Clifford Algebras

Clifford Algebras
Daniel Klawitter
Clifford Algebras
Geometric Modelling
and Chain Geometries
with Application in Kinematics
Foreword by Prof. Dr. Gunter Weiss
Dr. Daniel Klawitter

Dresden, Germany
Dissertation TU Dresden, 2014
ISBN 978-3-658-07617-7
DOI 10.1007/978-3-658-07618-4
ISBN 978-3-658-07618-4 (eBook)
The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie;

detailed bibliographic data are available in the Internet at http://dnb.d-nb.de.
Library of Congress Control Number: 2014953940
Springer Spektrum
Springer Fachmedien Wiesbaden 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole
or part of the material is concerned, specifically the rights of translation, reprinting, reuse of
illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical
way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed.
Exempted from this legal reservation are brief excerpts in connection with reviews or scholarly analysis or material supplied specifically for the purpose of being entered and executed
on a computer system, for exclusive use by the purchaser of the work. Duplication of this
publication or parts thereof is permitted only under the provisions of the Copyright Law of
the Publishers location, in its current version, and permission for use must always be obtained
from Springer. Permissions for use may be obtained through RightsLink at the Copyright
Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are
exempt from the relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date
of publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made. The publisher makes no warranty,
express or implied, with respect to the material contained herein.
Printed on acid-free paper
Springer Spektrum is a brand of Springer DE.
Springer DE is part of Springer Science+Business Media.
www.springer-Spektrum.de
To my parents,
Manuela and Burghard
Foreword
What is the right mathematical model to a real phenomenon of our
world? Do there exist criteria whether a model can be called elegant as well as practically ecient? These questions will surely be
answered dierently by e.g. a pure mathematician on the one hand
and an engineer on the other. But both will have to start with abstracting real world phenomena to objects of a more or less platonic
world. In this idealized world one constructs the structured geometric image of real world processes and objects in consideration. Thus,
on the way to constructing explicit mathematical models, geometric
representations, and geometric reasoning will always play an essential
part and provides insight into perhaps not that obvious interrelations
and structures.
The idea for this book took its origins from space kinematics and
robotics and the standardized way. Its aim is to be understood and
handled practically. There are two essentially dierent ways to model
space kinematics, although both ways are based on an Euclidean motion as the core element of the manifold of motions.
Loosely spoken, a seemingly clear model will interpret such a basic
element as a point and the manifold of motions as a point manifold.
(It turns out, that, via a clever transfer mapping, this point manifold becomes a hyper-quadric in a seven-dimensional real projective
space. This hyper-quadric is referred to as Studys quadric.) The
second model interprets a motion as a number rather than a point on
a line: In order to describe the position of it on that line, one needs a
pair of Hamilton quaternions as coordinates, which in addition fulll
conditions. Both models relate to well understood families of models.
The rst one can be seen e.g. as a higher dimensional analogue to the
geometry of lines and screws. Indeed, screws are closely connected
to motions. The second model relates to so-called circle geometries.
VIII
Foreword
Their structures are those of a line with coordinates of a ring. (A

standard example of such a circle geometry is the so-called (planar)
M
obius geometry; here it turns out that the coordinate ring of the
corresponding line is even a eld, namely the eld of complex numbers.)
Both standard models for space kinematics have their merits and can
be considered elegant. The rst model aims at a visualization of the
structured set of motions, where coordinates become inessential. In
spite of the transfer between the original set of motions and Studys
quadric, the seven-dimensional real projective space is described by
coordinates. Now the model induces objects of further research: For
example, planar intersections of Studys quadric lead to interesting
subsets of motions. The second model aims at a clearly performable calculus and the analytic treatment of forced motions. Here as
well, one feels the need for visualization. Such transfer from calculus
to geometry establishes a third model and can be derived as follows:
Pairs of quaternions are interpreted as elements of a four-dimensional
module over the ring of so-called dual numbers, which again are interpreted as points. Points originating from motions are contained
by a hypersphere in this four-dimensional module space and, again,
planar intersections of this sphere become a matter of interest. But
these planar intersections stem from helical motions in the original
world.
In this setting, space kinematics and its models are used to point
out an obvious fact: The models and model spaces, which are structured for their own, inuence the topics of research of the original real
world objects and re-structure these objects in a specic manner.
Therefore, the answer to the introductory question, which model is
best, will be rather a matter of taste and pre-knowledge of related
models than being based on objectied facts. But still the origin of
dierent models is the object of consideration! So the question about
relations between dierent models and their structures arises. Is is
possible to unify these view points?
This last question is the core topic of this book: The author introduces the seemingly universal tool of Cliord Algebras and their
Foreword
IX
group structures to model not only Euclidean kinematics and line geometry, but also circle geometries and their generalizations, so-called
chain geometries and, to take it even further, also non-Euclidean
kinematics and line geometry. Projective and chain geometric properties alternately relate to algebraic properties and concepts. Again,
the concept cross ratio becomes such an interface between Geometry and Algebra, being on the one hand visualized as quadruplets
of points and on the other being a number resp. a ring element as
well as responsible for structuring e.g. specic point sets of a line
to chains. The presented method allows to connect disciplines of
(Applied) Mathematics, that, up to now, are treated independently.
Furthermore, it provides an eective tool for solving practical problems of Mechanical Engineering and Robotics, too.
Is this book a Maths book on theoretical and practical applications
of Cliord Algebras? Yes, but in its rich details and examples it aims
at visualizing the more or less hidden Geometries of the Cliord Algebra calculus, and therefore, it could rightly be called a Geometry
book. By its wide range of examples it is protable to a broad span
of readers, be it mechanical engineers or mathematicians. Like me,
they surely consider its publication as an absolute stroke of luck.
Wien
Gunter Weiss
Preface
Usually, dual unit quaternions are used to describe Euclidean displacements in three-dimensional kinematics. Engineers widely use
this elegant calculus next to a real projective geometric model called
Studys quadric. Moreover, there exists a sphere-model in a fourdimensional module space over the ring of dual numbers. However,
there are no investigations connecting these three models up to now.
After recalling Studys quadric, the sphere-model , and other models, we present a survey-like introduction to Cliord algebras and their
Spin and Pin groups. The advantage of Cliord algebras is that geometric objects and transformation may be represented as algebra elements. Using the so-called sandwich operator, it is possible to apply
transformations that are represented by algebra elements directly to
geometric objects, also described by algebra elements. We introduce
the homogeneous and conformal Cliord algebra model. Furthermore, we show how to model special Cayley-Klein geometries and
their isometry groups in a homogeneous Cliord algebra model. As
an example, we focus on the homogeneous Cliord algebra model corresponding to line geometry and derive the correspondence between
projective transformations of three-dimensional projective space and
the Pin group Pin(3,3,0) . Therefore, we introduce the Cliord algebra
C(3,3,0) , constructed over the quadratic space R(3,3) , and describe how
points on Kleins quadric are embedded as null vectors. We discuss
how geometric entities that are known from Kleins model, i.e., linear line manifolds can be transferred to this homogeneous Cliord
algebra model. All entities known from line geometry occur naturally in this model and can be transformed projectively by the application of the sandwich operator. The action of grade-1 elements
corresponds to the action of null polarities on P3 (R), i.e., correlations
that are involutions as the basic elements building up the group of
regular projective transformations. It is proven that every regular
projective transformation of P3 (R) can be expressed as the product
XII
Preface
of six null polarities, i.e., skew-symmetric 4 4 matrices at the most.

The results achieved for Kleins quadric may be transferred to any
quadric, hence, we present the homogeneous Cliord algebra model
corresponding to Lie sphere geometry as an example. Additionally, a
new geometric algebra allowing the description of inversions with respect to quadrics in principal position as Pin group is presented. This
model serves as a generalization for the conformal geometric algebra
and is constructed for dimension two and three in detail. Furthermore, the generalization to arbitrary dimension is shown.
A further focus of the thesis is applying chain geometry to Cliord
algebras in order to examine the cross ratio in Cliord algebras. It
is well known that the cross ratio of four complex numbers is real
if, and only if, they all lie on a M
obius circle, i.e., a circle or a line
augmented by a point at innity. A generalization of the so called
Mobius geometry is obtained by using dierent algebras instead of
complex numbers. This leads to a branch of geometry called chain
geometry. Chains are subsets of the projective line over a ring which
can be parametrised with the cross ratio. Therefore, it is natural to
apply this theory to dual quaternions and to examine the kinematic
and geometric interpretation. A more general point of view can be
achieved by the use of Cliord algebras and Spin groups instead of
dual quaternions and dual unit quaternions. After recalling the fundamental chain geometric background, we dene the cross ratio for
Cliord algebras and their Pin and Spin groups. We present a quadric
model corresponding to the dual unit quaternions and homogeneous
Cliord algebra models of Kleins, Studys and Lies quadric where
chains that are contained in the grade-1 subspace correspond to conic
sections. Moreover, we derive an algebraic biarc construction with the
help of contact spaces. Chains of the grade-1 subspace that are in
contact at a certain point are parametrized with the cross ratio and
correspond to conic sections. Moreover, it has been proven that the
connected components of the Pin- and Spin groups dene subspaces
of chain spaces. Every element contained in a chain dened by three
elements of the same connected component of the Pin- or the Spin
group is contained in the same connected component of the Pin- or
the Spin group. The question for the cross ratio of dual unit quaternions has been answered in detail.
Preface
XIII
In the last chapter we recall the concept of kinematic mappings. Different kinematic mappings have been found, for example the kinematic mapping of Study that maps the group of Euclidean displacements to Studys quadric, or the kinematic mapping of Blaschke and
Gr
unwald. Furthermore, we use the Cliord algebra calculus to unify
dierent kinematic mappings. With this construction it is possible
to unify the concept of kinematic mappings for isometry groups of
Cayley-Klein geometries and for orthogonal groups SO(p, q). Collineations in any kinematic image and the corresponding Cayley-Klein
space can be derived from the homogeneous Cliord algebra model.
The kinematic images of Pin- and Spin groups are projective varieties. Due to the fact that projective varieties correspond to ideals,
methods of Grobner basis calculus can be applied to kinematic image spaces. We present kinematic mappings for Euclidean spaces of
dimensions two, three and four in detail.
Dresden
Daniel Klawitter
Contents
Foreword
Preface
Introduction
1 Models and Representations
1.1 Description of Displacements . . . . . . . . . . . . . . .
1.1.1 Homogeneous Matrices . . . . . . . . . . . . . .
1.1.2 Dual Quaternions . . . . . . . . . . . . . . . . .
1.1.3 Dual Orthogonal Matrices . . . . . . . . . . . .
1.2 Point Models for Lines and Displacements . . . . . . .
1.2.1 Kleins Quadric . . . . . . . . . . . . . . . . . . .
1.2.2 Studys Quadric . . . . . . . . . . . . . . . . . .
1.2.3 Studys Sphere . . . . . . . . . . . . . . . . . . .
1.3 Geometric Algebras, Cliord Algebras . . . . . . . . . .
1.3.1 Denition of a Geometric Algebra . . . . . . . .
1.3.2 Properties of Cliord Algebras . . . . . . . . . .
1.3.3 Pin and Spin Groups . . . . . . . . . . . . . . .
1.3.4 Matrix Representation of Cliord Algebras . .
1.3.5 Linear Transformation of the Vector Space . .
1.4 The Homogeneous Model . . . . . . . . . . . . . . . . .
1.4.1 The Construction . . . . . . . . . . . . . . . . .
1.4.2 Exterior Algebra . . . . . . . . . . . . . . . . . .
1.4.3 Homogeneous Model via projective Grassmann
Algebra . . . . . . . . . . . . . . . . . . . . . . .
1.5 The Conformal Model . . . . . . . . . . . . . . . . . . .
1.5.1 Construction of Conformal Geometric Algebra
1.5.2 Blades in CGA . . . . . . . . . . . . . . . . . . .
1.5.3 Conformal Transformations . . . . . . . . . . . .
VII
XI
1
5
5
5
6
13
15
15
21
25
26
26
28
33
36
37
38
38
39
41
42
43
45
48
XVI
Contents
1.6
A Cliord algebraic Approach to Line Geometry . . .

1.6.1 Collineations and Correlations in the Image Space
1.6.2 Algebra Representation of Linear Line Manifolds
1.6.3 Transformations . . . . . . . . . . . . . . . . . .
1.6.4 Collineations as Spin Group . . . . . . . . . . .
1.6.5 Correlations as Pin Group . . . . . . . . . . . .
1.6.6 Singular projective Transformations . . . . . . .
1.7 A Cliord algebraic Approach to Lie Sphere Geometry
1.7.1 Lies Quadric . . . . . . . . . . . . . . . . . . . .
1.7.2 The homogeneous Cliord Algebra Model corresponding to Lie Sphere Geometry . . . . . . .
1.8 A Cliord algebraic Approach to Studys Quadric . . .
1.9 A Cliord algebraic Approach to Studys Sphere . . .
1.10 Quadric Geometric Algebra . . . . . . . . . . . . . . . .
1.10.1 The Embedding . . . . . . . . . . . . . . . . . .
1.10.2 Geometric Entities . . . . . . . . . . . . . . . . .
1.10.3 Transformations . . . . . . . . . . . . . . . . . .
1.10.4 Eect on Lines and Points . . . . . . . . . . . .
1.10.5 Subgroups . . . . . . . . . . . . . . . . . . . . . .
1.10.6 Generalization to higher Dimensions . . . . . .
2 Chain Geometry over Cliord Algebras
2.1 Chain Geometry . . . . . . . . . . . . . . . . . .
2.1.1 Distance Spaces . . . . . . . . . . . . . .
2.1.2 The projective Line over an L-algebra .
2.1.3 The Projective Linear Group PGL(R, 2)
2.1.4 The projective Line over a Subring . . .
2.2 Chain Geometry as Incidence Geometry . . . .
2.2.1 Denition of a Cross Ratio . . . . . . . .
2.3 Quadric Chain Spaces . . . . . . . . . . . . . . .
2.4 Real Benz Planes . . . . . . . . . . . . . . . . . .
2.5 Jordan-Systems . . . . . . . . . . . . . . . . . . .
2.6 Contact Spaces . . . . . . . . . . . . . . . . . . .
2.7 Chain Geometries over Cliord Algebras . . . .
2.7.1 Grade-1 Subspace . . . . . . . . . . . . .
2.7.2 Pin and Spin Groups . . . . . . . . . . .
2.8 Quadric Chain Geometry . . . . . . . . . . . . .
2.8.1 Kleins Quadric . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
50
51
53
55
61
67
72
72
73
74
76
77
78
78
80
84
87
89
94
101
101
101
102
104
105
106
109
114
116
118
121
123
124
126
131
131
Contents
XVII
2.8.2 Lies Quadric . . . . . . . . . . . . . . . . . . . .

2.8.3 Studys Quadric . . . . . . . . . . . . . . . . . .
2.8.4 Studys Sphere . . . . . . . . . . . . . . . . . . .
2.9 Quadric Chain Spaces for certain Spin Groups . . . . .
2.9.1 A Quadric Model for dual unit Quaternions . .
2.9.2 Other possible Quadric Models . . . . . . . . .
2.10 Cross Ratio of dual unit Quaternions . . . . . . . . . .
2.10.1 Subspaces on Studys Quadric and Sub Chain
Geometries . . . . . . . . . . . . . . . . . . . . .
2.10.2 Application to line-symmetric Displacements .
2.10.3 Dual Quaternion Cross Ratio and Conics on
Studys Quadric . . . . . . . . . . . . . . . . . .
2.11 Chains of Geometric Entities . . . . . . . . . . . . . . .
2.11.1 Quadric Geometric Algebra . . . . . . . . . . .
2.12 Biarc Construction . . . . . . . . . . . . . . . . . . . . .
2.12.1 Biarcs as touching Chains . . . . . . . . . . . .
2.12.2 Biarcs on Quadrics in three-dimensional Space
2.12.3 Kleins Quadric . . . . . . . . . . . . . . . . . . .
2.12.4 Biarcs on the Dual Sphere . . . . . . . . . . . .
133
135
136
139
139
141
145
150
151
156
161
161
165
168
169
176
177
3 Kinematic Mappings for Spin Groups

3.1 Cayley-Klein Geometries and the homogeneous Model
3.1.1 Cayley-Klein Spaces . . . . . . . . . . . . . . . .
3.1.2 A homogeneous Model for Euclidean Geometry
3.2 Kinematic Mappings . . . . . . . . . . . . . . . . . . . .
3.2.1 Studys kinematic Mapping . . . . . . . . . . . .
3.2.2 A Mapping for planar Displacements . . . . . .
3.3 Kinematic Mappings via Cliord Algebras . . . . . . .
3.3.1 Studys Mapping via Cliord Algebra . . . . . .
3.3.2 Blaschkes and Gr
unwalds Mapping via Clifford Algebra . . . . . . . . . . . . . . . . . . . .
3.4 Kinematic mappings for other Cayley-Klein Spaces . .
3.4.1 Two-dimensional Cayley-Klein Spaces . . . . .
3.4.2 Three-dimensional Cayley-Klein Spaces . . . .
3.4.3 Higher dimensional kinematic Mappings . . . .
3.5 Projective Varieties via kinematic Algebra Elements .
181
181
182
184
185
185
185
187
187
Conclusion
201
190
192
192
193
195
197
XVIII
Contents
Index
203
List of Symbols
205
Acknowledgment
209
Bibliography
211
Introduction
Euclidean displacements in three-dimensional kinematics are usually
described by dual unit quaternions. Engineers widely use this elegant
calculus aside a real projective geometric model called Studys quadric. Moreover, there exists a sphere-model in a four-dimensional
module space over the ring of dual numbers. Seemingly there are no
investigations to connect these three models up to now and this was
the rst motivation for the present research, which turned out to lead
to a far more general view. One theme of interest was thereby to nd
interpretations of dierent projective closures of the two ring based
models and the real projective model.
The cross ratio is a fundamental invariant of projective geometry. It
is well known that the cross ratio of four complex numbers is real
if, and only if, they all lie on a M
obius circle, i.e., a circle or a line
augmented by a point at innity. A generalization of the so-called
Mobius geometry is obtained by using dierent algebras instead of
the complex numbers. A cross ratio for dual quaternions was dened
in [58] by
cr(a, b, c, d) := (a b) (b c)1 (c d) (d a)1 .
With this denition a lot of questions arise. What is the cross ratio
of dual unit quaternions and how can it be interpreted geometrically
and kinematically? Under which preliminaries is the cross ratio of
four dual unit quaternions a real or a dual number? Are there points
on a conic section on Studys quadric corresponding to dual unit quaternions with a real or dual cross ratio?
Kinematic and line geometric models that are used today date back
to E. Study [61] and F. Klein [42]. These models nd application in
modern treatises, see for example [13], [38]. Especially line geometry
Introduction
experienced a renaissance through computational technology, see [56].

Dierent representations of Euclidean displacements are well-known.
Chapter 1 deals with these models. Studys quadric, the spheremodel and other models are described. Furthermore, we present a
survey-like introduction to Cliord algebras and their Spin and Pin
groups. The advantage of Cliord algebras is that geometric objects
and transformation may be represented as algebra elements. With
the so-called sandwich operator it is possible to apply transformations
that are represented by algebra elements directly to geometric object
also described by algebra elements. We introduce the homogeneous
and conformal Cliord algebra model. Furthermore, we show how to
model special Cayley-Klein geometries and their isometry groups in
homogeneous Cliord algebra models. Therefore, we focus on the homogeneous Cliord algebra model corresponding to line geometry and
derive the relationship between projective transformations of threedimensional projective space and the Pin group Pin(3,3,0) .
Homogeneous Coordinates for lines of three-dimensional projective
space P3 (R) were introduced by Pl
ucker, see [7]. The Pl
ucker coordinates of all lines form a quadric in ve-dimensional projective space,
the so-called Klein quadric denoted by M24 P5 (R), see [66]. The
group of regular projective transformations of P3 (R) is isomorphic to
the group of projective automorphisms of Kleins quadric M24 , see [56].
Moreover, the group of automorphic collineations of Kleins quadric
is the isometry group of the Cayley-Klein space given by P5 (R) together with absolute gure M24 . This isometry group corresponds to
the Pin group of a special homogeneous Cliord algebra model, see
[28]. Therefore, we introduce the Cliord algebra C(3,3,0) constructed over the quadratic space R(3,3) and describe how points on Kleins
quadric are embedded as null vectors. We discuss how geometric entities that are known from Kleins model can be transferred to this
homogeneous Cliord algebra model. The action of grade-1 elements
corresponds to the action of null polarities on P3 (R). Moreover, we
prove that every regular projective transformation of P3 (R) can be
expressed as the product of at most six null polarities, i.e., skewsymmetric 4 4 matrices. The results that were achieved for Kleins
quadric may be transferred to any quadric, and therefore, we give
the homogeneous Cliord algebra model corresponding to Lie sphere
Introduction
geometry as example. Furthermore, a new geometric algebra which allows the description of inversions with respect to quadrics in principal
position as Pin group is presented. This new model is constructed for
dimension two and three in detail. Furthermore, the generalization
to arbitrary dimension is presented.
In chapter 2 we give the answer to the question for a cross ratio
of four dual unit quaternions. This leads to a branch of geometry
called chain geometry. Chains are subsets of the projective line over
a ring that can be parametrised with the cross ratio. The origins
of chain geometry can be found in Benz [5]. Benz dealt with twodimensional algebras over the real numbers, i.e., the double numbers
or split-complex numbers A, the complex numbers C, and the dual
numbers D. He classied the chains in these three algebras. More
recent results can be found in [11] or [33]. Chain geometry is an
essential tool for this work. The connection between quadric models
and chain geometries over kinematic algebras was found by H. Hotje,
see [37]. A. Blunck and M. Stroppel generalized this approach
to so-called Klingenberg chain spaces, see [9] and [12]. Therefore, it
is natural to apply this theory to dual quaternions and to examine
the kinematic and geometric interpretation. The cross ratio of four
dual unit quaternions corresponding to line-symmetric motions with
respect to a hyperboloid of revolution
x2
y2
z2
+
= 1, with a = b,
a2
b2
c2
is real-valued, see [58]. However, there is no explanation in general.

To answer the question for the cross ratio in all generality is a declared
aim of this research. A more general point of view can be achieved
with the use of Cliord algebras and Spin groups instead of dual quaternions and dual unit quaternions. The study of the cross ratio of
four dual unit quaternions can be translated to the study of the cross
ratio of four Spin group elements in a special Cliord algebra. Therefore, we study the cross ratio in the more general Cliord algebra
setting. Cliord algebras and especially homogeneous and conformal
models allow an elegant description of the group action on geometric
entities. There is a lot of literature on these topics see [19, 27, 28]
or [45] to mention just a few. With the help of this framework more
Introduction
general properties can be derived, too. Within the Cliord algebra

context Spin groups corresponding to more general transformation
groups can be discussed. The theory of chain geometry can be applied to every Cliord algebra and its Pin or Spin group. In chapter
2 we give the fundamental chain geometric background and dene
the cross ratio for Cliord algebras and their Pin and Spin groups.
Furthermore, we present a quadric model corresponding to the dual
unit quaternions and homogeneous Cliord algebra models whose
Pin group corresponds to the group of automorphic collineations of
Kleins, Studys and Lies quadric. Moreover, we derive an algebraic
biarc construction with the help of contact spaces. Therefore, the
parametrisation of chains with the cross ratio is used to model conic
sections on arbitrary quadrics. Biarcs on quadrics are well-known, see
[56, 64]. By means of contact relations between chains it is possible
to transfer this construction to an algebraic setting. The advantage
of this method is that the biarc construction is independent of the
model.
In the last chapter we recall the concept of kinematic mappings. Different kinematic mappings have been found, for example the kinematic mapping of Study [61] that maps the group of Euclidean displacements to Studys quadric, or the kinematic mapping of Blaschke
[6] and Gr
unwald [26]. Furthermore, we use the Cliord algebra
framework to unify dierent kinematic mappings. We derive a method
to map every Pin or Spin group element to a point on certain pseudoalgebraic variety in projective space. The dimension of the projective space depends of the dimension of the Cliord algebra. We dene
Cayley-Klein spaces and show how to present them in a homogeneous
Cliord algebra model. With this concept it is possible to unify the
concept of a kinematic mapping for the displacement group of CayleyKlein geometries and for orthogonal groups SO(p, q). We present kinematic mappings for Euclidean spaces of dimensions two, three and
four in detail. Projective varieties correspond to ideals, and therefore,
the theory of Grobner basis calculus can be applied to image spaces.

In this chapter we give a survey on methods that are used nowadays to
describe spatial displacements and other transformations. Therefore,
point models for lines and spatial displacements in three-dimensional
Euclidean space are presented. Moreover, dierent methods to describe displacements are examined. We review old concepts such as
dual numbers, quaternions, and dual quaternions. Furthermore, we
give a short review about homogeneous coordinates and how to linearize displacements. After that we consider Cliord algebras and
special Cliord algebra models. These models are a current eld of
research.
1.1 Description of Displacements

1.1.1 Homogeneous Matrices
To describe Euclidean transformations we use a familiar trick. We
consider the three-dimensional Euclidean space as ane part of the
projective space P(R R3 ) see [31]. The rst coordinate is dened
as homogeneous factor and we denote the projective space by P3 (R).
Points in P3 (R) are described by homogeneous coordinates
(p0 , p1 , p2 , p3 )T . Note that two homogeneous coordinate vectors x, y
describe the same point if x = f y, f R\{0}. The factor f is called
the homogeneous factor. A point p P3 (R), p = (0, 0, 0, 0) is called
proper if p0 = 0 else the point is called ideal. Proper points correspond
to ane points with inhomogeneous coordinates. Hence, the projective space without the hyperplane given by x0 = 0 can be identied
with the ane space. We can step between the homogeneous and
the inhomogeneous representation by (p0 , p1 , p2 , p3 )T
D. Klawitter, Clifford Algebras, DOI 10.1007/978-3-658-07618-4_1,

p1
p0 ,
p2
p0 ,
p3
p0
T
This allows us to describe Euclidean displacements with help of linear

transformations in P(1 R3 ) as
p =
1
t
0T
A
p,
A SO(3), t R3 , p = (1, x1 , x2 , x3 )T .
(1.1)
The group of Euclidean displacements that can be represented with

Eq. (1.1) is in the focus of this work. In the following we review
representations of this group.
1.1.2 Dual Quaternions

Dual quaternions are quaternions with dual number entries. Therefore, we introduce quaternions and dual numbers rst.
Quaternions Quaternions H, introduced by W.R. Hamilton in
1843, constitute an elegant tool for the representation of rotations
in three- and four-dimensional Euclidean case, see [49]. Furthermore,
quaternions form a skew eld, i.e., multiplication is not commutative.
Unit quaternions are a double cover of the group SO(3). A general
quaternion has the form
q = a + bi + cj + dk,
a, b, c, d R,
(1.2)
where i, j, and k are the quaternion units with

i2 = j2 = k2 = ijk = 1.
(1.3)
Denition 1.1. Let q be a quaternion as in Eq. (1.2). Then, a

is called the scalar part and (b, c, d)T is called the vector part. A
quaternion with vanishing scalar part is called a vectorial quaternion.
Quaternions form a four-dimensional vector space over the real numbers. Addition is dened component-wise by
q1 + q2 = (a1 + a2 ) + (b1 + b2 )i + (c1 + c2 )j + (d1 + d2 )k.
The multiplication rules for the quaternion units (1.3) are summarized in the following scheme:
i
j
k
i
1
k
j
j
k
k j
.
1
i
i 1
This scheme allows to extend the concept of quaternion multiplication to general quaternions. The real numbers form the center of the
quaternions, i.e., real numbers commute with all other quaternions.
Later, in section 1.3 we study algebras and we see that the quaternions are an example of an R-algebra.
Denition 1.2. The anti-involution
: H H,
a + bi + cj + dk = q q = a bi cj dk
is called quaternion conjugation. The norm of a quaternion is dened

by

q =

qq =
a 2 + b2 + c 2 + d 2 .
Every quaternion q = 0 has an inverse quaternion that can be calculated by

q 1 =
1
q
q
=
.
q2
qq
Unit quaternions are quaternions with norm equal to 1. With respect

to quaternion multiplication unit quaternions form a group. Every
unit quaternion can be represented by
q = cos + d sin with d = (d1 i + d2 j + d3 k),
where d is an unit vector and R. Moreover, unit quaternions can

be used to describe rotations in three-dimensional Euclidean space E3 .
This can be realized with the so called sandwich operator x qxq
where q is a unit quaternion and the coordinate vector x of a point
is considered as a vectorial quaternion. Therefore, we follow [38]
and apply the sandwich operator to the standard basis vectors of
R3 written as vectorial quaternions. This means the basis vectors e1
belonging to the x-component is expressed by the quaternion x = i
and the eect of the sandwich operator results in
qiq = (a + bi + cj + dk)j(a bi cj dk)
= (ai b ck + dj)j(a bi cj dk)
= (a2 + b2 c2 d2 )i + 2(bc + ad)j + 2(bd ac)k.
In the same way we see the action of this operator on the vectorial
quaternions j and k
qjq = 2(bc ad)i + (a2 b2 + c2 d2 )j + 2(cd + ab)k,
qkq = 2(bd + ac)i + 2(cd ab)j + (a2 b2 c2 + d2 )k.
The sandwich operator is linear and the image of a vectorial quaternion is a vectorial quaternion again. Furthermore, the scalar product of two vectors x, y R3 is invariant under the action of the sandwich operator and it is orientation preserving. When collecting the
images of the basis vectors e1 , e2 , e3 in a matrix A, we get
a 2 + b2 c 2 d 2
A=
2(bc + ad)
2(bd ac)
2(bc ad)
a2 b 2 + c 2 d 2
2(cd + ab)
2(bd + ac)
2(cd ab) .
2
a b2 c 2 + d 2
(1.4)
Matrix (1.4) is the well-known form of a rotation matrix. The components of a unit quaternion are the Euler parameters of a rotation.
Dual Numbers Like complex numbers dual numbers are an extension of the real numbers. A dual number has the form z = a + b,
where a and b are real numbers and is the dual unit that squares
to zero, 2 = 0. Addition is dened component-wise. For two dual
numbers the product is dened by
(a1 + b1 )(a2 + b2 ) = a1 a2 + (a1 b2 + a2 b1 ) + 2 (b1 b2 )
= a1 a2 + (a1 b2 + a2 b1 ).
The set of dual numbers

D := a + b | a, b R, 2 = 0
together with addition and multiplication forms a commutative ring

with identity. Moreover, the dual numbers form a two-dimensional
commutative unital associative algebra over the real numbers. Dual
numbers with vanishing real part are zero divisors
(a)(b) = 2 (ab) = 0.
Denition 1.3. For a dual number z = a + b the dual number

z = a b is called the conjugate dual number. The norm of a dual
number that is no zero divisor then is
z :=
z z =

(a + b)(a b) = |a|.
Any dual number without vanishing real part has an inverse dual
number
z1 = (a + b)1 :=
1
z .
a2
Analytic functions can be extended to dual functions with help of

their formal Taylor expansion. Note that any power of that is
bigger than one vanishes. Therefore, we get the Taylor expansion
f (a + b) = f (a) + bf (a),
which is the dual extension of the analytic function.

Remark 1.1. It is possible to calculate the inverse of a dual number
by the Taylor expansion of z1 to make the denition of the inverse
dual number clear
z1 = (a + b)1 =
b
1
1
1
2 = 2 (a b) = 2 z .
a
a
a
a
Dual Vectors Later we will use dual vectors. Therefore, we introduce the n-dimensional module
Dn := {v | v = v +
v , 2 = 0, v, v Rn }.
A dual vector is the sum of its real- and dual part

v = v +
v , with v, v Rn .
We dene a standard scalar product on this module by

v wT = v , w
= v, w
+ ( v, w
+
v , w
),
where ,
denotes the standard scalar product of Rn . Especially for
the dimension n = 3 we are able to dene a cross product by
v w = v w + (
v w + v w).
(1.5)
10
Dual Quaternions Dual quaternions were introduced by E. Study,

see [61]. Nowadays, dual quaternions form a frequently used tool
for the description of Euclidean kinematics in three dimensions, see
[7, 38, 50] or [59]. In this section we give a brief introduction to dual
quaternions. Our intention is to put this concept in a more general
context by using Cliord algebras. Quaternions with dual number
components are called dual quaternions and are denoted by
Hd := {a0 + a1 i + a2 j + a3 k + (c0 + c1 i + c2 j + c3 k)
| a0 , . . . , a3 , c0 , . . . , c3 R} .
Multiplication is dened with the relations for quaternion. Furthermore, the dual unit commutes with the quaternion units i = i,
j = j, k = k. Dual quaternions form an eight-dimensional vector
space over the real numbers. The basis elements are 1, i, j, k, , i, j, k.
Displacements Euclidean displacements can be described by dual
unit quaternions. A dual quaternion q = a0 + a1 i + a2 j + a3 k + (c0 +
c1 i + c2 j + c3 k) is normed or a dual unit quaternion, if the norm is
equal to one
N (q ) := q q = a20 + a21 + a22 + a23 + 2(a0 c0 + a1 c1 + a2 c2 + a3 c3 ) = 1,
where the conjugation is the quaternion conjugation

q = (q1 + q2 ) = q1 + q2 .
Therefore, a dual unit quaternion satises two relations in the components a0 , . . . , c3

a20 + a21 + a22 + a23 = 1,
a0 c0 + a1 c1 + a2 c2 + a3 c3 = 0.
(1.6)
(1.7)
Dual unit quaternions are denoted by Ud and form a group with

respect to multiplication. More details can be found in [38]. Usually,
a displacement is described by the sandwich operator. We start with
a dual unit quaternion
q = a0 + a1 i + a2 j + a3 k + (c0 + c1 i + c2 j + c3 k)
11
and a dual quaternion of the form p = 1 + (xi + yj + zk), representing

the point P = (x, y, z)T R3 . We apply the sandwich operator as
qpq = 1 + (xa20 + 2za0 a2 2ya0 a3 + 2c1 a0 + xa21 + 2ya1 a2
+ 2za1 a3 2c0 a1 xa22 + 2c3 a2 xa23 2c2 a3 )i
+ (ya20 2za0 a1 + 2xa0 a3 + 2c2 a0 ya21 + 2xa1 a2
2c3 a1 + ya22 + 2za2 a3 2c0 a2 ya23 + 2c1 a3 )j
+ (za20 + 2ya0 a1 2xa0 a2 + 2c3 a0 za21 + 2xa1 a3
+ 2c2 a1 za22 + 2ya2 a3 2c1 a2 + za23 2c0 a3 )k.
If we rewrite the result as s product of a matrix with a vector vector

in homogeneous coordinates we arrive at

1
1
l
x
=
m
y
n
z
0
a20 +a21 a22 a23
2a0 a3 +2a1 a2
2a1 a3 2a0 a2
0
2a1 a2 2a0 a3
a20 a21 +a22 a23
2a0 a1 +2a2 a3

0
1
x
2a0 a2 +2a1 a3
,
2a2 a3 2a0 a1 y
a20 a21 a22 +a23
z
where
l = 2c1 a0 2c0 a1 + 2c3 a2 2c2 a3 ,
m = 2c2 a0 2c3 a1 2c0 a2 + 2c1 a3 ,
n = 2c3 a0 + 2c2 a1 2c1 a2 2c0 a3 .
This matrix vector product represents an Euclidean displacement, see

[38] and Eq. (1.4). A parametrisation of the special Euclidean group
with the help of dual unit quaternions is given by
Q = a0 + a1 i + a2 j + a3 k + c0 + c1 i + c2 j + c3 k
= cos sin l0 i sin l1 j sin l2 k

2
2
2
2

v
v
sin sin l3 + cos l0 i
2 2 2
2 2
v
v
sin l4 + cos l1 j sin l5 + cos l2 k,
2
2
2
2
2
2
(1.8)
where is the rotation angle and v is the magnitude of a translation in

the direction dened by the Pl
ucker coordinate vector
L = (l0 : l1 : l2 : l3 : l4 : l5 ), see [13]. We will dene Pl
ucker coordinates
in section 1.2.1.
12
Matrix Representation
The multiplication of two dual quaternions
A = a0 + a1 i + a2 j + a3 k + (c0 + c1 i + c2 j + c3 k),
X = x0 + x1 i + x2 j + x3 k + (x4 + x5 i + x6 j + x7 k),
can be written in matrix form. Therefore, we perform the multiplication as follows

AX = (a0 x0 a1 x1 a2 x2 a3 x3 ) + (a0 x1 + a1 x0 + a2 x3 a3 x2 )i
+ (a0 x2 + a2 x0 a1 x3 + a3 x1 )j + (a0 x3 + a1 x2 a2 x1 + a3 x0 )k
+ (a0 x4 a1 x5 a2 x6 a3 x7 + c0 x0 c1 x1 c2 x2 c3 x3 )
+ (a0 x5 + a1 x4 + a2 x7 a3 x6 + c0 x1 + c1 x0 + c2 x3 c3 x2 )i
+ (a0 x6 + a2 x4 a1 x7 + a3 x5 + c0 x2 + c2 x0 c1 x3 + c3 x1 )j
+ (a0 x7 + a1 x6 a2 x5 + a3 x4 + c0 x3 + c1 x2 c2 x1 + c3 x0 )k.
If we rewrite AX as a product of a matrix and a vector, we get

x =
a0 a1 a2
a1 a0 a3
A=
a 2 a 3 a 0
a3 a2 a1

A O
x, with
C A
c0 c1 c2 c3
a3
a2
, C = c1 c0 c3 c2 ,
a1
c2 c3 c0 c1
a0
c3 c2 c1 c0
where x = (x0 , , x7 )T and O is the 4 4 zero matrix. In the same

way we can determine the matrix representation of the product
XA = (a0 x0 a1 x1 a2 x2 a3 x3 ) + (a0 x1 + a1 x0 a2 x3 + a3 x2 )i
+ (a0 x2 + a2 x0 + a1 x3 a3 x1 )j + (a0 x3 a1 x2 + a2 x1 + a3 x0 )k
+ (a0 x4 a1 x5 a2 x6 a3 x7 + c0 x0 c1 x1 c2 x2 c3 x3 )
+ (a0 x5 + a1 x4 a2 x7 + a3 x6 + c0 x1 + c1 x0 c2 x3 + c3 x2 )i
+ (a0 x6 + a2 x4 + a1 x7 a3 x5 + c0 x2 + c2 x0 + c1 x3 c3 x1 )j
+ (a0 x7 a1 x6 + a2 x5 + a3 x4 + c0 x3 c1 x2 + c2 x1 + c3 x0 )k.
The corresponding product of a matrix and a vector has the form

x =
A
C

O
x, with
A
a0 a1 a2 a3
c0 c1 c2 c3
a1 a0 a3 a2
c1 c0 c3 c2
A=
a2 a3 a0 a1 , C = c2 c3 c0 c1 ,
a3 a2 a1 a0
c3 c2 c1 c0
13
(1.9)
where x = (x0 , , x7 )T and O is the 4 4 zero matrix, cf. [54].
1.1.3 Dual Orthogonal Matrices

In the previous section we dened dual vectors. Naturally, we can
also dene matrices with dual entries. The set of matrices with n
rows, m columns, and dual entries shall be denoted by Dnm . We
introduce the concept of dual orthogonal matrices that can be used
to apply spatial displacements to lines with the help of a dual matrix
multiplication. Therefore, we follow [57] and examine dual orthogonal
3 3 matrices. A dual orthogonal matrix O D33 is a matrix such
that
OOT = OT O = I,
where I denotes the 3 3 identity matrix. The displacement of a line

L can be described by a dual orthogonal matrix, see [7] or [62]. With
dual orthogonal matrices we can write the displacement applied to
a line L that is represented as dual unit vector as matrix vector
product
= O with O = R + TR.
The real 3 3 matrix T is skew-symmetric and describes the linear
mapping x t x and R SO(3) is a 3 3 rotation matrix. For an
orthogonal dual matrix O and two arbitrary dual unit vectors v , w
we have v = Ov , w = Ow and
v , w
= Ov , Ow
= vT OT Ow = vT w = v , w
.
Hence, the dual scalar product is invariant with respect to the action of orthogonal dual matrices. Furthermore, the group of dual
orthogonal matrices is a Lie group. Its Lie algebra is the set of skewsymmetric dual matrices. In [57] the authors used this for motion
interpolation purposes. Ruled surfaces are interpreted as curves on
14
the dual unit sphere that we will introduce later on. Therefore, a generalization of the Rodriguez formula for dual skew symmetric 3 3
matrices is used:
O = exp g = I + sin g + (cos 1)g 2 ,
(1.10)
where = + d is a dual number called the dual angle, I the 3 3

identity matrix, and g a 3 3 skew symmetric matrix. It contains the
angle of the rotation and the magnitude d of the translation. With
Eq. (1.10) it is possible to obtain a spatial displacement that transforms one line into another. For this purpose we need the dual angle
between both lines and the common normal that can be computed
with Eq. (1.5). Furthermore, we need the adjoint mapping
ad : D3 g D33
0
g3
1 2 3 T
3
= (g , g , g ) g
0
g2 g1
(1.11)
g2
g1
=g
that maps dual vectors to dual skew-symmetric matrices. With the

help of this mapping we can parametrise the displacement that maps
one line into another. Let v , w be two dual unit vectors and let
v , w
= cos , = + d be the dual angle between them. With
the dual arccosine function we can compute
= cos1 () d sin1 (),
see right before remark 1.1. With the ad mapping Eq. (1.11) and the
dual Rodriguez formula (Eq. (1.10)) we write
O(t) = exp( (t))ad(n ) = I + sin( (t))ad(n ) + (cos( (t)) 1)(ad(n ))2 ,
where
n =
v w
v w
is the normalized dual vector representation of the common normal

of v (cf. [38]) and w (t) = t . The parameter t varies in the interval
t [0, 1]. Now we can apply this motion to the initial line v
(t) = O(t)v .
1.2 Point Models for Lines and Displacements
15
For t = 0 we obtain O(0) = I and the resulting line is v itself. For

t = 1 we obtain O(1)v = w . The resulting ruled surfaces is a helicoid
and the corresponding motion is a helical motion, see [57].
Remark 1.2. Helicoids correspond to great circles on Studys sphere,
see [57]. Therefore, helicoids can be interpreted as a special type of
geodesic curves on the dual sphere, see section 1.18.

First of all, a quotation from [56] is presented.
Working in a geometric point model enables better understanding and a simple interpretation of various objects
of line space. The design of ecient algorithms involving
lines is greatly simplied if it based on the right geometric
model.
Sometimes it seems confusing to introduce high dimensional spaces
as model spaces, but if things get easier these methods help.
1.2.1 Kleins Quadric

Lines in three-dimensional space form a four-dimensional manifold
called Kleins quadric, or Pl
uckers quadric. It is a special Grassmann
variety. In fact, every straight line in P3 (R) is mapped to a point in
ve-dimensional projective space P5 (R), see [7]. To make this explicit,
we think of R4 with its standard basis as model for P3 (R). A line is
spanned by two dierent points X = xR and Y = yR with homogeneous linearly independent coordinate vectors x = (x0 , x1 , x2 , x3 )T and
y = (y0 , y1 , y2 , y3 )T .
Denition 1.4. For two distinct points X = xR = (x0 , x1 , x2 , x3 )T R, Y =
yR = (y0 , y1 , y2 , y3 )T R P3 (R) we dene the Pl
ucker coordinates of the
line spanned by X and Y as:

xi xj

.
p = (p01 : p02 : p03 : p23 : p31 : p12 ), with pij =
yi yj
16
The mapping : L3 P5 (R), where L3 is the set of lines of P3 (R),

that maps each line L L3 to a point P = (p01 : . . . : p12 ) of P5 (R) is
called the Klein mapping.
Only those points of P5 (R) correspond to a line in P3 (R) that are
contained in Kleins quadric M24 . Its equation is derived by expanding
the determinant det(x, y, x, y) by complementary 2 2 minors:

x0

y0

x0

y0
x1
y1
x1
y1
x2
y2
x2
y2

x3

y3 x0 x1 x2 x3 x0 x2 x3 x1 x0 x3 x1 x2
+
+
= 0,
=
x3 y0 y1 y2 y3 y0 y2 y3 y1 y0 y3 y1 y2
y3
which yields
M24 : p01 p23 + p02 p31 + p03 p12 = 0.
(1.12)
Eq. (1.12) can be reformulated as

T
x Qx = 0, with Q =
O
1
2I
1
2I

, x = (x0 , x1 , x2 , x3 , x4 , x5 )T ,
(1.13)
where O is the 3 3 zero matrix and I is 3 3 identity matrix. We

have a one-to-one correspondence between lines in three-dimensional
projective space and points on this quadric. The polarity of Kleins
quadric can be expressed in matrix form by Q. When working
in ve-dimensional projective space we use and postx notation.
Moreover, the polarity of Kleins quadric is regular, since
det Q =
1
= 0.
64
The bilinear form induced by this polarity is denoted by and dened

by
(x, x) := xT Qx.
Two lines L1 and L2 given in Pl

ucker coordinates l1 R and l2 R intersect
if, and only if, their images under the Klein map satisfy (L1 , L2 ) :=
l1T Ql2 = 0. This means, they have to be conjugate with respect to
Kleins quadric.
17
Direction and Moment We write Pl

ucker coordinates in the form
(l, m). If we interpret l and m as vectors in R3 they are orthogonal
with respect to the standard scalar product and satisfy the Pl

ucker
condition (1.12). From Def. 1.4 we see that m = x y with x =
(x1 , x2 , x3 )T and y = (y1 , y2 , y3 )T . Furthermore, l is the direction of the
line. This follows by setting the homogeneous factors x0 and y0 to 1.
In this case the rst three Pl
ucker coordinates are l = x y .
Note that it is possible to calculate Pl
ucker coordinates from a given
direction vector l R3 and an arbitrary point p on the line by
(l, m) = (l, p l).
This is independent of the choice of p, since if p = p + l, R is

another point on the line the cross product simplies to
(p + l) l = p l + l l = p l.
The vector l R3 is called the direction vector and m R3 the

momentum vector. Moreover, we know from projective geometry
that the ideal point of a line, if not the line itself is ideal, is given by
(0, l)T R.
Subspaces contained in the Quadric An algebraic variety of dimension d and degree k is denoted by a capital letter with subscript k and
superscript d, say Vkd . Thus, Kleins quadric M24 is an example for
an algebraic variety. We are interested in the structure of Kleins
quadric. The quadric is hyperbolic with two-dimensional generator
spaces. Therefore, we take a closer look to one- and two-dimensional
subspaces contained in the quadric. In this paragraph we follow [56].
One-dimensional subspaces of M24 To classify one-dimensional subspaces we give a lemma from [56, section 2.1.3] without proof. Nevertheless, it is easy to verify that:
Lemma 1.1. The Klein mapping takes a pencil of lines to a straight
line contained in M24 . Vice versa, two points X = xR and Y = yR of
M24 correspond to intersecting lines in P3 (R) if, and only if, their span
is contained in M24 .
18
Two-dimensional subspaces of M24 There are two types of twospaces that are completely contained by M24 . The rst one is given
by the image of a eld of lines, i.e., all lines contained by the same
plane in P3 (R). The second type is given by all lines concurrent to the
same point, i.e., a bundle of lines. Now we ask for the intersection of
two two-spaces contained in Kleins quadric. Two-spaces of the same
type always intersect in one point. This means two dierent elds
of lines or two dierent bundles of lines contain one common line.
Two two-spaces of dierent type may have either empty intersection
or they intersect in a line. Thus, a eld of lines and a bundle of
lines may have empty intersection or they intersect in a common
pencil of lines. The type of the two-space, i.e., if it corresponds to a
bundle of lines or a eld of lines can be determined by its intersection
with the Klein image of the eld of ideal lines denoted by P . Note
that P corresponds to a eld of lines, and therefore, to a two-space
entirely contained in Kleins quadric. If a two-space corresponds to a
bundle of lines it has either no point or a whole line in common with
P . A two-space corresponding to a eld of lines has one point in
common with P . Fig. 1.1 shows the correspondence between lines
and two-spaces on Kleins quadric and their pre-images under the
Klein mapping. This gure is inspired by a gure from [56, p. 142].
Subspace Intersections of M24 With Kleins mapping sets of lines
can be studied as sets of points in ve-dimensional projective space.
Clearly, all points on the quadric are self-conjugate with respect to
the polarity . It is natural to ask for the intersections of Kleins
quadric with subspaces. In Kleins model k-space intersections, k 4
with M24 P5 (R) dene special sets of lines in P3 (R). These sets of
lines correspond to so-called linear line manifolds, cf. [56].
Conics on Kleins quadric Two-spaces which are neither tangent to
nor belong to Kleins quadric intersect in a conic section. Tangent
two-space intersections with M24 result in degenerate conics, i.e., two
intersecting lines on Kleins quadric that correspond to two pencils
of lines with one line in common in projective space P3 (R). In this
paragraph we discus the non-degenerate case. Therefore, we choose
three skew lines respectively their Pl
ucker coordinates L1 , L2 and L3

P3 (R)
M24
19
F
P5 (R)
Figure 1.1: Subspaces on Kleins quadric and their geometric interpretation in P3 (R). A pencil of lines P is mapped to the line
P on M24 . A bundle of lines B and a eld of lines F are
mapped to a two-spaces B and F that are contained entirely in M24 .
with (Li , Lj ) = 0, i, j = 1, 2, 3, i = j . These three lines possess image

points spanning a two-space
P12 ( : : ) = L1 + L2 + L3 , , , R.
The set of all points conjugate to this two-space is spanned by P12 =

L1 L2 L3 . This denes the polar two-space of P12 . Since this polar
two-space is the intersection of three tangent hyperplanes we conclude
that the intersection P12 M24 corresponds to the set of all lines in L3
intersecting L1 , L2 , and L3 . Now we can do this construction for three
points of the resulting conic again and get the same statement for the
original points L1 , L2 , and L3 . Furthermore, we derive that these two
sets of lines are dierent sets of generators of the same ruled surface.
We dene:
Denition 1.5. The set of all lines intersecting three given mutually
skew lines L1 , L2 , L3 L3 is called a regulus. A regulus is part of
a ruled quadric. The image of a regulus under the polarity that
denes the second family of generators of a ruled quadric is called the
opposite regulus.
Therefore, every regular conic on Kleins quadric corresponds to a
regulus. Reguli can be distinguished in the Klein model by their
20
intersection with the two-space corresponding to the eld of ideal

lines. In ane space a regulus that carries an ideal line is a hyperbolic
paraboloid. Otherwise if there is no ideal line the regulus corresponds
to a hyperboloid of one sheet.
Linear Line Congruences The two-parametric set of lines corresponding to a three-space intersection of Kleins quadric is called a
linear line congruence. In line geometry we distinguish between hyperbolic, parabolic, and elliptic linear line congruences, see [56].
Linear Line Complexes Linear line complexes correspond to hyperplanar intersections of M24 . Therefore, they dene three-dimensional
line manifolds.
Denition 1.6. A linear line complex is dened by a linear equation
in Pl
ucker coordinates:
(C, L) = c l + c l = 0,
with L = (l, l)T R and C = (c, c)T R, where C is given.

A linear line complex is called singular if the hyperplane is tangent
to M24 else regular. Note that the hyperplane dened by R(c, c)T is
polar to the point C = (c, c)T R. This point is contained in M24 in the
singular case.
Remark 1.3. We can derive any regular (singular) linear line complex
from one single regular (singular) linear line complex by a projective
transformation. With respect to ane geometry one has to distinguish
between two equivalence classes of regular linear line complexes, right
winded and left winded complexes. In a projectively enclosed Euclidean
space each linear complex is characterised by an Euclidean invariant,
the so-called pitch p R (see [56], [66]) and all proper lines of a
regular linear line complex can be interpreted as the path normals of
a helical motion. If (c, c)T R = ((c1 , c2 , c3 ), (c4 , c5 , c6 ))T R is the Pl
ucker
coordinate vector of a linear line complex C , then we get its pitch p by
p=
c c
.
c2
(1.14)
21
The axis of the helicoidal motion corresponding to the linear line

complex C is determined by
A = (a, a
)T R = ((a1 , a2 , a3 ), (a4 , a5 , a6 ))T R = (c, c pc)T R.
In case of a singular linear line complex the pitch equals 0, A = C , and

the complex contains all lines that are path normals of a one-parameter
pure rotation or pure translation group of three-dimensional Euclidean
space E3 .
The condition (C, L) = 0 is exactly the condition for the intersection
of lines in this case. With this knowledge we can identify the points of
P5 (R) that are not contained by M24 with regular linear line complexes.
The points on M24 correspond to singular linear line complexes. This
is called the extended Klein mapping , see [66].
1.2.2 Studys Quadric

The six-dimensional group of Euclidean displacements in threedimensional space can be mapped to a special hyperquadric S26 in
P7 (R) which is usually referred to as Studys quadric. Each displacement is represented by a point on this quadric, but not every point on
S26 corresponds to a displacement. There exists a three-dimensional
generator space V 3 S26 whose points do not correspond to displacements. If we denote a point P P7 (R) by P = (a0 , . . . , a3 , c0 , . . . , c3 )T R,
this exceptional space V 3 is given by a0 = a1 = a2 = a3 = 0. Therefore, the point set S26 \V 3 is the image of the Euclidean displacements.
Thus, the image space is a sliced quadric S26 \V 3 , a pseudo algebraic
variety, so to say. Hence, we have a bijective mapping
: SE(E) S26 \V 3 P7 (R),
SE(3) A = (a0 , a1 , a2 , a3 , c0 , c1 , c2 , c3 )T R.
Study himself [61] gave a method to compute the Study coordinates.

Let
m00
m10
M=
m20
m30
0
m11
m21
m31
0
m12
m22
m32
0
m13
m23
m33
22
be a homogeneous transformation matrix of an Euclidean displacement, see Eq. (1.1). The coordinates of a point on S26 \V 3 corresponding to this displacement can now be determined by the relations:
(1.15)
a 0 : a1 : a2 : a3
= m00 +m11 +m22 +m33 : m23 m32 : m31 m13 : m12 m21 ,
= m23 m32 : m00 +m11 m22 m33 : m12 +m21 : m31 +m13 ,
= m31 m13 : m12 +m21 : m00 m11 +m22 m33 : m23 +m31 ,
= m12 m21 : m31 +m13 : m23 +m32 : m00 m11 m22 +m33 ,
and
2m00 c0 = a1 m10 + a2 m20 + a3 m30 ,
(1.16)
2m00 c1 = a0 m10 + a3 m20 a2 m30 ,

2m00 c2 = a3 m10 a0 m20 + a1 m30 ,
2m00 c3 = a2 m10 a1 m20 a0 m30 .
These relations can be found in [23] or [61]. Note that each equation
of Eqs. (1.15) is important because the ratio could get (0 : 0 : 0 : 0)
in one equation, but not in all equations, what is not allowed since
the point would lie in the exceptional space V 3 . The entries of the
matrix representing the displacement can be calculated by
m00 = a20 + a21 + a22 + a22 ,
m10 = 2(a2 c3 a3 c2 a0 c1 + a1 c0 ),
m11 = a20 + a21 a22 a23 ,
m12 = 2(a1 a2 + a0 a3 ),
m13 = 2(a3 a1 a0 a2 ),
m20 = 2(a3 c1 a1 c3 a0 c2 + a2 c0 ),
m21 = 2(a1 a2 a0 a3 ),
m22 = a20 a21 + a22 a23 ,
m23 = 2(a2 a3 + a0 a1 ),
m30 = 2(a1 c2 a2 c1 a0 c3 + a3 c0 ),
m31 = 2(a3 a1 + a0 a2 ),
m32 = 2(a2 a3 a0 a1 ),
m33 =
a20
a21
a22
(1.17)
a23 .
The coordinates of all points on S26 satisfy the equation

S26 : a0 c0 + a1 c1 + a2 c2 + a3 c3 = 0.
This denes a hyperbolic quadric of maximum index three in

P7 (R) that carries two families of three-dimensional generator spaces
each being a projective three-dimensional space.
23
Subspaces contained in S26 A discussion of linear subspaces contained entirely in Studys quadric can be found in [4]. The maximal
possible dimension of subspaces contained in a six-dimensional quadric equals three. In fact S26 carries two three-parameter families of
three-dimensional subspaces. To classify the subspaces that are entirely contained in S26 we follow [24]. First we observe that a point
P S26 corresponds to the identity, a rotation, or a translation if,
and only if, c0 = 0, see [25]. This follows directly with Eq. (1.8).
Thus, all rotations and translations are contained in the hyperplane
x4 = 0, that is the tangent hyperplane at the point PI = (1, 0, . . . , 0)T R,
i.e., the point on S26 corresponding to the image (e) of the identity
element e SE(3) under the Study mapping . All points that are
conjugate to PI are contained in a quadratic cone with equation
C0 : x1 x5 + x2 x6 + x3 x7 = 0.
This equation can also be interpreted as the equation of M24 P5 (R)

after changing the indices, see [59, Chapter 11].
Lines within S26 Let A0 , A1 be two dierent points on S26 \V 3 . It is
possible to nd a collineation of the form (1.9) that maps A0 to the
identity PI and A1 to some point on the cone C0 . Since the mapping
preserves the exceptional generator there are two possibilities. The
image P11 of the line P11 = [A0 , A1 ] under has one or no point
in common with V 3 . With Eq. (1.8) we see that if the line has a
point in common with V 3 it corresponds to a one-parameter group
of translations, otherwise it corresponds to a one-parameter group
of rotations. Thus, a general line on Studys quadric corresponds to
the composition of a xed displacement 1 with a one-parameter
rotation or translation group.
Two-Spaces within S26 We can apply the same procedure that we
used for lines on the quadric. Following [24] we dene
Denition 1.7. (a) Let P12 be a plane in P3 (R) and O P12 be a point.
We call the two-parametric set of rotations with axes corresponding
to the pencil of lines P12 with vertex O a pencil of rotations. If P12
is a proper plane we call it a pencil of rotations with proper or
24

improper vertex depending on whether O is proper or improper.
If P12 is an improper plane, we speak of a pencil of rotations with
improper axes, i.e., a pencil of translations.
(b) The three-parametric set of all rotations with axes in a xed

plane P12 P3 (R) is called a eld of rotations and P12 is called its
supporting plane.
(c) The three-parametric set of rotations with axes through a xed
point O P3 (R) is called a bundle of rotations and O is called its
vertex, i.e., a coset of SO(3).
For every projective two-space spanned by three points P12 =
[A0 , A1 , A2 ] it is possible to nd a collineation of the form (1.9) with
A0 = PI . This results in three dierent cases for the intersection of
P12 with V 3 . If the intersection is empty P12 corresponds to a pencil
of rotations. If the intersection is a single point P12 corresponds to
a pencil of rotations with improper vertex, and if the intersection is
a whole line P12 corresponds to a pencil of rotations with improper
axes.
Three-Spaces within S26 There are two dierent types of threespaces that are contained entirely in Studys quadric. They are distinguished by their intersection with the exceptional generator V 3 .
Generators that do not intersect V 3 correspond to cosets of SO(3),
i.e., the composition of a xed displacement with SO(3) respectively
to bundles of rotations. This follows with a collineation of type (1.9)
that maps one point from the three-space to PI . If the intersection
of the three-space with the exceptional generator is a single point,
the corresponding displacements are contained by the composition
of a xed displacement with all displacements of a eld of rotations
with axes in a proper plane. If the intersection of the three-spaces
with V 3 is a line, the set of displacements is contained in the composition of a xed displacement with all displacements of a bundle
of rotations with improper vertex, i.e., a coset of SE(2). The last
case that may occur is if the intersection of the generator with the
exceptional generator is a two-space. In this case the corresponding
set of displacements is the composition of a xed displacement with
all displacements of a eld of rotations with improper axes, i.e., the
25
group of all translations. For proofs and a complete discussion we

refer to [24].
1.2.3 Studys Sphere

Another point model for lines or for Euclidean displacements is
Studys sphere, see [29, 56, 65]. Studys sphere is a point model
for spears, i.e., oriented lines. Each line in Euclidean space carries
two dierent spears. It can be oriented in two dierent ways. We
consider a dual vector v = v + v D3 . The canonical scalar product
of this dual vector with itself results in
v , v
= v, v
+ 2 v, v
.
(1.18)
If the dual vector is built from Pl

ucker coordinates (v, v) of a line in
three-dimensional projective space P3 (R), the real part contains the
direction and the dual part the momentum of a line. Then, the dual
part of Eq. (1.18) vanishes. Thus, the value of the scalar product is
a real number. Furthermore, we assume that the direction vector of
the line is normalized. In this case the scalar product (1.18) is equal
to one and we call a dual vector v with v , v
= 1, a normalized
dual vector or dual unit vector. Therefore, we can identify the set of
oriented lines in R3 with points of the dual unit sphere

2
SD
:= x D3 | x , x
= 1 .
Denition 1.8. The mapping that maps an oriented line L of Euclidean three-space to the dual vector v = v + v , where (v, v) are the
normalized Pl
ucker coordinates of L, is called the Study mapping. Its
image space is a model of the set of spears of the three-dimensional
Euclidean space and is called Studys sphere.
Remark 1.4. For dual unit quaternions the same construction can
be applied. In this case we get the set of all dual unit vectors of D4 .
Note, that the direction of the Euler coordinates denes the direction
of the displacement. We have to dier between two displacements
corresponding to dierent oriented Euler coordinates, and therefore,
we consider oriented displacements.
26
1.3 Geometric Algebras, Cliord Algebras

Geometric algebras are special Cliord algebras over the eld of real
numbers. A general introduction is given in [22]. Cliord algebras
are associative algebras that generalize complex numbers, Hamiltons
quaternions, and biquaternions. Geometric algebras, abbreviated by
GA, nd applications in computer graphics [19] robotics [28, 59, 50] ,
physics [36], and a lot of other disciplines. Here, we focus on their application in kinematics of Cayley-Klein geometries and on conformal
geometric algebras, and its generalization. The great advantage of
GA is that geometric entities such as points, lines, and planes can be
described as elements of the algebra. Furthermore, transformations
can also be described as special elements of the algebra and the action
of the transformations applied to a geometric entity is realized by an
algebra operation. A complete treatise of this topic is o the scope
of this work, but we try to give enough references for the interested
reader.
1.3.1 Denition of a Geometric Algebra

We start with a nite-dimensional real vector space V = Rn equipped
with a quadratic form : V R. The pair (V, ) is called a quadratic
space. The matrix of the quadratic form is denoted by (Qij )i,j with
1 j, j n. Therefore, (xi , xj ) = Qi,j for some basis vectors xi and
xj and we denote the quadratic form by its matrix representation Q.
The algebra is dened by the relations
xi xj + xj xi = 2Qij ,
1 i, j n.
(1.19)
Usually, the corresponding Cliord algebra is denoted by C(V, Q). We

shall use another notation. According to Sylvesters law of inertia, we
can always nd a basis {e1 , . . . , en } of V such that e2i {1, 1, 0}. This
basis is called the standard basis of the Cliord algebra. The number
of basis vectors that square to (1, 1, 0) is denoted by the signature
(p, q, r). Therefore, we use the notation C(p,q,r) instead of C(V, ). If
r = 0 we call the geometric algebra degenerate. An n-dimensional
real vector space equipped with a non-degenerate quadratic form of
27
signature (p, q, 0), n = p + q is denoted by R(p,q) . Furthermore, the

relations (1.19) become
ei ej + ej ei = 0,
i = j,
(1.20)
if we use Sylvesters law of inertia. For details we refer to [59]. Note

that Eq. (1.20) also shows that two basis elements anti-commute,
i.e.,
ei ej = ej ei , i = j.
(1.21)
In the remainder of this thesis we abbreviate the product of basis
elements with lists and call them generators
e12...k := e1 e2 . . . ek , with 0 k n.
In order to bring light into this denition, we give an example.

Example 1.1. We present the quaternions as elements of a Cliord
algebra. Therefore, we construct the Cliord algebra C(0,2,0) . This
means we have two basis vectors, that square to 1. The eld R is
embedded with the basis element e0 and a element of C(0,2,0) can be
written as:
a0 e0 + a1 e1 + a2 e2 + a12 e12 .
The basis {e0 , e1 , e2 , e12 } for C(0,2,0) is called a standard basis for this
geometric algebra. As we already know from the denition e21 = e22 =
1. The square of e12 evaluates to
e212 = e12 e12 = e12 e21 = e11 = 1.
If we now identify e1 with i, e2 with j, and e12 with k, we have an

isomorphism between C(0,2,0) and H. Of course the multiplication
rules for the quaternion units i, j and k have to be veried. This is
left to the reader as an exercise.
Denition 1.9. For every Cliord algebra over an n-dimensional vector space a general element is the sum of scalars, bivectors, trivectors
up to pseudoscalars, it is called a multivector and reads
a0 e0 +a1 e1 +an en +a12 e12 +. . .+an(n1) en(n1) +. . .+a1...n e1...n .
28
1.3.2 Properties of Cliord Algebras

A Cliord algebra is a graded algebra Every basis element e...
can be transformed to a basis element of the form eij...k , where i <
j < . . . < k . Each swap of two elements causes a multiplication with
1. If we arrive at eii we can insert 1, or 1, or 0 as prescribed by
the signature (p, q, r). The 2n monomials
ei1 i2 ...ik ,
0kn
form the standard basis of the Cliord algebra.

Denition 1.10. An algebra element that is the product of invertible
grade-1 elements is called a versor.
The Cliord algebra C(p,q,r) is the direct sum
i
n

i
V of all exterior
i=0
products V of V of any grade 0 i n where ek1 ...ki , k1 < . . . < ki

1
form a basis of i V . Especially 0 V is the scalar part
R and
V
2
is isomorphic to the vector space
V . Elements from
V are called
n
V are
bivectors and elements from
called pseudoscalars. The dii
V is ni . Thus, the dimension of the
mension of each subspace

Cliord algebra is ni=0 (dim i V ) = ni=0 ni = 2n . A Cliord algebra is called universal if dim C(p,q,r) = 2n , n = p + q + r. In [22,
p. 89] it is shown that there always exists an universal Cliord algebra for a given quadratic space (V, ). We restrict ourselves to the
standard basis. Furthermore, the Cliord algebra C(p,q,r) possesses a
Z2 -grading, i.e., it is the direct sum of an even and an odd part
C(p,q,r) =
C+
(p,q,r)
C
(p,q,r)
:=
n

i
i=0
i even
n

i
V.
(1.22)
i=0
i odd
Note that the even part C+

(p,q,r) is always a subalgebra, because the
product of two even-graded monomials must be even-graded and the
generators cancel only in pairs. The dimension of the even subalgebra
is 2n1 . Furthermore, we have the isomorphism
C(p,q,r)
= C+
(p,q+1,r) .
(1.23)
For details see [55]. To make this isomorphism more clear we give
another example.
29
Example 1.2. As we already know from Ex. 1.1 the Cliord algebra
C(0,2,0) is isomorphic to H. The even part of this algebra is generated
by {e0 , e12 }, and therefore, isomorphic to C. Furthermore, the Cliord
algebra C(0,1,0) is the algebra with one generator squaring to 1.
Therefore, C(0,1,0)
= C. All in all we have C+
(0,2,0) = C(0,1,0) .
Denition 1.11. The center of a ring R is the set of all elements
that commute with all other elements
C(R) := {c R | cx = xc for all a R} .
We are interested in the center of a Cliord algebra, see [22, p.95].

The center of a Cliord algebra C(p,q,r) is
(1) C(C(p,q,r) ) = {e0 + e12...n | , R} if n is odd,
(2) C(C(p,q,r) ) = {e0 | R} if n is even.
For the even part the center is
(3) C(C+
(p,q,r) ) = {e0 | R} if n is odd,
(4) C(C+
(p,q,r) ) = {e0 + e12...n | , R} if n is even.
Cliord Algebra Involution
The Conjugation Every Cliord algebra possesses an anti-automorphism, i.e., an automorphism of the algebra that reverses the order
of factors in a product. We follow [59] and denote the conjugation
by an asterisk. The action on any generator is given by ei = ei .
On scalars it has no eect. If we extend the conjugation by using
linearity to arbitrary algebra elements, we get the formula
(ei1 ei2 . . . eik ) = (1)k eik . . . ei2 ei1 ,
0 i1 < i2 < . . . < ik n.
(1.24)
Example 1.3. We apply the conjugation to an algebra element of

the Cliord algebra C(0,2,0)
= H. The standard basis is given by
{e0 , e1 , e2 , e12 }. The conjugation for the two grade-1 basis elements
e1 and e2 is given via denition by e1 = e1 and e2 = e2 . The
conjugation has no eect on scalars, so e0 = e0 . It remains to extend
30
these observations to the grade-2 basis element e12 . We use the antiinvolution property and the anti-commutativity
e12 = (e1 e2 ) = e2 e1 = (1)2 e2 e1 = e21 = e12 .
Hence, the conjugation of quaternions ts into the concept.

A general grade-1 element in C(p,q,r) has the form
x = x1 e 2 + x 2 e 2 + . . . + xn e n ,
n = p + q + r.
The product with its conjugate element is given by:

xx = x21 x22 . . . x2p + x2p+1 + . . . + x2p+q = (x, x).
(1.25)
If we identify V with 1 C(p,q,r) Eq. (1.25) gives the negative of the

quadratic form (x, x). Furthermore, the square of x 1 C(p,q,r)
results in the scalar product of the vector with itself
xx = x21 + x22 + . . . + x2p x2p+1 . . . x2p+q = (x, x).
The Main Involution Another involution of a Cliord algebra is the

main involution. It is denoted by and dened by
(ei1 ei2 . . . eik ) = (1)k ei1 ei2 . . . eik ,
0 i1 < i2 < . . . < ik n.
(1.26)
The main involution has no eect on the even subalgebra and it commutes with the conjugation. This means for an arbitrary algebra
element x, the equation (x ) = (x) holds.
Cliord Algebra Products We denote grade-1 elements, i.e., vectors, by small gothic letters. Elements of higher grade are denoted
by big gothic letters.
The inner Product The scalar product of two vectors, i.e., grade-1
elements can be written in terms of the geometric product
a b :=
1
(ab + ba).
2
(1.27)
31
A generalization of the
inner product
to blades, see Def. 1.12, can be

found in [36]. For A k V, B l V the generalized inner product
is dened by
A B := [AB]|kl| ,
where []m , m N denotes the grade-m part.

The outer Product The outer or exterior product for two vectors
is given by
1
a b := (ab ba).
(1.28)
2
This product can also be generalized to blades, cf. [36]. For A

l
k
V, B
V the exterior product is dened by
A B := [AB]|k+l| .
The geometric Product With the inner (1.27) and the exterior
product (1.28) the geometric product of vectors can be written in
the following form
ab = a b + a b.
(1.29)
is the k-fold exterior product of vectors
Denition 1.12. A k-blade

or grade-1 elements v 1 V . Therefore, a k-blade can be written as
A = a1 a2 . . . ak =:
k

ai .
i=1
Grade-k elements that are k-blades are called simple or decomposable.

A k-blade that squares to zero is called a null k-blade.
Remark 1.5. Not every grade-k element is also a k-blade. The twoblade e12 + e13 can be written as e1 (e2 + e3 ). As example for a grade-2
element that is not a two-blade we take e12 + e34 .
For treating geometric entities within geometric algebra context the
denition of the inner product null space and its dual the outer product
null space is needed.
32
Denition 1.13. The inner product null space (IPNS) of a k-blade

k
A
V , is dened by

NI(A) :=
1

V :vA=0 .
Moreover, the outer product null space (OPNS) of a k-blade A

is dened by
NO(A) :=
k

1
v
V :vA=0 .
The IPNS and the OPNS of k-blades can be used to describe subspaces of the vector space V , see [53]. Furthermore, we have the
property of the outer and inner product null space
NI(a b) = NI(a) NI(b),
NO(a b) = NO(a) NO(b).
(1.30)
Table 1.1: Cliord algebras C(p,q,0) , see [55].
p 0
1
2
3
4
5
6
7
q
0
R
R
R(2)
C(2)
H(2)
2
H(2)
H(4)
C(8)
C
R(2)
2
R(2)
R(4)
C(4)
H(4)
2
H(4)
H(8)
H
H
C(2) H(2)
R(4) C(4)
2
R(4) R(8)
R(8) 2 R(8)
C(8) R(16)
H(8) C(16)
2
H(8) H(16)
2
H(2)
H(2)
H(4)
C(8)
R(16)
2
R(16)
R(32)
C(32)
C(4)
H(4)
2
H(4)
H(8)
C(16)
R(32)
2
R(32)
R(64)
R(8)
R(8)
C(8)
R(16)
H(8)
C(16)
2
H(8) H(16)
H(16) 2 H(16)
C(32) H(32)
R(64) C(64)
2
R(64) R(128)
2
Table 1.1 shows the matrix representation of the non-degenerate Clifford algebras C(p,q,0) . In this table R(n) denotes the n n-matrices
over R and 2 R(n) denotes R(n) R(n).
33
1.3.3 Pin and Spin Groups

Not every element of a Cliord algebra has an inverse element. In
general there are zero divisors, this means for two non zero elements
A, B C(p,q,r) the geometric product AB is zero.
Example 1.4. A simple example for a zero divisor is given by an
element a = ae1 + ae2 C(1,1,0) , a R. This Cliord algebra has the
standard basis {e0 , e1 , e2 , e12 } with e21 = 1, e22 = 1 and e212 = 1. We
calculate
a2 = (ae1 + ae2 )2 = a2 e21 + a2 e12 + a2 e21 + a2 e22 = 0.
Note that C(1,1,0)

= R22 . We give an isomorphism by its action on
the standard basis, cf. [3].

1
e0

0

0
,
1

e1
1
0

0
,
1

e2
0
1

1
,
0

e12
0
1

1
.
0
The inverse element of a versor V is dened by V1 = NV(V) with

N (V) = VV . The map N : C(p,q,r) C(p,q,r) is called the norm of
the Cliord algebra. For general multivectors M C(p,q,r) inverse
elements exist and are dened by the relation MM1 = M1 M = 1,
but the computation is more dicult. This is discussed in section
1.3.4.
Cliord Group Invertible elements are called units. We denote the
set of units of a Cliord algebra by C
(p,q,r) . For a general ring R, we
denote the set of units by R . With respect to the geometric product
the units form a group. For a given Cliord algebra the Cliord group
is the group

(C(p,q,r) ) := g C
|(g)vg
V
v
V
.
(p,q,r)
(1.31)
It is not obvious that (C(p,q,r) ) is a group. For a proof we refer to

[21]. Elements of the Cliord group are denoted by the small gothic
letter g if not stated otherwise. We are interested in special subgroups
of the group of units.
34
The Pin Group The Pin group is dened by

Pin(p,q,r):= g C(p,q,r) |N (g) = 1, (g)vg V v V .
(1.32)
The action of the so called sandwich operator (g)vg with g Pin(p,q,r)

applied to vectors does not change the scalar product of two vectors.
This can be veried easily by direct calculation. Note thatthe condition N (g) = 1 guarantees that g is a unit. Let a, b 1 V and
g Pin(p, q, r). Furthermore, let a = (g)ag and b = (g)bg . The
scalar product can be expressed in terms of the geometric product as
in Eq. (1.27). We use this equation for the transformed vectors a
and b and nd
1
((g)ag (g)bg + (g)bg (g)ag )
2
1
= (g) [ag (g)b + bg (g)a] g .
2
a b =
The main involution applied to a Pin group element (g) is either g

or g. This follows, because a Pin group element g can be written as
g = g1 . . . gn , where g1 , . . . , gk , k n are grade-1 elements. The main

involution applied to a grade-1 element g1 1 V equals g1 . Therefore, we conclude that g (g) = (g)g is either 1 or 1 depending on
if k is even or odd. Therefore, the term in square brackets is a scalar
and consequently
a b =
1
(g)g [g (g)(ab + ba)] .
2
Thus, we get
1
1
((g)g )(g (g)) (ab + ba) = (ab + ba) = a b.
2
2

Linear transformations of 1 V that preserve distances and angles, i.e.,
the scalar product, are elements of the orthogonal group O(p, q, r). In
fact, Pin(p, q, r) is a double cover of the orthogonal group O(p, q, r),
a b =
see [21].
35
Remark 1.6. Usually, the sandwich operator for elements of the

Cliord group (p,q,r) is dened by (g)vg1 . For elements of the Pin
group the operator simplies because for g Pin(p,q,r) we have
g1 =
g
= g .
gg
In the homogeneous Cliord Algebra models we use g because multiplication with a homogeneous factors does not change the geometric
meaning of the sandwich operator.

Remark 1.7. The mapping v (g)vg for g Pin(p, q, r) 1 V is

a reection in the hyperplane perpendicular to NO(v) with respect to
through the origin. The composition of two reections is again an
element of the Pin group.
The Spin Group The second important subgroup of the group of
units is the Spin group dened via

Spin(p,q,r) : = g C+
(p,q,r) |N (g) = 1, gvg V v V
(1.33)
= Pin(p,q,r) C+
(p,q,r) .
Note that the main involution has no eect on the even subalgebra,
and therefore, can be neglected.
Remark 1.8. The Spin group is a subgroup of the Pin group. Thus,
the scalar product of vectors is preserved under the action of the Spin
group. Spin group elements are generated by pairs of reections, so
they are rotations.
The Spin group for degenerate Cliord algebras as in the case of
Euclidean geometry are semi-direct products of Spin groups for the
non-degenerate part and an additive matrix group, see [15]. Later we
will construct the Cliord algebra for the three-dimensional Euclidean
space.
Remark 1.9. Due to their denitions the Pin and the Spin group
consist of two connected components. The connected components are
36
given by N (g) = 1 and N (g) = 1. We distinguish the two connected

components by

,
Pin+
(p,q,r) := g C(p,q,r) |N (g) = +1, (g)vg V v V
Pin
.
(p,q,r) := g C(p,q,r) |N (g) = 1, (g)vg V v V
For the connected components of the Spin group we use the same
notation.
1.3.4 Matrix Representation of Cliord Algebras

In this section we follow [50]. Any multivector of a Cliord algebra
over an n-dimensional vector space n N can be represented by a
n
vector v R2 . Then the geometric product of two algebra elements
AB = C can be written as product of a matrix and a vector

C = A+ B,

C = B A,

A+ , B ,
(1.34)
where [A+ ] and [B ] are 2n 2n matrices with real entries and A, B, C

n
R2 . The columns of the matrix [A+ ] are dened by the products
ae0 , ae1 , . . . , aen , ae12 , . . . ae(n1)n , . . . , ae1...n .
Note that these columns are ordered from right to left to match the
chosen ordering of the vector representation for the Cliord algebra.
To obtain the columns of the matrix [B ] we multiply b from the left
side with each basis element
e0 b, e1 b, . . . , en b, e12 b, . . . e(n1)n b, . . . , e1...n b
again with the same ordering.

Remark 1.10. With the matrix representation it is possible to compute the inverse element for an arbitrary multivector. In order to
achieve this, we express the geometric product as product of a matrix
with a vector. If the 2n 2n matrix is invertible, the inverse algebra
element corresponds to the inverse matrix. For increasing vector space
dimension this calculation can be extremely expensive.
Remark 1.11. The matrix representation of the dual quaternion
product that was determined in section 1.1.2 is computed in the same
way.
37
1.3.5 Linear Transformation of the Vector Space

The action of the sandwich operator applied to vectors can be written
as a linear transformation of the vector space. Therefore, we rewrite
the sandwich product of an arbitrary versor g C(p,q,r) applied to a

vector v 1 V as product of a matrix with a vector
(g)vg = v K v = v ,
where v, v Rn and K Rnn is a linear mapping. This procedure

can be applied to every subspace of the Cliord Algebra C(p,q,r) . We
illustrate this at hand of the following example.
Example 1.5. The Cliord algebra C(0,2,0)
= H, see Ex. 1.1 serves
as an example. Hence, we take two general elements of the form
g = a0 e0 + a1 e2 + a2 e2 + a3 e12 ,
v = x0 e0 + x1 e2 + x2 e2 + x3 e12 ,
where g is a versor and v is arbitrary. We apply the sandwich operator

to v:
(g)vg = ((a20 a22 +a23 a21 )x0 +2(a0 a1 +a3 a2 )x1 +2(a0 a2 a1 a3 )x2 )e0
+ (2(a3 a2 a0 a1 )x0 +(a20 a21 +a22 a23 )x1 2(a0 a3 +a1 a2 )x2 )e1
+ (2(a1 a3 x0 +a0 a2 )x0 +2(a0 a3 a1 a2 )x1
+ (a21 a22 a23 +a20 )x2 )e2 +(a23 +a20 +a21 +a22 )x3 e12 .
If we now write this operation as a product of a matrix with a vector,

we obtain v = K v , with v = (x0 , x1 , x2 , x3 )T R4 and
a23 a21 +a20 a22 2a0 a1 +2a3 a2 2a0 a2 2a1 a3

0
2a0 a1 +2a3 a2 a22 a21 a23 +a20 2a0 a3 2a1 a2

0
.
K=
2a1 a3 2a0 a2 2a1 a2 +2a0 a3 a21 a22 a23 +a20

0
2
2
2
2
0
0
0
a0 +a1 +a2 +a3
If g P in(0,2,0) it yields a20 + a21 + a22 + a23 = 1. The matrix K represents

a rotation, where x3 contains the homogeneous component, see Eq.
(1.4). This means that in the projective space P3 (R) the hyperplane
x3 = 0 is identied as ideal plane and that the ane part is obtained
as P3 (R3 1) in P3 (R3 R), see [31].
38
1.4 The Homogeneous Model

In this section we follow [19, 27] and [48] and construct the Clifford algebra over a projective space Pn (R) modelled by an (n + 1)dimensional vector space V endowed with a quadratic form. The
resulting model is called the homogeneous model. In this model, it
is possible to describe ane subspaces, i.e., translated subspaces by
algebra elements respectively their IPNS or OPNS. Hence, we disengage ourselves from the origin. Points of Pn (R) are represented by
vectors respectively one-dimensional
subspaces of V . This enables us

to interpret the grade-1 subspace 1 V as a copy of a projective space.
The homogeneous model is nothing else but the geometric algebra
representation of homogeneous coordinates from projective geometry.
Meet and join operations are now available for ane subspaces.
1.4.1 The Construction

As suggested in the preliminaries of this section we use a vector space
of dimension n+1 to dene the geometric algebra corresponding to the
homogeneous model for an n-dimensional vector space. This means
we add one generator e to the standard basis of a given vector space
V . Later in chapter 3, we see that the choice of the square (e2 = 1, 1,
or 0) will determine a Cayley-Klein geometry. Furthermore, we dene
this direction to be orthogonal to all other directions of V : ei e =
0 for all i = 1, . . . , n with respect to the quadratic form . Points
of the space are then represented by grade-1 elements, i.e., vectors,
directed lines by two-blades and so on. The occurring geometric
entities represented by subspaces are described in the Grassmann
or exterior algebra. Let P = (p1 , . . . , pn )T V be a point. It has
homogeneous coordinates P = (1, p1 , . . . , pn )T R and as an element of
the algebra, it reads p = e + p1 e1 + . . . + pn en . With the wedge
product it is possible to dene algebra elements corresponding to the
join of other algebra elements. A line is the wedge product of two
points that are represented by vectors in the homogeneous Cliord
algebra model, and a plane can be described as the wedge product
of three non-collinear points. These subspaces can be described with
Grassmann coordinates.
39
1.4.2 Exterior Algebra

The exterior algebra or Grassmann algebra V over a real vector

space V assigns points to subspaces of projective spaces. Therefore,
we take a closer look on the wedge product that we have already met
in section 1.3. In the following we assume that one-dimensional subspaces in V represent points of the projective space P(V ). We denote
the vector space with V and its standard basis with {e1 , . . . , en }.
For an n-dimensional real vector space V the GrassDenition 1.14.

mann algebra V is a Cliord algebra with fully degenerate quadratic
Q of the quadratic form is the n n zero
form. This means the matrix

matrix. Multiplication in V is denoted by the wedge product and
sometimes referred to as exterior product.
From the properties of the geometric product we have for X, Y
X Y + Y X = 0.
The exterior product is an alternating multilinear operation. In fact,

the exterior product is the geometric product for this special algebra.
For vectors v1 , v2 , . . . vk it is zero if, and only if, they are linearly
dependent. If {e1 , . . . , en } isa basis of V , then ei1 i2 ...ik , 0 k n, i1 <
i2 < . . . < in is a basis for V , for k = 0 we dene ei0 := e0 , i.e., the
scalar part. Furthermore, the Grassmann algebra possesses the same
graduation as every universal Cliord algebra. We have the following
direct sum

0
1
n
V
V ...
V.

The dimension of k V equals nk , and therefore, the dimension of

0
V equals 2n . The subspace
V is identied with the real numbers,
1
i.e., scalars. Elements of
V are identied with elements of V re
spectively Pn (R). Elements from k V are called k-vectors. We recall
V =
a theorem from [56] without proof.

Theorem 1.1. For S k V the set of all a 1 V such that S a = 0

is a linear subspace denoted by L(S). If S is non-zero and simple with
S = v1 . . . vk , then L(S) equals the span [v1 , . . . , vk ]. For simple
S1 , S2 the spaces L(S1 ) and L(S2 ) are equal if, and only if, S1 is a
scalar multiple of S2 .
40
Th. 1.1 gives a one-to-one correspondence

between linear subspaces
k
L(S) and one-dimensional subspaces in
V spanned by simple elements. We dene:
Denition 1.15. For a k-dimensional linear subspace S spanned by
[v1 , . . . , vk ] we compute
v 1 v 2 . . . vk =
vi1 ...ik ei1 . . . eek .
i1 <...<ik
The coecients vi1 ...ik are called Grassmann coordinates of S .

Note that by Th. 1.1 Grassmann coordinates are dened only up
to a constant scalar factor. Therefore, Grassmann coordinates can
be interpreted
as homogeneous coordinates. Furthermore, not every

element of k V is simple, i.e., decomposable. Therefore, we need
more relations to describe simple elements. Here we do not investigate
these relations, the interested reader is referred to [56]. To clarify
these concepts, we give an example.
Example 1.6. Let V = R4 be the vector space model for P3 (R).
Moreover, two points P1 = (1, 3, 2, 0)T R and P2 = (1, 0, 2, 1)T R are
given in Grassmann algebra representation by p1 = e0 + 3e1 + e2 and
p2 = e0 +2e2 +e3 . Now we calculate the exterior product
(e0 + 3e1 + e2 ) (e0 + 2e2 + e3 ) = 3(e0 e1 ) + (e0 e2 ) + (e0 e3 )
+6(e1 e2 ) + 3(e1 e3 ) + (e2 e3 ).
These Grassmann coordinates can now be listed as (3, 1, 1, 6, 3, 1).

The coordinates are Pl
ucker coordinates if we order them in the right
way and use the historic convention that the coecient of e3 e1 is
ucker coordinates
used. The line spanned by P1 and P2 has the Pl
(3, 1, 1, 1 3, 6). Here, the Pl
ucker condition can be checked easily.
To test if an arbitrary point P3 with algebra representation p3 lies on
the line [P1 , P2 ] we just have to apply the exterior product one more
time. If the result is zero the point is incident with the line.
The example above also shows that in the case of Pl
ucker coordinates
simple elements satisfy an additional condition. In this case it is the
condition that they all have to lie on a hyperquadric M24 P5 (R), see
section 1.2.1.
41
Dual Exterior Algebra In some cases it is necessary to construct a

homogeneous Cliord algebra model with the use of the dual exterior
algebra. This is the reason why we give a short sketch for the construction of the dual of an exterior algebra. All that has to be done is to
do the same construction on the dual vector space V , i.e., the vector
space of all linear-forms on V . The dual exterior algebra is then the
algebra
of the k-multilinear forms and it is naturally isomorphic to

V again. In V we can also nd the subspace structure of Pn (R)

again. The dierence is that elements from 1 V represent hyperfrom
planes, elements from 2 V hyper lines. Furthermore, elements
n1
V represent points and so on. All in all the structure of V

is isomorphic to the structure of V turned on its head. Later we
will dene this isomorphism and use it to express the meet operation
with the wedge product.
1.4.3 Homogeneous Model via projective Grassmann

Algebra
From section 1.3 we know how to dene a Cliord algebra. A Cliord
algebra and
a Grassmann algebra have the same graduation. If we

now equip 1 V with an inner product, we dene a geometric product
way we could dene a geometric product
on vectors
for V . In a dual
n1

V space for V . In [28] this construction is presented
on the
in detail. The isomorphism presented now can also be found there.
In the literature duality in a Cliord algebra is often dened through
multiplication with the pseudoscalar. When we deal with degenerate
Cliord algebras this denition is not adequate. Therefore, we dene
duality by the so called Poincare isomorphism. This isomorphism
can be used to dene projective meet operations.
Denition 1.16. The isomorphism
J : C(p,q,r) C(p,q,r) ,
ei1 ...ik eI\{i
1 ,...,ik }
, 0 i1 < . . . < ik n,
where I is the ordered set {0, . . . , n} and n = p + q + r the dimension

of the vector space is called the Poincare duality. Grade-k elements
are mapped to grade-(n k) elements.
42
Example 1.7. If we apply this mapping to a grade-2 element p =

x0 e12 +x1 e23 + x2 e13 of C(2,0,1) we get a grade-1 element J(p) = x0 e3 +
x1 e1 +x2 e2 . The scalar product (1.27) results in
J(p) J(p) = x21 +x22 .
This expression belongs to the norm of vectors in the Euclidean plane.

Note, that in this case
J(p)J(p) = J(p) J(p),
since the exterior product of an element with itself vanishes. It can

be interpreted as the squared distance between the origin and the
hyperplane dened by NO(J(p)). Therefore, the geometric entity corresponding to the outer product null space of J(p) can be interpreted
as line, i.e., a hyperplane in P2 (R).
This example shows that for Euclidean geometry we have to dene the
inner product for hyperplanes and therefore, we have to use the dual
Grassmann algebra. A full description of the homogeneous model
lies beyond the scope of this work. The interested reader is invited
to consult the cited references. In chapter 3 we take a closer look on
special homogeneous Cliord algebra models to construct kinematic
mappings.
1.5 The Conformal Model

The next step was taken by D. Hestenes, see [34]. He designed a
geometric algebra that is capable to describe conformal transformations with its Pin group. Furthermore, a lot more geometric objects
nd their analogue as algebra elements. There is a huge amount of
literature on conformal geometric algebra, abbreviated by CGA, and
the interested reader is referred to [18, 19, 35, 34] or [46].
To arrive at the homogeneous model we had to add one extra dimension to get rid of of the origin. In the homogeneous Cliord algebra
model we are able to describe translations by Spin group elements.
The idea to model conformal geometry is to add an extra dimension
with special properties to the homogeneous model.
43
1.5.1 Construction of Conformal Geometric Algebra

We denote the n-dimensional Euclidean space by En . The associated
quadratic form shall be the standard scalar product. The basis vector
space is the Cartesian one with two additional generators e+ and e
squaring to 1 and 1 as prescribed by the index of the corresponding
additional generator. This means that we get a Minkowski algebra
and the quadratic form has signature (n+1, 1, 0). For our considerations it is more convenient to use a dierent basis for the quadratic
space R(n+1,1) , i.e., Rn+2 is equipped with the quadratic form . We
introduce the two isotropic vectors
o=
1
(e +e+ ),
2
= (e e+ ).
Note that these two generators square to zero. In geometric algebra

context isotropic vectors are called null vectors. Sometimes in the
literature o is denoted by no, for null vector representing the origin
and by ni, for null vector representing innity . Especially for
n = 3 the multiplication table is given by
o
e1
e2
e3
o e1
0 0
0 1
0 0
0 0
1 0
e2
0
0
1
0
0
e3
0
0
0
1
0
1
0
.
0
0
0
There are also other possibilities to dene the basis of R(4,1) . This
algebra can be thought of as the homogeneous model with one additional restriction in the -direction. The linear subspaces of Rn+1
that model ane subspaces in Rn are intersected with an (n + 1)dimensional paraboloid, called the horosphere, see [20, 35]. The intersection is then projected on the n-dimensional hyperplane spanned
by [e1 , . . . , en ]. The scalar product in conformal geometric algebra
(abbreviated by CGA) helps to give an analogue to the Euclidean distance. Therefore, points are modelled by null vectors, because they
have to satisfy p p = pp = 0, since p p = 0. For a point P Rn the
corresponding algebra element has the form
1
p = o+p+ p2 ,
2
44
where p = p1 e1 + . . . + pn en is a vector in Rn . With this denition it

easy to check that the scalar product of two such elements p and q
evaluates to
1
1
p q = (o+p+ p2 ) (o+q+ q 2 )
2
2
1 2
1 2
1
= q +p q p = (qp)2 .
2
2
2
(1.35)
With Eq. (1.35) we see that points are described by null vectors, since
pp = p2 = 0 and pp = 0. Furthermore, it now becomes clear that the
Euclidean distance is integrated in the algebra by the scalar product.
Note that the o-component can be understood as the homogeneous
factor. This factor can be extracted from a point by
1
p = (o+p+ p2 ) = .
2
Furthermore, a point is called normalized if the homogeneous factor

is equal to 1. We have to remember that if we want to calculate
Euclidean distances with the scalar product. The inner product null
space and the outer product null space with respect to the embedding
is called the geometric inner and geometric outer product null space.
It depends on the embedding of the base space into a higher dimensional space. We denote the embedding by . Especially, non-linear
embeddings give rise to interesting geometric entities. Therefore, we
recall the denition, see [53].
Denition 1.17. The geometric inner product null space (GIPNS)
and dual the geometric outer product null space (GOPNS) of a k
blade A k V is dened as

NIG (A) := x R(r,s) : (x) A = 0 ,

NOG (A) := x R(r,s) : (x) A = 0 .

The mapping : 1 R(r,s)

X, where X
embeds the base space into 1 R(p,q) .
1
R(p,q) , with r + s p + q
Note that we mean the Poincare duality when we talk about duality
in geometric algebra context.
45
Remark 1.12. In three-dimensional conformal geometric algebra the

embedding of three-dimen-sional space E3 is given by
: R3 R(4,1) ,
1
P o+p+ p2 .
2
1.5.2 Blades in CGA

Vectors andtheir geometric Interpretation Points P Rn are emvectors. Moreover, it is clear that there
bedded into 1 R(n+1,1) as null

are also other elements in 1 R(n+1,1) that are not null vectors. Here
we list these elements without discussing how to derive them. In the
grade-1 space we have the following geometric entities, see [19]:
Null vectors p = (o + p + 12 p2 ) representing a point.
Vector without o-component: = n + representing a dual hyperplane. To determine the point set in Rn that is described by we
examine its inner product null space with respect to the description
of points in CGA.

NIG () = (x) = 0 | x R3

1 2
3
= (o+x+ x ) (n+) = 0 | x R
2

= x n = 0 | x R3 .
(1.36)
All x NIG () given by Eq. (1.36) are contained by a hyperplane
to the origin. Note that

with normal vector n and distance n
is incident with every dual hyperplane.
A general vector s = (c 12 2 ) denes a dual hypersphere. This
can be interpreted as the sum of a point, i.e., a null vector and an
additional part = (c + ), where c is the normalized representation of a point. We can calculate the set of points contained in
this object through

NIG () = (x) = 0 | x R3 = (x c+x ) = 0 | x R3

1
2
3
= ( xc ) = 0 | x R .
2
46
Therefore, points contained in NIG () fullll x c2 = 2 . If

is negative we redene = 12 2 . This results in the equation of
a sphere with center c and radius . Note that a positive value of
results in a sphere with real equation and imaginary radius. It
carries no real point.
When speaking about dual objects we have to clarify what is meant
by dualization. In CGA the dual of a blade is the orthogonal complement. For non-degenerate algebras this is usually done by multiplication with the pseudoscalar I = o e1 . . . en , in the degenerate
case we have to use the Poincare duality cf. Def. 1.16.
Rounds and Flats In the homogeneous model lines can be dened
by the exterior product of two points. Furthermore, planes can be
dened by the exterior product of three non-collinear points and so
on. In CGA a point is described by a null vector. The outer product
of two points corresponds to a line in the homogeneous model that
now has to be intersected with the horosphere. The result is the
simplest round, i.e., a pair of points. Let us take a look on the
exterior product of three null vectors corresponding to three noncollinear points a = (a), b = (b) and c = (c), a, b, c R3 :
A = a b c.
The intersection of this three-blade with the horosphere is an ellipse

and the projection to Rn is a circle, except if one of the points is .
In the latter case the image of the ellipse is a line. Note that the
conformal model is produced by compactifying the Euclidean space,
i.e., the ane real space with one point at innity. Therefore, every
at has to contain this single point at innity. To do kinematics we
are interested in the three-dimensional Euclidean space modelled by
the conformal geometric algebra C(4,1,0) . Planes and spheres can be
described by the exterior product of the null vectors corresponding
to four non-concyclic respectively non-collinear points a, b, c, d R3 .
The corresponding null vectors are given by a = (a), b = (b), c =
(c), d = (d) and the exterior product results in
A = a b c d.
47
In the homogeneous model this results in a three-space that has to

be intersected with the horosphere and afterwards projected to R3
spanned by [e1 , e2 , e3 ]. The result is a sphere if non of the points a, b, c,
or d is and a plane if one is . All these objects can be understood
as geometric outer product null spaces of blades. Their dual algebra
elements AI can be understood as geometric inner product null spaces
of blades, where I denotes the pseudoscalar.
Dual Objects Planes and spheres are modelled by decomposable
elements of grade four. Dualization means multiplication with the
pseudoscalar. Therefore, the dual elements are of grade one. Since
the dual element describes the orthogonal complement of the object,
it is clear why we compute AI x = 0, x R3 to get the point set
described by NOG (A) = NIG (AI).
Tangent Blades There are also linear subspaces in the homogeneous
model that are tangent to the horosphere. It is clear that they cannot
represent a point, but they should be interpreted as set of points, since
they touch the horosphere in a specic null vector corresponding to
a point. Furthermore, these objects posses another component that
may be understood as a direction. One simple example of such a blade
is constructed by the outer product of the origin with an arbitrary
blade of the base space
T = o a.
The origin is the only intersection with the horosphere. Therefore, we

think of a as a direction appended to the origin. That is the reason
why we call such blades tangent blades.
Free Blades
A prototype of a free blade has the form

F = a .
Note that it has no o-component. Such a blade is interpreted as a

at at innity. Furthermore, translation has no eect on this blade,
rotation just eects the a-part. Therefore, we interpret these blades
as free direction vectors.
48
1.5.3 Conformal Transformations

The Pin group of C(n+1,1,0)

Pin(n+1,1,0) := g C(n+1,1,0) | N (g) = 1, (g)vg
1
V v
1
V
is a double cover of the group of Lorentz transformations, i.e., the

group of orthogonal transformations of the Minkowski space R(n+1,1) .
Furthermore, this group is isomorphic to the conformal group of Rn ,
see [55]. The Spin group

Spin(n+1,1,0) := g
C+
(n+1,1,0) | N (g) = 1,
gvg
1
1
V v
V
is a double cover of the orientation preserving conformal transformations. We construct this group with its subgroups by studying the
grade-1 elements, that generate this group. First we identify the
group SE(3) of Euclidean displacements as a subgroup of the conformal group. Therefore, we have to nd all non-null vectors that
leave the point at innity xed. This results in the condition
(g)g1 = .
We derive g (g) = 0. The most general vector satisfying this

condition is = n + , see [19]. The geometric inner product null
space corresponding to is a plane.
Reection in a Plane The most general versor xing the point at
innity corresponds to a plane in inner product null space representation, i.e., a dual plane. We study the action of this versor to a point.
Furthermore, we are free to chose the origin. Therefore, we choose a
plane passing through the origin, this means = 0. We calculate
1
1
n(o + p + p2 )n1 = o npn1 + p2
2
2
and see that this transformation leaves the point at innity and the
origin xed. Furthermore, this transformation is a reection in the
plane dened by n. Two reections in parallel planes result in a
49
translation written in versor form by t = 1 12 t, t = t1 e1 +t2 e2 +t3 e3

R3 . The composition of two reections in planes passing through the
origin yields a general rotation r = cos 12 sin( 12 )R, where R denotes
a bivector corresponding to Pl
ucker coordinates of the rotation axis,
see Ex. 1.6. The action on a point P is computed by
1
1
r(o + p + p2 )r1 = o + rpr1 + p2 .
2
2
Hence, the generators o and are invariant under this transformation

and the point P undergoes a rotation. It is clear that a general
Euclidean displacement can be obtained by composing translations
and rotations.
Reection in a Sphere Grade-1 elements and their geometric inner
product null spaces can be used to describe spheres. We do not
discuss this topic in detail here. A full classication of conformal
mappings and their generation can be found in [19]. The action of
a versor corresponding to a dual sphere applied to a point is, in
general, a reection in this sphere. A scaling is the composition of
two reections in spheres with the same center and dierent radii
(both real). A versor corresponding to a transformation that scales
with respect to the origin by the factor e is dened by:
s = cosh
+ sin (o ) = e 2 o .
2
2
By composing positive scalings and displacements the group of ane

transformations can be described as subgroup of Spin(4,1,0) .
Transversions A transversion is the composition of an inversion in
the unit sphere, a translation and a further inversion in the unit
sphere. This transformation can be written in exponential form by
1
1
1
1
(o )(1 t)(0 ) = e 2 ot .
2
2
2
Table 1.2 summarizes the basic operations in the conformal Cliord

algebra model and their versors, see [19, p. 476].
50
Table 1.2: Basic transformations in the conformal model
Type of Operation
re. in origin plane
re. in real unit sphere
re. in origin
rotation over in R-plane
translation over t
scaling by e
transversion over t
Explicit Form
n
c 12
o
Exp. Form
none
none
none
cos( 12 )sin( 12 )R
e 2 R
1
1
1 12 t
e 2 t
cosh( 21 )sinh( 12 )o
e 2 o
1 + 12 ot
e 2 ot
1
1
1.6 A Cliord algebraic Approach to Line

Geometry
In this section we aim at a Cliord algebraic description of the Klein
model, cf. section 1.6. Therefore, we recall results that are already
published in [40] and present the Cliord algebra C(3,3,0) as homogeneous model for the line space. It is well-known, that projective
automorphisms of Kleins quadric induce projective transformations
of P3 (R) and vise versa, see [56]. This is the reason why we consider
this homogeneous model. Hence, it is possible to describe projective
transformations as the action of the versor group of C(3,3,0) . Especially the action of a reection in the vector space, i.e., of a grade-1
element induces a null polarity in P3 (R), as we will show.
Remark 1.13. This construction works for any quadric in any projective space. In this and the following sections we are interested in
quadrics that serve as point models for other geometries.
To build up the homogeneous model we use the quadratic form of
Kleins quadric M24 , that is given by
Q=
O
I

I
,
O
where O is the 3 3 zero matrix and I the 3 3 identity matrix.

The matrix Q that is used here corresponds to the polarity of Kleins
1.6 A Cliord algebraic Approach to Line Geometry
51
quadric, since multiplication with real scalars has no eect, see Eq.
(1.13). As underlying vector space for the Cliord algebra we take R6
as vector space model for P5 (R). The corresponding Cliord algebra
has signature (p, q, r) = (3, 3, 0) (cf. [47]) and is of dimension 26 =
ucker coordinates, see Def. 1.4,
64. Lines of P3 (R) represented by Pl
correspond to null vectors in this algebra, i.e., vectors that square to
zero. A vector is given by
v = x1 e1 + x2 e2 + x3 e3 + x4 e4 + x5 e5 + x6 e6
and its square is computed by

vv = 2(x1 x4 + x2 x5 + x3 x6 ).
(1.37)
Eq. (1.37) evaluates to zero if, and only if, the Pl

ucker condition
(1.12) is fullled, i.e., if the point X = (x1 , . . . , x6 )T R P5 (R) is
contained by M24 , and therefore, describes a line in P3 (R). The norm
of a vector equals
vv = 2(x1 x4 + x2 x5 + x3 x6 ).
The polarity of the metric quadric is given by multiplication with

the pseudoscalar J := e123456 . Duality between subspaces of P5 (R) induced by the polarity is expressed by multiplication with the pseudoscalar. Outer product null spaces can be used to describe point sets
corresponding to algebra elements. Moreover, the dual geometric entity with respect to the polarity Q is obtained with the inner product
null space
NI(A) = NO(AJ).
1.6.1 Collineations and Correlations in the Image Space

The Pl
ucker line coordintes of a line spanned by two dierent points
X = xR = (x0 , x1 , x2 , x3 )T R, Y = yR = (y0 , y1 , y2 , y3 )T R P3 (R)
were dened in section 1.2.1 by
52
p01
p23

=

=
x0
y0
x2
y2

x
x1
, p02 = 0

y1
y0

x3
x
,p = 3
y3 31 y3

x
x2
, p03 = 0

y2
y0

x1
x
,p = 1
y1 12 y1

x3
,
y3

x2
.
y2
Now we ask for projective automorphisms of Kleins quadric induced

by collineations or correlations in P3 (R). First, we transfer projective transformations acting on P3 (R) to automorphic collineations of
M24 . Let C = (ckl ), k, l = 0, . . . , 3 be the matrix representation of a
collineation. We apply this collineation to the points X = xR, Y =
yR P3 (R) with x = (x0 , x1 , x2 , x3 )T , y = (y0 , y1 , y2 , y3 )T and compute
the Pl
ucker coordinates of the line joining x = Cx and y = Cy . The
Pl
ucker coordinates of the image line under this collineation are given
by:
pij

x x
= i j = xi yj xj yi
yi yj

=
cik xk
cjl yl
cjl xl
cik yk
k
l
l
k

cik cjl (xk yl xl yk ),
=
k,l
where (i, j) is one of (0, 1), (0, 2), (0, 3), (2, 3), (3, 1) or (1, 2), see [56, p.
139]. If we write the action of this transformation on the space of
lines as matrix vector product we get a 6 6 matrix L containing the
coecients from the equations above
(p01 , p02 , p03 , p23 , p31 , p12 )T = L (p01 , p02 , p03 , p23 , p31 , p12 )T .
When we repeat this procedure for a correlation the columns of the

matrix correspond to plane coordinates. Hence, we can compute the
collineation in the image space in the same way, but in this case we
have to calculate the Pl
ucker coordinates of the image lines by the
intersection of two planes instead of the connection of two points. We
get
(p01 , p02 , p03 , p23 , p31 , p12 )T , (1.38)
(p01 , p02 , p03 , p23 , p31 , p12 )T = L
53
denes an automorphic collineation of M 4 corresponding to

where L
2
a correlation in P3 (R).
Now we examine geometric entities that are described in this geometric algebra by inner product and outer product null spaces, see
[53].
1.6.2 Algebra Representation of Linear Line Manifolds

In this section we introduce the Cliord algebra representation for the
occurring linear line manifolds as outer product null spaces of (k + 1)blades. Using duality k-space intersections can also be described
as inner product null space of (n (k + 1))-blades where n = 6 is
the dimension of the vector space model of P5 (R). We start with
subspaces contained entirely in Kleins quadric.
Lines contained in M24 as null two-blades Null blades can be used
to describe subspaces contained entirely in M24 . Subspaces that are
contained in M24 are either lines or two-spaces in P5 (R). A null twoblade generated by the exterior product of two null vectors corresponding to conjugate points on M24 denes a line in M24 P5 (R). Its
outer product null space is the set of all null vectors corresponding
to a pencil of lines in P3 (R).
Two-spaces contained in M24 as null three-blades A two-space P12
in P5 (R) that is contained entirely in Kleins quadric can be expressed
as the exterior product of three null vectors corresponding to points
contained in the two-space. This results in a null three-blade. Its
outer product null space consists of all null vectors that correspond
to points contained in the two-space P12 M24 , i.e., a bundle of lines
or a eld of lines.
Conics on Kleins quadric as non-null three-blades Three-blades
corresponding to two-spaces in P5 (R)can be dened as exterior prod1
V corresponding to points on
uct of three null vectors v1 , v
2 , v3
3
4
M2 . If the three-blade B
V squares to zero it corresponds to
a two-space that is entirely contained in M24 , else it corresponds to
54
a two-space that intersects in a conic on M24 P5 (R). All points

contained by this two-space can be computed with the help of the
outer product null space
NO(v1 v2 v3 ) = NO(v1 ) NO(v2 ) NO(v3 ).
To get the null vectors located in the two-space p = v1 + v2 + v3

we determine the zero divisors by pp = 0. This results in a quadratic
equation involving the coecients , , and . The solution is given
by the intersection of Kleins quadric with the two-space. In P5 (R)
the dual of a two-space is a two-space and the points contained by
the dual of a two-space can be calculated by the inner product null
space of a three-blade corresponding to the two-space.
Linear line congruences as four-blades Three-spaces are polar to
lines. Thus, linear line congruences can be described by inner product
null spaces of two-blades that correspond to lines in P5 (R) or outer
product null spaces of four-blades that correspond to three-spaces in
P5 (R). Lines in P5 (R) are represented by the exterior product of two

vectors v1 , v2 1 V corresponding to points in P5 (R). A general line
in P5 (R) written as outer product of two arbitrary vectors
v 1 = x1 e 1 + x 2 e 2 + x 3 e 3 + x 4 e 4 + x 5 e 5 + x 6 e 6 ,
v2 = y1 e1 + y2 e2 + y3 e3 + y4 e4 + y5 e5 + y6 e6
has the form

6

xi xj

L = v1 v 2 =
yi yj eij .
i,j=1
i<j
With Eq. (1.30) we know that the two conditions

a1 x4 + a2 x5 + a3 x6 + a4 x1 + a5 x2 + a6 x3 = 0,
a 1 y4 + a 2 y5 + a 3 y6 + a 4 y1 + a 5 y 2 + a 6 y 3 = 0
are sucient to describe the inner product null space of L. All null
vectors a = a1 e1 + a2 e2 + a3 e3 + a4 e4 + a5 e5 + a6 e6 satisfying these
conditions correspond to the set of lines contained in a linear line
55
congruence. In a dual way, a linear line congruence can also be represented as the outer product null space of a four-blade that can be
constructed as exterior product of four linearly independent vectors.
Linear Line Complexes as ve-blades A four-space in P5 (R)can

be described as the outer product null space of a ve-blade B 5 V .
Using duality, the same four-space
is dened by the inner product null

space of the vector BJ 1 V . If the given vector is a null vector,
the linear line complex is singular, else regular. The outer product
null space of a vector and the inner product null space of its dual
correspond to the vector itself. Let v = x1 e1 + x2 e2 + x3 e3 + x4 e4 +
x5 e5 + x6 e6 be a general vector. Then its inner product null space is
given by

NI(v) =
1

V | x1 a 4 + x 2 a 5 + x3 a 6 + x 4 a 1 + x 5 a 2 + x6 a 3 = 0 ,
with a = a1 e1 + a2 e2 + a3 e3 + a4 e4 + a5 e5 + a6 e6 . The same set of lines

can be obtained as the outer product null space of the dual of v:

NO(vJ) =
1

V | x1 a 4 + x 2 a 5 + x 3 a 6 + x 4 a 1 + x 5 a 2 + x 6 a 3 = 0 ,
with a = a1 e1 + a2 e2 + a3 e3 + a4 e4 + a5 e5 + a6 e6 . Note, that it is not

important from which side we multiply with the pseudoscalar, since
Jv = vJ. We are working in a projective setting, and thus, the multiplication by 1 or by a real number does not change the geometric
meaning of the object and its inner product or outer product null
space.
1.6.3 Transformations
To describe the action of the versor group we recall the denition of
a versor, see Def. 1.10. A versor a is an algebra element that can be
expressed as the k-foldgeometric product of non-null vectors. Thus,
1
a = v1 . . . vk with vi
V, i = 1, . . . , k . Transformations are applied
with the so-called sandwich operator.
56
Remark 1.14. For a vector v

obtained by
v1 =
1
V the inverse element can be
v
.
vv
Thus, the inverse element diers from the conjugate element by a real
factor. This is also true for a k-fold product of vectors.
Since we are working in a homogeneous Cliord algebra model multiplication with a real factor does not change the geometric meaning of
an algebra element. Therefore, we use the conjugate element instead
of the inverse element. Hence, the sandwich operator that we use is
given by:
(a)va = aa ((a)va1 ).
This operator does not involve an inverse, and therefore, it can also
be applied if the element a is not invertible.

If a vector v 1 V corresponds to a line L L3 it is a null vector

respectively a zero divisor, i.e., it has no inverse. Applying the sandwich operator with an arbitrary element of the versor group to a zero
divisor results again in a zero divisor since the zero divisors forms an
ideal in the algebra. This means null vectors corresponding to lines
in P3 (R) are mapped to null vectors corresponding to lines in P3 (R).
The transformations induced by non-null vectors are reections in R6
as a model for P5 (R) that x Kleins quadric.
The advantage of this model is that transformations can be applied
via the
sandwich operator to every entity that can be represented as
k
A
V in inner product or outer product null space representation.
For example, we are able to apply the transformation to a four-blade
that corresponds to a linear line congruence with one sandwich operator. The resulting element is again a four-blade corresponding to
a linear line congruence that is of the same projective type (elliptic,
parabolic, or hyperbolic) as the original one. At this point we recall
a theorem from [56, theorem 2.1.10].
Theorem 1.2. Projective collineations and correlations of P3 (R) induce projective automorphisms of Kleins quadric, and Kleins quadric
does not admit other projective automorphisms.
57
The Cliord group of this Cliord algebra, i.e., the group of projective automorphisms of Kleins quadric is generated by non-null
vectors. Thus, we are interested in the action of non-null vectors on
null vectors. Let
a = a1 e 1 + a 2 e 2 + a 3 e 3 + a 4 e 4 + a 5 e 5 + a 6 e 6 ,
v = x1 e1 + x2 e2 + x3 e3 + x4 e4 + x5 e5 + x6 e6
be two vectors with

aa = a1 a4 + a2 a5 + a3 a6 = 0,
vv = x1 x4 + x2 x5 + x3 x6 = 0.
The action
of the sandwich operator (a)va is linear on the vector
1
V . The matrix acting on P5 (R) can be represented by
space
k1
a2 a4
a3 a4
M=
a4 a4
a5 a4
a6 a4
a1 a5
k2
a3 a5
a4 a5
a5 a5
a6 a5
a1 a6
a2 a6
k3
a4 a6
a5 a6
a6 a6
a1 a1
a2 a1
a3 a1
k4
a5 a1
a6 a1
a1 a2
a2 a2
a3 a2
a4 a2
k5
a6 a2
a1 a3
a2 a3
a3 a3
,
a4 a3
a5 a3
(1.39)
k6
with
k1 = a5 a2 a6 a3 ,
k2 = a6 a3 a4 a1 ,
k3 = a4 a1 a5 a2 ,
k4 = a5 a2 a6 a3 ,
k5 = a6 a3 a4 a1 ,
k6 = a4 a1 a5 a2 .
Naturally, we now ask for the corresponding projective mapping acting on P3 (R). Therefore, we examine the action of the collineation M
on the space of lines. The action on the base lines corresponding to
the Pl
ucker coordinate vectors
b1 = (1 : 0 : 0 : 0 : 0 : 0),
b2 = (0 : 1 : 0 : 0 : 0 : 0),
b3 = (0 : 0 : 1 : 0 : 0 : 0),
b4 = (0 : 0 : 0 : 1 : 0 : 0),
b5 = (0 : 0 : 0 : 0 : 1 : 0),
b6 = (0 : 0 : 0 : 0 : 0 : 1)
is given by the columns of the matrix M.

h1 = Mb1 ,
h2 = Mb2 ,
h3 = Mb3 ,
h4 = Mb4 ,
h5 = Mb5 ,
h6 = Mb6 .
If we look at the images of the three ideal lines b4 , b5 , b6 we see, that the
corresponding lines posses linear dependent direction vectors. This
58
means that the three image lines h4 , h5 , h6 intersect in the same ideal
point, and therefore, they belong to a bundle of lines. Since the lines
b4 , b5 , b6 dene a eld of lines and the image is a bundle of lines, the
mapping must be a correlation. With this knowledge we determine
the action of M on the bundle of lines concurrent to the origin spanned
by b1 , b2 , b3 . The image of this bundle of lines is a eld of lines. To
obtain the coordinates of the plane that carries the eld of lines we
intersect the image lines and get three points dening the plane in
P3 (R)
h1 h2 = s1 = (a3 , a5 , a4 , 0)T ,
h1 h3 = s2 = (a2 , a6 , 0, a4 )T ,
h2 h3 = s3 = (a1 , 0, a6 , a5 )T .
The plane coordinates p1 of the plane generated by these three points

have to satisfy
p1 , s1
= p1 , s2
= p1 , s3
= 0.
(1.40)
Solving Eq. 1.40 results in p1 = R(0, a4 , a5 , a6 )T as image of the origin
O = (1, 0, 0, 0)T R. We repeat this procedure for the bundle of lines
spanned by b1 , b5 , b6 , i.e., the bundle of lines concurrent to the ideal
point Xu = (0, 1, 0, 0)T R.
h1 h5 = (a2 , a6 , 0, a4 )T ,
h1 h6 = (a3 , a5 , a4 , 0)T ,
h5 h6 = (0, a1 , a2 , a3 )T ,
p2 = R(a4 , 0, a3 , a2 )T .
Repeating the procedure for the bundle of lines concurrent to the

ideal point Yu = (0, 0, 1, 0)T R spanned by the lines b2 , b4 , and b6 yields
h2 h4 = (a1 , 0, a6 , a5 )T ,
h2 h6 = (a3 , a5 , a4 , 0)T ,
h4 h6 = (0, a1 , a2 , a3 )T ,
p3 = R(a5 , a3 , 0, a1 )T .
Analogue computation for the bundle of lines spanned by b3 , b4 , and

b5 concurrent to the ideal point Zu = (0, 0, 0, 1)T R results in
h3 h4 = (a1 , 0, a6 , a5 )T ,
h3 h5 = (a2 , a6 , 0, a4 )T ,
h4 h5 = (0, a1 , a2 , a3 )T ,
p4 = R(a6 , a2 , a1 , 0)T .
Note, that the correlation is not determined through p1 , . . . , p4 since

the scaling of each pi , i = 1, . . . , 4 is not determined yet. Together
59
with the matrix representation of a correlation see Eq. (1.38) the

correlation in P3 (R) is xed. It can be written in matrix form as
0 a4 a5 a6
a4 0 a3 a2
.
K=
a5 a3
0 a1
a6 a2 a1
0
This correlation is a null polarity, see [65]. The determinant of this

null polarity is calculated by
det K = (a1 a4 + a2 a5 + a3 a6 )2 =
1
(aa)2 .
4
(1.41)
Thus, the square of a vector is related to the determinant of the

corresponding null polarity.
Remark 1.15. Note, that this procedure works also for null vectors,
because we do not use the inverse element to determine the sandwich
product in homogeneous Cliord Algebra models.
With these preliminaries we are now able to prove the following theorem:
Theorem 1.3. Each regular projective transformation, i.e., regular
correlation or regular colline-ation can be represented by the product
of at most six null polarities.
Proof. The group of automorphic collineations of M24 and projective
transformations of P3 (R) are isomorphic, see [56]. Furthermore, the
group of automorphisms of Kleins quadric can be described by the
Pin group of C(3,3,0) , see [28] for the general case. The Cliord algebra
model has at most grade six. Furthermore, the k-fold product of
grade-1 elements, that correspond to null polarities generate the Pin
or Spin group depending on whether k is odd or even. To reach
the maximum grade of six, we need a product of at least six grade-1
elements. That at most six elements are necessary follows from the
Cartan-Dieudonne theorem, cf. [22].
All together we have that the versor group generated by reections respectively non-null vectors corresponds to the group of automorphic
60
collineations of M24 induced by regular projective transformations in

P3 (R) that can be written as the product of invulotic automorphic
collineations of Kleins quadric corresponding to null polarities. Furthermore, all transformations that are generated by the product of
an even number of vectors are collineations, i.e., elements of the Spin
group when they are normalized. Transformations that are generated by an odd number of vectors correspond to correlations, and
therefore, to the Pin group when they are normalized.
Example 1.8. As example we present the action of a null polarity
applied to a conic on Kleins quadric. Thus, we give three null vectors
corresponding to Pl
ucker coordinates li , i = 1, 2, 3 of three lines
l1 = e1 + 2e6 ,
l2 = e2 + 2e4 ,
l3 = e3 + 2e5 .
The corresponding two-space in P5 (R) is represented as three-blade by

P = l1 l2 l3 = e123 +2e236 2e134 4e346 +2e125 +4e256 +4e145 +8e456 .
The outer product null space of this entity can be calculated as

NO(P) =

1
1
1
e2 +e4 +
e3 +e5 +
e1 +e6 | , , R .
2
2
2
Indeed, the outer product null space of P spans the same two-space
as [l1 , l2 , l3 ]. To get the regulus dened by the three lines Li or their
corresponding null vectors li we have to nd all null vectors in the
two-space p = l1 + l2 + l3 . Therefore, we compute
!
p2 = 4 + 4 + 4 = 0.
(1.42)
The parameters ( : : ) can be interpreted as homogeneous coordinates in a projective plane P12 . Eq. (1.42) denes a conic contained in
this plane. Now we apply an arbitrary null polarity given by
a = a1 e 1 + a 2 e 2 + a 3 e 3 + a 4 e 4 + a 5 e 5 + a 6 e 6 .
The outer product null space of the resulting three-blade P = (a)Pa

can be calculated as the set
NO(P ) = {v1 + v2 + v3 | , , R} ,
61
where the vectors vi are given by

2a5 a1 a6 a5 2a6 a1 +4a21
a6 a5 +2a5 a2 +4a1 a2 2a4 a1 +a6 a4
e1 +
e2
a5 a4 +a6 a4 +a6 a5
a5 a4 +a6 a4 +a6 a5
2a5 a3 a25 +4a1 a3 2a5 a1
+
+2e4 ,
a5 a4 +a6 a4 +a6 a5
4a1 a2 +2a6 a1 a26 2a6 a2
2a6 a2 2a4 a2 a6 a4 +4a22
v2 =
e1 +
e2
a5 a4 +a6 a4 +a6 a5
a5 a4 +a6 a4 +a6 a5
a5 a4 +2a6 a3 +a6 a4 +4a3 a2 2a5 a2
+
e3 +2e5 ,
a5 a4 +a6 a4 +a6 a5
2a4 a1 +4a1 a3 +a6 a5 2a6 a3 +a5 a4
2a4 a2 2a4 a3 +4a3 a2 a24
v3 =
e1 +
e2
a5 a4 +a6 a4 +a6 a5
a5 a4 +a6 a4 +a6 a5
2a4 a3 a5 a4 +4a23 2a5 a3
+
e3 +2e6 .
a5 a4 +a6 a4 +a6 a5
v1 =
We search for null vectors contained in the plane p (, , ) = v1 +

v2 + v3 . Therefore, we calculate p p = 0. This leads to the following
quadratic equation
((2a4 a3 + a5 a4 + 2a6 a3 + a6 a4 + 2a4 a2 + 8a3 a2 2a5 a2 a24 )
+ (8a1 a3 2a6 a3 2a5 a1 + 2a4 a1 + 2a5 a3 + a6 a5 a25 + a5 a4 ))
+ (2a4 a1 + 2a5 a2 + 2a6 a1 + a6 a4 + a6 a5 + 8a1 a2 2a6 a2 a26 )
+ (2a5 a1 a6 a5 2a6 a1 + 4a21 ) + (2a4 a3 a5 a4 + 4a23 2a5 a3 ) 2 2
+ (2a6 a2 2a4 a2 a6 a4 + 4a22 ) 2 = 0.
This equation denes a conic in the image of the projective plane P12
under the null polarity induced by a. Every point on this conic delivers
the parameters for a null vector that corresponds to a line in P3 (R).
1.6.4 Collineations as Spin Group

We are interested in the relationship between regular projective mappings and elements of the versor group of C(3,3,0) . The general approach (cf.[53]) does not work for this model. Therefore, we develop
a more line geometric approach by the examination of the action of
versors on null three-blades corresponding to two-spaces that are en-
62
tirely contained in M24 . A general element g C+

(3,3,0) corresponding
to a collineation is given by
g =g1 e0 +g2 e12 +g3 e13 +g4 e14 +g5 e15 +g6 e16 +g7 e23 +g8 e24 +g9 e25 +g10 e26
+g11 e34 +g12 e35 +g13 e36 +g14 e45 +g15 e46 +g16 e56 +g17 e1234 +g18 e1235
+g19 e1236 +g20 e1245 +g21 e1246 +g22 e1256 +g23 e1345 +g24 e1346 +g25 e1356
+g26 e1456 +g27 e2345 +g28 e2346 +g29 e2356 +g30 e2456 +g31 e3456 +g32 e123456 .
The conditions that this element is the product of invertible vectors

is obtained by
1
1
(g)vg
V for all v
V.
(1.43)
This results in 36 quadratic equations that occur as coecients of
the grade-5 element eJ , where J {1, 2, 3, 4, 5, 6} with |J| = 5. The
36 quadratic equations dened above dene a pseudo algebraic variety in P31 (R) that can be interpreted as image space of Spin(3,3) , see
chapter 3. Points of P3 (R) can be embedded in the homogeneous
Cliord algebra as null three-blades that correspond to two-spaces
on Kleins quadric. These two-spaces correspond to bundles of lines,
i.e., all lines concurrent to a point. The point contained by all lines
of a bundle can be determined and is described uniquely by the null
three-blade. Therefore, we examine the action of an arbitrary versor
corresponding to a collineation on a null three-blade corresponding to
a bundle of lines. This results in a null three-blade that corresponds
to a bundle of lines again. Afterwards we compute the point concurrent to all lines contained by the bundle of lines to get the image of
the start point under the transformation.
We start with the bundle of lines concurrent to the a general point
P = (y0 , y1 , y2 , y3 )T R. The Pl
ucker coordinates of three lines containing P are computed as
l1 = (y0 : 0 : 0 : 0 : y3 : y2 ), l2 = (0 : y0 : 0 : y3 : 0 : y1 ), l3 = (0 : 0 : y0 : y2 : y1 : 0).
The null three-blade corresponding to the two-space spanned by l1 , l2 ,

and l3 is obtained by
b = (y0 e1 + y3 e5 y2 e6 ) (y0 e2 y3 e4 + y1 e6 ) (y0 e3 + y2 e4 y1 e5 )
63
= y03 e123 + y3 y02 e235 y2 y02 e236 + y3 y02 e134 + y32 y0 e345 y2 y3 y0 e346
y02 y1 e136 + y3 y1 y0 e356 + y2 y02 e124 + y2 y3 y0 e245 y22 y0 e246
y0 y1 y2 e146 y02 y1 e125 + y0 y1 y2 e256 + y3 y1 y0 e145 + y0 y12 e156 .
We apply the sandwich operator with the arbitrary element g

C+
(3,3,0) . The outer product null space of the null three-blade b =
(g)bg is obtained by
NO(b ) = {v1 + v2 + v3 |, , R} ,
where

v1 = y0 (g1 g32 g20 + g4 g29 + g13 + g9 g24 )

+ 2y1 (g17 + g7 ) + 2y2 (g18 2g3 ) + 2y3 (g2 + 2g19 ) e1

2y0 (g31 + g14 ) + 2y1 (g27 g11 ) 2y2 (g12 y2 + g23 )

+ y3 (g9 g1 + g32 g29 + g4 g13 + g20 g24 ) e5

+ 2y0 (g15 g30 ) + 2y1 (g8 + g28 ) + 2y3 (g10 + g21 )

+ y2 (g9 g24 g4 + g1 g32 + g20 + g29 g13 ) e6 ,

v2 = y0 (g1 g29 + g13 + g9 g24 g32 g20 + g4 )

+ 2y1 (g17 + g7 ) + 2y2 (g18 g3 ) + 2y3 (g19 + g2 ) e2

+ 2y0 (g14 + g31 ) + 2y1 (g27 g11 ) 2y2 (g12 y2 + g23 )

+ y3 (g9 g1 + g32 g29 + g4 g13 + g20 g24 ) e4

2y0 (g16 + g26 ) 2y2 (g5 + g25 ) + 2y3 (g22 g6 )

+ y1 (g13 + g32 g4 + g29 + g9 g24 g20 g1 ) e6 ,

v3 = y0 (g1 g29 + g13 + g9 g24 g32 g20 + g4 )

+ 2y1 (g17 + g7 ) + 2y2 (g18 g3 ) + 2y3 (g2 + g19 ) e3

+ 2y0 (g15 g30 ) + 2y1 (g8 + g28 ) + 2y3 (g10 + g21 )

+ y2 (g1 g24 g4 g32 + g9 + g20 + g29 g13 ) e4

+ 2y0 (g16 + g26 ) + 2y3 (g22 g6 ) 2y2 (g5 + g25 )

+ y1 (g13 + g29 + g32 g4 + g9 g24 g20 g1 ) e5 .
The null vectors v1 , v2 , and v3 span a bundle of lines concurrent to the

image of the origin under the collineation corresponding to g. To get
the point we have to intersect two lines of the bundle of lines. Therefore, we change the model and transfer the null vectors corresponding
64
to v1 , v2 to two-blades L1 , L2 of the Grassmann algebra G(P3 ), cf. [27].

The representation of a line L = (p01 : p02 : p03 : p23 : p31 : p12 ) in G(P3 ) is
L = p01 e12 + p02 e13 + p03 e14 + p23 e34 p31 e24 + p12 e23 .
(1.44)
Note that due to historic conversion the coecient of e24 is multiplied

by 1. We obtain

L1 = y0 (g1 + g4 + g9 g29 + g13 g20 g32 g24 ) + 2y1 (g17 + g7 )

+ 2y2 (g18 g3 ) + 2y3 (g2 + g19 ) e12

+ 2y0 (g31 + g14 ) + 2(g27 g11 ) 2y2 (g12 + g23 )

+ y3 (g32 + g20 + g9 g1 g13 g29 + g4 g24 ) e24

+ 2y0 (g30 g15 ) 2y1 (g8 + g28 )

+ y2 (g24 + g4 g1 g9 g20 g29 + g32 + g13 ) 2y3 (g10 + g21 ) e23 ,

L2 = y0 (g4 + g9 g29 + g13 g20 + g1 g32 g24 ) + 2y1 (g17 + g7 )

+ 2y2 (g18 g3 ) + 2y3 (g2 + g19 ) e13

+ 2y0 (g14 + g31 ) + 2y1 (g27 g11 ) 2y2 (g12 + 2g23 )

+ y3 (g32 + g20 + g9 g1 g13 g29 + g4 g24 ) e34

+ 2y0 (g16 + g26 ) + y1 (g1 g9 g13 g32 + g24 g29 + g20 + g4 )

+ 2y2 (g25 + g5 ) + 2y3 (g6 g22 ) e23 .
To get the intersection point of these two lines we take an arbitrary

= xR P3 (R) with representation X = x0 e1 + x1 e2 + x2 e3 +
point X
1
G(P3 ) and compute the incidence condition with both lines
x 3 e4
L1 and L2 :
L1 X = 0,
L2 X = 0.
This results in a system of linear equations for x0 , x1 , x2 , x3 with solution

x0 = y0 (g4 +g9 g29 +g13 g20 g32 g24 +g1 )+2y1 (g17 +g7 )
+2y2 (g18 g3 )+2y3 (g2 +g19 ),
x1 = 2y0 (g16 +g26 )+y1 (g1 g9 g13 g29 +g20 +g4 g32 +g24 )
+ 2y2 (g5 +g25 )+2y3 (g6 g22 ),
x2 = 2y0 (g15 g30 )+2y1 (g8 +g28 )
+ y2 (g29 g24 g4 +g1 +g9 +g20 g32 g13 )+2y3 (g10 +g21 ),
65
x3 = 2y0 (g31 +g14 )+2y1 (g11 g27 )

+ 2y2 (g12 +g23 )+y3 (g1 g32 g20 g9 +g13 +g29 g4 +g24 ).
If we rewrite X as product of a matrix M with the vector (y0 , y1 , y2 , y3 )T

we get
(x0 , x1 , x2 , x3 )T = M (y0 , y1 , y2 , y3 )T ,
(1.45)
with
k1
2(g26 + g16 )
M=
2(g15 g30 )
2(g31 + g14 )
2(g7 + g17 )
k2
2(g8 + g28 )
2(g11 g27 )
2(g18 g3 )
2(g5 + g25 )
k3
2(g23 + g12 )
2(g19 + g2 )
2(g6 g22 )
,
2(g21 + 2g10 )
k4
where
k1 = g1 g20 g24 g32 g29 + g9 + g4 + g13 ,
k2 = g24 g9 + g20 g13 g32 + g1 + g4 g29 ,
k3 = g1 g13 g32 g4 + g29 + g9 g24 + g20 ,
k4 = g24 + g13 + g29 + g1 g4 g9 g20 g32 .
Therefore, we have found the correspondence between regular col+

lineations and elements of C+
(3,3,0) . For a given versor g C(3,3,0)
corresponding to a collineation we can determine the representation
as 4 4 matrix. If we start with a 4 4 matrix A with coecients
(aij ), i, j = 0, . . . , 3 representing a collineation we have to solve a system of 16 linear and 36 quadratic equations to get the corresponding
algebra representation. The linear equations are derived with the
help of the matrix representation determined above and result in
a01 = 2(g7 + g17 ),
a02 = 2(g18 g3 ),
a03 = 2(g19 + g2 ),
a10 = 2(g26 + g16 ),
a12 = 2(g5 + g25 ),
a13 = 2(g6 g22 ),
a20 = 2(g15 g30 ),
a21 = 2(g8 + g28 ),
a23 = 2(g21 + 2g10 ),
a30 = 2(g31 + g14 ),
a31 = 2(g11 g27 ),
a32 = 2(g23 + g12 ),
a00 = g1 g20 g24 g32 g29 + g9 + g4 + g13 ,

a11 = g24 g9 + g20 g13 g32 + g1 + g4 g29 ,
(1.46)
66

a22 = g1 g13 g32 g4 + g29 + g9 g24 + g20 ,
a33 = g24 + g13 + g29 + g1 g4 g9 g20 g32 .
These 16 equations have to be solved with the constraint equations

derived Eq. (1.43). For both cases gg = 1 and gg = 1 the resulting
system of 16 linear and 36 quadratic equations possesses an unique
solution. It can be solved analytically with the help of a computer
algebra system. We give an example:
Example 1.9. Let K PGL(P3 (R)) be given by
1
1
K=
1
1
0
1
2
1
3
0
1
2
0
1
.
0
1
To get a versor g C+

(3,3,0) corresponding to this collineation we have
to solve the system (1.46).
2(g7 + g17 ) = 0,
2(g18 g3 ) = 3,
2(g19 + g2 ) = 0,
2(g26 + g16 ) = 1,
2(g5 + g25 ) = 0,
2(g6 g22 ) = 1,
2(g15 g30 ) = 1,
2(g8 + g28 ) = 2,
2(g21 + 2g10 ) = 0,
2(g11 g27 ) = 1,
2(g23 + g12 ) = 2,
2(g31 + g14 ) = 1,
(1.47)
g1 g20 g24 g32 g29 + g9 + g4 + g13 = 1,

g24 g9 + g20 g13 g32 + g1 + g4 g29 = 1,
g1 g13 g32 g4 + g29 + g9 g24 + g20 = 1,
g24 + g13 + g29 + g1 g4 g9 g20 g32 = 1.
We have two possibilities to guarantee that the resulting versor is in

the Spin group, i.e., gg = 1 or gg = 1. We compute both solutions
and start with the constraint equations implied by Eq. (1.43) and
gg = 1. The corresponding Spin group element has the form:
1
g+ = 7e0 +6e12 6e13 +e14 2e15 6e23 +6e24 e25 2e26 +2e34
8 2
67
+6e35 5e36 4e45 +2e46 +6e1234 4e56 +6e1235 6e1236

5e1245 +2e1246 4e1256 +2e1345 e1346 +2e1356 2e2345

+2e2346 +e2356 2e2456 e123456 .
If we demand that gg = 1 the resulting Spin group element is

computed as
1
g = e0 6e12 6e13 e14 +2e15 +4e16 +6e23 +2e24 +e25 +2e26 +2e34
8 2
+2e35 +5e36 +2e46 6e1234 +6e1235 +6e1236 +5e1245 2e1246
+6e1345 +e1346 2e1356 4e1456 2e2345 +6e2346 e2356

2e2456 4e3456 7e123456 .
Both elements g+ and g correspond to the same collineation. With

Eq. (1.45) we get the matrix K back with the coecients of g+ and
g .
1.6.5 Correlations as Pin Group

We know that grade-1 elements respectively their action on null vectors correspond to null polarities. Thus, correlations are elements of
C
(3,3,0) . To transfer a 4 4 matrix representing a correlation into
the Cliord algebra we use a similar approach as we used for collineations. An arbitrary eld of lines contained in the plane determined by its plane coordinates X = R(x0 , x1 , x2 , x3 )T corresponds to
the outer product null space of a null three-blade. We apply an arbitrary versor h contained in C
(3,3,0) to this null three-blade. The image
is again a null three-blade whose outer product null space is a bundle
of lines. Thus, we compute the point concurrent to all lines of this
bundle of lines and describe the action of h as product of a matrix
with a vector. A general element h C
(3,3,0) is given by
h = h1 e1 +h2 e2 +h3 e3 +h4 e4 +h5 e5 +h6 e6 +h7 e123 +h8 e124 +h9 e125 +h10 e126
+h11 e134 +h12 e135 +h13 e136 +h14 e145 +h15 e146 +h16 e156 +h17 e234 +h18 e235
+h19 e236 +h20 e245 +h21 e246 +h22 e256 +h23 e345 +h24 e346 +h25 e356 +h26 e456
+h27 e12345 +h28 e12346 +h29 e12356 +h30 e12456 +h31 e13456 +h32 e23456 .
68
If this element shall be a versor we have constraint equations that

can be derived from
(h)vh
1
V for all v
1
V.
(1.48)
We start with an arbitrary plane given by its plane coordinates Y =

ucker coordinates of three lines contained in this
R(y0 , y1 , y2 , y3 )T . Pl
plane can be determined by
l1 = (0 : y3 : y2 : y0 : 0 : 0), l2 = (y3 : 0, y1 : 0 : y0 : 0), l3 = (y2 : y1 : 0 : 0 : 0 : y0 ).
The null three-blade corresponding to the plane [l1 , l2 , l3 ] whose outer

product null space is the eld of lines dened by l1 , l2 and l3 is computed as
b = (y3 e2 y2 e3 + y0 e4 ) (y3 e1 + y1 e3 + y0 e5 ) (y2 e1 y1 e2 + y0 e6 )
= y0 y1 y2 e134 + y0 y3 y2 e125 y22 y0 e135 + y02 y2 e145 + y0 y3 y1 e124
+ y0 y12 e234 + y0 y1 y2 e235 y02 y1 e245 + y32 y0 e126 y0 y3 y2 e136
+ y02 y3 e146 + y0 y3 y1 e236 y02 y1 e346 + y02 y3 e256 y02 y2 e356 + y03 e456 .
We apply the sandwich operator with the general element h C

(3,3,0) .
Since a versor from the odd part of the algebra corresponds to a
correlation the null three-blade b = (h)bh corresponds to a bundle
of lines. Thus, we can apply the same procedure that we used for
collineations to determine the image point, i.e., the point concurrent
to all lines of the bundle of lines. Therefore, we compute the outer
product null space of b
NO(b ) = {v1 + v2 + v3 |, , R} ,
where

v1 = 2y0 h7 + y1 (h29 h1 h9 h13 ) + y2 (h8 h2 h19 h28 )

+ y3 (h11 + h27 + h18 h3 ) e1

y0 (h3 + h11 h27 + h18 ) + y1 (h25 + h14 + h31 h5 )

+ y2 (h4 + h32 h24 + h20 ) + 2y3 h23 e5

+ y0 (h2 + h8 h19 + h28 ) + y1 (h6 + h22 h15 + h30 ) 2y2 h21

+ y3 (h20 h4 h32 h24 ) e6 ,
69

v2 = 2y0 h7 + y1 (h29 h1 h9 h13 ) + y2 (h8 h2 h28 h19 )

+ y3 (h27 + h18 h3 + h11 ) e2

+ y0 (h3 + h11 + h18 h27 ) + y1 (h25 + h14 + h31 h5 )

+ y2 (h4 + h32 h24 + h20 ) + 2y3 h23 e4

y0 (h1 h29 h13 h9 ) + 2h16 y1 + y2 (h22 h6 h15 h30 )

+ y3 (h25 + h5 h31 + h14 ) e6 ,

v3 = 2y0 h7 + y1 (h29 h13 h1 h9 ) + y2 (h8 h2 h28 h19 )

+ y3 (h11 h3 + h27 + h18 ) e3

y0 (h2 + h8 h19 + h28 ) + y1 (h6 + h22 h15 + h30 ) 2y2 h21

+ y3 (h20 h4 h32 h24 ) e4

+ y0 (h1 h29 h13 h9 ) + 2h16 y1 + y2 (h22 h6 h15 h30 )

+ y3 (h5 + h14 + h25 h31 ) e5 .
These three null vectors correspond to three points on Kleins quadric

that span the bundle of lines dened by the outer product null space
of b . The point concurrent to these three lines is computed with the
help of the exterior algebra G(P3 ). Hence, we transfer the null vectors
v1 , v2 to two-blades L1 , L2 of G(P3 ) with Eq. (1.44). The intersection
point of these two lines is the image of the plane Y = R(y0 , y1 , y2 , y3 )T .
A general point X = (x0 , x1 , x2 , x3 )T R is written as element of the
exterior algebra X = x0 e1 + x1 e2 + x2 e3 + x3 e4 G(P3 ). The point X is
incident with the lines L1 and L2 if L1 X = 0 and L2 X = 0. This
results in a system of linear equations for x0 , x1 , x2 , x3 with solution
x0 = 2y0 h7 +y1 (h1 +h9 +h13 h29 )+y2 (h2 +h19 +h28 h8 )
+y3 (h3 h18 h27 h11 ),
x1 = y0 (h13 h1 +h29 +h9 )2y1 h16 +y2 (h6 +h30 +h15 h22 )
+ y3 (h31 h25 h14 h5 ),
x2 = y0 (h19 h2 h8 h28 )+y1 (h15 h30 h6 h22 )+2h21 y2
+ y3 (h4 h20 +h32 +h24 ),
x3 = y0 (h27 h3 h11 h18 )+y1 (h5 h31 h25 h14 )
+y2 (h24 h4 h20 h32 )2h23 y3 .
70
If we rewrite X as a product of a matrix C with Y = (y0 , y1 , y2 , y3 )T

we get:

c00
x0
x1 c10
=
x2 c20
x3
c30
c01
c11
c21
c31
c02
c12
c22
c32
c03
y0
y1
c13
, with
c23 y2
c33
(1.49)
y3
c01 = h1 +h9 +h13 h29 ,
c02 = h2 +h19 +h28 h8 ,
c03 = h3 h18 h27 h11 ,
c10 = h13 h1 +h29 +h9 ,
c12 = h6 +h30 +h15 h22 , c13 = h31 h25 h14 h5 ,
c20 = h19 h2 h8 h28 ,
c21 = h15 h30 h6 h22 , c23 = h4 h20 +h32 +h24 ,
c30 = h27 h3 h11 h18 , c31 = h5 h31 h25 h14 , c32 = h24 h4 h20 h32 ,
and
c00 = 2h7 ,
c11 = 2h16 ,
c22 = 2h21 ,
c33 = 2h23 .
The matrix C describes the correspondence between regular correlations and elements of C
(3,3,0) . For a given versor that corresponds
to a correlation, we can compute the matrix representation by Eq.
(1.49). Furthermore, we can determine the versor corresponding to
a regular correlation represented by a 4 4 matrix A if we solve the
system of 16 linear equations given by
a01 = h1 +h9 +h13 h29 , a02 = h2 +h19 +h28 h8 , a03 = h3 h18 h27 h11 ,
a10 = h13 h1 +h29 +h9 , a12 = h6 +h30 +h15 h22 , a13 = h31 h25 h14 h5 ,
a20 = h19 h2 h8 h28 , a21 = h15 h30 h6 h22 , a23 = h4 h20 +h32 +h24 ,
a30 = h27 h3 h11 h18 , a31 = h5 h31 h25 h14 , a32 = h24 h4 h20 h32 ,
a00 = 2h7 ,
a11 = 2h16 ,
a22 = 2h21 ,
a33 = 2h23
(1.50)
with respect to the constraint equations derived by the condition

hh = 1 or hh = 1, see Eq. (1.48). With the help of a computer
algebra system it can be veried that this system possesses an unique
solution for each of the cases hh = 1.
Example 1.10. As example we take again the matrix
1
1
K=
1
1
0
1
2
1
3
0
1
2
0
1
,
0
1
71
but now the matrix describes a correlation. To get the corresponding

versor we have to solve the system of linear equations
h1 +h9 +h13 h29 = 0,
h2 +h19 +h28 h8 = 3,
h3 h18 h27 h11 = 0,
h13 h1 +h29 +h9 = 1,
h6 +h30 +h15 h22 = 0,
h31 h25 h14 h5 = 1,
h19 h2 h8 h28 = 1,
h15 h30 h6 h22 = 2,
h4 h20 +h32 +h24 = 0,
h27 h3 h11 h18 = 1, h5 h31 h25 h14 = 1,

2h7 = 1,
2h16 = 1,
2h21 = 1,
h24 h4 h20 h32 = 2,

2h23 = 1.
If we solve this system of linear equations with respect to the constraints

given by Eq. (1.48) and hh = 1 the resulting versor is given by
h+ =
1
e1 +e2 +e3 +2e4 4e6 2e123 3e124 +e125 +4e126 e134 +e136
8
2e145 2e146 2e156 2e234 e235 +5e236 +2e246 4e245 2e345

6e256 +4e456 2e356 +3e12345 +3e12346 +3e12356 6e23456 .
For hh = 1 we get
h =
1
3e1 +3e2 3e3 6e4 2e123 5e124 +e125 4e126 e134 +e136
8
2e145 +6e146 2e156 +2e234 e235 +3e236 +2e246 2e345 +2e256

+4e346 4e456 2e356 e12345 +e12346 e12356 4e12456 +2e23456 .
Note that both versors h+ and h correspond to the same correlation.

The correlation can be computed with Eq. (1.49).
Remark 1.16. With the knowledge which elements correspond to
collineations and which correspond to correlations we can apply a
transformation to a pencil of lines to study the action of a mapping
on a point. Since a general projective transformation maps pencils of
lines to pencils of lines, we could not distinguish between collineations
and correlations. Therefore, we used bundles of lines and elds of lines
to study the action of a general projective transformation represented
as element of C(3,3,0) .
72
1.6.6 Singular projective Transformations

Now we examine singular projective transformations. Every versor
can be written as the geometric product of grade-1 elements corresponding to null polarities. The square of a grade-1 element (see Eq.
(1.37)) is related to the determinant of the corresponding null polarity, see Eq. (1.41). Thus, we assume that one of these null polarities
is a singular one, and therefore, its determinant is equal to zero. We
dene:
Denition 1.18. An element g C(3,3) with g = v1 . . . vk with k 6

and vi 1 V is called a null versor if at least one vi squares to zero.
Constraint equations can be derived from
(g)vg
1
V for all v
1
V,
g C+
(3,3)
V,
h C
(3,3)
for a singular collineation and

(h)vh
1
V for all v
1
for a singular correlation. The matrix representations for general collineations (1.45) and correlations (1.49) are also valid for the singular
case. The system of linear equations
(1.46)
with the constraint equa
tions implied by (g)vg 1 V for all v 1 V with g C+
(3,3) can
not be solved for a singular collineation. For a singular correlation
the system of linear equations
(1.50)
with the constraint equations

implied by (h)vh 1 V for all v 1 V with h C
(3,3) has also
no solution. Nevertheless, we are able to write singular correlations
or collineations as null versors if we know the null polarities that
generate them. These null polarities can be transferred to vectors
in the homogeneous Cliord algebra model, and therefore, the whole
correlation or collineation can be transferred to the algebra model.
1.7 A Cliord algebraic Approach to Lie Sphere

Geometry
The same construction that we described for Kleins quadric can be
applied to any quadric. As a second example that shall demonstrate
73
the power of this calculus we examine Lie sphere geometry in a Clifford algebra context. Lie sphere geometry is the geometry of oriented
spheres. Especially, for the three-dimensional case the set of oriented
spheres can be mapped to a hyperquadric L41 in ve-dimensional projective space P5 (R). The construction goes back to S. Lie and was
treated again by W. Blaschke, cf. [8]. A modern treatment of this
topic can be found in [14]. Moreover, the Lie construction can be
achieved for arbitrary dimension.
1.7.1 Lies Quadric

A point model for the set of oriented hyperspheres, hyperplanes, and
points (considered as spheres of radius 0) of Rn is given by the projective quadric
Ln+1
: x20 + x21 + . . . + x2n+1 x2n+2 = 0.
1
For our purposes it is convenient that we restrict ourselves to the case

of oriented spheres in three-dimensional Euclidean space. Nevertheless, we formulate the calculus for arbitrary dimensions. The quadric
Ln+1
Pn+2 (R) is of dimension n + 1, degree 2, and is called Lies
1
is
quadric. The maximal dimension of subspaces contained by Ln+1
1
n+1
1, and therefore, there are no two-spaces contained entirely in L1 .
Oriented hyperspheres, hyperplanes, and points are represented in
Lie coordinates as shown in Table 1.3. It is not dicult to recover
the Euclidean representation from Lie coordinates. If x0 + x1 = 0 and
Table 1.3: Correspondence between Euclidean entities and Liecoordinates.
Euclidean
points: u Rn
Lie
T
, 1uu
R
2 , u1 , . . . , u n , 0
T
(1, 1, 0, . . . , 0, 0) R

T
1+ppr 2 1pp+r 2
sphere: center p Rn , signed
,
,
p
,
.
.
.
,
p
,
r
R
1
n
2
2
radius r
planes: uN = h, unit normal (h, h, N1 , . . . , Nn , 1)T R
N Rn
1+uu
2
74
if xn+2 = 0 we have the point at innity. If xn+2 = 0 we bring the

point to the form (h, h, N, 1)T R by dividing by xn+2 . If x0 + x1 = 0
and if xn+2 = 0, we have a proper point. We obtain its normal form
by dividing by x0 + x1 . The last case is if xn+2 = 0. In this case
we have an oriented sphere. Again we get its normal form through
division by x0 + x1 .
The fundamental invariant of Lie sphere geometry is the oriented
contact of spheres. It is not dicult to show that two spheres are in
satoriented contact if, and only if, their Lie coordinates s1 , s2 Ln+1
1
isfy (s1 , s2 ) = 0, where (, ) denotes the bilinear form corresponding
.
to Ln+1
1
Especially for n = 3 the lines on L41 correspond to so called parabolic
pencils of spheres. These pencils consist of all oriented spheres with
one common point of contact. Furthermore, each parabolic pencil
contains exactly one point, i.e., sphere of radius 0. If this point
sphere is not the pencil contains exactly one oriented hyperplane
.
Remark 1.17. Conics on Lies quadric correspond to Dupin cyclides,
that are the envelopes of two one-parameter families of spheres.
The group of Lie transformations shows up as the group of projective
. This group is isomorphic to O(n + 1, 2)/ 1,
automorphisms of Ln+1
1
see Cecil [14]. Since the Pin group of the Cliord algebra C(n+1,2,0)
is a double cover of O(n + 1, 2) we can use this group to describe Lie
transformations.
1.7.2 The homogeneous Cliord Algebra Model

corresponding to Lie Sphere Geometry
In this section we discuss the Cliord algebra model for Lie Sphere
Geometry in the three-dimensional case. Therefore, the projective
space we are dealing with is a ve-dimensional space P5 (R). The homogeneous Cliord algebra model is obtained with the six-dimensional
75
real vector space R6 as a model for the projective image space together
with the quadratic form of Lies quadric
1
0
0
Q=
0
0
0
0
1
0
0
0
0
0
0
1
0
0
0
0
0
0
1
0
0
0 0
0 0
0 0
.
0 0
1 0
0 1
This algebra has signature (p, q, r) = (4, 2, 0) and is of dimension

26 = 64. Again the advantage of the Cliord algebra lies in the common description of the application of Lie transformations. Arbitrary
projective subspaces of P5 (R) are transformed by the sandwich operator. As an example we determine all Lie inversions that leave the
point at innity xed, i.e., the subgroup of Laguerre transformations.
The point at innity has the form p = e1 e2 , compare to Table 1.3.
A general invertible grade-1 element is given by
a = a1 e1 + a2 e2 + a3 e3 + a4 e4 + a5 e5 + a6 e6 , a21 + a22 + a23 + a24 + a25 a26 = 0.
The application of the sandwich operator to p results in

(a)pa1 = 2(a23 + a24 + a21 + a22 a26 + 2a2 a1 + a25 )e1
2(a23 a24 + a26 + a21 + a22 + 2a2 a1 a25 )e2 4(a1 + a2 )a3 e3
4(a1 + a2 )a4 e4 4(a1 + a2 )a5 e5 4(a1 + a2 )a6 e6 .
To guarantee that this entity represents the point at innity, we rst

see that a1 + a2 = 0. With this condition the coecients of e3 , e4 , e5 ,
and e6 vanish. Moreover, the sum of the coecients e1 and e2 has to
vanish. This results in
4a21 4a22 8a2 a1 = 4(a1 + a2 )2 = 0.
Therefore, the only condition to a Lie inversion that it represents a

Laguerre transformation is given by a1 + a2 = 0 and the subgroup of
Laguerre transformations is generated by all vectors with a1 + a2 = 0.
Remark 1.18. Analogue to Th. 1.3 we can formulate a similar
theorem for Lie sphere geometry in arbitrary dimensions. Since the
76
projective model space for n-dimensional Lie sphere geometry has

dimension n + 2, the vector space for the homogeneous Cliord algebra
model has dimension n + 3. That means the highest grade is equal
to n + 3, and therefore, every group element can be written as the
composition of n + 3 grade-1 elements at the most. Especially for the
case n = 3, we have similar results as for the Klein quadric. In this
case six involutions are necessary to generate the whole group.
Let us reformulate this remark as theorem.
Theorem 1.4. Every Lie transformation in n-dimensional space is
the composition of n + 3 involutions, that correspond to the sandwich
action of grade-1 elements.
1.8 A Cliord algebraic Approach to Studys

Quadric
Studys quadric was discussed in section 1.2.2. To obtain a Cliord
algebra description of the Study model we take the eight-dimensional
real vector space R8 as a model for the seven-dimensional real projective space P7 (R) together with the quadratic form corresponding
to Studys quadric

Q=
O
I
I
,
O
where I denotes the 4 4 identity matrix and O the 4 4 zero matrix.

The corresponding Cliord algebra has signature (p, q, r) = (4, 4, 0)
and is of dimension 28 = 256. Null vectors correspond to points on
Studys quadric. Subspaces of P7 (R) of higher dimensions can be obtained by
the wedge product of grade-1 elements, as in the Grassmann

algebra P7 (R). The additional inner product structure allows us to
characterize points on the quadric as null vectors. With the sandwich
operator it is possible to apply automorphisms of Studys quadric to
subspaces. Especially, conics dened by intersections of two-spaces
with the quadric can be transformed at once.
1.9 A Cliord algebraic Approach to Studys Sphere
77
1.9 A Cliord algebraic Approach to Studys

Sphere
The same construction that was done for Kleins, Lies, and Studys
quadric can also be applied to Studys sphere, since it is a quadric over
the commutative ring of dual numbers. Furthermore, it is possible to
construct an ane and a projective model for Studys sphere SD2 .
Projective Model The Cliord algebra is dened by the module
M = D4 that serves as model for the projective space P3 (D) and the
quadratic form Q = M belonging to the unit sphere
1
0
2
T
3
SD : x Mx = 0, with X = xD P (D) and M =
0
0
0
1
0
0
0
0
1
0
0
0
.
0
1
Oriented lines are represented by null vectors.

Ane Model
The
model is obtained by using the module
ane
1 0 0
M = D3 and Q = 0 1 0. In this case oriented lines are represented
0 0 1
by grade-1
elements
that square to 1 respectively Pin group elements

of the 1 V subspace.
Remark 1.19. To obtain a Cliord algebra model of Studys sphere

of oriented displacements we have to increase the dimensions by one.
Dierent dual unit quaternions that describe the same displacement
correspond to dierent points in this model. For the projective model
we need M = D5 as model for P4 (D) and
1
0
Q=
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
1
78
as the matrix of the quadratic form for the projective model. Oriented
displacements correspond to null vectors in this model. For the ane
model we need M = D4 and
1
0
Q=
0
0
0
1
0
0
0
0
1
0
0
0
.
0
1
In this case oriented displacements correspond to grade-1 elements

that square to 1 respectively to Pin group elements from the 1 V
subspace.
1.10 Quadric Geometric Algebra

In this section we present a new geometric algebra model that was
introduced by Zamora-Esquivel [68]. With this model a generalization of the conformal model is achieved. We discuss the geometric
objects that can be represented. We extend this model and identify
the Pin group of this geometric algebra as the group of inversions
with respect to quadrics in principal axis position except for a translation. Moreover, we discuss the construction for the two- and threedimensional case in detail and give the construction for arbitrary dimension. We start with the planar, i.e., two-dimensional case before
we move on to higher dimensions.
1.10.1 The Embedding

The quadric geometric algebra for the two-dimensional case is constructed with a Cliord algebra over a six-dimensional vector space
V = R6 . In fact the term quadric could be replaced by the term
conic for the two-dimensional case. Without loss of generality we call
it quadric geometric algebra and abbreviate this term with QnGA,
where n denotes the dimension of the base space. The quadratic
form we are using is derived by the quadratic form of the conformal
geometric algebra used in [18]:
0
0
1
Q=
0
0
0
0
1
0
0
0
0
1
0
0
0
0
0
79
0
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
The signature of the resulting algebra is (p, q, r) = (4, 2, 0). For every
is performed,
axis, i.e., the x- and the y -axis a conformal embedding

see [19]. Therefore, we have the embedding : R2 1 V ,
R2 P p
1
V,
1
1
P = (x, y)T e1 + xe2 + x2 e3 + e4 + ye5 + y 2 e6 = p.
2
2
(1.51)
Ane points P = (x, y)T R2 are embedded as null vectors. This

means
(P )2 = 0 for P R2 .
(1.52)
The projection on the generator subspace spanned by e1 , e2 , and e3
is denoted by subscript x and the projection on e4 , e5 , e6 by subscript
y . Due to the fact, that the embedding is conformal (see [19, 68])for
both axes we get the additional conditions:
(P )2x = 0,
(P )2y = 0.
(1.53)
In the following we call grade-1 elements satisfying (1.52) and (1.53)

embedded points. Let P1 = (x1 , y1 )T R2 and P2 = (x2 , y2 )T R2 be
two point. The inner product of their images under results in
1
1
1
1
(P1 ) (P2 ) = x22 + x1 x2 x21 y22 + y1 y2 y12
2
2
2
2

1
= (x2 x1 )2 + (y2 y1 )2
2
1
= d2E (P1 , P2 ),
2
where dE (P1 , P2 ) denotes the Euclidean distance between the points

P1 and P2 . Note that this formula only is true for normalized null
vectors. This means that the homogeneous factors have to be equal
80
to one. These vectors can be interpreted as images of points under

and we call them normalized. They are characterized by
e3 p = 1,
e6 p = 1.
(1.54)
The inner product of an arbitrary embedded point with e3 or e6 is

constant. Therefore, we can interpret these elements as points at
innity. Furthermore, the combination of the conditions (1.54) results
in
(e3 + e6 ) p = 2.
Thus, the geometric entity corresponding e3 + e6 can be interpreted

as point at innity. The elements e3 and e6 represent the ideal points
corresponding to each axis and e3 + e6 represents a point at innity
contained in both axes. Geometrically, these three algebra elements
describe the same point, i.e., the point at innity although they
dier algebraically. There are grade-1 elements that satisfy the conditions (1.52) and (1.53) without having a pre-image in R2 . For example
e3 , e6 , e3 + e6 and algebra elements of the form:
1
u1 = e1 + x0 e2 + x20 e3 + e6 ,
2
1
u2 = e3 + e4 + y0 e5 + y02 e6 .
2
If we determine the Euclidean distance of an embedded point to u1

or u2 , the result is a complex number and depends on x0 respectively
y0 . Hence, these elements do not represent points.
1.10.2 Geometric Entities

To calculate the pre-image 1 (p) of p 1 V representing an embedded point, i.e., an algebra element fullling (1.52) and (1.53), we
recall the denition of the GIPNS and the GOPNS, see [53].
The geometric inner product null space (GIPNS) and dual the geo
metric outer product null space (GOPNS) of a k-blade A k V in
Q2GA is dened by

NIG (A) := (x, y)T R2 : (x, y) A = 0 ,

NOG (A) := (x, y)T R2 : (x, y) A = 0 .
81
Remark 1.20. When dealing with an algebra element and the corresponding geometric entity, we mention explicitly what null space is
meant. For example we talk about inner product conics. This means
the inner product null space denes a conic in R2 .
Before we start the examination of geometric objects occurring in this
model we dene special 5-blades that are necessary to change from
inner product to outer product null spaces and vice versa.
Denition 1.19. On the one hand the 5-blade
I = e2 e5 e1 e4 (e3 + e6 )
maps outer product null spaces to inner product null spaces

i
ki
I:
V
V,
ki
i
V v v I
V,
with i {1, . . . , 4} and k = 5 for the planar quadric geometric algebra.

On the other hand
I := e2 e5 e3 e6 (e1 + e4 )
maps dual elements to normal elements respectively inner product null

spaces to outer product null spaces. There is no dierence in leftor right multiplication with these 5-blades. The result diers by the
factor 1 and describes the same geometric entity.
With Def. 1.19 we get (see [68])
NIG (A) = NOG (A I),
NOG (A) = NIG (A I ).
Note that dualization is realized with the inner product. Now we take
a look at the inner product null space of grade-1 elements that are not
embedded points. Therefore, at least one of the conditions (1.52) or
(1.53) is not satised. Let c = 2a1 e1 +2a2 e2 a3 e3 2a4 e4 +2a5 e5 a6 e6
be a general grade-1 element. The GIPNS results
82

NIG (c) = (x, y)T R2 | (x, y) c = 0

= (x, y)T R2 | a1 x2 + 2a2 x + a3 + a4 y 2 + 2a5 y + a6 = 0 .
The GIPNS is a conic in principal position, because there is no term

containing xy . Any conic has a coecient matrix and is given by

a11
y a12
a13

a12
a22
a23

a13
1
a23 x = 0.
a33
y
Therefore, we can dene a bijection between those symmetric matrices

which represent conics in principal position and vectors by
a0
a2
a5
a2
a1
0
a5
0 2a1 e1 2a2 e2 +a3 e3 +2a4 e4 2a5 e5 +a6 e6 .
a4
For the bijection (1.55) we assume that a3 := 21 a0 and a6 := 12 a0 . It

would be sucient to demand that a3 + a6 = a0 to result in the same
conic, because the constant value is equal to a3 + a6 . This does not
change the GIPNS of the conic. With Eq. (1.55) embedded points
can be interpreted as circles whose radii are equal to zeor.
After dualization an inner product conic c becomes an outer product
conic c that is a four-blade and can be generated by the outer product
of four embedded points. Hence, these four points lie on the conic
because
pi NOG (p1 p2 p3 p4 ), for i = 1, . . . , 4.
The natural question that arises is: Is there a way to classify conics
in this model? For this purpose we study the incidence of the conics
with the three additional ideal points. If a conic contains both ideal
contains also e3 + e6 . First we
elements e3 and e6 it automatically

look at the entities a 1 V that contain the ideal points e3 , e6 , and
therefore, also e3 + e6 . Thus, we get the conditions
a e3 = 2a1 = 0,
a e6 = 2a4 = 0.
Hence, a1 and a4 have to vanish. The corresponding algebra element

has the form
l = 2a2 e2 a3 e3 + 2a5 e5 a6 e6 .
83
Its GIPNS is calculated by

NIG (l) = (x, y)T R2 | 2a2 x + 2a5 y + a3 + a6 = 0 .
Clearly, this entity describes an inner product line and every line
passes through e3 , e6 , and e3 + e6 . An algebra element that contains
just e3 or e6 is a parabola whose axis are parallel to the x-axis or the y axis. An element that contains e3 +e6 , but neither e3 nor e6 , is given by
the condition a (e3 + e6 ) = 2a1 + 2a4 = 0. This means a1 = a4 and the
corresponding conic is an equilateral hyperbola, i.e., the asymptotes
enclose an angle of 90 . All other conics in principal axes position
can be obtained by the wedge product of four embedded points or by
using the bijection between conics and the algebra elements, cf. Eq.
(1.55).
Remark 1.21. Note, that this description of conics also contains
conics without real points.
In the most general case two-blades correspond to inner product point
quadruples. This can be seen from
NIG (a b) = NIG (a) NIG (b),
see [53]. Therefore, two-blades represent all points belonging to both

conics that are represented by the vectors. If two non-degenerate conics do not intersect, the corresponding two-blade represents a complex
inner product point quadruple. Furthermore, we can see by dualization, that three-blades correspond to outer product point quadruples.
For two inner product lines l1 , l2 the corresponding two-blade l1 l2
represents an inner product pair of points, where one of the points is
their ane intersection point and the other the point at innity.
Example 1.11. Let us generate a conic through four points. Therefore, we choose four points and embed them via (1.51).
1
P1 = (1, 0)T p1 = e1 e2 + e3 + e4 ,
2
1
T
P2 = (1, 0) p2 = e1 + e2 + e3 + e4 ,
2
1
T
P3 = (0, 1) p3 = e1 + e4 e5 + e6 ,
2
84

1
P4 = (1, 0)T p4 = e1 + e4 + e5 + e6 .
2
The corresponding inner product representation is calculated by

c = I (p1 p2 p3 p4 ) = 4e1 e3 + 4e4 e6 .
The GIPNS is given by

NIG (c) = (x, y)T R2 | x2 + 1 y 2 = 0 .
With the bijection (1.55) we see easily that c is the image of the
conic given by the diagonal matrix diag(1, 1, 1), i.e., the unit circle
centered at the origin.
1.10.3 Transformations
In this section we discuss transformations in this algebra. The Clifford algebra C(4,2,0) corresponds to the quadratic space R(4,2) , and
therefore, the sandwich action of vectors represents reections in hyperplanes in this space. It is clear that the transformations induce
transformations that are not linear in the base space R2 because the
embedding is quadratic. From the last section we know that the
geometric entity corresponding to vectors are conics in principal position. Furthermore, a transformation acts via the sandwich operator
and results again in a k-blade when applied to a k-blade k = 1, . . . , 4.
We begin with an example:
Example 1.12. Let c be the circle from Ex. 1.11
c = 4e1 e3 + 4e4 e6
1
2 e3
and let p = e1 + e2 +
+ e4 + 2e5 + 2e6 be (1, 2)T . Applying the
sandwich operator to p results in
1
p = (c)pc1 = 5e1 + e2 e3 + 5e4 + 2e5 + e6 .
2
Now we check if this entity is still an embedded point. Therefore, the

conditions (1.52) and (1.53) have to be checked. Condition (1.52) is
satised, but condition (1.53) not
p x = 6,
2
p y = 6.
2
85
Hence, p cannot be interpreted as embedded point. The GIPNS of p

is given by
NIG (p ) =

5
5
1
(x, y)T R2 | x2 + x y 2 + 2y = 0 .
2
2
2
This represents a pair of complex lines intersecting in the real point

( 51 , 25 )T .
Ex. 1.12 shows that in general a conic is mapped to another conic.
We can not map a circle of radius zero to a circle of radius zero and
dene a mapping for points on this way. Thus, we study the action
of the transformations applied to vectors that correspond to conics.

Theorem 1.5. A conic represented by the vector a 1 V is xed

pointwise under the transformation induced by itself. Furthermore,
these transformation are involutions.
Proof. First, we show that the conic corresponding to the transformation is xed pointwise. Therefore, we look at the action of a general
vector
a = 2a1 e1 + 2a2 e2 a3 e3 2a4 e4 + 2a5 e5 a6 e6
to itself, and nd
(a)aa1 = (a) = a.
Multiplication with a homogeneous factor does not change the GIPNS.

Thus, the result is the conic represented by a again. To show that the
points of the conic a are xed under the transformation induced by
a, we examine the action of a on the intersection points of the conic
a with all lines containing the point (0, 0). These lines are given by
l(x, y) = I ((0, 0) (x, y) e3 e6 )

1 2
1 2
= I (e1 + e4 ) (e1 +xe2 + x1 e3 +e4 +ye5 + y e6 ) e3 e6
2
2
= 2ye2 2xe5 .
The intersection of all these lines with the conic a is represented by
86

l(x, y) a =4a1 ye12 4a4 ye24 4a1 xe15 + 4(a5 y + a2 x)e25
2a3 ye23 2a3 xe35 2a6 ye26 4a4 xe45 + 2a6 xe56 .
This two-blade represents the pair of common points of the conic and
l(x, y). The application of the transformation induced by a to l(x, y)a
results in
(a)(l(x, y) a)a1 =4a1 ye12 4a4 ye24 4a1 xe15 + 4(a5 y + a2 x)e25
2a3 ye23 2a3 xe35 2a6 ye26 4a4 xe45 + 2a6 xe56 .
This shows, that all pairs of common points of the pencil of lines with
the conic are xed, and therefore, the whole conic is xed pointwise.
To see that the transformation is an involution we have to apply it
twice to an arbitrary k-blade B
(a)(a)Ba1 a1 = (a2 )B(a2 )
= B.
The last equality follows because a2 is a real number.

Due to the fact that these transformations are represented as reections with respect to hyperplanes in R(4,2) , they are involutions and
x the corresponding hyperplane pointwise. This is the reason why
we interpret these transformations as reections or inversions with
respect to conics. Furthermore, the whole group of transformations
is generated by the action of vectors. Note that the image of a conic
in principal position is always a conic in principal position in this
model and that intersection point quadruples of a conic with the reection conic stay xed, no matter if the intersection points are real
or complex.
Remark 1.22. The group of conformal transformations of a quadratic space R(p,q) can be described as the Pin group of a Cliord algebra
C(p+1,q+1,0) , see [55]. Therefore, the group of conformal transformations of the Minkowski space R(3,1) is isomorphic to the group of
inversions with respect to conics in principal position except for a
translation. Especially for the planar quadric geometric algebra Q2GA
we have the signature (p, q, r) = (4, 2, 0) which is identical to the signature of the homogeneous model where we choose Lies quadric as
metric quadric. Thus, the Pin group of Q2GA is isomorphic to the
group of Lie transformations.
87
1.10.4 Eect on Lines and Points

In Ex. 1.12 we have seen that a circle with radius zero is mapped to a
pair of complex conjugate lines intersecting in a real point. Therefore,
we have to search for a better description of points in this model. One
way to describe points as two-blades is to examine the intersection of
two lines. If we take just the ane point of intersection, we can dene
an embedding of the ane plane as points of intersection of pairs of
lines. Therefore, we take two lines through a given point (x, y)T R2 .
We dene this point to be the point of intersection of the line parallel
to the x-axis and the line parallel to the y -axis.
p = (I ((x, y) (x, 0) e3 e6 )) (I ((x, y) (0, y) e3 e6 ))
= y 2 xe23 2yxe25 yx2 e35 + yx2 e56 y 2 xe26
= 2e25 + x(e35 e56 ) + y(e23 + e26 ).
(1.55)
Note that this element represents a pair of points, since lines intersect
also in the point . We parametrize ane lines as the sets of all lines
passing through two points p1 = (x1 , y1 )T and p2 = (x2 , y2 )T . The inner
product line is derived by
l(p1 , p2 ) = I ((p1 ) (p2 ) e3 e6 )
(1.56)
= 2(y1 y2 )e2 +(x1 y2 y1 x2 )e3 +2(x1 x2 )e5 +(x1 y2 y1 x2 )e6 .
The GIPNS of this line is determined by

NIG (l(p1 , p2 )) = (x, y) R2 | (x1 x2 )y+(y2 y1 )x+(y1 x2 x1 y2 ) = 0 .
Theorem 1.6. The image of a line under an inversion

with respect

to a conic represented by the non-null vector a 1 C(4,2,0) is a conic.
Moreover, for non-degenerate conics this conic is the image of a under
an ane transformation, i.e., translation and scalar multiplication.
Proof. To show this we concentrate on conics with no terms in x
and y . We can do this because we are just interested in the type of
the image conic. Furthermore, we can perform translations by two
reections in parallel lines, and thus, we can carry over the results
from the principal position to an arbitrary position. Furthermore, we
88
just show this theorem for non degenerate conics. A conic with no
terms in x or y is given by
1
1
a = 2a1 e1 + a0 e3 + 2a4 e4 + a0 e6 .
2
2
Since we are interested in real conics, the coecients a0 , a1 , and a4

are not allowed to have the same sign. In the sequel we assume that
the conic is real. Now we look at the matrix of the conic
1 0
0 a1
M = a0
0 0
0
.
a4
a0
The image of the set of lines (1.56) is calculated by

4a1 (x1 y2 y1 x2 )
e1 2(y1 y2 )e2
a0
4a4 (x1 y2 y1 x2 )
e4 +2(x1 x2 )e5 .
a0
(a)l(p1 , p2 )a1 =
The coecient matrix of the corresponding conic is given by
a0 (x1 x2 )
0
a0 (y2 y1 )
1
.
N(p1 , p2 ) =
0
a0 (y2 y1 ) 2c1 (x1 y2 y1 x2 )
a0
0
2a4 (x1 y2 y1 x2 )
a0 (x1 x2 )
From this representation we see immediately that lines through the

center of the conic are xed, but not pointwise. In order to transform
this matrix to diagonal form we apply the transformation
1 0 0
a (y1y2)
p 1 0 p, with = 2a (x0 y y x ) and =
1 1 2
1 2
0 1

!

a0 (x1x2)
2a4 (x1 y2 y1 x2 ) .
=:T
Here p = (1, x, y)T R3 is a point in the projective plane. The action

of this coordinate transformation applied to the coecient matrix of
the conic yields
N (p1 , p2 ) = TT N(p1 , p2 )T1
89
2
2
a2
0 (a4 (y1 y2 ) +a1 (x1 x2 ) )
2a4 a1 (x1 y2 y1 x2 ))
2a1 (x1 y2 y1 x2 )
a0
2a4 (x1 y2 y1 x2 )
a0
If we look at the ane part of the conic, we see that this results in
1
N (p1 , p2 ) = 0
0
0
ka1
0
0
2
4a2
1 (x1 y2 y1 x2 ) a4
.
0 , with k = 2
2
a0 (a4 (y1 y2 ) +a1 (x1 x2 )2 )
ka4
Therefore, the image is identical to the conic corresponding to a

except for a translation T and a scaling k. Furthermore, lines are
mapped to real conics.
To illustrate this, Fig. 1.2 shows the reections of three intersecting
lines with respect to a circle, an ellipse, a parabola, and a hyperbola.
The inversion conic ai , i = 1, . . . , 4 is shown in red while the pairs (line,
image of the line) are presented in another colour (but the same). The
inversion conics are from left to right and from up to down given by
a1 : x2 + y 2 = 1,
a3 : x2 y = 1,
25 2 16 2
x + y = 1,
16
25
3 2
2
a4 : x y = 1.
4
a2 :
1.10.5 Subgroups
In this section we examine some subgroups that are embedded naturally in the Pin and the Spin group of Q2GA.
Rotation First, we concentrate on the group that is generated by
inversions with respect to lines passing through the origin. Therefore,
we study the action of these mappings applied to points embedded
via (1.55). Two lines through the origin may be represented by
l1 = I ((0, 0) (cos , sin ) e3 e6 ) ,
l2 = I ((0, 0) (cos , sin ) e3 e6 ) .
90
l2
l2
c2
c1
e3
e1
c2
e2
o
a1
l1
l3
l1
a2
l3
h3
l2
l2
p2
h1
l1
h2
p1
a3
a4
p3
l3
l1
l3
Figure 1.2: Inversions with respect to conics.
Furthermore, we are interested in orientation preserving transformations, i.e., elements from the Spin group. The composition of two
reections in l1 and l2 is given by their geometric product
l1 l2 = (sin sin + cos cos ) + (cos sin sin cos )e25
and with the addition theorems for sine and cosine we conclude
l1 l2 = cos( ) + sin( )e25 .
(1.57)
The square of e25 is 1. This means that the consecutive reection

in two lines through the origin results in an algebra element that
91
can be interpreted as complex number with norm equal to 1. It

is a well-known result that rotations in the plane can be described
by normed complex numbers. Let us look at the action of such an
element applied to a point that is described by (1.55). Let R =
cos + sin e25 be an element in the form of Eq. (1.57) and p =
2e25 + x0 (e35 e56 ) + y0 (e23 + e26 ) a point of the form (1.55). We
compute
p =(R)pR1 = RpR1
=2e25 + (2 sin cos x0 + 2 cos()2 y0 y0 )(e23 + e26 )
+(2 sin cos y0 + 2 cos()2 x0 x0 )(e35 e56 ).
The GIPNS of this entity can be computed or we can simply read the
coordinates of the image point.
x = 2 cos()2 x0 + 2 cos sin y0 x0 = cos(2)x0 + sin(2)y0 ,
y = 2 cos()2 y0 2 cos sin x0 y0 = cos(2)y0 sin(2)x0 .
Therefore, we can see that this transformation, indeed, is a rotation

about the origin with the rotation angle 2. So these elements constitute a double cover of the group SO(2).
Remark 1.23. From the fact that vectors are mapped to vectors by
a reection in a line it follows that axis aligned conics have to be
mapped to axis aligned conics. Therefore, these mappings can not be
interpreted pointwise for conics.
Translations Now we aim at the group of planar Euclidean displacements SE(2). Therefore, we show that two consecutive reections in
parallel lines result in a translation. The group of planar Euclidean
displacements can be generated as the semi-direct product of SO(2)
and T(2), which describes the abelian translation group. Let l1 and
l2 be two parallel lines and let t1 , t2 be their distances from the origin.
The lines are given by
l1 (, t1 ) = 2 sin e2 t1 e3 2 cos e5 t1 e6 ,
l2 (, t2 ) = 2 sin e2 t2 e3 2 cos e5 t2 e6 .
92
The composition can be expressed with the geometric product as

(1.58)
T(, t1 , t2 ) = l1 (, t1 )l2 (, t2 )
= 2 + (t1 t2 ) sin (e23 +e26 )+(t1 t2 ) cos (e35 e56 ).
Applying the sandwich operator to a point p results in

(T)pT1 = TpT1
= 2e25 + (y0 2t2 cos + 2t1 cos )(e23 + e26 )
+ (x0 2t1 sin + 2t2 sin )(e35 e56 ).
The image is determined by

x = x0 sin (2(t1 t2 )),
y = y0 + cos (2(t1 t2 )).
Therefore, the transformation is a translation in the direction normal

to the given lines l1 and l2 .
The Group of planar Euclidean Displacements Translations and
rotations about the origin generate the entire group of planar Euclidean displacements SE(2). Furthermore, we can now examine the
group that is generated by rotations and translations as a subgroup
of the Spin group. These algebra elements have the form
a0 + a1 e25 + a2 (e23 + e26 ) + a3 (e35 e56 ).
The multiplication table of the geometric product for the generators

e0 , e25 ,
e23 + e26 , e35 e56 is given in Table 1.4. Hence, this is indeed a sub-
group of the Spin group. Furthermore, it is isomorphic to a subgroup

of the multiplicative group of dual quaternions, called planar dual
quaternions. We can dene a bijection by
e25 i,
(e23 + e26 ) j,
(e35 e56 ) k.
Remark 1.24. If we restrict ourself to reections in lines and circles,

we are able to describe the group of conformal transformations of the
plane.
93
Table 1.4: Multiplication table of planar displacements in C(4,2,0)

1
e25
e23 + e26
e35 e56
1
1
e25
e23 + e26
e35 e56
e25
e25
-1
(e35 e56 )
e23 + e26
e23 + e26
e23 + e26
e35 e56
e35 e56
e35 e56
(e23 + e26 )
0
0
0
0
Inversions applied to Points In this section we study the action of

reections with respect to a conic in principal position (centered at
the origin) on points. The points are embedded as points of intersection of two lines, as discussed in the previous section. Furthermore,
we have to note that the transformations map point pairs to point
pairs. Hence, the pair of points of intersection of two lines (the afne and the ideal point) are mapped to a pair of points. All lines
pass through and so the image of every line must pass through
the image of this point. The generalization to conics that are not
in principal position is obtained by the application of a coordinate
transformation. The inversion conic is given by
a=
1
c0 (e3 + e6 ) + 2c1 e1 + 2c2 e4 .
2
A point is represented as point of intersection of two lines (see (1.55)

by
p = 2e25 + y0 (e23 + e26 ) + x0 (e35 e56 ).
Applying the sandwich operator to the point results in

p = (a)pa1 =
2y0 c1
2x0 c1
2y0 c2
2x0 c2
e12 +
e15 2e25 +
e24 +
e45 .
c0
c0
c0
c0
This is not of the form (1.55), and therefore, it is not the representation of the intersection of two lines. The GIPNS is calculated

NIG (p ) = (x, y)T R2 | 2c1 (xy0 x0 y)e1 (y0 c1 x2 +2yc0 +y0 c2 y 2 )e2

2c2 (xy0 x0 y)e4 + (x0 c1 x2 +x0 c2 y 2 +2xc0 )e5 = 0
The solution set of this GIPNS can be written as

x=
2c0 x0
,
c1 x20 + c2 y02
y=
2c0 y0
.
c1 x20 + c2 y02
94
Note that we excluded the solution x = 0, y = 0, that is the image of

under the inversion.
Remark 1.25. Inversion with respect to conics that are not in principal position can be performed by the composition of an inversion
and a rotation. Note, that this rotation has to be applied pointwise.
1.10.6 Generalization to higher Dimensions

The main advantage of this geometric algebra model is its exibility.
It is no problem to change the dimension. We discuss the model for
the n-dimensional case and we show some examples for the threedimensional case. The construction is done in the same way as in
section 1.10. We start with a real vector space of dimension n. For
each axis we use a conformal embedding. Therefore, the dimension
of the geometric algebra is 23n and its quadratic form is given by
Q=
..
.
D
0 0 1
D = 0 1 0 .
1 0 0
n-times
The embedding is realized by

: Rn
1
(1.59)
V,
1
1
(x1 , . . . , xn ) e1 +x1 e2 + x21 e3 +. . .+e3n2 +xn e3n1 + x2n e3n .
2
2
The conditions for an embedded point (1.52) and (1.53) generalize to

(P )2 = 0,
(P )2x1 = 0,
(P )2x2 = 0, . . .
(P )2xn = 0.
We dene analogue to Def. 1.19:

Denition 1.20. The blade I that maps outer product null spaces to
inner product null spaces is dened by
n

I=
i
i=1
mod 3=2
n

ei
j
j=1
mod 3=1
n

ej
k
k=1
mod 3=0
ek .
95
Inner product null spaces can be mapped to outer product null spaces
with the blade
n

I :=
i=1
mod 3=2
n

ei
j
n

ej
j=1
mod 3=0
ek .
k=1
mod 3=1
Grade-1 elements correspond to inner product hyperquadrics in principal position. As the dimension is growing, the number of objects
that can be represented grows, too. Blades of grade k, (k n) correspond to the intersection of k hyperquadrics.
Quadrics in three-dimensional Space To construct the quadric geometric algebra for the three-dimensional space we use the quadratic
space R(6,3) given by the nine-dimensional real vector space R9 together with the quadratic form
Q=
D
D
0 0 1
D = 0 1 0 .
1 0 0
For three dimensions the embedding , see Eq (1.59), has the following form
: R3
1
V,
1
1
1
(x, y, z)T e1 + xe2 + x2 e3 + e4 + ye5 + y 2 e6 + e7 + ze8 + z 2 e9 .
2
2
2
The conditions for an embedded point (1.52) and (1.53) generalize to

(P )2 = 0,
(P )2x = 0,
(P )2y = 0,
(P )2z = 0.
Moreover, the blades I and I are given by

I = (e2 e5 e8 ) (e1 e4 e7 ) (e3 + e6 + e9 ),
I = (e2 e5 e8 ) (e3 e6 e9 ) (e1 + e4 + e7 ).
The corresponding geometric algebra has dimension 29 = 512. Any

quadric in principal position except for translation in R3 is uniquely
96
determined by six values. This can be seen from the symmetric matrix of the equation of the quadric that has, in general, ten free entries.
The fact that we are treating quadrics in principal position reduces
the number of free entries to seven. Furthermore, this matrix representation is homogeneous, and therefore, we have six degrees of
freedom. In analogy to Eq. (1.55) we obtain a bijection that is
dened as
1
1
1
C 2c1 e1 2c2 e2 + c0 e3 +2c4 e4 2c5 e5 + c0 e6 +2c7 e7 2c8 e8 + c0 e9 ,
3
3
3
with
c0
c 2
C=
c5
c8
c2
c1
0
0
c5
0
c4
0
c8
0
.
0
c7
As we did for the planar case, we can now pay our attention on the
intersection of three planes in order to get a pair of points containing one ane and one ideal point. We choose these planes to be
parallel to the coordinate planes and passing through a given point
P = (x0 , y0 , z0 )T . Expressed in terms of the quadric geometric algebra
C(6,3,0) we get
p = ((x0 , y0 , z0 )T (0, 0, z0 )T (x0 , 0, z0 )T e3 e6 e9 ) I
((x0 , y0 , z0 )T (0, y0 , 0)T (x0 , y0 , 0)T e3 e6 e9 ) I
((x0 , y0 , z0 )T (x0 , 0, 0)T (x0 , y0 , 0)T e3 e6 e9 ) I
= 3e258 + (e358 e568 + e589 )x0 + (e238 + e268 e289 )y0
+ (e235 + e256 + e259 )z0 .
Remark 1.26. A representation of the group of Euclidean displacements SE(3) can be obtained by studying the composition of reections in planes. Planes correspond to vectors that are obtained by
(P1 P2 P3 e3 e6 e9 ) I.
Now we dene an inner product inversion quadric with
9
, 0, 0)T ,
10
3
P4 = (0, , 0)T ,
4
P1 = (
9
, 0, 0)T ,
10
5
P5 = (0, 0, )T ,
4
P2 = (
3
P3 = (0, , 0)T
4
5
P6 = (0, 0, )T
4
97
a
a
Figure 1.3: Inversion of the unit cube with respect to an ellipsoid (left),
inversion with respect to a hyperboloid (right).
a = (P1 P2 P3 P4 P5 P6 ) I
3
3
36
3
25
e1 e3 + 4e4 e6 + e7 e9 .
=
9
8
8
25
8
The GIPNS of a is given by

NIG (a) =

100
16
16

x2 + y 2 + z 2 1 = 0 .
(x, y, z)T R3
81
9
25
In Fig. 1.3 (left) we can see the inversion of the unit cube [1, 1]3
with respect to the ellipsoid dened by a. The image of every face
of the cube, i.e., of every plane is an ellipsoid passing through the
origin. The action of the inversion given by a on pairs of points can
be written as
f (x, y, z) =
452
T
(x, y, z) .
22 (302 y 2 + 252 x2 + 182 z 2 )
Note that the point is mapped to the origin and that the origin is
mapped to . Therefore, we have a map from R3 \ {0} R3 . One
main advantage of this method is that we can calculate the image
ellipsoid of one face of the cube (one plane) directly by applying the
98
sandwich operator to the plane that is expressed as quadric. For example the plane passing through P1 = (1, 1, 1)T , P2 = (1, 1, 1)T , P3 =
(1, 1, 1)T can be expressed as vector by
p = (P1 P2 P3 e3 e6 e9 ) I = 3e2 + e3 + e6 + e9 .
To check that that this indeed the representation of the plane we

compute the GIPNS

NIG (p) = (x, y, z)T R3 | 1 x = 0 .
Applying the sandwich operator to p results in

c = (a)pa1 =
200
32
96
e1 3e2 e4 e7
27
3
25
with GIPNS
NIG (c) =

102
42
42
(x, y, z)T R3 | 2 x2 x + 2 y 2 + 2 z 2 = 0 .
9
3
5
This is one of the ellipsoids displayed in Fig. 1.3 (left). Furthermore, we can intersect two inner product planes p1 , p2 to get an inner
product line that is an edge of the cube. After that we can apply
the sandwich operator to the line and get the intersection curve (an
ellipse) of the two ellipsoids that are the images of p1 , p2 .
Remark 1.27. It is more convenient to compute the sandwich operator by a conjugation instead
inversion. This means we use (a)Xa
k
with a C(p,q,r) and X V . In general, the modied sandwich operator is easier to handle, because computing the inverse of a Cliord
algebra element is extremely expensive. Moreover, we are working in
a projective setting, and therefore, multiplication with a homogeneous
factor does not change the occurring geometric inner product and
outer product null spaces.
The second example in three-dimensional space is a hyperboloid of
two sheets in principal position that is generated by
9
9
a = (P1 P2 P3 P4 P5 P6 ) I = 6e1 + e3 + e4 + e6 + e7 + e9 .
8
8
99
a
a
Figure 1.4: Inversion with respect to a cylinder (left), inversion with

respect to an elliptic paraboloid (right).
Here, we have P1 = (1, 0, 0)T, P2 = (1, 0, 0)T, P3 = (2, 0, 4)T, P4 = (2, 0,4)T,
P5 = (2, 4, 0)T, P6 = (2,4, 0)T . We calculate the GIPNS
NIG (a) =

3
3
(x, y, z)T R3 | x2 1 y 2 z 2 = 0 .
16
16
The mapping applied to pairs of points results in

f (x, y, z) =
16x2
16
(x, y, z)T .
3y 2 3z 2
The image of the cube [1, 3] [1, 1] [1, 1] under this mapping
is shown in Fig. 1.3 (right). Fig. 1.4 (left) shows the image of an
inversion with respect to a cylinder given by a = 3e12e3+3e42e62e9
applied to a cube. The equation of the cylinder is derived as x2+y 2 = 4.
Planes that are not parallel to the axis of the cylinder are mapped to
paraboloids. Another example is presented in Fig. 1.4 (right). The
inversion quadric is an elliptic paraboloid given by a = 3e12e3+12e4
2e6 +6e8 2e9 respectively by x2 +y 2 4z +4 = 0.
2 Chain Geometry over Cliord

Algebras
In the following chapter we give an introduction to chain geometry.
It is not the aim of this work to give a complete treatise of this
topic, we just introduce the concepts we need for our purposes. For
a more detailed introduction the reader is referred to [11]. The roots
of chain geometry can be found in Benz [5]. Benz investigated
projective lines over commutative two-dimensional algebras and the
corresponding chain geometries. A more recent treatise is [33].
To provide the preliminaries we follow [11] in order to dene the setting that is necessary for our purposes. After this introduction to
chain geometry and chain geometric concepts we focus on chain geometry over Cliord algebras. We introduce chain geometries over Pin
and Spin groups. Moreover, we show that the connected components
of the Pin and the Spin group dene sub chain geometries of the Clifford algebra. Furthermore, we show that subgroups of these groups
also dene sub chain geometries. For application to kinematics we
study the chain geometry over the Cliord algebra C(3,0,1) and its
Spin group in detail and classify the occurring chains.
2.1 Chain Geometry

2.1.1 Distance Spaces
In this section we recall the concept of a distance space and give some
examples that we use later on. Projective lines over rings serve as
examples for distance spaces.

102
Denition 2.1. A pair (P, ), where P is a non empty set and

P P is a relation on P is called a distance space if the following
conditions are satised:
(1) is a symmetric relation, i.e., ab ba,
(2) is an anti-reexive relation, i.e., (aa) is valid for a P .
The elements of P are called points and the relation is the distance
relation. Two points a, b P are called distant if ab.
Denition 2.2. Elements a, b that are not distant are called parallel
(a b). The parallel relation is symmetric and reexive.
Examples for distance spaces will be given in the following sections.
2.1.2 The projective Line over an L-algebra

First, we give the denition of an L-algebra.
Denition 2.3. Let L be a subring of R contained in the center C(R),
where
C(R) := {a R | ax = xa for all x R} .
We call R an L-algebra. An L-algebra R is a module over the commutative ring L.
Remark 2.1. In the most cases we deal with algebras over the real
numbers. In this case we have L = R and we call the algebra an
R-algebra.
Denition 2.4. Let R be a K-algebra over the eld K.
(1) For x R we dene e(x) := {a K | x + a R }.
(2) The K-algebra R is called strong, if for all x R the inequality
card(e(x)) > card(K\e(x)) holds.
Denition 2.5. The general linear group GL(R, 2) over a ring R is
dened as the set of invertible 2 2 matrices with entries in R.
With the general linear group we can now dene the projective line
over a ring R.
2.1 Chain Geometry
103
Denition 2.6. The projective line over a ring R is dened as the

set P1 (R) of all cyclic submodules R(a, b) of R2 , where (a, b) is the
rst row of an invertible 2 2 matrix over R:
P1 (R) :=

a
R(a, b) | c, d :
c
b
d

GL(R, 2) .
Such pairs are called admissible, see [32]. Note, that we dene points
of the projective line as left-homogeneous equivalence classes.
Furthermore, we dene equivalence classes of admissible pairs, respectively points.
Denition 2.7. Two points (a, b), (c, d) P1 (R) are equivalent if there
is a unit r R with ra = c and rb = d. This relation is an equivalence
relation and denoted by . For an equivalence class we write R(a, b).
The homogeneous component of the pair is the second coordinate.
This means the algebra element 0 has the projective coordinates U =
R(0, 1), the neutral element 1 has the coordinates V = R(1, 1), and
the projective point W = R(1, 0) corresponds to an ideal point. In the
following we denote points of the projective line P1 (R) with capital
latin letters.
Remark 2.2. If R is a eld, Def. 2.6 describes the classical projective
line.
From Def. 2.6 we see, R(1, 0) is a point, since the 2 2 identity
matrix is in GL(R, 2). If we take a general matrix M GL(R, 2) all
points P P1 (R) can be obtained by R(1, 0) M. Hence, the point set
P1 (R) can also be described as the orbit of R(1, 0) under the action of
GL(R, 2). Two points X = R(x1 , x0 ), Y = R(y1 , y0 ) are called distant
(XY ) if they are complementary, this means if R2 = X Y . Hence,
(P1 (R), ) is a distance space called the projective line over R.
Remark 2.3. We consider only algebras where the left-inverse elements are also right-inverse elements. Thus, every point may be
represented by an admissible pair, see [32].
104
2.1.3 The Projective Linear Group PGL(R, 2)

The group of all linear mappings that map the projective line P1 (R)
onto itself can be described by the general linear group over R. The
kernel of GL(R, 2) is given by

ker GL(R, 2) =
q
0

0
| q C(R) R .
q
Note, that the kernel is equal to the center of the general linear group
ker GL(R, 2) = C(GL(R, 2)).
The projective linear group PGL(R, 2) acting on P1 (R) can be dened

as the quotient group
PGL(R, 2) := GL(R, 2)/C(GL(R, 2)).
Due to the fact, that the projective line is dened as left-module, a

mapping of P1 (R) onto itself can be described with the matrix vector
product (a, b) M, where the admissible pair (a, b) is a row vector.
Remark 2.4. In this chapter we denote the action of an element
PGL(R, 2) as superscript. For example the image of a point
P P1 (R) under is denoted by P .
Theorem 2.1. The group PGL(R, 2) acts on P1 (R) and leaves the
distance relation invariant. This means for PGL(R, 2) and P, Q
P1 (R) we have P Q P Q , where P and Q denote the images
of P and Q under .
A proof of this theorem can be found in [11]. Now we give another theorem that describes the action of PGL(R, 2) on P1 (R) more detailed.
Therefore, we need:
Denition 2.8. Let (P, ) be a distance space and a subgroup of
the group Aut(P, ), i.e., the group of automorphisms of the distance
space (P, ). The action of on (P, ) is called
(1) 2--transitive, if acts transitive on the set of all pairs of distant
points of P ,
2.1 Chain Geometry
105
(2) 3--transitive, if acts transitive on the set of all triples of distant

points of P .
Remark 2.5. A group acts transitive on a set if for each pair of
elements a, b there is a group element that maps a to b.
Theorem 2.2. The group PGL(R, 2) acts 3--transitive on P1 (R).
For a proof we refer to [33].
Table 2.1: Elements of PGL(R, 2) corresponding to addition, multiplication, and reciprocation.
addition
right-multiplic. left-multiplic. reciprocation
x x + t

1 0
R(x, 1)
t 1
x xq

q 0
R(x, 1)
0 1
x qx
x x1

1 0
0 1
R(x, 1)
R(x, 1)
0 q
1 0
Now we want to investigate a subgroup of the projective group that

corresponds to special operations on the ring R. Let R(x1 , x0 ) be an
arbitrary point on P1 (R) then addition, multiplication, and reciprocation in R can be described by the elements of PGL(R, 2) listed in
Table 2.1. These are all permutations of P1 (R) for that the inverse
map is of the same type. The subgroup of PGL(R, 2) generated by
these four elements is denoted by (R).
2.1.4 The projective Line over a Subring

For a chain geometry we need one more ingredient. Therefore, we
look at pairs of rings (L, R) where L is a subring of R. The projective
line over L is a distance space (P1 (L), L ). Hence, we are interested in
the relation between (P1 (L), L ) and (P1 (R), R ). We interpret the
projective line over L as a substructure of (P1 (R), R ). It is clear that
for a pair (a, b) L2 the equivalence classes have the form L(a, b) and
if we interpret the point as element of (P1 (R), R ), the equivalence
classes have the form R(a, b). Hence, we can dene the map
: P1 (L) P1 (R),
L(a, b) R(a, b).
106
This map is an injective morphism of distance spaces, see [11].

Remark 2.6. In general we can not expect that non-distant points
are mapped to non-distant points under . As example we can take a
look at the ring pair (Z, Q), cf. [11, p. 26].
If we equip P1 (L) with the distance relation R the inverse map 1
does not need to be a morphism of distance spaces. A criteria for the
inverse map 1 to be a morphism of distance spaces is given in:
Theorem 2.3. Let L R and let : (P1 (L), L ) (P1 (R), R ) be a
morphism of distance spaces. The following statements are equivalent:
(1) L = R L.
(2) maps non-distant points to non-distant points.
In this case, the map : P1 (L) P1 (L) is a isomorphism of distance
spaces.
2.2 Chain Geometry as Incidence Geometry

In this section we dene the term chain geometry. Therefore, we need
the projective line P1 (R) over an algebra as the point set P of our
chain geometry. Furthermore, there are special subsets of P that we
call chains.
Denition 2.9. Let R be an L-algebra and PGL(R, 2). The subset
1
1
1
P
(L) of P (R) is called a chain in P (R), if is induced by a matrix
a b
GL(R, 2). The set of all chains in P1 (R) is denoted by C(L, R)
c d
and dened as

C(L, R) := P1 (L) PGL(R, 2) .
We call the incidence structure (L, R) := (P1 (R), C(L, R)) chain geometry over the L-algebra R.
107
The set of points of (L, R) is additionally equipped with the distance

relation R . We need that the distance relation R on the chain
P1 (L) coincides with L . With Th. 2.3 this is the case if L = R L.
Hence, we assume this from now on and for R we simply write .
Furthermore, we provide the more general denition of a chain space
that also can be found in [11].
Denition 2.10. Let (P, C) be an incidence structure and a relation
on P such that (P, ) is a distance space.
(1) Then (P, C, ) is called incidence structure with distance relation.
(2) Let (P, C, ) be an incidence structure with distance relation. The
elements of C are called chains. (P, C, ) is called a weak chain
space, if the following axioms hold:
C1 Each chain c C contains at least three points and each point
p P is contained by at least one chain.
C2 Three pairwise distant points p, q, r P are incident with
exactly one chain c C. We denote this chain by c =: (pqr).
A weak chain space (P, C, ) is called a weak chain space in the
proper sense if the following additional axiom is satised.
C3 Two points p, q P are distant if, and only if, they are dierent
and connected by a chain.
We need a further denition.
Denition 2.11. Let A = (P, L, ) be an ane space and let L L
be a non-empty union of sets of parallel lines from A. The incidence
structure (P, L ) is called a partial ane space.
With Def. 2.10 and Def. 2.11, we are able to dene chain spaces.
Denition 2.12. Let = (P, C, ) be a weak chain space in the
proper sense.
(1) For P P let CP := {c\ {P } | P c C}. Let (P ) denote all
points distant from P P . Than P := ((P ), CP ) is called the
residuum of at the point P .
(2) If satises the axiom
108

C4 For each P P , the residuum P is a partial ane space.
is called a chain space.
Remark 2.7. All chain geometries over algebras are also chain spaces,
see [11] or [33]. When we talk about chain spaces we mention chain
geometries over algebras. A chain geometry over an L-algebra R is
denoted by (L, R), where the set of points is given by P = P1 (R) and
the set of chains by

C(L, R) := (P1 (L)) PGL(R, 2) .
In the following we cite some theorems concerning chain geometries

over algebras without proofs. For proofs we refer to [11] and [33].
Theorem 2.4. Let R be an L-algebra with L = R L. Then for
(L, R) it is true: Through three pairwise distant points there is exactly
one chain.
Theorem 2.5. For (L, R) it is equivalent:
(1) Let L = R L. Two pairwise distant points of P1 (R) are distant
if, and only if, they are incident with one chain.
(2) The ring L is a eld.
Lemma 2.1. Let R be an algebra over the ring L. Then the group
PGL(R, 2) is a subgroup of the group of automorphisms Aut((L, R))
of the incidence structure = (L, R). The group acts transitive on
the set F = {(p, c) (P C) | p c C(L, R)} of ags of .
Proposition 2.1. The parallel relation in (L, R) on P1 (R) is transitive, i.e., it is an equivalence relation, if R is a local ring.
A proof can be found in [33].
Denition 2.13. Let K be a eld.
(1) The chain geometry (K, R) is called a Mobius geometry if the
parallel relation is the equality relation. This means that any two
dierent points are incident with a single chain.
109
(2) (K, R) is called a Laguerre geometry provided that the parallel

relation is an equivalence relation on P1 (R) and every chain meets
every parallel class of points.
(3) (K, R) is called a Minkowski geometry of dimension n if
R = K K . . . K (n-times),
where addition and multiplication are dened component-wise.

Remark 2.8. The chain geometry (K, R) is a Laguerre geometry if
R is a Laguerre algebra.
2.2.1 Denition of a Cross Ratio

For a better description of chains and as natural invariant we dene
the cross ratio of four points.
Denition 2.14. For A, B, C, D P1 (R), where A, B, C are mutually
distant and A, D are distant, we dene the cross ratio cr(A, B, C, D)
as a subset of R in the following way:
d cr(A, B, C, D) PGL(R, 2) : (A, B , C , D ) = (U, V, W, R(d, 1)),
with U = R(0, 1), V = R(1, 1), W = R(1, 0).

Every cross ratio is a class of conjugates under R :

d cr(A, B, C, D) cr(A, B, C, D) = z 1 dz : z R .
Furthermore, the cross ratio is invariant under the action of PGL(R, 2):
cr(A, B, C, D) = cr(A , B , C , D ), for all PGL(R, 2).
If d C(R) the conjugacy class cr(A, B, C, D) becomes a single element

and we write cr(A, B, C, D) = d. A theorem taken from [33] states:
Theorem 2.6. (1) Four mutually distant points A, B, C, D P1 (R)
are cocatenal, i.e., incident with one and the same chain, if
cr(A, B, C, D) L .
110
(2) Let A, B, C be mutually distant points. The chain containing

A,B,C is the set

{A} X P1 (R) | XA and cr(A, B, C, X) L .
Th. 2.6 allows a parametrisation of chains with the help of the cross
ratio. Therefore, we take a closer look at the cross ratio. The map
PGL(R, 2) that takes three arbitrary mutually distant proper
points A, B, C P1 (R) to the points U, V, W can be constructed in the
following way:

M = M 1 M2 M3 M4 =

tv
1
v 0
1 0
0 1
,
=
(1 ct)v c
0 1
t 1
1 0
1 0
c 1
with t = (c a)1 and v = (t + (b c)1 )1 . In this product the rst

two matrices M1 M2 put the point C to W , where it is xed under the
action of M3 M4 . The third factor M3 maps the image of A under the
rst two factors to U . The last factor M4 does not aect the image
of A under the rst three matrices. Finally, the last factor has traced
B through the rst three factors and maps the image of B to V . To
check this we apply the mapping described by the matrix M to the
three points:
tv
1
R(a, 1)M = R(a, 1)
(1 ct)v c

= R(atv + (1 ct)v), a c)
= R((a c)t + 1)v, a c) = R((1 + 1)v, a c) R(0, 1) = U,

tv
1
= R(btv + (1 ct)v, b c)
R(b, 1)M = R(b, 1)
(1 ct)v c
= R((b c)t + 1)v, b c) R(b c)t + 1, (b c)v 1 )
= R((b c)t + 1, (b c)t + 1) R(1, 1) = V,

tv
1
= R(ctv + 1 ctv, c c)
R(c, 1)M = R(c, 1)
(1 ct)v c
= R(1, 0) = W.
Let us now consider the action on an arbitrary proper point R(x, 1)
P1 (R):

tv
(1 ct)v
1
c
111
= R (((x c)t + 1)v, x c)

= R (x c)(c a)1 + (c a)(c a)1 )v, x c

= R ((x c) + (c a))(c a)1 v, x c

= R (x a)(c a)1 ((c a)1 + (b c)1 )1 , x c

= R ((c a)(x a)1 )1 ((c a)1 + (b c)1 )1 , x c

= R (((c a)1 + (b c)1 )((c a)(x a)1 ))1 , x c

= R ((x a)1 + (b c)1 (c a)(x a)1 )1 , x c

= R ((1 + (b c)1 (c a))(x a)1 )1 , x c

= R (((b c)1 (b c) + (b c)1 (c a))(x a)1 )1 , x c

= R (((b c)1 ((b c) + (c a))(x a)1 )1 , x c

= R ((b c)1 (b a)(x a)1 )1 , x c

= R (x a)(b a)1 (b c), x c

(2.1)
= R (x a)(b a)1 (b c)(1, 0) + (x c)(0, 1) .

!

!
R(x, 1)
We repeat this calculation for an ideal point R(1, n), n N (R), where
N (R) denotes the set zero divisors of R.
tv
R(1, n)
(1 ct)v
1
c

= R (tv + n(1 ct)v, 1 nc)
= R ((t + n(1 ct))v, 1 nc)

= R (t + n(t1 t ct))v, 1 nc

= R (1 + n(t1 c))tv, 1 nc
= R ((1 na)tv, 1 nc)

= R (1 na)(c a)1 ((c a)1 + (b c)1 )1 , 1 nc

= R ((c a)(1 na)1 )1 ((c a)1 + (b c)1 )1 , 1 nc

= R (((c a)1 + (b c)1 )(c a)(1 na)1 ))1 , 1 nc

= R ((1 + (b c)1 (c a))(1 na)1 )1 , 1 nc

= R (((bc)1 (bc)+(bc)1 (ca))(1na)1 )1 , 1nc

= R ((b c)1 (b a)(1 na)1 )1 , 1 nc

= R (1 na)(b a)1 (b c), 1 nc
112

= R (1 na)(b a)1 (b c)(1, 0) + (1 nc)(0, 1) .

!

!
(2.2)
With the upper calculation we are able to compute the cross ratio as
f (X) = cr(A, B, C, X) = R(x1 , x0 )M = R(, ).
If the cross ratio is a proper element of the projective line, we can

nd normalized representatives of the equivalence class:
X proper:
X ideal:

R(, ) R (x a)(b a)1 (b c)(x c)1 , 1 ,

R(, ) R (1 na)(b a)1 (b c)(1 nc)1 , 1 .
(2.3)
(2.4)
If the cross ratio is an ideal element we can normalize to

X proper:
X ideal:

R(, ) R 1, (x c)(b c)1 (b a)(x a)1 ,

R(, ) R 1, (1 nc)(b c)1 (b a)(1 na)1 .
(2.5)
(2.6)
Remark 2.9. For a proper cross ratio R(, 1) we can calculate the
cross ratio directly from the ring elements by
cr(a, b, c, x) = (x a)(b a)1 (b c)(x c)1 .
(2.7)
This is exactly the cross ratio formula that is used for complex numbers
or in classical projective geometry. If C(R) the cross ratio is an
element and not a conjugacy class.
With the help of Th. 2.6 and the formulas (2.3)-(2.6) we can now
parametrise chains in the chain geometry (L, R)
(d a)(b a)1 (b c)(d c)1 =
(d a)(b a)1 (b c) = (d c)
d(b a)1 (b c) a(b a)1 (b c) = d c
d(b a)1 (b c) d = a(b a)1 (b c) c

d (b a)1 (b c) = a(b a)1 (b c) c

1

.
d = a(b a)1 (b c) c (b a)1 (b c)
(2.8)
113
Properties of the Cross Ratio Let a, b, c R be three pairwise

distant elements and let d R with dc. Furthermore, we assume
that the cross ratio is proper and not a conjugacy class. Therefore,
we can use Eq. (2.7) to compute the cross ratio. It follows
cr(a, b, c, a) = (a a)(b a)1 (b c)(a c)1 = 0,
cr(a, b, c, b) = (b a)(b a)1 (b c)(b c)1 = 1.
and with Eq. (2.1)

cr(a, b, c, c) = (c a)(b a)1 (b c)(1, 0) + (c c)(0, 1)
= (1, 0) = .
Thus, we can use chains parametrised by the cross ratio for interpolation purposes, since the elements a, b, c are interpolated for the cross
ratio values 0, 1, .
Theorem 2.7. Let a, b, c, d R be four mutually distant elements
incident with one chain. The exchange of two pairs of elements does
not change the value of the cross ratio:
cr(a, b, c, d) = cr(b, a, d, c) = cr(c, d, a, b) = cr(d, c, b, a).
All in all there are 24 permutations of the elements a, b, c, d. If we use

the symmetries given above, we have six possible values for the cross
ratios:
cr(a, d, c, b) = 1
cr(a, b, c, d) = ,
cr(a, c, d, b) = (1 )
cr(a, c, b, d) = ( 1)1
(2.9)
cr(b, a, c, d) = 1
cr(a, d, b, c) = ( 1)1 .
Note, that C(R) if a, b, c, d are cocatenal.

Proof. The proof is done be direct computation. We restrict ourselves
to the case where four elements are units. Then, Eq. (2.1) is valid.
If there are zero divisors, we have to take Eq. (2.4). The rst four
symmetries can be proofed in the same way. As an example we show
cr(a, b, c, d) = cr(c, d, a, b) and we use that cr(a, b, c, d) C(R):
114

cr(a, b, c, d) =(d a)(b a)1 (b c)(d c)1
=(d a)(b a)1 (b c)(d c)1 1
=(d a)(b a)1 (b c)(d c)1
(b c)(d c)1 (d c)(b c)1
=(b c)(d c)1 (d a)(b a)1
(b c)(d c)1 (d c)(b c)1
=(b c)(d c)1 (d a)(b a)1
=cr(c, d, a, b).
Now we show cr(a, b, c, d)1 = cr(c, b, a, d) which is also done by direct

computation.

1
cr(a, b, c, d)1 = (d a)(b a)1 (b c)(d c)1

1
1
= (b c)(d c)1
(d a)(b a)1
= (d c)(b c)1 (b a)(d a)1
= cr(c, b, a, d)
To complete the proof it is sucient to show one more of the identities

(2.9), since all other identities can be achieved by combining two
identities. This can be done be direct calculation and is left to reader
as an exercise.
2.3 Quadric Chain Spaces

Before we dene chain spaces on quadrics we give a short introduction
to quadrics in general, cf. [25].
Denition 2.15. Let V be a vector space over a commutative eld
K. A map : V K is called a quadratic form, if for all x, y V,
K\ {0}:
(1) (x) = 2 (x).
(2) (x, y) (x + y) (x) (y) is a bilinear form : V V K.
2.3 Quadric Chain Spaces
115
Remark 2.10. For a bilinear form : V V K we can explain

as
(x, x) =: (x).
Furthermore, we can dene the bilinear form with

(x + y) = (x + y, x + y) = (x, x) + (x, y) + (y, x) + (y, y)
= (x) + 2(x, y) + (y).
If char K = 2, we can dene the symmetric bilinear form

: V V K : (x, y)
1
((x + y) (x) (y)).
2
In the following we assume char K = 2.

Each quadratic form can be written as (x) = xT Ax, with A = AT .
The corresponding bilinear form reads (x, y) = xT Ay .
Denition 2.16. Let Pn (V ) be an n-dimensional projective space
over K and let : V K be a quadratic form.
(1) The set Q := Q() = {xK Pn (K) | (x) = 0} is called the projective
quadric corresponding to . This is well-dened since (x) = 0
implies that (xk) = 0, k K.
(2) Let l be a line in Pn (K). If |l Q| = 1, l is called tangent. If
|l Q| = 2, l is called secant. If the intersection consists of the
whole line, the line is tangent to and contained in the quadric.
(3) Let p be a point contained by Q. The set
Tp := {q l | l tangent to Q at the point p}
is called tangent space to Q at the point p.

(4) The symmetric matrix A corresponding to the quadratic form
denes a correlation in Pn (K), called a polarity. All points that
are incident with their image hyperplane are contained in Q.
116
Remark 2.11. With the polarity given by A, a point p Q is mapped

to the tangent hyperplane at the point p. Therefore, we can use the
polarity to describe tangent spaces. If the tangent space at a point
s Q is the whole space, s is called double point of Q. Tp is a
hyperplane if the point p Q is not a double point. A line through a
given point p Q is either a tangent or a secant.
With these preliminaries we want to dene a chain space on a quadric.
We follow [11] and assume dim V > 4. Furthermore, we demand that
the quadric Q has at least one secant and that it is not the union of
two hyperplanes.
Denition 2.17. Let Q be the set of points of Q that are not double
points. This set becomes the point set P of the chain geometry. A
plane E is called admissible if the intersection E Q contains at least
three points, but no whole line of Q. The set of all conic sections with
admissible planes is denoted by
C(Q) := {Q E | E admissible}
and becomes the chain set. Hence, we examine the incidence structure
(Q) := (Q , C(Q)) .
Theorem 2.8. The incidence structure (Q) := (Q , C(Q)) is a chain

space.
To proof this theorem, we refer to [11]. We just have to show the
axioms C1-C4.
2.4 Real Benz Planes

In this section we give famous examples for chain geometries. Hence,
we examine chain geometries over two-dimensional algebras over the
real numbers. The famous textbook [5] treated this topic exhaustively. Furthermore, we present the related quadric models. This
section follows the Wikipedia article about Benz planes [67].
There are only three dierent cases of two-dimensional algebras over
R. The algebras are the complex numbers C, the dual numbers D, and
2.4 Real Benz Planes
117
the double numbers also called the split-complex numbers A = R R.

Thus, we have to examine the three dierent chain geometries (R, C),
(R, D), and (R, A).
Theorem 2.9. The chain geometry (R, K) is a
(1) M
obius geometry, if K = C,
(2) Laguerre geometry, if K = D,
(3) Minkowski geometry, if K = A.
A proof can be found in [11]. All these planes arise from the real ane
plane with dierent compactications. The M
obius plane is the ane
plane with an additional point called , i.e., the conformally closed
ane plane. Chains are circles and lines that can be interpreted as
circles with innite radius. A corresponding quadric model is the unit
sphere with its planar sections as set of chains. The stereographic
projection provides a mapping between the planar model and the
quadric model. Chains in the Laguerre plane are parabolas of the
form y = ax2 +bx+c or lines, if a = 0. Here the real ane plane is closed
with a line at innity, i.e., the projective closure. To every parabola
y = ax2 +bx+c we have to add the point (, a), and therefore, the point
:= (, ) is also contained on the ideal line. As adequate quadric
model we have the so called Blaschke cylinder with its planar sections
as chains and as a mapping between the planar and the quadric model
a generalized stereographic projection from the plane to the cylinder.
In the Minkowski plane (R, A) the chains are hyperbolas of the
a
+ c, a = 0 and lines y = mx + n, m = 0. To every line we
form y = xb
add the point (, ) and to every hyperbola we add the two points
(b, ), (, c). Via a generalized stereographic projection we can map
the planar sections of a one-sheeted hyperboloid to this set of chains.
Remark 2.12. All these chains can be parametrised with Eq. (2.8)
and with three algebra elements corresponding to three points the circle,
the parabola, the hyperbola, or the line is determined uniquely.
118
2.5 Jordan-Systems
Now we take a look at sub structures of chain geometries.
Denition 2.18. Let = (P, C) be a chain space and let S P with
(1) S contains three mutually distant points.
(2) If p, q, r S are mutually distant, then the connecting chain c =
(pqr) is entirely contained in S . The set of all connecting chains
is denoted by C(S).
(3) The incidence structure (S, C(S)) is a chain space.
Under this conditions (S, C(S)) is a subspace of . If (S, C(S)) satises
just condition (2), (S, C(S)) is called a weak subspace of .
Trivial subspaces of = (P, C) are itself and subspaces containing
just a single chain c C. More examples can be generated with
subalgebras.
Denition 2.19. Let R be an algebra over the eld K. A subspace S
of R is called a subalgebra of R, if
(1) 1 S .
(2) For a, b S the product ab is in S .
(3) If a S R , then a1 S , i.e, S R = S .
Since chain geometries over algebras are chain spaces, the chain geometries over subalgebras are also chain spaces, and therefore, subspaces of the chain space. It is not common to demand condition
(3). For the description of subspaces of chain spaces this condition
is important. Just with this condition subalgebras correspond to
Jordan-systems. The following theorem states that subalgebras also
dene chain geometries on their own.
Theorem 2.10. Let R be a K-algebra and let S be a subalgebra of R.
The mapping : P1 (S) P1 (R) with S(a, b) R(a, b) is an injective
morphism of chain spaces (K, S) (K, R). The image ((K, S))
is a subspace of (K, R), that is embedded via and isomorphic to
(K, S).
2.5 Jordan-Systems
119

Remark 2.13. We identify ((K, S)) with (K, S).
Example 2.1. As example we take the skew eld of quaternions, see
section 1.1.2. A general quaternion has the form
q = a + bi + cj + dk,
a, b, c, d R.
A subalgebra is dened by the set {a + cj | a, c R}. In fact this subalgebra is isomorphic to the complex numbers C and it denes a subspace
(R, C) of the chain geometry (R, H).
There are also subspaces of chain spaces that do not correspond to
subalgebras. To describe some of these subspaces we need the denition of a Jordan-system.
Denition 2.20. Let R be a K-algebra and let J be a subspace of
the K-vector space R with 1 J . The subspace J is called
J1 Jordan-system in R, if b J R b1 J for all b J , we
dene J = J R .
J2 Jordan-closed in R, if a, b J aba J for all a, b J .
J3 strong in R, if for all b J the condition |e(b)| > |K\e(b)| is
satised, where

e(x) := f K | x + f R .
Remark 2.14. If R is a strong algebra, cf. Def. 2.4, then every

Jordan-system is strong in R.
Lemma 2.2. Let J be a Jordan-closed Jordan-system. For a, b J
the elements a2 and ab + ba are also elements of J . The set J together
with a b = 12 (ab + ba) is called Jordan algebra if char K = 2.
Remark 2.15. The term Jordan-algebra is named after the German
physicist P. Jordan.
Lemma 2.3. Let J be a strong Jordan-system in R. Then J is also
Jordan-closed in R.
120
Now we dene the projective line over the Jordan-system J .

Denition 2.21. Let J be a strong Jordan-system in the K-algebra
R. We dene the projective line over J as
P1 (J ) := {R(1 + ab, a) | a, b J } .
There is a subgroup of PGL(R, 2) that leaves P1 (J ) invariant.

2) that is genDenition 2.22. Let (J
the subgroupof PGL(R,

) be
1 0
0 1
erated by the matrices
. Then (J ) is a
, q J and
q 1
1 0
subgroup of Aut(P1 (R), ) that leaves P1 (J ) invariant.

Theorem 2.11. Let J be a strong Jordan-system in R. If there
are three mutually distant points p, q, r P1 (J ), their common chain
c = (pqr) C(K, R) lies completely in P1 (J ). Every such chain can be
written as c = ch , where (J ), h J and ch := {R(sh, t) | K(s, t)
P1 (K) . Conversely, for (J ), h J the chain ch is contained
in P1 (J ).
Proof. For a proof we refer to [11].
It can be shown that the subspace P1 (J ) is a subspace of (K, R), see
[11]. As distance relation on P1 (J ) we use the distance relation of
P1 (R).
Theorem 2.12. Let J be a strong Jordan-system in R. Then
= (K, R, J ) = (P1 (J ), C(K, R, J ))

with C(K, R, J ) = ch | h J , (J ) is a subspace of = (K, R).
We call (K, R, J ) the chain geometry over (K, R, J ).
Now we generalize the setting and assume that the set S does not
contain the three points R(1, 0), R(0, 1), and R(1, 1). Under these
assumptions S cannot be the projective line over a Jordan-system,
see Def. 2.20. Nevertheless, we can formulate the following theorem,
see [11]:
2.6 Contact Spaces
121
Theorem 2.13. Let R denote a strong K-algebra, = (K, R) and

S a non-trivially connected weak subspace of . Under this conditions
there exists PGL(R, 2) with S = P1 (J ) for a strong Jordan-system
J of R. The subspace (S, C(S)) is a strong subspace that is isomorphic
to (K, R, J ).
Proof. The theorem follows directly from the 3--transitivity of
PGL(R, 2), cf. [11].
2.6 Contact Spaces

Contact spaces are of certain interest on their own. Benz introduced
contact spaces and a contact relation for two-dimensional commutative algebras, see [5]. Furthermore, a survey-like description can be
found in [33]. The contact relation gives a method to compute chains
that are in contact with other chains at a certain point by means of
the cross ratio. First of all, we dene the contact relation following
[33].
Denition 2.23. Let = (P, C) be a chain geometry. A contact
relation on is a family (p ), p P , where each (p ) is an equivalence
relation on (p) := {c C | p c} with the following conditions (we write
aP b for ap b):
(1) aP b and a = b implies |a b| = 1,
(2) for a (p) and q P with p q there is exactly one b (p) with
q b and aP b.
A chain geometry, i.e., a chain space together with a contact relation
is called a contact space.
Remark 2.16. This denition is also adequate for weak chain spaces.
On the partial ane space p the contact relation p induces a parallelism. In every chain space = (P, C) we have the following natural
contact relation:
p P : aP b a\ {p} and b\ {p} are parallel lines in p .
122
Benz [5] dened a contact relation with the help of the residual spaces
without mentioning them explicitly. We also give his denition since
it is more intuitive.
Denition 2.24. Let = (P, C) be a chain geometry, let W =
R(1, 0) P1 (R) and let a, b C. The chains a and b are in contact at the point P if there is a PGL(R, 2), such that P = W and
the lines a \ {W }, b \ {W } are parallel.
We give a theorem from [5]:
Theorem 2.14. Let aP b, a = b. It follows a b = P . The contact
relation p is an equivalence relation, i.e., it is reexive, symmetric,
and transitive.
/ c be points
Contact theorem: Let c C be a chain and let p c, q
with p q . Then there is exactly one chain c incident with p and q
that is in contact with c.
Lemma 2.4. If aP b and PGL(R, 2), then a P b .
The inverse implication is not true in general, but, if there is a socalled beetle-gur (german: Kafergur), then it is true. If = (P, C)
is a chain space, then there is necessarily a beetle-gure. Now we give
two theorems that connect the contact relation with the cross ratio.
Both theorems can be found in [5].
Theorem 2.15. Let R be a K-algebra over a eld and let a, b C
be chains incident with P P1 (R). The following expressions are
equivalent:
(1) aP b,
(2) For two dierent points A, A a\ {P } and for two points B, B
b\ {P }
cr(P, A, A , B) cr(P, A, A , B ) K,
(3) There are two dierent points A, A a\ {P } and two dierent

points B, B b\ {P } with
cr(P, A, A , B) cr(P, A, A , B ) K.
2.7 Chain Geometries over Cliord Algebras
123
The next theorem allows a parametrisation of the chain b that is in

contact to another chain a at the point P .
Theorem 2.16. Let P a, Q
/ a, P Q, then there exists exactly
one chain b incident with P and Q that is in contact with a. This
chain consists of all points X P with
cr(P, A, A , X) cr(P, A, A , Q) K,
where A, A are dierent points on a\ {P }. Therefore,

b = {P } {X P | cr(P, A, A , X) cr(P, A, A , Q) K} .
A proof can be found in [5]. The proof given there is valid for commutative rings, but every statement can be transferred directly to the
non-commutative case. The proof uses Lemma 2.4 and the invariance
of the cross ration under the action of the group PGL(R, 2). With this
theorem we are able to calculate an explicit parametrisation of the
chain that is in contact to a given one.

Benz planes can be interpreted as chain geometries over two-dimensional Cliord algebras, i.e., chain geometries over C(1,0,0)
= C,
C(0,1,0)
= A, or C(0,0,1)
= D. Furthermore, we are dealing with
nite-dimensional Cliord algebras over the real numbers. For these
algebras the chain geometry is well understood, and therefore, we can
apply the methods easily, see [5]. Every Cliord algebra is an algebra
over a commutative ring. The dimension of the center depends on
the dimension of the algebra, see Def. 1.11. The real numbers are
always contained in the center. Therefore, we can always dene a
chain geometry (R, C(p,q,r) ). If the center of the Cliord algebra is
two-dimensional, it is isomorphic to A, C, or D. In these cases we can
dene dierent chain geometries. On one hand, we have (R, C(p,q,r) )
and on the other hand (R, C(p,q,r) ) where R is A, C, or D. The set
of points of the chain geometry is the projective line over the Cliord
algebra
P1 (R) :=

a
R(a, b) | c, d :
c
b
d

GL(R, 2) ,
124
where R denotes the Cliord algebra. The chains are dened as Rchains if not stated otherwise:
C(R, R) :=

P1 (R) | PGL(R, 2) .
Remark 2.17. Especially for dual unit quaternions, chains over

the center C(Hd )
= D are of interest, because the cross ratio of four
elements of one chain results in a dual number.
2.7.1 Grade-1 Subspace

If we are dealing
with a not fully degenerate Cliord algebra, we can

show that the 1 V subspace denes a chain geometry. Therefore, we
proof that another subspace of the Cliord algebra denes a Jordansystem.
After that we use Lemma 2.1 to map the Jordan-system to
1
V subspace. Hence, we proof the following:
the
Theorem 2.17. The subspace I of C(p,q,r) generated by the set
{e0 , e12 , e13 , . . . e1n } is a Jordan-closed and strong Jordan-system.
Proof. To show this we have to check J1-J3.
J1 Since every element of I can be written as the product of two
vectors, it is a versor. The inverse element for a versor v is given
v
by v1 = vv
.
J2
For n = 1 the theorem is clear. Our base case is n = 2. Let q =
q1 0
be the quadratic form that denes the Cliord algebra
0 q2
with q1 , q2 {1, 1, 0}. The set generated by I = {e0 , e12 } is the

set that shall be a Jordan-system. Two elements a = x1 e0 + x2 e12
and b = y1 e0 + y1 e12 satisfy J2:
aba = (x21 y1 (2x1 x2 y2 +x22 y1 )q1 q2 )e0 +(2x1 x2 y1 +x21 y2 x22 y2 q1 q2 )e12 .
Inductive step: Let q =diag(q1 , q2 , . . . , qn ) be the diagonal matrix

dening the qua-dratic form with q1 , . . . , qn {1, 1, 0}. Let
a = x1 e0 + x2 e12 + . . . + xn1 e1(n1) +xn e1n ,

!
A
125
b = y1 e0 + y2 e12 + . . . + yn1 e1(n1) +yn e1n .

!
B
Now we verify that the Jordan-system is Jordan-closed:

aba = (A + xn e1n )(B + yn e1n )(A + xn e1n )
(2.10)
= ABA + Ayn e1n A + xn e1n BA + xn yn e1n e1n A + ABxn e1n

+ Ayn e1n xn e1n + xn e1n Bxn e1n + xn yn e1n e1n xn e1n .
The statement is true for n1, and therefore, ABA I . It remains

to show that the other terms are also in I . We proof by direct
computation that all terms
xn yn e1n e1n xn e1n = x2n yn q1 qn e1n I,
xn e1n Bxn e1n = (x2n y1 e0 + x2n y2 e12 + . . . + x2n yn1 e1(n1) )q1 qn
Ayn e1n xn e1n = (x1 e0 x2 e12 . . . xn1 e1(n1) )xn yn q1 qn ,
xn yn e1n e1n A = (x1 e0 x2 e12 . . . xn1 e1(n1) )xn yn q1 qn ,
Ayn e1n A = (x21 + x22 q1 q2 . . . + x2n1 q1 qn1 )yn e1n ,
xn e1n BA+ABxn e1n = 2(x1 y1 x2 y2 q1 q2 . . .xn1 yn1 q1 qn1 )xn e1n
are contained in I . All in all we get aba I . Hence, I is Jordanclosed.

J3 We show that |e(a)| > |R\e(a)|. All elements a r with a I and
r R that are zero divisors can be found by
(a r)(a r) = 0.
This denes a quadratic equation in r and has two solutions at

the most. Hence, |R\e(a)| = 2 and condition J3 is satised.
Remark 2.18. The group PGL(R, 2) acts ag-transitive on the set

of ags F = {(p, C) P C | p C}, see Lemma 2.1. Thus, we can
use automorphisms of (L, R) to map the subspace dened by the
Jordan-system I elsewhere.
126

Without loss of generality let e21 = 0. We use the matrix
e1 0
0 1

=
PGL(C(p,q,r) , 2) that corresponds to a right-multiplication with e1 together with Lemma 2.1 and apply to I . This results in a new subset
that denes a chain geometry and is given by J = I =span(e1 , . . . , en ).
Therefore, I is isomorphic to a Jordan-system J and the isomorphism
is given by .

Remark 2.19. For n1 V the statement also holds, since we can

write a multiplication with the pseudoscalar as mapping
k
V with 1 < k < n 1 it is not
PGL(C(p,q,r) , 2). For subspaces
always possible to nd a Jordan-system that can be mapped to these
subspaces, but there are cases where it is possible. We proof that
certain subspaces are isomorphic to Jordan-systems, when we need
them.

Remark 2.20. A subspace S 1 V of a Cliord algebra over R is

also isomorphic to a Jordan-closed strong Jordan-system.
2.7.2 Pin and Spin Groups

Now we show that the Pin and the Spin group of a Cliord algebra
correspond to a subspace.
Lemma 2.5. The connected components of the Pin and the Spin
group of a Cliord algebra satisfy the properties J1-J3, but they form
no subspaces of the Cliord algebra C(p,q,r) .
Proof. In order to proof this theorem we have to check the condition
J1-J3 from the Def. 2.20 of a Jordan-system.
J1 Since the Pin and the Spin group are groups with respect to the
Cliord product, the inverse elements are also contained in the
group. Furthermore, every element possesses an inverse element
1
J.
and we have J = J = P in
(p,q,r) . We infer g J g
J2 The Pin respectively Spin group is closed under the Cliord product. Thus, for all a, b P in(p,q,r) aba P in(p,q,r) . Moreover, the
norm is not changed, because the number of factors is odd.
127
J3 We show |e(x)| > |R\e(x)|. The set R\e(x) is the set of all elements
r R with a r N , a P in(p,q,r) . An element g of a Cliord
algebra is a zero divisor if gg = 0. Hence, we look at the equation
(a r)(a r) = 0.
This denes a quadratic equation in r with two real solutions at

the most, and therefore, we have |R\e(x)| = 2. Thus, condition
J3 is satised.
We discussed the Pin group case, but for the Spin group the same
arguments hold.
Lemma 2.5 encourages the assumption that subsets of the projective
+
line P1 (C(p,q,r) ) dened by C

(p,q,r) (g, 1), g Pin(p,q,r) and C(p,q,r) (g, 1),
g Pin(p,q,r) are images of strong Jordan-closed Jordan-systems under

a map PGL(C(p,q,r) , 2). To show that the connected components
of the Pin group dene chain geometries, we show the theorem:
Theorem 2.18. A chain c C dened by three points of the form
C
(p,q,r) (g, 1), g Pin(p,q,r) is completely contained in the embedding of
1
Pin
(p,q,r) in the projective line P (C(p,q,r) ).
Proof. We dene a mapping PGL(R, 2) with R(1, 0) R(a, 1),
R(1, 1) R(b, 1), and R(0, 1) R(c, 1). Since the group PGL(R, 2)
acts ag-transitive on the set F, the standard chain R(s, r), s, r
R is mapped to another chain. Thus, we have to show that every
element of the chain P1 (R) corresponds to an element R(g, 1) with
g Pin(p,q,r) . Therefore, we give the map in matrix form

M=
v 1 c v 1
1 tc t
with t = (c a)1 , v = (t + (b c)1 )1 .
Now, we apply this mapping to the standard chain and nd

R(s, r)M = R(sv 1 c + r(1 tc), sv 1 tr) R(c + r(sv 1 rt)1 , 1).
Thus, we have to show that c + r(sv 1 rt)1 Pin

(p,q,r) for all a, b, c
Pin(p,q,r) . We denote the group that xes the set P1 (Pin

(p,q,r) ) by
(Pin(p,q,r) ). The group (Pin(p,q,r) ) is generated by the matrices
128

0 1
g 0
1 0
,
, and
with g Pin
(p,q,r) . This group acts trans1 0
0 1
0 g
itive on P1 (Pin
(p,q,r) ), and therefore, we can assume that a = 1. To
show that
(c + r(sv 1 rt)1 )(c + r(sv 1 rt)1 ) = 1,
we used the Cliord package from Maple, see [1, 2].

Remark 2.21. The proof gives a new parametrisation of the chain
cabc = c + r(sv 1 rt)1 , r, s R.
With Th. 2.18 each connected component of the Pin group denes a
weak subspace and with Th. 2.13 a subspace of (R, C(p,q,r) ).
Remark 2.22. Th. 2.18 also holds for null vectors. Thus, the null
vectors

N 1 :=
V | vv = 0
are the image of a Jordan-system under a mapping PGL(C(p,q,r) , 2).

Remark 2.23. It is obvious that also the restriction of subgroups of
the Pin and the Spin group to one connected component are images of
Jordan-systems, and therefore, chain geometries on their own. With
[16] every Lie group can be represented as Spin group. Hence, we have
a lot of examples for these chain spaces. Furthermore, Lie groups
correspond to Lie algebras and subgroups of Lie groups to subalgebras
of Lie algberas. The Lie algebra of a Spin group is given by its bivector
algebra, see [17, p. 402].
Now we give an example of a chain given by three elements of a
Pin group. Therefore, we derive an explicit formula for the chain
with Eq. (2.7). We can use this equation since all occurring algebra
elements are proper elements. Let R be the cross ratio of four
given elements. With Eq. (2.8) we get:
"
#"
#1
1
1
cabc () = a(b a) (b c) c (b a) (b c)
.
This gives a parametrisation of the chain cabc dened by the three

points a, b, c Pin+
(p,q,r) and the parameter equals the cross ratio.
129
Example 2.2. As example we look at the unit quaternions, that can

be represented as Pin group elements, see Ex. 1.1. Let
1
1
a = e0 + e12 ,
2
2
b = e1 ,
c = e2
be three algebra elements with aa = bb = cc = 1 of the Pin group of

C(0,2,0) . The chain dened by these three elements parametrised with
a real cross ratio is given by
cabc () =

1
.
2(1)e
+2e
+2(1)e
+
2(1)e
0
1
2
12
22 2+2
We can check, that the chain is entirely contained in the Pin group
cabc ()cabc () = 1,
for all R.
Remark 2.24. Since the group PGL(R, 2) acts ag-transitive on F,

we can use automorphisms of (L, R) to map the projective line over
the Pin or Spin group elsewhere.
Example 2.3. As an example we take the Cliord algebra C(0,2,0)
=
H. The elements
a=
2e0 + 2e12 ,
b = 2e1 ,
c = 2e2
are the elements of Ex. 2.2 multiplied by 2. Note, that the multiplication by 2 can also be expressed as mapping PGL(2, C(0,2,0) ). Now
the elements a, b, c satisfy aa = bb = cc = 4. Again we compute the
chain with Eq. (2.8)
cabc () =

1
.
2(1)e
+2e
+2(1)e
+
2(1)e
0
1
2
12
2 +1
This is exactly the chain of Ex. 2.2 multiplied by 2. Again all points
of cabc () satisfy
cabc ()cabc () = 4,
for all R.
Especially for the case of dual unit quaternions Ud

= Spin(3,0,1) we
are interested in the set of chains. Now we give an example for a
strong Jordan-closed Jordan-system of the quaternions and construct
a mapping that maps this Jordan-system to the Pin group.
130
Example 2.4. A general quaternion or an element of C(0,2,0) has

the form
q = q0 e0 + q1 e1 + q2 e2 + q12 e12 .
The set J = {q0 e0 + q1 e1 + q2 e2 | q0 , q1 , q2 R} forms a strong and

Jordan-closed Jordan-system. We check condition J1-J3.
J1 Let q = q0 e0 + q1 e1 + q2 e2 J , then its inverse is computed as
q1 =
q02
1
(q0 e0 q1 e1 q2 e2 ) J .
+ q12 + q22
J2 Let p = p0 e0 + p1 e1 + p2 e2 and q = q0 e0 + q1 e1 + q2 e2 be two elements

of J , then
qpq = (q02 p0 2q0 q1 p1 2q0 q2 p2 q12 p0 q22 p0 )e0
+ (q02 p1 2q1 q2 p2 q12 p1 + q22 p1 + 2q0 q1 p0 )e1
+ (2q0 q2 p0 + q02 p2 + q12 p2 q22 p2 2q1 q2 p1 )e2 J .
J3 Quaternions, and therefore, the Cliord algebra contains no zero

divisors. Thus, we infer
|e(q)| > |e(R\q)|, q J .
Let a = e0 , b = e1 and c = 12 e2 + 12 e12 . We construct the mapping

PGL(C(0,2,0) , 2) that maps the points R(1, 0) R(a, 1), R(1, 1)
R(b, 1) and R(0, 1) R(c, 1) and apply it to J . The matrix form of
can be derived as

M=

v 1 c v 1
, with t = (c a)1 , v = (t + (b c)1 )1 ,
1 tc t
cf. proof of Th. 2.18. We apply this mapping to a general element

q = q0 e1 + q1 e1 + q2 e2 J :
R(q, 1)M = R(qv 1 c+(1tc), qv 1 t) R(c+(qv 1 t)1 , 1).
The image of J can be described by
2.8 Quadric Chain Geometry
131

1
(2.11)
(1q0 +q1 )e0
1+q1 q0 +q22 +q12 +q02
2
(q1 q0 +q22 +q12 +q02 + 2q2 )e2
+ (q0 +q1 )e1 +
2
2
(q1 q0 +q22 +q12 +q02 2q2 )e12 .
+
2
c+(qv 1 t)1 =
Every element of J has norm equal to one and corresponds to an

unit quaternion. This can be checked easily with Eq. 2.11:

c+(qv 1 t)1

c+(qv 1 t)1 = 1, q J .

A. Blunck showed in [10] that chain spaces on quadrics can be represented with the help of Jordan-systems J contained in a Cliord
algebra. With the prerequisites from section 2.7.1 we are able to construct chain geometries corresponding to quadrics in a surprisingly
simple way. For this purpose, we consider a quadric Q Pn (R) given
by the symmetric matrix Q R(n+1)(n+1) corresponding to the quadratic form. To describe the quadric chain geometry algebraically we
construct the Cliord algebra C(p,q,r) corresponding to Q. The points

of the quadric can be represented as null vectors v 1 V with vv = 0.
So far we have collected all the results and settings that we need in
order to construct the quadric chain geometry. In the following we
give some examples to show the elegance of this apparatus. In section
2.12 we use this method to construct and describe biarcs on arbitrary
quadrics.
Remark 2.25. Since quadrics can be constructed over any commutative ring, we can generalize this method easily to quadric chain
geometries over commutative rings, see [9].

As a rst example of the algebraic description of a quadric chain
geometry we consider Kleins quadric. We compute a regular conic
dened by three points with methods of projective geometry. After
132
that we compute the chain dened by these three points described

as null vectors in the related Cliord algebra. With corresponding
parametrisations we see that both methods result in exactly the same
conic.
Example 2.5. Let A = aR = (1, 0, 0, 0, 1, 1)T R, B = bR = (0, 1, 0, 0, 0, 1)T R,
and C = cR = (0, 0, 1, 1, 0, 0)T R be Pl
ucker coordinates of three given
lines, see Def. 1.4. Pl
ucker coordinates satisfy the quadratic equation,
cf. Eq. 1.12, and describe the points contained in a quadric M24 . The
corresponding lines are visualized in Fig. 2.1. The quadratic form Q
corresponding to M24 is given by

O I
,
I O

Q=
where I is the 3 3-identity matrix and O the 3 3 zero matrix. We

denote the corresponding bilinear form by . A general parametrisation
of the conic section of M24 with the two-space dened by A, B, C is
given by
c(t) = t (A, B)C +(1t) (A, C)B +t(t1) (B, C)A.
(2.12)
Note, that the points A, B , and C correspond to the parameter values

t = , 0, 1. For the given points A, B, C the parametrisation (2.12)
becomes
c(t) = (t(t1), 2(1t), t, t, t(t1), t2 3t + 2)T .
(2.13)
Now we consider the Cliord algebra C(3,3,0) dened by R6 as vector

space model for P5 (R) and the quadratic form Q, see also section 1.6
or [40]. The null vectors corresponding to the points A, B , and C are
given by
a = e1 + e 5 + e 6 ,
b = e2 + e6 ,
c = e3 + e 5 .
If we use Eq. (2.8) the corresponding chain parametrised with a real

cross ratio R is given by
1
2(1)e1 +e2 +(1)e3

+ (2 3+2)e5 +(2 )e6 .
cabc () =
2 2+2
133
Since we are computing with homogeneous coordinates we can normalize to

cabc () = 2(1)e1 +e2 +(1)e3 +(2 3+2)e5 +(2)e6 .
(2.14)
Equating the coecients of Eq. (2.13) and Eq. (2.14) shows that
both
representations describe the same set of points if we identify 1 V
with P5 (R). Furthermore, the chain interpolates the null vectors a, b, c
for the cross ratio values = , 0, 1. A visualization of the resulting
regulus together with its striction curve (blue) is shown in Fig. 2.1.
b
a
Figure 2.1: Regulus corresponding to Eq. (2.13) and Eq. (2.14)
2.8.2 Lies Quadric

Lies quadric was introduced in section 1.7. Now we use the homogeneous Cliord algebra model corresponding to Lies quadric to describe
its quadric chain geometry algebraically. The Cliord algebra model
is given by the vector space V = R6 as an algebraic model of P5 (R)
and the symmetric matrix of the quadratic form
134
b
c
a
Figure 2.2: Dupin cyclide corresponding to the envelope of a conic on

Lies quadric
0
0
0
,
0
0
1

The subspace 1 V corresponds to the
1
0
0
Q=
0
0
0
0
1
0
0
0
0
0
0
1
0
0
0
0
0
0
1
0
0
0
0
0
0
1
0
compare to section 1.7.2.

projective space P5 (R). Furthermore, the points contained on Lies
quadric are null vectors in the homogeneous Cliord algebra model.
Example 2.6. Let A = aR = (0, 1, 0, 0, 0, 1)T R, C = cR = (8,7, 4, 0, 0, 1)T R,
and B = bR = (11, 10, 0, 5, 0, 2)T R be three points on L41 . According to
section 1.7.1 we interpret these points as Euclidean entities. The point
A corresponds to a sphere with center c = (0, 0, 0)T and signed radius
r = 1. Point B represents a sphere with center c = (0, 5, 0)T and signed
radius r = 2. Further, C represents a sphere with c = (4, 0, 0)T and r = 1.
A visualization of the three spheres corresponding to A, B and C is
given in Fig. 2.2 (left). We rewrite the points as null vectors
a = e2 +e6 ,
b = 11e1 10e2 +5e4 +2e6 ,
c = 8e1 7e2 +4e3 +e6 .
The chain dened by a, b, and c is computed with Eq. (2.8):

cabc () =
52

1
2( + 11)e1 + (52 6 20)e2 + 12e3
4 + 2
135

+ 10(1 )e4 + (52 6 + 4)e6 .
Note, that this parametrisation satises x1 + x2 = 1, where x1 and x2

are the coecients of e1 respectively e2 . Thus, we can read o the
curve of centers of the oriented spheres
c() =
52
1
T
(12, 10(1 ), 0)
4 + 2
and the radius function as

r() =
52 6 + 4
.
52 4 + 2
It is well-known that conics on Lies quadric correspond to a family

of spheres whose envelope is a Dupin cyclide. The envelope of this
one-parameter family of spheres is given by the implicit equation
0 =3(x4 +y 4 +z 4 )30y 3 24x3 +(36+6y 2 +6z 2 30y)x2
+(120y24y 2 24z 2 +48)x+(6z 2 +58)y 2 +(1030z 2 )y113+108z 2
and visualized in Fig. 2.2 (right).
2.8.3 Studys Quadric

The homogeneous model corresponding to Studys quadric was introduced in section 1.8. It is constructed with the eight-dimensional
vector space R8 as model for P7 (R) and the quadratic form
Q=
O
I

I
,
O
where O is the 44 zero matrix and I is the 44 identity matrix. Similar to the examples for Kleins and Lies quadric we give an example
of a conic on Studys quadric as chain in the homogeneous Cliord
algebra model.
Example 2.7. Let A = (1, 0, 0, 0, 0, 1, 1, 1)T R, B = (0, 1, 0, 0, 0, 0, 1, 1)T R
and C = (0, 0, 1, 0, 1, 0, 0, 0)T R be three points on S26 P7 (R). In analogy
to Eq. (2.12) the conic that interpolates A, B , and C for the parameter
values t = , 0, and 1 can be parametrised by
c(t) = t(A, B)C + (1 t)(A, C)B + t(t 1)(B, C)A,
where denotes the bilinear form belonging to Studys quadric. Especially for the chosen points we derive:
136

0
0
1
t(t 1)
0
1
0 2(1 t)
1
0
0
t
0
0
0
0
.
c(t) = t + 2(1 t) + t(t 1) =
1
0
0
t

0
0
1 t(t 1)

0
1
1 t2 3t + 2
0
t2 3t + 2
Now we calculate a parametrisation ofthe chain that is dened by

1
A, B , and C if we identify P7 (R) with
V . The corresponding null
vectors have the form
a = e1 +e6 +e7 +e8 ,
b = e2 +e7 +e8 ,
c = e3 +e5 .
The chain dened by these three null vectors is computed with Eq.
(2.8)
cabc () =
1
2 2+2
(1)e1 2(1)e2 +e3

+e5 +(1)e6 +(2 3+2)e7 +(2 3+2)e8 .
Normalizing this equation results in

cabc () = (1)e1 2(1)e2 +e3 +e5
+(1)e6 +(2 3+2)e7 +(2 3+2)e8 .
Equating the coecients shows that both representations are exactly

the same. The chain cabc is contained in the two-space spanned by the
points A, B, C that correspond to the null vectors a, b, c.
2.8.4 Studys Sphere

The dual sphere as point model for oriented lines was introduced in
section 1.2.3. Now we aim at a Cliord
algebra that embeds Studys

sphere in the grade-1 subspace 1 V in a natural way. Therefore,
137
we construct the Cliord algebra model over the module M = D3

together with the quadratic form
1
Q = 0
0
0
1
0
0
0 .
1
For a general grade-1 element v = x1 e1 +x2 e2 +x3 e3 with x1 = a1 +b1 ,

x2 = a2 + b2 and x3 = a3 + b3 we get
vv = x21 + x22 + x23 = a21 + a22 + a23 + 2(a1 b1 + a2 b2 + 2a3 b3 ).
Since we describe the dual unit sphere, the real part has to be equal to
1 and dual part has to vanish. We already showed that the connected
components of the Pin group dene subspaces of (C(C(p,q,r) ), C(p,q,r) ),
see section 2.7.2. Note that we use an ane Cliord algebra model
here. It is also possible to construct a homogeneous. We perform this
construction for the case of dual unit quaternions.
Figure 2.3: Line congruence with constant dual angle to an axis a (red)
Example 2.8. Now we give an example for a chain that corresponds to

a conic section on SD2 . Let A = (1, 0, 0, 0, 1, 1)T R, B = (0, 1, 0, 0, 0, 1)T R,
138
and C = (0, 0, 1, 1, 0, 0)T R be three lines given in Pl

ucker coordinates.
These Pl
ucker coordinates are the same that were used in Ex. 2.5.
The corresponding dual vectors

1
a = ,

0
b = 1 ,

c = 0
1
are already normed. Written as algebra elements the three oriented

lines get the form:
a = e1 + e2 + e3 ,
b = e2 + e3 ,
c = e1 + e3 .
Remember that the quadric is constructed over the ring of dual numbers
D. Therefore, the cross ratio may be a dual number = + . The
chain, computed with Eq. (2.8), has the form
1 + (2 1)e1
( 2 1 + 2 2 + + )
+ (3 1)e2
+
( 2 1 + 2 2 + + )
2 + + (2 2 + 1)e3
+
.
2 1 + 2 2 + + )
cabc () = cabc (, ) =
We split the dual numbers into real and dual parts and get

1
(3 22 + + 2)
+
e1
cabc (, ) =
(2.15)
2 + 1
(2 + 1)2

(1 )(22 2 + + 1 + )
+

e2
+
2 + 1
(2 + 1)2

( 1)
3 + 1 + 2
+

e3 .
+
2 + 1
(2 + 1)2
Study [61] and Klein [43] already knew that the intersection of
the dual sphere SD2 with two-dimensional ane submodules results in
congruences of lines, i.e., two-parameter families of oriented lines that
enclose a constant dual angle with an unique oriented line. Therefore,
Eq. (2.15) is a parametrisation for this congruence of lines, see Fig.
2.3 for a schematic view. This congruence can be interpreted as the
set of all tangents of a cylinder of revolution that enclose a constant
angle with its axis.
2.9 Quadric Chain Spaces for certain Spin Groups
139

We use the work of Blunck [9] which shows that algebras of the
form L + sL with s2 {1, 1, 0} and L a local ring posses a quadric
chain space over one of the rings A, C, or D. Furthermore, we restrict
ourselves to Spin group elements.
2.9.1 A Quadric Model for dual unit Quaternions

In this section we focus on the cross ratio of dual unit quaternions.
We know that the dual unit quaternions are isomorphic to Spin(3,0,1)
and that this Spin group is contained in the even part of the algebra
C+
(3,0,1) = C(2,0,1) . Furthermore, the algebra is of the form L + sL,
2
with s = 0, since C(2,0,1)
= C(2,0,0) D. For our investigations it is
important that the norm of a general element A C+
(3,0,1) is contained
+
in the center C(C(3,0,1) ) = D. Let

g = a0 + a3 e12 + a2 e13 + c1 e14 + a1 e23 + c2 e24 + c3 e34 + c0 e1234
be a general element from C+

(3,0,1) . We calculate its norm by
gg = (a20 + a21 + a22 + a23 ) + 2(c0 a0 a1 c1 + c2 a2 c3 a3 )e1234 .
(2.16)
Setting the pseudoscalar part of Eq. (2.16) to zero yields c0 a0 a1 c1 +

c2 a2 c3 a3 = 0. This describes a quadric that is the image of Studys
quadric under a collineation. This collineation denes a bijection
between Spin(3,0,1) and Ud , see chapter 3. A spin group element g
C+
(3,0,1) satises
gg = (a20 + a21 + a22 + a23 ) + 2(c0 a0 a1 c1 + c2 a2 c3 a3 )e1234 = 1.
Hence, we have the two conditions a20 +a21 +a22 +a23 = 1 and c0 a0 a1 c1 +
c2 a2c3 a3 = 0. In analogy to Ex. 2.8 we can construct an ane Cliord
algebra model that describes the group Spin(3,0,1) as the intersection
of the grade-1 subspace with its Pin group.
140
The ane Cliord Algebra Model The corresponding ane model

is constructed with the module M = D4 and
1
0
Q=
0
0
0
1
0
0
0
0
1
0
0
0
.
0
1
Hence, the square of grade-1 elements is the canonical scalar product

of dual vectors. Let v = (a0 +c0 )e1 +(a1 +c1 )e2 +(a2 +c2 )e3 +(a3 +c3 )e4
be a general element. we compute
vv = a20 + a21 + a22 + a23 + 2(a0 c0 + a1 c1 + a2 c2 + a3 c3 ).
This is exactly the norm of a dual quaternion q = a0 + a1 i + a2 j + a3 k +

(c0 + c1 i + c2 j + c3 k). The grade-1 subspace is isomorphic to a Jordansystem, the Pin group is a subspace, and therefore, the restriction
of the Pin-group to the grade-1 subspace is also a subspace. Thus,
chains of dual quaternions
with norm equal to 1 can be represented in

Pin(C(0,4,0) (D)) 1 (C(0,4,0) (D)). The chains are conic sections of the
dual sphere SD3 D4 with dual planes. A geometric interpretation of
the corresponding displacements is given in section 2.10.
Remark 2.26. The Study sphere SD2 is contained in this model. If
we set a0 = c0 = 0 the corresponding displacements are line-symmetric.
They still satisfy the quadric equation and can be interpreted as oriented lines. Hence, the point model for oriented lines SD2 is naturally
embedded in SD3 .
The homogeneous Cliord Algebra Model It is also possible to use
a projective quadric and the corresponding quadratic form to dene
a Cliord algebra model. In this setting the interesting elements are
embedded as null vectors. To construct the corresponding Cliord
algebra model we use M = D5 as a model for the projective space
P4 (D) together with the quadratic form corresponding to a sphere
1
0
Q=
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
.
0
1
141
The corresponding homogeneous Cliord algebra model is given by

C(1,4,0) (D).
Remark 2.27. For n = p + q + r = 4 we have some more cases that
allow a description as quadric model over a commutative ring.
2.9.2 Other possible Quadric Models
C(4,0,0) :
We start with the Cliord algebra C(4,0,0) . Its Spin

group is a double cover of the group SO(4). A general element of the
even part has the form
g = a0 + a3 e12 + a2 e13 + c1 e14 + a1 e23 + c2 e24 + c3 e34 + c0 e1234 .
The product with its conjugate element is calculated as

gg = (a20 +a21 +a22 +a23 +c20 +c21 +c22 +c23 )e0 +2(c0 a0 a1 c1 +c2 a2 c3 a3 )e1234 .
If this is an element of the Spin group, the real part has to equal
1 while the pseudoscalar part has to vanish. The equation in the
pseudoscalar part can, again, be understood as the equation of Studys
quadric. Therefore, the same set of points in projective space can
serve as a point model for dierent geometric entities, see chapter 3.
Ane Cliord Algebra Model Since the center of C(4,0,0) is isomorphic to A we construct the ane model by the use of M = A4
and the quadratic form
1
0
Q=
0
0
0
1
0
0
0
0
1
0
0
0
.
0
1
The chain geometry of Spin(4,0,0) can then be found in the grade-1

space. A general vector v 1 V has the form v = (a0 + jc0 )e1 + (a1
jc1 )e2 + (a2 + jc2 )e3 + (a3 jc3 )e4 with j 2 = 1. Its square is
vv = a20 + a21 + a22 + a23 + c20 + c21 + c22 + c23 + 2j(c0 a0 a1 c1 + c2 a2 c3 a3 ).
This is exactly the norm of a general element from C+

(4,0,0) . The chains
are conic sections with split-complex planes, i.e., two-dimensional
ane submodules in the module A4 .
142
Homogeneous Cliord Algebra Model To construct the homogeneous Cliord algebra model we use M = A5 as model for the projective
space P4 (A) and the quadratic form
1
0
Q=
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
.
0
1
Like in the case of dual quaternions the unit double quaternions are
in this model represented as null vectors.
C(3,1,0) :
For a geometric interpretation of this algebra and its Spin

group, see chapter 3. We construct the quadric model for the Spin
group in the same way. Thus, let
g = a0 + a3 e12 + a2 e13 + c1 e14 + a1 e23 + c2 e24 + c3 e34 + c0 e1234
be an arbitrary element in C(3,1,0) . The product with its conjugate

element results in
gg = (a20 +a21 +a22 +a23 c20 c21 c22 c23 )e0 +2(a0 c0 a1 c1 +a2 c2 a3 c3 )e1234 .
Ane Cliord Algebra Model In order to construct the ane model

we need V = C4 as vector space and
1
0
Q=
0
0
0
1
0
0
0
0
1
0
0
0
0
1
as quadratic form. A general vector v
1
V has the form v =

(a0 + ic0 )e1 + (a1 ic1 )e2 + (a2 + ic2 )e3 + (a3 ic3 )e4 . Its square reads
vv = a20 + a21 + a22 + a23 c20 c21 c22 c23 + 2(c0 a0 a1 c1 + c2 a2 c3 a3 )i.
This is exactly the norm of a general element of C+

(3,1,0) .
143
Homogeneous Cliord Algebra Model For the homogeneous quadric model of Spin(3,1,0) we use V = C5 as vector space model for P4 (C).
The corresponding quadratic form is given by
1
0
Q=
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
.
0
1
Remark 2.28. The Cliord algebra C(3,1,0) results in the same model.
We have to use the quadratic form Q = diag(1, 1, 1, 1) for C(3,1,0) .
If we use the standard form Q = diag(1, 1, 1, 1), the equality is not
obvious.
C(2,2,0) :
A general element of the even part of C(2,2,0) is given by
Its norm is determined by

gg = (a20 a21 a22 +a23 +c20 c21 c22 +c23 )e0 +2(a0 c0 a1 c1 +a2 c2 a3 c3 )e1234 .
We proceed analogue to the cases above and determine the ane and
the homogeneous Cliord algebra models for this Spin group.
Ane Cliord Algebra Model As module we use M = A4 . The
corresponding quadratic form is computed as
1
0
Q=
0
0
A general vector v
0 0 0
1 0 0
.
0 1 0
0 0 1
1
V is given by v = (a0 + c0 )e1 + (a1 + c1 )e2 +

(a3 + c3 )e3 + (a2 + c2 )e4 . Its square equals
vv = a20 a21 a22 + a23 + c20 c21 c22 + c23 + 2(+c0 a0 a1 c1 c2 a2 + c3 a3 ).
Again, we get exactly the norm of a general element from C+

(2,2,0) .
144
Homogeneous Cliord Algebra Model To represent the points of

the group Spin(2,2,0) as null vectors we construct a homogeneous
Cliord algebra model with M = A5 and
1
0
Q=
0
0
0
C(2,1,1) :
0 0
1 0
0 1
0 0
0 0
0 0
0 0
0 0
.
1 0
0 1
A general element from the even part is again given by
Its norm is calculated as

gg = (a20 a21 a22 + a23 )e0 + 2(c0 a0 a1 c1 + c2 a2 c3 a3 )e1234 .
Ane Cliord Algebra Model The ane Cliord algebra is constructed with M = D4 together with the quadratic form
1
0
Q=
0
0
0
0 0
1 0 0
.
0 1 0
0
0 1
A vector has the form v = (a0 + c0 )e1 + (a1 + c1 )e2 + (a2 c2 )e3 +
(a3 c3 )e4 and its square reads
vv = a20 a21 a22 + a23 + 2(c0 a0 a1 c1 + c2 a2 c3 a3 ).
Again the chain geometry (C(C(2,1,0) ), Spin(2,1,0) ) is now represented

as the set of conic sections respectively as chains in 1 V Pin(2,1,0) .
Homogeneous Cliord Algebra Model To represent this quadric
model in a homogeneous way, we embed this quadric in the projective space P4 (D). Thus, the homogeneous Cliord algebra model is
constructed with M = D5 as model for P4 (D) and the quadratic form
1 0 0
0 1 0
Q=
0 0 1
0 0 0
0 0 0
0 0
0 0
0 0
.
1 0
0 1
2.10 Cross Ratio of dual unit Quaternions
145
Remark 2.29. The Cliord algebra C(1,2,1) results in a quadric with

the same signature.
Remark 2.30. For Cliord algebras with more than one generator
squaring to zero this construction does not work.

In this section we focus on the question for the cross ratio of dual
unit quaternions. To give an answer to this question we use a Lemma
that is taken from [30].
Lemma 2.6. Assume we are given three points Pi S26 , i = 0, 1, 2.
Their Study coordinates are denoted by (ai , ci ), where ai =
(ai0 , ai1 , ai2 , ai3 )T , ci = (ci0 , ci1 , ci2 , ci3 )T , i = 0, 1, 2 are real vectors. Additionally, we assume
a
rk 0
c0
a1
c1
a2
c2

= 3.
If rk(a0 , a1 , a2 ) = 3, then an unique collineation , see Eq. (1.9), is

determined such that any Pi satises the equations ai0 = ci0 = 0.
A proof can be found in [30].
Remark 2.31. With the help of Lemma 2.6 we can say: There exists
exactly one coordinate transformation that maps three given displacements to line-symmetric displacements of the reference coordinate
system.
We do not need additional cases of Lemma 2.6, since a chain is dened
by three mutually distant points, and therefore, rk(a0 , a1 , a2 ) = 3 is
always true for any three points of a chain. With the help of Lemma
2.6 we can restrict ourselves to line-symmetric motions. Now we show
that the bivector part of C(3,0,1) is isomorphic to a Jordan-closed
strong Jordan-system.

Theorem 2.19. The bivector subspace 2 V of the Cliord algebra

C(3,0,1) is isomorphic to a Jordan-closed strong Jordan-system, i.e.,
2
(C(C(3,0,1) ),
V ) is a subspace of (C(C(3,0,1) ), C(3,0,1) ).
146
Proof. To proof this we show that the subspace I spanned by

{e0 , e13 , e14 , e23 , e24 , e1234
} is a Jordan-closed, strong Jordan-system.
With the mapping
e12
0
0
1
PGL(C(3,0,1) , 2) we can map the sub-
space to bivector subspace bijectively. Clearly 1 I . We show J1-J3

from Def. 2.20:
J1 A general element from I C
(3,0,1) can be written as
A = x1 e0 x2 e23 + x3 e13 x4 e24 + x5 e14 x6 e1234 .
Its inverse element is computed as

A1 =
+
x1
x2
x3
e0 + 2
e23 2
e13
2
2
2
2
2
x1 + x2 + x3
x1 + x2 + x3
x1 + x22 + x23
x21 x4 2x3 x1 x6 + x4 x22 x23 x4 + 2x3 x5 x2
e24
(x21 + x22 + x23 )2
2x2 x6 x1 + 2x2 x3 x4 x22 x5 + x5 x21 + x23 x5
e14
(x21 + x22 + x22 )2
2x3 x1 x4 + x6 x21 x23 x6 2x2 x5 x1 x22 x6 )e1234
I.
(x21 + x22 + x22 )2
It follows that x21 + x22 + x23 = 0 for A A := I C

(3,0,1) .
J2 Let B I be another general element with
B = y1 e0 y2 e23 y3 e24 + y4 e13 + y5 e14 y6 e1234 .
We compute
ABA = (x21 y1 2x3 x1 y4 2x1 x2 y2 x22 y1 x23 y1 )e0
+ (x22 y4 + x21 y4 + 2x3 x1 y1 x23 y4 2x2 x3 y2 )e13
(2x2 x3 y4 + 2x1 x2 y1 x22 y2 + x23 y2 + x21 y2 )e23
+ (2x1 x2 y6 + x23 y5 + x21 y5 2x5 x2 y2 2x4 x3 y2 2x5 x3 y4
2x2 x6 y1 + 2x1 x5 y1 2x1 x6 y2 + 2x2 x3 y3 + 2x4 x2 y4 x22 y5 )e14
(2x1 x4 y1 + 2x3 x1 y6 + 2x3 x6 y1 + x22 y3 + 2x2 x3 y5 2x4 x3 y4
x23 y3 + x21 y3 2x5 x2 y4 + 2x5 x3 y2 + 2x1 x6 y4 2x4 x2 y2 )e24
(2x1 x2 y5 + x21 y6 x23 y6 2x2 x6 y2 2x4 x3 y1 x22 y6 + 2x5 x2 y1
2x1 x4 y4 + 2x1 x5 y2 2x3 x6 y4 2x3 x1 y3 + 2x1 x6 y1 )e1234 I.
147
J3 We have to show that |e(b)| > |C(C(3,0,1) )\e(b)|. Therefore, we

search for all elements d = d1 e0 + d2 e1234 C(C(3,0,1)
= D with
A d N (C(3,0,1) ). If we compute the inverse of A d, we get the
condition x21 + x22 + x23 + 2x1 d1 + d21 = 0. Since this equation has at
most two real solutions at the most, we infer |C(C(3,0,1) )\e(b)| = 2.
Remark 2.32. With the identity C(2,0,1) = C(2,0,0) D, see Selig [59,
sect. 9.3], we see that the dual unit quaternions dene a sphere SD3 in
four-dimensional dual space.

A chain dened by three points in 2 V is completely contained in 2 V .

Therefore, we can interpret the corresponding set of line-symmetric
displacements as set of lines. Furthermore, in the quadric model,
introduced in section 2.9.1, this is a subspace that corresponds to
Studys sphere. Planar sections of Studys sphere with two-dimensional
ane submodules result in sets of lines with constant dual angle, i.e.,
a constant angle and a constant distance with respect to a xed line.
The next step is to show that this is also true in the general case.
Therefore, we proof the following:
Theorem 2.20. A chain cabc dened by three arbitrary mutually
distant elements a, b, c Spin(3,0,1) is characterized by an axis l and a
dual angle . All elements d contained in cabc satisfy d , l
= ,
where d , l D4 are the dual unit vectors corresponding to d and l.
Remark 2.33. The corresponding dual vectors can be obtained directly with the identity C(2,0,1) = C(2,0,0) D or as the dual number
coecients in the quaternion when using dual unit quaternions.
Proof. To proof this theorem we use Lemma 2.6. Three mutually
distant Spin group elements correspond to Study coordinates of points
that admit an unique collineation , see Eq. (1.9). This collineation
corresponds to a dual quaternion left- or right-multiplication which
does not change the
ratio. We can express this mapping
cross
g 0
with g Spin(3,0,1) . The collineation applied
0 1
to a point in P = (c, g)T P7 (R), c, g R4 gives

Ac
c
A 0
.
=
Bc + Ag
g
B A
PGL(C(3,0,1) , 2) as
148
If we write the elements as dual vectors in D4 , the action can be

written as
(A + B)(c + g) = Ac + (Ag + Bc).
The entries of the dual matrix O = A + B satisfy

3

3

i=0
a2i = 1 and
ai bi = 0 because the matrix O corresponds to the action of the Spin
i=0
group element g. Furthermore, the matrix O satises OOT = OT O = I

and det O
= 1. Thus, O SO(4, D) leaves the dual angle invariant.
Chains in 2 V correspond to intersections of two-dimensional ane
submodules with SD2 that correspond to the set of lines with constant
dual angle to an axis. If we now apply the inverse mapping O1
respectively 1 we get the original chain passing through a, b, and c.
Since the inverse mapping also preserves the dual angle, we have the
same property and the theorem is proved.
Geometrically, we can interpret the set of displacements corresponding
to a chain as the set of line-symmetric displacements with respect
to a congruence of lines in a reference coordinate system. This line
congruence is given by the set of all lines that enclose a constant dual
angle with a xed line.
Remark 2.34. With remark 2.23 subgroups of Spin groups dene
subspaces of the chain geometry (C(C(p,q,r) ), C(p,q,r) ). The twoparameter group of helical displacements with xed screw axis is both
a subgroup and a D-chain. Therefore, the cross ratio of four elements
of a helical displacements group is a dual number.
For the group Spin(3,0,1) there is a simple description for the set of
all displacements that correspond to elements of a chain. Let us
again start, without loss of generality,
with a chain of line-symmetric

displacements dened by a, b, c 2 V . That can be achieved with
Lemma 2.6. Every axis of each line-symmetric displacement contained
in the chain cabc is the image of one of the three initial lines under a
two-parameter screw group, i.e., all screws with respect to the axes l
of the cylinder. In fact this group is a one parameter group with a
dual parameter. This screw can be represented as the exponential of
a se(3), where a is the normed axis of the helical motion, = + d
149
the dual angle and se(3) denotes the Lie-algebra of SE(3). Then the
chain can be expressed with the sandwich operator
cabc (, d) = ea aea .
For a general chain, i.e., a set of line-symmetric displacements with

respect to another reference system, we need the transformation
g Spin(3,0,1) that corresponds to the collineation of Lemma 2.6.
cabc (, d) = gea aea .
Real Cross Ratio Now we ask for a real cross ratio of four dual
unit quaternions. First we restrict ourselves again to the case of linesymmetric displacements. For three given line-symmetric displacements a, b, c respectively lines the corresponding R-chain is a closed
ruled surface, since cr(a, b, c, a) = 0, cr(a, b, c, b) = 1 and cr(a, b, c, c) = .
Every normalized dual vector d of the ruled surface satises l , d
=
, where D is a constant dual angle and l is the axis of the
cylinder that denes the congruence, see Th. 2.20. This means that
the ruled surface is a closed ruled surface of constant slope. In general
its intersection with the ideal plane P is a conic with multiplicity two
that has two complex conjugate contact points with multiplicity four
with the absolute circle. This can be derived with the help of Pl
ucker
coordinates, see section 2.10.2. Hence, the order of the striction curve
is reduced by six, see [51, p. 152]. The parametrisation of the
congruence of lines in terms of Pl
ucker coordinates has degree four.
Therefore, the degree of the striction curve is two,
i.e., it is a planar

curve. In fact, the R-chain dened by a, b, c 2 V corresponds to
the set of line-symmetric displacements with respect to a closed ruled
surface of constant slope whose striction curve is an ellipse, see Fig.
2.4 for example. Note that the striction curve is dened by the planar
intersection of the cylinder that is the envelope of the congruence of
lines. The plane is dened by the contact points of the three given
lines with the cylinder.
150
2.10.1 Subspaces on Studys Quadric and Sub Chain

Geometries
From remark 2.23 we know that subgroups dene subchain geometries.
Naturally, there are subchain geometries contained in Ud that can
be derived from properties already known from Studys quadric. We
examine R-chains, because the set of points of P7 (R) corresponding to
elements of a D-chain spans a four-space in P7 (R) in general. Therefore,
D-chains are not adequate for a comparison with the Study model.
We start with generators G , i.e., three-spaces completely contained in
Studys quadric. A generator corresponds to the coset of a subgroup if,
and only if, the dim(G V 3 ) = 1, see section 1.2.2. If we use the group
Spin(3,0,1)
= Ud , we cannot determine chains contained in cosets of
SE(2). With elements of the Spin group it is just possible to compute
chains that are completely contained in cosets of SO(3), because every
element of a chain is distant to every other in cosets of SO(3). If we
want to investigate other cosets than cosets of SO(3) we have to use
the more general Cliord group, see Eq. (1.31). In this context we can
also parametrise chains that corresponds to cosets of SE(2) and even
chains contained in subgroups that correspond to lines on Studys
quadric. Of course, we lose the property that the whole chain is
contained in the Pin group, but we can still normalize after computing
the chain, if it is necessary to have Pin or Spin group elements. We
demonstrate this method in a simple example.
Example 2.9. As example we take three Spin group elements that
correspond to the same translation group. Let a = e0 + e14 , b = e0 +
2e14 , c = e0 + 4e14 be elements of Spin(3,0,1) . These elements are not
distant to each other, and therefore, there is no chain in the Spin group
through these elements. Now we change our setting and take multiples
of these element. Thus, let a = e0 + e14 , b = 2e0 + 4e14 , c = 4e0 + 16e14
be elements of (3,0,1) . These elements are mutually distant, and
therefore, we can compute the R-chain cab
c ():
cab
c () =
2(1 + 2)
4(2 + 1)2
e0 +
e14 ,
2+
( + 2)2
151
and c () = c. To get elements of the

with cab
a, cab
c (0) =
c (1) = b
ab
c
Spin group we can normalize and nd
cab
c () = 1 +
2(1 + 2)
e14 .
2+
2.10.2 Application to line-symmetric Displacements

In this section we give an example
of a chain dened by three linesymmetric displacements a, b, c 2 V . Let A = (1, 0, 0, 0, 1, 1)T R,
B = (0, 1, 0, 0, 0, 1)T R, C = (0, 0, 1, 1, 0, 0)T R be the axes of three linesymmetric displacements, cf. Ex. 2.8. The corresponding displacements are given by
a = e23 + e24 + e34 ,
b = e13 + e34 ,
c = e12 + e14 .
Note, that the minus in front of e13 is due to historic reasons, see Ex.
1.6. We compute the chain cabc (), = e0 + e1234 C(C(3,0,1) )
= D,
with Eq. (2.8)
(1 )e23
e13
( 1)e12
+
1 + 2
1 + 2
1 + 2
3
2
( 2 + + 2)e14
(1 2 + 3 + )e34
+
+
(1 + 2 )2
(1 + 2 )2
3
2
2
(2 4 + 3 + 1)e24
.
(1 + 2 )2
cabc () =
In Pl
ucker coordinates the line congruence can be written as
(1 + 2 )(1 )
2
(1 + )
(1 + )( 1)
.
l(, ) =
3
2
( 2 + + 2)
23 + 42 + 2 3 + 1
1 2 + 3 +
Intersection with the absolute Circle To determine the intersection

of the congruence of lines with the absolute circle, we intersect its
ideal curve given by

T
c() = (1 + 2 )(1 ), (1 + 2 ), (1 + 2 )( 1)
152
with the absolute circle determined by x21 + x22 + x23 = 0. This leads to
the equation
1 4 + 102 163 + 194 165 + 106 47 + 8 = 0
with solutions
1 = 2 = 3 = 4 =
i
i
1
1
3, 5 = 6 = 7 = 8 = +
3.
2 2
2 2
Hence, this congruence of lines has two points in common with the
absolute circle. At these points it has contact of order three.
Axis of the Congruence and constant dual Angle We are interested
in the axis of this congruence, i.e., in the dual vector a D3 with
a , cabc ()
= k1 + k2 , for all D
where cabc () is the chain dened a, b, c, and = + is the cross

ratio, and k1 , k2 R are constants. Therefore, we look for the axis
A = (c, c) = (c1 , c2 , c3 , c4 , c5 , c6 )T of the linear complex that contains
l(, ). This is done with the approach
(A, l(, )) = c1 l4 (, ) + c2 l5 (, ) + c3 l6 (, )
!
+ c4 l1 (, ) + c5 l2 (, ) + c6 l3 (, ) = 0.
(2.17)
For a more detailed description of this implicitization method, see

[63]. After expanding Eq. (2.17) and collecting the coecients of i j ,
i = 0, . . . , 4, j = 0, 1 we get
(A, l(, )) = (c6 + c1 )4 + (2c1 2c2 + c3 c4 + c5 2c6 )3
+ (c1 + c2 ) + (+c1 + 4c2 + 2c4 c5 + 2c6 )2
+ (2c1 2c3 ) + (3c2 c3 2c4 + c5 c6 )
!
+ (c2 + c3 ) + (c2 + c3 + c4) = 0.
In order to nd a solution, every coecient written in brackets has to

vanish. We can reformulate this as matrix vector product:
153

1 0 0 0 0 1
0

2 2 1 1 1 2
0
c
1

1 1 0 0 0 0 c2 0

1 4 0 2 1 2 c3 0

MA=
2 0 2 0 0 0 c4 = 0 .

0 3 1 2 1 1 c5 0

0 1 1 0 0 0
0
c6
0 1 1 1 0 0
0
The kernel of M is determined by

ker M = (, , , 2, , )T , R.
We see that the kernel of M is one-dimensional, geometrically this

means that l(, ) is contained in a hyperplane, i.e., a four-space.
In a next step, we could assume that there is a quadratic complex
of lines that contains the lines corresponding to elements of l(, )
and evaluate a quadratic approach to look for quadrics in P5 (R) that
contain l(, ). We know that the line congruence l(, ) is contained
in a linear complex, cf. Def. 1.6. If the linear complex is singular
then the Pl
ucker condition (1.12) is satised. If (A, A) = 0 the point
A corresponds to a line and the complex is singular, i.e., it is the set
of all lines that meet the line represented by (c, c). If (A, A) = 0, the
linear complex is regular and it consists of all path normals of a helical
motion. For a regular linear complex we can calculate the pitch p and
the axis (a, a) with Eq. (1.14). For our example this results in
p=
c c
4
= ,
c2
3

2 1 1
(a, a
) = (c, c pc) = (1, 1, 1)T , ( , , )T
3 3 3
=((3,
$
3, 3)T , (2, 1, 1)T ).
Remark 2.35. In Kleins model this hyperplane has a two-dimensional

intersection with the image-plane of the eld of ideal lines V 2 given
by x0 = x1 = x2 = 0. Due to the fact, that we can map every chain to
any other chain and the zero divisor is invariant under multiplication
with an arbitrary element, each chain is contained in a four-space that
intersects the exceptional generator in Studys model with dimension
two.
154
To compute the dual angle we switch the model to C(3,0,0) (D) and
rewrite the line congruence (see Eq. (2.15)) as:

1
(3 22 + + 2)
+

e1
2 + 1
(2 + 1)2

(1 )(22 2 + + 1 + )
+

+
e2
2 + 1
(2 + 1)2

( 1)
3 + 1 + 2
+

+
e3 .
2 + 1
(2 + 1)2
cabc (, ) =
This is done because the dual scalar product can be expressed in

C(3,0,0) (D) with help of the geometric product. The axis A = (a, a
)T R
is transferred to this model and normalized:
A=
1
3((3 2)e1 + (3 + )e2 + (3 + )e3 ).
9
The dual scalar product can be calculated with Eq. (1.27)

cabc (, ), a
= cabc (, ) A =
1
4
3+
3,
3
9
where a denotes the dual unit vector corresponding to A.

Ruled Surface of constant Slope Now we give an example of an
R-chain. Therefore, we have to assume a real cross ratio in Eq. (2.8).
In our example it is sucient to set = 0. This leads to a curve on
M24 in Pl
ucker coordinates:
(1 + 2 )(1 )
2
(1 + )
(1 + 2 )( 2)
.
l() =
3
2
( 2 + )
23 + 42 3 + 1
1 + 3
The striction curve of every R-chain in 2 V is an ellipse that may

degenerate into a line or a point. As example we calculate the striction
curve of the ruled surface given above. Therefore, we describe it in
Euclidean space. A point p on the line given by its Pl
ucker coordinates

(l, l) can be calculated by p =
155
l
l
.
ll
The vector r = lll is the direction

vector of the generator. Therefore, we parametrise the ruled surface
by using a curve on the surface together with the direction of each
generator with
s()
c
a
b
Figure 2.4: Ruled surface of constant slope with striction curve
+22 2+1
23 32 +2
1
.
4 23 +2 +1 +t
f (, t) = 2
3 2 +
( +1)2
3
2
4
3
2
+3 3+1
2 +2
The striction curve of this ruled surface results in

s() =

1
,
+1
2 1,
2 2 + 1
T
This is a rational parametrisation of an ellipse. A visualization of

the ruled surface of constant slope together with its striction curve is
presented in Fig. 2.4.
156

i
i
c
Figure 2.5: Intersection of the ideal curves with the absolute circle.
2.10.3 Dual Quaternion Cross Ratio and Conics on Studys

Quadric
With the interpretation of the cross ratio of dual unit quaternions,
we take a look at conic sections of Studys quadric. In this section
B,
C S 6 be three points
we answer the following question: Let A,
2
contained in Studys quadric that span the two-space P12 . Is there a
P 2 S 6 with cr(a, b, c, d) D, where a, b, c, d are the correspoint D
1
2
ponding Spin group elements or dual unit quaternions. As we know
from section 2.10 we can always map two-spaces that intersect S26
to two-spaces whose intersection with S26 can be interpreted as set
of line-symmetric displacements. Therefore, it is sucient to study
this question for line-symmetric displacements. Thus, we restrict
ourselves to Kleins quadric. A conic on Kleins quadric corresponds
to a regulus, see Def. 1.5. Hence, the question can be formulated as
follows:
Q: Is there an additional line except the three given lines that belongs to the regulus R and to the line congruence L given by the chain?
To answer this question we examine the intersections of R and L
interpreted as set of points in P3 (R) with the ideal plane P given by
x0 = 0. Furthermore, we denote the ideal points contained by the lines
corresponding to a, b, c by A, B, C . To classify the conic section c of R
with the ideal plane we intersect it with the absolute circle given by
i : x21 + x22 + x23 = 0. This leads to three dierent cases corresponding to
a general hyperboloid, see Fig. 2.5 (left), a hyperboloid of revolution,

c
B
l
157
B
l
B
l=c
A
C
Figure 2.6: Intersection of conic sections in P
Fig. 2.5 (middle) and a hyperbolic paraboloid or a degenerate conic,

i.e., two pencils of lines with one line in common, Fig. 2.5 (right).
The congruence of lines L contains a hyperboloid of revolution and
all other lines are parallel to the lines of this hyperboloid. Hence,
the intersection of L with P results in the same gure as for the
hyperboloid of revolution, see Fig. 2.6 (right). To answer Q we have
to intersect the conics in the ideal plane. Since two quadratic curves
have in general four, possibly complex, points of intersection and
we already know three of these four points, we can assume that the
fourth point of intersection is real. Therefore, there are three dierent
possible cases, see Fig. 2.6. For hyperbolic paraboloids the intersection
of the congruence of lines with the ideal plane and the intersection
of the ruled surface itself with the ideal plane are identical in every
case. The results we achieve for hyperboloids also hold for hyperbolic
paraboloids. All in all the following cases can occur: There are three
intersection points and one point is a double point, there are four
dierent points, or the conics are identical. Now we formulate this as
a theorem.
Theorem 2.21. Let A, B, C S26 P7 (R) be three points that are not
joined by a line contained in S26 , i.e., the conic c(t1 : t0 ) dened by
A, B, C is regular. Furthermore, let a, b, c Spin(3,0,1) be the corresponding Spin group elements. Then, one of the three statements is
true:
(1) There is no point D c(t1 : t0 ) with cr(a, b, c, d) D.
(2) There is exactly one point D c(t1 : t0 ) with cr(a, b, c, d) D.
158
(3) For every point D c(t1 : t0 ) the cross ratio cr(a, b, c, d) is a dual
number.
To show that each case occurs we give examples for every case.
Example 2.10. We take the chain from Ex. 2.8 given by Eq. (2.15).
This chain is generated by
a = e1 + e2 + e3 ,
b = e2 + e3 ,
c = e1 + e3 ,
with corresponding Pl
ucker coordinates given by A = (1, 0, 0, 0, 1, 1)T R,
T
B = (0, 1, 0, 0, 0, 1) R and C = (0, 0, 1, 1, 0, 0)T R. Now we parametrise
the conic on M24 dened by these three points with a homogeneous
parameter (t1 : t0 ).
c(t1 : t0 ) = t1 t0 (A, B)C + (t20 t1 t0 )(A, C)B + (t21 t1 t0 )(C, B)A
= (t21 t0 t1 , 2t20 2t1 t0 , t1 t0 , t1 t0 , t21 t1 t0 , t21 t1 t0 + t20 )T .
The chain from Ex. 2.8 is given by

1
(3 22 + + 2)
cabc (, ) =
+
e1
2 + 1
(2 + 1)2

(1 )(22 2 + + 1 + )
+
+
e2
2 + 1
(2 + 1)2

( 1)
3 + 1 + 2
+

+
e3 .
2 + 1
(2 + 1)2
If we compare this with the results of section 2.10.2, we see that the
resulting chain is exactly the same. The intersection of c(t1 : t0 ) with
P is given by
c1 (t1 : t0 ) = (t21 t0 t1 , 2t20 2t1 t0 , t1 t0 )T .
Moreover, the intersection of cabc (, ) with P can be represented in

homogeneous coordinates by
c2 (t1 : t0 ) = (t20 t0 t1 , t0 t1 , t21 t0 t1 )T .
To intersect c1 and c2 we need the implicit equation of one of them.

Therefore, we look for a collineation in P2 (R) that transforms the
159
standard conic parametrised by p(t1 : t0 ) = (t20 , t0 t1 , t21 )T with implicit

form y0 y2 y12 = 0 to c2 (t1 : t0 ). This can be realized by
2 2

t0
t0 t0 t1
x0
1 1 0
0 1 0 t0 t1 = t0 t1 = x1 .
t21
t21 t0 t1
x2
0 1 1
The inverse collineation maps the conic to the standard conic
1
0
0

x0
x0 + x1
y0
0
= y1 .
x1 =
x1
0
x2
x1 + x2
y2
1
1
1
1
With the implicit form of the standard conic we can now determine
the implicit form of c2 (t1 : t0 ) as
y0 y2 y12 = (x0 + x1 )(x1 + x2 ) x21 = x0 x1 + x0 x2 + x1 x2 = 0.
Now we are able to intersect both conics, i.e., we insert the parameter
form of c1 in the implicit form of c2 . This leads to the equation
t20 t21 t0 t31 = 0.
There are four solutions given by the homogeneous parameters

(t1 : t0 )1 = (0 : 1), (t1 : t0 )2 = (0 : 1), (t1 : t0 )3 = (1 : 1), (t1 : t0 )4 = (1 : 0).
We see that solution 1 and solution 2 are identical. Thus, there is

no point D c(t1 : t2 ) with corresponding algebra element d satisfying
cr(a, b, c, d) D, and we have case (1) of Th. 2.21. Now we do the
same calculation but we start with a, b, and

2 5
2 7
1 4
e2 +
+ e3 .
+ e1 +
3 9
3 9
3 9
c = cabc (2, 0) =
The corresponding Pl
ucker coordinate vector is given by
T
C = (3, 6, 6, 4, 5, 7) R.
Of course, this results in the same chain, and therefore, in the same
ideal conic of the congruence of lines. The regulus dened by the
points A, B, C can be parametrised by
c(t1 : t0 ) = (4t0 t1 + t21 , 10t1 t0 + 16t20 , 6t1 t0 ,
160

4t0 t1 , 6t1 t0 + t21 , 16t20 10t1 t0 + t21 )T R.
Its intersection with the ideal plane P of P3 (R) is

c3 (t1 : t0 ) = (t1 (4t0 t1 ), 2t0 (5t1 + 8t0 ), 6t1 t0 )T .
The intersection of c2 with c3 leads to

7t21 t20 + 8t1 t30 t31 t0 = 0.
This yields
(t1 : t0 )1 = (0 : 1), (t1 : t0 )2 = (1 : 0), (t1 : t0 )3 = (8 : 1), (t1 : t0 )4 = (1 : 1).
We see that there is a solution that does not correspond to the start
points A, B , or C . Therefore, we have
D = c(8 : 1) = 16(6, 6, 3, 2, 7, 10)T R.
As dual vector this line has the form

d=
1
((6 2)e1 + (6 + 7)e2 + (3 + 10)e3 )
16
and after normalization the corresponding Spin group element is given

by
d=
1
((6 2)e1 + (6 + 7)e2 + (3 + 10)e3 ) .
9
Finally, we verify that the cross ratio is a dual number

cr(a, b, c, d) = (d a)(b a)1 (b c)(d c)1 =
2
1
+ .
3
9
Hence, we arrive at case (2) of Th. 2.21.

In the third case, every point on the conic leads to a dual cross ratio.
This situation occurs if the regulus is a hyperboloid of revolution. In
fact, the cross ratio is a real number in this case.
Remark 2.36. For the hyperbolic paraboloid we have the same cases.
All lines dene a dual cross ratio with three given lines if all lines of
one regulus intersect the lines of the other regulus orthogonally. If this
is not the case, there may be one line with dual cross ratio or there
may be no one.
2.11 Chains of Geometric Entities
161
Remark 2.37. For general conics on S26 there are two possibilities to
perform this calculation. The rst one is based on the transformation
presented in section 2.10. After the transformation is performed we
can proceed with line-symmetric displacements. The second method is
to take the conic that is dened by the rst four Study coordinates and
intersect it with the conic dened by the rst four Study coordinates
corresponding to the chain that is dened by the three points.

Chains can be realized in every Cliord algebra. The corresponding
geometric entities dier. Thus, it is interesting to examine the properties of the chain with respect to the geometric meaning. Of course,
we can not do this in detail. Furthermore, it can be helpful to use
subgroups
to obtain specialized chains. This means that a chain in
1
V of QnGA dened by three general inversions in hyperquadrics in
principal position except for translation, delivers a chain where every
element is again an inversion in a hyperquadric in principal position.
If we take three inversions in hyperspheres, i.e., three elements of a
subgroup, the entire chain consists of inversions in hyperspheres, see
remark 2.23. Furthermore, we can interpret these inversions as the
corresponding geometric entities.
2.11.1 Quadric Geometric Algebra

In section 1.10 we introduced the quadric geometric algebra. In this
algebra we can perform inversions with respect to hyperquadrics in
principal position. In this section we give two examples for the threedimensional case to show properties of chains we have examined in the
previous sections. The rst example is the chain cabc that is dened
by three algebra elements. These
three elements were generated with
the help of the map : P R3 1 V (C(6,3,0) ), see Eq. (1.59).

a = (1, 0, 0)T (1, 0, 0)T (0, 1, 0)T

(0, 1, 0)T (0, 0, 1)T (0, 0, 1)T I
= 6e1 + e3 6e4 + e6 6e7 + e9 ,
162

7
13
3
b = ( , 0, 0)T ( , 0, 0)T (5, , 0)T
2
2
2
3 T
3 T
3
(5, , 0) (5, 0, ) (5, 0, )T I
2
2
2
32805
199017
6561
6561
e1
e2
e3
e4
=
64
64
512
64
199017
6561
199017
e6
e7
e9 ,
512
64
512

c = (4, 4, 0)T (6, 4, 0)T (5, 5, 0)T

(5, 3, 0)T (5, 4, 1)T (5, 4, 1)T I
= 6e1 30e2 40e3 6e4 24e5 40e6 6e7 40e9 .
After normalization we get

1
1
1
a = e1 + e3 e4 + e6 e7 + e9 ,
6
6
6
10
91
2
91
2
91
2
b = e1 e2 e3 e4 e6 e7 e9 ,
3
3
36
3
36
3
36
20
20
20
c = e1 5e2 e3 e4 4e5 e6 e7 e9 .
3
3
3
Note, that the normalization does not change the GIPNS of a, b, or c.

The geometric inner product null spaces are given by

NIG (a) = (x, y, z)T R3 | x2 +y 2 +z 2 1 = 0 ,

91
T
3
2
2
2
=0 ,
NIG (b) = (x, y, z) R | x +y +z 10x+
4

NIG (c) = (x, y, z)T R3 | x2 +y 2 +z 2 10x8y+40 = 0 .
The GIPNS of a is the unit sphere centered at the origin, the quadric
corresponding to b is the sphere centered at M = (5, 0, 0)T with radius
r = 23 and c is an inner product sphere with radius r = 1 and center
M = (5, 4, 0)T , see Fig. 2.7 (left). We compute the chain cabc dened
by the three elements:
5 (99 + 65)e2
1 (63 + 2 + 992 )e1
2
3 28 + 33 + 21
3 28 + 332 + 21
1 (63 + 2 + 992 )e4
1 (63 166 + 39602 )e3
18
28 + 332 + 21
3 (28 + 332 + 21)
cabc () =
163
b
a
Figure 2.7: Chain in quadric geometric algebra

1 (63 166 + 39602 )e6
132( 1)e5
2
28 + 33 + 21 18
(28 + 332 + 21)
1 (63 + 2 + 992 )e7
1 (63 166 + 39602 )e9
.
2
3 28 + 33 + 21
18
28 + 332 + 21
From Eq. (1.55) we see, that the geometric inner product null space
of a vector determines a sphere if the coecients of e1 , e4 and e7 are
equal. Thus, every entity contained in cabc corresponds to a sphere,
see Fig. 2.7 (right) for the cyclide that is the envelope of this set of
spheres. The GIPNS of the chain cabc () can be determined as

NIG (cabc ()) = (x, y, z)T R3 | 9902 x650x+63x2 +2x2 +992 x2
63166+39602 +63y 2 +2y 2 +992 y 2 7922 y

+792y + 63z 2 +2z 2 +992 z 2 = 0, R .
The implicit equation of the envelope is determined by

6236x4 61720x3 50688x2 y+12472x2 y 2 +137784x2 +12472x2 z 2
61720xy 2 +257400xy+8420x61720xz 2 256369+243409z 2 +6236y 4
50688yz 2 +12472y 2 z 2 +115632y+6236z 4 50688y 3 +86593y 2 = 0.
Furthermore, the chain is dened by three grade-1 elements corresponding to inner product spheres that can also be interpreted as Pin
group elements or as inversions in these spheres. Since the group of
conformal transformations is a subgroup of the group of reections in
quadrics in principal position it also denes a sub chain geometry, see
remark 2.23.
164

b
Figure 2.8: Chain in quadric geometric algebra
Remark 2.38. Algebra elements corresponding to inner product

spheres with identical radii form a subchain geometry. Thus, a chain
generated by three spheres with radius r = c0 R contains only spheres
with radius c0 . Furthermore, the centers of the spheres are contained
in a circle.
The second example we examine is given by the three quadrics that
are the GIPNS of
a = 36e1 + 6e3 36e4 + 6e6 18e7 9e8 + 6e9 ,
b = 24e1 120e2 91e3 24e4 91e6 24e7 91e9 ,
c = 3e1 15e2 20e3 3e4 12e5 20e6 3e7 20e9 .
The spheres corresponding to b and c are identical with the spheres

of the previous example. Only the quadric corresponding to a was
changed, it is an ellipsoid, see Fig. 2.8 (left) given by
2x2 + 2y 2 + z 2 z 2 = 0.
Furthermore, we did not normalize the elements since this does not
change the GIPNS and a calculation without normalization is much
faster. The chain cabc results in
15(1997 + 2147)e2
3(19972 817 + 2964)e1
1726 + 19972 + 247

1726 + 19972 + 247
3(19972 817 + 2964)e4
2(741 + 4340 + 199702 )e3
1726 + 19972 + 247

1726 + 19972 + 247
2(741 + 4340 + 199702 )e6
23964( 1)e5
1726 + 19972 + 247

1726 + 19972 + 247
cabc () =
2.12 Biarc Construction
165
2223( 1)e8
3(665 + 1482 + 19972 )e7
+
2
1726 + 1997 + 247
1726 + 19972 + 247
2
2(741 + 4340 + 19970 )e9
.
1726 + 19972 + 247
Again we compute the GIPNS of the chain cabc and get

NIG (cabc ()) = (x, y, z)T R3 | 520804446z2451x2 +4446z 2 +4446z
2451y 2 889264410x+1995z 2 +47928y+59912 x2
+ 2396402 +59912 z 2 479282 y599102 x+59912 y 2
+ 8892x2 + 8892y 2 = 0, R} .
As parametrisation of the family of quadrics in principal position. We

determine the envelope of this family, see Fig. 2.8 (right)
3745280640+2946608160x1095825792y1574951040z+1472358564x2
+2089543536y 2 +1273701252z 2 +2058028320xy+109803774y 2 z 2
347861424z 2 y+546057720zx+142058592zy28249884zx2
28249884zy 2 +138053658x2 y 2 +109803774x2 z 2 489920016x2 y
815538900xy 2 269481180xz 2 +69026829y 4 489920016y 3 +34187973z 4
41427828z 3 +69026829x4 815538900x3 = 0.
Note, the envelope of the one-parameter set of quadrics in principal

position may have more than one connected component, since not all
quadrics have to be real.

Originally, biarcs are circular arcs that interpolate given points Pi
with additional tangent information Ti , i = 0, 1. At each point, the
circular arcs have G1 -continuity, see Fig. 2.9. All transition points
are located on a xed transition circle j passing through the points
P0 and P1 . In Fig. 2.9 (left) two possible biarcs c0 and c1 for given
data (Pi , Ti ), i = 0, 1 are presented. On the right side of Fig. 2.9 the
orientation of the tangent at P1 is changed an the resulting biarc c2
has a cusp. A generalization are biarcs on quadrics, i.e., conic sections
on a quadric that have G1 -continuity at the transition points. Biarcs
166
on quadrics were investigated by W. Wang and B. Joe, see [64].

With the help of biarcs on quadrics circular arcs in three-dimensional
space can be achieved, because two points with skew tangents dene
a sphere. On this sphere the biarc can be obtained as conic section.
The basics of the biarc construction on quadrics can also be found
in [56]. Since we want to formulate biarcs in the context of chain
geometry we repeat the necessary basics.
P0
P0
j
L0
j
L0
c0
c1
c2
L1 P1
c0
P1
L1
Figure 2.9: Biarcs corresponding to (Pi , Li ), i = 0, 1, where Li is the

tangent dened by [Pi , Ti ] , i = 0, 1
The Biarc Construction in projective Space To describe the biarc

construction we follow [56]. Let P0 , T0 and P1 , T1 be points in projective space written in homogeneous coordinates with respect to some
coordinate system:
P0 = p0 R,
P1 = p1 R,
T0 = t0 R,
T1 = t1 R.
Let be the polarity and M the symmetric matrix describing . The

quadric Q is the set of points X = xR : xT Mx = 0. We assume that
the points P0 , P1 Q and [P0 , T0 ], [P1 , T1 ] are the tangents of Q at the
points P0 and P1 . Thus, we have
pT
0 Mp0 = 0,
pT
1 Mp1 = 0,
pT
0 Mt0 = 0,
pT
1 Mt0 = 0.
The connection point C = cR of both conic sections has to lie on Q:

cT Mc = 0. The tangents at P0 and P1 are given by
b = p0 + t0 ,
d = p1 + t1 ,
, R.
167
B
T0
C
P0
P1
T1
D
Figure 2.10: Biarc construction on a sphere. The transition point C is

located on the conic sections through P0 and P1 .
To obtain G1 -continuity the point C has to be located on the line

spanned by b and d, i.e., c = b + d. Therefore, we can formulate
one condition for G1 -continuity. The line joining B = bR and D = dR
has to be tangent to Q, see Fig. 2.10. This means it has exactly one
intersection point with Q. Thus, we compute
(b + d)T M(b + d) = 2 bT Mb + 2 dT Md + bT Md + dT Mb
(2.18)
2 T
2 T
= b Mb + d Md + 2b Md = 0.
As solution of this equation we need a double root for G1 -continuity.

This leads to
2
bT Md dT Md
+
= 0 =
+2 T
2
b Mb bT Mb
)

(bT Md)2
bT Md
dT Md
.
= T
+ T
T
2
1,2
b Mb
(b Mb)
b Mb
To obtain a double root the discriminant has to vanish:

0 = (bT Md)2 + (bT Mb)(dT Md)
(2.19)

T
T
= (p0 + t0 ) M(p0 + t0 ) (p1 + t1 ) M(p1 + t1 )

2
+ (p0 + t0 )T M(p1 + t1 )

*
T
T
T
T Mt tT Mt
= pT
Mp
+
p
Mt
+
t
Mp
+
t
Mt
t
1
1
1
1
0
1 .
0
0
0
0
0
1
168
T
If tT
0 Mt0 t1 Mt1 0 Eq. (2.19) is solvable and we can determine :
and : afterwards. The set of points that serves as connection
point is determined by the two conics through P0 and P1 that are
solutions of Eq. (2.18). Note, that the existence of a biarc cannot be
guaranteed in general. If both tangents are situated on dierent sides
of the quadric, there is no real point C = cR. This is the geometric
interpretation of the vanishing discriminant, see Eq. (2.19). If the
quadric Q is an oval one, there is no tangent in the interior of Q.
Thus, for oval quadrics we can always nd biarcs. Naturally, a biarc
is not determined uniquely. The set of all possible transition points is
contained in two dierent conics passing through P0 and P1 , see [56].
2.12.1 Biarcs as touching Chains

Conic sections on arbitrary quadrics can be described with the help
of chains in a geometric algebra corresponding to this quadric, see
section 2.8. Points are null vectors and the chains can be calculated
with Eq. (2.8). Furthermore, we introduced the concept of contact
spaces in section 2.6. Touching chains can be interpreted as conics
with G1 -continuity. We can use Th. 2.16 to parametrise the chain in
contact to another chain at a certain point. To apply this to biarc
construction, we have to change the setting. We do not assume that
we are given two points and their tangents, since we do not have
tangents in the Cliord algebra model. We start
with a chain cabc

determined by the three null vectors a, b, c 1 V and compute the
chain through a null vector d that has contact with cabc at c. The
advantage of this method is that we obtain a parametrisation for each
biarc. Furthermore, we can use the fact that connected components
of the Pin group and the grade-1 subspace form subchain geometries
to describe biarc interpolation in an ane Cliord algebra setting.
Biarc Construction with Cliord Algebra Let cabc be the chain
dened by a, b, c C(p,q,r) . Let d C(p,q,r) be an element with d
/ cabc .
With Th. 2.16 we can parametrise the chain cd that has contact with
cabc at c by
cr(a, b, s, c) cr(a, b, d, c) = ,
C(Cp,q,r ),
169
where s is contained by cd . Now we compute s():

cr(a, b, s, c) cr(a, c, d, b) =
(2.20)
(ca)(ba)1 (bs)(cs)1 cr(a, b, d, c = ,

(ca)(ba)1 b(ca)(ba)1 s = (+cr(a, c, d, b))(cs),

+cr(a, b, d, c)(ca)(ba)1 s = c+cr(a, b, d, c)c(ca)(ba)1 b ,

1

c+cr(a, b, d, c)c(ca)(ba)1 b ,
s = +cr(a, b, d, c)(ca)(ba)1

with
cr(a, b, d, c) = (c a)(b a)1 (b d)(c d)1 .
For our purpose the elements a, b, c, d are null vectors in a Cliord

algebra constructed with the matrix of a polarity of some quadric.
Furthermore, we assume R to obtain conic sections. Formula
(2.20) can be implemented in a computer algebra system to obtain
parametrisations of the chain that is in contact with cabc at the
element c. The same calculation can be performed in an ane setting.
Therefore, we model the quadric in the grade-1 space with an ane
Cliord algebra model. Points of the quadric are represented by
elements that square to 1.
2.12.2 Biarcs on Quadrics in three-dimensional Space

Now we give some examples for biarcs on quadrics in three-dimensional
space calculated with the help of chains that are in contact.
Sphere The rst example is the unit sphere S 2 R3 given by its
implicit equation
x2 + y 2 + z 2 = 1.
Projective Setting The unit sphere can be considered as projective

quadric
0
Q : xT Mx = 0, with X = xR P3 (R) and M =
0
0
0
1
0
0
0
0
1
0
0
0
.
0
1
170
Thus, we use the homogeneous Cliord algebra model constructed with

V = R4 as vector space model for P3 (R) together with the quadratic
form dened by M. Let a = e1 + e4 , b = e1 + e2 , and c = e1 + e3 be three
mutually distant null vectors. Note that the homogeneous factor is
contained in the e1 generator and that e2 , e3 , e4 correspond to the x-,
y -, z -coordinate. The chain cabc () is computed with Eq. (2.8):
cabc () = e1 +
(1 )e4
e2
( 1)e3
+
+
.
2
2
1+
(1 + ) 1 + 2
Let d = e1 + 12 e2 + 12 e3 be the null vector corresponding to D = dR =

(1, 12 , 12 , 0)T R. We compute the chain that has contact with cabc at
c with Eq. (2.20)
(1 + 2 )e2
((1 + 2) 2 2 1)e3
s() = e1
+
( 2 + 1) 2 2 2
( 2 + 1) 2 2 2
e4

.
( 2 + 1) 2 2 2
This conic section interpolates the point C = cR related to the null

vector c at = 0 and the point D = dR at the parameter = . A
visualisation of the biarc construction on the sphere is given in Fig.
2.11 (left).
Ane Setting
Now we compute the same biarc

ane Clif with an
1
ford algebra. Therefore, let V = R3 and let Q = 0 1 0. The Clifford algebra has signature (p, q, r) = (3, 0, 0). The square of a general
grade-1 element a = a1 e1 + a2 e2 + a3 e3 is computed as a2 = a21 + a22 + a23 .
If the square is equal to 1 the element corresponds to a point on S 2 .
The points A = (0, 0, 1)T , B = (1, 0, 0)T , and C = (0, 1, 0)T that were
also used in the example above correspond to the elements a = e3 ,
b = e1 and c = e2 with a2 = b2 = c2 = 1. The chain cabc () dened by
these three vectors is given by
cabc () =
e1
( 1)e2
(1 )e3
+
+
.
2
2
1+
1+
1 + 2
171
Figure 2.11: Biarcs on a sphere (left); on an ellipsoid (right).
Let D = ( 12 , 12 , 0)T be a point on S 2 with algebra representation

d = 12 e1 + 12 e2 . The chain s() through d that has contact with cabc
at c is computed with Eq. (2.20)
(1 + 2 )e1
((1 + 2) 2 2 1)e2
s() =
+
( 2 + 1) 2 2 2
( 2 + 1) 2 2 2
e3

.
( 2 + 1) 2 2 2
If we compare the results of the ane biarc construction with the

results of the projective setting, we observe that they are absolutely
identical. Therefore, it does not matter which method we take. The
ane version is faster because the corresponding algebra has half the
dimension of the algebra corresponding to the projective version. In
general it is good to have both versions since there are quadrics like
Kleins quadric that are naturally projective quadrics.
Ellipsoid The second example we discuss is an ellipsoid with the
equation
x2 + y 2 + 2z 2 = 1.
172
Projective Setting The ellipsoid is given by
1
0
T
3
Q : x Mx = 0, with X = xR P (R) and M =
0
0
0
1
0
0
0
0
1
0
0
0
.
0
2
We construct the homogeneous Cliord algebra model with V = R4 and

Q = M. Again, the homogeneous factor is stored in the e1 component
and the x-, y -, z -coordinate is contained in e2 , e3 , e4 . Let A = aR =
(1, 0, 0, 12 )T R, B = bR = (1, 1, 0, 0)T R, and C = cR = (1, 0, 1, 0)T R
be three points on Q. The corresponding null vectors are given by
a = e1 + 12 e4 , b = e1 + e2 , and c = e1 + e3 . The chain cabc () computed
via Eq. (2.8) results in
e2
( 1)e3
1 2(1 )e4
cabc () = e1 +
+
+
.
1 + 2
1 + 2
2 1 + 2
Let D = dR = (1, 12 , 12 , 0)T R be an additional point on Q with algebra

representation d = e1 + 12 e2 + 12 e3 . The chain cd () that has contact
with cabc () is calculated with the help of Eq. (2.20) as:
( 2 1)e2
( 2 1 2 2 + )e3
cd () = e1 +
+ 2 2 2 2
+ 2 2 2 2
2e4
1
2 + 2 2 2 2
Ane Setting For the ane biarc construction on the ellipsoid

we use the Cliord algebra
by V = R3 together with
constructed
1 0 0
the quadratic form Q = 0 1 0. The points A = (0, 0, 12 )T ,

0 0 2
B = (1, 0, 0)T , and C = (0, 1, 0)T on the ellipsoid correspond to elements
a = 12 e2 , b = e1 , and c = e2 , with a2 = b2 = c2 = 1. The chain cabc ()
is determined as
cabc () =
( 1)e2
1 2(1 )e3
e1
+
+
.
1 + 2
1 + 2
2 1 + 2
173
Let D = ( 12 , 12 , 0)T be another point on Q with algebra representation

d = 12 e1 + 12 e2 . The chain in contact with cabc that contains d is
computed via Eq. (2.20)
( 2 1)e1
( 2 1 2 2 + )e2
cd () =
+
+ 2 2 2 2
+ 2 2 2 2
2e3
1
2 + 2 2 2 2
The biarc construction performed above is displayed in Fig 2.11 (right).

One sheeted Hyperboloid Now we do the same on the one sheeted
hyperboloid given by the equation
x2 + y 2 z 2 = 1.
For quadrics that contain ideal points, it is advantageous to perform

the construction in the projective setting, because in this setting it
possible to describe also ideal elements. Therefore, we do not use the
ane setting for this example.
Projective Setting The Cliord algebra model is constructed with
V = R4 as vector space model for P3 (R) and the quadratic form Q = M
corresponding to the quadric
0
Q : xT Mx = 0, with X = xR P3 (R) and M =
0
0
0
1
0
0
0 0
0 0
.
1 0
0 1
We start with the points A = aR = (1, 12 , 12 , 0)T R, B = bR(1, 1, 0, 0)T R,

and C = cR(1, 0, 1, 0)T R Q. The corresponding algebra elements
are given by a = e1 + 12 e2 + 12 e3 , b = e1 + e2 , c = e1 e3 . We
compute the chain dened by these three null vectors as
(2 + 2) + 1 + 2
2 + 2 1 2
cabc () = e1 +
e2
e3 .
2 + 2 + 2 + 2
2 + 2 + 2 + 2
The next point that shall be interpolated is D = dR = (1, 2, 1, 2)T R
174
E
D
E
A
C
A
B
Figure 2.12: Biarc on an one sheeted hyperboloid (left); on a paraboloid

(right).
Q. Its algebra representation is d = e1 2e2 + e3 + 2e4 and the chain
through d in contact with cabc at c results in
2(3 + 4 + 22 + 2 2 + 2 2)e1
4(2 2 + 3 + 2 + 2)e2
cd () =
(2 + 2 + 2)2
(2 + 2 + 2)2
2(2 2 3 + 4 + 22 + 2 2)e3
4(2 2 + 3)e4
+
.
(2 + 2 + 2)2
(2 + 2 + 2)2
Note that cd contains a null vector

corresponding to an ideal Tpoint
1
for the parameter value = 1 2 2. Let E = eR = (1, 1, 3, 3) R be
another point on Q with algebra representation e = e1 e2 + 3e3 + 3e4 .
The chain ce in contact with cd at d has the form:
2(12 + 8 2 2 + (4 2 8))e2
ce () = e1
2 + (2 2 4) 24 + 16 2
(72 + 48 2 2 + (10 2 20))e3
+
2 + (2 2 4) 24 + 16 2
2(36 + 24 2 2 + (6 2 12))e4
.
+
2 + (2 2 4) 24 + 16 2
The corresponding biarc interpolation is presented in Fig. 2.12 (left).
175
Paraboloid As last example we examine the biarc construction on a

paraboloid with equation:
x2 + y 2 z = 0.
Projective Setting The homogeneous Cliord algebra model is constructed with V = R4 as model for P3 (R) together with the quadratic
form Q with the coecient matrix
0 0 0
0 1 0
T
3
Q : x Mx = 0, with X = xR P (R) and M =
0 0 1
1
0 0
2
1
2
0
.
0
0
As example
we again choose three points on the paraboloid. Let A =
aR = (1, 2, 0, 2)T R, B = bR = (1, 1, 0, 1)T R, and C = cR = (1,0, 1, 1)T R
be three dierent points with algebra representation a = e1 + 2e2 +2e4 ,
b = e1 + e2 + e4 , c = e1 + e3 + e4 . We compute the chain cabc with Eq.
(2.8).
( + 8 + 6 2)e2
(1 + )e3
cabc () = e1 +
2
2
+ (2 2 + 2) + 6 + 4 2 + (2 2 + 2) + 6 + 4 2
(2 + (2 2 + 4) 12 8 2)e4
2 + (2 2 + 2) + 6 + 4 2
Let D = dR = (1, 12 , 12 , 1)T R and E = eR = (1, 22 , 12 , 52 )T R be two

additional points on the paraboloid and d = e1 + 12 e2 + 12 e3 + e4 ,
e = e1 + 22 e2 + 12 e3 + 52 e4 the corresponding null vectors. The chain
cd that is in contact with cabc at c is computed with Eq. (2.20) and
has the form
( + 21 + 15 2)e2
cd () = e1
2 + (3 2 + 4) 21 2 30
(2 + ( 2 + 1) 21 15 2)e3
+
2 + (3 2 + 4) 21 2 30
(2 + ( 2 + 2) + 30 + 21 2)e4
2 + (3 2 + 4) 21 2 30
176
Moreover, we compute the chain ce that has contact with cd at d
1 (384 36 2 + 32 2 + 272 2 + 48)e2
ce () = e1 +
2
32 + (16 2 24) 96 2 + 136
1 ((20 16 2) + 3 22 + 136 2 192)e3
+
2
32 + (16 2 24) 96 2 + 136
2
(3 + (18 2 28) + 340 240 2)e4
.
+
32 + (16 2 24) 96 2 + 136
An illustration of the paraboloid together with the conic section

belonging to cabc , cd and ce is given in Fig. 2.12 (right).

To present an example in a higher-dimensional projective space we
construct biarcs on Kleins quadric M24 . We introduced Kleins quadric in section 1.2.1. Conics on M24 correspond to reguli. Thus, a
biarc construction on M24 can directly be transferred to an interpolation algorithm that interpolates a set of given lines by reguli. To
demonstrate the power of the Cliord algebra apparatus we apply
the concept of touching chains to the homogeneous Cliord algebra
model for Kleins quadric, see section 1.6. Hence, we use V = R6 as
model for P5 (R) and

Q=
O
I
I
,
O
where O denotes 3 3 zero matrix and I the 3 3 identity matrix. As example we take the same conic as in Ex. 2.5, but with
another parametrisation. We start with the Pl
ucker coordinates
A = aR = (0, 1, 0, 0, 0, 1)T R, B = bR = (1, 1, 1, 0, 0, 0)T R, and C = cR =
(0, 0, 1, 0, 1, 0)T R. Obviously, A, B, C M24 . The corresponding null
vectors are given by
a = e2 + e 6 ,
b = e1 + e2 + e3 ,
c = e 3 + e5 .
The chain dened by these three null vectors is computed with Eq.
(2.8)
cabc () =
2e1
(1 + )e2
(1 + )e3
( 1)e5
( 1)e6
+
+
+
.
1 + 2
1 + 2
1 + 2
1 + 2
1 + 2
177
s1
s2
B
D
C
Figure 2.13: Biarc construction with reguli.
Let D = dR = (552, 140, 371, 336, 827, 812)T R be another point on

M24 with algebra representation
d = 552e1 140e2 + 371e3 336e4 + 827e5 + 812e6 .
The chain cd containing d that has contact with cabc is given by:
24(23+28)e1
28(12+5)e2
(704+3362 +371)e3
+
607+1040+3362 607+1040+3362
607+1040+3362
2
336e4
(1376+336 +827)e5
28(12+29)e6
+
+
.
607+1040+3362
607+1040+3362
607+1040+3362
cd () =
Fig. 2.13 shows the lines corresponding to the points A, B, C, D, the

reguli corresponding to the conic sections, and their striction curves
s1 , s2 . These two reguli posses G1 -continuity at the transition line
given by its Pl
ucker coordinates C .
2.12.4 Biarcs on the Dual Sphere

With the concept of touching chains we are also able to compute conics
on SD2 that are in contact, i.e., congruences of lines having a constant
178
dual angle with respect to a xed line. In general, two such congruences with constant dual angles with respect to a xed line may have
zero, one, two, or innitely many oriented lines in common. The adequate ane model is the Cliord algebra C(0,3,0) (D) given by M = D3
0 0
ucker coordinates
1 0. As example we give the Pl
0 1
of four lines A = aR = (0, 1, 0, 0, 0, 1)T R, B = bR = ( 12 , 23 , 23 , 49 , 19 , 19 )T R,
C = cR = (0, 0, 1, 0, 1, 0)T R, D = dR = ( 35 , 0, 45 , 85 , 2, 65 )T R. The corres
1
and Q = 0
0
ponding algebra representations where we choose the orientation of

the related spears are
a = e2 +e3 ,
c = e2 +e3 ,
1 4
2 1
2 1
b = ( + )e1 +( + )e2 +( + )e3 ,
3 9
3 9
3 9
8
6
3
4
1
d = ( )e1 +( +2)e2 +( + )e3 .
5
5
5
5
5
We compute the chains cabc and cacd for a dual cross ratio with Eq.
(2.8).

(+1)(2 ++)
+

e1
2 +1+
(2 +1+)2

+1
(2+3 +2 1)
+
+

e2
2 +1+
(2 +1+)2

(+1) 2+3 1
+

e3 ,
2 +1+
(2 +1+)2

3(1)
6+3+42 122 483 +404 +4
e1
cacd (, ) =
1+52
(1+52 )2

(109+5152 +503 1)
1
+
+
e2
(1+52 )
(1+52 )2

(4+1)
(1)(303 312 +791)
+
+
e3 .
(1+52 )
(1+52 )2
cabc (, ) =
These two chains parametrise the congruences of lines that corresponds

to the intersection of two ane submodules spanned by a, b, c and
2
a, c, d with the dual unit sphere SD
. The congruences of lines have
exactly two oriented lines in common, i.e., the lines dened by a and
c. To construct a biarc we use again the chain geometric method. The
179
B
C
s2
s1
D
Figure 2.14: Biarc construction on dual unit sphere
chain cd that is in contact with cabc can be computed with a dual cross
ratio + , since the scalar domain of the algebra is the commutative
ring of dual numbers. Both chains correspond to congruences of lines
with constant dual angle to an uniquely determined oriented line.
Hence, the two congruences of lines dene two cylinders that intersect
each other in a single point if the congruences of lines are in contact.
Furthermore both congruences of lines have a whole bundle of parallel
lines in common. This is obvious because we can translate the line
that belongs to both congruences and that passes through the point
of contact of the two cylinders along the axis of one of the cylinders
and it is still contained in both congruences. A parametrisation of
this congruence can be obtained with Eq. (2.20):

( + 3)
4 + 16 + 40 + 6 + 2 + 3
cd (, ) =
e1
5 + 3 + 2
(5 + 3 + 2 )2

282 5 + 70 + 50 + 2 + 73 + 4
+
+
e2
5 + 3 + 2
(5 + 3 + 2 )2
2

+ 3 + 4 2 + 52 + 3 + 12 + 30 + 3
+
+
e3 .
5 + 3 + 2
(5 + 3 + 2 )2
This is a parametrisation of the entire congruence. Now we want

to examine the same problem with a real value for the cross ratio.
180
This can be achieved easily by setting = 0. The resulting chains

correspond to ruled surfaces of constant slope that have G1 -continuity
at the transition line. In Fig. 2.14 the ruled surfaces of constant
slope corresponding to the biarc are visualized together with the
lines corresponding to A, B, C, D and the striction curves of the ruled
surfaces s1 , s2 . Thus, we can use biarcs on the dual sphere to perform
G1 -interpolation of oriented lines. A parametrisation of the form
f (u, v) = c(u) + v r(u) where c(u) is a curve on the surface and r(u)
a curve on the unit sphere S 2 R3 can be computed with the help
of the corresponding Pl
ucker coordinates. For the chain cabc (, 0) we
obtain the parametrisation
1 u2
u
1
u + v u + 1 ,
f1 (u, v) = 2
u +u+1
(u + 1)u
u
with striction curve

s1 (u) =
1
(1 u2 , u, u)T .
u2 + u + 1
For the chain cd (, 0) we obtain
4
u 7u3 28u2 68u40
1
f2 (u, v) = 2
u3 u2 20u50
(u +3u+5)2
u3 +6u2 +26u+30
15+14u+u3 +6u2
,
+v
5uu3 3u2
3
4
2
20+27u+6u +u +18u
with striction curve

s2 (u) =
u2
1
(1 u2 4u, u 10, 13u + 18)T .
+ 3u + 5
Remark 2.39. The concept of chains that are in contact is not

restricted to the grade-1 space. We used contact chains to construct
biarcs on quadrics as they are understood classically. Naturally, chains
in the whole algebra can be in contact, and therefore, we can also use
this concept to construct chains in subgroups that are in contact with
each other. For example we can determine chains in contact within
Pin or Spin groups.
3 Kinematic Mappings for Spin

Groups
In this chapter we dene Cayley-Klein spaces and show how to describe
certain Cayley-Klein spaces within the homogeneous Cliord algebra
model. This construction is accomplished in detail for the threedimensional Euclidean space. Furthermore, the kinematic mapping of
Study and the mapping of Blaschke and Gr
unwald are constructed
in a unied method. Matrices of the collineations in the image and
pre-image space are derived. The construction is accomplished in
detail for the Euclidean spaces of dimension two and three. After that,
we give an overview of possible kinematic mappings for Cayley-Klein
spaces of dimension two and three. Moreover, the mapping for the
four-dimensional Euclidean space is presented. This chapter is already
published, see [41].
3.1 Cayley-Klein Geometries and the

homogeneous Model
Cayley-Klein spaces are metric spaces constructed within an n-dimensional projective space Pn (R) with a distinguished quadratic hypersurface given by a quadratic form
xT Mx = 0,
where M is a (n + 1) (n + 1) symmetric matrix and x Pn . This idea

goes back to A. Cayley and F. Klein. Comprehensive work on this
eld of geometry can be found in [44, 52] and [60]. With Silvesters
law of inertia we can always nd a diagonal matrix corresponding
to the quadratic hypersurface, see section 1.3.1. The Cayley-Klein
construction provides models for the Euclidean, hyperbolic, elliptic,
182
and many other geometries. An exhaustive treatise of this topic can

be found in [25]. The denition of a Cayley-Klein space that we will
use can also be found therein.
Denition 3.1. Let Pn (R) be the n-dimensional projective space, then
F Pn (C) is dened by
n 1
n 1
F : Qnr0 ,q0 An1 Qr11,q1 . . . An Qr,q .
Here ri is the rank, qi the index and ni 1 the dimension of the

n 1
quadratic variety Qrii,qi that is for i < a singular quadric and for
i = a regular one. Furthermore, the identity n + 1 = r0 + r1 + r
holds. Pn with absolute gure F is called a Cayley-Klein space. The
points contained in F are called ideal points and the points in Pn \F
n 1
proper points. Qrii,qi is called absolute quadric or absolute cone. Ani
is called absolute ni -subspace. A Cayley-Klein space where the set of
points W is removed is called sliced along W .
Remark 3.1. Points that are neither ideal nor proper are called improper points. These points occur for example in the case of hyperbolic
geometry.
The construction of a Cliord algebra over a projective space results
in the so-called homogeneous model, see section 1.4.
3.1.1 Cayley-Klein Spaces

At rst we present an example for a planar Cayley-Klein space. This
means we construct the homogeneous model over R3 with coordinates
x0 , x1 , x2 as vector space model for P2 (R) and an absolute gure. We
start with the Euclidean plane. For this purpose an absolute gure of
the form
F : Q11,0 A1 Q02,0
(3.1)
is prescribed. The quadric can be written as Q11,0 : x20 = 0. It is singular,
thus cone-shaped and has the vertex space A1 : x0 = 0 which is a line.
The regular quadric at the end of the chain is given by Q02,0 : x21 +x22 = 0.
The projective automorphisms of F form a subgroup PGL(P2 , F)
PGL(P2 ) and thus they constitute the group of isometries in this
Cayley-Klein geometry. In the Euclidean the subgroup of direct
3.1 Cayley-Klein Geometries and the homogeneous Model
183
isometries forms the group SE(2) of planar Euclidean displacements.

Now we construct the homogeneous model for this Cayley-Klein space.
Therefore, we take R3 as vector space and a quadratic form with
signature (2, 0, 1) to construct the Cliord algebra C(2,0,1) . Points
P = (x0 , x1 , x2 )T P2 (R) of the Euclidean plane are described in

the 2 V subspace of the algebra via p = x0 e12 + x1 e13 + x2 e23 . Note
that the homogenizing coordinate of the point is contained in the e12
component. The scalar product is calculated by p p = x20 . Hence,
we have the possibility to norm with operations in the algebra. Thus,
we nd Euclidean geometry considered as a Cayley-Klein geometry
within this model. The group PGL(P2 , F) can be found in the Cliord
algebra as the group Pin(2,0,1) . In fact the Pin group is a double cover
of the group PGL(P2 , F).
Three-dimensional Euclidean Space The three-dimensional Euclidean space is constructed in the same way. We start with P3 (R) and
the absolute gure
F : Q21,0 A2 Q12,0 ,
(3.2)
2
2
wherein the rst singular quadric in the chain is given by Q1,0 : x0 = 0
and its vertex is given by A2 : x0 = 0. Contained in A2 we have
the so called absolute circle Q12,0 : x21 + x22 + x23 = 0. The subgroup
PGL(P3 , F) PGL(P3 ) that xes this absolute gure can be identied
as the group of spatial Euclidean displacements SE(3). We aim at
a representation of this Cayley-Klein geometry with a homogeneous
Cliord algebra model. Therefore, we take R4 as vector space model
for P3 (R) together with a quadratic form of signature (3, 0, 1) to obtain
the Cliord algebra C(3,0,1) . The group Spin(3,0,1) is a double cover of
the group SE(3) of Euclidean motions, see for example [59].
Cayley-Klein Geometry via homogeneous Cliord Algebra Models
In general, a Cayley-Klein space can be described using the homogeneous Cliord algebra approach. All we need is the vector space model
of a projective space and a quadratic form given by its signature. Of
course, not every Cayley-Klein space can be described in this way, but
for our purposes this is convenient. The three-dimensional Gallilei
space considered as Cayley-Klein space serves as counter example
F : Q21,0 A2 Q11,0 A1 Q02,0 .
184
The quadrics and the subspaces can be described by

Q21,0 : x20 = 0, A2 : x0 = 0, Q11,0 : x21 = 0, A1 : x1 = 0, Q02,0 : x22 + x23 = 0.
It is not possible to describe this absolute gure with one quadratic

form, and therefore, we conclude that there exists no homogeneous
Cliord algebra representation for the Gallilei geometry.
3.1.2 A homogeneous Model for Euclidean Geometry

Now we construct the Cliord algebra for the three-dimensional Euclidean space. Therefore, we take V = R4 as model for the threedimensional projective space P3 (R). The signature of the Cliord
algebra is (3, 0, 1). Note that in the literature the signature (0, 3, 1)
is often used, because of the connection to dual quaternions, see
[59]. We will do the construction in a more natural way and take
the scalar product of Euclidean geometry. We have four basis elements e1 , e2 , e3 , e4 with e21 = e22 = e23 = 1 and e24 = 0. The grade-1
space is equipped with the quadratic form, therefore the points are
described by grade-3 elements. This approach is called a plane-based
approach, see [27]. Grade-2 and grade-1 elements correspond to lines
and planes, respectively. Note that not every grade-2 element corresponds to a line. The even part of the algebra C+
(3,0,1) is spanned by
e0 , e12 , e13 , e14 , e23 , e24 , e34 , e1234 and the Spin group is a double cover of
+
SE(3). Note that in this case we have C+
(3,0,1) = C(0,3,1) .
Isomorphism to Hd Dual quaternions were introduced in section
1.1.2. A more detailed introduction can be found in [38]. We give an
isomorphism between C+
(3,0,1) and Hd as introduced by E. Study, cf.
[61]. This isomorphism can also be found in [59] and is given by its
action on the basis:
e0 1,
e23 i,
e31 j,
e12 k,
e1234 ,
e14 i,
e24 j,
e34 k.
(3.3)
We can dene a group isomorphism between the group of dual unit

quaternions and Spin(3,0,1) if we restrict the isomorphism (3.3) to one
of these groups in the image or pre-image.
3.2 Kinematic Mappings
185
3.2 Kinematic Mappings

In this section we give a short introduction to the concept of kinematic mappings. We will treat two important examples. Kinematic
mappings map displacements to points in a certain space. The main
advantage is to work with points instead of displacements.
3.2.1 Studys kinematic Mapping

In section 1.2.2 we introduced the kinematic mapping of Study. We recall Studys mapping that maps displacements to a sliced hyperquadric
S26 \V 3
: SE(3) S26 \V 3 P7 (R),
(3.4)
SE(3) A = (a0 , a1 , a2 , a3 , c0 , c1 , c2 , c3 ) .
T
So far we did not mention that the exceptional generator V 3 is the

vertex of a singular quadric
N26 : a20 + a21 + a22 + a23 = 0.
The group of collineations in the image

that corresponds to
6 6space

7
SE(3) are the collineations
PGL(P
,
S
,
N
that
leave the pencil of
)
2
2

quadrics S26 , N26 invariant. Note that the image space is not a CayleyKlein space. On one hand we could interpret the seven-dimensional
projective space with its absolute gure
F : S26 V 3 Q24,0
as Cayley-Klein space when we use the real numbers as underlying

eld. On the other hand, if we look with complex glasses on the
scene, the exceptional
generator V 3 becomes the vertex space of a
3
6
singular quadric N2 : i=0 a2i = 0, that belongs to the absolute gure.
Furthermore, the quadric Q24,0 is a quadric in the subspace V 3 given
by the equation c20 + c21 + c22 + c23 = 0.
3.2.2 A Mapping for planar Displacements

Here, we present another kinematic mapping introduced by W.
nwald [26]. The interested reader is
Blaschke [6] and J. Gru
186
referred to [13, Ch. 11] for further information. Planar Euclidean

displacements can be decomposed into a rotation and a translation

x
x
cos

=
y
y
sin
sin
cos

a
x
.
+
b
y
The kinematic mapping maps planar Euclidean displacements to

points of P3 (R). Note that the homogeneous coordinates of points are
related to Cartesian coordinates (x, y, z)T via
x=
x1
,
x0
y=
x2
,
x0
z=
x3
,
x0
if x0 = 0.
Again not every point of the image space corresponds to a displacement.

In analogy to section 3.2.1 the line : x0 = x3 = 0 is the real vertex
space of a reducible quadric N22 : x20 + x23 = 0, which consists of the
complex conjugate planes x0 + ix3 = 0 and x0 ix3 = 0. Therefore,
is the line of intersection of these two planes. To get a bijection we
have to remove the line at innity from the image space. Under
these conditions, the map
: SE(2) P(R)\
(3.5)
T
SE(2) (, a, b) 2 cos , a sin b cos , a cos + b sin , 2 sin

2
2
2
2
2
2
is one-to-one and onto. Hence, the image space is a quasi-elliptic

space, cf. [25] with absolute gure
F : Q22,0 A1 Q02,0 ,
Q22,0 : x20 + x23 = 0,
A1 : x0 = x3 = 0,
Q02,0 : x21 + x22 = 0.
Planar Euclidean displacements can be described by a subgroup of

the group Ud , see section 1.1.2. For example: If we want to describe
planar Euclidean displacements in the [x, y]-plane, we have to restrict
the group Ud to rotations around axes that are parallel to the z -axis
and to translations in the [x, y]-plane.
3.3 Kinematic Mappings via Cliord Algebras
187

Both examples presented in the previous section can be treated together in the unifying and more general framework of Cliord algebras.
In the following we shall explain how this works and illustrate the
construction at hand of the previous examples. The Pin group corresponding to a certain homogeneous Cliord algebra model for a
Cayley-Klein space is a double cover of the group of collineations that
preserve the absolute gure F . Moreover, the Spin group forms a
double cover of the group of direct isometries of the corresponding
Cayley-Klein space. Therefore, we just have to look at the Spin group
as a subset of a projective space of the dimension 2n1 .
3.3.1 Studys Mapping via Cliord Algebra

Again we start with the homogeneous model for the three-dimensional
Euclidean space C(3,0,1) . The Spin group is located in the even part
C+
(3,0,1) of the algebra and every Spin group element satises the
condition N (g) = gg = 1.
Remark 3.2. Note, that for n < 6 the elements g of the Spin or
Pin group are characterized by g = 1. For higher dimensions the
sandwich action of an element with gg = 1 does not need to result in
a vector when it is applied to a vector, see [21, p. 30].
Thus, we take an arbitrary element of the even part of the algebra
g = a0 e0 + a3 e12 + a2 e13 + c1 e14 + a1 e23 + c2 e24 + c3 e34 + c0 e1234
and calculate the product with its conjugate element

gg = (a20 + a21 + a22 + a23 )e0 + 2(c0 a0 a1 c1 + c2 a2 c3 a3 )e1234 = 1. (3.6)
Eq. (3.6) shows that g Spin(3,0,1) if the quadratic equation in the

pseudoscalar part is equal to zero. With the isomorphism (3.3) we can
see that this condition is exactly the equation of Studys quadric. The
values ai , ci , i = 0, . . . , 3 are the coordinates of a Euclidean displacement,
for
c0 a0 a1 c1 + c2 a2 c3 a3 = 0.
(3.7)
188
Furthermore, the scalar part of Eq. (3.6) has to be equal to 1. Now

we identify the eight components a0 , . . . , a3 , c0 , . . . , c3 with coordinates
of a seven-dimensional projective space P7 (R). The condition in the
scalar part can now be relaxed to
a20 + a21 + a22 + a23 = 0.
This equation can be identied as the condition that a point is not

allowed to lie in the exceptional space V 3 . Hence, we have found a
mapping from Spin(3,0,1) S26 \V 3 , compare to Eq. (3.4). Note that g
and g represent the same displacement.
Matrices of Displacements In this paragraph we show how to describe the collineations of P3 (R) that preserve the absolute gure (3.2)
of the Euclidean space seen as Cayley-Klein space and correspond to
direct isometries. To do so, we look at the action of the Spin group on
a point P = (x0 , x1 , x2 , x3 )T that is represented by a grade-3 element
p = x0 e123 + x1 e234 + x2 e134 + x3 e124 .
An arbitrary element of the Spin group is given by

g = a0 e0 + a1 e23 + a2 e13 + a3 e12 + c0 e1234 + c1 e14 + c2 e24 + c3 e34 ,
gg = 1.
In
the projective representation the condition gg = 1 changes to
3
2
i=0 ai = 0 and further we have a0 c0 a1 c1 + a2 c2 a3 c3 = 0. The
action of a Spin group element can be written in terms of the sandwich
operator
gpg = (a20 + a21 + a22 + a23 )x0 e123

+ 2(a0 c1 a1 c0 a2 c3 a3 c2 )x0 + (a20 + a21 a22 a23 )x1

+ 2(a2 a1 a0 a3 )x2 + 2(a0 a2 + a3 a1 )x3 e234

+ 2(a0 c2 + a1 c3 a2 c0 a3 c1 )x0 + 2(a0 a3 + a2 a1 )x1

+ (a20 a21 + a22 a23 )x2 + 2(a3 a2 a0 a1 )x3 e134

+ (2a2 c1 + 2a1 c2 2a0 c3 2a3 c0 )x0 + 2(a3 a1 a0 a2 )x1

+ 2(a3 a2 + a0 a1 )x2 + (a20 a21 a22 + a23 )x3 e124 .
189
Here, we see that the action of g on p is linear. Hence, we can write

this action as a product of a matrix with a vector
x = M x, x = (x0 , x1 , x2 , x3 )T , x = (x0 , x1 , x2 , x3 )T
with
1
l
M=
m
n
0
a20 + a21 a22 a23
2(a0 a3 + a2 a1 )
2(a3 a1 a0 a2 )
0
2(a2 a1 a0 a3 )
a20 a21 + a22 a23
2(a3 a2 + a0 a1 )
0
2(a0 a2 + a3 a1 )
2(a3 a2 a0 a1 )
a20 a21 a22 + a23
and
= a20 + a21 + a22 + a23 ,
l = 2(a0 c1 a3 c2 a2 c3 a1 c0 ),
m = 2(a0 c2 + a1 c3 a2 c0 a3 c1 ),
n = 2(a2 c1 + a1 c2 a0 c3 a3 c0 ).
The factor 1 guarantees that the right lower 3 3 submatrix is an

element of SO(3).
Collineations in the Image Space Transformations in the underlying geometry induce collineations in the kinematic image space.
Therefore, we are interested in the matrix representation of Spin
group elements, see Eq. (1.34). Thus, we take an element from
Spin(3,0,1)
g = a0 e0 + a1 e23 + a2 e13 + a3 e12 + c0 e1234 + c1 e14 + c2 e24 + c3 e34 .
When
using homogeneous coordinates the norming condition becomes
3
2
i=0 ai = 0. Furthermore, the parameters (a0 , . . . , a3 , c0 , . . . , c3 ) have
to satisfy the condition (3.7). We get
ge0 = a0 e0 + a3 e12 + a2 e13 + c1 e14 + a1 e23 + c2 e24 + c3 e34 + c0 e1234 ,
ge23 = a1 e0 a2 e12 + a3 e13 c0 e14 + a0 e23 c3 e24 + c2 e34 + c1 e1234 ,
ge13 = a2 e0 + a1 e12 + a0 e13 c3 e14 a3 e23 + c0 e24 + c1 e34 c2 e1234 ,
ge12 = a3 e0 + a0 e12 a1 e13 c2 e14 + a2 e23 + c1 e24 c0 e34 + c3 e1234 ,
ge1234 = a1 e14 + a2 e24 a3 e34 + a0 e1234 ,
ge14 = a0 e14 a3 e24 a2 e34 + a1 e1234 ,
190
ge24 = a3 e14 + a0 e24 a1 e34 a2 e1234 ,

ge34 = a2 e14 + a1 e24 + a0 e34 + a3 e1234 .
If we write these equations in matrix form, we obtain a representation

of collineations
of P7 (R) that preserve the pencil of quadrics spanned
6 6
by S2 , N2 :
a0
a 1
a 2
+ a 3
G =
c0
c1
c
2
c3
a1
a0
a3
a2
c1
c0
c3
c2
a2
a3
a0
a1
c2
c3
c0
c1
a3
a2
a1
a0
c3
c2
c1
c0
0
0
0
0
a0
a1
a2
a3
0
0
0
0
a1
a0
a3
a2
0
0
0
0
a2
a3
a0
a1
0
0
0
.
a3
a2
a
1
a0
In order to write [G+ ] in the usual form, see Eq. (1.9) the isomorphism
(3.3) has to be used.
3.3.2 Blaschkes and Gr

unwalds Mapping via Cliord
Algebra
For planar Euclidean geometry the homogeneous Cliord algebra
model C(2,0,1) is adequate. We nd the Spin group in the even part
C+
(2,0,1) . A general element of the even part is given by
g = a0 e0 + a1 e12 + c0 e23 + c1 e13 .
The condition that this element is a Spin group element reads now
gg = a20 + a21 = 1.
In this case we have just one equation in the scalar part. If we change
to homogeneous coordinates, the condition gets a20 + a21 = 0. Again we
have the equation of a pair of complex conjugate planes intersecting
in a real line. This results in the same image space as in the previous
section, see Eq. (3.5).
191
Matrices of planar Euclidean Displacements With the Spin group

we can nd the matrix representation corresponding to the group
SE(2) as group of collineations xing the absolute gure, cf. Eq. (3.1).
All we have to do is to apply the sandwich operator to a point. As
we choose a plane based construction a point P = (x0 , x1 , x2 )T can be
described in the algebra C(2,0,1) via p = x0 e12 + x1 e23 + x2 e13 . A general
element of the Spin group is given by g = a0 e0 + a1 e12 + c0 e23 + c1 e13 .
Applying the sandwich operator to p results in

gpg = a20 + a21 x0 e12 + 2(a1 c0 + a0 c1 )x0 + (a20 a21 )x1 2a0 a1 x2 e23

+ 2(a1 c1 a0 c0 )x0 + 2a0 a1 x1 + (a20 + a21 )x2 e13 .
If we rewrite the action of this transformation as linear transformation

of the projective plane, we get
x0
a20 + a21
1
x1 =
2(a1 c0 + a0 c1 )
a20 + a21

x2
2(a1 c1 c0 a0 )
0
a20 a21
2a0 a1

0
x0
2a0 a1 x1 .
a20 a21
x2
(3.8)
In the rst line of Eq. (3.8) we see that the line at innity is xed under
these collineations. It is clear that all planar Euclidean displacements
are obtained in this way, because the Spin group is a double cover
of SE(2). The factor is added articially to guarantee that we just
look at matrices belonging to the Spin group. These two examples
show that the Spin group can be used to get the matrix group of the
direct isometries of a Cayley-Klein geometry, if it is representable as
homogeneous Cliord algebra model.
Collineations in the Image Space The Spin group elements induce
in the image space a group of transformations that x the pair of
complex conjugate planes, their intersection line and the two complex
conjugate points on this line. To get this group we have to look at the
matrix representation of the Spin group. Let g = a0 e0 + a1 e12 + c0 e23 +
c1 e13 Spin(2,0,1) . Calculating the matrix representation as described
in section 1.3.4 results in
a0 a1 0
0
a1 a 0
PGL(P2 (R), F) =
c1 c0 a0
c0 c1 a1
0

0
a0 , a1 , c 0 , c 1 R .
a1
a0
192
3.4 Kinematic mappings for other Cayley-Klein

Spaces
This procedure can be applied to homogeneous Cliord algebra models
of any signature. We will present one more example here. For the
remaining signatures, see Table 3.1 and Table 3.2. Note that it makes
no sense to look at the completely degenerate Cliord algebra C(0,0,3)
because the scalar product of any two vectors is zero.
3.4.1 Two-dimensional Cayley-Klein Spaces

Elliptic Plane The homogeneous model for the elliptic plane is given
by C(3,0,0) . Let g = a0 e0 + a1 e12 + c0 e23 + c1 e13 be an element of the
even part C+
(3,0,0) . The product of this element with its conjugate
element yields
gg = a20 + a21 + c20 + c21 = 1.
In homogeneous coordinates this condition gets a20 + a21 + c20 + c21 = 0.

Therefore, we can interpret the kinematic image of the congruences
in elliptic spaces as non-degenerate Cayley-Klein space with absolute
gure F : Q24,0 . Furthermore, this means, if we work over the real
numbers, every point has a pre-image. In Table 3.1 the absolute
Table 3.1: Planar Cayley-Klein geometries and belonging Cliord algebras
CK space
absolute Figure F
elliptic
Q13,0
hyperbolic
Q13,1
Euclidean
Q11,0 A1 Q02,0
pseudo-Euclidean
Q11,0 A1 Q02,1
quasi-elliptic
Q12,0 A0 Q1
1,0
1
0
quasi-hyperbolic
Q2,1 A Q1
1,0
totally isotropic Q11,0 A1 Q01,0 A0 Q1
1,0
Cliord alg.
C(3,0,0) , C(0,3,0)
C(2,1,0) , C(1,2,0)
C(2,0,1) , C(0,2,1)
C(1,1,1)
C(2,0,1) , C(0,2,1)
C(1,1,1)
C(1,0,2) , C(0,1,2)
3.4 Kinematic mappings for other Cayley-Klein Spaces
193
gures F and the corresponding Cliord algebras for possible twodimensional Cayley-Klein spaces are given. Note that the points of the
Cayley-Klein geometry are always written in the grade-2 subspace of
the Cliord algebra except the quasi-elliptic and the quasi-hyperbolic
case. In these two cases we use the Poincare isomorphism to describe
these Cayley-Klein geometries as dual partners of the Euclidean
respectively, the pseudo-Euclidean Cayley-Klein geometry. Therefore,
points of the quasi-elliptic (quasi-hyperbolic) Cayley-Klein geometry
are represented as grade-1 elements in the algebra corresponding to the
Euclidean (pseudo-Euclidean) Cayley-Klein space. Table 3.2 shows
the kinematic image spaces of the seven planar Cayley-Klein spaces.
Note that every image space is again a Cayley-Klein space.
Table 3.2: Kinematic image spaces of two-dimensional Cayley-Klein geometries presented as three-dimensional Cayley-Klein spaces
with their absolute gures F
pre-image
CK image space
absolute Figure F
elliptic
elliptic
Q24,0
hyperbolic
hyperbolic idx. 1
Q24,2
Euclidean
quasi-elliptic
Q22,0 A1 Q02,0
pseudo-Euclidean quasi-hyperb. idx. 0
Q22,1 A1 Q02,1
quasi-elliptic
quasi-elliptic
Q22,0 A1 Q02,0
quasi-hyperbolic quasi-hyperb. idx. 0
Q22,1 A1 Q02,1
totally isotropic
totally isotropic
Q21,0 A2 Q12,0 A0 Q1
1,0
3.4.2 Three-dimensional Cayley-Klein Spaces

The kinematic image space for the three-dimensional Euclidean space
was constructed in section 3.3.1. Here we want to repeat this construction for the elliptic three-dimensional Cayley-Klein space. In [59,
Ch. 11] the author showed that elements of the group SO(4) can be
identied with points of Studys quadric. In this case no exceptions
194
have to be made. Therefore, the kinematic mapping is one-to-one

and onto. Let us look what happens, if we construct the kinematic
mapping corresponding to the three-dimensional elliptic Cayley-Klein
geometry which is modelled through C(4,0,0) . This Cliord algebra is
of dimension 16 and the even part C+
(4,0,0) is eight-dimensional. An
arbitrary Spin group element is given by
g = a0 e0 + a1 e23 + a2 e13 + a3 e12 + c0 e1234 + c1 e14 + c2 e24 + c3 e34 .
The condition that it is an element of the Spin group reads

gg = (a20 +a21 +a22 +a23 +c20 +c21 +c22 +c23 )e0 +(a0 c0 a1 c1 +a2 c2 a3 c3 )e1234 = 1.
If we change to seven-dimensional projective space, the rst condition

is that the scalar part should not vanish
a20 + a21 + a22 + a23 + c20 + c21 + c22 + c23 = 0.
As in the case for Euclidean displacements this denes the exceptional

set, i.e., points that do not have a pre-image. Here, we can write this
exceptional set as hyperquadric
Q62 : a20 + a21 + a22 + a23 + c20 + c21 + c22 + c23 = 0.
The term in the pseudoscalar part is again the equation of Studys

quadric
S26 : a0 c0 a1 c1 + a2 c2 a3 c3 = 0.
Every point on Studys quadric stands for an element of SO(4) since

Spin(4,0,0) is a double cover of SO(4). In this case it is not necessary to
slice the quadric, since Q62 has no real
point.
Furthermore, this means
6
6
the pencil of quadrics spanned by S2 , Q2 is the absolute gure of
the kinematic image space. Surprisingly, every kinematic mapping of
three-dimensional Cayley-Klein spaces maps the Spin group elements
to points on Studys quadric. In the following we list the possible
Cayley-Klein spaces and the absolute quadric pencil in the kinematic
image space. In Table 3.3 the other possible homogeneous Cliord
algebra models are presented. Note that every Cayley-Klein space
is self-dual except the Euclidean and the pseudo-Euclidean. Their
dual Cayley-Klein spaces can be represented in the dual homogeneous
Cliord algebra model obtained by the Poincare duality.
3.4 Kinematic mappings for other Cayley-Klein Spaces
195
Table 3.3: Three-dimensional Cayley-Klein spaces with possible Cliord

algebra representation and exceptional quadric Q62
CK-space
elliptic
hyperbolic idx. 0
hyperbolic idx. 1
Euclidean
pseudo-Euclidean
quasi-elliptic
quasi-hyperb. idx. 0
double isotropic
C(p,q,r)
C(4,0,0) , C(0,4,0)
C(3,1,0) , C(1,3,0)
C(2,2,0)
C(3,0,1) , C(0,3,1)
C(2,1,1) , C(1,2,1)
C(2,0,2) , C(0,2,2)
C(1,1,2)
C(1,0,3) , C(0,1,3)
Q62
3
2
2
i=0 ai +
i=0 ci

3
3
2
2
i=0 ai
i=0 ci
a20 a21 a22 +a23 +c20 c21 c22 +c23
3
2
i=0 ai
a20 a21 a22 + a23
a20 + a23
a20 a23
a20
3
3.4.3 Higher dimensional kinematic Mappings

The procedure presented above can be generalized to arbitrary dimensions. Here, we present just one example, the four-dimensional
Euclidean space. Written as homogeneous Cliord algebra model we
get C(4,0,1) . The dimension of the algebra is 32, and therefore, the
dimension of the even part is 16. Now we take an arbitrary element
of the even part and formulate the conditions that it is element of
Spin(4,0,1) . Note that in this case two conditions are necessary, i.e.,
gg = g g = 1. The element has the form
g = a0 e0 + a1 e23 + a2 e13 + a3 e12 + a4 e15 + a5 e45 + a6 e25 + a7 e35
+ c1 e14 + c2 e24 + c3 e34 + c4 e1235 + c5 e1345 + c6 e1245 + c7 e2345 + c0 e1234 .
The product gg calculates to

gg = (c22 + c23 + a22 + a23 + a20 + c21 + a21 + c20 )e0
(3.9)
+ 2(c0 a0 a1 c1 c3 a3 + c2 a2 )e1234
+ 2(c7 c1 + c4 a0 c5 c2 c0 a5 + a6 a2 a1 a4 + c6 c3 a7 a3 )e1235
+ 2(c5 a1 + a6 c1 a5 a3 c2 a4 c7 a2 + c6 a0 + c0 a7 c4 c3 )e1245
+ 2(c4 c2 c6 a1 a5 a2 c3 a4 c0 a6 + c5 a0 + c7 a3 + a7 c1 )e1345
196

+ 2(c6 a2 + c0 a4 c4 c1 c5 a3 + a7 c2 c3 a6 + c7 a0 a5 a1 )e2345 .
Furthermore, we have to calculate g g

g g = (a21 + c22 + c23 + c20 + c21 + a23 + a22 + a20 )e0
(3.10)
+ 2(c2 a2 c3 a3 + c0 a0 a1 c1 )e1234
+ 2(c4 a0 a7 a3 c7 c1 + c0 a5 a1 a4 + a6 a2 c6 c3 + c5 c2 )e1235
+ 2(c7 a2 + a6 c1 c5 a1 a5 a3 + c6 a0 c2 a4 + c4 c3 c0 a7 )e1245
+ 2(c0 a6 c3 a4 c4 c2 + c6 a1 c7 a3 + a7 c1 + c5 a0 a5 a2 )e1345
+ 2(a7 c2 + c4 c1 a5 a1 + c5 a3 c6 a2 c3 a6 c0 a4 + c7 a0 )e2345 .
Vanishing coecients of the grade-4 generators in Eq. (3.9) and (3.10)

are the conditions for an even grade element to lay in the Spin group.
All coecients of grade-4 elements have to vanish and the scalar part
must be dierent from zero. Some quadric equations occur in both
expressions. All in all, the kinematic image space is a projective
variety in P15 (R) sliced along a quadric. It can be written as the
intersection of nine quadrics minus one quadric and therefore, as
pseudo algebraic variety:
V:
9
2
Qi \N1 P15 (R),
i=1
with
N1 : c22 + c23 + a22 + a23 + a20 + c21 + a21 + c20 = 0,
Q1 : c0 a0 a1 c1 c3 a3 + c2 a2 = 0,
Q2 : c7 c1 + c4 a0 c5 c2 c0 a5 + a6 a2 a1 a4 + c6 c3 a7 a3 = 0,
Q3 : c5 a1 + a6 c1 a5 a3 c2 a4 c7 a2 + c6 a0 + c0 a7 c4 c3 = 0,
Q4 : c4 c2 c6 a1 a5 a2 c3 a4 c0 a6 + c5 a0 + c7 a3 + a7 c1 = 0,
Q5 : c6 a2 + c0 a4 c4 c1 c5 a3 + a7 c2 c3 a6 + c7 a0 a5 a1 = 0,
Q6 : c4 a0 a7 a3 c7 c1 + c0 a5 a1 a4 + a6 a2 c6 c3 + c5 c2 = 0,
Q7 : c7 a2 + a6 c1 c5 a1 a5 a3 + c6 a0 c2 a4 + c4 c3 c0 a7 = 0,
Q8 : c0 a6 c3 a4 c4 c2 + c6 a1 c7 a3 + a7 c1 + c5 a0 a5 a2 = 0,
Q9 : a7 c2 + c4 c1 a5 a1 + c5 a3 c6 a2 c3 a6 c0 a4 + c7 a0 = 0.
3.5 Projective Varieties via kinematic Algebra Elements
197
Remark 3.3. With this method it is also possible to construct the

mapping SO(3) S 3 . In this case we do not need the homogeneous
Cliord algebra model. Therefore, we take the ane Cliord algebra
model C(3,0,0) and do the same construction for a Spin group element.
3.5 Projective Varieties via kinematic Algebra

Elements
Here, we present another method to construct projective varieties
corresponding to the Spin group via Cliord algebra. Therefore, a
denition is needed.
Denition 3.2. An element g of a Cliord algebra C(p,q,r) is called
kinematic, if it satises the following equation
g2 = tr(g)g N (g),
where tr(g) = g + g is the trace of the element.

Def. 3.2 generalizes the denition of kinematic algebras over elds,
see [39]. To obtain projective varieties by using this denition, we
rst show that every Spin group element is kinematic.
Lemma 3.1. Spin group elements g C(p,q,r) are kinematic, i.e.,
they full the equation
g2 = tr(g)g N (g).
Proof. The proof is done by direct calculation.

g2 = tr(g)g N (g) = (g + g )g gg = gg + g g gg = gg.
Here we have used, that gg = g g = 1 for every Spin group element.

Now we ask for general conditions that have to be satised by kinematic elements. We aim at a projective variety that corresponds to
the Spin group. Thus, we take an arbitrary element g C+
(p,q,r) and
calculate the conditions from
gg tr(g)g N (g) = 0.
198
This results in quadratic equations in several generators that all have

to vanish. Now we use these equations and check the condition gg = 0
to get more conditions that will guarantee that the element is in the
Spin group. To understand this method we present the kinematic
mapping for the group of four-dimensional Euclidean displacements,
i.e., Spin(4,0,1) .
Example 3.1. A general element g C+
(4,0,1) has the form
g = a0 e0 + a1 e23 + a2 e13 + a3 e12 + a4 e15 + a5 e45 + a6 e25 + a7 e35
+ c1 e14 + c2 e24 + c3 e24 + c4 e1235 + c5 e1345 + c6 e1245 + c7 e2345 + c0 e1234 .
For kinematic elements we have

0 = gg tr(g) N (g).
In expanded form this condition reads

0 = 4(c6 a2 + c0 a4 c4 c1 c5 a3 )e2345 4(c7 c1 c6 c3 + c5 c2 + c0 a5 )e1235
+ 4(c7 a2 + c0 a7 + c5 a1 c4 c3 )e1245 4(c0 a6 + c6 a1 c4 c2 c7 a3 )e1345 .
Now we use these four quadratic equations and formulate the Spin
group condition gg = 1,
gg = (a21 + c22 + c23 + c20 + c21 + a20 + a23 + a22 )e0
+ 2(a1 c1 c3 a3 + c0 a0 + c2 a2 )e1234
+ 2(c3 a6 + a7 c2 + c7 a0 a5 a1 )e2345
+ 2(c4 a0 a7 a3 + a6 a2 a1 a4 )e1235
+ 2(c6 a0 a5 a3 + a6 c1 c2 a4 )e1245
+ 2(a7 c1 a5 a2 c3 a4 + c5 a0 )e1345
and get ve more quadratic equations that have to vanish. All in all we
have the corresponding projective variety described by nine quadratic
equations and one exceptional quadric with equation
N1 : a20 + a21 + a22 + a23 + c20 + c21 + c22 + c23 = 0,
R1 : c6 a2 + c0 a4 c4 c1 c5 a3 = 0,
R2 : c7 c1 c6 c3 + c5 c2 + c0 a5 = 0,
3.5 Projective Varieties via kinematic Algebra Elements
199
R3 : c7 a2 + c0 a7 + c5 a1 c4 c3 = 0,
R4 : c0 a6 + c6 a1 c4 c2 c7 a3 = 0,
R5 : a1 c1 c3 a3 + c0 a0 + c2 a2 = 0,
R6 : c3 a6 + a7 c2 + c7 a0 a5 a1 = 0,
R7 : c4 a0 a7 a3 + a6 a2 a1 a4 = 0,
R8 : c6 a0 a5 a3 + a6 c1 c2 a4 = 0,
R9 : a7 c1 a5 a2 c3 a4 + c5 a0 = 0.
This method results in an ideal that describes the same projective

variety
V:
9
2
Ri \N1 P15 (R),
i=1
as in the previous section. Therefore, we give the Ri as linear combinations of the Qi

1
(Q9 Q5 ),
2
1
R4 = (Q8 Q4 ),
2
1
R7 = (Q6 + Q2 ),
2
R1 =
R2 =
1
(Q6 Q2 ),
2
R 5 = Q1 ,
R8 =
1
(Q3 + Q7 ),
2
1
(Q3 Q7 ),
2
1
R6 = (Q9 + Q5 ),
2
1
R9 = (Q8 + Q4 ).
2
R3 =
The advantage of both presented methods is that we can calculate

a Grobner basis for the ideal and apply the theory of ideals to the
constructed point models.
Conclusion
In chapter 1 we presented geometric models and representation that are
used today. Afterwards, Cliord algebras, the homogeneous Cliord
algebra model, and the conformal model were introduced. Moreover,
we presented a method to transfer general projective transformations
acting in three-dimensional projective space to elements of the homogeneous Cliord algebra model C(3,3,0) . All entities known from
line geometry occur naturally in this model and can be transformed
projectively by the application of the sandwich operator. It was
shown that the sandwich action of non-null vectors corresponds to
regular null polarities, i.e., correlations that are involutions as the
basic elements building up the group of regular projective transformations. The sandwich action of null vectors corresponds to singular null
polarities and we proved that every regular projective transformation
can be expressed as the product of six null polarities at the most.
Furthermore, we present a Cliord algebra model that serves as a
generalization of the conformal geometric algebra. A generalization
of inversions with respect to conics, quadrics, and even hyperquadrics
in any dimension is possible with the use of the sandwich operator.
Hyperquadrics in principal position are simply represented as grade-1
elements. Classical representations of groups are embedded in this
algebra naturally.
The basics of chain geometry were introduced in chapter 2. This theory was applied to Cliord algebras. We showed how the cross ratio
can be used to parametrize chains. Moreover, it was proven that the
connected components of the Pin- and Spin groups dene subspaces
of chain spaces. Therefore, every element contained in a chain that is
dened by three elements of the same connected component of the
Pin- or the Spin group is contained in the same connected component
of the Pin- or the Spin group. Of special interest was the question for
the cross ratio of dual unit quaternions that was answered in detail.
D. Klawitter, Clifford Algebras, DOI 10.1007/978-3-658-07618-4,
202
Conclusion
Moreover, we derived an algebraic biarc construction with contact

spaces. Chains of the grade-1 subspace that are in contact at a certain
point are parametrized with the cross ratio and correspond to conic
sections.
Old and well-known kinematic mappings were unied in chapter 3 in
one framework by the use of Cliord algebras. Collineations in any
kinematic image and the corresponding Cayley-Klein space can be
derived from the homogeneous Cliord algebra model. We presented
the Euclidean spaces of dimension two and three in detail as example.
Furthermore, a general method to construct pseudo algebraic varieties
in certain projective spaces was presented. These pseudo algebraic
varieties are point models for the Spin group of the homogeneous
Cliord algebra model respectively the collineation group of a certain
Cayley-Klein space. We performed the construction for the fourdimensional Euclidean. Moreover, the general construction enables
the examination of new kinematic image spaces for possible Pin and
Spin groups. This construction can also be done for Pin groups or
for any other subgroup of a Cliord group. Due to the fact that
projective varieties correspond to ideals, methods of Grobner basis
calculus can now be applied to kinematic image spaces.
Index
absolute gure, 182
admissible pairs, 103
algebra, 102
Cliord, 26
degenerate, 26
geometric, 26
Benz planes, 116
biarc, 165
Blaschke cylinder, 117
Cayley-Klein
geometry, 182
space, 182, 193
center, 102
chain, 106
geometry, 106
space, 107
Cliord algebra
conjugation, 29
even part, 28
main involution, 30
odd part, 28
universal, 28
Cliord group, 33
contact
relation, 121
space, 121
cross ratio, 109
distance
relation, 101
space, 101
dual
number, 8
orthogonal matrix, 13
quaternion, 10
Rodriguez formula, 14
sphere, 136
vector, 9
general linear group, 102
geometric inner product null
space, 44
geometric outer product null
space, 44
Grassmann
algebra, 39
coordinates, 39, 40
group
Spin group, 35
horosphere, 43
ideal, 199
inner product null space, 32
Jordan-closed, 119
Jordan-system, 118, 119

204
Index
kinematic algebra element, 197

kinematic mapping, 21, 185, 192
of Blaschke and Gr
unwald,
185
of Study, 185
Kleins quadric, 131
subalgebra, 118
Laguerre geometry, 109

Lies quadric, 73
Lorentz transformation, 48
versor, 28
Mobius geometry, 108

Minkowski geometry, 109
outer product null space, 31
parallel, 102
Pin group, 34
pitch, 20
Pl
ucker coordinates, 15
Poincare duality, 41
projective linear group, 104
projective variety, 199
pseudoscalar, 46
quadratic form, 114
quadric
Kleins quadric, 15, 50, 176
Lies quadric, 133
Studys quadric, 21, 76, 135
quaternion, 6
dual, 10
unit, 7
sandwich operator, 7
simple blade, 31
standard basis, 28
strong Jordan-system, 119
Studys sphere, 25
trace, 197
transitive
2--transitive, 104
3--transitive, 104
List of Symbols
(P, )
(l, m)
(p, q, r)

V
C+
(p,q,r)
C(p,q,r)
(J )

(, )
(C(p,q,r) )
(R)
,

[]m
[A+ ] , [B ]
i, j, k
N (R)
C(R)
F
G
distance space, p. 102

Pl
ucker coordinates with direction vector l and
moment vector m, p. 17
signature of a Cliord algebra, p. 28
main involution, p. 30
exterior algebra or Grassmann algebra, p. 39
bijection between quadrics in principal position
and vectors in QnGA, p. 82
even part of C(p,q,r) , p. 28
Cliord algebra of signature (p, q, r), p. 26
distance relation, p. 102
subgroup of PGL(R, 2), p. 120
dual unit vector representing an oriented line, p.
13
bilinear form corresponding to Lies quadric, p. 74
dual unit, p. 8
embedding of points in Q3GA, p. 79
element of PGL(R, 2), p. 109
Cliord group, p. 33
subgroup of PGL(R, 2), p. 105
extended Klein mapping, p. 21
morphism of distance spaces, p. 105
standard scalar product for dual vectors, p. 9
grade-m part of a Cliord algebra element, m N,
p. 31
matrix representations of geometric product, p. 36
quaternion units, p. 6
zero divisors of the ring R, p. 111
center of a ring R, p. 29
absolute gure of CK-space, p. 182
generator on Studys quadric, p. 150

206
J
K
M
N1
O
R
V
A
C
D
Dn
En
H
Hd
NI(A)
NIG (A)
NO(A)
NOG (A)
P1 (J )
P1 (R)
Pn (R)
Ud
A
a
C(L, R)
g
I
Q
Q
GL(R, 2)
I
PGL(R, 2)
Q
SE(3)
List of Symbols
Jordan-system, p. 119
eld, p. 102
module, p. 77
null vectors of a Cliord algebra, p. 128
dual orthogonal matrix, p. 13
set of units of a ring R , p. 33
projective variety, p. 199
split-complex numbers, p. 117
complex numbers, p. 117
ring of dual numbers, p. 8
n-dimensional module over D, p. 9
n-dimensional Euclidean space, p. 7
skew eld of quaternions, p. 6
dual unit quaternions, p. 10

inner product null space of a k-blade A k V , p.
32
geometric
inner product null space of a k-blade
k
A
V , p. 44

outer product null space of a k-blade A k V , p.
31
geometric
outer product null space of a k-blade
k
A
V , p. 44
projective line over the Jordan-system J , p. 120
projective line over the ring R, p. 103
n-dimensional real projective space, p. 5
group of dual unit quaternions, p. 10
blade in C(p,q,r) , p. 31
grade-1 elements resp. vectors in C(p,q,r) , p. 30
set of chains in P1 (R), p. 106
element of the Cliord group, p. 33
pseudoscalar, p. 46
projective quadric, p. 115
non-double points of Q, p. 116
general linear group over a ring R, p. 102
identity matrix, p. 13
projective linear group over P1 (R), p. 104
quadratic form, p. 26
group of spatial displacements, p. 21
List of Symbols
SO(3)
T(2)
(, )

Pin(p,q,r)
(K, R, J )
(L, R)
Spin(p,q,r)
An1
cabc
cr(a, b, c, d)
e + , e
ei
J
Ln+1
1
N (a)
N26
O(p, q, r)
PI
Qnr,q
S26
2
SD
tr()
V
V3
v = v +
v
Vkd
k
207
group of special orthogonal 3 3 matrices, p. 6
additive group of planar translations, p. 92
bilinear form corresponding to a quadric, p. 16
dual angle, p. 14
parallel relation, p. 102
Pin group, p. 34
chain geometry over (K, R, J ), p. 120
chain geometry over the L-algebra R, p. 106
Spin group, p. 35
map of Blaschke and Gr
unwald, p. 186
Studys kinematic mapping, p. 21
absolute n1 -subspace, p. 182
chain dened by a, b, c, p. 129
cross ratio of a, b, c and d, p. 1
generators of CGA, p. 43
generators of a Cliord algebra, p. 26
Poincare duality, p. 41
Lies quadric (ndimensional), p. 74
norm of a, p. 33
singular quadric in P7 (R), p. 185
orthogonal group, p. 34
image of identity e SE(3) under Study mapping,
p. 23
quadric of rank r, index q and dimension n, p. 182
Studys quadric, p. 21
Studys sphere, p. 25
trace of an algebra element, p. 197
vector space, p. 26
the exceptional generator, p. 21
small bold Latin letters denote dual vectors, with
real part v and dual part v, p. 9
algebraic variety of dimension d and degree k, p.
17
grade-k subspace of the exterior algebra corresponding to V , p. 28
Acknowledgment
Foremost, I would like to express my deepest thanks to my two supervisors, Professor Dr. Gunter Weiss and Prof. Dr. Marco Hamann.
Their patience, encouragement, and immense knowledge were key motivations throughout my PhD. They persuasively conveyed an interest
in my work, and I am grateful to my advisers.
Prof. Dr. Gunter Weiss has been my supervisor through three and a
half years of geometric research. I am truly thankful for his steadfast
integrity, and seless dedication to both my personal and academic
development. I cannot think of a better supervisor to have. Prof. Dr.
Marco Hamann is a mentor, from whom I have learnt the vital skill
of disciplined critical thinking. His forensic scrutiny of my work has
been invaluable. He has always found the time to propose consistently
excellent improvements. I owe a great debt of gratitude to Prof. Dr.
Gunter Weiss and Prof. Dr. Marco Hamann.
I would like to thank PD Dr. Boris Odehnal for oering thorough
and excellent feedback on an earlier version of this thesis. In addition,
a thank you to Prof. Dr. Brehm. He gave feedback on research,
and suggested general improvements to my models. Furthermore, I
thank Prof. Dr. Andrea Blunck and Prof. Dr. Hans Havlicek. Both
answered me questions on chain geometry during email correspondences. A special mention for Prof. Dr. Daniel Lordick, for our fruitful
discussions.
I thank my parents Manuela Klawitter and Burghard Klawitter for
the encouragement during the last years.
And last but not least, I thank my girlfriend Sarah Kunte. Her patience and her emotional support did always motivate me to go further.

210
Acknowledgment
This work was supported by the research project Line Geometry

for Lightweight Structures , funded by the DFG (German Research
Foundation) as part of the SPP 1542.
Bibliography
[1] Ablamowicz, R., and Fauser, B. Mathematics of CLIFFORD
- a Maple Package for Cliord and Grassmann Algebras. Advances
in Applied Cliord Algebras 15, 2 (2005), 157181.
[2] Ablamowicz, R., and Fauser, B. CLIFFORD/Bigebra, A
c
Maple Package for Cliord (Co)Algebra Computations: 19962011, RA&BF. Available at http://www.math.tntech.edu/rafal/
(2011).
[3] Ablamowicz, R., and Sobczyk, G. Lectures on Cliord
(geometric) algebras and applications. Birkh
auser, Boston, 2004.
[4] Beck, h. Uber

lineare Somenmannigfaltigkeiten. Mathematische
Annalen, 81 (1920), 187218.
ber Geometrie der Algebren. Springer,
[5] Benz, W. Vorlesungen u
Berlin, 1973.
[6] Blaschke, W. Euklidische Kinematik und nichteuklidische
Geometrie. Zeitschrift f
ur angewandte Mathematik und Physik,
60 (1911), 6191 und 203204.
[7] Blaschke, W. Kinematik und Quaternionen. Deutscher Verlag
der Wissenschaften, Berlin, 1960.
ber Dif[8] Blaschke, W., and Thomsen, G. Vorlesungen u
ferentialgeometrie und Geometrische Grundlagen von Einsteins
Relativit
atstheorie, Bd. 3: Dierentialgeometrie der Kreise und
Kugeln. Springer, Berlin, 1929.
[9] Blunck, A. A quadric model for Klingenberg chain spaces.
Geometriae Dedicata, 55 (1995), 237246.

212
Bibliography
ur
[10] Blunck, A. Chain spaces via Cliord algebras. Monatshefte f
Mathematik, 123 (1997), 93107.
[11] Blunck, A., and Herzer, A.
Einf
uhrung. Shaker, Aachen, 2005.
Kettengeometrien: Eine
[12] Blunck, A., and Stroppel, M. Klingenberg Chain Spaces.

Abhandlungen aus dem Mathematischen Seminar der Universit
at
Hamburg, 65 (1995), 225238.
[13] Bottema, O., and Roth, B. Theoretical kinematics. Dover
Publ, New York NY, 1990.
[14] Cecil, T. E. Lie Sphere Geometry, 2 ed. Springer, New York,
2008.
ak, S
., and Limoncu, M. Degenerate Spin
[15] Dereli, T., Koc
Groups as Semi-Direct Products. Advances in Applied Cliord
Algebras, 20 (2010), 565573.
[16] Doran, C., Hestenes, D., Sommen, F., and van Acker,
N. Lie groups as spin groups. J. Math. Phys. 34, 8 (1993),
36423669.
[17] Doran, C. J. L., and Lasenby, A. Geometric Algebra for
Physicists. Cambridge University Press, 2003.
[18] Dorst, L., and Fontijne, D. 3D Euclidean Geometry Through
Conformal Geometric Algebra (a GAViewer tutorial).
[19] Dorst, L., Fontijne, D., and Mann, S. Geometric Algebra
for Computer Science - An Object-oriented Approach to Geometry,
2 ed. Morgan Kaufmann, 978-0123749420.
[20] Fontijne, D. Ecient Implementation of Geometric Algebra.
PhD thesis.
[21] Gallier, J. Cliord Algebras, Cliord Groups and a Generalization of the Quaternions The Pin and Spin Groups.
[22] Garling, D. J. H. Cliord algebras: An Introduction. Cambridge University Press, Cambridge and New York, 2011.
Bibliography
213
[23] Gfrerrer, A. Studys Kinematic mapping - a tool for motion

design. Recent Advances in Robot Kinematics, Eds.: Lenarcic,
J., Stanisic, M.M., Kluwer Acad. Publ., Dordrecht - Boston London, 7-16 (2000).
[24] Gfrerrer, A. Studys Kinematic Mapping, unpublished manuscript. Institute of Geometry, Graz University of Technology,
Austria.
ber h
ohere Geometrie. Vieweg &
[25] Giering, O. Vorlesungen u
Sohn, Braunschweig, 1982.
nwald, J. Ein Abbildungsprinzip, welches die ebene Geo[26] Gru
metrie und Kinematik mit der r
aumlichen Geometrie verkn
upft.
Sitzungsberichte Akad. Wiss. Wien, 12 (1911), 677741.
[27] Gunn, C. Proceedings of AGACSE 2010: On the Homogeneous
Model of Euclidean Geometry. Springer, 2011.
[28] Gunn, C. Geometry, Kinematics, and Rigid Body Mechanics
in Cayley-Klein Geometries. PhD thesis, Technical University
Berlin, Berlin, December 9, 2011.
uttel
[29] Haack, W. Dierential-Geometrie Teil 1 und 2. Wolfenb
Verlagsanstalt G.m.b.H., Wolfenb
uttel and Hannover, 1948.
[30] Hamann, M. Line-symmetric motions with respect to reguli.
Mechanism and Machine Theory 46, 7 (2011), 960974.
ur Technische Mathematiker.
[31] Havlicek, H. Lineare Algebra f
Heldermann, Lemgo, 2006.
[32] Havlicek, H. From pentacyclic coordinates to chain geometries,
and back. Mitt. Math. Ges. Hamburg, 26 (2007), 7594.
[33] Herzer, A. Handbook of Incidence Geometry: Chain Geometries.
Elsevier Science, 1995.
[34] Hestenes, D. The Design of Linear Algebra and Geometry.
Acta Applicandae Mathematicae, 23 (1991).
214
Bibliography
[35] Hestenes, D. Old Wine in new Bottles - A new algebraic framework for computational geometry. E. Bayro-Corrochano & G.
Sobczyk (eds), Advances in Geometric Algebra with Applications
in Science and Engineering (2001), 114.
[36] Hestenes, D., and Sobczyk, G. Cliord Algebra to Geometric
Calculus: A unied Language for Mathematics and Physics. D.
Reidel, Dordrecht and Boston, 1984.
[37] Hotje, H. Zur Einbettung von Kettengeometrien in projektive
R
aume. Math. Zeitschr., 151 (1976), 517.
[38] Husty, M., Karger, A., Sachs, H., and Steinhilper, W.
Kinematik und Robotik. Springer, Berlin [u.a.], 1997.
[39] Karzel, H. Kinematische Algebren und ihre geometrischen
Ableitungen. Abhandlungen aus dem Mathematischen Seminar
der Universit
at Hamburg, 1 (1974), 158171.
[40] Klawitter, D. A Cliord algebraic Approach to Line Geometry.
Advances in Applied Cliord Algebras (2014).
[41] Klawitter, D., and Hagemann, M. Kinematic mappings
for Cayley-Klein geometries via Cliord algebras. Beitr
age zur
Algebra und Geometrie / Contributions to Algebra and Geometry
54, 2 (2013), 737761.
[42] Klein, F. Ueber Liniengeometrie und metrische Geometrie.
Mathematische Annalen, 5 (1872), 106126.
ber h
ohere Geo[43] Klein, F., and Blaschke, W. Vorlesungen u
metrie. Springer, Berlin, 1926.
[44] Kowol, G. Projektive Geometrie und Cayley-Klein Geometrien
der Ebene. Birkh
auser, Basel, 2009.
[45] Li, H., Hestenes, D., and Rockwood, A. A Universal Model
for Conformal Geometries of Euclidean, Spherical and DoubleHyperbolic Spaces. Geometric Computing with Cliord Algebras
(2001), 7799.
Bibliography
215
[46] Li, H., Hestenes, D., and Rockwood, A. Generalized Homogeneous Coordinates for Computational Geometry. Geometric
Computing with Cliord Algebras (2001), 2752.
[47] Li, H., and Zhang, L. Line geometry in terms of the null
geometric algebra over R(3, 3), and application to the inverse
singularity analysis of generalized stewart platforms. In Guide
to Geometric Algebra in Practice, L. Dorst and J. Lasenby, Eds.
Springer London, 2011, pp. 253272.
[48] Macdonald, A. A Survey of Geometric Algebra and Geometric
Calculus, 2008.
[49] McCarthy, J., and Ahlers, S. Dimensional synthesis of
robots using a double quaternion.
[50] McCarthy, J. M. An Introduction to Theoretical Kinematics.
MIT Press, Cambridge and Mass, 1990.
ller, E. Vorlesungen u
ber Darstellende Geometrie: Kon[51] Mu
struktive Behandlung der Regel
achen bearb. von Josef Leopold
Krames, vol. 3. Deuticke, Leipzig and Wien, 1931.
[52] Onishchik, A., and Sulanke, R. Projektive und CayleyKleinsche Geometrie. Springer, 2008.
[53] Perwass, C. B. U. Geometric Algebra with Applications in
Engineering. Springer, Berlin and Heidelberg, 2009.
[54] Pfurner, M. Analysis of spatial serial manipulators using
kinematic mapping. PhD thesis, University of Innsbruck, Austria,
Innsbruck.
[55] Porteous, I. R. Cliord Algebras and the Classical Groups.
Cambridge University Press, Cambridge, 2009.
[56] Pottmann, H., and Wallner, J. Computational Line Geometry. Springer, Berlin and New York, 2001.
[57] Ravani, B., and Sprott, K. Kinematic Generation of Ruled
Surfaces. Advances in Computational Mathematics, 17 (2002),
115133.
216
Bibliography
cker, H.-P., and Ju

ttler, B. The Simplest Truly Spa[58] Schro
tial Motion. Conference on Geometry: Theory and Applications,
Pilsen (June 29, 2009).
[59] Selig, J. M. Geometric Fundamentals of Robotics, 2nd ed.
Springer, New York, 2005.
[60] Struve, H., and Struve, R. Projective spaces with CayleyKlein metrics. Journal of Geometry 81, 1-2 (2004), 155167.
[61] Study, E. Geometrie der Dynamen. B.G. Teubner, 1903.
[62] Veldkamp, G. R. On the use of dual numbers, vectors and
matrices in instantanoues, spatial kinematics. Mech. Mach. Theory, 11 (1976), 141156.
[63] Walter, D. R., and Husty, M. On implicitization of Kinematic Constraint Equations.
[64] Wang, W., and Joe, B. Interpolation on quadric surfaces
with rational quadratic spline curves. Computer Aided Geometric
Design, 14 (1997), 207230.
[65] Weiss, G. Vorlesungen aus Liniengeometrie, unpublished manuscript. Institute of Geometry, Vienna University of Technology.
[66] Weiss, G. Zur euklidischen Liniengeometrie I-IV. Sitzungs
berichte der Osterreichischen
Akademie der Wissenschaften 19781982 .
[67] Wikipedia. Benz plane, http://en.wikipedia.org/wiki/benz planes.
Wikipedia, The Free Encyclopedia (April, 2014).
[68] Zamora-Esquivel, J. G(6,3) Geometric Algebra. 9th International Conference on Cliord Algebras and their Applications in
Mathematical Physics, Weimar (15-20 July 2011).

Clifford Algebras

Uploaded by

Copyright:

Available Formats

Clifford Algebras

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Clifford Algebras

Uploaded by

Copyright:

Available Formats

Clifford Algebras

Dr. Daniel Klawitter

Dissertation TU Dresden, 2014

ISBN 978-3-658-07618-4 (eBook)

The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie;

Their structures are those of a line with coordinates of a ring. (A

of six null polarities, i.e., skew-symmetric 4 4 matrices at the most.

A Cliord algebraic Approach to Line Geometry . . .

2.8.2 Lies Quadric . . . . . . . . . . . . . . . . . . . .

3 Kinematic Mappings for Spin Groups

experienced a renaissance through computational technology, see [56].

is real-valued, see [58]. However, there is no explanation in general.

general properties can be derived, too. Within the Cliord algebra

1 Models and Representations

1.1 Description of Displacements

D. Klawitter, Clifford Algebras, DOI 10.1007/978-3-658-07618-4_1,

1 Models and Representations

This allows us to describe Euclidean displacements with help of linear

The group of Euclidean displacements that can be represented with

1.1.2 Dual Quaternions

where i, j, and k are the quaternion units with

Denition 1.1. Let q be a quaternion as in Eq. (1.2). Then, a

1.1 Description of Displacements

is called quaternion conjugation. The norm of a quaternion is dened

Every quaternion q = 0 has an inverse quaternion that can be calculated by

Unit quaternions are quaternions with norm equal to 1. With respect

where d is an unit vector and R. Moreover, unit quaternions can

1 Models and Representations

The set of dual numbers

together with addition and multiplication forms a commutative ring

1.1 Description of Displacements

Denition 1.3. For a dual number z = a + b the dual number

Analytic functions can be extended to dual functions with help of

which is the dual extension of the analytic function.

A dual vector is the sum of its real- and dual part

We dene a standard scalar product on this module by

1 Models and Representations

Dual Quaternions Dual quaternions were introduced by E. Study,

where the conjugation is the quaternion conjugation

Therefore, a dual unit quaternion satises two relations in the components a0 , . . . , c3

Dual unit quaternions are denoted by Ud and form a group with

1.1 Description of Displacements

and a dual quaternion of the form p = 1 + (xi + yj + zk), representing

If we rewrite the result as s product of a matrix with a vector vector

This matrix vector product represents an Euclidean displacement, see

= cos sin l0 i sin l1 j sin l2 k

where is the rotation angle and v is the magnitude of a translation in

1 Models and Representations

The multiplication of two dual quaternions

can be written in matrix form. Therefore, we perform the multiplication as follows

If we rewrite AX as a product of a matrix and a vector, we get

where x = (x0 , , x7 )T and O is the 4 4 zero matrix. In the same

The corresponding product of a matrix and a vector has the form

1.1 Description of Displacements

where x = (x0 , , x7 )T and O is the 4 4 zero matrix, cf. [54].

1.1.3 Dual Orthogonal Matrices

where I denotes the 3 3 identity matrix. The displacement of a line

1 Models and Representations

where = + d is a dual number called the dual angle, I the 3 3

that maps dual vectors to dual skew-symmetric matrices. With the

The mapping : L3 P5 (R), where L3 is the set of lines of P3 (R),