Meschkowski - Unsolved and Unsolvable Problems in Geometry

H.
Meschkowski
Unsolved
and
Unsolvable
Problems
in
Geometry
UNSOLVED AND UNSOLVABLE PROBLEMS
IN G E O M E T R Y
UNSOLVED AND UNSOLVABLE
PROBLEMS IN GEOMETRY
HERBERT MESCHKOWSKI, Ph.D.

Professor of Mathematics, Padagogische Hochschule, Berlin
Lecturer at the Free University, Berlin
Translated by
J ane A. C. B urlak
FREDERICK UNGAR PUBLISHING CO.

N E W YORK
American edition 1966
by arrangement with Oliver and Boyd Ltd
First English Edition 1966
© 1966, Oliver & Boyd Ltd
Library of Congress Catalog Card No. 65-25028
P R IN T E D IN G R E A T BR IT A IN
PREFACE
At a time when new theories are constantly being developed in

mathematics, it seems worth while pointing out that there are still
unsolved problems in elementary geometry. We are thinking
especially of some problems which are so simple that even an
interested layman can understand them. Of course, it is only the
formulation of these problems that is quite so simple; and one
would need to have attained about the same standard in mathematics
as a student in his second year at university to understand the
mathematical arguments involved in discussing these problems.
We shall discuss some famous problems which occupied the
attention of mathematicians for thousands of years and which have
been shown to be unsolvable in recent decades and, in addition,
some very old classical problems which, up till now, have scarcely
been mentioned in books on the subject. We have endeavoured to
vary well-known methods of proof for classical problems and to
generalise their formulation.
Obviously a programme like this cannot be fully realised and we
have included only those open questions which seemed particularly
appropriate. We have completely omitted the problem of “ squaring
the circle ” in its usual form as there are already many descriptions
of it. The classical problem of the parallel postulate has also been
omitted since the proof that it is unsolvable would require the
fundamentals of non-Euclidean geometry to be developed and this
would be outside the scope of this book.
In the first chapter we list the problems which we consider in
detail. From the consideration of these follow further problems
which are formulated later in the text.
Thus, the object of this book is not the systematic representation
of a well-defined section of geometry. But rather our aim is of a
methodological nature: we want to develop the kind of arguments
which are suitable for the solution of the geometrical problems
considered and to discuss the questions implied by the existence of
unsolvable problems.
VI UNSOLVED AND U N S O L V A B L E PROBLEMS IN GEOMETRY
The numbers in square brackets refer to the bibliography at the

end of the book. This list may be useful for further study. In
connection with this we should like to make an important suggestion
to the reader.
In the experience of mathematical institutes and authors of
mathematical publications attempts are still made to solve problems
which are known to be unsolvable. And so we fear that there will
be some readers who, after reading this book, will believe that,
despite everything, they have found a solution to the problem of
trisecting an angle. We should like to advise these readers as
follows: Read through the proof showing that such a solution is
impossible three times and then try to find the mistake in your
argument. Better still, give up such attempts since they really
cannot lead anywhere. There are plenty of open questions—many
of which are stated in this book—to which mathematical intuition
may be applied with real prospect of success. However, it should
be noted that it is not particularly easy to solve the open questions
stated. It is sometimes much simpler to discover new results in a
new branch of mathematics rather than to solve one of the problems
left open in elementary geometry. However, according to E rhard
S ch m id t , it is better to solve old problems with new methods than
to solve new problems with old. Therefore we feel justified in
encouraging work on the “ elementary ” problems stated in this
book.
For the reader’s convenience there is, at the end of the Table of
Contents, a diagram showing the interrelationships of the chapters.
H erbert M eschkow ski

CONTENTS
PR EFA C E v
Chapter
I. THE PROBLEMS 1
II. REGULAR PACKINGS IN THE PLANE AND ON THE
SPHERE 5
1. Problems on the density of packing 5
2. Lemmas 8
3. Packing problems in the plane 13
4. Packing problems on the sphere 18
5. Further problems 25
III. IRREGULAR PACKINGS AND COVERINGS 29
1. The existence theorem 29
2. The packing problem graph 30
3. A point system for n = 7 32
4. The packing problem for n = 7 34
5. The covering problem graph 38
6. The covering problem for n = 5 40
7. The covering problem for n = 7 45
8. An outline of further problems 47
IV. PACKING OF CONGRUENT SPHERES 50
1. The densest lattice packing 50
2. The finest rigid packing 54
V. LEBESGUE’S “ TILE” PROBLEM 57
1. Formulation of the problem 57
2. The hexagonal tile 58
3. The tiles of Pal and Sprague 60
4. Curves and domains of constant width 62
5. Applications 65
6. The covering problem in 3$i 67
VI. ON THE EQUIVALENCE OF POLYHEDRA BY DISSECTION 71
1. The plane problem 71
2. Dehn’s theorem 73
3. Examples on Dehn’s theorem 75
4. The algebra of polyhedra 82
5. The basis polyhedra 84
VII. THE DECOMPOSITION OF RECTANGLES INTO 1NCON-
GRUENT SQUARES 91
1. The problem 91
2. The definition of a graph 95
3. Some calculations for decompositions 97
4. The interpretation in electrical engineering 100
5. The decomposition of rectangles with commensurable sides 101
vii
viii U N S OL V E D A N D U N S O L V A B L E PROBLEMS IN GEOMETRY
VIII. UNSOLVABLE EXTREMAL PROBLEMS 103

1. A theorem on the triangle 103
2. Besicovitch’s theorem 107
3. Sets of points at an integral distance apart 109
IX. CONSTRUCTIONS WITH RULER AND COMPASSES 113

1. Some comments on the use of ruler and compasses 113
2. Some construction problems 117
3. The proof of the impossibility of (8a) and (8b) 119
4. The quinquesection of the angle 123
5. The uniqueness of the construction 126
X. CONSTRUCTIONS ON THE SPHERE 128

1. Stereographic projection 128
2. Constructions with spherical compasses and spherical ruler 130
3. Squaring the circle on a sphere 133
4. A lemma 135
XI. PROBLEMS IN THE GEOMETRY OF SETS 137

1. A new definition of equivalence by dissection 137
2. Hausdorflf’s theorem 141
3. Paradoxical decompositions in ^3 150
4. Paradoxical decompositions in the plane 155
XII. THE METHODOLOGICAL SIGNIFICANCE OF “ UNSOLV

ABLE ” PROBLEMS 159
BIBLIOGRAPHY 163
INDEX 167
Chapter I
THE PROBLEM S
First of all we state the most important of the problems discussed

in subsequent chapters. Since this book deals with unsolved or
unsolvable problems our main concern is to present the material
associated with the solution of these problems. Thus we discuss the
solution of related (and usually simpler) problems, give an account
of partial results and so on. In the case of unsolvable problems we
have to prove that no solution is possible. However, we do not
indicate in this chapter which of the following problems are unsolv
able. We conclude by attempting to explain the significance of the
work on unsolvable problems for knowledge and culture in general.
We start then by enumerating the problems. These are numbered
for ease of reference. We denote related problems which are intro
duced later by adding the letters a, b, ... to the number.
(1) We are given a spherical planet and 10 («, in general) mutually
hostile dictators. How can the capital cities of these gentlemen
be distributed on the sphere so that the minimum distance between
any two of them is as great as possible?
This problem is equivalent to the following:—What is the smallest
radius r of a sphere on which 10 (n, in general) points can be placed
at a minimum distance of 1 unit apart ?
To see this equivalence it is best to choose as unit of length the
radius of the planet. Let d be the minimum distance apart for any
distribution of 10 (n, in general) points on this sphere; then a
dilatation by a factor <5-1 leads to a sphere of radius r = <5-1 and
distance of 1 unit between the points. Thus a minimum value of r
corresponds to a maximum value of S.
This problem (in the second form) was first tackled a few years
ago by S chutte and v a n der W aerden [III 1]. They gave solutions
for particular values of n.
l
2 U N S O L V E D A N D U N S O L V A B L E PROBLEMS IN GEOMETRY
(2) On any spherical planet n (n = 9, 10, 11, ...) fuel depots may be
placed so that the distance of any point on the surface from the
nearest depot is as small as possible.
This problem is certainly related to problem (1). It will be seen,
however, that the solution of (1) (for a definite value of n) does not
by any means always furnish a solution of (2).
(3) How should one pack a set of congruent spheres (in three-
dimensional space) so that the density of packing is as great as
possible ?
The “ density of packing ” can be defined as follows:—If r is the
radius of the congruent spheres to be packed without overlapping
each other and nR the number of the spheres of the given set which
lie completely within a sphere of radius R, and centre at some origin,
then
4 3
nR - nr3 3
d = lim ----------- = lim —
R-*a> 4 _3 J?->oo R
— 71K
3
is called the density of packing of the set. It can easily be shown
(see pp. 7 and 8) that this limit (if it exists) is independent of the
particular choice of the origin.
(4) What is the smallest plane figure with which every cluster o f
points o f diameter 1 can be covered?
A “ cluster of points ” is a finite set of points and its “ diameter ”
is the greatest distance between any two points of the set.
It is comparatively simple to determine the smallest circle, which
covers every cluster of points of diameter 1. Its radius, as H. W. E.
Jung [V 2] first showed, is equal to ^ /3 .
There are, however, still smaller plane figures (we shall call them
“ tiles ”) which cover the finite set of points under consideration.
The question, “ What is the tile of smallest area? ” has not yet been
answered.
(5) Under what hypotheses are polyhedra of equal content also
equivalent by dissection?
Two polyhedra are said to be “ equivalent by dissection ” if it is
possible to decompose them into finitely many sub-polyhedra which
THE PROBLEMS 3
are congruent in pairs and which have no interior points in common:
& = /n + /t2 + £ = <?■!+ <?2+ ••• + ?,,, (1)

= ^ 1.2, II.
It can easily be shown that plane polygons of equal area are also
equivalent by dissection.f The corresponding statement for
polyhedra is not valid. Necessary conditions are known for the
equivalence by dissection of polyhedra and these are not true for
many pairs of polyhedra of equal volume. The problem of finding
sufficient conditions for equivalence by dissection is not completely
solved.
(6) What is the smallest number of incongruent squares into which
a given square can be decomposed?
It is known that a square can be decomposed into 24 sub-squares,
no two of which are congruent. On decomposing a rectangle (with
commensurable sides) into squares, no two of which are congruent,
it is known that there are at least 9 such squares.
(7) In a domain Q) a line AB of length 1 may be moved (by trans
lations and rotations) so that the angle which AB makes with
a fixed line is changed by 360°. It is required to find a domain
of smallest surface area in which this is possible.
If AB is turned about A through 360° then the line traces out a
circle of area n. If AB is restricted to move as a chord of a circle
of radius R then the line traces out an annulus with radii R and
yjR 2 —1- The area of this annulus is independent of R and equal
to 7t/4. Next one asks whether there are such domains of still
smaller area and which of these domains has minimum surface area.
(8) Using ruler and compasses a given angle may be divided into five
equal parts.
(9) “ Squaring the circle ” on the sphere:— Using a spherical ruler
and spherical compasses one can construct on the unit sphere a
regular quadrilateral which is equal in area to a given spherical cap.
A “ spherical ruler ” is an instrument with which one can draw
on the sphere the great circle through two given points which are
not diametrically opposite. How one actually constructs the circle
t The equivalence by dissection of polygons SP and 2 is defined analogously.
is irrelevant to our problem. It is sufficient to agree that, corres

ponding to the rules for constructions with ruler and compasses in
the plane, it is possible to draw a circle about a (given or already
constructed) point P with radius PQ (Q is another given or already
constructed point) and that one can draw the great circle through
two (not diametrically opposite) points R and S.
Note, however, that both permissible steps of the construction
can be carried out using ordinary compasses. To draw the great
circle through R and S it is certainly necessary that one should
be given the span of the compasses for the great circle (s /l on
the unit sphere). With the compasses opened to a length of \/2
one is able first to determine the “ pole ” T of the great circle R S
and then draw about T, again using the span \/2 , the great circle
through R and S.
(10) The set-theoretical squaring problem'.— The circular disk (€
(x2+ y 2^ 1) and the square ST of equal area ( O ^ x ^ J n ;
O ^ y ^ J n ) may be sub-divided into a finite number n of subsets
and STx in such a way that the subsets with the same
subscript are congruent:—
+ sr = s r x+ y 2 + ..-+ srn,
T V=STV, v = 1,2, ..., n.
One should note that no special assumptions are made about the
subsets <TV and STv. They do not need to be polygons, triangles
composed of circular arcs, etc. Any two such sets and STv are
said to be congruent if it is possible to transform into STv by
congruence transformations (rotation, reflection, parallel displace
ment).
At this point it would be natural to consider also the four colour
problem as an “ unsolved problem We have not done this since
there have recently been good reports on the state of this problem.
Chapter II
REGULAR P A C K IN G S I N THE PLANE

AND ON THE SPH ER E
1. P roblems on the D ensity of P a ck ing

We now formulate in another way problem (1) which was stated on
p. 1. Let us consider first of all n points Pv (v = 1, 2, 3, n)
distributed arbitrarily on the unit sphere. Let the minimum
spherical distance between any two points PVP^ be d. Then it is
possible to draw about the points Pv spherical caps of spherical
radius r = - which have no points in common. By displacing the
points Pv, the minimum distance d is altered and, from continuity

considerations, becomes a maximum for (at least) one arrangement
of the points P v. The surface area of the circles (spherical caps) of
radius r = - drawn about the points P v (v = 1,2, ..., n) for such an
extremal configuration becomes a maximum together with d.

Therefore problem (1) is equivalent to problem (la):—
Problem (la): n congruent spherical caps must be packed without
overlapping on the unit sphere so that they are as large as possible.
The centres of such an extremal system of caps are then the
required capital cities for the dictators of problem (1).
Similarly we can reduce problem (2) to a covering problem. Let
us imagine spherical caps of spherical radius p, covering the whole
sphere t and placed about the points Pv which were introduced as
fuel depots. Then every point on the sphere is within a spherical
distance of at most p from one of the points P v and our problem is
t i.e. more precisely:—Every point of the sphere must be an interior point
or a boundary point of at least one of the caps. Therefore in the covering problem
we consider the caps to be closed whereas in the packing problem (la) we consider
them to be open: caps are allowed to touch each other.
5
to arrange the covering with the spherical caps so that one obtains
a covering of the whole sphere with caps of the smallest possible
spherical radius p. Again let us choose the spherical radius to be
unity so that (2) is equivalent t o :
Problem (2a): The unit sphere must be completely covered by ti
spherical caps of smallest possible radius.
The centres of such an extremal system of caps are then the
positions for the fuel depots mentioned in problem (2).
Now the area of a spherical cap of height h, or of spherical radius
r, on the unit sphere is
J(r) = 2n . \ . h = 2n(l — cos r).
Therefore J(r), and similarly the density of packing
_ n . J(r) n .
D„ = — — = - (1 -c o s /•) (1)
471 2
is a monotonically increasing function of r. Thus the solution of
problem (la) yields a system of caps which is packed on the sphere
and has maximal density of packing, while the solution of (2a) gives
a covering system of caps of minimal density of packing. The density
of packing Dn, defined by (1), for a system of caps has an intuitively
obvious geometrical meaning: 100 D„ gives the percentage of the
surface of the sphere which is covered by the caps. For the packing
problem we have Dn< 1; for the covering problem, on the other
hand, Z>„> 1.
Before we turn to the solution of problems (la) and (2a) for
definite natural numbers n, we want to formulate two simpler
problems, the treatment of which will be useful for the solution of
the original problems:—
Problem (lb): Let there be n congruent, non-overlapping spherical
caps packed on the unit sphere. We must estimate an upper
limit l(n), depending on n, for the density of packing (1) of this
system.
Problem (2b): Let the unit sphere be completely covered by n con
gruent spherical caps. A lower limit L(n) must be estimated for
the density of packing of the system.
We define these limit functions /(n) and L(n) so that, at least
for certain natural numbers n, they are equal to the density of packing
REGULAR PACKINGS (PLANE AND SPHERE) 7
of a well-defined system of caps. Then the original problem is
solved at least for these numbers n when the limit functions have been
determined.
However, we postpone for the time being the solution of problems
(lb) and (2b) and first of all, as a kind of preparatory exercise,
solve the corresponding packing problems in plane geometry.
This digression is in fact more than a preliminary exercise. We
shall see that packing problems in the plane, which are interesting
in themselves, lead to problems which are still unsolved (p. 26).
Let us consider therefore how one can pack a system of non
overlapping, unit circles as densely as possible in the plane and,
conversely, how one can completely cover a finite area of the plane
with as few unit circles as possible. We shall have to give a new
definition using a limit for the idea of density of packing in the plane.
One cannot avoid introducing a suitable limiting process in plane
geometry since the whole plane cannot be covered by a finite number
of circles.
Let {.c,} be a denumerable set of unit circles in the plane Let
# (R) be a circle of radius R about an arbitrary fixed point O of our
plane and n(R) the number of circles of which have an interior
point in common with # (R).
Then
Z J i = n £ i = n . n(R)
R R
is the sum of the areas of these circles.! The limit
lim - —- = lim = lim SR (2)

R-*<x> nR2 R-*oo R~ R->oo
is called the density of packing of the system {^J.
First of all we note that this definition of the density of packing
is independent of the particular choice of point O for the centre of
the circle # (R). Let 0 + be a second point (at a distance d from O)
which is the centre of the circle C
£ +{R) with radius R. If I + denotes
summation over all the circles of {^,} which have an interior point in
common with (&+{R), then
E
R-d
Ja E+
R
J i i E J,
R+d
t Here and in the following, summation over R means that the summation
must be taken over all the circles of our system which have an interior point in
common with # ( R).
8 UNSOLVED AND UNSOLVABLE PROBLEMS IN GEOMETRY
is true and hence
R J n(R + d)2
follows; therefore
lim 2R? _
R-oo 7lR
Now we are able to formulate the problems of plane geometry

which correspond to problems (lb) and (2b):—
Problem (lc): Let be a set of unit circles which have no interior
point in common in a plane 0>. An upper bound is required for the
density of packing o f this system.
Problem (2c): In a plane & let } be a set of unit circles with the
property that every circle V{R) about a point O e & will be
completely covered by finitely many circles of the system {-cj*’}.
It is required to find the lower bound for the density of packing o f
It is very simple to solve these problems if one assumes a lattice

like arrangement of the centres [II 3]. However, we want to avoid
such particular assumptions and we start the solution by deriving
some lemmas which are also important for the solution of further
problems.
2. L emmas
By a plane polygonal net we mean a set of simple, closed polygons
which lie in a plane and which satisfy the following properties:—
1. No two polygons have an interior point in common.
2. Any two polygons may have vertices or edges in common. If they
have edges in common, then they always have one or more
complete edges in common.
3. All free edges | form a single, simple, closed polygon.
This last condition excludes the possibility of using the term “ net ”
for a system of polygons which lie in a plane and whose union forms
a multiply connected domain. On the other hand if, in a case like
t “ Free ” edges of a net are those which belong to only one polygon of the
net.
R E G U L A R P A C K I N G S ( P L A N E A N D S P HE R E ) 9
this, one adds to the system the interior boundary polygons, then
one has a net in our sense. If we denote the number of polygons of a
net by A, the number of their edges (sides) by E and the number of
vertices by V, the number
C{Jf) = V —E + A
is called Euler’s characteristic number for the net J/". We now prove
Theorem 1: Euler's characteristic number for a plane net is equal
to 1:—
C (Jf) = V - E + A = 1. (3)
For the proof we note first that the characteristic number does
not change if one draws an (interior) diagonal in a polygon. When
F ig . 1
this is done the number of vertices remains unchanged while A and

E each increase by 1. By drawing diagonals like this one
can eventually triangulate all the polygons of the net ([II 1] p. 66)
and it therefore suffices to prove the theorem for a triangular
net.
Thus let J f n be a net made up of n triangles (fig. 1). We call
the triangles of jV n which have free edges boundary triangles of
the net (e .g .^ j a n d ^ 2 in fig- 1)- If among the boundary triangles
of the net one occurs with two free edges (e .g .^ j in fig. 1) then, by
separating off this triangle, one obtains a system of n —1 triangles
that similarly fulfil all the conditions required in the definition of a
net. It has the same characteristic number as J f n i.e. CG/fn-j)
= C (jV n), for V and A have decreased by 1 and E by 2. If in
the net only triangles with one free edge occur,! one can also exclude
a triangle in this case without disturbing the connectivity of the
t That is the case e.g. in net in Fig. 1 which is formed from by
removing F i.
remaining system of triangles. If this is done, E and A change by

1 and V not at all.
One can reduce the net in this way until it consists of only one
triangle. And, since the characteristic number of a triangle is
equal to 3 —3 + 1 = 1, our theorem is proved. Obviously it is also
valid if the sides of the polygons are replaced by continuous (non
intersecting) arcs.
Similarly one can construct on the sphere (say from great circle
arcs) polygonal nets. However, in this case there are also systems
of polygons which cover the whole sphere. We denote these
systems, which are important for the geometry of the sphere, as
nets on the sphere. One obtains the exact definition by varying
the definition for plane nets. Plane polygons are replaced of course
by polygons on the sphere and 3. is to be replaced by the rule:—
There are no free edges.
Now let us consider a net like this on the sphere mapped onto the
plane by stereographic projection (see e.g. [II 4]). Let the centre for
the projection be an interior point of a spherical polygon. Then
one obtains a plane net as the plane image of the spherical net.
However, care must be taken not to count the exterior of the polygon
with free edges as a plane polygon, though it is the image of a
spherical polygon, namely the image of the polygon which contains
the North pole of the stereographic projection. Thus there exist
between the vertices, edges and areas (u, e and a) of the spherical
net and the corresponding numbers (V, E and A) of the plane net
the following relations:—
v = V, e = E, a = A + 1.
Hence from (3) follows
Theorem 2: Euler's characteristic number for a spherical net is
equal to 2:
v —e + a = 2. (4)
A large number of geometrical extremal problems can be solved
by use of Jensen’s inequality for convex functions. A function / ( a')
which is defined on an interval [a, b~\ of the .v-axis is called convex
if, for any two points Xy and x 2 of the interval,
(5)
2
REGULAR PACKINGS ( PLANE AND S P HERE) 11
is always true. It is called concave, if, conversely, for any two points
* 1, * 2 ,
v f — 1) a/(x,)+^2). (so
Now for these functions we have
Theorem 3 (Jensen’s Inequality): If f (x) is convex in [a, b~] then,
for any points xv e [a, 6] (v = 1, 2, 3, ..., ri),
/(* .) + /(* 2) + ■+/{*,) a n f b Xl± — -+ (6)
On the other hand for concave functions f (x)
/(* i) + /( * 2) + ... +/(*„) (6-)
It is sufficient to prove (6). The proof for concave functions is

similar. We show first of all that the theorem is true for n = 2m
(m an integer greater than 1) then that, if the theorem is true for n,
it follows that it is also true for n —1. Hence the validity of (6) for
all natural numbers N is assured. In fact, if a number N lies between
2m~1 and 2m, then, since (6) is valid for 2m, it follows that it is valid
for 2m—1, hence that it is valid for 2m—2 and finally that it is true
for N.
For n = 4w e have first of all by repeated application of (5)
« / ( * ,) + /(* 2) +/(*a) +/(*„)) i i + y (~ Y ^ j)
Xi +x~ X i+X.
+
_j.( Xj + x 2+ x3 + x 4
The validity of (6) for n = 2m, m>2, is deduced by a standard

induction argument. It only remains to show that, if it is true for
n, it follows that it is true for n —1. Therefore, let Xj, x2, •••, xn_!
be any points of our integral [a, b~\. The validity of (6) will be
proved for these points assuming that the inequality for any n points
of the interval is true. We add to the n —1 points another point
X i+ X 2+ ...+X „_1
(7)
n —1
12 U N S O L V E D A N D U N S O L V A B L E PROBLEMS IN GEOME T R Y
which also lies in the interval [a, By our induction hypothesis

(6) is true for n points. We write this inequality in the form
f( x i) + f(x 2) + ... + f(xn_ i) ^ n -/(*„)• (8)
If one now multiplies (7) by n —1 it becomes clear that the following

representation is also true for x„:
x„ = Xl + *2— _ j 3 . (7')
n
From (8), using (7') and (7),
/(* 1) + f(x 2) + ••• +/(*„-1) ^ n f ( x n) - J \ x „) =

(n _ ++
This is the inequality we wanted to prove.

We apply Jensen’s inequality to prove the following theorem which
will be used later:—
Theorem 4: O f all the n-gons inscribed in a circle o f radius r the one
with the greatest area is the regular n-gon.
Let us denote (fig. 2) by 2av the angles subtended at the centre C
by the sides A 1A 2, A 2A 3, ... of the polygon A lA 2A 3...A n when it
is subdivided into isosceles triangles by the radii CAj, CA2, ....
Then the area of the whole polygon is
n j.2 n
A = r2 £ sin av cos av = — £ sin 2av.
V = 1 2 V=1
Now, in the interval [0 , n\, sin .r is a concave function. By a
well-known addition formula
sin x t + sin .\2 = 2 sin * l + * 2 5^2 sin * l+ * 2

0 L
Thus, by Jensen’s theorem (6'), it is true that
2a t + ... 4- 2a„ nr2 . 2n
A < — n sin — sm — .
“ 2 n 2 n
On the right hand side of this inequality we have just the area of the
regular «-gon inscribed in the circle.
In this case we used the addition formula to prove the concave
REGULAR PACKINGS (PLANE AND SPHERE) 13
character of sin x. In many cases it is convenient to use the
following criterion:—
A function f (x) which is twice continuously differentiable in the
interval [a, b~\, is convex (concave) if the second derivative f " ( x ) is
positive (negative) in the whole interval.
Since this important criterion is not proved in many text-books on
Analysis we include the proof here. By Taylor’s theorem
m = /(*) + /'( * ) « - x ) + £T (x + S(5 - *))« - X)2 (9)
is true for x e \ a , b \ £e[a,b'\. Now let f " ( x ) > 0 in the whole

interval. Then from (9) it follows that
M ) > J i x ) + f 'M Z - x ) .
Hence, for ^ and £ = x 2, respectively,
/(*v) > /(*) +/'(*)(*v “ x) (v = 1, 2),
and therefore, if X i + x 2 = 2x,
f i x 0 + f(x 2) > 2f(x) + f\x ){ x l + x 2- 2 x ) = 2f{x).
Similarly, one gets (5') iff " ( x ) is negative in [a, b~\.
3. P a c k in g P roblems in the P lane

Let be a set of non-overlapping unit circles in a plane. To
estimate the density of packing of such a system (problem (Ic)) we
can suppose without any loss of generality that the system is
“ saturated ” , i.e. it is not possible to pack another unit circle in the
plane without some overlapping.
We now join the centres of the circles by straight lines and so
form a triangular net whose vertices are all centres of the circles.
This net can be constructed so that there is no angle greater than
120° in any of the triangles. Let us begin to construct the net by

joining any one of the centres Cx with the closest (or one of the
closest) centres C2. Now we consider the circles having interior
points in common with the parallel strip formed by the lines through
Cx and C2 perpendicular to CXC2 (fig. 3). There must exist circles
with this property, otherwise the system would not be saturated.
The number 3 is given to the circle of this set whose centre is
closest to the straight lines. If there happen to be several of these
circles one is chosen arbitrarily. If its centre lies within the parallel
F ig . 3
strip, then the triangle CXC2C2 is acute angled. However, if C3

lies outside the strip and if c3 is cut by the perpendicular through Cl5
say, (fig. 3) then, for the angle e between CXC3 and the perpendicular,
it is true that
The angle C3CXC2 is therefore less than 120°. The net can now be
constructed by adjoining further triangles so that there is no angle
greater than 120° in any of the triangles. Then each triangle in the
net has an angle a for which 60°â<120° is true. Since the sides
of these triangles are at least of length 2, we get for the area J(?F) of
each of these triangles
J ( S T ) ^ . 2 . 2 .sina^V3. ( 10)
We have equality in (10) only for triangles CXCI1CV whose circles
cx, cn> cv touch each other in pairs. For later use we also note a
(quite crude) upper bound of the area of the triangle. Since the
system is saturated it follows that the base and height of the triangles
R E G U L A R P A C K I N G S ( P L A N E A N D SPHERE) 15
must be less than 4. Therefore, we have
J{F )< 8. (10')
Now let us draw the triangular net of the centres of all the circles
which have interior points in common with a circle C S{R) of radius
R and centre O. Then for the quotient 5R which arose in definition
(2) for the density of packing it is true that
V 7i < Vn J( j V)
( 11)
nR2 ~~ A . 7 3 nR 2 ’
where V is the number of vertices of our net, A the number of areas
and J (J f) the area of the whole net for which we have from (10)
and (10')
A . yjZ ^ J { A r)<%A. (10")
We also note that the second fraction on the right side of (11)
converges to 1 as R increases. From the definition of the net it
follows that the distances of its vertices from the centre O of # (R)
are less than R + 1. Since the system is saturated
n(R - 3)2 < J{j V) < tz(R + 1)2, (10'”)
and hence
lim K A0 = 1. ( 12)
R-*00 nR 2
Next we express the number of areas A in (11) in terms of V
using Theorem 1. To do this the edges of our net in each of the
triangles are enumerated including all the free edges once and all
the others twice. Therefore we have
ZA = 2 E - E ',
where E' indicates the number of free edges. Thus it follows from
(3) that
A = 2 V - E '- 2 (13)
and hence (11) becomes
7lV JU )
$r = (14)
(2V —E' —2)yJZ *r2
We will immediately get the desired bound for the density of
packing from (14) when we have proved that
F'
lim - = 0. (15)
K-* co E
For this proof we note that E ' is not only the number of edges but
also the number of angles of the boundary polygon. The area of
the circles corresponding to these vertices must, however, be smaller
than the area of the ring with radii R + 2 and R —4 in which the
circles lie. Therefore
E'n<n(R + 2)2- n ( R - 4 ) 2 = l2 n ( R - \) . (16)
Further, from (13), we have 2 V> A, and thus it follows from (16),
taking (10") and (10"') into account that
E' ^ 127t(jR —1) . 2 ^ \92n{R — 1) ^ 192(K-1)
V < A < J (J f) K ( R - 3)2
(15) immediately follows and from (14), (15) and (12) we obtain
n n
d = lim SR^ lim . 1=
Hence we have proved:—

Theorem 5: The density of packing o f a system {«,-} of non-intersecting
unit circles in the plane is at most equal to U
2^/3
nV n
d = lim _ < = 0-9069.... (17)
2 —
R-> cc 71jR 2V3
It is easy to satisfy oneself that this bound is attained for a system
of circles whose net contains only equilateral triangles with sides
of length 2 (fig. 4).
To solve problem (2c) (p. 8) we again construct a triangular net
from the centres of the system of unit circles. To do this we must
take all the centres of the circles in {■ci1'} which have at least one
interior point in common with a circle # ( R) about O. This is done
as follows:—Let C, and C2 be the centres of two such circles c f and
c2 with non-empty intersection. One vertex P of the common lune
must then be covered by a further circle c f Its centre is denoted by
C3. If P is covered by several circles then any one of them is chosen.
Now, by the same method, further triangles are added to the triangle
C ,C 2C3 so that finally all the centres in question are included.f
t It is possible that after completing the process of adding on triangles there
are still some centres of circles which lie in the interior of the triangles. By joining
lines to the sides of triangles already present these centres are subsequently
attached to the net.
REGULAR P A C K I N G S ( P L A N E A N D SPHERE. ) 17
In this way we obtain a triangular net in which every triangle has
a circumcircle o f radius at most equal to 1. The area of each of these
triangles is therefore, by Theorem 4, at most equal to the area of the
equilateral triangle inscribed in the unit circle:—
J ( S T ) ^ 3.
Therefore for the area of the whole net it is true that
However, the number A of areas of the net is, by (13), less than 2V
and hence we have for the density of packing, taking (12) into
account,
nV nV JfA f) > nV. 4
D = lim lim
nR1 R-> oo j u o nR 2 = 3V3 . 2 . V '
F ig . 5
We sum up in
Theorem 6: Let {r;+ } be a set of unit circles in a plane & with the
property that every circle T(R) with centre O e & is covered by
finitely many circles of the system }. Then we have for the
density of packing of this system
D = iim — ^ ~ = 1-209.... (18)

nR 3^/3
In this case also the extremal value is attained when we have a
packing for which the net of centres contains only equilateral
triangles (fig. 5).
4. P a c k in g P roblems on the S phere
To solve problems (lb) and (2b) we consider, as for plane problems,
a triangular net constructed from the centres of caps. In the
covering problem (problem (2b)) this net can be formed so that the
spherical radius of the circumscribed circle of each triangle is at
most equal to the radius r of the caps of the system. To construct
this net we follow a procedure that is completely analogous to that
for the corresponding plane problem (p. 13).
Then the surface area of a triangle of a net like this is at most
equal to that of an equilateral triangle inscribed in a spherical
circle of radius r. The proof of this extremal property of the
triangles of the net could be carried out using the Jensen inequality.
Instead we want to use the theorem about Lexell’s circle:—
Theorem 7: The geometrical locus of the apexes of all spherical
triangles with equal surface area which have a side AB in common
with a given triangle ABC, is the spherical circle passing through
C and the antipodal points A' and B' of A and B.
To prove this theorem we drop from A, B and C onto the great
circle passing through the midpoints D and E of AC and BC (the
“ middle-line ” *w) the perpendiculars which meet m in G and H,
say, (fig. 6). Then the following triangles are congruent:
ACDF=AADG, ACEF—ABEH,
since in each case they have one side and two corresponding angles
equal. Therefore the triangle ABC is equal in area to the quadrangle
ABHG (right-angled at G and H). t Let Ct be another point on the
t It is easy to see that this and the following statements are also valid for
obtuse-angled triangles ABC.
REGULAR PACK IN GS (PLANK AND SPHERE) 19
hemisphere which is determined by m and contains C and with its
perpendicular CXFX onto m equal to CF = AG = HB = d. Then,
simply by considering congruence, it is shown that m is also the
middle-line for the triangle ABCX. Therefore triangle ABC X has
also the same surface area as the quadrangle ABHG and triangle
ABC. The geometric locus for all points which are equidistant
from a great circle on the sphere, is, however, a spherical circle a,
the “ distance line
All triangles A B C + whose apexes C + lie on /*, have therefore

equal surface area and it is obvious that, conversely, triangles
A BC ++ with C ++ not lying on a have a smaller or greater area than
triangle ABC. It only remains to determine the position of the
distance line. Obviously the points A' and B' which are antipodal
to A and B are the same distance d from the great circle ni as A and
B. Since A' and B' lie on the hemisphere determined by m. and
containing C, the distance line a must also pass through A' and B'.
This completes the proof.
Now let ABC be an isosceles triangle inscribed in a circle s i
and with apex C. Then the Lexell circle passing through C must,
by symmetry, touch s i and clearly on the outside. This means,
however, that among all the triangles inscribed in circle s i and with
base AB the isosceles triangle has the greatest surface area. If ABC
is not equilateral then one can immediately find among the triangles
inscribed in s i a greater one choosing AC as base. From continuity
considerations there must be a greatest triangle among all those
t If one calls the given great circle “ the equator” of the sphere, then the
distance line is a circle of latitude (not a great circle!) lying in the northern or
southern hemisphere.
inscribed in stf. Hence the equilateral triangle must be this greatest

triangle. We have therefore:
Theorem 8: O f all the triangles inscribed in a spherical circle the
equilateral triangle has the greatest area.
Using this theorem we can now estimate the density of packing
for the given system of caps (problem (2b)). By Theorem 8 the
area of the triangles of the net is at most equal to the surface area
Sm of an equilateral triangle inscribed in the circle of spherical
radius r. Therefore let us calculate Sm.
Let L M N be an equilateral triangle with circumscribed circle of
radius r and angle 2a. Let U be the centre of the circumscribed
circle and O the foot of the perpendicular from U onto LM. Then
the triangle LUO has at U the angle n/3, at O the angle n jl and
n
cot a = cos r . tan
3
therefore
S,,, = 6a —;i = 6 c o t" 1 (^/3 cos r) —7i. (19)
Now let v, e and a again be the numbers of vertices, edges and
areas of our spherical net. Since it covers the whole sphere, we have
the surface area of this net
4ti = S (J T )^ a . Sm. (20)
Since the net consists only of triangles then 3a = 2e and from (4) it
follows that
a = 2v —4. (21)
Substituting (19) and (21) into (20) it follows that
6 cot-1 (J 3 cos r) —n> 'K—,

v “ 2v —4
or
cos r< — cot — —— . (22)
- ^3 3(2d —4)
Now we again replace the number of vertices v by the (equal)
number n of spherical caps of our system and from (22) obtain for
the density of packing defined by (1):—
Theorem 9: For the density of packing D of a system of n congruent
spherical caps covering the unit sphere
mt
1— — cot co, co„ (23)
V'3 6 ( n - 2)
R E G U L A R P A C K I N G S ( P L A N E ANI ) S P HE R E ) 21
There is equality in (22) and (23) only when all the triangles of the
net are equilateral. However, this is possible only if the n centres
are the n vertices of a regular triangular polyhedron inscribed in
the sphere. Therefore, there exist such (regular) packings (by the
known theorems about regular polyhedra, see e.g. [II 7]) for the
numbers n = 3, 4, 6 and 12. The bound (23) can be attained by
a suitable packing only in these cases. We set down in a table
the values for the smallest radius of the cap and the corresponding
density of packing which arise in these cases:—
n 3 4 6 12
cos r 0 } I v'3 W 3 + 1 /5
1
r 90° 70° 52' 54° 46' 37° 43'
D 1-5 1-33... 1-268... 1-254...
Hence the problem (2) is also solved for the cases n = 3, 4, 6 and
12. The situations for the fuel depots are the vertices of a regular
triangular polyhedron with the corresponding number of vertices.
Theorem 9 was first proved by F ejes Toth in 1943 [II 2] (by other
methods).
To solve the problem (la) (p. 5) one could try to carry over the
procedure used in the proof of Theorem 5. However, we choose to
describe as many different kinds of methods of solution as possible
and therefore the theorem of Jensen will be used this time. Thus,
let there be given a system of n congruent, non-overlapping spherical
caps of spherical radius r on the unit sphere. We assume that the
system is saturated and that therefore no more caps of radius r can
be packed without overlapping. Then the tangential planes touching
the sphere at the centres Cv (v = 1,2, 3, ..., n) of the caps form a
polyhedron whose central projection on the sphere produces a net
J f on it. If one joins the centres of the caps by arcs of great circles
which intersect the edges of the net jV at right-angles then one
obtains a second net J T which contains only triangles (fig. 7 shows
the corresponding picture in the plane).
Let v, e and a be as usual the numbers of vertices, edges and areas

of the first net J f , v', e' and a' the corresponding numbers for J f ' .
Then it is obvious that e = e', v = a', a = v'. Since J f ' is a
triangular net, we have moreover that
2e' = 2 e = 3a' = 3v. (24)
The polygons of J f can now be decomposed into isosceles
spherical triangles by joining the n centres to the vertices of the
F ig . 7
polygons. These triangles are then divided by the sides of J f ' into
right-angled triangles so that at each of the e edges of J f there are
4 right-angled triangles (fig. 7). In the right-angled triangles let
av be the angles at the centres, /?v the angles at the vertices of the
polygons and bv the sides opposite the /?v. Then the area of the
right-angled triangle with index v is:—
A(av) = aV+ 0V-
Since
it follows that
A(av) = av— sin-1 (cos bv sin av). (25)
Since bv^.r, we have for all v
A(av) ^ av— sin-1 (cos r sin av).
R E G U L A R P A C K I N G S ( P L A N E A N D S P HE R E ) 23
Now the 4e triangles just cover the whole sphere. Thus
4e 4e
4n^ ]T (av—sin-1 (cos r sin av)) = £ s(av) (26)
V = 1 V = 1
is true, where the function

5(a) = a — sin~1 (cos r sin a)
gives, by (25), the area of a right-angled spherical triangle with one
angle a and the adjacent side r. Now, using the criterion on p. 13,
one immediately establishes that 5(a) is a convex function for
Oâ^rc/2. Of course, j"(a) is positive as one can immediately
verify.! Applying Jensen’s theorem (Theorem 3), it follows from
(26) that
Tin . -i / . %n
4%^4e s \ ^ - \ = 4e ------sin cos r sin — (27)
2 c V 2 e,
since
a i + a2 + ... + a4e = 2n . n.
However, from (24) and (4)
e — 3{a —2) = 3(n —2)
and hence from (27) we have
7in . _1/ .Tin
4n'^\2(n —2) —sin cos r . sin
6(/i —2) 6(« —2),
or
n , . _i Tin
- ^ sin cos r . sin
6 “ 6( « - 2).
therefore
1___ Tin
cos r> oj„ = (28)
2 sin co„ 6( « ~ 2)
If one substitutes this estimate into the definition of the density
of packing, one has at once
Theorem 10: For the density of packing
d„ = -(1 —cos r)
2
t One can immediately derive the convexity of j(a) from the geometrical
meaning of this function. The details of the reasoning are left to the reader.
24 U N S O L V E D A N D U N S O L VAR I. E PROBLEMS IN GEOMETRY
of a system of n congruent, non-overlapping spherical caps of

spherical radius r
Tin
(29)
6(« —2)
is true.
Here too the equality sign is possible only when all the triangles
of the net J T are equilateral. If Q„ is the half-angle and 2r the
side of such a spherical triangle then we have
sin r 1
sin Q, (30)
therefore sin 2r 2 cos r
1
cos r = --------- .
2 sin Q„
This angle is just equal to the angle a>„ of (28). For the area of
our equilateral triangle is J(A) = 6Q„—n. If a' = 2(n—2) of such
triangles cover the whole sphere, we have
2(n - 2)(6Q„ - n) = 4n, Q„ = co„.
The converse is left for the reader to prove. Thus we have given a
geometrical interpretation for the angle con appearing in (23) and
(29) when n is the number of vertices of a regular triangular
polyhedron.
We note further that, when the inequality holds in (28) and (29),
>a. (31)
6(n —2)
In fact, in this case we have
1
sin con> = sin Q„,
2 cos r
therefore a>„>Q„.
Again we collect the results in a table for the cases n = 3, 4, 6 and
12:—
n 3 4 6 12
cos r i iv /3 W2 0-852
r 60° 54° 4 6 ' 45° 31° 4 3 '
d 0 -7 5 0 -8 4 5 0 -8 7 8 0-896
R E G U L A R P A C K I N G S ( P L A N E A N D SPHERE) 25
Thus the problem (1) is also solved for some values of n. In the cases
n = 3, 4, 6 and 12, just as for problem (2), the vertices of a regular
triangular polyhedron inscribed in the sphere form the point system
which is the solution of the problem. We shall see in the next chapter
that for other values of n the solutions of the two problems are by
no means always the same.
Another form of Theorem 10 is important for some problems:—
Theorem 10a: Given n (n> 3) points on the unit sphere there is always
a pair of points whose distance apart is
<5 < 4- (32)

sin2co„
By the distance <5„ between two points P and Q on the sphere we
mean here the length of the line-segment PQ. The spherical distance
corresponding to Sn is S* = 2 sin-1 —.
The bound (32) can easily be derived from (28). Obviously from
(28) it follows that
1
sin2 con=> , 2 ’
4 cos r
therefore
_ i / cot2 CO„ — 1
(5* = 2r < cos‘
From this we calculate the limit given in (32). In fact

. _ . 2 c*. cot2 co —1
1 —2 sin2 — = cos 5 ---------— ,
2 2
therefore
c* i
—4 sin2 — —> — -— - 4
2 sin2 co
or
= 2 sin — ^ 4 - - ^ -
2 v sin co„
5. Further Problems
To conclude this chapter we examine some unsolved problems
which are closely related to those already considered.
A packing of non-overlapping circles in the plane | is said to be

rigid if every circle is touched by at least three other circles so that
the points of contact on any of the circles are less than a semi-circle
apart. The packing shown in fig. 4 is certainly rigid: for every
circle is enclosed by six others and every circle need only be touched
by three others as described above to make a displacement of the
circles (without overlapping) impossible.
A natural example of a packing of circles in the plane in which
every circle is touched by three others is shown in fig. 8. The lines
from one centre to the centres of the neighbouring circles form the
edges of a trihedral whose arms enclose angles of 120°. All the
edges of the trihedrals form a net of hexagons. From this observa
tion we immediately obtain the density of packing of the system.
In each hexagon (e.g. P XP2P3P^P5P 6 in fig. 8) there are six circular
sectors with angle 120° and therefore two complete circles. Hence
the density of packing J of the system is
dt = ■2n ‘ -1 = — -0-6046. (33)

6 .- 7 3 V3
4v
However, this system SPx has by no means the smallest density

of packing of all rigid systems. One immediately obtains a system
PP2 with smaller density if one replaces every circle of SP^ by three
congruent circles which touch the first circle at its points of contact
(see < Pi in fig. 8). If, in addition, one requires that every two
of the three smaller circles must touch each other, then the
new system SP2 of smaller circles is uniquely determined. It is
also rigid in the sense of our definition since every circle < P 2 eSP2
is touched by three others, namely by two which belong to the same
circle of the system PPl as (€2 and by one from the neighbouring
circle. In fig. 8 the net of the centres o f ^ 2» consisting of triangles
and dodecagons is drawn in part of the figure. The edges of the
trihedrals of this system enclose an angle of 60° and two angles of
150°.
t Of course the corresponding definition is possible for a system of non

overlapping spherical caps on the sphere.
t It follows from an estimate, obtained similarly to that on p. 15, that the
density of packing as defined in (2) can be evaluated in this way from the covering
of the single hexagonal cell.
R E G U L A R P A C K I N G S ( P L A N E A N D S P HERE) 27
We calculate the radius p of the circles SP2, knowing the radius
r = 1 for the large circles, as follows: —
= sin 60° = p = r(2 ^ 3 -3 )-0 -4 6 4 . r. (34)

r-p
^np2
Therefore, since three circles of Sf 2 only cover the fraction----- of a
71
circle ^ we get for the density of packing of 2 from (33)
and (34)
d2 = 3p2d l = (7V 3-12)tt-0-393.
Probably this is the smallest possible value for the density of a rigid
packing of circles. In any case it is not possible to repeat the same
game of replacing a circle by three smaller ones. We know that
the angles formed by the edges of the trihedrals of 2 are 60°, 150°,
150°. Therefore it is not possible to place at each of the three points
F ig . 8
28 U N S O L V E D A N D U N S O L V A B L E P ROBLEMS IN GEOMETRY
of contact of an y 2 circle one of three congruent circles touching

the <$^2 circle and touching each other in pairs.
Other unsolved problems are referred to briefly in a publication
of A. H e p p e s [II 11]. If every point of the plane is covered by
at most k (congruent) circles, then we speak of a k-fold packing of
circles. If every point belongs to at least k circles then there is a
k-fold covering of the plane. Let the corresponding densities of
packing be bk and Ak and the extremal values
dk = lim dk, Dk = lim Ak.
Then from our results we have
n 2n
D,
n/27
Further it is obviously true that
k d l ^ d k< k, k D ^ D k> k. (35)
One obtains, for example, a k-fold packing of circles of density
kdl when one replaces each circle in the simple extremal packing by
k copies placed over one another. However, it is easy to see that
there exist k-fold packings of circles for which the density will be
greater that kd±. A. H e p p e s gives a simple example of this in his
work. Hence the equality sign cannot hold in the first inequality
in (35).
There still remains the open question:—
How large are the numbers
Chapter III
IRREGULAR PACKINGS AND COVERINGS
1. The Existence Theorem

Let us now deal with the solution of problems (la) and (2a) (pp. 5, 6)
for values of n for which n is not the number of vertices of
a regular triangular polyhedron. The corresponding packings of
spherical caps on the unit sphere and coverings of the unit sphere
by such caps are called irregular.
A simple consideration of convergence shows that both the
extremal problems stated must have (at least) one solution for each
integer n. For the packing problem the radius of the cap is in fact
bounded from above by II (28). There is therefore an upper bound
rn for the (spherical) radii of n congruent spherical caps which can
be packed on the unit sphere without overlapping. Let rnv be a
subsequence, converging to r„, from the set of radii of caps corres
ponding to all possible configurations of n points of the unit sphere.
Then there exists a subsequence r 'v of rnv such that all the sequences
of n centres C'Nv (N = 1, 2, ..., n) converge, say to Cl5 C2, •••, C„.
It is now easily seen that the spherical caps of radius r„ drawn about
these points CN (N = 1 ,2 ,...,« ) can have no interior points in
common. Moreover, for reasons of continuity, the systems of caps
with centres C'Nv (and the corresponding radii r'Nv) must have such
overlappings from a certain number v onwards.
Hence it has been shown that there exists a system of caps with
extremal radius for the integer n. The existence of a solution for
the covering problem can be established in a similar manner.
We may now find the extremal distributions of points on the
sphere and determine the corresponding radii rn for individual
integers n. Since the problem has already been solved for the
integers n = 3, 4, 6 and 12, let us consider first of all the integers
5, 7, 8, 9, 10, 11.
29
We start with the packing problem and first deal with the case
n — 5. It is obviously true that
( 1)
for the limiting radius r n of the cap corresponding to the number

n and therefore from the table onp. 24 we have:—54° 46'2:/-5^ 45°.
We want to show now that r 5 = r6 - 45°.
If three of the live centres of a packing of five non-overlapping
caps of (spherical) radius r are joined by great circles, then we have
a spherical triangle to which we can apply the cosine formula of
spherical trigonometry:
cos c —cos a cos b = sin a sin b cos y. (2)
Let us now assume that r is greater than 45°. Then the sides a, b and
c of the triangle would be greater than 90°, since the caps of radius
r must not overlap. Therefore the left, and hence also the right
side of (2), would be negative. Since sin a and sin b are positive,
cos y would be negative and therefore y an obtuse angle. However,
this would mean that all the angles in all the triangles which one
can construct from the 5 centres Cv (v = 1, 2, 3, 4, 5) would have
to be obtuse. But this is impossible:—If one joins, say, to the
4 other centres then one would have 4 obtuse angles at Cx. Therefore
the supposition r>45° is wrong. Thus we have r s ^45° and, from
(1), r 5 = 45°.
Thus we have obtained an astonishing result:—The radius of the
caps for n = 5 is no greater than that for n = 6. From this we
obtain an extremal packing for n = 5 by drawing caps of radius
r 5 = r 6 round any 5 of the 6 vertices of an octahedron inscribed in
the sphere.
2. The Packing Problem G raph

Now suppose we are given a system of n (/i^ 7) points on the unit
sphere with minimum spherical distance apart 2r. We introduce the
concept of the graph in order to determine the conditions under
which such a system of points represents a solution of our problem.
The graph is a system of points and lines (arcs of great circles) on
the sphere. The following belong to it:—
1. The n points of our system Cv (v = 1, 2, 3, ..., //),
2. all lines CVC,( for which the (spherical) length CVC„ = 2r.
IRREGULAR P A C K I N G S A N D C OV E R I N GS 31
First of all we note that the lines of a graph cannot intersect each
other in interior points. Obviously, if the lines CVCV+ 1 and t
had a point S in common, then the following would be true:—
4r ^ CVC„ + \ + CflCv+! < CVS + SCp +! + C^S + SC v+i
= CVCV+1 + CMC^ + 1 = 4r.
But this is a contradiction.
If X lines intersect at a point Cv, then we say that the point is o f
order, X. Isolated points are then of order 0.
F ig . 9
If a graph contains points of order 1 or order 2 then, by displacing

individual points on the sphere, one can get a new graph which has
fewer lines (for the same r). Let there be a point, say Ct (fig. 9), of
order 1 which is joined by a line of the graph to one point Cv . Then
one can displace with its cap a small distance in the opposite
direction to C,CV without the cap overlapping any other cap of the
system. Obviously the same is possible for points of order 2 (C2 in
fig. 9):—In this case we determine the admissible direction for the
displacement by the angle enclosed by the perpendiculars at C2 to the
two lines of the graph. One can carry out a small displacement of C2
(with cap!) within this sector without the “ misfortune ” of overlap
ping another cap. Such a displacement is not possible for the point
C3 (of order 3) in fig. 9 since the cap with centre Cv makes a
displacement in the direction of displacement permitted by the two
other caps impossible.
If this process of displacement can be continued until all the

points are isolated then we have obtained from the original point
system 9 a new system 9 ' also consisting of n points but in which
the least distance between any two points is greater than 2r. This
means that the given system 9 is not a configuration having
maximum density of packing.
We now introduce the term an irreducible graph for a connected
system of lines of a graph in which no further displacements are
possible. Therefore to solve the extremal problem in question we
have to investigate point systems with irreducible graphs. For these
investigations we need a f
Lemma: I f an irreducible graph contains a polygon with g vertices
then 2gr<360°.
To prove this we form a spherical triangle with the same perimeter
as the polygon. This is done most simply by imagining our polygon
to have rods as sides, these being circular arcs of length 2r and
curvature 1. We can now displace this polygon of rods so that we
obtain from it a triangle whose sides are multiples of 2r. Its peri
meter is the same as that of the polygon, namely 2rg. However,
since the perimeter of every spherical triangle is smaller than that
of a lune which is not convex, it is therefore smaller than 360°;
and hence the lemma has been proved.
In particular, for g — 5 we have 2r<72°. If it is established now
for a number n that there is an irreducible graph with 2r>72° then
it is also established that there cannot exist a pentagon in this graph
and, even more so, in the irreducible graph with greatest radius r.
3. A Point System for n = 7

To solve our extremal problem for n — 7, we mention first of
all a particular configuration of 7 points on the unit sphere for
which 2 r > lT . We shall see later that this configuration already
represents the solution of our problem. Hence it is expedient to
show first of all the existence of a particular solution with 2 r> ll°.
For we obtain from it an important result:—From the corollary
just deduced from the lemma, there cannot be any polygons with
more than 4 vertices in the graph of an extremal configuration (for
which, of course, 2rj^77° >72°).
t In llie following we measure the angles and hence the surface areas o f
spherical polygons in degrees.
I R R E G U L A R P A C K I N G S A N D C OV E R I N GS 33
To draw the 7-point graph we start from one of the regular
tetrahedra inscribed in the unit sphere. The projection of the
tetrahedron from the centre of the sphere yields a partitioning of
the unit sphere into 4 congruent, equilateral spherical triangles.
We call the centre of one of these 4 triangles the south pole of the
unit sphere.
Now we replace the tetrahedron by another figure:—We start
again from an equilateral triangle ABC whose centre must lie at
the south pole. However, its sides 2r must be smaller than the sides
F ig . 10
2rA of the tetrahedron. At each of the three sides of the triangle we

place a congruent, equilateral triangle: ABE, BCF and CAD. Then
the three points D, E and F do not coincide at the north pole of the
sphere this time, but they have the same geographical latitude. If
one chooses the base 2r small enough then the lines DN, E N and
FN will be greater than 2r. Since the lines were DN = EN = FN = 0
in the case of the tetrahedron, between 2r = 0 and 2r = 2r 4 there
must be a value of r for which the lines DN, E N and FA have length
2r exactly. The 7 points A, B, C, D, E, F and N obtained in this
way form the special point system Sf 7 which we now want to con
sider. Fig. 10 shows y n in stereographic projection.!
We have joined the 7 points by lines so that the whole sphere is
partitioned into 4 equilateral triangles and 3 rhombuses. Also,
since the opposite angles are equal in a spherical rhombus, we have
/_DNE = /_ENF = CFND = l_EAD
= LF B E = /_DCF = 120".
t In connection with this see Chapter X, section 1.
34 U N S O L V E D A N D U N S O L V A B L E PROBLEMS I N GEOMETRY
If we denote the angle of the 4 equilateral triangles by a, then it

follows for the sum of the angles at A, B or C that 3a+ 120° = 360°,
therefore a = 80°. From this we calculate 2r :—In the half equi
lateral triangle with side of length 2r and angle a = 80°, we have by
the rules of spherical trigonometry
sin - = -Sm V = — , 2r~77° 52' 10".
-
2 sin 2r 2 cos r
Therefore for the (still unknown) extremal radius of the cap for
« = 7 it is true that
2r?^ 2 r~ 7 7 ° 5 2 ' 10".
Therefore, by the lemma on p. 32 the only polygons in the extremal
graph are triangles and quadrangles. Now we still want to add that
there cannot be any isolated points of the graph in the interiors of
the triangles and quadrangles. To see this, we have only to observe
that every point in the interior of an equilateral spherical triangle
or quadrangle (with side 2r<90°) is at a distance of less than 2r
from the vertices. For the triangle this result is trivial. For a
rhombus K LM N with KL = 2r<90° the circles with radius 2r about
K and M are not great circles', therefore they overlap the whole
rhombus and every point in the interior of the rhombus is at a dis
tance of less than 2r from the vertices. Such a point cannot belong
to the graph of a packing system of caps of radius r. Hence, for
the case n — 7, we may assume that the extremal graph contains no
isolated points and consists only of triangles and quadrangles.
4. T h e P a c k i n g P r o b le m f o r n = 7
Having completed the preparatory work we can turn to the solution
of our extremal problem for n = 1. The required graph contains
only triangles and quadrangles: it has no isolated points. Not all
of its points can be of order 3, otherwise we would have for the
number of edges e the relation 3.7 = 21 = 2e. This is impossible
since e must be integral. Therefore there must be at least one of
the 7 given points which is of higher order. However, points of
order 5 cannot appear in an extremal 7-point graph. Of 5 lines
starting from a point at least two would enclose an angle less than
72°. However, in our example given in the preceding section, the
angle of the equilateral triangle was already 80° in magnitude.
Therefore the equilateral triangle in an extremal graph must have
an angle which is at least equal to 80°. An isosceles triangle with
equal sides of length 2r and an angle at the apex which is at most
equal to 72°, would therefore have base less than 2r. Hence we
would have two points in the graph which would intefere with each
other. Therefore we exclude points of order 5 (and of course points
of higher order).
Now let A be a point of an extremal graph of order 4. The 4 lines
of the graph starting from A may lead to the points B, C, D and E
(fig. 11a).
F F F
a b ! c
F igs , l l a - l l c
The graph contains two other points F a n d G. Since they are at

least of order 3, F and G are joined to at least two of the points
B, C, D and E. However, these points cannot be opposite each
other. Obviously, if one joined, say, F with B and D, then (since
intersections are impossible) C (or E) would be a point of order 1.
However, such points do not exist in an irreducible graph. There
fore let F be joined to B and C (fig. 11a). Then, from F (being a
point of at least third order) another line leads to the 7th point
G. G must then be joined to at least two of the 4 points B, C, D
and E, which are not opposite each other. Therefore, it would be
possible to take C and D, D and E, or E and B.
Let us consider the first case. Since B, E and D must be at least
of order 3, the lines BE and ED would also have to be drawn into
the graph (fig. lib ). However, in this way, we obtain a graph with
a pentagon (BEDGF), which is not divided by diagonals belonging
to the graph. Since the lines of the extremal graph are certainly not
greater than 90°, the diagonals FE, BD etc., lying in the interior
of the pentagon, have length greater than 2r. Since a pentagon
cannot appear in the graph, this case is excluded. This is also true
for the pair of points E and B, hence G must be joined to the points
D and E (fig. 1lc). Since only triangles and quadrangles may appear,
we must also include the lines BE and CD.
This graph has still one degree of freedom: the angle BAC is not
yet determined. Certainly it is clear that it cannot be smaller than
that of the equilateral triangles of the graph. Otherwise the points
B and C would interfere with each other: their distance apart would
be smaller than 2r.
Let us denote by j6 the (still undetermined) angle BAC and by a 7
the angle of the equilateral triangle. Then it is true that
a7^ /?^3 6 0 °—3oc7, (3)
since the angle EAD = = 360°—2a7 —/? also may not be less
than a 7. Now, if one divides the rhombus ACFB by its diagonals
into four right-angled, spherical triangles, one obtains from these
triangles the relation
tan - = tan 2r . cos -
2 2
for the diagonal d = AF. Now, a simple calculation shows that

the second derivative of
d = /(/?) = 2 tan-1 ^tan 2r . cos
is negative and therefore the function/ (/?) is concave (by the criterion
on p. 13). Likewise dx = AG = /(/?j) is a concave function of p x.
But
(h = /i(/0 =/(360°—2a7—/?)
is also a concave function of fi, as one can see by the same criterion.
Since the sum of two concave functions is also concave,! FG =
d + d x = d2 is a concave function of /?. Because a function which is
concave in an interval [a, 6] does not assume its minimum in the
interior of the interval, FG = d2(fi) does not become the minimum
for a value of ft in the interior of the interval (3).
Let us now assume that we are given an extremal graph for n = 7
with an angle ft from the interior of (3). Let us imagine the lines of
t This follows from the definition II (5').
this graph, with the exception of the line FG, to be formed in practice
by a system of rods on the unit sphere. By a small change of the
angle f = / _BAC one can now change this system of rods so that
d2 = FA + AG becomes smaller and therefore the complementary
line FG (belonging to the graph up till now) becomes larger. This is
possible because, as was just established, d2(fi) does not have a
minimum in the interior of (3). The graph is disrupted by this
displacement, for the line FG is now greater than 2r. However, one
can now construct a new system of lines with a greater side 2rl
retaining the angle ft. From continuity considerations there must
be a value ry >r for which the complementary line FG is just equal
to 2rx. Then, however, one has again obtained a graph as pre
scribed with side 2ri >2r. But this means that the original graph
given was not extremal.
Hence it follows that we obtain an extremal graph for n = 7 only
if we put the angle (I (or ^ respectively) equal to a 7 in one of the
two rhombuses. Then this rhombus consists of two equilateral
triangles and hence the diagonal (BC or ED respectively) of the
rhombus must still be drawn to make the graph complete:—This
diagonal joins the points which are exactly at a distance 2r apart.
If, in addition, in fig. 11c we plot the diagonal BC, we obtain just
the schematic representation of the graph already represented in
stereographic projection in fig. 10. In fact:—The equilateral triangle
ABC is adjoined in both cases by three congruent, equilateral
triangles ABE, BCF and CAD. The point denoted in fig. 10 by N
is called G in fig. 11c, otherwise the notation is the same. Hence it
has been shown that the graph introduced in section III 3 solves
our extremal problem.
We have now solved the problem (la) and hence also problem (1)
of p. 1, for a series of numbers n and we want to summarise the
result:—
Theorem 11: The problem of arranging n points on the unit sphere
so that the distance between any two of the points is as great as
possible is solved for the numbers n = 4, 6 and 12 by the vertices
of the regular, inscribed, triangular polyhedra (tetrahedron,
octahedron, icosahedron). For 5 points one obtains the maximum
for the minimal distance apart if one places them at 5 of the 6
vertices of an octahedron. For n — 7, the extremal graph consists
of four equilateral triangles (with angle a 7 = 80°) and three
rhombuses {fig. 10).
38 U N S O L V E D A N D U N S O L V A B L E PROBLEMS IN GEOME T R Y
We can describe the solution discovered by Schutte and van der

Waerden [III 1] for n = 7 in the language of mathematical
geography as follows:—
One puts an equilateral triangle with angle a 7 = 80° on the
sphere so that its centroid coincides with the south pole and adjoins
to each of the three sides of the triangle another congruent triangle.
Then the six vertices of these four triangles together with the north
pole form a 1-point configuration which solves the problems (1) and
(la) for n = 7.
To discover the extremal graph for larger values of n becomes ever
more complicated. The solutions for n = 8 and n = 9 have been
stated by Schutte and van der Waerden. There exist conjectures
([III 1] and [II 7]) for other numbers.
Later, we shall collect the results known up to the present time in a
table (p. 48). We now turn our attention to problems (2) and (2a).
5. The Covering Problem G raph

The covering problem (problems (2b) and (lb) of p. 6) has been
solved up till now only for numbers n which are the numbers of
vertices for regular triangular polyhedra (corollary of Theorem 9).
To find the solution for any number n, we start from an arbitrary
covering of the unit sphere by n congruent spherical caps (of spherical
radius r). If every point of the sphere is covered by at least one
interior point of the system of caps, it is obviously possible to replace
the system of caps by another, with the same centres but somewhat
smaller spherical radius r'(r' <r), which is also a complete covering
of the sphere. Then the original system given was not one of
minimal density of packing. Therefore, if we are looking for a
covering of the sphere by n caps with spherical radius as small as
possible, we can limit ourselves to those cases for which certain
points of the sphere are covered only by (3 or more) boundary
circles of the caps. In the following we shall call such points of
intersection “ I-points ” . For the centres of the caps we introduce
the abbreviated notation “ C-points
Now we join (fig. 12) the I-points of a covering of the sphere to thb
pertinent C-points by (spherical) lines of length r. In this way a
“ trihedral” (see in fig. 12) or a “ tetrahedral” (I2), “ penta-
t In Fig. J2 and the other figures of this chapter we indicate the I-points by
small circles and the C-points by large dots.
I R R E G U L A R P A C K I N G S A N D C OVERI NGS 39
hedral ” etc. is associated with each I-point. The configuration of all
such lines of length r is called a graph associated with the spherical
covering.f If there exist several trihedrals (/n-hedrals) in a covering
then it is possible that they combine together to form polygons.
The number of their vertices is always even since C- and I-points
alternate in the polygon (fig. 12).
Here also we can try to reduce the graph by a suitable transforma'
tion of caps and if possible reduce it completely. If this succeeds
then the given covering and the one obtained from it by displacement
are certainly not extremal. In this case our problem will again be
to discover coverings which have an irreducible graph.
Trihedrals which at all three C-points do not adjoin another
trihedral (m-hedral), can be eliminated by a displacement. If one
displaces for example the cap with centre Cx in fig. 12 in the direction
of the I-point I u we obtain at 7t a small spherical triangle which is
covered three times over. The covering of the sphere is not dis
turbed at any other place for a sufficiently small displacement.
In this way we obtain a new covering which has the same density of
packing and for which the graph has become smaller: the trihedral
f This definition of the graph differs from that which Schutte introduced in
his work [III 2] on this problem.
40 unsolved and unsolvable problems in geometry
at Tj must be cancelled since I j is not an I-point for the new covering.

It is also possible to reduce the graph by a displacement if a C-point
of order 2 appears in a graph, e.g. C2 in fig. 12. The two I-points
to which C2 is joined in the graph, are indicated in fig. 12 by / 3 and
/ 4. Now, if one displaces the cap with centre C2 into the interior
of the angle formed by the perpendiculars to C2/ 3 and C2/ 4 at C2
then for a sufficiently small displacement the complete covering of
the sphere is not disturbed but the points / 3 and / 4 are no longer
I-points. Again this causes a reduction of the graph.
From all this we see that an irreducible (therefore no reduction
possible) graph has the following properties:—
1. It is made up of lines of length r.
2. The lines form polygonal nets with even numbers of vertices.
3. The vertices of the graph are at least of order 3.
4. The lines of the graph do not intersect each other.
One can prove the last property as follows: If two lines Cv/ v
and CJp had a point of intersection T then we would have
Cv/„ + CM/V< C J + TI^ + C J + 77v = Cv/ v+ C ,/, = 2r.
But this means that at least one of the lines Cv/„ and C„/v is smaller
than r. Then / v (or 1^) would be covered by the circle about (Cv)
and would not be an I-point.
6. T h e C o v e r in g P r o b l e m f o r n = 5
Now let us find the graph for an extremal covering of the sphere
by 5 congruent caps. It is obvious that a hexagon cannot appear
in a graph with 5 C-points. All the polygons must be rhombuses
and all C-points are of order 3 or 4. If a C-point of order 4
appears then already all 5 C-points appear in the 4 rhombuses
meeting at Cx (fig. 13—a schematic diagram).
Then, from the 4 opposite vertices C2, C3, C4, C5 of the first
C-point Cu there are only lines leading to an I-point we can imagine
lying at the north pole of the sphere. Hence the sphere is partitioned
into 8 rhombuses.
However, a C-point of order 4 need not necessarily occur in the
graph. If we start from a C-point Ct where three rhombuses meet
(fig. 14)—C1I 1C2I2, and C1/ 3C3/ 2—, we can adjoin to
this “ s ta r” three further rhombuses (with I-points / 4, I 5 and / 6)
without using new C-points. Since only one C-point remains at

our disposal, the lines from I4,1 5 and / 6 must lead to the fifth C-point
which we again imagine at the north pole.
First of all we deal with the symmetrical graph of the type
represented in fig. 14. Let
L h C J 2 = L h C x h = ••• = Z-/4C5/ 5 = Z / 5 C 5 / 6 = 120°.
Then the 6 rhombuses meeting at Cx and C5 are congruent and
their diagonals CXC2 = C2C5 = ... = 90° in magnitude. The
three rhombuses meeting at Cl are each divided by their diagonals
into 4 congruent right-angled triangles with hypotenuse r, an angle
of 60° and the other arm of this angle 45°. Hence one determines
at once
1
tan r = = 2, r~63° 26'. (4)
cos 60c
In the general (non-symmetrical) case we have the rhombuses with
angles 2av (v = 1, 2, 3; v = 4, 5, 6) meeting 3 at a time at Cx and
C 5. Further, let
/_ IVCV+1I 3+V= 2a6+v, v = 1, 2, 3.
Finally, we denote the other angles of these 9 rhombuses by
2/?v (v = 1, 2, ..., 9). Then the following are true:
2 t «V= 2 £ av = 360° (5)

v = 1 v = 4
and
2 t «v + 4 E av = 3 . 360°,
V = 1 v = 7
therefore
2 £ av = 180°. ( 6)
v = 7
The 9 rhombuses of the graph are now decomposed into 18

triangles by the diagonals joining the C-points. Since
cot av• cot /?v = cos /• (v = 1, 2, ..., 9),
we can write the surface area of such a triangle as follows:
a(av) = 2(av+ cot-1 (cos r . tan av) ) —n.
Hence it follows that
47r = —1871 + 4 Y, (av+ cot 1 (cos r • *an av))- (7)
V = 1
Now the second derivative of o(av) is negative for av<90°, /• <90°:

4 cos r(cos2 r —1) sin 2av
« > v ) = < 0.
(cos2 av+ sin2 av cos2 r)2
Hence the function o(av) is concave and we can apply Jensen’s
theorem (Theorem 3) to each of the sums from 1 to 3, from 4 to 6
and from 7 to 9 in (7). Then, by (5) and (6), it follows from (7) that
2271 < 1071 + 24 cot 1 ( cos r . tan - ) +12 cot 1 ( cos r . tan - ).
I R R E G U L A R P A C K I N G S A N D COVERI NGS 43
therefore
2 ( J ”"COt 1 ('C° S r ’ ^ ) “ COt 1 ('C0S r ' ^

By the substitutions
tan <j>= cos r . yj3, cot p = cos r . \ y j 3, tan ip = u (9)
from (8) we have
. ~, 2 tan (p 2 .3
tan 2 <p = — .— -L- ^ ta n \p = u.
1 —tan2 <j>
Therefore we have
u = tan 5
and from (9)
r ^ cos-1 y^/5 = tan-1 2.
But this means that for every graph of the type represented in fig. 14,
r is not less than in the symmetrical case for which, by (4), tan r = 2
is valid.
The graphs of the type represented in fig. 13 (with a C-point of
order 4) also do not yield a smaller value for r. One quickly calcu
lates that in the symmetrical case
cos r = cot 45° cot 67-5°,
therefore r~65° 32'. For the graph of this type, applying Jensen’s
theorem also leads to the statement that it is not possible to have a
cap radius less than that of the symmetric type. Therefore the type
of symmetric graph represented in fig. 15 in stereographic projection
with no C-points of order 4 yields the best possible value f for r for
5 C-points:—
r = R s — cos- 1 1^/5 = tan -1 2~63° 26'.
In geographical terminology one can describe this graph simply
as follows:—If there is one point at the north pole and one at the
south pole then the three others are distributed symmetrically on the
equator. In the case n = 6 we have the corresponding solution:—
In this case the extremal configuration consists of the vertices of an
t We will denote by R„ the extremal value of r for the covering by n caps.
We have used the notation rn for the packing problem.
inscribed octahedron and, of course, one can also place this body so
that one point is at the north pole and one at the south pole and the
others lie on the equator.
Therefore it is natural to conjecture that for more than 6 points
the extremal configuration may also be formed by a corresponding
distribution. In any case it is not meaningless to calculate the
graph of this type for any n.
The (n —2) rhombuses meeting in the north or the south pole
respectively are each divided by their diagonals into 4 congruent,
c;
F ig . 15
right-angled triangles which have an angle (at the north or south
pole respectively) of The
side r for n = 7 is calculated from this triangle:—
r = p7 = cot-1 (cos 36°)~51° 2'.
In the case n = 8
p8 = cot-1 (cos 30°)-49° 7'. ( 10)
It remains to be shown whether or not these values p 7 and p 8 are

equal to the required extremal values R n and i?8.
7. The Covering Problem for n = 1

To survey the graphs coming into question in the case of the covering
problem for n = 7, we introduce a simpler system of lines and call
it the “ C-figure It corresponds to the graph | used in the packing
problem. If, in a graph made up of rhombuses, one draws the
diagonals joining the C-points in each rhombus then, on cancelling
the graph, there results a triangular net joining the centres which is in
fact the C-figure. The C-figure is indicated in fig. 14 by the dashed

lines.
If, conversely, a C-figure is given, one can recognise the graph, at
least in its topological structure:—One plots in each triangle an
I-point and joins it to the vertices of the triangle (fig. 16). On
cancelling the C-figure one obtains the graph. At the same time
one must take into account that I-points of order 4 might arise ;t
in this case a rhombus would degenerate into two lines, each
traversed twice, and I-points would concide e.g. 7j and I2, say, in
fig. 16.
t In Schutte [III 2] this figure is called the graph.
t One can certainly show easily that in the case n = 7 no I-points of order 4
can arise. Four cap circles which pass through the south pole, say, and are of
radius R t^ P t-^ 5 \° 2 'would leave uncovered a cap of radius 77° about the north
pole. Now the following theorem, whose proof we leave to the reader, is valid:—
i f a cap o f spherical radius p is covered by three congruent caps o f radius r, then
pgJ2 tan-1 (i tan r).
Hence it follows that the uncovered cap about the north pole cannot be covered
by the three remaining caps of radius R-j.
From Euler’s theorem (Theorem 2) for a polyhedron and the

relation 3a = 2e one immediately obtains for n = 7 the number a \—
The C-figure consists of 10 triangles. We assert that none of the
C-points can be of order 3. If there were,f then obviously one could
combine three of the 10 triangles into one triangle. Since the sides
of this triangle are diagonals of the graph, their length is less than
2r. The area of this triangle is therefore at most equal to the area
T of an equilateral triangle with sides of length 2r. However,
the area of the 7 remaining triangles is at most equal to the a r e a ^
of an equilateral triangle inscribed in the circle of radius r. For the
angle a x of the first triangle the following is true
cos r . sin — = cos 60°;

2
on the other hand we have for the angle of the second triangle
cot — . cot 60° = cos r.

2
Hence, it follows when one translates the radian measure into
degrees for r ^ p 7 <52° that
2T = 6 cot-1 (^ 3 cos r)—180°<80°
and
T = 6 sin "1 ( — — ) -180° < 150°.
\2 cos r j
Since the 10 triangles partition the whole sphere, it follows that
720°^ T + I F <710°.
Because of this contradiction, none of the C-points can be of
order 3. Since, obviously, no C-points of order 6 can arise, we
have, if cv denotes the number of C-points of order v,
4c4 + 5c5 = 2e = 30,
c4 + c5 = n = l.
Therefore cs = 2 , c4 = 5. But a C-figure with these numbers leads
to the type represented in fig. 16. Hence, in the case of symmetry,
one obtains the graph whose side p 7 we have already calculated.
t Here we are following Schutte [III 2],
Here two of the C-points are at the north and south pole of the
sphere and the other five are distributed at equal distances apart on
the equator. On applying Jensen’s theorem (Theorem 3), one
shows again, as in the case n = 5, that no unsymmetrical ordering
of C-points leads to a graph of this type having a smaller value for r.
Hence it has been shown that R-, = 'p7 = cot-1 (cos 36°) is the
smallest value for r in a graph consisting of only rhombuses. We
leave the reader to prove that in the case n = 7 the extremal graph
cannot contain a hexagon and we summarise our results in
Theorem 12: One obtains a covering of the unit sphere by 5, 6 or 1
congruent spherical caps o f smallest possible radius Rn (n = 5, 6, 7)
when two of the centres of the caps are placed at the north and
south poles of the sphere and the others are distributed at equal
distances apart on the equator. For the smallest possible radius
R„ of the spherical cap
is true.
8 . A n O utline of F urther P roblems

Anyone who imagines that this can be extended similarly is wrong.
For n = 8 there is [III 2] a graph whose radius r is smaller than the
value (10), which was calculated in Section III 6 and which is the
F ig . 17
value for the equatorial distribution of the points. This graph,

represented schematically in fig. 17, is not so easy to deal with,
even in the symmetrical case. One is led to a transcendental
equation with a solution
r~48° 9 '< p 8~49° 7'.
By an argument similar to that for n = 7 one can show, using this

result, that there cannot be a C-point of order 3 in the C-figure for
n = 8. Hence one can deduce that only a graph of the type
represented in fig. 17 must be considered for our extremal problem
when n = 8 p H n
Let us collect the most important results of Chapters H and
III in a table. The first two rows give the extremal radius and the
density of packing for the packing problem, the following rows refer
to the covering problem. We have noted not only the radius and
packing density (7?„ and D„) but also the number of the C- and
I-points of order v (cv and /v respectively).
4 5 6 7 8 9 10 11 12
fn 54° 46' 45° 45° 38° 56' 37° 26' 35° 16' 31° 43'
dn 0-845 0-732 0-878 0-777 0-823 0-825 0-896
Rn 70° 32' 63° 26' 54° 44' 51° 2' :S48° 9' 37° 43'
Dn 1-33... 1-382 1-268 1-299 gl-331 1-254
C} 4 2 0 0 0 o
c4 0 3 6 5 4 0
C5 0 0
0 2 4 12
h 4 6 8 10 12 20
I RREGULAR P ACKI NGS A ND COVERINGS 49
For some of the still unsolved cases of the packing problem
(n = 10, 11, 13, ...) there exist approximations and conjectures,
discussed in Fejes Toth [II 7]. Perhaps it is possible in the cases
which are still unsolved to make some progress by looking for
regularities in the structure of the graphs. We have also recorded
in the table the numbers cv and iv for the covering problem. /'4 is not
noted since an I-point of order 4 did not occur in any of the extremal
graphs found so far. However, at present, it has not been proved
that this must always be the case. It is natural to interpolate
between the values determined for n = 8 and n = 12 in the rows for
c3, c4, c5 and i3 as follows:
9 10 11
Cl 0 0 0
c4 3 2 1
Cs 6 8 10
h 14 16
'*
Chapter IV
PACKING OF CONGRUENT SPHERES
1. T he D ensest L attice P a ck ing

So far in dealing with packing problems it has always turned out
that the extremal configurations exhibit a high degree of symmetry.
If one assumes certain symmetry properties then finding the solution
becomes intrinsically easier: of course, the results obtained cannot
be formulated as generally as for the procedures “ without assump
tions ” , which we used in Chapters II and III.
To make this difference clear let us consider once again the
problem of packing unit circles in the plane. If we assume that
the centres of the circles have been arranged to form a lattice of
parallelograms, we can very quickly see how to get the densest
packing: one must choose the parallelograms to be rhombuses with
angles of 60° and 120° (fig. 4).
Let a point lattice like this be given by the position vectors
■i = ma + m> (1)
starting from some zero, where a and v- are (linearly independent)
vectors, whose length is at least equal to 2 and m and n are arbitrary
integers. If | a | or | v | is greater than 2 the density of packing can
obviously be improved. Therefore one can assume | ^ | = | v- \ = 2 .
Then the area of a single parallelogram in the net (1) is a{a) = 4 sin a
= 4 sin (a, v) and for the density of packing d(<x) we obtain f from
equation II (2).
m = lirn ?**> = lim ■ = _2_.

R~><n 71R 2 R~*oo ti(R) 4 sin a 4 sin a
The maximum of d(a) occurs when sin a is a minimum, a = 60°
is the condition for the latter. Obviously a smaller angle is not
t One may replace ttR* under the limit sign by n(R) . a(a). This is shown
by an argument like the one on p. 15.
50
P A C K I N G OF C O N G R U E N T S P HF R F S 51
possible for, if it were, the circles belonging to a pair of opposite
71
vertices would intersect each other. Hence it follows that d ^ -^ =
as in equation II (17).
Since the—as yet unsolved—general packing problem (3) (p. 2)
for congruent spheres in space is quite difficult, let us begin by
examining packings of congruent spheres in space when these are
formed on a lattice. Therefore we start from a lattice (fig. 18)
i = lu + + na> ( 2)
where t, v and u> are linearly independent vectors whose length is at
F ig . 18
least equal to 2 and /, m and n are integers. Let the points of (2)
be the centres of a system {d} of unit spheres. Since, here too, one
can obviously improve the density of packing by moving all the
spheres if one of the vectors is longer than 2, we can limit ourselves
to the case where the parallelepipeds formed by the lattice points
have only rhombuses as side faces: | a \ — \ v \ =\ u>\ = 2 .
Now, if a # 60°, ^ 120° were true, one could, as indicated by the
investigation of the density of packing of the unit circles, improve the
spherical packing by changing the angle a. The same is true for
the (a, «>)- and the (t>, «>)-planes. Therefore the angles of all
the rhombuses of the lattice (2) must be equal to 60° or 120° in the
lattice with extremal density.
By a discussion similar to that on p. 15, one obtains for the
density of packing of our system of spheres
N(R) - n N(R) - n
3 3
4n
5(d) = lim 4 lim N ( R ) . V =
r-» oo - nR 3V
3
52 U N S O L V E D A N D U N S O L V A D L E P ROBLEMS IN GEOME T R Y
where V is the volume of the parallelepiped determined by the

vectors a, v- and to and N(R) is the number of parallelepipeds lying
completely within the sphere of radius R. However, for the extremal
system of spheres {<?*} the volume is V = V* = 4^/2. Hence we
have
Theorem 13: Let
-) = let + mv -\-nto (| u | = 2, | v | = 2, \ to \ = 2)
he the lattice of the centres of a system {o} of unit spheres. For

the density of packing of this kind of spherical system
<S(«)S<5(d( , , ) = - V 2 (3 )
o
is true. The extremum is attained for an arrangement in the form
o f a lattice (2) with the vectors
a = (2,0,0), „ = (1 ,7 3 ,0 ), ^ = (1 ,1 7 3 ,1 7 6 ). (4)
There is an unproved conjecture that (3) also holds for arbitrary
packings of unit spheres.
Each rhombus of the extremal lattice decomposes into two
equilateral triangles with centroids Cv and C' (fig. 19).
In each parallelepiped, if one projects the edge parallel to the
vector to onto the rhombus parallel to the («, v)-plane, one obtains
a line P VCV which joins a vertex of the rhombus to the centroid.
Displacing the P v by the vector n = P jQ one obtains all these
centroids Cv. Therefore one can represent the extremal packing
of spheres so that firstly one places a layer of spheres in the (tt, v)-
plane (with the angle a = 60°) and then one imbeds the second layer
P A C K I N G OF C O N G R U E N T S P H E R E S 53
of spheres in the interspace denoted by the points Cv. On the other
hand, the centroids C' of the other triangles have no spheres. If one
applies the vector n again to the centres of the second layer one
obtains the centres of the next layer, and so on.
However, one can also proceed in another way:—If one applies
to the centres Pv not the vector n but —n then just the other points
C' (fig. 19) are reached by this displacement. One obtains a packing
of equal density if one imbeds the spheres in the interspace denoted
by C' and it is possible in going from layer to layer to change from
to —n in a regular or also in quite a random way. In each case one

n —
obtains for the density of packing d = - yJ2. Of course, arrange
ments like these would no longer be referred to as being “ lattice
formed ”.
For later applications we observe that one can also describe the
arrangement of the centres in the extremal lattice as a “ face-centred
cubical lattice ” . Moreover, we start from a cubical lattice
•i, = la+ m d + nc. (5)
The vectors a, 6, c must be perpendicular to each other in pairs and
have length 2 j2 . If one adds to the lattice points (5) the centres of
the side faces of each cube then one obtains an extremal lattice for
the packing of unit spheres.
To see this, we single out a cube of the lattice (5) and, in addition,
plot the centres of the faces (fig. 20). If one joins two opposite
vertices P x and P 8 of the cube with the centres P2, P 3 and P 4 and
P 5, P 6 and P 7 respectively of the adjoining side faces and also draws
in the linesP2P 6, P 3P6, P 4Ps and P4.P1, then one obtains a parallele
piped which is inscribed in the cube and whose edges are all of
length 2. Since P2P3 is also equal to 2, the whole face-centred cubical
lattice is just the lattice of an extremal packing of spheres. From this
fact one obtains once again the formula for the density of packing.
Of course, one sees immediately that the unit spheres placed about
the 6 face centres lie half inside the cube. One octant of each of the
spheres belonging to the eight vertices of the cube lies inside the
cube. There are f +1 = 4 full spheres. Therefore the density is
2. T h e F in e s t R ig id P a c k in g
By analogy with the packing of circles (p. 26), a spherical packing is
said to be rigid if each sphere is touched by 4 others so that the 4
points of contact do not lie on a hemisphere. The extremal packing
just investigated is certainly rigid: here, in fact, every sphere is
touched by 12 others.
There exists a rigid packing of spheres in which each sphere is
touched by exactly 4 others so that the points of contact form the
vertices of a regular tetrahedron.
To describe this system we turn once again to the system {j(1)}
with the densest packing. Suppose that a coordinate system is
specified in the face-centred cube of fig. 20 so that zero coincides
with P l5 the positive x-axis is in the direction from P x to P2 and
P 1P 4 lies in the xy-plane. Then the components of
a — P XP2, — P XP 4, rt> — PxP 3

are given in (4). Hence the vector x from P t to the centroid Qx of
the tetrahedron PXP2P3P4 has the components
-- = A e l = ( i . W 3 . W 6 ) ,
A new lattice of points {Qv} is defined by the displacement
n = P iQv = P xP v+ x = la + mv + +x (I, m, n integers).

P A C K I N G OF C O N G R U E N T SPHERES 55
Then each point Qv is the centroid of a tetrahedron which is formed
by 4 points of {Pv}: Qi of PyPiP^P^y Q2 ° f P2P3P4P 5 and s0 on-
Conversely, each point of {Pv} is the centroid of a regular tetra
hedron whose vertices belong to the lattice {Qv}. The union of the
lattices {Pv} and {£)v} is no longer a lattice in the sense of our
definition, but rather a system of points which one can take as the
centres of a rigid packing of spheres. One only needs to place a
sphere of radius
pi = i M = V «
about each point. We call this new system {<3(2)}.
Exactly 4 of the spheres about the points of the system {Qv} lie
wholly in the cube with sides 2 ^ 2 (fig. 20), namely those with centres
Qi> Q2> Q 3 and 64- The other spheres with centres Qv (v>4) lie
completely outside the cube. To see this, one need only note that
the vector ^ = P \Q \ is in the direction of the principal diagonal of

the cube. Since the spheres about the points P v and those about the
points Qv have, of course, the same density then just 4 of the spheres
about the points P v also lie in the cube. Therefore the density of
our packing is
8 . - np\
<5, = -0-340.
(2J2)3 16
Now this packing is by no means the finest of the rigid packings of
spheres. Here, in the same way as for the plane (p. 26), one can
find a new system in which one replaces each sphere of the system
{o(1)} of radius p x by 4 spheres with the following property:—These
4 spheres must have the same radius (p2) and touch the spheres of
internally at their points of contact. In addition, p2 must be
chosen so that each of the 4 spheres is touched by the other three.
Since 2a = 2 s i n “ 1x/ | is the angle between any two vectors
from the centroid of the regular tetrahedron to its vertices, we have
for the radius p2 of the spheres of the new system
sin a = V f = ~ 2 ,
P i-P i
therefore
1
P2 = P i ~~=---- •
V 2+1
56 U N S O L V E D AND U N S O L V A B L E P ROBLEMS IN G E O M E T R Y
Since each sphere of 1}} is replaced by four smaller ones from

{i>(2)}, the following is true for the density <52 of the new system
, 4 3
4 . - np2
3 4
d2 = — ------- <5i = ~ ~ -----5l -0-123.
U p\ U i+ 1 )1
3
It is conjectured that this is the smallest value for the density of a
rigid packing of spheres [II 3]. This has not yet been proved true.
Chapter V
L E B E S G U E ’S “ T I L E ” P R O B L E M
1. F orm ulation of the P roblem

Many statments for the “ tile problem ” (4) (p. 2) remain valid
when one considers any bounded point sets IT instead of clusters of
points {finite point sets). A segment of the plane (i.e. a closed
domain of the plane) is called a tile if a point set of diameter 1 in
the plane can be completely covered by it. By the diameter of any
bounded point set we mean the least upper bound of the distance
between any two points of the set. Our problem requires us to
determine a tile of minimal surface area. The problem was first
formulated in this form by Lebesgue in 1914 (see [V 3] p. 4). It is
easy to show that a positive lower bound exists for the surface area
of tiles (p. 65), but it is not immediately certain that there exists one
(or more) tile(s) having area equal to this lower bound.
When investigating such existence problems it is convenient to
restrict oneself to convex f tiles. Pal [V 3] specialised the Lebesgue
problem in this way:—
(4a) Let t 0 be the lower bound of the surface area t(IT) of all convex
tiles.
(a) The value o f t 0 is to be determined.
(b) It is to be ascertained whether there exists a tile of ST with
area equal to r 0.
(c) I f there exist tiles with minimal area, are all tiles ST with
surface measure t 0 of the form to be determined later ?
Since we have defined the tiles to be closed sets, we can also assume
the sets IT to be closed. In Sections 1-5 of this chapter we shall
always indicate by the letter IT a closed set of diameter 1.
t A domain is said to be convex if, when the points P and Q belong to the
domain, the line PQ also belongs to it.
E 57
58 U N S O L V E D A N D U N S O L V A B L E P R O B L E M S JN G E O M E T R Y
We want to approach our problem first of all by compiling

examples of tiles and trying to find those with the smallest possible
area. It is easy to show that the square with side of length 1 is a
suitable tile for each set £f. To prove this we introduce the concept
of the support line. This is a line / which contains at least one point
of and is such that a half plane to one side of it contains no point
of Sf. Then, to each line / in the plane of there exist two parallel
support lines and l2. They are at a distance apart dt which is at
most equal to 1. Then, if /3 and /4 are the support lines perpen
dicular to /j and at a distance apart d2tk 1, ^ is enclosed by these
4 lines, forming a rectangle with sides of length 1 and d2^ \ .
Thus it is clear that the square with side of length 1 is a tile. We
denote this tile by 3F2. We want to keep the suffix 1 for the Jung
circle of radius r = -^ 3 . Its surface area A = n/3 is somewhat
greater than 1 (A = 1*047197...).
It is immediately obvious that no circle with radius r< y 3 is a
tile:—The vertices of an equilateral triangle with side of length 1
forms a cluster of points of diameter 1, the closed triangular region
forms a convex set . It is covered by its circumscribed circle
(r = ^ /3 ) but no* by any smaller circle. One can easily prove
directly t h a t .^ actually is a tile (see e.g. [V 4]). We shall obtain
this property f o r y l from the fact that the regular hexagon inscribed
in Jung’s circle is a tile.
2. T he H exagonal T ile
To obtain another t i l e ^ 3 we enclosed in an equiangular hexagon
with sides (v = 1, 2, ..., 6) according to the following rule:—The
sides s2 and ,?4 must lie on parallel lines of support whose distance
apart is 1. s2 and s3 are on lines of support which enclose with
or s4, respectively, an angle of 120°. Finally s5 and s 6 must lie on
parallels to s2 or £3, respectively, at a distance of 1 unit. Therefore
s 5 and s 6 can be lines of support but need noi be (fig. 21). Numbering
the sides must produce a positive contour direction.
By projecting the sides and s2 onto a line which intersects s3
and s 6 at right angles, one obtains
(st + s2) cos 30° = 1.
In the same way
2
S2 + S3 ~ S 3 + 5 4 = S4 + S5 = S5 + 5 6 ~ s 6 + s l --------
P
l e b e s g u e ’s “ t i l e ” problem 59
and hence it follows that
s2 = s4 = s6(= s), Sj = s3 = s5(= a).
If Si = s2, then s = a and the equiangular hexagon is regular. We

want to show that in every case a regular hexagon (with distance 1
between opposite sides) can be used as a covering of Sf.
Let Sj >s2, say. Then we introduce a notation for direction in the
plane of Sf. We associate the direction <j>= 0 with the base line
and the direction <j> — 60° with the line s2. For each direction <j>
with 0 5 ^ ^ 6 0 ° , there can now be found a parallel strip which is
constructed from a line of support and a parallel at a distance of one
unit and which encloses 9*. The process just described motivates
constructing an equiangular hexagon enclosing 9*. Let the sides
of the hexagon be sn(4>), n = 1, 2, ..., 6. Then
S(<£) = Si (<£)- s2(<£)

is a continuous function of <j>. For <j>= 0 we have by hypothesis
S(0) = Si —s2>0. On the other hand
5 (6 0°) = Si(60°) —s2(60°) = s 2 —s 3 = s 2 — S i < 0.
Therefore, between </> = 0 and </> = 60°, by continuity, there must

be a value of </>for which S(<f>) is zero. Then Si(<f>) = s2((j>) and the
corresponding equiangular hexagon is regular.
Hence it has been shown that the regular hexagon with in-circle
of radius p = i , therefore with circumcircle of radius r = ^ 3 , is
a tile. We denote it by^"3. Of course the circumcircle of^"3 is also
a tile. This is just Jung’s circle which we have denoted b y ^ ^
60 UNSOLVED ANI ) U N S O L V A H L E P R O B L E M S IN G E O M E T R Y
In this way we have found 3 tiles whose areas decrease with the
suffix n :—
t , = F ( 3 r y) = 1-04797... > t 2 = F (3T 2) = 1 > t 3 = F ( ^ 3)
= = 0-8660.
3. The T iles of Pal and Sprague

The tile £T2 can easily be improved on. If one draws to the in-
circle o f ^ 3 the tangents which are perpendicular to the diagonals,
one obtains at the 6 vertices V of^"3 (fig. 22) 6 isosceles triangles
Av. Now let a point set Sf be covered b y ^ 3. Then all the triangles
Av cannot be occupied | by points of :—The interior points of
opposite triangles are at a distance apart which is greater than 1.
Therefore if Ax is occupied then A4 is free and conversely. Similarly,
both A2 and A5 and A3 and A6 respectively cannot be occupied
simultaneously. Therefore there are the following 8 possibilities
for 3 corners to be occupied:—
A i, A 2, a 3 a 4 , A 5, a 6
A 2, A 3, a 4 A 5, A g, A ! A i , A 3, a 5
A 3, A 4, a 5 A6> A i , A 2 a 2, A 4 , a 6
The other three respectively are free. It is immediately obvious that,

of the free corners, there are always two whose suffixes differ by
t We say that a triangle Av is “ occupied ” if a point of Sf lies in the interior
of the triangle or on a point of the boundary of Av which does not belong to the
tangent.
LLBLSGUL’S “ TILL” PROBLLM (>l
two or four. Hence ST3 does not cease to be a tile if one cuts off two
corners which are not adjacent and not opposite, say At and A3.
The remaining tile 4( = ABV2CDV4V 5V6 in fig. 23) given by Pal
[V 3] has area
t 4 = F(3T4) = 2 - ^ / 3 = 0-845299....
R. Sprague [Y 5] has shown that one can also improve on this
tile. One can show this as follows:—Let Sf be a point set of diameter
1 which is covered by 3~4. Only one of the two isosceles triangles

at V2 and V5, A2 = V2EF and A5 = J G V 5, can be occupied.!
First of all we suppose that a point P eSP lies on the line CF. Then
we draw about C the circular arc GH with radius 1 which must
touch V5V6 in G and intersect V4V5 in H. Then no point o f SP can
lie in the part of the tile cut off by this arc since all these points are
at a distance greater than 1 from the point of SP lying on CF. We
want to show that we achieve our purpose in each case with the
residual tile left when A = V5GH (GH being a circular arc) is cut
o ff.^ 4.
Therefore, let us suppose that no point of £P lies on CF. If one
does lie on EF then the whole triangle A5 is unoccupied and we
certainly achieve our purpose with the reduced tile ST s = FT4 —A. If a
point of PP lies on EB or AB, one can rotate the tile about the axis
V2V5 (fig. 23) and hence we have that CF or DC respectively lies
on a point of SP. In the first case—we have already seen this—we
certainly achieve our purpose with the “ residual tile ” . In the
f Fmust lie on ViV^ G on V5Vfi.
62 U N S O L V E D AND U N S O L V A U L E P R O B L E MS IN G E O ME T R Y
second case we displace the tile in the direction of the axis srf to the
left until a point of 9* lies on the polygon CFEBAL. If the broken
line CFEB is occupied then the residual tile forms the covering. If,
however, the displacement leads to BAL being occupied then we
rotate the tile about the axis stf and so have EFC occupied. This
rotation is permitted if CD is occupied, since, in this case, the
opposite corner (A6 = M L V 6) must be free.
Now let us suppose that in a covering of a set SP by.^”4 the whole
broken line ABEFCD (and, of course, the triangle A2 also) is free
of points of SP. Then one only needs to displace the whole tile in
the direction V2V5 (arrow in fig. 23) until a point of ABEFCD
reaches a point of SP. Carrying out this displacement does not
cancel the covering of the other points of SP by the tile. Hence
this general case has also led back to that already discussed and
we have shown that ST5 = 5 4 - A is also a tile.
FT $ is not symmetric about the axis V2V5. This suggests that
one can also cut off another piece of the triangle A5 by the circle
with centre B and radius BJ = 1 (fig. 23). Let K be the point of
intersection, lying on V2V 5, of the circles about B and C with
radius 1. Then, indeed, & 6 = ABV2CDV4JKGVb is a tile, as we
proceed to show in the next section.
4. C u r v e s a n d D o m a in s o f C o n s t a n t W i d t h
A bounded domain is said to be of bounded width b if each pair of
parallel lines of support has the same distance apart b. The
boundary of such a domain is called a curve of constant width.
The circular disk of diameter b is, of course, a domain of this kind.
Other simple examples are certain triangles with circular arcs as
sides. If one draws about each vertex of a (2/7+l)-gon the circular
arcs passing through the vertices of the opposite side one obtains
a domain of constant width d, where d is equal to the radius of the
circular arcs (fig. 24 for 2n+ 1 = 5).
Now the following theorem is significant for our tile problem:—
Theorem 14: Each cluster of points <€ is a subset of a closed, convex
domain with constant width 1.
To prove this we place about a polygon 0 which is constructed
from support lines of this set and whose vertices are points of the
cluster c€. Let <<? = {Qt , Qz, ..., Qm} be the cluster of points and
PvP2P2...Pn the polygon, where {P1,P 2, ..., P„} is a subset of c€.
L 1- iiLSGL'Ii’s “ T I L L ” PROBLEM 63
The distance between at least one pair of points PJ*^ must then be
equal to 1. Let this be the case for P xPi-
We denote by Ax the half-plane determined by the line P xPt and
containing the points P2,P 3, •••,P ;_x; let the other half-plane be
/?2 (fig. 25).
Let us now draw circles ( c x, c 2, ..., c,) with centres P l5 P 2, P 3, ■■■,Pi
and radius 1. and cj +l intersect at the point Rj of the half-plane
fi2. In this way there results in fi2 a polygon P iRl R2...R i_ lP l with
circular arcs as sides (curvilinear polygon). Now let us also draw
a circle with centre Rj and radius 1 then we also have in the half
plane ftx a curvilinear polygon P xP2P3...Pi (fig- 25).
Then the whole curvilinear polygon P lP2...P iR1...R ;_ XP X is

obviously a curve of constant width 1 :-O n e of a pair of parallel
support lines 1 unit apart always goes through a vertex of the
curvilinear polygon and the other is a tangent to the opposite arc.
All the points of our cluster must lie in the domain determined
by the curvilinear polygon since those lying in fi2 must also be at a
distance of at most 1 unit from all points P} of the half-plane
The convex domain of width 1 constructed in this way is also
called the Reuleaux curvilinear polygon belonging to c6.
Theorem 14 can now be used to prove that ST $ is a tile. Let # be
a cluster of points of diameter 1 and 01 a Reuleaux curvilinear
polygon associated with it. Then 0 can be covered by the tile 0~4.
Since there is a pair of parallel lines of support 1 unit apart corres
ponding to each direction, a boundary point of 0, must lie on each
of the lines BV2, V2C, DV4, V4V5, L5L6 and V6A (fig. 23). One
at least, of the triangles A2 and A5 must remain free of points. If A5
is free then the smaller tile S ' ^ is certainly sufficient to cover ci .
Therefore let us suppose that A2 = EV 2F, rather than A5, is free
of points of The point lying on V2C must therefore lie on
FC and that lying between B and V2 must lie on BE. Then no
further points of M can lie outside of the circle with centre C and
radius CG = 1 and with centre B, radius BJ = 1. However, this
means that the tile S ' 6 is also sufficient to cover £# and hence also c€.
In this way it has been shown, in the first instance, only that every
cluster of points cf> can be covered by ST6, but not unconditionally
every arbitrary set of diameter 1. Now one can easily extend the
statement of Theorem 14 to any set of diameter 1. But it is even
simpler to deduce directly the tile property for an arbitrary set SP
from that for a cluster of points.
For this purpose let us think of the points, enumerated in anyway,
of a subset which is everywhere dense in S ’ : P u P2, ..., Pn, ....
Then let %>„ be the cluster of points {Px, P2, ..., P„}. Each cluster
9>„ is covered by S~6. Let S ~ ^ be the domains which are congruent
to our tile S ' $ thought of as movable and which cover the corres
ponding clusters with the same suffix. From the sequence S ’%)
one can now select a sub-sequence S '(^ ) for which the (finitely many)
vertices S ‘(g ) converge to limit points. These limit points determine
a new position S'* of our tile S ' 6. It obviously covers all the points
Pn and hence also S ’ since each point of S ’ is a point of accumulation
of points of the set {/*„}.
The tile ^ e (also discovered by R. Sprague) has surface area
t 6 = A(S~6) = 0-844144....
In order to estimate the convergence of this tile,t the best to our
knowledge at present, to the lower limit r 0, we now try to find
lower limits for the area of all tiles.
Each tile which is to cover all sets of diameter 1 must contain a
circle of this diameter. The area of each tile is therefore at least
equal to n/4. In order to improve this limit we consider that each
tile must also cover an equilateral triangle with sides of length 1.
f One sees immediately that one cannot cut off any of the vertices P'2, V4 and
V6 of the tile § (Fig. 23) because they are required to cover the equilateral
triangle with sides of length 1. R. S prague conjectures that the absolute
necessity of the other vertices of the tile can also be proved by suitable, simple
clusters of points.
However, even if one cannot cut off any more from 2 T it is still possible that
there could exist yet another kind of tile with smaller area.
LE B E S GUE ’ s “ TILL " PROBLEM 65
Therefore each convex tile contains the convex hull of a union of
sets of circle and triangle. Pal has shown [V 3] that this set has
minimal surface area if the centre of the circle lies at the centroid
of the triangle. Therefore, the surface area of this figure & is a
lower limit for the surface area of all tiles. However, gF itself is not
a tile. In fact one can show that gF does not cover the regular
pentagon with constant width 1 (fig. 24). Hence A(iF) = 0-825711...
is t smaller than the lower bound t 0 of the areas of all the tiles and
we have as the prevailing result
0-825711... < t 0 ^ t 6 = A(gr6) = 0-844144...
or
t 0 = 0-834928 ± 0-009217.... (2)
Hence the numerical value mentioned in problem (4a) (p. 57) is in
fact not determined but is at least bounded. The question (b) in
problem (4a) is to be answered in the affirmative. One can show
using Blaschke’s theorem of choice £ that there actually exists at
least one minimal convex tile. We will not go into the proof of
this existence theorem since it is rather involved though not difficult^
The problem of the practical determination of one or more minimal
tiles is still open.
Theorem 15: Among all the convex tiles which cover each plane
domain of diameter 1 there exists at least one tile with the minimal
surface area r 0. For this numerical value
0-825711... < r 0^ r 6
is true, where x6 is the area of Sprague's tile .T 6.
5. A pplications
In 1933 G. Borsuk |] posed the question whether it is always
possible to decompose a point set of n-dimensional space of diameter
1 into n + 1 parts so that the diameter of each subset is less than 1.
One can immediately give an affirmative answer to this question for
the plane using our results on the tile problem. Each plane set of
diameter 1 can be completely covered by the hexagonal tile gF3.
If one can divide this tile into 3 parts, each with diameter smaller
t Here the positive answer to question (b) of problem (4a) (p. 57) has been
anticipated.
f See e.g. [V 10].
§ See e.g. [V 3],
|| Fund. Math., 20, pp. 177-190, 1933.
66 UNSOLVED AND U N S O L V A B L L F R O B L L M S IN G E O M E T R Y
than 1, then a corresponding decomposition is possible for each set of

diameter 1.
To divide the tile ZT3 in the required way, from the centre of this
tile we drop the perpendiculars onto three, non-adjacent sides.
Let the feet of these perpendiculars which enclose angles of 120° be
P \,P 2 and P3. ZT3 is divided by these perpendiculars into 3 sub-tiles
of diameter P^P^ = \yj2>. Hence we have an affirmative answer to
Borsuk’s problem for 0t2.
Theorem 16: Each plane set of diameter 1 can be decomposed into 3
subsets whose diameters are at most equal to ^ 3 .
Another application of our results for the covering problem leads
to the
Theorem 17: Let Q- be a set of open disks of radius 1, each two of
which have a point in common. Then there is an equilateral
triangle PQR with sides of length 1 such that each disk contains
at least one o f the points P, Q or R.
We can visualise this theorem as follows:—One imagines the
disks of the set Qs cut out of paper and laid on a drawing board.
Then one can fix the disks by pins in 3 points P, Q and R (which
form the vertices of an equilateral triangle with sides of length 1)
so that each disk is fastened to the board by (at least) one pin.
To prove this theorem we note that the centres of the disks of Q
form a cluster of points <6 with diameter d< 2, since each two disks
have a point in common. <€ can be covered by a tile of type ZT3 for
which the distance between opposite sides of the hexagon is equal
to 2. Let us denote the vertices of this tile by A v (v = 1, 2, ..., 6)
and the midpoints of A XA 3, A 3A 5 and A 5A X by P, Q and R. Then
PQR is an equilateral triangle with sides of length 1. Since each
point of the tile is at a distance of at most 1 unit from at least one
of the 3 points P, Q and R, this is also true for the centres of the
disks which are covered by the tile. This means that every disk
of Q) covers at least one of the three points.
One can generalise the problem considered here and ask by how
many points can a set Os' of disks be fixed if each two disks of this
set have an (interior) point in common but the radii maybe different.
One knows that 5 points suffice | and L. F ejes T oth conjectures
that one can also manage with 4 points.
f See H a d w ig e r in Ungeloste Probleme (Unsolved Problems), No. 19,
El. d. Math., XIF, pp. 109-110, 1957.
l h b e s g u e ’s “ tile” problem 67
6 . T h e C o v e r in g P ro b lem in &l3
Jung showed in 1903 [V 1] that every cluster of points of diameter 1
in three dimensional space can be placed in a sphere of radius
r = i>/§. K ir s c h gave a new, elementary proof [V 7]. This result
suggests the problem of finding a covering body (D-body) of smallest
volume for a cluster o f points of diameter 1 of d$3. Jung’s sphere is a
D-body Q}3 with volume V(Q)f) = -0-95.. We now prove
Theorem 18: The regular octahedron with edges of length a = ^J-2 is

a D-body Q!2 with volume
V(®2) = W 3 ~0-866....
To prove t this theorem we need the concept of the “ plane of
support ” of a cluster of points cd. By this we mean (in analogy to
the line of support, see p. 58) a plane 77 which has at least one point
in common with ^ and such that one of the two half-spaces
determined by 77 does not contain any points of cd.
Now let A lA 2A 3A 4B1B2 be an octahedron 0 whose axes A 3A 3,
A 2A 4 and B l B2 coincide with the x-, y- and z-axes of a coordinate
system. Let the edges be chosen of length 1, say. We use this
octahedron simply to have a convenient notation for certain planes
and angles. First of all we denote the planes determined by the
points A vA v+1Bl by ev (v = 1, ...,4 , A 5 = A j). Then the three
planes els e3 and e4 form at B1 a solid angle s i containing the
octahedron.
Then there are 3 well-defined planes of support rji , r)3 and 774 of #
which are parallel to el5 e3 and e4 and at a point S form a solid
angle s i t which is parallel to the angle s i and contains (i . Now let
us suppose that the plane ~r)2 through S' and parallel to e2 is not a
plane of support of T! (fig. 26). Then there are two possibilities:—
1. All the points C e ^ lie inside the 4-sided angle s i 2 which is
formed at S by the planes ^ (p = 1, 2, 3, 4). Then at least one
point C lies on the sides of s i 2 determined by the planes rj1, r)3
and 774 but none on rj2.
2. There is at least one point C e 'if which lies outside s i 2.
First of all let us consider the first case. There exists for this
position of s i 2 a plane of support r\'2 of (€ which is parallel to t)2.
t This theorem was stated without proof by D. G ale [V 6].
68 UNS OLVKD A N D U N S O L V A B L E PROBLEMS IN GEOMETRY
It must go through a point S' of the edge of determined by 171

V* (fig- 26).
We now have at S ’ a new solid angle which is located parallel
to the angle s#2. It is formed by the planes 77l51]'2, ri4 and a plane
t]3 parallel to r)3. Now, of these 4 planes tjlf tj2 and are planes of
support while rj2 does not have this property.
We now take an interior point of as the origin of a coordinate
system. Then we can introduce the directed distance of a point P
from a plane using the Hesse standard forms of the plane. With
this choice of origin the distance from the planes r\’2, 773, r\A for
all interior points of stf 2 becomes negative. Now let d be the

maximum of the directed distance of the points Cv e # from i)2 and
d' that from r]'2. Then in our case d < 0 but d' > 0 since at least one
point Cv e lies on the plane r]3 placed outside the angle s / 3.
Now let us imagine the octahedron 0 replaced by an octahedron
0(a) = A \A 1A lA ,l B lB2 which results from G = 0(0) by rotation
through the angle a about the axis B VB2. Let the corresponding
planes be e^(a). Here a distance d ( a) depending on a corresponds
to the maximal distance d from rj2. d(oc) is certainly a continuous
function of a. Now, for a = n/2, we get a rotation of the octahedron
through a right angle so that
Then the solid angle s / 2(n/2) corresponding to this octahedron has

the property that -rj^njl), 173(71/2) and 774(71/2), and therefore >f2,
and 77j, are the planes of support of However, the angle at S'
denoted just above by has this very property. Therefore we
L E B E S G U l '.S “ I 1 LI ” PROBLEM 69
have j^ 2(7r/2) = ^ 3 and correspondingly d{nj2) = d '. Hence
d(0) = d< 0, d ( ^ = d ’> 0 .
From continuity considerations there must therefore be an angle

a for which d(a) = 0. Then we have a solid angle sd2(°0 which
contains ^ completely and has at least one point of (8 in each side.|
If there is a point Cv e (6 lying outside s i 2, one comes to the
same result in the end. In the second case one starts with a parallel
displacement of s i 2 into a position sd3 for which three sides belong
to planes of support while the maximal distance d" from the last
side is negative. The details of this argument are left to the reader
to supply. In this case also one reaches the conclusion that there
exists a solid angle s i 2(oc) for which d'(a) is zero.
Therefore in each case a 4-sided solid angle exists which is
congruent to the 4-sided angle of the octahedron at B x and contains
on each of its sides at least one point of (€. However ^ has dia
meter 1. In each direction there are therefore two parallel planes at
distance 1 unit apart, of which at least one is a plane of support
and between which the whole cluster of points must lie. Then the
angle sd 2{d) (or respectively) and the planes parallel to its
sides and 1 unit apart determine an octahedron which contains (€ .
From the fact that the opposite sides are 1 unit apart, one determines
immediately the length of the edges a = Thus Theorem 18
has been proved.
If one places on each of the 8 surfaces of the octahedron a tetra
hedron with edges of length a = then one obtains a tetrahedron
with edge of length al = which is also of course a D-body. In
this way a theorem of D. G ale [V 6] for is proved: D. G ale
has shown that in n-dimensional space each cluster of points of
diameter 1 can be enclosed in a simplex with edge of length
/« (/z+ l) y
1 . However, a simplex is not a particularly good D-body
since, from the results of G ale , the edge length and hence also the
volume increases with n without bound. However, for Jung’s
sphere this is not the case: the generalisation of Jung’s theorem to
t This condition is also satisfied for two sides of the angle if at least one point
lies on the common edge.
/j-dimensional space gives for the radius (see e.g. [VI 11] p. 140)
=Jrr?’ lim
Y 2/7 + 2 n-+oo =^
It is possible to reduce the D-body 0)2 by cutting off parts in the
same way as for the hexagonal tile 3 in the plane. To see this we
place at a distance planes parallel to the planes of symmetry which
meet 3i 2 at 4 vertices and these cut off from Oi2 at each vertex a half
octahedron with height h = ] ( ^ /3 - l) , We denote these 4-sided
pyramids by (v = 1, They may belong to A t , A 2, A 3, A4,
2?j, B2 respectively. If, in the covering of a cluster of points ^ by
Qj2, a point of ^ lies | in one of the pyramids then the opposite
pyramid must be unoccupied since the distance between the bases
of opposite pyramids is exactly equal to 1. Therefore of the six
pyramids at most three can be occupied.
If one notes all the possible combinations of free and occupied
pyramids then one knows at once that in each case there remain
unoccupied three pyramids which belong to the vertices of a tri
angular face of 3)2. However, this means that there is a D-body
which arises from Q)2 by cutting off the pyramids 0>u $?2 and
Since the volume of each pyramid cut off is equal to ^(3N/5 —5), we
have for the volume of £23
V(®3) = V(®2) - 3 ,J(3V3 —5)~0-768.... (3)
Theorem 19: I f one cuts offfrom the D-body 0)2 three right pyramids
at vertices (which are not opposite each other) and of volume
^ (3 ^ 3 —5) then one again obtains a D-body S>3 with volume (3).
It is certainly possible to reduce this D-body still further by
carrying over to the arguments of R. S p r a g u e . The problem
of the D-body of smallest volume in is unanswered just as that
concerning the smallest tile in the plane.
One can use the D-body Q 2 to prove the Borsuk conjecture
(p. 65) for ^?3. B. G run ba um [V 11] has X shown recently, using
the covering property of Q 2, that one can decompose each point
set of diameter 1 in ^?3 into 4 subsets each of which has diameter
r/^0-88. The question whether one can improve this result by
using a D-body of smaller volume remains open.
t Here the points of the bases of the pyramids arc not considered as
belonging to
J See in this connection and for further basic problems of the theory of content
[II 1], [VI 2], [VI 11] and [VI 12],
Chapter VI
ON THE EQUIVALENCE OF
POLYHEDRA BY DISSECTION
1. T he P lane P roblem
In order to lead up to the problems of the theory of content in
(problem (5) p. 2), it is expedient to treat first of all the corresponding
questions in the plane.
One can associate with the triangle A the functional F(A) = \gh
(g\ base, height) as “ content” . Thus one can interpret the
product \gh either as a product of measures or in the sense of
Hilbert’s line multiplication as a product of lines. | One can easily
show that this product is independent of the special choice of base.
For a simple, closed, plane polygon & one can assign as area F(0f),
the sum of the areas of the constituent triangles which result from
any triangulation of the polygon. Hence FlfiF) is independent of the
particular kind of triangulation.
The functional F(&>) defined in this way has the following
properties:—
1. F(0>) = F(3), ?? = 3.
2. F(3>+ 3) = F(0>) + F(3) if & and 3 have no interior points in
common.
3. Fifii) = 1 for the unit square °U.
4. From FifiP) = F(3) it follows that & — 3 (see footnote J).
The last property is particularly noteworthy because, as a result
of it, the “ content” has a very clear geometrical meaning:—
Polygons of equal content can always be decomposed into finitely
many triangles which are congruent in pairs.
t In connection with this and for further basic problems of the theory of
content see [II 1], [VI 2] and [VI 12].
t The equivalence by dissection was defined already on p. 3.
71
72 UNSOLVLD AND UNSOLVABLE PROBLEMS IN GEOMETRY
To prove this important property 4. of our functional, we note
the following rules of equivalence by dissection (see e.g. [VI 2],
p. 134, et seq.):—
(a) From & = 2 and 2 = Sk it follows that & = 3%,
(b) F ro m ^ —&>x + .^ 2, 2 — 2 X+ 2 2, = 2 X, &2 — it follows
that &=■ 2.
We prove first of all the following lemma:—
Parallelograms with the same base and height are equivalent by
dissection.
A B
The proof of this lemma is particularly simple if, in the parallelo

grams ABCD and ABEF, the point E lies between C and D. Then
obviously
AEDB = AEDB, AEC = BFD,
therefore ABCD = ABEF. In the other case BD and AE will
intersect in G, say, (fig. 27). Now one measures off BG on BD and
AG on AE and obtains in this way the points Gx, G2, and
H x, H2, .... Now by the 1st axiom of continuity ([V I2] or [VI 11])
there is an integer n with the property that Gn coincides with D or
on the other hand D lies between Gn and Gn+X. Then one draws the
parallel lines noted in fig. 27 and reads off immediately from the
figure the congruence of triangles and trapezoids corresponding to
each other:—
A/ = Al, A ll = A2, ..., Tx = T2, DX= D 2,
AABG = A ABG.
EQUIVALENCE OF POLYHEDRA BY DISSECTION 73
Hence the equivalence by dissection of ABCD and A B E F is proved.
Note that at this point we have used an axiom of continuity.
As is well-known from H ilbert [VI 11] one can construct a theory
of content which does not use an axiom of continuity. Then at this
point one must of course be satisfied with the weaker statement that
parallelograms with common base and height are equivalent by
completion. |
Since each triangle is equivalent by dissection to a parallelogram
with half the height, it therefore follows by the rule (a): Triangles
with the same base and height are equivalent by dissection. If one
takes in addition the rule (b) one knows immediately that the process
(used in school teaching) of “ transforming ” an «-gon 0>n into an
(n —l)-gon leads to a figure which is not only equal in content
d
but also is equivalent by dissection: = &n- \ . Also the well-
known transformation of a triangle with base AB into a triangle
with base AB' leads to a figure equivalent by dissection. Now if one
changes two polygons and ^ 2 which are equal in content into
two triangles dTx and ST2 with unit bases (and height h = IFi&f )
= 2F(0>2))> one has the following relations between the figures
0> ,= 8T lv= 1,2), ^ x =2T2.
Hence it follows by rule (a) that ^ = ^ 2.
2. D e h n ’s T h eorem
In attempting to transfer these studies to space it is easy to discover
that prisms which are equal in content are equivalent by dissection
(see [VI 2] p. 190). However, in the 19th century people puzzled in
vain over the question whether the corresponding statement is true
also for pyramids (or for general polyhedra). Thus in 1900 H ilbert
posed the following problem to the Mathematical Congress in Paris:
It is required to find two pyramids with the same face as base and
equal height which are not equivalent by dissection.
This problem was solved a few months later by M ax D e h n . He
showed that there is a necessary condition for the equivalence by
dissection of polyhedra. Then it is no longer difficult to find a pair
c
t Two polygons ZP and 1 are “ equivalent by completion ” {0>= 1), if there
are finitely many triangles -Tv and congruent in pairs such that
3PyZ,$'y= 2.YZy\.
of pyramids with common base and equal height for which Dehn’s
condition is not satisfied.
Theorem 20: Let 0 and 2 be polyhedra equivalent by dissection,
having edge angles | a l9 a2, a„ and /?l5 ji2, ..., jim respectively.
Then there exist non-negative integers vl3 v2, v„ and
p i, p 2, ..., pm and an integer v such that
(v1CC1+V2<X2+ ... + VnXn) - ( p 1p 1+ p 2P2 + ...+ llmPm) = V7T. (1)
To prove this theorem we want to identify the sub-polyhedra of
0 and 2 and to demonstrate that we construct from the “ building
bricks” fr2, ..., first the polyhedron 0 then (from another
“ building plan ”) the polyhedron 2 equivalent by dissection to 0 .
We note the building plan for each of the sub-polyhedra by marking
the edges of the different, neighbouring polyhedra (in the 1st or 2nd
building plan) on y£p. For those who prefer a visual demonstration
one can colour the edges of /«p with a suitable dye and this clearly
indicates how to build up the polyhedra 0 and 2.
In this way a net of lines whose individual lines we will call
“ boundary lines ” is drawn on y£p. These boundary lines are
1. edges,
2. segments of edges,
3. lines which are noted on the faces of /*p.
An edge angle is now associated uniquely with each of these
boundary lines s :—If s is an edge (or part of an edge) of jip then the
edge angle is just the angle included by the two faces intersecting in
the edge. We associate the boundary line angle n with a boundary
line plotted on a face.
We now construct the sum S of all edge angles and calculate it
by two different methods. If we construct from the /ip the poly
hedron 0 , then certain boundary lines will lie on the edge eT of 0 .
The sum of these edge angles is an integral multiple of the edge angle
at of 0 . Other boundary lines which belong to edges of /?p can be
placed in the interior of 0 ; the sum of the corresponding edge angles
is an integral multiple of 2n. The boundary lines lying in the faces
of /{p finally yield an edge angle sum which must be a multiple of n.
Hence we have for the sum S of all edge angles
n
S= Z vn<^n + N 1it.
N= 1
t They are measured in radians.

EQUIVALENCE OF POLYHEDRA BY DISSECTION 75
Similarly, the construction of 3 from the sub-polyhedra yields
m
S= Z HmPm + N 2ti.
M = 1
If one forms the difference of these two representations for S then

relation (1) is obtained.
3. E xamples on D e h n ’s T heorem
If the second polyhedron is a right parallelepiped then the necessary
condition (1) for the equivalence by dissection is simplified to
t
N= 1
W« = N *.~ .
2
(2)
One easily sees that this condition is satisfied for Juel’s pyramids.
These are pyramids whose base is a square with side of length a and
whose vertex is on the normal to the base through the point of
intersection of the diagonals of the base (at the height a/2). The 8
edge angles of such a pyramid are
a1 = a2 = a3 = a4 = -, oc5 = a6 = a7 = a8 = —. (3)
4 3
Since one can combine 6 such pyramids to form a cube it follows
at once that the edge angles which do not belong to the base are
equal to 2n/3. The bases of the pyramids become side faces of the
cube. The edges with the angles a 5, a 6, a 7, a 8 lie on the principle
diagonals of the cube. At each half principal diagonal three edges
of congruent pyramids adjoin and the sum of their angles is 2n;
therefore each edge angle must be equal to 2n/3.
Now from (3) it follows that
Z av + 3 Z av =
V = 1 v = 5
The necessary condition (2) for the equivalence by dissection with a

cube (and/or a right parallelepiped) is therefore satisfied.! We shall
show later (p. 79) that each Juel’s pyramid is actually equivalent by
dissection to a cube.
First of all, we now give an example of two pyramids with
the same base and equal height for which condition (1) is not
satisfied and which therefore cannot be equivalent by dissection.
t M. G o l d b e r g [VI 13] h a s c o lle c te d fu rth er e x a m p le s o f p y r a m id s w h ic h
are e q u iv a le n t b y d isse c tio n to a c u b e .
76 U N S O L V E D A N D U N S O L V A B L E P R OB L E MS I N GE OME T R Y
Let £ be Juel’s pyramid with base edge of length 1 and g? a

pyramid with the same base and equal height but with its vertex
placed at the height \ above the midpoint of a side of the square
(fig. 28). Let the edge angles av of gP be numbered in the same way
as the edges of gP in fig. 28. Then
7t n
ay = a3 = - , a4 =
4 2
From tan a 2 = \ we calculate for later use
sin a2 = \ y j 5, cos a2 = 5. (4)

From the solid angle formed by the edges 1, 4 and 5 it follows that
a 5 — a 8 = n/2 is also true. Finally we calculate a 6 = a 7 from the

solid angle with edges 1, 2 and 6. Since the edges 1 and 2 of this
angle are perpendicular to each other one finds
cos a 6 —cos oq cos a2 — ^/lO; s in a 6 = i V 15- (5 )
Now if g? and # were equivalent by dissection then a relation

of the form (1) would have to be satisfied, therefore
n ( n . 2n
Vi . - + v 2a2 + v6a6 = V7r+ I - + n2 —
or (after multiplication by 12)

ra2 + sa6 — tn
where r and s are certain even numbers and t is integral. Hence we
must have
sin ra2 = ± sin sa 6. (6)
E Q U I V A L E N C E OF P O L Y H E D R A BY D I S S E C T I O N 77
Now f
-,n- 1 n
sin not = it cos" 4 a sin a — I ) cos_,n — 3 a sin a
w
+ ( ” j cos" “ 5 a sin5 a — b .... (7)
If one applies this formula (7) to both sides of (6) then one obtains
a contradiction. For by (4), for even r and s, sin m 2 is rational while
sin sa6 (from (5) and (7)) is of the form - . with integral a and b.
Since ^/6 is not rational, (6) cannot be correct. Our assumption that
there exists a relation of the form (1) for / and 0>, has therefore
led to a contradiction. & and ^ are therefore not equivalent by
dissection.
Theorem 21: There are pyramids with the same face as base and equal
height which are not equivalent by dissection.
There are various other examples of this theorem. We want to
mention yet another:—The regular tetrahedron J is not equivalent
by dissection to a tetrahedron 8 ' 1 which has the same height and
base as 2T but whose vertex is vertically above a vertex of the base.
The proof of this is left to the reader.
By similar methods one can show that no regular tetrahedron,
octahedron, rhombic dodecahedron or icosahedron is equivalent
by dissection to a cube [VI 2],
We have still to give the proof of the fact that Juel’s pyramids
really are equivalent to a cube. For a start it is ascertained only that
Dehn’s condition (1) is satisfied for them. To do this we require
an important theorem by S ydler [VI 3]:—
Theorem 22: Let 8P be a polyhedron and av (v = 1, 2 n
arbitrarily given, positive, real numbers which satisfy the equation
a j + a 2 8 . ■. + an — 1.
f This formula is obtained most simply from de Moivre’s theorem. From
(cos sin a)" = cos na-bf sin hoc = cos" a-i cos"-1 a i sin oc-b...
formula (7) follows by taking imaginary parts and by taking real parts
cos hoc = cos" a — c°s"-2 a sin2 oc+ cos"~4 a sin4 a — b.... (7')
J In the following we call each pyramid with triangular base a tetrahedron.
If we mean a triangular pyramid whose side faces are equilateral triangles, we
add the adjective “ regular ”.
78 U N S O L V E D A N D U N S O L V A B L E P ROBLEMS I N GEOMETRY
Then there are n polyhedra ft, and a polyhedron & with the following
properties:—
1. The f , are similar to FA and have the linear similarity factor av,
2. is equivalent by dissection to a right parallelepiped,
3. & = ft i +/«2 + ••• + /zn3“
First of all we assume that is a tetrahedron P (fig. 29). Now let
D
F i g . 29
the edge AD of & = ABCD be subdivided by the points A x, A 2,

A 3, ..., so that
A A 1: A 1A 2: A 2A 3... = a x \ a2: a3...
is true. Now, through the points A v are drawn planes parallel to
ABC and cutting the triangles A VBVCV on the tetrahedron. Next
we draw through A 2 the plane parallel to the plane BCD and inter
secting A XB X and A XCX in the points £ and F(fig. 29). Finally, let
FG be parallel to EBX parallel to A 2B2.
Thus the frustum A XBXCXA 2B2C2 of the pyramid is decomposed
into a tetrahedron ( / 2 = A XA 2EF) and two prisms (EB1GFA2B2
and A 2B2FGC2Cf). The tetrahedron / 2 is similar to the whole
tetrahedron ST (and to the tetrahedron /„ = DT„_ XB„_ XC„_ x at the
vertex) and has similarity factor a2. Now, if we undertake a
corresponding decomposition in the other parallel layers, we have
for ST a decomposition of the form
+ ^2 T •• ■ + (8)
where / v is a tetrahedron similar to^~ with similarity factor av and
the sum of all the prisms arising in this decomposition is combined
to form 0 . Since each prism is equivalent by dissection to a
rectangle, this is also true for 0 . If, now, 0 is any polyhedron, we
decompose it by suitable cuts into tetrahedra.| Then we have
0 = 0 ’1+&’2 + . . . + f N,
and from (8)
^ a = ^<r1+ ^<r2+---+^<rn+ ^<T-
Now one can again form polyhedra / v similar to 0 from the tetra-
hedra similar to STa and hence it follows that
I 1 />*+ &■
<7 = 1 V = 1
Now, from Theorem 22, the equivalence by dissection of Juel’s

pyramid and the right parallelepiped follows immediately: If # is
a Juel’s pyramid and av = %then we obtain
/ = 6 /+ 0 ,
where J is a pyramid similar to / with linear similarity factor
However one can now combine the 6 pyramids j into a cube and
hence
/ = W + & = &!.
One can also prove the equivalence by dissection directly by
suitable cuts in Juel’s pyramid.
The following theorem presents another application of Theorem
22 .
Theorem 23: Polyhedra which are equivalent by completion are
equivalent by dissection.
The corresponding theorem in the plane for polygons is a simple
consequence of the fact that polygons which are equivalent by
completion are equal in content and polygons with equal content
are also always equivalent by dissection (p. 71). In space this
t See e.g. [VI 2] on the decomposition of polyhedra to tetrahedra.
simple argument is not possible. We prove Theorem 23 as follows:

Let
®+ <# = &, V( j*) = V{0) = Vlt (9)
V(<&) = V{9) = V2.
Then we choose a natural number n for which
n2—1 > (10)

Vi
Now, by Theorem 22, $4 is decomposed into n similar (and con
gruent to each other) polyhedra juf' and a residual polyhedron 0
(equivalent by dissection to a right parallelepiped):—
s4 = nstf’+ 0 , (11)
where
V {s0 ') = A v x = v;. (12)
n
From the residual polyhedron 0t are now cut off n polyhedra
which are similar to 9? and have volume
V i = ~ V 2. (13)
n
Of course, it must first be shown that this is always possible. First
of all let be a tetrahedron = ABCD. Then 0~ is part of a
prism 0 * = ABCDEF which one can construct upon 3 of the 4
edges of 0~. Then we have V(0*) = 3 V(0~). The required separa
tion of n similar tetrahedra 0~' from 0t can now be carried out as
follows:—From 0 one removes n cubes of volume 2>V2, converts
them into prisms similar to 0 * and cuts from these the n tetrahedra
2T' similar to . This procedure is always possible if the volume of
0 is greater than that of the n prisms and this follows from our
choice of the number n. We have
V ( 0 ) = V 1- u V [ = Vl ^ 1 - - )
from (11), (9) and (12) and
3nV± = — V2
n2
from (13). Hence, from (10), it follows that
V(0) = Vt ( l - 1 ) > - V 2 = 3 n V i.
If is any polyhedron then one decomposes it by suitable cuts
into tetrahedra, carries out the decomposition of the individual
tetrahedra and then combines the sub-tetrahedra again into poly-
hedra < €' which are similar to (£. Then we can represent the
described partitioning of 9t as follows:—
St = nW+ST. (14)
Then we have from (11) and (14)
s i = n si' + n W ' + y = n { s i'+ c£’)+ 0 ’. (15)
From (9) it follows for the polyhedra s i ' and %>' similar to s i and
# that
si'+<6' = &' + &'. (16)
Since
= V(sT) = v;, Viffi') = V(<8') = Vi
from (15) and (16) we get
s i = n ( ^ ’ + ^ ') + 0 ' = nda' + inW + Si) = n8B' + W. (17)
If one applies to £8 the same decomposition as applied to s i (from
Theorem 22), one obtains
& = n&' + 9r. (18)
From (9), (16) and (18) it follows that ^2 and 01' have equal volumes.
But right parallelepipeds with equal volumes are equivalent by
dissection. Thus 01 = 0t' and from (17) and (18)
s i = n@' + 0l = 8
is true.
We will now use both Theorems 22 and 23 to derive necessary and
sufficient conditions, first obtained by H a d w i g e r , for the equivalence
by dissection of polyhedra.
For over fifty years it has been known that a Dehn’s condition of
the form (1) must be satisfied if two polyhedra are equivalent by
dissection. Whether this condition is also sufficient is so far still
not known. Now Hadwiger’s condition which we derive (Theorem
26, p. 88) is indeed necessary and sufficient but, unfortunately, it
cannot be used to decide whether a given pair of polyhedra of equal
content are really equivalent by dissection. The reason for this is
that, in the formulation of this condition, we use a set-theoretical
existence statement. Nevertheless, this theorem of Hadwiger is an
important contribution to the theory of content. For its derivation
we develop a kind of “ algebra of polyhedra ” .
4. T he A lgebra of P olyhedra
In the following we denote by X0 a polyhedron similar to 0 with
linear similarity factor X. Therefore for the volume V(0)
v(X0) = ;.3 . v{0>) (19)
is true. In general
(X + p )0 + X0 + [L0.
This can be seen, e.g. for X+p. = 1 using Theorem 22: thence
1.0 — fi2 + 0 = ^ 0 T h-0T 0 .
We proceed to develop an algebra of polyhedra for which the
distributive law holds and also multiplication by negative numbers
is meaningful. First of all we define multiplication by 0 by means
of the “ empty ” polyhedron 6 :—
0 0 = 0, 0 +6 = 0 .
Since each right parallelepiped is equivalent by dissection to a
cube, we can also formulate Sydler’s theorem (Theorem 22) as
follows:
n
Theorem 22a: Let av> 0 and Y av = 1. Then it is true for each

V = 1
polyhedron 0 that
0-/M+ £ av0 (20)

V = 1
where °U is the unit cube, A>0 and
x2 = v { 0 ) ( i - «v)-
Next we state the definition:— Two polyhedra 0 and 2L are said

to be equivalent modulo C
JU,
0&£> mod ^ (or simply 0 2),
if there exist two numbers a and fi for which the relation
0 + u.W = 2 + pW, aÔ , 0, a . p = 0,
exists. In other words:—Two polyhedra are considered to be
equivalent if the one with the larger volume can be made up from
parts of the other and of a right parallelepiped. With this new
notation (20) assumes the simple form :—
0& Y a^ - ( 20')
V = 1
There are the following rules for equivalence of polyhedra:
1. 0X0.
2. From 0 x £L it follows that 2 .x 0 .
3. From 0 x 2 and 2 x 0 it follows that 0 x 0 .
4. From 0 x 2 and 0 4 - 0 x 2 4 -0 , it follows that 0 x 0 .
These rules are proved directly from the definition. We will only
prove rule 3. Let
^ + oc^ = 24- P^, 2 + y<%= 04-SW
an d /?^ 7 , say. Then
P°U = yW + ZW (with p 3 = 73+ £3),
therefore
0 4-<0U = 2 + yaU+ ^% = m + 5w + &
or
04-aW = 04-qW , (t]3 = S 3 + ^3).
However, this means that 0 ^ 0 .
For each right parallelepiped stf it is true that s / x G for
d + m = ®4-Ml.
From Dehn’s theorem (Theorem 22) we have the result that there
exist polyhedra which are not equivalent by dissection to a cube.
We can formulate this statement in the new terminology as follows:
Theorem 20a: There exist polyhedra 0 which are not equivalent to
the empty polyhedron:—
0X&.
In order to have negative factors also in the algebra of polyhedra
for use in other investigations, we introduce the concept of the
negative polyhedron by the rule
- 0 = \W -0 ° , (21)
where X°U is a sufficiently large cube which covers 0 so that each
point of 0 is an interior point of the cube and 0 ° denotes the open
kernel of 0 . In this way the usual subtraction of sets in (21) means
that —0 is the closed set of points which is obtained from the cube
}.°U when the interior points of 0 are removed.
Of course —0 is not uniquely determined by 0 . If one chooses
another cube 0°U' (whose faces need not be parallel to those of
Xtfl) one obtains another negative polyhedron ( —0 )'. However,
the two negative polyhedra (-ZP) and ( —â)' belonging to & are
equivalent. For, from
/2?/ = 2 !, h'tfl' = 2 2, XJU « k ' W x (9,
therefore
2j
it follows from rules 4 and 1, that
0)j ✓ >_/ G) •
«z
Hence in our algebra of polyhedra negative scalar factors a, P, ...

are also possible: ( —a)2° means simply the negative polyhedron
—(a^3). Now we note the rules arising from this definition and the
earlier statements:—
5. 2 + ( - 2 ) x 0 .
6. The statements
2>+ 2& 0l and P?k,3/1—2 = PA+ { —2)
are equivalent.
7. a ( ^ + 2)& a0>+ a2.
8. a(pPP)x(ap)2>.
9. +
We omit the simple proofs for rules 5 to 8 and limit ourselves to
the proof of 9. For positive numbers a and P this rule follows from
Theorem 22a. For, according to it,
<x+ P a+P
Now let a > P > 0 be true and = a, p i = ~ p. Then from
Theorem 22a
( a -p )& + p&>na0>.
Hence, from rule 6, it follows that
(a —P)£A« aSP —pPP or (o^ + p i)2l>tt<xl2>+ p 12>.
If both coefficients are negative, one uses the fact that PP?z2 implies
the equivalence —ZPx —2.
5 . T h e B asis P o ly h e d r a
To derive his theorems on equivalence by dissection H adw ig er
[VI 6] uses set-theoretical methods which Hamel used in similar
form for real numbers.!
f See e.g. [VI 5] p. 135.
Using the axiom of choice one proves in set theory ([VI 5],
p. 130):—Each set can be well-ordered. As is well-known a set is
said to be well-ordered, if an ordering is defined in it so that each
non-empty subset has a first element.
The rational numbers between 0 and 1 can be well-ordered by the
usual enumeration by increasing denominators (for equal de
nominators by increasing numerators):—
J _ J _ 2 i _ 3 . J _ 2 . 3 _ _ 4 ( ' ) 0\
2 5 35 3 5 45 45 55 55 55 55 • • • • \ ^ J
If we write i\/_ r 2 when r2 in the ordering (22) occurs to the right

of r i then the set of rational numbers is well-ordered by this ordering.
If one takes the usual ordering as basis, instead of this, then each
non-empty subset of the set of rational numbers does not necessarily
have a first element.
Therefore one can really state the well-ordering for the rational
numbers. For the set of real numbers between 0 and 1 (and more
so for complex sets) the well-ordering theorem is purely a statement
of existence.!
Now let {&>} be the class (non-empty by Theorem 20a) of all
polyhedra which are not equivalent by dissection to any right
parallelepiped: ZP&0. Let this set be well-ordered
&KaL & K L & K L ~ .. (23)
We now say that a polyhedron e {£?} is “ pre-dependent ” if
there exists an equivalence relation of the form
(24)
where in the (finite) sum only 0>K for which SPk/J ? is true, are to
appear.
Next let {stf} be the class of polyhedra which are not pre-dependent.
Certainly t?K belongs to it and therefore {stfj is not empty, {stf}, as
a subset of a well-ordered set, is itself well-ordered with respect to
the same ordering relation. These polyhedra belonging to {stf} are
called (in analogy to Hamel’s “ basis numbers ”) basis polyhedra.
Theorem 24: The basis polyhedra are linearly independent, i.e. from
£ «*v^ r v~V (2 5 )
V = 1
t In an intuitionistic theory (see e.g. [VI 10]) non-constructive existence

statements are inadmissible. Under such hypotheses the conclusions of H amel
and the statements made in this section for content theory are not valid.
it follows that
°V1 = 2a T = = n
=0.
To prove this we can assume that n^.2. For sd x «& would
certainly contradict the definition of the basis polyhedra: {«*/} is,
of course, a subset of {0}. However, for nSffi2 and atn ^ 0, it would
follow from (25) that
"e
v= i aT
'n
But then would be, contrary to the hypothesis, pre-dependent.
Theorem 25: Each polyhedron 0 e {0} has a unique representation
as a finite | sum of the form
(26)
with (positive or negative) real coefficients <xx.
We show first of all that there exists at least one such representa
tion. The proof is carried through indirectly:—Let { 0 } be the
(non-empty) subset of {0} for which a relation (26) does not hold.
As a subset of a well-ordered set {0 } is well-ordered. Let the first
element of this set be 0 If 0 X were a basis polyhedron then we
would have 0± = therefore also 0 1&stfXp and hence a
representation of the form (26). Therefore we must suppose that
0 t does not belong to {s/}. However, 0 X is then pre-dependent,
i.e. there is a representation of the form
~ I /^ K (27)
K
where 0 Kf _ 0 1. Since 0 X is the first element in our well-ordering

for which a representation (26) does not hold, the 0 K appearing in
(27) are all representable as follows
Z 'V ^ p ’ d pL 0 K.
P
0 i « ZK Zp Pk^ kp^ p ~ ZP ^p L01

where
yp Z fix^xp'
K
Thus we would have obtained for 0 x a representation of the form

(26).
t Ali the sums appearing in this section are finite.
Now let there exist two representations of this kind for 0 . Then
one would have, say,
Z at ^r« Z PfS**'
t = 1 T = 1
where some aT or j8Z respectively, could vanish. Hence it would

now follow that
Z
t = 1
in contradiction to Theorem 24. Thus Theorem 25 is completely

proved. From the definition of equivalence it is possible to replace
the representation (26) by another in which only positive coefficients
appear:—
Theorem 25a: There exists a class {x/r} of basis polyhedra such that
to each polyhedron 0 e {0} can be related two sets of non-negative
(iuniquely determined) coefficients (£, f ) and (rj, y ) (f . t] = 0,
. yx = 0) for which the equivalence by dissection
0 + q® + Z y X - + Z&<< (28)
is true.
For this one only needs to write (26) so that the positive coefficients
at = /?Tare on the right and the negative a, = —yt on the left of the
equivalence sign:f
0 + I yTs / xx Z/frj / t.
Hence, immediately from the definition of equivalence one obtains
(28) and there remains only the uniqueness of the coefficients £ and
rj to be demonstrated. However, this follows from the equation for
the volume
V (0) = £3- i , 3+ Z ( ^ _ 73)FG<).
Then, since £ . rj = 0, £ and r) are uniquely determined.
The coefficients at in (26) are functionals of the polyhedra 0 .
To indicate this relation we write
= X rW (29)
Each basis polyhedron e {s/} gives rise to a functional iff? ).
If {s0f\ does not appear in the representation (26) for 0 then
Xz(0 ) = 0. For each 0 there are, from Theorem 25, only finitely
many functionals different from 0. Now we can formulate
t Note that for each number r at least one of the two numbers yT, j9r vanishes.
88 U NS OL VE D AND UNSOLVABLE PROBLEMS IN GEOMETRY
the criterion which we mentioned earlier for the equivalence by

dissection of polyhedra as follows:—
Theorem 26: Two polyhedra are equivalent by dissection if and only if
K&) = m x xl& ) = x M (3 0 )
is true for all f t\

The necessity of condition (30) follows immediately from Theorem
24 and the fact that polyhedra which are equivalent by dissection
are equivalent. That (30) is also sufficient for the equivalence by
dissection follows from Theorems 25a and 23:—From
&'LctX'Srfx, 2 ^h 'Zocxj2 x
we obtain from Theorem 25a the equivalence by dissection
& + rfU + 12yxs2x = 2 + rfli + 1yxs2x.

This means, however, that & and 2 are equivalent by completion.
Hence, from Theorem 23, it follows that & = 2.
The criterion expressed in Theorem 26 has the satisfactory
property of being necessary and sufficient. However, it cannot be
used in any way to decide for two polyhedra, which are given equal
in content and with their geometrical configuration (edges, angles,
etc.) known, whether or not they are equivalent by dissection.
Because the set of basis polyhedra is not known, one does not know
at once whether {srf} is denumerable or not. Nevertheless Hadwiger’s
theorem (Theorem 26) is of great significance in the theory of
content. From it we obtain important statements about the
cardinality of sets of polyhedra which are equal in content but not
equivalent by dissection, and also about the theory of functionals of
polyhedra. From Theorem 26 there follows at once
Theorem 26a: There is a system of polyhedra which are equal in
content with the cardinality of the continuum and these are not
equivalent by dissection in pairs.
Now let c be an arbitrary, positive, real number. Then, in the
representation
0>= t (" ^ 2 )
T = 1
f We w rite in o u r rep resen tatio n s in w hich, o f course, always only finite sums
ap p ear, a t o r Xx(&) w ith an integral index T. H ow ever this does n o t m ean th a t
th e set o f basis p o ly h ed ra is denum erable.
we choose the numbers at f so that
V(&)= t
X = 1
at3n < ) = c.
For the ar there are as many possibilities as the cardinal number
of the continuum. All the polyhedra obtained in this way for
different series of numbers at (t = 1, 2, 3, n) are equal in content
but not equivalent by dissection, since from
£
X = 1 X = 1
it would follow (Theorem 24) that a' = a" It might be noted that
Theorem 26a has also been derived by Sydler [VI 3] directly from
Theorem 22.
Another interesting consequence can be drawn for the theory of
functionals of polyhedra from Hadwiger’s theorem. The functional
(f)(0) = Xx(&) defined in (29) has, as one sees immediately, the
following properties:—■
1. <j)(0 ) = (f)(1) when 0 = 1.
2. (f>(0 + l ) = (f)(0) + (f)(1).
3. (f)(X0) is a continuous function of A for A>0.

The last property follows from f(X 0 ) = X(f>(0).
Another functional has these three properties, namely the volume
V(0). Now for the volume two other properties, which y^(0) does
not have, can be added:—
4. V(Û) = 1 for the unit cube,
5. V (0)Ô .
Property 4 is not particularly important since one can always
show, using a suitable numerical factor At, that x*(0) = Xxxx(0) has
this property. On the other hand 5 is an important characteristic
of the volume. One can show that there is exactly one functional
(f)(0) which has the properties 1, 2, 4 and 5, namely the volume
V(0) [VI 9]. In this way one obtains a particularly elegant basis
of elementary content theory.
We shall now show that the hypothesis 4>(0)^±O for the volume
is quite characteristic:—If one does without this condition then
t We limit ourselves to numbers at ^0.
there exists, besides the volume, the functional %T(^ ) which has the
properties 1, 2 and 3. H a dw ig er showed further [VI 6], that each
functional <f> ( w i t h the properties 1, 2 and 3 can be written in the
form
<f>(0>) = cV (^) + Z cjJ& ).
That means that V(&) and the coefficients xz(&) are essentially the
only functionals with the properties 1, 2 and 3.
Thus all these properties are important: the problem of finding
a sufficient criterion which can be applied for the equivalence by
dissection of polyhedra which are equal in content has not yet been
solved.
However, S ydler [VI 7, 8] achieved a noteworthy step in this
direction by a complicated decomposition process. He showed the
following: If Dehn’s condition for two polyhedra is satisfied, their
difference is equivalent to a polyhedron whose dihedral angles are
multiples of n/4. If one could manage to show also that such poly
hedra are equivalent to a cube, one would have proved that the
necessary condition (1) (together with equality of the volume) is
also sufficient.
Chapter VII
THE DECOMPOSITION OF RECTANGLES

INTO INCONGRUENT SQUARES
1. T h e P r o b l e m
In the 19th and 20th centuries attempts to demonstrate the unsolv
ability of some problems with which mathematicians had struggled
for many centuries, were successful. Among these problems were
the squaring of the circle (Ch. X), the parallel postulate and the
trisection of the angle (Ch. XI). The idea that the preceding
problems might perhaps be unsolvable evolved only after generations
of mathematicians had sought in vain for the solutions.
Sometimes it is in exactly the opposite way that one comes to a
new understanding. When the question emerged whether it might
be possible to decompose a square into incongruent sub-squares,
many mathematicians were of the opinion that this problem was not
solvable. So it was reported in 1930 by K r a it c h ik [VII 2] (p. 272):
“ ...il n’est pas possible de decomposer un carre donne en un
nombre fini de carres inegaux deux a deux. Cette proposition
quoique non demontree parait etre vraie. Elle nous a ete
communiquee par M. Lusin, professeur a Moscou.”
R. S p r a g u e also tackled the problem, first of all with the aim of
proving its unsolvability. But, he then found | a decomposition
of the square into 55 different sub-squares [VII 4]. A year later
an extensive work by R. L. B r o o k s , C. A. B. S m it h , A. H. S t o n e
and W. T . T u t t e was published and this contained inter alia the
decomposition into only 26 incongruent sub-squares [VII 7].
t In the literature it is sometimes reported (e.g. [VII 10]) that the work [V II7]
gave the first decom position o f a square into incongruent sub-squares. This
is not true, however. According to [VII 10] the first decom position o f a rect
angle into incongruent sub-squares w'as found in 1925. The book [VII 2] by
K raitchik , 1930, contains as an example only one decom position o f this kind
in which there appear pairs o f equal squares and also two equal squares with
a side in common.
91
In the mathematical journals dissection problems of the kind
discussed here are usually listed under “ recreational mathematics
But it is worth noting that our problem has a technical significance
which is not completely obvious at first sight. In the works [VII 7]
and [VII 10] the mathematical problem is interpreted immediately
in terms of electrical engineering and in the proofs there is a mixture
of mathematical and physical arguments. In this chapter we tackle
the problem first of all in purely mathematical terms and only at
the end discuss the possibility of a technical interpretation.
We start with some conventions for a suitable notation. Fig. 30
shows the decomposition of a rectangle with sides in the ratio 33:32
into 9 incongruent squares and fig. 31 that of a square into 24
incongruent squares. Since one cannot very well print diagrams
for all the decompositions under discussion, a simple and uniform
lettering for any decomposition of this kind is desired. From a
suggestion of C. J. B o u w k a m p [V II8] this can be done as follows:—
Rectangles are orientated so that the greater side lies horizontally
and the greatest of the four corner squares is placed in the upper
left-hand corner (fig. 30). Then in the decomposition figure we
call the lines which are made up of sides of squares lying in a
horizontal line, horizontal cuts. In fig. 30 these are the lines A VBV
(v = 1, 2, ..., 6). We now make a note of the decompositions by
recording the lengths of the sides of all the sub-squares, where we
begin with the squares lying below the first horizontal cut. They
are listed going from left to right. This is followed by the squares
(always collected in brackets) lying below the subsequent cuts. Our
code for fig. 30 is therefore
(18,15) (7,8) (14,4) (10,1) (9). (1)
Fig. 31 is to be coded as follows
(55,39,81) (16,9,14) (4,5) (3,1) (20) (56,18) (2)
(38) (30,51) (64,31,29) (8,43) (2,35) (33).
Then the following code (3) denotes a decomposition of the square
into 38 different sub-squares given by R. L. B rooks [VII 7]:—
(2132,1440,1348) (256,1092) (692,584,164) (420) (3)
(108,281,615) (840,758,534,627,173) (454) (199,893)
(120,694) (244,310) (217,984) (82,900) (922) (527)
(1587) (287,240) (7,1177) (104,1130) (1026).
THE DECOMPOSITION OF RECTANGLES 93
One should note the difference between the decompositions of
figs. 30 and 31:—In fig. 31 one can combine a subset of the sub-
squares to form a rectangle. In fig. 30 this is not possible. The
square of fig. 31 contains (upper left) a rectangle with sides
(55 + 39) . (55 + 56) which is subdivided into 13 sub-squares. A
dissection into squares in which a combination of the subset o f the
A, B,
15
18
B2
Dp
A, Bi 7
; 4
Bs
14
10
As Bs
c2
F ig . 30
F i g . 31
squares to form a rectangle is impossible is called “ simple ” . The
first decomposition into 55 sub-squares stated by R. S p r a g u e was
not simple, nor was that into 26 sub-squares stated in [VII 7]. The
smallest number n, for which a simple decomposition is known at
present is 38. The coding of this decomposition is quoted in (3).
Decompositions of rectangles into squares which are all different
are called “ perfect ” decompositions. So far we have not taken
any others into account. However, it makes perfectly good sense
to ask about simple, imperfect decompositions. Fig. 32 shows a
decomposition of this kind, which is not perfect but simple, of a
square into 13 sub-squares.
F ig. 32
The following questions now suggest themselves for our con

sideration:
(6) Which is the smallest number n (n', n") of squares used
(6a) in a perfect (simple, simple and perfect) decomposition of
(6b) a square?
(6c) Which is the smallest number iir o f sub-squares in a perfect
decomposition of rectangles ?
(6d) For which rectangles is a perfect decomposition into sub-squares
possible ?
The number n (problem (6)) is still unknown and so is n". On the
other hand n is known. This number is 13:—The decomposition
which is shown in fig. 32 and which is simple but imperfect cannot
be “ improved The problems (6d) and (6c) are also solved.
The answers to these are given at the end of this chapter.
THF. DECOMPOSITION OF RECTANGLES 95
2. T he D efinition of a G raph
A series of results from our range of problems was found obviously
by untiring investigations. However, it is natural to look for a
process by which one can discover all the possible decompositions
of rectangles. As a first step in this direction one can try to limit
the number of possibilities to be investigated.
For this purpose a graphical representation of the decomposition
is introduced, this being suggested by the code-method described
on p. 92. There the squares which were associated with the same
horizontal cut, were collected in brackets. Now, if one replaces
the horizontal cuts AB by points Pv (e.g. the mid-point of each
horizontal cut) and replaces each square by a line joining these
points then one obtains a graph, e.g. the decomposition of fig. 30
is represented schematically in fig. 33a. The numbers attached to
the lines are the lengths of the sides of the corresponding squares.
The lengths of the lines of the graph are immaterial for our purposes.
These depend only on the topological structure (and the noted
lengths of the sides of the squares). It is certainly useful to represent
the points of the graph so that (in a suitable orientation of the
drawing) P v is placed “ higher ” than P v+ 1.
For the later, systematic search for dissections of the rectangle
it is significant that in a graph with V vertices only the first and last
points (Pt and Pv) can be of order 2. An interior point Pm(l <m < V)
of order 2 would be possible only if two equal squares have a side
in common. However, this is expressly excluded. Therefore we
have for the vertices and edges of our graph the inequality
3K ^2E + 2. (4)
Hence, from Euler’s theorem for nets (Theorem 1, p. 9)
V - E + A = 1,
it follows for the number A of areas (“ mesh units ”)
E+ 1
A^ (5 )
3
Now one can introduce similarly the vertical cuts of a decomposi
tion of a rectangle. These are the lines formed from the vertical
sides of squares lying in a straight line. In fig. 30 only one vertical
cut, C2D2, is indicated. Each vertical cut joins two points of two
horizontal cuts A pBp and A aBa and the squares associated to the left
and right of the vertical cut correspond to straight lines which join
Pp and Pa in the graph. Hence there is established a unique relation
between the interior vertical cuts and the areas of the graph. The
number A appearing in (4) and (5) is also the number of interior
vertical cuts. If we take into account the two vertical sides of the
given rectangle then we have altogether v = A + 2 vertical cuts.
Now one can, of course, correlate to the rectangle, oriented
according to our convention, yet a second graph in which the points
correspond to the vertical cuts of the decomposition. The lines of
this graph which is dual to the first graph again correspond to the
sub-squares, but the mesh units of the net here correspond to the
horizontal cuts. Fig. 33b shows the graph which is dual to fig. 33a.
If we denote the vertices, edges and areas of the second graph
(dual to the first) by V1} A t , E l then we have, analogous to (4) and
(5 ),
3Vt ^ 2 E + 2 (4')
E+ 1
( 5 ')
3 ’
where we have already allowed for the number of edges in both
graphs being equal: E — £ ). Further the number of vertices Vt
in the second graph is equal to the number of vertical cuts, therefore
V1 = A + 2 = v. (6)
T H E D E C O M P O S I T I O N OF R E C T A N G L E S 97
Similarly we have for V

V = A l+2= h, (7)
where h is the number of horizontal cuts in the given decomposition.
Finally, from (4), (4'), (5), (5'), (6) and (7) one obtains for the
number of mesh units of both graphs the estimate
£+1 < 1 ^ 1 2E—4
( 8)
3 3
This inequality together with Euler’s equation (Theorem 1) makes
it possible to determine the possible numbers for the areas (“ mesh
units ”) and vertices of the graph for the decomposition of rectangles
into a given number E = E t of squares. Since only integral values
come into the question, we have collected for E — 5, 6, ..., 13, 14
the possibilities to be considered in the following table:—
E A V E A V
5 2 4 10 4; 5 7; 6
6 ------- — 11 4; 5; 6 8; 7; 6
7 3 5 12 5; 6 8; 7
8 3; 4 6; 5 13 5; 6; 7 9; 8; 7
9 4 6 14 5; 6 ; 7 ;8 10; 9; 8; 7
3. S o m e C a l c u l a t i o n s fo r D e c o m p o s it io n s
Now we must investigate which of the graphs possible from the

table actually arise in a dissection of rectangles. In this connection
we note the rules to which the side numbers of the squares are
subject. These numbers are always written by the side of the lines
of the graph (figs. 33a and 33b). The squares “ sitting on ” a
horizontal cut give the same sum of side lengths as those “ lying
below ” such a cut. The corresponding statement is true for the
squares placed to the right and left of a vertical cut. Since the
horizontal cuts correspond to the points and the (interior) vertical
cuts to the “ mesh units ” of the graph, there thus arise the rules for
the side lengths of the square.
In order to be able to formulate these rules conveniently we
agree that a negative side number x pa is correlated to a line
PpPa of a graph if p > a and a positive number if p<<x. Therefore,
in fig. 33a, we have written down the number 4 at the line P 3P 4;
accordingly —4 would be the corresponding side number for P 4P 3.
98 UNSOLVED AND UNS OL VABL E PROBLEMS IN GEOMETRY
Further we want to say that the side numbers which are correlated
to all the graph lines PpPav intersecting at Pp belong to an interior
p o in ty of a graph. Similarly, we correlate a mesh unit PniPni...P„Pni
of the graph with all the numbers xnv„v+1 for the lines P„vP„v+i,
always observing the rule for signs. Then there follows at once
from the definition of side numbers and the rule of signs:
The algebraic sum o f the side numbers corresponding to an
interior point or to a mesh unit is equal to 0:
9 h
E X P°V = 0? E *nv„v+1= 0 , «*+! = «!,
V = 1 V = 1
(9)
where g is the order o f the point, h the number o f vertices o f the
mesh unit.
Now the search for decompositions of rectangles (or possibly of
squares) can continue as follows:—One draws the graphs which are
topologically possible from the table on p. 97 and tests whether a
“ practicable ” solution of the diophantine equations (9) exists for
them. For example, solutions for which one x ik = 0 are not
“ practicable” . Also solutions with x ik = x ik+l are excluded
because they would lead to a dissection with two equal, adjacent
squares. Then this shows very quickly that most of the graphs
which are topologically possible do not give practicable dissections.
Let us take as an example the first case of our table E = 5 (fig. 34).
One can easily see directly that this figure cannot be the graph of
dissection of a rectangle. However, we want to derive the reason
from our statement about the algebraic sum of the side numbers in
order to show how one can also achieve one’s purpose in complicated
cases.
Let it be assumed (according to our convention about the orienta
tion of the rectangle) that x 13 > x 12. Then we must have
*21+ *23+*24 = 0> *31 + * 3 2 + * 3 4 = 0, (10)
* 1 3 + * '3 4 + * 4 2 + * 2 1 = 0-
Hence, it follows that
*12 = —*2l = * 2 3 *b* 2 4 > * 2 4 ; * 13 = * 3 4 —' * 2 3 < * 3 4 )
and, since * i3> * i2>therefore
*12 + *24 < 2 * 12< 2*! 3 < * 1 3 + *34
in contradiction to the 3rd equation of (10).

Now one can investigate the other graphs in a similar way. The
systematic investigation of the graphs for E = 6, 7, ... is not
THE D E C OMP OS I T I ON OF R E C T A N GL E S 99
particularly difficult but quite lengthy work. C. J. B o u w k a m p has
carried out this investigation completely for E = 5, 6, 13. There
it turns out that practicable solutions are possible first for E — 9.
Therefore there always appear for dissections of rectangles into
squares of different side lengths at least 9 squares. Fig. 30 shows
an example of such a dissection with E = 9.
With increasing E there are always more possibilities. In the
work of B o u w k a m p the enumeration of the
possible solutions by his code-procedure for
E = 13 occupies several printed pages. There
are ([VII 8], p. 1299) 585 decompositions of
rectangles into 9, 10, 11, 12 or 13 incongruent
squares. Among these rectangles there is
not a single square. On the other hand there
also exists among the 72 imperfect (but
not trivial) dissections for 9 ^ £ ^ 1 3 the
simplest “ simple” (but not perfect) dissec
tion into 13 squares which is represented in F ig . 34
fig. 32.
One can build perfect dissections of squares from simpler dis
sections of rectangles. The method was already used by R. S p r a g u e
in the construction of his first example [VII 4]. The basic idea is
the following: One decomposes a square (fig. 35) into two unequal
squares 2^ and 2 2 and two rectangles and &2. If one succeeds
in decomposing the rectangles and &t2 into squares which are all
different so that none of the squares of is equal to a square of
$2 and, in addition, all sub-squares of (v = 1, 2) are different
from 2 X and 2 2, a perfect (but not simple) decomposition of the
whole square is obtained. By means of two different decomposi
tions of the rectangle with sides in the ratio 13:16 R. S p r a g u e
obtained a decomposition of the square into 55 sub-squares in this
way. Using the results of B o u w k a m p one can also obtain by this
method a decomposition into 28 incongruent sub-squares, namely:
there exist two different dissections of rectangles with sides in the
ratio 593:422 into 13 different sub-squares:—
(222,164,207) (80,41,43) (39,2) (37,215) (200,22)
(78) 01)
m 2: (247,154,192) (116,38) (230) (175,72)
(49,67) (103,18) (85).
There are still some other methods (e.g. [VII 8], p. 74) of obtaining
a perfect partitioning of the square from decompositions of rectangles.
All decompositions found in this way are not simple. In the paper
[VII 7] a method is stated for finding simple and perfect decomposi
tions of squares. The method is certainly quite complicated and
leaves open the question of which is the
smallest number n" of sub-squares which
appear in a simple and perfect decom
& 3 t2 position.
4. T h e I n t e r p r e t a t i o n in E l e c t r i c a l
E n g in e e r in g
M, £U It was already mentioned on p. 92 that the
dissection of rectangles into incongruent
F ig . 35 squares is of significance for a problem in
electrical engineering. We have kept the
technical formulation separate from the mathematical considerations
but we do not want to omit this interesting interpretation of our
problem.
Let electrodes be placed at two opposite (horizontally oriented)
sides of a homogeneous metal plate of rectangular form. Let a
constant electrical current flow from one electrode to another.
We consider the unit of current and of voltage to be chosen so that
the horizontal sides of the rectangle are just equal to the current and
the vertical equal to the voltage. Then the current lines run
vertically and the potential lines horizontally. Therefore there is
no change in the behaviour of the current if we construct vertical
cuts in the plate and insert along a horizontal line in the plate a
conductor of resistance zero.
Let us suppose that our rectangle is decomposed into squares
which are all incongruent. Now we cut the plate at the vertical
sides of the sub-squares and apply electrodes of infinitely high
conductivity at the horizontal sides of the plate. Thus the current
is not changed. The ratio V\J (voltage rcurrent) is, with our choice
of units, just equal to 1 in each of the sub-squares. Therefore each
sub-square represents a resistance of magnitude 1. Now one can
concentrate all the horizontal electrodes at a point and roll together
the sub-squares to form a wire. Then there results from the rect
angular plate an electrical network which we have already
represented schematically in the graph. Then each non-trivial
THE DECOMPOSITION OF RECTANGLES 101
decomposition of a rectangle corresponds to a network of unit

resistances in which no simple parallel or series circuits occur. If the
rectangle to be decomposed is the particular case of a square, then
one obtains in this way an electrical network formed of resistances of
x ohms where the whole network also has x ohms resistance. The
rules (9) can then be interpreted as Kirchoff’s laws for electrical
networks.
For the problem in electrical engineering it is unimportant whether
the sub-squares appearing have all different side lengths or not.
What matters is only that the decomposition is non-trivial when
none of the squares has a side in common. Thus the graph belong
ing to the decomposition in fig. 32 represents the simplest net
formed of resistances of magnitude 1 and itself having resistance 1.
5. T h e D e c o m p o s it io n o f R ectangles w it h C o m m e n s u r a b l e S id e s
D ehnshowed [VII 1] already in the first work on decomposition of
rectangles that dissection into squares which are all unequal is
possible only for rectangles with commensurable sides. It was quite
a long time before anyone succeeded in proving that each rectangle
with commensurable sides is also in fact decomposable in the
prescribed way [VII 5].
We restrict ourselves to sketching the basic ideas of a method
of decomposition of this kind. Firstly, one shows that the unit square
can be decomposed in infinitely many different ways into incongruent
squares, where it turns out that in a series
... ( 12)
of decompositions of this kind each square in 2£^ is different from

each square in . One can obtain a series (12) of decompositions
e.g. by starting from a rectangle ^2 which can be decomposed in two
different ways into incongruent squares. Those rectangles assigned
with the decomposition into squares we call ^2° and
Now one obtains a series of pairs of rectangles and on
applying to one of the smaller sides of or 0tv2~ 1 respectively a
similar map of these rectangles (fig. 36)4
f R. Sprague has used in his work [VII 5] a pair of rectangles which yields
a square decomposed into 55 sub-squares. Using example (11) one gets a square
which is decomposed into 28 incongruent sub-squares.
t Also other methods of constructing perfect squares from decomposed
rectangles can be used to construct a series (12) e.g. those stated by Bouwkamp
([VII 8]) p. 74).
1 0 2 2 UNS OLVE D AND UNSOLVABLE PROBLEMS IN GEOMETRY
Besides, in each of the rectangles, the decomposition into sub
squares must be plotted. From the pairs of rectangles are now
constructed squares 2.v (v = 1, 2, 3, ...) according to the pattern of
fig. 35 and hence one obtains by a suitable similarity transformation
a series (12) of decompositions of the unit square.
Now it must be proved that the sub-squares appear
ing in the different decompositions 2£v are different
from one another, j
In this way one has obtained the possibility of
3% decomposing each rectangle with commensurable
sides into squares which are all different. In fact:—
One only requires to decompose the given rectangle
by equidistant parallel lines into equal squares and
in each of the squares qp (p = 1, 2, ..., m) so
F ig . 36 obtained to apply the decomposition p (p =
1, 2, 3 ,..., m). In this way one has obtained a perfect
but, of course, not simple, decomposition of the given rectangle.
Of the problems (6), (6a), (6b), (6c) and (6d) formulated on p. 94
(6) and (6b) are still open; and the others are dealt with by the
results of sections VII 3 and VII 5. For the numbers n, n and n"
(p. 94) we have the statements:—
1 3< /j^24, n' = 13, 13<n"^38.
f We omit this proof. See [VII 5].
Chapter VIII
UNSOLVABLE EXTREMAL PROBLEMS
1. A T heorem on the T riangle

Every mathematician knows that there are infinite sets which have
no greatest or no smallest element. If one poses, e.g., the problem
of determining the smallest element
greater than zero) then one has formulated an “ unsolvable ”

problem. In this case it would be obvious indeed to say that one
had posed a senseless question.
Now there exist extremal problems for which one would be
inclined, on looking any of them over, to give the opinion that
each must have a solution, but one finds on closer analysis of the
problem that one is looking for the smallest (or greatest) element
of a set SA which, in fact, has no minimum (or maximum). The
problem does not appear to nonsensical as that about the smallest
number statement of the problem is
quite similar.
This is the case, e.g., for the problem (7) on p. 3. Here the set Sf
is the set of surface areas of the domains in which the line of length 1
can be moved in the prescribed way. In fact, there exist domains of
arbitrarily small (but different from zero) content, in which the
rotation is possible. This must be proved in the following. In the
proof of the theorem only the simplest facts of elementary geometry
are used. Therefore the theorem could be made accessible to an
interested fifth form boy.
First of all we prove a lemma on the triangle.
Lemma: It is always possible to decompose a triangle ABC by
transversals from C into sub-triangles CAAi, CAlA 1, CAn^ l B
so that the union of the set (arising from a suitable parallel transla
tion of the sub-triangles) of displaced sub-triangles has arbitrary
small area.
103
To prove this we divide the side AB = c of the triangle ABC
into 2m equal parts (fig. 37a), where the number m can be chosen
arbitrarily large. Let the dividing points lying between A and B
be A u A 2, •••, A 2m-1. Then we draw m+ 1 equidistant lines which
are parallel to AB so that the height hc of the triangle ABC is divided
into m + 2 parts. We number the parallel lines gv (v = 0, 1, 2, ..., m)
so that the suffix increases with the distance from C.
Now we collect the 2m sub-triangles into 2m~l pairs:—The first
pair is = A A XC and A2 = A Â 2C, the next A3 = A 2A 3C,
A4 = A 2A^C, and so on. A parallel displacement of these pairs is

carried out (fig. 37b) so that the displaced triangles cut the same
segment on g r. In fig. 37b only the displacements for the first two
pairs are indicated. The triangles ^4(1)A(11)C(1) and A[2)A (22)C(2)
have, for example, the line Z)(1)Z)(2) as their common intersection
with g 1. The surface area of the figure resulting from the translation
is smaller than the sum of the surface areas of the two separate
triangles because the two displaced triangles have, of course, a
parallelogram in common. In fig. 37b the parallelograms
D ^ D (1)A[2U [ a n d D(4)D(3)A (34)A (33)
belonging to the first two pairs are shaded in.
Now the intersection d(g t) = DD' of the triangle ABC with g x is,
by the theorem on rays, equal to
2c
d(g i)
m + 2’
and the intersection of the 2m 1 pairs with g x is therefore
C
rf(1)G ?i)
m + 2’
unsolvable extremal problems 105
since, by the displacement, just two equal sub-lengths of the inter
section are always made to coincide.
We now calculate or find a limit for d(1)(g) for the other parallel
lines. For the parallels lying between C and g j obviously
d(lX g )^d (g )< d (g t) = ^ - . (1)

m+2
From the translation of the triangles we have, in addition, for the
parallels lying between g t and AB
d<‘Kg) = d(g)-d<‘>(g,) = d (g )- — . (2)
m+ 2
In particular, for g 2 we have
c 3c c _ 2c
da \ g 2) = d(g2)~ (3 )
m+2 m +2 m+2 m + 2’
and for the parallels lying between g x and g2
d>lXg)<d(g2) - d ^ ( g i) = <T)
m +2
Therefore (1) is true for all parallels between C and g 2.
Now the pairs of triangles are displaced into one another in
groups of fours. This must be done so that each two displaced
pairs cut the same length on the parallel line g2. The first two pairs
indicated in fig. 37b are, for example, to be displaced so that the
lines £ (1)£ C2) and £,(3)£ ,(4) are made to coincide. Since the other
translations occur in the same way, the surface area of the 2m~2
groups of four is smaller than that of the 2m_1 pairs of triangles and
the intersection d{-2\g ) of the line parallel to AB with the triangles
now collected into groups of four is likewise reduced by this
displacement process
(4 )
In particular, the intercept of g2 with the groups of four becomes
d,2\ g 2) = } df'K g J = (5)

m +2
The parts of the pairs of triangles lying between g 2 and AB are
combined by the second displacement into trapeziums in which a
parallelogram is twice covered (fig. 38). Thus e.g. v4(1)vl(22)£ (2)£ (1)
and v4(23)v4(44)£ (4)£ (3) (fig. 37b) are displaced together into the
quadrangle A (5)A i^ )E(6)E{5) (fig. 38), where E{2)A 22) goes into
£ (6M^6) and £ (3)^4(23) into £ (5)^ 5). The parallelogram
^ 5M^6)£ (6)£ (5) is therefore covered twice. This means that by
(5) and (2) for parallels g lying between g 2 and AB
d{2\g ) = d{l\ g ) - d {2\ g 2) = d{1\ g ) ------— = d(g)----- — .

m +2 m+ 2
For the straight lines g lying between g2 and g3, in particular, it is
true that
2c 4c 2c _ 2c
d(2\g )< d {g 3) - ( 6)
m +2 m+2 m+2 m+2
F ig . 38
From (T), this inequality (6) is true for all parallels between C and
03-
The displacement process is now continued. At the /ith step
2m~"+1 domains (which each consist of 2"-1 displaced triangles)
are translated so that a pair of these domains always has the same
interval in common with the straight line g ^ At the last stage one
obtains finally a last domain from two domains. From the
argument used above one deduces, in addition, that
2c
dw (g)<
m+2
is true for all parallel lines between C and gfl +1. Then in S>m we
have
2c
d^m\g )< (7)
m+2
for all the parallel lines.!
Hence we have a simple, practicable way to estimate the surface
area of the last domain B m. Of course, is decomposed by the
t The complete proof of (7) can be produced easily by mathematical induction
using the given method. The details are left to the reader.
UNSOLVABLE EXTREMAL PROBLEMS 107
parallel lines into trapeziums and triangles. However, the area of

the trapeziums and triangles can be calculated from the product
of the median and height.
For the area a(S>^) of the part of lying between any two
parallel lines and ^ + 1 (at distance 5 apart), from (7), it is true
that
2cS
m+ 2
Hence we have for the area of Q)m itself
2chc
a (3 m)<
m+2
By choosing a large number m we can therefore deduce that a(Sin)
becomes arbitrarily small. And this is what we had to prove.2
C
2 . B e s ic o v it c h ’s T h e o r e m
Using the lemma just proved we can demonstrate the unsolvability
of the extremal problem (7) (p. 3); namely that the theorem, first
proved by B e s ic o v it c h , is true:—
Theorem 27: There are domains of arbitrarily small surface area in
which a line UV of length 1 can be moved by rotations and transla
tions so that the direction angle between UV and a fixed line is
turned through 360°.
To prove this we use any triangle ABC with incircle of radius 1.
Let / be the incentre (fig. 39). Each of the sub-triangles ABI, BCI
and CAI are now divided into 2msub-triangles with vertex / in the way
described in the proof of the lemma in Section VII 1. One
chooses m large enough to ensure that each of the domains
^ „ 3) corresponding to the triangles ABI, BCI and CAI has
108 U N S O L V E D A N D U N S O L V A B L E P ROBLEMS I N GEOMETRY
an area which is smaller than e/6, where e is an arbitrarily small,

given, real number. We can combine these three domains to form
a domain S>m, say, by the following rule:— is adjoined to
by joining the first sub-triangle of BCI to the last sub-triangle of
ABI. Similarly, 3 ^ is adjoined to Thus we obtain a domain
Q)m whose area is at most equal to 3 .e/6 = e/2.
Now the basic idea of the proof is as
lv /„ follows:—One can move a line of length 1
through 360° around the incentre 7 in
the triangle ABC. Now this motion must
not be carried out in the triangle ABC itself
but in the domain 3>m arising from the
translations and adjoining of domains. As
we shall show this can be achieved only
if, using parallel displacements of the line,
we leave this domain. It will then be
shown, however, that there is a domain
A (containing of surface area
s
F ig . 40 in which rotations and translations can
be performed.
Let A VA Y+J Vand A'Y+lA'v+2rv, say, (fig. 40) be two triangles from
2!m which are obtained by displacing neighbouring triangles of ABI
(fig. 39), BCI or CAL Let IVFVbe the line of length 1 to be moved
and IYS a ray, making with A v+lIv an angle <5 which can be chosen
arbitrarily small. Now let this angle be formed at 7' on IYA'V+t and,
of course, in the half plane in which 7Vlies. Since I YA'Y+1 and IYA Y+1
are parallel, the free arms of the two angles 5 will intersect at S, say.
We now move our line of length 1 from the position 7VFV on the
side A VIV of the triangle A YA Y+1IY to the corresponding position
7'7^5) on the side I YA Y+l of the triangle A Y+lA Y+2T'Y. The motion
is carried out in several steps using rotations and translations (see
fig. 40):—
1. 7v7rv-^7vF^1) (Rotation about 7V)
2. 7v7ri 1)-*’7’v(2)S (Translation)
3. F^2)S-^F^3)S (Rotation about S)

UNSOLVABLE EXTREMAL PROBLEMS 109
4. Translation)
5. / v'irv(4)->/îrv(5) (Rotation about / ^).
This process can be continued through all the sub-triangles of 3)m.
In this way the line, in fact, experiences a rotation through 360°.
There only remains to be considered whether the path of the line
may also be imbedded in a domain of arbitrarily small surface
area. Q>mhas, of course, an area which is smaller than e/2. However,
the line is moved (partially at least) outside of this domain
1. in the displacement IvF$l)~+F$2)S or F$3)S~+IyF^\
2. in the rotation through the angle 28 about S.
In this rotation about S the line sweeps out a sector of area 8.
The area of all the 3 . 2msectors is therefore 3 . 2m . <5. If we choose
8 < >nt+2
then this area becomes smaller than s/4. There still remain the lines
i rv(1)irv(2) or i rv(3)irv(4) which (in fig. 40 this is not the case) can lie
partially outside Let us suppose that / is the total length of
all such lines. Then we can imbed these lines in rectangular strips
of arbitrarily small breadth 77. If we choose n< — then

41
I . t]<
Hence it is clear that the content of the union 3>* of sets 3)m, the
parallel strips and the sectors becomes arbitrarily small:—
fl( S : ) ^ f l ( S J + /iy + 3 . 2 - <5< - + 2 - = e.

2 4
3. S ets of P oints at a n I ntegral D istance A part

Recently in a mathematical journal [VIII 1] the following question
was posed:—
(7a) For the Euclidean space o f dimension d what is the maximal
number md o f points which are at an integral distance apart from
each other and do not all lie on the same straight line ?
To elucidate this problem we must make use of a theorem of
C hoquet and K reweras :—
Theorem 28: I f an infinite set of points 9* of the plane has the property
that the distance between any two points of the set is integral, all
the points of y lie on a straight line I.
To prove this theorem we denote by / the straight line determined
by any two points A and B of the set £f. Now let C be any point of
y which does not lie on / and let A C ^B C , say. Then the difference
of the distances AC and BC
dl = A C —BC ( <AB = d)
must also be integral. Thus the only possibilities for dr are the
numbers 0, 1, 2, ..., d —1. Therefore C (and every other point
D eSA but not lying on /) must lie on one of the d hyperbolae which
are determined by the foci A and B and the transverse axis a =
(<5 = 0 , 1,2, . . . , d - \ ) A
However, a corresponding argument is also possible for the line
segment AC — d* determining the straight line I*. Each of the
points of SP different from A and C must lie either on /* or on one
of the hyperbolae which are determined by the foci A and C and the
axis a* = i<5* (5* = 0, 1, 2, ..., d* —1). But there are only finitely
many points of intersection of the two straight lines /, /* and the
d+d* hyperbolae. However, since SP has to be an infinite set, the
supposition that a point C does not lie on the straight line / is false.
It is not difficult to extend this theorem, which is valid for the
plane, to Euclidean spaces of higher dimension. Therefore the
question noted above as number (7a) is a natural one and yet this
problem is also an unsolvable extremal problem. It will be shown
that, even in the plane, there is no maximum m2 for the number of
points which have integral distances between each two (without all
of them lying on a straight line /). In other words:—There are in
the plane an arbitrarily large number of points P u P2, P„ for
which all the distances PvPfl are integral without all the Pv lying on a
straight line. First of all Theorem 28 holds for an infinite set. We
prove in particular
Theorem 29: There exists in the plane an arbitrarily large number of
points P u P2, ..., P„ lying on a circle for which the distance PvPft
between any two points is integral.
| For § = 0 the hyperbola degenerates into the perpendicular bisector of AB.
UNSOLVABLF, EXTREMAL PROBLEMS
To prove this we start with any isosceles triangle with sides of

integral length. Of course, one can obtain any number of triangles
of this kind by adding a Pythagorean triangle (a right-angled triangle
with sides of integral length) to its reflection in a side which is not
a hypotenuse. Starting from any point P i on the circumcircle ^
of a triangle ABC of this kind we now measure off as a chord one
of the equal sides of ABC arbitrarily often. This produces points
which are all different. One can prove this for example using the
lemma proved on p. 135. Then any two of the points P l5 P 2, ..., P„
obtained in this way have a rational distance apart. We prove this
by mathematical induction. First of all our statement is correct
for the first three points since the triangle P XP2P3 is congruent to
the triangle ABC and therefore has sides of integral length. Now
let us suppose that our statement is true for the points P u P2, ■■■, Pn.
Then we have to show that Pn+l is also at a rational distance from
all the points P 1} P2, ..., Pn. Since the sequence of points P {, P2,
..., P„ is congruent to the sequence P 2, P 3, ..., Pn+l, we have only
to prove that PjPn+j is also rational. However, this follows from
Ptolemy’s theorem | according to which the product of the diagonals
in an inscribed quadrangle P1P2P3Pn+1 is equal to the sum of the
products of each of the two opposite sides:—
P{Pn +l • PlP^^PzPn+i ■P\Pl ~ P 1P3 ■P2-P/i+l-
From the induction hypothesis it therefore follows that P^Pn+x is
also rational.^
Now it is not difficult to obtain a set of points which are at
integral distances apart. Let Q be the least common multiple
appearing in any of the rational distances
p p — Eee
x \J fi
Qvn
and ^ ( 0 the circle with centre U and radius rQ (where U is the
centre and r the radius of # ); let 0 , Q2, ..., Qn be the points of
intersection of the radii UPy with <£(Q). Then { 0 , 0 , ..., 0 }
is a set of points with all the distances 0 0 integral.
One easily sees that the radius rQ of the circle ^ ( 0 increases
without bound with the number n. At most two of the distances
t See e.g. [IV 2] p. 104.
t Another form of this proof by H a d w ig e r [IV 4] avoids using Ptolemy’s
theorem. M O ller [VIII 3] originally gave the above proof.
P iP 2, P 1P3, PiPn can be equal. Therefore among the rational

numbers appear, with increasing n, some with arbitrarily large
denominator. Therefore the radius rQ of the circle &(Q) increases
without bound with the number n. This agrees with the statement
of Theorem 28, according to which the points lie on a straight line
if the set <9^ contains infinitely many elements. F. S teiger [VIII 4]
has shown that one can always find for each dimension d of a
Euclidean space, for N > d, a system of N points which are integral
distances apart and which do not all belong to the same linear
subspace of dimension d —1.
A problem which is still unsolved [VIII 5] and which should be
mentioned in this connection is the following:—
(7b) Is there a denumerable, infinite set which is everywhere dense in
the whole plane and for which every two points have a rational
distance apart ?
One can easily show that the denumerable set {Pn} on the circle
# (of radius r) used in the proof of Theorem 29 is everywhere dense.
On the straight line the points with rational coordinates form such
a set.
Another example of a denumerable plane set of points which are
at rational distances apart is the set SPu to which the points of the
(a, jp)-plane
belong, wherep and q are any integers which are different from zero.
The numbers of this set f which are denumerable on the x-axis
are everywhere dense there and the distance between any two points
of the set is rational. It has not yet been determined whether or not
there is a set of this kind which is dense in the whole plane.%
t R. Sprague is the originator of this example,
f On this range o f problem s see also [VIII 6].
Chapter IX
CONSTRUCTIONS WITH RULER

AND COMPASSES
1. S ome C omments o n the U se of R uler and C ompasses

In antiquity mathematicians were already concerned with con
structions using ruler and compasses. There were construction
problems of this kind which can be formulated quite simply but
whose solutions were not found at that time. One of these is the
trisection of the angle (see problem (8), p. 3) and another the “ Delian
problem ” :—The Delians, so the tale goes, had turned to Apollo
to request help against a plague and the deity had demanded that
they must “ double ” a sacred altar to him. When a cube with
edges of twice the length did not produce successful results, it
became clear to them that by doubling the edges the volume of the
cube had not been doubled but multiplied by eight. And because
the Delians were unsuccessful in their efforts to construct a cube of
twice the volume, they sent a delegation to Plato’s Academy
requesting the solution of the problem.
But the men of the Academy could not solve the problem and,
indeed, in the subsequent centuries right up to the present time
professionals and amateurs (in recent years only amateurs!) have
tried repeatedly to solve the Delian problem or the related problem
of the trisection of the angle. Today we know that both these
classical problems are unsolvable if one is allowed to use only ruler
and compasses in the construction as was customary in classical
times.
To avoid misunderstandings it is certainly necessary to say
exactly what may and may not be done with ruler and compasses.
If today mathematical outsiders still produce the solution of one or
another unsolvable construction problem, it is generally due to the
fact that the rules for constructions “ using ruler and compasses ”
113
are not clearly understood. Therefore, we want to make quite clear

exactly what is meant by this.
Let there be given n (n }z2) points A v (v = 1, 2 ,..., n) in a plane 0>.
The ruler may be used to draw lines through any two points A v and
Ap and the compasses to draw circles with centre at any point A v
and radius A XA Let the points of intersection of these lines and
circles be A„+1, A n+2, •••, A„+m, say. Then one can continue this
process (a finite number of times!), taking now as given the points
A x, A 2, •••, An, ..., A„+,„ and constructing additional points which
are intersections of lines and circles (or lines and lines or circles
and circles).
It will be useful to add what may not be done with ruler and
compasses. The ruler may not be provided with markings which
one “ inserts ” between lines or between a line and a circle or
between circles. One can, of course, make the convention that the
use of an “ insertion ruler ” is to be allowed. However, one must
be clear that this means introducing a new instrument of con
struction. If one refers simply to a ruler, one always means that the
usual ruler without markings is being used. We shall be concerned
exclusively with constructions in which the ruler is to be used in the
way described here.f
The compasses may not be used to discover by trial and error
which points of a given circle and a line are equally far apart. It
may be that able use of the method of trial and error can yield the
same accuracy or better than unskilful use “ according to regula
tions ” of the apparatus allowed. To investigate the possibilities
in construction this is just as unimportant as the fact that a
“ genuine ” straight line cannot be drawn by any means with a
pencil.
On elucidating what may be done using ruler and compasses we
did not exclude the possibility of drawing a circle with centre at any
point and arbitrary radius. In fact, methods involving this play a
role in the bisection and measuring off of an angle and in other
elementary problems. However, we deliberately refrain from using
this possibility throughout, since we can always (for at least two
given points) find among the given or constructed points two whose
joining line is suitable for use as the radius in a problem which is
usually solved using an arbitrary radius.
t On the theory of the insertion ruler and other means of construction such
as set square, parallel ruler and so on, see e.g. [IX 7].
C O N S T R U C T I O N S WI T H R U L E R A N D COMPASSES 115
When at least two points are given, it is also possible to designate
a given or constructed line as “ unit line ” and to introduce a
coordinate system. In this way one can translate all the steps of the
construction into algebraic ones by the methods of analytical
geometry. If one produces two lines, say those determined by the
points A v and Ap and A p and A a> to intersect then one obtains the
coordinates of the points of intersection A v by solving the system of
the two linear equations of the lines A VAP and A pAg. Quadratic
equations have to be solved to calculate the points of intersection
of a circle and a line or of two circles. Therefore the coordinates
of each point A n+i , obtained from the points A 1} A 2, •••, A„ by an
admissible construction step, can be determined from the co
ordinates of the points A v (v = 1, 2, 3, ..., n) by the four basic
arithmetical operations and the process of extracting square roots.
At this stage it is useful to introduce the idea, common in modern
algebra, of the “ number field ” . By this one means a set of numbers
which may be determined from a given set of numbers by addition,
subtraction, multiplication and division (by numbers other than
zero). The rational numbers form a field and so do all the numbers
of the form a + b jc , where a, b and c are rational. In this case one
says that the field of numbers a + b^Jc arises from the field S' ^
of rational numbers by “ adjoining ” J c . |
We want to prove here that the numbers of the form a + b^jc
form a field. To prove that addition, subtraction and multiplication
of numbers of this form lead to numbers of the form r1+ r2j c is
trivial. However, a quotient
a + b^Jc
q=
«i + îV c
may also be written in this way: q = r2-\-rA^jc. One sees this
immediately if one multiplies numerator and denominator of q by
ai —b\\]c. Then one obtains
_ {a + bsjc)(a1- b lsj c) _ aa1- b i bc bai - a b l j~c
^ aj —bjc a\ —b\c a] —b\c
If the coordinates of all the given points A u A 2, ..., A„ belong
t General statements about adjunction are to be found in textbooks of
algebra, e.g. [IX 1 ].
to the field of rational numbers then the coordinates of all

points constructed from them, using ruler and compasses, belong to
a field which is defined as follows:—SFm is obtained from
by adjoining a number where cm_! belongs to a
field dFm—i • Finally, SFx is obtained from by adjoining *Jc0,
where c0 is rational.
A simple example will illustrate this. Let us start with 2 given
points A x and A 2 whose coordinates we take to be (0,0) and (0,1).
Now A xA 2 is divided in the well-known way “ in golden section ” .
The point of division A 3 has the coordinates (0, —i+ W ^ ) - Then
these numbers 0 and —i + i^/5 belong to the field which results
from the field #"0 of rational numbers by adjoining J 5 .
Now one draws the circles with centres A t and A 3 and radius
A 1A 2 = 1. Let us call the two points of intersection A 4 and A 5.
Their coordinates
K -± + W 5 ), ± V i+ W l
belong to the field $F2 which results from by adjoining yjcx
= V f + i \ / 5 ’ ci e
In general, the coordinates of the given points A {, A 2, ..., A„ will
not all belong to F 0. Then we start from the field J 5" which results
from F 0 by adjoining the irrational coordinates of the points A v
(v = 1,2, ..., n). Then each construction using ruler and compasses
starting from the points A l} A 2, ..., A„ leads to points An+l, An+2,
..., A„+m whose coordinates belong to a field SFm which results from
8Fm_ x, S'm-i-, •••> 3F by repeated adjunction of square roots.
Conversely, one can construct each expression of the form
a + b j c from given lines of length a, b and c using ruler and com
passes. One learns this already in school.
We sum up in
Theorem 30: Let A x, A 2, ..., A„ be n given points in the plane and SF
the number field which results from the field 3FQ of rational
numbers by adjoining the coordinates of A {, A 2, ..., A n in a certain
coordinate system. Then using ruler and compasses one can
obtain those points of the plane whose coordinates belong to a
field ,Fm which has the following properties:— F m results from
,Fm_ { by adjoining J c „ - X, where cm_ x e 3Fm_, and, finally, 2Fx
is obtained from F by adjoining a number J c , where c e 2F.
2. Some Construction Problems

Now, how is one to decide in problem (8) (p. 3) or in the classical
problems mentioned in the preceding section whether the condition
of Theorem 30 is satisfied or not? First of all let us formulate
the problem of the quinquesection (or trisection) of the angle
somewhat differently.
Let /_a — /_ABC be the angle to be dissected and A r the inter
section of the circle with centre B and radius BC and of the arm
BA; in addition, let D be the foot of the perpendicular from A±
onto BC. If we now choose BC as unit of length, we have
BD = cos a. Our dissection problem will be solved if we succeed
in constructing a point E on the line BC with the property that
BE = cos a/5. Conversely, if the quinquesection is obtained in any
way by a ray BF then one can always construct BE = cos a/5.
Therefore our problem is solved if and only if the point E can be
constructed.
Now, we denote the unknown line segment cos a/5 by jc and use
the well-known formula |
cos ft sin4 p.
If one replaces sin2 p by 1 —cos2 P in this, it follows that

2 cos 5p = (2 cos P)5—5(2 cos P)3 + 5(2 cos p). (1)
Hence, for y — 2x and 5/1 = a, this becomes
y 5—5y3+ 5y —2 cos a = 0. ( 2)
If we denote by S' the field of numbers which results from the field
S?0 of rational numbers by adjoining cos a, we can now formulate
our problem as follows: Is there a solution of (2) belonging to a
field S' m which is an extension of S '.%
Before we turn to answering this question, we will establish
the equations corresponding to (2) for some simpler problems and
then advance by the easier problems to the answer to problem (8)
(p. 3).
f See in this connection VI (7')-
$ In this chapter by a field which is an extension of a field & we always
mean a field which is obtained from by successive adjunction of square root
expressions as described in Theorem 30.
118 U N S O L V E D A N D U N S O L V A B L E PROBLEMS I N GEOMETRY
Let us denote by (8a) the problem of the trisection of the angle.

Using the formula (see VI (7'))
cos 3/J = 4 cos3 (3—3 cos /3
for the classical problem of the trisection of the angle, substituting
3/J = a and y = 2x = 2 cos /?, one obtains the equation of third
degree
y3—3y —2 cos a = 0. (3)
One is also led to an equation of third degree for the Delian
problem of doubling the cube (problem (8b)). If the length of the
side of the given cube is denoted by 1 and the unknown length of the
cube with twice the volume by a, we have
x3—2 = 0. (4)
Finally, we want to deal with two construction problems which
lead to equations of 4th degree:—
(8c) It is required to draw (using ruler and compasses') an isosceles
triangle with the radius of its circumcircle r and the radius of its
incircle p.
(8d) It is required to construct (using ruler and compasses) an isosceles
triangle with the radius of its circumcircle r and areal.
In problem (8d) it is assumed that the area J is given by a line
segment of length J .
For problems (8c) and (8d) we introduce the following notation:
Let us call the base of the triangle to be constructed AB = a, the
side b, the base angle (3 and the angle at the vertex a. We note that
the circumcircle centre M lying on the axis of symmetry is charac
terised by the fact that the vertex angle between A M and the axis
of symmetry is equal to a. Then we have
sin a = sin (jz —2(3) = sin 2(3 — — (5)
2r
and
2 a
(6)
We put cos (3 = y, - = ti and note that one can construct the distance
r
u from p and r. Since
tan- = 1- ™ S P
2 1 + cos j3
C O N S T R U C T I O N S WI T H RULER A N D COMPASSES 119
it follows from (5) and (6) that
/ ( I - J O 2- - 4 = 0 -
This equation can also be written as follows
... \ u
X 1 -J)+ - = o.
From this form one sees immediately that the equation can be solved
in terms of square root expressions. Hence problem (8c) is
essentially solved.
To examine problem (8d) we substitute
a
x = sin a = —.
2r
Then for the height h of the required triangle it is true that
h = r + r cos a.
Therefore the area becomes
J = - r(l + cos a) = r2 sin a(l + cos a). 00
If we write g for the quotient —, we obtain from (7)

r2
x4 —2 / ix + /i 2 = 0 . (8 )
We have now to examine v/hether the equations (2), (3), (4) and (8)
can be solved in terms of square roots or not. For the equations
of third degree one can answer this question fairly simply by using
a theorem which we now consider.
3. The Proof of the Impossibility of (8a) and (8b)

Theorem 31: I f an equation of third degree with coefficients from a
field S' has a solution which belongs f to a field S' m which is an
extension of S' then it has also a solution which belongs to S'
itself.
Let us suppose for the proof of this theorem that the equation
x 3+ a 2x2+ u 1x + a0 = 0 (9)
f See in this connection footnote J p. 117.
120 U N S O L V E D A N D U N S O L V A B L E P ROBLEMS IN G E OME T R Y
has coefficients from a field and that a solution x t of this equation

(9) belongs to a field & m but not to the field S' m- 2. Then x t can
be written as follows
x t — a + b^Jc
where the numbers a, b and c belong to the field S rm_ l . In the
following we denote the number
x* = A —B y fc (A, B ,C e
as the number conjugate to x = A + B^JC in J 5'm. The numbers
from S' are then conjugate to themselves in S 'm.
One sees immediately the validity of the following rules
(x + y)* = x* + y*,
( 10)
(xy)* = X* . y*.
If /(x) is a polynomial with coefficients from JFm_ l5 it follows
from rules (10) that
(A x))* = /(**)• ( ii)
Since 0 = 0*, this means in particular that, besides x l = a + b^Jc
x 2 = x* = a —byjc
is also a root of (9). The two roots are different from one another
since otherwise x t = x2 = a would be true. However, by our
hypothesis, a belongs to ^ 'm_ 1 and x l must not belong to S rm_ l .
Now the coefficients of a polynomial are, as is well-known,|
elementary symmetric functions of the roots. In particular, for the
coefficient a 1 of (9)
—a l = Xj + x 2 T x 3 .
Therefore, we have for the third root x 3 of (9)

x3 = —(x'!+ x2) —al = —a1—2a.
Thus x 3 belongs to the field S rm_ 1. Therefore, this number can be
written as follows
X3 = U + VyJw, u, v, u- e S rm_ 2.
Hence it follows immediately that x* = u —v\/w must also be a

root of (9) which belongs to the field S rm_ 1. It cannot be identically
equal to x t or x2 since these numbers do not belong to S rm_ l .
t See e.g. [IX 1].
Since (9), being an equation of third degree, has only three roots,
x* = x 3 must be true. However, this means that w = 0 and
x3 e tF m_ 2- This argument can be repeated until one obtains the
result that x3 belongs to the field 2F of coefficients itself. Hence
Theorem 31 is proved.
In particular, it follows that
An equation of third degree with rational coefficients which can
be solved by a constructible number | has also always a rational
solution.
Hence we can immediately conclude that the Delian problem (the
doubling of the cube) is not solvable. To prove this we have only to
show that equation (4) cannot have a rational solution. Then by
Theorem 31 there cannot be a solution from a field !Fm.
Let us suppose that there is a solution x = - of (4) with a and b
b
relatively prime integers. Then we would have
a3 = 2b3.
However, this is impossible since a would be even and therefore b
odd. Then a3 would be divisible by 8 but lb 3 would not be.
Equation (3) for the trisection of the angle contains the parameter
cos a. For a = nj2 there exists the rational solution y — 0. The
angle a = tt/2 can, of course, be divided into three equal parts
using ruler and compasses since the angle n/6 can be constructed.
On the other hand there does not exist a rational solution for
a = 7i/3. In this case we have
y 3—2>y—l — 0. (3')
Let us suppose that (3') has a solution y — - where a and b are
b
relatively prime integers. Then we would have
a3—3ab2—b3 = 0.
But this means that each prime factor of a is also a prime factor of b
and conversely. For a and b to be relatively prime, we would have
a — b — 1. But y = 1 is not a solution of (3'). Thus we have
t A constructible number is a number from a field for which the field
.'W (see Theorem 30) is the field of rational numbers.
shown that (3') does not have a rational solution. It follows from
Theorem 31 that there does not exist a solution for equation (3')
from a field 3Fm. However, this means that one cannot divide the
71
angle - = 60° into three equal parts using ruler and compasses.
This establishes the fact that there cannot be a general method for
trisecting every angle. Our results are summarised in
Theorem 32: The Delian problem (doubling the cube using ruler and
compasses) is not solvable. There exist angles which cannot be
trisected using ruler and compasses.
We now investigate the solvability of problem (8d). It led to
an equation of fourth degree. As is well-known ([IX 1], vol. 2,
p. 74) one can solve these equations by radicals. One first reduces
the equation, using the Tschirnhaus transformation, to the form
y 4 + p y2+ qy + r = 0 (12)
and then obtains the 4 solutions
y = u +v+ w y = u —v —w
y = —u + v—w y = —u —v + w,
where u2, v2 and iv2 are the roots of the cubic equation
z 2+ lp z 2 + (1lt p 2- l r ) z - ls^ q 2= 0. (13)

Equation (8) of problem (8d) is already in the form (12). The
corresponding equation (13) in this case reads
z 3- } g 2z - 1^ g 2 = 0. (13')
Then t (8) has a solution which can be constructed using ruler and
compasses if and only if (13') has a root which belongs to the field
^ ( p ) of coefficients of (13'). For /i = 1, for example, if one assumes
that there is a rational solution z = - (where a and b are relatively

b
prime integers) one obtains the equation
16a3 —4ab2—b3 = 0.
f The reader should consider that this condition is not only sufficient but
also necessary.
From this one argues:—Each prime factor A of a is also a prime
factor of b. Since a and b are relatively prime, we have a = 1. For
each prime factor B of b, B 2 must be a divisor of 16. Hence only
the solutions z = 1, z = and z = \ are possible. Since none of
these numbers is a solution of equation (13') (for p = 1), there does
not exist a rational solution of this equation. Hence we have
established that problem (8d) has no general solution, since the
construction is not possible for p = l.f
Besides problem (8d) there exists a whole series of construction
problems for the triangle which cannot (in general) be solved using
ruler and compasses. We mention as examples the following (for
(a) and (b) see [II 1] and for (c) see [IX 2, 3]):—
(a) To construct an isosceles triangle, given
(a) s the half-perimeter (2s = a + b + c) and the incircle radius p,
(/J) 5 and the circumcircle radius r.
(b) To construct a right-angled triangle, given
(a) the height h from the vertex opposite to the hypotenuse c
and c + a,
(/?) h and a+q. (q is the projection of b on the hypotenuse c).
(c) To draw a triangle, given
(a) a, b and p (incircle radius),
(/?) a, c and wy (angle bisector).
4. T h e Q u i n q u e s e c t i o n o f t h e A n g l e
Theorem 31 referring to equations of third degree cannot be used
to say anything about the solution of equation (2). However, one
can derive a similar statement for this equation of the quinquesection
problem.
. a
Equation (2) was constructed so that it is satisfied by = 2 cos -.
5
One can see immediately that the numbers
fa 2n\ . fa 2n
2cos( i + T >
(1 4 )
. fa 47i\ ^ fa 4n''
y4 = 2 cos - -I----- , y 4 = 2 c o s ----------
V5 5 \5 5 ,
t One easily sees that the case fi = 1, therefore J = r2, is geometrically
realisable.
124 U N S O L V E D A N D U N S O L V A B L E P R O B L E MS I N GEOMETRY
are also solutions of (2). On substituting

a
-
2nk
- |---------------------
5 5
into (1), one obtains
cos 5fS = 2 cos - 2 cos a =
3
/ a 2nkT a 2nk
2 cos - 4------ -5 2 cos + 5 2 cos
[ V5 5 ,
or
yv-5^v + 5^v- 2 cos a = 0
for
v = 1, 2, 3, 4, 5.
The 5 roots jq, y 2, •••, y 5 are all different for a ^ ± n , ^ ±27t.
Now let us assume that the quinquesection of an angle a ^ n, # 2n
using ruler and compasses is possible. Then (2) must have a root
which belongs to a quadratic extension field of the field of
coefficients J27 = J%(cos a) of (2). Therefore let >q belong to a
field of this kind but not to the field
Now we note that, from (14),
H = 2 ( j ) cos 72°—2J 1 - sin 72"
= I - ( - j v i '+ i v '5- (l5 >
We obtain similar representations for y 3, >’4 and y 5 from (14) by

applying the addition theorem. Then the roots y 2, y 3, y 4 and y 5
belong in every case to the field # ”m+3 which one obtains from m
by adjoining ^ 5 , V f + W ^ an(* ^ can haPPen that

one (or some) of these roots already belong to a field where
s< m . This would be possible for example if ^1 - ,V|+W^
and J~5 already belong to J2 7 In each case there exists for every
number v (v = 1,2, 3, 4, 5) a number N(v) such that
( 16)
y v€Îs'(v)’
is true. An odd number of roots must belong to at least one number
N(v). Let s, say, be this number. Now, since (see p. 120) y*
= av—bv\Jcv is also a root if yv = av+ bvyJcv is, this is only possible
if y v = y t f°r some v. If is not & itself we would draw the
conclusion that y(v) also belongs to This contradicts (16)
and therefore at least one root must belong to F . Hence we have
Theorem 31a: An angle a (a ^ n, ^ 2n) can be divided into 5 equal
parts using ruler and compasses if and only if equation (2) has a
solution which belongs to the field F = ^ ( c o s a).
We have proved this theorem by a method which was suited
particularly to equation (2). Instead one can derive Theorem 31a
from the following general statement:—
I f an equation fix ) = 0 which is irreducible | in a field F has
at least one root from a field F,„ J, its degree must be a power of
2 [IX 7, p. 71].
Since (2) is of degree 5, this equation can have a root from F m
only if it is reducible in IF. Hence one deduces the existence of a
root from F .
Let us examine some of the special cases of quinquesection. For
a = n il, from (2) we have
y5—5y3+ 5y = 0. (2')
On the other hand for a = n/3 we have
y s - 5 y 3+ 5y —1 = 0. (2")
Both equations have a rational solution. For (2') it is y = 0, for
(2") y = 1. In both cases the solution denoted by y t (cos 18° or cos
12° respectively) is irrational. The rational solution is y2 = cos 90°
in the first case and y3 = cos ( —60°) in the second case.
Therefore for the angles 7r/2 and tt/3 quinquesection is possible.
This is trivial if one considers that an angle of 18° = \ ■. 36° and
12° = 12° —60° can be constructed using ruler and compasses. On
the other hand there exists no rational solution of (2) for a = cos - a-£.
The proof is similar to that for the examples of equations (3) and (8)
given in the preceding section.
Now it is natural to consider the question whether it is the rule
t For the definition see e.g. [IX I ], [IX 4].
j See footnote f, p. 117.
126 U N S O L V E D A N D U N S O L V A B L E P R OB L E MS IN GE OME T R Y
or the exception to be able to solve the dissection problem. This

question is answered in the following
Theorem 33: The set ST5 (STf) o f angles a(0<a<2n) for which the
quinquesection (trisection) is possible using ruler and compasses is
denumerable. The set ST5 (ST3) of angles for which quinquesection
(trisection) is not possible has the same cardinality as the continuum.
This is easily proved. If quinquesection of an angle a using ruler
and compasses is possible, there exists a root ,xv of equation (2)
which belongs to the field ^ ( c o s a). But this means that ,yv is a
rational function of cos a, say of the form
A 0 + A l cos a + ... + A n cos" a

A v, Bv integral. (17)
B0 + B x c o s a + ... + Bmcos'" a’
If one substitutes (17) in (2) one obtains for cos a an algebraic

equation with integral coefficients. But this means that cos a is an
algebraic number. Thus for all a for which cos a is transcendental
quinquesection cannot be carried out. The same is true for tri
section. Since the set of all algebraic numbers is denumerable, our
theorem is proved.
In addition we note that the denumerable set ST5 (STf) of numbers
for which quinquesection (trisection) is possible in the interval
0 < a< 27i is everywhere dense. For it is obvious that trisection is
possible for all a of the form a = ~n n, quinqesection for a = ~ ~ 71

in addition.
5. T he U n iq u e n e s s of the C o n s t r u c t io n
We know that there is no unique method for dividing a given angle
into three or five equal parts. However, having obtained Theorem
33, it is natural to ask whether there exists a unique method of
performing quinquesection or trisection at least for a denumerable
set {a„} of angles. In the case of trisection this question was first
dealt with by B ie b e r b a c h [IX 7, p. 55] using a beautiful method from
the theory of functions. We now take up the same question for
quinquesection for which the corresponding statements are possible.
Let us suppose that there exists a unique method of performing
quinquesection using ruler and compasses for a sequence {a„} of
angles. Then the uniqueness of the geometrical method would
C O N S T R U C T I O N S WI TH RULER A N D COMPASSES 127
correspond to a unique representation of
Vin) = 2 cos —
5
by constructible numbers. Then for the one root of (2), belonging
to the field .^ (c o s a„), by Theorem 31a (with cos a„ instead of cos a)
there must be a representation of the form
(") = A 0 + A, cos a„ + ... + A k cos* a_ ,, ,

y ~~— -------- 2-------- ------ = / ( cos a„),
Bo + Bi cos an+ ...+ B t cos a„
where the AK and Bx are rational numbers which are independent
of n. Hence, for some k (k = 0, +1, +2), the equation
F(cc„) = 2 cos + — j -/(co s an) = 0
would be satisfied for an infinite set of numbers n. Now, since the

magnitudes of the angles a„ have at least one limit point between
0 and 2n, it follows by a well-known theorem from the theory of
functions that the function
m = 2 cos g + ^ - /( z ) (18)
is identically zero. But /(z) has period 2n. Hence
0 = 2 cos g + ^ - /( z ) = 2 cos g + -A z).

therefore
(z 2kn \ (z 27i(k+l)\
cos - H------ ---- cos - -I----- --------
\5 5 ) \5 5 J
identically in z. But this is not true; therefore there cannot be a
unique method of construction for a sequence {a„} of angles.
However, as B ie b e r b a c h has shown, it is always possible to find
a unique method for a finite number of angles.
Chapter X
CONSTRUCTIONS ON THE SPHERE
1. S t e r e o g r a p h i c P r o j e c t i o n
Construction problems on the sphere (e.g. problem (9), p. 3) can be
reduced to problems of plane geometry using stereographic projec
tion. The properties of this projection are usually described in
detail in text-books on elementary and analytical geometry and also
in the introductory chapters of books on function theory.! There
fore we can limit ourselves here to collecting the rules of
stereographic projection which are important for our purposes.
We call an arbitrary great circle on the unit sphere the “ equator ”
£ and the points of the sphere on the diameter normal to the plane of
£ the “ north and south poles” (N and S). The stereographic
projection used is the central projection of the points of the sphere
from the centre N onto the plane p of the equator £. It has the
following properties:!—
1. There is exactly one image point in the plane p corresponding to
each point on the sphere other than N.
2. Circles on the sphere are mapped into circles or straight lines on
the plane p.
3. Angles are invariant under the projection.
4. If P and Q are diametrically opposite points on the sphere and P'
and Q' are their stereographic images in the plane,
S'P ' . S'Q ' = 1 (1)
is true where S' is the image of the south pole and therefore the
centre of the sphere. (1) is easily proved by applying the altitude
theorem to fig. 41.
5. If rj and ( are the coordinates of the point P on the sphere and
x and y are those of the image point P' in the plane p.
( 2)
t See e.g. [II 4] or [VI 2].

128
C O N S T R U C T I O N S ON THE SPHERE 129
and
___ 2x_ 2y x2 + )>2- l
2’ n= <== (2')
1 + x 2+ y 1 + x 2+ y 2 x 2+ y 2+ 1
Rule (1) can be used to perform constructions in the stereographic
image in the plane. In order to draw, say, the image of the spherical
quadrilateral mentioned in problem (9), we place the centre of this
(regular) quadrilateral at the south pole of the sphere for simplicity.
Then orthogonal great circles g t and g 2 through S are mapped into
perpendicular straight lines g\ and g'2 through S' and each “ circle
of latitude ” I of the sphere becomes a circle with centre S' by the
stereographic projection. Then the great circle arcs AB, BC, CD
S
F ig . 41
9/
and DA, determined by the points of intersection A, B, C and D

of and g 2 with /, form a regular quadrilateral. In order to draw
the images of the great circles through the image points A', B', C'
and D’, from (1), we only require to reflect these points in S' and the
unit circle. Then the images A", B", C" and D" obtained in this way
determine the images of the great circle arcs to be drawn in the plane
p (fig. 42): the image of the great circle through A and B for example
must pass through A" (and B") as well as A' and B ’. Since angles
are invariant under stereographic projection, the angle 2a in fig. 42
is equal to the (obtuse) angle of the quadrilateral on the sphere
itself, f
2. C onstructions w ith S pherical C ompasses and S pherical

R uler
To examine the possibilities of constructions using spherical
compasses and spherical ruler (problem (9), p. 3) in the stereographic
image, it is useful to specify a rectangular coordinate system on the
sphere. Then by (2') and (2) we can calculate the coordinates of the
original points from those of the images, and conversely. However,
we must first test whether it is always possible to construct three
mutually orthogonal great circles from n^.2 given points.
Let us suppose that two points P and Q are given on the sphere.
If these two points lie on a diameter, one cannot construct any other
point. We do not get much further if the distance between the two
points is exactly jn . The circles with centre A and radius + AB
and with centre B and radius BA touch each other in a point C of
the great circle AB and the distance of C from A and B is $n. In
this case it is not possible to construct other points.§ We exclude
these two cases in the following.
t Remember that figs. 10 and 15 are also stereographic images.
| The radius AB = r is the arc AB measured on the sphere, the span of the
compasses is the chord AB.
§ On page 4 it was mentioned that the spherical ruler is equivalent to the
use of ordinary compasses provided one is given the span \ ~2 for the great circle.
First one determines the pole of the great circle and with this as centre draws a
circle with span \/2 . One should note that drawing circles with the span \/2
is not in itself a permissible construction step. This possibility was mentioned
only to show how one can realise the “ spherical ruler ” in practice. Only the
circles (with a span < \ / 2) and the “ lines ” drawn by some means or other
through two points on the sphere are construction elements. From this stand
point it is not possible to construct the pole of the great circle ABC where the
(spherical) distances between the points A, B and C are -jrr.
If (at least) two points P and O are given whose distance apart is
not equal to n or $n, one can always construct three mutually
orthogonal great circles which then determine a coordinate system
on the sphere. One of these is the great circle PQ which we call the
“ equator If the arc PQ is less than the circles with centre P
and radius PQ and with centre Q and radius QP intersect each other
and the great circle, passing through the points of intersection, goes
through N and S, bisects the arc PQ, and is orthogonal to the equator
F ig . 43
8. Let the points of intersection with 8 be A and B (fig. 43). Thus

we have a pair of orthogonal great circles. To find the great circle
orthogonal to these two, we draw the circle with centre B and radius
AP which intersects 8 at P * and Q*. Then the perpendicular
bisector of the arc of the equator PP* (or QQ*) is orthogonal to
AB and PQ. However, since the arc PP* is greater than we
certainly cannot get this third great circle by the method we used
to obtain the second. The procedure here is to measure off the
radius AP from P and P* on the equator repeatedly and so obtain
the pairs of points P XP*, P2P*i ■■■■ After a finite number of steps
one obtains a pair P„P* whose spherical distance apart is smaller
than J-7T. Then the bisector of this arc of the equator can be con
structed in the usual way and is the third great circle which intersects
the equator orthogonally in C and D and goes through N and S.
Then the axes AB, CD and N S determine a rectangular coordinate
system which we can use to represent the stereographic projection
analytically by (2) or (2').
If the spherical distance between the two given points P and Q
is greater than jn , the method must be altered and we leave the
reader to supply the details of this. The coordinates <!;, rj and (
appearing in formulae (2) and (2') can be interpreted as half the

span of the compasses between two given or constructed points. In
fact, if we let P be a point on the sphere with coordinates <!;, r/, £, we
can construct from P and the points determining the coordinate
system f the reflected point F with coordinates rj, ( using
spherical compasses and spherical ruler. Then the span of the
compasses between P and P is exactly 2^. Similarly, 2rj and 2( can
be interpreted as the spans of the compasses between constructed
points.
Now, corresponding to each construction step on the sphere there
is a well-defined operation using ruler and compasses in the stereo
graphic image. Conversely, each step in terms of circles and straight
lines in the image plane has an equivalent on the sphere since, of
course, the circles and lines of the plane are mapped by the inverse
of the stereographic projection into circles or great circles on the
sphere. Since by (2) and (2') the relations between (x, .y) and (£, rj, ()
are rational, the possibilities for constructions using spherical ruler
and spherical compasses can be described from Theorem 30 as
follows:
Theorem 34: Let P X,P 2, ...,P„ be n (n~2i2) given points % on the
sphere with coordinates (£v,r)v, Cv) (v — 1,2, ...,/?). Let F be
the number field obtained from the field F 0 o f rational numbers
by adjoining the coordinates £v, rjv, £v, (v = 1,2, ...,«). Then
one can construct on the sphere, using spherical compasses and
spherical ruler, just the points whose coordinates belong to a
quadratic extension §field F m of F .
y
Since there is the simple relation s = sin - between the span s
of the compasses and the corresponding great circle arc r. Theorem

34 can also be formulated as follows [IX 7]:
Theorem 34a: Given m great circle arcs, one can construct, using
spherical compasses and spherical ruler, just the great circle
arcs whose trigonometric functions are numbers constructible from
the trigonometric functions o f the given great circle arcs.
t These are the points denoted above by A, B, C, D, S and N.

j We exclude the cases when only two points are given and the spherical
distance between them is n or ~v.
§ See footnote t, p. 117.
3 . S q u a r in g t h e C i r c le o n a S p h ere
Having dealt with these preliminaries we can go on to examine the
squaring problem on the sphere, problem (9). We transfer the
centre of the given spherical cap with radius p to the south pole of
the sphere and consider as given the centre S (0,0, —1) and a point
P (sin p, 0, —cos p) on the circumference.
The area of this spherical cap is
i7! = 27t(l — cos p). (3)
On the other hand the area of a regular spherical quadrilateral with
angle 2a (fig. 42) is
F2 = 8a—27i. (4)
Now we have the relation
71
cot a . cot - — cot a = cos r (5)
4
between the circumcircle radius r of the regular quadrilateral and
the angle a. Then, from Ft = F2, it follows that
71
- COS p ( 6)
4
or, from (5),
n
cos r = tan - cos p ( 6')
Now we can certainly draw the required quadrilateral if its circum

circle is known. Therefore we can also formulate our problem as
follows:
We are given the south pole and a point P with coordinates
(sin p, 0, —cos p). It is required to find a point Q with coordinates
(sin r, 0, —cos r) so that relation (6') between r and p is satisfied.
By (2) the stereographic images P' and Q' have the coordinates
sin p
y = o (7)
1 + cos p
and
sin r
= 77 ------ , y t = 0. (7')
1 + cos r
Then the relation
n 1 —x 2\
1 —tan
4 1 + x 2)
( 8)
7T 1 — X 2\
1 + tan
,4 1+ x 2)
between the coordinates of the stereographic image corresponds to

formula (6'). Therefore the construction on the sphere is practicable
if and only if x x in the plane can be constructed from given x using
ruler and compasses. This can be done if and only if the angle
n
= - cos p
4 1+x2
can be constructed using ruler and compasses. One sees immediately
that, for example, the conditions
cos p = 22 -", cos p = 3-1 . 23-n (« = 2 ,3 ,...)
are sufficient for this.
But the construction is also possible for some cases when cos /? is
irrational. Thus for example f = tan- 1 f can, of course, be
constructed although tan- 1 - = c o s"1y is not a rational multiple of n.
We can prove this fact using a lemma which is given in the last
section of this chapter.
Therefore, although the squaring problem in the plane f is
unsolvable, the corresponding problem on the sphere can be solved
in some cases.
Theorem 35: It is possible to construct a regular quadrilateral equal
in surface area to a spherical cap of radius p using spherical
compasses and spherical ruler if and only if tan cos p^j belongs
to a quadratic extension J field F m of the field IF = J%(cos p)
where F is the field obtained from the field F Qof rational numbers
by adjoining cos p.
There exists an analogue to this theorem in non-Euclidean
(hyperbolic) geometry.§
One can also state the converse of problem (9):
(9a) I f a regular quadrilateral which is inscribed in a spherical cap
of spherical radius r and which has angle 2a is given, a spherical
cap of equal surface area is to be constructed using spherical
compasses and spherical ruler.
To investigate the solvability of this problem it is convenient to
T See e.g. [1X7], [X 2], [X 3], [X 4],
J See footnote J, p. 117.
§ See [X 1]. However, the condition formulated there for the practicability
of the construction is not necessary, but only sufficient.
start from the angle a of the given quadrilateral. If one puts
a = from (6) we have
cos p — 2 —co, sin p = ^/4co—co2 —3.
Hence cos p, sin p and thence the y-coordinate of P' by (7) can be
constructed if and only if co belongs to a quadratic extension field J 27*
of the field J27* = J%(cos r).
4. A Lemma
We have still to prove that A in
P = ).n = tan " 1 y = cos- 1 y

is irrational. This can be done using the following general
Lemma: The vertex angles of a Pythagorean triangle are not rational
multiples of n.
To prove this theorem let us suppose that the Pythagorean
numbers u, g and r are relatively prime: i.e.
u2+g2 = r2, (u ,g ,r) = 1.

As is well-known, one of the side lengths of the legs of a right-angled
triangle must be even and one odd. Therefore let us assume that
u is odd and g even. We want to show that A in
/? = Atr = tan -1 - = cos-1 - (9)

u r
is irrational. Let us assume that A = —, (m, n) = 1.

n
1st case: n is odd.
From the formula (see VI (7))
sin rtf = cos" 1ft sin ft— cos"-3/? sin3/? + ... + ( —1)~sin"/?
( 10)
for n odd we obtain in our case
0=
:s(re " V “Y ~ 3
,3 J \ r
n—1
+ - ...+ ( - ! > 2 9

/ \ —
nun~ l = ( ” j u " “ V - + . . . + ( - l ) 2 gn~l = G
with G an even number. However, this is impossible since n . u"-1

is odd.
2nd case: n is even: n = 2 PU, where U is odd.
Hence we put |
mn
y = 2P . P = (ID
~U
and then note that sin y and cos y may be written in the form
sin y = cos y = - tu p
( 12)
j.2 p
where gp is even and up is odd. The validity of (12) is proved from

(9) using the addition theorem and induction. We now apply (10)
to sin Uy and obtain
U u -i
0= 9p_
r 2p
u -i U
U Q p_ 2 £p
j.2 p d--- ••• + ( ~ 1) j.2p
or
u+i
U
U . upU-l uPU —3
dl + ... + ( - 1 ) 2 gp _1 = G*
where G* is an even number and U and up are odd. From this

contradiction it follows that X cannot be rational in (9). Then
g in fin = - ~P is not rational and our lemma has been proved.
t U =£ 1 since U = 1 leads to sin y = sin imr = 0. Obviously sin 2P/? is

different from 0 for all p.
Chapter XI
PROBLEMS IN THE GEOMETRY OF SETS
1. A N ew D efinition of E quivalence by D issection

In elementary geometry the isosceles triangle ABC in fig. 44 is
equivalent by dissection (p. 2) to the rectangle A D C E : the triangle
Fig. 44
ACD is congruent to itself and the triangle ACE to triangle BCD.

Here we want to explain that the equivalence by dissection
A B C = ADCE
does not immediately establish a one-to-one correspondence
between the sets of points determined by the triangle and the
rectangle. Let i, say, be the set of points which lie in the interior
of the rectangle ADCE, / the corresponding set for the triangle
ABC and let / 2 and / 3 be the sets of interior points of AACE,
AACD and ABCD. Then to the dissection
ACD + ACE = ACD + BCD
there corresponds a one-to-one correspondence of onto / 3 and
of / 2 onto itself. The first of these two correspondences is realised
by the congruence mapping which brings the two figures into
superposition.
K 137
138 U N S O L V E D A N D U N S O L V A B L E P R OB L E MS I N GEOMETRY
However, no equivalent correspondence is established between a

and / . The union of the sets / x and / 2 is a proper subset of t since
the interior points of the line AC belong to i but not to / l + / 2-
Likewise the interior points of CD belong to t but not to / 2 + / 3.
It is not any simpler to consider, instead of / and (v = 1,2, 3),
the corresponding closed sets 0 , ST, STx (v = 1,2, 3).
Nevertheless it is possible to translate the equivalence by
dissection in elementary geometry into set theory. For this purpose
we give the definition
Two point sets 0 and 2 (of an n-dimensional Euclidean space)
are said to be “ equivalent by dissection into n parts ” if it is
possible to decompose 0 and 2 into n mutually disjoint sets 0 V
and 2 Vrespectively (v = 1,2, ...,//) so that the subsets 0 Vand 2 V
with the same suffix are congruentf :
0 = &l+0>2+ ... +0n, 0 V
0„ = O (v Ap),
2 = 2 i+ 2 1+...+2n, 2 v2/1= 0 (v + / / ) ,
0 V= 1 Y, v = 1,2
Then we write 0 = 2. If the number n of the parts is not known
(or there is no particular point in knowing this number), one can
say simply that the sets are finitely equivalent: 0 = 2,, where two
sets are said to be congruent if it is possible to transform the one
into the other by congruence mappings (displacements, rotations
and reflections).
This equivalence by dissection into n parts must certainly be
distinguished from the elementary equivalence by dissection of
polygons or polyhedra introduced earlier. In the new definition
any point sets of the considered space, not only polygons and
polyhedra, are permitted for the sets and the subsets. On the other
hand we require here for equivalence a unique correspondence of the
points and a dissection into mutually disjoint sets. In the dissection
of polygons into sub-polygons it was required only that the sub
polygons should have no interior points in common.
We now show that the rectangle ADCE and the triangle ABC
of fig. 44 are also equivalent by finite dissection as just defined,
where we now mean the rectangle or the triangle to be taken as the
f On the addition and subtraction of sets see e.g. [VI 5] or another text-book
on set theory.
P R OB L E MS IN THE GEOMETRY OF SETS 139
set of points which belong to the interior or to the boundary of the
figure. In the following the lines AC, BC and so on are enclosed
in brackets so that we can indicate whether we are referring to the
open, half-open or closed set. Thus [v4C] (in square brackets)
means the line AC including the end points, (AC) is the open line
and [AC) the “ half-open ” line (including A, excluding C).
We now try to obtain the correspondence
of the sets of points under consideration as
follows: The (closed) triangle ACD (2T2)
corresponds to itself, the open triangle ACE
( /, in fig. 44) to the open triangle BCD ( / 3):
2 2 1 ^ 1 = ^ 3 •
Now we have still to dispose of some lines. Let

F be the point on \_BC ] for which the congruence
[/!£ ] = [BF] is satisfied. Then we can decom
pose the sets (the closed rectangle AD CE) and
T (the closed triangle ABC) as follows:—
® = . r 2 + / l + (EC) + (A £\,
F = $ - 2 + / 2 + (DB) + (FB] + (CF],
where the congruence relations hold good:
3r2 =3T2, A = / 3, (EC) = (DB),(AE-] = (E B l
We can summarise this result as follows:—
y = (CF] + &. (1)

This means that the line (CF] is simply left over! However, we will
manage to accommodate it in the rectangle.
For this purpose we consider the following. Let M be the centre
of the rectangle ADCE (fig. 45) and (MN~\ a (half-open) line parallel
to EC and with length less than \EC. We denote this line by a
and by <r(va) (v integral and positive) its image obtained by rotation
about M through the angle va. Now we choose an arbitrary angle
a which is not a rational multiple of n and define the set i f as the
union of sets of infinitely many tr(va) (v = 1, 2, 3, ...):
Sf = cr + <r(a) + <r(2a) + ....
By rotating through the angle —a this set is transformed into
£ f( —a) = <j( —oi) + o + o(oi) + ... = £ A + o ( —a).
Therefore we have
£P = £? + a( —a). (2)
Now, if A D ^ C D in the rectangle 52, CF is less than %EC as one can
easily check. We now choose the line a to be congruent to (CF] so,
by (2), we have
y = s r + (C F '] . (3)
Now 52 is decomposed into <5* and the complementary set 52j.
Then from (3) we have
52 = 521+ ^ = 521+ ^ + (CF] = 52 + (CF]. (4)
A transitivity relation similar to that for equivalence by dissection
in elementary geometry is valid for finite equivalence:
Lemma A: I f two sets are finitely equivalent to a third, they are
finitely equivalent to each other.
i.e. If
j* = a , &=
then
One can carry out the proof of this lemma similarly (by decom
position into subsets) to that of the corresponding statement for
polygons. We omit it here.f
There follows now from this Lemma A by (1) and (4):—
Theorem 36: Each rectangle 52 (ADCE infig. 44) is finitely equivalent
to the isosceles triangle ST (ABC) formed by its two right-angled
sub-triangles.
Thus we have obtained a first theorem in the geometry of sets
analogous to a well-known statement of elementary geometry.
However, one achieves much more in the geometry of sets than
giving a new interpretation to theorems like this. In this theory
there are statements which must appear paradoxical from the
standpoint of elementary geometry. One of these is, for example,
the
Theorem 37: Each sphere in three dimensional space 523 is finitely
equivalent to two disjoint spheres of equal radius.
Perhaps it is necessary at this stage to emphasise that “ finitely
t It is in S ie r p in s k i [XI 3 ], p . 6 9 .
PROBLEMS IN THE GEOMETRY OF SETS 141
equivalent ” says essentially more than “ equivalent ” in the general

set-theoretical sense. Here it is asserted that each sphere can be
dissected into finitely many parts so that two spheres of equal radius
can be formed by rotations and parallel displacements from these
parts. Using S ierpinski’s terminology we call the decomposition
of a set J t into two disjoint sets J t t and J t 2 paradoxical if each of
the two subsets J t v (v = 1,2) is finitely equivalent to J t . One can
derive peculiar statements from the paradox of Theorem 37. For
example the paradox that each sphere of 523 is finitely equivalent to
each cube of 523 (Theorem 39, p. 154). In addition, we will show
(Theorem 40) that there are no paradoxical dissections in 52^ This
suggests considering what the situation is in {J2. We will not
anticipate any other results of the following sections but just
stress that problem (10), p. 4), the squaring problem in the geometry
of sets, is still unsolved. In order to solve this range of problems
one must undertake a somewhat complicated way of reasoning in
the next section.
2. H ausdorff ’s T heorem
One is able to tackle paradoxical dissections in three-dimensional
space by means of Hausdorlf’s theorem:
Theorem 38: The surface area of the unit sphere °lt of 523 may be
represented as the union of four mutually disjoint sets:
m = Q + m + S f+ F . (5)
Of these sets TL is denumerable and the others have the same
cardinal number as the continuum. These sets are related by the
congruences
I = + (5')
In this case the congruence motions are rotations of the sphere.
We divide the proof of this theorem into several parts.
A. Definitions of transformations
We place a rectangular coordinate system at the centre of the
sphere and call the positive w-axis a ray which starts from the origin
of the xz-plane and which makes an angle ^ 0 with the positive
z-axis and an angle ^ —J 0 with the positive x-axis. We will still
be able to choose the value of 0.
Now let
(X'\ /x\ lx \ ix\

P' = I y' I = <t>(P) = <l>\y and P ' = 1 / =«/<P) = <A y
be the rotations through 180° about the w-axis and through 120°
about the z-axis respectively. These rotations <j) and ip can be
represented as follows:
x' = —x cos 0 + z sin 0 ,
/ = -y , (6)
z' = x sin 0 + z cos 0 ,
and
x' = - \ x ~ \ y f i y ,
y' = \4 $ x -% y , (6')
Z -- ~y
respectively. Provided no misunderstandings are caused, we

will indicate simply by juxtaposition of the factors (j> and </< the
composition of motions (f>(P) and ip(P) such as
0)
Therefore instead of (7) we would write (£</». We agree that in
products like this the operation at the right must always be per
formed first.
Now it is true that
<t>2 = 1, <A3 = 1, (8)
and hence one can reduce operations like
</>50 4</>2i/r;
to
Now we assert that one can show, by suitable choice of the angle 0 ,
that differently written reduced transformations are also geometrically
different.
According to our convention reduced transformations are
represented in one of the following forms:—
a= p=
where the exponents mvare the numbers 1 or 2. Now if, for example,
- </»<£,
it would follow that
—ifj(j)2tfj2 = if)2 = 1.
Obviously it is sufficient for the proof of this assertion to verify that
none of the products just denoted by a, /?, y or 8 can be equal to 1.
We show first of all that, for suitable choice of 0 , a rotation of type
y. can never be equal to 1. These rotations are composed of rotations
of the forms ijj<f> and ip2<f>. Let us calculate first t(j<j> from (6) and
(6'). By combining the two transformations one obtains (most
simply by multiplying out the matrices) the representation
.v" = \ cos 0 . x + $yj3 ■y —\ sin © • z>
>•" = —\sj3 cos 0 . x + iy + Ly/3 sin 0 . z, (10)
z" — sin 0 . x +cos 0 . z
for the transformation
Since «/»2 indicates only the rotation through 240° or —120° about
the z-axis, one obtains the representation of this rotation from (10)
on replacing ^ 3 always by —J 3 .
The image of the point M 0 = (0, 0, 1) under the operation ip(j) is,
from (10), M v = ( - j sin 0 , ^ 3 sin 0 , cos 0). If one applies to
A/j the motion one obtains
M 2 = (sin 0 . / t(cos 0), sin 0 . / 2(cos 0 ), / 3(cos 0)),
where f v(v) is a polynomial in v. It is not necessary for our purposes
to evaluate these functions exactly. Then the repeated application
of these motions leads to
Mn = (sin 0 ./i(n)(cos 0 ), sin 0 ./ ) n)(cos 0 ) ,/3 n)(cos 0)).
In any case the third component zn of the point M n will be a poly

nomial of the type
z„ = a0 cos" 0 + fli cos"-1 0 + ... + an.

The coefficients are of the form a + b j 3, where a and b are rational
numbers. Now if a = 1, M n = M0 would have to be true and
therefore zn would also be equal to 1. Hence we have for £ = cos 0
A 0C + A 1? ' - 1+ ...+ A n = %/3(B0? + ...+ B n)

with rational numbers A v, Bv. By squaring one obtains from this
an equation which is satisfied only by algebraic numbers £.
If one chooses for example £ = cos 0 = n/4, this equation cannot
be satisfied and therefore a is not equal to 1.
It only remains to show that no product in (9) of the form /?, y
or 5 can be equal to 1. Therefore let us suppose that /? = 1. Then
^^(j) = (j)2 = 1 also and therefore
<£[<£</<nT .. <//""]<£ = 1.
However, this is a transformation of type a and we know from above
that it cannot be equal to 1. Let us now consider the case y = 1.
According to this we would also have <f>y<j> = (f>2 = 1. For the
special case y — <jxpmt<f>this means that </»'"> = 1 and this is certainly
false for m t = 1 or m 1 = 2. In the general case = 1 leads to
an equation of type <5 = 1. Therefore we will have completed our
proof if we have shown that, for the choice made for 0 , <5 = 1
also leads to a contradiction.
To verify this we divide the elements of <5 into two classes.
Transformations of type 5, for which the sum of the exponents
m, + m n # 3, belong to the class C\ and the others to the class C2.
Now let <5 = 1 where <5 belongs to Cx. Then
3= 1= = (j>i
However, for m i +mn ^ 3 this is a transformation of the type (1.
It cannot be equal to 1. Finally there still remains the case
S = 1 and <5e C2.
Here
= <f>2 = \
or
Because m y+mn = 3 and f 2 = f 2 = 1 we get from this
s'= , = i.
This is again a transformation of the type <5 but it has two terms
fewer than <5. 5' cannot, as already proved, belong to the class Ct .
Therefore the method just used can be continued to reduce factors
until finally f = 1 or f m* = 1 (niv = 1 or niv = 2) remains left
over and this is certainly false. Hence it is shown that all the
different reduced transformations for the particular choice of 0
are really different geometrically.
The transformations constructed by combining the factors f and
tp form a group (S . We continue the proof of Theorem 40 by proving
a lemma about this group (S .
B. Proof of a lemma
Lemma B: The elements of the group (S can be divided into three
disjoint classes A, B and C so that the following conditions are
satisfied:
1. I f a e A, f a £ B + C and conversely.
2. Each of the conditions a e A, f a e B, f 2o e C implies the
other two.
In order to be able to carry out the above division of the elements
of ^ into three classes we first of all perform another division of the
group (S . Let Gn be the set of elements of # whose reduced repre
sentation contains no more than n factors f , f , f 2 ( f 2 is considered
as one factor). Then obviously
Gt = ( i , f, <A, <A2).
We now divide this subset into classes A, B and C:
1 e A, (j) e B, f e B, f 2 e C.
The statement of lemma B is true for this division. In fact, if
o e A (and o e Gf ) , a = 1. Then, however, f o — <j>e BczB+C.
Conversely, if f o e B+ C , (po = (j) and therefore o = 1 e A.
If o, fa , f 2a are all from then only three values are possible
for a: a = 1, a = f , a = f 2. Now let one of the relations
a e A, f a e B, f 2a e C (11)
be correct. If a e A, a = 1 and the second and third relations are
satisfied. If f a s B, the first of the values possible for a, namely
a = 1, is taken. Finally, if f 2a e C, it follows again that a = 1.
The further division of ^ into the three classes A, B and C can

be carried out now by induction. Let us assume that Gn is divided
according to the lemma. Since Gn<=.Gn+[ it is sufficient to give the
classification for Gn+, —Gn. These are just the elements of ^ which
have exactly n+ 1 factors.
Now, for an element z e Gn+1 —G„, there are three possibilities:
t must be one of the three types
T = 4><J, Z = tfjo, Z = lfj2(T, (1 2 )
where a is an element of G„ with exactly n factors. Now we give a

table which, for each of the cases named in (12), allows one to read
off the class to which z belongs if that of o is known:
Class of r
T = <f><7 T — \p (J T -
A B B C
B A C A
C A A B
______________________________ (13)
With this the division of ^ into three classes A, B and C by induction
is completed. It will be shown that it has the property asserted in
Lemma B. Before doing this let us note as an example how,
starting from the division of G{, the division of (73 is obtained from
table (13):
Gy Gi—Gy g }- g 2
1
A ! 1 # , # 2, 'l'24>
B 4>> </> M'24>> '/"Ma i/ ' # 2
C i Ip2<fnf)2
PROBLLMS IN THt GEOMETRY OF SETS 147
Now we assume that all the statements of lemma B are correct for
G„ and show that they are also correct for Gn+,.
First we deal with statement 1 of the lemma. Let a and (f)o be
elements from Gn + Then the following four cases are to be taken
into consideration:
(a) o s G„, (j)o e G„\ (b) a e G„, <po£Gn;
(c) o $ G n, (j>o e G„; (d ) o $ G n, <j)o$G„.
In case (a) statement 1 is correct by the induction hypothesis. In
the case (b) a consists of exactly n factors of which the first is ip
or ifj2. If the first factor were (j) then, since <j)2 = 1, <j)o would have
fewer factors than a and so would have to belong to G„. If a e A,
(13) shows that (j)o e B e B + C. Conversely, if <j>oeB+C, (f>o e B
(from the column t = <po of (13)) and hence a s A.
In the case (c) the number of factors for <f>o = p is less than for
a. This is only possible if the following is true:
O = ( j) p , p = (j)<7, p = if j " ' . . . .
Now table (13) must always be read so that the element with the
smaller number of factors is inserted in the column under a. There
fore in our case a = <j>p is to be used and in (13) t is to be replaced
by a and a by p. If one observes this one reads from the table that
p = (f>o e B + C follows from a = <j)p e A and conversely.
Finally the case (d) is not possible, a would have to consist of
n+ 1 factors and ^ cannot be the first. (Otherwise we would have,
of course, (fxj e G„). Therefore a is of the form a = < But
then <j)o would have n + 2 factors and would not belong to (7,I+ 1.
Hence statement 1 of lemma B is proved.
For the proof of statement 2, suppose that a, i}jo, i^2o are elements
of Gn+ 1. If all three already belong to Gn then our statement is
correct by the induction hypothesis. It is impossible to have none
of the elements belonging to G„. For then a would be of the form
a — (j)..., and </»<r and ift2a would have n + 2 factors. Therefore there
remains only the case when some of the three elements a, t//a. i//2rr
belong to G„ and others do not.
First of all we suppose o s Gn. Then a must be of the form
a = (j).... Otherwise 0(T and </i2cr would also belong to G„. Then it
follows from (13) and a e A that
</*<t e B, i(j2o e C
and conversely.
148 U N S O L V E D A N D U N S O L V A B L E P R OB L E MS IN GE OME T R Y
Next suppose a $ Gn, f o e G„. Here we again have the case that
multiplying by </» decreases the number of factors. Then a is of the
form ijj2p, p = $■■■, p = </'cr. Using table (13) one can observe
again that p = ipo has fewer factors than a. Then from
a = if)2p e A it follows from (13) that p = ifjo e B, ip2o = f p e C .
Conversely, from f o e B or if>2a e C it follows that ifj2 . f o = cr e A.
There only remains the case: a $ Gn, f 2o = p e Gn. Here cr is
of the form cr = iftp and for a e A it follows that ifj2o = p e C and
f a = f 2p e B. From each of these last relations it follows again
that cr e A. Hence lemma B is completely proved.
C. Application of the axiom of choice

In the last part of our proof we shall use the axiom of choice of
set theory (see for example [XI 6], p. 53). It can be formulated as
follows:
Let J t be a set of non-empty, mutually disjoint sets J l T h e n
there is a set -m with the following property.—Each element m
of m is an element of a certain set J l x and m has exactly one element
in common with each o f the sets J t a.
This axiom is a statement about existence only and in practice it
is not always possible to determine the elements of this “ set of
choice ” . t Thus we shall have to satisfy ourselves in this case too
that the existence of the set mentioned in Theorem 38 is proved by
the axiom of choice; but no method of writing the division of the
spherical surface into 4 subsets exists in practice.
First we denote by J2 the denumerable set mentioned in Theorem
38. This is the set of points of the spherical surface which remains
fixed under one of the rotations of the group (S . Of course, a pair
of diametrically opposite points remains fixed under each spherical
rotation, and so we have a denumerable set 2, of such fixed points.
If M is a point of 21—2 then ^ <r2(M) for all rotations
<ti , cr2 (cr! 7 ^ <t2) of our group (S . For, if ux(M) = ct2( A it would
follow that <rj'1cr2(M) = 1. But <7j"V2 is also an element of ^ and
M would be a fixed point for a f 1 cr2. Since M does not belong to
2, this is impossible.
Now we denote by for M e t f / —2, the set of all points
cr(Af) which arise when o runs through the whole group From
f On the difficulty of the “ actual ” evaluation see for example [VI 10], pp. 44
el seq.
this definition it follows immediately that, if M e °ll —2, N e °lt —2,
M T N, the sets JT(M) and 0T{N) are either disjoint or identical.
Of course, if cr1(M) = it follows that M = o j 1o2(N) = i (N).
This means that M belongs to the setjT(N ), and, since a(M) = ax(N),
jT ( M ) and are identical. Therefore, in this way, Ql—2 is
divided into a collection (of cardinality equal to that of the con
tinuum) of classes of denumerable sets 0T(M). By the axiom of
choice there exists a set ST of “ representatives ”, i.e. a set which
contains exactly one element from each of the sets 0T(M).
Now, using this set of choice ST, we can define the sets Si, ST and
ST of Theorem 38. First we agree that er($T) means the set of points
which is obtained by the rotation o from SC. With this notation
we define
m = £ o{SC), ST = £ < 7 ( f ) , 3 T = Y J o{3C\ (14)
a £A ae B <reC
We now assert that

1. The sets 01, ST and ST are disjoint.
2. Each point of 21—2 belongs to one of the three sets.
We show first that to assume that 01 and ST have a common
point leads to a contradiction. If P were a point in the intersection
0ST, we would have
P = c f M ) = o2(N),
where
oêA, o2 eB, Me3S, NeSS
and M # N in any case, since otherwise = M would be
true and M would have to be in 2. And also from o x(M) = cr2(/V)
for M ^ N we get a contradiction. For N = «rj 1a 1(M) would hold
and therefore N = t(Af). But this is impossible since M and N as
elements of ST are representatives of different classes. Therefore the
intersection of Si and ST must be empty. Similarly one shows that
m sr = o, ST.sr = o.
Now we prove assertion 2. Let P be any point of °U —2. Then
t belongs in any case to 0T(P). Now, if M is the representative of
the class J>T(P) lying in ST, P e J T (M ) also, i.e. P = <
j (M) for a
certain o from C S . However, a must belong to one of the three
classes A, B or C and hence from definition (14) P belongs to ^2, ST
or 0 . Further for the sets 01, 0 and 0 defined by (14)

(j> {0)= 0 + 0 , ifi(O )= 0 , = 0- (15)
are now valid.
First of all let us prove the first of these relations. Let P e 0 so
p = g(M) with M e 0 and a e A. Then
= <j>a(M )e0+0,
since <f)a belongs to B + C from lemma B. Now, conversely, let
Q E 0 + 0 , therefore
Q = p(M) with M e i , p e B + C.
Therefore <})~1Q = (j)~1p(M). If we put (f>~lp = a or p = <fxr,
by lemma B, since 4>o e B + C , it follows that a e A and therefore
a(M) e 0 . And this means that <f> l(Q) e 01, Q e <f>(0t). Hence, we
have in fact
<t>(0) = 0 + 0-.
Similarly one shows, using lemma B, that the other relations of (15)
are also correct. The details of these arguments are left to the
reader.
Let us summarise:—We have discovered a division of the spherical
surface in which the sets 0, 0 1 ,0 and 0 are mutually disjoint. The
congruences stated in Theorem 38 are obtained from (15).
3. P a rado xical D ecompositions in
From Hausdorff’s theorem we obtain the paradoxical decomposition

of two sets which lie on the sphere and have the same cardinality as
the continuum: this is, from Theorem 38,
0 + 0- = 0, 0 + 0 = 0, 0 . 0 = 0.
By the definition on p. 141 this decomposition is paradoxical.
However, paradoxical decompositions of other simpler sets may
also be derived from Theorem 38. To derive these we require two
more lemmas.
Lemma C : Let 0 be any point set which is at most denumerable, on
the sphere 0 f . Then
x = .- r - 0 (16)
is true.
The proof of (16) is carried out similarly to that of (4). Here a
is replaced by 0 and <r(a) by 0(<x). One only needs to choose the
angle for the rotations about the z-axis, say, so that the sets
J2, -2(a), J2(2a),...
are all disjoint. One can easily see that this is always possible.
Let us denote by rjv the plane through the z-axis and any point Pv e 2L,
and by av the angle which the plane t]y makes with the xz-plane.
Then a may be chosen so that the relation
av+ «a = a / + 271/1 (17)
is not satisfied for any quadruple (//, v, v', f) of non-negative integral
numbers. Since the set of a satisfying (17) is denumerable, one can
certainly choose the angle of rotation so that all the sets H{va) are
disjoint.
Hence one can prove lemma C in the same way as equation (4)
above.
Lemma D: I f
si^>6=>2ft and s i = 2ft, (18)
then
s i — S.
By hypothesis the sets s i and 2ft can be decomposed as follows:—
s i = s i ^ 4- s i 2 T ••• T -s^n, 3ft = 3 8 3ft2~\~••• + 3ftn, (19)

where
s i ts i k = 2ft{2ftk = 0 for 1 ^ i < k ^ n (20)
and
s i k = 3ftk for k = 1,2, ..., n. (21)
Thus there is a congruence transformation ^ ( P ) for each k = l s 2,
3 ..., n which transforms the set s i k into 3ftk. Let us combine these
tran sform ations b y defining
0 (P ) = 0 * (P )fo rP e .s /t .
From (18) we have $>(si) = 28 c z g c s i and therefore
<S>'{si)^si, <&l{ s i-£)<=: s i ( / a positive integer).
jsjoW we define a new set s i' by the rule
s i ' = {si-<g) + $ > (si~ £ )+ ® \* t
(23)
and si" = s i - s i ' , £ ' = <$>{si').
(24)
152 U N S O L V E D A N D U N S O L V A B L E P R OB L E MS I N GEOMETRY
Then from (22) we have that s i ' ^ s i and

s i = si' + si", s i's i" = 0. (25)
Further
S' = <S>(sT)c:<S>(st) = @<^S, (26)
therefore
S'cS. (26')
In addition, from (23), it follows that
® (si') = <J>(^ - S) +<S>\si- S) + & (si -S ) + ... (27)

therefore
si' = (si-S)+<S>{si') = (s i-S ) + S' - s i - { 8 -S').
Hence, from (24),
S -S ' = s i - s i ' = si"
or
S = S' + si", S'si" = 0. (28)
After these preliminaries we can decompose si into (h+1)
mutually disjoint sets:
si = si' + si"= si 'si + s i '—si' si j +sf's/2+ ...+s/'s/H+ si". (29)
Now for the terms si' sik of the sum (29)
si'sik = S'^k, f c = l, 2, ...,n (30)
is true. This can be seen as follows: From the definition of the
congruence transformation
<&
k{si'sik) = <&(si'sik) = <S>(si'). U>K) = S '£ k.
Now we can write for S a decomposition similar to that for s i.
From (26) S' = S'iS and therefore, from (28) and (19), it follows
that
S = S'+si" = S'@ + si" = S ' 8 1+ S '® 2 + ... + S '8 n+ si". (31)
From (28) and (20) the terms of the sum on the right hand side are
disjoint. Therefore, from (29), (31) and (30),
si — S.
We can use the lemma just proved to derive other paradoxes.
From Theorem 38 it is possible to obtain a decomposition (5) of
the spherical surface. The sets denoted there by ffl, 9 and ST we
can now divide as follows: We have that 0 = 90 0 0 . Thus there
exists a decomposition 0 = 0 x + 0 2 for which
0tx . 0 2 = o, 0 X = 0 , 0t2 = 0 . (32)
On the other hand 0 = 90, 0, = 0~. 0 Xand 0 2 are transformed by
the appropriate rotations into the disjoint s e t s ^ a n d ^ 2 o r 0 x and
0 2 respectively:
0 = 0 x+ 0 2, 0 = 0 x+ 0 2, 0 x0 2 = 0 x0 2 = o, « t = y „ E y v,
v = 1, 2.
Hence we have from (5) the following decomposition of the
spherical surface :09:
09 = 0 X+ 0 20 0 ’\ 0 0 200~ 10 0 2 + 9L.
Now we put
09 x = jjf2 = 0 20S920 0 20 0 .
But from (5') and (32) the congruences
0X: 02 =0 =0 - 0,
0 X= 0 2 = =0,
0X- 02- 0 X~02=0
hold. Hence it follows that
09x = 0 9 - 0 092 = 09.

2
Now from lemma C 09 —9L — 09 and therefore from lemma A
6
jT , = J f . Finally we have the result that a spherical surface is
finitely equivalent to two spherical surfaces (of equal radius) :
09 —0 9 i + 0 9 2 , 0 9 x0 9 2 = 0, 09 1= 09 , 09 2 = 09 . (33)
Now it is no longer difficult to prove Theorem 37 formulated on
p. 140. In this theorem we are concerned with solid spheres, not
spherical surfaces. First of all let us denote by A' the solid sphere
without its centre and note that the outer set of boundary points is
the spherical surface just considered. If we replace the points
of the spherical surface by the radial lines (open at the centre M
of the sphere) through them, we obtain from (33) a corresponding
154 U N S O L V E D A N D U N S O L V A B L E P ROBLEMS I N GEOME T R Y
paradoxical decomposition for the solid sphere (without its centre):
/' = + / ' 2, = o, = £', t'2 = £'. (34)

Now, if we include the centre, we get for the solid sphere with its
centre:
£ — £[ + £2 + M, £ 1 = £ [ + M = £, £'2 = £ - M .
Hence, £ ’2 = £ —P, where P is a point of the spherical surface.

Finally from lemma C, it follows for the solid sphere that
£ = £ t + £'2, £ x =£ , £'2 = £,
i.e. Theorem 37 is proved.
Hence we discover
Theorem 39: I f J t x and J l 2 are two bounded sets of three-dimen
sional space which contain a solid sphere £, they are finitely
equivalent.
To prove this we show first of all that J l x is finitely equivalent
to / . Let r be the radius of the sphere £ and {IP} a cubical lattice
whose cubes have edges a = 3 . r. Then each cube is inscribed
in a sphere with radius equal to r. Let ^ cubes IP a {a = 1, 2, ..., s)
cover the set J l t completely. We have therefore
J l xc:iP x +IP2 + ... + iP s.
Now J l x is decomposed further into subsets ma which are contained

in'#""*:
J l i = m x + m 2+ ... + m s, m ac:IPa.
Now, from Theorem 37, £ = £ 1+ £ 2, where and £2 are two

separate spheres of radius r. It follows further that £2 = £3+ £x
therefore £ ^ £ 1+ £3-h£4 and by mathematical induction one
shows that there are s separate solid spheres of radius r for which
4 = £ 1 + £ 2 + ...+ £ s (35)
is true.
From the definition of the cubes IP a and the spheres £„ it now
follows immediately that J l x is finitely equivalent to a subset of
£\ +£2 + ••• + ^s* Then from (35) J l x is also finitely equivalent to
P ROBLEMS IN THE GEOMETRY OE SETS 155
a subset of / . Therefore we have
W '= J l x.
Thus from lemma D it follows that J i x = / . Hence we obtain

Theorem 39 immediately. For, if a sphere of radius r is contained
in J t j as well as in J ( 2, from our results we have that
J ( 2 = /,
and therefore, from lemma A, J ( x = J t 2.

From Theorem 39 it follows, moreover, that every polyhedron is
finitely equivalent to every sphere.
4. P a r a d o x i c a l D e c o m p o sitio n s in t h e P la n e
The results of the last section raise the question whether there exist
paradoxical decompositions in the plane. There are in fact such
decompositions.
We give a simple example:—
Let £ be a set of complex numbers of the form
C = a„e'n+ a„-1e'(n~1)-i- ... + a 1e' + a0. (36)
where n and av (v = 0, 1, 2, ..., n) are non-negative integers. Let
be the subset of £ for which a0 = 0 and = £ —stf the comple
mentary set. It is easily seen that the numbers of the set £ are just
those obtained from the number 0 by repeated application of the
operations
P(z) = 2 + 1, D(z) = el . 2.
Now, if one applies the transformation D(z), to the set £, one obtains
s4. If the numbers (36) are multiplied by e \ one again obtains a
number of the form (36) but with a power of e' appearing as a
factor in each term of the sum. Conversely, multiplying a number
C = anein+ an_ 1ei^ - i'>+... + a 1ei
o fstf by e~l leads to a number of £ ; so the congruence of sets s t and
£ for multiplication by e' means rotating the set £ through the angle
a = 1. Likewise P(z) = z+1 maps £ onto /?. Since this is a
congruence transformation too, we have
£ = j£ + a , s / . rn = 0, £ = J*, £ = &,
and therefore a paradoxical decomposition.
Now the set S is denumerable and unbounded in contrast to the

sets decomposed in 38^. It is natural to ask whether there are also
paradoxical decompositions of bounded plane sets which have the
same cardinality as the continuum. Using our results in ^?3 one
could try to find a paradoxical decomposition of a circle in a similar
way to that of the sphere. But it turns out that this is impossible.
In fact
Theorem 40: There are no paradoxical decompositions of the unit
circle or of its non-empty subsets.
is true.
To prove this let us suppose that S is a set of points of the complex
plane with | z | = 1 for which there exist decompositions of the
following kind:
£ = sd + 38, jd . 38 = 0\
S — S\-\-S2^~---~i~Sic = S ^ - \ - S ..~ h S 'i ', (37)
sd — sd i + sd 2+ ••. + sd j., 38 = 38j + 38, + ... + 38l.
sd K s i = 3 8 ,^ ,^ = $ i&j = S'i&'j = 0,
where
= 38,. (38)
The congruence transformations in (38) between S K and sdK or
S', and 38, are rotations or reflections. They can be written as
follows:
4>k{z ) = <t>K(ei9) = eiXK . z = e!^ +V or <}>K(z) = £ ri(a“+9) (39)
f,{ z ) = f,(e '3) = e'^k . z = e,(-Px+S) or 1p,(z) = e~‘(P*+!t\
Since S, sd and 38 are decomposed into mutually disjoint subsets
by (37), one can also combine the congruence transformations (39)
as follows :
K(z) for z e £ k, '¥(3) = f,{ z) for ze S ',.
We can assume, without loss of generality, that z = 1 belongs to
the set S. Then we have that 0(0) e j / , and therefore
0 ( 0 ) 'P(0), since the intersection of sd33 is empty. Since 0(zc) # ^(jc)
and 0(x) ^ vF(j>) for all x e S and y e S , we have also
00(0) =1=^0 (0 ), 00(0) 4='PF(0), 00(0) * OT(O),
O'F(O) =HT'O(O), O Y (0 )# W (0 ), 'P Ô ) * ¥ ^(0 ).
Therefore the four complex numbers <M>(0), <hvF(0), VFO(0), vFT(0)

are different from one another. Now we consider numbers of the
type
TS = 0 S0 S- ,..0 ,(O ), (40)
where 0 CT (a = 1,2, ...,5) stands for O or VF. Now one shows
easily by mathematical induction that any two numbers xs are
different from one another if different transformations (d> or VF)
are substituted for 0 ff for any a. Since there are two possible
substitutions for each <r, we have altogether 2s different numbers of
the form (40). From (39) we can also write these numbers as follows:
t , = eiU = ei(Cl +C2+- +c*\ (41)
where ca stands for any ±y.Kor + /?,. However, this means that the
r = 2k + 21 numbers
±<*1, ±^2> ± ak, ±Pi, •••»
come into the question for ca in (41). Let us denote these numbers
by up (p = 1 ,2 ,..., /■). Then Zs in (41) is a sum of 5 numbers up:
S = uvV1+ t/v
' 2+ ... + »
The number up may appear np times as a term of this sum. Obviously
np can take only the values 0, 1,2, ..., s. Therefore there are 5 + 1
possible values for np. Since we have a total of r numbers up at our
disposal, there must be less than (5+ l)r different sums I s. Therefore
there are less than (j+ l)r different numbers t s. But we proved
before that there exist 2s different numbers of this kind. Therefore
we must have 2s < ( t+ l)r. But this is wrong for large 5 (with r fixed)
since it is well-known that
lim (s + D r = 0.
.V-*00 2s
Hence the assumption that there exists a paradoxical decomposition

of a (non-empty) set 6° lying in | z | = 1 leads to a contradiction.
Similarly one can show ([XI 3], p. 56) that
There are no paradoxical decompositions of a non-empty linear set.
Hence it is shown that there is no theorem for the circle analogous
to Theorem 38. It can also be shown | that paradoxical decomposi
tions are impossible for all bounded plane sets.
t L in d e n b a u m , Fund. M ath., 8, p. 218.
158 U N S O L V E D A N D U N S O L V A B L E PROBLEMS IN GE OME T R Y
However, this still leaves completely open the possibility that two
plane sets are finitely equivalent even though they are not equivalent
by dissection in elementary geometry. Therefore it makes sense
to pose the squaring problem in the geometry of sets (problem
(10), p. 4). It is still unsolved. A series of other open problems in
the geometry of sets has been formulated by Sierpinski [XI 2],
[XI 3]).
Chapter XII
THE METHODOLOGICAL SIGNIFICANCE

OF “ UNSOLVABLE” PROBLEMS
W olfgang B olyai (1775-1856), a friend of C. F. G auss had spent

many years of his life trying to prove Euclid’s parallel postulate.)'
The existence of an unsolved problem like this in geometry was a
serious annoyance to him. In this connection he wrote to his son
Jo h a n n :
It is inconceivable that this pressing darkness, this eternal solar
eclipse, this defect, this continual flaw in pure truth should be
allowed in geometry.
Nowadays, if this is quoted in a lecture on the foundations of
geometry, the audience just smiles calmly. How is it that the
Hungarian mathematician’s emotion, certainly meant seriously,
appears so funny to modern students ? Herein lies a change in the
way of thinking which it is important to investigate.
For the mathematicians of past centuries, trained in Platonic
philosophy, the theorems of geometry were statements about the
world of ideas and the presence of an important and still unsolved
problem meant that the scholar did not have a completely clear view
in the world of absolutely valid truths. But the work of B olyai ’s
son (at approximately the same time as the Russian L obachevsky )
showed that the parallel problem cannot be solved because it is
possible to have a geometry free of contradictions in which all the
t The 5th postulate (“ parallel postulate ”) of E u c l i d is the following:
“ I f a straight line intersects two straight lines and the sum o f the interior
angles which it makes with them on the same side is smaller than two right
angles, the two lines intersect when they are produced on this side."
Translator’s note: This is equivalent to the usual form, viz.: There is a
unique straight line through a given point parallel to a given straight line in a
plane.
159
Euclidean axioms and postulates are valid up to the parallel
postulate: this is replaced by the statement that there exist at least
two parallels to a line through a point in a plane. Moreover, the
existence of “ non-Euclidean geometry like this meant that the
mathematicians in the 19th century learned to regard the nature of
their axioms differently. In Hilbert’s formalistic mathematics it is
no longer a matter of “ truth ” (in the metaphysical sense), but of
“ certainty ” ,f—the certainty that the construction of the theory
from the chosen axioms can never lead to a contradiction. The
fact that the parallel postulate cannot be proved has thus lost
its terror: Euclid’s 5th postulate is simply independent of the others
and the other axioms cannot by any means be adapted to
prove it.
P lato called mathematics the “ awakener of knowledge ” . Now
it appears that this is the case for modern mathematics in a different
sense:—In mathematics it is possible to determine the possibilities
and boundaries of a scientific method with a precision which is not
possible for example in the domain of historical scholarship. This is
particularly significant if one draws on the modern investigations
on the possibility of proving the consistency of mathematical
theories. There are possible subjects here for the teaching of
mathematics in colleges and schools which have not yet been fully
used.§
The unsolvable construction problems ((8) and for certain radii
(9)) considered in this book have a methodological significance
similar to that of the parallel problem. For thousands of years it
was believed that squaring the circle and trisecting the angle by ruler
and compasses must be possible. Today we know that ruler and
compasses are just not suitable instruments for solving these pro
blems. A modern geometer || compares the existence of the
unsolvable problems in geometry with the fact that a wood-saw is
unsuitable for shaving. The contemporary researcher is in fact
more prosaic than W olfgang B olyai but it is also clear that there
is good reason for this change of style.
t See e.g. [XII 1],
t H ilbert in his lecture “ Uber das Unendliche ” (“ On the infinite ”) which
he gave in Munich, 1925. Printed in the “ Grundlagen der Geometrie ” (Founda
tions of Geometry ”) 7th edition, 1930.
§ On this topic see the treatise [XII 3] which is also reprinted in the 2nd
edition of [VI 10],
|| W . B l a s c h k e in [X II 2 ], p . 7.
METHODOLOGICAL SIGNIFICANCE 161
For the “ unsolvable ” problems of Chapter VIII the position is.
somewhat different from that for the construction problems. Here
an unsolvable problem exists only because the problems were
formulated awkwardly:—In the formulation of problem (7) it is
implied that there exists in the set of domains 2 one with smallest
surface area. However, the set of surface areas of domains which
is bounded from below, does not necessarily have a minimum. But
a number set of this kind always has a lower bound. Let us replace
problem (7) by the following:
(7c) In a domain S> a line AB of length 1 may be moved (by translation

and rotation) so that the direction of AB is turned through 360°.
It is required to find the lower limit 1 of the surface areas of alt
the domains 3>.
This problem is solvable. From Chapter VIII we have / = 0 and

we can add that, in this case, the lower limit is not a minimum.
From the unsolvability of problem (7) one sees that one may
never assume the existence of a solution as obvious for extremal
problems if it concerns sets containing an infinite number o f
elements. And this also means that indirect proofs for extremal
problems need to be completed by an existence statement. We
illustrate this using the “ isoperimetric problem ” as an example.
Here we are concerned with the following classical problem:
(7d) Let M be the set of plane, simple, closed, continuous curves o f
length L. Which of the curves of the set M encloses a domain
of greatest surface area ?
There is a beautiful and simple proof by Jacob S teiner [V 4J

which says that a non-circular curve does not have this extremal
property. Each such curve (€ can be “ improved ” by giving a
curve c€' of equal length L, which includes a domain Q>' whose
surface area A(3>') is greater than that of the domain 3) enclosed by
the curve (€. Hence S teiner deduced that the problem (7c) is
solved by the circle of radius —. W eierstrass was the first to show

2n
that the conclusion is not correct. Of course it is not stated that
the Steiner extremal problem generally has a solution. First of all
it has only been shown that, if there exists a solution of this problem,
it is the circle. Of course, one can complete Steiner’s argument in,
this case by giving an “ existence proof ” or show directly in another
way that the circle is the solution of the isoperimetric problem.f
We merely emphasise by an example given first by O. P erron ,
that Steiner’s method of reasoning is inadmissible.
We can “ prove ” that 1 is the greatest of all natural numbers by
the method used for the isoperimetric problem. This is done as
follows: By squaring, one associates with each natural number n
different from 1 a greater natural number ri2. Each natural number
different from 1 can therefore (by giving a greater number n2) be
“ improved ” just as above in the geometric proof the closed curves
were “ improved ” by giving another with greater area. This is
true for all the natural numbers except 1. Therefore 1 is the greatest
natural number.
Here the fallacy is obvious: we have implied the existence of a
.greatest natural number, and this is certainly wrong.
f See in this connection, e.g. [II 7].
B IB L IO G R A P H Y
The bibliography is in the order of the chapters of the book. Those works
which contain full particulars about the literature on the subject are indicated
by a star at the number.
II 1. W. K il l in g a n d H . H o v e st a d t . Handbuch dcs matliematischen Unter-
ric/itsl. Leipzig a n d Berlin, 1910.
II 2. L. F fjes T o t h . Abschatzung des kurzesten Abstandes zweier Punkte
eines auf der Kugel liegenden Punktsystems, Jber. dt. Mat Verein, 53,
pp. 66-68, 1943.
II 3. D. H il b e r t and S. C o h n -V o ss e n . Anschauliche Geometrie. New
York, 1944.
IJ 4. W. L ie t z m a n n . Elementare Kitge /geometrie mil nurnerischen und
konstruktiven Methoden. Gottingen, 1949.
II 5. W. H a b ic h t and B. v a n d e r W a e r d e n . Lagerungen v o n Punkten auf
der Kugel., Math. Annin., 123, pp. 223-234, 1951.
II 6. B. v a n d e r W a e r d e n . Punkte auf der Kugel. Drei Zusatze., Math.
Annin., 125, pp. 213-222, 1952.
II 7.*L. F fjes T oth . Lagerungen in der Ebene, auf der Kugel und im Rauitt.
Berlin, Gottingen, Heidelberg, 1953.
II 8. L. F ejes T o t h . Kreisiiberdeckungen in der hyperbolischen Ebene.,
Acta Math Ac. Sc. Hung., IV, 1-2, 1953.
II 9. L. F ejes T o t h . Kreisausfiillungen in der hyperbolischen Ebene., Acta
Math. Ac. Sc. Hung., IV, 1-2, 1953.
II 10. H. M e s c h k o w s k i . Elementare Behandlung von Lagerungsproblemen.
Math.-phys. SemesterB., 4, pp. 256-262, 1955.
II 11. A. H e p p e s . Uber mehrfache Kreislagerungen., E/em. Math. 10,
pp. 125-127, 1955.
Ill1. K. S c h u t t e and B. v a n d e r W a e r d e n . Auf welcher Kugel haben

5, 6, 7 Oder 8 Punkte im Minimalabstand 1 Platz?, Math. Annin., 123,
pp. 96-123, 1951.
III 2. K. S c h u t t e . Oberdeckung der Kugel mit 8 Kreisen., Math. Annin.,
129, pp. 101-106, 1955.
IV 1. F. S u p n i c k . On th e d e n s e p a c k in g o f sp h e r e s, Trans. Amer. math.

Soc., 65, pp. 14-26, 1949.
IV 2. M. E. W is e . Dense random packing of unequal spheres, Philips Res.
Rep., 7, pp. 321-343, 1952.
IV 3. A. H. Boerdijk. Some remarks concerning close-packing of equal
spheres, Philips Res. Rep., 7, pp. 303-313, 1952.
IV 4 . H . H a d w i g e r . Einlagerung kongruenter Kugeln in eine Kugel.,
Elem. Math., 7, pp. 97-103, 1952.
V 1. H. W. E. Jung. Ober die kleinste Kugel, die eine raumliche Figur

einschlieflt., (Crellcs) J. reine angew. Math., 123, pp. 241-257, 1901.
V 2. H. W. E. J u n g . Ober den kleinsten Kreis, der eine ebene Figur
einschlieBt., (Crelles) J. reine angew. Math., 137, pp. 310-313, 1910.
163
164 UNSOLVED AND U N S O L V A B L E P ROBLEMS IN GEOME T R Y
V 3. J. Pal . Oberein elementares Variationsproblem, Kgl. danske Vidensk.
Selskab Mat.-fys. Meddr., Ill, 2, 1920.
V 4. H. R a d e m a c h e r and O. T o e p l it z . Von Zahlen und Figuren. Berlin,
1933. (The Enjoyment o f Mathematics. Princeton, 1957.)
V 5. R. S p r a g u e . Ober ein elementares Variationsproblem, Mat. Tide-
skrift, 1936, pp. 96-99.
V 6. D. G a l e . On inscribing /7-dimensional sets in a regular //-simplex^
Proc. Am. Math. Soc., 4, pp. 222-225, 1953.
V 7. A. K ir s c h . Die Pferchkugel eines Punkthaufens, Math.-phys.
ScmesterBer, 3, pp. 214-218, 1953.
V 8. H . G. E g g l e s t o n . Covering the three-dimensional set with sets o f
smaller diameter, J. London math. Soc., 30, pp. 11-24, 1955.
V 9. H . H a d w ig e r and H . D e b r u n n e r . Ausgewahlte Einzelprobleme der
kombinatorischen Geometrie in derEbene., Enseign. math. 1, p p . 5 6 -8 9 ,
1955.
V 10.*J. M. Ja g l o m and W. G. B o l t j a n s k i . Konvexe Figuren. Berlin, 1956.
V 11. B. G r u n b a u m . A simple proof of Borsuk’s conjecture in three
dimensions, Proc. Camb. phil. Soc., 53, pp. 776-778, 1957.
VI 1. M . D e h n . Oberden Rauminhalt., Math. Annin., 55, pp. 465-478, 1901.

VI 2. M. Z a c h a r ia s . Elementargeometric der Ebene und des Raumes.
Berlin, 1930.
VI 3. J. P. S y d l e r . Sur la decomposition des polyedres, Comment, math.
helvet., 16, pp. 266-273, 1943.
VI 4. H . H a d w i g e r . Bemerkungen zur elementaren Inhaltslehre des-
Raumes, Elem. Math., 4, pp. 3-7, 1949.
VI 5. E. K a m k e . Mengenlelire. Berlin, 1949.
V I 6. H . H a d w i g e r . Zerlegungsgleichheit und additive Polyederfunk-
tionale, Comment, math, hely., 24, pp. 204-218, 1950.
V I 7. J. P. S y d l e r . S u r le s c o n d itio n s n e c e ssa ir e s p o u r l ’e q u iv a le n c e d e s
p o ly e d r e s e u c lid ie n s, Elem. Math., 7 , p p . 4 9 -5 3 , 1952.
V I 8. J. P. S y d l e r . Sur (’equivalence des polyedres a diedres rationnels,
Elem. Math., 8, pp. 75-79, 1953.
VI 9. H. H a d w i g e r . Der InhaltsbegrifT, seine Begriindung und Wandlung
in alterer und neuerer Zeit., Mitt, naturf. Ges. Bern., 11, pp. 13-41, 1954.
VI 10.*H. M e s c h k o w s k i . Wand/ungen des niathematischen Denkens, Braun
schweig, 1956.
VI 11. D. H il b e r t . Gnindlagen der Geometrie. 8. Aufl. Stuttgart, 1956.
VI 1 2 .* H . H a d w i g e r . Vorleswtgen iiber Inhalt, Oberflaclte und Isoperimetrie.
Berlin, Gottingen, Heidelberg, 1957.
VI 13. M. G o l d b e r g . Tetrahedra equivalent to cubes by dissection, Elem.
Math., 13, pp. 107-109, 1958.
VII 1. M. D e h n . Ober die Zerlegung von Rechtecken in Rechtecke, Math.

Annin., 57, pp. 314-332, 1903.
VII 2. M. K r a it c h ik . La Mathematique des Jeux. Brussels, 1930.
VII 3. A. Stohr. Ober Zerlegungen von Rechtecken in inkongruente
Quadrate, Schr. Math. Inst, und Inst, angew. Math. Univ. Berlin, 4,
pp. 119-140, 1939.
VII 4. R. S p r a g u e . Beispiel einer Zerlegung des Quadrates in lauter ver-
schiedene Quadrate, Math. Z., 45, pp. 607-608, 1939.
VII 5. R. S p r a g u e . Ober die Zerlegung von Rechtecken in lauter verschie-
dene Quadrate, J. reine angew. Math., 182, pp. 60-64, 1940.
BIBLIOGRAPHY 165
VII 6. R. S prague . Zur Abschatzung der Mindestzahl inkongruenter
Quadrate, die ein gegebenes Rechteck ausfiillen, Math. 7.., 4 6 , pp.
460-471, 1940.
MI 7. R. L. B r o o k s , C. A. B. S m it h , A. H. S t o n e a n d W. T. T u t t e . The
d isse c tio n o f re c ta n g le s in to sq u a r e s, Duke J. o f Math., 7, pp. 312-340.
1940.
VII 8. C. J. Bouwkamp . On the dissection of rectangles into squares, Kon.
Ned. Ak. van Wet., 49, pp. 1176-1188, 1946; 50, pp. 58-78, pp. 1296-
1299, 1947.
VII 9. W. T. T utte . Squaring the square, Canad. J. Math., pp. 197-209,
1950.
VII 10. M. G o l d b e r g . The squaring of developable surfaces, Scr. math.,
1 8 , pp. 17-24, 1952.
VIII 1. E. T r o s t . Bemerkungen zu einem Satz liber Mengen von Punkten mit

ganzzahliger Entfernung, Elern. Math., 6, pp. 59-60, 1951.
VIII 2. M. A l t w e g g . Ein Satz liber Mengen m it ganzzahliger Entfernung.
Elem. Math., 7, pp. 56-58, 1952.
VIII 3. A. M u l l e r . Auf einem Kreis liegende Punktmengen ganzzahliger
Entfernungen, Elem. Math., 8, pp. 37-38, 1953.
VIII 4 . F. S t e ig e r . Z u einer Frage liber Mengen von Punkten mit ganz
zahliger Entfernung, Elem. Math., 8, pp. 66-68, 1953.
VIII 5. H. FIa d w ig e r . Ungeloste Probleme, Elem. Math., 13, p. 85, 1958.
VIII 6. W. S ie r p in s k i . Sur les ensembles de points aux distances rationelles
situes sur un cercle, Elem. Math., 14, pp. 25-27, 1959.
IX 1. O. P erron. Algebra, 1, 2. Berlin, 1932.

IX 2. P. B u c h n e r . Eine Aufgabe, die mit Zirkel und Lineal nicht losbar ist,
Elem. Math., 2, pp. 14-16, 1947.
IX 3. E. R oth -D f.smeules. Noch eine Aufgabe, die mit Zirkel und Lineal
nicht losbar ist. Elem. Math., 3, pp. 65-67, 1948.
IX 4. B. v a n d e r W a e r d e n . Moderne Algebra. Berlin, Gottingen, Heidel
berg, 1950.
IX 5. H. L ebesgue. Legons sur les constructions geometriques. Paris, 1950.
IX 6. E. K a s n e r and J. H a r r is o n . The trisection of horn angles, Scr. math.,
17, pp. 231-235, 1951.
IX 7. L. Bieberbach . Theorie der geometrischen Konstruktionen. Basle,
1952.
X 1. N. M. N estorovic. On the squaring of the circle and the circularisa-

tion of the square in the Lobachevskian plane (Russian), Dokl. Akad.
Nauk. SSSR, 63, pp. 613-614, 1948.
X 2. C. L. S ie g e l . Transcendental Numbers. Princeton, 1949.
X 3. E. Beutel . Die Quadratur des Kreises. Leipzig, 1951.
X 4. T h . S chneider . Einfiihrung in die transzendenten Zahlen. Berlin,
Gottingen, Heidelberg, 1957.
XI 1.
W. S ie r p in s k i . Sur quelques problemes concernant la congruence des
ensembles de points, Elem. Math., 5. pp. 1-4, 1950.
XI 2. W. S ie r p in s k i . Algebre des Ensembles. Warsaw, 1951.
XI 3. W. S ie r p in s k i . On the congruence of sets and their equivalence by
finite decomposition, Lucknow Unio. Studies XX, Fac. of Science, The
Lucknow University, 1954.
166 UNSOLVED A ND U N S O L V A B L E PROBLEMS IN GEOME T R Y
X I 4.*J. P. N atanson. Theorie der Funktionen einer reellen Veranderliclien.

Berlin, 1954.
X I 5. H. H adwiger . Ungeloste Probleme Nr. 5, Elem. Math.. 10, p. 66,
1955.
X I 6. P. S. A lexandroff. Einfiihrung in die Mengenlehre and die Theorie
der reellen Funktionen. Berlin, 1956.
X II 1. H. M eschkowski. Nichteuklidische Geonietrie. Brunswick, 1954.

X II 2. W. B laschke . Reden und Reisen ei/tes Geometers. Berlin, 1957.
XII 3. H . M e s c h k o w s k i . M enschenbildung im Zeitalter der A utom a-
tisierung, Erziehung und Bildung, 11, Part 4, 1958.
IN D E X
Adjunction, 117 Electrical engineering, interpretation

in, 100
Equivalence of polyhedra, 82 et seq.
Equivalent by completion, 73
Basis polyhedron, 84 et seq. Equivalent by dissection, 2
Besicovitch, 107 Euclid, 159
Bieberbach, 126 et seq. Euler, 9 et seq.
Blasclike, 160 Extension fields, 117
Bolyai, W. and J., 159 Extremal problems, 103 et seq.
Borsuk, 65
Bouwkamp, 92 et seq.
Brooks, 91, 92 Face-centred, 54
Fejes Toth, 21
Fields, 115
Free edges, 8 et seq.
Characteristic, 9 et seq. Functional, 71, 89
Choquet, 110
Cluster of points, 2, 57 et seq.
Compasses, 113 et seq. Gale, 67, 69
Concave function, 11 Golden section, 116
Construction on the sphere, 128 et seq. Graph, 30 et seq., 95 et seq.
Construction with ruler and com
passes, 113 et seq.
Continuum, 89 Hadwiger, 66, 81, 82, 111, 112
Convex domain, 57 Hamel, 84
Convex function, 10, 23 Hilbert, 73, 160
Coverings, 29 et seq.
Covering bodies, 67
Cube, 53, 82
Icosahedron, 77
Insertion ruler, 114
Irreducible equation, 125
D-bodies, 67 Irreducible graph, 32, 40
Decomposition of rectangles, 91 et seq.
imperfect, 94
perfect, 94 Jensen, 11, 23
simple, 94 duel, 76, 79
Dehn, 73 et seq., 101 Jung, 2, 67
Delian problem, 113, 122
Density of packing, 2, 5, 16 et seq., 50
et seq. Kirchhoff, 101
Diameter, 57 Kraitchik, 91 et seq.
Dual graph, 96 Kreweras, 110
168 I NDEX
Lattice packing, 50 et seq. Ruler, 113 et seq.

Lebesgue, 57
Lexell, 18
Lobachevsky, 159 Schiitte, 1, 38 et seq.
Set-theoretical squaring problem, 4
Sierpinskl, 140
Network, 101 Simplex, 69
Number field, 115 el seq. Smith, 91
Spherical compasses, 130 et seq.
Sprague, 60 et seq., 91 et seq., 112
Octahedron, 67 el seq., 11 Squaring the circle on the sphere, 3
Ohm, 101 Steiger, 112
Steiner, 161 et seq.
Stereographic projection, 128 et seq.
Packings, 5 et seq., 29 et seq., 50 el seq.
Pal, 57, 60 Stone, 91
Syc/ler, 11 et seq.
Parallel postulate, 159
Perron, 162
Plato, 113, 160
Tetrahedron, 78
Point sets, 57 et seq.
Tile, 2, 57 et seq.
Polyhedron, functional of, 88
Triangular polyhedron, 21
Polygonal net, 8
Pre-dependent, 85 et seq. Trisection, 121
Tschirnhaus, 122
Ptolemy, 111
Tulle, 91
Quinquesection, 117 et seq.

van der Waerden, 1, 38
Regular packings, 5
Reulaux, 63
Rhombic dodecahedron, 77 Width, 62

Meschkowski - Unsolved and Unsolvable Problems in Geometry

Uploaded by

Copyright:

Available Formats

Meschkowski - Unsolved and Unsolvable Problems in Geometry

Uploaded by

Document Information

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

Meschkowski - Unsolved and Unsolvable Problems in Geometry

Uploaded by

Copyright:

Available Formats

H.

HERBERT MESCHKOWSKI, Ph.D.

FREDERICK UNGAR PUBLISHING CO.

First English Edition 1966

© 1966, Oliver & Boyd Ltd

Library of Congress Catalog Card No. 65-25028

At a time when new theories are constantly being developed in

The numbers in square brackets refer to the bibliography at the

H erbert M eschkow ski

VIII. UNSOLVABLE EXTREMAL PROBLEMS 103

IX. CONSTRUCTIONS WITH RULER AND COMPASSES 113

X. CONSTRUCTIONS ON THE SPHERE 128

XI. PROBLEMS IN THE GEOMETRY OF SETS 137

XII. THE METHODOLOGICAL SIGNIFICANCE OF “ UNSOLV­

First of all we state the most important of the problems discussed

& = /n + /t2 + £ = <?■!+ <?2+ ••• + ?,,, (1)

is irrelevant to our problem. It is sufficient to agree that, corres­

REGULAR P A C K IN G S I N THE PLANE

1. P roblems on the D ensity of P a ck ing

radius r = - which have no points in common. By displacing the

points Pv, the minimum distance d is altered and, from continuity

radius r = - drawn about the points P v (v = 1,2, ..., n) for such an

extremal configuration becomes a maximum together with d.

lim - —- = lim = lim SR (2)

Now we are able to formulate the problems of plane geometry

It is very simple to solve these problems if one assumes a lattice­

this is done the number of vertices remains unchanged while A and

remaining system of triangles. If this is done, E and A change by

/(* .) + /(* 2) + ■+/{*,) a n f b Xl± — -+ (6)

On the other hand for concave functions f (x)

/(* i) + /( * 2) + ... +/(*„) (6-)

It is sufficient to prove (6). The proof for concave functions is

« / ( * ,) + /(* 2) +/(*a) +/(*„)) i i + y (~ Y ^ j)

The validity of (6) for n = 2m, m>2, is deduced by a standard

which also lies in the interval [a, By our induction hypothesis

f( x i) + f(x 2) + ... + f(xn_ i) ^ n -/(*„)• (8)

If one now multiplies (7) by n —1 it becomes clear that the following

/(* 1) + f(x 2) + ••• +/(*„-1) ^ n f ( x n) - J \ x „) =

This is the inequality we wanted to prove.

sin x t + sin .\2 = 2 sin * l + * 2 5^2 sin * l+ * 2

is true for x e \ a , b \ £e[a,b'\. Now let f " ( x ) > 0 in the whole

3. P a c k in g P roblems in the P lane

120° in any of the triangles. Let us begin to construct the net by

strip, then the triangle CXC2C2 is acute angled. However, if C3

Hence we have proved:—

D = iim — ^ ~ = 1-209.... (18)

All triangles A B C + whose apexes C + lie on /*, have therefore

inscribed in stf. Hence the equilateral triangle must be this greatest

6 cot-1 (J 3 cos r) —n> 'K—,

r 90° 70° 52' 54° 46' 37° 43'

D 1-5 1-33... 1-268... 1-254...

Let v, e and a be as usual the numbers of vertices, edges and areas

is true, where the function

of a system of n congruent, non-overlapping spherical caps of

r 60° 54° 4 6 ' 45° 31° 4 3 '

<5 < 4- (32)

From this we calculate the limit given in (32). In fact

A packing of non-overlapping circles in the plane | is said to be

dt = ■2n ‘ -1 = — -0-6046. (33)

However, this system SPx has by no means the smallest density

XII. THE METHODOLOGICAL SIGNIFICANCE OF “ UNSOLV

is irrelevant to our problem. It is sufficient to agree that, corres

It is very simple to solve these problems if one assumes a lattice

« / ( * ,) + /(* 2) +/(a) +/(„)) i i + y (~ Y ^ j)

t Of course the corresponding definition is possible for a system of non

The proof of this lemma is particularly simple if, in the parallelo