Vous êtes sur la page 1sur 26

Chapter 4

Euclidean geometry
Geometry deals with the properties of figures in space. Etymologically, geometry
means the practical science of measurement. Its roots can be traced to the ancient
Egypt, where the study of geometry was initiated in the need of obtaining reliable
methods for measuring areas. Pythagoras and other Greek scholars expanded the
concept of geometry to include and abstract positions, lines, angles and incidence
properties. Geometry, as it stood more than 2000 years ago, was presented in
terms of logical deductions from few definitions and assumptions in one of the
most widely read book in the world, namely Euclids Elements. Due to their
historic significance, these postulates are presented in Appendix B. Euclids
fifth postulate, sometimes called parallel postulate (which stands in the form
presented by Legendre as the sum of the angles in triangle is equal to two right
angles) is historically very significant. It led mathematicians to question the
traditional foundations of geometry, and led them circa 1830 to realize that there
are different kinds of geometries.
Due to the explosion on computer graphics, geometric methods are used now
more than ever before. In this chapter, we are concerned mostly with analytic
and projective geometry, which are needed in the practical applications.
Analytic geometry is concerned with the representation of geometric objects
by algebraic equations. Traditionally, the expression analytic geometry is a synonym for coordinate geometry, where geometrical figure is represented by a set
of scalars, i.e. coordinates. However, geometric algebra has offered us efficient
tools to describe geometric figures by their intrinsic properties alone.
115

116

CHAPTER 4. EUCLIDEAN GEOMETRY

Projective geometry deals with properties and invariants of geometric figures


under projection. It relates figures to their images in lower dimensional space.
Geometry can also be formulated projectively by adding extra dimensions to
the space under study. This often simplifies the necessary mathematics. For
example, all the conformal transformations can be linearized by adding two extra
dimensions (which produces the conformal model for geometry).

4.1

Geometry as set of transformations

Any two equivalent figures are related by transformations which preserve their
geometric properties. In some sense, as done by F. Klein in his Erlager program,
geometry can be defined as set of transformations of space. The transformations
included to the geometry in question specify the isometries, or properties which
are common to all equivalent figures. Let us denote the geometric equality by .
Generally, independently of the nature of the geometry
1. A figure F should be equal to itself, i.e. F F for all F .
2. If F equals F 0 , then F 0 should equal F , i.e. F F 0 F 0 F .
3. If F equals F 0 , and F 0 equals F 00 , then F should be equal to F 00 , i.e. F
F 0 &F 0 F 00 F 0 F 001 .
The Euclidean geometry is specified by including only the transformations
which preserve distances to the set of similarity transformations.

4.1.1

Discretenes vs. continuity

We have not yet specifified the nature of the objects which are related through
the isometries. Conventionally, it is thought that the basic object subjected to
1

An amusing exception to transitivity is the probability for which a dice selected from set

of very particular dices A={4,4,4,4,0,0}, B={3,3,3,3,3,3}, C={2,2,2,2,6,6} and D={5,5,5,1,1,1}


wins other dice in the set by the propability 2/3. Although A wins B, B wins C, and C wins D
by the propability 2/3, D still wins A by the same propability! This and other similar examples
are presented in the excellent book Innumeracy by J. A. Paulos (Hill & Wang, New York,
1988).

4.1. GEOMETRY AS SET OF TRANSFORMATIONS

117

the isometry transfornations is a point. Unfortunately, the concept of point has


turned out notoriously difficult to define. In the Weissteins Concise encyclopedia
of mathematics a point is defined
A zero-dimensional mathematical object, which can be specified
in N -dimensional space using N coordinates.
This definition is essentially equivalent to that given in Ref. [48], namely
Point: an object of no spatial extent (zero-dimensional) located at
a single position in space.
In essence, according to the definitions above, a point specifies a location
in space. In the past century it has become the prevailing practice among the
majority of set-theoretically oriented mathematicians to define continuous objects
as a collection of points. The continuity is not regarded as a property of a
collection per se, but it is derivable from certain additional structures (order
theoretical, topological, analytical etc.) imposed on it. For example, a geometric
straight line is identified with the real number line: for each point on the line
there correspond a unique real number (a line is a tray of points). Similarly,
any geometric object in N -dimensional space is regarded as a collection of sets
of N real numbers.
Unfortunately, this approach does not properly distinguish discreteness and
continuity. Most of the classical geometric objects are continuous by nature,
that is, they can be divided indefinitely without altering their essential nature.
Such an object is a segment of a straight line. If it is divided, the result is two
straight line segments (and so on). Consequently, the point based set-theoretical
approach often produces counterintuitive if not inherently contradictory results.
Consider again the straight line. One could ask why rational numbers do not
suffice to label every point at the line, since there are no holes in the rational
number system (that is, for each rational number there are bigger and smaller
rational number which are so close that their separation approaches zero). Yet, as
it can be show, there are irrational numbers between any two rational numbers2 .
2

There are many other counterintuitive implications in the set-theoretical real number ap-

proach to geometry. For example

118

CHAPTER 4. EUCLIDEAN GEOMETRY

In fact, irrational numbers originate from the proof given by the ancient

Greeks that the length of a diagonal of a unit square ( 2) is not a rational


number. This proof is by its very nature geometric, not set-theoretical. Only
after this observation it became necessary to include irrational numbers to label
positions in the geometric line. To make things still worse, we know that there
exists scalars between any two real numbers (although there is no empty space
between adjacent real numbers!), such as hyperreal and surreal numbers. From
a purely set theoretical point of view it is not obvious why one should prefer the
real line over, say, the hyperreal line as the representation of a geometric straight
line, if geometry is regarded as set of transformations relating points.
We can remedy many (if not all) paradoxes by defining the continuum as a
collection of locations, which can not be distinguished from each others. Nilpotent
infinitesimals are archetypes of arithmetic objects possessing this property (see
Appendix A.1). In this approach, continuum is not regarded as a collection of
points, but points locate the boundaries of continuum.
In what follows, we use the word point to strictly to stand for a location
in space. If we talk points in general, we denote them simply as p1 , p2 , ... . A
priviledged reference point, the origin is then referred as p0 .

4.2

Analytic geometry

Analytic geometry is concerned with the representation of internal properties of


geometric objects by algebraic equations of points. For the present purposes we
identify points as vectors
p = p0 + x

(4.1)

1. The number of points is the same in any two segments of real line, no matter how long
the segments are (Bolzano).
2. There are as many points in any line as there are points in any plane or more generally
in any N -dimensional solid (Cantor).
3. Any solid sphere can be decomposed into finitely many pieces which can themselves be
reassembled to form of two solid spheres of the same size as the original (Banach-Tarski
paradox).

4.2. ANALYTIC GEOMETRY

119

Figure 4.1: Points p1 and p2 .


which are bound to the specified reference point p0 (the origin). Furthermore the
distance of the point p from the origin is now given as
|p| = |x|

(4.2)

where it follows that p2 = (p0 + x)2 = x2 or p20 + p0 x + xp0 = 0, which in turn


implies that
p20 = 0

(4.3)

p0 x = xp0

(4.4)

and

for any vector x. In other words, p0 is a null vector, which is orthogonal to any
vector in E(N ). In effect, the inclusion of p0 expands the dimension of space to
N + 1.
However, the internal properties of any geometric figure depend, not on specific points, but their difference. Since the difference pj pi isa free vector xj xi ,
we can be formulate analytic geometry in terms of free vector variable x, and we
need not concern us with the distinction between a point p and the vector p used
to label it.

120

4.2.1

CHAPTER 4. EUCLIDEAN GEOMETRY

Lines

The point p = p0 + x in the line through points pi and pj can be determined


through an equation
(p pi ) xij = 0

(4.5)

where xij = xj xi is the direction vector of the line. This equation can be
further written as
x xij = M

(4.6)

M = xij xj = xi xj

(4.7)

where

is the moment of line, which in turn can be rewritten as a rectangle M = dxij =


dxij spanned by xij and the perpendicular support vector
d = xi xj x1
ij

(4.8)

which intersects the line at a right angle and whose length d is equal to the
1
distance of the line to the origin 0. Because x xij bx1
ij = x x xij xij , the

point in the line can be written in a parametric form as


x = (M + ) x1
ij

(4.9)

Division and intersections


Because all chords of a line are parallel, we can write
a x = (x b)

(4.10)

where the vectors {a, b, x} label three points pi = p0 + a, pj = p0 + b and


p = p0 + x on the line and the division ratio is a scalar. By solving for , we
obtain

ax
(4.11)
xb
or, by first taking the outer product of both sides of Eq. (4.10) with x, and then
=

solving for ,
=

ax
xb

(4.12)

4.2. ANALYTIC GEOMETRY

Figure 4.2: Moment of line.

Figure 4.3: Division ratio.

121

122

CHAPTER 4. EUCLIDEAN GEOMETRY

Figure 4.4: Division ratio when x and c are collinear.


The division ratio can be used as a coordinate by solving Eq. (4.10) for x,
which produces
a + b
(4.13)
1+
Because both A and B are proportional to the unit bivector i of the plane deterx=

mined by the points {p0 , pi , pj }, the division ratio can be written as = B/A,
and consequently Eq. (4.13) can be rewritten as
p = p0 +

Aa + Bb
A+B

(4.14)

where A = iA and B = iB are the line coordinates for the point p on the
line going through the points pi and pj . The division ratios are independent of
the location of origin. We can displace origin p0 to some point pk = p0 + c, and
obtain the division ratio as
=

B
B0
B B0
= 0 =
A
A
A A0

(4.15)

where A0 and B 0 are the line coordinates when the origin lies at c, see Fig. 4.3.
Now, the directions of the bivectors are implicated by the curved arrows in the
figure.
It is worth emphasizing that Eq. (4.11) is valid even when the origin p0 is
in the line and the ratio B/A is consequently B/A = 0/0. For example, by

4.2. ANALYTIC GEOMETRY

123

Figure 4.5: Cevas theorem.


comparing Fig. 4.4 to Fig. 4.5, we can extract the Cevas theorem, which states
that the division ratios for a triangle satisfy




a c0
b a0
c b0
=1
c0 b
a0 c
b0 a

(4.16)

where a, b and c label the vertices of the triangle, and a0 , b0 and c0 label the
points, lines drawn from the vertex and passing through a point s divide the
opposing side. To see why Eq. (4.16) is valid, note that
A1
C1 + C2
=
A2
B1 + B2
B1
A1 + A 2
=
B2
C1 + C2
C1
B1 + B2
=
C2
B1 + B2

4.2.2

(4.17)
(4.18)
(4.19)

Planes

The plane with the direction given by the bivector U passing through a point
q = p0 + a is determined by the equation
(x a) U = 0

(4.20)

124

CHAPTER 4. EUCLIDEAN GEOMETRY

where the vector x labels the point p = p0 + x on the plane. Parallel planes are
obtained as the solution set to the equation
xU=T

(4.21)

where the trivector T is the moment of the plane. The directance d from the
origin p0 to the plane is given by
d = TU1 = TbU1

(4.22)

The distance from the origin p0 to the plane is given by |d|.

4.2.3

Conic sections

The conic sections are intersections of a cone with a plane. Alternatively, they
can be defined as a set of all points p = f + x in a plane with the property
that the distance of each point from a fixed point, the focus f , is in fixed ratio
(the eccentricity) to the distance of that point from a fixed line, the directrix. In
vector form, this condition reads as
x
=e
d x ud

(4.23)

where d is the perpendicular support vector (the directance) from f to the directrix, d = |d| and e is the eccentricity parameter of the conic (see Fig. NN).
[FIGURE HERE!] Eq. (4.23) is often represented in an alternative form as
x=

l
1 + e ux

(4.24)

where l = ed is the semi-latus rectum and e = eud is the eccentricity vector .


Eq. (4.24) determines a curve, a conic, when x lies in a plane. If x is allowed to
range over all directions in E(3), then Eq. (4.24) determines a two-dimensional
surface, a conicoid. In the latter case, the directrix is a fixed plane instead of a
line. The range of the eccentricity parameter determines the type of the conic
section. If e > 1, the conic (conicoid) is hyperbola (hyperboloid). If e = 1, the
conic (conicoid) is parabola (paraboloid). If 0 < e < 1, the conic (conicoid) is
ellipse (ellipsoid). If e = 0, the conic (conicoid) is circle (sphere).

4.2. ANALYTIC GEOMETRY

4.2.4

125

Barycentric coordinates

In the three-dimensional space, we may choose 4 reference points {p0 , p1 , p2 , p3 },


and express any point p = p0 + x as a unique linear combination
p = p0 + t(1) a1 + t(2) a2 + t(3) a3

(4.25)

ai = pi p0

(4.26)

where

are the edges, and the points {p0 , p1 , p2 , p3 } are the vertices of a tetrahedron,
which can be characterized by the trivector a1 a2 a3 . The simplicial coordinates
can be obtained as
t(i) = x a(i)

(4.27)

where

aj ak
(4.28)
a1 a2 a3
is the ith edge of the reciprocal simplex a(1) a(2) a(3) , and the indices i, j, k
a(i) =

are in the cyclic order. If the point p lies inside the tetrahedron, the coordinates
t(i) are in the range 0 t(i) 1, and we may write
p = t(0) p0 + t(1) p1 + t(2) p2 + t(3) p3

(4.29)

where t(0) = 1 t(1) t(2) t(3) .


If we label the vertices {p0 , p1 , p2 , p3 } by the vectors {b0 , b1 , b2 , b3 }, we can
represent a vector x as unique linear combination

x = t(0) 1 b0 + t(1) b1 + t(2) b2 + t(3) b3

(4.30)

of the four vectors bi . The coordinates {t(0) 1, t(1) , t(2) , t(3) } are often called as
the tetrahedral coordinates (or, as Hamilton refers them, anharmonic coordinates). As the reader no doubt notices, the value of the coordinate t(i) depends
only on the difference bi b0 .
Barycentric coordinates generalize readily to N -dimensional space. Now, a
point p reads as
p = p0 +

N
X

t(i) ai

(4.31)

i=1

where t(i) is obtained by Eq. (4.27), and the frame reciprocal to {a1 , a2 , ..., aN }
is given by an equation analogous to that in Eq. (2.41).

126

CHAPTER 4. EUCLIDEAN GEOMETRY

4.2.5

A simple problem in plane geometry

In this section we illustrate the use of vector based methods in plane geometry3 .
We want to show that
Claim 8 An angle inscribed in a semicircle is a right angle.
First, we draw a figure of the inscribed angle (Fig. 4.6 [a]). Then we label the
sides of the angle by the vectors d and e in Fig. 4.6 [b]. But then, if d e = 0,
the angle is evidently a right angle. In order to evaluate the dot product, we
must somehow relate d to e. This is done in Fig. 4.6 [c], where we have drawn
from the midpoint of the base of the semicircle the vector c to the vertex of the
angle, and vectors a and a along the base of the semicircle. Now we can solve
d and e in terms of a and c, which produces the result in Fig. 4.6 [d]. In other
word, we have now solved two circuit equations (the equations relating the sides
of the two triangles). Note that we have deleted the arc of the semicircle since
it plays no more any role in the mathematical formulation of the problem. The
fact that tip of both a and c lies in the arc is encoded by setting their lengths
equal, i.e., |c| = |a|. From this, the claim follows, since
(c a) (c + a) = |c|2 |a|2 = |a|2 |a|2 = 0

(4.32)

so the angle is indeed /2.


The procedure followed above can be concentrated into simple set of rules:
1. Draw a figure of the given problem.
2. Restate the given information in vector form
3. Form the circuit equations and eliminate as many vector variables as you
can
4. From the information provided by the circuit equations solve the problem
analytically.
3

This Section is based on the article Patrick Reany, Proofs in plane geometry using vector

methods. Arizona J. Nat. Phil. 3, 8-12 (1991).

4.3. CONFORMAL MODEL

127

Figure 4.6: Inscribed angle.

4.3

Conformal model

In conformal model points are still represented as vectors. As we shall see in the
next chapter, the term conformal is well justified, because the angle preserving
transformations can be written in this model as linear transformations.
In the conformal model the inner product pi pj of two points4 pi and pj
equals to the square x2i + x2j 2xi xj of their Euclidean distance |xj xi |, up
to a constant non-zero scalar factor. Then p2 = 0, which implies that a non-zero
scalar multiple mp represents the same point as p. On the account of Eq. (4.2),
this requirement eliminates Eq. (4.1) as a possible candidate.
As a start for finding the respresentation, we write point p as
p = (x) e + x + f

(4.33)

Now, x is the position vector of point p from some chosen reference point p0 (the
origin), e and f are vectors and is a scalar independent of x, and (x) is some
scalar valued function of x. Note that we do not assume that the vectors e and f
would necessarily lie in the Euclidean space E(N), where all the points lie. Now,
pi pj = i j e2 + i e (xj + f ) + j e (xi + f ) + 2 xi xj + (xi + xj ) f + f 2 ,
4

To emphasize the conformal representation, the points are written in sans serif.

128

CHAPTER 4. EUCLIDEAN GEOMETRY

where i = (xi ). By setting = 2, we can require that


pi pj = 2 |xj xi |2 = 2(x2i + x2j ) + 4xi xj

(4.34)

It follows then that f 2 = 0. By further assuming that i 6= 0, it follows that


e2 = 0, e xi = e xj = 0, (x) = x2 and e f = 2.
Let us now introduce the notation 0 for the null-vector e, and 0 for the nullvector f . Then the point p can be represented as
p = 2x + x2 0 0

(4.35)

The vectors 0 and 0 in effect extend the N -dimensional Euclidean space, where
the points p lie, to (N + 2)-dimensional Minkowski space M(N +2) with signature
(N + 1, 1). The Minkowski space contains the orthonormal basis vectors
e = (0 + 0)/2

(4.36)

e = (0 0)/2

(4.37)

in addition to {u1 , u2 , u3 }. The signature (N + 1, 1) follows from the square of


the basis vectors: u21 = u22 = u23 , ..., u2N +1 = e2 = 1, and e2 = 1. We could
include the origin vector p0 to our model as a zero vector, which anticommutes
with every other vector in the model and write the point as
p = 2 (p0 + x) + x2 0 0

(4.38)

From an algebraic standpoint, the inclusion of p0 results in a degenerate space


with N + 1 positive, one negative and one square zero basis vectors.
Because any point p can be normalized as
p 2

p
p0

(4.39)

it is evident that 0 represents the point at infinity. The vector x can be obtained
from p as
x = p EE
where
E=

0 0
2

(4.40)

(4.41)

4.3. CONFORMAL MODEL

129

is a bivector, which fulfills E2 = 1.


By using the null-property of 0 and 0, and their anticommutation with any
vector x, which resides in the Euclidean space E(N ), we can easily write down the
following multiplication table for the geometric product of the relevant quantities
in the conformal model (the product is taken in the order first the left column,
then the top row)
xj

xi

xj 0

pj

xi 0

xi 0

xi xj

2xi xj

2xj 0

+x2j xi 0

x 0
i

2 (1 E)

2 (1 E)
(

xj 0

(
pi

4.3.1

2xi xj x2i xj 0
+xj 0

2 (1 + E)

2xi 0

2 (1 + E)

2xi 0
+2x2i (1 E)

2xj 0
+2x2j (1 + E)

2(xi xj )2
+4xi xj

+2 x2j xi x2i xj 0
+2 (xj xi ) 0

+2 x2j x2i E
(4.42)

Addition of points

Although pi and pj are vectors, the point, the points are not combined via the
vector addition. In other words, pi + pj does not represent the point labeled by
the position vector xi + xj (with respect to chosen origin p0 ) in the Euclidean
space E(N ). This can be seen by noticing that (xi + xj )2 = x2i + x2j + 2xi xj ,
which implies that the correct representation is
p = 2 (xi + xj ) + (x2i + x2j + 2xi xj )0 0

(4.43)

130

CHAPTER 4. EUCLIDEAN GEOMETRY

which is clearly not equal to pi + pj . Instead, as the reader no doubt can easily
verify, the points are combined via more complicated rule
pi pj = pi + pj + (pi EE) (pj EE) 0 + 0

(4.44)

We use the symbol to signify the combination of points in the conformal model
in order to clearly emphasize that it differs from the vector addition.

4.3.2

Euclidean transformations

Two figures in Euclidean geometry are said to be congruent if the distances between corresponding pairs of points are equal. If the figures can be superimposed,
they are directly congruent, or superposable. If their handness is opposite, the
figures are enanthiomorphous. Naturally, we can also specify Euclidean transformations to direct and opposite according if they change the orientation of the
pseudoscalar to opposite or not, or equivalently, if they change the bivector angles tj ij to tj ij . Translations and rotations are direct but the reflection along a
vector is opposite since it flips the the orientation of the pseudoscalar (Exercise
77).
If the geometric point p in E(N ) were represented as vectors x in E(N ), the
translation of the point p0 + x to the direction of a would be performed as
x x + a. Because the combination of two translations produces x x0 x00
= x + c + c0 6= x + x0 + c + c0 , it is evident that translations are not linear
operations. This implies further that although the reflection of a vector x about
p0 along a (unit) vector u is a linear transformation u1 xu, the reflection of any
(point-bound) vector, whose tail is at the point p is not, since it would require
translating (parallelly) the tail of the vector to the origin p0 , performing the
reflection along a vector u there and then translating the tail of the result vector
back to p. Similar comment applies also in the case a point is rotated about some
other point besides the origin.
Luckily, in our (N + 2)-dimensional model all the Euclidean transformations
are linear! Especially, the transformations, which preserve the angles and the orientation of the pseudoscalar IN i.e. the conformal transformations translations,
rotations can be expressed linear two sided spinor transformations.

4.3. CONFORMAL MODEL

131

Translation
The translation
pi pj = pi + xij

(4.45)

pi pj = 2 (xi + xij ) + (xi + xij )2 0 0

(4.46)

or

is handled by the spinor transformation


pj = T 1 (xij ) pi T (xij )

(4.47)

where the spinor T (xij ) is given by




0xij
T (xij ) = exp
2


=1

0xij
2

(4.48)

and its inverse by T 1 (xij ) = T (xij ). The series expansion terminates after
two terms, because 0 is a null-vector. Notice the minus sign in the exponent.
Proof. To prove this, it suffices to note that
T 1 (xij ) 0T (xij ) = 0

(4.49)

T 1 (xij ) 0T (xij ) = 0 2xij x2ij 0

(4.50)

T 1 (xij ) xi T (xij ) = xi + xij xi 0

(4.51)

and

By the direct substitution we obtain T 1 (xij ) pi T (xij ) = 2 (xi + xij xi 0)+x2i 0




0 2xij x2ij 0 = 2 (xi + xij ) + x2i + 2xij xi + x2ij 0 0, which completes the
proof.
Reflection
The reflection of the point p0 + x along the vector u about the origin p0 ,
p0 + x p0 u1 xu

(4.52)

is achieved in the conformal model via the transformation


p u1 pu

(4.53)

132

CHAPTER 4. EUCLIDEAN GEOMETRY

Rotation
Because the rotor R is an even multivector in E(N ), it obviously commutes with
0 and 0. Thus we can write
R pR = 2R xR + x2 0 0

(4.54)

which has the effect of rotating the point p0 + x in the rotation plane hRi2 about
the origin p0 .
Direct transformations
A direct transformation, or rigid displacement
x x0 = R (B) xR (B) + c

(4.55)

is a rotation R (B) in the plane B about the origin p0 followed by the displacement
c. In the conformal model rigid displacement in Eq. (4.55) can be written as a
linear transformation
p p0 = D1 (c) pD (c)

(4.56)

where the spinor D (c) is given as the product


D (c) = R (B) T (c)

(4.57)

D1 (c) = [R (B) T (c)]1 = T 1 (c) R (B)

(4.58)

and

Screw displacement Rigid displacement can also be expressed as a screw


displacement. It consists of a rotation Rpb (B) about an axis u, which passes
through some point pb = p0 + b, followed by a translation ck = c uu along the
rotation axis u (see Fig. 4.7). This is known as Chasles theorem. We can write

the screw displacement as Rpb (B) T ck , if we can find the vector b so that


D (c) = R (B) T (c ) T ck = Rpb (B) T ck

(4.59)

where R (B) Rp0 (B) is the rotation about the origin p0 , and c is component
of c in B, i.e,
c = ccBB1

(4.60)

4.3. CONFORMAL MODEL

133

Figure 4.7: Screw displacement {u01 , u02 , u03 } {u001 , u002 , u003 }.
which naturally is orthogonal to the rotation axis u. Because the rotation at
an arbitrary point pc = p0 + c can be achieved by translating c to the origin,
followed by a rotation R (B) about the origin, and then translating back, we may
write
0
Rpc (B) = T (c) R (B) T 1 (c) = R (B) + [R (B) , c]
(4.61)
2
where [A, B] = AB BA is the commutator of A and B. From Eq. (4.59) we
obtain Rpb (B) = R (B) T (c ). By Eq. (4.48) this can be written as Rpb (B) =
R (B) (1 0c /2) = R (B) 0R (B) c /2. On the other hand, by Eq. (4.61),
Rpb (B) can be written as Rpb (B) = R (B) + 0[R (B) , b]/2. Thus,
R (B) c = [b, R (B)]

(4.62)

Because b is in the rotation plane B, the commutator can be written as [b, R (B)] =
bR (B) R (B) b = b[R (B) R (B)] = 2b hRi2 , and we finally obtain the position vector b of the point pb in the screw axis as
R (B) c hRi1

2
b=
2

(4.63)

It should be emphasized that the result in Eq. (4.63) aplies in any N-dimensional
space E(N ) with N > 2.

134

CHAPTER 4. EUCLIDEAN GEOMETRY

Figure 4.8: Translation as two reflections about parallell planes (top view).

Opposite and direct transformations as combination of reflections

Geometrically, the product of reflections in two parallell but distinct planes is


a translation in the perpendicular direction through twice the distance of the
planes, and a product of reflections in two intersecting planes is a rotation about
the line of intersection through twice the angle between them.
The result of an odd number of reflections is an opposite transformation.
Similarly, the result of an even number of rotations is a direct transformation.
In three-dimensional space an Euclidean transformation can be represented as a
product of at most four reflections. First, the object is rotated by two successive
reflections about two intersecting planes. Then the result is reflected about a
plane, which is perpendicular to the rotation axis of the previous transformation. These three reflections amount in a screw displacement and the flip of the
handness of the object to the opposite. It is called the rotary-reflection. The
fourth reflection about a plane passing along the rotary-reflection axis restores
the handness of the figure back to the original, if needed.

4.3. CONFORMAL MODEL

4.3.3

135

Lines

The geometric product is pi pj is equal to


pi pj = pi pj + pi pj

= 2(xi xj )2 + 4xi xj + 2 x2j xi x2i xj 0

+2 (xj xi ) 0 + 2 x2j x2i E

(4.64)

The bivector part




Bij = pi pj = 4xi xj +2 x2j xi x2i xj 0+2 (xj xi ) 0+2 x2j x2i E (4.65)
of the product pi pj encodes the points pi and pj directly, since they are the only
solution to the equation
Bij p = 0

(4.66)

with p2 = 0. Bij has the square


B2ij = Bji Bij = (pj pi + pi pj ) (pi pj pi pj )
= pi pj (pj pi + pi pj ) (pi pj )2 = (pi pj )2

(4.67)

which, of course, equals to 4 |xj xi |4 > 0. The points pi and pj can be solved
k
from a given Bij by taking the projection PBij (a) = aaBB/ |B|2 of some arbik
k
trary vector to B. Then pi = P (a) (1 + B/ |B|) and pj = P (a) (1 B/ |B|).
Bij

Bij

By taking the outer product of Bij with 0 we obtain


pi pj 0
= 2 (xi xj 0 + xij E)
2

(4.68)

From this, we immediately recognize the moment of line M = xi xj (Eq. (4.7))


and the direction vector xij = xj xi of the line through the points pi and pj (in
that order). Accordingly, we identify the outer product pi pj 0 as the directed
line passing through the points pi and pj . We can neglect the numeric factor 1/2
in this interpretation, because generally both p and any its multiple represent the
same point so it makes no difference if both sides of the equation are multiplied
by a constant.
It follows from the interpretation given to Eq. (4.68) that a point p lies in the
line passing through the points pi and pj if
pA=0

(4.69)

136

CHAPTER 4. EUCLIDEAN GEOMETRY

where
A = pi pj 0

(4.70)

Eq. (4.69) can be alternatively interpreted as the condition for the linear
dependency of p from the three vectors pi , pj and 0. Then we can write
p = pi + pj + 0

(4.71)

where we assume the normalized form. By dotting this with 0 we obtain 1 = +,


and by squaring both sides we obtain 0 = x2ij 2 ( + ) . Thus we can
eliminate = 1 and = /[2( + )] = (1 ) /2 and obtain the
one-parameter equation for the line
p = pi + (1 ) pj +

(1 )
0
2

(4.72)

For the interval 0 1, this equation parametrizes the line segment pi pj .

4.4

Further reading

Thousands of books have been written on Euclidean geometry. The one by Brannan, Espen and Gray [12], Geometry, is written clearly and it has the advantage
of including full solutions to over 200 problem. The book follows Kleins Erlanger
program in defining geometry as a set of transformations in space.
A geometric algebra formulation of plane geometry is given in the on-line
course by Calvet [13].
A recent textbook on geometric algebra by Doran and Lasenby [19] includes
a chapter on Euclidean geometry.
Euclidean transformations and geometry in complex plane is explored in Needhams outstanding book Visual complex analysis [58]. The reader should carefully
compare his approach to the present one.
A multi-purpose program (GAViewer) for performing geometric algebra computations and visualizing geometric algebra can be obtained from the web page
http://www.science.uva.nl/ga/viewer/index.html maintained by L. Dorst. A beta
version of an interactive three-dimensional JAVA-based sketching program by E.
M. Hitzer is available in http://sinai.mech.fukui-u.ac.jp/gcj/software/KamiWaAi/
index.html.

4.5. EXERCISES

137

The interplay of points and continuum is explored the article by Bell [7].
The roots of the set-theoretic approach to geometry are explored in the biography of Cantor, The mystery of Aleph: Mathematics, the Kabbalah, and the
search for infinity, written by Aczel [1].

4.5

Exercises

Exercise 67 Prove that

2 is irrational number (i.e. it can not pe written as a

ratio of integers). Hint: Assume the length of a diagonal of a square is an m-fold


and the length of a side of a square is an n-fold multiple of a unit length. Use
the Pythagorean theorem to show that this assumption implies that the integers
m and n are both even and have no common factor, which is contradiction.
Exercise 68 Show that if a real number line is regarded as an archetype for a
geometric line there are as many points in any two segments of the line, no matter
how long the segments are. Hint: There is one-to-one relation of points in the
real line segment y and points in line segment x, given by y = x, where is the
ratio of the lengths of the segments.
Exercise 69 Assume again that points can be labeled by real numbers. Show that
the number of points in a unit line segment is equal to the number of points at a
unit square. Hint: Proceed by labeling a point in the unit square by its Cartesian
coordinates {x, y} = {0.a1 a2 ..., 0.b1 b2 ...} and the points at the unit line segment
by x = a1 b1 a2 b2 ..., where a1 , a2 , ..., b1 , b2 , ... are the digits from 0 to 9. Then use
the result derived in Exercise 68.
Exercise 70 At a first sight, it appears that the end point 1 is missing from the
line, as well as the lines (x, 1) and (1, y) in the boundary of the square in Exercise
69. Show that this is not the case by if the real number 0.9999... is understood
as shorthand for the infinite sum

9/10n = 9/10 + 9/100 + 9/1000 + ...

n=1

Exercise 71 Prove that lines drawn from the vertices to opposite sides intersect
at a common point if they are normal to the sides. Hint: use the fact that

138

CHAPTER 4. EUCLIDEAN GEOMETRY

x z = (b z) y x = (c x) and z y = (a y), where , and


are some scalar proportional constants, and x, y and z are vectors labeling
the vertices of a triangle formed by the intersections of the lines drawn from the
vertices a, b and c to the opposite sides.

Exercise 72 Show that in a right triangle the midpoint of the hypotenues is


equidistant from the vertices.
Exercise 73 Show that the perpendicular bisectors of a triangle are concurrent
at a point.
Exercise 74 Why is the vector b, which locates the position in the screw axis,
(Sect. 4.3.2) necessarily orthogonal to it?
Exercise 75 Prove that 000 = 40.
Exercise 76 Prove that A A = 16x2ij , where A is given by Eq. (4.70).
Exercise 77 Prove that reflection along a vector flips the rientation of a unit
pseudoscalar of the space. Hint: show that uIN u = uuIN for any unit vector u
in the space E(N ).
Exercise 78 Derive Eq. (4.49).

4.5. EXERCISES

139

Exercise 79 Prove that T 1 (q) eT (q) = (0 q) /2.


Exercise 80 Verify that A and B in Eq. (4.14) are the barycentric coordinates
t(1) = x a(1) and t(2) = x a(2) , respectively, where {a(1) , a(2) } is the reciprocal
frame to {a, b}.

140

CHAPTER 4. EUCLIDEAN GEOMETRY

Vous aimerez peut-être aussi