
tain at least o n e m e m b e r of S; hence, the density of S is at least
1 / 5 . O n the other hand, the set of the center squares of the crosses
in the tiling serves as a suitable family S.) This initiated his work in
(fc, n)-crosses and semicrosses, which first appeared in 1967.
G o l o m b and Welch showed that the (1, n)-cross tiles R . They

n
thought of the center of a cross as a code word and the other cubes
of the cross as words that might be received if there were an error
in one coordinate of the code word. A tiling then corresponds to a
perfect code.
In 1978, Szabo [14], stimulated by Molnär's work, considered
tilings by "lopsided" crosses, where at each facet of the central cube
either n o cube or one cube is attached. A r o u n d that time he r e a d
a Russian translation of a paper by Stein and became familiar with
the work of H a m a k e r and Stein [5]; in 1981 he proved that if 2n + 1
is not a prime, then there is a nonlattice Z-tiling by the ( 1 , n)-cross
and a Q-lattice tiling that is not a Z-tiling [14].
T h e semicross and cross have also shown that they deserve to
serve as archetypes of starbodies. For instance, Kasimatis in 1984
proved that any lattice covering of R? by translates of a (2,2)-cross
has density at least 9/7, which is less dense than the densest cover-
ing. This example, which is still in manuscript, is much simpler than
the earlier ad hoc example.
In 1985 Szabo [17] modified a (1,3)-cross by adding pyramids
at the ends of its six arms to obtain a starbody that tiles R but not by
3
a group of motions, thus providing a simple answer to Hubert's 18th

problem, "If congruent copies of a polyhedron Ρ tile Euclidean
space, is t h e r e a group of motions such that copies of Ρ u n d e r this
group of motions tile space?" (See Figure 7. T h e pyramids are lop-
sided t o prevent rotational symmetry.)
O t h e r examples have been constructed even in the plane, but
this is a particularly simple example. It also is an example of a set
that tiles by translates, but not by a lattice of translates. Stein in 1972
[11] had used a (4,10)-cross and (3,5)-semicross and finite fields to
provide the first examples of this p h e n o m e n o n .
In this chapter we answered some questions concerning the

tiling of π-space by the (k, n)-semicross and (k, n)-cross. However,
as is typical in mathematics, the few answers obtained raised many
questions. This suggests that our knowledge, measured by the num-
ber of answers, grows arithmetically, while o u r ignorance, measured
FIGURE 7
by the n u m b e r of questions, grows geometrically. So, strangely, we

are perhaps more ignorant at the e n d of the chapter than at the
beginning.
T h e tiling problems suggested questions about finite abelian
groups. Because of the rich structure of a splitting—each nonzero
element of a finite abelian group being represented exactly o n c e —
we could use counting arguments t o help settle some of the ques-
tions.
In the next chapter we will examine packing and covering by
the semicross and cross. T h e structures will not b e so rich, a n d the
problems will therefore b e m o r e difficult to approach.
References
1. R. W. Forcade and A. D. Pollington, What is special about 195?,
Groups, nth power maps and a problem of Graham, Number Theory,
Richard A. Mollin (Ed.), Walter de Gruyter, New York, 1990.
2. S. Galovich and S. Stein, Splittings of abelian groups by integers, Ae-
quationes Math. 22 (1981), 249-267.
3. S. Golomb and L. Welch, Perfect codes in the Lee metric, University
of Southern California, USCEE report 249 (1968), 1-24.
4. W. Hamaker, Factoring groups and tiling space, Aequationes Math. 9
(1973), 145-149.
5. W. Hamaker and S. Stein, Splitting groups by integers, Proc. Amer. Math.

Soc. 46 (1974), 322-324.
6. D. R. Hickerson, Splittings of finite abelian groups, Pacific J. Math. 107
(1983), 141-171.
7. F. Kärteszi, Szemläletes geometria, Gondolat, Budapest, 1966.
8. W. H. Mills, Characters with preassigned values, Canad. J. Math. 15
(1963), 169-171.
9. E. Molnär, Sui mosaici dello spazio di dimensione n, Atti delta
Academia Nazionale dei Lincei, Rend. Sc. Fis. Mat. e Nat. 51 (1971),
177-185.
10. S. K. Stein, Factoring by subsets. Pacific J. Math. 22 (1967), 523-541.
11. , A symmetric star body that tiles but not as a lattice, Proc. Amer.
Math. Soc. 36 (1972), 543-548.
12. , Tiling, packing and covering by clusters, Rocky Mountain J.
Math. 16(1986), 277-321.
13. , Lattice-tiling by certain star bodies, Studia Sei. Math. Hung. 20
(1985), 71-76.
14. S. Szabo, On mosaics consisting of multidimensional crosses, Acta Math.
Acad. Sei. Hung. 38 (1981), 191-203.
15. , Finite abelian groups and η-dimensional mosaics (Hungarian),
Mat. Lapok. 28 (1977/1980), 305-318 (MR 82i:20065).
16. , A bound of k for tiling by (fc, n)-crosses and semicrosses, Acta
Math. Acad. Set. Hung. 44 (1984), 97-99.
17. , A star polyhedron that tiles but not as a fundamental domain,
Colloquia Math. Soc. Jdnos Bolyai 48, Siofok, (1985).
18. W. Ulrich, Non-binary error correction codes, The Bell System Technical
Journal (1957), 1341-1388.
19. A. J. Woldar, A reduction theorem on purely singular splittings of cyclic
groups, Proc. Amer. Math. Soc., to appear.
Chapter 4
Packing and Covering

by the Semicross and Cross
W h e n a cluster C does not tile η-space, we immediately face two

questions: H o w densely can we pack translates of C ? H o w thinly
can we place translates of C such that they cover all of n-space?
Similar questions for the η-dimensional ball, going back t o Kepler,
have b e e n answered only for small values of n. We will get good es-
timates in all dimensions in the case of Z-lattice packings by semi-
crosses a n d even much m o r e information about packing by crosses.
Similar questions about Z-lattice coverings seem t o be far m o r e dif-
ficult, as we will see later in the chapter.
In o r d e r to quantify the efficiency of a packing or covering we
need the notion of the "density" of a family of translates of a cluster.
Let C b e a cluster in η-space consisting of q cubes and S =
{α!, a , . . . } a denumerable sequence of points in η-space without
2
an accumulation point. For each positive n u m b e r r let Q(r) b e t h e

cube { ( χ χ , ΐ 2 , · · • ,Χη) • \xi\ < r}, which has side 2r a n d volume
( 2 r ) " . Let N(r) b e the n u m b e r of points Oj in S such that a + C
{
lies in Q ( r ) . T h e fraction of Q(r) filled by these translates is
J K
' (2r)»
(In case of a packing / ( r ) < 1.) Assume that / ( r ) approaches a

limit as r —» oo and d e n o t e this limit d(S), which w e call the density
of the family of translates.
85
Exercise 1. Prove that the density of a covering is at least 1.
We may think of the density of a family of translates of a cluster

as the average n u m b e r of times that points in η-space are covered
by the translates.
T h e packing density of the cluster C, usually d e n o t e d 6(C), is
defined as the least upper b o u n d of the densities of packings by C.
T h e covering density of C, denoted 9(C), is defined as the greatest
lower b o u n d of the densities of coverings by C . If in these defini-
tions we restrict the families to b e lattices of translates of C, t h e n we
obtain the lattice packing density, 6L(C), and lattice covering density
0 (C).
L N o t e that
6 (C)
L < 6(C) < 1 < 0{C) < 6 {C).
L
If we restrict the translating vectors to be an integer lattice, then

we obtain the Z-latticepacking density i f (C) and Ζ-lattice covering
density θf(C).
Analogous densities have also b e e n defined when C is a con-
vex set and have been studied for over two centuries when C is a
ball. References for what has been discovered are [1, p p . 15-17], [6,
p.3], or [9, p p . 177-185]. For η = 1,2, and 3 the densest packing
of balls is achieved with a lattice packing; a proof for η = 3, d u e
to Hsiang, was obtained as recently as 1991, and its details are still
being checked.
Any denumerable packing of η-space by translates of a cluster
C can be shifted locally to produce another such packing in which
all the coordinates of the translating vectors are integers [3]. This
shift, which generalizes the idea of "rounding down," is defined as
follows.
Consider a denumerable family of translates of C that packs
η-space. E a c h cube in these clusters has the f j r m w + A, where
A = { ( x i , . . . , x ) : 0 < Xj < 1, 1 < i < n}.
n
Let W b e the set of w's.

Since W is denumerable, we may translate the family W by
some vector c so that n o vector c + w with w e W is o n the boundary
of z+A for any z e Z . Assume that this initial shift has b e e n made.
n
X2
XΟ
Xl
FIGURE 1
Then each w is in the interior of a unique cube of the form ζ + A

where ζ e Z . Call ζ + A the shift of w + A and ζ t h e shift of w. See
n
Figure 1.
N o t e that the set of shifts of the cubes in a cluster form a trans-
lation of that cluster. So the shift can be viewed as translating clus-
ters. Moreover, the density of the family obtained by the shift is the
same as the density of the original family.
Assuming that the family {w + A : w € W} is a packing, we
will show that the shift of this family is also a packing.
Let wi + A and w + A be distinct cubes in the packing, and
2
zi + A and z + A their respective shifts. If the interiors of z + A

2 x
and Z2 + A have a nonempty intersection, then z = z . T h u s w\ =

x 2
αϊ -I- z\ and w = a + z , where a\ and a are in the interior of A.

2 2 2 2
H e n c e υΐχ + a = w + αϊ, which violates the assumption that cubes

2 2
wi + A and w + A are part of a packing.

2
To show that the shift of a covering is again a covering, consider

a cube Z + A,ZG Z . We will show that t h e r e is a cube w + A in the
n
covering whose shift is ζ + A. Therefore, the densest packing by a

cluster has the same density as the densest integer packing.
T h e point ζ + ( 1 , 1 , . . . , 1) lies in the interior of w + A for some
w e W, as illustrated in Figure 2. Then it is not h a r d to show that
ζ + A is the shift of w + A.
3••2
ζ -K M )
ιυ
FIGURE 2
Exercise 2. Justify the last sentence.
T h e shift of a lattice is not necessarily a lattice. (Recall that the

cluster formed of two squares separated by o n e square lattice tiles
2-space but does not Z-lattice tile 2-space.)
T h e next exercise provides two m o r e illustrations of this fact.
Exercise 3.
(a) Let α > 1 be an irrational number. Translates of the interval
[0,1] by the n u m b e r s { n o : η e Z} form a lattice packing of
R . Show that the shift of this packing is not a lattice packing.
1
(b) Consider any lattice packing of R by cluster C where the de-

n
terminant of t h e lattice is irrational. Show that its shift is not a

lattice.
1. Packing by semicrosses
Let C be the (fc, n)-semicross, which has volume fcn + 1 . A packing
of R with translates of the (fc, n)-semicross by vectors in the lattice
n
Η of determinant d therefore has density

fcn+ 1
d '
Now restrict Η to being a sublattice of Z . T h e family of translates

n
of the (fc, n)-semicross by translates of the vectors of Η is a packing

if and only if t h e cosets {iej + Η : 1 < i < k, 1 < j < n } are
distinct and distinct from the coset H. Let G = Z /H. T h e n there
n
are η elements in G, s , s ,...,

x s , such that the fen elements isj,
2 n
1 < i < k, 1 < j < η are distinct and nonzero.
Exercise 4. Justify the last two statements.
Finding the densest Z-lattice packing by t h e (fe, n)-semicross

is equivalent to finding t h e smallest g r o u p G such that there are η
elements in G, s i , s , . . . , s , where the kn elements isj, 1 < i <
2 n
k, 1 < j < n, are distinct and distinct from 0. If such elements

«i, 32,..., s exist in G we will say that S(k) then η-packs G and
n
that { s i , s , 3 ) is a packing set.

2 n
Exercise 5. Show that for every pair of positive integers fe and η

there is a finite abelian group that S(fe) n-packs.
Let g(k, n) b e the o r d e r of t h e smallest abelian group G that

•S(fe) η-packs. Obviously, g(k, n) > kn +1 and equals kn +1 only if
S(k) splits an abelian group of o r d e r fen + 1 . Moreover, for fixed n ,
g(fc, n) is a nondecreasing function of fc. T h e density of t h e densest
Z-lattice packing of R" by the (fe, n)-semicross is
kn + 1
g(k,n)'
In dimensions 1 and 2, g(k, n) is easy to determine: g(k, 1) = fe + 1

and g{k, 2) = 2fe + 1. We will show that
= 1. (1)
T h e proof is based on two lemmas [7]. T h e first shows that g(k, 3) >
(fe + If' .
2
L e m m a 1. Let G be an abelian group of order m such that S(k)

3-packs G. Then (fc + l ) < m . 3 2
Proof. Let the packing set be {a, b, c}. Let G χ G b e t h e set of

ordered pairs (u, v), where u and υ are in G. Consider the (fc + l ) 3
elements (χα - yb, yb - zc), 0 < x, y, ζ < k, in t h e set G χ G.

If (fc + l ) > m , then by the pigeon-hole principle, two of these
3 2
elements coincide, say
xa — yb = xa — yb and yb — zc = yb — zc,
o < x,x,y,y,z,z < fc» (χ,ν,ζ) Φ (χ,ν,ζ)-

If χ = χ, then yo = yb. Since 0 < y, y < fc a n d 6 is an element in
the packing set {a, 6, c}, y = y. Similarly, z — z. But this contradicts
the fact that (x, y, z) and (x, y, z) a r e distinct. Similarly, y Φ y a n d
ζφζ.
Without loss of generality, we may assume that χ > x. T h e n
we have (x - x ) o = (y - y)b. Since α and b belong to a packing set,
y - y is negative. T h e equation (y - y)b = (z - z ) c then implies
that ζ - ζ is negative, hence ζ > ζ. We have
(χ - x ) a = (y - y)6 = (z - z)c,
hence (x - x ) a = (z - z)c, which contradicts the assumption that

{a, b, c] is a packing set. Thus (fc + l ) < τη , a n d t h e lemma is
3 2
proved. •
T h e next lemma puts a lower b o u n d o n g(k, 3) for an infinite

set of integers fc.
Lemma 2. Let a > 2 be an integer and let k = a - a. Then S{k) 2
3-packs the cyclic group C(a + 1) with packing set {1, - a , a } .

3 2
Proof. T h e proof that the congruence i · 1 =. j(-a) (mod a + 1), 3
has n o solutions for 1 < i, j < a - a is straightforward. T h e con-

2
gruence i · 1 = j(a ) (mod a + 1) reduces t o the previous o n e by

2 3
multiplying by - a . T h e congruence i(-a) = ja ( m o d a + 1) also

2 3
reduces to t h e first o n e by division by a (which is relatively prime to

the modulus, a + 1).
3
•
Incidentally, for a = 2 and 3 a n d fc = a - a, o(fc,3) equals

2
a + 1. It is not known whether this t r u e for all a.

3
Exercise 6. Fill in the details of the preceding proof.
L e m m a s 1 and 2 imply the following theorem.
Theorem 1.
l i m « = l
Outline of proof. By L e m m a 2, for k = a - a, g(k, 3) < a + 1 2 3
(1 + \ / l + 4fc) /8 + 1. Thus, for k of the form a - a,

3 2
(* )"'< *,3)<<i±4±!2 i .
+1 s( +
Therefore, for these values of k,
l b treat an arbitrary Λ, let ο b e the positive integer such that
a - a < k < (a + l ) - (a + 1).

2 2
Let ki = a - a and fc = (a + l ) - (o + 1). T h e n

2
2
2
fl(*i,3) < f l ( * , 3 ) <ff(fca,3).
N o t e that
lim £ = 1. (3)
Equations (2) and (3) imply that (1) holds when A; is not restricted
to b e of the form a - a. 2
•
Exercise 7. Fill in the details in the preceding proof.
Problem 1. If S(k) η-packs the finite abelian group G, must it n -

pack the cyclic group C(\G\)1 ( T h e o r e m 9 in the preceding chapter
shows that if the packing is a splitting of G, then the answer is "yes.")
Exercise 8.
(a) Show that for a positive odd integer b, S((b - l ) / 2 ) 4-packs 2
the cyclic group C{(b + l ) ( b + l ) / 2 ) with t h e splitting set

2
{1, - b , ( - b ) , ( - b ) } . (Hint: first show that the packing set is

2 3
a subgroup of C*((b + l)(b + l ) / 2 ) . 2
(b) D e d u c e that
limsup £!iM)< . 2
fc—»oo ™
Exercise 9.
(a) Show that for b s 1 (mod 6) and at least 7, S((b + b- 2
2)/3)
6-packs C((fc + b + 1)(6 + l ) / 3 ) with packing set
2
il,-b,(-b) ,(-bf,(-b)\(-b) }.
2 5
(b) D e d u c e that
lim sup y v
' ;
< 3.
k—*oo ™
In T h e o r e m 1 and Exercises 8 and 9 the m e t h o d rests o n the

fact that ι — 1, χ - 1, and χ - 1 have cubic factors with certain
3 4 6
properties. T h e n u m b e r k is expressed as a quadratic in b, a n d for

this k, S(k) is shown to pack a cyclic group whose o r d e r is a cubic
in b. (The packing set is generated by -b and forms a group u n d e r
multiplication.) In these cases
g (k,n) 2
fc—»00 k t
h m s u p 2
Λ
is shown to b e n o greater t h a n some specific rational n u m b e r . How-

ever, it is proved in [5] that for η > 3
limO^n) /ry =4cos2 !

fc->oo fc 3
\nJ
As Exercise 10 shows, for η > 3 this limit is rational only when η =

3,4, or 6. This means that the m e t h o d described has t o b e modified
to treat general n. T h e key modification to obtain a proof of (4) is to
consider packing sets of the form {1, -b, ( - f c ) , . . . , ( - f c ) " } that 2 - 1
d o not necessarily form a group. In other words, d o not d e m a n d

that ( - 6 ) " = 1.
Exercise 10. Let η be an integer and ω a primitive n t h root of

unity. Assume that [Q(u) : Q] = φ{η), where φ is the Euler "φ
function."
(a) Show that [Q(u) : Q(cos ( 2 π / η ) ] = 2 for η > 3.
(b) Show that £?(cos ( π / η ) ) = Q(cos ( 2 π / η ) ) for η > 1.
2
(c) Show that for η > 3, c o s ( π / η ) is rational if and only if η =

2
3,4, or 6.
Let us return to the case η = 3. Since
we can calculate how well the (fc, 3)-semicross Z-lattice packs R . 3
Letting 6% (k) b e the density of the densest such packing, we have

6f(k) = (3k + l)/g(k, 3). So we have
Um
6
JM = i. ( 5 )
Speaking loosely, we can say that when k is large, the densest Z -

lattice packing of the (k, 3)-semicross fills only about -j- of R . This
3
has b e e n generalized to arbitrary lattice packings in [7].

If we remove the condition that the packing be a lattice
packing, it turns out that the (k, 3)-semicross packs R much m o r e
3
densely. In fact, if 6(k) denotes the packing density of the (fc, 3)-
semicross, we have
lim*M=0.
6(k)
k-oc
(6)'
v
T h e combinatorial argument [4] rests on the construction of

certain designs, called "monotonic matrices," which we now d e -
scribe.
Consider a cube of width fc, which is composed of fc unit cubes.
3
Say that you place t(k) disjoint translates of the (fc, 3)-semicross in
such a way that each "corner c u b e " of a semicross coincides with
o n e of the fc unit cubes. T h e n these semicrosses lie in the cluster
3
formed of four copies of the cube of width fc, which is a magnifica-

tion of a (1,3)-semicross, as shown in Figure 3.
Of the 4fc unit cubes in this cluster, t(fc)(3fc + 1 ) are occupied
3
by the t(k) semicrosses. Since the (l,3)-semicross tiles Z , there 3
would be a packing of R with density

3
f(fc)(3fc + l )
4fc
3
Identify each of the fc unit cubes in the cube of width fc with

3
a triple (x, y, z), the coordinates of its center. Then x, y, and ζ are
integers, and 1 < x, y, ζ < k. Because the semicrosses are disjoint,
for a given (x, y) there is at most o n e triple (x, y, z). We can there-
fore record the configuration with the aid of a square array of fc 2
cells in which the cell corresponding to (x, y) holds the n u m b e r ζ if

the semicross with corner (x, y, z) is present. Otherwise cell (x, y)
is empty. This leads us to consider certain square arrays, which we
call "monotonic matrices."
A monotonic matrix of o r d e r k is a n array of fc cells in s o m e of
2
which an integer in the set { 1 , 2 , . . . , fc} is placed, subject t o three

conditions:
(a) For two filled-in cells in a row, the one further right has a larger
entry;
(b) For two filled-in cells in a column, the higher o n e has the larger
entry;
(c) For two filled-in cells with t h e same entry, the o n e further right
is higher (the "positive slope" condition).
Exercise 11. Show that conditions (a), (b), and (c) are equivalent
t o the fact that the corresponding semicrosses are disjoint.
Let m(k) b e the maximum n u m b e r of occupied cells in all
monotonic matrices of o r d e r k. N o t e that the (k, 3)-semicross tiles
Z with density at least m(k)(3k + l)/4ife .
3 3
Table 1 lists some of t h e values of m(k); n u m b e r s in parenthe-

ses are lower bounds. For instance, that m ( 3 ) > 5 is shown by the
array in Figure 4.
K. Joy programmed an exhaustive search for the largest m o n o -
tonic matrix of order 5, determining that m(5) = 11. Figure 5 dis-
plays o n e of the many arrays h e found.
Exercise 12. Show that m(2) = 2, m(3) = 5, a n d m ( 4 ) > 8.
Exercise 13. Show that if Λ is a square, then m(k) > k f . (Table
3 2
1 shows that m ( 4 ) = 4 / but m ( 9 ) > 9 / . )

3 2 3 2
Exercise 14. Show that m(k • I) > m(k)m(l)

(a) by using the geometry of semicrosses;
(b) by using only the definition of a monotonic matrix.
k 1 2 3 4 5 6 7 8 9 10
m(k) 1 2 5 8 11 (14) (19) (22) (28) (32)
TABLE 1
2 3
1
1 3
FIGURE 4
FIGURE 5
By Table 1, m ( 7 ) > 19 « 7 . Therefore, by Exercise 14,

1 5 1 3
m(7 ) > (m(7)) > ( 7 )

r r
, for any positive integer r . T h u s for
r 1 5 1 3
fc = 7 we have m(k) > fc - and there are packings by the (fc, 3)-
r 1 513
semicross with density at least

fc (3fc
1513
+ l)
4fc 3 ( , )
Comparing (7) with (5) establishes (6). T h e key is that 1.513

is greater than 1.5.
Exercise I S . Show that m(fc + 1) > m(fc).
In a m o m e n t we will apply the next exercise t o the function

m(fc).
Exercise 16. Let f(n) be a positive real n u m b e r for each positive

integer n. Assume that f(n + 1) > f(n), f(n) < n , and f(mn) 2
>
/ ( m ) / ( n ) , for any positive integers m , n. Define e by the equation
n
/ ( n ) = n " . Show that

e
lim e„
exists, and is at most 2. (Hint: It is the least u p p e r b o u n d of { e } . ) n
By Exercise 16, if we define e by setting m(fc) = k ,

k
ek
we know
that
lim ek
fc—»oo
exists. Call it L. Since e > 1.513, we have 1.513 < L < 2.

7
Problem 2. Find L.
Problem 3. Examine m{k)/k as k —> oo. If L < 2, then
2
m(fc)/fc -» 0. (Does the limit exist?)

2
Hickerson pointed out in conversation that if we knew that

the density of the densest packing of R by the (k, 3)-semicross ap-
3
proaches 0 as k —• oo, we would immediately have the analogous

result for all dimensions η > 3. To see this, assume that for some
dimension η > 3 there are arbitrarily large integers fc such that the
(fc, n)-semicross packs R with density greater than ε, a fixed pos-
71
itive number. Using the shift, we may assume that the translating
vectors lie in Z . n
T h e n u m b e r of corner cubes of the semicrosses in a packing

in some large cube of side s (which we take to b e an integer much
larger than fc) would then be greater than es /(nk + 1). Since the
n
large cube could b e viewed as the union of s " ( η — 1)-dimensional"

slabs of unit thickness, in some (η - 1)-dimensional cross sectional
slab of this large cube there would be at least e s / ( n f c + l ) corner
n - 1
cubes.
Now, the cluster similar to the (1, η — l)-semicross, but com-
posed of η cubes of side a tiles R ' . Of its n s
11 1
unit cubes at
n _ 1
least e s ( ( n - l)fc + l)/(nfc-|-1) are occupied by the ( l , n - 1)-

n _ 1
semicrosses whose corners lie in the corner of the large semicross.

Therefore there is a packing of i ? by the (1, η - l)-semicross of
n _ 1
density at least
ea^dn- l)fc + 1)
na - (nk+
n 1
1),
which approaches e(n - l ) / n as fc —• oo. Repeating this descend-
2
ing argument until we reach η = 3 produces a contradiction of

o u r assumption that the packing density of the (fc, 3)-semicross ap-
proaches 0 as fc —> co.
2. Packing by crosses
T h e case of Z-lattice packings of R by the (fc, n)-cross is quite dif-
n
ferent. For η > 2, let h(k, n) b e the order of a smallest abelian

group that F(k) η-packs. It is known [7] that
O n e difference between (4) and (8) is that the dimension η does not
influence (8) but does influence (4). T h e other is that the exponent
of A; is 2 instead of 3/2, which implies that the densest Z-lattice
packing of R by (fc, n)-crosses is much less dense than is the case
n
for semicrosses. If 6*(k) is that density,
6*(k)
fc-oo (2fcn + l)/fc 2
'
hence
lim
fc—oo
—V = 2n/k
1. (9)
A n o t h e r difference between the cross a n d semicross is that the

proof of (8) uses packing sets that form arithmetic progressions. We
will not present this proof, but will content ourselves by showing
that for η > 2, h(k, n) > (fc + l ) . 2
Assume that F(k) η-packs the finite abelian group G. T h e n

F(k) 2-packs G with packing set {a, b}. Assume that \G\ < ( f c + 1 ) . 2
Consider the (k + 1 ) elements ia + jb, 0 < i, j < k. By the pigeon-

2
hole principle, there are distinct pairs and ( i , j ) , 0 < <

k such that ia + jb = ia + jb, or equivalently (i - i)a = ( J - j)b.
N o t e that \i-i\<k and \j - j \ < k. If i = i, we have (J - j)b =
0, contradicting the fact that b is part of a set that packs G. Thus
i — ι φ 0 and, similarly, j - j Φ 0. This violates the assumption
that {a, b] is a packing set. Consequently h(k, n) > (k + l ) for all 2
η > 2.
Exercise 18.
(a) Show that for even fc, F(fc) 2-packs C ( f c + 2 f c + 2 ) with packing
2
set{l,fc + l}.
(b) Show geometrically that h(k, 2) = fc + 2fc -I- 2. 2
Exercise 19. Show that F(fc) 3-packs C(fc -1- 3fc + 3) with packing2
set {l,fc + l,fc + 2 } .

Exercise 20. Show that for even fc, F(k) 4-packs C(k 2
+ 4k + 5)
with packing set {1,fc+ 1, fc + 2, k + 3 } .
Exercise 21. Show that for fc = 0 or 4(mod 6), F(k) 5-packs

C(fc + 8fc + 9) with packing set {1,fc+ 1, fc + 3, fc + 5, fc + 7}.
2
Exercise 22. Show that for fc Ξ 0 (mod 6), F(fc) 6-packs C(fc 2
+
30fc+233) with packing set {1,fc+ 1 1 ,fc+ 1 3 ,fc+ 1 5 , fc-f-17, fc + 1 9 } .
Problem 4. Determine n(fc, n ) . Trivially, h(k, 1) = 2fc + 1.
Unlike the semicross case, the ratio between the density of the
densest Z-lattice packing and densest packing does not approach 0
as fc approaches infinity. Instead, it approaches 1. T h e argument is
not long.
First of all, for any packing of by the (fc, n)-cross, the shift
produces a packing of the same density. So we consider only inte-
ger packings. In a square tray of side length fc + 1 there can b e at
most o n e center of a cross in a packing. Since these trays tile R , n
t h e density of t h e packing is at most (2fcn + l)/(fc + l ) , which is

2
asymptotic to 2n/fc as fc —• oo. Comparing this with (8) shows that

Z-lattice packings by the (fc, n)-cross, when fc is large, can be just
about as dense as an arbitrary packing by the (fc, n)-cross.
Problem 5. Find the density of the densest packing of R n

by t h e
(fc, n)-cross.
Problem 6. Is the density of the densest packing of R n

by a cluster
always a rational number?
3. Covering by the semicross and cross

If a semicross or a cross does not tile R , n
we may also ask, " H o w
thinly can it cover R V n
We consider only covers of Z , and view a semicross or cross

n
as subset of Z , and deal only with translates by integer vectors.

n
Such a family of translates of a semicross or cross is said to

cover Z " if every point in Z is in at least o n e of t h e translates. T h e
n
density of the covering can b e viewed as "the average n u m b e r of

times each point is Z is covered."
n
For instance, t h e (2,2)-cross covers R with density 9 / 8 , since

2
two overlapping copies of this cross fill u p the cluster of 16 squares

shown in Figure 6, and this latter cluster tiles the plane [3]. H e n c e
the (2,2)-cross covers the plane with density 18/16 = 9 / 8 .
Exercise 23. Verify that the cluster in Figure 6 tiles the plane.
Exercise 24. Use a similar approach to get a covering of t h e plane

by the (3,2)-cross of density 13/11.
T h e construction of Z-lattice coverings of R by the (k, n ) -

n
semicross, just like Z-lattice tilings or packings, d e p e n d s on finite

groups. We say that S(k) η-covers a finite abelian group G with the
covering set {si, s , . . . , s } if every element in G \ {0} can b e ex-
2 n
pressed in t h e form isj, 1 < i < k, 1 < j < n. Let f(k,n) be

the order of the largest abelian group that S(k) η-covers. Clearly
/(Jb, n) < kn+1. Just as we can obtain a lattice packing of Z from n
a packing of a finite group, we can obtain a lattice covering of Z "

from a covering of a finite group.
Exercise 25. Show that the density of the least dense Z-lattice cov-
ering of Z " by the (k, n)-semicross is (kn + l)/f(k, n).
Determining the behavior of f(k, n) for large k is much h a r d e r

than determining the behavior of g(k, n), the corresponding func-
tion for packings. In the case of a packing, if g\ and g are two el-
2
ements of a packing set w e can say that the equation igi = jg , 2
1 < i,j < k has n o solution. However, for coverings, t h e struc-

ture is looser; there is nothing of use that we can say about a pair
of elements in a covering set. Even determining f(k, 3) is far from
trivial.
T h e next theorem provides very efficient coverings by semi-
crosses in (p + 1)-space, where ρ is a prime. It, as well as the other
results o n covering that we will describe, are d u e t o Dad-del [2].
Theorem 2. Let ρ be a prime and r be a positive integer. Let k =

r p - 1, m = r p , and η = ρ + 1. Then S(k) η-covers C(m) with the
2
covering set {p} U {1 + i r p : 0 < ί < ρ — 1}.
Proof. Consider an integer χ € [1, m - 1]. If ρ divides χ then χ is

of the form ip, w h e r e i e S(k).
Let A = {x : 1 < χ < m , ( x , p ) = 1}. We will show that A is
covered. For convenience, if g € C ( m ) , let g d e n o t e S(k)g, "the set
of elements covered by g."
For each i, 0 < i < ρ - 1,
\A Π 1 + irp\ = k — [k/p] = r ( p - 1).
Since \A\ = r p ( p - 1 ) , all that remains is to show that ifiî', then

1 -(- irp Π 1 + i'rp = 0.
Assume that there are integers j , f, 1 < j , f < k such that
j(l + irp) =. + i'rp) (mod rp ).

2
T h e n j = j ' (mod rp), hence j = f. Consequently
jirp = ji'rp (mod rp )

2
and we have
j(i — i') = 0 (mod p ) .

Since we assume that j (1+irp) is not divisible by p, (j, p) = 1. T h u s

i - i' Ξ 0 (mod p), from which it follows that i = ί'. This completes
the proof. •
It can b e shown that the covering set in the preceding t h e o r e m

is essentially unique.
In any case, the covering is quite economical. T h e τη — 1 =
rp - 1 nonzero elements of C(m) are covered by kn products, and
2
kn = (rp — l ) ( p + 1 ) = rp + (r - 1 )p -1, which is not much larger

2
(in ratio) than rp - 1. Holding η fixed, we have, since k = rp - 1

2
,. /(*,«) ^.. rp2
lim sup — - — > lim sup = p.

k—»oo k ,—>oo rp — 1
T h e proof of T h e o r e m 2 uses only cyclic groups. This suggests
the following problem.
Problem 7. If S(k) η-covers a finite abelian group G, does it τι-
cover C(|<?|)? (For fc = 2 the answer is "yes" [8].)
T h e o r e m 2 implies that when the dimension η is o n e m o r e than

a prime and fc is large there are Z-lattice coverings of R by the n
(fc, n)-semicross with density near n/(n - 1), which for large η is
n e a r 1. This is quite a contrast with the fact that Z-lattice packings
by these semicrosses have densities near 0.
T h e construction of Z-lattice coverings by the (fc, n)-cross re-
duces to the following problem on unite abelian groups. We say
F(k) = { ± 1 , ± 2 , . . . , ±fc} η-covers a finite abelian group G with
covering set {si, s ,...,
2 ns } if every element of G \ {0} can be ex-
pressed in the form isj, 1 < |i| < fc, 1 < j < n. Let c(fc, n ) b e
the order of the largest abelian group that F(k) η-covers. Clearly
c(fc, n) < 2fcn+1. A n argument similar to that for T h e o r e m 2 yields
the following theorem.
Theorem 3 . Let r be an even positive integer, q prime, k = rq/2,

η = q + 1, and m = rq . Then F(k) η-covers C(m) with covering set
2
{q} U {1 + irq : 0 < i < q - 1}.
Exercise 26. Prove T h e o r e m 3.

Returning t o the covering problem suggested by semicrosses,

we can easily check that /(fc, 1) = k + 1 and /(fc, 2) = 2fc + 1. We
might expect /(fc, 3) to b e approximately 3fc. If so, we are in for a
surprise. Very little is gained by adjoining another element to t h e
2-covering set. In fact /(fc, 3) = 2fc + 2.
Exercise 27. Show that S(fc) 3-covers C(2fc + 2) with covering set
{l,-l,fc + l}.
We thus have
^ ! ψ = 2 , uffiü.i
fc—>oo fc fc—>oo Κ fc—»oo fc
For η > 4, Η π ι ^ - κ » /(fc, n)/fc has not been determined. It has not
even b e e n shown to exist. As noted, when η = ρ + 1 for a prime p ,
,. /(Μ) .
lim s u p — - — > p .
fc—too
T h e o r e m 3, though it concerns t h e cross, also provides infor-
mation about t h e semicross, since when F(fc) η-covers a group G
with t h e covering set { s i , s , s } , S(k) 2n-covers G with t h e
2 n
covering set { ± s i , ± s , . . . , ± s } .
2 n
For instance, consider T h e o r e m 3 in the case q = 7 (and 2n is

therefore 16). According t o the theorem, when r is even, 5 ( 7 r / 2 )
16-covers C ( 4 9 r ) . Thus
l i m s u p /(M i> 4. 6
1
fc—»oo *
T h e o r e m 2 gives a weaker result: T h e case ρ = 13 gives us
limsup /(M i>13. 6
fc—»oo
Incidentally, η = 16 is t h e smallest dimension for which T h e o r e m

3 gives a stronger result for the function / than does T h e o r e m 2.
Exercise 28. F o r which values of η < 50 does T h e o r e m 3 give a
stronger result than does T h e o r e m 2?
O n t h e basis of extensive computations in low dimensions
as well as T h e o r e m s 2 a n d 3, it is tempting to conjecture that,
for η > 3,
k—*oo fc
is either the largest prime less than η or twice the largest prime less
than the greatest integer less than n / 2 , whichever is larger. How-
ever, it should be kept in mind that we do not even know that the
limit exists.
Problem 8. D o e s limk-,,» ^ ' " ^ exist? If so, what is its value?
Problem 9. Settle for the cross the analog of Problem 8.
Exercise 29. Show that S(k) 4-covers C(m) with covering set
{±1,±2}:
(a) if A; is even and m = 3fc + 1;
(b) if k is o d d and m = 3fc + 2.
Exercise 30. Show that for fe φ 2, S(fe) 5-covers C ( m ) with the
covering set given as follows (in the case fc = 61 + 1, χ is arbitrary):
fc m covering set
6Z + 0 3fc + 3 { ± l , ± 2 , f c + 1}
6i + l 3fc + 2 {±l,±2,x}
61 + 2 3fc + 5 {1,-2, ±3, -6}
6Z + 3 3fc + 4 { 1 , - 2 , - 3 , ±6}
6/+ 4 3fc + 5 { 1 , - 2 , - 3 , ±6}
6/+ 5 3fc + 4 { 1 , - 2 , - 3 , ±6}
If the conjecture m a d e after Exercise 28 is correct, we would

have the following table:
η 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
lim /<*·»> 1 2 2 3 3 5 5 7 7 7 7 11 11 13 13 14 14 17
fc-.oo fc
It would imply that while there are very efficient Z-lattice cov-
erings of R by the (fc, 12)-semicross for large fc, the situation in
n
R n
is quite different, where the most efficient Z-lattice coverings
have density approximately 11/7 « 1.57. For large η this density
approaches 1, which implies that R is t h e space that is " h a r d e s t "
n
t o Z-lattice cover by semicrosses.
U p to this point in the book we have considered only tilings,

packings, and coverings by clusters. Even these simple objects, com-
posed of a finite n u m b e r of cubes, even of a single cube, raise a host
of geometric, algebraic, and combinatorial problems, most of which
are unsolved. In the next two chapters we investigate tilings by tri-
angles. Chapter 5 is concerned with tiling a polygon with triangles
all of which have the same area. O n the other hand, Chapter 6 ex-
amines tilings by triangles where there is some restriction on their
shapes, b u t n o n e on their areas. For instance, it will show that a
square cannot be tiled by 30° - 60° - 90° triangles. As is to be ex-
pected, the algebraic techniques used in the coming two chapters
are quite different, l b treat triangles of equal areas we use valua-
tion theory, but t o deal with triangles of prescribed shapes we will
use isomorphisms of subfields of the complex numbers.
References
1. J. H. Conway and N. J. A. Sloane, Sphere packings, lattices, and groups,
Springer-Verlag, New York, 1988.
2. A. Dad-del, Covering abelian groups with cyclic subsets, Aequationes
Math., 1994 (to appear).
3. H. Everett and D. Hickerson, Packing and covering by translates of
certain nonconvex bodies, Proc. Amer. Math. Soc. 75 (1979), 87-91.
4. W. Hamaker and S. K. Stein, Combinatorial packing of R by certain
3
error spheres, IEEE Trans. Inf. Theory. IT-30 (1984), 364-368.

5. D. R. Hickerson and S. K. Stein, Abelian groups and packing by semi-
crosses, Pacific J. Math. 122 (1986), 95-109.
6. C. A. Rogers, Packing and Covering, Cambridge University Press,
Cambridge, 1964.
7. S. Κ. Stein, Packing of R by certain error spheres, IEEE Trans. Inf.

n
Theory, IT-30 (1984), 356-363.

8. S. Szabo, Lattice covering by semi-crosses of arm length two, Euro-
pean J. of Comb. 12 (1991), 263-266.
9. T. Thompson, From error correcting codes through sphere packings to
simple groups, Mathematical Association of America, Washington, D.
C , 1983.
Chapter 5
Tiling by Triangles of Equal Areas
In the first four chapters we examined tilings of Euclidean space by

translates of sets that are unions of cubes. In this chapter we con-
sider tilings of a much different character, namely tilings of poly-
gons by triangles of equal areas. (We d o not d e m a n d that the tri-
angles b e congruent.) We restrict our attention to simplicial dissec-
tions, though the theorems hold for all dissections.
In 1965 Fred Richman at the University of New Mexico at Las
Cruces wanted to include a geometric question in a master's degree
examination h e was preparing. "I noticed that it was easy to cut a
square into an even n u m b e r of triangles of equal areas, but I could
not see how to cut it into an odd number." Not solving his p r o b -
lem, h e chose not to put it on the examination. However, he did
show that the square could not b e cut into three or five triangles of
equal areas and mentioned the problem to his colleague and bridge
partner, J o h n Thomas. T h o m a s recalls that
Everyone to whom the problem was put (myself included)

said something like 'that is not my area but the question
surely must have been considered and the answer is p r o b -
ably well known.' Some thought they had seen it, but could
not r e m e m b e r where. I was interested because it reminded
m e of Sperner's L e m m a in topology, which has a clever
odd-even proof. I thought about it for several months, and
got t h e results in the 1968 paper by trying t o construct a
107
dissection into an odd number of triangles of equal areas.

They convinced m e that there is no such dissection, b u t I
could not prove it.
W h e n I sent the p a p e r to Mathematics Magazine, the
referee's reaction was predictable. H e thought the p r o b -
lem might be fairly easy (although h e could not solve it)
and was possibly well-known (although h e could find n o
reference to it).
H e r e c o m m e n d e d that I submit the problem to the
Monthly and, if no o n e came u p with a solution, that the
paper should b e published. This was d o n e , and the p a p e r
appeared three years after I wrote it.
We can introduce a coordinate system such that the four ver-

tices of the square are (0,0), (1,0), (1,1), and (0,1). T h o m a s [12]
proved that this square cannot be cut into an odd n u m b e r of trian-
gles of equal areas if all the vertices of the triangles have rational
coordinates with o d d denominators.
Paul Monsky [8] settled the question in 1970 by removing the
assumption on the coordinates of the vertices.
But even if Richman had not posed his question, t h e prob-
lem of tiling a polygon by triangles of equal areas would have been
raised. In 1984 Victor Klee was writing a paper o n convex poly-
topes. "In order to simplify a computation it was convenient to be
able to dissect an arbitrary convex polytope into simplices of equal
volumes," h e recalled in a letter written in 1990:
It seemed positively obvious to m e that this could b e d o n e ,

so I initially assumed it and proceeded to work out the
other details of the proof. Later, when I could not justify
the assumption, I had t o circumvent it, and found a m o r e
complicated argument that did not use the assumption.
However, I still felt strongly that a dissection into sim-
plices of equal volume should always b e possible. I was
familiar with Monsky's theorem. However, for my argu-
ment I did not have to restrict the parity of the n u m b e r of
simplices.
In the summer of 1984 at a conference at the Univer-

sity of Oregon I posed to a number of distinguished mathe-
maticians the problem of deciding (for example) whether an
arbitrary convex polygon could be dissected into triangles
of equal area. Several of them were attracted to the prob-
lem, and felt that they were On the verge' of the proof that
such a dissection was always possible. As far as I remember,
none of us seriously suggested that it might not always be
possible.
Klee managed to prove his polytope theorem, bypassing his dissection
question [6J.
In this chapter, we will construct quadrilaterals, even trapezoids,
that cannot be cut into triangles of equal areas. Actually, Hales and
Strauss in 1982 [2], two years before Klee raised his question, had
given a nonconstructive proof that such quadrilaterals exist. So there
are, all told, three separate and independent occasions when tiling by
triangles of equal areas was considered.
We first develop the algebraic and topological machinery used in
Monsky's proof, prove the Richman-Thomas-Monsky theorem and
then discuss a variety of related questions and results inspired by that
theorem.
1. Algebraic and topological preliminaries

We will make use of one theorem from topology, Sperner's Lemma,
which was proved by Sperner in 1928. Sperner applied his lemma
to study the dimensions of spaces. Soon after it was used to prove
Brouwer's fixed point theorem (any continuous mapping of an n-
dimensional ball into itself has a fixed point).
We will be using Sperner's Lemma in dimension two. It con-
cerns a simplicial dissection of a plane polygon into triangles. The
adjective "simplicial" means that two overlapping (closed) triangles
intersect either in a common vertex or in two vertices and the en-
tire edge that joins them. The dissection in Figure 1 is simplicial;
the one in Figure 2 is not. For convenience, all the dissections in
this chapter will be simplicial. However, Sperner's Lemma and the
theorems in this chapter can be generalized to non-simplicial dis-

sections.
Sperner's L e m m a involves the notion of "completeness." A
polygon whose vertices are labelled A, B, or C is complete if the
n u m b e r of edges whose vertices are labelled A and Β is odd. In
particular a triangle is complete if and only if its t h r e e vertices are
labelled A, B, and C. A n edge whose vertices are labelled A and Β
is also called complete. T h e substance of the following exercise will
b e n e e d e d later in the chapter.
Exercise 1. Introduce η dots on a complete edge Ε and label each

dot A or B. T h e η dots divide Ε into η + 1 sections.
(a) Show that the n u m b e r of sections that have vertices labelled A
and Β is odd.
(b) Show that if, o n the other hand, the vertices of Ε are both la-
belled A or both labelled B, then the n u m b e r of sections with
vertices labelled A and Β is even.
L e m m a 1. (Sperner) Consider a simplicial dissection of a plane

polygon. Assume that each vertex is labelled A, B, or C. Then the
number of complete triangles is congruent modulo 2 to the number
of complete sections on the polygon.
Proof. Consider any triangle in the dissection. Place a " p e b b l e "

in it next to any edge labelled AB. T h e procedure is illustrated in
Figure 3.
c Β c
A Β A Β A C
FIGURE 3
A complete triangle collects one pebble. All other triangles

collect either two or none. Therefore the total n u m b e r of pebbles
is congruent modulo 2 to the n u m b e r of complete triangles.
Next count the total n u m b e r of pebbles in terms of the edges
labelled AB. Next to each interior edge AB there are two pebbles.
Next to an AB edge on the boundary there is o n e pebble. H e n c e
the total n u m b e r of pebbles is congruent modulo 2 to the num-
ber of complete edges on the boundary. This establishes Sperner's
Lemma. •
Sperner's L e m m a tells us that if the polygon is complete then

there is at least o n e complete triangle in the dissection. For most
of our applications this is all we will need from Sperner's L e m m a .
Later we will state a stronger version, in which orientation plays a
role.
O u r other tool, valuation, is algebraic [1, pp. 1-15,13, pp. 197—
202]. A valuation on a field F is a function φ from F to the real
numbers augmented by oo that has the following properties:
(1) 0(xty) = φ(χ) + 0(y)
(2) 0(x + y) > m i n { 0 ( x ) , 0(j/)}
(3) 0(x) = oo if and only if χ = 0.
N o t e that if φ is a valuation on F, then for any positive n u m -
b e r k, kφ is also a valuation. T h e valuations φ and Ηφ are called
equivalent.
Since 0(1) = 0(1 · 1) = 0(1) + 0(1), it follows that 0(1) = 0.
Similarly, 0(x) = 0 for each root of unity and in particular 0 ( - l ) =
O . T h u s 0 ( - a ; ) = φ(-1·χ) = φ(-1)+φ(χ) = φ(χ). H e n c e φ(-χ) =

φ(χ). Also, when χ is not 0, φ{1/χ) = —φ(χ). Moreover, if φ(χ) <
φ(υ), then a stronger condition than (2) holds, namely φ(χ + y) =
φ(χ). To show this, first note that, by property (2), φ(χ) < φ(χ + y).
But we also have
φ(χ) = φ(χ + y-y)
> πύη{φ(χ + ν),φ{-υ)}
= τα\η{φ(χ + ν),φ{ν)}
> πιίη{φ(χ),φύ)}
= Φ(χ).
Thus φ(χ) - χώχί{φ(χ + j/),</>(j/)}. Since φ(υ) > φ(χ), we

conclude that φ(χ) — φ(χ + y), as claimed.
Exercise 2. Show that if φ is a valuation on a field F and χ and y,

yÔ, are in F, then φ(χ/ν) = φ(χ) - φ(ν).
To construct a valuation on Q, the field of rationale, select a

prime p.
Define φ as follows. First, define φ (0) to b e oo. If r e Q,
ρ ρ
r ^ 0, write r in the form p a/b, where η , a, and 6 are integers,

n
and ρ divides neither a nor o. Define φ {τ) to be n , which we can

ρ
think of as "the n u m b e r of times that ρ divides r." T h e function φ ρ
is called the p-adic valuation on Q.
Exercise 3. Prove that φ is well defined, that is, if p c / d =

ρ p a/b,m n
where m, n, a, b, c, and ci are integers, and ρ is relatively prime to

abed, then m = n.
Exercise 4. Prove that φ is a valuation on Q.

ρ
T h e next exercise shows that every valuation on Q is of the

form Ηφ for some constant fc and prime p .
ρ
Exercise 5. Let φ be a valuation on Q other than the trivial one

that assigns 0 to each number.
(a) Show that for every integer η, φ(η) > 0.
(b) Show that there is a prime ρ such that φ(ρ) > 0.
(c) Let p ' be a prime different from p. Show that φ(ρ') = 0. Sug-
gestion: T h e r e are integers α and b such that ap + bp' = 1.
(d) Letting φ(ρ) = fc, show that φ = kφ . v
T h e valuation φ is the one we need in the solution of Rich-

2
man's problem. If we restrict the vertices of the triangles in the

dissection to have only rational coordinates, that valuation would
suffice. To deal with arbitrary dissections of the square, we need to
extend φ from Q to a field that includes all the coordinates of the
2
vertices. Fortunately, each p-adic valuation can b e extended to all

of R, the field of the real numbers (and even to all of C, t h e field of
complex numbers). We pause to sketch how the domain of φ can ρ
b e extended beyond Q. T h e reader may skip over this interlude, for

the essential point is that each valuation on Q can b e extended in
at least o n e way to any subfield of the reals.
We first sketch briefly how a valuation φ on Q can b e extended ρ
to a finite algebraic extension F. Let A b e the ring of algebraic in-

tegers that lie in F. (A n u m b e r is an algebraic integer if it is a root
of a monic polynomial with integer coefficients.) T h e algebraic in-
tegers form a ring. Each ideal in A (other than 0 and A) is uniquely
expressible as a product of prime ideals. In particular, the ideal (p)
generated by the prime ρ has a factorization of t h e form
( p ) = pe(l)pe(2)...pe(r) )
where the e(i)'s are positive integers and the Pi's are prime ideals in
A. Incidentally, r is not larger than the dimension of F over Q. Pick
o n e of the r prime ideals, say P i . For χ in A, but n o t 0, define φ(χ)
t o be l / e ( l ) times the number of times P i appears in the represen-
tation of the ideal (x) as a product of prime ideals. Define 0(0) t o
b e oo. N o t e that φ(ρ) = e ( l ) / e ( l ) = 1 and therefore φ extends φ ρ
from Q to A. These r extensions are, u p to equivalence, the only

extensions of φ to A that have the properties of a valuation.
ρ
T h e extension of φ from A to F depends on the fact that every

element in F can b e expressed as a quotient of elements in A. Tb
see this, let 6 in Ρ be a root of
ax
n
n
+ α _ιΐ η
η 1
-1 h a = 0,0
where the coefficients are in Ζ and a is not 0. T h e n a b is a root n n
of
/ X \ n
f X \ ~ι
n
a (—
n +a„_i(— +... + a = 0
0
Va / n Va / n
or
a x n
n
+ α _ια χ η η
η _ 1
-\ \- α α " — 0.
0
Cancelling a „ yields a monic polynomial in Z[x\ of which t h e num-

b e r a b is a root. T h u s a b is an algebraic integer, and b —
n n (a b)/a n n
is a quotient of elements in A. Define 0(6) to be 0 ( a 6 ) - 0 ( a ) . n n
Since 0(xy) is equal to φ(χ) + φ(ν), this extension is single valued.

In particular, if F is a quadratic extension of Q, F = Q(s/D),
there are at most two extensions of φ to Q(\/D). For the n u m b e r ρ
of extensions see [3, p . 190]. For instance, for the o d d prime ρ there
are two extensions of φ to Q(\/D) if and only if ρ does not divide
ρ
D a n d D is a quadratic residue modulo p. For example, there are

two extensions of φ to Q{\/2) but only o n e to Q ( \ / 3 ) .
7
In the next few exercises assume that the valuations are defined
on all of R.
Exercise 6. Evaluate 0 at 7,8, 3/4, and 6. 2
Exercise 7. Evaluate 0 at v/2, v ^ . 3 / \ / 2 and 2 y/i.
Exercise 8. Evaluate φ at 1 + \ / 2 , \ / 2 + \ / 5 , and 2 + ν / ϊ ^ .

2
Exercise 9. Find 0 ( 1 + as follows. L e t u = l + \ / 3 a n d t ; =

2
1 - v/3-
(a) Find 0 (m>) and 0 (u + υ).
2 2
(b) Show that 0 (u) must equal φ (ν).

2 2
(c) Obtain 0 ( u ) . 2
Exercise 10. T h e r e are two possible values for 0 ( 1 + \/Ϊ7). 2 Find

them.
It is a much simpler m a t t e r to extend a valuation φ from a field

F t o the field F(x), where χ is transcendental over F. In fact, we
may preassign the value of the extension at χ to b e any real number,

say c. It suffices t o extend φ to F[x]. Let φ* be the extension we will
construct.
For any monomial ax in F[x], φ*(αχ ) must be φ(α) + nc.
n η
Define φ* o n any polynomial as t h e minimum of its values on any

of the monomials in the polynomial. T h e n extend φ* t o F(x) by
defining (£*(/(x)/g(x)) = Φ*(/(χ)) - Φ*( (χ)). (We will check it 9
for elements in F[x], leaving the check for elements in F(x) to t h e

reader.) We show that this extension is in fact a valuation.
We first show that
φ*(/(*) + g(x)) > min{^(/(*)),φ*(g(x))}.
Let f(x) = Σ η " α χ

and g(x) = £ b x . n
n
Then
(f{f(x)+g(x)) = Φ* ( ] > > " + M * " )
= min{</>(a + 6 ) + nc} n n
> min{min[0(a ),0(6„)] n +nc}
= ιτάη{φ(α ) η + nc, </>(6 ) + nc} n
= Ώύη{φ·(/(χ)),φ·( {χ))}. 9
In order to deal with φ* (/(x)o(x)) we will need the fact that

if φ* (f(x)) < φ* (g(x)), t h e n φ* (f(x) + g(x)) equals φ* ( / ( χ ) ) . To
establish this, let m be an index such that 0 * ( a x ) = 0*(/(x)). m
m
T h e n 0 * ( a x ) < 0 * ( 6 x ) since <£*(/(x)) < </>*(σ(χ)). H e n c e

m
m
m
m
0 ( a ) < <£(& ) a n d therefore φ(α + b ) - φ{α ). H e n c e

m m ιη m τη
Φ* ((a m + b )x )
m
m
= Φ{αχη + b ) + mc m
= φ(α ) + mc τη
= **(/(*)).
Thus
φ·{ί(χ)+ (χ))=φ·{ηχ)).
9
Now we are ready t o consider φ* (f(x)g(x)). F o r each nonneg-

ative integer η we have
> φ(α$ί) + n c if i+j = n

= φ(α ) + <A(6j) + nc
ί
= Φ(αί) + ic + 0(6j) + jc
> *·(/(*))+•
Next we produce a monomial in f(x)g(x) at which φ* takes on the
value φ·(Ηχ))+φ·( (χ)). 9
Let r b e t h e smallest index such that φ*(α χ ) is equal t o ί

ί
φ*(/(χ)) and let a b e the smallest index such that φ*(^χ*) is equal
to φ* (g(x)). Consider t h e value of φ* at t h e monomial in f(x)g(x)
of degree r + s:
φ*(αο^ .χ +
Γ+β
+ ••• + a b x
r a
r+a
+ • · • + a .b x ).
r+ 0
r+a
(1)
If i + j = r + s, b u t i is n o t r , then φ*(α^χ ') is strictly larger

τ+
than φ*(α ο χ ').

Γ 8
Γ+
T h u s (1) is equal t o 0*(α ί> χ *), a n d therefore
Γ 4
Γ+
equals 0 ' ( / ( x ) ) + 0* (<?(*)).
Exercise 1 1 . Verify that φ* is a valuation o n F(x).
We have seen how t o extend a valuation from Q t o a finite alge-

braic extension and from any field to an extension by a transcenden-
tal element. T h e extension of a valuation from a field t o an infinite
algebraic extension is fairly involved, and w e refer t h e r e a d e r t o the
references. In any case, it is possible t o extend any of the p-adic val-
uation on Q t o all of the real numbers. Actually, we n e e d t o extend
it only t o fields obtained from Q by an extension by a finite n u m b e r
of elements.
For each valuation φ on the field of real numbers we decom-

pose the coordinate plane into three sets, as follows:
50 = {(x,y) φ(χ) > 0 and φ(υ) > 0 }
51 = {(x,y) φ(χ)<0 and 0(y) > 0 ( x ) }
52 = {(x,y) φ(χ) > φ(υ) and φ{υ) < θ } .
Label any point Ρ that is in S*, P j . It is easy to see that if P t e

Sit then the point Pi - P is also in Si, 0 < i < 2.
0
Exercise 12. Show that Pj - P € Si. 0
Exercise 13. Let φ be a valuation on R that extends φ . D e t e r m i n e 2
in which set, SO, Si or S , each of these points lies: (0,0), (1,0),

2
(0,1), (1,1), (2,1), (2,2), (1,2).
T h e following lemma is the key to analyzing dissections of poly-

gons into triangles of equal areas.
Lemma 2. Let Τ be a triangle in the coordinate plane whose vertices

are (xn.yo). (xi,Vi), ond (x ,y ), where (xi,yi) is in Si, relative to a
2 2
valuation φ. Then
0(Area of Γ ) < - 0 ( 2 ) .
Proof. Since translation by (-xo, -yo) does not change areas, and
P\ — Po 6 S j , 0 < * < 2, we may assume for convenience that
(zo> i/o) is (0,0). T h e area of Γ is then the absolute value of
xi l/i
-Axiyi -Z2I/1).
X2 2/2
Now, φ(χχ) < 0 and φ(χ ) < 0(j/i). Also, </>(j/ ) < 0 and 0 ( y ) <
χ 2 2
φ{χ ). T h u s 0 ( x i j / ) < 0 ( x y i ) and 0 ( χ ι ί / ) < 0. H e n c e

2 2 2 2
0(Area of T ) = 0 ( 1 / 2 ) + 0 ( x i j / ) < 0 ( 1 / 2 ) = - 0 ( 2 ) .
2 •
Exercise 14. U s e L e m m a 2 to show that each line in R 2

meets at
most two of the sets So, Si, and S . 2
Exercise 14 shows, for example, that if the vertices of an edge

are labelled Po and P i , then any intermediate vertex is labelled P 0
orPj.
2. The square
All the machinery for analyzing dissections of the square is now in
place. But we still need a few definitions. We restrict our attention
to simplicial dissections, though t h e theorems hold for all dissec-
tions.
We call a dissection of a polygon into m triangles of equal areas
an m-equidissection. Let φ b e a valuation o n the reals. A triangle
whose vertices are labelled Po, P i , and P relative to t h e valuation φ
2
we will call complete. (The definitions of "complete," given relative

to t h e letters A, B, and C, carry over to t h e case when the letters
are Po, P i , and P . ) T h e next lemma is where Sperner's L e m m a ,
2
valuations, and equidissections come together.
Lemma 3. Let D be an m-equidissection of a polygon of area A and

let φ be a valuation on the real numbers. If D contains a complete
triangle relative to φ then
Φ(τη) > φ(2Α).
Proof. Let Τ be a complete triangle in the equidissection. Its area

is A/m. By L e m m a 2,
Φ(Α/η) < -φ(2), or φ(Α) - 0 ( m ) < - 0 ( 2 ) .
Thus
0 ( m ) > φ(Α) + 0(2) = φ(2Α). D
We illustrate t h e utility of L e m m a 3 by applying it to equidis-

sections of a square.
Theorem 1. ( R i c h m a n - T h o m a s - M o n s k y ) In any m-equidissection

of a square m is even.
Proof. Consider the square whose vertices are (0,0), (1,0), (1,1),
and (0,1). Relative to a valuation φ that extends the 2-adic valuation
on Q, these vertices are labelled P , P\, Pi, and P , respectively, as
0 2
shown in Figure 4.
Figure 5 shows a typical simplicial dissection of this square.
By Exercise 14, vertices on the b o t t o m edge of the square can
b e labelled either P or P i , on the left edge either P or P , o n the
0 0 2
top edge either P or P , and on the right edge either P j or P .

x 2 2
T h e b o r d e r is broken into sections by the vertices of the dissection

that lie on the border. Therefore the only sections whose ends are
labelled P and P i are on the b o t t o m edge of the square. T h e r e are

0
an odd n u m b e r of them. By Sperner's L e m m a , there is a complete

triangle. By L e m m a 3
φ(τη) > 0 ( 2 ( A r e a o f the square)) = φ(2) = 1.
Therefore m is even. •
3. Other polygons
T h e square can b e generalized in many directions: to n-dimensional
cubes, to regular polygons, to trapezoids, to quadrilaterals, to cen-
trally symmetric polygons, and so on.
In the first sequel to the t h e o r e m about the square, M e a d in
1979 [7] proved that if the η-dimensional cube is divided into sim-
plices of equal volumes, their n u m b e r must be a multiple of n!. (It
is easy to construct a dissection into any multiple of n! simplices.)
Exercise 15. Verify the statement in parentheses.
In 1985 Elaine Kasimatis, t h e n a graduate student, was look-

ing for some algebraic topic she could slip into G. D . Chakerian's
geometry seminar at Davis. Stein suggested that she report o n dis-
sections of the square and cube, a topic that Chakerian grudgingly
admitted was geometric. After the talk, Stein asked, " W h a t about
the regular p e n t a g o n ? " Kasimatis's answer was published in 1989
[4]:
In an equidissection of a regular n-gon, where η is at least
5, the n u m b e r of triangles must be a multiple of n.
At this point Kasimatis and Stein [5] posed the general ques-
tion, "If the plane polygon Κ has an m-equidissection, what can be
said about m ? " They introduced the spectrum of K, defined as the
set of integers m such that Κ has an equidissection into m triangles,
and d e n o t e d S(K). If m is in S{K) then so is any positive integer
multiple of m. If S(K) consists of the multiples of a single integer,
m, then Κ and S(K) are called principal and S(K) is also written
as (m). For instance, the q u o t e d results show that the square, the
(o,i) (α, 1)
The Trapezoid T(a)
χ
|(o,o) (1,0)
FIGURE 6
n-dimensional cube, and any regular polygon are principal. As we

will see, not every polygon is principal and, furthermore, we will
construct some polygons that have n o equidissections.
Let us begin with trapezoids. Any trapezoid can be carried by
an affine mapping into a trapezoid three of whose vertices are (0,0),
(1,0), and (0,1). Since an affine transformation magnifies all areas
by the same factor, it carries an equidissection into an equidissec-
tion. Therefore we consider a trapezoid of the type shown in Figure
6, whose fourth vertex is (α, 1).
A trapezoid is affinely equivalent to this trapezoid if the ratio
between the lengths of its parallel sides is a. Let T ( o ) d e n o t e the
trapezoid in Figure 6.
T h e next two theorems represent extreme cases, when a is tran-
scendental and when α is rational.
Theorem 2. If a is transcendental, T(a) has no equidissection.
Proof. Assume, on the contrary, that T(a) has an m-equidissection.

Let ρ be a prime larger than m . N o t e that the labelling of the four
vertices in Figure 6 is complete relative to any valuation φ , no mat-ρ
ter what φ {α) is. In any dissection of the trapezoid, all the sections
ρ
whose ends are labelled Po and P i occur on the b o t t o m edge and

t h e r e are an odd n u m b e r of them. Furthermore, the area of T(a) is
(l + a ) / 2 .
Applying Sperner's L e m m a and L e m m a 3, we see that
φ (τή)
ρ > φ (α + 1).
ρ
Since α -I-1 is transcendental, we can prescribe φ (α + 1) t o b e 1.

ρ
T h u s ρ divides m , contradicting the choice of p. H e n c e t h e r e is n o

equidissection of Τ (a). •
Exercise 16. A similar theorem does not hold for dissections into
quadrilaterals of equal areas, as mentioned in [2]. Show that any
polygon can be divided into quadrilaterals of equal areas. (For con-
venience, restrict the polygons to b e convex.)
T h e next theorem, in contrast with the preceding one, shows

that when a is rational, Τ (a) has equidissections and is principal.
Theorem 3 . Let a be a rational number, a = r/s, where r and s are

positive, relatively prime integers. Then the spectrum ofT(a) is (r+s).
Proof. First of all, T(a) has many (r + s)-equidissections. F o r in-

stance, a diagonal of T(a) cuts T(a) into two triangles whose areas
are in the ratio of r to a. Dividing the first triangle into r triangles of
equal areas and the second triangle into a triangles of equal areas
produces an (r + β)-equidissection of T ( a ) . T h u s (r -(- a) is a subset
ofS(T(o)).
Next we show that for any m-equidissection, m must be a mul-
tiple of r + a.
For convenience, consider the affine image of T(r/a) obtained
by magnification parallel to the χ axis by a factor a. T h a t is, consider
the trapezoid Τ whose vertices are (0,0), (a, 0), (r, 1), and (0,1),
shown in Figure 7.
Let ρ b e a prime dividing r + a. We wish to show that for an
m-equidissection of Τ, φ (τη) > φ (τ + a).
ρ ρ
Since ρ divides r + a, and r and a are relatively prime, φ (τ) = ρ
0 = 0 ( a ) . T h e vertices of Τ therefore have the complete labelling

p
relative to φ shown in Figure 8.

ρ
As with the square, in any equidissection of Τ t h e r e is

a complete triangle. Since the area of Τ is (r + a)/2 we have, by
L e m m a 3,
Φ (τη) > φ (2{τ

Ρ ρ + s)/2) = φ {τ + s).
ρ
Since φ (τη) > φ {τ + s) for each prime ρ that divides r + s, it

ρ ρ
follows that r + s divides m . •
T h e o r e m s 2 and 3 raise the question, "What about equidissec-

tions of the trapezoid T(a) when a is algebraic but not rational?"
T h e next few theorems sample some of what is known.
Theorem 4. Let t\, and £3 be positive integers such that t\ - 4t\t$

is positive and is not the square of an integer. Let a be one of the roots
of the equation t x - t x + t i = 0 . Then T(a) has an equidissection
3
2
2
into ti +1 + h triangles.
2
(0,1) (c, 1)
(0, b)
Τι
(0, 0) (1,0)
FIGURE 9
Proof. Let b be a number in the open interval (0,1) and c be a

positive number that will be determined later. Cut T(c) into three
triangles, Τχ, T , and T , defined as follows. Τχ has vertices (0,6),
2 3
( c , l ) , and (0,1); T has vertices (0,6), (1,0), and ( c , l ) ; T has

2 3
vertices (0,0), (1,0), and (0,6). These triangles are shown in Figure
9.
Tj has area Αχ = c(l - 6)/2; T has area A = (c6 + 1 - 6)/2;
2 2
T has area A = 6/2. We shall determine 6 and c so that the areas A\,
3 3
A , and A are proportional to the integers t\,t , and t , respectively.

2 3 2 3
For such a choice of numbers the trapezoid T(c) would then have a
(h +t + t ) -equidissection.
2 3
The equations
— = — and — = —
Ai h Αχ tx
become
cb+l-b t _, 6 t
= — and
2 3
c(l - 6) ii c(l - 6) ίχ'

Straightforward algebra shows that 6 = ct /(t + ct ), where c is 3 x 3
either root of the quadratic t x - t x + t = 0. Because t - 4txt is

3
2
2 x 2 3
positive, the root is real. (Note that c is positive and that 0 < 6 < 1.)
•
Let us examine the spectrum of T(a) more closely for three values
of a that satisfy the conditions of Theorem 4, namely, a = 2 — y/2,
a = 2 - a/3, and α = (6 + ν 2Ϊ)/3. /
In the first case, α is a root of the equation x — 4x + 2 = 0. Thus

2
7 is in S(T(2 - a/2)). Reasoning similar to that used before shows

that for any m-equidissection of the trapezoid Τ(2 — y/2),
Let d = 3 - a/2 and e = 3 + a/2. Then d e = 7 and ci + e = 6.

Hence 07(d) + 07(e) = 1 and 0 7 ( d + e) = 0. If 07(d) were equal
to 07(e), we would then have 0 7 ( d ) = 1/2 = 07(e) and therefore
0 7 ( d + e) > 1/2, contradicting the fact that 0 ( d + e) = 0. Therefore
7
07(d) and 07(e) are not equal. One of them must therefore be 0
and the other one must be 1. If 07(3 - y/2) = 1, we already have
0 ( m ) > 1. If 0 ( 3 - y/2) = 0, then 0 ( 3 + y/2) = 1. Let U be
7 7 7
the automorphism of Q(y/2) that takes y/2 to —y/2. Define a new

valuation 0* on Q(s/2) by letting φ*(χ) = φγ(υ~(χ)). Extend this
valuation to R. Then 0*(3 — y/2) = 1. Using this valuation instead of
07 again shows that 7 divides m. Therefore Τ(2 — y/2) is principal
with spectrum equal to (7).
If a = 2 - \/3, the argument is slightly different. In this case a is
a root of the equation x — Ax + 1. Therefore 6 is in the spectrum of
2
Τ(2 - y/3). Since the area of this trapezoid is (3 - y/3)/2, φ (τη) > ρ
0p(3 — y/3) for any of its m-equidissections and any prime p. If ρ = 3,

we obtain, since 0 (3) = 1 and 0 (λ/3) = 1/2, 0 ( 3 - y/3) = 1/2.
3 3 3
Thus 03 (τη) > 1/2, which shows that m is a multiple of 3.

Now consider 0 ( 3 - y/3). Let d = 3 - a/3 and e = 3 + y/3. We
2
have de = 6 and d + e = 6. Hence 02(d) + 0 ( e ) = 1 and 0 ( d + e) =

2 2
1. If 0 2 ( d ) and 0 ( e ) were unequal, then one of them would be less

2
than 1/2 and force 0 ( d + e) to be less than 1/2, contradicting the

2
fact that 0 ( d + e ) = 1. Therefore 0 ( d ) = 0 ( e ) = 1/2. Hence m is

2 2 2
a multiple of 2, and the spectrum is (6).

The case a = (6 + ν 2 Ϊ ) / 3 (or a = (6 - \/2Ϊ)/3), however, is
/
not settled. This trapezoid corresponds to t\ = 3, i = 12, and i = 5. 2 3

It has area (9 + ν 2~Γ)/6 and therefore <f> (m) > 0 ( ( 9 + ν 2 Ϊ ) / 3 ) .

/
p P
/
We know that h + t + h = 20 is in the spectrum.

2
Letting d = (9 + Λ / 2 Ί ) / 3 and e = (9 - \ / 2 Ϊ ) / 3 , we have de =

2 0 / 3 and d + e = 6. Reasoning as before, we obtain 0s(d) = 1/2,
from which it follows that 5 divides m . Turning to 0 , we have 2
02(d) + 02(e) = 2 and 0 (d + e) = l.

2
T h e assumption that 0 (d) is not equal to 0 (e) quickly leads to a

2 2
contradiction. Therefore they are equal and have the value 1. Thus
2 divides m and we can conclude that 10 divides m . All told, we
know that
In this case the spectrum is not determined, and we do not know

whether it is principal.
Problem 1. Find the spectrum of the trapezoid T ( (6 + α / 2 Ϊ ) / 3 ) .
Problem 2. D o e s the trapezoid T(y/2) have an equidissection? (It

can be shown by methods in the next chapter that the answer is n o
if we d e m a n d that the coordinates of the vertices lie in the held
Q(V2).)
Problem 3. For which algebraic irrational numbers a does T ( o )

have an equidissection? (Conjecture: W h e n the real conjugates of
a are all positive.)
Problem 4. Is every trapezoid principal?
T h e r e are in fact quadrilaterals that are not principal. For in-

stance, consider t h e quadrilateral whose vertices are (0,0), (1,0),
(0,1), and ( 3 / 2 , 3 / 2 ) . Its long diagonal cuts it into two triangles of
equal areas. O n the other hand, its short diagonal cuts it into a tri-
angle of area 1/2 and a triangle of area 1, which the long diagonal
divides into two triangles of areas 1/2. H e n c e both 2 and 3 are in
the spectrum of the quadrilateral, and the spectrum is not principal.
This quadrilateral suggests that we consider the class of quadrilat-

erals that are symmetric with respect to o n e diagonal.
Let Q(a) b e the quadrilateral whose vertices are ( 0 , 0 ) , (1,0),
(a, a ) , and (0,1), where α is a n u m b e r larger than 1/2 (so that the
quadrilateral is convex). T h e next theorem shows that an infinite
number of these quadrilaterals are not principal.
Theorem 5. Let a = r/(2s), where r and s are relatively prime pos-

itive integers, r is odd and is larger than s. Then the spectrum of the
convex quadrilateral Q(a) contains 2 and the integer r + 2sk for each
nonnegative integer k.
Proof. Let b e a nonnegative integer and let χ = a/(a+fc). Divide

Q(a) into three triangles with the aid of the vertex (a:, 0). (If fc = 0,
there are only two triangles.) T h e triangles are shown in Figure 10.
T h e areas of the three triangles are in the proportion
χ : χα — χ + a : a(l — x)
or
Replacing α by r/(2s) and simplifying, we see that the areas are in

the proportion
s : r — s + sk : ak.
V (α, a)
(0,1)'
•» χ
(x,0) (1.0)
FIGURE 1 0
Therefore the quadrilateral has an equidissection into a + (r- s +

sk) + ak = r + 2ak triangles, as claimed. •
In the preceding t h e o r e m 02 (a) < - 1 . T h e next t h e o r e m shows

that if 02(a) > - 1 , then Q(a) is principal.
Theorem 6. 7 / 0 ( a ) > - 1 , then the spectrum ofQ(a)is

2 (2).
Proof. If - 1 < 02(a) < 0, then Q(a) has only o n e edge labelled
PQP\, and we can quickly conclude that in any m-equidissection of
Q(a) m is even.
If 02(a) > 0, apply the shear (x,y) —• (x/a,y), obtaining a
quadrilateral with only o n e edge labelled PQP\. T h e area of this
quadrilateral is α / α = 1. Working with this quadrilateral, we see
again that m is even. •
Exercise 17. Fill in the details in the preceding proof.
T h e o r e m s 5 and 6 d o not cover irrational a for which 0 ( a ) <
2
- 1 , for instance, the case α = V3/2. So we have a problem similar

to the one concerning equidissections of the square.
Problem 5. Can Q ( \ / 3 / 2 ) b e divided into an odd n u m b e r of tri-

angles of equal areas?
4. Regular polygons
A s already mentioned, Kasimatis proved that the spectrum of the
regular n-gon, η > 5, is (n), and therefore is principal. In t h e gen-
eral case the proof employs valuations extended to the complex
numbers. However, for the regular hexagon it does not, a n d we
present the proof in this case, since it illustrates some of t h e ele-
ments of the general approach.
At first glance we would expect to use the well-known regu-
lar hexagon inscribed in the standard unit circle with center at the
origin, as shown in Figure 11.
However none of its vertices is labelled P relative to the valua-
2
tions 02 or 0 3 . Consequently we take an affine image of this hexagon.

O n e that works is shown in Figure 12, where three consecutive ver-
tices are (1,0), (0,0), and (0,1), and the other vertices are (2,1),
(2,2), and (1,2). Of course this step is opposite the m o r e custom-

ary approach of trying to analyze a problem in its most symmetric
form. In this case we are destroying the rotational symmetry of the
hexagon.
T h e area of the "distorted" polygon shown in Figure 12 is 3.
Relative to the valuation </> there is only o n e PQP\ edge. Therefore,
3
ΦΆ (m) > 03(2-3) = 1 for any m-equidissection of the polygon, and

we see that m is a multiple of 3.
T h e demonstration that m is even is not so direct. Relative t o
the valuation φ the vertices of the distorted hexagon are labelled
2
P , PI, P , P , Pi, P , where the vertices are listed counterclockwise

0 2 0 2
starting at the origin. T h e r e are two P Pi edges. N o t e that they are

0
swept out in the same direction, from P t o Pi. Fortunately, there

0
is a refined version of Sperner's L e m m a that implies that t h e r e is

a complete triangle in any dissection, from which it follows that 2
divides m. (We state this version below.)
T h e statement and proof of the stronger Sperner's L e m m a are
similar to those of the one already discussed.
Consider a simplicial dissection of a polygon in which each ver-
tex is labelled A, B, or C. Let n (ABC)
+ be the n u m b e r of complete
triangles in the dissection for which the order A to Β to C is coun-
terclockwise. Let n_ ( A B C ) b e the number for which that o r d e r is
clockwise. Similarly, let n (AB)
+ be the n u m b e r of sections on the
polygon labelled AB for which t h e order A to Β gives a counter-
clockwise orientation to the boundary. Let n_ (AB) b e the n u m b e r
for which that orientation is clockwise.
Theorem 7. Consider a simplicial dissection of a polygon in which

each vertex is labelled A B, or C. Then
n (ABC)
+ - n-(ABC) = n (AB)
+ - n-(AB).
T h e proof is about t h e same as that of the " n o n - o r i e n t e d "

Sperner's L e m m a . In each triangle place a 1 along a n edge AB in
which the orientation A to Β gives a counterclockwise orientation
to t h e triangle. Place a - 1 if the induced orientation is clockwise.
Summing u p all t h e l's and — l's in terms of the triangles a n d also
in terms of the edges immediately yields the theorem.
Exercise 18. Complete t h e proof of T h e o r e m 7.
5. Centrally symmetric polygons

In 1987 Stein, wondering whether t h e symmetry of a polygon in-
fluences its spectrum, first considered mirror symmetry. Since an
isosceles triangle is mirror symmetric and its spectrum is (1), clearly
this type of symmetry puts n o constraint o n the spectrum. O n the
o t h e r hand, by the t h e o r e m on equidissections of a square and by
Kasimatis's theorem, the spectrum of a centrally symmetric regular
polygon does not contain any odd numbers.
T h e R i c h m a n - T h o m a s - M o n s k y theorem implies that in any

m-equidissection of a centrally symmetric 4-gon, that is, a paral-
lelogram, m is even. Stein tried to construct a centrally symmetric
6-gon that has an m-equidissection with m odd. T h e attempt e n d e d
with a proof that the spectrum of a centrally symmetric hexagon
contains n o odd numbers. We present this proof, as simplified by
Monsky.
Let the vertices of the hexagon be, in order, D\, D , D , D , 2 3 4
D , and D . Consider the areas of the six triangles Α Α + ι A + 2 ,

5 e
where we interpret D as D\ and D as D . Since we are assuming

7 s 2
that the hexagon is not a parallelogram, n o n e of these areas is 0.

Let us index the vertices so that </> (Area D\D D ) is the minimum 2 2 3
of the six numbers </> (Area DiD iD ), 2 1 = (a, b + 1). Then D = (a + 1, b + 1), and

3 4 5
De — (α + 1, b), as in Figure 13.

T h e area of triangle DiD D is 1/2. Therefore 2 3
φ (α/2)
2 = 0 (Area DDD) 2 2 3 4 > φ (1/2),
2
and we conclude that φ (α) > 0. Similarly, by considering triangle

2
D D\D ,
6 we see that 0 ( b ) > 0. T h e labelling of the vertices of
2 2
t h e hexagon in Figure 13 relative t o the valuation φ is almost com- 2
pletely determined, as shown in Figure 14. T h e main ambiguity is

at£> . 5
(a, 6 + 1 ) (a+l,fa+l)
(a+1,6)
(1,0)
FIGURE 13
Po, Λ . or Pa
f
Pi orPj
FIGURE 14
If it is not labelled Po. then t h e r e is only o n e P o P i edge and

Sperner's L e m m a applies. If D is labelled P , Sperner's L e m m a
5 0
again applies, though we n e e d the refined version if D is labelled

6
P . Since the area of the hexagon is 1 + α + 6, we t h e n have

2
φ (τη)
2 > φ (2(1
2 + a + b)) >φ (2)2 = 1,
showing that in an m-equidissection m is even.

This technique works for a centrally symmetric octagon [11],
but the n u m b e r of separate cases t o b e considered grows rapidly
with the n u m b e r of sides. Using homology groups, Monsky [9] es-
tablished the theorem in full generality: A b o u n d e d centrally sym-
metric polygon (even with holes) cannot b e cut into a n odd n u m b e r
of triangles of equal areas.
Problem 6. Can a centrally symmetric η-dimensional polyhedron

be cut into a n odd n u m b e r of simplices of equal volumes?
A s in the previous two chapters, we find ourselves with more

questions than answers. In those chapters we saw that we d o not
fully understand finite cyclic groups, the most elementary groups
imaginable. In this chapter we discovered that even a trapezoid
raises a question that has not been answered. For that reason we
should be grateful that at least some questions have yielded t o the
available machinery.
T h e next chapter is almost the opposite of this one. In it we

consider dissections into triangles of certain shapes; their angles
rather than their areas will concern us.
References
1. G. Bachman, Introduction to p-adic numbers and valuation theory, Aca-
demic Press, New York, 1964.
2. A. W. Hales and E. G. Strauss, Projective colorings, Pacific J. Math.
99 (1982), 31-43.
3. K. Ireland and M. Rosen, A classical introduction to modern number
theory, Second Ed., Springer Verlag Graduate Texts in Mathematics
84, New York, 1982.
4. E. A. Kasimatis, Dissections of regular polygons into triangles of
equal areas, Discrete and Computational Geometry 4 (1989), 375-381.
5. E. A. Kasimatis and S. K. Stein, Equidissections of polytopes, Discrete
Math. 85 (1990), 281-294.
6. V. Klee, Facet-centroids and volume minimization, Studia Seien.
Math. Hung. 21 (1986), 143-147.
7. D. G. Mead, Dissection of a hypercube into simplices, Proc. Amer.
Math. Soc. 76 (1979), 302-304.
8. P. Monsky, On dividing a square into triangles, Amer. Math. Monthly
77 (1970), 161-164.
9. , A conjecture of Stein on plane dissections, Math. Zeit. 205
(1990), 583-592.
10. F. Richman and J. Thomas, Problem 5471, Amer. Math. Monthly 74
(1967), 329.
11. S. K. Stein, Equidissections of centrally symmetric octagons, Aequa-
tionesMath. 37 (1989), 313-318.
12. J. Thomas, A dissection problem, Math. Mag. 41 (1968), 187-190.
13. B. L. van der Waerden, Modem Algebra, Vol. 1, Ungar, New York,
1949.
Chapter 6
Tiling by Similar Triangles
Lajos Posa wrote his first paper when still in primary school and is
well known for his work in graph theory. A n enthusiastic teacher, h e
tries to convey the beauty of mathematics t o students of all abilities,
from t h e most talented t o the least able, a n d of all ages, from small
children t o candidates for the doctorate. Often h e holds irregular
classes in t h e most r e m o t e towns.
H e also organizes summer schools t o which h e invites students
from all over Hungary. In 1987, when preparing for such a session
h e decided h e needed a concrete geometry problem. Now, it is well
known [1] that it is possible t o cut any polygon into triangles in such
a way that t h e triangles can be assembled to form any preassigned
polygon of the same area as that of the original polygon. Posa won-
dered whether it is possible to cut an equilateral triangle into 30°-
60°-90° triangles that could be p u t together t o form a square. After
working o n t h e problem for five minutes h e started to like it. After
ten minutes h e decided that it was interesting enough to assign t o
his students. After half an hour h e grew a little upset, for he still
could n o t solve it. A t that point h e stopped, for geometry was far
from his main interest.
H e mentioned his experience to his good friend from school
days, Miklos Laczkovich, who found t h e problem appealing.
Laczkovich solved it and went far beyond. In a 25-page p a p e r pub-
lished in 1990 [2] he proved that it is impossible to cut a square into
a finite n u m b e r of 30°-60°-90° triangles. In fact, h e showed, as a
135
special case of o n e of his theorems, that it is impossible t o cut a

square into a finite n u m b e r of triangles all of whose angles, when
measured in degrees, are even. A s we will see in this chapter, his
proof is almost completely algebraic, involving such tools as fields,
vector spaces, isomorphisms between fields, and complex roots of
unity.
Throughout this chapter the dissections need not be simplicial.
1. The machinery
Let α b e the angle determined by lines of slopes m i and ma, neither
of which is parallel to the y axis, as in Figure 1. Let θ χ and θ be the 2
corresponding angles of incidence of the two lines, with θ > θχ. 2
T h e n α = θ - θχ and t a n θ - t a n θχ
2 2 τη - m i 2
tana =
1 + t a n 02 t a n θχ l + mim 2
However t a n a is not defined when a = π / 2 . Since we want to b e

able to work with any angle in a triangle, in particular, with a right
angle, we will use cotangents instead of tangents, a n d the equation
1 + τηχτη
cot α =
2
(1)
τη2 — mi
Rewriting (1) we have
(m2 — m i ) cot α = 1 + πΐχπι .

2 (2)
slope τπ2 slope m i

ν
FIGURE 1
So, if we know a and rn,2, we can determine m i , which then lies

in the field generated by cot α and m . Therefore, if we are given a
2
sketch of a dissection of a convex polygon Ρ into triangles, as in Fig-

ure 2, and the angles of all the triangles and the slopes of the edges
of P, we can calculate the slopes of all the edges of the triangles. We
could start by determining the slopes of all the edges that meet the
boundary, using (2), then gradually work inward from the bound-
ary, repeatedly using (2). Therefore the slopes of all the edges of
all the triangles belong to the field generated by the coordinates of
the vertices of Ρ and the cotangents of the angles of the triangles.
T h e following lemma, which will b e an important tool, says
much m o r e . T h e assumption in it that Ρ is convex is not necessary,
but simplifies the proof a little. By a "vertex of P" we m e a n a cor-
ner vertex, not a point situated in the interior of an edge of P. T h u s
adjacent edges of Ρ are not parallel and t h e interior angles of Ρ are
less than π.
L e m m a 1. Suppose that a convex polygon Ρ in the xy plane is tiled

by a finite set of triangles. Then the coordinates of the vertices of each
triangle belong to the field generated by the coordinates of the vertices
of Ρ and the cotangents of the angles of the triangles.
Proof. We may assume that n o n e of the edges of any of t h e tri-

angles is parallel to the χ o r y axes. Indeed, if any were parallel to
an axis, r o t a t e the polygon about t h e origin by an angle θ that has
rational cosine and sine in such a way that n o n e of t h e rotated lines
is parallel t o an axis. (Since there are an infinite n u m b e r of angles
0 < θ < 2π with rational cosine and sine, there is such a rotation.)
Now let x i < x < · · • < x b e the χ coordinates of t h e ver-
2 r
tices of t h e triangles. We will show that they all lie in the field F
generated by the cotangents of the angles of the triangles and the
coordinates of the polygon P. (A similar argument goes through
for the y coordinates of the vertices.)
First of all, x i and x are in F since they are t h e χ coordinates
r
of vertices of P. Consider a particular X j , 2 < t < r - 1 . T h e r e is a

n u m b e r y such that A = (x<, y) is a vertex of a triangle in t h e tiling,
as shown in Figure 3. Let Β and C b e the o t h e r vertices of o n e such
triangle such that t h e edge BC crosses the vertical line χ = χ* at
a point D = (x<, u). Let α , β, η, and δ b e t h e angles as labeled in
Figure 3. Observe that their cotangents lie in t h e field F since the
slopes of the lines o n the three sides of triangle ABC are in F. Let
FIGURE 3
Β have χ coordinate Xj a n d C, χ coordinate Xk. We may assume

that Xj < χ, < Xk, since t h e three numbers are distinct.
T h e length of AD is
\ u
~ y\ — { ί χ
~ Xj){cota + cot/3) ^
= (xfc - X j ) ( c o t 7 + cot δ).
Since \u — y\, Xi - Xj, a n d Xk - Xi are positive, t h e r e a r e positive

n u m b e r s s = c o t a + c o t β and t = cot 7 + c o t 6 in F such that
s(xi -Xj) = t(xk - Xi). (4)
Now let L b e the vector space generated by the numbers χχ,

x , . . . , x with coefficients in F, that is, all n u m b e r s of t h e form
2 r
c i x i + C 2 X 2 -I l· CrX , c, in F. Since at least o n e of the n u m b e r s

r
x i and x is n o t 0, L contains F. We wish t o show that t h e dimension

r
of L is o n e , for then L would b e just F.

Assume that the dimension of L is greater than o n e , and let
1 and b b e part of a basis. (Since L contains an element of F, we
may assume that 1 is a p a r t of a basis for L.) It then follows that b
is not in F. F o r each χ in L let c ( x ) b e t h e coefficient of b in t h e
representation of χ as a linear combination of t h e basis elements:
χ = c(x)6H .
T h e n C(XJ) is n o t 0 for at least o n e index i. We may assume that
C(XJ ) is positive for at least one i. (Otherwise replace b in t h e basis by
—b.) L e t m b e t h e maximum of t h e n u m b e r s c ( x 1 ) , c ( x 2 ) , . . . , c ( x ) . r
Let i b e t h e largest of those indices ρ such that c(x ) = τη. T h e n p
c(s(xi - Xj)) = c(t(x k - Xi)).
Hence
s(c(xi) - C(XJ)) = t(c(x ) k - c(xj)). (5)
By t h e definition of the index i, c(xk) - C(XJ) is negative a n d C(XJ) -

C(XJ) is nonnegative. This contradiction shows that t h e dimension
of L over F is o n e , which completes t h e proof. •
Exercise 1. Show that t h e r e are an infinite n u m b e r of angles in t h e

first q u a d r a n t whose sine and cosine are rational.
Exercise 2. Verify equations 4 and 5.
Exercise 3. Check the claim that for the vertex A t h e r e are vertices
Β and C such that the triangle ABC is in the tiling and the line
χ = Xi crosses the edge BC.
Exercise 4. In 1903 D e h n proved that if a square with a rational

width is tiled by finitely many rectangles such that the ratio of the
width to the length of each rectangle is rational, t h e n the dimen-
sions of each rectangle are rational. Show that this follows immedi-
ately from L e m m a 1.
T h e next exercise shows that the lemma could be phrased in

terms of slopes rather than cotangents.
Exercise 5. (Assume L e m m a 1.) Suppose that a convex polygon

Ρ in the xy plane is tiled by a finite set of triangles, n o n e of whose
sides is parallel to the y axis. T h e n the coordinates of the vertices
of each triangle belong to the held generated by the coordinates of
the vertices of Ρ and the slopes of the sides of the triangles.
T h e next lemma involves the notions of an oriented triangle

and the relation between the cotangent of an angle and the slopes
of its two arms. We pause t o describe these ideas.
Let A = ( α ι , α ) , Β = (6ι,ί> ), and C = ( c i , c ) b e t h r e e non-
2 2 2
collinear points in the xy plane. T h e triangle with vertices A, B, and

C is assigned an orientation by ordering its t h r e e vertices in a par-
ticular sequence. For instance, in Figure 4 the o r d e r "first A, then
ν ν
c c
χ X
FIGURE 4
Tiling by Similar Triangle! 141
B, then C , " which we abbreviate to A, B, C, is counterclockwise,

but the order A, C, B, for instance, is clockwise.
T h e notion of orientation can also be expressed in terms of the
sign of a determinant. Consider the case A = (0,0), Β = (1,0), and
C = (0,1). T h e order A, B, C is counterclockwise and the 2 by 2
determinant
B - A (1,0)-(0,0) 1 0
-1
II
C-A (0,1)-(0,0) 0 1
is positive. M o r e generally, if we are given the coordinates of t h r e e

noncollinear vertices, A, B, and C, we can determine whether the
order A, B, C is counterclockwise or clockwise by computing the
determinant
B - A
(6)
C-A
W h e n this determinant is positive, the order is counterclockwise;
when it is negative, the order is clockwise.
Exercise 6. Is the order (2,3), (1,4), (3,5) clockwise or counter-

clockwise? Solve by drawing the points and also by using the deter-
minant.
Exercise 7. Show by using vectors or elementary geometry that t h e

absolute value of (6) is twice the area of the triangle whose vertices
are A, B, and C.
We also need to express the cotangents of the angles of any

triangle ABC in terms of the coordinates of its vertices.
Assume that the order A, B, C is counterclockwise and let a
b e the angle at the vertex A, as in Figure 5.
T h e slope of AC is
C2 — 0,2
Ci - Ol
a n d the slope of AB is
62 ~ 0.2
61 - αϊ'
C = (ci,c ) 2 Β = (bi,f>2)
ν
(αι,α )2
FIGURE 5
Substituting these values in (1) gives
_ ( c - a ) ( b - Q2) + ( c i - Qi)(6i - Qi)

2 2 2
( c - a )(bi - a ) - ( c i - α ι ) ( 6 - a ) '
2 2 x 2 2
N o t e that t h e denominator, which is the determinant
B - A
C-A
is positive.
If, on t h e o t h e r hand, t h e o r d e r A, B, C is clockwise, t h e ex-
pression for cot a changes. In this case the o r d e r A, C, Β is counter-
clockwise a n d the formula for cot α is obtained from (7) by switch-
ing the roles of Β and C. T h e n u m e r a t o r in (7) remains t h e same,
but the d e n o m i n a t o r changes sign. This brings us to a key lemma.
L e m m a 2. Let A B, and C be noncollinear points and let F be a

subfield of R that contains their coordinates. Let φ : F —> Rbe an
isomorphism from F onto a subfield of R Define the mapping Φ :
F χ F -> Rx Rby
= (Φ(Χ), Φ(ν))·
Let A ' = Φ(Α), B' = Φ(Β), and C = Φ(<7). Let a be the angle at A
in triangle ABC and a' be the angle at A' in triangle A'B'C. Then
cot a' = 0 ( c o t a )
if the orientations A, B, C and Α', B', C are the same, and
cot a' = —0(cot a)
if they are different.
Exercise 8. Prove that t h e points Α', B', and C" in L e m m a 2 are

not collinear.
Exercise 9. Prove L e m m a 2 in the case that the orientation A, B,

C is counterclockwise and the orientation Α', B', C is clockwise.
Exercise 10. Prove that Φ takes collinear points to collinear points.
Keep in mind that Φ is defined only o n F x F , not necessarily

on the whole xy plane. It need not b e continuous. In fact, it might
not even preserve "betweeness." For instance, let F = Q(\/2) and
φ : F —• R be defined by φ(τι + r \/2) — r i — r \/2, where τ-χ
2 2
and r are rational numbers. T h e point (1,0) is between t h e points

2
(0,0) and (V2,0), but Φ(1,0) is not between Φ(0,0)and Φ(\/2,0),

as may easily b e checked.
Exercise 11. Show that the isomorphism φ : Q{\/2) -* R, defined

above, is not continuous.
Now, if α, β, and 7 are the three angles in a triangle ABC, t h e n

at least two of them are acute a n d therefore have positive cotan-
gents. If the mapping Φ in L e m m a 2 preserves t h e orientation of
the triangle, that is, the orientation Φ(Α), Φ(Β), Φ((7) is t h e same as
the orientation A, B, C, then at least two of the numbers φ(οοί a),
0(cot β), and 0(cot 7) are positive, since they are t h e cotangents of
the three angles in a triangle.
With these tools at our disposal, we are ready to analyze dis-
sections of a convex polygon Ρ into triangles.
Let t h e vertices of Ρ be Vi, V ,..., V in counterclockwise or-
2 n
der. We represent the boundary of Ρ by t h e formal sum
ViV + V V3 +
2 2 ----rV V .
n 1
We call such a formal sum a chain. (Appendix C gives a precise def-

inition of a formal sum.) T h e ordered pair Vi_i Vj may b e thought
of as a vector from Vi_i to Vi. If Ρ and Q are two points in the xy

plane, we regard the ordered pair QP as the negative of the o r d e r e d
pair PQ. I n other words, PQ + QP = 0. Also, if an edge AB is di-
vided by a n intermediate point C into two edges AC and CB, we
shall write AC + CB = AB. T h e boundary of triangle ABC with
orientation A, B, C will be written AB + BC + CA. As a result, if
two nonoverlapping triangles share a c o m m o n edge and both have
a counterclockwise orientation, their c o m m o n edges "cancel" alge-
braically. For instance, consider triangles ABC and ACD in Figure
6, b o t h with a counterclockwise orientation. T h e formal s u m of
their algebraic boundaries is
(AB + BC + CA) + (AC + CD + DA),
which reduces t o
AB + BC + CD + DA,
the formal boundary of their union. This cancellation property ex-

tends to any family of similarly oriented triangles that tiles a poly-
gon: t h e sum of their boundaries is the boundary of the polygon.
T h e sum of the boundaries of the five triangles in Figure 7 is
(V4V1 + ViB + BV ) + (ViC + CB + BV ) + (BC + CV + V B)

4 X 2 2
+(ΛΥ 2 + V V + V A) + (AV + V V + V*A),

2 3 3 3 3 4
which reduces, d u e to inner cancellations, to
ViC + CV 2 + V V + V3V4 + V4V1,

2 3
hence to
VI V + V V + V V + V4V1,
a 2 3 3 4
the counterclockwise boundary of the polygon in Figure 7.

T h e proof of the next lemma uses another way of recording
t h e orientation of a polygon. If Γ is the boundary of a convex poly-
gon, the value of the line integral ^xdy d e p e n d s o n w h e t h e r we
sweep out Γ in a counterclockwise or clockwise m a n n e r . Reversing
t h e orientation switches t h e sign of the integral.
Consider the case w h e n Γ is oriented counterclockwise, as in
Figure 8. Break Γ into a "right-hand" path Γ and a "left-hand"2
path Γ ι . O n T y increases, and o n Γι y decreases. We have

2
ψ xdy = ψ xdy+ψ xdy = I x dy— / X\dy = I

2 {x —x\)dy, 2
JT Jr 3 JTi Ja Ja Ja
which is the area of the region that Γ bounds. If Γ were clockwise,
then ff xdy would b e t h e negative of the area that Γ bounds.
L e m m a 3 . Let Ρ be a convex polygon with vertices V\, V~2,..., V~„,

listed counterclockwise. Suppose that Ρ is tiled by the triangles 7 \ ,
ν ν
X X
a
xi(v) xiivj b a b
FIGURE 8
T ,..., T . Let the vertices ofTj be Aj, Bj, and Cj, listed counter-
2 m
clockwise, 1 < j < m. Let F be a real field that contains the coor-
dinates of all the vertices of the triangles and φ : F —• Ran isomor-
phism that leaves the coordinates of the vertices of Ρ fixed. Let Φ be as
in Lemma 2. Then there is at least one triangle Tj such that the order
Φ{Α^), Φ(Β^), Φ (Cj) is counterclockwise.
Proof. T h e algebraic boundary of Tj is AjBj + BjCj + CjAj. T h e

sum of the boundaries for all the triangles Tj, 1 < j < m, is the
algebraic boundary of P , Vi V2 + V2V3 + ••• + VnVi. T h u s the sum
of the chains
6j = Φ(Λ·)Φ(β,·) + Φφ)Φ(σ,) + Φ(<7,)Φ(Α,·), 1< j < m,

is the chain
δ = VM + VM + '-' + VnY!,
since Φ(νί) = V 1 < i < η. Therefore

it
Since the right-hand side of (8) is positive, at least o n e summand

on the left is positive. Let j be an index such that $ xdy is pos-
s
itive. T h e n the orientation Φ ( Α , ) , &{Bj), Φ((7,-) is counterclock-

wise. This completes the proof. •
This lemma is the basis of the next one, which concerns angles
rather than orientations.
Lemma 4. Let Ρ be a convex polygon with vertices V\, V2, · · · , V . n
Suppose that Ρ is tiled by the triangles T\, T%,..., T . Let Fbea real
m
field that contains the coordinates of all the vertices of Ρ and the cotan-
gents of all the angles of the triangles. Let φ : F —> R be an isomor-
phism that leaves the coordinates of all the ν\ fixed. Then there is an
integer j such that at least two of the numbers 0 ( c o t α,·), c4(cot ßj),
and 0(cot7j) are positive, where ctj, ßj, and 7,· are the angles ofTj.
Proof. By L e m m a 1, F contains the coordinates of the vertices of

all the triangles. Let Φ b e as in L e m m a 2. By L e m m a 3, t h e r e is an
integer j such that Φ preserves t h e orientation of T j , that is, if Tj
is given the counterclockwise orientation A , , Bj, Cj, then the ori-
entation Φ ( Α , ) , $(Bj), ${Cj) is also counterclockwise. N o w let the
angles of Tj be ctj, ßj, a n d 7,. By L e m m a 2, φ(οοί atj), 0 ( c o t ßj), a n d
</>(cot 7 j ) are the cotangents of t h e angles in the triangle whose ver-
tices are Φ(-Α,), ${Bj), a n d Φ ( £ ^ ) . Since every triangle has at least
two acute angles, hence at least two angles with positive cotangents,
t h e lemma follows. •
2. Applications
We are now in position t o answer Posa's question.
Theorem 1. It is impossible to tile a square with a finite number of

30° - 60° - 90° triangles.
Proof. We assume, without loss of generality, that the vertices of

t h e square are (0,0), (1,0), (1,1), and ( 0 , 1 ) . L e t F be Q(s/3), a
field that contains the coordinates of the vertices of the square and
also the cotangents of the angles of the triangles in the tiling, since
cot 30° = \ / 3 , cot 60° = 1 / v ^ , a n d cot 90° = 0.
Define φ : F -* R by setting φ(τ\ + r -\/3) = η - Γ Ν / 3 , where

2 2
r i and r are rational numbers. T h e values of φ at the cotangents

2
of the three angles of every triangle are - V 3 , - l / \ / 3 , and 0. This

contradicts L e m m a 4 and therefore proves the t h e o r e m . •
Exercise 12.
(a) U s e Figure 9 t o show that the square can b e tiled by an infinite
n u m b e r of triangles similar to any given right triangle.
FIGURE 9
(b) W h e r e in the proof of T h e o r e m 1 is the assumption that t h e r e

are only a finite number of triangles used?
Exercise 13. Outline the flow of the key arguments that lead u p
t o t h e proof of T h e o r e m 1.
Exercise 14. (Contributed by M a r k Chrisman.) T h e following
steps outline an elementary proof that there is n o simplicial dissec-
tion of a square into 30°-60°-90° triangles.
(a) Assume that t h e r e is such a dissection and that the length of
a side of the square is 1. Let the length of the shortest side of
all the triangles b e β. Show that the length of the shortest side
in each of the triangles is of the form 2 3 / s , w h e r e m and η
m n 2
are integers.
(b) Obtain an equation for a by using the fact that t h e sum of the
areas of the triangles is the area of the square.
(c) Obtain another equation for s by using the fact that an edge of
the square is the union of edges of the triangles.
(d) Using (b) a n d (c), obtain a contradiction.
(e) W h a t properties of the square were used in this argument? To
what polygons does t h e argument apply?
(f) W h a t properties of the 30°-60°-90° triangle were used in this
argument? To what triangles does t h e argument apply?
To treat the more general case than that covered in T h e o r e m 1,

where we now assume only that all the angles of all the triangles in
the dissection have an even n u m b e r of degrees, we have to examine
the effect of an isomorphism o n cot Θ.
Recall that
e ie
+ e- i e
. . _ e -e~ie ie
cos θ — and sin θ = — .

2 2i
Thus
e ie
+ e' ie
cot θ — i
Let ω b e the complex n u m b e r of angle 2 π / 4 η = π / 2 η , where η

is a positive integer, and of modulus 1. T h u s ω = e"^ . N o t e that 2n
ω is a primitive (4n)th root of unity and that i = ω . If the integer η
α is n o t a multiple of n, then cot(a7r/n) is defined and we have
απ ηω +ω~
2α 2α
cot — = ω η
5-, (9)
which therefore lies in t h e field Q(u>).
Lemma 5. Let η be a positive integer and ω = e / . Let k be an wt 2n
odd integer relatively prime to n. Then there is an automorphism φ :

Q(u>) —• Q(UJ) such that
^ f c o t ^ ) = ( _ 1 ) (fc-l)/2 c o t ^![
V η / η
for every integer a not divisible by n.
Proof. Since (k, An) — 1, the n u m b e r u is a primitive (4n)th root k
of unity a n d there is therefore an automorphism φ : Q(u>) Q(u>)

such that φ(ω) = u A Consequently, in view of (9),

,2ofc _|_ ^j—2ak
1Γ>
η
= ω
' ω-)2ak _ u]—2ak
.2ak , , ,-2ak
flak _ ω -2αΑ:
= ( u ,2« ) ( f c - l ) / ,2 c o t
akn
η
(_l)(fc-D/2 akw
= c o t
η
•
Theorem 2. Suppose that each vertex of the convex polygon Ρ has

only rational coordinates. Assume that Ρ is tiled by a finite number of
triangles whose angles are rational multiples of π, hence of the form
α,π/η for some fixed integer η and integers Oj. Then η is a multiple of
4
Proof. Let F be the field generated by the cotangents of all the an-
gles of the triangles in the tiling, that is, by the numbers c o t ( a ^ / n ) .
As we observed, all these n u m b e r s lie in Q(LJ), where ω is a primi-
tive (4n)th root of unity.
Suppose that η is not a multiple of 4. Let k = 2n + 1 if η is
odd and let fc = η + 1 if η is even (hence η = 2 (mod 4)). Since fc is
odd and relatively prime to n , we have (fc, 4n) = 1. T h e definition
of fc assures us also that (fc - l ) / 2 is an o d d integer and that fc Ξ
1 (mod n ) .
By L e m m a 5 t h e r e is an automorphism φ : 0,{ω) —> Q{u) such
that
aikir
0(cot^) =(-l)< - )/ cot f c 1 2
(10)
η
for each a j . T h e restriction of φ to F is then an isomorphism of F
into R.
Now, since (fc - l ) / 2 is odd and fc = 1 (mod n ) , (10) reduces
to
Thus if θ is an angle in any of the triangles of the alleged tiling,

0(cot Θ) — - cot Θ. If α, β, and 7 are the angles in any of these tri-
angles, at least two of the numbers 0(cot a ) , 0(cot β), and 0(cot 7)
are negative. This contradiction of L e m m a 4 completes the proof.
•
Exercise 15. Show that T h e o r e m 2 implies that it is impossible t o
tile a square with triangles all of whose angles, when measured in
degrees, are even integers.
In a letter Laczkovich wrote in 1992, he remarked, "Next I

wanted t o determine all the triangles that tile a square. For a long
time I was convinced that only right triangles can tile a square or
even a rectangle. When I discovered [Figure 10] quite accidentally,
I was shocked." A polygon Ρ is said to tile a polygon Q if Q can b e
tiled by polygons similar t o P. T h e angles in each triangle in Figure
10 are π / 6 , π / 6 , and 2 π / 3 .
Exercise 16.
(a) Verify that Figure 10 is indeed a tiling of the 1 by y/3 rectangle.
(b) Prove that the π / 6 , π / 6 , 2 π / 3 triangle does not tile a square.
Laczkovich went on to show that there are only three triangles

that are not right triangles that tile the square. Their angles are
( π / 8 , π / 4 , 5 π / 8 ) , ( π / 4 , π / 3 , 5 π / 1 2 ) , and ( π / 1 2 , π / 4 , 2 π / 3 ) . T h e s e
and the triangle used in Figure 10 are the only triangles other than
right triangles that tile some rectangle.
Exercise 17. Show that three 15°-75°-90° triangles can tile a 1 by

4 rectangle, hence twelve of them can tile a square.
Laczkovich discovered a good deal m o r e about tilings by sim-

ilar triangles, and his results presented in this chapter may tempt
the reader to look at his paper. Since it appeared, h e has also con-
sidered the question, "Which pairs of triangles 7 \ and T have the2
property that similar copies of 1 \ tile T ." Working with Szekeres,

2
h e also investigated tilings of a square by a finite n u m b e r of rectan-

gles similar to a given rectangle R. l b describe one of their results,
let the ratio of the sides of R be r. They proved that a square can be
tiled by a finite n u m b e r of rectangles similar to R if and only if r is an
algebraic n u m b e r such that the real part of each of its conjugates is
positive. Using this result, they then produced a right triangle with
legs a and 6 such that similar copies of it tile the square, but rect-
angles similar to the o n e with sides a and b d o not. ( l b b e specific,
b/a is the real root of χ + χ - 1 = 0.)
3
3. Retrospective
With this chapter we complete our sampling of the algebra of tiling.
(The next chapter is the proof of Redei's T h e o r e m . )
As we look back over these chapters, we can see how t h e inter-
play of question and answer, followed by new questions, breathes
life into mathematics. In o u r case the questions came from within
mathematics, in particular, from geometry. But they can c o m e from
any discipline, such as physics, economics, or computer science.
Minkowski's question about cube tilings led to a question about
finite abelian groups! Hajos's answer, in turn, stimulated R e d e i to
ask, "Must the sets in the factorization be cyclic, or is there a more
general theorem, where the sets are arbitrary but of prime o r d e r s ? "
His t h e o r e m then gave us an insight into the tilings of space by clus-
ters that consist of a prime n u m b e r of cubes.
T h e question, "Which semicrosses or crosses tile space?,"

quickly suggested looking at splittings of finite abelian groups.
These splittings raised questions about exact sequences and cyclic
groups, especially groups of prime orders. We were left with m o r e
open questions than faced us when we started.
Packing and covering by the semicross and cross raised dif-
ferent questions, including further questions about finite abelian
groups, especially cyclic groups. T h e presence of such questions re-
minds us that even though we have a structure t h e o r e m for finite
abelian groups and know that each one is a product of cyclic groups
of prime power orders, they still keep many of their secrets.
In spite of a variety of results about cutting polygons into
triangles of equal areas, we were left with unanswered questions
as primitive as the ones we started with. For instance, we d o not
know whether the quadrilateral with vertices (0,0), (1,0), (0,1),
( \ / 3 / 2 , N/3/2) can be cut into an odd n u m b e r of triangles of equal
areas.
Tiling a polygon by triangles with prescribed angles led to alge-
braic questions completely different from those in the earlier chap-
ters. Though we met some surprising tilings, we are left with t h e
general question of when similar copies of one polygon tile another
polygon.
Clearly, we d o not need to venture into outer space o r the al-
most inaccessible regions of the A m a z o n forest t o enter territories
that have never b e e n explored. Mathematics still offers a cornu-
copia of challenges to even the most daring spirits, to anyone who
is compelled to enter worlds where no o n e has ever trod.
References
1. H. Eves, A Survey of Geometry, Allyn and Bacon, 1972.
2. M. Laczkovich, Tilings of polygons with similar triangles, Combina-
torica 10 (1990), 281-306.
Chapter 7
Redei's Theorem
A s a young mathematician, G. Hajos prepared a P h . D . thesis on

certain determinant identities. T h e chairman of his doctoral com-
mittee, L. Fejer, whose n a m e is closely associated with Fourier anal-
ysis, feeling that the result did n o t match the outstanding talent
of the candidate, rejected the thesis. This is why Hajos turned to
Minkowski's famous unsolved conjecture.
In 1938 Hajos formulated the problem in terms of factoriza-
tions of groups and, making use of this reformulation, refuted
Furtwängler's conjecture about multiple cube tilings, described in
C h a p t e r 1. This time his thesis m e t Fejer's legendary high standards.
Almost everyone, on first meeting t h e group theoretical equiv-
alent of Minkowski's conjecture, tends to think that the solution of
the problem should be immediate. So did Hajos. However, it took
him three years t o settle the conjecture. Looking back years later, he
said that the problem had been extremely deceiving. It had offered
many ways of attack but all but o n e led nowhere. Thinking about
the problem almost constantly, h e was able to pose it in many dif-
ferent versions. A s he said, " W h e n I had to walk u p to the 5th floor
I might m a k e u p my mind to find a new version on the way."
Eventually h e succeeded, obtaining his beautiful proof in 1941.
It is so algebraic that there is no discernible connection between its
lemmas and the geometry of the conjecture. This was why Hajos
was reluctant to talk about his proof. "Yes, I had a proof" he used
to say, "but I couldn't see what was really happening."
155
It seems that he eventually found a m o r e satisfactory view of

his proof, since some 30 years later, in 1970, he announced a sem-
inar on Minkowski's conjecture at Eötvös University in Budapest.
Unfortunately, Hajos died, at the age of fifty nine, in 1971 before
the seminar was to be held.
Hajos's proof is based on a d e e p investigation of the zero divi-
sors in a group ring Z(G). L. Redei wanted to prove the t h e o r e m us-
ing only group theoretical concepts, without referring to the group
ring, since the theorem itself is formulated in terms of groups. After
an extensive analysis he eventually found its group theoretical core
and discovered another approach to the problem: the technique of
replacing factors in a factorization. His proof, which does n o t rely
on the group ring, led to a broad generalization of Hajos's theorem.
This breakthrough, accomplished in 1965, is ample reward for his
25-year-long search.
Redei [3] proved the following theorem.
Theorem 1. Let G be a finite abelian group and Αχ, A , • • •, A„ sub-

2
sets ofG such that each contains the identity element, each has a prime
number of elements, and G = ΑχΑ · · · A is a factorization of G.
2 n
Then at least one of the factors A, tea subgroup ofG.
Since Hajos's t h e o r e m is equivalent t o its special case when

the cyclic factors are of prime cardinalities, Redei's t h e o r e m indeed
generalizes the Hajos-Minkowski theorem.
T h e proof we present in this chapter [1] is a much simplified
version of Redei's proof. Though he did not m a k e use of group
rings, we will, but only briefly.
W h e n each factor in a factorization contains the identity ele-

ment, we call the factorization normalized. If we omitted this con-
dition, then the t h e o r e m would state that at least o n e of the factors
A, is a coset of a subgroup of G.
T h e argument begins by establishing Redei's t h e o r e m in the
special case that G is a p-group and the factors are cyclic subsets.
(This is a special case of Hajos's theorem.)
Redei's Theorem 157
T h e n we establish Redei's theorem for cyclic p-groups, using

the irreducibility (over the field of rationale) of the cyclotomic poly-
nomials whose roots are the primitive ( p ) t h roots of unity.
n
A much more involved proof then treats the noncyclic group

of order p . A short induction then establishes Redei's t h e o r e m for
2
any finite abelian p-group.

Finally, another induction yields the proof for the general case
of any finite abelian group.
Critical to the arguments are replacement principles which
permit us, under certain conditions, to replace a factor in any fac-
torization of a group by a related subset of G. A s we proceed we
will develop these principles, all of which are special cases of o n e
general principle.
Exercises 1 a n d 2 show that t h e condition that the factors are
of prime cardinalities cannot be removed from Rodei's hypotheses.
Exercise 1. Let G be t h e cyclic group of order 8 with generator u.

(a) Show that neither A = {e, u, u , u } nor Β = {e, u } is a sub-
4 5 2
group of G.
(b) Verify that G = AB is a factorization of G.
Exercise 2. Let G b e the direct product of cyclic groups of order

8 and 2 a n d let u and ν a basis for G such that u — v = e.
8 2
(a) Show that neither A — {e, u, v, uv} n o r Β = {e, u , u , u i>} is

2 4 6
a subgroup of G.
(b) Verify that G = AB is a factorization of G.
1. Hajos's theorem for p-groups

O u r proof of Redei's theorem requires a special case of Hajos's
theorem, namely the case when G is a p-group. T h e proof depends
on a theorem that allows us to replace a factor in a factorization by
another subset.
If A and A' are subsets of a finite abelian group G such that for
every subset Β of G, if G = AB is a factorization, then G = A'Β
is also a factorization, we say that A is replaceable by A'. In this
section we need only a fairly simple replacement principle. Later

we develop m o r e general replacement principles..
Lemma 1. (First replacement principle) Let G = AB be a nor-

malized factorization of the finite abelian group G. Suppose that A =
{e, a, a ,...,
2
a } , where ρ is a prime. Let k be an integer relatively
p - 1
prime topand A' = {e, a , a , a } . Then G = A'Β is also

fc 2 k ( p _ 1 ) f c
a factorization of G.
Proof Since G = AB is a factorization, the sets
eB = B,aB,a B,...,a - B 2 p 1
(1)
form a partition of G. Thus aG, which is G again, is partitioned by

the sets
aB, a 2
B , a p
B ,
from which we conclude that a B = B. This implies that for any p
integer m (negative or positive), a B equals a'B, where 0 < j < m
p—\,j = m (modp).
Now consider A'B, which is the union of the sets
eB = B,a B,a B,...,a^-^ B.

k 2k k
(2)
Since fc and ρ are relatively prime the sets (1) are a permutation of
the sets (2). Thus G = A'B is a factorization, proving the lemma.
•
Exercise 3. Show that if fc and ρ are not relatively prime then A
cannot necessarily b e replaced by A'. (Hint: Use a factorization of
a group of o r d e r 4.)
Let G = Ai • • • A be a factorization of the finite abelian p-

n
group G, where A = {e, a a ,...,t a } . Any s of the elements

if
2 p _ 1
ai,...,an generate a subgroup of order at least p" and the η ele-

ments ο ι , α , . . . , α generate a subgroup of order exactly p , that
2 η
n
is, G itself.
Exercise 4. Verify the preceding observations.

Redei's Theorem 159
We introduce some notations. Let g be an element of the fi-

nite abelian group G. T h e o r d e r of g will b e denoted by |^| and
the subgroup generated by g will b e denoted (g). Let A b e a sub-
set of G. T h e cardinality of A and the subgroup spanned by A will
be denoted by \A\ and (A) respectively. Let A = { o i , . . . , a } . If n
G = Ai ••• A , where A = { e , a < , a , . . . , a f } , then | { ß ) | > p l l

n {
2 _ 1 ß
for each Β c A. This property of the factorization has a remark-

able consequence described in L e m m a 2, which is the key t o Hajos's
theorem for p-groups. Its proof makes use of the next exercise.
Exercise 5. Let G b e an abelian group and Η a p r o p e r subgroup

of G. Assume that there are factorizations Η = C\C -C and 2 T
G/H = D i £ > 2 · · · £ > , · For each i, 1 < i < s let D* b e a set of t
representatives of the cosets of Η in Ό . Show that χ
G = C C -C D{D* -D*
l 2 r 2 l)
is a factorization of G.
Lemma 2. Let Abe a subset of an abelian p-group G such that

\(A)\ = p W and \(B)\> p\ \forall B
Be A Then for each aeA
there exists a power of p, say s(a), such that
(A) = ft {e, α <°), a * W , . . . , a ^ - D ' W }

β 2
a£A
is a factorization of (A) and at least one of the factors is a subgroup of

G.
Proof T h e lemma holds when |i4| = 1 . This suggests an induction

o n η = |A|. For a given n, h(A) = Υ[ \a\ > p , with equality
αζΑ
n
only when \a\ = ρ for each α e A. In this case, also, the lemma
holds. So for a given η we will use an induction o n h(A).
If for each nonempty subset Β c Α, Β φ A, > pl l fl
replace o n e element of A by its pth power to get the set A'. T h e

conditions of the lemma are satisfied for A' and we have h(A') <
h(A). Noting that (Α') = (A), by induction o n h(A) we see that the
lemma holds for A. Now assume, on the other hand, that t h e r e is a
nonempty proper subset Β of A such that \{B)\ = p l L Clearly, Β B
satisfies the conditions of the lemma and \B\ < \ A |. By the inductive
assumption on the cardinality of A, (B) has a desired factorization
(B>= Π ί β , ο ^ , ^ , . . . , ^ - 1
^ } ,
b€B
where s(b) is a power of p. Consider the factor group G' = (A)/(B)

and t h e set of cosets A' = {a(B) : α e A \ B}. A straightforward
computation shows that \{A')\ = ρ ' Ι and that \(B')\ > p ' for
| Λ 1 0 1
all B' C A'. By induction on t h e cardinality of A, we obtain a fac-

torization of (Α') of the type in t h e statement of the lemma. Using
the factorizations of (B) a n d G' = (A)/(B), we obtain t h e desired
factorization of (A). •
Exercise 6. Verify L e m m a 2 when | A\ = 1.
Exercise 7. Verify L e m m a 2 when \a\ = ρ for each α € A.
Exercise 8. Fill in the details in t h e preceding proof.
With t h e aid of Lemmas 1 and 2 we now prove Hajos's t h e o r e m

for p-groups.
Lemma 3. Let G be a finite abelian p-group, and G = A\A% · • • A n
be a factorization, where Ai = { e , Oj, α , . . . , a ~ }. Then at least one

2 p 1
of the Aiisa subgroup ofG.
Proof. Let
G = A --A 1 n (3)
be a factorization of the finite abelian p-group, where
Ai = {e,ai,a ,...,a 2 p _ 1
}.
L e m m a 2 is applicable to A = {αχ,... , a } . S o f o r e a c h i , 1 < i < η

n
there is a power of p , say s(i), and a subset
i; = { ^ : ( i )
.^ ( i )
-^! " p I W i )
}
Redei's Theorem 161
such that G = A\ · · · A' is a factorization of G and at least o n e of

n
t h e factors A\ is a subgroup of G.
lis(i) = 1 for each i, 1 1. Since s ( i ) = 1 for each i, m + 1 
where 0 < t{i) < ρ - 1. So
e = a f ... ^ X W - i e S i 1
' · · · <(">. (4)
Now s ( i ) i ( i ) - 1 is relatively prime t o ρ for each i, 1 < i < m .

So by L e m m a 1, Aj can b e replaced by
A* - L N «W«(<)-l 2(»(i)t(T)-l) „(p-l)(a(i)t(i)-l)\

Λ
ί - I I °i
E
, i
a
, · · · Ι T
U
J
in t h e factorization (3) t o get t h e factorization
G - Aj · · • A A -)-i · · · A .
m m n
E q u a t i o n (4) violates this factorization unless m = 0. This com-

pletes t h e proof. •
2. Some examples and the general replacement

principle
To m a k e the proof of Redei's t h e o r e m m o r e accessible we pause t o
provide examples to introduce some of t h e ideas that appear in it.
Exercise 9. Show that if G is a 2-group, then Redei's t h e o r e m is

a consequence of Hajos's theorem, in t h e special case described in
L e m m a 3.
We introduce the notion of type of a finite abelian group. If G

is a direct product of cyclic groups of order m\,..., m , we say that t
G is of type ( m i , . . . , m ) . A group may have different types. For

t
instance the cyclic group of order 6 is of types (6), (2,3), and (3,2).
(If we insist that each rm is a prime power, the type of a group would
b e unique u p to order.)
Consider groups G of the types (9) or (3,3). Take the first case,
when G is the cyclic group of order 9 with generator g. Suppose
that G = A\A is a factorization of G, where Ai = {e, g , g } and
2
a b
Μ = {e,g ,g }.
c
H e r e we may assume that 1 < a,b,c,d
d
< 8. If
{a, b} or {c, d} is the set { 3 , 6 } , then A\ or A is a subgroup of G. 2
Exercise 10. How many choices d o we have for a, b, c, d to verify

R6dei's theorem for the cycle group of order 9 by b r u t e force?
T h e reader who attempts to verify R6dei's t h e o r e m in the style

of Exercise 10 will appreciate the n e e d for some devices to simplify
the bookkeeping. T h e basic law of exponents, x x = x , sug- a b a + b
gests that we introduce polynomials. (This idea goes back t o Euler

w h o used formal power series Σ ^ 1 f(n)x to record a function / 0
n
defined on the nonnegative integers.)

T h e factorization G = A\A 2 can be expressed in terms of the
product
( l + x + x ) ( l + x + x ).
a b c d
(5)
W h e n you expand (5), you will get a polynomial P(x) with nine
terms, each of which has coefficient 1. Furthermore, the nine expo-
nents are distinct modulo 9. If an exponent is larger than 8, we can
reduce it as follows. Say that x appears in P(x). Write 14 as 1-9+5.
1 4
Then x u
= x x , hence x - x = (x - l ) x which shows that
9 5 14 5 9 5
x is congruent to x modulo χ - 1 . This type of reduction implies

1 4 5 9
that there is a polynomial Q(x) with integer coefficients such that
(l + x° + x ) ( l + x + x ) = l + x + x + --- + x + Q ( x ) ( x - l ) .
b c d 2 8 9
R a t h e r than deal with this identity let us replace χ by ρ = ε , a 2 π < / 9
primitive 9th root of 1. We then obtain an equation in the field of

Redei's Theorem 163
complex numbers,
( l + p + p ) ( l + p + p ) = 0.
o b c d
(6)
A t least o n e of the factors in (6) must be 0. Assume, say, that 1 +
p + p = 0. Drawing three ninth roots of unity 1, p , and p in t h e
a b a b
complex plane we can convince ourselves that {a, b] = {3,6} is t h e

only case when their sum is 0.
Looking back on this argument, we see that we could have
omitted any reference to polynomials, a n d gone directly from t h e
factorization G = AiA to the equation
2
(1 + p + p )(l + p + p ) = l+p
a b c d
+ p + --- + p ,
2 s
(7)
In fact, (7) holds for any of the nine 9th roots of 1. For instance,
replacing ρ by 1, we obtain the equation 3 - 3 = 9, which comes as
n o surprise since the product of t h e orders of the factors equals t h e
order of t h e group.
Even though the original problem concerns groups, which have
only o n e composition, t h e solution exploits a ring of polynomials
or the complex field, which are richer structures, possessing both a
multiplication a n d an addition. T h e proof of Rodei's theorem uses
both structures as well as characters of a finite abelian group.
A function χ from the abelian group G to the complex n u m b e r s
of modulus 1 is called a character of G if x(gh) = x(g)x(h) for each
g,h€ G. T h e character for which x(g) — 1 for each g e G is called
the principal character of G. For any subset A of G define χ(Α) to
be the sum of the numbers χ(α) for all α e A. In particular, if χ
is the principal character of G, χ(Α) = \A\. (Appendix Β develops
the basic properties of characters.)
Exercise 11. Let G = A\ A be a factorization a n d χ be a character

2
of G. Prove that (G) =X χ(Α )χ{Α ). ι 3
We repeat the previous argument using the terminology of

characters.
Let ρ be a primitive 9th root of unity. Define the character χ
of G by setting x(g) = ρ and therefore x(g') = p\ 0 < i < 8.
Applying this character to the factorization G = ΑχΑ , 2 we obtain:
X(G) = x ( e ) + x(g) + x(g ) + ••• + 2

x(g ) e
= 1+ ρ + ρ + ··· + ρ = 0 2 8
χ ( Α χ ) = (e) + x(g ) + (g») = l+p

X
a
X
a
+ p»,
χ(Α )
2 = (e) + (g )
X X
c
+ (g )
X
d
= 1+ o + p, c d
and
X(G) = x(A )x(A ). 1 2
From 0 = x(G) = χ(Α )χ{Α ) it follows that χ ( Α χ ) = 0 or

χ 2
χ(Α ) = 0. For t h e sake of definiteness suppose that χ ( Α χ ) = 0.

2
H e n c e {a, b} = {3,6}.
Exercise 12. Show that if G is t h e cyclic group of order 3 a n d n
G = Αι · • · A is a n o r m e d factorization of G, where \ A\\ = •·• =

n
\A \ — 3, then o n e of the factors Α χ , . . . , A is a subgroup of G.

n n
In the next few examples w e will develop this tool further.

First we use characters to establish Redei's t h e o r e m for any
cyclic p-group.
Let ρ b e a prime a n d let G be the cyclic group of order p with n
generator g. Suppose that G = A\ • · · A is a n o r m e d factorization n
of G, w h e r e |AjJ = · · = | A | = p . Consider a primitive ( p ) t h

n
n
root of unity ρ and define the character χ of G by setting x(g) = p.

Now as before there is a factor, say
A = 1 {e,g \g \...,g '- }

a a a i
for which
0 = χ(Α,) = x(e) + x(g >) + (g > ) + ••• + {g *^)
a
X
a
X
a
=
= l + p ai
+p" 3
+ ··· + ρ *- = 0 . 0 1
We may assume that 1 < α< < p - 1 for each i,l<i<p—l n
We want to show that A is a subgroup of G, that is, t h e exponents

1
0, αχ, θ 2 , . . . , θρ_χ form a permutation of the numbers 0, p , n _ 1

Redei's Theorem 165
2 p ~ , . . . , ( p - l ) p . l b d o this consider the polynomial

n 1 n - 1
Αχ(χ)
associated with Ai, defined by
Ai(x) = 1+ χ αι
+ x° + ••• +a
χ ·" .
α 1
Next we need t h e fact that the ( p ) t h cyclotomic polynomial

n
F(x) = 1 + χ""- 1
+ ζ *"" + · · · +
2 1
xir-V'"- 1
is irreducible over the field of rational numbers [2, p.46]. (Appendix

D discusses the cyclotomic polynomials.) N o t e that F(x) a n d Αι (x)
have a common root, namely p. T h u s F(x) divides A\(x) in the ring
of polynomials with rational coefficients, a n d since F(x) is monic, in
the ring of polynomials with integer coefficients. T h e next exercise
completes the argument.
Exercise 13. Show that from t h e fact that F(x) divides Αχ(χ) it
follows that 0, α ϊ , a , . . . , α _ ι is a permutation of 0, p
2 ρ , n _ 1
2 p " , . . . , (p - l ) p . (Hint: Let Ai(x) = F(x)Q(x)

n 1 n _ 1
and first
show that the degree of Q(x) is less than p . T h e n show that Q(x)
n _ 1
must b e 1.)
We summarize this argument in L e m m a 4, which will play a

role in the proof of R i d e i ' s theorem.
Lemma 4. Let G be the cyclic group of order p where ρ is a prime n
and let G — A\A - • · A be a normed factorization. Then at least one

2 n
set Aiisa subgroup ofG.
T h e most important result about vanishing sums of roots of

unity we use in connection with factoring p-groups generalizes a
previous observation.
Lemma 5. Let ρ be a prime and pa(p )th root of unity. Ifp -\

n
h ai
ρ ° · = 0, where οχ = 0 and 1 < s < p, then s = ρ and p ,... ,p " is ai a
a permutation ofl, Θ,..., 0 , where θ = ε / .

P _ 1 2 π ί ρ
Exercise 14. Prove the lemma, using the technique in Exercise 13.
We have seen that characters a r e of use in t h e study of factor-

ization because if
G = ΑιΑ ···Α„,
2
then
x(G) = x(A )x(A ).-- (A ).

1 2 X n
But characters convey much more information, and we will use t h e

following fact: If A and Β a r e subsets of t h e finite abelian group G
and if χ(Α) = χ(Β) for each character χ of G then A = B. (That
χ(Α) = χ(Β) for t h e principal character tells us that \A\ = \B\.)
A n example will illustrate this observation. Let G be t h e cyclic
group of o r d e r four. D e n o t e a generator of G by g and a primitive
fourth root of unity by p, which could be chosen to b e i. T h e four
characters of G, χ ι , χ , χ , χ are given in Table 1.
2 3 4
e 9 92
9 3
Xi 1 1 1 1
X2 1 Ρ Ρ 2
ρ3
X3 1 Ρ2
1 Ρ 2
Xi 1 Ρ3
Ρ 2
ρ
TABLE 1
Exercise 15. Show that t h e columns of Table 1 a r e linearly inde-

pendent. (Appendix Β shows that t h e columns of such a character
table are always linearly independent.)
Let us check t h e claim that if χ(Α) = χ(Β) for each character

χ of G, then Λ = Β. F r o m x i ( A ) = χι{Β) it follows that |A| = \B\.
If for example |A| = \B\ = 2 and A = {αϊ, 02}, Β = {b\, b } then 2
the fact that χ(Α) — χ(Β) for each character χ of G is equivalent

Redei's Theorem 167
to the following four equations
Χι(αι) + χ ι ( ο ) 2 Xi(bi)+Xi(6 ) 2
Xa(ai) + ^2(02) X2(i»i) +Xa(i>2)
Χ3(αι) + X3(<12) X3(i>l) + X 3 ( & 2 )
Χ4(αι) + χ ( ο ) 4 2 X4(bi)+X4(6 ). 2
These equations imply that a linear combination of two columns of

Table 1 is equal to a linear combination of two columns of Table 1.
Since t h e columns of Table 1 are linearly i n d e p e n d e n t over t h e field
of complex numbers we conclude that A = B.
So far we have extended the domain of a character from ele-
ments of a group G to subsets of G. Now it is necessary t o extend
t h e domain further.
Let G = {gx, g ,..., g } be a finite abelian group and consider
2 n
the set of expressions of t h e form
ziSi + z g 2 2 + ••• + z g,
n (8)
w h e r e z e Z. (Such an expression is shorthand for the function

{
that assigns to & the integer Zi. See Appendix C.) We add two such
expressions coordinate by coordinate:
(2101 + z g 2 2 + · · · + zg)
n n + (iiffi + t g 2 2 + h tg)
n n
= (zi + h)gi + {z + t )g
2 2 2 + ··· + (ζ„ + t )g .
n n
T h e product of two such expressions is defined using the multipli-

cation in G:
This is similar to multiplication of polynomials.

T h e set of expressions (8), together with the addition and mul-
tiplication just introduced, is called the group ring of G; and is d e -
noted Z ( G ) .
A subset A in G can be identified with the element in the group

ring whose coefficients are 0 or 1: 1 at elements in A and 0 at ele-
ments in G \ A.
Exercise 16. Let G be the cyclic group of order 5 with generator

u and identity element e. In Z(G) compute
(a) (e - u)(e + u + u + u + u ) ,
2 s 4
(b) (e + 2u + u ) ( 3 u - u ) .
2 2 4
Now let A and Β b e subsets of a finite abelian group G. Con-

sidering A and Β as elements of the group ring Z(G), we may com-
p u t e their product AB, obtaining an element Y^, n(g)g. T h e co-
geG
efficient n(g) records the number of ordered pairs (a, b) such that
ab = g. That G = AB is a factorization is equivalent to the as-
sertion that n(g) = 1 for every g 6 G. T h a t the symbol AB has
two meanings, namely a product in the group ring and a factoring
of G will cause n o difficulty. T h e context will make it clear which is
intended.
Exercise 17. Let ζ = Σ i9% * * = Σ Ugi be two elements in

z a n c
Z(G) and χ a character on G. Prove that

(a) x(zt) = x(z)x(t),
(b) x(z + t) = (z) + x(t).
x
Exercise 18.
(a) Let ζ e Z(G) have the property that χ{ζ) = 0 for all characters
χ o n G. Prove that ζ = 0, the zero element of the ring Z(G).
(b) Prove that if ζ and t are in Z(G) and χ(ζ) = χ(ί) for all char-
acters χ on G, then ζ = t.
As Exercises 17 and 18 show, the characters of a group G con-

vey a great deal of information. In fact, as we will now show, they
even can tell us whether two subsets of G, A and B, provide a fac-
torization of G.
First of all, if χ is a nonprincipal character, x(G) = 0, as is
shown in Appendix B. If χ is the principal character of G, x(G) =
\G\. Now assume that G = AB is a factorization. T h e n if χ is a
Redei's Theorem 169
nonprincipal character of G, at least o n e of the numbers χ(Α) a n d

χ(Β) is 0. (If χ is the principal character of G, the equation \(G) —
χ(Α)χ(Β) says only that |G| = \A\\B\.)
Now consider any subsets A and Β of G. T h e n G = AB is
a factorization if and only if G = AB in Z(G). But G = AB in
Z(G) if and only if x(G) = χ(ΑΒ) for all characters χ of G. This is
equivalent to t h e assertion that x(G) = χ( Α)χ(Β) for all characters
of G. A n d this reduces t o t h e statement "G = AB is a factorization
if and only if \G\ = \A\\B\ and for every nonprincipal character χ of
G at least one ofx(A) and χ(Β) is 0." This observation is t h e basis
of the following replacement principle, which will be used several
times in t h e proof of Redei's theorem.
General Replacement Principle. Let A and A' be subsets ofG such

that \A\ = \A'\ and ifx(A) = 0 it follows that χ{Α') = 0. Then A
can be replaced by A' in all factorizations ofG; that is,ifG = AB is
a factorization, soisG = A'B.
Exercise 19. Prove the General Replacement Principle.
Exercise 20. Show that t h e first replacement principle ( L e m m a 1)

is a consequence of the General Replacement Principle.
Though we used t h e group ring t o develop t h e General R e -
placement Principle, the group ring does not a p p e a r in its state-
ment. We could have avoided mentioning the group ring, but only
by using some clumsy bookkeeping. We would have to define t h e
"product" AB of two subsets A a n d Β as a set in which each ele-
m e n t is assigned a weight, namely t h e n u m b e r of times it is repre-
sented in t h e form ab, a € A, b e B. T h e n we would show that
χ(ΑΒ) = χ(Α)χ(Β). It is far m o r e natural to introduce t h e ring
Z(G) a n d observe that each character provides a homomorphism
from it t o t h e ring of complex numbers. (Moreover, when studying
multiple tilings of R , t h e group ring is essential, for t h e elements
n
in it with negative coeffients and zero divisors play a role.)

As o u r first application of t h e General Replacement Principle
we have t h e following replacement principle of use in p-groups.
Lemma 6. (Second Replacement Principle) Let A be a subset of

the finite abelian p-group G such that \A\ = ρ and e e A Let k be
relatively prime to p. Then A can be replaced by
A' = {e,a ,a ,...,a^ }

k 2k k
in each factorization of G for each a € A \ {e}.
Exercise 21. Using L e m m a 5 and the General Replacement Prin-

ciple, prove L e m m a 6.
Problem 1. T h e proof of L e m m a 6 involves the use of characters,

though characters are not mentioned in it. Obtain a direct proof,
o n e that does not use characters or complex numbers.
Lemma 7. (Third replacement principle.) Let A have ρ elements,

where ρ is prime. Assume that each element in A has order lor μ Then
A can be replaced by a cyclic subset in any factorization.
Exercise 22. Prove L e m m a 7.
Let us illustrate the use of replacement techniques in the fac-

torings of a group of type (3,3), the non-cyclic group G of o r d e r 9.
Suppose that G = A\A is a n o r m e d factorization of G, where
2
I Αχ I = I A | = 3. We will use characters to show that Αχ or A

a 2
is a subgroup of G. T h e r e are simpler ways to prove this b u t our

purpose is t o shed light on the methods we will use later.
Let ρ b e a primitive third root of unity and h and k a basis of
G. E a c h character χ of G is determined by χ(η) and x(k). Both
x(h) and \(fc) can b e assigned only three possible values, 1, p, or
p . For the sake of concreteness we display the nine characters of
2
G in Table 2.
For each nonprincipal character χ of G we have χ(Αχ) = 0 or
χ ( Α ) = 0. O n the other hand, if Αχ and A are subsets of G such
2 2
that |J4I| = IAa| = 3 and for each nonprincipal character χ of G,

χ(Αχ) = 0οτχ(Α ) 2 = 0, then we have the factorization G = ΑχΑ . 2
Ridel's Theorem 171
e h h 2
k hk /i fc
2
fc 2
hk 2
hk
2 2
1 1 1 1 1 1 1 1 1
X2 1 Ρ ρ 2
1 Ρ 1 Ρ Ρ 2
X3 1 Ρ 2
Ρ 1 Ρ 2
Ρ 1 Ρ 2
Ρ
X4 1 1 1 Ρ Ρ Ρ Ρ 2
Ρ 2
Ρ 2
X5 1 Ρ ρ 2
Ρ 2
1 Ρ2
1 Ρ
X6 1 Ρ 2
Ρ 1 ρ 2
Ρ 2
Ρ 1
X7 1 1 1 ρ2
ρ 2
Ρ 2
Ρ Ρ Ρ
X8 1 Ρ ρ2
Ρ 2
1 Ρ Ρ Ρ2
1
X9 1 Ρ 2
Ρ Ρ 2
1 Ρ 1 Ρ 2
TABLE 2
Now let
Αι = {ε,α,η,α,ιζ} and Λ = {e, α ι , α } .

2 2 2 2
If χ is a character of G for which χ(.4χ) = 0, then χ ( β ) , χ ( α π ) ,

χ ( α χ ) must be a permutation of Ι , ρ , ρ . It follows that if χ(>1χ) = 0
2
2
then χ(Η) = 0, where Η = { β , α χ ι , α ^ } . Thus G = HA is also 2
a factorization of G. Η is a subgroup of G and so A is a com- 2
plete set of representatives of the cosets of Η ("modulo H" for

short). Similarly, G = ΑχΚ is also a factorization of G, where
Ä" = {e, α χ, a } . Κ is a subgroup of G and so Λχ is a complete set
2 2 1
of representatives modulo K. N o t e that G = Η Κ is also a factoriza-

tion. This implies that αχχ and a form a basis of G. For simplicity
2 1
we d e n o t e αχχ = Λ and α χ = A;. Thus

2
A\ = {e,h,h x} 2
and A = {e, 2 k,k y},
2
where ι e if, y e i i , i / = {e, h, h } and Ä" = {e, fc, fc }. If χ = e

2 2
or y = e, then A\ or A is a subgroup of G.
2
Exercise 23. Inspecting all t h e nine possible choices for χ and y,

verify that Αχ or A is a subgroup of G.
2
Characters of G simplify t h e bookkeeping again. Suppose that

χ φ e and y φ e. Pick a character χ of G for which χ(Αχ) = 0.
We know that for this character χ(Η) = 0. H e n c e x(h?x) = χ ( / ι ) , 2
that is, χ ( χ ) = 1. Since χ generates K, x(k) = 1. So there are at

most three characters χ such that χ(Αι) = 0. Similarly, there a r e
at most three characters χ such that χ(Α ) — 0. T h u s there are at 2
most 3 + 3 = 6 characters χ of G for which χ(ΑιΑ ) = 0. However, 2
this is not t h e case since there are 8 characters χ of G for which

χ(ΑχΑ )
2 = 0.
Let us look at o n e m o r e example before beginning the proof of
Redei's theorem. Let G b e of type (5,5). Suppose that G = A A is X 2
a n o r m e d factorization of G. As before, Αχ can b e replaced by t h e

subgroup Η = (h). H e n c e A is a complete set of representatives
2
modulo H. Similarly, A can be replaced by t h e subgroup Κ = (fc)

2
and so A i is a complete set of representatives modulo K. Since

G = Η Κ is also a factorization, h a n d k form a basis of G. T h u s
Αχ = {e, h, h x, h y, h z}
2 3 4
and A = {e, fc, fc u, fc i>, fc iu},
2
2 3 4
where x,y,z e Κ and u, v, w e H.
Exercise 24. How many choices d o we have for t h e elements x, y,

ζ a n d u, v, w in order to verify Redei's theorem by brute force? ( D o
not carry o u t t h e verification.)
Exercise 25. Let ρ be a prime and G be t h e group of type (p,p).

Suppose that G = A\A is a n o r m e d factorization, where | A i | =
2
IA 1 = p . Show that there is a basis, fc a n d h of G, such that

2
Αχ = {e,h,h k \...,h - k ^},

2 a p l a
A 2 = {e fc,fc /i V..,fc - /i "- }.

1
2 i p 1 b 1
R a t h e r than complete t h e case (5,5) we turn o u r attention to

the general case (ρ, ρ ) , ρ prime, which is far from trivial.
Ridel's Theorem 173
3. Redei's theorem for p-groups

With the tools w e have developed so far w e can prove Redei's the-
orem for p-groups. T h e proof proceeds in two stages. First we treat
the case when G is of type (p,p); then we prove the t h e o r e m for
any finite abelian p-group. Recall that we have already proved it for
groups of the type ( p ) , that is, for cyclic p-groups.
n
T h e next exercise establishes some facts about polynomials

whose coefficients lie in GF(p), the field with ρ elements. These
results will be used in the proof of L e m m a 8.
Exercise 26. Let E{x) = ΠΓ=ι( ~ <)> < GF{p). Let

:Ε α w n e r e a e
ci, C 2 , . . . , c , m < n, b e the distinct values occurring among the

m
ai's. Let F(x) = ΥΙΪΙ^χ - Ci). Show that

(a) F(x) divides E(x) and x - x. p
(b) E{x)/F{x) divides E'(x), t h e derivative of E(x).
Lemma 8. Ridei's theorem is true for groups of type (p,p), where ρ

is a prime.
Proof. Let G b e of type (p,p) a n d suppose that G = A\A is a 2
n o r m e d factorization of G, where |J4I| = \A \ — p. By Exercise 25 2
there is a basis k and hoiG such that
Αχ = {e,M fc V..,/i - fc »- },
2 a p 1 0 1
A2 = {e,k,k h ,...,k - h ''~ }.

2 b2 p 1 b 1
If ρ = 2, then both Ai and A are subgroups of G. From n o w on we

2
suppose that ρ is odd.

We wish to show that all t h e a / s a r e 0 modulo ρ o r that all
the bj's are 0 modulo p . For instance, in the first case, w e want t o
show that the polynomial (x - a ){x -as)---(x- 2 α _ ι ) , viewed as ρ
a polynomial over GF(p), is just the polynomial a ; . Introducing p _ 2
α = ai = 0, we want to show that Π?=ο ( ~ » ) simply x .

0
x a I S p
Tb begin the argument let ρ b e a primitive pth root of unity. For

each integer y,0<y<p-l define the (nonprincipal) character
χ on G by x (h) — p and x (k) = p~ ~ .
υ y
v
v Recalling that ρ is l v
odd, we see that x {Ai) y= 0 for at least (p 4 - 1 ) / 2 values of y o r

X (A )
y 2= 0 for at least (p + l ) / 2 values of y. We assume without
loss of generality that the first case occurs. From this we will deduce
that 0 2 = 0 3 = · · · = a -i = 0.
p
We have then
p-l
o = x {A ) v l = ,
£p -
i v a d l + y )
I=0
for at least (p + l ) / 2 values of y. Let Y be the set of these values of

y. Since χυ(Α\) = 0, the exponents iy - cn(l + y) form a complete
set of representatives modulo ρ for each y E Y.
Let CO, C I , . . . , C be the distinct values occurring among DO,
M
α ϊ , . . . , ap-i, taken modulo p. We define three polynomials with co-

efficients in GF(p):
p-l ρ
D(x,y) = Y[(x + iy - a,i(l + y)) = ^di{y)x\
i=0 i=0
p-l ρ
E(x)=D(x,0) = JJ(x-oO = ^ Β 4 Χ \
I=0 <=0
M
F(x) = [J(A; - ci),

I=0
and we wish to prove that E(x) = X , or equivalently, that E'(x) = 0.
P
We first show that many of the coefficients of E(x) are 0, by

examining D(x, y).
Let us write
D(x,y) =do(y) + d (y)x + x \-d (y)x .

p
p
Since D{x,y) = U jZl(x ~ (aj(l + y) ~ jv)), we have d {y) = 1

P
p
and each d -i(y), 0 < i < p, is an elementary symmetric function in

p
the ρ expressions a,j (1 + y) — jy. Let Si (zj ) denote the ith elementary
function of the expressions z0,..., zp-\. Then we have
d -i{y)
P = Si(Jy - a j ( l + y)).
Note that the degree of d -i(y) p is at most i — 1.
Reclei's Theorem 175
Now D(x, y) = —x + x for each y in Y. Thus for y £ Y we

p
have dp-i(y) = 0 for 1 < i < ρ — 2. Therefore for those values of i

the polynomial dp-i(y) has at least (p + l ) / 2 roots. Since the degree
of d -i(y) is at most i - 1 , we see that for 1 < i < (p+1)/2,
p d -i(y)
p
is the zero polynomial. That means that D(x, y) has the form
D{x, y) = d {y) + d (y)x + •••+ d _ (y)x -^

0 x (p l)/2
ip /2
+ x.
p
Thus
E(x) = e + e x + • • • + e _
0 x ( p 1 ) / 2 a; ( p _ 1 ) / 2
+ x,
p
where e* e GF(p). Moreover, since a0 = ax = 0 and are roots of

E(x), we may write
E(x) = e x + ••• + e _ x^-^

2
2
(p 1)/2
2
+ x".
All that remains is to show that e = e = • • • — eîy
2 = 0. To do
3 2
that, we show that E'(x) = 0.

Introduce G(x) = E(x) — (x —x), a polynomial of degree at most
p
(p - l ) / 2 . It is not the zero polynomial since it has a term of degree

1. Since E' has degree at most (p — 3 ) / 2 , the product G(x)E'(x) has
degree at most ρ — 2 or is the zero polynomial. We show that E{x),
which has degree p, divides G(x)E'(x).
By Exercise 26, E(x)/F(x) divides E'(x). Hence there is a
polynomial Q{x) such that
_ Q(x)E(x)
[
> ~ F(x) ·
Thus
G(x)E'{x) = <MQ( )E(X). x
Since F(x) divides x — x and E(x), it divides their difference, G(x).

p
Hence E(x) divides G(x)E'(x). Since the degree of G(x)E'(x) is

smaller than the degree of E(x), G(x)E'(x) is the zero polynomial,
hence E'(x) = 0. Therefore E(x) = x and the proof is complete. p
•
In order to prove Redei's theorem for any finite abelian p-

group it will be useful to note that if Redei's t h e o r e m is true, then
the following seemingly sharper L e m m a 9 also holds. T h e proof of
this lemma depends on the following exercise.
Exercise 27. Let G be a group and Ν a normal subgroup of G. Let

ß be a subset of G such that {bN : b e B} is a subgroup of G/N.
Then U ( , B { W V } is a subgroup of G.
€
Lemma 9. Let G = Αχ- • A bea n normalized factorization of the

finite abelian group G by subsets of prime orders. Assume Ridei's the-
orem is true. Then there is a rearrangement of the factors, Αι,..., An,
say Β i,..., B , such that all the subsets Βχ, Βχ B , • • •, Β χ B • • • B
n 2 2 n
are subgroups of G.
Exercise 28. Use Exercise 27 to prove L e m m a 9 for the cases η =

3 and η = 4.
We are now ready to obtain the main result in this section.
L e m m a 10. Redei's theorem holds for any finite abelian p-group.
Proof. Let G be a finite abelian p-group of order p a n d G = n
Αχ • • • A b e a normed factorization of G, where | A | = · · · =

n
I Ail — p. We wisn to prove that at least o n e A is a subgroup of

G. T h e case η = 1 is trivial. T h e case η = 2 is settled already, since
L e m m a 4 takes care of G of the type (p ) and L e m m a 8 the type
2
(p,p). We argue by induction, starting with the cases η = 1 and

η = 2, and consider η > 3.
By L e m m a 6 every factor A, can be replaced by a cyclic sub-
set. If each factor contains an element of order at least p , then 2
using them we can construct a factorization of G consisting of non-

subgroup cyclic subsets. This contradicts L e m m a 3. T h u s t h e r e ex-
ists a factor, say Αχ, whose nonidentity elements all have order
p. By L e m m a 7, Αχ can b e replaced by a subgroup H, which is
generated by a nonidentity element of Αχ, t o get t h e factorization
G = HA • · · A . Considering the factor group G/H we have the
2 n
Ridel's Theorem 177
factorization G/H = (A H)/H • • • (A H)/H. By the inductive as-

2 n
sumption and L e m m a 9 there is a permutation Βχ,...,Β of the η
factors H, A ,...,
2 A such that Βχ, ΒχΒ , •. • ,ΒχΒ · • • B are sub-
n 2 2 n
groups of G and Βχ — Η. To be concrete we will suppose that the

permutation B ,...,B of the factors A ,...,
2 n A is the identity 2 n
since this is only a matter of indexing the factors A*. Consider the
subgroup Κ = HA • • · A „ _ i . Since the identity element e is an el-
2
e m e n t of each factor, Η and each Aj is a subset of K. If Αχ c K,

then Κ = A\A • ·• A _ i is a factorization of K. (Note that both Κ
2 n
and A\A • · · A -\ have p

2 n elements.) By the inductive assump- n _ 1
tion, at least one of Αχ, A ,..., A -i is a subgroup of Κ and so of

2 n
G.
So suppose that Αχ is not a subset of K. T h e n replace Αχ in
the factorization of G by a subgroup L generated by an element of
Αχ that is not in K. Since L <£ K,we have Κ Π L = {e}.
F r o m the factorization G = Z/-A · · · A it follows, by applying 2 n
the induction assumption to the group G/L, there is a p e r m u t a -

tion C i , . . . , C of the factors L, A , . . . , A such that the subsets
n 2 n
Gi, C i C , . . . , C i C · · · C are subgroups of G and Ci = L. T h e r e

2 2 n
is an index j such that C — Aj and therefore L A , is a subgroup 2
of G. If j Φ η, then Κ Π LAj = A j , being the intersection of two

subgroups, is a subgroup of G. Therefore we may assume that LA n
is a subgroup of G.
Consider the intersection Κ Π LA . If it were just {e}, then G n
would contain the direct product of the two subgroups Κ and LA , n
a group whose order would b e p , which is greater than |G|. T h u s n + 1
| Α Γ Π £ Α „ | = p or p. 2
If \K Π Z A | = p , then I A c n hence L c K, which is a

2
n
contradiction. H e n c e \KnLA \ = p . Consider now two cases: LA n n
cyclic and LA of type (p, p).

n
Say that L A is cyclic. Then Κ Π LA is the unique subgroup

n n
of order ρ in L A „ , namely L. Since L <£ K, this case is ruled out.

Since LA is not cyclic, the nonidentity elements of A have
n n
order p. Consequently A„ can be replaced by a subgroup Μ gener-

ated by a nonidentity element of A„ to get the factorizations G =
Αι··· Α -ιΜ
η and G = HA • • • Α _ χ Μ . N o t e that G = KM is
2 η
also a factorization of G a n d so i f Π Μ = {e}.

From the factorization G = Ai • • • Α -χΜ it follows that thereη
is a permutation Dx, ...,D of the factors Αχ,..., Α ^χ, Μ such

n η
that D i = Μ and the subsets
D ,D D ,...,D D -D
1 l 2 l 2 n
are subgroups of G. T h e r e is an index j , 1 < j < η - 1 such that

D = Aj. H e n c e MAj is a subgroup of G.
2
If j Φ 1, then Aj c if. Recalling that i f Π Μ = {e}, we have

i f Π MAj = Aj and so Aj, being the intersection of two subgroups,
is a subgroup of G.
If j = 1, we have that Ν = Μ Αι is a subgroup of G.
Consider two cases: A c Ν and A <£. N. In the first case
n n
AiA n c N. Therefore, since both Ν and A A contain p ele- x n

2
ments, Ν = A\A . n By L e m m a 8 at least o n e of Αχ and A is a n
subgroup of N, hence of G.
T h e other case is A <f. N. T h e n A can b e replaced by a sub-
n n
group Τ such that ΤΠΝ = {e}. We then have G = AiA • · · A -iT. 2 n
Applying the induction to the group G/T shows that there is a sub-
group of G of the form TAj, where 1 < j < η - 1. Consider
i f (Ί TAj = HA · · • Α -ι Π TAj. A s argued before, if this inter-
2 η
section were just {e}, then G would contain a subgroup of p n+1
elements. T h e intersection cannot b e TAj since i f does not con-

tain T. T h u s the intersection is a subgroup of TAj with ρ elements.
If j φ 1, this intersection is Aj, establishing the theorem. If j = 1,
then both Τ Αχ and Ν = Μ Αχ are subgroups of G. Recalling that
ΤΠΝ = {e}, we conclude that Τ Αχ η Ν = Αχ, hence that Αχ is a
subgroup of G. This completes the proof. •
4. Vanishing sums of roots of unity

To prove Redei's t h e o r e m for non-p-groups we need another spe-
cial case of the G e n e r a l Replacement Principle. It depends o n an-
other property of vanishing sum of roots of unity, which is given in
L e m m a 11. T h e proof of this lemma uses the next exercise.
Redei's Theorem 179
Exercise 29. Show that if ρ is a primitive (r-s)th root of unity, where

r and s are relatively prime, then ρ is a product of a primitive r t h
and a primitive sth root of unity. (Hint: Since r a n d s are relatively
prime there are integers a and b such that 1 = ar + bs. H e n c e
p \ _ p ar+bs _ p ar b*j
p
Lemma 11. Let ρ be a primitive nth root of unity and let ρ be the least
prime factor of ru If ρ Η h ρ · = 0, αϊ = Oand 1 < s < p, then
αι α
s = ρ and there is a primitive pth root of unity θ such that p" ,..., p» 1 a
is a permutation of 1, Θ,..., 0 . P _ 1
Proof. We proceed by induction on n. T h e case η = p is settled e
in L e m m a 5.
Suppose that η — p r, where r is relatively prime t o p. Let
e
ρ = σ τ , where σ and τ are ( p ) t h and r t h primitive roots of unity

e
respectively. Now
S a
ο= Σ^ 0 <
= Σ ° σ < τ α <
·
t=l i=l
Divide p into a* to obtain the remainder bi such that 0 < bi <
e
p - 1. Let b[,..., b\ be the different numbers among

e
b\,...,b . s
Then
t=l i=l i=l
where oVs are nonempty sums of r t h roots of unity. Consider t h e

polynomial
t
A{x)^Y a x <. d i
h
(9)
t=l
Suppose that o n e of the coefficients a, equals 0. Divide it by one

of its terms to obtain a sum r + r + · · · + r *, where r is an
Cl Ca c
r t h root of unity and ci = 0. Note that k < s. Since this sum is 0,

by induction, k must be equal to the least prime factor of r , hence
k > p. T h u s s > k > p, contradicting the assumption that s < p.
We may therefore assume that n o n e of a<'s is zero and so A(x)

is not the zero polynomial. H e n c e 0 < degA(x) *- e1
is irreducible over t h e r t h cyclotomic field since r is relatively prime

to p . (For a proof see Appendix D.) N o t e that σ is a c o m m o n root
e
of A(x) a n d F(x). H e n c e A(x) is a multiple of F(x), that is, A(x) =

B(x)F(x) for a suitable polynomial B(x) with coefficients from the
r t h cyclotomic field.
Since t h e degree of F(x) is (p - l ) p , equation (9) tells us
e _ 1
that d e g ß ( x ) < pe _ 1
- 1. Therefore the nonzero terms of B(x)
occur among the nonzero terms of A(x). Since B(x) has at least
o n e nonzero term, A(x) has at least ρ terms. Thus ρ < t < a. O n
the other hand, p> a. H e n c e ρ = a. This completes the proof. •
Tb prove R i d e i ' s theorem for groups that are not p-groups we

need the concepts of p-part and p-part of an element of G for a
prime p . L e t Η b e t h e p-component of G a n d let Κ b e the comple-
mentary direct factor to U in G. E a c h g £ G is uniquely expressible
in t h e form g = hk, h £ H, k £ K. T h e elements h and k a r e t h e
p-part and t h e p-part of g a n d they will be denoted by g(p) a n d g(p)
respectively. T h u sg= g(p)g(p).
L e m m a 12. (Fourth replacement principle) Let A be α factor of G

such that \A\ = pis the least prime factor of\G\ and e € A Then A
can be replaced in any factorization of G by A' = {α(ρ)(δ(ρ))*( ) : α
a £ A} for any choice of integer exponents s(o).
Proof. Since A is a factor of G, t h e r e is a subset Β of G such that

G = AB is a factorization a n d e £ B. We first show that there is a
character of G that vanishes on A.
If there were n o such character, then for every nonprincipal
character χ, χ(Β) = 0. For the principal character χ of G, w e have
χ(Β) = \B\. Thus every character of G vanishes on t h e group ring
Redei's Theorem 181
element \B\G - \G\B, hence \B\G = \G\B. But this is impossible

since t h e coefficients of e on the two sides of the equation are dif-
ferent.
N o w let χ b e a character of G for which
aZA
T h e r e exists a minimal integer η such that each χ(α) is a power of a

fixed primitive n t h root of unity. Since η is a divisor of |G| the least
prime factor of η is at least p. T h u s by L e m m a 11 t h e r e is a primitive
pth root of unity θ such that {χ(α) : a e A} = {θ* : 0 < i < ρ - 1}.
This implies that \A'\ = ρ and that from χ(Α) = 0 it follows that
χ(Α') = 0. So, by the G e n e r a l R e p l a c e m e n t Principle, A can b e
replaced by A' in each factorization of G. •
Exercise 30. Justify t h e claim that χ(Α') = 0 in the preceding

proof.
A t o n e point near t h e end of the proof of R6dei's t h e o r e m we

will n e e d the following lemma.
L e m m a 13. (Fifth replacement principle) Let A be a subset of G

such that \A\ — pis an odd prime factor of \G\ and e € A Suppose
that the nonidentity elements of A have order ρ or 2 p Then A can be
replaced in any factorization ofG by A' = {α(ρ)(α(ρ))*( ) : a € A}α
for any integers s(a).
Proof Let A = { α ϊ , a ,...,

2 a }, with o j = e. Write each Oj in t h e
p
form o< = α<(2)α<(ρ). H e r e (a<(2)) = e and ( a ( p ) ) = e. N o t e

2
f
p
that a\ (2) = a\ (p) = e. A s argued in the proof of L e m m a 12, there

is a character χ of G such that χ{Α) = 0. For such a character we
have
χ ( α ( 2 ) ) χ ( α ( ρ ) ) + χ(α2(2))χ(θ20»)) + · · · + χ ( α ρ ( 2 ) ) χ ( α ( ) ) = 0.
ι 1 ρ Ρ
Now, x(eii(2)) = £j = ± 1 and χ(α*(ρ)) = ρ " , w h e r e rij may b e

4
chosen in the range from 0 to ρ - 1 and ρ is a primitive p t h root of

1. We have
ειρηι
+ ερ* 2
η
+ •••+ ε "
ρΡ
η
= Ο,
w h e r e ε! = 1 and ηχ = 0. T h u s ρ is a root of
g(x) = ε ι Χ
η ι
+ e x" + · • · +
2
a
e x *,
p
n
a polynomial of degree less than p, or the zero polynomial.

If it were the zero polynomial, the coefficient of x , which is Ui
a sum of l ' s and ( - l ) ' s would be 0. H e n c e there would b e an even

n u m b e r of terms corresponding to each exponent. Since t h e r e are
ρ terms, and ρ is odd, this would b e a contradiction. T h u s g(x) is
not the zero polynomial.
Since g(p) = 0 and g(x) has constant term 1, g(x) = F (x), the p
cyclotomic polynomial for t h e pth roots of 1. T h a t implies that ε< =

1 for alii and that n i , r i 2 , . . . , n is a r e a r r a n g e m e n t of 0 , 1 , . . . , p - l .
p
F r o m this it follows that \A'\ = N o t e that if χ(Α) = 0 then

χ(Α') = 0. By the General Replacement Principle, A may be re-
placed by A' in any factorization. •
5. Proof of Redei's theorem

We come now to the proof of Redei's t h e o r e m .
Let ρ b e a prime divisor of the o r d e r of t h e finite abelian g r o u p
G. We will say that an element g e G is a p-element if g(p) = e. In
o t h e r words, the p-elements are the elements in t h e p - c o m p o n e n t
ofG.
T h e proof also uses t h e result in the next exercise.
Exercise 3 1 . Let G = Αχ • • • A b e a normalized factorization of

n
the finite abelian group G and suppose that e r c h \ Ai\ is prime. Let
ρ b e a prime divisor of |G|. Show that if all t h e factors of cardinality
ρ contain only p-elements then these factors form a factorization of
the p-component of G. (Hint: Consider cardinalities.)
Theorem 1. (Redei's T h e o r e m ) Let Gbea finite abelian group and

G = ΑχΑ • • · A a normalized factorization such that each Ai con-
2 n
Redei's Theorem 183
tains a prime number of elements. Then at least one Aiis a subgroup

ofG.
Proof. By L e m m a 10 the theorem holds for p-groups, so let G b e

a non-p-group and G = Αχ · · · A a normed factorization of G,n
where the factors are of prime cardinalities. We use the inductive

assumption that the theorem holds for each p r o p e r subgroup Η of
G.
Let ρ be a prime divisor of \G\ and suppose that Αχ,..., A are t
the factors in the factorization G = Αχ ·• • A with cardinality p . n
T h e smallest value of
t
h (A ,...,A )
p 1 n = J\ JJ |S(p)|
ΐ=1 αξζΑ,
is 1, which is attained only when each Ai, 1 < i < t contains only
p-elements. In this case, by Exercise 31, Αχ • •• A is a factorization t
of the p-component of G. By Redei's theorem for p-groups, at least

o n e Ai is a subgroup of the p-component of G, hence of G. We now
proceed by induction on h (Ai,..., A ). p n
Let ρ b e the least prime factor of \G\. We may assume that

h ( A i , A ) > 1, that is, there is an i, 1 p.
Suppose that Ai contains m o r e than one non-p-element. By
L e m m a 12, Αχ can be replaced by a nonsubgroup A\ such that
h (A\, A , A ) < h ( A i , A ) . (Let all but one of the ex-
p 2 n p n
p o n e n t s s{a) in L e m m a 12 be 0.) By the inductive assumption o n e

of the factors A[, A , • • •, A is a subgroup of G. Since A[ is not a
2 n
subgroup, at least one of A ,..., A is a subgroup of G, and the

2 n
t h e o r e m is established in this case.

T h u s we may assume that a is the only non-p-element of A\.
Assume that |α(ρ) | > p . Then, by L e m m a 12, Αχ can b e replaced
by Αχ, the set which consists of the p-parts of the elements of Αχ.
(Let all the exponents s(a) in L e m m a 12 be 0.) Now, A[ is not
a subgroup of G and h (A[, A ,..., p A) < h (Ax,A ).
2 Byn p n
the inductive assumption o n e of t h e factors A ,..., A is a sub- 2 n
group of G. We may suppose therefore that \a(p)\ = p. Similarly, if

|ä(p) I > q, then Αχ can be replaced by a nonsubgroup A[ for which
h (A' ,A2,A )
p l n < h ( A i , A ) . O n c e again by the inductive
p n
assumption o n e of the factors A ,..., A is a subgroup of G. So, 2 n
finally, we suppose that \a\ = pq.

By L e m m a 12, for each integer i, 1 < i < q — 1, Αχ can be
replaced by
Ax <i = {a(p)(a(q)) }U(Ax\{a}) i
to get the factorization G = A A2•·• A . We will use this factlti n
soon. By L e m m a 12 A\ can be replaced by A\, which consists of the

p-parts of the elements of Αχ. Now
h (A'x,A ,...
p 2 ,A )n < h {Ax,...,A ).
p n
By the inductive assumption o n e of the factors Αχ,Α ,...,Α is a 2 η
subgroup of G. If this is not A[, then we are d o n e . H e n c e we assume

that Η = Α'χ is a subgroup of G.
From the factorization G = HA ·• · A it follows that there 2 n
is a permutation Η = Βχ, 02, · ·, Β of t h e factors Η, A ,..., A η 2 n
such that Βχ, ΒχΒ , • ••, BxB • •• B are subgroups of G. We may

2 2 n
assume that the permutation is just the identity as this is only a mat-
ter of reindexing the factors.
Consider the subgroup Κ = HA • • • Α -χ. If Αχ c K, then 2 η
Κ = ΑχΑ • · · Α -χ is a factorization of K. By the inductive as-

2 η
sumption about the subgroups of G, o n e of the factors Αχ,..., Α -χ η
is a subgroup of K, and so of G.
Thus we assume that Αχ φ Κ, that is, a(q) & K. From this it
follows that \G : K\ = q. So from the factorization G = KA we n
have |J4„| = q. Again, the factorization G = KA implies that A n n
is a complete set of representatives m o d u l o K. Therefore for each

i, 1 < i < q-1 there is an a* 6 A such that t h e coset a,iA contains
n n
the element (a(p))~ (a(q))~ . This means that a(p)(a(q)) ai

1 l
e K. l
Let
d = {α(ρ)(α( )γ }υ(Αχ\{α})
ς αί = {α(ρ)(α(ς))«α }υ(ΖΓ\{α( )}). 4 9
Redei's Theorem 185
Note that Κ = C A • · • Α -χ is a factorization of K. Indeed,

X 2 η
products coming from C A • • · Α _χ X occur among the products

2 η
coming from A\Â • · ·Α -χΑ

2 and these latter are distinct since
η η
Ai iA
t 2• · • Α -χΑ
η η is a factorization of G.
By the inductive assumption about the subgroups of G we have
that one of the factors Ci, A ,..., Α -χ is a subgroup of K. If this
2 η
is not d we are done. Hence we assume that Ci is a subgroup of K.

We distinguish two cases depending on whether ρ = 2 or ρ > 3.
If ρ > 3, then the subgroups C< and Η have a nonidentity
element in common and, both have prime order, Ci = H. T h u s
a(p)(a(q)Yai — a(p), that is, a = ( o ( o ) ) . This means that A =
x
- i
n
(a(q)) and therefore A is a subgroup of G.

n
Finally we suppose that ρ = 2. Now (a(p)(a(q)) ai) = e, i 2
hence a ^ g ) = ( a ( o ) ) ' . Thus ( a ( g ) ) are the g-parts of the non-

- _ i
identity elements of A . If the 2-part of each element of A is the

n n
identity element then A is a subgroup and we are done. Thus the

n
order of each nonidentity element of A is either q or 2g and A n n
contains an element whose order is 2g.

Making use of the replacement principle in L e m m a 13, we can
repeat the whole argument, starting with A instead oi Αχ. Since n
\A \ = q > 3 the procedure will not go on forever. This completes

n
the proof. •
Problem 2. Obtain a shorter or simpler proof of Redei's theorem.
References
1. K. Corrädi and S. Szabo, A new proof of Redei's theorem, Pacific J.
Math. 140(1989), 53-61.
2. D. S. Dummit and R. M. Foote, Abstract Algebra, Prentice Hall, En-
glewood Cliffs, 1991 (p. 466, irreducibility of the cyclotomic polyno-
mials).
3. L. Redei, Die neue Theorie der endlichen Abelschen Gruppen und
Verallgemeinerung des Hauptsatzes von Hajos, Acta Math. Acad. Sei.
Hung. 16 (1965), 329-373.
Epilog
While the central theme of the seven chapters has been the inter-
play of geometry with algebra, another theme, purely algebraic, also
ran through the book. Time and time again we have used mappings
from o n e structure to another to simplify a problem.
In Chapters 1 and 2, where we faced an infinite lattice L imbed-
ded in a rational lattice V, we formed the quotient group L'/L.
T h e natural homomorphism from V to L'/L enabled us to work in
L'/L, which is a finite group, instead of an infinite group.
A t t h e end of Chapter 2 the solution of a brick-tiling prob-
lem d e p e n d e d on the homomorphism from the ring of polynomials
Z[x, y] to the ring of complex numbers obtained by replacing χ a n d
y by a root of unity.
Two homomorphisms were the key to Chapters 3 and 4 o n the
semicross and cross. O n e homomorphism, defined on Z , was o n t o
n
a finite group G. It enabled us to lift a splitting of G to a tiling of Z .

n
O n e way to construct such a splitting when G is a cyclic group is to

find a "logarithm" φ : { 1 , 2 , . . . , fc} —• C(k). Such a logarithm turns
out, in certain circumstances, to be extendable to a h o m o m o r p h i s m
χ : C(p)* —> C(fc) for infinitely many primes p .
T h e key to the equidissections in Chapter 5 was the notion of
a valuation φ. Though φ is not a ring homomorphism from R to R,
it does behave fairly well with respect to the ring structure.
In Chapter 6, which concerns mainly tilings by similar trian-
gles, we relied on isomorphisms between subfields of the reals or of
187
the complex numbers. These induced certain transformations on

subsets of t h e plane, which w e r e also homomorphisms.
T h e proof of Redei's t h e o r e m in Chapter 7 repeatedly used
arguments that exploited characters. A character is yet another h o -
momorphism, this time from a finite abelian group to the complex
n u m b e r s of modulus 1. Any character can be extended to a ring
homomorphism from a group ring t o the field of complex numbers.
To apply algebra to a geometric problem, then, we first trans-
lated the geometric conditions into some algebraic structure. Then
we worked in that structure or, with the aid of a homomorphism, in
some other algebraic structure, chosen either because it is simpler
or because it is richer.
These observations show that the book indeed deserves t o bear
the subtitle, " H o m o m o r p h i s m s in the service of geometry."
Appendix A
Lattices
Let vi, v , • • •, v b e η linearly independent vectors in Euclidean n-

2 n
space, ß " . T h e set of vectors of the form mivi + 7 7 1 2 ^ 2 Η t-m„u , n
TBJ e Z , is called a lattice of dimension η and υ ι , t > 2 , . . . , v form a

n
basis for the lattice. T h e parallelepiped spanned by vi, v , • • •, v is 2 n
a fundamental parallelepiped for the lattice.

For example, the vectors v = ( 0 , . . . , 0 , 1 , 0 , . . . , 0), 1 < i < n,
t
where the 1 is in the ith place and 0's are in the other η - 1 places,
are a basis for the lattice Z . n
Exercise 1. W h e n d o the vectors w = (a, b) and w = (c, d) form

x 2
a basis for Z ? 2
Exercise 2. W h e n is the vector uii = (a, b) part of a basis for Z ? 2
Exercise 3. Let vi, v , • • •, v b e a basis for the lattice L. Let wi,

2 n
w ,...,
2 w be in L. T h e n there are integers a y such that
n
For which matrices A = ( a y ) is the set u>i, w ,..., 2 w n also a basis

for L?
Exercise 4. If u i , v ,...,
2 v and tui, ι υ > · · ·, vj are bases for the
n 2 n
same lattice, show that the determinant of the matrix whose rows
189
are v X) v ,...,
2 v„ is equal to + or - the determinant of the matrix
whose rows are w\, w ..., 2} w. n
ExerciseS. Prove that the volume of the fundamental paral-

lelepiped associated with a basis for a given lattice L is independent
of the particular basis.
Exercise 6. Let G = Z n
and let i f be a lattice of dimension η in
G. Prove that the index of Η in G is finite.
By Exercise 6, t h e r e is a finite set , g , • • •, g in Z such that

2 r
n
each element in Z is uniquely expressible in the form h+gi, he H,

n
1 < i < r. T h u s translates oi{gi,g ,...,g }by 2 elements of Η tile

r
Z . Therefore a large cube in R of volume V contains approxi-

n n
mately V/r elements of H. Let V* b e the volume of a fundamental

parallelepiped of H. T h e large cube of volume V contains approxi-
mately V/V* translates of the fundamental parallelepiped by trans-
lates of H. Therefore r = V*. This informal argument suggests that
the index of Η in Z is equal to the volume of a fundamental par-
n
allelepiped of the lattice Η c Z . n
Exercise 7. Fill in the details to make the informal argument rig-

orous.
We will also need fc-dimensional lattices in RJ , where k < n. 1
Let v\, v ,...,

2 Vk b e k linearly independent vectors in R . T h e set n
of vectors of the form mivi + m v + •·· + rrikVk, rrii G Z, is a

2 2
lattice of dimension k.
W h e n is a subset L of R a lattice? Necessarily, L is a group

n
u n d e r addition and contains no accumulation points (equivalently,

every ball contains only a finite n u m b e r of points of L). T h e s e con-
ditions are also sufficient, as the following t h e o r e m shows.
Theorem 1. Let Lbea subset of R that is a group under n

addition
and contains no accumulation points. Then Lisa lattice.
Lattices 191
Proof. Since L has n o accumulation points, there is a shortest

nonzero vector in L. Call one such vector v\. T h e r e are now two
cases. Either L is contained in the vector space generated by υχ or
it is not.
In t h e first case we show that {v } is a basis for L. To d o this,
x
let ν be any vector in L. Then ν is of the form xv\ for some real
n u m b e r x. Then χ = η + f, where η e Ζ and 0 < / < 1. Since
nv\ G L, we have fv\ = ν — nvi also in L. But is shorter than
« i , hence must b e the zero vector. So {v } is a basis for L.
x
In the second case, let Π be the plane through the origin per-
pendicular to vi. Every vector υ e R can be expressed uniquely in
n
the form ν = xvi + w, where χ e R and w e Π. Call w the projec-

tion of υ on Π, and denote it T(v). Let Π ι be the plane parallel to
Π through the tip of v\, as shown in Figure 1.
Let T(L) = {Τ{υ), υ e L}. We will show that T(L) is a lattice.
Clearly T(L) is a group. To show that it has no accumulation
points, n o t e that when υ is written in the form xvi + w, and χ =
n+f, as defined above, then T(v) = T{fv\ +w). T h u s every vector
in T ( L ) is the projection of a vector that lies between the planes
Π and Πχ. Now assume that Ρ e Π is an accumulation point of
T(L). Let B be a ball in Π a r o u n d P. A cylinder with base Β and
FIGURE 1
height I i>i I would then contain an infinite n u m b e r of points of L.

This contradiction shows that T(L) is a lattice.
We now may argue by induction on t h e dimension of t h e lat-
tice. Let w ,...,
2 Wk be a basis for T(L). Select Vi G L such that
T(vi) = Wj, 1 < i < k. T h e n vi,v ,... 2 ,v is a basis for L. To
k
show this, consider ν € L. T h e n T(v) = n w + • • • + nkWk for

2 2
some integers n , . . . , nk. H e n c e T(v - n v

2 2 2- · · · - nkVk) = 0.
Thus ν - n v 2 - . · . - rifcUfc is in the space spanned by v\. Since
2
ν — ri2t>2 - · · · - n Vk is also in L, it is of t h e form n i ^ i , for some

k
integer ηχ. This shows that ν = riii>i + 712^2 + · · · + n Vk, and the k
inductive step is complete. •
Theorem 2. Let Μ be a sublattice of the lattice L. Then Μ is a sum-

mand of L if and only if zl e Μ implies I € Μ for every nonzero
integer ζ and vector I e L.
Proof Assuming that the condition is satisfied, we will show that

Μ is a s u m m a n d of L. Consider the quotient g r o u p L/M, which is
a finitely generated abelian group without elements of finite o r d e r
(other than the identity element). T h u s L/M has a basis of t h e form
Zi + M,l + M,... , i + M . If T B I , T B 2 , . . . ,m, is a basis for M , then
2 r
m i , m ...,m ,h,l ,...,l

2 ) e 2 is a basis for L. Letting Ν b e the lattice
r
with basis h,h,--,lr, we have the direct sum decomposition,

L = Μ φ Ν.
Now assume that Μ is a s u m m a n d of L, L = Μ Θ Ν. Assume
that I e Lis not in M. T h e n I = m + n, m G Μ, η e Ν, η not 0.
If ζ is a nonzero integer, then zl is not 0. T h u s ζ is not in M. This
completes t h e proof. •
Appendix Β
The Character Group and

Exact Sequences
T h e character group of a finite abelian group is used in t h e proof

of R i d e i ' s theorem in C h a p t e r 7 and in the proof of T h e o r e m 4 in
C h a p t e r 3. We define this group here and develop its properties in
a sequence of exercises. Unless otherwise stated, all the groups a r e
finite a n d abelian.
Let C be t h e group of complex n u m b e r s of magnitude 1 un-
d e r multiplication and let G b e a finite abelian group. A h o m o m o r -
phism χ : G —» C is called a character of G. T h e product of t h e
characters χ ι and χ , denoted χ ι Χ 2 , is defined by the equation
2
(XiXaXff) = Xi(fl)X2(s),
for g e G. With this multiplication, the set of characters form a

group, called the character group of G, and is d e n o t e d G.
Exercise 1. Prove that if G = C(n), the cyclic g r o u p of order n ,

then G is isomorphic to G.
Exercise 2. Prove that if G is the direct product of Gι and G , then 2
G is isomorphic to G\ ® G . 2
Exercise 3 . Prove that any finite abelian group is isomorphic to its

character group.
193
Any character of a subgroup of G can b e extended to a char-

acter of G. This is a consequence of the following exercise.
Exercise 4. Let Η be a subgroup of the finite abelian group G

and a an element of G that is not in H. Let χ b e a character of H.
T h e n χ can b e extended to a character of the subgroup spanned by
a and H. (Suggestion: Let η be the minimum of the set of positive
integers m such that a e H. Let ζ be a complex n u m b e r such that
m
z = x(a ). Define χ* (a) to be ζ and χ* (a h) = z x(h). Show that

n n r r
χ* is well defined and extends χ.)
A sequence of abelian groups Go, G\, G , · · · , G 2 m with h o m o -

morphisms a* : d —• Gi-i, 1 < i < m,
a r, <*M-i ot , cx-i αϊ
Gm m
• Lr _l
m • · · •
3 r
• (_r 2
n
• Ln • Lro
n
is exact if the image of a, equals the kernel of α<_ι, 2 < i < m .

Consider an exact sequence of five groups where the first and last
are the group of o r d e r 1:
{0} —» A -^-» G -^-* Β — • {0}. (1)
This sequence is exact if and only if β is a homomorphism from G

o n t o Β and α is a one-to-one isomorphism of A o n t o the kernel of
β. Loosely put, "B is isomorphic to G/A."
If t h e r e is an exact sequence (1) it turns out that there is also
an exact sequence
{0} — • Β G A — » {0}. (2)
This implies that for finite abelian groups a group is a h o m o m o r -

phic image of G if and only if it is isomorphic to a subgroup of G.
(For infinite abelian groups, this is not true: C(2) is a homomorphic
image of Z, viewed as an additive group, but is not isomorphic to a
subgroup of Z.)
Exercise 5 shows that there is the reverse exact sequence (2).
For convenience, assume that A is a subgroup of G and a is the
inclusion mapping.
The Character Group and Exact Sequences 195
Define 5 : G -* A as follows. For χ 6 G, define &7(χ) to be the

restriction of χ t o A. Define β : Β —• G as follows. For χ e B, let
/3(χ) b e t h e composition χβ.
Exercise 5. Show that t h e sequence
{0} —> Β JL G A —+ {0}
is exact.
Combining Exercises 3 and 5, we see that if there is an exact

sequence (1), then there is an exact sequence (2).
T h e final exercise shows that a set of distinct characters is lin-
early independent.
Exercise 6. Let G be a finite abelian group and Xi,X2, - • • , χ dis- η
tinct characters of G. Show that if cy, C 2 , . . . , Cn are complex num-

bers such that
η
Y^CiXi{g) =Q
i=l
for all g e G, then CJ = 0, 1 < i < n, as follows. If there is a
nontrivial linear d e p e n d e n c e between s o m e of the characters, there
is o n e with the smallest n u m b e r of nonzero coefficients. A s s u m e
that
η
Σ ( !
ί»(ί) = ( 1
t=l
is such a relation, d φ 0, 1 < i < n. N o t e that η is greater than 1.

T h e n obtain a linear d e p e n d e n c e with fewer n o n z e r o coefficients.
(Suggestion: P i c k # € G such that x\{go)
0 φ Xiigo)- T h e n show
η
^CiXi{9o)Xi{9) =0
i=l
for all g Ε G. Using the two linear relations involving χ ι , - . - , χ , η
produce o n e that has fewer nonzero coefficients.)

Appendix C
Formal Sums
Let S = {si, 92,..., 8 ) b e a finite set and A an abelian group

n
written additively. We will give a precise definition of the group of

"formal sums," αιβι + (Z2S2 Η h o „ s „ , Oj G A.
Let / b e a function from S into A. If / ( s ^ = a 1 < i < n,
it
d e n o t e / by the formal sum
a\Si + a 82
2 + ···+ a8.
n n
D e n o t e the set of the formal sums by G(S, A).

Let / and g b e functions from S into A. Define their sum, f+g,
by ( / + g)(s) — f(s) + g(a) for each s e S . T h e r e is another way
t o describe this addition. Suppose that / ( S J ) = a< and g(8i) = bi,
1 < i < n. T h e n the sum of the corresponding formal sums is
defined by
(ÔiSi) + (Σ ) δ<β< =
^2(α, + bi)si.
In C h a p t e r 6 the set S consists of ordered pairs of the form PQ,

where Ρ and Q are vertices in a fixed dissection of a polygon and
the abelian group A is Z. In this case G(S, Z) is called the group of
chains. Let Η be the subgroup of G generated by the chains of the
form PQ + QP or PQ + QR+RP, where P, Q, and R are collinear
and Q is between Ρ and R.
T h e line integral f xdy is defined for each directed edge Ε
B
a n d therefore J xdy is defined by linearity for each chain C. N o t e

c
197
that if C is in H, then f xdy — 0. Therefore we may view / xdy as

c
being d e n n e d on the elements of t h e quotient g r o u p G/H.
Exercise 1. Show that if P, Q, and R are collinear vertices, then

PQ + QR + RP is in H, even if Q is not between Ρ and R.
In C h a p t e r 7 t h e set S is a finite group, whose composition

we d e n o t e multiplicatively and the abelian g r o u p A is Z. N o t e that
t h e r e is a multiplication in Z. In this case G ( S , Z) can b e m a d e into
a ring, called the group ring of the g r o u p S. Addition in G(S, Z) has
already b e e n defined. Define the product of £ % i d Σ
a s an
to b e
Σ CiSi, w h e r e c< = Oj&fc, where t h e summation is over all pairs
( j , k) such that SjSk — Si- This definition is similar t h e definition of
the product of two polynomials.
Appendix D
Cyclotomic Polynomials
Let η be a positive integer. T h e equation x - 1 = 0 has η com-

n
plex roots, called n t h mots of unity. For an n t h root of unity ρ let

m be the smallest of the positive integers j such that p> = 1, that
is, its order in the multiplicative group of n t h roots of unity. By La-
grange's theorem, m divides n. A n nth root of unity whose order is
η is called a primitive nth root of unity. T h e polynomial
F (x)
n = l[{x-u>),
as ω runs through the primitive n t h roots of unity, is called the n t h

cyclotomic polynomial. Its degree is given by the Euler phi-function,
φ(η), which is the n u m b e r of integers i, 1 < i < n, relatively prime
to n. Since each root of unity is a primitive dth root of unity for
some divisor d of n, we have
x - l = Y[F (x).
n
d (1)
d\n
With the aid of (1) we can show by induction that each cyclotomic
polynomial has integer coefficients. (Moreover, it is irreducible
over Q, as most algebra texts prove.)
N o t e that for a prime p,
F (x)
p = 1 + χ + • • • + x- p 2
+ x ~\
p
since each primitive pth root of unity is a root of (x p

- l)/(x - 1).
199
Exercise 1. Prove that F (x) n is in Z[x\.
Let Ω be the field obtained by extending Q by the nth roots of

η
unity, or, equivalently, by a single primitive nth root of unity. T h e

degree of this extension, [ Ω : Q], is φ(η), the degree of F (x).
η n
Lemma 1. If (m, n) = 1 then the smallest field containing il m and

Ω„ is n . mn
Proof Let Ω„, Ω d e n o t e the smallest field containing Cl a n d Ω .

η m η
Clearly Ω Ω c Ω „ , . Now, since ( m , n ) = 1 there are positive

η ι η η
integers a and b such that 1 = am - bn. Let ρ b e a primitive ( m n ) t h

root of unity. Then
P=(p r/(p ) -
m n b
But p is in Ω
m
η and p n
is in il - m H e n c e Ω„,„ c Ω , „ Ω , and the
η
lemma follows. •
In Chapter 7 we use the following theorem, which is an imme-

diate consequence of L e m m a 1.
Theorem 1. If ( m , n ) = 1, then F (x) m is irreducible in the ring

Ω„[4
Exercise 2. Prove the theorem.
Exercise 3. Show that if ρ is a prime and η is a positive integer,

then Fpn(x) = F ( a ; P ) . (Hint: If ρ is a primitive (p")th root of
p
n _ 1
unity, show that pP" is a primitive pth root of unity.)

1
Bibliography for Preface
1. F. W. Barnes, Algebraic theory of brick packing I, Discrete

Math. 42 (1982), 7-26, and II, ibid, 129-144.
2. R. Berger, The undecidability of the domino problem, Memoirs
A m e r . Math. Soc. N o . 66 (1966), 72 p p .
3. V. G. Boltanskii, Hubert's thirdproblem, Scripta Series in Math-
ematics, Wiley, New York, 1978. (Review in BulLAMS, 1 (1979),
646-650.)
4. , Equivalent and equidecomposable figures, D. C. H e a t h ,
Lexington, 1963.
5. J. H . Conway and J. C. Lagarias, Tiling with polyominoes and
combinatorial group theory, /. Comb. Theory Series A 53
(1990), 183-208.
6. N. G. de Bruijn, Algebraic theory of Penrose's non-periodic
tilings of the plane I, Proc. Akad. van Wetenschappen, Proc. Ser.
A 84 (1981), 39-52, and II, ibid., 53-66.
7. F. Μ. Dekking, Replicating superfigures and endomorphisms
of free groups, / Comb. Theory Series A 32 (1982), 315-320.
8. D . Gale, M o r e on squaring squares and rectangles, Mathemat-
ical Intelligencer 15, n u m b e r 4 (Fall, 1993), 6 0 - 6 1 .
9. S. Golomb, Polyominoes, Princeton University Press, 1994.
10. B. G r ü n b a u m and G. C. Shephard, Tiling with congruent tiles,
Bull. Amer. Math. Soc. 3 (1980), 951-973.
201
11. , Tiling and Patterns, Freeman, New York, 1987.

12. R. Kenyon, Tiling with squares and square-tileable surfaces,
Ecole Normale Superieure de Lyon, 119 (1993), 1-26.
13. D . A. Klarner and F. Göbel, Packing boxes with congruent fig-
ures, Indag. Math. 31 (1969), 465-472.
14. ML Laczkovich and G. Szekeres, Tilings of the square with simi-
lar rectangles, Conference in honor of Laszlo Fejes-Toth's 80th
Birthday, to appear in Discrete and Combinatorial Geometry.
15. D. G. M e a d and S. Stein, M o r e on rectangles tiled by rectan-
gles, Amer. Math. Monthly 100 (1993), 641-643.
16. H . Meschkowski, Unsolved and unsolvable problems in geome-
try, Ungar, New York, 1966.
17. R. Penrose, Pentaplexity, a class of non-periodic tilings of the
plane, Math. Intelligencer 2 (1979), 32-37.
18. R. M. Robinson, Undecidability and nonperiodicity of tilings
of the plane, Inventiones Mat. 12 (1971), 177-209.
19. D . Schattschneider, Will it tile? Try the Conway criterion,
Math. Magazine 53 (1980), 224-233.
20. S. Stein, T h e notched cube tiles R , Discrete Math. 80 (1990),
n
335-337.
21. S. Szabo, Cube tilings as contributions of algebra to geometry,
Contributions to Algebra and Geometry 34 (1993), 63-75.
22. W. P. Thurston, Conway's tiling groups, Amer. Math. Monthly
97 (1990), 757-773.
23. S. Wagon, Fourteen proofs of a result about tiling a rectangle,
Amer. Math. Monthly 94 (1987), 601-617.
Supplement to the Bibliography
1. J. Charlebois, Tiling by (k, n)-crosses, Electronic J. of Undergraduate

Math. 7 (2001), 1-11. (Related to Problem 2 on page 61.)
2. J. Mackey, A cube tiling of dimension eight with no facesharing, Dis-
crete Comput. Geom. 28 (2002), 275-279. (Contains an 8-dimensional
counterexample for Keller's conjecture. On page 28 a 10-dimensonal
example was reported.)
3. C. H. Jepsen, Equidissections of Trapezoids, Amer. Math. Monthly 103
(1996) 498-500 (Settles Problems 1 and 4, p. 126).
4. P. Monsky, Calculating a Trapezoidal Spectrum, Amer. Math. Monthly
103 (1996) 500-501.
5. A. Munemasa, On perfect i-shift codes in abelian groups, Designs,
Codes, and Cryptography 5 (1995), (Related to Chapter 3, uses splittings
to construct codes.)
6. I. Praton, Cutting Polyominoes into Equal-Area Triangles, Amer. Math.
Monthly 109 (2002) 818-826. (This and reference 8 generalize Theorem
1, page 118, to any polyomino.)
7. S. Saidi, Semicrosses and quadratic forms, European J. of Combinatorics
16 (1995), 191-196. (Connected to Chapter 3, uses quadratic forms to
study splittings.)
8. S. Stein, Cutting a Polyomino into Triangles of Equal Areas, Amer. Math.
Monthly 106 (1999) 255-257.
9. , A Generalized Conjecture About cutting a Polygon into Triangles
of Equal Areas, Discrete and Comp. Geometry 24 (2000) 141-145.
203
10. , Cutting a Polygon into Triangles of Equal Areas, Math. Intell. 26

(2004) 17-21.
11. Z. Su and R. Ding, Dissections of Polygons into Triangles of Equal
Areas, J. Appl. Math, and Computing 13 (2003) 29-36.
12. U. Tamm, Splittings of cyclic groups and perfect shift codes, IEEE
Transactions on Information Theory 44 (1998), 2003-2009. (Connected
to Chapter 3, uses splittings to construct codes.)
13. A. Tiskin, Packing Tripods: Narrowing the Density Gap, Discrete Math-
ematics (to appear) (Determines the limit in Problem 3, page 97.)
Name Index
Abrams, A. vii Gobel, F. 201

Alhocain, Abu Djafar Mohammed Ben Golomb, S. 53, 54, 80, 82, 200
1 Grunbaum, B. 201,202
Alkhadjandi, Abu Mohammed 1, 2
Hajos, G. vi, 22, 23, 26-29, 32, 152,
Bachet, CG. 2 155, 156, 159-161
Bachman, G. 133 Hales, A. W. 109, 133
Banach, S. vi Hamaker, W. 80, 82, 105
Barnes, F. W. vi, 201 Hermite, C. 9, 10-12, 15
Berger, R. 54, 201 Hickerson, D. R. vii, 65, 72, 76, 82,
Boltanskii, V. G. 201 97, 105
Bruijn, N. G. de vii, 48, 54, 201 Hilbert, D. 10, 15
Hocking, J. C. 55
Chakerian, G. D. vii, 120 Hsiang, W. Y. 86
Chrisman, M. vii, 148 Hurwitz, A. 10
Conway, J. H. vi, 31, 32, 53, 54, 105,
201 Ireland, Κ. 133
Corrädi, Κ. 28, 33, 185
Joy, Κ. 95
Dad-del, Α. 101,105
Dehn, Μ. W. vii, 140 Kärteszi, F. 80, 82
Dekking, F. M. 200 Kasimatis, Ε. Α. 81, 120, 128, 130,
Diophantos 1,2 133
Dummit, D. S. 185 Keller, Ο. Η. 28, 33
Kenyon, R. vii, 202
Euler, L. 2, 162 Kepler, J. 85
Everett, Η. 105 Klarner, D. A. 55, 202
Eves, Η. 153 Klee, V. vii, 108, 109, 133
Korchmaros, G. 80
Fejer, L. 155
Fejes-Toth, L. 202 Laczkovich, M. vii, 135, 151-153, 202
Fermat, P. 2 Lagarias, J. C. 28,31,33,53,54,201
Foote, R. M. 185 Lagrange, J. L. 2
Forcade, R. W. 78, 82 Lenstra, A. K. 31,33
Fourier, J. B. J 155 Lenstra, H. W. 31,33
Freiling, C. vii Loväsz, L. 31,33
Freiler, M. 80
Frenicle, B. 2 Mackinnon, N. 53, 55
Furtwangler, P. 29, 32, 33 Mead, D. G. 120,133,202
Medyanik, A. I. 80
Gale, D. vii, 201 Mersenne, M. 2
Galovich, S. 82 Mertens, F. 31,32
Gauss, C. F. 5, 6, 32 Meschkowski, H. vi, 202
205
Mills, W. H. 79, 82 Schmidt, T. 22, 34, 35, 55

Minkowski, H. v, vi, 1, 10-12, 15-19, Senechal, M. vii
22, 23, 26-31, 33, 36, 152, 155, 156 Sloane, N. J. A. 31,32,105
Molnär, Ε. 80, 82 Shephard, G. C. 201, 202
Monsky, P. 108, 109, 118, 131-133 Shor, P. 28, 33
Sperner, E. v, 107, 109-111, 118, 120,
Odlyzko, A. M. 31, 33
122, 130, 132
Pascal, B. 2 Stein, S. K. 34, 80-82, 105, 106, 120,
Penrose, R. 202 130, 131, 133, 202
Perron, O. 28, 33 Strauss, E. G. 109, 133
Pollington, A. D. 78, 82 Szabo, S. 28, 33, 34, 47, 55, 80-82,
Posa, L. vii, 135, 147 106, 185, 202
Pythagoras 1 Szekeres, G. vii, 152, 202
Redei, L. vi, vii, 28, 32, 33, 36, 152, Tarski, A. vi

155-157, 161-165, 169, 172, 173, Thomas, J. vii, 107,109,118, 131,133
176, 178, 180-182, 185, 188 Thompson, T. 106
Richman, F. vii, 107-109, 113, 118, Thurston, W. P. vi, 55, 202
131, 133
Ulrich, W. 80, 82
Riele, H. J. J. te 31, 33
Riemann, B. 31 van der Waerden, B. L. 133
Rinne, D. vii
Robinson, R. M. 29, 33, 54, 202 Wagon, S. 202
Rogers, C. A. 105 Welch, L. 80, 82
Rosen, M. 133 Woepcke, Μ. Ε 34
Woldar, A. J. 83
Schattschneider, D. 202
Young, G. S. 55
Subject Index
adjacent of cube 120

clusters 38 of regular hexagon 128-130
cubes 37 of regular n-gon 120
of square 118
Brouwer's fixed point theorem 109 of trapezoid 121-126
equivalent
centrally symmetric set 12
clusters 38
character (of finite abelian group) 163,
vectors 38
193
Euler phi-function 198
group 193
exact sequence 67, 194, 195
principal- 163
clusters 35 factorization (of a group) 26
equivalent 38 fc- 29
complete triangle normalized 156
with respect to a labeling 110 replaceable factor of 157
with respect to a valuation 118 formal sum 196
complete polygon 110 Furtwangler's conjecture 155
covering 19
by crosses 97 group ring 167
by semicrosses 99
η-covering (of a group) 100, 102 Hajos's theorem 26
set (of a group) 100, 102 for p-groups 160
cross 58 harmonic brick 52
covering by 97 multiple of 52
packing by 87
lattice 3, 189
tiling by 60-62
basis of 4, 189
cyclic subset 27
determinant of 5, 8
cyclotomic polynomial 198
dimension of 4
density fundamental parallelepiped of 5,
of a covering 86 189
of a family 85 fundamental parallelogram of 5
of a packing 86 integral 30
of a lattice covering 86 standard 5
of a lattice packing 86 logarithm on S(k) 78
of a Z-lattice covering 86
of a Z-lattice packing 86 Mertens' conjecture 31, 32
Minkowski's conjecture 22
equidissection 119 algebraic form 26, 27
m- 118 generalization of 28, 29
of centrally symmetric polygon multiplier set 67
130-132 monotonic matrix 94
207
packing 19 set
by crosses 97-99 centrally symmetric 12
η-packing (of a group) 89 closed 12
by semicrosses 88-97 convex 12
set (of a group) 89 star (also star body) 58
p-component (of finite abelian group) shift 87
74 simplicial dissection 109
p-dimension 74 spectrum 120
p-group 74 principal 120
p-part 180 Sperner's lemma 107, 109, 110
Pythagorean theorem 1 oriented version 130
splitting (of a group) 62, 67
quadratic form coset- 77
determinant of 8 and exact sequence 69, 70
positive definite 2 multipler set of 67
relation to lattice 5 nonsingular 70
purely singular 70
set 62,67
Redei's theorem 28, 156, 182
singular 70
for cyclic p-groups 160
star body 58
general 182
for p-groups 165, 176 tiling 19
for type (p,p) 173 integer 35
replacement principle fc-fold 29
first 158 lattice 35
second 169 Q- 35
third 170 Q-lattice- 35
fourth 180 rational 35
fifth 181 Z- 35
general 169 Z-lattice- 35
Riemann Hypothesis 31 translate (of a set) 11
root of unity 198 twin (cubes) 22
primitive 198 type (of a finite abelian group) 162
semicross 58 valuations (of a field) 111

covering by 99-102 and dissections 117
packing by 88-97 equivalent 111
tiling by 59-66 extension of 113-116
p-adic 112
Symbol Index
|Λ| 159
(A) 159
AB 168
Χ 193
X(A) 163
C(m) 67
<5(C) 86
S (C)
L 86
<Sf(C) 86
F(k) 62
3 101
G 193
lal 159
(g) 159
g(y) 180
ä(l/) 180
(m) 120
m(fc) 95
(mi,...,m ) t 162
Ωη 199
<£(n) 198
<3(α) 127
S(k) 62
S(AT) 120
Τ (a) 121
0(C) 86
0L(C) 86
91(C) 86

Algebra and Tiling - Homomorphisms

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Algebra and Tiling - Homomorphisms

Uploaded by

Copyright:

Available Formats

Homomorphisms in the Service of Geometry

Algebra and Tiling

Published and Distributed by

Library of Congress Catalog Card Number 2006925584

Hardcover (out of print) ISBN 978-0-88385-028-2

Printed in the United States of America

Current Printing (last digit):

Cams Mathematical Monographs Editorial Board

The following Monographs have been published:

1. Calculus of Variations, by G. A. Bliss (out of print)

MAA Service Center

If n-dimensional space is tiled by a lattice of parallel unit cubes,

" C a n we trisect every angle with straightedge and c o m p a s s ? " "Can

T h e beginner might start with Chapter 1, go to de Bruijn's har-

audience. T h e exposition also benefited from the advice of G.

Budapest Davis, California

T h e origins of most of the tiling problems we will explore go back

some of the problems that Minkowski's question, in turn, suggested.

and y be d ( 2 m n ) and the other be d(m -n ),

z . H e does not refer to the equation x + y = z , whose first

recorded mention is in a letter written about the year A.D. 1000 by

I have already explained that the arguments that A b u

Alkhodjandi is better known for his observation of the obliqueness

Fermat, who was familiar with Diophantos through Bachet's

to Mersenne, in 1640, he wrote, "Each prime n u m b e r that is one

Pascal in 1654 he stated, " E a c h prime that is o n e m o r e than a mul-

further, examining forms of the type Ax +Cy ,

integers A, B, and C have the property that Ax + Bxy + Cy is

2. Quadratic forms and lattices

n e o u s " set of isolated points. It is an example of a "regular" point

In Appendix A it is shown that if L is a lattice in η-space, then

where mi e Ζ, 1 < i < k. Such a set of vectors is called a basis

Let v\, V2, • • •, v b e a basis for an η-dimensional lattice L in τι-

space. T h e parallelepiped whose edges a r e v i , V 2 , . . . , v „ is the set

{ x i V i + X2V2 Η h i t ) : 0 < Xi < 1,1 < i <

It is called a fundamental parallelepiped of the lattice. T h e volume

A it is shown that this volume is independent of the particular basis

basis for this lattice consists of the η unit vectors

(1,0,...,0),(0 1,0,...,0),... (0,...,1).

Exercise 2. Let a and & b e integers. W h e n is the vector (a, b) p a r t

In a book review in 1831 Gauss [4] showed that positive defi-

T h e positive definite binary form axx + 2bxy + cyy repre-

trary points in a plane whose coordinates along axes that

In this way the plane is divided into congruent parallel-

Figure 6 shows what Gauss had in mind.

T h e square of the length of xv\ + yv 2 is the value of the dot

x (i>i · vi) + 2xy(vi

the form ax + 2bxy + cy indeed does represent t h e square of t h e

length of a typical vector in the lattice.

c > 0 a n d b < ac. 2

(b) Is the converse true?

\ani a n 2 . • Ann ) Un/

Denoting the row vector (xi, x , . . . , x„) by x, its transpose (a col-

u m n vector) by x , and the matrix (aij) by A, we see that a quadratic

form is denoted by the matrix product xAx . T h e determinant of T

A is called the determinant of the quadratic form, which we d e n o t e

is positive definite, then the diagonal entries in Ε are positive.

T h e quadratic form x^4x = xBB x T

Xlhn + Χ2&2η Η 1" ΧΚ

= £ι(&ιι, · · · ,bin) Η V x (b i,-

Denoting the j t h row of Β by Vj, we see that xB — χχνχΛ \-χ ν · η η

H e n c e as χ varies, χ Β sweeps out the lattice with basis υχ, u , . . . , v . 2 n