San Jose State University

SJSU ScholarWorks

Master's Theses Master's Theses and Graduate Research

Fall 2013

Geometric Algebra: An Introduction with Applications in Euclidean

and Conformal Geometry
Richard Alan Miller
San Jose State University

Follow this and additional works at: https://scholarworks.sjsu.edu/etd_theses

Recommended Citation
Miller, Richard Alan, "Geometric Algebra: An Introduction with Applications in Euclidean and Conformal
Geometry" (2013). Master's Theses. 4396.
DOI: https://doi.org/10.31979/etd.chd4-za3f
https://scholarworks.sjsu.edu/etd_theses/4396

This Thesis is brought to you for free and open access by the Master's Theses and Graduate Research at SJSU
ScholarWorks. It has been accepted for inclusion in Master's Theses by an authorized administrator of SJSU
ScholarWorks. For more information, please contact [email protected].
 GEOMETRIC ALGEBRA:

AN INTRODUCTION WITH APPLICATIONS IN EUCLIDEAN

AND CONFORMAL GEOMETRY

A Thesis

Presented to

The Faculty of the Department of Mathematics and Statistics

San José State University

In Partial Fulfillment

of the Requirements for the Degree

Master of Science

by

Richard A. Miller

December 2013
 ⃝
c 2013

Richard A. Miller

ALL RIGHTS RESERVED

 The Designated Thesis Committee Approves the Thesis Titled

GEOMETRIC ALGEBRA:
AN INTRODUCTION WITH APPLICATIONS IN EUCLIDEAN
AND CONFORMAL GEOMETRY

by

Richard A. Miller

APPROVED FOR THE DEPARTMENT OF MATHEMATICS AND STATISTICS

SAN JOSÉ STATE UNIVERSITY

December 2013

Dr. Richard Pfiefer Department of Mathematics and Statistics

Dr. Richard Kubelka Department of Mathematics and Statistics

Dr. Wasin So Department of Mathematics and Statistics

 ABSTRACT

GEOMETRIC ALGEBRA:
AN INTRODUCTION WITH APPLICATIONS IN EUCLIDEAN
AND CONFORMAL GEOMETRY

by Richard A. Miller

This thesis presents an introduction to geometric algebra for the uninitiated.

It contains examples of how some of the more traditional topics of mathematics can

be reexpressed in terms of geometric algebra along with proofs of several important

theorems from geometry. We introduce the conformal model. This is a current topic

among researchers in geometric algebra as it is finding wide applications in computer

graphics and robotics. The appendices provide a list of some of the notational

conventions used in the literature, a reference list of formulas and identities used in

geometric algebra along with some of their derivations, and a glossary of terms.
 DEDICATION

This thesis is affectionately dedicated to my wife, Denise.

v
 ACKNOWLEDGEMENTS

I wish to thank my thesis advisor, Professor Richard Pfiefer, whose willingness

to step outside of his comfort zone and venture into uncharted waters with me has

been greatly appreciated. It has been a pleasure learning with you. A big thanks to

my thesis committee members, Professor Richard Kubelka and Professor Wasin So.

Dr. Kubelka has spent hours with me scrutinizing every minute detail of this

document, and for that I am eternally grateful. A special thanks to my Math 100W

instructor, Judy Hilliard, who tried to teach this engineer/mathmatician how to

write and graciously proofread this document. I would also like to thank all of the

SJSU Mathematics Department instructors whom I’ve had the pleasure of learning

from over the past few years. Thank you to my fellow grad students whose

camaraderie and study groups have helped ease the pain of graduate school. I will

miss you all! And finally, thanks to my wife Denise for allowing me to follow a

dream and putting up with the insecurity of no income while I was in school. I’m

going to get a job, Honey!

vi
 TABLE OF CONTENTS

CHAPTER

1 INTRODUCTION 1

2 A BRIEF INTRODUCTION TO GEOMETRIC ALGEBRA 6

2.1 Scalars, Vectors, and Vector Spaces . . . . . . . . . . . . . . . . . . . 6

2.2 Oriented Lengths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2.1 The Inner Product . . . . . . . . . . . . . . . . . . . . . . . . 11

2.3 Oriented Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.3.1 The Outer Product . . . . . . . . . . . . . . . . . . . . . . . . 18

2.4 Oriented Volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.5 Scalars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.6 The G3 Vector Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2.6.1 The Geometric Product . . . . . . . . . . . . . . . . . . . . . 29

2.6.2 Multiplicative Inverse of a Vector . . . . . . . . . . . . . . . . 30

2.6.3 Applying the Geometric Product to Orthonormal Vectors . . . 32

3 APPLICATIONS OF THE VECTOR SPACE MODEL 36

3.1 The Vector Space Model . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.2 Plane Trigonometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.2.1 Law of Cosines . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.2.2 Law of Sines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.2.3 Area of a Triangle . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.2.4 Sum of the Interior Angles of a Triangle . . . . . . . . . . . . 41

vii
 3.3 Matrix Representation . . . . . . . . . . . . . . . . . . . . . . . . . . 42

3.4 Generalized Complex Numbers . . . . . . . . . . . . . . . . . . . . . 44

3.5 The Quaternions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.6 Reflections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

3.7 Rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3.8 Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3.8.1 The Inverse of a Bivector . . . . . . . . . . . . . . . . . . . . . 59

3.8.2 Lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.8.3 Ceva’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3.8.4 Menelaus’ Theorem . . . . . . . . . . . . . . . . . . . . . . . . 72

3.8.5 Desargues’ Two-Triangle Theorem . . . . . . . . . . . . . . . . 75

4 THE CONFORMAL MODEL 80

4.1 Development of the Conformal Model . . . . . . . . . . . . . . . . . . 80

4.1.1 Add a Dimension for Stereographic Projection . . . . . . . . . 81

4.1.2 Add a Dimension for Homogeneous Coordinates . . . . . . . . 83

4.1.3 Change Variables from e, ē to n, n . . . . . . . . . . . . . . . 87

4.1.4 Generalization to Euclidean 3-Space . . . . . . . . . . . . . . . 90

4.1.5 Geometric Interpretation of n and n . . . . . . . . . . . . . . 91

4.2 Conformal Transformations . . . . . . . . . . . . . . . . . . . . . . . 92

4.2.1 Translation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

4.2.2 Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

4.2.3 Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

4.2.4 Dilation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

4.3 Some Background Needed for Representing Geometric Objects with

the Conformal Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

viii
 4.3.1 Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

4.3.2 Inner Product and Outer Product Null Spaces . . . . . . . . . 111

4.3.3 Linear Transformations . . . . . . . . . . . . . . . . . . . . . . 112

4.3.4 Outermorphisms . . . . . . . . . . . . . . . . . . . . . . . . . 112

4.3.5 Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

4.4 Representing Geometric Objects with the Conformal Model . . . . . 114

4.4.1 Points in the Conformal Model . . . . . . . . . . . . . . . . . 114

4.4.2 Inner Product of Two Points in the Conformal Model . . . . . 115

4.4.3 Lines in the Conformal Model . . . . . . . . . . . . . . . . . . 116

4.4.4 Circles in the Conformal Model . . . . . . . . . . . . . . . . . 118

4.4.5 Planes in the Conformal Model . . . . . . . . . . . . . . . . . 120

4.4.6 Spheres in the Conformal Model . . . . . . . . . . . . . . . . . 123

BIBLIOGRAPHY 126

APPENDIX

A NOTATIONAL CONVENTIONS USED IN THE LITERATURE 127

B FORMULAS AND IDENTITIES OF GEOMETRIC ALGEBRA 128

B.1 Axioms and Definitions of Geometric Algebra . . . . . . . . . . . . . 128

B.2 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

B.3 Useful Formulas and Identities . . . . . . . . . . . . . . . . . . . . . . 131

B.3.1 The Grade Operator . . . . . . . . . . . . . . . . . . . . . . . 132

B.3.2 Reversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

B.3.3 Norm (in Positive Definite Spaces) . . . . . . . . . . . . . . . 132

ix
 B.3.4 Inverse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

B.3.5 Geometric Product . . . . . . . . . . . . . . . . . . . . . . . . 134

B.3.6 Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

B.3.7 Outer Product . . . . . . . . . . . . . . . . . . . . . . . . . . 136

B.3.8 Orthogonal Test . . . . . . . . . . . . . . . . . . . . . . . . . . 137

B.3.9 Parallel Test (Subspace Membership Text) . . . . . . . . . . . 138

B.3.10 Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

B.3.11 Projection and Rejection . . . . . . . . . . . . . . . . . . . . . 138

B.3.12 Other Useful Identities . . . . . . . . . . . . . . . . . . . . . . 139

B.4 Derivations of Formulas . . . . . . . . . . . . . . . . . . . . . . . . . 139

B.4.1 |av| = |a| |v| . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

B.4.2 |u · v| ≤ |u| |v| (Cauchy-Schwarz inequality) . . . . . . . . . 139

B.4.3 |u + v| ≤ |u| + |v| (triangle inequality) . . . . . . . . . . . . 140

B.4.4 u−1 = u/|u|2 (vector inverse) . . . . . . . . . . . . . . . . . 141

B.4.5 u · v = 12 (uv + vu) (inner product) . . . . . . . . . . . . . . 141

B.4.6 u ∧ v = 12 (uv − vu) (outer product) . . . . . . . . . . . . . 142

B.4.7 u · (v ∧ w) = (u · v) w − (u · w) v . . . . . . . . . . . . . . . . 142

B.4.8 u · (v ∧ w) + v · (w ∧ u) + w · (u ∧ v) = 0 (Jacobi Identity) 144

C GLOSSARY 145

x
 LIST OF FIGURES

Figure

2.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Oriented length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.3 The sum of two oriented lengths. . . . . . . . . . . . . . . . . . . . . 11

2.4 u · v = |u| |v| cos θ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.5 An orthonormal basis. . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.6 An oriented area B in a plane. . . . . . . . . . . . . . . . . . . . . . . 14

2.7 Oriented area equivalency. . . . . . . . . . . . . . . . . . . . . . . . . 15

2.8 Equivalent oriented areas. . . . . . . . . . . . . . . . . . . . . . . . . 16

2.9 Scalar multiplication of oriented areas. . . . . . . . . . . . . . . . . . 17

2.10 Addition of oriented areas. . . . . . . . . . . . . . . . . . . . . . . . . 18

2.11 Outer product. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.12 Oriented area basis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.13 Visual aid for Eq. (2.6). . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.14 Oriented volumes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.1 Plane trigonometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.2 Right multiplying Z by the unit pseudoscalar. . . . . . . . . . . . . . 46

3.3 Multiplying Z by an exponential. . . . . . . . . . . . . . . . . . . . . 48

3.4 Right multiplying v by the unit pseudoscalar. . . . . . . . . . . . . . 50

3.5 Left multiplying v by the unit pseudoscalar. . . . . . . . . . . . . . . 51

3.6 Projection and rejection of u relative to v. . . . . . . . . . . . . . . . 53

xi
3.7 Reflection of u in v. . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.8 Rotation in the I(2) plane by angle θ. . . . . . . . . . . . . . . . . . . 56

3.9 Rotation in any plane by angle θ. . . . . . . . . . . . . . . . . . . . . 57

3.10 Rotation viewed as two successive reflections. . . . . . . . . . . . . . 58

3.11 Altitude of a triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3.12 Lemma 3.8.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.13 Lemma 3.8.8, using vectors and bivectors . . . . . . . . . . . . . . . . 64

3.14 Lemma 3.8.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

3.15 Lemma 3.8.9 proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

3.16 Ceva’s Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3.17 Proof of Ceva’s Theorem. . . . . . . . . . . . . . . . . . . . . . . . . 69

3.18 Converse of Ceva’s Theorem. . . . . . . . . . . . . . . . . . . . . . . . 71

3.19 Menelaus’ Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.20 Proof of Menelaus’ Theorem. . . . . . . . . . . . . . . . . . . . . . . . 73

3.21 Converse of Menelaus’ Theorem. . . . . . . . . . . . . . . . . . . . . . 75

3.22 Desargues’ Two-Triangle Theorem. . . . . . . . . . . . . . . . . . . . 76

3.23 Proof of Desargues’ Two-Triangle Theorem. . . . . . . . . . . . . . . 77

4.1 Line in Euclidean 1-space . . . . . . . . . . . . . . . . . . . . . . . . 81

4.2 Stereographic projection . . . . . . . . . . . . . . . . . . . . . . . . . 81

4.3 Addition of the ē dimension . . . . . . . . . . . . . . . . . . . . . . . 84

4.4 The conformal model of a line. . . . . . . . . . . . . . . . . . . . . . . 89

4.5 Translation using vector a . . . . . . . . . . . . . . . . . . . . . . . . 94

4.6 Inversion in the origin. . . . . . . . . . . . . . . . . . . . . . . . . . . 101

4.7 Dilation with respect to the origin. . . . . . . . . . . . . . . . . . . . 104

xii
 CHAPTER 1

INTRODUCTION

In 1966, David Hestenes, a theoretical physicist at Arizona State University,

published the book, Space-Time Algebra, a rewrite of his Ph.D. thesis [DL03,

p. 122]. Hestenes had realized that Dirac algebras and Pauli matrices could be

unified in a matrix-free form, which he presented in his book. This was the first

major step in developing a unified, coordinate-free, geometric algebra and calculus

for all of physics. His book applied the long-forgotten geometric product of William

Clifford to the study of classical physics.

Hestenes’ success with applying Clifford’s geometric product and his

continued publications have inspired a new generation of physicists and

mathematicians to re-examine geometric algebra. Acceptance of geometric algebra

is growing and, while not everyone is in full agreement, it is now hard to find any

area of physics to which geometric algebra cannot or has not been applied without

some degree of success [DL03, p. 124]. Some of the better known successful

applications include fields such as classical physics, space-time, relativity, quantum

theory, differential geometry, computer graphics, and robotics.

What makes this geometric algebra so flexible that it can be applied to so

many areas? It turns out that geometric algebra provides a generalized theory that

encompasses many of the mathematical topics that have been around for years, such

as complex numbers, quaternions, matrix algebra, vectors, tensor and spinor

algebras, and the algebra of differential forms. Geometric algebra provides a method

of expressing geometrical relationships through algebraic equations.

 2

What is geometric algebra? Geometric algebra extends the concept of a vector

as a one-dimensional segment of a line with direction, orientation, and magnitude.

Geometric algebra extends this concept to multiple dimensions. A bivector, for

example, is like a two-dimensional vector. It is a two-dimensional segment of a

plane with direction, orientation, and magnitude. A trivector is like a

three-dimensional vector. It is a three-dimensional segment of a space with

direction, orientation, and magnitude.

Geometric algebra traces its history back to the nineteenth century. During

this century mathematicians like Hamilton, Grassmann, Gibbs, and Clifford

struggled with and developed the mathematics of complex numbers, quaternions,

and vectors [DL03]. A common thread to their research was to find the answer to

the question, “Just what does it mean to take the product of two vectors?”

William Rowan Hamilton (1805–1865) was an Irish mathematician and the

inventor of quaternions. He worked for years trying to extend his theory of complex

numbers to three dimensions [DL03, p. 8-10]. On October 16th , 1843, while out

walking with his wife along the Royal Canal in Dublin, he had an epiphany and

carved the equation i2 = j2 = k2 = ijk = −1 into the Broome Bridge. Although

quaternions found favor with physicists like James Clerk Maxwell, they were

weighed down by the increasingly dogmatic arguments over their interpretation and

were eventually displaced by the hybrid system of vector algebra promoted by

Gibbs.

Hermann Gunther Grassmann (1809–1877) was a German high school

mathematics teacher [DL03, p. 12,13,17]. He introduced the outer or exterior

product which, in modern notation, we write as a ∧ b, or “a wedge b.” The outer

product provided a means of encoding a plane, without relying on the notion of a

vector perpendicular to it. It was a new mathematical entity for encoding an

 3

oriented plane and today we call it a bivector. Grassmann published the first edition

of his geometric calculus, Lineale Ausdehnungslehre, in 1844, the same year

Hamilton announced his discovery of the quaternions. Grassmann’s work did not

achieve the same impact as the quaternions; however, and it was many years before

his ideas were understood and appreciated by other mathematicians. Perhaps he

was not taken seriously by his contemporaries because he was a high school teacher.

Josiah Willard Gibbs (1839–1903) was an American scientist/mathematician

who is credited with the invention of modern vector calculus [Kle08], [DL03,

p. 10,11,17]. He spent the majority of his career at Yale, where he was professor of

mathematical physics from 1871 until his death. Gibbs worked on developing

Grassmann’s exterior algebra from 1880 to 1884. He produced a vector calculus

well-suited to the needs of physicists. He distinguished the difference between the

dot and cross products of two vectors. Gibbs’s lecture notes on vector calculus were

privately printed in 1881 and 1884 for the use of his students and were later adapted

by Edwin Bidwell Wilson into a textbook, Vector Analysis, published in 1901. The

success of Gibbs’ vector calculus is evident, as today it is a major part of every

science, technology, engineering, and mathematics (STEM) curriculum.

William Kingdon Clifford (1845–1879) was a British mathematician and

professor of Applied Mathematics at University College, London [DL03, p.20,21].

Clifford appears to have been one of the small number of mathematicians at the

time to be significantly influenced by Grassmann’s work. Clifford introduced his

geometric algebra by uniting the inner and outer product into a single geometric

product. This product is associative like Grassmann’s product, but has the crucial

extra feature of being invertible. Although his work is often referred to as Clifford

Algebra, we use Clifford’s original choice—geometric algebra.

David Hestenes (1933–Present) is the chief architect of geometric algebra as

 4

we know it today [DL03, p. 122-124]. Throughout much of his career, he has worked

to develop geometric algebra as a unified language for mathematics and physics.

Hestenes, now retired, is the prime mover behind the contemporary resurgence of

interest in geometric algebra and other offshoots of Clifford algebras as ways of

formalizing theoretical physics. In 1984, he published Clifford Algebra to Geometric

Calculus. This book, coauthored by Garret Sobczyk and subtitled A Unified

Language for Mathematics and Physics, develops a complete system of Geometric

Calculus. In 1986 Hestenes published New Foundations for Classical Mechanics.

This book presents many of the topics covered in a classical undergraduate physics

class using the language of geometric calculus.

Chapter 2 of this thesis presents an introduction to geometric algebra for the

uninitiated. Chapter 3 contains examples of how some of the more traditional topics

of mathematics can be re-expressed in terms of geometric algebra and gives proofs

of several important theorems from geometry. Chapter 4 introduces the conformal

model. This is a current topic among researchers in geometric algebra as it is

finding wide applications in computer graphics and robotics. Because

geometric algebra is a relatively new topic, authors of the literature have not yet

standardized their notation. Appendix A provides a list of some of the notational

conventions used in the literature. Appendix B provides a reference list of formulas

and identities used in geometric algebra, along with some of their derivations.

Appendix C provides a glossary of terms.

Geometric algebra is really all about how we multiply vectors, i.e., about the

geometric product and how we interpret that product. Hestenes once wrote that,

“Physicists had not learned properly how to multiply vectors and,

as a result of attempts to overcome this, have evolved a variety of
mathematical systems and notations that has come to resemble
Babel.” [GLD93]
 5

It is our belief that geometric algebra is here to stay and may someday be a part of

every mainstream mathematics and physics program.

 6

CHAPTER 2

A BRIEF INTRODUCTION TO GEOMETRIC ALGEBRA

This chapter is intended to provide a brief introduction to geometric algebra

for the uninitiated. No proofs are provided. This material is intentionally presented

with a lack of rigor in hopes of bringing the reader up-to-speed as quickly as

possible on the objects, operations, and properties of geometric algebra. We limit

our discussion of geometric algebra to two and three dimensions for simplicity, but

part of the power and beauty of geometric algebra is that it extends to any

dimension. A primary motivation for geometric algebra is to unify all of modern

physics under one umbrella. For this reason, the theory has been well-developed for

higher dimensions. Much of the material and terminology used in this chapter

follows Macdonald [Mac10].

2.1 Scalars, Vectors, and Vector Spaces

When students begin their study of physics, they learn that there are two

types of physical quantities: scalar and vector. Those quantities that have only

magnitude and no direction, like temperature and volume, are called scalar

quantities. Quantities that have both magnitude and direction, like force and

velocity, are called vector quantities. This thesis uses lower case letters a, b, c to

represent scalars and lower case bold letters u, v, w to represent vectors.

Introductory physics classes emphasize the visualization of a vector quantity,

such as force or velocity, as a line segment with an arrowhead at the end. The
 7

length of the line segment shows the magnitude of the vector and the arrowhead

indicates the direction (see Figure 2.1a). This visualization of vectors as oriented

lengths is used to introduce the concepts of vector addition and scalar multiplication

(see Figures 2.1b and 2.1c).

w v
v 2v -v
v+w
.5v -.5v
(a) Examples (b) Vector Addition (c) Scalar Multiplication

Figure 2.1: Vectors

In upper-division math courses we are introduced to the more abstract

concept of a vector space. We learn that any set on which vector addition and scalar

multiplication operations are defined is a vector space, provided that certain

properties hold. This concept of a vector space is formally defined in

Definition 2.1.1.

Definition 2.1.1 (Vector space).

A vector space V is a set of objects called vectors. There are two operations

defined on V called vector addition and scalar multiplication. There also exists

a zero vector 0. Axioms V0–V7 must be satisfied for all vectors u, v, and w and all

scalars a and b ∈ R (in this thesis, we limit scalars to the field of real numbers):
 8

V0: v + w ∈ V and av ∈ V, closure

V1: u + v = v + u, commutativity

V2: (u + v) + w = u + (v + w), associativity

V3: v + 0 = v, additive identity

V4: ∀ v ∈ V, ∃ w ∈ V s.t. v + w = 0, additive inverse

V5: 1v = v, multiplicative identity

V6: (ab)v = a(bv), associativity

V7: a(u + v) = au + av distributive properties

(a + b)u = au + bu.

Along with this definition of a vector space comes the realization that there

are many different mathematical objects, other than oriented lengths, that qualify

as vectors. Take, for example, the set of all polynomials of order less than or equal

to n, where n is some positive integer. This set is a vector space. The abstraction of

the concept of a vector makes visualization much more difficult. But at the same

time, it vastly increases the range of applications to which the theory can be applied.

Geometric algebra introduces several new vector spaces that have important

significance to geometry. In addition, the direct sum of these vector spaces is also

very important in geometric algebra. The following sections describe these vector

spaces and introduce two new operations, the outer product and the geometric

product.

2.2 Oriented Lengths

For the next few sections, we refer to the common vectors of

Euclidean 3-space as oriented lengths.

 9

Definition 2.2.1 (Oriented length).

An oriented length v is an oriented segment of a line. The line segment has

an initial point and an end point with an arrowhead (see Figure 2.2). The length

of v is called its norm, denoted by |v|.

end point

v
initial point

Figure 2.2: Oriented length

Two oriented lengths are said to be equivalent if they have the same attitude,

orientation, and norm. These three properties of an oriented length are important

for future discussion, so we formally define them here.

Definition 2.2.2 (Properties of oriented lengths1 ).

attitude - Two oriented lengths have the same attitude if they lie on the same line

or on parallel lines.

orientation - Two oriented lengths have the same orientation if they have the same

attitude and their arrows point in the same direction. Thus, v and −v have

opposite orientations.
1
The terms attitude, orientation, and norm for oriented lengths are defined in some detail in
Dorst, Fontijne, and Mann [DFM07, p. 26].
 10

norm - Two oriented lengths have the same norm if they have the same length.

For example, if v is an oriented length, then −v is an oriented length with the

same attitude and norm, but with opposite orientation.

We now define scalar multiplication, vector addition, and zero for

oriented lengths.

Definition 2.2.3.

Scalar multiplication: The scalar multiple au is an oriented length parallel to u

with norm |a| |u|. If a > 0 then au has the same orientation as u. If a < 0

then au has the opposite orientation as u. If a = 0 then au = 0, the zero

oriented length (see below).

Vector addition: To determine the sum of two oriented lengths v and w, we move

the oriented length w parallel to itself so that its initial point coincides with

the end point of the oriented length v (see Figure 2.3). The sum v + w then

equals the oriented length whose initial point equals the initial point of v and

whose end point equals the end point of the moved oriented length w.

Zero: It is useful to define the zero oriented length, denoted by 0, with the

following properties:

a0 = 0, for all scalars a,

0u = 0, for all oriented lengths u, (2.1)

u + 0 = u, for all oriented lengths u.

Note that attitude and orientation are not defined for 0; the norm of 0 is 0.
 11

w
v
v
0
w v+w

Figure 2.3: The sum of two oriented lengths.

2.2.1 The Inner Product

An important operation defined on two oriented lengths is the inner product.

Definition 2.2.4 (Inner product).

The inner product of the nonzero oriented lengths u and v is the scalar

defined by

u · v = |u| |v| cos θ, 0≤θ≤π , (2.2)

where θ is the angle between the oriented lengths when one is moved parallel to itself

so that the initial points of the two oriented lengths coincide (see Figure 2.4). We

also define v · 0 = 0 · v = 0 for all v ∈ V. Note that v · v = |v|2 .

v
θ u
Figure 2.4: u · v = |u| |v| cos θ.

An inner product of oriented lengths has the following properties:

 12

Theorem 2.2.5 (Inner product properties).

Let u, v, and w be oriented lengths and let a be a scalar. Then

I1: u · v = v · u, commutativity

I2: (au) · v = a(u · v), homogeneity

u · (av) = a(u · v),

I3: (u + v) · w = u · w + v · w, distributive

w · (u + v) = w · u + w · v,

I4: If v ̸= 0, then v · v > 0. positivity

A vector space, along with an inner product, is called an inner-product space.

Now that we have the inner product, we can use it to test if two oriented lengths are

orthogonal.

Theorem 2.2.6 (Orthogonal vectors test).

Vectors u and v are orthogonal if and only if u · v = 0. Note that the zero

oriented length 0 is orthogonal to every vector.

Let e1 , e2 , and e3 be any three unit oriented lengths in Euclidean 3-space

such that 


0 if i ̸= j;
ei · ej =


1 if i = j.

then {e1 , e2 , e3 } forms an orthonormal basis for the vector space of oriented lengths

in Euclidean 3-space (see Figure 2.5). The dimension of any finite-dimensional

vector space is defined to be the the number of elements in any basis for that vector

space. Thus, Euclidean 3-space has dimension three.

With this basis, any oriented length v in Euclidean 3-space can be expressed
 13

e3

e2
e1

Figure 2.5: An orthonormal basis.

as a unique linear combination of e1 , e2 , and e3 .

v = ae1 + be2 + ce3

Oriented lengths in R3 satisfy axioms V0–V7 of Definition 2.1.1 and thus form a

vector space of dimension three.

2.3 Oriented Areas

Now that we have a clear understanding of what an oriented length is, we

would like to turn our attention to the vector space of oriented areas.

Definition 2.3.1 (Oriented area).

An oriented area B is an oriented segment or region of a plane. These

“segments” or “regions” of a plane must have well-defined, finite areas. The area of

B is called its norm, denoted by |B|. These “segments” or “regions” of a plane

must also have a well-defined orientation (usually denoted as clockwise or

counterclockwise).
 14

We use the uppercase bold letter B to represent oriented areas (see

Figure 2.6). Just like oriented lengths, two oriented areas B1 and B2 are considered

Figure 2.6: An oriented area B in a plane.

equivalent (B1 ∼ B2 ) if they have the same attitude, orientation, and norm. The

following definition captures these ideas.

Definition 2.3.2 (Properties of oriented areas2 ).

attitude - Two oriented areas have the same attitude if they lie on the same plane

or on parallel planes.

orientation - Two oriented areas have the same orientation if they have the same

attitude and the same rotational direction, clockwise or counterclockwise, as

we traverse the perimeter of the region.

norm - Two oriented areas have the same norm if they have the same area.

See Figure 2.7.

2
The terms attitude, orientation, and norm for oriented areas are defined in some detail in Dorst,
Fontijne, and Mann [DFM07, p. 27–28].
 15

B1
B

B1
B

B2
B

B2
B

(a) B1 ∼ B2 (b) B1  B2
Equivalent. Different Attitudes.

B2 B2
B1 B B
B1

(c) B1  B2 (d) B1 ∼ −B2

Different norms. Different Orientations.

Figure 2.7: Oriented area equivalency.

An oriented area B1 of any shape can be shown to be equivalent to an

oriented rectangle in the same plane. Suppose we have two perpendicular

oriented lengths u and v in the same plane as B1 . Suppose also that the rectangle,

B2 , formed by u and v, has the same area as B1 , i.e.,

|u| |v| = |B1 |.

We assign an orientation to B2 to be the same as that of B1 , thus the two

oriented areas are equivalent (see Figure 2.8).

 16

B1 B2 v
u
Figure 2.8: Equivalent oriented areas.

We now define scalar multiplication, vector addition, and zero for

oriented areas.

Definition 2.3.3.

Scalar multiplication: The scalar multiple aB is an oriented area in the same

plane as B or in a plane parallel to B with norm |a| |B|. If a > 0 then aB

has the same orientation as B. If a < 0 then aB has the opposite

orientation. If a = 0 then aB = 0, the zero oriented area (see Figure 2.9).

Vector addition: There are two cases.

(1) Suppose that the oriented areas B1 and B2 are in the same or parallel

planes. Let B be a unit oriented area—i.e., an oriented area with norm

of one—parallel to B1 and B2 . Then B1 ∼ b1 B and B2 ∼ b2 B, where

b1 and b2 are scalars with |bi | = |Bi |. Define B1 + B2 = (b1 + b2 )B.

(2) Suppose that the oriented areas B1 and B2 lie in nonparallel planes. In

Euclidean 3-space, two nonparallel planes must intersect in a line l (see

Figure 2.10).
 17

(a) Choose an oriented length w on the line of intersection of the two

planes containing B1 and B2 .

(b) Construct an oriented length u in the B1 plane such that when the

initial point of u is placed on the initial point of w, u indicates

the orientation of B1 , u ⊥ w, and |u| |w| = |B1 |.

(c) Likewise, construct an oriented length v in the B2 plane such that

when the initial point of v is placed on the initial point of w, v

indicates the orientation of B2 , v ⊥ w, and |v| |w| = |B2 |.

(d) Define B1 + B2 to be the oriented area formed from a rectangle

3
with sides u + v and w and with orientation given by u + v.

Zero: The zero oriented area, denoted by 0, has the following properties:
a0 = 0, for all scalars a,

0B = 0, for all oriented areas B,

B + 0 = B, for all oriented areas B.

As with the zero oriented length, attitude and orientation are not defined for

the zero oriented area; the norm of 0 is 0.

2B B
.5B

-2B
-.5B
Figure 2.9: Scalar multiplication of oriented areas.

3
Note that oriented area addition is similar to finding the diagonal of a parallelogram, but in
this case we are finding the diagonal plane of a parallelepiped.
 18

B1+B2

B1

u
u+v
w B2

Figure 2.10: Addition of oriented areas.

2.3.1 The Outer Product

As previously stated, two oriented lengths can be used to construct an

oriented area. We now formalize this concept by defining the outer product of two

oriented lengths. This new operation is a key concept of geometric algebra.

If a point is moved a distance and direction specified by a non-zero

oriented length u, it sweeps out a directed line segment. If this line segment is then

moved a distance and direction specified by a second non-zero, non-parallel

oriented length v, it sweeps out a parallelogram (see Figure 2.11a). Because the

oriented area corresponding to this parallelogram is uniquely determined by this

geometric construction, it may be regarded as a kind of “product” of the

oriented lengths u and v.4 If either of the oriented lengths happens to be zero, or if

they are parallel, then no parallelogram is constructed. In this case we define the

outer product to be the zero oriented area.

4
This description of the outer product comes from Hestenes [Hes03, p. 22–23].
 19

u
v
u^ v v v^ u
u

(a) u ∧ v. (b) v ∧ u.

Figure 2.11: Outer product.

Definition 2.3.4 (Outer product).

The outer product of two non-zero, non-parallel oriented lengths u and v,

denoted by u ∧ v, is an oriented area in the plane of u and v formed by sweeping the

oriented length u along the length of the oriented length v. Its norm is the area of

the parallelogram with edges u and v. Its orientation can be determined by sliding

the second oriented length v parallel to itself until its initial point touches the end

point of the first oriented length u. The orientation is then visualized as traversing

the perimeter of the parallelogram in the direction indicated by arrows on u and v.

If either of the oriented lengths happens to be zero, or if they are parallel, then we

define the outer product u ∧ v to be the zero oriented area. A “wedge” is used to

denote this new kind of multiplication to distinguish it from the “dot” denoting the

inner product.

Note that the orientation of v ∧ u is opposite that of u ∧ v. Thus,

v ∧ u = −(u ∧ v) (see Figure 2.11b). As long as neither oriented length is zero and

they are not parallel, the oriented area generated by an outer product has the shape
 20

of a parallelogram. It is important to remember that oriented areas, in general, have

no shape. It is the area that is important, not the shape.

Theorem 2.3.5 (Properties of the outer product).

O1: u ∧ u = 0, parallel

O2: u ∧ v = −(v ∧ u), antisymmetry

O3: (au) ∧ v = a(u ∧ v), homogeneity

u ∧ (av) = a(u ∧ v),

O4: (u + v) ∧ w = u ∧ w + v ∧ w, distributive

w ∧ (u + v) = w ∧ u + w ∧ v.

Definition 2.3.6 (Unit oriented area).

Let {b1 , b2 } be an orthonormal basis for a plane that contains the origin, i.e.,

of a two-dimensional homogeneous subspace of Rn . Because both b1 and b2 are unit

oriented lengths, b1 ∧ b2 forms a 1-by-1 square with area (norm) of one. This is

referred to as a unit oriented area.

Theorem 2.3.7.

Let {b1 , b2 } be an orthonormal basis for a homogeneous plane. Orient the

plane with b1 ∧ b2 . Let u and v be vectors in the plane. Let θ be the oriented angle

from u to v. Then

u ∧ v = |u| |v| sin θ (b1 ∧ b2 ), −π < θ ≤ π . (2.3)

We can express any oriented length in R3 as a linear combination of the

standard basis {e1 , e2 , e3 }. Let u = u1 e1 + u2 e2 + u3 e3 and v = v1 e1 + v2 e2 + v3 e3

be two oriented lengths in R3 , where ui and vi are scalars. The outer product u ∧ v
 21

can be calculated as:

u ∧ v = u1 v1 (e1 ∧ e1 ) + u2 v1 (e2 ∧ e1 ) + u3 v1 (e3 ∧ e1 ) (2.4)

+u1 v2 (e1 ∧ e2 ) + u2 v2 (e2 ∧ e2 ) + u3 v2 (e3 ∧ e2 )

+u1 v3 (e1 ∧ e3 ) + u2 v3 (e2 ∧ e3 ) + u3 v3 (e3 ∧ e3 )

= (u2 v3 − u3 v2 ) e2 ∧ e3 + (u3 v1 − u1 v3 ) e3 ∧ e1 + (u1 v2 − u2 v1 ) e1 ∧ e2 .

The following equation is useful for remembering the calculation of Eq. (2.4).

e2 ∧ e3 e3 ∧ e1 e1 ∧ e2
u∧v = u1 u2 u3 . (2.5)

v1 v2 v3

This calculation is very similar to that of the cross product u × v, only the cross

product results in a vector perpendicular to the plane of u and v, whereas the outer

product results in an oriented area in the plane of u and v.

Theorem 2.3.8.

The oriented areas {e2 ∧ e3 , e3 ∧ e1 , e1 ∧ e2 } form a basis for the

oriented areas in Euclidean 3-space (see Figure 2.12).

e3

e 2 ^e 3

e 3 ^e 1
e2

e 1 ^e 2

e1

Figure 2.12: Oriented area basis.

 22

Proof. Figure 2.8 shows that every oriented area is an outer product of two vectors.

Then Eq. (2.4) shows that the three oriented areas span the space of bivectors. Now

consider

a1 e2 ∧ e3 + a2 e3 ∧ e1 + a3 e1 ∧ e2 = e2 ∧ (a1 e3 − a3 e1 ) + a2 e3 ∧ e1 . (2.6)

The term on the right of the equal sign in Eq. (2.6) is the sum of two oriented areas.

Figure 2.13 can be helpful in visualizing this when ai = 1. According to Figure 2.10,

the sum of two nonzero oriented areas in different planes is nonzero. Therefore, the

sum of the terms in Eq. (2.6) is zero only when a1 = a2 = a3 = 0. Thus, the three

oriented areas are linearly independent.

e3-e1

e3 e2^(e3-e1)

e 3 ^e 1

e2

e1

Figure 2.13: Visual aid for Eq. (2.6).

Oriented areas satisfy axioms V0–V7 of Definition 2.1.1 and thus form a

vector space of dimension three.

 23

2.4 Oriented Volumes

In addition to oriented lengths and oriented areas, we now add a third vector

space, that of oriented volumes.

Definition 2.4.1 (Oriented volume).

An oriented volume T is an oriented segment of (i.e., solid in) three

dimensional space. The volume of T is called its norm, denoted by |T|. We use the

uppercase bold letter T to represent an oriented volume.

The theory of the outer product, as described so far, calls for a

generalization5 . Just as a plane segment is swept out by a moving line segment, a

“space segment” is swept out by a moving plane segment. Thus, the points on an

oriented parallelogram, specified by the oriented area u ∧ v moving a distance and

direction specified by an oriented length w (not in the u ∧ v plane,) sweep out an

oriented parallelepiped (see Figure 2.14a), which defines a new kind of directed

volume T called an oriented volume. In this way we define the outer product of the

oriented area u ∧ v with the oriented length w. Thus, we write

T = (u ∧ v) ∧ w.

w
v u
u
u^ v v^ w w
(a) (u ∧ v) ∧ w. (b) u ∧ (v ∧ w).

Figure 2.14: Oriented volumes.

5
This description comes from Hestenes [Hes03, p. 26–28].
 24

Oriented volumes contain no information about the shape of the solid. Two

oriented volumes are equivalent if they have the same attitude, orientation, and

norm.

Definition 2.4.2 (Properties of oriented volumes6 ).

attitude - In Euclidean 3-space, there is only one attitude for all oriented volumes.

In dimensions greater than three, the attitude represents the

three-dimensional subspace that contains the oriented volume.

orientation - The orientation of an oriented volume is usually referred to as

handedness. For an oriented volume formed by the outer product

(u ∧ v) ∧ w, we use the right-hand rule: we curl the fingers of our right hand

around u ∧ v consistent with its orientation. If the third oriented length w

is in the direction of our thumb, then volume is right-handed, otherwise it is

left-handed. Two oriented volumes have the same orientation if they have

the same handedness. The volume spanned by e1 , e2 , e3 (in that order), has

the opposite orientation from that spanned by e1 , e2 , −e3 (in that order),

i.e., moving the plane segment e1 ∧ e2 along the line segment e3 has the

opposite orientation from that of moving it along the line segment −e3 .

norm - Two oriented volumes have the same norm if they have the same volume.

We noted several properties of the outer product in Theorem 2.3.5. We now

need to add one new result, namely, the conclusion that the outer product should

obey the associative rule.

6
The terms attitude, orientation, and norm for oriented volumes are defined in some detail in
Dorst, Fontijne, and Mann [DFM07, p. 33–35].
 25

Theorem 2.4.3 (Properties of the outer product).

O5: u ∧ (v ∧ w) = (u ∧ v) ∧ w. associativity

In figure 2.14, it is easy to see that both u ∧ (v ∧ w) and (u ∧ v) ∧ w are

right-handed systems. The orientation of an oriented area depends on the order of

its factors. The anticommutative property together with the associative property

imply that exchanging any pair of factors in an outer product reverses the

orientation of the result. From this we can derive the following equalities:

T = u ∧ v ∧ w = v ∧ w ∧ u = w ∧ u ∧ v, (2.7)

−T = u ∧ w ∧ v = v ∧ u ∧ w = w ∧ v ∧ u. (2.8)

The outer product u ∧ v ∧ w of three oriented lengths u = u1 e1 + u2 e2 + u3 e3 ,

v = v1 e1 + v2 e2 + v3 e3 , and w = w1 e1 + w2 e2 + w3 e3 represents the oriented volume

of the parallelepiped with edges u, v, w:

u1 u2 u3
u ∧ v ∧ w = v1 v2 v3 (e1 ∧ e2 ∧ e3 ). (2.9)

w1 w2 w3

The oriented volume e1 ∧ e2 ∧ e3 has norm one. Any oriented volume in

Euclidean 3-space can be expressed in terms of e1 ∧ e2 ∧ e3 , thus the set

{e1 ∧ e2 ∧ e3 } forms an “orthonormal” basis for oriented volumes in

Euclidean 3-space, although orthogonality is a mute point here.

T = t (e1 ∧ e2 ∧ e3 ), where t is a scalar.

We now define scalar multiplication, vector addition, and zero for

oriented volumes.
 26

Definition 2.4.4.

Scalar multiplication: The scalar multiple aT is an oriented volume with norm

|a| |T|. If a > 0 then aT has the same orientation as T. If a < 0 then aT

has the opposite orientation as T. If a = 0 then aT = 0, the zero

oriented volume (see below). That is, if t = |T|, then

aT = a (t (e1 ∧ e2 ∧ e3 )) = a t (e1 ∧ e2 ∧ e3 ).

Vector addition: T1 + T2 is an oriented volume such that

T1 + T2 = t1 (e1 ∧ e2 ∧ e3 ) + t2 (e1 ∧ e2 ∧ e3 ) = (t1 + t2 )(e1 ∧ e2 ∧ e3 ).

Zero: The zero oriented volume, denoted by 0, has the following properties:
a0 = 0, for all scalars a,

0T = 0, for all oriented volumes T,

T + 0 = T, for all oriented volumes T.

Once again, attitude and orientation are not defined for the zero

oriented volume; the norm of 0 is 0.

Oriented volumes satisfy axioms V0–V7 of Definition 2.1.1 and thus form a

vector space of dimension one.

2.5 Scalars

Although we do not normally think of scalars as forming a vector space, a

closer examination reveals that scalars have much in common with the

oriented volumes of Euclidean 3-space. Two scalars are equivalent if they have the

same attitude, orientation, and norm.

 27

Definition 2.5.1 (Properties of scalars).

attitude - There is only one attitude for all scalars. We can visualize scalars as

points on the origin. All points on the origin have the same attitude.

orientation - Two scalars have the same orientation if they have the same sign

(i.e., positive or negative).

norm - Two scalars have the same norm if they have the same magnitude. The

magnitude or norm of a scalar is its absolute value denoted by |a|.

The set {1} forms an “orthonormal” basis for scalars. Any scalar s can be

expressed in terms of of this basis.

s = s 1, where s is a scalar.

Scalars satisfy axioms V0–V7 of Definition 2.1.1 and thus form a vector space

of dimension one.

2.6 The G3 Vector Space

We have used the terms oriented length, oriented area, and oriented volume

because they are geometrically descriptive, but they are not standard. From now

on, we use the standard terms vector for an oriented length v, bivector for an

oriented area B, and trivector for an oriented volume T.

The geometric algebra G3 is a vector space formed from the direct sum of the

scalar, vector, bivector, and trivector vector spaces, along with an additional

operation called the geometric product.

 28

Definition 2.6.1 (The G3 vector space).

The G3 vector space consists of objects of the form

M=s+v+B+T

where s is a scalar, v is a vector, B is a bivector, and T is a trivector. M is

referred to as a multivector.

Definition 2.6.2 (Grade).

Each component of a multivector has an associated grade. The grade

indicates the number of vector factors in a non-zero component, e.g.,

Component Grade Subspace

scalar 0 R

vector 1 R3
∧2 3
bivector 2 R
∧3 3
trivector 3 R
∧k
Where R3 represents the grade k subspace of G3 and G3 is the direct sum of the

four subspaces, i.e.,

∧2 ∧3
G3 = R ⊕ R 3 ⊕ R3 ⊕ R3

Two multivectors M1 and M2 are equivalent if and only if their components

are equal.

Let M1 = s1 + v1 + B1 + T1 and M2 = s2 + v2 + B2 + T2 .

Then M1 = M2 if and only if s1 = s2 , v1 = v2 , B1 = B2 , and T1 = T2 .

We now define scalar multiplication, vector addition, and zero for G3 .

 29

Definition 2.6.3.

Scalar multiplication: If M = s + v + B + T, then aM = as + av + aB + aT.

Vector addition: If M1 = s1 + v1 + B1 + T1 and M2 = s2 + v2 + B2 + T2 then

M1 + M2 = (s1 + s2 ) + (v1 + v2 ) + (B1 + B2 ) + (T1 + T2 ).

Zero: The zero multivector has zero for all its components (scalar, vector, bivector,

and trivector). We denote the zero multivector by 0 = (0, 0, 0, 0).

The set { 1, e1 , e2 , e3 , e2 ∧ e3 , e3 ∧ e1 , e1 ∧ e2 , e1 ∧ e2 ∧ e3 }
|{z} | {z } | {z } | {z }
1 scalar 3 vectors 3 bivectors 1 trivector

forms a basis for G3 . Any multivector M ∈ G3 can be expressed in terms of this

basis.

The space G3 satisfies axioms V0–V7 of Definition 2.1.1 and thus forms a

vector space of dimension eight.

2.6.1 The Geometric Product

The geometric product is the key innovation of geometric algebra. It is what

gives geometric algebra its power as a mathematical tool.

Definition 2.6.4 (Geometric product).

The geometric product uv of vectors u and v is a scalar plus a bivector:

uv = u · v + u ∧ v . (2.10)
 30

Theorem 2.6.5 (Properties of the geometric product).

G1: uv = u · v, if u ∥ v, inner product if parallel vectors

G2: uv = u ∧ v, if u ⊥ v, outer product if perpendicular vectors

G3: (au) v = a (uv), homogeneity

u (av) = a (uv),

G4: (u + v) w = uw + vw, distributive

w (u + v) = wu + wv,

G5: u (vw) = (uv) w. associativity

If we reverse the order of the terms in the geometric product, we get

vu = v · u + v ∧ u = u · v − u ∧ v. (2.11)

We now combine Eq. (2.10) with Eq. (2.11), through addition and subtraction, to

get the following important identities:

1
u · v = (uv + vu) , (2.12)
2
1
u ∧ v = (uv − vu) . (2.13)
2

2.6.2 Multiplicative Inverse of a Vector

In general, the geometric product of two vectors produces a scalar and a

bivector. Thus, the geometric product is not closed over the set of vectors but is

closed over the larger graded ring of multivectors. Although so far we have focused

on the inner, outer, and geometric products of two vectors, it can be shown that

these products are defined for all multivectors and form a closed algebra of

multivectors.
 31

To say that there exists a multiplicative identity element 1 for a vector v

under the geometric product implies that 1 v = v 1 = v. Although there is no

vector 1 with this property, if we consider the broader set of multivectors, the scalar

1 works nicely. We now extend our product definitions to include scalar-vector

products.

Definition 2.6.6 (Scalar-vector products).

Let α be a scalar and v be a vector, then

Inner: α·v =v·α=0 (2.14)

Outer: α ∧v = v ∧α = αv (2.15)

Geometric: αv = α ·v +α ∧v = 0+αv = αv (2.16)

Now that we have defined scalar-vector products, it is possible to define the

multiplicative inverse of a nonzero vector v with respect to the geometric product.

If v ̸= 0, then v v = v · v + v ∧ v = |v|2 + 0 = |v|2 . Thus, we have,

( ) ( )
vv 1 1 1
1= = vv = v v=v v ,
vv |v|2 |v|2 |v|2

where the scalar 1 plays the role of the multiplicative identity element. This leads

us to the following definition:

Definition 2.6.7 (Multiplicative inverse of a vector).

Let v be a nonzero vector. Then the multiplicative inverse of v is:

1
v−1 = v . (2.17)
|v|2
 32

2.6.3 Applying the Geometric Product to Orthonormal Vectors

When we apply the geometric product operation to the orthonormal basis

vectors of Euclidean 3-space {e1 , e2 , e3 }, we get several useful identities:

e21 = e1 e1 = e1 · e1 + e1 ∧ e1 = 1 + 0 = 1,

e22 = e2 e2 = e2 · e2 + e2 ∧ e2 = 1 + 0 = 1,

e23 = e3 e3 = e3 · e3 + e3 ∧ e3 = 1 + 0 = 1,

e2 e3 = e2 · e3 + e2 ∧ e3 = 0 + e2 ∧ e3 = e2 ∧ e3 ,

e3 e1 = e3 · e1 + e3 ∧ e1 = 0 + e3 ∧ e1 = e3 ∧ e1 ,

e1 e2 = e1 · e2 + e1 ∧ e2 = 0 + e1 ∧ e2 = e1 ∧ e2 .

These last three orthonormal bivectors show up so often in equations that going

forward we shall abbreviate them as:

e2 e3 ≡ e23 , e3 e1 ≡ e31 , e1 e2 ≡ e12 .

The highest grade component in a geometric algebra is called a pseudoscalar.

For example, a trivector—grade 3—is the the highest grade component in G3 , thus

a trivector is a pseudoscalar in G3 . In the next chapter we also discuss the

geometric algebra G2 where a bivector—grade 2—is a pseudoscalar. The name

pseudoscalar comes from the study of dual spaces where the pseudoscalar plays the

role that the scalar does in normal space. When discussing G3 , we sometimes

abbreviate the unit trivector (e1 ∧ e2 ∧ e3 )—also called the unit pseudoscalar —as

e1 e2 e3 ≡ e123 ≡ I.

Using these abbreviations it is possible to express the basis for G3 ,

{1, e1 , e2 , e3 , e2 ∧ e3 , e3 ∧ e1 , e1 ∧ e2 , e1 ∧ e2 ∧ e3 }, in the more compact form

{ 1, e1 , e2 , e3 , e23 , e31 , e12 , e123 }.

|{z} | {z } | {z } |{z}
1 scalar 3 vectors 3 bivectors 1 trivector
 33

So far, we have defined the geometric product for two vectors and for

scalar-vector combinations. If we define the geometric product for all of the basis

elements of G3 , we can then use that definition to extend the geometric product to

all multivectors in G3 . For example, consider the product of two unit bivectors:

e12 e23 = (e1 e2 )(e2 e3 ) = e1 (e2 e2 )e3 = e1 1e3 = e1 e3 = e13

e12 e31 = (e1 e2 )(e3 e1 ) = (−e2 e1 )(−e1 e3 ) = e2 (e1 e1 )e3 = e2 1e3 = e2 e3 = e23

e23 e31 = (e2 e3 )(e3 e1 ) = e2 (e3 e3 )e1 = e2 1e1 = e2 e1 = e21 = −e12

And lastly, consider the square of the unit pseudoscalar:

I2 = e2123 = e123 e123 = (e1 e2 e3 )(e1 e2 e3 ) = −1. (2.18)

Definition 2.6.8 (Geometric product of G3 basis elements).

The geometric product of any two basis elements of G3 is defined in Table 2.1.

Table 2.1: Multiplication table for the G3 geometric product.

GP 1 e1 e2 e3 e23 e31 e12 e123

1 1 e1 e2 e3 e23 e31 e12 e123
e1 e1 1 e12 −e31 e123 −e3 e2 e23
e2 e2 −e12 1 e23 e3 e123 −e1 e31
e3 e3 e31 −e23 1 −e2 e1 e123 e12
e23 e23 e123 −e3 e2 −1 −e12 e31 −e1
e31 e31 e3 e123 −e1 e12 −1 −e23 −e2
e12 e12 −e2 e1 e123 −e31 e23 −1 −e3
e123 e123 e23 e31 e12 −e1 −e2 −e3 −1

An element of a geometric algebra is called a blade if it can be expressed as

the outer product of k linearly independent vectors. A k-blade is a blade of grade k

whose constituent vectors span a k-dimensional subspace.

 34

For example, e1 , e2 , and e3 are all 1-blades because they each span a

1-dimensional subspace of Euclidean 3-space. In addition, any nontrivial linear

combination of these 1-blades is also a 1-blade. Likewise, e12 , e23 , and e31 are

2-blades because their constituent vectors each span a 2-dimensional subspace. Any

nontrivial linear combination of these 2-blades is also a 2-blade. The scalar 1 is

called a 0-blade and the unit pseudoscalar I is called a 3-blade.

In the literature, the term k-vector is often used. A k-vector is any linear

combination of k-blades. In G3 , all k-vectors are k-blades.

When all the vectors in a geometric product are orthogonal, the orientation is

changed if any two vectors are swapped.

e21 = e2 e1 = −e1 e2 = −e12 ,

e32 = e3 e2 = −e2 e3 = −e23 ,

e13 = e1 e3 = −e3 e1 = −e31 .

Likewise, for the unit trivector, all even (or odd) permutations are equivalent:

e123 = e231 = e312 = I,

e132 = e213 = e321 = −I.

We have now introduced all the fundamental objects, operations, and

properties of geometric algebra. The inner product, outer product, and geometric

product were all defined as operations on vectors. We extend the definitions of the

basic operations as needed to include bivectors, trivectors, and scalars. We started

with scalars and vectors. We were then able to define the bivector as the outer

product of two vectors. The trivector was shown to be the outer product of a vector

and a bivector or, via the associative property, the outer product of three vectors.
 35

We have obtained an 8-dimensional vector space, G3 , that is equipped with a

multiplication—the geometric product. We then have a ring over the field R, where

the underlying set, G3 , and the addition and 0 are the same in the ring and vector

space and

a(M1 M2 ) = (aM1 )M2 = M1 (aM2 )

holds for all a ∈ R and M1 , M2 ∈ G3 . A vector space with such a ring structure is

an associative algebra over the field R. An associative algebra that arises from a

vector space with a scalar product in the same manner as this example does from

Euclidean 3-space, is called a Clifford algebra.

 36

CHAPTER 3

APPLICATIONS OF THE VECTOR SPACE MODEL

3.1 The Vector Space Model

In the previous chapter we introduced the basic concepts of geometric algebra,

where G3 (a vector space of dimension eight) was used to model geometric objects

(scalars, vectors, bivectors, and trivectors) in Euclidean 3-space.

In the following chapters of this thesis, we learn that by adding one or two

additional dimensions to the space we are studying, we gain certain advantages. For

example, if we use G5 to represent Euclidean 3-space, then we can

• remove the origin as being a special point

• support points and lines at infinity

• provide a single geometric mechanism for representing lines, circles, planes,

and spheres

• enjoy all the normal features of Euclidean space.

The different approaches to representing Euclidean spaces are referred to as

models. When we represent R2 with G2 or R3 with G3 , we are using the

vector space model. When we represent R2 with G3 or R3 with G4 , we are using the

homogeneous model. When we represent R2 with G4 or R3 with G5 , we are using the

conformal model.

In all of the previous chapters we have been using the vector space model.

Using this model, we treat vectors, bivectors, and trivectors as having no fixed
 37

position, i.e., they can be moved anywhere parallel to themselves without changing

their orientation. With the vector space model we can perform many operations on

subspaces. We can span them, project them, and rotate them, but we do not move

them off the origin.

This chapter focuses on a few applications of the vector space model. We

present examples to help relate geometric algebra to other more familiar topics in

mathematics. Geometric algebra should be viewed as an abstract structure that

subsumes many of the more familiar concepts of mathematics, like vector calculus,

complex numbers, quaternions, matrix theory, tensors, and so on. Geometric

algebra facilitates a simple expression of a full range of geometric ideas.

3.2 Plane Trigonometry

Using what we currently know about geometric algebra, we now derive some

of the identities of plane trigonometry.1 In this section we are dealing exclusively

with Euclidean 2-space, R2 . We define G2 as a subalgebra of G3 generated by

{1, e1 , e2 , e1 ∧ e2 }. In G2 the unit pseudoscalar I(2) is a bivector2 , i.e., I(2) = e1 ∧ e2 .

3.2.1 Law of Cosines

Consider a triangle with sides of length a, b, and c and interior angles

α, β, and γ (see Figure 3.1a). If we redraw the triangle using vectors a, b, and c
1
This material is adapted from Hestenes [Hes03].
2
Note that the symbol I for the unit pseudoscalar can have different meanings depending on the
dimension of the vector space you are dealing with. We use the subscript (2) to distinguish the unit
pseudoscalar for G2 , I(2) = e1 ∧ e2 , from the unit pseudoscalar for G3 , I = e1 ∧ e2 ∧ e3 .
 38

for the sides so that

a + b + c = 0, (3.1)

and we assign directions to the angles so that they are all counterclockwise, we get

Figure 3.1b.

α
c c α

b γ
b
β γ β
a a
(a) Lengths and Angles (b) Vectors and Angles

Figure 3.1: Plane trigonometry

The lengths of the vectors are |a| = a, |b| = b, and |c| = c. The angle α starts

at c and ends on −b, thus the geometric product of c and − b is

−cb = −c · b − c ∧ b = cb cos α + cbI(2) sin α, where I(2) is the unit pseudoscalar

that defines the orientation of the of the plane containing the vectors a, b, c. We

can compute the geometric product of the vectors that define the angles β and γ in

a similar manner. This leads us to the following three equations:

−cb = −c · b − c ∧ b = cb cos α + cbI(2) sin α, (3.2a)

−ac = −a · c − a ∧ c = ac cos β + acI(2) sin β, (3.2b)

−ba = −b · a − b ∧ a = ba cos γ + baI(2) sin γ. (3.2c)

For each of these three equations we can separate the geometric product into its
 39

scalar and bivector parts.

a · b = −ab cos γ, a ∧ b = ab I(2) sin γ, (3.3a)

b · c = −bc cos α, b ∧ c = bc I(2) sin α, (3.3b)

c · a = −ca cos β, c ∧ a = ca I(2) sin β. (3.3c)

Now if we solve Eq. (3.1) for c and then square both sides, we get

c = −a − b,

c2 = (−a − b)2

= (−a − b)(−a − b)

= a2 + ab + ba + b2

= a2 + b2 + ab + ba
[ ]
= a2 + b2 + 2 12 (ab + ba)

c2 = a2 + b2 + 2 [a · b]

c2 = a 2 + b 2 + 2 a · b

c2 = a2 + b2 − 2ab cos γ. (3.4)

Where the last step is obtained by substitution of Eq. (3.3a) for a · b. This equation

(3.4) is the familiar Law of Cosines from trigonometry.

 40

3.2.2 Law of Sines

Next, we start with Eq. (3.1) and take the outer product first by a and then

by b or c. We get

a + b + c = 0, a + b + c = 0,

(a + b + c) ∧ a = 0 ∧ a (a + b + c) ∧ b = 0 ∧ b

a∧a+b∧a+c∧a=0 a∧b+b∧b+c∧b=0

0+b∧a+c∧a=0 a∧b+0+c∧b=0

−a ∧ b + c ∧ a = 0 a∧b−b∧c=0

a ∧ b = c ∧ a. a ∧ b = b ∧ c.

Combining these two equations we get

a ∧ b = b ∧ c = c ∧ a. (3.5)

Now substituting from Eq. (3.3a), (3.3b), and (3.3c) into Eq. (3.5) and multiplying

each term on the right by the inverse of abc I(2) and noting that the sine terms are

all scalars—we get

a∧b=b∧c=c∧a

ab I(2) sin γ = bc I(2) sin α = ca I(2) sin β

sin γ sin α sin β

= = . (3.6)
c a b
Thus, we have derived the Law of Sines, Eq. (3.6), using geometric algebra.

3.2.3 Area of a Triangle

Eq. (3.5) can be regarded as giving three equivalent ways of determining the

area of a triangle. We have previously noted that a ∧ b is the oriented area of a

 41

parallelogram; our triangle has only half that area. Hence, the oriented area A of

the triangle is given by

A = 21 a ∧ b = 12 b ∧ c = 12 c ∧ a.

Using Eq. (3.3a), (3.3b), and (3.3c) we can also show that

|A| = 21 (a ∧ b)I−1 −1 −1
(2) = 2 (b ∧ c)I(2) = 2 (c ∧ a)I(2) ,
1 1

|A| = 12 ab sin γ = 12 bc sin α = 21 ca sin β. (3.7)

Thus, the area is one-half the base times the height.

3.2.4 Sum of the Interior Angles of a Triangle

If we multiply Eq. (3.2a), (3.2b), and (3.2c) together and then divide by

a2 b2 c2 , we get

(−ac)(−cb)(−ba) = ac(cos β + I(2) sin β) cb(cos α + I(2) sin α) ba(cos γ + I(2) sin γ),

−accbba accbba (cos β + I(2) sin β) (cos α + I(2) sin α) (cos γ + I(2) sin γ)
2 2 2
= ,
abc a 2 b 2 c2

− a−1 c−1 cb−1 ba = [(cos β + I(2) sin β) (cos α + I(2) sin α)] (cos γ + I(2) sin γ),

− 1 = [(cos β cos α − sin β sin α) + I(2) (cos β sin α + sin β cos α)](cos γ + I(2) sin γ)

− 1 = [cos(α + β) + I(2) sin(α + β)] (cos γ + I(2) sin γ)

− 1 = (cos(α + β) cos γ − sin(α + β) sin γ)

+ I(2) (cos(α + β) sin γ + sin(α + β) cos γ)

− 1 = cos(α + β + γ) + I(2) sin(α + β + γ). (3.8)

This shows that successive rotations through the three interior angles of a

triangle are equivalent to a straight angle. We can conclude that α + β + γ = π.

 42

3.3 Matrix Representation

It turns out that the normal basis elements of G3 ,

{1, e1 , e2 , e3 , e12 , e23 , e31 , e123 }, can be represented by square matrices. 3

We represent the unit scalar 1 as the 4 × 4 identity matrix:

 
1 0 0 0
 
 
0 1 0 0
 
1= . (3.9)
 
0 0 1 0
 
0 0 0 1

The unit orthogonal vectors e1 , e2 , and e3 are represented as

     
0 0 0 1 0 0 1 0 1 0 0 0
     
     
0 0 1 0  0 0 0 −1   0 1 0 0 
    
e1 =   , e2 =   , e3 =  . (3.10)
     
0 1 0 0 1 0 0 0  0 0 −1 0 
     
1 0 0 0 0 −1 0 0 0 0 0 −1

The unit orthogonal bivectors e12 , e23 , and e31 are represented as
     
0 −1 0 0 0 0 −1 0 0 0 0 1
     
     
1 0 0 0  0 0 0 1 0 0 1 0
     
e12 =   , e23 =   , e31 =  .
     
0 0 0 −1 1 0 0 0 0 −1 0 0
     
0 0 1 0 0 −1 0 0 −1 0 0 0
(3.11)

And, the unit trivector e123 is represented as

 
0 −1 0 0
 
 
1 0 0 0
 
e123 = I =  . (3.12)
 
0 0 0 1
 
0 0 −1 0
3
Much of the material in this section follows that of Snygg [Sny10].
 43

Consider the vector subspace of M4 (R) = {4 × 4 real matrices} spanned by

these eight matrices. Representing scalars, vectors, bivectors, and trivectors in

terms of these square matrices enables us to visualize the geometric product as

matrix multiplication.

It is easy to verify that the set of matrices given above satisfies the definition

of a geometric product for G3 as given in Table 2.1. In particular,

e21 = e22 = e23 = 1, (3.13)

e12 + e21 = e23 + e32 = e31 + e13 = 0, (3.14)

I2 = e2123 = −1. (3.15)

Now let us consider the geometric product of two vectors. Suppose

u = u1 e1 + u2 e2 + u3 e3 and v = v1 e1 + v2 e2 + v3 e3 , then

uv = (u1 v1 + u2 v2 + u3 v3 ) +

u2 v3 e2 e3 + u3 v2 e3 e2 + u3 v1 e3 e1 + u1 v3 e1 e3 + u1 v2 e1 e2 + u2 v1 e2 e1 .

Using the relations of Eq. (3.14), we have

uv =(u1 v1 + u2 v2 + u3 v3 ) +

(u2 v3 − u3 v2 )e2 e3 + (u3 v1 − u1 v3 )e3 e1 + (u1 v2 − u2 v1 )e1 e2 . (3.16)

Notice that uv consists of a scalar part

u · v = u1 v1 + u2 v2 + u3 v3 , (3.17)

and a bivector part

u ∧ v = (u2 v3 − u3 v2 )e2 e3 + (u3 v1 − u1 v3 )e3 e1 + (u1 v2 − u2 v1 )e1 e2 . (3.18)

We note that the coefficients of e2 e3 , e3 e1 , and e1 e2 that appear in the outer

product u ∧ v are the three components of the cross product u × v. Instead of being
 44

the components of a vector perpendicular to u and v, they are components of the

bivector u ∧ v. It is this distinction which enables the outer product to be defined

in any dimension. There is a strong connection between the cross product and the

outer product, but whereas the cross product requires the introduction of a third

dimension perpendicular to the two vector factors, the outer product does not.

3.4 Generalized Complex Numbers

One of the achievements of Clifford’s geometric algebra is that it generalizes

complex arithmetic to spaces of arbitrary dimension.4

We initially limit our discussion to the G2 subalgebra of G3 consisting of the

e12 -plane. This subalgebra is a 4-dimensional real algebra with a basis

{1, e1 , e2 , e12 }. The basis elements obey the multiplication table shown in

Table 3.1.

Table 3.1: Multiplication table for the G2 geometric product.

GP 1 e1 e2 e12
1 1 e1 e2 e12
e1 e1 1 e12 e2
e2 e2 −e12 1 −e1
e12 e12 −e2 e1 −1

4
Much of the material in this section follows Lounesto [Lou03, p. 26], Doran and Lasenby [DL03,
p. 26–28] and Dorst, Fontijne, and Mann [DFM07, p. 177–178].
 45

The basis elements span the subspaces consisting of

basis subspace grade description

1 R 0 scalars

e1 , e2 R2 1 vectors
∧2 2
e12 R 2 bivectors,
∧2
where R2 represents the grade 2 subspace of G2 , i.e., the subspace of

bivectors.
∧2
Thus, the geometric algebra G2 contains copies of R, R2 , and R2 , and it is

the direct sum of its subspaces, elements of grades 0, 1, and 2:

∧2
G2 = R ⊕ R2 ⊕ R2 .

∧2
The geometric algebra G2 is also a direct sum of its even part G2+ = R ⊕ R2 and

its odd part G2− = R2 ,

G2 = G2+ ⊕ G2−

The even part is not only a subspace but also a subalgebra. It consists of elements

of the form x + ye12 , where x, y ∈ R and e212 = −1. This even subalgebra G2+ is

isomorphic to the complex numbers C.

Definition 3.4.1 (Unit pseudoscalar).

The bivector e1 ∧ e2 = e12 = I(2) is called the unit pseudoscalar of the

oriented plane containing the orthogonal unit vectors e1 and e2 .

This unit pseudoscalar uniquely determines an oriented plane in

Euclidean 3-space. An important property of unit pseudoscalar I(2) is that I2(2) = −1.

In G2+ we represent a complex number Z as Z = a + bI(2) , where a, b ∈ R.

Consider what happens when we take a complex number Z and multiply it on the
 46

right by the unit pseudoscalar I(2) .

ZI(2) = (a + bI(2) )I(2) = aI(2) + bI2(2) = aI(2) − b = −b + aI(2) .

The result of right multiplying by the unit pseudoscalar is to rotate the complex

number Z by 90◦ counterclockwise (see Figure 3.2).

I(2)
ZI(2)
a

b Z

-b a R

Figure 3.2: Right multiplying Z by the unit pseudoscalar.

Likewise, consider what happens when we multiply the same complex number

Z by the unit pseudoscalar I(2) on the left.

I(2) Z = I(2) (a + bI(2) ) = aI(2) + bI2(2) = aI(2) − b = −b + aI(2) .

Once again, the result of left multiplying by the unit pseudoscalar is to rotate

the vector by 90◦ counterclockwise. Just as with multiplication by i in the ordinary

complex numbers, multiplication of a complex number in G2+ by I(2) is

commutative and it rotates a complex number by 90◦ counterclockwise in the plane.

Suppose we would like to rotate a complex number Z = a + bI(2) in the

complex plane by some angle other than 90◦ . For this we need to define the

exponential function.
 47

Definition 3.4.2 (Exponential).

Define the exponential as eI(2) θ = cos θ + I(2) sin θ.

The exponential is the sum of a scalar and a bivector. For any given angle θ,

the norm of the exponential is always 1.

The exponential can also be visualized as a geometric product of two unit

vectors. It captures the angle between them and the orientation of the rotation with

respect to the unit pseudoscalar. If u1 and u2 are unit vectors in the I(2) -plane and

θ is the angle measured from u1 to u2 , then

u1 u2 = u1 · u2 + u1 ∧ u2 = u1 u2 cos θ + u1 u2 I(2) sin θ,

but u1 = u2 = 1, thus

u1 u2 = cos θ + I(2) sin θ = eI(2) θ .

By extension from a basis, the geometric product of any two vectors is a

complex number. Let u and v be any two vectors in the I(2) -plane. Then

Z = uv = u · v + u ∧ v = uv cos θ + uv sin θI(2) = uveI(2) θ ,

Z = a + bI(2) = reI(2) θ ,

where a = uv cos θ, b = uv sin θ,

( )
√ −1 b
r = uv = a2 + b2 , and θ = tan .
a

If we wish to rotate a complex number Z that exists in the I(2) -plane by an

angle θ, we can right- or left-multiply Z by eI(2) θ .

ZeI(2) θ = (a + bI(2) )(cos θ + I(2) sin θ)

= a cos θ + a sin θ I(2) + b cos θ I(2) + b sin θ I2(2)

= (a cos θ − b sin θ) + (a sin θ + b cos θ) I(2) .

 48

I(2)
ZeI(2) θ

Z
b θ

a R

Figure 3.3: Multiplying Z by an exponential.

See Figure 3.3.

Note that if θ = 90◦ , then

eI(2) θ = cos 90◦ + I(2) sin 90◦ = 0 + I(2) = I(2) ,

which confirms our previous example that multiplication of a complex number by

I(2) causes a 90◦ counterclockwise rotation.

In the preceding discussion we viewed the pseudoscalar I(2) as representing the

e12 -plane in Euclidean 3-space formed by the two standard orthonormal vectors e1

and e2 . If we consider {e1 , e2 , e3 } as the standard basis for R3 , then any plane in

Euclidean 3-space can be expressed as a linear combination of the basis bivectors

{e23 , e31 , e12 }, thus we should more properly think of bI(2) as any plane through the

origin in R3 where

bI(2) = b1 e23 + b2 e31 + b3 e12 ,

√
such that b = b21 + b22 + b23 and |I(2) | = 1.

Because we have defined a complex number in G2+ to be of the form

 49

Z = a + bI(2) , a generalized complex number has the form

Z = a + bI(2) = a + b1 e23 + b2 e31 + b3 e12 . (3.19)

Generalized complex numbers are just multivectors with a scalar part and a

bivector part. Table 3.2 does a side-by-side comparison of the algebra of complex

numbers with that of the geometric subalgebra of a plane, G2+ .

Table 3.2: Complex Numbers vs. G2+ Geometric Algebra

Property Complex Numbers G2+ Geometric Algebra

rectangular z = a + bi Z = a + bI(2)
polara z = reiθ Z = reI(2) θ
geometric product N/A Z = uv
rectangular conjugate z = a − bi Z = vu = a − bI(2)
polar conjugate z = re−iθ Z = vu = re−I(2) θ
i and I(2) squared i2 = −1 I2(2) = −1
rotate 90◦ CCW zi ZI(2)
rotate 90◦ CW −zi −ZI(2)
commutativity zi = iz ZI(2) = I(2) Z
norm squared |z|2 = zz = a2 + b2 |Z|2 = ZZ = a2 + b2
rotate by θ zeiθ ZeI(2) θ
√ (b)
a
Where r= a2 + b2 and θ = tan−1 a .

We have used the symbol I(2) for the unit pseudoscalar, rather than the

tempting alternative i. It is important to make this distinction. Even though G2+ is

isomorphic with the field of complex numbers, we see that i and I(2) behave

differently when we multiply them with vectors.

Consider what happens when we multiply a real vector v = ae1 + be2 by i.

vi = (ae1 + be2 )i = aie1 + bie2 ,

 50

we get a complex vector. Also note that

vi = iv,

thus multiplication of a vector by i is commutative.

Next, consider what happens when we take a vector v = ae1 + be2 in the

e12 -plane and multiply it on the right by the unit pseudoscalar I(2) .

vI(2) = (ae1 + be2 )e12 = ae1 e12 + be2 e12 = ae2 − be1 = −be1 + ae2 ,

The result of right-multiplying by the unit pseudoscalar is to rotate the vector by

90◦ counterclockwise (see Figure 3.4).

e2
vI(2)
a

v
b

−b a e1

Figure 3.4: Right multiplying v by the unit pseudoscalar.

Likewise, consider what happens when we multiply the same vector v by the

unit pseudoscalar I(2) on the left.

I(2) v = e12 (ae1 + be2 ) = ae12 e1 + be12 e2 = −ae2 + be1 = be1 − ae2 .

The result of left-multiplying by the unit pseudoscalar is to rotate the vector by 90◦

clockwise (see Figure 3.5).

 51

e2

b v

b
a e1

−a
I(2) v
Figure 3.5: Left multiplying v by the unit pseudoscalar.

It should be noted that our pseudoscalar I(2) anticommutes with e1 and e2

and thus I(2) anticommutes with every vector in the e12 -plane. From the above

discussion we see that if we limit ourselves to the algebra of G2+ , I(2) behaves

exactly the same as i, but if we broaden our algebra to include all of G2 , then I(2)

behaves quite differently from i.

Another name for a generalized complex number is spinor. The name spinor is

suggestive of its action on a vector under geometric multiplication. Each spinor can

be regarded as an algebraic representation of a rotation-dilation. Generalized

complex numbers, or spinors, form a subalgebra of G3 .

3.5 The Quaternions

It was shown in the previous section that complex numbers can be subsumed

in the broader frame of geometric algebra. The same is true of the quaternions.

William Rowan Hamilton (1805–1865) developed an algebra generated by four

fundamental units {1, i, j, k} with which many aspects of mechanics could be

 52

handled [DL03, p. 8-10]. Hamilton’s quaternion took the form

a + bi + cj + dk,

where i2 = j2 = k2 = ijk = −1,

and ij = k, jk = i, ki = j.

The quaternions form a skew field, a noncommutative division ring, under addition

and multiplication.

If we take our orthonormal basis for the bivectors {e23 , e31 , e12 } and reverse

their signs, we get the 2-blades

i = e32 , j = e13 , k = e21 .

With this definition, all of the identities listed above hold. Hamilton identified

pure quaternions (null scalar part) with vectors, but we now see that they are

actually bivectors. To set up an isomorphism between quaternions and bivectors, we

must flip the signs of the bivectors. This shows that the quaternions are a

left-handed set of bivectors, whereas Hamilton and others attempted to view i, j, k

as a right-handed set of vectors. Not surprisingly, this was a potential source of

great confusion and meant one had to be extremely careful when applying

quaternions in vector algebra [DL03, p. 34].

Quaternions are just multivectors with a scalar part and a bivector part. Like

generalized complex numbers, quaternions are spinors that form a subalgebra of G3 .

3.6 Reflections

We now take a closer look at the geometric product uv of two vectors u and

v. If we right multiply this product by the inverse of v (v−1 ), we should retrieve the
 53

vector u.

u = u (vv−1 ) = (uv) v−1

= (u · v + u ∧ v) v−1

= (u · v) v−1 + (u ∧ v) v−1 . (3.20)

Because u is a vector and (u · v) v−1 is a vector, we see that (u ∧ v) v−1 is also a

vector. The first term is the component of u in the v direction, i.e., the orthogonal

projection of u onto v (u∥v ). The second term represents the component of u that

contains no v component at all, i.e., the rejection of u by v (u⊥v ). Thus, we have:

u = u∥v + u⊥v ,

where u∥v = (u · v) v−1 ,

and u⊥v = (u ∧ v) v−1 .

We have decomposed the vector u relative to v, providing its v-component and

non-v-component (see Figure 3.6).

u⊥v = (u ∧ v)v−1

u∥v = (u · v)v−1

Figure 3.6: Projection and rejection of u relative to v.

We now consider what happens when we left multiply the geometric product
 54

uv by v−1 .

v−1 uv = v−1 (uv)

v
= (uv) (def. of vector inverse)
|v|2
v
= vu 2 (because scalars commute)
|v|
= vuv−1 = (vu) v−1

= (v · u + v ∧ u) v−1

= (v · u) v−1 + (v ∧ u) v−1

= (u · v) v−1 − (u ∧ v) v−1 . (3.21)

If we now compare Eq. (3.21) to Eq. (3.20), we see that the first term (the

projection) is the same, but the second term—the rejection—has the opposite sign.

The resulting of sandwiching the vector u between the vectors v and v−1 (v−1 uv)

has, in effect, reflected the u vector across the line of the v vector in the plane of

the two vectors (see Figure 3.7).

u⊥v = (u ∧ v)v−1

u∥v = (u · v)v−1

v−1 uv

−u⊥v = −(u ∧ v)v−1

Figure 3.7: Reflection of u in v.

The magnitude and orientation of the vector v are irrelevant to the outcome,
 55

because the inversion removes any scalar factor. Only the attitude of the line

matters. We have proved the following theorem:

Theorem 3.6.1.

The reflection of a vector u in a vector v is:

Reflection of u in v : u 7→ v−1 uv . (3.22)

Sandwiching operations are quite common in geometric algebra. It is

important to note that the objects of geometric algebra, vectors in this case, are also

the operators of a geometric transform, reflection in this case. One of the powers of

geometric algebra is that geometric objects and operators are all treated the same.

3.7 Rotations

The usual way to specify a rotation in Euclidean 3-space is by its axis n and a

scalar rotational angle θ. Instead we specify the rotation by the “plane I(2) ”, i.e.,

the plane with pseudoscalar I(2) orthogonal to the axis (the “plane of rotation”) and

θ (the “size of the rotation”).

Suppose that the rotation carries the vector u to a vector u′ .

Consider the first case where u is in the plane of rotation I(2) . Then so is u′

(see Figure 3.8).

From previous sections, we have:

uu′ = u · u′ + u ∧ u′

= |u| |u′ | cos θ + |u| |u′ | I(2) sin θ

= |u| |u′ | (cos θ + I(2) sin θ)

= |u| |u′ | eI(2) θ . (3.23)

 56

e2
u′ = ueI(2) θ

u
θ

e1

Figure 3.8: Rotation in the I(2) plane by angle θ.

If we multiply both sides of Eq. (3.23) by u on the left, and use the fact that

|u| = |u′ | and u2 = |u| |u| = |u| |u′ |, we obtain:

uu′ = |u| |u′ | eI(2) θ

|u| |u| u′ = u |u| |u| eI(2) θ

u′ = u eI(2) θ . (3.24)

Figure 3.8 depicts Eq. (3.24) in action: eI(2) θ rotates u to u′ in the plane of I(2) .

We now consider the more general case where the vector u is not necessarily

in the plane of rotation I(2) . We decompose u into its projection onto the plane

I(2) , u∥ and its rejection, the component perpendicular to the plane I(2) , u⊥ , thus

u = u∥ + u⊥ (see Figure 3.9).

Here is the key: as u rotates to u′ , u∥ rotates to u∥ eI(2) θ and u⊥ is carried

 57

u u′ = e−I(2) θ/2 ueI(2) θ/2

u⊥ n u⊥

θ
u∥ u∥ eI(2) θ

Figure 3.9: Rotation in any plane by angle θ.

along unchanged. Thus,

u′ = u∥ eI(2) θ + u⊥

= u∥ eI(2) θ/2 eI(2) θ/2 + u⊥ e−I(2) θ/2 eI(2) θ/2

= e−I(2) θ/2 u∥ eI(2) θ/2 + e−I(2) θ/2 u⊥ eI(2) θ/2

= e−I(2) θ/2 ueI(2) θ/2 . (3.25)

We have proved the following theorem which gives an elegant algebraic

representation of rotations in three dimensions.

Theorem 3.7.1.

Let RI(2) θ (u) denote the rotation of a vector u by the angle θ around an axis

perpendicular to the plane I(2) . Then

Rotation: RI(2) θ (u) = e−I(2) θ/2 ueI(2) θ/2 . (3.26)

We say the quaternion e−I(2) θ/2 represents the rotation RI(2) θ .

We know from geometry that two reflections make a rotation, so we can also

visualize this rotation as the result of two successive reflections. Suppose that v and
 58

w are two unit vectors in the plane of rotation and that the angle from v to w is

θ/2. We first reflect u in v, giving v−1 uv. Then we reflect this result in w giving

w−1 (v−1 uv) w = (w−1 v−1 )u(vw) = (vw)−1 u(vw).

See Figure 3.10.

The net result of the two reflections is the same as the rotation obtained with

e−I(2) θ/2 ueI(2) θ/2 , where vw = eI(2) θ/2 and (vw)−1 = e−I(2) θ/2 . Notice that to get a

rotation of angle θ, the angle between v and w must be θ/2 and both v and w must

be in the plane of rotation I(2) . The magnitude and orientation of the vectors v and

w are irrelevant to the outcome, because the inversion removes any scalar factor.

Only the attitude of the lines matters. In summary, to rotate an object by an angle

θ with respect to a plane, we need only choose any two vectors v and w in that

plane such that the angle between v and w is θ/2.

(vw)-1uvw
u θ

v w
θ/2

v-1uv
Figure 3.10: Rotation viewed as two successive reflections.

Once again, notice the sandwiching operation where the vector to be

transformed (u) is sandwiched between two operators. One of the most powerful

aspects of geometric algebra is the ability to realize isometries as sandwich

 59

operations of the form

X 7→ g−1 Xg

where X is any geometric element of the algebra and g is a specific geometric

element, unique to the isometry. The operator g is in general a versor, i.e., it can be

expressed as the product of 1-vectors.

3.8 Geometry

In this section we use geometric algebra to prove several theorems of

geometry. First we present several lemmas that are helpful in proving other lemmas

and theorems that follow.

3.8.1 The Inverse of a Bivector

Many of the lemmas and theorems in this section deal with inverse bivectors,

so we first discuss the meaning of the inverse of a multivector 5 .

The inverse of a multivector A, if it exists, is denoted by A−1 or 1

A
. We can

divide any multivector B by A in two ways, by multiplying by A−1 on the left,

1
A−1 B = B,
A
or on the right,
1
BA−1 = B = B/A.
A
The left division is not equivalent to the right division unless B commutes with

A−1 , in which case the division can be denoted unambiguously by

B
.
A
5
This description of multivector inverse comes from Hestenes [Hes03, p. 37].
 60

This is always the case when both A and B are bivectors in the same plane,

as shown in Lemma 3.8.3 and Lemma 3.8.4 below.

Definition 3.8.1 (Reversion).

Reversion ∼ is an operation that takes a k-blade B = b1 ∧ b2 ∧ · · · ∧ bk and

produces its reverse:

∼
reversion: B = bk ∧ bk−1 ∧ · · · ∧ b1 ,

which just has all vectors of B in reverse order. The notation of the tilde is chosen

to be reminiscent of an editor’s notation for an interchange of terms.6

Definition 3.8.2 (Inverse of a blade).

The inverse of a blade B is

∼
B
B−1 = ∼ . (3.27)
BB

3.8.2 Lemmas

Lemma 3.8.3.

The inverse of a bivector, if it exists, is a bivector in the same plane as the

original bivector.

Proof.

Let A be an invertible bivector. Without loss of generality, we assume that A

lies in the e1 ∧ e2 plane, viz.,

A = αe12 , where α is a scalar.

∼
6
Some literature denotes reversion by B† instead of B, because it is related to complex conju-
gation in a certain context.
 61

Then,
∼
−1 A αe21 1 1
A = ∼ = = e21 = − e12 .
AA α2 e
12 e21 α α
−1
Therefore, A is a bivector in the same plane as A.

Lemma 3.8.4.

If A and B are are bivectors in the same plane, then their geometric product

commutes, i.e.,

AB = BA.

Proof.

Let A and B be a bivectors that lie in the same plane. Without loss of

generality, we assume that A and B lie in the e1 ∧ e2 plane, i.e.,

A = αe12 , B = βe12 , where α and β are scalars.

Then,

AB = αe12 βe12 = −αβ and BA = βe12 αe12 = −αβ.

Therefore,

AB = BA.

Lemma 3.8.5.

Let A, B, A′ , B′ all represent bivectors in the same plane, where

A ̸= A′ , B ̸= ±B′ , B ̸= 0, B′ ̸= 0.

A A′ A A ± A′
If = ′, then = . (3.28)
B B B B ± B′
 62

Proof.

Because A and B are in the same plane, A commutes with B−1 by

Lemma 3.8.3 and Lemma 3.8.4. That is,

A
AB−1 = B−1 A = .
B
A′
Likewise, A′ commutes with (B′ )−1 . But A
B
= B′
implies that A′ = mA and

B′ = mB for some m ∈ R such that m ̸= 0, ±1. Then we have

A A (1 ± m) A A ± mA A ± A′
=1 = = = .
B B (1 ± m) B B ± mB B ± B′

Lemma 3.8.5 also works when A and B are scalars.

Definition 3.8.6 (Altitude of a triangle).

An altitude of a triangle is a segment from a vertex perpendicular to the

line in which the opposite side—the base—lies (see Figure 3.11 where the bases

AB, BC, CA have respective altitudes of a1 , a2 , a3 ). When referring the the length of

an altitude line segment, we use the term height.

a1
a3
A
B
a2

Figure 3.11: Altitude of a triangle

 63

Lemma 3.8.7.

When two triangles have the same height, the ratio of their areas equals the

ratio of their bases.

h h

A1 A2
A1 A2
b1 b2 b1 b2
(a) Configuration 1 (b) Configuration 2

Figure 3.12: Lemma 3.8.7

Proof.

Let A1 and A2 be the areas of two triangles with respective bases of b1 and b2

and sharing a common height of h (see Figure 3.12). Then A1 = 21 b1 h, A2 = 12 b2 h.

Therefore
1
A1 b1 h b1
= 21 = . (3.29)
A2 bh
2 2
b2

We use the configuration shown in Figure 3.12b for proving the theorems that

follow.

We now present an extension of Lemma 3.8.7 without proof.

 64

Lemma 3.8.8.

When two triangles have the same height, the ratio of the codirectional

bivectors representing their areas equals the ratio of the codirectional vectors

representing their bases, e.g.,

A1 a−x
= . (3.30)
A2 x−b
See Figure 3.13.

A1 A2
a x b
a-x x-b
Figure 3.13: Lemma 3.8.8, using vectors and bivectors

Lemma 3.8.9.

Three vectors x, a, and b, with initial points on the origin, have end points

that are collinear if and only if x can be expressed as a linear combination of a and

b, viz.,

x = αa + βb, (3.31a)

where α + β = 1. (3.31b)

See Figure 3.14 (a) and (b).

Proof.
 65

b x
x b
A a A a
B A+B

0 0

(a) Configuration a (b) Configuration b

Figure 3.14: Lemma 3.8.9

(⇒) Suppose that three vectors x, a, and b with initial points on the origin have

end points that are collinear. Let A be the oriented area of triangle △0xb,

A = 12 (x ∧ b) and let B be the oriented area of triangle △0ax, B = 12 (a ∧ x). From

Eq. (3.30) we have

A b−x
=
B x−a

B
x−a= (b − x)
A

B B
x−a= b− x
A A

B B
x+ x=a+ b
A A
B
a+ A b
x= B
1+ A
( ) ( )
A B
x= a+ b
A+B A+B
 66
( ) ( )
A B
x= a+ b.
A+B A+B

Where the last step is justified because A and B are in the same plane and are

codirectional.

A B
Let α= and β= ,
A+B A+B
then x = αa + βb and α + β = 1.

(⇐) Suppose that three vectors x, a, and b with initial points on the origin satisfy

the equation x = αa + βb where α + β = 1. Then α = 1 − β, thus

x = (1 − β)a + βb

x = a − βa + βb

x = a + β(b − a). (3.32)

Eq. (3.32) indicates that x is the sum of a and a scalar multiple of b − a, thus the

end point of x is collinear with the end points of a and b (see Figure 3.15).

b-a
b β(b-a)
x
a

Figure 3.15: Lemma 3.8.9 proof

 67

Lemma 3.8.10.

Three distinct points a, b, c lie on a line if and only if there exist scalars

α, β, γ, not all zero, such that αa + βb + γc = 0 and α + β + γ = 0.

Proof.

(⇒) Suppose that the three distinct points a, b, c lie on a line. Then by

Eqs. (3.31a and d), we have

c = αa + βb and α + β = 1,

0 = αa + βb − c,

0 = αa + βb + γc, where γ = −1,

0 = αa + βb + γc, where α + β + γ = 0.

(⇐) Suppose that three distinct points a, b, c are related by the equations

αa + βb + γc = 0, where α + β + γ = 0 and at least one of α, β, γ is nonzero.

Without loss of generality we may assume that γ ̸= 0. Then we have

−γc = αa + βb and − γ = α + β,
( ) ( )
α β
c= a+ b,
−γ −γ
( ) ( )
α β
c= a+ b,
α+β α+β

α β
c = α′ a + β ′ b, where α′ = and β ′ = ,
α+β α+β

c = α′ a + β ′ b, where α′ + β ′ = 1.

Therefore, the three points are collinear by Lemma 3.8.9.

 68

3.8.3 Ceva’s Theorem

Ceva’s theorem is a historically important theorem from geometry that was

first published in 1678 by the Italian Giovanni Ceva [Eve72, p. 63]. A cevian line is

a line passing through the vertex of a triangle that does not coincide with a side of

the triangle. A cevian line cuts the opposite side of a triangle at a point. Ceva’s

theorem examines the directed lengths of the sides of a triangle that have been cut

by cevian lines. It uses the product of ratios to determine if three cevian lines are

concurrent at a point (see Figure 3.16).

b'
a'

a c' b

Figure 3.16: Ceva’s Theorem.

Theorem 3.8.11 (Ceva’s Theorem).

Three cevian lines aa′ , bb′ , cc′ of an ordinary triangle △abc are concurrent if

and only if
( )( )( )
b − a′ c − b′ a − c′
= +1.
a′ − c b′ − a c′ − b

Proof of Ceva’s Theorem.

 69

b' B A2
1 a'
B2 0 A1

C1 C2
a c' b

Figure 3.17: Proof of Ceva’s Theorem.

(⇒) Suppose that three cevian lines aa′ , bb′ , cc′ of an ordinary triangle △abc

are concurrent at the point 0 as shown in Figure 3.17. Let A1 , A2 , B1 , B2 , C1 , C2 be

bivectors whose norms are equal to the areas of the smaller triangles.

Closer examination of Figure 3.17 reveals that each pair of smaller triangles

shares a common height, viz.,

triangles A1 and A2 , with respect to bases b − a′ and a′ − c,

triangles B1 and B2 , with respect to bases c − b′ and b′ − a,

triangles C1 and C2 , with respect to bases a − c′ and c′ − b.

By Eq. (3.30) we have

A1 b ∧ a′ b − a′
= ′ = ′
A2 a ∧c a −c

B1 c ∧ b′ c − b′
= ′ = ′ (3.33)
B2 b ∧a b −a

C1 a ∧ c′ a − c′
= ′ = ′ .
C2 c ∧b c −b

Next we consider the same ratios of sides, but this time we equate these ratios to
 70

areas of the larger triangle △abc. Once again, these larger pairs of triangles share a

common height. For example,

triangles △ baa′ and △ a′ ac, with respect to bases b − a′ and a′ − c,

triangles △ cbb′ and △ b′ ba, with respect to bases c − b′ and b′ − a,

triangles △ acc′ and △ c′ cb, with respect to bases a − c′ and c′ − b.

Then by Eq. (3.30) we have

b − a′ (b − a) ∧ (a′ − a) A1 + C1 + C2
= =
a −c
′ (a − a) ∧ (c − a)
′ A2 + B1 + B2

c − b′ (c − b) ∧ (b′ − b) B1 + A1 + A2
= = (3.34)
b −a
′ (b − b) ∧ (a − b)
′ B2 + C1 + C2

a − c′ (a − c) ∧ (c′ − c) C1 + B1 + B2
= = .
c −b
′ (c − c) ∧ (b − c)
′ C2 + A1 + A2

Equating the formulas in Eq. (3.33) with those in Eq. (3.34) and applying

Lemma 3.8.5, we get

A1 A1 + C1 + C2 C1 + C2
= =
A2 A2 + B1 + B2 B1 + B2

B1 B1 + A1 + A2 A1 + A2
= = (3.35)
B2 B2 + C1 + C2 C1 + C2

C1 C1 + B1 + B2 B1 + B2
= = .
C2 C2 + A1 + A2 A1 + A2

If we take the product of the factors in Eqs. (3.35), we get

( )( )( ) ( )( )( )
A1 B1 C1 C1 + C2 A1 + A2 B1 + B2
= = +1. (3.36)
A2 B2 C2 B1 + B2 C1 + C2 A1 + A2

And by Eqs. (3.33) this is equivalent to

( )( )( )
b − a′ c − b′ a − c′
= +1. (3.37)
a′ − c b′ − a c′ − b
 71

(⇐) Suppose that three cevian lines aa′ , bb′ , cĉ of an ordinary triangle △abc

divide their respective opposite sides such that the product

( )( )( )
b − a′ c − b′ a − ĉ
= +1. (3.38)
a′ − c b′ − a ĉ − b
Let cc′ be the cevian line concurrent with cevian lines aa′ and bb′ (see Figure 3.18).

In the first part of this proof, we showed that for concurrent cevian lines aa′ , bb′ , cc′

b'
a'

a b
^
c c'

Figure 3.18: Converse of Ceva’s Theorem.

we have Eq. (3.37). If we equate the left hand sides of Eq. (3.37) and (3.38) we get
( )( )( ) ( )( )( )
b − a′ c − b′ a − c′ b − a′ c − b′ a − ĉ
= .
a′ − c b′ − a c′ − b a′ − c b′ − a ĉ − b

After cancelation of like terms, we have

( ) ( )
a − c′ a − ĉ
= = r, where r is a scalar ratio.
c′ − b ĉ − b

Solving these equations for c′ and ĉ in terms of a, b, and r we get

a + rb a + rb
c′ = and ĉ = .
r+1 r+1
Therefore, ĉ must equal c′ , and the three cevian lines are concurrent.
 72

3.8.4 Menelaus’ Theorem

Another close companion to Ceva’s theorem is Menelaus’ theorem. This

theorem dates back to ancient Greece, where Menelaus of Alexandria was an

astronomer who lived in the first century A.D. [Eve72, p. 63]. A menelaus point is a

point lying on a side of a triangle but not coinciding with a vertex of the triangle.

Menelaus’ Theorem examines the directed lengths of the sides of a triangle that has

been cut by a transversal through three menelaus points. It uses the product of

ratios to determine if three menelaus points are collinear (see Figure 3.19). Note

that it is common in the literature for the terms point and vector to be

interchanged. A vector with its initial point at the origin determines a unique point

at its end point, e.g., points a, b, c can be uniquely represented by vectors a, b, c.

a' c'
b'
b
a

Figure 3.19: Menelaus’ Theorem.

Theorem 3.8.12 (Menelaus’ Theorem).

Three menelaus points a′ , b′ , c′ on the sides bc, ca, ab of an ordinary triangle

△abc are collinear if and only if

( )( )( )
b − a′ c − b′ a − c′
= −1.
a′ − c b′ − a c′ − b
 73

Proof of Menelaus’ Theorem.

(⇒) Suppose that three menelaus points a′ , b′ , c′ on sides bc, ca, ab of an

ordinary triangle △abc are collinear. Let A, B1 , B2 , B3 , B4 , C be bivectors whose

norms are equal to the areas of the smaller triangles (see Figure 3.20). Consider the

A c'
a'
b' B4
B3 C
B1
B2 b

Figure 3.20: Proof of Menelaus’ Theorem.

ratio of the side starting at b, through a′ , to c:

b − a′ 1 B2 + B3 2 B3 + B4 3 B2 − B4
= = = . (3.39)
a −c
′ A + B1 + B4 A B1 + B4
In Eq. (3.39), step (1) follows from the extended version of Lemma 3.8.7 where a is

the common vertex. Step (2) follows from Lemma 3.8.7 where b′ is the common
 74

vertex. Step (3) uses Lemma 3.8.5 to take the difference of the first two steps.

Consider the ratio of the side starting at c, through b′ , to a:

c − b′ 1 A + B3 + B4 2 A 3 B3 + B4
= = = . (3.40)
b −a
′ B1 + B2 B1 + B4 B2 − B4

In Eq. (3.40), step (1) follows from Lemma 3.8.7 where b is the common vertex.

Step (2) follows from Lemma 3.8.7 where a′ is the common vertex. Step (3) uses

Lemma 3.8.5 to take the difference of the first two steps.

Consider the ratio of the side starting at a, to c′ , and back to b:

a − c′ 1 B1 + B2 + B3 + B4 + C 2 B2 + B3 + C 3 B1 + B4
=− =− =− . (3.41)
c′ − b B3 + B4 + C C B3 + B4

In Eq. (3.41), step (1) follows from Lemma 3.8.7 where b′ is the common vertex.

Step (2) follows from Lemma 3.8.7 where a′ is the common vertex. Step (3) uses

Lemma 3.8.5 to take the difference of the first two steps. Notice that a negative sign

was introduced because the menelaus point c′ cuts the side ab exterior to the

triangle.

We now take the product of the right-most terms of Eq. (3.39), (3.40), and

(3.41) to get
( )( )( )
B2 − B4 B3 + B4 B1 + B4
− = −1.
B1 + B4 B2 − B4 B3 + B4
Therefore,
( )( )( )
b − a′ c − b′ a − c′
= −1. (3.42)
a′ − c b′ − a c′ − b
(⇐) Suppose that three menelaus points a′ , b′ , ĉ on sides bc, ca, ab of triangle

abc satisfy the equation

( )( )( )
b − a′ c − b′ a − ĉ
= −1. (3.43)
a′ − c b′ − a ĉ − b

Let c′ be a menelaus point on side ab that is collinear with a′ and b′ (see

Figure 3.21). In the first part of this proof, we showed that for collinear menelaus
 75

a' c'
b' ^
c
b
a

Figure 3.21: Converse of Menelaus’ Theorem.

points a′ , b′ , c′ we have Eq. (3.42). If we equate the left-hand sides of Eq. (3.42) and

(3.43) we get
( )( )( ) ( )( )( )
b − a′ c − b′ a − c′ b − a′ c − b′ a − ĉ
= .
a′ − c b′ − a c′ − b a′ − c b′ − a ĉ − b

After cancelation of like terms, we have

( ) ( )
a − c′ a − ĉ
= = r, where r is a scalar ratio.
c′ − b ĉ − b

Solving these equations for c′ and ĉ in terms of a, b, and r we get

a + rb a + rb
c′ = and ĉ = .
r+1 r+1

Therefore, ĉ must equal c′ and the three menelaus points are collinear.

3.8.5 Desargues’ Two-Triangle Theorem

Gérard Desargues was a French mathematician and engineer who is considered

one of the founders of projective geometry. In 1636 Desargues published a work on

 76

perspective that included his now famous two-triangle theorem [Eve72, p. 70]. This

theorem has become basic in the present-day theory of projective geometry. In 1639

Desargues published a book, Brouillon Projet d’une atteinte aux événemens des

rencontres du Cône avec un Plan, that marked the first advance in synthetic

geometry since the time of the ancient Greeks [Eve72, p. 59].

Two triangles △ABC and △A′ B ′ C ′ are said to be copolar if AA′ , BB ′ , CC ′

are concurrent at the point S; they are said to be coaxial if the points of

intersection P (of BC and B ′ C ′ ), Q (of CA and C ′ A′ ), R (of AB and A′ B ′ ) are

collinear (see Figure 3.22).

C
B
A

R Q P

A'
B'

C'

Figure 3.22: Desargues’ Two-Triangle Theorem.

Theorem 3.8.13 (Desargues’ Two-Triangle Theorem).

Two ordinary triangles are copolar if and only if they are coaxial.

Proof of Desargues’ Two-Triangle Theorem.

Let the vertices of triangles △ABC and △A′ B ′ C ′ be represented by the

vectors a, b, c and a′ , b′ , c′ respectively. Let the points P, Q, R, S be represented by

the vectors p, q, r, s respectively (see Figure 3.23).

 77

c
b
a

r q p

a'
b'

c'

Figure 3.23: Proof of Desargues’ Two-Triangle Theorem.

(⇒) Suppose that triangles △abc and △a′ b′ c′ are copolar at point s. Then

by Eq. (3.31a) and (3.31d) of Lemma 3.8.9, there exist scalars α, α′ , β, β ′ , γ, γ ′ such

that

s = αa + α′ a′ = βb + β ′ b′ = γc + γ ′ c′ , (3.44a)

α + α′ = β + β ′ = γ + γ ′ = 1. (3.44b)

Then from Eqs. (3.44a and b) we get

βb − γc = −(β ′ b′ − γ ′ c′ ),

and β − γ = −(β ′ − γ ′ ).

Thus, if β − γ ̸= 0, we have

βb − γc β ′ b′ − γ ′ c′
= = p. (3.45a)
β−γ β′ − γ′

Where p is the intersection of the line from c to b and the line from c′ to b′ . The

left-hand ratio is a point on the line through the points b and c because it can be

expressed as a linear combination of b and c where the coefficients sum to 1

(Lemma 3.8.9). Likewise, the right-hand ratio is a point on the line through the
 78

points b′ and c′ because it can be expressed as a linear combination of b′ and c′

where the coefficients sum to 1 (Lemma 3.8.9). Thus, these ratios equal p, the point

of intersection of these two lines.

Following similar derivations, we find expressions for q, the intersection of the

line from c to a and the line from c′ to a′ , and r, the intersection of the line from b

to a and the line from b′ to a′ ,

γc − αa γ ′ c ′ − α ′ a′
= = q, (3.45b)
γ−α γ ′ − α′

αa − βb α′ a′ − β ′ b′
= = r. (3.45c)
α−β α′ − β ′
Rearranging Eqs. (3.45a, b, and c), we get

(β − γ)p = βb − γc, (3.46a)

(γ − α)q = γc − αa, (3.46b)

(α − β)r = αa − βb. (3.46c)

Adding the left and right sides of Eqs. (3.46a, b, and c), we get

(β − γ)p + (γ − α)q + (α − β)r = γc − αa + βb − γc + αa − βb,

(β − γ)p + (γ − α)q + (α − β)r = 0,

where (β − γ) + (γ − α) + (α − β) = 0.

Thus, by Lemma 3.8.10, the points p, q, r are collinear, provided not all of

β − γ, γ − α, and α − β are zero.

If β − γ = 0, then the lines from b to c and b′ to c′ are parallel and p would

be a point at infinity. If β − γ ̸= 0, γ − α ̸= 0, and α − β ̸= 0, then we are assured

that each pair of lines intersects in an ordinary point and that we are dealing with

an ordinary triangle. This theorem can be extended to include points at infinity.

 79

(⇐) Suppose that triangles △abc and △a′ b′ c′ are coaxial through the points

p, q, r, and let s be the point of intersection of the line through a′ and a and the

line through b′ and b. Then the triangles △aa′ q and △bb′ p are copolar through r.

Therefore, by the first part of this proof, they are coaxial through c′ , c, s. Thus, the

coaxial triangles △abc and △a′ b′ c′ are copolar.

 80

CHAPTER 4

THE CONFORMAL MODEL

The conformal model is an alternative model to the Euclidean model for

solving three-dimensional geometric problems. The creation of the conformal model

involves embedding Euclidean space in a higher dimensional space. Two extra

dimensions are added to the Euclidean space being modeled. These extra

dimensions have particular properties such that linear subspaces in conformal space

represent geometric entities of interest in Euclidean space.

In this chapter we develop the conformal model and show how it can be used

to represent points, lines, circles, planes, and spheres.1 The conformal model can

support points and lines at infinity. It also supports the conformal transformations

of translation, rotation, reflection, inversion, and scaling using rotors.

4.1 Development of the Conformal Model

When first introduced to the conformal model, one often wonders “why go to

all this trouble?” It turns out that this model has some real advantages in its ability

to model geometric objects and express transformations. The reader can be assured

that any effort to understand this model will be rewarded in the end.

Suppose we have a line in Euclidean 1-space with a unit basis vector e1

(see Figure 4.1). Any point on this line can be expressed as a vector, a scalar
1
The organization of this chapter follows that of Vince [Vin08, Ch. 11] and
Doran & Lasenby [DL03, Ch. 10].
 81

e1
-3 -2 -1 0 1 2 3
Figure 4.1: Line in Euclidean 1-space

multiple of e1 , viz.,

x = xe1 .

4.1.1 Add a Dimension for Stereographic Projection

We can use stereographic projection to map points from a one-dimensional line

onto a unit circle in two-dimensional space. To do this, we introduce a new

dimension perpendicular to e1 with unit basis vector e. Next, we construct a unit

circle centered at the origin 0 of e1 and e. To map any point x on the e1 line to a

point x′ on the unit circle, we construct a line from the point x to the end of the

vector e. Wherever this line cuts the unit circle is the point x′ (see Figure 4.2).

0 e1 x

x'
e
Figure 4.2: Stereographic projection

Stereographic projection allows us to work with points at infinity. The point at

infinity on the e1 line is mapped to the point e on the unit circle. The stereographic
 82

projection x′ can be expressed as a vector in terms of its components e1 and e.

2x 1 − x2
x′ = e1 − e, where x′ · x′ = x′ 2 = 1. (4.1)
1 + x2 1 + x2

Equation (4.1) can be derived as follows:

Let x = xe1 and x′ = e + λ(x − e), where λ ∈ R. Note that

e · e = 1, x′ · x′ = 1, e · x = 0, x · x = x2 .

When we take the inner product of x′ with itself, we get

x′ · x′ = (e + λ(x − e)) · (e + λ(x − e))

x′ · x′ = e · e + 2λe · (x − e) + λ2 (x − e) · (x − e)

x′ · x′ = e · e + 2λ(e · x − e · e) + λ2 (x · x − 2e · x + e · e)

1 = 1 + 2λ(0 − 1) + λ2 (x2 − 0 + 1)

1 = 1 − 2λ + λ2 (1 + x2 )

0 = λ2 (1 + x2 ) − 2λ. (4.2)

Solving Eq. (4.2) for λ, we get

2
λ = 0 or λ = .
1 + x2

If λ = 0, then x′ = e and x would be the point at infinity. For all other points

x on the e1 axis, the equation for x′ would be

x′ = e + λ(x − e)
2
x′ = e + (x − e)
1 + x2

′ e(1 + x2 ) + 2(x − e)
x =
1 + x2

e + x2 e + 2x − 2e
x′ =
1 + x2
 83

2x + x2 e − e
x′ =
1 + x2

2x 1 − x2
x′ = e1 − e.
1 + x2 1 + x2

Thus, we have derived Eq. (4.1).

Using Eq. (4.1), any point x on the e1 line can be mapped to a point x′ on the

unit circle in the e1 , e plane, e.g.,

x = 0e1 7→ x′ = −e,

x = 1e1 7→ x′ = e1 ,

x = ∞e1 7→ x′ = e.

4.1.2 Add a Dimension for Homogeneous Coordinates

Before we go any further, we need to introduce the concept of signature. So

far we have been working with Inner-Product Spaces, i.e., vector spaces equipped

with positive definite, symmetric bilinear forms over the real numbers. At this

point, we find it useful to relax our assumptions a little bit by dispensing with the

assumption that the form under consideration is positive definite. Although we still

use the term “inner product,” along with the “dot product” notation to indicate the

value of our symmetric bilinear form, we now have the possibility that a vector v

satisfies v · v < 0 or v · v = 0.

Definition 4.1.1 (Signature).

If an n-dimensional vector space has p independent unit vectors satisfying

ei · ei = 1, and q independent unit vectors satisfying ei · ei = −1, then it is

customary to to say that the space has p “positive dimensions” and q “negative
 84

dimensions” (with n = p + q). Note that we assume that our symmetric bilinear

form is non-degenerate, thus avoiding “zero dimensions.” Instead of Rn , we then

write Rp,q and call (p, q) the signature of the space. An n-dimensional Euclidean

space is then written as Rn,0 , the representational space for the conformal model as

Rn+1,1 , and an n-dimensional Minkowski space as Rn−1,1 [DFM07, p. 586].

Next, we add another dimension to the stereographic projection described in

the previous section. This new dimension is identified by the unit vector ē, where ē

is perpendicular to both e1 and e. We give this new vector a negative signature,

meaning ē · ē = ē2 = −1 (ē2 = ēē = ē · ē + ē ∧ ē = ē · ē + 0 = ē · ē = |ē|2 = −1).

Finally, we lift the unit circle off the e1 e-plane so that its center lies one unit above

the origin at the end of the ē vector (see Figure 4.3).

x'
e x

e1
e
Figure 4.3: Addition of the ē dimension

The addition of this new dimension provides us with another degree of

freedom that we use in our expression for mapping a real point into the conformal
 85

model. Thus, we have

X = x′ + ē,
2x 1 − x2
X= e 1 − e + ē, (4.3)
1 + x2 1 + x2
where we have substituted Eq. (4.1) for x′ .

Note that
( )2
2x 1 − x2
2
X = e1 − e + ē
1 + x2 1 + x2
( )( )
2x 1 − x2 2x 1 − x2
= e1 − e + ē e1 − e + ē
1 + x2 1 + x2 1 + x2 1 + x2


 4x2 2x (1 − x2 ) 2x

 e2 − e1 e + e1 ē

 (1 + x2 )2 1 2
(1 + x )2 1 + x2



(1 − x2 ) 2x (1 − x2 )2 2 1 − x2
= − ee1 + e − eē

 (1 + x2 )2 (1 + x2 )2 1 + x2





 2x 1 − x2
 + ēe1 − ēe + ē2
1 + x2 1 + x2


 4x2 2x (1 − x2 ) 2x

 1 − e e + e1 ē

 (1 + x2 )2 (1 + x2 )2
1
1 + x2



2x (1 − x2 ) (1 − x2 )2 1 − x2
= + e1 e + 1 − eē

 (1 + x2 )2 (1 + x2 )2 1 + x2





 2x 1 − x2
 − e1 ē + eē − 1
1 + x2 1 + x2

4x2 (1 − x2 )2
= + −1
(1 + x2 )2 (1 + x2 )2

4x2 + (1 − x2 )(1 − x2 ) − (1 + x2 )(1 + x2 )

=
(1 + x2 )2

4x2 + 1 − 2x2 + x4 − 1 − 2x2 − x4

=
(1 + x2 )2

= 0.
 86

We see that the negative signature of ē ensures that X is always a null vector, i.e.,

X2 = XX = X · X + X ∧ X = |X|2 + 0 = |X|2 = 0.

Definition 4.1.2 (Null vector).

A null vector is a vector v whose inner product with itself is zero, i.e.,

v · v = 0. Null vectors show up when we work with spaces of mixed signature (both

positive and negative), as with the conformal model, for example. Null vectors may

be nonzero. However, the zero vector is a null vector. Null vectors are not invertible.

The equation X2 = 0 is homogeneous: if it is satisfied for X, it is satisfied for

any nonzero scalar multiple of X (e.g., σX, where σ ∈ R). Any nonzero scalar

multiple of the null vector X is also a null vector that lies on the line that passes

through the same point on the unit circle as X does. The extra degree of freedom

we get by adding the ē dimension has created what is called a homogeneous

coordinate system where we consider all such nonzero scalar multiples of X to

represent the same point on the unit circle. Thus, all nonzero scalar multiples of X

form an equivalence class.

The locus of all lines that pass from the origin through the unit circle is called

the null cone.

We can simplify Eq. (4.3) by multiplying each term by the scalar σ(1 + x2 ),

giving,

X = σ(2xe1 − (1 − x2 )e + (1 + x2 )ē), (4.4)

where σ is an arbitrary nonzero scalar. Thus, Eq. (4.4) defines an equivalence class

of null vectors X ∈ R2,1 used to represent a single vector x = xe1 ∈ R1 .

The vectors e and ē have extended the space R1 to R2,1 . The basis elements

for this new conformal space are {e1 , e, ē}, where e21 = 1, e2 = 1, and ē2 = −1.
 87

The inner, outer, and geometric products of {e1 , e, ē} are summarized in

Table 4.1.

Table 4.1: Multiplication tables for e1 , e, ē.

· e1 e ē ∧ e1 e ē GP e1 e ē
e1 1 0 0 e1 0 e1 ∧ e e1 ∧ ē e1 1 e1 e e1 ē
e 0 1 0 e −e1 ∧ e 0 e ∧ ē e −e1 e 1 eē
ē 0 0 −1 ē −e1 ∧ ē −e ∧ ē 0 ē −e1 ē −eē −1

4.1.3 Change Variables from e, ē to n, n

Here is a recap of what we have done so far:

(1) Stereographic Projection - We added the dimension e perpendicular to the

Euclidean dimension e1 . This allows us to map all points on the Euclidean

line e1 —including the point at infinity—onto the unit circle.

(2) Homogeneous Projection - We added the dimension ē perpendicular to the

Euclidean dimension e1 and perpendicular to e. We then raised the unit

circle off of the e1 e-plane, placing its center on the end point of the ē

vector. Any point on the unit circle can now be identified by an equivalence

class of nonzero vectors that lie on a line that passes from the origin

through that point on the unit circle. Thus, an equivalence class of vectors

all refer to the same point on the Euclidean line e1 vis-à-vis the point on the

unit circle. With the choice of the negative signature for ē, we insure that

all nonzero vectors X ∈ R2,1 lying on a line that passes from the origin

through the unit circle, and, therefore, representing points on the Euclidean

line e1 , are null vectors, i.e., X2 = 0.

 88

Unfortunately, the basis vectors e and ē that we used to get additional

dimensions have no geometric interpretation. If we use a change of variables to

replace these two basis vectors with two null vectors, then it is possible to assign a

geometric interpretation to these null vectors. We can achieve this by replacing our

basis vectors e and ē with the null vectors n and n defined as:

n = e + ē , n = e − ē . (4.5)

Note that just like e and ē, both n and n are perpendicular to the Euclidean

space spanned by e1 , i.e., n · e1 = 0 and n · e1 = 0. These vectors satisfy the

following relations:

n2 = n2 = 0 , n·n=2 , n ∧ n = −2eē . (4.6)

The inner, outer, and geometric products of {e1 , n, n} are summarized in

Table 4.2.

Table 4.2: Multiplication tables for e1 , n, n.

· e1 n n ∧ e1 n n GP e1 n n
e1 1 0 0 e1 0 e1 ∧ n e1 ∧ n e1 1 e1 n e1 n
n 0 0 2 n −e1 ∧ n 0 −2eē n −e1 n 0 2 − 2eē
n 0 2 0 n −e1 ∧ n 2eē 0 n −e1 n 2 + 2eē 0

Substituting Eqs. (4.5) into Eq. (4.4), we get

X = σ(2xe1 + x2 n − n), (4.7)

where σ is an arbitrary nonzero scalar.

This configuration can be visualized as shown in Figure 4.4. Points on the line

determined by e1 are mapped to null vectors on the null cone using Eq. (4.7).
 89

e n=e+e

e1
e

n=e-e

Figure 4.4: The conformal model of a line.

In the sections to follow, we have need of an identity related to n and n. This

identity is now derived.

Identity 4.1.3 (nnn = 4n).

nnn = (e + ē)(e − ē)(e + ē)

= (e2 − eē + ēe − ē2 )(e + ē)

= (1 − eē − eē + 1)(e + ē)

= (2 − 2eē)(e + ē)

= 2e + 2ē − 2eēe − 2eē2

= 2e + 2ē + 2e2 ē + 2e

= 2e + 2ē + 2ē + 2e

= 4e + 4ē = 4(e + ē) = 4n.

 90

∴ nnn = 4n . (4.8)

4.1.4 Generalization to Euclidean 3-Space

The preceding discussion, including Eq. (4.7), shows how to map a

one-dimensional Euclidean space into a three-dimensional conformal model. It is

easy to extend these findings to three-dimensional Euclidean space. First, for the

one-dimensional case, we let the vector x be defined as x = xe1 where

x2 = x · x = x2 . Then Eq. (4.7) becomes

X = σ(2x + x2 n − n), (4.9)

where σ is an arbitrary nonzero scalar.

Next, we replace x = xe1 with x = x1 e1 + x2 e2 + x3 e3 , and use Eq. (4.9) to

define X, the conformal mapping of x ∈ R3 into R4,1 . Note that e and ē are

perpendicular to all three Euclidean basis vectors e1 , e2 , e3 and that X lies on a line

that intersects a unit hypersphere.

Because σ can be any arbitrary nonzero scalar, we choose the value of one-half

as being convenient for expressing the standard form of the conformal image of

x = x1 e1 + x2 e2 + x3 e3 ,

Standard Form: X = 21 (2x + x2 n − n) .2 (4.10)

If we are given an arbitrary null vector X in conformal space, it is frequently

convenient to convert it into the standard form of Eq. (4.10) so that the Euclidean

coordinates can be easily extracted. If X is in standard form, then

−X · n = − 12 (2x + x2 n − n) · n = 1. (4.11)
2
Different authors use different standard forms of the conformal equation. See Appendix A.
 91

An arbitrary null vector X can be normalized by dividing it by −X · n.

b = X
Normalization: X . (4.12)
−X · n
For example, given the vector

X = 48e1 + 20e2 + 16e3 + 17e + 57ē, where X2 = X · X = 0,

we have

−X · n = −(48e1 + 20e2 + 16e3 + 17e + 57ē) · n

= −(48e1 + 20e2 + 16e3 + 17e + 57ē) · (e + ē)

= −(17 − 57)

= 40.

Therefore the normalized form of X is

b = X
X
−X · n

48e1 + 20e2 + 16e3 + 17e + 57ē

=
40
48 20 16 17 57
= e
40 1
+ e
40 2
+ e
40 3
+ 40
e + 40
ē

= 65 e1 + 12 e2 + 25 e3 + 17
40
e + 57
40
ē

= 56 e1 + 12 e2 + 25 e3 + 37
40
n − 21 n.

And from this formula, the Euclidean coordinates of the point x can be read as

x = ( 56 , 12 , 52 ).

4.1.5 Geometric Interpretation of n and n

We would like to be able to give a geometric interpretation to n and n in

Eq. (4.10). If x is at the origin of the Euclidean space, i.e., x = 0, then by

 92

Eq. (4.10), X = − 21 n, which implies that

n represents the origin .

To interpret n, we first need to take the inner product X · n:

X · n = 12 (2x + x2 n − n) · n

= 12 (2x · n + x2 (n · n) − n · n)

= 12 (0 + 2x2 + 0)

= x2 .

Multiplying the homogeneous vector X by a scalar doesn’t change the Euclidean

point it represents. We multiply X by the scalar σ = 2/(X · n) = 2/x2 and take the

limit as x approaches infinity.

( )
2
lim σX = lim X
x→∞ x→∞ X·n
( )
2
= lim 2
1
2
(2x + x2 n − n)
x→∞ x
( )
2x − n
= lim n +
x→∞ x2

= n. (4.13)

From Eq. (4.13) we see that

n represents the point at infinity .

4.2 Conformal Transformations

We now examine conformal transformations using the conformal model.

Conformal transformations are defined as transformations that leave angles

 93

invariant. Conformal transformations of R3 form a group under composition—the

conformal group—the main elements of which are translations, rotations, dilations,

and inversions. Translations and rotations form a subgroup of the larger conformal

group called the Euclidean group. The Euclidean group consists of transformations

that leave the distance between points invariant.

Definition 4.2.1 (Versor).

A versor is a geometric product of invertible vectors. Examples of versors are

u, uv, uvw, etc. A versor is called even or odd depending upon whether it has an

even or odd number of vector terms.

In the conformal model, conformal transformations are implemented with a

sandwiching operation using a versor, i.e.,

even versors: X 7→ VXV−1 ,

odd versors: X 7→ −VXV−1 ,

where V is a versor and X = 12 (2x + x2 n − n) is a conformal space representation of

a vector x ∈ R3 . In this section we drop the 1

2
from the standard form and use

X = 2x + x2 n − n when deriving the transformation versors. This makes the

formulas a little easier to follow and does not impact the final results.

We use the following approach in our study of conformal transformations:

• Present the versor V used to generate the transformation.

• Derive how the versor transforms x, n, and n, i.e.,

x 7→ ±VxV−1 , n 7→ ±VnV−1 , and n 7→ ±VnV−1 .

• Use the above information and the distributive property to derive how the

versor V transforms X = 2x + x2 n − n, the conformal model representation

 94

of x, i.e.,

±VXV−1 = ±(V (2x + x2 n − n) V−1 )

= ±(2VxV−1 + x2 VnV−1 − VnV−1 ).

4.2.1 Translation

Consider the operation of translating an object in R3 in a direction and a

distance indicated by the vector a (see Figure 4.5).

0
p+a
a

p
translated

original

Figure 4.5: Translation using vector a

Each point on the object is translated by the vector a. If the location of a

point on the object is indicated by the vector p, then p + a locates the same point

on the translated object.

If fa is a function that translates objects using the vector a then,

fa : R3 → R3 is defined by fa (x) = x + a.
 95

Definition 4.2.2 (Rotor).

A rotor is a geometric product of an even number of invertible unit vectors.

∼ ∼
Rotors have the useful property that R−1 = R (where R is the reversion of R).

Examples of rotors are uv, vw, uw, if |u| = |v| = |w| = 1. Rotors can always be

expressed as exponentials, e.g., uv = eI(2) θ , where I(2) is the unit pseudoscalar in the

uv-plane and θ is the angle from u to v. Rotors are even unit versors.

To implement translation in the conformal model, we use the rotor

R = Ta = ena/2 ,

where a ∈ R3 is the translation vector and n · a = 0.

The term na = n · a + n ∧ a = 0 + n ∧ a = n ∧ a is a bivector, and

(na)2 = nana = −nnaa = −0a2 = 0, thus na is a null bivector.

Consider the series expansion for the exponential,

x2 x3
ex = 1 + x + + + ··· .
2! 3!

Replacing x with na/2, we get

na ( na )2 ( na )3
ena/2 = 1 + + 2 + 2 + ··· .
2 2! 3!

Because na is a null bivector, all terms with power greater than one become zero

and we have

Translation na ∼ na
Ta = ena/2 = 1 + and Ta = e−na/2 = 1 − . (4.14)
rotors: 2 2

For clarity, we apply these translation rotors separately to the vectors x, n,

and n. We then use the distributive property,

∼ ∼ ∼ ∼ ∼
RXR = R (2x + x2 n − n) R = R2xR + Rx2 nR − RnR. (4.15)
 96
∼
to show that the sandwich operation Ta XTa corresponds to a translation of the

vector x in R3 .

For x ∈ R3 we have

∼ na na
Ta xTa = (1 + ) x (1 − )
2 2

nax na
= (x + )(1 − )
2 2

xna nax naxna

=x− + −
2 2 4

nxa nax n2 axa

=x+ + −
2 2 4
( )
xa + ax
=x+n − 0 = x + n(a · x)
2
∼
Ta xTa = x + (a · x)n . (4.16)

For n ∈ R4,1 , the point at infinity, we have

∼ na na
Ta nTa = (1 + ) n (1 − )
2 2

nan na
= (n + )(1 − )
2 2

n2 a na
= (n − )(1 − )
2 2

na
= (n − 0)(1 − )
2

na
= n (1 − )
2

n2 a
=n− =n−0=n
2
∼
Ta nTa = n . (4.17)
 97

For n ∈ R4,1 , the point at the origin, we have

∼ na na
Ta nTa = (1 + ) n (1 − )
2 2

nan na
= (n + )(1 − )
2 2

nna nan nanna

=n− + −
2 2 4

nna nna nnna2

=n− − −
2 2 4
( )
5 nn + nn 4na2
=n− a−
2 4

= n − (n · n) a − na2 = n − 2a − a2 n
∼
Ta nTa = n − 2a − a2 n . (4.18)

Where step (5) uses Identity 4.1.3, viz., nnn = 4n.

We now return to our equation for conformal mapping X = 2x + x2 n − n. Let

F : R3 → R4,1 , where F (x) = X = 2x + x2 n − n. Then,

∼ ∼
Ta F (x)Ta = Ta (2x + x2 n − n)Ta
∼ ∼ ∼
= Ta 2xTa + Ta x2 nTa − Ta nTa
3
= (2x + 2(a · x)n) + (x2 n) − (n − 2a − a2 n)

= 2(x + a) + (ax + xa)n + x2 n + a2 n − n

= 2(x + a) + (x2 + ax + xa + a2 ) n − n

= 2(x + a) + (x + a)2 n − n.
∼
Conformal Translation: Ta F (x)Ta = 2(x + a) + (x + a)2 n − n . (4.19)

Where step (3) comes from Eqs. (4.16), (4.17), and (4.18). Note that
∼
Ta F (x)Ta = F (x + a),
 98

which performs the conformal transformation corresponding to the translation

x 7→ x + a. In conformal space, translations are handled as rotations in the na

plane.

4.2.2 Rotation

In Section 3.7 we learned that rotating a vector x ∈ R3 about the origin, in

the I(2) -plane, can be accomplished by a sandwiching operation

∼
RxR.

The rotor R was defined as

∼
Rotation operators: R = eI(2) θ/2 and R = e−I(2) θ/2 . (4.20)

where θ is the angle of rotation and I(2) is a unit pseudoscalar for the plane of

rotation. The formula for rotating the vector x is

∼
RxR = eI(2) θ/2 x e−I(2) θ/2 . (4.21)

We now see what happens when we apply this rotor to the vector n ∈ R4,1 ,

the point at infinity. Without loss of generality we can assume that I(2) = b1 b2

where b1 and b2 are orthogonal unit vectors in the I(2) -plane. Then we have

nI(2) = nb1 b2

nI(2) = b1 b2 n

nI(2) = I(2) n. (4.22)

 99

and

I(2) nI(2) = I(2) I(2) n

I(2) nI(2) = −n. (4.23)

Then expanding Eq. (4.21) and using the results of Eqs. (4.22) and (4.23) we have,

∼
RnR = eI(2) θ/2 n e−I(2) θ/2

= (cos 2θ + I(2) sin 2θ ) n (cos 2θ − I(2) sin 2θ )

= n cos2 2θ − nI(2) cos 2θ sin 2θ + I(2) n sin 2θ cos 2θ − I(2) nI(2) sin2 θ
2

= n cos2 2θ − I(2) n cos 2θ sin 2θ + I(2) n cos 2θ sin 2θ + n sin2 θ

2

= n (cos2 2θ + sin2 2θ )

= n.

∼
RnR = n . (4.24)

If we replace n with n, in the above calculations, all results turn out the same.

Thus, we get
∼
RnR = n . (4.25)

The rotor R commutes with both n and n because it is an even element of R4,1 .

Now we apply the rotor R to the conformal space representation X of the

Euclidean vector x,

X = F (x) = 2x + x2 n − n
∼ ∼
RF (x)R = R(2x + x2 n − n)R
∼ ∼ ∼
= R2xR + Rx2 nR − RnR
∼
= 2RxR + x2 n − n.
 100

∼ ∼
Conformal Rotation: RF (x)R = 2RxR + x2 n − n , (4.26)
∼
where RxR is as shown in Eq. (4.21).

Suppose that we wish to rotate about a point other than the origin. This can

be achieved by translating that point to the origin, doing the rotation and then

translating back to the original point. In terms of X = F (x), the result is

∼ ∼ ∼
X 7→ Ta (R(T−a XT−a )R)Ta .

If we combine operators, we get

Generator for a rotation ( na ) I(2) θ/2 ( an )

R′ = Ta RT−a = 1 + e 1+ .
about a point a: 2 2

The combined transformation becomes

∼′
X 7→ R′ XR .

The conformal model has freed us from treating the origin as a special point.

Rotations about any point are handled in the same manner, and are still generated

by a bivector.

4.2.3 Inversion

An inversion in the origin in R3 consists of the map

x
x 7→ .
x2

The inverse of an invertible vector x, as shown in Section 2.6.2, is a vector

with the same attitude and orientation as x, but with norm 1/|x|. We use the
 101

b-1

b
a=a-1
-2 -1 0 1 2
c-1

Figure 4.6: Inversion in the origin.

notation x−1 to denote the inverse of x. If |x| is less than 1, then |x−1 | is greater

than one and vice versa. If |x| = 1, then |x−1 | = 1 (see Figure 4.6).

If we take the equation F (x) = 2x + x2 n − n and replace x with x−1 , we get

F (x−1 ) = 2x−1 + x−2 n − n,

1
= (2x + n − x2 n).
x2

But we are using the homogeneous representation for conformal space where
1
nonzero scalar multiples are arbitrary, so we can ignore the x2
factor and obtain

F (x−1 ) = 2x − x2 n + n.

We see that an inversion in the origin in R3 consists of the following maps in

conformal space,

n 7→ −n, and n 7→ −n.

This map can be achieved by a reflection in e, where the versor V = V−1 = e. For
 102

n we have

−VnV−1 = −ene

= −e (e + ē) e

= −eee − eēe

= −e + ē = −(e − ē) = −n.

−VnV−1 = −n . (4.27)

And for n we have

−VnV−1 = −ene

= −e (e − ē) e

= −eee + eēe

= −e − ē, = −(e + ē) = −n.

−VnV−1 = −n . (4.28)

If we apply this reflection to the vector x ∈ R3 , we get

−VxV−1 = −exe = eex = 1x = x.

−VxV−1 = x . (4.29)

We apply this reflection to the conformal space representation X of the

Euclidean vector x.

X = F (x) = 2x + x2 n − n,

−VF (x)V−1 = −e (2x + x2 n − n) e

= −e2xe − ex2 ne − (−ene)

= 2x − x2 n + n.
 103

Conformal Inversion: −VF (x)V−1 = 2x − x2 n + n . (4.30)

Note that

−VF (x)V−1 = x2 F (x−1 ).

Thus, a reflection in e corresponds to an inversion in the original Euclidean space.

To find the generator of an inversion in an arbitrary point a ∈ R3 , we

translate to the origin, invert, and translate back to the original point. In terms of

X = F (x), the result is

∼ ∼
X 7→ Ta (V(T−a XT−a )V−1 )Ta .

If we combine operators, the resulting generator is

(na ) ( an )
′
V = Ta VT−a = 1+ e 1+
2 2

ean nae naean

=e+ + +
2 2 4
( )
en + ne nea2 n
=e− a−
2 4

a2 nen
= e − (e · n) a −
4
[ ( ) ]
a2 n+n
= e − [e · (e + ē)] a − n n
4 2
[ ]
a2 nnn + nnn
= e − [e · e + e · ē] a −
4 2
[ ]
7 a2 0 + 4n
= e − [1 + 0] a −
4 2

a2
=e−a− n.
2
 104

Generator for an a2
V′ = Ta VT−a = e − a − n . (4.31)
inversion in a point a: 2

Where step (7) uses Identity 4.1.3, viz., nnn = 4n. The combined transform

becomes

X 7→ V′ XV′−1 .

4.2.4 Dilation

We use the term dilation to refer to either the contraction or expansion of

geometric objects. When we perform a dilation with respect to the origin, all

vectors from the origin are multiplied by the same scalar constant (see Figure 4.7).

e-αp
p

0 0

original
dilated

(a) Original image. (b) Dilated image.

Figure 4.7: Dilation with respect to the origin.

Although the size of the object has changed, all the angles remain the same.

Thus, dilation is a conformal transformation. If the absolute value of the scalar

 105

multiple is less than one, then we are shrinking the object. If it is greater than one,

then we are stretching the object.

We use the exponential e−α where α ∈ R to represent the scalar multiple. In

this way the scalar multiple is always a positive number. When α is positive, we

have a contraction. When α is negative we have an expansion. A dilation of the

vector x relative to the origin in R3 is expressed as

x 7→ e−α x.

If we take the equation F (x) = 2x + x2 n − n and replace x with e−α x, we get

F (e−α x) = 2e−α x + e−2α x2 n − n

= e−α (2x + e−α x2 n − eα n). (4.32)

Once again we can ignore the scalar multiple e−α due to the homogenous

coordinates used in R4,1 . If we compare Eq. (4.32) to the standard form Eq. (4.10),

we see that we are looking for a conformal space dilation operator that maps

x 7→ x, n 7→ e−α n, and n 7→ eα n.

Such an operator does exist and it takes the form of a rotor Dα , where
∼
Dα F (x)Dα = 2x + e−α x2 n − eα n.

The rotor Dα is defined as

Dα = cosh(α/2) + sinh(α/2) N
Dilation
and (4.33)
operators:
∼
Dα = cosh(α/2) − sinh(α/2) N .
 106

where N = eē = 12 (n ∧ n).

We now list several identities involving N that are used in the discussion that

follows.

nN = (e + ē) eē = n. (4.34a)

Nn = eē (e + ē) = −n. (4.34b)

nN = −Nn. (4.34c)

nN = (e − ē) eē = −n. (4.34d)

Nn = eē (e − ē) = n. (4.34e)

nN = −Nn. (4.34f)

N2 = (eē)2 = 1. (4.34g)

Next we apply the dilation operator Dα to the vectors x, n, and n. For

x ∈ R3 , we have

∼
xDα = x (cosh(α/2) − sinh(α/2) N)

= cosh(α/2)x − sinh(α/2) xN

= cosh(α/2)x − sinh(α/2) xeē

= cosh(α/2)x − sinh(α/2) eēx

= (cosh(α/2) − sinh(α/2) N) x
∼
= Dα x.
 107

This allows us to rewrite the transformation equations as follows:

∼ ∼
Dα xDα = Dα Dα x

= (cosh(α/2) + sinh(α/2) N) (cosh(α/2) − sinh(α/2) N) x

= (cosh2 (α/2) − sinh2 (α/2) N2 ) x

4
= (cosh2 (α/2) − sinh2 (α/2)) x
5
= 1x = x.

∼
Dα xDα = x . (4.35)

Where step (4) uses the identity in Eq. (4.34g) and step (5) uses the hyperbolic

identity cosh2 x − sinh2 x = 1.

For n ∈ R4,1 , we have

∼
nDα = n (cosh(α/2) − sinh(α/2) N)

= cosh(α/2)n − sinh(α/2) n
3
= cosh(α/2)n + sinh(α/2) Nn

= (cosh(α/2) + sinh(α/2) N) n

= Dα n.

where step (3) uses the identity in Eq. (4.34c). This allows us to rewrite the

transformation equations as follows:

∼
Dα nDα = D2α n

= (cosh(α/2) + sinh(α/2) N)2 n

= (cosh2 (α/2) + 2 cosh(α/2) sinh(α/2) N + sinh2 (α/2) N2 ) n

= cosh2 (α/2) n + 2 cosh(α/2) sinh(α/2) Nn + sinh2 (α/2) n

 108

5
= cosh2 (α/2) n − 2 cosh(α/2) sinh(α/2) n + sinh2 (α/2) n

= (cosh2 (α/2) − 2 cosh(α/2) sinh(α/2) + sinh2 (α/2)) n

7
= (2 cosh2 (α/2) − (cosh2 (α/2) − sinh2 (α/2)) − 2 cosh(α/2) sinh(α/2)) n
8
= (2 cosh2 (α/2) − 1 − 2 cosh(α/2) sinh(α/2)) n

= ([2 cosh2 (α/2) − 1] − [2 cosh(α/2) sinh(α/2)]) n

= (cosh(α) − sinh(α)) n = e−α n.

10

∼
Dα nDα = e−α n . (4.36)

Where step (5) uses Eq. (4.34b), step (7) adds and subtracts cosh2 (α/2), step (8)

uses the hyperbolic identity cosh2 x − sinh2 x = 1, and step (10) uses the hyperbolic

double-angle formulas.

A similar calculation for n produces the result

∼
Dα nDα = eα n . (4.37)

We now apply this dilation operator to the conformal space representation X

of the Euclidean vector x.

X = F (x) = 2x + x2 n − n,
∼ ∼
Dα F (x)Dα = Dα (2x + x2 n − n) Dα
∼ ∼ ∼
= Dα 2xDα + Dα x2 nDα − Dα nDα

= 2x + x2 e−α n − eα n.

∼
Conformal Dilation: Dα F (x)Dα = 2x + x2 e−α n − eα n . (4.38)

Note that
∼
Dα F (x)Dα = eα F (e−α x).
 109

4.3 Some Background Needed for Representing Geometric Objects

with the Conformal Model

Before we continue our discussion of how we represent geometric objects using

the conformal model, it is necessary that we provide some background information

on duality, null spaces, linear transformations, outermorphisms, and determinants.

4.3.1 Duality

In projective geometry we are introduced to the principle of duality. This

principle highlights the symmetry between geometric objects such as points and

lines.

With duality, if we have a true proposition about “points” and “lines,” we can

obtain another true proposition by interchanging the two words. For example:

Any two distinct points determine one, and only one, line on which
they both lie.

Any two distinct lines determine one, and only one, point through
which they both pass.

In geometric algebra there is also a symmetry between geometric objects of

different grades which allows us to use the principle of duality to extend propositions

about a set of objects to their duals. Scalars, vectors, bivectors, and trivectors all

have duals that depend on the dimension of the space we are working with.
 110

3
Definition 4.3.1 (Dual).

For a given geometric algebra Gn with a unit pseudoscalar I, the dual of a

multivector A, denoted by A* , is

A* = AI−1 . (4.39)

As an example, suppose we are working with R3 , where the unit pseudoscalar

is I = e1 e2 e3 . Then the duals of the vectors e1 , e2 , e3 are bivectors:

e*1 = e1 I−1 = e1 e3 e2 e1 = −e2 e3 ,

e*2 = e2 I−1 = e2 e3 e2 e1 = −e3 e1 ,

e*3 = e3 I−1 = e3 e3 e2 e1 = −e1 e2 .

The duals of the bivectors e2 e3 , e3 e1 , e1 e2 are vectors:

(e2 e3 )* = e2 e3 I−1 = e2 e3 e3 e2 e1 = e1 ,

(e3 e1 )* = e3 e1 I−1 = e3 e1 e3 e2 e1 = e2 ,

(e1 e2 )* = e1 e2 I−1 = e1 e2 e3 e2 e1 = e3 .

The dual of the scalar 1 is the trivector −I.

1* = 1I−1 = 1e3 e2 e1 = −I.

The dual of the trivector e1 e2 e3 = I is the scalar 1.

(e1 e2 e3 )* = e1 e2 e3 I−1 = e1 e2 e3 e3 e2 e1 = 1.

If Bk is a blade of grade k, then the dual of Bk returns the orthogonal

complement in Rn of the subspace represented by Bk . For example, if B ∈ G3 and

B = e1 ∧ e2 , then B* = e3 . The dual, B*k , has a grade of n − k, where n is the grade

of the pseudoscalar for the space. If Bk represents a subspace U , then B*k represents

U ⊥.

3
Different authors use different definitions for dual. See Appendix A.
 111

4.3.2 Inner Product and Outer Product Null Spaces

Before discussing the representations of lines, planes, circles, and spheres in

the conformal model, we need to look at how blades are used to represent geometric

objects.

Typically, in R3 , a non-zero vector a is used to “represent” a plane through

the origin with normal vector a. The plane is the set {x | x · a = 0}. Note that any

non-zero multiple of the vector a, λa, also “represents” the same plane, so the

representative, λa, is a homogeneous coordinate representative of the plane.

In geometric algebra, the vector a is a 1-blade and the plane {x | x · a = 0} is

called the Inner Product Null Space (IPNS) of the blade a. Let the linearly

independent vectors b and c be in the IPNS of the 1-blade a. In geometric algebra,

b ∧ c is a 2-blade (or bivector) and x ∧ b ∧ c = 0 if and only if x = βb + γc. Thus,

the set {x | x ∧ b ∧ c = 0}, called the Outer Product Null Space (OPNS) of the

bivector b ∧ c, is the same plane as the IPNS of the 1-blade a. We may take any

non-zero multiple of b ∧ c, λ(b ∧ c) as a “representative” of the plane.

Because (b ∧ c)* = (b ∧ c)I−1 = λa, we see that the OPNS of the bivector

b ∧ c, and the IPNS of the of the 1-blade a are related by duality in

geometric algebra. This occurs in all dimensions, and, in particular, in R4,1 .

In the conformal model of Euclidean space we have each point x in R3

mapped to an image point X = 2x + x2 n − n, in R4,1 on the null cone. We show

that certain blades can be taken as representatives of their corresponding (inner or

outer product) null spaces, and how their intersection with the null cone in R4,1

then represents geometric objects in R3 .

 112

4.3.3 Linear Transformations

Linear transformations form a major part of linear algebra. All of the

transformations we discussed in Section 4.2 are linear transformations on R4,1 .

Vector rotation as described in Section 3.7 is an example of a linear

transformation, e.g.,
∼
R(λu + µv) = R (λu + µv)R,
∼ ∼
= RλuR + RµvR,
∼ ∼
= λRuR + µRvR,

= λR(u) + µR(v).

4.3.4 Outermorphisms

Definition 4.3.2 (Outermorphism).

An outermorphism is a linear transformation F : Gm → Gn which preserves

outer products.

F (M1 ∧ M2 ) = F (M1 ) ∧ F (M2 ), (4.40)

for all multivectors M1 , M2 ∈ Gm .

Outermorphisms have nice algebraic properties that are essential to their

geometrical usage [DFM07, p. 101]:

• Blades remain blades. Oriented subspaces are transformed to oriented

subspaces.

• Grades are preserved. A linear transformation F maps vectors to

vectors. It follows that grade(F (B)) = grade(B) for all blades B.

 113

• Factorization is preserved. If blades A and B have a blade C in

common (so that they may be written as A = A′ ∧ C and B = B′ ∧ C, for

an appropriately chosen A′ and B′ .), then F (A) and F (B) have F (C) in

common. Geometrically this means that the intersection of subspaces is

preserved.

Reflections and rotations, as described in Sections 3.6 and 3.7, are examples of

outermorphisms. Macdonald [Mac10, p. 126] proves that reflection is an

outermorphism, i.e.,

R(u ∧ v) = R(u) ∧ R(v).

Thus, we can rotate a bivector by rotating each vector and then taking the outer

product of the result.

4.3.5 Determinants

Definition 4.3.3 (Determinant of a linear transformation [Mac10, p. 151]).

Let F be a one-to-one linear transformation on Rn , F : Rn → Rn . Let I be the

unit pseudoscalar of Gn , an n-vector. Because F preserves grades, F (I) is also an

n-vector. Thus, it is a scalar multiple of I,

F (I) = σ I.

The scalar σ is exactly the same as the determinant of an n × n matrix

representation of F . Thus, we have,

F (I) = det(F ) I. (4.41)

 114

The determinant has a simple geometric meaning:

The determinant of a linear transformation F on Rn is the factor by

which F multiplies oriented volumes of n-dimensional
parallelograms.

The following theorem is presented without proof:

Theorem 4.3.4 ([Mac12, p. 16]).

Let {v1 , v2 , · · · , vn } be vectors in Rn and let B = {e1 , e2 , · · · , en } be an

orthonormal basis for Rn . Then

 
v · e v2 · e1 ··· vn · e1
 1 1 
 
 v1 · e2 v2 · e2 · · · vn · e2 
 
v1 ∧ v2 ∧ · · · ∧ vn = det   I. (4.42)
 .. .. .. .. 
 . . . . 
 
v1 · en v2 · en · · · vn · e n
This theorem says that we can find the determinant of a matrix composed of

n column vectors from Rn , by taking the outer product of those vectors and

multiplying the result by I−1 .

4.4 Representing Geometric Objects with the Conformal Model

4.4.1 Points in the Conformal Model

As we have already seen in the conformal model, a point in Euclidean space is

represented by a null vector in R4,1 using the equation

Point: P = 21 (2x + x2 n − n) . (4.43)

Because this is a null vector in the conformal model, the inner product of any

point with itself is always zero.

P · P = 0.
 115

4.4.2 Inner Product of Two Points in the Conformal Model

Suppose we have two Euclidean points, x1 and x2 , that define a line in

Euclidean space. These two points map to vectors X1 and X2 respectively in

conformal space.

We now examine the scalar product X1 · X2 .

X1 · X2 = 21 (2x1 + x21 n − n) · 12 (2x2 + x22 n − n)

= (x1 + 12 x21 n − 21 n) · (x2 + 12 x22 n − 21 n)



 x1 · x2 + 1 2
x (x1 · n) − 1
· n)

 (x1
 2 2 2

=
 + 12 x21 (n · x2 ) + 14 x21 x22 (n · n) − 1 2
x (n
4 1
· n)



 − 1
2
(n · x2 ) − x (n · n) +
1 2
4 2
1
4
(n · n)


 x1 · x2 + 0 −

 0

=
 + 0 + 0 − 12 x21



 − 0 − 12 x22 + 0

= − 12 (x22 − 2x2 · x1 + x21 )

= − 12 (x2 − x1 ) · (x2 − x1 )

= − 12 |x2 − x1 |2 .

Because |x2 − x1 |2 is the square of the distance between the two points, we have

shown that the inner product of two points in the conformal model encodes the

distance between the two points on a Euclidean line.

This result is of fundamental importance because it follows that any

transformation of null vectors in conformal space which leaves the inner product

invariant can correspond to a transformation in Euclidean space which leaves

distance and angles invariant.

 116

For two points, P1 and P2 , in the conformal model, we have

P1 · P2 encodes the distance between the two points . (4.44)

4.4.3 Lines in the Conformal Model

In the conformal model, lines are represented by the outer product of two

ordinary points and the point at infinity, i.e., two points define a line and every line

includes a point at infinity.

Line: Ll = P1 ∧ P2 ∧ n , (4.45)

where P1 and P2 are the conformal mappings of two Euclidean points on the line,

i.e., P1 = F (p1 ) and P2 = F (p2 ).

The equation for a line is represented as an outer product null space (OPNS).

That is to say

OPNSline = {X | X ∧ P1 ∧ P2 ∧ n = 0}.

To show that this null set really does represent a line in R3 , we have the following

0 = X ∧ P1 ∧ P2 ∧ n

= 21 (2x + x2 n − n) ∧ 21 (2p1 + p21 n − n) ∧ 12 (2p2 + p22 n − n) ∧ n

4
= 18 [(2x − n) ∧ (2p1 − n) ∧ (2p2 − n) ∧ n]

= 81 [(4(x ∧ p1 ) − 2(x ∧ n) − 2(n ∧ p1 ) + n ∧ n) ∧ (2p2 − n) ∧ n]

6
= 81 [(4(x ∧ p1 ) − 2(x ∧ n) + 2(p1 ∧ n)) ∧ (2p2 − n) ∧ n]


 1 [(8(x ∧ p1 ∧ p2 ) − 4(x ∧ p1 ∧ n) − 4(x ∧ n ∧ p2 )
8
=

 + 2(x ∧ n ∧ n) + 4(p1 ∧ n ∧ p2 ) − 2(p1 ∧ n ∧ n)) ∧ n]
 117

8
= 81 [(8(x ∧ p1 ∧ p2 ) − 4(x ∧ p1 ∧ n) + 4(x ∧ p2 ∧ n) − 4(p1 ∧ p2 ∧ n)) ∧ n]


 1 [8(x ∧ p1 ∧ p2 ∧ n) − 4(x ∧ p1 ∧ n ∧ n)
8
=

 + 4(x ∧ p2 ∧ n ∧ n) − 4(p1 ∧ p2 ∧ n ∧ n)]


 1 [8(x ∧ p1 ∧ p2 ∧ n) + 4(x ∧ p1 ∧ n ∧ n)
8
=

 − 4(x ∧ p2 ∧ n ∧ n) + 4(p1 ∧ p2 ∧ n ∧ n)]

= 18 [8(x ∧ p1 ∧ p2 ∧ n) + 4(x ∧ p1 − x ∧ p2 + p1 ∧ p2 ) ∧ n ∧ n]

= 81 [8(x ∧ p1 ∧ p2 ∧ n) + 4(x ∧ (p1 − p2 ) + p1 ∧ p2 ) ∧ n ∧ n]

0 = x ∧ p1 ∧ p2 ∧ n + 21 (x ∧ (p1 − p2 ) + p1 ∧ p2 ) ∧ n ∧ n. (4.46)

Step (4) takes advantage of the fact that all terms with n become zero when we take

the outer product with n from the left. Steps (6) and (8) use the identities

n ∧ n = 0 and n ∧ pi = −pi ∧ n.

Because the two terms in Eq. (4.46) are linearly independent, they must both

be zero. The first term, x ∧ p1 ∧ p2 ∧ n = 0, implies that

x = αp1 + βp2 , where α, β ∈ R. (4.47)

The second term is zero if and only if

x ∧ (p2 − p1 ) = p1 ∧ p2

(αp1 + βp2 ) ∧ (p2 − p1 ) = p1 ∧ p2

αp1 ∧ p2 − αp1 ∧ p1 + βp2 ∧ p2 − βp2 ∧ p1 = p1 ∧ p2

αp1 ∧ p2 + βp1 ∧ p2 = p1 ∧ p2

(α + β)p1 ∧ p2 = p1 ∧ p2 .

Therefore,

α + β = 1. (4.48)
 118

From Eqs. (4.47) and (4.48) and Lemma 3.8.9 we see that x is a point on the line

through p1 and p2 .

A line can also be viewed as a circle with infinite radius.

4.4.4 Circles in the Conformal Model

In the conformal model, circles are represented by the outer product of three

non-collinear ordinary points, i.e., three non-collinear points define a circle.

Circle: Lc = A ∧ B ∧ C , (4.49)

where A, B, and C are the conformal mappings of three non-collinear Euclidean

points on the circle, i.e., A = F (a), B = F (b), and C = F (c).

The equation for a circle is represented as an outer product null space

(OPNS). That is to say

OPNScircle = {X | X ∧ A ∧ B ∧ C = 0}.

A circle is a two dimensional object, so for simplicity we consider a circle in

the xy-plane. The conformal model for a 2-dimensional space is R3,1 . We first note

that X ∧ A ∧ B ∧ C is a 4-blade in a 4-dimensional space R3,1 . From Theorem 4.3.4

we have
 
X · e1 A · e1 B · e1 C · e1
 
 
 X · e2 A · e2 B · e2 C · e2 
 
0 = X ∧ A ∧ B ∧ C = det   I,
 
 X·n A·n B·n C·n 
 
X·n A·n B·n C·n
 119

where I = e1 ∧ e2 ∧ n ∧ n. This implies that

 
X · e1 A · e1 B · e1 C · e1
 
 
 X · e2 A · e2 B · e2 C · e2 
 
det   = 0.
 
 X·n A·n B·n C·n 
 
X·n A·n B·n C·n

Because the determinant of a matrix equals the determinant of its transpose,

we have  
X · e1 X · e2 X · n X · n
 
 
 A · e1 A · e2 A·n A·n 
 
det   = 0.
 
 B · e1 B · e2 B·n B·n 
 
C · e1 C · e2 C·n C·n
By substituting all the inner products from the standard form Eq. (4.10) into

the matrix, we get  

2x1 2x2 −2 2x2
 
 
 2a1 2a2 −2 2a 
2
 
det   = 0,
 2 
 1
2b 2b 2 −2 2b 
 
2b1 2c2 −2 2c2

where subscripts, as in x1 , x2 , represent the e1 , e2 components of the vector x ∈ R2 .

We can factor out a 2 from each row and rearrange the columns of the matrix

without impacting the results to get

 
x2 x1 x2 −1
 
 
 a2 a1 a2 −1 
 
det   = 0, (4.50)
 2 
 b b2 b2 −1 
 
c2 c3 c2 −1

Equation (4.50) is in the standard form for representing a circle in R2 with a

 120

determinant [Che08, p. 3-14]. The variable x represents the locus of all points on a

circle passing through points a, b, and c.

We note that both circles and lines are represented by trivectors in the

conformal model. In fact, lines can be viewed as circles with infinite radius. If the

three points that define a circle are collinear, then the circle is a straight line, i.e.,

Lc ∧ n = 0 ⇒ straight line.

This explains why the earlier derivation of a line through p1 and p2 led to the

trivector P1 ∧ P2 ∧ n, which explicitly includes the point at infinity.

The center of the circle is the point

Center: P0 = Lc nLc . (4.51)

The operation Lc MLc reflects any object M in the circle Lc . Eq. (4.51) says that the

center of a circle is the image of the point at infinity under a reflection in a circle.

4.4.5 Planes in the Conformal Model

In the conformal model, planes are represented by the outer product of three

non-collinear ordinary points and the point at infinity, i.e., three non-collinear

points define a plane and every plane includes points at infinity.

Plane: Lp = A ∧ B ∧ C ∧ n , (4.52)

where A, B, and C are the conformal mappings of three Euclidean points on the

plane, i.e., A = F (a), B = F (b), and C = F (c).

The equation for a plane is represented as an outer product null space

(OPNS). That is to say

OPNSplane = {X | X ∧ A ∧ B ∧ C ∧ n = 0}.
 121

To show that this null set really does represent a plane in R3 , we would first note

that X ∧ A ∧ B ∧ C ∧ n is a 5-blade in a 5-dimensional space R4,1 . From

Theorem 4.3.4 we have

 
 X · e1 A · e1 B · e1 C · e1 n · e1 
 
 X · e2 A · e2 B · e2 C · e2 n · e2 
 
 
 
0 = X ∧ A ∧ B ∧ C ∧ n = det  X · e3 A · e3 B · e3 C · e3 n · e3  I,
 
 
 X·n A·n B·n C·n n·n 
 
 
X·n A·n B·n C·n n·n

where I = e1 ∧ e2 ∧ e3 ∧ n ∧ n. This implies that

 
 X · e1 A · e1 B · e1 C · e1 n · e1 
 
 X·e A·e B·e C·e n·e 
 2 2 2 2 2 
 
 
det  X · e3 A · e3 B · e3 C · e3 n · e3  = 0.
 
 
 X·n A·n B·n C·n n·n 
 
 
X·n A·n B·n C·n n·n

Because the determinant of a matrix equals the determinant of its transpose,

we have  
 X · e1 X · e2 X · e3 X · n X · n 
 
 A · e1 A · e2 A · e3 A · n A · n 
 
 
 
det  B · e1 B · e2 B · e3 B · n B · n  = 0.
 
 
 C · e1 C · e2 C · e3 C · n C · n 
 
 
n · e1 n · e2 n · e3 n · n n · n

By substituting all the inner products from the standard form Eq. (4.10) into
 122

the matrix, we get  

 2x1 2x2 2x3 −2 2x  2
 
 2a1 2a2 2a3 −2 2a2 
 
 
 
det  2  = 0,
2b1 2b2 2b3 −2 2b 

 
 2c1 2c2 2c3 −2 2c2 
 
 
0 0 0 0 2

where subscripts, as in x1 , x2 , x3 , represent the e1 , e2 , e3 components of the vector

x ∈ R3 .

Taking advantage of the zeros in row 5, the determinant of this matrix is 2

times the determinant of the submatrix formed by crossing out row 5, column 5.

This leads to  
2x 2x2 2x3 −2
 1 
 
 2a1 2a2 2a3 −2 
 
det   = 0,
 
 2b1 2b2 2b3 −2 
 
2c1 2c2 2c3 −2
We can factor out a 2 from each row without impacting the results to get
 
x1 x2 x3 −1
 
 
 a1 a2 a3 −1 
 
det   = 0, (4.53)
 
 1 2 3
b b b −1 
 
c1 c2 c3 −1

Equation (4.53) is in the standard form for representing a plane in R3 with a

determinant [Che08, p. 3-14]. The variable x represents the locus of all points on a

plane passing through points a, b, and c.

A plane can be viewed as a sphere with infinite radius.

 123

4.4.6 Spheres in the Conformal Model

In the conformal model, spheres are represented by the outer product of four

ordinary points, no three of which are collinear, i.e., four points, no three of which

are collinear, define a sphere.

Sphere: Ls = A ∧ B ∧ C ∧ D , (4.54)

where A, B, C, and D are the conformal mappings of four Euclidean points on the

sphere, i.e., A = F (a), B = F (b), C = F (c), and D = F (d).

The equation for a sphere is represented as an outer product null space

(OPNS). That is to say

OPNSsphere = {X | X ∧ A ∧ B ∧ C ∧ D = 0}.

To show that this null set really does represent a sphere in R3 , we would first note

that X ∧ A ∧ B ∧ C ∧ D is a 5-blade in a 5-dimensional space R4,1 . From

Theorem 4.3.4 we have

 
 X · e1 A · e1 B · e1 C · e1 D · e1 
 
 X · e2 A · e2 B · e2 C · e2 D · e2 
 
 
 
0 = X ∧ A ∧ B ∧ C ∧ D = det  X · e3 A · e3 B · e3 C · e3 D · e3  I,
 
 
 X·n A·n B·n C·n D·n 
 
 
X·n A·n B·n C·n D·n

where I = e1 ∧ e2 ∧ e3 ∧ n ∧ n. This implies that

 
 X · e1 A · e1 B · e1 C · e1 D · e1 
 
 X·e A·e B·e C·e D·e 
 2 2 2 2 2 
 
 
det  X · e3 A · e3 B · e3 C · e3 D · e3  = 0.
 
 
 X·n A·n B·n C·n D·n 
 
 
X·n A·n B·n C·n D·n
 124

Because the determinant of a matrix equals the determinant of its transpose,

we have  
 X · e1 X · e2 X · e3 X · n X · n 
 
 A · e1 A · e2 A · e3 A · n A · n 
 
 
 
det  B · e1 B · e2 B · e3 B · n B · n  = 0.
 
 
 C · e1 C · e2 C · e3 C · n C · n 
 
 
D · e1 D · e2 D · e3 D · n D · n
By substituting all the inner products from the standard form Eq. (4.10) into

the matrix, we get  

 2x1 2x2 2x3 −2 2x2 
 
 2a1 2a2 2a3 −2 2a2 
 
 
 
det  2  = 0,
2b1 2b2 2b3 −2 2b 

 
 2c1 2c2 2c3 −2 2c2 
 
 
2d1 2d2 2d3 −2 2d2
where subscripts, as in x1 , x2 , x3 , represent the e1 , e2 , e3 components of the vector

x ∈ R3 .

We can factor out a 2 from each row and rearrange the columns of the matrix

without impacting the results to get

 
2
 x x1 x2 x3 −1 
 
 a2 a a2 a3 −1 
 1 
 
 2 
det  b b1 b2 b3 −1  = 0, (4.55)
 
 
 c2 c1 c2 c3 −1 
 
 
d2 d1 d2 d3 −1

Equation (4.55) is in the standard form for representing a sphere in R3 with a

determinant [Che08, p. 3-15]. The variable x represents the locus of all points on a

sphere passing through points a, b, c, and d.

 125

We note that both spheres and planes are represented by quadvectors in the

conformal model. In fact, planes can be viewed as spheres with infinite radius. If

the four points that define a sphere are coplanar, then the sphere is a plane, i.e.,

Ls ∧ n = 0 ⇒ plane.

This explains why the earlier derivation of a plane through a, b, and c led to the

quadvector A ∧ B ∧ C ∧ n, which explicitly includes the point at infinity.

The center of the sphere is the point

Center: P0 = Ls nLs . (4.56)

The operation Ls MLs reflects any object M in the sphere Ls . Eq. (4.56) says that

the center of a sphere is the image of the point at infinity under a reflection in a

sphere.
 126

BIBLIOGRAPHY

[Che08] W. Chen, Linear Algebra, Macquarie University, 2008.

[DFM07] L. Dorst, D. Fontijne, and S. Mann, Geometric Algebra for Computer

Science, Morgan Kaufmann, 2007.

[DL03] C. Doran and A. Lasenby, Geometric Algebra for Physicists, Cambridge

University Press, 2003.

[Eve72] H. Eves, A Survey of Geometry, Allyn and Bacon, Inc., 1972.

[GLD93] S. Gull, A. Lasenby, and C. Doran, Imaginary Numbers are not Real –
The Geometric Algebra of Spacetime, Foundations of Physics 23 (1993),
no. 9, 1175–1201,
http://www.mrao.cam.ac.uk/˜clifford/introduction/intro/intro.html.

[Hes03] D. Hestenes, New Foundations for Classical Mechanics, Kluwer Academic

Publishers, 2003.

[HS84] D. Hestenes and G. Sobczyk, Clifford Algebra to Geometric Calculus, D.

Reidel Publishing Company, 1984.

[Kle08] M J Klein, Gibbs, Josiah Willard, Complete Dictionary of Scientific

Biography, Encyclopedia.com (2008),
http://www.encyclopedia.com/doc/1G2-2830901638.html.

[Lou03] P. Lounesto, Clifford Algebras and Spinors, Cambridge University Press,

2003.

[Mac10] A. Macdonald, Linear and Geometric Algebra, Self Published, 2010.

[Mac12] , A Survey of Geometric Algebra and Geometric Calculus,

http://faculty.luther.edu/˜macdonal/GA&GC.pdf, 2012.

[Sny10] J. Snygg, A New Approach to Differential Geometry Using Clifford’s

Geometric Algebra, Springer, 2010.

[Vin08] J. Vince, Geometric Algebra for Computer Graphics, Springer, 2008.

 127

APPENDIX A

NOTATIONAL CONVENTIONS USED IN THE LITERATURE

The subject of geometric algebra is new enough that standards have not yet

been established. This causes difficulties when perusing literature written by

different authors. Table A.1 provides a comparison of some of the notational

conventions used by different authors.

Table A.1: Notational Conventions Used in the Popular Literature

Property Hestenes Dorst, Fontijne Doran & Lasenby

[Hes03] & Mann [DFM07] [DL03]

orthonormal σ1 , σ2 , σ3 e1 , e2 , e3 e1 , e 2 , e 3
vectors
pseudoscalar i I I
dual iA A∗k = Ak ⌋I−1
n A∗ = IA
∼
reversion A† A∼ or A A†
infinity vector e ∞ n
origin vector e0 o n
conformal map x = x + 21 x2 e + e0 p = p + 12 p2 ∞ + o X = 2x + x2 n − n

Property Vince Macdonald

[Vin08] [Mac10] & [Mac12]

orthonormal e1 , e2 , e3 e1 , e2 , e3
vectors
pseudoscalar I i (2D) & I (3D)
dual A∗ = IA A∗ = AI−1
∼ ∼
reversion A∼ or A A∼ or A
infinity vector n e∞
origin vector n eo
conformal map X = 21 (2x + x2 n − n) p = eo + p + 21 p2 e∞
 128

APPENDIX B

FORMULAS AND IDENTITIES OF GEOMETRIC ALGEBRA

B.1 Axioms and Definitions of Geometric Algebra

In this section we present a set of axioms and definitions for

geometric algebra. This section closely follows the development presented in

Hestenes and Sobczyk [HS84]. For the sake of simplicity and ease of application, we

have not attempted to eliminate all redundancy from the axioms.

Definition B.1.1 (Multivector).

An element of the geometric algebra G is called a multivector.

We assume that G is closed under addition and geometric multiplication. The

sum or geometric product of any pair of multivectors is a unique multivector.

Axiom B.1.2.

Let A, B, and C be multivectors, then the geometric sum and product of

 129

multivectors have the following properties:

A1 : A+B =B+A addition is commutative, (B.1)

A2 : (A + B) + C = A + (B + C) addition is associative, (B.2)

(AB)C = A(BC) multiplication is associative, (B.3)

A3 : A(B + C) = AB + AC left distributive, (B.4)

(B + C)A = BA + CA right distributive, (B.5)

A4 : A+0=A ∃ a unique additive identity 0, (B.6)

1A = A ∃ a unique multiplicative identity 1, (B.7)

A5 : A + (−A) = 0 ∀A, ∃ a unique additive inverse − A. (B.8)

Geometric algebra is different from other associative algebras in that a

multivector is assumed to be composed of different types of objects that we classify

by grade. We assume that any multivector A can be written as the sum

∑
A = ⟨A⟩0 + ⟨A⟩1 + ⟨A⟩2 + · · · = ⟨A⟩k . (B.9)
k

Definition B.1.3 (k -vector).

The quantity ⟨A⟩k is called the k -vector part of A. If A = ⟨A⟩k for some

positive integer k, then A is said to be homogeneous of grade k and is called a

k -vector. We use the terms scalar, vector, bivector, and trivector as

synonyms for 0-vector, 1-vector, 2-vector, and 3-vector, respectively.

Axiom B.1.4.

The grade operator ⟨ ⟩k has the following properties

G1 : ⟨A + B⟩k = ⟨A⟩k + ⟨B⟩k , (B.10)

G2 : ⟨sA⟩k = s⟨A⟩k = ⟨A⟩k s, where s is a scalar (0-vector), (B.11)

G3 : ⟨⟨A⟩k ⟩k = ⟨A⟩k . (B.12)

 130

Axioms (B.10) and (B.11) imply that the space Gk of all k-vectors is a linear

subspace of G. Axiom (B.11) also implies that scalars compose a commutative

subalgebra of G. The space G0 of all scalars is identical with the set of real numbers

R. We regard the wider definition of the scalars (e.g. the complex numbers) as

entirely unnecessary and unfavorable to the purposes of geometric algebra. The

scalar grade operator is assigned the special notation

⟨M ⟩ ≡ ⟨M ⟩0 . (B.13)

The grade operator is a projection operator, so ⟨A⟩k can be regarded as the

projection of A into the space Gk .

Axiom B.1.5.

Multiplication of vectors is related to scalars by the assumption that the square

of any nonzero vector u is equal to the square of a unique positive1 scalar |u| called

the magnitude of u, that is

uu = u2 = ⟨u2 ⟩0 = |u|2 > 0. (B.14)

Definition B.1.6 (k -blade).

A multivector Ak is called a k -blade or a simple k -vector if and only if it

can be factored into a product of k anticommuting vectors u1 , u2 , · · · , uk , that is

Ak = u1 u2 · · · uk , (B.15)

where ui uj = −uj ui for i, j = 1, 2, . . . , k and i ̸= j. (B.16)

Any k -vector can be expressed as the sum of k-blades. We use a bold font to

emphasize that a variable is a k-blade as apposed to a k-vector or multivector.

1
Note that this axiom must be modified when working with spaces that are not positive definite,
i.e., spaces with negative or mixed signature.
 131

Axiom B.1.7.

For every nonzero k-blade Ak , there exists a nonzero vector u in G, such that

Ak u is an (k + 1)-blade.

This guarantees the existence of nontrivial blades of every finite grade. It

implies that G is is a linear space of infinite dimension.

B.2 Notation

With the variety of different types of objects in geometric algebra, it is

important to be consistent with the notation used. In the formulas and identities

that follow, we use the notation shown below.

scalar - lower-case letters, e.g. a, b, c, s.

vector - lower-case, bold characters, e.g. u, v, w.

blade - (when grade is of no consequence) upper-case, bold character, e.g. B, A.

k-blade - upper-case, bold character with subscript indicating grade, e.g. Bk , Ak .

k-vector - upper-case characters with subscript indicating grade, e.g. Ak , Bk .

multivector - upper-case character with no subscript, e.g. A, B.

B.3 Useful Formulas and Identities

In navigating the literature on geometric algebra, it is useful to have a list of

formulas and identities arranged for easy reference. This section provides such a list.

In the following, n is the dimension of the space we are modeling, i.e., Rn .

 132

B.3.1 The Grade Operator

The quantity ⟨A⟩k is called the k-vector part of the multivector A.

The grade operator ⟨ ⟩k has the following properties:

⟨A + B⟩k = ⟨A⟩k + ⟨B⟩k , (B.17)

⟨sA⟩k = s ⟨A⟩k = ⟨A⟩k s, where s is a scalar (0-vector), (B.18)

⟨⟨A⟩k ⟩k = ⟨A⟩k . (B.19)

B.3.2 Reversion

In algebraic computations, it is often desirable to reorder the factors in a

product. As such, we introduce the operation of reversion defined by the equations

∼
scalar reverse: s = s, (B.20)
∼
vector reverse: u = u, (B.21)

vector product reverse: (u1 u2 . . . uk−1 uk )∼ = uk uk−1 . . . u2 u1 , (B.22)

∼∼
multivector product reverse: (AB)∼ = B A, (B.23)
∼ ∼
multivector sum reverse: (A + B)∼ = A + B. (B.24)

B.3.3 Norm (in Positive Definite Spaces)

The norm |v| of a vector v in an inner-product space is given by |v|2 = v · v.

Consider the standard basis for G3 , B = {1, e1 , e2 , e3 , e12 , e23 , e31 , e123 }. Let eJ

represent one of these basis elements and aJ represent the scalar coefficient of the eJ

element in the multivector M , i.e.,

 133

M = a0 1 + a1 e1 + a2 e2 + a3 e3 + a12 e12 + a23 e23 + a31 e31 + a123 e123 .

The norm is defined by the equations

√
vector: |v| = v · v, (B.25)

k-blade: |Bk | = |b1 ||b2 | · · · |bk |, (B.26)

√
k-vector: |Ak | = Ak · Ak , (B.27)
v
√∑ u n
u∑
multivector: |M | = |aJ | = t
2 |⟨M ⟩k |2 . (B.28)
J k=0

Properties of the norm.

If v ̸= 0, then |v| > 0, positive definite (B.29)

|av| = |a||v|, scalar multiple (B.30)

|u · v| ≤ |u| |v|, Cauchy-Schwarz inequality (B.31)

|u + v| ≤ |u| + |v|. triangle inequality (B.32)

B.3.4 Inverse

Every non-null k-blade Bk has an inverse.

u
vector inverse: u−1 = (B.33)
|u|2
∼
Bk
blade inverse: B−1 = (B.34)
k
|Bk |2
 134

B.3.5 Geometric Product

The geometric product is the fundamental operation of geometric algebra.

The geometric product of two vectors has two components, a scalar (the inner

product) and a bivector (the outer product).

vector-vector: uv = u · v + u ∧ v (B.35)

vector-bivector: uB2 = u · B2 + u ∧ B2 (B.36)

vector-k -vector: uBk = u · Bk + u ∧ Bk (B.37)

j-vector-k -vector: Aj Bk = ⟨Aj Bk ⟩|j−k| + ⟨Aj Bk ⟩|j−k|+2 + · · ·

+ ⟨Aj Bk ⟩j+k−2 + ⟨Aj Bk ⟩j+k . (B.38)

Properties of the geometric product:

v2 = v · v = |v|2 , norm squared (B.39)

uv = u · v, if u ∥ v, parallel (B.40)

uv = u ∧ v = −v ∧ u, if u ⊥ v, perpendicular (B.41)

(au) v = a (uv), homogeneous (B.42)

u (av) = a (uv), (B.43)

(u + v) w = uw + vw, distributive (B.44)

w (u + v) = wu + wv, (B.45)

u (vw) = (uv) w. associative (B.46)

 135

B.3.6 Inner Product

The inner product is a grade-lowering operation, i.e., if Aj is a j-vector and

Bk is a k-vector with j ≤ k, then Aj · Bk is a multivector of grade k − j.

scalar-scalar: a·b=0 (B.47)

scalar-vector: a·v =0 (B.48)

vector-vector: u · v = 12 (uv + vu), (B.49)

vector-vector: u · v = |u||v| cos θ, 0≤θ≤π (B.50)

vector-bivector: u · B2 = 21 (uB2 − B2 u) = −B2 · u, (B.51)

vector-k -vector: u · Bk = 12 (uBk + (−1)k+1 Bk u) = ⟨uBk ⟩k−1 , (B.52)



⟨A B ⟩
j k |k−j| , if j, k > 0,
j-vector-k -vector: Aj · B k = (B.53)

 0, if j = 0 or k = 0,

∑
n ∑
n
multivector-multivector: A·B = ⟨A⟩j · ⟨B⟩k . (B.54)
j=0 k=0

Properties of the inner product:

u · v = v · u, commutative (B.55)

(au) · v = a(u · v), homogeneous (B.56)

u · (av) = a(u · v),

(u + v) · w = u · w + v · w, distributive (B.57)

w · (u + v) = w · u + w · v,

If v ̸= 0, then v · v > 0. for positive definite spaces (B.58)

 136

B.3.7 Outer Product

The outer product is a grade-raising operation, i.e., if Aj is a j-vector and Bk

is a k-vector, then Aj ∧ Bk is a multivector of grade j + k.

scalar-scalar: a ∧ b = a b, (B.59)

scalar-vector: a ∧ v = a v, (B.60)

vector-vector: u ∧ v = 12 (uv − vu), (B.61)

vector-vector: u ∧ v = |u||v| I sin θ, 0≤θ≤π (B.62)

vector-bivector: u ∧ B2 = 21 (uB2 + B2 u) = B2 ∧ u, (B.63)

vector-k -vector: u ∧ Bk = 12 (uBk − (−1)k+1 Bk u) = ⟨uBk ⟩k+1 , (B.64)



⟨A B ⟩ , if j + k ≤ n,
j k j+k
j-vector-k -vector: Aj ∧ B k = (B.65)

 0, if j + k > n,

∑
n ∑
n
multivector-multivector: A∧B = ⟨A⟩j ∧ ⟨B⟩k , (B.66)
j=0 k=0

1∑
arbitrary number v1 ∧ v2 ∧ · · · ∧ vr = (−1)ϵ vk1 vk2 · · · vkr , (B.67)
r!
of vectors:
where the sum runs over every permutation
k1 , . . . , kr of 1, . . . , r, and ϵ is +1 or −1 as
the permutation k1 , . . . , kr is an even or odd
respectively.
 137

Properties of the outer product:

v ∧ v = 0, parallel (B.68)

v ∧ u = −(u ∧ v), anticommutative (B.69)

(u ∧ v) ∧ w = u ∧ (v ∧ w), associative (B.70)

(au) ∧ v = a(u ∧ v), homogeneous (B.71)

u ∧ (av) = a(u ∧ v),

(u + v) ∧ w = u ∧ w + v ∧ w, distributive (B.72)

w ∧ (u + v) = w ∧ u + w ∧ v.

B.3.8 Orthogonal Test

In an inner-product space, the inner product can be used as a test for

orthogonality.

vector-vector: u·v =0 ⇔ u ⊥ v. (B.73)

vector-vector: vu = −uv ⇔ u ⊥ v. (B.74)

vector-blade: u · Bk = 0 ⇔ u ⊥ Bk . (B.75)


there is a nonzero vector
blade-blade: Aj · Bk = 0 ⇔ (B.76)

in Ak orthogonal to Bk .
 138

B.3.9 Parallel Test (Subspace Membership Text)

The outer product can be used to test for subspace membership.

vector-vector: u∧v =0 ⇔ u ∥ v. (B.77)

vector-vector: uv = u · v ⇔ u ∥ v. (B.78)

vector-k -blade: u ∧ Bk = 0 ⇔ u ∈ Bk . (B.79)

vector-k -blade: u · B*k = 0 ⇔ u ∈ Bk . (B.80)

j-blade-k -blade: Aj ∧ Bk = 0 ⇔ Aj ⊆ Bk for j < k (B.81)

B.3.10 Duality

Let I be the pseudoscalar for the space, e.g. for R3 , I = e1 e2 e3 and

I−1 = e3 e2 e1 . Then the dual M * of the multivector M is

M * = M I−1 . (B.82)

Properties of Duality:

j-vector-k -vector: (Aj · Bk )* = Aj ∧ Bk* , (B.83)

j-vector-k -vector: (Aj ∧ Bk )* = Aj · Bk* . (B.84)

B.3.11 Projection and Rejection

Formulas for determining the component of a vector u in a subspace Bk

(projection) and perpendicular to a subspace (rejection) are

vector-k -blade projection: u∥ = (u · Bk )/Bk (B.85)

vector-k -blade rejection: u⊥ = (u ∧ Bk )/Bk (B.86)

 139

B.3.12 Other Useful Identities

A · (B · C) = (A ∧ B) · C (B.87)

u · (v ∧ w) = (u · v)w − (u · w)v (B.88)

Jacobi Identity: u · (v ∧ w) + v · (w ∧ u) + w · (u ∧ v) = 0 (B.89)

B.4 Derivations of Formulas

This section provides some derivations of the formulas and assertions

presented in the previous section.

B.4.1 |av| = |a| |v|

|av|2 = (av) · (av) = a2 (v · v) = (|a| |v|)2 ,

therefore, |av| = |a| |v|.

B.4.2 |u · v| ≤ |u| |v| (Cauchy-Schwarz inequality)

This inequality holds for all positive definite spaces.

Case 1: Let v = 0, then the inequality becomes |u · 0| ≤ |u| |0|. This is true because

u · 0 = 0.

Case 2: Let v ̸= 0. For every scalar λ,

0 ≤ |u − λv|2 = (u − λv) · (u − λv) = |u|2 − 2λ(u · v) + λ2 |v|2 .

 140

Let λ = (u · v)/|v|2 , thus

0 ≤ |u|2 − 2λ(u · v) + λ2 |v|2 = |u|2 − 2(u · v)2 /|v|2 + (u · v)2 |v|2 /|v|4

= |u|2 − (u · v)2 /|v|2

0 ≤ |u|2 |v|2 − (u · v)2

0 ≤ |u|2 |v|2 − |u · v|2

|u · v|2 ≤ |u|2 |v|2

|u · v| ≤ |u| |v|.

B.4.3 |u + v| ≤ |u| + |v| (triangle inequality)

This inequality holds for all positive definite spaces. The proof uses the

Cauchy-Schwarz inequality from above.

|u + v|2 = (u + v) · (u + v)

=u·u+u·v+v·u+v·v

= |u|2 + 2 |u · v| + |v|2

≤ |u|2 + 2 |u| |v| + |v|2

= (|u| + |v|)2

therefore, |u + v| ≤ |u| + |v|.

 141

B.4.4 u−1 = u/|u|2 (vector inverse)

If u ̸= 0, then uu = u · u + u ∧ u = |u|2 + 0 = |u|2 . Thus, we have

( ) ( )
uu uu u u
1= = =u = u.
uu |u|2 |u|2 |u|2
u
Therefore, u−1 = .
|u|2

B.4.5 u · v = 12 (uv + vu) (inner product)

Method 1:

2(u · v) = u · v + u · v

=u·v+v·u

=u·u+u·v+v·u+v·v−u·u−v·v

= (u + v) · (u + v) − u · u − v · v

= (u + v)2 − u2 − v2

= u2 + uv + vu + v2 − u2 − v2

= uv + vu.

Therefore, u · v = 12 (uv + vu).

Method 2:

uv = u · v + u ∧ v, vu = v · u + v ∧ u = u · v − u ∧ v.

uv + vu = 2(u · v).

Therefore, u · v = 21 (uv + vu).

 142

B.4.6 u ∧ v = 12 (uv − vu) (outer product)

Following Method 2 above:

uv = u · v + u ∧ v, vu = v · u + v ∧ u = u · v − u ∧ v.

uv − vu = 2(u ∧ v).

Therefore, u ∧ v = 12 (uv − vu).

B.4.7 u · (v ∧ w) = (u · v) w − (u · w) v

The following identity is used in this derivation:

u · v = 12 (uv + vu) ⇔ uv = −vu + 2u · v.

Beginning with the geometric product of three vectors and using the

identity from above twice, we get

uvw = (uv)w

= (−vu + 2u · v)w

= −vuw + 2(u · v)w

= −v(uw) + 2(u · v)w

= −v(−wu + 2u · w) + 2(u · v)w

= vwu − 2(u · w)v + 2(u · v)w.

Rearranging terms and dividing by 2, we get

(u · v) w − (u · w) v = 21 [uvw − vwu]

= 12 [u(vw) − (vw)u]
 143

= 12 [u(v · w + v ∧ w) − (v · w + v ∧ w)u]

= 12 [u(v · w) + u(v ∧ w) − (v · w)u − (v ∧ w)u]

5
= 12 [(v · w)u + u(v ∧ w) − (v · w)u − (v ∧ w)u]

= 12 [u(v ∧ w) − (v ∧ w)u]
7
= 21 [u · (v ∧ w) + u ∧ (v ∧ w) − (v ∧ w) · u − (v ∧ w) ∧ u]
8
= 12 [u · (v ∧ w) + u ∧ (v ∧ w) + u · (v ∧ w) − u ∧ (v ∧ w)]

= 12 [2 u · (v ∧ w)]

= u · (v ∧ w).

∴ u · (v ∧ w) = (u · v) w − (u · w) v. (B.90)

Here is a description of some of the steps in the derivation.

Step 5: v · w is a scalar so we can bring it in front of u.

Step 7: We expand the geometric product of a vector and a bivector to the sum of

their inner product and outer product.

Step 8: The inner product of a vector and a bivector is anticommutative.

The outer product of a vector and a bivector is commutative.

 144

B.4.8 u · (v ∧ w) + v · (w ∧ u) + w · (u ∧ v) = 0 (Jacobi Identity)

We expand each term in the expression using Eq. (B.90).

u · (v ∧ w) + v · (w ∧ u) + w · (u ∧ v)

= [(u · v) w − (u · w) v] + [(v · w) u − (v · u) w] + [(w · u) v − (w · v) u]

= (u · v) w − (u · w) v + (v · w) u − (u · v) w + (u · w) v − (v · w) u

=0

∴ u · (v ∧ w) + v · (w ∧ u) + w · (u ∧ v) = 0. (B.91)

Note that in the next to last step, we used the fact that the inner product

of two vectors is commutative.

 145

APPENDIX C

GLOSSARY

Rn – a vector space of dimension n over the field R of real numbers with a

Euclidean signature.

Rp,q – a vector space of dimension n = p + q with p “positive dimensions” and

q “negative dimensions” over the field R of real numbers with signature (p, q).

Gn – a geometric algebra over Rn .

Gp,q – a geometric algebra over Rp,q .

∧k
Rn – the grade k subspace of Gn , i.e., the subspace of k-vectors.

AB – the geometric product of multivectors A and B.

A·B – the inner product of multivectors A and B.

A∧B – the outer product of multivectors A and B.

M −1 – the inverse of the multivector M .

⟨M ⟩k – the k-vector part of the multivector M .

⟨M ⟩ – the 0-vector (scalar) part ⟨M ⟩0 of the multivector M .

∼
M – the reversion of the multivector M .

M* – the dual of the multivector M .

 146

bivector – the outer product of two vectors, also called a 2-blade. We may think

of a bivector B = u ∧ v as an oriented area element with attitude, orientation,

and norm. A bivector is zero if either vector is zero or if the two vectors are

parallel. A bivector is not the same as a 2-vector. A 2-vector is a linear

combination of bivectors (2-blades). Bivectors are blades in G3 .

blade – A multivector Mk is called a k-blade or a simple k-vector if and only

if it can be factored into a product of k anticommuting vectors u1 , u2 , · · · , uk ,

that is

Mk = u1 u2 · · · uk ,

where ui uj = −uj ui for i, j = 1, 2, . . . , k and i ̸= j.

A k-blade is an outer product of k vectors that span a k dimensional subspace

of Rn . Blades are a generalization of the concept of scalars and vectors to

include simple bivectors, trivectors, and so on. A 0-blade is a scalar.

dual – for a given geometric algebra Gn with a unit pseudoscalar I, the dual of a

multivector M , denoted by M * , is defined as M * = M I−1 . If Bk is a blade of

grade k, then the dual of Bk represents the orthogonal complement in Rn of

the subspace represented by Bk . For example, if B ∈ G3 and B = e1 ∧ e2 ,

then B* = e3 . The dual, B*k , has a grade of n − k where n is the grade of the

pseudoscalar for the space. If Bk represents a subspace U , then B*k represents

U ⊥.

grade – the number of vector factors in a nonzero k-blade. We denote it by the

grade() symbol, e.g., grade(v1 ∧ · · · ∧ vk ) = k.

invertible – a vector is considered to be invertible if it is not the zero vector 0

 147

or a null vector.

k-blade – see blade.

k-vector – any linear combination of k-blades. For example, B1 = e1 e2 + e2 e3 is

a 2-vector in G3 . In G3 , 2-vectors are always 2-blades. This is not true in

higher dimensions. In G4 , the 2-blades e1 e2 and e3 e4 represent subspaces, but

the 2-vector B2 = e1 e2 + e3 e4 is not a blade and represents no subspace.

multivector – an element of the vector space Gn . A multivector M of grade n is

the sum of n + 1 k-vectors, i.e., M = 0-vector + 1-vector + ... + n-vector.

null vector – a vector whose inner product with itself is zero. Null vectors show

up when we work with spaces of mixed signature (both positive and negative),

as with the conformal model. Null vectors are not necessarily the zero vector.

However, the zero vector is a null vector. Null vectors are not invertible.

pseudoscalar – an n-vector in Gn . A unit pseudoscalar is denoted by I. In

Euclidean 3-space the unit pseudoscalar I = e1 ∧ e2 ∧ e3 . In Euclidean 2-space

the unit pseudoscalar I(2) = e1 ∧ e2 . The name pseudoscalar comes from the

study of dual spaces where the pseudoscalar plays the role that the scalar does

in normal space.

reversion – an unary operation on multivectors obtained by reversing each of its

∼
components. The reversion of a k-blade B, denoted by B, is just all the

vectors of B in reverse order, viz.,

If B = b1 ∧ b2 ∧ · · · ∧ bk ,
∼
then B = bk ∧ bk−1 ∧ · · · ∧ b1 .
 148

For multivectors A and B, reversion has the following properties:

∼ ∼ ∼∼
(A + B)∼ = A + B, (AB)∼ = B A,
∼ ∼
⟨A⟩0 = ⟨A⟩0 , ⟨A⟩1 = ⟨A⟩1 .

∼
Note that I = I−1 . The notation of the tilde is chosen to be reminiscent of an

editor’s notation for an interchange of terms. Some literature denotes

reversion with a dagger superscript, e.g., M † .

rotor – a geometric product of an even number of invertible unit vectors. Rotors

∼ ∼
have the useful property that R−1 = R (where R is the reversion of R, see

reversion). Examples of rotors are uv, vw, uw, where |u| = |v| = |w| = 1.

Rotors can always be expressed as exponentials, e.g., uv = eI(2) θ , where I(2) is

the unit pseudoscalar in the uv-plane and θ is the angle from u to v. Rotors

are even unit versors.

signature – if an n-dimensional vector space has p independent unit vectors

satisfying ei · ei = 1, and q independent unit vectors satisfying ei · ei = −1,

then it is customary to to say that the space has p “positive dimensions” and

q “negative dimensions” (with n = p + q). Rather than Rn , we then write Rp,q

and call (p, q) the signature of the space. An n-dimensional Euclidean space is

then written as Rn,0 , the representational space for the conformal model as

Rn+1,1 , and an n-dimensional Minkowski space as Rn−1,1 [DFM07, p. 586].

trivector – the outer product of three vectors, also called a 3-blade. We may

think of a trivector T = u ∧ v ∧ w as an oriented volume element with

attitude, orientation, and norm. A trivector is zero if any of the vectors are

zero or if all three vectors are coplanar. A trivector is not the same as a
 149

3-vector. A 3-vector is a linear combination of trivectors (3-blades).

versor – a geometric product of invertible vectors. Examples of versors are

u, uv, uvw, etc. A versor is called even or odd depending upon whether it has

an even or odd number of vector terms.