70

EX LIB RIS
« V
Projective Geometry
 A Blaisdell Book in the Pure and Applied Sciences

consulting editors George Springer, University of Kansas

Robert E. K. Rourke, St. Stephens, Rome, Italy
Projective Geometry

H. S. M. Coxeter
UNIVERSITY OF TORONTO

BLAISDELL PUBLISHING COMPANY

NEW YORK • LONDON * TORONTO

A Division of Ginn and Company

 TO MY WIFE

First Edition, 1964

© Copyright, 1964, by Blaisdell Publishing Company, a Division of Ginn and Company.
All rights reserved.
Library of Congress Catalog Card Number: 64-10589
Preface

In Euclidean geometry, constructions are made with the ruler and com¬
pass. Projective geometry is simpler: its constructions require only the ruler.
We consider the straight line joining two points, and the point of intersection
of two lines, with the further simplification that two lines never fail to
meet!
In Euclidean geometry we compare figures by measuring them. In projective
geometry we never measure anything; instead, we relate one set of points
to another by a projectivity. Chapter 1 introduces the reader to this important
idea. Chapter 2 provides a logical foundation for the subject. The third and
fourth chapters describe the famous theorems of Desargues and Pappus.
The fifth and sixth make use of projectivities on a line and in a plane, re¬
spectively. In the next three we develop a self-contained account of von
Staudt’s approach to the theory of conics, made more “modern” by allowing
the field to be general (though not of characteristic 2) instead of real or
complex. This freedom has been exploited in Chapter 10, which deals with the
simplest finite geometry that is rich enough to illustrate all our theorems
nontrivially (for instance, Pascal’s theorem concerns six points on a conic,
and in PG{2, 5) these are the only points on the conic). In Chapters 11 and 12
we return to more familiar ground, showing the connections between pro¬
jective geometry, Euclidean geometry, and the popular subject of “analytic
geometry.”
The possibility of writing an easy book on projective geometry was fore¬
seen as long ago as 1917, when D. N. Lehmer [1,* Preface, p. v] wrote:

The subject of synthetic projective geometry is . . . destined shortly to force

its way down into the secondary schools.

More recently, A. N. Whitehead [1, p. 133] recommended a revised cur¬

riculum beginning with Congruence, Similarity, Trigonometry, Analytic

* References are given on page 158.

Vll
Vlll PREFACE

Geometry, and then:

In this ideal course of Geometry, the fifth stage is occupied with the
elements of Projective Geometry . . .

This “fifth” stage has one notable advantage: its primitive concepts are so
simple that a self-contained account can be reasonably entertaining, whereas
the foundations of Euclidean geometry are inevitably tedious.
The present treatment owes much to the famous text-book of Veblen and
Young [1], which has the same title. To encourage truly geometric habits of
thought, we scrupulously avoid the use of coordinates and all metrical ideas
(Whitehead’s first four “stages”) except in Chapters 1, 11, 12, and a few of
the Exercises. In particular, the only mention of cross ratio is in three exer¬
cises at the end of Section 12.3.
I gratefully acknowledge the help of M. W. Al-Dhahir, W. L. Edge,
P. R. Halmos, S. Schuster and S. Trott, who constructively criticized the
manuscript, and of H. G. Forder and C. Garner, who read the proofs. I wish
also to express my thanks for permission to quote from Science: Sense and
Nonsense by J. L. Synge (Jonathan Cape, London).
H. S. M. COXETER
Toronto, Canada
February, 1963
Contents

A Note to the Reader xin

CHAPTER 1 Introduction

1.1 What is projective geometry? 1

1.2 Historical remarks 2
1.3 Definitions 5
1.4 The simplest geometric objects 6
1.5 Projectivities 8
1.6 Perspectivities 10

CHAPTER 2 Triangles and Quadrangles

2.1 Axioms 14
2.2 Simple consequences of the axioms 16
2.3 Perspective triangles 18
2.4 Quadrangular sets 20
2.5 Harmonic sets 22

CHAPTER 3 The Principle of Duality

3.1 The axiomatic basis of the principle of duality 24

3.2 The Desargues configuration 26
3.3 The invariance of the harmonic relation 28
 CONTENTS

3.4 Trilinear polarity 29

3.5 Harmonic nets 30

CHAPTER 4 The Fundamental Theorem and Pappus’s Theorem

4.1 How three pairs determine a projectivity 33

4.2 Some special projectivities 35
4.3 The axis of a projectivity 36
4.4 The Pappus configuration 38

CHAPTER 5 One-dimensional Projectivities

5.1 Superposed ranges 41

5.2 Parabolic projectivities 43
5.3 Involutions 45
5.4 Hyperbolic involutions 47

CHAPTER 6 Two-dimensional Projectivities

6.1 Projective collineations 49

6.2 Perspective collineations 52
6.3 Involutory collineations 55
6.4 Projective correlations 57

CHAPTER 7 Polarities

7.1 Conjugate points and conjugate lines 60

7.2 The use of a self-polar triangle 62
7.3 Polar triangles 64
7.4 A construction for the polar of a point 65
7.5 The use of a self-polar pentagon 67
7.6 A self-conjugate quadrilateral 68
7.7 The product of two polarities 68

CHAPTER 8 The Conic

8.1 How a hyperbolic polarity determines a conic 71

8.2 The polarity induced by a conic 75
8.3 Projectively related pencils 76
 CONTENTS

8.4 Conics touching two lines at given points 78

8.5 Steiner’s definition for a conic 80

CHAPTER 9 The Conic, Continued

9.1 The conic touching five given lines 81

9.2 The conic through five given points 85
9.3 Conics through four given points 87
9.4 Two self-polar triangles 88
9.5 Degenerate conics 89

chapter 10 A Finite Projective Plane

10.1 The idea of a finite geometry 91

10.2 A combinatorial scheme for PG(2, 5) 92
10.3 Verifying the axioms 95
10.4 Involutions 96
10.5 Collineations and correlations 97
10.6 Conics 98

chapter 11 Parallelism

11.1 Is the circle a conic? 102

11.2 Affine space 103
11.3 How two coplanar lines determine a flat pencil and a
bundle 105
11.4 How two planes determine an axial pencil 106
11.5 The language of pencils and bundles 107
11.6 The plane at infinity 108
11.7 Euclidean space 109

chapter 12 Coordinates

12.1 The idea of analytic geometry 111

12.2 Definitions 112
12.3 Verifying the axioms for the projective plane 116
12.4 Projective collineations 119
 CONTENTS

12.5 Polarities 122

12.6 Conics 124
12.7 The analytic geometry of PG(2, 5) 126
12.8 Cartesian coordinates 129

Answers to Exercises 133

References 158

Index 159
A Note to the Reader

Almost every italicized statement preceded by a

number, such as 1.63, is a theorem unless it is
explicitly stated to be an axiom. However, 2.32 and
3.11, which arise as theorems, are used later as
axioms.
 CHAPTER ONE

Introduction

If Desargues, the daring pioneer of the seventeenth century, could

have foreseen what his ingenious method of projection was to lead
to, he might well have been astonished. He knew that he had done
something good, but he probably had no conception of just how
good it was to prove.
E. T. Bell (1883-1960)
[Reference 3, p. 244]

1.1 What is Projective Geometry?

The plane geometry of the first six books of Euclid’s Elements may be
described as the geometry of lines and circles: its tools are the straight-edge
(or unmarked ruler) and the compasses. A remarkable discovery was made
independently by the Danish geometer Georg Mohr (1640-1697) and the
Italian Lorenzo Mascheroni (1750-1800). They proved that nothing is lost
by discarding the straight-edge and using the compasses alone. * For instance,
given four points A, B, C, D, we can still construct the point where the lines
AB and CD would meet if we had the means to draw them; but the actual
procedure is quite complicated. It is natural to ask how much remains if we
discard the compass instead, and use the straight-edge alone.f At a glance,
it looks as if nothing at all will remain: we cannot even carry out the construc¬
tion described in Euclid’s first proposition. Is it possible to develop a geometry

* See Reference 6, pp. 144-151, or Reference 8, p. 79.

,
f See Reference 16 pp. 41-43.
1
2 INTRODUCTION

having no circles, no distances, no angles, no intermediacy (or “betweenness”),

and no parallelism? Surprisingly, the answer is Yes; what remains is projec¬
tive geometry: a beautiful and intricate system of propositions, simpler than
Euclid’s but not too simple to be interesting. The passage from axioms and
“obvious” theorems to unexpected theorems will be seen to resemble Euclid’s
work in spirit, though not in detail.
This geometry of the straight-edge seems at first to have very little connec¬
tion with the familiar derivation of the name geometry as “earth measure¬
ment.” Though it deals with points, lines, and planes, no attempt is ever
made to measure the distance between two points or the angle between two
lines. It does not even admit the possibility that two lines in a plane might
fail to meet by being “parallel.”
We naturally think of a point as “position without magnitude” or “an
infinitesimal dot,” represented in a diagram by a material dot only just big
enough to be seen. By a line we shall always mean a straight line of unlimited
extent. Part of a line is reasonably well represented by a thin, tightly stretched
thread, or a ray of light. A plane is a flat surface of unlimited extent, that is,
a surface that contains, for any two of its points, the whole of the line joining
them. Any number of points that lie on a line are said to be collinear. Any
number of lines that pass through a point are said to be concurrent. Any
number of points or lines (or both) that lie in a plane are said to be coplanar.
People who have studied only Euclidean geometry regard it as an obvious
fact that two coplanar lines with a common perpendicular are parallel, in
the sense that, however far we extend them, they will remain the same dis¬
tance apart. By stretching our imagination we can conceive the possibility
that this is merely a first approximation: that if we could extend them for
millions or billions of miles we might find the lines getting closer together or
farther apart. When we look along a straight railroad we get the impression
that the two parallel rails meet on the horizon. Anyhow, by assuming that
two coplanar lines always meet, we obtain a system of propositions which (as
we shall verify in Chapter 11) is just as logically consistent as Euclid’s different
system. In the words of D. N. Lehmer (Reference 12, p. 12):

As we know nothing experimentally about such things, we are at liberty to make

any assumptions we please, so long as they are consistent and serve some useful
purpose.

1.2 Historical Remarks

The motivation for this kind of geometry came from the fine arts. It was
in 1425 that the Italian architect Brunelleschi began to discuss the geometrical
 PERSPECTIVE 3

theory of perspective, which was consolidated into a treatise by Alberti a few

years later. Because of this application, it is natural to begin the subject in
three-dimensional space; but we soon find that what happens in a single
plane is sufficiently exciting to occupy our attention for a long time. Plane
projective geometry may be described as the study of geometrical properties
that are unchanged by “central projection,” which is essentially what happens
when an artist draws a picture of a tiled floor on a vertical canvas. The square
tiles cease to be square, as their sides and angles are distorted by foreshorten¬
ing; but the lines remain straight, since they are sections (by the picture-plane)
of the planes that join them to the artist’s eye. Thus projective geometry
deals with triangles, quadrangles, and so on, but not with right-angled tri¬
angles, parallelograms, and so on. Again, when a lamp casts a shadow on a
wall or on the floor, the circular rim of a lampshade usually casts a large
circular or elliptic shadow on the floor and a hyperbolic shadow on the nearest
wall. (Such “conic sections” or conics are sections of the cone that joins the
source of light to the rim of the lampshade.) Thus projective geometry waives
the customary distinction between a circle, an ellipse, a parabola, and a
hyperbola; these curves are simply conics, all alike.
Although conics were studied by Menaechmus, Euclid, Archimedes and
Apollonius, in the fourth and third centuries b.c., the earliest truly projective
theorems were discovered by Pappus of Alexandria in the third century a.d.,
and it was J. V. Poncelet (1788-1867) who first proved such theorems by
purely projective reasoning.
More than two hundred years before Poncelet, the important concept of
a point at infinity occurred independently to the German astronomer Johann
Kepler (1571-1630) and the French architect Girard Desargues (1593-1661).
Kepler (in his Paralipomena in Vitellionem, 1604) declared that a parabola has
two foci, one of which is infinitely distant in both of two opposite directions,
and that any point on the curve is joined to this “blind focus” by a line
parallel to the axis. Desargues (in his Brouillon project. .. , 1639) declared
that parallel lines “sont entre elles (Tune mesme ordonnance dont le but est a
distance injinie.'’' (That is, parallel lines have a common end at an infinite
distance.) And again, “Quand en un plan, aucun des points d'une droit n'y est
a distance finie, cette droit y est a distance infinie.” (When no point of a line
is at a finite distance, the line itself is at an infinite distance). The groundwork
was thus laid for Poncelet to derive projective space from ordinary space by
postulating a common “line at infinity” for all the planes parallel to a given
plane. This ingenious device, which we shall analyze carefully in Chapter 11,
serves to justify our assumption that, in a plane, any two lines meet; for, if the
lines have no ordinary point in common, we say that they meet in a point at
infinity. But we are not really working in projective geometry until we are
prepared to forget the inferior status of such extra points and admit them into
4 INTRODUCTION

the community as full members having the same privileges as ordinary points.
This emancipation of the subject was carried out by another German,
K. G. C. von Staudt (1798-1867). The last vestiges of dependence on ordinary
geometry were removed in 1871, when Felix Klein provided an algebraic
foundation for projective geometry in terms of “homogeneous coordinates,
which had been discovered independently by K. W. Feuerbach and A. F.
Mobius in 1827.
The determination of a point by two lines nicely balances the determina¬
tion of a line by two points. More generally, we shall find that every state¬
ment about points and lines (in a plane) can be replaced by a dual statement
about lines and points. The possibility of making such a replacement is
known as the “principle of duality.” Poncelet claimed this principle as his
own discovery; but its nature was more clearly understood by another French¬
man, J. D. Gergonne (1771-1859). Duality gives projective geometry a
peculiar charm, making it more symmetrical than ordinary (Euclidean)
geometry.
Besides being a thing of beauty in its own right, projective geometry is
useful as supplying a fresh approach to Euclidean geometry. This is especially
evident in the theory of conics, where a single projective theorem may yield
several Euclidean theorems by different choices of the line at infinity; e.g.,
if the line at infinity is a tangent or a secant, the conic is a parabola or a
hyperbola, respectively. Arthur Cayley (1821-1895) and Felix Klein (1849-
1925) noticed that projective geometry is equally powerful in its application
to non-Euclidean geometries. With characteristic enthusiam, the former said:
Metrical geometry is a part of descriptive geometry, and descriptive geometry is all
geometry.

(Cayley, in 1859, used the word “descriptive” where today we would say
“projective.”)

EXERCISES

1. Which of the following figures belong to projective geometry:

(i) a parallelogram,
(ii) an isosceles triangle,
(iii) a triangle and its medians,
(iv) a figure consisting of 4 points, no 3 collinear, and the lines joining
them in pairs,
(v) a circle with a diameter,
(vi) a conic with a secant (i.e., a line meeting it twice),
(vii) a plane curve with a tangent,
(viii) a hexagon (consisting of 6 points named in cyclic order, and the
6 lines that join consecutive pairs)?
 VISH 5

2. How could a lampshade be tilted so that its circular rim would yield a
parabolic shadow on the wall?

3. Translate the following statements into the language of points at infinity:

(i) Through a given point there passes just one line parallel to a given line.
(ii) If two lines are parallel to a third line, they are parallel to each other.

1.3 Definitions

It is convenient to regard a line as a certain set of points, and a plane as a

certain set of points and lines. A point and a line, or a point and a plane,
or a line and a plane, are said to be incident if the former belongs to the latter.
We also say that the former lies on (or in) the latter, and that the latter passes
through the former. We shall consistently use capital italic letters for points,
small (lower case) italic letters for lines, and Greek letters for planes. If a line /
passes through two points P and Q, we say that it joins them and write
/ = PQ. Similarly, if a plane a passes through two lines / and m, or through /
and a nonincident point P, we say that a joins the two lines, or the line and
point, and write
a = Im = ml = IP — PI.
If P lies on both / and m, we say that these lines meet in P, or that P is their
common point (or “intersection”):
P — l • m.
(Notice the special use of the dot: Im is a plane, but / • m is a point.) Similarly,
a line and a plane may have a common point / • a, and two planes may have a
common line a • (3.
J. L. Synge (Reference 18, p. 32) has described an amusing and instructive
game called Yish (short for “vicious circle”):
The Concise Oxford Dictionary devotes over a column to the word “point” . . .
“that which has position but not magnitude.” This definition passes the buck, as
all definitions do. You now have to find out what “position” and “magnitude” are.
This means further consultation of the Dictionary, and we may as well make the
best of it by turning it into a game of Vish. So here goes.
Point = that which has position but not magnitude.
Position = place occupied by a thing.
Place = part of space. . ..
Space = continuous extension. . . .
Extension = extent.
Extent = space over which a thing extends.
Space = continuous extension... .
6 INTRODUCTION

The word Space is repeated. We have Vish In Seven. . . . Well, what about it?
Didn’t we see and prove that a vicious circle is inevitable, so why be surprised that
we get one here? If that is your reaction, I shout with joy.. . .
Vish illustrates the important principle that any definition of a word must
inevitably involve other words, which require further definitions. The only
way to avoid a vicious circle is to regard certain primitive concepts as being so
simple and obvious that we agree to leave them undefined. Similarly, the proof
of any statement uses other statements; and since we must begin somewhere,
we agree to leave a few simple statements unproved. These primitive state¬
ments are called axioms.
In addition to the primitive concepts and axioms, we take for granted the
words of ordinary speech, the ideas of logical argument, and the principle of
one-to-one correspondence. The last is well illustrated by the example of cups
and saucers. Suppose we had about a hundred cups and about a hundred
saucers and wished to know whether the number of cups was actually equal
to the number of saucers. This could be done, without counting, by the simple
device of putting each cup on a saucer, that is, by establishing a one-to-one
correspondence between the cups and saucers.

EXERCISES

1. Play Vish beginning with the words:

(i) Axiom,
(ii) Dimension,
(iii) Fraction.

2. Set up a one-to-one correspondence between the sequence of natural

numbers 1, 2, 3, 4, .. . and the sequence of even numbers 2, 4, 6, 8, ... .
Are we justified in saying that there are just as many even integers as there
are integers altogether ?

1.4 The Simplest Geometric Objects

A basis for projective geometry may be chosen in various ways. It seems

simplest to use three primitive concepts: point, line, and incidence. In terms
of these we can easily define “lie on,” “pass through,” “join,” “meet,”
“collinear,” “concurrent,” and so on. It is not quite so obvious that we can
define a plane; but if a point P and a line / are not incident, the plane PI may be
taken to consist of all the points that lie on lines joining P to points on /, and
all the lines that join pairs of distinct points so constructed.
 QUADRANGLE AND QUADRILATERAL 7

A triangle PQR consists of three noncollinear points P, Q, R, called its

vertices, and the three joining lines QR, RP, PQ, called its sides. (When we
have formulated the axioms and some of their simple consequences, we shall
see that the triangle PQR can be proved to lie in the plane PQR.) Thus, if
3 points are joined in pairs by 3 lines, they form a triangle, which is equally
well formed by 3 lines meeting by pairs in 3 points. The case of 4 points or
4 lines is naturally more complicated, and we will find it convenient to give
the definitions in “parallel columns” (although it is not seriously expected
that anybody will read the left column with the left eye and simultaneously
the right column with the right eye).

If 4 points in a plane are joined If 4 lines in a plane meet by

in pairs by 6 distinct lines, they are pairs in 6 distinct points, they are
called the vertices of a complete called the sides of a complete quad¬
quadrangle, and the lines are its 6 rilateral, and the points are its 6
sides. Two sides are said to be vertices. Two vertices are said to
opposite if their common point is be opposite if their join is not a side.
not a vertex. The common point The join of two opposite vertices is
of two opposite sides is called a called a diagonal line. There are 3
diagonal point. There are 3 diagonal diagonal lines. In Figure 1.4a, the
points. In Figure 1.4a, the quad¬ quadrilateral is pqrs, its vertices
rangle is PQRS, its sides are are
p ■ s, q ■ s, r • s,
PS, QS, RS,
q-r, r-p, p • q,
QR, RP, PQ,
and its diagonal points are and its diagonal lines are

A, B, C. a, b, c.

When there is no possibility of misunderstanding, we speak simply of

quadrangles and quadrilaterals, omitting the word “complete.” This word was

Figure 1.4a
8 INTRODUCTION

introduced to avoid confusion with an ordinary quadrangle, which has 4

vertices and 4 sides; for instance, the ordinary quadrangle PQRS has sides
PQ, QR, RS, SP. It is more usual to call this a “quadrilateral,” but to do so
is unreasonable, as the word “triangle” refers to its vertices rather than its
sides, and so too does the word “pentagon.” The only other polygon that
we shall have occasion to use is the (ordinary) hexagon, which has 6 vertices
and 6 sides.

EXERCISES

Regarding the triangle as a complete 3-point and the complete quadrangle

as a complete 4-point, define analogously (for any natural number n):

(i) a complete w-point, (ii) a complete «-line.

1.5 Projectivities

It is sometimes convenient to use the name range for the set of all points on
a line, and pencil for the set of all lines that lie in a plane and pass through a
point. Ranges and pencils are instances of one-dimensional forms. We shall
often have occasion to consider a one-to-one correspondence between two
one-dimensional forms. The simplest such correspondence between a range
and a pencil arises when corresponding members are incident. In this case it is
naturally understood that the line o on which the points of the range lie is not
incident with the point O through which the lines of the pencil pass. Thus
the range is a section of the pencil (namely, the section by the line o) and the
pencil projects the range (from the point O). As a notation for this elementary
correspondence we may write either

X 7\ x,
where A is a variable point of the range and x is the corresponding line of the
pencil (as in Figure 1.5a), or
ABC • * • a abc • • • ,

where A, B, C, .. . are particular positions of Xand a, b, c, .. . are the corre¬

sponding positions of jc (as in Figure 1.5b).
In such a relation, the order in which the symbols for the points or lines
are written does not necessarily agree with the order in which the points or
lines occur in the range or pencil. (In fact, the latter “order” is not defined!)
Corresponding symbols are placed in corresponding positions, but the
statement ABC • • • a abc • • •, has the same meaning as BAC • • • a bac • • • ,
and so forth.
 PROJECTIVITIES 9

Since the statement X /\ x means that X and x are incident, we can just as
well write
x a X;
but now it is convenient to make a subtle distinction. The correspondence
X a x is directed “from X to x”: it transforms X into x; but the inverse
correspondence x a X transforms x into X.

Figure 1.5 a Figure 1.5 b

A more sophisticated kind of transformation can be constructed by com¬

bining any number of elementary correspondences. For this purpose, we use a
sequence of lines and points occurring alternately:
o, O-^, o2, . . . , on, On.
We allow the sequence to begin with a point (by omitting o) or to end with a
line (by omitting On, as in Figure 1.5c), but we insist that adjacent members

On-1

shall be nonincident and that alternate members (such as O and Ol9 or ox

and o2) shall be distinct. This arrangement of lines and points enables us to
establish a transformation relating the range of points X on o (or the pencil of
lines x through O) to the pencil of lines x(n) through On (or the range of points
Xin) on on). We call such a transformation a projectivity.
Instead of
A A X A X' a x' A x" A • • • A *(n) A *'">
we write simply x x(n)

or x a *(n), or x a ^(n), or X a ^(n)-

10 INTRODUCTION

In other words, we extend the meaning of the sign a from an elementary

correspondence to the product (or “resultant”) of any number of elementary
correspondences.
This extension of meaning is comparable to the stage in arithmetic when
we extend the meaning of number from an integer to a fraction: the quotient
of two integers.

EXERCISES

1. In the elementary correspondence X a x, why is it necessary for the line

o and the point O to be nonincident?

2. Draw a version of Figure 1.5c using three points A, B, C, as in Figure 1.5b.

3. Draw a version of Figure 1.5c with n — 2 and o2 = o, so as to establish

a projectivity X a X" relating pairs of points on o. Where is X" when X
is on oxl Where is X" when X is on OOxl

1.6 Perspectivities

One kind of projectivity is sufficiently important to deserve a special

name and a slightly more elaborate sign: the product of two elementary
correspondences is called a perspectivity and is indicated by the sign x (with
two bars). Using Poncelet’s device of parallel columns to emphasize the
“principle of duality,” as in Section 1.4, we may describe this transformation
as follows:

Two ranges are related by a Two pencils are related by a

perspectivity with center O if they perspectivity with axis ox if they
are sections of one pencil (consist¬ project one range (consisting of all
ing of all the lines through O) by the points on ox) from two distinct
two distinct lines o and op, that is, points O and Op, that is, if the
if the join XX' of corresponding intersection x • x' of corresponding
points continually passes through lines continually lies on the line ov
the point O. In symbols: In symbols:

X x X' or X% X'. — /
x x * or a x x .
,
For instance, in Figure 1.6a (where A, B, C are particular instances of the
variable point X, and a, b, c of the vari, ible line x), we have the perspectivities
Oi
ABC ^ A’B'C, abc x a b'c'.

which can be analyzed in terms of elementary correspondences as follows:

ABC a abc a A'B'C', abc a A'B'C' a a'b'c'.

 PROJECTIVITIES 11

Figure 1.6a

Given three distinct points A, B, C on a line, and three distinct points

A", B", C" on another line, we can set up two perspectivities whose product
has the effect
ABC a A"B"C"
in the manner of Figure 1.6b, where the axis (or “intermediary line”) of the
projectivity joins the points

B' = AB" • BA", C' = AC" • CA",

so that if A' — AA" • B'C',

(1.61) ABC ^ A'B'C 4 A"B"C".

For each point X on AB, we can construct a corresponding point X" on A"B"
by joining A to the point X' — A"X • B'C', so that

(1.62) ABCX ^ A'B'C'X' 4 A"B"C"X"

We shall see, in Chapter 4, that this projectivity ABC a A"B"C" is unique,
in the sense that any sequence of perspectivities relating ABC to A"B"C" will
have the same effect on X.

Figure 1 .6b Figure 1.6c

12 INTRODUCTION

Figure 1.6d

Interchanging points and lines, we obtain an analogous construction

(Figure 1.6c) for the projectivity abc a a"b"c", where a, b, c are three distinct
lines through a point and a", b", c" are three distinct lines through another
point.
Another example of a projectivity is illustrated in Figure 1.6d, where
A, B, C, D are any four collinear points, R is a point outside their line,
T, Q, W are the sections of RA, RB, RC by an arbitrary line through D, and
Z is the point AQ ■ RC. In this case

ABCD tv ZRCW 4 QTDW ^ BADC.

Hence
ABCD a BADC.
Expressing this result in words, we have the following theorem:

1.63 Any four collinear points can be interchanged in pairs by a projectivity.

As a third instance, we have, in Figure 1.6e,

ABC ^ APS x AFB, ABC^AQR^AFB.

Figure 1.6e
 PROJECTIVITIES 13

In this projectivity ABC a AFB, the point A corresponds to itself. A point

that corresponds to itself is said to be invariant.

The idea of a projectivity is due to Poncelet. Its analysis into elementary

correspondences was suggested by Mathews (Reference 14, p. 39). The sign
a was invented by von Staudt. For the special case of a perspectivity, the

sign A was adopted by the great American geometer Oswald Yeblen (1880—
1960).

EXERCISES

1. Given three collinear points A, B, C, set up two perspectivities whose

product has the effect ABC a BAC.

2. Given three concurrent lines a, b, c set up two perspectivities whose

product has the effect abc a bac.

3. Given three collinear points A, B, C and three concurrent lines a, b, c, set

up five elementary correspondences (“two-and-a-half perspectivities”)
whose product has the effect ABC a abc.

4. Given four collinear points A, B, C, D, set up three perspectivities whose

product has the effect ABCD a DCBA.
 CHAPTER TWO

Triangles and Quadrangles

To construct a geometry is to state a system of axioms and to

deduce all possible consequences from them. All systems of pure
geometry . . . are constructed in just this way. Their differences . . .
are differences not of principle or of method, but merely of richness
of content and variety of application .... You must naturally be
prepared to sacrifice simplicity to some extent if you wish to be
interesting.
G. H. Hardy (1877-1947)
(“What is geometry ?” Mathematical Gazette,
12 (1925), pp. 314, 315.)

2.1 Axioms

As we saw in Section 1.3, the complete development of any branch of

mathematics must begin with some undefined entities (primitive concepts)
and unproved propositions (axioms). The precise choice is a matter of taste.
It is, of course, essential that the axioms be consistent (not contradicting one
another) and it is desirable that they be independent, simple, and plausible.
Such foundations for projective geometry were first proposed by two Italians:
Gino Fano (in 1892) and Mario Pieri (in 1899). The following eight axioms,
involving three primitive concepts (point, line, and incidence) differ only
slightly from those proposed by Veblen and Young (Reference 19. pp. 16,
18, 24, 45). (We have already seen how the words plane, quadrangle, and
projectivity can be defined in terms of the primitive concepts.)
14
 AXIOMS FOR PROJECTIVE SPACE 15

Axiom 2.11 There exist a point and a line that are not incident.

Axiom 2.12 Every line is incident with at least three distinct points.

Axiom 2.13 Any two distinct points are incident with just one line.

Axiom 2.14 If A, B, C, D are four distinct points such that AB meets CD,
then AC meets BD. (See Figure 2.1a.)

Axiom 2.15 If ABC is a plane, there is at least one point not in the plane
ABC.

Axiom 2.16 Any two distinct planes have at least two common points.

Axiom 2.17 The three diagonal points of a complete quadrangle are never
collinear. (See Figure 1.4a.)

Axiom 2.18 If a projectivity leaves invariant each of three distinct points

on a line, it leaves invariant every point on the line.

Figure 2.1a

The best possible advice to the reader is to set aside all his previously
acquired knowledge (such as trigonometry and analytic geometry) and use
only the axioms and their consequences. He may occasionally be tempted to
use the old methods to work out one of the exercises; but then he is likely to
be so engulfed in ugly calculations that he will return to the synthetic method
with renewed enthusiasm.

EXERCISES

1. Give detailed proofs of the following theorems, pointing out which axioms
are used:
(i) There exist at least four distinct points.
(ii) If a is a line, there exists a point not lying on a.
(iii) If A is a point, there exists a line not passing through A.
(iv) Every point lies on at least three lines.
2. Construct a projectivity having exactly two invariant points. [Hint: Use
Exercise 3 of Section 1.5.]
16 TRIANGLES AND QUADRANGLES

3. Draw an equilateral triangle ABC with its incircle DEF, medians AD, BE,
CF, and center G. Notice that the figure involves 7 points, 6 lines, and 1
circle. Consider a “geometry,” consisting entirely of 7 points and 7 lines,
derived from the figure by calling the circle a line (and ignoring the extra
intersections). Which one of the “two-dimensional” axioms (Axioms 2.11,
2.12, 2.13, 2.14, 2.17, 2.18) is denied? [Hint: Where are the diagonal points
of the quadrangles ABCG, AEFG, BCEF1]

2.2 Simple Consequences of the Axioms

Most readers will have no difficulty in accepting Axioms 2.11, 2.12, 2.13.
The first departure from Euclidean geometry appears in Axiom 2.14, which
rules out the possibility that AC and BD might fail to meet by being “parallel.”
This axiom, which resembles Pasch’s Axiom (12.27 of Reference 8, p. 178), is
Veblen’s ingenious device for declaring that any two coplanar lines have a
common point before defining a plane! (In fact, the line BD, whose inter¬
section with AC is asserted, must lie in the plane AEC, where E — AB • CD,
since B lies on AE, and D on EC.)
Given a triangle ABC, we can define a pencil of lines through C as consisting
of all the lines CX, where X belongs to the range of points on AB. The first
four axioms are all that we need in order to define the plane ABC as a certain
set of points and lines, namely, all the points on all the lines of the pencil, and
all the lines that join pairs of such points. We then find that the same plane is
determined when we replace C by another one of the points, and AB by one
of the lines not incident with this point.
Axiom 2.15 makes the geometry three-dimensional, and Axiom 2.16
prevents it from being four-dimensional. (In fact, four-dimensional geometry
would admit a pair of planes having only one common point!) It follows that
two distinct planes, a and (5, meet in a line, which we call the line a • /?.
In virtue of Axiom 2.17, the diagonal points of a quadrangle form a triangle.
This is called the diagonal triangle of the quadrangle. It will be found to play
an important role in some of the later developments. However (as we shall
see in the exercise at the end of Section 10.3), there are some interesting,
though peculiar, geometries in which the diagonal points of a quadrangle are
always collinear, so that Axiom 2.17 is denied.
The plausibility of Axiom 2.18 will appear in Section 3.5, where we shall
prove, on the basis of the remaining seven axioms, that a projectivity having
three invariant points leaves invariant, if not the whole line, so many of its
points that they have the “appearance” of filling the whole line.
As an indication of the way axioms lead to theorems, let us now state four
simple theorems and give their proofs in detail.
 THE SIMPLEST THEOREMS 17

2.21 Any two distinct lines have at most one common point.
Proof. Suppose, if possible, that two given lines have two common
points A and B. Axiom 2.13 tells us that each line is determined by these
two points. Thus the two lines coincide, contradicting our assumption that
they are distinct.

2.22 Any two distinct coplanar lines have at least one common point.
Proof. Let E be a point coplanar with the two lines but not on either
of them. Let AC be one of the lines. Since the plane ACE is determined by
the pencil of lines through E that meet AC, the other one of the two given
lines may be taken to join two points on distinct lines of this pencil, say B
on EA, and D on EC, as in Figure 2.1a. According to Axiom 2.14, the two
lines AC and BD have a common point.

2.23 If two lines have a common point, they are coplanar.

Proof. If two lines have a common point C, we may name them AC,
BC, and conclude that they lie in the plane ABC.

2.24 There exist four coplanar points of which no three are collinear.
Proof. By our first three axioms, there exist two distinct lines having
a comfnon point and each containing at least two other points, say lines
EA and EC containing also B and D, respectively, as in Figure 2.1a. The
four distinct points A, B, C, D have the desired property of noncollinearity.
For instance, if the three points A, B, C were collinear, E (on AB) would be
collinear with all of them, and EA would be the same line as EC, contra¬
dicting our assumption that these two lines are distinct.

Without Theorem 2.24, Axiom 2.17 might be “vacuous”: it merely says

that, if a complete quadrangle exists, its three diagonal points are not
collinear.
Notice the remarkably compact foundation which is now seen to suffice
for the erection of the whole system of projective geometry.

EXERCISES

1. Prove the following theorems:

(i) There exist four coplanar lines of which no three are concurrent.
(ii) The three diagonal lines of a complete quadrilateral are never con¬
current. (They are naturally said to form the diagonal triangle of the
quadrilateral.)

2. Draw complete quadrangles and quadrilaterals of various shapes, indi¬

cating for each its diagonal triangle.
18 TRIANGLES AND QUADRANGLES

2.3 Perspective Triangles

Two ranges or pencils are said to be perspective if they are related by a

perspectivity. This notion can be extended to plane figures involving more than
one point and more than one line, as follows. Two specimens of such a figure
are said to be perspective if their points can be put into one-to-one correspond¬
ence so that pairs of corresponding points are joined by concurrent lines, or
if their lines can be put into one-to-one correspondence so that pairs of

Figure 2.3a

corresponding lines meet in collinear points. For instance, the two triangles
PQR and P'Q'R' in Figure 2.3a are perspective since corresponding vertices
are joined by the three concurrent lines PP', QQ', RR', or since corre¬
sponding sides meet in the three collinear points

D = QR ■ Q’R’, E = RP • R'P', F = PQ-P'Q'.

When Theorems 2.31 and 2.32 have been stated and proved, we shall see
that either kind of correspondence implies the other. Meanwhile, let us say
tentatively that two figures are perspective from a point O if pairs of corre¬
sponding points are joined by lines through O, and that two figures are
perspective from a line o if pairs of corresponding lines meet on o. (It is some¬
times convenient to call O the center, and o the axis. Whenever we speak of
perspective figures we assume that the points, and also the lines, are all
distinct; for example, in the case of a pair of triangles, we assume that there
are six distinct vertices and six distinct sides.) The desired identification will
follow for all more complicated figures as soon as we have established it for
triangles. Accordingly, we begin with the following theorem:
 DESARGUES 19

2.31 If two triangles are perspective from a line they are perspective from
a point.
Proof. Let two triangles, PQR and P'Q'R', be perspective from a line
o. In other words, let o contain three points D, E, F, such that D lies on
both QR and Q'R', E on both RP and R'P', F on both PQ and P'Q'. We
wish to prove that the three lines PP', QQ', RR! all pass through one point
O, as in Figure 2.3a. We distinguish two cases, according as the given
triangles are in distinct planes or both in one plane.
(1) According to Axiom 2.14, since QR meets Q'R', QQ' meets RR'.
Similarly RR' meets PP', and PP' meets QQ'. Thus the three lines PP', QQ',
RR' all meet one another. If the planes PQR and P'Q'R' are distinct, the
three lines must be concurrent; for otherwise they would form a triangle,
and this triangle would lie in both planes.
(2) If PQR and P'Q'R' are in one plane, draw, in another plane through
o, three nonconcurrent lines through D, E, F, respectively, so as to form a
triangle PXQXRX, with QXRX through D, RXPX through E, and PXQX through
F. This triangle is perspective from o with both PQR and P'Q'R'. By the
result for noncoplanar triangles, the three lines PPX, QQX, RRX all pass
through one point S, and the three lines P'Pi, Q'Qi, R'RX all pass through
another point S'. (The points S and S' are distinct; for otherwise Px would
lie on PP' instead of being outside the original plane.) Since Px lies on both
PS and P'S', Axiom 2.14 tells us that SS' meets PP'. Similarly SS' meets
both QQ' and RR'. Hence, finally, the three lines PP', QQ', RR' all pass
through the point
O = PQR • SS1'.
The converse is

2.32 Desargues’s Theorem. If two triangles are perspective from a

point they are perspective from a line.
Proof. Let two triangles, PQR and P’Q’R' (coplanar or noncoplanar)
be perspective from a point O. We see from Axiom 2.14 that their three
pairs of corresponding sides meet, say in D, E, F. It remains to be proved
that these three points are collinear, as in Figure 2.3a. Consider the two
triangles PP'E and QQ'D. Since pairs of corresponding sides meet in the
three collinear points R’, R, O, these triangles are perspective from a line,
and therefore (by 2.31), perspective from a point, namely, from the point
PQ . p'Q' = E. That is, the three points E, D, F are collinear.

Theorem 2.31, the converse of Desargues’s theorem, happens to be easier

to prove ab initio than Desargues’s theorem itself. If, instead, we had
proved 2.32 first (as in Reference 7, p. 8), we could have deduced the con¬
verse by applying 2.32 to the triangles PP'E and QQ’D.
20 triangles and quadrangles

EXERCISES

1. Verify experimentally the correctness of Desargues’s theorem and its

converse for perspective triangles of various shapes in various relative
positions. If P, Q, R, P' and Q' are given, how much freedom do we have
in choosing the position of R' ?

2. If three triangles are all perspective from the same center, then* the three
axes are concurrent. [Hint: Let the three axes be DXEX, D2E2, D3E3. Apply
2.31 to the triangles D1D2D3 and E^E^.]

3. What happens to Theorem 2.31 if we allow corresponding sides of the two

triangles to be parallel, and admit points at infinity ?

2.4 Quadrangular Sets

A quadrangular set is the section of a complete quadrangle by any line g

that does not pass through a vertex. It is thus, in general, a set of six collinear
points, one point on each side of the quadrangle; but the number of points is
reduced to five or four if the line happens to pass through one or two diagonal
points.

Figure 2.4a

Following Veblen and Young (Reference 19, p. 49), we use the symbol

Q(ABC, DEF)

to denote the statement that the six points A, B, C, D, E, F, form a quadran¬

gular set in the manner of Figure 2.4a (that is, lying on the respective sides
PS, QS, RS, QR, RP, PQ of the quadrangle), so that the first three points lie
on sides through one vertex while the remaining three lie on the respectively
opposite sides, which form a triangle. This statement is evidently unchanged if

* Whenever an exercise is phrased as a statement, we understand that it is a theorem to

be proved. The omission of the words “prove that” or “show that” saves space.
 QUADRANGULAR SETS 21

we apply any permutation to ABC and the same permutation to DEF\ for
instance, Q{ABC, DEF) has the same meaning as Q(BAC, EDF), since the
quadrangle PQRS can equally well be calledQPRS. Similarly, the statement
Q(ABC, DEF) is equivalent to each of

Q(AEF, DBC), Q(DBF, AEC), Q(DEC, ABF).

Any five collinear points A, B, C, D, E may be regarded as belonging to a

quadrangular set. To see this, draw a triangle QRS whose sides RS, SQ, QR
pass, respectively, through C, B, D. (These sides may be any three noncon¬
current lines through C, B, D.) We can now construct P = AS ■ ER and
F = g ■ PQ. If we had chosen a different triangle QRS, would we still have
obtained the same point FI Yes:

2.41 Each point of a quadrangular set is uniquely determined by the

remaining points.
Proof. To show that F is uniquely determined by A, B, C, D, E
(Figure 2.4b), we set up another quadrangle P'Q'R'S' whose first five sides
pass through the same five points on g. Since the two triangles PRS and
P'R'S' are perspective from g, Theorem 2.31 tells us that they are also
perspective from a point; thus the line PP' passes through the point
O = RR' • SS'. Similarly, the perspective triangles QRS and Q'R’S' show
that QQ' passes through this same point O. (In other words, PQRS and
P’Q’R'S' are perspective quadrangles.) By Theorem 2.32, the triangles
PQR and P' Q'R', which are perspective from the point O, are also per¬
spective from the line DE, which is g; that is, the sides PQ and P’Q' both
meet g in the same point F.

Figure 2.4b
22 TRIANGLES AND QUADRANGLES

EXERCISES

1. A necessary and sufficient condition for three lines containing the vertices
of a triangle to be concurrent is that their sections A, B, C by a line g form,
with the sections D, E, F of the sides of the triangle, a quadrangular set.

2. If, on a given transversal line, two quadrangles determine the same

quadrangular set (in the manner of Figure 2.4b), their diagonal triangles
are perspective.

2.5 Harmonic Sets

A harmonic set of four collinear points may be defined to be the special

case of a quadrangular set when the line g joins two diagonal points of the

Figure 2.5a

quadrangle, as in Figure 2.5a or 1.6e. Because of the importance of this

special case, we write the relation Q(ABC, ABF) in the abbreviated form

H (AB, CF),

which evidently has the same meaning as H(BA, CF) or H{AB, FC) or
H{BA, FC), namely that A and B are two of the three diagonal points of a
quadrangle while C and F lie, respectively, on the sides that pass through the
third diagonal point. We call Fthe harmonic conjugate of C with respect to A
and B. Of course also C is the harmonic conjugate of F. From Theorem 2.41
we see that F is uniquely determined by A, B, C. (We shall prove in Theorem
3.35 that the relation H(T5, CF) implies H(CF, AB).) For a simple construc¬
tion, draw a triangle QRS whose sides QR, QS, RS pass through A, B, C (as
in Figure 2.5a); then P = AS • BR and F = AB • PQ.
Axiom 2.17 implies that these harmonic conjugates C and F are distinct,
except in the degenerate case when they coincide with A or B. In other words,
 HARMONIC SETS 23

2.51 If A, B, C are all distinct, the relation H(AB, CF) implies that F is
distinct from C.

It follows that there must be at least four points on every line; how many
more is not specified. We shall examine this question in Section 3.5.

EXERCISES

1. Assigning the symbol G to PQ • RS, name two other harmonic sets in

Figure 2.5a.

2. How should the points P, Q, R, S in Figure 2.5a be renamed if the names

of C and F were interchanged ?

3. Derive a harmonic set from a quadrangle consisting of a triangle PQR and

a point S inside.

4. What is the harmonic set determined by a quadrangle PQRS if Axiom 2.17

is denied and the diagonal points are collinear?

5. Suppose, for a moment, that the projective plane is regarded as an exten¬

sion of the Euclidean plane (as in Section 1.2). Referring to Figure 2.5a,
suppose PQ is parallel to AB, so that F is at infinity. What metric result
can you deduce about the location of C with reference to A and B1

6. Still working in the Euclidean plane, draw a line-segment OC, take G

two-thirds of the way along it, and E two-fifths of the way from G to C.
(For instance, make the distances in centimeters OG =10, GE = 2,
EC = 3.) If the segment OC represents a stretched string, tuned to the note
C, the same string stopped at E or G will play the other notes of the major
triad. By drawing a suitable quadrangle, verify experimentally that
H(OE, CG). (Such phenomena explain our use of the word harmonic.)
 CHAPTER THREE

The Principle of Duality

On November 18, 1812, the exhausted remnant of the French

army . . . was overwhelmed at Krasnoi. Among those left for dead
on the frozen battlefield was young Poncelet. ... A searching
party, discovering that he still breathed, took him before the
Russian staff for questioning. As a prisoner of war ... at Saratoff
on the Volga . . . , he remembered that he had received a good
mathematical education, and to soften the rigours of his exile he
resolved to reproduce as much as he could of what he had learned.
It was thus that he created projective geometry.

E. T. Bell (Reference 3, pp. 238-239.)

3.1 The Axiomatic Basis of the Principle of Duality

The geometry of points on a line is said to be one-dimensional. The

geometry of points and lines in a plane is said to be two-dimensional. The
geometry of points, lines, and planes in space is said to be three-dimensional.
It is interesting to observe that the only place where we made essential use of
three-dimensional ideas was in proving 2.31. This excursion “along the third
dimension” could have been avoided by regarding Desargues’s theorem, 2.32
(which implies 2.31) as a new axiom, replacing the three-dimensional axioms
2.15 and 2.16. For a purely two-dimensional theory, we can replace Axiom
2.14 by the simple statement:

3.11 Any two distinct lines have at least one common point.
24
 AXIOMS FOR THE PROJECTIVE PLANE 25

(This is derived from 2.22 by omitting the word “coplanar,” which is now
superfluous.) We shall develop such a theory out of the following seven
propositions:
2.11, 2.12, 2.13, 3.11, 2.17, 2.32, 2.18,

which will thus be seen to form a sufficient set of axioms for the projective
plane.
The two-dimensional principle of duality (i.e., the principle of duality in the
plane) asserts that every definition remains significant, and every theorem
remains true, when we interchange the words point and line (and consequently
also certain other pairs of words such as join and meet, collinear and concur¬
rent, vertex and side, and so forth). For instance, the dual of the point AB • CD
is the line (a • b)(c • d). (Since duality interchanges joining and meeting, it
requires not only the interchange of capital and small letters but also the
removal of any dots that are present and the insertion of dots where they are
absent.) The arrangement in Section 4.1 of definitions in “parallel columns”
shows at once that the dual of a quadrangle (with its three diagonal points)
is a quadrilateral (with its three diagonal lines). Still more simply, the dual of
a triangle (consisting of its vertices and sides) is again a triangle (consisting
of its sides and vertices); thus a triangle is an instance of a self-dual figure.
Axiom 2.11 is clearly self-dual. The dual of 2.12 is easily proved. The dual
of 2.13 consists of 2.21 and 3.11; therefore the dual of 3.11 is part of 2.13.
Similarly, the duals of 2.17 and 2.18 present no difficulty. The dual of 2.32
is its converse, 2.31 (which can be proved by applying 2.32 to the triangles
PP'E and QQ'D in Figure 2.3a). Thus all the axioms for the projective plane
imply their duals.
This remark suffices to establish the validity of the two-dimensional prin¬
ciple of duality. In fact, after using the axioms and their consequences in
proving a given theorem, we can immediately assert the dual theorem; for, a
proof of the dual theorem could be written down quite mechanically by
dualizing each step in the proof of the original theorem. (Of course, our proof
of 2.31 cannot be dualized in this sense, because it is three-dimensional; but
this objection is avoided by taking 2.32 as an axiom and observing that 2.31
can be deduced from it without either leaving the plane or appealing to the
principle of duality.)
One of the most attractive features of projective geometry is the symmetry
and economy with which it is endowed by the principle of duality: fifty
detailed proofs may suffice to establish as many as a hundred theorems.
So far, we have considered only the two-dimensional principle of duality.
But by returning to our original Axioms 2.11 through 2.18, we can just as
easily establish the three-dimensional principle of duality, in which points,
lines, and planes are interchanged with planes, lines, and points. For instance,
26 THE PRINCIPLE OF DUALITY

if lines a and b are coplanar, the dual of ab is a • b. However, for the sake of
brevity we shall assume, until the end of Chapter 9, that all the points and
lines considered are in one plane. (For a glimpse of the analogous three-dimen¬
sional developments, see Reference 8, p. 256.)

EXERCISES

1. Draw the duals of Figures 1.6d, 1.6e, and 2.4a.

2. Let the diagonal points of a quadrangle PQRS be

A = PS • QR, B= QS- RP, C = RS ■ PQ,

as in Figure 1.4a. Define the further intersections

= BC- QR, B1 = CA • RP, Cx = AB ■ PQ,

A2 = BC■ PS, B2 = CA - QS, C2 = AB- RS.

Then the triads of points A-J52C%, A2B1C2, A2B2C1, A^B^x lie on lines, say
p, q, r, s, forming a quadrilateral pqrs whose three diagonal lines are the
sides
a = BC, b = CA, c — AB

of the triangle ABC. In other words, the quadrangle PQRS and the quadri¬
lateral pqrs have the same diagonal triangle. (Reference 19, pp. 45-46.)

3.2 The Desargues Configuration

A set of m points and n lines in a plane is called a configuration (mc, nd) if

c of the n lines pass through each of the m points while d of the m points lie
on each of the n lines. The four numbers are not independent but evidently
satisfy the equation
cm = dn.

The dual of (mc, nd) is (nd, mc). For instance, (43, 62) is a quadrangle and (62, 43)
is a quadrilateral.
In the case of a self-dual configuration, we have m = n, c = d, and the
symbol (nd, nd) is conveniently abbreviated to nd. For instance, 32 is a triangle.
When the triangles PQR and P' Q'R' of Figure 2.3a are regarded as lying
in distinct planes, we have altogether five planes

a = OQR, 0 = ORP, y - OPQ, 8 = PQR, e = P'Q'R',

 THE COMPLETE PENTAHEDRON 27

their ten lines of intersection

/3 • 7, y • a, a • /?, a • <5, p • 7 • <3, a • £, p • e, 7 • e, 5 • e,

and their ten points of intersection

a • <5 • e, d • P ■ e, y 6 - s, p ■ 7 •£,... , p ■ y d, . . . , a • p • y.

When the whole figure is in one plane, we can use a variant of the same
notation by associating the letters a, p, y, d, e with the digits 1, 2, 3, 4, 5 so as
to call the ten lines

23 = OP, 31 = OQ, 12 = OR, 14 = QR, 24 = RP, 34 = PQ,

15 = Q'R', 25 = R'P', 35 = P'Q', 45 = DE

and the ten points

145 = D, 245 = E, 345 = F, 235 = P’, 315 = Q', 125 - R',

234 = P, 314 = Q, 124 = R, 123 = O.

We now have a self-dual configuration 103: ten points and ten lines, with
three points on each line and three lines through each point. For instance,
the three points on the line 45 are 145, 245, 345, and the three lines through
the point 123 are 23, 31, 12. The rule for incidence is simply that the point ijk
lies on the line ij (and the order of the digits in a symbol is arbitrary: 31 could
just as well be called 13). The self-duality of the configuration can be verified
by interchanging the point and line whose symbols together use up all the
five digits without repetition.

EXERCISES

1. Copy Figure 2.3a, marking the lines with the two-digit symbols.

2. Draw the five lines 12, 23, 34, 45, 51 in one color, and the five lines 14, 42,
25, 53, 31 in another color, thus exhibiting the configuration as a pair of
simple pentagons, “mutually inscribed” in the sense that consecutive
sides of each pass through alternate vertices of the other. [J. T. Graves,
Phil. Mag. (3), 5 (1839), p. 132.]

3. Beginning afresh, draw the four lines 15, 25, 35, 45 in one color, and the
remaining six in another color, thus exhibiting the configuration as a
complete quadrilateral and a complete quadrangle, so situated that each
vertex of the former lies on a side of the latter.

4. Is it possible to draw a configuration 73? [Hint: Look at Axiom 2.17.]

28 THE PRINCIPLE OF DUALITY

3.3 The Invariance of the Harmonic Relation

Dualizing Figure 2.5a, we find that any three concurrent lines a, b, c

determine a fourth line f concurrent with them, which we call the harmonic
conjugate of c with respect to a and b. To construct it, we draw a triangle

qrs (Figure 3.3a) whose vertices q ■ r, q • s, r-s lie on a, b, c, respectively;

then p = {a- s){b ■ r), f= (a- b)(p • q).

In fact, the quadrilateral pqrs has a and b for two of its three diagonal lines
while c andf respectively, pass through the vertices that would be joined by
the third diagonal line.
Figure 3.3b, which is obtained by identifying the lines
p, q, r, s, a, b, c
of Figure 3.3a with the lines
PQ, AB, QR, RP, PS, QS, RS
of Figure 2.5a, shows how the harmonic set of points ABCF arises as a section
of the harmonic set of lines abcf. Since such a figure can be derived from any
harmonic set of points and any point S outside their line,

3.31 A harmonic set ofpoints is projected from any point outside the line by
a harmonic set of lines.
Dually,

3.32 Any section of a harmonic set of lines, by a line not passing through
the point of concurrence, is a harmonic set of points.

Combining these two dual statements, we deduce that

3.33 If ABCF ^ A'B'C'F' and H(AB, CF), then H(A'B', C’Fj.

In other words, perspectivities preserve the harmonic relation. By repeated

application of this principle, we deduce:
 TRILINEAR POLARITY 29

3.34 If ABCF A A'B'C'F' and H(AB, CF), then H{A'B', C'F').

In other words, projectivities preserve the harmonic relation. (In von

Staudt’s treatment, this property serves to define a projectivity. See Reference
7, p. 42.)
By Theorem 1.63 in the form ABCF a FCBA, we can now assert:

3.35 If H{AB, CF) then H(FC, BA),

and therefore also H(CF, BA), H(CF, AB), H(FC, AB).

EXERCISES

1. Let ABC be the diagonal triangle of a quadrangle PQRS. How can PQR
be reconstructed if only ABC and S are given? [Hint: QR is the harmonic
conjugate of AS with respect to AB and AC.]

2. If PQR is a triangle and H(AAX, QR) and H(BBU RP), then P and Q are
harmonic conjugates with respect to

C = AB1 ■ BA^ and Cj = AB ■ AXBX.

3.4 Trilinear Polarity

It is interesting to see how Poncelet used a triangle to induce a correspond¬

ence between points not on its sides and lines not through its vertices. Let
PQR be the triangle, and S a point of general position. The Cevians SP, SQ,
SR determine points A, B, C on the sides QR, RP, PQ, as in Figure 3.4a. Let
D, E, F be the harmonic conjugates of these points A, B, C, so that
H{QR, AD), H(RP, BE), H(PQ, CF).

Comparing Figure 3.4a with Figure 2.5a, we see that these points D, E, F are
the intersections of pairs of corresponding sides of the two triangles PQR and
ABC, namely,

D = QR- BC, E= RP - CA, F = PQ • AB.

Since these two triangles are perspective from S, 2.32 tells us that D, E, Flie
on a line s. This line is what Poncelet called the trilinear polar of S.
Conversely, if we are given the triangle PQR and any line s, not through a
vertex, we can define A, B, C to be the harmonic conjugates of the points
D, E, Fin which s meets the sides QR, RP, PQ. By 2.31, the three lines PA,
QB, RC all pass through a point S, which is the trilinear pole of 5.
30 THE PRINCIPLE OF DUALITY

EXERCISES

1. What happens if we stretch the definitions of trilinear pole and trilinear

polar to the forbidden positions of S and s (namely S on a side, or s through
a vertex)?

2. Locate (In Figure 3.4a) the trilinear pole of s with respect to the triangle
ABC.

3. How should all the lines in this figure be named so as to make it obviously
self-dual?

4. In the spirit of Section 2.5, Exercise 5, what metrical property of the tri¬
angle PQR arises when we seek the trilinear polar of the line at infinity?

3.5 Harmonic Nets

A point P is said to be harmonically related to three distinct collinear points

A, B, C if P can be exhibited as a member of a sequence of points beginning
with A, B, C and proceeding according to this rule: each point (after C)
forms a harmonic set with three previous points. (Any three previous points
can be used, in any order.) The set of all points harmonically related to A, B,
C is called a harmonic net [or “net of rationality” (see Reference 19, p. 84)]
and is denoted by R(TRC). Of course, the same harmonic net is equally well
denoted by R(BAC) or R(BCA), and so forth. Since a projectivity transforms
any harmonic set into a harmonic set, it also transforms any harmonic net
 HARMONIC NETS 31

into a harmonic net. In particular,

3.51 If a projectivity leaves invariant each of three distinct points A, B, Con

a line, it leaves invariant every point of the harmonic net R(ABC).

This result, which we have deduced from our first seven axioms, may reason¬
ably be regarded as making Axiom 2.18 plausible.
By Theorem 1.63 (with B and D interchanged), any four collinear points
A, B, C, D satisfy ABCD DCBA. If D belongs to K(ABC), A must belong to
R(DCB), which is the same as R(BCD). It follows that not only A, but every
point of R(ABC), belongs also to R(BCD). Hence, if D belongs to R(ABC),
R (ABC) = R (BCD).
By repeated applications of this result we see that, if K, L, M are any three
distinct points of R(ABC),
R (ABC) = R(BCK) = R (CKL) = R (KLM).
Hence

3.52 A harmonic net is equally well determined by any three distinct points
belonging to it.
We could, in fact, define a harmonic net to be the smallest set of more than
two collinear points which includes, with every three of its members, the
harmonic conjugate of each with respect to the other two.
If we begin to make a careful drawing of a harmonic net, we soon find that
this is an endless task: the harmonic net seems to include infinitely many
points between any two given points, thus raising the important question
whether it includes every point on the line. There is some advantage
in leaving this question open, that is, answering “Yes and No.” Projective
geometry, as we are developing it here, is not categorical: it is really
not just one geometry but many geometries, depending on the nature
of the set of all points on a line. In rational geometry and the simplest
finite geometries, all the points on a line form a single harmonic net, so that
Axiom 2.18 becomes superfluous. In real geometry, on the other hand, the
points on a line are arranged like the continuum of real numbers, among
which the rational numbers represent a typical harmonic net (Reference 7,
p. 179). Between any two rational numbers, no matter how close, we can find
infinitely many other rational numbers and also infinitely many irrational
numbers. Analogously, between any two points of a harmonic net on the
real line we can find infinitely many other members of the net and also
infinitely many points not belonging to the harmonic net. Axiom 2.18 (or,
alternatively, some statement about continuity) is needed to ensure the
invariance of these extra points.
When the points of a line in rational geometry, or the points of a harmonic
32 THE PRINCIPLE OF DUALITY

•net in real geometry, are represented by the rational numbers, the natural
numbers (that is, positive integers) represent a harmonic sequence ABCD . . . ,
which is derived from three collinear points A, B, M by the following special
procedure. With the help of two arbitrary points P and Q on another line
through M, as in Figure 3.5a, we construct in turn,

A' = AP • BQ, B' — BP • A'M,

C — B'Q • AM, C' = CP • A'M,
D = C'Q • AM, D' = DP- A'M,
and so on. In view of the harmonic relations

H {BM, AC), H (CM, BD), ...,

the sequence ABC. . . , so constructed, depends only on the given points

A, B, M, and is independent of our choice of the auxiliary points P and Q.

EXERCISES

1. Is the harmonic sequence ABC. . . uniquely determined by its first three

members?

2. In the notation of Figure 3.5a, is A'B'C'... a harmonic sequence?

3. What happens to the sequence ABCD . . . (Figure 3.5a) when PQ is the

line at infinity, so that ABB'A', BCC'B’, CDD’C',..., are parallelograms?
 CHAPTER FOUR

The Fundamental Theorem

and Pappus’s Theorem

The Golden Age of Greek geometry ended with the time of Apol¬
lonius of Perga. . . . The beginning of the Christian era sees quite a
different state of things. . . . Production was limited to elementary
textbooks of decidedly feeble quality. . . . The study of higher ge¬
ometry languished or was completely in abeyance until Pappus arose
to revive interest in the subject. . . . The great task which he set
himself was the re-establishment of geometry on its former high
plane of achievement.
Sir Thomas Heath (1861-1940)
(Reference 11, p. 355)

4.1 How Three Pairs Determine a Projectivity

Given four distinct points A, B, C, X on one line, and three distinct points
A', B', C' on the same or another line, there are many possible ways in which
we may proceed to construct a point X' (on A'B') such that
ABCX a A'B'C'X'.
For instance, if the two lines are distinct, one way is indicated in Figure 4.1a
(which is Figure 1.6b in a revised notation):

(4.11) ABCX ^ GNMQ 4 A'B'C'X'.

This can be varied by using B' and B (or C' and C), instead of A' and A, as
centers of the two perspectivities. If, on the other hand, the given points are
33
34 THE FUNDAMENTAL THEOREM AND PAPPUS’S THEOREM

all on one line, as in Figure 4.1b, we can use an arbitrary perspectivity

ABCX tv A1B1CxX1 to obtain four points on another line, and then relate
A1B1C1X1 to A'B'C'X', so that altogether

ABCX % AxBxCxXx ^ GNMQ ^ A'B'C'X'.

We saw, in Figure 2.4b, that five collinear points A, B, C, D,E determine a

unique sixth point F such that Q(ABC, DEF). By declaring F to be unique,
we mean that its position is independent of our choice of the auxiliary triangle
QRS. Can we say, analogously, that seven points A, B, C, X, A', B', C’
(with the first four, and likewise the last three, collinear and distincs) deter¬
mine a unique eighth point X' such that

ABCX a A'B'C’X'?
If not, there must be two distinct chains of perspectivities yielding, respec¬
tively,
ABCX a A'B'C'X' and ABCX a A’B'C’X",
where X" # X'. Proceeding backwards along the first chain and then forwards
along the second, we would have
A'B'C'X' a A’B’C'X",
contradicting Axiom 2.18. Thus, by reductio ad absurdum, we have proved
4.12 The Fundamental Theorem of Projective Geometry: A projec-
tivity is determined when three collinear points and the corresponding three
collinear points are given.

Of course, either of the sets of “three collinear points” (or both) can be
replaced by “three concurrent lines.” Thus each of the relations
ABC a A'B'C', ABC a abc, abc p ABC, abc p ab'c’
 THE FUNDAMENTAL THEOREM 35

suffices to specify uniquely a particular projectivity. On the other hand, each

of the relations

ABCD a A B C'D', ABCD a abed, abed a a'b'c’d'

expresses a special property of eight points, or of four points and four lines,
or of eight lines, of such a nature that any seven of the eight will uniquely
determine the remaining one. We now have the proper background for
Theorem 1.63, which tells us that, if a projectivity interchanges A and B while
transforming C into D, it also transforms D into C, that is, it interchanges
C and D. 6

EXERCISES

1. Given three collinear points A, B, C, set up three perspectivities whose

product has the effect ABC a BCA.

2. If the projectivity ABC a BCA transforms D into E, and E into F, how

does it affect FI [Hint: Use Axiom 2.18.]

3. If A, B, C, D are distinct collinear points, then

ABCD a BADC a CDAB a DCBA.

4.2 Some Special Projectivities

The following simple consequences of 4.12 will be found useful.

4.21 Any two harmonic sets {of four collinear points or four concurrent
lines) are related by a unique projectivity.
Proof. If H{AB, CF) and V\{A'B', C'Fj, the projectivity ABC a A'B'C'
transforms F into a point F" such that, by 3.34, H(A'B', CF"). But the
harmonic conjugate of C' with respect to A’ and B’ is unique. Therefore
F" coincides with F'. The same reasoning can be used when either of the
harmonic sets consists of lines instead of points.

4.22 A projectivity relating ranges on two distinct lines is a perspectivity

if and only if the common point of the two lines is invariant.
Proof. A perspectivity obviously leaves invariant the common point
of the two lines. On the other hand, if a projectivity relating ranges on two
distinct lines has an invariant point E, this point, belonging to both ranges,
must be the common point of the two lines, as in Figure 4.2a. Let A and B
36 THE FUNDAMENTAL THEOREM AND PAPPUS’S THEOREM

be two other points of the first range, A' and B' the corresponding points
of the second. The fundamental theorem tells us that the perspectivity

ABE ^ A'B'E,
where O = AA' • BB', is the same as the given projectivity ABE a A'B'E.

Figure 4.2a

EXERCISES

1. Any two harmonic nets, R(ABC) and R(A'B'C) (or any two harmonic
sequences, ABC. . . and ABC'...) are related by a projectivity.

2. Let the side QR of a quadrangle PQRS meet the side BC of its diagonal
triangle ABC in Ax, as in Exercise 2 of Section 3.1. Regarding ABPCSA1 as
a hexagon whose six vertices lie alternately on two lines, w'hat can be said
about the intersections of pairs of “opposite” sides of this hexagon?

3. Dualize Theorem 4.22 and Figure 4.2a.

4. If H(AB, CD) then ABCD j. BACD. (Compare Section 4.1, Exercise 3.)

4.3 The Axis of a Projectivity

The fundamental theorem 4.12 tells us that there is only one projectivity
ABC a ABC relating three distinct points on one line to three distinct
points on the same or any other line. The construction indicated in 4.11
(see Figure 4.1a) shows how, when the lines AB and A'B' are distinct, this
unique projectivity can be expressed as the product of two perspectivities
whose centers may be any pair of corresponding points (in reversed order) of
the two related ranges. It is natural to ask whether different choices of the
 THE AXIS OF A PROJECTIVITY 37

two centers will yield different positions for the.line MN of the intermediate
range. A negative answer is provided by the following theorem:
4.31 Every projectivity relating ranges on two distinct lines determines
another special line, the “axis,” which contains the intersection of the cross¬
joins of any two pairs of corresponding points.
Proof. In the notation of 4.11, the perspectivities from A' and A
determine the axis MN, which contains the common point of the “cross¬
joins of the pair A A and any other pair; for instance, N is the common
point of the cross-joins (AB' and BAj of the two pairs A A' and BB'. What
remains to be proved is the uniqueness of this axis: that another choice of
the two pairs of corresponding points (such as BB' and CCj will yield
another crossing point” on the same axis. For this purpose, we seek a
new specification for the axis, independent of the particular pair AA'.

Figure 4.3a Figure 4.3b

Let E be the common point of the two lines. Suppose first that E is
invariant, as in Figures 4.2a and 4.3a. Referring to Figure 4.1a, we observe
that, when X coincides with E, so also do Q and X'. The axis EN is inde¬
pendent of AA', since it can be described as the harmonic conjugate of EO
with respect to the given lines EB and EB'. On the other hand, if the common
point E is noninvariant, as in Figure 4.3b, it is still a point belonging to
both ranges. Referring to Figure 4.1a again, we observe that, when X
coincides with E, Q and X' both coincide with the corresponding point
E'. Hence the axis passes through E'. For a similar reason the axis also
passes through the point E0 of the first range for which the corresponding
point of the second is E. Hence the axis can be described as the join E0E'.

This completes the proof of 4.31, which Forder (Reference 10, p. 72) has
named “the criss-cross theorem.” The final remark can be expressed as
follows:
4.32 If AA'E is a triangle, the axis of the projectivity AE0E a A'EE' is
the line E0E'.
38 THE FUNDAMENTAL THEOREM AND PAPPUS’S THEOREM

EXERCISE

Dualize Theorems 4.31, 4.32 and Figure 4.3b.

4.4 The Pappus Configuration

We are now ready to prove the earliest of all projective theorems (which was
originally proved by a laborious application of Euclid’s methods):

4.41 Pappus’s Theorem : If the six vertices of a plane hexagon lie alternately
on two lines, the three pairs of opposite sides meet in collinear points.
Proof. Let the hexagon be AB'CA'BC', as in Figure 4.4a (which
shows two of the many different ways in which it can be drawn). Since
alternate vertices are collinear, there is a projectivity ABC a A'B'C'. The
pairs of opposite sides, namely

B'C, BC'; C'A, AC'; A'B, AB',

are just the cross-joins of the pairs of corresponding points

BB', CC; CC', AA'; A A’, BB'.

By Theorem 4.31, their points of intersection

L = B'C ■ BC', M — C'A • CA', N = A'B- AB'

all lie on the axis of the projectivity.

It is important to notice the essential steps in this deduction: Axiom 2.18,

the fundamental theorem 4.12, the criss-cross theorem 4.31, and finally
Pappus’s theorem 4.41. An alternative procedure is to take Pappus’s theorem
as an axiom (instead of Axiom 2.18) and then reverse all these steps. Since
Pappus’s theorem implies Desargues’s theorem (in the plane),* this procedure
has the advantage of allowing us to develop a purely two-dimensional theory
without adopting 2.32 as an axiom. It has the disadvantage that we instinc¬
tively rebel against accepting, as an axiom, such a surprising result as Pappus’s
theorem. Moreover, the deduction of 4.31 from 4.41 (Reference 1, pp. 62-64)
is more complicated than anything that is done in this book.
Yet another procedure is to retain our original Axioms 2.11 through 2.18
and to postpone all the material in Chapter 3 until Pappus’s theorem has been
deduced from the fundamental theorem. This has the advantage of allowing

* See Reference 8, p. 238; also A. Cronheim, Proc. Am. Math. Soc., 4 (1953), pp. 219—
221. The details are really not as formidable as they look, but Cronheim complicates them
by writing A + B for AB, and (A + B)(C + D) for AB ■ CD.
 pappus 39

us to omit 2.32 from the list of two-dimensional axioms proposed in Section

3.1. (The list is thus seen to be redundant: only six of the seven are really
needed.) It has the disadvantage of making Theorem 3.51 depend on Axiom
2.18, so that the theorem can no longer be used to make the axiom plausible!

Figure 4.4a

We saw, in Section 3.2, that the figure for Desargues’s theorem is a self¬
dual configuration 103. Somewhat analogously, the figure for Pappus’s
theorem is a self-dual configuration 93: nine points and nine lines, with three
points on each line and three lines through each point. This fact becomes
evident as soon as we have made the notation more symmetrical by calling
the nine points

A1 = A, B1 — B, Q = C, A2 = A', B2 = B', C2 = C',

A3 = L, B3 — M, C3 — N
and the nine lines

ax = BL, b1 = AM, cx = A' B', a2 = CM, b2 — CL, c2 = AB,

a3 = AN, b3 = BN, c3 = LM,

as in Figure 4.4b. The shaded triangles, namely

A1B1C2, A2B2C3, A3B3C1 or a-JO]C2, a3b3c2, a2b2c3,

provide an instance of Graves triangles: a cycle of three triangles, each in¬

scribed in the next. (See page 130 of Graves’ paper mentioned in Exercise 2 of
Section 3.2. This aspect of the Pappus configuration was rediscovered by
G. Hessenberg. See also Reference 7, p. 75, Exercise 2.)
40 THE FUNDAMENTAL THEOREM AND PAPPUS’S THEOREM

EXERCISES

1. Using the “symmetrical” notation for the Pappus configuration, prepare

a table showing which are the 3 points on each line and which are the 3 lines
through each point. Observe that the 3 points At, Bx, Ck are collinear (and
the 3 lines at, bp ck, are concurrent) whenever i + j + k is a multiple of 3.
(F. W. Levi.*)

2. Given a triangle AXA2A3 and two points Bx, B2, locate a point B3 such that
the three lines AXBX, A2B3, A3B2 are concurrent while also the three lines
AxB3, A2B2, A3Bx are concurrent. Then also the three lines AXB2, A2BX, A3B3
are concurrent; in other words, if two triangles are doubly perspective, they
are triply perspective. (Reference 19, p. 100.)

3. What can be said about the three “diagonals” of the hexagon

AxB3A2B2A3Bx?

4. State the dual of Pappus’s theorem and name the sides of the hexagon in
the notation of Figure 4.4b.

5. Consider your solution to Exercise 2 of Section 4.2. Could this be developed

into a proof of Pappus’s theorem ?

6. If one triangle is inscribed in another, any point on a side of the former

can be used as a vertex of a third triangle which completes a cycle of
Graves triangles (each inscribed in the next).

7. Assign the digits 0, 1, . . . , 8 to the nine points of the Pappus configuration

in such a way that 801, 234, 567 are three triads of collinear points while
012, 345, 678 is a cycle of Graves triangles. (E. S. Bainbridge.+)

* Geometrische Konfigurationen (Hirzel: Leipzig, 1929), p. 108.

t In his answer to an examination question.
 CHAPTER FIVE

One-Dimensional Projectivities

The most original and profound of the projective geometers of the

German school was Georg Karl Christian von Staudt, long professor
at Erlangen. His great passion . . . was for unity of method.
J. L. Coolidge (1873-1954)
(Reference 4, p. 61)

5.1 Superposed Ranges

Axiom 2.18 tells us that a projectivity relating two ranges on one line (that
is, a projective transformation of the line into itself) cannot have more than
two invariant points without being merely the identity, which relates each
point to itself. The projectivity is said to be elliptic, parabolic, or hyperbolic
according as the number of invariant points is 0, 1, or 2. We shall see that
both parabolic and hyperbolic projectivities always exist. In fact, Figure 5.1a
(which is essentially the same as Figure 2.4a) suggests a simple construction for
the hyperbolic or parabolic projectivity AEC a BDC which has a given
invariant point C. Here A, B, C, D, E are any five collinear points, and we
draw a quadrangle PQRS as if we were looking for the sixth point of a quad¬
rangular set. The given projectivity can be expressed as the product of two
perspeotivities
AEC £ SBC % BDC,

and it is easy to see what happens to any other point on the line.
41
42 ONE-DIMENSIONAL PROJECTIVITIES

Evidently C (on RS) is invariant. If any other point is invariant, it must be

collinear with the centers P and Q of the two perspectivities; that is, it can
only be F. Hence the projectivity AEC a BDC is hyperbolic if C and F are
distinct (Figure 5.1a) and parabolic if they are coincident (Figure 5.1b). In
the former case we have AECF a BDCF, and in the latter we naturally write

AECC a BDCC,

the repeated letter indicating that the projectivity is parabolic. Thus

5.11 The two statements AECF a BDCF and Q(A BC, DEF) are equiva¬
lent, not only when C and F are distinct but also when they coincide.

D A

Figure 5.1a Figure 5.1b

Since the statement AECF a BDCF involves C and F symmetrically, the

statement Q(ABC, DEF) is equivalent to Q(ABF, DEC), and similarly to
Q{AEF, DBC) and to
Q(DEF, ABC).

This is remarkable because, when the quadrangular set is derived from the
quadrangle, the two triads in the symbol arise differently: the first from three
sides with a common vertex, and the second from three that form a triangle.
It is interesting that, whereas one way of matching two quadrangles (Figure
2.4b) uses only Desargues’s theorem, the other (Exercise 2, below) needs the
fundamental theorem.
Our use of the words elliptic, parabolic and hyperbolic arises from the
aspect of the projective plane as a Euclidean plane with an extra line at infinity.
In this aspect a conic is an ellipse, a parabola, or a hyperbola according as its
number of points at infinity is 0, 1, or 2, that is, according as it “goes off to
 PARABOLIC PROJECTIVITIES 43

infinity” not at all, or in one direction (the direction of the axis of the
parabola) or in two directions (the directions of the asymptotes of the
hyperbola). The names of the conics themselves are due to Apollonius
(see Reference 13, p. 10).

EXERCISES

1. Using Figure 5.1a again (and defining V = PS ■ QR), construct a hyper¬

bolic projectivity having A and D as its invariant points.
2. Take any five collinear points A, B, C, D, E. Construct F so that
Q(ABC, DBF). Then (on the other side of the line, for convenience) con¬
struct C' so that Q(DEF, ABC'). See how nearly your C' agrees with C.
3. Two perspectivities cannot suffice for the construction of an elliptic
projectivity. In other words, if an elliptic projectivity exists, its construction
requires three perspectivities.
4. Does an elliptic projectivity exist?

5.2 Parabolic Projectivities

The fundamental theorem, 4.12, shows that a hyperbolic projectivity is

determined when both its invariant points and one pair of distinct corre¬
sponding points are given. In fact, any four collinear points A, B, C, F
determine such a projectivity ACF a BCF, with invariant points C and F.
To construct it, we take a triangle QPS whose sides PS, SQ, QP pass,
respectively, through A, B, F. If the side through F meets CS in U (as in
Figure 5.1a), we have
ACF ^ SCU % BCF.

If we regard E as an arbitrary point on the same line AB, this construction

yields the corresponding point D. It remains effective when C, F and U
coincide (as in Figure 5.1b), that is, when the line AB passes through the
diagonal point U = PQ ■ RS of the quadrangle. Thus a parabolic projectivity
is determined when its single invariant point and one pair of distinct corre¬
sponding points are given. We naturally call it the projectivity ACC a BCC.
This notation is justified by its transitivity:
5.21 The product of two parabolic projectivities having the same invariant
point is another such parabolic projectivity (if it is not merely the identity).
Proof. Clearly, the common invariant point C of the two projectiv¬
ities is still invariant for the product, which is therefore either parabolic or
hyperbolic. The latter possibility is excluded by the following argument.
If any other point A were invariant for the product, the first parabolic
44 ONE-DIMENSIONAL PROJECTIVITIES

projectivity would take A to some different point B, and the second would
take B back to A. Thus the first would be ACC a BCC, the second would
be its inverse BCC a ACC, and the product would not be properly
hyperbolic but merely the identity.

Figure 5.2a

Thus the product of ACC a A'CC and A'CC a A"CC is ACC a A'CC,
and we can safely write out strings of parabolic relations such as
ABCC a A'B'CC a A"B"CC.
In particular, by “iterating” a parabolic projectivity ACC a A'CC we obtain
a sequence of points A, A', A", . . . such that
CCAA'A" ... a CCA'A" A"'
as in Figure 5.2a. Comparing this with Figure 3.5a, we see that A A’ A" ... is a
harmonic sequence.
We have seen that the statements
AECF a BDCF and Q(ABC, DEF)
are equivalent. Setting B = E and C = F, we deduce the equivalence of
ABCC a BDCC and H(BC, AD).
Hence, after a slight change of notation.

5.22 The projectivity AA'C a A’A"C is parabolic if H(A'C, AA'j, and

hyperbolic otherwise.
In other words, the parabolic projectivity ACC a A'CC transforms A'
into the harmonic conjugate of A with respect to A' and C.

EXERCISE

What happens to the parabolic projectivity ACC a A'CC (Figure 5.2a)

when PQ is the line at infinity (as in Exercise 3 of Section 3.5)?
 INVOLUTIONS 45

5.3 Involutions

Desargues’s treatise of 1639, which included his famous theorem and many
other projective ideas, was not well received during his lifetime. This lack of
appreciation was possibly a result of his obscure style; he introduced about
seventy new terms, of which only involution has survived. According to his
definition, formulated in terms of the nonprojective concept of distance and
the arithmetic concept of multiplication, an “involution” is the relation
between pairs of points on a line whose distances from a fixed point have a
constant product (positive or negative). He might well have added “or a con¬
stant sum.” An equivalent definition, not using distances, was given by von
Staudt: An involution is aprojectivity of period two, that is, a projectivity which
interchanges pairs of points. It is remarkable that this relation
XX' a X'X
holds for all positions of X if it holds for any one position:

5.31 Any projectivity that interchanges two distinct points is an involution.

Proof. Let X a X' be the given projectivity which interchanges two
distinct points A and A', so that

AA'X a A'AX',
where X is an arbitrary point on the line AA'. By Theorem 1.63, there is a
projectivity in which AA'XX' a A'AX'X. By the fundamental theorem
4.12, this projectivity, which interchanges X and X', is the same as the
given projectivity. Since X was arbitrarily chosen, the given projectivity is
an involution.

Any four collinear points A, A', B, B' determine a projectivity AA'B a

A'AB', which we now know to be an involution. Hence
5.32 An involution is determined by any two of its pairs.
Accordingly, it is convenient to denote the involution AA'B a A'AB' by
(AAj(BB')
or (A'A)(BBj, or (BBj(AA'), and so forth. This notation remains valid when
B' coincides with B; in other words, the involution AA'B a A'AB, for which
B is invariant, may be denoted by
(AAj(BB).
If Q(ABC, DEF), as in Figure 2.4a, we can combine the projectivity
AECF a BDCF of 5.11 with the involution (BD)(CF) to obtain
AECF A BDCF a DBFC,
46 ONE-DIMENSIONAL PROJECTIVITIES

which shows that there is a projectivity in which AECF p DBFC. Since this
interchanges C and F, it is an involution, namely

(.BE)(CF) or (CF)(AD) or (A D)(BE).

Thus the quadrangular relation Q(ABC, DEF) is equivalent to the statement
that the projectivity ABC a DEF is an involution, or that

ABC DEF a DEF ABC.

In other words,

5.33 The three pairs of opposite sides of a complete quadrangle meet any
line (inot through a vertex) in three pairs of an involution. Conversely, any
three collinear points, along with their mates in an involution, form a quad¬
rangular set.
It follows that the construction for F, when A, B, C, D, E are given (as in
the preamble to 2.41), may be regarded as a construction for the mate of C in
the involution (AD)(BE). (See Figure 2.4a or 5.1a.)
We have seen that CF is a pair of the involution (AD)(BE) if and only if
AECF a BDCF. We must get accustomed to using other letters in the same
context. For instance, MN is a pair of the involution (AB')(BA') if and only if
AA'MN a BB'MN. Since (AB')(BA') is the same as (AB')(A'B), it follows
that the two statements
AA’MN a BB'MN and ABMN p A’B’MN
are equivalent. (Notice that it is only the statements that are equivalent: the
two projectivities are, of course, distinct.)
If two involutions, (AA')(BB') and (AA^BBf, have a common pair MN,
we deduce
A’B'MN a BAMN p A^MN.
Hence

5.34 If MN is a pair of each of the involutions (AA')(BBr) and (AAflBBA,

it is also a pair of (A'Bj)(B'Mx).

All these results remain valid when M and Ncoincide, so that we are dealing
with parabolic (instead of hyperbolic) projectivities. Thus M is an invariant
point of the involution (AB')(BA’) if and only if A A'MM p BB'MM, that is,
if and only if ABMM p A!B'MM; and if M is an invariant point of each of
the involutions (AA')(BB') and (AAAiBBf, it is also an invariant point of
(A'BJiB'AJ.
If two involutions have a common pair MN, their product is evidently
hyperbolic, with invariant points M and N. In fact, by watching their effect
on A, M,N in turn, we see that the product of (AB){MN) and (BC)(MN) is
AMN a CMN. More interestingly,
 THE PRODUCT OF TWO INVOLUTIONS 47

5.35 Any one-dimensionalprojectivity is expressible as the product of two

involutions.
Proof. Let the given projectivity be ABC a A'B'C', where neither
A nor B is invariant. By watching what happens to A, B, C in turn, we see
that this projectivity has the same effect as the product of the two involutions

(AB')(BA') and (A'B')(C'D),

where D is the mate of C in (ABj(BAj. (Reference 10, p. 72.)

EXERCISES

1. Given six collinear points A, B, C, D, E, F, consider the three involutions

(AB)(DE), (BC)(EF), (CD){FA). If any two of these involutions have a
common pair, all three have a common pair.

2. The hyperbolic projectivity MNA a MNA' is the product of the involu¬

tions (AB)(MN) and (A'B)(MN), where B is an arbitrary point on the line.

3. Any involution (AAj(BB') may be expressed as the product of (AB)(A'B')

and (ABj(BAj.

4. Any projectivity that is not an involution may be expressed as the product

of (AA")(A'Aj and (AA'")(A'A").

5. Notice that Exercises 3 and 4 together provide an alternative proof for

Theorem 5.35. In the proof given in the text, why was it necessary to
insist that neither A nor B is invariant?

5.4 Hyperbolic Involutions

As we have seen, any involution that has an invariant point B (and a pair
of distinct corresponding points C and Cj may be expressed as BCC a BCC
or (.BB){CC’). Let A denote the harmonic conjugate of B with respect to C
and C. By 4.21, the two harmonic sets ABCC and ABC'C are related by a
unique projectivity ABCC' a ABC'C. The fundamental theorem identifies
this with the given involution. Hence

5.41 Any involution that has an invariant point B has another invariant
point A, which is the harmonic conjugate of B with respect to any pair of
distinct corresponding points.
Thus any involution that is not elliptic is hyperbolic: there are no “parabolic
involutions.” Moreover, any two distinct points A and B are the invariant
48 ONE-DIMENSIONAL PROJECTIVITIES

points of a unique hyperbolic involution, which is simply the correspondence

between harmonic conjugates with respect to A and B. This is naturally
denoted by
(AAxm-
The harmonic conjugate of C with respect to any two distinct points A
and B may now be redefined as the mate of C in the involution (AA)(BB).
Unlike the definition of harmonic conjugate in Section 2.5, this new definition
remains meaningful when C coincides with A or B:

5.42 Any point is its own harmonic conjugate with respect to itself and any
other point.

EXERCISES

1. If ABCD a BACD then H{AB, CD). (Compare Section 4.2, Exercise 4.)

2. If a hyperbolic projectivity relates two points that are harmonic conjugates

with respect to the invariant points, it must be an involution.

3. If H(AB, MN) and H(A'B', MN), then MN is a pair of the involution

(.AA')(BBj. [Hint: AB'MN a BA'MN.]

4. Given Q{ABC, DEF), let A', B', C', D', E', F' be the harmonic conjugates
of A, B, C, D, E, Fwith respect to two fixed points on the same line. Then
Q(A'B'C', D'E'Fj.
5. If ABCD a ABDE and H(CE, DDj, then H(AB, DDj. (S. Schuster.)

6. Let Xbe a variable point collinear with three distinct points A, B, C, and
let Y and X'be defined by H(AB, XY), H{BC, YXj. Then the projectivity
X a is parabolic. [Hint: When X is invariant, so that X' coincides
with it, Ytoo must coincide with it; for otherwise both A and C would be
the harmonic conjugate of B with respect to X and T. Therefore X must
be either A or B, and also either B or C; that is, X = B.]

7. If H{BC, AD) and H(CA, BE) and H(AB, CF), then Q{ABC, DEF).
[Hint: Apply 5.41 to the involution BCAD 7; ACBE. Deduce Y\{DE, CF).]
 CHAPTER SIX

Two-Dimensional Projectivities

History shows that those heads of empires who have encouraged

the cultivation of mathematics . . . are also those whose reigns have
been the most brilliant and whose glory is the most durable.
Michel Chasles (1793-1880)
(Reference 3, p. 163)

6.1 Projective Collineations

We shall find that the one-dimensional projectivity ABC a A'B'C' has two
different analogues in two dimensions: one relating points to points and lines
to lines, the other relating points to lines and lines to points. The names
collineation and correlation were introduced by Mdbius in 1827, but some
special collineations (such as translations, rotations, reflections, and dilata¬
tions) were considered much earlier. Another example is Poncelet’s “ho¬
mology”: the relation between the central projections of a plane figure
onto another plane from two different centers of perspective. We shall give
(in Section 6.2) a two-dimensional treatment of the same idea.
By a point-to-point transformation X -> X' we mean a rule for associating
every point X with every point X' so that there is exactly one X' for each X
and exactly one X for each X'. A line-to-line transformation x—>-x' is
defined similarly. A collineation is a point-to-point and line-to-line trans¬
formation that preserves the relation of incidence. Thus it transforms ranges
into ranges, pencils into pencils, quadrangles into quadrangles, and so on.
Clearly, it is a self-dual concept, the inverse of a collineation is a collineation,
and the product of two collineations is again a collineation.
49
50 TWO-DIMENSIONAL PROJECTIVITIES

A projective collineation is a collineation that transforms every one-dimen¬

sional form (range or pencil) projectively, so that, if it transforms the points Y
on a line b into the points Y' on the corresponding line b', the relation between
Y and Y' is a projectivity Y a Y'. The following remarkable theorem is
reminiscent of 5.31:

6.11 Any collineation that transforms one range projectively is a projective

collineation.

Figure 6.1a

Proof. Let a and a' be the corresponding lines that carry the pro¬
jectively related ranges. We wish to establish the same kind of relationship
between any other pair of corresponding lines, say b and b' (Figure 6.1a).
Let Tbe a variable point on b, and O a fixed point on neither a nor b. Let
O Y meet a in X. The given collineation transforms O into a fixed point O'
(on neither a! nor bj, and O Y into a line O' Y' that meets a! in X'. Since X
is on the special line a, we have X a X'. Thus

O O'
YkXkX'kY':

the collineation induces a projectivity Y a Y' between b and b', as desired.

To obtain the dual result (with “range” replaced by “pencil”) we merely

have to regard the ranges Y and Y' as sections of corresponding pencils.
Axiom 2.18 yields the following two-dimensional analogue:

6.12 The only projective collineation that leaves invariant 4 lines forming
a quadrilateral, or 4 points forming a quadrangle, is the identity.
Proof. Suppose the sides of a quadrilateral are 4 invariant lines.
Then the vertices (where the sides intersect in pairs) are 6 invariant points,
3 on each side. Since the relation between corresponding sides is projective,
every point on each side is invariant. Any other line contains invariant
points where it meets the sides and is consequently invariant. Thus the
collineation must be the identity. The dual argument gives the same result
when there is an invariant quadrangle.
 PROJECTIVE COLLINEATIONS 51

The analogue of the fundamental theorem 4.12 is as follows:

6.13 Given any two complete quadrilaterals (or quadrangles), with their
four sides (or vertices) named in a corresponding order, there is just one pro¬
jective collineation that will transform the first into the second.

Proof. Let DEFPQR and D'E'F'P'Q'R' be the two given quadri¬

laterals, as in Figure 6.IB. We proceed to investigate the effect that such a
collineation should have on an arbitrary line a. There are certainly two
sides of the first quadrilateral that meet a in two distinct points. For
definiteness, suppose a is XY, with Xon DEand Ton DQ. The projectivities
DEF a D'E'F' and DQR a D'Q'R' determine a line a' — X'Y', where
DEFX a D'E'F'X' and DQRY a D'Q'R'Y'.
To prove that this correspondence a-+ a' is a collineation, we have to
verify that it also relates points to points in such a way that incidences are
preserved. For this purpose, let a vary in a pencil, so that X 7^ Y. By our
construction for a', we now have
X' * X% Y * Y\

Since D is the invariant point of the perspectivity X ^ Y, D' must be an

invariant point of the projectivity X' a Y'. Hence, by 4.22, this projectivity
52 TWO-DIMENSIONAL PROJECTIVITIES

is again a perspectivity. Thus a , like a, varies in a pencil: that is, concurrent

lines yield concurrent lines. We have not only a line-to-line transformation
but also a point-to-point transformation, preserving incidences, namely,
a collineation. The projectivity X a X' suffices to make it a projective
collineation.
Finally, there is no other projective collineation transforming DEFPQR
into D'E'F'P'Q'R'; for, if another transformed a into ax, the inverse of the
latter would take a1 to a, the original collineation takes a to a , and al¬
together we would have a projective collineation leaving D'E'F'P' Q'R'
invariant and taking ax to a . By 6.12, this combined collineation can only
be the identity. Thus, for every a, ax coincides with a : the “other” collinea¬
tion is really the old one over again. In other words, the projective
collineation a -> a is unique.

In the statement of the theorem, we used the phrase “named in a corre¬

sponding order.” This was necessary, because otherwise we could have
permuted the sides of one of the quadrilaterals in any one of 4! = 24 ways,
obtaining not just one collineation but 24 collineations.
We happened to use quadrilaterals, but the dual argument would immedi¬
ately yield the same result for quadrangles.

EXERCISE

Let PQRS -^P'Q'R'S' denote the projective collineation that relates two
given quadrangles PQRS and P' Q'R'S'; for instance, PQRS —*■ PQRS is
the identity. Describe the special collineations

(i) PQRS —► QPRS, (ii) PQRS -> QRPS, (iii) PQRS -* QRSP.

Hints:
(i) What happens to the lines PQ and RSI
(ii) What does Figure 3.4a tell us about the possibility of an invariant
line?
(iii) What happens to the lines PR, QS, and to the points QR • PS, PQ • RSI

6.2 Perspective Collineations

In Section 2.3, we obtained the Desargues configuration, Figure 2.3a, by

taking two triangles PQR and P’Q'R', perspective from O. By 6.13, there is
just one projective collineation that transforms the quadrangle DEPQ into
DEP'Q'. This collineation, transforming the line o = DE into itself, and PQ
 PROJECTIVE COLLINEATIONS 53

into P’Q', leaves invariant the point o • PQ = F = o • P'Q'. By Axiom 2.18,

it leaves invariant every point on o. The join of any two distinct correspond-
ing points meets o in an invariant point, and is therefore an invariant line.
The two invariant lines PP' and QQ' meet in an invariant point, namely O.

The point R = DQ • EP is transformed into DQ' • EP' — R'. By the dual of

Axiom 2.18, every line through O is invariant.
This collineation, relating two perspective triangles, is naturally called a
perspective collineation. The point O and line o, from which the triangles are
perspective, are the center and axis of the perspective collineation. If O and o
are nonincident, as in Figure 2.3a, the collineation is a homology (so named by
Poncelet). If O and o are incident, as in Figure 6.2a, it is an elation (so named
by the Norwegian geometer Sophus Lie, 1842-1899). To sum up,

6.21 Any two perspective triangles are related by a perspective collineation,

namely an elation or a homology according as the center and axis are or are
not incident.
These ideas are further elucidated in the following six theorems.

6.22 A homology is determined when its center and axis and one pair of
corresponding points (collinear with the center) are given.
Proof. Let O be the center, o the axis, P and P' (collinear with O)
the given corresponding points. We proceed to set up a construction where¬
by each point R yields a definite corresponding point R'. If R coincides
with O or lies on o, it is, of course, invariant, that is, R' coincides with R.
If, as in Figure 2.3a, R is neither on o nor on OP, we have
R' = EP’ ■ OR, where E = o ■ PR.
Finally, if R is on OP, as in Figure 6.2b, we use an auxiliary pair of points
Q, Q' (of which the former is arbitrary while the latter is derived from it
the way we just now derived Rr from R) to obtain
R' = DQ' • OP, where D = o • QR.
54 TWO-DIMENSIONAL PROJECTIVITIES

Figure 6.2c

6.23 An elation is determined when its axis and one pair of corresponding
points are given.
Proof. Let o be the axis, P and P' the given pair. Since the collinea-
tion is known to be an elation, its center is o • PP'. We proceed as in the
proof of 6.22, using Figures 6.2a and 6.2c instead of 2.3a and 6.2b.
This elation, with center o ■ PP', is conveniently denoted by [o; P —► P'].
6.24 Any collineation that has one range of invariant points (but not more
than one) is perspective.
Proof. Since the identity is (trivially) a projectivity, 6.11 tells us that
any such collineation is projective. There cannot be more than one invariant
point outside the line o whose points are all invariant; for, two such would
form, with two arbitrary points on o, a quadrangle of the kind considered
in 6.12. If there is one invariant point O outside o, every line through O
meets o in another invariant point; that is, every line through O is invariant.
Any noninvariant point P lies on such a line and is therefore transformed
into another point P' on this line OP\ hence the collineation is a homology,
as in 6.22. If, on the other hand, all the invariant points lie on o, any two
distinct joins PP' and QQ' (of pairs of corresponding points) must meet o
in the same point O, and the collineation is an elation, as in 6.23.
Hence, also,
6.25 If a collineation has a range of invariant points, it has a pencil of
invariant lines.

6.26 All the invariant points of an elation lie on its axis.

6.27 For a homology, the center is the only invariant point not on the axis.

EXERCISES

1. Let P, P , Q, Q , D be five points, no three collinear. Then there is a unique

perspective collineation that takes P to P', and Q to Q', while its axis
passes through D.
 HARMONIC HOMOLOGY 55

2. What kind of projectivity does a perspective collineation induce on a line

through its center?

3. If, on a line through its center O, an elation transforms A into A', and
B into B', then Q(OAA', OB'B). (Reference 19, p. 78.) [Hint: Use 5.11.]

4. What kind of transformation is the product of two elations having the

same axis?

5. Elations with a common axis are commutative.

6. What kind of projective collineation will leave invariant two points and
two lines (the points not lying on the lines) ?

7. Referring to Theorem 6.12, discuss the significance of the phrase “forming

a quadrilateral.”

8. When the projective plane is regarded as an extension of the Euclidean

plane, what is the Euclidean name for a perspective collineation whose
axis is the line at infinity?

6.3 Involutory Collineations

Suppose a given transformation relates a point X to X', X' to X", X" to

X1", . . . , 3f(n_1) to X{n). If, for every position of X, Xin) coincides with X
itself, the transformation is said to be periodic and the smallest n for which
this happens is called the period. Thus the identity is of period 1, an involution
is (by definition) of period 2, and the projectivity ABC a BCA (for any three
distinct collinear points A, B, C) is of period 3.
We saw, in 6.22, that a homology is determined by its center O, axis o, and
one pair of corresponding points P, P'. In the special case when the harmonic
conjugate of O with respect to P and P' lies on o, we speak of a harmonic
homology. Clearly

6.31 A harmonic homology is determined when its center and axis are
given.

For any point P, the corresponding point P' is simply the harmonic con¬
jugate of P with respect to O and o ■ OP. Thus a harmonic homology is of
period 2. Conversely,

6.32 Every projective collineation of period 2 is a harmonic homology.

Proof. Given a projective collineation of period 2, suppose it inter¬
changes the pair of distinct points PP' and also another pair QQ' (not on
the lineRP'). By 6.13, it is the only projective collineation that transforms the
56 ■ TWO-DIMENSIONAL PROJECTIVIOES

quadrangle PP'QQ' into P'PQ'Q. The invariant lines PP' and QQ' meet in
an invariant point O, as in Figure 6.3a. Since the collineation interchanges
the pair of lines PQ, P'Q', and likewise the pair PQ', P' Q, the two points

M = PQ • P'Q' and N = PQ'-P'Q

are invariant. Moreover, the two invariant lines PP' and MN meet in a
third invariant point L on MN. By Axiom 2.18, every point on MN is
invariant. Thus the collineation is perspective (according to the definition
in Section 6.2). Since, by Axiom 2.17, its center O does not lie on its axis
MN, it is a homology. Finally, since H(PP', OL), it is a harmonic homology.

EXERCISES

1. What kind of transformation is the product of two harmonic homologies

having the same axis but different centers ?

2. Any elation with axis o may be expressed as the product of two harmonic
homologies having this same axis o.

3. What kind of collineation is the product of three harmonic homologies

whose centers and axes are the vertices and opposite sides of a triangle?

4. The product of two harmonic homologies is a homology if and only if

the center of each lies on the axis of the other. In this case the product
is again a harmonic homology. (W. Jonsson.*)

5. What is the Euclidean name for a harmonic homology whose axis is the
line at infinity?

* Private communication.
 PROJECTIVE CORRELATIONS 57

6.4 Projective Correlations

We have already considered a simple instance of a point-to-line trans¬

formation: the elementary correspondence that relates a range to a pencil
when the former is a section of the latter. We shall now extend this to a
transformation X -> x' relating all the points in a plane to all the lines in the
same plane, and its dual x -► X' which relates all the lines to all the points. A
correlation is a point-to-line and line-to-point transformation that preserves
the relation of incidence in accordance with the principle of duality. Thus it
transforms ranges into pencils, pencils into ranges, quadrangles into quadri¬
laterals, and so on. A correlation is a self-dual concept, the inverse of a
correlation is again a correlation, and the product of two correlations is a
collineation.
A projective correlation is a correlation that transforms every one-dimen¬
sional form projectively, so that, if it transforms the points Y on a line b into
the lines y' through the corresponding point B', the relation between Y and y'
is a projectivity Y a /• There is a theorem analogous to 6.11:

6.41 Any correlation that transforms one range projectively is a projective

correlation.
Proof. Let a and A' be the corresponding line and point that carry
the projectively related range and pencil X a x'. We wish to establish the
same kind of relationship between any other corresponding pair, say b
and B' (see Figure 6.4a). Let Tbe a variable point on b, and O a fixed point
on neither a nor b. Let O Y meet a in X. The given correlation transforms
O into a fixed line o' (through neither A' nor Bj, and O Yinto a point o' • y'
which is joined to A’ by a line x'. Since
O o'
Fa x a x' a/,

the correlation induces a projectivity Fa/ between b and B' as desired.

To obtain the dual result for a pencil and the corresponding range, we

Figure 6.4a
58 TWO-DIMENSIONAL PROJECT1VITIES

merely have to regard the range of points Y on b as a section of the given

pencil. This pencil yields a range which is a section of the pencil of lines y
through B'.
As a counterpart for 6.13, we have:

6.42 A quadrangle and a quadrilateral, with the four vertices of the former
associated in a definite order with the four sides of the latter, are related by
just one projective correlation.

Proof. Let defpqr and D’E'FP'Q'R! be the given quadrangle and

quadrilateral, as in Figure 6.4b. What effect should such a correlation have
on an arbitrary point A ? For definiteness suppose A is a: • y with jc through
d- e and y through d- q. The projectivities </e/ a D'E'F' and dqr a D'Q'R'
determine a line a = X' Y', where

defx a D'E'F'X', dqry a D'Q'R'Y'.

To prove that this correspondence A -»• a' is a correlation, we have to verify

that it also relates lines to points in such a way that incidences are preserved.
For this purpose, let A vary in a range, so that y. By our construction
for a', we now have
X' A X A y A Y'.
 PROJECTIVE CORRELATIONS 59

Since d is an invariant line of the perspectivity x ^ y, D' must be an in¬

variant point of the projectivity X' a Y'. Thus a varies in a pencil; that is,
collinear points yield concurrent lines. We have not only a point-to-line
transformation but also aline-to-point transformation, dualizing incidences,
namely, a correlation. The projectivity x p X' suffices to make it a projec¬
tive correlation.
Finally, there is no other projective correlation transforming defpqr into
D'E'F'P'Q'R'; for, if another transformed A into ax, the inverse of the
latter would take ax to A, the original correlation takes A to a', and al¬
together we would have a projective collineation leaving D'E'F'P'Q'R'
invariant and taking ax to a . As in Section 6.1, this establishes the unique¬
ness of the correlation A —*■ a .
The dual construction yields a projective correlation transforming a
given quadrilateral into a given quadrangle.

EXERCISE

If a correlation transforms a given quadrangle into a quadrilateral, it trans¬

forms the three diagonal points of the quadrangle into the three diagonal
lines of the quadrilateral.
 CHAPTER SEVEN

Polarities

Chasles was the last important member of the great French school
of projective geometers. After his time primacy in this subject
passed across the Rhine, never to return.
/. L. Coolidge (Reference 4, p. 58)

7.1 Conjugate Points and Conjugate Lines

A polarity is a projective correlation of period 2. In general, a correlation

transforms each point A into a line a and transforms this line into a new
point A". When the correlation is of period 2, A" always coincides with A and
we can simplify the notation by omitting the prime ('). Thus a polarity relates
A to a, and vice versa. Following J. D. Gergonne (1771-1859), we call a the
polar of A, and A the pole of a. Since this is a projective correlation, the polars
of all the points on a form a projectively related pencil of lines through A.
Since a polarity dualizes incidences, if A lies on b, the polar a passes through
the pole B. In this case we say that A and B are conjugate points, and that a
and b are conjugate lines. It may happen that A and a are incident, so that
each is self-conjugate: A on its own polar, and a through its own pole. How¬
ever, the occurrence of self-conjugate lines (and points) is restricted by the
following theorem:

7.11 The join of two self-conjugate points cannot be a self-conjugate line.

Proof. If the join a of two self-conjugate points were a self-conjugate
line, it would contain its own pole A and at least one other self-conjugate
point, say B. The polar of B, containing both A and B, would coincide with
60
 SELF-CONJUGATE POINTS 61

a; two distinct points would both have the same polar. This is impossible,
since a polarity is a one-to-one correspondence between points and lines.

The occurrence of self-conjugacy is further restricted as follows:

7.12 It is impossible for a line to contain more than two self-conjugate

points.
Proof. Let p and q (through C) be the polars of two self-conjugate
points P and Q on a line c, as in Figure 7.1a. Let R be a point onp, distinct
from C and P. Let its polar r meet q in S. Then S = q • r is the pole of
QR = s, which meets r in T, say. Also T = r • s is the pole of RS = t,

which meets c in B, say. Finally, B — c • t is the pole of CT = b, which

meets c in A, the harmonic conjugate of B with respect to P and Q.
The point B cannot coincide with Q or P. For, B = Q would imply
R = C; and B = P would imply S = C,r = p, R = P; but we are assuming
that R is neither C nor P. Hence, by 2.51, A ^ B, and jBisnotself-conjugate.
We thus have, on c, two self-conjugate points P, Q and a non-selfconjugate
point B.
Since the polars of a range form a projectively related pencil, each point
lone determines a conjugate point Y on c, which is where its polar a: meets
c (see Figure 7.1b), and this correspondence between X and Y is a projec-
tivity:
* A X A Y.
When X is P, x isp, and Y is P again; thus P is an invariant point of this
projectivity. Similarly, Q is another invariant point. But when A is B, Y is
the distinct point A; therefore the projectivity is not the identity. By Axiom
2.18, P and Q are its only invariant points; that is, P and Q are the only
self-conjugate points on c. This completes the proof that c cannot contain
more than two self-conjugate points.
62 POLARITIES

A closely related result is this:

7.13 A polarity induces an involution of conjugate points on any line that
is not self-conjugate.
Proof. On a non-selfconjugate line c, the projectivity X a Y, where
Y = c • x (as in Figure 7.1b), transforms any non-selfconjugate point B
into another point A — b • c, whose polar is BC. The same projectivity
transforms A into B. Since it interchanges A and B, it must be an involution.

Dually, the lines x and CX are paired in the involution of conjugate lines
through C.
Such a triangle ABC, in which each vertex is the pole of the opposite side
(so that any two vertices are conjugate points, and any two sides are con¬
jugate lines), is called a self-polar triangle.

EXERCISES

1. Every self-conjugate line contains just one self-conjugate point.

2. If a and b are nonconjugate lines, every point A on a has a conjugate point

Y on b. The relation between X and Fis a projectivity. It is a perspectivity
if and only if the point a • b is self-conjugate. (Reference 19, p. 124.) [Hint:
Use 4.22.]
3. Observe that the polarity of 7.12 (Figure 7.1a) relates the quadrangle
CRST to the quadrilateral erst, and the harmonic set of points P, Q, A, B
to the harmonic set of lines CP, CQ, CB, CA.
4. In the polarity of 7.12, find the poles of BA and SA.
5. In Section 3.4 we defined a “trilinear polarity.” Is this a true polarity?

7.2 The Use of a Self-Polar Triangle

We have referred to 6.11 as an analogue of 5.31. Another possible candidate

for the same distinction is as follows:
7.21 Any projective correlation that relates the three vertices of one tri¬
angle to the respectively opposite sides is a polarity.
Proof. Consider the correlation ABCP -* abep, where a, b, c are the
sides of the given triangle ABC and P is a point not on any of them. Then
P is a line not through any of A, B, C. The point P and line p determine
6 points on the sides of the triangle, as in Figure 7.2a:
Pa a ' dP, Pb = b • BP, Pc = c • CP, Av = a • p, Bp — b • p, Cv = c • p.
 SELF-POLAR TRIANGLE 63

The correlation, transforming A, B, C into a, b, c, also transforms a = BC

into b • c = A, AP into a • p = Av, Pa = a • AP into AAP, and so on. Thus
it transforms the triangle ABC in the manner of a polarity, but we still have
to investigate its effect on points (such as Ap) on a side of the triangle, and
on points (such as P) of general position; in other words, does it transform
Ap into AP, and p into PI
The correlation transforms each point X on c into a certain line which
intersects c in Y, say. Since it is projective correlation, we have X a Y.
When Xis A, Yis B; and when Xis B, Y is A. Thus the projectivity X a Y
interchanges A and B, and is an involution. Since the correlation transforms
Pc into CCP, the involution includes PcCp as one of its pairs. Hence the
correlation transforms CP into CPc, which is CP. Similarly, it transforms
AP into AP, and Bv into BP. Therefore it transforms p = APBP into
AP • BP = P, as required.
We have now proved that the correlation ABCP —> abcp is a polarity. An
appropriate symbol, analogous to the symbol (AB){PQ) for an involution, is

(ABCXPp).

Thus any triangle ABC, any point P not on a side, and any linep not through
a vertex, determine a definite polarity (ABC)(Pp), in which the polar x of
an arbitrary point X can be constructed by incidences.
This construction could be carried out by the method of 6.42, as applied to
the correlation ABCP —> abcp. More elegantly, we could adapt the notation
of Figure 7.2a so that

Xa = a- AX, Xb = b • BX, Ax = a-x, Bx = b ■ x.

64 POLARITIES

Then Ax is the mate of Xa in the involution (.BC)(PaAP), Bx is the mate of Xb

in (CA)(PbBp), and x is AXBX. A still simpler procedure will be described in
Section 7.4, but it seems desirable to deal with some other matters first.
Consider a polarity (ABCfPp), in which P does not lie on p (see Figure
7.2a). Since the polars of the points

Ap = a-p, Bv = b • p, Cp = c-p

are AP, BP, CP, the pairs of opposite sides of the quadrangle ABCP meet the
line p in pairs of conjugate points. Hence

7.22 In a polarity {ABC){Pp), where P is not on p, the involution of con¬

jugate points on p is the involution determined on p by the quadrangle ABCP.

EXERCISES

1. In the notation of Figure 2.4a, any correlation that relates the points
S, D, E, F to the respective lines g, SA, SB, SC is a polarity.
2. Consider a polarity (ABC)(Pp) in which P is on p. Find Q and q, not
incident, so that the same polarity can be described as (ABC){Qq).

7.3 Polar Triangles

From any given triangle we can derive a polar triangle by taking the polars
of the three vertices, or the poles of the three sides. M. Chasles observed that
a triangle and its polar triangle, if distinct, are perspective. In other words,

7.31 Chasles’s Theorem : If a triangle is not self-polar for a given polarity,

the polars of its vertices meet the respectively opposite sides in three collinear
points.
Proof. Let PQR be a triangle whose sides QR, RP, PQ meet the
polars p, q, r of its vertices in points Px, Qu Rx, as in Figure 7.3a. The polar

of R1 = PQ ■ r is, of course, r1 = {p • q)R. Define also the extra points

P' = PQ • q, R! — QR ■ q, and their polars p' = (p ■ q)Q, r' — (q • r)Q.
By Theorem 1.63 and the polarity, we have

RxPP'Q a PR^QP' a prxqp a PXRR'Q.

By 4.22 (since Q is invariant), RXPP' ^ PXRR'. The center of the perspec-

tivity, namely PR ■ P'R' = Qx, must lie on the line RXPX. Hence Px, Qlt Rx
are collinear, as desired.
 POLAR TRIANGLES 65

Figure 7.3a

This proof breaks down if Q lies on q. But in that case we permute the
names of P, Q, R (and correspondingly p, q, r), or call the first triangle pqr
and the second PQR, in such a way that the new Q and q are not incident. It
is evidently impossible for each triangle to be inscribed in the other.

EXERCISES

1. A triangle and its polar triangle (if distinct) are perspective from a line, and
therefore also from a point. The point is the pole of the line.

2. Two triangles ABP and abp, with A on b, B on a, but no other incidences,

are polar triangles for a unique polarity.

7.4 A Construction for the Polar of a Point

We are now ready to describe the “still simpler procedure” (Figure 7.4a)
which was promised in Section 7.2.

7-41 The polar of a point X (not on AP, BP, or p) in the polarity (.ABC)(Pp)
is the line XxX2 determined by
Ax — a • PX, Px—p- AX, Xx = AP-AxPx,
B2 = b • PX, P2=p- BX, X2 = BP- B2P2.
Proof.Applying 7.31 to the triangle PAX, we deduce that its sides
AX, XP, PA meet the polars p, a, x of its vertices in three collinear points,
66 POLARITIES

Figure 7.4a

the first two of which are Px and Ax. Hence x must meet PA in a point lying
on PXAX, namely, in the point PA • PXAX — Xx. That is, x passes through Xx.
Similarly (by applying 7.31 to the triangle PBX instead of PAX), x passes
through X2.
This construction fails when X lies on AP, for then A1P1 coincides with
AP, and Xx is no longer properly defined. However, since X2 can still be
constructed as above, the polar of X is now APX2 (where Av = a • p).
Similarly, when X is on BP, its polar is XxBv.

Finally, to locate the polar of a point X on p, we can apply the dual of the
above construction to locate the pole Y of a line y through X. This y may be
any line through Xexceptp or PX. (It is convenient to choose y — AX or, if
this happens to coincide with PX, to choose y = BX.) Then the desired polar
is x = P Y.

EXERCISES

1. For any point X, not on AP, BP, or p, the polar is

[AP • {a • PX){p - AX)][BP • (b • PX)(p ■ BX)].
Write out the dual expression for the pole of any line x, not through A v,
Bp, or P, and draw a figure to illustrate this dual construction.
2. If X lies on AB, x joins C to
AP-(a- PX)(p ■ c).
Deduce an alternative construction for the pole of a line WX, with W on
CA and X on AB.
 SELF-POLAR PENTAGON 67

7.5 The Use of a Self-Polar Pentagon

Instead of describing a polarity as (ABC){Pp), we can equally well describe

it in terms of a self-polar pentagon, that is, a pentagon in which each of the
five vertices is the pole of the “opposite” side, as in Figure 7.5a. This way of
specifying a polarity is a consequence of the following theorem, due to
von Staudt:
7.51 The projective correlation that transforms four vertices of a pentagon
into the respectively opposite sides is a polarity and transforms the remaining
vertex into the remaining side.

Proof. The correlation that transforms the four vertices Q, R, S, T

of the pentagon PQRST into the four sides q = ST, r = TP, s = PQ,
t — QR also transforms the three sides t = QR, p = RS, q = ST into the
three vertices T = q • r, P = r • s, Q = s • t, and the “diagonal point”
A = q • t into the “diagonal line” a = QT. Thus it transforms each vertex
of the triangle AQT into the opposite side. By 7.21, it is a polarity, namely
(since it transforms p into P), the polarity (AQT)(Pp).

EXERCISES

1. In the notation of Figure 7.2a, PBAvBvA is a self-polar pentagon.

2. Let Xbe any point on none of the sides p, r, s of a given self-polar pentagon
PQRST. Then its polar is the line

[r-(f PX)(p ■ TX)}[s ■ (q • PX)(p • QX)].

68 POLARITIES

7.6 A Self-Conjugate Quadrilateral

From Chasles’s theorem, 7.31, we can easily deduce

7.61 Hesse’s Theorem: If two pairs of opposite vertices of a complete

quadrilateral are pairs of conjugate points (in a given polarity), then the third
pair of opposite vertices is likewise a pair of conjugate points.
Proof. Consider a quadrilateral PQRP1Q1R1, as in Figure 7.3a,
with P conjugate to P1} and Q to Qx. The polars p and q (of P and Q) pass
through Px and Ql5 respectively. By Chasles’s theorem (7.31) the polar of R
meets PQ in a point that lies on PXQX, namely in the point PQ • PXQX = Rx.
Therefore the polar of R passes through Rx; that is, R is conjugate to Rx.

EXERCISE

In the notation of Figure 7.5a, let an arbitrary line through P meet ST in U,

and QR in V. Then RU and SV are conjugate lines.

7.7 The Product of Two Polarities

Figure 2.3a (or Figure 6.2b) shows the homology with center O and axis
o = DF that transforms P into P' (and consequently Q into Qj. This
homology may be expressed as the product of two polarities

0ODF)(Pp) and (ODF){P'p),

where p is any line not passing through a vertex of the common self-polar
triangle ODF. To prove this, we merely have to observe that, the homology
and the product of polarities both transform the quadrangle ODFP into
ODFP'.
Unfortunately, this expression for a homology as the product of two polar¬
ities cannot in any simple way be adapted to an elation. Accordingly, it is
worthwhile to mention a subtler expression that applies equally well to either
kind of perspective collineation. Figure 7.7a shows the homology or elation
with center O and axis o — CP that transforms A into another point A' on the
line c = OA. Here C and P are arbitrary points on the axis o (which passes
through O if the collineation is an elation). Let p be an arbitrary line through
O, meeting b = CA and b' = CA' in Q and Q'. Let B be an arbitrary point on
c. We proceed to verify that the given perspective collineation is the product of
the two polarities
(.ABC)(Pp) and (A'BC)(Pp).
 THE PRODUCT OF TWO POLARITIES 69

A’

In fact, the first polarity transforms the four points A, P, O = c • p, Q = b • p

into the four lines BC,p, CP, BP; and the second transforms these lines into
the four points A', P, c • p = O, b' • p — Q'. Thus their product transforms
the quadrangle APOQ into A'POQ'. By 6.22, this product is the same as the
given perspective collineation.
More generally,

7.71 Any projective collineation is expressible as the product of two

polarities.
Proof. By the above remarks, this is certainly true if the given collin¬
eation is perspective. Accordingly, we may concentrate our attention on a
given nonperspective collineation. Let A be a noninvariant point, and / a
noninvariant line through A. Suppose the given collineation transforms A
into A', A' into A", l into /', /' into /", and /" into /"'. Since the collineation
is not perspective, we may choose A and / (as in Figure 7.7b) so that
A A' is not an invariant line and / • /' is not an invariant point; more¬
over, A' does not lie on any of the three lines /, /'", and consequently A

Figure 7.7b
70 POLARITIES

does not lie on V nor on 1". Let l" meet / in B, l' in C. Of the two polarities

(.AA"B)(A'lj and (A'A"C)(Al'"),

the former transforms the four points A, A', B, C = l' • 1" into the four
lines A"B = l" = A"C, l' = CA', A"A, A'A, and the latter transforms
these lines into the four points A', A", l' ■ l" = B', 1" • t" = C. Hence their
product is the same as the given collineation, which is what we wished to
prove.

This theorem has an interesting corollary which includes 6.25 as a special

case:

In any projective collineation, the invariant points and invariant lines

7.72
form a self-dual figure.

EXERCISES

1. Can a projective collineation interchange two points without being a

harmonic homology?
2. If a projective collineation has three invariant points forming a triangle, it
is the product of two polarities having a common self-polar triangle.
 CHAPTER EIGHT

The Conic

Let us now pause to note that we have swung through a complete

circle, from Desargues and Poncelet who started with a conic and
defined a polar system, to von Staudt, who starts with a polar
system and reaches a conic.
J. L. Coolidge (Reference 4, p. 64)

8.1 How a Hyperbolic Polarity Determines a Conic

The study of conics is said to have begun in 430 b.c., when the Athenians,
suffering from a plague, appealed to the oracle at Delos and were told to
double the size of Apollo’s cubical altar. Attempts to follow this instruction
by placing an equal cube beside the original one, or by doubling the edge
length (and thus producing a cube of eight times the volume of the original
one), made the pestilence worse than ever. At last, Hippocrates of Chios
explained that what was needed was to multiply the edge length by the cube
root of 2. The first geometrical solution for this problem was given by
Archytas about 400 b.c. (Reference 2, p. 329). He used a twisted curve.
Menaechmus, about 340 b.c., found a far simpler solution by the use of
conics. For the next six or seven centuries, conics were investigated in great
detail, especially by Apollonius of Perga (262-200 B.c.), who coined the
names ellipse, parabola, and hyperbola. Interest in this subject was revived in
the seventeenth century (a.d.) when Kepler showed how a parabola is at once
a limiting case of an ellipse and of a hyperbola, and Blaise Pascal (1623-1662)
discovered a projective property of a circle which consequently holds just as
well for any kind of conic. We shall see how this characteristic property of a
71
72 THE CONIC

conic can be deduced from an extraordinarily natural and symmetrical defini¬

tion which was given by von Staudt in his Geometrie der Lage (1847). We
shall find that, in the projective plane, there is only one kind of conic: the
familiar distinction between the ellipse, parabola, and hyperbola can only be
made by assigning a special role to the line at infinity.

Figure 8.1a

Polarities, like involutions, are of two possible kinds. By analogy with

involutions, we call a polarity hyperbolic or elliptic according as it does or
does not admit a self-conjugate point. In the former case it also admits a
self-conjugate line: the polar of the point. Thus any hyperbolic polarity can
be described by a symbol {ABC)(Pp), where P lies on p. This self-conjugate
point P, whose existence suffices to make the polarity hyperbolic, is by no
means the only self-conjugate point: there is another on every line through P
except its polar p.
This can be proved as follows. By 7.11, the only self-conjugate point on a
self-conjugate line is its pole. Dually, the only self-conjugate line through a
self-conjugate point P is its polarp. By 7.13, it follows that every line through
P, exceptp, is the kind of line that contains an involution of conjugate points.
By 5.41, this involution, having one invariant point P, has a second invariant
point Q which is, of course, another self-conjugate point of the polarity.
Thus the presence of one self-conjugate point implies the presence of many.
Their locus is a conic, and their polars are its tangents. This simple definition
exhibits the conic as a self-dual figure: the locus of self-conjugate points and
also the envelope of self-conjugate lines (Figure 8.1a).
In some geometries (such as complex geometry) every polarity is hyperbolic,
that is, every polarity determines a conic. In other geometries (e.g., in real
geometry) both kinds of polarity occur, and then the theory of polarities is
more general than the theory of conics. From here on, we shall deal solely
with hyperbolic polarities, so that “pole” will mean “pole with respect to a
conic.” Similarly, instead of “conjugate for a polarity” we shall say “conjugate
with respect to a conic.”
A tangent justifies its name by meeting the conic only at its pole: the point of
contact. Any other line is called a secant or a nonsecant according as it meets
 EXTERIOR AND INTERIOR POINTS 73

the conic twice or not at all, that is, according as the involution of conjugate
points on it is hyperbolic or elliptic. The above remarks show that, of the lines
through any point P on the conic, one (namely p) is a tangent and all the
others are secants. Moreover, if P and Q are any two distinct points on the
conic, the line PQ is a secant.
Dually, a point not lying on the conic is said to be exterior or interior
according as it lies on two tangents or on none, that is, according as the
involution of conjugate lines through it is hyperbolic or elliptic. Thus an
exterior point H is the pole of a secant h, and an interior point E (if such a
point exists) is the pole of a nonsecant e. Of the points on a tangent p, one
(namely P) is on the conic, and all the others are exterior. If p and q are any
two distinct tangents, the point p • q (which is the pole of the secant PQ) is
exterior.

Figure 8.1b

Figure 8.1b may help to clarify these ideas, but we must take care not to be
unduly influenced by real geometry. For instance, it would be foolish to
waste any effort on trying to prove that every point on a nonsecant is exterior,
or that every line through an interior point is a secant; for in some geometries
these “obvious” propositions are false.

Consider the problem of drawing a secant through a given point A. If A

is interior, we simply join A to any point P on the conic. (Since AP cannot be a
tangent, it must be a secant.) If A is on the conic, we join it to another point on
the conic. Finally, if A is exterior, we join it to each of three points on the
conic. Since at most two of the lines so drawn could be tangents, at least one
must be a secant.
On a secant PQ, the involution of conjugate points is (PP)(QQ). Hence, by
5.41,

8.11 Any two conjugate points on a secant PQ are harmonic conjugates

with respect to P and Q.
74 THE CONIC

Conversely,

8.12 On a secant PQ, any pair of harmonic conjugates with respect to P

and Q is a pair of conjugate points with respect to the conic.

Dually,

8.13 Any two conjugate lines through an exterior point p • q are harmonic
conjugates with respect to the two tangents p, q; and any pair of harmonic
conjugates with respect to p and q is a pair of conjugate lines with respect to
the conic.

EXERCISES

1. Every point on a tangent is conjugate to the point of contact. Dually, the

tangent itself is conjugate to any line through the point of contact.

2. The polar of any exterior point joins the points of contact of the two tan¬
gents that can be drawn through the point. Dually, the pole of a secant PQ
is the point of intersection of the tangents at P and Q.

3. Is it true that every conic has exterior points?

4. Is it true that the polar of any interior point is a nonsecant?

5. If PQR is a triangle inscribed in a conic, the tangents at P, Q, R form a

triangle circumscribed about the conic. These are perspective triangles.
{Hint: Use Chasles’s theorem.)

6. Any two vertices of a triangle circumscribed to a conic are separated

harmonically by the point of contact of the side containing them and the
point where this side meets the line joining the points of contact of the
other sides. (Reference 19, p. 140.)

7. In the spirit of Section 2.5, Exercise 5, what would be natural names for:
(i) a conic of which the line at infinity is a nonsecant,
(ii) a conic of which the line at infinity is a tangent,
(iii) a conic of which the line at infinity is a secant,
(iv) the pole of the line at infinity with respect to any conic,
(v) the tangents of a hyperbola through its center,
(vi) a line (other than a tangent) through the center of any conic,
(vii) conjugate lines through the center?

8. Continuing to work in the “affine” geometry of Exercise 7, let D be the

point of intersection of the tangents at any two points P and Q on a pa¬
rabola, let C be the midpoint of PQ, and let S be the midpoint of CD.
Then S lies on the parabola.
 INSCRIBED QUADRANGLE 75

8.2 The Polarity Induced by a Conic

We have seen that any hyperbolic polarity determines a conic. Conversely,

any conic (given as a locus of points) determines a hyperbolic polarity. A
suitable construction will emerge as a by-product of the following theorem:

8.21 If a quadrangle is inscribed in a conic, its diagonal triangle is self-

polar.
Proof. Let the diagonal points of the inscribed quadrangle PQRS be

A = PS • QR, B= QS-RP, C = RS ■ PQ,

as in Figure 8.2a. The line BC meets the sides QR and PS in points Ax and
A2 such that H{QR, AAj) and H(P<S, AA2). By 8.12, both Ax and A2 are
conjugate to A. Thus the line BC, on which they lie, is the polar of A.
Similarly, CA is the polar of B, and AB of C.

Hence:

8.22 To construct the polar of a given point A, not on the conic, draw any
two secants QR and PS through A; then the polar joins the two points QS • RP
and RS • PQ.

In other words, we draw two secants through A to form an inscribed

quadrangle with diagonal triangle ABC, and then the polar of A is BC.
The dual construction presupposes that we know the tangents from any
exterior point. This presents no serious difficulty (since their points of con¬
tact lie on the polar of the given point); but the tangents are not immediately
apparent, for the simple reason that we are in the habit of dealing with
76 THE CONIC

loci rather than envelopes. If we insist on regarding the conic as a locus, we

can construct the pole of a given line as the common point of the polars of
any two points on the line. Then:

8.23 To construct the tangent at a given point A on the conic, join A to the
pole of any secant through A.

These constructions serve to justify the statement that any conic determines
a hyperbolic polarity whose self-conjugate points are the points on the conic.

EXERCISES

1. Let A and B be two conjugate points with respect to a given conic. Let an
arbitrary line through A meet the conic in Q and R, while BQ and BR
meet the conic again in S and P, respectively. Then A, S, P are collinear.

2. If PQR is a triangle inscribed in a conic, any point A on QR (except Q or R

or p • QR) is a vertex of a self-polar triangle ABC with B on RP and C on
PQ-
3. If ABC is a self-polar triangle for a given conic, any secant QR through A
provides a side of an inscribed triangle PQR whose remaining sides pass
through B and C, respectively.
4. A conic is transformed into itself by any harmonic homology whose
center is the pole of its axis.
5. The polars of the vertices of any quadrangle PQRS form a quadrilateral
pqrs such that the 3 diagonal points of PQRS are the poles of the 3 diagonal
lines of pqrs. (See the exercise at the end of Chapter 6.) What happens to
these 3 points and 3 lines in the special case when PQRS is an inscribed
quadrangle, so that p, q, r, s are tangents ?

6. There is one and only one conic through three given points and having
another given point as pole of a given line. (Reference 19, p. 137.) [Hint:
Let PQR be the given triangle, D and afthe point and line. Let DP meet d in
P'. Let S be the harmonic conjugate of P with respect to D and P'. Let ABC
be the diagonal triangle of the quadrangle PQRS. Consider the polarity
(ABC)(Dd).]

8.3 Projectively Related Pencils

The following theorem, due to Franz Seydewitz (1807-1852), provides a

useful connection between conjugate points and conjugate lines:
 SEYDEWITZ AND STEINER 77

8.31 Seydewitz’s Theorem: If a triangle is inscribed in a conic, any line

conjugate to one side meets the other two sides in conjugate points.
Proof. Consider an inscribed triangle PQR, as in Figure 8.2a. Any line
c conjugate to PQ is the polar of some point C on PQ. Let RC meet the
conic again in S. By 8.21, the diagonal points of the quadrangle PQRS
form a self-polar triangle ABC whose side c contains the conjugate points
A and B: one on QR and the other on RP.

Figure 8.3a

We are now ready for one of the most significant properties of a conic:

8.32 Steiner’s Theorem: Let lines x and y join a variable point on a conic
to two fixed points on the same conic; then x a y.
Proof. The tangentsp and q, at the fixed points P.and Q, meet in D,
the pole of PQ (see Figure 8.3a). Let c be a fixed line through D (but not
through P or Q), meeting x in B, andy in A. By 8.31, BA is a pair of the
involution of conjugate points on c. Hence, when the point R — x • y
varies on the conic, we have
x A B A A A y,
as desired.
The following remarks make it natural for us to include the tangents p
and q as special positions for x and y. (Notice that the idea of “continuity,”
though intuitively helpful, is not assumed.) Writing d — PQ and Cx = c • d,
consider the possibility of using P or Q as a position for R. When R is P, y is d,
A is Cl5 B is the conjugate point D, and therefore x is p. Similarly, when R is
Q, x is d, B is Cl5 A is D, and y is q. In other words, when y is d, x is p; and
when x is d, y is q.

EXERCISES

1. Dualize Seydewitz’s theorem and Steiner’s theorem (Figures 8.2a and

8.3a).
78 THE CONIC

2. (“The Butterfly Theorem.”) Let P, Q, R, S, T, U be 6 points on a conic,

such that the lines PS, QR, TU all pass through a point A. Also let TU
meet PR in E, and QS in B. Then

TEAU a TABU.

8.4 Conics Touching two Lines at Given Points

In the proof of 8.32, we chose an arbitrary line c through D (the pole of

PQ). For any particular position of R, we can usefully take c to be a side of the
diagonal triangle ABC of the inscribed quadrangle PQRS, where S is on RD,
as in Figure 8.4a; that is, we define C — PQ • RS and let c be its polar AB,
which passes through D since C lies on d. The point Cx = c • d, being the pole
of the line CD, is the harmonic conjugate of C with respect to P and Q.

If we are not given the conic, but only the points P, Q, R, D, we can still
construct C = PQ ■ RD and its harmonic conjugate Cx. Then c is the line
CXD, which meets QR and RP in A and B. The conic itself can be described
as the locus of self-conjugate points (and the envelope of self-conjugate lines)
in the polarity (ABC)(Pp), where p = PD. Since PD and QD are the tangents
at P and Q, our conclusion may be stated as follows:

8.41 A conic is determined when three points on it and the tangents at two
of them are given.

Retaining P, p, Q, q, but letting R vary, we obtain a “pencil” of conics

touching p at P and q at Q. Such conics are said to have double contact (with
one another). Let one of them meet a fixed line h in R and S' (see Figure 8.4b).
Let h meet the fixed line d = PQ in C. Let c (the polar of C) meet din Cx, and
 DOUBLE-CONTACT PENCIL OF CONICS 79

h in C2. The line c is fixed, since it joins D — p ■ q to Cl5 which is the harmonic
conjugate of C with respect to P and Q. In other words, the fixed point
C = d- h has the same polar for all the conics. Thus C2 = c • h is another
fixed point, and RS is always a pair of the hyperbolic involution (CC)(C2C2)
on h. Hence:

8.42 Of the conics that touch two given lines at given points, those which
meet a third line (not through either of the points) do so in pairs of an
involution.

EXERCISES

1. In Figure 8.4a, RCx is the tangent at R. [Hint: Use 8.23 with R for A.]

2. Given a quadrangle PQRD (as in Figure 8.4b), construct another point

on the conic through R that touches PD at P and QD at Q.

3. Given three tangents to a conic, and the points of contact of two of them,
construct another tangent.

4. Any two conics are related by a projective collineation and by a projective

correlation. More precisely, any three distinct points on the first conic can
be made to correspond to any three distinct points or tangents of the
second.

5. All the conics of a double-contact pencil are transformed into themselves

(each separately) by many harmonic homologies. In fact, the center of
such a homology may be any point (other than P or Q) on the line PQ
(Figure 8.4b). (E. P. Wigner.*)

* Private communication.
80 THE CONIC

8.5 Steiner’s Definition for a Conic

We have followed von Staudt in defining a conic by means of the self¬

conjugate points (and self-conjugate lines) in a hyperbolic polarity. An
alternative approach is suggested by Steiner’s theorem, 8.32. Could a conic be
defined as the locus of the common point of corresponding lines of two
projective (but not perspective) pencils? Of course, this construction would
only yield a conic locus: there would remain the problem of deducing that its
tangents join corresponding points of two projective (but not perspective)
ranges. The theorem that makes such an alternative procedure possible is as
follows:
8.51 Let variable lines x and y pass through fixed points P and Q in such a
way that x a y but not x' 7y y. Then the locus of x • y is a conic through P
and Q. If the projectivity has the effect pdx a dqy, where d = PQ, then p
and q are the tangents at P and Q.
Proof. Since the projectivity x a y is not a perspectivity, the line
d = PQ (Figure 8.3a) does not correspond to itself. Hence there exist lines
p and q such that the projectivity relatesp to d, and d to q. By 8.41, there is a
unique conic touching p at P, q at Q, and passing through any other
particular position of the variable point x • y. By 8.32, this conic determines
a projectivity relating all the lines through P to all the lines through Q. By
the fundamental theorem, the two projectivities must coincide, since they
agree for three particular positions of x and the corresponding positions
of y.

EXERCISES

1. Given a triangle PQR and a point O, not on any side, what is the locus of
the trilinear pole of a variable line through 01 [Hint: In Figure 3.4a, let
PQR be fixed while DE varies in a pencil. Then A a D x E a R ]
2. Let P and Q be two fixed points on a tangent of a conic. If x is a variable
line through P, and X is the (variable) pole of x, what is the locus of
x • QX1 (S. Schuster.*)
3. Give an explicit determination of the locus in Exercise 2, by naming a
sufficient number of special points on it.
4. Let P, Q, R, P', Q' be five points, no three collinear, and let x be a variable
line through P. Define
N = PQ' • P'Q, M — RP' • x, L = Q'R • MN, R' = QL-x.
What is the locus of R' 1

* Private communication.
 CHAPTER NINE

The Conic, Continued

Had [Pascal] confined his attention to mathematics he might have

enriched the subject with many remarkable discoveries. But after
his early youth he devoted most of his small measure of strength to
theological questions.
J. L. Coolidge (Reference 5, p. 89)

9.1 The Conic Touching Five Given Lines

Dualizing 8.51 (as in Figure 9.1a) we obtain

9.11 Let points X and Y vary on fixed lines p and q in such a way that
X a Y but not X yy Y. Then the envelope of XY is a conic touchingp and q.
If the projectivity has the effect PDX a DQY, where D = p • q, then P and
Q are the points of contact of p and q.
Let X1} X2, X3 be three positions of X on p, and Ylf Y2, Y3 the corre¬
sponding positions of Y on q, as in Figure 9.1b. By 4.12, there is a unique
D

Figure 9.1a Figure 9.1b

81
82 THE CONIC, CONTINUED

projectivity a Yx Y2 Y3 Y. By Theorem 9.11, the envelope of XT is a

conic, provided no three of the five lines X{ Y{,p, q are concurrent. Conversely,
if five such lines all touch a conic, any other tangent XY satisfies

a YxY2YzY.
Hence
9.12 Any five lines, of which no three are concurrent, determine a unique
conic touching them.

By 4.32, the line d = PQ (Figure 9.1a) is the axis of the projectivity X a Y;

that is, if A is a particular position of X and B is the corresponding position of
Y (Figure 9.1c), the point Z = AY - BX always lies on this fixed line d.
In fact, if AB meets d in G, we have an expression for the projectivity as the
product of two perspectivities:

APDX ^ GPQZ ^ BDQY.

We may regard XYZ as a variable triangle whose vertices run along fixed
lines p, q, d while the two sides YZ and ZX pass through fixed points A and B.
We have seen that the envelope of XY is a conic touching p at P, and q at Q.
More generally,

9.13 If the vertices of a variable triangle lie on three fixed nonconcurrent

lines p, q, r, while two sides pass through fixed points A and B, not collinear
with p • q, then the third side envelops a conic.
Proof. Let XYZ be the variable triangle, whose vertices X, Y, Z run
along the fixed lines p, q, r while the sides YZ and ZX pass through points
A and B (not necessarily on p or q), as in Figure 9.Id. Then

B A
X^Z^Y.
 A CONIC ENVELOPE 83

Figure 9.Id

Since neither r nor AB passes through D = p • q, the projectivity X a Y is

not a perspectivity. By Theorem 9.11, the envelope of AT is a conic touching
p and q.
In Figure 9. Id, each position for Z on r yields a corresponding position for
the tangent XY. Certain special positions are particularly interesting.
Defining

C — q ■ r, E = p • r, G = AB • r, I = AB ■ p, J = AB • q,

we see that, when Z is E, X also is E, AY is AE, and XY also is AE. Similarly,

when Z is C, Y also is C, BX is BC, and XY also is BC. Finally, when Z is G,
X is I, Y is J, and XY is AB. Thus the lines AE, BC, AB, like p and q, are
special positions for XY. In other words, all five sides of the pentagon ABODE
are tangents of the conic. We now have the following construction for any
number of tangents of the conic inscribed in a given pentagon ABCDE.

9.14 Let Z be a variable point on the diagonal CE of a given pentagon

ABCDE. Then the two points

X — ZB • DE, Y — ZA • CD,

determine a line XY whose envelope is the inscribed conic.

In Figure 9.Id, we see a hexagon ABCYXE whose six sides all touch a
conic. The three lines A Y, BX, CE, which join pairs of opposite vertices, are
naturally called diagonals of the hexagon. Theorem 9.14 tells us that, if the
diagonals of a hexagon are concurrent, the six sides all touch a cpnic.
Conversely, if all the sides of a hexagon touch a conic, five of them can be
identified with the lines DE, EA, AB, BC, CD. Since the given conic is the
only one that touches these fixed lines, the sixth side must coincide with one
of the lines XY for which BX • A Y lies on CE. We have thus proved

9.15 Brianchon’s Theorem: If a hexagon is circumscribed about a conic,

the three diagonals are concurrent.
84 THE CONIC, CONTINUED

Figure 9.1e

Figure 9.1e illustrates this in a more natural notation: the Brianchon

hexagon is ABCDEF and its diagonals are AD, BE, CF.

EXERCISES

1. Apply Figure 9.1c to Exercise 3 of Section 8.4.

2. Obtain a simple construction for the point of contact of any one of five
given tangents of a conic. [Hint: To locate the point of contact of p, in the
notation of Figure 9.Id, make Y coincide with D, as in Figure 9.If.]
3. Measure off points X0, Xu ..., X5 at equal intervals along a line, and
Y0, Ei,..., Y5 similarly along another line through X5 = yo. What kind
of conic do the joins XkYk appear to touch? Draw some more tangents.

Figure 9.If
 A CONIC LOCUS 85

9.2 The Conic Through Five Given Points

Dualizing 9.12 (as in Figure 9.2a), we obtain

9.21 Any five points, of which no three are collinear, determine a unique
conic through them.

The dual of 9.13 was discovered independently by William Braikenridge

and Colin MacLaurin, about 1733:

9.22 If the sides of a variable triangle pass through three fixed non-
collinear points P, Q, R, while two vertices lie on fixed lines a and b‘not
concurrent with PQ, then the third vertex describes a conic.

Figure 9.2b

This enables us (as in Figure 9.2b, where the variable triangle is shaded) to
locate any number of points on the conic through five given points.
The dual of 9.15 is the still more famous

9.23 Pascal’s Theorem: If a hexagon is inscribed in a conic, the three

pairs of opposite sides meet in collinear points.

In Figure 9.2c, the hexagon is abcdef and the three collinear points are
a • d, b • e, c • f. We have obtained Pascal’s theorem by dualizing Brianchon’s.
Historically, this procedure was reversed: C. J. Brianchon (1760-1854)
obtained his theorem by dualizing Pascal’s, at a time when the principle of
duality was just beginning to be recognized. Pascal’s own proof (for a
hexagon inscribed in a circle) was seen and praised by G. W. Leibniz (1646-
1716) when he visited Paris, but afterwards it was lost. All that remains of
Pascal’s relevant work is a brief Essay pour les coniques (1640), in which the
theorem is stated as follows:

Si dans le plan MSQ du point M partent les deux droites MK, MV, & du point
86 THE CONIC, CONTINUED

S partent les deux droites SK, S V • • • & par les points K, V passe la circonference
d’un cercle coupante les droites MV, MK* SV, SK es pointes O, P, Q, N: je dis
que les droites MS, NO, PQ sont de mesme ordre.

EXERCISES

1. Given five points (no three collinear), construct the tangent at each point
to the conic through all of them.

2. Given a quadrangle PQRS and a line s through S (but not through any
other vertex), construct another point on the conic through P, Q, R that
touches s at S.

3. Given six points on a conic, in how many ways can they be regarded as the
vertices of a Pascal hexagon ?

4. Name the hexagon in Pascal’s own notation.

5. Try to reconstruct Pascal’s lost proof (using only the methods that would
have been available in his time).

* CEuvres de Blaise Pascal, edited by L. Brunschvicg and P. Boutroux, 1 (Libraire Ha-

chette: Paris, 1908), p. 252. Pascal actually wrote MP for MK, but this was obviously a slip.
By “de mesme ordre” he meant “in the same pencil” or, in the terminology of modern
projective geometry, “concurrent.” Compare “d’une mesme ordonnance” in the passage
of Desargues that we quoted on page 3. Pascal was the first person who properly appre¬
ciated the work of Desargues. The complete statement may be translated as follows:
If, in the plane MSQ, two lines MK and MV are drawn through M, and two lines SK,
SV through S, and if a circle through K and V meets the four lines MV, MK, SV, SK in
points O, P, Q, N, then the three lines MS, NO, PQ belong to a pencil.
Although nobody knows just how Pascal proved this property of a circle, there is no
possible doubt about how he deduced the analogous property of the general conic. He
joined the circle and lines to a point outside the plane, obtaining a cone and planes; then
he took the section of this solid figure by an arbitrary plane.
 DESARGUES’S INVOLUTION THEOREM 87

9.3 Conics Through Four Given Points

Desargues not only invented the word involution (in its geometrical sense)
but also showed how the pairs of points belonging to an involution on a line
arise from a “pencil”of conics through four points. This is his “involution
theorem,” which is even more remarkable than his “two triangle theorem.”

9.31 Desargues’s Involution Theorem: Of the conics that can be drawn

through the vertices of a given quadrangle, those which meet a given line
{not through a vertex) do so in pairs of an involution.

Proof. Let PQRS be the given quadrangle, and g the given line,
meeting the sides PS, QS, QR, PR in A, B, D, E, and any one of the conics
in T and U (see Figure 9.3a). By regarding S, R, T, U as four positions of a
variable point on this conic, we see from 8.32 that the four lines joining them
to P are projectively related to the four lines joining them to Q. Hence

AETU a BDTU.

Since, by Theorem 1.63, BDTU a DBUT, it follows that

AETU a DBUT.

Hence TU is a pair of the involution (AD){BE). Since this involution

depends only on the quadrangle, all those conics of the pencil which inter¬
sect g (or touch g) determine pairs (or invariant points) of the same
involution.
Referring to Figure 9.3a again, we observe that, when S and Q coincide,
the line SQ (which determines B) is replaced by the tangent at Q. Every¬
thing else remains. Hence:
9.32 Of the conics that can be drawn to touch a given line at a given point
while passing also through two other given points, those which meet another
given line {not through any of the three given points) do so in pairs of an
involution.
88 THE CONIC, CONTINUED

Similarly, by letting R and P coincide, we obtain an alternative proof for

Theorem 8.42.

EXERCISES

1. Given five points P, Q, R, S, T, no three collinear, and a line g through P,

construct the second common point of the conic PQRST and the line g.
2. A given line touches at most two of the conics through the vertices of a
given quadrangle.
3. Let P, Q, R, S, T, U be 6 points on a conic, such that the lines PS, QR, TU
all pass through a point A. Also let TU meet PR in E, and QS in B. Then
EB is a pair of the involution (AA)(TU).
Can this be deduced directly from Exercise 2 of Section 8.3?

4. Let P, Q, R, S be four points on a conic, and t the tangent at a fifth point.

If no diagonal point of the quadrangle PQRS lies on t, there is another
conic also passing through P, Q, R, S and touching t.

9.4 Two Self-Polar Triangles

Combining 9.31 with 7.22, we see that the involution determined on g

(Figure 9.3a) by the quadrangle PQRS is not only the Desargues involution
determined by conics through P, Q, R, S but also the involution of conjugate
points on g for the polarity (PQR)(Sg). Hence:

9.41 If two triangles have six distinct vertices, all lying on a conic, there is a
polarity for which both triangles are self-polar.

And conversely (Figure 9.4a),

Figure 9.4a
 TWO SELF-POLAR TRIANGLES 89

9.42 If two triangles, with no vertex of either on a side of the other, are
self-polar for a given polarity, their six vertices lie on a conic and their six
sides touch another conic.

EXERCISES

1. How many polarities can be expected to arise in the manner of 9.41 from
six given points on a conic?

2. If two triangles have six distinct vertices, all lying on a conic, their six
sides touch another conic.

3. If two conics are so situated that there is a triangle inscribed in one and
circumscribed about the other, then every secant of the former conic that is
a tangent of the latter can be used as a side of such an inscribed-circum-
scribed triangle.

4. Let P, Q, R, S, T be five points, no three collinear. Then the six points

A = QR- PS, B = RP- QS, C = PQ - RS,

A' = QR- PT, B' = RP - QT, C = PQ ■ RT

all lie on a conic. (S. Schuster.)

9.5 Degenerate Conics

For some purposes it is convenient to admit, as degenerate conics, a pair of

lines (regarded as a locus) or a pair of points (regarded as an envelope: the set
of all lines through one or both). Visibly (Figure 9.5a) a hyperbola may differ
as little as we please from a pair of lines (its asymptotes), and the set of tangents
of a very thin ellipse is hardly distinguishable from the lines through one or
other of two fixed points.
By omitting the phrase “but not jc ^ y” from the statement of Steiner’s
construction 8.51, we could allow the locus to consist of two lines: the axis of
the perspectivity x ^ y, and the line PQ (any point of which is joined to P

Figure 9.5a
90 THE CONIC, CONTINUED

Figure 9.5b Figure 9.5c

and Q by “corresponding lines” of the two pencils, namely by the invariant

line PQ itself, as in Figure 9.5b).
Dually (Figure 9.5c), when the points P and Q of Figure 8.5a coincide
with D, we have a degenerate conic envelope consisting of two points,
regarded as two pencils: the various positions of the line XY when X and Y
are distinct, and the pencil of lines through D.
In the same spirit we can say that a conic is determined by five points, no
four collinear, or by five lines, no four concurrent.

The degenerate forms of Brianchon’s theorem (Figure 9. Id) and Pascal’s

theorem (Figure 9.2c) are as follows:
If AB, CD, EF are concurrent If a • b, c • d, e • f are collinear
and DE, FA, BC are concurrent, and d ■ e, /• a, b • c are collinear,
then AD, BE, CF are concurrent. then a ■ d, b • e, c • f are collinear.
Comparing Figure 9.5d with Figure 4.4b, we see that both these statements
are equivalent to Pappus’s theorem, 4.41.

EXERCISES

1. What kind of “polarity” is induced by a degenerate conic?

2. What happens to Exercise 4 of Section 9.3 if we omit the words “If no
diagonal point of the quadrangle PQRS lies on r?”
 CHAPTER TEN

A Finite Projective Plane

Our Geometry is an abstract Geometry. The reasoning could be

followed by a disembodied spirit who had no idea of a physical
point; just as a man blind from birth could understand the Electro¬
magnetic Theory of Light.
H. G. Forder (1889- )
(Reference 9, p. 43.)

10.1 The Idea of a Finite Geometry

The above words of Forder emphasize the fact that our primitive concepts
are defined solely by their properties as described in the axioms. This fact is
most readily appreciated when we abandon the “intuitive” idea that the
number of points is infinite. We shall find that all our theorems remain valid
(although the figures are somewhat misleading) when there are only 6 points
on each line, and 31 points in the plane.
In 1892, Fano described an ^-dimensional geometry in which the number of
points on each line is p + 1 for a fixed prime p. In 1906, O. Veblen and
W. H. Bussey gave this finite Projective Geometry the name PG(n,p) and
extended it to PG(n, q), where q — pk,p is prime, and k is any positive integer.
(For instance, q may be 5, 7, or 9, but cannot be 6.)
Without attempting to explain the necessity for restricting the possible
values of q to primes and their powers, we can deduce some numerical
results immediately. Since any range can be related to any other by a sequence
of perspectivities, the number of points on a line must be the same for all
lines. Let us agree to denote it by q + 1. By the two-dimensional principle of
91
92 A FINITE PROJECTIVE PLANE

duality, this is the same as the number of lines in a pencil (that is, lying in a
plane and passing through a point); by the three-dimensional principle of
duality, it is also the same as the number of planes through a line in three-
dimensional space. In a plane, any one point is joined to the remaining
points by a pencil which consists of q -f- 1 lines, each containing the one
point and q others. Hence the plane contains
q(q + 1) + l = q2 + q + l
points and (by duality) the same number of lines. In space, the q + 1 points
on one line are joined to the remaining points by q + 1 planes, each containing
the q + 1 points and q2 others. Hence the whole space contains

q\q + 1) + (q + 1) = q3 + q2 + q + 1
points and (by duality) the same number of planes.
The general formula for the number of points in PG(n, q) is

qn+qn-1 + ---+q + 1= -.
q — i
It was proved by J. Singer (Trans. Amer. Math. Soc. 43 (1938), pp. 377-385)
that every geometry of this kind (with Desargues’s theorem taken as an
axiom if n = 2) can be represented by a combinatorial scheme such as the
one exhibited in Section 10.2 for the special case PG(2, 5).

EXERCISES

1. In PG(3, q) there are q + 1 points on each line, how many lines (or planes)
through each point? How many lines in the whole space? [Hint: Every
two of the <73 + q2 + q + 1 points determine a line, but each line is
determined equally well by any two of its q + 1 points.]

2. How many triangles occur in PG(2, q) ?

3. In the notation of Section 3.2, PG(2, q) is a configuration nd. Express n and

d in terms of q.

10.2 A Combinatorial Scheme for PG(2, 5)

In accordance with the general formula, the finite projective plane PG(2, 5)
has 6 points on each line, 6 lines through each point,

s3 _ 1
52 + 5 + 1 =--- =31
5 — 1
 PERFECT DIFFERENCE SETS 93

points altogether, and of course also 31 lines. The appropriate scheme uses
symbols P0, Px,. . P30 for the 31 points, and /0, lx,. . ., /30 for the 31 lines,
with a table (page 94) telling us which are the 6 points on each line and which
are the 6 lines through each point.
For good measure, this table gives every relation of incidence twice: each
column tells us which points lie on a line and also which lines pass through a
point; e.g., the last column says that the line /„ contains the six points
P0, Pi) P» P8 P12) > Pia
and that the point P0 belongs to the six lines

4> 4 4 12
’ > ^ > ll8-

Thus the notation exhibits a polarity Pr<-> lr. Marshall Hall [Cyclicprojective
planes, Duke Math. J. 14 (1947), pp. 1079-1090] has proved that such a
polarity always occurs. (Our lr is his m_r.) By regarding the subscripts as
residues modulo 31, so that r + 31 has the same significance as r itself, we
can condense the whole table into the simple statement that the point Pr and
line lg are incident if and only if
10.21 r + s = 0, 1, 3, 8, 12, or 18 (mod 31).
The “congruence” a = b (mod n) is a convenient abbreviation for the
statement that a and b leave the same remainder (or “residue”) when
divided by n. The residues 0, 1, 3, 8, 12, 18 (mod 31) are said to form a
“perfect difference set” because every possible residue except zero (namely,
1, 2, 3, . . ., 30) is uniquely expressible as the difference between two of these
special residues:
1 = 1-0, 2 = 3-1, 3 = 3 — 0, 4 = 12-8, •••,
13 = 0 — 18,. . ., 30 = 0 — 1.
The impossibility of a PG(2, 6) is related in a subtle manner to the
impossibility of solving Euler’s famous problem of the 36 officers [Reference
2, p. 190]. '

EXERCISES

1. Set up a table of differences (mod 31) of the residues 0, 1, 3, 8, 12, 18,

analogous to the following table of differences (mod 13) of 0, 1, 3, 9:

0 1 3 9

0 0 1 3 9
1 12 0 2 8
3 10 11 0 6
9 4 5 7 0
 o O CO 00 CN oo

o O CN r-
co

<N ON O VO o VO
CN cn T 1 1—11
co 00 ON O »o ON to
(N <N

r- OO O 00
CN <N CO T ’
CO
c >0 vo r- ON CO
<D CN <N CN 1"H
r2
'o VO lO vo OO CN vo CN

_c
CN CN CN *“
r- to
o
I—
r-
(N
to
<N CN
03
oo co vo O o
CN <N CN ’

t3 ON <N cn to o CO ON
CN CN CN CO
C
c3
o _ <N ON CN 00
r—1 CN CN CN CN

_ O CO OO r-
CN (N CN CN

CN ON O CN r- o vo
CN CN CN
C
a OO ON
t—< CN
vo
CN
O
CO
to

r- OO O to ON
CN CN CN
c3
-C 00
*o VO r- ON CO
T—1 CN CN
*-*
-3
o 00 CO
3 VO to vo
T“t CN CN
CN

t"- in r- CN vo
CN CN
fi T-"
<0
> oo co vo
T—■(
»-H
CN
»o
CN
O

'5b
of ON CN co «o O O
CN CN CO

o o CN ON CO ON
</a CN CN CN
(U
_3 O CO OO CN 00
CN ""H r-H CN CN
1"H
> _
CN ON O CN r-
(N CN CN
3
co 00 ON 1—H vo O vo
IZ) CN T"H CN CN
O
a. r- OO o to On to
<N CN
Q-,
o
»o vo f" ON 00
jy (N CN
3
3 vo to vo 00 CO r- CO
y—4
H CN
*■*
CN

r- Tf- to r- CN vo CN
CN T—» CN

OO co Tj- VO _ to
CN r~' CN

ON CN cn in o O
CN CN

O _ <N ON CO ON
co

^ =o
94
 AXIOMS FOR THE PROJECTIVE PLANE 95

2. Set up an incidence table for PG(2, 3), assuming thatP,. and ls are incident
if and only if
r + s = 0, 1, 3, or 9 (mod 13).

10.3 Verifying the Axioms

The discussion in Section 3.1 indicates that the following seven axioms
suffice for the development of two-dimensional projective geometry:

Axiom 2.11 There exist a point and a line that are not incident.

Axiom 2.12 Every line is incident with at least three distinct points.

Axiom 2.13 Any two distinct points are incident with just one line.

Axiom 3.11 Any two distinct lines are incident with at least one point.

Axiom 2.17 The three diagonal points of a quadrangle are never collinear.

Axiom 2.32 If two triangles are perspective from a point, they are per¬
spective from a line.

Axiom 2.18 If a projectivity leaves invariant each of three distinct

points on a line, it leaves invariant every point on the line.

The fact that this is a logically consistent geometry can be established by

verifying all the axioms in one special case, such as PG(2, 5). Axioms 2.11
and 2.12 are obviously fulfilled. To verify Axioms 2.13 and 3.11, we observe
that any two residues are found together in just one column of the table, and
that any two columns contain just one common number. To check Axiom
2.17 for every complete quadrangle (or rather, for every one having P0 for a
vertex) is possible but tedious, so let us be content to take a single instance:
the diagonal points of P0P1P2P5 are

lo' 49 = 4 ■ 4 = Pu> 4 ■ 4o = p9*

For Axiom 2.32 (which is Desargues’s theorem), we shall again be content
to check a typical case: the triangles P2P3P4 and P28-P3oP29 are perspective
from P25 and from /25, as in Figure 10.3a (cf. Figure 2.3a). (When applied to a
finite geometry, such a “real”figure must of course be “taken with a grain of
salt”; for instance, it fails to indicate that P6 lies on /6, and P9 on /9.)
Axiom 2.18 is superseded by Theorem 3.51 because a harmonic net fills the
whole line. In fact, the harmonic net R(P0F1F18) contains the harmonic
sequence P0PiP3Pi2^8 ’ • • . To verify this, we use the procedure suggested by
Figure 3.5a, taking A, B, M, P, Q to be P0, P4, P18, P5, P30, so that C = P3,
96 A FINITE PROJECTIVE PLANE

D = P12, E = P8, F — P0 — A. Since there are only six points on the line,
the sequence is inevitably periodic: the five points

P0, Pit P3 P12 P»

> )

are repeated cyclically for ever. Instead of taking P and Q to be P5 and P30,
we could just as well have taken them to be any other pair of points on /13 or
lu or /16 or 4i or /25 (these being, with l0, the lines through P18); we would
still have obtained the same harmonic sequence.

EXERCISE

Set up an incidence table for PG(2, 2), assuming that Pr and ls are incident
if and only if
r + s = 0, 1, or 3 (mod 7).

Verify that this “geometry” satisfies all our two-dimensional axioms except
Axiom 2.17.

10.4 Involutions

Turning to Figure 2.4a, we observe that the sections of the quadrangles

P4P5P6P9, Pi4F15P16P19, P9P10PnPu by the line /„ yield the quadrangular and
harmonic relations

Q(W,„ PsPzPu), QiPuPsPzt PisPqPi)* bl(P12P18, P3P8).

 PROJECTIVITIES 97

The fundamental theorem 4.11 shows that every projectivity on l0 is

expressible in the form
W. A w*.
where i, j, k are any three distinct numbers selected from 0, 1, 3, 8, 12, 18.
Hence there are just 6 • 5 • 4 = 120 projectivities (including the identity).
Of these, as we shall see, 25 are involutions: 15 hyperbolic and 10 elliptic.
In fact, if i and j are any two of the six numbers, there is a hyperbolic involu¬
tion CPiPi)(.PjPj) which interchanges the remaining four numbers in pairs in a
definite way. The other two possible ways of pairing those four numbers
must each determine an elliptic involution which interchanges Pt and P,-.
For instance, the hyperbolic involution (P12Pi2)(P18P18), interchanging P3 and
P8, must also interchange P0 and P1; and is expressible as (P0Pi)(P3P8); but
both the involutions
(P1P8)(P0P3), (PoPsXPiPs)
interchange P12 and P18, and are therefore elliptic.

EXERCISES

1. In PG(2, 3), how many projectivities are there on a line ? How many of
them are involutions ? How many of the involutions are elliptic ?

2. In PG(2, q), the q3 — q projectivities on a line include just q2 involutions:

q(q + l)/2 hyperbolic and q(q — l)/2 elliptic.*

3. In PG(2, 3), the four points on a line form a harmonic set in every possible
order. (G. FanoX)

4. In PG(2, 5), any four distinct points on a line form a harmonic set in a
suitable order. In fact, H(AB, CD) if and only if the involution (AB)(CD) is
hyperbolic. In other words, each of the fifteen pairs of points on the line
induces a separation of the six into three mutually harmonic pairs. (W.L.
Edge, “31 -point geometry," Math. Gazette, 39 (1955), p. 114, section 3.)

10.5 CoUineations and Correlations

By 6.22 and 6.42, every projective collineation or projective correlation is

determined by its effect on a particular quadrangle, such as P0P1P2P5. The

* This problem, for q a prime, was solved more than a hundred years ago by J. A.
Serret, Corns d'algebre superieure. Tome 2, 3rd ed. (Gauthier-Villars: Paris, 1866), p. 355,
or 5th ed. (Paris, 1885), pp. 381-382.
t Giornale di Matematiche 30 (1892), p. 116.
98 A FINITE PROJECTIVE PLANE

collineation may transform P0 into any one of the 31 points, and P1 into any
one of the remaining 30. It may transform P2 int° any one of the 31 6 25
points not collinear with the first two. The number of points that lie on at
least one side of a given triangle is evidently 3 + (3 • 4) = 15; therefore the
number not on any side is 16. Hence PG(2, 5) admits altogether

31 • 30*25 • 16 = 372000

projective collineations, and the same number of projective correlations.

Of the 372000 projective collineations, 775 are of period 2 (see 6.32).
For, by 6.31, the number of harmonic homologies is

31 -25 = 775.

Apart from the identity, the two most obvious collineations are Pr —^► P5r
(of period 3, since 53 = 1 (mod 31)) and Pr-*Pr+1 (of period 31). The
criterion 6.11 assures us that they are projective. In fact, the corresponding
ranges of the former on P0Pi and P0P5 are related by the perspectivity

PqP\P^P%P\2^18 7\ P0P5P15P9P29P28

and the corresponding ranges of the latter on P0Pi and P4P2 are related by a
projectivity with axis P0P2:

A P0P2P20P11P17P7 A P1P2P4P97,i3JPi9-

EXERCISES

1. Express the collineation Pr P$r as a transformation of lines. What

happens to the incidence condition (10.21)?

2. How many projective collineations exist in PG(2, 3)? How many of them
are of period 2?

3. In PG(2, q), how many points are left invariant by

(i) an elation, (ii) a homology?

10.6 Conics

The most obvious correlation is, of course, Pr -> lr. To verify that it is
projective, we may use 6.41 in the form

P1P2PiPgP 13P19 A P„P29P28P9P5P15 A 444 h

Vl3 19-

Being a period 2, it is a polarity. Since P0 lies on /0, it is a hyperbolic polarity,

 SELF-POLAR TRIANGLES 99

and determines a conic. By 8.51 (Steiner’s construction), we see that the

number of points on a conic (in any finite projective plane) is equal to the
number of lines through a point, in the present case 6. By inspecting the
incidence table, or by halving the residues 0, 1,3, 8, 12, 18 (mod 31), we see
that the conic determined by the polarity Pr^> lr consists of the 6 points and
6 lines
PoP^P&P 9 P16 P175 hhhhhlhl-
The 6 lines are the tangents, By joining the 6 points in pairs, we obtain the

^2j =15 secants

4 PqPvii 4 = P&P\^ 4 = P()Py- 4 = PqP 45 I12= PqPki

44 PiPni 4s = Pi%Pn, 4s = PoPie, hi — P^Pin hi — PgPi6>

hi = P§P$ j hi — P&P 17^ hi — P&P\ii hi ~ PiPo, ho = P\P9-

It follows that the remaining 10 lines

hi hi hoi hi, hi, ho, ho, hi, hi, h29

are nonsecants, each containing an elliptic involution of conjugate points.
Any two conjugate points on a secant or nonsecant determine a self-
polar triangle. For instance, the secant lx, containing the hyperbolic involution
(PqPoXP^P^) or GP2JP30)(iVPn), is a common side of the two self-polar
triangles W30, P\PlP\\' These two triangles are of different types: of the
former, all three sides lx,h, ho are secants; but the sides l7 and llx of the latter
are nonsecants. We may conveniently speak of triangles of the first type and
second type, respectively. Since each of the 15 secants belongs to one self-
polar triangle of either type, there are altogether 5 triangles of the first type
and 15 of the second. (These properties of a conic are amusingly different
from what happens in real geometry, where the sides of a self-polar triangle
always consist of two secants and one nonsecant.)
There are, of course, many ways to express a given polarity by a symbol of
the form (A BC)(Pp); for example, the polarity Pr^> lr is (P1P2P30)(P34) or
(P1P7P11)(P34) or (P1P7P11)(P44)- Such symbols will enable us to find the
total number of polarities.
If ABC is given, there are 16 possible choices for P (not on any side) and 16
possible choices for p (not through a vertex), making 162 = 256 available
symbols (ABC)(Pp) for polarities in which ABC is self-polar. Since each of
the 16 lines contains 3 of the 16 points, just 48 of the 256 symbols have* P
lying on p, as in the case of (PiP7Pn)(P44)*
When the self-polar triangle is of the first type (with every side a secant) all
the six points on the conic are on sides of the triangle, P never lies on p, and
each hyperbolic polarity (with ABC of this type) is named 16 times by a
100 A FINITE PROJECTIVE PLANE

symbol (ABC)(Pp) with P not on p. When only one side is a secant, 2 of the 6
points are on this side and the remaining 4 are among the 16; therefore each
hyperbolic polarity (with ABC of the second type) is named 4 times with P on
p and 12 times with P not on p. Conversely, if P lies on p, ABC can only be of
the second type; therefore the number of such hyperbolic polarities (each
accounting for 4 of the 48 symbols) is 12. Since each hyperbolic polarity
(or conic) has 5 self-polar triangles of the first type and 15 of the second, the
number of hyperbolic polarities in which a given triangle ABC is of the first
type is one-third of 12, that is, 4. The total number of symbols (ABC)(Pp)
that denote hyperbolic polarities is thus
48 + 16 • 4 + 12 • 12 = 256.
Since we have accounted for all the available symbols,
There are no elliptic polaritu v in PG(2, 5).*
The total number of triangle s in PG(2, 5) can be found as follows. There are
31 choices for the first vertex, 30 for the second, and 31 — 6 = 25 for the
third; but the three vertices can be permuted in 3! = 6 ways. Hence the
number is
31 -30-25
-- = 31 • 125 = 3875.
6
We can easily deduce the number of conics. Each conic has 5 self-polar
triangles of the first type, and each triangle plays this role for 4 conics;
therefore the number of conics isf
31 -125-4
= 3100.
5
As an instance of 7.71, we observe that the collineation Pr->Pr+1 (or
lr -> lr_j) is the product of the polarities lr and Pr<-> lT_x, which may be
expressed as
(PzPbPzvliP Ji) and (P4P5P2 s)(P 3^2) •

EXERCISES

1. In PG(2, 5), express the collineation Pr —► P5r (or ls —► /5s+3) as the product
of two polarities.
2. Derive the number of conics in PG(2, 5) from the number of sets of 5
points, no 3 collinear.

* In other words, every polarity is hyperbolic. This is true not only in complex geometry
and in PG{2, 5) but in PG(2, q) for every q = pk. See P. Scherk, Can. Math. Bull., 2 (1959),
pp. 45-46, or Segre (Reference 15, pp. 266-268).
t Analogous reasoning shows that the number of conics in PG{2, q) is qh — q2. See B.
Segre, Le geometrie di Galois, Ann. Matematica (4), 48 (1959), pp. 1-96, especially p. 4.
 ACCESSIBILITY WITH RESPECT TO A POLARITY 101

3. In PG(2, 5), how many conics can be drawn through the vertices of a
given triangle? [Hint: Use 9.21.] How many triangles can be inscribed in a
given conic? (These results provide another method for determining the
total number of conics.)

4. How many conics exist in PG(2, 3)? In what sense does 9.21 remain valid
when there are only 4 points on a conic?

5. Does a conic in PG(2, 3) admit a self-polar triangle all of whose sides are
nonsecants?

6. In any projective plane, let us say that a point Q is accessible from a point
P if it is the harmonic conjugate of P with respect to some pair of distinct
points which are conjugate (to each other) in a given polarity. Then P is
accessible from itself; if Q is accessible from P, P is accessible from Q;
if Q is accessible from R, and R from P, Q is accessible from P. In other
words, the relation of accessibility is reflexive, symmetric, and transitive.

7. When the notion of accessibility (Exercise 6) is applied to a hyperbolic

polarity, which points are accessible from:
(i) a point on the conic,
(ii) an exterior point,
(iii) an interior point?

8. If the notion of accessibility could be applied to an elliptic polarity in

PG(2, q), how many points would be accessible from a given point P :
(i) on a line through P,
(ii) in the whole plane ?
(F. Bachmann, Aufbau der Geometrie aus dem Spiegelungsbegriff, Springer
Yerlag: Berlin, 1959, pp. 123-124.)

9. There are no elliptic polarities in PG(2, q).

 CHAPTER ELEVEN

Parallelism

Projective geometry was historically developed as a part of Eucli¬

dean geometry. . . . Von Staudt, for instance, still needed the
parallel axiom in his study of the foundations of projective geo¬
metry. Klein, in his work on non-Euclidean geometry, established
the independence of projective geometry from the theory of parallels
(in 1871). This opened the possibility of an independent foundation
for projective geometry. The pioneer work was done by Moritz
Pasch (in 1882).
D. J. Struik (1894— )
(Reference 17, p. 72.)

11.1 Is the Circle a Conic?

The attentive reader must have noticed that most of the conics appearing
in our figures look like something that has been familiar ever since he first
saw the full moon: they look like circles. This observation raises the important
question: Is the circle a conicl We can answer Yes as soon as we have found a
characteristic property of a conic that is also a property of the familiar circle.
One way to do this is to give a Euclidean definition for the pole-and-polar
relation with respect to a circle, and carry the discussion far enough to find
that this relation satisfies the projective definition for a polarity. A quicker
way is to give a Euclidean proof for the Braikenridge-MacLaurin construc¬
tion (Figure 9.2b). More precisely, we select five points on a circle and prove
that the unique conic that can be drawn through these points coincides with
the circle. For convenience we shall take the five points A, P, B, Q, C
102
 THE CIRCLE IS A CONIC 103

Figure 11.1a

(Figure 11.1a) to be five of the six vertices of a regular hexagon inscribed in

the circle.
The following simple proof is due to S. L. Greitzer. Let AQ and CP meet
in R (which is, of course, the center of the circle), let a variable diameter meet
AB in M, BC in N, and let PM meet QN in X. Since AB, being a median of
the equilateral triangle APR, is the perpendicular bisector of PR, and similarly
BC is the perpendicular bisector of QR, we have

f_XPA = f_MPA = /_ARM = /CQRN = f_NQR = AXQA.

By Euclid III.21 and 22, the locus of X is the circle APO. By the Braikenridge-
MacLaurin construction, the locus of X is the conic APBQC. Hence the
conic coincides with the circle.

EXERCISES

1. What happens when the diameter is parallel to AB ?

2. How does the proof have to be modified if A, P, B, Q, C are arbitrarily

placed on the circle?

11.2 Affine Space

The time has come for us to investigate the connection between projective
space, as determined by our axioms, and affine space, that is, the ordinary
space of elementary solid geometry, in which two coplanar lines, or a line
and a plane, or two planes, are said to be parallel if they do not meet. Instead
104 PARALLELISM

of deliberately modifying the customary notions of space as we did in Section

1.1, let us now exploit the powerful theory of parallelism as Euclid did in his
eleventh book. The idea of parallel lines leads immediately to that of a
parallelogram, and thence to ratios of distances along parallel lines, or on one
line (Reference 7, pp. 2, 115-128; Reference 8, pp. 175, 202, 222).
I. M. Yaglom, near the beginning of the second volume of his Geometric
Transformations (in Russian), shows two sketches of his house. The former
depicts his living-room on a sunny day. The window panes cast shadows on
the floor. The rectangles are distorted into parallelograms, but those that are
equal remain equal. Affine geometry deals with properties that are maintained
by such “parallel projection.” His other sketch is a night scene outside the
house. A lamp inside casts shadows of the same window panes on the lawn.
The rectangles are distorted into quadrangles of various sizes, but their
sides are still straight. Projective geometry deals with properties that are
maintained by such “central projection.”
One way of expressing the connection is to regard affine space as part of
projective space, namely, projective space minus one plane. For this purpose,
we specialize any one plane of the projective space and call it “the plane at
infinity.” (It is still, of course, a projective plane.) Two lines, or a line and a
plane, or two planes, are then said to be parallel if they meet on this special
plane. We soon see that parallelism, so defined, has all its familiar properties.
Another way is to begin with the affine space, regarding its properties as
known, and seeking certain figures which behave like the projective points
and lines. Such a representation of the projective plane is easy: we can use the
lines and planes through a fixed point O in the affine space. For instance, to
verify 2.22 we merely have to observe that any two distinct planes a and /?,
through O, have just one common line, namely the line a • /?. For projective
space, the appropriate figures are bundles and axial pencils, defined as
follows.
A bundle is the set of all lines and planes through a point.
An axial pencil is the set of all planes through a line.
When there is any possibility of confusion, the other kind of pencil (the set
of all lines that lie in a plane and pass through a point) is called a flat pencil.
In the terminology of Section 1.5, an axial pencil, like a flat pencil, is a
“one-dimensional form.” But a bundle is a combination of two two-
dimensional forms: the set of lines through a point (which is the space-dual
of the set of lines in a plane) and the set of planes through the same point
(which is the space-dual of the set of points in a plane). What makes these
forms useful in the present connection is that their description is precisely the
same in affine space as in projective space.
Clearly, a range is determined by any two of its points, and these may be
any two distinct points; a flat pencil is determined by any two of its lines, and
 THE BUNDLE IS DETERMINED BY 2 COPLANAR LINES 105

these may be any two lines that meet; an axial pencil is determined by any two
of its planes, and these may be any two planes that meet; finally, a bundle
(like a flat pencil) is determined by any two of its lines.

EXERCISE

A range, a flat pencil, and an axial pencil are three kinds of one-dimensional
form. Which two are space-duals of each other? Which one is its own dual
(that is, which one is “self-dual”)?

11.3 How Two Coplanar Lines Determine a Flat Pencil and a Bundle

If (in ordinary space) we are given a point P, and two coplanar lines a and b
whose point of intersection O is inconveniently far away, how can we construct
the line through P of the pencil or bundle determined by a and b ? If O were
available we could simply draw OP, but can we still locate this line without
using O ? If P is not in the plane ab, we merely have to draw the planes Pa and
Pb; these meet in a line p through P, which is the desired line of the bundle
(see Figure 11.3a). In a single symbol, the member through P (outside the
plane ab) is the line
p — Pa - Pb.

On the other hand, if P is in the plane ab, we can use an auxiliary point Q
outside the plane, locate the member q through Q (which is the line OQ), and
consider (as in Figure 11.3b) the line of intersection of the planes ab and Pq.
In other words, the line through P (of the bundle or pencil) is now

p — ab • Pq where q — Qa • Qb.

This construction owes its importance to the fact that it remains valid when a
and b are parallel, so that O does not exist! The lines a, b, q,p, which originally
passed through O, are now all parallel.
A practical model for this new situation is easily made by dividing a
rectangular card into four unequal strips (of widths roughly proportional to

Figure 11.3a Figure 11.3b

106 PARALLELISM

4 : 6 : 7 : 5, as in Figure 11.3c) by three cuts made half-way through the

thickness of the card. When suitably folded, this makes a solid version of
Figure 11.3b.
Since we can derivep without inquiring whether a and b are parallel or not,
we now feel justified in extending the meaning of the words pencil and
bundle so as to allow the determining lines a and b to be any two coplanar
lines. If a and b happen to be parallel, the bundle consists of all the lines and
planes parallel to them, and the pencil consists of all the lines parallel to them
in their own plane. Accordingly, we speak of a bundle of parallels and a (flat)
pencil of parallels.

q b a q p

Figure 11.3c

It must be remembered that two planes may be parallel to a line without

being parallel to each other. (For instance, in Figure 11.3b, the intersecting
planes Qa and Qb are both parallel to the line p.) Thus a bundle of parallels
contains a lot of lines, all parallel to one another, and a lot of planes, not all
parallel to one another but each containing two (and therefore infinitely many)
of the lines.
A familiar instance of a bundle of parallels is the set of all “vertical” lines
and planes. If we take a cosmic standpoint and insist that two vertical lines
are not strictly parallel but meet in the center of the earth, then we have an
ordinary bundle instead of a bundle of parallels.

EXERCISE

Two lines parallel to the same line are parallel to each other. Does this
remain true
(i) when the word “lines” is replaced by “planes,”
(ii) when the word “line” is replaced by “plane,”
(iii) when both substitutions are made simultaneously?

11.4 How Two Planes Determine an Axial Pencil

If we are given a point P, and two planes a and /3 whose line of intersection
is far away, how can we construct the member through P of the axial pencil
determined by a and /?? This can be done by means of two applications of
 THE AXIAL PENCIL DETERMINED BY 2 PLANES 107

Figure 11.4a

the previous construction. We take any two intersecting lines ax and a2 in a,

and a point B in /3 (but not in either of the planes Pax, Pa2), and draw the lines
bl = Bax • (3, b2 = Ba2 • /?,
Px = Pax • Pbx, p2 = Pa2 • Pb2,
as in Figure 11.4a. Then the desired plane through P is pxp2. For, if oc and /?
meet in a line o, we may assume ax and a2 to be chosen so as to meet o in two
distinct points Ox and 02. Since Ox — o • ax lies in both the planes Bax and (3,
it lies on their common line bx. Since Ox lies in both the planes Pax and Pbx, it
lies on their common line px. Similarly 02 lies on p2. Therefore the join
o — 0X02 lies in the plane pxp2.
If, on the other hand, a and (3 are parallel planes (conveyed by parallelograms
in Figure 11.4a), the construction makes the lines bx and px parallel to ax, and
the lines b2 and p2 parallel to a2 \ therefore the plane pxp2 is parallel to a and [3.
Allowing P to take various positions, we thus obtain a pencil of parallel
planes, consisting of all the planes parallel to a given plane.
A familiar instance is the set of all “horizontal”planes. If we insist that two
horizontal planes are not strictly parallel but intersect in a line called the
“horizon,” then we have an ordinary axial pencil instead of a pencil of
parallel planes.

EXERCISE

Can two lines be parallel to the same plane without being coplanar?

11.5 The Language of Pencils and Bundles

We have seen that a bundle can be derived from two of its lines, and an
axial pencil from two of its planes, by constructions that remain valid when
the two lines or planes are parallel.
108 PARALLELISM

Since an ordinary bundle consists of all the lines and planes through a
point, and an ordinary axial pencil consists of all the planes through a line,
any simple statement about points and lines can be “translated” into a
corresponding statement about bundles and axial pencils. For instance, the
statement
Any two distinct points lie on a line
becomes:

The common planes of any two distinct bundles form an axial pencil.

It is significant that the latter statement remains true when one of the
bundles is replaced by a bundle of parallels, and again when both are replaced
by bundles of parallels. In fact, the common planes of an ordinary bundle and
a bundle of parallels form the ordinary axial pencil whose axis is the common
line of the two bundles. (For instance, the bundle of lines and planes through
any point O shares with the bundle of vertical lines and planes the vertical
line through O and all the planes through this line.) Again, the common
planes of two bundles of parallels form a pencil of parallel planes. (For
instance, the bundle whose lines are horizontal in the north-south direction
and the bundle whose lines are horizontal in the east-west direction have in
common all the horizontal planes.)

EXERCISE

Translate the following statement into the language of pencils and bundles:
If two distinct lines have a common point they lie in a plane.

11.6 The Plane at Infinity

These considerations serve to justify a convenient extension of space by the

invention of an “ideal” plane whose points and lines represent the bundles of
parallels and pencils of parallel planes, respectively. Remembering that an
ordinary bundle consists of all the lines and planes through an ordinary
point, we regard a bundle of parallels as consisting of all the lines and planes
through an ideal point. Similarly, we regard a pencil of parallel planes as
consisting of all the planes through an ideal line, and we say that an ideal
point lies on an ideal line if the bundle contains the pencil. We can still
assert that any two distinct points lie on a line. If one of the points is ordinary,
so is the line; but if both are ideal, the line is ideal.
Since an ordinary bundle contains no pair of parallel planes, an ordinary
point cannot lie on an ideal line; that is, all the “points” on an ideal line are
 IDEAL ELEMENTS 109

ideal points. On the other hand, since a bundle of parallels contains ordinary
axial pencils as well as pencils of parallel planes, an ideal point lies on some
ordinary lines as well as on some ideal lines. Since any ordinary line belongs to
just one bundle of parallels (consisting of all the lines and planes parallel to it),
it contains just one ideal point, which we call its point at infinity. Thus we
regard any two parallel lines as meeting in an ideal point: their common
point at infinity. Since any plane belongs to just one pencil of parallel planes
(consisting of all the planes parallel to it), it contains just one ideal line, which
we call its line at infinity. Thus we regard any two parallel planes as meeting in
an ideal line: their common line at infinity. In a given plane, each point on
the line at infinity is the “center”of a pencil of parallel lines.
Since any two pencils of parallel planes belong to a bundle of parallels,
Any two ideal lines meet in an ideal point.
It follows that, if a and b are any two ideal lines, every other ideal line
meets both a and b. This state of affairs resembles what happens in a plane.
For, if a and b are two ordinary intersecting lines, every point in the plane ab
lies on a line that meets both a and b. Accordingly, it is appropriate to regard
the set of all ideal points and ideal lines as forming an ideal plane: the plane
at infinity. This makes it possible to assert that any two intersecting (or
parallel) lines determine a plane through both of them. If one of the lines is
ordinary this is an ordinary plane; if both are ideal it is the plane at infinity.
Since each point (or line) at infinity is joined to an ordinary point O by an
ordinary line (or plane), the points and lines of the projective plane may
simply be regarded as a “new language” for the lines and planes (respectively)
through O. In other words,
The projective plane is faithfully represented by a bundle.
Historically, points at infinity, lines at infinity, and the plane at infinity
were first thought of by Kepler, Desargues, and Poncelet, respectively.

EXERCISE

Examine all the Axioms 2.11-2.18 for projective space, verifying that each
is satisfied in our “extended” space, that is, in the ordinary affine space plus
the plane at infinity with all its points and lines.

11.7 Euclidean Space

Although elementary solid geometry operates in affine space (Section 11.2),

we must not imagine that affine geometry is merely another name for
110 PARALLELISM

Euclidean geometry! Affine geometry is the part of Euclidean geometry in

which distances are compared only on the same line or on parallel lines.*
Affine geometry becomes Euclidean geometry as soon as we have said what we
mean by perpendicular. (For, right angles lead to circles and spheres, and
thus enable us to compare distances.)
We have already referred to the set of all vertical lines as a familiar instance
of a bundle of parallels, and to the set of all horizontal planes as a familiar
instance of a pencil of parallel planes. More generally, every bundle of
parallels in Euclidean space determines a unique axial pencil (of parallel
planes), whose planes are perpendicular to the lines and planes of the bundle;
and, conversely, every pencil of parallel planes determines a perpendicular
bundle (of parallels). In the language of the plane at infinity, we thus have a
special one-to-one correspondence between points at infinity and lines at
infinity. As we have already remarked, the plane at infinity is a projective
plane. Accordingly, we are not surprised to find that this correspondence
between its points and lines is a polarity (called the absolute polarity). A line
and a plane are perpendicular if the point at infinity on the line is the pole of
the line at infinity in the plane. Two lines (or two planes) are perpendicular if
their sections by the plane at infinity are conjugate points (or lines). Since no
line or plane is perpendicular to itself, the polarity is elliptic. In fact, just as
affine space can be derived from projective space by singling out a plane (“at
infinity”) and using it to define parallelism, so Euclidean space can be derived
from affine space by singling out an elliptic polarity in the plane at infinity
and using it to define perpendicularity.

EXERCISE

Let P and Q be two distinct points in Euclidean space. What is the locus of
the point of intersection of a variable line through P and the perpendicular
plane through Q ?

* Melvin Hausner, “The Center of Mass and Affine Geometry,” Am. Math. Monthly,
69 (1962), p. 730.
 CHAPTER TWELVE

Coordinates

In any system of two-dimensional and homogeneous analytical

geometry a point is a class of triads (x, y, z), those triads being
classified together whose coordinates are proportional. . . . This is
a very obvious observation, but it is of fundamental importance,
since it marks the most essential difference between analytical geom¬
etries and “pure” geometries. . . . There are no axioms in any
analytical geometry. An analytical geometry consists entirely of
definitions and theorems.
G. H. Hardy
(“What is Geometry?” Math. Gazette, 12 (1925), p. 313.)

12.1 The Idea of Analytic Geometry

Analytic geometry gives geometric names to certain sets of numbers in

such a way that each geometric theorem is reduced to an algebraic theorem.
Often the algebraic theorem is easy to prove, whereas the search for a “pure
geometric” proof requires considerable ingenuity. For this reason many
mathematicians prefer the algebraic method, thereby running the risk of
producing a generation of students for whom a conic (for instance) is nothing
more than a certain kind of quadratic equation. There is much to be said for
an unbiassed attitude, in which each problem is solved by whichever method
seems to work better. In earlier chapters we have concentrated on the pure or
“synthetic” method; but now we restore the balance by showing how some of
the same results can be obtained analytically.
For Euclidean geometry it is natural to use the classical “non-homogeneous”
coordinates of Fermat, Descartes, and Newton, which may be illustrated by
111
112 COORDINATES

the description of a point in ordinary space (with reference to a chosen

origin) as being at distances xx east, x2 north, and x3 up. For projective
geometry it is more convenient to use the homogeneous coordinates of
Mobius, Grassmann, and Pliicker, which may be illustrated by the description
of a point in a plane (with reference to a triangle AXA2A3) as being at the
center of gravity of masses x1 at Ax, x2 at A2, and x3 at A3. However, such
illustrations are not at all essential; the important idea is to take an ordered
set of numbers (xx, x2, x3) and call it a point.
The “numbers” that we use may for simplicity be thought of as real
numbers, but actually they can be the elements of any commutative field
(Reference 15, p. 21) in which 1 + 1 ^ 0; in particular, they can form a
finite field, and this throws light on the subject of Chapter 10.
In order to be able to interpret lines as well as points, we consider two
types of ordered triads of numbers, say (x1( x2, x3) and [Xx, X2, A"3].
We agree to exclude the “trivial” triads (0, 0, 0), [0, 0, 0], and to regard
two triads of the same type as being equivalent (that is, geometrically
indistinguishable) if they are proportional; thus (xl5 x2, x3) is equivalent to
(Xxx, Xx2, 2x3), and [Xx, X2, X3] is equivalent to [XXx, XX2, XX3], for any
X ^ 0. With two triads of opposite types we associate a single number, their
“inner product”
(12.11) {xX} = {Ax} = JS^Xi + A2x2 + X3x3,
which may be zero.
EXERCISE

In Section 11.2, we considered the possibility of representing the points and

lines of the projective plane by the lines and planes through a fixed point O in
affine space. How is this representation related to homogeneous and non-
homogeneous coordinates?

12.2 Definitions

Setting aside all ideas of measurement, let us now build up the analytic
geometry of the projective plane in the manner proposed by Hardy, that is, as
consisting entirely of definitions and theorems, beginning with the definitions
of point, line, and incidence.
A point is the set of all triads equivalent to a given triad (xx, x2, x3). In
other words, a point is an ordered set of three numbers (xx, x2, x3), not all
zero, with the understanding that (2xx, Xx2, 2x3) is the same point for any
nonzero X. For instance, (2, 3, 6) is a point, and (——1) is another
way of writing the same point.
A line is the set of all triads equivalent to a given triad [Xx, X2, X3]. In
 HOMOGENEOUS LINEAR EQUATIONS 113

other words, a line is defined in the same manner as a point, but with square
brackets instead of ordinary parentheses, and with capital letters to represent
the three numbers. Thus [3, 2, -2] is a line, and [-1, -f, f] is the same line.
(We shall find that the point (xls x2, x3) and the line [x1; x2, x3] (with the
same x s) are related by a polarity, but the apparently special role of this
polarity is a notational accident.)
The point (x) and line [X], meaning (xl5 x2, x3) and [Xlt X2, X3], are said to
be incident (the point lying on the line and the line passing through the point)
if and only if
{xJT} = 0

in the notation of (12.11). For instance, (2, 3, 6) lies on [3, 2, —2], It follows
that any discussion can be dualized by interchanging small and capital letters,
round and square brackets.
The three numbers x{ are called the coordinates (or “homogeneous
coordinates,” or “projective coordinates”) of the point (x). The three numbers
Xi are called the coordinates (or “line coordinates,” or “envelope coordi¬
nates,” or “tangential coordinates”) of the line [X].
If (x) is a variable point on a fixed line [X], we call {Tx} = 0 the equation
of the line [X], because it is a characteristic property of points on the line.
For instance, the line [3, 2, —2] has the equation

3xx + 2x2 — 2x3 = 0, or 3xx + 2x2 = 2x3.

Dually, if [X] is a variable line through a fixed point (x), we call {x2f} = 0
(which is the same as {Tx} = 0) the equation of the point (x), because it is a
characteristic property of lines through the point. For instance, the point
(2, 3, 6) has the equation
2Xx + 3*2 + 6X3 = 0,

and the point (1, 0, 0) has the equation Xx — 0. Thus the coordinates of a
line or point are the coefficients in its equation (with zero for any missing
term).
The three points (1, 0, 0), (0, 1, 0), (0, 0, 1), or Xt — 0 (/ = 1, 2, 3), and
the three lines [1,0, 0], [0, 1, 0], [0, 0, 1], or x, = 0, are evidently the vertices
and sides of a triangle. We call this the triangle of reference (see Figure 12.2a).
The point (1, 1, 1) and line [1, 1, 1] are called the unit point and unit line. We
shall find, in Section 12.4, that there is nothing geometrically special about
this triangle and point and line, except that the point does not lie on a side,
the line does not pass through a vertex, and the point is the trilinear pole of
the line (see Section 3.4).
By eliminating Xx, X2> X3 from the equations

(xi) = 0, {yX} = 0, {zX} = 0

114 COORDINATES

Figure 12.2 a

of three given points (x), (y), (z), we find the necessary and sufficient condition

(12.21) yi y2 y3 = o

for the three points to be collinear. This condition is equivalent to the

existence of numbers X, /u, v, not all zero, such that

Xxi + (xyt + vZi = 0 (i = 1, 2, 3).

If (y) and (z) are distinct points, X # 0. Hence the general point collinear
with (y) and (z) is {Juy1 + vzx, fiy2 + vz2, fxyz + vz3) or, briefly,

(,uy + vz)
where /j, and v are not both zero.
When v = 0, this is the point (y) itself. For any other position, since (vz) is
the same point as (z), we can allow the coordinates of (z) to absorb the v, and
the point collinear with (y) and (z) is simply

(my + z).
If we are concerned with only one such point, we may allow the ju to be
absorbed too; thus three distinct collinear points may be expressed as (y), (z),
 LINEAR DEPENDENCE 115

(y + z)• However, this last simplification cannot be effected simultaneously

on two lines if thereby one point would have to absorb two different param¬
eters. The symbol (yy + z) can be made to include every point on the line
(y)(z) if we accept the convention that the point (y) is (yy + z) with y = ao.
Dually, the conditions for three lines [X], [7], [Z] to be concurrent is

Xi Xo X*

(12.22) Y1 Y3 0;
Z3

the general line concurrent with [ 7] and [Z] is [y Y + vZ]; and any particular
line concurrent with [ Y] and [Z], but distinct from them, may be expressed as
[Y + Z].

EXERCISES

1. Where does the unit line [1, 1, 1] meet the sides of the triangle of reference?

2. Copy Figure 12.2a (without the coordinate symbols) and add the lines
[0, 1, 1], [—1, 0, 1], [1, 1, 0]. Are these lines concurrent?

3. What line joins the points (1, 1, 1) and (1, —1, 0)? Where does it meet
[1,0, 0]?

4. The line joining (1, 0, 0) to (xx, x2, x3) is [0, x3, — x2]. Where does it
meet [1, 0, 0] ?

5. If the triangle of reference is the diagonal triangle of a quadrangle having

(1, 1, 1) for one vertex, where are the other three vertices? (See Exercise
1 of Section 3.3.)

6. The lines [Z] and (y)(z) meet in the point

({Xz}y - {Zy}z).

[Hint: What is the condition for (yy + z) to lie on [X] ?]

7. Obtain coordinates for the various points and lines in Figure 3.4a, begin¬
ning with
A = (1,0,0), B = (0,1,0), C = (0,0,1),

S = (*!, x2, X3).

Deduce the condition xxZx = x2X2 = x3X3 for the point (x) and line [Z]
to be trilinear pole and polar with respect to the triangle of reference.

8. Where is the harmonic conjugate of (xl5 x2, 0) with respect to (1, 0, 0) and
(0,1,0)?
116 COORDINATES

12.3 Verifying the Axioms for the Projective Plane

To show that this analytic geometry provides a model for the synthetic
geometry developed in earlier chapters, we must verify that Axioms 2.11,2.12,
2.13, 3.11, 2.17, 2.32, and 2.18 (page 95) are all satisfied.
The first four are easy. The point (1, 0, 0) and line [1, 0, 0] are certainly not
incident. A line [X] with X^Xa ^ 0 is incident with three points such as

(0, V3, - X2), (- X3, 0, XJ, (X2, - Xl5 0 );

[0, X2, X3] is incident with (0, X3, —X2), (1, X3, — X2), (1, 0, 0); and so on.
Two points (y) and (z) are joined by the line (12.21) or

" 72 73 73 71 7i 72 '
5
•^2 Z3 Z3 Z1 Zi z2 _

Two lines [7] and [Z] meet in the point (12.22) or

Y2 Y3 y, Y1 Y2

Z2 Z3 Z1 z2
To prove Axiom 2.17, we consider a quadrangle PQRS whose first three
vertices (p), (q), (r) satisfy

Pi P2 Pz
(12.31) <h ^2 ?3 ^0.
fi r2 r3

Since the side PS joins (p) to the diagonal point A — QR • PS, we may take A
(on QR, but distinct from Q and R) to be (q + r), and S (on PA, but distinct
from P and A) to be (p +
q + r), meaning

(Pi + qi + rx, p2 + q2 + r2, p3 + q3 + r3).

Then B, on both RP and QS, must be (r +
p), and C, on both PQ and RS,
must be (p +
q), The three diagonal points A, B, C are noncollinear since

qi + rx q% + r2 q% + r3 Pi P2 Pz
(12.32) fi + Pi r2 + p2 rz + Pz = 2 qi qz qz 7^0.

Pi + qi Pz + q* Pz + qz ri rz

As for Axiom 2.32 (i.e., Desargues’s theorem), let the first triangle PQR
and the center of perspective O be (p)(q){r) and (u). There is no loss of
 THE FUNDAMENTAL THEOREM 117

generality in taking the second triangle P'Q'R' to be

(p + u)(q + u)(r + u).

The point D = QR • Q'R', being collinear with (q) and (r) and also with
(q + u) and {r + «), can only be (q — r). Similarly, E is (r — p), and F is
(p — q)- These points D, E, F are collinear since

q1 ~ ri qz~ r2 q3 - r3

ri — Pi rz— p3 r3 — p3 = o
Pi ~ qi Pz q% Pz — qz
or, more simply, since

(3i ~ ri) + (ri - Pi) + (pt - qd = o a =1,2, 3).

When a range of points P arises as a section of a pencil of lines p, the

“elementary correspondence” P a p may be described as relating three
positions of P, say
(z), O’ + z)
to three positions of p, say

in [Z], [Y + Z].

From the information that P and p are incident in these three cases, can we
deduce that, when P is (juy + z),p is [pY + Z] with the same pi Yes! Since

\{(Y+ Z)(y + z)} = {7y} + {Yz} + {Zy} + {Zz},

the three given incidences imply

{ Yy} = 0, {Zz} - 0, { Yz} + {Zy} - 0,

whence
{(p Y + Z)(py + z)} = p2{ Yy} + p{{ Yz} + {Zy}) + {Zz} = 0,

showing that the line [pY + Z] is indeed incident with the point (juy + z).
Repeated application of this result shows that the relation

O0(z)(y + z)(py + z) a [Y][Z][Y + Z][pY + Z]

holds not only for an elementary correspondence but for any projectivity
from a range to a pencil; and of course we have also

(y)(z)(y + z)(py + z) A (/)0O(/ + z%“/ +

[Y][Z][Y + Z][pY + Z] a [Y')[Z']IY'+ Z'][pY' + Z'].
118 COORDINATES

This is the algebraic version of the Fundamental Theorem 4.12, from which
Axiom 2.18 can be deduced as a special case.

It is interesting to compare this verification of the axioms with what we did

in Section 10.3. An important difference is that, whereas PG{2, 5) is a single
(categorical) geometry, the analytic geometry that we are discussing now has
the same degree of freedom as the synthetic geometry itself. More advanced
treatises give synthetic proofs that the points on a line can be “added” and
“multiplied” so as to constitute the elements of a field. (Chapters 7 and 11 of
Reference 7 remain valid when all references to order and continuity have
been omitted.) In other words, projective geometry becomes categorical as
soon as the field of coordinates has been specified.

EXERCISES

1. The harmonic conjugate of (y + z) with respect to (y) and (z) is {y — z).

2. Where is the harmonic conjugate of (yy + z) with respect to (y) and (z)?
Dualize this result.

3. The hyperbolic involution with invariant points (y) and (z) is

OT + z) A {—py + z).
4. Any projectivity on the line (y)(z) is expressible in the form

(yy + z) a cu'y + z)>

where p = (ccp + 0)l(yp + (3), a<5 ^

5. Give an equation (or equations) for the general projectivity on the line
[0, 0, 1].
6. Under what circumstances will the projectivity described in Exercise 4 or 5
be:
(i) an involution, (ii) parabolic?
7. When four distinct collinear points are expressed in the form
00> 00> (y + z), (juy + z),

the number p is called the cross ratio of the four points. There is an analogous
definition for the cross ratio of four concurrent lines. Two such tetrads are
projectively related if and only if they have the same cross ratio.

8. For two points (y), (z) and two lines [ F], [Z], the expression

{yr}{zz}
{yZ}{zT}
 CROSS RATIO 119

is equal to the cross ratio of the two given points with the points in which
their join meets the two lines.

9. The cross ratio of (a, 1, 0), (b, 1, 0), (c, 1, 0), (d, 1, 0) is

(a — c)(b — d)
(a — d)(b - c)

12.4 Projective Collineations

We have seen that the condition (12.21) makes the points (x), (y), (z)
collinear; conversely (12.31) makes (p), (<q), (r) noncollinear, so that they
form a triangle. This triangle enables us to describe the position of any point
by means of barycentric coordinates A, fx, v, which are the coefficients in the
expression (Xp + [xq -f vr). This is an obvious generalization of the expression
(p + <7 + r) used in the proof of Axiom 2.17 in Section 12.3. Points on a side
of the triangle can be included by allowing X/xv = 0, and when fx = v = 0
we have the point (p) itself.
When (p)(q)(r) and {p + ? + r) are the triangle of reference and unit point,
(Xp + p,q + vr) is (X, ju, v), and the barycentric coordinates are the same as
the ordinary coordinates.
The correspondence (x) ->■ (x'), where

(12.41)

transforms (X, fx, v) into the point (Xp -(- [xq + vr) which has these same
barycentric coordinates referred to (p)(q)(r) instead of (1, 0, 0)(0, 1, 0)(0, 0, 1).
Since we are assuming (12.31), so that the equations (12.41) can be solved
for the x’s in terms of the x,5s, this is a point-to-point correspondence as
described at the beginning of Section 6.1. Since the equation {X x } = 0
[see (12.11)] is equivalent to

(12.42) {Xp}x1 {X q}x2 + {X r}x3 0,

it is a collineation. Since it transforms (0, [x, v) into (jxq + vr), the collineation
is projective (see 6.11). Since it transforms the quadrangle

(1,0,0)(0, 1,0)(0, 0, 1)(1, 1, 1)

into {p)(q)(r)(p + q + r), which may be identified with any given quadrangle
120 COORDINATES

by a suitable choice of the p's and ^’s and r’s, it is the general projective
collineation (see 6.13).
This collineation, which shifts the points (x) to new positions (*'), is the
active or alibi aspect of the linear homogeneous transformation (12.41).
There is also a passive or alias aspect: a coordinate transformation that gives
a new name to each point. In fact, we may regard {p){q)(r) as a new triangle of
reference, with respect to which the point that we have been calling (*') has
coordinates
Oi, *2, *3),
whereas its coordinates with respect to the original triangle are, of course,
(x l9 X 2, X 3).

One practical consequence of the “alias” aspect is that, when seeking an

analytic proof for a theorem concerning a triangle and a point of general
position, we are justified in using the triangle of reference and unit point.
Similarly, for a theorem concerning a quadrangle, it is often convenient to
take the vertices to be (1, ±1, ±1), so that the six sides have equations
*i ± Xj = 0 (/ < j) and the diagonal triangle is the triangle of reference.
Dually, a given quadrilateral may be taken to have sides [1, ±1, ±1] and
vertices X{ ± X} = 0.
Of course, a collineation is not only a point-to-point transformation but
also a line-to-line transformation. The latter aspect of the collineation (12.41)
is, by (12.42),
*1 = IpX'}, X2 = {qX'}, Xz = {rX'}.

A more systematic notation for the same two sets of equations is

px i cilxx "T Ci2x2 + CizX2 — 5j CijXj (/= 1,2, 3),

(12.43)
(*Xj — CxjX j + c2jX 2 + cZjX 3 = 2 cijX'i (;= 1,2,3),
where per # 0 and, by (12.31),

C11 c12 c13

(12.44) C21 C22 C23 7^0

C31 C32 C33

(Reference 19, p. 187; Reference 17, pp. 68, 85). The preservation of incidence
is verified as follows:

P{X'x'} = p2 X’^ = SS CijX'iXj = XjX} = a{Xx}.

Since our coordinates are homogeneous, there are many occasions when
 INVARIANT POINTS OF A COLLINEATION 121

we can omit the p and a in (12.43), that is, set p = <7 = 1. However, it is
important to retain them when we are looking for invariant points or invariant
lines. The invariant points are naturally given by x\ — x, or

pXi = 2 cijXj (/= 1,2,3).

Eliminating the x’s from these three equations, we obtain

C11 P c12 c13

C21 C22 P C23

C31 C32 C33 P

Any root p of this characteristic equation makes the three equations for the
x’s consistent, and then we can solve any two of them to obtain the coordinates
of an invariant point.
For instance, the collineation

(12.45) px\ = xx, px'2 = x2, px'z = p~xx3 (/t^ 1)

has the characteristic equation (p — l)2(p — pr1) — 0. The double root

p = 1 yields the range of invariant points (xx, x2, 0), and the remaining root
p = pr1 yields the isolated invariant point (0, 0, 1). By 6.24 and 6.27, this
collineation is a homology with center (0, 0, 1) and axis [0, 0, 1].
Again, the collineation

(12.46) px\ = Xx + axxz, px\ = x2 + a2x3, px'3 = x3

has the characteristic equation (p — l)3 = 0. If a3 and a3 are not both zero,
the triple root p = 1 yields the range of invariant points (xx, x2, 0); there are
no others. By 6.24 and 6.26, this collineation is an elation with axis [0, 0, 1].
Since the equation a2x\ — axx 2 = 0 implies a2x3 — a3x2 = 0, there is an
invariant line (other than [0, 0, 1]) through the point (a1; a2, 0). Hence this
point is the center of the elation.

Comparing the two parts of (12.43), we see that the expression for the
homology (12.45) as a line-to-line transformation is

0X1 — X x, csX2 — X 2, oX3 — p 1X 3

or, taking 0 = 1 for convenience and solving,

(12.47) X\ = Xlt X\ = X2, X'3 = pX3.

Similarly, the elation (12.46), qua line-to-line transformation, is

(12.48) X\ = Xlt X\ = X2, X'3 = X3- axXx - a2X2.

122 COORDINATES

EXERCISES

1. Find the projective collineations that transform

(1,0, 0)(0, 1, 0)(0, 0, 1)(1, 1, 1)
into the following quadrangles:
(i) (1, 0, 0)(0, 1, 0)(0, 0, \){p, q, r),
(ii) (-1, 1, 1)(1, —1, 1)(U, —1)(1, 1,1),
(iii) (0, 1, 0)(0, 0, 1)(1, 0, 0)(1, 1, 1),
(iv) (0, 1, 0)(0, 0, 1)(1, 1, 1)(1» 0, 0).
2. Give equations for the inverse of the collineation (12.43).
3. Prove Desargues’s theorem as applied to the triangle of reference and
1)(1, l,r).
4. Prove Pappus’s theorem. [Hint: Use Exercise 2 of Section 4.4, taking
as triangle of reference and B1 = (p, 1, 1), B2 = (1, 1, r).]

12.5 Polarities

Since the product of two correlations (for example, a polarity and another
projective correlation) is a collineation, any given projective correlation can
be exhibited as the product of an arbitrary polarity and a suitable projective
collineation. The most convenient polarity for this purpose is the one that
transforms each point (or line) into the line (or point) that has the same
coordinates. (This correlation is obviously projective, and of period 2.)
Combining the general projective collineation (12.43) with the polarity that
interchanges X\ and x\, we obtain the general projective correlation in the
form
P i ^ilX i T" Ci2X2 d~“ G'3-U — ^ ^iixj 0*=1,2, 3),
(12.51)
- cljX i -)- c2jx 2 T~ c3jX 3 = 2 ci]X i 0= 1,2, 3),
where again the coefficients satisfy (12.44). Incidences are dualized in the
proper manner for a correlation, since

The projective correlation (12.51) is a polarity if it is equivalent to the

inverse correlation oX',- = 2 ci)Xi or (interchanging i and j)

This means that

 SYMMETRIC BILINEAR FORMS 123

with the same alp for all i and j, so that, since the cu are not all zero,

_ a _ (a? /cr\2 . a
ca Cji ( I cij, I l -1, — zti-
p \p/ \p/ p
The lower sign is inadmissible, as that would make cH = — c{j and

C11 C12 C13 0 C12 C31

C21 C22 C23 = ^12 0 C23

C31 C32 C33 C31 C23 0

Hence a = p and cH — ci3. In other words, a projective correlation is a

polarity if and only if the matrix of coefficients ci3- is symmetric.
The nature of a polarity is such that no confusion can be caused by omitting
the prime (as in Section 7.1) and writing simply

(12.52) Xi = 2ciixi (/=1,2,3),

where ci3- = cH and det (ci3) = A =£ 0. These equations give us the polar [X]
of a given point (x). Solving them, we obtain the pole (x) of a given line [A]
in the form
Axi = 'ZCiiXi (i = l, 2, 3),

where C{j is the cofac> r of cu in the determinant A.

Two points (x) and (y) are conjugate if (x) lies on the polar [F] of (y).
Since F, = 2 ci3_y3, the condition { Yx} = 0 or S F^Xj = 0 becomes

(12.53) ci3-Xiy3- = 0,

which we shall sometimes write in the abbreviated form

(xy) = 0.
Letting (x) vary, we see that this is the equation of the polar of (y). Dually, the
condition for lines [X] and [ F] to be conjugate, or the equation of the pole of
[F], is
(12.54) [XY] = 0,
where [XY] = 22 CijXl Y3.
As a particular case of (xjy) = 0, the condition for (0, 1, 0) and (0, 0, 1) to
be conjugate is c23 = 0. Thus the triangle of reference is self-polar if and
only if
C23 = C31 = C12 = 0-
By choosing any self-polar triangle as triangle of reference, we reduce a given
polarity to its canonical form
Xi = CjjX; (^11^22^33 # 0)
124 COORDINATES

or, more conveniently,

(12.55) X^ = axj, X^ = bx2, X3 = cx3 (abc 0).

This is the polarity (ABC)(Pp), where ABC is the triangle of reference, P is

(1, 1, 1), and p is [a, b, c],

EXERCISES

1. Prove 7.21, using the triangle of reference.

2. Prove 7.13, using the polarity (12.52) and the line [0, 0, 1].

3. Prove 7.61 (Hesse’s theorem), using the quadrilateral [1, ±1, ±1], whose
pairs of opposite vertices are

(±1,1,0), (0, ±1, 1), (±1,0,1).

4. Prove 7.31 (Chasles’s theorem), using the triangle of reference and the
general polarity.

5. Any two perspective triangles are polar triangles for some polarity.

6. In the terminology of Exercise 6 at the end of Chapter 10, a point (y) is

accessible from (z) [with respect to the polarity (12.52)] if and only if there
is a number /z ± 0 such that

(yy) = p\zz):
either (yy) and (zz) are both zero or their product is a nonzero square.
(In real geometry this means that (yy) and (zz) have the same sign. But in
complex geometry, since every complex number is a square, all non¬
self-conjugate points are mutually accessible.)

12.6 Conics

The condition for a point (x) to be self-conjugate for the polarity (12.52) is
(jo:) = , or
0

CllXl2 + C22X2 + C33-*32 + 2C23X2X3 + 2c X .X ± 2CnXxX2 = 0.

31 3 1

Dually, the condition for a line [X] to be self-conjugate is [XX] = 0, or

QiV + C * 22 22 + C AV ± 2C23X2X3 ± 2C31X3X1 + 2C12X1X2
33 = 0.
Hence every conic (locus or envelope) has such an equation. In particular
[using (12.55) instead of (12.52)] every conic for which the triangle of reference
is self-polar has an equation of the form
axf ± bx22 ± cx 32 = 0 or a^X* + b~xX? + c~xXf = 0.
 INDEFINITE QUADRATIC FORMS 125

We can now clarify the statement (in Section 8.1) that in some geometries
every polarity is hyperbolic, whereas other geometries admit elliptic polarities
too. The polarity (12.52) is hyperbolic or elliptic according as the equation
(xx) = 0 does or does not have a solution (other than x1 — x2 = x3 = 0).
The distinction depends on the coordinate field. If this is the field of complex
numbers, every such equation can be solved; for example, the equation

*12 + X2 + *32 = 0

is satisfied by (1, V — 1, 0). Over such a field, every polarity is hyperbolic. In

the case of the field of real numbers, on the other hand, the quadratic form
(xx) may be “definite,” in which case the polarity (for instance, X* — xf) is
elliptic. (The only solution of the above equation in real numbers is 0, 0, 0.)
Some particular equations represent conics regardless of the field. For
example, the equation

(12.61) Xj2 + x22 — x32 = 0,

being satisfied by (1, 0, 1), cannot fail to represent a conic.

Since the condition for the conic (xx) = 0 to pass through (1, 0, 0) is
cn = 0, the most general conic circumscribing the triangle of reference is

c23*2*3 + ^31*3*1 + Cl2XlX2 = 0 (c23^31c12 ^ 0).

The coordinate transformation

X1 C23X1’ X2 ~*" C31X2’ X3 ~^ C12X3

converts this into

(12.62) x2x3 + x3Xi + xxx2 = 0.

Thus, in any problem concerning a triangle and a circumscribed conic, the

conic can be expressed in this simple form. Working out the cofactors in the
determinant, we obtain the envelope equation

Xx2 + X22 + X32 - 2X2X3 - 2X3X1 - 2X1X2 = 0

or
Xi* db X2i ± x3t = 0.
Dually, a conic inscribed in the triangle of reference is

X2X3 + X3Xx + XxX2 = 0

or

(12.63) *1* ± *2 ± x3 =
126 COORDINATES

EXERCISES

1. Prove 8.11, using the conic x2x3 + *3*1 + xix2 = 0.

2. Prove 8.41, taking the points P, Q, R and tangents p, q to be (1, 0, 0),

(0, 0, 1), (1,1,1) and [0, 0,1], [1,0, 0],

3. Prove 8.51 (Steiner’s construction), using the projectively related lines

[/1, 1, 0] and [0, y, 1] through the fixed points (0, 0, 1) and (1, 0, 0).

4. Write down the envelope equations for

(i) x22 + 2*3*! == 0, (ii) x22 = x3xv

5. Use Exercise 7 of Section 12.2 for an algebraic solution to Exercise 1 of

Section 8.5.

6 . Given a conic (xx) = 0 and an exterior point (y) (see Section 8.1), describe,
both algebraically and geometrically, the locus of a variable point (x) such
that the line (x)(y) is self-conjugate. Dualize this result.

7. If the four tangents through two exterior points are all distinct, their
points of contact and the two exterior points all lie on a conic.

8. Given a conic (xx) = 0, let (z) be any exterior point. If (zz) does not
happen to be a square, divide all the coefficients ctj in (xx) by (zz). We now
have the same conic expressed in such a way that (zz) is a square (namely,
1). When the equation (xx) = 0 has been thus normalized, any point (y)
not on the conic is exterior or interior according as (yy) is or is not a square.
(In real geometry this means, according as (yy) is positive or negative. In
complex geometry we know already that there are no interior points be¬
cause there are no elliptic involutions.) Illustrate this criterion by applying
it to the conic (12.61).

9. In real geometry all interior points of a conic are mutually accessible, but
in rational geometry the number of classes of mutually accessible points is
infinite.

12.7 The Analytic Geometry of PG(2, 5)

It is proved in books on algebra that, if q is any power of a prime (such as

2, 3, 4, 5, 7, 8, 9, or 11), there is a field having just q elements. A famous
theorem [Reference 15, p. 94] tells us that a finite field must necessarily be
commutative. We remarked, in Section 12.1, that our coordinates can belong
to any such field, provided q is odd (that is, not a power of 2). This proviso
 THE USE OF A FINITE FIELD 127

comes from (12.32), where the determinant of the coordinates of A, B, C,

being twice the determinant of the coordinates of P, Q, R, must be zero if
2 = 0 (which is what happens in a field having 2k elements).
We saw, in Section 12.2, that all the points on the general line (y)(z) can be
expressed in the form
<j*y + z),
where ^ runs over all the elements of the field and the extra element oo, which
yields (y). Hence the field with q elements (where q is any power of an odd
prime) yields the finite projective plane with q + 1 points on each line, that is,
in the notation of Section 10.1, PG(2, q). (In an ^-dimensional geometry such
as PG(n, q), a point has n + 1 coordinates.)

In PG(2, q), the number of coordinate symbols (xx, x2, x3), with q possible
values for each xi? is qz. From this number we subtract 1, since the symbol
(0, 0, 0) has no geometric meaning. Moreover, each point (x) is the same as
(Ax) for q values of A, namely all the nonzero elements of the field. In this
way we see again that the number of points in the plane is

= q*+q + l.
q — 1
For each point (x) there is, of course, a corresponding line [x]; so the number
of lines is the same.
When q is an odd prime (and not the square or higher power of such a
prime), the elements of the field are simply the residue-classes modulo q.
For instance, when q = 5, they may be denoted by the familiar symbols

0, 1,2, 3, 4,

with the usual rules for addition and multiplication except that

l+4 = 2 + 3 = 0, 2 + 4 — 3 + 3 —- 1,
3 + 4 = 2, 4 + 4 = 3,
2-3 = 4-4= 1, 3-4 = 2,
2-4 = 3, 3-3 = 4.

These rules are quite natural, provided we interpret the symbols as follows:
0 means the set of all multiples of 5, in other words, all integers whose final
digit is 0 or 5; 1 means the residue-class that includes all the positive integers
whose final digit is 1 or 6, and so on. In a more familiar notation, such equa¬
tions as
3 + 4 = 3- 4 = 2
are “congruences”
3 + 4 = 3- 4 = 2 (mod 5).
128 COORDINATES

Some rather subtle considerations, involving the field with 125 elements,
provide a systematic method for assigning coordinates to the 31 points Pr
and 31 lines ls of Section 10.2. The results are as follows

/'-V
©
p
/(> = [1,0, 0],

II
Pi = (0,1, 0), 4 = [0, i, 0],
A = (1, 0, 0), 4 = [4, 0, 1],
A = (0, 1, 1), 4 = [i,i,0],
A = 0, 1,0), 4 - [4, 1, 1],
A = (4, 1, 1), 4 = [1,3, 2],
A = ( 1,2, 1), 4 = [0, 2, 1],
A = (1, 0, 2), 4 = [1,0, i],
A = (0, 1, 4), 4 [1,4, 0],
=

A = (1, 4, 0), 4 = [1,1,4],

Pi 0 = (1, 1, 1), 4o = [0, i, i].
Pii = (4, 0, 1), 4i = [1,0, 2],
Aa = (0, 2, 1), 4a = [1,2, 0],
71!

bo
As = (2,1, 0), t—-
II
CO

Pu = (2,l, 1), 44 = [2, 3, 1],

As = (2, 3, 1), 4s = [1,2, 3],
Pu = (1, 3, 1), 4« = [0,1,2],
At = (2, 0, 1), 47 = [2, 0, 1],
p
>

4s = [2, 1, 0],
II
00

P19 = (1,2,0), 49 = [2, 1, 3],

P2 o = (3, 1, 1), 4o = [1,2,1],
Ai = (l>3, 2), 4i= [1,3,1],
^22 = (1,2, 3), 42 = [1, 1,3].
As = 0,1, 2), 43 = [1,1.1],
^24 = 0,1,4), 44 = [3, 1, 1],
As = (2, 1, 3), 4s = [1,1,2],
As = (1, 1,3), 4b = [2,1,1],
A? = (3, 1, 2), 4 =
? [3> 2, 1 ],
As = (3, 2, 1), 4s = [1,4, 1],
As = 0,4, 1), 49 = [0, i, 4],
,§5

o'
cT

Ao = (i,o, l).
II

EXERCISES

1. Which points of PG(2, 5) lie on the line [1, 0, 0]?

2. Which lines pass through the point (1, 1, 1)?

 THE HESSIAN CONFIGURATION 129

3. Find equations for the collineations

(i) Pr+1> 4 4-15

(ii) Pr —*■ P5r, Is —>■ 4s+3-

4. Find equations for the polarity lr.

5. Which points lie on the conics

(i) xf + x22 + x32 = 0, (ii) Xl2 + x22 - x32 = 0,
(iii) x22 = *3*i> (iv) x2x3 -f- x3x4 + x4x2 = 0,
(V) xf ± x2* ± x3* = 0?

[Hint: 2~x = 3, 3"1 = 2, 4~x = 4.]

6. What condition must the coordinate field satisfy if there exists a configura¬
tion 83 (in the notation of Section 3.2) consisting of points A, B, C, D, E,
F, G, H with the following triads collinear:

ABD, BCE, CDF, DEG, EFH, FGA, GHB, HAC?

When this condition is satisfied, the 8 points occur in 4 pairs of “opposites”

whose joins AE, BF, CG, DH are concurrent, so that the complete figure
is a (94, 123). In the special case of the plane PG(2, 3), this (94, 123) is
derived from the whole plane (which is a 134) by omitting one line and the
four points that lie on it. [Hint: Take ABC as triangle of reference and

D = (1, 1, 0), E — (0, 1, 1), H = { 1,0, o>).

Deduce F = (1, 1, co -f- 1), G — (1, co + 1, co). Obtain an equation for a>
from the collinearity of FGA.]

1. In PG(2, q), as in real geometry (Exercise 9 of Section 12.6), all interior

points of a conic are mutually accessible. (We are still assuming q to be odd.)

12.8 Cartesian Coordinates

In Section 11.2 we saw how to derive affine space from projective space by
specializing one plane, calling it “the plane at infinity,” and then omitting it.
The synthetic treatment required three dimensions, but the analytic treatment
can be carried out just as easily in two dimensions. Accordingly, let us con¬
sider the affine plane, that is, the ordinary plane of elementary geometry, in
which two lines are said to be parallel if they do not meet. We regard this as
part of the projective plane, namely, the projective plane minus one line:
“the line at infinity.” Two lines are said to be parallel if they meet on this
special line. We soon see that parallelism, so defined, has all its familiar
properties. The apparent inconsistency, of saying that parallel lines meet and
130 COORDINATES

yet do not meet, is resolved by regarding the affine plane as being derived from
the projective plane by omitting the special line (and all its points) while
retaining the consequent concept of parallelism. This modification of pro¬
jective geometry is called affine geometry.

Figure 12.8a

When coordinates are used, it is convenient to take the line at infinity to be

[0, 0, 1] or x3 = 0, so that the points at infinity are just all the points (x) for
which the third coordinate is zero. The points of the affine plane are thus all
the points (x) for which the third coordinate is not zero. By a suitable multi¬
plication (if necessary ), any such point can be expressed in the form

Oi, 1),
which can be shortened to (x1? x2). The two numbers xx and x2 are called the
affine coordinates of the point. In other words, if x3 =£ 0, the point (xx, x2, x3)
of the projective plane can be regarded as the point (xjx3, x2/x3) of the
affine plane; and the equation of any locus can be made into the corresponding
equation in affine coordinates by setting x3 = 1. In particular, the line [X]
has the equation
Xlx\ + ^2x2 + = 0,
and a pencil of parallel lines is obtained by fixing X1 and X2 (or, more pre¬
cisely, the ratio X1:X2) while allowing X3 to take various values.
Let us see what has happened to the triangle of reference. Its first two sides
have become the coordinate axes, namely the x2-axis x^ = 0 (along which x2
varies) and the xraxis x2 = 0 (along which xx varies). The third side, x3 = 0,
is the line at infinity. The first two vertices are the points at infinity on the axes:
(1, 0, 0) on the xraxis and (0, 1, 0) on the x2-axis. The third vertex is the
origin (0, 0), where the two axes meet. (See Figure 12.8a.)
The homology (12.45) becomes a transformation of affine coordinates

(l2-8!) x\ = ftx-L, x'2 = /ux2

 THE AFFINE AND EUCLIDEAN PLANES 131

when we set x3 = x'a = l, which requires p = prl. Similarly, the elation

(12.46), with p = 1, becomes

(12.82) x\ = x3 + a3, x' ^ = x2 + a2.

In either case [since both (12.47) and (12.48) involve X\ = X1 and X'2 = X2],
every line is transformed into a parallel line; in other words, directions are
preserved. The homology leaves the origin invariant and multiplies the
coordinates of every point by p; we call this a central dilatation. The elation,
leaving no (proper) point invariant, is a translation (or “parallel displace¬
ment”). These two affine transformations enable us to define relative dis¬
tances along one line, or along parallel lines (Reference 7, pp. 117-125); but
affine geometry provides no comparison for distances in different directions.
A conic is called a hyperbola, a parabola, or an ellipse, according as the line
at infinity is a secant, a tangent, or a nonsecant. (This agrees with the classical
definitions, since a hyperbola “goes off to infinity” in two directions, a
parabola in one direction, and an ellipse not at all.) The pole of the line at
infinity is called the center of the conic. In the case of a hyperbola, this is an
exterior point (see Section 8.1), and the two tangents that can be drawn from
it are the asymptotes, whose points of contact are at infinity. Thus

xxx2 = 1 and x,2 — x2 = 1

are hyperbolas, x22 = xx is a parabola, and in real geometry,

(12.83) xx2 + x22 = 1

is an ellipse.
To pass from affine geometry to Euclidean geometry we select, among all
the ellipses centered at the origin, a particular one, and call it the unit
circle. This provides units of measurement in all directions. To pass from
affine coordinates to Cartesian coordinates we choose, as unit circle, the ellipse
(12.83) . The dilatation (12.81) transforms this into a circle of radius p:

(12.84) xx2 + x22 = p2.

The translation (12.82) then yields the general circle
(xx - Ci)2 + (x2 - a2f = p2.
The rest of the story is told in every textbook on analytic geometry.

EXERCISES

1. Give equations (in affine or Cartesian coordinates) for

(i) the half-turn about the origin;
(ii) the half-turn about the point (av a2). (See Exercise 4 of Section 6.3.)
132 COORDINATES

2. Describe the collineation (in Cartesian coordinates)

x\ = x1 cos a — x2 sin a, x'2 = x1 sin a + x2 cos

Consider its effect on the circle (12.84).]

3. Give equations (in Cartesian coordinates) for

(i) the reflection in the j^-axis;
(ii) the reflection in the line xt = x2.
Answers To Exercises

Section 1.2

1. (iv), (vi), (vii), (viii).

2. So that just one “ray” (from the lamp to some point on the rim) is parallel
to the wall.

3. (i) An ordinary point and a point at infinity are joined by just one line,
(ii) Each ordinary line contains just one point at infinity.

Section 1.3

2. Yes. Such apparently paradoxical behavior is characteristic of infinite sets.

Section 1.4

(i) The configuration formed by n points (no 3 collinear) and the

(2) = in(n ~
1)

lines that join them in pairs.

(ii) The configuration formed by n lines (no 3 concurrent) and their
points of intersection.

Section 1.5
1. So that distinct points X will yield distinct lines x.

3. In both these special cases, X" coincides with X.

133
134 ANSWERS TO EXERCISES

Section 1.6
1. In the notation of Figure 1.6e, ABC 7^ SRC 77 BAC.
p_ x
2. In the notation of Figure 3.3a, abc 77 src 77 bac.
3. Use the method of Figure 1.6b, taking A", B", C" to be three collinear
points on a, b, c, respectively.
4. Proceed as in Figure 1.6d with the names of the points B and D inter¬
changed.

Section 2.1
1. (i) Axioms 2.11 and 2.12 yield one point and three others.
(ii) If this statement were not true, all the points that exist would lie on a.
By Axiom 2.12, every line would contain two (in fact, three) of these
points. By Axiom 2.13, every line would coincide with a, that is, a
would be the only line, contradicting Axiom 2.11.
(iii) Let P and / be the point and line described in Axiom 2.11. If A is not
on /, / fulfills our requirement. If A is on /, Axiom 2.12 yields another
point B on /. By Axiom 2.13, a line is equally well determined by any
distinct two of its points. The line / = AB, not containing P, is
distinct from BP. Therefore BP is a line not passing through A.
(iv) Consider the point A. By Axiom 2.12, the line whose existence is
proved in (iii) contains three distinct points, each of which (by Axiom
2.13) can be joined to A by a line. The three lines so obtained are
distinct; for, if two of them, AB and AC, were coincident, the same
line could be called BC, contradicting the fact that BC does not pass
through A.
2. In the notation of Figure 5.1a,

AECF ^ SRCU x BDCF.

This projectivity AECF a BDCF has C and F as invariant points. By
Axiom 2.18, it has no other invariant point.
3. 2.17.

Section 2.2
1. (i) By 2.24, there exists a quadrangle PQRS having 6 distinct sides
p = PQ, q ~ PS, r = RS, s = QR, u = QS, w = PR.
The first 4 of these 6 lines form a quadrilateral pqrs having 6 distinct
vertices
p = p-q> Q= P'S, R = r- s, S — q • r, U = q ■ s, W = p- r
[Reference 8, pp. 231-232],
 ANSWERS TO EXERCISES 135

(ii) If there exists a quadrilateral pqrs whose diagonal lines

u= QS, v — UW, w = PR
are concurrent, it follows that the point u • w lies on v, contradicting
Axiom 2.17 (which tells that the diagonal points
U = q • s, V = u • w, W=p • r
of the quadrangle PQRS are not collinear).

Section 2.3
1. We have O = PP' ■ QQ' \ R' may be anywhere on the line OR.

3. If the triangles are congruent by translation, they are perspective from a

point at infinity. If they are merely homothetic (differing in size, but still
similar and similarly situated) they are perspective from an ordinary point.

Section 2.4
1. If AP, BQ, CR are concurrent, as in Figure 2.4a, we have immediately
Q(ABC, DEF). Conversely, if Q(ABC, DEE) while the lines AP, BQ, CR
are not known to be concurrent, define S = AP ■ BQ, C' = AB ■ RS.
The quadrangle PQRS yields Q(ABC', DEF). By Theorem 2.14, C' = C.
Therefore CR passes through S.

2. Let the diagonal triangles be LMN and L'M'N', so that L = PS • QR,.. .,

L' = P’S' • Q'R',.... The triangles PQL and P'Q'L’ are perspective from
g, since pairs of corresponding sides meet in D, A, F. Therefore they are
perspective from O = PP' • QQ'. Thus O lies on LL', and similarly on MM',
NN'.

Section 2.5
1. H(PQ, FG), H(RS, CG).

2. As in Figure 1.6e.
3. Of course, the concept of a point “inside” a triangle does not belong to
projective geometry. But what happens in the Euclidean plane is that the
line AB separates R from the other three vertices of the quadrangle.

4. Since F coincides with C, the four points reduce to three. Each of the three
is its own harmonic conjugate with respect to the other two!

5. C is the midpoint of AB.

6. Measuring from O, we have the distances 10, 12, 15 in “harmonic pro¬

gression.” (Their reciprocals are in “arithmetic progression.”) Measuring
from C, we have another harmonic progression 3, 5, 15.
136 ANSWERS TO EXERCISES

Section 3.1
2. Certain pairs of triangles are perspective from points P, Q, R, S, and con¬
sequently also from lines p, q, r, s. For instance, ABC and SRQ are per¬
spective from P.

Section 3.2
4. No, not in the projective geometry that we have defined. A 73 would
consist of a quadrangle with collinear diagonal points, as in Exercise 3 of
Section 2.1.

Section 3.3
1. We construct, in turn, A2 = BC ■ AS, B2 = CA• BS, Cx — AB • A2B2,
P = AS • CCX, Q = BS • CCX, R = CS • AQ. (Compare Exercise 2 of
Section 3.1.)

2. The quadrangle BCB1C1 tells us that CCX meets AAX in the harmonic
conjugate of R (with respect to A and Ax), which is Q. Similarly, the quad¬
rangle CAC1A1 tells us that CCX meets BBX in the harmonic conjugate of
R, which is P. Finally, the quadrangle ABAXBX tells us that Q (on AAX) and
P (on BBX) are harmonic conjugates with respect to C and Cx.

Section 3.4

1. The trilinear polar of a vertex is indeterminate: any line through this

vertex will serve. The trilinear polar of any other point on a side is that
side itself. The trilinear pole of a side is indeterminate: any point on this
side will serve. The trilinear pole of any other line through a vertex is that
vertex itself.

2. S.

3. a= QR,b = RP, c = PQ, d = AP, e — BQ,f = CR,p = BC, q = CA,

r = AB, s = DE.

4. The three medians are concurrent.

Section 3.5
1. Yes, M is the harmonic conjugate of B with respect to A and C.
2. Yes, since H{B'M, A'C'), H{C'M, B'D'),_

3. The points of the sequence are evenly spaced along the line (so that the
segments AB, BC, . . . , are all congruent).

Section 4.1

1. Use the method of Figure 4.1b with A' = B, B' = C, C = A.

 ANSWERS TO EXERCISES 137

2. Suppose F is transformed into G. Applying the same projectivity three

times, we have
ABCD a BCAE a CABF a ABCG.
By Axiom 2.18, G = D. In other words, this is a projectivity of period
three.

3. Theorem 1.63 tells us that any “double interchange,” (AB)(CD) or

(AC)(BD) or (AD)(BC), can be effected by a projectivity.

Section 4.2
1. Any point K of the harmonic net R(ABC) (or of the harmonic sequence
ABC . . .) is derived from ABC by a finite sequence of harmonic construc¬
tions, each of which is carried over by the projectivity ABC a A'B'C'.
Hence K is transformed into a corresponding point K' of the harmonic net
R(A'B'C') (or of the harmonic sequence A’B’C’. . .).

2. These three points

Q = PC- AyA, C2 = CS- AB, D = SAl- BP

are collinear, since H(7?S', CC2), H(/?i?, PD), and RSC a RBP.

3. A projectivity relating pencils through two distinct points is a perspectivity

if and only if it has an invariant line.

4. Use 4.21.

Section 4.3
Every projectivity relating pencils through two distinct points determines
another special point, the “center,” which lies on the join of the cross¬
intersections of any two pairs of corresponding lines. If aa’e is a triangle,
the center of the projectivity ae0e a a’ee’ is the point e0 ■ e'.

Section 4.4
1. Ciy by Cy 02 ^2 C2 fifg 63 Cg

A3 Ay A% A2 A3 Ay Ay A2 A3
By B.} B2 B3 B2 By B2 By B^
C2 C3 C2 Cy Cy Cy C3 C3 C3

2. Construct the points Cy = AyBy - A3B2, C2 = A2B2 • A3By (Figure 4.4b)

and then B3 = A2Cy - AyC2. The concurrence of AyB2, A2BX, A3B3 follows
from 4.41, applied to the hexagon AyB2CyA2ByC2.
138 ANSWERS TO EXERCISES

3. The diagonals AXB2, B3A3, A2BX all pass through C3.

4. If the six sides of a plane hexagon pass alternately through two points, the
three diagonals are concurrent. One such hexagon is axb2c1a2b1c2.

5. No. The hexagon ABPCSA1 is very special, as APS 77 AXBC and APS 7:

AXCB.
6. Suppose A1B1C2 is inscribed in A2B2C3, as in Figure 4.4b. Any point Cx on
A1B1 determines two further points

A3 = B1C2 • B2Cx, B3 = CXA2 • C2Ax,

which, by Pappus’s theorem, are collinear with C3.

7. 8 = A, 0 = B, 1 = C, 2 = A', 3 = B', 4 = C\ 5 = L, 6 = M, 7 = N.
This notation has the advantage that the incidences are maintained when
we replace each number x by either 2x or x + 3 and reduce modulo 9.
Applying the former transformation to the given triads of collinear points,
we obtain the whole set of nine triads:

801, 234, 567,

702, 468, 135,
504, 837, 261.

Applying it to the given cycle of Graves triangles, we obtain the whole

set of six cycles:
012, 345, 678,
024, 681, 357,
048, 372, 615,
087, 654, 321,
075, 318, 642,
051, 627, 384.

Section 5.1

1. ADBF ^ AVSP ADCE.

3. The line joining the centers of the two perspectivities would inevitably
yield an invariant point (such as Fin Figure 5.1a, or C in Figure 5.1b).

4. Our axioms (not being categorical) provide no answer to this question.

In real geometry, the projectivity ABC a BCA (Exercise 2 of Section 4.1)
provides an instance of an elliptic projectivity. But in complex geometry
every projectivity is either hyperbolic or parabolic (because every quad¬
ratic equation can be solved).
 ANSWERS TO EXERCISES 139

Section 5.2

The parabolic projectivity becomes a translation, shifting every point along

the line through the same distance AA'. The point at infinity, C, is evidently
the only invariant point.

Section 5.3
1. This is merely 5.34 in a more symmetrical notation. The remarks about
Desargues at the beginning of Section 5.3 suggest an analogous theorem
concerning six numbers a, b, c, d, e,f (which we may think of as distances
of the points from an arbitrary “origin”):

If a + b = d+ e and b + c = e + /, then c + d =/+ a.

2. The product leaves both M and N invariant, and transforms A into A'.

3. Watch what happens to A, A', B.

4. Let A be any noninvariant point. The given projectivity, not being an

involution, can be expressed as

AA'A" a A'A"A'",

where the three points on either side are all distinct. (Conceivably A'"
coincides with A, and then the second involution, like the first, is expressed
in terms of one of its invariant points.)

5. If B were invariant, so that B' = B, the symbol {AB')(BA') would become

(AB)(BA'), which is meaningless.

Section 5.4
1. The given statement says that C and D are the invariant points of an
involution that interchanges A and B.

2. If H(AB, MN), the fundamental theorem serves to identify the hyperbolic

projectivity AMN a BMN with the involution (MM)(NN).

3. The involution (MM)(NN) has the effect AB’MN a BA'MN, which shows
that MN is a pair of the involution (AA')(BB').

4. A'E'C'F' a AECF a BDCF a B'D'C'F'.

5. Since the involution ABCD a BAED has CE for one of its pairs, its
invariant points are D and D'. Therefore H(AB, DD').

6. The hint shows that the projectivity X a X' has only one invariant point,
namely B.
140 ANSWERS TO EXERCISES

7. By 4.21, there is a projectivity BCAD a ACBE. Since A and B are inter¬

changed, 5.31 shows that this projectivity is an involution. Since C is
one invariant point, 5.41 shows that F is the other. Since DE is a pair
of this involution, H{DE, CF). By 4.21 again, there is a projectivity
ABCF a DEFC. By 5.31 again, this is an involution. By 5.33, it follows
that Q(ABC, DEF).

Section 6.1
(i) This is a collineation of period 2, transforming the side PQ into itself
according to the hyperbolic involution (PQ){UU), where U = PQ • RS.
On the opposite side RS, every point is invariant.
(ii) This is a collineation of period 3, transforming the trilinear polar of S
into itself according to the projectivity DEF a EFD.
(iii) This is a collineation of period 4, leaving invariant the point PR • QS
and the line (£>/? • PS){PQ • RS).

Section 6.2
1. The center is PP' • QQ' and the axis joins D to PQ • P'Q'.

2. Hyperbolic or parabolic, according as the collineation is a homology or an

elation.

3. Since ABOO a A'B'OO, we have Q(AA'0, B'BO).

4. Clearly, each point on the axis o is invariant. If any other point A were
invariant too, the first elation would take A to some different point A', and
the second would take A' back to A: the two elations would be [o; A A']
and [o; A'-+A], whose product is the identity. Apart from this trivial
case, the product can only be an elation (not a homology). In fact,

[o-A-+B][o-B->C)= [o-A^C].

Alternatively, the dual result can be proved as follows: If two elations have
the same center O, they induce parabolic projectivities on any given line
through O. By 5.21, their product, inducing another parabolic projectivity
on this line, must itself be an elation (if it is not merely the identity).

5. Let A and B be the centers of two such elations, a and ft, which transform
P into P and P&. If A and B are distinct, and P is any point not on the
axis AB, the point AP? ■ BPa is both P^ and P^; therefore xft = ftx. If
A and B coincide, let y be an elation having a different center (but the
same axis), so that a commutes with both y and fty. Then afty = ftyx =
ft<x.y; therefore xft = ftx.
 ANSWERS TO EXERCISES 141

6. A perspective collineation; namely, an elation or a homology according as

the join of the points and the intersection of the lines are or are not in¬
cident.

7. If four lines do not form a quadrilateral, they can be described as three

lines through a point O and a fourth line o. These are, of course, invariant
for a perspective collineation with center O and axis o.

8. A dilatation [Reference 8, p. 68]; more specifically, a translation or a

central dilatation (like a photographic enlargement) according as the center
is a point at infinity or an ordinary point, that is, according as the perspec¬
tive collineation is an elation or a homology. An elation is called a shear
if its axis is an ordinary line while its center is the point at infinity on this
line.

Section 6.3
1. An elation. In fact, since the product is evidently a perspective collineation
having the same axis, we merely have to eliminate the possibility of a
homology. For this purpose we apply, to the line joining the two centers,
Exercise 6 of Section 5.4, which tells us that the product of two hyperbolic
involutions with a common invariant point is a parabolic projectivity.
2. In the notation of Figure 6.2a, the elation [o; P -> P'] may be expressed as
the product of harmonic homologies with centers P and Ox, Ox being the
harmonic conjugate of O with respect to P and P'. Knowing, from Exercise
1, that this product is some kind of elation, we merely have to observe that

the first homology leaves P invariant and the second takes P to P'.
3. The identity. In fact, choosing four points P, Q, R, S, on the first two sides
of the given triangle ABC, in such a way that H(BC, PQ) and H(CA, RS),
we find PQRS —*■ PQSR-+ QPSR —> PQRS.
4. Let Ox, 02, ox, o2 be the centers and axes of two harmonic homologies
whose product is a homology with center O and axis o. By Exercise 1,
0l ^ o2 (for, if ox and o2 were the same, the product would be an elation,
not a homology). Dually, Ox ^ 02. Any point P on o, being invariant
for the product, is interchanged with the same point P' by both the
harmonic homologies. Hence Ox and 02 are collinear with P and P .
Since P is arbitrary on o, it follows that o = 0X02. Dually, O — ox% o2.
Both the harmonic homologies induce on o an involution P a P whose
invariant points are Ox and o • ox, also 02 and o • o2. But Ox 7^ 02.
Therefore Ox = o • o2 and 02 = o • ox. Finally, by applying Exercise 3 to
the triangle 0X020, we see that the product of the harmonic homologies
with centers Ox, 02 and axes ox, o2 is the harmonic homology with center
O and axis o.
 142 ANSWERS TO EXERCISES

5. A half-turn (or rotation through 180°, or reflection in a point). Similarly, the

reflection in a line is a harmonic homology having the line as axis while its
center is at infinity in the perpendicular direction.

Section 6.4
Each diagonal point of the quadrangle is determined by two opposite sides,
and these are transformed into two opposite vertices of the quadrilateral.

Section 7.1
1. A self-conjugate line contains its pole. By 7.11, it cannot contain any other
self-conjugate point.

2. Y = x ■ b.

4. r • BC and s ■ BC.

5. No; a trilinear polarity is not a correlation, as it transforms collinear

points into non-concurrent lines. Moreover, it is not a one-to-one corre¬
spondence, as it transforms any point on a side of the triangle into that side
itself.

Section 7.2
1. Such a correlation induces on g the involution DEF a ABC. Thus we can
apply 7.21 to the triangle SAD (or SBE, or SCF).

2. Since P andp are self-conjugate, 7.11 shows that Q may be any other point
on p (but not on a side of the triangle ABC). For instance, we may take

Q ~P ' (a - AP){b • BP), q = P[A{a ■ p) ■ B(b • /?)].

Section 7.3
1. The pole of the line (p ■ QR)(q • RP) is the point P(q • r) • Q(r • p).

2. The polarity is (ABC)(Pp), where C = a ■ b.

Section 7.4
1. (a ■ p)[A(p • x) ■ P{a ■ x)] • (b ■ p)[B{p ■ x) ■ P(b • *)].

2. The pole of such a line WX is

B[AP • (a • PW)(p ■ 6)] • C[AP • {a • PX)(p • c)].

Section 7.5
1. Each vertex is the pole of the opposite side.

2. Use Exercise 1 of Section 7.4, with b, AP, BP, a replaced by q, r, s, t.

 ANSWERS TO EXERCISES 143

Section 7.6
Apply the dual of Hesse’s theorem to the quadrangle RSUV, in which RS = p
is conjugate to UV(through P) and RV — t is conjugate to SU (through T).

Section 7.7
1. Yes. In the notation of Section 6.1, Exercise (iii), the projective collineation
PQRS —>■ QRSP interchanges PQ ■ RS and QR • SP.
2. The projective collineation ABCP —> ABCP' (where neither P nor P' is on
a side of the triangle ABC) is the product of the two polarities

(.ABC){Pp) and (.ABC)(P'p),

where p is any line not through a vertex.

Section 8.1
1. Every point on any line is conjugate to the pole of the line.

2. The polar of p • q is PQ.

3. Yes. Any two tangents meet in a point that is exterior.

4. Yes. If an interior point exists, its polar exists and is neither a tangent nor
a secant.

5. PQR and pqr are polar triangles.

6. Consider the vertices on the side r of the circumscribed triangle pqr. Apply
8.13 to the conjugate lines (p • q) R and (p • q)(r • PQ).

7. (i) an ellipse (being finite),

(ii) a parabola (extending to infinity in one direction),
(iii) a hyperbola (extending to infinity in two directions),
(iv) the center,
(v) the asymptotes,
(vi) a diameter (bisecting a set of parallel chords),
(vii) conjugate diameters.
[For a complete account, see Reference 7, pp. 129-159.]

8. The line CD, joining the midpoint of the chord PQ to the pole of the line
PQ, is a diameter. Its two intersections with the parabola are the center
(at infinity) and S.

Section 8.2
1. By 8.21, QR meets SP in a point conjugate to B. But the only such point on
QR is A.
144 ANSWERS TO EXERCISES

2. Let AP meet the conic again in S. Then the diagonal triangle of the quad¬
rangle PQRS has the desired properties.

3. Let BQ and BR meet the conic again in S’ and P, respectively. By Exercise

1, A and B are two of the diagonal points of the quadrangle PQRS. The
third diagonal point C lies on the side PQ, as in Figure 8.2a.

4. Let A and BC be the center and axis of such a homology. By Exercise 3,

any secant through A behaves like the line QR in Figure 8.2a. Since
H(AAV QR), the homology interchanges Q and R.

5. The inscribed quadrangle PQRS and the circumscribed quadrilateral pqrs

have the same diagonal triangle.

6. The polarity (ABC)(Dd) induces, on PD, the involution (PP)(SS). There¬

fore P and S’ are self-conjugate. By Exercise 3 (applied to PS instead of QR),
Q and R also are self-conjugate.

Section 8.3
1. If a triangle is circumscribed about a conic, any point conjugate to one
vertex is joined to the other two vertices by conjugate lines.
Let a variable tangent of a conic meet two fixed tangents in X and Y;
then X a Y.

2. If we regard PT, PE, PA, PU as four positions of x, the corresponding

positions ofy are QT, QA, QB, QU.

Section 8.4

2. Any line h through R determines C, Cj, c, C2, and finally 5, the harmonic
conjugate of R with respect to C and C2. Accordingly, we construct c by
joining D to P(q • h) • Q(p ■ h). Then S' is the point where h meets either of
the lines
P(c ■ QR), Q(C ■ PR).
3. Dualizing Exercise 2, let p, q, r be the given tangents, P and Q the points of
contact of p and q, and H any point on r. Let PQ meet (p • QH)(q ■ PH) in
C. Then another tangent ^ joins H to either of the points
P ’ c(q ■ r), q • C(p • r).
4. Let P, Q, R be three points on the first conic, and P', Q’, R’ (orp', q, /) the
corresponding points (or tangents) of the second. Let D be the pole of PQ
for the first conic, D' the pole of P'Q' for the second (and d’ the polar of
p’ ■ q). In view of Exercises 2 and 3, we merely have to relate the quadrangle
PQRD by a projective collineation (or correlation) to the quadrangle
P'Q'R' D' (or to the quadrilateral p'q'r'd').
 ANSWERS TO EXERCISES 145

5. Retaining P, p, Q, q, but letting C vary on PQ, we observe that any such

point has the same polar for all the conics, namely DCl5 where D = p ■ q
and H(PQ, CCj). Exercise 4 of Section 8.2 shows that each conic is trans¬
formed into itself by the harmonic homology with center C and axis c.

Section 8.5
1. Let 5 = DE be the variable line through O. After using the hint, we observe
that, by 8.51 with x = PA and y — QB, the locus of S — x • y is a conic
through P, Q, R.

2. Defining y = QX, we have x a A a y- But PQ is an invariant line of the

projectivity x sj. Hence x tv y, and the locus is a line.

3. The line joins the points of contact of the remaining tangents from P and Q.

4. By comparing ranges on RP' and Q'R, we see that, if y — QL,

N_
X A M A L A y.
Therefore the locus of R' = x • y is a conic.

Section 9.1
1. AB,p, q are three tangents; P, Q are the points of contact of the last two.
Each point Z on PQ yields another tangent XY.

2. If EA, AB, BC, CD, DE are the five tangents, the point of contact of the
last lies on the line B(AD • CE).

3. A parabola (since is the line at infinity).

Section 9.2
1. Dualizing Exercise 2 of Section 9.1, we see that the tangent at P to the conic
PQRST joins P to the point
RS-(TP- QR)(ST • PQ).
2. An arbitrary line through P, say z, meets the conic again where it meets the
line
S[(PQ ■ s)(RS - z) - QR].
3. 45! = 60.

4. KPQVON (or KNOVQP).

5. Consider the dissection of the two triangles RAP and RCQ (Figure 11.1a)
into three quadrilaterals (each) by means of the perpendiculars x, y, z
from M to RA, AP, PR ; and x', /, z' from N to RC, CQ, QR. (The same
symbols will serve for the segments and their lengths.) Comparison of
146 ANSWERS TO EXERCISES

angles shows that the quadrilateral Axy is directly similar to Cy'x', and Pyz
to Qz'y'. Therefore

xx' = yy' = zz',

and

Hence the quadrilateral Rxz is directly similar to Rz'x, and the diagonal
MR of the former is along the same line as the diagonal RN of the latter.
(See also H. G. Forder, Higher Course Geometry, Cambridge University
Press, 1949, p. 13.)

Section 9.3
1. The desired point lies on the line

T[QR • (PQ • ST)(g • i?5)].

2. Of the conics through P, Q, R, S (Figure 9.3a), any one that touches g
does so at an invariant point of Desargues’s involution.

3. In this case, A is an invariant point of Desargues’s involution. Yes,

TEAU a TABU a UBAT.

4. Since t is a tangent, Desargues’s involution on it is hyperbolic. Another

conic arises from the second invariant point.

Section 9.4

2. This follows from 9.41 along with the latter half of 9.42.

3. Let PQR (Figure 9.4a) be the given triangle, and ST any secant of the
former conic that is a tangent of the latter. Let the remaining tangents
from S and T meet in U. By the dual of Exercise 2, U lies on the conic
PQRST.

4. PQRS and PQRT are two quadrangles inscribed in the conic PQRST. By
8.21, their diagonal triangles, ABC and A'B'C', are self-polar. By 9.42, the
six vertices of these two triangles lie on a conic.

Section 9.5

1. With respect to a degenerate conic consisting of two lines a and b, the

polar of a general point P is the harmonic conjugate of (a • b)P with respect
to a and 6; the polar of a general point on a is a itself; the polar of a ■ b is
 ANSWERS TO EXERCISES 147

indeterminate; the pole of a general line is a • b; the pole of a general line

through a • b is any point on the harmonic conjugate line; and the pole of a
is any point on a. The other kind of degenerate conic has the dual properties.
2. The other conic may be degenerate.

Section 10.1
1. q2 + q + 1; {q2 + 1 ){q2 + q + 1).
2. q\q + 1 )(q2 + q + l)/6.
3. n = q2, q d = q + 1.

Section 10.2
2. r 12 11 10 9 8 7 6 5 4 3 2 1 0

s 1 2 3 4 5 6 7 8 9 10 11 12 0
2 3 4 5 6 7 8 9 10 11 12 0 1
4 5 6 7 8 9 10 11 12 0 1 2 3
10 11 12 0 1 2 3 4 5 6 7 8 9

Section 10.3
r 6 5 4 3 2 1 0

1 2 3 4 5 6 0
s 2 3 4 5 6 0 1
4 5 6 0 1 2 3

The diagonal points of the quadrangle P2P4P5P6 are the three points on

Section 10.4
1. 24; 9; 3. If ^4, B, C, D are the four points on a line, the three elliptic
involutions are
(BC)(AD), (CA)(BD), (.AB)(CD).

2. By considering the possible effect of a given involution on three distinct

points A, B, C, we see that there are q involutions (AA)(BX), where
X^A,q—1 involutions (AB)(CY), where Y =£ A and Y ^ B, and
(q- 1)2 involutions (A Y)(BZ), where Z # A and Z # Y: altogether
q 4- (q — i) + (r — i)2 = q1
involutions. The number of hyperbolic involutions, such as (AA)(BB), is

obviously therefore the remaining are elliptic.

3. This follows from 2.51.

148 ANSWERS TO EXERCISES

Section 10.5
1. 4 4s+3- In fac4 if r + 5 = 0, 1, 3, 8, 12, or 18 (mod 31),
5r + (5s + 3) = 5(r + s) + 3 = 3, 8, 18, 12, 1 or 0.

2. 13 • 12 • 9 • 4 = 5616; 13 • 9 - 117.

3- q + 1; q + 2.

Section 10.6
1. Since this collineation has period 3, Figure 7.7b must be modified so that
1"' = / and therefore C = B", C' = B. Choosing A = P10 and / = /2, we
obtain the polarities

(^10^2 ^4 (^19 4 3
) . ), (P19P2 P )(.P 4
26 lo )-

2. Since 5 of the 6 points on a conic can be chosen in 6 ways, the number of

conics is
31 -30 -25 -16 -6
= 3100
5! 6
[Reference 15, p. 267].

16 • 6 3875 • 16
= 16; = 20; = 3100.
3! 20
4. 234. In PG(2, 3), any 5 points include 3 that are collinear; thus 9.21 holds
vacuously. In Section 9.5 we saw that any 5 points, no 4 collinear, deter¬
mine a conic (possibly degenerate). This result remains significant in
PG(2, 3), but the conic is necessarily degenerate.

5. Yes. The polarity Pr^ lr determines the conic P0P7PaPn which, regarded
as a quadrangle, has the diagonal triangle P4P10P12. The sides of this tri¬
angle (namely, /4, lw, l12) are nonsecants: the points on them are just all
the points in the plane except those that form the conic.

6. Let C be any point on the polar of P. Then CP is a pair of conjugate points

and H(CP,PP); therefore P is accessible from itself. Since the relation
H(CC4, PQ) implies H(CCX, QP), accessibility is symmetric. To prove that
accessibility is transitive, let R be accessible from Q on one line and from
P on another, so that H(AAlf QR) and H(P54, RP), where AAX and BBX
are suitable pairs of conjugate points. By Exercise 2 of Section 3.3, P and Q
are harmonic conjugates with respect to the two points

C = ABX ■ BAX and Cx = AB • AXBX

which, being opposite vertices of the quadrilateral ABCAXBXCX, are

conjugate (by 7.61: Hesse’s theorem).
 ANSWERS TO EXERCISES 149

7. Let Q be accessible from P; that is, let H(CCls PQ), where CQ is a suitable
pair of conjugate points. The harmonic homology with center C and axis
the polar c (through C\) interchanges P and Q while transforming the conic
into itself (by Exercise 4 of Section 8.2). Since any tangent through P or Q
is transformed into a tangent through Q or P, two mutually accessible
points are always of the same type: both on the conic, or both exterior, or
both interior. By 8.11, any two points on the conic are mutually accessible.
By Exercise 1 of Section 8.1, any two exterior points on a tangent are
mutually accessible. Let P and Q be any two exterior points whose join is
not a tangent, and let either of the tangents from P meet either of the tan¬
gents from Q in R. Since R is accessible from both P and Q, Exercise 6
shows thatP and Q are accessible from each other. Thus the three answers
are:
(i) all the points on the conic,
(ii) all the exterior points,
(iii) a set of interior points (possibly all, as we shall see in Section 12.6,
Exercise 9, and Section 12.7, Exercise 7).
8. (i) On a given line through P, there are (q + l)/2 points accessible from P:
one for each pair of conjugate points on the line. (Two such pairs could
not both yield the same point Q ; for then the involution of pairs of
conjugate points could be described as (PP)(QQ), whereas we know
that it is elliptic.)
(ii) On each of the q -f- 1 lines through P, there are (q + l)/2 points
accessible from P, namely P itself and (q — l)/2 others. Hence the
total number of points accessible from P is
1 + {q + m - 1 )/2 = (.q2 + 1)12.
9. We use reductio ad absurdum. Suppose, if possible, that an elliptic polarity
exists. By Exercise 6, accessibility is an “equivalence relation” (G. Birkholf
and S. MacLane, A Survey of Modern Algebra, 2nd ed., Macmillan, New
York, 1953, pp. 155-156). Therefore the q2 + q + 1 points in PG{2, q) can
be distributed into a certain number of classes, each consisting of (q2 + l)/2
mutually accessible points. Dividing q2 + q + 1 by (q2 + l)/2 (and
remembering that q is an odd number, greater than 1), we obtain the
quotient 2 and a remainder q, which is absurd.

Section 11.1
1. X coincides with P.
2. See Exercise 5 of Section 9.2.

Section 11.2
The range and the axial pencil; the flat pencil.
150 ANSWERS TO EXERCISES

Section 11.3
(i) No. (ii) No. (iii) Yes.

Section 11.4
Yes.

Section 11.5
If two axial pencils belong to a bundle, they have a common plane.

Section 11.6
(2.14) If AB and CD are either intersecting or parallel, then AC and BD are
either intersecting or parallel.
(2.16) If two planes are parallel, their common points form a line at infinity.

Section 11.7
The sphere on PQ as diameter.

Section 12.1
When the triad (x1; x2, x3) is interpreted as a point in affine space, represented
by Cartesian coordinates, the equivalent triads all lie on a line through the
origin, and the equation {Xx} = 0 represents a plane through the origin.
When we pass from affine space to the projective plane, it is thus natural
to use xl5 x2, x3 as homogeneous coordinates for a point, and to regard
{Xx} = 0 as the equation for a line.

Section 12.2
1. (0, 1,-1), (-1,0, 1), (1, —1, 0).
2. Yes, they all pass through (1, —1, 1).

3. [1,1,-2]; (0,2,1).

4. (0, x2, x3).

5. (-1, 1,1), (1,-1, 1), (1, 1,-1).

6. The condition for (uy + z) to lie on [Y] is p{Xy) + {Xz} = 0, or

B = —{Xz]l{Xy}.
7. AS = [0, *3, — x2], . . . , AD = [0, x2, x2],
P = (—*1, *2> *3)> • • • , D = (0, x2, —X3),
s= (l/xx, l/x2, l/x3).

8. F = (*!, — x2, 0).

 ANSWERS TO EXERCISES 151

Section 12.3
1. In the notation of Figure 2.5a, we may take
A = (y\ B = (z), C = (y + z), R = (*), Q = (x + y).
Then S, on both BQ and CR, must be (x + y + z); P, on both BR and
AS, must be (x + z); and F, on both AB and PQ, must be (y — z).
2. In Exercise 1, write py for y. The point (jxy — z) is, of course, the same as
(~PY + z). The harmonic conjugate of [juY + Z\ with respect to [F] and
[Z] is [—pY + Z].

3. This is the correspondence between harmonic conjugates with respect to

0) and (z).

4. Any projectivity on (y)(z) is expressible as (juy + z) a ipy' + z'), where

00 = (aF + yz), (z) = Q3y + <5z) and, to make these points distinct,
a<5 — fiy 0.

Since py' + z' = ju(a.y + yz) + fiy + dz = (a(i + (i)y + (y/u + S)z,

the point (jzy' + z') is (ju'y + z), where

ix = (a/* + fifcyp + 8).

x\ _ axj + /?x2
5. (a^ M
x'2 yx1 + <5x2
or px'-L = axj + /5x2, px'2 = yxx + dx2 (p # 0).

6. In the notation of Exercise 4, the relation between [x and p! may be ex¬

pressed in the form
ypp — cup F dp1 — = 0 (a$ =£ (3y).

(i) This is an involution if p and p are interchangeable, that is if —a = d.

In other words, the general involution is given by
app + b(jx + p) + c = 0 (b2 7^ ac).
(ii) Any invariant point is given by a root of the quadratic equation
yp2 — (a — d)p — p = 0; therefore the projectivity is parabolic if
(a — S)2 + 4 (3y = 0.
7. This is simply a restatement of the above “algebraic version of the funda¬
mental theorem.” In particular, the cross ratio of a harmonic set is —1.

8. Since [F] passes through (y + z), and [Z] through (py + z), we have
{ Yy} + { Yz} = 0, p{Zy] + {Zz} = 0,
and the given expression reduces at once to p.
152 ANSWERS TO EXERCISES

9. Taking (y), (z), [Y], [Z] to be (a, 1, 0), (b, 1, 0), [1, —c, 0], [1, — d, 0],
we have
{yY}{zZ} __ Qa — c)(b — d)
{yZ}{zY} (a — d)(b — c)

Section 12.4
1. (i) x\ = pxv x'2 = qx2, x'3 = rx3.
(ii) x\ = —xx + x2 + x3, x\ = xx — x2 + x3, x'3 = x1 + x2 — x3.
(iii) x'i = x3, x'2 = xv x'3 = x2.
(iv) x' = x3, x'2 = —xx + x3, x'3 = —x2 + x3.

2. pXi 2^ijX j

or, solving the former set of equations in terms of Cih the cofactor of ci}
in the determinant 12.44,
Pxj —
3. Corresponding sides of the two triangles meet in the three points

(0, q — l, 1 — r), (1 — p, 0, r — 1), (p — 1,1— q, 0),

which are collinear by the criterion 12.21 or, still more simply, by addition.

4. The lines AyBx and A3B2 meet in Cx = (1, 1, 1). Therefore B3 is (1, q, 1) for
a suitable value of q. The three lines A3BX, AXB3, A2B2, having equations

Xi = px2, x2 = qx3, x3 - rxx,

all pass through the same point C2 if pqr = 1. The three lines A3B3, A2Blt
AxB2, having equations

x2 = qxx, xx = px3, x3 = rx2,

all pass through the same point C3 if qpr = 1. These two conditions agree,
sincepq = qp. (The connection between Pappus’s theorem and the commu¬
tative law of multiplication is one of the many important discoveries of
David Hilbert, 1862-1943.)

Section 12.5
1. If the correlation 12.51 transforms (1,0,0) into [1,0,0], (0, 1,0) into
[0, 1, 0], and (0, 0, 1) into [0, 0, 1], we must have ctj — 0 whenever i j.
Therefore ci3- = cH, and the correlation is a polarity.

2. Setting [X] = [7] = [0, 0, 1] in 12.54, we see that the condition for this
line to be self-conjugate is C33 = 0, or

C11C22 C122 — 0.
 ANSWERS TO EXERCISES 153

Thus the hypothesis of 7.13 requires cuc22 — c122 ± 0. Setting (x) =

1, 0) and (y) = (p , 1, 0) in 12.53, we see that the condition for these
two points to be conjugate is

cnW' + c12(p + p') + c22 = 0.

By Exercise 6(i) of Section 12.3, this relation between p and p represents
an involution provided c12 2 # cnc22.

3. Setting p — 1 and p = — 1 in the above solution of Exercise 2, we see

that the condition for the opposite vertices (±1, 1, 0) to be conjugate is

C11 — c22-
Similarly, if (0, ±1, 1) are conjugate, c22 = c33. Hence cn = c33, and this
makes (±1,0, 1) conjugate.

4. The polars [cu, c21, c31], [c12, c22, c32), [c13, c23, c33] of the vertices meet the
respectively opposite sides [1, 0, 0], [0, 1, 0], [0, 0, 1] in the collinear points

(0, ^3i) ^12)) ( ^23) 0, c12), (c23, c3X,0).

5. The points {p, 1, 1), (1 ,q, 1), (1, 1, r) (see Exercise 4 of Section 12.4) are the
poles of the sides of the triangle of reference for the polarity

*1 = pXx + X2 + X3, x2 = Xx + aX2 ± X3, a:3 = Xx + X2 + rX3,

in which every C0 = 1 except Cn = p, C22 = q, C33 = r. (This converse

of Chasles’s theorem was discovered by von Staudt. The polarity operates
on the Desargues configuration in the manner described at the end of
Section 3.2 [Reference 7, pp. 74-75].)

6. The point (y) is accessible from (z) if and only if there exists a number p
such that the harmonic conjugates (y ± pz) are conjugate for the polarity.
Replacing x andy in the expression (xy) byy + pz andy — pz, we see that
the desired condition for conjugacy is

(yy) — p(yz) + p(zy) — p2(zz) = 0,

that is,
(yy) = p\zz)

[cf. Reference 7, pp. 37-38, 192].

Section 12.6
1. When (xx) = 2{pc2x3 + x3xx + xxx2), the condition for conjugacy (xy) = 0
becomes (x2y3 + y2x3) + (x3yx + y3Xi) + (xxy2 + yiX2) = 0. Setting
x3 = y3 = 0, we see that, on the secant [0, 0, 1 ], (xl5 x2, 0) is conjugate
to (xl5 —x2, 0) (see Exercise 8 of Section 12.2).
154 ANSWERS TO EXERCISES

2. The tangent at (1, 0, 0) to the conic

^22X2 + 2c23X2X3 ""t- 2c3]XgXj -)- 2c12XlX2 = ®

is c31x3 + c12x2 = 0, which coincides with x3 = 0 if c12 = 0. Similarly,
the tangent at (0, 0, 1) is x, = 0 if c23 = 0. Finally, the conic

c22x2 + 2c3ix3xi = 0

passes through (1, 1, 1) if c22 = — 2c31, and we are left with the equation

x22 = x3xv

3. The point of intersection (1, —/x, /u2) traces the conic x22 = XgX^

4. (i) X22 + 2XzXx = 0. (ii) X2 = 4X3XV

5. Let PQR be the triangle of reference, and O the unit point (1,1, 1). The
point (Aq, x2, x3) whose locus we seek is the trilinear pole of the line
[jxq-1, x2-1, x3-1], which passes through O if

xr1 + x2-x + Xg-1 = 0.

Hence the locus is the conic 12.62.

6. If (x)(y) is a tangent, let (jux + y) be its point of contact. Since this is

conjugate to both (x) and (y), we have

Mxx) + (xy) = 0, /x(xy) + (yy) = 0.

Eliminating fx, we obtain the locus

(xx)(yy) - (xy)2 = o,
which is the combined equation of the two tangents that pass through (y).
Dually, if [T] is a secant PQ of the conic [XX] = 0,

[XX][YY] - [XY]2 = 0

is the combined equation of the two points P and Q.

7. Let (y) and (z) be the two exterior points, so that the points of contact are
the intersections of the original conic (xx) = 0 with the polars (xy) = 0
and (xz) = 0. The conic

(xx)(yz) - (xy)(xz) = 0

evidently passes through all these six points. Dually, two secants [ Y], [Z],
and the tangents at their four “ends,” all touch the conic

[XX][YZ] - [XY][XZ] = 0.
 ANSWERS TO EXERCISES 155

8. By Exercise 7(ii) at the end of Chapter 10, a point (y) is exterior if and only
if it is accessible from (z). By Exercise 6 of Section 12.5, this condition
means that (yy), like (zz), is a nonzero square.
The conic 12.61 has tangents xx = x3 and x2 = x3, meeting in the point
(1,1, 1), for which (zz) = 1. Therefore (y) is exterior or interior according
as yy2 + yy2 — y>32 is a nonzero square or a nonsquare.

9. In the field of real numbers, the product of any two negative numbers is
positive. But in the field of rational numbers, the product of two non¬
squares is not necessarily a square. To obtain an infinity of mutually
inaccessible points, we can take (xx) = xx2 + x22 — x32 and choose
points (y) for which the values of (yy) run through the sequence of primes.

Section 12.7
1. (0, 0, 1), (0, 1, 0), (0, 1, 1), (0, 1, 4), (0, 2, 1), (0, 1, 2).

2. [4, 0, 1], [1, 4, 0], [1, 3, 1], [1, 1, 3], [3, 1, 1], [0, 1, 4],

3. (i) Since this collineation has the effect

(1,0, 0)-* (0,1,1), [1,0, 0] —*■ [0, 0,-1],

(0, 1,0)-*(-1,0,0), [0,1,0]-*[1,0, 0],
(0, 0, l)-*(0,-1,0), [0, 0, l]-*[0, 1,-1],
(1,1, 1)-(-1,0, 1), [1,1,1]-*[1,1,-2],

its equations are

K.

X\ = 3T2,
II
1

w«

x 2 — Ay x3,
II
<M

X 3 = x1; x '3 — — xx — X3.

(Remember that, in this finite arithmetic, — 1 =4.)

(ii) In this case

(1,0, 0) —> (3, 3, 3), [1,0, 0] —[1, 1,0],

(0, 1,0) — (-1, E 1), [0, 1,0]-* [2, 3, 0],
(0, 0, 1) — (0, 0, 1), [0, 0, 1] — [0, —1, 1],
(1,1,1)- (2, 4, 0), [1, 1, 1] — [3, 3, 1];

x j — 3x^ -x2, X\ — X1 -]- 2X2,

x 2 == 3x^ -j- x2, X'2 — X1 -f- 3X2 X3,
x 3 —- 3x^ "E a2—f~ a3, X'3 = X3.
156 ANSWERS TO EXERCISES

4.
(1,0,0) — [-1,0, 1], [l,0,0]->(0, 0, 1),
(0,1,0) ->[0,1,0], [0,1,0] -(0,1,0),
(0, 0, 1)-[1,0, 0], [0, 0, 1] —> (1, o, 1),
(1, 1,1) -[0,1, 1], [1, 1, 1]-(1, 1,2);

X1 *3

II
= —

%2 — -^2,

*
II
X3 — xy, *3 = + X..

5. (i) (1, 0, 2), (0, 2, 1), (2, 1, 0), (2, 0, 1), (0, 1, 2), (1, 2, 0).
(ii) (0, 1, 1), (0, 1, 4), (4, 0, 1), (2, 1, 0), (1, 2, 0), (1, 0, 1).
(iii) (0, 0, 1), (1, 0, 0), (1, 1, 1), (2, 1, 3), (3, 1, 2), (1, 4, 1).
(iv) (0, 0, 1), (0, 1, 0), (1, 0, 0), (1, 2, 1), (2, 1, 1), (1, 1, 2).
(v) (0, 1, 1), (1, 1, 0), (4, 1, 1), (1, 1, 4), (1, 4, 1), (1, 0, 1).

6. The field must admit an element whose square is —3; for instance, such a
configuration exists in the complex plane but not in the real plane, in
PG(2,pk) if and only if p is a prime congruent to 1 or 3 modulo 6.
Following the hint, we obtain F = CD • EH, G = BH • DE. The colline-
arity of EGA makes

CO2 + CD + 1 = 0.

The joins of “opposite” points all pass through (—to, 1, 1). In the case of
PG{2, 3), we have co = 1, and the (94, 123) uses up all the points except

(0,1,2), (2,0,1), (1,2,0) and (1,1,1),

which are the points on the remaining line [1, 1, 1].

7. We normalize the equation of the conic, as in Exercise 8 of Section 12.6,

and then apply Exercise 6 of Section 12.5, recalling that, in a finite field,
the product of any two nonsquares is a square. In fact, since any nonzero
square a2 has two distinct square roots ±a, the number of nonzero squares
is just half the number of nonzero elements. Thus the q — 1 nonzero
elements of the finite field consist of just (q — l)/2 squares, say su s2,. . .,
and the same number of nonsquares, say nu n2,. . .. Multiplying all these
q — 1 elements by any one of them, we obtain the same q — 1 elements
again (usually in a different arrangement). Since the (q — l)/2 elements
sink> s2nk, ■ ■ ■ (for a given nonsquare nk) are the «’s, the remaining {q — l)/2
elements npik, n2nk,. . . must be the s’s. (G. H. Hardy and E. M. Wright,
An Introduction to the Theory of Numbers, 2nd ed., Oxford, 1945, p. 70.)
 ANSWERS TO EXERCISES 157

Section 12.8
1. (i) x\ = —xlt x'2 — —x2 (as in 12.81 with ^ = — 1).
(ii) x\ — 2ax — xv x'2 = la2 — x2.

2. A rotation about the origin.

3. (i) x\ = xl5
(ii) x\ = x2.
 References

1. H. F. Baker, An Introduction to Plane Geometry, Cambridge University Press, 1943.

2. W.W. R.BaW, Mathematical Recreations and Essays (llthed.), Macmillan and Company,
London,1959.
3. E. T. Bell, Men of Mathematics, Simon and Schuster, New York, 1937.
4. J. L. Coolidge, A History of the Conic Sections and Quadric Surfaces, Clarendon
Press, Oxford, 1945.
5. —-- The Mathematics of Great Amateurs, Clarendon Press, Oxford, 1949.
6. Richard Courant and Herbert Robbins, What is Mathematics?, Oxford University
Press, New York, 1958.
7. H. S. M. Coxeter, The Real Projective Plane (2nd ed.), Cambridge University Press,
1955.
8. ——— Introduction to Geometry, John Wiley & Sons, Inc., New York, 1961.
9. H. G. Forder, The Foundations of Euclidean Geometry, Cambridge University Press,
1927; Dover Publications, New York, 1958.
10. -—■ Geometry (2nd ed.), Hutchinson’s University Library, London, 1960; Harper
Torchbooks, London, 1963.
11. Sir Thomas Heath, A History of Greek Mathematics (2 vols.), Clarendon Press, Oxford,
1921.
12. D. N. Lehmer, An Elementary Course in Synthetic Projective Geometry, Ginn and
Company, Boston, 1917.
13. E. H. Lockwood, A Book of Curves, Cambridge University Press, 1961.
14. G. B. Mathews, Projective Geometry, Longmans, Green and Company, London, 1914.
15. Beniamino Segre, Lectures on Modern Geometry, Cremonese, Rome, 1961.
16. A. S. Smogorzhevskil, The Ruler in Geometrical Constructions, Blaisdell Publishing
Company, New York, 1962.
17. D. J. Struik, Lectures on Analytic and Projective Geometry, Addison-Wesley, Cambridge,
Massachusetts, 1953.
18. J. L. Synge, Science: Sense and Nonsense, Jonathan Cape, London, 1951.
19. Oswald Veblen and J. W. Young, Projective Geometry, Vol I, Ginn and Company,
Boston, 1910.
20. A. N. Whitehead, The Aims of Education and Other Essays, Williams and Norgate,
London,1929.
158
Index

[References are to pages; principal references are in boldface.]

Absolute polarity, 110 Canonical form:

Accessible points, 101, 124, 126, 149, for a conic, 125
153, 155 for a polarity, 123
Affine coordinates, 130 Cartesian coordinates, 131, 150
geometry, 23, 74, 104, 129 Categoricalness, 31, 118
space, 103-110, 112 Cayley, Arthur, 4
Alberti, L. B., 3 Center:
Al-Dhahir, M. W., xii of a conic, 74, 131
Alias and Alibi, 120 of gravity, 112
Analytic geometry, 111-132 of a perspective collineation, 53
Anharmonic ratio, see Cross ratio of a perspectivity, 10, 18
Apollo, 71 Central dilatation, 131, 141
Apollonius of Perga, 3, 33, 43, 71 projection, 3, 104
Archimedes of Syracuse, 3 Cevian, 29
Archytas, 71 Characteristic equation, 121
Asymptotes, 74, 131 Chasles, Michel, 49, 60, 64
Axial pencil, 104-109 Chasles’s theorem, 64, 74, 124
Axioms, 6 Circle, 102-103
for the projective plane, 25, 95, 116 Collinear points, 2
for projective space, 15 Collineation, 49, 98
Axis: Complete n-point, 8
of a perspective collineation, 53 Complex geometry, 31, 72, 100, 124,
of a perspectivity, 10, 18 138
of a projectivity, 11, 37 Concurrent lines, 2
Cone, 86
Configuration, 26, 92, 129
Bachmann, Friedrich, 101 of Desargues, 27, 52
Bainbridge, E. S., 40 of Pappus, 38-40
Baker, H. F., 38 Conic, 3, 72-90, 99-103, 124-126
Ball, W. W. R., 71 through five points, 85
Barycentric coordinates, 119 touching five lines, 82
Bell, E. T., 1, 24 Conjugate diameters, 74
Biratio, see Cross ratio lines, 60, 72, 123
Birkhoff, Garrett, 149 points, 60, 72, 123
Braikenbridge, William, 85, 102 Consistency, 95
Brianchon, C. J., 85 Construction :
Brianchon’s theorem, 83, 90 for a conic, 83
Brunelleschi, Filippo, 2 for a harmonic conjugate, 22
Bundle, 104 for an involution, 46
Bussey, W. H., 91 for the polar of a point, 65, 75
Butterfly theorem, 78 for a projectivity, 33-34

159
160 INDEX

Construction (continued) Equation:

for the sixth point of a quadrangular of a conic, 124-125
set, 20-21 of a line, 113
for a tangent, 76 of a point, 113
Continuity, 31, 77 Equilateral triangle, 16, 103
Coolidge, J. L., 41, 60, 71, 81 Erlangen, 41
Coordinate axes, 130 Euclid of Alexandria, 1, 3, 104
transformation, 120 Euclidean geometry, 23, 74, 102-103,
Coordinates, 111, 113 110, 111, 131
Coplanar (points or lines), 2 Euler, Leonhard, 93
Correlation, 49, 57, 98 Exterior point, 73, 101, 126
Correspondence, 6
Courant, Richard, 1
Criss-cross theorem, 37 Fano, Gino, 14, 91, 97
Cronheim, Arno, 38 Fermat, Pierre, 111
Cross ratio, 118-119 Feuerbach, K. W., 4
Finite field, 93, 126
geometry, 31, 91-101, 127-129
Degenerate conic, 89-90, 146 Forder, H. G., 47, 91, 146
Desargues, Girard, 1, 3, 26-27, 45, Four-dimensional geometry, 16
71, 86, 109, 122
Fundamental theorem, 34, 97, 117-118
Desargues’s involution theorem, 79,
87, 146
Desargues’s two-triangle theorem, 19, Galois, Evariste, 100
24, 38, 92, 95, 116-117 Garner, Cyril, xii
Descartes, Rene, 111 Gergonne, J. D., 4, 60
Diagonal lines, 7 Grassmann, Hermann, 112
points, 7 Graves, J. T., 27
triangle, 16, 17, 75, 115 Graves triangles, 39, 138
Dictionary, 5 Greitzer, S. L., 103
Dilatation, 55, 131, 141
Distance, 104
Double contact, 78 H{AB, CD), 22
Double point, see Invariant point Half-turn, 56, 131, 142
Duality, 4, 10, 25, 105 Hall, Marshall, 93
Duplication of the cube, 71 Halmos, P. R., xii
Hardy, G. H., 14, 111, 156
Harmonic conjugate, 22, 28, 48, 118
Edge, W. L., 97 homology, 55-56, 70, 76, 98, 141
Elation, 53, 98, 121, 140 net, 30, 95
Electro-magnetic theory of light, progression, 135
91 sequence, 32, 96
Elementary correspondence, 8 set, 22-23, 28-29, 96
Elements of Euclid, 1 Harmony, 23
Ellipse, 3, 42, 131 Hausner, Melvin, 110
Elliptic involution, 47 Heath, Sir Thomas, 33
line, see Nonsecant Hesse, L. O., 68, 124, 148
point, see Interior point Hessenberg, Gerhard, 39
polarity, 72, 100, 101 Hexagon, 8
projectivity, 41, 43 of Brianchon, 83
Envelope, 72, 83 of Pappus, 38
coordinates, 113 of Pascal, 85
 INDEX 161

Hilbert, David, 152 Mathews, G. B., 12

Hippocrates of Chios, 71 Medians of a triangle, 103, 136
History, 2-4, 24, 71, 102 Menaechmus, 3, 71
Homogeneous coordinates, 112, 150 Midpoint, 23, 74
Homography, 49 Mobius, A. F., 4, 49, 112
Homology, 53, 98, 121 Model, 104-108
Hyperbola, 3, 42, 74. 131 Mohr, Georg, 1
Hyperbolic involution, 47, 118
line, see Secant
point, see Exterior point ^-dimensional geometry, 127
polarity, 72, 98 Net of rationality, 30
projectivity, 41, 43 Newton, Sir Isaac, 111
Non-Euclidean geometry, 102
Nonsecant, 72, 99
Ideal elements, 108-109
Identity, 41
Incidence, 5, 94, 113 One-dimensional forms, 8
Interior point, 73, 101, 126 Opposite sides, 7
Intersection, 5 Opposite vertices, 7
Invariant line, 50, 53 Order, 31
Invariant point, 12, 50, 54 Origin, 130
Inverse correspondence, 9
Involution, 45, 97, 118, 147
Pappus of Alexandria, 33
of conjugate points, 62, 64, 73, 99
Pappus’s theorem, 38, 122, 152
of Desargues, 87
Parabola, 3, 42, 74, 131
Involutory collineation, 55
Parabolic projectivity, 41, 43-48, 118,
Involutory correlation, see Polarity
141
Parallel lines, 102, 129
Join, 5 Parallel planes, 103
Jonsson, Wilbur, 56 Parallel projection, 104
Parallelogram, 4
Pascal, Blaise, 71, 81, 85-86
Kepler, Johann, 3, 71, 109 Pascal’s theorem, 85, 90
Klein, Felix, 4, 102 Pasch, Moritz, 16, 102
Pencil:
of conics, 78, 87
Lehmer, D. N., 2
of lines, 8, 16
Leibniz, G. W., 85
of parallel lines, 106
Levi, F. W., 40
of parallel planes, 107
Lie, Sophus, 53
of planes, 104-109
Line, 2, 112
Pentagon, 8, 67, 83
Line at infinity, 3, 109
Perfect difference set, 93
Line coordinates, 113
Periodic projectivity, 55
Linear fractional transformation, 118
Perpendicular lines, 110
Linear homogeneous transformation, 119
Perspective, 3, 18
Lockwood, E. H., 43
Perspective collineation, 53
Locus, 72, 80, 85 Perspectivity, 10, 35
Pieri, Mario, 14
MacLane, Saunders, 149 Plane, 2, 6, 16
MacLaurin, Colin, 85, 102 Plane at infinity, 104, 109
Mascheroni, Lorenzo, 1 Plucker, Julius, 112
162 INDEX

Point, 2, 112 Section:

Point at infinity, 3, 109 of a cone, 86
Point of contact, 72 of a pencil, 8
Polar, 60 Segre, Beniamino, 100, 112, 126
Polar triangle, 64 Self-conjugate line, 60
Polarity, 60, 98, 110, 122-124 Self-conjugate point, 60
Pole, 60, 72 Self-duality, 25, 26, 39, 105
Poncelet, J. V., 3, 10, 12, 24, 29, 49, Self-polar pentagon, 67
53, 71, 109 triangle, 62-63, 88, 99
Primitive concepts, 6, 14 Serret, J. A., 97
Principle of duality, 4, 10, 25 Seydewitz, Franz, 76-77
Product: Shadow, 3, 104
of elations, 55 Side, 7
of elementary correspondence, 10 Singer, James, 92
of harmonic homologies, 56 Smogorzhevskii, A. S., 1
of involutions, 47 Staudt, K. G. C. von, 4, 12, 29, 41,
of polarities, 68-70 45, 67, 71-72, 80, 102
Projection, 1, 3, 104 Steiner, Jacob, 77, 80, 89, 99
Projective collineation, 50, 98, 119-122 Struik, D. J., 102, 120
correlation, 57, 98, 122 Symmetric matrix, 123
geometry, 2, 3, 91, 102, 104 Symmetry of the harmonic relation, 29
Projectivity, 9 Synge, J. L., 5
on a line, 10, 34, 41-48, 97, 118
relating two ranges or pencils, 9 Tangent of a conic, 72, 99
Pure geometry, 111 Transformation, 49, 104
of coordinates, 120
Transitivity:
Q(ABC, DEF), 20, 46
of accessibility, 101, 148
Quadrangle, 7, 51, 58, 87, 95
of parallelism, 106
Quadrangular involution, 46, 64
Translation, 55, 131, 139, 141
Quadrangular set of points, 20-22, 46,
Triangle, 7
96
of reference, 103, 130
Quadratic form, 124-125
Trilinear polarity, 29, 62, 136
Quadrilateral, 7, 51, 58
Trott, Stanton, xii

Radius of a circle, 131 Unit line, 113

Range, 8 Unit point, 113
Ratio, 23, 104
Rational geometry, 31, 126 Veblen, Oswald, 12, 14, 16, 20, 30,
Real geometry, 31, 131, 138 40, 74, 76, 91, 120
Reciprocation with respect to a circle, Vertex, 7
102 Vish, 5
Reflection, 142 Von Staudt, see Staudt

Right angle, 110

Robbins, Herbert, 1 Whitehead, A. N., xii
WlGNER, E. P., 79
Wright, E. M., 156
Scherk, Peter, 100
Schuster, Seymour, 80, 89 Yaglom, I. M., 104
Secant, 72, 99 Young, J. W., 14, 20, 30, 40, 74, 76,120

PRINTED IN THE UNITED STATES OF AMERICA

ABOUT THE AUTHOR

H. S. M. Coxeter is a professor of mathematics at the University of Toronto.

He is a Fellow of the Royal Society and former editor of the Canadian
Journal of Mathematics.

The author received his B.A. and Ph.D. degrees from the University of
Cambridge and an LL.D. degree in 1957 from the University of Alberta.
He has been visiting professor at the University of Notre Dame and Columbia
University. Professor Coxeter is a life member of the London Mathematical
Society and was once first vice-president of the Mathematical Association of
America. His areas of specialization are in ^-dimensional crystallography
and in connections between geometry and the theory of groups.

Other books by Professor Coxeter include: Introduction To Geometry

(1961), Regular Polytopes (1963), The Real Projective Plane (1955), and
Non-Euclidean Geometry (1957).
 DATE DUE

f £ 2^’^

RESERX t

N0 23 7C
OE 7 TO

GAYLORD PRINTED IN U.S.A.

 MARYGROUE COLLEGE LIBRARY
Projective geometry
515 C83

3 ns? DDD3D317 1

mm ^