Geometric Algebra and its Application

to Mathematical Physics
Chris J. L. Doran
Sidney Sussex College

A dissertation submitted for the

degree of Doctor of Philosophy in the
University of Cambridge.
February 1994
Preface
This dissertation is the result of work carried out in the Department of Applied Mathe-
matics and Theoretical Physics between October 1990 and October 1993. Sections of the
dissertation have appeared in a series of collaborative papers 1] | 10]. Except where
explicit reference is made to the work of others, the work contained in this dissertation is
my own.

Acknowledgements
Many people have given help and support over the last three years and I am grateful to
them all. I owe a great debt to my supervisor, Nick Manton, for allowing me the freedom
to pursue my own interests, and to my two principle collaborators, Anthony Lasenby and
Stephen Gull, whose ideas and inspiration were essential in shaping my research. I also
thank David Hestenes for his encouragement and his company on an arduous journey to
Poland. Above all, I thank Julie Cooke for the love and encouragement that sustained
me through to the completion of this work. Finally, I thank Stuart Rankin and Margaret
James for many happy hours in the Mill, Mike and Rachael, Tim and Imogen, Paul, Alan
and my other colleagues in DAMTP and MRAO.
I gratefully acknowledge
nancial support from the SERC, DAMTP and Sidney Sussex
College.
To my parents
Contents
1 Introduction 1
1.1 Some History and Recent Developments : : : : : : : : : : : : : : : : : : : 4
1.2 Axioms and De
nitions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7
1.2.1 The Geometric Product : : : : : : : : : : : : : : : : : : : : : : : : 11
1.2.2 The Geometric Algebra of the Plane : : : : : : : : : : : : : : : : : 12
1.2.3 The Geometric Algebra of Space : : : : : : : : : : : : : : : : : : : : 15
1.2.4 Reections and Rotations : : : : : : : : : : : : : : : : : : : : : : : 17
1.2.5 The Geometric Algebra of Spacetime : : : : : : : : : : : : : : : : : 21
1.3 Linear Algebra : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 23
1.3.1 Linear Functions and the Outermorphism : : : : : : : : : : : : : : 23
1.3.2 Non-Orthonormal Frames : : : : : : : : : : : : : : : : : : : : : : : 26
2 Grassmann Algebra and Berezin Calculus 29
2.1 Grassmann Algebra versus Cliord Algebra : : : : : : : : : : : : : : : : : : 29
2.2 The Geometrisation of Berezin Calculus : : : : : : : : : : : : : : : : : : : 30
2.2.1 Example I. The \Grauss" Integral : : : : : : : : : : : : : : : : : : : 33
2.2.2 Example II. The Grassmann Fourier Transform : : : : : : : : : : : 34
2.3 Some Further Developments : : : : : : : : : : : : : : : : : : : : : : : : : : 37
3 Lie Groups and Spin Groups 39
3.1 Spin Groups and their Generators : : : : : : : : : : : : : : : : : : : : : : : 39
3.2 The Unitary Group as a Spin Group : : : : : : : : : : : : : : : : : : : : : 44
3.3 The General Linear Group as a Spin Group : : : : : : : : : : : : : : : : : 48
3.3.1 Endomorphisms of <n : : : : : : : : : : : : : : : : : : : : : : : : : 54
3.4 The Remaining Classical Groups : : : : : : : : : : : : : : : : : : : : : : : 59
3.4.1 Complexi
cation | so(n,C) : : : : : : : : : : : : : : : : : : : : : : 59
3.4.2 Quaternionic Structures | sp(n) and so (2n) : : : : : : : : : : : : 61
3.4.3 The Complex and Quaternionic General Linear Groups : : : : : : : 64
3.4.4 The symplectic Groups Sp(n,R) and Sp(n,C) : : : : : : : : : : : : : 65
3.5 Summary : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 66
4 Spinor Algebra 68
4.1 Pauli Spinors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 69
4.1.1 Pauli Operators : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 73
4.2 Multiparticle Pauli States : : : : : : : : : : : : : : : : : : : : : : : : : : : 74

i
 4.2.1 The Non-Relativistic Singlet State : : : : : : : : : : : : : : : : : : : 77
4.2.2 Non-Relativistic Multiparticle Observables : : : : : : : : : : : : : : 78
4.3 Dirac Spinors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 79
4.3.1 Changes of Representation | Weyl Spinors : : : : : : : : : : : : : 83
4.4 The Multiparticle Spacetime Algebra : : : : : : : : : : : : : : : : : : : : : 85
4.4.1 The Lorentz Singlet State : : : : : : : : : : : : : : : : : : : : : : : 89
4.5 2-Spinor Calculus : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 91
4.5.1 2-Spinor Observables : : : : : : : : : : : : : : : : : : : : : : : : : : 93
4.5.2 The 2-spinor Inner Product : : : : : : : : : : : : : : : : : : : : : : 95
4.5.3 The Null Tetrad : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 96
4.5.4 The rA A Operator : : : : : : : : : : : : : :
0
: : : : : : : : : : : : : 98
4.5.5 Applications : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 99
5 Point-particle Lagrangians 102
5.1 The Multivector Derivative : : : : : : : : : : : : : : : : : : : : : : : : : : : 102
5.2 Scalar and Multivector Lagrangians : : : : : : : : : : : : : : : : : : : : : : 105
5.2.1 Noether's Theorem : : : : : : : : : : : : : : : : : : : : : : : : : : : 106
5.2.2 Scalar Parameterised Transformations : : : : : : : : : : : : : : : : 106
5.2.3 Multivector Parameterised Transformations : : : : : : : : : : : : : 107
5.3 Applications | Models for Spinning Point Particles : : : : : : : : : : : : : 108
6 Field Theory 121
6.1 The Field Equations and Noether's Theorem : : : : : : : : : : : : : : : : : 122
6.2 Spacetime Transformations and their Conjugate Tensors : : : : : : : : : : 124
6.3 Applications : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 128
6.4 Multivector Techniques for Functional Dierentiation : : : : : : : : : : : : 135
7 Gravity as a Gauge Theory 137
7.1 Gauge Theories and Gravity : : : : : : : : : : : : : : : : : : : : : : : : : : 138
7.1.1 Local Poincare Invariance : : : : : : : : : : : : : : : : : : : : : : : 140
7.1.2 Gravitational Action and the Field Equations : : : : : : : : : : : : 143
7.1.3 The Matter-Field Equations : : : : : : : : : : : : : : : : : : : : : : 148
7.1.4 Comparison with Other Approaches : : : : : : : : : : : : : : : : : : 152
7.2 Point Source Solutions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 155
7.2.1 Radially-Symmetric Static Solutions : : : : : : : : : : : : : : : : : 156
7.2.2 Kerr-Type Solutions : : : : : : : : : : : : : : : : : : : : : : : : : : 166
7.3 Extended Matter Distributions : : : : : : : : : : : : : : : : : : : : : : : : 169
7.4 Conclusions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 174

ii
List of Tables
1.1 Some algebraic systems employed in modern physics : : : : : : : : : : : : : 6
3.1 Bivector Basis for so(p,q) : : : : : : : : : : : : : : : : : : : : : : : : : : : : 44
3.2 Bivector Basis for u(p,q) : : : : : : : : : : : : : : : : : : : : : : : : : : : : 48
3.3 Bivector Basis for su(p,q) : : : : : : : : : : : : : : : : : : : : : : : : : : : : 48
3.4 Bivector Basis for gl(n,R) : : : : : : : : : : : : : : : : : : : : : : : : : : : 51
3.5 Bivector Basis for sl(n,R) : : : : : : : : : : : : : : : : : : : : : : : : : : : 51
3.6 Bivector Basis for so(n,C) : : : : : : : : : : : : : : : : : : : : : : : : : : : 60
3.7 Bivector Basis for sp(n) : : : : : : : : : : : : : : : : : : : : : : : : : : : : 62
3.8 Bivector Basis for so (n) : : : : : : : : : : : : : : : : : : : : : : : : : : : : 63
3.9 Bivector Basis for gl(n,C) : : : : : : : : : : : : : : : : : : : : : : : : : : : 64
3.10 Bivector Basis for sl(n,C) : : : : : : : : : : : : : : : : : : : : : : : : : : : : 65
3.11 Bivector Basis for sp(n,R) : : : : : : : : : : : : : : : : : : : : : : : : : : : 66
3.12 The Classical Bilinear Forms and their Invariance Groups : : : : : : : : : : 67
3.13 The General Linear Groups : : : : : : : : : : : : : : : : : : : : : : : : : : 67
4.1 Spin Currents for 2-Particle Pauli States : : : : : : : : : : : : : : : : : : : 79
4.2 Two-Particle Relativistic Invariants : : : : : : : : : : : : : : : : : : : : : : 91
4.3 2-Spinor Manipulations : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 95

iii
Chapter 1
Introduction
This thesis is an investigation into the properties and applications of Cliord's geometric
algebra. That there is much new to say on the subject of Cliord algebra may be a surprise
to some. After all, mathematicians have known how to associate a Cliord algebra with
a given quadratic form for many years 11] and, by the end of the sixties, their algebraic
properties had been thoroughly explored. The result of this work was the classi
cation of
all Cliord algebras as matrix algebras over one of the three associative division algebras
(the real, complex and quaternion algebras) 12]{16]. But there is much more to geometric
algebra than merely Cliord algebra. To paraphrase from the introduction to \Cli ord
Algebra to Geometric Calculus" 24], Cliord algebra provides the grammar from which
geometric algebra is constructed, but it is only when this grammar is augmented with a
number of secondary de
nitions and concepts that one arrives at a true geometric algebra.
In fact, the algebraic properties of a geometric algebra are very simple to understand, they
are those of Euclidean vectors, planes and higher-dimensional (hyper)surfaces. It is the
computational power brought to the manipulation of these objects that makes geometric
algebra interesting and worthy of study. This computational power does not rest on the
construction of explicit matrix representations, and very little attention is given to the
matrix representations of the algebras used. Hence there is little common ground between
the work in this thesis and earlier work on the classi
cation and study of Cliord algebras.
There are two themes running through this thesis: that geometric algebra is the nat-
ural language in which to formulate a wide range of subjects in modern mathematical
physics, and that the reformulation of known mathematics and physics in terms of geo-
metric algebra leads to new ideas and possibilities. The development of new mathematical
formulations has played an important role in the progress of physics. One need only con-
sider the bene
ts of Lagrange's and Hamilton's reformulations of classical mechanics, or
Feynman's path integral (re)formulation of quantum mechanics, to see how important the
process of reformulation can be. Reformulations are often interesting simply for the novel
and unusual insights they can provide. In other cases, a new mathematical approach can
lead to signi
cant computational advantages, as with the use of quaternions for combining
rotations in three dimensions. At the back of any programme of reformulation, however,
lies the hope that it will lead to new mathematics or physics. If this turns out to be
the case, then the new formalism will usually be adopted and employed by the wider
community. The new results and ideas contained in this thesis should support the claim

1
that geometric algebra oers distinct advantages over more conventional techniques, and
so deserves to be taught and used widely.
The work in this thesis falls broadly into the categories of formalism, reformulation
and results. Whilst the foundations of geometric algebra were laid over a hundred years
ago, gaps in the formalism still remain. To
ll some of these gaps, a number of new alge-
braic techniques are developed within the framework of geometric algebra. The process
of reformulation concentrates on the subjects of Grassmann calculus, Lie algebra theory,
spinor algebra and Lagrangian
eld theory. In each case it is argued that the geometric
algebra formulation is computationally more ecient than standard approaches, and that
it provides many novel insights. The new results obtained include a real approach to
relativistic multiparticle quantum mechanics, a new classical model for quantum spin-1/2
and an approach to gravity based on gauge
elds acting in a at spacetime. Through-
out, consistent use of geometric algebra is maintained and the bene
ts arising from this
approach are emphasised.
This thesis begins with a brief history of the development of geometric algebra and a
review of its present state. This leads, inevitably, to a discussion of the work of David
Hestenes 17]{34], who has done much to shape the modern form of the subject. A number
of the central themes running through his research are described, with particular emphasis
given to his ideas on mathematical design. Geometric algebra is then introduced, closely
following Hestenes' own approach to the subject. The central axioms and de
nitions
are presented, and a notation is introduced which is employed consistently throughout
this work. In order to avoid introducing too much formalism at once, the material in
this thesis has been split into two halves. The
rst half, Chapters 1 to 4, deals solely
with applications to various algebras employed in mathematical physics. Accordingly,
only the required algebraic concepts are introduced in Chapter 1. The second half of the
thesis deals with applications of geometric algebra to problems in mechanics and
eld
theory. The essential new concept required here is that of the dierential with respect to
variables de
ned in a geometric algebra. This topic is known as geometric calculus , and
is introduced in Chapter 5.
Chapters 2, 3 and 4 demonstrate how geometric algebra embraces a number of alge-
braic structures essential to modern mathematical physics. The
rst of these is Grass-
mann algebra, and particular attention is given to the Grassmann \calculus" introduced
by Berezin 35]. This is shown to have a simple formulation in terms of the properties
of non-orthonormal frames and examples are given of the algebraic advantages oered by
this new approach. Lie algebras and Lie groups are considered in Chapter 3. Lie groups
underpin many structures at the heart of modern particle physics, so it is important to
develop a framework for the study of their properties within geometric algebra. It is
shown that all (
nite dimensional) Lie algebras can be realised as bivector algebras and it
follows that all matrix Lie groups can be realised as spin groups. This has the interesting
consequence that every linear transformation can be represented as a monomial of (Clif-
ford) vectors. General methods for constructing bivector representations of Lie algebras
are given, and explicit constructions are found for a number of interesting cases.
The
nal algebraic structures studied are spinors. These are studied using the space-
time algebra | the (real) geometric algebra of Minkowski spacetime. Explicit maps are
constructed between Pauli and Dirac column spinors and spacetime multivectors, and

2
it is shown that the role of the scalar unit imaginary of quantum mechanics is played
by a
xed spacetime bivector. Changes of representation are discussed, and the Dirac
equation is presented in a form in which it can be analysed and solved without requiring
the construction of an explicit matrix representation. The concept of the multiparticle
spacetime algebra is then introduced and is used to construct both non-relativistic and
relativistic two-particle states. Some relativistic two-particle wave equations are consid-
ered and a new equation, based solely in the multiparticle spacetime algebra, is proposed.
In a
nal application, the multiparticle spacetime algebra is used to reformulate aspects
of the 2-spinor calculus developed by Penrose & Rindler 36, 37].
The second half of this thesis deals with applications of geometric calculus. The essen-
tial techniques are described in Chapter 5, which introduces the concept of the multivector
derivative 18, 24]. The multivector derivative is the natural extension of calculus for func-
tions mapping between geometric algebra elements (multivectors). Geometric calculus is
shown to be ideal for studying Lagrangian mechanics and two new ideas are developed |
multivector Lagrangians and multivector-parameterised transformations. These ideas are
illustrated by detailed application to two models for spinning point particles. The
rst,
due to Barut & Zanghi 38], models an electron by a classical spinor equation. This model
suers from a number of defects, including an incorrect prediction for the precession of
the spin axis in a magnetic
eld. An alternative model is proposed which removes many
of these defects and hints strongly that, at the classical level, spinors are the generators
of rotations. The second model is taken from pseudoclassical mechanics 39], and has the
interesting property that the Lagrangian is no longer a scalar but a bivector-valued func-
tion. The equations of motion are solved exactly and a number of conserved quantities
are derived.
Lagrangian
eld theory is considered in Chapter 6. A unifying framework for vectors,
tensors and spinors is developed and applied to problems in Maxwell and Dirac theory.
Of particular interest here is the construction of new conjugate currents in the Dirac
theory, based on continuous transformations of multivector spinors which have no simple
counterpart in the column spinor formalism. The chapter concludes with the development
of an extension of multivector calculus appropriate for multivector-valued linear functions.
The various techniques developed throughout this thesis are brought together in Chap-
ter 7, where a theory of gravity based on gauge transformations in a at spacetime is
presented. The motivation behind this approach is threefold: (1) to introduce gravity
through a similar route to the other interactions, (2) to eliminate passive transformations
and base physics solely in terms of active transformations and (3) to develop a theory
within the framework of the spacetime algebra. A number of consequences of this theory
are explored and are compared with the predictions of general relativity and spin-torsion
theories. One signi
cant consequence is the appearance of time-reversal asymmetry in
radially-symmetric (point source) solutions. Geometric algebra oers numerous advan-
tages over conventional tensor calculus, as is demonstrated by some remarkably compact
formulae for the Riemann tensor for various
eld con
gurations. Finally, it is suggested
that the consistent employment of geometric algebra opens up possibilities for a genuine
multiparticle theory of gravity.

3
1.1 Some History and Recent Developments
There can be few pieces of mathematics that have been re-discovered more often than
Cliord algebras 26]. The earliest steps towards what we now recognise as a geometric
algebra were taken by the pioneers of the use of complex numbers in physics. Wessel,
Argand and Gauss all realised the utility of complex numbers when studying 2-dimensional
problems and, in particular, they were aware that the exponential of an imaginary number
is a useful means of representing rotations. This is simply a special case of the more general
method for performing rotations in geometric algebra.
The next step was taken by Hamilton, whose attempts to generalise the complex num-
bers to three dimensions led him to his famous quaternion algebra (see 40] for a detailed
history of this subject). The quaternion algebra is the Cliord algebra of 2-dimensional
anti-Euclidean space, though the quaternions are better viewed as a subalgebra of the
Cliord algebra of 3-dimensional space. Hamilton's ideas exerted a strong inuence on
his contemporaries, as can be seen form the work of the two people whose names are most
closely associated with modern geometric algebra | Cliord and Grassmann.
Grassmann is best known for his algebra of extension. He de
ned hypernumbers
ei, which he identi
ed with unit directed line segments. An arbitrary vector was then
written as aiei, where the ai are scalar coecients. Two products were assigned to these
hypernumbers, an inner product
ei
ej = ej
ei =
ij (1:1)
and an outer product
ei ^ ej = ;ej ^ ei: (1:2)
The result of the outer product was identi
ed as a directed plane segment and Grassmann
extended this concept to include higher-dimensional objects in arbitrary dimensions. A
fact overlooked by many historians of mathematics is that, in his later years, Grassmann
combined his interior and exterior products into a single, central product 41]. Thus he
wrote
ab = a
b + a ^ b (1:3)
though he employed a dierent notation. The central product is precisely Cliord's prod-
uct of vectors, which Grassmann arrived at independently from (and slightly prior to)
Cliord. Grassmann's motivation for introducing this new product was to show that
Hamilton's quaternion algebra could be embedded within his own extension algebra. It
was through attempting to unify the quaternions and Grassmann's algebra into a sin-
gle mathematical system that Cliord was also led to his algebra. Indeed, the paper in
which Cliord introduced his algebra is entitled \Applications of Grassmann's extensive
algebra" 42].
Despite the eorts of these mathematicians to
nd a simple uni
ed geometric algebra
(Cliord's name for his algebra), physicists ultimately adopted a hybrid system, due
largely to Gibbs. Gibbs also introduced two products for vectors. His scalar (inner)
product was essentially that of Grassmann, and his vector (cross) product was abstracted
from the quaternions. The vector product of two vectors was a third, so his algebra
was closed and required no additional elements. Gibbs' algebra proved to be well suited
4
to problems in electromagnetism, and quickly became popular. This was despite the
clear de
ciencies of the vector product | it is not associative and cannot be generalised
to higher dimensions. Though special relativity was only a few years o, this lack of
generalisability did not appear to deter physicists and within a few years Gibbs' vector
algebra had become practically the exclusive language of vector analysis.
The end result of these events was that Cliord's algebra was lost amongst the wealth
of new algebras being created in the late 19th century 40]. Few realised its great promise
and, along with the quaternion algebra, it was relegated to the pages of pure algebra
texts. Twenty more years passed before Cliord algebras were re-discovered by Dirac in
his theory of the electron. Dirac arrived at a Cliord algebra through a very dierent
route to the mathematicians before him. He was attempting to
nd an operator whose
square was the Laplacian and he hit upon the matrix operator  @ , where the  -matrices
satisfy
 
+ 
 = 2I
: (1:4)
Sadly, the connection with vector geometry had been lost by this point, and ever since
the  -matrices have been thought of as operating on an internal electron spin space.
There the subject remained, essentially, for a further 30 years. During the interim
period physicists adopted a wide number of new algebraic systems (coordinate geometry,
matrix algebra, tensor algebra, dierential forms, spinor calculus), whilst Cliord algebras
were thought to be solely the preserve of electron theory. Then, during the sixties, two
crucial developments dramatically altered the perspective. The
rst was made by Atiyah
and Singer 43], who realised the importance of Dirac's operator in studying manifolds
which admitted a global spin structure. This led them to their famous index theorems, and
opened new avenues in the subjects of geometry and topology. Ever since, Cliord algebras
have taken on an increasingly more fundamental role and a recent text proclaimed that
Cliord algebras \emerge repeatedly at the very core of an astonishing variety of problems
in geometry and topology" 15].
Whilst the impact of Atiyah's work was immediate, the second major step taken in
the sixties has been slower in coming to fruition. David Hestenes had an unusual training
as a physicist, having taken his bachelor's degree in philosophy. He has often stated that
this gave him a dierent perspective on the role of language in understanding 27]. Like
many theoretical physicists in the sixties, Hestenes worked on ways to incorporate larger
multiplets of particles into the known structures of
eld theory. During the course of these
investigations he was struck by the idea that the Dirac matrices could be interpreted as
vectors, and this led him to a number of new insights into the structure and meaning of
the Dirac equation and quantum mechanics in general 27].
The success of this idea led Hestenes to reconsider the wider applicability of Cliord
algebras. He realised that a Cliord algebra is no less than a system of directed numbers
and, as such, is the natural language in which to express a number of theorems and results
from algebra and geometry. Hestenes has spent many years developing Cliord algebra
into a complete language for physics, which he calls geometric algebra. The reason for
preferring this name is not only that it was Cliord's original choice, but also that it serves
to distinguish Hestenes' work from the strictly algebraic studies of many contemporary
texts.
During the course of this development, Hestenes identi
ed an issue which has been
5
 coordinate geometry spinor calculus
complex analysis Grassmann algebra
vector analysis Berezin calculus
tensor analysis dierential forms
Lie algebras twistors
Cliord algebra

Table 1.1: Some algebraic systems employed in modern physics

paid little attention | that of mathematical design. Mathematics has grown into an
enormous group undertaking, but few people concern themselves with how the results of
this eort should best be organised. Instead, we have a situation in which a vast range of
disparate algebraic systems and techniques are employed. Consider, for example, the list
of algebras employed in theoretical (and especially particle) physics contained in Table 1.1.
Each of these has their own conventions and their own methods for proving similar results.
These algebras were introduced to tackle speci
c classes of problem, and each is limited
in its overall scope. Furthermore, there is only a limited degree of integrability between
these systems. The situation is analogous to that in the early years of software design.
Mathematics has, in essence, been designed \bottom-up". What is required is a \top-
down" approach | a familiar concept in systems design. Such an approach involves
identifying a single algebraic system of maximal scope, coherence and simplicity which
encompasses all of the narrower systems of Table 1.1. This algebraic system, or language,
must be suciently general to enable it to formulate any result in any of the sub-systems
it contains. But it must also be ecient, so that the interrelations between the subsystems
can be clearly seen. Hestenes' contention is that geometric algebra is precisely the required
system. He has shown how it incorporates many of the systems in Table 1.1, and part of
the aim of this thesis is to
ll in some of the remaining gaps.
This \top-down" approach is contrary to the development of much of modern math-
ematics, which attempts to tackle each problem with a system which has the minimum
number of axioms. Additional structure is then handled by the addition of further axioms.
For example, employing geometric algebra for problems in topology is often criticised on
the grounds that geometric algebra contains redundant structure for the problem (in this
case a metric derived from the inner product). But there is considerable merit to seeing
mathematics the other way round. This way, the relationships between
elds become
clearer, and generalisations are suggested which could not be seen form the perspective
of a more restricted system. For the case of topology, the subject would be seen in the
manner that it was originally envisaged | as the study of properties of manifolds that are
unchanged under deformations. It is often suggested that the geniuses of mathematics are
those who can see beyond the symbols on the page to their deeper signi
cance. Atiyah,
for example, said that a good mathematician sees analogies between proofs, but a great
mathematician sees analogies between analogies1 . Hestenes takes this as evidence that
these people understood the issues of design and saw mathematics \top-down", even if it
1 I am grateful to Margaret James for this quote.

6
was not formulated as such. By adopting good design principles in the development of
mathematics, the bene
ts of these insights would be available to all. Some issues of what
constitutes good design are debated at various points in this introduction, though this
subject is only in its infancy.
In conclusion, the subject of geometric algebra is in a curious state. On the one
hand, the algebraic structures keeps reappearing in central ideas in physics, geometry
and topology, and most mathematicians are now aware of the importance of Cliord
algebras. On the other, there is far less support for Hestenes' contention that geometric
algebra, built on the framework of Cliord algebra, provides a uni
ed language for much
of modern mathematics. The work in this thesis is intended to oer support for Hestenes'
ideas.

1.2 Axioms and Denitions

The remaining sections of this chapter form an introduction to geometric algebra and to
the conventions adopted in this thesis. Further details can be found in \Cli ord algebra
to geometric calculus" 24], which is the most detailed and comprehensive text on geo-
metric algebra. More pedagogical introductions are provided by Hestenes 25, 26] and
Vold 44, 45], and 30] contains useful additional material. The conference report on the
second workshop on \Cli ord algebras and their applications in mathematical physics"
46] contains a review of the subject and ends with a list of recommended texts, though
not all of these are relevant to the material in this thesis.
In deciding how best to de
ne geometric algebra we arrive at our
rst issue of math-
ematical design. Modern mathematics texts (see \Spin Geometry" by H.B Lawson and
M.-L. Michelsohn 15], for example) favour the following de
nition of a Cliord algebra.
One starts with a vector space V over a commutative
eld k, and supposes that q is a
quadratic form on V . The tensor algebra of V is de
ned as
X
1
T (V ) = rV (1:5)
r=0
where  is the tensor product. One next de
nes an ideal Iq (V ) in T (V ) generated by all
elements of the form v  v + q(v)1 for v 2 V . The Cliord algebra is then de
ned as the
quotient
Cl(V q)  T (V )=Iq (V ): (1:6)
This de
nition is mathematically correct, but has a number of drawbacks:
1. The de
nition involves the tensor product, , which has to be de
ned initially.
2. The de
nition uses two concepts, tensor algebras and ideals, which are irrelevant to
the properties of the resultant geometric algebra.
3. Deriving the essential properties of the Cliord algebra from (1.6) requires further
work, and none of these properties are intuitively obvious from the axioms.

7
 4. The de
nition is completely useless for introducing geometric algebra to a physi-
cist or an engineer. It contains too many concepts that are the preserve of pure
mathematics.
Clearly, it is desirable to
nd an alternative axiomatic basis for geometric algebra which
does not share these de
ciencies. The axioms should be consistent with our ideas of what
constitutes good design. The above considerations lead us propose the following principle:
The axioms of an algebraic system should deal directly with the objects of
interest.
That is to say, the axioms should oer some intuitive feel of the properties of the system
they are de
ning.
The central properties of a geometric algebra are the grading, which separates objects
into dierent types, and the associative product between the elements of the algebra. With
these in mind, we adopt the following de
nition. A geometric algebra G is a graded linear
space, the elements of which are called multivectors. The grade-0 elements are called
scalars and are identi
ed with the
eld of real numbers (we will have no cause to consider
a geometric algebra over the complex
eld). The grade-1 elements are called vectors, and
can be thought of as directed line segments. The elements of G are de
ned to have an
addition, and each graded subspace is closed under this. A product is also de
ned which
is associative and distributive, though non-commutative (except for multiplication by a
scalar). The
nal axiom (which distinguishes a geometric algebra from other associative
algebras) is that the square of any vector is a scalar.
Given two vectors, a and b, we
nd that
(a + b)2 = (a + b)(a + b)
= a2 + (ab + ba) + b2: (1.7)
It follows that
ab + ba = (a + b)2 ; a2 ; b2 (1:8)
and hence that (ab + ba) is also a scalar. The geometric product of 2 vectors a b can
therefore be decomposed as
ab = a
b + a ^ b (1:9)
where
a
b  21 (ab + ba) (1:10)
is the standard scalar, or inner, product (a real scalar), and
a ^ b  12 (ab ; ba) (1:11)
is the antisymmetric outer product of two vectors, originally introduced by Grassmann.
The outer product of a and b anticommutes with both a and b,
a(a ^ b) = 21 (a2b ; aba)
= 12 (ba2 ; aba)
= ; 21 (ab ; ba)a
= ;(a ^ b)a (1.12)
8
so a ^ b cannot contain a scalar component. The axioms are also sucient to show that
a ^ b cannot contain a vector part. If we supposed that a ^ b contained a vector part c,
then the symmetrised product of a ^ b with c would necessarily contain a scalar part. But
c(a ^ b) + (a ^ b)c anticommutes with any vector d satisfying d
a = d
b = d
c = 0, and
so cannot contain a scalar component. The result of the outer product of two vectors
is therefore a new object, which is dened to be grade-2 and is called a bivector. It can
be thought of as representing a directed plane segment containing the vectors a and b.
The bivectors form a linear space, though not all bivectors can be written as the exterior
product of two vectors.
The de
nition of the outer product is extended to give an inductive de
nition of the
grading for the entire algebra. The procedure is illustrated as follows. Introduce a third
vector c and write
c(a ^ b) = 12 c(ab ; ba)
= (a
c)b ; (b
c)a ; 12 (acb ; bca)
= 2(a
c)b ; 2(b
c)a + 12 (ab ; ba)c (1.13)
so that
c(a ^ b) ; (a ^ b)c = 2(a
c)b ; 2(b
c)a: (1:14)
The right-hand side of (1.14) is a vector, so one decomposes c(a ^ b) into
c(a ^ b) = c
(a ^ b) + c ^ (a ^ b) (1:15)
where
c
(a ^ b)  21 c(a ^ b) ; (a ^ b)c] (1:16)
and
c ^ (a ^ b)  12 c(a ^ b) + (a ^ b)c] : (1:17)
The de
nitions (1.16) and (1.17) extend the de
nitions of the inner and outer products
to the case where a vector is multiplying a bivector. Again, (1.17) results in a new object,
which is assigned grade-3 and is called a trivector. The axioms are sucient to prove that
the outer product of a vector with a bivector is associative:
c ^ (a ^ b) = 12 c(a ^ b) + (a ^ b)c]
= 41 cab ; cba + abc ; bac]
= 41 2(c ^ a)b + acb + abc + 2b(c ^ a) ; bca ; cba]
= 12 (c ^ a)b + b(c ^ a) + a(b
c) ; (b
c)a]
= (c ^ a) ^ b: (1.18)
The de
nitions of the inner and outer products are extended to the geometric product
of a vector with a grade-r multivector Ar as,
aAr = a
Ar + a ^ Ar (1:19)
where the inner product
a
Ar  haArir;1 = 21 (aAr ; (;1)r Ar a) (1:20)
9
lowers the grade of Ar by one and the outer (exterior) product
a ^ Ar  haArir+1 = 12 (aAr + (;1)r Ar a) (1:21)
raises the grade by one. We have used the notation hAir to denote the result of the
operation of taking the grade-r part of A (this is a projection operation). As a further
abbreviation we write the scalar (grade 0) part of A simply as hAi.
The entire multivector algebra can be built up by repeated multiplication of vectors.
Multivectors which contain elements of only one grade are termed homogeneous, and will
usually be written as Ar to show that A contains only a grade-r component. Homogeneous
multivectors which can be expressed purely as the outer product of a set of (independent)
vectors are termed blades.
The geometric product of two multivectors is (by de
nition) associative, and for two
homogeneous multivectors of grade r and s this product can be decomposed as follows:
ArBs = hAB ir+s + hAB ir+s;2 : : : + hAB ijr;sj : (1:22)
The \
" and \^" symbols are retained for the lowest-grade and highest-grade terms of
this series, so that
Ar
Bs  hAB ijs;rj (1.23)
Ar ^ Bs  hAB is+r  (1.24)
which we call the interior and exterior products respectively. The exterior product is
associative, and satis
es the symmetry property
Ar ^ Bs = (;1)rs Bs ^ Ar: (1:25)
An important operation which can be performed on multivectors is reversion, which
reverses the order of vectors in any multivector. The result of reversing the multivector
A is written A~, and is called the reverse of A. The reverse of a vector is the vector itself,
and for a product of multivectors we have that
(AB )~= B~ A:
~ (1:26)
It can be checked that for homogeneous multivectors
A~r = (;1)r(r;1)=2Ar: (1:27)
It is useful to de
ne two further products from the geometric product. The
rst is the
scalar product
A  B  hAB i: (1:28)
This is commutative, and satis
es the useful cyclic-reordering property
hA : : : BC i = hCA : : :B i: (1:29)
In positive de
nite spaces the scalar product de
nes the modulus function
jAj  (A  A)1=2: (1:30)
10
The second new product is the commutator product, de
ned by
A  B  12 (AB ; BA): (1:31)
The associativity of the geometric product ensures that the commutator product satis
es
the Jacobi identity
A  (B  C ) + B  (C  A) + C  (A  B ) = 0: (1:32)
Finally, we introduce an operator ordering convention. In the absence of brackets,
inner, outer and scalar products take precedence over geometric products. Thus a
bc
means (a
b)c and not a
(bc). This convention helps to eliminate unruly numbers of
brackets. Summation convention is also used throughout this thesis.
One can now derive a vast number of properties of multivectors, as is done in Chapter 1
of 24]. But before proceeding, it is worthwhile stepping back and looking at the system
we have de
ned. In particular, we need to see that the axioms have produced a system
with sensible properties that match our intuitions about physical space and geometry in
general.

1.2.1 The Geometric Product

Our axioms have led us to an associative product for vectors, ab = a
b + a ^ b. We call
this the geometric product. It has the following two properties:
Parallel vectors (e.g. a and a) commute, and the the geometric product of parallel
vectors is a scalar. Such a product is used, for example, when
nding the length of
a vector.
Perpendicular vectors (a, b where a
b = 0) anticommute, and the geometric product
of perpendicular vectors is a bivector. This is a directed plane segment, or directed
area, containing the vectors a and b.
Independently, these two features of the algebra are quite sensible. It is therefore reason-
able to suppose that the product of vectors that are neither parallel nor perpendicular
should contain both scalar and bivector parts.
But what does it mean to add a scalar to a bivector?
This is the point which regularly causes the most confusion (see 47], for example).
Adding together a scalar and a bivector doesn't seem right | they are dierent types of
quantities. But this is exactly what you do want addition to do. The result of adding a
scalar to a bivector is an object that has both scalar and bivector parts, in exactly the
same way that the addition of real and imaginary numbers yields an object with both
real and imaginary parts. We call this latter object a \complex number" and, in the same
way, we refer to a (scalar + bivector) as a \multivector", accepting throughout that we
are combining objects of dierent types. The addition of scalar and bivector does not
result in a single new quantity in the same way as 2 + 3 = 5! we are simply keeping
track of separate components in the symbol ab = a
b + a ^ b or z = x + iy. This type
11
of addition, of objects from separate linear spaces, could be given the symbol
, but it
should be evident from our experience of complex numbers that it is harmless, and more
convenient, to extend the de
nition of addition and use the ordinary + sign.
Further insights are gained by the construction of explicit algebras for
nite dimen-
sional spaces. This is achieved most simply through the introduction of an orthonormal
frame of vectors fig satisfying
i
j =
ij (1:33)
or
ij + j i = 2
ij : (1:34)
This is the conventional starting point for the matrix representation theory of
nite Clif-
ford algebras 13, 48]. It is also the usual route by which Cliord algebras enter particle
physics, though there the fig are thought of as operators, and not as orthonormal vec-
tors. The geometric algebra we have de
ned is associative and any associative algebra
can be represented as a matrix algebra, so why not de
ne a geometric algebra as a matrix
algebra? There are a number of aws with this approach, which Hestenes has frequently
drawn attention to 26]. The approach fails, in particular, when geometric algebra is used
to study projectively and conformally related geometries 31]. There, one needs to be able
to move freely between dierent dimensional spaces. Matrix representations are too rigid
to achieve this satisfactorily. An example of this will be encountered shortly.
There is a further reason for preferring not to introduce Cliord algebras via their
matrix representations. It is related to our second principle of good design, which is that
the axioms af an algebraic system should not introduce redundant structure.
The introduction of matrices is redundant because all geometrically meaningful results
exist independently of any matrix representations. Quite simply, matrices are irrelevant
for the development of geometric algebra.
The introduction of a basis set of n independent, orthonormal vectors fig de
nes a
basis for the entire algebra generated by these vectors:
1 fig fi ^ j g fi ^ j ^ k g : : :  1 ^ 2 : : :^ n  I: (1:35)
Any multivector can now be expanded in this basis, though one of the strengths of geo-
metric algebra is that it possible to carry out many calculations in a basis-free way. Many
examples of this will be presented in this thesis,
The highest-grade blade in the algebra (1.35) is given the name \pseudoscalar" (or
directed volume element) and is of special signi
cance in geometric algebra. Its unit is
given the special symbol I (or i in three or four dimensions). It is a pure blade, and a
knowledge of I is sucient to specify the vector space over which the algebra is de
ned
(see 24, Chapter 1]). The pseudoscalar also de
nes the duality operation for the algebra,
since multiplication of a grade-r multivector by I results in a grade-(n ; r) multivector.

1.2.2 The Geometric Algebra of the Plane

A 1-dimensional space has insucient geometric structure to be interesting, so we start
in two dimensions, taking two orthonormal basis vectors 1 and 2. These satisfy the
12
relations
(1)2 = 1 (1.36)
(2)2 = 1 (1.37)
and
1
2 = 0: (1:38)
The outer product 1 ^ 2 represents the directed area element of the plane and we assume
that 1, 2 are chosen such that this has the conventional right-handed orientation. This
completes the geometrically meaningful quantities that we can make from these basis
vectors:
1 f1 2g 1 ^ 2: (1:39)
scalar vectors bivector
Any multivector can be expanded in terms of these four basis elements. Addition of
multivectors simply adds the coecients of each component. The interesting expressions
are those involving products of the bivector 1 ^ 2 = 12. We
nd that
(12)1 = ;211 = ;2 (1:40)
(12)2 = 1
and
1(12) = 2 (1:41)
2(12) = ;1:
The only other product to consider is the square of 1 ^ 2,
(1 ^ 2)2 = 1212 = ;1122 = ;1: (1:42)
These results complete the list of the products in the algebra. In order to be completely
explicit, consider how two arbitrary multivectors are multiplied. Let
A = a0 + a11 + a22 + a31 ^ 2 (1.43)
B = b0 + b11 + b22 + b31 ^ 2 (1.44)
then we
nd that
AB = p0 + p11 + p22 + p31 ^ 2 (1:45)
where
p0 = a0b0 + a1b1 + a2b2 ; a3b3
p1 = a0b1 + a1b0 + a3b2 ; a2b3
p2 = a0b2 + a2b0 + a1b3 ; a3b1 (1:46)
p3 = a0b3 + a3b0 + a1b2 ; a2b1:

Calculations rarely have to be performed in this detail, but this exercise does serve to
illustrate how geometric algebras can be made intrinsic to a computer language. One can
even think of (1.46) as generalising Hamilton's concept of complex numbers as ordered
pairs of real numbers.
13
 The square of the bivector 1^2 is ;1, so the even-grade elements z = x + y12 form
a natural subalgebra, equivalent to the complex numbers. Furthermore, 1 ^ 2 has the
geometric eect of rotating the vectors f1 2g in their own plane by 90 clockwise when
multiplying them on their left. It rotates vectors by 90 anticlockwise when multiplying
on their right. (This can be used to de
ne the orientation of 1 and 2).
The equivalence between the even subalgebra and complex numbers reveals a new
interpretation of the structure of the Argand diagram. From any vector r = x1 + y2 we
can form an even multivector z by
z  1r = x + Iy (1:47)
where
I  12: (1:48)
There is therefore a one-to-one correspondence between points in the Argand diagram
and vectors in two dimensions,
r = 1z (1:49)
where the vector 1 de
nes the real axis. Complex conjugation,
z  z~ = r1 = x ; Iy (1:50)
now appears as the natural operation of reversion for the even multivector z. Taking the
complex conjugate of z results in a new vector r given by
r = 1z~
= (z1)~
= (1r1)~
= 1r1
= ;2r2: (1.51)
We will shortly see that equation (1.51) is the geometric algebra representation of a
reection in the 1 axis. This is precisely what one expects for complex conjugation.
This identi
cation of points on the Argand diagram with (Cliord) vectors gives ad-
ditional operational signi
cance to complex numbers of the form exp(i ). The even mul-
tivector equivalent of this is exp(I ), and applied to z gives
eI z = eI 1r
= 1e;I r: (1.52)
But we can now remove the 1, and work entirely in the (real) Euclidean plane. Thus
r0 = e;I r (1:53)
rotates the vector r anticlockwise through an angle . This can be veri
ed from the fact
that
e;I 1 = (cos ; sin I )1 = cos 1 + sin 2 (1:54)
14
and
e;I 2 = cos 2 ; sin 1: (1:55)
Viewed as even elements in the 2-dimensional geometric algebra, exponentials of \imag-
inaries" generate rotations of real vectors. Thinking of the unit imaginary as being a
directed plane segment removes much of the mystery behind the usage of complex num-
bers. Furthermore, exponentials of bivectors provide a very general method for handling
rotations in geometric algebra, as is shown in Chapter 3.

1.2.3 The Geometric Algebra of Space

If we now add a third orthonormal vector 3 to our basis set, we generate the following
geometric objects:
1 f1 2 3g f12 23 31g 123: (1:56)
scalar 3 vectors 3 bivectors trivector
area elements volume element
From these objects we form a linear space of (1 + 3 + 3 + 1) = 8 = 23 dimensions. Many
of the properties of this algebra are shared with the 2-dimensional case since the subsets
f1 2g, f2 3g and f3 1g generate 2-dimensional subalgebras. The new geometric
products to consider are
(12)3 = 123
(123)k = k (123) (1.57)
and
(123)2 = 123123 = 121232 = ;1: (1:58)
These relations lead to new geometric insights:
A simple bivector rotates vectors in its own plane by 90, but forms trivectors
(volumes) with vectors perpendicular to it.
The trivector 1^2^3 commutes with all vectors, and hence with all multivectors.
The trivector (pseudoscalar) 123 also has the algebraic property of squaring to ;1. In
fact, of the eight geometrical objects, four have negative square, f12, 23, 31g and
123. Of these, the pseudoscalar 123 is distinguished by its commutation properties
and in view of these properties we give it the special symbol i,
i  123: (1:59)
It should be quite clear, however, that the symbol i is used to stand for a pseudoscalar
and therefore cannot be used for the commutative scalar imaginary used, for example,
in quantum mechanics. Instead, the symbol j is used for this uninterpreted imaginary,
consistent with existing usage in engineering. The de
nition (1.59) will be consistent with
our later extension to 4-dimensional spacetime.

15
 The algebra of 3-dimensional space is the Pauli algebra familiar from quantum me-
chanics. This can be seen by multiplying the pseudoscalar in turn by 3, 1 and 2 to

nd
(123)3 = 12 = i3
23 = i1 (1:60)
31 = i2
which is immediately identi
able as the algebra of Pauli spin matrices. But we have
arrived at this algebra from a totally dierent route, and the various elements in it have
very dierent meanings to those assigned in quantum mechanics. Since 3-dimensional
space is closest to our perception of the world, it is worth emphasising the geometry of
this algebra in greater detail. A general multivector M consists of the components
M =  + a + ib + i
(1:61)
scalar vector bivector pseudoscalar
where a  ak k and b  bk k . The reason for writing spatial vectors in bold type is
to maintain a visible dierence between spatial vectors and spacetime 4-vectors. This
distinction will become clearer when we consider relativistic physics. The meaning of the
fk g is always unambiguous, so these are not written in bold type.
Each of the terms in (1.61) has a separate geometric signi
cance:
1. scalars are physical quantities with magnitude but no spatial extent. Examples are
mass, charge and the number of words in this thesis.
2. vectors have both a magnitude and a direction. Examples include relative positions,
displacements and velocities.
3. bivectors have a magnitude and an orientation. They do not have a shape. In Fig-
ure 1.1 the bivector a^b is represented as a parallelogram, but any other shape could
have been chosen. In many ways a circle is more appropriate, since it suggests the
idea of sweeping round from the a direction to the b direction. Examples of bivec-
tors include angular momentum and any other object that is usually represented as
an \axial" vector.
4. trivectors have simply a handedness and a magnitude. The handedness tells whether
the vectors in the product a^b^c form a left-handed or right-handed set. Examples
include the scalar triple product and, more generally, alternating tensors.
These four objects are represented pictorially in Figure 1.1. Further details and discussions
are contained in 25] and 44].
The space of even-grade elements of the Pauli algebra,
 =  + ib (1:62)
is closed under multiplication and forms a representation of the quarternion algebra.
Explicitly, identifying i, j , k with i1, ;i2, i3 respectively, the usual quarternion
relations are recovered, including the famous formula
i2 = j 2 = k2 = ijk = ;1: (1:63)
16
 c

a a b a b

scalar vector bivector trivector

line segment plane segment volume segment

Figure 1.1: Pictorial representation of the elements of the Pauli algebra.

The quaternion algebra sits neatly inside the geometric algebra of space and, seen in
this way, the i, j and k do indeed generate 90 rotations in three orthogonal directions.
Unsurprisingly, this algebra proves to be ideal for representing arbitrary rotations in three
dimensions.
Finally, for this section, we recover Gibbs' cross product. Since the  and ^ symbols
have already been assigned meanings, we will use the ? symbol for the Gibbs' prod-
uct. This notation will not be needed anywhere else in this thesis. The Gibbs' product
is given by an outer product together with a duality operation (multiplication by the
pseudoscalar),
a ? b  ;ia^ b: (1:64)
The duality operation in three dimensions interchanges a plane with a vector orthogonal
to it (in a right-handed sense). In the mathematical literature this operation goes under
the name of the Hodge dual. Quantities like a or b would conventionally be called \polar
vectors", while the \axial vectors" which result from cross-products can now be seen to be
disguised versions of bivectors. The vector triple product a ? (b ? c) becomes ;a
(b^c),
which is the 3-dimensional form of an expression which is now legitimate in arbitrary
dimensions. We therefore drop the restriction of being in 3-dimensional space and write
a
(b ^ c) = 12 (ab ^ c ; b ^ ca) (1.65)
= a
bc ; a
cb (1.66)
where we have recalled equation (1.14).

1.2.4 Reections and Rotations

One of the clearest illustrations of the power of geometric algebra is the way in which it
deals with reections and rotations. The key to this approach is that, given any unit vector
n (n2 = 1), an arbitrary vector a can be resolved into parts parallel and perpendicular to
n,
a = n2a
= n(n
a + n ^ a)
= ak + a? (1.67)

17
where
ak = a
nn (1.68)
a? = nn ^ a: (1.69)
The result of reecting a in the hyperplane orthogonal to n is the vector a? ; ak, which
can be written as
a? ; ak = nn ^ a ; a
nn
= ;n
an ; n ^ an
= ;nan: (1.70)
This formula for a reection extends to arbitrary multivectors. For example, if the vectors
a and b are both reected in the hyperplane orthogonal to n, then the bivector a ^ b is
reected to
(;nan) ^ (;nbn) = 21 (nannbn ; nbnnan)
= na ^ bn: (1.71)
In three dimensions, the sign dierence between the formulae for vectors and bivectors
accounts for the dierent behaviour of \polar" and \axial" vectors under reections.
Rotations are built from pairs of reections. Taking a reection
rst in the hyperplane
orthogonal to n, and then in the hyperplane orthogonal to m, leads to the new vector
;m(;nan)m = mnanm
= RaR~ (1.72)
where
R  mn: (1:73)
The multivector R is called a rotor. It contains only even-grade elements and satis
es the
identity
RR~ = RR~ = 1: (1:74)
Equation (1.74) ensures that the scalar product of two vectors is invariant under rotations,
(RaR~ )
(RbR~ ) = hRaRRb ~ R~ i
~ RR
= haRRb ~ i
= habi
= a
b: (1.75)
As an example, consider rotating the unit vector a into another unit vector b, leaving
all vectors perpendicular to a and b unchanged. This is accomplished by a reection
perpendicular to the unit vector half-way between a and b (see Figure 1.2)
n  (a + b)=ja + bj: (1:76)

18
 b

-nan

Figure 1.2: A rotation composed of two reections.

This reects a into ;b. A second reection is needed to then bring this to b, which must
take place in the hyperplane perpendicular to b. Together, these give the rotor
R = bn = j1a++babj = q 1 + ba  (1:77)
2(1 + b
a)
which represents a simple rotation in the a ^ b plane. The rotation is written
b = RaR~ (1:78)
and the inverse transformation is given by
~
a = RbR: (1:79)
The transformation a 7! RaR~ is a very general way of handling rotations. In deriving
this transformation the dimensionality of the space of vectors was at no point speci
ed. As
a result, the transformation law works for all spaces, whatever dimension. Furthermore,
it works for all types of geometric object, whatever grade. We can see this by considering
the image of the product ab when the vectors a and b are both rotated. In this case, ab
is rotated to
~ R~ = RabR:
RaRRb ~ (1:80)
In dimensions higher than 5, an arbitrary even element satisfying (1.74) does not
necessarily map vectors to vectors and will not always represent a rotation. The name
\rotor" is then retained only for the even elements that do give rise to rotations. It can
be shown that all (simply connected) rotors can be written in the form
R = eB=2 (1:81)
where B is a bivector representing the plane in which the rotation is taking place. (This
representation for a rotor is discussed more fully in Chapter 3.) The quantity
b = eB=2ae;B=2 (1:82)

19
is seen to be a pure vector by Taylor expanding in ,
b = a + B
a + 2! B
(B
a) +


:
2
(1:83)
The right-hand side of (1.83) is a vector since the inner product of a vector with a bivector
is always a vector (1.14). This method of representing rotations directly in terms of the
plane in which they take place is very powerful. Equations (1.54) and (1.55) illustrated this
in two dimensions, where the quantity exp(;I ) was seen to rotate vectors anticlockwise
through an angle . This works because in two dimensions we can always write
e;I=2reI=2 = e;I r: (1:84)
In higher dimensions the double-sided (bilinear) transformation law (1.78) is required.
This is much easier to use than a one-sided rotation matrix, because the latter becomes
more complicated as the number of dimensions increases. This becomes clearer in three
dimensions. The rotor
R  exp(;ia=2) = cos(jaj=2) ; i jaaj sin(jaj=2) (1:85)
represents a rotation of jaj = (a2)1=2 radians about the axis along the direction of a.
This is already simpler to work with than 3  3 matrices. In fact, the representation of
a rotation by (1.85) is precisely how rotations are represented in the quaternion algebra,
which is well-known to be advantageous in three dimensions. In higher dimensions the
improvements are even more dramatic.
Having seen how individual rotors are used to represent rotations, we must look at
their composition law. Let the rotor R transform the unit vector a into a vector b,
b = RaR:~ (1:86)
Now rotate b into another vector b0, using a rotor R0. This requires
b0 = R0bR~ 0 = (R0R)a(R0 R)~ (1:87)
so that the transformation is characterised by
R 7! R0R (1:88)
which is the (left-sided) group combination rule for rotors. It is immediately clear that
the product of two rotors is a third rotor,
R0 R(R0R)~ = R0RR~ R~0 = R0R~0 = 1 (1:89)
so that the rotors do indeed form a (Lie) group.
The usefulness of rotors provides ample justi
cation for adding up terms of dierent
grades. The rotor R on its own has no geometric signi
cance, which is to say that no
meaning should be attached to the individual scalar, bivector, 4-vector : : : parts of R.
When R is written in the form R = eB=2, however, the bivector B has clear geometric
signi
cance, as does the vector formed from RaR~ . This illustrates a central feature of
geometric algebra, which is that both geometrically meaningful objects (vectors, planes
: : : ) and the elements that act on them (rotors, spinors : : : ) are represented in the same
algebra.
20
1.2.5 The Geometric Algebra of Spacetime
As a
nal example, we consider the geometric algebra of spacetime. This algebra is
suciently important to deserve its own name | spacetime algebra | which we will
usually abbreviate to STA. The square of a vector is no longer positive de
nite, and we
say that a vector x is timelike, lightlike or spacelike according to whether x2 > 0, x2 = 0
or x2 < 0 respectively. Spacetime consists of a single independent timelike direction, and
three independent spacelike directions. The spacetime algebra is then generated by a set
of orthonormal vectors f g,  = 0 : : : 3, satisfying


= 
= diag(+ ; ; ;): (1:90)
(The signi
cance of the choice of metric signature will be discussed in Chapter 4.) The
full STA is 16-dimensional, and is spanned by the basis
1 f g fk  ik g fi g i: (1:91)
The spacetime bivectors fk g, k = 1 : : : 3 are de
ned by
k  k 0: (1:92)
They form an orthonormal frame of vectors in the space relative to the 0 direction. The
spacetime pseudoscalar i is de
ned by
i  0123 (1:93)
and, since we are in a space of even dimension, i anticommutes with all odd-grade elements
and commutes with all even-grade elements. It follows from (1.92) that
123 = 102030 = 0123 = i: (1:94)
The following geometric signi
cance is attached to these relations. An inertial system
is completely characterised by a future-pointing timelike (unit) vector. We take this to
be the 0 direction. This vector/observer determines a map between spacetime vectors
a = a  and the even subalgebra of the full STA via
a0 = a0 + a (1:95)
where
a0 = a
0 (1.96)
a = a ^ 0: (1.97)
The even subalgebra of the STA is isomorphic to the Pauli algebra of space de
ned in
Section 1.2.3. This is seen from the fact that the k = k 0 all square to +1,
k 2 = k 0k 0 = ;k k 00 = +1 (1:98)
and anticommute,
j k = j 0k 0 = k j 00 = ;k 0j 0 = ;k j (j 6= k): (1:99)
21
There is more to this equivalence than simply a mathematical isomorphism. The way we
think of a vector is as a line segment existing for a period of time. It is therefore sensible
that what we perceive as a vector should be represented by a spacetime bivector. In this
way the algebraic properties of space are determined by those of spacetime.
As an example, if x is the spacetime (four)-vector specifying the position of some point
or event, then the \spacetime split" into the 0-frame gives
x0 = t + x (1:100)
which de
nes an observer time
t = x
0 (1:101)
and a relative position vector
x = x ^ 0: (1:102)
One useful feature of this approach is the way in which it handles Lorentz-scalar quantities.
The scalar x2 can be decomposed into
x2 = x00x
= (t + x)(t ; x)
= t2 ; x2 (1.103)
which must also be a scalar. The quantity t2 ; x2 is now seen to be automatically
Lorentz-invariant, without needing to consider a Lorentz transformation.
The split of the six spacetime bivectors into relative vectors and relative bivectors is
a frame/observer-dependent operation. This can be illustrated with the Faraday bivector
F = 21 F
 ^ 
, which is a full, 6-component spacetime bivector. The spacetime split
of F into the 0-system is achieved by separating F into parts which anticommute and
commute with 0. Thus
F = E + iB  (1:104)
where
E = 21 (F ; 0F0) (1.105)
iB = 2 (F + 0F0):
1 (1.106)
Here, both E and B are spatial vectors, and iB is a spatial bivector. This decomposes F
into separate electric and magnetic
elds, and the explicit appearance of 0 in the formulae
for E and B shows that this split is observer-dependent. In fact, the identi
cation
of spatial vectors with spacetime bivectors has always been implicit in the physics of
electromagnetism through formulae like Ek = Fk0.
The decomposition (1.104) is useful for constructing relativistic invariants from the E
and B
elds. Since F 2 contains only scalar and pseudoscalar parts, the quantity
F 2 = (E + iB)(E + iB)
= E 2 ; B 2 + 2iE
B (1.107)
is Lorentz-invariant. It follows that both E 2 ; B 2 and E
B are observer-invariant
quantities.
22
 Equation (1.94) is an important geometric identity, which shows that relative space
and spacetime share the same pseudoscalar i. It also exposes the weakness of the matrix-
based approach to Cliord algebras. The relation
123 = i = 0123 (1:108)
cannot be formulated in conventional matrix terms, since it would need to relate the
2  2 Pauli matrices to 4  4 Dirac matrices. Whilst we borrow the symbols for the
Dirac and Pauli matrices, it must be kept in mind that the symbols are being used in
a quite dierent context | they represent a frame of orthonormal vectors rather than
representing individual components of a single isospace vector.
The identi
cation of relative space with the even subalgebra of the STA necessitates
developing a set of conventions which articulate smoothly between the two algebras. This
problem will be dealt with in more detail in Chapter 4, though one convention has already
been introduced. Relative (or spatial) vectors in the 0-system are written in bold type to
record the fact that in the STA they are actually bivectors. This distinguishes them from
spacetime vectors, which are left in normal type. No problems can arise for the fk g,
which are unambiguously spacetime bivectors, so these are also left in normal type. The
STA will be returned to in Chapter 4 and will then be used throughout the remainder of
this thesis. We will encounter many further examples of its utility and power.

1.3 Linear Algebra

We have illustrated a number of the properties of geometric algebra, and have given ex-
plicit constructions in two, three and four dimensions. This introduction to the properties
of geometric algebra is now concluded by developing an approach to the study of linear
functions and non-orthonormal frames.

1.3.1 Linear Functions and the Outermorphism

Geometric algebra oers many advantages when used for developing the theory of linear
functions. This subject is discussed in some detail in Chapter 3 of \Cli ord algebra to
geometric calculus" 24], and also in 2] and 30]. The approach is illustrated by taking
a linear function f (a) mapping vectors to vectors in the same space. This function in
extended via outermorphism to act linearly on multivectors as follows,
f (a ^ b ^ : : :^ c)  f (a) ^ f (b) : : :^ f (c): (1:109)
The underbar on f shows that f has been constructed from the linear function f . The def-
inition (1.109) ensures that f is a grade-preserving linear function mapping multivectors
to multivectors.
An example of an outermorphism was encountered in Section 1.2.4, where we consid-
ered how multivectors behave under rotations. The action of a rotation on a vector a was
written as
R(a) = eB=2ae;B=2 (1:110)

23
where B is the plane(s) of rotation. The outermorphism extension of this is simply
R(A) = eB=2Ae;B=2: (1:111)
An important property of the outermorphism is that the outermorphism of the product
of two functions in the product of the outermorphisms,
f g(a)] ^ f g(b)] : : :^ f g(c)] = f g(a) ^ g(b) : : :^ g(c)]
= f g(a ^ b ^ : : :^ c)]: (1.112)
To ease notation, the product of two functions will be written simply as f g(A), so that
(1.112) becomes
fg(a) ^ fg(b) : : : ^ fg(c) = f g(a ^ b ^ : : :^ c): (1:113)
The pseudoscalar of an algebra is unique up to a scale factor, and this is used to de
ne
the determinant of a linear function via
det(f )  f (I )I ;1 (1:114)
so that
f (I ) = det(f )I: (1:115)
This de
nition clearly illustrates the role of the determinant as the volume scale factor.
The de
nition also serves to give a very quick proof of one of the most important properties
of determinants. It follows from (1.113) that
f g(I ) = f (det(g)I )
= det(g)f (I )
= det(f ) det(g)I (1.116)
and hence that
det(fg) = det(f ) det(g): (1:117)
This proof of the product rule for determinants illustrates our third (and
nal) principle
of good design:
Denitions should be chosen so that the most important theorems can be proven
most economically.
The de
nition of the determinant clearly satis
es this criteria. Indeed, it is not hard to
see that all of the main properties of determinants follow quickly from (1.115).
The adjoint to f , written as f , is de
ned by
f (a)  eihf (ei )ai (1:118)
where feig is an arbitrary frame of vectors, with reciprocal frame feig. A frame-invariant
de
nition of the adjoint can be given using the vector derivative, but we have chosen not
to introduce multivector calculus until Chapter 5. The de
nition (1.118) ensures that
b
f (a) = a
(b
eif (ei))
= a
f (b): (1.119)
24
A symmetric function is one for which f = f .
The adjoint also extends via outermorphism and we
nd that, for example,
f (a ^ b) = f (a) ^ f (b)
= ei ^ ej a
f (ei)b
f (ej )
1 ei ^ ej a
f (e )b
f (e ) ; a
f (e )b
f (e )


= 2 i j j i
= 1 ei ^ ej (a ^ b)
f (e ^ e ): (1.120)
2 j i
By using the same argument as in equation (1.119), it follows that
hf (A)B i = hAf (B )i (1:121)
for all multivectors A and B . An immediate consequence is that
det f = hI ;1f (I )i
= hf (I ;1)I i
= det f: (1.122)
Equation (1.121) turns out to be a special case of the more general formulae,
Ar
f (Bs ) = f f (Ar)
Bs] rs (1:123)
f (Ar )
Bs = f Ar
f (Bs)] r  s
which are derived in 24, Chapter 3].
As an example of the use of (1.123) we
nd that
f (f (AI )I ;1) = AIf (I ;1) = A det f (1:124)
which is used to construct the inverse functions,
f ;1(A) = det(f );1 f (AI )I ;1 (1:125)
f ;1(A) = det(f );1 I ;1f (IA):
These equations show how the inverse function is constructed from a double-duality op-
eration. They are also considerably more compact and ecient than any matrix-based
formula for the inverse.
Finally, the concept of an eigenvector is generalized to that of an eigenblade Ar, which
is an r-grade blade satisfying
f (Ar ) = Ar (1:126)
where  is a real eigenvalue. Complex eigenvalues are in general not considered, since
these usually loose some important aspect of the geometry of the function f . As an
example, consider a function f satisfying
f (a) = b (1:127)
f (b) = ;a

25
for some pair of vectors a and b. Conventionally, one might write
f (a + jb) = ;j (a + jb) (1:128)
and say that a + bj is an eigenvector with eigenvalue ;j . But in geometric algebra one
can instead write
f (a ^ b) = b ^ (;a) = a ^ b (1:129)
which shows that a ^ b is an eigenblade with eigenvalue +1. This is a geometrically more
useful result, since it shows that the a^b plane is an invariant plane of f . The unit blade
in this plane generates its own complex structure, which is the more appropriate object
for considering the properties of f .

1.3.2 Non-Orthonormal Frames

At various points in this thesis we will make use of non-orthonormal frames, so a number
of their properties are summarised here. From a set of n vectors feig, we de
ne the
pseudoscalar
En = e1 ^ e2 ^ : : :^ en: (1:130)
The set feig constitute a (non-orthonormal) frame provided En 6= 0. The reciprocal frame
feig satis
es
ei
ej =
ji  (1:131)
and is constructed via 24, Chapter 1]
ei = (;1)i;1e1 ^ : : : e$i : : :^ enE n  (1:132)
where the check symbol on e$i signi
es that this vector is missing from the product. E n is
the pseudoscalar for the reciprocal frame, and is de
ned by
E n = en ^ en;1 ^ : : :^ e1: (1:133)
The two pseudoscalars En and E n satisfy
EnE n = 1 (1:134)
and hence
E n = En =(En )2: (1:135)
The components of the vector a in the ei frame are given by a
ei, so that
a = (a
ei)ei (1:136)
from which we
nd that
2a = 2a
eiei
= eiaei + aeiei
= eiaei + na: (1.137)
26
The fact that eiei = n follows from (1.131) and (1.132). From (1.137) we
nd that
eiaei = (2 ; n)a (1:138)
which extends for a multivector of grade r to give the useful results:
eiAr ei = (;1)r (n ; 2r)Ar 
ei(ei
Ar) = rAr  (1.139)
ei(e ^ Ar) = (n ; r)Ar :
i

For convenience, we now specialise to positive de

nite spaces. The results below are
easily extended to arbitrary spaces through the introduction of a metric indicator function
28]. A symmetric metric tensor g can be de
ned by
g(ei) = ei (1:140)
so that, as a matrix, it has components
gij = ei
ej : (1:141)
Since
g(E n ) = E~n (1:142)
it follows from (1.115) that
det(g) = En E~n = jEnj2: (1:143)
It is often convenient to work with the ducial frame fk g, which is the orthonormal
frame determined by the feig via
ek = h(k ) (1:144)
where h is the unique, symmetric
ducial tensor. The requirement that h be symmetric
means that the fk g frame must satisfy
k
ej = j
ek  (1:145)
which, together with orthonormality, de
nes a set of n2 equations that determine the k
(and hence h) uniquely, up to permutation. These permutations only alter the labels for
the frame vectors, and do not re-de
ne the frame itself. From (1.144) it follows that
ej
ek = h(ej )
k =
kj (1:146)
so that
h(ej ) = j = j : (1:147)
(We are working in a positive de
nite space, so j = j for the orthonormal frame fj g.)
It can now be seen that h is the \square-root" of g,
g(ej ) = ej = h(j ) = h2(ej ): (1:148)
It follows that
det(h) = jEnj: (1:149)
27
The
ducial tensor, together with other non-symmetric square-roots of the metric tensor,

nd many applications in the geometric calculus approach to dierential geometry 28].

We will also encounter a similar object in Chapter 7.
We have now seen that geometric algebra does indeed oer a natural language for
encoding many of our geometric perceptions. Furthermore, the formulae for reections
and rotations have given ample justi
cation to the view that the Cliord product is a
fundamental aspect of geometry. Explicit construction in two, three and four dimensions
has shown how geometric algebra naturally encompasses the more restricted algebraic
systems of complex and quaternionic numbers. It should also be clear from the preceding
section that geometric algebra encompasses both matrix and tensor algebra. The following
three chapters are investigations into how geometric algebra encompasses a number of
further algebraic systems.

28
Chapter 2
Grassmann Algebra and Berezin
Calculus
This chapter outlines the basis of a translation between Grassmann calculus and geo-
metric algebra. It is shown that geometric algebra is sucient to formulate all of the
required concepts, thus integrating them into a single unifying framework. The transla-
tion is illustrated with two examples, the \Grauss integral" and the \Grassmann Fourier
transform". The latter demonstrates the full potential of the geometric algebra approach.
The chapter concludes with a discussion of some further developments and applications.
Some of the results presented in this chapter
rst appeared in the paper \Grassmann
calculus, pseudoclassical mechanics and geometric algebra" 1].

2.1 Grassmann Algebra versus Cliord Algebra

The modern development of mathematics has led to the popularly held view that Grass-
mann algebra is more fundamental than Cliord algebra. This view is based on the idea
(recall Section 1.2) that a Cliord algebra is the algebra of a quadratic form. But, whilst
it is true that every (symmetric) quadratic form de
nes a Cliord algebra, it is certainly
not true that the usefulness of geometric algebra is restricted to metric spaces. Like all
mathematical systems, geometric algebra is subject to many dierent interpretations, and
the inner product need not be related to the concepts of metric geometry. This is best
illustrated by a brief summary of how geometric algebra is used in the study of projective
geometry.
In projective geometry 31], points are labeled by vectors, a, the magnitude of which is
unimportant. That is, points in a projective space of dimension n ; 1 are identi
ed with
rays in a space of dimension n which are solutions of the equation x ^ a = 0. Similarly,
lines are represented by bivector blades, planes by trivectors, and so on. Two products
(originally de
ned by Grassmann) are needed to algebraically encode the principle con-
cepts of projective geometry. These are the progressive and regressive products, which
encode the concepts of the join and the meet respectively. The progressive product of two
blades is simply the outer product. Thus, for two points a and b, the line joining them
together is represented projectively by the bivector a ^ b. If the grades of Ar and Bs sum
to more than n and the vectors comprising Ar and Bs span n-dimensional space, then
29
the join is the pseudoscalar of the space. The regressive product, denoted _, is built from
the progressive product and duality. Duality is de
ned as (right)-multiplication by the
pseudoscalar, and is denoted Ar . For two blades Ar and Bs , the meet is then de
ned by
(Ar _ Bs ) = Ar ^ Bs (2:1)
) Ar _ Bs = Ar
Bs: (2:2)
It is implicit here that the dual is taken with respect to the join of Ar and Bs. As an
example, in two-dimensional projective geometry (performed in the geometric algebra of
space) the point of intersection of the lines given by A and B , where
A = ai (2.3)
B = bi (2.4)
is given by the point
A _ B = ;a
B = ;ia^ b: (2:5)
The de
nition of the meet shows clearly that it is most simply formulated in terms
of the inner product, yet no metric geometry is involved. It is probably unsurprising to
learn that geometric algebra is ideally suited to the study of projective geometry 31].
It is also well suited to the study of determinants and invariant theory 24], which are
also usually thought to be the preserve of Grassmann algebra 49, 50]. For these reasons
there seems little point in maintaining a rigid division between Grassmann and geometric
algebra. The more fruitful approach is to formulate the known theorems from Grassmann
algebra in the wider language of geometric algebra. There they can be compared with, and
enriched by, developments from other subjects. This program has been largely completed
by Hestenes, Sobczyk and Ziegler 24, 31]. This chapter addresses one of the remaining
subjects | the \calculus" of Grassmann variables introduced by Berezin 35].
Before reaching the main content of this chapter, it is necessary to make a few com-
ments about the use of complex numbers in applications of Grassmann variables (particu-
larly in particle physics). We saw in Sections 1.2.2 and 1.2.3 that within the 2-dimensional
and 3-dimensional real Cliord algebras there exist multivectors that naturally play the
r^ole of a unit imaginary. Similarly, functions of several complex variables can be studied in
a real 2n-dimensional algebra. Furthermore, in Chapter 4 we will see how the Schr&odinger,
Pauli and Dirac equations can all be given real formulations in the algebras of space and
spacetime. This leads to the speculation that a scalar unit imaginary may be unneces-
sary for fundamental physics. Often, the use of a scalar imaginary disguises some more
interesting geometry, as is the case for imaginary eigenvalues of linear transformations.
However, there are cases in modern mathematics where the use of a scalar imaginary is
entirely superuous to calculations. Grassmann calculus is one of these. Accordingly, the
unit imaginary is dropped in what follows, and an entirely real formulation is given.

2.2 The Geometrisation of Berezin Calculus

The basis of Grassmann/Berezin calculus is described in many sources. Berezin's \The
method of second quantisation" 35] is one of the earliest and most cited texts, and a
30
useful summary of the main results from this is contained in the Appendices to 39]. More
recently, Grassmann calculus has been extended to the
eld of superanalysis 51, 52], as
well as in other directions 53, 54].
The basis of the approach adopted here is to utilise the natural embedding of Grass-
mann algebra within geometric algebra, thus reversing the usual progression from Grass-
mann to Cliord algebra via quantization. We start with a set of n Grassmann variables
fig, satisfying the anticommutation relations
fi  j g = 0: (2:6)
The Grassmann variables fig are mapped into geometric algebra by introducing a set of
n independent Euclidean vectors feig, and replacing the product of Grassmann variables
by the exterior product,
i j $ ei ^ ej : (2:7)
Equation (2.6) is now satis
ed by virtue of the antisymmetry of the exterior product,
ei ^ ej + ej ^ ei = 0: (2:8)
In this way any combination of Grassmann variables can be replaced by a multivector.
Nothing is said about the interior product of the ei vectors, so the feig frame is completely
arbitrary.
In order for the above scheme to have computational power, we need a translation for
for the calculus introduced by Berezin 35]. In this calculus, dierentiation is de
ned by
the rules
@j =
 (2.9)
ij
@i
@;
j @ =
ij  (2.10)
i
together with the \graded Leibnitz rule",
@ (f f ) = @f1 f + (;1) f1]f @f2  (2:11)
@i 1 2 @i 2 1
@i
where f1] is the parity of f1. The parity of a Grassmann variable is determined by
whether it contains an even or odd number of vectors. Berezin dierentiation is handled
within the algebra generated by the feig frame by introducing the reciprocal frame feig,
and replacing
@ ( $ ei
( (2:12)
@i
so that
@j $ ei
e =
i : (2:13)
j j
@i
It should be remembered that upper and lower indices are used to distinguish a frame from
its reciprocal frame, whereas Grassmann algebra only uses these indices to distinguish
metric signature.
31
 The graded Leibnitz rule follows simply from the axioms of geometric algebra. For
example, if f1 and f2 are grade-1 and so translate to vectors a and b, then the rule (2.11)
becomes
ei
(a ^ b) = ei
ab ; aei
b (2:14)
which is simply equation (1.14) again.
Right dierentiation translates in a similar manner,
@;
) @ $ )
ei (2:15)
i
and the standard results for Berezin second derivatives 35] can also be veri
ed simply.
For example, given that F is the multivector equivalent of the Grassmann variable f ( ),
@ @ f ( ) $ ei
(ej
F ) = (ei ^ ej )
F
@i @j
= ;ej
(ei
F ) (2.16)
shows that second derivatives anticommute, and
 ! ;
@f @ $ (ei
F )
ej = ei
(F
ej ) (2:17)
@ @
i j
shows that left and right derivatives commute.
The
nal concept needed is that of integration over a Grassmann algebra. In Berezin
calculus, this is de
ned to be the same as right dierentiation (apart perhaps from some
unimportant extra factors of j and 2 52]), so that
Z @; @; @;
f ( )dn dn;1 : : : d1  f ( ) @ @ : : : @ : (2:18)
n n;1 1
These translate in exactly the same way as the right derivative (2.12). The only important
formula is that for the total integral
Z
f ( )dn dn;1 : : :d1 $ (: : : ((F
en)
en;1) : : :)
e1 = hFE n i (2:19)
where again F is the multivector equivalent of f ( ), as de
ned by (2.6). Equation (2.19)
picks out the coecient of the pseudoscalar part of F since, if hF in is given by En , then
hFE n i = : (2:20)
Thus the Grassman integral simply returns the coecient .
A change of variables is performed by a linear transformation f , say, with
e0i = f (ei) (2:21)
) En0 = f (En) = det(f )En : (2:22)
32
But the feig must transform under f ;1 to preserve orthonormality, so
ei0 = f ;1(ei) (2:23)
) E n 0 = det(f );1E n  (2:24)
which recovers the usual result for a change of variables in a Grassmann multiple integral.
That En0 E n 0 = 1 follows from the de
nitions above.
In the above manner all the basic formulae of Grassmann calculus can be derived in
geometric algebra, and often these derivations are simpler. Moreover, they allow for the
results of Grassmann algebra to be incorporated into a wider scheme, where they may

nd applications in other
elds. As a further comment, this translation also makes it clear
why no measure is associated with Grassmann integrals: nothing is being added up!

2.2.1 Example I. The \Grauss" Integral

The Grassmann analogue of the Gaussian integral 35],
Z
expf 12 ajk j k g dn : : :d1 = det(a)1=2 (2:25)
where ajk is an antisymmetric matrix, is one of the most important results in applications
of Grassmann algebra. This result is used repeatedly in fermionic path integration, for
example. It is instructive to see how (2.25) is formulated and proved in geometric algebra.
First, we translate
2 a j k $ 2 a ej ^ ek = A say,
1 jk 1 jk (2:26)
where A is a general bivector. The integral now becomes
Z
expf 12 ajk j k g dn : : :d1 $ h(1 + A + A2!^ A + : : :)E n i: (2:27)
It is immediately clear that (2.27) is only non-zero for even n (= 2m say), in which case
(2.27) becomes
h(1 + A + A2!^ A + : : :)E n i = m1 ! h(A)mE ni: (2:28)
This type of expression is considered in Chapter 3 of 24] in the context of the eigen-
value problem for antisymmetric functions. This provides a good illustration of how the
systematic use of a uni
ed language leads to analogies between previously separate results.
In order to prove that (2.28) equals det(a)1=2 we need the result that, in spaces with
Euclidean or Lorentzian signature, any bivector can be written, not necessarily uniquely,
as a sum of orthogonal commuting blades. This is proved in 24, Chapter 3]. Using this
result, we can write A as
A = 1A1 + 2A2 + : : : mAm (2:29)
where
Ai
Aj = ;
ij (2.30)
Ai Aj ] = 0 (2.31)
A1A2 : : : Am = I: (2.32)
33
Equation (2.28) now becomes,
h(12 : : : m)IE ni = det(g);1=212 : : : m (2:33)
where g is the metric tensor associated with the feig frame (1.140).
If we now introduce the function
f (a) = a
A (2:34)
we
nd that 24, Chapter 3]
f (a ^ b) = (a
A) ^ (b
A)
= 21 (a ^ b)
(A ^ A) ; (a ^ b)
AA: (2.35)
It follows that the Ai blades are the eigenblades of f , with
f (Ai) = 2i Ai (2:36)
and hence
f (I ) = f (A1 ^ A2 ^ : : :Am) = (12 : : : m)2I (2:37)
) det(f ) = (12 : : :m )2: (2:38)
In terms of components, however,
fjk = ej
f (ek )
= gjlalk  (2.39)
) det(f ) = det(g) det(a): (2:40)
Inserting (2.40) into (2.33), we have
1 h(A)m E ni = det(a)1=2 (2:41)
m!
as required.
This result can be derived more succinctly using the
ducial frame i = h;1(ei) to
write (2.27) as
1 h(A0)mI i (2:42)
m!
where A0 = 21 ajk j k . This automatically takes care of the factors of det(g)1=2, though it
is instructive to note how these appear naturally otherwise.

2.2.2 Example II. The Grassmann Fourier Transform

Whilst the previous example did not add much new algebraically, it did serve to demon-
strate that notions of Grassmann calculus were completely unnecessary for the problem.
In many other applications, however, the geometric algebra formulation does provide for
important algebraic simpli
cations, as is demonstrated by considering the Grassmann
Fourier transform.
34
 In Grassmann algebra one de
nes Fourier integral transformations between anticom-
muting spaces fk g and fk g by 39]
G( ) = exp
R fj P  k gH ()dn : : :d1
H () = n R expf;j P k k gG( )dn : : : d1
k (2:43)
where n = 1 for n even and j for n odd. The factors of j are irrelevant and can be
dropped, so that (2.43) becomes
R
G( ) = expfRP k k gHP()dn : : : d1
H () = (;1)n expf; k k gG( )dn : : :d1 : (2:44)
These expressions are translated into geometric algebra by introducing a pair of anticom-
muting copies of the same frame, fek g, ffk g, which satisfy
ej
ek = fj
fk (2.45)
ej
fk = 0: (2.46)
The full set fek  fk g generate a 2n-dimensional Cliord algebra. The translation now
proceeds by replacing
k $ ek  (2:47)
k $ f k 
where the fk g have been replaced by elements of the reciprocal frame ff k g. From (2.45),
the reciprocal frames must also satisfy
ej
ek = f j
f k : (2:48)
We next de
ne the bivector (summation convention implied)
J = ej ^ f j = ej ^ fj : (2:49)
The equality of the two expressions for J follows from (2.45),
ej ^ f j = (ej
ek)ek ^ f j
= (fj
fk )ek ^ f j
= ek ^ fk : (2.50)
The bivector J satis
es
ej
J = fj fj
J = ;ej  (2:51)
ej
J = f j f j
J = ;ej 
and it follows that
(a
J )
J = ;a (2:52)
for any vector a in the 2n-dimensional algebra. Thus J generates a complex structure,
which on its own is sucient reason for ignoring the scalar j . Equation (2.52) can be
extended to give
e;J=2aeJ=2 = cos a + sin a
J (2:53)
35
from which it follows that expfJ=2g anticommutes with all vectors. Consequently, this
quantity can only be a multiple of the pseudoscalar and, since expfJ=2g has unit mag-
nitude, we can de
ne the orientation such that
eJ=2 = I: (2:54)
This de
nition ensures that
EnF n = E nFn = I: (2:55)
Finally, we introduce the notation
Ck = k1! hJ k i2k : (2:56)
The formulae (2.44) now translate to
X n
G(e) = (Cj ^ H (f ))
Fn
j =0
Xn
H (f ) = (;1) (C~j ^ G(e))
E n
n (2.57)
j =0

where the convention is adopted that terms where Cj ^ H or C~j ^ G have grade less than
n do not contribute. Since G and H only contain terms purely constructed from the fekg
and ff k g respectively, (2.57) can be written as
Xn
G(e) = (Cn;j ^hH (f )ij )
Fn
j =0
X n
H (f ) = (;1)j (hG(e)ij ^ Cn;j )
E n: (2.58)
j =0
So far we have only derived a formula analogous to (2.44), but we can now go much
further. By using
eJ = cosn + cosn;1 sin C1 + : : : + sinn I (2:59)
to decompose eJ (+=2) = eJ I in two ways, it can be seen that
Cn;r = (;1)r Cr I = (;1)r ICr (2:60)
and hence (using some simple duality relations) (2.58) become
Xn
G(e) = Cj
Hj En
j =0
X n
H (f ) = (;1) Gj
Cj F n:
n (2.61)
j =0
Finally, since G and H are pure in the fek g and ff k g respectively, the eect of dotting
with Ck is simply to interchange each ek for an ;fk and each fk for an ek . For vectors
36
this is achieved by dotting with J . But, from (2.53), the same result is achieved by a
rotation through =2 in the planes of J . Rotations extend simply via outermorphism, so
we can now write
Cj
Hj = eJ=4Hj e;J=4
Gj
Cj = e;J=4Gj eJ=4: (2.62)
We thus arrive at the following equivalent expressions for (2.57):
G(e) = eJ=4H (f )e;J=4En
H (f ) = (;1)ne;J=4 G(e)eJ=4F n: (2.63)
The Grassmann Fourier transformation has now been reduced to a rotation through =2
in the planes speci
ed by J , followed by a duality transformation. Proving the \inversion"
theorem (i.e. that the above expressions are consistent) amounts to no more than carrying
out a rotation, followed by its inverse,


G(e) = eJ=4 (;1)ne;J=4 G(e)eJ=4F n e;J=4 En
= G(e)E nEn = G(e): (2.64)
This proof is considerably simpler than any that can be carried out in the more restrictive
system of Grassmann algebra.

2.3 Some Further Developments

We conclude this chapter with some further observations. We have seen how most aspects
of Grassmann algebra and Berezin calculus can be formulated in terms of geometric al-
gebra. It is natural to expect that other
elds involving Grassmann variables can also be
reformulated (and improved) in this manner. For example, many of the structures stud-
ied by de Witt 52] (super-Lie algebras, super-Hilbert spaces) have natural multivector
expressions, and the cyclic cohomology groups of Grassmann algebras described by Co-
quereaux, Jadczyk and Kastler 53] can be formulated in terms of the multilinear function
theory developed by Hestenes & Sobczyk 24, Chapter 3]. In Chapter 5 the formulation of
this chapter is applied Grassmann mechanics and the geometric algebra approach is again
seen to oer considerable bene
ts. Further applications of Grassmann algebra are consid-
ered in Chapter 3, in which a novel approach to the theory of linear functions is discussed.
A clear goal for future research in this subject is to
nd a satisfactory geometric algebra
formulation of supersymmetric quantum mechanics and
eld theory. Some preliminary
observations on how such a formulation might be achieved are made in Chapter 5, but a
more complete picture requires further research.
As a
nal comment, it is instructive to see how a Cliord algebra is traditionally built
from the elements of Berezin calculus. It is well known 35] that the operators
Q^ k = k + @@  (2:65)
k

37
satisfy the Cliord algebra generating relations
fQ^ j  Q^ k g = 2
jk  (2:66)
and this has been used by Sherry to provide an alternative approach to quantizing a
Grassmann system 55, 56]. The geometric algebra formalism oers a novel insight into
these relations. By utilising the
ducial tensor, we can write
Q^ k a( ) $ ek ^ A + ek
A = h(k ) ^ A + h;1 (k )
A
= h(k ^ h;1(A)) + h(sk
h;1(A))
= hk h;1(A)] (2.67)
where A is the multivector equivalent of a( ) and we have used (1.123). The operator Q^ k
thus becomes an orthogonal Cliord vector (now Cliord multiplied), sandwiched between
a symmetric distortion and its inverse. It is now simple to see that
fQ^ j  Q^ k ga( ) $ h(2j
k h;1(A)) = 2
jk A: (2:68)
The above is an example of the ubiquity of the
ducial tensor in applications involving
non-orthonormal frames. In this regard it is quite surprising that the
ducial tensor is
not more prominent in standard expositions of linear algebra.
Berezin 35] de
nes dual operators to the Q^ k by
P^k = ;j (k ; @@ ) (2:69)
k
though a more useful structure is derived by dropping the j , and de
ning
P^k = k ; @@ : (2:70)
k
These satisfy
fP^j  P^k g = ;2
jk (2:71)
and
fP^j  Q^ kg = 0 (2:72)
so that the P^k  Q^ k span a 2n-dimensional balanced algebra (signature n n). The P^k can be
translated in the same manner as the Q^ k , this time giving (for a homogeneous multivector)
P^k a( ) $ ek ^ Ar ; ek
Ar = (;1)r hh;1(Ar )k ]: (2:73)
The fk g frame now sits to the right of the multivector on which it operates. The factor
of (;1)r accounts for the minus sign in (2.71) and for the fact that the left and right
multiples anticommute in (2.72). The Q^ k and P^k can both be given right analogues
if desired, though this does not add anything new. The fQ^ kg and fP^k g operators are
discussed more fully in Chapter 4, where they are related to the theory of the general
linear group.

38
Chapter 3
Lie Groups and Spin Groups
This chapter demonstrates how geometric algebra provides a natural arena for the study of
Lie algebras and Lie groups. In particular, it is shown that every matrix Lie group can be
realised as a spin group. Spin groups consist of even products of unit magnitude vectors,
and arise naturally from the geometric algebra treatment of reections and rotations
(introduced in Section 1.2.4). The generators of a spin group are bivectors, and it is
shown that every Lie algebra can be represented by a bivector algebra. This brings the
computational power of geometric algebra to applications involving Lie groups and Lie
algebras. An advantage of this approach is that, since the rotors and bivectors are all
elements of the same algebra, the discussion can move freely between the group and its
algebra. The spin version of the general linear group is studied in detail, revealing some
novel links with the structures of Grassmann algebra studied in Chapter 2. An interesting
result that emerges from this work is that every linear transformation can be represented
as a (geometric) product of vectors. Some applications of this result are discussed. A
number of the ideas developed in this chapter appeared in the paper \Lie groups as spin
groups" 2].
Throughout this chapter, the geometric algebra generated by p independent vectors
of positive norm and q of negative norm is denoted as <pq. The grade-k subspace of
this algebra is written as <kpq and the space of vectors, <1pq , is abbreviated to <pq. The
Euclidean algebra <n0 is abbreviated to <n, and the vector space <1n is written as <n .
Lie groups and their algebras are labeled according to the conventions of J.F. Cornwell's
\Group Theory in Physics", Vol. 2 57]. (A useful table of these conventions is found on
page 392).

3.1 Spin Groups and their Generators

In this chapter we are interested in spin groups. These arise from the geometric algebra
representation of orthogonal transformations | linear functions on <pq which preserve
inner products. We start by considering the case of the Euclidean algebra <n. The
simplest orthogonal transformation of <n is a reection in the hyperplane perpendicular
to some unit vector n,
n(a) = ;nan (3:1)

39
where we have recalled equation (1.70). (A convenient feature of the underbar/overbar
notation for linear functions is that a function can be written in terms of the multivector
that determines it.) The function n satis
es
n(a)
n(b) = hnannbni = a
b (3:2)
and so preserves the inner product. On combining n with a second reection m, where
m(a) = ;mam (3:3)
the function
m n(a) = mnanm (3:4)
is obtained. This function also preserves inner products, and in Section 1.2.4 was identi
ed
as a rotation in the m ^ n plane. The group of even products of unit vectors is denoted
spin(n). It consists of all even multivectors (rotors) satisfying
RR~ = 1 (3:5)
and such that the quantity RaR~ is a vector for all vectors a. The double-sided action of
a rotor R on a vector a is written as
R(a) = RaR~ (3:6)
and the R form the group of rotations on <n , denoted SO(n). The rotors aord a spin-1/2
description of rotations, hence rotor groups are referred to as spin groups.
In spaces with mixed signature the situation is slightly more complicated. In order to
take care of the fact that a unit vector can now have n2 = 1, equation (3.1) must be
modi
ed to
n(a) = ;nan;1: (3:7)
Taking even combinations of reections now leads to functions of the type
M (a) = MaM ;1 (3:8)
as opposed to MaM~ . Again, the spin group spin(p q) is de
ned as the group of even
products of unit vectors, but its elements now satisfy M M~ = 1. The term \rotor"
is retained for elements of spin(p q) satisfying RR~ = 1. The subgroup of spin(p q)
containing just the rotors is called the rotor group (this is sometimes written as spin+ (p q)
in the literature). The action of a rotor on a vector a is always de
ned by (3.6). Spin
groups and rotor groups are both Lie groups and, in a space with mixed signature, the spin
group diers from the rotor group only by a direct product with an additional subgroup
of discrete transformations.
The generators of a spin group are found by adapting the techniques found in any of
the standard texts of Lie group theory (see 57], for example). We are only interested
in the subgroup of elements connected to the identity, so only need to consider the rotor
group. We introduce a one-parameter set of rotors R(t), so that
R(t)aR~ (t) = hR(t)aR~ (t)i1 (3:9)
40
for all vectors a and for all values of the parameter t. On dierentiating with respect to
t, we
nd that the quantity
R0aR~ + RaR~0 = R0R~ (RaR~ ) + (RaR~ )RR~0
= R0R~ (RaR~ ) ; (RaR~)R0 R~ (3.10)
must be a vector, where we have used RR~ = 1 to deduce that
R0R~ = ;RR~0 : (3:11)
The commutator of R0 R~ with an arbitrary vector therefore results in a vector, so R0R~ can
only contain a bivector part. (R0R~ cannot contain a scalar part, since (R0R~ )~ = ;R0R~ .)
The generators of a rotor group are therefore a set of bivectors in the algebra containing
the rotors.
A simple application of the Jacobi identity gives, for vectors a, b, c, and d,
(a ^ b)  (c ^ d) = (a ^ b)
c] ^ d ; (a ^ b)
d] ^ c (3:12)
so the commutator product of two bivector blades results in a third bivector. It follows
that the space of bivectors is closed under the commutator product, and hence that the
bivectors (together with the commutator product) form the Lie algebra of a spin group.
It should be noted that the commutator product, , in equation (3.12) diers from the
commutator bracket by a factor of 1=2. The commutator product is simpler to use, since
it is the bivector part of the full geometric product of two bivectors A and B :
AB = A
B + A  B + A ^ B (3:13)
where
A
B + A ^ B = 21 (AB + BA) (3.14)
A  B = 12 (AB ; BA): (3.15)
For this reason the commutator product will be used throughout this chapter.
Since the Lie algebra of a spin group is generated by the bivectors, it follows that all
rotors simply connected to the identity can be written in the form
R = eB=2 (3:16)
which ensures that
R~ = e;B=2 = R;1: (3:17)
The form of a rotor given by equation (3.16) was found in Section 1.2.4, where rotations
in a single Euclidean plane were considered. The factor of 1=2 is included because rotors
provide a half-angle description of rotations. In terms of the Lie algebra, the factor of
1=2 is absorbed into our use of the commutator product, as opposed to the commutator
bracket.

41
 It can be shown that, in positive de
nite spaces, all rotors can be written in the form
of (3.16). The bivector B is not necessarily unique, however, as can be seen by considering
the power series de
nition of the logarithm,
H 3 H5
ln X = 2H + 3 + 5 +


] (3:18)
where
H=X X ; 1: (3:19)
+1
It is implicit in this formula that 1 + X is invertible, and the logarithm will not be well-
de
ned if this is not the case. For example, the pseudoscalar I in <40 is a rotor (I I~ = 1),
the geometric eect of which is to reverse the sign of all vectors. But 1+ I is not invertible,
since (1 + I )2 = 2(1 + I ). This manifests itself as a non-uniqueness in the logarithm of I
| given any bivector blade B satisfying B 2 = ;1, I can be written as
I = expfB (1 ; I ) 2 g: (3:20)
Further problems can arise in spaces with mixed signature. In the spacetime algebra, for
example, whilst the rotor
R = (0 + 1 ; 2)2 = 1 + (0 + 1)2 (3:21)
can be written as
R = expf(0 + 1)2g (3:22)
the rotor
; R = expf12 2 gR = ;1 ; (0 + 1)2 (3:23)
cannot be written as the exponential of a bivector. The problem here is that the series
for ln(;X ) is found by replacing H by H ;1 in equation (3.18) and, whilst 1 + R =
2 + (0 + 1)2 is invertible, 1 ; R = ;(0 + 1)2 is null and therefore not invertible.
Further examples of rotors with no logarithm can be constructed in spaces with other
signatures. Near the identity, however, the Baker-Campbell-Hausdor formula ensures
that, for suitably small bivectors, one can always write
eA=2eB=2 = eC=2: (3:24)
So, as is usual in Lie group theory, the bulk of the properties of the rotor (and spin)
groups are transferred to the properties of their bivector generators.
In the study of Lie groups and their algebras, the adjoint representation plays a partic-
ularly important role. The adjoint representation of a spin group is formed from functions
mapping the Lie algebra to itself,
AdM (B )  MBM ;1 = M (B ): (3:25)
The adjoint representation is therefore formed by the outermorphism action of the linear
functions M (a) = MaM ;1 . For the rotor subgroup, we have
AdR(B ) = R(B ) = RB R: ~ (3:26)
42
It is immediately seen that the adjoint representation satis
es
AdM1 AdM2 (B )] = AdM1M2 (B ): (3:27)
The adjoint representation of the Lie group induces a representation of the Lie algebra as
adA=2(B ) = A  B (3:28)
or
adA (B ) = 2A  B: (3:29)
The Jacobi identity ensures that
2 (adA adB ; adB adA )(C ) = 2A  (B  C ) ; B  (A  C )]
1
= 2(A  B )  C
= adAB (C ): (3.30)
The Killing form is constructed by considering adA as a linear operator on the space of
bivectors, and de
ning
K (A B ) = Tr(adAadB ): (3:31)
For the case where the Lie algebra is the set of all bivectors, we can de
ne a basis set of
bivectors as BK = ei ^ ej (i < j ) with reciprocal basis B K = ej ^ ei . Here, the index K
is a simplicial index running from 1 to n(n ; 1)=2 over all combinations of i and j with
i < j . A matrix form of the adjoint representation is now given by
(adA)K J = 2(A  BJ )
B K (3:32)
so that the Killing form becomes
n(nX
;1)=2
K (A B ) = 4 (A  BJ )
B K (B  BK )
B J
JK =1
= 2A  (B  (ei ^ ej ))]
(ej ^ ei)
= hABei ^ ej ej ^ ei ; Aei ^ ej Bej ^ eii (3.33)
Now,
ei ^ ej ej ^ ei = eiej ej ^ ei
= n(n ; 1) (3.34)
and
ei ^ ej Bej ^ ei = eiej Bej ^ ei
= eiej Bej ei ; eiej ei
ej B
= (n ; 4)2 ; n]B (3.35)
where we have used equations (1.139). On recombining (3.34) and (3.35), the Killing form
on a bivector algebra becomes
K (A B ) = 8(n ; 2)hAB i (3:36)
43
 Eij = ei ^ ej (i < j i j = 1 : : : p)
Fij = fi ^ fj (i < j i j = 1 : : : q)
Gij = ei ^ fj (i = 1 : : : p, j = 1 : : : q):

Table 3.1: Bivector Basis for so(p,q)

and so is given by the scalar part of the geometric product of two bivectors. The constant
is irrelevant, and will be ignored. The same form will be inherited by all sub-algebras of
the bivector algebra, so we can write
K (A B )  A
B (3:37)
as the Killing form for any bivector (Lie) algebra. This product is clearly symmetric, and
is invariant under the adjoint action of any of the group elements. The fact that both
the scalar and bivector parts of the geometric product of bivectors now have roles in the
study of Lie algebras is a useful uni
cation | rather than calculate separate commutators
and inner products, one simply calculates a geometric product and reads o the parts of
interest.
As an example, the simplest of the spin groups is the full rotor group spin(p q) in
some <pq . The Lie algebra of spin(p q) is the set of bivectors <2pq . By introducing a
basis set of p positive norm vectors feig and q negative norm vectors ffig, a basis set
for the full Lie algebra is given by the generators in Table 3.1. These generators provide
a bivector realisation of the Lie algebra so(p,q). When the feig and ffig are chosen to
be orthonormal, it is immediately seen that the Killing form has (p(p ; 1) + q(q ; 1))=2
bivectors of negative norm and pq of positive norm. The sum of these is n(n ; 1)=2, where
n = p + q. The algebra is unaected by interchanging the signature of the space from <pq
to <qp. Compact Killing metrics arise from bivectors in positive (or negative) de
nite
vector spaces.
We now turn to a systematic study of the remaining spin groups and their bivector
generators. These are classi
ed according to their invariants which, for the classical
groups, are non-degenerate bilinear forms. In the geometric algebra treatment, bilinear
forms are determined by certain multivectors, and the groups are studied in terms of these
invariant multivectors.

3.2 The Unitary Group as a Spin Group

It has already been stressed that the use of a unit scalar imaginary frequently hides useful
geometric information. This remains true for the study of the unitary groups. The basic
idea needed to discuss the unitary groups was introduced in Section 2.2.2. One starts in
an n-dimensional space of arbitrary signature, and introduces a second (anticommuting)
copy of this space. Thus, if the set feig form a frame for the
rst space, the second space
is generated by a frame ffig satisfying equations (2.45) and (2.46). The spaces are related
by the \doubling" bivector J , de
ned as (2.49)
J = ej ^ f j = ej ^ fj : (3:38)
44
We recall from Section 2.2.2 that J satis
es
(a
J )
J = ;a (3:39)
for all vectors a in the 2n-dimensional space. From J the linear function J is de
ned as
J (a)  a
J = e;J=4aeJ=4: (3:40)
The function J satis
es
J 2(a) = ;a (3:41)
and provides the required complex structure | the action of J being equivalent to mul-
tiplication of a complex vector by j .
An important property of J is that it is independent of the frame from which it was
constructed. To see this, consider a transformation h taking the feig to a new frame
e0i = h(ei) (3.42)
;1 i
) e = h (e )
i 0
(3.43)
so that the transformed J is
J 0 = h(ej ) ^ h;1(f j )
= (ek ek
h(ej )) ^ h;1(f j ): (3.44)
But h(ej ) remains in the space spanned by the feig, so
ek
h(ej ) = f k
h(fj )
= fj
h(f k ) (3.45)
and now


J 0 = ek ^ fj
h(f k )h;1(f j )
= ek ^ h;1 h(f k )
= J: (3.46)
We now turn to a study of the properties of the outermorphism of J . A simple
application of the Jacobi identity yields
(a ^ b)  J = (a
J ) ^ b + a ^ (b
J )
= J (a) ^ b + a ^ J (b) (3.47)
and, using this result again, we derive
(a ^ b)  J ] J = J 2(a) ^ b + J (a) ^ J (b) + J (a) ^ J (b) + a ^ J 2(b)
= 2(J (a ^ b) ; a ^ b): (3.48)
It follows that
J (B ) = B + 12 (B  J )  J (3:49)
45
for all bivectors B . If the bivector B commutes with J , then we see that
J (B ) = B (3:50)
so that B is an eigenbivector of J with eigenvalue +1. The converse is also true | all
eigenbivectors of J with eigenvalue +1 commute with J . This result follows by using
J (B ) = B (3:51)
to write the eigenbivector B as
B = 12 (B + J (B )): (3:52)
But, for a blade a ^ b,
a ^ b + J (a ^ b)]  J = J (a) ^ b + a ^ J (b) + J 2(a) ^ J (b) + J (a) ^ J 2(b)
= 0 (3.53)
and the same must be true for all sums of blades. All bivectors of the form B + J (B )
therefore commute with J , from which it follows that all eigenbivectors of J also commute
with J . In fact, since the action of J on bivectors satis
es
J 2(a ^ b) = J 2(a) ^ J 2(b) = (;a) ^ (;b) = a ^ b (3:54)
any bivector of the form B + J (B ) is an eigenbivector of J .
The next step in the study of the unitary group is to
nd a representation of the
Hermitian inner product. If we consider a pair of complex vectors u and v with components
fuk g and fvkg, where
uk = xk + jyk (3:55)
v = r + js 
k k k
then
(u v) = uyk vk = xk rk + yk sk + j (xk sk ; yk rk ): (3:56)
Viewed as a pair of real products, (3.56) contains a symmetric and a skew-symmetric
term. The symmetric part is the inner product in our 2n-dimensional vector space. Any
skew-symmetric inner product can be written in the form (a ^ b)
B , where B is some
bivector. For the Hermitian inner product this bivector is J , which follows immediately
from considering the real part of the inner product of (ja b). The form of the Hermitian
inner product in our 2n-dimensional vector space is therefore
(a b) = a
b + (a ^ b)
Jj: (3:57)
This satis
es
(b a) = a
b ; (a ^ b)
Jj = (a b)  (3:58)
as required. The introduction of the j disguises the fact that the Hermitian product
contains two separate bilinear forms, both of which are invariant under the action of the

46
unitary group. All orthogonal transformations leave a
b invariant, but only a subset will
leave (a ^ b)
J invariant as well. These transformations must satisfy


f (a) ^ f (b)
J = (a ^ b)
f (J ) = (a ^ b)
J (3:59)
for all vectors a and b. The invariance group therefore consists of all orthogonal transfor-
mations whose outermorphism satis
es
f (J ) = J: (3:60)
This requirement excludes all discrete transformations, since a vector n will only generate
a symmetry if
n(J ) = nJn;1 = J
) n
J = 0 (3.61)
and no such vector n exists. It follows that the symmetry group is constructed entirely
from the double sided action of the elements of the spin group which satisfy
MJ = JM: (3:62)
These elements aord a spin group representation of the unitary group.
Equation (3.62) requires that, for a rotor R simply connected to the identity, the
bivector generator of R commutes with J . The Lie algebra of a unitary group is therefore
realised by the set of bivectors commuting with J , which we have seen are also eigen-
bivectors of J . Given an arbitrary bivector B , therefore, the bivector
BJ = B + J (B ) (3:63)
is contained in the bivector algebra of u(p,q). This provides a quick method for writing
down a basis set of generators. It is convenient at this point to introduce an orthonormal
frame of vectors fei fig satisfying
ei
ej = fi
fj = ij (3.64)
ei
fj = 0 (3.65)
where ij = i
jk (no sum) and i is the metric indicator (= 1 or ;1). This frame is used
to write down a basis set of generators which are orthogonal with respect to the Killing
form. Such a basis for u(p,q) is contained in Table 3.2. This basis has dimension
2 n(n ; 1) + 2 n(n ; 1) + n = n : (3:66)
1 1 2

Of these, p2 + q2 bivectors have negative norm, and 2pq have positive norm.
The algebra of Table 3.2 contains the bivector J , which commutes with all other
elements of the algebra and generates a U(1) subgroup. This is factored out to give the
basis for su(p,q) contained in Table 3.3. The Hi are written in the form given to take care
of the metric signature of the vector space. When working in <2n one can simply write
Hi = Ji ; Ji+1: (3:67)
47
 Eij = eiej + fi fj (i < j = 1 : : : n)
Fij = eifj ; fiej 00

Ji = eifi (i = 1 : : : n):

Table 3.2: Bivector Basis for u(p,q)

Eij = eiej + fifj (i < j = 1 : : : n )

Fij = eifj ; fiej 00

Hi = eif i ; ei+1f i+1 (i = 1 : : : n ; 1):

Table 3.3: Bivector Basis for su(p,q)

The use of Hermitian forms hides the properties of J in the imaginary j , which makes
it dicult to relate the unitary groups to other groups. In particular, the group of linear
transformations on <2n whose outermorphism leaves J invariant form the symplectic
group Sp(n,R). Since U(n) leaves a
b invariant as well as J , we obtain the group relation
U(n) 
= O(2n) \ Sp(n,R): (3:68)
More generally, we
nd that
U(p q) 
= O(2p 2q) \ Sp(p q,R) (3:69)
where Sp(p q,R) is group of linear transformations leaving J invariant in the mixed-
signature space <2p2q. The geometric algebra approach to Lie group theory makes rela-
tions such as (3.69) quite transparent. Furthermore, the doubling bivector J appears in
many other applications of geometric algebra | we saw one occurrence in Section 2.2.2 in
the discussion of the Grassmann-Fourier transform. Other applications include multipar-
ticle quantum mechanics and Hamiltonian mechanics 32]. Consistent use of geometric
algebra can reveal these (often hidden) similarities between otherwise disparate
elds.

3.3 The General Linear Group as a Spin Group

The development of the general linear group as a spin group parallels that of the unitary
groups. Again, the dimension of the space is doubled by introducing a second space, but
this time the second space has opposite signature. This leads to the development of a
Grassmann structure, as opposed to a complex structure. Vectors in <pq are then replaced
by null vectors in <nn , where n = p + q. Since a (positive) dilation of a null vector can
also be given a rotor description, it becomes possible to build up a rotor description of the
entire general linear group from combinations of dilations and orthogonal transformations.
The required construction is obtained by starting with a basis set of vectors feig in
<pq, and introducing a second space of opposite signature. The second space is generated
48
by a set of vectors ffig satisfying
ei
ej = ;fi
fj (3.70)
ei
fj = 0 (3.71)
and the full set fei fig form a basis set for <nn . The vector space <nn is split into two
null spaces by introducing the bivector K de
ned by
K = ej ^ fj = ;ej ^ f j : (3:72)
Again, K is independent of the initial choice of the feig frame. The bivector K determines
the linear function K by
K (a)  a
K: (3:73)
The function K satis
es
K (ei) = fi K (fi ) = ei (3:74)
K (ei) = ;f i K (f i ) = ;ei
and
K 2(a) = (a
K )
K = a (3:75)
for all vectors a.
Proceeding as for the complexi
cation bivector J we
nd that, for an arbitrary bivector
B,
K (B ) = ;B + 12 (B  K )  K: (3:76)
Any bivector commuting with K is therefore an eigenbivector of K , but now with eigen-
value ;1.
An arbitrary vector a in <nn can be decomposed into a pair of null vectors,
a = a+ + a; (3:77)
where
a+ = 21 (a + K (a)) (3.78)
a; = 12 (a ; K (a)): (3.79)
That a+ is null follows from


(a+)2 = 41 a2 + 2a
(a
K ) + (a
K )
(a
K )
= 14 (a2 ; (a
K )
K ]
a)
= 14 (a2 ; a2)
= 0 (3.80)
and the same holds for a;. The scalar product between a+ and a; is, of course, non-zero:
a+
a; = 14 (a2 ; (a
K )2) = 12 a2: (3:81)
49
This construction decomposes <nn into two separate null spaces, V n and V n , de
ned by
K (a) = a 8a 2 V n (3:82)
K (a) = ;a 8a 2 V n 
so that
<nn = V n
V n : (3:83)
A basis is de
ned for each of V n and V n by
wi = 21 (ei + K (ei)) (3.84)
w i = 21 (ei ; K (ei)) (3.85)
respectively. These basis vectors satisfy
wi
wj = w i
w j = 0 (3:86)
and
w i
wj = 21
ji : (3:87)
In conventional accounts, the space V n would be recognised as a Grassmann algebra (all
vector generators anticommute), with V n identi
ed as the dual space of functions acting
on V n. In Chapter 2 we saw how both V n and V n can be represented in terms of functions
in a single n-dimensional algebra. Here, a dierent construction is adopted in which the V n
and V n spaces are kept maximally distinct, so that they generate a 2n-dimensional vector
space. This is the more useful approach for the study of the Lie algebra of the general
linear group. We shall shortly see how these two separate approaches are reconciled by
setting up an isomorphism between operations in the two algebras.
We are interested in the group of orthogonal transformations which keep the V n and
V spaces separate. For a vector a in V n, the orthogonal transformation f must then
n
satisfy
f (a) = f (a)
K: (3:88)
But, since a = a
K and f ;1 = f , equation (3.88) leads to
a
K = f f (a)
K ]
= a
f (K ) (3.89)
which must hold for all a. It follows that
f (K ) = K (3:90)
and we will show that the f satisfying this requirement form the general linear group
GL(n,R). The orthogonal transformations satisfying (3.90) can each be given a spin de-
scription, which enables the general linear group to be represented by a spin group. The
elements of this spin group must satisfy
MK = KM: (3:91)

50
 Eij = eiej ; e^ie^j (i < j = 1 : : : n)
Fij = eie^j ; e^iej 00

Ji = eie^i (i = 1 : : : n):

Table 3.4: Bivector Basis for gl(n,R)

Eij = eiej ; e^ie^j (i < j = 1 : : : n)

Fij = eie^j ; e^iej 00

Hi = eie^i ; ei+1e^i+1 (i = 1 : : : n ; 1):

Table 3.5: Bivector Basis for sl(n,R)

The generators of the rotor part of the spin group are therefore the set of bivectors which
commute with K , which are eigenbivectors of K with eigenvalue ;1.
Before writing down an orthogonal basis for the Lie algebra, it is useful to introduce
some further notation. We now take feig to be an orthonormal basis for the Euclidean
algebra <n, and fe^ig to be the corresponding basis for the anti-Euclidean algebra <0n.
These basis vectors satisfy
ei
ej =
ij = ;e^i
e^j (3:92)
e
e^ = 0:
i j
The hat also serves as a convenient abbreviation for the action of K on a vector a,
a^  K (a): (3:93)
Since all bivectors in the Lie algebra of GL(n,R) are of the form B ; K (B ), an orthogonal
basis for the Lie algebra can now be written down easily. Such a basis is contained in
Table 3.4. The algebra in Table 3.4 includes K , which generates an abelian subgroup.
This is factored out to leave the Lie algebra sl(n,R) contained in Table 3.5.
The form of the Lie algebra for the group GL(n,R) is clearly very close to that for
U(n) contained in Table 3.2. The reason can be seen by considering the bilinear form
generated by the bivector K ,
(a b) = a
K (b): (3:94)
If we decompose a and b in the orthonormal basis of (3.92),
a = xiei + yie^i (3.95)
b = riei + sie^i (3.96)
we
nd that
(a b) = xisi ; yiri  (3:97)
which is the component form of the symplectic norm in <2n. We thus have the group
relation
GL(n,R)  = O(n,n) \ Sp(n,R) (3:98)
51
which is to be compared with (3.68) and (3.69). The dierences between the Lie algebras
of GL(n,R) and U(n) are due solely to the metric signature of the underlying vector
space which generates the bivector algebra. It follows that both Lie algebras have the
same complexi
cation, since complexi
cation removes all dependence on signature. In
the theory of the classi
cation of the semi-simple Lie algebras, the complexi
cation of the
su(n) and sl(n,R) algebras is denoted An;1 .
An alternative basis for sl(n,R) can be given in terms of the fwig and fwi g null frames,
which are now de
ned as
wi = 21 (ei + e^i) (3:99)
w = 1 (e ; e^ ):
i 2 i i
The fwig and fwi g frames satisfy
wi
wj = wi
wj = 0 (3:100)
and
wiwj + wj wi =
ij  (3:101)
which are identi
able as the relations of the algebra of fermionic creation and annihilation
operators. The pseudoscalars for the the V n and V n spaces are de
ned by
Wn = w1w2 : : :wn (3:102)
Wn = w1 w2 : : : wn
respectively. If we now de
ne
Iij+ = 21 (Eij + Fij )
= 12 (ei ; e^i)(ej + e^j )
= 2wi wj (3.103)
and
Iij; = 21 (Eij ; Fij )
= 12 (ei + e^i)(ej ; e^j )
= ;2wj wi (3.104)
we see that a complete basis for sl(n,R) is de
ned by the set fIij+ Iij; Hi g. This corresponds
to the Chevalley basis for An;1. Furthermore, a complete basis set of generators for
GL(n,R) is given by the set fwi ^ wj g, de
ned over all i, j . This is perhaps the simplest
of the possible basis sets for the Lie algebra, though it has the disadvantage that it is not
orthogonal with respect to the Killing form.
We now turn to a proof that the subgroup of the spin group which leaves K invariant
does indeed form a representation of GL(n,R). With a vector a in <n represented by the
null vector a+ = (a + a^) in <nn , we must prove that an arbitrary linear transformation
of a, a 7! f (a), can be written in <nn as
a+ 7! Ma+ M ;1 (3:105)

52
where M is a member of the spin group spin(n n) which commutes with K . We start by
considering the polar decomposition of an arbitrary matrix M . Assuming that det M 6= 0,
the matrix M M ' can be written (not necessarliy uniquely) as
M M' = S S' (3:106)
where S is an orthogonal transformation (which can be arranged to be a rotation matrix),
and is a diagonal matrix with positive entries. One can now write
M = S 1=2R (3:107)
where 1=2 is the diagonal matrix of positive square roots of the entries of and R is a
matrix de
ned by
R = ;1=2S' M : (3:108)
The matrix R satis
es
RR' = ;1=2S' M MS ' ;1=2
= ;1=2 ;1=2
= I (3.109)
and so is also orthogonal. It follows from (3.107) that an arbitrary non-singular matrix
can be written as a diagonal matrix with positive entries sandwiched between a pair of
orthogonal matrices. As a check, this gives n2 degrees of freedom. To prove the desired
result, we need only show that orthogonal transformations and positive dilations can be
written in the form of equation (3.105).
We
rst consider rotations. The Eij generators in Table 3.4 produce rotors of the form
R = expf(E ; E^ )=2g (3:110)
where
E = ij Eij (3:111)
and the ij are a set of scalar coecients. The eect of the rotor R on a+ generates
R(a+) = R(a + ^a)R~
= RaR~ + (RaR~ )
K
= eE=2ae;E=2 + (eE=2ae;E=2)
K (3.112)
and so accounts for all rotations of the vector a in <n. To complete the set of orthogonal
transformations, a representation for reections must be found as well. A reection in
the hyperplane orthogonal to the vector n in <n is represented by the element nn^ in <nn .
Since nn^ n^ n = ;1, nn^ is not a rotor and belongs to the disconnected part of spin(n n).
That nn^ commutes with K , and so is contained in spin(n n), is veri
ed as follows,
nn^ K = 2nn^
K + nK n^
= 2n2 + 2n
K n^ + Knn^
= 2(n2 + n^ 2) + Knn^
= Knn^ : (3.113)
53
The action of nn^ on a vector is determined by (3.8), and gives
nn^ a+ nn^ = ;nn^ an^n ; (nn^ an^n)
K
= ;nan ; (nan)
K (3.114)
as required.
Finally, we need to see how positive dilations are given a rotor description. A dilation
in the n direction by an amount e is generated by the rotor
R = e;nn^=2 (3:115)
where the generator ;nn^ =2 is built from the Ki in Table 3.4. Acting on the null vector
n+ = n + n^ , the rotor (3.115) gives
Rn+ R~ = e;nn^=2n+ e+nn^=2
= e;nn^ (n + n^)
= (cosh  ; nn^ sinh )(n + n^ )
= (cosh  + sinh )(n + n^ )
= en+: (3.116)
In addition, for vectors perpendicular to n in <n, the action of R on their null vector
equivalents has no eect. These are precisely the required properties for a dilation in
the n direction. This concludes the proof that the general linear group is represented by
the subgroup of spin(n n) consisting of elements commuting with K . As an aside, this
construction has led us to the Eij and Ki generators in Table (3.4). Commutators of the
Eij and Ki give the remaining Fij generators, which are sucient to close the algebra.
The determinant of a linear function on <n is easily represented in <nn since
f (e1) ^ f (e2) ^ : : :^ f (en) = det f En (3:117)
becomes
MWn M ;1 = det fWn  (3:118)
in the null space of V n. Here M is the spin group element representing the linear function
f . From the de
nitions of Wn and Wn (3.102), we can write
det f = 2n hW~ n MWn M ;1i (3:119)
from which many of the standard properties of determinants can be derived.

3.3.1 Endomorphisms of <n

We now turn to a second feature of <nn , which is its eectiveness in discussing endomor-
phisms of <n . These are maps of <n onto itself, and the set of all such maps is denoted
end(<n). Since the algebra <n is 2n -dimensional, the endomorphism algebra is isomorphic
to the algebra of real 2n  2n matrices,
end(<n) 
= R(2n): (3:120)
54
But the Cliord algebra <nn is also isomorphic to the algebra of 2n  2n matrices, so
every endomorphism of <n can be represented by a multivector in <nn 1. Our
rst task is
therefore to
nd how to construct each multivector equivalent of a given endomorphism.
Within <n, endomorphisms are built up from the the primitive operations of the inner
and outer products with the feig. It is more useful, however, to adopt the following basis
set of functions,
ei(A)  ei
A + ei ^ A = eiA (3.121)
^ei(A)  ;ei
A + ei ^ A = Ae ^ i (3.122)
where the hat (parity) operation in <n is de
ned by
A^r  (;1)r Ar (3:123)
and serves to distinguish even-grade and odd-grade multivectors. The reason for the use
of the hat in both <n and <nn will become apparent shortly. The feig and f^eig operations
are precisely those found in Section 2.3 in the context of Berezin calculus, though with
the
ducial tensor h now set to the identity. They satisfy the relations
eiej + ej ei = 2
ij
^ei^ej + ^ej ^ei = ;2
ij (3.124)
ei^ej + ^ej ei = 0 (3.125)
which are the de
ning relations for a vector basis in <nn . This establishes the isomorphism
between elements of end(<n ) and multivectors in <nn . Any element of end(<n) can be
decomposed into sums and products of the feig and f^eig functions, and so immediately
speci
es a multivector in <nn built from the same combinations of the feig and fe^ig basis
vectors.
To complete the construction, we must
nd a 2n -dimensional subspace of <nn on
which endomorphisms of <n are faithfully represented by (left) multiplication by elements
of <nn . The required subspace is a minimal left ideal of <nn and is denoted I n. It is
constructed as follows. We de
ne a set of bivector blades by
Ki  eie^i: (3:126)
Here, and in the remainder of this section, we have dropped the summation convention.
The Ki satisfy
Ki
Kj =
ij (3.127)
Ki  Kj = 0 (3.128)
and the bivector K is can be written as
X
K = Ki : (3:129)
i
1 I am grateful to Frank Sommen and Nadine Van Acker for pointing out the potential usefulness of
this result.

55
A family of commuting idempotents are now de
ned by
Ii  21 (1 + Ki) = wi wi (3:130)
and have the following properties:
Ii2 = Ii (3.131)
IiIj = Ij Ii (3.132)
eiIi = wi = e^iIi (3.133)
Iiei = wi = ;Iie^i (3.134)
KiIi = Ii: (3.135)
From the Ii the idempotent I is de
ned by
Yn
I  Ii = I1I2 : : :In = w1 w1w2 w2 : : : wnwn = Wn W~ n: (3:136)
i=1
I has the following properties:
I2 = I (3.137)
eiI = e^iI (3.138)
and
EnI = E^nI = Wn I = Wn (3:139)
where En is the pseudoscalar for the Euclidean algebra <n and E^n is the pseudoscalar
for the anti-Euclidean algebra <0n. The relationships in (3.139) establish an equivalence
between the <n, <0n and V n vector spaces.
Whilst the construction of I has made use of an orthonormal frame, the form of I is
actually independent of this choice. This can be seen by writing I in the form
1  K ^ K K ^ K ^ : : : ^ K 
I = 2n 1 + K + 2! + : : : + n! (3:140)
and recalling that K is frame-independent. It is interesting to note that the bracketed term
in (3.140) is of the same form as the Grassmann exponential considered in Section 2.2.1.
The full 2n -dimensional space I n is generated by left multiplication of I by the entire
algebra <nn ,
I n = <nnI: (3:141)
Since multiplication of I by ei and e^i are equivalent, every occurrence of an e^i in a multi-
vector in <nn can be replaced by an ei, so that there is a simple 1 $ 1 equivalence between
elements of <n and I n. The action of an element of end(<n) can now be represented in
<nn by left multiplication of I n by the appropriate multivector. For a multivector Ar in
<n the equivalence between the basic operators (3.122) is seen from
eiArI $ eiAr (3:142)

56
and
e^iArI $ A^r ei: (3:143)
The parity operation on the right-hand side of (3.143) arises because the e^i vector must
be anticommuted through each of the vectors making up the Ar multivector. This is
the reason for the dierent uses of the overhat notation for the <n and <nn algebras.
Symbolically, we can now write
eiI n $ ei<n (3.144)
e^iI $ <^ nei:
n (3.145)
Also, from the de
nitions of wi and wi (3.99), we
nd the equivalences
wiI n $ ei ^<n (3.146)
wi I $ ei
<n
n (3.147)
which establishes contact with the formalism of Grassmann/Berezin calculus given in
Chapter 2. We can now move easily between the formalism with dot and wedge products
used in Chapter 2 and the null-vector formalism adopted here. The chosen application
should dictate which is the more useful.
We next consider the quantity nn^ , where n is a unit vector. The action of this on I n
gives
nn^ I n $ n<^ nn: (3:148)
The operation on the right-hand side is the outermorphism action of a reection in the
hyperplane perpendicular to n. In the previous section we used a double-sided application
of nn^ on null vectors to represent reections in <n . We now see that the same object can
be applied single-sidedly in conjunction with the idempotent I to also produce reections.
The same is true of products of reections. For example, the rotor (3.110) gives
e(E;E^)=2MI = eE=2Me;E= ^ 2
I $ eE=2Me;E=2 (3:149)
demonstrating how the two-bladed structure of the Eij generators is used to represent
concurrent left and right multiplication in <n .
The operation <n 7! <^ n is performed by successive reections in each of the ei direc-
tions. We therefore
nd the equivalence
e1e^1e2e^2 : : : ene^nI n $ <^ n : (3:150)
But
e1e^1e2e^2 : : : ene^n = en : : : e2e1e^1e^2 : : : e^n = E~nE^n = Enn (3:151)
is the unit pseudoscalar in <nn , so multiplication of an element of I n by Enn corresponds
to the parity operation in <n. As a check, (Enn )2 is always +1, so the result of two parity
operations is always the identity.
The correspondence between the single-sided and double-sided forms for a dilation are
not quite so simple. If we consider the rotor expf;nn^ =2g again, we
nd that, for the
vector n,
e;nn^=2nI = e=2nI $ e=2n (3:152)
57
For vectors perpendicular to n, however, we
nd that
e;nn^=2n? I = n?e;=2nn^ I $ e;=2n? (3:153)
so the single-sided formulation gives a stretch along the n direction of expfg, but now
combined with an overall dilation of expf;=2g. This overall factor can be removed by
an additional boost with the exponential of a suitable multiple of K . It is clear, however,
that both single-sided and double-sided application of elements of the spin group which
commute with K can be used to give representations of the general linear group.
Finally, we consider even products of the null vectors wi and wi . These generate the
operations
wiwi I n $ ei
(ei ^<n)
wi wiI n $ ei ^ (ei
<n) (3.154)
which are rejection and projection operations in <n respectively. For a vector a in <n ,
the operation of projecting a onto the ei direction is performed by
Pi (a) = eiei
a (3:155)
and for a general multivector,
Pi(A) = ei ^ (ei
A): (3:156)
This projects out the components of A which contain a vector in the ei direction. The
projection onto the orthogonal complement of ei (the rejection) is given by
Pi? (A) = ei
(ei ^ A): (3:157)
Projection operations correspond to singular transformations, and we now see that these
are represented by products of null multivectors in <nn . This is sucient to ensure that
singular transformations can also be represented by an even product of vectors, some of
which may now be null.
Two results follow from these considerations. Firstly, every matrix Lie group can be
represented by a spin group | every matrix Lie group can be de
ned as a subgroup
of GL(n,R) and we have shown how GL(n,R) can be represented as a spin group. It
follows that every Lie algebra can be represented by a bivector algebra, since all Lie
algebras have a matrix representation via the adjoint representation. The discussion of
the unitary group has shown, however, that subgroups of GL(n,R) are not, in general,
the best way to construct spin-group representations. Other, more useful, constructions
are given in the following Sections. Secondly, every linear transformation on <n can be
represented in <nn as an even product of vectors, the result of which commutes with
K . It is well known that quaternions are better suited to rotations in three dimensions
than 3  3 matrices. It should now be possible to extend these advantages to arbitrary
linear functions. A number of other applications for these results can be envisaged. For
example, consider the equation
u0(s) = M (s)u(s) (3:158)

58
where u(s) and M (s) are vector and matrix functions of the parameter s and the prime
denotes the derivative with respect to s. By replacing the vector u by the null vector u
in <nn, equation (3.158) can be written in the form
u0 = B (s)
u (3:159)
where B (s) is a bivector. If we now write u = Ru0R~ , where u0 is a constant vector, then
equation (3.158) reduces to the rotor equation
R0 = 21 BR (3:160)
which may well be easier to analyse (a similar rotor reformulation of the Lorentz force
law is discussed in 20]).

3.4 The Remaining Classical Groups

We now turn attention to some of the remaining matrix Lie groups. Again, all groups
are realised as subgroups of the orthogonal group and so inherit a spin-group represen-
tation. The various multivectors and linear functions which remain invariant under the
group action are discussed, and simple methods are given for writing down the Bivector
generators which form the Lie algebra. The results from this chapter are summarised in
Section 3.5.

3.4.1 Complexi
cation | so(n,C)
Complexi
cation of the Orthogonal groups O(p q) leads to a single, non-compact, Lie
group in which all reference to the underlying metric is lost. With the uk and vk de
ned
as in Equation (3.55), the invariant bilinear form is
(u v) = uk vk = xk rk ; yk sk + j (xk sk + yk rk ): (3:161)
This is symmetric, and the real part contains equal numbers of positive and negative norm
terms. The Lie group O(n,C) will therefore be realised in the \balanced" algebra <nn . To
construct the imaginary part of (3.161), however, we need to
nd a symmetric function
which squares to give minus the identity. This is in contrast to the K function, which is
antisymmetric, and squares to +1. The solution is to introduce the \star" function
a  (;1)n+1 EnaEn;1 (3:162)
so that
ei = ei (3:163)
e^ = ;e^ :
i i
The use of the  notation is consistent with the de
nitions of fwig and fwi g bases (3.99).
The star operator is used to de
ne projections into the Euclidean and anti-Euclidean
subspaces of <nn :
E n (a) = 21 (a + a ) = a
EnEn;1 (3:164)
E^ (a) = 1 (a ; a ) = a ^ E E ;1:
n 2 n n

59
 Eij = eiej ; e^ie^j (i < j = 1 : : : n)
Fij = eie^j + e^iej : 00

Table 3.6: Bivector Basis for so(n,C)

The Euclidean pseudoscalar En anticommutes with K , so the star operator anticommutes
with the K function. It follows that the combined function
K (a)  K (a ) (3:165)
satis
es
K 2(a) = K K (a ) ]
= ;K K (a )]
= ;a (3.166)
and
K (a) = ;K (a)]
= K (a) (3.167)
and so has the required properties. The complex symmetric norm can now be written on
<nn as
(a b) = a
b + ja
K (b) (3:168)
which can veri
ed by expanding in the fei e^ig basis of (3.92).
An orthogonal transformation f will leave (a b) invariant provided that
K f (a) = f K (a) (3:169)
which de
nes the group O(n,C). Each function f in O(n,C) can be constructed from the
corresponding elements of spin(n n), which de
nes the spin-group representation. The
bivector generators must satisfy
K eB=2ae;B=2] = eB=2K (a)e;B=2 (3:170)
which reduces to the requirement
K (B
a) = B
K (a) (3.171)
) B
a = ;K B
K (a)] = ;K (B )
a (3.172)
) K (B ) = ;B: (3.173)
Since K 2(B ) = B for all bivectors B , the generators which form the Lie algebra so(n,C)
are all of the form B ; K (B ). This is used to write down the bivector basis in Table 3.6.
Under the commutator product, the Eij form a closed sub-algebra which is isomorphic
to so(n). The Fij ful
l the role of \jEij ". The Killing metric has n(n ; 1)=2 entries of
positive and signature and the same number of negative signature.
60
3.4.2 Quaternionic Structures | sp(n) and so (2n)
The quaternionic unitary group (usually denoted Sp(n) or HU(n)) is the invariance group
of the Hermitian-symmetric inner product of quaternion-valued vectors. By analogy with
the unitary group, the quaternionic structure is introduced by now quadrupling the real
space <n or <pq to <4n or <4p4q. We deal with the Euclidean case
rst and take feig to
be an orthonormal basis set for <n . Three further copies of <n are introduced, so that
fei e1i  e2i  e3i g form an orthonormal basis for <4n. Three \doubling" bivectors are now
de
ned as
J1 = eie1i + e2i e3i
J2 = eie2i + e3i e1i (3:174)
J3 = eiei + ei ei 
3 1 2

which de
ne the three functions
J i(a) = a
Ji: (3:175)
(The introduction of an orthonormal frame is not essential since each of the Ji are inde-
pendent of the intial choice of frame. Orthonormal frames do ease the discussion of the
properties of the Ji, however, so will be used frequently in this and the following sections).
The combined eect of J 1 and J 2 on a vector a produces
J 1J 2(a) = J 1(a
eie2i ; a
e2i ei + a
e3i e1i ; a
e1i e3i )
= a
eie3i ; a
e2i e1i ; a
e3i ei + a
e1i e2i
= J 3(a): (3.176)
The J i functions therefore generate the quaternionic structure
J 21 = J 22 = J 23 = J 1J 2J 3 = ;1: (3:177)
The Hermitian-symmetric quaternion inner product can be realised in <4n by
(a b) = a
b + a
J 1(b)i + a
J 2(b)j + a
J 3(b)j  (3:178)
where fi j  kg are a basis set of quaterions (see Section 1.2.3). The inner product (3.178)
contains four separate terms, each of which must be preserved by the invariance group.
This group therefore consists of orthogonal transformations satisfying
f (Ji) = Ji i = 1:::3 (3:179)
and the spin group representation consists of the elements of spin(4n) which commute
with all of the Ji. The bivector generators of the invariance group therefore also commute
with the Ji. The results established in Section 3.2 apply for each of the Ji in turn, so an
arbitrary bivector in the Lie algebra of Sp(n) must be of the form
BHU = B + J 1(B ) + J 2(B ) + J 3(B ): (3:180)
This result is used to write down the orthogonal basis set in Table 3.7. The algebra has
dimension 2n2 + n and rank n.

61
 Eij = eiej + e1i e1j + e2i e2j + e3i e3j (i < j = 1 : : : n)
Fij = eie1j ; e1i ej ; e2i e3j + e3i e2j 00

Gij = eie2j ; e2i ej ; e3i e1j + e1i e3j 00

Hij = eie3j ; e3i ej ; e1i e2j + e2i e1j 00

Fi = eie1i ; e2i e3i (i = 1 : : : n )

Gi = eie2i ; e3i e1i 00

Hi = eie3i ; e1i e2i 00

Table 3.7: Bivector Basis for sp(n)

The above extends easily to the case of sp(p q) by working in the algebra <4p4q. With
feig now a basis for <pq, the doubling bivectors are de
ned by
J1 = eie1i + e2i e3i etc (3:181)
and the quaternion relations (3.177) are still satis
ed. The Lie algebra is then generated
in exactly the same may. The resultant algebra has a Killing metric with 2(p2 + q2)+ p + q
negative entries and 4pq positive entries.
The properties of the K function found in Section 3.4.1 suggests that an alternative
quaternionic structure could be found in <2n2n by introducing anticommuting K and J
functions. This is indeed the case. With feig and ffig a pair of anticommuting orthonor-
mal bases for <n , a basis for <2n2n is de
ned by fei fi e^i f^ig. The hat operation is now
de
ned by
^a = K (a) = a
K (3:182)
with
K = eie^i + fif^i: (3:183)
A complexi
cation bivector is de
ned by
J = eif i + e^if^i = eifi ; e^if^i (3:184)
and additional doubling bivectors are de
ned by
K1 = eie^i ; fif^i (3:185)
K1 = eif^i + fie^i:
The set fJ K1 K2g form a set of three bivectors, no two of which commute.
With pseudoscalars En and Fn de
ned by
En = e1e2 : : :en (3:186)
Fn = f1f2 : : : fn
the star operation is de
ned by
a = ;En FnaF~nE~n : (3:187)
62
 Eij = eiej + fifj ; e^ie^j ; f^if^j (i < j = 1 : : : n)
Fij = eifj ; fiej + e^if^j ; f^ie^j 00

Gij = eie^j + e^iej + fif^j + f^ifj 00

Hij = eif^j + f^iej ; fi e^j ; e^ifj 00

Hi = eifi + e^if^i (i = 1 : : : n)

Table 3.8: Bivector Basis for so (n)

K i operations are now de
ned by
K i (a) = K i(a ) = a
Ki: (3:188)
These satisfy
K i 2(a) = ;a (3:189)
and
K 1K 2(a) = K 1K 2(a ) ]
= ;K 1K 2(a)
= J (a): (3.190)
The J and K i therefore form a quaternionic set of linear functions satisfying
K 12 = K 22 = J 2 = K 1K 2J = ;1: (3:191)
Orthogonal functions commuting with each of the J and K i functions will therefore leave
a quaternionic inner product invariant. This inner product can be written as
(a b) = a
J (b) + ia
b + j a
K 1(b) + ka
K 2(b) (3:192)
which expansion in the fei fi e^i f^ig frame shows to be equivalent to the skew-Hermitian
quaternionic inner product
(u v) = uyk ivk : (3:193)
The invariance group of (3.193) is denoted SO (2n) (or Sk(n,H)). The bivector generators
of the invariance group must satisfy J (B ) = B and K i (B ) = ;B and so are of the form
BH = B + J (B ) ; K 1(B ) ; K 2(B ):

(3:194)
This leads to the orthogonal set of basis generators in Table 3.8.
The bivector algebra so (n) has dimension n(2n ; 1) and a Killing metric with n2
negative entries and n2 ; n positive entries. This algebra is one of the possible real
forms of the complexi
ed algebra Dn. Some of the properties of so (2n), including its
representation theory, have been discussed by Barut & Bracken 58].

63
 Eij = eiej + fifj ; e^ie^j ; f^if^j (i < j = 1 : : : n)
Fij = eifj ; fiej ; e^if^j + f^ie^j 00

Gij = eie^j ; e^iej + fif^j ; f^ifj 00

Hij = eif^j + f^iej ; fi e^j ; e^ifj 00

Ji = eifi ; e^if^i (i = 1 : : : n)
Ki = eie^i + fif^i 00

Table 3.9: Bivector Basis for gl(n,C)

3.4.3 The Complex and Quaternionic General Linear Groups
The general linear group over the complex
eld, GL(n,C), is constructed from linear
functions in the 2n-dimensional space <2n which leave the complex structure intact,
h(a)
J = h(a
J ): (3:195)
These linear functions can be represented by orthogonal functions in <2n2n using the
techniques introduced in Section 3.3. Thus, using the conventions of Section 3.4.2, a
vector a in <2n is represented in <2n2n by the null vector a+ = a + ^a, and the complex
structure is de
ned by the bivector J of equation (3.184). These de
nitions ensure that
the J function keeps null vectors in the same null space,
K J (a) = J K (a) (3:196)
) (a
J )
K ; (a
K )
J = a
(J  K ) = 0 (3:197)
which is satis
ed since J  K = 0. The spin group representation of GL(n,C) consists of
all elements of spin(2n 2n) which commute with both J and K and hence preserve both
the null and complex structures. The bivector generators of the Lie algebra gl(n,C) are
therefore of the form
BC = B + J (B ) ; K (B ) ; K J (B ) (3:198)
which yields the set of generators in Table 3.9. This algebra has 2n2 generators, as is to
be expected. The two abelian subgroups are removed in the usual manner to yield the
Lie algebra for sl(n,C) given in Table 3.10. The Killing metric gives n2 ; 1 terms of both
positive and negative norm.
The general linear group with quaternionic entries (denoted U (2n) or GL(n,H)) is
constructed in the same manner as the above, except that now the group is contained
in the algebra <4n4n. Thus we start in the algebra <4n and introduce a quaternionic
structure through the Ji bivectors of equations (3.174). The <4n algebra is then doubled
to a <4n4n algebra with the introduction of a suitable K bivector, and the Ji are extended
to new bivectors
Ji0 = Ji ; J^i: (3:199)
The spin-group representation of U (2n) then consists of elements of spin(4n 4n) which
commute with all of the Ji0 and with K . The bivectors generators are all of the form
BH = B + J 01(B ) + J 02(B ) + J 03(B ) ; K B + J 01(B ) + J 02(B ) + J 03(B )]: (3:200)
64
 Eij = eiej + fifj ; e^ie^j ; f^if^j (i < j = 1 : : : n)
Fij = eifj ; fiej ; e^if^j + f^ie^j 00

Gij = eie^j ; e^iej + fif^j ; f^ifj 00

Hij = eif^j + f^iej ; fi e^j ; e^ifj 00

Gi = Ji ; Ji+1 (i = 1 : : : n ; 1)
Hi = Ki ; Ki+1 00

Table 3.10: Bivector Basis for sl(n,C)

The result is a (4n2)-dimensional algebra containing the single abelian factor K . This is
factored out in the usual way to yield the bivector Lie algebra su (2n).

3.4.4 The symplectic Groups Sp(n,R) and Sp(n,C)

The symplectic group Sp(n,R) consists of all linear functions h acting on <2n satisfying
h(J ) = J , where J is the doubling bivector from the <n algebra to the <2n algebra. A
spin-group representation is achieved by doubling to <2n2n and constructing Sp(n,R) as
a subgroup of GL(2n,R). In <2n, the symplectic inner product is given by (a ^ b)
J . In
<2n2n, with K de
ned as in Equation (3.183), the vectors a and b are replaced by the
null vectors a+ and b+. Their symplectic inner product is given by
(a+ ^ b+)
JS = K (a+) ^ K (b+)]
JS = (a+ ^ b+)
K (JS ): (3:201)
The symplectic bivector in <2n2n satis
es
K (JS ) = JS (3:202)
and so is de
ned by
JS = J + J^ = eifi + e^if^i: (3:203)
(This diers from the J de
ned in equation (3.184), so generates an alternative complex
structure). The group Sp(n,R) is the subgroup of orthogonal transformations on <2n2n
which leave both JS and K invariant. The spin-group representation consists of all ele-
ments which commute with both JS and K . The bivector generators of Sp(n,R) are all
of the form
BSp = B + J S (B ) ; K (B ) ; K J S (B ): (3:204)
An orthogonal basis for the algebra sp(n,R) is contained in Table 3.11. This has dimension
n(2n + 1) and a Killing metric with n2 negative entries and n2 + n positive entries. The
same construction can be used to obtain the algebras for sp(p q,R) by starting from <pq
and doubling this to <2p2q.
The group Sp(n,C) consists of functions on <4n satisfying h(J1) = J1 and which also
preserve the complex structure,
h(a
J3) = h(a)
J3: (3:205)

65
 Eij = eiej + fifj ; e^ie^j ; f^if^j (i < j = 1 : : : n)
Fij = eifj ; fiej ; e^if^j + f^ie^j 00

Gij = eie^j ; e^iej ; fif^j + f^ifj 00

Hij = eif^j ; f^iej + fi e^j ; e^ifj 00

Fi = eifi ; e^if^i (i = 1 : : : n)
Gi = eie^i ; fif^i 00

Hi = eif^i + fie^i 00

Table 3.11: Bivector Basis for sp(n,R)

The complex and symplectic structures satisfy J 3(J1) = ;J1, so J3 and J1 do not com-
mute. Instead they are two-thirds of the quaternionic set of bivectors introduced in
Section 3.4.2. The C-skew inner product on <4n is written
(a b) = a
J 1(b) ; ja
J 1J 3(b) = a
J 1(b) + ja
J 2(b): (3:206)
By analogy with Sp(n,R), a spin-group representation of Sp(n,C) is constructed as a
subgroup of GL(2n,C)in <4n4n. With the null structure de
ned by K , the symplectic
structure is now determined by
JS = J1 + K (J1) (3:207)
and the complex structure by
J = J2 ; K (J ): (3:208)
The Lie algebra sp(n,C) is formed from the set of bivectors in <24n4n which commute with
all of the K , J and JS bivectors. With this information, it is a simple matter to write
down a basis set of generators.

3.5 Summary
In the preceding sections we have seen how many matrix Lie groups can be represented
as spin groups, and how all (
nite dimensional) Lie algebras can be realised as bivector
algebras. These results are summarised in Tables 3.12 and 3.13. Table 3.12 lists the
classical bilinear forms, their invariance groups, the base space in which the spin group
representation is constructed and the general form of the bivector generators. The re-
maining general linear groups are listed in Table 3.13. Again, their invariant bivectors
and the general form of the generators are listed. For both tables, the conventions for
the various functions and bivectors used are those of the section where the group was
discussed.
A number of extensions to this work can be considered. It is well known, for example,
that the Lie group G2 can be constructed in <07 as the invariance group of a particular
trivector (which is given in 46]). This suggests that the techniques explored in this chapter
can be applied to the exceptional groups. A geometric algebra is a graded space and in
Chapter 5 we will see how this can be used to de
ne a multivector bracket which satis
es
66
 Base Form of Bivector
Type Form of (a b) Group Space Generators
R-symmetric a
b SO(p q) <pq B
R-skew a
J (b) Sp(n,R) <2n2n B + J S (B ) ; K (B + J S (B ))
C-symmetric a
b + ja
K (b) SO(n,C) <nn B ; K (B )
C-skew a
J 1(b) + ja
J 2(b) Sp(n,C) < 4n4 n B + J (B ) + J S (B ) + J J S (B )
;K ( 00 )
C-Hermitian a
b + ja
J (b) U(p q) <2p2q B + J (B )
H-Hermitian a
b + a
J 1(b)i + Sp(n) <4n B + J 1(B ) + J 2(B ) + J 3(B )
a
J 2(b)j + a
J 3(b)j
H-Skew a
J (b) + a
K 1(b)+ SO (2n) <2n2n B + J (B ) ; K 1(B ) ; K 2(B )
a
bi + a
K 2(b)k

Table 3.12: The Classical Bilinear Forms and their Invariance Groups
the super-Jacobi identities. This opens up the possibility of further extending the work of
this chapter to include super-Lie algebras. Furthermore, we shall see in Chapter 4 that the
techniques developed for doubling spaces are ideally suited to the study of multiparticle
quantum theory. Whether some practical bene
ts await the idea that all general linear
transformations can be represented as even products of vectors remains to be seen.

Base Form of Bivector

Group Space Invariants Generators
GL(n,R) <nn K B ; K (B )
GL(n,C) < 2 n2 n K J B + J (B ) ; K (B + J (B ))
GL(n,H) / SU (n) <4n4n K J10  J20  J30 B + J 01(B ) + J 02(B ) + J 03(B )
;K ( 00 )

Table 3.13: The General Linear Groups

67
Chapter 4
Spinor Algebra
This chapter describes a translation between conventional matrix-based spinor algebra
in three and four dimensions 59, 60], and an approach based entirely in the (real) geo-
metric algebra of spacetime. The geometric algebra of Minkowski spacetime is called the
spacetime algebra or, more simply, the STA. The STA was introduced in Section 1.2.5 as
the geometric algebra generated by a set of four orthonormal vectors f g,  = 0 : : : 3,
satisfying


= 
= diag(+ ; ; ;): (4:1)
Whilst the f g satisfy the Dirac algebra generating relations, they are to be thought
of as an orthonormal frame of independent vectors and not as components of a single
\isospace" vector. The full STA is spanned by the basis
1 f g fk  ik g fi g i (4:2)
where
i  0123 (4:3)
and
k  k 0: (4:4)
The meaning of these equation was discussed in Section 1.2.5.
The aim of this chapter is to express both spinors and matrix operators within the
real STA. This results in a very powerful language in which all algebraic manipulations
can be performed without ever introducing a matrix representation. The Pauli matrix
algebra is studied
rst, and an extension to multiparticle systems is introduced. The Dirac
algebra and Dirac spinors are then considered. The translation into the STA quickly yields
the Dirac equation in the form
rst found by Hestenes 17, 19, 21, 27]. The concept of
the multiparticle STA is introduced, and is used to formulate a number of two-particle
relativistic wave equations. Some problems with these are discussed and a new equation,
which has no spinorial counterpart, is proposed. The chapter concludes with a discussion
of the 2-spinor calculus of Penrose & Rindler 36]. Again, it is shown how a scalar unit
imaginary is eliminated by the use of the real multiparticle STA. Some sections of this
chapter appeared in the papers \States and operators in the spacetime algebra" 6] and
\2-Spinors, twistors and supersymmetry in the spacetime algebra 4].

68
4.1 Pauli Spinors
This section establishes a framework for the study of the Pauli operator algebra and Pauli
spinors within the geometric algebra of 3-dimensional space. The geometric algebra of
space was introduced in Section 1.2.3 and is spanned by
1 fk g fik g i: (4:5)
Here the fk g are a set of three relative vectors (spacetime bivectors) in the 0-system.
Vectors in this system are written in bold type to distinguish them from spacetime vectors.
There is no possible confusion with the fk g symbols, so these are left in normal type.
When working non-relativistically within the even subalgebra of the full STA some nota-
tional modi
cations are necessary. Relative vectors fk g and relative bivectors fikg are
both bivectors in the full STA, so spatial reversion and spacetime reversion have dierent
eects. To distinguish these, we de
ne the operation
~ 0
Ay = 0A (4:6)
which de
nes reversion in the Pauli algebra. The presence of the 0 vector in the de
nition
of Pauli reversion shows that this operation is dependent on the choice of spacetime frame.
The dot and wedge symbols also carry dierent meanings dependent on whether their
arguments are treated as spatial vectors or spacetime bivectors. The convention adopted
here is that the meaning is determined by whether their arguments are written in bold
type or not. Bold-type objects are treated as three-dimensional multivectors, whereas
normal-type objects are treated as belonging to the full STA. This is the one potentially
confusing aspect of our conventions, though in practice the meaning of all the symbols
used is quite unambiguous.
The Pauli operator algebra 59] is generated by the 2  2 matrices
 !  !  !
0 1 0 ; j
^1 = 1 0  ^2 = j 0  ^3 = 0 ;1 : 1 0 (4:7)
These operators act on 2-component complex spinors
 !
ji = 12  (4:8)
where 1 and 2 are complex numbers. We have adopted a convention by which standard
quantum operators appear with carets, and quantum states are written as kets and bras.
We continue to write the unit scalar imaginary of conventional quantum mechanics as j ,
which distinguishes it from the geometric pseudoscalar i.
To realise the Pauli operator algebra within the algebra of space, the column Pauli
spinor ji is placed in one-to-one correspondence with the even multivector  (which
satis
es  = 00) through the identi
cation1
 0 3 !
a + ja
ji = ;a2 + ja1 $  = a0 + ak ik : (4:9)
1 This mapping was
rst found by Anthony Lasenby.

69
In particular, the basis spin-up and spin-down states become
 !
1 $ 1 (4:10)
0
and  !
0 $ ;i2: (4:11)
1
The action of the four quantum operators f^k  j g can now be replaced by the operations
^k ji $ k 3 (k = 1 2 3) (4:12)
and
j ji $ i3: (4:13)
Verifying these relations is a matter of routine computation, for example
 2 1 !

; a +
^1 ji = a0 + ja3 ja $ ;;aa0i+2a+ia33i1 = 1 a0 + akik 3:
2 1
(4:14)
With these de
nitions, the action of complex conjugation of a Pauli spinor translates to
ji $ 22: (4:15)
The presence of a
xed spatial vector on the left-hand side of  shows that complex
conjugation is a frame-dependent concept.
As an illustration, the Pauli equation (in natural units),


j@t ji = 21m (;j r ; eA)2 ; e^k B k ji + eV ji  (4:16)
can be written (in the Coulomb gauge) as 22]
@ti3 = 21m (;r2 + 2eA
ri3 + e2A2) ; 2em B 3 + eV  (4:17)
where B is the magnetic
eld vector B k k . This translation achieves two important
goals. The scalar unit imaginary is eliminated in favour of right-multiplication by i3,
and all terms (both operators and states) are now real-space multivectors. Removal of
the distinction between states and operators is an important conceptual simpli
cation.
We next need to
nd a geometric algebra equivalent of the spinor inner product h ji.
In order to see how to handle this, we need only consider its real part. This is given by
<h ji $ hyi (4:18)
so that, for example,
h ji $ hyi = h(a0 ; iaj j )(a0 + iakk )i
= (a0)2 + ak ak: (4.19)
70
Since
hji = <hji ; j <hjji (4:20)
the full inner product becomes
h ji $ ( )S  hyi ; hyi3ii3: (4:21)
The right hand side projects out the f1 i3g components from the geometric product
y. The result of this projection on a multivector A is written hAiS . For Pauli-even
multivectors this projection has the simple form
hAiS = 21 (A ; i3Ai3): (4:22)
As an example of (4.21), consider the expectation value
hj^k ji $ hyk 3i ; hyk iii3 = k
h3yi1 (4:23)
which gives the mean value of spin measurements in the k direction. The STA form
indicates that this is the component of the spin vector s = 3y in the k direction,
so that s is the coordinate-free form of this vector. Since 3y is both Pauli-odd and
Hermitian-symmetric (reverse-symmetric in the Pauli algebra), s contains only a vector
part. (In fact, both spin and angular momentum are better viewed as bivector quantities,
so it is usually more convenient to work with is instead of s.)
Under an active rotation, the spinor  transforms as
 7! 0 = R0 (4:24)
where R0 is a constant rotor. The quantity 0 is even, and so is a second spinor. (The
term \spinor" is used in this chapter to denote any member of a linear space which is
closed under left-multiplication by a rotor R0.) The corresponding transformation law for
s is
s 7! s0 = R0sRy0 (4:25)
which is the standard double-sided rotor description for a rotation, introduced in Sec-
tion 1.2.4.
The de
nitions (4.9), (4.12) and (4.13) have established a simple translation from the
language of Pauli operators and spinors into the geometric algebra of space. But the STA
formulation can be taken further to aord new insights into the role of spinors in the
Pauli theory. By de
ning
 = y (4:26)
the spinor  can be written
 = 1=2R (4:27)
where R is de
ned as
R = ;1=2: (4:28)
R satis
es
RRy = 1 (4:29)

71
and is therefore a spatial rotor. The spin vector can now be written
s = R3Ry (4:30)
which demonstrates that the double-sided construction of the expectation value (4.23)
contains an instruction to rotate and dilate the
xed 3 axis into the spin direction. The
original states of quantum mechanics have now become operators in the STA, acting on
vectors. The decomposition of the spinor  into a density term  and a rotor R suggests
that a deeper substructure underlies the Pauli theory. This is a subject which has been
frequently discussed by David Hestenes 19, 22, 23, 27]. As an example of the insights
aorded by this decomposition, it is now clear \why" spinors transform single-sidedly
under active rotations of
elds in space. If the vector s is to be rotated to a new vector
R0sRy0 then, according to the rotor group combination law, R must transform to R0R.
This produces the spinor transformation law (4.24).
We should now consider the status of the
xed fk g frame. The form of the Pauli
equation (4.17) illustrates the fact that, when forming covariant expressions, the fkg
only appear explicitly on the right-hand side of . In an expression like
Ak ^k ji $ A3 (4:31)
for example, the quantity A is a spatial vector and transforms as
A 7! A0 = R0ARy0: (4:32)
The entire quantity therefore transforms as
A3 7! R0ARy0R03 = R0A3 (4:33)
so that A3 is another spinor, as required. Throughout this derivation, the 3 sits on
the right-hand side of  and does not transform | it is part of a
xed frame in space.
A useful analogy is provided by rigid-body dynamics, in which a rotating frame fek g,
aligned with the principal axes of the body, can be related to a
xed laboratory frame
fk g by
ek = Rk Ry: (4:34)
The dynamics is now completely contained in the rotor R. The rotating frame fek g is
unaected by the choice of laboratory frame. A dierent
xed laboratory frame,
k0 = R1k Ry1 (4:35)
simply requires the new rotor
R0 = RRy1 (4:36)
to produce the same rotating frame. Under an active rotation, the rigid body is rotated
about its centre of mass, whilst the laboratory frame is
xed. Such a rotation takes
ek 7! e0k = R0ek Ry0 (4:37)
which is enforced by the rotor transformation R 7! R0R. The
xed frame is shielded from
this rotation, and so is unaected by the active transformation. This is precisely what
72
happens in the Pauli theory. The spinor  contains a rotor, which shields vectors on the
right-hand side of the spinor from active rotations of spatial vectors.
Since multiplication of a column spinor by j is performed in the STA by right-sided
multiplication by i3, a U(1) gauge transformation is performed by
 7! 0 = ei 3 : (4:38)
This right-sided multiplication by the rotor R = expfi3g is equivalent to a rotation
of the initial (
xed) frame to the new frame fRk Ryg. Gauge invariance can therefore
now be interpreted as the requirement that physics is unaected by the position of the 1
and 2 axes in the i3 plane. In terms of rigid-body dynamics, this means that the body
behaves as a symmetric top. These analogies between rigid-body dynamics and the STA
form of the Pauli theory are quite suggestive. We shall shortly see how these analogies
extend to the Dirac theory.

4.1.1 Pauli Operators

In our geometric algebra formalism, an arbitrary operator M^ ji is replaced by a linear
function M () acting on even multivectors in the algebra of space. The function M ()
is an example of the natural extension of linear algebra to encompass linear operators
acting on multivectors. The study of such functions is termed \multilinear function the-
ory" and some preliminary results in this
eld, including a new approach to the Petrov
classi
cation of the Riemann tensor, have been given by Hestenes & Sobczyk 24]. Since
 is a 4-component multivector, the space of functions M () is 16-dimensional, which
is the dimension of the group GL(4,R). This is twice as large as the 8-dimensional Pauli
operator algebra (which forms the group GL(2,C)). The subset of multilinear functions
which represent Pauli operators is de
ned by the requirement that M () respects the
complex structure,
j M^ (j ji) = ;M^ ji
) M (i3)i3 = ;M (): (4.39)
The set of M () satisfying (4.39) is 8-dimensional, as required.
The Hermitian operator adjoint is de
ned by
 E
h M
^ = hM^ y ji : (4:40)
In terms of the function M (), this separates into two equations
hyM ()i = hMHA
y
()i (4:41)
and
hyM ()i3i = hMHA
y
()i3i (4:42)
where the subscript on MHA labels the STA representation of the Pauli operator adjoint.
The imaginary equation (4.42) is automatically satis
ed by virtue of (4.41) and (4.39).

73
The adjoint of a multilinear function is de
ned in the same way as that of a linear function
(Section 1.3), so that
hM' ()i = hM ()i: (4:43)
The Pauli operator adjoint is therefore given by the combination of a reversion, the
geometric adjoint, and a second reversion,
MHA() = M' y(y): (4:44)
For example, if M () = AB , then
M' () = BA (4:45)
and
MHA() = (ByA)y
= AyB y (4.46)
Since the STA action of the ^k operators takes  into k 3, it follows that these operators
are, properly, Hermitian. Through this approach, the Pauli operator algebra can now be
fully integrated into the wider subject of multilinear function theory.

4.2 Multiparticle Pauli States

In quantum theory, 2-particle states are assembled from direct products of single-particle
states. For example, a basis for the outer-product space of two spin-1=2 states is given
by the set
 !  !  !  !  !  !  !  !
1  1  0  1  1  0  0  0 : (4:47)
0 0 1 0 0 1 1 1
To represent these states in the STA, we must consider forming copies of the STA itself.
We shall see shortly that, for relativistic states, multiparticle systems are constructed by
working in a 4n-dimensional con
guration space. Thus, to form two-particle relativis-
tic states, we work in the geometric algebra generated by the basis set f 1  2 g, where
the basis vectors from dierent particle spacetimes anticommute. (The superscripts la-
bel the particle space.) If we wanted to adopt the same procedure when working non-
relativistically, we would set up a space spanned by fi1 i2g, where the basis vectors from
dierent particle spaces also anticommute. This construction would indeed suce for an
entirely non-relativistic discussion. The view adopted throughout this thesis, however,
is that the algebra of space is derived from the more fundamental relativistic algebra of
spacetime. The construction of multiparticle Pauli states should therefore be consistent
with the construction of relativistic multiparticle states. It follows that the spatial vectors
from two separate copies of spacetime are given by
i1 = i101 (4.48)
i2 = i202 (4.49)
74
and so satisfy
i1j2 = i101j202 = i1j20201 = j202i101 = j2i1: (4:50)
In constructing multiparticle Pauli states, we must therefore take the basis vectors from
dierent particle spaces as commuting. In fact, for the non-relativistic discussion of this
section, it does not matter whether these vectors are taken as commuting or anticommut-
ing. It is only when we come to consider relativistic states, and in particular the 2-spinor
calculus, that the dierence becomes important.
Since multiparticle states are ultimately constructed in a subalgebra of the geometric
algebra of relativistic con
guration space, the elements used all inherit a well-de
ned
Cliord multiplication. There is therefore no need for the tensor product symbol ,
which is replaced by simply juxtaposing the elements. Superscripts are used to label
the single-particle algebra from which any particular element is derived. As a further
abbreviation i111 is written, wherever possible, as i11 etc . This helps to remove some of
the superscripts. The unit element of either space is written simply as 1.
The full 2-particle algebra generated by commuting basis vectors is 8  8 = 64 dimen-
sional. The spinor subalgebra is 4  4 = 16 dimensional, which is twice the dimension of
the direct product of two 2-component complex spinors. The dimensionality has doubled
because we have not yet taken the complex structure of the spinors into account. While
the role of j is played in the two single-particle spaces by right multiplication by i31 and
i32 respectively, standard quantum mechanics does not distinguish between these opera-
tions. A projection operator must therefore be included to ensure that right multiplication
by i31 or i32 reduces to the same operation. If a two-particle spin state is represented by
the multivector , then  must satisfy
i31 = i32 (4:51)
from which we
nd that
 = ;i31i32 (4.52)
)  =  12 (1 ; i31i32): (4.53)
On de
ning
E = 21 (1 ; i31i32) (4:54)
it is seen that
E2 = E (4:55)
so right multiplication by E is a projection operation. It follows that the two-particle
state  must contain a factor of E on its right-hand side. We can further de
ne
J = Ei31 = Ei32 = 12 (i31 + i32) (4:56)
so that
J 2 = ;E: (4:57)
Right-sided multiplication by J takes over the role of j for multiparticle states.

75
 The STA representation of a 2-particle Pauli spinor is now given by 12E , where 1
and 2 are spinors (even multivectors) in their own spaces. A complete basis for 2-particle
spin states is provided by
 !  !
1  1 $ E
0! 0!
0  1 $ ;i21E
1
 !  ! 0 (4:58)
1  0 $ ;i2E 2
0! 1!
0  0 $ i21i22E:
1 1
This procedure extends simply to higher multiplicities. All that is required is to
nd
the \quantum correlator" En satisfying
En i3j = Eni3k = Jn for all j , k: (4:59)
En can be constructed by picking out the j = 1 space, say, and correlating all the other
spaces to this, so that
Yn
En = 21 (1 ; i31i3j ): (4:60)
j =2
The form of En is independent of which of the n spaces is singled out and correlated to.
The complex structure is de
ned by
Jn = Eni3j  (4:61)
where i3j can be chosen from any of the n spaces. To illustrate this consider the case of
n = 3, where
E3 = 41 (1 ; i31i32)(1 ; i31i33) (4.62)
= 41 (1 ; i31i32 ; i31i33 ; i32i33) (4.63)
and
J3 = 14 (i31 + i32 + i33 ; i31i32i33): (4:64)
Both E3 and J3 are symmetric under permutations of their indices.
A signi
cant feature of this approach is that all the operations de
ned for the single-
particle STA extend naturally to the multiparticle algebra. The reversion operation, for
example, still has precisely the same de
nition | it simply reverses the order of vectors
in any given multivector. The spinor inner product (4.21) also generalises immediately,
to
( )S = hEn i;1hyEn iEn ; hyJn iJn]: (4:65)
The factor of hEn i;1 is included so that the operation
P (M ) = hEn i;1hMEn iEn ; hMJn iJn ] (4:66)
76
is a projection operation (i.e. P (M ) satis
es P 2(M ) = P (M )). The fact that P (M ) is a
projection operation follows from the results
P (En ) = hEn i;1hEn EniEn ; hEn Jn iJn]
= hEn i;1hEn iEn ; hEn i3j iJn ]
= En (4.67)
and
P (Jn) = hEn i;1 hJnEn iEn ; hJnJn iJn ]
= Jn : (4.68)

4.2.1 The Non-Relativistic Singlet State

As an application of the formalism outlined above, consider the 2-particle singlet state
ji, de
ned by ( !  !  !  !)
ji = p2 10  01 ; 01  10 :
1 (4:69)
This is represented in the two-particle STA by the multivector
 = p12 (i21 ; i22) 21 (1 ; i31i32): (4:70)
The properties of  are more easily seen by writing
p
 = 21 (1 + i21i22) 21 (1 + i31i32) 2i21 (4:71)
which shows how  contains the commuting idempotents 12 (1 + i21i22) and 12 (1 + i31i32).
The normalisation ensures that
( )S = 2hyiE2
= 4h 21 (1 + i21i22) 21 (1 + i31i32)iE2
= E2: (4.72)
The identi
cation of the idempotents in  leads immediately to the results that
p
i21 = 12 (i21 ; i22) 12 (1 + i31i32) 2i21 = ;i22 (4:73)
and
i31 = ;i32 (4:74)
and hence that
i11 = i31i21 = ;i22i31 = i22i32 = ;i12: (4:75)
If M 1 is an arbitrary even element in the Pauli algebra (M = M 0 + M k ik1), then it follows
that  satis
es
M 1 = M 2y: (4:76)
This provides a novel demonstration of the rotational invariance of . Under a joint
rotation in 2-particle space, a spinor  transforms to R1 R2, where R1 and R2 are copies
77
of the same rotor but acting in the two dierent spaces. The combined quantity R1R2 is a
rotor acting in 6-dimensional space, and its generator is of the form of the Eij generators
for SU(n) (Table 3.3). From equation (4.76) it follows that, under such a rotation, 
transforms as
 7! R1R2 = R1R1y =  (4:77)
so that  is a genuine 2-particle scalar.

4.2.2 Non-Relativistic Multiparticle Observables

Multiparticle observables are formed in the same way as for single-particle states. Some
combination of elements from the
xed fkj g frames is sandwiched between a multiparticle
wavefunction  and its spatial reverse y. An important example of this construction is
provided by the multiparticle spin current. The relevant operator is
Sk () = k131 + k232 +


+ kn3n
= ;ik1i31 + ik2i32 +


+ ikni3n] (4.78)
and the corresponding observable is
( Sk ())S = ;hEn i;1 hy(ik1i31 +


+ ikni3n)En iEn
+hEn i;1 hy(ik1i31 +


+ ikni3n)JniJn
= ;2n;1 h(ik1 +


+ ikn)JyiEn + h(ik1 +


+ ikn)yiJn ]
= ;2n;1 (ik1 +


+ ikn)  (Jy)En : (4.79)
The multiparticle spin current is therefore de
ned by the bivector
S = 2n;1 hJyi2 (4:80)
where the right-hand side projects out from the full multivector Jy the components
which are pure bivectors in each of the particle spaces. The result of projecting out from
the multivector M the components contained entirely in the ith-particle space will be
denoted hM ii , so we can write
S i = 2n;1 hJyii2: (4:81)
Under a joint rotation in n-particle space,  transforms to R1 : : :Rn  and S therefore
transforms to
R1 : : : Rn SRn y : : :R1 y = R1S 1R1y +


+ Rn SnRn y: (4:82)
Each of the single-particle spin currents is therefore rotated by the same amount in its
own space. That the de
nition (4.80) is sensible can be checked with the four basis
states (4.58). The form of S for each of these is contained in Table 4.1.
Other observables can be formed using dierent
xed multivectors. For example, a
two-particle invariant is generated by sandwiching a constant multivector ( between the
singlet state ,
M = (y: (4:83)
78
 Pauli Multivector Spin
State Form Current
j ""i E2 i31 + i32
j "#i ;i22E2 i31 ; i32
j #"i ;i21E2 ;i31 + i32
j ##i i21i22E2 ;i31 ; i32

Table 4.1: Spin Currents for 2-Particle Pauli States

Taking ( = 1 yields
M = y = 2 21 (1 + i21i22) 12 (1 + i31i32) = 12 (1 + i11i12 + i21i22 + i31i32) (4:84)
and ( = i1i2 gives
M = i1i2y = 12 (i1i2 + 1112 + 2122 + 3132): (4:85)
This shows that both ik1ik2 and k1k2 are invariants under two-particle rotations. In
standard quantum mechanics these invariants would be thought of as arising from the
\inner product" of the spin vectors ^i1 and ^i2. Here, we have seen that the invariants
arise in a completely dierent way by looking at the full multivector y. It is interesting to
note that the quantities ik1ik2 and k1k2 are similar in form to the symplectic (doubling)
bivector J introduced in Section 3.2.
The contents of this section should have demonstrated that the multiparticle STA
approach is capable of reproducing most (if not all) of standard multiparticle quantum
mechanics. One important result that follows is that the unit scalar imaginary j can be
completely eliminated from quantum mechanics and replaced by geometrically meaningful
quantities. This should have signi
cant implications for the interpretation of quantum
mechanics. The main motivation for this work comes, however, from the extension to
relativistic quantum mechanics. There we will part company with operator techniques
altogether, and the multiparticle STA will suggest an entirely new approach to relativistic
quantum theory.

4.3 Dirac Spinors

We now extend the procedures developed for Pauli spinors to show how Dirac spinors can
be understood in terms of the geometry of real spacetime. This reveals a geometrical role
for spinors in the Dirac theory (a role which was
rst identi
ed by Hestenes 19, 21]).
Furthermore, this formulation is representation-free, highlighting the intrinsic content of
the Dirac theory.
We begin with the  -matrices in the standard Dirac-Pauli representation 59],
 !  !
I
^0 = 0 ;I 0 and ^k = ^ 0 ; 
^ k
k 0 : (4:86)

79
A Dirac column spinor ji is placed in one-to-one correspondence with an 8-component
even element of the STA via 4, 61]
0 a0 + ja3 1
B 2 1 C
ji = BB@ ;;ab3 ++ ja
jb0 C
C $  = a0 + ak ik + i(b0 + bk ik ):
A (4:87)
;b1 ; jb2
With the spinor ji now replaced by an even multivector, the action of the operators
f^  ^5 j g (where ^5 = ^5 = ;j ^0^1^2^3) becomes
^ ji $  0 ( = 0 : : : 3)
j ji $  i3 (4:88)
^5 ji $ 3
which are veri
ed by simple computation! for example
0 ;b3 + jb0 1
B ;b1 ; jb2 C ;b3 + b03 + b1i2 ; b2i1 = 3:
^5 ji = B
B@ a0 + ja3 C C
A $ +a03 + a3i ; a21 + a12 (4:89)
;a2 + ja1
Complex conjugation in this representation becomes
ji $ ;22 (4:90)
which picks out a preferred direction on the left-hand side of  and so is not a Lorentz-
invariant operation.
As a simple application of (4.87) and (4.88), the Dirac equation
^ (j@ ; eA ) ji = m ji (4:91)
becomes, upon postmultiplying by 0,
ri3 ; eA = m0 (4:92)
which is the form
rst discovered by Hestenes 17]. Here r =  @ is the vector derivative
in spacetime. The properties of r will be discussed more fully in Chapter 6. This
translation is direct and unambiguous, leading to an equation which is not only coordinate-
free (since the vectors r =  @ and A =  A no longer refer to any frame) but is
also representation-free. In manipulating (4.92) one needs only the algebraic rules for
multiplying spacetime multivectors, and the equation can be solved completely without
ever having to introduce a matrix representation. Stripped of the dependence on a matrix
representation, equation (4.92) expresses the intrinsic geometric content of the Dirac
equation.
To discuss the spinor inner product, it is necessary to distinguish between the Hermi-
tian and Dirac adjoint. These are written as
h'j ; Dirac adjoint (4:93)
hj ; Hermitian adjoint
80
which translate as follows,
h'j $ ~ (4:94)
hj $ ~ 0:
y = 0
This makes it clear that the Dirac adjoint is the natural frame-invariant choice. The inner
product is handled in the same manner as in equation (4.21), so that
h' ji $ h ~ i ; hi
~ 3ii3 = h~ iS  (4:95)
which is also easily veri
ed by direct calculation. In Chapters 6 and 7 we will be interested
in the STA form of the Lagrangian for the Dirac equation so, as an illustration of (4.95),
this is given here:
L = h'j(^ (j@ ; eA ) ; m)ji $ hri3~ ; eA0~ ; m~i: (4:96)
By utilising (4.95) the STA forms of the Dirac spinor bilinear covariants 60] are readily
found. For example,
h'j^ ji $ h ~ 0i ; h ~ i3ii3 = 
h0~i1 (4:97)
identi
es the vector 0~ as the coordinate-free representation of the Dirac current. Since
~ is even and reverses to give itself, it contains only scalar and pseudoscalar terms. We
can therefore de
ne
ei
 :~ (4:98)
Assuming  6= 0,  can now be written as
 = 1=2ei
=2R (4:99)
where
R = (ei
);1=2: (4:100)
The even multivector R satis
es RR~ = 1 and is therefore a spacetime rotor. Double-sided
application of R on a vector a produces a Lorentz transformation. The STA equivalents
of the full set of bilinear covariants 33] can now be written as
Scalar h' ji $ h~i =  cos

Vector h'j^ ji $ 0~ = v
Bivector h'jj ^
ji $ i3~ = ei
S (4:101)
'
Pseudovector hj^ ^5 ji $ 3 = s ~
Pseudoscalar h'jj ^5 ji $ hi ~ i = ; sin

where
v = R0R~ (4:102)
s = R3R~
and
S = isv: (4:103)
81
These are summarised neatly by the equation
(1 + 0)(1 + i3)~ =  cos
+ v + ei
S + is + i sin
: (4:104)
The full Dirac spinor  contains (in the rotor R) an instruction to carry out a rotation
of the
xed f g frame into the frame of observables. The analogy with rigid-body
dynamics discussed in Section 4.1 therefore extends immediately to the relativistic theory.
The single-sided transformation law for the spinor  is also \understood" in the same way
that it was for Pauli spinors.
Once the spinor bilinear covariants are written in STA form (4.101) they can be manip-
ulated far more easily than in conventional treatments. For example the Fierz identities,
which relate the various observables (4.101), are simple to derive 33]. Furthermore, recon-
stituting  from the observables (up to a gauge transformation) is now a routine exercise,
carried out by writing
hiS = 41 ( + 00 ; i3( + 00)i3)
= 41 ( + 00 + 33 + 33) (4.105)
so that
h~iS = 41 (ei
+ v0 ; ei
Si3 + s3): (4:106)
The right-hand side of (4.106) can be found directly from the observables, and the left-
hand side gives  to within a complex multiple. On de
ning
Z = 41 (ei
+ v0 ; ei
Si3 + s3) (4:107)
we
nd that, up to an arbitrary phase factor,
 = (ei
)1=2Z (Z Z~ );1=2: (4:108)
An arbitrary Dirac operator M^ ji is replaced in the STA by a multilinear function
M (), which acts linearly on the entire even subalgebra of the STA. The 64 real dimen-
sions of this space of linear operators are reduced to 32 by the constraint (4.39)
M (i3) = M ()i3: (4:109)
Proceeding as at (4.44), the formula for the Dirac adjoint is
MDA () = M~' (~): (4:110)
Self-adjoint Dirac operators satisfy M~ () = M' (~) and include the ^ . The Hermitian
adjoint, MHA , is derived in the same way:
MHA() = M' y(y) (4:111)
in agreement with the non-relativistic equation (4.44).
Two important operator classes of linear operators on  are projection and symmetry
operators. The particle/antiparticle projection operators are replaced by
1 (m  ^ p )ji $ 1 (m  p ) (4:112)
0
2m 2m
82
and the spin-projection operators become
1 (1  ^ s ^ )j i $ 1 (  s ): (4:113)
2 5 2 3
Provided that p
s = 0, the spin and particle projection operators commute.
The three discrete symmetries C , P and T translate equally simply (following the
convention of Bjorken & Drell 59]):
P^ ji $ 0('x)0
C^ ji $ 1 (4:114)
^T ji $ i0(;x')1
where x' = 0x0 is (minus) a reection of x in the time-like 0 axis.
The STA representation of the Dirac matrix algebra will be used frequently throughout
the remainder of this thesis. In particular, it underlies much of the gauge-theory treatment
of gravity discussed in Chapter 7.

4.3.1 Changes of Representation | Weyl Spinors

In the matrix theory, a change of representation is performed with a 4  4 complex matrix
S^. This de
nes new matrices
^0 = S^^ S^;1 (4:115)
with a corresponding spinor transformation ji 7! S^ ji. For the Dirac equation, it is
also required that the transformed Hamiltonian be Hermitian, which restricts (4.115) to
a unitary transformation
^0 = S^^ S^y S^S^y = 1: (4:116)
The STA approach to handling alternative matrix representations is to
nd a suitable
analogue of the Dirac-Pauli map (4.87) which ensures that the eect of the matrix oper-
ators is still given by (4.88). The relevant map is easy to construct once the S^ is known
which relates the new representation to the Dirac-Pauli representation. One starts with a
column spinor ji0 in the new representation, constructs the equivalent Dirac-Pauli spinor
S^yji0, then maps this into its STA equivalent using (4.87). This technique ensures that
the action of j and the f^  ^5g matrices is still given by (4.88), and the C^ , P^ and T^
operators are still represented by (4.114). The STA form of the Dirac equation is always
given by (4.92) and so is a truly representation-free expression.
The STA from of the Dirac and Hermitian adjoints is always given by the formu-
lae (4.110) and (4.111) respectively. But the separate transpose and complex conjugation
operations retain some dependence on representation. For example, complex conjugation
in the Dirac-Pauli representation is given by (4.90)
ji $ ;22: (4:117)
In the Majorana representation, however, we
nd that the action of complex conjugation
on the Majorana spinor produces a dierent eect on its STA counterpart,
jiMaj $ 2: (4:118)
83
In the operator/matrix theory complex conjugation is a representation-dependent con-
cept. This limits its usefulness for our representation-free approach. Instead, we think
of  7! ;22 and  7! 2 as distinct operations that can be performed on the
multivector . (Incidentally, equation 4.118 shows that complex conjugation in the Ma-
jorana representation does indeed coincide with our STA form of the charge conjugation
operator (4.114), up to a conventional phase factor.)
To illustrate these techniques consider the Weyl representation, which is de
ned by
the matrices 60]
 !  !
^0 = ;I 0 0 ; I and ^k = ^ 0 ;^ k
0 : (4:119)
0 0
k
The Weyl representation is obtained from the Dirac-Pauli representation by the unitary
matrix  !
1
u^ = p ;I I : I I (4:120)
2
A spinor in the Weyl representation is written as
 !
ji = jj'ii 
0
(4:121)
where ji and j'i are 2-component spinors. Acting on ji0 with u^y gives
 !
j 
u^ ji = p2 ji + j'i:
y 0 1 i ; j 
' i (4:122)
Using equation (4.87), this is mapped onto the even element
 !
u^ ji = p2 ji + j'i
y 0 1 j i ; j 
' i $  =  p12 (1 + 3) ; ' p12 (1 ; 3) (4:123)
where  and ' are the Pauli-even equivalents of the 2-component complex spinors ji and
j'i, as de
ned by equation (4.9). The even multivector
 =  p12 (1 + 3) ; ' p12 (1 ; 3) (4:124)
is therefore our STA version of the column spinor
 !
ji = jj'ii 
0
(4:125)
where ji0 is acted on by matrices in the Weyl representation. As a check, we observe
that  !
^00 ji0 = ;ji ;j 'i $ ;' p12 (1 + 3) +  p12 (1 ; 3) = 00 (4:126)
and  !
^k ji = ^ ji ; 
^ k j 
' i $ ; k '3 p12 (1 + 3) ; k 3 p12 (1 ; 3) = k 0: (4:127)
k
(We have used equation (4.12) and the fact that 0 commutes with all Pauli-even ele-
ments.) The map (4.123) does indeed have the required properties.
84
4.4 The Multiparticle Spacetime Algebra
We now turn to the construction of the relativistic multiparticle STA. The principle is
simple. We introduce a set of four (anticommuting) basis vectors f i g,  = 0 : : : 3,
i = 1 : : : n where n is the number of particles. These vectors satisfy
 i

j =
ij 
(4:128)
and so span a 4n-dimensional space. We interpret this as n-particle con
guration space.
The construction of such a space is a standard concept in classical mechanics and non-
relativistic quantum theory, but the construction is rarely extended to relativistic systems.
This is due largely to the complications introduced by a construction involving multiple
times. In particular, Hamiltonian techniques appear to break down completely if a strict
single-time ordering of events is lost. But we shall see that the multiparticle STA is ideally
suited to the construction of relativistic states. Furthermore, the two-particle current no
longer has a positive-de
nite timelike component, so can describe antiparticles without
the formal requirement for
eld quantisation.
We will deal mainly with the two-particle STA. A two-particle quantum state is rep-
resented in this algebra by the multivector  = )E , where E = E2 is the two-particle
correlator (4.54) and ) is an element of the 64-dimensional direct product space of the
two even sub-algebras of the one-dimensional algebras. This construction ensures that 
is 32-dimensional, as is required for a real equivalent of a 16-component complex column
vector. Even elements from separate algebras automatically commute (recall (4.50)) so a
direct product state has  = 12E = 21E . The STA equivalent of the action of the
two-particle Dirac matrices ^i is de
ned by the operators

i () =  i 0i : (4:129)
These operators satisfy

1
2() =  1 2 0201 =  2  1 0102 =
2
1() (4:130)
and so, despite introducing a set of anticommuting vectors, the
i from dierent particle
spaces commute. In terms of the matrix theory, we have the equivalences
^  I ji $
1() (4.131)
I  ^ ji $
(): 2 (4.132)
Conventional treatments (e.g. Corson 62]) usually de
ne the operators

() = 21 
1() +
2()] (4:133)
which generate the well-known Dun-Kemmer ring




 +



= 

+ 

: (4:134)
This relation is veri
ed by
rst writing



() = 14 (
)1 + (
)2 + 
120201 + 
210102] (4.135)
)



() = 81  1
01 +  2
 02 + 
1  2 02 + 
2  1 01 +
 1
202 +  2
101 +  1 
202 +  2 
101] (4.136)
85
where 
=  
etc. In forming



 +



we are adding a quantity to its reverse,
which simply picks up the vector part of the products of vectors sitting on the left-hand
side of  in (4.136). We therefore
nd that
(


 +



) = 14 h 1
 + 
2  1 +  2
1 +  2
1i101 +
h 2
 + 
1  2 +  1
2 +  1 
2i102]
= 12 
1 + 
  1]01 + 21 
2 + 
  2]02
= 

() + 

(): (4.137)
The realisation of the Dun-Kemmer algebra demonstrates that the multiparticle STA
contains the necessary ingredients to formulate the relativistic two-particle equations that
have been studied in the literature.
The simplest relativistic two-particle wave equation is the Dun-Kemmer equation
(see Chapter 6 of 62]), which takes the form
@
()J = m: (4:138)
Here,  is a function of a single set of spacetime coordinates x , and @ = @x . Equa-
tion (4.138) describes a non-interacting
eld of spin 0
1. Since the wavefunction is
a function of one spacetime position only, (4.138) is not a genuine two-body equation.
Indeed, equation (4.138) has a simple one-body reduction, which is achieved by replacing
 by a 4  4 complex matrix 62, 63].
The
rst two-particle equation to consider in which  is a genuine function of position
in con
guration space is the famous Bethe-Salpeter equation 64]
(j r^ 1 ; m1)(j r^ 2 ; m2)j(x1 x2)i = jI j(x1 x2)i (4:139)
where r^ 1 = ^1 @x 1 etc. and I is an integral operator describing the inter-particle in-
teraction (Bethe & Salpeter 64] considered a relativistic generalisation of the Yukawa
potential). The STA version of (4.139) is
r1r20201 + m1r202 + m2r101 ; I ()]J = m1m2 (4:140)
where r1 and r2 are vector derivatives in the particle 1 and particle 2 spaces respectively.
An alternative approach to relativistic two-body wave equations was initiated by
Breit 65] in 1929. Breit wrote down an approximate two-body equation based on an
equal-time approximation and applied this approximation to the
ne structure of ortho-
helium. Breit's approach was developed by Kemmer 66] and Fermi & Yang 67], who
introduced more complicated interactions to give phenomenological descriptions of the
deuteron and pions respectively. More recently, this work has been extended by a number
of authors (see Koide 68] and Gale~oa & Leal Ferriara 63] and references therein). These
approaches all make use of an equation of the type (in STA form)
E + (01 ^r1 + 02 ^r2)J ; m1001 ; m20202 ; I () = 0 (4:141)
where  = (x1 x2) is a function of position in con
guration space and I () again
describes the inter-particle interaction. Equation (4.141) can be seen to arise from an
equal-time approximation to the STA equation
01(r1J + m101) + 02(r2J + m102) ; I () = 0: (4:142)
86
In the case where the interaction is turned o and  is a direct-product state,
 = 1(x1)2(x2)E (4:143)
equation (4.142) recovers the single-particle Dirac equations for the two separate particles.
(This is also the case for the Bethe-Salpeter equation (4.139).) The presence of the 01 and
02 on the left-hand side mean that equation (4.142) is not Lorentz covariant, however,
so can at best only be an approximate equation. From the STA form (4.139), one can
immediately see how to proceed to a fully covariant equation. One simply removes the
0's from the left. The resultant equation is
(r101 + r202)J ; I () = (m1 + m2) (4:144)
and indeed such an equation has recently been proposed by Krolikowski 69, 70] (who did
not use the STA).
These considerations should make it clear that the multiparticle STA is entirely su-
cient for the study of relativistic multiparticle wave equations. Furthermore, it removes
the need for either matrices or an (uninterpreted) scalar imaginary. But, in writing
down (4.144), we have lost the ability to recover the single-particle equations. If we set
I () to zero and use (4.143) for , we
nd that


2(ri3)1 + 1(ri3)2 ; (m1 + m2)12 E = 0: (4:145)
On dividing through by 12 we arrive at the equation
(1);1(ri3)1 + (2);1(ri3)2 ; m1 ; m2 = 0 (4:146)
and there is now no way to ensure that the correct mass is assigned to the appropriate
particle.
There is a further problem with the equations discussed above. A multiparticle action
integral will involve integration over the entire 4n-dimensional con
guration space. In
order that boundary terms can be dealt with properly (see Chapter 6) such an integral
should make use of the con
guration space vector derivative r = r1 + r2. This is not the
case for the above equations, in which the r1 and r2 operators act separately. We require
a relativistic two-particle wave equation for particles of dierent masses which is derivable
from an action integral and recovers the individual one-particle equations in the absence of
interactions. In searching for such an equation we are led to an interesting proposal | one
that necessitates parting company with conventional approaches to relativistic quantum
theory. To construct a space on which the full r can act, the 32-dimensional spinor
space considered so far is insucient. We will therefore extend our spinor space to the
the entire 128-dimensional even subalgebra of the two-particle STA. Right multiplication
by the correlator E then reduces this to a 64-dimensional space, which is now sucient
for our purposes. With  now a member of this 64-dimensional space, a suitable wave
equation is
r1 r2
( m1 + m2 )J ; (01 + 02) ; I () = 0: (4:147)
The operator (r1=m1 + r2=m2) is formed from a dilation of r, so can be easily in-
corporated into an action integral (this is demonstrated in Chapter 6). Furthermore,
87
equation (4.147) is manifestly Lorentz covariant. In the absence of interactions, and with
 taking the form of (4.143), equation (4.147) successfully recovers the two single-particle
Dirac equations. This is seen by dividing through by 12 to arrive at
 !
1 r1 1i1 + 1 r2 2i2 ;  1 ;  2 E = 0: (4:148)
1 m1 3 2 m2 3 0 0
The bracketed term contains the sum of elements from the two separate spaces, so both
terms must vanish identically. This ensures that
1 r1 1i1 =  1
 1 m1 3 0
) r11i31 = m1101 (4.149)
with the same result holding in the space of particle two. The fact that the particle
masses are naturally attached to their respective vector derivatives is interesting, and will
be mentioned again in the context of the STA gauge theory of gravity (Chapter 7).
No attempt at solving the full equation (4.147) for interacting particles will be made
here (that would probably require a thesis on its own). But it is worth drawing attention
to a further property of the equation. The current conjugate to gauge transformations is
given by
j = mj 1 + mj 2
1 2
(4:150)
where j 1 and j 2 are the projections of h(01 + 02)~i1 into the individual particle spaces.
The current j satis
es the conservation equation
r
j = 0 (4:151)
or
r1 + r2 )
h( 1 +  2)~i = 0:
(m (4:152)
1 m2 0 0 1
For the direct-product state (4.143) the projections of j into the single-particle spaces
take the form
j 1 = h2~2i(101~1) (4:153)
j 2 = h1~1i(202~2):
But the quantity h~i is not positive de
nite, so the individual particle currents are no
longer necessarily future-pointing. These currents can therefore describe antiparticles. (It
is somewhat ironic that most of the problems associated with the single-particle Dirac
equation can be traced back to the fact that the timelike component of the current is
positive de
nite. After all, producing a positive-de
nite density was part of Dirac's initial
triumph.) Furthermore, the conservation law (4.151) only relates to the total current in
con
guration space, so the projections onto individual particle spaces have the potential
for very strange behaviour. For example, the particle 1 current can turn round in space-
time, which would be interpreted as an annihilation event. The interparticle correlations
induced by the con
guration-space current j also aord insights into the non-local aspects
of quantum theory. Equation (4.147) should provide a fruitful source of future research,
as well as being a useful testing ground for our ideas of quantum behaviour.
88
4.4.1 The Lorentz Singlet State
Returning to the 32-dimensional spinor space of standard two-particle quantum theory,
our next task is to
nd a relativistic analogue of the Pauli singlet state discussed in
Section 4.2.1. Recalling the de
nition of  (4.70), the property that ensured that  was a
singlet state was that
ik1 = ;ik2 k = 1 : : : 3: (4:154)
In addition to (4.154), a relativistic singlet state, which we will denote as , must satisfy
k1 = ;k2 k = 1 : : : 3: (4:155)
It follows that  satis
es
i1 = 112131 = ;322212 = i2 (4:156)
so that
 = ;i1i2 (4.157)
)  = 2 (1 ; i i ):
1 1 2 (4.158)
Such a state can be formed by multiplying  by the idempotent (1 ; i1i2)=2. We therefore
de
ne
   12 (1 ; i1i2) = p12 (i21 ; i22) 21 (1 ; i31i32) 12 (1 ; i1i2): (4:159)
This satis
es
ik1 = ik1 12 (1 ; i1i2) = ;ik2 k = 1 : : : 3 (4:160)
and
k1 = ;k1i1i2 = i2ik2 = ;k2 k = 1 : : : 3: (4:161)
These results are summarised by
M 1 = M~ 2 (4:162)
where M is an even multivector in either the particle 1 or particle 2 STA. The proof that
 is a relativistic invariant now reduces to the simple identity
R1R2 = R1R~1 =  (4:163)
where R is a relativistic rotor acting in either particle-one or particle-two space.
Equation (4.162) can be seen as arising from a more primitive relation between vectors
in the separate spaces. Using the result that 0102 commutes with , we can derive
 1 01 =  10102020101
= 02( 0)102
= 0202 2 02
=  202 (4.164)
and hence we
nd that, for an arbitrary vector a,
a101 = a202: (4:165)
89
Equation (4.162) follows immediately from (4.165) by writing
(ab)1 = a1b10101
= a1b20201
= b2a10102
= b2a20202
= (ba)2: (4.166)
Equation (4.165) can therefore be viewed as the fundamental property of the relativistic
invariant .
From  a number of Lorentz-invariant two-particle multivectors can be constructed by
sandwiching arbitrary multivectors between  and ~. The simplest such object is
~ =  12 (1 ; i1i2)~
= 21 (1 + i11i12 + i21i22 + i31i32) 12 (1 ; i1i2)
= 14 (1 ; i1i2) ; 14 (k1k2 ; ik1ik2): (4.167)
This contains a scalar + pseudoscalar term, which is obviously invariant, together with
the invariant grade-4 multivector (k1k2 ; ik1ik2). The next simplest object is
0102~ = 21 (1 + i11i12 + i21i22 + i31i32) 12 (1 ; i1i2)0102
= 14 (0102 + i1i2k1k2 ; i1i20102 ; k1k2)
= 14 (0102 ; k1k2)(1 ; i1i2): (4.168)
On de
ning the symplectic (doubling) bivector
J   1 2 (4:169)
and the two-particle pseudoscalar
I  i1i2 = i2i1 (4:170)
the invariants from (4.168) are simply J and IJ . As was disussed in Section (3.2), the
bivector J is independent of the choice of spacetime frame, so is unchanged by the two-
sided application of the rotor R = R1R2. It follows immediately that the 6-vector IJ is
also invariant.
From the de
nition of J (4.169), we
nd that
J ^ J = ;20102k1k2 + (k1k2) ^ (j1j2)
= 2(k1k2 ; ik1ik2) (4.171)
which recovers the 4-vector invariant from (4.167). The complete set of two-particle invari-
ants can therefore be constructed from J alone, and these are summarised in Table 4.2.
These invariants are well-known and have been used in constructing phenomenological
models of interacting particles 63, 68]. The STA derivation of the invariants is quite new,
however, and the role of the doubling bivector J has not been previously noted.
90
 Type of
Invariant Interaction Grade
1 Scalar 0
J Vector 2
J ^J Bivector 4
IJ Pseudovector 6
I Pseudoscalar 8

Table 4.2: Two-Particle Relativistic Invariants

4.5 2-Spinor Calculus
We saw in Section 4.3.1 how spinors in the Weyl representation are handled within the
(single-particle) STA. We now turn to a discussion of how the 2-spinor calculus developed
by Penrose & Rindler 36, 37] is formulated in the multiparticle STA. From equation (4.87),
the chiral projection operators 21 (1  ^50 ) result in the STA multivectors
2 j i0 $  21 (1 + 3) =  p12 (1 + 3)
1 (1 + ^ 0 ) 
5 (4:172)
2 ; j i0 $  21 (1 ; 3) = ;' p12 (1 ; 3):
1 (1 ^ 0 ) 
5

The 2-spinors ji and j'i can therefore be given the STA equivalents
ji $  p12 (1 + 3) (4:173)
j'i $ ;' p12 (1 ; 3):
These dier from the representation of Pauli spinors, and are closer to the \minimal left
ideal" de
nition of a spinor given by some authors (see Chapter 2 of 13], for example).
Algebraically, the (1  3) projectors ensure that the 4-dimensional spaces spanned by
elements of the type  p12 (1 + 3) and ' p12 (1 ; 3) are closed under left multiplication by
a relativistic rotor. The signi
cance of the (1  3) projectors lies not so much in their
algebraic properties, however, but in the fact that they are the 0-space projections of
the null vectors 0  3 . This will become apparent when we construct some 2-spinor
\observables".
Under a Lorentz transformation the spinor  transforms to R, where R is a relativistic
rotor. If we separate the rotor R into Pauli-even and Pauli-odd terms,
R = R+ + R; (4:174)
where
R+ = 21 (R + 0R0 ) (4.175)
R; = 21 (R ; 0R0) (4.176)

91
then we can write
R p12 (1 + 3) = R+  p12 (1 + 3) + R; 3 p12 (1 + 3)
R' p12 (1 ; 3) = R+ ' p12 (1 ; 3) ; R; '3 p12 (1 ; 3): (4:177)

The transformation laws for the Pauli-even elements  and ' are therefore
 !7 R+ + R; 3 (4.178)
' !7 R+' ; R; '3 (4.179)
which con
rms that ji transforms under the operator equivalent of R, but that j'i
transforms under the equivalent of
~ 0)~= (R;1)y:
R+ ; R; = 0R0 = (0R (4:180)
This split of a Lorentz transformations into two distinct operations is an unattractive
feature of the 2-spinor formalism, but it is an unavoidable consequence of attempting to
perform relativistic calculations within the Pauli algebra of 2  2 matrices. The problem
is that the natural anti-involution operation is Hermitian conjugation. This operation is
dependent on the choice of a relativistic timelike vector, which breaks up expressions in
a way that disguises their frame-independent meaning.
The 2-spinor calculus attempts to circumvent the above problem by augmenting the
basic 2-component spinor with a number of auxilliary concepts. The result is a language
which has proved to be well-suited to the study of spinors in a wide class of problems and it
is instructive to see how some features of the 2-spinor are absorbed into the STA formalism.
The central idea behind the 2-spinor calculus is that a two-component complex spinor ji,
derived form the Weyl representation (4.121), is replaced by the complex \vector" A .
Here the A is an abstract index labeling the fact that A is a single spinor belonging to
some complex, two-dimensional linear space. We represent this object in the STA as
A $  21 (1 + 3): (4:181)
p
(The factor of 1=2 replaces 1= 2 simply for convenience in some of the manipulations
that follow.) The only dierence now is that, until a frame is chosen in spin-space, we
have no direct mapping between the components of A and . Secifying a frame in
spin space also picks out a frame in spacetime (determined by the null tetrad). If this
spacetime frame is identi
ed with the f g frame, then the components A of A specify
the Pauli-even multivector  via the identi
cation of equation (4.9). A second frame in
spin-space produces dierent components A, and will require a dierent identi
cation
to equation (4.9), but will still lead to the same multivector  21 (1 + 3). 2-Spinors are
equipped with a Lorentz-invariant inner product derived from a metric tensor AB . This
is used to lower indices so, for every 2-spinor A , there is a corresponding A. Both of
these must have the same multivector equivalent, however, in the same way that a and
a both have the STA equivalent a.
To account for the second type of relativistic 2-spinor, j'i (4.121), a second linear
space (or module) is introduced and elements of this space are labeled with bars and

92
primed indices. Thus an abstract element of this space is written as !' A . In a given basis,
0

the components of !' A are related to those of !A by complex conjugation,

0

!' 0 = !0
0
!' 1 = !1:
0
(4:182)
To construct the STA equivalent of !' A we need a suitable equivalent for this operation.
0

Our equivalent operation should satisfy the following criteria:

1. The operation can only aect the right-hand side !(1 + 3)=2, so that Lorentz
invariance is not compromised!
2. From equation (4.173), the STA equivalent of !' A must be a multivector projected
0

by the (1;3)=2 idempotent, so the conjugation operation must switch idempotents!

3. The operation must square to give the identity!
4. The operation must anticommute with right-multiplication by i3.
The only operation satisfying all of these criteria is right-multiplication by some combi-
nation of 1 and 2. Choosing between these is again a matter of convention, so we will
represent 2-spinor complex conjugation by right-multiplication by ;1. It follows that
our representation for the abstract 2-spinor !' A is 0

!' A 0
$ ;! 12 (1 + 3)1 = ;!i2 21 (1 ; 3): (4:183)
Again, once a basis is chosen, ! is constructed using the identi
cation of equation (4.9)
with the components !0 = !' 0 and !1 = !' 1 .
0 0

4.5.1 2-Spinor Observables

Our next step in the STA formulation of 2-spinor calculus is to understand how to rep-
resent quantities constructed from pairs of 2-spinors. The solution is remarkably simple.
One introduces a copy of the STA for each spinor, and then simply multiplies the STA
elements together, incorporating suitable correlators as one proceeds. For example, the
quantity A'A becomes
0

A ' A
0
$ ;1 21 (1 + 31)2i22 12 (1 ; 32) 12 (1 ; i31i32): (4:184)
To see how to manipulate the right-hand side of (4.184) we return to the relativistic
two-particle singlet  (4.159). The essential property of  under multiplication by even
elements was equation (4.162). This relation is unaected by further multiplication of 
on the right-hand side by an element that commutes with E . We can therefore form the
object
 =  21 (1 + 31) (4:185)
(not to be confused with the non-relativistic Pauli singlet state) which will still satisfy
M 1 = M~ 2 (4:186)

93
for all even multivectors M . The 2-particle state  is still a relativistic singlet in the sense
of equation (4.163). From (4.185) we see that  contains
2 (1 ; i i ) 2 (1 + 3 ) 2 (1 ; i3 i3 ) = 2 (1 ; i3 i ) 2 (1 + 3 )E
1 12 1 1 1 1 2 1 12 1 1
= 12 (1 ; i32i2) 21 (1 + 31)E
= 12 (1 + 32) 21 (1 + 31)E (4.187)
so we can write
 = p12 (i21 ; i22) 21 (1 + 32) 12 (1 + 31)E: (4:188)
A second invariant is formed by right-sided multiplication by (1 ; 31)=2, and we de
ne
' =  21 (1 ; 31): (4:189)
Proceeding as above, we
nd that
' = p12 (i21 ; i22) 21 (1 ; 32) 12 (1 ; 31)E: (4:190)
This split of the full relativistic invariant  into  and ' lies at the heart of much of the
2-spinor calculus. To see why, we return to equation (4.184) and from this we extract the
quantity 21 (1 + 31) 12 (1 ; 32) 12 (1 ; i31i32). This can be manipulated as follows:
1 (1 +  1 ) 1 (1 ;  2 )E =  1 1 (1 ;  1) 1 (1 ;  1) 1 (1 ;  2)E 1
2 3 2 3 02 3 2 3 2 3 0
= 0 i2 2 (1 ; 3 )(;i2 ) 2 (1 ; 3 ) 2 (1 ; 32)E01
1 2 1 1 2 1 1 1
= 01i22 21 (1 ; 31)(i21 ; i22) 12 (1 ; 31) 12 (1 ; 32)E01
= 01i22 p12 (1 ; 31)'01
= ; p12 (01 + 31)i21'01 (4.191)
which shows how an ' arises naturally in the 2-spinor product. This ' is then used to
project everything to its left back down to a single-particle space. We continue to refer
to each space as a \particle space" partly to stress the analogy with relativistic quantum
states, but also simply as a matter of convenience. In 2-spinor calculus there is no actual
notion of a particle associated with each copy of spacetime.
Returning to the example of A ' A (4.184), we can now write
0

;1 21 (1 + 31)2i22 12 (1 ; 32)E = ;12i22 12 (1 + 31) 21 (1 ; 32)E

= 12 p12 (01 + 31)'01
=  p12 (0 + 3)~]1'01: (4.192)
p
The key part of this expression is the null vector (0 + 3)~= 2, which ispformed in the
usual STA manner by a rotation/dilation of the
xed null vector p (0 + 3)= 2 by the even
multivector . The appearance of the null vector (0 + 3 )= 2 can be traced back directly
to the (1 + 3)=2 idempotent, justifying the earlier comment that these idempotents have
a clear geometric origin.
There are three further manipulations of the type performed in equation (4.191) and
the results of these are summarised in Table 4.3. These results can be used to
nd a
single-particle equivalent of any expression involving a pair of 2-spinors. We will see
shortly how these reductions are used to construct a null tetrad, but
rst we need to
nd
an STA formulation of the 2-spinor inner product.
94
 2 3 2 ;
1 (1 +  1) 1 (1  2)E
3 = ; p12 (01 + 31)i21'01
2 ;
1 (1  1 ) 1 (1 +  2)E
3 2 3 = ; p12 (01 ; 31)i2101
1 (1 +  1 ) 1 (1 +  2)E
2 3 2 3 = ; p12 (11 + i21)
2 ; 3 2 ;
1 (1  1) 1 (1  2)E
3 = ; p12 (;11 + i21)'
Table 4.3: 2-Spinor Manipulations
4.5.2 The 2-spinor Inner Product
.
Spin-space is equipped with an anti-symmetric inner product, written as either A !A
or A !B AB . In a chosen basis, the inner product is calculated as
A!A = A!A = 0!1 ; 1!0 (4:193)
which yields a Lorentz-invariant complex scalar. The antisymmetry of the inner product
suggests forming the STA expression
1 (A ! B ; ! A B ) $ 21 (1!2 ; 2!1) 12 (1 + 31) 21 (1 + 32)E
2

1
= ; 12  p12 (1 + i2)~! ; p12 !(1 + i2)~ 
= ; p12 h(1 + i2)~!i104: (4.194)
The antisymmetric product therefore picks out the scalar and pseudoscalar parts of the
quantity (11 + i21)~!. This is sensible, as these are the two parts that are invariant
under Lorentz transformations. Fixing up the factor suitably, our STA representation of
the 2-spinor inner product will therefore be
A!A $ ;h(1 + i2)~!i04 = ;hi2!~ i + ihi1!~ i: (4:195)
That this agrees with the 2-spinor form in a given basis can be checked simply by ex-
panding out the right-hand side of (4.195).
A further insight into the role of the 2-spinor inner product is gained by assembling
the full even multivector (an STA spinor)
 =  12 (1 + 3) + !i2 21 (1 ; 3): (4:196)
The 2-spinor inner product can now be written as
~ =  12 (1 + 3) + !i2 21 (1 ; 3)]; 21 (1 + 3)i2!~ + 12 (1 ; 3)~]
= ; 12 (1 + 3)i2!~ + !i2 21 (1 ; 3)~
= ;h(1 + i2)~!i04 (4.197)
which recovers (4.195). The 2-spinor inner product is therefore seen to pick up both the
scalar and pseudoscalar parts of a full Dirac spinor product ~. Interchanging  and ! in
 (4.196) is achieved by right-multiplication by 1, which immediately reverses the sign
95
of ~. An important feature of the 2-spinor calculus has now emerged, which is that the
unit scalar imaginary is playing the role of the spacetime pseudoscalar. This is a point
in favour of 2-spinors over Dirac spinors, but it is only through consistent employment of
the STA that this point has become clear.
The general role of the AB tensor when forming contractions is also now clear. In the
STA treatment, AB serves to antisymmetrise onp the two particle indices carried by its
STA equivalent. (It also introduces a factor of 2, which is a result of the conventions
we have adopted.) This antisymmetrisation always results in a scalar + pseudoscalar
quantity, and the pseudoscalar part can always be pulled down to an earlier copy of
spacetime. In this manner, antisymmetrisation always removes two copies of spacetime,
as we should expect from the contraction operation.

4.5.3 The Null Tetrad

An important concept in the 2-spinor calculus is that of a spin-frame. This consists of
a pair of 2-spinors, A and !A say, normalised such that A !A = 1. In terms of the
full spinor  (4.196), this normalisation condition becomes ~ = 1. But this is simply
the condition which ensures that  is a spacetime rotor! Thus the role of a \normalised
spin-frame" in 2-spinor calculus is played by a spacetime rotor in the STA approach. This
is a considerable conceptual simpli
cation. Furthermore, it demonstrates how elements of
abstract 2-spinor space can be represented in terms of geometrically meaningful objects
| a rotor, for example, being simply a product of an even number of unit vectors.
Attached to the concept of a spin-frame is that of a null tetrad. Using A and !A as
the generators of the spin frame, the null tetrad is de
ned as follows:
la = A'A 0
$ ;12 21 (1 + 31)i22 12 (1 ; 32)E
= p12 (0 + 3)~]1'01
=  p12 (0 + 3)~]1'01 (4.198)

na = !A !' A 0
$ ;!1!2 21 (1 + 31)i2 12 (1 ; 32)E
= p12 !(0 + 3)~! ]1'01
=  p12 (0 ; 3)~]1'01 (4.199)

ma = A !' A 0
$ ;1!2 21 (1 + 31)i2 21 (1 ; 32)E
= p12 (0 + 3)~!]1'01
=  p12 (1 + i2)~]1'01 (4.200)
and
m' a = !A 'A 0
$ ;!12 21 (1 + 31)i2 21 (1 ; 32)E
= p12 !(0 + 3)~]1'01
=  p12 (1 ; i2)~]1'01: (4.201)
96
The key identity used to arrive at the
nal two expression is
(1 + i2)~ = (1 + 3)1~
= (1 + 3)1~
= ;1(1 + 3)i2!~
= ;1(1 + 3)1!~
= (0 + 3)~!: (4.202)
The simplest spin frame is formed when  = 1. In this case we arrive at the following
comparison with page 120 of Penrose & Rindler 36]!
la = p12 (ta + za) $ p1
2 (0 + 3 )
na = p12 (ta ; za) $ p (0
1 ; 3)
ma = p12 (xa ; jya) $
2
p (1 + i2 )
1 (4:203)
2
m' a = p12 (xa + jya) $ p1 (1
2 ; i2):
The signi
cant feature of this translation is that the \complex vectors" ma and m' a have
been replaced by vector + trivector combinations. This agrees with the observation that
the imaginary scalar in the 2-spinor calculus plays the role of the spacetime pseudoscalar.
We can solve (4.203) for the Minkowski frame fta xa ya zag (note how the abstract
indices here simply record the fact that the t : : :z are vectors). The only subtlety is
that, in recovering the vector ya from our expression for jya, we must post-multiply our
2-particle expression by i31. The factor of (1 + 31) means that at the one-particle level
this operation reduces to right-multiplication by i. We therefore
nd that
ta $ 0 ya $ ;2 (4:204)
xa $ 1 za $ 3:
The only surprise here is the sign of the y-vector 2. This sign can be traced back
to the fact that Penrose & Rindler adopt an usual convention for the 2 Pauli matrix
(page 16). This is also reected in the fact that they identify the quaternions with vectors
(page 22), and we saw in Section 1.2.3 that the quaternion algebra is generated by the
spatial bivectors fi1 ;i2 i3g.
An arbitrary spin-frame, encoded in the rotor R, produces a new null tetrad simply
by Lorentz rotating the vectors in (4.203), yielding
l = R p12 (0 + 3)R ~ m = R p12 (1 + i2)R ~
n = R p2 (0 ; 3)R
1 ~ m' = R p2 (1 ; i2)R:
1 ~ (4:205)

In this manner, the (abstract) null tetrad becomes a set of four arbitrary vector/trivector
combinations in (4.205), satisfying the anticommutation relations 4]
1
2 fl ng = 1 1
2 fm m' g = 1 all others = 0: (4:206)

97
4.5.4 The rA A Operator 0

The
nal 2-spinor object that we need a translation of is the dierential operator rA A . 0

The translation of rA A will clearly involve the vector derivative r =  @x and this must
0

appear in such a way that it picks up the correct transformation law under a rotation in
two-particle space. These observations lead us to the object
rA A $ r101
0
(4:207)
so that, under a rotation,
r101 7! R1R2r101 = R1rR201
= (RrR~ )101 (4.208)
and the r does indeed inherit the correct vector transformation law. In this chapter
we are only concerned with the \at-space" vector derivative r! a suitable formulation
for \curved-space" derivatives will emerge in Chapter 7. A feature of the way that the
multiparticle STA is employed here is that each spinor (1+ 3)=2 is a function of position
in its own spacetime,
j 21 (1 + 3j ) = j (xj ) 12 (1 + 3j ): (4:209)
When such an object is projected into a dierent copy of spacetime, the position depen-
dence must be projected as well. In this manner, spinors can be \pulled back" into the
same spacetime as the dierential operator r.
We are now in a position to form the contraction rA AB AB . We know that the role
0

of the AB is to antisymmetrise on the relevant p particle spaces (in this case the 2 and 3
spaces), together with introducing a factor of 2. Borrowing from the 2-spinor notation,
we denote this operation as 23. We can now write
rA A A = rA AB AB $ r112013 21 (1 + 33)E323
0 0
(4:210)
where we have introduced the notation ij for the  invariant (singlet state) under joint
rotations in the ith and j th copies of spacetime. Equation (4.210) is manipulated to give
r112013 21 (1 + 33)E323
= r1 p12 (i21 ; i22)3 12 (1 + 31) 12 (1 + 32) 12 (1 + 33)E32301


= r1 p12 ;i21h(1 + i2)~i204 + hi2(1 + i2)~i204 12 (1 + 31)2301E3(4.211)
and projecting down into particle-one space, the quantity that remains is
rA A A $ r p12 i2h(1 + i2)i04 + h(1 + i2)i2i04] 12 (1 + 3)0: (4:212)
0

We now require the following rearrangement:

i2h(1 + i2)i04 + h(1 + i2)i2i04] 21 (1 + 3)
= i2(hi2i ; ihi1i) ; hi + ihi3i] 12 (1 + 3)
= ;hi + ikhik i] 21 (1 + 3)
= ; 12 (1 + 3): (4.213)
98
Using this, we
nd that
rA A A $ ; p12 r 12 (1 + 3)0 = ; p12 r0 12 (1 ; 3)
0
(4:214)
where pulling the 0 across to the left-hand side demonstrates how the rA A switches
0

between idempotents (modules). Equation (4.214) essentially justi

es the procedure de-
scribed in this section, since the translation (4.214) isp\obvious" from the Weyl represen-
tation of the Dirac algebra (4.119). The factor of 1= 2 in (4.214) is no longer a product
of our conventions, but is an unavoidable aspect of the 2-spinor calculus. It remains to

nd the equivalent to the above for the expression rAA !'A . The translation for rAA is
0
0
0

obtained from that for rA A by switching the \particle" indices, so that

0

rAA $ ;r202 = ;r1'01:

0
(4:215)
Then, proceeding as above, we
nd that
rAA !'A $ ; p12 r!i2 12 (1 + 3)0:
0
0 (4:216)

4.5.5 Applications
The above constitutes the necessary ingredients for a complete translation of the 2-spinor
calculus into the STA. We close this chapter by considering two important applications.
It should be clear from both that, whilst the steps to achieve the STA form are often
quite complicated, the end result is nearly always more compact and easier to understand
than the original 2-spinor form. An objective of future research in this subject is to
extract from 2-spinor calculus the techniques which are genuinely powerful and useful.
These can then be imported into the STA, which will suitably enriched by so doing. The
intuitive geometric nature of the STA should then make these techniques available to a
wider audience of physicists than currently employ the 2-spinor calculus.
The Dirac Equation
The Dirac equation in 2-spinor form is given by the pair of equations 36, page 222]
rA A A = !' A
0 0

(4:217)
rAA !'A = A :
0
0

p p
The quantity  is de
ned to be m= 2, where m is the electron mass. The factor of 1= 2
demonstrates that such factors are intrinsic to the way that the rA A symbol encodes the
0

vector derivative. The equations (4.217) translate to the pair of equations

r 121 (1 + 3)0 = m!i 2 2 (1 ; 3)
1
(4:218)
;r!i2 2 (1 ; 3)0 = m 21 (1 + 3):
If we now de
ne the full spinor  by
 =  21 (1 + 3) + !2 12 (1 ; 3) (4:219)

99
we
nd that
r0 = m!2 21 (1 ; 3) ; m 12 (1 + 3)]i
= ;mi3: (4.220)
We thus recover the STA version of the Dirac equation (4.92)
ri3 = m0: (4:221)
Of the pair of equations (4.217), Penrose & Rindler write \an advantage of the 2-spinor
description is that the  -matrices disappear completely { and complicated  -matrix identi-
ties simply evaporate! " 36, page 221]. Whilst this is true, the comment applies even more
strongly to the STA form of the Dirac equation (4.221), in which complicated 2-spinor
identities are also eliminated!
Maxwell's Equations
In the 2-spinor calculus the real, antisymmetric tensor F ab is written as
F ab = AB A B + AB A B 
0 0 0 0
(4:222)
where AB is symmetric on its two indices. We
rst need the STA equivalent of AB .
Assuming initially that AB is arbitrary, we can write
AB $  12 (1 + 31) 12 (1 + 32)E =  21 (1 + 31)11 (4:223)
where  is an arbitrary element of the product space of the two single-particle Pauli-
even algebras. A complete basis for  is formed by all combinations of the 7 elements
f1 ik1 ik2g. The presence of the singlet  allows all elements of second space to be
projected down into the
rst space, and it is not hard to see that this accounts for all
possible even elements in the one-particle STA. We can therefore write
 21 (1 + 31)11 = M 1  (4:224)
where M is an arbitrary even element. The condition that AB is symmetric on its two
indices now becomes (recalling that  is antisymmetric on its two particle indices)
M 1  = ;M 2 = ;M~ 1 (4:225)
~
) M = ;M: (4:226)
This condition projects out from M the components that are bivectors in particle-one
space, so we can write
AB $ F 1 (4:227)
where F is now a bivector. For the case of electromagnetism, F is the Faraday bivector,
introduced in Section (1.2.5). The complete translation of F ab is therefore
F ab $ F 1 + F 11112 = F 1 (4:228)
100
where  is the full relativistic invariant.
The 2-spinor form of the Maxwell equations an be written
rA B AC BC = ;J AA
0 0
(4:229)
where J AA is a \real" vector (i.e. it has no trivector components). Recalling the con-
0

vention that ij denotes the singlet state in coupled fi j g-space, the STA version of
equation (4.229) is
r11201F 33424 = ;J 11301: (4:230)
This is simpli
ed by the identity
123424 = 13 (4:231)
which is proved by expanding the left-hand side and then performing the antisymmetri-
sation. The resultant equation is
r1F 313 = ;J 113 (4:232)
which has a one-particle reduction to
rF = J: (4:233)
This recovers the STA form of the Maxwell equations 17]. The STA form is remarkably
compact, makes use solely of spacetime quantities and has a number of computational
advantages over second-order wave equations 8]. The 2-spinor calculus also achieves a

rst-order formulation of Maxwell's equations, but at the expense of some considerable

abstractions. We will return to equation (4.233) in Chapters 6 and 7.

101
Chapter 5
Point-particle Lagrangians
In this chapter we develop a multivector calculus as the natural extension of the calculus
of functions of a single parameter. The essential new tool required for such a calculus is
the multivector derivative, and this is described
rst. It is shown how the multivector
derivative provides a coordinate-free language for manipulating linear functions (forming
contractions etc.). This supersedes the approach used in earlier chapters, where such
manipulations were performed by introducing a frame.
The remainder of this chapter then applies the techniques of multivector calculus
to the analysis of point-particle Lagrangians. These provide a useful introduction to the
techniques that will be employed in the study of
eld Lagrangians in the
nal two chapters.
A novel idea discussed here is that of a multivector-valued Lagrangian. Such objects are
motivated by the pseudoclassical mechanics of Berezin & Marinov 39], but can only
be fully developed within geometric algebra. Forms of Noether's theorem are given for
both scalar and multivector-valued Lagrangians, and for transformations parameterised
by both scalars and multivectors. This work is applied to the study of two semi-classical
models of electron spin. Some aspects of the work presented in this chapter appeared
in the papers \Grassmann mechanics, multivector derivatives and geometric algebra" 3]
and \Grassmann calculus, pseudoclassical mechanics and geometric algebra" 1].

5.1 The Multivector Derivative

The idea of a vector derivative was partially introduced in Chapter 4, where it was seen
that the STA form of the Dirac equation (4.92) required the operator r =  @x , where
x = 
x. The same operator was later seen to appear in the STA form of the Maxwell
equations (4.233), rF = J . We now present a more formal introduction to the properties
of the vector and multivector derivatives. Further details of these properties are contained
in 18] and 24, Chapter 2], the latter of which is particularly detailed in its treatment.
Let X be a mixed-grade multivector
X
X = Xr  (5:1)
r
and let F (X ) be a general multivector-valued function of X . The grades of F (X ) need
not be the same as those of its argument X . For example, the STA representation of a
102
Dirac spinor as (x) is a map from the vector x onto an arbitrary even element of the
STA. The derivative of F (X ) in the A direction, where A has the same grades as X , is
de
ned by
A  @X F (X )  lim F (X + A) ; F (X ) : (5:2)
!0 
(It is taken as implicit in this de
nition that the limit exists.) The operator A  @X satis
es
all the usual properties for partial derivatives. To de
ne the multivector derivative @X , we
introduce an arbitrary frame fej g and extend this to de
ne a basis for the entire algebra
feJ g, where J is a general (simplicial) index. The multivector derivative is now de
ned
by X
@ X = eJ e J  @ X : (5:3)
J
The directional derivative eJ@X is only non-zero when eJ is of the same grade(s) as X , so
@X inherits the multivector properties of its argument X . The contraction in (5.3) ensures
that the quantity @X is independent of the choice of frame, and the basic properties of
@X can be formulated without any reference to a frame.
The properties of @X are best understood with the aid of some simple examples. The
most useful result for the multivector derivative is
@X hXAi = PX (A) (5:4)
where PX (A) is the projection of A on to the grades contained in X . From (5.4) it follows
that
~ i = PX (A~)
@X hXA
~ i = PX (A):
@X~ hXA (5:5)
Leibniz' rule can now be used in conjunction with (5.4) to build up results for the action
of @X on more complicated functions. For example,
@X hX X~ ik=2 = khX X~ i(k;2)=2X:
~ (5:6)
The multivector derivative acts on objects to its immediate right unless brackets are
present, in which case @X acts on the entire bracketed quantity. If @X acts on a multivector
that is not to its immediate right, we denote this with an overdot on the @X and its
argument. Thus @_X AB_ denotes the action of @X on B ,
@_X AB_ = eJ AeJ  @X B: (5:7)
The overdot notation is an invaluable aid to expressing the properties of the multivector
derivative. In particular, it neatly encodes the fact that, since @X is a multivector, it does
not necessarily commute with other multivectors and often acts on functions to which it
is not adjacent. As an illustration, Leibniz' rule can now be given in the form
@X (AB ) = @_X AB
_ + @_X AB: _ (5:8)
The only drawback with the overdot notation comes in expressions which involve time
derivatives. It is usually convenient to represent these with overdots as well, and in such
instances the overdots on multivector derivatives will be replaced by overstars.
103
 The most useful form of the multivector derivative is the derivative with respect to a
vector argument, @a or @x. Of these, the derivative with respect to position x is particularly
important. This is called the vector derivative, and is given special the symbol
@x = r = rx: (5:9)
The operator r sometimes goes under the name of the Dirac operator, though this name
is somewhat misleading since @x is well-de
ned in all dimensions and is in no way tied
to quantum-mechanical notions. In three dimensions, for example, @x = r contains all
the usual properties of the div, grad and curl operators. There are a number of useful
formulae for derivatives with respect to vectors, a selection of which is as follows:
@a a
b = b
@aa2 = 2a
@a
a = n
@a ^ a = 0 (5:10)
@aa
Ar = rAr
@aa ^ Ar = (n ; r)Ar
@aAra = (;1)r (n ; 2r)Ar 
where n is the dimension of the space. The
nal three equations in (5.10) are the frame-
free forms of formulae given in Section (1.3.2).
Vector derivatives are very helpful for developing the theory of linear functions, as
introduced in Section (1.3). For example, the adjoint to the linear function f can be
de
ned as
f (a)  @bhaf (b)i: (5:11)
It follows immediately that
b
f (a) = b
@chaf (c)i = hf (b
@cc)ai = f (b)
a: (5:12)
Invariants can be constructed equally simply. For example, the trace of f (a) is de
ned by
Trf  @a
f (a) (5:13)
and the \characteristic bivector" of f (a) is de
ned by
B = 12 @a ^ f (a): (5:14)
An anti-symmetric function f = ;f can always be written in the form
f (a) = a
B (5:15)
and it follows from equation (5.10) that B is the characteristic bivector.
Many other aspects of linear algebra, including a coordinate-free proof of the Cayley-
Hamilton theorem, can be developed similarly using combinations of vector derivatives 24,
Chapter 3].

104
5.2 Scalar and Multivector Lagrangians
As an application of the multivector derivative formalism just outlined, we consider
Lagrangian mechanics. We start with a scalar-valued Lagrangian L = L(Xi  X_ i ), where
the Xi are general multivectors, and X_ i denotes dierentiation with respect to time. We
wish to
nd the Xi (t) which extremise the action
Z t2
S = dt L(Xi  X_ i ): (5:16)
t1
The solution to this problem can be found in many texts (see e.g. 71]). We write
Xi (t) = Xi0(t) + Yi(t) (5:17)
where Yi is a multivector containing the same grades as Xi and which vanishes at the
endpoints,  is a scalar and Xi0 represents the extremal path. The action must now
satisfy @ S = 0 when  = 0, since  = 0 corresponds to Xi (t) taking the extremal values.
By applying the chain rule and integrating by parts, we
nd that
Z t2

@ S = dt (@ Xi)  @Xi L + (@ X_ i )  @X_ i L
Zt1t2

= t dt Yi  @Xi L + Y_i  @X_ i L
Z 1t2

= dt Yi  @Xi L ; @t(@X_ i L) : (5.18)
t1
Setting  to zero now just says that Xi is the extremal path, so the extremal path is
de
ned by the solutions to the Euler-Lagrange equations
@Xi L ; @t(@X_ i L) = 0: (5:19)
The essential advantage of this derivation is that it employs genuine derivatives in place
of the less clear concept of an in
nitessimal. This will be exempli
ed when we study
Lagrangians containing spinor variables.
We now wish to extend the above argument to a multivector-valued Lagrangian L.
Taking the scalar product of L with an arbitrary constant multivector A produces a scalar
Lagrangian hLAi. This generates its own Euler-Lagrange equations,
@Xi hLAi ; @t(@X_ i hLAi) = 0: (5:20)
A \permitted" multivector Lagrangian is one for which the equations from each A are
mutually consistent, so that each component of the full L is capable of simultaneous
extremisation.
By contracting equation (5.20) on the right-hand side by @A , we
nd that a necessary
condition on the dynamical variables is
@Xi L ; @t(@X_ i L) = 0: (5:21)
For a permitted multivector Lagrangian, equation (5.21) is also su
cient to ensure that
equation (5.20) is satis
ed for all A. This is taken as part of the de
nition of a multivector
Lagrangian. We will see an example of how these criteria can be met in Section 5.3.
105
5.2.1 Noether's Theorem
An important technique for deriving consequences of the equations of motion resulting
from a given Lagrangian is the study of the symmetry properties of the Lagrangian itself.
The general result needed for this study is Noether's theorem. We seek a form of this
theorem which is applicable to both scalar-valued and multivector-valued Lagrangians.
There are two types of symmetry to consider, depending on whether the transformation
of variables is governed by a scalar or by a multivector parameter. We will look at these
separately.
It is important to recall at this point that all the results obtained here are derived in
the coordinate-free language of geometric algebra. Hence all the symmetry transforma-
tions considered are active. Passive transformations have no place in this scheme, as the
introduction of an arbitrary coordinate system is an unnecessary distraction.

5.2.2 Scalar Parameterised Transformations

Given a Lagrangian L = L(Xi  X_ i ), which can be either scalar-valued or multivector-
valued, we wish to consider variations of the variables Xi controlled by a single scalar
parameter, . We write this as
Xi0 = Xi0(Xi ) (5:22)
and assume that Xi0( = 0) = Xi. We now de
ne the new Lagrangian
L0(Xi  X_ i) = L(Xi0 X_ i0) (5:23)
which has been obtained from L by an active transformation of the dynamical variables.
Employing the identity L0 = hL0Ai@A, we proceed as follows:
@L0 = (@Xi0)  @Xi hL0Ai@A + (@aX_ i0)  @X_ i hL0Ai@A



0 0

= (@aXi0)  @Xi hL0Ai ; @t(@X_ i hL0Ai) @A + @t (@aXi0)  @X_ i L0 : (5.24)

0 0 0

The de
nition of L0 ensures that it has the same functional form of L, so the quantity
@Xi hL0Ai ; @t(@X_ i hL0 Ai)L0
0 0 (5:25)
is obtained by taking the Euler-Lagrange equations in the form (5.20) and replacing the
Xi by Xi0. If we now assume that the Xi0 satisfy the same equations of motion (which
must be checked for any given case), we
nd that


@L0 = @t (@aXi0)  @X_ i L00 (5:26)
and, if L0 is independent of , the corresponding quantity (@aXi0)  @X_ i L0 is conserved.
0

Alternatively, we can set  to zero so that (5.25) becomes


@X hL0Ai ; @t(@ _ hL0Ai)L0] = @X hLAi ; @t(@ _ hLAi)
0
i Xi
0
=0 i Xi (5:27)
which vanishes as a consequence of the equations of motion for Xi . We therefore
nd that


@aL0j=0 = @t (@aXi0)  @X_ i L0 =0 
0 (5:28)
106
which is probably the most useful form of Noether's theorem, in that interesting conse-
quences follow from (5.28) regardless of whether or not L0 is independent of . A crucial
step in the derivation of (5.28) is that the Euler-Lagrange equations for a multivector-
valued Lagrangian are satis
ed in the form (5.20). Hence the consistency of the equa-
tions (5.20) for dierent A is central to the development of the theory of multivector
Lagrangians.
To illustrate equation (5.28), consider time translation
Xi0(t ) = Xi(t + ) (5.29)
) @aXi j=0 = X_ i:
0
(5.30)
Assuming there is no explicit time-dependence in L, equation (5.28) gives
@tL = @t(X_ i  @X_ i L) (5:31)
from which we de
ne the conserved Hamiltonian by
H = X_ i  @X_ i L ; L: (5:32)
If L is multivector-valued, then H will be a multivector of the same grade(s).

5.2.3 Multivector Parameterised Transformations

The most general single transformation for the variables Xi governed by a multivector M
can be written as
Xi0 = f (Xi M ) (5:33)
where f and M are time-independent functions and multivectors respectively. In gen-
eral f need not be grade-preserving, which provides a route to deriving analogues for
supersymmetric transformations.
To follow the derivation of (5.26), it is useful to employ the dierential notation 24],
f M (Xi  A)  A  @M f (Xi  M ): (5:34)
The function f M (Xi  A) is a linear function of A and an arbitrary function of M and Xi .
With L0 de
ned as in equation (5.23), we derive
A  @M L0 = f M (Xi A)  @Xi L0 + f M (X_ i  M )  @X_ i L0



0 0

= f M (Xi A)  @Xi hL0 B i ; @t(@X_ i hL0B i) @B + @t f M (Xi  A)  @X_ i L0

0 0 0

= @t f M (Xi  A)  @X_ i L0 
0 (5.35)
where again it is necessary to assume that the equations of motion are satis
ed for the
transformed variables. We can remove the A-dependence by dierentiating, which yields


@M L0 = @t @Af M (Xi A)  @X_ i L0 0 (5:36)
and, if L0 is independent of M , the corresponding conserved quantity is
@Af M (Xi A)  @X_ i L0 =@M f (Xi  M )  @X_ i L0
0 0 (5:37)
107
where the overstar on M denote the argument of @M .
It is not usually possible to set M to zero in (5.35), but it is interesting to see that
conserved quantities can be found regardless. This shows that standard treatments of
Lagrangian symmetries 71] are unnecessarily restrictive in only considering in
nitesimal
transformations. The subject is richer than this suggests, though without multivector
calculus the necessary formulae are hard to
nd.
In order to illustrate (5.37), consider reection symmetry applied to the harmonic
oscillator Lagrangian
L(x x_ ) = 21 (x_ 2 ; !2x2): (5:38)
The equations of motion are
x& = ;!2x (5:39)
and it is immediately seen that, if x is a solution, then so to is x0, where
x0 = ;nxn;1: (5:40)
Here n is an arbitrary vector, so x0 is obtained from x by a reection in the hyperplance
orthogonal to n. Under the reection (5.40) the Lagrangian is unchanged, so we can
nd
a conserved quantity from equation (5.37). With f (x n) de
ned by
f (x n) = ;nxn;1 (5:41)
we
nd that
f n(x a) = ;axn;1 + nxn;1an;1: (5:42)
Equation (5.37) now yields the conserved quantity
@a(;axn;1 + nxn;1 an;1)  (;nxn
_ ;1 ) = @ahaxxn _ ;1 ; axxn
_ ;1i
= hxxn
_ ;1 ; xxn _ ;1i1
= 2(x ^ x_ )
n;1: (5.43)
This is conserved for all n, from which it follows that the angular momentum x ^ x_
is conserved. This is not a surprise, since rotations can be built out of reections and
dierent reections are related by rotations. It is therefore natural to expect the same
conserved quantity from both rotations and reections. But the derivation does show
that the multivector derivative technique works and, to my knowledge, this is the
rst
time that a classical conserved quantity has been derived conjugate to transformations
that are not simply connected to the identity.

5.3 Applications | Models for Spinning Point Par-

ticles
There have been numerous attempts to construct classical models for spin-half particles
(see van Holten 72] for a recent review) and two such models are considered in this
section. The
rst involves a scalar point-particle Lagrangian in which the dynamical

108
variables include spinor variables. The STA formalism of Chapter 4 is applied to this
Lagrangian and used to analyse the equations of motion. Some problems with the model
are discussed, and a more promising model is proposed. The second example is drawn
from pseudoclassical mechanics. There the dynamical variables are Grassmann-valued
entities, and the formalism of Chapter 2 is used to represent these by geometric vectors.
The resulting Lagrangian is multivector-valued, and is studied using the techniques just
developed. The equations of motion are found and solved, and again it is argued that the
model fails to provide an acceptable picture of a classical spin-half particle.

1. The Barut-Zanghi Model

The Lagrangian of interest here was introduced by Barut & Zanghi 38] (see also 7, 61])
and is given by
'_ ; )' ))
L = 21 j ()) _ + p (x_ ; )'  )) + qA (x))'  ) (5:44)
where ) is a Dirac spinor. Using the mapping described in Section (4.3), the Lagran-
gian (5.45) can be written as
_ 3~ + p(x_ ; 0~) + qA(x)0~i:
L = hi (5:45)
The dynamical variables are x, p and , where  is an even multivector, and the dot
denotes dierentiation with respect to some arbitrary parameter  .
The Euler-Lagrange equation for  is
@ L = @ (@_ L)
) @ (i3~) = ;i3_~ ; 20p ~ + 2q0A
~
) i_ 3 = P0 (5.46)
where
P  p ; qA: (5:47)
In deriving (5.46) there is no pretence that  and ~ are independent variables. Instead
they represent two occurrences of the same variable  and all aspects of the variational
principle are taken care of by the multivector derivative.
The p equation is
x_ = 0~ (5:48)
but, since x_ 2 = 2 is not, in general, equal to 1,  cannot necessarily be viewed as the
proper time for the particle. The x equation is
p_ = qrA(x)
(0~)
= q(r^ A)
x_ + qx_
rA
) P_ = qF
x:_ (5.49)
We now use (5.28) to derive some consequences for this model. The Hamiltonian is
given by
H = x_  @x_ L + _  @_ L ; L
= P
x_ (5.50)
109
and is conserved absolutely. The 4-momentum and angular momentum are only conserved
if A = 0, in which case (5.45) reduces to the free-particle Lagrangian
L0 = hi _ 3~ + p(x_ ; 0~)i: (5:51)
The 4-momentum is found from translation invariance,
x0 = x + a (5:52)
and is simply p. The component of p in the x_ direction gives the energy (5.50). The
angular momentum is found from rotational invariance, for which we set
x0 = eB=2xe;B=2
p0 = eB=2pe;B=2 (5:53)
 = e :
0 B= 2

It is immediately apparent that L00 is independent of , so the quantity

(B
x)  @x_ L0 + 21 (B)  @_ L0 = B
(x ^ p + 21 i3~) (5:54)
is conserved for arbitrary B . The angular momentum is therefore de
ned by
~
J = p ^ x ; 21 i3 (5:55)
which identi
es ;i3= ~ 2 as the internal spin. The factor of 1=2 clearly originates from
the transformation law (5.53). The free-particle model de
ned by (5.51) therefore does
have some of the expected properties of a classical model for spin, though there is a
potential problem with the de
nition of J (5.55) in that the spin contribution enters with
the opposite sign to that expected from
eld theory (see Chapter 6).
Returning to the interacting model (5.45), further useful properties can be derived
from transformations in which the spinor is acted on from the right. These correspond
to gauge transformations, though a wider class is now available than for the standard
column-spinor formulation. From the transformation
0 = ei 3 (5:56)
we
nd that
@ h~i = 0 (5:57)
and the transformation
0 = e 3 (5:58)
yields
@ hi~i = ;2P
(3~): (5:59)
Equations (5.57) and (5.59) combine to give
@ (~) = 2iP
(3~): (5:60)
Finally, the duality transformation
0 = ei (5:61)
110
yields
_ 3~i = 0:
2h (5:62)
A number of features of the Lagrangian (5.45) make it an unsatisfactory model a clas-
sical electron. We have already mentioned that the parameter  cannot be identi
ed with
the proper time of the particle. This is due to the lack of reparameterisation invariance
in (5.45). More seriously, the model predicts a zero gyromagnetic moment 61]. Further-
more, the P_ equation (5.49) cannot be correct, since here one expects to see p_ rather than
P_ coupling to F
x. Indeed, the de
nition of P (5.47) shows that equation (5.49) is not
gauge invariant, which suggests that it is a lack of gauge invariance which lies behind
some of the unacceptable features of the model.

2. Further Spin-half Models

We will now see how to modify the Lagrangian (5.45) to achieve a suitable set of classical
equations for a particle with spin. The
rst step is to consider gauge invariance. Under
the local gauge transformation
 7!  expf;i3( )g (5:63)
_ 3~i transforms as
the \kinetic" spinor term hi
_ 3~i 7! hi
hi _ 3~i + h~_ i: (5:64)
The
nal term can be written as
h~_ i = h~x_
(r)i (5:65)
and, when r is generalised to an arbitrary gauge
eld qA, (5.64) produces the interaction
term
LI = qh~x_
Ai: (5:66)
This derivation shows clearly that the A
eld must couple to x_ and not to 0~, as it is
not until after the equations of motion are found that 0~ is set equal to x_ . That there
should be an x_
A term in the Lagrangian is natural since this is the interaction term for
a classical point particle, and a requirement on any action that we construct is that it
should reproduce classical mechanics in the limit where spin eects are ignored (i.e. as
h' 7! 0). But a problem still remains with (5.66) in that the factor of ~ is unnatural
and produces an unwanted term in the  equation.. To remove this, we must replace the
_ 3~i term by
hi
_ 3;1i
L0 = hi (5:67)
where, for a spinor  = (ei
)1=2R,
~
;1 = (ei
);1=2R: (5:68)
In being led the term (5.67), we are forced to break with conventional usage of column
spinors. The term (5.67) now suggests what is needed for a suitable classical model. The
_ 3;1i is unchanged by both dilations and duality transformations of  and
quantity hi
111
so is only dependent on the rotor part of . It has been suggested that the rotor part
of  encodes the dynamics of the electron
eld and that the factor of ( expfi
g)1=2 is
a quantum-mechanical statistical term 29]. Accepting this, we should expect that our
classical model should involve only the rotor part of  and that the density terms should
have no eect. Such a model requires that the Lagrangian be invariant under local changes
of  expfi
g, as we have seen is the case for L0 (5.67). The remaining spinorial term is
the current term 0~ which is already independent of the duality factor
. It can be
made independent of the density  as well by dividing by . From these observations we
are led to the Lagrangian
_ 3;1 + p(x_ ; 0=
L = hi ~ ) ; qx_
Ai: (5:69)
The p equation from (5.69) recovers
~ = R0R
x_ = 0= ~ (5:70)
so that x_ 2 = 1 and  is automatically the ane parameter. This removes one of the
defects with the Barut-Zanghi model. The x equation gives, using standard techniques,
p_ = qF
x_ (5:71)
which is now manifestly gauge invariant and reduces to the Lorentz force law when the
spin is ignored and the velocity and momentum are collinear, p = mx_ . Finally, the 
equation is found by making use of the results
@ hM;1i = M;1 + @_ hM _ ;1i = 0 (5:72)
) @ hM;1 i = ;;1M;1 (5:73)
and
@  = 21 @ (0
~ 0~)
= 2 0ei
0~
= 2;1 (5.74)
to obtain
_ 3;1 ; 1 (20p
; ;1i ~ ; 2;1hp0~i) = @ (i3;1): (5:75)

By multiplying equation (5.75) with , one obtains
1 S_ = p ^ x
_ (5:76)
2
where
~
S  i3;1 = Ri3R: (5:77)

112
Thus the  variation now leads directly to the precession equation for the spin. The
complete set of equations is now
S_ = 2p ^ x_ (5.78)
x_ = R0 R ~ (5.79)
p_ = qF
x_ (5.80)
which are manifestly Lorentz covariant and gauge invariant. The Hamiltonian is now p
x_
and the free-particle angular momentum is still de
ned by J (5.55), though now the spin
bivector S is always of unit magnitude.
A
nal problem remains, however, which is that we have still not succeeded in con-
structing a model which predicts the correct gyromagnetic moment. In order to achieve
the correct coupling between the spin and the Faraday bivector, the Lagrangian (5.69)
must be modi
ed to
_ 3;1 + p(x_ ; 0=
L = hi ~ ) + qx_
A ; q Fi3;1i: (5:81)
2m
The equations of motion are now
S_ = 2p ^ x_ + mq F  S
x_ = R0R~ (5.82)
q
p_ = qF
x_ ; 2m rF (x)
S
which recover the correct precession formulae in a constant magnetic
eld. When p is set
equal to mx_ , the equations (5.82) reduce to the pair of equations studied in 72].

3. A Multivector Model | Pseudoclassical Mechanics Recon-

sidered
Pseudoclassical mechanics 39, 73, 74] was originally introduced as the classical analogue
of quantum spin one-half (i.e. for particles obeying Fermi statistics). The central idea is
that the \classical analogue" of the Pauli or Dirac algebras is an algebra where all inner
products vanish, so that the dynamical variables are Grassmann variables. From the
point of view of this thesis, such an idea appears fundamentally awed. Furthermore, we
have already seen how to construct sensible semi-classical approximations to Dirac theory.
But once the Grassmann variables have been replaced by vectors through the procedure
outlined in Chapter 2, pseudoclassical Lagrangians do become interesting, in that they
provide examples of acceptable multivector Lagrangians. Such a Lagrangian is studied
here, from a number of dierent perspectives. An interesting aside to this work is a new
method of generating super-Lie algebras, which could form the basis for an alternative
approach to their representation theory.
The Lagrangian we will study is derived from a pseudoclassical Lagrangian introduced
by Berezin & Marinov 39]. This has become a standard example in non-relativistic pseu-
doclassical mechanics 73, Chapter 11]. With a slight change of notation, and dropping
an irrelevant factors of j , the Lagrangian can be written as
L = 21 i _i ; 21 ijk !ij k  (5:83)
113
where the fig are formally Grassmann variable and the f!ig are a set of three scalar
constants. Here, i runs from 1 to 3 and, as always, the summation convention is implied.
Replacing the set of Grassmann variables fig with a set of three (Cliord) vectors feig,
the Lagrangian (5.83) becomes 1]
L = 21 ei ^ e_i ; ! (5:84)
where
! = 12 ijk !i ej ek = !1(e2 ^ e3) + !2(e3 ^ e1) + !3(e1 ^ e2): (5:85)
The equations of motion from (5.84) are found by applying equation (5.21)
@ei 21 (ej ^ e_j ; !) = @t@e_i 12 (ej ^ e_j ; !)]
) e_i + 2ijk !j ek = ;@tei
) e_i = ;ijk !j ek : (5.86)
We have used the 3-dimensional result
@aa ^ b = 2b (5:87)
and we stress again that this derivation uses a genuine calculus, so that each step is
well-de
ned.
We are now in a position to see how the Lagrangian (5.84) satis
es the criteria to
be a \permitted" multivector Lagrangian. If B is an arbitrary bivector, then the scalar
Lagrangian hLB i produces the equations of motion
@ei hLB i ; @t(@e_i hLB i) = 0
) (_ei + ijk !j ek )
B = 0: (5.88)
For this to be satis
ed for all B , we simply require that the bracketed term vanishes.
Hence equation (5.86) is indeed sucient to ensure that each scalar component of L is
capable of simultaneous extremisation. This example illustrates a further point. For a

xed B , equation (5.88) does not lead to the full equations of motion (5.86). It is only
by allowing B to vary that we arrive at (5.86). It is therefore an essential feature of the
formalism that L is a multivector, and that (5.88) holds for all B .
The equations of motion (5.86) can be written out in full to give
e_1 = ;!2e3 + !3e2
e_2 = ;!3e1 + !1e3 (5:89)
e_3 = ;!1e2 + !2e1
which are a set of three coupled
rst-order vector equations. In terms of components,
this gives nine scalar equations for nine unknowns, which illustrates how multivector
Lagrangians have the potential to package up large numbers of equations into a single,
highly compact entity. The equations (5.89) can be neatly combined into a single equation
by introducing the reciprocal frame feig (1.132),
e1 = e2 ^ e3En;1 etc. (5:90)
114
where
En  e1 ^ e2 ^ e3: (5:91)
With this, the equations (5.89) become
e_i = ei
! (5:92)
which shows that potentially interesting geometry underlies this system, relating the
equations of motion of a frame to its reciprocal.
We now proceed to solve equation (5.92). On feeding (5.92) into (5.85), we
nd that
!_ = 0 (5:93)
so that the ! plane is constant. We next observe that (5.89) also imply
E_ n = 0 (5:94)
which is important as it shows that, if the feig frame initially spans 3-dimensional space,
then it will do so for all time. The constancy of En means that the reciprocal frame (5.90)
satis
es
e_1 = ;!2e3 + !3e2 etc. (5:95)
We now introduce the symmetric metric tensor g, de
ned by
g(ei) = ei: (5:96)
This de
nes the reciprocal bivector
!  g;1 (!)
= !1(e2 ^ e3) + !2(e3 ^ e1) + !3(e1 ^ e2) (5.97)
so that the reciprocal frame satis
es the equations
e_i = ei
! : (5:98)
But, from (1.123), we have that
ei
! = ei
g;1 (!) = g;1 (ei
!): (5:99)
Now, using (5.92), (5.98) and (5.99), we
nd that
g(e_i) = ei
! = e_i = @tg(ei) (5:100)
) g_ = 0: (5:101)
Hence the metric tensor is constant, even though its matrix coecients are varying. The
variation of the coecients of the metric tensor is therefore purely the result of the time
variation of the frame, and is not a property of the frame-independent tensor. It follows

115
that the
ducial tensor (1.144) is also constant, and suggests that we should look at the
equations of motion for the
ducial frame i = h;1 (ei). For the fig frame we
nd that
_ i = h;1(_ei)
= h;1(h;1(i)
!)
= i
h;1(!): (5.102)
If we de
ne the bivector
+ = h;1(!) = !123 + !231 + !312 (5:103)
(which must be constant, since both h and ! are), we see that the
ducial frame satis
es
the equation
_ i = i
+: (5:104)
The underlying
ducial frame simply rotates at a constant frequency in the + plane. If
i(0) denotes the
ducial frame speci
ed by the initial setup of the feig frame, then the
solution to (5.104) is
i(t) = e;t=2i(0)et=2 (5:105)
and the solution for the feig frame is
ei(t) = h(e;t=2i(0)et=2) (5:106)
ei(t) = h;1(e;t=2i(0)et=2):
Ultimately, the motion is that of an orthonormal frame viewed through a constant (sym-
metric) distortion. The feig frame and its reciprocal representing the same thing viewed
through the distortion and its inverse. The system is perhaps not quite as interesting as
one might have hoped, and it has not proved possible to identify the motion of (5.106)
with any physical system, except in the simple case where h = I . On the other hand, we
did start with a very simple Lagrangian and it is reassuring to recover a rotating frame
from an action that was motivated by the pseudoclassical mechanics of spin.
Some simple consequences follow immediately from the solution (5.106). Firstly, there
is only one frequency in the system,  say, which is found via
 2 = ;+2
= !12 + !22 + !32: (5.107)
Secondly, since
+ = i(!11 + !22 + !33) (5:108)
the vectors
u  !1e1 + !2e2 + !3e3 (5:109)
and
u = g;1 (u) (5:110)
are conserved. This also follows from
u = ;En! (5.111)
u = E !: n (5.112)
116
Furthermore,
eiei = h(i)h(i)
= ig(i)
= Tr(g) (5.113)
must also be time-independent (as can be veri
ed directly from the equations of motion).
The reciprocal quantity eiei = Tr(g;1 ) is also conserved. We thus have the set of four
standard rotational invariants, ii, the axis, the plane of rotation and the volume scale-
factor, each viewed through the pair of distortions h, h;1. This gives the following set of
8 related conserved quantities:
feiei, eiei, u, u , !, ! , En , E ng: (5:114)
Lagrangian Symmetries and Conserved Quantities
We now turn to a discussion of the symmetries of (5.84). Although we have solved the
equations of motion exactly, it is instructive to derive some of their consequences directly
from the Lagrangian. We only consider continuous symmetries parameterised by a single
scalar, so the appropriate form of Noether's theorem is equation (5.28), which takes the
form

@aL0j = @t 1 ei ^ (@ae0 )  :
=0 2 i =0
(5:115)
In writing this we are explicitly making use of the equations of motion and so are
nding
\on-shell" symmetries. The Lagrangian could be modi
ed to extend these symmetries
o-shell, but this will not be considered here.
We start with time translation. From (5.32), the Hamiltonian is
H = 21 ei ^ e_i ; L = ! (5:116)
which is a constant bivector, as expected. The next symmetry to consider is a dilation,
e0i = eei: (5:117)
For this transformation, equation (5.115) gives


2L = @t 12 ei ^ ei = 0 (5:118)
so dilation symmetry shows that the Lagrangian vanishes along the classical path. This
is quite common for
rst-order systems (the same is true of the Dirac Lagrangian), and
is important in deriving other conserved quantities.
The
nal \classical" symmetry to consider is a rotation,
e0i = eB=2eie;B=2: (5:119)
Equation (5.115) now gives

B  L = @t 21 ei ^ (B
ei) (5:120)

117
but, since L = 0 when the equations of motion are satis
ed, the left hand side of (5.120)
vanishes, and we
nd that the bivector ei ^ (B
ei) in conserved. Had our Lagrangian
been a scalar, we would have derived a scalar-valued function of B at this point, from
which a single conserved bivector | the angular momentum | could be found. Here our
Lagrangian is a bivector, so we
nd a conserved bivector-valued function of a bivector |
a set of 3  3 = 9 conserved scalars. The quantity ei ^ (B
ei) is a symmetric function of
B , however, so this reduces to 6 independent conserved scalars. To see what these are
we introduce the dual vector b = iB and replace the conserved bivector ei ^(B
ei) by the
equivalent vector-valued function,
f (b) = ei
(b ^ ei) = ei
bei ; beiei = g(b) ; bTr(g): (5:121)
This is conserved for all b, so we can contract with @b and observe that ;2Tr(g) is constant.
It follows that g (b) is constant for all b, so rotational symmetry implies conservation of
the metric tensor | a total of 6 quantities, as expected.
Now that we have derived conservation of g and !, the remaining conserved quantities
can be found. For example, En = det(g)1=2i shows that En is constant. One interesting
scalar-controlled symmetry remains, however, namely
e0i = ei + !i a (5:122)
where a is an arbitrary constant vector. For this symmetry (5.115) gives
1 a ^ u_ = @ 1 e ^ (! a)


2 t 2 i i (5:123)
) a ^ u_ = 0 (5:124)
which holds for all a. Conservation of u therefore follows directly from the symmetry
transformation (5.122). This symmetry transformation bears a striking resemblance to
the transformation law for the fermionic sector of a supersymmetric theory 75]. Although
the geometry behind (5.122) is not clear, it is interesting to note that the pseudoscalar
transforms as
En0 = En + a ^ ! (5:125)
and is therefore not invariant.
Poisson Brackets and the Hamiltonian Formalism
Many of the preceding results can also be derived from a Hamiltonian approach. As a
by-product, this reveals a new and remarkably compact formula for a super-Lie bracket.
We have already seen that the Hamiltonian for (5.84) is !, so we start by looking at
how the Poisson bracket is de
ned in pseudoclassical mechanics 39]. Dropping the j and
adjusting a sign, the Poisson bracket is de
ned by
@; @
fa( ) b( )gPB = a @ @ b: (5:126)
k k
The geometric algebra form of this is
fA B gPB = (A
ek) ^ (ek
B ) (5:127)
118
where A and B are arbitrary multivectors. We will consider the consequences of this
de
nition in arbitrary dimensions initially, before returning to the Lagrangian (5.84).
Equation (5.127) can be simpli
ed by utilising the
ducial tensor,
(A
h;1(k )) ^ (h;1(k )
B ) = hh;1(A)
k] ^ hk
h;1(B )]
= h(h;1(A)
k) ^ (k
h;1(B ))]: (5.128)
If we now assume that A and B are homogeneous, we can employ the rearrangement
(Ar
k ) ^ (k
Bs ) = 41 h(Ar k ; (;1)r k Ar )(k Bs ; (;1)sBs k )ir+s;2
= 41 hnAr Bs ; (n ; 2r)Ar Bs ; (n ; 2s)Ar Bs
+n ; 2(r + s ; 2)]ArBsir+s;2
= hAr Bs ir+s;2 (5.129)
to write the Poisson bracket as
fAr Bs gPB = hhh;1(Ar )h;1(Bs)ir+s;2 : (5:130)
This is a very neat representation of the super-Poisson bracket. The combination rule is
simple, since the h always sits outside everything:
D E
fAr fBs  CtgPB gPB = h h;1(Ar)hh;1(Bs)h;1(Ct)is+t;2 r+s+t;4 : (5:131)
Cliord multiplication is associative and
hAr Bsir+s;2 = ;(;1)rshBs Arir+s;2  (5:132)
so the bracket (5.130) generates a super-Lie algebra. This follows from the well-known
result 76] that a graded associative algebra satisfying the graded commutator rela-
tion (5.132) automatically satis
es the super-Jacobi identity. The bracket (5.130) there-
fore provides a wonderfully compact realisation of a super-Lie algebra. We saw in Chap-
ter 3 that any Lie algebra can be represented by a bivector algebra under the commutator
product. We now see that this is a special case of the more general class of algebras closed
under the product (5.130). A subject for future research will be to use (5.130) to extend
the techniques of Chapter 3 to include super-Lie algebras.
Returning to the system de
ned by the Lagrangian (5.84), we can now derive the
equations of motion from the Poisson bracket as follows,
e_i = fei H gPB
= h(i
+)
= ei
!: (5.133)
It is intersting to note that, in the case where h = I , time derivatives are determined
by (one-half) the commutator with the (bivector) Hamiltonian. This suggests an inter-
esting comparison with quantum mechanics, which has been developed in more detail
elsewhere 1].
119
 Similarly, some conservation laws can be derived, for example
fEn H gPB = hhi+i3 = 0 (5:134)
and
f! H gPB = hh++i2 = 0 (5:135)
show that En and ! are conserved respectively. The bracket (5.130) gives zero for any
scalar-valued functions, however, so is no help in deriving conservation of eiei. Further-
more, the bracket only gives the correct equations of motion for the feig frame, since
these are the genuine dynamical variables.
This concludes our discussion of pseudoclassical mechanics and multivector Lagran-
gians in general. Multivector Lagrangians have been shown to possess the capability to
package up large numbers of variables in a single action principle, and it is to be hoped
that further, more interesting applications can be found. Elsewhere 1], the concept of a
bivector-valued action has been used to give a new formulation of the path integral for
pseudoclassical mechanics. The path integrals considered involved genuine Riemann inte-
grals in parameter space, though it has not yet proved possible to extend these integrals
beyond two dimensions.

120
Chapter 6
Field Theory
We now extend the multivector derivative formalism of Chapter 5 to encompass
eld
theory. The multivector derivative is seen to provide great formal clarity by allowing
spinors and tensors to be treated in a uni
ed way. The relevant form of Noether's theorem
is derived and is used to
nd new conjugate currents in Dirac theory. The computational
advantages of the multivector derivative formalism are further exempli
ed by derivations
of the stress-energy and angular-momentum tensors for Maxwell and coupled Maxwell-
Dirac theory. This approach provides a clear understanding of the role of antisymmetric
terms in the stress-energy tensor, and the relation of these terms to spin. This chapter
concludes with a discussion of how the formalism of multivector calculus is extended to
incorporate dierentiation with respect to a multilinear function. The results derived
in this section are crucial to the development of an STA-based theory of gravity, given
in Chapter 7. Many of the results obtained in this chapter appeared in the paper \A
multivector derivative approach to Lagrangian eld theory" 7].
Some additional notation is useful for expressions involving the vector derivative r.
The left equivalent of r is written as r and acts on multivectors to its immediateleft.
(It is not always necessary to use r, as the overdot notation can be used to write A r as
_Ar_ .) The operator r
$
acts both to its left and right, and is taken as acting on everything
within a given expression, for example
A r B = A_ r_ B + Ar_ B: _
$
(6:1)
Transformations of spacetime position are written as
x0 = f (x): (6:2)
The dierential of this is the linear function
f (a) = a
rf (x) = f x(a) (6:3)
where the subscript labels the position dependence. A useful result for vector derivatives
is that
rx = @aa
rx
= @a(a
rxx0)
rx 0

= @af (a)
rx 0

= f x(rx ):
0 (6.4)
121
6.1 The Field Equations and Noether's Theorem
In what follows, we restrict attention to the application of multivector calculus to rela-
tivistic
eld theory. The results are easily extended to include the non-relativistic case.
Furthermore, we are only concerned with scalar-valued Lagrangian densities. It has not
yet proved possible to construct a multivector-valued
eld Lagrangian with interesting
properties.
We start with a scalar-valued Lagrangian density
L = L(i  a
ri) (6:5)
where fig are a set of multivector
elds. The Lagrangian (6.5) is a functional of i and
the directional derivatives of i. In many cases it is possible to write L as a functional of
 and r, and this approach was adopted in 7]. Our main application will be to gravity,
however, and there we need the more general form of (6.5).
The action is de
ned as Z
S = jd4xjL (6:6)
where jd4xj is the invariant measure. Proceeding as in Chapter 5, we write
i(x) = i0(x) + i(x) (6:7)
where i contains the same grades as i, and i0 is the extremal path. Dierentiating,
and using the chain rule, we
nd that
Z
@ S = jd4xj (@ i)  @i L + (@ i )  @i
L]
Z
= jd4xj i  @i L + (i )  @i
L]: (6.8)
Here, a
xed frame f g has been introduced, determining a set of coordinates x  
x.
The derivative of i with respect to x is denoted as i . The multivector derivative
@i
is de
ned in the same way as @i . The frame can be eliminated in favour of the
multivector derivative by de
ning
@i
a  a @i
 (6:9)
where a = 
a, and writing
Z
@ S = jd4xj i  @i L + (@a
ri)  @i
a L]: (6:10)
It is now possible to perform all manipulations without introducing a frame. This ensures
that Lorentz invariance is manifest throughout the derivation of the
eld equations.
Assuming that the boundary term vanishes, we obtain
Z
@ S = jd4xj i  @i L ; @a
r(@i
a L)]: (6:11)

122
Setting  = 0, so that the i takes their extremal values, we
nd that the extremal path
is de
ned by the solutions of the Euler-Lagrange equations
@i L ; @a
r(@i
a L) = 0: (6:12)
The multivector derivative allows for vectors, tensors and spinor variables to be handled
in a single equation | a useful and powerful uni
cation.
Noether's theorem for
eld Lagrangians is also be derived in the same manner as in
Chapter 5. We begin by considering a general multivector-parameterised transformation,
i0 = f (i M ) (6:13)
where f and M are position-independent functions and multivectors respectively. With
L0  L(i0 a
ri0), we
nd that
A  @M L0 = f M (i A)  @i L0 + f M (@a
ri A)  @i
a L0
0 0

= r
@af M (i A)  @i
a L0] + f M (i A)  @iL0 ; @a
r(@i
a L0)]: (6.14)

0 0 0

If we now assume that the i0 satisfy the same

eld equations as the i (which must again
be veri
ed) then we
nd that
@M L0 = @Ar
@af M (i A)  @i
a L0]:
0 (6:15)
This is a very general result, applying even when i0 is evaluated at a dierent spacetime
point from i,
i0(x) = f i(h(x)) M ]: (6:16)
By restricting attention to a scalar-parameterised transformation, we can write

@L0j = r
@b(@0)  @ L] 
=0 i i
b =0
(6:17)
which holds provided that the i satisfy the
eld equations (6.12) and the transformation
is such that i0( = 0) = i. Equation (6.17) turns out, in practice, to be the most useful
form of Noether's theorem.
From (6.17) we de
ne the conjugate current
j = @b ( @i0j=0)  @i
b L: (6:18)
If L0 is independent of , j satis
es the conservation equation
r
j = 0: (6:19)
An inertial frame relative to the constant time-like velocity 0 then sees the charge
Z
Q = jd3xjj
0 (6:20)
as conserved with respect to its local time.

123
6.2 Spacetime Transformations and their Conjugate
Tensors
In this section we use Noether's theorem in the form (6.17) to analyse the consequences of
Poincare and conformal invariance. These symmetries lead to the identi
cation of stress-
energy and angular-momentum tensors, which are used in the applications to follow.

1. Translations
A translation of the spacetime
elds i is achieved by de
ning new spacetime
elds i0 by
i0(x) = i(x0) (6:21)
where
x0 = x + n: (6:22)
Assuming that L is only x-dependent through the
elds i(x), equation (6.17) gives
0

n
rL = r
@a(n
ri)  @i
a L] (6:23)

and from this we de
ne the canonical stress-energy tensor by
T (n) = @a(n
ri)  @i
a L ; nL: (6:24)
The function T (n) is linear on n and, from (6.23), T (n) satis
es
r
T (n) = 0: (6:25)
To write down a conserved quantity from T (n) it is useful to
rst construct the adjoint
function
T (n) = @bhnT (b)i
= @bhn
@a(b
ri)  @i
a L ; n
bLi
= rh _ _i@i
n Li ; nL: (6.26)
This satis
es the conservation equation
T_ (r_ ) = 0 (6:27)
which is the adjoint to (6.25). In the 0 frame the
eld momentum is de
ned by
Z
p = jd3xj T (0) (6:28)
and, provided that the
elds all fall o suitably at large distances, the momentum p is
conserved with respect to local time. This follows from
Z
0
rp = jd3xj T_ (00
r_ )
Z
= ; jd3xj T_ (00 ^ r_ )
= 0: (6.29)
The total
eld energy, as seen in the 0 frame, is
Z
E = jd3xj0
T (0): (6:30)

124
2. Rotations
If we assume initially that all
elds i transform as spacetime vectors, then a rotation of
these
elds from one spacetime point to another is performed by
i0(x) = eB=2i(x0)e;B=2 (6:31)
where
x0 = e;B=2xeB=2: (6:32)
This diers from the point-particle transformation law (5.53) in the relative directions of
the rotations for the position vector x and the
elds i. The result of this dierence is a
change in the relative sign of the spin contribution to the total angular momentum. In
order to apply Noether's theorem (6.17), we use
@i0j=0 = B  i ; (B
x)
ri (6:33)
and
@L0j=0 = ;(B
x)
rL = r
(x
B L): (6:34)
Together, these yield the conjugate vector
J (B ) = @aB  i ; (B
x)
ri]  @i
a L + B
xL (6:35)
which satis
es
r
J (B ) = 0: (6:36)
The adjoint to the conservation equation (6.36) is
J_ (r_ )
B = 0 for all B
) J_ (r_ ) = 0: (6.37)
The adjoint function J (n) is a position-dependent bivector-valued linear function of the
vector n. We identify this as the canonical angular-momentum tensor. A conserved
bivector in the 0-system is constructed in the same manner as for T (n) (6.28). The
calculation of the adjoint function J (n) is performed as follows:
J (n) = @B hJ (B )ni
= @B h(B  i ; B
(x ^r)i)  @i
n L + B
xLni
= ;x ^ r_ _ i  @i
n L ; nL] + hi  @i
n Li2
= T (n) ^ x + hi  @i
n Li2: (6.38)
If one of the
elds , say, transforms single-sidedly (as a spinor), then J (n) contains the
term h 21 @
n Li2 .
The
rst term in J (n) (6.38) is the routine p ^ x component, and the second term is
due to the spin of the
eld. The general form of J (n) is therefore
J (n) = T (n) ^ x + S (n): (6:39)
125
By applying (6.37) to (6.39) and using (6.27), we
nd that
T (r_ ) ^ x_ + S_ (r_ ) = 0: (6:40)
The
rst term in (6.40) returns (twice) the characteristic bivector of T (n). Since the
antisymmetric part of T (n) can be written in terms of the characteristic bivector B as
T ; (a) = 21 B
a (6:41)
equation (6.40) becomes
B = ;S_ (r_ ): (6:42)
It follows that, in any Poincare-invariant theory, the antisymmetric part of the stress-
energy tensor is a total divergence. But, whilst T; (n) is a total divergence, x ^ T;(n)
certainly is not. So, in order for (6.37) to hold, the antisymmetric part of T (n) must be
retained since it cancels the divergence of the spin term.

3. Dilations
While all fundamental theories should be Poincare-invariant, an interesting class go be-
yond this and are invariant under conformal transformations. The conformal group con-
tains two further symmetry transformations, dilations and special conformal transfor-
mations. Dilations are considered here, and the results below are appropriate to any
scale-invariant theory.
A dilation of the spacetime
elds is achieved by de
ning
i0(x) = edii(x0) (6:43)
where
x0 = ex (6:44)
) ri(x) = e
0 (d i +1)
0
 rx i(x ):
0
(6:45)
If the theory is scale-invariant, it is possible to assign the \conformal weights" di in such
a way that the left-hand side of (6.17) becomes
@L0j=0 = r
(xL): (6:46)
In this case, equation (6.17) takes the form
r
(xL) = r
@a(di i + x
ri)  @i
a L] (6:47)
from which the conserved current
j = di@ai  @i
a L + T (x) (6:48)
is read o. Conservation of j (6.48) implies that
r
T (x) = @aT (a) = ;r
(di@ai  @i
a L) (6:49)
so, in a scale-invariant theory, the trace of the canonical stress-energy tensor is a total
divergence. By using the equations of motion, equation (6.49) can be written, in four
dimensions, as
di hi@i Li + (di + 1)(@a
ri)  @i
a L = 4L (6:50)
which can be taken as an alternative de
nition for a scale-invariant theory.
126
4. Inversions
The remaining generator of the conformal group is inversion,
x0 = x;1: (6:51)
As it stands, this is not parameterised by a scalar and so cannot be applied to (6.17). In
order to derive a conserved tensor, the inversion (6.51) is combined with a translation to
de
ne the special conformal transformation 77]
x0 = h(x)  (x;1 + n);1 = x(1 + nx);1: (6:52)
From this de
nition, the dierential of h(x) is given by
h(a) = a
rh(x) = (1 + xn);1a(1 + nx);1 (6:53)
so that h de
nes a spacetime-dependent rotation/dilation. It follows that h satis
es
h(a)
h(b) = (x)a
b (6:54)
where
(x) = (1 + 2n
x + 2x2n2);2: (6:55)
That the function h(a) satis
es equation (6.54) demonstrates that it belongs to the con-
formal group.
The form of h(a) (6.53) is used to postulate transformation laws for all
elds (including
spinors, which transform single-sidedly) such that
L0 = (det h)L(i (x0) h(a)
rx i(x0))
0 (6:56)
which implies that
@L0j=0 = @ det hj=0 L + ( @x0j=0)
rL: (6:57)
Since
det h = (1 + 2n
x + 2x2n2);4 (6:58)
it follows that
@ det hj=0 = ;8x
n: (6:59)
We also
nd that
@x0j=0 = ;(xnx) (6:60)
and these results combine to give
@L0j=0 = ;8x
nL ; (xnx)
rL = ;r
(xnxL): (6:61)
Special conformal transformations therefore lead to a conserved tensor of the form
T SC (n) = @ah(;(xnx)
ri + @i0(x))  @i
a L + xnxLi=0
= ;T (xnx) + @ah(@i0(x))  @i
a Li=0 : (6.62)
127
The essential quantity here is the vector ;xnx, which is obtained by taking the constant
vector n and reecting it in the hyperplane perpendicular to the chord joining the point
where n is evaluated to the origin. The resultant vector is then scaled by a factor of x2.
In a conformally-invariant theory, both the antisymmetric part of T (n) and its trace
are total divergences. These can therefore be removed to leave a new tensor T 0(n) which
is symmetric and traceless. The complete set of divergenceless tensors is then given by
fT 0(x), T 0(n), xT 0(n)x, J 0(n)  T 0(n) ^ xg (6:63)
This yields a set of 1 + 4 + 4 + 6 = 15 conserved quantities | the dimension of the
conformal group. All this is well known, of course, but it is the
rst time that geometric
algebra has been systematically applied to this problem. It is therefore instructive to
see how geometric algebra is able to simplify many of the derivations, and to generate a
clearer understanding of the results.

6.3 Applications
We now look at a number of applications of the formalism established in the preceding
sections. We start by illustrating the techniques with the example of electromagnetism.
This does not produce any new results, but does lead directly to the STA forms of the
Maxwell equations and the electromagnetic stress-energy tensor. We next consider Dirac
theory, and a number of new conjugate currents will be identi
ed. A study of coupled
Maxwell-Dirac theory then provides a useful analogue for the discussion of certain aspects
of a gauge theory of gravity, as described in Chapter 7. The
nal application is to a two-
particle action which recovers the
eld equations discussed in Section 4.4.
The essential result needed for what follows is
@
a hb
rM i = a @
h(b

M i
= a
bP(M ) (6.64)
where P (M ) is the projection of M onto the grades contained in . It is the result (6.64)
that enables all calculations to be performed without the introduction of a frame. It
is often the case that the Lagrangian can be written in the form L(i ri), when the
following result is useful:
@
a hrM i = @
a hb
rM@bi
= a
bP (M@b)
= P (Ma): (6.65)

1. Electromagnetism
The electromagnetic Lagrangian density is given by
L = ;A
J + 21 F
F (6:66)

128
where A is the vector potential, F = r^A, and A couples to an external current J which
is not varied. To
nd the equations of motion we
rst write F
F as a function of rA,
F
F = 41 h(rA ; (rA)~)2i
= 12 hrArA ; rA(rA)~i: (6.67)
The
eld equations therefore take the form
;J ; @b
r 21 hrAb ; (rA)~bi1 = 0
) ;J ; @b
rF
b = 0
) r
F = J: (6.68)
This is combined with the identity r^F = 0 to yield the full set of Maxwell's equations,
rF = J .
To calculate the free-
eld stress-energy tensor, we set J = 0 in (6.66) and work with
L0 = 12 hF 2i: (6:69)
Equation (6.26) now gives the stress-energy tensor in the form
_ AF
T (n) = rh _
ni ; 21 nhF 2i: (6:70)
This expression is physically unsatisfactory as is stands, because it is not gauge-invariant.
In order to
nd a gauge-invariant form of (6.70), we write 60]
_ AF
rh _
ni = (r^ A)
(F
n) + (F
n)
rA
= F
(F
n) ; (F
r_ )
nA_ (6.71)
and observe that, since r
F = 0, the second term is a total divergence and can therefore
be ignored. What remains is
T em(n) = F
(F
n) ; 21 nF
F
= 21 FnF ~ (6.72)
which is the form of the electromagnetic stress-energy tensor obtained by Hestenes 17].
The tensor (6.72) is gauge-invariant, traceless and symmetric. The latter two properties
follow simultaneously from the identity
@aT em(a) = @a 21 FaF~ = 0: (6:73)
The angular momentum is obtained from (6.38), which yields
J (n) = (rh _ AFn
_ i ; 21 nhF 2i) ^ x + A ^ (F
n) (6:74)
where we have used the stress-energy tensor in the form (6.70). This expression suers
from the same lack of gauge invariance, and is
xed up in the same way, using (6.71) and
; (F
n) ^ A + x ^ (F
r_ )
nA_ ] = x ^ (F
r
$
)
nA] (6:75)
129
which is a total divergence. This leaves simply
J (n) = T em(n) ^ x: (6:76)
By rede
ning the stress-energy tensor to be symmetric, the spin contribution to the
angular momentum is absorbed into (6.72). For the case of electromagnetism this has the
advantage that gauge invariance is manifest, but it also suppresses the spin-1 nature of
the
eld. Suppressing the spin term in this manner is not always desirable, as we shall
see with the Dirac equation.
The free-
eld Lagrangian (6.69) is not only Poincare-invariant! it is invariant under
the full conformal group of spacetime 7, 77]. The full set of divergenceless tensors for
free-
eld electromagnetism is therefore T em(x), T em(n), xT em(n)x, and T em(n) ^ x. It is a
simple matter to calculate the modi
ed conservation equations when a current is present.

2. Dirac Theory1
The multivector derivative is particularly powerful when applied to the STA form of the
Dirac Lagrangian. We recall from Chapter 5 that the Lagrangian for the Dirac equation
can be written as (4.96)
L = hri3~ ; eA0~ ; m~i (6:77)
where  is an even multivector and A is an external gauge
eld (which is not varied). To
verify that (6.77) does give the Dirac equation we use the Euler-Lagrange equations in
the form
@ L = @a
r(@
a L) (6:78)
to obtain
(ri3)~ ; 2e0A~ ; 2m~ = @a
r(i3a ~)
= i3~ r :

(6.79)
Reversing this equation, and postmultiplying by 0, we obtain
ri3 ; eA = m0 (6:80)
as found in Chapter 4 (4.92). Again, it is worth stressing that this derivation employs a
genuine calculus, and does not resort to treating  and ~ as independent variables.
We now analyse the Dirac equation from the viewpoint of the Lagrangian (6.77).
In this Section we only consider position-independent transformations of the spinor .
Spacetime transformations are studied in the following section. The transformations we
are interested in are of the type
0 = eM  (6:81)
where M is a general multivector and  and M are independent of position. Operations
on the right of  arise naturally in the STA formulation of Dirac theory, and can be
1 The basic idea developed in this section was provided by Anthony Lasenby.

130
thought of as generalised gauge transformations. They have no such simple analogue in
the standard column-spinor treatment. Applying (6.17) to (6.81), we obtain
r
hMi3~i1 = @L0j=0  (6:82)
which is a result that we shall exploit by substituting various quantities for M . If M is
odd both sides vanish identically, so useful information is only obtained when M is even.
The
rst even M to consider is a scalar, , so that hMi3~i1 is zero. It follows that


@ e2L =0 = 0
) L = 0 (6.83)
and hence that, when the equations of motion are satis
ed, the Dirac Lagrangian vanishes.
Next, setting M = i, equation (6.82) gives

;r
(s) = ;m@he2iei
i=0 
) r
(s) = ;2m sin
 (6.84)
where s = 3~ is the spin current. This equation is well-known 33], though it is not
usually observed that the spin current is the current conjugate to duality rotations. In
conventional versions, these would be called \axial rotations", with the role of i taken by
5. In the STA approach, however, these rotations are identical to duality transformations
for the electromagnetic
eld. The duality transformation generated by ei is also the
continuous analogue of the discrete symmetry of mass conjugation, since  7! i changes
the sign of the mass term in L. It is no surprise, therefore, that the conjugate current,
s, is conserved for massless particles.
Finally, taking M to be an arbitrary bivector B yields
r
(B
(i3)~) = D2hriB
3~ ; eAB
E0~i
= eA(3B3 ; B )0~  (6.85)
where the Dirac equation (6.80) has beed used. Both sides of(6.85) vanish for B = i1 i2
and 3, with useful equations arising on taking B = 1 2 and i3. The last of these,
B = i3, corresponds to the usual U (1) gauge transformation of the spinor
eld, and gives
r
J = 0 (6:86)
where J = 0~ is the current conjugate to phase transformations, and is strictly con-
served. The remaining transformations generated by e 1 and e 2 give
r
(e1) = 2eA
e2 (6:87)
r
(e2) = ;2eA
e1
where e =  ~. Although these equations have been found before 33], the role of
e1 and e2 as currents conjugate to right-sided e 2 and e 1 transformations has not
been noted. Right multiplication by 1 and 2 generates charge conjugation, since the
transformation  7! 0  1 takes (6.80) into
r0i3 + eA0 = m00 : (6:88)
It follows that the conjugate currents are conserved exactly if the external potential van-
ishes, or the particle has zero charge.
131
3. Spacetime Transformations in Maxwell-Dirac Theory
The canonical stress-energy and angular-momentum tensors are derived from spacetime
symmetries. In considering these it is useful to work with the full coupled Maxwell-Dirac
Lagrangian, in which the free-
eld term for the electromagnetic
eld is also included. This
ensures that the Lagrangian is Poincare-invariant. The full Lagrangian is therefore
L = hri3~ ; eA0~ ; m~ + 21 F 2i (6:89)
in which both  and A are dynamical variables.
From the de
nition of the stress-energy tensor (6.26) and the fact that the Dirac part
of the Lagrangian vanishes when the
eld equations are satis
ed (6.83), T (n) is given by
T (n) = rh _ 3n
_ i ~ i + rh
_ AFn
_ i ; 21 nF
F: (6:90)
Again, this is not gauge-invariant and a total divergence must be removed to recover a
gauge-invariant tensor. The manipulations are as at (6.71), and now yield
T md(n) = rh _ 3n
_ i ~ i ; n
JA + 21 FnF
~ (6:91)
where J = 0~. The tensor (6.91) is now gauge-invariant, and conservation can be
checked simply from the
eld equations. The
rst and last terms are the free-
eld stress-
energy tensors and the middle term, ;n
JA, arises from the coupling. The stress-energy
tensor for the Dirac theory in the presence of an external
eld A is conventionally de
ned
by the
rst two terms of (6.91), since the combination of these is gauge-invariant.
Only the free-
eld electromagnetic contribution to T md (6.91) is symmetric! the other
terms each contain antisymmetric parts. The overall antisymmetric contribution is
T;(n) = ;
1 T (n) T (n)]
2 md md
=
^ ; r^h
1 n A J
2
_ _ 3~ 1]
i i
=
h ; r
1 n AJ
2 i3~ + _ i
rh_ 3~ 3 2
ii
=
r

n ( ( 41 is))
=
; r^
n ( i ( 41 s)) (6.92)
and is therefore completely determined by the exterior derivative of the spin current 78].
The angular momentum is found from (6.39) and, once the total divergence is removed,
the gauge-invariant form is
J (n) = T md(n) ^ x + 12 is ^ n: (6:93)
The ease of derivation of J (n) (6.93) compares favourably with traditional operator-based
approaches 60]. It is crucial to the identi
cation of the spin contribution to the angular
momentum that the antisymmetric component of T md(n) is retained. In (6.93) the spin
term is determined by the trivector is, and the fact that this trivector can be dualised to
the vector s is a unique property of four-dimensional spacetime.

132
 The sole term breaking conformal invariance in (6.89) is the mass term hm~i, and
it is useful to consider the currents conjugate to dilations and special conformal transfor-
mations, and to see how their non-conservation arises from this term. For dilations, the
conformal weight of a spinor
eld is 32 , and equation (6.48) yields the current
jd = T md(x) (6:94)
(after subtracting out a total divergence). The conservation equation is
r
jd = @a
T md(a)
= hm~i: (6.95)
Under a spacetime transformation the A
eld transforms as
A(x) 7! A0(x)  f A(x0)] (6:96)
where x0 = f (x). For a special conformal transformation, therefore, we have that
A0(x) = (1 + nx);1 A(x0)(1 + xn);1: (6:97)
Since this is a rotation/dilation, we postulate for  the single-sided transformation
0(x) = (1 + nx);2 (1 + xn);1(x0): (6:98)
In order to verify that the condition (6.56) can be satis
ed, we need the neat result that


r (1 + nx);2 (1 + xn);1 = 0: (6:99)
This holds for all vectors n, and the bracketed term is immediately a solution of the
massless Dirac equation (it is a monogenic function on spacetime). It follows from (6.56)
that the conserved tensor is
T SC (n) = ;T md(xnx) ; n
(ix ^ (s)): (6:100)
and the conservation equation is
r
T SC (xnx) = ;2hm~in
x: (6:101)
In both (6.95) and (6.101) the conjugate tensors are conserved as the mass goes to zero,
as expected.

4. The Two-Particle Dirac Action

Our
nal application is to see how the two-particle equation (4.147) can be obtained from
an action integral. The Lagrangian is
r1 + r2 )J ( 1 +  2)~ ; 2~i
L = h( m (6:102)
1 m2 0 0

133
where  is a function of position in the 8-dimensional con
guration space, and r1 and
r2 are the vector derivatives in their respective spaces. The action is
Z
S = jd8xjL: (6:103)
If we de
ne the function h by
h(a) = i1 am
i1 + i2 am
i2 
1 2
(6:104)
where i1 and i2 are the pseudoscalars for particle-one and particle-two spaces respectively,
then we can write the action as
Z
S = jd8xjhh(@b)b
rJ (01 + 02)~ ; 2~i: (6:105)
Here r = r1 + r2 is the vector derivative in 8-dimensional con
guration space. The
eld
equation is
@ L = @a
r(@
a L) (6:106)
which leads to
h(@a)a
rJ (01 + 02)]~ ; 4~ = @a
rJ (01 + 02)h
~ (a)]: (6:107)
The reverse of this equation is
h(@a)a
rJ (01 + 02) = 2 (6:108)
and post-multiplying by (01 + 02) obtains
r + r )J = ( 1 +  2)
(m
1 2
(6:109)
1 m2 0 0
as used in Section 4.4.
The action (6.102) is invariant under phase rotations in two-particle space,
 7! 0  e;J  (6:110)
and the conserved current conjugate to this symmetry is
j = @a(;J )  @
aL
~ (a)i
= @ahE (01 + 02)h
= hhE (01 + 02)~i1: (6.111)
This current satis
es the conservation equation
r
j = 0 (6:112)
or, absorbing the factor of E into ,
r + r )
h( 1 +  2)~i = 0:
(m
1 2
(6:113)
1 m2 0 0 1

Some properties of this current were discussed briey in Section 4.4.

134
6.4 Multivector Techniques for Functional Dieren-
tiation
We have seen how the multivector derivative provides a natural and powerful extension
to the calculus of single-variable functions. We now wish to extend these techniques to
encompass the derivative with respect to a linear function h(a). We start by introducing
a
xed frame feig, and de
ne the scalar coecients
hij  h(ei)
ej : (6:114)
Each of the scalars hij can be dierentiated with respect to, and we seek a method of
combining these into a single multivector operator. If we consider the derivative with
respect to hij on the scalar h(b)
c, we
nd that
@hij h(b)
c = @hij (bk clhkl)
= bicj : (6.115)
If we now multiply both sides of this expression by a
eiej we obtain
a
eiej @hij h(b)
c = a
bc: (6:116)
This successfully assembles a frame-independent vector on the right-hand side, so the op-
erator a
eiej @hij must also be frame-independent. We therefore de
ne the vector functional
derivative @h(a) by
@h(a)  a
eiej @hij  (6:117)
where all indices are summed over and hij is given by (6.114).
The essential property of @h(a) is, from (6.116),
@h(a)h(b)
c = a
bc (6:118)
and this result, together with Leibniz' rule, is sucient to derive all the required properties
of the @h(a) operator. The procedure is as follows. With B a
xed bivector, we write
@h(a)hh(b ^ c)B i = @_h(a)hh_ (b)h(c)B i ; @_h(a)hh_ (c)h(b)B i
= a
bh(c)
B ; a
ch(b)
B
= ha
(b ^ c)]
B (6.119)
which extends, by linearity, to give
@h(a)hh(A)B i = h(a
A)
B (6:120)
where A and B are both bivectors. Proceeding in this manner, we
nd the general formula
X
@h(a)hh(A)B i = hh(a
Ar)Br i1: (6:121)
r

135
For a
xed grade-r multivector Ar, we can now write
@h(a)h(Ar) = @h(a)hh(Ar)Xr i@Xr
= hh(a
Ar)Xr i1@Xr
= (n ; r + 1)h(a
Ar) (6.122)
where n is the dimension of the space and a result from page 58 of 24] has been used.
Equation (6.121) can be used to derive formulae for the functional derivative of the
adjoint. The general result is
@h(a)h(Ar ) = @h(a)hh(Xr )Ar i@Xr
= hh(a
X_ r )Ari1@_Xr : (6.123)
When A is a vector, this admits the simpler form
@h(a)h(b) = ba: (6:124)
If h is a symmetric function then h = h, but this cannot be exploited for functional
dierentiation, since h and h are independent for the purposes of calculus.
As two
nal applications, we derive results for the functional derivative of the deter-
minant (1.115) and the inverse function (1.125). For the determinant, we
nd that
@h(a)h(I ) = h(a
I )
) @h(a) det(h) = h(a
I )I ;1
= det(h)h;1(a) (6.125)
where we have recalled the de
nition of the inverse (1.125). This coincides with standard
formulae for the functional derivative of the determinant by its corresponding tensor. The
proof given here, which follows directly from the de
nitions of the determinant and the
inverse, is considerably more concise than any available to conventional matrix/tensor
methods. The result for the inverse function is found from
;1
@h(a)hh(Br )h;1(Ar )i = hh(a
Br)h;1(Ar )i1 + @_h(a)hh_ (Ar)h(Br )i = 0 (6:126)
from which it follows that
@h(a)hh;1(Ar )Br i = ;hha
h;1(Br )]h;1(Ar)i1
= ;hh;1 (a)
Brh;1(Ar)i1 (6.127)
where use has been made of results for the adjoint (1.123).
We have now assembled most of the necessary formalism and results for the application
of geometric calculus to
eld theory. In the
nal chapter we apply this formalism to
develop a gauge theory of gravity.

136
Chapter 7
Gravity as a Gauge Theory
In this chapter the formalism described throughout the earlier chapters of this thesis is
employed to develop a gauge theory of gravity. Our starting point is the Dirac action, and
we begin by recalling how electromagnetic interactions arise through right-sided transfor-
mations of the spinor
eld . We then turn to a discussion of Poincare invariance, and
attempt to introduce gravitational interactions in a way that closely mirrors the intro-
duction of the electromagnetic sector. The new dynamical
elds are identi
ed, and an
action is constructed for these. The
eld equations are then found and the derivation
of these is shown to introduce an important consistency requirement. In order that the
correct minimally-coupled Dirac equation is obtained, one is forced to use the simplest
action for the gravitational
elds | the Ricci scalar. Some free-
eld solutions are ob-
tained and are compared with those of general relativity. Aspects of the manner in which
the theory employs only active transformations are then illustrated with a discussion of
extended-matter distributions.
By treating gravity as a gauge theory of active transformations in the (at) spacetime
algebra, some important dierences from general relativity emerge. Firstly, coordinates
are unnecessary and play an entirely secondary role. Points are represented by vectors,
and all formulae given are coordinate-free. The result is a theory in which spacetime does
not play an active role, and it is meaningless to assign physical properties to spacetime.
The theory is one of forces, not geometry. Secondly, the gauge-theory approach leads
to a
rst-order set of equations. Despite the fact that the introduction of a set of coor-
dinates reproduces the matter-free
eld equations of general relativity, the requirement
that the
rst-order variables should exist globally has important consequences. These are
illustrated by a discussion of point-source solutions.
There has, of course, been a considerable discussion of whether and how gravity can
be formulated as a gauge theory. The earliest attempt was by Utiyama 79], and his
ideas were later re
ned by Kibble 80]. This led to the development of what is now
known as the Einstein-Cartan-Kibble-Sciama (ECKS) theory of gravity. A detailed review
of this subject was given in 1976 by Hehl et al. 81]. More recently, the
bre-bundle
approach to gauge theories has been used to study general relativity 82]. All these
developments share the idea that, at its most fundamental level, gravity is the result
of spacetime curvature (and, more generally, of torsion). Furthermore, many of these
treatments rely on an uncomfortable mixture of passive coordinate transformations and

137
active tetrad transformations. Even when active transformations are emphasised, as by
Hehl et al., the transformations are still viewed as taking place on an initially curved
spacetime manifold. Such ideas are rejected here, as is inevitable if one only discusses the
properties of spacetime
elds, and the interactions between them.

7.1 Gauge Theories and Gravity

We prepare for a discussion of gravity by
rst studying how electromagnetism is introduced
into the Dirac equation. We start with the Dirac action in the form
Z
SD = jd4xjhri3~ ; m~i (7:1)
and recall that, on de
ning the transformation
 7! 0  e;i 3 (7:2)
the action is the same whether viewed as a function of  or 0. This is global phase
invariance. The transformation (7.2) is a special case of the more general transformation
0 = R~ 0 (7:3)
where R0 is a constant rotor. We saw in Chapter 4 that a Dirac spinor encodes an
instruction to rotate the f g frame onto a frame of observables fe g. 0 is then the
spinor that generates the same observables from the rotated initial frame
 0 = R0 R~0: (7:4)
It is easily seen that (7.3) and (7.4) together form a symmetry of the action, since
hr0i30 ~0 ; m0~0i = hrR~ 0iR03R~0R0~ ; mR~ 0R0~i
= hri3~ ; m~i: (7.5)
The phase rotation (7.2) is singled out by the property that it leaves both the time-like
0-axis and space-like 3-axis unchanged.
There is a considerable advantage in introducing electromagnetic interactions through
transformations such as (7.3). When we come to consider spacetime rotations, we will
nd
that the only dierence is that the rotor R0 multiplies  from the left instead of from the
right. We can therefore use an identical formalism for introducing both electromagnetic
and gravitational interactions.
Following the standard ideas about gauge symmetries, we now ask what happens if
the phase  in (7.2) becomes a function of position. To make the comparison with gravity
as clear as possible, we write the transformation as (7.3) and only restrict R~0 to be of
the form of (7.2) once the gauging has been carried out. To study the eect of the
transformation (7.3) we
rst write the vector derivative of  as
r = @a(a
r) (7:6)
138
which contains a coordinate-free contraction of the directional derivatives of  with their
vector directions. Under the transformation (7.3) the directional derivative becomes
a
r0 = a
r(R~ 0)
= a
rR~ 0 + a
rR~ 0
= a
rR~ 0 ; 21 R~0(a) (7.7)
where
(a)  ;2R0a
rR~ 0: (7:8)
From the discussion of Lie groups in Section (3.1), it follows that (a) belongs to the
Lie algebra of the rotor group, and so is a (position-dependent) bivector-valued linear
function of the vector a.
It is now clear that, to maintain local invariance, we must replace the directional
derivatives a
r by covariant derivatives Da, where
Da  a
r + 21 +(a): (7:9)
We are thus forced to introduce a new set of dynamical variables | the bivector
eld
+(a). The transformation properties of +(a) must be the same as those of (a). To
nd
these, replace 0 by  and consider the new transformation
 7! 0 = R~ (7:10)
so that the rotor R~ 0 is transformed to R~ 0R~ = (RR0)~. The bivector (a) then transforms
to
0(a) = ;2(RR0 )a
r(RR0)~
= R(a)R~ ; 2Ra
rR: ~ (7.11)
It follows that the transformation law for +(a) is
+(a) 7! +0(a)  R+(a)R~ ; 2Ra
rR ~ (7:12)
which ensures that
Da0(0) = a
r(R~ ) + 21 R~ +0(a)
= a
r(R~ ) + 12 R~ (R+(a)R~ ; 2Ra
rR~)
= a
rR~ + 21 +(a)R~
= Da()R: ~ (7.13)
The action integral (7.1) is now modi
ed to
Z
S = jd4xjh@aDai3~ ; m~i
0
(7:14)
from which the
eld equations are
ri3 + 21 @a+(a)
(i3) = m: (7:15)
139
For the case of electromagnetism, the rotor R is restricted to the form of (7.2), so the
most general form that +(a) can take is
+(a) = 2a
(eA)i3: (7:16)
The \minimally-coupled" equation is now
ri3 ; eA0 = m (7:17)
recovering the correct form of the Dirac equation in the presence of an external A
eld.

7.1.1 Local Poincare Invariance

Our starting point for the introduction of gravity as a gauge theory is the Dirac action
(7.1), for which we study the eect of local Poincare transformations of the spacetime

elds. We
rst consider translations
(x) 7! 0(x)  (x0) (7:18)
where
x0 = x + a (7:19)
and a is a constant vector. To make these translations local, the vector a must become a
function of position. This is achieved by replacing (7.19) with
x0 = f (x) (7:20)
where f (x) is now an arbitrary mapping between spacetime positions. We continue to
refer to (7.20) as a translation, as this avoids overuse of the word \transformation". It
is implicit in what follows that all translations are local and are therefore determined by
completely arbitrary mappings. The translation (7.20) has the interpretation that the

eld  has been physically moved from the old position x0 to the new position x. The
same holds for the observables formed from , for example the current J (x) = 0~ is
transformed to J 0(x) = J (x0).
As it stands, the translation de
ned by (7.20) is not a symmetry of the action, since
r0(x) = r(f (x))
= f (rx )(x0)0 (7.21)
and the action becomes
Z
S 0 = jd4x0j (det f );1hf (rx )0i3~0 ; m0~0i:
0 (7:22)
To recover a symmetry from (7.20), one must introduce an arbitrary, position-dependent
linear function h. The new action is then written as
Z
Sh = jd4xj(det h);1 hh(r)i3~ ; m~i: (7:23)

140
Under the translation
(x) 7! 0(x)  (f (x)) (7:24)
the action Sh transforms to
Z
Sh = jd4x0j(det f );1(det h0);1hh0 f (rx )0i3~0 ; m0~0i
0
0 (7:25)
and the original action is recovered provided that h has the transformation law
hx 7! h0x  hx f ;x 1
0 where x0 = f (x): (7:26)
This is usually the most useful form for explicit calculations, though alternatively we can
write
hx 7! h0x  hx f x
0 0 where x = f (x0) (7:27)
which is sometimes more convenient to work with.
In arriving at (7.23) we have only taken local translations into account | the currents
are being moved from one spacetime position to another. To arrive at a gauge theory
of the full Poincare group we must consider rotations as well. (As always, the term
\rotation" is meant in the sense of Lorentz transformation.) In Chapter 6, rotational
invariance of the Dirac action (7.1) was seen by considering transformations of the type
(x) 7! R0(R~ 0xR0), where R0 is a constant rotor. By writing the action in the form
of (7.23), however, we have already allowed for the most general type of transformation
of position dependence. We can therefore translate back to x, so that the rotation takes
place at a point. In doing so, we completely decouple rotations and translations. This is
illustrated by thinking in terms of the frame of observables fe g. Given this frame at a
point x, there are two transformations that we can perform on it. We can either move
it somewhere else (keeping it in the same orientation with respect to the f g frame),
or we can rotate it at a point. These two operations correspond to dierent symmetries.
A suitable physical picture might be to think of \experiments" in place of frames. We
expect that the physics of the experiment will be unaected by moving the experiment
to another point, or by changing its orientation in space.
Active rotations of the spinor observables are driven by rotations of the spinor
eld,
 7! 0  R0: (7:28)
Since h(a) is a spacetime vector
eld, the corresponding law for h must be
h(a) 7! h0(a)  R0h(a)R~ 0: (7.29)
By writing the action (7.23) in the form
Z
Sh = jd4xj(det h);1 hh(@a)a
ri3~ ; m~i (7:30)
we observe that it is now invariant under the rotations de
ned by (7.28) and (7.29). All
rotations now take place at a point, and the presence of the h
eld ensures that a rotation

141
at a point has become a global symmetry. To make this symmetry local, we need only
replace the directional derivative of  by a covariant derivative, with the property that
Da0 (R) = RDa (7:31)
where R is a position-dependent rotor. But this is precisely the problem that was tackled
at the start of this section, the only dierence being that the rotor R now sits to the left
of . Following the same arguments, we immediately arrive at the de
nition:
Da  (a
r + 21 +(a)) (7:32)
where +(a) is a (position-dependent) bivector-valued linear function of the vector a. Under
local rotations  7! R, +(a) transforms to
+(a) 7! +(a)0  R+(a)R~ ; 2a
rRR: ~ (7:33)
Under local translations, +(a) must transform in the same manner as the a
rRR~ term,
so
+x(a) 7! +x f x(a)
0 if x0 = f (x) (7:34)
+ (a) 7! + f (a) if x = f (x0):
x x x
0
;1
0

(The subscript x on +x(a) labels the position dependence.)

The action integral
Z
S = jd4xj(det h);1 hh(@a)Dai3~ ; m~i (7:35)
is now invariant under both local translations and rotations. The
eld equations derived
from S will have the property that, if f h +g form a solution, then so too will any
new
elds obtained from these by local Poincare transformations. This local Poincare
symmetry has been achieved by the introduction of two gauge
elds, h and +, with a
total of (4  4) + (4  6) = 40 degrees of freedom. This is precisely the number expected
from gauging the 10-dimensional Poincare group.
Before turning to the construction of an action integral for the gauge
elds, we look
at how the covariant derivative of (7.32) must extend to act on the physical observables
derived from . These observables are all formed by the double-sided action of the spinor

eld  on a constant multivector ; (formed from the f g) so that

~
A  ;: (7:36)
The multivector A therefore transforms under rotations as
A 7! RAR ~ (7.37)
and under translations as
A(x) 7! A(f (x)): (7.38)
We refer to objects with the same transformation laws as A as being covariant (an example
is the current, J = 0~). If we now consider directional derivatives of A, we see that
these can be written as
a
rA = (a
r);~ + ;(a
r)~ (7:39)
142
which immediately tells us how to turn these into covariant derivatives. Rotations of A are
driven by single-sided rotations of the spinor
eld , so to arrive at a covariant derivative
of A we simply replace the spinor directional derivatives by covariant derivatives, yielding
(Da);~ + ;(Da )~
= (a
r);~ + ;(a
r)~+ 21 +(a);~ ; 12 ;~+(a)
= a
r(;~) + +(a)  (;~): (7.40)
We therefore de
ne the covariant derivative for \observables" by
DaA  a
rA + +(a)  A: (7:41)
This is applicable to all multivector
elds which transform double-sidedly under rotations.
The operator Da has the important property of satisfying Leibniz' rule,
Da(AB ) = (DaA)B + A(DaB ) (7:42)
so that Da is a derivation. This follows from the identity
+(a)  (AB ) = (+(a)  A)B + A(+(a)  B ): (7:43)
For notational convenience we de
ne the further operators
D  h(@a)Da (7.44)
DA  h(@a)DaA (7.45)
and for the latter we write
DA = D
A + D^ A (7:46)
where
D
Ar  hDAir;1 (7.47)
D^ Ar  hDAir+1 : (7.48)
The operator D can be thought of as a covariant vector derivative. D and D have the
further properties that
a
D = Dh(a) (7.49)
a
D = Dh(a): (7.50)

7.1.2 Gravitational Action and the Field Equations

Constructing a form of the Dirac action that is invariant under local Poincare transfor-
mations has required the introduction of h and +
elds, with the transformation proper-
ties (7.26), (7.29), (7.33) and (7.34). We now look for invariant scalar quantities that can
be formed from these. We start by de
ning the
eld-strength R(a ^ b) by
1 R(a
2 ^ b)  Da Db] (7:51)
143
so that
R(a ^ b) = a
r+(b) ; b
r+(a) + +(a)  +(b): (7:52)
R(a ^ b) is a bivector-valued function of its bivector argument a ^ b. For an arbitrary
bivector argument we de
ne
R(a ^ c + c ^ d) = R(a ^ b) + R(c ^ d) (7:53)
so that R(B ) is now a bivector-valued linear function of its bivector argument B . The
space of bivectors is 6-dimensional so R(B ) has, at most, 36 degrees of freedom. (The
counting of degrees of freedom is somewhat easier than in conventional tensor calculus,
since two of the symmetries of R
 are automatically incorporated.) R(B ) transforms
under local translations as
Rx(B ) 7! Rx f x(B )
0 where x0 = f (x) (7:54)
and under local rotations as
R(B ) 7! R0R(B )R~0: (7:55)
The
eld-strength is contracted once to form the linear function
R(b) = h(@a)
R(a ^ b) (7:56)
which has the same transformation properties as R(B ). We use the same symbol for
both functions and distinguish between them through their argument, which is either a
bivector (B or a ^ b) or a vector (a).
Contracting once more, we arrive at the (\Ricci") scalar
R = h(@b)
R(b) = h(@b ^ @a)
R(a ^ b): (7:57)
R transforms as a scalar function under both rotations and translations. As an aside,
it is interesting to construct the analogous quantity to R for the electromagnetic gauge
sector. For this we
nd that
Daem  Dbem ] = e(b ^ a)
Fi3 (7:58)
) h(@b ^ @a)Daem  Dbem ] = ;2eh(F )i3: (7:59)
Interestingly, this suggests that the bivector h(F ) has a similar status to the Ricci scalar,
and not to the
eld-strength tensor.
Since the Ricci scalar R is a covariant scalar, the action integral
Z
SG = jd4xj(det h);1 R=2 (7:60)
is invariant under all local Poincare transformations. The choice of action integral (7.60) is
the same as that of the Hilbert-Palatini principle in general relativity, and we investigate
the consequences of this choice now. Once we have derived both the gravitational and
matter equations, we will return to the subject of whether this choice is unique.

144
 From (7.60) we write the Lagrangian density as
LG = 21 R det h;1 = LG (h(a) +(a) b
r+(a)): (7:61)
The action integral (7.60) is over a region of at spacetime, so all the variational princi-
ple techniques developed in Chapter 6 hold without modi
cation. The only elaboration
needed is to de
ne a calculus for +(a). Such a calculus can be de
ned in precisely the
same way as the derivative @h(a) was de
ned (6.117). The essential results are:
@(a)h+(b)B i = a
bB (7.62)
@(b)
a hc
r+(d)B i = a
cb
dB (7.63)
where B is an arbitrary bivector.
We assume that the overall action is of the form
L = LG ; LM  (7:64)
where LM describes the matter content and  = 8G. The
rst of the
eld equations is
found by varying with respect to h, producing
@h(a)LM = 21 @h(a)(hh(@b ^ @c)R(c ^ b)i det h;1 )
= (R(a) ; 21 h;1(a)R) det h;1: (7.65)
The functional derivative with respect to h(a) of the matter Lagrangian is taken to de
ne
the stress-energy tensor of the matter
eld through
T h;1(a) det h;1  @h(a)LM  (7:66)
so that we arrive at the
eld equations in the form
R(a) ; 12 h;1(a)R = T h;1 (a): (7:67)
It is now appropriate to de
ne the functions
R(a ^ b)  Rh(a ^ b) (7.68)
R(a)  Rh(a) = @a
R(a ^ b) (7.69)
G  R(a) ; 12 aR: (7.70)
These are covariant under translations (they simply change their position dependence),
and under rotations they transform as e.g.
R(B ) 7! R0R(R~ 0BR0)R~0: (7:71)
Equation (7.71) is the de
ning rule for the transformation properties of a tensor, and
we hereafter refer to (7.68) through to (7.70) as the Riemann, Ricci and Einstein tensors
respectively. We can now write (7.67) in the form
G (a) = T (a) (7:72)
145
which is the (at-space) gauge-theory equivalent of Einstein's
eld equations.
In the limit of vanishing gravitational
elds (h(a) 7! a and +(a) 7! 0) the stress-
energy tensor de
ned by (7.66) agrees with the canonical stress-energy tensor (6.24), up
to a total divergence. When +(a) vanishes, the matter action is obtained from the free-

eld L(i a
ri) through the introduction of the transformation de

ned by x0 = h(x).
Denoting the transformed
elds by i0, we
nd that
 
@h(a)(det h);1L0]h=I = @h(a)L(i0  h(a)
ri0)h=I ; aL (7:73)
and

@h(a)L(i0 h(b)
ri0)h=I
= @h(a)i0  @i L + (@b
h(r)i0)  @i
b L]h=I
= @b(a
ri)  @i
b L + @h(a)_ i0  @i L + h(@b)
r_ @i
b L]h=I : (7.74)
When the
eld equations are satisifed, the
nal term in (7.74) is a total divergence, and
we recover the stress-energy tensor in the form (6.24). This is unsurprising, since the
derivations of the functional and canonical stress-energy tensors are both concerned with
the eects of moving
elds from one spacetime position to another.
The de
nition (7.66) diers from that used in general relativity, where the functional
derivative is taken with respect to the metric tensor 83]. The functional derivative with
respect to the metric ensures that the resultant stress-energy tensor is symmetric. This is
not necessarily the case when the functional derivative is taken with respect to h(a). This
is potentially important, since we saw in Chapter 6 that the antisymmetric contribution
to the stress-energy tensor is crucial for the correct treatment of the spin of the
elds.
We next derive the
eld equations for the +(a)
eld. We write these in the form
@(a)LG ; @b
r(@(a)
b LG )
n o
=  @(a)LM ; @b
r(@(a)
b LM )  S (a) det h;1 (7.75)
where the right-hand side de
nes the function S (a). Performing the derivatives on the
left-hand side of the equation gives



det h;1+(b)  h(@b) ^ h(a) + @b
r h(b) ^ h(a) det h;1 = S (a) det h;1: (7:76)
On contracting (7.76) with h;1(@a), we
nd that
h;1(@a)
S (a) det h;1 n
o
= @a
+(b)  (h(@b) ^ a)] det h;1 + h;1(@a)
@b
r h(b) ^ h(a) det h;1
= 2h(@a)
+(a) det h;1 ; 3h(r det h;1) ; h(r_ )h;1 (@a)
h_ (a) det h;1
$

+h_ (r_ ) det h;1: (7.77)

If we now make use of the result that
ha
rh(b)h;1 (@b) det h;1i = ;(a
rh(@b))  @h(b) det h;1
= ;a
r det h;1 (chain rule) (7.78)
146
from which it follows that
h(r_ )hh_ (a)h;1(@a) det h;1i = ;h(r det h;1) (7:79)
we
nd that
h;1(@a)
S (a) det h;1 = 2h(@a)
+(a) det h;1 ; 2h(r det h;1)
$

= ;2Dah(@a det h;1): (7.80)

We will see shortly that it is crucial to the derivation of the correct matter
eld equations
that
Dah(@a det h;1) = 0: (7:81)
This places a strong restriction on the form of LM , which must satisfy


h;1(@a)
@(a)LM ; @b
r(@(a)
b LM ) = 0: (7:82)
This condition is satis
ed for our gauge theory based on the Dirac equation, since the
bracketed term in equation (7.82) is
(@(a) ; @b
r @(a)
b )hDi3~ ; m~i = 21 h(a)
(i3~)
!

= ; 12 ih(a) ^ s: (7.83)
It follows immediately that the contraction in (7.82) vanishes, since
h;1 (@a)
(ih(a) ^ s) = ;i@a ^ a ^ s = 0: (7:84)
We de
ne S by
S  21 i3~ (7:85)
so that we can now write
S (a) = h(a)
S : (7:86)
Given that (7.81) does hold, we can now write (7.76) in the form



S (a) = h_ (r) ^ h_ (a) + +(b)  h(@b) ^ h(a) ; +(b)
h(@b) ^ h(a)


= h(@b) ^ b
rh(a) + +(b)
h(a)
= D^ h(a): (7.87)
The right-hand side of this equation could be viewed as the torsion though, since we are
working in a at spacetime, it is preferable to avoid terminology borrowed from Riemann-
Cartan geometry. When the left-hand side of (7.87) vanishes, we arrive at the simple
equation
D^ h(a) = 0 (7:88)
valid for all constant vectors a. All dierential functions f (a) = a
rf (x) satisfy r^f (a) =
0, and (7.88) can be seen as the covariant generalisation of this result. Our gravitational

eld equations are summarised as

G (a) = T (a) (7.89)
D^ h(a) = S (a) (7.90)
which hold for all constant vectors a.
147
7.1.3 The Matter-Field Equations
We now turn to the derivation of the matter-
eld equations. We start with the Dirac
equation, and consider the electromagnetic
eld equations second.
The Dirac Equation
We have seen how the demand for invariance under local Poincare transformations has
led to the action Z
S = jd4xj(det h);1 hh(@a)Dai3~ ; m~i: (7:91)
Applying the Euler-Lagrange equations (6.12) to this, and reversing the result, we
nd
that


det h;1 h(@a)a
ri3 + h(@a) ^ +(a)i3 ; 2m = ;@a
r(h(a)i3 det h;1) (7:92)
which can be written as
Di3 = m ; 12 Dah(@a det h;1 )i3: (7:93)
We now see why it is so important that Dah(@a det h;1) vanishes. Our point of view
throughout has been to start from the Dirac equation, and to introduce gauge
elds to
ensure local Poincare invariance. We argued initially from the point of view of the Dirac
action, but we could equally well have worked entirely at the level of the equation. By
starting from the Dirac equation
ri3 = m (7:94)
and introducing the h and +(a)
elds in the same manner as in Section 2.1, we
nd that
the correct minimally coupled equation is
Di3 = m: (7:95)
If we now make the further restriction that our
eld equations are derivable from an
action principle, we must demand that (7.93) reduces to (7.95). We are therefore led to
the constraint that Da h(@a det h;1) vanishes. To be consistent, this constraint must be
derivable from the gravitational
eld equations. We have seen that the usual Hilbert-
Palatini action satis
es this requirement, but higher-order contributions to the action
would not. This rules out, for example, the type of \R + R2 " Lagrangian often considered
in the context of Poincare gauge theory 84, 85, 86]. Satisfyingly, this forces us to a theory
which is
rst-order in the derivatives of the
elds. The only freedom that remains is the
possible inclusion of a cosmological constant, though such a term would obviously violate
our intention that gravitational forces should result directly from interactions between
particles.
The full set of equations for Dirac matter coupled to gravity is obtained from the
action Z
S = jd4xj(det h);1 ( 12 R ; hh(@a)Da i3~ ; m~i) (7:96)

148
and the
eld equations are
G (a) = ha
Di3~i1 (7.97)
D^ h(a) = h(a)
( 12 i3~) = h(a)
( 21 is) (7.98)
Di3 = m0: (7.99)
It is not clear that self-consistent solutions to these equations could correspond to any
physical situation, as such a solution would describe a self-gravitating Dirac uid. Self-
consistent solutions have been found in the context of cosmology, however, and the solu-
tions have the interesting property of forcing the universe to be at critical density 10].
The Electromagnetic Field Equations
We now return to the introduction of the electromagnetic
eld. From the action (7.91),
and following the procedure of the start of this chapter, we arrive at the action
Z
SD+EM = jd4xj(det h);1 hh(@a)(Dai3~ ; ea
A0~) ; m~i: (7:100)
The
eld equation from this action is
Di3 ; eA = m0 (7:101)
where we have introduced the notation
A = h(A): (7:102)
It is to be expected that A should appear in the
nal equation, rather than A. The
vector potential A originated as the generalisation of the quantity r. If we examine
what happens to this under the translation (x) 7! (x0), with x0 = f (x), we
nd that
r 7! f (rx (x0)):
0 (7.103)
It follows that A must also pick up a factor of f as it is moved from x0 to x,
A(x) 7! f (A(x0)) (7.104)
so it is the quantity A that is Poincare-covariant, as are all the other quantities in equa-
tion (7.101). However, A is not invariant under local U (1) transformations. Instead, we
must construct the Faraday bivector
F = r^ A: (7:105)
It could be considered a weakness of conventional spin-torsion theory that, in order to
construct the gauge-invariant quantity F , one has to resort to the use of the at-space
vector derivative. Of course, in our theory background spacetime has never gone away,
and we are free to exploit the vector derivative to the full.
The conventional approach to gauge theories of gravity (as discussed in 81], for ex-
ample) attempts to de
ne a minimal coupling procedure for all matter
elds, preparing
149
the way for a true curved-space theory. The approach here has been rather dierent, in
that everything is derived from the Dirac equation, and we are attempting to put electro-
magnetic and gravitational interactions on as similar a footing as possible. Consequently,
there is no reason to expect that the gravitational
eld should \minimally couple" into the
electromagnetic
eld. Instead, we must look at how F behaves under local translations.
We
nd that
F (x) 7! r^ fA(x0) = f (rx ^ A(x0))
0 (7.106)
= fF (x0) (7.107)
so the covariant form of F is
F  h(F ): (7:108)
F is covariant under local Poincare transformations, and invariant under U (1) transfor-
mations. The appropriate action for the electromagnetic
eld is therefore
Z
SEM = jd4xj(det h);1 h 12 FF ; A
J i (7:109)
which reduces to the standard electromagnetic action integral in the limit where h is the
identity. To
nd the electromagnetic
eld equations, we write
LEM = det h;1h 12 FF ; A
J i = L(A a
rA) (7:110)
and treat the h and J
elds as external sources. There is no +-dependence in (7.109), so
LEM satis
es the criteria of equation (7.82).
Variation of LEM with respect to A leads to the equation


@a
r(a
h(F ) det h;1) = r
hh(r^ A) det h;1 = det h;1 J (7:111)
which combines with the identity
r^ F = 0 (7:112)
to form the Maxwell equations in a gravitational background. Equation (7.111) corre-
sponds to the standard second-order wave equation for the vector potential A used in
general relativity. It contains only the functions hh  g;1 and det h;1 = (det g)1=2,
where g is the symmetric \metric" tensor. The fact that equation (7.111) only involves h
through the metric tensor is usually taken as evidence that the electromagnetic
eld does
not couple to torsion.
So far, we only have the Maxwell equations as two separate equations (7.111) and
(7.112). It would be very disappointing if our STA approach did not enable us to do better
since one of the many advantages of the STA is that, in the absence of a gravitational

eld, Maxwell's equations

r
F = J r^ F = 0 (7:113)
can be combined into a single equation
rF = J: (7:114)

150
This is more than a mere notational convenience. The r operator is invertible, and can be
used to develop a
rst-order propagator theory for the F -
eld 8]. This has the advantages
of working directly with the physical
eld, and of correctly predicting the obliquity factors
that have to be put in by hand in the second-order approach (based on a wave equation
for A). It would undermine much of the motivation for pursuing
rst-order theories if
this approach cannot be generalised to include gravitational eects. Furthermore, if we
construct the stress-energy tensor, we
nd that
TEM h;1(a) det h;1 = 21 @h(a)hh(F )h(F ) det h;1i


= h(a
F )
F ; 12 h;1(a)F
F det h;1 (7.115)
which yields
TEM (a) = ;(F
a)
F ; 21 aF
F
= ; 12 F aF : (7.116)
This is the covariant form of the tensor found in Section (6.2). It is intersting to see how
the de
nition of TEM as the functional derivative of L with respect to @h(a) automatically
preserves gauge invariance. For electromagnetism this has the eect of forcing TEM to
be symmetric. The form of TEM (7.116) makes it clear that it is F which is the genuine
physical
eld, so we should seek to express the
eld equations in terms of this object. To
achieve this, we
rst write the second of the
eld equations (7.90) in the form
D^ h(a) = h(r^ a) + h(a)
S  (7:117)
which holds for all a. If we now de
ne the bivector B = a ^ b, we
nd that
D^ h(B ) = D^ h(a)] ^ h(b) ; h(a) ^D^ h(b)
= h(r^ a) ^ h(b) ; h(a) ^ h(r^ b) + (h(a)
S ) ^ h(b)
;h(a) ^ (h(b)
S )
= h(r^ B ) ; h(B ) S  (7.118)
which is used to write equation (7.112) in the form
D^F ; SF = h(r^ F ) = 0: (7:119)
Next, we use a double-duality transformation on (7.111) to write the left-hand side as
r
(h(F ) det h;1) = ir^ (ih(F ) det h;1)
= ir^ (h;1(iF ))
= ih;1 (D^ (iF ) + (iF ) S )  (7.120)
so that (7.111) becomes
D
F ; S
F = ih(Ji) det h;1 = h;1(J ): (7:121)

151
Writing
J = h;1 (J ) (7:122)
we can now combine (7.119) and (7.121) into the single equation
DF ; SF = J  (7:123)
which achieves our objective. The gravitational background has led to the vector deriva-
tive r being generalised to D ; S . Equation (7.123) surely deserves considerable study.
In particular, there is a clear need for a detailed study of the Green's functions of the
D ; S operator. Furthermore, (7.123) makes it clear that, even if the A equation does
not contain any torsion term, the F equation certainly does. This may be of importance
in studying how F propagates from the surface of an object with a large spin current.

7.1.4 Comparison with Other Approaches

We should now compare our theory with general relativity and the ECKS theory. In
Sections 7.2 and 7.3 a number of physical dierences are illustrated, so here we concentrate
on the how the mathematics compares. To simplify the comparison, we will suppose
initially that spin eects are neglible. In this case equation (7.90) simpli
es to D^h(a) = 0.
This equation can be solved to give +(a) as a function of h(a). This is achieved by
rst
\protracting" with h;1(@a):
h i
h;1(@a) ^ (D^ h(a)) = h;1 (@a) ^ h(r) ^ h(a) + h(@b) ^ (+(b)
h(a))
= h;1 (@b) ^ h(r) ^ h(b) + 2h(@b) ^ +(b) = 0: (7.124)
Contracting this expression with the vector h;1(a) and rearranging yields
2+(a) = ;2h(@b) ^ (+(b)
h;1(a)) ; h;1(a)
h;1(@b)h(r) ^ h(b)
+h;1(@b) ^ (a
rh(b)) ; h;1(@b) ^ h(r_ )h_ (b)
h;1 (a)
= ;2h(r^ g(a)) + h(r) ^ h;1(a) ; h(r_ ) ^ hg _ (a)
+h;1(@b) ^ (a
rh(b))
= ;h(r^ g(a)) + h;1(@b) ^ (a
rh(b)) (7.125)
where
g(a)  h;1 h;1(a): (7:126)
The quantity g(a) is the gauge-theory analogue of the metric tensor. It is symmetric, and
arises naturally when forming inner products,
h;1 (a)
h;1(b) = a
g(b) = g(a)
b: (7:127)
Under translations g(a) transforms as
gx(a) 7! f xgx f x(a) where x0 = f (x)
0 (7:128)

152
and under an active rotation g(a) is unchanged. The fact that g(a) is unaected by active
rotations limits its usefulness, and this is a strong reason for not using the metric tensor
as the foundation of our theory.
The comparison with general relativity is clari
ed by the introduction of a set of 4
coordinate functions over spacetime, x = x (x). From these a coordinate frame is de
ned
by
e = @ x (7:129)
where @ = @x . The reciprocal frame is de
ned as
e = rx (7:130)
and satis
es
e

e = (@ x)
rx
= @x x
=

: (7:131)
From these we de
ne a frame of \contravariant" vectors
g = h;1(e ) (7:132)
and a dual frame of \covariant" vectors
g = h(e ): (7:133)
These satisfy (no torsion)
g
g
=

 (7.134)
D^ g = 0 (7.135)
and
g
Dg
; g

Dg = 0: (7:136)
The third of these identities is the at-space equivalent of the vanishing of the Lie bracket
for a coordinate frame in Riemannian geometry.
From the fg g frame the metric coecients are de
ned by
g
= g
g
 (7:137)
which enables us to now make contact with Riemannian geometry. Writing + for +(e ),
we
nd from (7.125) that
2+ = g ^ (@ g) + g ^ g
@
g : (7:138)
The connection is de
ned by
;
 = g
(D
g) (7:139)
so that, with a  a
g,
@
a ; ;
 a = @
(a
g) ; a
(D
g)
= g
(D
a) (7.140)

153
as required | the connection records the fact that, by writing a = a
g , additional
x-dependence is introduced through the g.
By using (7.138) in (7.139), ;
 is given by
;
 = 21 g (@
g + @g
; @g
 ) (7:141)
which is the conventional expression for the Christoel connection. In the absence of
spin, the introduction of a coordinate frame unpackages our equations to the set of scalar
equations used in general relativity. The essential dierence is that in GR the quantity g

is fundamental, and can only be de
ned locally, whereas in our theory the fundamental
variables are the h and +
elds, which are de
ned globally throughout spacetime. One
might expect that the only dierences that could show up from this shift would be due
to global, topological considerations. In fact, this is not the case, as is shown in the
following sections. The reasons for these dierences can be either physical, due to the
dierent understanding attached to the variables in the theory, or mathematical, due often
to the constraint that the metric must be generated from a suitable h function. It is not
always the case that such an h function can be found, as is demonstrated in Section 7.2.1.
The ability to develop a coordinate-free theory of gravity oers a number of advantages
over approaches using tensor calculus. In particular, the physical content of the theory
is separated from the artefacts of the chosen frame. Thus the h and +
elds only dier
from the identity and zero in the presence of matter. This clari
es much of the physics
involved, as well as making many equations easier to manipulate.
Many of the standard results of classical Riemannian geometry have particularly simple
expressions in this STA-based theory. Similar expressions can be found in Chapter 5 of
Hestenes & Sobczyk 24], who have developed Riemannian geometry from the viewpoint
of geometric calculus. All the symmetries of the Riemann tensor are summarised in the
single equation
@a ^R(a ^ b) = 0: (7:142)
This says that the trivector @a ^R(a ^ b) vanishes for all values of the vector b, and so
represents a set of 16 scalar equations. These reduce the 36-component tensor R(B ) to a
function with only 20 degrees of freedom | the correct number for Riemannian geometry.
Equation (7.142) can be contracted with @b to yield
@a ^R(a) = 0 (7:143)
which says that the Ricci tensor is symmetric. The Bianchi identity is also compactly
written:
_ R_ (B ) = 0
D^ (7:144)
where the overdot notation is de
ned via
D_ T_ (M )  DT (M ) ; @aT (a
DM ): (7:145)
Equation (7.144) can be contracted with @b ^ @a to yield


_ R_ (a ^ b) = @b
R_ (D^ _ b) ; D^
_ R_ (b)


(@b ^ @a)
D^
= ;2R_ (D_ ) + DR = 0: (7.146)
154
It follows that
G_ (D_ ) = 0 (7:147)
which, in conventional terms, represents conservation of the Einstein tensor. Many other
results can be written equally compactly.
The inclusion of torsion leads us to a comparison with the ECKS theory, which is
certainly closest to the approach adopted here. The ECKS theory arose from attempts
to develop gravity as a gauge theory, and modern treatments do indeed emphasise active
transformations 81]. However, the spin-torsion theories ultimately arrived at all involve
a curved-space picture of gravitational interactions, even if they started out as a gauge
theory in at space. Furthermore, the separation into local translations and rotations is
considerably cleaner in the theory developed here, as all transformations considered are

nite, rather than in

nitessimal. The introduction of a coordinate frame can be used
to reproduce the equations of a particular type of spin-torsion theory (one where the
torsion is generated by Dirac matter) but again dierences result from our use of a at
background spacetime. The inclusion of torsion alters equations (7.142) to (7.147). For
example, equation (7.142) becomes
@a ^R(a ^ b) = ;b
DS + 21 D^S b (7:148)
equation (7.143) becomes
@a ^R(a) = ;D
S (7:149)
and equation (7.144) becomes
_ R_ (B ) + SR(B ) = 0:
D^ (7:150)
The presence of torsion destroys many of the beautiful results of Riemannian geometry
and, once the connection between the gauge theory quantities and their counterparts in
Riemannian geometry is lost, so too is much of the motivation for adopting a curved-space
viewpoint.
Finally, it is important to stress that there is a dierence between the present gauge
theory of gravity and Yang-Mills gauge theories. Unlike Yang-Mills theories, the Poincare
gauge transformations do not take place in an internal space, but in real spacetime | they
transform between physically distinct situations. The point is not that all physical ob-
servables should be gauge invariant, but that the
elds should satisfy the same equations,
regardless of their state. Thus an accelerating body is subject to the same physical laws
as a static one, even though it may be behaving quite dierently (it could be radiating
away electromagnetic energy, for example).

7.2 Point Source Solutions

In this section we seek solutions to the
eld equations in the absence of matter. In this
case, the stress-energy equation (7.67) is
R(a) ; 21 aR = 0 (7:151)

155
which contracts to give
R = 0: (7:152)
Our
eld equations are therefore
D^ h(a) = h(r^ a) (7:153)
R(a) = 0:
As was discussed in the previous section, if we expand in a basis then the equations for
the coordinates are the same as those of general relativity. It follows that any solution
to (7.153) will generate a metric which solves the Einstein equations. But the converse
does not hold | the additional physical constraints at work in our theory rule out certain
solutions that are admitted by general relativity. This is illustrated by a comparison of
the Schwarzschild metric used in general relativity with the class of radially-symmetric
static solutions admitted in the present theory. Throughout the following sections we use
units with G = 1.

7.2.1 Radially-Symmetric Static Solutions

In looking for radially-symmetric solutions to (7.153), it should be clear that we are ac-
tually
nding possible
eld con
gurations around a
-function source (a point of matter).
That is, we are studying the analog of the Coulomb problem in electrostatics. In gen-
eral, specifying the matter and spin densities speci
es the h and +
elds completely via
the
eld equations (7.89) and (7.90). Applying an active transformation takes us to a
dierent matter con
guration and solves a dierent (albeit related) problem. This is not
the case when symmetries are present, in which case a class of gauge transformations
exists which do not alter the matter and
eld con
gurations. For the case of point-source
solutions, practically all gauge transformations lead to new solutions. In this case the
problem is simpli
ed by imposing certain symmetry requirements at the outset. By this
means, solutions can be classi
ed into equivalence classes. This is very natural from the
point of view of a gauge theory, though it should be borne in mind that in our theory
gauge transformations can have physical consequences.
Here we are interested in the class of radially-symmetric static solutions. This means
that, if we place the source at the origin in space, we demand that the h and +
elds only
show dependence on x through the spatial radial vector (spacetime bivector)
x = x ^ 0: (7:154)
Here 0 is a
xed time-like direction. We are free to choose this as we please, so that a
global symmetry remains. This rigid symmetry can only be removed with further physical
assumptions! for example that the matter is comoving with respect to the Hubble ow of
galaxies (i.e. it sees zero dipole moment in the cosmic microwave background anisotropy).
To facilitate the discussion of radially-symmetric solutions, it is useful to introduce a
set of polar coordinates
t = 0
x ;1 < t < 1
r = jx ^ 0j 0r<1 (7:155)
cos = ;3
x=r 0 
tan  = 2
x=(1
x) 0   < 2
156
where the f1 2 3g frame is a
xed, arbitrary spatial frame. From these coordinates,
we de
ne the coordinate frame
et = @tx = 0
er = @rx = sin cos1 + sin sin2 + cos 3 (7:156)
e = @ x = r(cos cos1 + cos sin2 ; sin 3)
e = @x = r sin (; sin1 + cos2):
The best-known radially-symmetric solution to the Einstein equations is given by the
Schwarzschild metric,
ds2 = (1 ; 2M=r)dt2 ; (1 ; 2M=r);1 dr2 ; r2(d 2 + sin2 d2) (7:157)
from which the components of g
= g
g
can be read straight o. Since g = h;1(e
),
we need to \square root" g
to
nd a suitable h;1 (and hence h) that generates it. This
h;1 is only unique up to rotations. If we look for such a function we immediately run into
a problem | the square roots on either side of the horizon (at r = 2M ) have completely
dierent forms. For example, the simplest forms have
)
gt = (1 ; 2M=r)1=2et g = e for r > 2M (7:158)
gr = (1 ; 2M=r);1=2 er g = e
and )
gt = (2M=r ; 1)1=2er g = e for r < 2M: (7:159)
gr = (2M=r ; 1);1=2et g = e
These do not match at r = 2M , and there is no rotation which gets round this problem. As
we have set out to
nd the
elds around a
-function source, it is highly undesirable that
these
elds should be discontinuous at some
nite distance from the source. Rather than
resort to coordinate transformations to try and patch up this problem, we will postulate
a suitably general form for h and +, and solve the
eld equations for these. Once this is
done, we will return to the subject of the problems that the Schwarzschild metric presents.
We postulate the following form for h(a)
h(et) = f1et + f2er h(e ) = e (7:160)
h(er) = g1er + g2et h(e) = e
where fi and gi are functions of r only. We can write h in the more compact form
h(a) = a + a
et ((f1 ; 1)et + f2er ) ; a
er ((g1 ; 1)er + g2et)  (7:161)
and we could go further and replace er and r by the appropriate functions of x. This
would show explicitly how h(a) is a linear function of a and a non-linear function of x^0.
We also postulate a suitable form for +(a), writing + for +(e ),
+t = er et + = (
1er +
2et)e =r (7:162)
+r = 0 + = (
1er +
2et)e=r
with  and
i functions of r only. More compactly, we can write
+(a) = a
eter et ; a ^ (eret)(
1et +
2er )=r: (7:163)
157
We could have used (7.138) to solve for +(a) in terms of the fi and gi, but this vastly
complicates the problem. The second-order form of the equations immediately introduces
unpleasant non-linearities, and the equations are far less easy to solve. The better asp-
proach is to use (7.138) to see what a suitable form for +(a) looks like, but to then leave
the functions unspeci
ed. Equations (7.160) and (7.162) do not account for the most
general type of radially-symmetric static solution. The trial form is chosen to enable us
to
nd a single solution. The complete class of solutions can then be obtained by gauge
transformations, which will be considered presently.
The
rst of the
eld equations (7.153) can be written as
D^ g = h(r) ^ g + g
^ (+

g
) = 0 (7:164)
which quickly yields the four equations
g1f10 ; g2f20 + (f12 ; f22) = 0 (7.165)
g1g2 ; g1g2 + (f1g2 ; f2g1) = 0
0 0 (7.166)
g1 =
1 + 1 (7.167)
g2 =
2 (7.168)
where the primes denote dierentiation with respect to r. We immediately eliminate
1
and
2 using (7.167) and (7.168). Next, we calculate the
eld strength tensor. Writing
R
for R(e ^ e
), we
nd that
Rtr = ;0er et
Rt = (g1et + g2 er)e =r
Rt = (g1et + g2 er)e=r (7:169)
Rr = (g10 er + g20 et)e =r
Rr = (g10 er + g20 et)e=r
R = (g12 ; g22 ; 1)e e=r2:
Contracting with g and setting the result equal to zero gives the
nal four equations
2 + 0r = 0 (7.170)
2g10 + f10r = 0 (7.171)
2g2 + f2 r = 0
0 0
(7.172)
r(f1g1 ; f2g2) + r(g1 g1 ; g2g2) + g1 ; g2 ; 1 = 0:
0 0 2 2 (7.173)
The
rst of these (7.170) can be solved for  immediately,
= M r2  (7:174)
where M is the (positive) constant of integration and represents the mass of the source.
Equations (7.171) and (7.172) now de
ne the fi in terms of the gi
f1 = g10 (7.175)
f2 = g2:
0
(7.176)
158
These are consistent with (7.166), and substituted into (7.165) yield
(f1g1 ; f2g2)0 = 0: (7:177)
But the quantity f1g1 ; f2g2 is simply the determinant of h, so we see that
det h = f1g1 ; f2g2 = constant: (7:178)
We expect the eect of the source to fall away to zero at large distances, so h should tend
asymptotically to the identity function. It follows that the constant det h should be set
to 1. All that remains is the single dierential equation (7.173)
1 r@ (g 2 ; g 2 ) + g 2 ; g 2 = 1 ; M=r (7:179)
2 r 1 2 1 2
to which the solution is
g12 ; g22 = 1 ; 2M=r (7:180)
ensuring consistency with det h = 1.
We now have a set of solutions de
ned by
 = M=r2
g1 ; g22 = 1 ; 2M=r
2
(7:181)
Mf1 = r2g10
Mf2 = r2g20 :
The ease of derivation of this solution set compares very favourably with the second-order
metric-based approach. A particularly pleasing feature of this derivation is the direct
manner in which  is found. This is the coecient of the +t bivector, which accounts
for the radial acceleration of a test particle. We see that it is determined simply by the
Newtonian formula!
The solutions (7.181) are a one-parameter set. We have a free choice of the g2 function,
say, up to the constraints that
g22(r)  2M=r ; 1 (7:182)
and )
f1 g1 ! 1 as r ! 1: (7:183)
f g ! 0
2 2
As an example, which will be useful shortly, one compact form that the solution can take
is
g1 = cosh(M=r) ; eM=rM=r f1 = cosh(M=r) + eM=rM=r (7:184)
g2 = ; sinh(M=r) + eM=rM=r f2 = ; sinh(M=r) ; eM=rM=r:
The solution (7.181) can be substituted back into (7.169) and the covariant
eld
strength tensor (Riemann tensor) is found to be
R(B ) = ;2 M r 3 B + 3 M B  (e e )e e
r3 r t r t

= ; 2Mr3 (B + 3er etBeret): (7.185)

159
It can now be con
rmed that @a
R(a ^ b) = 0. Indeed, one can simultaneously check
both the
eld equations and the symmetry properties of R(B ), since R(a) = 0 and
@a ^R(a ^ b) = 0 combine into the single equation
@aR(a ^ b) = 0: (7:186)
This equation greatly facilitates the study of the Petrov classi
cation of vacuum solutions
to the Einstein equations, as is demonstrated in Chapter 3 of Hestenes & Sobczyk 24].
There the authors refer to @a
R(a^b) as the contraction and @a^R(a^b) as the protraction.
The combined quantity @aR(a ^ b) is called simply the traction. These names have much
to recommend them, and are adopted wherever necessary.
Verifying that (7.185) satis
es (7.186) is a simple matter, depending solely on the
result that, for an arbitrary bivector B ,
@a(a ^ b + 3Ba ^ bB ;1) = @a(a ^ b + 3BabB ;1 ; 3Ba
bB ;1)
= @a(a ^ b ; 3a
b)
= @a(ab ; 4a
b)
= 0: (7.187)
The compact form of the Riemann tensor (7.185), and the ease with which the
eld
equations are veri
ed, should serve to demonstrate the power of the STA approach to
relativistic physics.
Radially-Symmetric Gauge Transformations
From a given solution in the set (7.181) we can generate further solutions via radially-
symmetric gauge transformations. We consider Lorentz rotations
rst. All rotations
leave the metric terms g
= g
g unchanged, since these are de
ned by invariant inner
products, so g12 ; g22, f12 ; f22, f1g2 ; f2g1 and det h are all invariant. Since the
elds
are a function of x^et only, the only Lorentz rotations that preserve symmetry are those
that leave x ^ et unchanged. It is easily seen that these leave the Riemann tensor (7.185)
unchanged as well. There are two such transformations to consider! a rotation in the
e ^ e plane and a boost along the radial axis. The rotors that determine these are as
follows:
Rotation: R = exp((r)ieret=2)! (7.188)
Radial Boost: R = exp((r)eret=2): (7.189)
Both rotations leave +t untransformed, but introduce an +r and transform the + and
+ terms.
If we take the solution in the form (7.184) and apply a radial boost determined by the
rotor  
R = exp M 2r r t 
e e (7:190)
we arrive at the following, highly compact solution
h(a) = a + Mr a
e;e;
+(a) = M r2 (e; ^ a + 2e;
aeret) (7.191)

160
where
e; = et ; er : (7:192)
Both the forms (7.191) and (7.184) give a metric which, in GR, is known as the (advanced-
time) Eddington-Finkelstein form of the Schwarzschild solution,
ds2 = (1 ; 2M=r)dt2 ; (4M=r)dr dt ; (1 + 2M=r)dr2 ; r2(d 2 + sin2 d2): (7:193)
There are also two types of transformation of position dependence to consider. The

rst is a (radially-dependent) translation up and down the et-axis,

xy = f (x) = x + u(r)et: (7:194)
(We use the dagger to denote the transformed position, since we have already used a
prime to denote the derivative with respect to r.) From (7.194) we
nd that
f (a) = a ; u0a
eret (7.195)
;1
f (a) = a + u0a
eter (7.196)
and that
xy ^ et = x ^ et: (7:197)
Since all x-dependence enters h through x^et it follows that hx = hx and +x = +x. The
y y

transformed functions therefore have

hy(et) = (f1 + u0g2)et + (f2 + u0g1)er (7.198)
h (er ) = h(er)
y
(7.199)
+ (et) = +(et)
y
(7.200)
+y(er ) = (Mu0=r2 )eret (7.201)
with all other terms unchanged. The fi's transform, but the gi 's are
xed. A time
translation can be followed by a radial boost to replace the +y(er ) term by +(er ), and so
move between solutions in the one-parameter set of (7.181).
The
nal transformation preserving radial symmetry is a radial translation, where the

elds are stretched out along the radial vector. We de

ne
xy = f (x) = x
etet + u(r)er (7:202)
so that
ry = jxy ^ etj = u(r) (7.203)
eyr = jxxy ^^ eetj et = er :
y
(7.204)
t
The dierential of this transformation gives
f (a) = a
etet ; u0a
erer + ur a ^ (eret)er et (7.205)
f ;1(a) = a
etet ; u10 a
erer + ur a ^ (eret)er et (7.206)

161
and
det f = u0(u=r)2: (7:207)
The new function hy = hx f ;1 has an additional dilation in the e e plane, and the
y

behaviour in the eret plane is de

ned by
fiy(r) = fi (ry) (7.208)
giy(r) = u10 gi(ry): (7.209)
The horizon has now moved from r = 2M to ry = 2M , as is to be expected for an
active radial dilation. The physical requirements of our theory restrict the form that
the transformation (7.202) can take. The functions r and u(r) both measure the radial
distance from a given origin, and since we do not want to start moving the source around
(which would change the problem) we must have u(0) = 0. The function u(r) must
therefore be monotomic-increasing to ensure that the map between r and r0 is 1-to-1.
Furthermore, u(r) must tend to r at large r to ensure that the
eld dies away suitably. It
follows that
u0(r) > 0 (7:210)
so the transformation does not change the sign of det h.
We have now found a 4-parameter solution set, in which the elements are related via
the rotations (7.188) and (7.189) and the transformations (7.194) and (7.202). The
elds
are well-de
ned everywhere except at the origin, where a point mass is present. A second
set of solutions is obtained by the discrete operation of time-reversal, de
ned by
f (x) = ;etxet (7.211)
) f (x) ^ et = ;(etxet) ^ et = x ^ et: (7.212)
This translation on its own just changes the signs of the fi functions, and so reverses the
sign of det h. The translation therefore de
nes
elds whose eects do not vanish at large
distances. To correct this, the h and +
elds must also be time-reversed, so that the new
solution has
hT (a) = ;ethf (x)(;etaet)et
= eth(etaet)et (7.213)
and
+T (a) = et+f (x)(;etaet)et
= ;et+(etaet)et: (7.214)
For example, the result of time-reversal on the solution de
ned by (7.191) is the new
solution
hT (a) = etetaet + Mr (etaet)
e;e;]et
= a + Mr a
e+e+ (7.215)

162
and
+T (a) = ; M2 et (e; ^ (etaet) + 2e;
(etaet)er et) et
r
M
= r2 (2a
e+eret ; e+ ^ a)  (7.216)
where e+ = et + er . This new solution reproduces the metric of the retarded-time
Eddington-Finkelstein form of the Schwarzschild solution. Time reversal has therefore
switched us from a solution where particles can cross the horizon on an inward journey,
but cannot escape, to a solution where particles can leave, but cannot enter. Covariant
quantities, such as the
eld strength (7.169), are, of course, unchanged by time reversal.
From the gauge-theory viewpoint, it is natural that the solutions of the
eld equations
should fall into sets which are related by discrete transformations that are not simply con-
nected to the identity. The solutions are simply reproducing the structure of the Poincare
group on which the theory is constructed.
Behaviour near the Horizon
For the remainder of this section we restrict the discussion to solutions for which det h = 1.
For these the line element takes the form
ds2 = (1 ; 2M=r)dt2 ; (f1g2 ; f2g1)2dr dt ; (f12 ; f22)dr2
;r2(d 2 + sin2 d2): (7.217)
The horizon is at r = 2M , and at this distance we must have
g1 = g2: (7:218)
But, since det h = f1g1 ; f2g2 = 1, we must also have
f1g2 ; f2g1 = 1 at r = 2M (7:219)
so an o-diagonal term must be present in the metric at the horizon. The assumption that
this term can be transformed away everywhere does not hold in our theory. This resolves
the problem of the Schwarzschild discontinuity discussed at the start of this section. The
Schwarzschild metric does not give a solution that is well-de
ned everywhere, so lies
outside the set of metrics that are derivable from (7.181). Outside the horizon, however,
it is always possible to transform to a solution that reproduces the Schwarzschild line
element, and the same is true inside. But the transformations required to do this do
not mesh at the boundary, and their derivatives introduce
-functions there. Because
the Schwarzschild line element is valid on either side of the horizon, it reproduces the
correct Riemann tensor (7.185) on either side. Careful analysis shows, however, that
the discontinuities in the + and +
elds required to reproduce the Schwarzschild line
element lead to
-functions at the horizon in R(a ^ b).
The fact that the f1g2 ; f2g1 term must take a value of 1 at the horizon is interesting,
since this term changes sign under time-reversal (7.213). Once a horizon has formed, it
is therefore no longer possible to
nd an h such that the line element derived from it is
163
invariant under time reversal. This suggests that the f1g2 ; f2g1 term retains information
about the process by which the horizon formed | recording the fact that at some earlier
time matter was falling in radially. Matter infall certainly picks out a time direction,
and knowledge of this is maintained after the horizon has formed. This irreversibility is
apparent from the study of test particle geodesics 9]. These can cross the horizon to
the inside in a
nite external coordinate time, but can never get back out again, as one
expects of a black hole.
The above conclusions dier strongly from those of GR, in which the ultimate form
of the Schwarzschild solution is the Kruskal metric. This form is arrived at by a series
of coordinate transformations, and is motivated by the concept of \maximal extension"
| that all geodesics should either exist for all values of their ane parameter, or should
terminate at a singularity. None of the solutions presented here have this property. The
solution (7.191), for example, has a pole in the proper-time integral for outgoing radial
geodesics. This suggests that particles following these geodesics would spend an in
nite
coordinate time hovering just inside the horizon. In fact, in a more physical situation
this will not be the case | the eects of other particles will tend to sweep all matter
back to the centre. The solutions presented here are extreme simpli
cations, and there
is no compelling physical reason why we should look for \maximal" solutions. This is
important, as the Kruskal metric is time-reverse symmetric and so must fail to give a
globally valid solution in our theory. There are a number of ways to see why this happens.
For example, the Kruskal metric de
nes a spacetime with a dierent global topology to
at spacetime. We can reach a similar conclusion by studying how the Kruskal metric
is derived from the Schwarzschild metric. We assume, for the time being, that we are
outside the horizon so that a solution giving the Schwarzschild line element is
g1 = -1=2 g2 = 0 (7:220)
f1 = - ;1=2
f2 = 0
where
- = 1 ; 2M=r: (7:221)
The
rst step is to re-interpret the coordinate transformations used in general relativity
as active local translations. For example, the advanced Eddington-Finkelstein metric is
reached by de
ning
ty ; ry = t ; (r + 2M ln(r ; 2M )) (7.222)
ry = r (7.223)
or
xy = x ; 2M ln(r ; 2M )et (7:224)
which is now recognisable as a translation of the type of equation (7.194). The result of
this translation is the introduction of an f2y function
f2y = ; 2M ;1=2
r -  (7:225)
which now ensures that f1yg2y ; f2yg1y = 1 at the horizon. The translation (7.224), which
is only de
ned outside the horizon, has produced a form of solution which at least has a
164
chance of being extended across the horizon. In fact, an additional boost is still required
to remove some remaining discontinuities. A suitable boost is de
ned by
R = exp(er et=2) (7:226)
where
sinh  = 12 (-;1=2 ; -1=2) (7:227)
and so is also only de
ned outside the horizon. The result of this pair of transformations
is the solution (7.191), which now extends smoothly down to the origin.
In a similar manner, it is possible to reach the retarted-time Eddington-Finkelstein
metric by starting with the translation de
ned by
ty + ry = t + (r + 2M ln(r ; 2M )) (7.228)
ry = r: (7.229)
The Kruskal metric, on the other hand, is reached by combining the advance and retarded
coordinates and writing
ty ; ry = t ; (r + 2M ln(r ; 2M )) (7.230)
ty + ry = t + (r + 2M ln(r ; 2M )) (7.231)
which de
nes the translation
xy = x
etet + (r + 2M ln(r ; 2M ))er : (7:232)
This translation is now of the type of equation (7.202), and results in a completely dif-
ferent form of solution. The transformed solution is still only valid for r > 2M , and the
transformation (7.232) has not introduced the required f1g2 ; f2g1 term. No additional
boost or rotation manufactures a form which can then be extended to the origin. The
problem can still be seen when the Kruskal metric is written in the form
ds2 = 32M
3
;r=2M
(dw2 ; dz2) ; r2(d 2 + sin2 d2)
r e (7:233)
where
z 2 ; w 2 = 2M 1 (r ; 2M )e;r=2M (7.234)
w = tanh t  
z 4M (7.235)
which is clearly only de
ned for r > 2M . The loss of the region with r < 2M does not
present a problem in GR, since the r-coordinate has no special signi
cance. But it is a
problem if r is viewed as the distance from the source of the
elds, as it is in the present
theory, since then the
elds must be de
ned for all r. Even in the absence of torsion, the
at-space gauge-theory approach to gravity produces physical consequences that clearly
dier from general relativity, despite the formal mathematical similarities between the
two theories.
165
7.2.2 Kerr-Type Solutions
We now briey discuss how the Kerr class of solutions
t into the present theory. The
detailed comparisons of the previous section will not be reproduced, and we will simply
illustrate a few novel features. Our starting point is the Kerr line element in Boyer-
Lindquist form 87]

ds2 = dt2 ; 2( dr- + d 2) ; (r2 + L2) sin2 d2 ; 2Mr

2
2 (L sin d ; dt) 
2 2 (7:236)
where
2 = r2 + L2 cos2 (7.237)
- = r2 ; 2Mr + L2: (7.238)
The coordinates all have the same meaning (as functions of spacetime position x) as
de
ned in the preceding section (7.155), and we have diered from standard notation
in labelling the new constant by L as opposed to the more popular a. This avoids any
confusion with our use of a as a vector variable and has the added advantage that the
two constants, L and M , are given similar symbols. It is assumed that jLj < M , as is
expected to be the case in any physically realistic situation.
The solution (7.236) has two horizons (at - = 0) where the line element is singular and,
as with the Schwarzschild line element, no continuous function h exists which generates
(7.236). However, we can
nd an h which is well-behaved outside the outer horizon, and
a suitable form is de
ned by
r 2 + L2
h(et) = -1=2 et ; r Le

1=2
h(er) = - er
h(e ) = r e
h(e) = r e ; Lr-sin
2 2
1=2 et : (7.239)
The Riemann tensor obtained from (7.236) has the remarkably compact form
M
R(B ) = ; 2(r ; iL cos )3 (B + 3er etBeret): (7:240)
(This form for R(B ) was obtained with the aid of the symbolic algebra package Maple.)
To my knowledge, this is the
rst time that the Riemann tensor for the Kerr solution has
been cast in such a simple form.
Equation (7.240) shows that the Riemann tensor for the Kerr solution is algebraically
very similar to that of the Schwarzschild solution, diering only in that the factor of
(r ; iL cos )3 replaces r3. The quantity r ; iL cos is a scalar + pseudoscalar object and
so commutes with the rest of R(B ). It follows that the
eld equations can be veri
ed in
166
precisely the same manner as for the Schwarzschild solution (7.187). It has been known
for many years that the Kerr metric can be obtained from the Schwarzschild metric via
a complex coordinate transformation 88, 89]. This \trick" works by taking the Schwarz-
schild metric in a null tetrad formalism and carrying out the coordinate transformation
r 7! r ; jL cos : (7:241)
Equation (7.240) shows that there is more to this trick than is usually supposed. In
particular, it demonstrates that the unit imaginary in (7.241) is better thought of as a
spacetime pseudoscalar. This is not a surprise, since we saw in Chapter 4 that the role of
the unit imaginary in a null tetrad is played by the spacetime pseudoscalar in the STA
formalism.
The Riemann tensor (7.240) is clearly de
ned for all values of r (except r = 0). We
therefore expect to
nd an alternative form of h which reproduces (7.240) and is also
de
ned globally. One such form is de
ned by
h(et) = et + 212 (2Mr + L2 sin2 )e; ; r Le


h(er) = er + 212 (2Mr ; L2 sin2 )e;

h(e ) = r e
h(e) = r e ; Lr sin
2 2
2 e;  (7.242)
with
e; = (et ; er ): (7:243)
This solution can be shown to lead to the Riemann tensor in the form (7.240). The
solution (7.242) reproduces the line element of the advanced-time Eddington-Finkelstein
form of the Kerr solution. Alternatives to (7.242) can be obtained by rotations, though
at the cost of complicating the form of R(B ). One particular rotation is de
ned by the
rotor ( )
L
R = exp 2r e ^ (et ; er)  (7:244)
which leads to the compact solution
h(a) = a + Mr
2 a
e ; e ; ; L a
e e + ( r ; 1)a ^ (e e )e e :
r r   r t r t (7:245)
None of these solutions correspond to the original form found by Kerr 90]. Kerr's
solution is most simply expressed as
h(a) = a ; a
nn (7:246)
where  is a scalar-valued function and n2 = 0. The vector n can be written in the form
n = (et ; net) (7:247)
167
 3

2.5

1.5

0.5

0
-3
-2
-1
0
y
1
-2 -3
2 0 -1
2 1
3 3 x

Figure 7.1: Incoming light paths for the Kerr solution I | view from above. The paths
terminate over a central disk in the z = 0 plane.
where n is a spatial vector. The explicit forms of n and  can be found in Schier et al. 89]
and in Chapter 6 of \The mathematical theory of black holes" by S. Chandrasekhar 91].
These forms will not be repeated here. From the
eld equations it turns out that n
satis
es the equation 89]
n
rn = 0: (7:248)
The integral curves of n are therefore straight lines, and these represent the possible
paths for incoming light rays. These paths are illustrated in
gures (7.1) and (7.2). The
paths terminate over a central disk, where the matter must be present. The fact that the
solution (7.246) must represent a disk of matter was originally pointed out by Kerr in a
footnote to the paper of Newman and Janis 88]. This is the paper that
rst gave the
derivation of the Kerr metric via a complex coordinate transformation. Kerr's observation
is ignored in most modern texts (see 91] or the popular account 92]) where it is claimed
that the solution (7.246) represents not a disk but a ring of matter | the ring singularity,
where the Riemann tensor is in
nite.
The transformations taking us from the solution (7.246) to solutions with a point
singularity involve the translation
f (x) = x0  x ; Lr x
(i3) (7:249)
which implies that
(r0)2 = r2 + L2 cos2 : (7:250)
Only the points for which r satis
es
r  jL cos j (7:251)

168
 3

z 0

-1

-2

-3
-1
-0.5
0
y
0.5 -2 -3
1 0 -1
1 3 2 x

Figure 7.2: Incoming null geodesics for the Kerr solution II | view from side on.
are mapped onto points in the transformed solution, and this has the eect of cutting out
the central disk and mapping it down to a point. Curiously, the translation achieves this
whilst keeping the total mass
xed (i.e. the mass parameter M is unchanged). The two
types of solution (7.242) and (7.246) represent very dierent matter con
gurations, and
it is not clear that they can really be thought of as equivalent in anything but an abstract
mathematical sense.

7.3 Extended Matter Distributions

As a
nal application of our at-space gauge theory of gravity, we study how extended
matter distributions are handled. We do so by concentrating on gravitational eects in
and around stars. This is a problem that is treated very successfully by general relativ-
ity (see 93, Chapter 23] for example) and, reassuringly, much of the mathematics goes
through unchanged in the theory considered here. This is unsurprising, since we will as-
sume that all eects due spin are neglible and we have already seen that, when this is the
case, the introduction of a coordinate frame will reproduce the
eld equations of GR. It
will be clear, however, that the physics of the situation is quite dierent and the central
purpose of this section is to highlight the dierences. Later in this section we discuss some
aspects of rotating stars, which remains an active source of research in general relativity.
Again, we work in units where G = 1.
We start by assuming the simplest distribution of matter | that of an ideal uid.
The matter stress-energy tensor then takes the form
T (a) = ( + p)a
uu ; pa (7:252)
where  is the energy density, p is the pressure and u is the 4-velocity
eld of the uid.
We now impose a number of physical restrictions on T (a). We
rst assume that the
169
matter distribution is radially symmetric so that  and p are functions of r only, where r
is the (at-space!) radial distance from the centre of the star, as de
ned by (7.155). (We
use translational invariance to choose the centre of the star to coincide with the spatial
point that we have labelled as the origin). Furthermore, we will assume that the star is
non-accelerating and can be taken as being at rest with respect to the cosmic frame (we
can easily boost our
nal answer to take care of the case where the star is moving at a
constant velocity through the cosmic microwave background). It follows that the velocity

eld u is simply et, and T now takes the form

T (a) = ((r) + p(r))a
etet ; p(r)a: (7:253)
This must equal the gravitational stress-energy tensor (the Eintein tensor), which is gen-
erated by the h and + gauge
elds. Radial symmetry means that h will have the general
form of (7.160). Furthermore, the form of G (a) derived from (7.160) shows that f2 and g2
must be zero, and hence that h is diagonal. This conclusion could also have been reached
by considering the motions of the underlying particles making up the star. If these follow
worldlines x( ), where  is the ane parameter, then u is de
ned by
u = h;1(x_ ) (7.254)
) x_ = h(et): (7.255)
A diagonal h ensures that x_ is also in the et direction, so that the consituent particles
are also at rest in the 3-space relative to et. That this should be so could have been
introduced as an additional physical requirement. Either way, by specifying the details of
the matter distribution we have restricted h to be of the form
h(a) = (f (r) ; 1)a
etet ; (g(r) ; 1)a
erer + a: (7:256)
The ans&atz for the gravitational
elds is completed by writing
+t = (r)er et + = (g(r) ; 1)er e =r (7:257)
+r = 0 + = (g(r) ; 1)er e=r
where again it is convenient to keep (r) as a free variable, rather than solving for it in
terms of f and g. The problem can now be solved by using the
eld equations on their
own, but it is more convenient to supplement the equations with the additional condition
T_ (D_ ) = 0 (7:258)
which reduces to the single equation
p0(r) = f
g ( + p): (7:259)
Solving the
eld equations is now routine. One can either follow the method of Section 3.1,
or can simply look up the answer in any one of a number of texts. The solution is that
g(r) = (1 ; 2m(r)=r)1=2 (7:260)
170
where Zr
m(r) = 4r02(r0) dr0 : (7:261)
0
The pressure is found by solving the Oppenheimer-Volkov equation
p0 = ; ( + rp()(r m;(2rm
) + 4r3p) 
(r)) (7:262)
subject to the condition that p(R) = 0, where R is the radius of the star. The remaining
term in h is then found by solving the dierential equation
f 0(r) = ; m(r) + 4r3p (7:263)
f (r) r(r ; 2m(r))
subject to the constraint that
f (R) = (1 ; 2m(R)=R);1=2 : (7:264)
Finally, (r) is given by
(r) = (fg);1(m(r)=r2 + 4rp): (7:265)
The complete solution leads to a Riemann tensor of the form
R(B ) = 4 ( + p)B
etet ; B  (er et)er et]
; m2r(r3) (B + 3er etBeret) (7.266)
which displays a neat split into a surface term, due to the local density and pressure, and
a (tractionless) volume term, due to the matter contained inside the shell of radius r.
The remarkable feature of thie solution is that (7.261) is quite clearly a at-space
integral! The importance of this integral is usually downplayed in GR, but in the context
of a at-space theory it is entirely natural | it shows that the
eld outside a sphere of
radius r is determined completely by the energy density within the shell. It follows that
the
eld outside the star is associated with a \mass" M given by
ZR
M = 4r02(r0) dr0 : (7:267)
0
We can understand the meaning of the de
nition of m(r) by considering the covariant
integral of the energy density
Z
E0 = eti h;1(d3x)
ZR 2
= 4r0 (1 ; 2m(r0)=r0);1=2(r0) dr0 : (7.268)
0
This integral is invariant under active spatial translations of the energy density. That
is to say, E0 is independent of where that matter actually is. In particular, E0 could
be evaluated with the matter removed to a suciently great distance that each particle
171
making up the star can be treated in isolation. It follows that E0 must be the sum of
the individual mass-energies of the component particles of the star | E0 contains no
contribution from the interaction between the particles. If we now expand (7.268) we
nd
that
ZR 2
E0  4r0 ((r0) + (r0)m(r0)=r0) dr0
0
= M ; Potential Energy: (7.269)
The external mass M is therefore the sum of the mass-energy E0 (which ignored interac-
tions) and a potential energy term. This is entirely what one expects. Gravity is due to
the presence of energy, and not just (rest) mass. The eective mass seen outside a star
is therefore a combination of the mass-energies of the constituent particles, together with
the energy due to their interaction with the remaining particles that make up the star.
This is a very appealing physical picture, which makes complete sense within the context
of a at-space gauge theory. Furthermore, it is now clear why the de
nition of M is not
invariant under radial translations. Interpreted actively, a radial translation changes the
matter distribution within the star, so the component particles are in a new con
guration.
It follows that the potential energy will have changed, and so too will the total energy.
An external observer sees this as a change in the strength of the gravitational attraction
of the star.
An important point that the above illustrates is that, given a matter distribution in
the form of T (a) and (more generally) S (a), the
eld equations are sucient to tie down
the gauge
elds uniquely. Then, given any solution of the
eld equation G (a) = 8T (a), a
new solution can always be reached by an active transformation. But doing so alters T (a),
and the new solution is appropriate to a di erent matter distribution. It is meaningless
to continue talking about covariance of the equations once the matter distribution is
speci
ed.
Whilst a non-vanishing T (a) does tie down the gauge
elds, the vacuum raises a
problem. When T (a) = 0 any gauge transformation can be applied, and we seem to have
no way of specifying the
eld outside a star, say. The resolution of this problem is that
matter (energy) must always be present in some form, whether it be the sun's thermal
radiation, the solar wind or, ultimately, the cosmic microwave background. At some level,
matter is always available to tell us what the h and +
elds are doing. This
ts in with
the view that spacetime itself does not play an active role in physics and it is the presence
of matter, not spacetime curvature, that generates gravitational interactions.
Since our theory is based on active transformations in a at spacetime, we can now
use local invariance to gain some insights into what the
elds inside a rotating star might
be like. To do this we rotate a static solution with a boost in the e direction. The rotor
that achieves this is
R = expf!(r )^etg (7:270)
where
^  e=(r sin ): (7:271)
The new matter stress-energy tensor is
T (a) = ( + p)a
(cosh ! et + sinh ! ^)(cosh !et + sinh ! ^) ; pa (7:272)
172
and the Einstein tensor is similarly transformed. The stress-energy tensor (7.272) can
only properly be associated with a rotating star if it carries angular momentum. The
de
nitions of momentum and angular momentum are, in fact, quite straight-forward.
The ux of momentum through the 3-space de
ned by a time-like vector a is T (a) and
the angular momentum bivector is de
ned by
J (a) = x ^T (a): (7:273)
Once gravitational interactions are turned on, these tensors are no longer conserved with
respect to the vector derivative,
T_ (r_ ) 6= 0 (7:274)
and instead the correct law is (7.258). This situation is analogous to that of coupled
Dirac-Maxwell theory (see Section 6.3). Once the
elds are coupled, the individual (free-

eld) stress-energy tensors are no longer conserved. To recover a conservation law, one
must either replace directional derivatives by covariant derivatives, or realise that it is
only the total stress-energy tensor that is conserved. The same is true for gravity. Once
gravitational eects are turned on, the only quantity that one expects to be conserved is
the sum of the individual matter and gravitational stress-energy tensors. But the
eld
equations ensure that this sum is always zero, so conservation of total energy-momentum
ceases to be an issue.
If, however, a global time-like symmetry is present, one can still sensibly separate
the total (zero) energy into gravitational and matter terms. Each term is then separately
conserved with respect to this global time. For the case of the star, the total 4-momentum
is the sum of the individual uxes of 4-momentum in the et direction. We therefore de
ne
the conserved momentum P by Z
P = d3x T (et) (7:275)
and the total angular momentum J by
Z
J = d3x x ^T (et): (7:276)
Concentrating on P
rst, we
nd that
P = Mrotet (7:277)
where
ZR Z h i
Mrot = 2 dr d r2 sin (r) cosh2 !(r ) + p(r) sinh2 !(r ) : (7:278)
0 0
The eective mass Mrot reduces to M when the rotation vanishes, and rises with the
magnitude of !, showing that the internal energy of the star is rising. The total 4-
momentum is entirely in the et direction, as it should be. Performing the J integral next,
we obtain
ZR Z
J = ;i3 2 dr d r3 sin2 ((r) + p(r)) sinh !(r ) cosh !(r ) (7:279)
0 0

173
so the angular momentum is contained in the spatial plane de
ned by the ^et direc-
tion. Performing an active radial boost has generated a
eld con
guration with suitable
momentum and angular momentum properties for a rotating star.
Unfortunately, this model cannot be physical, since it does not tie down the shape of
the star | an active transformation can always be used to alter the shape to any desired
con
guration. The missing ingredient is that the particles making up the star must satisfy
their own geodesic equation for motion in the
elds due to the rest of the star. The simple
rotation (7.270) does not achieve this.
Attention is drawn to these points for the following reason. The boost (7.270) produces
a Riemann tensor at the surface of the star of


R(B ) = ; M2rrot3 B + 3er (cosh ! et + sinh ! ^)Ber(cosh ! et + sinh ! ^)  (7:280)
which is that for a rotated Schwarzschild-type solution, with a suitably modi
ed mass.
This form is very dierent to the Riemann tensor for the Kerr solution (7.240), which
contains a complicated duality rotation. Whilst a physical model will undoubtedly require
additional modi
cations to the Riemann tensor (7.280), it is not at all clear that these
modi
cations will force the Riemann tensor to be of Kerr type. Indeed, the dierences
between the respective Riemann tensors would appear to make this quite unlikely. The
suggestion that a rotating star does not couple onto a Kerr-type solution is strengthened
by the fact that, in the 30 or so years since the discovery of the Kerr solution 90], no-one
has yet found a solution for a rotating star that matches onto the Kerr geometry at its
boundary.

7.4 Conclusions
The gauge theory of gravity developed from the Dirac equation has a number of inter-
esting and surprising features. The requirement that the gravitational action should be
consistent with the Dirac equation leads to a unique choice for the action integral (up to
the possible inclusion of a cosmological constant). The result is a set of equations which
are
rst-order in the derivatives of the
elds. This is in contrast to general relativity,
which is a theory based on a set of second-order partial dierential equations for the met-
ric tensor. Despite the formal similarities between the theories, the study of point-source
solutions reveals clear dierences. In particular, the
rst-order theory does not admit
solutions which are invariant under time-reversal.
The fact that the gauge group consists of active Poincare transformations of spacetime

elds means that gauge transformations relate physically distinct situations. It follows
that observations can determine the nature of the h and +
elds. This contrasts with
Yang-Mills theories based on internal gauge groups, where one expects that all observables
should be gauge-invariant. In this context, an important open problem is to ascertain
how the details of radial collapse determine the precise nature of the h and +
elds around
a black hole.
A strong point in favour of the approach developed here is the great formal clarity that
geometric algebra brings to the study of the equations. This is illustrated most clearly in
the compact formulae for the Riemann tensor for the Schwarzschild and Kerr solutions
174
and for radially-symmetric stars. No rival method (tensor calculus, dierential forms,
Newman-Penrose formalism) can oer such concise expressions.
For 80 years, general relativity has provided a successful framework for the study of
gravitational interactions. Any departure from it must be well-motivated by sound phys-
ical and mathematical reasons. The mathematical arguments in favour of the present
approach include the simple manner in which transformations are handled, the alge-
braic compactness of many formulae and the fact that torsion is perhaps better viewed
as a spacetime
eld than as a geometric eect. Elsewhere, a number of authors have
questioned whether the view that gravitational interactions are the result of spacetime
geometry is correct (see 94], for example). The physical motivation behind the present
theory is provided by the identi
cation of the h and +
elds as the dynamical variables.
The physical structure of general relativity is very much that of a classical
eld theory.
Every particle contributes to the curvature of spacetime, and every particle moves on
the resultant curved manifold. The picture is analogous to that of electromagnetism, in
which all charged particles contribute to an electromagnetic
eld (a kind of global ledger).
Yet an apparently crucial step in the development of Q.E.D. was Feynman's realisation
(together with Wheeler 95, 96]) that the electromagnetic
eld can be eliminated from
classical electrodynamics altogether. A similar process may be required before a quantum
multiparticle theory of gravity can be constructed. In the words of Einstein 97]
: : : the energy tensor can be regarded only as a provisional means of represent-
ing matter. In reality, matter consists of electrically charged particles : : :
The status of the h and +
elds can be regarded as equally provisional. They may simply
represent the aggregate behaviour of a large number of particles, and as such would not
be of fundamental signi
cance. In this case it would be wrong to attach too strong a
physical interpretation to these
elds (i.e. that they are the result of spacetime curvature
and torsion).
An idea of how the h
eld could arise from direct interparticle forces is provided by the
two-particle Dirac action constructed in Section 6.3. There the action integral involved
the dierential operator r1=m1 + r2=m2, so that the vector derivatives in each particle
space are weighted by the mass of the particle. This begins to suggest a mechanism by
which, at the one-particle level, the operator h(r) encodes an inertial drag due to the
other particle in the universe. This is plausible, as the h
eld was originally motivated
by considering the eect of translating a
eld. The theory presented here does appear
to contain the correct ingredients for a generalisation to a multiparticle quantum theory,
though only time will tell if this possibility can be realised.

175
Bibliography
1] A.N. Lasenby, C.J.L. Doran, and S.F. Gull. Grassmann calculus, pseudoclassical
mechanics and geometric algebra. J. Math. Phys., 34(8):3683, 1993.
2] C.J.L. Doran, D. Hestenes, F. Sommen, and N. van Acker. Lie groups as spin groups.
J. Math. Phys., 34(8):3642, 1993.
3] C.J.L. Doran, A.N. Lasenby, and S.F. Gull. Grassmann mechanics, multivector
derivatives and geometric algebra. In Z. Oziewicz, A. Borowiec, and B. Jancewicz,
editors, Spinors, Twistors and Cli ord Algebras, page 215. Kluwer, 1993.
4] A.N. Lasenby, C.J.L. Doran, and S.F. Gull. 2-spinors, twistors and supersymmetry
in the spacetime algebra. In Z. Oziewicz, A. Borowiec, and B. Jancewicz, editors,
Spinors, Twistors and Cli ord Algebras, page 233. Kluwer, 1993.
5] S.F. Gull, A.N. Lasenby, and C.J.L. Doran. Imaginary numbers are not real | the
geometric algebra of spacetime. Found. Phys., 23(9):1175, 1993.
6] C.J.L. Doran, A.N. Lasenby, and S.F. Gull. States and operators in the spacetime
algebra. Found. Phys., 23(9):1239, 1993.
7] A.N. Lasenby, C.J.L. Doran, and S.F. Gull. A multivector derivative approach to
Lagrangian
eld theory. Found. Phys., 23(10):1295, 1993.
8] S.F. Gull, A.N. Lasenby, and C.J.L. Doran. Electron paths, tunnelling and diraction
in the spacetime algebra. Found. Phys., 23(10):1329, 1993.
9] C.J.L. Doran, A.N. Lasenby, and S.F. Gull. Gravity as a gauge theory in the space-
time algebra. In F. Brackx and R. Delanghe., editors, Third International Confer-
ence on Cli ord Algebras and their Applications in Mathematical Physics., page 375.
Kluwer, 1993.
10] A.N. Lasenby, C.J.L. Doran, and S.F. Gull. Cosmological consequences of a at-
space theory of gravity. In F. Brackx and R. Delanghe., editors, Third International
Conference on Cli ord Algebras and their Applications in Mathematical Physics.,
page 387. Kluwer, 1993.
11] M.F. Atiyah, R. Bott, and A. Shapiro. Cliord modules. Topology, 3(Supp. 1):3,
1964.
12] I.R. Porteous. Topological Geometry. Van Nostrand Reinhold Company, 1969.
176
13] I.W. Benn and R.W. Tucker. An Introduction to Spinors and Geometry. Adam
Hilger, 1988.
14] C. Chevalley. The Algebraic Theory of Spinors. Columbia University Press, 1954.
15] H.B. Lawson and M.-L. Michelsohn. Spin Geometry. Princeton University Press,
1989.
16] F. Reese. Harvey. Spinors and Calibrations. Academic Press, San Diego, 1990.
17] D. Hestenes. Space-Time Algebra. Gordon and Breach, 1966.
18] D. Hestenes. Multivector calculus. J. Math. Anal. Appl., 24:313, 1968.
19] D. Hestenes. Vectors, spinors, and complex numbers in classical and quantum physics.
Am. J. Phys., 39:1013, 1971.
20] D. Hestenes. Proper dynamics of a rigid point particle. J. Math. Phys., 15(10):1778,
1974.
21] D. Hestenes. Observables, operators, and complex numbers in the Dirac theory. J.
Math. Phys., 16(3):556, 1975.
22] D. Hestenes and R. Gurtler. Consistency in the formulation of the Dirac, Pauli and
Schr&odinger theories. J. Math. Phys., 16(3):573, 1975.
23] D. Hestenes. Spin and uncertainty in the interpretation of quantum mechanics. Am.
J. Phys., 47(5):399, 1979.
24] D. Hestenes and G. Sobczyk. Cli ord Algebra to Geometric Calculus. D. Reidel
Publishing, 1984.
25] D. Hestenes. New Foundations for Classical Mechanics. D. Reidel Publishing, 1985.
26] D. Hestenes. A uni
ed language for mathematics and physics. In J.S.R. Chisholm
and A.K. Common, editors, Cli ord Algebras and their Applications in Mathematical
Physics, page 1. D. Reidel, 1986.
27] D. Hestenes. Cliord algebra and the interpretation of quantum mechanics. In J.S.R.
Chisholm and A.K. Common, editors, Cli ord Algebras and their Applications in
Mathematical Physics, page 321. D. Reidel, 1986.
28] D. Hestenes. Curvature calculations with spacetime algebra. Int. J. Theor. Phys.,
25(6):581, 1986.
29] D. Hestenes. On decoupling probability from kinematics in quantum mechanics.
In P. Foug.ere, editor, Maximum Entropy and Bayesian Methods, page 161. Kluwer,
1990.
30] D. Hestenes. The design of linear algebra and geometry. Acta Appl. Math., 23:65,
1991.
177
31] D. Hestenes and R. Ziegler. Projective geometry with Cliord algebra. Acta. Appli.
Math., 23:25, 1991.
32] D. Hestenes. Hamiltonain mechanics with geometric calculus. In Z. Oziewicz,
A. Borowiec, and B. Jancewicz, editors, Spinors, Twistors and Cli ord Algebras,
page 203. Kluwer, 1993.
33] D. Hestenes. Real Dirac theory. In Preparation, 1994.
34] D. Hestenes. Dierential forms in geometric calculus. In F. Brackx and R. Delanghe.,
editors, Third International Conference on Cli ord Algebras and their Applications
in Mathematical Physics. Kluwer, 1993.
35] F.A. Berezin. The Method of Second Quantization. Academic Press, 1966.
36] R. Penrose and W. Rindler. Spinors and space-time, Volume I: two-spinor calculus
and relativistic elds. Cambridge University Press, 1984.
37] R. Penrose and W. Rindler. Spinors and space-time, Volume II: spinor and twistor
methods in space-time geometry. Cambridge University Press, 1986.
38] A.O. Barut and N. Zanghi. Classical models of the Dirac electron. Phys. Rev. Lett.,
52(23):2009, 1984.
39] F.A. Berezin and M.S. Marinov. Particle spin dynamics as the Grassmann variant
of classical mechanics. Annals of Physics, 104:336, 1977.
40] M. Kline. Mathematical Thought from Ancient to Modern Times. Oxford University
Press, 1972.
41] H. Grassmann. Der ort der Hamilton'schen quaternionen in der ausdehnungslehre.
Math. Ann., 12:375, 1877.
42] W. K. Cliord. Applications of Grassmann's extensive algebra. Am. J. Math., 1:350,
1878.
43] M.F. Atiyah and I.M. Singer. The index of elliptic operators on compact manifolds.
Bull. A.M.S., 69:422, 1963.
44] T.G. Vold. An introduction to geometric algebra with an application to rigid body
mechanics. Am. J. Phys., 61(6):491, 1993.
45] T.G. Vold. An introduction to geometric calculus and its application to electrody-
namics. Am. J. Phys., 61(6):505, 1993.
46] R. Ablamovicz, P. Lounesto, and J. Maks. Second workshop on Cliord algebras and
their applications in mathematical physics. Found. Phys., 21(6):735, 1991.
47] E.T. Jaynes. Scattering of light by free electrons. In A. Weingartshofer and
D. Hestenes, editors, The Electron, page 1. Kluwer, 1991.

178
48] N. Salingaros. On the classi
cation of Cliord algebras and their relation to spinors
in n-dimensions. J. Math. Phys., 23(1):1, 1982.
49] I. Stewart. Hermann grassmann was right (News and Views). Nature, 321:17, 1986.
50] M. Barnabei, A. Brini, and G.-C. Rota. On the exterior calculus of invariant theory.
J. Algebra, 96:120, 1985.
51] F.A. Berezin. Introduction to Superanalysis. D. Reidel, 1987.
52] B. de Witt. Supermanifolds. Cambridge University Press, 1984.
53] R. Coquereaux, A. Jadczyk, and D. Kastler. Dierential and integral geometry of
Grassmann algebras. Reviews in Math. Phys., 3(1):63, 1991.
54] A. Connes and J. Lott. Particle models and non-commutative geometry. Nucl. Phys.
B (Proc. Suppl.), 18B:29, 1990.
55] G.C. Sherry. A generalised phase space for classical fermions arising out of Schon-
berg's geometric algebras. Found. Phys. Lett., 2(6):591, 1989.
56] G.C. Sherry. Algebraic model of a classical non-relativistic electron. Found. Phys.
Lett., 3(3):267, 1990.
57] J.F. Cornwell. Group Theory in Physics II. Academic Press, 1984.
58] A.O. Barut and A.J. Bracken. The remarkable algebra so (2n), its representations,
its Cliord algebra and potential applications. J. Phys. A, 23:641, 1990.
59] J.D. Bjorken and S.D. Drell. Relativistic Quantum Mechanics, vol 1. McGraw-Hill,
1964.
60] C. Itzykson and J-B. Zuber. Quantum Field Theory. McGraw-Hill, 1980.
61] S.F. Gull. Charged particles at potential steps. In A. Weingartshofer and D. Hestenes,
editors, The Electron, page 37. Kluwer, 1991.
62] E.M. Corson. Introduction to Tensors, Spinors, and Relativistic Wave-Equations.
Blackie & Son, Ltd. (Glasgow), 1953.
63] A.P. Galeao and P. Leal Ferreira. General method for reducing the two-body Dirac
equation. J. Math. Phys., 33(7):2618, 1992.
64] E.E. Salpeter and H.A. Bethe. A relativisitc equation for bound-state problems.
Phys. Rev., 84(6):1232, 1951.
65] G. Breit. The eect of retardation on the interaction of two electrons. Phys. Rev.,
34(4):553, 1929.
66] N. Kemmer. Zur theorie der neutron-proton wechselwirkung. Helv. Phys. Acta.,
10:48, 1937.
179
67] E. Fermi and C.N. Yang. Are mesons elementary particles? Phys. Rev., 76(12):1739,
1949.
68] Y. Koide. Exactly solvable model of relativistic wave equations and meson spectra.
Il Nuovo Cim., 70A(4):411, 1982.
69] W. Krolikowski. Dun-Kemmer-Petiau particle with internal structure. Acta Phys.
Pol. B, 18(2):111, 1987.
70] W. Krolikowski. Cliord algebras and the algebraic structure of fundamental
fermions. In Z. Oziewicz, A. Borowiec, and B. Jancewicz, editors, Spinors, Twistors
and Cli ord Algebras, page 183. Kluwer, 1993.
71] H. Goldstein. Classical Mechanics. Addison Wesley, 1950.
72] J.W. van Holten. On the electrodynamics of spinning particles. Nucl. Phys.,
B356(3):3, 1991.
73] P.G.O. Freund. Supersymmetry. Cambridge University Press, 1986.
74] R. Casalbuoni. The classical mechanics for Bose-Fermi systems. Il Nuovo Cimento.,
33A(3):389, 1976.
75] L. Brink, S. Deser, B. Zumino, P. di Vecchia, and P. Howe. Local supersymmetry
for spinning particles. Phys. lett. B, 64(4):435, 1976.
76] J.F. Cornwell. Group Theory in Physics III. Academic Press, 1989.
77] S. Coleman. Aspects of Symmetry. Cambridge University Press, 1985.
78] H. Tetrode. Der impuls-energiesatz in der Diracschen quantentheorie des elektrons.
Z. Physik, 49:858, 1928.
79] R. Utiyama. Invariant theoretical interpetation of interaction. Phys. Rev.,
101(5):1597, 1956.
80] T.W.B. Kibble. Lorentz invariance and the gravitational
eld. J. Math. Phys.,
2(3):212, 1961.
81] F.W. Hehl, P. von der Heyde, G.D. Kerlick, and J.M. Nestev. General relativity with
spin and torsion: Foundations and prospects. Rev. Mod. Phys., 48:393, 1976.
82] D. Ivanenko and G. Sardanashvily. The gauge treatment of gravity. Phys. Rep.,
94(1):1, 1983.
83] S.W. Hawking and G.F.R. Ellis. The Large Scale Structure of Space-Time. Cambridge
University Press, 1973.
84] R.T. Rauch. Equivalence of an R + R2 theory of gravity to Einstein-Cartan-Sciama-
Kibble theory in the presence of matter. Phys. Rev. D, 26(4):931, 1982.

180
85] R.D. Hecht, J. Lemke, and R.P. Wallner. Can Poincare gauge theory be saved? Phys.
Rev. D, 44(8):2442, 1991.
86] A.V. Khodunov and V.V. Zhytnikov. Gravitational equations in space-time with
torsion. J. Math. Phys., 33(10):3509, 1992.
87] R. d'Inverno. Introducing Einstein's Relativity. Oxford University Press, 1992.
88] E.T. Newman and A.I. Janis. Note on the Kerr spinning-particle metric. J. Math.
Phys., 6(4):915, 1965.
89] M.M. Schier, R.J. Adler, J. Mark, and C. Sheeld. Kerr geometry as complexi
ed
Schwarzschild geometry. J. Math. Phys., 14(1):52, 1973.
90] R.P. Kerr. Gravitational
eld of a spinning mass as an example of algebraically
special metrics. Phys. Rev. Lett., 11(5):237, 1963.
91] S. Chandrasekhar. The Mathematical Theory of Black Holes. Oxford University
Press, 1983.
92] W.J. Kaufmann. The Cosmic Frontiers of General Relativity. Penguin Books, 1979.
93] C.W. Misner, K.S. Thorne, and J.A. Wheeler. Gravitation. W.H. Freeman and
Company, 1973.
94] S. Weinberg. Gravitation and Cosmology. John Wiley and Sons, 1972.
95] J.A. Wheeler and R.P. Feynman. Interaction with the absorber as the mechanism of
radiation. Rev. Mod. Phys., 17:157, 1945.
96] J.A. Wheeler and R.P. Feynman. Classical electrodynamics in terms of direct inter-
particle action. Rev. Mod. Phys., 21(3):425, 1949.
97] A. Einstein. The Meaning of Relativity. Princeton University Press, New Jersey,
1945.

181