An Introduction To Clifford Algebras and

www.Ebook777.
com
AN INTRODUCTION TO CLIFFORD
ALGEBRAS AND SPINORS
www.Ebook777.com
www.Ebook777.com
An Introduction
to Clifford Algebras and Spinors
Jayme Vaz, Jr.
IMECC, Universidade Estadual de Campinas, Campinas, SP, Brazil
Roldão da Rocha, Jr.

CMCC – Universidade Federal do ABC, Santo André, SP, Brazil
www.Ebook777.com
3
Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© Jayme Vaz, Jr. and Roldão da Rocha, Jr. 2016
The moral rights of the authors have been asserted
First Edition published in 2016
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2015959276
ISBN 978–0–19–878292–6
Printed and bound by
CPI Group (UK) Ltd, Croydon, CR0 4YY
Links to third party websites are provided by Oxford in good faith and
for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.
To Maria Clara and Liliane
J.
To my Family
R.
Preface
In 1878 William Kingdom Clifford published an article in the American Journal

of Mathematics entitled ‘Applications of Grassmann’s Extensive Algebra’ (Clifford,
1878). Clifford presented a new mathematical structure which he called ‘Geometric
Algebra’, but which is now usually known as Clifford Algebra. Some years earlier,
in 1873, Clifford had published in Proceedings of the London Mathematical Society
an article called ‘Preliminary Sketch of Biquaternions’, where he presented the first
rudiments of his ideas. These later evolved in 1876 in two unpublished manuscripts –
called ‘Further note on so-called biquaternions’ and ‘On the classification of geomet-
ric algebra’, respectively (published posthumously as a single article entitled ‘On the
Classification of Geometric Algebras’ (Clifford, 1882)) – culminating in the ideas pre-
sented in the 1978 article. Briefly, what Clifford accomplished was the synthesis of two
apparently dissociated mathematical structures: the quaternions of Sir William Rowan
Hamilton, and the algebra of extensions (Ausdehnungslehre) of Hermann Grassmann.
We can trace the origin of the Clifford algebras to the efforts to geometrically represent
a complex number.
The idea of unifying geometric and algebraic operations was first advocated by Got-
tfried Leibniz in 1679, in a letter to Christiaan Huygens (this letter was later published
in 1833). This idea was manifested for complex numbers in terms of the Argand, or
Argand–Gauss plane, which has been given the name tribute to the results published
on this topic in 1806 by Jean-Robert Argand and then in 1831 by Carl Friedrich Gauss.
Meanwhile, the first to succeed in representing the complex numbers in a plane was
the Norwegian Caspar Wessel, who presented his results in 1797 to the Royal Danish
Academy of Sciences and Letters, in his work ‘Om Directionens analytiske Betegning’.
This work was published in 1799 but remained in obscurity until it was translated into
French in 1897. The success of the geometrical representation of complex numbers led
Hamilton to search for a generalisation of this approach to three-dimensional space.
After some years of failed attempts, Hamilton succeeded in 1843 with the discovery of
quaternions.1 Afterwards, Hamilton (and his followers) dedicated years to developing
the theory and applications of quaternions. Meanwhile, in 1844 Grassmann published
his outstanding work, Die lineale Ausdehnungslehre, (Grassmann, 1844, 1894). In this
work – influenced by ideas from his father, Justus Grassmann – he introduced an
algebraic system based on a geometric product that, because of its innovative and
abstract character, proved to be difficult to understand at the time. In an attempt
to facilitate the knowledge of his work, Grassmann presented in 1862 a new version
of Die lineale Ausdehnungslehre, which he called Die Ausdehnungslehre: Vollstandig
und in strenger Form bearbeitet (Grassmann, 1862). Nevertheless, Grassmann’s work
1 The present-day formalism of vector algebra was extracted from the quaternion product of two
vectors, by Josiah Willard Gibbs in 1901.
viii Preface
remained obscure for but a few like Clifford. Interestingly, in 1844, after the publi-
cation in 1833 of Leibniz’s 1679 letter to Huygens, the Jablonowskischen Gesellschaft
der Wissenschaft offered an award to those who developed Leibniz’s ideas. The prize
was granted to Grassmann in 1846 for taking Leibniz’s ideas into the realm of the
algebraic system presented in Die lineale Ausdehnungslehre and was later published
as a separate work (Grassmann, 1847). Yet, this award was not enough to draw the
attention of the majority of the scientific community to Grassmann’s work.
More information on the work of Hamilton, Grassmann and other important names
in the history of the vector analysis can be found in, for example, the book by Crown
(1994). Clifford’s major achievement in 1878 was essentially introducing a quaternion
framework into Grassmann’s algebra of extensions, thus obtaining a system naturally
adapted to the orthogonal geometry of an arbitrary space: in this system, quaternions
represent merely a very particular case of a Clifford algebra. Furthermore, one of the
many (at least ten) products defined by Grassmann is the so-called ‘stereometrisches
Produkt’, which implements quaternion multiplication for three generators, and is thus
itself a Clifford product. Unfortunately, Clifford was unable to continue his work as he
died in the following year, at the age of 33. However, it was not long before the Clifford
algebras were reinvented. In 1880, Lipschitz studied representations of rotations by
complex numbers and quaternions, and their generalisations to high dimensions led to
the Clifford geometric algebra and to the group Spin.
Many scientists have contributed to the development of the Clifford algebras. One
of the most important is certainly Élie Cartan. As well as making other contributions,
he described Clifford algebras as algebras of matrices and found the periodicity 8 of
these algebras (the periodicity theorem). Cartan introduced in 1913 the concept of the
spinor, and in 1938 the concept of the pure spinor. Although the term ‘spinor’ had
been coined by Paul Ehrenfest in the 1920s, the intrinsic concept of a spinor is much
older than that, as a spinor is a special linear structure, which had been studied (and
had been used in civil engineering for calculating statics) long before Ehrenfest used
it in quantum theory.
It was through the concept of the spinor that Clifford algebras had a decisive
presence in science. It was prominently Wolfgang Pauli who introduced the concept of
spin to physics, first to explain, for example, the Zeeman effect, and later to formulate
the exclusion principle (his Nobel-prize winning work). Paul Dirac showed in 1928 that
the main equation in relativistic quantum mechanics is written in terms of a Clifford
algebra and that the electron is described by a spinor, or rather, by a spinor field (Dirac,
1928). Cornelius Lanczos rewrote Dirac’s equation in terms of quaternions (Gsponer
and Hurni, 1994) in 1929, and Gustave Juvet and Fritz Sauter in 1930 replaced column
spinors by square matrix spinors, in minimal left ideals, where only the first column
contained non-zero elements (Lounesto, 2001a). Marcel Riesz in 1947 was the first one
to consider spinors as elements of a minimal left ideal in a Clifford algebra. Thereafter,
Feza Gürsey (1956, 1958) expressed the Dirac equation in terms of 2 × 2 quaternionic
matrices. After Kustaanheimo presented the spinor regularisation of Kepler motion in
1964, proposing the so-called Kustaanheimo–Stiefel transformation, David Hestenes
(1966, 1967) reformulated the Dirac theory in this context, providing new aspects
and insights. Since then, a variety of other applications of Clifford algebras have been
Preface ix
found, not only in physical sciences, but also in engineering and computation (Doran
and Lasenby, 2003; Baylis, 1996; Dorst, Fontijne, and Mann, 2007; Perwass, 2008).
This book is divided into six chapters. In the first one, a review of the main concepts
and results from linear algebra is presented, as these are essential in order to develop
the subsequent ideas in this book. Then, tensor algebra is introduced, as it is required
for the definitions of exterior algebras and Clifford algebras. The second chapter is
dedicated to the presentation of exterior algebras and Grassmann algebras. Although
some authors use these terms as synonyms, that will not be the case here. The exterior
algebra is understood in this book as a structure devoid of a metric, whereas Grass-
mann algebras are structures constructed on a vector space endowed with a metric. In
this chapter, some fundamental concepts, such as the exterior product, the contrac-
tion, and the quasi-Hodge isomorphism, and the Hodge isomorphism are introduced.
Clifford algebras are then defined in the third chapter, together with their properties.
Three different definitions of real Clifford algebras are introduced, emphasising the
number of excellent features possessed by Clifford algebras. In the fourth chapter, the-
orems about the structure of Clifford algebras are presented; these theorems are later
used for the classification and representation of the Clifford algebras in terms of ma-
trix algebras. Moreover, we introduce procedures for explicitly building these matrix
representations, as these are important when it comes to computations in physics and
other applications. The groups associated with the Clifford algebras are the objects of
discussion in the fifth chapter. The groups Pin and Spin are comprehensively exam-
ined. In addition, the Lie algebras associated with those groups are discussed and some
of their applications provided. In particular, conformal transformations and twistors
are introduced. The sixth chapter is dedicated to the study of spinors. Three different
definitions of spinors are presented, as well as the properties associated with each def-
inition. In addition, the relations between the different definitions are examined. The
so-called pure spinors are also introduced and the triality principle and the Penrose
flagpole are discussed. A detailed study of the so-called Weyl spinors, which form the
basis of the Penrose and Rindler formalism (Penrose and Rindler, 1984), concludes this
chapter. Finally, in an appendix, the Lorentz transformations and the classical coun-
terparts of dotted/undotted Weyl spinors, the Penrose flagpole and supersymmetry
algebra are presented, in the context of the Van der Waerden framework.
This book is based on lecture notes written by Jayme Vaz Jr for two courses on
Clifford algebras in 1999 and 2005 at the Institute of Mathematics, Statistics and Sci-
entific Computation (IMECC) at the University of Campinas (Unicamp). From 2008
to 2015 Roldão da Rocha Jr used and improved the lecture notes, in particular for
the courses ‘Clifford Algebras and Spinors’, ‘Spinors in Hilbert Spaces’, and ‘Quan-
tum Field Theory’ for graduate programs at the Center of Mathematics, Computation
and Cognition and at the Center of Natural and Human Sciences at ABC Federal
University, Santo André, Brazil, in the process of completing and ameliorating the
manuscript, as well as adding topics. Those who are familiar with the Clifford alge-
bras and their applications certainly will realise that an important topic, the Dirac
operators, which play a significant role in many areas of modern mathematics and
physics, is missing. The reason for its absence is that, although a discussion of the
Dirac operators would have been welcome in the course, it was beyond the scope of
x Preface
these lecture notes, which aimed to provide material for a 30-hour course. For a study
of Dirac operators, we recommend Gilbert and Murray (1991) book, which contains
a complete introduction to Clifford algebras as well as to representations of the Spin
group. Moreover, it defines Dirac-type operators in analysis and geometry, as well as
providing their applications. Examples of Dirac-type operators include, for instance,
the Hodge–Dirac operator on a Riemannian manifold, the Laplacian in Euclidean
spaces, and the Atiyah–Singer–Dirac operator on a spin manifold. Therefore, the local
Atiyah–Singer index theorem can be demonstrated. Moreover, Gilbert and Murray
(1991) book offers a systematic and comprehensive presentation of spinor fields, Dirac
operators, and the Atiyah–Singer index theorem.
Acknowledgements
We would like to thank all colleagues and friends who in some way or other helped us
in the preparation of this work. In particular, we wish to thank R. T. Cavalcanti, R. A.
Mosna, E. C. de Oliveira, W. A. Rodrigues Jr, and M. A. Traesel for many important
discussions and suggestions. We also thank S. Adlung from OUP for all support in
the publishing process, and the referees for their invaluable advice and suggestions.
Finally, we acknowledge the support of the CNPq, CAPES Foundation, and FAPESP
research agencies over the years during which this work was accomplished.
Contents
1 Preliminaries 1
1.1 Vectors and Covectors 1
1.2 The Tensor Product 10
1.3 Tensor Algebra 16
1.4 Exercises 19
2 Exterior Algebra and Grassmann Algebra 21
2.1 Permutations and the Alternator 21
2.2 p-Vectors and p-Covectors 22
2.3 The Exterior Product V 24
2.4 The Exterior Algebra (V ) 29
2.5 The Exterior Algebra as the Quotient of the Tensor Algebra 33
2.6 The Contraction, or Interior Product 36
2.7 Orientation, and Quasi-Hodge Isomorphisms 41
2.8 The Regressive Product 47
2.9 The Grassmann Algebra 48
2.10 The Hodge Isomorphism 50
2.11 Additional Readings 54
2.12 Exercises 54
3 Clifford, or Geometric, Algebra 57
3.1 Definition of a Clifford Algebra 57
3.2 Universal Clifford Algebra as a Quotient of the Tensor Algebra 60
3.3 Some General Considerations 66
3.4 From the Grassmann Algebra to the Clifford Algebra 73
3.5 Grassmann Algebra versus Clifford Algebra 78
3.6 Notation 83
3.8 Exercises 85
4 Classification and Representation of the Clifford Algebras 87
4.1 Theorems on the Structure of Clifford Algebras 87
4.2 The Classification of Clifford Algebras 93
4.3 Idempotents and Representations 102
4.4 Clifford Algebra Representations 108
4.6 Exercises 120
xiv Contents
5 Clifford Algebras, and Associated Groups 121

5.1 Orthogonal Transformations and the Cartan–Dieudonné
Theorem 121
5.2 The Clifford–Lipschitz Group 126
5.3 The Pin Group and the Spin Group 131
5.4 Conformal Transformations in Clifford Algebras 138
5.6 Exercises 144
6 Spinors 145
6.1 The Babel of Spinors 145
6.2 Algebraic Spinors 148
6.3 Classical Spinors 149
6.4 Spinor Operators 152
6.5 A Comparison of the Different Definitions of Spinors 158
6.6 The Inner Product in the Space of Algebraic Spinors 165
6.7 The Triality Principle in the Clifford Algebraic Context 170
6.8 Pure Spinors 179
6.9 Dual Rotations, and the Penrose Flagpole 186
6.10 Weyl Spinors in Cℓ3,0 190
6.11 Weyl Spinors in the Clifford Algebra Cℓ0,3 ≃ H ⊕ H 193
6.12 Spinor Transformations 194
6.13 Spacetime Vectors as Paravectors of Cℓ3,0 from Weyl Spinors 196
6.14 Paravectors of Cℓ4,1 in Cℓ3,0 via the Periodicity Theorem 199
6.15 Twistors as Geometric Multivectors 200
6.16 Spinor Classification According to Bilinear Covariants 203
6.18 Exercises 206
Appendix A The Standard Two-Component Spinor Formalism 209
A.1 Weyl Spinors 214
A.2 Contravariant Undotted Spinors 214
A.3 Covariant Undotted Spinors 215
A.4 Contravariant Dotted Spinors 215
A.5 Covariant Dotted Spinors 216
A.6 Null Flags and Flagpoles 217
A.7 The Supersymmetry Algebra 223
Appendix B List of Symbols 224
References 231
Index 239
1
Preliminaries
In this chapter, we briefly review essential concepts regarding vector spaces and their
associated dual vector spaces. Moreover, the tensor algebra, including its universal
character, is introduced, with some computational examples as well. Moreover, we fix
the notation to be used throughout the text. Some demonstrations can be also found
in the standard literature – for a good reference, see for example the work by Hoffman
and Kunze (1971), Birkhoff and MacLane (1997), or Kostrykin and Manin (1997).
1.1 Vectors and Covectors

The concept of vector space is fundamental throughout the text, and so deserves to
be discussed in detail. A vector space V over a field K (from here on, elements of a
field are called scalars) is a set of elements called vectors, endowed with an additive
operation +: V × V → V and also equipped with a product by scalars . : K × V → V ,
satisfying the following properties:
(1) With each pair of vectors u, v ∈ V , an element u + v ∈ V is associated, denom-
inated the sum of u and v, with the properties:
(a) commutativity: u + v = v + u, for all u, v ∈ V ;
(b) associativity: u + (v + w) = (u + v) + w, for all u, v, w ∈ V ;
(c) there exists a vector denoted by 0, named the null vector, such that u + 0 = u for
all u ∈ V ;
(d) for each u ∈ V , there exists a unique vector denoted by (−u) such that u + (−u) =
0.
(2) For each pair (a, u) ∈ K × V , there exists a vector denoted by a.u ∈ V , satisfying
the properties:
(a) associativity: a.(b.u) = (ab).u, for all a, b ∈ K and u ∈ V , where ab is the product
of a and b in K;
(b) 1.u = u for all u ∈ V , where 1 denotes the unity of K;
(c) distributivity: a.(u + v) = a.u + a.v, for all a ∈ K, and all u, v ∈ V ;
(d) distributivity: (a + b).u = a.u + b.u, for all a, b ∈ K, and u ∈ V .
This text concerns vector spaces V over the field R of real numbers and over the
field C of complex numbers as well and, in some specific cases, we consider modules
over the quaternions H.1 Some attention is needed when H is regarded, to circumvent
1 The quaternions H are defined by the set of elements {a + bi + cj + dk | a, b, c, d ∈ R}, with the
associative product defined as ij = −ji = k; jk = −kj = i; ki = −ik = j; and i2 = j 2 = k2 = −1
(Hamilton, 1866).
An Introduction to Clifford Algebras and Spinors. First Edition. Jayme Vaz, Jr. and Roldão da Rocha, Jr.
© Jayme Vaz, Jr. and Roldão da Rocha, Jr. 2016. Published in 2016 by Oxford University Press.
2 Preliminaries
some subtleties, since multiplication by scalars (in this case, elements of H) is not com-
mutative. Consequently, both the left and the right multiplications must be evaluated.
Such details, however, will not be discussed here, being postponed to section 6.10. We
restrict our discussion to vector spaces over R for most of our purposes, taking into
account that C is a straightforward generalisation that is sometimes simpler in the
context of Clifford algebras than in other contexts. In addition, finite-dimensional vec-
tor spaces, where we denote dim(V ) = n, will be examined. It is worth emphasising
that various demonstrations throughout the book are general and also hold for infinite-
dimensional spaces. From now on, we denote the product a.v by the juxtaposition av.
Moreover, unless otherwise stated, we use the Einstein summation convention. For
instance, if B = {e1 , . . . , en } is a basis for V , then
n
X
v= v i ei = v i ei .
i=1
Not so much discussed as the vector space, although as important and inseparable
from it, is the dual vector space. A mapping α : V → K is said to be a linear functional
if it is a homomorphism of vector spaces,2 namely if it satisfies
α(au + v) = aα(u) + α(v),
for all a ∈ K and for all u, v ∈ V . A linear functional is also known as a covector,
and the latter term will be preferentially used throughout the text. When the sum of
covectors
(α + β)(v) = α(v) + β(v)
and the multiplication by scalars
(aα)(v) = a(α(v))
are defined, the space of covectors is endowed with a vector space structure. Such a
covector space is called the dual vector space associated with the vector space V , and
it is denoted by V ∗ .
The evaluation of the covector α on a vector v = v i ei is performed by
α(v) = α(vi ei ) = v i α(ei ) = v i αi ,
where αi = α(ei ). This expression reveals that a covector α is completely defined by

its components αi with respect to a basis B = {ei } of the vector space V .
2 A homomorphism is a mapping of a structure X into a structure Y such that the structural

properties in the domain are preserved in the image. More explicitly, if ∗ is an operation in X and
• is another operation in Y , then ϕ : X → Y is a homomorphism when ϕ(a ∗ b) = ϕ(a) • ϕ(b). If
Y = X, then this homomorphism is called an endomorphism.
Vectors and Covectors 3
Consider now the set of covectors {ei } (i = 1, . . . , n) defined by

(
i 1 if i = j
e (ej ) = δji = . (1.1)
0 6 j
if i =
The set {ei } in eqn (1.1) is a basis for the dual vector space V ∗ . The basis B∗ = {ei }
is said to be the dual basis associated with the basis B = {ei }. As an immediate
consequence, it follows that
dim(V ∗ ) = dim(V ). (1.2)
Covariant Transformations and Contravariant Transformations

The intrinsic difference between vectors and covectors can be explored by considering,
for instance, the effect of a change of basis. Let us consider a change of basis B 7→ B′
as described by e′j = B i j ei . A vector v ∈ V has components {vi } with respect to the
basis B, and components {v ′i } with respect to the basis B′ , namely v = v i ei = v ′i e′i .
The vector components are related by v j = B j i v′i .
Now, let B∗ = {ei }, and B′∗ = {e′i }, be the dual bases respectively associated
with the bases B = {ei }, and B′ = {e′i }. By definition, we have ei (ej ) = e′i (e′j ) = δji .
The components of α ∈ V ∗ in the bases B∗ and B′∗ are given by the evaluation of
α on the bases B and B′ , respectively. Hence, α = αi ei = α′i e′i , where αi = α(ei ),
and α′i = α(e′i ). Since e′j = B i j ei , the components of α transform as αj′ = B i j αi , and
consequently the dual bases are related by ej = B j i e′i .
To summarise, with a change of basis, the components of a covector transform as
the basis vectors
e′j = B i j ei , α′j = B i j αi ,
whereas the components of a vector transform as the basis covectors
ej = B j i e′i , v j = B j i v ′i .
The inverse transformations are given by
ej = (B −1 )i j ei , αj = (B −1 )i j αi ,
and
e′j = (B −1 )j i ei , v ′j = (B −1 )j i v i ,
accordingly.
The transformations regarding these basis vectors are usually called covariant
transformations, and the transformations of the vector components are called con-
travariant transformations. Such a denomination asserts that the dual basis vectors
transform in a contravariant way, and the covectors components transform in a covari-
ant way.
Example 1.1 Let B = {ei } be the standard basis of R3 , namely e1 = (1, 0, 0)⊺ ; e2 = (0, 1, 0)⊺ ; e3 =
(0, 0, 1)⊺ ; and let B′ = {e′i } be another basis such that e′1 = (−1, −1, 1)⊺ ; e′2 = (−1, 0, 1)⊺ ; and
4 Preliminaries
e′3 = (2, 1, −1)⊺ . Given a vector v = v i ei = v ′i e′i , the components {v i } and {v ′i } with respect to these
bases are expressed as
 1    ′1   ′1     1
v −1 −1 2 v v 1 −1 1 v
v 2  = −1 0 1  v ′2  , v ′2  = 0 1 1 v 2  ,
v3 1 1 −1 v ′3 v ′3 1 0 1 v3
and the bases vectors are related by
e′1 = −e1 − e2 + e3 , e1 = e′1 + e′3 ,
e′2 = −e1 + e3 , e2 = −e′1 + e′2 ,
e′3 = 2e1 + e2 − e3 , e3 = e′1 + e′2 + e′3 .
Let B∗ = {ei } be the dual basis related to B. By writing the vectors {ei } as
     
1 0 0
e1 = 0 , e2 = 1 , e3 = 0 ,
0 0 1
the dual basis reads
e1 = 1 0 0 , e2 = 0 1 0 , e3 = 0 0 1 .
Let now B′∗ = {e′i } be the dual basis associated with B′ ; therefore, e′i (e′j ) = δji , and in particular
e′1 (e′1 ) = 1; e′1 (e′2 ) = e′1 (e′3 ) = 0; and so on. In order to find a relationship among the dual basis
vectors, it is possible to evaluate the action of the covector e′i on the vectors ej , expressed in terms of
B′ . For instance,
e′1 (e1 ) = e′1 (e′1 + e′3 ) = e′1 (e′1 ) + e′1 (e′3 ) = 1,
e′1 (e2 ) = e′1 (−e′1 + e′2 ) = −e′1 (e′1 ) + e′1 (e′2 ) = −1,
e′1 (e3 ) = e′1 (e′1 + e′2 + e′3 ) = e′1 (e′1 ) + e′1 (e′2 ) + e′1 (e′3 ) = 1,
where we can conclude that e′1 = e1 − e2 + e3 . An analogous procedure can be used to express the set
{ei } with respect to the basis B′∗ . The result yields
e′1 = e1 − e2 + e3 , e1 = −e′1 − e′2 + 2e′3 ,
e′2 = e2 + e3 , e2 = −e′1 + e′3 ,
1 3
e′3
= e +e , e3 = e′1 + e′2 − e′3 .
Consider now the linear functional α acting on a vector v:
α(v) = α1 v 1 + α2 v 2 + α3 v 3 ,
where the set {αi } contains the components of α with respect to the basis B∗ . If the components of v
are placed in a row matrix, then the components of the linear functional α can be written as the column
matrix
[α]B∗ = α1 α2 α3 ;
therefore,  1
v2
α(v) = α1 α2 α3 v  = α1 v 1 + α2 v 2 + α3 v 3 .
v3
As usual, by denoting by α′i the components of α with respect to the basis B′∗ (α = αi ei = α′i e′i ), the
respective components are related by:
 
1 −1 1
′ ′
α1 α2 α3 = α1 α2 α3 ′  0 1 1 ,
1 0 1
 
−1 −1 2
α1 α2 α3 = α1 α2 α3 −1 0 1  .
′ ′ ′
1 1 −1
When a matrix is multiplied from the right by a column vector, we relate the components of a vector in a
basis A as components with respect to a basis B. Moreover, when a matrix is multiplied from the left by
a row vector, the components of a covector in a basis B are related to the components with respect to a
basis A.
Example 1.2 Let F be the space of continuous functions f : R → R. The integral

Z x1
L(f ) = f (x) dx
x0
defines a linear functional L over F . Consider the subset P2 of F and which consists of the polynomial
functions P of degree less than or equal to 2 and is expressed as P (x) = a + bx + cx2 , where a, b, c ∈ R.
The canonical basis for P2 is B = {1, x, x2 }, and we denote
e1 = 1, e2 = x, e3 = x2 .
Let us define the following linear functionals on P2 :
Z 1 Z 2 Z 3
L1 (P ) = p(x) dx, L2 (P ) = p(x) dx, L3 (P ) = p(x) dx.
0 0 0
Explicitly, these integrals read
1 1 8 9
L1 (P ) = a + b + c, L2 (P ) = 2a + 2b + c, L3 (P ) = 3a + b + 9c.
2 3 3 2
If {ei } is the dual basis of {ei }, then these equations yield
1 1 8 9
L1 = e1 + e2 + e3 , L2 = 2e1 + 2e2 + e3 , L3 = 3e1 + e2 + 9e3 .
2 3 3 2
Let now {Li } be the basis related to the dual basis {Li }, so that Li (Lj ) = δji . Since Li = ej Li (ej ), and
ej = Li (ej )Lj , the expressions for {ei } in terms of {Li } are thus given by
1 9 1 8
1 = e1 = L1 + 2L2 + 3L3 , x = e2 = L1 + 2L2 + L3 , x2 = e3 = L1 + L2 + 9L3 .
2 2 3 3
The inverse relation can be obtained and derived by usual manipulation, which results in
3 3 3 1 1
L1 = 3 − 5x + x2 , L2 = − + 4x − x2 , L 3 = − x + x2 .
2 2 2 3 2
Since Li = ej (Li )ej , and ej = ej (Li )Li , from these expressions, {ei } can be expressed from {Li } as
3 1 3 3 1
e1 = 3L1 − L2 + L3 , e2 = −5L1 + 4L2 − L3 , e3 = L1 − L2 + L3 .
2 3 2 2 2
Finally, an arbitrary functional L, Z x1
L(P ) = p(x) dx,
x0
can be, for instance, written in the form
L = λ1 e1 + λ2 e2 + λ3 e3 = l1 L1 + l2 L2 + l3 L3 ,
where
x21 − x20 x31 − x30
λ1 = (x1 − x0 ), λ2 = , λ3 = ,
2 3
and the components in the set {li } are given by
 
3 −3/2 1/3
l1 l2 l3 = λ1 λ2 λ3  −5 4 −1  .
3/2 −3/2 1/2
The Bidual
Since V ∗ is a vector space, one can argue whether it is also possible to define a
dual space associated with the dual vector space V ∗ . Define the linear functionals
ξv : V ∗ → R as
ξv (α) = α(v). (1.3)
The sum of such linear functionals is given by ξv + ξu = ξ(v+u) , and the multiplication
by a scalar by aξv = ξ(av) . Such space is denoted by (V ∗ )∗ and dim((V ∗ )∗ ) = dim(V ).
6 Preliminaries
The fundamental result here is that the spaces V and (V ∗ )∗ are isomorphic. Although
it is well known that all vector spaces having the same dimension are isomorphic, in
this context, the assertion is stronger than usual: there exists a natural (namely, a
canonical identification) isomorphism between V and (V ∗ )∗ , given by ξ and which is
the mapping ξ : V → (V ∗ )∗ , and ξ : v 7→ ξv . By ‘natural’ we mean that such an
isomorphism does not depend upon the choice of a basis. In fact, eqn (1.3) defines the
natural isomorphism, since α(v) is a scalar, as being invariant under basis change.
However, there is no natural isomorphism (in this sense) between a vector space V
and its respective dual V ∗ . An additional structure is needed to define the isomorphism
between V and V ∗ ! This additional structure is called a correlation. In other words,
an isomorphism between V and V ∗ must be chosen.
The Kernel and the Image of a Linear Transformation

Given two sets X and Y , a mapping f : X → Y is said to be a surjective (or onto)
mapping if, for each y ∈ Y , there exists an element x ∈ X such that f (x) = y.
However, the mapping is said to be an injective (or one-to-one) mapping if, for each
y ∈ Y , there exists at most a unique element x ∈ X such that f (x) = y.3 A mapping
that is onto and one-to-one is said to be bijective.
Consider now the two vector spaces V and W and let f be a linear mapping
f : V → W . The rank of a linear mapping f , denoted by rank f , is the dimension
of the image of the mapping (the image is itself a vector subspace f (V ) ⊆ W ). The
kernel of f (denoted by ker f ) consists of the set {v ∈ V | f (v) = 0}. The dimension
of the kernel of f is also called the nullity of f . For finite-dimensional vector spaces,
the expression
rank f + dim ker f = dim V
holds. It is straightforward to show that a linear mapping f is injective if and only if
ker f = {0}. Moreover, when V and W are vector spaces of the same finite dimension,
any one-to-one surjective linear mapping f : V → W is an isomorphism – and, in this
case, ker f = {0}. For spaces of infinite dimension, we must also impose the mapping
f to be onto (surjective), for the definition of isomorphism.
Correlations and Bilinear Functionals

We can introduce the linear mapping τ : V → V ∗ , the so-called correlation, which
naturally defines a bilinear functional B : V × V → R by the equation
B(v, u) = τ (v)(u), ∀ u, v ∈ V. (1.4)
Bilinearity means linearity in each one of the entries, namely B(av+u, w) = aB(v, w)+
B(u, w), and B(v, au + w) = aB(v, u) + B(v, w), ∀ u, v, w ∈ V , and a, b ∈ R.
If ker τ = {0}, the correlation is said to be non-degenerate. In this case, the vec-
tor space V and the bilinear functional associated with τ are also said to be non-
degenerate. It is possible to show that the bilinear functional B is non-degenerate if
and only if, for each vector v 6= 0, there exists a vector u 6= 0 such that B(v, u) 6= 0.
3 There can be a situation wherein there does not exist any element.
Since dim V = dim V ∗ , if ker τ = {0}, then τ is an isomorphism; thus, the non-
degenerate correlation provides an isomorphism between a vector space and its respec-
tive dual space.
Quadratic Spaces and Symplectic Spaces

A bilinear functional, in this case denoted by g, is said to be symmetric if
g(v, u) = g(u, v), ∀ u, v ∈ V. (1.5)
The correlation τ associated with g and defined by τ (v)(u) = g(v, u) satisfies τ (v)(u) =
τ (u)(v). A vector space endowed with a non-degenerate symmetric bilinear functional
is said to be a quadratic space. A symmetric bilinear functional is completely deter-
mined by the quadratic form Q(v) = g(v, v) via the so-called polarisation procedure.
Indeed, by using the bilinearity property to calculate Q(v + u) = g(v + u, v + u), we
can write
1
g(v, u) = (Q(v + u) − Q(v) − Q(u)) . (1.6)
2
A bilinear functional, in this case denoted by σ, is said to be antisymmetric when
it satisfies the property
σ(v, u) = −σ(u, v). (1.7)
The correlation τ in this case satisfies τ (v)(u) = −τ (u)(v). A vector space endowed
with a non-degenerate antisymmetric bilinear functional is said to be a symplectic
space.
If g(v, u) = 0, the vectors v and u are said to be orthogonal with respect to g,
when g is non-degenerate. A non-trivial vector v can be orthogonal to itself, namely
g(v, v) = 0, and such vectors are called isotropic. It is immediately obvious that, in a
symplectic space, all the vectors are isotropic.
Musical Isomorphisms
In this section, only quadratic spaces are considered with respect to the symmetric
correlations τ : V → V ∗ , and τ −1 : V ∗ → V . Here, we shall use a notation which
is much more convenient than the one we have used so far: we shall denote those
correlations respectively by
♭ : V → V ∗, ♯ : V ∗ → V, (1.8)
where ♭ = ♯−1 , and ♯ = ♭−1 . Such isomorphisms are called musical isomorphisms. In
general, the expression
v♭ = ♭(v), α♯ = ♯(α) (1.9)
will be used. By definition, it follows that
v♭ (u) = g(v, u), g(α♯ , v) = α(v) . (1.10)
For the vectors v = v i ei , and u = ui ei , we can write g(v, u) = gij v i uj , where gij =
g(ei , ej ) = gji . As v♭ (u) = v♭i ui , where v♭i represents the covector v♭ components in
the basis {ei }, namely v♭ = v♭i ei , it follows that
8 Preliminaries
v♭i = gij v j . (1.11)
Similarly,
ei♭ = gij ej . (1.12)
On the other hand, by writing α♯ = αi ei♯ , it follows from g(α♯ , v) = α(v) that
αk g(ek♯ , ei )v i = αi v i ; thus,
ei♯ = g ij ej , (1.13)
where g ij = g ji is defined as
g ik gkj = δji . (1.14)
This ‘numerical inverse’ is not an inverse mapping, but an adjoint. It is also possible
to write
α♯i = g ij αj , (1.15)
where α♯i is the ith component of the vector α♯ in the basis {ei }.
Observation ☞ In some texts, these equations read
vi = gij v j , ei = gij ej , (1.16)
and
αi = g ij αj , ei = g ij ej . (1.17)
In such texts, it is common to assert that such equations are responsible for the indices
raising and lowering, respectively. However, rigorously speaking, these equations make
no sense, since g ij is not the inverse (as a mapping) of gij (although it is a numerical
inverse with respect to both a chosen basis and its dual basis); for instance, there is
a vector at the left-hand side of the equation ei = gij ej but a linear combination
of covectors at the right-hand side. Like the mapping gij : R → R, these functions
could be defined properly, for example by using a closed structure on the category of
finite-dimensional vector spaces over R; however this approach outside the scope of
this book.
However, as ♯ and ♭ are isomorphisms and sometimes isomorphism symbols can be
concealed in order to avoid a heavy notation, we shall use the notation outlined in
this section. Indeed, the explicit use of the indices in these equations should virtually
eliminate any confusion there might otherwise be. However, when there is no reference
to any particular basis, there is no way to avoid using notation using ♯ and ♭.
Example 1.3 Let B = {ei } (i = 1, 2, 3) be the standard basis of R3 and let B∗ = {ei } be its respective
dual basis. Let us define a correlation ♭ as
e1 ♭ = 3e1 + e2 , e2 ♭ = e1 + 3e2 , e3 ♭ = 2e3 .
If we represent a vector v = viei by a column matrix, this correlation reads
 1
v
♭ v 2  = (3v 1 + v 2 ) (3v 2 + v 1 ) (2v 3 ) .
v 3
This correlation is non-degenerate, in view of ker ♭ = {0}. Indeed, if ♭(v) = 0, then 3v 1 + v 2 = 0;

3v 2 + v 1 = 0; and 2v 3 = 0; thus v 1 = v 2 = v 3 = 0. Moreover, the correlation ♭ is symmetric, since
 1
u
v♭ (u) = (3v 1
+ v2 ) (3v 2
+ v1 ) u2  = 3v 1 u1 + v 2 u1 + 3v 2 u2 + v 1 u2 + 2v 3 u3
(2v 3 )
u3
 1
v
= (3u1 + u2 ) (3u2 + u1 ) (2u3 ) v 2  = u♭ (v).
v3
Therefore, a symmetric bilinear form g can be defined as g(v, u) = v♭ (u). Since

 
310
v♭ = v v v 1 3 0 ,
1 2 3
002
it follows that g can be represented in this basis by the matrix

 
310
1 3 0 .
002
On the other hand, the inverse correlation ♯ is given by

1 1 1
e1♯ = (3e1 − e2 ), e2♯ = (3e2 − e1 ), e3♯ = e3 ,
8 8 2
which can be represented by  
(3α1 − α2 )/8
♯ α1 α2 α3 = (3α2 − α1 )/8 .

α3 /8
Since     
(3α1 − α2 )/8 3/8 −1/8 0 α1
(3α2 − α1 )/8 = −1/8 3/8 0  α2  ,
α3 /2 0 0 1/2 α3
it follows that, in this basis the symmetric bilinear form g −1 (here it must be clear that this inverse is
taken in the group of automorphisms Aut(R3 )) such that g −1 (α, β) = α(β ♯ ) = β(α♯ ) = g −1 (β, α) can
be represented by  
3/8 −1/8 0
−1/8 3/8 0  .
0 0 1/2
As expected, the matrices g and g −1 are inverse with respect to each other, as can be clearly seen. Indeed,
the elements gij which are entries of the matrix representation of g, and the elements g ij of the matrix
which represents g −1 , must satisfy eqn (1.14); therefore, in this case, the matrices are the inverse of each
other.
In addition, the basis {ei } is not orthogonal with respect to g, inasmuch as g(e1 , e2 ) = g(e2 , e1 ) =
1 6= 0. Frequently, it is convenient to work with an orthogonal basis, as in this example. In order to obtain
such an orthogonal basis, the eigenvalues of g are calculated when the characteristic equation
3−r 1 0
1 3−r 0 = 0,
0 0 2−r
is solved, having r1 = 2; r2 = 2; and r3 = 4 as solutions. The corresponding eigenvectors are

     
1 0 1
ξ 1 = −1 , ξ 2 = 0 , ξ3 = 1 ,
0 1 0
so the matrix T that diagonalises g, and its inverse T −1 , are respectively represented by
   
1 01 1/2 −1/2 0
T = −1 0 1 , T −1
= 0 0 1 .
0 10 1/2 1/2 0
10 Preliminaries
The set of vectors {ξi } is an orthogonal basis. In addition, an orthonormal basis can be constructed as
g(ξ1 , ξ1 ) = 4, g(ξ2 , ξ2 ) = 2, g(ξ3 , ξ 3 ) = 8,
and the orthonormal basis B′ = {e1 , e2 , e3 } is forthwith elicited by
     
1 1 1 1 0 1 1 1 1
′
e1 =  
−1 = (e1 − e2 ), e2 = √ ′  0 = √ e2 , e3 = √ 1 = √ (e1 + e2 ).
 ′
2 0 2 2 1 2 2 2 0 2 2
1.2 The Tensor Product

Let us consider the vector spaces U , V , and W . Lin(V, W ) denotes the space of the
linear mappings from V to W , and Lin(2) (V, W ; U ) denotes the space of the bilinear
mappings from V × W to U . The tensor product of the two vector spaces (over the
same field) V and W can be defined from the bilinear transformations φ : V × W → U .
One of these transformations is universal, in the sense that it describes all others. A
bilinear mapping φ : V ⊗W → U is said to be universal if the following linear mapping
is bijective for every vector space S:
Lin(U, S) → Lin(2) (V, W ; S), f 7→ f ◦ φ .
It is straightforward to apply this definition to vector spaces of finite dimensions.
Theorem 1.1 ◮ If {e1 , . . . , en } is a basis of V and {e′1 , . . . , e′p } is a basis of W , the

following two assertions are equivalent:
(a) the bilinear mapping φ : V ⊗ W → U is universal;
(b) the set of np elements {φ(ej , e′k )} constitutes a basis of U .
Proof: Remember that a family {ǫ1 , . . . , ǫm } of elements of U is a basis if and only if
the following linear mapping is bijective for every vector space S:
Lin(U, S) → S × S × · · · × S = S m , f 7→ {f (ǫ1 ), . . . , f (ǫm )} .

| {z }
m times
Now, let us consider this sequence of two linear mappings:
Lin(U, S) → Lin(2) (V, W ; S) → S np ,
where the first arrow is defined as above, and the second arrow maps every F to the
family of all F (ej , e′k ). The second arrow is well known to be bijective. Therefore,
the bijectiveness of the first arrow, namely the universality of φ, is equivalent to the
bijectiveness of the mapping Lin(U, S) → S np , which maps every f to the family of
all f (φ(ej , e′k )). The bijectiveness of Lin(U, S) → S np means that the set of elements
{φ(ej , e′k )} constitutes a basis of U . ✓
An immediate corollary of this theorem states that property (b) is true for every
basis of V and every basis of W as soon as it is true for a particular basis of V and a
particular basis of W .
The Tensor Product 11
In particular, given a bilinear mapping φ : V ×W → U , there exists a unique linear

transformation ψ : V ⊗ W → U such that
φ(v, w) = ψ(v ⊗ w), v ∈ V, w ∈ W. (1.18)
We can also denote the tensor product by V ⊗K W when we explicitly want to
emphasise on which field the tensor product is performed. However, unless otherwise
stated, we shall not use this notation; instead, from here on, V ⊗ W will be used to
denote the tensor product when K = R.
The tensor product between covectors can be defined mutatis mutandis, and the
tensor product defines a bilinear functional acting on the Cartesian product V × V by
(α ⊗ β)(v, u) = α(v)β(u). (1.19)
The quantity α⊗β is denominated the tensor product between α and β. Consequently,
from this definition, the tensor product is not commutative, in general:
α ⊗ β 6= β ⊗ α.
It is also possible to define a tensor product between vectors acting on covectors. In this
case, the tensor product between vectors must be a quantity that acts on the Cartesian
product V ∗ × V ∗ and results in a scalar. Indeed, we can use the natural isomorphism
between V and the bidual V ∗∗ and, since ξv (α) = α(v), the tensor product v ⊗ u can
be defined as
(v ⊗ u)(α, β) = α(v)β(u). (1.20)
The expression (v ⊗ u)(α, β) formally means that (ξv ⊗ ξu )(α, β) = ξv (α)ξu (β) =
α(v)β(u), since V ∗∗ and V are canonically isomorphic.
The space defined by the tensor product between covectors is itself a vector space,
denoted by T2 (V ) = V ∗ ⊗ V ∗ . Similarly, the space defined by the tensor prod-
uct between vectors is also a vector space, denoted by T2 (V ) = V ⊗ V . It follows
immediately that T2 (V ) ≃ T2 (V ∗ ); this result implies that, if dim V = n, then
dim T2 (V ) = dim T2 (V ) = n2 . The spaces T2 (V ) and T2 (V ) can be straightforwardly
related by the correlation.
Example 1.4 Let v, u ∈ R2 be the vectors v = 2e1 − e2 and u = e1 + 3e2 . Then, the following tensor
products can be evaluated as
v ⊗ u = 2e1 ⊗ e1 + 6e1 ⊗ e2 − e2 ⊗ e1 − 3e2 ⊗ e2 ,
u ⊗ v = 2e1 ⊗ e1 − e1 ⊗ e2 + 6e2 ⊗ e1 − 3e2 ⊗ e2 .
This result explicitly shows that the tensor product is, in general, not commutative: v ⊗ u 6= u ⊗ v. Indeed,
v ⊗ u − u ⊗ v = 7(e1 ⊗ e2 − e2 ⊗ e1 ).
Moreover, v ⊗ u = u ⊗ v if and only if u and v are collinear. Indeed, P if u iand v are not collinear, there
is a basis {e1 , . . . , en } such that e1 = u. We can then write v = n i=1 v ei and, if u ⊗ v = v ⊗ u, it
follows that
n
X n
X
0 = (v 1 − v 1 )e1 ⊗ e1 + v i e1 ⊗ ei − v i ei ⊗ e1 .
i=2 i=2
This result implies that v i = 0 for i = 2, 3, . . . , n and that v 1 can be arbitrary. Hence, v and u are
collinear. Conversely, if u and v are collinear, there exists λ ∈ K such that u = λv, and therefore
u ⊗ v = λv ⊗ v = v ⊗ (λv) = v ⊗ u .
12 Preliminaries
Bases
Let B = {ei } be a basis for V and let B∗ = {ei } be its dual basis. We know that
α(v) = αi v i , and β(u) = βi ui , where αi = α(ei ); βi = β(ei ); v i = ei (v); and
ui = ei (u). In this way,
(α ⊗ β)(v, u) = αi v i βj uj .
On the other hand, we have
(ei ⊗ ej )(v, u) = v i uj .
Comparing these equations, we can write
α ⊗ β = αi βj ei ⊗ ej .
The set of bilinear functionals {ei ⊗ ej } (i, j = 1, . . . , n) is a basis for the space
2
T (V ). If B is an arbitrary bilinear functional, it follows that
B = bij ei ⊗ ej , (1.21)
where the scalars bij , which are the components of B in this basis B, are given by
bij = B(ei , ej ).
Consequently,
B(v, u) = bij v i uj .
Analogously, the set {ei ⊗ ej } (i, j = 1, . . . , n) forms a basis for the space T2 (V ),
namely A ∈ T2 (V ), which can be written as
A = aij ei ⊗ ej , (1.22)
where aij = A(ei , ej ) are components of A in this basis.

The bilinear functionals in T2 (V ) and T2 (V ) can be decomposed into the sum of
symmetric and antisymmetric bilinear functionals. Given B ∈ T2 (V ), a symmetric
bilinear functional Bsym can be written as
B(v, u) + B(u, v)
Bsym (v, u) = ,
2
and an antisymmetric bilinear functional Balt can be expressed as
B(v, u) − B(u, v)
Balt (v, u) = .
2
Indeed, it is always possible to decompose B = Bsym + Balt .
Still considering the space of the symmetric bilinear functionals T2sym (V ) and the
space of the antisymmetric bilinear functionals T2alt (V ), it is clear that the sets
{ei ⊗ ej + ej ⊗ ei } and {ei ⊗ ej − ej ⊗ ei }
are, respectively, bases for T2sym (V ) and T2alt (V ).

Mixed Bilinear Functionals

It is also possible to define a tensor product between a vector and a covector; this
product is called a mixed bilinear functional. We denote this vector space by T11 (V ) =
V ∗ ⊗V . Obviously, dim T1 1 (V ) = n2 . A basis for T1 1 (V ) is given by the tensor products
{ei ⊗ ej } . Thus, a tensor C ∈ T1 1 (V ) can be written as
C = ci j ei ⊗ ej , (1.23)
where ci j = C(ei , ej ).
Similarly, we can define the tensor product V ⊗ V ∗ , denoting it by T1 1 (V ). An
element D ∈ T11 (V ) therefore can be written as
D = di j ei ⊗ ej , (1.24)
i i
where d j = D(e , ej ).
There exists an isomorphism between the spaces T1 1 (V ) and T1 1 (V ). However,
correct positioning of the indices is necessary in order to avoid ambiguity about which
space we are considering when there is a correlation between lowering and raising
indices (see Example 1.5). By convention, we do not take into account the position of
the indices, and the notation T11 (V ) refers to the space T1 1 (V ).
Example 1.5 The isomorphism between the spaces T11 (V ) and T11 (V ) is defined by µV ∗ ,V : T11 (V ) →
T11 (V ) given by
µV ∗ ,V (α ⊗ v) = v ⊗ α.
With respect to a basis of T11 (V ), it follows that µV ∗ ,V (ei ⊗ ej ) = ej ⊗ ei . It is straightforward to see
that this definition does not depend upon the chosen basis. Indeed, given another basis {e′i } for V , and
{e′i } for V ∗ , this equation reads µV ∗ ,V (e′i ⊗ e′j ) = e′j ⊗ e′i . Despite the isomorphism, care is demanded
with respect to the notation in order to distinguish the spaces T11 (V ) and T11 (V ). Suppose that, given
a symmetric bilinear form g defined with respect to a basis {e1 , e2 } of R2 by
g11 = 1, g12 = 1, g21 = 1, g22 = −1,
where gij = g(ei , ej ), it is then possible to raise and lower indices, respectively, via
e1 = e1 + e2 , e2 = e1 − e2 ,
and
1 1
e1 = (e1 + e2 ), e2 = (e1 − e2 ).
2 2
Let us consider B ∈ T11 (V ) given by
B = e1 ⊗ e2 .
In this case,
B1 1 = 0, B1 2 = 1, B2 1 = 0, B2 2 = 0.
By raising and lowering indices via these formulæ, we can write
1 1 1 1
B = e1 ⊗ e1 − e1 ⊗ e2 + e2 ⊗ e1 − e2 ⊗ e2 ,
2 2 2 2
so that
1 1 1 1
B 11 = , B 12 = − , B 21 = , B 22 = − .
2 2 2 2
It follows that Bi j 6= B ji . Hence, it is necessary to distinguish the spaces T11 (V ) and T11 (V ) and thus
to adopt the convention of writing Bij for T11 (V ). Thus, it is then legitimate to write
B11 = 0, B12 = 1, B21 = 0, B22 = 0.
This example is one where using the notation ♯ and ♭ for the correlations associated with g was very useful
for avoiding possible misunderstandings. Indeed, what we accomplished here was defining another tensor
14 Preliminaries
B ′ given by B ′ = (♯ ⊗ ♭)B, where ♯ ⊗ ♭ : T11 (V ) → T11 (V ) denotes the tensor product between the
mappings ♯ and ♭ and which is defined by (♯ ⊗ ♭)(α ⊗ v) = ♯(α) ⊗ ♭(v). As seen in this example, we thus
obtain ♯ ⊗ ♭ 6= µV ∗ ,V .
Tensors of Type (p, q)

The above definitions can be straightforwardly generalised for the tensor product of an
arbitrary number of covectors and vectors, defining in such a way the spaces Tp (V ) =
p q p q q p
(V ∗ )⊗ ; Tq (V ) = V ⊗ ; Tpq (V ) = (V ∗ )⊗ ⊗ V ⊗ ; Tq p (V ) = V ⊗ ⊗ (V ∗ )⊗ ; T1 q p (V ) =
q p
V ∗ ⊗ V ⊗ ⊗ (V ∗ )⊗ ; and so on. We adopt the convention that Tpq (V ) refers to the
space Tpq (V ).
Consider the space Tpq (V ). A basis for this space is given by the set of tensor
products
{eµ1 ⊗ eµ2 ⊗ · · · ⊗ eµp ⊗ eν1 ⊗ eν2 ⊗ · · · ⊗ eνq },
where {µ1 , . . . , µp , ν1 , . . . , νq } = {1, . . . , n}. An arbitrary element T ∈ Tpq (V ) can be
written as
ν ν ···ν
T = Tµ11µ22 ···µqp eµ1 ⊗ eµ2 ⊗ · · · ⊗ eµp ⊗ eν1 ⊗ eν2 ⊗ · · · ⊗ eνq , (1.25)
where
ν ν ···ν
Tµ11µ22 ···µqp = T (eµ1 , eµ2 , . . . , eµp , eν1 , eν2 , . . . , eνq ).
The multilinear functional T ∈ Tpq (V ) is called a tensor of type (p, q). The quantities
ν ν ···ν
Tµ11µ22 ···µqp
are the components of the tensor T in the given basis.
Tensors of type (p, 0) are sometimes called covariant tensors, and tensors of type
(0, q) are called contravariant tensors. A covector is therefore a tensor of type (1,0) –
an example of covariant tensor – whereas a vector is a tensor of type (0,1) and thus a
contravariant tensor.
Transformations
Under a change of basis B 7→ B′ , described by e′j = Bji ei , the dual basis transforms
according to e′j = Bij e′i , and the components of a vector v and a covector α respec-
tively transform according to v ′j = Bij v i , and α′j = Bji αi . The basis vectors B and
the components of a covector transform in a covariant way, and the dual basis vectors
B∗ and the vector components transform in a contravariant way. A generalisation of
those results for a tensor of type (p, q) is straightforward.
Indeed, considering a tensor T of type (p, q), the expression for this tensor in the
bases B and B′ is given by
ν ν ···ν
T = Tµ11µ22 ···µqp eµ1 ⊗ eµ2 ⊗ · · · ⊗ eµp ⊗ eν1 ⊗ eν2 ⊗ · · · ⊗ eνq
ν ν ···ν (1.26)
= (T ′ )µ11 µ22 ···µqp e′µ1 ⊗ e′µ2 ⊗ · · · ⊗ e′µp ⊗ e′ν1 ⊗ e′ν2 ⊗ · · · ⊗ e′νq .
Now, when we substitute the basis change e′µi = Bµνii eνi and the corresponding trans-
formation e′µi = (B −1 )µνii e′νi in this expression, it then reads
ρ ρ ···ρ ν ν ···ν
Tσ11σ22···σqp = (T ′ )µ11 µ22 ···µqp (B −1 )µσ11 · · · (B −1 )µσpp (B −1 )ρν11 · · · (B −1 )ρνqq . (1.27)
Hence, the covariant components (the ones with index p) transform in the same
way as covector components, that is, e. g., in a covariant way. On the other hand, the
contravariant components (that ones with index q) transform in the same way as the
vector components do, that is, in a contravariant way.
Example 1.6 Let g be the symmetric bilinear form shown in Example 1.3. For this case,
g(v, u) = 3v 1 u1 + v 2 u1 + 3v 2 u2 + v 1 u2 + 2v 3 u3
= 3(e1 ⊗ e1 )(v, u) + (e2 ⊗ e1 )(v, u) + 3(e2 ⊗ e2 )(v, u)
+ (e1 ⊗ e2 )(v, u) + 2(e3 ⊗ e3 )(v, u),
and, therefore,
g = 3e1 ⊗ e1 + e2 ⊗ e1 + 3e2 ⊗ e2 + e1 ⊗ e2 + 2e3 ⊗ e3 ,
characterising a covariant type (2, 0) tensor. Now, g can be expressed in terms of another basis. In
Example 1.3, the basis B′ = {e′i },
1 1 1
e′1 = (e1 − e2 ), e′2 = √ e3 , e′3 = √ (e1 + e2 ),
2 2 2 2
is orthonormal with respect to g. From these expressions, we can straightforwardly express {ei } in terms
of {e′i } by merely calculating ei (e′j ), since ei = ei (e′j )e′j . This calculation yields
1 ′1 1 1 1 1
e1 = e + √ e′3 , e2 = − e′1 + √ e′3 , e3 = √ e′2 .
2 2 2 2 2 2 2
Substituting these expressions into the one for g, we obtain
g = e′1 ⊗ e′1 + e′2 ⊗ e′2 + e′3 ⊗ e′3 ,
thus showing that the basis B′ is orthonormal with respect to g. On the other hand, g −1 is a type (0, 2)
contravariant tensor, given by
3 1 1 3 1
g −1 = e1 ⊗ e1 − e1 ⊗ e2 − e2 ⊗ e1 + e2 ⊗ e2 + e3 ⊗ e3 .
8 8 8 8 2
In order to express g −1 in terms of the basis {e′i } we use that
√ √ √ ′
e1 = e′1 + 2e′3 , e2 = −e′1 + 2e′3 , e3 = 2e2 ,
which implies, as expected, that
g −1 = e′1 ⊗ e′1 + e′2 ⊗ e′2 + e′3 ⊗ e′3 .
Direct Sum of Vector Spaces

Let Wi be a subspace of a vector space V and let vi ∈ Wi (i = 1, 2, . . . , n). The sum
W1 + W2 + · · · + Wn is defined by the set of all the sums v1 + v2 + · · · + vn . This sum
is a subspace of V . If Wi ∩ Wj = {0} for i 6= j, and V = W1 + W2 + · · · + Wn (i.e. for
each v ∈ V , there are wi ∈ Wi with v = v1 + · · · + vn ) (Rotman, 2000), then V is
said to be the direct sum of the Wi subspaces, denoted by
n
M
V = W1 ⊕ W2 ⊕ · · · ⊕ Wn = Wi .
i=1
16 Preliminaries
Let now P be a linear operator in V . This operator is said to be a projection

operator, or a projection, if P 2 = P.
Consider a vector space V defined as the direct sum V = W1 ⊕ W2 ⊕ · · · ⊕ Wk , in
other words, that v = v1 + v2 + · · · + vk ∈ V , where vi ∈ Wi (i = 1, . . . , k). Let Pj
(j = 1, . . . , k) be a linear mapping defined by
Pj (vi ) = δji vi .
It follows that
Pj (v) = Pj (v1 + v2 + · · · + vk ) = vj .
Thus, the operator Pj is a projection.
Lk
If V = i=1 Wi , there exists k operators P1 , . . . , Pk in V such that (i) Pj is a
projector (Pj2 = Pj ), i = 1, . . . , k; (ii) Pi Pj = 0 for i 6= j; (iii) P1 + P2 + · · · + Pk = 1 ,
where 1 denotes the identity operator (11(v) = v); and (iv) Pj (V ) = Wj , that is,
the image of Pj is Wj . Conversely, every family (P1 , . . . , Pk ) such that Pi2 = Pi for
i = 1, . . . , k and Pi Pj = 0 , if i 6= j, determines a decomposition V = W1 ⊕ · · · ⊕ Wk .
Example 1.7 Let us consider the space T2 (V ) of the covariant tensors of order 2. We characterise the
spaces T2alt (V ) and T2sym (V ) by the so-called alternator and symmetriser operators, respectively, also
adducing the concepts of direct sum and of projection operators. Let P be the permutation operator
P (α ⊗ β) = β ⊗ α,
and Alt and Sym the operators defined in T2 (V ) by
1 1
Alt = (id −P ), Sym = (id +P ),
2 2
where id is the identity operator id(α ⊗ β) = α ⊗ β. Then, the operators Alt and Sym are projection
operators: Alt2 = Alt ◦ Alt = Alt, and Sym2 = Sym ◦ Sym = Sym. In order to prove this assertion, we
can use the property P 2 = P ◦ P = id; for instance,
1 1
Alt2 = (id −P ) ◦ (id −P ) = (id ◦ id − id ◦P − P ◦ id +P ◦ P )
4 4
1 1
= (id −P − P + id) = (id −P ) = Alt .
4 2
In addition,
Alt ◦ Sym = Sym ◦ Alt = 0, Alt + Sym = id,
which implies that T2 (V
) = T2alt (V
)⊕ T2sym (V
), where Alt : T2 (V ) → T2alt (V ), and Sym : T2 (V ) →
Tsym (V ). Furthermore, ker Sym = Talt (V ), and ker Alt = T2sym (V ).
2 2
1.3 Tensor Algebra
Preliminaries
An algebra A over a field K consists of a vector space over K (here, K = R, C) and
endowed with a bilinear product mA : A × A → A.
Let G be an abelian group. An algebra
L A is said to be G-graded if there exists a
subspace Ak (k ∈ G) such that A = k Ak and, if given xk ∈ Ak , yl ∈ Al , it follows
Tensor Algebra 17
that xk yl ∈ Ak+l . The elements of Ak are said to be homogeneous of degree k. In

general, we use the notation
k = deg(xk ), xk ∈ Ak . (1.28)
Since G is abelian,
deg(xk yl ) = deg(xk ) + deg(yl ). (1.29)
Note that, for the scalar a ∈ K = R, C, it is assumed that deg(a) = 0 and that
the null vector must be considered homogeneous for all degrees, since every subspace
Ak contains it. If the unique element which is negative graded is the null vector, the
algebra is said to be positive graded.
The Tensor Algebra

Given two tensors T and S of type (p, q), it is possible to define their sum as the tensor
T + S of type (p, q) in terms of their components by
ν ν ···ν ν ν ···ν ν ν ···ν
(T + S)µ11 µ22 ···µqp = Tµ11µ22 ···µqp + Sµ11 µ22 ···µqp . (1.30)
If T is a tensor of type (p, q), and S is a tensor of type (r, s), we can define a tensor
product T ⊗ S which is a tensor of type (p+r, q +s). In terms of components, it follows
that
ν ν ···ν ρ ρ ···ρ ν ν ···ν
(T ⊗ S)µ11 µ22 ···µqp σ11 σ2 2 ···σs r = Tµ11µ22 ···µqp Sσρ11 σρ22···ρ
···σr .
s
(1.31)
The tensor product is distributive with respect to the sum, namely (T + S) ⊗ R =
T ⊗ R + S ⊗ R and T ⊗ (S + R) = T ⊗ S + T ⊗ R; in addition, it is associative:
(T ⊗ S) ⊗ R = T ⊗ (S ⊗ R). As already seen, the tensor product is not commutative: in
general, T ⊗S 6= S ⊗T . The direct sum of all the vector spaces Tpq (V ) endowed with the
operations of sum and of tensor product is called the tensor algebra associated with
the vector space V . The tensor algebra is a graded algebra. In the general case, the
grading is given by G = Z×Z, and it is positive. Two cases are particularly important:
the algebra of the covariant tensors and that of the contravariant tensors.
L∞ The algebra
of the covariant tensors – of type (p, 0) – is denoted by T∗ (V ) = p=0 T p
(V ). We
denote T0 (V ) = R, and T1 (V ) = VL ∗
. The algebra of the contravariant tensors – of
∞
type (0, q) – is denoted by T(V ) = q=0 Tq (V ), where T0 (V ) = R, and T1 (V ) = V .
The algebra of the covariant tensors, and the algebra of the contravariant tensors, are
Z-graded algebras.
Let us consider the algebra of the covariant tensors T∗ (V ). Since it is Z-graded, it
allows us to define a mapping called the grade involution as
#(Tp ) = (−1)deg Tp Tp = (−1)p Tp , (1.32)
where Tp ∈ Tp (V ) ⊂ T∗ (V ). Another notation that can be used for the grade involution
is
cp = #(Tp ).
T (1.33)
Observation ☞ There are reasons for using both notations for the grade involution. The
second notation (the ‘hat’) is much more convenient than the first, for its usefulness in
18 Preliminaries
formulæ involving the action of this mapping on an element. Nevertheless, when we

solely want to denote the mapping itself, the first notation is much more appropriate
than the second. As it will be seen, there are situations where one notation is more
suitable than the other. In addition, there is no possibility of confusing the symbol
# for the grade involution with the correlation ♯. Besides the noticeable typographic
difference, in general the correlation is written α♯ instead of ♯(α), and in this text we
never use # as an index, as in x# .
The mapping # is an automorphism. Indeed,
#(Tp ⊗ Sq ) = (−1)deg(Tp ⊗Sq ) (Tp ⊗ Sq ) = (−1)(deg(Tp )+deg(Sq )) (Tp ⊗ Sq )
(1.34)
= (−1)deg(Tp ) Tp ⊗ (−1)deg(Sq ) Sq = (#Tp ) ⊗ (#Sq ).
Besides, it also satisfies #2 = 1, where here 1 denotes the identity mapping, which
justifies the denomination grade ‘involution’.
Since #2 = 1, there is a refinement in the grading of T∗ (V ). An element Tp ∈ Tp (V )
is said to be even or odd if (−1)p is respectively even or odd. In this way, the operators
Π± can be defined as
1
Π± = (1 ± #). (1.35)
2
These operators Π± are projectors, as can be straightforwardly verified. The sub-
space T∗+ (V ) = Π+ (T∗ (V )) consists of the even elements in T∗ (V ), and the sub-
space T∗− (V ) = Π− (T∗ (V )) consists of the odd elements. It is then possible to write
T∗ (V ) = T∗+ (V ) ⊕ T∗− (V ), and
T∗+ (V ) ⊗ T∗+ (V ) ⊂ T∗+ (V ), T∗+ (V ) ⊗ T∗− (V ) ⊂ T∗− (V ),
(1.36)
T∗− (V ) ⊗ T∗+ (V ) ⊂ T∗− (V ), T∗− (V ) ⊗ T∗− (V ) ⊂ T∗+ (V ).
The grade involution endows both the algebras of covariant and contravariant tensors
with a Z2 -grading.
Finally, another prominent and very useful mapping is called reversion, denoted
by a tilde and defined by
(T^ f f
p ⊗ Sq ) = Sq ⊗ Tp , (1.37)
for all Tp ∈ Tp (V ), and Sq ∈ Tq (V ), where
e
a = a, ∀a ∈ R,
(1.38)
e = α,
α ∀α ∈ T1 (V ) = V ∗ .
This definition implies that
(α ⊗ β^
⊗ · · · ⊗ ω) = ω ⊗ · · · ⊗ β ⊗ α, (1.39)
∗
for α, β, . . . , ω ∈ V , which justifies the name reversion.
The composition of the grade involution and the reversion is called conjugation
and is denoted by a bar:
f c
cp = T
fp .
T̄p = T (1.40)
Now, the tensor product of linear mappings can be defined. Let fi : Vi → Wi
(i = 1, 2) be homomorphisms. Because of the universality of V1 × V2 → V1 ⊗ V2 ,
Exercises 19
there is a linear bijection from Lin(V1 ⊗ V2 , W1 ⊗ W2 ) onto Lin(2) (V1 , V2 ; W1 ⊗ W2 ).

The mapping f1 ⊗ f2 is the Lin(V1 ⊗ V2 , W1 ⊗ W2 ) element that corresponds to the
Lin(2) (V1 , V2 ; W1 ⊗ W2 ) element which maps every (v1 ⊗ v2 ) to f1 (v1 ) ⊗ f2 (v2 ). This
definition can be straightforwardly extended for an arbitrary number of factors.
1.4 Exercises
(1) Let {e1 , e2 , e3 } be a basis of V = R3 , where e1 = (1, 0, 1)⊺ ; e2 = (1, 1, −1)⊺ ;
and e3 = (0, 1, 2)⊺ . Let also α be the covector given by α(e1 ) = 4; α(e2 ) = 1; and
α(e3 ) = 1. Calculate α(v) for the vector v = (a, b, c) and express α in terms of the
dual basis associated to the standard basis of R3 .
(2) Let {e1 , e2 } be a basis of R2 , where e1 = (1, −1)⊺ , and e2 = (2, −1)⊺ , and let g be
the non-degenerate symmetric bilinear form given by g = 2e1 ⊗ e1 + e1 ⊗ e2 + e2 ⊗ e1 +
2e2 ⊗ e2 . Compute the correlation τ associated with g in terms of v = (a, b) and find a
basis {e′1 , e′2 } in terms of which g can be written in the form g = e′1 ⊗ e′1 + e′2 ⊗ e′2 .
(3) Let M(n, R) be the vector space of the real matrices n × n. Given
P A = {Aij } ∈
M(n, R), define the trace function Tr : M(n, R) → R as Tr(A) = i Aii . (a) Show
that Tr is a linear function over M(n, R). (b) Show that Tr(AB) = Tr(BA) for all
A, B ∈ M(n, R). (c) Show that there do not exist matrices A and B such that AB −
BA = I, where I ∈ M(n, R) is the identity matrix of order n. (d) Consider now
an infinite-dimensional space W . Exhibit operators A, B ∈ in the space End(W ) of
endomorphisms of W such that AB − BA = IdW . (e) Suppose that AB − BA = IdW .
Show that Am B − BAm = mAm−1 , m ∈ N.
(4) Let M(n, R) be the space of the real matrices n × n. Consider the set of the n2
matrices Eij (i, j = 1, . . . , n) defined in the following way: all the matrix entries Eij
equal 0 except the entry that corresponds to the ith row and j th column, as it equals
1. In other words, if the pairs kl denote the entries corresponding to the k th row and
the lth column, then the matrix Eij has the form (Eij )kl = δik δjl , where δik equals
1 if i = k and equals 0 if i 6= k. (a) Show that the set of the matrices B = {Eij }
(i, j = 1, . . . , n) is a basis for M(n, R). (b) Show that Eij Emn = δjm Ein . (c) Define
the dual basis B∗ = {E ij } such that E ij (Ekl ) = δki δlj . Show that the components
P of
the trace function Tr in this basis are given by Trij = δij , so that Tr = i E ii .
(5) Let Lin(V, V ) be the set of the linear mappings of a vector space V in itself V .
Consider the mapping φV : V ⊗ V ∗ → Lin(V, V ), defined as (φV (v ⊗ α))(u) = vα(u),
for any v, u ∈ V . (a) Show that any linear transformation T : V → V can be written
as T = φV (Tji ei ⊗ ej ) (here the summation convention is assumed!), where B = {ei } is
an arbitrary basis of V , B = {ei } is its corresponding dual basis, and the scalar Tji is
given by T (ei ) = Tij ej . (b) Show that ker φV = {0}, in such a way that, when it is taken
together with the result in item (a), we can conclude that φV is an isomorphism, namely
that the spaces Lin(V, V ) and V ⊗ V ∗ are isomorphic. (c) Show that φV (ei ⊗ ei ) = idV .
(d) Consider the mappings evV : V ∗ ⊗ V → R, and µV,V ∗ : V ⊗ V ∗ → V ∗ ⊗ V , defined
by evV (α ⊗ v) = α(v), and µV,V ∗ (v ⊗ α) = α ⊗ v, for all v ∈ V and α ∈ V ∗ . Show
that the trace function can be defined by the following composition of the mappings:
Tr = evV ◦µV,V ∗ ◦ φ−1V .
20 Preliminaries
(6) Let V = M(n, K). Construct explicitly the isomorphism End(V ) ≃ (End(V ))∗
(Hint: show that, for any α ∈ V ∗ , there exists a unique matrix A ∈ V with the
property α(X) = Tr(AX), ∀X ∈ V ).
(7) Although the tensor product between two vector spaces V and W is not commu-
tative, it is possible to establish the isomorphism
µV,W : V ⊗ W → W ⊗ V,
v ⊗ w 7→ w ⊗ v (1.41)
when the basis {ei ⊗ fj } of V ⊗ W is changed to the basis {fj ⊗ ei } of W ⊗ V .

(a) Show that µV,W ◦ µW,V = idW ⊗V and that µW,V ◦ µV,W = idV ⊗W .
(b) Given another vector space U , considering the vector space U ⊗ V ⊗ W , show
that (µV,W ⊗ idU ) ◦ (idV ⊗ µU,W ) ◦ (µU,V ⊗ idW ) = (idW ⊗ µU,V ) ◦ (µU,W ⊗ idV ) ◦
(idU ⊗ µV,W ). This expression is called the Yang–Baxter equation and it defines
the braid group.
(8) (a) Find the value of the tensor φ⊗ψ −ψ ⊗φ ∈ T 5 (V ) applied to (v1 , v2 , . . . , v5 ) ∈
V × V × V × V × V , where φ = e1 ⊗ e2 + e2 ⊗ e3 + e2 ⊗ e2 ∈ T 2 (V ); ψ = e1 ⊗ e1 ⊗
(e1 − e3 ) ∈ T 3 (V ); v1 = e1 ; v2 = e1 + e2 ; v3 = e2 + e3 ; and v4 = v5 = e2 . (b) Find
12
the components T̃123 of a tensor in T2 3 (V ), if all its components in the basis {ei } are
equal to 2, and the bases {ẽi } and {ei } are related by
 
123
(ẽ1 , ẽ2 , ẽ3 ) = (e1 , e2 , e3 ) 0 1 2 .
001
(9) Let A = [aij ] be a matrix associated with a linear operator A : V → V in the

basis {e1 , . . . , en } of V and let B = [bkl ] be a matrix associated with B : V →
V in the basis {f1 , . . . , fm } of W . The matrix associated with A ⊗ B in the basis
{e1 ⊗ f1 , e1 ⊗ f2 , . . . , e1 ⊗ fm , e1 ⊗ f1 , e2 ⊗ f2 , . . . , e2 ⊗ fm , . . . , en ⊗ fm } of V ⊗ W is
given by  
a11 B a12 B · · · a1n B
 a21 B a22 B · · · a2n B 
 
A⊗B =  . .. . . ..  .
 .. . . . 
an1 B an2 B · · · ann B
(a) Show that Tr(A ⊗ B) = Tr A . Tr B. Calculate
Tr(A ⊗ A ⊗· · · ⊗ A). (b) Given

10 01 0 −i 1 0
I = σ0 = , and the Pauli matrices σ1 = , σ2 = , and σ3 =
01 10 i 0 0 −1
in M(2, C), compute σi ⊗ σj (i, j = 0, 1, 2, 3).
(c) Show that, if A : V → V is diagonalisable, then A ⊗ A ⊗ · · · ⊗ A is also diagonalis-
able. If {λi } denotes the spectrum of A, what are the eigenvalues associated with the
operator A ⊗ A ⊗ · · · ⊗ A?
2
Exterior Algebra and Grassmann
Algebra
In this chapter, exterior algebras and Grassmann algebras are discussed. In some texts,
the terms exterior algebra and Grassmann algebra are considered to be synonyms;
however, this is not the case here! We aim to present the differences between the
exterior algebra and the Grassmann algebra as clearly as possible, in order to avoid
future confusion. The quasi-Hodge isomorphism is thus presented and studied. Basing
our approach on the Ausdehnungslehre of 1862 (Grassmann, 1862), where Grassmann
used the Hodge star operator to define the regressive product, here we will employ the
quasi-Hodge operator to accomplish the quasi-Hodge isomorphism. We will end the
chapter by examining the Hodge isomorphisms.
2.1 Permutations and the Alternator

A permutation of the set of p elements {1, 2, . . . , p} is a bijection σ : {1, 2, . . . , p} →
{1, 2, . . . , p}, represented by the cycle

1 2 ··· p
.
σ(1) σ(2) · · · σ(p)
The composition of two permutations is obviously another permutation, and the set of
all permutations is a group called the symmetric group, denoted by Sp . The number
of elements in Sp is p!.
A permutation σ of the set {1, 2, . . . , p} such that σ(k) = k for all k 6= i, and
k 6= j, and moreover σ(i) = j, and σ(j) = i, is called a transposition. A permutation
of n elements is even or odd if the permutation is obtained respectively by an even
or odd number of transpositions. The sign ε(σ) of the permutation σ is defined to be
ε(σ) = +1 if the permutation is even, and ε(σ) = −1 if the permutation is odd.
Let us consider now a tensor that is either contravariant or covariant of the form
X1 ⊗ X2 ⊗ · · · ⊗ Xp , where X denotes respectively either a vector or a covector, and
the indices enumerate such elements. The operator Alt called alternator is defined in
the following way:
1 X
Alt(X1 ⊗ X2 ⊗ · · · ⊗ Xp ) = ε(σ)Xσ(1) ⊗ Xσ(2) ⊗ · · · ⊗ Xσ(p) . (2.1)
p!
σ∈Sp
22 Exterior Algebra and Grassmann Algebra
For other cases, this definition is generalised by linearity. The operator Alt is a pro-
jection operator (Alt2 = Alt) and is a straightforward generalisation of the operator
defined in example 1.7.
Example 2.1 Permutations of three elements form the symmetric group S3 . These permutations can be
represented by

123 123 123
, , ,
123 231 312

123 123 123
, , .
321 213 132
The permutations represented by the matrices in the first row are all even permutations, whereas the ones
in the second row are all odd permutations.
For a tensor X1 ⊗ X2 ⊗ X3 , which is either covariant or contravariant, this alternator is therefore
written as
1
Alt(X1 ⊗ X2 ⊗ X3 ) = X1 ⊗ X2 ⊗ X3 + X2 ⊗ X3 ⊗ X1 + X3 ⊗ X1 ⊗ X2
6

− X3 ⊗ X2 ⊗ X1 − X2 ⊗ X1 ⊗ X3 − X1 ⊗ X3 ⊗ X2 .
There is another way to express the action of the alternator. Let us, for instance,
suppose that covariant tensors are taken into account. These objects are multilinear
functionals whose arguments are vectors. Without loss of generality, let us consider a
covariant tensor of the form α1 ⊗ α2 ⊗ · · · ⊗ αp . Therefore,
(α1 ⊗ α2 ⊗ · · · ⊗ αp )(v1 , v2 , . . . , vp ) = α1 (v1 )α2 (v2 ) · · · αp (vp ).
The action of Alt on a contravariant tensor is then defined by writing the resulting
tensor acting on vectors as
α1 (v1 ) α1 (v2 ) . . . α1 (vp )
1 α2 (v1 ) α2 (v2 ) . . . α2 (vp )
Alt(α1 ⊗· · · ⊗αp )(v1 , . . . , vk ) = .. .. .. .. . (2.2)
p! . . . .
αp (v1 ) αp (v2 ) . . . αp (vp )
At the right-hand side of this equation, we denote the determinant of the associated
matrix. This result follows immediately from the definition of the determinant, namely,
if A is the matrix of order p with entries Aij , then the determinant det A reads as
follows (Hoffman and Kunze, 1971):
X
det A = ε(σ)A1σ(1) A2σ(2) · · · Apσ(p) .
σ∈Sp
2.2 p-Vectors and p-Covectors

A p-vector is an alternating contravariant tensor of order p. A p-vector is denoted by
A[p] and characterised by
A[p] = Alt(A[p] ). (2.3)
The brackets here are used to indicate the alternation of the p indices set. Thus, given
Ap ∈ Tp (V ), Alt(Ap ) is a p-vector, since Alt(Alt(Ap )) = Alt(Ap ).
p-Vectors and p-Covectors 23
Similarly, a p-covector is an alternating covariant tensor of order p. Denoted by

Ψ[p] , it is intrinsically alternating as well:
Ψ[p] = Alt(Ψ[p] ). (2.4)

V Vp
The symbols p (V ) and (V ) respectively denote the space of the p-vectors and
p-covectors; 1-vector is a synonym for vector, and 1-covector is a synonym for covector.
Both 0-vectors or 0-covectors are scalars. Using this notation for all p, we thus consider
^ ^ ^ ^
0 1
(V ) = 0 (V ) = R, (V ) = V ∗ , 1 (V ) = V. (2.5)
Observation ☞ It is common also to use the term bivector for a 2-vector, trivector for
a 3-vector, and so on (analogously for the p-covectors).
V2
Example 2.2 Let Ψ = Ψij ei ⊗ ej ∈ T2 (V ). A 2-covector Ψ[2] ∈ (V ) is alternating by definition:
1
Ψ[2] = Ψij (ei ⊗ ej − ej ⊗ ei )
2
1
= (Ψij − Ψji )ei ⊗ ej .
2
Notice that, in this case, the inequality dim V ≥ 2 must hold since, if dim V = 1, it follows that
Ψ[2] = (1/2)Ψ11 (e1 ⊗ e1 − e1 ⊗ e1 ) = 0. V
Let now A = Aijk ei ⊗ ej ⊗ ek ∈ T3 (V ) be a 3-tensor. A 3-vector A[3] ∈ 3 (V ) can be written as
an alternator Alt(A):
1 ijk
A[3] = A ei ⊗ ej ⊗ ek + ej ⊗ ek ⊗ ei + ek ⊗ ei ⊗ ej
6

− ek ⊗ ej ⊗ ei − ej ⊗ ei ⊗ ek − ei ⊗ ek ⊗ ej
1 ijk
= A + Ajki + Akij − Akji − Ajik − Aikj ei ⊗ ej ⊗ ek .
6
In this case, dim V ≥ 3; otherwise, A[3] = 0.
Vp V
The Dimensions of (V ) and p (V )
Vp V
Obviously, dim (V ) = dim p (V ), but what are the dimensions of such spaces? Let
us consider a basis for the space Tp (V ) of the covariant tensors of order p:
Bp = {ei1 ⊗ ei2 ⊗ · · · ⊗ eip },
where the indices ik (k = 1, 2, . . V

. , p) take the values ik = 1, 2, . . . , n, and where
p
n = dim V . A p-covector Ψ[p] ∈ (V ) can be written [p]
Vp as Ψ = Alt(Ψ ), where
p
p p
Ψ ∈ T (V ). In order to calculate the dimension of (V ), it
Vpsuffices to realise how
many elements in the basis of Tp (V ) contribute to the space (V ).
Among the elements of Bp , we must select all the ones which are annihilated by
the alternator. Hence, for the index i1 , we have n possible choices; for the index i2 ,
there are n − 1 possible choices; and so on, until the pth index ip , for which we have
n−p+1 choices. Consequently, the number of Bp elements that contributes is reduced
to n(n − 1) · · · (n − p + 1).
Meanwhile, not all the n(n − 1) · · · (n − p + 1) elements of Bp present distinct

contributions! Any one of them can be obtained from another one by a permutation and
contributes in the same way, because of the action of the alternator via permutations.
For p elements, the alternator has p! distinct permutations, in such a way that the
p
quantity n(n − 1) · · · (n − p + 1) must be divided by p!. The number of elements
Vp of B
that contribute in a different way to the calculations for the dimension of (V ) is
therefore
n(n − 1) · · · (n − p + 1) n! n
= = ,
p! (n − p)!p! p
which is the well-known number of p-combinations of n elements. Thus, we conclude
that
^ ^ n
dim p (V ) = dim p (V ) = , (p = 0, 1, . . . , n). (2.6)
p
There can be no (n + 1)-vectors or (n + 1)-covectors if dim V = n. This result is
obvious since, in such a case, at least two indices of an alternating tensor of order
(n + 1) would be equal, thus
annihilating
the alternating tensor.
Remembering that np = n−p n
, we can then obtain the following important result:
^ ^
p n−p
dim (V ) = dim (V ) . (2.7)
Vp Vn−p
Although the spaces (V ) and (V ) are isomorphic, there is no natural isomor-
phism between them. One isomorphism which is extremely useful and which requires
the consideration of additional structures on the vector space V is called the Hodge
isomorphism, and will be examined in section 2.10.
Observation ☞ It would be at the very least irritating (to the reader, and to the
authors as well) if we were to repeat every construction involving p-vectors for p-
covectors. Once the construction for one case has been accomplished, the same for the
other case can be regarded mutatis mutandis. Hence, from this point on, only the case
involving p-vectors shall be considered.
2.3 The Exterior Product

Let A[p] be a p-vector, and B[q] a q-vector. Since such quantities are alternating covari-
ant tensors, it is natural to take the tensor product between these quantities, namely
A[p] ⊗ B[q] . The result of such tensor product, although it is a covariant tensor of
order p + q, is not alternating. Meanwhile, the object Alt(A[p] ⊗ B[q] ) is an alternating
covariant tensor of order p + q, namely it is a (p + q)-vector.
V V
Definition 2.1 ◮ Let A[p]V∈ p (V )Vbe a p-vector
V and let B[q] ∈ q (V ) be a q-vector.
The exterior product ∧ : p (V ) × q (V ) → p+q (V ) is defined as
A[p] ∧ B[q] = Alt(A[p] ⊗ B[q] ) . (2.8)

The Exterior Product 25
Some consequences of this definition are now explored. First, the exterior product
is associative, that is,
(A[p] ∧ B[q] ) ∧ C[r] = A[p] ∧ (B[q] ∧ C[r] ); (2.9)
this associativity is an inherited property, elicited from the associativity

V of the tensor
product. Obviously, the exterior product is also bilinear. If a ∈ 0 (V ) = R is a scalar,
it follows that a ∧ A[p] = aA[p] . Equation (2.9) can be shown by starting with the
action of the group of permutations Sp on the tensor space Tp (V ). It may be either
an action on the left side (such that σ(τ (A[p] )) = (στ )(A[p] )) or an action on the right
side (such that (A[p]σ )τ = A[p]στ ):
either σ(v1 ⊗ · · · ⊗ vp ) = ǫ(σ)vσ−1 (1) ⊗ · · · ⊗ vσ−1 (p) ,

or (v1 ⊗ · · · ⊗ vp )σ = ǫ(σ)vσ(1) ⊗ · · · ⊗ vσ(p) .
Vp
Thus, (V ) is the subspace of all A[p] ∈ Tp (V ) such that either σ(A[p] ) = A[p] or
A[p]σ = A[p] for all σ ∈ Sp .
Moreover, the projector Alt is theVp classical projector from the space Tp (V ) onto
the subspace of invariant elements (V ). The definition of the exterior product is
equivalent to
p!q! X σ
A[p] ∧ B[q] = A[p] ⊗ B[q] ,
(p + q)!
σ∈Sp,q
where Sp,q denotes the subset of Sp+q containing all σ such that σ(i) < σ(i + 1) if
0 < i < p, or p < i < p + q. Subsequently, both sides of equality (2.9) prove to be
equal to
p!q!r! X
A[p] ⊗ B[q] ⊗ C[r] ,
(p + q + r)!
σ∈Sp,q,r
where Sp,q,r is a subset of Sp+q+r and contains all σ such that σ(i) < σ(i + 1) if (a)
0 < i < p or p < i < p + q; and (b) 0 < i < p + q or p + q < i < p + q + r.
The case involving the exterior product between two vectors is fundamental. Ac-
cording to the definition of an exterior product, the exterior product between two
vectors reads
1
v ∧ u = (v ⊗ u − u ⊗ v). (2.10)
2
From this equation, it follows that
v ∧ u = −u ∧ v , (2.11)
namely, that the exterior product involving two vectors is anti-commutative. In par-
ticular,
v ∧ v = 0. (2.12)
It is straightforward to obtain the generalisation regarding eqn (2.11). Because of
the bilinearity and associativity of the exterior product, it is enough to consider a
p-vector A[p] and a q-vector B[q] in the form
A[p] = v1 ∧ · · · ∧ vp , B[q] = u1 ∧ · · · ∧ uq . (2.13)
The exterior product A[p] ∧ B[q] can be thus written as
A[p] ∧ B[q] = v1 ∧ · · · ∧ vp ∧ u1 ∧ · · · ∧ uq . (2.14)
Now using eqn (2.11), in order to reorder the vectors involved in the exterior products
on the right-hand side, we find that
v1 ∧ · · · ∧ vp ∧ u1 ∧ · · · ∧ uq = (−1)pq u1 ∧ · · · ∧ uq ∧ v1 ∧ · · · ∧ vp ,
namely,
A[p] ∧ B[q] = (−1)pq B[q] ∧ A[p] . (2.15)
A p-vector that can be written as the exterior product of a p number of 1-vectors –

as in eqn (2.13) – is called a simple p-vector. In vector spaces V such that n = dim V ≤
3, every p-vector is simple. For dim V ≥ 4, not all p-vectors are simple. For instance,
let V be a vector space of dimension 4, and B = {e1 , e2 , e3 , e4 } a basis for V . Let A[2]
be the 2-vector given by A[2] = e1 ∧ e2 + e3 ∧ e4 . There is no linear combination of
the vectors {ei } (i = 1, 2, 3, 4) that allows us to write A[2] = v1 ∧ v2 . This prominent
and useful resultVdeserves a detailed exposition. V
Let 0 6= ψ ∈ 2 (V ). Then, ψ is simple if and only if ψ ∧ ψ = 0 ∈ 4 (V ). Indeed, if
ψ = u ∧ v, for u, v ∈ V , then ψ ∧ ψ = u ∧ v ∧ u ∧ v = 0. The reciprocal assertion
V can be
shown using induction in the dimension of V . If dim V = 0 or 1, then 2 (V ) = {0},
and
V therefore the first case to be considered is when dim V = 2. In this case, dim
2 (V ) = 1, and v1 ∧ v2 is a non-trivial element if {v1 , v2 } is a basis of V , and ψ
is simple. V Now let us consider the case where dim VV= 3 is considered now. Given
0 6= Vψ ∈ 2 (V ), let us define a mapping A : V → 3 (V ) by A(v) = ψ ∧ v. Since
dim 3 (V ) = 1, therefore dim ker A ≥ 2. Now let u1 and u2 be linearly independent
vectors which are in ker A and can be extended to a basis {u1 , u2 , u3 } of V , so
we can write ψ = au1 ∧ u2 + bu1 ∧ u3 + cu2 ∧ u3 . By definition, A(u1 ) = 0, and
therefore 0 = ψ ∧ u1 = cu1 ∧ u2 ∧ u3 , which implies that c = 0. In the same way,
A(u2 ) = 0, and so 0 = ψ ∧ u2 = −bu1 ∧ u2 ∧ u3 , which means that b = 0. It follows
that ψ = au1 ∧ u2 , which is simple. Suppose now by induction that the assumption
holds for dim V ≤ n − 1, and consider the case where dim V = n. Using the basis
{v1 , . . . , vn }, it follows that
n n−1
! n−1
X X X
ψ= aij vi ∧ vj = ain vi ∧ vn + aij vi ∧ vj
1≤i<j i=1 1≤i<j
′
= u ∧ vn + ψ ,
V
where U is the subspace generated by {v1 , . . . , vn−1 }, u ∈ U , and ψ ′ ∈ 2 (U ). Now,
0 = ψ ∧ ψ = (u ∧ vn + ψ ′ ) ∧ (u ∧ vn + ψ ′ ) = 2u ∧ ψ ′ ∧ vn + ψ ′ ∧ ψ ′ ,
but vn appears neither in the expansion of u ∧ ψ ′ nor in the expansion of ψ ′ ∧ ψ ′ , and

separately one obtains u ∧ ψ ′ = 0 = ψ ′ ∧ ψ ′ . By induction, 0 = ψ ′ ∧ ψ ′ implies that
The Exterior Product 27
ψ ′ = u1 ∧ u2 , and so u ∧ u1 ∧ u2 = 0. Hence, there exists µ, λ1 , λ2 ∈ K such that

µu + λ2 u2 + λ1 u1 = 0. If µ = 0, then u1 and u2 are linearly dependent, and therefore
ψ ′ = u1 ∧ u2 = 0, which means that ψ = u ∧ vn , and therefore ψ is simple. If µ 6= 0,
then u = − λµ2 u2 − λµ1 u1 , and
λ1 λ2
ψ=− u1 ∧ vn − u2 ∧ vn + u1 ∧ u2 ,
µ µ
corresponding to the three-dimensional case, which was shown to be always simple.
Observation ☞ There is no consensus with respect to the definition of the exterior
product. Some authors define the exterior product of vectors as in eqn (2.10), whereas
other authors define it without the factor 1/2, namely, v ∧ u = v ⊗ u − u ⊗ v. The
source of this difference lies in the definition of the alternating operator: some authors
use the factor 1/p! as we use it in eqn (2.1), whereas other authors prefer not to use it.
However, this difference is completely irrelevant when the necessary attention is taken
and the corresponding adaptations made in the definitions. Without the factor 1/p! ,
the operator Alt is not a projection operator. Therefore, in order for us to use it as
a projection operators, we must employ a factor p in the definition of the contraction
(or interior product; see eqn (2.34)). Authors that do not use the factor 1/p! in the
definition of a p-vector do not need to use the factor p in eqn (2.34) but must use
the factor 1/p!q! in eqn (2.8). For those mathematicians who need exterior algebras
over arbitrary fields see, for instance, the work by Lam (1980), Lounesto (2001a), and
Helmstetter and Micali (2008).
Bases
Suppose that dim V = n and let B = {e1 , . . . , en } be a basis
V for V . From this basis,
we can then construct a basis for
V each one of the spaces p (V ).
Consider at first the space 2 (V ) and the exterior products of the form ei ∧ ej .
Because of the anti-commutativity of the exterior product between basis vectors, the
linearly independent set of 2-vectors is provided by
e1 ∧ e2 , e1 ∧ e3 , e1 ∧ e4 , . . . , e1 ∧ en ,
e2 ∧ e3 , e2 ∧ e4 , . . . , e2 ∧ en ,
··· ··· ···
en−1 ∧ en .
Therefore,
V there exist (n−1)+(n−2)+· · ·+1 = n(n−1)/2
V elements, which is precisely
dim 2 (V ). This set of 2-vectors forms a basis for 2 (V ). An arbitrary 2-vector A[2]
can thus be written as
1 X ij X
A[2] = A ei ∧ ej = Aij ei ∧ ej . (2.16)
2 i,j i<j
Notice in the first expression the presence

P of the factor 1/2, which is absent in the
second expression. Since, in the sum ij , we consider all the values for the indices
i, j = 1, 2, . . . , n and, as Aij = −Aji , and ei ∧ ej = −ej ∧ ei , we are indeed counting

the same term twice. However, P this factor is not necessary in the second expression,
since we consider the sum i<j restricted to all values of i and j such that i < j, so
that no term is counted twice. V
This result can be generalised for p (V ). A basis for this space consists of elements
of the form eµ1 ∧ eµ2 ∧ · · · ∧ eµp , and the number of distinct elements is the number
V
of p-combinations of n elements, denoted by np . An arbitrary element A[p] ∈ p (V )
can be written as
1 X
A[p] = Aµ1 µ2 ···µp eµ1 ∧ eµ2 ∧ · · · ∧ eµk
p! µ µ ···µ
1 2 p
X (2.17)
= Aµ1 µ2 ···µp eµ1 ∧ eµ2 ∧ · · · ∧ eµp ,
µ1 <µ2 <···<µp
where, in the first case, we consider the sum over all possible values for the indices
µi = 1, 2, . . . , n (i = 1, 2, . . . , p) and, in the second case, we consider the sum over the
indices with the restriction that µ1 < µ2 < · · · < µp .
Observation ☞ Sum convention: In order to apply the sum convention to p-vectors
as efficiently as possible, we impose the same restriction on the sum convention with
respect to the indices, namely, that
X
Aµ1 µ2 ···µp eµ1 ∧ eµ2 ∧ · · · ∧ eµp = Aµ1 µ2 ···µp eµ1 ∧ eµ2 ∧ · · · ∧ eµp . (2.18)
µ1 <µ2 <···<µp
Thus, we avoid using expressions with the factor p!.
n-Vectors, or Pseudoscalars
The exterior product of m vectors is 0 whenever m > n, where n = dim V . Indeed, let
us consider the exterior product of n + 1 vectors, v1 ∧ · · · ∧ vn ∧ vn+1 . If dim V = n,
there are at most n linearly independent vectors. Therefore, the n + 1 given vectors
are necessarily linearly dependent and we can write one of those vectors as a linear
combination of the others. Without loss of generality,
Pn it is possible to choose such a
vector as being vn+1 . It follows that vn+1 = i=1 ai vi . Using the anti-commutation
relation (2.11), it reads
v1 ∧ · · · ∧ vn ∧ vn+1 =v1 ∧ v2 ∧ · · · ∧ vn ∧ (a1 v1 + a2 v2 + · · · + an vn )

=(−1)n−1 a1 v1 ∧ v1 ∧ v2 ∧ · · · ∧ vn
+ (−1)n−2 a2 v1 ∧ v2 ∧ v2 ∧ v3 ∧ · · · ∧ vn
+ · · · + an v1 ∧ · · · ∧ vn−1 ∧ vn ∧ vn = 0,
where vi ∧ vi = 0 is used. Consequently, we have
v1 ∧ v2 ∧ · · · ∧ vm = 0 if m > n. (2.19)
Indeed, a stronger result than that can be shown, namely, that

V
The Exterior Algebra (V ) 29
v1 ∧ · · · ∧ vp = 0 ⇐⇒ {v1 , . . . , vp } is linearly dependent.

V
This result shows that the vector space p (V ) does not exist if p > n. The spaces that
can be constructed are
^ ^ ^ ^ ^
0 (V ), 1 (V ), 2 (V ), . . . , n−1 (V ), n (V ),
V V V
such that dim p (V ) = dim n−p (V ). The space n (V ) is particularly important.
V
The vector space n (V ) has dimension nn = 1. A basis for this space consists of
the element v1 ∧ · · · ∧ vn , where {v1 , . . . , vn } is a set of linearly independent
V vectors.
If B = {e1 , . . . , en } is a basis of V , it is natural to take as a basis for n (V ) the
exterior product of these n vectors of B. Whatever order these n vectors are taken,
their product can be transformed into
±e1 ∧ · · · ∧ en ,
because of the anti-commutativity associated with the exterior product. For instance,
en ∧ · · · ∧ e1 = (−1)n(n−1)/2 e1V∧ · · · ∧ en , where (−1)n(n−1)/2 = ±1, depending upon
the value of n. For a basis for n (V ), we shall consider the element
e1 ∧ · · · ∧ en ,
and hence the exterior product of n linearly independent vectors can be always written
as
v1 ∧ · · · ∧ vn = pe1 ∧ · · · ∧ en , (2.20)
V
where p ∈ R. The elements of n (V ) are called n-vectors, as before, and another usual
denomination for the n-vectors is pseudoscalars.
V
2.4 The Exterior Algebra (V )
V
Let
V us consider the vector space (V ) defined by the direct sum of the vector spaces
p (V ), (p = 0, 1, 2, . . . , n):
^ ^ ^ ^ ^ n ^
M
(V ) = 0 (V )⊕ 1 (V )⊕ 2 (V ) ⊕ ··· ⊕ n (V )= p (V ).
p=0
V V
V The exterior productV of a p-vector and a q-vector is such that ∧ : p (V )× q (V ) →
p+q (V ). The space (V ) is thereby closed with respect to the exterior product, when
one extends by linearity what was discussed in section 2.3.
V
Definition 2.2 ◮ The pair ( (V ), ∧) is called the exterior algebra associated with
the vector space V .
V
Arbitrary elements of (V ) will be called multivectors. An arbitrary multivector
is therefore the sum of a scalar, a 1-vector, a 2-vector, . . . , up to an n-vector or
pseudoscalar. We can write in a general manner
a + vi ei + F ij ei ∧ ej
A = |{z}
|{z} | {z }
scalar vector 2-vector
^ (2.21)
ijk
+T ei ∧ ej ∧ ek + · · · + pe1 ∧ · · · ∧ en ∈ (V ) .
| {z } | {z }
3-vector n-vector
V
The dimension of (V ) can be straightforwardly calculated by using the well-
known result
X n
n n n n n n
+ + +··· + + = = 2n .
0 1 2 n−1 n p=0
p
V n
The dimension of (V ) is 2 , since
^ Xn ^ Xn
n
dim (V ) = dim p (V ) = = 2n . (2.22)
p=0 p=0
p
V V
We denote by h ip the projector h ip : (V ) → p (V ), in such a way that
hAip = A[p] , (2.23)
V
where A[p] ∈ p (V ) is the p-vector component of the multivector A.
The operations grade involution, reversion, and conjugation, as defined in the ten-
sor algebra, descend to the exterior algebra, since they leave the ideal IE in eqn (2.29)
invariant (see section 2.5). The grade involution is given by
b[p] = (−1)p A[p] .
#(A[p] ) = A (2.24)
For the reversion, it follows that
(v1 ∧^
· · · ∧ vp ) = vp ∧ · · · ∧ v1 , (2.25)
which implies that
e[p] = (−1)p(p−1)/2 A[p] .
A (2.26)
The conjugation is known to be the composition of the two operations
eb be
Ā[p] = A [p] = A[p] . (2.27)
Now let us define the following notation:
^ ^
∗
(V ) = exterior algebra of V (V ) = exterior algebra of V ∗
v, u, . . . = vectors α, β, . . . = covectors
A, B, . . . = multivectors Ψ, Φ, . . . = multicovectors
One more word about this notation: whereas A1 and A2 denote two distinct multivec-
tors, A[1] and A[2] respectively denote the 1-vector and 2-vector parts of the multivector
A.
V
The Exterior Algebra (V ) 31
V
Example 2.3 Consider V = R3 , and A, B ∈ (R3 ) given by
A = 2 + 3e1 − e1 ∧ e2 ∧ e3 ,
B = −10 + e2 + 4e1 ∧ e2 + 3e2 ∧ e3 .
The multivector A has scalar, 1-vector and 3-vector (or pseudoscalar) parts, and B has scalar, 1-vector
and 2-vector components. For instance, the projector h i2 acting on those multivectors reads
hAi2 = 0 , hBi2 = 4e1 ∧ e2 + 3e2 ∧ e3 .
The exterior product A ∧ B is then
−20 − 30e1 + 2e2 + 11e1 ∧ e2 + 6e2 ∧ e3 + 19e1 ∧ e2 ∧ e3 ,
whereas the exterior product B ∧ A is
−20 − 30e1 + 2e2 + 5e1 ∧ e2 + 6e2 ∧ e3 + 19e1 ∧ e2 ∧ e3 .
It is then clear that the exterior product between multivectors is neither commutative nor anti-commutative.
Such properties generally only hold when the multivectors are simple, as shown in eqn (2.15). Furthermore,
A ∧ A = 4 + 12e1 − 4e1 ∧ e2 ∧ e3 .
V
Indeed, v ∧ v = 0 always when v ∈ V ⊂ (V ), since in this case the exterior product is anti-commutative
and consequently the product of an element with itself must equal 0. If the exterior product of the
multivectors is commutative or if it is neither commutative nor anti-commutative, in general the exterior
product of an element with itself does not necessarily equal 0.
Example 2.4 Let W be a vector subspace of V = R4 , generated by the vectors v1 = (2, 1, 0, 1)⊺ and
v2 = (1, 0, 0, −1)⊺ . The 2-vector IW = v1 ∧v2 is the pseudoscalar in W . A vector v is in W if v∧IW = 0.
Indeed, IW is given by
IW = v1 ∧ v2 = (2e1 + e2 + e4 ) ∧ (e1 − e4 )
= −e1 ∧ e2 − 3e1 ∧ e4 − e2 ∧ e4 .
When v ∧ IW is calculated, it follows that
v ∧ IW =(−v 1 + 3v 2 − v 4 )e1 ∧ e2 ∧ e4 + (−v 3 )e1 ∧ e2 ∧ e3
+ (3v 3 )e1 ∧ e3 ∧ e4 + (v 3 )e2 ∧ e3 ∧ e4 ,
where we write v = v i ei . The condition v ∧ IW = 0 then reveals that
v 1 − 3v 2 + v 4 = 0, v 3 = 0.
Taking v 2 = a, and v 4 = a − b, it follows that v 1 = 2a + b, v can be written as v = av1 + bv2 , that is,
v ∈ W . The space W can be characterised by the 2-vector v1 ∧v2 . On the other hand, if v∧IW 6= 0, there
exists another vector space W ′ characterised by its pseudoscalar IW ′ = v ∧ IW . Let v3 = (1, 0, −1, 1)⊺
be such a vector. Indeed,
IW ′ =v3 ∧ IW = v1 ∧ v2 ∧ v3 = −e1 ∧ e2 ∧ e3
− 2e1 ∧ e2 ∧ e4 + 3e1 ∧ e3 ∧ e4 + e2 ∧ e3 ∧ e4 .
The condition v ∈ W′ is equivalent to v ∧ IW ′ and implies that
v ∧ IW ′ = (v 1 + 3v 2 − 2v 3 − v 4 )e1 ∧ e2 ∧ e3 ∧ e4 .
Taking v2= a, together with v 3 = −c, and v 4 = a − b + c, it follows that v 1 = 2a + b + c, and
v = av1 + bv2 + cv3 .
Example 2.5 An interesting and prominent application of the exterior algebra is the solution of a linear
system of equations. Consider a system of 3 equations and 3 variables (the generalisation for a system of
n equations and n variables is trivial):
a11 x1 + a12 x2 + a13 x3 = y1 ,
a21 x1 + a22 x2 + a23 x3 = y2 ,
a31 x1 + a32 x2 + a33 x3 = y3 ,
that can be forthwith written as

    
a11 a12 a13 x1 y1
a21 a22 a23  x2  = y2  .
a31 a32 a33 x3 y3
Now let us take a basis {e1 , e2 , e3 } of R3 and define the following vectors:
a1 = a11 e1 + a21 e2 + a31 e3 ,
a2 = a12 e1 + a22 e2 + a32 e3 ,
a3 = a13 e1 + a23 e2 + a33 e3 ,
y = y1 e1 + y2 e2 + y3 e3 .
The linear systems of equations can thus be expressed as
x1 a1 + x2 a2 + x3 a3 = y.
The exterior product between the vector equation and the 2-vector a1 ∧ a2 can be calculated. Since
a1 ∧ a2 ∧ a1 = a1 ∧ a2 ∧ a2 = 0, it follows that
x3 a1 ∧ a2 ∧ a3 = a1 ∧ a2 ∧ y.
Now let us calculate the exterior products. First,
a1 ∧ a2 =(a11 a22 − a21 a12 )e1 ∧ e2 + (a11 a32 − a31 a12 )e1 ∧ e3 + (a21 a32 − a31 a22 )e2 ∧ e3 .
The exterior product of this 2-vector and a3 is given by
a1 ∧ a2 ∧ a3 =[a33 (a11 a22 − a21 a12 ) − a23 (a11 a32 − a31 a12 )
+ a13 (a21 a32 − a31 a22 )]e1 ∧ e2 ∧ e3 ,
which can be regarded, once the determinant function is introduced, as
a11 a12 a13
a1 ∧ a2 ∧ a3 = a21 a22 a23 e1 ∧ e2 ∧ e3 .
a31 a32 a33
On the other hand, the exterior product a1 ∧ a2 ∧ y implies that
a1 ∧ a2 ∧ y =[y3 (a11 a22 −a21 a12 )−y2 (a11 a32 −a31 a12 )+y1 (a21 a32 −a31 a22 )]e1 ∧ e2 ∧ e3 ,
which can be expressed as
a11 a12 y1
a1 ∧ a2 ∧ y = a21 a22 y2 e1 ∧ e2 ∧ e3 .
a31 a32 y3
Therefore, the vector equation x3 a1 ∧ a2 ∧ a3 = a1 ∧ a2 ∧ y has a solution if a1 ∧ a2 ∧ a3 6= 0, that is, if
a11 a12 a13
a21 a22 a23 6= 0,
a31 a32 a33
where in this case x3 is given by
a11 a12 y1
a21 a22 y2
a31 a32 y3
x3 = .
a11 a12 a13
a21 a22 a23
a31 a32 a33
Repeating the same procedure but using instead the 2-vectors a1 ∧ a3 and a2 ∧ a3 , we find that
a11 y1 a13 y1 a12 a13
a21 y2 a23 y2 a22 a23
a31 y3 a33 y3 a32 a33
x2 = , x1 = .
a11 a12 a13 a11 a12 a13
a21 a22 a23 a21 a22 a23
a31 a32 a33 a31 a32 a33
The Exterior Algebra as the Quotient of the Tensor Algebra 33
The final expressions for x1 , x2 , and x3 comprise the well-known Cramer’s rule. Clearly, this procedure can
be immediately extended to high dimensions, corresponding to systems of linear equations of high order
and can be straightforwardly computed.
2.5 The Exterior Algebra as the Quotient of the Tensor Algebra

In this section, we present another construction of the exterior algebra. Besides it
being a prominent construction itself, a construction that is analogous to it can be
used for Clifford algebras. Therefore, this approach is particularly interesting.
Equivalence Relations
An equivalence relation R in a set X is a relation that is (i) reflexive, (ii) symmetric,
and (iii) transitive, namely, (i) xRx; (ii) if xRy then yRx; and (iii) if xRy and yRz,
then xRz, ∀x, y, z ∈ X. The set of all elements that are equivalent to x constitutes
an equivalence class of x, denoted by [x] = {y ∈ X | yRx}. The set of such (disjoint)
equivalence classes is denoted by X = X/R = {[x] | x ∈ X}. A notation that is
frequently used for xRy is x ∼ y.
Example 2.6 In the set of integers Z, we can define the following equivalence relation:
m ∼ n ⇔ m − n = kN, k = 0, ±1, ±2, . . . ,
namely, that the integers m and n are considered equivalent if they differ by a multiple of a fixed natural
number N . The equivalence class of m consists therefore of the set
[m] = {n ∈ Z|n = m + kN } = {m, m ± N, m ± 2N, m ± 3N, . . .}.
The set of all equivalence classes is the set ZN = Z/ ∼, the set of integers modulo N .
Ideals
Let A denote an algebra. A set IL ⊂ A is said to be a left ideal of A if, for all a ∈ A
and for all x ∈ IL , the relation ax ∈ IL holds. Analogously, IR ⊂ A is said to be a
right ideal of A if, for all a ∈ A and for all x ∈ IR , it follows that xa ∈ IR . The set
I ⊂ A is said to be a bilateral ideal (or simply an ideal) of A if, ∀a, b ∈ A and ∀x ∈ I,
it implies that axb ∈ I.
Given an algebra A, let us write it as A = B + C. Here, it is not supposed a priori
that B and C are subalgebras. Notwithstanding, only the properties of the vector space
which underlies the algebra A are emphasised. Moreover, this sum is not supposed to
be a direct sum. Notice that only A and C are relevant, as it is always possible to set
B = A. Of course, if A = B ⊕ C, then the quotient A/C can be identified as B. Given
a, b ∈ A, the following equivalence relation can be defined:
a ∼ b ⇐⇒ a = b + x, x ∈ C. (2.28)
The set A/ ∼ of the equivalence classes has a natural vector space structure, with the
sum of vectors defined by
[a] + [b] = [a + b]
and the multiplication by scalars d ∈ K defined by
d[a] = [da].
Now, the set A/ ∼ is also intended to present an algebra structure. A natural way to
define the product between equivalence classes is
[a][b] = [ab].
According to the definition given in eqn (2.28), regarding the equivalence relation for
[a], [b] ∈ A/ ∼, it follows that [a] = [a+x], and [b] = [b+y], where x, y ∈ C; in addition,
[a][b] = [a + x][b + y] = [(a + x)(b + y)] = [ab + ay + xb + xy].
The last two equations result in
[ab] = [ab + ay + xb + xy],
namely
ay + xb + xy ∈ C, ∀a, b ∈ A, ∀x, y ∈ C ,
which only holds if C is an (bilateral) ideal. This algebra is called the quotient algebra
of A by the ideal C and is denoted by A/C.
The Exterior Algebra as a Quotient of the Tensor Algebra

Let T(V ) denote the algebra of contravariant tensors. Let IE be the ideal of T(V ),
generated by the elements v ⊗ v, v ∈ V . In other words, the ideal IE consists of the
sums X
Ai ⊗ vi ⊗ vi ⊗ Bi , (2.29)
i
where vi ∈ V , and Ai , Bi ∈ T(V ). Moreover, the ideal IE is generated by the elements

v ⊗ u + u ⊗ v, where v, u ∈ V . In addition,
v ⊗ u + u ⊗ v ∈ ker Alt,
and ker(Alt) = IE . Let us now show that the exterior algebra is isomorphic to the
quotient algebra T(V )/IE . The respective equivalence class is given by eqn (2.28) with
the identification C = IE , namely
A ∼ B ⇐⇒ A = B + x, x ∈ IE .
The equivalence class of A is denoted by [A], and the product is denoted by
[A] ∧ [B] = [A ⊗ B]. (2.30)
For all v, u ∈ V , it is clear that [v] ∼ v, and [u] ∼ u. It is possible to calculate v ∧ u

according to the definition given in eqn (2.30), since
1 1
v⊗u = (v ⊗ u − u ⊗ v) + (v ⊗ u + u ⊗ v),
2 2
The Exterior Algebra as the Quotient of the Tensor Algebra 35
where 12 (v ⊗ u + u ⊗ v) ∈ IE . Indeed,
(v ⊗ u + u ⊗ v) = (v + u) ⊗ (v + u) − v ⊗ v − u ⊗ u,
showing that this quantity is an element of IE . We conclude that
1
v⊗u∼v∧u= (v ⊗ u − u ⊗ v), (2.31)
2
or [v ⊗ u] = [v ∧ u], and [v] ∧ [u] = [v ∧ u].

This result can be generalised as
v1 ⊗ · · · ⊗ vp ∼ Alt(v1 ⊗ · · · ⊗ vp ) = v1 ∧ · · · ∧ vp . (2.32)
In fact, v1 ⊗ · · · ⊗ vp is congruent to every (v1 ⊗ · · · ⊗ vp )σ , since this assertion holds

when σ is the transposition of two consecutive elements of {1, 2, . . . , p}, and these
transpositions generate the group Sp .
The congruence (2.32) has prominent consequences, summarised in the following
theorem:
V
Theorem ◮ T(V ) is the direct sum of (V ) and ker(Alt), V which is equal to the ideal
IE . Moreover, A ∧ B ∼ A ⊗ B mod IE for all A, B ∈ (V ).
V
Proof: It is clear that T(V ) is the direct sum of (V ) (the imageVof Alt) and ker(Alt),
and that IE ⊂ ker(Alt). Besides, eqn (2.32) proves that T(V ) = (V ) + IE because it
shows that A ∼ Alt(A) modulo IE for all A ∈ T(V ), proving the first assertion. The
second assertion follows if we remember that A[p] ∧ B[q] = Alt(A[p] ⊗ B[q] ). ✓
Theorem 2.1 allows to prove by induction on p that Alt(v1 ⊗· · ·⊗vp ) = v1 ∧· · ·∧vp .
In addition, an immediate consequence of this theorem is that (up to an isomorphism)
we have
^
(V ) = T(V )/IE . (2.33)
V
The universal property of (V ) follows from eqn (2.33). In fact, since IE is also the
ideal of T(V ), as generated by all v⊗v, this property can be stated as follows:
V for every
(associative unital) algebra A, by mapping every algebra homomorphism (V ) → A
to its restriction V → A, we obtain a bijection from the set of all such homomorphisms
onto the set of all linear mappings f : V → A such that f (v)2 = 0A , for all v ∈ V .
Here, by definition, an algebra homomorphism must map the unit element to the unit
element.
Besides the tensor algebra and, as we shall also see, the Clifford algebras and
the exterior algebras, other objects have universal properties that are often useful
in the study of Clifford algebras, such as, for example, the tensor product and the
twisted tensor product of two (or several) algebras, the quotient of a vector space by a
subspace, and the quotient of an algebra by an ideal. Moreover, the grade involution,
the reversion, and consequently the conjugation
V preserve the ideal IE and thus pass
to the quotient, being also defined in (V ).
2.6 The Contraction, or Interior Product

It has already been seen that, given a vector v ∈ V and a covector α ∈ V ∗ , there exists
an associated scalar, α(v). Likewise, this property can be interpreted as representing
the action of the linear functional α on v (in terms of the mapping α : V → R), or
the action of the linear functional ξv on α via the relation ξv (α) = α(v) (in this case,
in terms of the mapping ξv : V ∗ → R). The quantity α(v) can thus be interpreted as
a kind of product between α and the vector v, namely, the operation V × V ∗ → R.
Similarly, this operation can be defined in such a way that V ∗ × V → R. V
V In the context of the exterior algebra, we can equivalently write α : 1 (V ) →
0 (V ). The natural question arising is whether
V this viewpoint
V can be generalised,
in order to define an operation such that p (V ) → p−1 (V ), for all p such that
(0 ≤ p ≤ n). In what follows this generalisation is accomplished.
The Left Contraction

Let A[p] be a p-vector and
V let α be a covector. We V aim to construct an operation that
when acting on A[p] ∈ p (V ) yields an element of p−1 (V ), namely, a (p − 1)-vector.
The left contraction of a p-vector A[p] by a covector α, denoted from this point on by
α⌋, is defined as
(α⌋A[p] )(α1 , α2 , . . . , αp−1 ) = pA[p] (α, α1 , α2 , . . . , αp−1 ) , (2.34)
where α1 , . . . , αp−1 are arbitrary covectors. Taking A[p] = v1 ∧ · · · ∧ vp on the right-

hand side of eqn (2.34) means that
(v1 ∧ · · · ∧ vp )(α, α1 , . . . , αp−1 )

1 X
= ε(σ)α(vσ(1) )α1 (vσ(2) ) . . . αp−1 (vσ(p) ).
p!
σ∈Sp
This definition clearly shows that α⌋A[p] is a (p − 1)-vector. In the case where a
covector v is considered, this definition obviously leads to
α⌋v = α(v). (2.35)

V
For an element of 0 (V ), it is assumed that
α⌋1 = 0. (2.36)
Observation ☞ The left contraction is also called the interior product and can be
denoted by i(α), for example, the interior product between A[p] and the covector α
is denoted by i(α)(A[p] ) (Frankel, 2012). However, this notation is not as suitable for
our purposes as the other is. As will be shown, the right contraction emulates the left
contraction. Using the notation α⌋ to denote the left contraction makes this fact clear,
so we will continue to use this notation.
The Contraction, or Interior Product 37
Properties of the Left Contraction

Although the definition given in eqn (2.34) is appropriate for the formalism, it is not
the best one to use for computation. However, an important and useful property can
be introduced from this definition. First, let us consider the contraction of a 2-vector,
which can be considered, without loss of generality, to be a simple 2-vector v ∧ u.1
Using the definition given in eqn (2.34) as well as the definition of the exterior product,
given α, β ∈ V ∗ , it follows that
(α⌋(v ∧ u))(β) = 2(v ∧ u)(α, β) = α(v)β(u) − α(u)β(v)

= α(v)u − α(u)v (β) = (α⌋v)u − (α⌋u)v (β).
Moreover, as this expression must hold for any covector β, it follows that
α⌋(v ∧ u) = (α⌋v)u − v(α⌋u) = α(v)u − vα(u) . (2.37)
Now this result can be generalised for the 3-vector v∧u∧w. Again, using eqn (2.34)
together with the definition of the exterior product, we obtain
(α⌋(v ∧ u ∧ w))(β, γ) = 3(v ∧ u ∧ w)(α, β, γ)

= (3/3!)[α(v)β(u)γ(w) + β(v)γ(u)α(w) + γ(v)α(u)β(w)
− γ(v)β(u)α(w) − β(v)α(u)γ(w) − α(v)γ(u)β(w)]
= α(v)(1/2!)[β(u)γ(w) − β(w)γ(u))]
− α(u)(1/2!)[(β(v)γ(w) − β(w)γ(v)]
+ α(w)(1/2!)[β(v)γ(u) − β(u)γ(v)]
= α(v)(u ∧ w)(β, γ) − α(u)(v ∧ w)(β, γ) + α(w)(v ∧ u)(β, γ).
Since expression must hold for any covectors β and γ, it follows that
α⌋(v ∧ u ∧ w) = (α⌋v)u ∧ w − (α⌋u)v ∧ w + (α⌋w)v ∧ u

(2.38)
= α(v)u ∧ w − α(u)v ∧ w + α(w)v ∧ u.
Using now eqns (2.37) and (2.38), together with the associativity of the exterior
product, we obtain
α⌋((v ∧ u) ∧ w) = (α⌋(v ∧ u)) ∧ w + v ∧ u(α⌋w), (2.39)

α⌋(v ∧ (u ∧ w)) = (α⌋v)u ∧ w − v ∧ (α⌋(u ∧ w)). (2.40)
where the contraction of the exterior product between a 1-vector and a 2-vector is
considered. It is important to note on the right-hand side of these equations the pres-
ence of either a positive or negative sign, according to the presence of a 2-vector (v ∧ u
in eqn (2.39)) or a 1-vector (v in eqn (2.40)), respectively. The different signs can be
taken into account by the use of the grade involution given by (2.24).
1 It is enough to consider a simple 2-vector, because of the linearity of the contraction, which can
be forthwith verified.
The generalisation of these equations for the exterior product of a p-vector and a
q-vector is provided by
α⌋(A[p] ∧ B[q] ) = (α⌋A[p] ) ∧ B[q] + (−1)p A[p] ∧ (α⌋B[q] ),
b[p] = #A[p] = (−1)p A[p] is used. Because of the linearity of the

where the expression A
contraction, we can write in a general manner that
b ∧ (α⌋B) ,
α⌋(A ∧ B) = (α⌋A) ∧ B + A (2.41)
where A and B are arbitrary multivectors. Notice that this expression resembles the
Leibniz rule in differential calculus for the derivative of the product of functions. In
fact, it is the graded Leibniz rule, so called because of the presence of the grade
involution.
The Right Contraction

In eqn (2.34), which defines the left contraction, the covector α is the first argument
of a p-vector. However, it would also be correct to locate the covector α in the last
argument of that p-vector; there is no reason to prefer one or the other case. This
observation gives rise to another definition: the right contraction of a p-vector A[p] by
the covector α, denoted by ⌊α, is defined as
(A[p] ⌊α)(α1 , α2 , . . . , αp−1 ) = pA[p] (α1 , α2 , . . . , αp−1 , α) , (2.42)
where α1 , . . . , αp−1 are arbitrary covectors.

Given the definition in eqn (2.42), in the case where p = 1, the contraction of a
1-vector v by the covector α is immediately reduced to
v⌊α = α(v) = α⌋v . (2.43)

V
Analogously, for an element of 0 (V ), the result equals 0:
1⌊α = 0. (2.44)
The differences between the left contraction and the right contraction can be seen
when the right contraction of a 2-vector v ∧ u is considered. Using the definition
given in eqn (2.42), and a reasoning similar to that used to obtain eqn (2.37), we can
conclude that
(v ∧ u)⌊α = v(u⌊α) − u(v⌊α) = α(u)v − α(v)u. (2.45)
It is not difficult to demonstrate the analogue of eqn (2.41) for the right contraction.
The result is
(A ∧ B)⌊α = A ∧ (B⌊α) + (A⌊α) ∧ B b, (2.46)
where A and B are arbitrary multivectors.

The Contraction, or Interior Product 39
The relationship between the left and right contractions can be straightforwardly
obtained from their respective definitions. Indeed, it is enough to realise that
A[p] (α1 , . . . , αp−1 , α) = (−1)p−1 A[p] (α, α1 , . . . , αp−1 ),
which implies that

α⌋A[p] = (−1)p−1 A[p] ⌊α. (2.47)
In addition, the general expression
b
α⌋A = −A⌊α (2.48)
can be obtained for an arbitrary multivector A.
Generalisation of the Left and the Right Contractions

It is well-known that the exterior product of covectors is a p-covector (or, in general
terms, a multicovector). It is thus natural to generalise the definitions of the left and
the right contractions for a p-covector.
∗
V First, we need to verify that α⌋α⌋A[p] = 0 = A[p] ⌊α⌊α, for α ∈ V and for A[p] ∈
(V ). As usual, let us consider covectors α1 , . . . , αp−2 . Thus, eqn (2.34) implies that
(α⌋(α⌋A[p] ))(α1 , . . . , αp−2 ) = (p − 1)(α⌋A[p] )(α, α1 , α2 , . . . , αp−2 )

= p(p − 1)A[p] (α, α, α2 , . . . , αp−2 )
1 X
= ε(σ)α(vσ(1) ) . . . αp−2 (vσ(p) ) (2.49)
(p − 2)!
σ∈Sp
= 0.
The case A[p] ⌊α⌊α can be proved analogously. Moreover,
1⌊A = A = A⌋1 ,
where A denotes
V an arbitrary multivector, and weVmust be aware of the universal
property of (V ∗ ). In fact, the mapping V ∗ → End( (V V )), which maps every
V covector
A 7→ α⌋A, extends to an algebra homomorphism (V ∗ ) → End( (V )), where
α to V
End(V (V )) stands for the space of endomorphisms V of V . In addition, the unit element
1 of (V ∗ ) is mapped to the unit element of End( (V )).
Hence, given a p-covector α1 ∧ α2 ∧ · · · ∧ αp , we can define
(α1 ∧ α2 ∧ · · · ∧ αp )⌋ = α1 ⌋α2 ⌋ · · · αp ⌋ , (2.50)
and
⌊(α1 ∧ α2 ∧ · · · ∧ αp ) = ⌊α1 ⌊α2 · · · ⌊αp . (2.51)
From these definitions, it immediately follows that the (left or right) contraction of a
p-vector by a q-covector equals 0 if q > p.
These definitions shall now be explored for the case of a 2-covector α ∧ β. Let us
consider a 2-vector v ∧ u. Equation (2.50) means that
(α ∧ β)⌋(v ∧ u) = α⌋(β⌋(v ∧ u)), (2.52)
and the result following from the use of eqn (2.37) is
(α ∧ β)⌋(v ∧ u) = α(u)β(v) − α(v)β(u) . (2.53)
On the other hand, the action of the 2-covector α ∧ β on the 2-vector v ∧ u can be
defined. For the case involving covectors, the definition of tensor products of linear
mappings, given at the end of chapter 1, reads
(α ⊗ β)(v ⊗ u) = α(v)β(u).
Thus, using the definition of the exterior product, we can define (α ∧ β)(v ∧ u) as
follows:
1
(α ∧ β)(v ∧ u) = α(v)β(u) − α(u)β(v) . (2.54)
2
Comparing eqns (2.53) and (2.54) yields
1
(α ∧ β)(v ∧ u) = (β ∧ α)⌋(v ∧ u), (2.55)
2
where the order of the exterior product of the covectors on the two sides of the equation
must be taken into account. Moreover, the appropriate form for this expression is
1 ^
(α ∧ β)(v ∧ u) = (α ∧ β)⌋(v ∧ u). (2.56)
2
These results can be immediately generalised. If A[p] is a p-vector, and Ψ[p] a
p-covector, this generalisation is provided by
1 e [p]
Ψ[p] (A[p] ) = Ψ ⌋A[p] . (2.57)
p!
In the case involving a simple p-vector and a simple p-covector, this equation can be
written as
1
(α1 ∧ · · · ∧ αp )(v1 ∧ · · · ∧ vp ) = (αp ∧ · · · ∧ α1 )⌋(v1 ∧ · · · ∧ vp )
p!
α1 (v1 ) α1 (v2 ) . . . α1 (vp )
(2.58)
1 α2 (v1 ) α2 (v2 ) . . . α2 (vp )
= .. .. .. .. .
p! . . . .
αp (v1 ) αp (v2 ) . . . αp (vp )
The (left or right) contraction involving a p-covector and a q-vector can be similarly
defined. This construction shall not be shown here, since it presents neither difficulties
nor novelties. However, two related equations that are interesting are
Orientation, and Quasi-Hodge Isomorphisms 41
Ψ[p] ⌋A[q] = (−1)p(q−1) A[q] ⌊Ψ[p] , (2.59)
where p ≤ q, and
A[q] ⌋Ψ[p] = (−1)q(p−1) Ψ[p] ⌊A[q] , (2.60)
where q ≤ p. Another result is given by
Ψ^
[p] ⌋A [p]
[q] = Ã[q] ⌊Ψ̃ . (2.61)
In addition, when p = q, it follows that
Ψ[p] ⌋A[p] = A[p] ⌊Ψ[p] = A[p] ⌋Ψ[p] = Ψ[p] ⌊A[p] , (2.62)
which can be used to write eqn (2.57) as other, equivalent expressions.
Example 2.7 Let α and β be the covectors

α = 5e1 − 2e2 , β = e2 + 3e3 − e4 ,
and let A be the multivector
A = e1 ∧ e2 ∧ e3 + 2e1 ∧ e4 .
First, α⌋A is calculated:
α⌋A = (5e1 − 2e2 )⌋(e1 ∧ e2 ∧ e3 ) + (5e1 − 2e2 )⌋(2e1 ∧ e4 )
= 5e1 ⌋(e1 ∧ e2 ∧ e3 ) − 2e2 ⌋(e1 ∧ e2 ∧ e3 ) + 10e1 ⌋(e1 ∧ e4 ) − 4e2 ⌋(e1 ∧ e4 )
= 5e2 ∧ e3 + 2e1 ∧ e3 + 10e4 ,
since e2 ⌋(e1 ∧ e4 ) = 0. On the other hand, β⌋A is given by
β⌋A = (e2 + 3e3 − e4 )⌋(e1 ∧ e2 ∧ e3 ) + (e2 + 3e3 − e4 )⌋(2e1 ∧ e4 )
= e2 ⌋(e1 ∧ e2 ∧ e3 ) + 3e3 ⌋(e1 ∧ e2 ∧ e3 ) − e4 ⌋(e1 ∧ e2 ∧ e3 )
+ e2 ⌋(2e1 ∧ e4 ) + 3e3 ⌋(2e1 ∧ e4 ) − e4 ⌋(2e1 ∧ e4 )
= −e1 ∧ e3 + 3e1 ∧ e2 + 2e1 ,
where we use e4 ⌋(e 1 ∧ e2 ∧ e3 ) = e2 ⌋(2e1 ∧ e4 ) = e3 ⌋(2e1 ∧ e4 ) = 0. In addition,
β⌋(α⌋A) = 5e3 − 15e2 − 6e1 − 10,
and
α⌋(β⌋A) = −5e3 + 15e2 + 10 + 6e1 .
These two last results are expected, since
β⌋(α⌋A) = (β ∧ α)⌋A = −(α ∧ β)⌋A = −α⌋(β⌋A).
Finally, with respect to the right contraction, it follows that
A⌊α = 5e2 ∧ e3 − 10e4 + 2e1 ∧ e3 .
The relationship between α⌋A and A⌊α is obvious, and, in this case, given by eqn (2.48).
2.7 Orientation, and Quasi-Hodge Isomorphisms

Before discussing the Hodge isomorphism, let us define an isomorphism which is called
the ‘quasi-Hodge’ isomorphism. In order to introduce it, it is necessary to choose an
n-vector (pseudoscalar ).
Orientation
There is no criteria that would induce us to prefer the choice of some specific n-vector.
If a basis B = {e1 , . . . , en } of V is taken into account, it is natural to choose the
n-vector given by the exterior product of the basis vectors B. No matter the order
that such exterior products are taken – for instance, e1 ∧ · · · ∧ en , or en ∧ · · · ∧ e1 , or
even eσ(1) ∧ · · · ∧ eσ(n) , ∀σ ∈ Sn – the result can be always written as
±e1 ∧ e2 ∧ · · · ∧ en .
Indeed, much more information can be extracted from it. Given any set of n linearly
independent vectors {v1 , . . . , vn }, it follows that
v1 ∧ · · · ∧ vn = Je1 ∧ · · · ∧ en ,
where J > 0, or J < 0.2 This result implies that, once an n-vector has been chosen,
any other set of linearly independent n vectors can be represented by one of the two
following classes: (i) that one for which J > 0, and (ii) that one such that J < 0.
In other words, when an n-vector Ω has been chosen, we can define two equivalence
classes, [Ω] and [−Ω], as
^
[Ω] = {A[n] ∈ n (V ) | A[n] = JΩ, J > 0},
^
[−Ω] = {A[n] ∈ n (V ) | A[n] = JΩ, J < 0}.
These two equivalence classes comprise the vector space orientation. There exists
two possible orientations for a vector space, completely determined once an n-vector
has been selected. The choice of an n-vector ΩV in an equivalence class defines a pair
constituted by a vector space V endowed with an n-vector ΩV .
We can immediately define an orientation for the dual space V ∗ via the n-covector
ΩV ∗ . The question to be posed is whether the orientation of V ∗ can be related to the
orientation of V . In order to answer this question, we write
ΩV = Je1 ∧ · · · ∧ en , ΩV ∗ = J ′ e1 ∧ · · · ∧ en .
Using eqn (2.57) and (2.58), we can see that

1e 1 1
ΩV ∗ (ΩV ) = ΩV ∗ ⌋ΩV = JJ ′ (en ∧ · · · ∧ e1 )⌋(e1 ∧ · · · ∧ en ) = JJ ′ ,
n! n! n!
in such a way that
(
> 0 if J > 0 and J ′ > 0; or J < 0 and J ′ < 0,
0 6= ΩV ∗ (ΩV ) =
< 0 if J > 0 and J ′ < 0; or J < 0 and J ′ > 0.
As ΩV ∗ (ΩV ) > 0, or ΩV ∗ (ΩV ) < 0, we can use this fact to relate both the orientations
of V and V ∗ .
2 Let us remember that, if these n vectors are not linearly independent, the exterior product among
them equals 0.
The orientations of V and V ∗ determined by the n-vector ΩV and the n-covector

ΩV ∗ , respectively, are called compatible if ΩV ∗ (ΩV ) > 0. Hence, if the orientations of
V and V ∗ are compatible, once the orientation for one of those spaces are chosen, the
orientation for the other space is forthwith defined.
In order to define ΩV ∗ given ΩV , in such a way that both orientations are com-
patible, let us choose the most natural way. If with respect to a basis B = {ei } we
have
ΩV = e1 ∧ · · · ∧ en , (2.63)
then, in terms of a dual basis B = {ei },
ΩV ∗ = e1 ∧ · · · ∧ en . (2.64)
It follows that
e V ∗ ⌋ΩV = 1.
Ω (2.65)
Quasi-Hodge Isomorphisms
Given the pair (V, ΩV ), it is possible to define the two operators ⋆ and ⋆, which will
be called quasi-Hodge isomorphisms:
^ ^ ^ ^
n−p p
⋆: p (V ) → (V ), ⋆: (V ) → n−p (V ).
V
Given A[p] ∈ p (V ), we define ⋆A[p] as
p
^
[p] [p] [p]
Ψ ∧ ⋆A[p] = p!Ψ (A[p] )ΩV ∗ , ∀Ψ ∈ (V ) . (2.66)
Vp
Similarly, given Ψ[p] ∈ (V ), we define ⋆Ψ[p] as
^
A[p] ∧ ⋆Ψ[p] = p!Ψ[p] (A[p] )ΩV , ∀A[p] ∈ (V ) . (2.67)
p
Now let us focus on some of the properties and applications of these definitions,
considering only the case for ⋆, since both cases can be similarly analysed.
V0 In eqns
(2.66)
V and (2.67), when p = 0, it is implicit that a(b) =
V ab for a ∈ (V ) = R, and
b ∈ 0 (V ) = R. Taking this fact into account, for 1 ∈ 0 (V ), eqn (2.66) reads
a ∧ ⋆1 = a(1)ΩV ∗ = aΩV ∗ ,
so
⋆1 = ΩV ∗ . (2.68)
For v ∈ V , let us calculate ⋆v. According to eqn (2.66), it follows that
α ∧ ⋆v = α(v)ΩV ∗ = (α⌋v)ΩV ∗ = (v⌋α)ΩV ∗ , (2.69)
where eqn (2.62) was used in order to express the right-hand side in terms of the
contraction of a covector by a vector; the reason for this approach will be made clear
Vn V1
in what follows. Since ΩV ∗ ∈ (V ), thus α ∧ ΩV ∗ = 0, ∀α ∈ (V ). The contraction
v⌋ satisfies the graded Leibniz rule
b ∧ (v⌋Φ),
v⌋(Ψ ∧ Φ) = (v⌋Ψ) ∧ Φ + Ψ
where Ψ and Φ are multicovectors, and it implies that
0 = v⌋(α ∧ ΩV ∗ ) = (v⌋α)ΩV ∗ − α ∧ (v⌋ΩV ∗ ).
Hence, eqn (2.69) can be written as
α ∧ ⋆v = α ∧ (v⌋ΩV ∗ ). (2.70)
V1
Since this expression holds for all α ∈ (V ), it reads
⋆v = v⌋ΩV ∗ . (2.71)
Now let us consider the p-vector A[p−1] ∧ v. According to eqn (2.66),

^∧ v)⌋Ψ[p] ΩV ∗ ,
Ψ[p] ∧ ⋆(A[p−1] ∧ v) = p!Ψ[p] (A[p−1] ∧ v)ΩV ∗ = (A[p−1] (2.72)
where eqns (2.57) and (2.62) were employed; consequently, we have

Ψ[p] ∧ ⋆(A[p−1] ∧ v) = (v ∧ A e[p−1] )⌋Ψ[p] ΩV ∗

= (−1)p−1 (A e[p−1] ∧ v)⌋Ψ[p] ΩV ∗ (2.73)

= (−1)p−1 A e[p−1] ⌋(v⌋Ψ[p] ) ΩV ∗ ,
where eqn (2.15) and the analogue for eqn (2.50) for the case involving vectors were
used. Again by eqn (2.66), the last term of eqn (2.73) reads

(−1)p−1 A e[p−1] ⌋(v⌋Ψ[p] ) ΩV ∗ = (−1)p−1 (v⌋Ψ[p] ) ∧ ⋆A[p−1] . (2.74)
On the other hand, according to the graded Leibniz rule,
v⌋(Ψ[p] ∧ ⋆A[p−1] ) = (v⌋Ψ[p] ) ∧ ⋆A[p−1] + (−1)p Ψ[p] ∧ (v⌋⋆A[p−1] ). (2.75)

Vn−(p−1) Vn−p+1
But ⋆A[p−1] ∈ (V ) = (V ); consequently, it follows that
Ψ[p] ∧ ⋆A[p−1] = 0.
Finally, the expression
(v⌋Ψ[p] ) ∧ ⋆A[p−1] = −(−1)p Ψ[p] ∧ (v⌋⋆A[p−1] ) (2.76)
is obtained, and we can substitute it into eqn (2.74) and subsequently eqn (2.73),
obtaining
Ψ[p] ∧ ⋆(A[p−1] ∧ v) = Ψ[p] ∧ (v⌋⋆A[p−1] ). (2.77)
Vp
As it must hold ∀Ψ[p] ∈ (V ), it then reads
⋆(A[p−1] ∧ v) = (v⌋⋆A[p−1] ). (2.78)
Now eqn (2.71) can be used recursively, together with eqn (2.78), in order to obtain
⋆(v1 ∧ · · · ∧ vp ) = (vp ∧ · · · ∧ v1 )⌋ΩV ∗ , (2.79)
and, when extended by linearity, it yields
e[p] ⌋ΩV ∗ .
⋆A[p] = A (2.80)
This equation and eqn (2.68) completely determine the quasi-Hodge isomorphism ⋆.
The calculations for ⋆ are similar and they imply that
⋆1 = ΩV , (2.81)
and
e [p] ⌋ΩV .
⋆Ψ[p] = Ψ (2.82)
It is possible to show that the isomorphisms ⋆ and ⋆ are the inverse of each other,
up to a sign. Specifically:
⋆⋆ = ⋆⋆ = (−1)p(n−p) 1, (2.83)
where 1 denotes the identity operator, and it is natural to define
⋆ −1 = (−1)p(n−p) ⋆, ⋆−1 = (−1)p(n−p) ⋆. (2.84)
Remembering the assumption about the compatibility between the orientations of
V and V ∗ , and therefore the validity of the relation Ω̃V ∗ ⌋ΩV = Ω̃V ⌋ΩV ∗ = 1, by using
eqn (2.66) together with eqn (2.57 and eqn (2.62), we obtain
(Ψ[p]^
∧ ⋆A[p] )⌋ΩV = Ã[p] ⌋Ψ[p] (Ω̃V ∗ ⌋ΩV ) = Ã[p] ⌋Ψ[p] . (2.85)
Alternatively, from eqn (2.82), we obtain
⋆(Ψ[p] ∧ ⋆A[p] ) = Ã[p] ⌋Ψ[p] . (2.86)
On the other hand, the definition given in eqn (2.67) implies that
(A[p]^
∧ ⋆Ψ[p] )⌋ΩV ∗ = Ã[p] ⌋Ψ[p] (Ω̃V ∗ ⌋ΩV ∗ ) = Ã[p] ⌋Ψ[p] , (2.87)
and, by using eqn (2.80), we obtain
⋆(A[p] ∧ ⋆Ψ[p] ) = Ã[p] ⌋Ψ[p] . (2.88)
When we compare eqns (2.86) and (2.88), we can see that
⋆(Ψ[p] ∧ ⋆A[p] ) = ⋆(A[p] ∧ ⋆Ψ[p] ). (2.89)
Example 2.8 Let us prove eqn (2.83). Because of linearity, it is enough to prove this equation for A[p]
being a simple p-vector. Given B = {ei } (i = 1, . . . , n), let us write an arbitrary simple p-vector as
A[p] = eµ1 ∧ · · · ∧ eµp , where µi = 1, . . . , n; µi 6= µj (i 6= j); i, j = 1, . . . , p. Let σ be a permutation

of n elements σ : {1, . . . , n} → {σ(1), . . . , σ(n)} such that σ(i) = µi (i = 1, . . . , p). In this way, we can
write this arbitrary simple p-vector as
A[p] = eσ(1) ∧ · · · ∧ eσ(p) .
The n-vectors ΩV and ΩV ∗ can be written as
ΩV = e1 ∧ · · · ∧ ep ∧ ep+1 ∧ · · · ∧ en
= ǫ(σ) eσ(1) ∧ · · · ∧ eσ(p) ∧ eσ(p+1) ∧ · · · ∧ eσ(p) ,
ΩV ∗ = e1 ∧ · · · ∧ ep ∧ ep+1 ∧ · · · ∧ en
= ǫ(σ) eσ(1) ∧ · · · ∧ eσ(p) ∧ eσ(p+1) ∧ · · · ∧ eσ(p) ,
where ǫ(σ) is the sign of the permutation σ. Now for ⋆A[p] we have

⋆A[p] = (eσ(p) ∧ · · · ∧ eσ(1) )⌋ ǫ(σ) eσ(1) ∧ · · · ∧ eσ(p) ∧ eσ(p+1) ∧ · · · ∧ eσ(n)
= ǫ(σ) eσ(p+1) ∧ · · · ∧ eσ(n) .
In a similar way,

⋆(⋆A[p] ) = ǫ(σ) eσ(n) ∧ · · · ∧ eσ(p+1) ⌋ ǫ(σ) eσ(1) ∧ · · · ∧ eσ(p) ∧ eσ(p+1) ∧ · · · ∧ eσ(n)

= eσ(n) ∧ · · · ∧ eσ(p+1) ⌋ (−1)n(p−n) (eσ(p+1) ∧ · · · ∧ eσ(n) ) ∧ (eσ(1) ∧ · · · eσ(p) )
= (−1)p(n−p) eσ(1) ∧ · · · ∧ eσ(p) = (−1)p(n−p) A[p] .
The proof that ⋆⋆ = (−1)p(n−p) is analogous.
Example 2.9 In what follows, quasi-Hodge isomorphisms are used to find the inverse of a matrix. Specif-
ically, consider the matrix A given by  
1 2 0
A = 0 4 1 .

1 −1 1
Take the entries aji of this matrix as the vector components ai = aji ej , namely
a 1 = e1 + e3 ,
a2 = 2e1 + 4e2 − e3 ,
a 3 = e2 + e3 .
In this way, the matrix A is defined as
A = a1 a2 a3 .
By using quasi-Hodge isomorphisms, we can define a new set of covectors {⋆(ai ∧ aj )} (i 6= j):
⋆(ai ∧ aj ) = (aj ∧ ai )⌋ǫ,
where the n-covector ǫ is given by
ǫ = e1 ∧ e2 ∧ e3 .
Now the vectors ā1 , ā2 , and ā3 are defined as
ā1 = ⋆(a2 ∧ a3 ), ā2 = ⋆(a3 ∧ a1 ), ā3 = ⋆(a1 ∧ a2 ).
Computing these quantities, we obtain
ā1 = a3 ⌋(a2 ⌋(e1 ∧ e2 ∧ e3 )) = 5e1 − 2e2 + 2e3 ,
ā2 = a1 ⌋(a3 ⌋(e1 ∧ e2 ∧ e3 )) = e1 + e2 − e3 ,
ā3 = a2 ⌋(a1 ⌋(e1 ∧ e2 ∧ e3 )) = −4e1 + 3e2 + 4e3 .
It straightforwardly follows that
āi (aj ) = 7δji , (i, j = 1, 2, 3),
and that
det A = 7.
These results can be used to obtain the inverse matrix of A, as follows. First, place the vector components
(according to our convention) into a matrix column; then, the covector components must be placed in the
The Regressive Product 47
corresponding rows of the matrix. Thus, using the covectors ā1 , ā2 , and ā3 , the matrix Ā can be defined
as  1  
ā 5 −2 2
Ā = ā  =  1 1 −1 .
2
ā3 −4 3 4
As can be immediately verified,
ĀA = AĀ = 71 = (det A)1.
Hence, the inverse matrix of A is given by
A−1 = Ā/(det A),
and the determinant can be implicitly written as
a1 ∧ a2 ∧ a3 = (det A)e1 ∧ e2 ∧ e3 .
2.8 The Regressive Product

A close look at the Ausdehnungslehre of 1844 (Grassmann, 1844) shows that Grass-
mann actually used the Hodge star operator (which he called the Ergänzung) to de-
fine the regressive product. He also used it in the revised Ausdehnungslehre, which
was published in 1862 (Grassmann, 1862), and indeed his Ergänzung depended on the
determinant of an otherwise arbitrary metric (Grassmann, 1894; Fearnley-Sandre and
Stokes,
V 1997). Here, by using ⋆ and ⋆, we can define a new product among elements
of (V ) from the product ∧. Let us define the so-called regressive product, which we
shall denote by ∨, as
A[p] ∨ B[q] = ⋆ −1 (⋆A[p] ∧ ⋆B[q] ) . (2.90)
Vn−p Vn−q
The combination of these operations shows that ⋆A[p] ∈ (V ) and ⋆B[q] ∈ (V ).
V2n−p−q −1
Hence, the exterior product in eqn (2.90) belongs to (V ). V
Since ⋆ differs
from ⋆ only by a sign (see eqn (2.84)) it follows that A[p] ∨ B[q] ∈ p+q−n (V ). This
result leads us to define
A{p} = A[n−p] , (2.91)
and _ ^
p (V )=
). n−p (V (2.92)
W W
Thus,
V the regressive product
V betweenWelements of p (V ) and q (V ) is an element of
(n−p)+(n−q)−n (V ) = n−p−q (V ) = p+q (V ), that is,
_ _ _
∨: (V ) × (V ) → (V ).
p q p+q
The expression involving p-covectors is similar:
Ψ[p] ∨ Φ[q] = ⋆−1 (⋆Ψ[p] ∧ ⋆Φ[q] ) . (2.93)
The regressive product is associative. Indeed,

(A[p] ∨ B[q] ) ∨ C[r] = ⋆ −1 (⋆(A[p] ∨ B[q] ) ∧ ⋆C[r] )

= ⋆ −1 (⋆ ⋆ −1 (⋆A[p] ∧ ⋆B[q] ) ∧ ⋆C[r] ) = ⋆ −1 ((⋆A[p] ∧ ⋆B[q] ) ∧ ⋆C[r] )
= ⋆ −1 (⋆A[p] ∧ (⋆B[q] ∧ ⋆C[r] )) = ⋆ −1 (⋆A[p] ∧ ⋆ ⋆ −1 (⋆B[q] ∧ ⋆C[r] ))
= ⋆ −1 (⋆A[p] ∧ ⋆(B[q] ∨ C[r] ))
= A[p] ∨ (B[q] ∨ C[r] ).
In addition, the regressive product can be related to the contraction as follows:
A[p] ∨ ⋆Ψ[q] = ⋆ −1 (⋆A[p] ∧ ⋆⋆Ψ[q] ) = (−1)q(n−q) ⋆ −1 (⋆A[p] ∧ Ψ[q] )

= (−1)q(n−q) (−1)(n+q−p)(p−q) ⋆(⋆A[p] ∧ Ψ[q] )
^
= (−1)p(n−p) (⋆A[p] ]
∧ Ψ[q] )⌋ΩV = (−1)p(n−p) Ψ̃[q] ⌋(⋆A[p] ⌋ΩV )
= (−1)p(n−p) Ψ̃[q] ⌋(⋆⋆A[p] )

= Ψ̃[q] ⌋A[p] ,
namely,
Ψ̃[q] ⌋A[p] = A[p] ∨ ⋆Ψ[q] . (2.94)
Analogously,
Ã[p] ⌋Ψ[q] = Ψ[q] ∨ ⋆A[p] . (2.95)
2.9 The Grassmann Algebra

V
The exterior
V algebra is defined as being the pair ( (V ), ∧), consisting of the vector
space (V ) endowed with the exterior product ∧. In addition, we can define a sym-
metric bilinear functional g : VV× V → R – or, equivalently, a symmetric correlation
♭ : V → V ∗ . Once the space (V ) is constructed from V , if V is endowed with a
bilinear functional g, it is natural
V to ask whether it would be possible to generalise
the definition of g for the space (V ). The Grassmann algebra, in this context, is the
exterior algebra endowed with the extension of g, which we intend to define in what
follows. The bilinear functional g is often called a metric, so the pair (V, g) is then a
metric space.
Instead of taking the bilinear functional g, we prefer to take into account the
correlation, since the expressions seem straightforward in this case. Let us therefore
∗
consider the correlation ♭ : V → V V and define
Vp its extension, which is denoted by the
3
same symbol, as the mapping ♭ : p (V ) → (V ) given by
(v1 ∧ v2 ∧ · · · ∧ vp )♭ = v1♭ ∧ v2♭ ∧ · · · ∧ vp♭ . (2.96)
By the extension of ♭, we can immediately define the extension of the bilinear

functional g : VV× V → R,
V namely g(v, u) = v♭ (u). Let us denote this extension by G.
We define G : p (V ) × p (V ) → R for the case of simple p-vectors as
3 Evidently, it is a usual abuse of notation. Instead of writing this extension as ♭, we could write
♭p ; however, writing it in this way seems a preciousness that simply overloads the notation.
The Grassmann Algebra 49
G(v1 ∧ · · · ∧ vp , u1 ∧ · · · ∧ up ) = p!(v1 ∧ · · · ∧ vp )♭ (u1 ∧ · · · ∧ up ), (2.97)

V
and generalise this expression for all p (V ) by bilinearity. V The
V presence of the factor
p! is a convention. Previously, the duality mapping (V ∗ ) × (V ) → K was defined
V ∗ mapping∗ T(V ) × T(V ) → K, as well as by the inclusions
∗
by
V means of the duality
(V ) ⊂ T(V ) and (V ) ⊂ T(V ). However, this method generates the inopportune
factor p! when elements of degree p are taken into account. The bilinear functional G
just defined is symmetric:
G(v1 ∧ · · · ∧ vp , u1 ∧ · · · ∧ up ) = G(u1 ∧ · · · ∧ up , v1 ∧ · · · ∧ vp ). (2.98)
Equivalently, we can write
G(v1 ∧ · · · ∧ vp , u1 ∧ · · · ∧ up ) = (vp ∧ · · · ∧ v1 )♭ ⌋(u1 ∧ · · · ∧ up ), (2.99)
or even
G(v1 ∧ · · · ∧ vp , u1 ∧ · · · ∧ up )
g(v1 , u1 ) g(v1 , u2 ) . . . g(v1 , up )
g(v2 , u1 ) g(v2 , u2 ) . . . g(v2 , up ) (2.100)
= .. .. .. .. .
. . . .
g(vp , u1 ) g(vp , u2 ) . . . g(vp , up )
Equation (2.99) is still valid over fields of non-zero characteristic. V

Finally, these definitions can be generalised
V for the vector
V space (V ). In order to
do so, we just need to consider the case of p (V ) and q (V ) with p 6= q, as
G(A[p] , B[q] ) = 0, when p 6= q, (2.101)

V V
where A[p] ∈ p (V ), and B[q] ∈ q (V ).
V
V 2.3 ◮ The exterior algebra ( (V ), ∧) endowed with the extension G of g
Definition
for all (V ) is the Grassmann algebra of the vector space V and is denoted by G(V ).
Example 2.10 Let ♭ be the correlation defined in chapter 1, example 1.3, namely, 1 2
V e1 ♭ = 3e + e ;
e2 ♭ = e1 + 3e2 ; and e3 ♭ = 2e3 . Equation (2.96) provides the extension of ♭ to p (V ) (p = 2, 3). For
V
2 (V ), it yields
(e1 ∧ e2 )♭ = 8e1 ∧ e2 , (e1 ∧ e3 )♭ = 6e1 ∧ e3 + 2e2 ∧ e3 ,

(e2 ∧ e3 )♭ = 2e ∧ e + 6e2 ∧ e3 ,
1 3
V3
and for (V ) it reads
(e1 ∧ e2 ∧ e3 )♭ = 16e1 ∧ e2 ∧ e3 .
For the symmetric bilinear functional G, one finds
G(e1 ∧ e2 , e1 ∧ e2 ) = 8, G(e1 ∧ e2 , e1 ∧ e3 ) = 0, G(e1 ∧ e2 , e2 ∧ e3 ) = 0,
and
G(e1 ∧ e2 ∧ e3 , e1 ∧ e2 ∧ e3 ) = 16.
2.10 The Hodge Isomorphism

V V
Previously, the vector spaces p (V ) and n−p (V ) were shown to have the same di-
mension and therefore to be isomorphic, although this isomorphism is not canonical.
We must therefore define this isomorphism and, among numerous possible choices, the
Hodge isomorphism has prominent importance. The Hodge isomorphism is obviously
closely related to the quasi-Hodge isomorphisms. In order to define the quasi-Hodge
isomorphisms, an orientation must be chosen for the vector space V . In order to define
the Hodge isomorphism, one additional structure is demanded: a symmetric bilinear
functional on V . Consequently, the Hodge isomorphism is defined only in the con-
text of the Grassmann algebra and not in the exterior algebra, as only quasi-Hodge
isomorphisms exist in the latter.
The presence of a symmetric bilinear functional g makes it possible to relate the
spaces V V and V ∗Vvia the correlation ♭ : V → V ∗ . This correlation can be generalised
p
via ♭ : p (V ) → (V ) from eqn (2.96). The same reasoning applies to the correlation
−1
Vp
♯V= ♭ , which can be similarly generalised (see eqn (2.96)), so that ♯ : (V ) →
p (V ).
V Vn−p
Now let us consider the quasi-Hodge isomorphism ⋆ : p (V ) → (V ). Since
Vn−p V
♯: (V ) → n−p (V ), the composition ♯ ◦ ⋆ satisfies
^ ^
♯◦⋆: p (V )→ n−p (V ). (2.102)
Furthermore, let us consider the composition ⋆ ◦ ♯ such that

^ ^
p n−p
⋆◦♯: (V ) → (V ). (2.103)
Vp V
With respect to the quasi-Hodge isomorphism ⋆ : (V ) → n−p (V ), we can
consider two compositions: (i) ♭ ◦ ⋆, and (ii) ⋆ ◦ ♭. The first composition is provided by
^ ^
p n−p
♭◦⋆: (V ) → (V ), (2.104)
whereas the second one reads

^ ^
⋆◦♭: p (V )→ n−p (V ). (2.105)
V V
Let us now define the isomorphism p (V ) → n−p (V ). From what we have previ-
ously discussed, there exist two isomorphisms which can be used to define it: (i) ♯ ◦ ⋆,
and (ii) ⋆ ◦ ♭. Which one to choose? In order to settle this question, we demanded that
♯ ◦ ⋆ = ⋆ ◦ ♭. (2.106)
However, which conditions must hold in order for this expression to be valid?
Let us first consider the isomorphism ⋆ defined by eqn (2.66). By applying the
correlation ♯ at both sides of this equation and using ♯(Ψ ∧ Φ) = Ψ♯ ∧ Φ♯ , we can see
that
Ψ[p]♯ ∧ (♯ ◦ ⋆)A[p] = p!Ψ[p] (A[p] )Ω♯V ∗ , (2.107)
The Hodge Isomorphism 51
which can be written as
B[p] ∧ (♯ ◦ ⋆)A[p] = p!A[p]♭ (B[p] )Ω♯V ∗ . (2.108)
On the other hand, using the definition of ⋆ given in eqn (2.67), we find that
A[p] ∧ (⋆ ◦ ♭)B[p] = p!B[p]♭ (A[p] )ΩV . (2.109)
Comparing those two last equations, if the equality ♯ ◦ ⋆ = ⋆ ◦ ♭ holds, then Ω♯V ∗ = ΩV
or equivalently
ΩV ∗ = Ω♭V . (2.110)
The relationship between ΩV ∗ and ΩV is given by eqn (2.65). With this last con-
dition, it therefore follows that
Ωf♭ (Ω ) = 1. (2.111)
V V
In order to understand the meaning of this equation, the definition of G is used together
with eqn (2.99), implying that
G(ΩV , ΩV ) = 1. (2.112)
Then ♯ ◦ ⋆ = ⋆ ◦ ♭ holds if the n-vector ΩV is unitary. The Hodge isomorphism, denoted

by ⋆, is defined by
⋆=♯◦⋆=⋆◦♭ . (2.113)
Using eqn (2.97), we can thus define ⋆ as
A[p] ∧ ⋆B[p] = G(A[p] , B[p] )ΩV , (2.114)
where G(ΩV , ΩV ) = 1; or, even better than that, as
A ∧ ⋆B = G(A, B)ΩV , (2.115)

V
where A, B ∈ (V ). Here we used the fact that G(A[p] , B[q] ) = 0 if p 6= q. The Hodge
Vp Vn−p
isomorphism between the spaces (V ) and (V ) is similarly defined. In fact, the
∗ Vp Vn−p
mapping ⋆ : (V ) → (V ) can be written as
∗
⋆ = ♭ ◦ ⋆ ◦ ♯. (2.116)
Evidently, the calculations for ⋆ and ⋆ can be applied to ⋆ by careful considering

the correlations ♭ and ♯, in the context of either eqns (2.68) and (2.80), or eqns (2.81)
and (2.82). For the case of the Hodge isomorphism ⋆, it is immediately clear that
⋆1 = ΩV ,
, (2.117)
⋆A = Ae♭ ⌋ΩV
V
where A ∈ (V ).
The regressive product can be written using the Hodge isomorphism. From eqn
(2.113), it reads
⋆ = ⋆ ◦ ♯, ⋆ = ♭ ◦ ⋆. (2.118)
Taking it and ⋆ −1 = ⋆ ◦ ♭−1 = ⋆ ◦ ♯ into account, from eqn (2.90) we find that
A[p] ∨ B[q] = ⋆−1 (⋆A[p] ∧ ⋆B[q] ) . (2.119)
To conclude this discussion, let us explicitly write the unit pseudoscalar ΩV . Let
B = {e1 , . . . , en } be a basis of V and let g be the symmetric bilinear functional
g(v, u) = v♭ (u) = gij v i uj , where gij = g(ei , ej ) = gji . Let B′ = {e′1 , . . . , e′n } be an
orthonormal basis,4
g(e′i , e′j ) = λi δij , (2.120)
where (
1, (i = 1, . . . , p),
λi = (2.121)
−1, (i = p + 1, . . . , n),
defining the quadratic space Rp,q (p + q = n).
Since the vectors {e′i } are unitary, from the definition of G, it immediately follows
that the n-vector
ΩV = e′1 ∧ · · · ∧ e′n (2.122)
is unitary. We can then use this expression for ΩV with respect to the basis B′ in
order to explicitly calculate the Hodge isomorphism. Sometimes it is appropriate not
to take into account an orthonormal basis B′ but rather an arbitrary basis B. Let us
then express ΩV with respect to the basis B, writing the relation between B and B′
as
e′i = h(ei ) = hji ej , (2.123)
where the demonstration of
e′1 ∧ · · · ∧ e′n = (det h)e1 ∧ · · · ∧ en (2.124)
is left as an exercise, and h denotes the matrix
 1 1 
h1 h2 · · · h1n
 h21 h22 · · · h2n 
 
h= . . . . . (2.125)
 .. .. . . .. 
hn1 hn2 · · · hnn
On the other hand, we can write from eqn (2.120) that
λi δij = hki gkl hli . (2.126)
This equation can be written in matrix form as
λ = h⊺ gh, (2.127)
4 Equation (2.120) does not denote the sum convention, since the index i, although repeated, is
not placed as a raised or lowered index, which would be the case if it did.
The Hodge Isomorphism 53
where g denotes the matrix

 
g11 g12 · · · g1n
 g21 g22 · · · g2n 
 
g= . .. . . ..  , (2.128)
 .. . . . 
gn1 gn2 · · · gnn
h⊺ denotes the transposed matrix associated with h, and λ is the diagonal matrix
with entries λii = λi given by eqn (2.121). Using the well-known properties det(AB) =
det A det B, and det A⊺ = det A, as well as the fact that det λ = (−1)q , where q = n−p,
it follows that (−1)q = (det h)2 det g or
1 = (det h)2 | det g|. (2.129)
There are now two possibilities: (i) det h = +| det g|−1/2 , or (ii) det h = −| det g|−1/2 .
Since the set of matrices h must include the identity matrix 1, the first possibility
must be chosen:
det h = | det g|−1/2 . (2.130)
Thus, the unit n-vector ΩV can be written in terms of the basis B as
1
ΩV = p e1 ∧ · · · ∧ en . (2.131)
| det g|
Another useful expression is

p
ΩV = | det g −1 |e1 ∧ · · · ∧ en , (2.132)
where g −1 is the inverse matrix related to g,

 11 12 
g g ··· g 1n
 g 21 g 22 ··· g 2n 
 
g −1 =  . . .. ..  , (2.133)
 .. .. . . 
g n1 g n2 · · · g nn
where g ij = g(ei , ej ), g ij gjk = δki .
Example 2.11 Let us establish the Hodge isomorphism for the case where the space R3 is endowed with
the usual Euclidean scalar product. The standard basis {ei } satisfies g(ei , ej ) = δij . The unit 3-vector
is given by ΩR3 , corresponding to the usual orientation of R3 given by ΩR3 = e1 ∧ e2 ∧ e3 . Using eqn
(2.117), we find that
⋆ 1 = e1 ∧ e2 ∧ e3 , ⋆e1 ∧ e2 ∧ e3 = 1,
⋆ e1 = e2 ∧ e3 , ⋆e2 ∧ e3 = e1 ,
2 3 1
⋆e =e ∧e , ⋆e3 ∧ e1 = e2 ,
⋆ e3 = e1 ∧ e2 , ⋆e1 ∧ e2 = e3 .
Example 2.12 This example aims to establish the Hodge isomorphism for the space R3 endowed with the
correlation investigated in example 1.3. The results in example 2.9 show that the unit 3-vector ΩR3 with
the usual orientation of R3 reads
1
ΩR3 = e1 ∧ e2 ∧ e3 .
4 V
Using the definition of ♭ for this case and its extension for p (V ) calculated in the example 2.8 in eqn
(2.117) , we find that
1 3 1
⋆1= e1 ∧ e2 ∧ e3 , ⋆e1 = e2 ∧ e3 + e3 ∧ e1 ,
4 4 4
1 3 1
⋆ e2 = e2 ∧ e3 + e3 ∧ e1 , ⋆e3 = e1 ∧ e2 ,
4 4 2
3 1
⋆ e1 ∧ e2 = 2e3 , ⋆e3 ∧ e1 = e2 − e1 ,
2 2
3 1
⋆ e2 ∧ e3 = e1 − e2 , ⋆e1 ∧ e2 ∧ e3 = 4.
2 2
2.11 Additional Readings

A thorough discussion of the exterior algebra can be found in Greub’s (1978) book,
whereas some results obtained by using the exterior product in linear algebra are
revisited in Winitzki’s (2010) book. The exterior algebra is the cornerstone of the cal-
culus of differential forms. A good introduction to the formalism of differential forms
is provided by do Carmo (1994). For the use of such forms in differential geometry via
Cartan’s moving frame method, a great introductory text has been provided by O’Neill
(2006) whereas, for a modern treatment, including vector bundles, a nice introduction
has been given by Darling (1994). For applications of differential forms in physics,
the classic work is Flanders (1963), whereas Frankel (2012) is modern and complete.
Another text with an emphasis on applications in thermodynamics, electromagnetism,
and gauge theories is the one by Edelen (1985). The exterior calculus without a met-
ric g is fundamental for a metric-free formulation of classical electrodynamics (Hehl
and Obukhov, 2003), with subsequent applications (da Rocha and Rodrigues, 2010).
The Grassmann algebra also underlies supermanifolds and supersymmetry; for an in-
troduction to this subject, from the geometric point of view, we suggest the work
by Rodrigues Jr, da Rocha, Bernardini and Vaz Jr (2005) as well as that by Rogers
(2007).
2.12 Exercises
(1) Let B = {e1 , e2 , e3 , e4 , e5 } be a basis of V ∗ and let α, β, and γ the following
multicovectors:
α = e1 + 2e1 ∧ e2 + 5e3 ∧ e4 ,
β = e3 − e1 ∧ e2 + 3e1 ∧ e3 ∧ e4 − e3 ∧ e4 ∧ e5 ,
γ = e5 − 2e4 ∧ e5 .
Calculate the following: (a) α ∧ α; (b) β ∧ β; (c) γ ∧ γ; (d) α ∧ β; (e) β ∧ α; (f) α ∧ γ;
(g) γ ∧ α; (h) β ∧ γ; (i) γ ∧ β; (j) α ∧ α ∧ α; (k)α ∧ α ∧ β; (l) α ∧ α ∧ γ; (m) β ∧ β ∧ α;
(n) β ∧ β ∧ β; and (o) β ∧ β ∧ γ.
Exercises 55
V2
(2) Show that an arbitrary 2-vector F ∈ (V ) can be written as the exterior product
V1
of two 1-vectors v, u ∈ V = (V ), that is, F = v ∧ u, only if dim V ≤ 3.
(3) Use the method described in example 2.8 to find the inverse of the matrix
 
1 2 1 −1
0 3 −1 1 
 
2 1 2 0  .
1 1 0 −1
(4) Let B = {ei }, and B′ = {e′i }, be two bases of V related by ei = Bij ej and
let B = {Bji } be the matrix that changes the two bases, where Bji corresponds to
the element of the j th row and the ith column (i, j = 1, . . . , n). Such change of basis
obviously induces a change of basis for the space of the k-vectors. (a) Show that
X
e′i ∧ e′j = (det ∆kl
ij )ek ∧ el ,
k<l
where ∆klij denotes the matrix of order 2 obtained from the matrix B in the following
way: the first row of ∆kl ij is given by the i
th
row of B, and the second row of ∆kl
ij by
the j row of B; and the first column of ∆ij is given by the k th column of B, and the
th kl
P
second column of ∆kl ij by the l
th
column of B. In this expression, j<l denotes the
sum over all j and l such that k < l. (b) Generalise the previous result, showing that,
in the space of k-vectors (k < n), the property
X
e′µ1 ∧ · · · ∧ e′µk = (det ∆νµ11···ν
···µk )eν1 ∧ · · · ∧ eνk
k
ν1 <···<νk
holds, where ∆νµ11···ν

···µk denotes the matrix of order k obtained from the matrix B when
k
we take the i row of ∆νµ11···ν

th th
···µk as the µi row of B and the j
k th
column of ∆νµ11···ν
···µk
k
th
as being the νj column of B. (c) Show that, as a consequence of these results, for
pseudoscalars, the expression
e′1 ∧ · · · ∧ e′n = (det B)e1 ∧ · · · ∧ en
holds.
(5) From this last expression (which can be taken as the definition of the determinant
of a linear transformation), deduce the Laplace rule for the calculation of determinants.
(6) Let V be a vector space (dim V = n) and let W be the k-dimensional subspace
generated by {v1 , . . . , vk }. The k-vector IW = v1 ∧ · · · ∧ vk completely defines this
subspace, and a vector v is in W if and only if v∧IW = 0 (see example 2.4). If B = {ei }
(i = 1, . . . , n) is a basis of V , the components
V of the k-vector IW with respect to the
basis {eµ1 ∧ · · · ∧ eµk | µ1 < · · · < µk } of k (V ) are called the Plücker coordinates of
the subspace W in the basis B, that is, the Plücker coordinates V µ1 ···µk are given by
IW = v1 ∧ · · · ∧ vk = V µ1 ···µk eµ1 ∧ · · · ∧ eµk .

(a) Show that, in general, the Plücker coordinates are not all independent but satisfy
a set of identities called the Plücker correlations and given by
V [µ1 ···µk V ν1 ]ν2 ···νk = 0,
where the brackets denote antisymmetrisation of the indices, that is,
V [µ1 ···µk V ν1 ]ν2 ···νk

= V µ1 ···µk V ν1 ν2 ···νk − V µ1 ···µk−1 ν1 V µk ν2 ···νk
+ V µ1 ···µk−2 ν1 V µk−1 ν2 ···νk − V µ1 ···µk−3 µk−1 ν1 V µk−2 ν2 ···νk + · · ·
+ (−1)k−1 V µ1 µ3 ···µk−1 ν1 V µ2 ν2 ···νk + (−1)k V µ2 µ3 ···µk−1 ν1 V µ1 ν2 ···νk .
(b) Not all Plücker correlations are trivial. Some of them, for instance, follow from the
anti-commutativity of the components V µ1 ···µk = V [µ1 ···µk ] of the k-vector IW . Show
that, for k = n, and k = n − 1, not all Plücker correlations are trivial. (c) Consider the
case where n = 4, and k = 2. Show that, for this case, there exists only one non-trivial
Plücker correlation.
^ = A⌊α, e V
(7) Show that (α⌋A) for all α ∈ V ∗ , and A ∈ (V ). This result implies that
^ = A⌊
(Ψ⌋A) e Ψ̃, for all Ψ ∈ V(V ∗ ), and A ∈ V(V ).
V Vq
(8) Consider T[p] ∈ p (V ), and S [q] ∈ (V ). Show that

T[p] ⌊S [q] = (−1)q(p+1) S [q] ⌋T[p] .
˙ denote the usual

(9) Let R3 be endowed with the usual Euclidean scalar product. Let ×
vector product between vectors (which defines the Gibbs–Heaviside vector algebra) and
the vector product × defined as
v × u = ⋆(v ∧ u),
where v, u ∈ R3 . (a) Show that the objects defined by v×u and v×u ˙ present the same
˙
components. (b) Show that the object defined by v×u corresponds to what sometimes
is called the axial vector or pseudovector (i. e. it is an object that does not change sign
under the coordinate system inversion), and that, in contrast, the object defined by
v × u is indeed a vector (sometimes called polar vector, in order to be distinguished
from the axial vector, since in this case there is a change sign under the coordinate
system inversion).
(10) Consider {u1 , u2 , . . . , u2r−1 , u2r } vectors in V . Given v = u1 ∧ u2 + u3 ∧ u4 +
u5 ∧ u6 + · · · + u2r−1 ∧ u2r , show that
| ∧v∧
v {z· · · ∧ v} = r! (u1 ∧ u2 ∧ · · · ∧ u2r−1 ∧ u2r ).
r times
3
Clifford, or Geometric, Algebra
In this chapter, we introduce the so-called Clifford, or geometric, algebras. We are here
interested in the universal Clifford algebras and, in this text, except for the first section
in this chapter, ‘Clifford algebra’ will be a synonym for ‘universal Clifford algebra’.
Besides the standard definition of a Clifford algebra via the so-called Clifford mapping,
we present an explicit construction of the (universal) Clifford algebra associated with
a quadratic space as a quotient of the tensor algebra. Moreover, Clifford algebras are
also presented in the context of Grassmann algebras (Clifford, 1878). The prominent
features of Clifford algebras are presented and creation operators and annihilation
operators are introduced. For a discussion regarding Clifford algebras over infinite-
dimensional spaces, see the book by Plymen and Robinson (1990).
3.1 Definition of a Clifford Algebra

Let V be a vector space over R, endowed with a symmetric bilinear form g. Let A be
an associative algebra with unity 1A and let γ be the linear mapping γ : V → A.
Definition 3.1 ◮ The pair (A, γ) is a Clifford algebra for the quadratic space (V, g)
when A is generated as an algebra by {γ(v) | v ∈ V } and {a1A | a ∈ R}, and γ
satisfies
γ(v)γ(u) + γ(u)γ(v) = 2g(v, u)1A (3.1)
for all v, u ∈ V .
Equation (3.1) holds for all u, v ∈ V if and only if γ(v)2 = Q(v) holds for all v ∈ V .
In many cases, it is easier to verify the equality γ(v)2 = Q(v) than the equality in
eqn (3.1).
When g is not degenerated, the Clifford algebra for the quadratic space (V ∗ , g −1 )
is defined in a completely analogous way, where now the linear mapping γ : V ∗ → A
is defined, satisfying
γ(α)γ(β) + γ(β)γ(α) = 2g −1 (α, β)1A , (3.2)
for all α, β ∈ V ∗ , and g −1 : V ∗ × V ∗ → R.

The mapping γ is a kind of ‘square root’ of the quadratic form Q(v) = g(v, v),
since
(γ(v))2 = Q(v) = g(v, v) . (3.3)
Such a mapping γ is said to be a Clifford mapping.
58 Clifford, or Geometric, Algebra
Now let us consider an orthonormal basis B = {e1 , . . . , en } of V . In the Clifford

algebra (A, γ) for (V, g), we have
γ(ei )γ(ej ) + γ(ej )γ(ei ) = 0A , i 6= j, (3.4)
and
γ(ei )2 = Q(ei )1A , (3.5)
where Q(ei ) = g(ei , ei ). By using such relations, any product involving γ(ei ) (i =
1, . . . , n), and their powers as well, can be reordered to yield
γ(e1 )µ1 γ(e2 )µ2 · · · γ(en )µn , µi = 0, 1, (i = 1, . . . , n),
where γ(e1 )0 γ(e2 )0 · · · γ(en )0 is the identity 1A of A. As A is generated as an algebra

by {γ(v) | v ∈ V } and {a1A | a ∈ R}, it is generated by these products:
A = span{γ(e1 )µ1 γ(e2 )µ2 · · · γ(en )µn | µi = 0, 1}. (3.6)
However, since the number of elements of type γ(e1 )µ1 γ(e2 )µ2 · · · γ(en )µn , where
µi = 0, 1 is 2n , therefore dim A ≤ 2n and so the maximal dimension of a Clifford
algebra is 2n . Although there exist examples of Clifford algebras with fewer than 2n -
dimensions(Porteous, 1995), Clifford algebras of maximal dimension will be the focus
here. Indeed, such algebras have a prominent property that distinguishes them from
algebras with fewer than 2n dimensions. This property is called universality, and the
Clifford algebras that have such property have dimension 2n .
Definition 3.2 ◮ A Clifford algebra (A, γ) for the quadratic space (V, g) is said to be
a universal Clifford algebra if, for each Clifford algebra (B, ρ) for (V, g), there exists
an isomorphism φ : A → B such that ρ = φ ◦ γ, and φ(1A ) = 1B . A universal Clifford
algebra for the quadratic space (V, g) is denoted by Cℓ(V, g).
The universal Clifford algebra Cℓ(V, g), if it exists, is unique up to a unique iso-
morphism. In fact, since Cℓ(V, g) is a Clifford algebra, there is a unique isomorphism
such that φ : A → B such that ρ = φ ◦ γ, and φ(1A ) = 1B whereas, since (B, ρ) is
also a Clifford algebra, there is a unique isomorphism such that φ′ : B → A such that
γ = φ′ ◦ ρ, and φ′ (1B ) = 1A . Now, the composition φ′ ◦ φ : Cℓ(V, g) → Cℓ(V, g) is such
that γ = (φ′ ◦ φ) ◦ γ and the identity 1A : Cℓ(V, g) → Cℓ(V, g), whence φ′ ◦ φ = 1A . A
similar argument shows that φ ◦ φ′ = 1B , whence φ : A → B is an isomorphism.
Theorem 3.1 ◮ The Clifford algebra (A, γ) for the quadratic space (V, g) is universal
when dim A = 2n , where n = dim V .
Proof: Consider an orthonormal basis B = {e1 , . . . , en } of V . For the Clifford algebra
(A, γ), it follows that γ(ei )γ(ej ) + γ(ej )γ(ei ) = 0A , (i 6= j) and γ(ei )2 = Q(ei )1A . It
is supposed that dim A = 2n . In this case, the set {γ(e1 )µ1 γ(e2 )µ2 · · · γ(en )µn } with
µi = 0, 1 for i = 1, . . . , n does not only generate A but is in addition a basis for A. Let
now (B, ρ) be an arbitrary Clifford algebra. Then ρ(ei )ρ(ej ) + ρ(ej )ρ(ei ) = 0B (where
Definition of a Clifford Algebra 59
i 6= j), and ρ(ei )2 = Q(ei )1B , and the set {ρ(e1 )µ1 ρ(e2 )µ2 · · · ρ(en )µn | µi = 0, 1}
generates B. Define the mapping φ as a linear mapping φ : A → B such that
φ(γ(e1 )µ1 γ(e2 )µ2 · · · γ(en )µn ) = ρ(e1 )µ1 ρ(e2 )µ2 · · · ρ(en )µn .
It is straightforward to see that φ as defined is an algebra isomorphism, that is,

φ(aa′ ) = φ(a)φ(a′ ), ∀ a, a′ ∈ A, satisfying φ(ei )φ(ej ) + φ(ej )φ(ei ) = 2g(ei , ej )1B .
From the definition of a Clifford algebra, (A, γ) is therefore a universal Clifford alge-
bra Cℓ(V, g). ✓
Theorem 3.2 ◮ For all quadratic space (V, g), there exist a universal Clifford algebra,
and every universal Clifford algebra has dimension 2n .
This theorem is demonstrated from the universal Clifford algebra associated with a
quadratic space. Its explicit construction is accomplished by a quotient of the tensor
algebra by a specific ideal. However, this discussion is reserved for section 3.2, as a
point that deserves special attention. Since the universal Clifford algebra has been
shown to be unique up to a uniquely given isomorphism, it is legitimate to say the
universal Clifford algebra associated with a quadratic space.
Example 3.1 Let V be a 1-dimensional vector space. In this case, any vector v ∈ V can be written as
v = ye, where {e} is a basis of V . Let now g be the symmetric bilinear functional such that g(e, e) = −1.
Then, g(v, v) = −y2 . Consider now the subalgebra of the matrix algebra defined by

x y
A= | x, y ∈ R .
−y x
Obviously, this algebra is generated by

0 1 10
y |y∈R and x = x1A | x ∈ R .
−1 0 01
Now let us define the mapping γ : V → A as

0 1
γ(e) = .
−1 0
Consequently,
0 y
γ(v) = .
−y 0
The mapping γ is a Clifford mapping, since

0 y 0 y 10
[γ(v)]2 = = −y 2 = g(v, v)1A .
−y 0 −y 0 01
The algebra A is isomorphic to the algebra of complex numbers algebra C, and C is an example of a
Clifford algebra.
Example 3.2 Let V = R3 be endowed with the usual scalar product g(v, u) = v1 u1 + v2 u2 + v3 u3 , where
v = (v1 , v2 , v3 )⊺ , and u = (u1 , u2 , u3 )⊺ . Consider now the algebra of the 2 × 2 complex matrices:

z1 z2
A= | zi ∈ C(i = 1, 2, 3, 4) ,
z3 z4
with the matrix product. This algebra is generated by {x1 σ1 , x2 σ2 , x3 σ3 | xi ∈ R} and {x0 1 | x0 ∈ R},
where 1 denotes the identity matrix, and σi denotes the matrices

01 0 −i 1 0
σ1 = , σ2 = , σ3 = .
10 i 0 0 −1
These matrices are called the Pauli matrices. In order to realise that this algebra is indeed generated by
these matrices, it is enough to use the fact that
σ1 σ2 = iσ3 , σ2 σ3 = iσ1 , σ3 σ1 = iσ2 ,
which can be straightforwardly verified. Let us now define the mapping γ : R3 → A:
γ(1, 0, 0) = σ1 , γ(0, 1, 0) = σ2 , γ(0, 0, 1) = σ3 .
The mapping γ is a Clifford mapping, and the algebra A is a Clifford algebra. Indeed,

v3 v1 − iv2
γ(v) = ,
v1 + iv2 −v3
which implies that

10
γ(v)γ(u) + γ(u)γ(v) = 2(v1 u1 + v2 u2 + v3 u3 ) = 2g(v, u)1A .
01
3.2 Universal Clifford Algebra as a Quotient of the Tensor Algebra

This construction was proposed by Chevalley in 1954 (Chevalley, 1954) as well as
by Bourbaki in 1959 (Bourbaki, 1989); In it, the Clifford algebra is presented as a
quotient of the tensor algebra by a two-sided ideal. This approach provides a proof of
existence by construction, and it is very suitable to a fast access to the main properties
of Clifford algebras over commutative rings (Helmstetter and Micali, 2008).
Let (V, g) be a quadratic space and let T(V ) be the algebra of the contravariant
tensors. Let us consider the ideal IC of T(V ) generated by elements of type
v ⊗ v − Q(v)1,
where Q(v) = g(v, v), and 1 is the identity of the tensor algebra. The ideal IC consists
therefore of all sums X
Ai ⊗ (v ⊗ v − Q(v)1) ⊗ Bi ,
i
where Ai , Bi ∈ T(V ). We can also realise that the ideal IC is generated by the elements
v ⊗ u + u ⊗ v − 2g(v, u)1.
In order to construct the quotient algebra T(V )/IC , we must consider the equiva-
lence relation
A ∼ B ⇔ A = B + x, x ∈ IC .
Let us denote the product of equivalence classes by ⋄:
[A] ⋄ [B] = [A ⊗ B].
Let v, u ∈ V and consider the tensor product v ⊗ u suitably expressed by

1 1
v ⊗ u = (v ⊗ u − u ⊗ v) + g(v, u) + (v + u) ⊗ (v + u)
2 2
− g(v + u, v + u) − v ⊗ v + g(v, v) − u ⊗ u + g(u, u) .
Universal Clifford Algebra as a Quotient of the Tensor Algebra 61
The term in the square brackets is an element of the ideal IC , so

1
v⊗u∼ (v ⊗ u − u ⊗ v) + g(v, u), (3.7)
2
or, equivalently,
v ⊗ u ∼ v ∧ u + g(v, u). (3.8)
Hence, by denoting the quotient mapping π : T → T(V )/IC ≃ Cℓ(V, g), the product
of vectors in the quotient algebra T(V )/IC can be written as
v ⋄ u = π(v ∧ u) + g(v, u) . (3.9)
In
V eqn (3.8), v∧u is the element mentioned in eqn (3.7), since we have already identified
(V ) with a subspace of T(V ). Although in eqn (3.9) we have come down into the
quotient
V T(V )/IC , the element π(v∧u) can be interpreted as v∧u, when every element
of (V ) is identified with its image by V π. Nevertheless, this interpretation
V is allowed
only if the mapping π is injective on (V ), and we must prove that IC ∩ (V ) = {0}.
In other words, if A is an P element of IC and is other than the zero multivector, then
n
Alt(A) 6= A. In fact, Alt[ i=1 Ai ⊗ v ⊗ v ⊗ Bi ] = 0 since, for each permutation σ,
there exists a permutation σ ′ with opposite sign (that exchanges v in σ(j) and v in
σ(j + 1)), so the two cancel each other out. Thus,
" n #
X
Alt Ai ⊗ (v ⊗ v − Q(v)1) ⊗ Bi
i=1
" n
# " n #
X X
= Alt Ai ⊗ v ⊗ v ⊗ Bi − Alt Ai ⊗ Q(v)1 ⊗ Bi
i=1 i=1
" n #
X
= −Q(v)Alt Ai ⊗ 1 ⊗ B i = 0 if and only if Q(v) = 0 .
i=1
Therefore, when Q(v) 6= 0, then Alt(A) 6= A.

V
Keeping in mind that every element of (V ) is identified with its image by π, we
can thus simplify the notation in what follows, writing in particular eqn (3.9) as
v ⋄ u = v ∧ u + g(v, u). (3.10)

V
Moreover, (V ) is identified V with Cℓ(V, g). In fact, the quotient mapping π : T(V ) →
Cℓ(V, g) induces a mapping (V ) → Cℓ(V, g) which, by a reasoning similar to that
presented after eqn (3.9), can be shown to be a bijection.
The next step is to generalise these expressions, considering v, u, w ∈ V . The
exterior product v ∧ u ∧ w can be written as (see chapter 2, examples 2.1 and 2.2)
1
v∧u∧w = v ⊗ (u ∧ w) − u ⊗ (v ∧ w) + w ⊗ (v ∧ u) . (3.11)
3
Now eqn (3.8) implies that the two last terms in eqn (3.11) respectively read
u ⊗ (v ∧ w) ∼ u ⊗ v ⊗ w − g(v, w)u,
(3.12)
w ⊗ (v ∧ u) ∼ w ⊗ v ⊗ u − g(v, u)w.
By using the equivalence classes
u ⊗ v + v ⊗ u ∼ 2g(u, v),
w ⊗ v + w ⊗ u ∼ 2g(w, v),
eqn (3.12) yields

u ⊗ (v ∧ w) ∼ −v ⊗ u ⊗ w + 2g(u, v)w − g(v, w)u,
(3.13)
w ⊗ (v ∧ u) ∼ −v ⊗ w ⊗ u + 2g(w, v)u − g(v, u)w.
Now, substituting eqn (3.13) into eqn (3.11), we obtain
v ∧ u ∧ w = v ⊗ (u ∧ w) + g(v, w)u − g(v, u)w. (3.14)
It is worth emphasising that, by using the expression for v♭ ⌋(u ∧ w) (see e.g. eqn
(2.37)) we can finally express
v ⊗ (u ∧ w) ∼ v ∧ u ∧ w + v♭ ⌋(u ∧ w), (3.15)
namely, the product between a 1-vector and a 2-vector in the quotient algebra T(V )/IC
can be written as
v ⋄ (u ∧ w) = v ∧ u ∧ w + v♭ ⌋(u ∧ w), (3.16)
which is a clear generalisation of eqn (3.9).
Equation (3.16) can be generalised as
v ⋄ A[p] = v ∧ A[p] + v♭ ⌋A[p] . (3.17)
This formula can be demonstrated by induction, and the cases for p = 1 and for p = 2
have already been shown.
The quotient algebra T(V )/IC is a Clifford algebra. Indeed, from eqn (3.9), it
follows that
v ⋄ u + u ⋄ v = 2g(v, u). (3.18)
Moreover, Clifford algebra is universal, as will be shown in ‘Universality’. Hence,
Cℓ(V, g) = T(V )/IC . (3.19)
Example 3.3 Let us prove eqn (3.17) by induction. Because of linearity, it is sufficient to prove this
equation for the case where A[p] is a simple p-vector. First, we slightly change the notation, so that it
shall be suitable for this purpose. Let us denote a simple p-vector by A[1···p] , that is,
A[1···p] = v1 ∧ · · · ∧ vp .
In addition, we introduce the notation
A[1···ı̂···p] = v1 ∧ · · · ∧ vi−1 ∧ vi+1 ∧ · · · ∧ vp
to denote the (p − 1)-vector obtained when the vector vi is taken out of the product. Thus, the ‘hat’ here
does not denote the grade involution but rather the omission of the corresponding vector in the wedge
product.
Using this notation and the definition of the wedge product, we can write
" p
#
1 X
i
u ∧ A[1···p] = u ⊗ A[1···p] + (−1) vi ⊗ u ∧ A[1···ı̂···p] .
p+1 i=1
Assuming that the inductive hypothesis holds for (p − 1)-vector, this equation implies that
u ∧ A[1···ı̂···p] ∼ u ⊗ A[1···ı̂···p] − u♭ ⌋A[1···ı̂···p] .
Hence, we have

vi ⊗ u ∧ A[1···ı̂···p] ∼ vi ⊗ u ⊗ A[1···ı̂···p] − vi ⊗ u♭ ⌋A[1···ı̂···p]

∼ 2g(vi , u)A[1···ı̂···p] − u ⊗ vi ⊗ A[1···ı̂···p] − vi ⊗ u♭ ⌋A[1···ı̂···p] .
Now, keeping in mind the definition of the wedge product, we obtain
p
X
(−1)i+1 vi ⊗ A[1···ı̂···p] = pA[1···p] ,
i=1
which yields

1
u ∧ A[1···p] ∼ u ⊗ A[1···p] + u ⊗ pA[1···p]
p+1
p
X p
X

− (−1)i+1 2g(vi , u)A[1···ı̂···p] − (−1)i vi ⊗ u♭ ⌋A[1···ı̂···p] .
i=1 i=1
However, we can also write

p
X
(−1)i vi ⊗ u♭ ⌋A[1···ı̂···p]
i=1
p
X X
i−1 p
X
= (−1)i vi ⊗ (−1)j+1 g(u, vj )A[1···̂···ı̂···p] + (−1)j g(u, vj )A[1···ı̂···̂···p]
i=1 j=1 j=i+1
p
X p
X
= (−1)i (−1)j+1 g(u, vj )vi ⊗ A[1···̂···ı̂···p]
j=1 i=j+1
p j−1
X X
+ (−1)i (−1)j g(u, vj )vi ⊗ A[1···ı̂···̂···p]
j=1 i=1
p
X j−1
X p
X
= (−1)j+1 g(u, vj ) (−1)i+1 vi ⊗ A[1···ı̂···̂···p] + (−1)i vi ⊗ A[1···̂···ı̂···p]
j=1 i=1 i=j+1
p
X
= (−1)j+1 g(u, vj )(p − 1)A[1···̂···p] ,
j=1
where in the last equality we have used again the definition of the exterior product. Finally, we obtain
X p
1
u ∧ A[1···p] ∼ u ⊗ A[1···p] − 2 (−1)i+1 g(vi , u)A[1···ı̂···p]
p + 1 i=1
Xp
+ (p − 1) (−1)j+1 g(vj , u)A[1···̂···p]
j=1
∼ u ⊗ A[1···p] − u♭ ⌋A[1···p] ,
which proves eqn (3.17).

Observation ☞ The Clifford mapping γ is obviously γ = π ◦ ı, where ı denotes the

inclusion ı : V ֒→ T(V ), and π : T(V ) → Cℓ(V, g) = T(V )/IC , in such a way that
γ(v) = [v]. In these expressions, we did not explicitly write [v] or γ(v), since doing so
would have overloaded the notation. Although there are situations where the notation
γ(v) is useful and where using it can help detailed understanding, in this text, except
when it is convenient, we avoid explicitly writing the Clifford mapping γ.
Observation ☞ Notation: At first glance, using the notation ⋄ for the product in the
Clifford algebra Cℓ(V, g) seems convenient. However, the side effect of using ⋄ is that
it would have to be written on every single page in this text, to denote the Clifford
product. Hence, in order to employ a straightforward notation, we shall abandon the
use of ⋄ and simply denote the product in Cℓ(V, g) by juxtaposition. Equations (3.17)
and (3.18) then read
vu + uv = 2g(v, u) , (3.20)
and
vA[p] = v ∧ A[p] + v♭ ⌋A[p] , (3.21)
where A[p] is a p-vector.
Similarly, we can write
A[p] v = A[p] ∧ v + A[p] ⌊v♭ . (3.22)
Using eqn (2.15) for the exterior product, namely A[p] ∧ v = (−1)p v ∧ A[p] , as
well as eqn (2.47) which relates the left and the right contractions, that is A[p] ⌊v♭ =
−(−1)p v♭ ⌋A[p] , we obtain
A[p] v = (−1)p v ∧ A[p] − (−1)p v♭ ⌋A[p] . (3.23)
When we now compare eqns (3.21) and (3.23), we find that
1
v ∧ A[p] = vA[p] + (−1)p A[p] v) , (3.24)
2
and
1
v♭ ⌋A[p] = vA[p] − (−1)p A[p] v) . (3.25)
2
These equations can be further generalised into expressions involving an arbitrary
multivector A, as, respectively,
1 b ,
v∧A = vA + Av) (3.26)
2
and
1 b .
v♭ ⌋A = vA − Av) (3.27)
2
Observation ☞ The deep meaning of these last two equations is much more subtle
that what it might seem at first sight: they illustrate the relationship between the
Grassmann and the Clifford algebras. This fact is a prominent point that requires a
special discussion. Nevertheless, the relationship between these two algebras can be
interpreted incorrectly when the notation is used carelessly. We will see how this can
happen in what follows!
Universality
Let us now introduce the result that ensures that the Clifford algebra constructed as
the quotient of the tensor algebra is indeed a universal Clifford algebra.
Theorem 3.3 ◮ Let (V, g) be a quadratic space, let Cℓ(V, g) be the Clifford algebra
Cℓ(V, g) = T(V )/IC , and let (B, ρ) a Clifford algebra for (V, g). Then there exists a
homomorphism φ : Cℓ(V, g) → B such that ρ : φ ◦ γ, where γ is a Clifford mapping
γ : V → Cℓ(V, g).
Proof: For the Clifford algebra (B, ρ), let us consider the function ρ : V → B, where
ρ(v)2 = Q(v). This mapping ρ can be extended to T(V ) as the linear mapping ρ′ :
T(V ) → B given by
ρ′ (v1 ⊗ · · · ⊗ vk ) = ρ′ (v1 ) · · · ρ′ (vk ) = ρ(v1 ) · · · ρ(vk ).
There is a linear bijection from Lin(Tk (V ), B) onto Lin(k) (V, V, . . . , V ; B); the restric-
tion ρ′k : Tk (V ) → A is an element of the first space and corresponds to the element
defined by (v1 , . . . , vk ) 7→ ρ(v1 ) · · · ρ(vk ) in the second space.
Let us consider now the quotient space T(V )/ ker ρ′ , where the elements of T(V )/ ker ρ′
consist of the equivalence classes [x] constructed from x ∼ y ⇔ x − y ∈ ker ρ′ . We can
express [x] = π(x), where x ∈ T(V ), and π : T(V ) → T/ ker ρ′ . Hence, in this case,
there exists a mapping φ : T/ ker ρ′ → B given by
φ([x]) = ρ′ (x), ∀x ∈ T(V ),
which is the homomorphism
φ([x][y]) = φ([x ⊗ y]) = ρ′ (x ⊗ y) = ρ′ (x)ρ′ (y) = φ([x])φ([y]).
On the other hand, looking at ρ′ , we can immediately see that
ρ′ (v ⊗ v − Q(v)) = 0,
namely, IC ⊆ ker ρ′ . This result implies that there is a surjective homomorphism

T(V )/IC → T(V )/ ker ρ′ . Indeed, if Ui is a subspace of a vector space Wi (for i = 1, 2)
and if f : W1 → W2 is a linear mapping such that f (U1 ) ⊂ U2 , then f induces a
linear mapping f ′ : W1 /U1 → W2 /U2 . Moreover, when f is surjective, then f ′ is
surjective too. The surjective homomorphism T(V )/IC → T(V )/ ker ρ′ shows that
dim T(V )/IC ≥ dim T(V )/ ker ρ′ .
The homomorphism Cℓ(V, g) → B follows from

φ
Cℓ(V, g) = T(V )/IC → T(V )/ ker ρ′ −→ B .
The homomorphism φ is injective and even bijective if B is generated by ρ(V ). For
v ∈ V , it follows that φ([v]) = ρ′ (v) = ρ(v). Since [v] = γ(v), we obtain
φ ◦ γ = ρ,
which demonstrates the universality of Cℓ(V, g). ✓
Observation ☞ Although we refer to Cℓ(V, g) simply as the Clifford algebra, we mean
that it is the universal Clifford algebra.
3.3 Some General Considerations

Let B = {e1 , . . . , en } be an orthogonal basis of V . Equation (3.21) yields
ei ej = ei ∧ ej , (i 6= j), (3.28)
which can be used recursively to show that
eµ1 eµ2 · · · eµp = eµ1 ∧ eµ2 ∧ · · · ∧ eµp (µ1 6= µ2 6= · · · 6= µp ). (3.29)
Pn
The dimension of a universal Cℓ(V, g) is p=0 np , namely,
dim Cℓ(V, g) = 2dim V . (3.30)
ThereVexists an isomorphism between the Clifford algebra Cℓ(V, g) and the exterior
algebra (V ) or the Grassmann algebra G(V ). Clearly, this isomorphism is not an
algebra isomorphism (we will discuss this point in more details later on), but a vector
space isomorphism, ^
Cℓ(V, g) ≃ (V ). (3.31)
V
Hence, for the operations of the vector space structure, it is natural to use for the
Clifford algebra Cℓ(V, g) the same notation used for
V the exterior algebra, although the
subspace of the p-vectors will still be denoted by p (V ). Then the following equality
must be realised as an equality of vector spaces:
n ^
M
Cℓ(V, g) = p (V ). (3.32)
p=0
The projection operators are equivalently denoted by

^
h ip : Cℓ(V, g) → p (V ). (3.33)
The projection operator into the scalar part h i0 has prominent importance in the
formalism of Clifford algebras. It can be shown that, for A, B ∈ Cℓ(V, g), the scalar
part is such that
hABi0 = hBAi0 . (3.34)
Observation ☞ The underlying multivector structure of the Clifford algebra Cℓ(V, g)
is immediately seen in eqn (3.29), where an orthogonal basis is taken into account.
Some General Considerations 67
When the basis is not orthogonal, the multivector structure is not as straightforward
to realise: a Clifford algebra is a Z2 -graded algebra, whereas the exterior algebra and
Grassmann algebra are Zn -graded algebras. However, as the multivector structure is
basis independent, there is no reason to encounter this problem. Nonetheless, we want
to emphasise that calculations with Clifford algebras are more straightforward when
we work with an orthogonal basis than when we work with a non-orthogonal basis.
Hence, in most mappings, it is preferred to take an orthogonal basis into account – in
some cases, indeed, it is indispensable.
Let g be a symmetric bilinear form in Rn of signature (p, q), where p + q = n,
such that B = {e1 , . . . , en } is an orthonormal basis. On any vector v = v i ei ∈ V , the
symmetric bilinear form can be evaluated as
g(v, v) = (v1 )2 + · · · + (v p )2 − (v p+1 )2 − · · · − (vn )2 . (3.35)
We denote this quadratic space by Rp,q , and the corresponding Clifford algebra by
Cℓp,q :
Cℓp,q = Cℓ(Rp,q ). (3.36)
The centre Cen(Cℓp,q ) of the Clifford algebra Cℓp,q is defined as being the set of
elements in Cℓp,q that commute with all elements of Cℓp,q :
Cen(Cℓp,q ) = {a ∈ Cℓp,q | ax = xa, ∀x ∈ Cℓp,q }.
When a basis {e1 , . . . , en } is taken into account, every ei either commutes or anti-
commutes with eµ1 eµ2 · · · eµp , according to whether p is even or odd, respectively,
and whether i belongs or does not belong, respectively, to the set {µ1 , . . . , µp }. This
property provides the elements of the centre in Cℓ(V, g). It is left as an exercise to
show that
(V
(Rp,q ), if dim Rp,q is even,
Cen(Cℓp,q ) = V0 p,q V (3.37)
0 (R ) ⊕ n (R ), if dim Rp,q is odd.
p,q
The grade involution, the reversion, and the conjugation preserve the ideal IC . Then,
as in the exterior algebras, such operations pass to the quotient. Hence, for the grade
involution, it follows that
b[p] = (−1)p A[p] .
#A[p] = A (3.38)
For the reversion, we have the property
(A^ e e
[p] B[q] ) = B[q] A[p] , (3.39)
e[0] = A[0] , A
with A e[1] = A[1] . This result implies that
e[p] = (−1)p(p−1)/2 A[p] ,

A (3.40)
and, for the conjugation,
eb be
Ā[p] = A [p] = A[p] . (3.41)
Example 3.4 Equation (3.39) is straightforward to prove when A[p] and B[q] are products of vectors of
the orthogonal basis {e1 , e2 , . . . , en }. When some vectors ei are factors both in A[p] and in B[q] , we can,
without loss of generality, choose ep as the common vector that composes A[p] and B[q] . Up to a sign,
we can write A[p] and B[q] as
A[p] = e1 ∧ · · · ∧ ep , B[q] = ep ∧ · · · ∧ ep+q .
Then we obtain
A^
[p] B[q] = [(e1 ∧ · · · ∧ ep )(ep ∧ ep+q )]
∼
= [[(e1 ∧ · · · ∧ ep−1 )ep − (e1 ∧ · · · ∧ ep−1 )⌊ep ](ep ∧ · · · ∧ ep+q )]∼ .

Nevertheless, (e1 ∧ · · · ∧ ep−1 )⌊ep = 0, and hence
A^
[p] B[q] = [(e1 ∧ · · · ∧ ep−1 )ep (ep ∧ · · · ∧ ep+q )]
∼
= [(e1 ∧ · · · ∧ ep−1 )[ep ⌋(ep ∧ · · · ∧ ep+q )+ep ∧ (ep ∧ · · · ∧ ep+q )]]∼

= [(e1 ∧ · · · ∧ ep−1 )[ep ⌋(ep ∧ · · · ∧ ep+q )]]∼
= [(e1 ∧ · · · ∧ ep−1 )(ep+1 ∧ · · · ∧ ep+q )]∼
= [(e1 ∧ · · · ∧ ep−1 ) ∧ (ep+1 ∧ · · · ∧ ep+q )]∼
= (ep+q ∧ · · · ∧ ep+1 ) ∧ (ep−1 ∧ · · · ∧ e1 )
= [(ep+q ∧ · · · ∧ ep+1 )(ep−1 ∧ · · · ∧ e1 )]
= [(ep+q ∧ · · · ∧ ep+1 )(ep )2 (ep−1 ∧ · · · ∧ e1 )]
= [(ep+q ∧ · · · ∧ ep+1 )ep ][ep (ep−1 ∧ · · · ∧ e1 )]
= [(ep+q ∧ · · · ∧ ep+1 ) ∧ ep ][ep ∧ (ep−1 ∧ · · · ∧ e1 )]
e[q] A
=B e[p] . (3.42)
The Clifford algebra Cℓ(V, g) is clearly a Z2 -graded algebra. We can write
Cℓ(V, g) = Cℓ+ (V, g) ⊕ Cℓ− (V, g), (3.43)
where
1
Cℓ± (V, g) = Π± (Cℓ(V, g)) = (1 ± #)(Cℓ(V, g)). (3.44)
2
Consequently, it reads
Cℓ± (V, g) Cℓ± (V, g) ⊂ Cℓ+ (V, g), Cℓ± (V, g) Cℓ∓ (V, g) ⊂ Cℓ− (V, g), (3.45)
Since Cℓ± (V, g) Cℓ± (V, g) ⊂ Cℓ+ (V, g), the set Cℓ+ (V, g) is a subalgebra of the Clif-
ford algebra Cℓ(V, g). This subalgebra is prominent in the Clifford algebra formalism
and applications. It is called the even subalgebra:
b} .
Cℓ+ (V, g) = {A ∈ Cℓ(V, g) | A = #A = A (3.46)
The assertion that Cℓ+ (V, g) is a subalgebra can also be proved in a different
way. Let A[p] and B[q] be a p-vector and a q-vector, respectively. The repeated use of
eqn (3.21) permits us to conclude that
A[p] B[q] = hA[p] B[q] i|p−q| + hA[p] B[q] i|p−q|+2 + · · · + hA[p] B[q] ip+q . (3.47)
In general, the product of a p-vector and a q-vector in Cℓ+ (V, g) is not a graded-defined
multivector but a sum involving a |p − q|-vector, a (|p − q| + 2)-vector, and so on up
to a (p + q)-vector. From this equation, we obtain the result explicitly described by

eqn (3.45).
It is left as an exercise to show that, for A, B ∈ Cℓ(V, g), the following expression
holds:
e 0 = G(A, B) ,
hABi (3.48)
where G is the extension of g for the Grassmann algebra as given by eqns (2.99–2.101).
Using eqn (3.34) and the definition of the reversion, we can realise that hABi e 0 =
e
hBAi0 , according to the symmetry of G.
For elements of the Clifford algebra, if we try to define a norm |A|2 by
e 0,
|A|2 = hAAi (3.49)
it would not be the unique norm that we might define in Cℓ(V, g). Besides, it can have
negative values when g is not Euclidean. We can however define
|A|′2 = hĀAi0 . (3.50)

V
e[p] for A[p] ∈
Since Ā[p] = (−1)p A p (V ),
|A[p] |′2 = (−1)p |A[p] |2 . (3.51)
Nevertheless, these definitions do not ensure that the functions are non-negative.
Therefore, we opt to define, for simple multivectors, the norm as
|A|2 = |hĀAi0 | . (3.52)
When K is either R or C, the positive root is chosen to be the norm.

In the case where the multivectors are not simple, we have two possible definitions
for the norm:
e 0| ,
|A|2 = |hAAi (3.53)
|A|′2 = |hĀAi0 |. (3.54)
We can show that, for multivectors that are not simple, the norms in eqn (3.53) and
in eqn (3.54) can be considered, as they are in general, indeed, distinct. This fact is
illustrated in the following example.
Example 3.5 Let us consider the Clifford algebra Cℓ1,2 generated by {e2 , e2 , e3 } (here we adopt the
convention that e21 = −1), and take u = e1 − e2 e3 . The expressions for ũu and ūu then read
ũu = (e1 + e2 e3 )(e1 − e2 e3 ) = −1 + e22 e23 = 0,

ūu = (e1 − e2 e3 )(−e1 + e2 e3 ) = 2 + 2e1 e2 e3 6= 0.
Hence, |u|2 = 0, and |u|′2 = 2.

The Hodge isomorphism ⋆ can be written in terms of Cℓ(V, g) simply as
e V .
⋆A = AΩ (3.55)
In order to show this result, we first observe from eqn (3.47) that it is possible to write
A[p] ∧ B[q] = hA[p] B[q] ip+q . (3.56)
Equation (3.47) allows us to conclude that, as ΩV is an n-vector, therefore
hA[p] B[q] i0 ΩV = hA[p] B[q] ΩV in . (3.57)
Using the definition of the Hodge isomorphism and eqns (3.48), (3.56), and (3.57) we
find that
hB[p] ⋆ A[p] in = B[p] ∧ ⋆A[p] = G(B[p] , A[p] )ΩV
(3.58)
= hB[p] Ae[p] i0 ΩV = hB[p] A
e[p] ΩV in ,
from which eqn (3.55) ensues.
Example 3.6 Let V = R3 be endowed with the usual Euclidean scalar product. Consider the Clifford
algebra Cℓ3 , which is generated by 1 and {γi = γ(ei )}, where we take {ei } as the standard basis of R3 .
Take two vectors, for example, v = (3, 2, −1)⊺ , and u = (1, −1, 3)⊺ . In Cℓ3 , we write v = 3γ1 + 2γ2 − γ3 ,
and u = γ1 − γ2 + 3γ3 . The products vu and uv provide
vu = −2 − 5γ1 γ2 + 10γ1 γ3 + 5γ2 γ3 ,
uv = −2 + 5γ1 γ2 − 10γ1 γ3 − 5γ2 γ3 .
The symmetric part of the geometric product corresponds to the scalar product, and the alternating part
corresponds to the exterior product. Indeed,
1
(vu + uv) = −2 = g(v, u),
2
and
1
(vu − uv) = −5γ1 γ2 + 10γ1 γ3 − 5γ2 γ3 .
2
In particular, since g(ei , ej ) = 0 for i 6= j,
γi γj = −γj γi , (i 6= j) , γi2 = 1.
An arbitrary element of Cℓ3 can be written as
A = a0 + a1 γ1 + a2 γ2 + a3 γ3 + a12 γ1 γ2 + a13 γ1 γ3 + a23 γ2 γ3 + a123 γ1 γ2 γ3 .
Since γi γj = −γj γi , if i 6= j, then the element γ1 γ2 γ3 commutes with all elements of Cℓ3 . In addition,
(γ1 γ2 γ3 )2 = −1, so γ1 γ2 γ3 can be denoted by i:
i = γ1 γ2 γ3 .
Hence, iA = Ai (∀ A ∈ Cℓ3 ), and i2 = −1. The Hodge isomorphism applied to the multivector A then
reads
e
⋆A = Ai,
and, consequently,
γ1 γ2 = γ1 γ2 γ3 γ3 = iγ3 = ⋆γ3 ,
γ1 γ3 = γ1 γ2 γ2 γ3 = −iγ2 = − ⋆ γ2 ,
γ2 γ3 = γ1 γ1 γ2 γ3 = γ1 i = ⋆γ1 ,
Finally, an arbitrary element ψ ∈ Cℓ3 can be written as

A = (a0 + ia123 ) + (a1 + ia23 )γ1 + (a2 + ia31 )γ2 + (a3 + ia12 )γ3 .
As we remember from example 3.3, the Clifford algebra Cℓ3 is isomorphic to the algebra of complex
matrices of order 2. The isomorphism ρ is explicitly given by ρ(1) = 1 (the matrix identity) and ρ(γi ) = σi
(the Pauli matrices). An arbitrary element A ∈ Cℓ3 is mapped in A = ρ(A) given by
0
(a + a3 ) + i(a12 + a123 ) (a1 + a31 ) + i(a23 − a2 ) z1 z2
ρ(A) = A = 1 31 23 2 0 3 123 12 = .
(a − a ) + i(a + a ) (a − a ) + i(a −a ) z3 z4
In terms of the matrix representation, the reversion, the grade involution, and the conjugation correspond
to
∗ ∗
ρ(A)e = A† = z1∗ z3∗ ,
z2 z4
∗ ∗

ρ(A)b = adj(A† ) = z4∗ −z∗3 ,
−z2 z1

z4 −z2
ρ(Ā) = adj(A) = ,
−z3 z1
where the asterisk denotes complex conjugation.
Example 3.7 The Clifford algebra structure is very rich and exhibits a large number of particularities
because of the dimension and the signature associated with the quadratic space. For instance, the equality
uū = ūu does not always hold. Indeed, consider the Clifford algebra Cℓ3,1 ≃ M(4, R) of R3,1 . An element
u = (1 + e1 )(1 + e234 ) = 1 + e1 + e234 + e1234 ,
presents a Clifford conjugate
ū = (1 + e234 )(1 − e1 ) = 1 − e1 + e234 + e1234 .
By calculating the products between u and ū, we obtain
uū = 4(e234 + e1234 ),
ūu = 0 .
In the Clifford algebra associated with finite-dimensional spaces, the norm can be naturally defined as
q
|u| = huũi0 , for u ∈ Cℓn,0 ,
q
|u| = huūi0 , for u ∈ Cℓ0,n , (3.59)
where we used eqn (3.49) and eqn (3.50). Indeed, in Cℓ0,2 ≃ H, the corresponding conjugate q̄ =
√
w − ix − jy − kz is associated with q = w + ix + jy + kz ∈ H. The norm is given by |q| = q q̄, and
−1 2
the inverse is q = q̄/ |q| . Nevertheless, it is not always possible to calculate the inverse of an arbitrary
element u.
Example 3.8 Considering the Clifford algebra Cℓ3,1 , take u = e1 − e2 e3 . Computing uũ and uū we obtain
uũ = (e1 − e2 e3 )(e1 + e2 e3 ) = −1 + e22 e23 = 0,

uū = (e1 − e2 e3 )(−e1 + e2 e3 ) = 2 + 2e1 e2 e3 6= 0.
V V
Since uũ = 0, and uū ∈ 0 ⊕ 3 , no inverse element can be constructed for u = e1 − e2 e3 by using
an anti-automorphism of the Clifford algebra. Some particular and illustrative examples are investigated in
the work by Lounesto (1996).
In order to finish those considerations, now two general results concerning Clifford
algebras are presented.
Theorem 3.4 ◮ (M. Riesz) An orthonormal basis B = {e1 , e2 , . . . , en } generates a

2n -dimensional algebra, unless the pseudoscalar e1 e2 . . . en is a scalar multiple of the
identity.
Proof: At first, with at most one exception, the product involving one or more elements
of the basis B anti-commutes with at least one element of B. This property holds only
if the product of an element of the basis anti-commutes with any element appearing in
the product. Moreover, the product of an odd number of such elements anti-commutes
with any element that does not appear in the product. The unique product that
commutes with any element of the set {ei } is the pseudoscalar e1 e2 . . . en when n is
odd, according to eqn (3.37). Now the proof is by reductio ad absurdum. Consider the
products that are not linearly independent:
a0 + ai ei + aij ei ej + · · · + pe1 e2 · · · en = 0,
which can be also written in the notation of multiple indices. In this case, there exists
a set of coefficients ai1 i2 ...ik , with some of them non-zero, such that
n
X
ai1 i2 ...ik ei1 ei2 · · · eik = 0 . (3.60)
k=0
If the coefficient of 1 does not equal zero in eqn (3.60), we can divide the equation by
this coefficient, obtaining
n
X
1+ bi1 i2 ...ik ei1 ei2 · · · eik = 0, (3.61)
k=1
where the sum does not include the identity. On the other hand, if the coefficient of
1 in eqn (3.61) equals zero, we can take a term with a non-zero coefficient, let us say,
em1 em2 · · · emk ≡ em1 m2 ...mk , out and then multiply the equation by e−1 m1 m2 ...mk =
±em1 m2 ...mk . In this way, it is always possible to obtain an equation of the same type
as eqn (3.61) from eqn (3.60). If all the coefficients bi1 i2 ...ik are equal to zero, the
equality 1 = 0 is already a contradiction. If the sum in eqn (3.61) contains a product,
let us say, em1 m2 ...mk , which anti-commutes with some em , then we can multiply eqn
(3.61) from the left by em and from the right by e−1 m and thus obtain
n
2X −1
1+ bi1 i2 ...ik em ei1 i2 ...ik e−1
m = 0. (3.62)
k=0
Notice that
em em1 m2 ...mk e−1 −1

m = −em1 m2 ...mk em em = −em1 m2 ...mk .
Therefore, eqn (3.62) can be added to eqn (3.61), yielding a new equation that up
to a factor 2 is identical to eqn (3.60), except that now at least one term of type
(em1 m2 ...mk ) is not in the sum. If n is even, the process can be repeated until the sum
From the Grassmann Algebra to the Clifford Algebra 73
is reduced to 0, and we obtain the contradiction 1 = 0. If n is odd, then the procedure

can be repeated until
1 + ae1 e2 · · · en = 0,
for some scalar a 6= 0. Hence, we again obtain a contradiction, unless e1 e2 · · · en is
a scalar multiple of the identity. In fact, when we apply the grade involution to this
equation, we find that 1 − ae1 e2 · · · en = 0. Adding both equations, we arrive at the
contradiction; therefore, e1 e2 · · · en is a scalar multiple of the identity. This property
in particular holds for the real and complex fields, which are our main interests. ✓
The Clifford algebras that are not universal are restricted by the following theorem:
Theorem 3.5 ◮ If the Clifford algebra Cℓ(Rp,q ), generated by some orthonormal basis
{e1 , . . . , en }, is not universal, then n = p + q is odd. Moreover, if the Clifford algebra
is real and not universal, then p − q − 1 is an integer multiple of 4.
Proof: The first part of this theorem comes from the proof of theorem 3.4 (Riesz,
1993). The second part follows from the condition that η 2 = 1, where η = e1 e2 · · · en .
Then,
η 2 = e12...n e12...n = (−1)n(n−1)/2 en...21 e12...n = (−1)n(n−1)/2 (−1)q 1 .
Since η 2 = 1, it follows that

1
n(n − 1) + q = 2k, k ∈ N.
2
As n is odd, by hypothesis, n = 2m + 1, m ∈ N, and
(2m + 1)(2m) + 2q = 4k ⇒ 2m + 2q = 4(k − m2 ).
However, as 2m = n − 1 = p + q − 1, this equation reads
p + 3q − 1 = 4(k − m2 ) ⇒ p + 3q − 1 − 4q = 4(k − m2 ) − 4q
⇒ p − q − 1 = 4(k − m2 − q),
which proves the theorem. ✓
3.4 From the Grassmann Algebra to the Clifford Algebra
The Annihilation and the Creation Operators

For what will be discussed in this section, the exterior algebra would suffices. However,
if we endow the exterior algebra with a bilinear form, we get the Grassmann algebra
structure; therefore, we will situate our discussion in the context of the Grassmann
algebra. Let us then consider the Grassmann algebra G(V ) and define the annihilation
and the creation operators.1
1 The terms ‘creation’ and ‘annihilation’ are derived from the second quantisation formalism in
quantum field theory. Their use here is motivated by the algebraic relations that such operators
satisfy, as these relations are the similar to the ones in the second quantisation formalism.
The Operators E and E†

V V
Let us consider v ∈ V ⊂ (V ), and A ∈ (V ). In the Grassmann algebra G(V ), it
is possible to perform the exterior product between the vector v and the multivector
A, either as v ∧ A or as A ∧ v. Let us at first take the exterior product v ∧ A. Since
the result of this
V productV is a multivector, we can interpret the exterior productVas an
operation of (V ) on (V ), that is, an element ofV the endomorphism space of (V );
this space is denoted from this point on by End( (V )). Let us define E(v) as
E(v)(A) = v ∧ A. (3.63)
V V
The operator E is thus an object such that E : V → End( (V )), and E(v) : (V ) →
V
(V ). This operator E is called the creation operator. Moreover, the exterior product
between v and A can be taken in the reverse order, namelyVA ∧ v. Hence, we define
another operator, denoted by E† , such that E† : V → End( (V )). Define E† (v) as
E† (v)(A) = A ∧ v. (3.64)
There is a close relationship between the operators E and E† . Let us first consider
the case where A is a k-vector A[k] . In this case,2 eqn (2.15) yields
v ∧ A[k] = (−1)k A[k] ∧ v,

which implies that
v ∧ A = (#A) ∧ v. (3.65)
Hence, it follows that
E(v) = E† (v)#,
(3.66)
E† (v) = E(v)#.
The Operators I and I†

Other operations defined on the space of multivectors are the left and the right con-
tractions by a covector.
V ∗ Using those V operations, the operators I and IV† can be defined.
∗ ∗
Given α ∈ V ⊂ (V ), and A ∈ (V ), the operator I : V → End( (V )), called the
annihilation operator, is defined by
I(α)(A) = α⌋A. (3.67)
V
In addition, the operator I† : V ∗ → End( (V )) is defined by
I† (α)(A) = A⌊α. (3.68)
The relationship between those operators follows from eqn (2.48):
I(α) = −I† (α)#,
(3.69)
I† (α) = −I(α)#.
2 For b[k] .
the content of this section, the notation #A[k] is more appropriate than A
Commutation Relations
The anti-commutativity of the exterior product between two linearly independent
covectors implies that creation operators anti-commute:
E(v)E(u) + E(u)E(v) = 0 (3.70)
for all v, u ∈ V . Similarly,
E† (v)E† (u) + E† (u)E† (v) = 0. (3.71)
From eqn (2.50), the commutation relation between annihilation operators is given
by
I(α)I(β) + I(β)I(α) = 0 . (3.72)
In full compliance with eqn (2.51), we have
I† (α)I† (β) + I† (β)I† (α) = 0. (3.73)
Finally, there is a commutation relation between creation and annihilation opera-

tors. In this case,
I(α)E(v) + E(v)I(α) = α(v) , (3.74)
which follows from eqn (2.41), and
I† (α)E† (v) + E† (v)I† (α) = α(v), (3.75)
which also follows from eqn (2.46).
The Clifford Algebras Cℓ(V, +g) and Cℓ(V, −g)

Let us suppose that the vector space V is endowed with a symmetric correlation
♭ : V → V ∗ (or, equivalently, a symmetric bilinear form g : V ×V → R). In the previous
V
section the operatorsVE and I were defined in such a way that E : V → End( (V ))
and I : V ∗ → End( (V )). Since ♭ : V → V ∗ , the composition V of I ◦ ♭ can now be
introduced. It is immediately obvious Vthat I ◦ ♭ : V → End( (V )).V
Given the operators E : V → End( (V )), and I ◦ ♭ : V → End( V(V )), we can then
consider the sum of those operators. Let us define γ + : V → End( (V )) as
γ+ = E + I ◦ ♭ , (3.76)
V
and γ − : V → End( (V )) as
γ− = E − I ◦ ♭ . (3.77)
Using these definitions, we can assert the following theorem:

Theorem 3.6 ◮ The mappings γ ± are Clifford mappings. The quantities γ + (v)
(v ∈ V ) satisfy
γ + (v)γ + (u) + γ + (u)γ + (v) = 2g(v, u) , (3.78)
generating the Clifford algebra Cℓ(V, +g). On the other hand, the quantities γ − (v)
(v ∈ V ) satisfy
γ − (v)γ − (u) + γ − (u)γ − (v) = −2g(v, u) , (3.79)
generating the Clifford algebra Cℓ(V, −g).
In eqns (3.78) and (3.79), we omit (in order to simplify the notation) the presence
of an operator 1 in the right-hand side Vof these operations. Formally, we should have
written g(v, u)1, where 1A = A, ∀A ∈ (V ).
Proof: Only the definitions of γ ± and the commutation relations in eqns (3.70), (3.72),
and (3.74) need to be employed. First,
γ ± (v)γ ± (u) = [E(v) ± I(v♭ )][E(u) ± I(u♭ )]

= E(v)E(u) ± E(v)I(u♭ ) ± I(v♭ )E(u) + I(v♭ )I(u♭ ).
By using the commutation relations, we can see that
γ ± (v)γ ± (u) + γ ± (u)γ ± (v)

= [E(v)E(u) + E(u)E(v)] ± [E(v)I(u♭ ) + I(u♭ )E(v)]
± [I(v♭ )E(u) + E(u)I(v♭ )] + [I(v♭ )I(u♭ ) + I(u♭ )I(v♭ )]
= ±v♭ (u) ± u♭ (v) = ±2g(v, u),
which shows eqns (3.78) and (3.79). In the last step, the relation v♭ (u) = g(v, u) =
g(u, v) = u♭ (v) was used. In the book by Gilbert and Murray (1991), it is shown that
the Clifford algebra constructed in this way is indeed universal. ✓
Sufficiency of the Algebra Cℓ(V, +g)
The Clifford algebras Cℓ(V, +g) and Cℓ(V, −g) have been already defined; however, we
shall now show that it is not necessary to take into account both algebras.
The Clifford Mappings γ †±
The operator E(v) consists of the left exterior multiplication by the vector v, whereas
the operator E† (v) consists of the right exterior multiplication by the vector v. In
the same way, the operator I(v♭ ) is the left contraction by the covector v♭ , and the
operator I† (v♭ ) consists of the right contraction by the same vector. Since we defined
γ ± using the operations acting on the left, it is also possible to define γ †± employing
the same operations but now using the right action.
V
Let us thus define γ †± : V → End( (V )) as
γ †± = E† ± I† ◦ ♭. (3.80)
By using the commutation relations in eqns (3.71), (3.73), and (3.75), we obtain
γ †± (v)γ †± (u) + γ †± (u)γ †± (v) = ±2g(v, u). (3.81)
This result shows that the quantities {γ †± (v) | v ∈ V } satisfy the same relations as
{γ ± (v) | v ∈ V } do.
Let us now denote by Cℓ† (V, ±g) the algebras generated by {γ †± (v) | v ∈ V }. Since
the generators {γ †± (v)} satisfy the same commutation relations as {γ ± (v)}, therefore
Cℓ† (V, +g) and Cℓ(V, g) are isomorphic, as they are both universal.
The Relationship between γ ± and γ †±

Two important results established by the creation and the annihilation operators were
summarised in eqns (3.66) and (3.69). Using these equations,3 we find that
γ †± = E# ± (−I#) ◦ ♭ = E# ∓ (I ◦ ♭)#, (3.82)
namely,
γ †± = γ ∓ #,
. (3.83)
γ ∓ = γ †± #
This result shows that the algebra generated by γ − is isomorphic to the algebra
generated by γ †+ .4 Likewise, the algebra generated by γ + is isomorphic to the algebra
generated by γ †− . Explicitly,
Cℓ† (V, +g) ≃ Cℓ(V, −g), Cℓ† (V, −g) ≃ Cℓ(V, +g). (3.84)
We can thus conclude that it is not necessary to consider both of the Clifford algebras
Cℓ(V, +g) and Cℓ(V, −g). In each of these algebras, the generators satisfy distinct com-
mutation relations – namely, the relationship given by eqns (3.78) and (3.79), respec-
tively. Therefore, it is sufficient to consider just one of them, for instance, Cℓ(V, +g),
and to take into account the left and the right actions of the generators.
Consider now the quantities γ + (v) and γ †+ (u). It is immediately obvious that the
product between these quantities commutes, namely,
γ + (v)γ †+ (u) − γ †+ (u)γ + (v) = 0. (3.85)
This result is equivalent to
γ + (v)γ − (u) + γ − (u)γ + (v) = 0. (3.86)

3 Care must be taken not to confuse the symbols # and ♯ in eqn (3.82).
4 The presence of the grade involution # in eqn (3.83) is not relevant for this type of argument.
Equation (3.85) is particularly important, since it allows us to write the product which
acts on both sides and satisfies the property of associativity. Indeed, writing
^
Aγ + (v) = γ †+ (v)(A), ∀v ∈ V , ∀A ∈ (V ) (3.87)
yields
(γ + (v)A)γ + (u) = γ + (v)(Aγ + (u)) = γ + (v)Aγ + (u). (3.88)
To summarise, it is enough to examine the algebra generated by {γ + (v) | v ∈ V },
via both left and right multiplications, since the same quantity satisfies the same
commutation relation, no matter whether the left or the right actions are considered.
Since we are just taking Cℓ(V, +g) = Cℓ(V, g) into account, we can omit the index ‘+’
in the generators {γ(v) | v ∈ V }.
Chevalley was the first to propose
V the definition Cℓ(V, g) = T(V )/IC in the context
of the mapping γ + : V → End( (V )) defined in eqn (3.76). By the universal property
of theV algebra T(V ), the mapping γ + extends to an algebra homomorphism T →
End( (V )), which can also be denoted by γ + , vanishing V on IC . Consequently, it
factorises through Cℓ(V, g), as T(V ) → Cℓ(V, g) → End( (V )). It suffices to remember
eqn (3.17), which asserts that γ + (vp ⊗· · ·⊗v2 ⊗v1 )(1),
V where v1 , v2 , . . . , vp ∈ V , is the
sum of vp ∧ · · · ∧ v2 ∧ v1 and some other elements of (V ) of degree less than V p, which
we proved in example 3.3. This result gives a surjective mapping
V T(V ) → (V ), which
factorises through Cℓ(V, g), whence dim(Cℓ(V, g)) ≥ dim( (V )). Since we proved V just
after eqn (3.6) that the maximal dimension
V of a Clifford algebra is less than dim( (V )),
it is clear that dim(Cℓ(V, g)) = dim( (V )).
3.5 Grassmann Algebra versus Clifford Algebra
The Relationship between Clifford and Grassmann Products

The Clifford product (or geometric product) between two multivectors A and B will
be denoted by γ(A)γ(B).5 In order to realise the meaning of this product, we start
by considering the case where there are two vectors v and u originating the Clifford
product γ(v)γ(u), where γ : V → Cℓ(V, g). The associativity of Clifford algebras can
be expressed by the equation
γ(v)γ(u) = γ[γ(v)(u)]. (3.89)
Using now the definition of γ, we obtain
γ(v)(u) = v ∧ u + g(v, u), (3.90)
and then
γ(v)γ(u) = γ(v ∧ u) + g(v, u) , (3.91)
where we omit the presence of the operator 1, that is, we use γ(g(v, u)) = g(v, u)1,
in order to simplify the notation.
5 In this section, the explicit use of the Clifford mapping γ in the formulæ involving the product in
the Clifford algebra is intended to help (at least we believe it will!) in comprehending the relationship
between the Clifford and the Grassmann algebras.
Grassmann Algebra versus Clifford Algebra 79
This equation is well known, although by another notation: it is eqn (3.9). This
equation shows that the Clifford product can be thought as being composed of two
parts. The first part, corresponding to the term γ(v ∧ u), is the image in Cℓ(V, g) of
the exterior product v ∧u.6 The second part is the image in Cℓ(V, g) of the contraction
v♭ ⌋u, which equals in this case the scalar product g(v, u).
Equation (3.91) is useful for understanding the meaning regarding the Clifford
product between a vector and a bivector, namely, the product γ(v)γ(u ∧ w). Using
eqn (3.91), we can write
γ(v)γ(u ∧ w) = γ(v)γ(u)γ(w) − g(u, w)γ(v). (3.92)
On the right-hand side of eqn (3.92), the first term can be calculated by using again
eqn (3.89). It reads
γ(v)γ(u)γ(w) = γ[γ(v)[γ(u)(w)]] = γ[γ(v)(u ∧ w + g(u, w))]
= γ(v ∧ u ∧ w + v♭ ⌋(u ∧ w) + g(u, w)γ(v))
= γ(v ∧ u ∧ w)+g(v, u)γ(w)−g(v, w)γ(u)+g(u, w)γ(v).
Using now this result in eqn (3.92), we can obtain
γ(v)γ(u ∧ w) = γ(v ∧ u ∧ w) + γ(v♭ ⌋(u ∧ w)), (3.93)
where
γ(v♭ ⌋(u ∧ w)) = g(v, u)γ(w) − g(v, w)γ(u). (3.94)
Equation (3.93) shows that, as in the case involving two vectors, the Clifford prod-
uct between a 1-vector and a 2-vector can be split in two parts: one corresponding to
the image of the exterior product and the other part corresponding to the image of
the contraction.
A similar reasoning can be repeated for the case involving a 1-vector and a 3-vector,
a 4-vector, or in general any p-vector. The result is
γ(v)γ(A[k] ) = γ(v ∧ A[k] ) + γ(v♭ ⌋A[k] ), (3.95)
V
where A[k] ∈ k (V ). Moreover, since the mapping γ is linear, we can write for an
arbitrary multivector A the expression
γ(v)γ(A) = γ(v ∧ A) + γ(v♭ ⌋A) . (3.96)
Another result that follows from a completely analogous procedure is
γ(A)γ(v) = γ(A ∧ v) + γ(A⌊v♭ ) , (3.97)
providing the meaning to the product γ(A)γ(B), in particular when A = v (vector),
and B = u ∧ w (bivector). Equations (3.96) and (3.97) are equivalent to eqns (3.21)
and (3.22).
6 We must realise that it does not make sense to write something like γ(α) ∧ γ(β). In Cℓ(V, g),
there exists only one product – the Clifford product – denoted by juxtaposition most of the time
here. In fact, it is well known that the exterior product ∧ only makes
V sense in the exterior algebra or
in the Grassmann
V algebra. However, the identification between (V ) and Cℓ(V, g), coming from the
bijection (V ) → Cℓ(V, g) induced V by the quotient mapping π : T(V ) → Cℓ(V, g), further provides
the identification between A ∈ (V ) and π(A) ∈ Cℓ(V, g). Hence, we can write the image of the
exterior product in Cℓ(V, g) by using the Clifford product, as we shall show soon.
It is worthwhile emphasising that the structure of these equations – namely, the

decomposition of the geometric product into a part involving the exterior product and
a part involving the contraction – holds only when one of the elements in the product
is a vector. Let us analyse the following example. The geometric product involving
two 2-vectors can be calculated with the aid of eqn (3.91), yielding
γ(v1 ∧ v2 )γ(u1 ∧ u2 ) = γ(v1 ∧ v2 ∧ u1 ∧ u2 ) + g(v1 , u2 )γ(v2 ∧ u1 )

− g(v1 , u1 )γ(v2 ∧ u2 ) + g(v2 , u1 )γ(v1 ∧ v2 ) − g(v2 , u2 )γ(v1 ∧ u1 ) (3.98)
+ g(v2 , u1 )g(v1 , u2 ) − g(v2 , u2 )g(v1 , u1 ).
Hence, the Clifford product involving two 2-vectors does not consist simply of the sum
of a term involving the exterior product and another one involving the contraction.
Those two parts are indeed present, where the exterior product is expressed in the
first term on the right-hand side of the equation and the contraction in the two last
terms on the right-hand side:
(v1♭ ∧ v2♭ )⌋(u1 ∧ u2 ) = v1♭ ⌋(v2♭ ⌋(u1 ∧ u2 ))

= g(v2 , u1 )g(v1 , u2 ) − g(v2 , u2 )g(v1 , u1 ). (3.99)
Such terms contribute to 4-vector and 0-vector (scalar) parts, and the other terms
contribute with a 2-vector part. In general, the Clifford product between a p-vector
and a q-vector splits into the sum of a (p + q)-vector, a (p + q − 2)-vector, a (p + q − 4)-
vector, . . . , up to a |p − q|-vector part, according to eqn (3.47).
It is worthwhile emphasising a peculiarity of the Clifford product between two
2-vectors. From eqn (3.98), we can observe that
γ(v1 ∧ v2 )γ(u1 ∧ u2 ) − γ(u1 ∧ u2 )γ(v1 ∧ v2 ) = 2[g(v1 , u2 )γ(v2 ∧ u1 )

(3.100)
− g(v1 , u1 )γ(v2 ∧ u2 ) + g(v2 , u1 )γ(v1 ∧ v2 ) − g(v2 , u2 )γ(v1 ∧ u1 )].
Here it is opportune to define the commutator [γ(A), γ(B)] as
[γ(A), γ(B)] = γ(A)γ(B) − γ(B)γ(A). (3.101)
Equation (3.100) shows that the commutator of two 2-vectors is a 2-vector. The Clifford
product of two 2-vectors yields
γ(v1 ∧ v2 )γ(u1 ∧ u2 ) = γ(v1 ∧ v2 ∧ u1 ∧ u2 )

1 (3.102)
+ [γ(v1 ∧ v2 ), γ(u1 ∧ u2 )] + (v1♭ ∧ v2♭ )⌋(u1 ∧ u2 ).
2
The Exterior Product in Clifford Algebra

Since it was previously demonstrated that the CliffordV product contains a part that
is the image, in Cℓ(V, g), of the exterior product in (V ) or G(V ), it is expected that
Grassmann Algebra versus Clifford Algebra 81
the exterior product (namely, the image of the exterior product) can be expressed by
the geometric product. Indeed, eqn (3.97) reads
γ(A)γ(v) = γ(v ∧ #A) − γ(v♭ ⌋#A),
where eqns (2.48) and (3.65) were used. Hence, it is equivalent to
γ(#A)γ(v) = γ(v ∧ A) − γ(v♭ ⌋A). (3.103)
When we take this equation together with eqn (3.96), we obtain
1
γ(v ∧ A) = [γ(v)γ(A) + γ(#A)γ(v)] . (3.104)
2
This equation is actually eqn (3.26) in a different notation.
Contractions in Clifford Algebra

The expression for the left contraction is straightforward:
1
γ(v♭ ⌋A) = [γ(v)γ(A) − γ(#A)γ(v)] . (3.105)
2
It is also presented in another notation by eqn (3.27).
The expression for the right contraction is also immediately obvious:
1
γ(A⌊v♭ ) = [γ(A)γ(v) − γ(v)γ(#A)]. (3.106)
2
Grassmann Algebra in Clifford Algebra and Vice Versa

The last results in the previous section show that the Grassmann algebra G(V ) op-
erations can be completely reproduced by operations in the Clifford algebra Cℓ(V, g),
and vice versa.7 This fact, however, does not mean that these algebras are equivalent
– or, in a technical sense, isomorphic. Specifically, the Grassmann and Clifford alge-
bras are isomorphic from the point of view of vector spaces but are not algebraically
isomorphic.
The Clifford product can be represented in the Grassmann algebra as a function
of eqn (3.90). When g is defined as
α g A = γ(α)A, (3.107)
it immediately follows that
α g β + β g α = 2g(α, β). (3.108)
It is important to emphasise that the product g is associated with operations in the
Grassmann algebra. In other words, the underlying algebra concerning the product g is
7 We must remember here the distinction between the exterior and the Grassmann algebras. Their
equivalence just holds in the context of the Grassmann algebra – the exterior algebra endowed with
the respective generalisation of the bilinear functional in V – not the exterior algebra.
the Grassmann algebra, not the Clifford algebra. Hence, the product g provides a rep-
resentation of the geometric product of the Grassmann algebra. Although eqns (3.108)
and (3.18) are similar, the products g and ⋄ involved in those equations are different.
As algebras endowed with the products g and ⋄, they are clearly isomorphic but, since
g is defined in eqn (3.107) by an operator acting on the Grassmann algebra G(V ), the
product ⋄ is the image of the tensor product in the quotient algebra T(V )/IC .
The isomorphism of vector spaces Cℓ(V, g) ≃ G(V ) is provided by the mapping
λ : Cℓ(V, g) → G(V ) defined as
λ(γ(v)γ(A)) = v g λ(γ(A)) = v ∧ λ(γ(A)) + v♭ ⌋λ(γ(A)), (3.109)
where
λ(1) = 1. (3.110)
Then eqn (3.110) implies that
λ(γ(v)) = λ(γ(v)γ(1)) = v g λ(1) = v,
and subsequently that
λ(γ(v)γ(u)) = v g u = v ∧ u + g(v, u).
The mapping λ is not an algebraic isomorphism, since it does not satisfy λ[γ(Ψ)γ(Φ)] =
[λ(γ(Ψ))] ∧ [λ(γ(Φ))].
To summarise, the exterior product and the contractions of the Grassmann algebra
can be ‘simulated’ in the Clifford algebra, and the product of the Clifford algebra can
be emulated in the Grassmann algebra. Thus, any calculation in either one of those
structures can be reproduced in the other.
Nevertheless, sometimes it is advantageous to work in one or the other. One of
the great advantages of the Clifford algebra as compared to the Grassmann algebra
is that, via the Clifford product, we can define the inverse of a multivector, which we
cannot do via the exterior product. Given a multivector A, there exists an inverse if
there exists another multivector B such that, when it is left or right multiplied by A,
the result is unity. In some cases, it is possible to accomplish it in the Clifford algebra;
one of the simplest examples is that concerning a non-isotropic vector v (g(v, v) 6= 0).
Clearly, the inverse of v does not exist in the Grassmann algebra, since v ∧ v = 0.
Thus, for any multivector B, it follows that B ∧ v = ±v ∧ B 6= 1. In contrast, in the
Clifford algebra, there always exists an inverse of v. Indeed,
γ(v)γ(v) = γ(v ∧ v) + g(v, v) = g(v, v),
and, since g(v, v) 6= 0, we can define
γ(v)−1 = γ(v)/g(v, v).
In addition, γ(v)−1 γ(v) = γ(v)γ(v)−1 = 1 holds.

From the computational point of view, it is more straightforward to regard a struc-
ture where the inverse does exist than to take a framework where it does not. Doing
so can sometimes reduce many pages of calculations to just few lines.
Notation 83
However, the Grassmann algebra has some advantages over the Clifford algebra.
In particular, the exterior product permits us to define the multivector structure in a
natural manner which is independent of a vector space basis. This advantage comes
from the fact that the exterior algebra is the most basic and general structure when
we take into account the Clifford and Grassmann algebras, since it does not demand
the existence of a correlation. Indeed, in this text, we used the exterior product to
define the multivector structure of p-vectors.
As for the question of whether it is possible to define a multivector structure by
using the Clifford product, it is straightforward to do so when an orthogonal basis is
taken into account. Indeed, if v and u are orthogonal, that is, g(v, u) = 0, then
γ(v)γ(u) = γ(v ∧ u).
Hence, the geometric product can be used to define a multivector structure since, in
this case, γ(v)γ(u) is a 2-vector. In this case, the Clifford product of p orthogonal
vectors is a p-vector.
Obviously, this construction does depend upon the orthogonality between vectors.
If v and u are not orthogonal, then the geometric product between them is not (only)
a 2-vector. Indeed, in order to establish the multivector structure, we must use the
geometric product given by eqn (3.104) as the expression for the exterior product.
However, this is just a way to disguise using the exterior product! To summarise, in
order to define the multivector structure in a basis-independent way, it is best to use
the exterior product.
3.6 Notation
Notations seem to satisfy a kind of ‘uncertainty principle’: the more precise they are,
the less operational, and vice versa. Thus, it is important to fully understand the
notation and know how to use it appropriately. Let us establish in this section a
notation that is adapted for use in calculations.
Let B = {ei } be a basis of V and let B∗ = {ei } be its associated dual basis (where
i = 1, . . . , n). Let g be a symmetric bilinear functional defined by gij = g(ei , ej ) = gji ,
and g ij = g −1 (ei , ej ) = g ji . It follows that ♭(ei ) = ei♭ = gij ej , and ♯(ei ) = ei♯ = g ij ej .
Then,
v♭ = v i ei♭ = vi gij ej = vi ei ,
where vi = gij v j .
The operators E and I are linear. Thus, the action of one of them on elements of
a basis of the vector space V suffices to completely determine the action on V . Let us
denote
Ei = E(ei ), Ii = I(ei ). (3.111)
Hence, I ◦ ♭(ei ) = gij Ij can be written as
γi = Ei + gij Ij , (3.112)
where
γi = γ(ei ). (3.113)
The quantities {γi } (i = 1, . . . , n) are the generators of the Clifford algebra Cℓ(V, g)
and thus satisfy
γi γj + γj γi = 2gij . (3.114)
In addition,
γ(v) = v i γi . (3.115)
Using an analogous reasoning, we write
γ(α) = αi γ i , (3.116)
where γ(ei ) = g ij γ(ej ), that is,
γ i = g ij γj , (3.117)
which satisfies
γ i γ j + γ j γ i = 2g ij . (3.118)
Finally, it is common to conceal the mapping γ in the expressions given in eqns
(3.115) and (3.116) by directly writing
v = v i γi , α = αi γ i . (3.119)
Sometimes we need to go further and actually omit the mapping. In that case, we can
express v = v i ei , where
ei ej + ej ei = 2gij . (3.120)
The choice of one or the other notation clearly depends on the context.

There are many texts that we can suggest for additional reading on the content of
this chapter. Besides the references already provided throughout the text, a clear and
concise introduction to Clifford algebras, with elements of the theory of Dirac opera-
tors and Clifford analysis, has been provided by Garling (2011); a set of introductory
lectures about Clifford algebras and analysis, as well as applications in physics and
engineering can be found in Ablamowicz and Sobczyk (2004). Further mathematical
developments of Clifford algebras are described in the work by Helmstetter and Micali
(2008). With respect to applications, there are many, mainly in physics but also in
different areas of engineering. An introduction to Clifford algebras, written for physi-
cists, including applications in mechanics, electromagnetism, quantum mechanics, and
gravitation, has been provided by Doran and Lasenby (2003). Moreover, a series of
papers by Hiley (2011) and by Binz et al. (2013) deal with applications in quantum me-
chanics. Another great book about Clifford algebras, with an emphasis on differential
geometry, and on applications in physics, has been written by Benn and Tucker (1987).
Many different applications of Clifford algebras in physics, engineering and computer
science have been discussed, for example, in the work by Baylis (1996), Sommer, and
Dorst and Lasenby (2011). General properties of Clifford algebras can be also found
in, for example, the work by Gallier (1997) and Todorov (2011). For a Clifford bundle
approach and in particular for the interplay among the Dirac, the Einstein, and the
Maxwell equations, see the work by Rodrigues and de Oliveira (2007).
Exercises 85
3.8 Exercises
V V
(1) Let W be a vector subspace of (R5,0 ), given by W = R ⊕ R5,0 ⊕ 2 (R5,0 ) and
for which the dimension is 1 + 5 + 5(5 − 1)/2 = 25 /2 = 24 . Define a product ∗ in W
as (Lounesto, 2001a)
a ∗ b = hab(1 + e1 e2 e3 e4 e5 )i0⊕1⊕2
= hab(1 + e1 e2 e3 e4 e5 )i0 + hab(1 + e1 e2 e3 e4 e5 )i1 + hab(1 + e1 e2 e3 e4 e5 )i2 ,
where {ei } (i = 1, 2, 3, 4, 5) is an orthonormal basis of R5,0 , and ab is the usual Clifford
algebra product Cℓ(R5 , g).
(a) Show that the algebra A = (W, ∗) is associative. (b) Show that v ∗ v = g(v, v) for
v ∈ R5,0 . The algebra A is thus an example of a Clifford algebra that is not universal.
(c) Show that e1 ∗ e2 ∗ e3 ∗ e4 ∗ e5 = 1. (d) Show that the generators of A can be
represented by the following matrices:

0 −i 0 −j 0 −k 1 0 01
e1 = , e2 = , e3 = , e4 = , e5 = ,
i 0 j 0 k 0 0 −1 10
where i, j, k denote the quaternionic units.
V V
(2) In (R2 ) = R ⊕ R2 ⊕ 2 (R2 ), define a product ⊙ as
e1 ⊙ e1 = 1, e1 ⊙ e2 = e1 ∧ e2 + b,
e1 ⊙ (e1 ∧ e2 ) = e2 − be1 , e2 ⊙ e1 = −(e1 ∧ e2 ) − b,
e2 ⊙ e2 = 1, e2 ⊙ (e1 ∧ e2 ) = −e1 − be2 ,
(e1 ∧ e2 ) ⊙ e1 = −e2 − be1 , (e1 ∧ e2 ) ⊙ e2 = e1 − be2 ,
(e1 ∧ e2 ) ⊙ (e1 ∧ e2 ) = −1 − b2 − 2b(e1 ∧ e2 ).
V
(a) Show that the algebra ( (R2 ), ⊙) is a Clifford algebra for the quadratic space R2 .
(b) Show that this algebra is isomorphic to the Clifford algebra Cℓ(R2 ). (c) Show that
the generators of this algebra can be represented by

1 0 01 −b 1
e1 = , e2 = , e1 ∧ e2 = .
0 −1 10 −1 −b
V
(d) Although the Clifford algebras ( (R2 ), ⊙) and Cℓ(R2 ) are isomorphic, they are
clearly different Clifford algebras. Which is the relationship between these algebras?
(Hint: how could you define a Clifford algebra for a vector space equipped with a
bilinear form that is not necessarily symmetric?)
(3) Let {v1 , . . . , vn } be an arbitrary basis of (V, g) and let {θ 1 , . . . , θn } be its associated
dual basis. Prove that the product ψφ in Cℓ(V, g) between the multivectors ψ and φ
can be written as
Xn
(−1)p(p−1)/2
ψφ = gi1 j1 . . . gip jp #p (θi1 ∧ · · · ∧ θip )⌋ψ ∧ (θ j1 ∧ · · · ∧ θ jp )⌋φ ,
p=0
p!
where the sum convention is assumed in the indices i1 , . . . , ip and j1 , . . . , jp . The

Clifford algebra represented by the product defined by the right-hand side in this
equation is sometimes called Kähler–Atiyah algebra.
V
(4) Show that, if n = dim Rp,q is even,
V then Cen(Cℓ
V p,q ) = 0 (R
p,q
). In addition, prove
p,q p,q
that, if n is odd, then Cen(Cℓp,q ) = 0 (R ) ⊕ n (R ).
(5) Let η = e1 e2 · · · en be a unit n-vector of the Clifford algebra Cℓ(V, g), where
V = Rp,q is a n-dimensional vector space and {e1 , . . . , en } is an orthonormal basis of
V , namely
g(ei , ei ) = 1, i = 1, . . . , p,
g(ei , ei ) = −1, i = p + 1, . . . , p + q = n,
g(ei , ej ) = 0, i 6= j.
(a) Show that
η 2 = 1, if p − q = 0, 1 mod 4,
2
η = −1, if p − q = 2, 3 mod 4.
This results also holds, of course, for η = −e1 e2 · · · en , in such a way that it does not
depend on the orientation of V . (b) Show that, for p − q = 0, 1 mod 4, we can write
Cℓ(V, g) = + Cℓ(V, g) ⊕ − Cℓ(V, g), where
± Cℓ(V, g) = {ψ ∈ Cℓ(V, g) | ψ = ±ψη}.

(c) Show that, if p − q = 1 mod 4, then ± Cℓ(V, g) are bilateral ideals of Cℓ(V, g), that
is, that
± Cℓ(V, g) Cℓ(V, g) ⊂ ± Cℓ(V, g) and Cℓ(V, g) ± Cℓ(V, g) ⊂ ± Cℓ(V, g) .
(6) Let {ei } (i = 1, . . . , n) be an orthogonal basis for the quadratic space (V, g). Use
multi-index notation, that is, write {eI } where
I = {1, . . . , n, 12, . . . , 1n, 123, . . . , . . . , 12 · · · n}
and
ei1 ···ik = ei1 · · · eik = e1 ∧ · · · ∧ eik
V
such that {eI } is an orthonormal basis with respect to the extension G of g for (V )
(eqns (2.97)–(2.100)). (a) Prove eqn (3.48). (b) Show that any element ψ ∈ Cℓ(V, g)
can be written as X
ψ= hψeeI i0 eI ,
I
I
where {e } is the dual basis of {eI }.
(7) Let {v1 , v2 , . . . , vn } be a basis for the quadratic space (V, g) and let Cℓ(V, g) be
the associated Clifford algebra. The multivectors φ0 , φ1 , φ2 , . . . , φn are defined as
φ0 = 1, φ1 = v1 , φ2 = hv1 v2 i2 , . . . , φn = hv1 v2 · · · vn in .
Show that the vectors
ei = φei−1 φi , i = 1, . . . , n,
are orthogonal and that the process of showing that they are orthogonal exactly cor-
responds to the Gram–Schmidt procedure.
4
Classification and Representation of
the Clifford Algebras
In this chapter, the classification and representation of Clifford algebras are introduced
and discussed. We denote by M(n, K) the algebra of n × n matrices with entries
in K = R, C, H, where H denotes the set of quaternions. Some important theorems
regarding the structure of Clifford algebras are presented; these are used later on for
the classification and construction of the representations of the Clifford algebras. In
addition, procedures to explicitly build Clifford algebras matrix representations are
presented.
4.1 Theorems on the Structure of Clifford Algebras

In this section, some important and essential theorems regarding the structure and
classification of Clifford algebras are introduced and investigated. As well as being
important in themselves, these theorems are useful in the classification of the Clifford
algebras as matrix algebras.
Let us start by defining the alternating tensor product of two graded algebras, A
and B. The alternating tensor product A⊗B ˆ between those algebras is defined as the
algebra generated by the product a⊗b, ˆ a ∈ A, b ∈ B, with the product being defined
by
(a1 ⊗b ˆ 2 ) = (−1)deg (b1 ) deg (a2 ) a1 a2 ⊗b
ˆ 1 )(a2 ⊗b ˆ 1 b2 . (4.1)
This definition applies to homogeneous elements and can be extended by linearity. The
tensor product ⊗ˆ is the usual tensor product and the hat notation reminds us that
the product must take into account the grading of the elements involved.
The theorem that is key for Clifford algebra mod 8 periodicity shall be proved in
what follows. However, first we shall enunciate a lemma that is important for proving
it. Details on the proof of the lemma can be found, for example, in the book by
Helmstetter and Micali (2008).
Lemma 4.1 ◮ Let A and B be graded algebras, and let f : A → C, and f ′ : B → C,

be algebra homomorphisms such that ∀a ∈ A and ∀b ∈ B,
f (a)f ′ (b) = (−1)deg (a) deg (b) f ′ (b)f (a).
There exists a unique algebra morphism f˚ : A⊗B

ˆ → C such that f (a) = f˚(a ⊗ 1B ),
′ ˚
and f (b) = f (1A ⊗ b).
88 Classification and Representation of the Clifford Algebras
Now we can enunciate and prove the prominent theorem that endows us with the
technique necessary to introduce the mod 8 periodicity of Clifford algebras, as well as
other structure theorems:
Theorem 4.1 ◮ Let (V, g) and (V ′ , g ′ ) be two quadratic spaces and let Cℓ(V, g) and
Cℓ(V ′ , g ′ ) be their respective Clifford algebras. Then
Cℓ(V ⊕ V ′ , g ⊕ g ′ ) ≃ Cℓ(V, g)⊗
ˆ Cℓ(V ′ , g ′ ), (4.2)
′
where ≃ denotes an isomorphism between Clifford algebras, and V ⊕ V stands for the
orthogonal direct sum of V and V ′ .
Proof: Let g ⊕ g ′ be the symmetric bilinear form defined in V ⊕ V ′ , considered to be
the orthogonal direct sum of V and V ′ . Therefore
(g ⊕ g ′ )(v + v′ , u + u′ ) = g(v, u) + g ′ (v′ , u′ ), ∀ v, u ∈ V, ∀ v′ , u′ ∈ V ′ .
ˆ Cℓ(V ′ , g ′ ) can be defined as
The mapping Γ : V ⊕ V ′ → Cℓ(V, g)⊗
Γ(v + v′ ) = v⊗1 ˆ ′,
ˆ v′ + 1v ⊗v
where it is implicit that v = γ(v), and v′ = γ ′ (v′ ). In addition, since γ : V → Cℓ(V, g),
and γ ′ : V ′ → Cℓ(V ′ , g ′ ), are Clifford mappings, then the mapping Γ is also a Clifford
mapping. Indeed,
2
Γ(v + v′ ) = (v⊗1 ˆ v′ + 1v ⊗vˆ ′ )(v⊗1 ˆ ′)
ˆ v′ + 1v ⊗v
= (−1)0·1 (v)2 ⊗1
ˆ v′ + (−1)0·0 v⊗v
ˆ ′
+ (−1)1·1 v⊗vˆ ′ + (−1)0·1 1v ⊗(v ˆ ′ )2
= g(v, v)⊗1 ˆ ′ (v′ , v′ )
ˆ v′ + 1v ⊗g

= g(v, v) + g ′ (v′ , v′ ) 1v ⊗1
ˆ v′
= (g ⊕ g ′ )(v + v′ , v + v′ )1v ⊗1
ˆ v′ .
Moreover, for finite-dimensional Clifford algebras, we have dim Cℓ(V, g)⊗ ˆ Cℓ(V ′ , g ′ ) =
dim V dim V ′ dim(V ⊕V ′ ) ′ ′
2 2 = 2 . On the other hand, let Cℓ(V ⊕ V , g ⊕ g ) be the Clif-
ford algebra associated to the quadratic space (V ⊕ V ′ , g ⊕ g ′ ). From the universality
theorem, there exists a homomorphism
φ : Cℓ(V ⊕ V ′ , g ⊕ g ′ ) → Cℓ(V, g)⊗
ˆ Cℓ(V ′ , g ′ ) . (4.3)
′ ′ dim(V ⊕V ′ ) ˆ Cℓ(V , g ), it follows that φ is
′ ′
As dim Cℓ(V ⊕ V , g ⊕ g ) = 2 = dim Cℓ(V, g)⊗
surjective.
Now, the canonical injections ι : (V, g) → (V ⊕ V ′ , g ⊕ g ′ ), and ι′ : (V ′ , g ′ ) → (V ⊕
V , g ⊕ g ′ ), respectively induce the homomorphisms f : Cℓ(V, g) → Cℓ(V ⊕ V ′ , g ⊕ g ′ ),
′
and f ′ : Cℓ(V ′ , g ′ ) → Cℓ(V ⊕ V ′ , g ⊕ g ′ ). Since the vectors v + 0, and 0 + v′ , are

orthogonal in (V ⊕ V ′ , g ⊕ g ′ ), therefore f (γ(v)) and f ′ (γ ′ (v)) anti-commute. Con-
sequently, f and f ′ satisfy the property required in lemma 4.1 to define an algebra
homomorphism f˚ : Cℓ(V, g)⊗ ˆ Cℓ(V ′ , g ′ ) → Cℓ(V ⊕ V ′ , g ⊕ g ′ ). The action of the ap-
′
plication φ : Cℓ(V ⊕ V ) → Cℓ(V, g)⊗ ˆ Cℓ(V ′ , g ′ ), which maps every Γ(v + v′ ) to
γ(v) ⊗ 1v′ + 1v ⊗ γ (v ) on the generators γ(v) ⊗ 1v′ , and 1v ⊗ γ ′ (v′ ), shows that φ
′ ′
and f˚ are inverse homomorphisms; therefore, φ is an isomorphism. ✓

Theorems on the Structure of Clifford Algebras 89
Up to this point, only real Clifford algebras have been examined, namely, Clifford
algebras for a quadratic space over the reals R. Let us now consider the relationship
between real and complex Clifford algebras. However, before that, let us define the com-
plexification of a vector space. Let V be a real vector space of dimension dim V = n. Its
complexification VC is the space of the elements v + iu, where v, u ∈ V , and i denotes
the imaginary unit. The space VC is a vector space with addition and multiplication
by the complex scalar (a + ib), defined by
(v1 + iu1 ) + (v2 + iu2 ) = (v1 + v2 ) + i(u1 + u2 ),
(a + ib)(v + iu) = (av − bu) + i(bv + au).
The dimension of VC is dimC VC = n over C, and dimR VC = 2n over R. We can see that
VC = C ⊗ V.
Given a symmetric bilinear form g endowing V , its extension gC to VC is defined as

gC (v1 + iu1 , v2 + iu2 ) = g(v1 , v2 ) − g(u1 , u2 ) + i g(v1 , u2 ) + g(u1 , v2 ) .
The next theorem concerns Clifford algebras over complex vector spaces.
Theorem 4.2 ◮ Let (V, g) be a quadratic space over R and let Cℓ(V, g) be its asso-
ciated real Clifford algebra. Consider the complex Clifford algebra Cℓ(VC , gC ) for the
complexified quadratic space (VC , gC ). Then,
Cℓ(VC , gC ) ≃ CℓC (V, g), (4.4)
where CℓC (V, g) = C ⊗ Cℓ(V, g) denotes the complexification of Cℓ(V, g).
Proof: Let us first observe that the complexification of CℓC (V, g) = C ⊗ Cℓ(V, g) is an
algebra endowed with a product given by
(a ⊗ A)(b ⊗ B) = ab ⊗ AB, ∀a, b ∈ C, ∀ A, B ∈ Cℓ(V, g).
Since dimR Cℓ(V, g) = 2dim V , its dimension CℓC (V, g) over R is given by dimR CℓC (V, g) =
2 dimR Cℓ(V, g) = 2 · 2dim V . If γ denotes a Clifford mapping γ : V → Cℓ(V, g), we can
define a mapping Γ : VC → CℓC (V, g) as a linear mapping (in C)
Γ = 1 ⊗ γ,
where 1 denotes the identity of VC . Hence, for a ⊗ v ∈ C ⊗ V = VC , it follows that
Γ(a ⊗ v) = a ⊗ v,
where it is implicit that v = γ(v). Such a mapping Γ is a Clifford mapping. In fact,
2
Γ(1 ⊗ v) = (1 ⊗ v)(1 ⊗ v) = 1 ⊗ g(v, v).
On the other hand, if Cℓ(VC , gC ) is a Clifford algebra for (VC , gC ), it satisfies the universal
property and there exists a homomorphism φ : Cℓ(VC , gC ) → CℓC (V, g). The dimension
of Cℓ(VC , gC ) is dimC Cℓ(VC , gC ) = 2dim V or, equivalently, dimR Cℓ(VC , gC ) = 2 · 2dim V .
Since dim Cℓ(VC , gC ) = dim CℓC (V, g), therefore φ is an isomorphism. ✓
This result shows us that, in order to describe the Clifford algebra structure, it
suffices to understand the real Clifford algebra structure. The complex Clifford algebra
structure follows from the real case when we use this last theorem. In other words,
once the real Clifford algebras structure is known, the complex Clifford algebras are
immediately obtained via complexification, as shown in eqn (4.4). Let us then introduce
some theorems involving real Clifford algebras.
Theorem 4.3 ◮ Let Cℓp,q be a Clifford algebra associated with the quadratic space
Rp,q . Then, the following isomorphisms hold:
Cℓp+1,q+1 ≃ Cℓ1,1 ⊗ Cℓp,q

Cℓq+2,p ≃ Cℓ2,0 ⊗ Cℓp,q (4.5)
Cℓq,p+2 ≃ Cℓ0,2 ⊗ Cℓp,q
where either p > 0 or q > 0, and ⊗ denotes the usual tensor product and not the
alternating tensor product.
Proof: Let U be a two-dimensional space endowed with a symmetric bilinear form
gU such that, with respect to an orthonormal basis {f1 , f2 }, we can write, for u =
u1 f 1 + u2 f 2 ∈ U ,
gU = λ1 (u1 )2 + λ2 (u2 )2 , λ1 = ±1, λ2 = ±1,
where the values of λ1 and λ2 are chosen according to whether U = R2,0 ; R1,1 ; or
R0,2 . For the quadratic space Rp,q , the notation used in eqn (3.35) is adopted. Now,
the linear mapping Γ : Rp,q ⊕ U → Cℓ(U, gU ) ⊗ Cℓp,q can be defined as
Γ(v + u) = f1 f2 ⊗ v + u ⊗ 1, ∀ u ∈ U, ∀ v ∈ Rp,q ,
where f1 = ρ(f1 ); f2 = ρ(f2 ); u = ρ(u); v = γ(v); and where ρ and γ are the Clifford
mappings ρ : U → Cℓ(U, gU ), and γ : Rp,q → Cℓp,q . Let us show that Γ is also a Clifford
mapping:
2
Γ(v + u) = f1 f2 ⊗ v + u ⊗ 1) f1 f2 ⊗ v + u ⊗ 1)
= (f1 f2 )2 ⊗ v2 + (uf1 f2 + f1 f2 u) ⊗ v + u2 ⊗ 1.
The term uf1 f2 + f1 f2 u can be computed if we remember that {f1 , f2 } is orthogonal –

that is, that f1 f2 = f1 ∧ f2 = −f2 f1 – and use eqn (3.26). It then yields
uf1 f2 + f1 f2 u = u(f1 ∧ f2 ) + (f1 ∧ f2 )u = 2u ∧ f1 ∧ f2 = 0,
since u ∈ U is written as u = af1 + bf2 . Using this result and that
(f1 f2 )2 = −(f1 )2 (f2 )2 = −λ1 λ2 ,

2
for Γ(v + u) , we obtain
Theorems on the Structure of Clifford Algebras 91
2
Γ(v + u) = −λ1 λ2 ⊗ g(v, v) + gV (u, u) ⊗ 1

= λ1 (u1 )2 + λ2 (u2 )2

− λ1 λ2 [(v 1 )2 + · · · + (v p )2 − (v p+1 )2 − · · · − (v n )2 ] 1 ⊗ 1.
The mapping Γ is thus a Clifford mapping Γ : Rp,q ⊕ U → Cℓ(W, gW ), where W is a

(n + 2)-dimensional space endowed with a symmetric bilinear functional gW given by
gW (w, w) = λ1 (u1 )2 + λ2 (u2 )2

− λ1 λ2 (v1 )2 + · · · + (v p )2 − (v p+1 )2 − · · · − (v n )2 ,
where w = u1 f1 + u2 f2 + v i ei . Hence, three possibilities arise: (i) if U = R2,0 , then

W = Rq+2,p ; (ii) if U = R1,1 , then W = Rp+1,q+1 ; and (iii) if U = R0,2 , then
W = Rp,q+2 . The isomorphisms follow from the Clifford algebraVuniversality. In the
last paragraph of section 3.4, we proved that dim Cℓ(V, g) ≥ dim (V ) = 2dim V = 2n .
However, we also proved in the beginning of chapter 3 that, from the general definition,
dim Cℓ(V, g) ≤ 2n . Hence, dim Cℓ(V, g) = 2n , so the Clifford algebra is universal. ✓
By combining these isomorphisms, a plethora of other ones can be obtained. For
instance:
Cℓp,p ≃ ⊗p Cℓ1,1 , (4.6)
which follows from repeated applications of the first isomorphism in eqn (4.5). Simi-
larly, we have
Cℓp,q ≃ Cℓp,p ⊗ Cℓ0,q−p (q > p), Cℓp,q ≃ Cℓq,q ⊗ Cℓp−q,0 (p > q). (4.7)
Some particular cases of great interest are given by the expressions
Cℓ0,2 ⊗ Cℓ2,0 ≃ Cℓ0,4 , Cℓ0,4 ⊗ Cℓ4,0 ≃ Cℓ0,8 , (4.8)
and
Cℓ2,2 ≃ Cℓ0,2 ⊗ Cℓ0,2 ≃ Cℓ1,1 ⊗ Cℓ1,1 . (4.9)
From these isomorphisms, it follows in addition that
Cℓ0,4 ⊗ Cℓp,q = Cℓp,q+4 , Cℓ0,8 ⊗ Cℓp,q = Cℓp,q+8 . (4.10)
These isomorphisms are due to Cartan (1908). An isomorphism which does not follow
from this analysis but which has outstanding importance is
Cℓ2,0 ≃ Cℓ1,1 , (4.11)
which can be explicitly constructed. In fact, the elements of Cℓ2,0 are written as
a0 + a1 e1 + a2 e2 + a12 e1 e2 ∈ Cℓ2,0 ,
where (e1 )2 = 1, and (e2 )2 = 1. On the other hand, the elements of Cℓ1,1 are expressed
as
b0 + b1 f1 + b2 f2 + b12 f1 f2 ∈ Cℓ1,1 ,
where (f1 )2 = 1 and (f2 )2 = −1. It is straightforward to realise that the linear mapping
φ : Cℓ2,0 → Cℓ1,1 defined by
φ(1) = 1, φ(e1 ) = f1 , φ(e2 ) = f1 f2 , φ(e1 e2 ) = f2 ,
is an isomorphism. Combining this last isomorphism with the others, we can see that
Cℓp+1,q ≃ Cℓq+1,p . (4.12)
Other combinations arising from these isomorphisms can be made, although it is

not important to do so now. However, these results indicate that, from the Clifford
algebras
Cℓ1,0 , Cℓ0,1 , Cℓ0,2 , Cℓ1,1 ≃ Cℓ2,0 , (4.13)
all other Clifford algebras in arbitrary finite dimensions can be constructed by using
these isomorphisms, thus providing a method for classifying Clifford algebras.
Finally, the following result is of paramount importance:
Theorem 4.4 ◮ Let Cℓp,q be the Clifford algebra associated with the quadratic space
Rp,q and let Cℓ+
p,q be its even subalgebra (eqn (3.46)). Then,
Cℓ+ +
p,q ≃ Cℓq,p−1 ≃ Cℓp,q−1 ≃ Cℓq,p . (4.14)
Proof: Let {ei , fk } (i = 1, . . . , p, k = 1, . . . , q) be an orthonormal basis of V such

that Cℓp,q is generated by 1 and {ei , fk } (a Clifford mapping again implied), where
(ei )2 = 1; (fk )2 = −1; ei ej +ej ei = 0 (i 6= j); V fk fl +fl fk = 0 (k 6= l); and ei fk +fk ei = 0
(i, j = 1, . . . , p, k, l = 1, . . . , q). The space 2 (Rp,q ) consists of the elements {ei ej (i 6=
j), fk fl (k 6= l), ei fk }. Not all of these quantities generate the even subalgebra Cℓ+ p,q .
However, there is redundancy: for example, all the 2-vectors {fk fl } (k 6= l) can be
written in terms of the 2-vectors of type {ei fk }, since (ei fk )(ei fl ) = −(ei )2 fk fl =
−fk fl (k 6= l). Choosing an arbitrary vector, for instance e1 , V it is clear that the set
{e1 em , e1 fk } (m = 2, . . . , p, k = 1, . . . , q) generates the space 2 (Rp,q ) and therefore
generates the even subalgebra Cℓ+ +
p,q . Let us write such generators of Cℓp,q as ξa =
e1 ea+1 for a = 1, . . . , p − 1 and as ζb = e1 fb for b = 1, . . . , q. It is then straightforward
to check that (ξa )2 = −(e1 )2 (ea+1 )2 = −1; (ζb )2 = −(e1 )2 (fb )2 = 1; ξa ξc + ξc ξa = 0
(a 6= c); ζb ζd + ζd ζb = 0 (b 6= d); and ξa ζb + ζb ξa = 0. The quantities {ζb , ξa } (b =
1, . . . , q, a = 1, . . . , p − 1) are then the generators of a Clifford algebra associated with
a quadratic space Rq,p−1 , namely, Cℓ+ p,q ≃ Cℓq,p−1 . The other isomorphisms naturally
follow from the isomorphism in eqn (4.12). ✓
Previously, the grade involution, the reversion, and the conjugation were defined in
the context of the Clifford algebras. Their notations are now redefined: the reversion
by α1 , and the conjugation by α−1 , so that we can unify the two operations in the
notation αǫ (ǫ = ±1). Taking this notation into account, we now assert a generalisation
of the periodicity theorem (Maks, 1989):
The Classification of Clifford Algebras 93
Theorem 4.5 ◮
(Cℓp+1,q+1 , αǫ ) ≃ (Cℓp,q , α−ǫ ) ⊗ (Cℓ1,1 , αǫ ) . (4.15)
Proof: The sets {ei } and {fj } are generators for the algebras Cℓp,q and Cℓ1,1 , respec-
tively. Take {ei ⊗ f1 f2 , 1 ⊗ fj } as a set of generators for Cℓp+1,q+1 ≃ Cℓp,q ⊗ Cℓ1,1 .
Consider now a mapping (α−ǫ ⊗ αǫ )(ei ⊗ f1 f2 ) = α−ǫ (ei ) ⊗ αǫ (f1 f2 ) = ǫ(ei ⊗ f1 f2 ).
In addition, we have (α−ǫ ⊗ αǫ )(1 ⊗ fj ) = α−ǫ (1) ⊗ αǫ (fj ) = ǫ(1 ⊗ fj ). Hence, the
generators are multiplied by ǫ; thus the theorem is proved. ✓
This theorem is very useful for describing of the conformal transformations. For
example, in the case of Minkowski spacetime, as it uses only the Dirac algebra C⊗Cℓ1,3 ,
conformal transformations can be represented by 2 × 2 matrices with entries in Cℓ3,0 .
Thus, operations which act on the Clifford algebra elements of high dimensions can
be led to operations on algebras of low dimensions.
4.2 The Classification of Clifford Algebras
Representations
Definition 4.1 ◮ Let A be a real algebra and let V be a vector space over K =
R, C, H. A linear mapping ρ : A → EndK (V ) satisfying ρ(1A ) = 1V , and ρ(ab) =
ρ(a)ρ(b), ∀a, b ∈ A, is called a K-representation of A. The vector space V is called the
representation space (or carrier space) of A.
The two representations ρ1 : A → EndK (V1 ), and ρ2 : A → EndK (V2 ), are equiva-
lent if there exists a K-isomorphism φ : V1 → V2 satisfying ρ2 (a) = φ ◦ ρ1 (a) ◦ φ−1 ,
∀a ∈ A.
A representation is said to be faithful if ker ρ = {0}. A representation is irreducible
or simple if the only invariant subspaces of ρ(a), ∀a ∈ A, are V and {0}. It is said to
be reducible or semisimple if V = V1 ⊕ V2 , where V1 and V2 are invariant subspaces
under the action of ρ(a), ∀a ∈ A.
Example 4.1 Let us consider the algebra C of complex numbers. As an algebra over C, it has two repre-
sentations: ρ(a + ib) = a + ib, and ρ̄(a + ib) = a − ib. These two C-representations are not equivalent.
Indeed, there does not exist any linear mapping φz : C → C; (a + ib) 7→ φz (a + ib) = z(a + ib), where
z = x + iy, such that ρ̄(a + ib) = zρ(a + ib)z −1 . On the other hand, every C-representation (and also
every H-representation) is an R-representation. We can define two real representations σ : C → M(2, R)
and σ̄ : C → M(2, R) as

a b a −b
σ(a + ib) = , σ̄(a + ib) = ,
−b a b a
respectively. These two representations are equivalent, namely, there exists an isomorphism φ : R2 → R2
such that σ̄(a + ib) = φσ(a + ib)φ−1 . For instance,

1 1 1
φ= √ = φ−1 .
2 1 −1
Those R-representations are irreducible. An example of a reducible R-representation is
 
a b 0 0
−b a 0 0 
ξ(a + ib) =  .
0 0 a −b
0 0 b a
The Clifford Algebra Cℓ0,1

A Clifford algebra associated with the quadratic space R0,1 was introduced in example
3.2. If e is an unit vector such that g(e, e) = −1, an arbitrary element in Cℓ0,1 reads
ψ = a + be ∈ Cℓ0,1 , (4.16)
where e2 = −1. This algebra is isomorphic to the algebra of complex numbers C,
namely, the set of pairs (a, b), a, b ∈ R, with multiplication given by
(a, b)(c, d) = (ac − bd, ad + bc).
The isomorphism is provided by ρ : Cℓ0,1 → C such that ρ(1) = (1, 0) and ρ(e) =
(0, 1) = i. Therefore,
Cℓ0,1 ≃ C. (4.17)
The Clifford Algebra Cℓ1,0

Let us consider the quadratic space R1,0 . Taking the unit vector e, g(e, e) = 1, an
arbitrary element of Cℓ1,0 can be written as
ψ = a + be ∈ Cℓ1,0 , (4.18)
where now e2 = 1. The difference between this case and the previous one is that here
e2 = 1, before, whereas, e2 = −1. In order to make this difference explicit, let us
consider the pair of numbers (a, b), a, b ∈ R, with multiplication defined by
(a, b)(c, d) = (ac + bd, ad + bc). (4.19)
This set might at first sight seem like C, but the difference of sign at the term bd in the
right-hand side of the equation has drastic consequences. In particular, this set is not
a field, although it is a ring. It is also not a division ring, since (1, 1)(1, −1) = (0, 0).
Nonetheless, this set is relevant, as it has some interesting applications in physics –
for example, in the theory of relativity (Fjelstad, 1986; Baylis, 1998; da Rocha and
Vaz, 2006). Let us denote this set by D, whose elements have distinct denominations:
double numbers, perplex numbers, duplex numbers, or Lorentz numbers.
The multiplication defined in eqn (4.19) is appropriate for comparing D to C, but
is not suitable for the classification of Clifford algebras. Instead, let us consider the
pairs (a, b), a, b ∈ R, with multiplication ∗ defined by
(a, b) ∗ (c, d) = (ac, bd). (4.20)
This algebra consists of the direct sum R ⊕ R and is isomorphic to the algebra of 2 × 2
diagonal matrices. This isomorphism is given by

a0
φ(a, b) = . (4.21)
0b
The algebras D and R ⊕ R are isomorphic. The isomorphism ϕ : D → R ⊕ R reads
ϕ(a, b) = (a + b, a − b). (4.22)
It is trivial to verify that indeed ϕ((a, b)(c, d)) = ϕ(a, b) ∗ ϕ(c, d).
Now going back to the Clifford algebra Cℓ1,0 , this algebra is clearly isomorphic to
D: the isomorphism is ρ(1) = (1, 0), and ρ(e) = (0, 1). Hence, Cℓ1,0 is isomorphic to
R ⊕ R:
Cℓ1,0 ≃ R ⊕ R. (4.23)
The Clifford algebra Cℓ0,2

Let us consider the quadratic space R0,2 and an orthonormal basis {e1 , e2 },
g(e1 , e1 ) = g(e2 , e2 ) = −1, g(e1 , e2 ) = g(e2 , e1 ) = 0.
An arbitrary element of Cℓ0,2 reads
ψ = a + be1 + ce2 + de1 e2 ∈ Cℓ0,2 , (4.24)
where a, b, c, d ∈ R, and
(e1 )2 = (e2 )2 = −1, e1 e2 + e2 e1 = 0. (4.25)
These relations imply that (e1 e2 )2 = −1.

It is clear to see that Cℓ0,2 is isomorphic to the quaternion algebra H. This isomor-
phism is given by
ρ(1) = 1, ρ(e1 ) = i, ρ(e2 ) = j, ρ(e1 e2 ) = k, (4.26)
where i, j, and k are the quaternion units i2 = j 2 = k 2 = −1; ij = −ji = k;

jk = −kj = i; and ki = −ik = j. Hence, it explicitly follows that
Cℓ0,2 ≃ H. (4.27)
The Clifford algebra Cℓ2,0 ≃ Cℓ1,1

The Clifford algebras associated with the quadratic spaces R2,0 and R1,1 have been
shown to be isomorphic: Cℓ2,0 ≃ Cℓ1,1 . Hence, it suffices to consider just one of these
spaces, for instance, R2,0 . Given an orthonormal basis {e1 , e2 },
g(e1 , e1 ) = g(e2 , e2 ) = 1, g(e1 , e2 ) = g(e2 , e1 ) = 0.
An arbitrary element of Cℓ2,0 can be written as
ψ = a + be1 + ce2 + de1 e2 ∈ Cℓ0,2 , (4.28)
where a, b, c, d ∈ R, and
(e1 )2 = (e2 )2 = 1, e1 e2 + e2 e1 = 0. (4.29)
These relations further imply that
(e1 e2 )2 = −1. (4.30)

Let M(2, R) be the algebra of the real 2 × 2 matrices. The set

10 1 0 01 0 1
, , ,
01 0 −1 10 −1 0
generates M(2, R) and, furthermore,

2 2 2
1 0 10 01 0 1 −1 0
= = , = . (4.31)
0 −1 01 10 −1 0 0 −1
Comparing eqns (4.29) and (4.30) to (4.31), ρ can be defined as the linear mapping

10 01
ρ(1) = , ρ(e2 ) = ,
01 10
(4.32)
1 0 0 1
ρ(e1 ) = , ρ(e1 e2 ) = .
0 −1 −1 0
As ρ is an isomorphism,
Cℓ2,0 ≃ Cℓ1,1 ≃ M(2, R). (4.33)
Classifying Arbitrary Clifford Algebras

Once the isomorphisms (i) Cℓ0,1 ≃ C, (ii) Cℓ1,0 ≃ R ⊕ R, (iii) Cℓ0,2 ≃ H, and
(iv) Cℓ2,0 ≃ Cℓ1,1 ≃ M(2, R) have been established, we can, by using the isomorphisms
in the previous section, proceed to the classification of arbitrary Clifford algebras. For
instance, by using eqn (4.6) and
M(m, R) ⊗ M(n, R) ≃ M(mn, R), (4.34)
we obtain
Cℓp,p ≃ M(2p , R). (4.35)
This result and eqn (4.9) allow us to conclude that
H ⊗ H ≃ M(4, R). (4.36)
In addition, eqn (4.8) can be used to obtain
Cℓ0,4 ≃ H ⊗ M(2, R) ≃ M(2, H) ≃ Cℓ4,0 , (4.37)
and
Cℓ0,8 ≃ M(2, H) ⊗ M(2, H) ≃ M(2, R) ⊗ H ⊗ H ⊗ M(2, R)
(4.38)
≃ M(2, R) ⊗ M(4, R) ⊗ M(2, R) ≃ M(16, R).
This result, together with eqn (4.10), implies that
Cℓp,q+8 ≃ Cℓp,q ⊗ M(16, R). (4.39)
Equation (4.39) has an important consequence: we only need to explicitly obtain

the classification of the Clifford algebras up to dim V = p + q = 8 since, for dimensions
higher than that, the isomorphism Cℓp,q+8 ≃ Cℓp,q ⊗M(16, R) (called the periodicity
theorem) can be used to immediately obtain any other isomorphism. In this way, by
using all the previous results in this section, we obtain
Cℓ0,0 ≃ R,
Cℓ0,1 ≃ C,
Cℓ1,0 ≃ R ⊕ R,
Cℓ0,2 ≃ H,
Cℓ2,0 ≃ M(2, R),
Cℓ1,1 ≃ Cℓ2,0 ≃ M(2, R),
Cℓ0,3 ≃ Cℓ0,2 ⊗ Cℓ1,0 ≃ H ⊕ H,

Cℓ3,0 ≃ Cℓ2,0 ⊗ Cℓ0,1 ≃ M(2, C),
Cℓ1,2 ≃ Cℓ1,1 ⊗ Cℓ0,1 ≃ M(2, R) ⊗ C ≃ M(2, C),
Cℓ2,1 ≃ Cℓ1,1 ⊗ Cℓ1,0 ≃ M(2, R) ⊗ R ⊕ R ≃ M(2, R ⊕ R),
Cℓ0,4 ≃ Cℓ0,2 ⊗ Cℓ2,0 ≃ H ⊗ M(2, R) ≃ M(2, H),

Cℓ4,0 ≃ Cℓ2,0 ⊗ Cℓ0,2 ≃ M(2, R) ⊗ H ≃ M(2, H),
Cℓ1,3 ≃ Cℓ1,1 ⊗ Cℓ0,2 ≃ M(2, R) ⊗ H ≃ M(2, H),
Cℓ3,1 ≃ Cℓ1,1 ⊗ Cℓ2,0 ≃ M(2, R) ⊗ M(2, R) ≃ M(4, R),
Cℓ2,2 ≃ Cℓ1,1 ⊗ Cℓ1,1 ≃ M(2, R) ⊗ M(2, R) ≃ M(4, R),
Cℓ5,0 ≃ Cℓ2,0 ⊗ Cℓ0,3 ≃ M(2, R) ⊗ H ⊕ H ≃ M(2, H ⊕ H),

Cℓ0,5 ≃ Cℓ0,2 ⊗ Cℓ3,0 ≃ H ⊗ M(2, C) ≃ H ⊗ C ⊗ M(2, R)
≃ M(2, C) ⊗ M(2, R) ≃ C ⊗ M(2, R) ⊗ M(2, R)
≃ C ⊗ M(4, R) ≃ M(4, C),
Cℓ4,1 ≃ Cℓ1,1 ⊗ Cℓ3,0 ≃ M(2, R) ⊗ M(2, C) ≃ M(4, C),
Cℓ1,4 ≃ Cℓ1,1 ⊗ Cℓ0,3 ≃ M(2, R) ⊗ (H ⊕ H) ≃ M(2, H ⊕ H),
Cℓ3,2 ≃ Cℓ1,1 ⊗ Cℓ1,1 ⊗ Cℓ1,0 ≃ M(2, R) ⊗ M(2, R) ⊗ (R ⊕ R)
≃ M(4, R ⊕ R),
Cℓ2,3 ≃ Cℓ1,1 ⊗ Cℓ1,1 ⊗ Cℓ0,1 ≃ M(2, R) ⊗ M(2, R) ⊗ C ≃ M(4, C),
Cℓ6,0 ≃ Cℓ2,0 ⊗ Cℓ0,4 ≃ Cℓ2,0 ⊗ Cℓ0,2 ⊗ Cℓ2,0

≃ M(2, R) ⊗ H ⊗ M(2, R) ≃ M(4, H),
Cℓ0,6 ≃ Cℓ0,2 ⊗ Cℓ4,0 ≃ Cℓ0,2 ⊗ Cℓ2,0 ⊗ Cℓ0,2 ≃ H ⊗ M(2, R) ⊗ H
≃ M(2, H) ⊗ H ≃ M(2, R) ⊗ H ⊗ H ≃ M(2, R) ⊗ M(4, R)
≃ M(8, R),
Cℓ5,1 ≃ Cℓ1,1 ⊗ Cℓ4,0 ≃ M(2, R) ⊗ M(2, H) ≃ M(4, H),
Cℓ1,5 ≃ Cℓ1,1 ⊗ Cℓ0,4 ≃ M(2, R) ⊗ M(2, H) ≃ M(4, H),
Cℓ4,2 ≃ Cℓ1,1 ⊗ Cℓ1,1 ⊗ Cℓ2,0 ≃ M(2, R) ⊗ M(2, R) ⊗ M(2, R) ≃ M(8, R),

Cℓ2,4 ≃ Cℓ1,1 ⊗ Cℓ1,1 ⊗ Cℓ0,2 ≃ M(2, R) ⊗ M(2, R) ⊗ H ≃ M(4, H),
Cℓ3,3 ≃ Cℓ1,1 ⊗ Cℓ1,1 ⊗ Cℓ1,1 ≃ M(2, R) ⊗ M(2, R) ⊗ M(2, R) ≃ M(8, R),
Cℓ7,0 ≃ Cℓ2,0 ⊗ Cℓ0,5 ≃ M(2, R) ⊗ M(4, C) ≃ M(8, C),

Cℓ0,7 ≃ Cℓ0,2 ⊗ Cℓ5,0 ≃ H ⊗ M(2, H ⊕ H)

M(4, R) M(4, R) M(4, R) M(4, R)
≃ ⊕
M(4, R) M(4, R) M(4, R) M(4, R)
≃ M(8, R) ⊕ M(8, R) ≃ M(8, R ⊕ R),
Cℓ6,1 ≃ Cℓ1,1 ⊗ Cℓ5,0 ≃ M(2, R) ⊗ M(2, H ⊕ H) ≃ M(4, H ⊕ H),
Cℓ1,6 ≃ Cℓ1,1 ⊗ Cℓ0,5 ≃ M(2, R) ⊗ M(4, C) ≃ M(8, C),
Cℓ5,2 ≃ Cℓ1,1 ⊗ Cℓ1,1 ⊗ Cℓ3,0 ≃ M(2, R) ⊗ M(2, R) ⊗ M(2, C) ≃ M(8, C),
Cℓ2,5 ≃ Cℓ1,1 ⊗ Cℓ1,1 ⊗ Cℓ0,3 ≃ M(2, R) ⊗ M(2, R) ⊗ (H ⊕ H) ≃ M(4, H ⊕ H),
Cℓ4,3 ≃ Cℓ1,1 ⊗ Cℓ1,1 ⊗ Cℓ1,1 ⊗ Cℓ1,0
≃ M(2, R) ⊗ M(2, R) ⊗ M(2, R) ⊗ (R ⊕ R) ≃ M(8, R ⊕ R),
Cℓ3,4 ≃ Cℓ1,1 ⊗ Cℓ1,1 ⊗ Cℓ1,1 ⊗ Cℓ0,1
≃ M(2, R) ⊗ M(2, R) ⊗ M(2, R) ⊗ C ≃ M(8, C),
where the following isomorphisms hold:
Cℓ2,0 ≃ Cℓ1,1 ≃ M(2, R),

Cℓ3,0 ≃ Cℓ1,2 ≃ M(2, C),
Cℓ0,4 ≃ Cℓ4,0 ≃ Cℓ1,3 ≃ M(2, H),
Cℓ3,1 ≃ Cℓ2,2 ≃ M(4, R),
Cℓ5,0 ≃ Cℓ1,4 ≃ M(2, H ⊕ H),
Cℓ0,5 ≃ Cℓ4,1 ≃ Cℓ2,3 ≃ M(4, C),
Cℓ6,0 ≃ Cℓ5,1 ≃ Cℓ1,5 ≃ Cℓ2,4 ≃ M(4, H),
Cℓ0,6 ≃ Cℓ4,2 ≃ Cℓ3,3 ≃ M(8, R),
Cℓ7,0 ≃ Cℓ1,6 ≃ Cℓ5,2 ≃ Cℓ3,4 ≃ M(8, C),
Cℓ0,7 ≃ Cℓ4,3 ≃ M(8, R ⊕ R),
Cℓ6,1 ≃ Cℓ2,5 ≃ M(4, H ⊕ H).
Furthermore, by supposing that p > q and taking p − q = 8k + r with r < 8, we

can use eqns (4.5) and (4.39) to obtain:
Cℓp,q ≃ Cℓq,q ⊗ Cℓp−q,0 ≃ Cℓq,q ⊗ Cℓ8k+r,0

≃ Cℓq,q ⊗ Cℓ8(k−1)+r+6+2,0
≃ Cℓq,q ⊗ Cℓ2,0 ⊗ Cℓ0,8(k−1)+r+6
≃ Cℓq,q ⊗ Cℓ2,0 ⊗ Cℓ0,8(k−1) ⊗ Cℓ0,r+6
≃ M(2q , R) ⊗ M(2, R) ⊗ M(16k−1 , R) ⊗ Cℓ0,r+6
≃ M(2q+4(k−1)+1 , R) ⊗ Cℓ0,r+6
≃ M(2q+4k−3 , R) ⊗ Cℓ0,r+6 .
Since
Cℓ0,r+6 ≃ Cℓ0,2 ⊗ Cℓ2,0 ⊗ Cℓ0,2 ⊗ Cℓr,0

≃ M(4, R) ⊗ M(2, R) ⊗ Cℓr,0 ≃ M(8, R) ⊗ Cℓr,0 ,
it reads
Cℓp,q ≃ M(2q+4k−3 , R) ⊗ M(8, R) ⊗ Cℓr,0 ≃ M(2q+4k , R) ⊗ Cℓr,0 .
Now, if q > p, and q − p = 8k + r, we obtain
Cℓp,q ≃ Cℓp,p ⊗ Cℓ0,q−p = Cℓp,p ⊗ Cℓ0,8k+r ≃ Cℓp,p ⊗ Cℓ0,8k ⊗ Cℓ0,r

≃ M(2p , R) ⊗ M(24k , R) ⊗ Cℓ0,r ≃ M(2p+4k , R) ⊗ Cℓ0,r .
Hence, we can see that the Clifford algebra is determined by r = p − q mod 8. The
possibilities are analysed in what follows:
• r = 0, and p ≥ q (p − q mod 8 = 0):
Cℓp,q ≃ M(2q+4k , R) ⊗ Cℓ0,0 ≃ M(2q+4k , R)
• r = 0, and p < q (p − q mod 8 = 0):
Cℓp,q ≃ M(2p+4k , R) ⊗ Cℓ0,0 ≃ M(2p+4k , R)
• r = 1, and p > q (p − q mod 8 = 1):
Cℓp,q ≃ M(2q+4k , R) ⊗ Cℓ1,0 ≃ M(2q+4k , R) ⊗ (R ⊕ R)

≃ M(2q+4k , R) ⊕ M(2q+4k , R)
• r = 1, and p < q (p − q mod 8 = 7):
Cℓp,q ≃ M(2p+4k , R) ⊗ Cℓ0,1 ≃ M(2p+4k , R) ⊗ C ≃ M(2p+4k , C)

h i n
In these cases, q + 4k = 2q+8k+r
2 = p+q
2 = 2 , where [s] denotes the integer
part of s.
• r = 2, and p > q (p − q mod 8 = 2):
Cℓp,q ≃ M(2q+4k , R) ⊗ Cℓ2,0 ≃ M(2q+4k , R) ⊗ M(2, R) ≃ M(2q+4k+1 , R)
• r = 2, and p < q (p − q mod 8 = 6):
Cℓp,q ≃ M(2p+4k , R) ⊗ Cℓ0,2 ≃ M(2q+4k , R) ⊗ H ≃ M(2q+4k , H)
• r = 3, and p > q (p − q mod 8 = 3):
Cℓp,q ≃ M(2q+4k , R) ⊗ Cℓ3,0 ≃ M(2q+4k , R) ⊗ M(2, C) ≃ M(2q+4k+1 , C)

• r = 3, and p < q (p − q mod 8 = 5):

Cℓp,q ≃ M(2q+4k , R) ⊗ Cℓ0,3 ≃ M(2q+4k , R) ⊗ (H ⊕ H)
≃ M(2q+4k , H) ⊕ M(2q+4k , H)
h i
In the cases where r = 2, 3, if p > q then q + 4k + 1 = 2q+8k+r
2 = n2 ; and, if
h i
p < q, then p + 4k = 2p+8k+r
2 − 1 = n2 − 1
• r = 4, and p > q (p − q mod 8 = 4):
Cℓp,q ≃ M(2q+4k , R) ⊗ Cℓ4,0 ≃ M(2q+4k , R) ⊗ M(2, H) ≃ M(2q+4k+1 , H)
• r = 4, and p < q (p − q mod 8 = 4):
Cℓp,q ≃ M(2p+4k , R) ⊗ Cℓ0,4 ≃ M(2q+4k , R) ⊗ M(2, H) ≃ M(2q+4k+1 , H)
• r = 5, and p > q (p − q mod 8 = 5):
Cℓp,q ≃ M(2q+4k , R) ⊗ Cℓ5,0 ≃ M(2q+4k , R) ⊗ M(2, R) ⊗ (H ⊕ H)
≃ M(2q+4k , R) ⊗ (M(2, H) ⊗ M(2, H))
≃ M(2q+4k+1 , H) ⊗ M(2q+4k+1 , H)
• r = 5, and p < q (p − q mod 8 = 3):
Cℓp,q ≃ M(2q+4k , R) ⊗ Cℓ0,5 ≃ M(2q+4k , R) ⊗ H ⊗ M(2, C)
≃ M(2q+4k+2 , C) .
n
In the cases where r = 4, 5, if p > q or r= 4, then q + 4k + 1 = 2
− 1, whilst,
if p < q and r = 5, then q + 4k + 2 = n2 .
• r = 6, and p > q (p − q mod 8 = 6):
Cℓp,q ≃ M(2q+4k , R) ⊗ Cℓ6,0 ≃ M(2q+4k , R) ⊗ M(4, H) ≃ M(2q+4k+2 , H)
• r = 6, and p < q (p − q mod 8 = 2):
Cℓp,q ≃ M(2p+4k , R) ⊗ Cℓ0,6 ≃ M(2q+4k , R) ⊗ M(8, R) ≃ M(2q+4k+3 , R)
• r = 7, and p > q (p − q mod 8 = 7):
Cℓp,q ≃ M(2q+4k , R) ⊗ Cℓ7,0 ≃ M(2q+4k , R) ⊗ M(4, R) ⊗ H ⊗ C
≃ M(2q+4k , R) ⊗ M(8, C) ≃ M(2q+4k+3 , C)
• r = 7, and p < q (p − q mod 8 = 1):
Cℓp,q ≃ M(2q+4k , R) ⊗ Cℓ0,7
≃ M(2q+4k , R) ⊗ H ⊗ H ⊗ M(2, R) ⊗ (R ⊕ R)
≃ M(2q+4k , R) ⊗ M(8, R) ⊗ (R ⊕ R)
≃ M(2q+4k+3 , R) ⊕ M(2q+4k+3 , R)

In the cases where r = 6 or 7, if p < q, then q + 4k + 3 = n2 ; and, if p > q and
r = 6, then q + 4k + 2 = n2 − 1.
We can organise the algebras of dimension n < 8 according to p − q, obtaining:
p−q = 0: Cℓ0,0 , Cℓ1,1 , Cℓ2,2 , Cℓ3,3 M(2[n/2] , R)

2
p−q = 1: Cℓ1,0 , Cℓ2,1 , Cℓ3,2 , Cℓ4,3 , Cℓ0,7 M(2[n/2] , R)⊕
p−q = 2: Cℓ2,0 , Cℓ3,1 , Cℓ4,2 , Cℓ0,6 M(2[n/2] , R)
p−q = 3: Cℓ3,0 , Cℓ4,1 , Cℓ5,2 , Cℓ0,5 , Cℓ1,6 M(2[n/2] , C)
p−q = 4: Cℓ4,0 , Cℓ5,1 , Cℓ0,4 , Cℓ1,5 M(2[n/2]−1 , H)
2
p−q = 5: Cℓ5,0 , Cℓ6,1 , Cℓ0,3 , Cℓ1,4 , Cℓ2,5 M(2[n/2]−1 , R)⊕
p−q = 6: Cℓ6,0 , Cℓ0,2 , Cℓ1,3 , Cℓ2,4 M(2[n/2]−1 , H)
p−q = 7: Cℓ7,0 , Cℓ0,1 , Cℓ1,2 , Cℓ2,3 , Cℓ3,4 M(2[n/2] , C)
2
where A⊕ denotes A ⊕ A. Table 4.1 shows the Clifford algebra classification obtained
from these isomorphisms.
Let us provide some examples of how this table can be used. First, take the Clifford
algebra Cℓ3,0 , where p − q = 3 and which is isomorphic to M(2[n/2] , C). Since n =
p+q = 3, it follows that [n/2] = [3/2] = 1, and the related matrix algebra is M(21 , C).
Hence, Cℓ3,0 ≃ M(2, C). Let us now consider the algebra Cℓ0,2 . In this case p −
q = −2 = 6 mod 8 and the corresponding matrix algebra is M(2[n/2]−1 , H). Since
n = p + q = 2, and [n/2] = [2/2] = 1, therefore M(21−1 , H) = M(1, H) = H, namely,
Cℓ0,2 ≃ H, as seen previously. The same reasoning leads, for instance, to the conclusion
that Cℓ0,3 ≃ H ⊕ H; Cℓ1,3 ≃ M(2, H); Cℓ3,1 ≃ M(4, R); Cℓ4,1 ≃ M(4, C); and so on.
With respect to the complex case, the classification can be obtained from Cℓ(VC , gC ) ≃
C ⊗ Cℓ(V, g). The complex Clifford algebra depends only on the parity of n = p + q.
We therefore denote C ⊗ Cℓp,q = CℓC (n). If n is even, p − q = 0, 2, 4, 6:
Table 4.1 Real Clifford Algebra Classification, Where p + q = n, and [n/2] Denotes
the Integer Part of n/2
p−q
0 1 2 3
mod 8
M(2[n/2] , R)
Cℓp,q M(2 [n/2]
, R) ⊕ M(2[n/2] , R) M(2[n/2] , C)
M(2[n/2] , R)
p−q
4 5 6 7
mod 8
M(2[n/2]−1 , H)
Cℓp,q M(2 [n/2]−1
, H) ⊕ M(2[n/2]−1 , H) M(2[n/2] , C)
M(2[n/2]−1 , H)
p−q = 0, C ⊗ M(2n , R) ≃ M(2n , C),

p−q = 2, C ⊗ M(2n , R) ≃ M(2n , C),
p−q = 4, C ⊗ M(2n−1 , H) ≃ M(2n , C),
p−q = 6, C ⊗ M(2n−1 , H) ≃ M(2n , C).
If n is odd, p − q = 1, 3, 5, 7:
p−q = 1, C ⊗ (M(2n , R) ⊕ M(2n , R)) ≃ M(2n , C) ⊕ M(2n , C),

p−q = 3, C ⊗ M(2n , C) ≃ M(2n , C) ⊕ M(2n , C),
p−q = 5, C ⊗ (M(2n−1 , H) ⊕ M(2n−1 , H)) ≃ M(2n , C) ⊕ M(2n , C),
p−q = 7, C ⊗ M(2n , C) ≃ M(2n , C) ⊕ M(2n , C).
Therefore, for the complex Clifford algebras CℓC (n) it follows the classification given
in table 4.2.
Although this table provides the isomorphisms between Clifford and matrix alge-
bras (or, in some cases, the direct sums of some algebras), it does not explicitly say
how to write such isomorphisms. In other words, the table shows that Cℓ2,0 ≃ M(2, R)
but does not supply a way to explicitly obtain the isomorphism given in eqn (4.32).
In those cases involving, for instance, low-dimensional algebras like Cℓ2,0 , it is not
that complicated to find an isomorphism; however, as the number of dimensions of
the space increases, the procedure becomes non-trivial. The main aim here is to find
and exhibit a matrix representation for a Clifford algebra. It is not only important
from the ‘theoretical’ point of view but also from the ‘practical’ viewpoint. From the
theoretical point of view, the concept of the spinor is associated with such representa-
tions (we will discuss spinors in chapter 6). From the practical point of view, it is often
better (computationally) to use matrices than to consider the abstract Clifford algebra
generators directly. Later on, we will discuss how to explicitly obtain a representation
of a Clifford algebra.
4.3 Idempotents and Representations
The Regular Representation

Let A be an algebra, where there is involved naturally an underlying vector space
structure. We can further consider the set of endomorphisms End(A) in order to
construct the representations of the algebra A. We can define a representation L :
A → End(A) as
L(a)b = ab, ∀ b ∈ A. (4.40)
Table 4.2 Classification of the Complex Clifford Al-

gebras
n
CℓC (2k) = M(2k , C)
even
n
CℓC (2k + 1) = M(2k , C) ⊕ M(2k , C)
odd
Idempotents and Representations 103
It is straightforward to see that L is indeed a representation: L(1) = 1, and L(ab) =

L(a)L(b). We denominate L a regular representation. A regular representation is faith-
ful. Indeed, if L(a)c = L(b)c, then L(a − b)c = (a − b)c = 0, and when c = 1, it follows
that a = b. In other words, ker L = {0}. We can also define R : A → End(A) as
R(a)b = ba, ∀ b ∈ A. (4.41)
In this case, R(1) = 1 but R(ab) = R(b)R(a), namely, R is a representation of the

opposed algebra Aop , where Aop is an algebra with multiplication opposed with respect
to A, namely, mAop (a, b) = mA (b, a) = ba. Therefore, R(A) = Aop . The sets L(A)
and R(A) are subalgebras of End(A) that commute L(a)R(b)c = L(a)cb = acb =
(L(a)c)b = R(b)L(a)c, since A is associative.
Now let us suppose that there exist subspaces B1 and B2 of A that are invariant
with respect to the regular representation, that is, L(a)(B1 ) ⊂ B1 , and L(a)(B2 ) ⊂ B2 ,
∀ a ∈ A. Hence, it is possible to express L = L1 ⊕ L2 , where L1 : A → End(B1 ),
and L2 : A → End(B2 ). If there exist other invariant subspaces L1 and L2 , the same
procedure is employed until a space S and a representation L : A → End(S) are
obtained, such that the unique invariant subspaces are S and {0}. This representation
is then irreducible. The space S satisfies L(a)(S) ⊂ S, ∀ a ∈ A, that is, ∀ a ∈ A
and ∀ x ∈ S, it follows that ax ∈ S. By definition, S is a left ideal with respect to
the algebra A (see chapter 2, section 2.5), and indeed a minimal left ideal, where
by minimal it is meant that S does not contain any non-trivial subideal (the unique
subideals of S are either S or {0}).
To summarise: the representation space associated with an irreducible regular rep-
resentation is a left minimal ideal of the algebra.
Idempotents
An element f ∈ A is an idempotent if f 2 = f . If the algebra A is a division algebra,1
then the unique idempotent is the identity 1. Indeed, if f 2 = f , where f 6= 0 and A is
a division algebra, then f = f −1 f 2 = f −1 f = 1. However, most Clifford algebras are
not division algebras.
Two idempotents, f1 and f2 , are called orthogonal if f1 f2 = f2 f1 = 0. An idempo-
tent f is said to be primitive if it cannot be written as the sum of other two orthogonal
idempotents, that is, f 6= f1 + f2 , where (f1 )2 = f1 ; (f2 )2 = f2 ; and f1 f2 = f2 f1 = 0
(Lounesto and Wene, 1987).
Simple Algebras
Looking at table 4.1 we can see that every Clifford algebra can be expressed either by
K ⊗ M(N, R) or by [K ⊗ M(N, R)] ⊕ [K ⊗ M(N, R)], for K = R, C, H, and for some
1 A division algebra is defined as an algebra where every non-zero element has an inverse. Equiva-
lently, an algebra A is a division algebra when, if ab = 0 (or if ba = 0), ∀ b ∈ A, then a = 0. A theorem
attributed to Frobenius asserts that the unique, real, associative finite-dimensional algebras are R, C,
and H (Schafer, 1954). The octonion algebra (or Cayley algebra) O is also a division algebra, but it is
not associative. For the Clifford algebraic formulation of octonions see Lounesto (2001a, de Andrade
and Toppan (1999, da Rocha and Vaz (2007); an advanced discussion of this topics, see the articles
by da Rocha et al. (2012, da Rocha and Traesel (2012, da Rocha and Vaz Jr (2006).
N = 0, 1, . . . . In addition, an algebra that can be represented by K ⊗ M(N, R) is a

simple algebra. An algebra A is said to be simple if the unique bilateral ideals of A
are A and {0}. In order to check that a matrix algebra A = K ⊗ M(N, R) is a simple
algebra, let us consider an element x in an ideal I ∈ A. According to exercise 4 in
chapter 1, the matrices {EAB } defined by (EAB )CD = δAC δBD and satisfying EAB ECD =
P M(N, R). Since
δBC EAD (A, B, C, D = 1, . . . , N ) form a basis for the space of matrices
I ⊂ A = K ⊗ M(N, R), therefore x ∈ I can be written as x = AB xAB EAB , where
xAB ∈ K. If x 6= 0, then at least one of the components xAB is null. Now we can see
that
X X
ECA xEBC = xDE ECA EDE EBC
C CDE
X X
= xDE δAD δBE ECC = xAB ECC = xAB 1,
CDE C
P
where 1 = C ECC denotes the identity matrix. Since xAB ∈ K – which is a division
algebra – then there exists x−1
AB
and
X
x−1
AB
ECA xEBC = 1.
C
This last equation means that 1 ∈ AxA. However, AxA ∈ AIA ⊂ I, and then 1 ∈ I.
Taking x = 1, it follows that AA = A ⊂ I. Since A ⊂ I and I ⊂ A, therefore I = A.
The other possibility is the case where x 6= 0 and therefore I = {0}. Hence, an algebra
A = K ⊗ M(N, R) is a simple algebra. It is possible to show – indeed, it will be shown
in the next section – that the reciprocal holds, namely, if an algebra is simple, then it
can be written (not uniquely) as K ⊗ M(N, R).
Therefore, every Clifford algebra is either a simple algebra or the direct sum of
simple algebras. An algebra that is the sum of simple algebras is said to be a semisimple
algebra.
Idempotents and Simple Algebras

Consider a set of N primitive idempotents {f1 , . . . , fN } which are mutually orthogonal:
fA fB = δAB fA . Let us denote by AAB the set AAB = fA AfB , (A, B = 1, . . . , N ). For
those sets, we have
AAB ACD = fA AfB fC AfD = δBC fA AfB AfD .
In addition, the set AfB A is a bilateral ideal of A. Since the algebra A is simple,
therefore AfB A equals either A or {0}. Since fB 6= 0, therefore AfB A = A and thus
AAB ACD = δBC AAD . (4.42)
On the other hand, there exists idempotents fA ∈ AAA (A = 1, . . . , N ). Let us choose

one of those idempotents, for example, f1 . Because of eqn (4.42), it follows that f1 ∈
A11 = A1A AA1 for any value of A. Hence, there exist E 1A ∈ A1A , and E A1 ∈ AA1 , such
that f1 = E 1A E A1 . Now, choose E 1A and E A1 such that
fB E A1 = δAB E A1 , E 1A fB = δAB E 1A . (4.43)
Let us define the quantities E AB as
E AB = E A1 E 1B , (4.44)
which yields
fC E AB = δAC E AB , E AB fC = δBC E AB . (4.45)
Now let us take into account the product E AB E CD . With these definitions it reads
E AB E CD = E A1 E 1B E C 1 E 1D = δBC E A1 E 1B E B1 E 1D
= δBC E A1 f1 E 1D = δBC E A1 E 1D ,
namely,
E AB E CD = δBC E AD . (4.46)
This equation shows that the quantities {E AB } (A, B = 1, . . . , N ) comprise a basis for
the space of N × N matrices. In addition, it shows that E AA is an idempotent. The
identity 1A de A is given by X
1A = E AA , (4.47)
A
P P
since 1A E BC = A E AA E BC = A δAB E AC = E BC .
The set AAA = fA AfA is an ideal of A. For xA ∈ AAA , we have xA fA = fA xA = xA ,
that is, the idempotent fA is the unit of AAA . This idempotent fA is the unique
idempotent of AAA if and only if fA is a primitive. This assertion follows from the
following: (i) if fA is not a primitive, then fA = gA + hA , where hA gA = gA hA = 0;
therefore, fA gA = gA fA = gA , and fA hA = hA fA = hA , namely, gA , hA ∈ AAA , which
negates the hypothesis that fA is the unique idempotent of AAA ; (ii) if there exists
another idempotent gA ∈ AAA besides fA , then fA − gA is another idempotent, since,
if gA ∈ AAA , then gA fA = fA gA = gA . This result implies that (fA − gA )(fA − gA ) =
fA − fA gA − gA fA + gA = fA − gA . In addition, the idempotents gA and fA − gA
are orthogonal: gA (fA − gA ) = gA − gA = 0, and (fA − gA )gA = gA − gA = 0; since
fA = (fA −gA )+gA , the idempotent fA is not primitive. Moreover, since E AA ⊂ fA AfA ,
therefore the idempotent fA is primitive, yielding
E AA = fA . (4.48)
Moreover, AAA = fA AfA is a division algebra with identity fA , and the algebras
AAA (A = 1, . . . , N ) are isomorphic. Let us first show that fA AfA is a division algebra.
Let IA = AfA be a left ideal of A. Since fA is primitive, this ideal is minimal; thus,
the unique primitive subideals are IA and {0}. Let now JA be a (non-null) left ideal
of AAA . Obviously, JA ⊂ AAA , and AJA ⊂ AfA AfA = fA AIA ⊂ IA , that is, AJA ⊂
IA . However, since IA is minimal and JA is non-null, the unique possibility is that
AJA = IA . On the other hand, AAA = fA AfA = fA IA = fA AJA fA AfA JA ⊂ JA . Since
JA ⊂ AAA and we just proved that AAA ⊂ JA , it follows that JA = AAA . We then
conclude that the unique left ideal of AAA is either AAA itself or {0}. Consider now a
non-null element z ∈ AAA . The set AAA z is a left ideal of AAA . However, since AAA
does not contain non-trivial subideals, and z 6= 0, therefore AAA = AAA z. This result
means that there exists w, w′ ∈ AAA such that wz = w′ . Hence, there exists a non-null
z ′ ∈ AAA such that z ′ z = fA , since fA is the identity in AAA . Similarly, there exists
z ′′ ∈ AAA such that z ′′ z ′ = fA . Now z ′′ = z ′′ z ′ z = z, namely, z ′ z = zz ′ = fA ; thus,
AAA is a unital division algebra fA . In order to show that AAA ≃ ABB , given xA ∈ AAA ,
and xB ∈ ABB , we define φAB : AAA → ABB as
φAB (xA ) = E BA xA E AB . (4.49)
This mapping is linear and by using eqn (4.46) and given the fact that fA = E AA is
the identity in AAA , we can see that it satisfies
φAB (xA yA ) = E BA xA yA E AB = E BA xA fA yA E AB
= E BA xA E AB E BA yA E AB = φAB (xA )φAB (yA ).
Moreover, φ−1
AB = φBA . Therefore, the division algebras AAA (A = 1, . . . , N ) are iso-
morphic.
Given, for instance, x1 ∈ A11 , and xA = E A1 x1 E 1A , we define x ∈ K as
X X
x= xA = E A1 x1 E 1A . (4.50)
A A
This expression defines a linear mapping A11 → K whose inverse is provided by

E 11 xE 11 = x1 . Hence, it implies that K ≃ A11 ≃ AAA (A = 1, . . . , N ). Moreover,
x ∈ K commutes with any matrix in the set generated by {E AB }. Indeed,
X
xE AB = E C 1 x1 E 1C E AB = E A1 x1 E 1A E AB = E A1 x1 E 1B
C
X
= E AB E B1 x1 E 1B = E AB E C 1 x1 E 1C = E AB x.
C
P P
Consider now x ∈ A; since 1A = A fA = A E AA , it is then possible to write
X X X X
x= E AA x E BB = E AA xE BA E AB = E CA xE BC E AB , (4.51)
A B AB ABC
and subsequently X
x= (xC )AB E AB , (4.52)
ABC
where we define
(xC )AB = E CA xE BC . (4.53)
Since (xC )AB = fC (xC )AB fC , it follows that (xC )AB ∈ ACC . As expected, eqn (4.49)
holds, that is,
(xD )AB = E DC (xC )AB E CD . (4.54)
According to eqn (4.50), we define

X
xAB = (xC )AB , (4.55)
C
namely
X
xAB = E CA xE BC (4.56)
C
Finally, we can write eqn (4.52) as
X
x= xAB E AB (4.57)
AB
where xAB ∈ K, and {E AB } forms a basis for the space of matrices N × N . This result
shows that, if A is simple, then A can be written as K ⊗ M(N, R). The element x
can be represented by the matrix with coefficients xAB , accordingly. In fact, there is a
representation ρ given by
 
x11 x12 . . . x1N
 x21 x22 . . . x2N 
 
ρ(x) =  . .. .. ..  (4.58)
 .. . . . 
xN 1 xN 2 . . . xN N
Clearly, the expression (4.56) is not unique. In fact, any invertible element u ∈ A
can be used to define another basis of M(N, R) by
E ′AB = uE AB u−1 , (4.59)
and so yields
X
x= x′AB E ′AB , (4.60)
AB
P P
where x′AB = C E ′CA aE ′BC = C uE CA u−1 auE BC u−1 = u(u−1 au)AB u−1 .
The representation space here is a minimal left ideal Af1 , because of the choice of
the idempotent f1 when the quantities E AB = E A1 E 1B (A, B = 1, . . . , N ) are defined
via eqn (4.44). If an idempotent fC were chosen, the representation space would clearly
be AfC . The set of objects {E AB } acts on E A1 (A = 1, P. . . , N ), and {E A1 } is
Pa basis for
the ideal Af1 . Indeed, for x ∈ A, we can write xf1 = AB xAB E ABE 11 = A xA1 E A1 ,
which shows that {E A1 } generates A1 . It is clear that the set {E A1 } is linearly indepen-
dent, so there is a basis for the minimal left ideal Af1 . Since A ≃ K ⊗ M(N, R), there-
fore Af1 ≃ K ⊗ RN , a result which follows from E A1 ≃ eA , where eA (A = 1, . . . , N )
is a basis of RN . The isomorphism Af1 ≃ K ⊗ RN , although obvious, has leading and
substantial consequences, which shall be explored in chapter 6.
4.4 Clifford Algebra Representations

The Clifford algebras are already known to be either isomorphic to simple algebras
or isomorphic to the direct sum of simple algebras (namely, a semisimple algebra).
As a result of the discussion in section 4.3, we now have a method to obtain matrix
representations of Clifford algebras. Let us summarise the steps of this procedure:
P Choose a set of N primitive idempotents fA (A = 1, . . . , N ) of Cℓp,q such that

(1)
f = 1, with one of them being a primitive idempotent, for instance, f1 .
A A
(2) Choose elements {E A1 } and {E 1A } (A = 1, . . . , N ) such that f1 = E 1A E A1 and

that eqn (4.43) holds, namely, fB E A1 = δAB E A1 , and E 1A fB = δAB E 1A .
It is possible to switch these steps with other, equivalent ones:
(1) Choose a primitive idempotent f1 of Cℓp,q .
(2) Find a basis of the ideal Cℓp,q f1 , denoted by {E A1 }, and the associated dual basis
{E 1A } which satisfies E 1A E B1 = δAB f1 .
The third step, whichever of the first two steps is chosen, is
(3) Define a basis for M(N, R) as E AB = E A1 E 1B . If {γi = γ(ei )} (i = 1, . . . , n) are
the generators of Cℓp,q (p + q = n), its matrix representation is given by eqn (4.56):
X
(γi )AB = E CA γi E BC . (4.61)
C
The scalars are isomorphic to the set f1 Cℓp,q f1 , where f1 is the identity. The repre-
sentation space is isomorphic to the minimal left ideal Cℓp,q f1 .
Example 4.2 Let us consider the Clifford algebra Cℓ2,0 with {e1 , e2 } being an orthonormal basis of R2,0
satisfying (e1 )2 = (e2 )2 = 1, and e1 e2 + e2 e1 = 0. Obviously,
1
f± = (1 ± e1 )
2
are primitive idempotents of Cℓ2,0 and satisfy 1 = f+ + f− . Other idempotents are g± = (1/2)(1 ± e2 ),
as well as suitable linear combinations of f± and g± .
Let us begin with the first procedure. The idempotents are f1 = f+ , and f2 = f− , and let us choose
f1 . Now let us choose elements {E A1 } and {E 1A } (A = 1, 2) such that
f1 = E 1A E A1
and
fB E A1 = δAB E A1 , E 1A fB = δAB E 1A ,
for A, B = 1, 2. Since {E A1 } and {E 1A } are elements of Cℓ2,0 , they can be written as φ = a + be1 +
ce2 + de1 e2 . Let us calculate the products f1 φ, f2 φ, φf1 , and φf2 :
f1 φ = (a + b)f1 + (c + d)f1 e2 , f2 φ = (a − b)f2 + (c − d)f2 e2 ,
φf1 = (a + b)f1 + (c − d)f2 e2 , φf2 = (a − b)f2 + (c + d)f1 e2 .
These equations imply that f1 φf1 = (a + b)f1 = a′ f1 , (a′ ∈ R), in such a way that f1 φf1 ≃ R.
Using these results, it can be straightforwardly seen that the conditions f2 E 11 = E 11 f2 = 0, and
f1 E 11 = E 11 f1 = E 11 hold, and hence E 11 = a(1 + e1 ). The conditions f1 E 21 = E 21 f2 = 0 and
f2 E 21 = E 21 f1 = E 21 are valid for E 21 = c(1 − e1 )e2 . Finally, the conditions f2 E 12 = E 12 f1 = 0, and
f1 E 12 = E 12 f2 = E 12 hold for E 12 = c ′ (1 − e1 )e2 . The condition f1 = E 11 E 11 holds if a = 1/2 in E 11 .
On the other hand, the condition f1 = E 12 E 21 implies that cc ′ = 1/4. This is the same condition that
Clifford Algebra Representations 109
is obtained with f2 = E 22 = E 21 E 12 . The simplest solution is obviously c = c ′ = 1/2. Furthermore, we

can express
1 1
E 11 = (1 + e1 ), E 12 = (1 + e1 )e2 ,
2 2
1 1
E 21 = (1 − e1 )e2 , E 22 = (1 − e1 ).
2 2
In
P order to find a matrix representation of an element φ, let us calculate the matrix components φAB =
C
E CA φE BC = E 1A φE B1 + E 2A φE B2 . This calculation is certainly a lot of work. A straightforward
manner is to use directly the equations for E 11 , E 12 , E 21 , and e22 . From the first and last equations, it
can be seen that
1 = E 11 + E 22 , e1 = E 11 − E 22 .
From the second and third ones, we can see that
e2 = E 12 + E 21 , e1 e2 = E 12 − E 21 .
It is immediately clear that

1 0 1 0
ρ(1) = , ρ(e1 ) = ,
0 1 0 −1

0 1 0 1
ρ(e2 ) = , ρ(e1 e2 ) = .
1 0 −1 0
This representation is precisely that described in the discussion of the isomorphism Cℓ2,0 = M(2, R).
Let us consider the second procedure. Since φf1 = (a + b)(1/2)(1 + e1 ) + (c − d)(1/2)(1 − e1 )e2 , it
follows that the ideal Cℓ2,0 f1 is given by
1 1
Cℓ2,0 f1 = {a′ (1 + e1 ) + b′ (1 − e1 )e2 | a′ , b′ ∈ R}.
2 2
It is clear that B = {(1/2)(1 + e1 ), (1/2)(1 − e1 )e2 } = {E 11 , E 21 } is a basis for the ideal Cℓ2,0 f1 .
The dual basis can be straightforwardly obtained, and the action
of the dual basis elements {E 1A } can
be written in this case by the multiplication, as E 1A E B1 = δAB f1 = E 1A E B1 . The dual basis is
∗
B = {(1/2)(1 + e1 ), (1/2)(1 + e1 )e2 } = {E 11 , E 12 }. The objects E AB (A, B = 1, 2) are exactly the
ones already obtained and, therefore, the representation is completely derived.
Example 4.3 Another important example involves the Clifford algebra Cℓ3,0 . If {e1 , e2 , e3 } denotes an
orthonormal basis of R3,0 , we have (ei )2 = 1 (i = 1, 2, 3), and ei ej + ej ei = 0 (i 6= j), and an arbitrary
element of Cℓ3,0 is given by
φ = a0 + a1 e1 + a2 e2 + a3 e3 + a12 e1 e2 + a13 e1 e3 + a23 e2 e3 + a123 e1 e2 e33 .
This algebra has an interesting property: the pseudoscalar (or 3-vector) e1 e2 e3 commutes with any element
of Cℓ3,0 . Indeed, it is immediate to see that e1 e2 e3 ei = ei e1 e2 e3 for i = 1, 2, 3, so that if e1 e2 e3
commutes with {e1 , e2 , e3 } which are generators of the algebra,
V then Vit commutes with any element in
this algebra. The centre of Cℓ3,0 is therefore Cen(Cℓ3,0 ) = 0 (R3,0 ) ⊕ 3 (R3,0 ). On the other hand, the
algebra Cℓ3,0 presents a property that is not shared by any Clifford algebra. Indeed, calculating (e1 e2 e3 )2 ,
we find that
(e1 e2 e3 )2 = −1.
The centre of Cℓ3,0 is therefore such that Cen(Cℓ3,0 ) ≃ C. This result suggests the following notation:
I = e1 e2 e3 .
A set of primitive idempotents of Cℓ3,0 is
1 1
f1 = (1 + e3 ), f2 = (1 − e3 ).
2 2
The choice of the vector e3 is arbitrary. The vector e1 could be chosen as well as e2 or, in the general
case, any vector v such that v2 = 1 can be placed, instead. Let us calculate f1 φ, f2 φ, φf1 , and φf2 . In
order to do so, some preliminaries are required. First,
e3 f1 = f1 e3 = f1 , e3 f2 = f2 e3 = −f2 .
It then follows that

e1 e2 f1 = f1 e1 e2 = If1 , e1 e2 f2 = f2 e1 e2 = −If2 .
In addition,
e2 f1 = f2 e2 = e1 If1 , e2 f2 = f1 e2 = −e1 If2 .
By using those results and some other manipulations, we find that
f1 φ = [(a0 + a3 ) + (a12 + a123 )I]f1 + [(a1 − a13 ) + (−a2 + a23 )I]f1 e1 ,
f2 φ = [(a0 − a3 ) + (−a12 + a123 )I]f2 + [(a1 + a13 ) + (a2 + a23 )I]f2 e1 ,
φf1 = [(a0 + a3 ) + (a12 + a123 )I]f1 + [(a1 + a13 ) + (a2 + a23 )I]e1 f1 ,
φf2 = [(a0 − a3 ) + (−a12 + a123 )I]f2 + [(a1 − a13 ) + (−a2 + a23 )I]e1 f2 .
Furthermore, we can observe that
f1 φf1 = [(a0 + a3 ) + (a12 + a123 )I]f1 ≃ C.
The element 1 is represented by the identity matrix I, and the pseudoscalar I is represented by the matrix
iI, where i denotes the imaginary unit.
Given these results, the conditions f2 E 11 = E 11 f2 = 0 and f1 E 11 = E 11 f1 = E 11 hold for E 11 =
(a + bI)(1 + e3 ). In addition, E 11 E 11 = E 11 , in such a way that 4(a2 − b2 ) = 2a and 2ab = b, the solution
of which is a = 1/2, and b = 0. Therefore, E 11 is given by
1
E 11 = (1 + e3 ).
2
The conditions f1 E 21 = E 21 f2 = 0, and f2 E 21 = E 21 f1 = E 21 , hold for E 21 = (a′ e1 + b′ e2 )(1 + e3 ),
inasmuch as the conditions f2 E 12 = E 12 f1 = 0, and f1 E 12 = E 12 f2 = E 12 , hold for E 12 = (a′′ e1 +
b′′ e2 )(1 − e3 ). The condition e12 E 21 = E 11 implies that a′ b′′ = a′′ b′ , and a′ a′′ + b′ b′′ = 1/4. One
possible solution is b′ = b′′ = 0, and a′ = a′′ = 1/2. Then,
1 1
E 12 = e1 (1 − e3 ), E 21 = e1 (1 + e3 ).
2 2
When E 22 = E 21 E 12 , it reads
1
E 22 = (1 − e3 ).
2
The equations for E AB (A, B = 1, 2) imply that
1 = E 11 + E 22 , e3 = E 11 − E 22 ,
e1 = E 12 + E 21 , e1 e3 = E 21 − E 12 .
For the remaining elements, we need just to take the representation of I = e1 e2 e3 as being the imaginary
unit i, namely, IE AB = iE AB . In addition to these equations, we can express
e1 e2 e3 = I1 = iE 11 + iE 22 , e1 e2 = Ie3 = iE 11 − iE 22 ,
e2 e3 = Ie1 = iE 12 + iE 21 , e2 = Ie1 e3 = iE 21 − iE 12 .
A matrix representation of the vectors e1 , e2 , and e3 is given by

01 0 −i 1 0
ρ(e1 ) = , ρ(e2 ) = , ρ(e3 ) = ,
10 i 0 0 −1
which are the Pauli matrices σi = ρ(ei ) (i = 1, 2, 3) already seen in example 3.3.
Using the other procedure described in the text, we must regard the ideal Cℓ3,0 f1 , whose elements
are expressed by
Cℓ3,0 f1 = {(a + bI)f1 + (c + dI)e1 f1 | a, b, c, d ∈ R}.
A basis for this ideal is provided by
B = {f1 , e1 f1 } = {E 11 , E 21 }.
Since e1 f1 = f2 e1 , and f1 f2 = f2 f1 = 0, and E 1A (E B1 ) = E 1A E B1 = δAB f1 , it follows that the dual
basis is
B∗ = {f1 , f1 e1 } = {E 11 , E 12 },
and therefore E 22 = E 21 E 12 = e1 f1 f1 e1 = e1 f1 e1 = (e1 )2 f2 = f2 . The representation is derived in the
same way as the previous procedure was.
In these methods for obtaining a representation of a Clifford algebra, the starting

point is a primitive idempotent that generates – by right multiplication – a minimal
left ideal Cℓp,q f , which is the space of representation of a simple Clifford algebra. In
order to close this topic, we shall establish a result that makes it possible to obtain
this idempotent.
In what follows, the multi-index notation shall be used. Let {ei } (i = 1, . . . , n =
p + q) be an orthonormal basis of Rp,q . An element eµ1 · · · eµk of Cℓp,q is denoted by
eµ1 ···µk = eµ1 · · · eµk . (4.62)
and the set of multi-indices by I.
Theorem 4.6 ◮ Let Cℓp,q be the Clifford algebra associated with Rp,q and let {ei }
(i = 1, . . . , n) be an orthonormal basis of this quadratic space. A primitive idempotent
of Cℓp,q is given by
1 1
f = (1 + eI1 ) · · · (1 + eIk ), (4.63)
2 2
where {eI1 , . . . , eIk } is a set of elements which are in Cℓp,q , which commute, and such
that (eIα )2 = 1 for α = 1, . . . , k. This idempotent then generates a group of order 2k ,
where k = q − rq−p , and rj are the Radon–Hurwitz numbers defined by table 4.3 with
the recurrence relation rj+8 = rj + 4.
Table 4.3 The Radon–Hurwitz Numbers
j 0 1 2 3 4 5 6 7
rj 0 1 2 2 3 3 3 3
Proof: Let us take an element eI1 of Cℓp,q such that (eI1 )2 = 1. Hence, f1± = (1/2)(1 ±
eI1 ) are orthogonal idempotents: in fact, f1+ + f1− = 1, and f1+ f1− = f1− f1+ = 0.
The Clifford algebra Cℓp,q can be therefore decomposed into the sum of two ideals
Cℓp,q f1+ ⊕ Cℓp,q f1− , and the dimension of each one of these ideals is 2n /2 = 2n−1 .
Let now eI2 be another element of Cℓp,q such that (eI2 )2 = 1. In this way, f2± =
(1/2)(1 ± eI2 ) are idempotents satisfying f2+ + f2− = 1, and f2+ f2− = f2− f2+ = 0.
If eI1 and eI2 commute, four mutually orthogonal idempotents can be constructed:
f1+ f2+ , f1− f2− , f1− f2+ , and f1− f2− , the sum of which is the identity. The algebra Cℓp,q
can be thus decomposed into the sum of four ideals, each one with dimension 2n−2 .
Continuing this procedure, a set of 2k idempotents
1 1
(1 ± eI1 ) · · · (1 ± eIk )
2 2
can be obtained, and we can decompose Cℓp,q in the sum of 2k ideals, each one with
dimension 2n−k . Such reasoning can be again used until 2n−k equals the dimension
of the irreducible representation space of Cℓp,q , a value that can be obtained from
table 4.1. With a little work, it can be concluded that k is given by this formula. ✓
Example 4.4 Let us consider the Clifford algebra Cℓ0,7 . According to theorem 4.6, k = 7 − r7 = 7 − 3 = 4.
We must look for a set of four elements, that is, eI1 , eI2 , eI3 , and eI4 , which are of Cℓ0,7 , which
commute, and the square of which equals 1. Since, for all vectors in Cℓ0,7 , we have (ei )2 = −1 (i =
1, . . . , 7), the elements that we search for are not 1-vectors and, furthermore, cannot be 2-vectors, since
(ei ej )2 = −(ei )2 (ej )2 = −1 (i 6= j). Meanwhile, the four elements can be 3-vectors, since (ei ej ek )2 =
−(ei )2 (ej )2 (ek )2 = 1 (i 6= j 6= k). Let us choose one of those elements as eI1 = e123 = e1 e2 e3 .
Moreover, 3-vectors, having only one of their indices equal, commute. Hence, we can identify, for instance,
eI2 = e145 . Moreover, eI3 = e167 , and the fourth element might be eI4 = e347 , for example. Therefore,
the idempotent
1 1 1 1
f = (1 + e1 e2 e3 ) (1 + e1 e4 e5 ) (1 + e1 e6 e7 ) (1 + e3 e4 e7 )
2 2 2 2
is primitive in Cℓ0,7 .
Example 4.5 Let us consider Cℓ1,3 (C) ≃ M(4, C) and let {e0 , e1 , e2 , e3 } be an orthonormal basis of
R1,3 . Four primitive idempotents f1 , f2 , f3 , and f4 must be obtained such that 1 = f1 + f2 + f3 + f4 .
By theorem 4.6, we know that two elements, eI1 , eI2 , which are of Cℓ1,3 (C), which commute, and for
which (eI1 )2 = (eI2 )2 = 1, must be obtained in such a way that such four primitive idempotents are
(1/2)(1 ± eI1 )(1/2)(1 ± eI2 ). For the standard representation of the so-called Dirac gamma matrices,
these elements are eI1 = e0 , and eI2 = ie1 e2 . Hence,
1 1 1 1
f1 = (1 + e0 ) (1 + ie1 e2 ), f2 = (1 + e0 ) (1 − ie1 e2 ),
2 2 2 2
1 1 1 1
f3 = (1 − e0 ) (1 + ie1 e2 ), f4 = (1 − e0 ) (1 − ie1 e2 ). (4.64)
2 2 2 2
These four primitive idempotents are similar. Indeed,
e13 f1 (e13 )−1 = f2 , e30 f1 (e30 )−1 = f3 , e10 f1 (e10 )−1 = f4 . (4.65)
Thus, e13 f1 ⊂ f2 Cℓ1,3 (C)f1 ; e30 f1 ⊂ f3 Cℓ1,3 (C)f1 ; and e10 f1 ⊂ f4 Cℓ1,3 (C)f1 ; so
E 11 = f1 , E 21 = −e13 f1 , E 31 = e30 f1 , E 41 = e10 f1 . (4.66)
With the restriction that E 1j ⊂ f1 Cℓ1,3 (C)fj , and E 1j E j1 = f1 , this result implies that
E 11 = f1 , E 12 = e13 f2 , E 13 = e30 f3 , E 14 = e10 f4 . (4.67)
All the E ij are shown in table 4.4. From these identities, the matrix representations of eµ , denoted by
γµ = γ(eµ ), can be immediately obtained:
(i) e0 = f1 + f2 − f3 − f4 = E 11 + E 22 − E 33 − E 44 . Hence,
 
1 0 0 0
0 1 0 0 
γ(e0 ) = γ0 =  . (4.68)
0 0 −1 0 
0 0 0 −1
(ii) e10 = e10 f1 + e10 f2 + e10 f3 + e10 f4 = E 41 + E 32 + E 23 + E 14 . Therefore,
   
0001 0 0 0 −1
0 0 1 0  0 0 −1 0  0 −σ1
γ(e10 ) = γ10 =   , γ(e1 ) = γ1 = γ10 γ0 =   = . (4.69)
0100 01 0 0 σ1 0
1000 10 0 0
Table 4.4 The Elements E ij for i, j = 1, 2, 3, 4.
E ij E i1 E i2 E i3 E i4
E 1j f1 e13 f2 e30 f3 e10 f4
E 2j −e13 f1 f2 e10 f3 −e30 f4
E 3j e30 f1 e10 f2 f3 e13 f4
E 4j e10 f1 e03 f2 −e13 f3 f4
(iii) e30 = e30 f1 + e30 f2 + e30 f3 + e30 f4 = E 31 − E 42 + E 13 − E 24 . This result implies that
 
0 0 1 0
0 0 0 −1  0 −σ3
γ(e30 ) = γ30 =   , and γ3 = −γ0 γ03 = . (4.70)
1 0 0 0 σ3 0
0 −1 0 0
(iv) f1 +f3 −f2 −f4 = ie1 e2 =⇒ e2 = ie0 (e01 f1 +e01 f3 −e01 f2 −e01 f4 ) = ie0 (E 14 +E 32 −E 23 −E 41 ).
It follows that   
10 0 0 0 0 0 1
 0 1 0 0   0 0 −1 0  0 −σ2
γ2 = i     = . (4.71)
0 0 −1 0 0 1 0 0 σ2 0
0 0 0 −1 −1 0 0 0
The Dirac matrices standard representation is consequently given by

I 0 0 −σk
γ0 = , γ(ek ) = γk = (4.72)
0 −I σk 0
Example 4.6 In view of the isomorphism Cℓ4,1 ≃ C ⊗ Cℓ1,3 and the matrix representation of C ⊗ Cℓ1,3 ≃
Cℓ1,3 (C) obtained in example 4.5, we can construct a matrix representation of Cℓ4,1 . But in order to
implement this representation, we need to explicitly construct such an isomorphism. Let us denote the
generators of Cℓ4,1 by EA (A = 0, 1, 2, 3, 4) such that E12 = E22 = E32 = E42 = −E02 = 1. An isomorphism
ρ1 : Cℓ4,1 → C ⊗ Cℓ1,3 can be defined by
ρ1 (Eµ ) = −i ⋆ γµ = −iγµ γ0123 (µ = 0, 1, 2, 3), ρ1 (E4 ) = −i ⋆ 1 = −iγ0123 .
If ̺ : Cℓ1,3 (C) → M(4, C) is the standard representation from example 4.5, we have a representation
̺1 = ̺ ◦ ρ1 : Cℓ4,1 → M(4, C) such that, if Z ∈ Cℓ4,1 is an arbitrary element of the form
Z = H + H A EA + H AB EAB + H ABC EABC + H ABCD EABCD + H 01234 E01234 ,
its matrix representation ̺1 (Z) reads
 
z11 z12 z13 z14
z z22 z23 z24  φ1 φ2
̺1 (Z) =  21  = ,
z31 z32 z33 z34 φ3 φ4
z41 z42 z43 z44
where
z11 = (H + H 04 + H 034 − H 3 ) + i(H 01234 − H 123 + H 12 + H 0124 ),
z12 = (−H 13 − H 0134 + H 014 − H 1 ) + i(−H 024 + H 2 + H 23 + H 0234 ),
z13 = (H 03 − H 34 + H 4 + H 0 ) + i(H 124 + H 012 + H 0123 − H 1234 ),
z14 = (H 01 − H 14 + H 134 − H 013 ) + i(H 234 + H 023 − H 02 + H 24 ),
z21 = (H 13 + H 0134 + H 014 − H 1 ) + i(H 024 − H 2 + H 23 + H 0234 ),
z22 = (H + H 04 − H 034 + H 3 ) + i(H 01234 − H 123 − H 12 − H 0124 ),
z14 = (H 01 − H 14 + H 134 + H 013 ) + i(H 234 + H 023 + H 02 − H 24 ),
z24 = (−H 03 + H 34 + H 4 + H 0 ) + i(−H 124 − H 012 + H 0123 − H 1234 ),
z31 = (H 03 + H 34 + H 4 − H 0 ) + i(H 124 − H 012 + H 0123 + H 1234 ),
z32 = (H 01 + H 14 − H 134 + H 013 ) + i(H 234 − H 023 − H 02 − H 24 ),
z33 = (H − H 04 + H 034 + H 3 ) + i(H 01234 + H 123 + H 12 − H 0124 ),
z34 = (−H 13 + H 0134 + H 014 + H 1 ) + i(−H 024 − H 2 + H 23 − H 0234 ),
z41 = (H 01 + H 14 + H 134 − H 013 ) + i(H 234 − H 023 + H 02 + H 24 ),
z42 = (−H 03 − H 34 + H 4 − H 0 ) + i(−H 124 + H 012 + H 0123 + H 1234 ),
z43 = (H 13 − H 0134 + H 014 + H 1 ) + i(H 024 + H 2 + H 23 − H 0234 ),
z44 = (H − H 04 − H 034 − H 3 ) + i(H 01234 + H 123 − H 12 + H 0124 ).
The matrix representations of the conjugation Z̄, the reversion Z̃, and the graded involution Ẑ are given,
respectively, by !
φ†4 −φ†2 adj(φ4 ) adj(φ2 )
̺1 (Z̄) = , ̺ 1 ( Z̃) = ,
−φ†3 φ†1 adj(φ3 ) adj(φ1 )
and
cof(φ∗1 ) − cof(φ∗2 ) a b d −c
̺1 (Ẑ) = ∗ ∗ , with cof = ,
− cof(φ3 ) cof(φ4 ) c d −b a
where † denotes Hermitian conjugation, and cof(φ) = adj(φT ).
Another isomorphism, ρ2 : Cℓ4,1 → C ⊗ Cℓ1,3 , is given by
ρ2 (Eµ ) = −iγµ (µ = 0, 1, 2, 3), ρ2 (E4 ) = −iγ0123 .
Now we have the matrix representation ̺2 = ̺ ◦ ρ2 such that
 ′ ′ z ′ z′ 
z11 z12 13 14 ′ ′
′ z ′ z ′ z′
z21 22 23 24  = φ1 φ2 ,
̺2 (Z) =  ′ ′ ′ ′ 
z31 z32 z33 z34 φ′3 φ′4
′ z ′ z ′ z′
z41 42 43 44
where
′
z11 = (H − H 1234 + H 034 + H 012 ) + i(H 01234 − H 0 + H 12 + H 34 ),
′
z12 = (−H 13 + H 24 + H 014 + H 023 ) + i(−H 024 + H 013 + H 23 + H 14 ),
′
z13 = (H 03 − H 0124 + H 4 − H 123 ) + i(H 124 + H 3 + H 0123 − H 04 ),
′
z14 = (H 01 + H 0234 − H 134 + H 2 ) + i(H 234 + H 1 − H 02 + H 0134 ),
′
z21 = (H 13 − H 24 + H 014 + H 023 ) + i(H 024 − H 013 + H 23 + H 14 ),
′
z22 = (H − H 1234 − H 034 − H 012 ) + i(H 01234 − H 0 − H 12 − H 34 ),
′
z23 = (H 01 + H 0234 + H 134 − H 2 ) + i(H 234 + H 1 + H 02 − H 0134 ),
′
z24 = (−H 03 − H 0124 + H 4 + H 123 ) + i(−H 124 − H 3 + H 0123 + H 04 ),
′
z31 = (H 03 − H 0124 + H 4 + H 123 ) + i(H 124 − H 3 + H 0123 + H 04 ),
′
z32 = (H 01 − H 0234 − H 134 − H 2 ) + i(H 234 − H 1 − H 02 − H 0134 ),
′
z33 = (H + H 1234 + H 034 − H 012 ) + i(H 01234 + H 0 + H 12 − H 34 ),
′
z34 = (−H 13 − H 24 + H 014 − H 023 ) + i(−H 024 − H 013 + H 23 − H 14 ),
′
z41 = (H 01 − H 0234 + H 134 + H 2 ) + i(H 234 − H 1 + H 02 + H 0134 ),
′
z42 = (−H 03 + H 0124 + H 4 + H 123 ) + i(−H 124 + H 3 + H 0123 + H 04 ),
′
z43 = (H 13 + H 24 + H 014 − H 023 ) + i(H 024 + H 013 + H 23 − H 14 ),
′
z44 = (H + H 1234 − H 034 + H 012 ) + i(H 01234 + H 0 − H 12 + H 34 ).
The matrix representation of the conjugation Z̄, the reversion Z̃, and the graded involution Ẑ are given,
respectively, by ′†
φ1 −φ′3 † adj(φ′1 ) adj(φ′3 )
̺2 (Z̄) = ′ † ′ † , ̺2 (Z̃) = ′ ′ ,
−φ2 φ4 , adj(φ2 ) adj(φ4 )
and
cof(φ′1 ∗ ) − cof(φ′2 ∗ )
̺2 (Ẑ) = ′ ∗ ′ ∗ .
− cof(φ3 ) cof(φ4 )
Now a few remarks are deserved. First, notice that the matrix representations ̺1 (Z) and ̺2 (Z) are
different, although we have used the same standard representation of the gamma matrices. Second, ̺1 (Z̄)
and ̺2 (Z̄) are obtained from ̺1 (Z) and ̺2 (Z), respectively, by different matrix operations. The same is
true for ̺1 (Z̃) and ̺2 (Z̃). However, ̺1 (Ẑ) and ̺2 (Ẑ) are obtained from ̺1 (Z) and ̺2 (Z), respectively,
by the same matrix operations. The first fact follows from ρ1 and ρ2 being different isomorphisms; the
images of Z ∈ Cℓ4,1 in C ⊗ Cℓ1,3 are different, and it is not difficult to see that
ρ1 (Z) = T+ [ρ2 (Z)], ρ2 (Z) = T− [ρ1 (Z)],
where
1 1
T± [ψ] = (ψ + ψ̂) ± (ψ − ψ̂)γ0123 .
2 2
The second follows from the fact that there are different images of the anti-automorphisms but the same
one for the graded involution, that is
ρ1 (Cℓ4,1 ) ≃ C∗ ⊗ Cℓ1,3 , ]
ρ2 (Cℓ4,1 ) ≃ C∗ ⊗ Cℓ 1,3 ,
]
ρ1 (Cℓ 4,1 ) ≃ C ⊗ Cℓ1,3 , ]
ρ2 (Cℓ ]
4,1 ) ≃ C ⊗ Cℓ 1,3 ,
[
ρ1 (Cℓ ∗
4,1 ) ≃ C ⊗ Cℓ1,3 , [
ρ2 (Cℓ ∗
4,1 ) ≃ C ⊗ Cℓ1,3 ,
]
Notice that T± [Cℓ ]
1,3 ] = Cℓ1,3 , and T± [Cℓ1,3 ] = Cℓ1,3 . It is also opportune to observe that the complex
conjugation in C∗ ⊗ Cℓ1,3 is not the same as the complex conjugation in M(4, C), as will be seen in what
follows. Moreover, the graded involution in Cℓ1,3 is the image of the automorphism
Z △ = E4 ZE4 ,
that is,
ρ1 (Cℓ△ [ ρ2 (Cℓ△ [
4,1 ) ≃ C ⊗ Cℓ1,3 4,1 ) ≃ C ⊗ Cℓ1,3 .
Their matrix representations are, respectively,
′ ′
φ4 φ3 φ4 φ3
̺1 (Z) = , ̺2 (Z) = .
φ2 φ1 φ′2 φ′1
Another interesting isomorphism is ρ3 : Cℓ4,1 → C ⊗ Cℓ1,3 , which will be useful when discussing
twistors from the Clifford algebraic point of view (section 6.15), is
ρ3 (E0 ) = iγ0 , ρ3 (Ei ) = γi0 (i = 1, 2, 3), ρ3 (E4 ) = −γ123 .
The Hermitian Conjugation

The Hermitian conjugation is defined in the matrix algebra as a composition of the
complex conjugation and the transposition operations. Let us discuss the features
related to each of these operations in this context.
It is well-known that complex Clifford algebras are isomorphic either to the complex
algebra of the matrices or to the sum of two such algebras. In addition, the operation
of complex conjugation necessarily means that the complex conjugate of the matrices
components is taken into account. Consider an algebra A ≃ C ⊗ M(r, R) with an
involutive automorphism which is denoted by ∗ and which induces a non-trivial auto-
morphism at the centre of the algebra, and let B be a real subalgebra of A – namely,
a ∈ B if and only if a = a∗ (Benn and Tucker, 1987). Since any element a ∈ A can
be expressed as the sum of real and imaginary parts, it follows that A = C ⊗ B, and
therefore C ⊗ B ≃ C ⊗ M(r, R). In order for these statements to be true, the algebra B
must be simple; and, since only real algebras are isomorphic to the tensor product of
matrix algebras by the algebras R, C or H, either B ≃ M(r, R), or B ≃ H ⊗ M( r2 , R).
Now let us define in A another involutive automorphism ⋆ which leaves elements of
M(r, R) invariant, and which conjugates the elements in the centre. Choose the bases
{E ij } and {E ∗ij } for M(r, C), so that E ∗ij = mE ij m−1 for some m ∈ A. Hence, if
r
X
a= aij E ij , (4.73)
i,j=1
then
 
r
X r
X
a∗ = a∗ij mE ij m−1 = m  a∗ij E ij  m−1 , (4.74)
i,j=1 i,j=1
which implies that

a∗ = ma⋆ m−1 . (4.75)
Since ∗ and ⋆ are involutions, eqn (4.75) asserts that m∗ m = ρ, where ρ is an element
of the centre of the algebra. The involutions ∗ and ⋆ induce the same automorphism
in the centre, and (m∗ m)∗ = (m∗ m)⋆ = m−1 (m∗ m)m, namely, mm∗ = m∗ m, and ρ
is real. Thus, we can redefine m up to a scale factor, and m∗ m = ±1, or, equivalently,
m⋆ m = ±1. In what follows, we shall discuss how the sign chosen in the relation
m∗ m = ±1 corresponds to the classification of the real algebra B (Benn and Tucker,
1987).
Theorem 4.7 ◮ If A ≃ C ⊗ B, where ∗ and ⋆ are automorphisms that conjugate the

centre and leave B and M(r, R) invariant, then a∗ = ma⋆ m−1 , where mm∗ = 1 if and
only if B ≃ M(r, R), and mm∗ = −1 if and only if B ≃ H ⊗ M( 2r , R).
Proof: Since there are two mutually exclusive possibilities for B, and similarly for m,
if we prove that mm∗ = 1 ⇔ B ≃ M(r, R), it immediately follows that mm∗ = −1 ⇔
H ⊗ M( 2r , R). Suppose first that B ≃ M(r, R). Using bij and E ij to denote bases of B
and M(r, R) respectively, then we obtain E ij = sbij s−1 , for some s ∈ A. In addition,
by eqn (4.73), it follows that
r
X r
X
a∗ = a∗ij (sbij s−1 )∗ = a∗ij s∗ bij s∗ −1
i,j=1 i,j=1
r
(4.76)
X
= a∗ij s∗ s−1 E ij ss∗ −1 =s s∗ −1 ⋆
a (s s∗ −1 −1
) .
i,j=1
Therefore, m = s∗ s−1 can be chosen, so m∗ = m−1 . The reciprocal can be shown when
a linear mapping – complex conjugation – is introduced into A by ac = a∗ m = ma⋆ ,
and therefore c preserves the columns of M(r, R). When m∗ = m−1 , the mapping c
is involutive, and for any a ∈ A, we write a = 21 (a + ac ) + 12 (a − ac ). In particular,
the left minimal ideals of A – which are columns of M(r, R) with entries in C –
can be decomposed into eigenspaces of c ; since the dimension (over the real field)
of a left minimal ideal of A equals 2r, the associated eigenspaces have dimension r.
Let ψ be an element of such eigenspaces. If a ∈ B, it follows that aψ is in a left
minimal ideal of A and, since (aψ)c = a∗ ψ c = aψ c , it is indeed an autospace carrying
representations of B. Thus, if m∗ = m−1 , then irreducible representations of A induce
reducible representations in B. However, either B ≃ M(r, R) – in which case, its
irreducible representations are r-dimensional – or B ≃ H ⊗ M( 2r , R), with irreducible
representations of dimension 2r. Thus, m∗ = m−1 implies that B ≃ M(r, R). ✓
Let us now consider the transposition. Define
Ak = ei1 ···ik = ei1 · · · eik . (4.77)
The transposition of Ak , denoted by ATk , is defined by

(
T Ak , if A2k = 1,
Ak = (4.78)
−Ak , if A2k = −1,
and satisfies
(AB)T = B T AT , (4.79)
for A, B arbitrary multivectors.
An interesting question to be posed is whether it is possible to write the transpo-
sition as an inner automorphism, that is, if we can write AT as U AU −1 , U ÃU −1 , or
even as U ĀU −1 . By the property in eqn (4.79), the first possibility is excluded.
If p = 0 or q = 0, the transposition can be written in terms of the Clifford conju-
gation and the reversion, respectively. Indeed, since
Ãk = eik · · · ei1 = (−1)k(k−1)/2 Ak , (4.80)
it follows that
A2k = (−1)k(k−1)/2 e2i1 · · · e2ik . (4.81)
If p = 0, then e2i = −1 and therefore
A2k = (−1)k(k−1)/2 (−1)k , (4.82)
or, equivalently,
AT
k = (−1)
k(k−1)/2
(−1)k Ak = Āk . (4.83)
On the other hand, if q = 0, then e2i = 1, yielding
A2k = (−1)k(k−1)/2 , (4.84)
that is,
AT
k = (−1)
k(k−1)/2
Ak = Ãk . (4.85)
Now the cases where p 6= 0 and q 6= 0 are analysed. First, let us suppose that
the transposition is given by AT = U ÃU −1 . For a basis {ei } of Cℓp,q , in this case
eT T
i = ei for i = 1, . . . , p, and ei = −ei for i = p + 1, . . . , p + q, in such a way that,
T −1
if A = U ÃU , then U has to commute with ei for i = 1, . . . , p and must anti-
commute with ei for i = p + 1, . . . , p + q. If p is odd, there exists U satisfying the
properties given by U = e1···p = e1 · · · ep . If q is even, then there also exists U given
by U = ep+1...p+q = ep+1 · · · ep+q . If p is odd and q even, then a linear combination
of these elements also satisfies the properties. On the other hand, if p is even and q is
odd, there exists U such that AT = U ÃU −1 .
In addition, there exists another possibility, namely, AT = U ĀU −1 . In this case,
U must anti-commute with ei for i = 1, . . . , p and has to commute with ei for i =
p + 1, . . . , p + q. Therefore, if p is even, there exists U for which such properties

hold, given by U = e1···p = e1 · · · ep . If q is odd, there in addition exists U given by
U = ep+1...p+q = ep+1 · · · ep+q . Finally, if p is even and q is odd, a linear combination
of these elements also satisfies the properties.
To summarise:
p is odd =⇒ AT = U ÃU −1 , U = e1 · · · ep , (4.86)
p is even=⇒ AT = U ĀU −1 , U = e1 · · · ep , (4.87)
T −1
q is odd =⇒ A = U ĀU , U = ep+1 · · · ep+q , (4.88)
T −1
p is even=⇒ A = U ÃU , U = ep+1 · · · ep+q . (4.89)
Now the Hermitian conjugation, denoted by †, is defined by
A† = (A∗ )T = (AT )∗ , (4.90)
where ∗ is the complex conjugation in the Clifford algebra (when it is the case), and
the transposition is expressed according to eqns (4.86–4.89).
Example 4.7 In example 4.5, we obtained a matrix representation of the Clifford algebra C ⊗ Cℓ1,3 ≃
Cℓ1,3 (C) ≃ M(4, C). What was discussed there will be illustrated here using this algebra as an example.
Consider an arbitrary multivector
A = (α + iβ) + (α0 + iβ0 )e0 + (α1 + iβ1 )e1 + (α2 + iβ2 )e2 + (α3 + iβ3 )e3
+ (α01 + iβ01 )e0 e1 + (α02 + iβ02 )e0 e2 + (α03 + iβ03 )e0 e3
+ (α12 + iβ12 )e1 e2 + (α13 + iβ13 )e1 e3 + (α23 + iβ23 )e2 e3 (4.91)
+ (α012 + iβ012 )e0 e1 e2 + (α013 + iβ013 )e0 e1 e3 + (α023 + iβ023 )e0 e2 e3
+ (α123 + iβ123 )e1 e2 e3 + (α0123 + iβ0123 )e0 e1 e2 e3 .
Using this matrix representation, this multivector is represented by the matrix [A], whose components Aij
are
A11 = (α + α0 + β12 + β012 ) + i(β + β0 − α12 − α012 ),
A12 = (α13 + α013 + β23 + β023 ) + i(β13 + β013 − α23 − α023 ),
A13 = (−α3 − α03 − β123 − β0123 ) + i(−β3 − β03 + α123 + α0123 ),
A14 = (−α1 − α01 − β2 − β02 ) + i(−β1 − β01 + α2 + α02 ),
A21 = (−α13 − α013 + β23 + β023 ) + i(−β13 − β013 − α23 − α023 ),
A22 = (α + α0 − β12 − β012 ) + i(β + β0 + α12 + α012 ),
A23 = (−α1 − α01 + β2 + β02 ) + i(−β1 − β01 − α2 − α02 ),
A24 = (α3 + α03 − β123 − β0123 ) + i(β3 + β03 + α123 + α0123 ),
(4.92)
A31 = (α3 − α03 + β123 − β0123 ) + i(β3 − β03 − α123 + α0123 ),
A32 = (α1 − α01 + β2 − β02 ) + i(β1 − β01 − α2 + α02 ),
A33 = (α − α0 + β12 − β012 ) + i(β − β0 − α12 + α012 ),
A34 = (α13 − α013 + β23 − β023 ) + i(β13 − β013 − α23 + α023 ),
A41 = (α1 − α01 − β2 + β02 ) + i(β1 − β01 + α2 − α02 ),
A42 = (−α3 + α03 + β123 − β0123 ) + i(−β3 + β03 − α123 + α0123 ),
A43 = (−α13 + α013 + β23 − β023 ) + i(−β13 + β013 − α23 + α023 ),
A44 = (α − α0 − β12 + β012 ) + i(β − β0 + α12 − α012 ).
The complex conjugation is now considered in C ⊗ Cℓ1,3 ≃ Cℓ1,3 (C). The multivector A∗ is obtained
when we take i 7→ −i in eqn (4.91). Then we can obtain a matrix representation [A∗ ] of the multivector
A∗ according to these expressions. Let us denote the matrix components [A∗ ] by Bij . It follows that
Additional Readings 119
B11 = A∗22 , B31 = −A∗42 ,

B12 = −A∗21 , B32 = A∗41 ,
B13 = −A∗24 , B33 = A∗44 ,
B14 = A∗23 , B34 = −A∗43 ,
(4.93)
B21 = −A∗12 , B41 = A∗32 ,
B22 = A∗11 , B42 = −A∗31 ,
B23 = A∗14 , B42 = −A∗34 ,
B24 = −A∗13 , B44 = A∗33 .
If we denote by [A]∗ the matrix obtained from the complex conjugation A∗ij and evaluated on its compo-
nents, it is clear to see that
[A∗ ] 6= [A]∗ , (4.94)
as already discussed. By theorem 4.7 and since B = Cℓ1,3 ≃ M(2, H), we know that there must exist
m ∈ Cℓ1,3 satisfying mm∗ = −1 such that [mA∗ m−1 ] = [A]∗ . Furthermore, m = e013 = e0 e1 e3 ; or,
equivalently,
[e013 A∗ e−1 ∗ ∗ ∗ −1
013 ] = [A] ⇐⇒ [A ] = γ013 [A] γ013 . (4.95)
Regarding the Hermitian conjugation, as p = 1, it follows that U = e0 , that is,
A† = e0 Ã∗ e0 . (4.96)
By accomplishing this operation on the multivector A and then taking the matrix representation, we clearly
see that
[A† ] = [e0 Ã∗ e0 ] = [A]† , (4.97)
†
where [A] denotes the matrix which is Hermitian conjugated and associated with [A]. Consequently, it
follows that
¯
[Ã∗ ] = γ0 [A]† γ0 = [A], (4.98)
¯
which is the expression for the Dirac adjoint [A]. Using the expression for the complex conjugation, we
then obtain
−1
[Ã] = γ13 [A]T γ13 , (4.99)
where [A]T is the matrix transposed with respect to [A]. Finally, we can prove that
−1
[Â] = γ0123 [A]γ0123 . (4.100)

Some texts with a focus on Clifford algebras (and applications) in terms of their matrix
representations are those by Charlier, Bérard, Charlier, and Fristot (1992) and Snygg
(1997, 2010). In the first two works, one can find many applications in physics – for
example, in relativity, electromagnetism, quantum mechanics, and son on – whereas
the third contains applications in differential geometry. Garling (2011) uses Pauli spin
matrices to represent the angular momentum of particles with spin 1/2 to construct
the Dirac equation and expresses Maxwell’s equations as a single equation using the
Dirac operator.
4.6 Exercises
(1) Consider the Clifford algebra Cℓ1,3 ≃ M(2, H) and let {eµ } (µ = 0, 1, 2, 3) be an
orthonormal basis of R1,3 such that e20 = 1, e2i = −1, (i = 1, 2, 3). Using the idempotent
f = (1/2)(1 + e0 ), obtain the following matrix representation for the generators of this
algebra:

1 0 0i
ρ(e0 ) = , ρ(e1 ) = ,
0 −1 i0

0j 0k
ρ(e2 ) = , ρ(e3 ) = ,
j0 k0
where i, j, k are the quaternionic units.

(2) Consider the complexification C ⊗ Cℓ1,3 of the Clifford algebra Cℓ1,3 regarded in
the previous exercise. The standard representation of the Dirac matrices is such that
eI1 = e0 , and eI2 = ie1 e2 . Using
eI1 = e5 = e0123 , eI2 = ie1 e2 ,
obtain the so-called Weyl representation or chiral representation of the gamma matri-
ces, namely,

0 I 0 −σ1
γ0 = , γ1 = ,
−I 0 σ1 0

0 −σ2 0 −σ3
γ2 = , γ3 = ,
σ2 0 σ3 0
where I is the identity matrix 2 × 2, and σi denotes the Pauli matrices.

(3) Consider the Clifford algebra Cℓ3,1 ≃ M(4, R). Using
eI1 = e1 eI2 = e0 e2 ,
obtain the Majorana representation of the gamma matrices:

0 −iσ2 I 0
γ0 = , γ1 = ,
−iσ2 0 0 −I

0 −I 0 σ3
γ2 = , γ3 = .
I 0 σ3 0
5
Clifford Algebras, and Associated
Groups
In this chapter, we study the groups that can be defined within a Clifford algebra.
These groups, which deserve special attention, are the Clifford–Lipschitz, the Pin
group, and Spin group. The Lie algebras associated with those groups are hence
constructed and implemented, together with some of their applications. Conformal
transformations and the standard twistors of Penrose (Penrose, 1967; Penrose and
Rindler, 1984) are introduced from the algebraic point of view (Klotz, 1974; Keller,
1997; da Rocha and Vaz Jr, 2007) in a straightforward formulation, as geometric mul-
tivectors.
5.1 Orthogonal Transformations and the Cartan–Dieudonné

Theorem
Orthogonal Transformations
Let g be a symmetric bilinear form endowing the vector space V . A linear mapping
T : V → V is said to be an isometry, or orthogonal transformation, if
g(T (v), T (u)) = g(v, u), ∀ v, u ∈ V. (5.1)
Defining Tij by T (ei ) = Tij ej and letting gij = g(ei , ej ), we can consequently write
eqn (5.1) as Tik gkl Tjl = gij . When matrices are used, this equation is equivalent to
T ⊺ GT = G, where now T denotes a matrix with entries {Tij }, and T ⊺ denotes its
associated transposed matrix. Since det(AB) = det A det B, and det A = det A⊺ , it
immediately follows that
2
det T = 1. (5.2)
The orthogonal transformations for which det T = 1 are called rotations, and those
for which det T = −1 are reflections. The set of isometries forms a group called the
orthogonal group and is denoted by O(p, q) for V = Rp,q . The subgroup of O(p, q)
formed only by rotations is called the special orthogonal group and is denoted by
SO(p, q) – in general, the use of the notation S indicates that the group is restricted
to the case where det T = 1.
122 Clifford Algebras, and Associated Groups
The Components of the Orthogonal Group

Given a group G, define a path in this group as a continuous mapping χ : [0, 1] → G.
A subset G′ of G is said to be connected if, for any elements g0 , g1 ∈ G′ , there exists
a path χ(t) linking these elements, namely, χ(0) = g0 and χ(1) = g1 . A connected
subset that is not contained in any other connected subset is called a component of
the group G.
The orthogonal groups O(n, 0) and O(0, n) have two components. Indeed, given
orthogonal transformations T0 and T1 such that, for instance, det T0 = 1 and det T1 =
−1, there is no continuous path that links such transformations. Between these two
components, the one satisfying det T = 1 is either the subgroup SO(n, 0) or the sub-
group SO(0, n).
The orthogonal groups O(p, q), where p 6= 0 or q 6= 0, have four components. In
fact, let us consider an orthonormal basis {e1 , . . . , ep+q } of Rp,q in such a way that,
in terms of this basis, the symmetric bilinear functional g can be represented by the
matrix
1p 0
G= , (5.3)
0 −1q
where 1p and 1q denote the identity matrices of order p and q, respectively. Repre-
senting T by the matrix
Ap Bp,q
T = , (5.4)
Cq,p Dq
where Ap is the matrix p × p, and Bp,q is the matrix p × q, the condition T ⊺ GT = G
implies that
A⊺p Ap − Cp,q
⊺
Cq,p = 1p ,
Dq⊺ Dq − Bq,p
⊺
Bp,q = 1q , (5.5)
A⊺p Bp,q = Cp,q
⊺
Dq ,
⊺
where Bq,p = (Bp,q )⊺ , and Cp,q
⊺
= (Cq,p )⊺ .
The matrices Ap and Dq satisfy det Ap 6= 0, and det Dq 6= 0. Let us consider the
case for the matrix Ap – the case involving Dq is completely analogous. The matrix
Ap must satisfy A⊺p Ap = 1p + Cp,q⊺
Cq,p . This result implies that
(det Ap )2 = det(1p + Cp,q

⊺
Cq,p ). (5.6)
⊺ ⊺
Let us suppose that det(1p +Cp,q Cq,p ) = 0. In this case, the equation (1p +Cp,q Cq,p )X =
0 has a non-trivial solution X. However, X = −Cp,q Cq,p X, and therefore
⊺
X ⊺ X = −X ⊺ Cp,q
⊺
Cq,p X = −(Cq,p X)⊺ (Cq,p X). (5.7)
On the right-hand side, X ⊺ X = (X1 )2 +· · ·+(Xn )2 > 0, where Xi represents the com-
ponents of X, since X 6= 0. Similarly, the left-hand side yields −(Cq,p X)⊺ (Cq,p X) ≤ 0,
where the equality corresponds to the possibility Cq,p X = 0. Then we obtain a con-
⊺
tradiction; thus we must conclude that det(1p + Cp,q Cq,p ) 6= 0.
Orthogonal Transformations and the Cartan–Dieudonné Theorem 123
In addition, this result implies that
(det Ap )2 6= 0, (det Dq )2 6= 0. (5.8)
It is possible to divide the orthogonal transformations T ∈ O(p, q) into four classes:
(i) O↑+ (p, q) : det Ap > 0, det Dq > 0,

(ii) O↑− (p, q) : det Ap > 0, det Dq < 0,
(5.9)
(iii) O↓+ (p, q) : det Ap < 0, det Dq > 0,
(iv) O↓− (p, q) : det Ap < 0, det Dq < 0.
It is possible to show (it is left as an exercise) that, if det Ap det Dq > 0, then det T > 0.
However, if det Ap det Dq < 0, then det T < 0. In addition, the following sets can be
shown to be subsets of O(p, q):
(i) O↑+ (p, q),

(ii) O↑ (p, q) = O↑+ (p, q) ∪ O↑− (p, q),
(5.10)
(iii) O+ (p, q) = O↑+ (p, q) ∪ O↓+ (p, q),
(iv) O↑+ (p, q) ∪ O↓− (p, q).
We can moreover verify that
SO+ (p, q) = SO↑ (p, q) = SO↑+ (p, q) = O+ (p, q) ∩ O↑ (p, q). (5.11)
Example 5.1 Let us consider the orthogonal group O(1, 1). A matrix

a b
T = ∈ O(1, 1)
c d
must satisfy
a c 1 0 a b 1 0
= ,
b d 0 −1 c d 0 −1
so
a2 − c2 = 1, d2 − b2 = 1, ab = cd.
The two first equations hold when
a = ± cosh α, c = sinh α,
d = ± cosh β, b = sinh β,
and the last one provides the relation between α and β. It reads as follows: (i) if a = cosh α, and
d = cosh β, then α = β; (ii) if a = − cosh α, and d = cosh β, then α = −β; (iii) if a = cosh α, and
d = − cosh β, it implies that α = −β; and (iv) if a = − cosh α, and d = − cosh β, then α = β. Four
possibilities then exist:

cosh α sinh α cosh α − sinh α
T1 = , T2 = ,
sinh α cosh α sinh α − cosh α

− cosh α − sinh α − cosh α sinh α
T3 = , T4 = .
sinh α cosh α sinh α − cosh α
It is clear that there is no path that links any two matrices Ti and Tj , for i 6= j. According to the notation
used, we have
T1 ∈ O↑+ (1, 1), T2 ∈ O↑− (1, 1),

T3 ∈ O↓+ (1, 1), T4 ∈ O↓− (1, 1).
Orthogonal Symmetries and Reflections

Let us write V as V = U ⊕ U ⊥ , where U is a non-isotropic subspace of V , that is,
g(u, u) 6= 0, ∀ u ∈ V . We define the orthogonal symmetry SU with respect to U as
SU (vk + v⊥ ) = −vk + v⊥ , ∀ vk ∈ U, v⊥ ∈ U ⊥ . (5.12)
It is clear that SU = − idU + idU ⊥ , where id denotes the identity operator. Since the
matrix representation of id is the identity matrix, the determinant det SU of the matrix
that represents SU is obviously
det SU = (−1)dim U . (5.13)
Let us choose a vector u ∈ V such that g(u, u) 6= 0 and take U as the subspace
generated by u, namely, U = {au | a ∈ R}. We can write a vector v ∈ V as v = vk +v⊥ ,
where g(v⊥ , u) = 0. It is immediately obvious that the expression
g(v, u)
v⊥ = v − u (5.14)
g(u, u)
satisfies g(v⊥ , u) = 0; therefore, we can split v = vk + v⊥ , where v⊥ is provided by
eqn (5.14), and
g(v, u)
vk = u. (5.15)
g(u, u)
The symmetry Su is thus given by
Su (v) = Su (vk + v⊥ ) = −vk + v⊥
g(v, u) g(v, u) (5.16)
=− u+v− u,
g(u, u) g(u, u)
that is,
g(v, u)
Su = v − 2 u. (5.17)
g(u, u)
The symmetry Su is a reflection – indeed, from eqn (5.13) it follows that det Su =
(−1)1 = −1. This reflection occurs on the hyperplane orthogonal to the vector u.
Two arbitrary non-isotropic vectors v and u with the same norm – g(v, v) =
g(u, u) 6= 0 – can be related by at most two reflections. Indeed, let us first suppose
the case where g(v − u, v − u) 6= 0. Hence,
g(v, v − u)
Sv−u (v) = v − 2 (v − u)
g(v − u, v − u)
g(v, v) − g(v, u)
=v−2 (v − u)
g(v, v) + g(u, u) − 2g(v, u) (5.18)
g(v, v) − g(v, u)
=v− (v − u)
g(v, v) − g(v, u)
= v − (v − u) = u.
Orthogonal Transformations and the Cartan–Dieudonné Theorem 125
In the case where g(v − u, v − u) = 0, it follows that g(v, v) = g(u, u) = g(v, u) 6= 0,

and g(v + u, v + u) 6= 0. Then,
g(v, v + u)
Sv+u (v) = v − 2 (v + u)
g(v + u, v + u)
g(v, v) + g(v, u) (5.19)
=v− (v + u)
g(v, v) + g(v, u)
= v − (v + u) = −u ,
implying that
g(−u, u)
Su Sv+u (v) = Su (−u) = −u − 2 u = u. (5.20)
g(u, u)
The Cartan–Dieudonné Theorem

We have just proved that two arbitrary non-isotropic vectors with the same norm –
g(v, v) = g(u, u) 6= 0 – can be related by at most two reflections. This result is a
particular case of the Cartan–Dieudonné theorem, presented in what follows as its
‘weak version’:
Theorem 5.1 ◮ Any orthogonal transformation T in a finite dimensional vector space

V can be expressed as the product of symmetries (reflections) with respect to non-
isotropic hyperplanes.
Proof: The first assertion can be demonstrated by using finite induction, since it holds
for n = 1. Let us then assume that the assertion holds for dim V = n and show that, in
this case, it holds for dim V = n+1. Let v ∈ V be such that g(v, v) 6= 0; dim V = n+1;
and U = span{v}. The vector subspace U ⊥ is n-dimensional. If we denote by T an or-
thogonal transformation, then, by definition, g(T (v), T (v)) = g(v, v). Moreover, from
eqns (5.18) and (5.20), T (v) and v are related by at most two symmetries (reflections)
S; thus, S(T (v)) = v. Since the space U is S ◦ T -invariant, the orthogonal comple-
ment U ⊥ is also S ◦ T invariant; therefore, S ◦ T is an orthogonal transformation. Since
dim U ⊥ = n, supposing that the first assertion holds for dimension n, we can conclude
that S ◦T is the product Σ of a finite number of symmetries and therefore T = S −1 ◦Σ,
that is, the orthogonal transformation T in a space V such that dim V = n + 1 is the
product of a finite number of symmetries. ✓
Observation ☞ The ‘strong version’ of the Cartan–Dieudonné theorem asserts that, if
dim V = n, then T (T 6= id) can be expressed as the product of at most n symmetries.
The proof of this version is not as straightforward as that for the weak version. Thus,
since it will not be necessary to use the strong version in this book, we shall not go
into it in detail.
Now let us focus on eqn (5.17). In Cℓp,q we have vu + uv = 2g(v, u), and u2 =
g(u, u). The object u/g(u, u) can be interpreted as the inverse of the element u:
u u
u−1 = = 2, (5.21)
g(u, u) u
where u−1 u = uu−1 = 1. Hence, eqn (5.17) can be written as
(vu + uv)
Su (v) = v − u = v − (vu + uv)u−1 = v − v − uvu−1 , (5.22)
g(u, u)
namely,
Su (v) = −uvu−1 = u
b vu−1 . (5.23)
This equation has prominent usefulness in what follows.
5.2 The Clifford–Lipschitz Group
Group of The Invertible Elements

Various groups can be defined in a Clifford algebra Cℓp,q . The largest one is the group
of invertible (or regular) elements Cℓ∗p,q ,
Cℓ∗p,q = {a ∈ Cℓp,q | ∃a−1 }. (5.24)
The Clifford–Lipschitz Group

A subgroup of Cℓ∗p,q of great interest is the Clifford–Lipschitz group Γp,q , defined by
Γp,q = {a ∈ Cℓ∗p,q | ava−1 ∈ V, ∀v ∈ V = Rp,q } . (5.25)
It is straightforward to see that this set presents a group structure – in general, a

non-abelian structure.
Adjoint Representation
A representation ρ of the Clifford algebra Cℓp,q obviously defines a representation of the
Clifford–Lipschitz group Γp,q . Other representations of this group can also be defined;
a particular representation is called the adjoint representation, or vector representation
σ : Γp,q → Aut(Cℓp,q ), which is defined by,1
σ(a)(x) = axa−1 . (5.26)
Here, σ(a) is an element of the group of the automorphisms Aut(Cℓp,q ) of Cℓp,q .2
1 Another notation commonly employed is σ = Ad.

2 An endomorphism is a homomorphism of a set X on X. If this homomorphism is also an iso-
morphism, it is said to be an automorphism. The set of all automorphisms is the group of the
automorphisms.
The Clifford–Lipschitz Group 127
Consider v, u ∈ V = Rp,q , such that
vu + uv = 2g(v, u). (5.27)
Let us consider the vectors σ(a)v and σ(a)u. We have
2g(σ(a)v, σ(a)u) = σ(a)vσ(a)u + σ(a)uσ(a)v

= ava−1 aua−1 + aua−1 ava−1
(5.28)
= avua−1 + auva−1 = 2ag(v, u)a−1
= 2g(v, u).
Since σ(a) satisfies

g(σ(a)v, σ(a)u) = g(v, u), (5.29)
therefore,
σ(a) ∈ O(p, q) , (5.30)
where O(p, q) denotes the group of orthogonal transformations of Rp,q . Hence, σ is
both a mapping σ : Γp,q → O(p, q), and a group homomorphism. In fact,
σ(ab)(v) = abv(ab)−1 = abvb−1 a−1 = σ(a)σ(b)(v), (5.31)
and σ(ab) = σ(a)σ(b). We will show in what follows that σ is, furthermore, surjective.
In some cases, the image of σ(Γp,q ) may not be the whole group O(p, q) but instead
merely a subgroup, depending upon the dimension n = p + q of the vector space. To
determine which is the case, det σ(a) must be calculated.
The determinant of a linear transformation T can be usually defined by T (e1 ) ∧
· · · ∧ T (en ) = (det T )e1 ∧ · · · ∧ en , where {e1 , . . . , en } is a basis of V = Rp,q . Hence,
det σ(a) is given by

σ(a)(e1 ) ∧ · · · ∧ σ(a)(en ) = det σ(a) e1 ∧ · · · ∧ en . (5.32)
On the other hand, for vectors v and u, it follows that
σ(a)(v) ∧ σ(a)(u) = (ava−1 ) ∧ (aua−1 )

1
= (ava−1 aua−1 − aua−1 ava−1 ) (5.33)
2
1
= a (vu − uv)a−1 = a(v ∧ u)a−1 ,
2
that is,
σ(a)(v) ∧ σ(a)(u) = σ(a)(v ∧ u), (5.34)
and, for eqn (5.32), it follows that

a(e1 ∧ · · · ∧ en )a−1 = det σ(a) e1 ∧ · · · ∧ en . (5.35)
In order to compute the left-hand

V side of (5.35), let us take a as the homoge-
neous multivector a = a[k] ∈ k (V ) and then generalise the result for an arbitrary
multivector by linearity. Since e1 ∧ · · · ∧ en is an n-vector (pseudoscalar), it follows

that
a[k] (e1 ∧ · · · ∧ en ) = a[k]♭ ⌋(e1 ∧ · · · ∧ en ). (5.36)
Now by using eqn (2.60), we see that
a[k]♭ ⌋(e1 ∧ · · · ∧ en ) = (−1)k(n−1) (e1 ∧ · · · ∧ en )⌊a[k]♭ , (5.37)
which yields
a[k] (e1 ∧ · · · ∧ en ) = (−1)k(n−1) (e1 ∧ · · · ∧ en )a[k] . (5.38)
If n is odd, then n − 1 is even and consequently
a[k] (e1 ∧ · · · ∧ en ) = (e1 ∧ · · · ∧ en )a[k] , k = 0, 1, . . . , n. (5.39)
Hence, if n is odd, it follows that
a(e1 ∧ · · · ∧ en ) = (e1 ∧ · · · ∧ en )a (n odd), (5.40)
which can be substituted into eqn (5.35), so that

(e1 ∧ · · · ∧ en ) = det σ(a) e1 ∧ · · · ∧ en , (5.41)
and therefore
det σ(a) = 1, (n odd). (5.42)
On the other hand, if n is even, then n − 1 is odd and, from eqn (5.38), we conclude
that
a[k] (e1 ∧ · · · ∧ en ) = (−1)k (e1 ∧ · · · ∧ en )a[k] (n even). (5.43)
For n even, the expression det σ(a) = ±1 holds. Specifically, when n is even, we have
det σ(a) = 1 if a ∈ Cℓ+

p,q ,
(5.44)
det σ(a) = −1 if a ∈ Cℓ−
p,q .
Thus, we can conclude that

σ(Γp,q ) = O(p, q) (n even),
(5.45)
σ(Γp,q ) = SO(p, q) (n odd).
This result is somewhat disappointing: we have two distinct situations, depending
upon the dimension of the vector space V . It is clear that, if there is no way to deal
with this situation, we must separately consider each situation. It would be preferable
to deal with it independently of the dimension of the space V .
Twisted Adjoint Representation

b is
Instead of considering the representation σ, the Clifford group representation σ
defined by
σ ava−1 .
b(a)(v) = b (5.46)
The mapping σ b denotes the twisted adjoint (or vector) representation. Clearly, the
Clifford–Lipschitz group could be defined by the condition b ava−1 ∈ V , instead of
The Clifford–Lipschitz Group 129
ava−1 ∈ V for v ∈ V . However, the difference is irrelevant if only the group definition
is taken into account. Either way, the twisted Clifford–Lipschitz group Γ bp,q is defined
as
b p,q = {a ∈ Cℓp,q | b
Γ ava−1 ∈ V, ∀v ∈ V = Rp,q } . (5.47)
Obviously, the isomorphism Γ b p,q ≃ Γp,q holds. In terms of a representation, the pres-
ence of the grade involution ba drastically modifies the rule of each one of these groups,
as we will discuss.
The results obtained for σ can be straightforwardly
V adapted to σb. In order to avoid
unnecessary complication, let us take a = a[k] ∈ k (V ) and then generalise the results
b is
by linearity. First, the analogue of eqn (5.34) for σ
σ b(a[k] )(vi ) = (−1)k(i−1) σ

b(a[k] )(v1 ) ∧ · · · ∧ σ b(a[k] )(v1 ∧ · · · ∧ vi )
= (−1)k(i−1) (−1)k σ(a[k] )(v1 ∧ · · · ∧ vi ) (5.48)
= (−1)ki σ(a[k] )(v1 ∧ · · · ∧ vk ).
By using eqn (5.38), we obtain
σ b(a[k] )(en ) = (−1)nk a[k] (e1 ∧ · · · ∧ en )a−1

b(a[k] )(e1 ) ∧ · · · ∧ σ [k]
(5.49)
= (−1)nk (−1)k(n−1) e1 ∧ · · · ∧ en ,
which yields
σ b(a[k] )(en ) = (−1)k (e1 ∧ · · · ∧ en ).
b(a[k] )(e1 ) ∧ · · · ∧ σ (5.50)
Equation (5.35) implies that
b(a[k] ) = (−1)k .
det σ (5.51)
What should be realised here is that det σ b(a[k] ) does not depend on n, which is the
b(a[k] ) = ±1, therefore
case for det σ(a[k] ). Since det σ
b(Γp,q ) = O(p, q).
σ (5.52)
However, if we define
+
Γ+
p,q = Γp,q ∩ Cℓp,q , (5.53)
so that Γ+
p,q consists of the even elements of the Clifford–Lipschitz group, then
b(Γ+
σ p,q ) = SO(p, q). (5.54)
The mapping σ b is a surjective homomorphism. We show it for σ b : Γp,q → O(p, q),
and we conclude that the same holds for σ b : Γ+
p,q → SO(p, q) – and consequently also
for σ, as already asserted. Indeed, in section 5.1, it was shown (eqn (5.23)) that a
reflection with respect to the orthogonal hyperplane to a given vector u is given by
b vu−1 . Therefore,
Su (v) = u
b(u) = Su .
σ (5.55)
Clearly u ∈ Γp,q . Nevertheless, according to the Cartan–Dieudonné theorem, any
orthogonal transformation T ∈ O(p, q) can be written as the product of a finite number
of reflections of type Su = σ b(u). It follows that there exists a finite number of vectors
{u1 , . . . , uk } such that any orthogonal transformation reads
b(u1 ) · · · σ
σ b(u1 · · · uk ) .
b(uk ) = σ (5.56)
Hence, u1 · · · uk ∈ Γp,q , showing that σb is surjective. The same holds for σ and for the
restriction to the even elements σ b : Γ+
p,q → SO(p, q) as well.
Another equivalent characterisation of the Clifford–Lipschitz group can be obtained
by the last result, namely, as the group consisting of the product of all non-null vectors
of Cℓp,q :
Γp,q = {a ∈ Cℓ∗p,q | a = v1 · · · vk , where vi ∈ Rp,q and g(vi , vi ) 6= 0}. (5.57)
The decomposition a = v1 · · · vk is not unique. Moreover, we have
k = odd ⇐⇒ reflection,
(5.58)
k = even ⇐⇒ rotation .
We can further prove that

b = R∗ ,
ker σ (5.59)
where R∗ = R\{0} (and ker σ = R∗ ). Indeed, the condition σb(a) = 1,3 where a ∈ ker σ
b,
can be equivalently written as σ b(b), where a, b ∈ ker σ
b(a) = σ b; therefore
b(a)(v) = σ
σ b(b)(v) ,
ava = bbvb−1 ,
b −1
(5.60)
\
(b −1 a)v = v(b−1 a).
By writing b−1 a = (b−1 a)+ + (b−1 a)− , where (b−1 a)+ denotes the even part of b−1 a,
and (b−1 a)− denotes its odd part, we can see that
(b−1 a)+ v = v(b−1 a)+ , (b−1 a)− v = −v(b−1 a)− . (5.61)
The second condition holds only if (b−1 a)− = 0, since there is no element of Cℓp,q
that anti-commutes with all generators ei (i = 1, . . . , n) of Cℓp,q . The first condition
holds for (b−1V a)+ ∈ Cen(Cℓp,q ). As shown in chapter 4, exercise 1, if n is Veven, then
p,q
Cen(Cℓ
V p,q ) = 0 (R ); on the other hand, if nVis odd, then Cen(Cℓp,q ) = 0 (Rp,q ) ⊕
p,q
n (R ). However, if n is odd, an element of n (Rp,q ) is odd and, consequently, the
second condition must hold, V instead of the first one. Hence, the two conditions are only
satisfied when b−1 a ∈ 0 (Rp,q ) = R. Moreover, b−1 a is invertible and consequently
non-null. It follows that b−1 a ∈ R∗ , and ker σ b = R∗ .
3 The definition given in chapter 1 – ker f = {v ∈ V | f (v) = 0} for f : V → W – was provided with
respect to a linear mapping that exists between vector spaces V and W and where the unity is the
null vector 0. Here, the mapping σb : Γp,q → O(p, q) must be denoted by ker σ b = {a ∈ Γp,q | σ
b(a) = 1},
where 1 is the unity in O(p, q).
The Pin Group and the Spin Group 131
Example 5.2 Let us consider the Clifford algebra Cℓ2,0 , where an arbitrary element is given by
ψ = a0 + a1 e1 + a2 e2 + a12 e1 e2 .
In addition,
ψ̃ψ = a20 + a21 + a22 + a212 + 2(a0 a1 − a2 a12 )e1 + 2(a0 a2 − a1 a12 )e2 ,
ψ̄ψ = a20 − a21 − a22 + a212 .
The last expression shows that, if
ψ̄ψ = a20 − a21 − a22 + a212 6= 0 ,
then ψ −1 can be defined as
ψ̄ a0 − a1 e1 − a2 e2 − a12 e1 e2
ψ −1 = = .
ψ̄ψ a20 − a21 − a22 + a212
The group of the invertible elements Cℓ∗2,0 consists of the elements of Cℓ2,0 such that a20 −a21 −a22 +a212 6= 0.
In order to determine the Clifford–Lipschitz group, the relation ψvψ −1 ∈ R2,0 must hold for all
v ∈ R2,0 . Since ψ −1 = ψ̄/(ψ̄ψ), we just need to consider the similar condition ψvψ̄ ∈ R2,0 . By calculating
ψvψ̄, we obtain
ψvψ̄ =[v1 (a20 − a21 + a22 − a212 ) + v2 (2a0 a12 − 2a1 a2 )]e1
+ [v1 (−2a0 a12 − 2a1 a2 ) + v2 (a20 + a21 − a22 − a212 )]e2
+ [v1 (−2a1 a12 − 2a0 a2 ) + v2 (2a0 a1 − 2a2 a12 )]e1 e2 .
Hence, in order for ψvψ̄ ∈ R2,0 , the equalities

a0 a2 + a1 a12 = 0, a0 a1 − a2 a12 = 0
must hold. Here, there are two possibilities: either a1 = a2 = 0, or a0 = a12 = 0. In the first case,
elements of Γ2,0 are written as ψ = a0 + a12 e1 e2 , with the condition a20 + a212 6= 0. In the second case,
the elements of Γ2,0 are a1 e1 + a2 e2 = (a2 + a1 e1 e2 )e2 , with the condition a21 + a22 6= 0.
In order to understand what is being accomplished here, we use the matrix representation of Cℓ2,0 ,
as studied in chapter 4. The element ψ ∈ Cℓ2,0 can be represented by the matrix Ψ = ρ(ψ) ∈ M(2, R)
given by
a0 + a1 a2 + a12
Ψ= .
a2 − a12 a0 − a1
The condition ψ̄ψ 6= 0, necessary for the existence of ψ −1 , is equivalent to det Ψ 6= 0, where
det Ψ = ψ̄ψ = a20 − a21 − a22 + a212 .
The inverse matrix Ψ−1 corresponds to

1 a0 − a1 −a2 − a12
Ψ−1 = ,
det Ψ −a2 + a12 a0 + a1
which is exactly the matrix representation of ψ −1 . Notice that

a0 − a1 −a2 − a12
ρ(ψ̄) = .
−a2 + a12 a0 + a1
5.3 The Pin Group and the Spin Group

The property ker σb = R∗ can be interpreted as an assertion that the Clifford–Lipschitz
group Γp,q is ‘bigger than necessary’. It is not necessary to use the whole group Γp,q
to describe orthogonal transformations; instead, a subgroup can be used, which can
be obtained via a suitable ‘normalisation process’ can be used.
In chapter 3, we showed that the norm of a multivector – obtained by the ‘Grass-

mannian’ extension of g – reads in Cℓp,q as eqn (3.49)
N (a) = |he
aai0 |, (5.62)
where a ∈ Cℓp,q . In addition, another definition for the norm in Cℓp,q was shown to be
N ′ (a) = |hāai0 |. (5.63)
From eqn (3.51), it follows that

N ′ a[k] = N a[k] . (5.64)
It is worth emphasising here that, as we proved in section 3.3, this equality does not
hold when the multivectors are not simple.
The mappings N and N ′ are the homomorphisms N : Γp,q → R∗ , and N ′ : Γp,q →
R , as can be shown by eqn (5.57). Let us now consider the norm N , since the case
∗
involving N ′ is completely similar. By denoting a = v1 · · · vk ,

N (a) = hvk · · · v1 v1 · · · vk i0
(5.65)
= vk · · · v1 v1 · · · vk = N (v1 ) . . . N (vk ).
It follows that
g 0 = hb̃ãabi0 = hãabb̃i0
N (ab) = h(ab)abi
= hvk · · · v1 v1 · · · vk u1 · · · ul ul · · · u1 i0
(5.66)
= N (v1 ) . . . N (vk )N (u1 ) . . . N (ul )
= N (a)N (b).
The elements of Γp,q can thus be normalised to obtain subgroups whose kernels
are smaller than that for Γp,q .
The Pin Group

The group Pin(p, q) is defined as
Pin(p, q) = {a ∈ Γp,q | N (a) = ±1} . (5.67)
Using this definition, we have

b : Pin(p, q) → O(p, q) ,
σ (5.68)
where
b
ker σ Pin(p,q)
= {±1} = Z2 . (5.69)
′
By using the norm N , we can define the group Pinˆ(p, q) as
Pinˆ(p, q) = {a ∈ Γp,q | N ′ (a) = ±1} . (5.70)
From eqn (5.64), it can be seen that N ′ (a) = N (a) if a ∈ Cℓ+ (p, q). Consequently,
Pin(p, q) ∩ Cℓ+ (p, q) = Pinˆ(p, q) ∩ Cℓ+ (p, q). (5.71)
The Spin Group

The group Spin(p, q) is defined as
Spin(p, q) = {a ∈ Γ+
p,q | N (a) = ±1} , (5.72)
which implies that

b : Spin(p, q) → SO(p, q) ,
σ (5.73)
where
b
ker σ Spin(p,q)
= {±1} = Z2 . (5.74)
It immediately follows that
Spin(p, q) = Pin(p, q) ∩ Cℓ+

p,q . (5.75)
The definition of the group Spinˆ(p, q) using the norm N ′ is no different, since, in
this case,
Spinˆ(p, q) = Pinˆ(p, q) ∩ Cℓ+
p,q , (5.76)
and, from eqn (5.71), it follows that
Spin(p, q) = Spinˆ(p, q). (5.77)
The Reduced Pin Group and the Reduced Spin Group

The Pin+ (p, q) and the Pin+ ˆ(p, q) subgroups are respectively defined as
Pin+ (p, q) = {a ∈ Γp,q | N (a) = 1} , (5.78)
and
Pinˆ+ (p, q) = {a ∈ Γp,q | N ′ (a) = 1} ; (5.79)
the subgroup Spin+ (p, q) is defined as
Spin+ (p, q) = {a ∈ Γ+
p,q | N (a) = 1} . (5.80)
From these definitions it is straightforward to see that
Spin+ (p, q) = Pin+ (p, q) ∩ Pinˆ+ (p, q). (5.81)
Let us look at these groups in more detail. Concerning v ∈ Rp,q , eqn (5.55) asserts
that σ̂(v) = Sv is an orthogonal reflection with respect to the orthogonal hyperplane to
v. On the other hand, N (v) = v2 = g(v, v) and, if N (v) = 1, then g(v, v) = 1. Hence,
the elements of Pin+ (p, q) are such that σ̂(a) consists of the product of reflections on
hyperplanes orthogonal to vectors of type g(v, v) = 1. If T (a) is the matrix of type (5.4)
representing this orthogonal transformation, then det Dq cannot change sign under this
transformation. Consequently, according to the notation established in section 5.1, it

yields
σ̂(Pin+ (p, q)) = O+ (p, q). (5.82)
On the other hand, N ′ (v) = −v2 = −g(v, v), and N (v) = 1 implies that g(v, v) =
−1. Thus, the elements of Pinˆ+ (p, q) are such that σ̂(a) consists of the product of
reflections with respect to orthogonal hyperplanes to vectors v such that g(v, v) = −1.
In this case, the matrix T (a) representing this orthogonal transformation is given by
det Ap , which does not change sign. Consequently, according to the notation adopted
in section 5.1,
σ̂(Pinˆ+ (p, q)) = O↑ (p, q). (5.83)
Equations (5.11) and (5.81) imply that
σ̂(Spin+ (p, q)) = SO+ (p, q). (5.84)
The kernel of the mappings (5.82–5.84) is Z2 . In this case, the group Spin+ (p, q)
is said to be the two-fold covering of the group SO+ (p, q). Finally, we can write
Pin+ (p, q)/Z2 ≃ O+ (p, q)

Pinˆ+ (p, q)/Z2 ≃ O↑ (p, q) . (5.85)
Spin+ (p, q)/Z2 ≃ SO+ (p, q)
A very useful result, since it involves cases of great interest – mostly involving
important applications in physics – is the following:
Theorem 5.2 ◮ Let Cℓp,q be the Clifford algebra associated with the quadratic space
Rp,q and let Cℓ+
p,q be the associated even subalgebra. If n = p + q ≤ 5, then
Spin+ (p, q) = {R ∈ Cℓ+

p,q | RR̃ = R̃R = 1} . (5.86)
Proof: Let us first consider the case n < 5. Take x = R̂vR−1 for v ∈ V = Rp,q , and
R ∈ Cℓ+
p,q , such that RR̃ = R̃R = 1. Since R̂ = R, and R
−1
= R̃, therefore x = RvR̃.
ˆ ˜ R̃ = RvR̃ = x. Only
In addition, we have x̂ = R̂v̂R̃ = R(−v)R̃ = −x, and x̃ = R̃ṽ
vectors of Cℓp,q can satisfy the conditions x̂ = −x and x̃ = x, if n < 5, and in this
case x = RvR̃ ∈ V . In addition, N (R) = |hR̃Ri0 | = |R̃R| = 1, and R ∈ Spin+ (p, q).
For n = 5, the elements of Cℓp,q are, such that x̂ = −x, and x̃ = x, in general a sum
of 1-vectors and 5-vectors: x = u + pη, where η = e1 e2 e3 e4 e5 . Since η commutes with
all elements of Cℓp,q , namely, η ∈ Cen(Cℓp,q ), therefore x2 = u2 + p2 η 2 + 2puη. All
the terms x2 , u2 , and p2 η 2 are scalars, and uη is a 4-vector. Hence, uη = 0, and thus
either u = 0, or η = 0. If u = 0, and consequently RvR−1 = η, on the left-hand
side there is a quantity that is not in the centre of Cℓp,q , but on the right-hand side
there is a quantity that is in the centre of Cℓp,q , a result that is absurd. Hence, the
unique possibility is η = 0, and therefore x = RvR̃ ∈ V for n = 5. Since N (R) = 1,
it follows that R ∈ Spin+ (p, q). This reasoning cannot be generalised for n = 6 since,
in this case, √
a 5-vector is not an element of the centre of the algebra. For instance,
for R = (1/ 2)(e1 e2 + e3 e4 e5 e6 ) ∈ Cℓp,q , it follows that R̃R = RR̃ = 1 but that
Re1 R̃ = −e2 e3 e4 e5 e6 . ✓
We can actually obtain a slightly more general result than the one in eqn (5.86):
Pin(p, q) = {R ∈ Cℓp,q | R̂ = ±R, RR̃ = R̃R = ±1}. (5.87)
As the proof of eqn (5.87) is quite similar to that for eqn (5.86), the details of the
proof have been omitted.
Example 5.3 Let us consider the Clifford algebra Cℓ1,1 . According to eqn (5.86), the group Spin+ (1, 1)
is defined by the elements R ∈ Cℓ+
1,1 such that RR̃ = R̃R = 1. Hence,
Spin+ (1, 1) = {a + be1 e2 | a, b ∈ R, a2 − b2 = 1},

that is, Spin+ (1, 1) ≃ R ⊕ R. This group has two components, with representatives R = cosh α +
sinh αe1 e2 , and R = − cosh α + sinh αe1 e2 , respectively, −∞ < α < ∞. Therefore, the two-fold
covering Spin+ (1, 1) is not connected, although SO+ (1, 1) is connected. Indeed, the groups Spin+ (p, q)
with p + q ≥ 2 are connected, and the exception is precisely Spin+ (1, 1).
Example 5.4 The group Spin(3) is given by
Spin(3) = {R ∈ Cℓ+
3,0 |R̃R = RR̃ = 1} = {R ∈ Cℓ3,0 |R = R̂, R̃R = RR̃ = 1}.
In terms of the matrix representation of Cℓ3,0 , we have from section 3.3 that R̃R = 1 is translated for
R = ρ(R) as R† R = 1, that is, R ∈ U(2). But the condition R = R̂ is translated as R = adj(R† ), and it
is straightforward to see that, from both conditions, it follows that det R = 1. As a consequence,
Spin(3) ≃ SU(2).
The Lie Algebra of the Associated Groups

The groups that have been presented in this section are called Lie groups. The Lie
algebra associated with these groups can be identified as a subspace of the Clifford
algebra Cℓp,q , and the product of the Lie algebra – the Lie bracket – is given by the
commutator [a, b] = ab − ba, for a, b in this subspace of Cℓp,q . In fact, it is expected,
since a Clifford algebra is isomorphic to a matrix algebra, and therefore the group
of invertible elements Cℓ∗p,q is isomorphic to some subgroup in the invertible matrices
group, which is itself a Lie group. From the formalism of Lie algebras, the exponential
mapping maps an element of the Lie algebra to the Lie group component connected
to the identity of this group (Meinrenken, 2013). It is thus pertinent to consider the
definition of the exponential mapping inside a Clifford algebra. The exponential of
a ∈ Cℓp,q is defined by
X∞
an
exp a = ea = . (5.88)
n=0
n!
Since exp (−a) = (exp a)−1 , the exponential mapping links an element of Cℓp,q with
an element from the group of invertible elements Cℓ∗p,q ; namely, exp : Cℓp,q → Cℓ∗p,q .
The vector space V = Cℓp,q endowed with a product defined by [a, b] = ab − ba can be
identified as the Lie algebra associated with Cℓ∗p,q .
The Clifford–Lipschitz group Γp,q is the Lie subgroup of the Lie group Cℓ∗p,q . Its
associated Lie algebra is a vector subspace of Cℓp,q . Let us suppose that X is an element
in the Lie algebra of Γp,q , in such a way that exp (tX) is an element of Γp,q , that is,
f (t) = Ad exp(tX)(v) = exp (tX)v exp (−tX) ∈ Rp,q , ∀v ∈ Rp,q . (5.89)
Let us analyse f (t). It is straightforward to see by finite induction that
dn f (t) n
= exp (tX) ad(X) (v) exp (−tX), (5.90)
dtn
where
(ad(X) (v) = [X, v] = Xv − vX. (5.91)
Constructing a Taylor expansion to f (t) around the point t = 0, we can write
t2 2
f (t) = v + t ad(X) (v) + ad(X) (v) + · · ·
2!
X∞ n (5.92)
t n
= ad(X) (v) = exp ad(tX) (v).
n=0
n!
This expression is equivalent to

Ad exp(tX) = exp ad(tX) , (5.93)
which is a classical result in group theory.

Accordingly, f (t) ∈ Rp,q if and only if
ad(X)(v) = [X, v] = Xv − vX ∈ Rp,q . (5.94)
Let us split X in even and odd parts: X = X+ +X− , where X+ = X̂+ , and X− = −X̂− .
It is thus possible to express
Xv = X⌊v + X ∧ v = X+ ⌊v + X− ⌊v + X+ ∧ v + X− ∧ v
(5.95)
= −v⌋X+ + v⌋X− + v ∧ X+ − v ∧ X− ,
and therefore
Xv − vX = 2v⌋X+ − 2v ∧ X− . (5.96)
V V V
If
V X+p,q ∈ b (R ), then v⌋X+ ∈ b−1 (R ). In order for the expression b−1 (Rp,q ) =
p,q p,q
1 (R ) to hold, we must have the condition b = 2, which means that X+ is a 2-

vector. For the otherV cases, we must have v⌋X+ = 0, ∀v ∈ Rp,q . This equation only
holds for X+ ∈ 0 (Rp,q ) = R. The most general V expression for
V X+ is the sum of a
scalar Vand a 2-vector.VIn addition, v∧X− ∈ b+1 (Rp,q ) if X− ∈ n (Rp,q ). In this case,
when b+1 (Rp,q ) = 1 (Rp,q ) holds, we must have b = 0. However, X− is odd, and
b = 1, 3, 5, . . ., which is not possible. Therefore, we must have v ∧ X− = 0, ∀v ∈ Rp,q .
It only happens if X− is an n-vector, and X− is odd only if n is odd; in this situation,

an n-vector is an element of the centre of Cℓp,q . Hence, if n = p + q is even, then
^ ^
p,q p,q
X∈ 0 (R )⊕ 2 (R ) (5.97)
and, if n = p + q is odd,
^ ^ ^
p,q p,q p,q
X∈ 0 (R )⊕ 2 (R )⊕ n (R ). (5.98)
In either case, we can write

^
p,q
X ∈ Cen(Cℓp,q ) ⊕ 2 (R ). (5.99)
In this way, exp (tX) ∈ Γp,q .

A subgroup that is particularly important is the group Spin+ (p, q), which is the
connected component to the identity of Spin(p, q). Since R ∈ Spin+ (p, q) consists of
the even elements, it is equivalent to R̂ = R. Hence, for R = exp (tX), we must
have VX̂ = X, and therefore X must be of the form X = a + B, where a ∈ R, and
B ∈ 2 (R). Moreover, the condition N (R) = 1 implies that 1 = exp (tX̃) exp (tX) =
exp [t(a − B)] exp [t(a + B)] = exp (2ta), implying that a = 0. Hence, we can conclude
that ^
p,q
R = exp (tB) ∈ Spin+ (p, q) , B ∈ 2 (R ). (5.100)
The Lie algebra associated with the group Spin+ (p, q) consists of the vector space de-
fined by the 2-vectors and endowed with the commutator. Moreover, the commutator
of two 2-vectors is a 2-vector. In fact, if B and C are 2-vectors, then
BC = hBCi0 + hBCi2 + hBCi4 (5.101)
e = −B, C
and, since B e = −C,
^=C
(BC) eB
e = CB. (5.102)
By applying the reversion in eqn (5.101), we obtain
CB = hBCi0 − hBCi2 + hBCi4 . (5.103)
Now, by subtracting eqn (5.103) from eqn (5.103), we find that
BC − CB = [B, C] = 2hBCi2 . (5.104)

p,q
V
Hence, the commutator of 2-vectors is another 2-vector, and the algebra 2 (R ), [ , ]
is indeed the Lie algebra of Spin+ (p, q). From eqn (5.104), it is often convenient to con-
sider (1/2)[B, C] instead of the commutator [B, C]. For another approach to Clifford
algebras and Lie theory, the reader may consult, for example, the work by Meinrenken
(2013).
Example 5.5 Consider Cℓ3,0 , the Clifford algebra associated with the Euclidean space R3,0 . If {e1 , e2 , e3 }
is an orthonormal basis of R3,0 , the space of the 2-vectors has elements of the form
{B = ae1 e2 + be2 e3 + ce3 e1 | a, b, c ∈ R}.
The element exp (tX) ∈ Spin+ (3, 0) and indeed it is straightforward to realise that Spin+ (3, 0) =
Spin(3, 0). In order to further present the Lie algebra of Spin(3, 0), let us denote by Li (i = 1, 2, 3)
the objects
1 1 1
L1 = e2 e3 , L2 = e1 e3 , L3 = e1 e2 ,
2 2 2
satisfying
[L1 , L2 ] = L3 , [L2 , L3 ] = L1 , [L3 , L1 ] = L2 .
These show that the Lie algebra of Spin(3, 0) is isomorphic to the Lie algebra associated with the Lie
group SU(2). Indeed, these groups are isomorphic: Spin(3, 0) ≃ SU(2).
5.4 Conformal Transformations in Clifford Algebras

The main goal of this section is to describe the conformal transformations in the
context of the Clifford algebras. As a main application, we aim to study the conformal
transformations associated with the Minkowski spacetime R1,3 and derived from the
action of elements of the group $pin+ (2, 4) on elements of R ⊕ R4,1 , as an immediate
application of the periodicity theorem. Given a vector space V and its associated real
universal Clifford algebra Cℓ(V, g), the subspace R ⊕ V of Cℓ(V, g) is defined to be the
paravector space of V (Porteous, 1969; Maks, 1989; Baylis, 1996; da Rocha and Vaz Jr,
2007), denoted by Vπ .
Möbius Transformations on the Plane

The algebra Cℓ0,1 ≃ C is appropriate for describing rotations in R2 . Using the
periodicity theorem Cℓp+1,q+1 ≃ Cℓ1,1 ⊗ Cℓp,q , we see that the Lorentz transforma-
tions in Minkowski spacetime, generated from the action of elements of the group
$pin+ (1, 3) = {s ∈ Cℓ3,0 |ss̄ = 1} ≃ SL(2, C), are closely associated with the Möbius
transformations in the plane, since
Cℓ3,0 ≃ Cℓ2,0 ⊗ Cℓ0,1 ≃ Cℓ1,1 ⊗ Cℓ0,1 . (5.105)
The conformal transformations in the plane are described by the algebra Cℓ1,1 ≃
M(2, R). Using the isomorphism in (5.105), we can represent a paravector a ∈ R ⊕ R3
of Cℓ3,0 by an element of M(2, C):

zλ
a= ∈ R ⊕ R3 ,
µ z̄
where z, λ, µ ∈ C. Consider now an element in the group $pin+ (1, 3). The generalised
periodicity theorem asserts that, in this particular case of R3 , the reversion of a matrix
which represents an element of Cℓ3,0 is accompanied by the conjugation of each one of
its entries. It is in fact represented by
^
ac d¯ c̄
= , where a, b, c, d ∈ C. (5.106)
bd b̄ ā
Conformal Transformations in Clifford Algebras 139
The rotation of a paravector a ∈ R ⊕ R3 can be then performed by a transformation

a 7→ a′ = ηaη̃ for η ∈ $pin+ (1, 3). A rotation can be represented by
^
ac zλ ac ac zλ d¯ c̄
= . (5.107)
bd µ z̄ bd bd µ z̄ b̄ ā
When µ = 1, and λ = z z̄, it follows that the paravector a is mapped into

′ ′
ac z z z̄ d¯ c̄ z z z¯′
=ω ,
bd 1 z̄ b̄ ā 1 z¯′
where z ′ := az+c
bz+d , and ω := |bz + d| ∈ R.
2
The formalism of Möbius transformations in the plane, constructed from the gen-
eralised periodicity theorem, leads to the classical framework regarding rotations per-
formed by the group SL(2, C). It can be further generalised, to describe conformal
transformations in Minkowski spacetime, as we shall see.
Example 5.6 Consider the space R2,0 . The image of the mapping
R2,0 → Rd 0,4
(x, y) 7→ [x, y, 1, x2 + y2 ]
lies in the quadric defined by the equation x2 + y 2 − µν = 0, which is the conformal compactification
of R2,0 (Porteous, 1969, 1995), denoted by Rd 0,4 . Suppose now that a change of variables is taken into
account in order to express the quadric equation as the sum of two squares. An appropriate choice reads
x = x; y = y; z = µ − ν; and t = µ + ν. The equation of the quadric is led to x2 + y2 + z 2 − t2 , and
the image lies in the chart t = 1, under the mapping R2 → R3 given by

2x 2y 1 − x2 − y 2
(x, y) 7→ 2 2
, 2 2
, 2 2
, (5.108)
1+x +y 1+x +y 1+x +y
which image is a subset of the sphere S 2 in R3 . Indeed, the image is the sphere devoid of the south pole
(0,0,−1).
Conformal Compactification
In this section, we revisit some prominent results from the work by Porteous (1969),
Maks (1989), and Crumeyrolle (1990). Given the quadratic space Rp,q , consider the
injective mapping given by
κ : Rp,q → Rp+1,q+1
x 7→ κ(x) = (x, x · x, 1) = (x, λ, µ) . (5.109)
The image of Rp,q is a subset of the quadric Q ֒→ Rp+1,q+1 , described by the equation:
x · x − λµ = 0 . (5.110)
This quadric is a high-dimensional generalisation of the quadric equation and is called

the Klein absolute. The mapping κ induces an injective mapping from Q into the
projective space RPp+1,q+1 . In addition, Q is compact and is defined as the conformal
compactification R dp,q of Rp,q . (Here, the hat denotes the conformal compactification,
instead of the grade involution, in order to preserve the usual notation).

It has been shown that Q ≈ R dp,q is homeomorphic to (S p × S q )/Z (Porteous,

2
1969). In the particular case where p = 0, and q = n, the quadric is homeomorphic
to the n-sphere S n , seen as the compactification of Rn by the addition of a point at
infinity.
There also exists an injective mapping
s : R ⊕ R3 → R ⊕ R4,1 ,

v vv̄
v 7→ s(v) = . (5.111)
1 v̄
The following theorem was introduced by Porteous (1969, 1995); also see Maks (1989):
Theorem 5.3 ◮ (i) The mapping κ : Rp,q → Rp+1,q+1 ; x 7→ (x, x · x, 1) is an isometry.

(ii) The mapping π : Q → Rp,q ; (x, λ, µ) 7→ x/µ is defined where λ 6= 0 is conformal.
(iii) If U : Rp+1,q+1 → Rp+1,q+1 is an orthogonal mapping, the map Ω = π ◦ U ◦ κ :
Rp,q → Rp,q is conformal.
The mapping Ω maps conformal spheres onto conformal spheres, which can be either
quasi-spheres or hyperplanes. A quasi-sphere is a submanifold of Rp,q , defined by the
equation a x · x + b · x + c = 0, where a, c ∈ R, and b ∈ Rp,q . A quasi-sphere is a sphere
when a quadratic form g in Rp,q is positive defined, and a 6= 0. Otherwise, a quasi-
sphere is a plane when a = 0. From assertion (iii) of theorem 5.3, we see that both U
and −U induce the very same conformal transformation in Rp,q . The conformal group
is defined as
Conf(p, q) ≃ O(p + 1, q + 1)/Z2 , (5.112)
where O(p + 1, q + 1) has four components and, in the Minkowski spacetime case,
where p = 1, and q = 3, the group Conf(1,3) has four components (Porteous, 1995;
Anglès, 2008). The component of Conf(1, 3) connected to the identity, denoted by
Conf+ (1, 3), is known as the Möbius group of R1,3 . In addition, SConf+ (1,3) denotes
the time-preserving and future-pointing component connected to the identity.
Möbius Transformations in Minkowski Spacetime

Let the group $pin+ (2, 4) be defined as
$pin+ (2, 4) = {g ∈ Cℓ4,1 | gḡ = 1, gbg −1 ⊂ R ⊕ R4,1 , ∀b ∈ R ⊕ R4,1 }. (5.113)

ac
The matrix g = is in the group $pin+ (2, 4) if, and only if, its entries a, b, c, d ∈
bd
Cℓ3,0 satisfy the following conditions (Maks, 1989):
(i) aā, bb̄, cc̄, dd¯ ∈ R,
(ii) ab̄, cd¯ ∈ R ⊕ R3 ,
(iii) avc̄ + cv̄ā, cv d¯ + dv̄c̄ ∈ R, ∀v ∈ R ⊕ R3 ,
(iv) av d¯ + cv̄ b̄ ∈ R ⊕ R3 , ∀v ∈ R ⊕ R3 ,
(v) ac̃ = cã, bd˜ = db̃,
(vi) ad˜ − cb̃ = 1. (5.114)
Conformal Transformations in Clifford Algebras 141
Conditions (i), (ii), (iii), and (iv) are equivalent to the condition σ̂(g)(b) := gbg̃ ∈
R ⊕ R4,1 , ∀b ∈ R ⊕ R4,1 (Ahlford, 1986; Fillmore and Springer, 1990) where σ̂ :
$pin+ (2, 4) → SO+ (2, 4) denotes the twisted adjoint representation. Indeed,

ac xλ d¯ c̄
gbg̃ =
bd µ x̄ b̄ ā

axd + λab̄ + µcd¯ + cx̄b̄ axc̄ + λaā + µcc̄ + bxā
¯
=
bxd¯ + λdb̄ + µdd¯ + dx̄b̄ bxc̄ + λbā + µdb̄ + dx̄ā

w λ′
= ∈ R ⊕ R4,1 , (5.115)
µ′ w̄
where the last equality (considering w ∈ R ⊕ R3 and λ′ , µ′ ∈ R) comes from the

requirement that g ∈ $pin+ (2, 4), that is, gbg̃ ∈ R ⊕ R4,1 . If these conditions are
required, (i), (ii), (iii), and (iv) follow from the results given by Vahlen (1902) and
Ahlford (1986). More details can be seen in the work by Maks (1989) and Fillmore
and Springer (1990).
Conditions (v) and (vi) lead to gḡ = 1 since, for all g ∈ $pin+ (2, 4), we have

ad˜ − cb̃ ac̃ − cã 10
gḡ = 1 ⇔ = . (5.116)
bd˜ − db̃ dã − bc̃ 01
Conformal Transformations
The paravector b ∈ R ⊕ R4,1 ֒→ Cℓ4,1 is represented as

x xx̄ xλ
= , (5.117)
1 x̄ µ x̄
where x ∈ R ⊕ R3 is a paravector of Cℓ3,0 .

Consider an element of the group $pin+ (2, 4). It is
possible to represent it as an
ac
element g ∈ Cℓ4,1 ≃ Cℓ1,1 ⊗Cℓ3,0 , that is, g = , where a, b, c, d ∈ Cℓ3,0 . The
bd
rotation of b ∈ R ⊕ R 4,1
֒→ Cℓ4,1 is performed by the use of the twisted adjoint
representation σ̂ : $pin+ (2, 4) ⇒ SO+ (2, 4), defined as
σ̂(g)(b) = gbĝ −1
= gbg̃, g ∈ $pin+ (2, 4). (5.118)
Using the matrix representation, the action of $pin+ (2,4) is given by

^
ac xλ ac ac xλ d¯ c̄
= . (5.119)
bd µ x̄ bd bd µ x̄ b̄ ā
By fixing µ = 1, the paravector b is mapped on
′ ′
ac x xx̄ d¯ c̄ x x x̄′
=∆ , (5.120)
bd 1 x̄ b̄ ā 1 x̄′
where
x′ := (ax + c)(bx + d)−1 , ∆ := (bx + d)(bx + d) ∈ R. (5.121)
The transformation (5.121) is conformal (Vahlen, 1902; Lawson and Michelson, 1990;
Hestenes, 1991).
From the isomorphisms
Cℓ4,1 ≃ C ⊗ Cℓ1,3 ≃ M(4, C), (5.122)
elements of $pin+ (2,4) are elements of the Dirac algebra C ⊗ Cℓ1,3 . From eqn (6.192),
we denote x ∈ R ⊕ R3 a paravector. The conformal mappings are expressed by the
action of $pin+ (2,4), via the matrices given in table 5.1 (Vahlen, 1902; Porteous, 1969;
Maks, 1989; Hestenes, 1991).
Table 5.1 Representation of the Conformal Mappings
Explicit Transformation Matrix of $pin+ (2, 4)

1h
Translation x 7→ x + h, h ∈ R ⊕ R3 0 1
√
ρ 0
Dilation x 7→ ρx, ρ ∈ R 0 1/ ρ
√

g0
Rotation x 7→ gxĝ−1 , g ∈ $pin+ (1, 3) 0 ĝ

0 −1
Inversion x 7→ −x 1 0

1 0
Transvection x 7→ x + x(hx + 1)−1 , h ∈ R⊕ R3 h1
This index-free geometric formulation makes it possible to trivially generalise the con-
formal mappings of R1,3 to the ones of Rp,q , if the periodicity theorem of Clifford
algebras is used.
The group SConf+ (1,3) is fourfold covered by SU(2, 2) (Laufer, 1997), and the
identity element idSConf + (1,3) of the group SConf + (1, 3) corresponds to the following
elements of SU(2, 2) ≃ $pin+ (2,4):

12 0 −12 0 i2 0 −i2 0
, , , . (5.123)
0 12 0 −12 0 i2 0 −i2
The element 12 denotes the 2 × 2 identity matrix, and i2 denotes the matrix diag(i, i).
In this way, elements of $pin+ (2,4) generate the orthochronous Möbius transfor-
mations. The isomorphisms
Conf(1, 3) ≃ O(2, 4)/Z2 ≃ Pin(2, 4)/{±1, ±i} (5.124)
are constructed as in the work by Porteous (1969); consequently,
SConf + (1, 3) ≃ SO+ (2, 4)/Z2 ≃ $pin+ (2, 4)/{±1, ±i}. (5.125)
The homomorphisms
2−1 2−1
$pin+ (2, 4) −→ SO+ (2, 4) −→ SConf + (1, 3) (5.126)
are explicitly constructed in the work by Klotz (1974) and Laufer (1997).
The Lie Algebra of the Conformal Group

V2 2,4
The Lie algebra of $pin+ (2,4) is generated by (R ), which has dimension 15. Since
dim Conf(1,3) = 15, the relation between these groups is investigated now. In chapter
4, example 4.6, we introduced the isomorphisms
Eµ = −iγµ (µ = 0, 1, 2, 3), E4 = −iγ0123 , (5.127)
and
EA = εA ε5 , (5.128)
where {εÅ }5Å=0 is basis of R2,4 ; {EA }4A=0 is basis of R4,1 ; and {γµ }3µ=0 is basis of R1,3 .
V
The generators of Conf(1,3), as elements of 2 (R2,4 ), are defined as follows:
i 1
Pµ = (εµ ε5 + εµ ε4 ), D = − ε4 ε5 , (5.129)
2 2
i i
Kµ = − (εµ ε5 − εµ ε4 ), Mµν = εν εµ . (5.130)
2 2
From the relations in (5.127) and (5.128), the generators of Conf(1,3) are expressed
from the {γµ } ∈ Cℓ1,3 as
1 1
Pµ = (γµ + iγµ γ5 ), D= iγ5 , (5.131)
2 2
1 1
Kµ = − (γµ − iγµ γ5 ), Mµν = (γν ∧ γµ ). (5.132)
2 2
They satisfy the following relations:
[Pµ , Pν ] = 0, [Kµ , Kν ] = 0, [Mµν , D] = 0, (5.133)

[Mµν , Pλ ] = −(gµλ Pν − gνλ Pµ ), (5.134)
[Mµν , Kλ ] = −(gµλ Kν − gνλ Kµ ), (5.135)
[Mµν , Mσρ ] = gµρ Mνσ + gνσ Mµρ − gµσ Mνρ − gνρ Mµσ , (5.136)
[Pµ , Kν ] = 2(gµν D − Mµν ), [Pµ , D] = Pµ , [Kµ , D] = −Kµ . (5.137)
These commutation relations are invariant under Pµ 7→ −Kµ ; Kµ 7→ −Pµ ; and D 7→

−D. In chapter 6, we shall apply what has been developed here.

There are some good texts for those who want to go deeply into the material discussed
in this chapter but keep connection to the theory of Clifford algebras: the classical
groups are regarded in the book by Porteous (1995); a detailed study of the theory of
Lie groups and Lie algebras is given by Meinreken (2013); and a detailed study of the
conformal group, by Anglès (2008). Problems related to the construction of Clifford
algebras and of Spin groups on differentiable manifolds are examined, for example, in
the book by Lawson and Michelson (1990).
5.6 Exercises
(1) Consider the spacetime algebra Cℓ3,1 ≃ M(4, R) generated by the basis {e1 , e2 , e3 , e4 }
such that e2i = 1, e24 = −1. Show that the matrix

1 1 + e1 e4 −e1 + e4
M=
2 e1 + e4 1 − e1 e4
can be written as the composition of a transvection, a translation, and again a transvec-
tion. In addition, show that
1 1 e1
√
2 −e1 1
can be written as the product of a transvection, a dilatation, and a translation.
(2) Show that, in an Euclidean space Rn , the conditions ab̃, b̃d, dc̃, c̃a ∈ Rn , and
¯ dc,
āb, bd, ¯ cā ∈ Rn , are equivalent, where a, b, c, d ∈ Cℓn .
(3) Define the Clifford and the exterior exponential mappings, respectively, by
1 2 1
eA = 1 + A + A + A3 + · · · , A ∈ Cℓp,q
2! 3!
1 1
e∧A = 1 + A + A ∧ A + A ∧ A ∧ A +··· .
2! 3!
V2
Given B ∈ (Rp,q ), show that eB ∈ Spin+ (p, q) and that e∧B ∈ Γp,q .
(4) Define the groups
Pin± (n, C) = {g ∈ Cℓ(Cn ) | N (g) = ±1, gvĝ −1 ∈ Cn } ,

Spin± (n, C) = {g ∈ Cℓ+ (Cn ) | N (g) = ±1, gvĝ −1 ∈ Cn } .
Show that those groups are not connected when n = 1. In other words, show that
Pin+ (1, C) = {±1, ±ie1 } and that Spin+ (1, C) = {±1}.
(5) Show that Spin+ (p, q) ≃ Spin+ (q, p) and that
Spin(2) ≃ U(1), Spin+ (1, 1) ≃ R ⊕ R,

Spin(3) ≃ SU(2), Spin+ (1, 2) ≃ SL(2, R),
Spin+ (4) ≃ SU(2) × SU(2), Spin+ (1, 3) ≃ SL(2, C),
Spin+ (2, 2) ≃ SL(2, R) × SL(2, R), Spin(5) ≃ Sp(2, H),
Spin+ (1, 4) ≃ Sp(1, 1, H), Spin+ (2, 3) ≃ Sp(4, R).
6
Spinors
In this chapter, we introduce and discuss the theory of spinors as well as some of
their prominent applications. Three different definitions of spinors and their properties
are presented. Pure spinors are also introduced and the triality principle, twistors,
and Penrose flagpoles are discussed. A detailed study of the so-called Weyl spinors,
which are the basis of the Penrose and Rindler formalism (Penrose, 1967; Penrose and
Rindler, 1984; da Rocha and Vaz Jr, 2007), concludes this chapter. The comparison
between the Clifford algebraic framework and the van der Waerden notation is left to
the appendix.
First, we must assert that there is not a unique definition of spinor in the literature.
Although the differences between the definitions are small, they do exist. Perhaps
the origin of these differences is the fact that the theory of spinors was developed
independently by physicists and mathematicians. To try to clarify this situation, we
begin with a brief discussion about these differences, to pave the way for the definitions
and the developments to be presented in this chapter.
6.1 The Babel of Spinors

There are essentially three different definitions of spinors, each one emphasising a
different point of view. Two of them are well established, whereas the third one is
gradually becoming known in the literature. We classify these three definitions as (i)
classical, (ii) algebraic, and (iii) operatorial. The ‘well-established’ definitions are the
classical and the algebraic ones. A comparative study of these different definitions
in some situations of interest in physics is presented in the article by Figueiredo, de
Oliveira, and Rodrigues Jr (1990).
In physics, spinors effectively emerged as a result of the study of quantum me-
chanics, specifically, from the Pauli theory (1926), about non-relativistic quantum
mechanics (Pauli, 1927), and Dirac’s (1928) theory regarding relativistic quantum me-
chanics (Bethe and Salpeter, 1957). The first confusion arises when we naively think
that because of these origins, the spinor is necessarily and closely related to the spin
of a particle, in the same way that the electron. Spinors had already appeared be-
fore in physics, although within the context of the classical mechanics, precisely, in
the context of rigid-body dynamics. In a standard classical mechanics textbook like
that by Goldstein, Poole, and Safko (2001), we can find a detailed discussion about
the so-called Cayley–Klein parameters, which are widely used to describe spatial ro-
tations. These parameters are in fact entries of a 2 × 2 unitary matrix A ∈ SU(2). The
group SU(2) is well known to be the double covering of the special orthogonal group
SO(3). The representation space of the group SU(2) is obviously C2 . Such elements
146 Spinors
of C2 are called spinors – and in our classification they are called classical spinors.
However, there is no relationship between this name and the emergence of spinors in
classical mechanics. The adjective ‘classical’ suggests that it was because of the classi-
cal approach that these objects appeared initially in physics and mathematics as well.
Spinors, seen as elements of C2 , were largely used by Pauli to describe the behaviour
of an electron according to quantum mechanics, taking into account the spin of this
electron (which is not the case in the Schrödinger theory). In the literature of quantum
mechanics, such objects are known as Pauli spinors.
The essential fact here is that the group SU(2) is isomorphic to the group Spin(3) =
Spin(3, 0). From the isomorphism SU(2) ≃ Spin(3), we conclude that C2 is the rep-
resentation space of the group Spin(3). Hence, a Pauli spinor is an element of the
representation space of the group Spin(3), which is the spin 1/2 representation of
the group of three-dimensional spatial rotations (Goldstein, Poole, and Safko, 2001).
Moreover, the group SO(3) is the representation of spin 1 regarding these rotations,
with Spin(3) ≃ SU(2), and Spin(3)/Z2 ≃ SO(3).
Let us now analyse the relativistic case. Spacetime rotations are described by el-
ements of the group SO+ (1, 3), the so-called orthochronous proper Lorentz group –
remembering that SO+ (1, 3) ≃ SO↑ (1, 3) ≃ SO↑+ (1, 3) – or, simply, the Lorentz group.
A similar role is played here by the group SL(2, C) of 2 × 2 complex matrices and de-
terminant equal to 1. The group SL(2, C) is the double covering of SO+ (1, 3). We have
here the isomorphism SL(2, C) ≃ Spin+ (1, 3). Obviously, the representation space of
SL(2, C) is C2 . However, here the situation is quite different. Indeed, there are two
representations of SL(2, C) that are not equivalent: we can define ρ(A) and ρ̄(A) for
A ∈ SL(2, C) as ρ(A)(z) = Az, and ρ̄(A)(z) = Āz, where Ā is the complex conjugate
matrix associated with A, and z ∈ C2 . These two representations should be equiva-
lent if there exists an isomorphism φ : C2 → C2 such that ρ̄(A) = φ ◦ ρ(A) ◦ φ−1 .
In other words, those two representations should be equivalent if there exists an in-
vertible 2 × 2 complex matrix T such that ρ̄(A)T = T ρ(A). By explicit computation,
we can show that ρ̄(A)T = T ρ(A) does not have solution for A ∈ SL(2, C) – on the
other hand, ρ̄(A)T = T ρ(A) does have a solution for A ∈ SU(2). Hence, there are
two inequivalent representations of SL(2, C), denoted by D(1/2,0) and D(0,1/2) , respec-
tively. The elements of a space that carries each one of such representations are called
Weyl spinors. As in the previous case, such spinors are elements of the representation
space of a Spin group, that is, Spin+ (1, 3) ≃ SL(2, C), and they fit into what we call
the classical definition of a spinor. Moreover, the so-called Dirac spinors, according
to the classical definition, are elements of C4 and carry a reducible representation of
Spin+ (1, 3) ≃ SL(2, C) composed of the sum of two Weyl spinors, each one correspond-
ing to one of the irreducible representations D(1/2,0) and D(0,1/2) .
To summarise, from the classical point of view, spinors can be defined as objects
which carry an irreducible representation of the Spin group, which is the double cov-
ering of the special orthogonal group and therefore the spin 1/2 representation of the
group of rotations in a quadratic space.
On the other hand, we already discussed the Spin group in a Clifford algebra.
Hence, another definition of spinor can be introduced: the algebraic one. When we
discussed the representations of a Clifford algebra (section 4.3), the representation
space associated with an irreducible regular representation was shown to be a minimal
The Babel of Spinors 147
left ideal related to the Clifford algebra. An algebraic spinor is an element of a minimal
left ideal in a Clifford algebra. The representation of the Clifford algebra obtained is
called a spinor representation.
There is a little variation when the algebraic definition of spinors is regarded.
We know that a Clifford algebra is isomorphic either to a simple algebra or to the
direct sum which is a semisimple algebra – of two simple algebras. In the latter case,
there is a minimal left ideal associated with each simple algebra. Let us denote such
ideals by I1 and I2 . In this case, we have two inequivalent irreducible representations:
one associated with the representation space I1 , and the other associated with I2 .
According to the definition, elements in these ideals I1 and I2 are spinors. Some authors
call elements of I1 and I2 semispinors, reserving the term spinors for the elements of
I1 ⊕ I2 . Hence, the two inequivalent irreducible representations are called semispinor
representations. Increasing the amount of confusion, some authors use the term pinors
for the elements of a minimal left ideal of a Clifford algebra and use the term spinors
for the elements of a minimal left ideal of the even subalgebra/ associated with this
Clifford algebra.
A spinor representation of a Clifford algebra naturally induces a representation in
any subset, by simply restricting to the elements of this subset the left multiplication
on the ideal. An irreducible representation of the Clifford algebra induces therefore
an irreducible representation of the Clifford–Lipschitz group – the irreducibility arises
from the fact that non-isotropic vectors generate the Clifford algebra and the Clifford
group. Moreover, it also induces a representation of the even subalgebra related to this
Clifford algebra. The question is whether this representation is reducible or irreducible.
In what follows, we will see that, if the even subalgebra representation is reducible,
then it consists of a sum of two irreducible representations.
Similarly, an irreducible representation of the even subalgebra induces a representa-
tion in its subsets. In particular, an irreducible representation of the even subalgebra
induces an irreducible representation of the Spin group. In this point, the contact
between the classical and algebraic definitions of a spinor is accomplished. The repre-
sentation space of the Spin group – whose elements are classical spinors – is a minimal
left ideal of the even subalgebra. This minimal left ideal of the even subalgebra is not
necessarily a minimal ideal of the whole Clifford algebra. If a primitive idempotent of
the even subalgebra is also a primitive in the whole Clifford algebra, then the ideals are
similar. However, it does not hold in general, and the algebraic and classical definitions
do differ.
Another possible definition of spinor, which is denominated operatorial, can be
introduced from another representation – distinct from the regular representation – of
a Clifford algebra. This representation uses the representation space associated with
the even subalgebra of a Clifford algebra. Although this definition seems to be distinct
from the algebraic and classical definitions, it is equivalent to these in most cases, in
particular in the cases of great interest for physical applications.
Because these definitions are so different, we must be careful to clearly state which
one which we are using. The best way to do this is to attribute different names to
each case: algebraic spinors, classical spinors, and operator spinors. These adjectives
are not standard, but it seems extremely appropriate to use them in order not to be
lost in this Babel of spinors.
148 Spinors
6.2 Algebraic Spinors
Definition 6.1 ◮ An element of a minimal left ideal associated with a Clifford algebra
Cℓ(V, g) is said to be an algebraic spinor if Cℓ(V, g) is a simple algebra, and an algebraic
semispinor if Cℓ(V, g) is semisimple.
Given this definition, the algebraic spinors can be now identified according to
the classification of the Clifford algebras. For a simple Clifford algebra, we have the
isomorphism Cℓp,q ≃ M(N, K), and a minimal left ideal of Cℓp,q is isomorphic to KN ,
which is used for algebraic spinor classification. In the case of a semisimple Clifford
algebra, we have the isomorphism Cℓp,q ≃ M(N, K)⊕M(N, K). In this case, a minimal
left ideal of Cℓp,q is isomorphic to KN . The algebraic semispinors classification follows
from this isomorphism. The sum of algebraic semispinors is called an algebraic spinor
– although, for this case, the ideal is not minimal. For a semisimple Clifford algebra,
an algebraic spinor can be classified as KN ⊕ KN .
Table 6.1 shows the algebraic spinor classification.
p − q = 0, 2 mod 8 In this case, Cℓp,q ≃ M(2[n/2] , R). An algebraic spinor is thus
[n/2]
an element of a minimal left ideal isomorphic to R2 .
p − q = 4, 6 mod 8 Here, the isomorphism Cℓp,q ≃ M(2[n/2]−1 , H) holds. The
[n/2]−1
space of algebraic spinors is isomorphic to H2 .
p − q = 3, 7 mod 8 In this case, Cℓp,q ≃ M(2[n/2] , C). Hence, the space of alge-
[n/2]
braic spinors is isomorphic to the representation space C2 . The possibility p − q =
3, 7 mod 8 holds only if the dimension n = p + q is odd. In this case, the pseudoscalar
(n-vector) η commutes with all elements of Cℓp,q and satisfies η 2 = −1, defining a
complex structure.
p − q = 5 mod 8 For this metric signature, the Clifford algebra Cℓp,q is semisim-
ple, and Cℓp,q ≃ M(2[n/2]−1 , H) ⊕ M(2[n/2]−1 , H). This is a situation where alge-
[n/2]−1
braic semispinors exist. The space of algebraic semispinors is isomorphic to H2 ,
[n/2]−1 [n/2]−1
whereas the space of algebraic spinors is isomorphic to H2 ⊕ H2 . For
p − q = 5 mod 8, a n-vector η commutes with all elements of Cℓp,q and satisfies
η 2 = 1. We can thus write (see chapter 3, exercise 4) Cℓp,q = + Cℓp,q ⊕ − Cℓp,q , where
± Cℓp,q ≃ M(2
[n/2]−1
, H).
p − q = 1 mod 8 In this case, the Clifford algebra Cℓp,q is semisimple, and Cℓp,q ≃
M(2[n/2] , R) ⊕ M(2[n/2] , R). The space of the algebraic semispinors is isomorphic to
[n/2] [n/2] [n/2]
R2 , and the space of the algebraic spinors is isomorphic to R2 ⊕ R2 . For
p − q = 1 mod 8, an n-vector η commutes with all the elements of Cℓp,q and is such
that η 2 = 1. In this case we can write (see again chapter 3, Exercise 4) Cℓp,q =
+ Cℓp,q ⊕ − Cℓp,q , where ± Cℓp,q ≃ M(2
[n/2]
, R).
This analysis is summarised in table 6.1.
For the case involving complex Clifford algebras, the situation is much simpler
than that for reals. When dim V = n is an even number, we have the complex Clifford
Classical Spinors 149
algebra isomorphism CℓC (2k) ≃ M(2k , C). Hence, the space of algebraic spinors is
k
isomorphic to C2 . For dim V = n odd, the isomorphism CℓC (2k + 1) ≃ M(2k , C) ⊕
M(2 , C) holds for the complex Clifford algebra. The space of the algebraic semispinors
k
k
is then isomorphic to C2 , and the space of algebraic spinors in this case is isomorphic
k k
to C2 ⊕ C2 as summarised in table 6.2.
Example 6.1 Some of the most outstanding and important examples in physics regard dimensions 3 and 4.
For the quadratic space R3,0 , we have the Clifford algebra Cℓ3,0 ≃ M(2, C). The space of algebraic spinors
of Cℓ3,0 is isomorphic to C2 . For the quadratic space R0,3 , we have Cℓ0,3 ≃ H ⊕ H. Algebraic semispinors
are elements of a space isomorphic to H. For the quadratic space R1,3 , we have Cℓ1,3 ≃ M(2, H) and
therefore the space of the algebraic spinors is H2 . In the case of R3,1 , we have Cℓ3,1 ≃ M(4, R), and
the space of the algebraic spinors is provided by R4 . For complex Clifford algebras, we have CℓC (3) ≃
M(2, C) ⊕ M(2, C), and the space of algebraic semispinors is C2 . Regarding CℓC (4) ≃ M(4, C), the
algebraic spinors space is given by C4 .
6.3 Classical Spinors
Definition 6.2 ◮ Consider Rp,q a quadratic space, the Clifford algebra Cℓp,q associated
with this space, and the reduced Spin group Spin+ (p, q), associated with Cℓp,q . An
element of the irreducible representation space of Spin+ (p, q) is said to be a classical
spinor.
The group Spin+ (p, q) = {a ∈ Γ+p,q | N (a) = 1} is the set of even elements of the
Clifford–Lipschitz group. Then, an irreducible representation of Spin+ (p, q) descends
from an irreducible representation of the even subalgebra Cℓ+ p,q . On the other hand,
Table 6.1 Algebraic Spinors Classification: The Real Case, Where

p + q = n, and [n/2] Denotes the Integer Part of n/2
p−q
0 1 2 3
mod 8
[n/2] [n/2] [n/2] [n/2] [n/2]
Sp,q
A
R2 R2 ⊕ R2 R2 C2
p−q
4 5 6 7
mod 8
[n/2]−1 [n/2]−1 [n/2]−1 [n/2]−1 [n/2]
Sp,q
A
H2 H2 ⊕ H2 H2 C2
Table 6.2 Algebraic Spinors: The Complex Case
k
n = 2k C2
k k
n = 2k + 1 C2 ⊕ C2
150 Spinors
when we discussed the classification of the Clifford algebras, an important result was
established: Cℓ+ + +
p,q ≃ Cℓq,p−1 ≃ Cℓp,q−1 ≃ Cℓq,p . An irreducible representation of Cℓp,q ≃
Cℓ+q,p is thus obtained from an irreducible representation of Cℓq,p−1 ≃ Cℓp,q−1 , as
was previously established. In other words, a classical spinor in a quadratic space
Rp,q or Rq,p is an algebraic spinor (or an algebraic semispinor) in a quadratic space
that is either Rq,p−1 or Rp,q−1 . In order to make our analysis easier, let us use such
isomorphisms to construct a classification of the even Clifford algebras, as shown in
table 6.3.
We can now describe the classification of classical spinors in what follows.
p − q = 1, 7 mod 8 For p − q = 1, 7 mod 8, it follows that Cℓ+

p,q ≃ Cℓp,q−1 =
Cℓp′ ,q′ , where p′ − q ′ = p − q + 1 = 0, 2 mod 8, namely, Cℓ+
p,q ≃ M(2
[(n−1)/2]
, R),
where n = p + q. Hence, a classical spinor is an element of the representation space
[(n−1)/2]
R2 .
p − q = 2, 6 mod 8 In this case, p′ − q ′ = p − q + 1 = 3, 7 mod 8, yielding Cℓ+

p,q ≃
[(n−1)/2]
M(2[(n−1)/2] , C). A classical spinor is therefore an element of C2 . In this case,
the n-vector η defines a complex structure in the classical spinors space. Hence, there
are two inequivalent irreducible representations: one of them with η equals the complex
structure induced by i, whereas the other is achieved when η is equal to the complex
structure induced by −i. The two corresponding classical spinors are conjugate.
p − q = 3, 5 mod 8 Here, p′ −q ′ = 4, 6 mod 8, and then Cℓ+

p,q ≃ M(2
[(n−1)/2]−1
, H).
[(n−1)/2]−1
A classical spinor is an element of H2 .
p − q = 4 mod 8 We have p′ − q ′ = 5 mod 8. The even subalgebra is semisim-

ple, and Cℓp,q ≃ M(2[(n−1)/2]−1 , H) ⊕ M(2[(n−1)/2]−1 , H). There are two inequiva-
lent representations of Spin+ (p, q). In this case, we can write Cℓ+ + +
p,q = + Cℓp,q ⊕ − Cℓp,q
and denominate as positive classical spinors the elements of the representation space
Table 6.3 Real Even Subalgebra Classification Table, Where p + q = n, and

[κ] Denotes the Integer Part of κ = (n − 1)/2
p−q
0 1 2 3
mod 8
M(2[κ] , R)
Cℓ+
p,q ⊕ M(2[κ] , R) M(2[κ] , C) M(2[κ]−1 , H)
M(2[κ] , R)
p−q
4 5 6 7
mod 8
M(2[κ]−1 , H)
Cℓ+
p,q ⊕ M(2[κ]−1 , H) M(2[κ] , C) M(2[κ] , R)
M(2[κ]−1 , H)
Classical Spinors 151
of + Cℓ+ +
p,q , and the elements of the representation space of − Cℓp,q as negative clas-
sical spinors. A classical spinor, positive or negative, in this case is an element of
[(n−1)/2]−1
H2 .
p − q = 0 mod 8 In this case, p′ − q ′ = 1 mod 8. The even subalgebra Cℓ+
p,q is
semisimple, and Cℓ+ p,q ≃ M(2
[(n−1)/2]
, R) ⊕ M(2[(n−1)/2] , R). There are two inequiv-
alent representations of Spin+ (p, q). As in the previous case, we can write Cℓ+ p,q =
+ +
+ Cℓp,q ⊕ − Cℓp,q and denominate as positive classical spinors the elements of the repre-
sentation space of + Cℓ+ p,q , and as negative classical spinors the elements of the repre-
sentation space − Cℓ+p,q A positive or negative classical spinor in this case is an element
.
[(n−1)/2]
of R2 .
This analysis is summarised in table 6.4.
For the case involving complex Clifford algebras, the situation is again straight-
forward (see Table 6.5). Clearly, Cℓ+C
(n) ≃ CℓC (n − 1). Hence, if dim V = n is even,
it follows that Cℓ+C (2k) = CℓC (2k − 1) ≃ M(2
k−1
, C) ⊕ M(2k−1 , C). There are two
inequivalent irreducible representations, and the classical positive or negative spinors
k−1
are elements of C2 . If dim V = n is odd, we have Cℓ+ C
(2k+1) ≃ CℓC (2k) ≃ M(2k , C).
2k
The classical spinors are therefore elements of C .
Regarding the space of classical spinors Sp,q , an idempotent endomorphism R
∈ End(Sp,q ) can be defined, which for some dimensions and signatures is usually identi-
fied with the volume element η (Lazaroiu, Babalic, and Coman, 2013; Bonora, de Brito,
Table 6.4 Classical Spinors Classification: The Real Case, Where

p + q = n, and [n/2] Denotes the Integer Part of n/2
p−q
0 1 2 3
mod 8
[(n−1)/2]
R2
[(n−1)/2] [(n−1)/2] [(n−1)/2]−1
Sp,q
C
⊕ R2 C2 H2
[(n−1)/2]
R2
p−q
4 5 6 7
mod 8
[(n−1)/2]−1
H2
[(n−1)/2]−1 [(n−1)/2] [(n−1)/2]
Sp,q
C
⊕ H2 C2 R2
[(n−1)/2]−1
H2
Table 6.5 Classical Spinors: The Complex Case
k−1 k−1
n = 2k C2 ⊕ C2
k
n = 2k + 1 C2
152 Spinors
and da Rocha, 2015). Spin projectors are then defined by Π± = 21 (I ± R), where I de-
+ −
notes the identity operator on Sp,q , providing the direct sum Sp,q = Sp,q ⊕ Sp,q , where
± ±
Sp,q = Π± (Sp,q ). Elements of Sp,q are called (symplectic) Majorana–Weyl spinors
+
when p − q = 0 mod 8 (p − q = 4 mod 8), whereas elements of Sp,q are known as
(symplectic) Majorana spinors when p − q = 7 mod 8 (p − q = 6 mod 8). More de-
tails shall be discussed in section 6.6 and can be verified in the article by de Andrade,
Rojas, and Toppan (2001).
6.4 Spinor Operators

Given a Clifford algebra Cℓp,q , the Z2 -grading can be taken into account to use the
even subalgebra Cℓ+p,q as a representation space for the algebra Cℓp,q . In other words,
we can define a representation ρ : Cℓp,q → End(Cℓ+ p,q ), which is called a graded regular
representation. An arbitrary element a ∈ Cℓp,q can be split as a = a+ + a− , where
1
a± = (a ± â). (6.1)
2
Let us further split the representation ρ as ρ = ρ+ + ρ− , such that
ρ(a) = ρ+ (a+ ) + ρ− (a− ). (6.2)
The part ρ+ is a regular representation of a+ in Cℓ+

p,q , namely,
ρ+ (a+ )(φ) = a+ φ, ∀φ ∈ Cℓ+

p,q . (6.3)
On the other hand, for a− ∈ Cℓ− − +

p,q , we have a− φ ∈ Cℓp,q , for any φ ∈ Cℓp,q . In order
to define ρ− (a− ) we must have ρ− (a− )(φ) ∈ Cℓ+ p,q . It is possible if an odd element κ
is taken into the following definition
ρ− (a− )(φ) = a− φκ, ∀φ ∈ Cℓ+

p,q . (6.4)
If we choose κ such that

κ2 = 1, κ ∈ Cℓ−
p,q , (6.5)
then ρ = ρ+ + ρ− is a representation of Cℓp,q .
In order to see that ρ is a representation of Cℓp,q , let us calculate ρ(ab) for a, b ∈
Cℓp,q :
ρ(ab) = ρ(a+ b+ + a+ b− + a− b+ + a− b− )
(6.6)
= ρ+ (a+ b+ ) + ρ− (a+ b− ) + ρ− (a− b+ ) + ρ+ (a− b− ).
Moreover, each part in the above sum can be further analysed:
ρ+ (a+ b+ )(φ) = a+ b+ φ = ρ+ (a+ )ρ+ (b+ )(φ),

ρ− (a+ b− )(φ) = a+ b− φκ = ρ+ (a+ )ρ− (b− )(φ),
(6.7)
ρ− (a− b+ )(φ) = a− b+ φκ = ρ− (a− )ρ+ (b+ )(φ),
ρ+ (a− b− )(φ) = a− b− φ = a− b− φκ2 = ρ− (a− )ρ− (b− )(φ),
Spinor Operators 153
which therefore proves that ρ is indeed a representation:
ρ(ab) = ρ(a)ρ(b). (6.8)
The definition of the irreducible regular representation depends upon the existence
of an odd element κ such that κ2 = 1. In two cases, this element does not exist: when
Cℓ0,1 and Cℓ0,2 , which are exactly the cases Cℓ0,1 ≃ C and Cℓ0,2 ≃ H. For the other
cases, we can find an element κ.
Now we want to know whether the representation ρ is reducible or not. Suppose
that there exists elements η, ǫ ∈ Cℓ+
p,q such that
η 2 = ±1, ηκ = κη, (6.9)

2
ǫ = 1, ǫκ = κǫ, ǫη = −ηǫ. (6.10)
In this case, we can write

Cℓ+
p,q =
+
Cℓ+ − +
p,q ⊕ Cℓp,q , (6.11)
where
± 1 +
Cℓ+
p,q = [Cℓ ±η Cℓ+
p,q η
−1
]. (6.12)
2 p,q
±
Hence, for arbitrary elements φ± ∈ Cℓ+
p,q , we have
ηφ± = ±φ± η. (6.13)
These subspaces consist of Cℓ+

p,q elements which either commute or anti-commute with
η. Since
±
Cℓ+
p,q
±
Cℓ+
p,q ⊂
+
Cℓ+
p,q ,
±
Cℓ+
p,q
∓
Cℓ+
p,q ⊂
∓
Cℓ+
p,q ,
+
we can see that only Cℓ+ +
p,q is a subalgebra of Cℓp,q . In other words, an involution
aη = ηaη−1 (6.14)
can be defined (note that (aη )η = a) where an element a such that aη = a is said to
be η-even, and aη = −a is said to be η-odd, where + Cℓ+ p,q is an η-even subalgebra of
+
Cℓp,q . Note that ǫ is η-odd.
If there exists η, ǫ ∈ Cℓ+ p,q satisfying these conditions, we can define a representation
ρ : Cℓp,q → End( + Cℓ+ p,q ). Given a ∈ Cℓp,q , we can express a = + a+ + − a+ + + a− + − a− ,
where
± 1
a = (a ± ηaη −1 ), (6.15)
2
and a± are given by eqn (6.1). Now let us write ρ as
ρ = + ρ+ + − ρ+ + + ρ− + − ρ− ,
in such a way that

+
ρ(a) = ρ+ ( + a+ ) + −
ρ+ ( − a+ ) + +
ρ− ( + a− ) + −
ρ− ( − a− ), (6.16)
154 Spinors
+
where, for φ+ ∈ Cℓ+
p,q , it follows that
+
ρ+ ( + a+ )(φ+ ) = +
a+ φ + ,
− − −
ρ+ ( a+ )(φ+ ) = a+ φ+ ǫ,
+ + +
(6.17)
ρ− ( a− )(φ+ ) = a− φ+ κ,
−
ρ− ( − a− )(φ+ ) = −
a− φ+ κǫ.
We can see that ± ρ± ( ± a± )(φ+ ) ∈ + Cℓ+ p,q and is in fact a representation, namely,
ρ(ab) = ρ(a)ρ(b).
If there still exist other odd elements and η-even elements η ′ and ǫ′ such that
(η ) = ±1, and (ǫ′ )2 = 1, and η ′ ǫ′ = −ǫ′ η ′ , and which furthermore commute with
′ 2
κ, η, and ǫ, we can construct a new subalgebra ++ Cℓ+ p,q which is invariant under the
action of ρ in a completely similar way.
When even elements satisfying such conditions do not exist, we then arrive at an
irreducible representation. The space which carries this graded irreducible represen-
tation is a subalgebra of Cℓp,q and is called spinor algebra. The spinor algebra is a
subalgebra of the even subalgebra, and, in some cases, it can be the even subalgebra
itself. Meanwhile, it is worth emphasising that the subalgebras + Cℓ+ p,q or
++
Cℓ+
p,q are
not in general Clifford algebras (as we will see with an example in what follows).
Definition 6.3 ◮ An element of the graded irreducible representation space of Cℓp,q

is said to be a spinor operator.
Example 6.2 Let us consider the case Cℓ3,0 . Let {e1 , e2 , e3 } be an orthonormal basis in such a way that
(ei )2 = 1, and ei ej = −ej ei (i 6= j). An odd element such that κ2 = 1 is, for instance, κ = e3 .
In this case, there are no even elements (besides the scalar 1) such that η 2 = 1. The even subalgebra
Cℓ+
3,0 ≃ Cℓ0,2 ≃ H is therefore an irreducible representation space of Cℓ3,0 . Spinor operators are elements
of Cℓ+
3,0 , reading a + a12 e1 e2 + a13 e1 e3 + a23 e2 e3 . These are the so-called Pauli spinor operators.
Example 6.3 Another interesting example, particularly for physical applications, is the algebra Cℓ1,3 .
Let {e0 , e1 , e2 , e3 } be an orthonormal basis which satisfies (e0 )2 = 1; (ei )2 = −1 (i = 1, 2, 3); and
eµ eν = −eν eµ (µ 6= ν). An odd element satisfying κ2 = 1 is given by κ = e0 . The unique even elements
that squared equals 1 are e0 ei (i = 1, 2, 3). However, these elements do not commute with κ. The even
subalgebra Cℓ+ 1,3 ≃ Cℓ3,0 is then an irreducible representation space of Cℓ1,3 , and its elements are called
Dirac spinor operators.
Example 6.4 Let us now consider Cℓ4,1 with (e0 )2 = −1; (ea )2 = 1 (a = 1, 2, 3, 4); and eµ eν = −eν eµ
(µ 6= ν). We can choose κ as, for instance, κ = e4 . Here, there already exist even elements of type η and ǫ
commuting with κ; for example η = e1 e2 , ǫ = e0 e1 , in such a way that η 2 = −1; ǫ2 = 1; ηǫ = −ǫη. We
can further verify that there is no other element of this same type. Hence,+ Cℓ+ 4,1 ≃ Cℓ3,0 is an irreducible
representation space of Cℓ4,1 .
Example 6.5 Consider the Clifford algebra Cℓ2,1 . The element κ can be taken as κ = e1 e2 e3 , since κ is
odd, and κ2 = 1. In this way, we can take Cℓ+ 2,1 as being the representation space. This representation
is not irreducible. Indeed, there are elements ǫ = e1 e3 , and η = e2 e3 , which commute with κ and such
that ηǫ = −ǫη, and ǫ2 = 1, besides the relation η 2 = 1. Thus, the subalgebra + Cℓ+ 2,1 carries a graded
irreducible representation of Cℓ2,1 . In addition, we have + Cℓ+
2,1 ≃ R ⊕ R. On the other hand, there exists
another possible choice for η that also defines an irreducible representation. In fact, let ή = e1 e2 be the
element that commutes with κ and anti-commutes with ǫ, and ή 2 = −1. We have another irreducible
representation carried by the subalgebra 0 Cℓ+
2,1 , where we used another index to improve the notation.
The interesting fact to be mentioned here is that such representations are not equivalent. Indeed, we can
see that 0 Cℓ+
2,1 ≃ C.
Example 6.6 Consider now the algebra Cℓ2,2 . In this case, we can choose κ = e1 ; ǫ = e2 e3 ; and η = e3 e4 .
The subalgebra + Cℓ+
2,2 hence carries an irreducible graded representation of Cℓ2,2 . The interesting fact
here is that + Cℓ+
2,2 ≃ R ⊕ R ⊕ R ⊕ R, namely,
+ +
Cℓ2,2 is not a Clifford algebra, an observation which can
be straightforwardly verified from the classification table. There is still another possible choice given by κ
and ǫ, and where ή = e3 e4 ; this choice leads us to a inequivalent representation in terms of the subalgebra
0 +
Cℓ2,2 ≃ C ⊗ C ≃ C ⊕ C. This algebra is not a real Clifford algebra, although it is the complex Clifford
algebra CℓC (1) ≃ C ⊕ C.
Example 6.7 The graded irreducible representation of a complex Clifford algebra is relatively easy to study.
Let us consider, for example, Cℓ2k (C) and let us choose κ = e1 ; ǫj = ie2j e2j+1 ; and ηj = e2j−1 e2j for
2k
j = 1, . . . , k − 1. We then obtain +···+
Cℓ+
2k (C) = C ⊕ C ⊕ · · · ⊕ C = C⊕ .
| {z }
2k times
A Generalised Spinor Algebra

A detailed study concerning Z2 -gradings in Clifford algebras can be seen in the article
by Mosna, Miralles, and Vaz Jr (2003). Let us denote an arbitrary Z2 -grading of
Cℓ(V, g) by Cℓ0 ⊕ Cℓ1 and let α be a vector space isomorphism defined by α|Cℓi =
(−1)i idCℓi , where idCℓi is the identity map on Cℓi (i = 0, 1). We say that Cℓ0 and
Cℓ1 are the α-even and α-odd parts, respectively, of Cℓ(V, g). For the usual Z2 -grading
given by Cℓ(V, g) = Cℓ+ ⊕ Cℓ− , we have α = #.
A grading automorphism α is said to preserve the multivector structure of Cℓ(V, g)
Vk Vk
if α( (V )) ⊂ (V ). In the article by Mosna, Miralles, and Vaz Jr (2003), it is proven
that, if Cℓ
V p,q p,q (R) = Cℓ0p,q ⊕ Cℓ1p,q is a Z2 -grading preserving the multivector structure
of (R ), then
Cℓ0p,q ≃ Cℓp0 ,q0 ⊗ Cℓ+
p−p0 ,q−q0 , (6.18)
where p0 (q0 ) is the number of α-even elements of an orthonormal basis of Rp,q squaring
to +1(−1).
Given Cℓ0p,q , we can use it to define a graded regular representation, as we did
with Cℓ+ i
p,q , that is, we write a = a0 + a1 , with ai ∈ Cℓp,q (i = 0, 1) and define a
representation ρ = ρ0 + ρ1 such that
ρ0 (a0 )(φ) = a0 φ, ρ1 (a1 )(φ) = a1 φκ, (6.19)
where φ ∈ Cℓ0p,q and κ ∈ Cℓ1p,q such that κ2 = 1. Moreover, if there exist elements
η, ǫ ∈ Cℓ0p,q satisfying relations as in eqns (6.9) and (6.10), the representation space
can be reduced to 0 Cℓ0p,q , and so on.
The question now is how to actually find the spinor algebra using Cℓ0p,q as in
eqn (6.18); in order to do so, we need to find the α grading automorphism. First,
let us denote by {ei , fk } (i = 1, . . . , p, k = 1, . . . , q) an orthonormal basis of V in
such a way that Cℓp,q is generated by {ei , fk } (and 1) with (ei )2 = 1; (fk )2 = −1; and
156 Spinors
ei fk +fk ei = 0. Let us suppose that Cℓp0 ,q0 is generated by {e1 , . . . , ep0 , f1 , . . . , fq0 } and
that Cℓp−p0 ,q−q0 is generated by {ep0 +1 , . . . , ep , fq0 +1 , . . . , fq }. Now we remember that,
if V = span{v1 , . . . , vn }, then v ∈ V if and only if v∧ΩV = 0, where ΩV = v1 ∧· · ·∧vn
is the pseudoscalar of V . Let us denote the pseudoscalars of Rp,q , Rp0 ,q0 , and Rp−p0 ,q−q0
by I, Ω, and Θ, respectively, that is,
I = e1 · · · ep f1 · · · fq ,
Ω = e1 · · · ep0 f1 · · · fq0 , (6.20)
Θ = ep0 +1 · · · ep fq0 +1 · · · fq ,
where we used the fact that {ei , fk } is an orthonormal basis. Note that
I = (−1)(p−p0 )q0 ΩΘ. (6.21)
The condition that v ∈ Rp0 ,q0 is equivalent to v ∧ Ω = 0 (Mosna, Miralles, and Vaz Jr,
2003), which can be written as
b = 0,
vΩ + Ωv (6.22)
yields
b −1 = −ΩvΩ
v = −ΩvΩ b −1 . (6.23)
This result suggests the definition of α as
b −1 ,
α(v) = −ΩvΩ (6.24)
in such a way that v ∈ Rp0 ,q0 if and only if α(v) = v. We can extend this definition
to φ ∈ Cℓp,q in such a way to satisfy α(vu) = α(v)α(u) as either
b −1 ,
α(φ) = ΩφΩ b = Ω,
if Ω (6.25)
or
α(φ) = ΩφΩ−1 , if b = −Ω.
Ω (6.26)
Hence, if φ ∈ Cℓp0 ,q0 , then we have α(φ) = φ.
Similarly, we have that u ∈ Rp−p0 ,q−q0 if and only if u ∧ Θ = 0, which can be
written as
u = Θb
uΘb −1 , (6.27)
and generalised to ψ ∈ Cℓp−p0 ,q−q0 as either
b −1 ,
ψ = ΘψΘ if b = Θ,
Θ (6.28)
or
ψ = ΘψΘ−1 , if b = −Θ.
Θ (6.29)
Let us see now what happens in the four different situations according to whether
the pseudoscalars are even or odd elements.
b = Θ. Then, for ψ ∈ Cℓp−p ,q−q , we have

(1) Let us suppose that Θ 0 0
b −1 = Ω−1 I ψI
ψ = ΘψΘ b −1 Ω = ΩI ψI
b −1 Ω−1 , (6.30)
where we used eqn (6.21). We must also distinguish two other cases.
b
b = Ω. In this case, I is an even element, and IψI −1 = ψ.
(a) Let us suppose that Ω
Then, (
b −1 ,
ΩψΩ if ψb = ψ,
−1
ψ = ΩψΩ = (6.31)
−ΩψΩ , if ψb = −ψ,
b −1
that is, (
α(ψ), if ψb = ψ,
ψ= (6.32)
−α(ψ), if ψb = −ψ.
b = −Ω. In this case, I is an odd element, and IψI −1 = ψ.

(b) Let us suppose that Ω
Accordingly, (
b −1 = ΩψΩ ,
−1
if ψb = ψ,
ψ = ΩψΩ (6.33)
−ΩψΩ−1 , if ψb = −ψ,
that is (
ψ= (6.34)
−α(ψ), if ψb = −ψ.
b = −Θ. Then, for ψ ∈ Cℓp−p ,q−q we have

(2) Let us suppose that Θ 0 0
ψ = ΘψΘ−1 = Ω−1 IψI −1 Ω = ΩIψI −1 Ω−1 , (6.35)
where we used eqn (6.21). We must also distinguish two other cases.
b = Ω. In this case, I is an odd element, and IψI −1 = ψ.

(a) Let us suppose that Ω
Then, (
ΩψΩb −1 , if ψb = ψ,
−1
ψ = ΩψΩ = (6.36)
−ΩψΩ , if ψb = −ψ,
b −1
that is, (
ψ= (6.37)
−α(ψ), if ψb = −ψ.
b
b = −Ω. In this case, I is an even element, and IψI −1 = ψ.
(b) Let us suppose that Ω
Thus, we have (
ΩψΩb −1 , if ψb = ψ,
ψ = ΩψΩ−1 = b −1 , if ψb = −ψ, (6.38)
−ΩψΩ
that is,
158 Spinors
(
ψ= (6.39)
−α(ψ), if ψb = −ψ.
Then, from all the four cases, we conclude that
Cℓ+ 0
p−p0 ,q−q0 ∈ Cℓp,q , Cℓ− 1
p−p0 ,q−q0 ∈ Cℓp,q , (6.40)
in relation to the α grading automorphism. Therefore, in order to use eqn (6.18) to

define a generalised spinor algebra, we define the α grading automorphism using either
eqn (6.25) or eqn (6.26), depending upon whether Ω is even or odd, respectively.
Example 6.8 Let us consider Cℓ1,1 . We have that Cℓ+ 1,1 ≃ Cℓ1,0 , and the spinor algebra obtained using
the usual Z2 -grading is R ⊕ R. However, if we use the α grading automorphism with Ω = e2 ((e2 )2 = −1)
in eqn (6.26), that is, α(ψ) = −e2 ψe2 , we obtain that Cℓ01,1 ≃ C. In this case, we can choose either
κ = e1 , or κ = e1 e2 .
Example 6.9 Consider the case Cℓ3,0 . We saw that Cℓ+ 3,0 ≃ H is an irreducible representation space of
Cℓ3,0 . Consider now Ω = v, where v is a unit vector. Then, α(φ) = vφv. We choose κ = u, where u
is an unit vector orthogonal to v. We have that Cℓ03,0 is generated by {1, v, iv, i}, where i = e1 e2 e3 ,
and Cℓ03,0 ≃ C ⊕ C. We can also choose Ω = uv, with u and v orthogonal to each other. Then,
b
α(ψ) = uvψvu. In this case, we choose κ = iuv. The spinor algebra Cℓ0 in this case is generated by
3,0
{1, u, v, uv} and Cℓ03,0 ≃ Cℓ2,0 ≃ M(2, R).
Example 6.10 We noticed that H ⊕ H and M(2, C) are irreducible representation spaces of Cℓ4,1 by using
the usual Z2 -grading. If we use the α grading, we can obtain another spinor algebra, that is, C ⊕ C ⊕ C ⊕ C,
which is not a Clifford algebra. In fact, we can define α(φ) = e1 φe1 , and κ = e2 . But we can also define an-
other grading as α′ (φ) = e1 e2 e3 φe3 e2 e1 , with κ′ = e3 e5 . Note that κκ′ = κ′ κ. Then, the space 0 Cℓ04,1
of α-even and α′ -even elements is generated by {1, e1 , e2 e3 , e1 e2 e3 , e4 e5 , e1 e4 e5 , e2 e3 e4 e5 , e1 e2 e3 e4 e5 },
and it is not difficult to see that 0 Cℓ04,1 = C ⊕ C ⊕ C ⊕ C.
Further applications and development of the α grading (6.24) can be found in the
article by da Rocha and Vaz Jr (2006c).
6.5 A Comparison of the Different Definitions of Spinors

A question that naturally arises now regards the relationship among the different defi-
nitions of spinors. Let us first consider the case of real Clifford algebras. Unfortunately,
we do not have a classification table for the spinor operators for real Clifford algebras,
as it would make it easy to compare the different definitions. In fact, for spinor opera-
tors, we have even cases with different and non-equivalent spinor algebras, and we were
not able to design a table that could take into account all possible cases. Thus, we will
study the problem for each case up to dimension 5, leaving for the interested readers
the analysis for dimensions higher than that. The spinor operators for dimensions 1–5
can be found in table 6.6.1
We can easily see that, given Cℓp,q , algebraic spinors, classical spinors, and spinor
operators are in general different. However, in most of the cases, all these different
1 The notation A | B in the S O column in table 6.6 means that A uses the usual parity automor-
p,q
phism, and B uses the α grading automorphism.
A Comparison of the Different Definitions of Spinors 159
spinor spaces are isomorphic from the point of view of vector spaces. In fact, given
Cℓp,q , we see that, when p − q = 0, 2, 3, 4 mod 8, the dimensions of these spaces are
equal. This case includes the distinctive example of the three-dimensional Euclidean
space. When p − q = 1, 5 mod 8, if we consider the space of algebraic semispinors
instead of the space of algebraic spinors, we see that the dimensions of the spaces are
also equal. However, when p − q = 6, 7 mod 8, the dimension of the space of classical
spinors is half that of the algebraic spinors and the spinor operators. The space of
spinor operators has therefore the same dimension as the space of algebraic spinors or
the space of algebraic semispinors, when the latter is defined.
A C
Table 6.6 Algebraic Spinors (Sp,q ), Classical Spinor s (Sp,q )
O
and Spinor Operators (Sp,q ) for Cℓp,q , with p + q ≤ 5 (Where
4
A⊕ = A ⊕ A ⊕ A ⊕ A)
Sp,q
A
Sp,q
C
Sp,q
O
Cℓ1,0 R⊕R R R
Cℓ0,1 C R C
Cℓ2,0 R 2
C C|R⊕R
Cℓ1,1 R 2
R⊕R R⊕R| C
Cℓ0,2 H C H
Cℓ3,0 C2 H H | C ⊕ C; M(2, R)
Cℓ2,1 R ⊕R
2 2
R 2
C; R ⊕ R
Cℓ1,2 C 2
R 2
M(2, R) | C ⊕ C; H
Cℓ0,3 H⊕H H H
Cℓ4,0 H2 H⊕H H ⊕ H | M(2, C)
4
Cℓ3,1 R4 C2 C ⊕ C | R⊕
4
Cℓ2,2 R4 R2 ⊕ R2 R⊕ ; C ⊕ C
Cℓ1,3 H2 C2 M(2, C) | H ⊕ H
Cℓ0,4 H 2
H⊕H H ⊕ H | M(2, C)
Cℓ5,0 H ⊕H
2 2
H 2
M(2, C) | H ⊕ H
4
Cℓ4,1 C4 H2 H ⊕ H; M(2, C) | C⊕
4
Cℓ3,2 R4 ⊕ R4 R4 C ⊕ C; R⊕
4
Cℓ2,3 C4 R4 M(2, C) | C⊕
4
Cℓ1,4 H2 ⊕ H2 H2 H ⊕ H; M(2, C) | C⊕
4
Cℓ0,5 C4 H2 H ⊕ H; M(2, C) | C⊕
160 Spinors
There is a suitable way to understand this relation between spinor operators and
algebraic spinors or semispinors. First, let us remember that a minimal left ideal of
Cℓp,q is of the form Cℓp,q f , where f is given in eqn (4.63), that is, we have
A
Sp,q = Cℓp,q f, (6.41)
where f has the form
1 1
f= (1 + eI1 ) · · · (1 + eIk ), (6.42)
2 2
and {eI1 , . . . , eIk } is a set of Cℓp,q elements which commute with each other and such
that (eIα )2 = 1 (α = 1, . . . , k). Moreover, k = q − rq−p , where rj are the Radon–
Hurwitz numbers. However, note that
f = eIα f, α = 1, . . . , k. (6.43)
Since eI1 , . . . , eIk commute with each other, and (eIα )2 = 1, we can choose κ = ǫ, and
ǫ, ǫ′ , and so on in the definition of a graded representation as the elements eI1 , . . . , eIk .
Let us take, for example, κ = eI1 . Then, Cℓp,q = Cℓ0p,q ⊕ Cℓ1p,q , and κ = eI1 ∈ Cℓ1p,q .
But then
Cℓp,q f = (Cℓ0p,q ⊕ Cℓ1p,q )f = Cℓ0p,q f ⊕ Cℓ1p,q f = Cℓ0p,q f ⊕ Cℓ1p,q eI1 f, (6.44)
and since
Cℓ1p,q eI1 ∈ Cℓ0p,q , (6.45)
we have
Cℓp,q f ⊂ Cℓ0p,q f. (6.46)
Next, we choose ǫ = eI2 , and a grading such that Cℓ0p,q = 0
Cℓ0p,q ⊕ 1
Cℓ0p,q , with eI2 ∈
1
Cℓ0p,q . Since 1 Cℓ0p,q eI2 ∈ 0 Cℓ0p,q , we have
Cℓp,q f ⊂ 0 Cℓ0p,q f. (6.47)

We can continue with this procedure up to the last element eIk , to obtain
0···0 (0)k−1
Cℓp,q f ⊂ Cℓ0p,q f = Cℓ0p,q f. (6.48)
Nevertheless, dim Cℓ0p,q = (dim Cℓp,q )/2; dim 0 Cℓ0p,q = (dim Cℓp,q )/(2 · 2); and so on; in
addition,
k−1 dim Cℓp,q
dim (0) Cℓ0p,q = . (6.49)
2k
Since dim Cℓp,q = 2n , and the dimension of Cℓp,q f is 2n−k , we can conclude that
(0)k−1
Cℓp,q f = Cℓ0p,q f. (6.50)
Thus, from the point of view of vector spaces, we have
(0)k−1
Cℓp,q f ≃ Cℓ0p,q . (6.51)
V
Let Λ be the pseudoscalar of Cℓp,q . We leave it as an exercise to show that Λ2 = 1

when p−q = 0, 1 mod 4. The semispinors φ± ∈ Sp,q ±
are eigenvalues of Λ when Λ2 = 1,
that is, Λφ± = ±φ± for ψ ± ∈ Sp,q

±
. But, when p − q = 1 mod 4, the volume element
Λ ∈ Cen(Cℓp,q ) (Λ is an odd element), and then Λφ± = φ± Λ. Consequently, we can
write the semispinors as
1
φ± = φ(1 ± Λ), (6.52)
2
where φ ∈ Sp,q , and
1 k−1 1
±
Sp,q = Sp,q (1 ± Λ) = ( (0) Cℓ0p,q )f (1 ± Λ). (6.53)
2 2
Now we consider a grading such that
k−1 k−1
(0)k
( (0) Cℓ0p,q )1 Λ ⊂ ( (0) Cℓ0p,q )0 = Cℓ0p,q , (6.54)
and then
(0)k
(Cℓp,q f )± ≃ Cℓ0p,q . (6.55)
V
As we see, this situation happens when p −q = 1 mod 4, or p− q = 1, 5 mod 8, which
was the condition we started with.
The case of complex Clifford algebras is much simpler than that of the reals. The
comparison of algebraic spinors, classical spinors, and spinor operators for complex
Clifford algebras is given in table 6.7. Note that, when n is even, all spaces have the
same dimension but, when n is odd, the classical spinor and the spinor operator spaces
have the same dimension as the space of the algebraic semispinor.
When comparing the different spinor definitions, one may be tempted to ask
whether any of the definitions is better than the others. This kind of question is
meaningful only if we clearly specify the aspects of the concept we want to examine
or emphasise. In this sense, when we are interested in calculations with spinors, the
concept of spinor operators proves to be very useful, since the representation space is
an algebra. This fact is particularly clear in those cases where the spinor algebra is a
Clifford algebra.
There are two cases where the spinor algebra is a Clifford algebra, and which have
very important physical applications: (i) the Clifford algebra of the three-dimensional
Euclidean space Cℓ3,0 , and (ii) the Clifford algebra of the four-dimensional Minkowski
spacetime Cℓ3,1 . The concept of the spinor operator proves to be very useful in these
cases.
Example 6.11 In order to discuss the case of the three-dimensional Euclidean space, let us establish some
notation, that is, Ψ ∈ Cℓ3,0 f (where f is a primitive idempotent); |Ψi ∈ C2 ; Ψ+ ∈ Cℓ+ 0
3,0 ; Ψ0 ∈ Cℓ3,0 ;
Table 6.7 Algebraic Spinors (SnA ), Classical Spinors

(SnC ), and Spinor Operators (SnO ) for Complex Clifford
Algebras Cℓn (C)
SnA SnC SnO

k k−1 k−1 2k
Cℓ2k (C) C2 C2 ⊕ C2 C⊕
k k k 2k
Cℓ2k+1 (C) C2 ⊕ C2 C2 C⊕
162 Spinors
and Ψ ∈ H. The Clifford algebra Cℓ3,0 is generated by {e1 , e2 , e3 }, and σi = ρ(ei ) (i = 1, 2, 3) are the
Pauli matrices, as seen in example 3.2. The pseudoscalar of Cℓ3,0 is I = e1 e2 e3 .
Let ψ ∈ Cℓ3,0 , where
ψ = s + v1 e1 + v2 e2 + v3 e3 + b12 e12 + b13 e13 + b23 e23 + t123 e123 ,
and where e12 = e1 e2 , and so on, as usual. The matrix representation of ψ, obtained by using the
idempotent f = 21 (1 + e3 ), is

(s + v3 ) + i(t123 + b12 ) (v1 − b13 ) + i(b23 − v2 )
ρ(ψ) = .
(v1 + t13 ) + i(b23 + v2 ) (s − v3 ) + i(t123 − b12 )
The algebraic spinor Ψ = ψf reads
Ψ = [(s + v3 ) + I(b12 + t123 )]f + [(v1 + b13 ) + I(v2 + b23 )]e1 f
= (w1 + Iw2 )f + (w3 + Iw4 )e1 f.
Its matrix representation is provided by

(s + v3 ) + i(t123 + b12 ) 0 w1 + iw2 0
ρ(Ψ) = = .
(v1 + t13 ) + i(b23 + v2 ) 0 w3 + iw4 0
A
The space S3,0 is clearly isomorphic as a vector space to C2 , with the identification

w1 + iw2
Ψ = (w1 + Iw2 )f + (w3 + Iw4 )e1 f ↔ = |Ψi.
w3 + iw4
The spinor operator Ψ+ ∈ Cℓ+

3,0 has the form
Ψ+ = s + b12 e12 + b13 e13 + b23 e23 .

Let us conveniently rewrite the coefficients of Ψ+ as
Ψ+ = w1 + w2 e12 + w3 e13 + w4 e23 .
The matrix representation of Ψ+ is given by

w1 + iw2 −w3 + iw4
ρ(Ψ+ ) = ,
w3 + iw4 w1 − iw2
which shows the vector space isomorphism Cℓ+ 2 +

3,0 ≃ Cℓ3,0 f ≃ C . We also know that Cℓ3,0 ≃ Cℓ0,2 ≃ H,
which can be made explicitly by the identification
e12 ↔ i, e23 ↔ j, e13 ↔ k,
that is,
Ψ = w1 + w2 i + w4 j + w3 k.
For the spinor algebra Cℓ03,0 , we define the α grading by α(ψ) = vψv, where v is a unit vector and we
choose κ = u, where u is a unit vector orthogonal to v. Specifically, we choose v = e1 , and u = e3 .
Then, the spinor operator Ψ0 ∈ Cℓ03,0 has the form
Ψ0 = s + v1 e1 + b23 e23 + t123 e123 .
Let us also conveniently rewrite the coefficients of Ψ0 as
Ψ0 = w1 + w2 e123 + w3 e1 + w4 e23 .
It is represented by
w1 + iw2 w3 + iw4
ρ(Ψ0 ) = ,
w3 + iw4 w1 + iw2
revealing the vector space isomorphism of Cℓ03,0 with these spinor spaces. Note that, as an algebra,
Cℓ03,0 ≃ C ⊕ C, with the identification i ↔ e1 e2 e3 and (1, 0) ↔ 1, (0, 1) ↔ e1 .
It is important for physical applications to consider bilinear quantities constructed from spinors. In
terms of |Ψi ∈ C2 , these quantities are given by
1
σ = hΨ|Ψi, Ji = hΨ|σi |Ψi, Sij = hΨ| [σi , σj ]|Ψi, ω = hΨ|σ1 σ2 σ3 |Ψi,
2
with i, j = 1, 2, 3. It follows that

σ = w12 + w22 + w32 + w42 ,
J1 = 2(w1 w4 + w2 w3 ),
J2 = 2(w1 w3 − w2 w4 ),
J3 = w12 + w22 − w32 − w42 .
The transformation (w1 , w2 , w3 , w4 ) 7→ (J1 , J2 , J3 ) is known as the Kustaanheimo–Stiefel transformation.
Now, let us see how to write the bilinear quantities using the other spinor definitions. Clearly, it is
enough to consider Ji (i = 1, 2, 3). In terms of Ψ ∈ Cℓ3,0 f , we first note that Ψ e ∈ f Cℓ3,0 (a right
minimal ideal). The matrix representation of Ψ e is

e = (s + v3 ) − i(t123 + b12 ) (v1 + b13 ) − i(b23 + v2 ) = w1 − iw2 w3 − iw4 .
ρ(Ψ)
0 0 0 0
Then, we can identify

e = f (w1 − Iw2 ) + f e1 (w3 − Iw4 ) ↔ w1 − iw2 w3 − iw4 = hΨ|.
Ψ
Moreover, since Tr[ρ(f )] = 2hf i0 = 1, we have that
e i Ψi0 = 2hΨΨe
Ji = 2hΨe e i i0 ,
where we used hABi0 = hBAi0 . But Ψ = Ψf , with f = (1/2)(1 + e3 ), and then

e i i0 .
Ji = hΨ(1 + e3 )Ψe
Moreover, for spinor operators, we also have
e + ei i0 ,
Ji = hΨ+ (1 + e3 )Ψ e 0 e i i0 .
Ji = hΨ0 (1 + e3 )Ψ
It is worth noting that the expression for Ji takes a very interesting and simple form using Ψ+ . In
fact, we have
Ji = hΨ+ Ψ e + ei i0 + hΨ+ e3 Ψ
e + e i i0 .
− e −
However, ei ∈ Cℓ , and then Ψ+ Ψ+ ei ∈ Cℓ , which gives hΨ+ Ψ e + ei i0 = 0, since scalar elements
3,0 3,0
belong to Cℓ+
p,q . Hence,
e + ei i0 = hΨ+ e3 Ψ
Ji = hΨ+ e3 Ψ e + i1 · ei .
Let us denote
J = Ψ+ e3 Ψe +.
Note that Je = J; J b = −J; and J̄ = −J. The only Cℓ3,0 elements which that satisfy these properties are
vectors, that is, J = hJi1 . Then,
Ji = J · ei .
The spinor operator Ψ+ itself has a simple and interesting interpretation. First, we note that
e + = w2 + w2 + w2 + w2 = σ ≥ 0.
Ψ+ Ψ 1 2 3 4
Then, we can write √

Ψ+ = σR,
where R ∈ Cℓ+ e
3,0 and such that RR = 1. But this means that R ∈ Spin(3) (see eqn (5.86)). The operation
Re3 R̃ therefore represents a rotation of the vector e3 . Consequently,
e + = σRe3 R̃
J = Ψ + e3 Ψ
can be interpreted as giving the vector J as the result of the composition of a rotation of the vector e3 and
a dilation by a factor σ (Vaz, 2013). This is an amazing interpretation for a spinor in the three-dimensional
Euclidean space. For an example of an application of this result, see the article by Vaz Jr (2013).
Example 6.12 Let us consider now the four-dimensional Minkowski spacetime with signature (1, 3). This
is a case where the dimension of the space of classical spinors is half of the dimensions of the space of
algebraic spinors and spinor operators. Moreover, this case discussed in section 6.1; in it, there are two
164 Spinors
non-equivalent representations of the spacetime rotations in terms of 2 × 2 complex matrices, an idea

which takes us to the concept of Weyl spinors. Since this is a very important subject, its discussion from
the Clifford algebra point of view will be addressed later in sections 6.10 and 6.11. Let us now focus our
attention on algebraic spinors and spinor operators.
Let us use a notation similar to that used in example 6.11, that is, Ψ ∈ Cℓ1,3 f (where f is a primitive
idempotent); |Ψi ∈ H2 ; Ψ+ ∈ Cℓ+ 0 2
1,3 ; Ψ0 ∈ Cℓ1,3 ; and Ψ ∈ C . The Clifford algebra Cℓ1,3 is generated
by {e0 , e1 , e2 , e3 } such that (e0 ) = −(e1 ) = −(e2 ) = −(e3 )2 = 1. Let ψ ∈ Cℓ1,3 be given as
2 2 2
ψ = s + v0 e0 + v1 e1 + v2 e2 + v3 e3 + b01 e01 + b02 e02 + b03 e03 + b12 e12 + b13 e13 + b23 e23
+ t012 e012 + t013 e013 + t023 e023 + t123 e123 + q0123 e0123 ,
whose matrix representation using f = (1/2)(1 + e0 ) is (see chapter 4, exercise 1 of chapter 4)

pr
ρ(ψ) = ,
q s
where
p = (s + v0 ) + i(b23 + t023 ) + j(−b13 − b013 ) + k(b12 + t012 ),
q = (q0123 − t123 ) + i(v1 − b01 ) + j(v2 − b02 ) + k(v3 − b03 ),
r = (−q0123 − t123 ) + i(v1 + b01 ) + j(v2 + b02 ) + k(v3 + b03 ),
s = (s − v0 ) + i(b23 − t023 ) + j(t013 − b13 ) + k(b12 − t012 ).
The algebraic spinor Ψ ∈ Cℓ1,3 f is
Ψ = [(s + v0 ) + (b12 + t012 )e12 + (b13 + b013 )e13 + (b23 + t023 )e23 ]f
+ [(q0123 − t123 ) + (v3 − b03 )e12 + (−v2 + b02 )e13 + (v1 − b01 )e23 ]e0123 f,
and its matrix representation is

p0 p0 + p1 i + p2 j + p3 k 0
ρ(Ψ) = = .
q 0 q0 + q1 i + q2 j + q3 k 0
A
The vector space isomorphism between S1,3 and H2 is provided by identifying
i ↔ e23 , j ↔ e31 , k ↔ e12 , |1i ↔ f, |2i ↔ e0 e1 e2 e3 f.
For the spinor operator Ψ+ , we have
Ψ+ = s + b01 e01 + b02 e02 + b03 e03 + b12 e12 + b13 e13 + b23 e23 + q0123 e0123 .
We conveniently rewrite the coefficients of Ψ+ as
Ψ+ = p0 − q1 e01 − q2 e02 − q3 e03 + p3 e12 − p2 e13 + p1 e23 + q0 e0123 .
Then, its matrix representation is
p −q
ρ(Ψ+ ) = .
q p
Note that as an algebra we have Cℓ+ 1,3 ≃ Cℓ3,0 ≃ M(2, C).
Besides the spinor operator Ψ+ , we can define a spinor operator Ψ0 using the α grading automorphism
defined by α(ψ) = e123 ψe123 . In this case, we use κ = e0 . Then, Ψ0 has the form
Ψ0 = s + v1 e1 + v2 e2 + v3 e3 + b12 e12 + b13 e13 + b23 e23 + t123 e123 .
Note that, as an algebra, Cℓ01,3 ≃ Cℓ0,3 ≃ H ⊕ H. If we conveniently rewrite the coefficients of Ψ0 as
Ψ0 = p0 + q1 e1 + q2 e2 + q3 e3 + p1 e12 − p2 e13 + p3 e23 − q0 e123 ,
its matrix representation is written as
pq
ρ(Ψ0 ) = .
q p
As in the three-dimensional Euclidean space, the use of spinor operators here has some significant
computational advantages, as well as an amazing interpretation. In order to see this, we first note that
e + = (s2 + b2 + b2 + b2 − b2 − b2 − b2 − q2 )
Ψ+ Ψ 12 13 23 01 02 03 0123
+ (sq0123 − b01 b23 + b02 b13 − b03 b12 )e0123 .
The Inner Product in the Space of Algebraic Spinors 165
We usually denote
e + i0 ,
σ = hΨ+ Ψ e + i0 ,
ω = hΨ+ e0123 Ψ
in such a way that
e + = σ − ωe0123 .
Ψ+ Ψ
If we define
ρ cos β = σ, ρ sin β = −ω,
we have that Ψ+ can be written as
√
Ψ+ = ρe(β/2)e0123 R,
where R ∈ Cℓ+ e
1,3 , and RR = 1. Thus, R ∈ Spin+ (1, 3). We see that Ψ+ has an interpretation very similar
to that of the three-dimensional case, except for the term e(β/2)e0123 , which has no obvious interpretation.
This term has no effect in an expression like Ψ+ eµ Ψ e + because e0123 anti-commutes with vectors, but
e + (µ 6= ν). In the case of a bivector, we can interpret the term eβe0123 as a duality
it has in Ψ+ eµ eν Ψ
rotation (Rainich, 1925; Misner and Wheeler, 1957; Vaz Jr and Rodrigues Jr., 1993).
6.6 The Inner Product in the Space of Algebraic Spinors

Let us consider in this section the definition of the inner product in the space of
spinors according to the algebraic definition of a spinor. This case deserves particular
attention, since a classical spinor in a quadratic space that is Rp,q or Rq,p is an algebraic
spinor (or an algebraic semispinor) in a quadratic space that is either Rq,p−1 or Rp,q−1 ,
respectively. Furthermore, the space of spinor operators is isomorphic as vector space
to the space of algebraic spinors or semispinors.
The Spinor Structure Map

An R-linear mapping ς in the space of algebraic spinors S is said to be a spinor
structure mapping if ς 2 = ±1 and if â = ςaς −1 , for all a ∈ Cℓp,q ⊂ EndK (S).V
If n = p + q is even, the n-vector η is such that vη = −ηv, for all v ∈ 1 (Rp,q ),
in such a way that η commutes with the even elements and anti-commutes with the
odd elements. Thus, for n even, this spinor structure map exists and is given by the
multiplication by η, in such a way that ηaη −1 = â.
If n = p + q is odd, then the n-vector η commutes with all elements of Cℓp,q . Any
odd element of Cℓp,q can be written in the form of a product of an even element and η,
namely, Cℓ− + + +
p,q = Cℓp,q η, and we can write Cℓp,q = Cℓp,q ⊕ Cℓp,q η. Let us now distinguish
two cases. In the first case, p − q = 3, 7 mod 8. In this case, we have η 2 = −1, and
η defines a complex structure. We also have ηS = iS, where S denotes the space of
algebraic spinors. The spinor structure map corresponds to the complex conjugation
in S. In the second case, p − q = 1, 5 mod 8. In this case, η 2 = 1, and ηS = IS,
where I ∈ D, and I 2 = 1. The conjugation in D (like the conjugation in C) is defined
by I ∗ = −I. The application of spinor structure corresponds to the conjugation in S
accordingly.
The Two Types of Inner Products in Spinor Space

The spinor inner product is an inner product in the – algebraic – spinor space, with
the property that the adjoint with respect to this inner product corresponds to an
anti-automorphism in the corresponding Clifford algebra. Since we have two types of
166 Spinors
anti-automorphisms, namely, the reversion and the conjugation, we have consequently

two types of spinor inner products. Let S be the space of algebraic spinors. We can
define the spinor inner products h̃ : S × S → K, and h̄ : S × S → K. Given x, y ∈ S,
and a ∈ Cℓp,q , in such a way that Cℓp,q ≃ EndK (S), define h̃ as the spinor inner product
with the property that
h̃(ax, y) = h̃(x, ãy). (6.56)
In this case, the adjoint corresponds to the reversion. The spinor inner product h̄ is
defined in such a way that
h̄(ax, y) = h̄(x, āy). (6.57)
In this case, the adjoint corresponds to the conjugation.
The conjugation corresponds to the composition between the reversion and the
grade involution. As shown in the previous subsection, there always exists a spinor
structure map in S – an R-linear mapping ς in the space of algebraic spinors – such
that the grade involution reads â = ςaς −1 , ∀a ∈ Cℓp,q . Consequently, one of these inner
products determines the other, and vice versa, by the expression
h̄(x, y) = h̃(ςx, y), (6.58)
where ς denotes the application of spinor structure. Indeed, h̄(ax, y) = h̃(ςax, y) =

h̃(âςx, y) = h̃(ςx, āy) = h̄(x, āy). Let us denote by ◦ any one of the anti-automorphisms
˜ or ¯, that is, let ψ ◦ denote ψ̃ or ψ̄ for ψ ∈ Cℓp,q . As seen in chapter 4, Cℓp,q is a simple
algebra; therefore, f Cℓp,q f ≃ K, where K = R, C, H; and, if Cℓp,q is a semisimple
algebra, then f Cℓp,q f ≃ K ⊕ K. For ψ, φ ∈ S, we have ψf = ψ, and φf = φ; and for
ψ ◦ φ, it follows that
ψ ◦ φ = (ψf )◦ φf = f ◦ ψ ◦ φf. (6.59)
Now, notice that the quantity h(ψ, φ) = ψ ◦ φ satisfies h(aψ, φ) = h(ψ, a◦ φ). Therefore,
if f ◦ = f , then ψ ◦ φ is a spinor scalar product. If f ◦ 6= f , we can take an element
s ∈ Cℓ∗p,q such that sf ◦ s−1 = f and then sψ ◦ φ is a spinor scalar product.
In this way, the applications h̃ and h̄, defined as
h̃(ψ, φ) = sψ̃φ,
(6.60)
h̄(ψ, φ) = sψ̄φ,
with s ∈ Cℓ∗p,q , are spinor scalar products. The existence of either an element s ∈ Cℓp,q
satisfying sf˜f s−1 = f or s ∈ Cℓp,q satisfying sf¯f s−1 = f can be explicitly verified for
low dimensions and induced for high dimensions via the periodicity theorem.
Example 6.13 Let us consider as an example Cℓ1,3 , with (e0 )2 = 1; (ei )2 = −1 (i = 1, 2, 3); and
eµ eν = −eν eµ (µ 6= ν). Let us take algebraic spinors as elements of the minimal left ideal S = Cℓ1,3 f ,
where f = (1/2)(1 + e0 ). Let us denote ξ1 = f , and ξ2 = e5 f , where e5 = e0 e1 e2 e3 . The left ideal S
has a basis
{ξ1 , e2 e3 ξ1 , e3 e1 ξ1 , e1 e2 ξ1 , ξ2 , e2 e3 ξ2 , e3 e1 ξ2 , e1 e2 ξ2 } .
The set f Cℓ1,3 f is isomorphic to H and has a basis {ξ1 , e2 e3 ξ1 , e3 e1 ξ1 , e1 e2 ξ1 }. The ideal S is thus a
right H-module with the basis {ξ1 , ξ2 }. Now we can see that
ξ˜1 ξ1 = ξ1 , ξ˜1 ξ2 = 0, ξ¯1 ξ1 = 0, ξ̄1 ξ2 = ξ2 ,

and
ξ̃2 ξ1 = 0, ξ̃2 ξ2 = ξ1 , ξ¯2 ξ1 = ξ2 , ξ̄2 ξ2 = 0.
The spinor scalar product h̃ can be defined by

h̃(ψ, φ) = ψ̃φ,
and the spinor scalar product h̄ can be defined as
h̄(ψ, φ) = e5 ψ̄φ.
The Spinor Inner Product, and Charge Conjugation

As the two types of spinor inner products have now been examined, let us now focus
on the complex Clifford algebras in this context. Given a spinor ψ in C ⊗ Cℓp,q , the
associated adjoint spinor ψ ◦ is represented by
ψ ◦ (φ) = h◦ (ψ, φ), (6.61)
where ψ is an arbitrary spinor. For all a ∈ C ⊗ Cℓp,q , we can verify that h◦ (aψ, φ) =
h◦ (ψ, a◦ φ), by the equations in (6.60). Except in the cases where p is odd, and q
is even, the mapping ∗ , which is the composition of the Clifford conjugation and the
complex conjugation – seen as an involutive automorphism which induces a non-trivial
automorphism in the centre of the algebra – is an adjoint involution associated with
some Hermitian spinor product. When p is even, and q is odd, the Clifford conjugation
is substituted by the reversion, in the previous case.
Let us first consider the case where p is odd, and q is even. For a choice of basis
for the Clifford algebra representations, consider the Hermitian conjugation † , which
is related to ∗ by
ā∗ = s−1 a† s, ∀a ∈ C ⊗ Cℓp,q , (6.62)
where s̄∗ = s ∈ C ⊗ Cℓp,q – or, equivalently, s† = s. As a particular case, given spinors
ψ and φ, we define the spin-invariant product
h∗ (φ, ψ) = sφ∗ ψ,
which is invariant by elements of the Clifford–Lipschitz group and which takes values
in the complex numbers algebra, whose identity is f Cℓp,q f ≃ C. It follows that we can
obtain a product which is a complex number and is given by hφ, ψi = Tr h∗ (φ, ψ). The
adjoint of ψ with respect to the product in eqn (6.62) is given by
ψ △ = sψ ∗ = ψ † s . (6.63)
If a basis for the representation of C ⊗ Cℓp,1 is chosen such that (e0 )† = −e0 , and
(e ) = ei , i = 1, . . . , p, then eqn (6.62) is equivalent to ea † = −sea s−1 . Hence, we
i †
can choose s = ie0 , which implies the well-known relation

ψ △ = iψ † e0 .2 (6.64)
For the case C ⊗ Cℓ1,q , the factor i is absent from eqn (6.64).
2 Here, we denote the adjoint spinor by ψ △ , unlike the case in quantum mechanics textbooks
where, in general, the authors adopt the notation ψ̄ for the case where ψ is a Dirac spinor. We
use this notation in order to reduce the possibility of confusing the adjoint spinor with the Clifford
conjugation.
168 Spinors
When p is even, and q is odd, the Clifford conjugation is substituted by the rever-
sion, and, therefore
ã∗ = s−1 a† s, ∀a ∈ C ⊗ Cℓp,q , (6.65)
where s̃∗ = s† = s. The adjoint of ψ with respect to the product in eqn (6.65) is
defined by
ψ △ = sψ̃ ∗ = ψ † s . (6.66)
The adjoint spinor in (6.61) is associated with a pseudo-Hermitian product for
which ψ̄ ∗ or ψ̃ ∗ are the adjoint spinors. There also exists spin-invariant products for
which the involutions ψ̄ or ψ̃ are the adjoint spinors. Except in the case when n = 1
mod 8 or n = 5 mod 8, the involution (¯· ) induces an involution in the reducible
Clifford algebra simple components. If (·)T represents the transposition in some matrix
basis, then, except in those cases, it follows that
ā = b−1 aT b, ∀a ∈ C ⊗ Cℓp,q , (6.67)
where b̄ = bT = ±b ∈ C ⊗ Cℓp,q . The symmetry of b determines the symmetry of a

complex bilinear product defined by
h̄(φ, ψ) = bφ̄ψ. (6.68)
Equation (6.67) is thus equivalent to
eaT = −bea b−1 , (6.69)
whose matrix entries are usually taken as the charge conjugation definition. If ψ △ is
the adjoint of ψ with respect to the product (6.68), then
ψ △ = bψ̃ = ψ T b . (6.70)
This kind of adjoint spinor is called a Majorana spinor. Except when n = 3 or 7

mod 8, the reversion induces an involution in the reducible Clifford algebra simple
components. We can define
ã = b−1 aT b, ∀ a ∈ C ⊗ Cℓp,q , (6.71)
where b̃ = bT = ±b, and therefore ea T = bea b−1 .

The automorphism ∗ regarding the complex conjugation leaves invariant the real
subalgebra generated by the real orthogonal space of signature (p, q). Hence, the au-
tomorphism ⋆ is defined as the complex conjugate of the matrix entries in some basis.
The automorphism ⋆ depends on the choice of basis. Except when the real subalge-
bra is isomorphic to the complex matrix algebra for some m ∈ C ⊗ Cℓp,q , these two
automorphisms are related by
a∗ = ma⋆ m−1 , ∀a ∈ C ⊗ Cℓp,q . (6.72)
When the real subalgebra is represented by the algebra of real matrices, we can choose
m such that mm∗ = 1. When the real subalgebra is either the tensor product between
a matrix algebra and the quaternions, or the sum of two of these algebras, we can
choose mm∗ = −1. The real subalgebra is isomorphic to the complex matrix algebra
when p − q = 3 or 7 mod 8. In this case, the complex conjugation transposes the
simple components and, except in this case, we define the charge conjugate spinor
ψc = ψ∗m . (6.73)
This expression can be written by using the Dirac adjoint and the charge conjugation
matrix as well. Except in the case when n = 1 or 5 mod 8, or when p is odd, and q is
even, we can use eqns (6.65) and (6.67):
ā∗ = s̄b−1 a†T bs̄−1 . (6.74)
Now, the complex conjugation operator commutes with the Clifford conjugation op-
erator, and †T = T† = ⋆ , yielding
a∗ = ma⋆ m−1 , m = s−1 ∗ b−1 , (6.75)
where we used the fact that s̄∗ = s. We can choose m in such a way that mm∗ = ±1,
which is immediately obtainable from a suitable choice of C. Using eqn (6.73), eqn
(6.75) can be expressed as
ψ c = b−1 ψ̄ T . (6.76)
In the same way, for the case where n = 3 or 7 mod 8, or when p is odd, and q is odd,
eqn (6.73) reads
ψ c = b−1 ψ̄ T , (6.77)
where ψ̄ is given by eqn (6.66).
In the particular case concerning Dirac spinors, since they carry a reducible rep-
resentation of the real Clifford subalgebra, elements of the spaces that carry the ir-
reducible representation are called Majorana spinors. This case is manifested either
when the real subalgebra is a real matrix algebra or the direct sum of two such alge-
bras, namely, when p − q = 0, 1, 2 mod 8. In this signature, Dirac spinors can be split
into eigenspaces of the charge conjugation operator c . Hence, a Majorana spinor is an
eigenspinor of the charge conjugation operator:
ψ c = ±ψ . (6.78)
It is worth emphasising that, for the complex case, eqn (6.78) reads ψ c = eiθ ψ, where
θ ∈ R. When the dimension of the underlying vector space is even, the irreducible
representations of the complex Clifford algebra induce a reducible representation of
the even subalgebra, where the spinor representation is split into two inequivalent
semispinor representations of the even subalgebra. The central idempotents that cause
the even subalgebra to be split into simple components are given by P± = 12 (1 + η̊),
where either η̊ = η, or η̊ = iη, ensuring that η̊ 2 = 1, where η denotes the n-volume
element. A Dirac spinor can thus be split into subspaces which transform as irreducible
representations under the even subalgebra:
ψ = ψ+ + ψ− , where ψ ± = P± ψ , (6.79)
where ψ± are called chiral spinors, or Weyl spinors.
170 Spinors
When the dimension of the vector space is odd, the irreducible representation of the
complexified Clifford algebra induces irreducible representations of the even subalgebra
and can also induce a reducible representation of the real even subalgebra. This is the
case where p − q = 1 mod 8, where Dirac spinors carry a reducible representation
of the real subalgebra. For the case where p − q = 7 mod 8, the Dirac spinors carry
irreducible representations of the real subalgebra and the even subalgebra, but such
spinors carry a reducible representation of the real even subalgebra. For p − q = 7 mod
8, we know that
Cℓp,q ≃ C ⊗ M(2(n−1)/2 , R), Cℓ+

p,q ≃ M(2
(n−1)/2
, R) . (6.80)
Hence, we can choose a basis for the Clifford algebra in such a way that the grade invo-
lution plays the role of complex conjugation of the components. The complexified Clif-
ford algebra Cℓp,q (C) is reducible and, in this case, the grade involution interchanges
the simple components. The complex conjugation also interchanges the components
∗
of the algebras, and the automorphism (d · ) preserves the simple components.
6.7 The Triality Principle in the Clifford Algebraic Context

In this section, the triality principle is introduced in the Clifford algebraic context; the
geometric point of view is further explored in the work by Benn and Tucker (1987)
and Knus (1998). For other geometric and topological approaches, see, for instance,
the work by Porteous (1969) and Harvey (1990).
Let us consider a field K of characteristic that does not equal 2. Suppose that
the spinor space S can be expressed as the direct sum S = S + ⊕ S − , where S ± are
semispinor spaces – carrying non-equivalent irreducible representations of the even
subalgebra Cℓ+ (V, g). The aim here is to search for a vector space V such that the
associated semispinor spaces S ± have the same dimension of V . When K = R or C,
it must happen when dim V = dim S + = dim S − = n and also when the index
of g equals the index of h, where h denotes the spinor metric associated with the
reversion – when K = C. The first condition holds when 2n/2−1 = n, namely, when
n = 8. When K = R, the spinor metric h associated with the reversion exists only
in one of the following three cases: V ≃ R8,0 ; R0,8 ; or R4,4 , although this last case
demands a careful approach, since the metric index is not maximal in this case. For
more details, see the work by Benn and Tucker (1987) and de Andrade, Rojas, and
Toppan (2001). Hence, in what follows, we consider V ≃ C8 . Moreover, in the real
case we have V ≃ R8,0 ; R0,8 ; or R4,4 .
Let us define the space E = V ⊕ S + ⊕ S − , which is 24-dimensional. Its elements
read φ = x + u + v, where in this section we fix the notation, x ∈ V ; u ∈ S + ; and
v ∈ S − . We first introduce a result that will be used throughout this section.
Lemma 6.1: If x ∈ V is such that x2 = 1, then for all v ∈ S − there exists u ∈ S + such
that v = xu.
Proof: Indeed, consider the volume element η ∈ Cℓ+ (V, g). Since S + and S − are
spaces that carry the irreducible representations of Cℓ+ (V, g), and dim V = 8, then
S = S + ⊕ S − carries an irreducible representation of Cℓ(V, g). In this case, it follows
The Triality Principle in the Clifford Algebraic Context 171
that S ± = 12 (1±η)S ± since, in the matrix algebra, every element of the even subalgebra
Cℓ+ (V, g) can be represented
by a block diagonal matrix ∗ 0
0 ∗ such that η can be
represented by η = 0I −I 0
. For any vector x ∈ V , it follows that ηx = −xη. Hence,
since u ∈ S + , therefore xu = x 21 (1 + η)u = 21 (1 − η)xu, and we conclude that xu ∈ S − .
Similarly, given v ∈ S − , we can prove that xv ∈ S + . ✓
The spinor metric h : S ± × S ± → K is defined in each one of these spaces S ± as
⌢
h(u, xv) = h(x̃u, v) = xuv , (6.81)
where the dual adjoint is given by
⌢
( · ) : S ± −→ (S ± )∗ = Hom(S ± , K) ,
⌢ ⌢
ψ 7→ ψ, where ψ(φ) = h(φ, ψ), ∀φ ∈ S ± .
From this definition, a symmetric bilinear form B endowing the vector space E can
be introduced, from the spinor metric h and also from the metric g that endows the
space V . Let xi ∈ V ; ui ∈ S + ; vi ∈ S − , i = 1, 2, 3. Then, we can define (Cartan, 1937;
Chevalley, 1954; Benn and Tucker, 1987)
B(φ1 , φ2 ) = g(x1 , x2 ) + h(u1 , u2 ) + h(v1 , v2 ) . (6.82)
Now, a totally symmetric trilinear tensor T : E × E × E → K can be defined as
T (φ1 , φ2 , φ3 ) = h(u1 , x2 v3 ) + h(u1 , x3 v2 ) + h(u2 , x3 v1 )
(6.83)
+ h(u2 , x1 v3 ) + h(u3 , x2 v1 ) + h(u3 , x1 v2 ).
The Chevalley Product

It is possible to endow the space E with a commutative and non-associative product
◦ : E × E → E, called the Chevalley product and implicitly defined as
T (φ1 , φ2 , φ3 ) = B(φ1 ◦ φ2 , φ3 ) . (6.84)
From eqn (6.83) – defining the total symmetry property of T – we conclude that the
product ◦ is indeed commutative (φ1 ◦ φ2 = φ2 ◦ φ1 ). The pair (E, ◦) is a commutative
and non-alternative algebra – in particular, the algebra (E, ◦) is not associative. In
order to prove this last property, we shall show that x ◦ (x ◦ u) 6= (x ◦ x) ◦ u, for all
non-isotropic vectors x ∈ V and for all u ∈ S + . In fact, x ◦ (x ◦ u) = x ◦ (xu) = x(xu),
since xu ∈ S − . Then it yields x ◦ (x ◦ u) = x2 u = g(x, x)u, whereas (x ◦ x) ◦ u = 0.
It is useful to observe that, if φ1 and φ2 are elements in the same subspace, then
T (φ1 , φ2 , φ3 ) = 0 and thus φ1 ◦ φ2 = 0. Moreover, for all x ∈ V , for all u ∈ S + , and
for all v ∈ S − we can assert that
x ◦ u = xu, x ◦ v = xv. (6.85)
Indeed, it forthwith follows for the definitions (6.82) and (6.83) that
B(x ◦ u, v) = T (x, u, v) = h(u, xv) = h(xu, v) = B(xu, v) ,
B(x ◦ v, u) = T (x, v, u) = h(v, xu) = h(xv, u) = B(xv, u) .
The spinor norm x ◦ u can be derived from the norms of x and u, since h(x ◦ u, x ◦
u) = h(xu, xu) = g(x, x)h(u, u). Since the Chevalley product between vectors and
172 Spinors
semispinors has been defined from the action of the regular representation of S ± the
equations in (6.85), it is useful to express the Chevalley product between semispinors
from the Clifford product between spinors, by the relations
⌢ ⌢
B(u ◦ v, x) = T (u, v, x) = h(xu, v) = (xu)v = h u xvi0
⌢ ⌢ ⌢
(6.86)
= hxv u i0 = h(xhv u i1 )i0 = B(x, hv u i1 ) ,
which implies that

⌢
u ◦ v = hv u i1 . (6.87)
A prominent result that introduces the triality principle is that the inclusions
V ◦ S+ ⊆ S−, S + ◦ S − ⊆ V, S− ◦ V ⊆ S+ (6.88)
hold. Indeed, if φ1 ∈ V , φ2 ∈ S + , and φ3 ∈ V ⊕ S + , we denote φ1 = x1 ; φ2 = u2 ;

and φ3 = x3 + u3 . Hence, T (φ1 , φ2 , φ3 ) = T (x1 , u2 , φ3 ) = 0 = B(φ1 ◦ φ2 , φ3 ), namely,
φ1 ◦ φ2 is an element in the space orthogonal to V ⊕ S + with respect to the metric B,
since E = S − ⊕ (S − )⊥ – here we suppose that ker h = {0}. A similar reasoning holds
for the other cases.
A spinor representation σ of the Clifford–Lipschitz group in S ± , and the vector
representation χ of the Clifford–Lipschitz group in V , induce an irreducible represen-
tation Y in E. Given a unitary element of the Clifford–Lipschitz group s ∈ Γ+ p,q , the
action of this representation in Y (s) : E → E is defined as
Y (s)(x + u + v) = χ(s)x + σ(s)u + σ(s)v = sxs−1 + su + sv, (6.89)
where such a mapping is orthogonal with respect to the bilinear form B, since
B(Y (s)φ1 , Y (s)φ2 )

= B(χ(s)x1 + σ(s)u1 + σ(s)v1 , χ(s)x2 + σ(s)u2 + σ(s)v2 )
= g(χ(s)x1 , χ(s)x2 ) + h(σ(s)u1 , σ(s)u2 ) + h(σ(s)v1 , σ(s)v2 )
= g(sx1 s−1 , sx2 s−1 ) + h(su1 , su2 ) + h(sv1 , sv2 )
= g(x1 , x2 ) + h(u1 , u2 ) + h(v1 , v2 )
= B(φ1 , φ2 ) .
In addition, if x0 ∈ V is such that x20 = 1, then Y 2 (x0 ) = I. Obviously, x0 ∈ Γ+

p,q , in
such a way that the notation Y (x0 ) does indeed make sense. Since x20 = 1, therefore
x−1
0 = x0 , yielding
Y 2 (x0 )(x + u + v) = Y (x0 )(x0 xx0 + x0 u + x0 v)

= x20 xx20 + x20 u + x20 v = x + u + v.
For the tensor T to be invariant under Y (s), it is necessary that Y (s) be an auto-
morphism with respect to the Chevalley product. In fact,
T (Y (s)φ1 , Y (s)φ2 , Y (s)φ3 ) = B([Y (s)φ1 ] ◦ [Y (s)φ2 ], Y (s)φ3 )

= B(Y (s)[φ1 ◦ φ2 ], Y (s)φ3 ) = B(φ1 ◦ φ2 , φ3 ) (6.90)
= T (φ1 , φ2 , φ3 ).
The Triality Principle
In this section, the triality principle is introduced, in order that the algebraic character
of this approach can be made explicit. Consider a spinor u0 ∈ S + of unitary norm with
respect to the spinor metric, namely, h(u0 , u0 ) = 1. Define the linear mapping
ζ : S+ → V
u0 7→ ζ(u0 ) : V → S − (6.91)
x 7→ ζ(u0 )(x) := x ◦ u0
with the consequent property that ζ(u0 ) is orthogonal with respect to B; this result
can be straightforwardly verified:
h(ζ(u0 )x1 , ζ(u0 )x2 ) = h(x1 ◦ u0 , x2 ◦ u0 ) = h(x1 u0 , x2 u0 )

⌢ ⌢⌢
= (x1 u0 )x2 u0 = u0 x1 x2 u0 = g(x1 , x2 )h(u0 , u0 ) = g(x1 , x2 ) .
This result relates the spinor x ◦ u0 ∈ S − norm to the norms of x ∈ V , and v ∈ S − .

The mapping ζ(u0 ) is uniquely extended to an involutive automorphism in V ⊕S − .
If v ∈ S − is such that v = ζ(u0 )x for an unique x ∈ V , then it is possible to define
ζ(u0 )(v) = x . (6.92)
Furthermore, the mapping ζ(u0 ) is defined in S + as a reflection with respect to the

spinor u0 ∈ S + :
ζ(u0 )(u) = 2h(u, u0 )u0 − u (6.93)
One more important result necessary to the existence of triality is the following:
Lemma 6.2: Let u0 ∈ S + and let x0 ∈ V such that x20 = 1, and h(u0 , u0 ) = 1. Then,
ζ(u0 )Y (x0 )ζ(u0 ) = Y (x0 )ζ(u0 )Y (x0 ).
Proof: Consider v ∈ S − . Then,

174 Spinors
ζ(u0 )Y (x0 )ζ(u0 )(v)

= ζ(u0 )Y (x0 )(ζ(u0 )(v)) = ζ(u0 )(x0 (ζ(u0 )(v))x0 )
= (x0 (ζ(u0 )(v))x0 ) ◦ u0 , since (x0 (ζ(u0 )(v))x0 ) ∈ V
= (x0 (ζ(u0 )(v))x0 )u0 ,
= 2g(x0 , ζ(u0 )v)x0 u0 − ((ζ(u0 )v)x0 )x0 u0 ,
since (ζ(u0 )v)x0 + x0 (ζ(u0 )v) = 2g(x0 , ζ(u0 )v)
= 2h(ζ(u0 )x0 , ζ(u0 )ζ(u0 )v)x0 u0 − (ζ(u0 )v))u0
= 2h(x0 ◦ u0 , v)x0 u0 − ζ(u0 )(vu0 )
= 2h(x0 u0 , v)x0 u0 − ζ(u0 )(vu0 )
= 2h(u0 , x0 v)x0 u0 − (vu0 ) ◦ u0 , since vu0 ∈ V
= 2h(u0 , x0 v)x0 u0 − vu0 u0
= 2h(u0 , x0 v)x0 u0 − v.
On the other hand,
Y (x0 )ζ(u0 )Y (x0 )(v) = Y (x0 )ζ(u0 )(x0 v)

= Y (x0 )[2h(x0 v, u0 )u0 − x0 v] (6.94)
= 2h(x0 v, u0 )x0 u0 − x0 x0 v.
In this case, the lemma is demonstrated for S − . Now, by taking x ∈ V , we obtain:
ζ(u0 )Y (x0 )ζ(u0 )(x)

= ζ(u0 )Y (x0 )(xu0 ) = ζ(u0 )(x0 xu0 )
= 2h(x0 xu0 , u0 ) − x0 xu0 = 2h(xu0 , x0 u0 )u0 − x0 xu0
= 2h(ζ(u0 )x, ζ(u0 )x0 )u0 − x0 xu0 = 2g(x, x0 )u0 − x0 xu0
= xx0 u0 .
In addition,
Y (x0 )ζ(u0 )Y (x0 )(x) = Y (x0 )ζ(u0 )(x0 xx0 )

= Y (x0 )(x0 xx0 )u0 = x20 xx0 u0 = xx0 u0 .
Finally,
ζ(u0 )(x0 u0 ) = ζ(u0 )(x0 ◦ u0 ) = ζ(u0 )(ζ(u0 )x0 ) = x0 . (6.95)
Since ζ(u0 ) is an involutive automorphism, given u ∈ S + , it follows that
ζ(u0 )Y (x0 )ζ(u0 )(u)

= ζ(u0 )Y (x0 )(2h(u, u0 )u0 − u) = ζ(u0 )(2h(u, u0 )x0 u0 − x0 u)
= 2h(u, u0 )ζ(u0 )x0 u0 − ζ(u0 )(x0 u)
= 2h(u, u0 )x0 − ζ(u0 )(x0 u),
from eqn (6.95). On the other hand,
Y (x0 )ζ(u0 )Y (x0 )(u)

= Y (x0 )ζ(u0 )(x0 u) = x0 ζ(u0 )(x0 u)x0
= −ζ(u0 )(x0 u)x0 x0 + 2g(ζ(u0 )(x0 u), x0 )x0
= 2h(x0 u, ζ(u0 )x0 )x0 − ζ(u0 )(x0 u)
= 2h(x0 ◦ u, x0 ◦ u0 )x0 − ζ(u0 )(x0 u)
= 2g(x0 , x0 )h(u, u0 )x0 − ζ(u0 )(x0 u)
= 2h(u, u0 )x0 − ζ(u0 )(x0 u). ✓
Using those results, define now the operator Θ : E → E as follows:
Θ(x0 , u0 ) = Y (x0 )ζ(u0 ) . (6.96)
Theorem 6.1 ◮ The operator Θ(x0 , u0 ) is an order 3 automorphism.

Proof: Indeed, using lemma 6.2, we have
Θ3 (x0 , u0 ) = Y (x0 )ζ(u0 )Y (x0 )ζ(u0 )Y (x0 )ζ(u0 )

= Y (x0 )ζ(u0 )Y (x0 )Y (x0 )ζ(u0 )Y (x0 ) = 1. ✓
Moreover, we can prove that Θ(x0 , u0 ) is orthogonal with respect to the bilinear form
B
B(Θ(x0 , u0 )φ1 , Θ(x0 , u0 )φ2 ) = B(φ1 , φ2 ) (6.97)
and that
T (Θ(x0 , u0 )φ1 , Θ(x0 , u0 )φ2 , Θ(x0 , u0 )φ3 ) = T (φ1 , φ2 , φ3 ) , (6.98)
since both mappings Y (x0 ) and ζ(u0 ) satisfy each of the relations in eqns (6.97) and
(6.98).
In addition, the subspaces V , S + , and S − in E are cyclically permuted by Θ(x0 , u0 ):
Θ(x0 , u0 )V ⊂ S + , Θ(x0 , u0 )S + ⊂ S − , Θ(x0 , u0 )S − ⊂ V . (6.99)
Indeed, given x ∈ V , it follows that
Θ(x0 , u0 )(x) = Y (x0 )ζ(u0 )x = Y (x0 )(xu0 ) = x0 xu0 ∈ S + ,

Θ(x0 , u0 )(u) = Y (x0 )ζ(u0 )u = Y (x0 )[2h(u, u0 )u0 − u]
= 2h(u, u0 )xu0 − x0 u ∈ S − ,
Θ(x0 , u0 )(v) = Y (x0 )ζ(u0 )v = Y (x0 )x = x0 xx0 ∈ V.
The space of spinors associated with V is written as S + ⊕ S − . From these results,

we can assert that, if the space S ± is taken as a vector space, the spinor space as-
sociated with S ± is V ⊕ S ∓ . In this sense, semispinors in S ± – if we consider the
176 Spinors
underlying vector space structure of S ± – are vectors in V . We can moreover prove

the isomorphisms of Clifford algebras:
Cℓ(V, g) ≃ Cℓ(S + , h) ≃ Cℓ(S − , h). (6.100)
Indeed, consider the automorphism obtained from Θ(x0 , u0 ) on the algebras Cℓ(V, g),
Cℓ(S + , h), and Cℓ(S − , h), already defined. Consider S − ⊕ V as the spinor space of
Cℓ(S + , h) and, using the notation in the book by Benn and Tucker (1987), let us denote
by ⋄ the Clifford product in the algebra Cℓ(S + , h). Given x ∈ V , and ψ ∈ S = S + ⊕S − ,
[xψ] = [x] ⋄ [ψ].
For more details see the book by Benn and Tucker (1987).
Triality, and Octonionic Realisations
One of the most prominent aspects regarding the triality principle is its underlying
geometric content. The realisation of triality in the Clifford algebraic context can be
accomplished in the octonionic algebra O, defined as the space R ⊕ R0,7 endowed with
the product ◦ : (R ⊕ R0,7 ) × (R ⊕ R0,7 ) → R ⊕ R0,7 , the so-called octonionic standard
product (Lounesto, 2001b; da Rocha and Vaz Jr, 2006). Moreover, trialities can be
composed upon different octonion products.
Concerning the vector space R ⊕ R0,7 with the basis {e0 = 1, ea }7a=1 , where it is
usual to identify e0 = 1 with the basis of R, the octonionic product reads (Harvey,
1990; Ivanova, 1993; Baez, 2002; da Rocha and Vaz Jr, 2006).
ea ◦ eb = ǫcab ec − δab (a, b, c = 1, . . . , 7), (6.101)
where ǫcab = 1 for the cyclic permutations
(abc) = (126), (237), (341), (452), (563), (674), (715).
Explicitly, the multiplication is given by table 6.8, wherein all the relations can be
expressed as ea ◦ ea+1 = ea+5 mod 7 .
Table 6.8 The Octonionic Product between Units in the O+5 Convention
1 e1 e2 e3 e4 e5 e6 e7
e1 −1 e6 e4 −e3 e7 −e2 −e5
e2 −e6 −1 e7 e5 −e4 e1 −e3
e3 −e4 −e7 −1 e1 e6 −e5 e2
e4 e3 −e5 −e1 −1 e2 e7 −e6
e5 −e7 e4 −e6 −e2 −1 e3 e1
e6 e2 −e1 e5 −e7 −e3 −1 e4
e7 e5 e3 −e2 e6 −e1 −e4 −1
The octonionic product can be constructed using the Clifford algebra Cℓ0,7 as
u ◦ v = huv(1 − ψ)i0⊕1 , u, v ∈ R ⊕ R0,7 , (6.102)
where the 3-vector ψ is given by
ψ = e1 e2 e6 +e2 e3 e7 +e3 e4 e1 +e4 e5 e2 +e5 e6 e3 +e6 e7 e4 +e7 e1 e5 . (6.103)
In a close analogy, the octonionic product can be also expressed with respect to
the Clifford algebra on the Euclidean space R8,0 , according to Lounesto (2001a), in
terms of a basis {e1 , . . . , e8 } of R8,0 . The octonionic product is given in this case by
u ◦ v = hu e8 v(1 + ψ)(1 − e12...8 )i1 , u, v ∈ R8,0 , (6.104)
where 18 (1 + ⋆ψ) 21 (1 − e12...8 ) is an idempotent. Both the approaches are equivalent:

bivectors in Cℓ8,0 correspond to the elements in R⊕R0,7 ⊂ Cℓ0,7 , when the isomorphism
eσ e8 7→ eσ , σ = 1, 2, . . . , 7 is considered and e8 e8 = 1 = e0 denotes the octonionic
unit in R ֒→ R ⊕ R0,7 . In fact, e2σ = (eσ e8 )2 = −eσ eσ e8 e8 = −1. More details can
be seen, for example, in the work by Lounesto (2001a); da Rocha and Traesel (2012),
and da Rocha, Traesel, and Vaz Jr (2012).
Table 6.8 can be obtained by the octonionic product defined either by eqn (6.102)
or by eqn (6.104). The definition in eqn (6.102) is regarded from hereon, where, in this
case R ⊕ R7 is considered instead of the usual R8 underlying vector space, concerning
the definition in eqn (6.104).
Some useful identities follow from eqn (6.101):
ǫabc ǫdcf + ǫdbc ǫacf = δab δdf + δaf δdb − 2δad δbf .
Moreover, an analogue of the Jacobi formula in this context reads
[ei , [ej , ek ]]+[ek , [ei , ej ]]+[ej , [ek , ei ]] = 3ǫijkl el , (6.105)
where ǫijkl := −ǫmij ǫmkl − δil δjk + δik δjl (Gunaydin and Ketov, 1996). Since the un-
derlying vector space of O can be considered as being R ⊕ R0,7 ֒→ Cℓ0,7 , the Clifford
conjugation of v = v0 + va ea ∈ O is given by v̄ = v 0 − v a ea , where v0 and v a are real
coefficients. The underlying structure of the vector space is unable to assert whether
the O-conjugation is equivalent to the grade involution, since the octonionic conjuga-
tion v̄ can be written either as v b or v̄, with respect to Clifford algebra morphisms.
However, the octonionic conjugation v̄ is involutive and thus an anti-automorphism,
which immediately excludes the graded involution.
To invoke an explicit realisation of the triality principle, we observe that all au-
tomorphisms of SO(n) are of the form A 7→ SAS −1 , where S ∈ O(n). Moreover, the
automorphisms of Spin(n) can be written as a 7→ sas−1 , for s ∈ Pin(n), for n 6= 8.
However, the group Spin(8) has exceptional automorphisms, which cyclically permute
the elements of the set {−1, ±e12...8 }, which lies in the centre of Spin(8). Such an
order 3 automorphism of Spin(8) is the triality automorphism. Moreover, an order 2
automorphism of Spin(8) interchanges the element −1 with either of ±e12...8 . Such an
178 Spinors
automorphism of Spin(8) is the so-called swap automorphism, denoted by swap(a), for

a ∈ Spin(8) (Lounesto, 2001b).
From the representation point of view, triality can be viewed as permuting the
vector space R8 and the two even spinor spaces, namely, the minimal left ideals
Cℓ+ 1
8 8 (1 + ⋆ψ)(1± e12...8 ), where ψ is given by eqn (6.103). Moreover, it corresponds to
a 120◦ rotation of the Coxeter–Dynkin diagram of the Lie algebra D4 ≃ so(8) (Knus,
1998; Lounesto, 2001b).
Since the octonionic framework is helpful in this presentation, it is worth regarding
triality in the context of the Clifford algebra Cℓ0,7 ≃ M(8, R) ⊕ M(8, R) and the $pin
group
$pin(8) = {a ∈ Cℓ+
0,7 | |hāai0 | = 1, and avâ
−1
∈ R ⊕ R0,7 , ∀v ∈ R ⊕ R0,7 } .
For a ∈ $pin(8), and v ∈ R ⊕ R0,7 , the two linear transformations A1 and A2 of

O(R ⊕ R0,7 ) can be defined, respectively, by
A1 (v) = 16havf i0⊕1 , A2 (v) = 16havfî0⊕1 , (6.106)
where f = 81 (1 + ⋆ψ)(1 ± e12...8 ). The action of a on the left ideal Cℓ0,7 18 (1 + ⋆ψ) has
the matrix representation

A1 0
[a] = ∈ $pin(8) . (6.107)
0 A2
Now, given u ∈ R ⊕ R0,7 , the linear transformation U ∈ End(R ⊕ R0,7 ) can be

defined by
U (v) = 16huvf i0⊕1 . (6.108)
Thus,
A(v) = a ◦ v .
Therefore, the space R ⊕ R0,7 is the underlying vector space for the octonion algebra
O.
Now, using a notation similar to the one used in the article by Lounesto (2001b), let
us define the so-called companion matrix Ǔ that he associated with U ∈ SO(R ⊕ R0,7 )
in the same article:
[
Ǔ (v) := U (v̂), v ∈ R ⊕ R0,7 . (6.109)
Since Ǔ ⊺ (v) = 16huvfî0⊕1 , therefore,

U 0
[u] = ∈ $pin(8) , (6.110)
0 Ǔ ⊺
which from now on, in this section, we shall denote by u ∼ U . Now, by computing the
matrix product
−1
−1 A1 0 U 0 Ǎ2 0
A(u) = auâ ∼ , (6.111)
0 A2 0 Ǔ ⊺ 0 Ǎ−1
1
Pure Spinors 179
we find the correspondence A(u) ∼ A1 U Ǎ−1

2 . By denoting A0 = Ǎ, for all u, v ∈
R ⊕ R0,7 we obtain (Lounesto, 2001b)
Ǎ0 (u) ◦ v = A1 U Ǎ−1 −1

2 (v) = A1 (u ◦ Ǎ2 (v)). (6.112)
The ordered triple (A0 , A1 , A2 ) in SO(8) is called a triality triplet with respect to the
octonion product of O. Lounesto (2001b) observed that, if (A0 , A1 , A2 ) is a triality
triplet, then (A2 , A0 , A1 ) and (Ǎ2 , Ǎ1 , Ǎ0 ) are also triality triplets. Hence, Cartan’s
triality principle reads
Ǎ0 (u ◦ v) = A1 (u) ◦ A2 (v), u, v ∈ R ⊕ R0,7 . (6.113)
6.8 Pure Spinors

In this section, we introduce the main concepts involving pure spinors. For more details
and further developments see the work by Benn and Tucker (1987), Cartan (1937),
Chevalley (1954), Budinich and Trautmann (1989), Crumeyrolle (1990), and Budinich
(2002). It is also worth pointing, out that, in four and in six dimensions, pure spinors ac-
cidentally coincide with Weyl spinors (Lounesto, 2001a; da Rocha and da Silva, 2010),
while there are quadratic constraints that pure spinors obey in higher dimensions (Bu-
dinich, 2002). In particular, the constraints in ten dimensions play an important role
in Berkovits’s approach to superstrings (Berkovits and Howe, 2002; Berkovits, 2004).
Given a complex vector space C2r and its associated Clifford algebra Cℓ(2r, C), we
have already seen from the classical definition of spinors that a spinor u is a vector of
the 2r-dimensional representation space of the Spin group, associated with the Clifford
algebra Cℓ(2r, C) = End S. A pure spinor is defined by the Cartan equation vu = 0
(Cartan, 1937). For u 6= 0, v ∈ C2r is isotropic. Before introducing this concept, let us
remember some prerequisites in what follows.
Given a bilinear form B : V × V → K, the set {v ∈ V | B(u, v) = 0, ∀ u ∈ V }
is a subspace of the space V – denominated radical of V (rad V ). A vector space V
endowed with a bilinear form B – either symmetric or antisymmetric – is said to be
a direct
Lr sum of orthogonal vector subspaces Vi (i = 1, . . . , r) – with respect to B – if
V = i=1 Vi , and B(vi , vj ) = 0 for all vi ∈ Vi , and vj ∈ Vj . A subspace U ⊆ V is said
to be isotropic (totally isotropic) if the restriction of B to the subspace U is degenerate
(null). A vector v ∈ V is said to be isotropic if B(v, v) = 0, and a subspace U ⊆ V
is said to be isotropic if it contains a non-trivial isotropic vector. A two-dimensional
quadratic space that has a null radical is called a hyperbolic plan (Crumeyrolle, 1990),
and the orthogonal direct sum of r hyperbolic plans Pi is a 2r-dimensional hyperbolic
space over a field K.
Now we exhibit the Witt decomposition, relating hyperbolic and isotropic sub-
spaces of V . For details about proofs and correlated topics see, for example, the work by
Chevalley (1954), Lam (1980), Crumeyrolle (1990), and Ablamowicz (1995). Consid-
ering (V, B) a finite-dimensional vector space endowed by a non-degenerate quadratic
form B over a field K, and W ⊂ V a maximal isotropic subspace, the Witt decompo-
sition shows that there exists a maximal isotropic subspace U ⊂ V such that
(a) dim U = dim W, U ∩ W = {0} .
180 Spinors
(b) V = W ⊕ U ⊕ (W ⊕ U )⊥ .
(c) For all v ∈ (W ⊕ U )⊥ \{0}, we have B(v, v) 6= 0 .
In addition, among all bases {wj } ⊂ W , there exists a basis {ui } ⊂ U satisfying
B(ui , wj ) = δij , 1 ≤ i, j ≤ k, called the Witt basis.
Another result is based upon the straightforward definition of a vector space V
whose correlation is non-degenerate – namely, the mapping v ∈ V 7→ B(v, ·) is one-
to-one – and is endowed with the quadratic form Q : V → K. Then,
(a) Every r-dimensional totally isotropic subspace U ⊂ V is an element of a hyper-
bolic subspace H2r of V .
(b) V is isotropic if and only if V contains a hyperbolic plan.
(c) If V is isotropic, the quadratic form Q takes values in all non-null elements in K,
namely, for all a ∈ K∗ , there exists v ∈ V \{0} such that Q(v) = B(v, v) = a.
A hyperbolic plan P has a basis (u, v) consisting of a pair of isotropic vectors (u2 =
0 = v2 ) satisfying B(u, v) = 1 and called a hyperbolic pair. Any hyperbolic space
H2r has a basis of hyperbolic vectors (ui , vi ) such that B(ui , uj ) = 0 = B(vi , vj ),
and B(ui , vj ) = δij , i, j = 1, . . . , r. Moreover, any quadratic space can be split in
an orthogonal direct sum, consisting of a totally isotropic subspace – rad V – an
anisotropic subspace, and a hyperbolic subspace. For more details see the work by
Chevalley (1954), Benn and Tucker (1987), and Crumeyrolle (1990). Consequently,
given any totally isotropic maximal subspace F of V , we can find another maximal
totally isotropic subspace F ′ ⊂ V such that F ⊕ F ′ = V . In particular, if B is anti-
symmetric and non-degenerate, then V is the hyperbolic space – H2r = V – and the
hyperbolic pairs (ui , vj ) form the so-called symplectic basis of H2r .
Given a non-degenerate bilinear form B, any isometry σ : F → F ′ can be extended
to an isometry of V . In particular, this result implies that, when F and F ′ are maximal
totally isotropic subspaces, they have the same dimension r, which is called the Witt
index of V . Spaces where the associated Witt index is maximal are called neutral.3
According to the Sylvester theorem, the index of V is given by the maximum between
p and q, where p − q is the metric signature.
Example 6.14 Given an orthonormal basis {e1 , . . . , en } of Rp,q , where p + q = n, and denoting by r the
index of Q, the Witt decomposition for Rp,q is given by (F ⊕ F ′ )⊥G, where F and F ′ are maximal totally
isotropic subspaces, dim F = dim F ′ = r, and the subspace F ⊕ F ′ is generated by the hyperbolic pairs
(ui , vj ) defined by
1 1 1
u1 = (e1 + en ), u2 = (e2 + en−1 ), . . . , ur = (er + en−r+1 ),
2 2 2
(6.114)
1 1 1
v1 = (e1 − en ), v2 = (e2 − en−1 ), . . . , vr = (er − en−r+1 ).
2 2 2
These elements satisfy the relations
ui uj + uj ui = 0 = vi vj + vj vi , (⇒ vi2 = 0 = u2i ), (6.115)
ui vj + vj ui = δij , (6.116)
where i, j = 1, . . . , r, and G is an anisotropic subspace generated by {er+1 , er+2 , . . . , en−r }. The sets
{ui } and {vj } generate, respectively, the maximal totally isotropic subspaces F and F ′ . In particular,
when p = q = r, the vector space Rr,r is neutral and equal to H2r .
3 Quadratic spaces over algebraically closed fields are always neutral.

Pure Spinors 181
In the complex case, the space is VC = C2r , and we can consider a basis {e1 , e2 , . . . , e2r } for
V = Rp,q , with p + q = 2r, and p ≤ q. Hence, the quadratic form QC has maximal Witt index r, and
VC = F ⊕ F ′ , where the sets constituted by the vectors
1 1
u1 = (e1 + e2r ), . . . , uk = (ek + e2r−k+1 ),
2 2
1 1
uk+1 = (iek+1 − e2r−k ), . . . , ur = (ier + er+1 ),
2 2
1 1
v1 = (e1 − e2r ), . . . , vk = (ek − e2r−k+1 ),
2 2
1 1
vk+1 = (iek+1 − e2r−k ), . . . , vr = (ier − er+1 )
2 2
generate, respectively, the maximal totally isotropic subspaces F and F ′ for VC , and (ui , vj ) are then
hyperbolic pairs with respect to QC .
In the particular case of the Minkowski spacetime R1,3 with orthonormal basis {eµ }, the quadratic
form Q has index 1, and the Witt decomposition of R1,3 is given by (F ⊕ F ′ )⊥G, where u1 = 12 (e0 + e1 ),
v1 = 12 (−e0 + e1 ). The space G is generated by e2 and e3 – this case is commonly defined in the bosonic
strings formalism. For a gentle introduction see, for example, the work by Zwiebach (2001) – whereas, for
the case C1+3 = F ⊕ F ′ , the maximal totally isotropic subspaces F and F ′ are, respectively, generated
by (Zwiebach, 2001)
1 1
u1 = (e0 + e1 ), u2 = (ie2 + e3 ),
2 2
1 1
v1 = (e1 − e0 ), v2 = (ie2 − e3 ).
2 2
The spinor space S for Cℓr,r can be constructed from the Witt decomposition
Rr,r = F ⊕ F ′ , where the maximal totally isotropic subspaces F and F ′ – both of
dimension r – are generated by
1 1 1
u1 = (e1 + e2r ), u2 = (e2 + e2r−1 ), . . . , ur = (er + er+1 ),
2 2 2 (6.117)
1 1 1
v1 = (e1 − e2r ), v2 = (e2 − e2r−1 ), . . . , vr = (er − er+1 ).
2 2 2
Define now the volume element associated with the maximal totally isotropic subspaces
F ′ as ΩF ′ = v1 v2 . . . vr so that Ω2F ′ = 0. A basis for S is given by ui1 i2 ...ik ΩF ′ ,
1 ≤ i1 ≤ · · · ≤ ir ≤ r. The elements {ui1 i2 ...ik , vi1 i2 ...ik } form a basis of the algebra
Cℓr,r . It then follows that dim Cℓr,r ΩF ′ = 2r and that S ≃ Cℓr,r ΩF ′ . Hence, S is a
minimal left ideal of Cℓr,r . Since
V B|F ≡ 0 ≡ B|F , in particular, the elements {ui }
′
generate the exterior algebra (F ) ֒→ Cℓr,r .

Furthermore, using the relation given in eqn (6.116), the elements of F can be
positioned on the right side of any term in Cℓ(V, g). Hence, we obtain
^
Cℓ(V, g)ΩF = Cℓ(F ′ , g)ΩF = (F ′ )ΩF ,
where the left ideal Cℓ(V, g)ΩF has the dimension of the exterior algebra associated to
F ′ – equal to 2r – which thus is a minimal ideal (Ablamowicz, 1995). Hence Cℓ(F ′ , g)ΩF
is the spinor space.
Although the r-vector ΩF defined in this way is isotropic, e1 e2 . . . er ΩF is a prim-
itive idempotent, and it can be written as the sum of mutually annihilating idempo-
tents.
182 Spinors
Example 6.15 Consider the subspaces F and F ′ given by (6.114), and define new elements v̊i = ei vi ,
which satisfy v̊i2 = v̊i , and v̊i v̊j = v̊j v̊i . The product ǫ = v̊1 v̊2 . . . v̊r = (−1)r(r−1)/2 e1 e2 . . . er ΩF ′
is a primitive idempotent, and a complete set of primitive idempotents {ǫ1 , ǫ2 , . . . , ǫr } can be obtained
from ǫ by applying the reversion in Cℓr,r to one or more factors vi of ǫ, with the properties
ǫ1 + ǫ2 + · · · + ǫr = 1, ǫi ǫj = 0 (i 6= j), ǫ2i = ǫi , i, j = 1, 2, . . . , r. (6.118)

From this decomposition, we can write
Cℓr,r = Cℓr,r ǫ1 ⊕ Cℓr,r ǫ2 ⊕ · · · ⊕ Cℓr,r ǫr , (6.119)
where the direct sum corresponds to the sum of non-decomposable ideals. Therefore, ǫCℓr,r ǫ ≃ R, and
Cℓr,r ǫ is a minimal ideal.
Since Cℓr,r is a simple algebra, then a representation ρ : Cℓr,r → End S is faithful and induces a
representation ρ+ of Cℓ+ r,r in End S. The spinor space S decomposes in two subspaces S
± = Cℓ± Ω ′
r,r F
of dimension 2r−1 and which are irreducible and invariant with respect to ρ+ . Moreover, the elements of
S ± are called semispinors, where the spinor space is written as the direct sum of two semispinor spaces
S = S + ⊕ S − . The representation ρ+ is an isomorphism between Cℓ+ r,r and End S
+ × End S − .
Considering again the subspaces F and F ′ given in (6.114), the Witt decomposition of Rr,r+1 is
given by (F ⊕ F ′ )⊥e2r+1 . The algebra Cℓr,r+1 ≃ M(2r , C) is simple, and its centre is generated
by {1, e123...2r+1 } ≃ C. The spinor space S = Cℓr,r+1 ΩF ′ has dimension 2r+1 over R. The algebra
Cℓr+1,r ≃ M(2r , R) ⊕ M(2r , R) is semisimple, and can be decomposed as the direct sum of minimal
ideals Cℓr,r+1 21 (1 ± e123...2r+1 ) of dimension 2r , from the central primitive idempotents.
Definition 6.4 ◮ If F and T are maximal totally isotropic subspaces in V , an element

of ΩF Cℓ(V, g)ΩT is a representative spinor for T – with respect to F . A spinor that
represents some subspace T is called a pure spinor.
Since we used a maximal totally isotropic subspace F to define the spinor space,
any other maximal totally isotropic subspace T can be used to define a right minimal
ideal and consequently to define a 1-dimensional subspace ΩF Cℓ(V, g)ΩT in the spinor
space.
In general, a mapping ρ : Cℓ(V, g) → End(S) defined by ρ(u) : vΩF 7→ uvΩF
is used to provide a representation of Cℓ(V, g) in S = Cℓ(V, g)ΩF . Such a represen-
tation also induces equivalent representations in its subgroups, that is, the Clifford–
Lipschitz group, the Pin group, and the Spin group as well. Since ψ −1 vψ ∈ V , for
all ψ ∈ Γp,q and for all v ∈ V , a Witt basis {ui , vi } can be led to another Witt
basis {u′i , vi′ }, where u′i = ψ −1 ui ψ, and vi′ = ψ −1 vi ψ. Therefore, another basis for
S is given by u′i1 i2 ...ik ΩF ′ , where ΩF ′ = ψ −1 ΩF ψ, and such bases are said to be ge-
ometrically equivalent. The choice of basis is unique, and the geometric equivalence
between any maximal totally isotropic subspaces leads to the concept of pure spinors.
Indeed, every maximal totally isotropic subspace F defining a spinor space S is an
equivalence class of pure spinors, characterised as the non-trivial elements in the in-
tersection (Cℓ(V, g)ΩF ) ∩ (ΩF ′ Cℓ(V, g)) between minimal ideals. The pure spinor space
is 1-dimensional in K. Theorem 6.2 is used in some texts as the definition of a pure
spinor.
Theorem 6.2 ◮ If the space of the representative spinors for F is generated by ΩF ,

and if ψ is a representative spinor of F , then uψ = 0, ∀u ∈ F .4
4 Note that uψ is the regular representation of the group Spin(p, q).
Pure Spinors 183
V
Proof: If ψ is any element in Cℓ(V, g)ΩF , then ψ = φΩF , for some V φ ∈ (F ′ ). For
any vector vi ∈ F ′ , we can split φ = vV i φ1 + φ2 , where φ1 , φ2 ∈ (F ′ )hvi i. Hence,
′
ui ψ = ui φΩF = κui ΩF , for some κ ∈ (F ), where in the last equality we used the
relations in (6.115) and (6.116). It follows that ui ψ = 0, since ui ΩF = ui u1 . . . ur =
(−1)i−1 u1 . . . u2i . . . ur = 0. Since this result holds for any i = 1, . . . , r, therefore
uψ = 0, for all u ∈ F . ✓
Given a maximal totally isotropic subspace F , there is no unique space F ′ such that
V = F ⊕ F ′ . If T is another maximal totally isotropic subspace where dim (T ∩ F )
= h, then it is possible to choose a Witt basis such that {ui } is a basis of F and
{u1 , . . . , uh , vh+1 , . . . , vt } is a basis of T . From some basis {u1 , . . . , uh } for T ∩ F , the
Witt basis can be completed by the Gram–Schmidt procedure. A representative for T
is therefore given by u = vh+1 . . . vr ΩF , which is known as the canonical form for a
pure spinor.
Now, given s ∈ Γp,q , if T = σ(s)(F ) = sF s−1 , a representative spinor for T can
be always written as ψ = sΩF . Indeed, any isomorphism between vector (sub)spaces
F and T can be extended to an isometry σ(s) in O(V ) by the Witt theorem. Hence,
ΩT = sΩF s−1 6= 0, and ΩT is a product of elements of a basis of T . Since sΩF ∈
Cℓ(V, g)ΩF , and sΩF = ΩT s ∈ ΩT Cℓ(V, g), it follows that sΩF generates the intersec-
tion (Cℓ(V, g)ΩF ) ∩ (ΩT Cℓ(V, g)).
It then follows that pure spinors have definite parity. Indeed, when n = p + q is
even, all elements of the twisted Clifford–Lipschitz group Γ̂p,q have defined parity since,
given s ∈ Γ̂p,q , we have ŝvs−1 ∈ V, ∀v ∈ V . On the other hand, ŝvs ^ −1 = ŝvs−1 ⇒
s̃−1 vs̄ = ŝvs−1 ⇒ s̃ŝv = vs̄s ⇒ (s̄s)v(s̄s) d −1

= v ⇒ s̄s ∈ ker σ̂ ⇒ s̄s ∈ R∗ . Hence,
s−1 = λs̄ for some λ ∈ R∗ . However, ŝvs−1 = −ŝvs−1 ⇒ s−1 vs̃ = ŝvs−1 ⇒ s̄ŝv =
d
vs̃s ⇒ (s̃s)v(s̃s) −1
= v ⇒ s̃s ∈ ker σ̂ ⇒ s̃s ∈ R∗ , and therefore s−1 = ks̃, for some
k ∈ R∗ . It follows that s̄ = cs̃ for some c ∈ R∗ . Now, by calculating the reversion of
both members in the previous equation, we obtain ŝ = cs, which implies that c = ±1.
Hence, ŝ = ±s, and it follows that a representative spinor for T – written as ψ = sΩF
– is respectively even or odd.
In addition, σ̂(s) ∈ O(p, q), for all s ∈ Γ̂p,q . In fact, given v ∈ V , we have kσ̂(v)k2 =
ŝvs−1 ŝvs−1 = svs−1 svs−1 , which can be derived from the previous paragraph. Hence,
it yields kσ̂(v)k2 = svvs−1 = kvk2 ss−1 = kvk2 . Moreover, the twisted Clifford–
Lipschitz group can also be characterised as
Γ̂p,q = {v1 . . . vk ∈ Cℓp,q | vi are non-isotropic vectors, i = 1, . . . , k},
since from the previous result we have σ̂(s) ∈ O(p, q) and, from the Cartan–Dieudonné
theorem, there exist non-isotropic vectors v1 , . . . , vk such that σ̂(s) = Sv1 ◦ · · · ◦ Svk ,
where Sv denotes a reflexion with respect to the hyperplane orthogonal to v. Since
σ̂(v) = Sv , it follows that σ̂(s) = σ̂(v1 ) . . . σ̂(vk ) = σ̂(v1 . . . vk ). In addition, since ker
σ̂ = R∗ , we arrive at s = λv1 . . . vk , for some λ ∈ R∗ . Finally, if we redefine v1 7→ λv1 ,
then we obtain the aforementioned result.
Let us suppose now that u is a pure spinor.5 Supposing that u is an even semispinor,
then uC is an odd semispinor, where the notation ( )C indicates charge conjugation.
5 Considering n = dimR V , if n = 4 or 6, all Weyl spinors are pure, namely, xu = 0, ∀x ∈ F ⊂ VC .
184 Spinors
As the spinor u is pure, it represents some maximal totally isotropic subspace, let
us say M1 . Since uC is also a pure spinor, it represents some other maximal totally
isotropic subspace, M2 . If M1 and M2 are two maximal totally isotropic subspaces,
then M1 and M2 have the same parity if and only if dimK (M1 ∩ M2 ) = r mod 2.
Indeed, by the Witt theorem, there exists σ ∈ O(p, q) – equivalently, s ∈ Γ̂r,r – which
leads M2 to F via the application σ̂(s). If ψ1 and ψ2 are representatives of M1 and M2 ,
respectively, then ΩF = sψ2 (since ψ2 = s′ ΩF for some s′ ∈ Γ) and, if ψ = sψ1 , then
ψ is a representative for M = σ̂(s)(M1 ). However, since s ∈ Γ̂r,r has defined parity,
then ψ and ΩF have the same parity with respect to the grade involution if and only
if ψ1 and ψ2 have the same parity. In addition, M ∩ F = σ̂(s)(M1 ∩ M2 ), and it is
sufficient to prove that the representative spinors for M and F are either both odd
or both even if dim(M ∩ F ) = r mod 2. When a Witt basis is adopted for M and
F , then a representative ψ for M has the canonical form ψ = vh+1 . . . vr ΩF , where
dim(M ∩ F ) = h. Therefore, ψ and ΩF are either both even or both odd if r − h = 0
mod 2, namely, h = r mod 2.
Another property asserts the conditions under which a linear combination of pure
spinors comprises a pure spinor, the demonstration of which can be seen in the work
by Chevalley (1954), Benn and Tucker (1987), and Crumeyrolle (1990). Given spinors
ψ1 and ψ2 which respectively represent the maximal totally isotropic subspaces T1 and
T2 of V , a necessary and sufficient condition for ψ1 + ψ2 be pure is that dim(T1 ∩ T2 )
= r or r − 2.
Observation ☞ In general, it is always possible to choose a set of pure spinors as a basis
for the spinor space, and any semispinor is a linear combination of pure spinors with
definite parity. From the result dim(T1 ∩ T2 ) = r, or r − 2, another result previously
obtained asserts that dim(T1 ∩ T2 ) = r mod 2. This result then implies that a linear
combination of pure spinors of same parity is a pure spinor if r ≤ 3, and therefore all
semispinors are pure for r ≤ 3.
Now, every (r − 1)-dimensional maximal totally singular subspace M1 is contained
in exactly one maximal totally isotropic subspace, which is either even or odd, of V .
To see this fact, remember that it is always possible to transform M1 in a subspace of
F from the action of an element of the orthogonal group. Hence, it suffices to consider
this assertion for a subspace M1 ⊂ F . Let Z be a maximal totally isotropic subspace
containing M1 . If Z has the same parity as F , then we have dim (Z ∩ F ) = r mod 2.
However, since dim (Z ∩ F ) ≥ r − 1, and dim Z = r, and taking into account that F
contains at least a subspace M1 of dimension r − 1, then Z = F . If Z does not have
the same parity as F , then for the same reasoning it follows that Z ∩ F = M1 . Let u
be a representative spinor for Z, written as
u = av1 . . . vr−1 + bv1 . . . vr , a, b ∈ K, vr ∈ F, vr ∈

/ M1 . (6.120)
Since u is an element of defined parity, then u = av1 . . . vr , and therefore Z is uniquely

determined by the representative spinor.
A very interesting case for its applications for the use of the Clifford algebras
Cℓ1,3 (C) ≃ Cℓ3,1 (C) is to consider the space V = R3,1 . If F is a maximal totally
isotropic subspace of VC , then, supposing that M2 is another maximal totally isotropic
Pure Spinors 185
subspace, F and M2 have the same dimension, dimC M2 = 2, like all maximal totally
isotropic subspaces of VC . Let M1 and M2 have, respectively, semispinors u1 – even –
and u2 – odd – which represent them and which are pure, from the previous obser-
vation. We already demonstrated that dimC (M1 ∩ M2 ) = r mod 2 if and only if M1
and M2 have the same parity. Now, since in this case we have r = 2, then necessarily
dimC (M1 ∩ M2 ) must be odd. Indeed, dimC (M1 ∩ M2 ) = 1, since M1 and M2 have
different parities. Hence, M1 ∩ M2 = span{y}, where y ∈ VC , and it follows from a
former result that y2 = 0.
Defining Ω̊ as a 2r-form such that Ω̊2 = 1, the idempotents P± = 12 (1 ± Ω̊) re-
duce Cℓ+ 3,1 (C) to simple ideals. Then, Ω̊ = ie0123 = iΩ. Denoting uC as the charge
conjugation of u, then both u and uC have different parity.
If u is a pure spinor, then
yu = 0, ∀y ∈ M1 ∩ M2 . (6.121)
In the same way, uC is a pure spinor, and eqn (6.121) implies that
y∗ uC = 0 . (6.122)
In addition, y ∈ M1 ∩M2 implies that y∗ ∈ M1 ∩M2 , where y∗ denotes the C-conjugate

of y, since M1 ∩ M2 = span{y} has complex dimension equal to 1. Hence, y∗ = λy,
where λ ∈ C; since (y∗ )∗ = y, it follows that
y = (y∗ )∗ = (λy)∗ = λ∗ y∗ = λ∗ λy ⇒ λ∗ λ = 1,
and we can conclude that λ ∈ S 1 . There exists µ ∈ S 1 such that λ = µ2 and there
also exists an element p ∈ VC such that p = µy. This result yields
p∗ = (µy)∗ = µ∗ y∗ = µ∗ λy
(6.123)
= µ∗ (µ2 )y = (µ∗ µ)µy = µy = p.
Therefore, p is a real isotropic vector lying in the intersection of two maximal totally
isotropic subspaces. Moreover, the sum of a pure spinor with its charged conjugated
spinor is annihilated by p. In fact, we have
p(u + uC ) = pu + puC = 0, (6.124)
where eqns (6.121) and (6.122) are taken into account. The vector p is determined up
to the multiplication by a real constant, defining thus a projective space.
Let us suppose now that u represents T – we know that dimC M1 = 2 – and let
us consider a basis {p, x} of T . Then, p(xuC ) = −x(puC ) = 0, from the properties
puC = 0, and px = −xp. Moreover, obviously the equation
x(xuC ) = x2 uC = 0 (6.125)
holds, since x is isotropic. Therefore, p(xuC ) = 0; pu = 0; and xuC and u are pure
spinors that represent T .
186 Spinors
Since the space of the spinors that represent T is one-dimensional over the complex
field, there exists η ∈ C∗ – that the possibility that η = 0 does not exist since x is not
in the subspace represented by uC – such that xuC = ηu. Defining ω = η −1 x ∈ M1 ,
we then obtain ωuC = η −1 xuC = η −1 ηu = u. It follows that
ωuC = u . (6.126)
By taking the charge conjugation in eqn (6.126), we obtain
(ωuC )C = uC ⇒ ω ∗ u = uC . (6.127)
Moreover, ωω ∗ u = ωuC = u; since ω ∈ T , then ωu = 0, yielding ω ∗ ωu = 0. In

addition, (ωω ∗ + ω ∗ ω)u = u, implying that
(ωω ∗ + ω ∗ ω) = 1. (6.128)
Define a vector a ∈ VC as a = ω + ω ∗ . We assert that a is a unit vector. Indeed,
a2 = (ω + ω ∗ )(ω + ω ∗ ) = ω 2 + ωω ∗ + ω ∗ ω + (ω ∗ )2
= ωω ∗ + ω ∗ ω = 1.
On the other hand,
a(u + uC ) = au + auC = (ω + ω ∗ )u + (ω + ω ∗ )uC

= ωu + ω ∗ u + ωuC + ω ∗ uC = u + uC .
The equivalence class [a] is defined from the relation a ∼ a + σp, where σ ∈ C. Indeed,
the unit vector a is defined up to the sum of a scalar multiple of y, since
(a + σp)(u + uC ) = a(u + uC ) + σp(u + uC ) = a(u + uC ). (6.129)
The sum u + uC is a Majorana spinor, as in fact this spinor is an eigenspinor of

the charge conjugation operator C:
C[ψ] ≡ ψC = (u + uC )C = uC + u = ψ. (6.130)
6.9 Dual Rotations, and the Penrose Flagpole

Penrose flagpoles can be characterised immediately from the formalism introduced
in section 6.8. For more details, see, for example, the work by Penrose and Rindler
(1984) and Benn and Tucker (1987). First, let u1 and u2 be pure spinors representing
⌢
M1 and M2 , where dimC (M1 ∩ M2 ) = k; then, hu2 u1 ik = ΩM1 ∩M2 . Here, ΩM1 ∩M2
⌢
denotes the volume element M1 ∩ M2 , and ψ denotes the adjoint spinor associated
with ψ with respect to the spinor inner product h̃(ψ, φ) in eqn (6.56). This result is a
theorem whose proof can be seen in the work by Benn and Tucker (1987). Hence, since
span{y} = M1 ∩ M2 , and dimC√ (M1 ∩ M2 ) = 1, therefore hiuuC i1 is a scalar multiple
of p = ΩM1 ∩M2 . The factor i = −1 is introduced in order to turn p = hiuuC i1 into a
real vector. In the Lorentzian case, we can always choose a basis for the spinor space
Dual Rotations, and the Penrose Flagpole 187
such that the charge conjugation is equivalent to the spinor component conjugation
(Chevalley, 1954). In this way,
(u1 , u2 )∗ = ((u1 )C , (u2 )C ), (6.131)
where ( , ) is the spin product with adjoint induced by the reversion, namely, (ψ, φ) =
⌢
ψ(φ) = h̃(ψ, φ). By taking a basis {ej } of VC , it follows that
⌢ ⌢
hiu u C i1 = hiu u C ej i0 ej = huC iej ui0 ej = (uC , iej u)ej . (6.132)
Then,
⌢ ⌢
hiu u C i∗1 = −(u, iej uC )ej = −(iej u, uC )ej = (uC , iej u)ej = hiu u C i1 . (6.133)
Therefore, p∗ = p, where p ∈ V is determined up to a real scalar. The vector p is

identified modulo a real scalar to a family of coplanar vectors, which determine the
Penrose flagpole. V
Let now {p, ω} be another basis of T , where ω ∈ 1 (C3,1 ) satisfies ωuC = u and is
well defined up to a scalar multiple of x, since xuC = 0. From the previously obtained
⌢
results, we know that iu u is a complex multiple of xω. Since puC = 0, therefore
[ω] = ω + ζp, ζ ∈ C. Hence, ω is determined up to the sum with a scalar multiple
of p.
Likewise, another possible characterisation for a pure spinor is given by the asser-
⌢
tion that a spinor u is pure if and only if hu u ik = 0, for all k 6= r, where r = dimC M1
(Benn and Tucker, 1987). Therefore, in this case, if M1 = M2 , then M1 ∩ M2 = M1 ,
⌢
and dimC (M1 ∩ M2 ) = dimC M1 = 2, where one concludes that hu u i2 , which is the
volume element of M1 ∩ M2 = M1 , is a scalar multiple of pω. Suppose that this scalar
is 2eiθ . It follows that
⌢ ⌢
hiu u i2 = iu u = 2 exp(iθ)pω. (6.134)
When the flagpole is defined as
1 ⌢ ⌢
G= (iu u − iuC u C ) , (6.135)
2
it is possible to write it as
G = exp(iθ)pω + exp(−iθ)pω ∗ , (6.136)
and, therefore,
G(ω + ω ∗ ) = (exp(iθ)pω + exp(−iθ)pω ∗ )(ω + ω ∗ )
= exp(iθ)pω 2 + exp(iθ)pωω ∗ + exp(−iθ)pω ∗ ω + exp(−iθ)p(ω ∗ )2
= cos θpωω ∗ + i sin θpωω ∗ + cos θpω ∗ ω − i sin θpω ∗ ω
= cos θp(ωω ∗ + ω ∗ ω) + i sin θp(ωω ∗ − ω ∗ ω)
= cos θp + 2i sin θpω ∧ ω ∗
= cos θp + 2i sin θp ∧ ω ∧ ω ∗ .
(6.137)
188 Spinors
The last equality comes from the fact that M1 is a maximal totally isotropic subspace,
and p · ω = 0. From the expression (6.136), when θ = 0 we obtain
G θ=0
= p(ω + ω ∗ ) = pa = p ∧ a, (6.138)
since p and a are elements of M1 , which is isotropic. Hence,
G θ=0
= F = p ∧ a. (6.139)
From the definition given in eqn (6.135), we can assert that

⌢ ⌢
2G(ω + ω ∗ ) = iu u (ω + ω ∗ ) − iuC u C (ω + ω ∗ )
= iu[(ω + ω ∗ )u]⌢ − iuC [(ω + ω ∗ )uC ]⌢
⌢ ⌢ ⌢ ⌢ (6.140)
= iu(ωu + ω ∗ u) − iuC (ωuC + ω ∗ uC )
⌢ ⌢
= iuuC − iuC u ,
since ω ∗ u = uC ; ωuC = u; and ωu = 0.

Equation (6.139) can be written as
⌢
F = Re(iu u ) . (6.141)
Indeed, considering ϕ, ξ, φ, and ψ as arbitrary spinors, we know that
] ⌢ ⌢
(ϕ, (φ ψ) ξ) = ((φ ψ)ϕ, ξ) = (ψ, ϕ)(φ, ξ) = −(ϕ, ψ)(φ, ξ)
(6.142)
] ⌢
= −(ϕ, (ψ φ) ξ),
] ⌢ ] ⌢
and then (φ ψ) = −(ψ φ ). It follows from eqn (6.140) that
⌢ ⌢
G(ω + ω ∗ ) = iuuC − iuC u
(6.143)
⌢ ^ ⌢
= iuuC + (iuuC ),
d⌢ ⌢ g⌢ g⌢ ⌢
and, since iΩψ = ψ, therefore uuC = −ΩuuC Ω = −Ωu(ΩuC ) = −iΩu(iΩuC ) = −uuC ,
⌢ V V ⌢
where we conclude that uuC ∈ 1 (C3+1 ) ⊕ 3 (C3+1 ). Since huuC i3 changes sign under
reversion, in order for eqn (6.143) to give a non-trivial solution for G(ω+ω ∗ ), it follows
⌢
that G(ω + ω ∗ ) = huuC i1 = p, from eqn (6.133). For the equation G θ=0 = F = p ∧ a
to hold, the required result must follow.
Now, both the real vector p and the bivector F can be written from a Majorana
spinor and the volume element of a maximal totally isotropic subspace C3,1 as p =
⌢ ⌢
1 1
2 hψ(Ωψ)i1 , and respectively F = 2 hψ(Ωψ)i2 . Here, ψ is a Majorana spinor, and
Dual Rotations, and the Penrose Flagpole 189
ΩF = u0123 is the volume element of F ⊂ VC . In fact, a Majorana spinor can be

written as ψ = u + uC ; since u = iΩu, therefore
⌢ ⌢ ⌢ ⌢
ψ(Ωψ) = ((u + uC )(z(u + uC ))) = (u + uC )((u + uC ) z )
⌢ ⌢ ⌢ ⌢⌢ ⌢⌢
= (u + uC )(( u + uC ) z ) = (u + uC )( u z + uC z )
⌢ ⌢
= (u + uC )(−i u + iuC ) (6.144)
⌢ ⌢ ⌢ ⌢
= −iu u + iuC uC + iuuC − iuC u
⌢ ⌢ ⌢
= (−iu u + iuC uC ) + (iuuC ^
+ i(uuC )) ,
] ⌢ ⌢ ⌢
since u = izu implies that uC = −izuC . Now, since (ψ φ) = −φ ψ, the term −iu u +
⌢ ⌢
^
iuC uC is odd under a reversion. However, the term iuuC + i(uuC ) is even; this result
⌢ ⌢
implies from eqn (6.133) that p = 21 hψ(zψ)i1 , and F = 21 hψ(zψ)i2 .
Here, we stated that iΩu = u, where Ω = u0123 is the volume element associated
with the maximal totally isotropic subspace F ⊂ VC ; this statement can be forthwith
proved. The object Ω̊ was defined as a 2r-form such that Ω̊2 = 1, so that the idempo-
tents P± = 12 (1 + Ω̊) reduce Cℓ+3,1 (C) to simple ideals; consequently, Ω̊ = ie0123 = iΩ.
Since u was defined as an even spinor, therefore Ω̊u = Ω̊sΩF = sΩ̊ΩF = sΩF = u, since
u = sΩF is the form of u that represents F . Finally, in order to see that ΩΩF = ΩF ,
it suffices to see that, in general, given a Witt basis {ui , vj } for VC , we have that
Ω = u1 ∧ v1 ∧ u2 ∧ v2 ∧ · · · ∧ ur ∧ vr = [u1 , v1 ][u2 , v2 ] . . . [ur , vr ] and, therefore,
ΩΩF = [u1 , v1 ][u2 , v2 ] . . . [ur , vr ]u1 u2 . . . ur

= [u1 , v1 ]u1 [u2 , v2 ]u2 . . . [ur , vr ]ur (6.145)
= u1 u2 . . . ur = u1 ∧ u2 ∧ · · · ∧ ur = ΩF ,
since [ui , vi ]ui = ui vi ui = (1 − vi ui )ui = ui . Here, the obvious notation for the
commutator [a, b] = ab − ba is used, for a, b ∈ Cℓp,q .
If v denotes a pure spinor related to u by
v = exp(iθ)u, (6.146)
then ⌢ ⌢ ⌢
hivvC i1 = hi exp(iθ)u exp(−iθ)uC i1 = hiuuC i1 = p, (6.147)
and therefore the spinor v determines the same null direction that u does.
If v determines the 2-form F ′ , then
⌢ ⌢
F ′ = Re(iv v ) = Re(i exp(iθ)u exp(iθ) u )
⌢ ⌢
= Re(i cos(2θ)u u − sin(2θ)u u )
⌢ ⌢
(6.148)
= cos(2θ)Re(iu u ) − sin(2θ)Re(Ωiu u )
= cos(2θ)F − sin(2θ)ΩF.
Consequently,
190 Spinors
F ′ = exp(−2θΩ)F. (6.149)
′
Therefore, F is associated with F by a dual rotation, since
⋆ F = F̃ Ω = ΩF̃ = −ΩF, (6.150)
and it follows that
ΩF = − ⋆ F . (6.151)
It finally reads
F ′ = cos(2θ)F + sin(2θ) ⋆ F , (6.152)
and we thus identify F with the Penrose flagpole structure. In addition, the flagpole F
rotates by the angle 2θ, and the spinor u associated with F rotates by θ. This flagpole
is a generalisation of the Penrose flagpole, which is a particular case when θ = 0.
6.10 Weyl Spinors in Cℓ3,0

This section aims to uniquely introduce the Weyl spinors from the Clifford algebra
Cℓ3,0 , which is exactly the Penrose formalism in an algebraic language which is more
general and accessible than that used in the Penrose formalism (Penrose and Rindler,
1984). Moreover, this formulation emulates the van der Waerden framework (van der
Waerden, 1928; Veblen, 1933), which is revisited in the appendix. First, let us define
the idempotents f± = 21 (1 ± e3 ) – which satisfy the relations f+ f− = f− f+ = 0, and
2
f± = f± . Using the isomorphism Cℓ3,0 ≃ M(2, C) given by ei 7→ σi , where σi denotes
the Pauli matrices, it is immediately obvious that

10 00
f+ 7→ , e1 f+ 7→ ,
00 10
(6.153)
00 01
f− 7→ , e1 f− 7→ .
01 00
The isomorphism Cℓ3,0 f+ ≃ Cℓ+

3,0 f+ is straightforward to realise

w1 −w2∗ w1 0
Cℓ+ f
3,0 + ∋ φ f
+ + = f + =
w2 w1∗ w2 0

w1 w3 10
≃ ∈ Cℓ3,0 f+ .
w2 w4 00
An algebraic spinor can be written as K = ψf+ , where ψ = s+b12 e12 +b13 e13 +b23 e23 ∈
Cℓ+
3,0 , and the coefficients are real numbers. We can define the
• undotted contravariant spinor:

K = ψf+ = (s + b12 e123 )(f+ ) + (b13 + b23 e123 )(e1 f+ )
(6.154)
= k 1 (f+ ) + k 2 (e1 f+ ),
where k 1 = s + b12 e123 , and k 2 = b13 + b23 e123 . Spinors are expressed in such a way
that their components commute with the basis {f+ , e1 f+ } of the space of the algebraic
spinors. Hence, all components are written as elements of the centre of Cℓ3,0 .
From the spinor K, three other types of spinors can be defined. The first is the
Weyl Spinors in Cℓ3,0 191
• undotted covariant spinor:
K∗ = e1 K = e1 (k1 f+ + k2 e1 f+ ) = e1 (f− k 1 + f− (−e1 )k 2 )

(6.155)
f = (−k 2 )f+ + (k 1 )(f+ e1 ).
Since K∗ ∈ f+ Cℓ3,0 , we can write K∗ = k1 (f+ ) + k2 (f+ e1 ). The relationships between

the upper and lower components read
k1 = −k 2 , k2 = k 1 ,
and are similar to those in the standard van der Waerden formalism. For more details,
see the work by van der Waerden (1928), Veblen (1933), Penrose (1967), and Penrose
and Rindler (1984).
Given an algebraic spinor K∗ ∈ f+ Cℓ3,0 and L = η 1 f+ + η 2 e1 f+ ∈ Cℓ3,0 f+ – where
η and η 2 are the spinor L components that are elements of the centre Cℓ3,0 – the
1
spinor metric associated with the idempotent f+ can be defined as follows:
Gf+ : f+ Cℓ3,0 × Cℓ3,0 f+ → f+ Cℓ3,0 f+ ≃ Cf+ (6.156)

(K, L) 7→ Gf+ (K, L) = K∗ L
= (−k 2 f+ + k 1 f+ e1 )(η 1 f+ + η 2 e1 f+ ) ,
which leads to
Gf+ (K, L) = K∗ L = (−k 2 η 1 + k 1 η 2 )f+ . (6.157)
This definition coincides with the classical spinor definition, where the scalar product
has mixed and antisymmetric components, and f+ plays the role of the unit in the
algebra f+ Cℓ3,0 f+ ≃ C.
Moreover, from K we define the
• dotted contravariant spinor:
K = e1 K̃ = e1 (k 1 f+ + k 2 e1 f+ )e = e1 (f+ k˜1 + f+ e1 k˜2 )

(6.158)
= k˜1 (e1 f+ ) + k˜2 f− = k˜1 (f− e1 ) + k˜2 (f− ).
1̇ 2̇
Since K ∈ f− Cℓ3,0 , we can write K = k (f− e1 )+k (f− ), obtaining the equivalences
1̇ 2̇
k = k˜1 , k = k˜2 .
In addition,6 k˜α = (a + be123 )e = (a + be321 ) = a − be123 , which suggests the notation

k˜α = k α .7
Finally, we construct the
6α = 1, 2.
7 By denoting Cℓs the scalars and Cℓp the pseudoscalars, we obtain the isomorphism Cℓs ⊕ Cℓp ≃ C,
since (e2123 = −1). The notation k˜α = kα is therefore evident, since e123 plays the role of the imaginary
unit of C, and the reversion in Cℓp is equivalent to the complex conjugation.
192 Spinors
• dotted covariant spinor:

1̇ 2̇
K∗ = (e1 K) = −(K)e1 = −k (f− e1 ) + k (f− )e1
1̇ 2̇ 1̇ 2̇ (6.159)
= −(−e1 f+ k + f+ k )e1 = k f− − k f+ e1
2̇ 1̇
= (−k )(e1 f− ) + (k )(f− ).
Since K∗ ∈ Cℓ3,0 f− , we can write K∗ = (k 1′ )(e1 f− ) + (k 2̇ )(f− ), and the expressions
2̇ 1̇
k 1̇ = −k , k 2̇ = k (6.160)
hold.
Given K ∈ f− Cℓ3,0 , and L∗ = η 2̇ e1 f+ + η 1̇ f− ∈ Cℓ3,0 , the spinor metric associated
with the idempotent f− is defined as follows:
Gf− : f− Cℓ3,0 × Cℓ3,0 f− → f− Cℓ3,0 f− ≃ Cf−
1̇ 2̇
(K, L∗ ) 7→ KL∗ = (k f− e1 + k f− )(−η 2̇ e1 f− + η 1̇ f− ),
and, therefore,
2̇ 1̇
Gf− = (η 1̇ k − η 2̇ k )f− . (6.161)
The expressions for the four types of Weyl spinors, as elements of an ideal of Cℓ3,0 ,
are as follows:
• undotted contravariant spinor:
K = k 1 (f+ ) + k 2 (e1 f+ ) ∈ Cℓ3,0 f+
• undotted covariant spinor:
K∗ = e1 K = k1 (f+ ) + k2 (f+ e1 ) ∈ f+ Cℓ3,0
• dotted contravariant spinor:

1̇ 2̇
K = e1 K̃ = k (f− e1 ) + k (f− ) ∈ f− Cℓ3,0
• dotted covariant spinor:
K∗ = −(K)e1 = (e1 K) = k 1̇ (e1 f− ) + k 2̇ (f− ) ∈ Cℓ3,0 f−
Taking now into account the operations already defined, the diagram in figure 6.1
illustrates how to pass from one ideal to another in Cℓ3,0 .
K −→ K∗ −→ K −→ b
K∗ = K
∗ −( b ) ∗
↑ ↑ ↑ ↑
contravariant covariant contravariant covariant
undotted undotted dotted dotted
Cℓ3,0 f+ f+ Cℓ3,0 f− Cℓ3,0 Cℓ3,0 f−
Figure 6.1 Passing from One Ideal to Another in Cℓ3,0
Weyl Spinors in the Clifford Algebra Cℓ0,3 ≃ H ⊕ H 193
6.11 Weyl Spinors in the Clifford Algebra Cℓ0,3 ≃ H ⊕ H

Consider the Euclidean space R3 , and an orthonormal basis {e1 , e2 , e3 }. The Clifford
algebra Cℓ0,3 is defined by the relations g(ei , ej ) = −δij = 21 (ei ej + ej ei ), where i, j =
1, 2, 3, and, in particular, e2i = −1.
By using the very same procedure as we did for Cℓ3,0 , we first reduce the redundant
dimensions, proving that Cℓ0,3 f+ = Cℓ+ 1
0,3 f+ , where f± = 2 (1 ± e123 ). Indeed, the left
ideal Cℓ0,3 f+ is algebraically isomorphic to H. In addition, writing an arbitrary element
of Cℓ0,3 as
A = a0 + ak ek + b1 e23 + b2 e31 + b3 e12 + b0 e123

= a0 + ak ek − bk ek e123 + b0 e123 ∈ Cℓ0,3 ,
we can see that
Af+ = [(a0 + b0 ) + (ak − bk )ek ]f+ = [(a0 + b0 ) + (ak − bk )ek e123 ]f+ .
Hence, given Af+ ∈ Cℓ0,3 f+ , by writing A′ = (a0 + b0 ) + (ak − bk )ek e123 , we can see
that A′ ∈ Cℓ+ ′ +
0,3 , and Af+ = A f+ . These results show that Cℓ0,3 f+ ⊂ Cℓ0,3 f+ . The other
inclusion immediately follows.
Consider an even element Q ∈ Cℓ+ 0,3 :
Q = a + be12 + ce13 + de23 = (a + be12 ) + e13 (c − de12 )

= k 1 + e13 k 2 .
Another way to describe Weyl spinors is to consider the algebra Cℓ0,3 ≃ H ⊕ H, where
a spinor K = Qf+ ∈ Cℓ0,3 f+ can be expressed as an undotted contravariant spinor:
K = (k 1 f+ + e13 k 2 f+ ).
Likewise, we can also define a dotted contravariant spinor K = (f+ k̄ 1 − f+ k̄ 2 e13 ).
Multiplying the conjugate of K to the left by e13 , we obtain
e13 K = e13 (f+ k 1 + f+ k 2 e31 ) = f+ k 1 e13 + f+ k 2 ,
and by performing the product with another spinor L ∈ Cℓ0,3 f+ , we obtain
e13 KL = k 1 η 1 e13 f+ − k 1 η 2 f+ + k 2 η 1 f+ + k 2 η 2 e13 f+ .
Hence, the spinor metric in Cℓ0,3 can be obtained:
G(K, L) = 2h(e13 K)Li0 = (k 2 η 1 − k 1 η 2 )f+
Consider now a mapping σ : Cℓ+ +

0,3 → Cℓ0,3 defined by the expression
σ(Q) = e32 Qe23 .

194 Spinors
The mapping σ turns a right module into a right module. Indeed,
σ(k 1 + e13 k 2 ) = σ(a + be12 + ce13 + de23 )

= e32 (a + be12 + ce13 + de23 )e23
= (a + be12 ) + (c − de12 )e13 = k 1 + k 2 e13 .
For a spinor K = Qf+ , it follows that
σ(ψ) = σ(Qf+ ) = e32 (Qf+ )e23 = e32 (f+ Qe23 )

= f+ e32 Qe23 = f+ σ(Q)
= f+ (k 1 + k 2 e13 ).
In this way,
σ(K)e13 = f+ (k 1 e13 − k 2 ) = f+ (−k 2 + k 1 e13 ) = K∗ .
The spinor metric can also be defined as follows:
1
G(ψ, φ) = hσ(ψ)e13 φiC = [σ(ψ)e13 φ + e21 σ(ψ)e13 φe12 ] .
2
The algebra Cℓ0,3 is not as natural for the spacetime metric description as Cℓ3,0 is. In
fact, it is suitable for the Euclidean space R4 since, for an element u ∈ R4 , we have
uū = u20 + ~u2 , where u0 ∈ R, ~u ∈ R3 . In addition, Cℓ0,3 ≃ H ⊕ H is a semisimple
algebra, and the ring H is not commutative. Thus, we must distinguish the left and
right products in H. Furthermore, it has been proven that there exists a mapping σ
that leads one module to the other.
6.12 Spinor Transformations

Consider now an arbitrary element
R = s + v i ei + bij eij + pe123 = α + βe12 + γe13 + δe23 ∈ Cℓ3,0 ,
where
α = s + pe123 , β = b12 − v 3 e123 , γ = b13 + v 2 e123 , and δ = b23 − v1 e123
are elements of the centre of Cℓ3,0 . Under the action of R, an undotted contravariant
spinor K behaves as RK = R(ψf+ ) = k 1 (Rf+ ) + k 2 (Re1 f+ ), where
Rf+ = (α + βe123 )f+ + (γ + δe123 )(e1 f+ ),

Re1 f+ = (α − βe123 )f+ + (−γ + δe123 )(e1 f+ ).
A matrix representation ρ : Cℓ3,0 → M(2, C) of R is given by

α + βi −γ + δi
ρ(R) = .
γ + δi α − βi
Spinor Transformations 195
It follows that det ρ(R) = α2 +β 2 +γ 2 +δ 2 . Under the morphisms in Cℓ3,0 , the element
R ∈ Cℓ3,0 transforms as
R̂ = α + βe12 + γe13 + δe23 ,

R̃ = α − βe12 − γe13 − δe23 ,
R = α − βe12 − γe13 − δe23 .
Hence, the relation

RR = α2 + β 2 + γ 2 + δ 2 = det ρ(R)
holds. In this way, given R ∈ Cℓ3,0 and demanding that R ∈ $pin+ (1, 3), namely,
RR = 1, we find that det ρ(R) = 1, and hence R ∈ SL(2, C). Consequently, the
isomorphism
$pin+ (1, 3) ≃ SL(2, C)
has been established. From the condition RR = 1, it follows that R = R−1 . Therefore,
the transformation rules
K 7−→ RK, (6.162)

∗ ∗ −1
K 7−→ e1 (RK) = e1 K̄R̄ = K R , (6.163)
] = e1 K̃R̃ = K(R̂) = K(R̂)−1 ,
K 7−→ e1 (RK) (6.164)
d = R̂K∗ ,
K∗ 7−→ (RK) (6.165)
hold for Weyl spinors. Such transformations on algebraic Weyl spinors are well known
from the point of view of classical spinors. Indeed,
†
ρ(R̂) = [ρ(R) )]−1 .
When the representation of each one of the four spinors is taken into account, the
correspondences
1 !
k 0 0 0 0 0
K ←→ , K ←→ 1̇ 2̇ = ,
k2 0 k k −k 2̇ k 1′
2 1
∗ −k k ∗ 0 k 1′
K ←→ , K ←→
0 0 0 k 2̇
hold. To summarise, in order to formulate the four types of Weyl spinors – and sub-
sequently the Dirac spinor (Hladik, 1999) – it suffices to consider the ideal Cℓ3,0 f+ .
Thus, the right and the left Clifford multiplication by the element e1 yield the other
ideals – f+ Cℓ3,0 , Cℓ3,0 f− , f− Cℓ3,0 , whose elements are the other three types of Weyl
spinors. The results obtained in this section can be compared to those in the appendix,
which uses the dotted/undotted van der Waerden notation.
196 Spinors
6.13 Spacetime Vectors as Paravectors of Cℓ3,0 from Weyl Spinors
Paravectors
One of the main benefits of the formalism presented in this chapter is the Penrose
interpretation of a spinor as being a structure that is more basic than a point in
spacetime is (Penrose and Rindler, 1984). Thus, it is natural to construct spacetime
vectors from Weyl spinors.
Let us discuss first the concept of the paramultivector (Baylis, 1996), in particular,
the paravector, which is the sum of a scalar and a vectors. One of the main purposes
for the introduction of this concept is to reduce the dimension of the algebra, in
order to formulate a theory. The main advantage of a minimalist formulation is that
redundancies are precluded and only those elements that are strictly needed are used.
A paravector is defined as an element of R ⊕ Rp,q ⊂ Cℓp,q . We are here interested in
the case n = 3, where Cℓ3,0 is the Pauli algebra. We shall now show how an arbitrary
element ψ of Cℓ3,0 can be written as a complex paravector.
When the orthonormal basis for Cℓ3,0 is {1, ei , ei ej , e123 }, the pseudoscalar e1 e2 e3 ≡
I satisfies I2 = −1 and is an element in the centre of Cℓ3,0 . Hence, the algebra Cℓ3,0 is
isomorphic to an algebra with half of the number of elements – when considered over
C – where I plays the role of the imaginary unit.
With respect to a C-structure, every bivector in Cℓ3,0 can be written as an imag-
inary vector, for instance, e1 e2 = e1 e2 e3 e3 = Ie3 = e3 I. This vector is precisely
the dual Hodge operator acting on basis vectors, and a basis for Cℓ3,0 in this con-
text is given by {e2 e3 , e3 e1 , e1 e2 }, where we use the isomorphism ρ : Cℓ3,0 → Cℓ+ 1,3 ,
ei 7→ ρ(ei ) = γi γ0 . In addition, we see that
ei ej = −γi γj , ej ek el = γ0 γj γk γl . (6.166)
Remember that {γµ }3µ=0 denotes a basis for Cℓ1,3 .

Parabivectors andVparatrivectors
V are also V in Cℓ3,0 and defined respec-
V introduced
tively as elements of 1 (R3 ) ⊕ 2 (R3 ), and 2 (R3 ) ⊕ 3 (R3 ). From (6.166), bivectors
of Cℓ1,3 correspond to parabivectors of Cℓ3,0 . Here, we will focus on paravectors; for
further details, see, for instance, the work by Baylis (1996).
An element ψ ∈ Cℓ3,0 is the sum of a scalar p0 and a vector p, both of which are,
in general, complex structures, where p0 = Re(p0 ) + Im(p0 ), and p = Re(p) + Im(p).
In addition, Re(p0 ) = hψi0 ; Im(p) = hψi2 ; Re(p) = hψi1 ; and Im(p0 ) = hψi3 . The
object ψ is said to be a C-paravector.
Paravector Automorphisms
We shall describe now how the grade involution, the reversion, and the conjugation
act on paravectors. The conjugation reverses the vector part of ψ:
ψ = p0 + p 7→ ψ̄ = p0 − p. (6.167)
The conjugation is then called spatial reversion in this context. Any element of Cℓ3,0
can be written as ψ = 21 (ψ + ψ̄) + 12 (ψ − ψ̄) = hψi0 + hψi1 , where hψi0 and hψi1
Spacetime Vectors as Paravectors of Cℓ3,0 from Weyl Spinors 197
are, respectively, the scalar and the vector components of ψ. They can be regarded,
respectively, as the scalar and the vector products
1 1
hψφi0 = (ψφ + ψφ), hψφi1 = (ψφ − ψφ). (6.168)
2 2
Thus, a multivector ψ ∈ Cℓ3,0 is a scalar if and only if ψ = ψ̄.
Likewise, the complex conjugation is given by the reversion in Cℓ3,0 . Indeed, if ψ ∈
Cℓ3,0 is written in the paravector basis {e0 = 1, e1 , e2 , e3 }, the complex conjugation is
obtained when the complex conjugate is taken with respect to each coefficient:
ψ = ψ µ eµ 7→ ψ̃ = ψ µ∗ eµ , µ = 0, 1, 2, 3. (6.169)
The complex conjugation is used to split multivectors into real and imaginary parts:
1 1
ψ= (ψ + ψ̃) + (ψ − ψ̃) = Re(ψ) + Im(ψ). (6.170)
2 2
The composition between the complex conjugation and the Clifford conjugation results
in the grade involution, which decomposes a multivector into even and odd parts:
1 b + 1 (ψ − ψ)
b = hψi+ + hψi− .
ψ= (ψ + ψ) (6.171)
2 2
The norm of a paravector is given by the quadratic form
ψ ψ̄ = (p0 + p)(p0 − p) = p20 − p2 . (6.172)
The Lorentzian metric tensor components ηµν associated with the Minkowski space-
time defines the norm of a vector in the same way. We conclude that vectors of R1,3 in
special relativity can be represented by real paravectors of Cℓ3,0 . The metric is given
by 

1, if µ = ν = 0,
ηµν = heµ ēν i0 = −1, if µ = ν = 1, 2, 3, (6.173)


0, if µ 6= ν.
Using Weyl Spinors

Weyl spinors can generate arbitrary paravectors as
2Ke1 K = 2Ke1 e1 K̃ = 2KK̃ ∈ R ⊕ R3 .

1̇ 2̇
Given Weyl spinors K = k 1 f+ + k 2 e1 f+ ∈ Cℓ3,0 f+ , and K = k (f− e1 ) + k (f− ), it
follows that
1̇ 2̇ 1̇ 2̇
KK̃ = Ke1 K = k 1 k f+ + k 1 k f+ e1 + k 2 k f− e1 + k 2 k f− , (6.174)
which leads, from eqns (6.153) and (6.154), to the expression

198 Spinors
1̇ 2̇
!
k1 k k1 k
KK̃ = 1̇ 2̇
. (6.175)
k2 k k2 k
Indeed, a paravector a ∈ R ⊕ R3 ∈ Cℓ3,0 can be written as
a = 2Ke1 K = 2KK̃ , (6.176)
since, given a spinor operator ψ ∈ Cℓ+

3,0 , it follows that
a = 2KK̃ = 2ψf+ f+ ψ̃ = 2ψf+ ψ̃ = ψ(1 + e3 )ψ̃

(6.177)
= ψ ψ̃ + ψe3 ψ̃ = a0 + ai ei .
The paravector a is future pointed, since
R ∋ ψ ψ̃ = a0 = (a + be12 + ce13 + de23 )(a − be12 − ce13 − de23 )

= a2 + b2 + c2 + d2 > 0.
In addition, since ai ai = (ψ ψ̄)2 = (a0 )2 , the paravector a is null: a2 = (a0 )2 − ai ai = 0.

The last equality in eqn (6.177) arises from the fact that any vector in R3 can be
written as x = xi ei = ψe3 ψ̃, which is derived from the reference vector e3 via a
rotation and a dilatation. This expression is the spin density multiplied by the factor
~/2.
Now we will obtain the spacetime metric from Weyl spinors. According to eqn
(6.174), let the two paravectors a and b read
1̇ 2̇ 1̇ 2̇
a = k 1 k f+ + k 1 k f+ e1 + k 2 k f− e1 + k 2 k f− = a0 + ai ei ,
(6.178)
b = r 1 r 1̇ f+ + r 1 r 2̇ f+ e1 + r2 r 1̇ f− e1 + r 2 r 2̇ f− = b0 + bi ei ,
and their respective conjugations be given by
[1̇ [2̇ [1̇ [2̇

a = k 1 k f− − k 1 k e1 f− − k 2 k e1 f+ + k 2 k f+ = a0 − ai ei ,
(6.179)
d d d d
b = r 1 r 1̇ f− − r 1 r 2̇ e1 f− − r 2 r 1̇ e1 f+ + r2 r 2̇ f+ = b0 − bi ei .
The Clifford relation is then obtained for paravectors:

1̇ 2̇ 1̇ 2̇
ab̂ + bâ = (k 1 k r 2 r 2̇ + k 1 k r 2 r 1̇ + k 2 k r 1 r 2̇ + k 2 k r 1 r 1̇ )
(6.180)
= 2(a0 b0 − ai bi ) = 2g(a, b).
From light-like paravectors obtained from the generators {f+ , f− , e1 f+ , e1 f− } of

the four ideals of Cℓ3,0 , it is possible to obtain the tetrad {e0 = 1, e1 , e2 , e3 } in the
paravector space. To start, let us define a basis of light-like paravectors {π µ }:
π 0 = f+ ff
+ = f+ , π 1 = f− ff
− = f− ,
(6.181)
π 2 = e1 f+ ff
+ = e1 f+ , π 3 = f+ e]
1 f+ = e 1 f− .
Paravectors of Cℓ4,1 in Cℓ3,0 via the Periodicity Theorem 199
The orthonormal tetrad is obtained from the basis of paravectors {π µ }:
e0 = π 0 + π 1 , e1 = π 2 − π 3 ,
(6.182)
e2 = π 3 − π 2 , e3 = π 0 − π 1 ,
since
e0 = f+ + f− , e1 = e1 f+ + e1 f− ,
(6.183)
e2 = −e123 (e1 f− − e1 f+ ), e 3 = f+ − f− .
Hence, both the Minkowski space tetrad and the paravector space R ⊕ R3 tetrad are
obtained.
6.14 Paravectors of Cℓ4,1 in Cℓ3,0 via the Periodicity Theorem

Consider the basis {εÅ }5Å=0 of R2,4 ; it obviously satisfies the relations
ε20 = ε25 = 1, ε21 = ε22 = ε23 = ε24 = −1, εÅ · εB̊ = 0 (Å 6= B̊).
Moreover, consider also R4,1 , with the basis {EA }4A=0 , where
E02 = −1, E12 = E22 = E32 = E42 = 1, EA · EB = 0 (A 6= B).
The basis {EA } can be obtained from the basis {εÅ } if we define the isomorphism
^
2,4
ξ : Cℓ4,1 → 2 (R ),
EA 7→ ξ(EA ) = εA ε5 . (6.184)
The basis {EA } defined by eqn (6.184) obviously satisfies eqn (6.14).
Given a vector α = αÅ εÅ ∈ R2,4 , we obtain a paravector b ∈ R ⊕ R4,1 ֒→ Cℓ4,1 if
the element ε5 is left multiplied by b:
b = αε5 = αA EA + α5 . (6.185)
From the periodicity theorem, the isomorphism Cℓ4,1 ≃ Cℓ1,1 ⊗Cℓ3,0 follows and
thus it is possible to express an element of Cℓ4,1 as a 2 × 2 matrix with entries in Cℓ3,0 .
A homomorphism ϑ : Cℓ4,1 → Cℓ3,0 can be defined by
Ei 7→ ϑ(Ei ) = Ei E0 E4 ≡ ei . (6.186)
It follows that e2i = 1, Ei = ei E4 E0 and that E4 = E+ + E− , E0 = E+ − E− , where

E± := 12 (E4 ± E0 ). Hence, the paravector b can be split into
b = α5 + (α0 + α4 )E+ + (α4 − α0 )E− + αi ei E4 E0 . (6.187)

01
If we choose E4 and E0 to be, respectively, represented by E4 = , and E0 =
 
10
0 −1
 , then
1 0
200 Spinors

00 01 1 0
E+ = , E− = , E4 E0 = , (6.188)
10 00 0 −1
and hence the paravector b ∈ R ⊕ R4,1 ֒→ Cℓ4,1 in eqn (6.187) is represented by

5
α + αi e i α4 − α0
b= . (6.189)
α0 + α4 α5 − αi ei
The vector α ∈ R2,4 is an element of the Klein absolute, that is, α2 = 0. In

addition, this condition implies that α2 = 0 ⇔ bb̄ = 0, since α2 = αα = α1α =
αε25 α = αε5 ε5 α = bb̄. We denote
λ = α 4 − α0 , µ = α4 + α0 . (6.190)
Using the matrix representation of bb̄, the entry (bb̄)11 of the matrix is given by
(bb̄)11 = xx̄ − λµ = 0, (6.191)
where
x := (α5 + αi ei ) ∈ R ⊕ R3 ֒→ Cℓ3,0 . (6.192)
If we fix µ = 1, then λ = xx̄ ∈ R. This choice corresponds to a projective description.
Then, the paravector b ∈ R ⊕ R4,1 ֒→ Cℓ4,1 can be represented as

xλ x xx̄
b= = . (6.193)
µ x̄ 1 x̄
From eqn (6.191), we obtain (α5 +αi ei )(α5 −αi ei ) = (α4 −α0 )(α4 +α0 ), which implies
that
(α5 )2 − (αi ei )(αj ej ) = (α4 )2 − (α0 )2 , (6.194)
yielding
(α5 )2 + (α0 )2 − (α1 )2 − (α2 )2 − (α3 )2 − (α4 )2 = 0 , (6.195)
which is the Klein absolute (eqn (5.110)).
6.15 Twistors as Geometric Multivectors

In this section, we present and discuss the Keller approach and also introduce our
definition of twistors, showing how the twistor formulation can be led to the Keller
approach and, consequently, to the Penrose classical twistor framework. The twistor
defined as a minimal lateral ideal is further examined in the book by Crumeyrolle
(1990). Robinson congruences and the incidence relation, which determines a point in
spacetime as a secondary concept obtained from the intersection between two twistors,
are also investigated here.
The Keller Approach

The twistor approach by Keller (1997) uses the projectors PR,L := 21 (1 ± iγ5 ) (where
R and L, respectively, denote the right the left projections, as they are usually called)
Twistors as Geometric Multivectors 201
and the element Tx = 1 + γ5 x, where x = xµ γµ ∈ R1,3 . Now we introduce some of the

results obtained by Keller (1997). Define the reference twistor ηx , which is associated
with the vector x ∈ R1,3 and with the Weyl covariant dotted
spinor (written as the
left-handed projection of a Dirac spinor ω) Π = PL ω = 0ξ by
ηx = Tx PL ω = (1 + γ5 x)PL ω = (1 + γ5 x)Π . (6.196)
In order to show the equivalence of this definition with the Penrose classical twistor
formalism, the Weyl representation is used:

I0 −i2 0 0 ~x 0
ηx = (1 + γ5 x)Π = + . (6.197)
0I 0 i2 ~xc 0 ξ
Each entry in these matrices denotes a 2 × 2 matrix, and the vector representation
0
x + x3 x1 + ix2
~x = (6.198)
x1 − ix2 x0 − x3
is related to the point x ∈ R1,3 , where ~xc is the H-conjugation of ~x ∈ R1,3 , given by
eqn (6.198). Hence, the reference twistor reads

−i~xξ
ηx = , (6.199)
ξ
which is the index-free version of the Penrose classical twistor (Penrose, 1967). The
sign in the first component is different, since we use the Weyl representation:

0I 0 −σk
γ(e0 ) = γ0 = , γ(ek ) = γk = . (6.200)
I0 σk 0
In order to get the correct sign, Keller uses a representation which is similar to the
Weyl one, but in which the vectors in R3 are reflected (~x 7→ −~x) through the origin:

0I 0 σk
γ(e0 ) = γ0 = , γ(ek ) = γk = . (6.201)
I0 −σk 0
Thus, it is possible to get the Penrose twistor

i~xξ
ηx = . (6.202)
ξ
Therefore, twistors are completely described by the multivector structure of the

Dirac algebra C ⊗ Cℓ1,3 ≃ Cℓ4,1 ≃ M(4, C).
202 Spinors
An Alternative Approach to Twistors

We now define twistors as a special class of algebraic spinors in Cℓ4,1 . The isomor-
phism Cℓ4,1 ≃ C ⊗ Cℓ1,3 , where
E0 = iγ0 , E1 = γ10 , E2 = γ20 , E3 = γ30 , E4 = γ5 γ0 = −γ123 , (6.203)
explicitly gives rise to the relations E02 = −1, and E12 = E22 = E32 = E42 = 1. A
paravector x ∈ R ⊕ R4,1 ֒→ Cℓ4,1 is written as
x = x0 + xA EA = x0 + α0 E0 + x1 E1 + x2 E2 + x3 E3 + α4 E4 . (6.204)
V0 4,1 V1 4,1 V2 4,1
We also define an element χ := xE4 ∈ (R ) ⊕ (R ) ⊕ (R ) as
χ = xE4 = x0 E4 + α0 E0 E4 + x1 E1 E4 + x2 E2 E4 + x3 E3 E4 + α4 .
It can be seen that
1 1
χ (1 + iγ5 ) = Tx (1 + iγ5 ) = Tx PL . (6.205)
2 2
We define the twistor as the algebraic spinor
χPL U f ∈ (C ⊗ Cℓ1,3 )f ,
where f is a primitive idempotent of C ⊗ Cℓ1,3 ≃ Cℓ4,1 , and U ∈ Cℓ4,1 denotes an
arbitrary element. Hence, U f is a Dirac spinor, and PL U f = 0ξ = Π ∈ 12 (1 + iγ5 )(C ⊗
Cℓ1,3 ) is a covariant dotted Weyl spinor. The twistor reads
χΠ = xE4 Π
= (x0 E4 + α0 E0 E4 + x1 E1 E4 + x2 E2 E4 + x3 E3 E4 + α4 )Π.
From the relation E4 Π = γ5 γ0 Π = −γ0 γ5 Π = −iγ0 Π, it follows that
χΠ = (x0 E4 + α0 E0 E4 + x1 E1 E4 + x2 E2 E4 + x3 E3 E4 + α4 )Π
= x0 (E4 Π) + xk Ek (E4 Π) + α0 E0 (E4 Π) + α4 Π
= −ix0 γ0 Π − ixk γk Π + α0 Π + α4 Π
= (1 + γ5 )Π

i~xξ
= . (6.206)
ξ
Hence, our definition is shown to be equivalent to Keller’s and, therefore, to the Penrose
classical twistor, by eqn (6.202).
The incidence relation, which determines a point in spacetime from the intersection
between two twistors (Penrose, 1967; Penrose and Rindler, 1984), is given by
Jχ̄χ = xE4 U xE4 U = −Ū E4 x̄xE4 U = 0,
since the paravector x ∈ R ⊕ R4,1 ֒→ Cℓ4,1 is an element in the Klein absolute;
consequently, xx̄ = 0. Finally, the Robinson congruence is defined in our formalism
from the product
Jχ̄χ′ = xE4 U x′ E4 U = −Ū E4 x̄x′ E4 U. (6.207)
This product is null if x = x′ , and the Robinson congruence is defined when we fix x
and let x′ vary.
Spinor Classification According to Bilinear Covariants 203
6.16 Spinor Classification According to Bilinear Covariants

There is a spinor classification, due to Lounesto (2001a), which is particularly inter-
esting both for mathematicians and physicists. The essence of this classification relies
upon the so-called bilinear covariants, which describe physical observables in field
theory. Moreover, the so-called Fierz aggregate can bring a robust geometrical inter-
pretation of these quantities (Lounesto, 2001a). Within the Lounesto classification, a
specific bilinear covariant plays a prominent role. In fact, the 1-covector current density
has such interpretation, at least for the case of a regular spinor describing the electron
in Dirac theory (Dirac, 1928). The current density reads J = Jµ eµ = ψ † γ0 γµ ψeµ ,
where ψ denotes a classical spinor, {γµ } stands for the gamma matrices, and {eµ } is a
dual basis in C ⊗ Cℓ1,3 . Regarding the electron theory, J is a conserved current. Con-
sequently, the time component J0 = ψ † ψ provides the probability density associated
with the electron and which should be non-null.8 Hence, we must have J 6= 0.
The Lounesto spinor classification is derived when a classical spinor ψ is taken
into account. Indeed, the well-known Lounesto spinor classification is based on bilin-
ear covariants and the underlying multivector structure (Crawford, 1985; Lounesto,
2001a). The physical nature of the classification focuses on the bilinear covariants,
which are physical observables which describe physical features of fermionic particles.
The observable quantities are given by the following multivector structure:
σ = ψ † γ0 ψ, (6.208)
J = Jµ eµ = ψ † γ0 γµ ψ eµ , (6.209)
1
S = Sµν eµ ∧ eν = ψ † γ0 iγµν ψ eµ ∧ eν , (6.210)
2
K = Kµ eµ = ψ † γ0 iγ0123 γµ ψ eµ , (6.211)
ω = −ψ † γ0 γ0123 ψ, (6.212)
where γ0123 = iγ5 . The expression for these quantities using algebraic spinors and
spinor operators can be found in the book by Lounesto (2001a).
The bilinear covariants have a physical interpretation in the Dirac theory, after a
suitable multiplication by some physical constants. Indeed, eJ0 is interpreted as charge
density, ecJk (k = 1, 2, 3) as electric current density, (e~/2mc)S ij as magnetic moment
density, (e~/2mc)S 0j as electric moment density, and (~/2)Kµ as spin density. The
interpretation of the scalar σ and pseudoscalar ω bilinear covariants is less clear than
this, but when combined in ρ2 = σ 2 + ω 2 = |J|2 (by the Fierz–Pauli–Kokink (FPK)
identities), ρ can be interpreted as probability density. A prominent requirement for
the Lounesto spinor classification is that the bilinear covariants satisfy quadratic alge-
braic relations known as the FPK identities (Holland, 1986; Crawford, 1985; Lounesto,
2001a):
8 It is worth emphasising that the reason for considering J as the current density is clear when the
spinor obeys the usual dynamics ruled by the Dirac equation (Villalobos, da Silva, and da Rocha,
2015). The mass dimension in this case is the same mass dimension 3/2 associated with the usual spin-
1/2 fermions in the standard model of elementary particles. When J = 0 is required, the underlying
dynamics are not be provided by the Dirac equation. Since the construction is relativistic, the emergent
spinors with J = 0 respect the Klein–Gordon equation.
204 Spinors
J2 = ω 2 + σ 2 , K2 = −J2 , JyK = 0, J ∧ K = −(ω + σγ0123 )S . (6.213)
By taking a classical spinor ξ which satisfies ξ † γ0 ψ 6= 0, the original spinor ψ can

be recovered from its aggregate Z, which is provided by
Z = σ + J + iS + iKγ0123 + ωγ0123 , (6.214)
using the Takahashi algorithm (Takahashi, 1982; Vaz, 1998). In fact, the spinor ψ can
be written as the multivector Zξ by
1
ψ= p e−iθ Zξ, (6.215)
2 ξ † γ0 Zξ
where e−iθ = 2(ξ † γ0 Zξ)−1/2 ξ † γ0 ψ ∈ U(1). For more details see, for example, the article
by Crawford (1985). Moreover, if the bilinear covariants satisfy the Fierz identities,
then the complex multivector Z is called a Fierz aggregate. If γ0 Z† γ0 = Z, then Z is
said to be a boomerang (Lounesto, 2001a).
Within the Lounesto classification scheme, the condition J 6= 0 is fundamental for
defining the so-called boomerang (Lounesto, 2001a). With this condition, there exist
just six types of spinors, according to the bilinear covariants.9 Indeed, in the Lounesto
scheme, spinors are classified as regular or singular spinors. Regular spinors present
either σ 6= 0, or ω 6= 0 (or even both non-null quantities). On the other hand, singular
spinors present σ = 0 = ω, in which case the Fierz identities are in general replaced
by the following conditions, which are more general than those for regular spinors
(Crawford, 1985):
Z2 = 4σZ, Zγµ Z = 4Jµ Z, Zγ0123 Z = −4ωZ,

iZγµν Z = 4Sµν Z, iZγ0123 γµ Z = 4Kµ Z . (6.216)
The aggregate plays an essential role within the Lounesto classification, since Z has
to be promoted to a boomerang, satisfying
Z2 = 4σZ. (6.217)
For regular spinors, eqn (6.217) holds and Z is a boomerang. However, for singular
spinors, we must ensure that the aggregate is a boomerang. In fact, J must be parallel
to K, and both must lie in the plane defined by S. Hence, by using eqn (6.214) and
taking into account singular spinors, we can clearly see that the aggregate reads
Z = J(1 + is + ihγ0123 ), (6.218)
where s is a space-like vector orthogonal to J, and h is a real number. The multivector

expressed in eqn (6.218) is a boomerang (da Rocha and da Silva, 2010). Equation
(6.217) yields, for singular spinors, Z2 = 0. However, for the FPK identities to hold,
9 More three classes can be obtained when the condition J 6= 0 is not demanded (da Rocha, Fabbri,
da Silva, Cavalcanti, and Silva-Neto, 2013).
both conditions J2 = 0, and (s + hγ0123 )2 = −1, must hold in order to constrain the
possible spinor classes.10
Equation (6.215) implies that different bilinear covariants combinations may lead
to different spinors, by the constraints imposed by the FPK identities. Hence, the
algebraic constraints reduce the possibilities to six different spinor classes, namely:
(1) σ 6= 0, ω 6= 0
(2) σ 6= 0, ω = 0
(3) σ = 0, ω 6= 0
(4) σ = 0 = ω, K 6= 0, S 6= 0
(5) σ = 0 = ω, K = 0, S 6= 0
(6) σ = 0 = ω, K 6= 0, S=0
Classes (1), (2), and (3) contain regular spinors. Class (4) spinors are called flag-
dipole spinors (da Rocha, Fabbri, da Silva, Cavalcanti, and Silva-Neto, 2013), while
class (5) spinors are called flagpole spinors (Benn and Tucker, 1987). Majorana spinor
(Majorana, 1932) and Elko dark spinors are elements of the fifth class (da Rocha and
Rodrigues, 2006; Ahluwalia and Grumiller, 2005; Ahluwalia, Lee, and Schritt, 2010;
da Rocha and da Silva, 2009; da Rocha and da Silva, 2010; Cavalcanti, 2014; da Rocha
and da Silva, 2014). Finally, class (6) dipole spinors are exemplified by Weyl spinors.
New physical particles have been proposed via the use of this classification in, for
example, the work by da Rocha and Rodrigues Jr (2006) and by da Rocha, Fabbri,
da Silva, Cavalcanti, and Silva-Neto (2013), and also in exotic structure frameworks
(da Rocha, Bernardini, and da Silva, 2011; Bernardini and da Rocha, 2012; Cavalcanti,
da Silva, and da Rocha, 2014).
Note that there are only six different spinor fields. In fact, for regular spinors, since
J 6= 0, it follows that S 6= 0, and K 6= 0, from the identities in (6.213). On the other
hand, for the singular case, the geometry asserts that J(s + hγ0123 ) = S + Kγ0213 .
Hence, J 6= 0 provides all the possibilities. The most general form of the respective
spinors in each class have been introduced in the article by Cavalcanti (2014). High-
dimensional spaces have a similar spinor classification (Bonora, de Brito, and da Rocha,
2015; Bonora and da Rocha, 2016); however, the so-called geometric Fierz identities
(Lazaroiu, Babalic, and Coman, 2013) obstruct the proliferation of new spinor classes
in high-dimensions (Bonora, de Brito, and da Rocha, 2015).

The standard references for spinor theory, from the classical point of view, are (Brauer
and Weyl, 1935; Cartan, 1937), and those from the algebraic point of view are the work
by Chevalley (1954) and that by Riesz (1993). One of the most important applications
of spinor theory is the Dirac theory of the electron, and Lounesto (2001a) gives a
detailed comparison of the concepts of classical spinors, algebraic spinors, and spinor
operators in the realm of the Dirac theory. The concept of spinor operators is further
approached in the book by Dorac and Lasenby (2003), who compared that concept
10 Note that J must be non-zero in the Lounesto classification.

206 Spinors
with the classical and algebraic definitions. Pauli spinors and Dirac spinors are dis-
cussed in detail in the book by Hladik (1999). A classical reference for Weyl spinors
and field theory is the book by Penrose and Rindler (1984). For more information
about spinor theory we also suggest the book by Harvey (1990), mainly for a different
approach to triality, as well as the book by Knus (1998). Moreover, additional details
on generalisations of the octonionic algebra can be seen in the articles by da Rocha
and Vaz Jr (2006a), da Rocha and Traesel (2012), and da Rocha, Traesel, and Vaz Jr
(2012). Particular applications of triality to physics are further presented by Hasiewicz
and Kwasniewski (1985), de Andrade, Rojas, and Toppan (2001), and Baez (2002).
Pure spinors have been also used in superstring theory (Berkovits, 2004) and parti-
cle physics (Hasiewicz and Kwasniewski, 1985; Benn and Tucker, 1987; Budinich and
Trautman, 1989; de Andrade, Rojas, and Toppan, 2001; Budinich, 2002; Ahluwalia,
Lee, and Schritt, 2010; da Rocha, Bernardini, and Vaz, 2010). For additional informa-
tion about the Lounesto spinor classification and its applications in field theory and
theories of gravity, see, for example, the work by Rocha and Rodrigues Jr (2006), da
Rocha and da Silva (2009), da Rocha and da Silva (2010), da Rocha, Bernardini, and
da Silva (2011), Bernardini and da Rocha (2012), da Rocha, Fabbri, da Silva, Caval-
canti, and Silva-Neto (2013), da Rocha and da Silva (2014), Cavalcanti, da Silva, and
da Rocha (2014); for their high-dimensional extensions, see the article by Bonora, de
Brito, and da Rocha, (2015). In addition, spinor classification in the context of quan-
tum Clifford algebras is presented by Ablamowicz, Gonçalves, and da Rocha (2014).
(Ablamowicz, Gonçalves, and da Rocha, 2014)
6.18 Exercises
(1) The Clifford algebra Cℓ2,1 is isomorphic to M(2, R ⊕ R). (a) Show that f =
1 1
2 (1+e1 ) 2 (1+e2 e3 ) is a primitive idempotent in Cℓ2,1 and that the subalgebra f Cℓ2,1 f
is isomorphic to the scalars. (b) Show that the elements
1 1
f1 = (1 + e1 + e2 e3 + e1 e2 e3 ), f2 = (e2 − e1 e2 + e3 − e1 e3 )
4 4
form a basis for the space of algebraic spinors in Cℓ2,1 and that (i) f˜1 f1 = 0 = f˜1 f2 =
f˜2 f1 = f˜2 f2 ; (ii) f¯1 f1 = 0 = f¯2 f2 ; (iii) f¯2 f1 = −f2 ; and (iv) f¯1 f2 = f2 . (c) Given
arbitrary algebraic spinors in Cℓ2,1 , show that the spinor scalar products ψ̃φ and e2 ψ̄φ
take values in f Cℓ2,1 f and that ψ̃φ is identically null. Show that e2 ψ̄φ is antisymmetric.
(2) Given an orthonormal basis {γ0 , γ1 , γ2 , γ3 } of Cℓ1,3 ≃ M(2, H), show that
1 1
h1 = (1 + γ0 ), h2 = (−γ1 γ2 γ3 + γ0 γ1 γ2 γ3 ),
2 2
1 1
i1 = (γ2 γ3 + γ0 γ2 γ3 ), i2 = (γ1 − γ0 γ1 ),
2 2
1 1
j1 = (γ3 γ1 + γ0 γ3 γ1 ), j2 = (γ2 − γ0 γ2 ),
2 2
1 1
k1 = (γ1 γ2 + γ0 γ1 γ2 ), k2 = (γ3 − γ0 γ3 )
2 2
Exercises 207
form a basis for the real vector space Cℓ1,3 12 (1 + γ0 ). Show that the set {h1 , i1 , j1 , k1 }
is a basis for the real vector space f Cℓ1,3 f and is, in addition, a ring isomorphic to H.
Moreover, show that the right module f Cℓ1,3 f -linear Cℓ1,3 f is two-dimensional, with
the basis {h1 , h2 }. In this basis, show that the left multiplication by γµ is represented
by the matrices with quaternionic entries

1 0 0i 0j 0k
γ0 = , γ1 = , γ2 = , γ3 = .
0 −1 i0 j0 k0
Show also that
h̃1 h1 = h1 , h̃1 h2 = 0 = h̃2 h1 , h̃2 h2 = −h1 ,

h̄1 h1 = 0 = h̄2 h2 , h̄2 h1 = h2 , h̄1 h2 = h2 .
(3) Let us introduce in C ⊗ Cℓp,q the complex conjugation such that u∗ = A − iB for
u = A + iB, where A, B ∈ Cℓp,q . Show according to the Clifford algebras in table 6.9
that f (C ⊗ Cℓp,q )f ≃ C, where the idempotents f associated with each of the Clifford
algebras are given in the second column of the table. Prove in addition that, given
ψ 6= 0 and φ arbitrary algebraic spinors in (C ⊗ Cℓp,q )f , the properties given in the
third column of Table 6.9 hold.
Table 6.9 Clifford Algebras and Associated Idempotents for u∗ = A − iB for
u = A + iB, Where A, B ∈ Cℓp,q
Clifford Algebra Idempotent (f ) Property
1
C ⊗ Cℓ0,2 2
(1 + ie1 ) ψ̄ ∗ ψ > 0
1
C ⊗ Cℓ1,1 2
(1 + e1 )
1
C ⊗ Cℓ2,0 2 (1 + e1 ) ψ̃ ∗ ψ > 0
1 1
C ⊗ Cℓ0,3 2 (1 + ie1 ) 2 (1 + ie2 e3 ) ψ̄ ∗ ψ > 0, ψ̃ ∗ φ = 0
1 1
C ⊗ Cℓ1,2 2 (1 + e1 ) 2 (1 + ie2 e3 ) ψ̄ ∗ φ = 0
1
C ⊗ Cℓ2,1 2
(1 + e1 ) 12 (1 + ie2 e3 ) ψ̄ ∗ φ > 0, ψ̃ ∗ ψ > 0
1
C ⊗ Cℓ3,0 2
(1 + e1 ) 12 (1 + ie2 e3 ) ψ̄ ∗ φ = 0, ψ̃ ∗ ψ > 0
1
C ⊗ Cℓ0,4 2
(1 + ie1 ) 12 (1 + ie2 e3 ) ψ̄ ∗ ψ > 0
1
C ⊗ Cℓ1,3 2
(1 + e1 ) 12 (1 + ie2 e3 )
Define now (
e1 e2 . . . ep if p = 0, 1 mod 4
A=
ie1 e2 . . . ep if p = 2, 3 mod 4
and show the properties in the second and third columns of table 6.10, for all v ∈ Rp,q ,
for ψ 6= 0, and for φ arbitrary algebraic spinors in (C ⊗ Cℓp,q )f .
208 Spinors
Table 6.10 Clifford Algebras and Associated Idempotents for A =

e1 e2 . . . ep if p = 0, 1 mod 4 and for A = ie1 e2 . . . ep if p = 2, 3 mod 4
Clifford Algebra Idempotent (f ) Property
∗
C ⊗ Cℓ0,2 ∗
ψ̄ Aψ > 0 vψ φ + ψ̄ ∗ vφ = 0
∗
C ⊗ Cℓ1,1 f φ − ψ̃ ∗ vφ = 0
vψ
∗
C ⊗ Cℓ2,0 ψ̃ ∗ Aψ > 0 vψ φ + ψ̄ ∗ vφ = 0
∗
C ⊗ Cℓ0,3 ψ̄ ∗ Aψ > 0 vψ φ + ψ̄ ∗ vφ = 0
∗
C ⊗ Cℓ1,2 ψ̄ ∗ Aφ = 0 f φ − ψ̄ ∗ vφ = 0
vψ
∗
C ⊗ Cℓ2,1 ψ̃ ∗ Aφ = 0 vψ φ + ψ̄ ∗ vφ = 0
∗
C ⊗ Cℓ3,0 ψ̄ ∗ Aφ = 0, ψ̃ ∗ Aψ > 0 f φ − ψ̃ ∗ vφ = 0
vψ
∗
C ⊗ Cℓ0,4 ψ̄ ∗ Aψ > 0 vψ φ + ψ̄ ∗ vφ = 0
∗
C ⊗ Cℓ1,3 f φ − ψ̃ ∗ vφ = 0
vψ
(4) Given the Clifford algebra Cℓ1,7 ≃ M(16, R) with generators {γ0 , γa } such that
γ02 = 1, γa2 = −1, a = 1, . . . , 7 and the idempotent
1 1 1 1
f= (1 + γ0 ) (1 + iγ1 γ2 ) (1 + iγ3 γ4 ) (1 + iγ5 γ6 ),
2 2 2 2
an open question is whether there exists a spinor ψ ∈ (C⊗Cℓ1,7 )f such that (hψ ψ̃ ∗ i0 )2 <
0. Can the reader answer that? Consider an element
ψ = (1 − e1 e2 e3 e4 e5 e6 e7 e8 )(1 + w) ∈ Cℓ8 ,
where
w = e1 e2 e3 e6 − e1 e2 e5 e7 − e1 e3 e4 e5 + e1 e4 e6 e7 + e2 e3 e4 e7 − e2 e4 e5 e6 − e3 e5 e6 e7 .
Show that ψ satisfies ψvψ̃ = 0, for all v ∈ R8 , and that ψ 2 = 16ψ. Show that the
factor (1 + w) is not invertible (hint: show that (1 + w)2 = 8(1 + w)).
(5) Compute the matrix associated with the multivector 14 (w − 3), where w is given
as in exercise 4, in the basis (g1 , g2 , . . . , g8 ), where ga = ea e8 f , a = 1, 2, . . . , 8, and
f = 18 (1 + w)(1 + e12...8 ). In other words, prove that
4hg̃a (w − 3)gb i0 = diag(−1, −1, −1, −1, −1, −1, −1, 1) .
(6) Show that the quantities given by eqn (6.208) in terms of a classical spinor ψ can
be written in terms of the spinor operator Ψ = Ψ+ (see example 6.11) as
1
σ + e5 ω = ΨΨ̃, J = Ψe0 Ψ̃, S= Ψe21 Ψ̃, e5 K = Ψe3 Ψ̃,
2
and, supposing that ΨΨ̃ 6= 0, prove the FPK identities in (6.213).
Appendix A
The Standard Two-Component
Spinor Formalism
In chapter 6, we presented Weyl spinors in Cℓ3,0 ; now, the connection with the ordinary
notation in either field theory or supersymmetry books is briefly presented.
A spacetime vector v ∈ R1,3 can be expressed as v = x0 e0 + x1 e1 + x2 e2 + x3 e3 ,
where (x0 , x1 , x2 , x3 ) denotes components of v with respect to an orthonormal basis
{e0 , e1 , e2 , e3 }. Null vectors are isotropic vectors and satisfy (x0 )2 − (x1 )2 − (x2 )2 −
(x3 )2 = 0. They present null directions in R1,3 with respect to the origin O of an
arbitrary frame in R1,3 .
The space of null directions that are future (past) pointed are denoted by S + [S − ],
and represented by the intersections E+ [E− ] of the future (past) light cones with
the hyperplanes x0 = 1 (x0 = −1).1 The space S± is a sphere with the equation
x2 + y 2 + z 2 = 1, where (x, y, z) are coordinates in E± (Penrose and Rindler, 1984).
Generally, the direction of any null vector v ∈ R1,3 , unless the vector is an element
of the plane defined by the equation x0 = 0, can be represented by two points. This
description results from the intersection of v and the hyperplanes x0 = ±1. The future-
pointed v is thus represented by (x1 /kx0 k, x2 /kx0 k, x3 /kx0 k). The inner points of E+
(E− ) represent the set of future-pointed (past-pointed) light-like directions.
By considering E+ and by performing a stereographic projection on the Argand–
Gauss plane, we obtain a representation of the union between the set of complex
numbers and the point at infinity, the latter corresponding to the north pole of E+ .
By defining the complex number
x + iy
β= , (A.1)
1−z
x2 +y 2
we obtain ββ = (1−z)2
and, consequently,
β+β β−β ββ − 1
x= , y= , z= . (A.2)
ββ + 1 i(ββ + 1) ββ + 1
The correspondence between points of E+ and the Argand–Gauss plane is injective if
the point β ∼ ∞ is added to the complex plane, making it correspond to the north pole
with components (1, 0, 0, 1). However, to avoid this point, it is convenient to associate
a point of E+ not to a complex number β but to a pair of complex numbers 2 (ξ, η),
1 This space is a Riemann sphere.
2 With the condition that both numbers are not simultaneously equal to 0.
210 The Standard Two-Component Spinor Formalism
where
β = ξ/η. (A.3)
The pairs (ξ, η) and (λξ, λη), where λ ∈ C, represent the same point in E+ . Such
components are called projective coordinates.
The point β = ξ/η ∼ ∞ corresponds to the point of coordinates ηξ = 10 . The
equations in (A.2) can then be expressed as
ξη + ηξ ξη − ηξ ξξ − ηη
x= , y= , z= . (A.4)
ξξ + ηη i(ξξ + ηη) ξξ + ηη
The point P = (1, x, y, z) is an arbitrary point of the light-cone transversal section with
constant time and represents a null future-pointed direction, which can be represented
by any point of the line OP . In particular,
√ if a point R is taken in the line OP by
multiplying P by the factor (ξξ + ηη)/ 2, then R has coordinates
1 1
x1 = √ (ξη + ηξ), x2 = √ (ξη − ηξ),
2 i 2
1 1
x3 = √ (ξξ − ηη), x0 = √ (ξξ + ηη). (A.5)
2 2
Unlike the point P , the point R is not invariant under (ξ, η) 7→ (rξ, rη), r ∈ R, although
it is independent of phases (ξ, η) 7→ (eiθ ξ, eiθ η), θ ∈ R.
Consider now the following complex linear transformation
ξ 7→ ξ˜ = αξ + µη,
(A.6)
η 7→ η̃ = γξ + δη,
where α, µ, γ, δ ∈ C satisfy αδ − µγ 6= 0, so that the transformation is invertible. It

can be rewritten as
αβ + µ
β 7→ f (β) = , (A.7)
γβ + δ
and called a Möbius transformation, from the set C\{−δ/γ} to C\{α/γ}. Moreover,
if f (−δ/γ) ∼ ∞, and f (∞) ∼ α/γ, then f is an injective function from the complex
plane, compactified by the point at the infinity; this point is denoted by (C ∪ {∞}).
Hence, the space of light-like vectors on Minkowski spacetime is naturally a Rie-
mann sphere. The restricted Lorentz group L+ is, on the other hand, the automorphism
group of the Riemann sphere. The equations in (A.6), when taken with the condition
αδ − µγ = 1
are called spinor transformations, where β = ξ/η is related to the null vectors by the
equations in (A.5), implying that
x1 + ix2 x0 − x3
β= 0 3
= 1 . (A.8)
x −x x − ix2
The Standard Two-Component Spinor Formalism 211
The spinor matrix A ∈ SL(2, C) is defined as

αµ
A= , det A = 1. (A.9)
γ δ
The equations in (A.6), with respect to A, read

ξ˜ ξ
=A .
η̃ η
The spinor matrices {±A} induce the same transformation of β = ξ/η. The equations
in (A.5) yield
0
1 x + x3 x1 + ix2 ξξ ξη ξ
√ 1 2 0 3 = = (ξ η). (A.10)
2 x − ix x − x ηξ ηη η
√
Hence, up to a factor 1/ 2, it follows that
0 !
x +x3 x1 +ix2 x˜0 + x˜3 x˜1 +ix˜2
7→
x1 −ix2 x0 −x3 x˜1 −ix˜2 x˜0 − x˜3
0
x +x3 x1 +ix2
= A A† . (A.11)
x1 −ix2 x0 −x3
The transformation acting on the point v = (x0 , x1 , x2 , x3 ) is real and preserves the
light-cone structure (x0 )2 − (x1 )2 − (x2 )2 − (x3 )2 = 0. Thus, this relation defines a
restricted Lorentz transformation.
Hence, the group SL(2,C) is the twofold covering of the restricted Lorentz group
SO+ (1, 3) ≃ L+ .
A more general case than this is the spin space Gα , which has three basic operations
(Penrose and Rindler, 1984):
• multiplication by scalars: C × Gα → Gα (λ ∈ C, kα ∈ Gα 7→ λk α ∈ Gα )
• sum: Gα × Gα → Gα (k α , ω α ∈ Gα 7→ k α + ω α ∈ Gα )
• scalar product: Gα × Gα → C (k α , ωα ∈ Gα 7→ {k α , ω α } ∈ C)
The dual spin space Gα is similarly defined: Gα ∋ πα : Gα → C. Thus, kα ω α ≡
{k α , ω α } ∈ C. With respect to the null vectors of Minkowski spacetime, Penrose
proposed that the algebra of null vectors must be contained in the algebra of spinors.
The spin space Gα̇ is defined by the application ̺ : Gα → Gα̇ such that
̺(k α + ω α ) = ̺(k α ) + ̺(ω α ),

̺(λk α ) = λ̺̄(k α ), ∀k α , ω α ∈ Gα , λ ∈ C (λ̄ is the C-conjugate of λ).
Instead of by the notation ̺(k α ), this transformation is denoted by k α , according to

the Penrose notation, characterised by the composition between the C-conjugation
and the transposition. From now on, the notation
k α ≡ k̄ α̇ ∈ Gα̇ (A.12)
shall be used. From (A.10), the space R1,3 endowed with the coordinates (x0 , x1 , x2 , x3 )
can be expressed from the components of the spin vector k (ξ = k 0 , η = k 1 ). The basic
operations in this space are defined by
λ(k 0 , k1 ) = (λk 0 , λk1 ), (A.13)
(k , k ) + (ω 0 , ω 1 ) = (k 0 + ω 0 , k 1 + ω 1 ),
0 1
(A.14)
{(k 0 , k1 ), (ω 0 , ω1 )} = k 0 ω 1 − k 1 ω 0 . (A.15)
With the antisymmetric bilinear form in eqn (A.15), the representation of a spin
vector is given by the choice of a pair of normalised spin vectors oα and ια :
{oα , ια } = oα ια = 1 = −ια oα = −{ια , oα }, (A.16)
where (oα , ια ) stands for the dual pair of (oα , ια ). Moreover, the antisymmetry of
(A.15) implies that oα oα = ια ια = 0. The pair (oα , ια ), with the condition in eqn
(A.16), is called the spin basis, and the components of k with respect to the spin basis
are provided by (Figueiredo, de Oliveira, and Rodrigues, 1990)
k 0 = {k α , ια }, k 1 = −{k α , oα } . (A.17)
Hence, k α = k 0 oα + k 1 ια .
The antisymmetric element
Gαβ ∋ ǫαβ : Gα → Gβ
k α 7→ kβ = k α ǫαβ , (A.18)
is responsible for lowering and raising indices, such that
{k α , ω α } = ǫαβ k α ω β = k 0 ω 1 − k 1 ω 0 = −{ω α , kα }. (A.19)
We can write
ǫαβ = oα ιβ − ια oβ (A.20)
since, for any spin basis, it follows that
k α = k 0 oα + k 1 ια = (oα ιβ − ια oβ )(k 0 oβ + k 1 ιβ )
= ǫαβ kβ . (A.21)
Here, Gαβ denotes the tensor product Gα ⊗Gβ . The dual tensor Gαβ ∋ ǫαβ : Gβ → Gα
satisfies
ǫαβ ǫγβ = δαγ , (A.22)
where Gα → Gβ → Gγ is defined. Similarly, we can express
ǫαβ = oα ιβ − ια oβ , (A.23)
and

β 0 1 0 1 0 1 k0 = −k 1
{k, ω} = kβ ω = k0 ω + k1 ω = k ω − ω k ⇒ . (A.24)
k1 = k 0
Now, given arbitrary spinors kα , ωα ∈ Gα such that kα ω α = 1, we can write
The Standard Two-Component Spinor Formalism 213
ǫαβ = kα ωβ − ωα kβ = (k0 ια + k1 oα )(ω0 ιβ + ω1 oβ )

= (k1 ω0 − k0 ω1 )oα ιβ − (k1 ω0 − k0 ω1 )ια oβ
= oα ιβ − ια oβ , (A.25)
where the equivalence to eqn (A.20) is then accomplished.

Now the spacetime metric and the spacetime vectors as well can be constructed
from spin vectors. Spacetime vectors of R1,3 present a spinor description via spin
vectors. Latin indices are employed here to label the elements xµ of a real vector space,
when the indices α and α̇ are grouped together. Moreover, the notation {a, b, . . .} =
{αα̇, β β̇, . . .} will be used. The null tetrad (lµ , nµ , mµ , mµ )
lµ = oα oα̇ , nµ = ια ια̇ , mµ = oα ια̇ , mµ = ια oα̇ (A.26)
and the metric ηµν = ǫαβ ǫα̇β̇ can be defined by verifying that the vectors of the null
tetrad are null vectors with respect to ηab :
ηµν lµ lν = lµ lµ = 0 . (A.27)
Similarly, nµ nµ = mµ mµ = mµ mµ = 0. In addition, the following expressions hold:
lµ nµ = 1, mµ mµ = −1, lµ mµ = lµ mµ = nµ mµ = nµ mµ = 0 ,
ηµν ≡ ηµρ g ρν
= nµ lν + lµ nν − mµ mν − mµ mν . (A.28)
It is sometimes convenient to define another tetrad (tµ , xµ , yµ , z µ ) as
tµ = √12 (lµ + nµ ) = √12 (oα oα̇ + ια ια̇ ),

xµ = √12 (mµ + mµ ) = √12 (oα ια̇ + ια oα̇ ),
(A.29)
y µ = √i2 (mµ + mµ ) = √i2 (oα ια̇ − ια oα̇ ),
z µ = √12 (lµ − nµ ) = √12 (oα oα̇ − ια ια̇ ).
We obtain from (A.28) that
tµ xµ = tµ yµ = tµ zµ = xµ yµ = y µ zµ = z µ xµ = 0,
tµ tµ = 1 = −xµ xµ = −y µ yµ = −z µ zµ . (A.30)
Consequently, the metric components ηµν (A.28) have the form ηµν = tµ tν − xµ xν −
yµ y ν −zµ z ν , thus identifying the tetrad (tµ , xµ , y µ , z µ ) as the Minkowski tetrad. Hence,
we can write
K µ = K 0 tµ + K 1 x µ + K 2 y µ + K 3 z µ . (A.31)
By considering the spin basis {oα , ια }, we can express the vector K µ with respect to
the basis as
K µ = K 00̇ lµ + K 01̇ mµ + K 10̇ mµ + K 11̇ nµ . (A.32)
When the two last equations are compared, it follows that

!
µ 1 K 0 + K 3 K 1 + iK 2 K 00̇ K 01̇
K =√ = . (A.33)
2 K 1 − iK 2 K 0 − K 3 K 10̇ K 11̇
Thus,
K µ = ±k α k̄ α̇ , (A.34)
µ
where the sign defines a future (+) (past (−)). The vector K is real and null, since
K µ Kµ = |kα k α |2 = 0. If ξ = k 0 , and η = k 1 , it follows that
K 00̇ = ξξ, K 01̇ = ξη, K 10̇ = ηξ, K 11̇ = ηη , (A.35)
which causes eqn (A.33) to be led to
0
1 x + x3 x1 + ix2 ξξ ξη ξ
√ 1 2 0 3 = = (ξ η), (A.36)
2 x − ix x − x ηξ ηη η
where x0 = K 0 ; x1 = K 1 ; x2 = K 2 ; and x3 = K 3 .
A.1 Weyl Spinors

Given the formalism presented for spin vectors, which are known as 2-spinors, it is
also necessary to study then from the point of view of representations of the Lorentz
group SL(2, C), the 2-fold covering of the restricted Lorentz group L+ ≃ SO+ (1, 3).
Linear transformations with a unit determinant, with respect to the spin space, de-
termine the group SL(2, C). We have already shown that there are two non-equivalent
representations of SL(2, C); these are denoted by D(1/2,0) and D(0,1/2) , respectively,
and the elements of the carrier space associated with them are called Weyl spinors.
Both the left-handed D(1/2,0) and the right-handed D(0,1/2) representations of the
Lorentz group determine the rules of transformation obeyed by fermions of spin-1/2.
It is well known that the Hermitian conjugation can be used to interchange these
two representations. Dirac spinors take into account reducible representations of the
form D (1/2,0) ⊕ D(0,1/2) . From here on, D (1/2,0) spinors will carry undotted indices
α, β, . . . = 1, 2, and D(0,1/2) spinors will carry dotted indices α̇, β̇, . . . = 1, 2.
A.2 Contravariant Undotted Spinors

Contravariant undotted spinors are elements of a complex two-dimensional space en-
dowed with the spinor metric
G : C2 × C2 → C,
(ζ, χ) 7→ G(ζ, χ) = ζ † Jχ, (A.37)
where
0 1
J= . (A.38)
−1 0
The spinor ζ is represented by the column vector
1
ζ
ζ= . (A.39)
ζ2
This spinor can be identified with its algebraic counterpart in eqn (6.154).
Covariant Undotted Spinors 215
It also carries the D(1/2,0) representation of SL(2, C) and is transformed under

R ∈ SL(2, C) as
ζ 7→ Rζ.
Moreover, this transformation corresponds to the rule established in chapter 6, eqn

(6.162).
A.3 Covariant Undotted Spinors
Covariant undotted spinors are elements of a complex two-dimensional dual space C2♭ ,
and are defined by
C2♭ ∋ ζ♭ : C2 → C ,
χ 7→ ζ♭ (χ) = ζ♭ χ = G(ζ, χ) = ζ † Jχ , (A.40)
which implies that ζ♭ = ζ † J. Hence, the spinor ζ♭ is represented by
ζ♭ = (ζ1 , ζ2 ) = (ζ 2 , −ζ 1 ) . (A.41)
This spinor can be identified with its algebraic counterpart in eqn (6.155). For a spinor
metric to be invariant under R ∈ SL(2, C), it is necessary that
ζ♭ 7→ ζ♭ R−1 , R ∈ SL(2, C) , (A.42)
which corresponds to the transformation in eqn (6.163).

Contravariant undotted spinors and covariant undotted spinors represent respec-
tively elements of Gα and Gα .
A.4 Contravariant Dotted Spinors
Covariant dotted spinors are elements of a complex two-dimensional space (Ċ2 ) ∋ ζ̇ ,

ζ ∈ C2 , endowed with the spinor metric
Ġ : Ċ2 × Ċ2 → C ,
(ζ̇, χ̇) 7→ Ġ(ζ̇, χ̇) = ζ̇J χ̇† . (A.43)
1̇ 2̇
A covariant dotted spinor is represented by ζ̇ = (ζ , ζ ) and can be identified with its
algebraic counterpart in eqn (6.158).
A.5 Covariant Dotted Spinors

Covariant dotted spinors are elements of a complex two-dimensional dual space Ċ2♭ ,
which is defined by
Ċ2♭ ∋ χ̇♭ : Ċ2 → C ,

ζ̇ 7→ ζ̇(χ̇♭ ) = ζ̇ χ̇♭ = Ġ(ζ̇, χ̇) = ζ̇J χ̇♭ , (A.44)
which implies that χ̇♭ = J(χ̇)† ; a covariant dotted spinor is represented by

!
χ1̇ χ2̇
χ̇♭ = = . (A.45)
χ2̇ −χ1̇
Hence, we can identify it with its algebraic equivalent counterpart in eqn (6.160).
Clearly, the transformation rule for dotted spinors under the transformation R ∈
SL(2, C) is provided by
ζ̇ 7→ ζ̇R† , ζ̇♭ 7→ (R† )−1 ζ̇♭ , (A.46)
carrying the D(0,1/2) representation of SL(2, C). In fact, it corresponds to the trans-
formation in eqn (6.164).
Dotted spinors are elements of Gα̇ and Gα̇ . The action of the Lorentz group on
Weyl spinors can be depicted as follows:
ζ 7→ Rζ , (A.47)
7 ζ♭ R−1 ,
ζ♭ → (A.48)
ζ̇ 7→ ζ̇R† , (A.49)
ζ̇♭ 7→ (R† )−1 ζ̇♭ . (A.50)
These transformations are emulated in eqns (6.162–6.165), in the algebraic spinor

framework.
From 2-spinors, Dirac spinors can defined as elements of Gα ⊕ Gα̇ . They are clas-
sically realised as elements of C4 , equipped with the spinor metric
G : C4 × C4 → C
(ψ1 , ψ2 ) 7→ G(ψ1 , ψ2 ) = ψ1† J d ψ2 , (A.51)
where ψ is defined by


ζ1
 ζ2 
C2 ⊕ (Ċ2 )♭ ∋ ψ = ζ + χ̇♭ =  
 χ̄1̇  . (A.52)
χ̄2̇
With respect to the standard basis of C4 , the matrix J d is the representation J d =

diag(J, J), where J denotes the symplectic matrix defined by eqn (A.38).
Null Flags and Flagpoles 217
Dirac spinors carry the D(1/2,0) ⊕ D(0,1/2) representation of SL(2, C). Under the
condition G(ψ1 , ψ2 ) = G(ρ(R)ψ1 , ρ(R)ψ2 ), requiring that the spinor metric G be in-
variant under R, the following important representation is obtained:

R 0
ρ(R) = , R ∈ SL(2, C). (A.53)
0 (R† )−1
A.6 Null Flags and Flagpoles

This section describes the classical framework corresponding to the algebraic formu-
lation provided by section 6.9.
We have already associated, via eqn (A.34), a future-pointed null vector, which
contains components K µ , with the spin-vector k α , which has coordinates (ξ, η). From
eqn (A.35), the 2-uple (ξ, η) can be further identified as the coordinates associated with
the components K µ , which are invariant under transformations (ξ, η) 7→ (eiθ ξ, eiθ η).
Moreover, this ambiguity can be reduced up to a sign by introducing a structure that
is composed of a null vector K = K µ eµ , called a pole, and a null half-plane – tangent
to the light cone and having K as the intersection – called a flagpole.
Given a contravariant undotted spinor k α , a geometric object can be constructed,
namely, the flagpole. As in eqn (A.36), we shall change the notation by establishing
K µ = xµ . The pole is defined by eqn (A.34), namely,
0 !
µ 1 αα̇ α α̇ 1 x + x3 x1 + ix2 k 1 k̄ 1̇ k 1 k̄ 2̇
x = √ x = k k̄ = √ 1 2 0 3 = .
2 2 x − ix x − x k 2 k̄ 1̇ k 2 k̄ 2̇
The vector xµ is dilated by ρ2 , when k α is multiplied by λ = ρeiθ . Notwithstanding,

the vector xµ does not change its direction and is independent of the choice of θ. Hence,
the null vector xµ is uniquely determined by the spinor k α . However, the spinor k α is
not uniquely determined by xµ , which corresponds to a family of spinors. They form
a projective space and differ from each other by a phase eiθ .
The momentum is defined as
F µν = F αβ α̇β̇ = k α k β ǫα̇β̇ + ǫαβ k̄ α̇ k̄ β̇ . (A.54)
The antisymmetric tensor F ab is real and determines a half-plane that is tangent to

the light cone along the vector xµ = k α k̄ α̇ .
By taking a spin basis {k α , ω α }, where kα ω α = 1, we find that ǫαβ = k α ω β −ω β k α .
Thus, the quantity F µν can be characterised as the angular momentum, since
F µν = F αβ α̇β̇ = k α k β (k̄ α̇ ω̄ β̇ − ω̄ α̇ k̄ β̇ ) + (k α ω β − ω α k β )k̄ α̇ k̄ β̇

= k α k̄ α̇ (k β ω̄ β̇ − ω β k̄ β̇ ) + (k α ω̄ α̇ + ω α k̄ α̇ )k β k̄ β̇ = X αα̇ Y β β̇ − Y αα̇ X β β̇
= xµ y ν − y µ xν . (A.55)
The tensor F ab hence represents a bivector constituted of two vectors with compo-
nents xµ and y µ in R1,3 . The pole xµ is the null flagpole vector, uniquely determined
by the spinor oα . The second vector, given by
y µ = Y αα̇ = (k α ω̄ α̇ + ω α k̄ α̇ ), (A.56)
is also determined by k α , although not uniquely, since the pair (k α , ω α ) is not the only
way to construct ǫαβ . Indeed, any spinor of type
ω0α = ω α + λk α , λ∈C (A.57)
satisfies kα ω0α = 1. With this freedom, the vector y µ transforms as y0µ = y µ +(λ+ λ̄)xµ .
Each scalar (λ + λ̄) can be thus associated to a family of coplanar vectors y0µ . This is
the flagpole, as proposed by Penrose. Some prominent properties can be now derived.
The vector y µ is orthogonal to the null vector xµ . Indeed,
1 1
xµ · y µ = xµ y µ = − Xαα̇ Y αα̇ = − kα k̄α̇ (k α ω̄ α̇ + ω α k̄ α̇ ) = 0 . (A.58)
2 2
Moreover, y µ is a space-like unit vector. In fact,
1
y µ · y µ = yµ y µ = (kα ω̄α̇ + ωα k̄α̇ )(k α ω̄ α̇ + ω α k̄ α̇ )
2
1 1
= (kα ω α )(ω̄α̇ k̄ α̇ ) + (ωα k α )(k̄α̇ ω̄ α̇ ) = −1. (A.59)
2 2
By multiplying the spinor k α by eiθ , the vector y µ spins around the pole by the angle
2θ. Actually, we have
µ Aα̇
yrot = Yrot = e2iθ k α ω̄ α̇ + e−2iθ ω α k̄ α̇
= cos 2θ(k α ω̄ α̇ + ω α k̄ α̇ ) + sin 2θ(ik α ω̄ α̇ − iωα k̄ α̇ )
= y µ cos 2θ + z µ sin 2θ . (A.60)
In addition, z µ = i(k α ω̄ α̇ − ω α k̄ α̇ ) is a space-like unit vector, orthogonal to the vectors

xµ and y µ . Together with y µ , it constitutes the flagpole.
In order to fix the notation, we know that two-dimensional spinor representations of
the Lorentz group can be derived from the property that, under a Lorentz transforma-
tion, a contravariant 4-vector xµ transforms as xµ 7→ x′ µ = Rµ ν xν , where R ∈ SO(1,3)
satisfies Rµ ν ηµρ Rρ λ = ηνλ . The corresponding covariant 4-vector xµ 7→ ηµν xν satisfies
xν = x′µ Rµ ν . The most general proper orthochronous Lorentz transformation, corre-
sponding to a rotation by an angle of θ about an axis n̂, where ~θ = θn̂, and a boost
vector ζ 7→ v̂ tanh−1 β, where v̂ = v/kvk and where β = kvk, is a 4 × 4 matrix given
by
i ρσ
R = exp − θ Sρσ = exp −i~θ · S ~ − iζ · K
~ , (A.61)
2
where θi = 21 ǫijk θjk ; ζ i = θ i0 = −θ0i ; S i = 12 ǫijk Sjk ; K i = S 0i = −S i0 ; and
(Sρσ )µ ν = i(ηρ µ ησν − ησ µ ηρν ) . (A.62)
Here, the indices i, j, k = 1, 2, 3, and ǫ123 = +1 (Dreiner, Haber, and Martin, 2010).
It follows from (A.61, A.62) that an infinitesimal orthochronous Lorentz transfor-
mation is given by Rµ ν ≈ δνµ + θµ ν . Moreover, the infinitesimal boost parameter reads
βv̂, since β ≪ 1 for an infinitesimal boost. Hence, the actions of the infinitesimal
boosts and rotations on the spacetime coordinates are respectively given by
~x 7→ ~x′ ≈ ~x + (~
θ × ~x) , (t 7→ t′ ≈ t) ,
~ t , (t 7→ t′ ≈ t + β
~x 7→ ~x′ ≈ ~x + β ~ · ~x) . (A.63)
For contravariant 4-vectors, the reasoning is similar.
With respect to the Lorentz transformation R, a general n-component field Φ
transforms according to a representation R of the Lorentz group as Φ(xµ ) 7→ Φ′ (x′ µ ) =
[R] Φ(xµ ), where [R] is the corresponding (finite) d-dimensional matrix representation.
Equivalently, the functional form of the transformed field Φ obeys
Φ′ (xµ ) = [R]Φ([R−1 ]µ ν xν ) . (A.64)
For proper orthochronous Lorentz transformations,

i
R = exp − θµν J µν
≈ Id×d − i~θ · J~ − iζ~ · K
~, (A.65)
2
where Id×d is the d × d identity matrix, and θµν parameterises the Lorentz transforma-
tion R by (A.61). The six independent components of the matrix-valued antisymmetric
tensor J µν are the d-dimensional generators of the Lorentz group and satisfy the com-
mutation relations
[J µν , J λκ ] = i(g µκ J νλ + g νλ J µκ − g µλ J νκ − g νκ J µλ ) . (A.66)
The vectors J~ and K ~ are defined as the generators of rotations parameterised by ~θ
~ respectively, where J i = 1 ǫijk Jjk , and K i = J 0i .
and the boosts parameterised by ζ, 2
Here, we focus on the inequivalent non-trivial irreducible representations of the
Lorentz algebra D (1/2,0) and D(0,1/2) . In the D(1/2,0) representation, J~ = ~σ /2, and
~ = −i~σ /2, in eq. (A.65), so
K
R( 12 ,0) 7→ R ≈ I2×2 − iθ~ · ~σ /2 − ζ~ · ~σ /2 , (A.67)
where ~σ = (σ 1 , σ 2 , σ 3 ) represents the Pauli matrices. The transformation R carries
undotted spinor indices, as indicated by Rα β . A two-component spinor in the D(1/2,0)
representation is already denoted by ψα , which transforms as ψα 7→ Rα β ψβ .
On the other hand, in the D(0,1/2) representation, J~ = −~σ ∗ /2, and K~ = −iσ̃ ∗ /2, in
∗
eqn (A.65). Hence, its representation matrix is R , the complex conjugate of eqn (A.67).
By definition, the indices carried by R∗ are dotted, as indicated by (R∗ )α̇ β̇ . It is already
known that a two-component D (0,1/2) spinor ψα̇† transforms as ψα̇† 7→ (R∗ )α̇ β̇ ψβ̇† .
It follows that the D(1/2,0) and D(0,1/2) representations are related by Hermitian
conjugation. In fact, if ψα denotes a D(1/2,0) spinor, then (ψα )† transforms as a D(0,1/2)
spinor. In combining spinors to make Lorentz tensors, it is useful to regard ψα̇† as a
row vector, and ψα as a column vector, with
ψα̇† 7→ (ψα )† . (A.68)
The Lorentz transformation property of ψα̇† then follows from (ψα )† 7→ (ψβ )† (R† )β̇ α̇ ,
where (R† )β̇ α̇ = (R∗ )α̇ β̇ .
In the dotted-index notation, the dagger is used to denote Hermitian conjugation,

as in (A.68). In fact, the dagger is used to denote the Hermitian conjugation of spinors
in most textbooks (Srednicki, 2007). However, it is worth emphasising that many
references in supersymmetry e.g. (Sohnius, 1985; Srivastava, 1986; West, 1990; Wess
and Bagger, 1992; Bailin and Love, 1994; Mohapatra, 2003) employ the Wess and
Bagger (1992) notation, where ψ α̇ ≡ ψα̇† = (ψα )† .
There are two additional spin-1/2 irreducible representations of the Lorentz group,
namely, (R−1 )⊺ and (R−1 )† , However, they are equivalent to the D(1/2,0) and the
D(0,1/2) representations, respectively. The spinors that transform under these repre-
sentations have the raised spinor indices ψ α and ψ †α̇ , respectively, with the transfor-
mation laws ψ α 7→ [(R−1 )⊺ ]α β ψ β , and ψ †α̇ 7→ [(R−1 )† ]α̇ β̇ ψ †β̇ , respectively. Lorentz
tensors can be derived from spinors by regarding ψ α as a row vector, and ψ † α̇ as a
column vector, with
ψ † α̇ 7→ (ψ α )† . (A.69)
The Lorentz transformation property ψ † α̇ then follows from
(ψ α )† 7→ [(R−1 )† ]α̇ β̇ (ψ β )† .
The spinor indices are raised and lowered with the two-index antisymmetric epsilon
symbol with non-zero components ǫ12 = −ǫ21 = ǫ21 = −ǫ12 = 1 , and similar sign
conventions for the dotted spinor indices. In particular, ǫα̇β̇ = (ǫαβ )∗ , and ǫα̇β̇ = (ǫαβ )∗ ,
as well.
Moreover, the Kronecker delta symbol reads δα̇β̇ = (δαβ )∗ . The epsilon symbols with
undotted and with dotted indices, respectively, satisfy
ǫαβ ǫγδ = −δαγ δβδ + δαδ δβγ , ǫα̇β̇ ǫγ̇ δ̇ = −δα̇γ̇ δβ̇δ̇ + δα̇δ̇ δβ̇γ̇ , (A.70)
yielding the so-called Schouten identities
ǫαβ ǫβγ = ǫγβ ǫβα = δαγ , (A.71)

ǫαβ ǫγδ + ǫαγ ǫδβ + ǫαδ ǫβγ = 0 . (A.72)
The same equations hold for dotted indices. To construct Lorentz invariant Lagrangians
and observables, in particular, Lorentz vectors are obtained by introducing the sigma
matrices σαµβ̇ and σ̄ µ α̇β defined by

0 0 10 1 1 01
σ = σ̄ = , σ = −σ̄ = ,
01 10

2 2 0 −i 3 3 1 0
σ = −σ̄ = , σ = −σ̄ = . (A.73)
i 0 0 −1
Hence, eqn (A.73) is equivalent to σ µ = (I2×2 , ~σ ), and σ̄ µ = (I2×2 , −~σ ), which can
be related by
σαµα̇ = ǫαβ ǫα̇β̇ σ̄ µ β̇β , σ̄ µ α̇α = ǫαβ ǫα̇β̇ σβµβ̇ ,
ǫαβ σβµα̇ = ǫα̇β̇ σ̄ µβ̇α , ǫα̇β̇ σαµβ̇ = ǫαβ σ̄ µα̇β . (A.74)
There is a one-to-one correspondence between each 2-spinor construction Vαβ̇ and the
associated Lorentz 4-vector V µ , provided by the Infeld–van der Waerden symbols
1 µβ̇α
Vµ = σ̄ Vαβ̇ , Vαβ̇ = V µ σµαβ̇ . (A.75)
2
In particular, if V µ is a real 4-vector, then Vαβ̇ is Hermitian. Moreover, it is often
useful to further simplify the notation by defining V α̇β = (Vαβ̇ )∗ . In this notation, an
Hermitian 2-spinor satisfies Vαβ̇ = V α̇β . Then,
(V ⊺ )αβ̇ = Vβ α̇ , (V ∗ )α̇β = (Vαβ̇ )∗ , (V † )αβ̇ = (V β α̇ )∗ = (V ∗ )β̇α .
A Hermitian 2-spinor satisfies V = V † or, equivalently, Vαβ̇ = (V ∗ )β̇α .

In addition, 2-spinors can be interpreted as 2 × 2 matrices. It is indeed convenient
to define the following:
(V ⊺ )α β = V β α , (V ∗ )α̇ β̇ = (Vα β )∗ , (V † )β̇ α̇ = (Vα β )∗ = (V ∗ )α̇ β̇ .
Note that the matrix transposition of Vα β interchanges the rows and columns of W
without modifying the relative heights of the α and β indices. Similar results hold for
Vαβ and V αβ by either lowering or raising the spinor indices.
For an anti-commuting two-component spinor ψ, the product ψ α ψ β is antisym-
metric with respect to the interchange of the spinor indices α and β. Hence, it must
be proportional to ǫαβ . Similar conclusions hold for the corresponding spinor products
with raised undotted indices and with lowered and raised dotted indices, respectively.
Thus,
1 1
ψ α ψ β = − ǫαβ ψψ , ψα ψβ = ǫαβ ψψ ,
2 2
1 † † 1
ψ † α̇ ψ † β̇ = ǫα̇β̇ ψ † ψ † , ψα̇ ψβ̇ = − ǫα̇β̇ ψ † ψ † , (A.76)
2 2
where ψψ = ψ α ψα and ψ † ψ † = ψα̇† ψ † α̇ .
The van der Waerden symbols in eqns (A.75) provide antisymmetrised products,
from the sigma matrices (Dreiner, Haber, and Martin, 2010):
i µ ν ρ̇β
(σ µν )α β = σαρ̇ σ̄ − σαν ρ̇ σ̄ µρ̇β , (A.77)
4
i µ α̇ρ ν
(σ̄ µν )α̇ β̇ = σ̄ σρβ̇ − σ̄ ν α̇ρ σ ρµβ̇ . (A.78)
4
Now we can introduce the infinitesimal forms for the 4 × 4 Lorentz transformation
1
matrix; the corresponding matrices R and (R−1 )† which transform the D( 2 ,0) and
1
D(0, 2 ) spinors, respectively, are given by
1
Rµ ν ≈ δνµ +
θαν g αµ − θνβ g βµ , (A.79)
2
1
R ≈ I2×2 − iθµν σ µν , (A.80)
2
1
(R ) ≈ I2×2 − iθµν σ̄ µν .
−1 †
(A.81)
2
The inverses of these quantities are obtained up to the first order in θ by replacing
θ 7→ −θ in the formulæ . Equations (A.80) and (A.81) yield
(R−1 )ρ σ = ǫσα Rα β ǫβρ , (R−1 † )ρ̇ σ̇ = ǫσ̇α̇ (R† )α̇ β̇ ǫβ̇ ρ̇ . (A.82)
These results prove the covariance of the spinor index raising and lowering properties
of the epsilon symbols. The infinitesimal forms given by (A.79) and (A.81) imply that
R† σ̄ µ R = Rµ ν σ̄ ν , R−1 σ µ (R−1 )† = Rµ ν σ ν . (A.83)
Using the Lorentz transformation properties of the undotted and dotted two-component
spinor fields, eqn (A.83) yields the proof that the spinor products ξ † σ̄ µ η and ξσ µ η †
transform as Lorentz 4-vectors.
The usual framework use in field theory regards a pure boost from the rest frame to
a frame where pµ = (Ep , p), which corresponds to θij = 0, and ζ i = θ i0 = −θ0i . The
so-called mass-shell condition is satisfied: p0 = Ep = (kpk2 + m2 )1/2 . The matrices
Rα β and [(R−1 )† ]α̇ β̇ , which describe Lorentz transformations of spinor fields, are given,
respectively, for the D (1/2,0) and D (0,1/2) representations by
 r
 R = exp − 1 ζ~ · ~σ = p · σ ,

i 2 m
exp − θµν J µν = r (A.84)
2 
 1 ~ p · σ̄
(R−1 )† = exp ζ · ~σ = ,
2 m
where
√ (Ep + m) I2×2 − ~σ · p
p
p·σ = ,
2(Ep + m)
√ (Ep + m) I2×2 + ~σ · p
p · σ̄ = p . (A.85)
2(Ep + m)
√ √
According to (A.84), the spinor index structure of p · σ and p · σ̄ corresponds to
that of Rα β and [(R−1 )† ]α̇ β̇ , respectively. Hence, the equations in (A.85) yield
√ β
p β (p · σαα̇ )σ̄ 0 α̇β + mδαβ
p·σ α
= p · σ σ̄ 0 α
= p , (A.86)
2(Ep + m)
√ α̇ p α̇ (p · σ̄ α̇α )σα0 β̇ + mδβ̇α̇

p · σ̄ β̇ = p · σ̄ σ 0 β̇ = p , (A.87)
2(Ep + m)
since σ 0 = σ̄ 0 = I2×2 .
The Supersymmetry Algebra 223
A.7 The Supersymmetry Algebra

The two operators φa and φb in a graded Lie algebra satisfy
φa φb − (−1)|a||b| φb φa = Cab
d
φd ,
where either |a| = 0 for even (bosonic) φa or |a| = 1 for odd (fermionic) operators and
d
the Cab denote the algebra structure constants. The Poincaré generators J µν in eqn
(A.66), together with the P µ , are bosonic generators. Nevertheless, in supersymmetry,
β
fermionic generators QA α and Q̄α̇ are introduced, respectively denoting elements of the
( 12 ,0) (0, 21 )
D and D representations of the Lorentz group, and A, B = 1, . . . , N label the
number of supercharges (West, 1990; Wess and Bagger, 1992).
For N = 1, the supersymmetry algebra reads
[M µν , M σρ ] = i(M µν η νρ + M νρ η µσ − M µρ η νσ − M νσ η µρ ) , (A.88)
[P µ , P ν ] = 0 = [Qα , Pµ ] = {Qα , Qβ } , (A.89)
[M µν , P σ ] = i(P µ η νσ − P ν η µσ ) , (A.90)
[Qα , M µν ] = (σ µν )αβ Qβ , (A.91)
{Qα , Q̄β̇ } = 2(σ µ )αβ̇ Pµ . (A.92)
The first three equations describe the usual Poincaré algebra.

A spinor Qα transforms under an infinitesimal Lorentz transformation as
β
′ − 2i ωµν σ µν β i µν
Qα 7→ Qα = (e )α Qβ ≈ I − ωµν σ Qβ . (A.93)
2 α
i µν
From the operator point of view, it transforms, by denoting U = e− 2 ωµν M , as

′ † i µν i µν
Qα 7→ Qα = U Qα U ≈ I + ωµν M Qα I − ωµν M . (A.94)
2 2
When eqn (A.94) is compared to eqn (A.93), eqn (A.91) can be derived. Indeed,
i i
Qα − ωµν (σ µν )αβ Qβ = Qα − ωµν (Qα M µν − M µν Qα ) + O(ω 2 ), (A.95)
2 2
which implies that
[Qα , M µν ] = (σ µν )αβ Qβ . (A.96)
The commutator for the right-handed representation reads
[Q̄α̇ , M µν ] = (σ̄ µν )α̇β̇ Q̄β̇ . (A.97)
For more detail, the reader can see, for example, the references by West (1990),
Wess and Bagger (1992), Bailin and Love (1994), and Dreiner, Haber, and Martin
(2010). Moreover, the Clifford–Hopf algebra associated with the super-Poincaré algebra
was formulated in the article by da Rocha, Bernardini, and Vaz Jr (2010).
Appendix B
List of Symbols
a (paravector), 1981
Â (grade involution of A ∈ T(V )), 17
Ā (conjugation of A ∈ T(V )), 18
Ā (Hermitian conjugation), 118
Ã (reversion of A ∈ T(V )), 18
|A| (norm of A), 69
|A|′ (equivalent norm of A), 69
A (algebra), 16
A/ ∼ (quotient algebra), 34
A[p] (p-vector), 22
A{p} (= A[n−p] ), 47
AAB (= fA AfB , bilateral ideal of A), 104
Aop (opposed algebra), 103
A⊺ (transposition of a matrix A), P3
Aµ1 µ2 ···µp eµ1 ∧ eµ2 ∧ · · · ∧ eµp = µ1 <µ2 <···<µp Aµ1 µ2 ···µp eµ1 ∧ eµ2 ∧ · · · ∧ eµp (arbitrary
p-vector), 28
Alt (alternator operator), 16, 21
Aut(Cℓp,q ) (group of the automorphisms), 126
B (bilinear functional), 162

B (basis of a vector space), 2
B∗ (basis of a dual vector space), 2
B i j (change of basis matrix components), 3
Bp (basis of the space Tp (V )), 23
( )c (complex conjugation in M(r, C)), 116; (charge conjugation operator), 169

C (complex field), 2
C2 , 145
C (ideal of an algebra A), 34
Cℓ(S ± , h) (Clifford algebra over semispinor spaces), 176
Cℓ(V, g) (universal Clifford algebra), 65
± Cℓ(V, g) (the set {ψ ∈ Cℓ(V, g) | ψ = ±ψη}), 86
1 The Greek letters and some special symbols (e. g. tensor product ⊗, etc.) can be found at the
end of this list.
List of Symbols 225
Cℓ3 (three-dimensional Euclidean Clifford algebra), 70

Cℓp,q (Clifford algebra associated with the quadratic space Rp,q ), 90
CℓC (V, g) (complexified universal Clifford algebra), 89
Cℓ+ (V, g) (even subalgebra), 68
Cℓ− (V, g) (odd subspace), 68
Cℓ0 ⊕ Cℓ1 (arbitrary Z2 -grading of Cℓ(V, g)), 155
Cℓ+
p,q (even Clifford algebra associated with the quadratic space R
p,q
), 92
∗
Cℓp,q (group of the invertible (or regular) elements), 126
Cen(Cℓp,q ) (centre of a Clifford algebra), 67
Conf(p, q) (conformal group), 140
SConf(p, q) (time-preserving and future-pointing component of the conformal group
connected to the identity), 140
Conf + (p, q) (Möbius group), 140
D, Pµ , Kµ , Mµν (generators of the Lie algebra of the conformal group), 143

D (duplex numbers), 94
deg(xk ) (degree of an element xk ∈ Ak ), 17
det (determinant of a matrix), 22
dim(V ) (dimension of V ), 2
D(1/2,0) and D(0,1/2) (non-representations of the Lorentz group), 146
e (canonical basis of a one-dimensional vector space), 94

E (triality space), 170
E (creation operator), 74
e (basis of Cℓ0,3 ), 193
ei (canonical basis of a vector space), 2
EA (generators of Cℓ4,1 ), 113
Eij (basis of the space of matrices n × n), 19
ei (canonical basis of a dual vector space), 2
eβe0123 (duality rotation), 165
E† (creation operator), 74
End(A) (space of endomorphisms of the algebra A), 102
EndK (S) (space of endormorphisms of the algebraic spinors space), 165
End(V ) (space of endomorphisms of V ), 19
exp (exponential mapping inside a Clifford algebra), 135
f (idempotent), 103
fA (primitive idempotent), 104
f± (idempotents in Cℓ3,0 ), 190
F (space of continuous functions), 5
F, F ′ (maximal totally isotropic subspaces), 180
g (symmetric bilinear form),

V 7
G (bilinear functional in (V )), 49
226 List of Symbols
gC (complexified symmetric bilinear form), 89

gij (symmetric bilinear form components), 7
G(V ) (Grassmann algebra), 49
h (spinor metric on semispinor spaces), 170

h̃ (spinor inner product), 166
h̄ (spinor inner product), 166
H (quaternion ring), 1
i (volume element in Cℓ3 ), 70

i, j, k (quaternionic units), 120
I (identity operator), 19
I (bilateral ideal), 33
I (annihilation operator), 74
I† (annihilation operator), 74
IdW , 19
IL (left ideal), 33
IR (right ideal), 33
IC (ideal of T(V ) generated by the elements v ⊗ v − Q(v)1, where v ∈ V ), 67
IE (ideal of T(V ) generated by the elements v ⊗ v, where v ∈ V ), 34
J (sign of an orientation), 42
J (current density), 203
Ji (components of the vector bilinear covariant), 162
Jχ̄χ′ (Robinson congruence), 202
ker f (kernel of a linear mapping f ), 6

K (field), 1
K (undotted contravariant spinor), 190
L(a) (left regular representation in End(A)), 102

Lin(V, W ) (space of the linear mappings from V to W ), 10
Lin(2) (V, W ; U ) (space of the bilinear mappings from V × W to U ), 10
M1 , M2 (maximal totally isotropic subspaces), 184

M(n, K) (vector space of the n × n matrices with entries in K), 19
mA (algebra product), 16
N (a) (norm in Cℓp,q ), 132

N ′ (a) (norm in Cℓp,q ), 132
O(p, q) (orthogonal group of Rp,q ), 121

O+ (p, q) (orthogonal group of Rp,q ; connected to the identity), 123
List of Symbols 227
O↑ (p, q) (orthogonal group of Rp,q ; future oriented), 123

SO(p, q) (special orthogonal group of Rp,q ), 121
O (octonion algebra), 103
p! (p factorial), 21
Pin(p, q) (Pin group), 132
Pin+ (p, q) (reduced Pin group), 133
Pinˆ(p, q) (Pin group with the norm N ′ ), 132
Pinˆ+ (p, q) (reduced Pin group with the norm N ′ ), 133
P (projector), 15
P2 (space of polynomial functions of degree less than or equal to 2), 5
Pi (projectors), 15
Q (quadratic form), 7
rj (Radon-Hurwitz numbers), 160

R (Lorentz transformation), 194
R (real field), 2
R(a) (right regular representation in End(A)), 103
d
R p,q (conformal compactification of Rp,q ), 139
rank f (rank of a linear mapping f ), 6
Sij (components of the bivector bilinear covariant), 162

Sp (symmetric group), 21
Sp,q (space of classical spinors), 152
SU (orthogonal symmetry), 124
S 2 (2-sphere), 139
S p (p-sphere), 139
SnA (algebraic spinor of Cℓn (C)), 161
SnC (classical spinor of Cℓn (C)), 161
SnO (spinor operator Cℓn (C)), 161
Sp,q
A
(algebraic spinor), 159
Sp,q (classical spinor) 159
C
Sp,q
O
(spinor operator), 159
±
Sp,q (semispinors), 152
SL(2, C) (Lorentz group), 138
SO(3), 145
Spin(p, q) (Spin group), 133
Spin(3), 145
Spin+ (p, q) (Spin group associated to the even Clifford algebra), 133
$pin+ (1, 3) ({s ∈ Cℓ3,0 |ss̄ = 1}), 138
Spinˆ(p, q) (Spin group with the norm N ′ ), 132
Spinˆ+ (p, q) (reduced Spin group with the norm N ′ ), 132
228 List of Symbols
SU(2), 145
Sym (symmetriser operator), 16
ν ν ···ν
T = Tµ11µ22 ···µqp eµ1 ⊗ eµ2 ⊗ · · · ⊗ eµp ⊗ eν1 ⊗ eν2 ⊗ · · · ⊗ eνq (arbitrary tensor of type
(p, q)), 14
T(V ) [T∗ (V )] (algebra of the contravariant [covariant] tensors), 17
T(V )/IE (exterior algebra), 34
T2 (V ) = V ∗ ⊗ V ∗ , T2 (V ) = V ⊗ V , 11
Tp (V ), Tq (V ), Tpq (V ), Tq p (V ), 14
T∗± (V ) = Π± (T∗ (V )), 18
T1 1 (V ) = V ∗ ⊗ V , T1 1 (V ) = V ⊗ V ∗ , 13
T2ant (V ) (space of antisymmetric bilinear functionals), 12
T2sym (V ) (space of symmetric bilinear functionals), 12
Tr (trace operator), 19
ui (right chiral semispinor), 171

u0 (unit right chiral semispinor), 173
U ⊥ (orthogonal complement of s subspace U ⊆ V ), 124
v (vector), 1
V (vector space), 1
v × u (vector product), 56
v ∧ u (2-vector), 40
v♭ (dual vector), 7
vi (left chiral semispinor), 171
Vπ (paravector space of V ), 138
VC (complexified vector space), 89
v i (components of a vector), 3
V ∗ (dual vector space), 2
∗
(V ∗ ) (bidual vector space), 5
µ1 ···µk
V (Plücker coordinates), 55
[x] (equivalence class of an element x), 33
Z (maximal totally isotropic subspaces), 184

Z (aggregate/boomerang), 204
Z (integers), 17
Z2 (integers modulo 2), 67
ZN (integers modulo N ), 33
α (dual vector or covector), 2

α1 (reversion), 92
α−1 (conjugation), 92
List of Symbols 229
α♯ (dual covector), 7
αi (components of a covector), 3
α ∧ β (2-covector), 40
γ (Clifford mapping acting of vectors), 57
γ (Clifford mapping), 57
γ ± (Clifford mappings), 75
γ5 (volume element of Minkowski spacetime), 143
γi (= γ(ei )), 70
γ †± (Clifford mappings), 76
Γ (Clifford mapping), 88
Γp,q (Clifford–Lipschitz group), 126
Γ̂p,q (twisted Clifford–Lipschitz group), 128
Γ̂+
p,q (even twisted Clifford–Lipschitz group), 128
δji (Kronecker delta symbol), 3
ε(σ) (sign of the permutation σ) 22
ζ (mapping from the right chiral spinor space to a vector space), 173
η (volume element in Rp,q ), 73
ηx (reference twistor), 201
Θ (triality order 3 automorphism), 175
ı (inclusion V ֒→ T(V )), 64
λ (vector space isomorphism between the Clifford and Grassmann algebras), 82
Λ (pseudoscalar of Cℓp,q ), 160
ξ (bidual mapping), 5
ξi (eigenvectors), 9
ξv (bidual vector), 5
π (quotient mapping T → T(V )/IC ≃ Cℓ(V, g)), 61
Π± (projectors), 18, 152
ρ (graded regular representation), 152
ρ+ (even graded regular representation), 152
σ (scalar bilinear covariant), 162
b (twisted adjoint representation), 128
σ
ς (spinor structure mapping), 165
τ (correlation), 6
φV (isomorphism V ⊗ V ∗ → Lin(V, V )), 19
χ (path on a group), 122
Ψ[p] (p-covector), 22
ω (pseudoscalar bilinear covariant), 162
Ω (n-vector), 42
ΩV [ΩV ∗ ], (n-vector [n-covector]), 42
♭ (correlation), 7
♯ (inverse correlation), 7
⊗ (tensor product), 10
ˆ (alternating tensor product), 88
⊗
∧ (exterior product), 24
∨ (regressive product), 47
230 List of Symbols
g (representation of the geometric product of the Grassmann algebra), 82

#
V (grade V involution), 17
∗
V (V ) [ (V
Vp )] (exterior algebra of V [V ∗ ]), 29
Wp (V ) [ V (V )] (space of p-vectors [p-covectors]), 23
p (V ) ( n−p (V )), 47
⌋ (left contraction), 36
⌊ (right contraction), 38
∼ (equivalence relation), 33
⋆, ⋆ (quasi-Hodge isomorphism), 43
≃ (vector space isomorphism), 66
V
⋆ (Hodge isomorphism), 51
⋄ (Clifford product), 60
×˙ (vector product), 56
∗ (deformed Clifford product in R5 ), 85
( )∗ (involutive automorphism in M(r, C)), 115
( )⋆ (equivalent involutive automorphism in M(r, C)), 115
◦ (Chevalley product), 171; (octonionic product), 176
( )◦ (anti-automorphisms of Cℓp,q ), 166
⊙ (deformed Clifford product in R2 ), V 85
h ip (projection operators Cℓ(V, g) → p (V )), 30, 66
△ (inner automorphism in Cℓ4,1 ), 115
[a, b] (Lie algebra commutator in Cℓ∗p,q ), 135
0A (zero element of the algebra A), 57
1A (unit element of the algebra A), 57
References
Ablamowicz, R. (1995). Construction of spinors via Witt decomposition and prim-

itive idempotents: A review. In Clifford Algebras and Spinor Structures (ed.
R. Ablamowicz and P. Lounesto), pp. 113–123. Kluwer, Amsterdam.
Ablamowicz, R., Gonçalves, I., and da Rocha, R. (2014). Bilinear covariants
and spinor fields duality in quantum Clifford algebras. Journal of Mathematical
Physics, 55, 103501.
Ablamowicz, R., and Sobczyk, G. (eds) (2004). Lectures on Clifford (Geometric)
Algebras and Applications. Birkhäuser, Berlin.
Ahlford, L. V. (1986). Möbius transformations in Rn expressed through 2×2 matrices
of Clifford numbers. Complex Variables, Theory and Application, 5(2–4), 215–224.
Ahluwalia, D. V., and Grumiller, D. (2005). Spin-half fermions with mass dimension
one: Theory, phenomenology, and dark matter. Journal of Cosmology and Astropar-
ticle Physics, 2005(07), 012.
Ahluwalia, D. V., Lee, C.-Y., and Schritt, D. (2010). Elko as self-interacting fermionic
dark matter with axis of locality. Physics Letters B , 687(2–3), 248–252.
Anglès, P. (2008). Conformal Groups in Geometry and Spin Structures. Birkhäuser,
Providence, RI.
Baez, J. (2002). The octonions. Bulletin of the American Mathematical Society, 39(2),
145–205.
Bailin, D., and Love, A. (1994). Supersymmetric Gauge Field Theory and String
Theory. Institute of Physics Publishing, Bristol.
Baylis, W. (1996). Clifford (Geometric) Algebras with Applications in Physics, Math-
ematics and Engineering. Birkhäuser, Boston, MA.
Baylis, W. (1998). Electrodynamics: a Modern Geometric Approach. Birkhäuser,
Boston, MA.
Benn, I., and Tucker, R. (1987). An Introduction to Spinors and Geometry with
Applications in Physics. Adam Hilger, Bristol.
Berkovits, N. (2004). Multiloop amplitudes and vanishing theorems using the pure
spinor formalism for the superstring. Journal of High Energy Physics, 2004(09),
047.
Berkovits, N., and Howe, P. S. (2002). Ten-dimensional supergravity constraints from
the pure spinor formalism for the superstring. Nuclear Physics B , 635(1–2), 75–105.
Bernardini, A. E., and da Rocha, R. (2012). Dynamical dispersion relation for ELKO
dark spinor fields. Physics Letters B , 717(13), 238–241.
Bethe, H. A., and Salpeter, E. E. (1957). Quantum Mechanics of One- and Two-
Electron Atoms. Springer, Berlin.
Binz, E., de Gosson, M. A., and Hiley, B. J. (2013). Clifford algebras in symplectic
geometry and quantum mechanics. Foundations of Physics, 43(4), 424–439.
232 References
Birkhoff, G., and MacLane, S. (1997). A Survey of Modern Algebras (4th edn).
Macmillan Publishing Company, New York, NY.
Bonora, L., and da Rocha, R. (2016). New spinor fields on lorentzian 7-manifolds.
Journal of High Energy Physics, 2016(39), 1–17.
Bonora, L., de Brito, K. P. S., and da Rocha, R. (2015). Spinor fields classification
in arbitrary dimensions and new classes of spinor fields on 7-manifolds. Journal of
High Energy Physics, 2015(2).
Bourbaki, N. (1989). Elements of Mathematics: Algebra I, Chapters 1–3. Number 1.
Springer, Berlin.
Brauer, R., and Weyl, H. (1935). Spinors in n dimensions. American Journal of
Mathematics, 35(1), 425–449.
Budinich, P. (2002). From the geometry of pure spinors with their division algebra
to fermion physics. Foundations of Physics, 32(9), 1347–1398.
Budinich, P., and Trautman, A. (1989). The Spinorial Chessboard. Springer, New
York, NY.
Cartan, É. (1908). Nombres complexes. In Encyclopédie des sciences mathématiques
(ed. J. Molk), Vol. 1. Gauthiers-Villars, Paris.
Cartan, É. (1937). Leçons sur la theorie des spineurs. Hermann, Paris. Reprinted
as: The Theory of Spinors, Dover Publications, New York, NY, 1981.
Cavalcanti, R. T. (2014). Classification of singular spinor fields and other mass di-
mension one fermions. International Journal of Modern Physics D, 23(14), 1444002.
Cavalcanti, R. T., da Silva, J. M. H., and da Rocha, R. (2014). VSR symmetries
in the DKP algebra: The interplay between Dirac and ELKO spinor fields. The
European Physical Journal Plus, 129(246), 1–10.
Charlier, A., Bérard, A., Charlier, M.-F., and Fristot, D. (1992). Tensors and the Clif-
ford Algebra; Application to the Physics of Bosons and Fermions. Marcek Dekker,
Inc, New York.
Chevalley, C. (1954). The Algebraic Theory of Spinors. Columbia University Press,
New York, NY. Reprinted as The Algebraic Theory of Spinors and Clifford Algebras,
Collected Works, Vol. 4, Springer, Berlin, 1997.
Clifford, W. K. (1878). Applications of Grassmann’s extensive algebra. American
Journal of Mathematics, 1(4), 350–358.
Clifford, W. K. (1882). On the classification of geometric algebras, pp. 397–401.
Macmillan, London. In Mathematical Papers (ed. R. Tucker), Reprinted by Chelsea,
New York, NY 1985.
Crawford, J. P. (1985). On the algebra of Dirac bispinors densities: Factorization and
inversion theorems. Journal of Mathematical Physics, 26(7), 1439–1441.
Crowe, M. J. (1994). A History of Vector Analysis. Dover Publications, New York,
NY.
Crumeyrolle, A. (1990). Orthogonal and Symplectic Clifford Algebras: Spinor Struc-
tures. Kluwer, Dordrecht.
da Rocha, R., Bernardini, A. E., and da Silva, J. M. H. (2011). Exotic dark spinor
fields. Journal of High Energy Physics, 2011(4), 1–26.
da Rocha, R., Bernardini, A. E., and Vaz, J. Jr. (2010). κ-deformed Poincare algebras
and quantum Clifford–Hopf algebras. International Journal of Geometric Methods
References 233
in Modern Physics, 7(5), 821–836.

da Rocha, R., and da Silva, J. M. H. (2009). Elko spinor fields: Lagrangians for
gravity derived from supergravity. International Journal of Geometric Methods in
Modern Physics, 6(3), 461–477.
da Rocha, R., and da Silva, J. M. H. (2010). Elko, flagpole and flag-dipole spinor fields,
and the instanton Hopf fibration. Advances in Applied Clifford Algebras, 20(1), 847–
870.
da Rocha, R., and da Silva, J. M. H. (2014). Hawking radiation from Elko particles
tunnelling across black-strings horizon. EPL (Europhysics Letters), 107(5), 50001.
da Rocha, R., Fabbri, L., da Silva, J. M. H., Cavalcanti, R. T., and Silva-Neto,
J. A. (2013). Flag-dipole spinor fields in ESK gravities. Journal of Mathematical
Physics, 54(10), 102505.
da Rocha, R., and Rodrigues, W. A. Jr. (2006). Where are ELKO spinor fields in
lounesto spinor field classification? Modern Physics Letters A, 21(1), 65–74.
da Rocha, R., and Rodrigues, W. A. Jr. (2010). Pair and impair, even and odd form
fields, and electromagnetism. Annalen der Physik , 19(1–2), 6–34.
da Rocha, R., and Traesel, M. A. (2012). Generalized non-associative structures on
the 7-sphere. Journal of Physics: Conference Series, 343(1), 012026.
da Rocha, R., Traesel, M. A., and Vaz, J. Jr. (2012). Non-associativity in the Clif-
ford bundle on the parallelizable torsion 7-sphere. Advances in Applied Clifford
Algebras, 22(3), 595–610.
da Rocha, R., and Vaz, J. Jr. (2006). Extended Grassmann and Clifford algebras.
Advances in Applied Clifford Algebras, 16(1), 103–125.
da Rocha, R., and Vaz, J. Jr. (2007). Isotopic liftings of Clifford algebras and appli-
cations in elementary particle mass matrices. International Journal of Theoretical
Physics, 46, 2464–2487.
da Rocha, R., and Vaz Jr, J. (2006). Clifford algebra-parametrized octonions and
generalizations. Journal of Algebra, 301(2).
da Rocha, R., and Vaz Jr, J. (2007). Conformal structures and twistors in the par-
avector model of spacetime. International Journal of Geometric Methods in Modern
Physics, 4(4), 547–576.
Darling, R. W. R. (1994). Differential Forms and Connections. Cambridge University
Press, Cambridge.
de Andrade, M. A., Rojas, M., and Toppan, F. (2001). The signature triality of
Majorana–Weyl space-times. International Journal of Modern Physics A, 16(27),
4453–4479.
de Andrade, M. A., and Toppan, F. (1999). Real structures in Clifford algebras
and Majorana conditions in any space-time. Modern Physics Letters A, 14(26),
1797–1814.
Dirac, P. A. M. (1928). The quantum theory of the electron. Proceedings of the Royal
Society of London A: Mathematical, Physical and Engineering Sciences, 117(778),
610–624.
do Carmo, M. P. (1994). Differential Forms and Applications. Springer, Berlin.
Doran, C., and Lasenby, A. (2003). Geometric Algebra for Physicists. Cambridge
University Press, Cambridge.
234 References
Dorst, L., Fontijne, D., and Mann, S. (2007). Geometric Algebra for Computer Sci-
ence. Morgan Kaufmann, Burlington, MA.
Dorst, L., and Lasenby, J. (eds.) (2011). Guide to Geometric Algebra in Practice.
Springer, Berlin.
Edelen, D. G. B. (1985). Applied Exterior Calculus. John Wiley & Sons, New York,
NY.
Fearnley-Sandre, D., and Stokes, T. (1997). Area in Grassmann geometry. Lecture
Notes in Computer Science, 1360(1), 141–170.
Figueiredo, V. L., de Oliveira, E. C., and Rodrigues, W. A. Jr. (1990). Covariant,
algebraic and operator spinors. International Journal of Theoretical Physics, 29(2),
371–395.
Fillmore, J. P., and Springer, A. (1990). möbius group over general fields using
Clifford algebras associated with spheres. International Journal of Theoretical
Physics, 29(3), 225–246.
Fjelstad, P. (1986). Extending special relativity via the perplex numbers. American
Journal of Physics, 54(5), 416–422.
Flanders, H. (1963). Differential Forms with Applications to the Physical Sciences.
Academic Press, New York, NY.
Frankel, T. (2012). The Geometry of Physics: An Introduction (3rd edn). Cambridge
Gallier, J. (1997). Clifford algebras, Clifford groups, and a generalization of the
quaternions. arXiv:0805.0311 [math.GM].
Garling, D. J. H. (2011). Clifford Algebras. An Introduction. Cambridge University
Press, Cambridge.
Gilbert, J. E., and Murray, M. A. M. (1991). Clifford Algebras and Dirac Operators
in Harmonic Analysis. Cambridge University Press, Cambridge.
Goldstein, H., Poole, C. P., and Safko, J. L. (2001). Classical Mechanics (3rd edn).
Addison-Wesley, Harlow.
Grassmann, H. (1844). Die lineale Ausdehnungslehre. Wiegand, Leipzig.
Grassmann, H. (1847). Geometrische Analyse: Geknüpft an die von Leibniz erfundene
geometrische Charakteristik. Weidmannsche Buchhandlung, Leipzig.
Grassmann, H. (1862). Die Ausdehnungslehre: Vollstandig und in strenger Form
bearbeitet. Enslin, Berlin.
Grassmann, H. (1894). Hermann Grassmanns gesammelte Mathematische und
physikalische Werke. Druck und Verlag von B. G. Teubner, Leipzig. (3 volumes
1894–1911).
Greub, W. H. (1978). Multilinear Algebra (2nd edn). Springer, Berlin.
Gsponer, A., and Hurni, J.-P. (1994). Lanczos’s equation to replace Dirac’s equation?,
pp. 509–512. SIAM, Philadelphia, PA. In Proceedings of the Cornelius Lanczos
International Centenary Conference (ed. J. D. Brown, M. T. Chu, D. C. Ellison,
and R. J. Plemmons).
Gunaydin, M., and Ketov, S. V. (1996). Seven-sphere and the exceptional n = 7 and
n = 8 superconformal algebras. Nuclear Physics B , 467(1–2), 215–246.
Gürsey, F. (1956). Correspondence between quaternions and four-spinors. Review of
the Faculty of Science, University of Istanbul, Series A, 21(1), 33–54.
References 235
Gürsey, F. (1958). Relation of charge independence and baryon conservation to

Pauli’s transformation. Il Nuovo Cimento (1955–1965), 7(3), 411–415.
Hamilton, W. R. (1866). Elements of Quaternions. Longmans Green, London.
Harvey, F. R. (1990). Spinors and Calibrations. Academic Press, New York, NY.
Hasiewicz, Z., and Kwasniewski, A. K. (1985). Triality principle and g2 group in
spinor language. Journal of Mathematical Physics, 26(1), 6–11.
Hehl, F. W., and Obukhov, Y. N. (2003). Foundations of Classical Electrodynamics:
Charge, Flux and Metric. Birkhäuser, Berlin.
Helmstetter, J., and Micali, A. (2008). Quadratic Mappings and Clifford Algebras.
Birkhäuser, Berlin.
Hestenes, D. (1966). Spacetime Algebra. Gordon and Breach, New York.
Hestenes, D. (1967). Real spinor fields. Journal of Mathematical Physics, 8(4), 798
– 808.
Hestenes, D. (1991). The design of linear algebra and geometry. Acta Applicandae
Mathematica, 23(1), 65–93.
Hiley, B. J. (2011). Process, distinction, groupoids and Clifford algebras: An alter-
native view of the quantum formalism. Lecture Notes in Physics, 813(1), 705 –
750.
Hladik, J. (1999). Spinors in Physics. Springer, Berlin.
Hoffman, K., and Kunze, R. (1971). Linear Algebra (2nd edn). Pearson, London.
Holland, P. R. (1986). Relativistic algebraic spinors and quantum motions in phase
space. Foundations of Physics, 16(2), 701–719.
Ivanova, T. (1993). Octonions, self-duality and strings. Physics Letters B , 315(3),
277–282.
Keller, J. (1997). Spinors, twistors, mexors and the massive spinning electron. Ad-
vances in Applied Clifford Algebras, 7, 439 – 455.
Klotz, F. S. (1974). Twistors and the conformal group. Journal of Mathematical
Physics, 15, 2242–2247.
Knus, M.-A. (1998). The Book of Involutions. Vol. 44, American Mathematical
Society Colloquium Publications.
Kostrykin, A. I., and Manin, Yu. I. (1997). Linear Algebra and Geometry. Gordon
and Breach, New York, NY.
Lam, T. Y. (1980). The Algebraic Theory of Quadratic Forms. Benjamin, Reading,
MA.
Laufer, A. (1997). The Exponential Map, Clifford Algebras, Spin Representations and
Spin Gauge Theory of U(1)×T 4 . Phd thesis, Universität Konstanz, Konstanz.
Lawson, H. B., and Michelson, M.-L. (1990). Spin Geometry. Princeton University
Press, Princeton, NJ.
Lazaroiu, C. I., Babalic, E. M., and Coman, I. A. (2013). The geometric algebra
of Fierz identities in arbitrary dimensions and signatures. Journal of High Energy
Physics, 2013(9), 1–74.
Lounesto, P. (1996). Counter-examples in Clifford algebras. Advances in Applied
Clifford Algebras, 6(1), 69–104.
Lounesto, P. (2001a). Clifford Algebras and Spinors (2nd edn). Cambridge University
Press, Cambridge.
236 References
Lounesto, P. (2001b). Octonions and triality. Advances in Applied Clifford Alge-

bras, 11(2), 191–213.
Lounesto, P., and Wene, G. P. (1987). Idempotent structure of Clifford algebras.
Acta Applicandae Mathematica, 9(3), 165–173.
Majorana, E. (1932). Oriented atoms in a variable magnetic field. Il Nuovo Ci-
mento, 9(1), 43–50.
Maks, J. (1989). Modulo (1,1) Periodicity of Clifford Algebras and Generalized
(anti)Möbius Transformations. Phd thesis, Technische Universiteit Delft, Delft.
Meinrenken, E. (2013). Clifford Algebras and Lie Theory. Springer, Berlin.
Misner, C. W., and Wheeler, J. A. (1957). Classical physics as geometry. Annals of
Physics, 2(1), 525–603.
Mohapatra, R. N. (2003). Unification and Supersymmetry: The Frontiers of Quark-
Lepton Physics (3rd edn). Springer, Berlin.
Mosna, R. A., Miralles, D., and Vaz Jr, J. (2003). z2 -gradings of Clifford algebras and
multivector structures. Journal of Physics A: Mathematical and General , 36(15),
4395–4405.
O’Neill, B. (2006). Elementary Differential Geometry (revised 2nd edn). Academic
Press, New York, NY.
Pauli, W. (1927). Zur Quantenmechanik des magnetischen elektrons. Zeitschrift für
Physik , 43(1), 601–623.
Penrose, R. (1967). Twistor algebra. Journal of Mathematical Physics, 8(2), 345–366.
Penrose, R., and Rindler, W. (1984). Spinors and Space-Time: Volume 1, Two-Spinor
Calculus and Relativistic Fields. Cambridge University Press, Cambridge.
Perwass, C. (2008). Geometric Algebra with Applications in Engineering. Springer,
Berlin.
Plymen, R. J., and Robinson, P. L. (1990). Spinors in Hilbert Spaces. Cambridge
Porteous, I. R. (1969). Topological Geometry. Cambridge University Press, Cam-
bridge.
Porteous, I. R. (1995). Clifford Algebras and the Classical Groups. Cambridge Uni-
versity Press, Cambridge.
Rainich, G. Y. (1925). Electrodynamics in the general relativity theory. Transactions
of the American Mathematical Society, 27(1), 106–136.
Riesz, M. (1993). Clifford Numbers and Spinors. Kluwer, Dordrecht.
Rodrigues, W. A. Jr, da Rocha, R., Bernardini, A. E., and Vaz, J. Jr. (2005). Hidden
consequence of active local Lorentz invariance. International Journal of Geometric
Methods in Modern Physics, 2(22), 305–357.
Rodrigues, W. A. Jr, and de Oliveira, E. C. (2007). The Many Faces of Maxwell,
Dirac and Einstein Equations: A Clifford Bundle Approach. Volume 722, Lecture
Notes in Physics. Springer, Berlin.
Rogers, A. (2007). Supermanifolds: Theory and Applications. World Scientific, Sin-
gapore.
Rotman, J. J. (2000). Advanced Modern Algebra. American Mathematical Society,
Providence, RI.
References 237
Schafer, R. D. (1954). On the algebras formed by the Cayley-Dickson process. Amer-

ican Journal of Mathematics, 76(2), 435–446.
Snygg, J. (1997). Clifford Algebra: A Computational Tool for Physicists. Oxford
University Press, Oxford.
Snygg, J. (2010). A New Approach to Differential Geometry Using Clifford’s Geo-
metric Algebra. Oxford University Press, Oxford.
Sohnius, M. F. (1985). Introducing supersymmetry. Physics Reports, 128(2–3), 39–
204.
Srednicki, M. (2007). Quantum Field Theory. Cambridge University Press, Cam-
bridge.
Srivastava, P. P. (1986). Supersymmetry, Superfields and Supergravity: An Introduc-
tion. Adam Hilger, Bristol.
Takahashi, Y. (1982). Reconstruction of a spinor via Fierz identities. Physical Review
D, 26(8), 2169–2171.
Todorov, I. (2011). Clifford algebras and spinors. Bulgarian Journal of
Physics, 38(1), 3–28.
Vahlen, K. T. (1902). Über Bewegungen und complexe Zahlen. Mathematische
Annalen, 55(1), 585–593.
van der Waerden, B. L. (1928). Spinoranalyse. Nachrichten von der Gesellschaft der
Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse, 1929(1), 100–
109.
Vaz, J. Jr. (1998). Construction of monopoles and instantons by using spinors and
the inversion theorem. In Clifford Algebras and Their Application in Math. Physics
(ed. V. Dietrich, K. Habetha, and G. Jank), Fundamental Theories of Physics 94,
pp. 401–421. Kluwer, Amsterdam.
Vaz, J. Jr. (2013). A Clifford algebra approach to the classical problem of a charge
in a magnetic monopole field. International Journal of Theoretical Physics, 52(5),
1440–1454.
Vaz Jr, J., and Rodrigues Jr., W. A. (1993). Equivalence of Dirac and Maxwell equa-
tions and quantum mechanics. International Journal of Theoretical Physics, 32(6),
945–959.
Veblen, O. (1933). Geometry of two-component spinors. Proceedings of the National
Academy of Sciences, 19(4), 462–474.
Villalobos, C. H., da Silva, J. M. H., and da Rocha, R. (2015). Questing mass
dimension 1 spinor fields. The European Physical Journal C , 75(6), 1–7.
Wess, J., and Bagger, J. (1992). Supersymmetry and Supergravity. Princeton Uni-
versity Press, Princeton.
West, P. (1990). Introduction to Supersymmetry and Supergravity (2nd edn). World
Scientific, Singapore.
Winitzki, S. (2010). Linear algebra via exterior products.
https://sites.google.com/site/winitzki/linalg, accessed 2 December, 2015.
Zwiebach, B. (2001). A First Course in String Theory. Cambridge University Press,
Cambridge.
Index
(p + q)-vector, 24 classification – complex case, 151

Z2 grading – even/odd, 18 classification — real case, 151
0-covectors, 23 Clifford algebra, 57
0-vector, 23 alternating tensor product, 88
1-covector, 23 classification
1-vector, 23, 30 complex case, 102
2-spinor, 214 real case, 101
2-vector, 23, 30 complex, 89
even subalgebra, 92
aggregate, 204 isomorphisms, 90
algebra periodicity theorem, 96
Clifford Clifford mapping, 64, 76
primitive idempotent, 111 Clifford–Lipschitz
matrix representation, 108 group, 126
structure – simple/semi-simple, 104 twisted group, 129
Dirac, 93 companion
Grassmann, 49 matrix, 178
non-alternative, 171 compatible
quotient, 34 orientation, 43
algebraic complex
semispinor, 165 conjugation, 116
spinor, 165 complex numbers, 94
algebraic spinor conformal compactification, 139
classification – real case, 149 conformal group, 140
algebraic spinors conformal sphere, 140
classification – complex case, 149 conjugation, 18, 30
alternator, 21 charge, 167
annihilation complex, 115
operator, 74 Hermitian, 115
automorphism, 126 contraction
left, 36
basis right, 38
change, 3 contravariant
geometrically equivalent, 182 tensor, 14
bidual, 5 transformation, 3, 15
bilateral correlation, 6
ideal, 33, 86 covariant
boomerang, 204 tensor, 14
braid group, 20 transformation, 3, 15
covector, 2
carrier space, 93 creation
Cartan equation, 179 operator, 74
Cayley–Klein cycle, 21
parameters, 145
centre diagram
Clifford algebra, 67 Coxeter–Dynkin, 178
charge conjugation, 168, 169 dilatation, 144
operator, 169 dimension
chiral spinor, 169 Clifford algebra, 66
classical spinor Dirac
240 Index
adjoint, 169 spinor, 165

algebra, 93 interior product, 36
matrices irreducible representation, 154
chiral representation, 120 graded, 154
Majorana representation, 120 isometry, 121, 180
standard representation, 112 isomorphism
Weyl representation, 120 Hodge, 50, 51, 70
spinor, 201 musical, 7
direct sum, 15 quasi-Hodge, 43
division algebra isotropic
unital, 106 vector, 7
dual
basis, 3 Kähler–Atiyah algebra, 85
spin space, 211 kernel, 6
Klein absolute, 139, 200
eigenspinor, 169
equation left contraction, 64, 81
Yang-Baxter, 20 left ideal, 103
equivalence class, 33 minimal, 103
even subalgebra Leibniz rule
Clifford algebra, 68 graded, 38
exponential mapping, 135 Lie
exterior algebra, 135
algebra, 21, 29, 34 conformal group, 143
product, 24, 64, 81 bracket, 135
Lorentz group, 146
Fierz
aggregate, 204 Möbius
identities, 204 group, 140
flagpole, 186, 187, 190, 217 transformation, 210
transformations, 142
graded orthochronous, 142
algebra, 16 Majorana
involution, 17, 30 spinor, 169, 186, 188
Grassmann mapping
algebra, 21, 48, 49 Clifford, 57
group spinor structure, 165
invertible elements – regular, 126 mass-shell condition, 222
orthogonal, 122 matrix
Pin, 132 spinor, 211
reduced, 133 maximal
Spin, 133 totally isotropic subspaces, 181
reduced, 133 metric
symmetric, 21 index, 170
Minkowski spacetime, 181
Hermitian conjugation, 115 momentum, 217
Hodge isomorphism, 50 multicovectors, 30
hyperbolic multivector, 30
pair, 180 structure, 66
plan, 179
neutral
ideal space, 180
left, 33 null
right, 33 direction, 209
idempotent, 103 vector, 209
primitive, 103, 111, 182 number
identities double/perplex/duplex/Lorentz, 94
Fierz, 205
inequivalent representation, 155 octonionic algebra, 176
inner product octonionic product, 177
Index 241
octonionic unit, 177 algebraic, 148

orientation, 43 simple
vector space, 42 p-vector, 26
orthogonal algebra, 104
group, 121 singular
symmetry, 124 spinor, 204
transformation, 121 space
vector, 7 isotropic, 179
neutral, 180
paramultivectors, 196 representation, 93, 103
paravector, 138, 196 totally isotropic, 179
path, 122 vector, 1
Penrose twistor, 201 special orthogonal
permutation, 21 group, 121
Plücker spin
coordinates, 55 basis, 212
pole, 217 space, 211
product spinor
Chevalley, 171, 172 matrix, 211
regressive, 52 adjoint, 168
projector, 15, 66 algebra, 154
pseudoscalar, 29 algebraic, 147
pure spinor, 182, 186 charge conjugate, 169
canonical form, 183 Dirac, 146
space, 182 dotted contravariant, 191
dotted covariant, 192
quadratic Majorana, 168
space, 7 operator, 152, 154
quasi-sphere, 140 Pauli, 146
quaternion, 1, 95 pure, 182
representative, 182
radical, 179 space, 181
null, 179 structure, 165
Radon–Hurwitz application, 165
numbers, 111 transformation, 210
rank, 6 undotted contravariant, 190
reference twistor, 201 undotted covariant, 191
reflection, 121, 124, 173 Weyl, 146
regressive spinors
product, 47 algebraic, 148
regular classical, 147, 149
spinor, 204 operator, 147
regular representation pure, 179
irreducible, 152 Weyl, 190, 193, 197
representation standard representation, 112
adjoint, 126 subspace
regular, 103 totally isotropic, 180
faithful, 103 supercharge, 223
graded, 152 swap automorphism, 178
irreducible, 103 symmetric group, 21
twisted adjoint, 128 symplectic
reversion, 18, 30 space, 7
Robinson congruence, 202 symplectic basis, 180
rotation, 121
tensor
scalar part, 66, 69 algebra, 17
Schouten identities, 220 product, 11
semispinor, 148 tensor product
semispinor space, 170 alternating, 87
semispinors tetrad, 199
242 Index
The Cartan–Dieudonné theorem, 125 universal

totally isotropic subspace Clifford algebra, 58, 65
maximal, 180
totally isotropic subspaces van der Waerden formalism, 191
maximal, 182 vector
trace, 19 bivector, 23
transformation isotropic, 179
Möbius, 138 product, 56
spinor, 194, 210 space, 1
translation, 144 trivector, 23
transposition, 21, 117 vector space
transvection, 142, 144 dual, 2
triality, 170, 172, 173
principle, 179 Weyl
triplet, 179 covariant dotted spinor, 201
triplets spinor, 169, 190, 214
triality, 179 Witt
twistor, 200, 202 basis, 180, 183
Keller approach, 200 decomposition, 179, 181
type (p, q) index, 180
tensor, 14 theorem, 183

An Introduction To Clifford Algebras and

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

An Introduction To Clifford Algebras and

Uploaded by

Copyright:

Available Formats

www.Ebook777.

Roldão da Rocha, Jr.

In 1878 William Kingdom Cliﬀord published an article in the American Journal

5 Clifford Algebras, and Associated Groups 121

1.1 Vectors and Covectors

α(au + v) = aα(u) + α(v),

and the multiplication by scalars

α(v) = α(vi ei ) = v i α(ei ) = v i αi ,

where αi = α(ei ). This expression reveals that a covector α is completely deﬁned by

2 A homomorphism is a mapping of a structure X into a structure Y such that the structural

Consider now the set of covectors {ei } (i = 1, . . . , n) deﬁned by

Covariant Transformations and Contravariant Transformations

The inverse transformations are given by

Example 1.2 Let F be the space of continuous functions f : R → R. The integral

The Kernel and the Image of a Linear Transformation

Correlations and Bilinear Functionals

B(v, u) = τ (v)(u), ∀ u, v ∈ V. (1.4)

Quadratic Spaces and Symplectic Spaces

g(v, u) = g(u, v), ∀ u, v ∈ V. (1.5)

v♭ (u) = g(v, u), g(α♯ , v) = α(v) . (1.10)

v♭i = gij v j . (1.11)

vi = gij v j , ei = gij ej , (1.16)

This correlation is non-degenerate, in view of ker ♭ = {0}. Indeed, if ♭(v) = 0, then 3v 1 + v 2 = 0;

Therefore, a symmetric bilinear form g can be defined as g(v, u) = v♭ (u). Since

it follows that g can be represented in this basis by the matrix

On the other hand, the inverse correlation ♯ is given by

is solved, having r1 = 2; r2 = 2; and r3 = 4 as solutions. The corresponding eigenvectors are

1.2 The Tensor Product

Lin(U, S) → Lin(2) (V, W ; S), f 7→ f ◦ φ .

It is straightforward to apply this deﬁnition to vector spaces of ﬁnite dimensions.

Theorem 1.1 ◮ If {e1 , . . . , en } is a basis of V and {e′1 , . . . , e′p } is a basis of W , the

Lin(U, S) → S × S × · · · × S = S m , f 7→ {f (ǫ1 ), . . . , f (ǫm )} .

Now, let us consider this sequence of two linear mappings:

Lin(U, S) → Lin(2) (V, W ; S) → S np ,

In particular, given a bilinear mapping φ : V ×W → U , there exists a unique linear

Comparing these equations, we can write

where aij = A(ei , ej ) are components of A in this basis.

{ei ⊗ ej + ej ⊗ ei } and {ei ⊗ ej − ej ⊗ ei }

are, respectively, bases for T2sym (V ) and T2alt (V ).

Mixed Bilinear Functionals

Tensors of Type (p, q)

Direct Sum of Vector Spaces

Let now P be a linear operator in V . This operator is said to be a projection

1.3 Tensor Algebra

that xk yl ∈ Ak+l . The elements of Ak are said to be homogeneous of degree k. In

The Tensor Algebra

#(Tp ) = (−1)deg Tp Tp = (−1)p Tp , (1.32)

formulæ involving the action of this mapping on an element. Nevertheless, when we

there is a linear bijection from Lin(V1 ⊗ V2 , W1 ⊗ W2 ) onto Lin(2) (V1 , V2 ; W1 ⊗ W2 ).

when the basis {ei ⊗ fj } of V ⊗ W is changed to the basis {fj ⊗ ei } of W ⊗ V .

(9) Let A = [aij ] be a matrix associated with a linear operator A : V → V in the

2.1 Permutations and the Alternator

(α1 ⊗ α2 ⊗ · · · ⊗ αp )(v1 , v2 , . . . , vp ) = α1 (v1 )α2 (v2 ) · · · αp (vp ).

2.2 p-Vectors and p-Covectors

Similarly, a p-covector is an alternating covariant tensor of order p. Denoted by

Ψ[p] = Alt(Ψ[p] ). (2.4)

Bp = {ei1 ⊗ ei2 ⊗ · · · ⊗ eip },

where the indices ik (k = 1, 2, . . V

Meanwhile, not all the n(n − 1) · · · (n − p + 1) elements of Bp present distinct

2.3 The Exterior Product

A[p] ∧ B[q] = Alt(A[p] ⊗ B[q] ) . (2.8)

(A[p] ∧ B[q] ) ∧ C[r] = A[p] ∧ (B[q] ∧ C[r] ); (2.9)

this associativity is an inherited property, elicited from the associativity