Projective Geometry and Symmetry Notes

GEOMETRY
Projective Geometry
Symmetry
Ruler and Compass

COURSE NOTES FOR MATH300

4
th
EDITION 2010

by Dr C. D. H. Cooper
2

What is Projective Geometry?

We all learnt Euclidean Geometry at school. This is the geometry that Euclid and his
co-workers developed over two thousand years ago and it has to do with points, lines, circles
and simple shapes like triangles and parallelograms. Euclids methods were axiomatic. He
wrote down a number of intuitively obvious facts and then proceeded to prove theorems in
a systematic way. Over the centuries Euclidean geometry became one of the pillars of
mathematics and was taught, not so much for its practical applications, but as an exercise in
pure thought.
Several centuries ago Ren Descartes showed how the same theorems could be
proved by introducing coordinates and using basic algebra. This method allowed parabolas,
ellipses and hyperbolas to be treated with almost as much ease as the circle.
The theorems of Euclidean geometry mainly deal with metric properties lengths,
angles and areas. But a few theorems are independent of measurement. They deal just with
incidence points lying on lines.
Affine geometry is Euclidean geometry stripped of all measurement. Not much
remains, but we can still talk of collinearity (points lying on the same line) and concurrence
(lines passing through the same point) and even the concept of lines being parallel.
Parallel lines, in the affine plane, are defined to be lines that dont intersect. In
Euclidean geometry they can be viewed as lines that remain the same distance apart and in
coordinate geometry wed view them as lines having the same slope. But in affine
geometry we must rely on the non-intersecting property.
Now its really rather bothersome that there are lines that dont cut. Through any two
distinct points there passes exactly one line. It would be nice to be able to say that any two
distinct lines intersect in exactly one point, but we cant, in the affine plane. Somehow it
seems like an artificial limitation, just like being told that negative numbers dont have
square roots. Well, they dont not in the real number field. But what did mathematicians
do there? They said bother that well, if they dont exist well invent them and they
proceeded to extend the real number field to the field of complex numbers.
Its a pity that points of intersection dont exist for parallel lines. Well, well just
invent them! And thats exactly what happened. The affine plane was extended to the
projective plane. Plane geometry was now complete in the same way that the complex
numbers completed arithmetic.
It happened in the 17
th
century, when perspective was being discovered by the
Renaissance artists. Remember that in a perspective painting parallel lines do meet at the
horizon. Projective geometry enables incidence properties of the Euclidean plane to be
proved much more easily.
As with Euclidean geometry there are two main approaches to projective geometry
through axioms and using algebra. Weve chosen to take the algebraic approach because it
makes good use of linear algebra. After such an introduction to projective geometry a
student truly understands linear independence and orthogonal complements.
Our proofs of Desargues Theorem and Pascals Theorem are built on the Collinearity
Lemma, something that appears here for the first time. Moreover it turns out that a certain
scalar that appears in this lemma is projective cross ratio and so the Collinearity Lemma
unifies all this material. We discuss perspectivities and projectivities and the Fundamental
Theorem of Projective Geometry.

3
Who Needs Symmetry?
Symmetry goes far beyond the mirror symmetry that everyone looks for in a beautiful
face. Theres rotational symmetry, translational symmetry and glide symmetry. In three
dimensions there are even more types of symmetry.
Who needs symmetry? Well an expectation of symmetry seems to be hard-wired into
our brains and affects the way we look at the world. In art, symmetry plays an important role
in creating an expectation. Its up to the artist whether to satisfy that expectation, or to
challenge it by creating beautiful asymmetry. For crystallographers, a branch of chemistry,
symmetry in 3 dimensions is very much their bread and butter. Particle physicists use
symmetry in higher dimensions. And architects and engineers, when using their CAD
software, need to know about symmetry.
Well explore isometries (distance preserving functions) in 2 and 3 dimensions and
well define symmetry in terms of shapes that remain invariant under certain isometries.
Groups are introduced as a tool for studying symmetry, and we describe the classification of
frieze patterns (linear patterns such as you might find on a border) into seven types and
wallpaper patterns (patterns that extend indefinitely in two dimensions) into 17 types.
We then look at finite rotation groups in 3 dimensions, which leads us to the seven
Platonic solids.

What are Ruler and Compass
Constructions?
Certain geometric constructions can be made using just a ruler and compass, such as
bisecting any given angle. The ruler is just used to draw straight lines, not to measure. Ruler
and compass methods have some practical use, but their main role in mathematics has been as
a theoretical challenge. Can a ruler and compass construction be found to carry out such and
such a construction. The method has to be perfectly exact. A very good approximation
might be sufficient for practical purposes but, for the Greeks and for those who followed
after, if it isnt perfectly exact its useless!
A classical problem in this area was to find a ruler and compass construction to trisect
any given angle. Geometers struggled for many centuries to find a way of doing it. Then
finally it was proved to be impossible. We develop field extensions as vector spaces and use
the concept dimension to provide a proof. This material is a very useful preparation for the
deeper study of Galois Theory.
Mostly we follow a very standard path through this topic. However, unlike many
treatments, we give as much attention to what can be done as to what cant and include a
large number of constructions. One novelty is the inclusion of constructions involving a
parabola. While rulers and compasses cant draw a parabola, if were given one, we can
construct its axis, vertex, focus and tangents.

4

5
CONTENTS

PART A: PROJECTIVE GEOMETRY

1. THE REAL PROJECTIVE PLANE
1.1 The Real Affine Plane . 9
1.2 Intuitive Construction of the Real Projective Plane 10
1.3 The Real Projective Plane is Complete ... 12
1.4 The Artists View of the Real Projective Plane ... 13
1.5 Embedding the Real Affine Plane in the Real Projective Plane .. 14
1.6 Review of Relevant Linear Algebra .... 16
1.7 The Algebraic Version of the Real Projective Plane ... 21
Exercises for Chapter 1 ..... 22
Solutions for Chapter 1 ...... 24

2. DESARGUES THEOREM
2.1 Orthogonality ... 27
2.2 The Collinearity Lemma .. 28
2.3 Perspective Triangles ... 29
2.4 Desargues Theorem 30
2.5 Euclidean Interpretation of Desargues Theorem 31
Solutions for Chapter 2 ..... 33

3. PAPPUS THEOREM
3.1 Duality . 37
3.2 Pappus Theorem . 38
3.3 Finite Projective Planes ... 39
3.4 Combinatorial Applications 43
Solutions for Chapter 3 .. 46

4. CROSS RATIO
4.1 Euclidean Cross Ratio . 53
4.2 Projective Cross Ratio . 54
4.3 Reconciliation of Euclidean and Projective Cross Ratios ... 56
4.4 Cross Ratio of Rearrangements ... 58
Exercises for Chapter 4 ................. 60
Solutions for Chapter 4 ..................... 60

6

5. PERSPECTIVITIES AND PROJECTIVITIES
5.1 Perspectivities .. 63
5.2 Projectivities 64
5.3 The Fundamental Theorem of Projective Geometry .. 65
5.4 Harmonic Conjugates .. 69
Solutions for Chapter 5 ...................... 71

PART B: SYMMETRY

6. ISOMETRIES
6.1 Isometries . 75
6.2 Central Isometries 76
6.3 Central Isometries of the Plane .. 78
6.4 Products of Central Isometries 80
6.5 Plane Isometries ... 81
6.6 Identifying Products of Plane Isometries 85

7. SYMMETRY GROUPS
7.1 What is Symmetry? . 93
7.2 Symmetry Groups 94
7.3 The Symmetry Group of a Square ... 96
7.4 Cyclic and Dihedral Groups 98
7.5 Finite Symmetry Groups in the Plane .. 101
7.6 The Seven Frieze Patterns ... 103
7.7 The 17 Wallpaper Patterns . 106

8. ISOMETRIES OF R
3

8.1 Central Isometries of R
3
... 115
8.2 Platonic Solids . 119
8.3 Rotation Groups of the Platonic Solids ... 121
8.4 Groups Acting on a Set 122
8.5 Finite Rotation Groups of Subsets of R
3
. 124
Exercises for Chapter 8 .................. 125

7
PART C: RULER AND COMPASS CONSTRUCTIONS

9. RULER AND COMPASS CONSTRUCTIONS
9.1 Ruler and Compass Constructions .. 131
9.2 Examples of Ruler and Compass Constructions . 133
9.3 Some More Advanced Constructions .. 137
9.4 Constructible Numbers 139

10. IMPOSSIBLE CONSTRUCTIONS
10.1 Number Fields and Field Extensions . 143
10.2 Fields as Vector Spaces . 144
10.3 Dimensions of Field Extensions 145
Exercises for Chapter 10 ................... 147
Solutions for Chapter 10 ................... 149

8

9
1. THE REAL PROJECTIVE PLANE

1.1 The Real Affine Plane
The Euclidean Plane involves a lot of things that can be measured, such as distances,
angles and areas. This is referred to as the metric structure of the Euclidean Plane. But
underlying this is the much simpler structure where all we have are points and lines and the
relation of a point lying on a line (or equivalently a line passing through a point). This
relation is referred to as the incidence structure of the Euclidean Plane.
The Real Affine Plane is simply the Euclidean Plane stripped of all but the incidence
structure. This eliminates any discussion of circles (these are defined in terms of distances)
and trigonometry (these need measurement of angles). It might seem that theres nothing left
but this isnt the case.
There is, of course, the concept of collinearity. Three or more points are collinear if
there is some line that passes through them all. Certain theorems of affine geometry state that
if theres such and such a configuration and certain triples of points are collinear then a
certain other triple is collinear. Another affine concept is that of concurrence three or
more lines passing through a single point.
Then theres the concept of parallelism, though we cant define it in terms of lines
having a constant distance between them or lines having the same slope.

Definition: Two lines h, k are parallel in the Real Affine Plane if either they coincide or
they dont intersect (meaning that no point lies on them both).

Here are two basic properties of the Real Affine Plane:
(1) Given any two distinct points there is exactly one line passing through both.
(2) Given any two distinct lines there is at most one point lying on both.

Theres an obvious similarity between these two properties. But while we can say
exactly one in property (1) the best we can do in property (2) is at most one because there
is no common point when the lines are parallel.

This is reminiscent of the situation with the real number system, where every real
number (except zero) has at most two square roots. Why not exactly two? The reason is
because the negative numbers dont have any square roots.
This was considered to be a defect of the real number system and so imaginary
numbers were invented to provide square roots for negative numbers. The field of real
numbers was expanded to form the complex numbers, and for this system we can say that
every non-zero number has exactly two square roots.

We do the same thing with the Real Affine Plane. We invent imaginary points
where parallel lines can meet. We call these extra points, ideal points. And we invent an
imaginary line called the ideal line, and specify that all the ideal points lie on this line. The
Real Affine Plane is thereby extended to the Real Projective Plane. In the Real Projective
Plane there are no longer such things as distinct parallel lines since every pair of distinct lines
10
now intersect in exactly one point. This then mirrors exactly the fact that every pair of
distinct points lie on exactly one line.

1.2. Intuitive Construction of the Real Projective Plane
We start with the Real Affine Plane.

(This trapezium is merely a representative picture of the affine plane. If we drew an accurate
picture it would cover the whole page, and then some! There would be no space for any extra
points, let alone the text. You should view this as a perspective view of a very large rectangle
and then pretend that it extends infinitely far in all directions.)

We call the points on the Real Affine Plane ordinary points.

. . . . . . .
. . . . . . . . . . . .
. . . . . . . . .

We call the lines on the Real Affine Plane ordinary lines.

We sort these ordinary lines into parallel classes.

A parallel class consists of a line together with all lines parallel to it.

11
For each parallel class we invent a new point, called an ideal point.

These ideal points dont lie on the Real Affine Plane. Where are they then? The answer is
simply in our minds. However, to assist our imagination, we can put these ideal points on
our diagram outside of the shape that represents the Real Affine Plane.

As well as ordinary points lying on ordinary lines in the usual way

we decree that all lines in a given parallel class (and no others) pass through the
corresponding ideal point.

We also invent a new line called the ideal line

and decree that this line passes through all the ideal points (and no others).

12
The resulting geometry is called the Real Projective Plane. It contains all of the real affine
plane, as well as the ideal points and the ideal line. Any theorem that we can prove for the
real projective plane will be true for the real affine plane simply by taking the points and lines
to be ordinary ones.

1.3. The Real Projective Plane is Complete
Theorem 1A: In the real projective plane:
(i) any two distinct points lie on exactly on line;
(ii) any two distinct lines intersect in exactly one point.
Proof: (i) Case I: two ordinary points lie on an ordinary line, as in the affine plane.

Case II: an ordinary point P and an ideal point Q lie on the line through P parallel to the lines
in the parallel class corresponding to Q. (Remember that in Euclidean Geometry theres a
unique line which passes through a given point and is parallel to a given line.)

Case III: Two ideal points lie on the ideal line.

(ii) Case I: Two ordinary lines:
If these lines are non-parallel they intersect in an ordinary point.

If theyre parallel they intersect in the ideal point that corresponds to their parallel class.

P
Q
13
Case II: An ordinary line and the ideal line:
An ordinary line intersects the ideal line in the ideal point corresponding to the parallel class
in which it lies.

1.4. The Artists View of the Real Projective Plane
Renaissance artists had no problem with the concept of parallel lines meeting a point.
This happens all the time in a perspective drawing.

Consider what an artist does when he sketches a scene. (If this sounds sexist why not she
allow me to point out this is a Renaissance artist, and they were mostly male!) You might
think that he represents points in the scene by points on the canvas, but it would be more
accurate to say that he represents rays not points. Every ray emanating from his eye
corresponds to a single point on his canvas.

This leads to the next way of thinking about the Real Projective Plane. Consider the 3-
dimensional Euclidean space, R
3
.

Definition: A projective point is a line through the origin. A projective line is a plane
through the origin.

It might seem strange at first to call a line a point and to call a plane a line but to the
artist a line in 3-dimensional space, passing through his viewpoint, is what he depicts as a
point on his canvas. [This is not completely true, in that an artist (unless he has eyes in the
back of his head) only depicts rays (half-lines) not whole lines.]
14

rays lines

The Real Projective Plane is defined as the set of all projective points and all
projective lines. It can be thought of as a sort of porcupine in 3-dimensional space, bristling
with lines going in all directions through the origin. To complete the description we must
define incidence, that is, we must explain what it means for a projective point to lie on a
projective line.

Definitions: A projective point P (line through the origin) lies on a projective line h (plane
through the origin) if, when considered as a line, it lies on h, considered as a plane.
A projective line h passes through a projective point P if P lies on h.
Three (or more) projective points are collinear if they lie on a common projective line.
Three (or more) projective lines are concurrent if they pass through the same projective
point.

Theorem 1B: In the Real Projective Plane:
(i) any two distinct projective points lie on exactly on projective line;
(ii) any two distinct projective lines intersect in exactly one projective point.
Proof: Translated into ordinary terms these state that:
(i) any two lines through the origin lie on exactly one plane through the origin and
(ii) any two planes through the origin intersect in exactly one line.
Both of these are clearly true statements for 3-dimensional Euclidean Space.

1.5. Embedding the Real Affine Plane in the Real Projective
Plane
Its all very well to have constructed a geometry whose incidence structure is
complete, but is it really the same structure that we created earlier, by extending the Real
Affine Plane by adding ideal points? After all, with this new version, which projective points
are the ideal points? With this bunch of lines through the origin they all seem pretty much
the same.
Well, suppose we take an ordinary Real Affine Plane in R
3
, one that doesnt pass
through the origin. We can think of the origin as the artists eye and the plane as his (infinite)
canvas. Every (ordinary) point P on corresponds to a unique line through the origin,
namely the line OP. But this is a projective point according to our second point of view.
line OP

P ordinary point P

O

But if we take the totality of affine points the corresponding projective points (lines
through O) do not use up all the available lines through the origin. Only a projective point
15
that, viewed as a line through the origin, cuts will correspond to a point on (viz. the
intersection of that line with ). Left over will be the projective points (lines through O) that
are parallel to . These are in fact the ideal points in the real projective plane.

no corresponding ordinary point

ideal projective point
O
Suppose we have two ordinary parallel lines on . They dont intersect on . But the
corresponding projective lines do intersect. These corresponding projective lines will be the
planes through the origin that pass through the respective lines on . These lines will itersect
in a line (just as two pages of a book intersect in the spine of the book).

two parallel ordinary lines corresponding projective
lines intersect in a projective point that doesnt
correspond to an ordinary point

This line will pass through O so it can be considered as a projective point. But it will
be parallel to and hence wont correspond to an ordinary point. So what were two parallel,
non-intersecting lines in the ordinary plane , correspond to two projective lines (planes
through O) that do intersect in an ideal point (line through O). The ideal points are in fact
the lines through O that are parallel to and the ideal line is thus the plane through O that is
parallel to .

ordinary plane embedded in R
3

the ideal line (corresponding to )

In this way we have, in effect, completed the Real Affine Plane to produce the Real
Projective Plane. But notice that the plane that becomes the ideal line depends on which
affine plane we took in R
3
. In fact any projective line can become the ideal line if we take a
suitable affine plane and a suitable origin O.
This is important. The distinction between ordinary point and ideal point is not one that
is intrinsic to the Real Projective Plane itself. It only reflects the way we view the Real
Projective Plane from the point of view of a Real Affine Plane.
The beauty of the Real Projective Plane is that it is wonderfully uniform. We can prove
Euclidean theorems by projective means without the messiness of having to consider cases
where lines are parallel and where lines are not parallel. In the Real Projective Plane, all
points are equivalent and all lines are equivalent.
O

O

16
Example 1: Suppose we take the plane to be z = 1.
(i) Which of the axes (x-, y- and z-) are ideal points and which are ordinary (relative to )?
(ii) What is the projective point corresponding to the ordinary point (1, 1, 1) on ?
(iii) What point on corresponds to the projective point x = 2y = 3z?
(iv) Relative to which planes is x + y + z = 0 the ideal line?
Solution: (i) The x- and y-axes are ideal, the z-axis is ordinary.
(ii) The line x = y = z;
(iii) (3, 3/2, 1) (this is where the line x = 2y = 3z intersects );
(iv) any plane of the form x + y + z = c for c 0.

1.6. Review of Relevant Linear Algebra
Were going to make this last view of the Real Projective Plane more rigorous by
introducing linear algebra. Instead of talking about projective points as lines through the
origin we shall think of them as one-dimensional subspaces of R
3
. This will free us from
needing geometric intuition, and allow us to use algebraic methods. But also we can change
the underlying field of our vector space, giving rise to other projective planes even finite
ones. Before we do this we will review the relevant linear algebra.
A vector space over a field F is a set (with the elements being called vectors)
together with operations of addition and scalar multiplication (a scalar being an element of F)
such that certain axioms are satisfied. Vectors are usually printed in bold type, such as v, to
distinguish them from scalars (but there are times when this distinction cant be maintained).
Practically any mathematical object can be an element of a useful vector space. You
can have vector spaces of numbers, of matrices, and of functions. But here we need consider
only those vectors of the form (x
1
, , x
n
), and then only for the case n = 3.
So our vectors will all have the form (x, y, z) with the components x, y, z coming
from some field. Well mainly use the field R, of real numbers, though at one point well
use various finite fields such as Z
p
, the field of integers modulo a prime p.
Vector addition and scalar multiplication are defined in the usual way, component by
component.

A linear combination of the vectors v
1
, , v
n
is a vector of the form
1
v
1
+
n
v
n
for some scalars
1
, ,
n
.
A linear combination is non-trivial if at least one of the scalars is non-zero. A set of vectors
{v
1
, v
n
} is linearly independent if
1
v
1
+
n
v
n
= 0 implies that each
i
= 0.
Otherwise they are linearly dependent.

Note that its the set thats linearly dependent or independent, not the vectors
themselves. In other words it refers to the relationship between the vectors. A set of vectors
is linearly dependent if there exists a non-trivial linear combination of them that is equal to
zero. In such a case one of the vectors can be written as a linear combination of the others. If
theyre linearly independent this isnt possible.

Example 2: Are the vectors (1, 5, 2), (6, 1, 9), (9, 17, 24) linearly independent?
Solution: Sometimes students learn to answer such questions by putting the three vectors into
a determinant and concluding that, because the determinant is zero, the vectors must be
17
linearly dependent. That technique is valid in certain cases but its best to go back to the
definition and use the techniques of solving systems of linear equations. Apart from the fact
that the determinant method only works for n vectors in F
n
, its seen by many students to
be a piece of hocus-pocus a formula to be invoked, without any understanding. The
following technique reveals whats happening, and need involve no more work.
Suppose a(1, 5, 2) + b(6, 1, 9) + c(9, 17, 24) = (0, 0 0). This is equivalent to the
system of equations:
)
a + 6b + 9c = 0
5a b 17c = 0
2a + 9b + 24c = 0

which can be represented by an augmented matrix:

\
|
.
|
|
|
1 6 9 0
5 1 17 0
2 9 24 0
.
As is usual with systems of homogeneous equations we can omit the column of 0s and
simply write the system as:
\
|
.
|
|
|
1 6 9
5 1 17
2 9 24
.
To solve this system we carry out a sequence of elementary row operations to put the
matrix into echelon form. This simpler matrix represents a system of homogeneous linear
equations thats equivalent to the original system (equivalent in the sense of having precisely
the same solutions).
In this example we obtain:
\
|
.
|
|
|
1 6 9
5 1 17
2 9 24

\
|
.
|
|
|
1 6 9
0 31 62
0 21 42

\
|
.
|
|
|
1 6 9
0 31 62
0 1 2

\
|
.
|
|
|
1 6 9
0 1 2
0 0 0

R
2
5R
1
, R
3
+2R
1
R
2
21 R
2
+31R
3
, R
2
R
3

Clearly this system has infinitely many solutions, and hence non-trivial ones. This is
not so much because of the row of zeros at the bottom as the fact that we have an equivalent
system with fewer equations than variables. In solving the above equivalent system:
)
`
a + 6b + 9c = 0
b + 2c = 0

we may choose c arbitrarily (and hence non-zero) and use the two equations to calculate the
corresponding values of a and b.
The conclusion is that the vectors are linearly dependent. Its not necessary to exhibit
an explicit relationship, but if called upon to do this we simply choose a convenient non-zero
value of c and calculate the corresponding values of a and b.
For example if we take c = 1 we get b = 2 from the second equation and hence
a = 12 9 = 3 from the first equation. This gives the non-trivial linear relationship:
3(1, 5, 2) 2(6, 1, 9) + (9, 17, 24) = (0, 0, 0), which can be written as:
(9, 17, 24) = 3(1, 5, 2) + 2(6, 1, 9).
18
Example 3: Are the vectors (2, 5, 9), (3, 1, 0), (5, 2, 7) linearly independent?
Solution: Writing them as the columns of a matrix (to get the coefficient matrix of the
corresponding system of homogeneous linear equations) we have:
\
|
.
|
|
|
2 3 5
5 1 2
9 0 7

\
|
.
|
| 1
3
2

5
2
5 1 2
9 0 7

\
|
.
|
|
|
1
3
2

5
2
0
13
2

29
2
0
27
2

31
2

\
|
.
|
|
|
1
3
2

5
2
0 1
29
13
0
27
2

31
2

\
|
.
|
|
|
1
3
2

5
2
0 1
29
13
0 0
31
2
+
27
2
.
29
13
.
R
1
2 R
2
5R
1
, R
3
9R
1
R
2
(13/2) R
3
+ (27/2)R
2

Since
31
2
+
27
2
.
29
13
0 this system has a unique solution (the zero one).
Note that if were doing the calculations by hand we can avoid most fractions if we
get leading 1s by subtracting suitable multiples of rows from others rather than by division.
This greatly simplifies the arithmetic. Re-doing the above example we can obtain:
\
|
.
|
|
|
2 3 5
5 1 2
9 0 7

\
|
.
|
|
|
2 3 5
1 5 12
9 0 7

\
|
.
|
|
| 1 5 12
2 3 5
9 0 7

\
|
.
|
|
| 1 5 12
0 13 29
0 45 115

\
|
.
|
|
| 1 5 12
0 13 29
0 6 28

\
|
.
|
|
|
1 5 12
0 1 27
0 6 28

R
2
2R
1
R
1
R
2
R
2
2R
1
, R
3
9R
1
R
3
3R
2
R
2
2R
3

|
.
|
|
|
1 5 12
0 1 27
0 0 190
.
R
3
6R
2

As before this system has only the trivial solution so the vectors are linearly independent.

Example 4: Is the vector (3, 18, 4) a linear combination of the vectors (1, 5, 3) and
(4, 22, 1)?
Solution: The question is whether there exist scalars x, y such that:
(3, 18, 4) = x (1, 5, 3) + y (4, 22, 1).
We can write this equation as a non-homogeneous system of linear equations:
)
x + 4y = 3
5x + 22y = 18
3x + y = 4

Now we can write this system as an augmented matrix:
\
|
.
|
|
|
1 4 3
5 22 18
3 1 4
and carrying out a
sequence of elementary row operations we put this in echelon form:
\
|
.
|
|
|
1 4 3
5 22 18
3 1 4

\
|
.
|
|
|
1 4 3
0 2 3
0 13 5

\
|
.
|
|
|
1 4 3
0 2 3
0 1 13

\
|
.
|
|
|
1 4 3
0 1 13
0 2 3

\
|
.
|
|
|
1 4 3
0 1 13
0 0 29
.
R
2
5R
1
, R
3
+ 3R
1
R
3
6R
2
R
2
R
3
R
3
3R
2

This echelon form represents an equivalent system thats clearly inconsistent. In other words
theres no solution for x, y and so (3, 1, 4) is not a linear combination of the vectors
(1, 5, 9) and (4, 2, 1).
19
A set of vectors spans a vector space V if every vector in V is a linear combination
of them. A vector space is finite-dimensional if it has a finite spanning set. A set of vectors
is a basis for V if its linearly independent and also spans V.
If we have a finite spanning set for V we can remove suitable vectors one by one, as
long as the set is linearly dependent, without affecting the spanning, until we reach a basis.
That is, every spanning set contains a basis. In a similar way, every linearly independent
subset of V can be suitably extended to a basis. But any two bases of a finite-dimensional
vector space contain the same number of vectors. This unique number of vectors in a basis is
called the dimension of the vector space. The vector space R
3
clearly has dimension 3
since we have the basis {(1, 0, 0), (0, 1, 0), (0, 0, 1)}.
If we have a vector space V of dimension n, then any set of more than n vectors
must clearly be linearly dependent (otherwise we could extend them to a basis with more
vectors than the dimension). And if we have a set of less than n vectors they cant possibly
span V (otherwise the dimension would be less than n).

Example 5: Are the vectors (1, 4, 6), (2, 5, 1), (6, 1, 9), (4, 8, 3) linearly independent?
Solution: The answer is clearly NO since you cant have a set of four linearly independent
vectors inside a 3-dimensional vector space. Theres no need to do any calculation in this
case other than to count the number of vectors!

A vector space V is said to be the sum of subspaces U and W if every vector
v V can be expressed as v = u + w where u U and w W. It is said to be the direct
sum if as well, U W = 0. In the first case we write V = U + W and in the case of direct
sum we write V = U W. The special feature of a direct sum is that if V = U W then
every vector v V can be expressed uniquely as v = u + w where u U and w W.
If we take a basis for each of the subspaces in a direct sum, and combine them, well
produce a basis for the whole space. This means that dim(U V) = dimU + dimV.

A Euclidean space is a vector space V over R on which there is defined an inner
product, that is a function that assigns to every pair of vectors u, v V a real number u.v
such that the following axioms hold:
(1) u.v = v.u for all u, v;
(2) u.(v + w) = u.v + u.w, for all u, v, w;
(3) u.u > 0 when u 0.

Where the space is a space of functions the inner product is often defined as a certain definite
integral, but in the simple case of row vectors over R we usually take the dot product:
(x
1
, , x
n
).(y
1
, , y
n
) = x
1
y
1
+ + x
n
y
n
.
Vectors in a Euclidean space are orthogonal if their dot product is zero. The
orthogonal complement of a subspace U of a vector space V is defined to be the set of all
vectors that are orthogonal to every vector in U, that is U
= {v V | u.v = 0 for all u U}.

The orthogonal complement U
is also a subspace of V and its easy to show that every

vector in V can be expressed uniquely as the sum of a vector in U and a vector orthogonal
to every vector in U. In symbols this can be expressed as V = U U
. One can also easily

show that for all subspaces of a Euclidean space U
= U (the orthogonal complement of the

orthogonal complement is the original subspace).
20
For the purpose of Projective Geometry well be dealing with R
3
with the usual dot
product. This means that well have to consider (apart from the zero subspace and the whole
of R
3
) only subspaces of dimension 1 and 2. Moreover the orthogonal complement of a 1-
dimensional subspace will be a 2-dimensional one and vice-versa.

Vectors in R
3
have a simple geometric interpretation whereby (x, y, z) denotes the
point with coordinates x, y, z in 3-dimensional Euclidean space. A 1-dimensional
subspace of R
3
will then represent a line through the origin and a 2-dimensional subspace
will represent a plane through the origin.
Often we interpret (x, y, z) as representing not just the point, but the directed line
segment from the origin to that point. In this case the vector has both magnitude (length) and
direction. With this interpretation, non-zero vectors are orthogonal if and only if theyre
perpendicular.
The orthogonal complement of a 1-dimensional subspace of R
3
(a line through the
origin) is a 2-dimensional subspace (a plane through the origin), being the plane
perpendicular to that line. The orthogonal complement of a plane through the origin is a line
through the origin.

In this special case of R
3
we have another product of vectors. The cross product
u v of two vectors u, v in a 3-dimensional vector space is a vector that is orthogonal to
both u and v. It is defined by:
(x
1
, x
2
, x
3
) (y
1
, y
2
, y
3
) = (x
2
y
3
x
3
y
2
, (x
1
y
3
x
3
y
1
), x
1
y
2
x
2
y
1
).
Its more easily remembered as the 3 3 determinant:
i j k
x
1
x
2
x
3
y
1
y
2
y
3

The top row are the standard basis vectors i = (1, 0, 0), j = (0, 1, 0), k = (0, 0, 1) and
not scalars like the second and third rows. Nevertheless, using the usual rules for expanding
determinants, we obtain a valid expression for the cross-product.

Example 6: If u = (1, 5, 3) and v = (4, 1, 2) find u v.
Solution: u v =
i j k
1 5 3
4 1 2
= [5.2 (3)(1)] i [1.2 (3).4)] j + [1.(1) 5.4] k
= 7i 14j 21k = (7, 14, 21).
Check: (7, 14, 21).(1, 5, 3) = 7 70 + 63 = 0.
(7, 14, 21).(4, 1, 2) = 28 + 14 42 = 0.

Since its possible to make an error in sign at some stage its important that you check
your answer to a cross-product against at least one of the two vectors.

Another concept from linear algebra that well find useful is that of a linear
transformation. A function f : U V is a linear transformation if
f(u + v) = f(u) + f(v)
for all u, v in U and all scalars and .
21
The kernel of such a linear transformation is ker f = {u U | f(u) = 0} and its
image is im f = {f(u) | u U}. The kernel is a subspace of U and its dimension is called
the nullity of f. The image is a subspace of V and its dimension is called the rank of f.
One can show that
rank(f) + nullity(f) = dim U.

1.7. The Algebraic Version of the Real Projective Plane
The description of the Real Projective Plane as lines and planes through the origin is
better than the intuitive one, where we simply invented ideal points, in that it frees us from
having to maintain the artificial distinction between the ordinary and the ideal. But its still
somewhat difficult to work with because it involves 3-dimensional geometry. The effective
way of dealing with this is to use the linear algebra of R
3
. This third development of the
Real Projective Plane is essentially the same as the previous one, except that things are
expressed algebraically instead of geometrically.
We work in the 3-dimensional real vector space R
3
. The vectors have the form
(x, y, z) where x, y and z are real numbers.

Definitions:
A projective point is a 1-dimensional subspace of R
3
.
A projective line is a 2-dimensional subspace of R
3
.
A projective point P lies on a projective line h if the 1-dimensional subspace P is a
subspace of the 2-dimensional subspace h.
The Real Projective Plane is the set of all projective points and lines, together with the
above incidence structure (relation of point lying on a line).

You should be able to see that this is simply an algebraic restatement of the previous
description of the artists real projective plane.

Now a projective point P, being a 1-dimensional subspace, has a basis consisting of
one non-zero vector p so P = (p). The elements of P are scalar multiples of p.
projective point (p)

If P = (p) and Q = (q) are projective points then P = Q doesnt necessarily imply
that p = q. What it does mean is that each is a non-zero scalar multiple of the other, or to use
the language of vector spaces, the set {p, q} is linearly dependent. On the other hand if P,
Q are distinct points then p, q are linearly independent.

If P = (p) and Q = (q) are distinct then the smallest subspace containing them both is
the 2-dimensional subspace (p, q). This is the line PQ.

(p, q) = {p + q | , R}

0
p
q
0
p
22
Theorem 1C: The Real Projective Plane is complete. That is:
(i) any two projective points lie on exactly one projective line;
(ii) any two projective lines intersect in exactly one projective point.
Proof:
(i) Let P = (p) and Q = (q) be two distinct projective points. Then P + Q = (p, q) is
clearly the only 2-dimensional subspace containing both P and Q. It is thus the only
projective line containing the projective points P, Q.
(ii) Let h, k be two distinct projective lines. Now h + k contains both h and k.
If h + k = h then k h + k = h. Since h, k both have dimension 2 we must have
h = k. [Recall that if two subspaces have the same dimension and one is inside the other, they
must be equal.] This contradicts our assumption.
Hence h is a proper subspace of h + k, so dim(h + k) > 2. On the other hand the
whole thing is going on in R
3
which has dimension 3, so dim(h + k) = 3. Thus we must
have dim(h + k) = 3.
Now recall the theorem from linear algebra that states that:
dim(h + k) = dim(h) + dim(k) dim(h k).
Therefore 3 = 2 + 2 dim(h k), and so dim(h k) = 1. So h k is a projective point,
and being their intersection, it lies on both h and k.

Theorem 2: Three projective points P = ( (( (p) )) ), Q = ( (( (q) )) ) and R = ( (( (r) )) ) are collinear if and
only if the vectors p, q and r are linearly dependent.
Proof: For (p), (q), (r) to be collinear they must lie on a common projective line. That
means they must lie in a 2-dimensional subspace. For three vectors to lie in a 2-dimensional
subspace they must be linearly dependent. Conversely if p, q and r are linearly dependent
theyll span a subspace of dimension at most 2. If (p, q, r) has dimension 2 then its a
projective line passing through the projective points P, Q, R.

EXERCISES FOR CHAPTER 1

TRUE/FALSE QUIZ
(1) An ordinary point cannot lie on the ideal line.
(2) An ideal point cannot lie on an ordinary line.
(3) Every ordinary line cuts the ideal line.
(4) Two non-parallel ordinary lines cannot intersect in an ideal point.
(5) There are infinitely many lines parallel to the ideal line.
(6) There are infinitely many parallel classes.
(7) There are infinitely many ideal points.
(8) There are infinitely many ideal lines.
(9) There is exactly one ideal point on every projective line.
(10) There is exactly one ideal line through every projective point.

Exercise 1: Which of the following sets of vectors are linearly dependent?
With R as the field of scalars:
(i) {(3, 4, 7), (2, 8, 3), (10, 8, 25)};
(ii) {(4, 6, 2), (6, 0, 1), (5, 1, 3), (1, 1, 0)};
23
(iii) {(6, 9, 11), (2, 1, 1)};
(iv) {(12, 15, 9), (20, 25, 15)}.
With Z
2
as the field of scalars:
(v) {(1, 1, 1), (1, 1, 0), (0, 0, 1)};
(vi) {(1, 0, 0), (1, 1, 0), (0, 1, 1), (1, 1, 1)}.

Exercise 2: Write (10, 8, 25) as a linear combination of the vectors (3, 4, 7) and (2, 8, 3).

Exercise 3:
(a) What is the dimension of the subspace of R
3
spanned by (3, 4, 7), (2, 8, 3) and (10, 8, 25)?
(b) What is the dimension of the space of all vectors which are orthogonal to all three of the
vectors (3, 2, 7), (1, 1, 0), (2, 3, 7)?
(c) With Z
5
as the field of scalars find the dimension of the subspace spanned by the vectors
(1, 3, 2), (4, 1, 0) and (0, 1, 3).

Exercise 4: If a = (6, 3, 2) and b = (2, 1, 5),find
(i) a.a;
(ii) a.b;
(iii) b.a;
(iv) b.b;
(v) a a;
(vi) a b;
(vii) b a;
(viii) b b.

Exercise 5: (a) Prove that for all u, v
3
u v = (v u).
(b) Hence show that for all v
3
v v = 0.
Exercise 6: Find ((4, 7, 8))
((3, 1, 2))
.

Exercise 7: Let a = (1, 2, 3). Find the rank and nullity of the following (where they exist):
(a) f: R
3
R defined by f(v) = v.a.
(b) g: R
3
R
3
defined by g(v) = v a.
(c) h: R
3
R defined by h(v) = v.v.

Exercise 8: If A = ((1, 2, 3)) and B = ((5, 1, 5)) find AB.

Exercise 9: If P = ((1, 1, 1)) and Q = ((1, 2, 3)) does R = ((3, 1, 4)) lie on PQ?

Exercise 10: If A = (a), B = (b) and C = (a + b), show that C lies on AB.

Exercise 11: Let P = ((1, 1, 1)), Q = ((1, 2, 3)), R = ((2, 3, 1)) and S = ((3, 1, 2)).
Find PQ RS

Exercise 12: Let A = ((4, 10, 5)), B = ((0, 7, 3)), C = ((1, 1, 0)) and D = ((1, 2, 2)).
24
Let M = AB CD and let N = AC BD. Find MN and determine whether AD BC lies
on MN.

Exercise 13: Is ((1, 0, 1)) an ideal point?

Exercise 14: If we embed the Euclidean plane in the projective plane by taking the plane
= {(x, y, z) | x + 2y + 3z = 4}, determine whether ((1, 2, 1)) is an ideal point.

SOLUTIONS FOR CHAPTER 1

TRUE/FALSE QUIZ:
(1) T (2) F (3) T (4) F (5) F (6) T (7) T (8) F (9) F (dont forget the ideal line) (10) F.

Exercise 1: (i) Suppose x (3, 4, 7) + y (2, 8, 3) + z (10, 8, 25) = (0, 0, 0).
Then x, y, z satisfy the system of homogeneous linear equations with coefficient matrix:
\
|
.
|
|
|
3 2 10
4 8 8
7 3 25

\
|
.
|
|
|
3 2 10
1 6 2
7 3 25

\
|
.
|
|
| 1 6 2
3 2 10
7 3 25

\
|
.
|
|
| 1 6 2
0 16 16
0 39 39

\
|
.
|
|
|
1 6 2
0 1 1
0 0 0
.
R2-R1 R1 R2 R23R1, R37R1 R2(16), R3+39R2
This system has the non-trivial solution z = 1, y = 1, x = 4.
Hence 4 (3,4,7) = (2, 8, 3) + (10, 8, 25) and so the set is linearly dependent.

(ii) The set is linearly dependent since there are 4 vectors in a 3-dimensional vector space.

(iii) A set of two vectors is linearly dependent if and only if one is a scalar multiple of the
other. This is not the case here so the set is linearly independent.

(iv) (12, 15, 9) = (3/5)(20, 25, 15) so the set is linearly dependent.

(v) The sum of the vectors is 0 so the set is linearly dependent.

(vi) The set is linearly dependent for the same reason as in (ii)

Exercise 2:
Let x (3, 4, 7) + y (2, 8, 3) = (10, 8, 25).
Then x, y satisfy the system of linear equations:
\
|
.
|
|
|
3 2 10
4 8 8
7 3 25
. Proceeding as in Question 1 (i)
we get the echelon form
\
|
.
|
|
|
1 6 2
0 1 1
0 0 0
which gives y = 1, x = 4.
Hence (10, 8, 25) = 4 (3, 4, 7) (2, 8, 3).

Exercise 3:
25
(a) Since the vectors are linearly dependent (done in Question 1 (i)) the dimension < 3. As
they are not all scalar multiples of the same vector the dimension > 1.
Hence the dimension = 2.

(b) Suppose x (3, 2, 7) + y (1, 1, 0) + z (2, 3, 7) = (0, 0, 0).
Then x, y, z satisfy the homogeneous system:
\
|
.
|
|
|
3 1 2
2 1 3
7 0 7

\
|
.
|
|
|
1 0 1
2 1 3
7 0 7
. Since R3 = 7R1 this system has a non-zero solution and so the set of
vectors is linearly dependent. So they span a 2-dimensional subspace and their orthogonal
complement has dimension 3 2 = 1.

(c) Suppose x (1, 3, 2) + y (4, 1, 0) + z (0, 1, 3) = (0, 0, 0).
Then x, y, z satisfy the homogeneous system:
\
|
.
|
|
|
1 4 0
3 1 1
2 0 3

\
|
.
|
|
|
1 4 0
0 4 1
0 2 3

\
|
.
|
|
|
1 4 0
0 1 4
0 2 3

\
|
.
|
|
|
1 4 0
0 1 4
0 0 0
..
R23R1, R32R1 R24 R32R2
This system has only the zero solution so the set of vectors is linearly independent. So they
span a 3-dimensional subspace.

Exercise 4:
a.a = 36 + 9 + 4 = 49; a.b = 12 + 3 + 10 = 1; b.a = 1 (always b.a = a.b);
b.b = 4 + 1 + 25 = 30; a a = 0 (true for all a); a b =
i j k
6 3 2
2 1 5
= (13, 34, 12)
(check: 78 102 + 24 = 0); b a = ( 13, 34, 12) (always b a = a b); b b = 0.

Exercise 5:
(a) The computations of u v and v u as 3 3 determinants are identical, except for the
swapping of the 2
nd
and 3
rd
rows. This changes the sign of a determinant.
(b) By (a) v v = v v, so v v = 0.

Exercise 6:
(x, y, z) ((4, 7, 8))
((3, 1, 2))
if and only if it is orthogonal to both vectors. Such

vectors are scalar multiples of the cross product.
(4, 7, 8) (3, 1, 2) =
i j k
4 7 8
3 1 2
= (6, 32, 25).
(Check: 4.6 7.32 + 8.25 = 24 224 + 200 = 0.)
So the answer is ( (( ((6, 32, 25)) )) ).

Exercise 7:
(a) f is a linear transformation. ker f is the orthogonal complement of ((1, 2, 3)). Hence dim
ker f = 2. Thus the nullity = 2 and the rank is thus 1 (ie 3 2).
(b) g is a linear transformation. ker g = {v | v a = 0} = (a).
26
So the nullity is 1 and the rank is 2.
(c) h is not a linear transformation and so rank and nullity are not defined.
[eg h(2v) = 2v.2v = 4(v.v) = 4h(v), not 2h(v) ]

Exercise 8: AB = ((1, 2, 3) (5, 1, 5))
= ((7, 10, 9))
.

Exercise 9: PQ = ((5, 2, 3))
. Since (3, 1, 4).(5, 2, 3) = 15 2 12 = 1 0, R does not

lie on PQ. (Alternatively we can put the vectors for P, Q and R as the rows of a 3 3 matrix
and show that the determinant is non-zero).

Exercise 10: Since a, b, c are clearly linearly dependent.

Exercise 11:
(i) PQ = ((1, 1, 1) (1, 2, 3)) = ((1, 2, 1))
,
RS = ((2, 3, 1) (3, 1, 2)) = ((5, 1, 7))
,
PQ RS = ((1, 2, 1) (5, 1, 7)) = ((15, 12, 9)) = ((5, 4, 3)).
(ii) PQ is the line y = 1, RS is the line x + y = 2, PQ RS = (1, 1).
(iii) PQ = R
, RS = N
, PQ RS = A.
(ii) P = (2, 1), Q = the ideal point on horizontal lines, R = (2, 0), S = the ideal point on lines
of slope 1. These are points in the Euclidean plane extended to the real projective plane.

Exercise 12: AB = ((5, 12, 28))
and CD = ((2, 2, 1))
. If M = (m) then m is orthogonal

to both (5, 12, 28) and (2, 2, 1) and so is parallel to (5, 12, 28) (2, 2, 1)
= (68, 61, 14). Hence M = ((68, 61, 14)). Similarly AC = ((5, 5, 14))
and
BD = ((20, 3, 7))
and N = ((7, 245, 85)). Then MN = ((1755, 5682, 16233)). Now AD

= ((30, 3, 18))
and BC = ((3, 3, 7))
so AD BC = ((4155, 2424, 7329)). It would

lie on MN if the vectors (1755, 5682, 16233) and (4155, 2424, 7329) were orthogonal.
Since their dot product is 125452800 0 they are not orthogonal and so AD BC does not
lie on MN.

Exercise 13: Sorry to trick you! Of course theres no distinction between ideal points and
ordinary points in the projective plane itself. The concepts only make sense when we embed
a Euclidean plane in the projective plane. Depending on which embedding we take this point
could be ordinary or ideal.

Exercise 14: Since (1, 2, 1) lies on the plane x + 2y + 3z = 0 the line joining the origin to
(1, 2, 1) is parallel to and so does not cut . Hence ((1, 2, 1)) is an ideal point.

27
2. DESARGUES THEOREM

2.1. Orthogonality
Recall that two vectors a, b in R
3
are orthogonal if a.b = 0. Recall, too, that the
orthogonal complement of a subspace U is U

= {v | u.v = 0 for all u U}. In other words
U
is the set of all vectors that are orthogonal to every vector in U.

Now U + U
is always the total vector space, which in this case is R

3
, and clearly
U U
= 0 since the only vector that is orthogonal to itself is the zero vector.
Since dim(U + U
) = dim(U) + dim(U
) dim(U U
) we have
3 = dim(U) + dim(U
) 0 so dim(U) + dim(U
) = 3.
This means that if U is a projective point (dimension 1) then U
is a projective line
(dimension 2) and vice versa. So U U
is a 1-1 correspondence between the projective

points and the projective lines in the real projective plane.
This means that, just as we can represent any projective point as (p) using one vector,
so any projective line can be represented in the form (q)
. Geometrically you can think of

(p) as the line through the origin that passes through p and (q)
as the plane through the

origin that is perpendicular to q. So the projective point (p) lies on the projective line (q)

if and only if p and q are orthogonal.

Theorem 1:
(i) If P = ( (( (p) )) ) and Q = ( (( (q) )) ) are two distinct projective points then the projective line PQ
is ( (( (p q) )) )

.
(ii) If h = ( (( (a) )) )

and m = ( (( (b) )) )

are two distinct projective lines then h m = ( (( (a b) )) ).
Proof: (i) Since p and q are each orthogonal to p q then (p) and (q) both lie on
(p q)
.
(ii) Since a b is orthogonal to both a and b, (a b) (a)
(b)
. But both of these

subspaces are projective points and so have dimension 1, which means they must be equal.

Example 1: Does ((1, 3, 2)) lie on the line ((11, 1, 7))
?
Solution: Yes, since (1, 3, 2).(11, 1, 7) = 11 3 + 14 = 0.

Example 2: If A = ((1, 3, 2)) and B = ((5, 1, 9)) find the line AB in the form (p)
.
Solution: AB = ((1, 3, 2), (5, 1, 9)) = (p)
where p is a non-zero vector orthogonal to both

(1, 3, 2) and (5, 1, 9). Clearly such a vector is the cross-product.
(1, 3, 2) (5, 1, 9) =
i j k
1 3 2
5 1 9
= (25, 1, 14) so AB = ((25, 1, 14))
.

Example 3: If h = ((1, 3, 2))
and k = ((5, 1, 9))
find h k.
Solution: We need to find a vector that is orthogonal to both vectors (1, 3, 2) and
(5, 1, 9). Once again we may take the cross-product (1, 3, 2) (5, 1, 9) = (25, 1, 14).
So h k = ((25, 1, 14)).
28
Example 4: Suppose A = ((1, 3, 5)), B = ((1, 1, 1)) and h = ((2, 1, 0))

. Find AB h.
Solution: (1, 3, 5) (1, 1, 1) =
i j k
1 3 5
1 1 1
= (8, 4, 4) so AB = ((8, 4, 4))
= ((2, 1, 1))
.
(2, 1, 1) (2, 1, 0) =
i j k
2 1 1
2 1 0
= (1, 2, 0) so AB h = ((1, 2, 0)).

The following table translates properties of projective points and lines into the concepts of
linear algebra and back again. In proving theorems of projective geometry by linear algebra
techniques we convert our assumptions into the language of linear algebra, work with them
using the standard techniques of linear algebra and then convert our conclusions back to the
language of projective geometry.

Let P = (p), Q = (q), R = (r) be projective points and h = (u)
, k = (v)
, m = (w)
be
projective lines (where all these vectors are non-zero).

projective point 1-dimensional subspace
projective line 2-dimensional subspace

P = Q p = q for some real 0
P lies on h p.u = 0
P, Q, R are collinear p, q, r are linearly dependent
h, k, m are concurrent u, v, w are linearly dependent
projective line PQ 2-dimensional subspace P + Q = (p q)
intersection of projective lines h, k 1-dimensional subspace h k = (u v)

2.2. The Collinearity Lemma
Definition: If P = ((x, y, z)) is a projective point, we say that x, y, z are homogeneous
coordinates for P.
Of course theyre not unique. For example ((1, 2, 3)) = ((2, 4, 6)). This fact is
exploited by the following lemma which provides a particularly simple set of homogeneous
coordinates for three or four collinear points.

Theorem 2: If P = ( (( (p) )) ), Q, R, S are collinear projective points such that P, Q, R are
distinct and P S, then for a suitably chosen vector q and scalar we may express
the four points as: P = ( (( (p) )) ), Q = ( (( (q) )) ), R = ( (( (p + q) )) ), S = ( (( ( p + q) )) ).
Proof: We break the proof into 12 separate steps.
(1) Since P = (p) has dimension 1, p 0.
(2) Since P lies on QR = Q + R we may write p = q
1
+ r for some q
1
Q, r R.
(3) Now if r = 0 we would have p = q
1
and so P = Q, a contradiction. Hence r 0.
(4) Thus R = (r).
(5) Now r = p q
1
= p + q if we define q = q
1
.
(6) If q = 0 then r = p and so R = P, a contradiction. Hence q 0.
29
(7) Since any non-zero vector in a 1-dimensional subspace spans that subspace we must have
Q = (q) and R = (r) = (p + q).
(8) Since Q lies on PS, q P + S = (p) + S.
(9) Thus q =
1
p + s for some
1
R and s S.
(10) Now s = (
1
)p + q = p + q if we define =
1
.
(11) If s = 0 then q =
1
p and so Q = P, a contradiction. Hence s 0.
(12) Therefore S = (s) = (p + q).

2.3. Perspective Triangles
Definition: A triangle ABC is a set of distinct projective points A, B, C together with
the projective lines AB, AC, BC.

Definition: ABC is in perspective with ABC from the point P if:
{P, A, A}, {P, B, B}, {P, C, C} are three sets of collinear points.

Definition: ABC is in perspective with ABC from the line h if:
{h, AB, AB}, {h, AC, AC}, {h, BC, BC} are three sets of concurrent lines.

A
P
A
B
B
C
C
A
B
C
A
A
B
C
C
B
h
30
2.4. Desargues Theorem
Girard Desargues (1591-1661), a French architect and mathematician who lived in
Lyons and Paris, was one of the founders of projective geometry. In 1639 he introduced
many of the basic concepts projective geometry.

Theorem 3: (Desargues) Two triangles are in perspective from a point if and only if they
are in perspective from a line.
Proof:

Suppose {P, A, A}, {P, B, B} and {P, C, C} are three collinear sets of points on distinct
lines. (The result is trivial if we allow two of the lines to coincide.)
Let L = AB AB, M = AC AC and N = BC BC.
We must show that L, M, N are collinear.
By the collinearity lemma we may write, for suitable vectors p, a, b, c:
P = (p), A = (a), A = (p + a),
B = (b), B = (p + b),
C = (c), C = (p + c)
Now a b A + B and a b = (p + a) (p + b) A + B. Since a b, a b is a non-
zero vector. Let L = (a b). Since L is a subspace of A + B as a projective point L lies
on the projective line AB. Similarly L lies on AB and so L = (a b) = AB AB.
By defining M, N similarly we have:
M = (c a) = AC AC and
L = (b c) = BC BC.
Now (a b) + (b c) + (c a) = 0 so a b, b c and c a are linearly dependent.
Hence L, M, N are collinear.

This proves only one half of the theorem, namely that if two triangles are in perspective from
a point they are in perspective from a line. We ought now to prove the converse. But shortly
well develop the very powerful Principle of Duality which will enable us to do that with
virtually no further effort!
P
A
A
B
B
C
C
L
M
N
31
2.5. Euclidean Interpretation of Desargues Theorem
Since Desargues Theorem is true for the Real Projective Plane it must hold for the
Real Affine Plane inside it, that is, ordinary points and ordinary lines relative to some
embedding of the Real Affine Plane in the Real Projective Plane. This is the case illustrated
by the above diagram. However if some of the points are taken to be ideal points we obtain
different affine interpretations of the same theorem. If we had attempted to prove them just
for the affine plane wed need separate proofs for each of them. But by working in the real
projective plane we get them all with no extra effort!

Interpretation 1: L ideal

If AB || AB and M = AC AC and N = BC BC then MN || AB.

Interpretation 2: Both L, M ideal

Since L, M, N are collinear N is an ideal point.
So if AB || AB and AC || AC then BC || BC.

Interpretation 3: P is ideal

M
P
N
L
N
M
A
A
B
B
C
C
P
L
P
32
If AA, BB and CC are parallel then L = AB AB, M = AC AC and
N = BC BC are collinear.
The great power of Projective Geometry is illustrated here. Quite apart from the fact
that algebraic methods are simpler than geometric ones, and what could be simpler than
(a b) + (b c) + (c a) = 0, we can prove several affine theorems by taking suitable
interpretations of a single projective theorem.

But wait! Theres more! The Principle of Duality means that every time we prove a
projective theorem we get a second one for free. And this, too, will have several affine
interpretations.


Exercise 1: Let A = ((1, 2, 4)), B = ((5, 3, 2)), C = ((3, 7, 6)) and D = ((13, 13, 2)).
(a) Find AB and show that C, D both lie on AB.
(b) Use the proof of the Collinearity Lemma to find vectors a, b and a scalar such that
A = (a), B = (b), C = (a + b) and D = (a + b).

Exercise 2: If P = ((1, 2, 8)), Q = ((4, 1, 5)), R = ((5, 4, 14)), S = ((14, 7, 31)), find vectors
p, q and a scalar such that P = (p), Q = (q), R = (p + q), S = (p + q).

Exercise 3:
(a) In the following diagram AF is parallel to BD and AG is parallel to BE. Find two
triangles that are in perspective from a point.

(b) State the Euclidean interpretation of Desargues Theorem for this configuration.

Exercise 4: Let A = ((1, 0, 0)), B = ((0, 1, 0)), C = ((0, 0, 1)),
A = ((2, 1, 1)), B = ((2, 3, 2)), C = ((3, 3, 4)).
Let L = AB AB, M = AC AC and N = BC BC.
Find L, M and N and verify that they are collinear.

A
B
D
E
F
G
C
33
Exercise 5: In each of the following cases find a point O and a line h such that the
triangles ABC and ABC are in perspective from the point O and from the line h.
(For simplicity corresponding points have corresponding labels.)
(i) A = (0, 2), B = (1, 1), C = (1, 0), A = (1, 2), B = (2, 1), C = (0, 0).
These points are in the Euclidean plane embedded in the real projective plane.

(ii) A = ((1, 4, 7)), B = ((2, 1, 3)), C = ((3, 5, 3)),
A = ((2, 3, 5)), B = ((1, 1, 1)), C = ((1, 4, 2)).

Exercise 6: In the following diagram A, B, P are not collinear while
ACXYZB, AQP, RQX, RSZ, RPB, QSB are straight lines.
Let
A = (a), R
B = (b),
C = (a + b), P
X = (a + xb), Q
Y = (a + yb),
Z = (a + zb), S
P = (p),
Q = (a + p).
(i) Prove that R = (p xb ); A C X Y Z B
(ii) Find S in terms of a, b, p, x, y, z;
(iii) Use the fact that RSZ are collinear to prove that x + y = z.

Exercise 7:
Let A, B, C and D be four distinct points, no three of which are collinear. Let S and R be
points on AB and AD respectively, and let K = AC BD, M = KS BC and N = KR CD.
Prove that the lines BD, MN and RS are concurrent.

Exercise 1: AB = ((1, 2, 4) (5, 3, 2))
= ((16, 18, 13))
. Since (3, 7, 6).(16, 18, 13) =

0 and (13, 13, 2).(16, 18, 13) = 0, C, D lie on AB.
Suppose (3, 7, 6) = x(1, 2, 4) + y(5, 3, 2).
Then
x + 5y = 3
2x 3y = 7
4x + 2y = 6
.
Solving, we get x = 2, y = 1.
Hence (3, 7, 6) = 2(1, 2, 4) (5, 3, 2). Therefore we let a = 2(1, 2, 4) = (2, 4, 8) and
b = (5, 3, 2) = (5, 3,2). Then A = (a), B = (b) and C = (a + b).

Suppose (13, 13, 2) = x(1, 2, 4) + y(5, 3, 2).
Then
x + 5y = 13
2x 3y = 13
4x + 2y = 2
.
34
Solving, we get x = 2, y = 3.
Hence (3, 7, 6) = 2(1, 2, 4) + 3(5, 3, 2) = a 3b.
Let c = (1/3)(3, 7, 6). Then c = (1/3)a + b.
Hence a = (2, 4, 8), b = (5, 3, 2) and = 1/3.

Exercise 2:
Let x (1, 2, 8) + y (4, 1, 5) + z (5, 4, 14) = (0, 0, 0).
\
|
.
|
|
| 1 4 5
2 1 4
8 5 14

\
|
.
|
|
| 1 4 5
0 7 14
0 27 54

\
|
.
|
|
|
1 4 5
0 1 2
0 0 0
. Let z = 1. y = 2, x = 3.
3(1, 2, 8) + 2(4, 1, 5) + (5, 4, 14) = (0, 0, 0).
(5, 4, 14) = 3(1, 2, 8) 2(4, 1, 5).
Let p = (3, 6, 24) and q = (8, 2, 10).
P = (p), Q = (q), R = (p + q).
Let x (1, 2, 8) + y (4, 1, 5) + z (14, 7, 31) = (0, 0, 0).
\
|
.
|
|
|
1 4 14
2 1 7
8 5 31

\
|
.
|
|
|
1 4 14
0 7 21
0 27 81

\
|
.
|
|
|
1 4 14
0 1 3
0 0 0
. Let z = 1. y = 3, x = 2.
(14, 7, 31) = 2(1, 2, 8) + 3(4, 1, 5) = (2/3)p (3/2)q
(3/2)(14, 7, 31) = (4/9)p + q.
So S = (p + q) where = 4/9.

Exercise 3: Triangles AFG and BDE are in perspective from C.
Let L = AF BD, M = AG BE and N = FG DE. Then L, M, N are collinear. Since AF
is parallel to BD and AG is parallel to BE, the points L, M are ideal points and so LM is the
ideal line. Thus N must also be an ideal point. It follows that FG is parallel to DE.

Exercise 4:
AB = ((0, 0, 1))
, AB = ((1, 2, 4))
so L = ((2, 1, 0)).
AC = ((0, 1, 0))
, AC = ((1, 5, 3))
so M = ((3, 0, 1)).
BC = ((1, 0, 0))
, BC = ((6, 2, 3))
so N = ((0, 3, 2)).
LM = ((1, 2, 3))
= ((1, 2, 3))
, MN = ((3, 6, 9))
= LM.
[We could have simply observed that 3(2, 1, 0) + 2(3, 0, 1) + (0, 3, 2) = (0, 0, 0) and so
L, M, N are collinear.]

Exercise 5:
(i)
A A

B B
O

C C

35
AA, BB, CC are all horizontal. So O is the ideal point on the horizontal lines. AC
AC = (1/2, 1). AB AB is the ideal point on the lines with slope 1. So h is the line
through (1/2, 1) with slope 1, ie the line x + y = 3/2.

(ii) (1, 4, 7) (2, 3, 5) = (1, 9, 5) so AA = ((1, 9, 5))
.
(2, 1, 3) (1, 1, 1) = (2, 5, 3) so BB = ((2, 5, 3))
.
O = ((1, 9, 5) ((2, 5, 3))
= ((52, 13, 13)) = ((4, 1, 1)).

CC = ((3, 5, 3) (1, 4, 2))
= ((2, 9, 17))
.
Since (2, 9, 17).(4, 1, 1) = 0, O lies on CC.
Thus the triangles are in perspective from O = ( (( ((4, 1, 1)) )) ).

AB = ((19, 11, 9))
and AB = ((8, 7, 1))
. So AB AB = ((52, 91, 221)) =

((4, 7, 17)).
BC = ((18, 3, 13))
and BC = ((6, 1, 5))
so BC BC = ((28, 168, 0)) =

((1, 6, 0)).
Thus h is ((4, 7, 17) (1, 6, 0))
= ((102, 17, 17))
= ( (( (( 6, 1, 1)) )) )

.
We check that AC AC lies on h.
AC = ((23, 18, 7))
and AC = ((14, 9, 11))
so AC AC = ((135, 351, 459))

= ((5, 13, 17))
. Since (5, 13, 17).(6, 1, 1) = 0 AC AC lies on h.

Exercise 6: (i) p xb = (a + p) (a + xb) PB QX so R = (p xb).
(ii) (a + p) + yb = (a + yb) + p QB YP = S so S = (a + p + yb).
(iii) Since R, S, Z are collinear there exist scalars , , , not all zero, such that:
(p xb) + (a + p + yb) + (a + zb) = 0.
Since p, a, b are linearly independent we have:
+ = 0, + = 0, x + y + z = 0. Thus (x + y z) = 0.
Now 0 so x + y = z.

Exercise 7:

Triangles DNR and BMS are in perspective from the line AKC and hence they are in
perspective from a point, by the converse of Desargues Theorem. Hence the lines BD, MN
and RS are concurrent.
A
B
S
M
D
R
N
K
C
36

37
3. PAPPUS THEOREM

3.1. Duality
If P is a projective point (1-dimensional subspace) then P
is a projective line (2-

dimensional subspace) and if h is a projective line then h
is a projective point. The

relation P P
(or equivalently h h
) establishes a 1-1 correspondence between the

points and lines of the Real Projective Plane. Moreover it interacts with the incidence
structure in a very nice way.

Theorem 1: The projective point P lies on the projective line h if and only if the
projective point h

lies on the projective line P

.
Proof: Let P = p and let h = v
. Then P lying on h is equivalent to p and v being

orthogonal. But h
= v
= v and so h
lying on P
is also equivalent to p and v

being orthogonal. So the two geometric statements are equivalent to one another.

As a consequence of this very innocent-looking theorem we can establish the following
very powerful Principle of Duality.

The Principle of Duality
Any theorem in projective geometry that can be expressed in terms of the following six
concepts remains true if these concepts are interchanged as follows:
projective point projective line
lies on passes through
collinear concurrent

A concept or theorem that is obtained by the above interchanges is called the dual of
the original one. What this principle means is that every time we prove a theorem in
projective geometry we automatically have proved another theorem, its dual. Well thats not
quite true because sometimes a theorem is its own dual. But in the case of Desargues the
dual of what we proved is the converse.
The dual of the property of two triangles being in perspective from a point is two
triangles being in perspective from a line. (Its not just because we have changed the word
point to line. You need to look at their definitions to see that the definitions are duals of one
another.) We proved that if triangles are in perspective from a point then theyre in
perspective from a line. The dual, which must be true by the principle of duality, is that if
triangles are in perspective from a line then theyre in perspective from a point.
Sometimes, as in Desargues theorem, the dual turns out to be the converse.
Sometimes it turns out to be the same theorem (a self-dual theorem). Sometimes its a totally
different theorem. This is the case with the next theorem.
38
3.2 Pappus Theorem
Pappus of Alexandria (c. 300 A.D.) was a Greek mathematician who provided a
particularly simple proof of the equality of the base angles of an isosceles triangle. His great
work A Mathematical Collection is an important source of information about ancient Greek
mathematics.

Theorem 2: Suppose {A, B, C} and {A , B , C } are two collinear sets where the 6
points are distinct and the two lines are distinct.

Let Q = AB A B, R = AC A C and S = BC B C. Then Q, R, S are collinear.
Proof: Let P = AB AB. The theorem is trivial if P coincides with any of the six points
so we may assume that it is distinct from each of them.
By the Collinearity Lemma we may choose vectors a, a and scalars , such that:
P = p;
A = a, A = a;
B = p + a, B = p + a;
C = p + a, C = p + a.

Let q = (p + a) + a = (p + a) + a (A + B) (A + B). Hence Q = q.
Let r = p + a + a = (p + a) + a = (p + a) + a (A + C) (A + C).
Hence R = r.
Finally let s = q r = p + a + a p a a
= (1 )(p + a) + (1 )(p + a) B + C
= (1 )(p + a ) + (1 )(p + a) B + C.
Hence S = s.
Since q = r + s it follows that q, r, s are linearly dependent and so Q, R, S are collinear.
A
P
B
C
A
B
C
Q
S
R
39
3.3 Finite Projective Planes
So far weve only discussed the Real Projective Plane. We took the field of real
numbers R and formed the 3-dimensional vector space over it, R
3
. Our projective points
were 1-dimensional subspaces of R
3
and our lines were 2-dimensional subspaces.
Everything weve done so far would have worked if wed taken any field F and
worked within the 3-dimensional vector space F
3
. Our scalars would have been the elements
of F. Every piece of algebra we carried out would have been valid for any such field. (Note
that there was one place where we needed to assume that the scalars commute under
multiplication, which of course is valid in any field. Can you find where we invoked the
commutative law for multiplication?)
We denote the projective plane that is formed from the field F by the symbol (F). So
the real projective plane is (R).
We could, instead of the field of real numbers, take the field of complex numbers and
so produce the Complex Projective Plane (C). Or we could have manufactured the
Rational Projective Plane (Q). All our theorems, including Desargues and Pappus
theorems would remain valid. (We may have had difficulty in drawing diagrams in the
Complex Projective Plane but the diagrams are not essential to the proofs.)
We could also take our field to be finite and so produce finite projective planes. Again
diagrams would be difficult to draw but again there would be no need. In a finite projective
plane, projective lines have finitely many points so we need only list the points to completely
describe the line.
The simplest finite fields are the integers modulo a prime. The field of integers modulo
p is denoted by Z
p
. The smallest field is Z
2
, the field with 2 elements. These elements are
written 0, 1 and addition and multiplication can be described by the tables:

+ 0 1

0 1
0 0 1 0 0 0
1 1 0 1 0 1

Lets examine the smallest projective plane, (Z
2
). The field Z
2
has 2 elements. The
vector space Z
2
3
has 2
3
= 8 elements. Of these one, the zero vector (0, 0, 0), doesnt span
a projective point. The other 7 do. Moreover, because theres only one non-zero scalar,
each of these 7 vectors spans a different projective point. So (Z
2
) has 7 points:
(0, 0, 1), (0, 1, 0), (0, 1, 1), (1, 0, 0), (1, 0, 1), (1, 1, 0), (1, 1, 1).
By duality there must be the same number of projective lines:
(0, 0, 1)
, (0, 1, 0)

, (0, 1, 1)

, (1, 0, 0)

, (1, 0, 1)

, (1, 1, 0)

, (1, 1, 1)

.
For example the projective point (0, 1, 1) lies on the projective line (1, 1, 1)

since
(0, 1, 1).(1, 1, 1) = 0 + 1 + 1 = 0.
How many points are there on each of these 7 lines? A projective line is a
2-dimensional subspace and so has 2
2
= 4 vectors. Of these, the zero vector doesnt span a
projective point but the other three do. They span three distinct projective points. So on each
line in (Z
2
) we have three points. And how many lines through each point? Thats easy
by duality theres the same number, three.
40
Now its very messy working with these vectors and having to do a modulo 2 dot
product every time we want to see if a point lies on a line. Its much easier to code them and
to construct a Collinearity Table.
We code the points as A, B, C, D, E, F, G and the lines as a, b, c, d, e, f, g.

A = (0, 0, 1), a = (0, 0, 1)
;
B = (0, 1, 0), b = (0, 1, 0)

;
C = (0, 1, 1), c = (0, 1, 1)

;
D = (1, 0, 0), d = (1, 0, 0)

;
E = (1, 0, 1), e = (1, 0, 1)

;
F = (1, 1, 0), f = (1, 1, 0)

;
G = (1, 1, 1). g = (1, 1, 1)

.
Now, with a certain amount of calculation we can work out which points lie on which
line and so produce the following Collinearity Table.

a b c d e f g
B A C A B A C
D D D B E F E
F E G C G G F

The columns of this table list the seven lines, and for each one, the three points on the
line. Look closely at the table. Any two capital letters occur together in exactly one column
(any two points lie on exactly one line) and any two columns have exactly one capital letter
in common (any two lines intersect in exactly one point).
Theres no longer any need to go back to the original vectors. Every geometric
question for (Z
2
) can be answered by just examining this table.

Question 1: Does B lie on the line g?
Answer: No, since B isnt in column g.

Question 2: What is c f?
Answer: G, since only G is in both columns.

Question 3: What is the line AD?
Answer: b since thats the column where both A and D occur.

Question 4: Are the points A, F, G collinear?
Answer: Yes, since they all lie on the line f.

Question 5: Are the lines b, c, g concurrent?
Answer: No, since theres no letter common to all three columns.

Question 6: Are the triangles CDE and AFG in perspective from a point?
Answer: Yes, theyre in perspective from B since {B, D, F}, {B, C, A} and {B, E, G} are
three sets of collinear points.

41
Question 7: Are the triangles CDE and AFG in perspective from a line?
Answer: Yes, by question 6 and Desargues Theorem.

Question 8: Which line?
Answer: CD AF = c f = G
CE AG = g f = F
DE FG = b f = A.
Note that A, F, G are collinear, as predicted by Desargues Theorem. From the table we see
that this line is f. So the triangles are in perspective from the line f.

In the above triangle no attempt has been made to put the points in the right places
right places doesnt make sense for the 7-point projective plane. In fact weve had to
place E in two places. The diagram is merely an aid for us to remember the three collinear
sets of projective points.
Note that theres a certain amount of degeneracy in this example in that the triangle A,
F, G consists of 3 collinear points. Its impossible in this tiny example, with only seven
points, to avoid getting a degenerate triangle. Nevertheless Desargues Theorem still works
even if one of the triangles is degenerate.

Question 9: Which of the seven lines is the ideal line?
Answer: This is a non-question. Remember that any line can be considered to be the ideal
line by placing the relevant affine plane appropriately. For example if we want d to be the
ideal line then its three points A, B, C will become the ideal points. Removing the line d
and the three points A, B, C we are left with the four ordinary points D, E, F, G and the
six ordinary lines a, b, c, e, f, g. There are two ordinary points on each ordinary line, as
follows:

a b c e f g
D D D E F E
F E G G G F

Now we have lines which dont intersect, and which are therefore considered to be parallel.
For example a || e. These lines dont intersect in an ordinary point, but, going back to the
original table, they intersect in the ideal point B.
It must be emphasized that this distinction between ideal and ordinary points depended
on our arbitrary choice of d as ideal line. Any line could have been chosen and whether a
point is ordinary or ideal would change. In the projective plane itself theres no such
distinction.
D
B
E
C
A
E
G
42
It would be nice if we could draw a picture of the 7-point projective plane. Of course
we cant, not in the true sense, because our pictures must live in a Euclidean plane. But the
following picture can be useful.

Here we have a diagram with seven points and seven lines. OK so one of the lines
looks more like a circle, but thats the best we can do in the Euclidean plane. There are three
points on each line and three lines through each point. Each pair of distinct points lies on
exactly one line and each pair of distinct lines intersect in exactly one point.

We can consider the projective plane (Z
3
) over the field of integers modulo 3. This
field has the following addition and multiplication tables:

+ 0 1 2

0 1 2
0 0 1 2 0 0 0 0
1 1 2 0 1 0 1 2
2 2 0 1 2 0 2 1

There will now be 3
3
= 27 vectors. Removing the zero vector we are left with 26 non-
zero vectors. But because there are now 2 non-zero scalars each projective point will be
spanned by 2 different vectors. So the 26 non-zero vectors in fact only give us 26/2 = 13
distinct points. And, by duality, there must be 13 lines. Each line is a 2-dimensional
subspace containing 3
2
= 9 vectors, of which 8 are non-zero. But these will only give us 8/2
= 4 points on the line because there are 2 non-zero scalars and 2v = v.
So the 13-point projective plane has 13 points and 13 lines with 4 points on each line
and 4 lines through each point.
By working through the process of constructing finite projective planes one can obtain,
by suitably coding the 1-dimensional subspaces, the following collinearity table. Just for fun
well use the names of the 13 cards in a suit in a pack of cards to denote the 13 points and the
same labels for the 13 lines. The entries at the top of the columns denote the lines and the
entries in the body of the table give the 4 points on that line.

A 2 3 4 5 6 7 8 9 10 J Q K
9 7 Q Q J 10 K Q J A K 5 4
J 6 8 10 8 A 8 9 10 Q 10 K A
7 5 5 7 6 8 7 6 5 J 6 9 2
4 A 4 3 3 9 2 2 2 K 4 3 3

Note that every pair of points occurs in exactly one column and that every pair of
lines has exactly one common point.
43
The next smallest projective plane is the 21-point plane that arises from the field with 4
elements. But this field is not Z
4
because 4 isnt prime. Its denoted by the symbol GF(4)
(and is also called a Galois field after the famous mathematician Galois). It arises by
starting with the field Z
2
and adjoining to it a symbol x that satisfies the rule that
x
2
= x + 1. There are four elements 0, 1, x and x + 1. (Any higher powers are unnecessary
because x
2
= x + 1.) The addition and multiplication tables for GF(4) are as follows.

THE FIELD GF(4)
+ 0 1 x x + 1

0 1 x x + 1
0 0 1 x x + 1 0 0 0 0 0
1 1 0 x + 1 x 1 0 1 x x + 1
x x x + 1 0 1 x 0 x x + 1 1
x + 1 x + 1 x 1 0 x + 1 0 x + 1 1 x

We can code the expressions x and x + 1 as 2, 3 respectively to give more compact tables:
THE FIELD GF(4)
+ 0 1 2 3 0 1 2 3
0 0 1 2 3 0 0 0 0 0
1 1 0 3 2 1 0 1 2 3
2 2 3 0 1 2 0 2 3 1
3 3 2 1 0 3 0 3 1 2

3.4 Combinatorial Applications
Finite projective planes are useful in certain combinatorial situations. The following is
a very simple example where such a combinatorial problem can be solved.

Problem: A chemist shop operates 12 hours a day 7 days a week. It needs to have 3 people
on duty on any given day and, because of the long days, they each work 3 days a week. Now
its desirable, for the sake of continuity, that at least one of the three people on duty on any
given day was one of the three working on any other day (so that a customer could come in
on Friday, for example, and say I had this prescription dispensed last Tuesday and one of
the Friday staff would know what she was talking about, having worked on the Tuesday).
Draw up such a roster.
Solution: With 7 days, and 3 people required each day, there are 21 slots to be filled on a
weekly roster. With a 3 day working week its clear that exactly 7 staff are needed
provided we could meet the requirement that the staff for any 2 days have at least one person
in common.
Well, of course, we simply assign each member of staff to a point in the 7-point
projective plane. The staff roster for the 7 days can be obtained by simply taking the 7
projective lines. So the table above for the 7-point projective plane will give a suitable roster.
(Simply allocate the 7 labels A to G to the staff and the 7 labels a to g to the days of the
week.)
44

Exercise 1: Write out, with diagrams, at least six Euclidean interpretations of Pappus
Theorem.

Exercise 2: State the dual of Pappus Theorem and illustrate it with a diagram.

Exercise 3: In (Z
7
), let A = (1, 2, 3), B = (1, 1, 4), C = (5, 0, 1), D = (1, 1, 1).
Find AB CD.

Exercise 4: (a) Find the 13 points of the projective plane (Z
3
). Label them A, 2, 3, 10,
J, Q, K (as in a pack of cards).
(b) Construct a collinearity table for (Z
3
) in terms of these labels. It will have 13 columns
and 4 rows. The columns correspond to the 13 lines, giving the 4 points on each line. (The
order of the rows and columns does not matter, but try to be systematic.)
(c) Use your table to illustrate theorems of Desargues and Pappus.

Exercise 5: If F is the field Z
p
of integers modulo p how many points and how many lines,
are there in (F)? How many points lie on each line? How many lines pass through each
point?

Exercise 6:
Let A, B, C and D be four points in the real projective plane, no three of which are collinear.
Let U, V and W denote the points of intersection of various lines as indicated in the diagram.
Suppose that A, B, C, V and W have vector representations A = <a>, B = <b>, C = <c>,
V = <a + b> and W = <a + c>.

(i) Show that D = <a + b + c> and find a vector for U.

(ii) Show that U, V and W are never collinear in (R), whereas they are always collinear in
(Z
2
). What happens in (Z
3
)?
A
B
V
U
D
C
W
45
Exercise 7: Suppose P, A, C are three non-collinear points in the following configuration.
B
A

P
Q
C

D

R
Let P = p, A = a, B = p + a, C = c, D = p + c and let Q = AD BC, R = AC BD.

(i) Explain why R = a c and Q = p + a + c.
(ii) Prove that if F is the field of real numbers then P, Q, R cannot be collinear.
(iii) Find a field F in which P, Q, R are collinear.
(iv) Let S = PR BC, T = AS BR, U = AR PQ. Write down S, T, U in terms of p, a, c.
(v) Prove that if F is the field Z
3
then P, U, T are collinear.

Exercise 8:
In the following diagram in (R), let P = p, A = a, B = a + c, C = c, D = p + c.
Construct the points p + 2a and p + (3/2)a.

Exercise 9: In (Z
5
), let P = p, A = a, B = a + c, C = c, D = p + c.
Give a construction for p + 3a, as a sequence of line intersections. (Dont attempt to draw
the construction.)
P
A
C
D
B
46
SOLUTIONS TO CHAPTER 3

Exercise 1:
(1) Suppose P, A, B, C, A, B, C are distinct points with P, A, B, C and P, A, B, C forming
two distinct lines.
Let Q = AB AB, R = AC AC and S = BC BC. Then Q, R, S are collinear.

(2) Suppose A, B, C, A, B, C are distinct points with A, B, C and A, B, C forming two
distinct lines and with ABC parallel to ABC.
Let Q = AB AB, R = AC AC and S = BC BC. Then Q, R, S are collinear.

(3) Suppose P, A, B, A, B, C are distinct points with P, A, B and P, A, B, C forming
distinct lines. Let m be the line through A parallel to PAB and let n be the line through B
parallel to PAB. Let Q = AB AB, R = m AC and S = n BC. Then Q, R, S are
collinear.

A
B
C
A B C
Q S
R
P
A
P
B
C
A
B
C

Q
S
R
A
P
B
A
B
C
Q
R
C
47
(4) Suppose A, B, C are distinct points with A, B, C collinear. Let m, n, r be lines such
that no two of ABC are parallel. Let u be the line through A parallel to r. Let v be the
line through B parallel to n. Let w be the line through A parallel to m. Let x be the line
through C parallel to n. Let y be the line through B parallel to m. Let z be the line
through C parallel to r. Let Q = u v, R = w x and S = y z. Then Q, R, S are
collinear.

(5) Suppose P, A, B, A, B are distinct points with P, A, B and P, A, B forming distinct
lines. Let m be the line through A parallel to PAB and let n be the line through A
parallel to PAB. Let u be the line through B parallel to PAB and let v be the line through
B parallel to PAB. Let Q = AB AB, R = m n and S = u v. Then Q, R, S are
collinear.

(6) Suppose P, A, B, C, A, B, C are distinct points with P, A, B, C and P, A, B, C forming
distinct lines. Suppose AB is parallel to AB.
Let R = AC AC and S = BC BC. Then RS is parallel to AB.

A
R
B
Q
C
u
v
P
S
w
x
y
z
A
P
B
A
B
C
Q
R
C
S
48

Exercise 2:
Suppose {a, b, c} and {a, b, c} are two concurrent sets where the 6 lines are distinct and
the two points of concurrency are distinct.
Let q = (a b)(a b), r = (a c)(a c) and S = (b c)(b c).
Then q, r, s are concurrent.

Exercise 3: Although cross products only have a true geometric interpretation in R
3
,
algebraically they give a vector orthogonal to two given vectors in F
3
for any field, provided
we do the arithmetic in that field.
AB = (1, 2, 3) (1, 1, 4)
= (5, 6, 6)
and CD = (5, 0, 1) (1, 1, 1)
= (6, 3, 5)
.
Hence AB CD = (5, 6, 6) (6, 3, 5) = (5, 4, 0).
A
P
B
C
B
A
C
Q
R
S
a
p
b
a
b
c
q
s
r
c
49
Exercise 4:
A = (0, 0, 1); 2 = (0, 1, 0);
3 = (0, 1, 1); 4 = (0, 1, 2);
5 = (1, 0, 0); 6 = (1, 0, 1);
7 = (1, 0, 2); 8 = (1, 1, 0);
9 = (1, 1, 1); X = (1, 1, 2);
J = (1, 2, 0); Q = (1, 2, 1);
K = (1, 2, 2).

A
2

3

4

5

6

7

8

9

X

J

Q

K

2 A 4 3 A 2 2 A 4 3 A 3 4
5 5 5 5 2 7 6 J 7 6 8 7 6
8 6 9 9 3 X 9 Q 9 X 9 8 8
J 7 K K 4 K Q K J J X Q K

(c) Desargues:
Triangles 528 and 739 are in perspective from A.
52 73 = 8; 58 79 = J; 28 39 = 5 and 5, 8, J are collinear (lie on A
).

Pappus:
{A, 3, 4} and {6, 8, 4} are sets of collinear points.
A8 36 = X, 34 48 = 4, A4 46 = 4 and X, 4, 4 are collinear.

Exercise 5: There are p
3
vectors in Z
p
3
. Of these p
3
1 are non-zero. Since there are
p 1 non-zero scalars a given 1-dimensional subspace is spanned by any one of p 1 non-
zero vectors.
So the number of distinct 1-dimensional subspaces is
p
3
1
p 1
= p
2
+ p + 1. So (Z
p
) has
p
2
+ p + 1 points and, by duality, p
2
+ p + 1 lines.
Each line is a 2-dimensional subspace containing p
2
1 non-zero vectors and hence, by the
above argument,
p
2
1
p 1
= p + 1 one-dimensional subspaces. Thus each line contains
p + 1 points and, by duality, each point has p + 1 lines passing through it.

Exercise 6:
(i) D = BW CV. Since a + b + c = b + (a + c), a + b + c BW.
Since a + b + c = c + (a + b), a + b + c) CV. Hence it is BW CV.
U = BC AD so U = b + c since b + c = (a + b + c) a.
(ii) If U, V, W are collinear then b + c, a + b and a + c are linearly dependent.
If x(b + c) + y(a + b) + z(a + c) = 0 then (y + z)a + (x + y)b + (x + z)c = 0.
Since a, b, c are linearly independent:
y + z = 0
x + y = 0
x + z = 0

50
Hence x = z from the first two equations and so 2x = 0 from the third. If the field is R, this
means that x = 0, and so y = z = 0. Thus b + c, a + b and a + c are linearly independent,
and so U, V, W are not collinear.
On the other hand, if F = Z
2
then (b + c) + (a + b) + (a + c) = 2(a + b + c) = 0 and
so U, V and W are collinear. If F = Z
3
then U, V and W are not collinear, as for R.

Exercise 7:
(i) a c = (p + a) (p + c) and so a c AC BD = R.
p + a + c = (p + a) + c = (p + c) + a BC AD = Q.
(ii) Suppose P, Q, R are collinear. Then p, p + a + c, a c are collinear.
Suppose xp + y(p + a + c) + z(a c) = 0. Then (x + y)p + (y + z)a + (y z)c = 0.
Hence:
x + y = 0
y + z = 0
y z = 0

From the last two equations, 2y = 0 and hence y = 0. It follows that x = z = 0, contradicting
the linear dependence.
(iii) If F = Z
2
then p (p + a + c) + (a c) = 2c = 0 and so P, Q and R are collinear.
(iv) S = p + a c, T = p + 2a c and U = a + c.
(v) 2p + (p + 2a c) + (a + c) = 3(p + a) = 0 if F = Z
3
.

Exercise 8:

Let E = AD BC = p + a + c.
Let F = AC BD = a c.
Let G = EF PA = p + 2a.
Let H = GC BD = p + 2a c.
Let K = HE PA = 2p + 3a = p + (3/2)a.
P
A
C
D
B
E
F
G
H
K
51
Exercise 9:
Let E = AD BC = p + a + c.
Let F = AC BD = a + 4c.
Let G = EF PA = p + 2a.
Let H = GC BD = p + 2a + 4c.
Let K = HE PA = 2p + 3a = p + 4a.
Let L = KC BD = p + 4a + 2c.
Let M = LE PA = p + 3a.
52

53
4. CROSS RATIO

4.1 Euclidean Cross Ratio
Very few quantitative properties of an object remain fixed when its drawn in
perspective. For example the following is a perspective drawing of a railway track.

The picture is very much smaller than the original, so lengths and areas are not
preserved. The sleepers on the actual track are at right angles to the lines yet in the
perspective drawing theyre not, so angles are not preserved.
Ratios of lengths along a line are not preserved because the sleepers are equally
spaced while they get closer and closer in the perspective drawing. But ratios of ratios along
any line are preserved.

Definition: Suppose A, B, C, D are four collinear points. Choose a direction along the line
as the positive direction (so that distances measured in the opposite direction are treated as
negative). The cross-ratio of these four points, in the given order, is defined to be:
(A, B; C, D) =
AC
BC

AD
BD
where AC, BC etc denote the signed distances between the
respective points. (Clearly the cross ratio is independent of the chosen direction, but does
depend on the order in which the points are taken.)

Example 1: 3 1 2

A B C D

(A, B; C, D) =
AC
BC

AD
BD
=
4
1

6
3
= 2.
As mentioned earlier, the cross ratio depends on the order of the points. In the above
example (C, A; D, B) =
CD
AD

CB
AB
=
2
6

1
3
= 1.
The fact (which we have yet to prove) that cross ratios are preserved when things are
drawn in perspective means that if four collinear points are projected from any point onto
another line, the image points have the same cross ratio as the original ones.
54

(A,B;C,D) = (A,B;C,D)

Example 2: Show that the following drawing cannot be a true perspective drawing of any
piece of railway track.

D
4
C
8
B

11

A

Solution: The cross ratio (A, B; C, D) on the drawing is
19
8

23
12
=
57
46
.
The corresponding four points on the original railway track have cross ratio
2
1

3
2
= 4/3.
Since these are different this picture cant be an accurate picture of the railway track.

4.2. Projective Cross Ratio
Theorem 1: The scalar in the Collinearity Lemma is unique (ie depends only on the
four points and the order in which they are listed.)
Proof: Suppose, using the Collinearity Lemma twice,
P = (p) = (p);
Q = (q) = (q);
R = (p + q) = (p + q);
S = (p + q) = (p + q).
Then for some scalars , , , :
p = p;
q = q;
A
B
C
D
A
B
C
D
55
p + q = (p + q);
p + q = (p + q).
Hence p + q = p + q, and since p, q are linearly independent, = = .
Also p + q = (p + q) and so
= ;
= .
Thus = = and so = .
Now 0 (since p 0) and so = .

Definition: The value of in the Collinearity Lemma is the Projective Cross Ratio of the
four points, (P, Q; R, S).

Example 3:
Find (P, Q; R, S) when P = ((1, 2, 3)), Q = ((4, 5, 6)), R = ((2, 1, 0)), S = ((2, 3, 4)).
Solution: Since were now working in the projective plane we cant use distances. Instead
we use the Collinearity Lemma.
Write (2, 1, 0) = x (1, 2, 3) + y (4, 5, 6).
Hence:
x + 4y = 2;
2x + 5y = 1;
3x + 6y = 0.
Normally three equations in two variables are inconsistent. The fact that theyre consistent
here reflects the fact that A, B, C are collinear.
\
|
.
|
|
|
1 4 2
2 5 1
3 6 0

\
|
.
|
|
|
1 4 2
0 3 3
0 6 6

\
|
.
|
|
|
1 4 2
0 1 1
0 0 0
giving the solution y = 1, x = 2.
R2 2R1, R3 3R1 R3 2R2, R2 (3)
Thus (2, 1, 0) = (2, 4, 6) + (4, 5, 6).
[In this case we could have simply substituted x = 2y from the third equation to give y = 1
from the others.]
So take p = (2, 4, 6) and q = (4, 5, 6).
Then P = (p), Q = (q), R = (p + q).

Now write (2, 3, 4) = x (2, 4, 6) + y (4, 5, 6).
Proceeding as above we get x = 1/3, y = 1/3.
Hence (2, 3, 4) = (1/3)p + (1/3)q.
Thus 3(2, 3, 4) = p + q.
So S = ( p + q), having the form (p + q) where = 1.
Hence (P, Q; R, S) = 1.
56
4.3. Reconciliation of Euclidean and Projective Cross Ratios
We must now show that the concepts of Euclidean and Projective Cross Ratios are
equivalent. We do this by showing that if the Euclidean Plane is embedded in the Real
Projective Plane and four collinear projective points correspond to four collinear ordinary
points, then the Projective Cross Ratio and Euclidean Cross ratio are equal.

But first we need to recall a fact from vector geometry about points that divide a line
segment in a given ratio.
If a, b represent two distinct points A, B in R
3
the typical point on the line AB is
represented by: p = (1 ) a + b where =
AP
PB
and where AP and PB represented the
directed lengths of the respective sub-intervals (taking the direction from A to B as
positive).

B
1

P

A

Special cases are: = 0, where P = A; = , where P is the midpoint of AB) and
= 1, where P = B.

For > 1 P lies beyond B (on the opposite side to A) and for < 0 P lies beyond A (on
the opposite side to B.) As ranges from to + the point P traverses the line in the
direction from A to B.

< 0 = 0 = = 1 > 1

A B

Theorem 2: Suppose is a Euclidean Plane that does not pass through the origin.
If A, B, C, D are distinct collinear points on and A*, B*, C*, D* are the
corresponding projective points OA, OB, OC, OD then the projective cross ratio of
A*, B*, C*, D* is equal to the Euclidean Cross Ratio of A, B, C, D.

A B C
D
A
*
B
*

C
*
D
*
O
57

Proof: By the Collinearity Lemma we can choose vectors a, b and a scalar such that:
A* = (a);
B* = (b);
C* = (a + b);
D* = (a + b).
Moreover is the projective cross ratio of the four projective points
A*, B*, C*, D*.
Now A, B, C, D lie on the vectors a, b, a + b, a + b so theyre represented by the vectors
a, b, (a + b), (a + b) for suitable scalars , , , .
Since C, D lie on the line AB we may write
(a + b) = (1 s)a + sb
and (a + b) = (1 t)a + tb
where s =
AC
AB
and t =
AD
AB
.

Now a, b are linearly independent so, equating coefficients:
= (1 s)
= s
= (1 t)
= t
from which it follows that
s
1 s
=
and
t
1 t
=
.
Since s =
AC
AB
, 1 s =
CB
AB
. Similarly 1 t =
DB
AB
.

A C B A D B
Thus
AC
BC
=
\
|
.
|
|
s
1 s
and
AD
BD
=
\
|
.
|
|
t
1 t
and so
AC
BC

AD
BD
=
s
1 s
.
1 t
t
=
= .

Example 4: Find the cross ratio (P, Q; R, S) for the four projective points in example 3 by
using the corresponding points in the plane x = 1.
Solution: The four projective points are:
P ((1, 2, 3)), Q = ((4, 5, 6)), R = ((2, 1, 0)), S = ((2, 3, 4)).
The corresponding points on the plane x = 1 are:
P = (1, 2, 3), Q = (1, 5/4, 3/2), R = (1, , 0), S = (1, 3/2, 2).
Using just the y- and z-coordinates we can write these as:
P = (2, 3), Q = (5/4, 3/2), R = (, 0), S = (3/2, 2).
Thus (P, Q; R, S) = (P, Q; R, S) =
PR
QR

PS
QS

58
=
PR.QS
QR.PS
=
PR.SQ
QR.PS
=
\
|
.
|
|
3
2
2
+ 3
2
.
\
|
.
|
|
1
4
2
+
\
|
.
|
|
1
2
2
\
|
.
|
|
3
4
2
+
\
|
.
|
|
3
2
2
.
\
|
.
|
|
1
2
2
+ 1
2
=
45
4
.
5
16
45
16
.
5
4
= 1.

4.4. Cross Ratio of Rearrangements
The cross ratio of four projective points depends on the order in which theyre
specified. There are 4! = 24 orders in which they could be given. However there are at
most 6 different values of the cross ratio that can be obtained because certain rearrangements
dont change the value of the cross ratio.
Lemma: If P = ( (( (p) )) ), Q = ( (( (q) )) ), R = ( (( (ap + bq) )) ), S = ( (( (cp + dq) )) ) then (P, Q; R, S) =
bc
ad
.
Proof: Let u = ap, v = bq. Then cp + dq =
\
|
.
|
|
c
a
u +
\
|
.
|
|
d
b
v and so S =
< >
bc
ad
u + v and the
cross ratio is
bc
ad
.

Example 5: (A, B; C, X) =
0 if X = B
1 if X = C

Theorem 3: If (P, Q; R, S) = then
(R, S; P, Q) = ;
(P, Q; S, R) = 1/ ;
(P, R; Q, S) = 1 .
Proof: Let P = (p), Q = (q), R = (r), S = (s) where r = ap + bq, s = cp + dq.
Then = bc/ad.
(P, Q; S, R) = ad/bc = 1/.
Now P = (dr bs) and Q = (cr as) and so (R, S; P, Q) =
bc
ad
= .
Finally Q = (ap + r) and S = (bs) = (bcp + bdq) = (bcp + dr adp) = ((bc ad)p + dr).
Hence (P, R; Q, S) =
bc ad
ad
= 1 .

Theorem 4: If (P, Q; R, S) = then the 24 arrangements of these points can be
partitioned into six classes of four, each class corresponding to one of the values , 1
,
1

,
1
1
,
1

and

1
as cross ratio.
Proof: (P, Q; S, R) =
1
; (P, R; Q, S) = 1 ; (P, R; S, Q) =
1
(P, R; Q, S)
=
1
1
;
(P, S; Q, R) = 1 (P, Q; S, R) = 1
1
=
1
;
59
(P, S; R, Q) = 1 (P, R; S, Q) = 1
1
1
=
1
;
(Q, P; R, S) = 1 (Q, R; P, S) = 1 (P, S; Q, R) = 1
1
=
1
.
The verification for the remaining arrangements is left as an exercise.

Example 6: If (A, B; C, D) = find (A, D; B, C).
Solution: (A, B; D, C) = 4. (A, D; B, C) = 1 4 = 3.

Example 7: Suppose A = ((3, 1, 0)), B = ((4, 5, 3)), C = ((1, 0, 1)), D = ((3, 4, 5)) are
projective points in the projective plane (Z
7
). Calculate (A, B; C, D) and (B, D; A, C).
Solution: (1, 0, 1) = 3(3, 1, 0) + 5(4, 5, 3) (this requires some work).
Let u = 3(3, 1, 0) = (2, 3, 0) and v = 5(4, 5, 3) = (6, 4, 1).
Then A = (u), B = (v), C = (u + v).
Finally (3, 4, 5) = 4(2, 3, 0) + 5(6, 4, 1) = 5(5(2, 3, 0) + (6, 4, 1)) = 5(5u + v).
So D = (5u + v) so (A, B; C, D) = 5.
Hence (B, A; C, D) = 1/5 = 3 (mod 7).
Thus (B, A; D, C) = 1/3 = 5.
Thus (B, D; A, C) = 1 5 = 3 (mod 7).
60

Exercise 1: (i) If A = (0, 1), B = (3, 7), C = (5, 11) and D = (6, 13) find (A, B; C, D).
(ii) For the above four points find (C, B; A, D).

Exercise 2: In the projective plane (Z
7
) let A = ((1, 1, 0)), B = ((5, 2, 3)), C = ((0, 2, 5) and
D = ((4, 5, 6)). Find (A, B; C, D).

Exercise 3:
Let A, B, C and D be distinct collinear points such that (A, B; C, D) = 2. Use the
Collinearity Lemma to find (C, A; B, D).

Exercise 4:
Suppose A, B, C, D, E are five distinct collinear points.
Prove that (B, C; D, E) =
(A, C; B, E)
(A, C; B, D)
.
[HINT: Use the Collinearity Lemma on A, B, C, D and on B, C, D, E.]


Exercise 1: (i) (A, B; C, D) =
AC/BC
AD/BD
=
125/ 20
180/ 45
=
5 5/2 5
6 5/3 5
=
5/2
6/3
=
5
4
.
[NOTE: We could have avoided the square roots by projecting the four points onto the x-axis,
but the relevant theory has to wait until the next chapter.]
(ii) Rather than work it out afresh we can proceed as follows by Theorem 3:
(A, B; C, D) = 5/4 so (C, D; A, B) = 5/4 and hence (C, D; B, A) = 4/5 and so
(C, B; D, A) = 1 4/5 = 1/5 and so, finally, (C, B; A, D) = 5.

Exercise 2: (A, B; C, D) = 2 so (C, D; A, B) = 2 and hence (C, D; B, A) = 1/2 and
so (C, B; D, A) = 1 (1/2) = 3/2 and so, finally, (C, B; A, D) = 2/3.

Exercise 3: Let (0, 6, 5) = x(1, 1, 0) + y(5, 2, 3). Then
x + 5y = 0
x + 2y = 2
3y = 5
.
The last equation gives 3y = 5 = 12 so y = 4 and x = 1.
Hence (0, 6, 5) = (1, 1, 0) + 4(5, 2, 3).
Let a = (1, 1, 0) and b = 4(5, 2, 3) = (6, 1, 5). Then A = (a), B = (b) and C = (a + b).
Let (4, 5, 6) = x(1, 1, 0) + y(6, 1, 5). Then
x + 6y = 4
x + y = 5
5y = 6
.
The last equation gives 5y = 6 = 13 = 20, so y = 4 and x = 1.
Thus (4, 5, 6) = (1, 1, 0) + 4(6, 1, 5) = a + 4b.
Dividing by 4, mod 7, is equivalent to multiplying by 2 since 2.4 = 1(mod 7).
So 2(4, 5, 6) = (1, 3, 5) = 2a + b and so D = (2a + b). Hence (A, B; C, D) = 2.
61
Exercise 4:
Let A = (a), B = (b), C = (a + b), D = (a + b) where = (A, B; C, D).
Then D = (xa + yb) for some scalars x, y where y 0 and so D = (a + b) where = x/y.
Then = (A, B; C, E). Since A, B, C, D. E are distinct , 1 , , 1 are all non-zero,
so the fractions in the following all exist.
We want to find x, y such that a + b = xb + y(a + b). Then y = and x + y = 1, which gives
y = and x = 1 .
So a + b = (1 )b + (a + b). So if b = (1 )b and c =(a + b) then B = (b), C = (c)
and D = (b + c).
We now want to write a + b as a linear combination of b and c.
Now b =
1
1
b and a = c b = c
1
b so a + b =
\
|
.
|
|
|
.
|
|
c
1
b +
1
1
b
=
\
|
.
|
|
1
1
b +
\
|
.
|
|
c.
Hence E =
< >
\
|
.
|
|
1
1
\
|
.
|
|
b + c and so (B, C; D, E) =
\
|
.
|
|
1
1
\
|
.
|
|
=
\
|
.
|
|
1
|
.
|
|
1
.
Now
\
|
.
|
|
1
=
\
|
.
|
|
1
1
and
\
|
.
|
|
1
=
\
|
.
|
|
1
1
.
Finally (A, B; C, D) = so (A, B; D, C) =
1
and hence (A, C; B, D) = 1

1
.
Similarly (A, B; C, E) = so (A, B; E, C) =
1
and hence (A, C; B, E) = 1

1
.
It follows that (B, C; D, E) =
(A, C; B, E)
(A, C; B, D)
.
62

63
5. PERSPECTIVITIES AND
PROJECTIVITIES

5.1. Perspectivities
Definition: A perspectivity from a line h to a line k with centre C, on neither line, is the
map f(P) = CP k. h
C
Q
P

k
f(P) f(Q)

Theorem 1: Cross ratios are preserved by perspectivities.
Proof: Let f: h k be a perspectivity with centre C. Let P, Q, R, S be four points on h
with P, Q, R distinct and S P. Let P = f(P), Q = f(Q), R = f(R) and S = f(S).

By the Collinearity Lemma, given p there exists a vector q and a scalar such that
P = p,
Q = q,
R = p + q,
S = p + q
where = (P, Q; R, S).
Also there exists c such that C = c; P = p + c .
Now C, Q, Q are collinear so Q = q + c for scalars , . Since Q C, 0 so
without loss of generality we may take = 1 and Q = q + c.
Since R = CR PQ and (p + c) +(q + c) = (p + q) + (1 + )c, we have
R = (p + c) +(q + c) .
Since S = PQ CS and (p + c) +(q + c) = (p + q) + ( + )c, we have
S = (p + c) + (q + c.
P
Q
R
S
P
Q R S k
C
64
So P = p + c,
Q = q + c,
R = (p + c) + (q + c),
S = (p + c) + (q + c).
Hence (P, Q; R, S) = = (P, Q; R, S).

Remarks:
(1) If two points and their images are known, one can determine the lines h and k and the
centre of perspectivity, viz. C = A.f(A) B.f(B).
(2) Perspectivities are 1-1 and onto mappings and hence have inverses. Moreover the inverse
of a perspectivity is a perspectivity with the same centre.

Example 1: Find the cross ratio (P, Q; R, S) of the four projective points given in example
13 using the above theorem.
Solution: Using example 14, (P, Q; R, S) = (P, Q; R, S) where P = (2, 3),
Q = (5/4, 3/2), R = (1/2, 0), S = (3/2, 2) which lie in the x-y plane. Let Y be the ideal
point on the y-axis and take the projectivity from PQRS to the x-axis with centre Y. The
images of P, Q, R, S are 2, 5/4, , 3/2 on the x-axis.
The cross ratio of these points is:
3/2
3/4

1/2
1/4
= 1.

5.2. Projectivities
The product of two perspectivities need not be a perspectivity. The simplest way to see this
is to take two lines h, k and points, O and O, that lie on neither of these lines.
If f is the perspectivity from h to k with centre O and g is the perspectivity from k
back to h with centre O then fg is a map from h to itself. If O O then fg cannot be the
identity map, yet the only perspectivity from a line to itself is the identity map.
O

h

k

O

Definition: A projectivity is a product of a finite number of perspectivities.

Theorem 2: Projectivities preserve cross ratios.
Proof: This follows immediately from the fact that perspectivities preserve cross ratios.

65
It follows that given any four points on one line and any four points on another, it is
not possible to find a projectivity taking the first four to the second four unless both sets have
the same cross ratio.
On the other hand, given any two points A, B on one line and any two points on
another, theres a perspectivity (not just a projectivity) which takes A to A and B to B. We
simply take AA BB as the centre. [If A = A or B = B there will be more than one
possible centre.]
A B

A B

If the lines are the same a perspectivity is no longer enough (unless A = A and
B = B) but all we need to do is to take A, B off to another line by one perspectivity and back
to the original line by another. Thus a 2-step projectivity is sufficient.

The projectivity which does the trick here is not unique because in fact any three
points on one line can be sent to any three points on another (or the same) line by a unique
projectivity.

5.3. The Fundamental Theorem of Projective Geometry
Theorem 3: If h, k are disjoint projective lines and P, Q, R are distinct points on h
and P , Q , R are distinct points on k then there exists a unique projectivity f : h k
such that f(P) = P , f(Q) = Q and f(R) = R . Moreover f is the product of at most 3
perspectivities (2 perspectivities if h k).
Proof: Case I: h k but one of {P, Q, R} and {P , Q , R } have one point in common.
Without loss of generality we may assume P = P or P = Q.

Case IA: P = P : Any perspectivity from h to k must fix P (ie send P to P) and as we
have seen there is a perspectivity which sends Q to Q and R to R.
B A B A
66

Case 1B: P = Q :
Choose any line m through P different to k. Let Q = QR l and R = RR m. Then
the perspectivity from h to m with centre R takes P to P, Q to Q, R to R. By
case 1A theres a perspectivity from m to k that takes Q to Q and R to R. The product
of these two perspectivities is the required projectivity.

Case 2: h k and {P, Q, R} and {P , Q , R } are disjoint.
We may assume that h k P. (If h k = P we may proceed with Q instead.) Choose
any line h through P with h k and h PP. Choose any point O on PP different from
P and P.

Q
R
P
R
P = Q
Q
R
h
k
m
P = P
Q
R
Q
R
h
k
O
P
Q
R
P
Q
R
O
R
O
Q
h
h
k
67
Let f: h h be the perspectivity with centre O. Then f(P) = P, f(Q) = Q, f(R) = R.
We now need to get from Q to Q and from R to R. But any perspectivity from h to k
must fix P and there is a perspectivity g: h k (with centre O = QQ RR) which
sends Q to Q and R to R. Thus fg: h k is a suitable (2-step) projectivity.

Case III: h = k
In this case we simply take P, Q, R off to another line by any perspectivity and proceed as in
Case II.

The uniqueness of the projectivity follows from the fact that once the images of
three points P, Q, R are given all other images are determined because of the fact that cross-
ratios must be preserved. If X is any point on h then f(X) must be that unique point X
on k such that (P, Q; R, X) = (P, Q; R, X).

WARNING: The fact that the projectivity is unique doesnt mean that its factorization into
perspectivities is unique.

Example 2: Find a projectivity f taking A to A etc where:
A = (1, 0) A = (0, 1)
B = (2, 0) B = (0, 0)
C = (3, 0) C = (0, 1).
If D = (4, 0), find f(D).

Solution: Here the points are in a Euclidean plane, regarded as being embedded in the real
projective plane. We follow Case II of the proof. Choose h to be the line y = 1. Choose
O to be the ideal point on the line AA. (We could have chosen any ordinary point on the
line, apart from A or A themselves but using an ideal point will give us good practice in
viewing the Euclidean plane as embedded in the real projective plane.)
Let f: h h be the perspectivity with centre O. Then B = f(B) = (1, 1) and
C = f(C) = (2, 1).
C B A
D
A
B
C
k
h
68

Then take as g: h k the perspectivity with centre BB CC. Since the lines are
parallel in the Euclidean Plane their intersection in the projective plane is the ideal point O
on the line y = x. Then fg: h k is the required projectivity.
Now f(D) = (3, 1) and so fg(D) = g(f(D)) = (0, 2).

5.4. Harmonic Conjugates
Definition: If A, B, C, H are collinear points then H is called the harmonic conjugate of C,
relative to A, B, if (A, B; C, H) = 1.

Construction for the Harmonic Conjugate
Theorem 4: Let A, B, C be distinct collinear points.
(1) Choose D not on AB
(2) Choose E on BD distinct from B and D themselves.
(3) Let F = AE CD.
(4) Let G = AD BF.
(5) Let H = AB EG.
Then H is the harmonic conjugate of C relative to A, B.
Proof: This is left as an exercise. [Apply the Collinearity Lemma to the lines ABC and BED
and then find vectors to represent all the other points. The harmonic conjugate, H, should be
a b = a + b.]

C B A
D
A
B
C
B C
k
h
h
C B A
D
A
B
C
B C
k
h
h
69
G
D
F
E

A H B C

Example 3: There is a straight road which ends at the base of Uluru (Ayers Rock) in Central
Australia. It is desired to continue the road on the other side of the rock so that both portions
are in the same straight line. How can this be done using only standard surveying equipment
which can effectively draw a straight line between two points (provided the rock is not in the
way)?

existing road future road

Solution: Take points A, B, C on the existing road and carry out the Harmonic Conjugate
construction to the point where the line EG is drawn. Repeat the construction with different
choices of D, E named D and E to obtain the line EG. Then EG EG will intersect at the
Harmonic Conjugate of A, B relative to C and this point will lie on the line AB of the road.
If the choices are suitably made this point will lie on the far side of the rock and
none of the construction lines will pass through the rock and so will be possible. Repeat all
of this again, using different points A, B, C on the existing road and a second point on the
line of the road can be obtained (again on the far side if suitable choices are made). Well
now have two points on the line AB on the opposite side of the rock to the existing road and
these can now be joined.
In order to get H to lie on the far side of the rock wed need to take C between A and
B. Choosing it to be the midpoint would result in H being the ideal point on AB, which
would not suit a surveyor. On the other hand choosing C too close to B might result in H
being too close to B and lie underneath the rock. So a choice of C just to the right of the
midpoint seems to be suitable,
D

G
E
F

rock
A C B H H for A, B, C

F
G E
D
70

Exercise 1: Suppose m is the line y = 2x + 1 in the real affine plane, and n is the line
y = 6 x. Let A be the point (1, 3) on m and let P be the point (2, 1). Let f:mn be the
perspectivity from m to n with centre P. Find f(A).

Exercise 2: Suppose m is the line y = 2x + 1 in the affine plane over Z
5
, and let n be the
line y = 2. Let P be the point (1, 1). Let f:mn be the perspectivity from m to n with
centre P.
(i) If A is the point (2, 0), find f(A);
(ii) If B is the point (4, 1), find f(B).

Exercise 3: Suppose O = (1, 1, 1), A = (2, 1, 0), B = (3, 4, 3), P = (2, 1, 3) and
X = (3, 2, 1) in the real projective plane. Let f be the projectivity from OA to OB with
centre P. Find f(X).

Exercise 4: Let m be the line y = x + 1, let n be the line x = 5 and let r be the line
y = 2 3x in the real affine plane. Let O = (1, 0), Q = (0, 1) and X = (3, 2). Let f be the
perspectivity from m to n with centre O and g is the perspectivity from n to r with
centre Q and h = fg. Find h(X).

Exercise 5: In the real projective plane, find the centre of the perspectivity f that takes
P(0, 1) to P(0, 2) and Q(0, 2) to Q(1, 0). Does f take R(0, 3) to R(3, 0)?
Exercise 6: In the real projective plane, find a projectivity f that takes P(0, 1) to P(0, 2),
Q(0, 2) to Q(1, 0) and R(0, 3) to R(3, 0). Is there such a projectivity that also takes S(0, 4)
to S(5/2, 0)? Illustrate your solution with a diagram.

Exercise 7:
(i) If (A, B; C, D) = , find (B, C; D, A) in terms of .
(ii) If A, B, C, D are distinct collinear points, for what value or values of (if any) does there
exist a projectivity f from the real line to itself such that:
f(A) = B, f(B) = C, f(C) =D, f(D) = A?
(iii) Let A, B, C, D be the points 0, 1, 4, and 2 respectively on the real line. Use the
Euclidean interpretation of cross ratio to calculate (A, B; C, D).
(iv) Construct a projectivity, from the real line to itself, such that:
A B, B C, C D, D A.
(Describe the construction in words, as a sequence of perspectivities, and illustrate your
construction with a diagram.)

Exercise 8:
(a) Let A = (2, 1), B = (2, 3), C = (0, 2) and D = (6, 1) be points in the Euclidean plane.
Find (A, B; C, D).
(b) Let h denote the line y = 2 x on which the points in (a) lie. Exhibit a projectivity
from h to h under which A C, B A, C B.
(c) Determine the image of D under the projectivity in part (b).
71
Exercise 9: In the Euclidean plane extended by ideal points and the ideal line let
A = (0, 0), B = (1, 0) and let C be the ideal point on the x-axis. Carry out a construction to
find the harmonic conjugate of C relative to A and B.

Exercise 1: The line joining P to A is y = 2x + 5. It cuts y = 6 x when x = 1. So f(A)
is (1, 7).

Exercise 2:
(i) The line joining P to A is y = 4x + 2. It cuts y = 2 when x = 0. So f(A) is (0, 2).
(ii) The line joining P to B is y = 1. It cuts y = 2 at the ideal point on that line. Hence f(A)
is the ideal point on the lines y = c, for constants c.

Exercise 3: PX = (2, 1, 3) (3, 2, 1)
= (5, 7, 1).
OB = (1, 1, 1) (3, 4, 3)
= (6, 6, 12).
Hence f(X) = (5, 7, 1) (6, 6, 12)
= (90, 54, 72) = (5, 3, 4).

Exercise 4: The line OX is y = x + 1. This cuts x = 5 at (5, 4), so f(X) = (5, 4).
The line Qf(X) is y = x + 1. This cuts y = 2 3x at ( , ), so h(X) = g(f(X)) = ( , ).

Exercise 5: The centre is PP QQ = (1/3, 5/6). The line joining this point to R has
equation y = (13/2)x + 3. It cuts PQ (the x-axis) at (6/13, 0). This is f(R) but not R.
Exercise 6: Let h be the y-axis and let k be the x-axis. Take h (any line through P not equal
to k) to be the line x = 2. The line PP is y = 1 x so choose O (any point on PP except
those two points themselves) to be (1, ).
The line joining O to (0, b) on h is y = ( b)x + b and it cuts h when x = 2 and y
= 1 b. So the images of P, Q, R and S under the perspectivity from h to h with centre O
are: P, Q(2, 1), R(2, 2), S(2, 3).
We now project from h to k. The centre is O = QQ RR. This is the
intersection of the lines y = x + 1 and y = 2x 6. So O is (7/3, 4/3).
Projecting R through O onto k we get (13/5, 0). This is the image of R under
the projectivity, and so is not R. Could there be another projectivity that works (after all we
made some choices in constructing the one we did)?
(P, Q; R, S) =
PR/QR
PS/QS
=
4
3
, while (P, Q; R, S) =
PR/QR
PS/QS
=
3
2
. Since
projectivities preserve cross ratios there can be no projectivity that takes each of P, Q, R, S to
P, Q, R and S respectively.

Exercise 7:
(i) (B, A; C, D) =1/. (B, C; A, D) = 1 1/ = ( 1)/.
(B, C; D, A) = /( 1)..
(ii) We must have
1
= so = 2 (NB 0 since the points are distinct.)
72
(iii) (0, 1; 4, 2) =
4/3
2/3
= 2.
(iv) By (i) and (ii) (A, B; C, D) = (B, C; D, A) and so there exists such a projectivity.

(1) Take the perspectivity from the x-axis to the line y = 1 with centre the ideal point on
vertical lines.
This takes A, B, C, D to A = (0, 1), B = (1, 1), C = (4, 1), D = (2, 1) respectively.

(2) Take the perspectivity from the line y = 1 to the line x = 1 with centre the ideal point on
lines with slope 1. This takes A, B, C, D to A = (1, 0), B = (1, 1), C = (1, 4),
D = (1, 2) respectively.

(3) The line BC is y = (1/3)x + 4/3. The line CD is y = (4/3)x + 8/3. They intersect at
(4/5, 8/5). So we take the perspectivity from the line x = 1 to the x-axis again, with centre
(4/5, 8/5).
This is the required projectivity. The line AD is y = (1/2)x. Since this passes through
(4/5, 8/5) this verifies that f(D) is indeed A.

C

D A B = B C

D A B = A C

D

Exercise 8:
(a) To avoid awkward square roots with distances we can project the points onto the x-axis.
This amounts to taking a perspectivity whose centre is the ideal point on the y-axis. In other
words, we can simply work with the x-coordinates.
(A, B; C, D) =
AC/BC
AD/BD
= 2.
(b) Using the perspectivity described above we take the four points off onto another line (the
x-axis) to A(2, 0), B(2, 0), C(0, 0) and D(6, 0). We must now take A C, B A and
C B.
73
We take y = 2 as our line through C and we choose O to be (1, 1). Taking the
perspectivity from the x-axis to y = 2 with centre O, the point (t, 0) maps to (2 t, 2). So A
maps to C, B maps to B(4, 2), C maps to C(2, 2).
Now O = BA CB = (10/3, 5/3). So we take the perspectivity from the line
y = 2 to the x-axis with centre (10/3, 5/3). The product of these three perspectivities is the
required projectivity. To summarise, these perspectivities are:
From line to line with centre
y = 2 x x-axis ideal point on y-axis
x-axis y = 2 (1, 1)
y = 2 y = 2 x (10/3, 5/3)
(c) Under this projectivity D (6, 0) (4, 2) (2/5, 9/5).

Exercise 9:

F D

G

A H B C

Choose D = (0, 1) (not on AB). Choose E = ideal point on BD. E
F = AE CD = (1, 1), G = AD BF = (0, ), H = AB EG = (, 0).
74

75
6. ISOMETRIES

6.1. Isometries
Fundamental to the theory of symmetry are the concepts of distance and angle.
So we work within R
n
, considered as an inner-product space. This is the usual n-
dimensional real vector space, together with the inner product defined by the dot product
u.v. We use this dot product to define distances and angles.
Recall that the length of a vector v is given by |v| = v.v , that the distance
between two points u, v is |u v| and that the cosine of the angle between two
vectors u, v is given by cos =
u.v
|u|.|v|
.
These correspond to the usual concept of distances and angles in 3-dimensional
space and can be taken to be the definitions of lengths and angles in higher dimensional
spaces (or in any inner product space for that matter).

Definition: An isometry on an inner-product space V is a distance-preserving function
from V to V, that is, it is a function f : V V such that |f(u v)| = |u v| for all
v V. Examples include reflections, rotations and translations.
Now a fundamental tool in the study of isometries is linear algebra. However
isometries need not be linear transformations since they neednt fix the origin for
example translations dont. But those isometries that fix the origin do turn out to be
linear, as we shall prove.

Definition: A central isometry is an isometry that fixes the origin.

A reflection is a central isometry if the axis of reflection passes through the origin
and a rotation is a central isometry if the centre of the rotation is the origin. A translation
can never be central because it moves every point. However translations provide the link
between a general isometry and a central one.

Definition: A translation, in an inner product space V, is a function of the form
f(v) = v + a for some fixed a V. (If a = 0 the isometry is simply the identity map.)

Theorem 1: Every isometry is a central isometry or a central isometry followed by a
translation.
Proof: Suppose f is an isometry. Define c(v) = f(v) f(0). This is an isometry followed
by a translation and since the product of two isometries is an isometry, c is also an
isometry. But c(0) = 0, so c is a central isometry. Thus f(v) = c(v) + f(0) is a central
isometry followed by a translation.
76
6.2. Central Isometries
Theorem 2: Central isometries are linear transformations.
Proof: Suppose f is a central isometry. Isometries map straight lines to straight lines
and so they map quadrilaterals to quadrilaterals. Since a parallelogram is a quadrilateral
with opposite sides equal in length, isometries map parallelograms to parallelograms. A
central isometry thus maps the parallelogram with vertices 0, u, v and u + v to the
parallelogram with vertices 0, f(u), f(v) and f(u) + f(v).
Thus f(u + v) = f(u) + f(v) for all u, v V.

Suppose 0 and v 0. Clearly isometries take straight lines to straight lines and so,
since 0, v, v are collinear, so are f(0) = 0, f(v) and f(v). Hence f(v) must be a scalar
multiple of f(v). Suppose f(v) = f(v).
Clearly, if v = 0 or = 0 then f(v) = f(v). So suppose that 0 and v 0.
Now |f(v)| = ||.|f(v)| = ||.|v|. But |f(v)| = |v| = ||.|v|. Hence = and so f(v) =
f(v). But if f(v) = f(v) then f((1 + )v) = f(v) + f(v) = f(v) f(v)
= (1 )f(v), and, taking lengths, (1 + )|v| = (1 )|f(v)| = (1 )|v|, a contradiction.
Thus f(v) = f(v) for all and v. Hence f is a linear transformation.

Recall that an orthogonal matrix is a square real matrix A such that A
T
A = I.
Recall too that a set of vectors is an orthonormal set if they are mutually orthogonal unit
vectors. By considering the product A
T
A and comparing this with the identity matrix
we see that the columns of an orthogonal matrix form an orthonormal set. It also follows
from the equation A
T
A = I that an orthogonal matrix A is invertible and that A
T
= A
1
.
Now matrices dont commute in general, but every matrix commutes with its
inverse, so we must also have AA
T
= I. Writing this as (A
T
)
T
A
T
we can see that if A
is orthogonal then so is A
T
. But the columns of A
T
are the rows of A so the rows of an
orthogonal matrix also form an orthonormal set.
Taking determinants of both sides of the equation A
T
A = I, and using the fact that
|A
T
| = |A|, we can see that the determinant of an orthogonal matrix A must satisfy
|A|
2
= 1 and so must be 1.

Example 1:
2
3

1
3

2
3
2
3

2
3

1
3
1
3

2
3

2
3
is an orthogonal matrix. The columns
0
u
v
u+ v
f(v)
f(u + v)
f(u)
77

2
3
2
3
1
3
,
1
3
2
3
2
3
,
2
3
1
3
2
3
form an orthonormal set.

Theorem 3: The linear transformation f(v) = Av, on R
n
is an isometry if and only if
A is orthogonal.
Proof: Suppose that f is an isometry. The standard basis vectors e
1
= (1, 0, , 0),
e
2
= (0, 1, , 0), e
1
= (0, 0, , 1) are an orthonormal set. The columns of A are
their images under f and so also form an orthonormal set. Thus A is orthogonal.
Conversely suppose that A is orthogonal and let u, v R
n
. Let w = u v.
Then |Aw|
2
= (Aw)
T
(Aw) = w
T
A
T
Aw = w
T
w = |w|
2
so |Aw| = |w|.
Now |Au Av| = |Aw| = |w| = |u v| so f is an isometry.

Since the determinant of the matrix of a central isometry is 1 we can divide
central isometries into two types.

Definition: A central isometry is direct if it has determinant 1.
It is opposite if it has determinant 1.

We can extend these ideas to isometries in general.

Definition: An isometry is direct if it is a direct central isometry followed by a
translation and opposite if it is an opposite central isometry followed by a translation.

The identity isometry is direct, and so are translations. To determine whether
rotations and reflections in 2 and 3 dimensions are direct or opposite we need to find their
matrices. But first let us consider isometries in 1 dimension.

We consider the very simple case of isometries of a line. The only 11 real
matrices with determinant 1 are the matrices (1) and (1) and so the only central
isometries of R are the identity function I(x) = x and the reflection in the origin
M(x) = x.
Translation through a distance a is the isometry T
a
(x) = x + a. The only other
possible isometries are MT
a
(for various values of a). This is a reflection in the origin
followed by a translation through a. If we let M
a
= MT
a
we get M
a
(x) = T
a
(M(x))
= T
a
(x) = a x. This is a reflection in the point a/2. Clearly these are all opposite
isometries.
So the isometries of a line can be summarized as follows:
78

LINEAR ISOMETRIES
ISOMETRY Equation DIRECT? Fixed Points
Identity I(x) = x

all
Translation by a T
a
(x) = x + a

none
Reflection in a/2 M
a
(x) = a x
a/2

6.3. Central Isometries of the Plane
Definition:

denotes the rotation about the origin through the angle .

denotes reflection in the line, through the origin, that is inclined at the
angle to the positive x-axis.

These are linear transformations of R
2
and so will correspond to a certain matrix.
We shall find these matrices relative to the standard basis.

Theorem 4: The matrix for

is R

=
cos sin
sin cos
.
Proof: Let P be the point (x, y) where x = rcos and y = rsin.

Rotating it about the origin through the angle takes it to (X, Y) where
X = rcos( +) and Y = rsin( +). We can write these equations as:
X = rcos.cos rsin.sin = xcos ysin and
Y = rsin.cos + rcos.sin = xcos + ysin.
Writing this in matrix form we get:
X
Y
=
cos sin
sin cos
x
y

Theorem 5: The matrix for
/2
is M

=
cos sin
sin cos
.
Proof: Let x = r cos and y = r sin.

The angles represented by are each equal to /2 .
P = (X, Y)
P = (x, y)

X
O

P = (X, Y)
P = (x, y)

X
O
/2
79
Hence POX = /2 + (/2 ) = and so:
X = r cos( ) and
Y = r sin( ).
Expanding we get:
X = r cos( ) = cos.r cos + sin.r sin = x cos + y sin and
Y = r sin( ) = sin.r cos cos.r sin = x sin y cos.
In matrix form this becomes:
X
Y
=
cos sin
sin cos
x
y
.

Since
cos sin
sin cos
= cos
2
+ sin
2
= 1, rotations are direct isometries and since
cos sin
sin cos
= 1, reflections are opposite isometries.

Theorem 6: Every central isometry of the plane is either a rotation about the origin
or a reflection in a line through the origin.
Proof: Let f be a central isometry of R
2
. Write the corresponding orthogonal matrix as:
A =
a b
c d
.
Because of the orthogonality we have:
a
2
+ c
2
= 1,
b
2
+ d
2
= 1, and
ab + cd = 0.
From the first equation we can write a = cos, c = sin for some .
From the second we have that b = cos, d = sin for some .
From the third equation we have cos.cos + sin.sin = 0. Thus cos( ) = 0.
So = /2 + , in which case b = sin, d = cos,
or = 3/2 + , in which case b = sin, d = cos.
Thus A =
cos sin
sin cos
or
cos sin
sin cos
.
The eigenvalues of the first matrix are e
i
and of the second are 1.

So the central isometries are:

ISOMETRY Matrix Eigenvalues
Identity I(v) = v 1, 1
Rotation through
about the origin
R

=
cos sin
sin cos

e
i
, e
i

Reflection in the line
y = tan(/2).x
M
cos sin
sin cos

1, 1

80
6.4. Products of Central Isometries
We can identify the product of some central isometries by multiplying the
relevant matrices. But we must remember that the product of matrices must be in reverse
order since the one thats applied first is the right hand factor.

Example 2: Identify the product, f, of a reflection in the line y = x, a rotation through
60 and a reflection in the y-axis.
Solution:
The matrices of each of these central isometries are (angles are in degrees):
cos90 sin90
sin90 cos90
,
cos60 sin60
sin60 cos60
,
cos180 sin180
sin180 cos180
.
So we calculate M
90
R
90
M
180
=
cos180 sin180
sin180 cos180

cos60 sin60
sin60 cos60

cos90 sin90
sin90 cos90

=
1 0
0 1

1
2

3
2

3
2

1
2

0 1
1 0

=
1
2

3
2

3
2

1
2

0 1
1 0

=
3
2

1
2

1
2

3
2

=
cos30 sin30
sin30 cos30

So the product is a central rotation through 30.

Theorem 7:
R

R

= R
+
;
R

M

= M
+
;
M

R

= M

;
M

M

= M

.
Proof: Just multiply the relevant matrices.

These results can be used to simplify the calculation of the product of central isometries.

Example 3: Identify the product, f, of a reflection in the line y = x, a rotation through
60 and a reflection in the y-axis.
Solution: The matrices of these isometries are M
90
, R
60
, M
180
. We must multiply these
in reverse order since the rightmost matrix is the one that corresponds to the first
isometry to be applied.
M
180
R
60
M
90
= M
120
M
90
= R
30
.
81
The effect of the above identities is the application of the following rules.

(1) In a product of isometry matrices Rs and Ms the product is a:
rotation if theres an even number of reflections in the product and a
reflection if theres an odd number of reflections.

(2) If the successive subscripts are
1
,
2
, ,
n
the subscript for the product is:
is
1

2

n
where each sign is:
+ if there is an even number of Ms preceding the factor and
if there is an odd number of Ms preceding the factor.

Example 4: M
= M
++

Example 5: R
= R
+++

Example 6: M
120
M
45
M
90
R
60
M
180
= R
12045+9060180
= R
75
= R
285
, a rotation about the
origin through 285.

6.5. Plane Isometries
The central isometries of the plane are the identity I, the rotations
and the
reflections in a line through the origin
. To complete the picture we need to consider

these followed by translations.

Definition: A translation is an isometry of the form
a
where
a
(v) = v + a. This is
said to be the translation by the vector a.

To complete the picture we need to follow rotations and reflections by
translations. Perhaps a rotation followed by a translation is a rotation about some point
other than the origin. If so, it will fix some point, and that will be the centre of the
rotation. Lets see if we can solve the equation
a
(v) = v or, in other words,
R
v + a= v.
This gives (I R
)v = a. If I R
is invertible we can then write

v = (I R
)
1
a.
82
Now, if R
is a proper rotation, that is not through 0, its eigenvalues are non-real.

Hence 1 is not an eigenvalue and so |I R
| 0. Thus it is indeed the case that I R
is
invertible.
Weve established that
a
fixes c = (I R
)
1
a but we have yet to establish that
its a rotation about the point c.
Rotating the point v about c = (I R
)
1
a through the angle produces the point
R
(v c) + c. (We first make c the temporary new origin, by subtracting c, then rotate,
and finally add c to revert to the original origin.)

Now
a
(v) = R
v + a
= R
(v c) + R
c + a
= R
(v c) + R
(I R
)
1
a + a
= R
(v c) + (R
I)(I R
)
1
a + (I R
)
1
a + a
= R
(v c) a + c + a
= R
(v c) + c
Hence the product of this rotation and translation is a rotation through about a new
centre c = (I R
)
1
a .

Finally we must consider a central reflection followed by a translation. It would
be natural to guess that this will be a reflection in some other line. However theres a
surprise!
If the product is a reflection it will fix all the points on the axis of the reflection.
So lets consider the equation
/2
a
(v) = M
v + a = v or, equivalently, (I M
)v = a.
This would give v = (I M
)
1
a if I M
was invertible. But alarm bells should be

ringing, for if that was the case then v would be unique whereas were expecting a
whole line of fixed points.
The trouble is that I M
is invertible. Remember that the eigenvalues of a

reflection matrix are 1. This makes sense because a central reflection fixes two lines
the axis of reflection, which is the eigenspace corresponding to the eigenvalue 1 and the
perpendicular to that axis, which corresponds to the eigenvalue 1. Since 1 is an
eigenvalue, |I M
| = 0 and so I M
is not invertible.
c
v
v c
R
(v c) + c
R
(v c)
83
Lets first consider the special case where the direction of translation is
perpendicular to the axis of reflection. In this case each point v will stay on the line
through v perpendicular to the axis of reflection. The situation will become equivalent
to the 1-dimensional case, which turned out to be a reflection.
Let p be the foot of the perpendicular to the axis of the reflection. Then
following the reflection v will be sent to 2p v. (We can see that this is correct
because the midpoint of v and 2p v is p, as it should be.) Translating by a will take
the point to 2p v + a. The midpoint of v and 2p v + a is 2p + a. Since this is
independent of v each point will be mapped to its mirror image in the line through
2p + a.

Now lets consider the special case where the direction of translation is parallel to
the axis of reflection.

Its clear that in this case, assuming of course that a 0, the product fixes no
points. Those that are not on the axis go to the other side, and those that are on the axis
are fixed by the reflection and are then moved by the translation. So if there are no fixed
points this composite cant be a reflection. Nor can it be a rotation or the identity. It
must be a totally new type of isometry.

Definition: A glide is a reflection in a line followed by a translation parallel to that line.

v
p
2p v
2p v + a

v
p
2p v
2p v + a
axis of the product

axis of the central reflection

84
It remains to consider a central reflection followed by a general translation, not
necessarily in the direction perpendicular or parallel to the axis of reflection. In this case
we simply resolve the vector that represents the translation in the parallel and
perpendicular directions. That is, we write a = b + c where b is perpendicular to the
axis and c is parallel to it. Then
/2
a
(v) = M
(v) + a = (M
v + b) + c. By the
perpendicular case, M
v + b is a reflection in an axis at an angle /2 to the positive x-

axis. Then since c is parallel to this axis the composite is a glide in that axis.

PLANE ISOMETRIES
DIRECT? FIXED POINTS FIXED LINES
IDENTITY
All all
ROTATION
none
TRANSLATION
None none

REFLECTION

all points on a line

a line and all lines to it
GLIDE

None
a line

We denote the isometries of the plane as follows:
I = identity isometry;
R
a,
= rotation about a through the angle ;
T
a
= translation by the vector a;
M
a,b
= reflection in the line joining a to b;
G
a,b
= glide in the line joining a to b, that takes a to b.

Theorem 8: Let p be the foot of the perpendicular from 0 to the line joining a to b.
Then (i) T
a
(v) = v + a.
(ii) R
a,
(v) = R

(v a) + a.
(iii) M
a,b
(v) = M
b a
(v) + 2p
(iv) G
a,b
(v) = M
b a
(v) + 2p + b a.
Proof: (i) follows from the parallelogram law for vector addition:
(ii) can be proved by first translating a to the origin, then performing the corresponding
central isometry, and then translating back.
(iii) If v is the reflection of v in the line joining a to b and v is the reflection of v
in the line joining 0 to b a then v = v + 2p. Part (iv) is now obvious.

v
b
v
t p
v
t
s t
b a
s
s t
0
a
85
Theorem 9: Every plane isometry is a product of at most 3 reflections.
Proof:
Identity: If M is any reflection, I = M
2
.
Rotation: The product of two reflections whose axes intersect at P is a rotation about P
through twice the angle between them.

.
.

Reflecting P in OA gives P. Reflecting this in OB gives P. The combined effect
of these two reflections is to rotate P through 2AOB.

Translation: The product of two reflections in parallel axes is a translation in the
direction perpendicular to the axes, through twice the distance between them.

Reflection: A reflection is a product of one reflection.

Glide: A glide is a reflection followed by a translation, so from the above it is a product
of three reflections.

6.6. Identifying Products of Plane Isometries
Suppose were given a sequence of plane isometries, to be performed
sequentially, and we wish to identify the product. One way would be to perform the
necessary matrix and vector calculations. But then we would want to convert the
resulting formula to a geometric description. Now that we know all the possible
isometries of the plane we can avoid a lot of the algebra and proceed geometrically.
We know how to multiply products of central isometries easily, but what happens
if we have some translations in the product?

Theorem 10: If f is a central isometry then
a
f = f
b
for some b.
Proof: Suppose F is the matrix of the central isometry. Then
a
f(v) = F(v + a)
= Fv + Fa. Then
a
f = f
b
for b = Fa.

This means that if we have a product of central isometries and translations the
product will have the form f where f is the product of the central isometries, ignoring
the translation, and will be some translation.
Since we know how to easily multiply central isometries well easily be able to
determine the factor f in the answer. All that remains is to determine the .
P
P
P
O
A
B
P
P P
x x y y
86
Example 7: Identify the product of a reflection in the line y = x, a glide in the y-axis
through a distance 1 vertically and then the rotation through 90 about (1, 1).
Solution: The reflection is
45
, with matrix M
90
, the glide is
90
(0,1)
and the rotation will
be
90
for some translation . The product is
45
90
(0,1)
90
=
45
90
90
for some
translation . The matrix for
45
90
90
will be R
90
M
180
M
90
= R
90+18090
= R
180
. So the
product will be
180
and so will be a 180 rotation about some point.
To find the centre of rotation we choose a point. The midpoint of this point and
its image will clearly be the centre of the rotation. Take P = (0, 0). This maps to (0, 1).
Hence the centre will be (0, ). So the product is the rotation through 180 about (0, ).

Example 8: Identify the product of a rotation through 30 about the origin, O, followed
by a rotation through 60 about A = (1, 0).
Solution: This is clearly a 90 rotation. The only question is: what is the centre?
The origin O rotates to the point B where AO = AB and angle OAB is 60. The
centre must be the point C such that CO = CB and OCB = 90.

O A
C

B

Since OAB is an isosceles triangle with a 60 angle its equilateral, and so OB = 1.
OCB is a right-angled triangle and so OC = 1/2. Angle AOC = 60 45 = 15.
So C is the point (x, y) where x =
1
2
cos15 and y =
1
2
sin15 .

Example 9: Identify the product, f, of a reflection in the line y = x, the rotation through
90 about (1, 1) and a glide in the y-axis through a distance 1 vertically.
Solution: Note that these factors are the same as in example 7, but the order is different.
The product can be written as
45
90
90
for some translations , . The matrix for
45
90
90
is M
180
R
90
M
90
(remember to reverse the order when using the matrices and to
double the angle for the Ms). The product is clearly R
180 90 90
= R
0
and so is the
identity and so the final answer is a translation (or possibly the identity).
Take P = (1, 1). Then f(P) = (1, 2). Since (1, 2) (1, 1) = (2, 1) the product is a
translation by the vector (2, 1).

87
by a rotation through 270 about A = (1, 0).
Solution: This is clearly
360
= for some translation .

The origin is mapped to (1, 1) so the product is the translation by the vector (1, 1).

by the translation f(v) = v + (1, 1).
Solution: The product is clearly a rotation through 90. The only question is: what is the
centre? The origin maps to (1, 1) so the centre C is such that angle OCA is 90.

Clearly C must be (0, 1). (C = (1, 0) would also make OCA a right angle, but in that case
it would need a rotation through 270 to rotate O to A.) So the product is a 90 rotation
about (0, 1).

Example 12: Identify the product of the reflection in the line y = x + 3 followed by the
reflection in the line x + y = 1.
Solution: The product is
45
45
for some translations , a direct isometry. The
matrix of
45
45
is M
90
M
90
= M
180
= M
180
. So it is a 180 rotation. Clearly it must fix
the intersection of the two lines, namely (1, 2). So the product is the rotation through
180 about (1, 2).

Example 13: Identify the product of the reflection in the line x + y = 1 followed by the
rotation through 45 about the point (1, 0).
Solution: R
45
M
90
= M
45
so the product must be in a line inclined to the positive x-axis
by the angle 22.5. Since the point (1, 0) lies on the axis of reflection it is fixed by the
product. Therefore the product must be a reflection in the line through (1, 0) at an angle
of 22.5.
88

Example 14: Identify the product of the reflection in the line y = x followed by the
rotation through 90 about the point (1, 0).
Solution: R
90
M
90
= M
180
so the product is either a reflection or a glide in a vertical axis.
The origin moves to the point (1, 1) so the product must be the glide which consists of a
reflection in the line x = followed by a translation vertically downwards through a
distance of 1 unit.

Exercise 1:
Consider the following isometries of the plane.
T is the translation to the right through a distance of 1 unit;
R is the 90 (anti-clockwise) rotation about (1, 1);
M is the reflection in the line x + y = 1.
Identify each of the following products as a translation, a rotation, a reflection or a glide.
If a product is a translation or a glide state the direction and the distance. If it is a
rotation state the centre and the angle. If it is a reflection state the axis.

(i) TR;

(ii) MT;
89
Exercise 2: The point P(1, 1) is rotated through 30 about the point (2, 3) and then
translated in the direction of (1, 2) through a distance of 3 units. Find the coordinates of
the resulting point.

Exercise 3: ABCD is a unit square and a point P is successively rotated through 90
about each of the four points, in the given order. Show after the four rotations, the net
effect will be to translate P in the direction AD through a distance 4.

Exercise 3: ABCD is a unit square (with the corners being given in anticlockwise order).
A point P is rotated through 90 successively about the four vertices in the given order.
Show that the net effect is to translate the point in the direction AD through a distance 2.

Exercise 4: Let O = (0, 0), A = (0, 1) and let B be such a point in the upper half plane
that makes OAB an equilateral triangle. A point P is rotated through 30 about O, then
through 30 about A and finally through 30 about B. It is clear that the result is a 90
rotation. Find the centre of this rotation.

Exercise 1:
(a) (i) TR is a direct isometry which rotates all lines by 90 so it must be a 90 rotation
about some point C.
Under TR: (0, 0) (1, 0) (2, 1) so C lies on the perpendicular bisector of the interval
from (0, 0) to (2, 1).
The midpoint is (1, ) and the slope of the perpendicular is 2. Thus C lies on the line y
= 2x + 5/2.
Similarly (1, 1) (2, 1) (1, 2). The midpoint is (1, 3/2) and the perpendicular
bisector is horizontal. So its equation is y = 3/2.
Solving these two equations we get x = and y = 3/2.
So TR is a 90 rotation about (1/2, 3/2).
(ii) MT is an opposite isometry.
Under (MT)
2
: (0, 0) (1, 1) (1, 1) (0, 0) (2, 0) so (MT)
2
I. Hence MT is not
a reflection. Therefore it must therefore be a glide. (MT)
2
must therefore be a
translation. It must, in fact, be a translation through 2 units to the right. So MT must be a
glide in a horizontal line through a distance of 1 unit.
So MR is a glide along the axis y = through a distance of 1 unit to the right.

Exercise 2:
Translate (2, 3) to the origin. So (1, 1) (1, 1). The rotation matrix for a 30 rotation
about the origin is
3
2

1
2
1
2
3
2
so (1, 1)
3
2

1
2
1
2
3
2

1
1
=

1 + 3
2
3 1
2
.
90
Now translate the origin back to (2, 3). The point now moves to
3 3
2
,
3 + 5
2
.
Now translate by
3
2
(1, 2) to get
3 3
2
+
3 5
5
,
3 + 5
2
+
6 5
5

Exercise 3: Let the four points be given by the vectors 0, a, a + b and b, with a, b being
an orthonormal basis of R
2
. Let R = R
90
=
0 1
0 1
be the matrix of a 90 rotation about
the origin. Then Ra = b and Rb = a. If P is represented by v then the successive
positions are:
v Rv
R(Rv a) + a = R
2
v Ra + a = v b + a
R[v b + a (a + b)] + a + b = Rv 2Rb + a + b = Rv + 3a + b
R[Rv + 3a + b b] + b = R
2
v + 3Ra + b = v + 4b. So the product of the four
rotations is a translation in the direction AD through a distance of 4.

Its clear that the product of four 90 rotations about any centres will result in directed
lines being fixed and so will be a translation. If we take the A and carry out the rotations
we can see geometrically that the direction and distance of the translation are as claimed.]

Exercise 4: Let A be represented by the vector a and let B be represented by b.
Let c = cos30 = 3/2 and s = sin30 = and let R = R
30
=
c s
s c
be the matrix of a 30
rotation about the origin. Then b = R
2
a.
If P is represented by the vector v then:
v Rv
R(Rv a) + a = R
2
v Ra + a
R(R
2
v Ra + a R
2
a) + R
2
a = R
3
v + Ra R
3
a.
Suppose the centre is represented by the vector z.
Then v R
3
(v z) + z = R
3
v R
3
z + z.
Hence (I R
3
)z = Ra R
3
a
Hence z = R(1 R)(1 + R + R
2
)
1
a
Now 1 + R + R
2
=
1 0
0 1
+
c s
s c
+
s c
c s
=
1+c+s (s+c)
(s+c) 1+c+s
=
1+t t
t 1+t

where t = c + s.
(I + R + R
2
)
1
=
1
1+2t+2t
2
1+t t
t 1+t
.
(I + R + R
2
)
1
a =
1
1+2t+2t
2
1+t t
t 1+t

1
0
=
1
1+2t+2t
2
1+t
t
.
R + R
2
=
t t
t t
so z =
1
1+2t+2t
2
t t
t t

1+t
t
=
1
1+2t+2t
2
t+2t
2
t

So the centre is
t+2t
2
1+2t+2t
2,
t
1+2t+2t
2 .
91
Now t = c + s =
1 + 3
2
and t
2
=
2 + 3
2
.
1+2t+2t
2
= 4 + 2 3 , t + 2t
2
=
5+3 3
2
.
So the centre of the equivalent 90 rotation is
5+3 3
8+4 3
,
1+ 3
8+4 3
=
1+ 3
28
,
1+ 3
28
.
92

93
7. SYMMETRY GROUPS

7.1. What is Symmetry?
When we say that something is symmetric were usually thinking of left-right
symmetry. For example, a design is symmetric in this way if the right half is the mirror
image of the left. The axis of symmetry separates the two halves and, if we place a
mirror along this line, the design seems complete. The reflection of
the left half makes up for the hidden right half. The human face is
generally considered to have this mirror symmetry even though we
can detect subtle differences between the two sides.

But this is only one type of geometric symmetry. A snowflake has
more than just mirror symmetry. You can rotate it through a 60
angle about its centre and it fits exactly onto its original form. This is
called rotational symmetry. The number of times this needs to be
performed in order for all the points to actually return to their original
positions is called the order or degree of the rotation. A 60 rotation
would need to be performed 6 times to produce a full 360 rotation and so has degree 6.
We say that a snowflake has 6-fold symmetry. Of course a
120 rotation rotates a snowflake to exactly the same position,
even though individual points are occupying different
positions. So why dont we say that it has 3-fold symmetry as
well? The reason is that we quote the highest degree of
symmetry which, for a snowflake, is 6. A square has 4-fold
symmetry since its fixed by a 90 rotation but no smaller
positive angle.
The Isle of Man has a motto to the effect that
whichever way I fall, I stand and an appropriate insignia with 3-
fold symmetry, but no mirror symmetry.
A row of trees gives the appearance of translational
symmetry. If all the trees are absolutely identical, and if the row
extends forever in either direction, then moving each tree to the
right a certain distance has no effect on the overall pattern.

Reflections, rotations and translations are the three basic operations involved in
symmetry but there are more. However lets digest these three before considering
glides, screw rotations etc.
94
7.2. Symmetry Groups
Definition: A symmetry operation, for a subset X of the plane, is an isometry f of the
plane that maps X to itself, meaning that f(X) = X. The individual points in X are not
necessarily fixed but the set as a whole is. The set of all symmetry operations of X is
denoted by Sym(X) and is called the symmetry group of X.

Example 1: If X is the following picture of a house then Sym(X) = {I, } where is the
reflection in the axis of symmetry.

(If you have a careful eye youll detect that the mirror symmetry isnt quite perfect since
the chimney and door-handle dont have a corresponding chimney or door-handle on the
other side of the axis. Symmetry, as employed in Western Art, often has small
asymmetry, to make the image more interesting. In the art of the east imperfections in
the symmetry are deliberately included because only God can create perfection.)

Example 2: If X is the Isle of Man insignia then Sym(X) = {I, ,
2
} where is the
rotation about the centre through 120 and
2
is the rotation through 240. The 240
rotation is written as
2
because it is equivalent to performing twice in succession.

Example 3: If X is the following infinite pattern of equally spaces cars arranged in a line
then Sym(X) = { ,
2
,
1
, I, ,
2
,
3
, } where is the translation to the right
through the distance between successive cars. So takes each car to the next on the right,
and so the whole infinite pattern is mapped to itself, or is fixed. The translation
2

moves each car two places to the right,
5
takes each car 5 places to the left, and so on.

95
Example 4: If X is the following infinite pattern then Sym(X) contains translations, but
no reflections or rotations.

However it does contain glides. If is the glide in the horizontal axis, consisting of a
reflection in this axis followed by a suitable translation then each car maps to the nearest
car to the right, on the opposite side of the axis. This glide generates the translations
since
2
is the translation that sends each car to the next, on the same side of the axis.
So Sym(X) is { ,
2
,
1
, I, ,
2
,
3
, }.

The word group was used in the phrase symmetry group because these
symmetry groups are more than just sets of isometries. They have an algebraic structure.
We can multiply any two symmetries, and if each of them fixes a certain subset of the
plane then so does their product. We say that the set of symmetry operations for a given
planar set is closed under multiplication.

A group is a certain type of algebraic system. Its a set on which a binary
operation is defined that satisfies certain properties, called the group axioms. The
operation is called multiplication but it doesnt have to have anything to do with
multiplication of numbers. In fact the elements of the set neednt be numbers. They
could be matrices or functions or points. For the sort of groups well be considering the
elements of the set are symmetry operations, such as rotations, reflections and
translations. The relevant binary operation consists of performing one symmetry
operation after another (in the given order). So the product of a 30 rotation and a 60
rotation (about the same axis) is the effect of doing the 30 rotation and then following
this by a 60 rotation. The net effect is a 90 rotation (about the same axis).

Before an algebraic system with a binary operation is allowed to be called a group
it needs to satisfy the four group axioms.

Closure: The product of any two elements in the set must be again in the set.

Associative Law: A(BC) = (AB)C.
This holds for symmetry operations because in both cases the effect is to perform the
operations A, B, C one after the other, in that order.

Identity: The identity operation is the operation of doing nothing. Clearly its a
symmetry operation for every set because if the individual points are fixed, the overall
pattern must be too. We denote this identity by I. Note that IA = AI for all symmetry
operations, A.
96
We could regard I as a 0 degree rotation, though wed be hard-pressed to work
out the centre. Or we could regard I as a translation through a distance of 0, though wed
have difficulty deciding in which direction wed moved. Instead we treat it as something
special. So when we refer to a rotation here we exclude a 0 degree rotation or a 360.
Both of these would be the identity. And whenever we refer to a translation well never
mean one that moves points through a zero distance.
We can now write equations such as R
4
= I to indicate the fact that if a 90
degree rotation R is performed 4 times in succession the effect on all the points (in so far
as where they end up, not what sort of journey theyve been on) is to fix them (i.e. map
them to themselves).

Inverse: In a group every element has to have an inverse relative to the operation. So for
all A there must exist a B in the set such that AB = I = BA. This is true for symmetry
operations because they can all be undone by a symmetry operation. The inverse of a
rotation through is a rotation through (about the same axis). The inverse of a
translation through a distance h to the right is the translation through of a distance h to
the left. And, of course, the inverse of a reflection is simply the reflection itself.

7.3. The Symmetry Group of a Square
A square has four axes of symmetry. They all run through the centre. Two of
these axes run through the corners and the other two run through the midpoints of
opposite sides.
A B C

R D

Lets call the reflections about these axes A, B, C and D as in the diagram.
Then theres the 90 rotation anti-clockwise. (As usual, for rotations in the plane
we take anti-clockwise as the positive direction.) Lets call this R. A 180 rotation can
then be expressed as R
2
and a 270 rotation (i.e. 90 clockwise) is R
3
. Finally we have
the identity, I.
These eight operations constitute the entire symmetry group of a square.

If X is a square Sym(X) = {I, R, R
2
, R
3
, A, B, C, D}. As with any finite group, we can
exhibit the group table which sets out every possible product. This is the group table for
the symmetry group of a square.
97

I R R
2
R
3
A B C D
I I R R
2
R
3
A B C D
R R R
2
R
3
I B C D A
R
2
R
2
R
3
I R C D A B
R
3
R
3
I R R
2
D A B C
A A D C B I R
3
R
2
R
B B A D C R I R
3
R
2
C C B A D R
2
R I R
3
D D C B A R
3
R
2
R I

You should check some of these out with a small square cut out of paper.
Number the corners on one side of the paper 1, 2, 3, 4 and number the corners on the
back of the paper in such a way that each corner has the same number, front or back.

Place the square so that the corners are as follows:

1 2

3 4

Now perform a pair of operations. For example take AB. Do A first and then B. Its
important to remember that the axes are fixed in space. Dont write them on your paper.
So, for example, B is always the vertical axis.
After doing AB look at the numbers at the corners. You should get:

3 1

4 2

Now ask yourself what single operation, from the set of eight, would have
achieved the same effect. Clearly, in this case, its R
3
so youll have demonstrated that
AB = R
3
.
Examine the table carefully. Notice, for example, that AB BA. Symmetry
operations frequently fail to commute. In other words we often get a different answer if
we multiply in the opposite order.
Youll find that the table has a very definite pattern, with A, B, C advancing in
order, either forwards or backwards as you go across the rows or down the columns. In
particular note that RA = B, R
2
A = B and R
3
A = D. So all eight operations can be
generated by just R and A. Now R
4
= I and A
2
= I. Also note that AR = R
3
A. We
usually write this as AR = R
1
A. (Because R
4
= I it follows that R
3
= R
1
.) These
relations are sufficient to compute all products in the group.
98
R
4
= I
A
2
= I
AR = RA
1
.
If we have any product of Rs and As and their powers the last relation enables us
bring all the Rs to the front and all the As to the back and so we can write it as R
i
A
j
.
The first two relations mean that we can restrict ourselves to i = 0, 1, 2, 3 and j = 0, 1.
Thus we can identify the product as one of the eight combinations: I, R, R
2
, R
3
, A, RA,
R
2
A, R
3
A. These are the eight elements of the group.
For this reason we write this group in the form R, A | R
4
= A
2
= I, AR = R
1
A.
This means that its the group generated by R and A subject to the relations specified. It
is a well-known group, belonging to a family of groups, called the dihedral groups.

7.4. Cyclic and Dihedral Groups
The simplest family of groups is the family of cyclic groups. These are groups
with just one generator. If we just stuck to the rotations in the symmetry group of a
square wed get the group {I, R, R
2
, R
3
} = R | R
4
= I.

Definition: A cyclic group of order n is a group of the form A | A
n
= I, where A is a
generator. (Saying of order n is simply saying that it has size n.) The elements are:
I, R, R
2
, , R
n1
. The infinite cyclic group is generated by a single generator with no
relations and can be expressed as A | . Here the elements are not just I, A, A
2
, A
3
,
but also include the negative powers A
1
, A
2
,

Notation: The cyclic group of order n is denoted by C
n
and the infinite cyclic group is
denoted by C

.

The symmetry groups of each of the sets in examples 1 and 2 are cyclic. Those involving
the infinite patterns of cars are infinite cyclic groups.

Since any two powers of the same generator commute we can say that a cyclic
group is commutative. But a more usual adjective that we use is abelian, in honour
of the Norwegian mathematician Abel (1802 1829), so all cyclic groups are abelian.

The next simplest family of groups, and the place where we meet non-abelian
groups for the first time, is the family of dihedral groups.

Definition: A dihedral group of order 2n is a group of the form
A, B | A
n
= B
2
= I, BA = A
1
B.
The infinite dihedral group is A, B | B
2
= I, BA = A
1
B.
Notation: The dihedral group of order 2n is denoted by D
2n
and the infinite dihedral
group is denoted by D

.
99
Theorem 1: The elements of D
2n
= A, B | A
n
= B
2
= I, BA = A
1
B are:
I, A, A
2
, , A
n 1
, B, AB, A
2
B, , A
n 1
B.
Proof: Because of the relation BA = A
1
B any product involving powers of A and B
can be written as A
i
B
j
for some i, j. Since A
n
= I we can limit i to the range 0 i < n
and since B
2
= I we need only have j = 0, 1. Clearly these 2n elements are distinct.

Theorem 2: D
2n
is abelian if and only if n = 1 or 2.
Proof: Mostly, dihedral groups are non-abelian because BA = A
1
B. But in D
4
, A
2
= I in
which case A = A
1
so BA = AB and so D
4
is abelian. Also D
2
is identical to C
2
and
so is also abelian. We have thus proved the following theorem.

Example 5: Simplify A
4
BA
3
BA
5
BA
2
in the dihedral group D
12

A, B | A
6
= B
2
= I, BA = A
1
B.
Solution: Using the dihedral relation we can write:
A
5
BA
3
BA
5
BA
2
= A
5
BA
3
BA
5
A
2
B = A
5
BA
3
BA
7
B = A
5
BA
3
A
7
B
2
= A
5
BA
10
B
2
= A
5
B
3

and using the other relations we can write this as AB.

A similar analysis to that we made for the square will apply to any regular
polygon. So the symmetry group of an n-sided polygon is D
2n
where the A generator is a
rotation through 360/n and B is any reflection. If X is an equilateral triangle, Sym(X) =
D
6
.
Theres no 2-sided polygon so D
4
cant arise in this way. But the group of
symmetries of a proper rectangle (one that isnt a square) is D
4
.

If X is a proper rectangle Sym(X) = D
4
= A, B | A
2
= B
2
= I, BA = AB = {I, A, B, AB}
where V is a vertical reflection (in the horizontal axis), H is a horizontal reflection (in the
vertical axis) and R = AB is the 180 rotation about the centre.

Definition: A group H is a subgroup of G if its a subset that forms a group in its own
right. For example the rotations in a symmetry group of a set X form a subgroup
Rot(X).

Definition: If G and H are subgroups of some larger group and if:
(1) gh = hg for all g G, h H and
(2) only the identity element belongs to both G and H
then we define G H to be {gh | g G, h H} and we call it the direct product of G
and H.

100
Example 6: A, B, C | A
4
= B
2
= C
3
= I, BA = A
1
B, CA = AC, CB = BC is the direct
product of A, B | A
4
= B
2
= I, BA = A
1
B and C | C
3
= I and so may be written as
D
8
C
3
.

Example 7: Show that D
4
is the same group as C
2
C
2
.
Solution: D
4
= A, B | A
2
= B
2
= I, BA = A
1
B. Since A
2
= I in this group it follows
that A
1
= A, so D
4
= A, B | A
2
= B
2
= I, BA = AB = A | A
2
= I B | B
2
= I
= C
2
C
2
.

Example 8: Find the symmetry group of:
(i) a rhombus;
(ii) a proper parallelogram (one that isnt a rhombus or a rectangle);
(iii) an equilateral triangle;
(iv) an isosceles right-angled triangle.
Solution: (i) A rhombus has two axes of reflection as well as a 180
rotation about the centre. So its symmetry group is D
4
= C
2
C
2
just
like the proper rectangle.

(ii) With a proper parallelogram these reflections no longer fix the figure. All we have is
a 180 rotation and, of course, the identity. The symmetry group is thus C
2
.

(iii) The symmetry group of an equilateral triangle is
D
6
= A, B | A
3
= B
2
= I, BA = A
1
B
where A is a 120 rotation about the centre and B is a 180 rotation
about any of the three axes of symmetry.

(iv) An isosceles right-angled triangle is not equilateral so we have C
2
as the
symmetry group. But in this case the element of order 2 is a reflection instead of
a 180 rotation.

While dihedral groups occur as the symmetry groups of many patterns
(e.g. the Mercedes Benz logo has D
6
symmetry) some patterns have cyclic
symmetric groups. These patterns have rotational symmetry but no mirror
symmetry. The most famous (infamous) of these is the Swastika. All such
patterns come in two varieties, each being the mirror image of the other.
Its a pity this ancient symbol was soiled by the Nazis because its a symbol with
a long history, yet one cannot now help feel uncomfortable seeing it. The earliest
Swastika was found on pots from Persia which are dated from about the
fourth millennium BC. It has also been found Greece, India, China and
Japan, generally as a good-luck charm. (The word swastika is from
Sanskrit meaning all will be well.) Fortunately there is a left-handed and
a right-handed version. The one shown here is not the Nazi symbol but is
its mirror image. The symmetry group of either Swastika is C
4
.

101
The insignia for the Isle of Man (a small island between England and
Ireland a part of the U.K. but with its own parliament) has C
3
as its symmetry
group.

7.5. Finite Symmetry Groups in the Plane
A symmetry group is a way of describing the type of symmetry of a set. For
example the shapes that have just mirror symmetry can be considered to have the same
type of symmetry. Its natural therefore to ask what are the possible types of symmetry
for subsets of the plane. In this section well restrict our attention to just finite symmetry
groups, so that will preclude translations and glides.
Weve seen that any finite cyclic group can occur as the symmetry group of some
planar set, as can any dihedral group.

Example 9: Find a subset of the plane that has D
16
= A, B | A
8
= B
2
= I, BA = A
1
B
as its symmetry group.

Solution: Clearly an example of such a set is a regular octagon, where A is a rotation
through 360/8 = 45 degrees and B is a reflection in any of the eight axes of symmetry.

Example 10: Find a subset of the plane that has C
7
as its symmetry group.
Solution: One can modify the Isle of Man insignia by taking 7 legs, equally spaced
around a circle.

One can build up a set with symmetry group C
n
by taking as a basis any shape
with trivial symmetry group. Wed normally say that such a shape has no symmetry, but
when talking about the symmetry group we must always include the identity.
An example of such a shape is:

We then take any point in the plane and rotate this basic shape through various
multiples of 360/n degrees. The union of all these rotated copies will generally have
symmetry group C
n
. However particular choices may introduce further symmetry.
102
Example 11: The following sets have symmetry group C
6
.

Example 12: Choosing to rotate each piece about the corner we get the following which
has symmetry group D
12
.

Theorem 3: If G is a finite group of rotations about a common point then G is cyclic.
Proof: Let R G be the rotation through the smallest positive angle . Let S be any
non-trivial rotation in G and suppose it is through the positive angle . Let n be the
integer part of /. Then 0 n < . But S(R
n
)
1
G and its a rotation through the
angle n. If n > 0 this contradicts the choice of R. Hence = n and so S = R
n
.

Theorem 4: The finite symmetry groups for subsets of the plane are cyclic or
dihedral.
Proof: Let S be a subset of the plane with a finite symmetry group G. Take a single point
in S and operate on it by all the elements of G. This gives us a finite set of points which
are permuted amongst themselves by the elements of G. Call this set X.
If X = {(x
1
, y
1
), , (x
m
, y
m
)} the centre of gravity is the point P = (x
i
/m, y
i
/m).
While this point neednt be in X, every element of Sym(X) must fix P. We can thus take
P to be the origin and so Sym(X) consists entirely of central isometries.
The rotation group, Rot(X), consists entirely of rotations about the origin and so is
C
n
= R | R
n
= I for some n. Any opposite isometry is a reflection in an axis through the
origin. If theres no mirror symmetry Sym(X) is just C
n
.
Suppose that theres mirror symmetry and let M Sym(X) be any reflection.
Then Sym(X) is generated by R and M. If N is any other reflection then MN is a
rotation. Hence MN = R
k
for some k and hence N = M
1
R
k
= MR
k
. It follows that
Sym(X) is generated by R and M, where R
n
= I and M
2
= I.
Now MR is an opposite isometry and so must be a reflection. Hence (MR)
2
= I
from which we get MRM = R
1
.
Thus Sym(X) = R, M | R
n
= I, M
2
= I, MR = R
1
M which is isomorphic to D
2n
.
103
7.6. The Seven Frieze Patterns:
Definition: A frieze pattern is any subset X of the plane whose translational symmetry
group Sym(X) contains a translation, T, that generates all translations in Sym(X). In
other words, its a linear pattern that repeats in one direction.
Were excluding such subsets as a single line because, although the translations
have a common direction, there are translations in the Symmetry group of a line through
arbitrarily small distances, none of which can generate them all.
We define two patterns to be equivalent if their symmetry groups are isomorphic
in such a way that corresponding isometries have the same type (rotation, reflection etc.)
The following is a list of seven frieze patterns, no two of which are equivalent. Well
later show that these are the only ones. Every frieze pattern is equivalent to one of these
seven.

F:

This has no symmetry, other than translational symmetry in one direction. Its symmetry
group is the infinite cyclic group, C
= T | generated by a translation to the right by one

unit.

E:

This pattern has mirror symmetry in an axis parallel to the direction of the translation. Its
symmetry group is the infinite dihedral group:
D
= T, M | M
2
= I, MT = T
1
M
where T is a generator for the translations and M is the reflection in the axis.

A:

Like the pattern E, above, this has mirror symmetry, as well as the translation symmetry.
But it has infinitely many axes of mirror symmetry, not just one. Its symmetry group is:
T, M | M
2
= I, MT = TM
where T is a generator for the translations and M is any reflection.
It is isomorphic to C
C
2
.
104
pb:

This pattern has no mirror symmetry but it does have glide symmetry. The axis of the
glide is the centre line. A reflection in this line, followed by a translation through half a
unit to the right, is a glide which fixes the whole pattern. Every other glide is the product
of this one and a suitable translation.
The symmetry group for this pattern is also the infinite cyclic group
C
= G | generated by the glide described above. Every translation in the symmetry

group is an even power of this glide while the odd powers are other glides.

N:

This pattern has no mirror or glide symmetry. But it has rotational symmetry about the
centre of each letter N. If we pick one of these rotations, the others can be obtained by
multiplying it by a suitable translation. So the symmetry group is again the infinite
dihedral group:
D
= T, R | R
2
= I, R
1
TR = T
1

where T is the generating translation and R is one of the rotations.

MW:

This pattern has 2-fold rotational symmetry about the points on the horizontal
axes half-way between successive scrolls. In addition theres glide symmetry in the
horizontal axis. The symmetry group is generated by this glide, G, and the 2-fold
rotation, R. The translations are even powers of G, while the odd powers of G are other
glides. The reflections are simply the product of a glide and the rotation.
The symmetry group is again the infinite dihedral group:
D
= G, R | R
2
= 1, M
2
= 1, M
1
GM = G
1

where G is the glide described above and R is one of the rotations.

H:

This pattern has both rotational and mirror symmetry in two directions. The
symmetry group also contains glides, but these are just products of the reflection in the
horizontal axis and a translation.
The symmetry group is generated by a translation, T, through 1 unit, the
reflection M in the horizontal axis, and the 180 reflection about one of the centres of 2-
105
fold rotation. (The reflections in the vertical axes can be expressed in terms of these
generators. The symmetry group of this pattern is thus:
T, M, R | M
2
= R
2
= 1, MT = TM, MR = RM, R
1
TR = T
1

where T is a generating translation, R is a rotation and M is a reflection. This group is
isomorphic to D
C
2
.

Theorem 5: Every frieze pattern is equivalent to one of the above.
Proof: Let X be a frieze pattern where Sym(X) contains a translation T along the
horizontal axis.
The only possible rotational symmetries are 2-fold rotations.
The frieze type is determined by the existence or non-existence of the following
symmetry features:
V = a vertical reflection (in the horizontal axis);
H = a horizontal axis (in a vertical axis);
R = a 2-fold rotation;
G = a horizontal glide.
Thus there are at most 2
4
= 16 possible types of frieze symmetry. But certain
combinations are impossible.

H G: If Sym(X) contains the reflection in the horizontal axis it also contains a glide.
RG H: The product of a 2-fold rotation and a horizontal glide fixes all horizontal
lines, so it must be a horizontal reflection (in a vertical axis).
GV R: The product of a horizontal glide and a vertical reflection fixes a point. Being
a direct isometry it must be a rotation.
RV G: The product of a rotation and a vertical reflection fixes only one horizontal line
(the horizontal axis). It is therefore either a horizontal reflection or a horizontal glide. In
either case Sym(X) contains a glide.

It follows that only the following combinations are possible:
combination frieze type
none F
V A
R N
G pb
HG E
VG MW
VRGH H

Example 13: Classify the following frieze patterns:
(i) (ii)

(iii) (iv)

106
(v) (vi)

(vii)

Solution: (i) pb; (ii) H; (iii) E; (iv) F; (v) N; (vi) A; (vii) MW.

Example 14: Classify the graphs of the following functions as frieze patterns:
(i) sin x; (ii) cos x; (iii) tan x; (iv) |sin

x| .
Solution: (i) MW; (ii) MW; (iii) N; (iv) A.

7.7. The 17 Wallpaper Patterns
A wallpaper pattern is any subset, X, of the plane whose translational symmetry
group Sym(X) contains translations, S, T, in two different directions, which together
generate all the translations in Sym(X). In other words theyre patterns that cover the
whole plane and are repetitive in two independent directions.
There are 17 different types of wallpaper patterns. We wont provide a proof, but
we do describe these 17 types. There are several different naming conventions for them.
Well use three parameters to describe each one.

xyz will mean that:
x is the maximal rotational symmetry about any point;
y is the maximum number of mirror axes through any point;
z is the maximum number of glide axes through any point.
represents a mirror axis
represents a glide axis
represent centres of rotational symmetry

100:

107
101:

111:

110:

200:

Here we have 2-fold rotational symmetry but no reflectional or glide symmetry
108
202:

211:

222:

There are mirror and glide axes in both directions as well as 2-fold
poles.

220:

Here there are 2-fold poles and horizontal and vertical mirror symmetry. But there are no
proper glides.

300:

This pattern has 3-fold rotational symmetry but, because of the asymmetry of the blades
of the propeller, there is no reflectional or glide symmetry.
109

332A:

Here we have 3-fold rotational symmetry together with both reflectional and glide
symmetry.

332B:

Here theres 3-fold rotational symmetry as well as both reflectional and glide symmetry.
But unlike the previous pattern, all the poles lie on mirror axes.

400:

In this pattern there are some poles with 4-fold symmetry and others with only 2-fold
symmetry. There is no reflectional or glide symmetry.

423:

110
For this pattern there are both 2-fold and 4-fold poles as well as rotational and glide
symmetry in both directions.

442:

600:

664:

Here, for clarity, weve only shown the glide axes through a single point.
111

Exercise 1:
Describe the following wallpaper patterns by the code xyz where:
x = maximum rotational symmetry about any point;
y = the maximum number of mirror axes through any point;
z = the maximum number of (proper) glide axes through any point;

(i) (ii) (iii)

Exercise 2:
A certain room has exposed brick walls, a wooden parquet floor, square acoustic tiles in
the ceiling and diamond-pane leaded glass windows (each pane is a rhombus).

Write down the symbol xyz for each of these patterns (assuming is a portion of a
complete wall-papering of the plane) where:
x is the maximum degree of rotational symmetry about any point,
y is the maximum number of mirror axes through any point and
z is the maximum number of proper glide axes through any point.

(i) bricks (ii) parquet flooring (iii) ceiling tiles (iv) window panes
112
Exercise 3: Identify the rotational, mirror and glide symmetry for each of the following
frieze patterns:
(i)

(ii)

(iii)

HHHHHHHHHHHHHHHHHHH


Exercise 1: (i) 222, (ii) 664; (iii) 222.

Exercise 2:
(i) bricks 222; (ii) parquet floor 423; (iii) ceiling 442; (iv) window panes 222.

Exercise 3:

(i)

Rotational symmetry: This has 2-fold rotational symmetry about the midpoint of each
line.
Mirror Symmetry: none
Glide Symmetry: none

(ii)

Rotational symmetry: This has 2-fold rotational symmetry about the points on the
horizontal axis midway between successive scrolls.
Mirror Symmetry: This has vertical axes of symmetry through the middle of each
scroll.
113
Glide symmetry: This has a horizontal axis of glide symmetry. If the scrolls on each
side of the line are 1 unit apart, reflecting the pattern in the horizontal axis and translating
to the right through a distance of fixes the pattern.

(iii)

HHHHHHHHHHHHHHHHHHH
Rotational symmetry: This has 2-fold rotational symmetry about the midpoints of each
H and about the points midway between two successive Hs
Mirror Symmetry: This has a horizontal axis of symmetry and infinitely many vertical
axes of symmetry. The vertical axes go through the centres of rotational symmetry.
Glide Symmetry: There are no proper glides in the symmetry group of this pattern.
114

115
8. ISOMETRIES OF R
3

8.1 Central Isometries of R
3

In addition to the familiar rotations and reflections we need to consider two new
types of central isometries of R
3
.

Definition: A rotary reflection is a rotation about an axis followed by a reflection in a
plane perpendicular to that axis. A special case of a rotary reflection is the antipodal
map A(v) = v. This is a 180 rotation followed by a reflection in a plane perpendicular
to the axis.

Theorem 1: Every central isometry of R
3
is one of the following:
the identity;
a rotation;
a reflection in a plane;
a rotary reflection.
Proof: Let f be a central isometry of R
3
and let A be its matrix relative to the standard
basis. Then A is orthogonal, which means that its diagonalizable, and its eigenvalues lie
on the unit circle.
We first consider the case where the eigenvalues are real. Lying on the unit circle
they must be 1. We can choose an orthonormal basis for R
3
that consists of three,
mutually orthogonal, eigenvectors each with unit length and if we represent a typical
point as (x, y, z), relative to this basis then f must map it to (x, y, z) where ,
and are the respective eigenvalues. This gives rise to the following four cases.

Case 1 (A has eigenvalues 1, 1, 1): Since A is similar to the identity matrix, A = I
and so f is the identity map.

Case 2 (A has eigenvalues 1, 1, 1): If the coordinates of a point, relative to the
eigenvectors, are (x, y, z) the point is mapped to (x, y, z). Here f is a reflection in the
plane spanned by the eigenvectors corresponding to the eigenvalue 1.

v
f(v)
0
eigenspace for eigenvalue 1
116
Case 3 (A has eigenvalues 1, 1, 1): If the coordinates of a point, relative to the
eigenvectors, are (x, y, z) the point is mapped to (x, y, z). Here f is a reflection in
the line spanned by an eigenvector corresponding to the eigenvalue 1. Here f is a 180
rotation about the line through an eigenvector corresponding to the eigenvalue 1.

Case 4 (A has eigenvalues 1, 1, 1): Here f is the antipodal map, a special case of
a rotary reflection.

If the eigenvalues arent all real then they must consist of one real eigenvalue plus
a conjugate pair of non-real ones. The real eigenvalue must be 1 and the non-real ones
must have the form e
i
.

Case 5 (A has eigenvalues 1, e
i
): Relative to a corresponding (orthonormal) basis
of eigenvectors A =
\
|
.
|
| 1 0
0 R
where R is a 22 real orthogonal matrix with non-real
eigenvalues. Thus R =
\
|
.
|
|
cos sin
sin cos
for some and so
A =
\
|
.
|
|
|
1 0 0
0 cos sin
0 sin cos
or
\
|
.
|
|
| 1 0 0
0 cos sin
0 sin cos
.
In the first case f is a rotation about the line through the eigenvector corresponding to the
eigenvalues 1. In the second case it is a rotary reflection.

Definition: The group O(3) is the group of all central isometries of R
3
. The group
SO(3) is the group of all direct central isometries (ie the identity plus all the rotations).

The product of two rotations in R
3
is a rotation (or the identity). If their axes are
different the composite rotation will be about a different axis altogether.

Example 1: If is the rotation through 90 about (0, 1, 0) and is the rotation through
90 about (1, 0, 0), describe .
Solution: is clearly a rotation. But we need to find its axis and the angle of rotation.
Let R and S be the respective matrices of and relative to the standard basis of R
3
.
Then R =
\
|
.
|
|
|
0 0 1
0 1 0
1 0 0
and S =
\
|
.
|
|
|
1 0 0
0 0 1
0 1 0
. The matrix for (first and then ) will
f(v)
v
0
eigenspace for eigenvalue 1
117
be SR =
\
|
.
|
|
|
0 0 1
1 0 0
0 1 0
. Clearly
\
|
.
|
|
|
1
1
1
is an eigenvector with corresponding eigenvalue 1. This
gives the axis of the rotation.
The vector
\
|
.
|
|
|
1
1
0
(a convenient point in the plane x + y + z, perpendicular to the axis)
maps to
\
|
.
|
|
|
0 0 1
1 0 0
0 1 0

\
|
.
|
|
|
1
1
0
=
\
|
.
|
|
|
0
1
1
. The angle between these is cos
1
1
2. 2

= cos
1
(1/2) = 120. So is a rotation about (1, 1, 1) through 120.

Example 2: Find the matrix of the rotation about the vector (1, 1, 1) through 40.
Solution: Let p = (1, 1, 1). We choose q = (1, 0, 1) as a convenient vector orthogonal
to a. Then r = p q = (1, 2, 1) is mutually orthogonal to both. Moreover p, q, r form a
positive triple.
We now make these unit vectors. For convenience well reuse the same names.
So we now have p =
1
3
(1, 1, 1), q =
1
2
(1, 0, 1) and r =
1
6
(1, 2, 1).

r
R(r)

R(q)
40

40 q

Thus, if c = cos40 and s = sin 40, R(p) = p; R(q) = cq + sr; R(r) = cr sq.

So p = (0.5774, 0.5574, 0.5574),
q = (0.7071, 0, 0.7071),
r = (0.4082, 0.8165, 0.4082),
R(p) = (0.5774, 0.5574, 0.5574),
R(q) = (0.2793, 0.5248, 0.8041)
R(r) = (0.7672, 0.6255, 0.1418)
Thus R
\
|
.
|
|
|
0.5774 0.7071 0.4082
0.5574 0 0.8165
0.5574 0.7071 0.4082
=
\
|
.
|
|
|
0.5774 0.2793 0.7672
0.5774 0.5248 0.6255
0.5574 0.8041 0.1418
.
Hence R =
\
|
.
|
|
|
0.5774 0.2793 0.7672
0.5774 0.5248 0.6255
0.5574 0.8041 0.1418

\
|
.
|
|
|
0.5774 0.7071 0.4082
0.5574 0 0.8165
0.5574 0.7071 0.4082

1
=
\
|
.
|
|
|
0.5774 0.2793 0.7672
0.5774 0.5248 0.6255
0.5574 0.8041 0.1418

\
|
.
|
|
|
0.5774 0.7071 0.4082
0.5574 0 0.8165
0.5574 0.7071 0.4082

T
118
=
\
|
.
|
|
|
0.5774 0.2793 0.7672
0.5774 0.5248 0.6255
0.5574 0.8041 0.1418

\
|
.
|
|
|
0.5774 0.5774 0.5774
0.7071 0 0.7071
0.4082 0.8165 0.4082

=
\
|
.
|
|
| 0.8441 0.2930 0.4491
0.4491 0.8441 0.2931
0.2931 0.4492 0.8441
=
\
|
.
|
|
| 0.844 0.293 0.449
0.449 0.844 0.293
0.293 0.449 0.844
to 3 decimal places.

Example 3: The matrix of the rotation through 90 about the vector (0, 3, 4) is:
S =
\
|
.
|
|
|
0 0.8 0.6
0.8 0.36 0.48
0.6 0.48 0.64
.
If R is the rotation in the previous exercise describe the rotation RS by giving the axis
and the angle of rotation.

Solution: RS =
\
|
.
|
|
| 0.844 0.293 0.449
0.449 0.844 0.293
0.293 0.449 0.844

\
|
.
|
|
|
0 0.8 0.6
0.8 0.36 0.48
0.6 0.48 0.64
=
\
|
.
|
|
| 0.504 0.565 0.653
0.851 0.196 0.487
0.147 0.801 0.580
.
We now find an eigenvector corresponding to the eigenvalues 1.
\
|
.
|
|
| 1.504 0.565 0.653
0.851 1.196 0.487
0.147 0.801 0.420

\
|
.
|
|
| 1 0.376 0.434
0.851 0.804 1.196
0.147 0.801 0.420

\
|
.
|
|
| 1 0.376 0.434
0 1.516 0.857
0 0.857 0.484

|
.
|
|
| 1 0.376 0.434
0 1 0.565
0 0 0
. Let z = 1. Then y = 0.565, x = 0.646.
Thus the axis of the rotation is the line through the origin and (0.646, 0.565, 1).
The plane through the origin perpendicular to this line is 0.646 0.565y + z = 0 and a
suitable point on this plane is (0, 1, 0.565).
This rotates to
\
|
.
|
|
| 0.504 0.565 0.653
0.851 0.196 0.487
0.147 0.801 0.580

\
|
.
|
|
|
0
1
0.565
=
\
|
.
|
|
| 0.196
0.079
1.129
.
As a check we verify that the square of the length of these two vectors is the same: 1.319.
If is the angle between these vectors then cos =
0.717
1.319
= 0.544 and hence = 57
(to the nearest degree).
So RS is a rotation about the line
x
0.646
=
y
0.565
= z through the angle 57.
119
8.2. Platonic Solids
Definition: A polyhedron is a closed surface R
3
made up of plane faces, each of which
is a polygon.

A regular polygon is one, like a square, in which all edges have the same length
and hence, all angles are equal. We shall consider polyhedra in which the faces are
regular polygons, all of the same size.

Definition: A Platonic solid (or regular polyhedron) is a polyhedron whose faces are
identical regular polygons.

The following is a list of regular polyhedra. We shall prove later that this list is
complete and that only these five Platonic solids exist. Here V is the number of vertices,
F is the number of faces and E is the number of edges. By Eulers Theorem:
V + F E = 2.

TETRAHEDRON: V = 4, E = 6, F = 4

Each face is an equilateral triangle. This shape is sometimes used for cardboard drink
containers.

CUBE: V = 8, E = 12, F = 6

OCTAHEDRON: V = 6, E = 12, F = 8

120
ICOSAHEDRON: V = 12, E = 30, F = 20

DODECAHEDRON: V = 20, E = 30, F = 12

The dodecahedron was featured in the closing ceremony of the Sydney Olympics.

Theorem 2: There are five Platonic solids: the tetrahedron, the cube, the octahedron, the
icosahedron and the dodecahedron.
Proof: Let P be a polyhedron with V vertices, E edges and F faces. Let d be the
number of edges coming in to each point and let f be the number of edges around each
face.
There are f edges around each face and F faces. But each edge is shared by two
faces. Thus E =
f F
2
. Also, there are v edges around each vertex and V vertices, but
each edge is shared by two vertices, so E =
v V
2
. Thus f F = dV = 2E.
Thus F =
2E
f
and V =
2E
d
. Since V + F E = 2, we have
2E
d
+
2E
f
E = 2 .
So E
\
|
.
|
|
2
d
+
2
f
1 = 2 . Now d, f are integers and each is at least 3. So there are only five
solutions to this equation:
d = 3, f = 3, E = 6 (and so V = 4, F = 4, so P is a tetrahedron);
d = 3, f = 4, E = 12 (and so V = 8, F = 6, so P is a cube);
d = 3, f = 5, E = 30 (and so V = 20, F = 12, so P is a dodecahedron);
d = 4, f = 3, E = 12 (and so V = 12, F = 8, so P is an octahedron);
d = 5, f = 3, E = 30 (and so V = 12, F = 20, so P is an icosahedron).

Definition: Let P be a polyhedron. The dual of P is the polyhedron obtained from P as
follows: take a vertex in the middle of each face and join these new vertices by an edge
whenever the corresponding faces are adjacent in P.

121
The dual of a Platonic solid is clearly a Platonic solid. The tetrahedron is self
dual. The other four fall into two dual pairs. The cube and the octahedron are duals of
one another, as are the dodecahedron and the icosahedron.

8.3. Rotation Groups of the Platonic Solids
TETRAHEDRON:
There are four axes of 3-fold rotational symmetry, one through each vertex and the centre
of the opposite face. Since two rotations share each axis, a 120 rotation and a 240
rotation, this gives eight rotations of order 3.
But, in addition, there are three axes of 2-fold symmetry (180
rotations). These join the midpoint of one edge to the midpoint of the
opposite edge (the one that they dont meet). This gives an additional
three rotations in the symmetry group. The identity makes the 12
th

element so |Rot(tetrahedron)| = 12 and |Sym(tetrahedron)| = 24.
For those who know more group theory than weve given
here we point out that in fact Sym(tetrahedron) is isomorphic to S
4
,
the group of permutations on {1, 2, 3, 4} since the map that takes
every isometry to the corresponding permutation of the four vertices
is a homomorphism onto S
4
(i.e. every permutation can be produced by an isometry) and
its kernel is trivial. Rot(tetrahedron) is isomorphic to A
4
(the group of even permutations
on {1, 2, 3, 4}).

CUBE:
The cube has three axes of 4-fold symmetry, one through the centres of each pair
of opposite faces. For each of these three axes there are three rotations, two of them 4-
fold and the remaining one 2-fold. This gives altogether six 4-fold rotations and three 2-
fold rotations.
There are four axes of 3-fold rotation, one through each pair of opposite vertices,
with two 3-fold rotations about each. This makes eight 3-fold rotations.
There are six axes of 2-fold rotation, one through each pair of opposite edges,
giving six more 2-fold rotations. And last, but not least, is the identity making 24
rotations altogether in Rot(cube). So |Rot(cube)| = 24 and |Sym(cube)| = 48.
The group Rot(cube) is, in fact, S
4
, the group of all permutations on {1, 2, 3, 4}.
To see this we have to find four things that are permuted by the isometries in Rot(cube)
in all possible ways. There are six faces, but not all 6! permutations are possible. For
example, opposite faces must always remain opposite. There are eight vertices, but only
four pairs of opposite vertices or, in other words, four diagonals. These are permuted by
Rot(X) in all 4! possible ways.

122
DODECAHEDRON:
The dodecahedron has twelve faces, and so six axes of 5-fold rotational
symmetry, one from the centre of each face to the centre of the opposite face. Each of
these axes has four 5-fold rotations, giving altogether 24
rotations.
From each vertex to the opposite vertex theres an axis of
3-fold symmetry. With ten such axes and two rotations about
each, we get twenty more rotations.
From the midpoint of each edge to the midpoint of the
opposite edge we get an axis of 2-fold rotational symmetry.
There are fifteen such axes and so fifteen more rotations.
Then, there is the identity, giving 1 + 24 + 20 + 15 = 60
elements altogether. So |Rot(dodecahedron)| = 60 and
|Sym(dodecahedron)| = 120.
Each of the twelve faces has five diagonals, making 60 in all. These form the
vertices of five internal cubes that are permuted by the isometries in Sym(dodecahedron).
In this way it can be shown that Sym(dodecahedron) is S
5
and Rot(dodecahedron) is A
5
.

OCTAHEDRON & ICOSAHEDRON:
These are duals of the cube and the dodecahedron. Its not difficult to see that
dual polyhedra have identical symmetry. So Sym(octahedron) is S
4
and
Sym(icosahedron) is S
5
.

8.4. Groups Acting on a Set
Definition: If G is a group we say that G acts on a set, X if there is a function
: X G X such that:
(1) x 1 = x for all x X and
(2) (x g) h = x (gh) for all x X and g, h G.

We have here a primitive analogue of the vector space from linear algebra. We
can think of the elements of X as being like vectors and the elements of G as scalars, with
* as a scalar multiplication. There are many examples of G-sets, and the concept is very
useful in Group Theory. However the only example that will concern us here is the fact
that isometries act on sets of points.

Definition: Suppose G acts on X and let x X. The orbit containing x is defined by:
xG = {xg | g G}. Its the set of all those elements of X that can be reached from x by
multiplying it by some element of G.

Definition: The stabiliser of x X is G
x
= {g G | x g = x}. Clearly its a subgroup
of G.

Definition: The fixed set of g G is X
g
= {x X | x * g = x}.

123
Definition: If a central rotation acts on the unit sphere its fixed set consists of the two
points where the axis of rotation cuts the sphere. These are called the poles of the
rotation.

Example 4: Let X be the set consisting of the corners of a cube, together with the face
centres and the centre of the cube. Clearly |X| = 15. The symmetry group of the cube, G,
acts on X and there are 3 orbits. The eight corners form one orbit
since you can get from any corner to any other by a suitable element
of G. The other orbits are the six face centres and, the smallest orbit
of all, consisting of just the centre of the cube.

The stabilizer of a corner is the group of order 6, consisting of the
identity and the two 3-fold rotations about the diagonal through the corner as well as
three reflections. The stabilizer of a face centre has order 8 (the identity, three rotations
and four reflections). The stabilizer of the centre of the cube is the whole of G.

The fixed set of a reflection in the horizontal mirror plane is the set of five points on that
plane. The fixed set of a 4-fold rotation about the axis joining two face centres has three
elements those two points and the centre of the cube.

Definition: Every subgroup H of a group G acts on itself by multiplication (on the right).
The orbits are called the right cosets.
gH is the right coset containing g G, is defined by: gH = {gh| h H}.

Theorem 3: If the finite group G acts on the finite set X:
|xG| = |G|/|G
x
|.
Proof: x*g = x*h x*(gh
1
) = x gh
1
G
x
gG
x
= hG
x
. So f(x*g) = gG
x
is a well-
defined 1-1 and onto map between xG and the set of right cosets of the stabiliser G
x
.

Example 5: If x is a corner of the cube, in the above example, |xG| = 8, since the orbit xG
consists of all 8 corners. The stabilizer G
x
, as weve seen, has order 6. And |G| = 48.
Note that 8 = 48/6.

Theorem 4: If the finite group G acts on a set X the number of orbits is the average
number of elements of X fixed by the elements of G.
Proof: Let X
g
be the set of elements of X which are fixed by g G. Then the average
number of elements fixed by the elements of G is
1
|G|
.
gG
|X
g
| .
Now the number of pairs (x, g) such that g fixes x is
gG
|X
g
| =
xX
|G
x
| =
xX
|G|
|xG|
.
The elements of the orbit xG each contribute
|G|
|xG|
to this sum, and as there are |xG| of
them, together they contribute |G|. Thus
gG
|X
g
| = |G| number of orbits.

124
Example 6: Suppose now that G is the rotational symmetry group of a cube, so |G| = 24.
It acts on the set X of 27 points we defined earlier. We can take a census of the fixed
points of the elements of G.

Type axis n = # isometries f = #fixed points nf
Identity 1 15 15
4-fold face
centres
6 3 18
2-fold face
centres
3 3 9
3-fold corners 8 3 24
2-fold edge
midpoints
6 1 6
TOTAL 24 72

# orbits = 3 = 72/24.

8.5. Finite Rotation Groups of Subsets of R
3

Theorem 5: Let G be a finite group of rotations of a subset of R
3
. Then G is either
cyclic or dihedral or has order 12, 24 or 60.
Proof: We may clearly suppose that |G| 2. Let X be the set of poles. Since G contains
|G| 1 rotations, each of which has two poles, |X| 2(|G| 1). Now the elements of G
act on these poles. Let k be the number of orbits.
Then k is the average number of elements fixed by the elements of G.
So k =
2(|G| 1) + |X|
|G|
and so |X| = (k 2)|G| + 2.
Now 2 |X| 2 (|G| 1) and hence 2 k 4
2
|G|
. Hence k = 2 or 3.

Case I: k = 2: Hence |X| = 2. Thus the non-trivial elements of G have a common axis
and so, by the Lemma, G is cyclic.

Case II: k = 3: Hence |X| = |G| + 2. Let the three orbits be P, Q, R with sizes p q r.
Then p + q + r = |X| = |G| + 2. Choose x P, y Q, z R.
Their stabilizers have sizes a =
|G|
p
, b =
|G|
q
, c =
|G|
r
. Clearly a b c.
Now
|G|
a
+
|G|
b
+
|G|
c
= p + q + r = |G| + 2, so
1
a
+
1
b
+
1
c
= 1 +
2
|G|
. Thus a = 1 or 2.
But since |X| > 2 the rotations in G do not all have the same axis and so the stabilizers are
proper subgroups of G. Thus b, c < |G| and so a > 1. So a = 2 and so
1
b
+
1
c
=
1
2
+
2
|G|
.
Thus b = 2 or 3.
125
Case IIA: a = 2, b = 2: Hence c = |G|/2 and hence r = 2. So R = {z, z} for some pole z.
The rotations in G
z
have a common axis and so it is cyclic.
Let g G
z
and suppose g has poles t. Clearly t z so t P or t Q.
|G
t
| = 2. Thus G
t
= {I, t}. Thus t
2
= I.
But also gt G
z
and so (gt)
2
= I. gt = (gt)
1
= t
1
g
1
= tg
1
so t
1
gt = g
1
.
Thus, for some n, G = (g, t | g
n
= 1, t
2
= 1, t
1
gt = g
1
). Thus G is the dihedral group of
order 2n.
Case IIB: a = 2, b = 3: Thus
1
c
=
1
6
+
2
|G|
so |G| =
12c
6 c
. Clearly c < 6.
Case IIB(i): a = 2, b = 3, c = 3: |G| = 12.
Case IIB(ii): a = 2, b = 3, c = 4: |G| = 24.
Case IIB(iii): a = 2, b = 3, c = 5: |G| = 60.

With a bit more group theory in these last three cases we could show that G is
either A
4
(the rotation group of the regular tetrahedron or octahedron), S
4
(the rotation
group of a cube) or A
5
(the rotation group of an icosahedron or dodecahedron).


Exercise 1: Find the order (size) of Sym(X) when X is a rectangular box
of dimensions 1 1 2. Describe each of the elements of Sym(X) in
terms of rotations, reflections etc, giving axes and angles of rotations and
mirror planes.

Exercise 2:
(a) Show that the matrix of the rotation R about the axis x = y = z through 40 when
viewed from (1, 1, 1) towards the origin is (to 3 decimal places):
R =
\
|
.
|
|
| 0.844 0.293 0.449
0.449 0.844 0.293
0.293 0.449 0.844
.
(b) Find the matrix of the rotation S about the axis x = 0, 4y = 3z through 90 when
viewed from (0, 3, 4) towards the origin.

(c) Describe the rotation RS by giving the axis and the angle of rotation.

(d) If T is the translation T(v) = v + (1, 0, 0) describe the isometry ST.
126
Exercise 3: Describe the rotational symmetry of the five Platonic solids, as illustrated
below. What is the size of the rotation group in each case?

Tetrahedron Cube Octahedron Dodecahedron Icosahedron


Exercise 1: There is an axis of 4-fold rotational symmetry through the centres of the
square faces. This gives two 90 rotations R, R
3
plus one 180 rotation R
2
. There are
axes of 2-fold symmetry about the lines joining the midpoints of the longer sides to the
midpoint of the opposite side. Also there is an axis of 2-fold rotational symmetry through
the centres of opposite non-square faces. Together with the identity, this makes up the
rotation group of order 8. Thus the order of Sym(X) is 16.
The opposite isometries include reflections in the following planes:
the vertical planes through the diagonals of the square faces;
the vertical planes through the midpoints of opposite edges of the square faces;
the reflection M in the horizontal plane through the centre of X.
This makes 5 reflections. Then there is the reflectional rotations RM and R
3
M.
Finally there is the antipodal isometry v v (which is R
2
M).
Thus |Sym(X)| = 16. The elements of Sym(X) can be summarised as follows:

TYPE #
identity 1
4-fold rotations 2
2-fold rotations 5
reflections 5
rotary reflections 2
antipodal isometry 1
TOTAL 16

Although the question did not ask for a presentation of Sym(X) we can provide it as:
(A, B, C |A
4
= B
2
= C
2
= 1, BA = A
1
B, AC = CA, BC = CB), which is D
8
C
2
.

Exercise 2:
(a) Let p = (1, 1, 1). We choose q = (1, 0, 1) as a convenient vector orthogonal to a.
Then r = p q = (1, 2, 1) is mutually orthogonal to both. Moreover p, q, r form a
positive triple.
127
We now make these unit vectors. For convenience we will reuse the same names.
So now we have p =
1
3
(1, 1, 1), q =
1
2
(1, 0, 1) and r =
1
6
(1, 2, 1).
r
R(r)

R(q)
40

40 q

Thus, if c = cos40 and s = sin 40, R(p) = p; R(q) = cq + sr; R(r) = cr sq.
So p = (0.5774, 0.5574, 0.5574),
q = (0.7071, 0, 0.7071),
r = (0.4082, 0.8165, 0.4082),
R(p) = (0.5774, 0.5574, 0.5574),
R(q) = (0.2793, 0.5248, 0.8041)
R(r) = (0.7672, 0.6255, 0.1418)
Thus R
\
|
.
|
|
|
0.5774 0.7071 0.4082
0.5574 0 0.8165
0.5574 0.7071 0.4082
=
\
|
.
|
|
|
0.5774 0.2793 0.7672
0.5774 0.5248 0.6255
0.5574 0.8041 0.1418
.
Hence R =
\
|
.
|
|
|
0.5774 0.2793 0.7672
0.5774 0.5248 0.6255
0.5574 0.8041 0.1418

\
|
.
|
|
|
0.5774 0.7071 0.4082
0.5574 0 0.8165
0.5574 0.7071 0.4082

1
=
\
|
.
|
|
|
0.5774 0.2793 0.7672
0.5774 0.5248 0.6255
0.5574 0.8041 0.1418

\
|
.
|
|
|
0.5774 0.7071 0.4082
0.5574 0 0.8165
0.5574 0.7071 0.4082

T
=
\
|
.
|
|
|
0.5774 0.2793 0.7672
0.5774 0.5248 0.6255
0.5574 0.8041 0.1418

\
|
.
|
|
|
0.5774 0.5774 0.5774
0.7071 0 0.7071
0.4082 0.8165 0.4082

=
\
|
.
|
|
| 0.8441 0.2930 0.4491
0.4491 0.8441 0.2931
0.2931 0.4492 0.8441
=
\
|
.
|
|
| 0.844 0.293 0.449
0.449 0.844 0.293
0.293 0.449 0.844
to 3 decimal places.

(b) Let p = (0, 3, 4). We choose q = (0, 4, 3) as a convenient vector orthogonal to a.
Then r = p q = (25, 0, 0) is mutually orthogonal to both. Moreover p, q, r form a
positive triple.
We now make these unit vectors. For convenience we will reuse the same names.
So now we have p = (0, 0.6, 0.8), q = (0, 0.8, 0.6) and r = (1, 0, 0).
Thus S(p) = p; S(q) = r; S(r) = q.
So S
\
|
.
|
|
|
0 0 1
0.6 0.8 0
0.8 0.6 0
=
\
|
.
|
|
|
0 1 0
0.6 0 0.8
0.8 0 0.6
.
128
Hence S =
\
|
.
|
|
|
0 1 0
0.6 0 0.8
0.8 0 0.6

\
|
.
|
|
|
0 0 1
0.6 0.8 0
0.8 0.6 0

1
=
\
|
.
|
|
|
0 1 0
0.6 0 0.8
0.8 0 0.6

\
|
.
|
|
|
0 0 1
0.6 0.8 0
0.8 0.6 0

T
=
\
|
.
|
|
|
0 1 0
0.6 0 0.8
0.8 0 0.6

\
|
.
|
|
|
0 0.6 0.8
0 0.8 0.6
1 0 0

=
\
|
.
|
|
|
0 0.8 0.6
0.8 0.36 0.48
0.6 0.48 0.64
.
(c) RS =
\
|
.
|
|
| 0.844 0.293 0.449
0.449 0.844 0.293
0.293 0.449 0.844

\
|
.
|
|
|
0 0.8 0.6
0.8 0.36 0.48
0.6 0.48 0.64
=
\
|
.
|
|
| 0.504 0.565 0.653
0.851 0.196 0.487
0.147 0.801 0.580
.
We now find an eigenvector corresponding to the eigenvalues 1.
\
|
.
|
|
| 1.504 0.565 0.653
0.851 1.196 0.487
0.147 0.801 0.420

\
|
.
|
|
| 1 0.376 0.434
0.851 1.196 0.487
0.147 0.801 0.420

\
|
.
|
|
| 1 0.376 0.434
0 1.516 0.857
0 0.857 0.484

|
.
|
|
|
1 0.376 0.434
0 1 0.565
0 0 0
. Let z = 1. Then y = 0.565, x = 0.222.
Thus the axis of the rotation is the line through the origin and (0.222, 0.565, 1).
The plane through the origin perpendicular to this line is 0.222 + 0.565y + z = 0 and a
suitable point on this plane is (0, 1, 0.565). This rotates to

\
|
.
|
|
| 0.504 0.565 0.653
0.851 0.196 0.487
0.147 0.801 0.580

\
|
.
|
|
|
0
1
0.565
=
\
|
.
|
|
|
0.934
0.471
0.473
.
As a check we verify that the square of the length of these two vectors is the same, 1.148.
If is the angle between these vectors then cos =
0.738
(1.148)
2 = 0.560
and hence = 124 (to the nearest degree).
So RS is a rotation about the line:
x
0.222
=
y
0.565
= z through the angle 124 .

(d) ST is clearly a 90 rotation. We need to find its axis, so we solve RS(v) = v.
If v = (x, y, z) this system of equations becomes:
0.8y + 0.6z + 1 = x
0.8x + 0.36y + 0.48z = y
0.6x + 0.48y + 0.64z = z
Thus
\
|
.
|
|
| 1 0.8 0.6 1
0.8 0.64 0.48 0
0.6 0.48 0.36 0

\
|
.
|
|
| 1 0.8 0.6 1
0 1.28 0.96 0.8
0 0.96 0.72 0.6

\
|
.
|
|
|
1 0.8 0.6 1
0 1 0.75 0.625
0 0 0 0
.
So z = k, y = 0.75k, x = k. Thus ST is a 90 rotation about the vector (4, 3, 4).
129

Exercise 3: (i) 12; (ii) 24; (iii) 24; (iv) 60; (v) 60.
130

131
9. RULER AND COMPASS
CONSTRUCTIONS

9.1. Ruler and Compass Constructions
Many geometric constructions can be carried out with just two tools a ruler and a
compass (and, of course, a sharp pencil!) The classic examples are bisection bisection of
intervals and of angles. Strictly speaking, instead of a ruler we should be talking about a
straight edge, because a ruler can be used to make measurements, whereas the only use that
we allow for the straight edge is the drawing of a straight line between distinct points.
Although many ruler and compass constructions have a practical use, were taking
here a purely theoretical approach where we use idealised instruments. Our pencil is
assumed to be infinitely sharp, our straight edge is a perfect straight line and our compass
draws arcs of perfect circles.
A ruler and compass construction is carried out in the Euclidean Plane. We begin
with a finite set of points which we call marked points because they have been specifically
identified. As we proceed with our constructions we mark additional points. Although our
pencil produces infinitely many additional points when we draw a line, only those that occur
as intersections in the construction will be considered as marked.

At each stage in the construction well have a finite set of marked points. We can
extend the construction as follows:

Draw the line through any two distinct marked points. (In practice this line would be
a line segment, with two endpoints but in theory the lines would extend indefinitely.)

Draw the circle with one marked point as centre, passing through another marked
point.

Having additional marked points in our construction we can now draw more lines and
more circles and so continue the construction as far as we wish. Note that we dont have to
include all possible lines or circles at each stage in our construction. We usually have a goal
and we choose just what we need to construct to achieve our goal.

Example 1: Bisect an interval by ruler and compass.
Solution: Suppose the endpoints of the interval are P and Q.
(1) With centre P draw the circle passing through Q.
(2) With centre Q draw the circle passing through P.
Suppose that the two points of intersection of these circles are F
and G.
(3) Draw FG. The point of intersection, M, of PQ and AB is the
required midpoint.

P Q
A
B
M
132
This is, in fact, the perpendicular bisector. The line AB will be perpendicular to the
original line PQ. The proof that PM = MQ and that AB PQ uses the elementary
geometry of congruent triangles.
Note that APQ is equilateral, so this is also a construction for the equilateral triangle
on a given base.

Example 2: Bisect an angle by ruler and compass.
Solution:

The Greeks considered arithmetic and geometry as being two different ways of
looking at the same number system, and geometrical constructions to perform arithmetic
operations were considered very natural. Since the only numbers they could conceive of
arithmetically were rational numbers they assumed that thats all that could be obtained
geometrically.
So it came as a great shock for them to discover that 2 is irrational, that is a number
that had a proper geometric existence did not have a proper arithmetic one. But then
irrational numbers came to be accepted though they didnt lose their geometric flavour.
Number was a geometric concept. And it was accepted as an unwritten axiom that only
numbers which arose in the context of geometric constructions could possibly exist.
Certain problems were posed where a construction must exist but is hard to find.
The three most famous were the doubling of a cube, the trisection of an angle and the
squaring of a circle. The numbers involved in these three problems clearly exist, they
thought (its interesting that their intuition must have been using some primitive notion of
continuity a concept that took another couple of thousand years to become fully developed),
so clearly there had to be a corresponding construction. And construction meant a
construction using ruler and compass.

Doubling the Cube:
Theres a legend that, when asked for a way of stopping a plague that was attacking
the city of Delos, the Oracle of Delphi advised that the altar of Apollo should be doubled in
size. The altar was in the shape of a cube and although its sides were doubled the plague
continued. The Oracle then revealed that the citizens of Delos had not done as instructed
since they had increased the altar eight-fold. What was required was to double the volume.

The problem of doubling the cube is thus:
Given the side of a cube, construct the side of a cube with twice the volume.

Trisecting an Angle:
Divide any given angle into three equal pieces.

For certain angles, such as 90, it can easily be done (a 30 angle can easily be
constructed). But the problem is to do it for any given angle.
133
Squaring the Circle:
Construct a square whose area is exactly equal to that of a given circle.

The methods allowed in all these constructions are the use of a ruler and compass.
The compass is to be used to draw circles through given points and passing through others.
The ruler must be used solely as a straight-edge for joining points by straight lines, not for
measurement. For this reason, ruler and compass constructions are often called constructions
by straight-edge and compass.
The reason for disallowing measurement is the question of accuracy. The accuracy
with which we can measure lengths is limited by the scale of the markings. Even if we had
the means to magnify the scale wed have to end up making judgements. The whole
philosophy of ruler and compass constructions is to have a procedure that is theoretically
exact.
There do exist ruler and compass methods for getting quite good approximations to
the solutions to all three problems. But thats not the point. With a question of existence its
no good saying that these numbers approximately exist. We need methods that are exact.
And its been shown that all three problems are insoluble by ruler and compass methods. The
geometric concept of number in terms of constructability thus proved to be inadequate and so
the more general concept of real number gradually emerged.

The objects in a ruler and compass construction are points (denoted A, B, C, ... ) and
lines (denoted by a, b, c ... ). Lines include straight lines and circles, which can either be
given, or can be drawn in the course of the construction. Other curves, such as parabolas, can
only occur if theyre given at the beginning of the construction.

LINE(A, B) = the line through A,B (where A B)

CIRC(A, B) = the circle with centre A, passing through B (where A B)

INTERSECTIONS of lines with lines.
A = p q indicates that A is the point of intersection of straight lines p and q or one of the
points of intersection when one of p, q is a circle.
A, B = p q indicates that A and B are the distinct points of intersection of lines p and q
where one of them is a circle.
A p describes A as an arbitrary points on p (distinct from any other marked point).
A, B p describes A, B as distinct arbitrary points on p.

9.2. Examples of Ruler and Compass Constructions
The following are some examples of some standard ruler and compass constructions.

(1) EQU(A, B) is a point C that makes ABC an equilateral triangle

c = CIRC(A, B)
d = CIRC(B, A)
EQU(A, B) = C = d e
A B
C
c d
134
(2) BIS(A, B) is the perpendicular bisector of AB.

c = CIRC(A, B)
d = CIRC(B, A)
C, C' = c d
b = LINE(C, C')

(3) MID(A, B) is the midpoint of the interval AB.

a = LINE(A, B)
b = PERP(A, B)
MID(A, B) = M = a b

(4) REFL(A,B) is the reflection of A in B

a = LINE(A, B)
c = CIRC(B, A)
REFL(A, B) = A' = a c

(5) PERP(A, B) is the line through B that's perpendicular to AB

A = REFL(A, B)
PERP(A, B) = p = BIS(A, A)

(6) PGRM(A, B, C) is the point D such that ABCD is a
parallelogram

M = MID(A, C)
D = REFL(B, M)

(7) PARR(A, m) is the line through A that's parallel to m

B, C m
D = PGRM(C, B, A)
p = LINE(A, D)

(8) BIS(a, b) is the bisector of one of the angles formed between straight lines a, b.
A B

C
c d
C
B
a
b
M

A
B
a
c
A
A
B
A
p
B
A
C
M
D
A
B
C
a
p
D
135

C = a b
D a
c = CIRC(C, D)
E = b c
M = MID(D, E)
BIS(a, b) = m = LINE(C, M)

(9) SQU(A, B) is the pair of points C, D such that ABCD is a square.

p = PERP(A, B)
c = CIRC(B, A)
C = p c
D = PGRM(A, B, C)

(10) FOOT(A, m) is the foot of the perpendicular from A to m

B m
c = CIRCLE(A, B)
C = m c
FOOT(A, m) = F = MIDPT(B, C)

NOTE: This fails if B happens to be the foot of the perpendicular already for then there is
only one point of intersection of the circle with the line. In practice it is easy to choose a
suitable point. But he precise formulation that we have given there is no place for this. Is
there a modification to the above construction which will work in all circumstances? Yes,
there is.

9.3. Some More Advanced Constructions
Any circle that we construct in the course of a Ruler and Compass Construction will
already have its centre identified. But suppose we are given a circle at the outset. How do
we find its centre?
We simply take three distinct points P, P, P on the circle and join P to P and P to P.
The perpendicular bisectors of PP and PP intersect at the centre.

(11) CENTRE(c) is the centre of circle c

P, P', P" c
b = BIS(P, P')
b = BIS(P,P")
C = b' b"

a
b
C
D
E
M
m
A
B
p
c
C
D
A
F
c
C
m
B
P
P
P
b
b
C
136
If were given a circle and a point on it, how do we construct the tangent at that point?
We construct the centre, join the point to the centre and then draw the line perpendicular to
this line.

(12) TANG(c, P) is the tangent to circle c at P lying on the circle.

C = CENTRE(c)
t = PERP(C,P)

Now we could be given other curves at the outset, curves that cannot be constructed.
For example we might be given a parabola and be asked to construct its axis. Here we need
to have some knowledge of the geometry of the parabola such as is normally taught in high-
school coordinate geometry.
You may recall that the midpoints of parallel chords of a parabola lie on a line that is
parallel to the axis. This leads to the following construction.

(13) AXIS(p) is the axis of parabola p
A, B, C p
a = LINE(B, C)
b = PARR(A, a)
D = b p
E = MID(A, D)
F = MID(B, C)
c = LINE(F, G)
d = PERP(F, G)
e = PERP(G, F)
G, H = e p
M = MID(H, K)
S, T = d p
N = MID(S, T)
AXIS(p) = x = LINE(M, N)

(14) VERTEX(p) is the vertex of parabola p

x = AXIS(p)
V = x p

C
P
t
A
B
C
D
E
F
a
b
c
p
F
G
G
H
S
M
N
d
e
T
x
V
x
137
TANGENT (P, p) is the tangent to p at the point P.
V = VERTEX(p)
P p (ensure that P V)
h = AXIS(p)
b = PERP(P, h)
Q = a b
R = REFL(Q, P)
d = PERP(R, b)
S = d p
c = LINE(V, S)
t = PAR(P, c)
[For the parabols x
2
= 4ay, take P = (2at, at
2
) where t 0. The axis, h, is x = 0 and the line, b,
is y = at
2
. Q is (0, at
2
) and R is (4at, at
2
). The line d is x = 4at so S is (4at, 4at
2
). The slope
of c is t, which is the slope of the tangent.]

9.4. Constructible Numbers
We identify the complex number with its corresponding point on the complex plane.
A constructible number is a complex number whose corresponding point on the complex
plane can be constructed by ruler and compass, starting with two points representing 0 and 1.

Theorem 1: A complex number is constructible if and only if its real and imaginary
parts are constructible.
Proof: The real and imaginary axes are a = LINE(0, 1) and b = PERP(1, 0) respectively.
Let Z = X + iY. Then X = FOOT(Z, a) and iY = FOOT(Z, b).
Hence Y = CIRC(0, iY) a so if Z is constructible then so are Re(Z) and Im(Z).
Conversely suppose X and Y are constructible. Then iY = CIRC(0, Y) b and so Z =
PRGM(X, 0, iY).

Theorem 2: The set of constructible numbers is a field.
Proof: Firstly the set of all real constructible numbers is a field since:
X + Y = PGRM(X, 0, Y);
X = REFL(X, 0)
Let X = e
i
and Y = e
i
.
a = LINE(0, Y)
F = FOOT(X, a)
XY = REFL(X, F)
represents XY = e
i(+)
.

V
p
h
b
P
Q
R
c
S
t
d
Y X
Z = X + iY
i
1
0
iY
a
b
X
Y
F
XY
a
138
If Y is a complex number and X is real we can construct XY as follows:

a = LINE(0, 1)
b = PERP(1, 0)
c = CIRC(0, Y)
D = b c
E = PRGM(D, 0, 1)
d = LINE(0, E)
e = PERP(0, X)
F = d e
G = PRGM(0, X, F)
f = CIRC(0, G)
g = LINE(0, Y)
Z = f g

Multiplication of two complex numbers X and Y can be achieved by first multiplying X by
|Y| and then multiplying by
Y
|Y|
(except when Y = 0, but the constructability is obvious in this
case). Since each of these takes constructible numbers to constructible numbers the
constructability of XY follows. Of course we must also check that |Y| and
Y
|Y|
are
constructible if Y is, but this is easy.

1/x = LINE(0, PYTHAG(0, x, 1) PERP(0, 1)
We shall leave it as an exercise to show that if X is a non-zero constructible complex
number,
1
|X|
is constructible.
Corollary: Rational numbers are constructible.
D
a
b
c
1
E
d
X
e
F
G
f
g
Z
139
Theorem 3: The square roots of a constructible number are constructible.
Proof: The theorem is true for positive real constructible numbers as the following
construction shows (we leave it as an exercise to obtain an explicit construction).

1 + x
2

1 x
2

x

For a non-real constructible number, z = re
i
we construct r = CIRC(0, z) LINE(0, 1).
Then we construct r as above.
The square roots of z are BIS(LINE(0, 1), LINE(0, z) CIRCLE(0, r).

Example 1: 3 5
3 2
7
5
3 2
4
+
+
+
+
i is constructible (though it would probably be a
nightmare to obtain an explicit construction).

While there are infinitely many constructible numbers, of arbitrary complexity, most
complex numbers are not constructible. In a later chapter well prove that 2
3
and e
2i/9
are
not constructible, thereby proving the impossibility of doubling the cube and trisecting the
angle 2/3, by ruler and compass. Another famous non-constructible number is whose
non-constructability establishes the impossibility of squaring the circle.


Exercise 1: Describe a ruler and compass construction to draw the chord of contact of
tangents from the point P, lying outside the circle c.

Exercise 2: Find ruler and compass constructions for the inscribed circle for a triangle ABC.

Exercise 3: Find a ruler and compass construction to construct the common tangents (where
they exist) to two given circles.

Exercise 4: Find an explicit ruler and compass construction for finding the square root of an
arbitrary positive real number x.

Exercise 5: Show that cos(2/5) is constructible.
HINT: Use De Moivres Theorem to find a polynomial with integer coefficients having
cos(2/5) as a root and hence express cos(2/5) as a surd expression involving square roots.

140

Exercise 1:
The tangents will be perpendicular to the radius. Since the angle in a semicircle is a
right angle, we can construct a circle whose diameter is the line segment joining P to the
centre of the circle. Where this cuts c are the points of contact.

C = CENTRE(c)
M = MID(P, C)
d = CIRC(M, P)
T, T = d c
t = LINE(P, T)
t = LINE(P, T)

Exercise 2:
The centre of the inscribed circle is the common intersection of the perpendicular bisectors of
the sides.
p = BIS(C, A, B)
q = BIS(A, B, C)
C = p q
F = FOOT(C, A, B)
c = CIRC(C, F)

Exercise 3: Construct a right-angled triangle whose hypotenuse is the interval joining the
centres and which has, as one of its sides, the difference between the radii.
A
B
C
F
c
A
q
C
p
141
Exercise 4:
Since x =
x + 2
2
2

x 2
2
2
we construct a right-angled triangle with hypotenuse of length
x + 1
2
and another side of length
x 1
2
.

Let O, A, B represent 0,
x 1
2
and
x + 1
2
respectively.
M = MID(O, B)
c = CIRC(M, O)
d = CIRC(O, A)
P, P = c d
Then PB = PB have length x.

Exercise 5: If c = cos(2/5) and s = sin(2/5) then by De Moivres Theorem,
(c + is)
5
= cos(2) + isin(2 = 1.
Hence c
5
+ 5isc
4
10s
2
c
3
10is
3
c
2
+ 5s
4
c + is
5
.
Equating imaginary parts, and dividing by s we get 5c
4
10s
2
c
2
+ s
4
= 0.
So 5c
4
10(1 c
2
)c
2
+ (1 c
2
)
2
= 0.
Hence 16c
4
12c
2
+ 1 = 0.
So c
2
=
12 80
32
so c =
12 80
32
. By eliminating three of these four alternatives we
see that in fact c =
12 80
32
. This is clearly constructible by ruler and compass.
O
A B
P
c
d
P
M
142

143
10. IMPOSSIBLE CONSTRUCTIONS

10.1. Number Fields and Field Extensions
One of the classic non-constructability problems is that of trisecting an angle. Ruler
and compass constructions exist for certain angles, such as 60, and constructions exist that
approximately trisect any angle, but there are none that exactly trisect any angle. This is not
an open question, awaiting for some brilliant mathematician to solve it. It is a logical
impossibility. One can be as sure that no such construction will be produced as one can be
sure that no-one will ever find a solution to the equation 0.x = 1.
Well demonstrate the impossibility of trisecting an angle by translating it into an
algebraic problem, using the concept of a number field.

Definition: A number field is any subfield of the complex numbers.

Example 1: Q, R and C are number fields but Z isnt a field. The field Z
p
, that is the
integers modulo the prime p, form a field but as Z
p
is not a subfield of C (the operations are
different) it isnt a number field.
To prove that a given set, F, of complex numbers is a number field its sufficient to
check that 0, 1 F and that x + y, x and x
1
(if x 0) all belong to F whenever x, y F.
This is because all other field axioms (associative, commutative and distributive laws) hold
automatically for subfields.

Example 2: {a + b2 | a, b Q} is a number field.
The only axiom thats not immediately obvious is the existence of multiplicative
inverses. Suppose a + b2 0. Then a b2 0 and so:
1
a + b2
=
a b2
(a + b2)(a b2)
=
a b2
a
2
2b
2
=
a
a
2
2b
2
+
b
a
2
2b
2
2 .

If F, K are number fields with F K, we say that K is an extension of F and we
denote the extension by the symbol K:F.
A simple extension of a field F is a field F[] where this denotes the smallest number
field containing F (as a subfield) and (as an element). Such a smallest number field will
always exist because it will be the intersection of all number fields containing F and (and
the intersection of any collection of fields is itself a field).
Suppose we start with a number field F. Then F[] would have to contain such
numbers as 1 + and
2
and, in fact, all polynomial expressions in . In addition it must
contain
2
1 +
and, more generally, all expressions that have the form
f()
g()
where f() and
g() are polynomial expressions in with g() 0. But some of these may be equal to
others.

Example 3: Q[2] = {a + b2 | a, b Q}. We showed above that K = {a + b2 | a, b Q}
is a field. Clearly any number field that contains Q and 2 must contain every element of K.
So K must be the smallest number field containing Q and 2.

Example 4: R[i] = C since a field that contains every real number as well as the complex
number i must contain every complex number.
144

Example 5: R[]= R. More generally F[] = F whenever F.

Theorem 1: If = a + b for some a, b F, with b 0, then F[ ] = F[ ].
Proof: F[] contains and F so F[] F[]. But = (a/b) + (1/b), so F[] contains F and
and hence F[] F[]. Thus F[] = F[].

Example 6: Q[e
2i/3
] = Q[3i] since e
2i/3
=
1
2
+
3
2
.

10.2. Fields as Vector Spaces
Were attempting to solve the geometric problem of angle trisection using the
algebraic theory of fields. But theres a slightly more basic structure thats very useful here,
and one that we know a lot more about vector spaces. Every field extension can be viewed
as a vector space.
In fact the field extension K:F (this just means one field F contained inside another
field K) can be viewed as a vector space over F. The vectors are the elements of K and the
scalars are the elements of F. Never mind that K contains F so that some of the vectors are
also scalars. Theres nothing in the vector space axioms which prevents this. Of course well
have to abandon the convention of writing vectors with a ~ underneath or printing them in
bold type.
To be a vector space we need to be able to add two vectors and to multiply a vector by
a scalar. This we can do because all vectors and scalars live inside the field K. In fact we
can even multiply two vectors, something thats not normally possible in a vector space.
We also need to check out the many vector space axioms. But these will just be the
field axioms. For example the axioms (u + v) = u + v and ( + )v = v + v are just two
instances of the distributive law.
If V is a finite-dimensional vector space over the field F we denote its dimension by
the symbol |V:F|. If K is a field extension of F the degree of the extension is simply the
dimension |K:F| of this vector space. If K = F then |K:F| = 1. It can in fact be infinite, but
well be mainly interested in the case of finite-dimensional extensions.
The degree of a complex number over a number field F is defined to be the degree
of the corresponding simple extension, that is, |F[]:F|.

Example 7: Find the degree of 2 over Q.
Solution: We must find a basis for Q[2] over Q. Weve shown that every element of Q[2]
has the form a + b2 for a, b Q. Writing this as a.1 + b.2 we can view this as a linear
combination of the vectors 1 and 2, with the scalar coefficients being the rational
numbers a, b. So 1 and 2 span Q[2] over Q. But are they linearly independent? Suppose
a + b2 = 0 for rational a, b. If b 0 this gives 2 =
a
b
which is impossible since 2 is
irrational. So b = 0. But then this forces a = 0. So 1, 2 are indeed linearly independent.
and so they form a basis for this field extension. The fact that the basis consists of two
elements shows that the degree of the extension is 2.

Example 8: What is the degree of n over Q (where n is an integer)?
Solution: If n is irrational (as is the case for n = 2, n = 3, n = 5 etc) then Q[n] has degree 2
over Q. But if n Q (eg if n = 4) then Q[n] = Q and so n has dimension 1 over Q.
145

Theorem 2: If is a zero of some quadratic equation ax
2
+ bx + c, where a, b, c F,
and F, then has degree 2 over F.
Proof: =
b b
2
4ac
2a
. If = b
2
4ac then F, but F (otherwise F). A
basis for F[] = F[] over F is 1, and so |F[]:F| = 2.

Example 9: What is the degree of i over C and what is its degree over R?
Solution: C[i] = C so i has degree 1 over C. R[i] = C which has degree 2 over R and so
i has degree 2 over R. (It can be shown that i has infinite degree over Q.)

10.3. Dimensions of Field Extensions
Theorem 3: Suppose V is a finite-dimensional vector space over the field K, which in
turn is a finite-dimensional extension of the field F. Then V can be viewed as a vector
space over F and |V:F| = |V:K| |K:F|.
Proof: Let |V:K| = n and |K:F| = m. Let
1
,
2
, ... ,
n
be a basis for V as a vector space over
K and let
1
,
2
, ... ,
m
be a basis for K as a vector space over F. Well show that the mn
products
i
j
form a basis for V as a vector space over F. The theorem then follows.
The
i
j
span V over F. For, let v V. Then v = k
1
1
+ ... + k
n
n
for some
scalars in K. But each of these is a vector in K, regarded as a vector space over F.
Hence each k
i
can be expressed in the form k
i
= h
i1
1
+ ... + h
im
m
where each h
ij
F.
Substituting into the previous equation we obtain v h
ij i j
i j
=

,
showing that the
i
j
span V
over F.
On the other hand the
i
j
are linearly independent over F. Suppose h
ij i j
i j

,
= 0
for h
ij
s F. Then h
ij j
j i
i

= 0 and since each h

ij j
j
K and
1
, ... ,
n
is a basis
for V over K, each h
ij j
j
= 0. Now each h
ij
F and
1
, ... ,
m
is a basis for V over F
and so each h
ij
= 0. Thus the set of mn products
i
j
is linearly independent, and is therefore
a basis for V over F.

Theorem 4: If a point ( , ) is constructible by ruler and compass with rational
coordinates then the degrees of and over Q are powers of 2.
Proof: Suppose that at some stage in the ruler and compass construction the coordinates of
all points generate some number field F. Then any point (, ) that can be constructed from
these points in one step is a point of intersection of two curves of the form
ax
2
+ ay
2
+ 2fx + 2gy + c = 0.
(For a 0 this represents a circle with centre (f/a, g/a) with radius
f
2
a
2 +
g
2
a
2 c .
For a = 0 it represents a straight line.)
The coefficients of the equations are expressible in terms of the coordinates of the
points from which the circles/lines were constructed using only the operations of addition,
subtraction, multiplication and division and so they belong to F.
146
Eliminating y from these two equations we find that is a zero of some quadratic
(or perhaps linear equation) with coefficients in F. It follows that the degree of the minimum
polynomial of over F is 1 or 2. Hence |F[]:F| = 1 or 2. Similarly |F[]:F| = 1 or 2.
As the ruler and compass construction proceeds we build up a sequence of fields, each
having degree 2 over the previous one. By Theorem 2, the degree of each of these fields over
Q must be a power of 2. If is now a coordinate of any point that is constructible by ruler
and compass (starting with points with rational coordinates) then K for some number
field with |K:Q| = 2
n
for some n. By Theorem 2, |F:Q[]|.|Q[]| = |F:Q| = 2
n
and so
|Q[]:Q| is a power of 2.

To show that an angle of 20 cant be constructed by ruler and compass we simply
need to show that |Q[cos(2/9)]:Q| is not a power of 2. But how do we compute this degree?

We use the trigonometric identity:
cos(3) = 4cos
3
3cos.
If = 2/9 (corresponding to the 20 angle that would result from a trisection of 60) then
3 = 2/3 and cos(3) = . So, if = cos(2/9) then 4
3
3 = and so is a zero of
the polynomial 8x
3
6x + 1.
This polynomial has in fact three zeros: cos(2/9), cos(8/9) and cos(14/9). If any
of these are rational then the cubic would factorise over Q.
Using a calculator we find that:
cos(2/9) 0.7660444431, cos(8/9) 0.9396926208 and cos(14/9) 0.1736481777.
These dont look rational, but thats not very convincing. In actual fact these approximate
values are rational (because they have finitely many decimal places) but then theyre only
very good approximations and not the exact values. What then?
Suppose a/b (where a, b are coprime integers and a > 0) is a zero of this cubic. Then:
8
a
3
b
3 6
a
b
+ 1 = 0 and so 8a
3
6ab
2
+ b
3
= 0.
Suppose r 1 and let p be a prime divisor of a. Then p divides b
3
and hence divides b.
This contradicts the coprimeness of a, b. Hence a = 1.
So if f(x) = x
3
6x
2
+ 8 = 0, f(b) = 0. But f(2) = 24, f(1) = 1 and so there is a
root between 2 and 1. Similarly theres a root between 1 and 2 and between 5 and 6.
Being a cubic there can be no other roots and hence f(x) has no integer zeros.
Thus 8x
3
6x + 1 has no rational zeros and hence it cant factorise over Q. It must therefore
be prime over Q.

Theorem 5: If p(x) = px
3
+ qx
2
+ rx + s is a prime cubic over Q and p( ) = 0 then
|Q[ ]:Q| = 3.
Proof: Let K = {a
2
+ b + c | a, b, c Q}. By the closure properties K must be a subset of
Q[]. In fact K is a field. Its clearly closed under addition and subtraction.
The fact that p
3
+ q
2
+ r + s = 0 means that
3
can be expressed as a rational linear
combination of 1, and
2
. So, multiplying two elements of K we get an expression
involving powers of up to
4
, but then
3
, and hence
4
, can be expressed in terms of 1,
and
2
. The only thing left to check is closure under inverses.
Is
1
a
2
+ b + c
K if the denominator is non-zero?
Suppose the denominator is non-zero and let h(x) be the polynomial ax
2
+ bx + c.
If p(x) divides h(x) then the denominator is h() = 0.
147
So p(x) doesnt divide h(x) and hence theyre coprime. It follows that there exist
polynomials e(x), f(x) Q[x] such that
e(x)p(x) + f(x)h(x) = 1.
Now substituting x = we get e()p() + f()h() = f()h() = 1.
Hence
1
h()
= f() K.
This means that Q[] = {a
2
+ b + c | a, b, c Q} and so 1, and
2
span Q[].
Are they linearly independent?
Suppose there exist rational numbers a, b and c such that a
2
+ b + c = 0.
Let u(x) be the polynomial ax
2
+ bx + c. Dividing p(x) by the quadratic u(x) we get
p(x) = u(x)q(x) + (rx + s), where rx + s is the remainder.
Substituting x = we get r + s = 0. If r 0 we get = s/r Q, a contradiction.
Hence r = 0 and so s = 0. But then p(x) = u(x)q(x), contradicting the assumption that p(x) is
prime. It follows that {1, ,
2
} is a basis for Q[] over Q and so has degree 3 over Q.

The theorem can be easily generalised to any field, and any prime polynomial.

Theorem 6: If is a zero of a prime polynomial of degree n over a field F then
|F[ ]:F| = n and 1, ,
2
, ,
n 1
is a basis.

To wrap things up, its impossible to have a ruler and compass construction for
trisecting any given angle, because it would have to be able to trisect 60 in particular. But
this would involve constructing a point having cos(2/9) as its x-coordinate. Since ruler and
compass constructions involve intersecting lines with lines and lines with circles and circles,
the coordinates of newly constructed points would have degree 1 or 2 over the field generated
by the coordinates of the existing points. No matter how many steps were involved in the
construction the degree of the coordinates of any point so constructed, over Q, would have to
be a power of 2. But the degree of cos(2/9) is 3, which is not a power of 2. Therefore no
such trisection construction is possible.


Exercise 1: Determine which of the following are TRUE and which are FALSE
(1) Every angle can be trisected by ruler and compass.
(2) A regular polygon with 18 sides can be constructed by ruler and compass.
(3) {a + b 2
3
| a, b Q} is a number field.
(4) Q[8] = Q[2].
(5) If F is a number field then so is {f() | f(x) F[x]}.
(6) 1, ,
2
are linearly independent over Q.
(7) Q[1 + 2
3
]= Q[ 4
3
].
(8) If H, K are number fields then H K a number field.
(9) {a + b 2
3
| a, b Q} is a vector space over Q.
(10) |Q[e
2i/11
]:Q| = 11.
Exercise 2: Prove that there is no number field F such that R < F < C.

Exercise 3: Prove that all the zeros of x
8
+ x
4
+ 1 are constructible.
[HINT: x
8
+ x
4
+ 1 is a quadratic in x
4
.]
148

Exercise 4: Prove that { a b c d + + + 2 4 8
4 4 4
| a, b, c, d Q} is a number field.
[HINT: Of course you could check it directly, but would Theorem 6 help?]

Exercise 5: Prove that the volume of a cube cannot be doubled by a ruler and compass
construction. That is, show that 2
3
is not constructible.

Exercise 6:
Show that the following construction, attributed to Archimedes, will exactly trisect any given
angle and explain why this doesnt contradict the theorem that angles cant be trisected by
ruler and compass.

To trisect the angle AOB, with A, B on a circle of radius r with centre O, construct C so
that COB is a diameter of the circle and construct L on this diameter so that LC = r.
Having a straight edge lying along the diameter COB, with points L and C marked on them,
move it so that C slides around the circle, and L slides along the diameter. Being a rigid
straight edge, the distance between L and C will always be equal to r. Continue sliding
until the straight edge passes through A and let L, C be the respective positions of L and C
when this occurs. Then ALO will be exactly one third of AOB.

Exercise 7: Prove that the regular pentagon can be constructed by ruler and compass.
[HINT: Find a polynomial with integer coefficients that has cos(2/5) as a root,]

Exercise 8: Prove that the regular heptagon (seven equal sides) is not constructible by ruler
and compass.
B
O
A
C
L
3
C
L
149

Exercise 1:
(1) FALSE; (2) FALSE That would make 20 constructible; (3) FALSE 4
1/3
is not in the set;
(4) TRUE; (5) FALSE Only true if is algebraic; (6) FALSE 1 + +
2
= 0; (7) TRUE;
(8) TRUE; (9) TRUE; (10) FALSE (the minimum polynomial has degree 10 since x 1 is a
factor of x
11
1).

Exercise 2: |C: R| = 2. If R < F < C, |C:R| = |C:F|.|F:R| but each of these factors is at least 2,
a contradiction.
Exercise 3: If is a zero of x
8
+ x
4
+ 1 then
4
=
1 3
2
. So |Q[
4
]:Q| = 2.
Clearly |Q[]:Q[
2
]| = |Q[
2
]:Q[
4
]| = 2 so |Q[]:Q[
4
]| = 4 and Q[]:Q| = 8. Since this is a
power of 2, is contructible..

Exercise 4: Let =
4
2 . Then x
4
2 = (x
2
2)(x
2
+ 2) = (x )(x + )(x i)(x + i).
None of these factors has rational coefficients, and no pair of factors can multiply to give a
polynomial with rational coefficients. So x
4
2 is prime over Q. Hence, by Theorem 6,
{1, ,
2
,
3
} is a basis for Q[] over Q. Hence{ a b c d + + + 2 4 8
4 4 4
| a, b, c, d Q} is a
field.

Exercise 5: Let = 2
3
. The roots of x
3
2 are , ,
2
where = e
2i/3
. Since none
of these is rational (wed prove that 2
3
is irrational in a similar way to 2) this polynomial
has no rational roots and hence no linear factors and hence is prime over Q. Therefore, by
Theorem 5, |Q[]:Q| = 3. Since this is not a power of 2, 2
3
is not constructible.

Exercise 6:

If CLO = then COL = as CLO is isosceles. Hence ACO = 2, being the exterior
angle of OCL. Since OAC is isosceles, CAO = 2 and so AOB = 3 being the
exterior angle of OAL.
Although this method exactly trisects a given angle, it is not a ruler and compass
construction. The sliding of L and C until the line passes through A is not a permissible
operation according to the definition of ruler and compass constructions.
B
O
A

2
3
2
C
L
150
Exercise 7: Let c = cos
2
5
and let s = sin
2
5
. Then (c + is)
5
= cos(2) + isin(2) = 1.
Equating imaginary parts we have 5c
4
s 10c
2
s
3
+ s
5
= 0. Dividing by s (clearly non-zero)
and writing s
2
= 1 c
2
we get 5c
4
10c
2
(1 c
2
) + (1 c
2
)
2
= 0, that is, 16c
4
12c
2
+ 1 = 0.
Considering this as a quadratic in c
2
we get c
2
=
12 80
32
, whence c =
12 80
32
. With
a calculator we can check that, of the four alternatives, c =
12 80
32
. Hence c is
constructible.

Exercise 8:
Let c = cos
2
7
and let s = sin
2
7
. Then (c + is)
7
= cos(2) + isin(2) = 1.
Equating imaginary parts we have 7c
6
s 35c
4
s
3
+ 21c
2
s
5
s
6
= 0. Dividing by s (clearly
non-zero) and writing s
2
= 1 c
2
we get 7c
6
35c
4
(1 c
2
) + 21c
2
(1 c
2
)
2
(1 c
2
)
3
= 0, that
is,
64c
6
80c
4
+ 24c
2
1 = 0. Let x = c
2
. Then 64x
3
80x
2
+ 24x 1 = 0. If the regular
heptagon is constructible then so is c and hence x is constructible.
Let
a
b
be a rational root of this cubic (with a, b coprime and a > 0).
Then 64
a
3
b
3 80
a
2
b
2 + 24
a
b
1 = 0 and so 64a
3
80a
2
b + 24ab
2
b
3
= 0. If p is any
prime divisor of a then p divides b
3
and hence divides b. If a > 1 this contradicts the
coprimeness of a, b and hence a = 1.
So if g(x) = x
3
24x
2
+ 80x 64 then g(b) = 0. But g(1) = 7, g(2) = 8 and
g(3) = 13 so there are zeros between 1 and 2 and between 2 and 3. Clearly there must be a
third zero beyond 3, since ultimately g(x) becomes positive again. With a bit of persistence
we find that it lies between 20 and 21.
So g(x) has no integer zeros and hence f(x) = 64x
3
80x
2
+ 24x 1 has no rational zeros. It
must therefore be prime over Q. So the degree of cos(2/7) over Q is 3 and hence it is not
constructible.

Projective Geometry and Symmetry Notes

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Projective Geometry and Symmetry Notes

Uploaded by

Copyright:

Available Formats

GEOMETRY

= {v V | u.v = 0 for all u U}.

is also a subspace of V and its easy to show that every

. One can also easily

= U (the orthogonal complement of the

if and only if it is orthogonal to both vectors. Such

= ((7, 10, 9))

. Since (3, 1, 4).(5, 2, 3) = 15 2 12 = 1 0, R does not

and CD = ((2, 2, 1))

. If M = (m) then m is orthogonal

and N = ((7, 245, 85)). Then MN = ((1755, 5682, 16233)). Now AD

and BC = ((3, 3, 7))

so AD BC = ((4155, 2424, 7329)). It would

is the set of all vectors that are orthogonal to every vector in U.

is always the total vector space, which in this case is R

is a 1-1 correspondence between the projective

. Geometrically you can think of

as the plane through the

. But both of these

where p is a non-zero vector orthogonal to both

and k = ((5, 1, 9))

= ((16, 18, 13))

. Since (3, 7, 6).(16, 18, 13) =

= ((52, 13, 13)) = ((4, 1, 1)).

and AB = ((8, 7, 1))

. So AB AB = ((52, 91, 221)) =

and BC = ((6, 1, 5))

so BC BC = ((28, 168, 0)) =

= ((102, 17, 17))

and AC = ((14, 9, 11))

so AC AC = ((135, 351, 459))

. Since (5, 13, 17).(6, 1, 1) = 0 AC AC lies on h.

is a projective line (2-

is a projective point. The

) establishes a 1-1 correspondence between the

. Then P lying on h is equivalent to p and v being

is also equivalent to p and v

and CD = (5, 0, 1) (1, 1, 1)

and hence (A, C; B, D) = 1

and hence (A, C; B, E) = 1

= (90, 54, 72) = (5, 3, 4).

. To complete the picture we need to consider

is invertible we can then write

is a proper rotation, that is not through 0, its eigenvalues are non-real.

| 0. Thus it is indeed the case that I R

was invertible. But alarm bells should be

is invertible. Remember that the eigenvalues of a

v + b is a reflection in an axis at an angle /2 to the positive x-

= T | generated by a translation to the right by one

= G | generated by the glide described above. Every translation in the symmetry

= 0 and since each h

You might also like