Clifford Algebra and Spinors

London Mathematical Society
Lecture Note Series 286
Clifford Algebras
and Spinors
Second Edition
LBNPOF4 MATHEMATICAL SOCIETY
LECTURE NOTE SERIES
London Mathematical Society Lecture Note Series. 286
Clifford Algebras and Spinors

Second Edition
Pertti Lounesto
Helsinki Polytechnic Stadia
CAMBRIDGE
UNIVERSITY PRESS
PUBLISHED BY THE PRESS SYNDICATE OF THE UNIVERSITY OF CAMBRIDGE
The Pitt Building, Trumpington Street, Cambridge, United Kingdom
CAMBRIDGE UNIVERSITY PRESS
The Edinburgh Building, Cambridge, CB2 2RU, UK www.cup.cam.ac.uk
40 West 20th Street, New York, NY 10011-4211, USA www.cup.org
10 Stamford Road, Oakleigh, Melbourne 3 166, Australia
Ruiz de AlarcQ 13,28014 Madrid, Spain
O P. Lounesto 2001
This book is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without
the written permission of Cambridge University Press.
First published 1997
Reprinted 1998,1999
Second edition 2001
Printed in the United Kingdom at the University Press, Cambridge
A catalogue record for this book is available from the British Library
ISBN 0 521 0055 1 5 paperback

Contents
Preface vii
Mathematical Notation ix
1. Vectors and linear spaces 1
2. Complex numbers 18
3. Bivectors and the exterior algebra 33
4. Pauli spin matrices and spinors 50
5. Quaternions 67
6. The fourth dimension 80
7. The cross product 92
8. Electromagnetism 100
9. Lorentz transformations 118
10. The Dirac equation 135
11. Fierz identities and boomerangs 152
12. Flags, poles and dipoles 162
13. Tilt to the opposite metric 174
14. Definitions of the Clifford algebra 188
15. Witt rings and Brauer groups 195
16. Matrix representations and periodicity of 8 205
17. Spin groups and spinor spaces 219
18. Scalar products of spinors and the chessboard 23 1
19. Mobius transformations and Vahlen matrices 244
20. Hypercomplex analysis 255
21. Binary index sets and Walsh functions 279
22. Chevalley's construction and characteristic 2 288
23. Octonions and triality 300
A history of Clifford algebras 320
Selected reading 33 1
Index 335
Preface
This book is intended for people who are not primarily algebraists, but nonethe-
less get involved in the subject through other areas - having backgrounds, in
say, engineering, physics, geometry or analysis. Readers of this book may have
different starting levels, backgrounds and goals.
Chapters 1-2 form a unit for an undergraduate course: Vectors, the scalar
product; complex numbers, a geometrical interpretation of the imagi-
nary unit i = a.
Chapters 3-7 guide the reader through bottlenecks and provide necessary
building blocks: Bivectors and the exterior product. Pauli spin ma-
trices and Pauli spinors. Quaternions and the fourth dimension. The
cross product is generalized to higher dimensions.
Chapters 8-10 aim to serve readers with different backgrounds: Electromag-
netism, special relativity, Dirac theory. The Dirac equation is formu-
lated with complex column spinors, spinors in minimal left ideals and
considering spinors as operators.
Chapters 11-13 discuss physical applications of spinors: In the case of an
electron, Fierz identities are sufficient to reconstruct spinors from their
physical observables, but this is not the case for the neutrino. Boome-
rangs are introduced to handle neutrinos. A new class of spinors is
identified by its bilinear observables: the flag-dipole spinors which re-
side between Weyl, Majorana and Dirac spinors.
Chapters 14-15 are more algebraic than the previous chapters. Clifford al-
gebras are defined for the first time. Finite fields. Isometry classes of
quadratic forms and their Witt rings. Tensor products of algebras and
Brauer groups are discussed.
Chapters 16-19 view Clifford algebras through matrix algebras: Clifford al-
gebras are given isomorphic images as matrix algebras, Cartan's peri-
odicity of 8, spin groups and their matrix images in lower dimensions,
scalar products of spinors with a chessboard of their automorphism
groups, Mobius transformations represented by Vahlen matrices.
Chapters 20-23 discuss miscellaneous mathematical topics. A one-variable
higher-dimensional generalization of complex analysis: Cauchy's inte-
gral formula is generalized to higher dimensions. Multiplication rule
of standard basis elements of a Clifford algebra and its relation to
Walsh functions. Multivector structure of Clifford algebras and the
Preface
>
non-existence of k-vectors, k 2, in characteristic 2. In the last chap-
ter, we come into contact with final frontiers science: an exceptional
phenomenon in dimension 8, triality, which has no counterpart in any
other dimension.
The first parts of initial chapters are accessible without knowledge of other
parts of the book - thus a teacher may choose his own path for his lectures
on Clifford algebras. The latter parts of the chapters are sometimes more
advanced, and can be left as independent study for interested students.
Introduction of the Clifford algebra of multivectors and spinors can be mo-
tivated in two different ways, in physics and in geometry:
(i) In physics, the concept of Clifford algebra, as such or in a disguise, is
a necessity in the description of electron spin: Spinors cannot be con-
structed by tensorial methods, in terms of exterior powers of the vector
space.
(ii) In geometry, information about orientation of subspaces can be encoded
in simple multivectors, which can be added and multiplied. Physicists
are familiar with this tool in the special case of one-dimensional sub-
spaces of oriented line-segments, which they manipulate by vectors (not
by projection operators, which lose information about orientation).
Acknowledgements
In preparing this book I have enjoyed the help of many friends and colleagues:
Jerry Segercrantz from Finland, Peter Morgan, Ian Porteous and Ronald Shaw
from England, Rafal Ablamowicz, Michael Becker, Geoffrey Dixon, Tevian
Dray, Michael Kinyon and Arvind Raja from the USA, Roland Bacher (Switzer-
land), Helga Baum (Germany), Bernard Jancewicz (Poland), Alphonse Charlier
and Yvon Siret (France), Josep Parra (Spain) and Garret Sobczyk (Mexico).
I thank Ross Jones and Goran Lgngstedt for revising my English. For their
assistance with TEX, I am indebted to Timo Korvola and Martti Nikunen.
I appreciate feedback from the readers of the first edition: Leo Dorst, Kynn
Jones, Joseph Riel and Perttu Puska.
In particular, I would like to thank Jacques Helrnstetter for his comments on
my DEA lectures in Grenoble as well as Ismo Lindell at the Electromagnetics
Laboratory in Helsinki, and Waldyr Rodrigues at IMECC in Campinas. For
financial support, I am indebted to the NOKIA Foundation in Helsinki and
CNPq in Brazil.
Helsinki, November 2000 PERTTI

LOUNESTO
Mat hematical Not ation
v linear space
IF field
D division ring (typically R, C, W)
W division ring of quaternions
0 division algebra of octonions
Mat(d, IF) d x d matrix algebra over IF
Mat(d,D) matrix algebra over Cen(D) with entries in D
21F double ring F x IF of the field IF
2Mat(d,IF) direct sum Mat(d,IF) ;) Mat(d,IF) Mat(d, 21F)
A an algebra
Cen(A) the center of an algebra A
A R, C, W, 2R, or 2W
Rn n-dimensional real linear space
Rn = Rn,o n-dimensional Euclidean space
P n-dimensional complex linear space
Wn n-dimensional module over W
RP9P real quadratic space (p for positive and q for negative)
C ~# P
P Clifford algebra of Rplq
Q quadratic form
C4Q) Clifford algebra of the quadratic form Q; x 2 = Q(x)
Ces cz Mat(2, C) Clifford algebra of the Euclidean space R3
-
ce, = ce, ,O
--
Clifford algebra of the Euclidean space Rn = RntO
CelS3 Mat(2, W) Clifford algebra of the Minkowski space R1t3
Ces,~ Mat(4,R) Clifford algebra of the Minkowski space R3t1
u
-
u
grade involute of u E Ce(Q) = Ce+(Q) $ Ce-(Q)
reverse of u E CC(Q); 5 = x for x E V
-
u Clifford-conjugate of u E CC(Q); ji = -x for x E V
u* complex conjugate of u (in chapter 2 also G )
f primitive idempotent of C%,q
s = cep,,f spinor space (minimal left ideal of CeP,,)
s, u s
left ideal S $ or companion of U E SO(8)
Zn {0,1,. . . , n - 1) or (1, ei2=/",. . . , ei2a(n-1)/n)
1
Vectors and Linear Spaces
Vectors provide a mathematical formulation for the notion of direction, thus

making direction a part of our mathematical language for describing the physi-
cal world. This leads to useful applications in physics and engineering, notably
in connection with forces, velocities of motion, and electrical fields. Vectors
help us to visualize physical quantities by providing a geometrical interpreta-
tion. They also simplify computations by bringing algebra to bear on geometry.
1.1 Scalars a n d vectors

In geometry and physics and their engineering applications we use two kinds
of quantities, scalars and vectors. A scalar is a quantity that is deterrnined
by its magnitude, measured in units on a suitable scale. For instance, mass,
temperature and voltage are scalars.
A vector is a quantity that is determined by its direction as well as its mag-
nitude; thus it is a directed quantity or a directed line-segment. For instance,
force, velocity and magnetic intensity are vectors.
We denote vectors by boldface letters a , b, r , etc. [or indicate them by arrows,
a', g, Z, etc., especially in dimension 31. A vector can be depicted by an arrow,
a line-segment with a distinguished end point. The two end points are called
the initial point (tail) and the terminal point (tip):
1. length (of the line-segment OA)
JIA
2. direction
- attitude (of the line OA)
- orientation (from O to A)
0
The length of a vector a is denoted by la[. Two vectors are equal if and only
1 In this chapter scalars are real numbers (elements of R).
2 Vectors and Linear Spaces
if they have the same length and the same direction. Thus,
a=b w lal=lbl and a t t b .
Two vectors have the same direction, if they are parallel as lines (the same
attitude) and similarly aimed (the same orientation). The zero vector has
length zero, and its direction is unspecified. A unit vector u has length one,
IuI = 1. A vector a and its opposite -a are of equal length and parallel, but
have opposite orientations.
1.2 Vector a d d i t i o n a n d subtraction

Given two vectors a and b , translate the initial point of b to the terminal
point of a (without rotating b). Then the sum a + b is a vector drawn from the
initial point of a to the terminal point of b. Vector addition can be visualized
by the triangle formed by vectors a, b and a b. +
b b
Vector addition b
Vector addition is commutative, a + b = b +a, as can be seen by inspection of

+ +
the parallelogram with a and b as sides. It is also associative, (a b ) c =
a+ ( b + c ) , and such that two opposite vectors cancel each other, a + (-a) = 0.
Instead of a + (-b) we simply write the difference as a - b. Note the order
t t - + -+ -+
in BA = O A - O B when a = OA and b = OB.
Vector subtraction 0 b B
R e m a r k . To qualify as vectors, quantities must have more than just direction
1.3 Multiplication by numbers (scalars) 3
and magnitude - they must also satisfy certain rules of combination. For in-
stance, a rotation can be characterized by a direction a, the axis of rotation,
and a magnitude cr = lal, the angle of rotation, but rotations are not vectors
because their composition fails to satisfy the commutative rule of vector addi-
tion, a + b = b + a . The lack of commutativity of the composition of rotations
can be verified by turning a box around two of its horizontal axes by 90° :
The terminal attitude of the box depends on the order of operations. The axis
of the composite rotation is not even horizontal, so that neither a+ b nor b a +
can represent the composite rotation. We conclude that rotation angles are not
vectors - they are a different kind of directed quantities. I
1.3 Multiplication b y n u m b e r s (scalars)

Instead of a + a we write 2a, etc., and agree that (-l)a = -a, the opposite of
a. This suggests the following definition for multiplication of vectors a by real
numbers X E R : the vector Xa has length lXal = lXllal and direction given by
(for a # 0)
Xatta if A > 0,
XaTJa if X < 0.
Numbers multiplying vectors are called scalars. Multiplication by scalars, or
+ + +
scalar multiplication, satisfies distributivity, X(a b ) = Xa Xb, (A p)a =
Xa + pa, associativity, (Xp)a = X(pa), and the unit property, la = a, for all
real numbers A, p and vectors a, b .
1.4 Bases and coordinates

In the plane any two non-parallel vectors el, e 2 form a basis so that an arbitrary
vector in the plane can be uniquely expressed as a linear combination a =
+
ale1 a2e2. The numbers a l , a2 are called coordinates or components of the
vector a with respect to the basis {el, en).
When a basis has been chosen, vectors can be expressed in terms of the
coordinates alone, for instance,
If we single out a distinguished point, the origin 0, we can use vectors to label
t
the points A by a = OA. In the coordinate system fixed by 0 and {el,e2)
we can denote points and vectors in a similar manner,
point A = (al, a%), vector a = (01, a2),
since all the vectors have a common initial point 0 .
In coordinate form vector addition and multiplication by scalars are just
coordinate-wise operations:
Conversely, we may start from the set R x R = {(z, y) I t ,y E R ) , and

equip it with component-wise addition and multiplication by scalars. This
construction introduces a real linear structure on the set R2 = R x R making
it a 2-dimensional real linear space R2. The real linear structure allows us to
view the set R 2 intuitively as a plane, the vector plane R2. The two unit points
on the axes give the standard basis
of the 2-dimensional linear space R2.

In our ordinary space a basis is formed by three non-zero vectors el, e2, e3
which are not in the same plane. An arbitrary vector a can be uniquely
represented as a linear combination of the basis vectors:
The numbers a l , a2, a3 are coordinates in the basis {el,e 2 , es). Conversely,
coordinate-wise addition and scalar multiplication make the set
R x R x R = { ( t , y,z) 1 t , y , z E R )
a 3-dimensional real linear space or vector space R3. In a coordinate system
fixed by the origin 0 and a standard basis {el,e2,e3) a point P = ( t , y, z )
and its position vector
t
O P = tel + ye2 + ze3
have the same coordinates.
2 Some authors speak about components of vectors and coordinates of points.
3 Since a vector beginning at the origin is completely determined by its endpoints, we will
sometimes refer to the point r rather than to the endpoint of the vector r.
1.5 Linear spaces and linear functions 5
1.5 Linear spaces a n d linear functions
Above we introduced vectors by visualizing them without specifying the grounds
of our study. In an axiomatic approach, one starts with a set whose elements
satisfy certain characteristic rules. Vectors then hecome elements of a math-
ematical object called a linear space or a vector space V. In a linear space
vectors can be added to each other but not multiplied by each other. Instead,
vectors are multiplied by numbers, in this context called scalars.
Formally, we begin with a set V and the field of real numbers R. We associate
with each pair of elements a , b E V a unique element in V, called the sum and
denoted by a + b , and to each a E V and each real number X E R we associate
a unique element in V, called the scalar multiple and denoted by Xa. The set
V is called a linear space V over R if the usual rules of addition are satisfied
for all a, b , c E V
a+b=b+a commutativity
+ +
(a b) c = a + (b + c) associativity
a+O=a zero-vector 0
a+(-a) = O opposite vector -a
and if the scalar multiplication satisfies
+
X(a b) = Xa Xb + distributivity
(A + p ) a = Xa+,ua
( X P )=~ X(P~> associativity
la =a unit property
for all A, ,u E IW and a , b E V. The elements of V are called vectors, and the
linear space V is also called a vector space. The above axioms of a linear space
set up a real linear structure on V.
A subset U of a linear space V is called a linear subspace of V if it is closed
under the operations of a linear space:
a+bEU for a , b E U ,
Xa E U for X E R, a E U.
For instance, R2 is a subspace of R3.
A function L : U + V between two linear spaces U and V is said to be
linear if for any a , b E U and X E R,
L ( a + b ) = L(a) + L(b) and
L(Xa) = X L(a).
4 Vectors are not scalars, and scalars are not vectors. Vectors belong to a linear space V ,
and scalars belong to a field IF. In this chapter IF = R.
Linear functions preserve the linear structure. A linear function V + V is
called a linear transformation or an endomorphism. An invertible linear func-
tion U + V is a linear isomorphism, denoted by U 2: V.
The set of linear functions U +V is itself a linear space. A composition
of linear functions is also a linear function. The set of linear transformations
V + V is a ring denoted by End(V). Since the endomorphism ring End(V)
is also a linear space over R , it is an associative algebra over R, denoted by
Endne(V).
1.6 Linear independence; dimension

A vector b E V is said to be a linear combination of vectors al, a 2 , . . . ,a k if
it can be written as a sum of multiples of the vectors al, aa, . . . , ak, that is,
b = Xlal+ X2a2 + ...+ Xkak where XI, X2,.. . , Xk E R.
A set of vectors {al, a2,. . . ,ak)is said to be linearly independent if none of the
vectors can be written as a linear combination of the other vectors. In other
words, a set of vectors {al,a 2 , . . . ,ak) is linearly independent if X1 = X2 =
. . . = Xk = 0 is the only set of real numbers satisfying
In a linear combination
of linearly independent vectors al, a2,. . .,a k the numbers X I , X2,. . . , Xk are
unique; we call them the coordinates of b.
Linear combinations of {al,a2, . . . ,ak) c V form a subspace of V; we say
that this subspace is spanned by {al, an, . . . ,ak). A linearly independent set
{al, a2,. . . ,ak) C V which spans V is said to be a basis of V. All the bases
for V have the same number of elements called the dimension of V.
QUADRATIC
STRUCTURES
Concepts such as distance or angle are not inherent in the concept of a linear
structure alone. For instance, it is meaningless to say that two lines in the
linear space R2 meet each other at right angles, or that there is a basis of
5 Finite-dimensional real linear spaces are isomorphic if they are of the same dimension.
6 A ring R is a set with the usual addition and an associative multiplication R X R + R
which is distributive with respect to the addition. An algebra A is a linear space with a
bilinear product A x A + A.
1.7 Scalar product 7
equally long vectors el, e:, in R2. The linear structure allows comparison of
lengths of parallel vectors, but it does not enable comparison of lengths of non-
parallel vectors. For this, an extra structure is needed, namely the metric or
quadratic structure.
The quadratic structure on a linear space Rn brings along an algebra which
makes it possible to calculate with geometric objects. In the rest of this chapter
we shall study such a geometric algebra associated with the Euclidean plane
R2.
1.7 Scalar p r o d u c t
We will associate with two vectors a real number, the scalar product a . b E R of
+
a, b E R2. This scalar valued product of a = ale1 a2e2 and b = blel b2e2 +
is defined as
i n coordinates a - b = albl + a2b2
geometrically a - b = lallblcoscp
where cp [0 5 cp 5 180°] is the angle between a and b. The geometrical
construction depends on the prior introduction of lengths and angles. Instead,
the coordinate approach can be used to define the length
which equals la1 = d m , and the angle given by

a.b
cos cp = -
lal lb l
Two vectors a and b are said to be orthogonal, if a . b = 0. A vector of
length one, la1 = 1, is called a unit vector. For instance, the standard basis
vectors el = (1, O), en = ( 0 , l ) are orthogonal unit vectors, and so form an
ort1~onormalbasis for R2.
The scalar product can be characterized by its properties:
(a+b).c=a.c+b-c
linear in the first factor
(Xa) - b = X(a. b)
a . b =b . a symmetric
a . a > O for a f O positive definite.
Symmetry and linearity with respect to the first factor together imply bilin-
earity, that is, linearity with respect to both factors. The real linear space R2
endowed with a bilinear, symmetric and positive definite product is called a
Euclidean plane R2.
All Euclidean planes are isometric to the one with the metriclnorm
r = xe1+ ye2 + Irl= Jm.
In the rest of this chapter we assume this metric structure on our vector plane
R2.
+
Remark. The quadratic form r = xel ye2 -+ lrI2 = x2 y2 enables us to +
compare lengths of non-parallel line-segments. The linear structure by itself
allows only comparison of parallel line-segments. I
1.8 The Clifford product of vectors; the bivector

It would be useful to have a multiplication of vectors satisfying the same axioms
as the multiplication of real numbers - distributivity, associativity and commu-
tativity - and require that the norm is preserved in multiplication, lab( = JaJJbJ
Since this is impossible in dimensions n >- 3, we will settle for distributivity and
associativity, but drop commutativity. However, we will attach a geometrical
meaning to the lack of commutativity.
Take two orthogonal unit vectors el and e2 in the vector plane R2. The
+
length of the vector r = xel ye2 is Irl = d m . If the vector r is
multiplied by itself, rr = r2, a natural choice is to require that the product
equals the square of the length of r,
1,' = (r(2 .
In coordinate form, we introduce a product for vectors in such a way that
(xel+ ye2)' = x2 + Y2.
Use the distributive rule without assuming commutativity to obtain
This is satisfied if the orthogonal unit vectors el, e2 obey the multiplication
rules
which correspond to
Use associativity to calculate the square (ele2)' = -efe: = -1. Since the
square of the product ele2 is negative, it follows that ele2 is neither a scalar
7 An isometry of quadratic forms is a linear function f : V + V' such that Q'( f (a)) = Q(a)
for all a E V.
8 The scalar product a . b is not the same as the Clifford product a b . Instead, the two
+
products are related by a . b = $ ( a b ba).
1.9 The Clifford algebra Ce2 9
nor a vector. The product is a new kind of unit, called a bivector, represent-
ing the oriented plane area of the square with sides el and e2. Write for short
el2 = ele2.
+
We define the Cliffordproduct of two vectors a = ale1 a2e2 and b = blel +
b2e2 to be ab = albl +a2b2+(alb2-a2bl)e12, asum of a scalar and a bivector.
1.9 The Clifford algebra Ce2
The four elements

1 scalar
el, e2 vectors
el2 bivector
form a basis for the Clifford algebra Ce2 of the vector plane R2, that is,
an arbitrary element
is a linear combination of a scalar uo, a vector ulel + u2e2 and a bivector

10
'1112'312.
2
Example. Compute elel2 = elele2 = e2, enel = ele2el = -e1e2 = -e2,
e2e12 = egele2 = -elei = -el and el2e2 = elei = el. Note in particular
that el2 anticommutes with both el and e2. I
The Clifford algebra Cf2 is a 4-dimensional real linear space with basis elements
9 These algebras were invented by William Kingdon Clifford (1845-1879). T h e first an-
nouncement of the result was issued in a talk in 1876, which was published posthumously
in 1882. T h e first publication of the invention came out in another paper in 1878.
10 T h e Clifford algebra Cf!, of Rn contains 0-vectors (or scalars), 1-vectors (or just vec-
tors), 2-vectors, ..., n-vectors. The aggregates of k-vectors give the linear space Cen a
.
multivector structure Cen = R $ Rn $ A2 Rn $ . . $ An Rn.
1, el, e2,el2 which have the multiplication table
1.10 Exterior product = bivector part of the Clifford product

Extracting the scalar and bivector parts of the Clifford product we have as
+
products of two vectors a = ale1 a2e2 and b = blel b2e2 +
a . b = albl + a2b2, the scalar product 'a dot b',
a A b = (alb2- a2bl)e12, the exterior product 'a wedge b'
The bivector a A b represents the oriented plane segment of the parallelogram
with sides a and b. The area of this parallelogram is lalb2 - azbll, and we will
take the magnitude of the bivector a~ b to be this area laA bl = lalb2- a2blI.
Area = lalb2- a2bl1
The parallelogram can be regarded as a kind of geometrical product of its

sides:
The bivectors a A b and b A a have the same magnitude but opposite senses
of rotation. This can be expressed simply by writing
1.11 Components of a vector in given directions 11
Using the multiplication table of the Clifford algebra C12 we notice that the
Clifford product
+ +
of two vectors a = ale1 a2e2 and b = blel b2e2 is a sum of a scalar
+
a . b = al bl a2b2and a bivector a A b = (alb2- a2bl)e12.l1 In an equation,
The commutative rule a . b = b . a together with the anticommutative rule

a A b = -b A a implies a relation between ab and ba. Thus,
Adding and subtracting equations (a) and (b), we find

1 1
a.b=-(ab+ba) and a A b = - ( a b - b a ) .
2 2
Two vectors a and b are parallel, a 11 b, when they commute, ab = ba, that
is, a A b = 0 or al b2 = a2bl, and orthogonal, a l b , when they anticommute,
ab = -ba, that is, a . b = 0. Thus,
ab=-ba 9 a l b a.b=O 9 ab=aAb.
1.11 Components of a vector in given directions

Consider decomposing a vector r into two components, one parallel to a and
the other parallel to b, where a b. This means determining the coefficients
a and 8 in the decomposition r = aa+/?b. The coefficient cr may be obtained
+
by forming the exterior product r A b = (cra pb) A b and using b A b = 0;
this results in r A b = cr(a A b). Similarly, a A r = p(a A b). In the last two
equations both sides are multiples of el2 and we may write, symbolically, l 2
11 The bivector valued exterior product a /\ b = (alb2 - a 2 b l ) e 1 2 ,which represents a plane

area, should not be confused with the vector valued cross product a x b = (a1b2 - a2 b l ) e s ,
which represents a line segment.
12 As a n element of the exterior algebra /\R2 the bivector a b is not invertible. As a n
element of the Clifford algebra Ce2 a non-zero bivector a A b is invertible, but since t h e
multiplication in Ce2 is non-commutative, it is more appropriate to write
a = ( r ~ b ) ( a ~ b ) and
-' P= (a~r)(a~b)-I.
However, since r A b , a A r and a A b commute, our notation is also acceptable.
The coefficients a and P could be obtained visually by comparing the oriented
areas (instead of lengths) in the following figure:
a
Exercise 5
1.12 Perpendicular projections a n d reflections

Let us calculate the component of a in the direction of b when the two vectors
diverge by an angle p , 0 < p < 180'. The parallel component all is a scalar
multiple of the unit vector b/lbl:
In other words, the parallel component all is the scalar product a . b =

lallblcosp multiplied by the vector b-' = b/lb12, called the inverse l3 of
the vector b . Thus,
I
The last formula tells us that the length of b is irrelevant when projecting into
the direction of b.
The perpendicular component a1 is given by the difference
a1 = a - all = a - (a.b)b-'
= (ab - a - b)b-' = (a A b)b-'
13 The inverse b-' of a non-zero vector b E R2 C C& satisfies b-'b = bb-' = 1 in the
Clifford algebra C e z . A vector and its inverse are parallel vectors.
1.12 Perpendicular projections and reflections 13
Note that the bivector el2 anticommutes with all the vectors in the ele2-plane,
therefore
The area of the parallelogram with sides a, b is seen to be

la1bl = la A bl = lal]b]sin cp
where 0 < cp < 180'.
+
The reflection of r across the line a is obtained by sending r = rll r l to
r' = r11- r l , where rll = (r . a)a-'. The mirror image r' of r with respect to
a is then
=(r.a-r~a)a-l
= ( a - r+aAr)a-'
= ara-'
and further
r' = ( 2 a . r - ra)a-'
r'
a.r
= 2-a -r.
a2
The formula r' = ara-' can be obtained directly using only commutation
+
properties of the Clifford product: decompose r = rll r l , where arlla-' =
rllaa-' = rll, while aria-' = -rlaa-l = -rl.
The composition of two reflections, first across a and then across b, is given
by
The composite of these two reflections is a rotation by twice the angle between
a and b . As a consequence, if a triangle ABC with angles a,@,-y is turned
about its vertices A, B, C by the angles 2a, 2P, 27 in the same direction, the
result is an identity rotation.
Exercises 6,7
1.13 M a t r i x representation of Ce2

In this chapter we have introduced the Clifford algebra Ce2 of the Euclidean
plane R2. The Clifford algebra Ct2 is a 4-dimensional algebra over the reals
R. It is isomorphic, as an associative algebra, to the matrix algebra of real
2 x 2-matrices Mat(2,R), as can be seen by the correspondences
However, in the Clifford algebra Ce2 there is more structure than in the matrix
algebra Mat(2, R). In the Clifford algebra Ct2 we have singled out by definition
a privileged subspace, namely the subspace of vectors or 1-vectors R 2 C Ce2.
No similar privileged subspace is incorporated in the definition of the matrix
algebra Mat(2, R). l4
For arbitrary elements the above correspondences mean that
and
1
-((a+d)
2
+ (a- d ) e l + (b+c)e2+ ( b - c)e121 2 (: i) a
In this representation the transpose of a matrix,
corresponds to the reverse
14 For instance, we might choose ul = f i e l+elz, uz = ez. This also results in the
commutation relations u: = 1, ul = 1, uluz +U ~ U I= 0, which define a different
representation of Ct2 as Mat(2, R).
1.13 Matrix representation of Cl2 15
of u = uo + u l e l + u2e2 + ul2el2 in C12. The complementary (or adjoint)
matrix
( ,b) [=(ad-bc)( ) for a d b c # 0 11

corresponds to the Clifford-conjugate l5
The reversion and Clifford-conjugation are anti-involutions, that is, involutory

anti-automorphisms,
-
fi = u, CiG = 62,
- -
u = u , uv=vii.
We still need the grade involute
for which ii = ii- = ii-
Exercises
+ +
1. Let a = e2 - e l 2 , b = el e 2 , c = 1 e2. Compute ab, ac. What did you
learn by completing this computation?
+
2. Let a = en e12, b = :(I+ e l ) . Compute ab, ba. What did you learn?
+
3. Let a = 1 el, b = -1 + e l , c = el + e 2 . Compute ab, ba, ac, ca, bc and
cb. What did you learn?
+
4. Let a = :(1 + el), b = el e l 2 . Compute a 2 , b2.
+
5. Let a = el - 2e2, b = el e 2 , r = 5el - e 2 . Compute cr,P in the
+
decomposition r = cra p b .
+
6. Let a = 8el - e2, b = 2el e2. Compute all, a l .
+
7. Let r = 4el - 3e2, a = 3el - en, b = 2el e2. Reflect first r across a
and then the result across b.
8. Show that for any u E C12, uii = iiu E R, and that u is invertible, if
uii # 0, with inverse
-
U
u-l --
uii
9. Let u = 1 + el + el2. Compute u-l. Show that
u-I = ii(uij)-l # (uii)-lii, u-I = (iiu)-lii # ij(iiu)-I and
u-I = ii(uii)-I # (uii)-lii, u-I = (iiu)-lii # ii(iiu)-l.
+
15 In some countries a vector u = ulel u2e2 E R2 is denoted by ii in handwriting, but
this practice clashes with our notation for the Clifford-conjugate.
10. Consider the four anti-involutions of Mat(2, R) sending
Define two anti-automorphisms a,p to be similar, if there is an

intertwining automorphism y such that a y = yo. Determine which ones of
these four anti-involutions are similar or dissimilar to each other. Hint:
keep track of what happens to the matrices
with squares I, I, and -I.

R e m a r k . In completing the exercises, note that an arbitrary element of C12
is most easily perceived when written in the order of increasing indices as
+ +
uo u l e l + u2e2 u l ~ e l 2 . I
Solutions
+ +
1. ab = ac = 1 - el e 2 - el2; one can learn that ab = ac b = c.
+
2. ab = 0, ba = e 2 + el2; one can learn that ab = 0 ba = 0 (and also that
ba = a + b = 1).
+ + + + +
3. ab = ba = 0, ac = 1 el ea elz, ca = 1 el e 2 - e12,
+
bc = 1 - el - e 2 el2, cb = 1 - el - e2 - el2; one can learn that
+
ab = ba = 0 ac = 0 or ca = 0.
4. a2 = a , b2 = 0.
5. r = 2 a + 3 b .
6. all = 6 e l + 3e2, a1 = 2el - 4e2.
7. r' = ara-' = sell rN= br'b-I = 3el 4e2. +
8. uii = iiu = u i - uf - u i+ E R.
+
9. u-' = 1 - el - el2 and (uQ)-lG = G(Gu)-' = 1 3el - 4e2 - 5e12 and
+ +
C(uu)-l = (uG)-'G = 1 3el 4e2 - 5e12.
10. Only two of the anti-involutions are similar,
as can be seen by choosing the intertwining automorphism
for which a y = yp.

1.13 Matrix representation of C t 2 17
Bibliography
W. K. Clifford: Applications of Grassmann's extensive algebra. Amer. J . Math. 1
(1878), 350-358.
W. K. Clifford: On the classification of geometric algebras; pp. 397-401 in R. Tucker
(ed.): Mathematical Papers by William Ii'ingdon Cliflord, Macmillan, London, 1882.
(Reprinted by Chelsea, New York, 1968.) Title of talk announced already in Proc.
London Math. Soc. 7 (1876), p. 135.
M.J.Crowe: A History of Vector Analysis. University of Notre Dame Press, 1967.
Reprinted by Dover, New York, 1985.
D.C. Lay: Linear Algebra and its Applications. Instructor's Edition. Addison-Wesley,
Reading, MA, 1994.
M.Riesz: Cliflord Numbers and Spinors. The Institute for Fluid Dynamics and Ap-
plied Mathematics, Lecture Series No. 38, University of Maryland, 1958. Reprinted
as facsimile (eds.: E.F. Bolinder, P. Lounesto) by Kluwer, Dordrecht, The Nether-
lands, 1993.
G. Strang: Introduction to Linear Algebra. Wellesley-Cambridge Press, Cambridge,
MA. 1993.
2
Complex Numbers
The feature distinguishing the complex numbers from the real numbers is that
the complex numbers contain a square root of -1 called the imaginary unit
'
i = G. Complex numbers are of the form
where x, y E IR and i satisfies i2 = -1. The real numbers x, y are called the
real part x = Re(%)and the imaginary part y = Im(z). To each ordered pair
of real numbers x, y there corresponds a unique complex number x i y . +
A complex number x + iy can be represented graphically as a point with
rectangular coordinates (x, y). The xy-plane, where the complex numbers are
represented, is called the complex plane @. Its x-axis is the real axis and y-axis
the imaginary axis.
A complex number t = x + iy has an opposite - z = -x - iy and a complex
conjugate 2 = x - iy, obtained by changing the sign of the imaginary part.
Im I z = x + yi
1 Electrical engineers denote the square root of -1 by j = G.

2 In quantum mechanics the complex conjugate is denoted by t* = x - iy.
Complex Numbers 19
The sum of two complex numbers is computed by adding separately the real
parts and the imaginary parts:
Addition of complex numbers can be illustrated by the parallelogram law of

vector addition.
The product of two complex numbers is usually defined to be
although this result is also a consequence of distributivity, associativity and

the replacement i2 = -1.
Examples. 1. i3 = -i, i4 = 1, i5 = i. 2. (1 + i)2 = 2i. I
+
The product of a complex number t = x iy and its complex conjugate
+
5 = x - iy is a real number z.2 = x 2 Y2. Since this real number is non-zero
for z # 0, we may introduce the inverse
or in coordinate form
1 - x - iy
---
x + iy +
x 2 y2 '
Division is carried out as multiplication by the inverse: zl/z2 = z l z ~ l .

If we introduce polar coordinates r, cp in the complex plane by setting x =
+
r cos cp and y = r sin cp, then the complex number z = x iy can be written as
z = r(coscp+ isincp).
This is the polar form of z. The distance r = from z to 0 is
denoted by lzl and called the norm of z. Thus
The real number cp is called the phase-angle or argument of z [sometimes all

the real numbers cp + 27rk, k E Z,are assigned to the same phase-angle].
The familiar addition rules for the sine and cosine result in the polar form
of multiplication,
3 Electrical engineers denote the polar form by r k .

4 The scalar product Re(zlZ2) is compatible with the norm lzl. Incidentally, Im(zlE2)
measures the signed area of the parallelogram determined by zl and z2.
20 Complex Numbers
of complex numbers
= rl (cos p l + i sin pl) and z2 = r2(cos cp2 + i sin p2).

Thus, the norm of a product is the product of the norms,
and the phase-angle of a product is the sum of the phase-angles (mod 27r).
The exponential function can be defined everywhere in the complex plane by
We write ez = exp(z). The series expansions of trigonometric functions result

in Euler's formula
e'9 = cosp + i s i n p
which allows us to abbreviate z = r(cos cp + i sin cp) as z = re'p
The exponential form of multiplication seems natural:

( ~ ~ e ~ 9 l ) ( r ~=
e ~(rl9 r2)ei(91+92)
2)
Powers and roots are computed as
(rei'+')n = r n e i n 9 and = fiei9/n+i2rk/n, k E z n .
Examples. (1 + i)-' = i ( 1 - i), & = &$(l+ i), eirl2 = i. I
2.1 The field @ versus the real algebra @

Numbers are elements of a mathematical object called a field. In a field numbers
can be both added and multiplied. The usual rules of addition
a+b=b+a commutativity
+
(a + b) c = a + (b + c ) associativity
a+O=a zero 0
a + (-a) = 0 opposite -a of a
2.1 The field C versus the real algebra C 21
are satisfied for all numbers a, b, c in a field IF. The multiplication satisfies
(a + b)c = ac+ bc
distributivity
a(b + c) = ab + ac
(ab)c= a(bc) associativity
la = a unity 1
aa-' = 1 inverse a-' of a # 0
ab= ba commutativity
for all numbers a, b, c in a field IF. The above rules of addition and multiplica-
tion make up the axioms of a field F.
Examples of fields are the fields of real numbers R , complex numbers C ,
rationals Q , and the finite fields JFq where q = pm with a prime p.
It is tempting to regard R as a unique subfield in C . However, C contains
several, infinitely many, subfields isomorphic to R ; choosing one means intro-
ducing a real linear structure on C , obtained by restricting a in the product
C x C + C , ( a ,b) -+ ab to be real, a E R . Such extra structure turns the field
(C into a real algebra C.
Definition. An algebra over a field IF is a linear space A over IF together with

a bilinear function A x A -+ A , ( a ,b) + ab. I
To distinguish the field C from a real algebra C let us construct C as the

set IR x R of all ordered pairs of real numbers z = ( x ,y) with addition and
multiplication defined as
The set R x R together with the above addition and multiplication rules makes
up the field C. The imaginary unit ( 0 , l ) satisfies (0,l)' = (- 1 , O ) .
+ +
Since ( X I , 0 ) (x2,O) = ( 2 1 x2,O) and ( X I , 0 )(x2,O) = (21x2,O ) , the real
field R is contained in C as a subfield by R + C , x -+ ( x ,0 ) . If we restrict
multiplication so that one factor is in this distinguished copy of R ,
then we actually introduce a real linear structure on the set R 2 = R x R . This

5 The finite fields IFp., where q = pm with a prime p, are called Galois fields GF(pm).
6 Bilinear means linear with respect to both arguments. This implies distributivity. In other
words, distributivity has no independent meaning for an algebra.
7 Note that associativity is not assumed.
22 Complex Numbers
real linear structure allows us to view the field of complex numbers intuitively
as the complex plane C.
The above construction of C as the real linear space R 2 brings in more
structure tban just the field structure: it makes C an algebra over R. We
often identify R with the subfield {(x, 0) I x E R) of @, and denote the
standard basis of R2 by 1 = (1, O), i = ( 0 , l ) in @.
A function a : C + C is an automorphism of the field C if it preserves addition
and multiplication,
a(z1 + z2) = a(z1) + +2),
as well as the unity, a(1) = 1. A function a : C + C is an autoniorphism of

the real algebm C if it preserves the real linear structure and multiplication (of
complex numbers),
as well as the unity, a(1) = 1.

The field C has an infinity of automorphisms. In contrast, the only auto-
morphisms of the real algebra C are the identity automorphism and complex
conjugation.
Theorem. Complex conjugation is the only field automorphism of C which is
different from the identity but preserves a fixed subfield R.
Proof. First, note that a ( i ) = f i for any field automorphism a of C, since
a ( i ) 2 = a ( i 2 ) = a(-1) = -1. If a : C + C is a field automorphism such that
a ( R ) C R , then a ( x ) = x for all x E R , because the only automorphism of
+
the real field is the identity. It then follows that, for all x iy with x, y E R ,
where a ( i ) = fi. The case a ( i ) = i gives the identity automorphism, and the
case a ( i ) = -i gives complex conjugation. I
The other automorphisms of the field C send a real subfield R onto an iso-
morphic copy of R , which is necessarily different from the original subfield R.
However, any field automorphism of C fixes point-wise the rational subfield Q.
8 The geometric view of complex numbers is connected with the structure of (C as a real
algebra, and not so much as a field.
9 In the above construction we introduced a field structure into the real linear space WZ and
arrived at an algebra (C over W, or equivalently at a field (C with a distinguished subfield
W.
2.2 The double-ring 'R of R 23
Example. It is known that there is a field automorphism of C sending fi to
-.\/Zand i/Z to ifi, but no one has been able to construct such an automor-
phism explicitly since its existence proof calls for the axiom of choice. I
If a field automorphism of C is neither the identity nor a complex conjugation,
then it sends some irrational numbers outside R , and permutes an infinity of
subfields all isomorphic with R. Related to each real subfield there is a unique
complex conjugation across that subfield, and all such automorphisms of finite
order are complex conjugations for some real subfield. The image a ( R ) under
such an automorphism a of a distinguished real subfield R is dense in C [in the
topology of the metric 1 . ~ 1 = f i given by the complex conjugation across R].
This can be seen as follows: An automorphism a must satisfy cr(rx) = r a ( x )
when r E Q. So if there is an irrational x E R with t = a ( x ) 6 R , and nec-
+
essarily t $ Q iQ, the image a ( R ) of R contains all numbers of the form
+ +
a(r SX) = r st with r, s E Q . This is a dense set in C.
The above discussion indicates that there is no unique complex conju-
gation in the field of complex numbers, and that the field structure of C
does not fix by itself the subfield R of C. The field injection R + C is an extra
piece of structure added on top of the field C. If a privileged real subfield R
is singled out in C, it brings along a real linear structure on C, and a unique
complex conjugation across R , which then naturally imports a metric structure
to @.
Our main interest in complex numbers in this book is C as a real algebra,
not so much as a field.
2.2 The double-ring 2R of R

There is more than one interesting bilinear product (or algebra structure) on
the linear space R2. For instance, component-wise multiplication
(21, ~ 1 ) ( 2 2 ~, 2 =
) (21x2, ~ 1 ~ 2 )
results in the double-ring 2R of R. The only automorphisms of the real algebra

2R are the identity and the swap
2R + 2 ~ , P) -,swap(& P) = (P, A).
(A,
The swap acts like the complex conjugation of C, since
swap[a(l, 1) + b(1, -I)] = a ( 1 , l ) - b(1, -1).
The multiplicative unity 1 = ( 1 , l ) and the reflected element j = (1, -1) are
now related by j2 = 1.
24 Complex Numbers
Alternatively and equivalently we may consider pairs of real numbers (a, b) E
R2 as Study numbers
Study numbers have Study conjugate (a + jb)- = a - j b , Lorentz squared

norm (a + jb)(a - j b ) = a2 - b2, and the hyperbolic polar form a j b = +
p(cosh x + j sinh X) for a2 - b2 2 0. l o In products Lorentz squared norms
are preserved and hyperbolic angles added. Study numbers have the matrix
represent ation
Exercise 4
2.3 Representation by m e a n s of real 2 x 2-matrices

Complex numbers were constructed as ordered pairs of real numbers. Thus we
can replace
making explicit the real linear structure on C.The product of two complex
numbers c = a + ib and z,
can be replaced by / factored as
One is thus led to consider representing complex numbers by certain real 2 x 2-

matrices in Mat(2, R) : l 1
10 The linear space R2 endowed with a n indefinite quadratic form (a, b) + a2 - b2 is the
hyperbolic quadratic space R1ll.The Clifford algebra of R 'l' is CP1,l which has Study
numbers as the even subalgebra ~e:,~.
11 In this matrix representation, the complex conjugate of a complex number becomes the
transpose of the matrix and the (squared) norm becomes the determinant. The norm is
preserved under similarity transformations, but 'transposition = complex conjugation' is
only preserved under similarities by orthogonal matrices.
2.3 Representation by means of real 2 x 2-matrices 25
The multiplicative unity 1 and the imaginary unit i in C are represented by
the matrices
I=(: y)
and J
0 -1
,).
However, this is not the only linear representation of C in Mat(2,R). A sim-
ilarity transformation by an invertible matrix U , det U # 0, sends the repre-
sentative of the imaginary unit J to another 'imaginary unit' J' = U J U - I in
Mat(2, R).
(i
E x a m p l e . Choosing U = :) , we find J' = (: I;), and the matrix repre-
+
sentation x iy + (5:Y i!Yy). I
In the rest of this chapter we shall study introduction of complex numbers by

means of the Clifford algebra Ce2 of the Euclidean plane R2. This approach
gives the imaginary unit i = flvarious geometrical meanings. We will see
that i represents
(i) an oriented plane area in R2,
(ii) a quarter turn of R2.
The Euclidean plane R 2 has a quadratic form
We introduce an associative product of vectors such that
Using distributivity this results in the multiplication rules
The element ele2 satisfies
and therefore cannot be a scalar or a vector. It is an example of a bivector, the

unit bivector. Denote it for short by e l 2 = ele2.
26 Complex Numbers
2.4 @ as t h e even Clifford algebra ~lt
The Clifford algebra C12 is a Cdimensional real algebra with a basis (1, el, e 2 , e12).
The basis elements obey the multiplication table
The basis elements span the subspaces consisting of l2
1 R scalars
el, e 2 R2 vectors
el2 R2 bivectors.
Thus, the Clifford algebra C12 contains copies of R and R2, and it is a direct
sum of its subspaces of elements of degrees 0,1,2:
The Clifford algebra is also a direct sum C12 = C l t $ Ct; of its

even part C l t = R $ R2,
odd part Cl; = R2.
The even part is not only a subspace but also a subalgebra. It consists of
+
elements of the form x yela where x, y E R and e:, = -1. Thus, the even
subalgebra C t t = R CB R2 of C12 is isomorphic to @. The unit bivector e l 2
shares the basic property of the square root i of -1, that is i2 = -1, and we
could write i = e l 2 . It should be noted, however, that our imaginary unit e l 2
anticommutes with el and e 2 and thus e l 2 anticommutes with every vector
in the ele2-plane: l3
re12 = -e12r for r = xel + ye2 and el2 = elez.
12 In higher dimensions the Clifford algebra Cen of W n is a sum of its subspaces of k-vectors:
ce, = a e W n e ~ ~ j ~ n e . . . e / \ ~ a ~ .
13 In a complex linear space, or complex algebra, where scalars are complex numbers, the
imaginary unit commutes with all the vectors, ir = ri.
2.5 Imaginary unit = the unit bivector
2.5 I m a g i n a r y unit = the unit b i v e c t o r

+
Multiplying the vector r = xel ye2 by the unit bivector el2 gives another
vector re12 = xe2 - ye1 which is perpendicular to r . The function r + re12 is
a left turn, and the effect of two left turns [e12.e12]is to reverse direction [ - I ] ;
or, in a more picturesque way, is a U-turn. The statement ' e f 2= -1' is just
an arithmetic version of the obvious geometric fact that the sum of two right
angles, 90' +90°, is a straight angle, 180'. In the vector plane R2 the sense of
+
rotation depends on what side the vector r = xel ye2 is multiplied by el2 so
that the rotation r + el2r = ye1 -xe2 is clockwise and r + re12 = -yel +xe2
is counter-clockwise.
In the complex plane C = R@/\' IW' both the rotations sending z = x+ ye12
to el22 and tel2 are counter-clockwise. Multiplying a complex number z =
+
x ye12 by the unit bivector el2 results in a left turn, %el2= -y xel2,+
and the effect of two left turns [el2. el2]is direction reversal [ - I ] ; that is a
half-turn in the complex plane C:
The square root of -1 has two distinct geometric roles in IW2: it is the
generator of rotations, i = ele2 E C e z , and it represents a unit oriented plane
area el A e2 E /\' It2. l4
+
A complex number t = x ye12 E R @ /\' R2 is a sum of
- a real number x = Re(%) and
- a bivector ye12 = el2 Im(z) .
14 In an n-dimensional vector space W n rotations can be represented by multiplications
in Clifford algebras Ce,, while certain simple elements of the exterior algebra /\Rn =
-. An
R $ W n $ /\' Rn $ . $ W n represent oriented subspaces of dimensions 0 , 1 , 2 , . ..,n .
28 Complex Numbers
2.6 E v e n a n d odd parts
The Clifford algebra Ce2 of IR2 contains both the complex plane C and the
vector plane R 2 so that
R2 is spanned by el and e2,
C is spanned by 1 and el2 .
The only common point of the two planes is the zero 0 . The two planes are both
parts of the same algebra Ce2. The vector plane R2 and the complex field C are
incorporated as separate substructures in the Clifford algebra Ce2 = $Ce;
so that the complex plane C is the even part Ce; and the vector plane EX2 is
the odd part Ce; .
Vector plane IR2 = Ce; Complex plane C =
The names even and odd mean that the elements are products of an even or
odd number of vectors. Parity considerations show that
- complex number times complex number is a complex number,
- vector times complex number is a vector,
- complex number times vector is a vector, and
- vector times vector is a complex number.
The above observations can be expressed by the inclusions
ce;ce; c ce;,
ce; ce; c ce; ,
ce;ce; c ce; ,
ce; ce; c ce;.
By writing ( C e 2 )=
~ and (Ce2)l = Ce; , this can be further condensed to
(Ce2)j(Ce2)kc (Ce2)j+k,where j , k are added modulo 2. These observations
are expressed by saying that the Clifford algebra Ce2 has an even-odd grading
or that it is graded over Z2 = ( 0 , l ) . l 5
15 We have already met a &-graded algebra, namely the real algebra C = R $ iR with even
part R = Re(C) and odd part iR = i Im(C).
2.7 Involutions and the norm 29
2.7 Involutions a n d t h e n o r m
The Clifford algebra Cl2 has three involutions similar to complex conjugation
+
in @. For an element u = ( u ) ~ ( u ) l + (u)2 E Ce2, ( u ) ~E l\kEX2, we define
grade involution ii = ( u ) -
~ (u)~ +
reversion +
6 = ( 4 0 ( 4 1 - (42,
Clifford-conjugation ii = (u)O - ( u ) ~-
The grade involution is an automorphism, = Ci, while the reversion and
the Clifford-conjugation are anti-automorphisms, = G i i , iE = Zlu.
+
For a complex number z = x ye12 the complex conjugation z + 2 =
x - ye12 is a restriction of the Clifford-conjugation u + ii in C12 and also of
the reversion u + 6 in C12. Likewise, the norm lzl= J - in @, obtained
+
as the square root of z2 = x2 y2, is a restriction of the norm ( u (=
in Ce2.
a
A complex number is a product of its norm r = lzl and its phase-factor
+
cos cp el2 sin cp, where x = r cos cp and y = r sin cp. The expression z =
+
r(cos cp e l 2 sin cp) can be abbreviated as z = r exp(e12cp), and read as ' r in
phase cp.'
2.8 Vectors multiplied b y complex n u m b e r s

+
The product of a vector r = xel ye2 and a unit complex number eiq =
+
cos cp i sin cp, where for short i = el2, is another vector in the ele2-plane:
r cos cp + r i sin cp = re1'+'.
The vector r i = xe2 - ye1 is perpendicular to r so that a rotation to the left
by ~ / carries
2 r to ri.
Since the unit bivector i anticommutes with every vector r in the ele2-
plane, the rotated vector could also be expressed as
+ 5
Furthermore, we have cos cp i sin cp = (cos $. i sin 5)2
and thus the rotated
vector also has the form s-lrs where s = ei'+'I2and s-' = e-j'+'I2. The ro-
tation of r to the left by the angle cp will then result in r z = z-lr = s-'rs
where z = ei'+',t - l = e-i'+' and s2 = Z. There are two complex numbers s and
-s which result in the same rotation s-'rs = (-s)-lr(-s). In other words,
there are two complex numbers which produce the same final result but via
different actions.
Complex Numbers
rZ
2.9 T h e g r o u p Spin(2)
The unit complex numbers z E @, JzJ= 1, form the unit circle S1 = { z E
@ I Izl = I), which with multiplication of complex numbers as the product
becomes the unitary group U(1) = {z E C I zZ = 1). A counter-clockwise
rotation of the complex plane C by an angle cp can be represented by complex
number multiplication:
A counter-clockwise rotation of the vector plane R2 by an angle cp can be

represented by a matrix multiplication:
(;)+ ~ 0
(sincp
-+) (;), (
s ~
coscp
coscp - sincp
sin cp cos cp )E SOP]
where SO(2) = {R E Mat(2,R) I R ~ R = I, det R = I), the rotation group.
The rotation group SO(2) is isomorphic to the unitary group U(1).
Rotations of R2 can also be represented by Clifford multiplication: l6
xel + ye2 i (cos -cp2 + e l 2 sin E)-'(zel
2
+ ye2)(cos-cp2 + e l 2 sin -)cp2
+
where cos $ e l 2 sin $ E Spin(2) = { s E Cll I sS = 11, the spin group.
The fact that two opposite elements of the spin group Spin(2) . . represent the
-
same rotation in SO(2) is expressed by saying that Spin(2) is a two-fold l7

cover of S0(2), and written as Spin(2)/{f 1) 2 SO(2). Although SO(2) and
Spin(2) act differently on EX2, they are isomorphic as abstract groups, that is,
+
16 We use this particular form t o represent the rotation because the expression x e l ye2 -+
(COS$+el2 sin $)-'(xel+ye2)(cos :+el2 sin $) can be generalized t o higher dimensions.
+ +
T h e expression x e l ye2 -+ (xel + yez)(coscp el2 sincp) is not generalizable t o higher-
dimensional rotations.
17 You are already familiar with two-fold covers: 1. A position of the hands of your watch
corresponds t o two positions of the Sun. 2. A rotating mirror turns half t h e angle of
the image. 3. Circulating a coin one full turn around another makes t h e coin turn twice
around its center.
2.9 The group Spin(2)
Spin(2) 2: SO(2). Is
Exercise 6
History
Imaginary numbers first appeared around 1540, when Tartaglia and Cardano
expressed real roots of a cubic equation in terms of conjugate complex num-
bers. The first one to represent complex numbers by points on a plane was
a Norwegian surveyor, Caspar Wessel, in 1798. He posited an imaginary axis
perpendicular to the axis of real numbers. This configuration came to be known
as the Argand diagram, although Argand's contribution was an interpretation
of i = as a rotation by a right angle in the plane. Complex numbers got
their name from Gauss, and their formal definition as pairs of real numbers is
due to Hamilton in 1833 (first published 1837).
Exercises
+
1. (3 4i)-l, d m , G, 6, + log(-1 i).
2. Let zk = ei2"k/n, k = 1 , 2 , . . . ,n - 1. Compute
(1 - zl)(l - zz) . . . (1 -
3. An ordering of a field IF is an assignment of a subset P C IF such that
(i) 0 $2 p,
(ii) for all non-zero a E IF either a E P or -a E P, but not both,
+
(iii) a b E P and ab E P for all a , b E P.
It is customary to call P the set of positive numbers, and the set
-P = {-a I a E P ) the set of negative numbers. The statement a - b E P
>
is also written a > b (and a - b E P U (0) is written a b). Show that the
field C cannot be ordered.
4. Two automorphisms a , /3 of an algebra are similar if there exists an
intertwining automorphism y such that a y = y/3. The identity
automorphism is similar only to itself.
a) Show that the two involutions of the real algebra C are dissimilar, and
that the two involutions of the real algebra 21W are dissimilar.
b) Show that the two involutions a(X,p) = (p,X) and P(X,p) = (,3, A) are
similar involutions of the real or complex algebra 2 C [that is, find an
intertwining automorphism y of 2@ such that a y = yP].
5. A rotation is called rational if it sends a vector with rational coordinates to
18 Both SO(2) and Spin(2) are homeomorphic to S'.
32 Complex Numbers
another vector with rational coordinates. Determine all the rational
rotations of R2. Hint: R E SO(2) \ {-I) can be written in the form
+
R = (I A ) ( I -A)-' where = -A.
+ + +
6. Write 6 = ( u ) ~ ( u ) ~- (u)2 for u = ( u ) ~ ( u ) ~ ( u ) ~E C12,
(u)k E R2. Let Pin(2) = {u E Ce2 I 6u = I),
EX2 -+ R 2 , x + R(x) = uxu-l, and O(2) = {R E Mat(2,R) I R ~ = R I).
Show that Pin(2)/{f 1) 2 O(2) and Pin(2) rz O(2).
7. Show that a 2-dimensional real algebra with unity 1 is both commutative
and associative. Hint: First show that there is a basis {1,a) such that
a2 = a l , a E R .
8. Show that a 2-dimensional real algebra with unity 1 and no zero-divisors
[ab = 0 implies a = 0 or b = 0] is isomorphic to C.
Solutions
1. i ( 3 - 4i), *(2+ i), f 1 f i, fi= {i,*$ - ii),
log(-1 + +
i) = log2 i? i 27rk.+
2. Note that the roots of xn - 1 = 0 are zk = eiZrkfn, k = 0 , 1 , . . . ,n - 1
Therefore (x - tO)(x- zl) (x - t2). . . (x - ~ ~ - = ) - 1. Define
1 xn
f (x) = (X - zl)(x - t2) (x - tn-1) which equals
+ + +
xn - 1
f (x) = ~ - - lfor x # 1
+ +
and f (x) = xn-I + . . . x 1 in general. Compute f (1) = n.
3. In an ordered field non-zero numbers have positive squares, and the sum of
such squares is positive, and therefore non-zero. The equality i2 + 1 = 0 in
C can also be written as i2 + l2= 0, which excludes the inequality
i2 + l2 > 0. Consequently, it is impossible to order the field C.
4. b) Choose y(X,p) = (X,p) or y(X,p) = (X,,Q) to find a y = yp.
Bibliography
L.V. Ahlfors: Complex Analysis. McGraw-Hill, New York, 1953.
E. Cartan (expos6 d'aprks l'article allemand de E. Study): Nombres complexes; pp.
329-468 in J. Molk (red.): Encyclope'die des sciences mathe'matiques, Tome I, vol.
1, Fasc. 4, art. 15, 1908. Reprinted in E. Cartan: CEuvres complktes, Partie 11.
Gauthier-Villars, Paris, 1953, pp. 107-246.
R.V. Churchill, J.W. Brown, R.F. Verhey: Complex Variables and Applications.
McGraw-Hill, New York, 1984.
H.-D. Ebbinghaus et al. (eds.): Numbers. Springer, New York, 1991.
B. L. van der Waerden: A History of Algebra. Springer, Berlin, 1985.
3
Bivectors and the Exterior Algebra
There are other kinds of directed quantities besides vectors, most notably bivec-
tors. For instance, a moment of a force, angular velocity of a rotating body,
and magnetic induction can be described with bivectors. In three dimensions
bivectors are dual to vectors, and their use can be circumvented. Scalars, vec-
tors, bivectors and the volume element span the exterior algebra AIR3, which
provides a multivector structure for the Clifford algebra C13 of the Euclidean
space R3.
3.1 Bivectors as directed plane segments

In three dimensions bivectors are oriented plane segments, which have a di-
rection and a magnitude, the area of the plane segment. Two bivectors have
the same direction if they are on parallel planes (the same attitude) and are
similarly oriented (the same sense of rotation).
Vector (directed line segment)

1. magnitude (length of PQ)
2. direction
- attitude (line PQ)
- orientation (toward the point Q)
P
Bivector (directed plane segment)
Q 1. magnitude (area of OPQR)
2. direction
- attitude (plane OPQ)
- orientation (sense of rotation)
0 P
34 Bivectors and the Exterior Algebra
Bivectors are denoted by boldface capital letters A, B , etc. The area or
norm of a bivector A is denoted by IAl. Two bivectors A and B in parallel
planes have the same attitude, and we write A 11 B. Parallel bivectors A and
B can be regarded as directed angles turning either the same way, A TT B , or
the opposite way, A TJ. B. If two plane segments have the same area and the
same direction (= parallel planes with the same sense of rotation), then the
bivectors are equal:
A=B lAl=lBl and A T T B
A bivector A and its opposite -A are of equal area and parallel, but have
opposite orientations. A unit bivector A has area one, [ A [= 1.
The shape of the area is irrelevant.
Representing a bivector as an oriented parallelogram suggests that a bivector
can be thought of as a geometrical product of vectors along its sides. With this
in mind we introduce the exterior product a A b of two vectors a and b as the
bivector obtained by sweeping b along a.
a
The bivectors a A b and b A a have the same area and the same attitude but
opposite senses of rotations. This can be simply expressed by writing
3.2 Addition of bivectors

The geometric interpretation of bivector addition is most easily seen when
the bivectors are expressed in terms of the exterior product with a common
L L
1 In handwriting, bivectors can be distinguished by an angle on top of the letter, A, B .
3.3 Basis of the linear space of bivectors 35
vector factor. In three dimensions this is always possible because any two
planes will either be parallel or intersect along a common line. Thus let
A = a A c and B = b A c; then the bivector A B is defined so that +
A + B = a A c + b A c = (a + b) A c. The geometric significance of this can be
depicted as follows:
By decomposing the vectors a and b into components parallel and perpendic-

+
ular to a b, SO that
a = all +a l and b = bll +b l
where bl = - a l , we are able to reduce the general addition of bivectors in
three dimensions to the addition of coplanar bivectors. This is evident in the
equality
3.3 Basis of the linear space of bivectors

Bivectors can be added and multiplied by scalars. This way the set of bivectors
becomes a linear space, denoted by / \ 2 ~ 3 . A basis for the linear space /\2 R3
can be constructed by means of a basis {el,e2, e3) of the linear space R3.
The oriented plane segments of the coordinate planes, obtained by taking the
exterior products
2 T h e two bivectors are first translated in the affine space R3 so t h a t they induce opposite
orientations to their common edge, t h a t is, the terminal side of A = a A c is opposite t o
the initial side of B = (-c) A b.
3 A depiction of addition of bivectors does not require a metric, or perpendicular compo-
+
nents. I t is sufficient that one component of both a and b is parallel t o a b, so t h a t
the two components sum up t o a + b, while the other component can be any non-parallel
component.
form a basis for the linear space of bivectors /\'It3.
e3
An arbitrary bivector is a linear combination of the basis elements,
and such linear combinations form the space of bivectors /\'IR3. The con-
struction of bivectors calls only for a linear structure, and no metric is needed.
The scalar product on a Euclidean space R3 extends to a symmetric bilinear
product on the space of bivectors /\'It3,
In particular, <a A b , a A b > = la121b12- (a - b ) 2 .The norm or area of B =

+ +
Bl2el A ea Bl3el A e3 BZ3e2A e3 is seen to be
3.4 T h e oriented volume element

The exterior product a A b A c of three vectors a = ale1 a282 (3383, + +
+ +
b = blel b2e2+ b3e3 and c = clel + c2e2 c3e3 represents the oriented
volume of the parallelepiped with edges a, b , c :
It is an element of the 1-dimensional linear space of 3-vectors /\3R3 with basis

el A e2 A e3. The exterior product is associative,
4 In three dimensions all bivectors are simple, that is, they are exterior products of two
vectors, B = x A y for some x, y E R3.This is no longer true in four dimensions; for
+
instance el A e? e3 A e4 is not simple.
3.5 The cross product
and antisymmetric,
for a , b , c E IR3.
The exterior product of the orthogonal unit vectors el, e 2 , e 3 E R3 is the
unit oriented volume element el A e 2 A e 3 E /\3 R3. The norm or volume JVI
of a 3-vector
is IVI = IVI, that is, l V e l A e 2 A e s l = V for V 2 0 and lVel A e 2 A e 3 1 = -V

for V < 0.
More formally, the scalar product on R3 extends to a symmetric bilinear
product on /\3 R3 by
Xl.Yl Xl'Y2 Xl'Y3
<xi A x 2 A ~ 3Y i, A ~2 A y 3 > = ~2 . yi ~2 . ~2 ~2 .y 3
X3'Yl X3'Y2 X3'Y3
giving the norm as IVI = Jw.

3.5 The cross p r o d u c t
Let a = a l e 1 + a 2 e 2 + use3 and b = blel + b2e2+ b3e3. The bivector
can be expressed as a 'determinant'

e2 A e3 e3 A el
el A e 2
aAb= a1 a2 as .
bl b2 b3
It is customary to introduce a vector with the same coordinates. Thus, we
define the cross product
of a and b . The cross product can also be represented by a 'determinant'

el ez ea
axb= a1 a2 a3 .
bi bz b3
5 V is a real number, positive or negative, while V is a 3-vector. The usual volume is IVI.
Bivectors and the Ezterior Algebra
a x b
The direction of a x b is perpendicular to the plane of aAb and the lengthlnorm

of a x b equals the area/norm of a A b ,
where cp, 0 5 cp 5 180°, is the angle between a and b .

In spite of the resemblance between the determinant expressions for the ex-
terior product a A b and the cross product a x b there is a difference: the
exterior product does not require a metric while the cross product requires or
induces a metric. The metric gets involved in positioning the vector a x b
perpendicular to the bivector a A b.
3.6 T h e Hodge dual

Since the vector space R3 and the bivector space /\2 R3 are both of dimension
3, they are linearly isomorphic. We can use the metric on the vector space R3
to set up a standard isomorphism between the two linear spaces, the Hodge
dual sending a vector a E R3 to a bivector *a E /\2 IR3, defined by
b A * a = ( b . a ) e l A e 2 A e 3 for all b E R3.
The Hodge dual depends not only on the metric but also on the choice of
orientation - it is customary to use a right-handed and orthonormal basis
{el,e 2 , e3).
3.6 T h e Hodge dual
Vector a and its dual bivector A = ae123
Thus, we have assigned to each vector
a bivector
2
A = *a = ale2 A e3 + aze3 A el + aael A e2 E AIR3
Using the induced metric on the bivector space lR3 we can extend the Hodge
dual to a mapping sending a bivector A E lR3 to a vector *A E lR3, defined
by
2
B A *A = < B , A>el A e2 A e3 for all B E A IR3.
Using duality, the relation between the cross product and the exterior pro-
duct can be written as
a A b = * ( a x b),
a x b=*(aAb).
6 In terms of the Clifford algebra C13 the relation between the exterior product and the
cross product can be written as
The metric gets involved in multiplying by e l 2 3 = e l e z e g . Using the Clifford algebra

Ce3 the Hodge dual can be computed as *u = G e 1 2 3 . This gives rise to the Clifford dual
defined as u e 1 2 3 for u E C e 3 . Later we will see that in actual computations the Clifford
dual is more convenient than the Hodge dual (although in three dimensions the Hodge
dual happens to be symmetric/involutory).
3.7 T h e exterior algebra and t h e Clifford algebra
The exterior algebra AIR3 of the linear space IW3 is a direct sum of the
subspaces of I with basis
scalars IR 1
vectors B3 e l , e2, e3
bivectors /\2 IR3 el A e 2 , el A e3, e 2 A e3
volume elements /\31W3 el A e2 A e3
We also write R = / \ O I W 3 and IW3 = /\'It3. Thus, /\IW3 is a direct sum of its
subspaces of homogeneous degrees 0,1,2,3:
The dimensions of B , R3, /\2 IR3, /\3 IW3 and /\ IW3 are 1,3,3,1 and Z3 = 8,
respectively.
The exterior algebra /\IW3 is an associative algebra with unity 1 satisfying
for a basis {el, e 2 , e3) of the linear space IW3. The exterior product of two
homogeneous elements satisfies
i+j i j
a/Ib€AIR3 for a € A g 3 , b€/\IW3.
The product of two elements u and v in the Clifford algebra Ce3 of the
Euclidean space IW3 is denoted by juxtaposition, uv, to distinguish it from
the exterior product u A v. An orthonormal basis {el,e 2 , es) of the Euclidean
space IW3 C Ce3 satisfies
7 These rules were invented by W.K. Clifford in 1882. In an earlier paper Clifford 1878 had
considered an associative algebra of dimension 8 with the rules e,e, = -1 for i = 1,2,3.
3.8 The Clifford product of two vectors
and generates a basis of Ce3, corresponding to a basis of AIR3,
The above correspondences induce an identification of the linear spaces Ces

and IR3, and we shall write
This decomposition introduces a multivector structure into the Clifford algebra

Ces. The multivector structure is unique, that is, an arbitrary element u E Ce3
can be uniquely decomposed into a sum of k-vectors, the k-vector parts ( u ) ~
of u,
k
u = (u)o + (U)I +(u)z +(u)3 where (u)k E AIR3.
3.8 The Clifford product of two vectors

A new kind of product called the Clifford product of vectors a and b is obtained
by adding the scalar a.b and the bivector a A b :
The commutative rule a . b = b . a together with the anticommutative rule

a A b = -b A a implies a relation between ab and b a . Thus,
Two vectors a and b are parallel, a 11 b , when their product is commutative,

a b = b a , and perpendicular, a Ib, when their product is anticommutative,
ab = -ba.
Note that if a is decomposed into components parallel, all, and perpendic-
ular, a l , to b, then ab = a l l b + a L b= a . b + a A b .
Bivectors and the Exterior Algebra
a . b = i(ab+ba)
a Ab = i ( a b - ba)
Compute the product abba to get a2b2= (a . b)2 - (a A b)2 and use
(a A b)2 = -la A bI2 to obtain the identity
a2b2= ( a -b)2+ la^ bI2.
3.9 Even and odd parts

The Clifford algebra is, like the exterior algebra, a direct sum of two of its
subspaces,
the even part R $ /\2 R3,
the odd part R3 $ /\3 R3.
For both algebras the even part is also a subalgebra. The even subalgebra
(/\ R3)+ = R $ /\2 R3 of /\ R3 is commutative, but the even subalgebra ~ 1 ;=
R3 of Ce3 is not commutative; instead it is isomorphic to the quaternion
algebra: W E ~e;. The odd parts are denoted by Ce; and (/\ R3)-.
3.10 T h e center
The center of an algebra consists of those elements which commute with all the
elements of the algebra. The center Cen(Ce3) = I W $ / \ ~ I W ~of Ce3 is isomorphic
to C, and the center of /\ R3 is Cen(/\ R3) = R $ /\2 R3 $ /\3 R3.
3.11 Gradings and t h e multivector structure

The exterior products of homogeneous elements satisfy the relations
i +j I j
a h b AIR3
~ for a € A I R 3 and b € A R 3 .
Such a property of an algebra is usually referred to by saying that the algebra
is graded over the index group Z. We shall refer to this property of the exterior
algebra /\R3 as the dimension grading, because simple homogeneous elements
3.11 Gradings and the multivector structure 43
represent subspaces of specified dimension. The homogeneous elements in /\ R3
satisfy
i j
a A b = (-1)"bha for a € / \ R 3 , b E / \ R ~ ,
that is, the exterior algebra /\R3 is graded commutative.

The Clifford products of even and odd subspaces satisfy the inclusion rela-
tions
ce; ce; c ce; , ce; ce; c ce; ,

ce, ceg c ce; , ce; ce; c ce;.
These relations can be summarized by saying that the Clifford algebra Ce3 has
an even-odd grading, or that it is graded over the index group Z2= ( 0 , l ) .
The exterior algebra /\R3 is also even-odd graded.
The Clifford algebra Ce3 is not graded over Z. However, we can reconstruct
the exterior product from the Clifford product in a unique manner. We shall
refer to the dimension grading of the associated exterior algebra by saying that
the Clifford algebra has a multivector structure. Recall that R and R3 have,
by definition, unique copies in Ce3. The exterior product of two vectors equals
the antisymmetric part of their Clifford product,
2
1
x A y = -(xy - y x ) E / \ R ~ for x , y € R 3 ,
2
whence the space of bivectors /\' R3 has a unique copy in Ce3. The subspace
of 3-vectors /\3 R3 can be uniquely reconstructed within Ce3 by a completely
antisymmetrized Clifford product
of three vectors x , y , z E R3
Thus, we have established a linear isomorphism sending /\ R3 to Ce3 defined
8 The graded opposite algebra of /\ R3 is the linear space /\ R3 with a new product uov
defined by
for uo,vo E (/\R3)+and u1 ,vl E (/\R3)-. Since /\ R3 is graded commutative, that is

u o v = u A v , the graded opposite of /\R3is just /\ R3.
44 Bivectors and the Ezterior Algebra
for k-vectors:
AR3 I ce3
There is another construction of the subspace of 3-vectors /\3 R3, obtained by

using the reversion, xAy Az = $(xyz-zyx) E /\3R3 for x, y , z E R3, related
to the following recursive construction, via an intermediate step in /\2 R3 :
3 2
1
X A B = - ( X B + B X ) E A R ~for x € R 3 ,
2
BEAR^
3.12 Products of vectors and bivectors, visualization
A vector a E R3 and a bivector B E /\ R3 can be multiplied to give a 3-vector
a A B = B A a E /\3R3. The exterior product of a vector and a bivector can
be depicted as an oriented volume:
aAB BAa
The orientation is obtained by putting the arrows in succession. The commu-
tativity of the exterior product a A B = B A a means that the screws of a A B
and B A a can be rotated onto each other (without reflection).
A vector x E R3 and a bivector B E /\2R3 can also be multiplied so that
the result is a vector B L x E R3. Consider a vector x tilted by an angle cp out
of the plane of a bivector B. Let a be the orthogonal projection of x in the
plane of B. Then la1 = 1x1coscp. The right contraction of the bivector B by
the vector x is a vector y = B L x in the plane of B such that
(1 lYl=IBllal,
(ii) y l a and aAyTTB.
3.13 Contractions and the derivation
By convention, we agree that
that is, the left and right contractions have opposite signs.
[The inverse vector a-l of a has a geometrical meaning in this figure: it gives
= Ja-IJ J1.y1
the area of the rectangle, JBJ
Write XI( = a and x l = x - xll. Then x J B = xllB and x A B = xlB so
that
XII = (X J B)B-' parallel component
x l = (x A B)B-' perpendicular component
where B-I = B/B2, B2 = - 1~1~.
3.13 Contractions and the derivation

The Clifford product of two vectors a and b is a sum of a scalar a - b and a
bivector a A b,
so that the terms on the right hand side can be recaptured from the Clifford
product:
1 1
a . b = -(ab+ba), a A b = -(ab-ba).
2 2
The product of a vector a and a bivector B is a sum of a vector and a 3-vector:
where
In general, the Clifford product of a vector x E R3 and an arbitrary element

u E C13 can be decomposed into a sum of the left contraction and the exterior
product as follows:
where we can write, in the case where u is a k-vector in l\kR3,
The exterior product and the left contraction by a homogeneous element, re-
spectively, raise or lower the degree, that is,
i +.i j-i
~ A ~ E A R ~ , ~ J ~ E A B ~
for a E /\iR3 and b E /\'R3.
The left contraction can be obtained from the exterior product and the Clif-
ford product as follows:
This means that the left contraction is dual to the exterior product. The left
contraction can be directly defined by its characteristic properties
where x , y E R3 and u, v, w E /\ R3. Recalling that Q = (- 1)" for u E l\kR3,

the second rule can also be written as
when u E l\kR3. The second rule means that the left contraction by a vector is
a derivation of the exterior algebra /\R3. It happens that the left contraction
by a vector is also a derivation of the Clifford algebra, that is,
x J (UV)= ( x J u ) v + i i ( x J v ) for x ER3, U , VE C13.
9 A scalar product on R3 C /\ W3 induces a contraction on /\ R3 which can be used to
+
introduce a new product x u = x J u x /\ u for x E R3 and u E I\ R 3 , which extends by
linearity and associativity to all of /\ W3. The linear space /\ R3 provided with this new
product is the Clifford algebra Ce3.
3.14 The Clifford algebra versus the exterior algebra 47
3.14 T h e Clifford algebra versus t h e exterior algebra
Both the Clifford algebra CC3 and the exterior algebra /\R3 contain a copy
of R3, which enables application of calculations to the geometry of R3. The
feature distinguishing CC3 from /\R3 is that the Clifford multiplication of
vectors preserves the norm, lab1 = lallbl for a11 a, b E R3, whereas la A bl 5
lallbl. The equality lab1 = lallbl enables more calculations to be carried out
in R3, most notably rotations become represented as operations within one
algebra, the Clifford algebra CC3.
Historical survey
The exterior algebra /\R3 of the linear space IW" was constructed by Grass-
mann in 1844. Grassmann's exterior algebra /\R3 has a basis
satisfying the multiplication rules

ej A ej = -ej A ej for i # j,
ei Aei = 0.
Clifford introduced a new product into the exterior algebra; he kept the first
rule
eiej = -ejej for i # j,
that is eiej = ei A ej, but replaced the second rule by
ejei =1 in 1882, and
ejej = -1 in 1878.
These two algebras generated are Clifford's geometric algebras
CC3 = C&,O21 Mat(2, C) and CCo,3 2 W @ W
of the positive definite and negative definite quadratic spaces R3 = R3p0 and
R073, respectively.
Exercises
1. Find the area of the triangle with vertices (1, -4, -6), (5, -4, -2) and
(O,O, 0).
2. Find the volume of the parallelepiped with edges a = 2el - 3e2 + 4e3,
+
b = el 2e2 - eg, c = 3el - e2 2e3. +
3. Compute the square of the volume element el23 = elegeg (square with
respect to the Clifford product).
4. Show that el23 commutes with e l , eg, eg.
5. Find the inverse of the bivector B = 3e12+ e23 (inverse with respect to
the Clifford product).
+ + +
6. Let a = 2e1 3e2 7e3 and B = 4e12 5e13- e23. Compute a A B and
a J B.
+ + +
7. Let a = 3el 4e2 7e3 and B = 7e12 el3. Compute the perpendicular
and parallel components of a in the plane of B.
8. Show that the Clifford product of a bivector B E /\2 R3 and an arbitrary
element u E C13 can be decomposed as
9. Reconstruct the dot product a . b with the help of the cross product a x b
and the exterior product a A b . Hint: a x ( a x b ) = ( a .b ) a - a2b.
Define the right contraction by u L v = e ~ ; ~ [ ( e ~A~v]g for
u ) u , v E C13.
10. Show that the following properties - the characteristic properties - of the
right contraction hold:
1) xLy=x.y,
2) (u~v)Lx=u~(vLx)+(uLx)A~,
3) u L ( v A w) = ( u L v ) Lw,
for x , E R3 and u, v, w E /\R3.
11. Show that a L b E A"' R3 for a E /\h3 and b E /\j R3.
12. Show that ( u J v )L w = u J (v Lw).
13. Show that u J v =*(*-'(v) ~ i i and ) uLv =*-~(GA*(U)).
14. Show that
where, for a k-vector u E
15. Show that u A v - v A u ~ / \and

~ ~u v~ - v u ~ R ~ $ / \ ~ R ~ .
Let a € R 3 , B E / \ ~ I Wu =
~ ,l + a + B .
3.14 The Clifford algebra versus the exterior algebra 49
+
16. The exterior inverse of u is uA(-') = 1 - a - B aa A B with some
a E R . Determine a. Hint: use power series or u A uA(-'1 = 1.
+ +
17. The exterior square root of u is uA('f2) = 1 $a $B +pa B with
some ,8 E R. Determine p. Hint: u " ( ' / ~ ) u " ( ' / ~ ) = U.
18. Show that 1 J u = u for all u E AIR3.
Solutions
1. a = el - 4e2 - 6e3, b = 5el - 4e2 - 2e3, a A b = 16e12 + 28e13 - 16e23,
+ +
:la A bl = i d 1 6 2 282 162 = 18.
2. a A b A c = -7e123, la A b A cl = 7.
3. efz3= -1.
5. B~ = -10, IBI = m, B-' = -&(3e12 + eZs).
6. a A B = lle123, a J B = -47el + 15e2 +7e3.
7. a1 = -0.9e2 +
6.3e3, all = 3el + 4.9e2 + 0.7e3.
9. Take a wedge product with b to obtain (a x (a x b)) A b = ( a . b ) ( a A b),
and
(a x (a x b)) A b
a.b= for a b
aAb
(the division is carried out in the Clifford algebra C&, or it is just a ratio
of two parallel bivectors).
16. a = 2.
17. ,8 = --.41
18. 1J u = (1 A 1) J u = 1J (1J u) and so the contraction by 1 is a projection
with eigenvalues 0 and 1. The only idempotents of A R 3 are 0 and 1, and
so 1 J u = 0 or 1 J u = u, identically. The latter must be chosen, since
e R3.
1 J ( ~ . ~ ) = 1 J ( x J y ) = ( 1 ~ x ) J y = x J ~ = x ~ y # O f o r sx o, ym E
Bibliography
R. Deheuvels: Formes quadratiques et groupes classiques. Presses Universitaires de
France, Paris, 1981.
J. Dieudonnk: The tragedy of Grassmann. Linear and Multilinear Algebra 8 (1979),
1-14.
W. Greub: Multilinear Algebra, 2"d edn. Springer, Berlin, 1978.
J. Helmstetter: Alg6bres de Clifford et algkbres de Weyl. Cahiers Math. 25, Mont-
pellier, 1982.
G. Sobczyk: Vector Calculus with Complex Variables. Spring Hill College, Mobile,
AL, 1982.
I. Stewart: Hermann Grassmann was right. Nature 321, 1 May (1986), 17.
D. Sturmfels: Algorithms of Invariant Theory. Springer, Wien, 1993.
4
Pauli Spin Matrices and Spinors
In classical mechanics kinetic energy fmv2 = &, jj = m l , and potential

energy W = W(?) sum up to the total energy '
Inserting differential operators for total energy and momentum,
E=ifi-
a and p'=-ihV,
at
into the above equation results in the Schrodinger equation
a quantum mechanical description of the electron. The Schrodinger equation

explains all atomic phenomena except those involving magnetism and relativity.
The wave function 11, is complex valued, $(?, t ) E C. The square norm 1 ~ 1 2
integrated over a region in space gives the probability of finding the electron
in that region.
The Stern & Gerlach experiment, in 1922, showed that a beam of silver atoms
splits in two in a magnetic field [there were two distinct spots on the screen,
instead of a smear of silver along a line]. Uhlenbeck & Goudsmit in 1925 pro-
posed that silver atoms and the electron have an intrinsic angular momentum,
the spin. The spin interacts with the magnetic field, and the electron goes up
or down according as the spin is parallel or opposite to the vertical magnetic
field.
1 This holds in a conservative system.
2 The SchrGdinger equation arose out of the hypothesis that if light has both wave and
particle properties, then perhaps particles might have wave properties such as interference
and diffraction.
3 This is the Born interpretation.
Pauli Spin Matrices and Spinors 51
In an electromagnetic field l?,Jwith potentials V, A the Schrlidinger equa-
tion becomes
ih-84 = --[(-ihV
1 - eA)']$ - eV$,
dt 2m
or after 'squaring'
iti-a* = -[-h2v2
1
+
e 2 ~ ' ihe(V . A + V)]$ - eV$. +A.
dt 2m
This equation does not yet involve the spin of the electron. The differential
operator, known as the generalized momentum,
d
n'= 5- eA where 3 = -ihV

is such that its components nk = pk - eAk satisfy the commutation relations
nln2 - n2nl = itieB3 (permute 1 , 2 , 3 cyclically).
Pauli 1927 introduced the spin into quantum mechanics by adding a new
term into the Schrijdinger equation. The Pauli spin matrices
satisfy
ulu2 = iu3 (permute 1,2,3 cyclically)
and the anticommutation relations
Applying the above commutation and anticommutation relations, and tem-

porarily using the old-fashioned notation
we may see that

( 2 . q2= n2 - tie(:. J)
where
n 2 -- p 2 + e 2 ~-2 ,(a. A + A . a ) .
Pauli replaced n2 by (a' . ii)' in equation (1):
4 A Schrijdinger equation with W =0 is brought into this form by a gauge transformation

+(?, t ) -,q(?, t)eia('lt), when eV = h-
aff
and e A = hVa.
dt
52 Pauli Spin Matrices and Spinors
This Schrijdinger-Pauli equation describes the spin by virtue of the term
he
-(.'
-
- B).
2m
+
The matrix a'. B operates on two-component column matrices with entries in
C. The wave function sends space-time points to Pauli spinors
that is, it has values in the complex linear space C2.

The SchrGdinger-Pauli equation in the Clifford algebra C&. The mul-
tiplication rules of the Pauli spin matrices al, a 2 , a3 E Mat(2, C) imply the
matrix identity
(.'.I?)~ = (B;+ B; + B:)I.
Thus, we may regard the set of traceless Hermitian matrices as a Euclidean
space R3 with an orthonormal basis { a l l a 2 , a3).
The length (of the representative) of a vector 2 is preserved under a simi-
larity transformation U ( . ' . ~ ) U - ' by a special unitary matrix U E SU(2),
SU(2) = {U E Mat(2, C) I U ~ U
= I, det U = 1).
In this way, not only vectors but also rotations become represented within the
matrix algebra Mat(2,C). In fact, each rotation R E SO(3) becomes repre-
sented by two matrices fU E SU(2), and we say that SU(2) is a two-fold
covering of SO(3) :
Pauli spinors could also be replaced by square matrices with only the first
column being non-zero,
Such square matrix spinors form a left ideal S of the matrix algebra Mat(2, C),
that is, for U E Mat(2, C) and $ E S we also have U$ E S.
The matrix algebra Mat(2, C) is an isomorphic image of the Clifford algebra
CC3 of the Euclidean space R3. Thus, not only vectors in R and rotations in
5 The left ideal can be written as S = Mat(2,C) f , where f = ;(I + as) is an idempotent
satisfying f 2 = f . The idempotent is primitive and the left ideal is minimal.
4 .I Orthogonal unit vectors, orthonormal basis 53
SO(3) have representatives in Ct3, but also spinor spaces or spinor represen-
tations of the rotation group SO(3) can be constructed within the Clifford
algebra Ce3.
In the notation of the Clifford algebra Ce3 we could describe Pauli's achieve-
+
ment by saying that he replaced ?r2 = ii . ii by ii2 = ii- ii ii A ii= ?r2 - h e 2
and came across the equation
where E R3 c Ce3 and $ ( F , t) E S = Ce3 f , f = $(l+e3). All the arguments

and functions now have values in one algebra, which will facilitate numerical
computations.
In this chapter we shall study more closely the Clifford algebra Cea and the
spin group Spin(3), and reformulate once more the Schrodinger-Pauli equation
in terms of Ct3.
4.1 Orthogonal u n i t vectors, orthonormal basis

The 3-dimensional Euclidean space R3 has a basis consisting of three ortho-
gonal unit vectors el, e2, e3. The Clifford algebra Ce3 of IR3 is the real asso-
ciative algebra generated by the set {el, e 2 , e3) satisfying the relations
The Clifford algebra Ce3 is 8-dimensional with the following basis:

1 the scalar
el, e 2 , e3 vectors
el e g , ele3, e g e 3 bivectors
8 1 ~ 3 a volume element.
We abbreviate the unit bivectors as eij = eiej, when i # j, and the unit
oriented volume element as el23 = ele2e3. An arbitrary element in Ct3 is a
sum of a scalar, a vector, a bivector and a volume element, and can be written
+ + +
as a a be123 Belz3, where a,,B E R and a, b E R3.
Example. Compute the product elze13. By definition elgel3= (ele2)(eles)
6 Actually, spinor representations are representations of the universal covering group
S U ( 2 ) 2 S p i n ( 3 ) of S O ( 3 ) . The spinor representations cannot be reached by tensor
methods, as irreducible components of tensor products of antisymmetric powers of R3.
7 The orthogonal group O ( 3 ) also has a non-trivial covering group Pin(3) residing within
ces.
and by associativity (ele2)(ele3) = ele2ele3. Use anticommutativity, ele2 =

-e2el, and substitute e: = 1 to get elezele3 = -efe2e3 = -e23. I
I m a g i n a r y units. The three unit bivectors elea, eleg, e ~ e 3represent unit
oriented plane segments as well as generators of rotations in the coordinate
planes, and share the basic property of the imaginary unit, (eiej)2 = -1 for
i # j. The oriented volume element ele2e3 also shares the basic property of
the imaginary unit, (ele2e3)2= -1, and furthermore it commutes with all
the elements in CCs The unit oriented volume element ele2e3 represents the
duality operator, which swaps plane segments and line segments orthogonal to
the plane segments. I
4.2 M a t r i x representation of Cks

The set of 2 x 2-matrices with complex numbers as entries is denoted by
Mat(2, C). Mostly we shall regard this set a s a real algebra with scalar multi-
plication taken over the real numbers in R although the matrix entries are in
the complex field C. The Pauli spin matrices
satisfy the multiplication rules

,2- - a2 2 --1
2 -- u3 and
ala2 = ia3 = -'72'71,
'73'71 = 2'72 = -'71'73,
6 2 ' 7 3 = 2261 = -U3U2.
They also generate the real algebra Mat(2, C). The correspondences el N al,
e2 N '72, e3 N '73 establish an isomorphism between the real algebras, CC3 N
Mat(2, C), with the following correspondences of the basis elements:
Mat (2, C) ce3
Note that e,j = -eji for i # j. The essential difference between the Clifford
algebra CC3 and its matrix image Mat(2, C) is that in the Clifford algebra CC3
we will, in its definition, distinguish a particular subspace, the vector space R3,
4.3 The center of C13 55
in which the square of a vector equals its length squared, that is, r2= (rI2.No
such distinguished subspace has been singled out in the definition of the matrix
algebra Mat(2, C). Instead, we have chosen the traceless Hermitian matrices
to represent R3, and thereby added extra structure to Mat(2, C).
4.3 T h e center of C&

The element e l 2 3 commutes with all the vectors el, e2, e3 and therefore with
every element of el3. In other words, elements of the form
commute with all the elements in C&. The subalgebra of scalars and 3-vectors
3
R d A F L 3 = {x+ye123 12, Y €8)
is the center Cen(C13) of C13, that is, it consists of those elements of C13 which
commute with every element of C13. Note that ulu2u3 = iI. Since ef23= -1,
the center of C13 is isomorphic to the complex field @, that is,
3
Cen(C13) = R e A R3 2 C.
4.4 T h e even subalgebra

The elements 1 and e l 2 = elen, el3 = 6183, e23 = e2e3 are called even,
because they are ~roductsof an even number of vectors. The even elements
are represented by the following matrices:
w+iz ix+y
W + ze23 + ye31 + zel2 E ix-y w-iz
The even elements form a real subspace
2
~ e ~ ~ ~ = { w + x e 2 3 + y e 3 l + zI W
e i, zX , Y , Z E R )
21 {wI + xiul + yiu2 + ziu3 1 w, x, y, z E R)
8 We could also have chosen, for the representatives of the anticommuting (and therefore
orthogonal) unit vectors in R3, the following matrices:
+
that is, u1 = 3(5u1 3u1u2), 212 = 62, 213 = i(5u3 - 3 0 ~ 0 3 ) . These matrices are
+
non-Hermitian and satisfy uluk ukul = 2Slk1.
which is closed under multiplication. Thus, the subspace R $ R3 is a sub-
algebra, called the even subalgebra of C13. We will denote the even subalgebra
by even(Cl3) or for short by ~ l ; . The even subalgebra is isomorphic to the
division ring of quaternions W, as can be seen by the following correspondences:
R e m a r k . The Clifford algebra Ct3 contains two subalgebras, isomorphic to @

[the center] and W [the even subalgebra], in such a way that [temporarily we
denote these subalgebras by their isomorphic images]
1. a b = b a f o r a E C a n d b E W ,
2. C13 is generated as a real algebra by @ and W,
3. (dim @) (dim W) = dim C& .
These three observations can be expressed as
4.5 Involutions of Ct3

The Clifford algebra Ct3 has three involutions similar to complex conjugation.
Take an arbitrary element
written as a sum of a scalar (u)o, a vector ( U ) ~a, bivector ( u ) ~and a volume

element ( u ) ~ We
. introduce the following involutions:
ii = ( u ) - +
~ (u)~ ( u )-~ ( u ) ~ , grade involution,
+
6 = (U)O ( u ) -~ ( u )-~ ( u ) ~ , reversion,
U = (u)o - (u)i - ( u ) +
~ ( u ) ~ , Clifford-conjugation.
Clifford-conjugation is a composition of the two other involutions: ii = u" =
- A
U.
The correspondences ol E e l , a 2 2 ez, 03 2. e3 fix the following represen-
tations for the involutions:
4.6 Reflections and rotations 57
where the asterisk denotes complex conjugation. We recognize that the reverse
fi is represented by the Hermitian conjugate ut and the Clifford-conjugate G
by the matrix u-' det u E Mat(2, R) [for an invertible u].
The grade involution is an automorphism, that is,
while the reversion and the conjugation are anti-automorphisms, that is,
The grade involution induces the even-odd grading of Ct3 = ~ t $$Ct; .

The reversion can be used to extend the norm from R3 to all of Ct3 by
setting
lu12 = (ufi)o.
The norm of
can be obtained from
luI2 = Iu012 + I ~ 1 +1 ~ + 1 ~ 3 +1 ~1 ~ 1 2 +1 ~1~131+

1.11~1~
2
+l
1 ~ 2 3 1 ~ ~ ~
The norm satisfies the inequality
I U V -<~ ~ ~ I U I I V I for U, v E Ct3.

The conjugation can be used to determine the inverse
of u E Ct3, ufi # 0. The element ufi = Gu is in the center R $ /\3 R3 of Cts,

so that division by it is unambiguous.
4.6 Reflections a n d rotations

In the Euclidean space IR3 the vectors r and ara-l = 2 ( a . r)av1- r are
symmetric with respect to the axis a [use the definition of the Clifford product,
ar + ra = 2 a . r]. The opposite of ara-l, the vector
is obtained by reflecting r across the mirror perpendicular to a [reflection

across the plane aelZ3].
Pauli Spin Matrices and Spinors
Two successive reflections in planes perpendicular to a and b result in a

rotation r + bara-'b-' around the axis which is perpendicular to both a
+
and b. Indeed, r can be decomposed as r = rll r l where rll and r l are
parallel and perpendicular, respectively, to the plane of a and b. The per-
pendicular component r l remains invariant under both the reflections while
the two successive reflections together rotate the parallel component rll in the
plane of a and b by twice the angle between a and b.
+ +
Consider a vector a = ale1 a2e2 ase3 and the bivector ae123 = ale23 +
+
a2e3l asel2 dual to a. The vector a has positive square
a2 = laI2, where la1 = Ja: + a: + a:,

but the bivector ael23 has negative square
2
(ae123)2= -la[ .
It follows that
exp(ael23) = cos a + e123-a sin
ff
where a = la1 A spatial rotation of the vector r = xel + ye2 + ze3 around
the axis a by the angle a is given by
The sense of the rotation is clockwise when regarded from the arrow-head of
a. The axis of two consecutive rotations around the axes a and b is given by
the Rodrigues formula
C =
a' + b' + a' x b' where al = -at a n - .a
1 - a' . b' CY 2
4.7 The group Spin(3)
This result is obtained by dividing both sides of the formula
by their scalar parts and then by inspecting the bivector parts.
4.7 The g r o u p Spin(3)

The Clifford algebra Ce3 of R3 can be employed to construct the universal
covering group for the rotation group SO(3) of R3. A vector x E R3 can be
rotated by the formula
R3 + R3, x -,p(s)x = sxs-l
where s is an element of the group
Spin(3) = {S E Ce3 I as = 1, ss = 1).
The group Spin(3), called the spin group, is a two-fold covering group of the
rotation group SO(3).
In the matrix formulation provided by the Pauli spin matrices, the spin group
Spin(3) has an isomorphic image, the special unitary group
SU(2) = { S E Mat(2,C) I s t s = I, d e t s = 1).
For an element s E SU(2) the function x + p(s)x = sxst is a rotation of the
Euclidean space of traceless Hermitian matrices,
Every element in SO(3) can be represented by a matrix in SU(2). There are

two matrices s and -s in SU(2) representing the same rotation R = p ( f s) E
SO(3). In other words, the group homomorphism p : Spin(3) + SO(3) is
surjective with kernel {f 1). This can be depicted by a sequence
which is exact, that is, the image of a homomorphism coincides with the kernel
of the successive homomorphism.
The spin group Spin(3) is a universal cover of the rotation group S 0 ( 3 ) ,
that is, the Lie group Spin(3) is simply connected. The group SO(3) is
doubly connected. l o
9 A Lie group is simply connected if it is connected and evely loop in the group can be
shrunk to a point.
LO Rotations in S O ( 3 ) can be represented by vectors a E R3, la1 5 a. Each rotation, la1 < a,
has a unique representative, and each half-turn, la1 = a, is represented twice, fa. A loop
connecting the identity and a half-turn does not shrink to a point.
4.8 P a u l i spinors
In the non-relativistic theory of the spinning electron one considers column
matrices, the Pauli spinors
An isomorphic complex linear space is obtained if one replaces Pauli spinors

by the square matrix spinors
where only the first column is non-zero. The fact that only the first column is
non-zero can be expressed as
$EMat(2,C)f where f = (i :).

We shall regard the correspondences e l N ul, e2 N u2, e3 cz u3 as an
identification between Ct3 and Mat(2, C). If we multiply $ E Mat(2, C)f on
the left by an arbitrary element u E Ct3 = Mat(2, C), then the result is also of
the same type:
$1 0
Such matrices, with only the first column being non-zero, form a left ideal S
of Ct3, that is,
u$ES forall u E C t 3 and $EScC&.
This left ideal S of Ct3 contains no left ideal other than S itseif and the zero
ideal (0). Such a left ideal is called minimal in Ct3.
As a real linear space, S has a basis { fo, fi,f 2 , f3) where
The element f = fo is an idempotent, that is, f2 = f.

4.8 Pauli spinors
The subset
of Ce3 is a subring with unity f , that is, af = f a for a E IF. None of the
elements of IF is invertible as an element of Ces, but for each non-zero a E IF
there is a unique b E IF such that ab = f . Thus, IF is a division ring with
-
unity f [this follows from the idempotent f being primitive in Ce3]. As a 2-
dimensional real division algebra IF must be isomorphic to C. The isomorphism
IF C is seen by the equation fi = -f0 relating the basis elements i f o , f3)
of the real algebra IF.
C o m m e n t . The multiplication of an element $ of the real linear space S on
the left by an arbitrary even element u E c@, expressed in coordinate form in
the basis {fo, f l , f2, f3),
corresponds to the matrix multiplication
The square matrices corresponding to the left multiplication by even elements

constitute a subring of Mat(4,R); this subring is an isomorphic image of the
quaternion ring W. I
The minimal left ideal
has a natural right IF-linear structure defined by
We shall provide the minimal left ideal S with this right IF-linear structure,
and call it a spinor space. ''
The map Ce3 + Endr S, u + ~ ( u ) where , T(U) is defined by the relation
T(u)$ = u$, is a real algebra isomorphism. Employing the basis i f o , -f2) for
the IF-linear space S, the elements r ( e l ) , 7(e2),r(e3) will be represented by
the matrices ul ,u2, u3. In this way the Pauli matrices are reproduced.
11 Note that multiplying a matrix $ in S, a left ideal, on the left by X E F does not result
in a left F-linear structure.
There is a natural way to introduce scalar products on the spinor space
S c Ct3. First, note that for all $, cp E S the product
falls in the division ring IF (z + z* means complex conjugation). To show that

the map
s x s + p , ($,cp)+4cp
defines a scalar product we only have to verify that the reversion 11, + is 4
a right-to-left F-semilinear map. For all y5 E S, X E F we have ($A)" = i q
where the map X + is an anti-involution of the division algebra IF (actually
complex conjugation).
Multiplying a spinor +
E S C Ct3 by an element s E Ct3 is a right IF-
linear transformation S + S, $ + s$. The automorphism group of the scalar
product is formed by those right F-linear transformations which preserve the
scalar product, that is,
(s+)-(sp) = &o for all $, cp E S.

The automorphism group of the scalar product 4cp
is seen to be the group
{s E Ct3 I as = 1) which is isomorphic to the group of unitary 2 x 2-matrices,
We can also use the Clifford conjugate u + C of Ct3 to introduce a scalar

product for spinors. In this case, the element
does not appear in the division ring?!I = fCt3f. However, we can find an
invertible element a E Ct3 so that a& E IF, e.g. a = el or a = esl. The map
S x S + IF, ($, cp) + a4cp
defines a scalar product. Writing
we find that a& .Y ~ ( $ ) ~ J r ( c pHence,

). the automorphism group { s E Ct3 I
Ss= 1) of the scalar product a6cp is the group of symplectic 2 x 2-matrices,
Sp(2, C) = { s E Mat(2, @) I s T ~ =
s J).
4.9 Spinor operators 63
4.9 S p i n o r o p e r a t o r s
Up till now spinors have been objects which have been operated upon. Next we
will replace such passive spinors by active spinor operators. Instead of spinors
in minimal left ideals we will consider the following even elements:
also computed as 4 = ++4

for Ijl E Ce3f . Classically, the expectation values
of the components of the spin have been determined in terms of the column
spinor 11, E C2 by computing the following three real numbers:
In terms of $ E Ce3f this computation could be repeated as
However, in terms of Q E ~ e wei may compute s = slel+ s 2 e 2 + s3es directly
Since 4 acts here like an operator, we call it a spinor operator. It should be

emphasized that not only did we get all the components of the spin vector s
at one stroke, but we also got the entity s as a whole.
R e m a r k . The mapping ~1: + R3, 4 + Qu3Qt = Qe34 is the Ii'S-
transformation (introduced by Kustaanheimo&Stiefel 1965) for spinor regu-
larization of Kepler motion, and its restriction to norm-one spinor operators
Q satisfying QQ = 1 (or equivalently QQt = I) results in a Hopf fibration
S3 + S2 (the matrix 9 u 3 9 t is both unitary and involutory and represents a
reflection of the spinor space with axis $).
The above mapping should not be confused with the 'Cartan map', see Car-
tan 1966 p. 41 and I<eller &Rodriguez-Romo 1991 p. 1591. A 'Cartan map'
C2 x C2 + Cl3, ($, 9 ) + 2$el(p, where C2 = C&f, sends a pair of square
matrix spinors to a complex 4-vector xo x , +
When 11, = (o, x 2 = 0. I

Note also that trace($$t) = 2($4)0 = 44 which equals Q% = det(4).
In operator form the Schrodinger-Pauli equation
shows explicitly the quantization direction e3 of the spin. The explicit occur-
rence of e3 is due to the injection C2 + C13f , $ + @; technically 2 even(l?$) =
h e 3 . If we rotate the system 90' around the y-axis, counter-clockwise as seen
from the positive y-axis, then vectors and spinors transform to
K
B'/ = ugu-l 9' = u@ where u = exp(-els),
and
4
and the Pauli equation transforms to
If this equation is multiplied on the right by u-l, then e3 goes to el = ue3u11,

and the equation looks like
ih-
a@" = -nt2@"
1 he
- -B -1 11
@ el - eV@I1,
dt2m 2m
where 9" = u@u-l. Both the transformation laws give the same values for
observables, that is, @'e3$' = @"el$".
Exercises
1. +
Compute the square of a belz3 where a, b E R3.
2. Compute p2, q2 and pq for p = + ( I +e3) and q = $(I -es).
3. + ;
Compute the squares of i ( 1 e3) f (1 - es)e12.
4. +
Find all the four square roots of cos p e l 2 sin p. Hint: elze3 = esel2.
5. Find the exponentials of f5(1 - e3)e12.Hint: e l 2 and el23 commute [or
q = i ( 1 - e3) is an idempotent satisfying q2 = q].
+ + +
6. Let u = a a bela3 /?elas [a,/? E R and a , b E R3]. Compute ufi.
+ +
7. Find the inverse of u = o a belz3+ Bela3. Hint: ufi is of the form
+
z yelas, x, Y E R.
8. Find the exponential of u = a + a + belz3+ /?elas. Hint: compute
+
(a be123)'.
9. Show that each non-zero even element in ~i?:is invertible.
10. Show that uii E R $ R 3 for all u 6 C&.
11. Show that JuaiiJ= J u J 2 ) afor
J a E R3, u E R $ /\'R3.
12. Show that the norm on C13, defined by luI2 = ( ~ i i ) agrees
~, with the
4.9 Spinor operators 65
norm given by I U =~ <u,
~ u> where the symmetric bilinear product is
determined by
<a,p>=ap for a , P E R ,
<a,b>=a.b for a , b € R 3
and by
Xl.yl ... X1 ' Yk
<xlA ... AXk,ylA ... A Y k > =

xk ' y 1 . . . xk yk'
in l\kIR3, k 2 2. [One also needs to assume orthogonality of the

components in Ce3 = R $ IW3 $ /\2 R3 $ /\3 R3.]
13. Show that the reflection across the plane of the bivector A is obtained by
+ r1= - A ~ A - ~ .
14. Let x , y , z E R3. Compute (xyz)l and (xyz)3. Hint: use reversion.
Solutions
+ +
1. (a be123)~= a . a - b . b 2 ( a . b)eI23.
2. p2 = p and q2 = q , that is, p and q are idempotents; and pq = 0 [and so
there are zero-divisors in the Clifford algebra Ce3].
3. e3 [this shows that vectors can have square roots].
4. f(cos 5+ el2 sin z), fe3(cos 5+ el2 sin 5 ) .
5 . e3 [this shows that vectors also have logarithms].
6 . a 2-P2 - a . a + b . b + 2 ( a p - a . b ) e l Z 3 .
8. Denote r = - J +
E IR $ /\3R3, v = (a belZ3)/r, v2 = 1. Then
+ +
exp(u) = exp(a pe123)[+(I+ v) exp(r) ;(I - v) exp(-r)] when r # 0.
+
When r = 0: exp(u) = exp(a ,Be123)(l a belz3). + +
10. u = a + a + b e 1 2 3 + P e l z 3 , u 3 = a 2 + P 2 + a 2 + b 2 + 2 ( a a + p b + a x b )
which is in IW $ IR3. Direct proof:
which implies u3 E R $ R3, since the reversion sends bivectors and

3-vectors to their opposites.
13. Decompose r into components parallel, rll, and perpendicular, r l , to A ,
and note that A anticommutes with vectors in its plane,
+ +
A(rll rl) = (-rll rl)A. Then
+
A(rll r1)Ad1 = (-rll + rL)AA-' = -rl.
14. First, (xyz)- = zyx and (xyz)- = ( x y z ) ~- ( x y z ) ~ Therefore,
.
+
(xyz)l = $(xyz zyx) and ( X Y Z ) ~= $(xYz- ZYX), and also
+
(xyz)l = (y . z)x - (z . x)y (x . y)z and ( X Y Z ) ~= x A y A Z .
Bibliography
W. Baylis: Theoretical Methods in the Physical Sciences: an Introduction to Problem
Solving with MAPLE V . Birkhauser, Boston, MA, 1994.
H.A Bethe, E.E. Salpeter: Quantum Mechanics of One- and Two-Electron Atoms.
Springer, Berlin, 1957.
E. Cartan: The Theory of Spinors. The M.I.T. Press, Cambridge, MA, 1966.
A. Charlier, A. Bkrard, M.-F. Charlier, D. Fristot: Tensors and the Clifiord A!gebra,
Applications to the Physics of Bosons and Fermions. Marcel Dekker, New York,
1992.
R.C. Feynman, R.B. Leighton, M. Sands: The Feynman Lectures on Physics, Vol. III,
Quantum Mechanics. Addison-Wesley, Reading, MA, 1965.
D. Hestenes: Space-Time Algebra. Gordon and Breach, New York, 1966, 1987, 1992.
J. Keller, S. Rodriguez-Romo: Multivectorial generalization of the Cartan map. J .
Math. Phys. 32 (1991), 1591-1598.
E. Merzbacher: Quantum Mechanics. Wiley, New York, 1970.
W. Pauli: Zur Quantenmechanik des magnetischen Elektrons. Z. Physik 4 3 (1927),
601-623.
M. Riesz: Clifford Numbers and Spinors. The Institute for Fluid Dynamics and Ap-
lands, 1993.
Quaternions
We saw in the chapter on Complex Numbers that it is convenient to use the real
algebra of complex numbers C to represent the rotation group SO(2) of the
plane R2. In this chapter we shall study rotations of the 3-dimensional space
R3. The composition of spatial rotations is no longer commutative, and we
need a non-commutative multiplication to represent the rotation group SO(3).
This can be done within the real algebra of 3 x 3-matrices Mat(3, R), or by
the real algebra of quaternions, W,invented by Hamilton.
The complex plane @ is a real linear space lR2, and multiplication by a
+
complex number c = a ib, that is, the map C + C,z -+ cz, may be regarded
as a real linear map with matrix ( :b) operating on () in R'. The
complex plane is also a real quadratic space R210,in short R2, with a quadratic
form
C+R, z = x + i y + z ~ = x ~ + ~ ~ ,
and norm (z(= m. Multiplication of complex numbers preserves the norm,
that is, lczl = lcllzl for a11 c, z E @, and so multiplication by c is a rotation of
R2 if, and only if, Ic( = 1. Conversely, any rotation of R2 can be represented
by a unit complex number c, Icl = 1, in C.The unit complex numbers form a
group
-
U(1) = {cE @ 1 cF= 11,
called the unitary group, which is isomorphic to the rotation group SO(2) =
{U E Mat(2,lR) 1 U ~ = U I, det U = I), that is, U(l) SO(2). The unitary
group U(l) can be visualized as the unit circle
S1 = {x + iy E C 1 x2 + Y2 = 1)
Quaternions
of the complex plane C.
Similarly, the algebra of quaternions W may be used to represent rotations

of the 3-dimensional space JR3. It will turn out that quaternions are also con-
venient to represent the rotations of the Cdimensional space JR4.
Quaternions as hypercomplex numbers

Quaternions are generalized complex numbers of the form q = w ix j y k z + + +
where w, x, y, z are real numbers and the generalized imaginary units i, j , k
satisfy the following multiplication rules:
2.2 --3 .2 - - k2 -
- -1 ,
..
i j = k = -32, jk = i = -kj, ki = j = -ik.
Note that the multiplication is by definition non-commutative. One can
show that quaternion multiplication is associative. The above multiplication
rules can be condensed into the following form:
2' 2 -
- 3 ' 2 --k 2 = i j k = -1
where in the last identity we have omitted parentheses and thereby tacitly
assumed associativity.
The generalized imaginary units will be denoted either by i , j , k or by i ,j, k.
They have two different roles: they act as generators of
rotations, that is, they are bivectors, or
translations, that is, they are vectors.
This distinction is not clear-cut since bivectors are dual to vectors in JR3
5.1 Pure part and cross product 69
5.1 P u r e p a r t a n d cross product
+
A quaternion q = w + ix jy + kz is a sum of a scalar and a vector, called
the real part, Re(q) = w E J R , and the pure part, Pu(q) = ix + jy + kz E R3.
The quaternions form a 4-dimensional real linear space W which contains the
real axis R and a 3-dimensional real linear space R3 so that MI = R $ JR3. We
+
denote the pure part also by a boldface letter so that q = qo q where qo E R
and q = iql +jqg+kq3 E R3. The real linear space R@R3 with the quaternion
product is an associative algebra over R called the quaternion algebra W. The
+ +
product of two quaternions a = a0 a and b = bo b can be written as
+
A quaternion q = qo q is pure if its real part vanishes, qo = 0, so that
q = q E R3. A product of two pure quaternions a = ial jag + +
ka3 and
+ +
b = ibl j b g kbg is a sum of a real number and a pure quaternion:
+ +
where we recognize the scalar product a . b = albl a2b2 a3b3 and the cross
+ +
product a x b = i(a2b3- a3b2) j(asbl - alb3) k(alb2 - a2bl).
The vector space R3 with the cross product a x b is a real algebra, that is,
it is a real linear space with a bilinear map
Kt3 x Ik3 + Kt3, (a, b ) + a x b.
The cross product satisfies two rules
the latter being called the Jacobi identity; this makes R3 with the cross product
a Lie algebra. In particular, the cross product is not associative, a x (b x c) #
(a x b) x c.
We can reobtain the cross product of two pure quaternions a, b E R3 as the
pure part of their quaternion product: a x b = Pu(ab).
5.2 Quaternion conjugate, n o r m a n d inverse

The conjugate of a quaternion q = w + ix +jy + kz is obtained by changing
the sign of the pure part:
We shall also refer to q as the quaternion conjugate of q. The conjugation is

an anti-automorphism of W; = 6ii for a, b E W.
70 Quaternions
A quaternion q multiplied by its conjugate g results in a real number qg =
+ + + + + +
w 2 x2 y2 z 2 called the square norm of q = w ix jy k z . The norm
191 of q is given by Iql = fi so that
The norm of a product of two quaternions a and b is the product of their

norms - as an equation, lab1 = lallbl for a , b E W - which turns W into a
normed algebra.
The inverse q-I of a non-zero quaternion q is obtained by q-' = q/lqI2 or
more explicitly by
1 - w-ix-jy-kz
-
w+ix+jy+kz w ~ + x ~ + Y ~ + z ~ '
In particular, ab = 0 implies a = 0 or b = 0, which means that the quaternion
algebra is a division algebra (or that the ring of quaternions is a division ring).
5.3 T h e center o f W
The set of those elements in W which commute with every element of W forms
the center of W,
Cen(W) = { w E W I wq = qw for all q E W).
The center is of course closed under multiplication. The center of the division
ring W is isomorphic to the field of real numbers R. In contrast to the case
of the complex field @, the real axis in W is the unique: subfield which is the
center of W.
5.4 R o t a t i o n s i n t h r e e dimensions
Take a pure quaternion or a vector
r = i x + j y + k z ~ R 3 , where I H = R @ R 3 ,
+
of length Irl = d x 2 + y2 2 2 . For a non-zero quaternion a E W, the expression
ara-l is again a pure quaternion with the same length, that is,
ara-' E R3 and lara-'1 = lrl.
In other words, the mapping
R3-+R3, r+araP1
is a rotation of the quadratic space of pure quaternions R3. Each rotation in
SO(3) = {U E Mat(3,R) I U ~ = U I , det U = 1) can be so represented,
5.5 Rotations in four dimensions 71
and there are two unit quaternions a and -a representing the same rotation,
ara-' = (-a)r(-a)-'. In other words, the sphere of unit quaternions,
is a two-fold covering group of S 0 ( 3 ) , that is, SO(3) E S3/{f 1).

A rotation has three parameters in dimension 3. In other words, SO(3) and
S3 are 3-dimensional manifolds. The three parameters are the angle of rotation
and the two direction cosines of the axis of rotation.
To find the axis of this rotation we take a unit quaternion a , la1 = 1, and write
it in the form a = eaI2 where a E R3. Note that
a a . a
2 a
+
eaJ2 = cos - - sin -
2
where a = la[. The rotation r + ara-' turns r about the axis a by the
angle a . The sense of the rotation is counter-clockwise when regarded from
the arrow-head of a .
The composite of two consecutive rotations, first around a by the angle
a = la1 and then around b by the angle P = Ibl, is again a rotation around
some axis, say c. The axis of the composite rotation can be found by inspection
of the real and pure parts of the formula e C l 2 = ebJ2eaf2. Divide both sides by
their real parts and substitute
c Y
C' = -tan -, where 7 = lcl,
Y 2
to obtain the Rodrzgues formula
5.5 R o t a t i o n s i n f o u r dimensions
The mapping W + W,q + aqb-', where a , b E W are unit quaternions la1 =
Ibl = 1, is a rotation of the 4-dimensional space R4 = W. In other words, the
real linear mapping
W -,W, q -,aqb-l, where a , b E W and la1 = lbl = 1,
is a rotation of R4. Each rotation in SO(4) can be so represented, and there
are two elements (a, b) and (-a, -b) in S3x S3 representing the same rotation,
that is, aqb-I = (-a)q(-b)-l. In other words, the group S3x S3 is a two-fold
covering group of S 0 ( 4 ) , that is,
72 Quaternions
A rotation in dimension 4 can be represented by a pair of unit quaternions,
and so it has six parameters, in other words, dimSO(4) = dim(S3 x S3)= 6.
A rotation has two completely orthogonal invariant planes; both the invariant
planes can turn arbitrarily; this takes two parameters. Fixing a plane in R4
takes the remaining four parameters: three parameters for a unit vector in
S3, plus two parameters for another orthogonal unit vector in S2,minus one
parameter for rotating the pairs of such vectors in the plane.
5.6 M a t r i x representation of quaternion multiplication

The product of two quaternions q = w+ix+jy+kz and u = uo+iul+jug+ku3
can be represented by matrix multiplication:
where qu = v. Swapping the multiplication to the right, that is, uq = v', gives
a partially transformed matrix:
Let us denote the above matrices respectively by L, and R,, that is,
L,(u) = qu (= v) and R,(u) = uq (= v').

We find that
LiLjLk=-I and R i R j R k = I .
The sets {L, E Mat(4,IR) I q E W) and {R, E Mat(4, R) I q E W) form two
subalgebras of Mat(4, IR), both isomorphic to W. For two arbitrary quaternions
a , b E W these two matrix representatives commute, that is, LaRb = RbL,. Any
real 4 x 4-matrix is a linear combination of matrices of the form LORb.The
above observations together with (dimW)2 = dimMat(4,R) imply that
Mat(4,IR) 2: W 8 W,
or more informatively Mat(4, R) = W @I W* .
1 Note that R'RTR: = -I.
2 For unit quaternions a , b E W such that J a J= JbJ= 1 we may choose La E Q and Rb E Q *
or alternatively La E Q* and Rb E Q. For a discussion about the meaning of Q and Q * ,
see the chapter on The Fourth Dimension.
5.7 Linear spaces over W 73
Take a matrix of the form U = La Rb in Mat(4,R). Then U T u = la121b121,
+
but in general U uT # CYI. Take a matrix of the form V = La Rb in +
+ +
Mat(4, R). Then V vT = 2(Re(a) Re(b))I, but in general V ~ # V PI.
+
Conversely, if U E Mat(4, R) is such that U uT = CYIand uTu = PI then
the matrix U belongs either to W or to W*.
Besides real 4 x 4-matrices, quaternions can also be represented by complex
2 x 2-matrices:
w+iz ix+ y
w+iz+jy+kz~
is- y w-iz
The orthogonal unit vectors i , j, k are represented by matrices obtained by
multiplying each of the Pauli matrices u l , a 2 , us by i = fl:
5.7 Linear spaces over W

Much of the theory of linear spaces over commutative fields extends to W.
Because of the non-commutativity of W it is, however, necessary to distinguish
between two types of linear spaces over W, namely right linear spaces and left
linear spaces.
A right linear space over W consists of an additive group V and a map
such that the usual distributivity and unity axioms hold and such that, for all
A, p E W and x E V,
(xX11.1= x(Xp).
A left linear space over W consists of an additive group V and a map
such that the usual distributivity and unity axioms hold and such that, for all
A, p E W and x E V,
X(px) = ( X P ) ~ .
A mapping L : V - Uibetween two right linear spaces V and U is a right
linear map if it respects addition and, for all x E V, X E W, L(xX) = (L(x))X.
C o m m e n t . In the matrix form the above definition means that

74 Quaternions
R e m a r k . Although there are linear spaces over W, there are no algebras

over W, since non-commutativity of W precludes bilinearity over W: X(xy) =
(XX)Y # (xX)Y = ~ ( X Y )# ~ ( Y X )= ( x Y ) ~ . I
5.8 Function theory of quaternion variables

The richness of complex analysis suggests that there might be a function theory
of quaternion variables. There are several different ways to generalize the
theory of complex variables to the theory of quaternion functions of quaternion
variables, f : W + W. However, many generalizations are uninteresting, the
classes of functions are too small or too large. In the following we will first
eliminate the uninteresting generalizations.
First, consider quaternion differentiable functions such that
fl(c) = fl$f(u + h) - f(q)lh--l, where q, h E W,
exists. The derivative f'(q) is a real linear function

R4 + R4 : h -+ fl(q)h
corresponding to multiplication by a quaternion a E W on the left, jl(q)h = ah
for h E W = R4.However, since ah # ha the only quaternion differentiable
functions are the affine right W-linear functions
f (q) = aq +b where a, b E W.
We conclude that the set of quaternion differentiable functions reduces to a
small and uninteresting set.
+ + +
Second, if we consider power series in a quaternion variable q = w ix j y
kz, then we get the set of all power series in the four real variables w, x, y, z.
For instance, the coordinates are first-order functions
w = +(q - iqi - jqj - kqk),
+ + kqk)i-',
x = :(q - iqi jqj
y = ;(q + + kqk)j-l,
iqi - jqj
= +(q+iqi+jqj-kqk)k-l,
and so the set of power series in q, with left and right quaternion coefficients,
is the set of all power series in the real variables w, x, y, z. This set is too big
to be interesting.
Third, we could consider power series in q with real coefficients, that is,
+ +
functions of type f(q) = a0 + alq a2q2 ... where ao, a l , a2, ... are real.
5.8 Function theory of quaternion variables 75
Restrict such a function to the complex subfield @ C W, and send z = x iy +
to f (z)= u+iv, where u = u(x, Y) and v = v(x, Y). Decompose the quaternion
+
q into real and vector parts, q = qo q , and note that 9/lql is a generalized
imaginary unit, (q/lq()2 = -1. Then
So this generalization just rotates the graph of C + C, z + f(z), or rather

makes i = i sweep all of S2 = {r E R3 I Irl = 11, and thus gives only (a
subclass of) axially symmetric functions.
Fourth, we could consider functions which are conformal almost everywhere
in R4. This leads to Mobius transformations of R4, or its one-point compact-
ification R4 U {co). The Mobius transformations are compositions of the four
mappings sending q to
aqb-' a , b E S3 rotations
+
q b bEW translations
9X X >0 dilations
(q-l + c)-l cEW transversions.
A nice thing about quaternions is that all Mobius transformations of R4 can
+
be written in the form (aq b)(cq + d)-', where a, b, c, d E ]HI.
Fifth, we could focus our attention on a generalization of the Cauchy-Riemann
equations,
df d f .Of
-+i-+J-+k-=O df where f :lHI+IH[.
dw dx dy dz
Using the differential operator
the above equation can be put into the form
+
where f = fo f with fo : W + R and f : W + R3. This decomposes into
scalar and vector parts
dfo
- - df
V - f = O and - + V f o + V x f = o .
dw dw
There are three linearly independent first-order solutions to these equations
qI = x - iw, qy = y - jw, qz = z - kw.
76 Quaternions
Higher-order homogeneous solutions are linear combinations of symmetrized
products of q,, qy,q,. For instance, the symmetrized product of degrees 2,1,O
with respect to qx, qy,q, is seen to be
qxqy
2
+ qxqyqx + qyq; = 3(x2 - w ' ) ~- 6wzyi + (w3 - 3wx2)j.
This already shows that the last alternative results in an interesting class of
new functions, to some extent analogous to the class of holomorphic functions
of a complex variable.
Historical survey
Hamilton invented his quaternions in 1843 when he tried to introduce a product
for vectors in R3 similar to the product of complex numbers in 6. The present-
day formalism of vector algebra was extracted out of the quaternion product
+
of two vectors, ab = - a . b a x b, by Gibbs in 1901.
Hamilton tried to find an algebraic system which would do for the space
R3 the same thing as complex numbers do for the plane R2. In particular,
+
Hamilton wanted to find a multiplication rule for triplets a = ali a 2 j ask +
and b = bli + b j+ b3k so that lab1 = lallbl, that is, a multiplicative product
of vectors a, b E R3. However, no such bilinear products exist (at least not
over the rationals), since 3 x 21 = 63 # np+nz+n: for any integers n l , n2, n3
though 3 = 12+12+12 and 21 = 12+22+42 (no integer ofthe form 4a(8b+7),
> >
with a 0, b 0, is a sum of three squares, a result of Legendre in 1830).
Hamilton also tried to find a generalized complex number system in three
dimensions. However, no such associative hypercomplex numbers exist in three
dimensions. This can be seen by considering generalized imaginary units i and
j such that i2 = j2 = -1, and such that 1, i , j span R3. The product must
+
be of the form i j = a + i p jy for some real a , p, y. Then
i(ij) = i a - p + ($7 = i a - P + (a +iP+jy)y
+ + +
= -P a y i ( a Py) jy2, +
whereas by associativity i(ij) = i2j= -j which leads to a contradiction since
>
y2 0 for all real y.
Hamilton's great idea was to go to four dimensions and consider elements of
the form q = w + ix +jy + kz where the hypercomplex units i , j , k satisfy the
following non-commutative multiplication rules
3 Actually, it is not necessary t o assume that j2 = -1. The computation shows that there
is no embedding C C LR3, where LR3 is an associative algebra.
5.8 Function theory of quaternion variables 77
Hamilton named his four-component elements quaternions. Quaternions form
a division ring which we have denoted by W in honor of Hamilton.
Cayley in 1845 was the first one to publish the quaternionic representation of
rotations of R3 + R3, r + ara-l, but he mentioned that the result was known
to Hamilton. Cayley, in 1855, also discovered the quaternionic representation
of 4-dimensional rotations:
R4 + R4, q + aqb-l,
where we have identified R4 = W.
d d d
The differential operator V = i-+j-+k- is due to Hamilton, although
dx dy dz
his symbol for nabla was turned 30'. The first one to study solutions of
-df
+ i - + df
j - + k -df
=0, df where f :W+W,
dw dx dy dz
was Fueter 1935.
Comment
The quaternion formalism might seem awkward to a physicist or an engineer,
for two reasons: first, the squares of i , j , k are negative, i2 = j2 = k2 = -1,
and second, one invokes a 4-dimensional space which is beyond our ability of
visualization.
Exercises
1. Let u be a unit vector in R3, lul = 1. Show that R3 + R3, x + uxu is a
reflection across the plane u L .
2. Determine square roots of the quaternion q = qo + q.
3. Hurwitz integral quaternions q = w + ix + jy + k z are Z-linear
+ +
combinations of i, j,k and (1 + i j k), that is, either all w , x , y, z are
t.
integers or of the form n + Show that lqI2 is an integer, and that the set
is closed under multiplication.

4. Clearly, ab = ba implies eaeb = ea+b, but does eaeb = ea+b imply ab = ba?
5. Denote
Quaternions
Show that
acid - bdc c6b - dba
bEc - aEd daa - cab
+
for a non-zero A = la121d12 lb121~12- 2 ~ e ( a ~ d b ) .
6 . Verify that only one of the matrices
a= (j ) 1 j
and b = ( i k)
is invertible.
7. Does an involutory automorphism of the real algebra Mat(2, W) necessarily
send a diagonal matrix of the form
(i (d) where a E W
to a diagonal matrix?
8. Suppose A (# R) is a simple real associative algebra of dimension 5 4
with center R. Show that A is W or Mat(2,R).
9. Suppose A (# R) is a simple real associative algebra with center R and
an anti-automorphism a: + a ( z ) such that a: + a ( x ) E IW and a:a(a:) E R.
Show that A is W or Mat(2,R).
10. Show that all the subgroups of Q8 = {f1, fi, fj, fk) are normal, that is,
for a subgroup H c Q8 and elements g E Q8, h E H , ghgvl E H .
11. Take two vectors a , b in R3, such that la1 = Ibl, and a = ea, b = eb in
S3. Determine the point-wise invariant plane of the simple rotation
q + agb-l of R4.
Solutions
2. If q = 0, then there is only one square root, 0. If q = 0, go > 0, then there
are two square roots, f6. If q = 0, go < 0, then there is an infinity of
square roots, f i u , where u is a unit pure quaternion u E R3 c W,
IuI = 1. If q # 0, then there are two square roots,
and its opposite.

4. Hint: consider the quaternions a = 3ni and b = 4nj, or the matrices
0 -1 0 i
a = 3 ~ ( 0~ ) and b = 4 ~ ( O
~).
5.8 Function theory of quaternion variables
6. a is invertible, but b is not.
11. If a = b-l, the point-wise invariant plane is a' in R3.Otherwise the
point-wise invariant plane is spanned by a + b and
lallbl- ab = lallbl + a . b - a x b.
Bibliography
S.L. Altmann: Rotations, Quaternions, and Double Groups. Oxford University Press,
Oxford, 1986.
G.M. Dixon: Division Algebras: Octonions, Quaternions, Complex Numbers and the
Algebraic Design of Physics. Kluwer, Dordrecht, The Netherlands, 1994.
P. du Val: Homogmphies, Quaternions and Rotations. Oxford University Press, Ox-
ford, 1964.
R. Fueter: ~ b e die
r analytische Darstellung der regularen Funktionen einer Quaternion-
envariablen. Comment. Math. Helv. 8 (1935), 371-378.
K. Giirlebeck, W. Sprossig: Quaternionic Analysis and Elliptic Boundary Value Prob-
lems. Akademie-Verlag, Berlin, 1989. Birkhauser, Basel, 1990.
W.R. Hamilton: Elements of Quaternions. Longmans Green, London, 1866. Chelsea,
New York, 1969.
T.L. Hankins: Sir William Rowan Hamilton. Johns Hopkins University Press, Balti-
more, MD, 1980.
K. Imaeda: Quaternionic Formulation of Classical Electrodynamics and the Theory of
Functions of a Biguaternion Variable. Department of Electronic Science, Okayama
University of Science, 1983.
I.R. Porteous: Topological Geometry. Van Nostrand Reinhold, London, 1969. Cam-
bridge University Press, Cambridge, 1981.
lands, 1993.
A. Sudbery: Quaternionic Analysis. Math. Proc. Cambridge Philos. Soc. 85 (1979),
199-225.
6
The Fourth Dimension
In this chapter we study the geometry of the Euclidean space R4. The purpose
is to help readers to get a solid view, or as solid a view as possible, of the first
dimension beyond our ability to visualize. This is an important intermediate
step in scrutinizing higher dimensions. We start by reviewing regular figures
in lower dimensions.
6.1 Regular polygons in R2

The equilateral triangle, the square, the regular pentagon, ..., are regular poly-
gons. We shall also call them a 3-cell, 4cell,5-cell, ..., denoted by (31, (41, (51,
..., respectively. Therefore, we call a regular p-gon a p-cell, denoted by {p).
As p grows toward infinity, we get in the limit an co-cell, where the line is
divided into line segments of equal length. As a degenerate case we get a 2-cell,
which is bounded by 2 line segments in the same place. The interior angle of a
regular p-gon at a vertex is (1 - 2/p)1r.
6.2 Regular polyhedra in R3

A regular polyhedron is a convex polyhedron bounded by congruent regular
polygons, for instance, by p-gons. The number of regular p-gons meeting a t a
vertex is the same, say q ; it satisfies
because the sum of angles of faces meeting a t a vertex cannot exceed 21r. The
above inequality can also be written in the form
6.2 Regular polyhedra in R3 81
The same result is obtained by inspection of the topological properties of a
regular polyhedron: the numbers No, Nl, N2 of vertices, edges and faces satisfy
the Euler formula:
No - N1+ N2 = 2.
On the other hand, each edge of a regular polyhedron is a boundary of two
faces, each with p sides, so that 2N1 = pN2; and a vertex is a meeting point
of q edges, each with 2 end points, so that qNo = 2N1. The above inequality
is a consequence of the Euler formula and the equation
A regular polyhedron (p, q 2 3) must satisfy the foregoing inequality, and so

only a few pairs p, q are possible. These regular polyhedra are called Platonic
solids, or p, q-cells with Schl$li symbols {p, q}. There are five Platonic solids.
Name {P,Q} No N1 N2
Tetrahedron {3,3} 4 6 4
Octahedron {3,4} 6 12 8
Cube {4,3} 8 12 6
Icosahedron {3,5) 12 30 20
Dodecahedron {5,3} 20 30 12
When q = 2 in the above inequality we get a dihedron with Schlafli symbol

{p, 2). A dihedron is bounded by two regular polygons positioned in the same
place.
When a plane is covered by regular polygons so that at each vertex there
meet q regular p-gons, we are solving the equation
There are three solutions to the above equation; they have Schlafli symbols
82 The Fourth Dimension
{4,4), {3,6) and {6,3) corresponding to tilings of the plane by squares, equi-
lateral triangles and regular hexagons. These regular tilings are called tessel-
lations.
6.3 Regular polytopes in R4

A polyhedron is regular if its faces and vertices (= parts of the polyhedron near
a vertex point) are regular. A regular polyhedron with Schlafli symbol {p, q}
has p-cells as faces and q-cells as vertices. A vertex is regular, if a plane cuts
off a regular polygon whose central normal passes through the vertex.
A regular vertex
A polytope is a higher-dimensional analog of a polyhedron. A polytope is
regular if its faces and vertices are regular. A 4-dimensional regular polytope
with p, q-cells as faces must have q, r-cells as vertices. This drops the number
of 4-dimensional regular polytopes from 52 = 25 to 11. The sum of the solid
angles of the faces meeting at a vertex cannot exceed 4n. As a consequence,
there remain six possible combinations of p, q and q, r. A closer inspection
shows that all these six combinations are in fact 4-dimensional regular poly-
topes; we shall call them p, q, r-cells with Schlafli symbols {p, q, r).
{p,q,r) No N1 N2 N3 Face Vertex

--
{3,3,3) 5 10 10 5 Tetrahedron Tetrahedron

{3,3,4) 8 24 32 16 Tetrahedron Octahedron
{4,3,3) 16 32 24 8 Cube Tetrahedron
{3,4,3) 24 96 96 24 Octahedron Cube
{3,3,5} 120 720 1200 600 Tetrahedron Icosahedron
{5,3,3) 600 1200 720 120 Dodecahedron Tetrahedron
There are the regular simplex {3,3,3) and the hypercube {4,3,3), also called
a tesseract. There is the octahedron analog {3,3,4), a dipyramid with octahe-
dron as a basis. There are the analogs of the icosahedron and the dodecahedron,
{3,3,5) and {5,3,3); and there is an extra regular polytope {3,4,3).
The 3-dimensional space can be filled with cubes, a configuration with
6.4 T h e spheres 83
Schliifli symbol {4,3,4). The 4-dimensional space can be filled with hyper-
cubes, dipyramids and the extra regular polytope, configurations with Schlafli
symbols {4,3,3,4), {3,3,4,3) and {3,4,3,3).
In a higher-dimensional space, n > 4, there are only the regular simplex,
dipyramid and hypercube, and it can only be filled with hypercubes.
6.4 The spheres

A circle with radius r in R2 has circumference 2nr and area nr2. A sphere with
radius r in R3 has surface 4nr2 and volume %nr3.A hypersphere with radius r
in R4 has 3-dimensional surface 2n2r3 and 4-dimensional hypervolume ;n2r4.
For lower-dimensional spheres we have the following table:
n I surface I volume
If the volume of the sphere in Rn is denoted by wnrn then its surface is

nwnrn-I. Observe a rule mwmrm-l = 2nr - wnP between the surface in di-
mension m = n + 2 and the volume in dimension n. This leads to the recursion
and the formula

nnI2
Wn =-
!
which can be computed for odd n by recalling that (1/2)! = f i / 2 .
6.5 Rotations in four dimensions

Let A be an antisymmetric 4 x 4-matrix, that is, A E Mat(4,R), AT =
-A. Then the matrix eA represents a rotation of the 4-dimensional Euclidean
space R4. In general, a rotation of R4 has two invariant planes which are
completely orthogonal; in particular they have only one point in common.
The antisymmetric matrix A has imaginary eigenvalues, say ficr and fiP,
the eigenvalues of the rotation matrix eA are unit complex numbers e f and
e f " , and the invariant planes turn by angles cr and P under e A . First, assume
that a > p 2 0 (and o < K). Each vector is turned through at least an angle
/? and at most an angle a. In the case /? = 0 we have a simple rotation leaving
one plane point-wise fixed. If @/ais rational, then etA = I for some t > 0. If
P/a is irrational, then etA # I for any t > 0.
By the Cayley-Hamilton theorem eA is a linear combination of the matrices
I, A, A2 and A3 SO that
and direct computation shows that

1
ho = - ( a 2cos ,O - p2cos a ) ,
a 2 -82
Letting a now approach P and computing the coefficients in the limit give
lim eA = I(cos o
ff+D
+ -a2 sin a )
+;A (Z3 sin o - -2 cos a1
+-A2 (- sin o)
a2 2
0.
Observe that in the limit A2 = -021, which cancels some terms and results in
A
lim eA = I c o s a + -sina.
ff-P o
These rotations with only one rotation angle a have a whole bundle of invariant
rotation planes. In fact, every point of IR4 stays in some invariant plane, but
not every plane of R4 is an invariant plane of e A .
If a rotation U of IR4 has rotation angles a and ,8 we shall denote it by
U ( o , p ) . Consider the set J' = {U(o,P) E SO(4) I a = P} and the relation
'N' in the set J' = J' \ {I,-I),
which can be seen to be an equivalence relation. The equivalence class of a

matrix U E J" is the set {X E J" I X -- U}. This equivalence class together
6.6 Rotating ball in R4 85
with the center {I,-I) of the rotation group SO(4) forms a subgroup of
S 0 ( 4 ) , denoted in the sequel by the letter Q. Also ( J \ Q) U {I,-I) is a
subgroup of SO(4); denote it by Q*. Observe that UV = VU for U E Q and
V E Q*. It can be shown that Q and Q* are isomorphic to the group of unit
quaternions S3= {q E IHI I Iql= 1).
Each rotation L E SO(4) of R4 can be written in the form L = UV, where
U E Q, V E Q*. The rotation angles of L are a f,B when the rotation angles
of U and V are cr and ,B. A pair of completely orthogonal planes, both with
a fixed sense of rotation, induces a pair of senses of rotations for all pairs of
completely orthogonal planes. There are two classes of such pairs of oriented
planes: those of the type Q and those of type Q*.
-
Furthermore, we have an isomorphism of algebras,
{Xq I > 0, q E Q) U (01,

which we shall regard a s an identification. Introduce the algebra
w*= {Xq I X > 0, q E Q*) u (0).
and observe an isomorphism of algebras, W EW.
6.6 Rotating ball in R4

A rotating ball in IW3 has an axis of rotation, like the axis going through the
North and South Poles, and a plane of rotation, like the plane of the equator. A
rotating ball in IK4 has two lanes of rotation, which are completely orthogonal
to each other in the sense that they have only one point in common. Let the
angular velocities in these planes be bivectors w l and wz. The total angular
+
velocity is a bivector w = w l w2. The velocity v of a point x on the surface
of the ball is
v = x J w l +x-Iw2.
Assume that cp is the angle between the direction x and the plane of w l . Then
Therefore, the local angular velocity IvI/IxI is always between lwll and Iwzl.
If IwlI = Iw21, then every point on the sphere is rotating at the same veloc-
ity and furthermore every point is travelling along some great circle, that is,
everybody is living on an equator!
6.7 The Clifford algebra Ct4
The Clifford algebra Ct4 of R4 with an orthonormal basis {el,en, e3, e 4 ) is

generated by the relations
el2 -- e2
2- 2-
- e3 2-
- e4 - 1 and eiej = -ejei for i # j.
It is a 16-dimensional algebra with basis consisting of
scalar 1
vectors el, e 2 , e3, e 4
bivectors e n , el3, e l 4 , e 2 3 , e 2 4 , e 3 4
3-vectors e m , e l 2 4 , e l 3 4 , e 2 3 4
volume element e l 2 3 4
where e i j = eiej for i # j and e1z34 = e l e 2 e 3 e 4 .
An arbitrary element u E Ct4 is a sum of its k-vector parts:
k
u = (u)o + (U)I + (U)Z + (u)3 + (u)4 where (u)k E A R4.
There are three important involutions of Ct4:
= (U)O - (u)i + (u)2 - (u)3 + ( u ) 4 grade involution

6 = (U)O + (U)I -( ~ ) 2- (u)3 +(u)4 reversion
= (U)O - (U)I - (U)Z + +
( ~ ) 3 (u)4 Clifford-conjugation.
The Clifford algebra Ct4 is isomorphic to the real algebra of 2 x 2-matrices
Mat(2, E-JI) with quaternions as entries,
6.8 Bivectors in R4 C C 4
The essential difference between 3-dimensional and 4-dimensional spaces is
that bivectors are no longer products of two vectors. Instead, bivectors are
6.9 The group Spin(4) and its Lie algebra 87
sums of products of two vectors in lR4. In the 3-dimensional space IW3 there
are only simple bivectors, that is, all the bivectors represent a plane. In the
4-dimensional space R4 this is not the case any more.
+
Example. The bivector B = e l 2 e34 E /\2 lR4 is not simple. For all simple
elements the square is real, but B 2 = -2 +
2e1234$ IR. I
If the square of a bivector is real, then it is simple. '
Usually a bivector in /\2 R4 can be uniquely written as a sum of two simple
bivectors, which represent completely orthogonal planes. There is an excep-
tion to this uniqueness, crucial to the study of four dimensions: If the simple
components of a bivector have equal squares, that is equal norms, then the
decomposition to a sum of simple components is not unique.
+
Example. The bivector ele2 e3e4 can also be decomposed into a sum of
two completely orthogonal bivectors as follows:
6.9 The g r o u p Spin(4) a n d i t s Lie algebra

The group Spin(4) = {s E c~Z
I sS = 1) is a two-fold covering group of the
rotation group SO(4) so that the map
R4 + lR4, x + sxs-' , where s E Spin(4),
is a rotation, and each rotation can be so represented, the same rotation being
obtained by s and -s. The Lie algebra of Spin(4) is the subspace of bivectors
/\2 lR4 with commutator product as the product. The two sets of basis bivectors
a1 (e23 + e14) 1
a(e23 - e14)
1
(e3l+ e24) and a(e31 - e24)
1
(el2 + e34) 1
z(e12 - e34)
in /\2 lR4 c C.t4 both span a Lie algebra isomorphic to the subspace /\' lR3 c
Ct3 with basis {+e23, he31,+el,), that is, they satisfy the same commutation
relations. In other words, the Lie algebras
1 Although the square of a 3-vector is real, it need not be simple. For instance, V =
el23 +e456 E /\3 IR6 is not simple [this can be seen by computing V e i V - l , i = 1,2,.. ., 6 ,
and observing that they are not all vectors].
are both isomorphic to /\2 R3. The two subspaces i ( 1 f e1234) /\2 R4 of C14
annihilate each other, and consequently,
At the group level this means the isomorphism
where Spin(3) N S3N SU(2).
6.10 T h e m a p p i n g F -+ (1 + F)(1 - F)-' for F E /\2R4

The exponential eFl2 E Spin(4) of a bivector F E /\2R4 corresponds to the
rotation eA E S 0 ( 4 ) , where A(x) = F L x , for x E R4. Every rotation of
R4 can be so represented, and the two elements feF/%epresent the same
rotation.
+ +
The exterior exponential eF = 1 F i F A F of a bivector F E /\2 R4 is a
multiple of an element in Spin(4), that is,
Up to a sign, every element in Spin(4) can be so represented, except fe1234.

The exterior exponential eF of the bivector F corresponds to the rotation
+
(I A ) ( I - A)-l; every rotation of R4 can be so represented, except -I.
The above observations raise the question: What is the rotation correspond-
+
ing to (1 F)(1 - F)-l € Spin(4)? This is an interesting and non-trivial
question in dimension 4. Here follows the answer.
Let F E /\2 R4. The antisymmetric function induced by F is denoted by
A, that is, A(x) = F L x for all x E R4. Write s = (1 F ) ( 1 - F)-l. +
The rotation induced by s E Spin(4) is denoted by U E S 0 ( 4 ) , that is,
+
U = ( I A ) ( I - A)-l. In other words, U(x) = sxs-l for all x E R4. The
following cases can be distinguished:
(i) If F E /\2 R3 then U = (-)II -+AA 2

.
(ii) If F E /\2 R4 is simple, then U =
I+2A
(iii) If F E /\2 R4 is isoclinic, then U = -
I - 2A'
2 It is also a non-trivial question in dimension 5. In dimension 6, (1 + F)(1 - I?)-' !$
Spin(6).
6.10 The mapping F + (1 + F ) ( 1 - F)-l for F E R4 89
(iv) In the case of an arbitrary F E R4 we cannot express U as a rational
function of A [although U still has the same eigenplanes as A ] . Instead,
where B(x) = (FelZs4) L X, the dual of A. The denominator of U is a

multiple of the identity I.
Summary
There are three different kinds of rotations in four dimensions depending on
> >
the values of the rotation angles a , ,!? satisfying ./r > a ,!? 0. Let R : R4 +
R4 be a rotation and a a non-zero vector with iterated images b = R(a),
c = R ( b ) , d = R(c). In general, a, b , c, d are linearly independent, that is,
a A b A c A d # 0. In the case of a simple rotation with ,f? = 0, only the vectors
a, b , c are linearly independent, that is, a A b A c # 0 but a A b A c A d = 0. In
the case of an isoclinic rotation with a = ,!?, only the vectors a, b are linearly
independent, that is, a A b # O but a A b A c = O and a ~ b A d = O .
In general, a rotation of R4 has six parameters, computed as
The number 3 comes from picking up a unit vector a; the number 2 comes
from picking up a unit vector b in the orthogonal complement of a; the unit
bivector ab = a A b fixes a plane but the same plane is obtained by rotating
a and b in the plane of a A b , thus subtract 1; then finally add 2 for the two
rotation parameterslangles a and p. On the other hand, an isoclinic rotation
has three parameters, computed as
The number 3 comes from picking up a unit vector a in S3;but in an isoclinic

rotation a stays in a plane or a great circle S1,so subtract 1; and finally add
1 for the rotationlangle a = ,!?.
A simple bivector, an exterior product of two vectors, corresponds to simple
3 In dimension 5 the rotation U is given by the same expression, when
FAF
B(x)= (F-) L x.
IFAFI
The denominator is no longer a multiple of 1, although it still commutes with the numer-
ator by virtue of A B = B A .
4 An isoclinic rotation with equal rotation angles corresponds to a multiplication by a quater-
rotation turning only one plane. A simple bivector multiplied by one of the
idempotents i ( l f corresponds to an isoclinic rotation. An isoclinic
rotation has an infinity of rotation planes, and in fact, each vector is in some
invariant rotation plane of an isoclinic rotation.
The two-fold cover Spin(4) of SO(4) has three different subgroups isomor-
phic to Spin(3), each with a Lie algebra
There is an automorphism of Spin(4) which swaps the last two copies of

Spin(3), but there is no automorphism of Spin(4) swapping the first copy
of Spin(3) with either of the other two copies.
Exercises
+ +
1. Compute the squares of i ( 1 el2 e34 f
+
2. Take a vector a E R4 and a bivector B = ael2 pes4 E A2 R4. Show that
B a B E R4.
+
3. Compute exp(ae12 pe34).
+ + + +
4. Let a = ale1 a2e2 use3 and b = blel b2e2 b3e3. Compute
+
A = ae123 and B = be123. Determine i ( 1 e1234)A and i ( 1 - e1234)B,
and show that these bivectors commute.
+ +
5. Compute C = 3(1 e1234)A 3(1 - e1234)B, and express cxp(C) using
la1 and Ibl. What are the two rotation angles of the rotation
R4 + R4, x + cxc-l where c = exp(C)?
6 . Consider the Lie algebra A2R4 with the commutator product
[a, b] = ab - ba, and its three subalgebras spanned by
each isomorphic to R3. Show that there is no automorphism of the Lie

algebra R4 which permuts V, Z1, Z2 cyclically or swaps V for Z1 or Z2.
7. In two dimensions we can place 4 circles of radius r inside a square of side
4r, and put a circle of radius (fi - l ) r in the middle of the 4 circles. In
three dimensions we can place 8 spheres of radius r inside a cube of side
4r, and put a sphere of radius (a - l ) r in the middle of the 8 circles. In
n dimensions we can place 2n spheres of radius r inside a hypercube of
side 4r, and put a sphere of radius (fi - 1)r in the middle of the 2n
spheres.
6.10 The mapping F + (1 + F ) ( 1 - F)-' for F E /\2 lR4 91
Dimensions 2 and 3 differ topologically: in dimension 3 one can see the
middle sphere from outside the cube. Let the dimension be progressively
increased. In some dimension the middle sphere actually emerges out of
the hypercube. In some dimension the middle sphere becomes even bigger
than the hypercube, in the sense that its volume is larger than the volume
of the hypercube. Determine those dimensions.
Solutions
+
1. e1234, e l 2 e34.
+ + +
3. cos a cos p e l 2 sin a cos p e34 cos a sin p el234 sin a sin p.
5. The rotation angles are a = (la1+ lb1)/2 and P = (la1- lb1)/2, and
= cos a cos p - el234 sin a sin p

+C a -
a2 - p2
(sin a cos P + e l 2 3 4 cos a sin P).
7. In dimension 9 the middle sphere touches the surface of the hypercube,
and in dimension 10 it emerges out of the hypercube. In dimension 1206
the volume of the middle sphere is larger than the volume of the hypercube.
Bibliography
S.L. Altmann: Rotations, Quaternions, and Double Groups, Oxford University Press,
Oxford, 1986.
H.S.M. Coxeter: Regular Polytopes, Methuen, London, 1948.
P. du Val: Homographies, Quaternions and Rotations, Oxford University Press, Ox-
ford, 1964.
D. Hilbert, S. Cohn-Vossen: Anschauliche Geometrie, Dover, New York, 1944. Ge-
ometry and the Imagination, Chelsea, New York, 1952.
The Cross Product
The cross product is useful in many physical applications. It measures the

angular velocity w' = 5: x ii about 0 of a body moving a t velocity v' a t the
-+
position P, 5: = O P . It is used to describe the torque 5: x @ about 0 of a
force @ acting at 5:. It also gives the force = qii x 2 acting on a charge q
moving a t velocity ii in a magnetic field 2.
The usefulness of the cross product in three dimensions suggests the following
questions: Is there a higher-dimensional analog of the cross product of two
vectors in R3? If an analog exists, is it unique?
The first question is usually responded to by giving an answer to a modified
question by explaining that there is a higher-dimensional analog of the cross
product of n- 1 vectors in Rn.However, such a reply not only does not answer
the original question, but also gives an incomplete answer to the modified
question. In this chapter we will give a complete answer to the above questions
and their modifications.
7.1 Scalar product in R3

The linear space R3 can be given extra structure by introducing the scalar
product or dot product
+
for vectors a = ale1 + a2e2 use3 and b = blel + b2e2 + b3e3 in R3. The
scalar product is scalar valued, a . b E R , and satisfies
(a+b).c=a.c+b.c
linear in the first factor
(Xa) . b = X(a. b)
a.b=b.a symmetric
a . a > O for a#O positive definite.
'7.2 Cross product in R3 93
Linearity with respect to the first argument together with symmetry implies
that the scalar product is linear with respect to both arguments, that is, it
is bilinear. The symmetric bilinear scalar valued product gives rise to the
quadratic form
which makes the linear space R3 into a quadratic space R3. The quadratic
form is positive definite, that is, a . a = 0 implies a = 0, which allows us to
introduce the length la1 = &i-Z of a vector a E R3. The real linear space
R3 with a positive definite quadratic form on itself is called a Euclidean space
IR3. The length and the scalar product satisfy
la+bl<lal+lbl triangleinequality
la. bl 5 lallbl Cauchy-Schwarz inequality
where the latter inequality gives rise to the concept of angle. The angle cp
between two directions a and b is obtained from
cos cp = -
lal lbl
Thus, we can write the scalar product in the form
a formula which is usually taken as a definition of the scalar product, although

this requires prior introduction of the concepts of length and angle.
7.2 Cross product in R3

In the Euclidean space R3 it is convenient to introduce a vector valued product,
the cross product a x b E El3 of a, b E R3, with the following properties:
(a x b) Ia, (a x b ) l.b orthogonality
la x bl = lallbl sinp length equals area
a,b, a x b right-hand system.
In other words, the vector a x b is perpendicular to a and b, its length is
equal to the area of the parallelogram with a and b as edges, and the vectors
+ +
1 The function R3 + R, a -t la1 is a norm satisfying IXa) = IXllal, la bl 5 !a[ Ibl,
la1 = 0 j a = 0. Since this norm can be obtained from a scalar product, it sat~sfiesthe
+ + +
parallelogram law la bI2 la - bI2 = 21aI2 21bI2.
94 The Cross Product
a, b and a x b are oriented according to the right hand rule.
a x b
The above definition results in the following multiplication rules:

el x e2 = e3 = -e2 x el,
e2 x e3 = e l = -es x e2,
e3 x el = e 2 = -el X e3.
It is convenient to write the cross product in the form
el '32 e3
a x b = a1 a2 a3 .
bl 62 63
The cross product is uniquely determined by
(a x b ) . a = 0, (a x b ) . b = 0 orthogonality
la x bI2 = la121b12- ( a . b)2 Pythagorean theorem
together with the right hand rule. The Pythagorean theorem can be written
using the Gram determinant as
which in coordinate form means Lagmnge's identity
The cross product satisfies the following rules for all a, b , c E R3 :

axb=-bxa antisymmetry
(a x b ) . c = a . ( b x c) interchange rule.
The antisymmetry of the cross product has a geometric meaning: the lack of
'7.3 Cross product of n - 1 vectors i n IRn 95
symmetry measures how much the two directions diverge. The cross product
is not associative, a x (b x c) # (a x b) x c, which results in an inconvenience
in computation, because parentheses cannot be omitted.
The cross product is dual to the exterior product of two vectors:
Taking the exterior product of a x ( a x b) = a ( a . b ) - laI2b and b one finds

that
(ax (ax b ) ) ~ b
a.b= for a y i b ,
aAb
that is, the scalar product can be recaptured from the cross product [you can
also replace A by x in the above formula].
7.3 Cross p r o d u c t of n - 1 vectors in Rn

We can associate to three given vectors a, b , c in IR4 a fourth vector
which is orthogonal to the factors a, b , c and whose length is equal to the

volume of the parallelepiped with a, b , c as edges, that is,
The cross product a x b x c of three vectors a, b , c in R4 is completely anti-

symmetric and obeys the interchange rule slightly modified:
where d E R4.The oriented volume of the 4-dimensional parallelepiped with

a, b, c , d as edges is the scalar
multiplied by (the unit oriented volume) e1234.

2 The cross product is antisymmetric, a x b = - b x a, and satisfies the Jacobi identity
+ -+
a x (b x C ) b x (c x a) c x (a x b) = 0, which makes the linear space R3,with cross
product on R3,a non-associative algebra, called a L i e algebra. The Jacobi identity can
be verified using a x (b x c) = (a.c)b - ( a . b)c.
The cross product of three vectors in R4 is dual to the exterior product:
where the latter product is computed in the Clifford algebra C.&.

In a similar manner we can introduce in n dimensions a cross product of
n - 1 factors. The result is a vector orthogonal to the factors, and the length
of the vector is equal to the hypervolume of the parallelepiped formed by the
factors.
7.4 Cross p r o d u c t of two vectors in R7

Is there a cross product in n dimensions with just two factors? If we require
the cross product to be orthogonal to the factors and have length equal to the
area of the parallelogram, then the answer is no, unless n = 3 or n = 7.
The cross product of two vectors in R7 can be defined in terms of an ortho-
normal basis e l , ez, . . . , e7 by antisymmetry, ei x e j = -ej x ei, and
The above table can be condensed into the form
where the indices are permuted cyclically and translated modulo 7.

This cross product of vectors in R7 satisfies the usual properties, that is,
( a x b) . a = 0, (a x b) . b = 0 orthogonality
la x bI2 = la121b12- ( a . b)' Pythagorean theorem
where the second rule can also be written as l a x bl = lallbl sin 6(a,b ) . Unlike
the 3-dimensional cross product, the 7-dimensional cross product does not
+ +
satisfy the Jacobi identity, (a x b) x c ( b x c) x a (c x a) x b # 0, and
so it does not form a Lie algebra. However, the 7-dimensional cross product
satisfies the Malcev identity, a generalization of Jacobi, see Ebbinghaus et al.
1991 p. 279.
In R3 the direction of a x b is unique, up to two alternatives for the ori-
entation, but in R7 the direction of a x b depends on a 3-vector defining the
cross product; to wit,
7.4 Cross product of two vectors in R7
depends on
In the 3-dimensional space a x b = c x d implies that a, b, c, d are in the

same plane, but for the cross product a x b in IW7 there are also other planes
than the linear span of a and b giving the same direction as a x b .
The 3-dimensional cross product is invariant under all rotations of S 0 ( 3 ) ,
while the 7-dimensional cross product is not invariant under all of S 0 ( 7 ) , but
only under the exceptional Lie group G 2 , a subgroup of SO(7). When we let
a and b run through all of R7, the image set of the simple bivectors a A b is a
manifold of dimension 2.7-3 = 11 > 7 in A' R7, dim(/\' R7) = 37(7- 1) = 21,
while the image set of a x b is just R7. So the mapping
is not a one-to-one correspondence, but only a method of associating a vector

to a bivector.
The 3-dimensional cross product is the pure/vector part of the quaternion
product of two pure quaternions, that is,
a x b = Im(ab) for a, b E R3 c W.
In terms of the Clifford algebra Ct3 2 Mat(2, C) of the Euclidean space R3
the cross product could also be expressed as
a ~ b = - ( a b e ~ for
~ ~ a) ,~b E R 3 C C t 3 .
In terms of the Clifford algebra Cto,3 W x W of the negative definite quadratic
space R0p3 the cross product can be expressed not only as
a x b = -(abelzs)l for a, b E R013 c Ceo,3
but also as
a x b = (ab(1 - elz3))l for a, b E R013 C Ct0,3.
Similarly, the 7-dimensional cross product is the pure/vector part of the
octonion product of two pure octonions, that is, a x b = (ao b ) l . The octonion
algebra 0is a norm-preserving algebra with unity 1, whence its purelimaginary
part is an algebra with cross product, that is, a x b = $(ao b - b o a) for
a, b E R7 C 0= R @ R7. The octonion product in turn is given by
3 This expression is also valid for a, b E R 3 C C e 3 , but the element 1 - e l 2 3 does not pick
up an ideal of C e 3 . Recall that C e 3 is simple, that is, it has no proper two-sided ideals.
+ +
for a = a a and b = ,f? b in R $ R7. If we replace the Euclidean space R7
by the negative definite quadratic space R017, then not only
aob=a/?+ab+ap+a.b+axb
for a , b E R $ R017, but also
7.5 Cross products of k vectors in Rn

If one reformulates the question about the existence of a cross product of two
vectors in Rn, and also allows n - 1 factors, then one is led to a more general
problem on the existence of a cross product of k factors in Rn. If we were
looking for a vector valued product of k factors in Rn, then we should first
formalize our problem by modifying the Pythagorean theorem, a candidate
being the Gram determinant. A natural thing to do is to consider a vector
valued product a1 x a 2 x . . - x a k satisfying
(al x a 2 x . - . x ak) . ai = 0 orthogonality
lal x a 2 x 3 .x a k 1' = det(ai . aj)
. Gram determinant
where the second condition means that the length of a1 x a 2 x . . . x a k equals
the volume of the parallelepiped with alla 2 , . . . , a k as edges.
The solution to this problem is that there are vector valued cross products
in
3 dimensions with 2 factors
n dimensions with n - 1 factors
and no others - except if one allows degenerate solutions, when there would
also be in all even dimensions n, n E 22, a vector product with only one factor
(and in one dimension an identically vanishing cross product with two factors).
The cross product of three vectors in R8 can be expressed as
a x b x c = (a A b A c) J (w - ve8) = ((a A b A c)(l - el2...8)w)l
where
w = -(el24 +
e235 e346 + + e457 + e m + e672 + e713)en...7
= el236 - el257 - el345 + el467 + - - e3567
~ 3 4 7
and w E /\4R7 c
7.5 Cross products of k vectors i n Rn 99
The trivial cross product with one factor in an even number of dimensions
rotates all vectors by 90'. Thus, let n be even and let a be the only factor
of a trivial cross product with value b , Ibl = la[, b . a = 0. This can be
accomplished by
Exercises
1. Show that the cross product a x r can be represented by a matrix
multiplication Ar = a x r, where
2. Express the rotation matrix e A in terms of I,A and A2. Hint: use the
Cayley-Hamilton theorem, A3 + laI2A = 0.
3. Express the rotated vector e A r as a linear combination of r, a x r and
( a .r ) a . Hint: A2r = ( a . r)a - a 2 r .
4. Compute the square of w = -ve12,,,7 E /\4 IR7.
5. Show that i ( 1 + w) is an idempotent of C17 2 Mat(8, C).
Solutions
A A2
2. eA = I + - s i n a + - ( 1 - c o s a ) , where a = la[.
a a2
sin a 1 - cos cu
+
3. e A r = c o s a r -a x r +
a
a2 ( a .r)a.
+
4. w2 = 7 6w.
Bibliography
F.R. Harvey: Spinors and Calibrations. Academic Press, San Diego, 1990.
W.S. Massey: Cross products of vectors in higher dimensional Euclidean spaces.
Amer. Math. Monthly 90 (1983), #lo, 697-701.
R.D. Schafer: On the algebras formed by the Cayley-Dickson process. Amer. J. Math.
76 (1954), 435-446.
8
Electromagnetism
The Maxwell equations can be formulated with vectors or more advanced no-
tation like tensors, differential forms or Clifford bivectors. In these advanced
formalisms the Maxwell equations become more uniform and easier to manip-
ulate; for instance, relativistic covariance is more apparent. However, the cost
of the convenience is that one has to master new concepts in addition to scalars
and vectors; and antisymmetric tensors have to be untangled for physical in-
terpretation.
8.1 The Maxwell equations

The electric field l? and the magnetic induction 3 act on a charge q moving
at velocity v' by the Lorentz force
F=q(E-ta~Z).
The electric displacement 6 and the magnetic intensity are related to
and 3 by the constitutive relations
6=~l?,B = p H .
-, -
J. C. Maxwell brought together the following four equations in 1864:
-+
V.D=p A6 . dZ= Q Gauss' law
VXH=T l? . dl?= 1 Amp6re's law
V . ~ = O A~.~z=o no magnetic sources
VxE=--
- - d@
E . d l = -- Faraday's law
dt
8.1 The Maxwell equations 101
Maxwell also complemented Ampbre's law by a new term, which observed time-
dependence. Ampitre had developed a mathematical formulation for producing
'
magnetism by electricity, a phenomenon detected by 0rsted in 1820, but his
law is not valid in a time-varying situation: take the divergence of both sides
to obtain
v.(vxEi)=o=v.T
which violates charge conservation. Maxwell corrected this equation into the
form
-. -
V . ( V X H ) = O = V . J + - , ap
at
applied Gauss' law, and got
This predicted the existence of a displacement current d 6 / d t , which was first

detected experimentally by H. Hertz in 1888, when he radiated electromagnetic
waves by a dipole antenna. The electromagnetic field is now described by the
Maxwell equations
These equations are linear, and the last two equations with a vanishing right-
hand side are homogeneous.
If E,p are constants, so that they do not depend on position, then the
medium is homogeneous. If E , p are scalars, and not matrices or tensors, then
the medium is isotropic. In a medium that is uniform in space, i.e. homo-
geneous and isotropic, and stationary in time, the Maxwell equations can be
1 In the paper of 1820, 0rsted's name is printed as ~ r s t e dbecause
, the printer had no 0 .
2 Charge conservation requires that the continuity equation
holds for the charge density p and the current density f in R3.
[a 5,
3 W e u s e S I units: = [dl = 5, 5,
[dl = [Ef] =&.
4 In the case that the material is non-isotropic, D, = € i J E J , Bi = p i J H J , where the
matrices are symmetric € i J = e 3 i , piJ = p J i .
5 Stationary means that E and p do not depend on time. In an explosion E and p are time
dependent.
Electromagnetism
expressed in terms of E and I? alone:
where 1/c2 = ~ p These

. equations hold in a vacuum. In a vacuum it is cus-
tomary to set E = 1, p = 1.
8.2 T h e M i n k o w s k i space-time R3v1

The electromagnetic quantities depend on time t E R and position x = x l e l +
+
x2e2 x3e3 E R3. Position and time can be united into a single entity
a vector in a 4-dimensional real linear space R4 = R 3 x R . In this linear space

we introduce a metric (or a quadratic form)
2: + 2; + 2; - c2t2
which makes it a quadratic space, called the Minkowski space-time R3v1.
In the Minkowski space-time it is customary to set x4 = ct = - 2 4 and agree
that the indices are raised and lowered as follows:
2' = 2 1 , x 2 = 2 2 , x3 = 2 3 and x4 = -x4
With this convention the quadratic form xf + x i + x i - c2t2 becomes
where in the last step we have used the summation convention.

E x a m p l e s . 1. The two densities p and T can be combined into a single
quantity
J = T + cpe4 in R3J
with four components J 1 ,J 2 , J3, J4 = cp = -J4 and the quadratic form
+ +
51" 52" 53" - J&
2. We can combine the two potentials V and A
in R 3 into a single quantity
with four components A', A 2 ,A3 and A4 = $V = -A4, a space-time vector
with a quadratic form Af + A; + A; - A:.

8.3 Antisymmetric tensor of the electromagnetic field 103
8.3 Antisymmetric tensor of t h e electromagnetic field
H. Minkowski combined the two vectors l? and l? into a single quantity, a
4 x 4-matrix with entries FaP given by
and antisymmetry, FaP = -FPa, so that
The matrix entries FaP are coordinates of an antisymmetric tensor of rank 2,

namely the electromagnetic field in space-time IW3l1.
With this change of notation from g, l? to FaP we can write the Maxwell
equations in a vacuum:
The last displayed equation can also be written as
by employing antisymmetry and the lowering convention 2 4 = -x4.

Using the summation convention the Maxwell equations for Fa@ can be
condensed to
dFP7 dFYa
-+-
dx, dxp
+--dFaP
dx,
- 0,
d d
and further adopting the notations d, = - and da = - to
dxa 8%
104 Electromagnetism
Similarly, 6 and can be combined to a second-rank antisymmetric tensor
Using Gap the general Maxwell equations (non-homogeneous, non-isotropic,

time-varying) can be written in tensor/index form, due to Minkowski:
Exercises lab,2ab,3a
8.4 Electromagnetic potentials

Because of V .2? = 0 there exists, in a contractible region, a vector-potential
A such that
i=~xA.
If this equation is substituted into Faraday's law, we get
This curl-free quantity is up to a sign the gradient of a scalar, called the electric
potential V,
We have shown that E and 2? can be expressed in terms of the potentials V

and A
as follows:
Combine the two potentials V and A

in IR3 into a single quantity with four
components A', A', A3 and A4 = :V, a space-time vector
8.5 Gauge transformations 105
T h e above equations mean that FaP can be expressed in terms of the potential
Aa as follows:
which can be condensed to
We can now verify that dffFPr dPFra+ + dYFaP = 0 by computing

a " ( a P ~ r- YAP) + a P ( a r -~ aaAY)
~ + d T ( d f f ~-P PA,) = 0 .
Exercises 2c,3b
8.5 Gauge transformations

The vector-potential is not unique, since we can add to it, without changing
physics, any vector with a vanishing curl. Adding to
-1
A
a curl-free vector,
the gradient of a scalar @, gives us A = + V @ . In order to keep E =
A
-VV - aAlat we also change V to V',
which implies V' = V - d@/dt.The change of potentials
is called a gauge transformation. In coordinate form this means
or swapping the sign of the time component
A', = A,
a@ = A, + a,@.
+-
axa
The fact that l?,l? remain unchanged in a gauge transformation is called
gauge invariance. In quantum electrodynamics gauge invariance is used t o
deduce the existence of a zero-mass carrier for the electromagnetic field.
8.6 The Lorenz condition for potentials

The two homogeneous Maxwell equations guaranteed existence of potentials
for the electromagnetic field. Now we shall find out conditions imposed on the
potentials by the remaining Maxwell equations. Substitute 3 = -VV - a A l a t
into V .i= PIEto obtain
in a vacuum. Substitute l? = -VV - a A p t and I? = V x A into V x 2 -

$dl?/dt = p y , and use the identity V x ( V x A) = V ( V .A) - V2A, to obtain
The last two displayed equations couple V and A.

Although the curl of /i is designated to I?, we are still a t liberty to choose
the divergence of A,
which ensures the choice
called the Lorent condition. In coordinate form,
When the Lorenz condition is satisfied, the above two second-order differen-
tial equations, which coupled V and A, can be decoupled
into wave equations with the d'Alembert operator v2- $& = dada
6 The Lorenz ~ondition/~auge was discovered by the Danish physicist Ludwig Lorenz in
1867, and not by the Dutch physicist H. A. Lorentz, who demonstrated covariance of the
Maxwell equations under Lorentz transformations in 1903. See 3 . van Bladel: Lorenz or
Lorentz? IEEE Antennas and Propagation Magazine 33 (1991) p. 69 and The Radiosci-
entist 2 (1991) p. 55.
8.7 The vector l? and the bivector l?e123 107
In the rest of this chapter we shall discuss electromagnetism in terms of the

Clifford algebras. Clifford algebras automatically take care of the manipulation
of indices. The Clifford algebra approach allows various degrees of abstraction
- -
which gradually become more and more distant from classical vector analysis.
We reformulate the Maxwell equations first in terms of the Clifford algebras
Ce3 Mat(2, @) of the Euclidean space R3 and then in terms of the Clifford
algebra Ce3,l Mat(4,R) of the Minkowski space R311. In the Euclidean
space R3 we shall deal with the vector l? and the bivector and in the
Minkowski space R311 we shall deal with the bivector
8.7 T h e vector l? a n d t h e bivector gel23

The work W done by an electric field l? in moving a charge q along a path C
is given by the line integral
We conclude that the electric field l? is a vector, because it is integrated along

a path.
Similarly, the magnetic induction 5 is integrated over a surface S in order
to get the magnetic flux:
Since we are integrating over a surface, we conclude that we are actually dealing
+ +
with the bivector gel23 = BleZ3 B2esl B3e12, rather than the vector
4
+
B = Blel B2e2 B3e3.+
8.8 Differentiating vectors a n d bivectors

Differentiate the vector I?, in R3 C Cl3, to find
v~?=v.~?+vA~?
=V .z+ el23(V x l?)
where ~ x l =
? - e 1 2 3 ( ~ ~ zDifferentiate
). the bivector to find v(Zelz3)=
+
V A (2e123) V J (Zel23) where
V A (gem)= . Z) ,
v J (2e123)= e123(VA B) = -V x
4
8.
8.9 Single equation in Ce3

Recall the Maxwell equations in a vacuum:
4
V.E=p,
-
Multiply the last two equations by e123, use the following replacements V A ~ =
4
(V x l?)e123, V J (gel23) = -V x 6 and V A (ge123) (V . B)e123, and you

will get
?
0 v.l?=p
The numbers on the left indicate the dimension degrees of the equations. Sum-
+
ming up these four equations we get (use ~ l =?V . l? V A l?)
Use V F = V J F + V A F tofind
and we have condensed all the Maxwell equations into a single equation in terms
of the Clifford algebra C13. Taking the grade involute of both sides results in
The potentials V and 1,a scalar and a vector, can be united into a para-
vector V + /i. Differentiate the paravector V + by the paravector differential
A
8.10 The use of the Clifford algebra Cestl
operator,
a + V,
where VA = V . A + (V x A)el23, and you will get

( aB+ V)(V + A) = -2 + Je123.
Taking the grade involute of both sides results in
a
(- - V)(V - A) = + sel23.
at
8.10 T h e use of t h e Clifford algebra Ces,l

Consider the Clifford algebra Ce3 z Mat(2, @) as a subalgebra of the Clifford
algebra Ce3,1 generated by el, e2, e3, e4 with the relations
e 2 - - e2
2 -- e3
2 --1 , e j = -1,
e,ep = -epe, for a # P.
The Clifford algebra Ce3,1 is isomorphic, as an associative algebra, with the

algebra of real 4 x 4-matrices Mat(4,R). In the Clifford algebra Ce3,1 we
consider the electromagnetic bivector
2
14
F = -Ee4 - Be123 E
C
A IR3t1
and the space-time current vector
From F we can find 2 by l? = ce4 J F, and from J we can find f by

J'= ( J ~ e 4 ) e ; l .
We introduce the differential operator
a +
For a function f : R311 + Ce3,~we have df = A f 8 J f , where 8 A f is the
raising differential and 13 J f is the lowering differential.
--
7 If we use the orthonormal basis {eo,e l , ez, e3} with e: = e; = eg = 1 and e%= -1,
that is, = e 4 , then we find by reordering the indices that
eo = -el234 and F =
eo123
gee4 +
- (Be4)e1234= gee0 (geo)eolm.
Compute the raising differential
Define G = c6e4 - I?elZ3and compute the lowering differential
The Maxwell equations now have a particularly succinct form
corresponding to
- j,
-7' (He123) - 8 6 -
4 4
V.D=p,
d
-V A (Ze123) = 0, VA2 + -(Ze123)
at
= 0.
8.11 Single equation in a vacuum, Ce3,1
In a vacuum the Maxwell equations can be further compressed into a single

equation
d F = J,
which decomposes into two parts, d A F = 0 and d J F = J . Also, d A A = -F
and the Lorenz condition 13. A = 0 imply
a A = -F.
8.12 The energy-momentum tensor

Marcel Riesz in 1947 wrote the energy-momentum tensor in the form
1
T,, = - - (e, Fe, F)o.
2
8 The 3D formulation differs from this 4D formulation in the sense that G and F are
bivectors in /\2 R3t1.
8.12 The energy-momentum tensor
D. Hestenes 1966 p. 31 introduced the vectors
for which T,, = T , . e, = e, . T,, and also the mapping T x = - ; F X F

where ( T x ) , = T,,x". The energy-momentum tensor is symmetric, that
- l T , where g,, = e, . e,, and traceless, that is,
is, T,, = T,, or 9 ~ ~ g =
T,, = 0. lo
Note that the Poynting vector T q = I? x 2 $(I?' + +
Z2)e4 is not a space-
time vector, in the sense that it does not transform properly under Lorentz
transformations, but rather it is just the last column of the energy-momentum
matrix T = (T",) which transforms as T = LTL-l.
ELECTROMAGNETISM
IN DIFFERENTIAL
FORMS
Electromagnetism can also be formulated with differential forms, based on
Grassmann's exterior algebra. In this context it is customary to invoke the
dual space
V* = {f : V +R 1 f linear)
of the real linear space V = R311. Instead of vectors and bivectors, in V and
/\2 V, one considers 1-forms and 2-forms, in V* and /\2 V*.
In theoretical physics one applies differential forms to electromagnetism, but
in electrical engineering one uses almost exclusively the vector analysis of Gibbs
and Heaviside. l 1 Electrical engineers are not interested in transformation
''
laws, and so it is convenient for them to place all vectors in V = R3s1 [and
disregard the dual space V*]. However, a theory without the dual space V*
cannot be generalized to curved space-times. In a curved space-time it is not
possible to differentiate vector valued functions, only differential forms can be
differentiated [in general relativity vectors are differentiated covariantly].
Although differential forms are not of practical value for electrical engineers,
+
9 Juvet&Schidlof 1932 p. 141 gave Tfiv= FfiaFa, $ g f i , ~ , p ~ a pbut did not consider
T x = - $ F X F , compare this to Bolinder p. 469 in Chisholm & Common (eds.) 1986.
10 T h e tracelessness of T f i , = - $ ( e f i F e , ~ ) is
o a n accident in dimension 4, since efiFe,, = 0,
k
and in general efiaep= ( n - 2 k ) l for a E /\ Rn.
11 As far as the author knows the only university where electrical engineers have used dif-
ferential forms in teaching is Helsinki Urliversity of Technology, see lecture notes Lin-
dell & Lounesto 1995.
12 For instance, the spacetime position x = Z + cte4 and the current density J = J+ cpe4
transform differently under the Lorentz group; one transforms contravariantly and the
other covariantly. In tensor calculus elements of V are called vectors and elements of t h e
dual space V * are called covectors.
we shall close this chapter with a short discussion on the formulation of electro-
magnetism with differential forms, see Lindell1995. But first some observations
about functions with values in /\ V = Ce3,1.
8.13 Using o n l y raising o r lowering differentials

Since the current density f integrates over a surface S,
we can replace it by a bivector fel23, and since the charge density p integrates
over a 3-volume, we can replace it by a 3-vector pel23. Similarly, we can regard
I? as a vector, but replace the vector d by a bivector 8el23.
The two Maxwell equations with a source-term on the right hand side can
be rewritten in the form
V A (6e123) = pel23,
- a -
-V A H - -(Del23)
+
= Je123.
dt
Take the Hodge dual
- +
*G = Gel234 = -cDe123 - He4 and
4
+
*J = h 2 3 4 = C p 1 2 3 (fe123)e4,
and compute the raising differential
4
d A *G = V - e4- - A (-cDe123 - d e 4 )
( c aat )
+ ad
= -c(V . 6)e123 - e m ( V x H)en + - p l 2 3 4
The Maxwell equations can now be expressed in terms of the raising differential
alone:
~A*G=-*J,
dAF=O.
Dually, we can write down the Maxwell equations using only the lowering
differential:
d J G = J,
dJ*F = 0.
These equations are invariant under the general linear group GL(4,R), and
the solutions are independent of the choice of metric. l3
13 In the absence of a metric it is customary to invoke the dual algebra V * of the exterior
algebra A V and take exterior differentials of differential forms rather than differentials
of multivector valued functions.
8.14 The constitutive relations
8.14 T h e constitutive relations
The constitutive relations of the medium are
~=EE+cYB,
= pE + p - 1 z .
Here E , a , p and pW1 are 3 x 3-matrices. To find the rules imposed on them,
write the above equations in coordinate form:
Then, if XnXpv is an irreducible tensor, l4 we must have
where the brackets [ ] mean complete alternation of indices. The second re-
lation implies cT = E, pT = p and a = -pT and the third relation im-
plies trace(&) = trace(p), which together with the former implies trace(&) =
trace(/?) = 0. These considerations can be condensed into saying that the in-
dices of the constitutive tensor X n X p v = $:
can be arranged into a Young
tableau
The irreducible tensor x has 20 components, where 20 = &n2(n2- 1 ) for

n = 4. In chiral media the tensor x need not be irreducible, and the number
of components may rise to 36.
8.15 The derivative a n d t h e exterior differential

Let U and V be real linear spaces with norms. The derivative of f : U +V
at x U is a linear function
such that
f ( x + h ) - f (x) = f l ( x ) h + I1 h II c(x, h )
where E(X,h ) -+ 0 as h + 0. The linear function f'(x) : U -+ V can be
identified with an element of U* 8 V.
14 The factor x need not be a tensor. For instance, magnetic saturation and hysteresis are
not expressible with a tensor X.
Consider now a function f 8 : V + /\ V*. Its derivative at x ,
is no longer an element of the dual exterior algebra /\ V* C @V*.The alter-

nation, which antisymmetrizes tensor product of vectors, is a linear function
projecting /\ V* out of @V* so that
u ~ u = A l t ( u @ v ) for u , u E A V *
We define the exterior differential of f : V + /\ V* at x by l5
Next, we will replace vector valued functions V -+ V by 1-forms V +

V*, and bivector valued functions V + /\' V by 2-forms V + A' V*. The
electromagnetic bivectors F and G are replaced by 2-forms F and G. The
current vector J is replaced by a I-form J.
The exterior differential raises the degree. The dual of the exterior differen-
tial, called the contraction differential d J f = * - ' d ~ * f , l 6 lowers the degree.
In differential forms the Maxwell equations look like
8.16 General linear covariance of t h e Maxwell equations

Using the differential forms we may find the most general expression of the
Maxwell equations:
These equations include only the exterior differential, and no contraction differ-
ential, so that a metric is not involved. This makes the equations independent
of any coordinate system. The metric gets involved by the constitutive relations
of the medium
and the Hodge dual.

This form of the Maxwell equations is not only relativistically covariant,
15 The exterior differential is usually denoted by d j .
16 The contraction differential is commonly called the co-differential and denoted by 6 j .
8.16 General linear covariance of the Maxwell equations 115
under the Lorentz group O(3, I), l 7 but also covariant under any linear trans-
formation of space-time coordinates, that is, under the general linear group
GL(4,R). This general linear covariance of the Maxwell equations, and their
independence of metriclmedium, were recognized by Weyl 1921, Cartan 1926
and van Dantzig 1934.
Historical Survey
The Maxwell equations have been condensed into a single equation using com-
plex vectors (Silberstein 1907), complex quaternions (Silberstein 1912/1914,
Lanczos 1919), spinors (Laporte & Uhlenbeck 1931, Bleuler & Kustaanheimo
1968) and using Clifford algebras (Juvet & Schidlof 1932, Mercier 1935, M.
Riesz 1958). Marcel Riesz 1947 wrote the energy-momentum tensor in the
form Tp, = -+(epFe,~)o.
Exercises
Metric x: + +x i xg - x i :
1. Recall that (F14,F24,F34) = (f El, $E2, $ ~ 3 and)
~ , F ' ~=
( F ~ F31, ) (-B1, -B2, -B3). Compute the matrices
a) F a p , b) F,P, and the vector
c) v* Fa@for (vl, v2,v3, v4) = (vl, 212, v3, c). l8
Metric -xi + x: + x i + xg :
I d d d d
In this metric da = (-- - - - -).
cdt'dx'dy'dz
2. Replace l? and l? by F"@= -F@" so that
(Fol,Fo2, Fo3) = (-:El, - f E 2 , -fE3) and
( F ~F31,
~ F12)
, = (-B1, -B2, -B3), and determine
a) the antisymmetric matrix F f f p ,
b) the Maxwell equations in terms of Fa@,
C) F*@ in terms of A",
17 The Maxwell equations describe massless particles, photons, and as such they are con-
formally covariant, as was demonstrated by Cunningham and Bateman in 1910. The
conformal transformations are not linear in general, that is, they are not in GL(4,R).
18 For simplicity we have omitted the factor
which makes both sides of the equation of the Lorentz force, J P = u a ~ , P , ua = y v a ,

properly transforming space-time vectors.
Electromagnetism
d) va Fa@for ( v Ov, l , v2,v3) = ( c ,v1, v2,213).
Metric xi - x: - xz - xi :
Id d d d
In this metric da = (--, --, -- --).
c dt dz du'" d z
3. Replace l? and by Fa@= -FPa so that
( ~ 0 ~1 0 1 ~2 0 3 =
) ( - 1 ~ 1-1E2,-1E3
j
C ' C c ) an d
(F23,F31,F12) = ( - B 1 , - B ~ -B3),
, and determine
a) the Maxwell equations in terms of Fa@,
b) Fad in terms of Aa.
[Note that A' = A, = -A1, A2 = Ay = -A2, A3 = Ae = -A3 but
A O = L V = A0 $1
4. Electrical engineers use the pairs l?,Ei and 5 , d . The constitutive
relations sending l?,2 to 8,d are then
Show that fi = p , & = crp, ) = -pP and 2 = E - apP.
Solutions
la.
c. The space-component is l? + Z x and the time-component iv' .l?.

8.16 General linear covariance of the Maxwell equations 117
d. T h e time-component is tii. I? and the space-component I? + x fi.
3a. d , ~ ~= p ~ p , P F P Y +d +
P ~ 7 @FaP
~ = 0.
b. Fa@ = daAP -
Bibliography
W.E. Baylis: Electrodynamics: A Modern Geometric Approach. Birkauser, Boston,
1999.
E.F. Bolinder: Unified microwave network theory based on Clifford algebra in Lorentz
space, pp. 25-35 in Conference Proceedings, 1 2 European
~ ~ Microwave Conference
(Helsinki 1982). Microwave Exhibitions and Publishers, Tunbridge Wells, Kent,
1982.
E.F. Bolinder: Clifford algebra: What is it? IEEE Antennas and Propagation Society
Newsletter 29 (1987), 18-23.
D.K. Cheng: Field and Wave Electromagnetics. Addison-Wesley, Reading, MA, 1989,
1991.
G.A. Deschamps: Electromagnetics and differential forms. Proc. IEEE, 69 (1981),
676-696.
D. Hestenes: Space- Time Algebra. Gordon and Breach, Philadelphia, PA, 1966, 1987,
1992.
J.D. Jackson: Classical Electrodynamics. Wiley, New York, 1962, 1975.
B. Jancewicz: Multivectors and Clifiord Algebra in Electrodynamics. World Scientific,
Singapore, 1988.
G. Juvet, A. Schidlof: Sur les nombres hypercomplexes de Clifford et leurs applications
B l'analyse vectorielle ordinaire, B 1'6lectromagnetisme de Minkowski et B la thkorie
de Dirac. Bull. Soc. Neuchat. Sci. Nat., 57 (1932), 127-147.
I.V. Lindell: Methods for Electromagnetics Field Analysis. Clarendon Press, Oxford,
1992.
I.V. Lindell, A. Sihvola, S. Tretyakov, A. Viitanen: Electromagnetic Waves in Chiral
and Bi-Isotropic Media. Artech House, Boston, MA, 1994.
I. Lindell, P. Lounesto: Difierentiaalimuodot sahkomagnetiikassa. Helsinki University
of Technology, Electromagnetics Laboratory, 1995.
P. Lorrain, D.R. Corson: Electromagnetic Fields and Waves: Principles and Applica-
tions. W.H. Freeman, San Francisco, 1990.
A. Mercier: Expression des e'quations de l'e'lectromagne'tisme au moyen des nombres
de Clifiord. Thesis, Universit6 de Genkve, 1935.
E.J. Post: Formal Structure of Electromagnetics. North-Holland, Amsterdam, 1962.
with comments as facsimile by E.F. Bolinder, P. Lounesto (eds.), Kluwer, Dordrecht,
The Netherlands, 1993.
9
Lorent z Transformations
According to the Galilean principle of relativity the laws of classical mechanics

are the same for all observers (moving at constant velocity with respect to each
other). More precisely, the laws of classical mechanics remain the same under
Galilean transformations
direct inverse
x' = x - vt +
x = x' vt'
t' = t t =t'
relating two frames (3,t) and (x', t') moving at relative velocity v. The equa-
tions on the left show that the origin of the second frame x' = 0 corresponds
to uniform motion x = vt in the first frame. There is no privileged inertial
frame or absolute rest for moving bodies, but time is preserved, that is, time
is absolute.
The Galilean principle or invariance does not govern all of physics, most
notably electromagnetism and in particular light propagation. For instance,
the wave equation
is not preserved in a Galilean change of variables (3,t) + (x', t'). The wave
equation is instead invariant under another transformation, named after H.A.
Lorentz. In 1887, Michelson & Morley carried out an experiment which indi-
cated that light travels at the same velocity independent of the motion of the
source. In 1905, Einstein took the constancy of the velocity of light as a postu-
late, and showed that this postulate, together with the principle of relativity, is
sufficient for deriving the kinematical formulas of Lorentz. In so doing, Einstein
had to revise the notion of time, and abandon the concept of absolute time.
9.1 Lorentz transformations in one space dimension 119
9.1 Lorentz transformations in one space dimension
The simplest modification of the Galilean transformation, preserving linearity
and the implication x' = 0 + x = vt, is obtained by multiplying with a factor
7:
XI = y(x - vt), x = y(xt + vt')
where y is independent of x and t but may depend on v. We require that y is
the same in both equations since the inverse transformation should be identical
to the direct one except for a change of v to -v.
In computing y we use the observation of equal velocity of light. Consider a
light-signal travelling at velocity c in both frames, so that x = ct and x' = ct',
which substituted into the right-hand side of the previous equations results in
or, substituting x' = ct' and x = ct also into the left-hand side,
a formula admitting explicitly a transformation of time. Divide the two equa-

tions
which gives the factor

1
Next, compute the transformation of the time coordinate of events. Substi-

tute x' = y(x - vt) into x = y(xl + vt'),
x = y2(x - vt) + yvt',
use the explicit form of y, and solve for 2 ' )
Similarly, compute the inverse transformation,
The fact that time is also transformed is referred to as the relativity of time.
120 Lorentz Transformations
Summarizing, we have the following transformation laws for the space and
time coordinates
direct inverse
x - ut 2' ut' +
t' =
t-Lx
c2
t = t' + $2'
Jiq ,/-
known as the Lorentz transformation. Lorentz transformations preserve the
quadratic form x2 - c2t2= zr2- c2tr2and orthogonality of events; two events
2 c2tlt2 = 0. In particular,
x l , ctl and x2, ct2 are said to be orthogonal if ~ 1 x -
time and space are orthogonal.
It should be noted that time and space do not diverge by 90°, that is, they
are not 'perpendicular' or 'rectangular'. If we draw space-time coordinates
x, ct on paper so that the time-axis is 'perpendicular' to the space and perform
a Lorentz transformation, then the transformed coordinate-axes x', ct' are no
longer 'rectangular' (but they are orthogonal, by definition).
ct'
Write the direct Lorentz transformation in matrix form:
This resumes the composition of Lorentz transformations into multiplication

of matrices:
The composition of two Lorentz transformations at velocities u l and u2 results

9.2 The Minkowski space-time R311
in a Lorentz transformation L, = L,, L,, at velocity
a formula known as the relativistic composition of parallel velocities.
9.2 T h e Minkowski space-time R311

Space-time events can be labelled by points (xl, x2,x3, x4) or vectors x =
xlel + x2e2+ x3e3 +x4e4 in the Minkowski space-time R311. Indices are raised
and lowered according to
The Minkowski space-time R311 has a quadratic form sending a vector x E R3p1
to a scalar which we shall denote by x 2 ,
x2 = x; +x; + x i - x;.
Solutions to the equation
x; + x; + x$ = x: or x2 = 0
form the null-cone or light-cone.
Light-cone and light-like vectors

The set of non-zero vectors, or space-time intervals, x E R311 can be divided
into
x 2 > 0, space-like vectors,
x 2 < 0, time-like vectors,
x 2 = 0, null vectors or light-like vectors.
The set of time-like and light-like vectors can be divided into future oriented
x4 > 0, and past oriented x4 < 0 [recall that x4 = ct = -x4].
Planes passing through the origin can be divided into time-like, light-like
and space-like according as they intersect the light-cone along two, one or zero
light-like vectors.
Space-like unit vectors x2 = 1 form a connected hyperboloid, and time-like
unit vectors x 2 = -1 form a two-sheeted hyperboloid. Future oriented time-
'
like unit vectors correspond to observers; an observer travelling at velocity
v' E R3 is associated to the time-axis
v=
v' ce4 + where v2 = -1.
JzT
9.3 Lorentz boost at velocity v' E IR3

Let us review how a space-time event (?, t ) of an observer 0 is seen by another,
0', moving at velocity v' with respect to 0 .
0 = Observer
To do this we first decompose r' E JR3 into components. r' = Zll + 71 which are
v'
parallel = (F - v')- and perpendicular ? I = ? - Fll to Z. The transformed
ii2
space-time event is ( 7 ,t') where r" = ?Il + r': and
This transformation is called a boost at velocity v'.

The scalar r'2 - c2t2, where r'2 = (r'I2, remains invariant under a boost. A
boost leaves untouched the perpendicular component Fl, but alters the parallel
component Sll.
1 We consider only inertial observers; inertial = free of forces (no acceleration).
9.4 Lorentz transformations of the electromagnetic field 123
A boost at velocity v' = vlel + v2e2 + v3e3 can be represented by matrix
multiplication zta = ~ ~ p or
z pz' = Lz where
and
A boost is a special case of a Lorentz transformation, which can in general

also rotate the space JR3.
9.4 Lorentz transformations of t h e electromagnetic field

The electromagnetic field 2,sexperiences a boost at velocity v' as
A boost of the electromagnetic field leaves invariant two scalars

E'2-~2B'2 and l?.l?,
called Lorentz invariants.
A Lorentz transformation of the electromagnetic field can be written in coor-
dinate form as Flap= L ~ ~ F ~ ' ~ ( L or -concisely
~ ) " ~ as matrix multiplication
F' = LFL-'. The matrix F = (Fap) satisfies g ~ T g - l= -F, where
F is said to be Minkowski-antisymmetric.
2 The Lorentz invariants remain the same also under rotations of R3, and therefore under
the special Lorentz group S O ( 3 , l ) . This can be seen by squaring F = Ee4 - Be123 :
+
F2 = F J F F A F = I?' - 3' - 2(I?. d ) e 1 2 3 4The . scalar part remains invariant under
L E 0 ( 3 , 1 ) and the 4-volume part remains invariant under L E SL(4,R). Note that
O ( 3 , l )n SL(4,R ) = SO(3, l ) , SL(4,R ) / & z SO+ ( 3 , 3 ) and /\2 L E SO+ ( 3 , 3 ) acts on
/\2 ~ 3 , 2,
1 ~393.
9.5 T h e Lorentz group 0 ( 3 , 1 )
A matrix L satisfying L ~ =Lg ~ is said to be Minkowski-orthogonal or a
Lorentz transformation. The Lorentz transformations form the Lorentz group
0 ( 3 , 1 ) = { L E Mat(4,R) I L ~ =Lg ) . ~
A Lorentz transformation has a unit determinant: det L = f 1. The subgroup
with positive determinant,
is called the special Lorentz group. The special Lorentz group S O ( 3 , l ) has
two components. The component connected to the identity I is denoted by
S 0 + ( 3 , 1 ) ; it preserves orientations of both space and time. The other compo-
nent S O ( 3 , l ) \ SO+ ( 3 , l ) reverses orientations of both space and time.
The Lorentz group 0 ( 3 , 1 ) has four components; these form three two-
component subgroups preserving space orientation, time orientation or space-
time orientation. Time-orientation-preserving Lorentz transformations form
the orthochronous Lorentz group Ot(3, 1). A restricted or special orthochronous
Lorentz transformation L E S 0 + ( 3 , 1 ) preserves space-time orientation (orien-
tation of both space and time); its opposite - L E S O ( 3 , l )\SO+(3,1) reverses
space-time orientation, while gL reverses time orientation, and -gL E Ot(3, 1)
reverses space orientation, -gL E Ot(3, 1) \ SO+ ( 3 , l ) .
The Lorentz transformations, which stabilize a time-like vector, form a sub-
group 0 ( 3 ) , the orthogonal group of R3 = R310.The Lorentz transformations,
which stabilize a space-like vector, form a subgroup O(2, I), the small Lorentz
group of R211.The Lorentz transformations, which stabilize a light-like vector,
form a subgroup isomorphic to the group of rigid movements of the Euclidean
plane R2.
Any special orthochronous Lorentz transformation L E S 0 + ( 3 , 1 ) can be
written as an exponential L = e A of a Minkowski-antisymmetric matrix
0 b3 -b2 a1
-b3 0 bl a2
A=
+ +
-
which satisfies g ~ T g - l = -A. The matrix A can be characterized by two
+ +
vectors a' = ale1 a2e2 a3e3 and b = blel b2e2 b3e3 in R3. If = 0,
then L is a boost at velocity
= c tanhla'l.
3 Note the resemblance between LgLTg-' = I and the condition of orthogonality R R =~ I
of a matrix R, R E O ( n ) .
9.6 Simple Lorentz transformations 125
If a' = 0, then L E SO(3) is a rotation of the Euclidean space R3 around the
c
axis by the angle I;]. Boosts and rotations are special cases of simple Lorentz
transformations.
9.6 Simple Lorentz transformations

A special orthochronous Lorentz transformation L E S 0 + ( 3 , l ) , L # I, has
one or two light-like vectors as eigenvectors. If there are two light-like eigenvec-
tors, then they span a time-like eigenplane, which is preserved by the Lorentz
transformation; there is also a space-like eigenplane, which is completely ortho-
gonal to the time-like eigenplane. A special orthochronous Lorentz trans-
formation is called simple, if it turns vectors only in one eigenplane, leaving
the other eigenplane point-wise invariant. Disregarding the case L = I, a spe-
cial orthochronous Lorentz transformation L = eA, where A is characterized
as before by a', E R3, is simple if and only if a' . = 0. A simple Lorentz
transformation is called
hyperbolic, la'] > IS],
elliptic, -
la'l < I ~ I ,
parabolic, la'l = Ibl.
A hyperbolic Lorentz transformation is a boost for an observer, whose time-axis
is in the time-like eigenplane of the Lorentz transformation. An elliptic Lorentz
transformation is a rotation of the Euclidean space R3, which is orthogonal
to an observer, whose time-axis is orthogonal to the space-like eigenplane of
the Lorentz transformation. A parabolic Lorentz transformation has only one
light-like eigenvector; it is of the form
1
L=I+A+-A', since A 3 = 0 ,
2
and has only one eigenplane, which is light-like and tangent to the light-cone
along the light-like eigenvector. A non-parabolic Lorentz transformation can be
written as a product of two commuting simple transformations, one hyperbolic
and the other elliptic.
Lorentz transformations can be described within the Clifford algebras C.e3 --

Mat(2,6), C.e3,1 N Mat(4,R) and Cllr3 N Mat(2,W).
4 Completely orthogonal planes have only one point in common, the origin 0 . For two
+
vectors x , y in completely orthogonal planes, the scalar product x . y = xlyl +
x2y2
XBYQ - x4y4 vanishes: x . y = 0.
9.7 In t h e Clifford algebra C t 3 2: Mat(2, C )
Events in time and space can be labelled by sums of scalars and vectors,
in R $ R3 c C t 3 . A paravector x = x0 + xlel + x2e2+ x3e3 can be provided

with a quadratic form
making IIB $ R3 isometric to the Minkowski time-space R1t3. This quad-

ratic form is preserved in a special orthochronous Lorentz transformation L €
S0+(1,3),
L:IIB$IR~-,IW$IIB~,x - , ~ ( x ) = s x i - ~ ,
where s is in the spin group ti
$pin+(l,3) = { s E C t 3 I sS = 1 ) E SL(2, C ) .
+
The time-space event x = t 2 E R $ R3 and the electromagnetic field
F = I? - g e l a s E R3 $ /\'R3 behave slightly differently under restricted
Lorentz transformations:
x' = L(x) = s x i - l ,
F' = sFs-'.
The spin group $pin+ ( 1 , 3 )is a two-fold covering of the special orthochronous
Lorentz group S 0 + ( 1 , 3 ) .In other words, there are two elements f s in the
group $pin+ ( l , 3 ) inducing the same Lorentz transformation L in SO+ ( 1 , 3 ) .
This can be expressed by saying that the kernel of the group homomorphism
consists of two elements {f1 ) E $pin+(l,3) [the kernel is the pre-image of the
identity element I E SO+ ( 1 , 3 ) ] .
Every element s in the spin group $pin+(l,3) is of the form
1
s = f exp ;;(a'+ be123)
-
where a' and are vectors in R3. The minus sign in front of the exponential
5 The raising and lowering conventions are different in R $ R3 and I R 1 1 3 . In ~ ' 1 ~
, l , x 2 ,x3) = (10, - X I , -12, - 1 3 )

( x Ox whereas in R3 C IR $ R3 there is a prescribed
metric such that ( x 1 , x 2 , x 3 )= ( X I , x 2 , x3).
6 The groups $pin+ ( l , 3 ) C Ce3 and Spin+(l,3) C Cel,3 are isomorphic, and so are their
Lie algebras R3 $ A2 R3 and A2 ~ ' 1 ~ .
9.8 In the Clifford algebra C&,l Mat(4, R) 127
is needed, because not all the elements in the two-fold cover $pin+(l, 3) of
SO+ (1,3) can be written as exponentials of para-bivectors in R3 @ /\2 IK3. In
the case a' = 0 the Lorentz transformation is a rotation of IK3, and in the case
+
b = 0 we have a boost at velocity
a'
v' = tanh a' = -5- tanhla'l.
1.
For any s E $pin+(l, 3) the product sg is a boost, that is, s i E R @ IK3.
Since ( ~ s ' >
) ~0, there is a unique square root of u = sg, a boost such that
f i = a ( l + u) , a > 0. Squaring both sides and using u l = 1 results in
Write b l = 6.The product r = b l 1s satisfies f r = 1 and Fr = 1, and so

it is a rotation, r E Spin(3). A special orthochronous Lorentz transformation
can be uniquely decomposed into a product of a boost and a rotation,
called the polar decomposition. Similarly computing b2 = 6 and r = sbzl,

we find that s = rba with the same rotation r, that is,
Both the decompositions have as a factor the same rotation r E Spin(3) =

{s E ti I sg = l), but the boosts are different: bl # b2.
9.8 I n t h e Clifford algebra C t 3 , ~E Mat(4,R)

A boost b E R @ R3e4, at velocity v' E R3, can be computed by
and results in
1 + y ( l + v'e4) 1
obtained also by taking a square root of b2 = y(1 v'e4). +

The restricted Lorentz group S 0 + ( 3 , 1 ) has a double cover
7 If la'l = lq and a'. $ = 0 then (a'+ $e123)2 = 0, and for a non-zero F = a' + gel23 there
is no para-bivector B E R3 $ A2R3 such that eB = - e F .
8 The first tanh-function is evaluated in the Clifford algebra Ce3.
Under a Lorentz transformation induced by s E Spin+(3,1) the space-time
vector
F=;
-, -
x transforms according to x' = sxs-l and the electromagnetic bivector
~ -eBelz3
~ transforms according to F' = sFs-l.
9.9 I n the Clifford algebra Celts Mat(2, W)

Consider the Lorentz group of the Minkowski time-space IR1l3 in the Clifford
algebra Cel,3 which is isomorphic, as an associative algebra, to the real algebra
of 2 x 2-matrices Mat(2,W) with quaternions as entries. The Clifford algebra
Ct1,3 is generated as a real algebra by the Dirac gamma-matrices yo, yl ,y2, y3
satisfying
In this case the Lorentz groups 0 ( 1 , 3 ) , S0(1,3), S0+(1,3) are doubly

covered by
-
P i n ( l , 3 ) = {s E cet3u C t , , I sii = f1),
Spin(l,3) = {s E Cet3 I sii = f1),
Spin+(l, 3) = {s E cet3 I sZ = 1) SL(2, C).
A Lorentz transformation L E 0(1,3) is given by L(x) = sxi-l in general,
but a special Lorentz transformation L E S0(1,3) corresponds to an even
s and can also be written as L(x) = sxs-l. The group homomorphism p :
P i n ( l , 3 ) + 0(1,3) is fixed by L = p(s), L(x) = sxi-l, and its kernel is
{f1), that is, each L E 0 ( 1 , 3 ) has two pre-images fs in P i n ( l , 3 ) .
An element s E Spin+(l, 3) has a unique polar decomposition
where the boosts are different,
bl = 4- and b2 = J-,
but the rotation is the same, r = b l l s = sbzl.
Penrose & Rindler 1984. On pp. 31-32 the authors give a geometric inter-
pretation for Lorentz transformations, reviewed here in terms of the Clifford
algebra C t l , ~Take
. four distinct light-like vectors a, b , c, d such that a . b = 1
and c . d = 1. The bivector a A b represents a time-like plane, since (aA b)' =
( a . b)2 - a 2 b 2= 1; the bivector a A b belongs to Spin(l,3) \ Spin+ ( l , 3 ) and
represents a Lorentz transformation, which reverses the space-time orientation.
Therefore, the product
s = (aAb)(cAd)
9.9 In the Clifford algebra Ctl,3 E Mat(2, IN) 129
is in Spin+(l,3). Let the light-like eigenvectors of the corresponding Lorentz
transformation be l1 and 12, and choose 11 .12= 1 SO that (11 A 12)' = 1. The
bivector l1 A 12 anticommutes with a A b and c A d, that is, it is the unique
'normal' to aA b and c Ad. The bivector F = log(s) in is determined
/ \ 2 ~ 1 1 3
up to a multiple of 2~70123(11 A 12). The square root 4 $70123 = @ is +

such that F = f(4 $70123)11+ A 12; it is determined up to a sign; choosing
>
4 0, the Lorentz transformation L = p(s) has velocity v = tanh(24) and
eigenvalues
The planes a A b and c A d 'differ' in the sense that (a A b)s = c A d and

s(c A d) = a A b by a sum of an elliptic angle 1C, about the plane 11 A 12 and
a hyperbolic angle 4 in the plane l1 A 12. Indicating the transformed light-like
vectors by primes,
we find that c' A d' = s(c A d)s-I = (a A b)(c A d)(a A b)-l, that is, the
Lorentz transformation reflects the plane c A d across the plane a A b. But,
s ( a ~ b ) s -= ) - I ( c ~ d ) ( a ~ b ) ( c ~=
' ( c ' ~ d ' ) ( a ~ b ) ( c ' ~ d 'and d )s - ' ( a ~ b ) s ,
that is, the inverse of s(a A b)s-l . Take a square root of the inverse of s,
within Spin+(l,3), and find that

UaU-l = c and ubu-I = d,
a kind of 'half' of the reflection above. I
Jancewicz 1988. On pp. 252-256 the author shows how to decompose a non-
simple bivector F into simple components. He defines
and sets S2 = a 2 + P2. Then he gives the simple components

F
F1,2 = -(s f ff P70123)
26
+
for F = F1 F2 so that Fq > 0 and F: < 0, that is, F1 is hyperbolic or
time-like and F2 is elliptic or space-like. Observing that
enables us to work out the decomposition in another way:
F1 =
FdJ and F2 = F$~0123
4+ $yo123 4 +$~0123 '
Hestenes & Sobczyk 1984 p. 81 note that 4' = F:, $' = -Fz and
(their formula 4.16 concerns only the positive definite case). I

Hestenes 1966. The author gives on pp. 52-53 a method to find out s E
Spin+ ( l , 3 ) from the coordinates LP, of a special orthochronous Lorentz trans-
formation L ( x ) = sxs-' , L E SO+ ( 1 , 3 ) . Recall that L(y,) = y, L", , and
deduce
LP v - YP . L (7")= ( y P s 7 v ~ ) 0 .
To compute s in terms of LP, define first
It follows that
L = L(y,)yV = sy,SyV .
In computing r , i y Y , note that in general e,uev = ( n - 2k)Q for u E Ak R n ,
+ +
and deduce that for s = ( s ) ~ (s)2 (s)4
Therefore,
L = 4s[(s)o- ( ~ ) 4 ] .
- ( s ) ~ ] 'and
Since i s = 1, EL = 16[(s)~, ,
Substituting (1) this gives s explicitly as a function of LP,. This construction

is an accident in dimension n = 4, because only then does the sum e,uev
vanish for a bivector u E A2 Rn. I
9.9 In the Clifford algebra Celts E Mat(2, W) 131
Historical survey
In 1881, A.A. Michelson carried out, for the first time, measurements intended
to determine the motion of the Earth relative to an absolute, imaginary 'light
medium'. For this purpose he measured the velocity of light in different direc-
tions. Michelson& Morley repeated the experiment in 1887 and came to the
conclusion that light travels at the same velocity independent of the motion of
the source with respect to the 'light medium'.
Voigt in 1887 was the first to recognize that the wave equation
is invariant with respect to the change of variables
where also time is transformed. Voigt's formulas are not identical for direct
and inverse transformations; symmetry was restored later by introducing the
factor d-. This factor was first encountered in another connection:
FitzGerald and Lorentz 1892 gave independently an explanation of the Michel-
son & Morley experiment by suggesting that moving bodies are contracted in
the direction of motion by the ratio d m - .
The Lorentz transformations of space-time events were introduced by Larmor
in 1900, while the relativistic covariance of the Maxwell equations was demon-
'
strated by H.A. Lorentz 1903 (and conformal covariance by Cunningham
1909/1910 and Bateman 1910).
In 1905 l o Einstein supplemented the principle of relativity by postulating the
principle of independence of the velocity of light (of the motion of the source).
These two principles led Einstein to a revision of the notion of time and enabled
him to deduce the kinematical transformation laws of Lorentz; his predecessors
had obtained the transformation laws by considering transformations which do
not change the form of the Maxwell equations.
9 Poincari noticed that restricted Lorentz transformations of space-dimension 1 form a
group SO+ ( 1 , l ) consisting of the elements
coshx sinhx
sinh x coshX
where x E R.
10 A. Einstein: Zur Elektrodynamik bewegter Korper. Ann. Physik 17 (1905), 891-921. In
this paper Einstein compared the same phenomenon when observed in two different frames:
a magnet moving near a closed conductor and a closed conductor moving near a magnet.
In another paper of 1905 Einstein gave a relation between mass and energy, which was
later popularized as the formula E = mc2, written today as E = mc2 /J- or
+
E2 = m2c4 p 2 ~ 2 .
Later Einstein reformulated the principle of relativity so that it embraces
not only mechanical but also electromagnetic phenomena:
] the laws of physics have the same form in all reference frames 1
When this Einsteinian principle of relativity is applied to the Maxwell equa-
tions, one is compelled to conclude that the velocity of light is the same in all
reference frames. In other words, the principle of constancy of the velocity of
light becomes superfluous as an amendment to the principle of relativity. The
principle of relativity and knowledge of the Maxwell equations are enough to
deduce the transformation laws of Lorentz.
Nowadays the terms 'relativistic' and 'relativity' almost invariably refer to
the Einsteinian principle.
Questions
1. How many light-like eigenvectors does a Lorentz transformation have?
2. Are all L E SO+ ( 3 , l ) of the form L = exp(A), 9 ~ ~ g =- -A?
l
3. Are all s E Spin+(3,1) of the form s = exp(B/2), B E A2JR31'?
4. A special orthochronous Lorentz transformation can be written as a
product of a boost and a rotation, in two different orders. In the two
expressions, which factor is the same: the boost or the rotation?
5. Are all the special orthochronous Lorentz transformations products of two
commuting transformations, one hyperbolic and one elliptic?
Let B E
6. Is 3 x + .xu-',
JR3t1 +
u = 1 B + ;B A B , a Lorentz transformation?
7. Do the Lorentz transformations induced by exp(B/2) and
+ +
1 B ;B A B , B2 # 1, have the same eigenvectors?
+
8. Does (1 B)(1 - B)-' represent a Lorentz transformation?
9. Do the Lorentz transformations induced by exp(B/2) and
(1 + B)(1 - B)-' , B2 # 1, have the same eigenvectors?
Answers
1. In general two, parabolic has one, fI have all of them.
2. Yes. 3. No. 4. Rotation. 5. No (parabolic are not).
6. Yes, if B2 # 1.
7. Yes, because both the Lorentz transformations are functions (= power
series with real coefficients) of A, A(x) = B L x; namely e A and
+
(I A)(I - A)-', respectively.
9.9 I n the Clifford algebra C.tl,3 2: Mat(2, W) 133
8. Yes, if B2 # 1 (but this is no longer true in dimension 6).
9. No, because the latter Lorentz transformation is not a rational function of
A alone (but also of AT).
Exercises
1. Derive the composition rule for non-parallel velocities,
Hint: use the inverse Lorentz transformation
t i = J(t2 + y).
1- 3
2. Show that v", = tanh(2 log(b',)) where b', = m, s', = sls2 E R $ R3
and sl = exp($il), Zl = tanh(a'1).
3. Show that the composite of two boosts is a hyperbolic transformation.
4. Consider a time-space event x = ct + i in R $ IK3 corresponding to
y = 2 + cte4 in IR3l1. Define s = exp(a'12) and u = exp(a'e4/2) for a' E R3.
Show that the boost ixs-l = s-'xi corresponds to the boost uyu-'.
5. Show that for u E Spin+(3, l ) , when decomposed into a product of a
boost and a rotation, u = blr = rb2, the rotation-factor r E Spin(3) can
be obtained by normalizing (u A e4)e41.
6. Take a bivector F = a'e4+ ;e123 E /\2R311 such that la'l = 151. Consider
the antisymmetric linear transformation R311 + R3t1, x + Ax = ( F x ) ~ .
Show that (A3x) 1) X.
7. Take a non-simple bivector F E /\2 R311 with simple components
+
F = F1 F2, Fq > 0, Fg < 0. Show that
+ R3J, FxF
R3J X +-
Ff - FE
is a Lorentz transformation, a reflection across the plane of F1.
+ +
8. Show that (4 $~0123)(6 a - P~oiz3)= 264.
9. Show that as topological spaces Spin+ ( l , 3 ) 2: IK3 x S3.
10. Show that as groups $pin+ ( 3 , l ) 2: Spin+ ( 3 , l ) 2: SL(2, C) and
S 0 + ( 3 , 1 ) 2: SO(3, C) = {R E Mat(3, C) I RRT = I, det R = 1).
11. Show that for u E Spin+(3,1) there is a square root in Spin+(3,1) given
Lorentz Transformations
+
Hint: for s E Spin+(3, l ) , s2 1 = s2 +Zs = ( s + Z ) s = 2((s)o + (s)4)s.
+ +
Therefore, (a pel234)fi = u 1 with a , P E R.
Bibliography
V. Bargmann, E.P. Wigner: Group theoretical discussion of relativistic wave equa-
tions. Proc. Nat. Acad. Sci. U.S.A. 34 (1948), 211-223. Discussion on unitary
representations of the PoincarC group.
Solving with MAPLE V. Birkhauser, Boston, MA, 1994.
A. Einstein: Zur Elektrodynamik bewegter Korper. Ann. Physik. (4) 17 (1905),
891-921.
I.M. Gelfand, R.A. Minlos, Z.Ya. Shapiro: Representations of the Rotation and Lorentz
Groups and their Applications. Pergamon, New York, 1963.
D. Hestenes: Space-Time Algebra. Gordon and Breach, Philadelphia, PA, 1966, 1987,
1992.
B. Jancewicz: Multivectors and Clifford Algebra in Electrodynamics. World Scientific,
Singapore, 1988.
W. Kopczyliski, A. Trautman: Spacetime and Gravitation. PWN, Warszawa; John
Wiley, Chichester; 1992. Includes a historical survey on relativity.
M.A. Naimark: Linear Representations of the Lorentz Group. Macmillan, New York,
1964.
R. Penrose, W. Rindler: Spinors and Space-Time. Vol. 1. Cambridge University
Press, Cambridge, 1984.
with comments as facsimile by E.F. Bolinder, P. Lounesto (eds.), Kluwer, Dordrecht,
The Netherlands, 1993.
L. Silberstein: The Theory of Relativity. Macrnillan, London, 1914.
R.F. Streater, A.S. Wightman: PCT, Spin and Statistics, and all that. Benjamin,
New York, 1964. Contains a discussion on the two-fold cover of the Lorentz group.
E.P. Wigner: On unitary representations of the inhomogeneous Lorentz group. Ann.
of Math. 40 (1939), 149-204. A historical article in which is discussed, for the first
time, the physical role of the Poincare'group.
10
The Dirac Equation
The Schrodinger equation describes all atomic phenomena except those involv-
ing magnetism and relativity. The Schrodinger-Pauli equation takes care of
magnetism by including the spin of the electron.
The relativistic phenomena can be taken into consideration by starting from
the equation E 2 / c 2 - j52 = m2c2. Inserting energy and momentum operators
into this equation, results in the Klein-Gordon equation
which treats time and space on an equal footing. Dirac 1928 linearized the
Klein-Gordon equation, or replaced it by a first-order equation,
The above Dirac equation implies the Klein-Gordon equation provided the
symbols y,, satisfy the relations
Dirac found a set of 4 x 4-matrices satisfying these relations, namely, the

following Dirac matrices:
In terms of the Pauli spin-matrices uk the Dirac gamma-matrices y,, can be

136 The Dirac Equation
expressed as '
Writing xo = ct, the Dirac equation can be condensed into the form
where 8, = =. 8
An interaction with the electromagnetic field
cluded via the space-time potential (A0,A', A2,A3) = ( t V ,A,, Ay, At) of F""
FpV is in-
by employing the replacement itid" + itidp - eAp. This leads to the conven-
tional Dirac equation
yp (itid" - eAj')$ = mc$
where the wave function is a column spinor, that is,
The Dirac equation takes into account the relativistic phenomena and also spin;
it describes spin-$ particles, like the electron.
10.1 Bilinear covariants

The Dirac adjoint of a column spinor 1C, E @ is a row matrix
A column spinor $(x) and its Dirac adjoint $t(x)70 can be used to define four
real valued functions
J"(4= $+(x)707"$(x)
which are components of a space-time vector, the Dirac current,
J (x) = y,, Jp(x).
Under a Lorentz transformation
1 The above matrix representation is called the Pauli-Dirac representation (although it

should be called the Pauli-Dirac basis).
2 Note that +'a,, = Y,,Wwhere = &.
3 The Dirac adjoint of 4 is often denoted rby 4, but we have reserved this bar-notation for
the Clifford-conjugation.
10.1 Bilinear covariants
the Dirac spinor transforms according to
and the Dirac current according to
Thus, the Dirac current is covariant under the Lorentz transformations. The
components J p = $tyoyj'$ are called bilinear covariants.
The physical state of the electron is determined by the following 16 bilinear
covariants:
Their integrals over space give expectation values of the physical observables.
>
The quantity J0 = $t$, J0 0, integrated over a space-like domain gives
the probability of finding the electron in that domain. The quantities Jk =
$tyoyk$ (k = 1,2,3) give the current of probability f = ykJ ~ they ; satisfy
the continuity equation
l d J O+ - = o .
--
c dt dxk
>
The Dirac current J is a future-oriented vector, J2 0. The time-component
uo = yo . u of the unit vector u = J/.\/S;~, J2# 0, gives the probable velocity
of the electron,
The bivector S = $SpVyP, is usually interpreted as the electromagnetic

moment density, while it gives the probability density of the electromagnetic
moment of the electron.
4 The quantities $tyoyfi$ are actually quadratic in $. Also their polarized forms $tyoyfip
are not bilinear but rather sesquilinear, while anti-linear in ++!J.
5 Or rather the probability multiplied by the (negative of the) charge of the electron. In the
case of a large number of particles J0 can be interpreted as the charge density.
6 Recall that JZ= (5')' - (J1)' - (5')' - ( J ~ ) ' .
fr
7 This is a shorthand notation for S = Cp,u Sfiuyfiu= Ell<,, Sfiuyfiu.
The vector K = ICfly, is space-like, and such that K 2 = -J2. It is ortho-
gonal to J , K . J = 0, and gives the direction of the spin of the electron, the
spin vector $fi~K /2G # 0. Note
, that ICfl = $tyOyfliy0123$.
The first and last of the bilinear covariants were combined into a single
quantity by de Broglie:
I fi = a1 + fi2~01231
Note that a2 = -$ty~Yo123$.
SPINORS
IN IDEALS
Here we shall take a new view on spinors and regard them as elements of
minimal left ideals, first in matrix algebras, then in complexified Clifford
algebras, and finally in real Clifford algebras.
10.2 Square matrix spinors

Usually the wave function is a column spinor $ E c,
but we shall also regard
it as a 4 x 4-matrix with only the first column being non-zero; that is, $ E
Mat(4, @) f where f is the primitive idempotent
More explicitly, a Dirac spinor might appear as a column spinor or as a square

matrix spinor: l o
8 We shall reject ideal spinors later in favor of spinor operators.

9 The factors i(l +yo) and $ ( I + iylz)are energy and spin projection operators.
10 We replace column spinors by square matrix spinors in order to be able to get everything
- vectors, rotations and spinors - represented within one mathematical system, namely
the Clifford algebra.
10.3 Real structures and involutions 139
+ + +
where 11,, E C. In the latter case 11, = $1 fl $2 f2 $3 f3 11,4f4 expressed in
a basis of the complex linear spinor space S = (C @ Cel,3) f ,
10.3 Real structures and involutions

Although we have an isomorphism of real algebras C 8 Mat(4, R) C @ Cll,3,
the complex conjugations are not the same in C @ Cel,3 and C @ Mat(4, R) E
Mat(4, C). In the matrix algebra Mat(4, C) we take complex conjugates of
-
the matrix entries u* = (ujk)* = (ujtk), whereas in the complexified Clifford
algebra C@Cel,3complex conjugation has no effect on the real part Cel,3, and
we have u* = (a + ib)* = a - ib for a, b E CllP3. Thus there are two different
complex conjugations (real parts) in the algebra C8Ce1,3 N Mat(4, C). This is
referred to by saying that there are two different real structures l1 in the same
complex algebra.
To make this point more explicit, the following table lists some correspon-
dences of involutions.
complex conjugate u* ~ 0 1 3 ~ * ~ 2 3
~013~*~$3 u* complex conjugate
grade involute ii
reverse u
- ^l0123~~;\3
7134
'
T -1
713
T -1
Clifford-conjugate ii Y02U Yo2
~13~~1-,l uT transpose
Yoii*~;~ ut = u * ~ Hermitian conjugate
ii* youty;' Dirac adjoint
11 Not to be confused with the complex structure of an even-dimensional real linear space, a
real linear transformation J such that J' = -I.
dimension degrees (u)k E I \ JR113,
~ has three important involutions:
+
fi = (U)O- ( ~ ) 1 ( ~ ) - + (u)4,
2 (u)3 grade involution,
6 = (U)O+ (u)i - (u)2 - (u)3 + (u)4, reversion,
= (U)O- ( ~ ) -i (U)Z+ (u)3 + ( ~ ) 4 , Clifford-conjugation.
The reversion and Clifford-conjugation are anti-automorphisms satisfying .li?; =
56, ?iF= Gii, whereas the grade involution is an automorphism iiij = u6. These
three involutions are extended to C @ Cel,3 as complex linear functions, that
is, for X E C and u E C11,3 we have (Xu)^ = Xfi, (Xu)' = XG, (Xu)- = Xii,
whereas the complex conjugation is by definition anti-linear: (Xu)* = Y u .
Complex conjugation is of course an automorphism, (uv)* = u*v* for u, v E
c 8 celS3.
10.4 C o m p a r i s o n of real p a r t s / s t r u c t u r e s
Note that the real part and the complex conjugate of a Dirac spinor depend
on the decomposition (in the real structure) singling out the real part. For
$ E Mat(4, C ) f :
For $ E (C 8 cel,3)
f [viewed as a matrix]:
The Dirac spinor $ might appear as a column spinor $ E C' or else as a square
matrix spinor $ E Mat(4, C)f or as a Clifford algebraic spinor $ E (C@Cel,3)f
where the last two differ in their real structures.
I m p o r t a n t Note. To indicate in what real structure the real part and the
complex conjugate are taken we write
Re($) in Mat(4, C)f and $* in Mat(4, C)f

or
Re($) in @ @ Cel,3 and $* in C 8 Cel,3.
Other contextual indicators are the Hermitian conjugation [either $t yo is a row
spinor or else it is in Mat(4, C)] and for instance the reversion [the composite
10.5 Bilinear covariants via algebraic spinors 141
of the reversion and the complex conjugation 4*

in C . 8 corresponds to
$tro in Mat(4, C)].
The reader should also observe that the real part Re($) in @ 8 carries
the same information as the original Dirac spinor $ E C? [in contrast to Re($)
in Mat(4, @) f]. I
Exercises 1,2,3,4,5
10.5 Bilinear covariants via algebraic spinors

For a column spinor $ E C? the Dirac adjoint is a row matrix
but for a square matrix spinor $ E Mat(4, C)f the Dirac adjoint is a square
matrix
with only the first row being non-zero.

The components of the Dirac current can be computed as follows for column
spinors, square matrix spinors and Clifford algebraic spinors
where the factor 4 appeared because
has scalar part $, that is, (f)o = a, while trace(f) = 1. The current vector is
the resultant
Similarly $ E C? carries a real bivector S with components
for which S,, = -y,, JS and S = $ S , , y ~ " . In various formalisms
The Dirac adjoint $tyo of a column spinor $ E C? corresponds to $* of an

algebraic spinor II, E (@ 8 Cel,3) f . The current vector J and the bivector S
are examples of bilinear covariants listed below for a column spinor $ E @
and for an algebraic spinor )I E (C 8 Cel,3) f .
Later we shall need the following aggregate of bilinear covariants Z = O1 J + +

+ +
i s i K ~ 0 1 2 3 fi2~0123.
SPINORS
AS OPERATORS
Here we shall view spinors as new kinds of objects: rather than being some-
thing which are operated upon they are regarded as active operators. The
big advantage is that the physical observables, which were earlier calculated
component-wise, can now be obtained at one stroke.
10.6 Spinor operators Q E c",~ 143
10.6 Spinor operators Q E CcP3
We will associate to a Clifford algebraic spinor $ E (@ @ CelT3)f [viewed here
as a matrix] the mother spinor [this will be the mother of all real spinors]
and the spinor operator
From the mother spinor @ E Cll,33(l +yo) we may reobtain the original Dirac
spinor
[that is, the square matrix spinor, not the column spinor], and from the spinor
operator Q we may reobtain the mother spinor @ = Q(l +yo) and the original
Dirac spinor
1 1
II,= Q-(1
2 +
YO)-(1
2 +
iylz) E ((C @ f
Note that the spinor operator is invertible if
1$112+ - 1$312 - 1$412 # O and 21m($;$3+$;$4) # 0,

or equivalently !P\k # 0, the inverse being
Multiplication by i = flcorresponds to right multiplication by the bivec-

tor
that is, ill, = $72 y1 for $ E C4. In other words, the real part of i$, $ E
(@@Cel,3)f , is the mother spinor @y2yl whose even part is the spinos operator
QYZYI,4 Re(even(i$)) = ' 3 ~ 2 ~ 1 .
Decompose the mother spinor @ E Cel,3i(l + y o ) into even and odd parts
@ + + + +
= 0 0 @1= (@o @1)$(I yo) = $(a0 %yo) $(@I @0^)'0). + + It follow^
that 00 = Qlyo and = Ooyo. Taking the real part [in C 8 Cll,3] of the
Dirac equation (id - eA)$ = m$ results in
which decomposes into even and odd parts [a0 = even(@), = odd(@)]
Therefore, the even part of the mother spinor, the spinor operator, satisfies the
equation l 2
where Q : R1s3 + ~ l t ,In~ this

. Dimc-Hestenes equation the role of the Dirac
column spinors is taken over by real even multivectors, which are not in any
proper left ideal of the Clifford algebra Cll,3.
Comments. 1. Under a Lorentz transformation x -+ sxi-l , s E P i n ( l , 3 ) ,
x E R1s3, a Dirac spinor $ E Mat(4, C)f 2 (C 8 Cll,3)f transforms according
to $ -+ s$, and a spinor operator Q E clt3 transforms like this:
Q + sQ when s E Spin(l,3),
Q -+ sQyo when s E Pin(l,S)\Spin(l, 3).
This can be seen by the definition Q = 4Re(even($)) and using $ = $f,
+
f = $0 yo)$(l+ iy12).
Note that the so-called Wigner time-reversal is not represented by any s E
Pin(1, 3)\Spin(l, 3).
2. The Dirac-Hestenes equation has been criticized on the basis that it is
not Lorentz covariant because of an explicit appearance of the two basis ele-
ments yo and 712. This criticism does not hold. The Dirac-Hestenes equation
is Lorentz covariant in two different ways: first, we can regard yo and 712
as constants and transform Q to sQ; secondly, we can transform yo, 712 to
syos-', sy12s-l and 9 to sQs-l, s E Spin+(l, 3).
3. In curved space-times spinor fieldslbundles [functions with values in a mini-
mal left ideal of a Clifford algebra] exist globally only under certain topological
conditions: the space-time must be a spin manifold [have a spinor structure]. It
has been argued that since even multivector functions exist on all oriented man-
ifolds, the theory of spin manifolds is superfluous. This argument is misplaced
since yo and 712 do not exist globally.
12 Note that 4Re(even(x$)) = X Q T ~for x E R1'3.
10.7 Bilinear covariants via spinor operators 145
However, the physical justification of the theory of spin manifolds could be
questioned on the following basis: why should we need to know the global
properties of the universe if we want to explore the local properties of a single
electron?
4. The explicit occurrence of yo and 712 is due to our injection U? + S =
(C 8 Cel,3)f. In curved manifolds it is more appropriate to use abstract rep-
resentation modules as spinor spaces and not minimal left ideals [nor the even
subalgebras] of Clifford algebras. The injection ties these spaces together in a
manner that singles out special directions in R1v3. I
10.7 Bilinear covariants via spinor operators

Write as before 4 = 4 Re(even(+)) [real part taken in C 8 Cel,3]. Because of
the identities l3
we call 4 a spinor operator. In the non-null case 4% # 0 the element 4

operates like a Lorentz transformation composed with a dilation [and a duality
transformation]. In coordinate form
For later convenience we introduce P = 0 J , fi = fi1 + + fi270123, and

+
Q = S KyOlz3.We have the following identities:
9 ( 1 +yo)@ = P, + yo)y12% = &,

~ ( 1 yo)yij@
+ = Q Q ~ (ijk CYC~.,
Q3 = Q ) ,
4 ( 1 + yo)(l + iylz)@= Z [= P + iQ].

Hestenes 1986 p. 334 gives P, -Q and Z in (2.26), (2.27) and (2.28).
Exercises 6,7,8
13 The Dirac-Hestenes equation d Y y Z l -eAF = mY-yo coptains 721 , y o explicitly. It follows
that y o , y 1 2 must be explicit in J = Q y o 8 , S = Qy12 Q .
10.8 Higher-dimensional analogies f o r spinor o p e r a t o r s
In the case of the Minkowski space-time IR1l3 the spinor space is a minimal left
ideal (C@CelP3) 4
f induced by the primitive idempotent f = (l+yo) $(l+iy12).
In the primitive idempotent we have projection operators for energy $(I yo) +
+
and spin $(I iylz). In other words, spin is quantized in the y3-direction or
more precisely in the yl y2-plane.
In the case of a higher-dimensional space-time, say R1v5 with an orthonormal
basis {yo, yl, . . . ,751, the spinor space is a minimal left ideal (C @ CelB5)f
induced, for instance, by the primitive idempotent
The spin is quantized in the yly2-plane and the y3y4-plane. The procedure of
taking the real part and the even part does not result in an invertible operator,
since 4 Re(even(f)) = $(I - x234). In other words, for a spinor in a minimal
left ideal+ E (C @ Ct?1,5)f the 'spinor operator' is also in a left ideal, 4 =
4 Re(even(+)) E ~!:,~3(1 - ylZ34). We conclude that there is no analogy for
spinor operators in higher dimensions.
A p p e n d i x 1: Discussion o n t h e role of i = f l i n QM
Are there superj7vous complex numbers in the present formulation of quantum
mechanics? Is it possible to get rid of some complex numbers in QM? To
answer these questions, we present analogies which become step by step closer
to the present situation in quantum mechanics.
Analogy # 1. Consider someone who uses only the line y = x in the complex
plane C, that is, someone who does not use all the complex numbers z = x+iy,
but instead restricts himself to complex numbers of the form x+ix. This person
could equally well restrict himself to the real axis and consider instead only the
+
real part x = Re($ ix). In terms of the picture
Im y=x
x+iy =x+ix while y = x
Re
2
this would mean a projection from the line y = x onto the real axis y = 0 with
no information lost.
10.8 Higher-dimensional analogies for spinor operators 147
This analogy/picture could be criticized by arguing that the product w of
+
two complex numbers of the form x ix is not of the same type, that is,
Re(w) # Im(w). I
Analogy # 2. The sums and products of matrices of type
provide an isomorphic image for addition and multiplication of the reals x E R.

It is of course more economical to use just the real numbers x E IW instead of
the real 2 x 2-matrices X E Mat(2,R). I
Analogy # 3. If we have a complex matrix
1 x+iy -y+ix
s =2- (y - i x X + ~ Y
then the real part, multiplied by two, Z = 2 Re(S), that is,
obeys the same addition and multiplication rules as S and carries the same
information as S [contained in the pair (x, Y)].Note that for a complex number
+
z = x iy we have S = $ ( I - a n ) , where the matrix f = ;(I - a 2 ) is an
idempotent satisfying f = f . I
Situation in QM. In the present formulation of quantum mechanics one uses
column spinors 4 E @, which could be replaced without loss of generality by
spinors in a minimal left ideal of the complex Clifford algebra $ E (@@ C e l , ~f),
+ i(l+
f = i ( 1 yo) iy12). Spinors in minimal left ideals 11 E (@ @ Cel,3)f
can be replaced without reduction of information by spinor operators 4 =
4 Re(even(4)) E eel3. No information is lost in this replacement, because the
+ +
original spinor can be recovered as 4 = 4 f , f = $(I yo) - ( 1 ir12). I
Appendix 2: Real ideal spinors 4E i ( 1 - 703)

This appendix is included mainly for historical reasons. The concept of a spinor
operator 4 E Ce13 was introduced by Hestenes 1966. In his invention he used
as an intermediate step the real ideal spinor
and not the mother spinor E Cel,3 i ( 1 + T ~ ) , = 4 Re($), 4 E (@@ Cel,~)

f.
The ideal spinor contains the same information as the mother spinor, since
+
= 4(1 yo). Note that 4 = @ & ( I- 703) or 4 = 2 Re($)(l - 703) which
implies 4y2y1 = 2 Re(i$)(l - yo3), and so the Dirac equation has the form
In contrast to the mother spinor a, the real ideal spinor 4 satisfies 4~~~~~=
4721, and so we could rewrite the Dirac equation in the same way as Hestenes
1966:
C o m m e n t s . The ideal spinors might be useful in conjunction with conformal

transformations of the Dirac equation. Decompose the ideal spinor 4 = $0
into its even and odd parts and separate the parts,
which can be put into the matrix form
where we have used the fact that the matrix
0
O
-70123
) commutes with
0 d
(dl
0 ) and
40 0
0)
and takes the role of an overall commuting imaginary unit Q. I
Exercises 9,10
Historical survey
Pauli 1927 and Dirac 1928 presented their spinor equations for the description
of the electron spin. Juvet 1930 and Sauter 1930 replaced column spinors by
square matrix spinors, where only the first column was non-zero. Marcel Riesz
1947 was the first one to consider spinors as elements in a minimal left ideal of a
Clifford algebra (although the special case of pure spinors had been considered
earlier by Cartan in 1938).
Giirsey 1956-58 rewrote the Dirac equation with 2 x 2 quaternion matrices
in Mat(2, W ) [Lanczos 1929 had used pairs of quaternions, see Gsponer & Hurni
19931. Kustaanheimo 1964 presented the spinor regularization of the Ke-
pler motion, the KS-transformation, which emphasized the operator aspect
of spinors. This led David Hestenes 1966-74 to a reformulation of the Dirac
theory, where the role of spinors [in columns (l?
or in minimal left ideals of the
complex Clifford algebra C @ Cel,3 2 Mat(4, C)] was taken over by operators
in the even subalgebra Cet3 of the real Clifford algebra Cel,s 2: Mat(2, JR).
Exercises
1. Show that for u E C 8 Cel,3 the real part Re(u) corresponds to
1 (u+yol3u*y02~)E Mat(4, C).
+
2. Show that if 2~ E C 8 Ce1,3 satisfies the condition u = u t ( 1 iylz) then
+
u = Re(u)(l iylz) and iu = uyzyl.
3. Show that Im($) = Re($)y12 in C 8 Ce1,3.
4. Show that the charge conjugate $c = -iy2$* of the Dirac spinor
$ E Mat(4, C) f corresponds to 11,~= $*yl E (C @ CelB3)f .
5. Show that although for 11, E Mat(4, C)f , Re($) E Mat(4, C) f , for a
non-zero 11, E (C 8 Cel,3)f , Re($) !$ (C 8 Ce1,3)f.
6. Show that in terms of the ideal spinor 4 = 6 t ( 1 - yo3) E Cel,3 k(1 - yO3),
where 6 = 4 Re($), the bilinear covariants can be expressed as
Ri = (647o)o = ($4) .Y3,
Jp = ( $ 7 4yO)O
~ = ( $ 7 4)
~ .73,
s p v = -(8ypv4~123)0 = -(67pv4)3 ' 7123,
= -($~0123~p4~123)0,
0 2 = -($yO1234yO)O
and the aggregates P = R +J and Q = S + KyOlz3 as
[Hestenes 1986 p. 334 gives P in (2.26) and -Q in (2.27) .]

7. Show that 44, 44, 4yl$, 4y14, 4yz$, 4y26 all vanish.
8. Show that for a different choice of sign in 4 = i ( 1 - y03), namely
cp = 6 $ ( I + yO3), we have cpyo@= P = -cpy3@ and
~73@ = -&^/0123 = -P"IO@.
9. Show that 4 = 6 t ( 1 - 73).
10. Show that 4 = 4 ;(I+ yo)(l - 70s).
Bibliography
Solving with MAPLE V. Birkhauser, Boston, MA, 1994.
J.D. Bjorken, S.D. Drell: Relativistic Quantum Mechanics. McGraw-Hill, New York,
1964.
R. Boudet: Les alg6bres de Clifford et les transformations des multivecteurs, pp.
343-352 in A. Micali et al. (eds.): Proceedings of the Second Workshop on 'Clif-
ford Algebras and their Applications in Mathematical Physics' (Montpellier, 1989).
Kluwer, Dordrecht, The Netherlands, 1992.
R. Brauer, H. Weyl: Spinors in n dimensions. Amer. J. Math. 57 (1935), 425-449.
Reprinted in Selecta Hermann Weyl, Birkhauser, Basel, 1956, pp. 431-454.
A. Charlier, A. BCrard, M.-F. Charlier, D. Fristot: Tensors and the Clifford Algebra,
Applications to the Physics of Bosons and Fermions. Marcel Dekker, New York,
1992.
P.A.M. Dirac: The quantum theory of the electron. Proc. Roy. Soc. A117 (1928),
610-624.
A. Gsponer, J.-P. Hurni: Lanczos' equation to replace Dirac's equation? pp. 509-512
in J.D. Brown (ed.): Proceedings of the Cornelius Lanczos Centenary Conference
(Raleigh, NC, 1993).
F. Giirsey: Correspondence between quaternions and four-spinors. Rev. Fac. Sci.
Univ. Istanbul A21 (1956), 33-54.
F. Giirsey: Relation of charge independence and baryon conservation to Pauli's trans-
formation. Nuovo Cimento 7 (1958), 411-415.
J.D. Hamilton: The Dirac equation and Hestenes' geometric algebra. J. Math. Phys.
25 (1984), 1823-1832.
D. Hestenes: Space- Time Algebra. Gordon and Breach, New York, 1966, 1987, 1992.
D. Hestenes: Real spinor fields. J . Math. Phys. 8 (1967), 798-808.
D. Hestenes: Local observables in the Dirac theory. J. Math. Phys. 14 (1973), 893-
905.
D. Hestenes: Proper dynamics of a rigid point particle. J. Math. Phys. 15 (1974),
1778-1786.
D. Hestenes: Clifford algebra and the interpretation of quantum mechanics, pp. 321-
346 in J.S.R. Chisholm, A.K. Common (eds.): Proceedings of the Workshop on 'Clif-
ford Algebras and their Applications in Mathematical Physics' (Canterbury 1985).
Reidel, Dordrecht, The Netherlands, 1986.
P.R. Holland: Minimal ideals and Clifford algebras in the phase space representation
of spin-h fields, pp. 273-283 in J.S.R. Chisholm, A.K. Common (eds.): Proceedings of
the Workshop on 'Clifford Algebras and their Applications in Mathematical Physics'
(Canterbury 1985). Reidel, Dordrecht, 1986.
P.R. Holland: Relativistic algebraic spinors and quantum motions in phase space.
Found. Phys. 16 (1986), 708-709.
G. Juvet: OpCrateurs de Dirac et kquations de Maxwell. Comment. Math. Helv. 2
(1930), 225-235.
J. Keller, S. Rodri'guez-Romo: A multivectorial Dirac equation. J. Math. Phys. 31
(1990), 2501-2510.
J. Keller, F. Viniegra: The multivector structure of the matter and interaction field
theories. pp. 437-445 in A. Micali et al. (eds.): Proceedings of the Sccond Workshop
on 'Clifford Algebras and their Applications in Mathematical Physics' (Montpellier
1989). Reidel, Dordrecht, The Netherlands, 1992.
P. Kustaanheimo, E. Stiefel: Perturbation theory of Kepler motion based on spinor
regularization. J. Reine Angew. Math. 218 (1965), 204-219.
R. Penrose, W. Rindler: Spinors and Space-Time, Vol. 1. Cambridge University Press,
Cambridge, 1984.
M. Riesz: Sur certaines notions fondamentales en thCorie quantique relativiste; pp.
123-148 in C . R . 10" Congris Math. Scandinaves (Copenhagen, 1946). Jul. Gjellerups
Forlag, Copenhagen, 1947. Reprinted in L. Ghding, L. Hiirmander (eds.): Marcel
Riesz, Collected Papers. Springer, Berlin, 1988, pp. 545-570.
F. Sauter: Losung der Diracschen Gleichungen ohne Spezialisierung der Diracschen
Operatoren. Z. Phys. 63 (1930), 803-814.
11
Fierz Identities and Boomerangs
Fierz identities are quadratic relations between the bilinear covariants (or phys-
ical observables) of a Dirac spinor. They are used to recover the original Dirac
spinor from its bilinear covariants, up to a phase. The Fierz identities are suffi-
cient to examine the non-null case, when either $tyo$ # 0 or $t Y0Y0123$ # 0.
However, they are insufficient for the null case when both $+-yo$ = 0 and
$tYOY0123$ = 0. In this chapter, we introduce a new object called the boome-
rang, which enables us to study also the null case.
11.1 Fierz identities

The bilinear covariants satisfy certain quadratic relations called Fierz identi-
ties [see Holland 1986 p. 276 (2.8)]
= R; + R;, K 2 = -J2
In coordinate form the Fierz identities are as follows [see Crawford 1985 p. 1439
(141
where (*S),, = - 3 ~ ~ u a (with

~ ~ €0123
~ ' = 1) or *S = sYOlz3
[in general,
*v = 670123 given by 2~ A xv = <u, v > Y o A Y ~ A Y ~ A Y ~ ] .
In the non-null case R # 0 the Fierz identities result in [Crawford 1985 p. 1439
(1.3) and 1986 p. 356 (2.14)]
11.2 Recovering a spinor from its bilinear covariants
and
In the index-notation some of these identities look like
Note also that in general S L K = -K J S, 70123(SA K) = ( Y 0 1 2 3 ~ ) L K =

-(*S) L K = K J (*S), ~oi23(SA S) = (~0123s) J S = -(*S) J S and that
(JJS)AS=&JJ(SAS).
Fierz identities via spinor operators. By direct computation we can see
that
which gives one of the Fierz identities. Computing in a similar manner we find
Since the result is a bivector, we find that J A K = -(a2+ 01~0123)S

and
J.K=O. I
Exercise 1
11.2 Recovering a spinor f r o m i t s bilinear covariants

Let the spinor II, have bilinear covariants R1, J , S, K, R2 [a scalar, a vector,
a bivector, a vector, a scalar]. Take an arbitrary spinor q such that ij*$ # 0
154 Fierz Identities and Boomerangs
in C 18 Cel,3 or equivalently qtyo$ # 0 in Mat(4, C). Then the spinor $ is
proportional to
$ 2 Zq where ( Z = !&+ J + iS + iKmlz3 + Rzyo123 1
that is, $ and Zq differ only by a complex factor. The original spinor $ can
be recovered by the algorithm [see Takahashi 1983, and Crawford 1985, who
also gave a proof in the non-null case a# 0]
[For the choice q = f we get simply
which are not the same N, e-" as those for an arbitrary q.] Once the spinor
$ has been recovered, we may also write
A spinor 11 is determined by its bilinear covariants ill, J , S , K, i-22 up to a

phase-factor e-ia, and
projects/extracts out of q the relevant part parallel to $.

Recovery via m o t h e r spinors +
E Cel,3 i ( 1 yo). Take two arbitrary
elements in the real Clifford algebra, a , b E Ct?l,s, in such a way that 4 =
+
(a ib)f , f = :(I + +
y ~ ) % ( l y12). Then $$* = O and
where we have written g = i ( 1 + yo). Next, we introduce a real spinor, called

the mother spinor [for all real spinors]
11.3 Fierz identities and the recovery of spinors
Compute 04 = 0 and
to find
1 1
-06
2
=4~e($q*), and similarly - 0 ~ 1 2 6 = 41m($q*).
2
Recall that Z = 444' is sufficient to reconstruct the original Dirac spinor $
+
and conclude that the real mother spinor 0 E C.f!l,s i ( 1 yo) carries all the
physically relevant information of the Dirac spinor $. In fact,
1
$ = 0 -(1 + iy12) and 0 = 4 Re($)
4
where the real part is taken in the decomposition C @ C.f!l,3[and not in the
decomposition C @ Mat(4, R)].
+
Write as before Z = P + iQ where P = R J and Q = S Ky0123. We +
will show how to recover the real mother spinor 0 from its bilinear covariants
[G = g = $(I +yo)]:
[the same N as for the choice f E C @ Ci?1,3]or for an arbitrary spinor q 6

Cel,3g, 60 # 01
[ql E (C @ CC1,3)f and q = 2 Re(ql) result in the same numerical value for N].
Note that the role of i = a is played by multiplication by 7271 on the right
hand side, that is, 0y2y1 = 4 Re($).
Exercises 2,3,4
11.3 Fierz identities and the recovery of spinors

It might be interesting to know if given multivectors f i l l J , S , K , [a scalar,
a vector, a bivector, a vector, a scalar] are bilinear covariants for some spinor.
The answer is postponed to the next section in the case R1 = 0 = R2. Writing
+
R = R1 S/2701231we are left with the remaining case R # 0, in which we can
say that the multivectors are bilinear covariants essentially if they satisfy the
Fierz identities.
156 Fierz Identities arid Boomerangs
If the multivectors R1, J , S , K , R2 satisfy the Fierz identities, then their
aggregate
Z = Rl + J + i s + iK70123+ 0270123
can be factored as [see Crawford 1985 p. 1439 (2.2)]
This factorization is valid only in the non-null case R # 0. Using this factor-
ization Crawford proved that if the multivectors R1, J, S , K , R2 satisfy the
Fierz identities [and J0 > 0 with 4(ij*Zq)o = qtyoZq > 0 for all non-zero
spinors q], then R1, J , S , K, R2 are bilinear covariants for some spinor $, for
instance,
1
d=4NZq, N = I / ~ = ~ J w ~
[and two such spinors $ obtained by distinct choices of q differ only in their
phases].
Hamilton 1984 p. 1827 (4.2) mentioned how 11, determines Z = 4$qhtyo, see
also Holland 1986 p. 276 (2.9), Keller & Rodriguez-Romo 1990 p. 2502 (2.3b)
and Hestenes 1986 p. 334 (2.28).
11.4 Boomerangs
Definition. If the multivectors R1, J , S , K, R2 [a scalar, a vector, a bivector,
a vector, a scalar] satisfy the Fierz identities, then their aggregate Z = 01 +
+ + +
J i s iK-y0123 S/270123 is called a Fierz aggregate. I
Definition. A multivector Z = R1+J+iS+iKyo123+R2m23, which is Dirac
self-adjoint 2' = Z, is called a boomerang, if its components R1, J , S , K , R2
are bilinear covariants for some spinor $ E C'. I
Both in the non-null case R # 0 and in the null case R = 0 a spinor II,
is determined up to a phase-factor by its aggregate of bilinear covariants Z =
R1+J+iS+iK~0123+R2~0123 [as 11, = & e - i a ~ q ] , which in turn is determined
by the original spinor $ via the formula Z = 4$4* = 4$$t-yo [thus we have a
boomerang, which comes back].
If Z is a boomerang so that Z = 4$$t yo then Z2 = 4R1Z where R1 = (Z)O,
because
+ +
1 In Crawford's factorization Z = (Q J)(1 iQ-'Ky0123) the first factor P = Q J +
is Dirac self-adjoint, p-* = P. Writing r +
1 i ~ i l - ' - y 0 ~ 2 3 we
, can write Crawford's
factorization as Z = Pr*, and note that PT* = TP # Pr. Crawford 1985 posed an open
problem of decomposing Z into a ~ r o d u c tof two commuting Dirac self-adjoint factors.
This problem is solved at the end of this chapter.
11.4 Boomerangs
z2= 4$$ty~ 4$$t yo = 16$($ty~$)$tr~

= 16 tra~e($~yo$)$$~yo [since $tyo$ = trace($tyo$) f ]
= 16 tra~e($$~yo)$$tyo= trace(4$$tyo) 4$$tyo.
Conversely if R1 # 0 then Z2 = 4R1Z ensures a boomeranging Z. If Z is a
Fierz aggregate and R # 0 , then it boomerangs back to Z. Crawford's results
say that in the non-null case R # 0 we have a boomeranging Z if and only if Z
is a Fierz aggregate. However, in the null case R = 0, there are such Z which
are Fierz aggregates but still do not boomerang [for instance Z = J , J 2 = 0,
J # 01.
If R = 0 and J , S, K satisfy the Fierz identities, then for a spinor con-
structed by
1
$ = -Zrj
4N
+ +
where Z = J i s iKyolz3
we have in general Z # 4$4* (the Fierz identities are reduced to J~ = K 2 = 0,

J . K = J A K = 0 which impose no restriction on S). Even if the Fierz
identities were supplemented by all the conditions presented in section 12.2
(in the non-null case these conditions are consequences of the Fierz identities)
these extended identities would not result in a boomeranging Z. To handle also
the null case R = 0 we could replace the Fierz identities by the more restrictive
conditions
[see Crawford 1986 p. 357 (2.16)], but this would result in a tedious checking
+ +
process. If Z = J iS iKyo123 is a boomerang, then Z2 = 0, and so each
dimension degree vanishes,
(Z~)~=J~-SJS-K~
( z 2 ) i = +2~0123(SA K ) K in the plane of S
(z2)2 = +i2~0123(JA K ) J and K are parallel
(z2)3= +i2J A S J in the plane of S
(z2)4 = -S S S is simple.
The bivector part implies that J and K are parallel, the 4-vector part implies
that S is simple, and the vector and 3-vector parts imply that J and K are
in the plane of S. Altogether we must have
where h is a real number and s is a space-like vector orthogonal to J , J . s = 0.
+ +
We again compute Z2 = J2(1 (s hYo123)2) = 0 and conclude that either
1. J 2 = 0 orelse
2. + = -1.
(S hyo123)2
Neither condition alone is sufficient to force Z to become a boomerang [Z is
not even a Fierz aggregate if J' # 01. However, such a Z is a boomerang if
both conditions are satisfied simultaneously.
Counter-examples. 1. In the case 521 = 0, the element Z = J - S/270123,
= 52; > 0, is such that Z2 = 0, but Z is not a Fierz aggregate.
+
2. Z = J iS with J' > 0, S = -yOlz3Js,J s = 0, s2 = -1, is not aFierz
+
aggregate, and Z2 # 0, but we have Z-yO123Z= 0.

+ +
3. Z = J iS iK-yOlz3where J2= K2 = 0, J - K = 0, J A K = 0, S A S # 0,
is a Fierz aggregate but does not satisfy Z2 = 0, Z - ~ O=~0.~ ~ Z
+ +
4. Z = J ( l is ih~0123)with J 2 = 0, J . s = 0, (S hY0123)2 + # -1, is
a Fierz aggregate and satisfies Z2 = 0 and Z Y ~ =~0, ~but~still Z we do not
have a boomeranging Z. I
Throughout this chapter we assume that 521, J, S, K, Slz are real multivectors
+ + +
or equivalently that Z = 521 J iS iK-yOlz3 fi23/0123 + is Dirac self-adjoint
[2*= Z or in matrix notation -yoZt-yo= Z]. This implies that r]tyoZr] [=
4 ( ~ j * Z q )is~ a] real number for all spinors 77.
>
For a boomerang Z we have r,~t-yoZr] 0, for all spinors 77, and also J0 > 0
[the grade involute 2 of Z is such that (z)o.-~o < 0 and 4(ij*~r])0= r ] t - y o ~ r l5
01.
Theorem. Let Z be such that r l t - y o ~2v0 for all spinors r], and that J0> 0.
Then the following statements hold.
1. Z is a boomerang if and only if Zy02*= 4J0Z or equivalently
ZZtyO= 45'2.
2. In the non-null case 52 # 0, Z is a boomerang if and only if it is a Fierz
aggregate.
3. In the null case 52 = 0, Z is a boomerang if and only if
+ +
Z = J ( l is ihyolz3) where J is a null-vector, J~= 0, s is a space-like
vector, s2 < 0 or s = 0, orthogonal to J , J . s = 0, and h is a real number
such that h = Ihl 5 1. I
The condition Z-y02*= 4J0Z could also be written with an arbitrary time-
like vector v as follows: Zvg* = 4(v . J)Z.
11.5 Decomposition and factorization of boomerangs
11.5 Decomposition and factorization of boomerangs
Write
+
P = Qi + J fi23/0123, +
Q = S K70123 = S - 70123K
C = 1 - ~ J K = 1- - iKJ-1Y0123
~ ~ ~ [when
~ ~Sl #~01
so that Z = P + iQ = PC. Then PC = C P and we have found a solution
to the open problem posed by Crawford 1985 p. 1441 ref. (10). [Crawford's
second factor in (4.1)'
1 + i(% + %.y0123)-~K~0123
= 1 - i(Q2 - Sll-y0123)-~K
= 1 - is-lJ = 1 + iJ-lS,
did not commute with P unless S12 = 0.1 In the case Sll #0 there is another
factorization
where the factors commute and are Dirac self-adjoint, but in this factorization
the second factor is not an idempotent (even though it behaves like one when
multiplied by P).
11.6 Multiplication by the imaginary unit i = f l

We have found that i$ = $ 7 2 ~(# ~ $70123) corresponds to @y2yl = 4 Re(@)
and further to $7271 = @-y2-y1 $(I - 703). In the non-null case Sl # 0 write
[Using per = 47,@, p2 = Slf + Sl; Hestenes 1986 p. 333 (2.23) gives s =
yol23e3e0, Boudet 1985 p. 719 (2.6) gives -s = ele2.1 Note that
and s2 = -1, k2 = -1 and sk = ks. Both s and k play the role of the
imaginary unit (multiplication on the left side):
i$ = s$ = k$ = $7271 # $70123 Dirac spinor

SO = k@ = @727i mother spinor
s$ = 14 = $7271 = $70123 # $(-~012) ideal spinor
s4 = kqyo = 4y2y1 spinor operator
and P = sQ = kQ. [Hestenes 1986 p. 334 (2.24) reports s 9 = 9yzyl and also
s4 = 47271 = 4 ~ 0 1 2 3 . 1
Question
Do the conditions Z 2 = 0 and Z70123Z= 0 imply that Z is a Fierz aggregate?
Exercises
1. Compute K 2 , when K = *y3*
Show that
2. Im(11,) = Re(11,)ylz in C 8 Ctl,s.
3. a @ y l 6= 4 ~e(11,yl4 ) = -4 1m($y24) [no complex conjugation]
$@y26 = 4~e(11,y%6) = 4Im(11,yl$).
4. Q i = -2R1P, QiQj = 2fllQk (ijk cycl.) for Qk = $(@Yk6)Y0123
[P, Q1, Q2,Q3 = Q span a quaternion algebra when R1 # 01.
Show that for Z = 411,4*, where 11, E (C @ Ct1,3)f , the following hold:
5. Z y O Z= 4 J 0 Z (# 0 for 11, # 0).
6. PyOP= 2 J 0 P = -QyOQ.
7. PyoQ = QyoP.
8. even(2)Z = 0, o d d ( 2 ) =~ 0.
9. 22 = 0 + P2= -Q2, P Q = Q P [no complex conjugation].
lo. 22 = 0, Z 2 = 4 Q l z j P2= 2RlP, P Q = 2fi1Q.
11. ZZ = o + PP = QQ, PQ = -QP.
12. ZZ = 0, Z*Z = 4Qzy0123z j PP = 2S/2~0123P,P Q = 2S/2y0123&
Write
and show that:

+
13. c = s(nl n2yo123)-1= 1 - is.
+ +
14. r = 1 - ik, F* = 1 i k =-1 - i(R1 ~ z y o i ~ 3 ) - ~ y o i z 3 K .
15. z = I(C = CI( = rs = sr*.
16. Z = ns = sfI= n K = KfI [no complex conjugation needed]
= r p = PF* [this is Crawford's factorization]
-
[= Pr only if 522 = 0 since then I' = I"].
11.6 Multiplication by the imaginary unit i = f l 161
Bibliography
R. Boudet: Conservation laws in the Dirac theory. J. Math. Phys. 26 (1985), 718-724.
R. Boudet: Les alghbres de Clifford et les transformations des multivecteurs; pp.
343-352 in A. Micali et al. (eds.): Proceedings of the Second Workshop on 'Clif-
ford Algebras and their Applications in Mcithematical Physics' (Montpellier, 1989).
J.P. Crawford: On the algebra of Dirac bispinor densities: Factorization and inversion
theorems. J. Math. Phys. 26 (1985), 1439-1441.
J.P. Crawford: Dirac equation for bispinor densities; pp. 353-361 in J.S.R. Chisholm,
A.K. Common (eds.): Proceedings of the Workshop on 'Clifford Algebras and their
Applications in Mathematical Physics' (Canterbury 1985). Reidel, Dordrecht, 1986.
J.D. Hamilton: The Dirac equation and Hestenes' geometric algebra. J . Math. Phys.
25 (1984), 1823-1832.
D. Hestenes: Clifford algebra and the interpretation of quantum mechanics; pp. 321-
346 in J.S.R. Chisholm, A.K. Common (eds.): Proceedings of the Workshop on 'Clif-
ford Algebras and their Applications in Mathematical Physics' (Canterbury 1985).
Reidel, Dordrecht, The Netherlands, 1986.
of spin- $ fields; pp. 273-283 in J.S.R. Chisholm, A.K. Common (eds.): Proceedings of
Found. Phys. 16 (1986), 708-709.
J. Keller, S. Rodriguez-Romo: A multivectorial Dirac equation. J. Math. Phys. 31
(1990), 2501-2510.
theories; pp. 437-445 in A. Micali et al. (eds.): Proceedings of the Second Workshop
Y. Takahashi: A passage between spinors and tensors. J . Math. Phys. 24 (1983),
1783-1790.
12
Flags, Poles and Dipoles
The classification of spinors is commonly based on representation theory, irre-

ducible representations of the Lorentz group S0+(1,3). Thus, one customarily
speaks about Dirac, Majorana and Weyl spinors. In this chapter spinors are
classified in a new way by their bilinear covariants, multivectors of observables.
The new classification is geometric, since it is based on multivectors, and phys-
ical, since it is based on observables. The classification reveals new spinors,
called Jag-dipole spinors, which reside between the Weyl, Majorana and Dirac
spinors.
Dirac spinors describe the electron, and for them fl # 0. Weyl and Majorana
spinors describe the neutrino. Weyl spinors are eigenspinors of the helicity
projection operators $(I k Y ~ ~ and ~ their
~ ) bilinear
, covariants satisfy R = 0,
S = 0, K # 0. Majorana spinors are eigenspinors of the charge conjugation
operator, with eigenvalues f 1, and their bilinear covariants satisfy fl =
0, S # 0, K = 0. [Weyl and Majorana spinors are usually introduced by
properties of matrices, see Benn &Tucker 1987 and Crumeyrolle 1990.]
The flag-dipole spinors satisfy R = 0 [and cannot be Dirac spinors] and
S # 0, K # 0 [and so they are neither Weyl nor Majorana spinors]. Unlike
Weyl and Majorana spinors, the flag-dipole spinors do not form a real linear
subspace, because they are characterized by a quadratic constraint. There-
fore the superposition principle is violated, and the flag-dipole spinors cannot
describe fermions. It has been conjectured (G. Trayling, Windsor) that the
flag-dipole spinors are related to the quark confinement.
1 The charge conjugation operator C is conventionally defined by C(+) = -iy2+* for +E
6.
12.1 Classification of spinors by their bilinear covariants 163
12.1 Classification of spinors b y t h e i r bilinear covariants
In the following we shall present a classification of spinors ll, based on properties
of their bilinear covariants R1, J , S , K , R2, collected as
In other words, we classify the boomerangs Z = 4ll,$*.

+ + +
Recall that Z = P iQ, P = R1 J S/z-y0123,Q = S + K~oi23.
D i r a c spinors of t h e electron:
1. R1 # 0, R2 # 0: Using P2 = 2 R l P = -Q2 we find the relationship
P = &(-+Q~)/ Jw where the sign is given by J0 > O (and coincides
with the sign of R1). P = kQ, where k = -(a2+ f11Yo123)-1K, ill, = kll,.
2. R1 # 0, R2 = 0 : P is a multiple of &(a1 +
J ) which looks like a proper
energy projection operator and which commutes with the spin projection opera-
tor $(I- iKY0123/Rl). Z = R1+ J+iS+iK-y0123 = (R1+ J ) ( l - i-y0123K/Rl),
S = Y ~ ~ ~ ~P J= K / ~ ~ ~ k .= -yOlz3K/R1. In this class the Yvon-
-yOlz3&KQ,
Takabayasi angle /3 gets only two values, 0 and T ; and the charge superselec-
tion rule applies.
3. R1 = 0, R2 # 0: Using P2 = 2RlP we find that P is nilpotent: P2 = 0.
+ +
Z = J iS iK-yOlz3+fl2-yOlz3,S = -JK/R2. P = - 1522K Q = fK Q /
(opposite to the sign of R2), k = -K/R2.
Singular spinors w i t h a light-like pole/current:
+
4. R1 = R2 = 0, K # 0, S # 0: Flag-dipole spinors. Z = J i J s - ih-yo123J,
J' = 0, s is a space-like vector, s2 < 0, orthogonal to J , J . s = 0, S = Js
+
(= J A s ) , K = h J , h2 = 1 s2 < 1 (h real, h # 0). P = J , Q =
J(s + +
(1 + i s ih-yOlz3)Z= 0. Note that $ ( I - is - ih-yo123)ll,= ll, and
(1 +is+ih-yOlz3)ll,= 0. 2*= Z and z2= 0 imply Z = J(l +is+ih-y0123) etc.
Let ll, = &Zq, then Z = 4$$* implies (s+h-y0~23)~ = -1. P = (s+h-y0123)Q,
+
ill,= (s h~0123)$.
5. R1 = R2 = K = 0, S # 0: Flag-pole spinors for which Z = J i J s is a +
pole J plus a flag S = J s (= J As), J . s = 0, s 2 = -1. P = sQ, ill, = sll,.
The flag-pole spinors are eigenspinors of the charge conjugation operator with
eigenvalues X 6 U(1), thus C(ll,) = All,, IXI = 1.
- Write K k = @-ykg, Sk = @-yij6
(ijk cycl.) with K3 = K , S3 = S . Then
K1 = J , K 2 = K3 = 0 and S1 = 0, S2 = J s 2 (= J A s2), S3 = Js3 where
s 3 = s , s; = - I , s 2 - s 3 = 0 .
- Given an arbitrary Dirac spinor ll, with covariants J , K (and with K1, Sg)
we may construct, as special cases of flag-pole spinors, two Majorana spinors
164 Flags, Poles and Dipoles
$* = $($ f $C), which are seen to be eigenspinors of the charge conjugation
C($*) = f $*, and whose bilinear covariants J*, S* satisfy K . J* = 0,
K . S * = -R2Jh and J = J + + J - , S = S + + S - . [Note that J* = $ ( ~
K1)
f
and S* = ;(s 7 S2y0123).] The charge conjugations of i$* are C(i$+) =
-i$+ # &$- and C(i$-) = i$- # f$+.
6. R1 = R2 = S = 0, K # 0: Weyl spinors (of massless neutrinos) are eigen-
spinors of the chirality operator y0123$* = f i$*. Z = J T iy0123J, J =
fK , h = f1, $* = i ( 1 T iy0123)$*. Note that even($*) = $* b(1 F yoa),
odd($*) = $*;(I f 7 o ~ ) .P = fyoizaQ.
- Write K k = QykQ, Sk = QYij\k as before. Then K1 = K 2 = 0, S1 = Jsl
(= J A s l ) , S2 = Js2 (= J As2) where sf = si = -1, sl . s 2 = 0.
- Given an arbitrary Dirac spinor 11, with covariants J, K we may construct
two Weyl spinors $* = k(1 iyOl2a)$ with covariants J* = $(J f K),
K* = 3 ( ~ f J). Weyl spinors are pure: &yo123yIc$* = 0 [no complex con-
jugation; for arbitrary Dirac spinors $$ = 0, $y,$ = 0, &0123$ = 0 though
4yrv$ # 0, & 0 1 2 3 ~ p $ # 0 (and also $11, = 0, $ Y O I ~ ~ Y , $ = 0, $ ~ 0 1 2 3 $ = 0,
though $y,$ # 0, $y,,$ # O)]. C($*) is of helicity h = ~1 with covariants
J*, -K*.
In addition to the above six classes there are no other classes based on distinc-
tions between bilinear covariants. This can be seen by the following reasoning.
First, we always have J # 0, because J0 > 0. Secondly, S1 # 0 implies
+ +
S # 0 and K # 0. Thirdly, R = 0 implies Z = J(l i(s hyOl23)) where
(s+ hYo123)2= - 1 SO that we have a non-vanishing J(s+hy0123) = S + K ~ 0 1 2 ~ .
Comments :
For classes 1 , 2 the element &Z is a primitive idempotent in C @ Ctl,3.
Classes 1 , 2 , 3 are Dirac spinors for the electron. A spinor operator Q has
a unique (up to a sign) polar decomposition Q = a u , u E Spin+ ( l , 3 ) . In
particular, writing K k = Qyk\k we have an orthogonal basis {J,K1, K 2 , K3)
(K3 = K ) of IW1l3.
Class 4 consists of flag-dipole spinors with a flag S on a dipole of two poles
J and K . Class 5 consists of flag-pole spinors with a flag S on a pole J . Class
6 consists of dipole spinors with two poles J and K .
In classes 4,5,6 the vectors J , K1, K 2 , K3 no longer form a basis but col-
lapse into a null-line J (also S1, S2, S3 intersect along J ) . The even elements
J(yo - syl2 - hy3) and Q differ only up to a complex factor x - yyl2 (on the
right).
In addition to the electron (classes 1,2,3) the massless neutrino (class 6)
has also been discussed by Hestenes [I967 p. 808 (8.13) and 1986 p. 3431 who
12.1 Classification of spinors by their bilinear covariants 165
quite correctly observed that JTt= $@(yof y3)\t; note also that \tky,Qk =
(7, .J*)(yo 773). Hestenes has not discussed classes 4 and 5. Holland in Found.
Phys. 1986, pp. 708-709, does not discuss classes 3,4,5,6 with a nilpotent Z,
Z 2 = 0, but focuses on a nilpotent $J, $J2 = 0.
Majorana spinors Q E cet3+(1T yol) are not stable under the U (1)-gauge
transformation Q + QeaY126 cet3i ( 1 T 701).
Given a Weyl spinor $ with bilinear covariants J , K we can associate to
it two Majorana spinors $J*= $($ f $Jc) with Penrose flags 2%= +(Jr
is2~0123).
The number of parameters in the sets of bilinear covariants (or spinors with-
out U(1)-gauge) is seen to be
class no. 1 2 3 4 5 6
parameters 7 6 6 5 4 3
If the U(1)-gauge is taken into consideration, then the number of parameters
will be raised by one unit in all classes except in class 5 of Majorana spinors
[Weyl spinors with U(1)-gauge and Majorana spinors both have four param-
eters and can be mapped bijectively onto each other - which enables Penrose
flags also to be attached to Weyl spinors].
The Weyl and Majorana spinors can be written with spinor operators in the
form
where u = f73 for Weyl spinors and u = hyl for Majorana spinors. The
flag-pole spinors can be written in a similar form with u = yl cos 4 + 72 sin 4.
+
It is easy to see that all elements of the form @ $ ( l you), Q E ceTB3are
flag-dipole spinors, when u is a spatial unit vector, us yo = 0, u 2 = - 1, which
is not on the y3-axis or in the yly2-plane. About the converse the following
has been presented:
Conjecture (C. Doran, 1995): All the flag-dipole spinors can be written in
the form * $ ( I + you), where Q E cet3, u E R3, u 2 = -1. I
When u varies in the unit sphere S2 in IR3 (orthogonal to yo), the flag-dipole
spinor sweeps around the 'paraboloid' Q\k = 0. If the conjecture is true, it
would be nice to know the relation between s, h and u . [Clearly, h = u . 73.1
2 Our flag-pole Z = J + iS is invariant under rotations 1l + 4eaY12, whereas the Penrose
flags Z* = + ( J iS2y0123)make a 720' turn under a rotation of 360'.
12.2 Projection operators in E n d ( C t ~ , ~ )
+ +
Write as before P = R1 J ~270123,Q = S KyOlz3, + = P + i& = P C =
CP, C = 1 - iYolZ3JK-l. Then
1 1
- 2 = % P + = $, when Rl # 0
4%
1
VC* = d , when R # 0.
Define for u E Cll,3 (or u E @ 18Ct1,3)
1
Z+(U) = V(Uf ~ o I z s J K - ~ u ~ ~ z Zit
) , E End(Ctl,3).
Then
In general, for u E Cll,3, P: (u) = P* (u), C: (u) = C* (u) and P* (C* (u)) =
Ck (P*(u)), that is, P* and CTt are commuting projection operators. For an
+
arbitrary 9 in Cll,s i ( 1 yo) [or in Cll,3 i ( 1 - y03)] the spinor P+(C+(q))
is parallel to @ [or to 41, that is, the bilinear covariants of P+(C+(q)) are
proportional to P, Q. However, for an arbitrary u E Ctl,3, P+(C+(u)) $!
+ 4
ct1,3i ( 1 m) [or P+ (Z+ (u)) $! ct1,3 (1 - YO^)].
Define
1
~ : ( u ) = V(u y0123JK-1~y0123)
where I stands for ideal spinor. Then for an arbitrary u E ellt3we have
c+(c:(u)) E c11,3i ( 1 - yO3), and P+(c+ (C:(u))) is an ideal spinor parallel
to 4 (with bilinear covariants proportional to P, Q). Furthermore, c:($) = 4,
C i (4) = 0, and C: are projection operators commuting with P* , C*.
Define (0 stands for spinor operator)
which are projection operators commuting with C*.

12.3 Projection operators for Majorana and Weyl spinors
Exercise. (Inspired by Crawford 1985) Define
and show that rrt are projection operators commuting with Cf [but not with
Pk unless Q2 = 0; recall here the factorization of Crawford]. Show that
P+(r+(@))= @, P+(F+(4)) = 4. How would you define rh for a spinor
operator @?
+
[Answer: I'$(u) = $(u r (Q2 Q ~ Y ~ I ~ ~ ) - ~ Kfor
U uY EOCef,3.]
I~) I
R e m a r k . Define I'k for an ideal spinor 4 (I stands for ideal):
In the special case Q2 = 0 of type 2 these take the form
and commute with P* [this special case was also observed by Hestenes 1986
p. 336 (2.32)]. I
12.3 P r o j e c t i o n o p e r a t o r s f o r M a j o r a n a a n d Weyl spinors

Treat first the general case (class 4) Ql = 0 = Q2, K # 0 # S. Recall that
+ +
(1 is i h ~ ~ ~ = ~ 0~ or l ,= (s + h ~ ~ ~ ~Define
) lill, ~)ll,.
Then c$(@) = @, C$(d) = 4. Majorana and Weyl spinors are now the limiting
cases
Exercises 1,2,3,4,5
12.4 Charge conjugate +c = C(+)
The charge conjugate spinor = -iy2$* sits in C4 or in the same minimal
left ideal Mat(4, C) f ; it satisfies
Charge conjugation is an anti-linear operation, that is, C(i$) = -iC(+) . Other

t
characteristics of charge conjugation are +cyo$c = -+tyo$ and +EYoY,+c =
In the notation of C @ CelP3 we have Gc = -ir2ro13+*r~3 = qLY1,

which
also sits in the minimal left ideal (C@Cel,3)f , while yl swaps the signs of both
+ +
factors of the primitive idempotent f = +(1 y o ) i ( l iyl y2). The bilinear
covariants are transformed as follows under the charge conjugation
(?J~/)c)o = -($*$)o (as above)

(~EY~*C)O = +(4*r,+)o (as above)
(?ZE~Y,"$C)O = +(4* iy,v*)o
and 4 ~ =- +
~2' 4= -P~ i~ = -S11 + + +
J iS iy0123K - ~0123Q2,since
4 + c 4 ~= 4&*yl(&* -yl)-*= 4 & * ~ 1 ~ =
1 $-4(+4*) * *
The charge conjugate of the mother spinor is
+
it satisfies @c E C.el,si ( 1 yo), and has the same properties as were listed
above for the charge conjugation. The charge conjugate of the ideal spinor is
4c = dYi = 6y1 i ( 1 - yo3). Its bilinear covariants are (as above)
The charge conjugate of the spinor operator is Qc = Qylyo = even(&yl),

where 6 y l = (ao- @oyo)yl.
Exercise. Show that the operator form of a Majorana spinor i(+f +c) is
3 In this case the complex conjugate is
12.4 Charge conjugate Gc = C ( $ ) 169
The Wigner time-reversal is $7 = -iyr3$* E Mat(4, @)f or +7 = Y123$*Y1 E

(C @ C t l , 3 )f and the parity involution of space is $ p = y o $ . So
$7 = 7123$*7I $p = -yo$ $C = 4*71 Dirac

@7 = 7123471 @P = YO@ @c = mother
47 = 7 1 2 3 4 ~ 1 4~ = 704 4c = $71 ideal
*T =~123*~1 QP = YOQY,' 9c =~ 7 1 7 0 operator
and GTpC = - f O l z 3 $ . Note that charge conjugation C anticommutes with both

parity involution P and time-reversal 7.
Exercises 6,7,8,9,10
Appendix: Crumeyrolle's spinoriality transformation

Crumeyrolle introduced a number of spinoriality groups to be able to treat the
complicated situations with spinors. However, one relevant problem remained
unsolved: how can the usual bilinear covariants be obtained from Crumeyrolle's
spinors? The bilinear covariants of Crumeyrolle's spinors mix the Dirac cur-
rent vector J and the electromagnetic moment bivector S. A solution to this
problem can be given by a variation of Crumeyrolle's spinoriality group. In
this appendix it is shown how to extract the standard bilinear covariants (see
the standard textbooks on quantum mechanics, like Bjorken & Drell 1964) from
Crumeyrolle's or Cartan's pure spinors.
Crumeyrolle considered the complexificatioil C @ C e l , 3 of the Clifford alge-
bra C t l , 3 of the Minkowski space R193. In the complex linear space C @ R1p3
Crumeyrolle picked up a maximal totally null subspace spanned by the ortho-
gonal null vectors
1 1
5 (70- %) and (7' - i W )
Denote the product of these vectors by
which is the volume element of the totally null subspace. Crumeyrolle chose
as his spinor space the minimal left ideal (C @ C t l , 3 ) v of the complex Clifford
algebra C @C t 1 , 3 . The difficulty with this choice is that the bilinear covariants
of such a spinor are not directly related to those of a column spinor in standard
textbooks on quantum mechanics. To overcome this difficulty, first note that
the element
is a primitive idempotent generating Crumeyrolle's spinor space, that is,
+
Unfortunately, g = i ( 1 - yo3)i(l iy12) is an even element and does not
contain as a factor the 'energy projection' operator i ( 1 yo). The physical +
observables are obtained from column spinors sitting in C4 which are related
to Clifford algebraic spinors sitting in (C 8 Cfl,3)f , where
In order to move from the spinor space (C 8 Cfl,3)g to the spinor space
(C 8 cll,3)f we must find a transformation law for spinors $g E (C 8 Cel,3)g
and $f E (C 8 Cll,3)f . This transformation law is a slight variation of
Crumeyrolle's spinoriality transformation.
Before giving our variation of the spinoriality transformation let us recall
that $g is a sum of two Weyl spinors
so that the components are of homogeneous parity [the correspondence between

the even/odd parts and the negative/positive helicities is irrelevant, since it
could be swapped by a different choice of g, for instance, by g = (1 yo3) (1 4 + 4 +
+
iylz) for which g = y o ~ vwith v = ;(-yo Y 3 ) 1Z ( ~ 1 - iY2)].
Our variation of the spinoriality transformation is carried out by the element
4 T h e bilinear covariants of Crumeyrolle's spinors either vanish identically o r else, as in

Crumeyrolle 1990 p. 229 formula 24, mix the Dirac current J and the electromagnetic
moment S .
5 Recall t h a t , for instance, J p = 4 ( 4 ; ~ , , $ J ~In) ~contrast,
. these bilinear covariants of
$Jg vanish: 4 ( 4 : y p + g ) o = 0. However, +g does carry all the information of J : J p =
4 ( 4 i ~ p d ' g ) .Yo.
i
6 For Crumeyrolle the spinoriality group meant a number of things, with different adjectives
added as specification. First, it is thesubgroupof those s E S p i n ( l , 3 ) for which sv = fV ,
see Crumeyrolle 1990 p. 145. Secondly, it is the group of invertible elements z in C@Ct!l,s
such t h a t the primitive idemPotents g and zgz-' determine the same minimal left ideal
( C @ CelB3)g= ( C @ Ct!1,3)g',see p. 277. Thirdly, it is, if normalized, the intersection of
the previous group with S p i n ( l , 3 ) , see p. 281.
12.4 Charge conjugate $c = C($) 171
for which g = z f z-l or f = z-lgz. The latter rule gives us a relation between
Crumeyrolle's nilpotent induced spinors $g and idempotent induced spinors $f
(directly related to the standard column spinors like those in Bjorken & Drell
1964),
(Earlier we wrote $ = lCtf but here it is necessary to indicate to which minimal

left ideal the spinor belongs.)
Now we can compute the spinor operator @ = even(4Re(Gf)) and the bi-
linear covariants, for instance, J = For later convenience note that
@ = Oper($f) where
Recall the aggregate of bilinear covariants
and note that 4$f 4;= 4$f % d j . Our variation of the spinoriality group is the
group of those elements z in Cel,3 or (I: 8 Cel,3 which preserve the aggregate
Z under the transformation +hf + lCtf z-l . Crumeyrolle's spinoriality groups
preserve the ideals whereas our spinoriality groups preserve the physical ob-
servable~. The spinoriality groups are seen to be the following (see Lounesto
1981 p. 733):
where, as an example, the Lie algebra of Sp(4)

elements
71, 7 2 , 73
- Spin(5) is spanned by the
712, 713, 7 2 3
7012, 7013, 7023
70123.
For those z in Cel,3 which preserve Z = 4+hfYo4j, under the replacement
Gf + z-l, that is z E Sp(4), we may find that the spinor operator is
preserved under the following transformations:
where w = zz', that is,
To put all this in a nutshell: our variation of

I spinoriality transformation preserves bilinear covariants I
However, this preservation should be distinguished from our use of the par-
ticular element t = &(I
-
+
73) E Sp(4) to retrieve the aggregate of bilinear
covariants Z = 4$ $j by sending $, to $, t = $f.
Exercises
+
1. Recall that 4y2yl = s9yo h ~ ~ HOW ~ would
~ ~YOU9 define
. for a CZ
spinor operator 9 ?
2. Recall that 4-fOlz3= 4 ~ ~ HOW
7 ~would
. YOU define another pair ~4 for an
ideal spinor 4?
3. Show that up to a unit complex factor eY12a:
Q N &(RI + +
Jm - S712 - K73 Mz-~oIz~), when N = # 0.
4. Show that the operator form of a Weyl spinor is Qk(1 yo3).
5. Show that Weyl spinors $ ( li-yOlz3)$
~ correspond to even and odd parts
of the ideal spinor 4 = 40 41. +
Write W = 4 $ a , note that $4; = 0, and show that
6. W = -(&I + I I Z ~ &k = $ ( @ ~ k ~ ) ~ o01
~ Q Z ) Y C where i z&k3 = Q ( 1 4 YO)Yij\jt
( i j k cycl.).
7. W = K - S Y O where + +
I Z ~ K = K1 iK2 and S = S1 i s 2 , where as before
Kk = 9-yk\jt, Sk = ( i j k cycl.).
8. W2 = 0.
9. W Z = 0.
10. Z W = 4R1W and so the 3-vector part vanishes:
+ +
(ZW)s = -J A (~0123s) i s A K ~0123Q2x i K A S = 0. +
11. Show that 4$,4; = 0.
12. Write 9 = Oper($f) and 9, = Oper ($,). Show that
Q ~ 0 %= 29gy0\jtg.
13. Write 9f = even(4 Re($/ )) = Oper($f) and
Q, =?ven(4 Re($,)) # Oper($,). Show that 9,\jtg =P, ~ ~ ~= 0 ~and ~ \
Qgyo9,= J + K where J=9f70\jtfand K = Q f y 3 Q f .
7 4;= (*c)-* # ( S * ) c .
12.4 Charge conjugate $c = C($)
Solutions
1. C* (u) = +(u f suy012 f hy0123uy12) for u E ~ e f , ~ .
GO
2. c4*(~) +
= ;(u r (S ~ Y ~ ~ ~ ~u E) u Y$(I~ -~y03). ~ ~ ) ,
$ r iy0123)$) in the decomposition
4. Hint: compute the even part of 4 ~ e ( (1
c 8 ce,,,.
Bibliography
I.M. Benn, R.W. Tucker: A n Introduction to Spinors and Geometry with Applications
in Physics. Adam Hilger, Bristol, 1987.
J.D. Bjorken, S.D. Drell: Relativistic Quantum Mechanics. McGraw-Hill, New York,
1964.
J.P. Crawford: On the algebra of Dirac bispinor densities: factorization and inversion
J.P. Crawford: Dirac equation for bispinor densities; pp. 353-361 in J.S.R. Chisholm,
A.K. Common (eds.): Proceedings of the Workshop on 'Clifford Algebras and their
Applications in Mathematical Physics' (Canterbury 1985). Reidel, Dordrecht, The
Netherlands, 1986.
A. Crumevrolle: Groupes de spinorialitC. Ann. Inst. H. Poincare' 14 (1971), 309-323.
A. ~ r u m e ~ r o l l eorthogonal
: -and Symplectic Clifford Algebras, Spinor structures.
C. Doran: Geometric Algebra and its Applications to Mathematical Physics. Thesis,
Univ. Cambridge, 1994.
25 (1984), 1823-1832.
of spin-: fields; pp. 273-283 in J.S.R. Chisholm, A.K. Common (eds.): Proceedings of
Found. Phys. 16 (1986), 708-709.
J. Keller, S. Rodriguez-Romo: A multivectorial Dirac equation. J. Math. Phys. 31
(1990), 2501-2510.
theories; pp. 437-445 in A. Micali et al. (eds.): Proceedings of the Second Workshop
13
Tilt to the Opposite Metric
Physicists usually go from Ct1,3 --N Mat(2,W) to its opposite algebra C13,1
Mat(4,R) by replacing y, by iy, [within Mat(4, C)]. However, such a tran-
sition to the opposite metric does not make sense within the space-time R4,
because it calls for i R 4 which is outside of R4. We will instead regard the
linear space R4 as one and the same space-time, endowed with two different
metrics or quadratic structures, R113 and TW311.
Let the basis {yo,yl, 72, y3) of the space-time generating Ctl,3 corre-
spond to the basis {eo,el, e 2 , e3) of the space-time R311 generating Cts,l,
+ + +
So the vectors A0 yo ~ l y i ~ ' ~ A3y3 2 and AOeo Alel + + A2e2+ A3e3
correspond to each other but have opposite squares
We shall go further and regard Aoyo+ A1yl +AZy2+ A3y3 E R113 c Ct1,3 and
+ + +
AOeo Ale1 ~ ' e z A3e3 E R311 C Ct3,1 as one and the same vector A E R4
embedded in two non-isomorphic algebras Ctl,3 and Ct3,1 which are identified
as linear spaces by the correspondences
[Note that yo = yo and A0 = A0 in Ct1,3 whereas in C&,r we have eo = -eO

and A. = -A0 and that the numerical values of A0 are the same in Ctl,3 and
Ct3,l .I
13.1 The Maxwell equations in R311 175
The products in the Clifford algebras Cll,3 and CtaBl are related to each
other by (all the terms in this table are computed in Ctl,3)
where a0 = even(a) and a1 = odd(a). For a E Celts we may sometimes

emphasize that opp[a] E Ce3,1 (or the other way round). In this notation, the
products in the (graded) opposite algebras are related by
In this chapter we shall study the Maxwell equations and the Dirac equation
in opposite metrics, in the quadratic spaces R1s3 and R3t1. In particular, we
do not consider curved space-times, only flat space-times. In a flat space-time
it is also possible to differentiate multivector fields, not only differential forms;
we will focus on differentiating multivector fields.
THE MAXWELL
EQUATIONS
IN OPPOSITEMETRICS
There are a few changes of sign in the Maxwell equations in the quadratic
spaces and
IR3t1 IR113.
13.1 The Maxwell equations in R3t1

We use the following definitions for the potential A, the current J , the differ-
ential operator d, and the electromagnetic field F :
This leads to the d'Alembert operator
and the equation

d*d,~P = - p ~ P .
1 The symbol 'opp' is not a function of one or two variables; it is rather an indicator signaling
that all the computations in the brackets will be computed in the opposite algebra.
176 Tilt to the Opposite Metric
The equations
- alAO = - L a A X - a E = L E
at axe c
d 1 A 2 - d 2 A 1 = &Ay - &A, = B z , . . .
give
F*P = -(daAP - dPAa),
and
4
~,F"o=LL+LE,+L%=Lv.E=L=~J 0
ax c ay c az a
d , ~ , l = - Lc&aE t $ + & ~ z - & ~ y = p J 1 , . . . .
give the Maxwell equations
d,FaP = ~ J P
In the Clifford algebra C.&,l we have
and the equations
which lead to
d A = -F, dF = pJ, d 2 =~-pJ
The computations can be related to Gibbs' vector algebra as follows (here
c = 1, 6./ = 1 ) :
o ~-
where in the last step we used - e aA - zaAe o .
13.2 The Maxwell equations in R113 177
13.2 The Maxwell equations in R1s3
We use the following definitions for the potential A, the current J, the differ-
ential operator 8,and the electromagnetic field F:
This leads to the d'Alembert operator
and the equations

dad,~P =~ 1 ~ 0 .
The equations
give
F ~ P= d a ~ P- PA^,
and
ax c c a~ c c
-
~ , F ~ O = L & + ~ & + ~ & = I . V . E = ~ = ~ J
ay a
0
&Fal = -la&
at c
+ &B, - a~
aa Y --C ~ J ' , ...
give the Maxwell equations
da~a=
P
In the Clifford algebra C.ll,3 we have
and the equations

which lead to
13.3 Comparison of R q l and R1t3

The equation d A = fF has opposite signs in opposite metrics. This means
that the raising part d A A = fF has opposite signs in opposite metrics,
whereas the lowering part d J A = 0 is independent of metric. The Maxwell
equations d F = pJ have the same signs in both metrics. This means that
the lowering part d J F = pJ is invariant under the metric swap (and that in
both metrics d A F = 0). The above unexpected results are consequences of
our definition: the differential operator d experiences a sign change under the
metric swap, that is,
IGEq
In the rest of this chapter we shall study the Dirac equation, spinors and
observables in IR311. Our special concern will be the behavior of spinors under
the transition to the opposite metric.
Since going to the opposite algebra interchanges left and right ideals, we
will study real ideal spinors E Ct1,3$(1 - 703) in conjunction with their
opposite-reverses
(both and the opposite of its reverse 4 are in left ideals). Clearly, opp[J] =
O P P [ ~- .]
For instance, the Dirac equation for the real ideal spinors, 84721 = eA4+m4,
E Ct1,33(l - -yO3), is transformed by the opposite-reversion to d4e2l - =
e ~ +$m$, $ E Cts,1 i(l
+ eo3), and further by grade involution to
However, this is not a nice formula, because we have to explain the occurrence
of the grade involution in the last term. There are even more interpretational
difficulties for the opposite-reverses qb = opp[$] [of Dirac spinor qb = +
2 Note that 4 E Cel,3 fr(1 - 703) is in a graded minimal left ideal of Cel,3 2. Mat(2, W),
+
which is also an ungraded minimal left ideal, while $ E Ces,l fr(1 eo3) is in a graded
minimal left ideal of Ce3,1 2. Mat(4,R), which is not an ungraded minimal left ideal [the
minimal left ideals of C!3,1 are not graded].
13.3 Comparison of I R 3 p 1 and IR1t3 179
+
yo)$(l iy12)] and 9 = opp[6] [of mother spinor O = @$(I yo)], since +
$ $! (C 8 Ces,l)$(l 2c ieo) and $ is not in any proper left ideal of Ct3,l. An
obvious attempt for a possible solution would be to study q- = $- 4 ( 1 - ieo),
but then (like evaluating 4$4* = Z in the case IR1t" for the aggregate of
observables
and we would have the inconvenience of an extra factor i.

This shortcoming
could be circumvented by multiplying q- by fi,but then the relation to the
original Dirac spinor $ would be irrational. Again there is an obvious solution:
multiply r]- by 1 - i, which has absolute value dj,and study instead the flip
of the opposite-reverse, that is, the tilted spinor
for which
The opposite-reverse of $yo = $ is qeo - Therefore, we find

- = $.
(1 + i ) ? = $(1 - ieo) = $ + i$ = (1 +i)$, + (1 - i)$,
[$, = even($), = odd($)] which implies - - - or since i$- = $el2,
$ = $o-i$l, -
- =$ -e l (9 = O P P [ ~ I ,$ E c 8
Similarly, we define the tilted spinor for the mother of all real spinors @ and
for the real ideal spinor 4 :
Of course, for the spinor operator 9 = (= opp[@]).

-
The transition back from $ E (C 8 Ce3,1)$ ( l - ieo) $ ( l - ie12) to $ E
+
(C 8 Ct1,3);(I yo);(I +
iY12) is given by tilting again,
since
-
I Z(40)- - o p ~ [ ( - ~ ~ e ~ 2 ) - ] y l 2
OPP[$OI- O P P [ $ ~ ] ~=
= $0 - (-~12$1)-~12= $0 - $1~12712.
13.4 The D i r a c equation i n R3t1
The opposite-reverse Dirac equation
1
3 e 2 1 = eA$ - m$, $ E C&,l 5(1 + e03),
splits into even and odd parts [40 - 41
- = even($), - = odd($)]
a&e21 = e q l-
a$,ezl= eAgo + m$,.
Recalling that -$0 = $o- and -$1 = $le21
- we find
d o = eAjoe12 - m j 1
3 1 = eAjie12 - m4o -
which added together result in
1
a j = e A i e l 2 - m -4, j E Ck.1 2(1 + e03).
Similarly, the flip of the opposite-reverse, or simply tilted, Dirac spinor $ -
obeys
-
a$=ieA$-m$, - - -$ ~ ( @ B C t ~ , 11 ) ~ ( 1 - i e21o ) - ( l - i e l z ) ,
a formula found essentially in Benn& Tucker 1987 p. 284 (and p. 256). So the
-
tilted mother spinor 9 = 4 Re( $) = g ( 1 - eo12) obeys
which decomposed into even and odd parts (and recalling that go= - gleo12,
-
Q1 = - o e o 1 2 )results in the Dirac-Hestenes equation in the opposite metric,
where 9 : R3p1 + c e l l .
13.5 Bilinear covariants i n R3>l

- C&J) (1 - ieo)+(1 - iel 2)
Recall that for $ E (@18
Compute the bilinear covariants in the opposite metric R311 :

3 Note that 2= ( 1 - e m ) and = + eo3).
13.6 Fierz identities in IR3t1
J, = 4($*ie, 2)o ( J O= -Jo)

S," = -4($*ie,v 2)o
K, = -4($*eo123ep $)o
a2 = -4($*e0123$)0.
(Observe that the coordinates J, of J have the same numerical values in C l l , ~
and C&,J. In contrast, the coordinates J, are opposite in C-t1,3 and Clstl.)
13.6 Fierz identities i n R3p1

The Fierz identities in the opposite metric are
J2= - ( q + n;), K2 = -J2
J.K=O, J A K = -(R2 + e0123Rl)S.
Note also that S2 = (Rz - e0123Rl)~.
Exercise. Derive the real theory from the mother of all real tilted spinors
-
4 Re($) = = $ ( I - eon) E Cl3,l i ( 1 - eon). I
13.7 Decomposition of boomerangs i n R3t1

-
Write for $ E (C @ Clstl)$ ( l - ieo)i ( 1 - ie12)
2=4$$* --
IC=R1+Keo12s+R~eo123, L=-J-S
-S = R1 - iS + R2e0123, -C = 1 - ieo123JK-1
[Z, K , P , II, I' are not the same as those in the case of R113 but instead as a
sample Z in C @ C13,1 is obtained by sending Z in C @ Cllt3 to even(2) -
i odd(i)]. Then
Z = I( + iL = K _C = _CK = III( = K ~ I * (# IcII unless Q2 = 0)
= P _ c =_ ~ ~ = n _ s = _ s n *
= I'P = PF (no complex conjugation)
11'2 = 201K = -L2, ICL = LI( = 2O1L
n = P(Ol + O~e0123)-l, -C = _S(O1+ O~e0123)-'.
13.8 Multiplication b y i = f l in C 8 C13,J
Write
= -eolZ3SK-l = ' e o g-l
j = (O1 - 02e~123)-1J
-s = eolz3JK-I = (a1+ 02eo123)-1S= 9 e 1 2 9 - l .
Then Ii' = jL = s L , ll= 1 - i j and _C = 1 - is. Also
- -
i $ =j $ = -$ = -$el2 = Pea # -$eolZ3 tilted Dirac
j = S =el2=e o tilted mother
- - -$el2 = e 0 1 2 3 # e o
j$=?$= tilted ideal
-s -9 = 9 e 1 2 (but j 9 e o = - 9 ) operator.
13.9 S o m e differences between Celts and C&,J

Compute as a sample in C.e3,J
1 - 1 -
-0 3 = 0, -2 @
- 2= I(, -
- @ e o = -L
2
1
- @e3 = 0,
1 -
- @e3 = -Keol23.
2 - 2 -
Note that
o ( Q ~ ~ Q >Y o~ ) for
( ~ y o p y o )= ~ non-zero $E c 8 cel,3
while
( ~ e o k e=~( )~~ e ~ i$ 0e ~ for

) non-zero
~ -11, E C 8 C13,,
( ~ i ' e ~ ~#e( ;~~ e) ~~ l e>;0~ ) ~
so that it is possible that Ii' = 0 while L # 0 (this happens in the case
Ol = Q 2= 0, K = 0 of Majorana spinors).
13.10 Charge conjugate in R311
13.10 C h a r g e conjugate in R3t1
The charge conjugate of the tilted Dirac spinor is obtained as follows:
$C - = $*el and tilt

= $*yl take opposite-reverse qhc
-$c = (1 - i)$*e1;(1 - ieo) = (1 - i)$*;(l+

- ieo)el
= [(I+ i)$i(1 - ieo)]*el = -i[(l - i)$;(l
- - ieo)]*el
= -i - $*ele12= - -
$*el = - - $*e2 or
-
= - $*eleo = $*em E (C 8 C.t3,1)i(l - ieo)f ( I - ie12).
NUMERICAL
EXAMPLE
Start from Ctl,3. Take a column spinor
Then Z = 4$$+ 70
=4 ( 25
22+ 21i
22-21i
37
26 - 7i 17 - 28i
11-2i 8-Ili
-26-7i
-29
-12-i
-11-2i
-17- 282' -8 - l l i
-12
-5
+ i
where
In the opposite algebra C t 3 , ~we must first fix the matrix representation, for
instance, using the Pauli spin matrices a k ,
corresponding to the tilted primitive idempotent
1 1
f = - ( 1 - ieo)-(1 - ie12)
- 2 2
and the tilted spinor basis
Then the tilted spinor is

4+3i
-- --
and the boomerang Z = 4 $ $t(-ieo) = 4 $ $tie0
25 22-21i 7-26i 2-lli
22 + 21i 37 28-17i 11-8i
-7 - 262' -28 - 172' -29 -12+i
- 2 - I l i -11 - 8 i -12-2' -5
where
Note that for u E C 8 C.&,I, G* = e o u t e i l .

Note that Z for R1p3 and Z for R3s1 are related via a similarity transfor-
mation by & ( I +h).
Summary
To realize the transition to the opposite metric we use the rules
opp[d] = -8 and opp[ab] = boa0 + boa1 + blao - b l a ~
13.10 Charge conjugate in R311 185
and apply reversion to get the tilted spinors also in left ideals. Thus, the two
sides of dA = F in IR113 are transformed as
opp[dA]- = - ( - d ) ~ = dA
-
opp[FIe = F = -F
and so we have dA = -F in IR3t1. The two sides of d F = J in IR113 are
transformed as
opp[dF]' = ( - d ) ~= d F
-
opp[J]- = J = J
and so we have d F = J in R311. The terms of d4yZ1- eAS = mSyo in IR1t3
are transformed as
0 ~ p [ d S ~ 2 1=
] - (-d)@& = a g e z l
opp[AB]- = A$ = A@
opp[Syo]- = $60 = Gee
and so we have d@ezl- e ~ 4= m$eo in R3s1. [Earlier in this chapter we wrote
-S for 4.1 Note in particular that the Dirac-Hestenes equation has the same
form in both metrics, only the spinor operators are reversed. For complex ideal
spinors the situation is more complicated, an extra flip is needed to complete
the metric tilt.
In our differential operator
we have used an orthonormal basis, but this formula gives the same d for any
basis {eo,e l , e2, e3} for IR3t1 when e, . e, = 6;, that is, when {eo,e l , e2, e3}
and {eO,e l , e2,e3} are reciprocal. In this sense our differential operator is not
only Lorentz covariant but also invariant under all of GL(4, R).
Note that the raising differential d A f is metric dependent and therefore it
is not related to the exterior differential d A f [in a metric inpendent way]. In
general, in dimension n, not necessarily n = 4, the lowering differential d J f
is metric independent and related to the exterior differential by
for an n-volume w E An V such that w*Jw = 1 for w* E An V*. The relation

(8J f ) J w = d A (f J w*) requires identification of multivector-valued functions
with differential forms, which is possible only in flat spaces, while multivectors
cannot be differentiated on curved spaces. Such an identification can be carried
out by lowering the coordinate-indices of a multivector by means of the metric
tensor g,, (or raising the indices of a differential form by the inverse gpV).
Tilt to the Opposite Metric
Exercises
Show that
l.geo%=~. 2.e125=s.
3. $('I - ieo)(l- ie12)9 = I( iL. + 4. e o = -L.
-
5. 40 = - $ l e o l z , 91 = -$0e012.
6. qo + qle12 E (C 8 C&,i)$(l+ ieo).
Write flip(u) = uo - ulelz and recall that the opposite-reverse of A11, is AG-
(for a vector A). Show that
- - -
) -A 11,e12 [+8 11, = eA ll,elz - m $ ] .
7. f l i p ( ~ G=
8. A11, and AO = 4Re(A11,) correspond to AByo = even(AO), and in the
opposite algebra to -A gee.
- -6;
9. 4 ll,c = -4( 11, $*)*. [This means that J , S are preserved under charge
conjugation whTle Q1, K, Q2 swap their signs - as in R19. This should be
contrasted with Crumeyrolle (1990 p. 135,l. -9)) who considers charge
conjugation in conjunction with a scalar product of spinors induced by the
reversion (composed with complex conjugation), in which case S , K are
preserved and Ql, J , Q2 swap signs. Crumeyrolle's numerical results are
not directly related to the Bjorken & Drell formulation of the Dirac theory,
as he induces spinor spaces by totally isotropic subspaces of C 8 1W3l1.TO
relate the results one must permute the primitive idempotents by an
algebra automorphism of the Clifford algebra in such a way that dimension
grades are mixed. The next exercise gives a hint on how the scalar product
of spinors induced by the reversion (composed with complex conjugation)
can be used to find the bilinear covariants.]
10. K, = 4(4*e,- - - e3 (find similar formulas for J,, S,,).
In the next exercise we have a scalar product of spinors induced by the reversion
alone without composing it with complex conjugation.
- -
11. Take a Majorana spinor 11, = ll,c with bilinear covariants J , S = J A s.
-
Then the Weyl spinor u = $ ( I + ieolz3)11, has charge conjugate
- -
uc = $ ( I - ieolZ3)11, so that 11, = u + uc. Show that
J(u + uc) = 0 (3.1.21)
S ( U +UC)= -(u+ UC) (3.1.25)
4
4(iuelsiic)l = J (3.1.28)
Re(4iue13ii) = -4 S (3.1.29)
13.10 Charge conjugate in R3s1 187
The numbering on the right refers to Benn & Tucker 1987 pp. 113-116. Try
to work out a translation to their notation, and discuss the physical
relevance of the connection between the Majorana and Weyl spinors. Hint:
-
uc = i i * e 1 3 while u = + ( I + i e o l 2 3 ) $. Benn&Tucker use the scalar
product of spinors
Bibliography
This material grew out of discussions with Bill Pezzaglia at a meeting in Banff, 1995.
W. M. Pezzaglia Jr.: Classification of multivector theories and the modification of
the postulates of physics; pp. 317-323 in F. Brackx, R. Delanghe, H. Serras (eds.):
Proceedings on the Third Conference on 'Clifford Algebras and their Applications in
Mathematical Physics' (Deinze 1993). Kluwer, Dordrecht, The Netherlands, 1993.
W. M. Pezzaglia Jr.: Multivector solutions to the hyperholomorphic massive Dirac
equation; pp. 345-360 in J. Ryan (ed.): 'Clifford Algebras in Analysis and Related
Topics' (Fayetteville, AR, 1993). CRC Press, Boca Raton, FL, 1996.
I.R. Porteous: Clifford Algebms and the Classical Groups. Cambridge University
Press, Cambridge, U.K., 1995.
14
Definitions of the Clifford Algebra
In this chapter we shall for the first time give a formal definition of the Clif-
ford algebra. There are several definitions, suitable for different purposes. In
mathematics, definitions serve as premises for deductions; in physics, however,
definitions are more or less secondary and serve as characterizations. We shall
review Clifford's original definition, its basis-free variation given as a deforma-
tion of the exterior algebra, definition by the universal property, which does not
guarantee existence, and the definition as an ideal of the tensor algebra. The
construction of Chevalley, where Clifford algebra is regarded as a subalgebra
of the endomorphism algebra of the exterior algebra, is postponed till the dis-
cussion on characteristic 2. The definitions by the multiplication table of the
basis elements, and by index sets, are postponed till the chapter on the Walsh
functions. The definition of Clifford algebras as group algebras of extra-special
groups will be omitted.
14.1 Clifford's original definition

Grassmann's exterior algebra l\Rn of the linear space Rn is an associative
algebra of dimension 2n. In terms of a basis {el, e 2 , . . . , en) for Rn the exterior
algebra l\ Rn has a basis
14.2 Basis-free version of Clifford's definition
The exterior algebra has a unity 1 and satisfies the multiplication rules
ei A e j = -ej A e, for i # j,
ei Aei = 0.
Clifford 1882 kept the first rule but altered the second rule, and arrived at
the multiplication rules
e ,. e . - -e.Jei for i # j ,
J -
eiei = 1.
This time {el, ez, . . . , e n ) is an orthonormal basis for the posit,ive definite
Euclidean space Rn. An associative algebra of dimension 2" so defined is the
Clifford algebra Cen.
Clifford had earlier, in 1878, considered the multiplication rules
of the Clifford algebra Ceo,, of the negative definite space ROln.
14.2 Basis-free version of Clifford's definition

Here we consider as an example the exterior algebra AIR4 of the 4-dimensional
real linear space R4. Provide the linear space R4 with a quadmtic form
and associate to Q the symmetric bilinear form
This makes IW4 isometric with the Minkowski space-time R1s3. Then define the
left contraction u J v E by
/\IW1t3
for x, y E R113 and u, v , w E AR1t3. The identity ( b ) says that x operates

like a derivation and the identity (c) makes /\RIP a left module over /\IWIJ.
Then introduce the Clifford product of x E and u E
IW1t3 by the formula
/\IW113
1 Recall that .t, is the grade involute of u E AV, defined for a k-vector u E V by
a = (-qku.
190 Definitions of the Clifford Algebra
and extend this product by linearity and associativity to all of /\ IW113. Provided
with the Clifford product (the linear space underlying) the exterior algebra
/\ R1>3becomes the Clifford algebra C11,3.
14.3 Definition b y generators and relations

The following definition is favored by physicists. It is suitable for non-degenerate
quadratic forms, especially the real quadratic spaces IWPsq.
Definition. An associative algebra over IF with unity 1 is the Clifford algebra
Ce(Q) of a non-degenerate Q on V if it contains V and IF = IF. 1 as distinct
subspaces so that
(1) x2 = Q(X) for any x E v

(2) V generates C1(Q) as an algebra over IF
(3) Ce(Q) is not generated by any proper subspace of V. I
The third condition (3) guarantees the universal property [see below], and di-
mension 2". Using an orthonormal basis {el, e ~. ., ., e,) for W"q, generating
Cepsp,the condition (1) can be expressed as
while condition (3) becomes ele2 . . .en # f1, as in Porteous 1969. Condition
(3) is needed only in signatures p - q = 1 mod 4 where (elez . . .e n ) 2= 1. The
relations (1.a) without (3) also generate a lower-dimensional non-universal
algebra of dimension 2n-1 in any signature p - q = 1 mod 4 in which all
the basis elements ei commute with e l 2 , , , , = elez.. .en. No similar non-
universal algebra exists in even dimensions, and so it is correct to introduce the
Clifford algebra of the Minkowski space-time without condition (3). However,
in arbitrary dimensions it is controversial to omit condition (3).
The above definition gives a unique algebra only for non-degenerate (non-
singular) quadratic forms Q. In particular, the definition is not good for a
degenerate Q, for which e l e z . . .en = 0, as is shown by the following two
counter-examples where Q = 0.
1. Define for x, y E V, dimV = n, the product x y = 0. This makes the direct
sum IF $ V an associative algebra with unity 1. It is of dimension n 1. +
2. Introduce a product in /\R3 by eiej = ei A ej for all i, j = 1 , 2 , 3 and
eleze3 = 0. Thus the subspace R $ R3 $ /\2 IW3 of /\IW3 is a 7-dimensional
associative algebra with unity, generated by R and R3.
This shows that it is not possible to replace condition (3) by the requirement
14.3 Definition b y generators and relations 191
that only parallel vectors commute. We could include arbitrary quadratic forms
Q by requiring instead of condition (3) that the product of any set of linearly
free vectors in V should not belong to F. However, even this would leave
some 'ambiguity' in the definition by generators and relations. The above
definition results in a unique algebra only 'up to isomorphism'. Here are two
more examples to clarify the meaning of this statement:
3. The multiplication table of the exterior algebra AIR2 with respect to the
basis (1, e l , e2, el A e2) is
Introduce a second product on /\R2 with multiplication table
where b > 0. Denote the second product by U A V Rearrange

. the multiplication
table of the second product into the form
which shows that we have generated a new exterior algebra AE%~ on R2, differ-
ent from AIR2 but isomorphic with AIR2. In other words, we have introduced
a linear mapping a : R2 + j\IR2 for which a(ei) = ei, i = 1 , 2 , and
&(elA e2) = el A e 2 = el A e 2 +b so that it is the identity on R2, pre-
serves even-odd grading and gives an isomorphism between the two products,
a ( u A v ) = a(u) ~ a ( v ) .
+
4. An orthonormal basis el, e 2 for R2 satisfying eiej eje; = 2Sij gener-
ates the Clifford algebra CL2 = CL2,0 with basis (1, el, e 2 , e12) where e l 2 =
192 Definitions of the CliffordAlgebra
Introduce a second product on

1:"
ele2 (= el A e 2 ) .We have the following multiplication table for
e2
el2
-12
-e2 el
-el
-1
el2 with multiplication table
el2:
I el e2 '312
el 1 el2 b + e2 - bel
e2 -el2-b 1 -el - be2
el2 -e2 - be1 el - be2 - 1 - b2 - 2be12
The anticommutation relations eiej + e j e ; = 2Sij are also satisfied by the new
product, and one may directly verify associativity. As the real number b varies
we have a family of different but isomorphic Clifford algebras on R2.
14.4 Universal object o f quadratic algebras

The Clifford algebra CC(Q) is the universal associative algebra over IF generated
by V with the relations x2 = Q(x), x E V.
Let Q be the quadratic form on a linear space V over a field IF, and let A
be an associative algebra over IF with unity 1 ~ A. linear mapping V + A,
x + px such that
(cpx)2 = Q(x) . 1~ for all x E V
is called a Cliffordmap. The subalgebra of A generated by IF = IF. l A and V
(or more precisely by the images of F and V in A) will be called a quadratic
algebra. The Clifford algebra el(&) is a quadratic algebra with a Clifford
map V + el(&), x + yx such that for any Clifford map cp : V + A there
exists a unique algebra homomorphism II, : Cl(Q) + A making the following
diagram commutative:
This definition says that all Clifford maps may be obtained from y : V + Cl(Q)
which is thereby universal.
2 The term quadratic algebra is commonly used for something else: in a quadratic algebra
each square x2 is linearly dependent on x and 1.
14.5 Clifford algebra as a quotient of the tensor algebra 193
The definition by the universal property is meaningful for an algebraist who
knows categories and morphisms up to the theory of universal objects. A
category contains objects and morphisms between the objects. Invertible mor-
phisms are called isomorphisms. In a category there is an initial (resp. final)
universal object U, if for any object A, there is a unique morphism a : U + A
(resp. A + U). The universal objects are unique up to isomorpl~isin.In many
categories there exists trivially the final universal object, which often reduces
to 0. The Clifford algebra is the initial universal object in the category of
quadratic algebras.
Example. Consider the category of quadratic algebras on WsQ. In this cate-
gory the initial universal object is the Clifford algebra C%,q of dimension 2n
and the final universal object is 0. Between these two objects there are no
other objects, when p - q # 1 mod 4. However, there are four objects in this
category, when p - q = 1 mod 4; between Cep,, and 0 there are two algebras
both of dimension 2"-'; in one we have the relation elez . . . e n = 1 and in the
other el e 2 . . .en = - 1; these two algebras are not isomorphic in the category
of quadratic algebras (the identity mapping on R P J ~does not extend to an
isomorphism from one algebra to the other); however, they are isomorphic as
associative algebras (in the category of all real algebras). I
The above definition of Clifford algebras is most suitable for an algebraist who
wants to study Clifford algebras over commutative rings (and who does not
insist on injectivity of mappings IF + A and V + A). However, this approach
does not guarantee existence, which is given by constructing the Clifford alge-
bra as the quotient algebra of the tensor algebra (which in turn is regarded by
algebraists as the mother of all algebras).
14.5 Clifford algebra as a quotient of t h e tensor algebra

Chevalley 1954 p. 37 constructs the Clifford algebra C&(Q) as the quotient
algebra @V/I(Q) of the tensor algebra @V with respect to the two-sided ideal
I(&) generated by the elements x@x-Q(x) where x E V. See also N. Bourbaki
1959 p. 139 and T.Y. Lam 1973 p. 103. The tensor algebra approach gives a
proof of existence by construction - suitable for an algebraist who is interested
in rapid access to the main properties of Clifford algebras over commutative
rings.
In characteristic zero we may avoid quotient structures by making the exterior

algebra /\ V concrete as the subspace of antisymmetric tensors in @V. For
example, if x , y E V, then x A y = i ( x @ y - y @ x) E V. More generally,
194 Definitions of the Clifford Algebra
a simple k-vector xl A x2 A . . . A xk is identified with
where the linear operator Alt : @V + /\ V, called alternation, is a projection

operator Alt(@V) = /\ V satisfying u A v = Alt(u @ v).
Similarly, we may obtain an isomorphism of linear spaces /\ V + CL(Q) by
identifying simple k-vectors with antisymmetrized Clifford products
thus splitting the Clifford algebra CC(Q) into fixed subspaces of k-vectors
l\kV c CL(Q). Any orthogonal basis el, e2, . . . , en of V gives a correspon-
dence
of bases for /\ V and CL(Q).
Exercises
1. Show that the subspace Alt(@V) of @ V is not closed under the tensor
product.
2. Show that A @ B - B @ A = ~ ( A - B BA) for bivectors A, B E V.
Bibliography
E. Artin: Geometric Algebra. Interscience, New York, 1957, 1988.
N. Bourbaki: Algdbre, Chapitre 9, Formes sesquiline'aires et formes quadratiques. Her-
mann, Paris, 1959.
C. Chevalley: Theory of Lie Groups. Princeton University Press, Princeton, NJ, 1946.
C. Chevalley: The Algebraic Theory of Spinors. Columbia University Press, New
York, 1954.
W.K. Clifford: Applications of Grassmann's extensive algebra. Arner. J . Math. 1
(1878), 350-358.
W.K. Clifford: On the classification of geometric algebras; pp. 397-401 in R. Tucker
(ed.): Mathematical Papers by William Kingdon Clifford, Macmillan, London, 1882.
Reprinted by Chelsea, New York, 1968. Title of talk announced already in Proc.
J. Helmstetter: Algkbres de Clifford et algkbres de Weyl. Cahiers Math. 25, Mont-
pellier, 1982.
I.R. Porteous: Clifford Algebras and the Classical Groups. Cambridge University
3 Another alternative is to omit the factor 1.This gives in all characteristics a correspon-
dence between the exterior product and tfi; antisymmetrized tensor product.
15
Witt Rings and Brauer Groups
Quadratic forms can be classified by their Witt classes in Witt rings (of con-
cerned fields). This is a slightly coarser classification than the one given by the
Clifford algebras (of quadratic forms).
Associative algebras with unity can be studied by means of Brauer groups
(of fields); for this one needs to know tensor products of algebras. These topics
will be discussed in this chapter.
15.1 Quadratic forms

A quadratic form on a linear space V over a field F is a map Q : V + IF such
that for any X E F and x E V
Q(W= X2Q(x)
and such that the map
is bilinear, that is, linear in both arguments. A linear space with a quad-
ratic form on itself is called a quadratic space. A quadratic form obeys the
parallelogram law
In characteristic # 2 the quadratic form may be recaptured from its sym-

metric bilinear form
since Q(x) = B(x, x). In characteristic # 2 the theory of quadratic forms is

the same as the theory of symmetric bilinear forms, but in characteristic 2 there
196 Witt Rings and Braver Groups
are quadratic forms which are not induced by any symmetric bilinear form.
Example. Consider the 2-dimensional space IF: over F2 = ( 0 , l ) . For the
quadratic form Q : F; -+ IF2, x = (XI,2 2 ) -+ 21x2 there is no symmetric
bilinear form B such that Q(x) = B(x, x ) . However, there is a bilinear form
B , not symmetric, such that Q(x) = B(x, x ) , namely B(x, y ) = xlyz, but the
matrix
cannot be symmetrized by adding an alternating matrix with entries in IFz =

{O, 11. I
R e m a r k . We shall not be concerned with characteristic 2, or non-syrr~metric
B , except at the end of this book when considering the relation between the
exterior algebra and the Clifford algebras. The role of non-symmetric B will
be described in Chevalley's construction CL(Q) c End(A V). I
A non-zero vector x is null or isotropic if Q(x) = 0. A quadratic form is
anisotropic if Q(x) = 0 implies x = 0, and isotropic if Q(x) = 0 for some
x # 0. A bilinear form is non-degenerate if B(x, y ) = 0 for all y E V i~nplies
x = 0. An anisotropic quadratic form is always non-degenerate.
A 2-dimensional isotropic but non-degenerate quadratic space is known as
the hyperbolic plane. A hyperbolic plane has a quadratic form yly2 or equiv-
+
alently xf - x i (choose yl = xl 2 2 and y2 = X I - 2 2 in characteristic
# 2)-
A subspace with a vanishing quadratic form is totally isotropic. In a non-
degenerate quadratic space with a totally isotropic subspace S, there is another
totally isotropic subspace S' such that S n S' = (0) and d i m s = dims'. A
non-degenerate quadratic space is neutral or hyperbolic if it is a direct sum of
two totally isotropic subspaces (necessarily of the same dimension). A neutral
quadratic space is even-dimensional.
Examples. 1. A Euclidean space Rn has an anisotropic (and positive defi-
nite) quadratic form on itself, i.e., x # 0 implies x . x > 0. This enables us to
introduce the norm or length 1x1 = of x E Rn.
2. The real quadratic space RPlq is non-degenerate and for non-zero p, q also
isotropic (indefinite). The dimension of its maximal totally isotropic subspace
is p or q according as p 5 q or p 2 q, respectively. This number is called the
isotropy index (or Witt index) of RPfQ.
+
3. The quadratic form $4 x; on Fi is non-degenerate but isotropic, since
+
l 2 22 = 0 mod 5. It is also neutral.
+ +
4. The quadratic forms x4 22; and xf 32.; are anisotropic on IF:. I
15.1 Quadratic forms 197
Two quadratic spaces (V, Q) and (V', Q') are isometric if there is a linear
isomorphism L : V -+ V' such that Q1(Lx) = Q(x), or equivaiently, in char-
acteristic # 2,
B1(Lx,Ly) = B(x, y) for all x , y E V.

We will express the isometry as (V, Q) z (V', Q') or simply Q z Q'. A self-
isometry (or automorphism) of Q on V is a linear isomorphism L : V -+ V
such that Q(L(x)) = Q(x) for all x E V; these self-isometries of Q form the
orthogonal group O(V,Q) .
Two vectors x , y such that B(x, y) = 0 are said to be orthogonal (in the
case of a symmetric B). If a quadratic space is a direct sum of two subspaces,
(Vl , Q1) $ (V2,Q2), such that B(xl, x2) = 0 for all x l E Vl and x2 E V2, it is
an orthogonal sum denoted by Vi I V2 or Q1 IQz.
There is also, for two quadratic spaces (Vl,Q1) and (V2,Q2), the tensor
product Vl @V2of dimension (dim K ) (dim V2), with a quadratic form satisfying
for decomposable elements x l 8 xz with x l E Vl and x2 E V2.

The symmetric matrix B(ei, ej) can be diagonalized (in characteristic # 2);
as a consequence any quadratic form is isometric to a diagonal form d1x2 +
+ +
d 2 ~ ; . . . dnx i for some dl, d2, . . . ,dn E IF. We shall write
(dl, d 2 , .. . , d,) to denote dlxi + d2xi + . . . + dnx:.
The orthogonal sum and the tensor product appear in diagonal form as
Examples. 1. The real quadratic space RPlP is neutral and an orthogonal

sum of p copies of hyperbolic planes R1ll each with a quadratic form (1, -1).
2. The hyperbolic plane over IF, char IF # 2, is isometric with xlx2 and
x4 - 2; 2 (1, -1).
3. In IFs, (1) $ (2) since 2 $ ,'$I the set of non-zero squares, but (2) E (3)
since 2xi = 3(2x1)'.
4. The quadratic forms (1,2) and (1,3) are isometric on FE as can be seen
+
by the identity 24 22; = x; + ) ~ 5 and the linear isomorphism
3 ( 2 x ~mod
( X I ,22) -) (21,222).
5. The quadratic forms (1,l) and (1, -1) are isometric on IF; since z: - x i =
XI + (2~2)~. I
198 Wztt Rings and Brauer Groups
15.2 Witt rings
Two quadratic spaces V, Q and V', Q' are said to be in the same Witt class if
Q I(-Q') is neutral. A non-degenerate quadratic space is an orthogonal sum
V = Va IVh of an anisotropic subspace Va and a neutral subspace Vh. The
anisotropic part Va is unique up to isometry. It follows that V and V' are in
the same Witt class iff Q, I (-Qb) is neutral, or equivalently iff Qa z Qb.
This results in a correspondence
( W i t t class of V = Va @ Vh t)isometry class of Va (
exactly one anisotropic isometry class being included in each Witt class.
Examples. Let the ground field be IF5.
1. From the orthogonal sum (1) I(1,2) z (1,1,2) we can cancel the hyperbolic
plane ( 1 , l ) to extract the anisotropic part (2). So the sum (1) I(1,2) is in
the Witt class of (2).
2. From the tensor product (1,2) 8 (3) 21 (3,2 . 3 = 1) there is nothing to
cancel since ( 3 , l ) z (1,2) is anisotropic. So (1,2) 8 (3) is in the Witt class of
(42). I
The orthogonal sum I and the tensor product 8 provide the set of all the
Witt classes over F with a ring structure yielding the Witt ring W(IF).
The opposite of Q is represented by -Q in W(IF). The neutral quadratic
forms, in particular (0), represent the zero of W(IF). The 1-dimensional form
(1) corresponds to the multiplicative unity of W(F). The zero (0) and the
unity (1) are the only idempotents in W(IF). The even-dimensional quadratic
forms induce an ideal of W(IF).
Examples. 1. The Witt ring W(IF5) contains four anisotropic isometry classes
(0)) (I), (2)) (1,2). The addition and multiplication tables of W(IF5) are
The Witt ring W(IF5) is isomorphic to the group algebra Z2p,X/IFF]; and the
additive group of W(IF5) is Z2 x Z2, but as a ring W(IF5) $ 'Z2 = Z2 x Z2.
2. The 1-dimensional line IF5 with quadratic form (2) has the quadratic field
extension IF5 (A)z F25 as its Clifford algebra. The Clifford algebras of both
(1) and (-1) split as the double-ring 21F5 = 1Fg x IF5. Therefore, as algebras
Ce((2)iF5) $ ce((l)l F5) ce((-1),IF5).
1 Denoting the Clifford algebra of the n-dimensional quadratic space p(1) I q(-1) by
15.2 Watt rings 199
3. The Witt ring W(F7) contains four anisotropic isometry classes (0), (I),
(1, I), ( l , l , I), and W(F7) = Z4. I
The finite fields IFq. The Witt ring of IFq, q = pm, char p # 2, contains four
anisotropic isometry classes
(O),(l)() ( 1 where s$!IFy for p = l m o d 4 ,
('4, (I), (1, I), ( & I >1) for p = 3 mod 4.
The corresponding Witt rings are W(IFq) 2 Zz[IF:/IFy], q = 1 mod 4, and
W(IFq) 21 Z4, q = 3 mod 4. All quadratic forms over the finite fields are
isotropic in dimensions n > 3.
The real field R. A field IF is ordered if there is a subset P c IF (of positive
+
numbers) such that for all a, b E P also a b,ab E P , and, for all a E F
exactly one of a E P , a = 0, and -a E P holds. The statement a - b E P is
also written a > b. An ordered field IF has an absolute value IF -+ P, x + 1x1
defined by setting 101 = 0, 1x1 = x for x > 0, and 1x1 = -x for -x > 0.
In an ordered field, IFo c P. If all the positive numbers have square roots,
then there is a unique ordering with P = p. The following holds for any
ordered field F such that P = IFo, but we shall only consider the real field R.
There exist exactly two anisotropic forms on R n , namely the positive definite
(1,1, . . . ,1) and the negative definite (-1, -1, . . . , -1). A non-degenerate
quadratic form on Rn is isometric to
which we abbreviate as p(1) I q(-1). The real quadratic space with this
quadratic form is denoted by RPlq. The integer p - q is called the signature of
RPI~.
The signature map sending RPlq to p-q gives a ring isomorphism W(R) Z. =
As a consequence, the Clifford algebras of non-degenerate real quadratic spaces
can be listed by the symbols ClPtq,denoted more fully by ClPjq(Rn)= Cl(Rf"q)
or Ct(p(1) l q(-I), Rn).
The complex field @. The Witt ring of @ contains only two anisotropic
isometry classes, namely (0) and (I), and W(@) 2: Z2. We only have to
distinguish between even- and odd-dimensional spaces over @.
Exercises 1,2
Cep,q(lE'n), this example shows that t l ~ enotion does not reach all the Clifford algebras
over arbitrary fields F.
2 T h e real linear space Rn with the positive definite quadratic form n ( 1 ) = ( 1 , 1 , . . ., 1) is
t l ~ eEuclidean space Rn.
15.3 Algebras
An algebra A over a field IF is a linear (that is a vector) space A over F
together with a bilinear map A x A + A, (a, b) + ab, the algebra product.
+ + +
Bilinearity means distributivity (a b)c = ac bc, a(b c) = ab a c and +
(Xa)b = a(Xb) = X(ab) for all a , b , c A
~ and X E F .
Examples. 1. The 2-dimensional real linear space R2 together with the pro-
+
duct (el, y1)(x2,y2) = (x1x2 - y1y2, X I y2 x2y1) results in the real algebra of
complex numbers C.
2. The double-ring 2~ of a field F has a product (al, bl)(a2, b2) = (a1a2,blb2)
making it a 2-dimensional algebra over the subfield denoted by IF(1,l) =
{(A, A) I E F).
3. The matrix algebra of real 2 x 2-matrices Mat(2, R) is a 4-dimensional real
associative algebra with unity I.
4. The real linear space R3 together with the cross product a x b is a (non-
associative) Lie algebra. I
An algebra is without zero-divisors if ab = 0 implies a = 0 or b = 0. In a
division algebra D the equations ax = b and ya = b have unique solutions
x, y for all non-zero a , b E D. A division algebra is without zero-divisors, and
conversely, every finite-dimensional algebra without zero-divisors is a division
algebra. If a division algebra is associative, then it has a multiplicative unity
and each non-zero element admits a unique inverse (on both sides).
An algebra with a multiplicative unity is said to admit inverses if each non-
zero element admits an inverse.
Examples. 1. The quaternions W form a real associative but non-commutative
division algebra with unity 1.
2. Define the following product for pairs of quaternions:
This makes the real linear space W x W a real algebra, the Cayley algebra of
octonions 0. The Cayley algebra is non-associative, a o ( b o c ) # (a o b) o c , but
alternative, (a o a ) o b = a o (a o b), a o (bob) = (a o b) o b. It is a division algebra
with unity 1.
3. Consider a 3-dimensional real algebra with basis (1, i, j) such that 1 is
the unity and i2 = j2 = -1 but i j = j i = 0. The algebra is commutative,
non-associative and non-alternative. It admits inverses, but the inverses of the
+
elements of the form xi y j are not unique, (xi $- yj)-l= X(yi - x j ) - ir+jy. z2+y2
It has by definition zero-divisors, and cannot be a division algebra. I
An isomorphism or anti-isomorphism of algebras A and B is a linear isomor-
15.4 Tensor products of algebras, Braver groups 20 1
phism f : A -+ B such that
respectively. An automorphism or anti-automorphism of an algebra A is an

isomorphism or anti-isomorphism A -+ A, respectively. An automorphism or
anti-automorphism f of A such that f (f (x)) = x for all x E A is an involution
or an anti-involution, respectively.
The only algebra automorphisms of @, regarded as a real algebra, are the
identity and the complex conjugation t -+ 2.
The only automorphisms of the real algebra 'R are the identity and the swap
The swap acts like the complex conjugation of 6,since
Two automorphisms or anti-automorphisms a ,,8 of an algebra A are said

to be similar if there is an automorphism y of A such that cry = yp. If no
such y exists then a and p are said to be dissimilar.
The identity automorphism is similar only to itself. Consequently, the two
involutions of the real algebra @ are dissimilar, and the two involutions of the
real algebra 'R are dissimilar.
Exercises 3,4
15.4 Tensor products of algebras, Brauer groups

The tensor product of two algebras A and B over a field IF is the linear space
A @ B made into an algebra with the product satisfying
(a 8 b) (a' @ b') = (aa') @ (bb')
for a , a l E A and b, b' E B. This algebra is also denoted by A 8 B or, to
emphasize the ground field, by A @rB.
In the special case of finite-dimensional associative algebras with multiplica-
tive unity, the statement C = A @ B can be tested by the following conditions
posed on the subalgebras A and B of C:
(i) ab = ba for any a E A and b E B,
(ii) C is generated as an algebra by A and B,
(iii) dim C = (dim A) (dim B).
Examples. 1. G BlwW 2 Mat(2, @) the real matrix algebra of 2 x 2-matrices
with complex numbers as entries.
202 Witt Rings and Brauer Groups
I
Exercises 5,6
A two-sided ideal T of an algebra A is a subalgebra such that ta E T and
at E T for all t E T and a E A. An algebra A is called simple if it has no two-
sided ideals other than 0 and A. The center Cen(A) of an algebra A consists
of the elements commuting with all the elements of A:
Cen(A) = {c E A I ac = ca for all a E A}.
An algebra with multiplicative unity 1 is called central if Cen(A) = IF. 1 2: F.
A finite-dimensional central simple associative F-algebra A with multiplicative
unity is isomorphic to Mat(d,D) for some suitable division ring D (and division
algebra over IF).
The opposite AOPp of an algebra A is the linear space A with a new product
opp[abj of a, b E A given by opp[ab] = ba. Two central simple associative IF-
algebras A and B are in the same Brauer class if A @ BOPP E Mat(d, F) for
some integer d. The tensor product of algebras induces a product for Brauer
classes, making the set of Brauer classes a group, called the Brauer group Br(IF)
of the field IF.
Examples. Br(R) E {R, W}, Br(C) ci {C}, Br(F5) .v {IFs). I
An algebra A is graded over Z2 = {0,1} if it is a direct sum of two subalgebras
A = A. $ A1 so that AiAj c Ai+j [the indices are added modulo 21. For two
graded algebras A = A. $ A1 and B = Bo $ B1 the graded tensor product
A C$B is the linear space A @ B provided with the product determined by the
formula
(a @ 6) (a' 8 6') = (- l)"(aa') @ (66')
for homogeneous a' E Ai and b E Bj. The graded opposite AO" of a graded
algebra A = A. $ Al is the linear space A with a new product opp[ab] of
a, b E A given by opp[ab] = (-1)"ba for homogeneous elements a E Ai and
b E Bj.
Exercises
1. Determine the addition and the multiplication tables of the anisotropic
isometry classes (0), (I), (1, I), (1,1,1) of W(IF7).
2. Identify as matrix algebras all the Clifford algebras of non-degenerate
quadratic forms over F g .
15.4 Tensor products of algebras, Braver groups 203
3. Show that the two involutions a ( A , ,u) = (p, A) and P(A, p) = (p,X) are
similar involutions of the real or complex algebra 'C.
4. Consider the four anti-involutions of Mat (2,R) sending
Determine which ones of these four anti-involutions are similar or

dissimilar to each other. Hint: keep track of what happens to the matrices
with squares I , I , and -I.

=
5. Show that @ @B @ @ $ (C.
6. Show that W BlwW E Mat(4, R).
Solutions
2. A non-degenerate quadratic form (al, a2,. . . , a,) over IF5 is isometric to

p(1) l q(2) where the numbers p and q mean, respectively, occurrences of
1 , 4 and 2,3 in a l , a2, . . . , a,. The Clifford algebra Ct(p(1) I q(2), I&?) is
isomorphic, as an associative algebra, to the matrix algebra
at (2"f2,IF5) p and q even
at (2,f2, IF5) p and q odd
2 ~ a t ( 2 ( n - 1 ) fIF5)
2, p odd, q even
~at(2("-')/', IF5 (d))p even, q odd.
For instance, Ct((1,2), IF:) = Mat(2, IF5) by the correspondences
of an orthogonal basis {el, ez) of (1,2) on IF;.
3. Choose y(A, p) = (X,p) or y(A, p) = (A, ,!i) to find a y = yp.
4. Only two of the anti-involutions are similar,
as can be seen by choosing the intertwining automorphism
for which a y = yp.

5. We must have 1 8 1 2 ( 1 , l ) and may choose i 8 i z (1, - 1) or
i 8 i z (- 1 , l ) . If we choose the latter, we may still choose 1 8 i N_ (i,i),
i 8 1 N_ (i, -i) or 1 8 i N_ (i, -i), i 8 1 z (i, i) or opposites of both.
+ + + + +
6. Choose for a = ao ial jaz ka3, b = bo ibl jb2 kb3 in W the +
matrix representations
and check that the matrices commute and form two isomorphic images of
the ring W.
Bibliography
M.-A. Knus: Quadratic Forms, Clifford Algebras and Spinors. Univ. Estadual de
Campinas, SP, 1988.
T.Y. Lam: The Algebraic Theory of Quadratic Forms. Benjamin, Reading, MA, 1973,
1980.
E. Witt: Theorie der quadratischen Formen in beliebigen Korpern. J. Reine Angew.
Math. 176 (1937), 31-44.
16
Matrix Representations and Periodicity of 8
The Clifford algebra Ce(Q) of a quadratic form Q on a linear space V over

a field IF contains an isometric copy of the vector space V. In this chapter we
will temporarily forget this special feature of the Clifford algebra el(&). Then
the Clifford algebra of a non-degenerate quadratic form is nothing but a matrix
algebra or a direct sum of two matrix algebras. We have already identified the
following Clifford algebras:
c e 2 N M a t ( 2 , ~ ) ,ceo,2Nw,
Ct3~Mat(3,C),C~O,~EW$W,
Ceq N Mat(2,W)) Ce3,1 N Mat(4,R), Ctl,3 2: Mat(2, W).
We will find a general pattern for matrix images of Clifford algebras C$,, of
non-degenerate quadratic spaces IWPpQ. We will see that ClP,, are isomorphic
to real matrix algebras with entries in R , C, W or in 2R = R $ R, 2W = W$ W,
that is, their matrix images are
Mat(d, R), Mat(d, C), Mat(d, W ) or
'Mat(d, R) = Mat(d, 'R), 2Mat(d,W) = Mat(d, 'W).
REVIEWOF MATRIXIMAGES
OF Cl,,,, p +q <5
The quadratic space RPlq is an n-dimensional real vector space Rn, n = p + q,
with a non-degenerate symmetric scalar product
x .Y = xlyl + - .. + XPYP - xp+lyp+l - . . . - xp+,yp+,.

The scalar product induces the quadratic form
X .X = x12 + . . . + xp2 - xp+l
2
- . . . - XP+,.
2
A real associative algebra with unity 1 is the Clifford algebra Clp,, on RPJQ
206 Matrix Representations and Periodicity of 8
if it contains R P l q and R = R . 1 $ Wjq as subspaces so that RplQ generates
CtpPqas a reaI algebra and
for all x E RP'Q.Furthermore, we require that Ctp,, is not generated by any

proper subspace of IWP'Q.
+
The identity x2 = x . x has a polarized form xy yx = 2x . y. In an
orthonormal basis el, e2, . . . ,en of RPlq this means
where gjj = ei . e j or gjj = 1, i L p , gii = -1, i > p, and gjj = 0 , i # j. The

above identity is a condensed form of the relations
The requirement that no proper subspace of RPjq generates Ctp,, results in the
constraint ele2.. . en # f1, needed only in the case p - q = 1 mod 4.
+
The Clifford algebra Clp,,, p q = n, is of dimension 2n. If the constraint
ele2.. .en # f 1 is omitted, then the resulting algebra could be of dimension
2n or 2n-1,the lower value being possible only if p-q = 1 mod 4. In the lower-
dimensional case we have ele2. . .en = f1, the algebra itself being isomorphic
to the two-sided ideal $ ( 1 f el2,,,,)Ctp,,.For instance, the negative definite
quadratic space R0p3 has an 8-dimensional Clifford algebra Cto,3 2 W $ W,
which is a direct sum of two ideals i ( 1 f e123)Ct~,3, both isomorphic to the
4-dimensional quaternion algebra W.
16.1 The Euclidean spaces Rn

In the positive definite case, p = n, q = 0, of the Euclidean space we abbreviate
Rn10to Rn and its Clifford algebra Ctntoto Ctn. In the Euclidean case we can
speak of the length 1x1 of a vector x E Rn given by 1x1' = x . x.
The Euclidean plane R2. Consider the Euclidean plane R2. The Clifford
algebra Ct2 of R2 is generated by an orthonormal basis e l ,e2 of R2. We have
the multiplication rules
e 4 = 1 , e2,=1 [ell = 1 , le21 = 1
ele2 = -egel corresponding to el Ie2.
Using ele2 = -e2el and associativity we find (ele2)2= -eqez which implies
(ele2)2= -1. This indicates that ele2 is neither a scalar nor a vector, but a
1 In the negative definite case we can also speak of the length 1x1 of x E R ~ given
> ~by
(x12 = -X . X.
16.1 The Euclidean spaces Rn 207
new kind of unit, called a biuector. The Clifford algebra C12 is 4-dimensional
with a basis consisting of
1 a scalar
el, e 2 vectors
ele2 a bivector.
Write for short el2 = ele2. The Clifford algebra Ct2 has the following multi-
plication table:
The Clifford algebra C12 of the Euclidean plane R2 is isomorphic, as an asso-

ciative algebra, to the matrix algebra of real 2 x 2-matrices Mat(2, R). This is
seen by the correspondences
It should be emphasized that the Clifford algebra Ct2 has more structure than
the matrix algebra Mat(2,lR). The Clifford algebra C12 is the matrix algebra
Mat(2, R) with a specific subspace singled out (and a quadratic form on that
subspace making it isometric to the Euclidean plane R2). I
T h e 3-dimensional Euclidean space IR3. Consider the 3-dimensional Eu-
clidean space lR3. The Clifford algebra C13 is generated by an orthonormal
basis {el,e2,e3) of lR3. This time there are three linearly independent bivec-
tors elz,el3,e23, each being a square root of -1. In addition, there is the
volume element el23 = eleze3 which squares to -1 and commutes with all
the vectors el, e 2 , e3 and thereby also with all the elements of the algebra Cl3.
The Clifford algebra Ct3 is 8-dimensional over R and has a basis consisting of
1 a scalar
el, e 2 , es vectors
e12,e13,e23 bivectors
el23 a volume element.
The Clifford algebra Ct3 is isomorphic, as a real associative algebra, to the
matrix algebra Mat(2, C) of 2 x 2-matrices with entries in C. The isomorphism
Ct3 z Mat(2, C) of real associative algebras is fixed by the correspondences
The matrices above are known as Pauli spin matrices. The multiplication of
the unit vectors, ele2e3 = el23, results in the correspondence
As noted above, the volume element e123, such that e:23 = -1, commutes
with all the elements of the algebra Ct3; that is, it belongs to the center of
Ct3. This enables us to view Ct3 as a complex algebra isomorphic, as an asso-
ciative algebra, to the matrix algebra of complex 2 x 2-matrices Mat(2, C). 1
T h e 4-dimensional Euclidean space R4. The Clifford algebra Ct4 of the
Euclidean space R4 is isomorphic, as an associative algebra, to the real algebra
Mat(2, NI) of 2 x 2-matrices with entries in the division ring of quaternions NI.
Using an orthonormal basis {el, e 2 , e3, e4) of R4 we can find the correspon-
dences
The Clifford algebra Ct4 is of dimension 16 and has a basis consisting of

1 a scalar
el, e2, e3, e4 vectors
e12, e13, e14, e23, e24, e34 bivector~
e123, e124, e134, e234 3-vectors
el234 a 4-volume element.
+ + + +
An arbitrary element u = ( u ) ~ (u)l (u)2 ( u ) ~ (u)4 in Ct4 is a sum
of a scalar (u)o, a vector ( u ) ~ a, bivector ( u ) ~a, 3-vector ( u ) ~
and a volume
16.1 The Euclidean spaces Rn 209
element (u)4.
Split complex numbers IW $ R. The Clifford algebra Cll of the Euclidean
line IW1 = R is spanned by 1, el where e: = 1. Its multiplication table is
The Clifford algebra Cll is isomorphic, as an associative algebra, to the double-

field R $ IW of split complex numbers. The product of two elements ( a l ,a 2 )
and (PI,P2) in R $ R is defined component-wise:
The isomorphism Cll 2: R $ IW can be seen by the correspondences
The Clifford algebra Cll E IW $ R is a direct sum of two ideals spanned by the
+
idempotents a(l el) E (1,O) and i ( 1 - el) 21 (0,l). I
T h e 5-dimensional Euclidean space R5. The Clifford algebra C15 of IW5 is
isomorphic to 2Mat(2,W) = Mat(2, 2W), as can be seen by the correspondences
The Clifford algebra C15 has two central idempotents
which both project out of C15 an isomorphic copy of Mat(2, W), that is, i ( l f
e12345)C152: Mat(2, W).An isomorphic copy of i(l
fe12345)C15is constructed
within another subspace of C15 in the following counter-example.
Counter-example. Consider the subspace of scalars, vectors and bivectors R $
+ +
R5 $ /\2 R5 of dimension 1 5 45(5 - 1) = fr25. Introduce in this subspace
a new product u o v defined by (one of the following)
210 Matrix Represeritations and Pzriodicity of 8
where (w)0,1,2 = (W)O (w)1 + + ( w ) ~ This

. new product is associative and
satisfies
xox = lx12 for x E IR5.
However, el o e 2 o e3 o e4 o e5 = f1. As a sample, this new product satisfies
el o e 2 = el2, el o e l 2 = e 2 , e l 2 o el2 = -1, el2 0 e23 = el3
, o e34 = fe5.
el o e23 = ~ e 4 5 el2
The multiplication table of this new ~ r o d u c tis given by the following matrices
This serves as a counter-example to a belief that the Clifford algebra would be

uniquely generated by its subspaces IR and IRn. I
T h e 3-dimensional anti-Euclidean space R0l3
The anti-Euclidean space I R 0 t 3 has a negative definite quadratic form sending
+ +
a vector x = xlel x2e2 x3e3 to the scalar
X . x = -(xf + xi +xi).
An orthonormal basis {el, e 2 , e3) of R0p3 obeys the multiplication rules
e2 2 = e; = -1
- e2 and
8182 = -€!281, €383 = -8361, 8283 = -€!3e2.
These relations are satisfied by the unit quaternions
in MI. The rule i j k = -1, or eleze3 = -1, means that the real algebra
W = IR $ IR0t3 is generated by a proper subspace of ~ ~In other
1 ~words,
IR012 .
each quaternion can be expressed in the form x = xo xlel x2e2 236182 + + +
where e3 = ele2. This matter is expressed by saying that W is an algebra of
the quadratic form
xlel + 2282 + 2383 + -(x; + x i + 2:)

although it is not the Clifford algebra Cto,3. The 8-dimensional Clifford algebra
C ~ Ois , isomorphic,
~ as an associative algebra, to the direct sum W $ W.This
16.2 Indefinite metrics RPlY
can be seen by the correspondences
ct0,3
1 (1,l)
el, e2, e3 (i, -i), (j,-j), (k, -k)
e23, e x , e12 (i,i), (j,j), (k, k)
el23
The Clifford algebra Cto,3 of R0v3 is the universal object in the category of
algebras of the quadratic form
or for short in the category of quadratic algebras. If there are other objects
in this category, they are quotients of the universal object with respect to a
two-sided ideal. This gives us two other algebras of dimension 4; in one of them
we have the relation elenes = 1 and in the other eleze3 = -1. These two
algebras of dimension 4 are linearly isomorphic to R $ R0p3. In the category
of quadratic algebras these two algebras of dimension 4 are not isomorphic
with each other, which means that the relations ele2e3 = 1 and ele2e3 = -1
prevent the identity mapping on R013 being extended to an isomorphism in this
category. However, in the category of all associative algebras these two algebras
of dimension 4 are isomorphic with each other (and with the quaternion algebra
= R $ R0p3). The isomorphism can be seen by the mappings
el + el, e 2 + e 2 and e3 -+ -es. I
16.2 Indefinite metrics RPJY

The hyperbolic plane R1ll. The hyperbolic plane is the linear space R2
endowed with a quadratic form
(u, v) -$ uv
which by change of variables u = X I + x2, v = xl - 2 2 is seen to be
(xl, x2) + x: - x;.
Thus the hyperbolic plane is indefinite, neutral and has the Lorentz signature
R1pl. The Clifford algebra CtlPl of R1pl is isomorphic, as an associative algebra,
2 The term quadratic algebra is customarily used for something else: in a quadratic algebra
z2 is linearly dependent on z and 1.
to the matrix algebra Mat(2, R) by the correspondences
ce1,1 Mat(2,R)
(A Y )
e1?e2
(: ) ( ;)
Note that the Clifford algebras Celtl and Ce2 21 Mat(2, R) are isomorphic as
associative algebras but non-isomorphic as quadratic algebras. I
T h e Minkowski space-time lR3l1. The elements of an orthonormal basis
{el, e2, e3, e4) of R3t1 anticornmute, e,ev = -eve,, and have unit squares,
e 2 - - e2 2 --1 , e i = -1. The basis vectors are often given the following
2 -- e3
representation by complex 4 x 4-matrices:
where we recognize the 2 x 2 Pauli spin matrices al, an, a3. It is possible to
represent Cestl by real matrices as follows:
This implies Cespl 21 Mat(4, R). I

T h e Minkowski time-space R1l3. In the signature R113 one usually gives
the following representation by complex 4 x 4-matrices:
-yo = - y o = ( l0 O
-1 )' and
16.2 Indefinite metrics RPlq 213
In addition to the above matrix representation one can represent the Clifford
algebra Cel,3 by the following 2 x 2-matrices with quaternion entries:
Since the Clifford algebra Cel,3 and the matrix algebra Mat(2,W) of 2 x 2-
matrices with entries in W are both real algebras of dimension 16, the above cor-
respondences establish an isomorphism of associative algebras, that is, Ct1,3 N
Mat(2,W).
+
A short look at physics: A vector u = uoyO+ulyl u2y2 +u3y3 with square
u2 = U; - uf - U: - U: can be time-like u2 > 0, null u2 = 0, or space-like
u2 < 0. A time-like vector or non-zero null vector can be future oriented uo > 0
or past oriented uo < 0. A time-like future oriented unit vector u, u2 = 1,
gives the velocity u < c of a real particle by
Physicists might want to observe that the Clifford algebras C!3,1 21 Mat(4, R)
and Cel,3 N Mat(2, W) are not isomorphic as associative algebras, even though
both of them have the same complexification Mat(4, C) with the same complex
structure but with different real structures (= different real subalgebras). The
complexified Clifford algebras C 8 Ctl,3 E C 8 Ce3,1 have a 4-dimensional
irreducible left ideal (8-dimensional real subspace). As a graded left ideal
this ideal is also irreducible. The real algebra Cel,3 has an 8-dimensional
irreducible left ideal, which is also graded. However, the real algebra C&,1 has
a 4-dimensional irreducible ideal, which is not graded (that is CesPl does not
have primitive idempotents sitting in Clt1), and an 8-dimensional irreducible
graded ideal.
THE TABLEOF CLIFFORDALGEBRAS

The Clifford algebra Clp,q, where p - q # 1 mod 4, is a simple algebra of
+
dimension 2n, where n = p q , and therefore isomorphic with a full matrix
algebra with entries in R , C, or W.The Clifford algebra Ctp,q,where p - q =
1 mod 4, is a semi-simple algebra of dimension 2n so that the two central
idempotents 3 ( 1 It ele2 . . . e n ) project out two copies of a full matrix algebra
with entries in R or W. To put it slightly differently, the Clifford algebra C!p,q
has a faithful representation as a matrix algebra with entries in R, C, W or
R $ R, W $ W. In the rings 'R = R $ R and 2W = W $ W the multiplication
is defined component-wise:
16.3 M a t r i x representation Clp+l,q+l N Mat(2, Cepe,,,)

Let {el,e2,. . . , e n ) be an orthonormal basis of RPlQ, n = p + q, generating
the Clifford algebra elp,,. The 2 x 2-matrices
for i = 1,2,..., n,
anticommute and generate the Clifford algebra Clp+l,q+l.In other words, the
Clifford algebra Clp+l,q+l is isomorphic, as an associative algebra, to the al-
gebra of 2 x 2-matrices with entries in the Clifford algebra ClP,,. This can be
condensed by writing Clp+l,q+lN Mat (2, elp,,).
Examples. Recall that Cll z 2R = R $ R by setting el N (1, -1). This
implies the isomorphism C12,121 2Mat(2,R). Recall that Clo,s 21 2W = W $ W,
which implies z 'Mat(2, W).Recall that Celts N Mat(2, W) which implies
Ce2,4N Mat(4,W). I
Supplement an orthonormal basis {el, e 2 , . . . , e n ) of RppQ with two more
anticommuting basis vectors e+ and e- such that e: = 1 and e! = -1 to
form an orthonormal basis of Rp+l*q+l.The generators el, e 2 , . . . , en,e+,e-
of Clp+l,q+l correspond to the generators
of Mat(2, ClP,,), so that the element a E Clp,, is represented by a matrix
where the hat means the grade involution ii = a0 - a1 with a0 = even(a) and
a1 = odd(a). There is another possibility to embed elp,, into Mat(2, C$,,),
so that a E ClP,, is represented by
16.3 Matrix representation Clp+l,q+l 2 Mat(2, CG,,) 215
which is just a multiple of the identity matrix. Since a' = a0 + ale+e- com-
mutes with
we have the correspondence Mat(2,CtP,,) 2: C%+l,q+l, given by
To put all this in another way: The Clifford algebra C$+l,q+l contains an
isomorphic copy of Ce,,, generated by the elements e: = eje+e-, where i =
+
1 , 2 , . . . , n = p q, in such a way that each element of Up,, commutes with
every element of a copy of Ctljl generated by e+ and e-, and further that
C$,, and Ctl,l together generate all of C$+l,q+l . These considerations can
be condensed by writing
where CllS1 2 Mat(2,R).

S y m m e t r y C$,, N Ctq+l,p-l. Take an orthonormal basis {el,e 2 , . . . , en) of
RPpq where p _> 1 and set
ei = el and e: = eiel for i > 1.
The elements e: where i = 1 , 2 , . . .,n anticommute with each other so that
ei2 = e4 and ei2 = -e; for i > 1. Therefore, the subset {ei,ei, . . . , e;) of
Cep,, is a generating set for Ctq+l,p-l. This proves the isomorphism
when p > 1.
Examples. Recall that Ct3 N Mat(2, C), which by symmetry implies Ctl,2 2:
Mat(2,C). Recall that CesSl N Mat(4,R), which implies Ct2,2 N Mat(4,R).
From Ceo,4 2: Mat(2, W) we can first deduce Ctljs 2: Mat (4, W) (by adding a
hyperbolic plane) which implies Ces 2 Mat(4, W). I
16.4 Periodicity of 8
Table 1, of Clifford algebras, contains or continues with two kinds of peri-
odicities of 8, namely for algebras of the same dimension Clp,T N Clp-4,q+4
>
where p 4, and for algebras of different dimension Cept8,, N Mat(l6, Ce,,,).
>
Let us first prove Ce,,, E Ctp-4,q+4 where p 4. Take an orthonormal basis
{el,e2,. . . , e n ) of RPlq and set
e:=eih for i = 1 , 2 , 3 , 4 ,
e: = ei for i > 4,
where h = ele2e3e4. Then the subset {ei, eh, . . . , e;} of Ce,,, is a generating
set for Ctp-4,q+4,which implies the isomorphism
where p > 4. These isomorphisms are due to Cartan 1908 p. 464.

Examples. Recall that Cts N Mat(4,W), which implies Ce2,4 N Mat(4,W).
From Ce3 2: Mat(2, @) deduce first Ce4,1 N Mat(4, @) , which implies Cto,s N
Mat(4, C). From Ce2,2 E Mat(4, R) we first deduce C13,3 2: Mat(8, R); then
by CesP3E Ce4,2 and Ce4,2N Ceo,6 we find Clo,6 N Mat(8,IR). From Ce3,3 N
Mat(8,R) we find Cl4,4 2 Mat(l6,R) and also Clg N Mat(l6,R) and Cto,8N
Mat(l6, R). I
Next, prove Cep+8,q2 Mat(l6, Ce,,,) by showing that Ctp,q+8 N Mat(l6, Ce,,,).
Take an orthonormal basis {el,e2,. . . , e n ,e,+l,. . . , ent8) of RP>q+' where
n = p + q and set
e: = ejen+l...en+8 for i = 1 , 2 , . .. , n = p + q .
Then the subset {ei,e6, . . . , ek) of Cep,q+8generates a subalgebra isomorphic
to Ct,,,. The subalgebra generated by e,+l, . . . ,en+8is isomorphic to Ceo,g N
Mat(l6, R). These two subalgebras commute with each other element-wise and
generate all of Cep,,+8, which shows that
Similarly, Cep+8,qN C!, 8 Mat(l6,R) N Mat(l6, C!,,). These isomorphisms

are due to Cartan 1908.
Example. Note that C l O ,N~ C which implies C t 8 , ~E Mat(l6, C). Recall that
Ce1,l N Mat(2, R) and so CelP9N Mat(32, R). I
16.5 Complex Clifford algebras and their periodicity of 2 217
Table 1. Clifford Algebras Cep,,, p + q < 8.
A(d) means the real algebra of d x d-matrices Mat(d, A) with entries in the
ring A = R, C, W,'R, 'W.
16.5 Complex Clifford algebras and their periodicity of 2

Complex quadratic spaces Cn have quadratic forms
The type of their Clifford algebras Ct(Cn) depends only on the parity of n.
Denote t = [n]. In even dimensions Ct(Cn) N Mat(2', C) and in odd dimen-
sions Ct(Cn) 21 'Mat(2', C).
Table 2. Complex Clifford Algebras Ct(Cn), n < 8.

Exercises
1. Show that Ct;, =
z Ctp,g-l and C t t = ~ t : , ~Cto,n-l.
2. Show that all the algebra isomorphisms presented in this chapter are
special cases of the following:
where Q2 is non-degenerate and V2 is even-dimensional, dim V2 = 2k,

w w~=AER\{o).
Bibliography
M.F. Atiyah, R. Bott, A. Shapiro: Clifford modules. Topology 3, suppl. 1 (1964),
3-38. Reprinted in R. Bott: Lectures on I < ( X ) . Benjamin, New York, 1969, pp.
143-178. Reprinted in Michael Atiyah: Collected Works, Vol. 2. Clarendon Press,
Oxford, 1988, pp. 301-336.
E. Cartan (expos6 d'aprbs l'article d e m a n d de E. Study): Nombres complexes; pp.
329-468 in J. Molk (red.): EncyclopLdie des sciences mathe'matiques, Tome I, vol.
1, Fasc. 4, art. I 5 (1908). Reprinted in E. Cartan: U3uvres complites, Partie 11.
W.K. Clifford: On the classification of geometric algebras, pp. 397-401 in R. Tucker
(ed.): Mathematical Papers by William Kingdon Clifford, Macmillan, London, 1882.
1980.
I.R. Porteous: Topological Geometry. Van Nostrand Reinhold, London, 1969. Carn-
I.R. Porteous: Cliflord Algebras and the Classical Groups. Cambridge University
17
Spin Groups and Spinor Spaces
We have already met in some lower-dimensional special cases the spinor spaces,
minimal left ideals of Clifford algebras, and the spin groups, which operate on
spinor spaces. In this chapter we shall study the general case of Rplq.
SPIN GROUPSAND THE TWOEXPONENTIALS

Review first the special case of the 3-dimensional Euclidean space R3.
17.1 S p i n g r o u p Spin(3) and SU(2)

+ +
The traceless Hermitian matrices xu1 yu2 zu3, with x, y, z E R , represent
+ +
vectors r = xel ye2 ze3 E R3. The group of unitary and unimodular
matrices
SU(2) = {U E Mat(2,C) I U'U = I, det U = I)
represents the spin group Spin(3) = {u E Ce3 1 uii = 1, uii = 1) or
Spin(3) = {u E ti I uii = 1).
Both these groups are isomorphic with the group of unit quaternions S3= {q E
W I qq = 1). For an element u E Spin(3) the mapping r + urii is a rotation
of R3. Every element of SO(3) can be represented in this way by an element
in Spin(3). In fact, there are two elements u and -u in Spin(3) representing
the same rotation of R3. This can be written as Spin(3)/{f 1) 2 SO(3) and
one can say that Spin(3) is a double covering of SO(3).
220 Spin Groups and Spinor Spaces
17.2 T h e Lipschitz groups a n d t h e s p i n groups
The Lipschitz group also called the Clifford group although invented by
Lipschitz 1880186, could be defined as the subgroup in Ctp,, generated by
invertible vectors x E RPJQ,or equivalently in either of the following ways:
Note the presence of the grade involution s + 0, and/or the restriction to

the even/odd parts ~t;,. For s E rp,,, SB E R. The Lipschitz group has a
normalized subgroup
The group Pin(p, q) has an even subgroup
The spin group Spin(p, q) has a subgroup

Spin+ (p, q) = {s E Spin(p, q) I sB = 1).
Write Spin(n) = Spin(n,O), and note that Spin+(n) = Spin(n). Because
of the algebra isomorphisms CeL, .E CQp we have the group isomorphisms
Spin(p, q) E Spin(q, p). However, in general Pin(p, q) 9 Pin(q, p). In partic-
ular, Pin(1) 2 Z2 x Z2 and P i n ( 0 , l ) 2: Z4.
The groups Pin(p, q), Spin(p, q), Spin+ (p, q) are two-fold covering groups
of O(p, q), SO(p, q), SO+ (p, q). Although SO+(p, q) is connected, its two-fold
cover Spin+ (p, q) need not be connected. However, the groups Spin+(p,q),
+
p q 2 2, are connected with the exception of
which has two components, two branches of a hyperbola. The group
has four components.

> >
The groups Spin(n), n 3, and Spin+(n - 1 , l ) E Spin+ (1, n - I ) , n 4,
are simply connected and therefore universal covering groups of SO(n) and
SO+(n - 1 , l ) 2 S 0 + ( 1 , n - 1). However, the maximal compact subgroup of
S 0 + ( 3 , 3 ) is SO(3) x SO(3) which has a four-fold universal cover Spin(3) x
Spin(3). Consequently, Spin+ (3,3) is not simply connected, but rather doubly
connected, and therefore not a universal cover of SO+ (3,3).
17.3 The two ezponentials of bivectors 221
17.3 T h e t w o e x p o n e n t i d s of bivectors
The Lie algebra of Spin+(p, q) is the space of bivectors r\2Rp14. Fbr two
bivectors A, B E /\2 RP14 the commutator is again a bivector,
This can be seen by considering the reverse of

2 4
A B = (AB), + (AB)2 + (AB), E R $ A R p l q$ ARP~~
for which
(AB)- = (AB)o - (AB)2 + (AB)4
and on the other hand
(AB)' = BA = (-B)(-A) = BA.
The exponentials of bivectors generate the group Spin+(p, q).
In this section we consider two different exponentials of bivectors, the ordi-
nary or Clifford exponential
where B 2 = B B , and the exterior exponential
where B A 2 = B A B. The series of the exterior exponential is finite. The

ordinary exponential is always in the spin group, that is,
2
eB E Spin+ (p, q) for B E A Illplq.
The exterior exponential is in the Lipschitz group, if it is invertible in the

Clifford algebra,
2
eAB E I ? , , for B E AIRpgq such that e A B e A ( - B )# 0.
Note that eAB A e A ( - B ) = 1 , and so the exterior inverse of eAB is e A ( - B ) .

The reverse of s = eAB is s" = e A ( - B ) ,and so the exterior inverse sA(-'1 of s
equals 8. The ordinary inverse of s, in the Clifford algebra C t s , is given by
where sB E R .
The Euclidean spaces R n . The bivector B can be written as a sum
of at most t = [n/2J simple bivectors Bi, Bi A B, = 0, which are mutually

completely orthogonal so that their planes have only one point in common,
B i A B j = BiBj, i # j. This decomposition is unique unless B: = B i .
Notwithstanding, the product
(1 + B1) A (1 + B2) A .. . A (1 +Be) = (1 + B i ) ( l + Bz) . . . (1 + Be)
depends only on B and equals the exterior exponential eAB. The square norm
of eAB is seen to be leABI2= (1 - Bq)(l - B i ) . . . (1 - B:).
The Cayley transform. An antisymmetric n x n-matrix A is sent by the Cayley
transform to the rotation matrix
There corresponds to A a bivector B E Rn such that A(x) = B L x for all

x E IWPlq. If y = UX, then y - Ax = x + A x , or equivalently
Next, compute s A (x + B L x) =sAx + s A ( B L x) for s = eAB. Sum up
+
for k = 0 , 1 , 2 , . . . ,t to obtain s A ( B L x) = s L x. Since s A x s L x = sx, it
follows that s A (x + B L x) = sx. Similarly, s A (Y + y J B) = s A y - s L y =
y A s + y J s = ys. Therefore, the equation (1) is equivalent to s x = y s or
This representation of rotations was first discovered by R. Lipschitz 1880186.

Thus we have the following result: An antisymmetric n x n-matrix A and the
rotation matrix U = (I+A)(I-A)-' E SO(n) correspond, respectively, to the
bivector B E /\'Rn, A(x) = BLx, and its exterior exponential s = eABE I?,+,
which is the unique element of , :'I with scalar part 1, inducing the rotation
U, U(x) = sxs-l. For every rotation U E SO(n), which does not rotate any
plane by a half-turn (all eigenvalues are different from -I), there is a unique
element s E I';, ( s ) ~= 1, such that U(x) = sxs-l.
For an element s E,:?I sS E R , s0 > 0. Therefore Is1 = 6 ,and
S AB
- E Spin(n) for s = e .
Is1
17.3 The two exponentials of bivectors
Every element u E Spin(n), (ujo # 0, can be written in the form
eAB
u=f-
leABI'
which corresponds to the rotation U = (I+A)(I-A)-' E SO(n). This should
be contrasted with the fact that every element in Spin(n) can be written in
the form eBI2, which corresponds to the rotation eA in SO(n).
Lorentz signatures. In the Lorentz signatures the decomposition
still exists and can be used to test invertibility of eAB.The exterior exponential
eAB= (1 + B1)(1+ B2) . . .(1 + Be)
is invertible in the Clifford algebra if B; # 1 for all i. In other words, eABE
I'~-,,,if all B; # 1.
Indefinite metrics. Every isometry U of IWP99, connected to the identity of
SO+(p, q), is an exponential of an antisymmetric transformation A of IWP*g,
U = eA, if and only if
see M. Riesz 1958193 pp. 150-152. In these Euclidean and Lorentz signatures
there is always a bivector B, B L x = A(x) such that U (x) = eBxe-B, see M.
Riesz 1958193 p. 160.
Given a bivector B one can, in general, find other bivectors F such that
eB = -eF and hence eBxe-B = eFxe-F. The only exceptions concern the
following cases:
R I J for all B
R 2 ~ ' and R112 for all B #0 such that B2 >_ 0
R3p1 and ~for all B
' 9 ~ #0 such that B 2 = 0,
see M. Riesz 1958193 p. 172.
TO summarize with special cases: All the elements of the cornpact spin groups
Spin(n) are exponentials of bivectors [when n >
21. Among the other spin
groups the same holds only for Spin+ (n - 1 , l ) 2 Spin+ (1, n - I), n 2 5.
In particular, the two-fold cover Spin+(l, 3) 2 SL(2, C) of the Lorentz group
SO+ (1,3) contains elements which are not exponentials of bivectors: take (yo +
2 1,3
y1)y2 € A R , [ ( y ~ + y ~ ) y ~ ] ~then +
= O- ,~ ( Y O + Y ~ ) Y ~= -1- (YO ~ 1 ) #~ eB
2
for any B E /\2 R1l3. However, all the elements of Spin+(l, 3) are of the
1 In contrast, -e(el+es)e2 = -1-(el+es)ez = e(el+es)ez+re34 in Spin+ ( 4 , l ) 2: Sp(2, 2).
form feB, B E /\'lR1p3. Therefore, the exponentials of bivectors do not form
a group.
Every element L of the Lorentz group SO+ (1,3) is an exponential of an an-
tisymmetric matrix, L = eA, g ~ T S - l= -A; a similar property is not shared
by SO+ (2,2). There are elements in Spin+ (2,2) which cannot be written in
the form feB, B E /\2 R2~'; for instance fe1234ePB, B = e l 2 2e14 e34, + +
P > 0, see M. Riesz 1958193 p. 168-171. '
Lower-dimensional spin groups. The dimension of the Lie group Spin(n) is
$n(n - 1). The groups Spin(n), n <
6, and Spin,. (p, q), p q 5 6, are +
identified in Table 1.
Table 1. Spin Groups Spin+ (p, q), p + q < 6.
Note that Spin+ (p, q) = {s E Ce,: I sS = 1) for p+q 5 5. In dimension 6 the

group {s E I sS = 1) E U(4) has a proper subgroup Spin(6) 2 SU(4).
The groups Spin(7) and Spin(8) are not directly related to classical matrix
groups; their study will be postponed till the discussion on triality.
In the case of the complex quadratic spaces Cn we define the complex pin
group slightly differently:
2 Riesz also showed, by the same construction on pp. 170-171,that there are bivectors which
cannot be written as sums of simple and completely orthogonal bivectors; for instance
+ +
B = e l 2 2e14 esr E A2 R212.
3 The structures of square classes are different for R and C. In R = R \ {0), R X = fR O ,
R 0 = (A2 I X E R X ) ; so to pick up one representative out of each square class we set
sd = f1. In contrast, in C X = C \ {0), C X = CO; so to pick up one representative out of
each square class we set sg = 1.
17.4 Pauli spinors
The complex spin groups Spin(n, C), n 1: 6, are seen to be as follows:
We also define the Lipschitz group $I'q+l,p for paravectors in IR @ IRPpq as

the group containing the products of invertible paravectors, or equivalently,
For any non-null paravector a E R $ IRPpq, the mapping x + a x i - l is a spe-

cial orthogonal transformation of IW $ IWPtq with metric x + xZ. Therefore
$rq+l, N I'itllp. Note that I'p,qc $&+l,p and $r&l,p = The nor-
+
malized subgroup $pin(q 1,p) = {s E $&+lPp I sS = f1) is isomorphic to
+
Spin((? 1, P).
IDEMPOTENTS,
LEFT IDEALSAND SPINORS
Review first the Clifford algebra Ct3 of the Euclidean space IR3.
17.4 P a u l i spinors
In the non-relativistic theory of the electron, spinors are regarded as columns
We shall instead introduce spinors as square matrices
If we multiply $ on the left by an arbitrary element u in Ct3 we obtain another

element u$ = cp in Ct3 whose matrix is also of spinor type:
The spinors make up a left ideal S of Ct3, that is,

for all u E Ct3 and $ E S we also have u$ E S.
The left ideal S contains no left ideals of Ct3 other than the zero ideal (0)
and S itself. Such a left ideal is called minimal in Ct3.
+
The element f = $(1 e3) is an idempotent, that is, f 2 = f , which is
primitive in Ct3, that is, it is not a sum of two annihilating idempotents,
f # f l + f 2 , f l f 2 = f2 fl = 0. The left ideal S = Ct3f can be provided with a
D + S,
right linear structure over the division ring D = fCe3f as follows: S x I
($J,A) + $A. With this right linear structure over ED C the left ideal S
becomes a spinor space.
17.5 Primitive idempotents a n d minimal left ideals

An orthonormal basis of RPlQ induces a basis of Cep,,, called the standard
basis. Take a non-scalar element e ~ e$ , = 1, from the standard basis of Ce,,,.
Set e = i ( l + e ~ )and f = 3 1 - e T ) , then e + f = 1 and ef = f e =O. So
Cep,, decomposes into a sum of two left ideals Ce,,, = Cep,,e @ Ce,,, f , where
dim Ce,,,e = dimCep,, f = iCep,, = 2n-1 . Furthermore, if {eTl, e ~. . .~, e , ~ ~
is a set of non-scalar basis elements such that
e$i = 1 and e ~e ,~=, eTjeT, ,
then letting the signs vary independently in the product i ( l f e T l ) i ( l f
eT2). . . i ( l f eT,.), one obtains 2k idempotents which are mutually annihi-
lating and sum up to 1. The Clifford algebra Ce,,, is thus decomposed into
a direct sum of 2k left ideals, and by construction, each left ideal has dirnen-
sion 2n-k. In this way one obtains a minimal left ideal by forming a maximal
product of non-annihilating and commuting idempotents.
The Radon-Hurwitz number ri for i E 23 is given by
and the recursion formula
For the negative values of i one may observe that r-1 = -1 and r-i =
1 - i + r i - 2 for i > 1.
T h e o r e m . In the standard basis of Cep,, there are always k = q - rq+ non-
scalar elements e ~, e$, , = 1, which commute, eT,eT, = eTjeT,, and generate a
group of order 2k. The product of the corresponding mutually non-annihilating
idempotents,
is primitive in Ce,,,. Thus, the left ideal S = Cep,, f is minimal in Ce,,,. g

Examples. 1. In the case of I R 0 t 7 we have k = 7 - r7 = 4. Therefore the
+ +
idempotent f = $(I + elz4)i ( 1 e235)$(1+ eS4,j)i ( 1 e457) is primitive in
C e o , 21
~ 'Mat(8,R).
17.5 Primitive idempotents and minimal left ideals 227
-
2. In the case of R2t1 we have k = 1 - r-1 = 1 - (r7 - 4) = 2. Therefore the
Ce2,1 2Mat(2,R).
+ + + +
idempotent f = i ( 1 + e l ) $ ( l e23) = a ( l el e 2 3 e123) is primitive in
If e and f are commuting idempotents of a ring R, then ef and e + f - e f

I
are also idempotents of R. The idempotents e f and e + f - e f are a greatest

lower bound and a least upper bound relative to the partial ordering given by
e <f if andonly if ef = fe = e .
-
A set of commuting idempotents induces a lattice of idempotents.
Example. In the Clifford algebra Ce3,i, k = 1 - r-2 = 1 - (rs - 4) = 2. Since
2k = 4, Mat(4,R) and there are 22k = 16 commuting idempotents
in the lattice generated by the following four mutually annihilating primitive
idempotents:
The lattice induced by the primitive idempotents f l , f 2 , f3, f4 looks like a rhom-
bidodecahedron, see diagram. I
17.6 Spinor spaces

For a primitive idempotent f E elp,, the division ring D = f elp,,f is isomor-
phic to
R for p - q = 0 , 1 , 2 m o d 8
C for p - q = 3 m o d 4
W for p - q = 4 , 5 , 6 m o d 8
and the map
S x D + S , ($,X)+$X
defines a right D-linear structure on the minimal left ideal S = Ce,,, f . Pro-
vided with this right D-linear structure the minimal left ideal S becomes a
spinor space.
The spinor space provides an irreducible representation
of elp,,. This representation is also faithful for all simple Clifford algebras
Ce,,,, p - q # 1 mod 4.
Next, we construct a faithful representation for semi-simple Clifford algebras
Cep,,, p - q = 1 mod 4, which are direct sums of two simple ideals 4(l f
el2..,) Cep,,. Take a primitive idempotent f and an idempotent e = f f in +
Ce,,,. The ring IE = eCep,,e is the direct sum IE = D $ lb, lb = {i I X E D),
isomorphic to the double ring of the division ring D, more precisely,
ReIW for p - q = l m o d 8
W$W for p - q = 5 m o d 8 .
To find a faithful representation for a semi-simple Clifford algebra Cep,, with
p - q = 1 mod 4 take a left ideal S $ s where s = {?i, I $ E S ) . The map
defines a right E-linear structure on S $ S. Provided with this right IE-linear

4 Similarly, beginning with a minimal left ideal of the even subalgebra C P $ , ~we obtain an
even spinor space. The dimension of the even spinor space is lower than the dimension of
the spinor space, when p - q = 0 mod 4. In this case, even spinors are called semi-spinors.
17.6 Spinor spaces 229
s
structure the left ideal S $ of Clp,q becomes a double s p i n o r space. The
double spinor space provides a faithful but reducible representation
for a semi-simple Clp,q, p - q = 1 mod 4.

In order to be able to consider faithful representations of simple and semi-
simple Clifford algebras a t the same time, we adopt the following notation:
according as Clp,q is simple or semi-simple, respectively. Thus, the ring fi is

isomorphic to R , @, W,'R or 'W.In this way we have a faithful representation
for all CtpPq.However, this representation is reducible in the cases p - q =

1 mod 4.
Questions
1. Do the exponentials of bivectors form a group?
2. Do the exterior exponentials of bivectors form a group?
+ >
3 . Are Spin+( p , q ) , p q 3 , p , q # 2, universal covers of SO+( p , q ) ?
4 . Are double spinor spaces needed to construct a faithful representation for
ce2,5 ?
Answers
1. No. 2. No. 3 . No. 4. Yes.
Bibliography
P. Budinich, A. Trautman: The Spinorial Chessboard. Springer, Berlin, 1988.
York, 1954.
J.S.R. Chisholm, A.K. Common (eds.): Proceedings of the NATO and SERC Work-
shop on '~liflo'rdAlgebras and thei; ~ p p l i c a t i o k sin Mathematical Physics' (Can-
terbury, 1985). Reidel, Dordrecht, The Netherlands, 1986.
5 Similarly, by doubling of a minimal left ideal of the even subalgebra ~ e : ,we~ obtain a
double even spinor space.
A. Crumeyrolle: Orthogonal and Symplectic Clifford Algebras, Spinor Structures.

Kluwer, Dordrecbt, The Netherlands, 1990.
R. Deheuvels: Formes quadratiques et groupes classiques. Presses Universitaires de
R. Lipschitz: Principes d'un calcul algCbrique qui contient comme esphces particulibres
le calcul des quantitCs imaginaires et des quaternions. C.R. Acad. Sci. Paris 91
(1880), 619-621, 660-664. Reprinted in Bull. Soc. Math. (2) 11 (1887), 115-120.
R. Lipschitz: Untersuchungen uber die Summen uon Quadraten. Max Cohen und
Sohn, Bonn, 1886, pp. 1-147. The first chapter of pp. 5-57 translated into French
by J. Molk: Recherches sur la transformation, par des substitutions rCelles, d'une
somme de deux ou troix carr6s en ellemtme. J. Math. Pures Appl. (4) 2 (1886),
373-439. French rCsumC of all three chapters in Bull. Sci. Math. (2) 10 (1886),
163-183.
R. Lipschitz (signed): Correspondence. Ann. of Math. 69 (1959), 247-251.
P. Lounesto: Cayley transform, outer exponential and spinor norm, in Proc. Winter
School of Geometry and Physics, Srni. Suppl. Rend. Circ. Mat. Palerrno, Ser. I1
(1987), 191-198.
P. Lounesto, G.P. Wene: Idempotent structure of Clifford algebras. Acta Applic.
Math. 9 (1987), 165-173.
M. Riesz: Sur certaines notions fondamentales en thCorie quantiqile relativiste. C . R .
l o e Congrks Math. Scandinaves, Copenhagen, 1946. Jul. Gjellerups Forlag, Copen-
hagen, 1947, pp. 123-148. Collected Papers, pp. 545-570.
as facsimile (eds.: E.F. Bolinder, P. Lounesto) by Kluwer, The Netherlands, 1993.
18
Scalar Products of Spinors and the Chessboard
+
The Euclidean space R3 has a scalar product x . y = x l y l + x2y2 x3y3 with
the automorphism group O(3). Pauli spinors of R3 are of the form
and belong to a complex linear space C2. There are two kinds of scalar products
for Pauli spinors $, cp E C2,
$*T cp = $:(PI + $;cp2 and
which have automorphism groups U(2) and Sp(2, C) = SL(2,C), respectively.

The Minkowski space R1p3 has a scalar product
with the automorphism group 0 ( 1 , 3 ) . Dirac spinors of R1>3belong to a com-

plex linear space C4. There is a scalar product of Dirac spinors $, cp e C4,
with the automorphism group U(2,2).

One might wonder about the following things:
(i) Why do spinors with complex entries arise in conjunction with the real
quadratic spaces R3 and R1t3?
(ii) If we consider generalizations to arbitrary RPJQ,are the scalar products
of spinors still Hermitian or antisymmetric?
(iii) Are the scalar products of spinors definite or neutral for all RP>Q?
(iv) Is there a general pattern in higher dimensions for the changes from R3
to C2 or from R1p3 to C2p2?
232 Scalar Products of Spinors and the Chessboard
We will answer these questions in the following general form: What is the
automorphism group of the scalar product of spinors in the case of the quadratic
space IRPJQ?The scalar products of spinors can be collected into two equivalence
classes when p andq are kept fixed in RPJQ.There are altogether
different kinds of scalar products of spinors when we let p and q vary in WpQ.
The situation is much simplified if we consider instead of the real quadratic
spaces RPJQtheir complexifications C @ WpQ. Then there remain only four
different types of scalar products of spinors to be considered.
The reader will notice that the unitary group U(2,2) can be adjoined to the
Minkowski space-times R1t3 and R3p1 in two different ways by
- complexifying, or
- adding one extra dimension (of positive signature),
which respectively result in
- C@ IR113 and C @ IR3p1, or
- I R 2 p 3 and IR411.
In both cases U(2,2) is the automorphism group of the scalar product of
spinors. The latter case gives a hint of a relation to the conformal group
of the Minkowski space.
18.1 Scalar products on spinor spaces

We start with spinors $, cp in spinor spaces S = Cl!p,qf which are linear spaces
over division rings ID = fCtptqf . We will consider two cases:
(i) The minimal left ideals S = Ctp,qf providing irreducible representations

for all C.lp,q; these representations are also faithful for simple Ctp,,.
+
(ii) The left ideals S $ s = CeP,,e, e = f f , providing faithful represen-
tations for semi-simple Cl!p,q.
1 The Vahlen matrices of the Minkowski space are such that Mat(2,CPlV3)E CP2,4 and
Mat(2, CP3,1) E CP4,2, where the even subalgebras are isomorphic: CP;,, z CP:,
Mat(4,C) or CP2,3 z Ce4,1 z Mat(4,C). The (connected components of the) conformal
groups of R1p3 and R3v1 are isomorphic to
The automorphism group U(2,2) of the scalar product of Dirac spinors contains as a
subgroup the universal cover SU(2,2) of the conformal group of the Minkowski space.
18.1 Scalar products on spinor spaces
As before, let
fi beeither ID or D @ f i ,
S be either S or S @ s
according as CtP,¶ is simple or semi-simple, respectively.
Let /3 be either of the anti-automorphisms u + ii and u + ti of C.lp,,. The
real linear spaces
P+ = {$ E S I P($) = +$),
P- = {$ E S 1 P($) = -$)
have real dimensions 0, 1, 2 or 3 and
-
has real dimension 0, 1, 2 or 4 no matter how large the dimension of S is. To
prove this we may use periodicity, CtP,¶ 8 Ceo,s CtP,¶+s,and the fact that
for Ceo,8 the dimension of P = P+ is 1 (over R).
Define the real linear space
which has real dimension 1, 2, 3 or 4. For all $, cp in S or s we have p($)cp in

P or P . There is an invertible element s in Cep,, with the property P C s-'D
and which is, in the case dim P # 0, such that for all X in D also Xu = s@(X)s-'
is in D. To prove that such an element s exists in every Cep,, we may first
consider the lower-dimensional cases and then proceed by making use of the
fact that P(f) = f for
in Cto,s, and therefore s = 1 is such an element in Ceo,8.

In the same way, there is an invertible element s in Cb,, with the property
P = s-lfi and which is moreover such that for all X in Ib also Xu = sp(X)s-I
is in 6. Both the maps
are scalar products on S. Similarly, we may construct a scalar product on S.

The element s can be chosen from the standard basis of CtP,¶ [when f is
. .-- In particular, P(s) = rts, and so
constructed by the standard basis of Ce,,,].
the scalar product is symmetric or antisymmetric [on both S and s]. The
2 The mapping X -+ X u is an (anti-)auto~norphismof the division ring D.
3 More precise!^, the scalar product p n S is Dm-symmetric or Dm-skew, and the scalar
product on S is IDm-symmetricor Dm-skew.
scalar product on s is more interesting; it is
symmetric or antisymmetric
non-degenerate
Ce,,, with S&
positive definite (for the choice s = 1) on
Ce,,, with s$p
neutral except on
ce,,,, ce,,, , ce,,,, ce,,, with s 4 p
ce,,,, ce,,, with s&.
The scalar product is definite or neutral except for Ceo,~,Clo,2, Ce0,3 or CCI,~.
In these lower-dimensional exceptional cases neutrality is not possible, because
the spinor space S is 1-dimensional over Ib = C, W, or 'R, respectively.
For a fixed C.t2p,q,the neutral scalar products on S, induced by arbitrary
anti-automorphisms of Cep,¶,can be collected into two equivalence classes, the
equivalence relation being
for all $, p S. In each class there is a scalar product induced by such an anti-
automorphism of c.&,¶(extending an orthogonal transformation of that
does not single out any distinguished direction in RplQ,namely, the reversion
u -+ ii or the Clifford-conjugation u + ii of Ctp,¶.
18.2 Automorphism groups of scalar products of spinors

Examples. 1. The Clifford algebra Ce2,1 is isomorphic to Mat(2, 'R). The
+ +
idempotent f = i ( 1 e l ) $ ( l eZ3) is primitive in Ceztl. The subalgebra
D = fCe2,lf is just the line {Xf I X E R); with unity f it is isomorphic to the
division ring R. The basis elements
fl = + +
+ e l e23 e123)
f2 +
= i(e2 - el2 e3 - e13)
of S = C&,lf are such that
The products s&p, s = 1, and s$p, s = e 2 , have values in D; they are scalar
products on S. The scalar product $p vanishes identically; its automorphisrn
group is the full linear group GL(2,R). The scalar product ez$p is antisym-
metric; its automorphirns group is Sp(2,R). If we consider s = S $ ,S instead
of S, then the automorphism group of the scalar product s4p becomes non-
degenerate (because of the swap) and the automorphism group of the scalar
18.2 Automorphism groups of scalar products of spinors 235
product s$p splits: ' S p ( 2 ,R ) = Sp(2,R) x Sp(2,R ) .
2. The Clifford algebra C t l , 3 of the Minkowski space R113 is isomorphic to
the real matrix algebra Mat(2, W). Take an orthonormal basis {yo,yl , y2,y3)
for R1p3. The idempotent f = i(1 + y o ) is primitive in Ce1,3. AS a real linear
space the minimal left ideal S = Cel,3 f is 8-dimensional and the elements
form a basis for SR. The set {hl ,il ,jl , k l ) is a basis for the real linear space
D = f C e l , 3 f . As a ring ID is isomorphic to the quaternion ring W, and the
right D-linear module SD is two-dimensional with basis { h l ,h2). In the basis
{ h l ,h 2 ) left multiplication by yo, yl, yg,y3 is represented by the following 2x2-
matrices with quaternion entries:
The real linear spaces P+ and P- have bases
In the scalar products S x S + D, ($,p) + sp($)cp one can take s = 1 for

s q p and s = 7123 for sqp. Direct computation shows that
- -
klhl = h l , k l h 2 = 0 h l h l = O , hlh2=h2
and
k2hl = 0, k2h2 = -hl L2hl = h 2 , L2h2 = 0.
Both the scalar products have the automorphism group Sp(2,2). I
The Tables 1 and 2 list automorphism groups of the scalar products on S; they
are nothing but the groups
{S I S ; = 1) and { S E C ~1 sS= ~ , 1).
~
If the Clifford algebra CG,, is semi-simple and if the automorphism group on
s is a direct product 2G = G x G , then the automorphism group on S is G.
Table 1. Automorphism Groups of sqcp on s in CG,,.
Table 2. Automorphism Groups of s$cp on s in Cep,,.
0 O(1)
1 U(1) GL(11R)
2 s ~ ( 2 ) s ~ ( 2 1R, s~(21IW)
3 2 s ~ ( 2 ) s ~ ( 2 1 @ )2 s ~ ( 2 1 R ) s~(21@)
4 s ~ ( 4 ) SP(~ 2), s ~ ( 4R), S P ( ~R) , SP(% 2)
5 U(4) GL(2,IHI) U(2,2) GL(4,R) U(2,2) GL(2, W)
6 O(8) SO*(8) SO*(8) 0 ( 4 , 4 ) 0 ( 4 , 4 ) SO*(8) SO*(8)
7 2 0 ( 8 ) O(8, @) 2SO*(8) O(8, @) 20(4,4) O(8, @) 2SO*(8) O(8, @)
Examples. 1. Ceo,2, s&: S0*(2) = {U E S0(2,@) I U * J = J U } 2 SO(2).

2. Ce2, sqcp: Sp(2,R) = {U E Mat(2,R) I U ~ J U= U) rr SL(2,R).
3. ce5, S ~ P 2: ~ p ( 4=
) sp(4) x sp(4), S P ( ~ ) / { ~ I2}1 s o ( 5 ) .
4. Ce1,3, Sp(2,2) = U(2,2) n Sp(4, C), Sp(2,2)l{f 1) 2 SO+(4,1). I
Note that the group U(2,2) appears as an automorphism group of the scalar
product s$cp for Ce2,3 and C.t4,~. To explain the presence of U (2,2) in the
Dirac theory by the real Clifford algebras C$,,, we must add one dimension of
positive square to the Minkowski spaces R113 and EX3?'.
There is another explanation: use complexifications G @ Ctp,,. For a fixed
n = p + q we have the isomorphisms of algebras @ 8 Cep,, 2 Ce(Cn). Although
the complex linear space Cn has a symmetric (= not sesquilinear) bilinear form
on itself, we may equip the spinor spaces of @ 18CtP,, with sesquilinear forms
18.2 Automorphism groups of scalar products of spinors
s$* cp and sG* cp. These sesquilinear products have automorphism groups
{S E @ 8 Cep,, 1 sSf = 1) and { s E @ @ C.tp,, I sS* = 1).
+
For a fixed n = p q these groups depend on the values of p andq [although
the algebra C @ C t p , , is independent of p and q ] .
Table 3. Automorphism Groups of s@cpin C* 8 Cep,q.
Table 4. Automorphism Groups of s$*p in @' 8 Cep,,.
See Porteous 1969 p. 271 11. 1-8. Note that complexification explains the oc-
currence of U(2,2) in conjunction with the Minkowski spaces.
In complexifications of real algebras we replaced the ground field IW by C , a
field extension with an involution, the complex conjugation [to emphasize that
C comes with a complex conjugation we denote @ or C*].
We could also tensor Cep,, by the real algebra 21R, a commutative ring with
an irreducible involution, the swap. See Porteous 1969 pp. 193, 251. This leads
to the automorphism groups shown in Table 5 [isomorphic to the subgroup of
invertible elements in Cl,,,].
Table 5. Automorphism Groups for 'R @ Clp,,.
See Porteous 1969 p. 271 11. 11-18.

In the case of the complex Clifford algebras Cl(Cn) we may further equip
the spinor space with a symmetric (= not sesquilinear) form on itself, sending
($,p) to sqcp or s+t&, see Table 6.
Table 6. Automorphism Groups for Cn
See Porteous 1969 p. 271 1. 9.

As the last extension we consider the tensor product '@ @cP. The scalar
products of spinors are formed by reversion or Clifford-conjugation composed
with swap (no complex conjugation), see Table 7.
18.3 Brauer- Wall-Porteous groups
Table 7. Automorphism Groups for 'C @ C"
See Porteous 1969 p. 271 1. 10.
18.3 Brauer-Wall-Porteous groups

As before, we consider only finite-dimensional associative algebras.
Central simple algebras over R are isomorphic to the real matrix algebras
Mat(d, R) and Mat(d, W). A tensor product of two matrix algebras with entries
in W is a matrix algebra with entries in R.This can be expressed by saying
that the Brauer group Br(R) of R is a two-element group {R, W).
Tensor products of graded central simple algebras over R lead to the Brauer-
Wall group BW(R) of R ; this is a cyclic group of eight elements,
Here we use the abbreviation A(v) = Mat(v, A); the notation
means that B is the even subalgebra of A. The elements of B W ( R ) can be

represented by the graded algebras
where n is taken modulo 8. This is just another way of expressing Cartan's

periodicity of 8.
Graded algebras are algebras with an involution (= involutory automor-
phism). We could further consider tensor products in graded central simple
algebras with an anti-involution (= involutory anti-automorphism). When the
involution and the anti-involution commute, this leads to the Brauer-Wall-
Porteous group BWP(IW) of IW; its elements are graded subgroups (of a graded
algebra A)
G
H
where G is the subgroup determinded by the anti-involution P, G = { s E A I
p(s)s = I ) , and H is its even subgroup, H = B n G ( B is the even part of A ) .
Table 8. Scalar Product sqcpin Clp,, and BWP(IW).

- -
p-q 0 1 2 3 4 5 6 7
- -
O(v,v ) O(v,v ) SO* ( 2 4 SO* (2v)

20(u,v ) 0 ( 2 v ,(C) 2 S 0 *(2v) 0 ( 2 v ,C)
O(v,v ) O(v,v ) SO* (2v) SO*(2v)
GL(2v,R ) U ( v ,v ) GL(v,W) U ( v ,v )
Sp(2v,R ) S P ( ~ R V ), S P ( ~v ), S P ( ~v ),
2s~(2v,IW)S P ( ~ @) V , 2 s ~ ( v ), S P ( ~ C)
V,
s p p , R ) S P ( ~ RY ), S P ( ~v ), S P ( ~v ),
GL (2v,R ) U( v ,v ) GL ( v ,W) U( v ,v )
Table 9. Scalar Product s&,oin ClPtpand B W P ( R ) .
p-q 0 1 2 3 4 5 6 7
O(24 0 O(v,v ) O(v,v ) SO*(2v) S 0 * ( 2 v )

U(2v) 1 GL(2v,IR) U ( v ,v ) GL(v,W) U ( v ,v )
Sp(2v) 2 Sp(2v,R ) Sp(2v,R ) S P ( ~ v ), S P ( ~v ),
2Sp(2v) 3 2sp(2v,IR) spp,(C) 2 s ~ ( v ), S P ( ~ V ,
SP(~U) 4 S P ( ~ IW)
V, SP(~R V ), S P ( ~v ), S P ( ~v ),
U(2v) 5 GL(2v,IR) U ( v ,v ) GL(v,W) U ( v ,v )
o(2v) 6 O(v,v ) O(v,v ) SO"(2v) SO'(2v)
20(2v) 7 20(v,u) 0 ( 2 v ,(C) 2so*(2v) 0 ( 2 v ,C)
The Brauer-Wall-Porteous group BWP(IW) is a commutative group of 32

elements,
18.3 Brauer- Wall-Porteous groups 24 1
We see that the elements of B W P ( R ) are (graded) automorphism groups of
scalar products of spinors for C%,q,
The even subgroup { s E Ce2q I sp(s) = 1) is isomorphic to { s E C%,q-l I

sp(s) = I ) , obtained by taking a step to the North-East. Tensor products
of real graded central simple algebras with an anti-involution correspond to
movements of a bishop on the chessboard.
Recall that the Brauer group B r ( 6 ) of C is a one-element group { C ) . The
Brauer-Wall group B W ( 6 ) of C is a group of two elements
{Ma",y) Z}
' C .
Thus, complex Clifford algebras have a periodicity of 2. The Brauer-Wall-
Porteous group B W P ( 6 ) of C is a cyclic group of eight elements; in other
words complex Clifford algebras with an anti-involution have a periodicity of
8, see Table 10.
Table 10. Ce(Cn)and B W P ( 6 ) .
n s?Zv n s4cp
0 0(2v,q 0 0(2u,6)
1 '0(2u,6) 1 GL(2u,C)
2 0(2u,@) 2 SP(~V,~)
3 GL(2u,C) 3 'Sp(2v,C)
4 SP(~V,~) 4 SP(~V,@)
5 ' s p ( 2 v ,C ) 5 GL(2u,C)
6 S P ( ~ @)
V, 6 0(2u,6)
7 GL(2u,C) 7 '0(2u,@)
The Brauer-Wall-Porteous group B W P ( ' R ) of the double ring ' R with swap
is also a cyclic group of eight elements, see Table 11.
Table 11. ' E X @ Clp,q and BwP('R).

Tensoring Ct,,, by C*, the complex field with complex conjugation, results
in a Brauer-Wall-Porteous group isomorphic to Z2 x Z2, see Table 12.
Table 12. C* @CtP,, and BWP(@').
As our last extension we tensor CC(Cn) by 'C (Table 13).
Table 13. '@@cCn and BWP('C).
n s&o and &p
In total, we have the following Brauer-Wall-Porteous groups (of IW and C

and their extensions with an irreducible involution).
It is convenient to be able to characterize the automorphism groups of scalar

products on spinor spaces S directly by making use of real dimensions of the
subspaces P* = {$ E S I P($) = f$1, see Table 14.
18.3 Brauer- Wall-Porteous groups
Table 14. Scalar products on S.
d i m P-
O GL(v,R) O(v,v) 0(2v,C)

1 S P ( ~ V , R ) U(v, v) SO*(4v)
2 SP(~~,C)
3 S P P ,2v)
Bibliography
I.M. Benn, R.W. Tucker: An Introduction to Spinors and Geometry with Applications
J.S.R. Chisholm, A.K. Common (eds.): Proceedings of the NATO and SERC Work-
shop on 'Clifford Algebras and their Applications in Mathematical Physics' (Cun-
P. Lounesto: Scalar products of spinors and an extension of Brauer-Wall groups.
Found. Phys. 11 (1981), 721-740.
19
Mobius Transformations and Vahlen Matrices
Classical complex analysis can be generalized from the complex plane to higher
dimensions in three different ways: function theory of several complex variables
(commutative), higher-dimensional one-variable hypercomplex analysis (anti-
commutative), and conformal transformations (geometric). In this chapter we
study the third possibility: conformal transformations in n dimensions, n 2 3.
A function f sending a region in IFX2 = @ into C is conformal at z, if it is
complex analytic and has a non-zero derivative, f'(z) # 0 (we consider only
sense-preserving conformal mappings). The only conformal transformations of
the whole plane @ are affine linear transformations: compositions of rotations,
dilations and translations. The Mobius mapping
is affine linear when c = 0; otherwise it is conformal at each z E C except

when z = -$. The Mobius mapping f sends C \ {-$1 onto C \ { z). If
we agree that f(-$) = co and f(co) = f , then f becomes a (one-to-one)
transformation of C U {co), the complex plane compactified by the point at
infinity. These transformations are called Mobius transformations of @u{KI).
Mobius transformations are compositions of rotations, translations, dilations
and transversions. Mobius transformations send circles (and affine lines)
to circles (or affine lines). The derivative of a Mobius transformation is a
composition of a rotation and a dilation.
By definition, a conformal mapping preserves angles between intersecting
curves. Formally, let D be a region in a Euclidean space Rn. A continuously
1 MGbius mappings f ( z ) = are defined almost everywhere in C. The set of MGbius
mappings can be used to compactify C, the compactification being C U {co).
2 A transversion is a composition of an inversion in the unit circle, a translation and another
inversion. Thus, transversions are conjugate (by the inversion) to translations.
19.1 Quaternion representation of conformal transformations of R 4 245
differentiable function g : D + R n is conformal in D if there is a continuous
function X : D + R X = R \ (0) such that
for all x E D and a, b E R n . In higher-dimensional Euclidean spaces R n ,

n 2 3, the only conformal mappings [sending a region in Rn into Rn] are
restrictions of Mobius transformations of Rn U {m). The case n = 3 was
proved by Liouville 1850. The analogous statement for indefinite quadratic
spaces is also true by a theorem of Haantjes 1937.
19.1 Quaternion representation of conformal transformations of R4

Conformal transformations of R4 can be represented by quaternion cornputa-
tion:
R4 = W + W,q + (aq + b)(cq + d)-l, a , b, c, d E W.
In order to exclude constant functions we require the matrix
+ - 2 ~ e ( cad b) # 0. This matrix rep-

to be invertible, that is, ( a ( 2 ( d ( 2 (b121~(2
resentation renders composition of non-linear conformal transformations into
multiplication of matrices.
19.2 Mobius transformations of Rn

Mobius transformations might be sense-preserving with det fl(x) > 0 or sense-
reversing with det f'(x) < 0. The Mobius transformations form a group, the
full Mobius group, which has two components, the identity component being
the sense-preserving Mobius group. The full Mobius group of Rn is generated
by translations, reflections and the inversion
or equivalently, by reflections in affine hyperplanes and inversions in spheres

(not necessarily centered a t the origin). The sense-preserving Mobius group is
generated by the following four types of transformations:
3 We shall often refer to Mobius transformations of the Euclidean space Rn whereby we
tacitly mean transformations of the cornpactification Rn U (co).
246 Mobius Transformations and Vahlen Matrices
rotations axa-' a E Spin(n)
translations x+b beRn
dilations XS 6>0
transversions
x x2c + c ERn.
1+2x.c+x2c2
Rewriting the transversion into the form (x-' +
c)-l, one sees that it is a
composition of the inversion, a translation and the inversion. Using the multi-
plicative notation of the Clifford algebra CC,, the transversion can further be
written in the form
+
This might suggest the following: Let a , b, c, d be in Cen . If (ax x) (cx d)-' +
is in R n for almost all x E Rn and if the range of
g(x) = (ax + b)(cx + d)-'
is dense in R n , then g is a Mobius transformation of Rn. Although this is
true, the group so obtained is too large to be a practical covering group of the
full Mobius group. Therefore, we introduce:
Definition (Maass 1949, Ahlfors 1984). The matrix ( 1) E ~ a t ( 2ten)

,
fulfilling the conditions
(i) a , b, c, d E I', U (0)
(ii) a&,&dldc', c'a E Rn
(iii) a 2 - bc' E R \ (0)
is called a Vahlen matrix of the Mobius transformation g of Rn given by
g(x) = ( a x + b)(cx+d)-l. I
By condition (i)
. the diagonal entries of a Vahlen matrix are either even or
,
odd. Conditions (i) and (ii) imply that if the diagonal entries are even then
the off-diagonal entries must be odd, and if the diagonal entries are odd then
4 This group is the Vahlen group multiplied by the group generated by invertible matrices
of the form
5 Vahlen 1902 originally wrote the second condition in the form
(ii) 86, 6 4 &, CZ E Rn,

which gives an equivalent characterization of the Vahlen group.
19.3 Opposite of a Euclidean space 247
the off-diagonal entries must be even. Condition (iii) tells us that the pseudo-
determinant ad- bE is real and non-zero, in particular, that the Vahlen matrix
is invertible.
The Vahlen matrices form a group under matrix multiplication, the Vahlen
group. The Vahlen group has a normalized subgroup where condition (iii) is
replaced by
-
(iii') ad - bE = f1.
The normalized Vahlen group is a four-fold, or rather double two-fold, covering
group of the full Mobius group of R n ; the identity Mobius transformation is
represented by the following four matrices:
The sense-preserving Mijbius group has a non-trivial two-fold covering group

formed by normalized Vahlen matrices with even diagonal (and odd off-diagonal)
and pseudo-determinant equal to 1. The full Mobius group has a non-trivial
two-fold covering group with two components, the non-identity component con-
sisting of normalized Vahlen matrices with odd diagonal (and even off-diagonal)
and pseudo-determinant equal to -1.
19.3 Opposite of a Euclidean space

Consider the (n- 1)-dimensional real quadratic space R0pn-' having a negative
definite quadratic form
The sums of scalars and vectors are called paravectors. Paravectors span the
linear space R $ ROJ"-~,which we denote by
$Rn =R$IRO~"-~.
The linear space of paravectors, $Rn, can be made isometric to the Euclidean
+
space R n by introducing for x = xo x E R $ R O ~ n - lwhere
, xo E R and
x E Roln-l, a quadratic form
As an extension of the Lipschitz group Porteous 1969 pp. 254-259 introduced

the group of products of invertible paravectors, defined equivalently by
For a non-zero paravector a E $Rn the mapping a: + axii-' is a rotation of

248 Mobius Transforntatzons and Vahlen Matrices
$Rn. Thus, we have a group isomorphism $rn c I'i.Note that I'o,n-lC $Tn
and I'tn-l= $r:.
Vahlen originally considered the sense-preserving Mobius group of the para-
vector space $Rn.
Definition (Vahlen 1902). The matrix ('I i) E Mat(2, Cla,c-l) fulfilling

the conditions
(i) a , b, c,d E $rn U (0)
- bd, dc, cti E $Rn
(ii) lib,
(iii) ad - b t = 1
is a Vahlen matrix, with pseudo-determinant or norm 1, of the sense-preserving
+
Mobius transformation g of $Rn given by g(x) = (ax b)(cx + d)-'. I
These Vahlen matrices with norm 1 form a group, which is a non-trivial two-
fold cover of the sense-preserving Mobius group of $Rn.
19.4 Indefinite quadratic spaces

The full Mobius group of IWppQ contains two components (if either p or q is
even) or four components (if both p and q are odd).
The identity component of the Mobius group of RPlq is generated by rota-
tions, translations, dilations and transversions which are represented, respec-
tively, as follows:
On the right we have the Vahlen matrices of the respective Mobius transfor-
mations.
6 Maass 1949 and Ahlfors 1984 presented an equivalent characterization of Vahlen matrices
where the second condition was replaced by
(ii) a6, i d , dE, Ea E $Rn.

19.4 Indefinite quadratic spaces 249
T h e o r e m (J. Maks 1989). Consider four Vahlen matrices which represent a
rotation, a translation, a dilation and a transversion. A product of these four
matrices, in any order, always has an invertible entry in its diagonal (there are
4! = 24 such products).
Proof. To complete the proof of the fact that a product of a rotation, a transla-
tion, a dilation and a transversion, in any order, is such that its Vahlen matrix
always has an invertible entry in its diagonal, one can (or rather must) check
the claim for all the 24 orderings. For instance, in the product
the lower right-hand diagonal element a/& is invertible. We leave the verifi-
cation of the remaining 23 orderings to the reader. I
C o u n t e r - e x a m p l e (Maks 1989). In the general case (p # 0, q # 0) J . Maks
1989 p. 41 gave an example of a Vahlen matrix where none of the entries is
invertible (and all are non-zero).
Consider the Minkowski space-time R3s1 and its Clifford algebra C&,l z
Mat(4,R) generated by e l , ea, e3, e4 satisfying e? = ei = e$ = 1, ei = -1.
Take a Vahlen matrix
By the theorem of Maks the matrix M cannot be a product of just one rotation,
one translation, one dilation and one transversion (in any order). However, the
matrix M is in the identity component of the normalized Vahlen group, the
four-fold covering group of the Mobius group of the Minkowski space-time. This
can be concluded while M has pseudo-determinant equal to 1 and even diago-
nal. This can also be deduced by factoring M into a product of a transversion,
a translation and a transversion as follows:
Topologically, we can see this by connecting M to the identity matrix by the

following path (here /? grows from 0 to ~ 1 4 ) :
Maks' counter-example proves that condition (i) has to be modified in the

250 MGbius Tmnsfomzations and Vahlen Matrices
definition of a Vahlen matrix. I
Recall that the Lipschitz group rp,,consists of products of non-isotropic vec-
tors of RPpQ.In the sequel we need the set nP,,of products of vectors, possibly
isotropic, of Rplq. The set nP,,is the closure of rP,,.
Definition (Fillmore & Springer 1990). The matrix ( i)E Mat(2,Ce~,q)

fulfilling the conditions
(i) a , b, c, d E np,,
(ii) iib, bd", dc, ca E RPlQ
(iii) ad" - bZ E R \ (0)
is a Vahlen matrix of the Mobius transformation g of RPlq given by g(x) =
+ +
(ax b)(cx d)-l. I
The Vahlen matrices form a group under matrix multiplication, the Vahlen
group. The normalized Vahlen matrices, with pseudo-determinant satisfying
ad" - bZ = f1, form a four-fold, possibly trivial, covering group of the full
Mobius group of Rplq. When both p and q are odd, the normalized Vahlen
group is a non-trivial four-fold covering group of the full Mijbius group of RPlq.
When either p or q is even, we may find a non-trivial two-fold covering group
of the full Mobius group of RP9Q. It consists of the identity component of the
normalized Vahlen group, that is, normalized Vahlen matrices with even diago-
nal and pseudo-determinant equal to 1, and another component not containing
the (non-trivial) pre-images of the identity:
The identity component of the normalized Vahlen group is a two-fold (either

p or q is even) or four-fold (both p and q are odd) covering group of the
sense-preserving Mobius group.
Conditions (i), (iii) and iib, bd, dc, ca E RPlq imply a i , id, dc', Za E RPlQ. In
contrast to the Euclidean case, conditions (i), (iii) and a&,i d , dZ, Za E RPlQ do
-
not imply iib, bd, dc, cii E RPpq.
Counter-example (Cnops 1996). Consider the Minkowski space-time R3p1
and its Clifford algebra Ce3,1 Mat(4, R ) generated by el, en, e3, e4 satisfying
.2- - e2 2 --1 , ez = -1. The Vahlen matrix
2 -- e3
7 The set lTp,q C Ct,,, rr I\ Rn, considered as a subset of the exterior algebra I\ Rn, is
+
independent of p, q for a fixed n = p q.
19.5 Indefinite paravectors 25 1
ad" - bZ = 1 and a&,id, dc', Za = 0 E R3p1, but
satisfies a , b, c, d E 113,1,
+
even then lib, bd, dc, cli $ R3p1. The mapping gc(x) = (ax b)(cx d)-' is +
conformal. If the matrix C is multiplied on either side by
then B = C D = DC is such that g ~ ( x = ) gc(x) for almost all x E R3s1.

Furthermore, B does satisfy ab, bd, dc, cG E R3p1.
The matrices satisfying a , b,c, d E n3,~,
ai-bc' = 1 and a&,id, dc', Za E R3s1
do not form a group, but only a set which is not closed under multiplication.
This set generates a group which is the Vahlen group with norm 1 multiplied
by the group consisting of the matrices
(cos cp + ;234~inc
cos cp - el234 sin cp
All these matrices are pre-images of the identity Mobius transformation. I
19.5 Indefinite paravectors

Let $nq+1,p be the set of products of paravectors in $RQ+l@= R @ RPpQ.
Definition. The matrix ( i) E Mat(2, fulfilling the conditions
(i) a , b , c , d E $&+I,,
(ii) iib,- bd, dc, cli E $RQtlJ'
(iii) ad - bZ = 1
is a Vahlen matrix with norm 1 of the sense-preserving Mobius transformation
g of $RQ+lJ' given by g(x) = (ax + b)(cx d)-l. + I
The Vahlen matrices with norm 1 form a two-fold or four-fold covering group
of the sense-preserving Mobius group of $RQt1?P.Conditions (i), (ii), (iii) imply
a i l i d , dZ, Za E $RQtlJ' [although (i), (iii) and a&,&dldc', c'a E $RQtlJ' do not
imply Scb, bd, dc, c7i E $RQ+'J'].
19.6 T h e derivative of a Mobius transformation

The difference of the Mobius transformations of x, y in $RQ+'~Pis given by
252 Mobius Tmnsformations and Vahlen Matrices
Letting x approach y we may compute the derivative of a Mijbius transforma-
+
tion. Denoting z = cx d and using N(z) = zE E R , we see that in the case
N(z) # 0 the derivative of x + g(x) is the composition of the rotation
x + ixz-l
and the dilation
19.7 The Lie algebra of the Vahlen group
If the matrix (t ): E Mat(2,Ctn) is in the Lie algebra of the Vahlen

+
group of Rn, then Ax B - x C x - x D E Rn for all x E Rn. It follows
that B , C E Rn and A, D E R $ R2 $ An Rn so that (A)2 = (D)2 and
( A ) , = f(D),. Actually, for the Lie algebra of the Vahlen group (A),, (D),
vanish and for the Lie algebra of the normalized Vahlen group (A)o = -(D)o.
In fact, matrices in the Lie algebra of the normalized Vahlen group can be
characterized by
(i) A , D E R $ A ~ R ~
(ii) B, C E Rn
(iii) A + f i = O .
The Lie algebra is spanned by the matrices
These matrices represent rotations, translations, dilations and transversions.
19.8 Compactification and the isotropic cone at infinity

The set of Mijbius mappings on RPlQ can be used to compactify RPpQ. The
compactification is homeomorphic to
s p X SQ
z2
In particular, the compactification of a Euclidean space Rn is the sphere Sn,
and the compactification of the hyperbolic plane R1ll is the torus S1x S1. The
conformal compactification adjoins an isotropic cone at infinity to the quadratic
19.8 Compactification and the isotropic cone at infinity 253
Questions
1. Is an element in the identity component of the conformal group necessarily
a product of a rotation, a translation, a dilation and a transversion?
2. The group SU(2,2) is a covering group of the identity component of the
conformal group of IW1p3.IS it a two-fold or a four-fold covering group?
Answers
1 . No, as the counter-example of Maks shows.
2. As the identity component of the normalized Vahlen group it is a four-fold
covering group.
Exercises
1. The counter-example M of Maks can be factored into a product of two
'diversions':
Show that a 'diversion' is a product of just one transversion, one dilation,

one translation and one rotation.
2. Show that in the case of a Euclidean space IWn the conditions
a6, i d , dZ, Ea E Rn and lib, bd, dc, cti E Rn are equivalent.
3. Show that the conformal compactification of the Minkowksi space IW1l3is
homeomorphic to U(2).
Solutions
1. The first factor is a product of just one transversion, one dilation and one
translation as follows:
One can insert the identity rotation as the last factor.

2. For a E I',, li = 6. If iib E R n , then we have two cases to consider: either
a is zero, and so a&is a vector, or a is in the Lipschitz group I',, but then
a(6b)c is a vector and a6 E R \ {0), and so b6 is a vector, which implies
6b = (b6)- E IWn.
3. This follows as a special case from the matrix isomorphism
254 Mobius Tmnsforntations and Vahlen Matrices
Bibliography
L.V. Ahlfors: Old and new in Mobius groups. Ann. Acad. Sci. Fenn. Ser. A I Math.
9 (1984), 93-105.
L.V. Ahlfors: Mobius transformations and Clifford numbers; pp. 65-73 in I. Chavel,
H.M. Farkas (eds.): Differential Geometry and Complex Analysis. Springer, Berlin,
1985.
L.V. Ahlfors: Mobius transformations in Rn expressed through 2 x 2 matrices of
Clifford numbers. Complex Variables Theory Appl. 5 (1986), 215-224.
P. Anglks: Construction de revctements du groupe conforme d'un espace vectoriel
muni d'une 'm6trique' de type (p,q). Ann. Inst. H. Poincare' Sect. A33 (1985),
33-51.
P. Anglks: Real conformal spin structures on manifolds. Studia Scientiarum Mathe-
maticarum Hungarica 23 (1988), 11-139.
J. Cnops: Vahlen matrices for non-definite metrics; pp. 155-164 in R. Ablamowicz, P.
Lounesto, J.M. Parra (eds.): Clifford Algebras with Numeric and Symbolic Compu-
tations, Birkhauser, Boston, MA, 1996.
Kluwer, Dordrecht, The Netherlands, 1990. [Conformal group on pages 149-164.1
R. Deheuvels: Groupes conformes et algkbres de Clifford. Rend. Sem. Mat. Univ.
Politec. Torino 43 (1985), 205-226.
J. Elstrodt, F. Grunewald, J. Mennicke: Vahlen's groups of Clifford matrices and
spin-groups. Math. Z. 196 (1987), 369-390.
J.P. Fillmore, A. Springer: Mobius groups over general fields using Clifford algebras
associated with spheres. Internat. J. Theoret. Phys. 29 (1990), 225-246.
J. Gilbert, M. Murray: Clifford Algebras and Dirac Operators in Harmonic Analy-
sis. Cambridge Studies in Advanced Mathematics, Cambridge University Press,
Cambridge, 1991. [Conformal transformations on pp. 37-38, 278-296.1
J. Haantjes: Conformal representations of an n-dimensional euclidean space with a
non-definite fundamental form on itself. Nederl. Akad. Wetensch. Proc. 40 (1937),
700-705.
J. Liouville: Extension au cas de trois dimensions de la question du trace gkographique;
pp. 609-616 in G. Monge (ed.): Application de l'analyse d la ge'ometrie, Paris, 1850.
P. Lounesto, A. Springer: Mobius transformations and Clifford algebras of Euclidean
and anti-Euclidean spaces; pp. 79-90 in J. Lawrynowicz (ed.): Deformations of
Mathematical Structures. Kluwer, Dordrecht, The Netherlands, 1989.
H. Maass: Automorphe Funktionen von mehreren Verhderlichen und Dirichletsche
Reihen. Abh. Math. Sem. Univ. Hamburg 16 (1949), 72-100.
J. Maks: Modulo (1,1) periodicity of Clifford algebras and the generalized (anti-)Mobius
transformations. Thesis, Technische Universitei t Delft, 1989.
J. Maks: Clifford algebras and Mobius transformations; pp. 57-63 in A. Micali et al.
(eds.): Clifford Algebras and their Applications in Mathematical Physics. Kluwer,
Dordrecht, The Netherlands, 1992.
J. Ryan: Conformal Clifford manifolds arising in Clifford analysis. Proc. Roy. Irish
Acad. Sect. A85 (1985), 1-23.
J. Ryan: Clifford matrices, Cauchy-Kowalewski extension and analytic functionals.
Proc. Centre Math. Anal. Austral. Nat. Univ. 16 (1988), 284-299.
K. Th. Vahlen: ~ b e Bewegungen
r und complexe Zahlen. Math. A n n . 55 (1902),
585-593.
20
Hypercomplex Analysis
Complex analysis has applications in the theory of heat, fluid dynamics and
electrostatics. Such versatility gives occasion to explore whether function the-
ory of complex variables can be generalized from the plane to higher dimen-
sions. Are there hypercomplex number systems which could provide a higher-
dimensional analog for complex analytic functions?
Function theory can be generalized to higher dimensions in several different
ways, for instance, to quasiconformal mappings, several complex variables or
to hypercomplex analysis. Clearly, these generalizations cannot maintain all
the features of complex analysis.
In the theory of quasiconformal mappings one retains some geometric fea-
tures, related to similar appearance of images, and renounces some algebraic
features, like multiplication of complex numbers. In the theory of quasiconfor-
ma1 mappings one does not multiply vectors in Rn.
The starting point of hypercomplex analysis is the introduction of a suitable
multiplication of vectors in Rn.In contrast to the theory of several complex
variables, which commute, hypercomplex analysis is a one-variable theory - the
argument being in higher dimensions, where orthogonal vectors anticommute.
20.1 Formulation of complex analysis i n C l z

+ +
For a complex valued function u iv = f (x iy) of the complex variable x + iy
the Cauchy-Riemann equations are
256 Hypercomplex Analysis
The second equation tells us that the vector (u, -v) is the gradient of a function
q5:R2+R:
Using the first relation of the Cauchy-Riemann equations we obtain
that is, 4 is a harmonic function, V2q5 = 0. Conversely, if q5 is harmonic, then

u and v defined by the relation (1) satisfy the Cauchy-Riemann equations.
The Cauchy-Riemann equations can be condensed into a single equation as
follows:
a
- + 2-
(ax aBy ) ( u + iv) = 0.
Recall that i = ele2 and multiply this equation on the left and on the right
by el, then use associativity and anticommutativity to get
As we know, this relation holds if and only if the vector (u, -v) is the gradient
of a harmonic function. It follows that
if and only if (u, v) is the gradient of a harmonic function.

There are three possible ways to formulate the Cauchy-Riemann equations
employing the Clifford algebra CL2 (these possibilities will be generalized to
higher dimensions in three different ways).
1) Firstly, we may consider the Cauchy-Riemann equations to be a condition

+ +
on vector fields, sending a vector xel ye2 in R2 to a vector uel w e 2 in R2.
The above condition (2),
gives us Cauchy-Riemann equations up to sign and results in those conformal

maps which reverse the orientation of R2. In higher dimensions this alternative
means the study of those vector fields, that is, mappings from Rn to R n , which
are gradients of harmonic functions, mappings from Rn to R.
2) Secondly, we may reformulate the Cauchy-Riemann equations as a condition

20.1 Formulation of complex analysis in Ce2 257
+
on the even fields sending a vector xel ye2 in R2 t80an even element in
c;! +
= {U vele2 I u, v E R) 2: C of the Clifford algebra Ce2. The condition
(ii)
a
(el ;j; + -)aya (u + uelz) = 0
e2
gives us the Cauchy-Riemann equations. This alternative has non-trivial gen-

eralizations in higher dimensions sending the vector space Rn to the even
subalgebra :C
! of the Clifford algebra C.,!
3) Thirdly, we may focus our attention on spinor fields sending the vector
plane R 2 to a minimal left ideal of CL2. Before studying this alternative closer,
let us recall that the Clifford algebra Ce2 is isomorphic to the matrix algebra of
real 2 x 2-matrices Mat(2, R). The isomorphism is seen by the correspondences
In this case one sends a vector xel + ye2 in R 2 to a spinor
where
1 1
fl = p(l + e l ) 2: and f2= -(e2-e12)
2
N-
Here ff = f l , SO fl is an idempotent, and the spinor space S = Ct2fl = {a fl (

a E CL2) is a left ideal of CC2, for which a11, E S for all a E Ce2 and 11, E S.
Since
elf1 = fi, elf2 = -f2
e2f1 = f2, e2f2 = fi
one verifies that
(iii)
is equivalent to (i)
To summarize, there are three alternatives for the 2-dimensional target:

(i) the Euclidean vector space itself R2,
(ii) the even subalgebra Ce; of the Clifford algebra CL2,
(iii) the spinor space S = Cezfi.
In the next section, we will generalize, as a preliminary construction, the first
alternative to higher dimensions.
20.2 Vector fields

T h e Dirac operator. It is possible to extract a certain kind of square root
of the n-dimensional Laplace-operator
and consider instead a first-order differential operator
called the Dirac operator. Since the Dirac operator applied twice equals the
Laplace operator, the elements e l , e 2 , . . . , en are subject to the relations
+ +
The linear combinations x = xlel x2e2 + . . . xnen can be considered as
vectors building up an n-dimensional vector space Rn with quadratic form
+ + +
x 2 = x: xi . . . x;. The above relations generate an associative algebra of
dimension 2", the Clifford algebra Cl, of Rn [or of dimension $2n, isomorphic
to an ideal t(l f el2...,)Ctn of CL,].
Operating on a vector field f with V gives
where V .f is the divergence of f and V A f is the curl, which in this approach

is bivector valued.
Sourceless a n d irrotational vector fields. Consider a steady motion of
incompressible fluid in an n-dimensional Euclidean space Rn. Represent the
velocity of the flow by the vector field f . The integral
= l d ~ A f
over an orientable hypersurface S, d i m s = n - 1, is the stream across S.

We regard dS as a tangent (n - 1)-vector measure, rather than the normal
vector measure; this makes the stream n-vector valued. If a vector field f is
20.2 Vector fields 259
sourceless, V . f = 0, its stream across S depends in a contractible domain
only on the boundary dS of S. In particular, no stream emerges through a
closed hypersurface. If n 2 2, a sourceless vector field f has a bivector valued
potential v such that f = V J V. If n 2 3, the bivector potential can be
subjected to a supplementary condition V A v = 0, in which case f = Vv.
The circulation of the vector field f around a closed path C is given by the
line integral
If a vector field f is irrotational, VAf = 0, the circulation vanishes in a simply

connected domain, and the line integral
u(x)=-
l: f.dx
is independent of path. The function u is called the scalar potential of f . The

irrotational vector field f is the gradient of its scalar potential u , f = -Vu.
If a vector field f is sourceless and irrotational, that is V.f = 0 and VAf = 0,
its scalar potential is harmonic, v 2 u = 0. A vector field f is called monogenic,
+
if Vf = 0. For a monogenic vector field Vf = V . f V A f = 0, and so it is
sourceless and irrotational. A monogenic vector field has a potential which is a
+
sum or complex of the scalar and bivector potentials: w = u v . The complex
of potentials is also monogenic, Vw = 0.
Example. A monogenic vector field f , homogeneous of degree l , has a scalar
potential
and a bivector potential
In the singular case, a monogenic vector field f , homogeneous of degree e =

-(n - l ) , might still have a bivector/complex potential. For instance, the
Cauchy kernel q(x) = x / r n , r = 1x1, has a complex potential
- l o g r + i 8 for n = 2 ,
1 8
-(1 + i t a n - ) for n = 3 ,
r 2
where 8 is the angle between x and a fixed direction a , and i is the imaginary
unit of the plane x A a, given by i = x A a/lx a!. I
The plane case. In the plane the stream is a bivector valued line integral.
If the plane vector field is sourceless, its stream across any line in a simply
connected domain depends only on the two end points of the line. Integrating
from a fixed point xo to a variable point x the stream becomes a bivector
valued function, called the stream function,
which can serve as the bivector potential of f = VJ $ [v = $1.

I f f is monogenic, Vf = 0, then there is an even valued function
which serves as the complex potential of f . This complex potential is also

monogenic, Vw = 0. (In higher dimensions there is no correspondence for such
line integrals representing complex potentials - unless one is confined to axially
symmetric vector fields.) I
E v e n fields. Instead of vector fields, we could instead examine the even fields
Here we replace the target R n of dimension n by a wider target Ce: of

dimension 32n. The even subalgebra Cl: is a direct sum of the k-vector
spaces l\kR n with even k. If we require the even functions to be monogenic,
V f (x) = 0, then we have a system of coupled equations:
where fk is the homogeneous part of degree k of f = f (x). These equations

are invariant under the rotation group SO(n).
Irreducible fields. Instead of vector fields or even fields, we could examine
functions with values in an irreducible representation of SO(n) or Spin(n). In
the Clifford algebra realm this would mean studying k-vector fields or spinor
fields. Important physical fields fall into this category: the Maxwell equations
are of the form V A F = 0, VJ F = J , where F E l\2R3~1,and the Dirac field
has as its target the spinor space, a minimal left ideal of C 8 Ce3,1. It should
be emphasized though that in modern treatment of the Dirac theory the spinor
space is replaced by the even subalgebra ~1:~. I
20.3 Tangential integration 261
20.3 Tangential integration
Here we consider integration over surfaces, that is, smooth manifolds embedded
in a linear space R n . Our surface S is compact, connected, orientable and
contractible. The surface S is k-dimensional if there are k linearly independent
vectors tangent to S at each point x of S. The tangent vectors span a tangent
space T,.
In a Euclidean space R n the tangent space T, generates a tangent algebra
CL(T,) isomorphic to Cek. A multivector field on S is a smooth function f :
S 4 Ce,; it is tangential if f ( x ) 6 Ce(Tx) for each x 6 S.
There are exactly two continuous tangential unit k-vector fields on an ori-
entable k-dimensional surface S, each corresponding to one of the two orienta-
tions attached to S. So tangent to each point x of an oriented k-dimensional
surface S there is a unique unit k-vector ~ ( x characterizing
) the orientation
of S at x E S. The value of the map T at x is called the tangent of S at x.
Consider a multivector field f : S + Cen on a k -dimensional surface S C R n ,
1 5 k 5 n. Define the tangential integral of f over S by
where dV is the usual scalar measure of the k-dimensional volume element of

S and dS is the k-vector valued tangential measure,
So the tangential integral of f ( x ) is equivalent to the usual (Riemann) integral

of T ( x )f ( x ) over S.
Since multiplication of multivectors is not commutative, the above equation
is not the most general form for a tangential integral. The appropriate gener-
alization is the following:
Consider an oriented k-dimensional surface S with boundary d S of dimen-

sion k - 1. Set at the point x E d S a tangent ras(x) of the boundary d S and
a tangent rS( x ) of the surface S. Then the expression ( r ~ ( x ) ) - ~ r a s (isx a)
vector normal to the boundary dS. There are two alternatives: the normal vec-
tor points inwards, in which case it is tangent to the surface S, or outwards, in
which case it is opposite to the inward tangent of S. The orientations of S and
d S are compatible, when the vector ( r s ( x ) ) - ' r a s ( x ) points outwards. Define
the normal integral of f over the boundary d S by
where ds is the (k - 1)-vector valued tangential measure on the boundary dS,

and a s is an outward pointing vector normal to dS,
ds = (rs(x))-l ds for x E dS.
T h e D i r a c o p e r a t o r without coordinates. Take a k-dimensional surface
S which is contractible to a point x E S in such a way that the tangent of S
at x remains a fixed k-vector r. Define a differential operator
V, f (x) = lim
where d(S) is the diameter of S and vol(S) is the scalar volume of S.

For instance, when r is a 1-vector, the partial derivative 8, in the direction
of r could be expressed as 8, = rV,.
The case when r is an oriented volume element is important. The same result
is obtained for +el2..., and -elz...,. So it is convenient to drop the subscript
and write V. The differential operator V is called the Dirac operator. Applying
an orthonormal basis {el,ez, . . . ,en) of Rn the Dirac operator is seen to be
d
where - = d,, .
dxi
The relation of V, to the Dirac operator V is obtained by computing
where
v, = r - l ( 7 L V).
20.4 Stokes' theorem

Consider a compact, contractible and oriented surface S c Rn with boundary
d S and a real differentiable function f : S + CL,. Stokes' theorem relates
tangential integrals over S and as, with compatible orientations,
20.4 Stokes' theorem
where the left hand side becomes, using V, = r - l ( r L V),
Here dS and ds are of dimension degree k and k - 1, respectively.

Examples. 1. Consider a 2-dimensional surface S in R3, and a vector field f
on S. Then the scalar part of Stokes' theorem says
Use a vector measure dA = eZ3ds, normal to the surface S, to write the left
hand side as
then use the interchange rule (of dot and cross) to get the usual Stokes' theorem
where ds and dA form a right-hand system.

2. Consider a 2-dimensional surface S E Rn,with bounding line C = dS, and
a circulation of vector field f around C. First, convert the line integral to a
surface integral by Stokes' theorem,
and then compare the homogeneous components of degree 0 to obtain
This shows that in a simply connected domain the circulation vanishes if the
divergence vanishes. I
By convention V differentiates only quantities to its right, unless otherwise
indicated. Because of non-commutativity of multiplication, it is good to have
a notation indicating differentiation both to the right and to the left, when
desired. Accordingly, we have, for instance, the Leibniz rule,
where the dots indicate where the differentiation is applied. Stokes' theorem is
now generalized to the form
Here d i m s = k E {1,2, ...,n). The minus sign on the last term comes from
a ~ b -=( - l ) k b ~ a for a € R n and b E A ~ I W " .
20.5 Positive a n d negative definite metrics

The monogenic homogeneous polynomials play a part in hypercomplex analysis
similar to that of powers of a complex variable in the classical function theory.
In constructing an explicit basis for the function space of monogenic polynomi-
als it is customary to single out a special direction, say en, in an orthonormal
basis {el,e2, . . . , en- 1 , en) of the Euclidean space Rn. The unit bivectors
generate the even subalgebra Ce: ; they anticommute and square up to -1.
Thus, they form an orthonormal basis {il, i2, . . . , in-1) of a negative definite
quadratic space RO~"-lgenerating a Clifford algebra C ~ O , , - ~ Ce;. A closer
contact with the classical function theory is obtained if a vector
in the Euclidean space Rn is replaced by a sum of a vector and a scalar, a

paravec tor,
2 + + . . . + xn-lin-l + Y,
= ~ l i l x2i2 Y = Xnl
in IWO~"-l @ R [the special direction is the scalarlreal part y, also denoted
by xo = x,]. By the above correspondence 2 +) e n x , we have established a
correspondence between the following two mappings:
R~ +ce;, x+~(x),
R C B R O >+~ ce0,,-,,
-~ +f ( ~ ) ;
both are denoted for convenience by f .
In the case of a Euclidean space R n , the Dirac operator is homogeneous,
d
V =el-+e2-+
a ...+ en-1-
a +en-,
d
dxl 8x2 ax,-1 ax,
but it is replaced by a differential operator (inhomogeneous in the dimension
degrees)
d
D = -+ V , in the paravector space R $ ROJ~-'.
8x0
20.6 Cauchy's integral formula 265
20.6 Cauchy's integral formula
Consider a region S c Rn of dimension n with boundary i3S of dimension
n - 1 and multivector function f : S 4 Ce, . In this case Stokes' theorem is of
the form
JTsdsvf = l s d s f .
If a multivector function f is monogenic, V f = 0, then
ds f = 0 or equivalently
which means that the 'stream' of a monogenic function across any closed hy-
persurface vanishes. This is Cauchy's theorem.
In the following we need the Cauchy kernel
which is both left monogenic, Vq = 0, and right monogenic, 4~= 0, at x # 0.

Substitute the Cauchy kernel g(x) = q(x - a) and a left monogenic function
f (x), V f (x) = 0, into Stokes' theorem. On the right hand side the first term
vanishes, and the second term can be evaluated by a limiting process. One
obtains Cauchy's integral formula
where w, = nnI2/(n/2)! and the sign in -(-l)ne12...n comes from the choice
of orientation TS = e l 2 ..., for S.
Example. In the special case n = 2 the above formula is
ks - ds f (x) = -e122rf (a),

since according to our convention e l 2 is compatible with the clockwise orienta-
tion. The classical formula corresponds to both the special cases f : R 2 + R 2
and f : R2 + Ce:. [A better matching with the classical case would be ob-
+
tained by mappings f (a: + yel) = u v e l , where e: = -1, in the Clifford
algebra Cto,l E @ of R'J~.] I
Cauchy's integral formula can also be written in the form
,
S,, ,(x - a) a s f (x) = nw, f (a) for f : Rn -+ Ce:
and in the form
= nu, f (a) for f : IR $ Roln-l + Ce0,~-1,

J,,,(z - a) 8 s f (z)
where the Cauchy kernel is
The paravector z E R@RO~n-lhas a norm lzl = a,

and an arbitrary element
u E CtOtn-l has a norm given by luI2 = (uG)o.
As in the classical case, we conclude that in a simply connected domain
the values of a left monogenic function are determined by its values on the
boundary.
20.7 Monogenic homogeneous functions

A function f : Rn + Cen homogeneous of degree t? satisfies
f (AX) = At f (x) for X E R,
which implies Euler's formula
d
T-f ( ~=)tf (x), where r = 1x1.
dr
If a multivector function f (x) is monogenic, V f (x) = 0, then also xvf (x) =
(x . V + x A V) f (x) = 0, and so
where
and
If a multivector function f is monogenic, Vf(x) = 0, and homogeneous of

d
degree t , that is r-f = ef, then Lf = ~ fwhere
, K = -1. If a multivector
dr
function f : Rn + Ct, is harmonic, V2f = 0, and homogeneous of degree 1,
20.7 Monogenic homogeneous functions
d
that is r-f = ef, then Lf = ~ f where
,
dr
~ = e + n - 2 spinup
K = -l spin down.
This can be seen by writing the Laplace operator V2 as the square of the Dirac
d
operator V = x-'(r-
dr
+ L) and factoring it by the relations
as follows:
If a multivector function f (x) is monogenic, Vf (x) = 0, then h(x) = x f (x)

is harmonic, V2h(x) = 0.
If a multivector function f : Rn + Ce, is monogenic at x # 0, that is
d 8
(r- + L)f(x) = 0, then f(x-l) satisfies (-r-
dr dr
+ L)f(x) = 0, and the
function
is monogenic,
T h e plane case. Consider conformal mappings sending x E R2 to f (x) E R2.

d
For sense-preserving conformal mappings (-r-
dr
+ L)f = 0, and for sense-
d
reversing conformal mappings (r- + L)f = 0. I
dr
Example. Using the vector identity a J ( b A c) = (a - b)c - b ( a . c) we find
that
V J ( X A ~ ) (=~ . x ) f - x ( ~ . f ) + ( x . ~ ) f - i ( ~ . f )
d
=nf-x(V.f)+r-f-f
dr
for a vector field f : Rn + Rn.
If the vector field f is sourceless, V . f = 0, and homogeneous of degree l ,
d
that is r-f = e f , then
dr
V J ( X A ~=) ( t + n - l ) f ,
which shows that
is a bivector potential for f .

If the vector field f is irrotational, V A f = 0, and homogeneous of degree t ,
d
that is r-f = t f , then
dr
+
V(X . f ) = (e i ) f ,
which shows that
is a scalar potential for f . I
20.8 A basis f o r monogenic homogeneous polynomials

A monogenic function is real analytic, that is, a power series of the compo-
nents of the argument. The homogeneous part of degree 1 of the Taylor series
expansion
can be written, in virtue of a monogenic f , as follows:
fl(a)z = (xi - yi1)-
The functions
af
8x1 I+
a
- yi2)-
( ~ 2
sf
8x2 I+
a
. . . + (xn-1 - yin-1)-
+
are monogenic; note that zk = xk yeken, k = 1 , 2 , . . . , n - 1. Write =
(11,12, . . . , In-I), and define the symmetrized polynomials
each term being homogeneous of degree lk with respect to zk , and homogeneous

of degree 1 = l1 + l2 + . . .+ ln-l with respect to z . The symmetric polynomials
are monogenic, Dpl-(z) = 0; they appear as multipliers of partial derivatives
in the Taylor series expansion of f (a + z).

20.8 A basis for monogenic homogeneous polynomials
Examples. 1. For 1 = ( 2 , 1 , 0 , 0 , .. . , 0 ) we have
P&) + +
= i(z12z2 Z l Z Z Z l 22212)
= (xf - y 2 ) x 2- 2x1x2yil - (xf - Slyi2
= ( x : - y2)x2 + 2xixzyeien + (x: - $)yeZen.
2. The functions ql ( z ) = dlq(z) are homogeneous of degree -(I + n - 1) and
monogenic when z # 0. I
For all z E R CBR ~ ~we~also ~ , pl-( I ) E R $ Rotn-l.
- have
The monogenic polynomials homogeneous of degree 1 span a right module
over the ring Cto,,-l; the polynomials pl(z)- form a basis of this module of
dimension (I+;-'). Harmonic polynomials homogeneous of degree 1 form a
module over the rotation group S O ( n ) of R $ I W O J ~ - ~ namely
, the irreducible
module of traceless symmetric tensors of degree 1, the dimension of this module
being
The Laurent series expansion is formulated as follows:
where
when f ( z ) is monogenic in a region S C R $ IIBOJ~-' except at a E S.
Single out a distinguished direction or axis a in Rn, say
Write y = xn and x = x l e l + x2e2 + . . . + xn-len-l scj that
Write x = 1x1 and let

be the unit bivector of the plane determined by a and r . Since V(ra) =
(Vr)a = n a and V(ar) = -(n - 2)a, it is evident that the polynomial
1
w = - [ n a r + ( n - 2 ) r a ] = ( n - l ) a . r + a A r = ( n - 1)y-xi
2
satisfies the equation Vw = 0; such a w is called monogenic. The polynomial
w is a complex of its scalar part u = (n - 1 ) a . r = (n - l)y and its bivector
part v = a A r = -xi. The vector field
is irrotational, that is V A f = 0, with a (scalar) potential u = (n - 1)y,

such that f = -VU, and also sourceless, that is V . f = 0, with a stream
function 4 = -xn-' and a bivector potential v = -xi,such that f = VJv and
V A v = 0. In fact, f is monogenic, Vf = V . f + V A f = 0, with a complex
+
potential w = u vi = (n - 1)y - xi, which is also monogenic, Vw = 0.
The vector functionq(r) = r / r n , where r = Irl, is called the Cauchy kernel,
and
is the field of an n-dimensional dipole. These vector fields can be used to

reproduce the complex potential
of the vector field f = -(n - 1)a.
Axial monogenic polynomials of degree 2. In the above we just found

that
V(ar) = -(n - 2)a
V r a = na.
+
Next, we differentiate the second powers of r . The product ara = (ar ra)a -
a2r= 2(a - r)a - a 2 r is a vector in the plane determined by a and r . So
and (ar)(ra)= a2r2. Since

20.9 Axial monogenic polynomials of homogeneous degree
it is evident that the polynomial
is monogenic, Vpz(r) = 0. In cylindrical coordinates y, x, the polynomial
is a complex potential of the monogenic vector field f = n ( x - (n - 1)ya) which

has a stream function II, = -nyxn-l.
20.9 A x i a l m o n o g e n i c p o l y n o m i a l s o f h o m o g e n e o u s d e g r e e
In the following, monogenic polynomials, homogeneous of degree 1 in the factors
ar and ra (and then also in r or in y and x ) , will be introduced. First,
To verify these, observe that e l f e l + . . . + enfen = -(n - 2)f for any vector f .
Then
xe;
n
i=l
(ar)j-l ae, (ar)'-j = -(n - 2) (ar)j-I a(ar)'-j
= -(n - 2)a(ra)j-l (ar)'-j
because ( a r ) j - l a is a vector in the plane determined by a and r, as can be

seen by inspection on ( r a ) j r = r ( ( a r ) j - l a ) r and induction on j. The equation
(3) is now proved. Similarly, observe that Cy=leiuei = (-l)'(n - 21)u for
any element u, homogeneous of grade 1. Then
because (ra)j + (ar)j is a scalar of grade 0 and (ra)j - (ar)j is a bivector of

grade 2. It follows that
which multiplied by (ra)'-j-' on the right gives the equation ( 4 ) after summing
up the terms j = 0 , 1 ; ~ ~ , 1 1-.
) j that (ar)'-j(ra)j= r 2 j a 2 j ( a r ) 1 - 2when

To calculate ~ ( a r ) ' - j ( r a note j 12
2 j and (ax-)'-j(ra)j= r 2 ( ' - j ) ( r a ) 2 j - when
1 1 5 2 j . Use results ( 3 ) ) ( 4 ) and
for 1 even
( 1 - 1 ) r 1 for l o d d .
Then
1
v (ar)'-j(ra)j = C mjka(ar)'-*(ra)*
k=l
where the (I + 1 ) x I-matrix mjk is
To get the coefficients pl,j in
such that Vpl(r)= 0 , multiply the rows of mjk by the corresponding coeffi-
cients P1,j and determine pl j so that the sum of the resulting elements in each
column is zero. To calculate the coefficients plj one has useful algorithms such
as
1 n-2
. 1 -(T
~ 1 ' , -- +i-.i)(Pl-l,j +
PI-I,[-j)
1 It is worth noting that the formula (5) for the coefficients p l , ) is valid for all signatures
and not only for positive definite quadratic forms.
20.10 Diflerential equations in cylindrical coordinates
which gives
Example. Let n = 3 and 1 = 2. Then

V(ar)' = -a(ar) - a(ra) 5
V(arra) = 2a(ar) 2
V(ra)' = a(ar) + 5a(ra) 1
where the right-hand sides of the identities give the matrix
and the right column gives
with coefficients p2,0 = 5, p 2 , =

~ 2 and p2,' = 1. I
In some cases it is worth knowing the smallest integer coefficients, which in a
few lower-dimensional cases are as follows:
20.10 Differential equations in cylindrical coordinates

Consider an axially symmetric vector field f in the cylindrical coordinates y, x .
+
Write r = x + ya, V = V, + a(d/dy) and f = g ha with g = ( x / x ) ~ Then .
If now f is monogenic, Vf = 0, then there is a complex potential w = u vi =+

u + v such that f = -Vu = VJ v , V A v = 0. The condition to be monogenic,
V w = 0, means that
(vx+ a")d y (u+ ,?)

x du dv xa
= --
xdx
+ x---
xdxx
+ v(n - 2)-a +a-d u + a-dv
x dy
-
xa
= 0,
dy x
which decomposed gives an n-dimensional analog of the Cauchy-Riemann equa-
tions
The components of f = g + ha are then expressed as
Of course, u is harmonic,
and also v is harmonic,
and so
As an axially symmetric and sourceless vector field f has a stream function

$ = x " - ~ v , satisfying
such that
The stream function can be expressed by a line integral
$ - $ J=
~ 1 P
Po
x"-2(gdy - h d x )
20.10 Differential equations in cylindrical coordinates 275
independent of path in a fixed plane containing the symmetry axis a. Let the
path of integration sweep around the symmetry axis a and form an axially
symmetric hypersurface S. The stream
~ = l d ~ / \ f
across S is @ = e 1 2 . . . ~(n - l)wn-I($ - $0) where (n - l)wn-1 is the measure

of the unit sphere Sn-2in the (n - 1)-dimensional space (orthogonal to the
axis a E R n ) .
The n-dimensional Cauchy-Riemann equations ( 6 ) , by a change of variables
x = r sin 8, y = r cos 0, become
du I dv v
-+--+(n-2)cotB-=O
d r r dB r
- 1 du
dv - -- v
d r r dB
+( n - 2)- = 0 .
r
If the complex potential w = u + vi is homogeneous of degree I, then
Differentiation with respect to B and a further change of variable /I = cosB

then result in
and
The solutions u/rl of this last equation can be expressed as hypergeometric

series or ultraspherical (Gegenbauer) functions
and v = [d=/(l+ n - 2)](du/dp). The previously introduced monogenic

polynomials pl (r),homogeneous of degree 1, are now pl (r) = u vi.+
In some lower-dimensional cases the scalar part u of pl(r) divided by r1 is
n 1 u/rl
21c0sls=T (cos 0) Chebyshev
3 1 lJ,(cos 0) Legendre
4 1 (1+
( 1 + l ) ( c o s l ~ + c o t ~ s i=
n ~ ) 1)
+
sin(1 1)O
sin 0
20.11 Inversion of multipoles i n u n i t sphere

If a function f ( r ) with values in the Clifford algebra is monogenic, then the
function Q(r)f (r-l), obtained by inversion, is also monogenic for r # 0. The
vector field
is axially symmetric, monogenic and homogeneous of degree -(1+ n - I ) , and it

describes an axial multipole of order 2'. The previously introduced monogenic
complex polynomials are obtained by inversion in the unit sphere,
These axially symmetric monogenic complex polynomials pl(r) should be dis-

tinguished from the monogenic complex polynomials introduced by Haefeli,
who defined symmetrized products of the functions zi = xi yeje,, where +
i = 1 , 2 ,..., n - 1 .
E x a m p l e . The polynomial ~ ( z ~ z 2 + z l z 2 z l + z 2 = +
z ~x2(x~-y2)+2x1x2yelen
)
Y(X;- (y2/3))ezen is such a monogenic symmetrized product. I
Haefeli's monogenic symmetrized products form a basis of the right module
(over the Clifford algebra) which consists of monogenic polynomials.
E x a m p l e . The axially symmetric monogenic polynomials, homogeneous of
degree 1 and 2, can be expressed in this basis in the forms
n
PI(.) = - and p2(r) = -?
respectively. I
Finally, the complex polynomial pl(r) is such that its bivector part deter-
mines the plane spanned by a and r = x + ya. So the function rpl (r) is vector
valued in this same plane. Since the complex function pl(r) is monogenic,
20.11 Inversion of multipoles in unit sphere 277
Vpl(r) = 0, the vector function rpl(r) is harmonic, V2rpl(r) = 0. This can be
seen from
+
where r V = r . V r A V , r(d/dr) = r . V and L = r A V , which has axially
+
symmetric eigenfunctions Lpl(r) = -lpl(r) and L[rpl(r)] = (1 n - l)rpl(r.)
H i s t o r y and survey of research

Hypercomplex analysis attempts to generalize one-variable complex analysis to
a higher-dimensional one-variable theory using Clifford algebras of Euclidean
spaces. It was first examined by Moisil (in terms of integrals), and rediscovered
in quaternion form by Fueter, who introduced the symmetrized polynomials.
In quaternion analysis the central result was Cauchy's integral formula in di-
mension 4. The notion of monogenic functions with values in a Clifford algebra
is due to Iftimie and Bosshard. Habetha showed that if an algebra gives rise to
Cauchy's integral formula, then it is sufficient that it contains a linear subspace
where all non-zero vectors are invertible in the algebra; that is, the algebra is
almost a Clifford algebra.
Lounesto & Bergh 1983 initiated a study of axially symmetric functions with
values in a Clifford algebra. The research was later taken over by Sommen.
Presently, there are several schools studying hypercomplex analysis with dif-
ferent emphasis: harmonic analysis (J. Ryan, J . Gilbert), functional analysis
(R. Delanghe, F. Brackx), and function theory (K. Habetha, R. Gilbert).
Exercises
+
1. Show that V x = n, Vx2 = 2x, Vx3 = (n 2)x2, Vx4 = 4x3.
+
2. Show that vxk= kxk-' for k even, and vxk= (n k - l)xk-l for k
odd. 1
3. Show that Vcosx = -sinx, V s i n x = c o s x + (n - 1 ) L s i n x .

X
1
4. Show that V exp x = exp x + (n - 1): sinh x ,

X
1 1
V log(1 + x) = -+ (n - 1) - arctanx.
l+x X
Bibliography
P. Bosshard: Die Cliffordschen Zahlen, ihre Algebra und ihre Funktionentheorie.
Ph.D. Thesis, Univ. Ziirich, 1940.
D. Constales: The relative position of LZ domains in complex and in Clifford analysis.
Ph.D. Thesis, Univ. Gent, 1989.
R. Delanghe, F. Brackx: Hypercomplex function theory and Hilbert modules with
reproducing kernel. Proc. London Math. Soc. 37 (1978), 545-576.
R. Delanghe, F. Sommen, V. Soutek: Cliflord Algebra and Spinor Valued Functions:
A Function Theory for the Dirac Operator. Kluwer, Dordrecht, The Netherlands,
1992.
J. Gilbert, M. Murray: Cliflord Algebras and Dirac Operators i n Harmonic Analy-
Cambridge, 1991.
R.P. Gilbert, G.N. Hile: Hypercomplex function theory in the sense of L. Bers. Math.
Nachr. 72 (1976), 187-200.
K. Giirlebeck, W. Sprossig: Quaternionic Analysis and Elliptic Boundary Value Prob-
lems. Akademie-Verlag, Berlin, 1989. Birkhauser, Basel, 1990.
K. Giirlebeck, W. Sprossig: Quaternionic and Cliflord Calculus for Physicists and
Engineers. Wiley, Chichester, 1997.
K. Habetha: Function theory in algebras; pp. 225-237 in E. Lanckau, W. Tutschke
(eds.): Complex Analysis: Methods, Trends and Applications, Akademie-Verlag, Ber-
lin, 1983.
G. Haefeli: Hyperkomplexe Differentiale. Comm. Math. Helu. 20 (1947), 382-420.
D. Hestenes: Multivector calculus. J. Math. Anal. Appl. 24 (1968), 313-325.
V. Iftimie: Fonctions hypercomplexes. Bull. Math. Soc. Sci. Math. R.S. Roumanie
9,57 (1965)) 279-332.
P. Lounesto, P. Bergh: Axially symmetric vector fields and their complex potentials.
Complex Variables, Theory and Application 2 (1983), 139-150.
G.C. Moisil: Les fonctions monogdnes duns espaces 6 plusieurs dimensions. C.R.
Congrhs des Soc. Savantes de Clermont-Ferrand, 1931.
J. Ryan (ed.): Clifford Algebras i n Analysis and Related Topics. CRC Press, Boca
Raton, FL, 1996.
F. Sommen: Special functions in Clifford analysis and axial symmetry. J. Math. Anal.
~ p p l .130 (i988), 110-133.
A. Sudbery: Quaternionic analysis. Math. Proc. Camb. Phil. Soc. 85 (1979), 199-225.
21
Binary Index Sets and Walsh Functions
The present chapter scrutinizes how the sign of the product of two elements
in the basis for the Clifford algebra of dimension 2" can be computed by the
Walsh functions of degree less than 2n. In the multiplication formula the basis
elements are labelled by binary n-tuples, which form an abelian group Q, which
in turn gives rise to the maximal grading of the Clifford algebra. The group of
binary n-tuples is also employed in the Cayley-Dickson process.
WALSHFUNCTIONS
Consider n-tuples = a l a z . . . a n of binary digits a, = 0 , l . For two such
n-tuples g and b the sum g $ b = c is defined by termwise addition modulo
2, that is,
ci = ai + bi mod 2.
These n-tuples form a group so that the group characters are Walsh functions
- = (-1)x:=l
wa(b)
The Walsh functions have only two values, f1, and they satisfy wk(g $ b) =
wk(g)wk(b), as group characters, and w,(b) = wb(g). The Walsh functions
- labelled by binary n-tuples & = klk2.. . k,, can be ordered by integers
wk,
k = C;="=,i2n-'.
21.1 Sequency order

In applications one often uses the sequency order of the Walsh functions,
~ - ~= ( - (l ) k l ~~l + C : =1~ ( k ~ - l + k , ) ~ ,
9
280 Binary Index Sets and Walsh Functions
for instance, in special analysis of time series, signal processing, communica-
tions and filtering, Harmuth 1977 and Maqusi 1981. In the sequency order the
index is often replaced by an integer k = Cyz1ki2n-i and the argument :
by a real number on the unit interval x = 2-n Cy=p~'-l(Fig. 1 and Fig. 2).
-(x), k = 0,1, . . . , 7
Figure 1. The first eight Walsh functions Gk
In Figure 1 the first eight Walsh functions are given:

6 k-( x ) = (-l)kl~1+(kl+kz)~z+(kz+k3)~3
+
with k = 4k1+ 2k2 kg and x = ;(XI +2x2 + 4x3). Observe that the number
of zero crossings per unit interval equals k.
g(g $ b) = g(g) $ g(b). The inverse h of the Gray code is obtained by
h(g))i = a j mod 2.
OF CLIFFORDALGEBRAS
BINARYREPRESENTATIONS
As a preliminary example, consider the Clifford algebra Cto,2,isomorphic to the
division ring of quaternions W. Relabel the basis elements of CtoP2by binary
2-tuples
1 eoo
el, e 2 elo, e01
e12 ell
and verify the multiplication rule
e,eb = w,(h(b))e,@k.
For an alternative representation reorder the basis elements by the formula
to get the correspondences
This yields the multiplication rule
21.3 Clifford multiplication

In general, consider the Clifford algebra Cto,, with n generators el, e 2 , . . . , en
such that
e:=-1 for i = 1 , 2 ,..., n,
e.e,
3~ = -eje)i for i # j.
T h e o r e m 1. If a real 2n-dimensional algebra A has the multiplication rule
e,eb = w,(h ( b ))eg@b

between the basis elements labelled by the binary n-tuples, then A is isomor-
phic to the Clifford algebra Cto,,,
21.3 Clifford multiplication 283
Proof. It is sufficient to show that A is associative, has a unit element and is
generated by n anticommuting elements with square -1.
The element eo- = eooo,,,oois the unit, since e,eo = w,(h(Q))eCBo = wo(Q)e, =
+e,- and similarly eoe,- = +e,. The n basis elements
generate by definition all of A. Each generator has square -eo;

- in particular
for the i :th generator e -
,
a = 00.. .010.. .00, h(g) = 0 0 . . .011.. . l l
-
1 2 n 1 i n
and SO w-a ( h ( g ) )= -1, from which one concludes that e,e, = w,(h(g))eCe, =
-eo.- In a similar manner one finds that generators anticommute with each
other.
Finally, A is associative, since for three arbitrary basis elements e,, eb, e ,
-
the condition (eaeb)e, = e,(ebe3 is equivalent to
= w,(h(b@ c))wb(h(c)),
~,(h(b))~,$b(h(C))
which is a consequence of wCeb(z)= w,(z)wb(z) and w,(z @ y) = w,(:)w,(y)
and h being a group isomorphism. I
It is convenient to assume the correspondences
ei =coo...040...00 for i = 1 , 2 , . . . , n
1 a n
between the ordinary and binary representations of the generators of the Clif-
ford algebra Ceo,,. Then the basis elements of C l o p , are labelled by the binary
n-tuples = alaz . . .an as follows:
Since the Gray code is a group isomorphism among the binary n-tuples, we
can reorder the basis of the Clifford algebra Ceo,, by
-
e , = eg(g).
This reordering results in a simple multiplication formula:
Corollary. The product of the basis elements of the Clifford algebra Cto,, is
given by
- -
eseb = ~ a ( b ) ~ c e b .
Proof.
6a-b = eg(g)eg(b)= ~g(g)(h(g(b)))eg(a)eg(b)
- I
- wg(c3)(b)eg(@b) = &(b)eg$b.
If you choose the signs in e,ek - = in some other way, you get other
algebras than Ceo,, . For instance, the Clifford algebra Cep,, over the quadratic
form zq + . . . + x i - z;+, - . . . - xi+, has the multiplication formula
e,eb = (-1)xLx aabaw,(h(b))ec3ek.
Of course, this might also be written without Walsh functions:

e,ek = ( - 1 ) x h ~ + laiba(-l)x:>J
albjec3eb,
a formula essentially obtained by Brauer & Weyl 1935. See also Artin 1957 and
Delanghe & Brackx 1978 for a related definition of the product on the Clifford
algebras (based on sums of multi-indices).
21.4 An iterative process t o f o r m Clifford algebras

Clifford algebras can be obtained by a method analogous to the Cayley-Dickson
process. Consider pairs (u, v) of elements u and v in the Clifford algebra C$,,.
Define a product for two such pairs,
where u + u is the grade involution of Cep,,. This results in an algebra iso-

morphic to the Clifford algebra
according to the f sign. This iterative process could be repeated by noting

that (u, v)^ = (C, -6).
For more details on the Clifford algebras see Micali & Revoy 1977 and Por-
teous 1969, 1981.
All the above algebras are special cases of the following. Let A be a real linear
space of dimension 2". Label a basis for A by binary n-tuples g to get the
basis elements e,. Then define a multiplication between the basis elements e,-
and extend it to all of A by linearity. The definition is of the form
21.5 Ca yle y-Dickson process 285
for a certain choice of signs. Then the algebra A is a direct sum of the 1-
dimensional subspaces Ua-, spanned by ea-, satisfying
In other words A is an algebra graded by the abelian group of binary n-tuples

Q. This grading is maximal (Kwasniewski 1985), and these algebras will be
called Clifford-like algebras. Next we shall study some Clifford-like algebras.
21.5 Cayley-Dickson process

Consider a generalized quaternion ring Q with i2 = 71, j2 = 72 and k2 = 7172,
where 71, 72 = f1. The conjugation-involution u 4 uL of Q is given by
Introduce a multiplication in the 8-dimensional real linear space Q x Q by the

formula
where 73 = f1. Inducing an anti-involution (u, v ) = ~ (uL,-v) of Q x Q =

CD(yl,yz,73) makes it possible to repeat this Ca yley-Dickson process to get
an algebra CD(y1,72,. . . , 7"), where 7j = f 1. In fact, the Cayley-Dickson
process could be started with IR to give CD(yl) and Q = CD(yl, 72).
Example. CD(-1) 2: C, CD(-1, -1) z W,and CD(-1, -1, -1) 2: 0, the
real 8-dimensional alternative division algebra of octonions (Porteous 1969,
1981). I
The algebras CD(yl, 72,. . . , 7,) obtained by the Cayley-Dickson process are
simple flexible algebras of dimension 2" (Schafer 1954). Every element of such
an algebra satisfies a quadratic equation with real coefficients.
21.6 Binary representation of t h e Cayley-Dickson process

The algebras formed by the Cayley-Dickson process are Clifford-like algebras.
For instance, choose a basis of CD(yl) = IR x IR,
and introduce the multiplication rule

The involution is given by
- = (-l)alea.
or in a condensed form et -
Theorem 2. A Clifford-like algebra A , dim A = 2n, with multiplication rule
<
where Si(g) is the maximum of aj for 1 5 j i, is isomorphic to the Cayley-
Dickson algebra CD(yl, 72, . . . , 7,). The anti-involution is
Proof. The first case of the mathematical induction is proved in the example
above.
Assume that the statement holds up to the n:th step, and apply the Cayley-
Dickson process. If the new basis elements are denoted by
eaoebo = (e,, O)(eb,0) = (eaeb,0) = f (a,b)(eEeb,0) = f (a,b)eEebo

ealebo = (0, ed(eb,0) = (0, ea+) = ( - ~ ) ~ ~ ( (a, f - -
b ) b)ea@bl
e~eb-1= (e,, 0)(0, ed = (0, eked = f (b, a)eEebl
- = (0, ed(0,ed = ( ~ + l e i e 0)
e~leb-1 (b, ~)egeb-o.
a ,= % t i (-l)Sn(b)f
These four equations can be condensed into one equation
where
f (b, a) = f (-,a -b)(- l)Sn(E)+Sn(b)+Sn(C@b),
which is a consequence of (eaedL= eke:. Thus we have proved the desired

multiplication rule in the case n + 1. he induced anti-involution is also of the
assumed type:
21.6 B i n a r y representation of the Cayley-Dickson process
or in a condensed form
L
el?an+l
= ( - I ) max (sn(d,an+l)eaan+l
-
. I
The algebra CD(y1, 72, . . . , 7n)is generated by an n-dimensional vector space,

whose elements
+ + +
have squares (-ylxq -y2x; . . . yn x:)eo.- In contrast to the Clifford algebras,
different orderings of the parameters yi in C D ( y l , 7 2 , . . . , 7,) may result in
non-isomorphic algebras in the case where n > 3.
Another construction relating Clifford algebras and Cayley-Dickson algebras
is found in Wene 1984.
For more details of the algebraic extensions of the group of binary n-tuples
R see Hagmark 1980.
Bibliography
E. Artin: Geometric Algebra. Interscience, New York, 1957.
R. Brauer, H. Weyl: Spinors in n dimensions. Amer. J . Math. 5 7 (1935), 425-449.
R. Delanghe, F. Brackx: Hypercomplex function theory and Hilbert modules with
reproducing kernel. Proc. London Math. Soc. (3) 3 7 (1978),545-576.
N.J. Fine: On the Walsh functions. Trans. Amer. Math. Soc. 6 5 (1949), 372-414.
P.-E. Hagmark: Construction of some 2"-dimensional algebras. Helsinki UT, Math.
Report A177, 1980.
H.F. Harmuth: Sequency Theory, Foundations and Applications. Academic Press,
New York, 1977.
A.K. Kwasniewski: Clifford- and Grassmann-like algebras - old and new. J. Math.
Phys. 26 (1985), 2234-2238.
M. Maqusi: Walsh Analysis and Applications. Heyden, London, 1981.
A. Micali, Ph. Revoy: Modules quadratiques. Cahiers MathCmatiques 10, Montpellier,
1977. Bull. Soc. Math. France 63, suppl. (1979), 5-144.
I.R. Porteous: Topological Geometry. VNR, London, 1969. Cambridge University
76 (1954), 435-446.
K. Th. Vahlen: ~ b e hijhere
r komplexe Zahlen. Schriften der phys.-okon. Gesellschaft
zu Konigsberg 3 8 (1897), 72-78.
G.P. Wene: A construction relating Clifford algebras and Cayley-Dickson algebras. J.
Math. Phys. 25 (1984), 2351-2353.
22
Chevalley's Construction and Characteristic 2
Consider an n-dimensional linear space V over a field IF, char IF # 2, and the
symmetric bilinear form
associated with the quadratic form Q. Later in this chapter we will discuss the
case charIF = 2. As before, we denote the exterior algebra of V by /\ V and
the Clifford algebra of Q, with x2 = Q(x), by Ce(Q) or Cep,, when V = RPl"
and
We shall construct a natural linear isomorphism /\ V + Ce(Q), review how

Riesz goes backwards and derives /\ V from Ce(Q) and compare Riesz's method
to an alternative construction due to Chevalley but known to some theoretical
physicists in the disguise of the Kahler-Atiyah isomorphism.
22.1 Construction of t h e linear isomorphism /\ V 4 Ce(Q)

Here we start from the exterior algebra /\ V. Recall that a k-vector a E l\kV
has the grade involute i = (-l)ka and the reverse a = ( - 1 ) 3 ~ ( ~ - ' ) a The
.
symmetric bilinear form associated with Q on V can be extended to simple
k-vectors in l\kV by way of
where
<xl,yl> <x1,y2> ...
<x2, y1> <x2,y2> . . . <x2,Y k >
det<xi, yj> = ,
<xk,yl> <xk,Y2> . . . < x k ,Y k >

22.1 Construction of the linear isomorphism /\ V + Ct(Q) 289
and further by linearity to all of l\kV and by orthogonality to all of /\ V.
+
Example. Let Q(xlel+ x2e2) = ax: bx;. Then <x, y> = axlyl bx2~2 +
and x ~ =y(xly2-x2yl)elAe2. The identity ( a x ~ + b x ~ ) ( a y ~ +=
b y(axlyl+
~)
bx2y2)2+ab(z1y2 - ~ 2 ~ 1 )can
' be written as Q(x)Q(Y)= <x, y>2 Q(xAY), +
where Q(x A y) = ab(xly2 - ~ 2 ~ 1 ) ~ . I
In the case of a non-degenerate Q on V we can introduce the dual of the
exterior product called the left contraction u J v of v E /\ V by u E /\ V
through the requirement
Examples. 1. Let x , y E V , w ~ l F Then

. <xJy,w>=<y,xAw>=
<y,xw> = <y,x>w and since <x J y,w> = <x J y, l>w we have the rule
x-l y = < x , y > .
2. Let x, y, z, w E V. Then <x J (y A z), w> = <y A z , x A w>
= ( x J y)<z,w> - ( x J z)<y,w> = <(x J y)z - (x J z)y, w> and so we have

the rule x J ( y A z ) = ( x J y ) z - ( X J Z ) ~ .
3. Let x E V, xi E V and w = w ~ A w ~...Awk-1 A E A ~ - ' V . Then
< ~ J ( x ~ A x ~ A . . . A x ~ )< ,x ~wA>x :=! A ... A x ~ , x A wAw2A
~ ... A ~ k - 1 >
and so
Cheualley's Construction and Characteristic 2
5. < ( u A v ) J w , z > = < w , ( u A v ) - A z > = < w , i i A G A z > = < u J w , G A z >

= < u J ( v J w ) , z > andso ( u ~ v ) J w = ~ J ( v J w ) . I
In the case of a non-degenerate Q we have verified the following properties of

the contraction:
(Q) x ~ y = < x , y > for X , ~ E V

PI xJ(uAv) = ( x J u ) A v + u ~ ( x J u )
(c) ( u A v ) J w = ~ J ( u J w ) for u,v,wE/\V
(see Helmstetter 1982). These properties also determine the contractionuniquely
for an arbitrary, not necessarily non-degenerate Q. The identity (c) introduces
a scalar multiplication on /\ V making it a left module over /\ V. The identity
(b) means that contraction by x E V operates like a derivation. Evidently,
XJ~E/\~-'V for a ~ l \ ~ b ' .
Introduce the Clifford product of x E V and u E /\ V by
and extend this product by linearity and associativity to all of /\ V, which

then becomes, as an associative algebra, isomorphic to CC(Q). For instance, the
product of a simple bivector x A y E /\2 V and an arbitrary element u E /\ V
is given by
where we have first expanded ( x A y)u = (xy - x J y)u, then used
and the derivation rule x J ( y A u) = (x J y ) A u - y A (x J u).

Exercises 1,2,. . . ,10
Remark. Some authors use, instead of the left and right contractions, a
more symmetric dot product in CC(Q), defined by the Clifford product for
homogeneous elements as (char # 2)
22.2 Chevalley's identification of Cl(Q) C End(/\ V) 29 1
and extended by linearity to all of CC(Q). The relation between the dot product
and the contractions,
shows that the dot product cannot be expected to have properties, which can
be easily proved (using the more fundamental contractions). Some authors try
to make the dot product look like derivation, and define exceptionally A.u = 0,
u - A = 0 for A E F. This only makes things worse, because for this dot product
the relation to the contractions is still more complicated:
22.2 Chevalley's identification of Cl(Q) c End(/\ V)

Chevalley 1954 pp. 38-42 tried to include the characteristic 2 by embedding
the Clifford algebra Cl(Q) into the endomorphism algebra End(/\ V) of the
exterior algebra /\ V. He introduced a linear operator LL = cp, E End(/\ V)
such that
( o , ( u ) = x ~ u + x J u for X E V , u EAV.
From the derivation rule x J ( x A u ) = ( x J x ) A u - x A ( x J u ) and x A x A u = 0,
x J ( x J u ) = 0 we can conclude the identity (cp,)2 = Q(x). Chevalley's inclusion
map V 4 End(/\V), x + cp, is then a Clifford map and can be extended
to an algebra homomorphism 1C, : Cl(Q) + End(/\ V), whose image evaluated
a t 1 E / \ V yields the map 4 : End(/\ V) 4 /\ V. The composite linear map
6' = 4 o 1C, is the right inverse of the natural map /\ V 4 Cl(Q) and
is the identity mapping on /\V. The faithful representation II, sends Cl(Q)
onto an isomorphic subalgebra of End(/\ V).
Chevalley's identification also works well with a contraction defined by an
arbitrary - not necessarily symmetric - bilinear form B such that B ( x , x) =
Q(x) and
(Q) x-ly=B(x,y) for x , y ~ V
(c) (uAv)Jw=uJ(vJw) for u,v,wE/\V

292 Chevalley's Construction and Characteristic 2
(see Helmstetter 1982). As before, x J a E l\k-lV for a E l\kV and
and the faithful representation 1C, sends the Clifford algebra Cl(Q) onto an
isomorphic subalgebra of End(/\ V), which, however, as a subspace depends
on B.
R e m a r k . Chevalley introduced his identification CC(Q) c End(/\ V) in or-
der to be able to include the exceptional case of characteristic 2. In char-
acteristic # 2 the theory of quadratic forms is the same as the theory of
symmetric bilinear forms and Chevalley's identification gives the Clifford alge-
+
bra of the symmetric bilinear form <x, y > = ;(B(X, y ) B(y, x)) satisfying
xy+yx=2<x,y>. I
For arbitrary Q but charF # 2 there is the natural choice of the unique sym-
metric bilinear form B such that B(x,x) = Q(x) giving rise to the canon-
ical/privileged linear isomorphism CC(Q) 4 /\ V. The case char IF = 2 is
quite different. In general, there are no symmetric bilinear forms such that
B(x,x) = Q(x) and when there is such a symmetric bilinear form, it is not
'
unique since any alternating bilinear form is also symmetric and could be
added to the symmetric bilinear form without changing Q. Hence the contrac-
tion is not unique if charlF # 2, and there is an ambiguity in cp,.
In characteristic 2 the theory of quadratic forms is not the same as the t,heory
of symmetric bilinear forms.
+
Example. Let dimp V = 2, B(x, y ) = axlyl bxly2 cxzyl dxzy2 and + +
Q(x) = B(x, x). The contraction x J y = B(x, y) gives the Clifford product
+
xv = x A v x J v of x E V, v E /\ V. We will determine the matrix of v + uv,
+
u = uo u1e1+ uze2 f 211281 A e 2 with respect to the basis (1, el, ez, el ~ e ~
for /\ V. The matrix of el is obtained by the following computation:
ell=elAl=el (first column = 0100)
elel = el J el = a (second column = a000)
eiez = el A e 2 + el = el A e 2 b
J e2 +(third column = b001)
el(e1 A ez) = el J (el A ez) = (el J e l ) A ez - (el J e2) A el = aez - bel.
1 Recall that antisymmetric means B ( x ,y ) = - B ( y , x ) and alternating B ( x , x ) = 0; alter-

nating is always antisymmetric, though in characteristic 2 antisymmetric is not necessarily
alternating.
22.2 Cheualley's identification of Cl(Q) C End(/\ V) 293
The matrix of el A e2 is obtained b y
(el A ez) 1 = el A e2
(el A e2)el = (el A e2) L el = el A (ez L e l ) - ez A (el L e l ) = cel - aez
(el A e2)ez = ( 8 1 8 2 - el J ez)ez = ele; - (el J e2)e2= del - be2
(el A ez)(el A ez) = (elez - el J ez)(el A ez)
+
= el(e2 A (el A ez) e2 J (el A ez))- (el J e2)(elA e 2 )
= el (ce2 - del) - b(e1 A e2) = -ad + bc + (-b + c)(el A e2).
So we have the following matrix representations:
O a b 0 c d O
el =
0 0 1 0 0 - 1 0 0
or in general
+ +
Evidently, the commutation relations elez e2el = b c and e: = a, e; = d
are satisfied, and we have the following multiplication table:
el a el A e2 b + -bel +
aez
ez -el Aez c + d -del +
cez
el A ez ce~- aez del - be2 -ad + +bc (-b +
c)el A e2
In characteristic #2 we find
1 1
-(ele2 - e2el)= el A ez + - ( b - c)
2 2
+
and more generally for x = xlel x2e2, y = ylel yzez +
1
-(XY - Y X )
2
=( ~ ~2 2 ~el1 A) e2
~ 1- + l1( b - C ) ( X I Y ~- ~ 2 ~ 1 )
294 Cheualley's Construction and Characteristic 2
with an alternating scalar valued form A(x, Y) = ( ~ ( xY), -B(y, x)). For non-
zero A(x, Y) the quotient x A Y/A(x, Y) is independent of x, y E V. Note that
the matrix of $(xy - YX) is traceless. The symmetric bilinear form associated
with Q(x) is
1 1
+ + + + +
x . Y = -(B(x, y ) B(y, x)) = a x l y ~ 2(b c)(x1~2 ~ 2 ~ 1d)x 2 ~ 2
2
and we have x y + y x = 2x . y for x , y E V c Ct(Q). Orthogonal vec-
tors x I y anticommute, x y = -yx and (xy)' = - x ~ even ~ ~ though
,
x y = x A y + A(x, y). [In this special case A(x, y ) = B(x, y ) # 0 while
x - y = 0 implies B ( y , x ) = -B(x, y).] I
It is convenient to regard /\ V as the subalgebra of End(/\ V) with the canon-
ical choice of the symmetric B = 0. We may also regard Ct(Q) as a subalgebra
of End(A V) obtained with some B such that B(x, x) = Q(x) and choose the
symmetric B in char # 2.
The following example shows that for Q = 0 and B = 0, Chevalley's process
results in the original multiplication of the exterior algebra /\ V, but that for
Q = 0 and alternating B # 0, the process gives an isomorphic but different
exterior multiplication on /\ V.
Example. Take a special case of the previous example, the Clifford algebra
with Q = 0 and B(x, y) = b(xly2 - x2y1). Send the matrix of the exterior
product (with the symmetric bilinear form = 0) to a matrix of the isomorphic
Clifford product (determined by the alternating bilinear form = B ) :
In this case
+
In particular, P(el)P(ez) = P(e1 A ez b), P(ez)P(el) = -P(el A ez b) +
+
and P(el e 2 b)P(ei) = 0, P(ei)P(el A ez + b) = 0. We have already met
this situation in the chapter on the Definitions of the Clifford Algebra in the
section on the Uniqueness and the definition by generators and relations except
that here the exterior algebra and the Clifford algebra (determined by the
alternating B) are regarded as different subspaces of End(/\ V) [although they
are isomorphic subalgebras of End(/\ V)]. I
22.2 Chevalley's identification of CL(Q) c End(/\ V) 295
The above example shows that those who do not accept the existence of k-
vectors in a Clifford algebra Cl(Q) over IF, charIF # 2, should also exclude
fixed subspaces /\k V C /\ V.
In general, consider two copies of CC(Q) in End(/\ V) so that Q(x) equals
Bl(x, x ) = B2(x, x), which determine & (x)Pl (y) = Pl(x A y +
Bl (x, y))
and P2(x)P2(y)= P ~ ( A +
x y B2(x,y)). A transition between the two copies
is given by an alternating bilinear form B(x, y ) = Bl(x, y ) - B2(x,y ) and
+
P(x)P(y) = P ( ~ YB(x, Y)).
In characteristic # 2 this means that the symmetric bilinear forrn such that
< x , x > = Q(x) gives rise to the natural choice x y = x A y +
<x, y > among
+
the Clifford products x y B(x, y ) with an alternating B. In other words, the
Clifford product x y has a distinguished decomposition into the sum x A y +
+
<x, y > where <x, y > = 3(xy yx) is a scalar and x A y = ;(xy - yx) is a
bivector [this decomposition is unique among all the possible decompositions
with antisymmetric part x A y = $(xy - yx) equaling a new kind of bivector
. 2
+
x A y = x A y B(x, y) where X A y E /\ V and x A y E A ~ V ] [Similarly,
.
a completely antisymmetric product of three vectors equals a new kind of 3-
+
vector x A y A z = x A y A z xB(y, z) + YB(z,X) zB(x, y).] +
Example. Let IF= {0,1}, dimFV = 2 and Q(xlel+x2e2) = 21x2. There are
only two bilinear forms Bi such that Bi(x, x) = Q(x), namely Bl (x, y ) = x1y2
and B2(x, y ) = x2 yl, and neither is symmetric. The difference A = B1 - B2,
A(x, y ) = xly2 - x2y1 (= x1y2 + x2y1) is alternating (and thereby symmetric).
"
Therefore, there are two representations of CL(Q) in End(/\ V) :
for B1 : u = ( uo
u1
u2
U12
0
uo
0
-u2
u1
0
uo - 2112
u1
0
uo - 2112
for B2 : u = ( 210
u1
u2
U12
uo
0
-u2
0
212
+ u12
0
uo u2
u1 uo 2112 +
These representations have the following multiplication tables with respect to
the basis (1, el, e 2 , el A e2}for /\ V :
In this case there are only two linear isomorphisms /\ V -+ CL(Q) which are
identity mappings when restricted to IF $ V and which preserve parity (send
even elements to even elements and odd to odd). It is easy to verify that
the above multiplication tables actually describe the only representations of
CL(Q) in A V. In this case t h e r e are n o canonical linear isomorphisms
/\ V -+ CL(Q), in other words, neither of the above multiplication tables can be
preferred over the other. In particular, /\'V cannot be canonically embedded
in CL(Q), and t h e r e are n o bivectors i n characteristic 2. I
The need for a simplification of Chevalley's presentation is obvious. For in-
stance, van der Waerden 1966 said that 'the ideas underlying Chevalley's proof
(p. 40) are not easy to discern' and gave another proof, equivalent but easier
to follow. [Also Crumeyrolle 1990 p. xi claims that 'Chevalley's book proved
too abstract for most physicists' and in a Bull. AMS review Lam 1989 p. 122
admits that 'Chevalley's book on spinors is ... not the easiest book to read.']
It might be helpful to get acquainted with a simpler and more direct method
of relating /\ V and CL(Q) due to M. Riesz 1958/1993 pp. 61-67. Riesz in-
troduced a second product in CL(Q) making it isomorphic with /\ V without
resorting to the usual completely antisymmetric Clifford product of vectors and
constructed a privileged linear isomorphism CL(Q) + /\ V.
22.3 Riesz's introduction of a n exterior p r o d u c t i n CL(Q)

In the following we review a construction of M. Riesz 1958/1993 pp. 61-67. Start
from the Clifford algebra CC(Q) over IF, char IF # 2. The isometry x -+ -x of
V when extended to an automorphism of CL(Q) is called the grade involution
u + u. Define the exterior product of x E V and u E CL(Q) by
and extend it by linearity to all of CL(Q), which then becomes isomorphic to

/\ V. The exterior products of two vectors x A y = $(xy - yx) are simple
bivectors and they span A2 V. The exterior product of a vector and a bivector,
x A B = ~ ( X B+ Bx), is a 3-vector in r\3 V. The subspace of k-vectors is
22.3 Riesz's introduction of an exterior product in el(&)
constructed recursively by
k k-1
1
2
/\
x A a = - ( ~ a + ( - l ) ~ - l a x E) / \ V for a € 1/
We may deduce associativity of the exterior product as follows. First, the

definition implies for x, y, z E V
Then x A (y A 2) = (x A y) A z since
xyz - zyx = xyz - zyx + (zy + yz)x - x(yz + zy) = yzx - xzy.
This last result implies
when char IF # 2,3 (note the resemblance with antisymmetric tensors). [Simi-
larly, we may conclude that xyz+zyx = x(yz+zy) - (xz+zx)y +z(xy +yx)
is a vector in V.] Riesz's construction shows that bivectors exist in all char-
acteristics # 2.
Introduce the contraction of u E el(&)by x E V so that
and show that this contraction is a derivation of el(&)since

1 1
x J (uv) = -(xuv - &x) = -(xuv - G x )
2 2
Thus one and the same contraction is indeed a derivation for both the exterior
product and the Clifford product. [Kahler 1962 p. 435 (4.4) and p. 456 (10.3)
was aware of the equations
x J ( u A v ) = ( x J u ) ~ ~ + U ~ ( x Jand
v)
+
xJ(uv) = (xJu)v U(xJv).]
Provided with the scalar multiplication (u A v) J w = u J (v J w), the exterior
algebra V and the Clifford algebra el(&)are linearly isomorphic as left
A V -modules.
Exercises 11,12,. . . ,20
Exercises
Show that
1. X A ( Y J U ) - Y ~ ( ~ ~ ~ ) = ~ ~ ( Y for ~ X
~ ,)Y -E V
Y.~ ( ~ ~ ~
2. X A Y A ( Z J U ) - X A Z A ( Y J U ) + Y A Z A ( X J U )
=~J(~AzAu)-~J(XAZAU)+ZJ(XA~AU).
3. x A ( y J (z J u)) - y A (XJ (z J u)) + z A (x J (y J u))
=(xAY)J(zAu)-(XAZ)J(YAU)+(YAZ)J(XAU)
=XJ(YJ(ZAU))-XJ(ZJ(~AU))+~J(ZJ(XAU)).
4. (XAYAZ)U=XAYA(ZJU)-XAZA(~JU)+~AZA(XJU)
+
+x A (y J (z J u)) - y A (x J (z J u)) z A (x J (y J u))
+XA~AZAU+XJ(YJ(ZJU)).
5. a J b E /\'-jV for a € /\%, b E A ' V (charF# 2).
In the next five exercises we have a non-degenerate Q. Define the right con-
traction u L v by <u L v, w> = <u, w /\ G> for all w € /\ V (we say that u is
contracted by the contractor v). Show that
6. u J v = v o L u o - v o L u l + v l L U O + V ~ L U ~ = V L U - ~ V O L U ~
(VO= even(v), ul = odd(u)).
7. u L v = v ~ J u ~ + ~ ~ J u ~ - v ~ J u ~ + v ~ J u ~ = v J u - ~ v ~ J u ~
(vl = odd(v), uo = even(u)).
Show that (when char IF # 2)
8. a J b = (-l)"j-i)b L a for a E /\%, b E /\j V.
9. ~ E / \ ~ va#O, , X E V , x J a = O u x = a J b forsome b€/\"+'V.
10. b E /\k v is simple w ( a J b ) A b = 0 for all a E /\k-l V.
11. x and x A y anticommute for vectors x, y E V.
12. x and x J B anticomrnute for a bivector B E /\' V.
13. (x A y ) 2 = (x J y ) 2 - x2y2 (Lagrange's identity).
14. ( x A y A z ) J u = ( x A y ) J ( z J u ) = ~ J ( y J ( z J u ) ) .
15. (xyz - zyx)' E IF, x A y A z = $(xyz - zyx).
16. a A b = ( - l ) j j b A a for a € h i V and ~ E / \ ' v .
17. u A v = vo A uo + vo A ul + v1 A uo - v1 A u1 = v A u - 2211 A u1 where
u, v E /\ V, uo = even(u), vo = even(v), u l = odd(u), vl = odd(v).
18. B U = B A U + $ ( B U - U B ) + B J U for B€/\'v.
[Hint: (x A y ) A u + (xA y ) J u = X A (y A U)+ x J (y Ju).]
19. Q(u) = <u, v> = (fi J v)o (= the scalar part of fi J v).
In the last exercise we have a non-degenerate Q :
20. Q on V extends to a neutral or anisotropic Q on Ct(Q).
22.3 Riesz's introduction of an exterior product in C l ( Q ) 299
Bibliography
N. Bourbaki: Algt?bre, Chapitre 9, Formes sesquiline'aires et formes quadratiques. Her-
mann, Paris, 1959.
C. Chevalley: Theory of Lie Groups. Princeton University Press, Princeton, N J , 1946.
York, 1954.
J. Helmstetter: Alggbres de Clifford et algkbres de Weyl. Cahiers Math. 25, Mont-
pellier, 1982.
E. Kiihler: Der innere Differentialkalkiil. Rendiconti di Matematica e delle sue Appli-
cazioni (Roma) 2 1 (1962), 425-523.
plied Mathematics, Lecture Series No. 38, University of Maryland, i958. Reprinted
lands, 1993.
B.L. van der Waerden: On Clifford algebras. Neder. Akad. Wetensch. Proc. Ser. A69
(1966), 78-83.
Math. 176 (1937), 31-44.
23
Octonions and Triality
Complex numbers and quaternions form special cases of lower-dimensional Clif-

ford algebras, their even subalgebras and their ideals
In this chapter, we explore another generalization of C and W, a non-associative

real algebra, the Cayley algebra of octonions, 0. Like complex numbers and
quaternions, octonions form a real division algebra, of the highest possible
dimension, 8. As an extreme case, 0 makes its presence felt in classifications,
for instance, in conjunction with exceptional cases of simple Lie algebras.
Like C and W, 0 has a geometric interpretation. The automorphism group
of W is S 0 ( 3 ) , the rotation group of R3 in W = R $ R3. The automorphism
group of 0= R $R7 is not all of S 0 ( 7 ) , but only a subgroup, the exceptional
Lie group G2. The subgroup G2 fixes a 3-vector, in /\,EX7, whose choice
determines the product rule of 0.
The Cayley algebra 0 is a tool to handle an esoteric phenomenon in di-
mension 8, namely triality, an automorphism of the universal covering group
Spin(8) of the rotation group SO(8) of the Euclidean space R8. In general, all
automorphisms of SO(n) are either inner or similarities by orthogonal matrices
in O(n), and all automorphisms of Spin(n) are restrictions of linear transfor-
mations Cen + Ctn, and project down to automorphisms of SO(n). The only
exception is the triality automorphism of Spin(8), which cannot be linear while
it permutes cyclically the three non-identity elements - 1, el2...g, - e 1 2 . . .s in the
center of Spin(8).
We shall see that triality is a restriction of a polynomial mapping Ces -+ Ces,
of degree 2. We will learn how to compose trialities, when they correspond to
different octonion products. We shall explore triality in terms of classical linear
algebra by observing how eigenplanes of rotations transform under triality.
23.1 Division algebras 301
23.1 Division algebras
An algebra A over R is a linear (that is a vector) space A over R together with
a bilinear map A x A + A, (a, b) + ab, the algebra product. Bilinearity means
+ +
distributivity (a b)c = ac+ bc, a(b c) = ab+ ac and (Xa)b = a(Xb) = X(ab)
for all a , b, c E A and X E R. An algebra is without zero-divisors if ab = 0
implies a = 0 or b = 0. In a division algebra D the equations ax = b and
ya = b have unique solutions x, y for all non-zero a E D. A division algebra is
without zero-divisors, and conversely, every finite-dimensional algebra without
zero-divisors is a division algebra. If a division algebra is associative, then it
has unity 1 and each non-zero element has a unique inverse (on both sides).
An algebra with a unity is said to admit inverses if each non-zero element
admits an inverse (not necessarily unique). An algebra is alternative if a(ab) =
a2b and (ab)b = ab2, and flexible if a(ba) = (ab)a. An alternative algebra is
flexible. An alternative division algebra has unity and admits inverses, which
are unique. The only alternative division algebras over R are R , C, W and 0.
An algebra A with a positive-definite quadratic form N : A + R , is said to
preserve norm, or admit composition, if for all a , b E A, N (ab) = N (a)N (b).
The dimension of a norm-preserving division algebra D over R is 1, 2, 4 or 8;
if furthermore D has unity, then it is R , C, W or 0.
Examples. 1. Define in C a new product a o b by a o b = ab. Then C becomes
a non-commutative and non-alternative division algebra over R , without unity.
2. Consider a 3-dimensional algebra over R with basis (1, i,j) such that 1 is
the unity and i2 = j2 = -1 but i j = j i = 0. The algebra is commutative and
flexible. but non-alternative. It admits inverses. but inverses of the elements of
+
the form xi y j are not unique, (xi + ~ j ) - = +
ix jy where
l X ( Y ~- x j ) - -
+
x2 y2 '
X E R. It has zero-divisors, by definition, and cannot be a division algebra,
although all non-zero elements are invertible.
3. Consider a 3-dimensional algebra over Q with basis (1, i, i2), unity 1 and
multiplication table
The algebra is commutative and flexible, but non-alternative. Each non-zero

+ +
element has a unique inverse. Multiplication by x iy i2z has determinant
+
x3 3Y3- 18z3, which has no non-zero rational roots (Euler 1862). Thus, the
algebra is a division algebra, 3 0 over Q. I
302 Octonions and Triality
23.2 The Cayley-Dickson doubling process
Complex numbers can be considered as pairs of real numbers with cornponent-
wise addition and with the product
Quaternions can be defined as pairs of complex numbers, but this time the
product involves complex conjugation
Octonions can be defined as pairs of quaternions, but this time order of multi-
plication matters
This doubling process, of Cayley-Dickson, can be repeated, but the next al-
gebras are not division algebras, although they still are simple and flexible
(Schafer 1954). Every element in such a Cayley-Dickson algebra satisfies a
quadratic equation with real coefficients.
Example. The quaternion q = w+ix+ jy+kz satisfies the quadratic equation
The Cayley-Dickson doubling process
provides a new imaginary unit C, C2 = -1, which anticommutes with i, j, k.

The basis (1, i, j, k) of W is complemented to a basis (1, i, j, k, C, it, j C , kt)
of 0 = IHI $ WC. Thus, 0 is spanned by 1 E R and the 7 imaginary units
i, j, k, C, iC, jC, kt, each with square -1, so that 0= R$R7.Among subsets of
3 imaginary units, there are 7 triplets, which associate and span the imaginary
part of a quaternionic subalgebra. The remaining 28 triplets anti-associate.
The multiplication table of the unit octonions can

be summarized by the Fano plane, the smallest pro-
jective plane, consisting of 7 points and 7 lines, with
orientations. The 7 oriented lines correspond to the
7 quaternionic/associative triplets.
jC k ie
23.3 Multzplzcation table of 0 303
23.3 Multiplication table of 0
Denote the product of a , b E 0 by a o b. Let 1 , e l , ez, . . . , e7 be a basis of 0.
Define the product in terms of the basis by
eioei=-1, and e i o e j = - e j o e i for i # j ,
and by the table
The table can be condensed into the form

If ei o ej = fe k , then ei, e j , ek generate a subalgebra isomorphic to IHI.
The sign in ei o ej = f e k can be memorized by rotating the triangle in the
following picture by an integral multiple of 2x17 :
e2
e5
Example. The product e2 o e5 = -e3 corresponds to a triangle obtained by
rotating the picture by 2x17. I
In the Clifford algebra Ceo,7 of R017,octonions can be identified with paravec-
tors, 0 = R $ R0p7, and the octonion product may be expressed in terms of
the Clifford product as
a o b = (ab(1- v))o,l,
+ +
where v = el24 e235 e346 e457 e561 + + + e672 + e713 E /\3R017.In Ce0,7,
the octonion product can be also written as
a o b = (ab(1 + w ) ( l - el2...7 ) ) 0 , l for a , b E R @ R017
-1
+
where g(l w) i ( 1 - el^...^) is an idempotent, w = ve;;,,, E A ~ Rand ~ J ~
el2...7 = el2...7.
In the Clifford algebra Ct.8 of R8, we represent octonions by vectors, O = R8.
As the identity of octonions we choose the unit vector e8 in R8. The octonion
product is then expressed in terms of the Clifford product as
a o b = (aesb(l+ w ) ( l - el2...8))l for a, b E R8

V ~ GE. A4R8,
where i ( 1 + w ) i ( l - el2...8) is an idernpotent, w = . ~ e ~ ; , .=~
+ + + +
-e 1 2...7 and v = el24 e235 e346 e457 e m + em2 + en3 E A ~ I R ~ .
23.4 The octonion product and the cross product in R7
A product of two vectors is linear in both factors. A vector-valued product of
two vectors is called a cross product, if the vector is orthogonal to the two
factors and has length equal to the area of the ~arallelogramformed by the
two vectors. A cross product of two vectors exists only in 3 0 and 7 0 .
The cross product of two vectors in R7 can be constructed in terms of an
orthonormal basis el, e2,. . ., e7 by antisymmetry, ei x ej = -ej x ei, and
The above table can be condensed into the form

This cross product of vectors in R7 satisfies the usual properties, that is,
(a x b) . a = 0, (a x b) . b = 0 orthogonality
la x bI2 = laI21bl2- ( a . b)' Pythagorean theorem
where the second rule can also be written as la x bl = lallbl sin <(a,b). Unlike
the 3-dimensional cross product, the 7-dimensional cross product does not
+ +
satisfy the Jacobi identity, (a x b) x c (b x c) x a (c x a) x b # 0, and
so it does not form a Lie algebra. However, the 7-dimensional cross product
satisfies the Malcev identity, a generalization of Jacobi, see Ebbinghaus et al.
1991 p. 279.
In R3, the direction of a x b is unique, up to two alternatives for the ori-
entation, but in R7 the direction of a x b depends on a 3-vector defining the
23.4 The octonion product and the cross product in IR7
cross product; to wit,
+ + + +
depends on v = elad e235 e346 e457 e561 e672 e713 E /\3 R7. In + +
the 3-dimensional space a x b = c x d implies that a, b , c, d are in the same
plane, but for the cross product a x b in R7 there are also other planes than
the linear span of a and b giving the same direction as a x b.
The 3-dimensional cross product is invariant under all rotations of S 0 ( 3 ) ,
while the 7-dimensional cross product is not invariant under all of S0(7), but
only under the exceptional Lie group G2, a subgroup of SO(7). When we let
a and b run through all of IR7, the image set of the simple bivectors a A b is a
manifold of dimension 2-7-3 = 11 > 7 in /\' R7, dim(/\' IR7) = &7(7- 1) = 2 1,
while the image set of a x b is just R7. So the mapping
is not a one-to-one correspondence, but only a method of associating a vector

to a bivector.
The 3-dimensional cross product is the vector part of the quaternion product
of two pure quaternions, that is,
a x b = Im(ab) for a, b E IR3 c W.
In terms of the Clifford algebra Cl3 E Mat(2, C) of the Euclidean space R3
the cross product could also be expressed as
axb=-(abe123)1 for a , b ~ ~ ~ c ~ t ~ .
In terms of the Clifford algebra Cto,32 W x W of the negative definite quadratic
space R0s3 the cross product can be expressed not only as
a x b = -(abelns)l for a, b E IR013 c CtoS3
but also as
a x b = (ab(1 - e123))i for a, b E IR013 c Ceo,3.

Similarly, the 7-dimensional cross product is the vector part of the octonion
product of two pure octonions, that is, a x b = (a o b)l. The octonion algebra
0 is a norm-preserving algebra with unity 1, whence the vector part R7 in
0 = R $ R7 is an algebra with cross product, that is, a x b = $(a o b - b o a )
for a, b E R7 C 0 = R $ IR7. The octonion product in turn is given by
1 This expression is also valid for a, b E R3 C C e 3 , but the element 1 - e l 2 3 does not pick
up an ideal of C 4 3 . Recall that C t 3 is simple, that is, it has no proper two-sided ideals.
+ +
for a = cr a and b = p b in R $ R7. If we replace the Euclidean space R7
by the negative definite quadratic space Rot7, then not only
aob=crp+crb+ap+a.b+ax b
for a, b E IR $ I R 0 s 7 , but also
a o b = (ab(1 - v))oBl
where v = e l 2 4 + e235 + e346 + e457 + e561 + e672 + e713 E /\3 ~'9~.
23.5 Definition of triality

Let n 2 3. All automorphisms of SO(n) are of the form U + S US-I where
S E O(n). All automorphisrns of Spin(n), n # 8, are of the form u + sus-'
where s E Pin(n). The group Spin(8) has exceptional automorphisms, which
permute the non-identity elements - 1, e12,..8,-e12.,,8 in the center of Spin(8) :
Such an automorphism of Spin(8), of order 3, is said to be a triality automor-

phism, denoted by trial(u) for u E Spin(8).
Regard Spin(8) as a subset of Ct8. In Ct8, triality sends the lines through
1, -el2...8 and -1, e12,,,8,which are parallel, to the lines through 1, -1 and
el2...8, -e12,..8, which intersect each other. Thus, a triality automorphism of
Spin(8) cannot be a restriction of a linear mapping Ct8 + Cts.
A non-linear automorphism of Spin(8) might also interchange -1 with ei-
ther of fe12...8 . Such an automorphism of Spin(8), of order 2, is said to be a
swap automorphism, denoted by swap(u) for u E Spin(8).
On the Lie algebra level, triality acts on the space of bivectors /\'IR8, of
dimension 28. Triality stabilizes point-wise the Lie algebra G2 of G2, which is
the automorphism group of 0. The dimension of is 14. In the orthogonal
complement G i of 6 2 , triality is an isoclinic rotation, turning each bivector
by the angle 120°. A swap stabilizes point-wise not only G2 but also a 7-
dimensional subspace of 621, and reflects the rest of the Lie algebra so(8) D4,
that is, another 7-dimensional subspace of G i . For a bivector F E IR8, we
denote triality by Trial(F) and swap by Swap(F).
On the level of representation spaces, triality could be viewed as permuting
the vector space IR8 and the two even spinor spaces, that is, the minimal
left ideals ~t8+ +
i ( 1 w)3(1 f e12,,,8), which are sitting in the two-sided ideals
23.5 Definition of triality 307
C l i i ( l &e12...8)2 Mat(8, R) of ~ lEi2 ~ a t ( 8R)
,. This means a 120' rotation
of the Coxeter-Dynkin diagram of the Lie algebra Dq :
Rather than permuting the representation spaces, triality permutes elements

of Spin(8), or their actions on the vector space and the two spinor spaces.
Because of its relation to octonions, it is convenient to view triality in terms
of the Clifford algebra C e o ~2 2Mat(8,R), the paravector space $R8 = R@R0p7,
having an octonion product, the spin group
$pin(8) = {u E Ce0,7 I uG = 1; for all x E $R8 also uxii-l E $R8),
+
the minimal left ideals Ce0,7i(l w ) $ ( l F el2...7) of Ceot72:'~at(8,IW), and
the primitive idempotents
For u $pin(8), define two linear transformations U1, U2 of $R8 by
(1 + w) of CeoB7results in the matrix

The action of u on the left ideal Ceo,7&
representation
Qin(8) 3 u E (2 ) where Ul ,UZ E $0 (8).
For U E S0(8),
-
define the companion u by
~ ( x =) U (f) for all x E $R'.
. f,
2 Choose the bases (elj,ez j, . . ,e7f, j) for Ce0,7 j and (elf,ez . . . , e7f, j) for ~ ~ 0 f,
+ + +
where j = i ( 1 w ) $ ( l - e12...7) and f^ = i ( 1 w ) $ ( l e l 2 ...7). Then the matrices
of Ul and U2 are the same as in the basis ( e l , e2,. . .,e 7 , 1) of $R'. Denoting j; = e , j,
..
i = 1 , 2 , . , 7 , and j8 = j , (Ul)i1 = 16(fiujl)0, and denoting gi = e i f , i = 1 , 2 , . .., 7,
and ga = f, (U2)ij = 1 6 ( g i ~ g j ) 0 .
. A,
3 If we had chosen the bases (11, j2,. .,j7, j) for Ceo,7 j and ( f l , f z , . . . , f) for ~ ~ 0 , 7 f ^ ,
where ji = e i j and ji = -e;f^ for i = 1 , 2 , . . . , 7 , then we would have obtained the
following matrix representation
where Ul (x) = 16(uxf ) o , l as before but w2(x) = 16(6xf ) o , l . This representation is used
by Porteous 1995,
4 Or, for U E Mat(8,R).
The companion ii of u E $pin(8) is just its main involution, fi = 6, and
corresponds to the matrix
For a paravector a E $R', define the linear transformation A of $R' by

A(x) = 16(axf)o,l, that is, A(x) = a 0 x,
making $Rs the Cayley algebra 0. Since AT(x) = l 6 ( a ~ f ) ~we
, ~ have
, the
correspondence
a 2 (t lT)
, abbreviated as a - A.
Computing the matrix product
U (a) = uau-' = (Ul0 O ) ( AO

U2 O)(Uil
AT
Ul-I
0 )'
we find the correspondence U(a)
operate on x E $R8, to get
- U~AWT'.Denote Uo = i?, and let u0(a)
uo (a) o x = (u~Au;') (x) = Ul (a o U; '(1)).

The ordered triple (Uo,Ul , U2) in SO(8) is called a triality triplet with respect
to the octonion product of 0.
If (Uo, Ul , U2) is a triality triplet, then also (Ul , U2, Uo), (U2, UO,Ul) and
( u ~Ul,, UO) are a triality triplets. This results in
UO(Xo y) = Ul (x) o Uz(y) for all x, y E 0 = $R',
referred to as Cartan's principle oftriality. Conversely, for a fixed Uo E $0(8),
the identity u0(x o y) = Ul(x) o U2(y) has two solutions U1, U2 in $0(8),
resulting in the triality triplet (Uo, Ul, U2) and its opposite (Uo, -UI , -U2).
5 Recall that for x E $R8 = W $ IW0s7, U ( x ) = U X G - ~and
, so ~ ( x=) G ~ u - ' .
6 The matrix of A can be computed as A;j = 1 6 ( f ; a J j ) o .The paravector a = a0 +ale1 +
+
. . a7e7 has the matrix
23.5 Definition of triality 309
Thus, Uo corresponds to two triality triplets (Uo,Ul, U2) and (UO,-U1, -U2),
while -Uo, corresponds to (-Uo , -U1, U2) and (-Uo, Ul , -U2).
The rotations Ul, U2 E $O(8) are represented by fu l , fu2 E $pin(8). We
choose the signs so that
Uo 21 (h
0 o),
u2 ..
U1 N (h
0 o),
u0 ..
Ug N 0 u1
(h 0 )
where uo = u and Uo = U . Using the notion of triality triplets,
The rotation Uo in $0(8) corresponds to uo cz (Uo, U1, U2) and its opposite
-uo 2: (UO,-U1, -U2) in $pin(8), and the opposite rotation -Uo corresponds
to el2...7uo cz (-Uo, -Ul, U2) and -el2...7uo 2: (-UO, U1, -U2). Triality is
defined as the mapping
trial : $pin(8) + $pin(8), ul N

0 li,
Triality is an automorphism of $pin(8); it is of order 3 and permutes the non-
identity elements -1, e12,,,7,-e12...7 in the center of $pin(8).
E x a m p l e . Take a unit paravector a E $R' = IR $ I R 0 t 7 , la1 = 1. The action
x + axii-l is a simple rotation of $R'. Thus, a E $pin(8). Denote a0 = ii,
a1 = trial(ao) and a2 = trial(a1) so that 16(iilxf)o,l = a2xiiz1, 16(iizxf)o,l =
alxii,'. Then
a o x = alxiill and x o a = a2xiiz1
represent isoclinic rotations of $R8. Left and right multiplications by a E S7C
0 are positive and negative isoclinic rotations of $R8 = 0. The Moufang
identity
results in a special case of Cartan's principle of triality
In this special case, ao, a l , a2 commute, a2 = Zil (= 6,') and aoala2 = 1

implying a = ala2 = aliill = a2ay1. I
7 Note that a o x o a = ax;, ~5= ci-' and a o x o a-' = SXS-' where s = a1 a;' E $pin(7).
8 Any four mutually orthogonal invariant planes of an isoclinic rotation of $R8 induce the
same orientation on $R8.
9 To prove Cartan's principle of triality, in the general case, iterate the Moufang identity,
like b o a o ( x o y ) o a o b = ( b o ( a o x ) ) o ( ( y o a ) o b ) . Observe thenesting b o ( a o x ) = sx8-',
where s = trial(b)trial(ic) = trial(ba).
Triality sends a simple rotation to a positive isoclinic rotation and a positive
isoclinic rotation to a negative isoclinic rotation. The isoclinic rotations can
be represented by octonion multiplication having neutral axis in the rotation
plane of the simple rotation:
positive isoclinic rotation
7
simple rotation 4-
negative isoclinic rotation
23.6 Spin(7)
Let uo E Spin(7) c C18, and ul = trial(uo), u2 = trial(u1). Then uz = 41,
that is, trial(trial(u0)) = e8trial(uo)e;l. l o Thus, u14i1 = 1 and u l u z l =
-1 -1
ule8ul e, E IR $ IR7e8, being a product of two vectors, represents a simple
rotation. l1 l 2 Since WO = Uo, U2 = ~ 1 ,
Comparing matrix entries of ulu;luz, we find ulu;lu2 E Spin(7) and so

u~u['u~E SO(7).
Let therotationanglesof Uo E SO(7) be oo,Po,yo so that cro 2 Po 2 y~3>
- 0.
Then the rotation angles of U1 E SO(8) are
I f f l = ;(a0 +Po + yo)

P1 = ;(a0 +Po - yo)
71 = $ ( a 0 - Po + yo)
61 = $(a0 -Po -yo).
Since eigenvalues change in Uo + Ul, triality cannot be a similarity, Ul #
SUoS-l. Represent the rotation planes of Uo E SO(7) c SO(8) by unit bivec-
tors A o , B o , Co, and choose the orientation of D o = ueg, u E IR7, IuI = 1 SO
that A. A B o A C o A D o = el2,,,,. The rotation planes of Ul can be expressed
as unit bivectors l3
A1 = $Trial(Ao + B o Co - Do) +
B1 = ;Trial(Ao +
Bo - Co Do) +
C1 = iTrial(A0 - Bo Co Do) + +
D l = ;T'rial(Ao - Bo - C o - Do).
10 Note that trial(trial(u1)) # e a t r i a ~ ( u ~ ) trial(trial(u2))
e~~, # estrial(u2)e;l.
11 In C I ! ~ , ~u2, = ti1, and so 2~10;~= 1, but 2 ~ 1 2 ~ ; E~ R $ R09'.
12 Recall that for u E Spin(8), u-' = G, and for u E $pin(8), u-' = G.
13 Trial : 1\2 R8 -+ 1\2R8 sends negative isoclinic bivectors to simple bivectors.
23.7 The exceptional Lie algebra G2
The rotation angles and planes of U2 are
The rotation planes of Uo, U1, U2 induce the same orientation on lR8, that is,
For uo E Spin(7), ul, u2 E Spin(8), so that
23.7 The exceptional Lie algebra G2

A rotation U E SO(7) such that
U (x o y) = U(x) o U(y) for all x, y E 0
is an automorphism of the Cayley algebra 0. The automorphisms form a group
G2 with Lie algebra G2 C Ilk8, dimg2 = 14. A bivector U E G 2 acts on the
octonion product as a derivation
The double cover of G2 C SO(7) in Spin(7) consists of two components, G 2

and -G2. The groups G 2 and G2 = p(G2) are isomorphic, G2 2 G2. l 4
A rotation Uo E G2 C SO(7) has only one preimage in G 2 c Spin(7), say
uo, p(uo) = Uo. Since trial(u0) = uo, ul = trial(uo) equals uo, and U1 = p(u1)
equals Uo. The rotation angles ao,Po,yo of Uo, such that ao 2 Po 2 yo 2 0,
satisfy the identities
f f l = ;(a0 + Po +yo) = ffo
P1 = +o1 +Po -yo) =Po
71 = Z ( f f-~PO+ 70) = 70
1
61 = $ ( a 0 -Po- yo) = 0
each of which implies
ffo =Po +yo.
14 Note that - I 4 G2 because - 1 4 SO(7) and -1 4 G2 because triality stabilizes point-
wise G2 but sends -1 to fe12...8 .
This can also be expressed by saying that the signed rotation angles cr, P, y of
U 6 Gz satisfy
Represent the rotation planes of U by unit bivectors A , B , C and choose

+
orientations so that u = e x p ( $ ( c r ~+ PB yC)), when U = p(u). Then
A J w = B + C. Conversely, for an arbitrary rotation U 6 SO(7) to be in G2
it is sufficient that
AJw=B+C and cr+P+y=O.
In order to construct a bivector U E G2, pick up a unit bivector A E /\'IR7,
A 2 = -1, decompose the bivector A J w into a s u m of two simple unit bivectors
B + C (this decomposition is not unique), choose cr, 0,y E R so that a+P+ y =
+
0, and write U = crA P B yC. +
For u E G 2 , trial(u) = u, and for U E G2, Trial(U) = U , in other words,
triality stabilizes point-wise G 2 and G2. Multiplication by u E G 2 stabilizes
+ +
the idempotent z( 1 w ) , u i ( 1 w) = ; ( I + w)u = :(I +
w ) , while a
+
bivector U E g2 annihilates i ( l +w), ~ : ( 1 +w) = $(I W)U = 0, and thus
~ i ( -7w ) = k(7 - w ) U = U. Conversely, a rotation U E SO(7) is in G2 if
it fixes the 3-vector
+ +
for which w = v e ~ l .=, e~l 2 3 6 - el257 - el345 e l 4 6 7 e2347 - e2456 - e3567.
A bivector F E /\' R8, dim(/\' R8) = 28, can be decomposed as
1
F=G+H where G E G 2 and H = - W J ( W A F ) E G $ .
3
+
Under triality, F goes to Trial(F) = G Trial(H), Trial(H) E Gf, where the
angle between H and Trial(H) is 120'. In ~articular,triality is an isoclinic
rotation when restricted to 621, dim(Gk) = 14.
+
A bivector F E /\' R7 can be decomposed as F = G H , where G E G 2 ,
For a vector a E R7, v L a E ~ 2 n/\2

1 R7. The mapping a + v L a is one-to-one,
since a = ~ V ( Lv ~ a )The
. element u = exp(vLa) E Spin(7) induces a rotation
I
of R7, which has a as its axis and is isoclinic in aL = {x E R7 x . a = 0). A
miracle happens when la1 = 2x13. Then the rotation angles of U = p(u) are
4x13, which is the same as 4x13-2x = - 2 ~ 1 3in the opposite sense of rotation.
For the signed rotation angles we can choose cr = 4x13, /3 = y = - 2 ~ 1 3which
23.8 Components of the automorphism group of Spin(8) 313
+ + +
satisfy cr /3 y = 0. Since also A J w = B C , it follows that u E G 2 .
Therefore, u = exp(v L a ) , where a E JR7 and la1 = 2 ~ 1 3 belongs
, to
+ +
Note that cr /3 y = 0 in G 2 while cr = /3 = y in exp(G: n R7). An
element u E G 2 n exp(Gi n JR7) can be also constructed by choosing a unit
bivector A E R7, A' = - 1, decomposing A J w = B C , constructing +
bivectors
2
27r 27r
-(2A-B-C)=G€G2 and -(-A-B-c)=HEG:~AJR~
3 3
and exponentiating
= ,G = ,H = -1 + . . .
8
The elements u are extreme elements in G 2 in the sense that ( u ) ~= - while i,

for allother g E G z , (g)o > -:.
The elements u = exp(v L a), where a E R0p7 and la1 = 2 ~ 1 3 ,satisfy
u3 = 1, and they are the only non-identity solutions of u3 = 1 in G2. The
octonion a = eoa (= ea) satisfies ao3 = 1 and aO-I o x o a = UXU-' for all
x E $R8 = R $ R017. Conversely, the only unit octonions a E S7 c 0 = $Rs
satisfying
a'-' o (x o y) o a = (a0-' o x o a) o (a0-' o o a) for all x, y E $ R ~
are solutions of ao3 = f1.
23.8 C o m p o n e n t s of the a u t o m o r p h i s m g r o u p of Spin(8)

In general, the only exterior automorphisms of Spin(n), n # 8, are of type
u + sus-' , where s E P i n ( n ) \ Spin(n). Thus, Aut(Spin(n))/Int(Spin(n)) =
Z2, when n # 8. However, in the case n = 8, the following sequence is exact
that is, Aut(Spin(8))/Int(Spin(8)) !x S3, a non-commutative group of order

6.
For u E Spin(8), denote swapl (u) = e8trial(u)ei1 = trial(trial(e8uei1))
and swap,(,) = e8trial(trial(u))ei1= trial(esueil). Then trial, swapl, swap2
generate S3 :
trial o trial o trial = swapl o swapl = swap2 0 swap2 = identity
swap, o swap2 = trial and swap2 0 swap, = trial 0 trial-
The automorphism group of Spin(8) contains 6 components, represented by
the identity, trial, trial o trial, swapl, swap2 and the companion. l5 In the
component of trial, all automorphisms of order 3 are trialities for some octonion
product.
23.9 Triality is q u a d r a t i c
Triality of u E Spin(8) C Ce8 is a restriction a polynomial mapping C18 -,Ce8,
of degree 2,
trial(u) = triall (u)trial2(u)
+
triall(u) = + ( I +el2 ...8)[(u(1 w)(1 - el2...8))0,6 A e8]ei1
+;(I - el2...8)
tria12(u) = (w - 3)[(u(l+ el^...^)) A e8]ei1(w - 3)-'.
The first factor is affine linear and the second factor is linear. To verify that
trial is a triality, it is sufficient to show that it is an automorphism of order 3
sending - 1 to e12...8.
In the Lie algebra level, the triality automorphism of a bivector F E R8
is
+ +
Trial(F) = e8(F - i F ( 1 w ) ( l e12.,.8))2ei1
= ;e8(F - FJ w - ( F A w) ~ e ~ ~ . . . 8 ) e i l .
The triality automorphism of a para-bivector F E R0s7$ A2 Rot7 is
For u E $pin(8), triality is

trial(u) = triall (u)tria12(u)
+
triall(u) = i ( 1 - el2 ...7 ) ( ~ ( l w ) ( l + el2...7))o,s
+i( 1 + e l 2 ...7)
tria12(u) = (w - 3) even(u(1 - el^...^)) (W - 3)-'.
23.10 Triality i n t e r m s of eigenvalues a n d invariant planes

Triality can also be viewed classically, without Clifford algebras, by inspection
of changes in eigenvalues and invariant planes of rotations. Consider Uo E
SO(8) and a triality triplet (Uo,Ul, Uz). Let the rotation angles cro, Po,yo, 60
15 The subgroup of linear automorpl~ismscontains 2 components, represented by the identity
and the companion, u + e s u e l l .
23.11 Trialities with respect to different octonion products 315
of Uo be such that (YO 2 Po 2 yo 2 60 2 0. Represent the rotation planes of
Uo by the unit bivectors Ao, Bo,Co,Do. Then the rotation angles of Ul and
U2 are
ai = $(ao+Po +yo -6)
Pi = $(ao+Po - yo + 6 )
71 = $(a0 - Po +yo + 6 )
61 = $((yo-Po -yo -6)
and { a2=
P2
$((YO
= $((yo
72 = $((yo
62 = $(-(Yo
+Po + yo + 6)
+ Po - YO - 6)
-Po +yo -6)
+Po + yo - 6)
and the rotation planes are
+
A1 = $~rial(AO Bo Co - Do) +
+
B1 = $Trial(Ao Bo - Co D O ) +
C1 = $Trial(Ao - Bo Co DO) + +
D l = $Trial(Ao - Bo - Co - Do)
and
+ + +
A2 = $Trial(~rial(A~Bo Co Do))
+
B2 = $Trial(Trial(Ao Bo - Co - Do))
+
C2 = $Trial(Trial(Ao - Bo Co - Do))
D2 = ;Trial(Trial(-A. + +
Bo Co - Do)).
23.11 Trialities with respect t o different octonion products

+
An arbitrary 4-vector w E /\4 IRs, for which i ( l w) is an idempotent in ~.
is called a calibration. l 6 A direction n E IRs, In1 = 1, fixed by the calibration,
w n = nw, is called the neutral axis of the calibration. A calibration together
with its neutral axis can be used to introduce an octonion product on R8 and
a triality of Spin(8).
Let w l , w:, be calibrations, with neutral axes n l , n2 E IRs. Denote the octo-
nion products by
+
x Owl,nl Y = (xnly(1 w i ) ( l - el2...s))i,
x ow2,n2Y = (xnzy(l+ wz)(l - el2...s))i
and the trialities by trial,, ,,, (u) , trial,,,,, (u). Denote the opposite of the
composition of the trialities by tria1,,,,,,,(u) so that
tria1WlZ,nl2 (u))= trial,, ,, (trial,,,,,
(tria1W12,,l12 (u)).
Then
1
w12 = Z +
( ~ lw2) $(-WI + +
w2)elz ...a
+
= $(1 - el2...8)wl :(I+ el2...8 ) ~ 2
16 Note that w satisfies w2 = 7 + 6w.

and
n12
Y for a non-zero y = x - a w l 2 L (w12 L x), where x E R8
=-
IYI
23.12 Factorization of u E Spin(8)
S
Take u E Spin(8) \ Spin(7). Denote s = (u A eg)egl, and 217 = -. Then
Is1
u7 E Spin(7), and u = ugU7 = u ~ u ; , where ug, uk E R @ R7eg. l7 These
factorizations are unique, up to a sign (-1 is a square root of 1 in Spin(7)) :
The following factorizations are unique, up to a cube root of 1 in G 2 :
where
h: = u7trial(u7)-l u7trial(trial(u7))-l,
h13 - trial(u7)-'u7triaI(trial(u7))-'u7,
and
hl = hog, h2 = hog2, hi = g'hb, h i = gr2hb,
2n L h
g = exp(-p d),
g r = e x p (2ny v L hi
ho = ( H o A Ho A Ho)el2...7, h6 = (Hb A Hb A Hb)e12...7~
h o = e H 0 , hb=eHb.
In this factorization, go, gl, g2 E G 2 and ho, hl, h2, hb, h i , h i E e ~ ~ ( & j - nIR7)
/\~
and g, g' E G 2 n exp(G$ n R7) 2: S 6 . These factorizations are unique, up
to a factor g,g1 E G 2 , g3 = g'3 = 1. l8
Appendix: C o m p a r i s o n of formalisms i n R8 and R @ R017

We use the 3-vector
v = el24 + e235 + e346 + e457 + e56l + e672 + e713
in /\3R8 or / \ 3 ~ 0 9 7 , and the 4-vector w = ve;;,,, in /\4R8 or /\ 4 R 0,7,
w = e l 2 3 6 - e l 2 5 7 - el345 + el467 + e2347 + e2456 - e3467.
1
Note that e ~ ; , ,=
, ~-e12...7 in Clg while e;2,,.7 = e 1 2 , , , 7 in Clo,7.
17 Note that us =
18 Recall that -1
-e4 Gz.
and u( = J-.
23,12 Factorization of u E Spin(8) 317
We use the octonion product
x o y = ( x e s y ( l + w ) ( l - e l 2 ...8))l for vectors x , y ERs,
x o y = (xy(l + w ) ( l - e12,,,7))o,l for paravectors x, y E R $ RO1'
Note that in C!0,7, also x o y = (xy(1 - v))o,l.
+
The bivector F = A B E A2 Rs, with B E A2 R7 and A = ae8, a E R7,
corresponds to the para-bivector F = a - B E R 0 * 7 @R017,
~ 2 with a = ~ e g El
+
R0l7. Let u = u+ u-es E Spin(8), where u* E c!?. Then u E Spin(8)
corresponds to
The companion 4 of u is
4 = esueil for u in Spin(8) or c!$ (or C!8),
4 =G for u in $pin(8) or C!o,7.
For uo, ul = trial(uo), u2 = trial(ul), Cartan's principle of triality says
In the Lie algebra level, Freudenthal's principle of triality says l9
The non-identity central elements of the Lie groups are permuted as follows
Exercises
Show that
1. For U E 62, U J w = -U, UW = -U, U J U = -IUI2, IU A UI = IUI2.
+
2. For Uo E /\2R8, U1 = Trial(Uo), U2 = Trial(U1), Uo U1 + U2 E 6 2 ,
2Uo - U1 - u2€6;.
+
3. In Ce0,7, (1 w ) ( l - e12.,.7)= (1 - el24)(1 - e235)(1- e346)(1 - e457),
+
(1 w)(l - e12,..7)= (1 - v ) ( l - el2...7).
4. w 2 = 7 + 6 w .
-1)+1+Q(7"-l)w, neven.
8 ( 7 n + 1 ) - 1 + i ( 7 n + 1 ) ~ , nodd.
19 Note that x J F corresponds to (zJ P)o,I.
6. For x E R , f (xw) = f (-x)i(7 - w ) + f ( 7 x ) i ( l + w). Hint: the minimal
polynomial of w , x2 = 7 + 6x, has roots x = -1, x = 7.
7. For G E 62, i ( 1 + w ) ~ i ( l w + ) = 0, i ( 7 - w ) ~ $ ( -
7 w) = G ,
eG = 1 ,(7-w)e g( ~ - W ) + ~ ( I + W ) .
+
8. v 2 = -7 - 6w in Clg, v 2 = 7 6w in Cl0,7.
9. For v E A~IW',errv= -1. Hint: v4 + 50v2 + 49 = 0, and
x4 + 50x2 + 49 = 0 has roots fi, f7i.
10. For v E /\3R0~7,COS(?TV) = -1, sin(?rv) = 0. Hint: v4 - 50v2 + 49 = 0,
and x4 - 50x2 + 49 = 0 has roots f7, f 1.
11. For F E / \ ~ R ' , F = G + H , G ~ 6 2 H, E 6 ; : H = ; W J ( W A F ) .
12. or HEG;, H = + W L ( W L H ) .
13. For uo E $pin(8), ul = trial(uo), u2 = trial(u1) : U O U ~ E~ $pin(7). U ~
Hint: a = u;'io E R $ and so a = trial(a)trial(a)-l.
.
lR017
14. For the opposite product x y = y o x of X, y E 0= lRg,

fio(x. y ) 4 i 1 = (u2xu,') (ulyuyl).
15. uo(x o y ) u i 1 = (fi2xiiy1) 0 (filYiill).
+
16. (fiox(1 w ) ( l - el2 ...7))o,i = u l x f i l l ,
( Q o z ( l + w ) ( l + el2 ...7))0,l = u 2 x ~ 2 ~ .
+
17. (Soxy(1 w ) ( l - el2 ...7))0,l = ul(x 0 y)fi;',
+ +
(fiOxy(1 w ) ( l el2 ...7))0,l = UZ(Yo x ) u ~ l .
18. For B E lR8 n Spin(8), (trial(B))O = $. For C E /\4 lR8 n Spin(8),
(trial(C))O = 0. For D E /\6 lR8 n Spin(8), (trial(D))o = -$.
19. For C E /\4 R8 fl Spin(8), trial((=) E /\4 R8.
20. For u E Spin(8), inducing a simple rotation U = p(u) :
. 8 (trial(trial(u)))gelz...a < 0.
( t r i a l ( ~ ) ) ~ e ~>~ 0, . and
21. (G2)0 2 -i, (trial(Spin(7)))o 2 -:.
-
22. G 2 n exp(6t n /\'R7) is homeomorphic but not isometric to s6.
23. diam(G2) = 1, 4;.
diam(trial(Spin(7))) =
24. Triality does not extend to an automorphism of Pin(8).
25. [$(w - 3)12 = 1.
Determine
26. The matrix of $(w - 3) in the basis (fl, f 2 , . . . , fa) of C l i f , where
fi = eie8f, i = 1 , 2,..., 8, and f = i ( l + w ) g ( l +e12...8).
Solutions
13. u,'Q0 = 0 ~ ~ u so~ 1 0= u~ ~ ~ Qu ~ ~ ~ ~ ~= ~( ~o f~ i ' ~~ ~~ ~Q ~~
which implies u o 0 ~ 1 = = (u0iiy1u2).
~ ii0u~1ii2
2
24. Triality sends -1 E Cen(Pin(8)) to el2 ...8 !$ Cen(Pin(8)).
23.12 Factorization of u E Spin(8)
Bibliography
J.F. Adams: On the non-existence of elements of Hopf invariant one. A n n . of Math.
72 (1960), 20-104.
J.F. Adams: Vector fields of spheres. Ann. of Math. 75 (1962), 603-632.
F. van der Blij: History of the octaves. Simon Stevin 34 (1961), 106-125.
C. Chevalley: T h e Algebraic Theory of Spinors. Columbia University Press, New
York, 1954. T h e Algebraic Theory of Spinors and Clifford Algebras. Spiinger,
Berlin, 1997.
G.M. Dixon: Division Algebras: Octonions, Quaternions, Complex Numbers and the
Algebraic Design of Physics. Kluwer, Dordrecht, 1994.
A.J. Hahn: Cayley algebras and the automorphisms of POL(V) and Pnb(V). Amer.
J. Math. 98 (1976), 953-987.
A.J. Hahn: Cayley algebras and the isomorphisms of the orthogonal groups over
arithmetic and local domains. J. Algebra 45 (1977), 210-246.
M.-A. Knus, A. Merkurjev, M. Rost, J.-P. Tignol: The Book of Involutions. American
Mathematical Society, Colloquium Publications 44, 1998.
W.S. Massey: Cross products of vectors in higher dimensional Euclidean spaces.
Amer. Math. Monthly 90 (1983), #lo, 697-701.
A. Micali, Ph. Revoy: Modules quadratiques. Cahiers MathCmatiques 10, Montpellier,
1977. Bull. Soc. Math. France 63, suppl. (1979), 5-144.
S. Okubo: Octonions and Non-associative Algebras i n Physics. Cambridge University
76 (1954), 435-446.
T.A. Springer, F.D. Veldkamp: Octonions, Jordan Algebras and Exceptional Groups.
G.P. Wene: A construction relating Clifford algebras and Cayley-Dickson algebras. J.
Math. Phys. 25 (1984), 2351-2353.
A History of Clifford Algebras
Clifford's geometric algebras were created by William K. Clifford in 1878/1882,

when he introduced a new multiplication rule into Grassmann's exterior al-
gebra /\ R n , by means of an orthonormal basis (el,en, . . . , e,) of Rn. Clif-
ford also classified his algebras into four classes according to the signs in
(ele2 . . . = f1 and (ele2 . . .e,)ej = fej(ele2.. .en). In the special case
of n = 3, Clifford's construction embodied Hamilton's quaternions, as bivec-
tors i = eze3, j = esel, k = ele2. Clifford algebras were independently redis-
covered by Lipschitz 1880/1886, who also presented their first application to
geometry, while exploring rotations of Rn, in terms of Spin(n), a normalized
subgroup of the Lipschitz group I'i.
+
Spinor representations of the orthogonal Lie algebras, Bl = so(2n 1) and
Dl = so(2n), were observed by E. Cartan in 1913, but without using the term
"spinor". Two-valued spinor representations of the rotation groups SO(n)
were re-constructed recursively by Brauer& Weyl in 1935, but without using
the term "Clifford algebra".
In the Schrodinger equation, Pauli 1927 replaced lr2 = ii. ii,where ii =
-inv - eA, by
?2=?.?+;~ii,
where the exterior part does not vanish: (ii A ii)$ = -heI?$. Pauli explained
interaction of a spinning electron with a magnetic field I? by means of a spinor-
valued wave function 11, : IW3 x R + C2. Thus, besides tensors, nature required
new kinds of objects: spinors, whose construction calls for Clifford algebras.
Clifford algebras are not only necessary but also offer advantages: their mul-
tivector structure enables controlling of subspaces without losing information
about their orientations. Physicists are familiar with this advantage in the
special case of 1-dimensional oriented subspaces, which they manipulate by
vectors, not by projection operators, which forget orientations.
23.13 Algebras of Hamilton, Gmssmann and Clifford 32 1
23.13 Algebras of Hamilton, Grassmann a n d Clifford
The first step towards a Clifford algebra was taken by Hamilton in 1843 (first
published 1844), when he studied products of sums of squares and invented his
quaternions while searching for multiplicative compositions of triplets in R3.
The present formulation of vector algebra was extracted from the quaternion
product of triplets/vectors x y = - x . y + x x y by Gibbs 1881-84 (first published
1901). Hamilton regarded quaternions as quotients of vectors and wrote a
rotation in the form y = ax using a unit quaternion a E W, la1 = 1. In such a
rotation, the axis a had to be perpendicular to the vector x (which turned in
the plane orthogonal to a). Cayley 1845 published the formula for rotations
about an arbitrary axis a E R3 by angle a = la1 (Cayley ascribed the discov-

ery to Hamilton). Cayley thus came into contact with half-angles and spin
representation of rotations in R3. However, in 1840 Olinde Rodrigues had
already recognized the relevancy of half-angles in his study on the compo-
sition of rotations in R3. Cayley 1855 also discovered the quaternionic rep-
resentation q -+ aqb of rotations in R4, equivalent to the decomposition
Spin(4) E Spin(3) x Spin(3).
The quaternion algebra W is isomorphic to the even Clifford algebras ~ t :2:
on R3. Note the algebra isomorphisms Ct3 E Mat(2, C) and Cto,3 W $ W. -

ctt3 as well as to the proper ideals Cto,33(lf e123) of Ct0,3. Hamilton also
considered complex quaternions C @I W, isomorphic to the Clifford algebra Ct3
Bivectors were introduced by H. Grassmann, when he created his exterior

algebras in 1844. The exterior product of two vectors, the bivector a A b , was
interpreted geometrically as the parallelogram with a and b as edges, and two
such exterior products were equal if their parallelograms lay in parallel planes
and had the same area with the same sense of rotation (from a to b). Thus the
exterior product of two vectors was anticomrnutative, a A b = -b A a. Using
a basis (el, e2, . . ., en) for Rn, the exterior algebra /\Rn had a basis
and was thereby of dimension 2".

W. K. Clifford studied compounds (tensor products) of two quaternion alge-
bras, where quaternions of one algebra commuted with the quaternions of the
322 A History of Cliflord Algebras
other algebra, and applied exterior algebra to grade these tensor products of
quaternion algebras. Clifford coined his geometric algebra in 1876 (first pub-
lished in 1878). Clifford's geometric algebra was generated by n orthogonal
unit vectors el, e 2 , . . . , en which anticommuted eiej = -ejei (like Grass-
mann) and satisfied all ef = -1 (like Hamilton) [or then all e: = 1 as in
Clifford's paper 18821. The number of independent products eiej = ej A ej,
i < j, of degree 2 was +n(n - 1) = (;). Clifford summed up the numbers of
independent products of various degrees 0, 1, 2, . . . , n and thereby determined
the dimension of his geometric algebra to be
Clifford distinguished four classes of these geometric algebras characterized by

the signs of (ele2. . .en)ei = fei (ele2. . .e n ) and (ele2 . . .e n ) 2= f1. He also
introduced two algebras of lower dimension 2"-l, namely, the subalgebra of
even elements and, for odd n, a reduced (non-universal) algebra obtained by
putting ele2 . . . e n = f1 [instead of letting ele2.. .en = el A e 2 A . . . A en be
of degree n with (ele2 . . .en)2= 11.
23.14 Rudolf Lipschitz

Clifford's geometric algebra was reinvented in 1880, just two years after its
first publication, by Rudolf Lipschitz, who later acknowledged Clifford's prior
discovery in his book, see R. Lipschitz: Untersuchungen uber die Summen von
Quadraten, 1886. In his study on sums of squares, Lipschitz considered a repre-
sentation of rotations by complex numbers and quaternions and generalized this
to higher dimensional rotations in Rn . Lipschitz thus gave the first geometric
application of Clifford algebras in 1880. He expressed a rotation
(written here in modern notation with an antisymmetric matrix A) in the form

y - A y = x + A x or as y + y J B = x + B L x where B E /\2Rn is the bivector
determined by Ax = B L x ( = -x J B). Lipschitz rewrote (I- A)y = ( I A)x +
using an even Clifford number a E cttn in the form y a = a x , thus representing
the rotation as
y = axa-l, aE rtn.
[Lipschitz wrote yal = ax where 2: = x e l l , y = a1 = e l a e l l . ] In mod-
ern terms (introduced by Chevalley), Lipschitz used the exterior exponential
23.15 Theodor Vahlen 323
of the bivector B so that the normalized element allal was in the spin group
Spin(n) .
23.15 T h e o d o r Vahlen
Vahlen 1897 found an explicit expression for the multiplication rule of two basis
elements in C&,n
1 2 . . .eEw) = (-l)Citj aiPje:l+Pler2+P2
. . . eZn)(ePleP* . . .enQn+Pn
QI QZ
(el e2
where the exponents are 0 or 1 (added here modulo 2, although for Vahlen 1+1
= 2, so that summation was over i > j). Vahlen's formula has frequently been
reinvented afterwards: for positive metrics by Brauer & Weyl 1935, for arbitrary
metrics by Deheuvels 1981 p. 294, disguised with index sets as in Chevalley 1946
p. 62, Artin 1957 p. 186 and Brackx& Delanghe& Sommen 1982 p. 2, or hidden
among permutations as in Kahler 1960162 and Delanghe & Sommen & SouEek
1992, pp. 58-59.
Although Brauer & Weyl 1935 reinvented (in the case of the Clifford alge-
bra Ct,) the above explicit multiplication formula of Vahlen, they did not
observe the connection to the Walsh functions (discovered in the meantime by
Walsh 1923). The connection to the Walsh functions was observed by Hag-
mark & Lounesto 1986.
In 1902, Vahlen initiated the study of Mobius transformations of vectors in
Rn (or paravectors in R @ Rn) by 2 x 2-matrices with entries in Clot,. This
study was re-initiated by Ahlfors in the 1980's.
23.16 Elie C a r t a n
Besides detecting spinors in 1913 (and pure spinors in 1938)) Cartan made two
other contributions to Clifford algebras: their periodicity of 8 and the triality
of Spin(8).
Cartan 1908 p. 464, identified the Clifford algebras Ceprqas matrix algebras
with entries in R , C, W, R @ R , W@W and found a periodicity of 8. To decipher
Cartan's notation:
h
324 A History of Clifford Algebras
+
where h = 1 - p q (mod 8). Clifford's original notion of 4 classes was thus
refined to 8 classes (and generalized from Ce, and Ceo,, to Ce,,,). 20
Cartan's periodicity of 8 for real Clifford algebras, with an involution, was
extended by C.T.C Wall 1968 and Porteous 1969 (rediscovered by Harvey 1990).
Wall considered real graded Clifford algebras, with an anti-involution, and
found a 2-way periodicity of type (8 x 8)/2, like the movements of a bishop on
a chessboard. Porteous used the anti-involution to induce a scalar product for
spinors, and classified the scalar products of spinors into 32 classes, according
to the signature types (p, 4) of real quadratic spaces RPJ.
In 1925, Cartan came into contact with the triality automorphism of Spin($).
Lounesto 1997 (in the first edition of this book) showed that triality is a re-
striction of a polynomial mapping Ce8 + Ceg, of degree 2.
23.17 E r n s t W i t t
Witt 1937 started the modern algebraic theory of quadratic forms. He iden-
tified Clifford algebras of non-degenerate quadratic forms over arbitrary fields
of characteristic # 2. The Witt ring W(IF), of a field IF, consists of similarity
classes of non-singular quadratic forms over IF (similar quadratic forms have
isometric anisotropic parts). In characteristic # 2, the structure of Clifford
algebras of certain quadratic forms was studied by Lee 1945148 (ep = I),
Chevalley 1946 (ep = -I), and Kawada & Iwahori 1950 (e; = f1). These
authors did not benefit the Witt ring (although they already had it at their
disposal), and so they did not consider all the isometry classes of anisotropic
quadratic forms.
Example. The Witt ring W(IF5) of the finite field IF5 = {0,1,2,3,4) of
characteristic 5 contains four isometry classes 0, (I), (s), (1, s) where s = 2 or
s = 3. Chevalley 1946, Lee and Kawada&Iwahori did not notice that none of
+
the quadratic forms xq x;, x; - x;, -x: - x i on the plane IF; is isometric
with x: +
SX; 2 (1, S) (but in fact they are all neutral, and thereby in the
same isometry class as 0). A simpler example is the line IF5 where the Clifford
algebra of 2x2 N (2) is the quadratic extension IF5(&) whereas the Clifford
algebras of both fx2 2 (f1) split IF5 x IF5. I
20 Cartan's ~eriodicitvof 8 for Clifford algebras. from 1908. is often attributed to Bott, who
was born'in 1923 and proved his perGdicit; of homotdpy groups of rotation groups in
1959.
23.18 Claude Chevalley 325
23.18 Claude Chevalley
Chevalley 1954 constructed Clifford algebras as subalgebras of the endomor-
phism algebra of the exterior algebra, Cl(Q) c End(/\ V), by means of a not
necessarily symmetric bilinear form B on V such that Q(x) = B ( x , x ) . By
this construction, Chevalley managed to include the exceptional case of char-
acteristic 2, and thus amended the work of Witt.
Chevalley 1954 went on further and gave the most general definition, as a
factor algebra of the tensor algebra, Ct(Q) = @V/ZQ, valid also when ground
fields are replaced by commutative rings. From the pedagogical point of view,
this approach is forbidding, while it refers to the infinite-dimensional tensor
algebra @V.
Chevalley 1954 introduced exterior exponentials of bivectors and used them
to scrutinize the Lipschitz group, unfairly naming it a 'Clifford group'. Thus,
there were two exponentials such that
and for
23.19 Marcel Riesz

Marcel Riesz 1958, pp. 61-67, reconstructed Grassmann's exterior algebra from
the Clifford algebra, in any characteristic # 2, by
where x E V and u E /\k V. Riesz' contribution enhanced Chevalley's result

of 1946, which related exterior products of vectors to antisymmetrized Clifford
products of vectors (Chevalley's result was valid only in characteristic 0).
Clifford algebras admit a parity grading or even-odd grading with even
and odd parts, C!(Q) = Ce+ (Q) eCe- (Q). Riesz's construction of 1958 showed
that there also exists a dimension grading
in all characteristics # 2 (because there is a privileged linear isomorphism

between a Clifford algebra and the exterior algebra). Thus, bivectors exist in
all characteristics # 2.
M. Riesz 1947 expressed the Maxwell stress tensor as T, = - i(e,Fe,F)o.
The first one to consider spinors as elements in a minimal left ideal of a
Clifford algebra was M. Riesz 1947 (although the special case of pure spinors
had been considered earlier by Cartan 1938).
23.20 Atiyah & Bott & Shapiro
In 1964, Atiyah & Bott & Shapiro reconsidered spinor spaces as modules over a
Clifford algebra, instead of regarding them as minimal left ideals in the Clifford
algebra. This permitted differentiation of spinor valued functions on manifolds,
not just on flat spaces.
They re-identified the definite real Clifford algebras Ce, and Ceo,, as matrix
algebras with entries in R, @, W, R $ R, W $ W (identified by Cartan in
1908). They rediscovered the periodicity of 8 (found by Cartan in 1908) with
respect to the graded tensor product (used earlier by Chevalley in 1955). They
emphasized the importance of the Z2-graded structure (the even-odd parity
in Clifford's works) and used it to simplify Chevalley's approach to Lipschitz
groups (the role of parity grading and/or grade involution in reflections was
not observed by Chevalley).
Atiyah & Bott & Shapiro's paper of 1964 is known for its dirty joke: they at-
tributed the introduction of Pin(n) to Serre from France, where pronunciation
of 'pin group' in English sounds the same as 'pine groupe' in French: pine [pin]
is a slang expression for male genitals.
E. Kahler 1960162 introduced a second product for Cartan's exterior differ-
ential forms making Grassmann's exterior algebra isomorphic with a Clifford
algebra. This re-interpretation of Chevalley's definition of the Clifford algebra
(extending xu = x J u + x A u) was renamed as liahler-Atiyah algebra by
W . Graf 1978 (Graf's Kahler-Atiyah algebra was again reinvented and applied
to the Kahler-Dirac equation by Salingaros & Wene 1985).
23.21 T h e Maxwell Equations

In the special case of a homogeneous and isotropic medium, Maxwell equations
can be condensed into a single equation. This has been done by means of com-
plex vectors (Silberstein 1907), complex quaternions (Silberstein 1912/1914),
spinors (Laporte & Uhlenbeck 1931) and in terms of Clifford algebras. Ju-
vet & Schidlof 1932, Mercier 1935 and Riesz 1958 condensed the Maxwell equa-
tions into a single equation by bivectors in the Clifford algebra Cel,3 of R19.
In the Clifford algebra Ce3,1 2: Mat(4, R),the single Maxwell equation
where d = V - eodo and F = Eeo - Bel23 E / \ 2 ~ 3 * 1could

, be decomposed
into two parts
d J F = J , dAF=O.
23.22 Spinors in ideals, as operators and recovered 327
Similarly, dA = -F could be decomposed into two parts, dA A = -F and the
Lorenz gauge/condition d . A = 0 discovered by the Danish physicist Ludwig
Lorenz and not by the Dutch physicist H. A. Lorentz. 21 Marcel Riesz 1947
expressed the Maxwell stress tensor as
1
Tpv= --(e,Fe,F)o
2
and Hestenes 1966, p. 31, introduced the vectors T, = - + F ~ , F for which
Tpv= T,-e, = e,-T, and the mapping T x = - ~ F X F where (Tx), = TpyxV.
[Juvet & Schidlof 1932, p. 141, gave
but did not observe that T x = -;FXF.] Note that To = $ ( E ~ + B ~ ) ~ ~ + E
23.22 Spinors i n ideals, a s o p e r a t o r s a n d recovered

Juvet 1930 and Sauter 1930 replaced column spinors by square matrices in
which only the first column was non-zero - thus spinor spaces became minimal
left ideals in a m a t r i x algebra. Marcel Riesz 1947 was the first one to consider
spinors as elements in a minimal left ideal of a Clifford algebra (although the
special case of pure spinors had been considered earlier by Cartan 1938).
Giirsey 1956158 rewrote the Dirac equation with 2 x 2 quaternion matrices
in Mat(2,W) (see also Gsponer&Hurni 1993). Kustaanheimo 1964 presented
the spinor regularization of the Kepler motion, the KS-transformation, which
-
emphasized the operator aspect of spinors. This led David Hestenes 1966-74 to
a reformulation of the Dirac theory, where the role of spinors [in columns C4
or in minimal left ideals of the complex Clifford algebra C @I Cel,3 Mat (4, C)]
was taken over by operators in the even subalgebra cet3 of the real Clifford
algebra Cel,3 21 Mat (2, W ) .
Spinors were reconstructed from their bilinear covariants by Y. Takahashi
1983 and J. Crawford 1985, in the case of the electron. Lounesto 1993 general-
ized the reconstruction of spinors to the null case of the neutron, and predicted
existence of a new particle residing in between electrons and neutrons.
Bibliography

21 J. van Bladel: Lorenz or Lorentz? IEEE Antennas and Propagation Magazine 33 (1991)
p. 69 and The Radioscientist 2 (1991) p. 55.
A History of Cl?flod Algebras
M.F. Atiyah, R. Bott, A. Shapiro: Clifford modules. Topology 3 , suppl. 1 (1964),
3-38. Reprinted in R. Bott: Lectures on I<(X). Benjamin, New York, 1969, pp.
Oxford, 1988, pp. 301-336.
R. Brauer, H. Weyl: Spinors in n dimensions. Amer. J. Math. 5 7 (1935), 425-449.
Reprinted in Selecta Hermann Weyl, Birkhiser, Basel, 1956, pp. 431-454.
E. Cartan (expos&d'aprks l'article allemand de E. Study): Nombres complexes, pp.
1, Fasc. 4, art. 15, 1908.
E. Cartan: Sur les groupes projectifs qui ne laissent invariante aucune multiplicitd
plane. Bull. Soc. Math. France 4 1 (1913), 53-96.
E. Cartan: Le principe de dualitk et la thkorie des groupes simples et semi-simples.
Bull. Sci. Math. 49 (1925), 161-174.
A. Cayley: On certain results relating to quaternions. Phil. Mag. (3) 26 (1845),
141-145.
A. Cayley: Recherches ultdrieures sur les ddterminants gauches. J. Reine Angew.
Math. 50 (1855), 299-313.
C. Chevalley: Theory of Lie Groups. Princeton University Press, Princeton, 1946.
York, 1954. The Algebraic Theory of Spinors and Clifford Algebras. Springer,
Berlin, 1997.
C. Chevalley: The Construction and Study of Certain Important Algebras. Mathe-
matical Society of Japan, Tokyo, 1955.
W.K. Clifford: Applications of Grassmann's extensive algebra. Amer. J. Math. 1
(1878), 350-358.
W.K. Clifford: On the classification of geometric algebras; pp. 397-401 in R. Tucker
(ed.): Mathematical Papers by William Kingdon Clifford. Macmillan, London, 1882.
(Reprinted by Chelsea, New York, 1968.) Title of talk announced already on p. 135
in Proc. London Math. Soc. 7 (1876).
J.P. Crawford: On the algebra of Dirac bispinor densities: Factorization and inversion
P.A.M. Dirac: The quantum theory of the electron. Proc. Roy. Soc. A 117 (1928),
610-624.
A. Gsponer, J.-P. Hurni: Lanczos' equation to replace Dirac's equation? in pp. 509-512
of J.D. Brown (ed.): Proceedings of the Cornelius Lanczos Centenary Conference,
(Raleigh, NC, 1993).
F. Giirsey: Correspondence between quaternions and four-spinors. Rev. Fac. Sci.
Univ. Istanbul A21 (1956), 33-54.
F. Giirsey: Relation of charge independence and baryon conservation to Pauli's trans-
formation. Nuovo Cimento 7 (1958), 411-415.
P.-E. Hagmark, P. Lounesto: Walsh functions, Clifford algebras and Cayley-Dickson
process, pp. 531-540 in J.S.R. Chisholm, A.K. Common (eds.): Clifford Algebras and
their Applications in Mathematical Physics. Reidel, Dordrecht, The Netherlands,
1986.
J. Helmstetter: Algdbres de Clifford et algbbfes de Weyl. Cahiers Math. 25, Mont-
pellier, 1982.
D. Hestenes: Real spinor fields. J. Math. Phys. 8 (1967), 798-808.
D. Hestenes, G. Sobczyk: Clifford Algebra to Geometric Calculus. Reidel, Dordrecht,
1984, 1987.
23.22 Spinors in ideals, as operators and recovered 329
B. Jancewicz: Multivectors and Clifford Algebra in Electrodynamics. World Scientific
Publ., Singapore, 1988.
G. Juvet: Opkrateurs de Dirac et kquations de Maxwell. Comment. Math. Helv. 2
(1930), 225-235.
G. Juvet, A. Schidlofi Sur les nombres hypercomplexes de Clifford et leurs applications
B l'analyse vectorielle ordinaire, B l'dlectromagnetisme de Minkowski et B la thkorie
de Dirac. Bull. Soc. Neuchat. Sci. Nat., 57 (1932), 127-147.
E. Kidder: Der innere Differentialkalkiil. Rendiconti di Matematica e delle sue Ap-
plicazioni (Roma) 21 (1962), 425-523.
Y.Kawada, N. Iwahori: On the structure and representations of Clifford algebras. J.
Math. Soc. Japan 2 (1950), 34-43.
P. Kustaanheimo, E. Stiefel: Perturbation theory of Kepler motion based on spinor
regularization. J. Reine Angew. Math. 218 (1965), 204-219.
0.Laporte, G. E. Uhlenbeck: Application of spinor analysis to the Maxwell and Dirac
equations. Phys. Rev. 37 (1931). 1380-1397.
H. C. Lee: On Clifford algeb~asakd their representations. Ann. of Math. 49 (1948),
760-773.
R. Lipschitz: Principes d'un calcul algkbrique qui contient comme espkces particulikres
le calcul des quantitks imaginaires et des quaternions. C.R. Acad. Sci. Paris 91
R. Lipschitz: Untersuchungen iiber die Summen von Quadraten. Max Cohen und
Sohn, Bonn, 1886, pp. 1-147. (The first chapter of pp. 5-57 translated into French
by J. Molk: Recherches sur la transformation, par des substitutions rkelles, d'une
somme de deux ou troix carrks en elle-mCme. J. Math. Pures Appl. (4) 2 (1886),
373-439. French rksumk of all three chapters in Bull. Sci. Math. (2) 10 (1886),
163-183.)
R. Lipschitz (signed): Correspondence. Ann. of Math. 69 (1959), 247-251. Reprinted
on pages 557-561 of A. Weil: @uvres Scientifipues, Collected Papers, Volume 11.
Springer, 1979.
P. Lounesto: Clifford algebras and Hestenes spinors. Found. Phys. 23 (1993), 1203-
1237.
A. Mercier: Expression des dquation de ~'dlectrorna~netisrne au Moyen des Nombres
de Clifford. Thkse, Universitk de Genkve, 1935.
W. Pauli: Zur Quantenmechanik des magnetischen Elektrons. Z. Phys. 42 (1927),
601-623.
M. Riesz: Sur certain notions fondamentales en thdorie quantique relativiste. C . R . loe
Congris Math. Scandinaves, (Copenhagen, 1946). Jul. Gjellerups Forlag, Copen-
hagen, 1947, pp. 123-148. &printed in L. Gkding, L. Hormander (eds.): Marcel
Riesz, Collected Papers. Springer, Berlin, 1988, pp. 545-570.
M. Riesz: Clifford Numbers and Spinors. University of Maryland, 1958. Reprinted
as facsimile by Kluwer, Dordrecht, 1993.
F. Sauter: Losung der Diracschen Gleichungen ohne Spezialisierung der Diracschen
Operatoren. Z. Phys. 63 (1930), 803-814.
Y. Takahashi: A passage between spinors and tensors. J. Math. Phys. 24 (1983),
1783-1790.
K. Th. Vahlen: ~ b e hohere
r komplexe Zahlen. Schriften der phys.-Ckon. Gesellschaft
zu KGnigsberg 38 (1897), 72-78.
r und complexe Zahlen. Math. Ann. 55 (1902),
585-593.
C.T.C. Wall: Graded algebras, antiinvolutions, simple groups and symmetric spaces.
Bull. Amer. Math. Soc. 74 (1968), 198-202.
B.L. van der Waerden: On Clifford algebras. Nederl. Akad. Wetensch. Proc. Ser. A
69 (1966), 78-83.
E. Wi tt: Theorie der quadratischen Formen in beliebigen Korpern. J. Reine Angew.
Math. 176 (1937), 31-44.
Selected Reading
R. Ablamowicz, P. Lounesto, J. Maks: Conference Report, Second Workshop on 'Clif-

ford Algebras and their Applications in Mathematical Physics' (Montpellier, 1989).
Found. Phys. 2 1 (1991), 735-748.
L. Ahlfors, P. Lounesto: Some remarks on Clifford algebras. Complex Variables,
Theory and Application 1 2 (1989), 201-209.
S.L. Altmann: Rotations, Quaternions and Double Groups. Clarendon Press, Oxford,
1986.
M.F. Atiyah, R. Bott, A. Shapiro: Clifford modules. Topology 3 , suppl. 1 (1964),
3-38. Reprinted in R. Bott: Lectures on IC(X). Benjamin, New York, 1969, pp.
Oxford, 1988, pp. 301-336.
W.E. Baylis: Electrodynamics: A Modern Geometric Approach. Birkhauser, Boston,
1999.
E.F. Bolinder: Unified microwave network theory based on Clifford algebra in Lorentz
space; pp. 25-35 in Conference Proceedings, 1.2" European Microwave Conference
(Helsinki 1982). Microwave Exhibitions and Publishers, Tunbridge Wells, Kent,
1982.
E.F. Bolinder: Clifford algebra: what is it? IEEE Antennas and Propagation Society
Newsletter 29 (1987), 18-23.
N. Bourbaki: AlgBbre, Chapitre 9, Formes sesquiline'aires et formes quadratiques. Her-
mann, Paris, 1959.
F. Brackx, R. Delanghe, F. Sommen: Clifford Analysis. Research Notes in Mathe-
matics 76, Pitman, London, 1982.
R. Brauer, H. Weyl: Spinors in n dimensions. Amer. J. Math. 57 (1935), 425-449.
Reprinted in Selecta Hermann Weyl, Birkhauser, Basel, 1956, pp. 431-454.
E. Cartan (expos6 d'apr6s l'article d e m a n d de E. Study): Nombres complexes; pp.
1, Fasc. 4, art. I 5 (1908). Reprinted in E. Cartan: (Euvres Complktes, Partie 11.
C. Chevalley: Theory of Lie Groups. Princeton University Press, Princeton, NJ, 1946.
332 Selected Reading
York, 1954.
C. Chevalley: The Construction and Study of Certain Important Algebras. Mathe-
matical Society of Japan, Tokyo, 1955.
J.S.R. Chisholm, A.K. Common (eds.): Proceedings of the N A T O and S E R C Work-
shop o n 'Clifford Algebras and their Applications in Mathematical Physics' (Can-
W.K. Clifford: Applications of Grassmann's extensive algebra. Amer. J. Math. 1
(1878), 350-358.
W.K. Clifford: On the classification of geometric algebras, pp. 397-401 in R. Tucker
(ed.): Mathematical Papers by William Kingdon Clifford,Macmillan, London, 1882.
J. Crawford: On the algebra of Dirac bispinor densities: factorization and inversion
R. Deheuvels: Formes quadratiques et groups classiques. Presses Universitaires de
R. Delanghe, F. Sommen, V. SouEek: Clifford Algebra and Spinor Valued Functions:
A Function Theory for the Dirac Operator. Kluwer, Dordrecht, The Netherlands,
1992.
V.L. Figueiredo, E. Capelas de Oliveira, W.A. Rodrigues, Jr.: Covariant, algebraic
and operator spinors. Internat. J. Theoret. Phys. 29 (1990), 371-395.
L. Gkding, L. Hormander (eds.): Marcel Riesz, Collected Papers. Springer, Berlin,
1988.
J. Gilbert, M. Murray: Clifford Algebras and Dirac Operators i n Harmonic Analy-
Cambridge, 1991.
W. Greub: Multilinear Algebra, 2nd edn. Springer, Berlin, 1978.
P.-E. Hagmark, P. Lounesto: Walsh functions, Clifford algebras and Cayley-Dickson
process, pp. 531-540 in J.S.R. Chisholm, A.K. Common (eds.): Clifford Algebras and
their Applications i n Mathematical Physics. Reidel, Dordrecht, The Netherlands,
1986.
25 (1984), 1823-1832.
J. Helmstetter: Algkbres de Clifford et algkbres de Weyl. Cahiers Math. 25, Mont-
pellier, 1982.
D. Hestenes: Real spinor fields. J. Math. Phys. 8 (1967), 798-808.
D. Hestenes: Observables, operators, and complex numbers in the Dirac theory. J.
Math. Phys. 16 (1975), 556-571.
D. Hestenes, G. Sobczyk: Clifford Algebra to Geometric Calculus. Reidel, Dordrecht,
The Netherlands, 1984, 1987.
D. Hestenes: New Foundations for Classical Mechanics. Reidel, Dordrecht, 1986,
(2nd ed.) 1999.
Found. Phys. 16 (1986), pp. 708-709.
D.J. Hurley, M.A. Vandyck: Geometry, Spinors, and Applications. Springer, Berlin,
1999.
Selected Reading
B. Jancewicz: Multivectors and Clifford Algebra in Electrodynamics. World Scientific,
Singapore, 1988.
E. Kahler: Der innere Differentialkalkiil. Rendiconti di Matematica e delle sue Appli-
cazioni (Roma) 2 1 (1962), 425-523.
J. Keller, S. Rodriguez-Romo: A multivectorial Dirac equation. J. Math. Phys. 3 1
(1990), 2502.
M.-A. Knus: Quadratic Forms, Cligord Algebras and Spinors. Univ. Estadual de
Campinas, Brazil, 1988.
1980.
H.B. Lawson, M.-L. Michelsohn: Spin Geometry. Universidade Federal do Cearb,
Brazil, 1983. Princeton University Press, Princeton, NJ, 1989.
R. Lipschitz: Principes d'un calcul algdbrique qui contient comme espkces particulihres
le calcul des quantitds imaginaires et des quaternions. C.R. Acad. Sci. Paris 9 1
R. Lipschitz: Untersuchungen iiber die Summen von Quadraten. Max Cohen und
Sohn, Bonn, 1886, pp. 1-147. The first chapter of pp. 5-57 translated into French
by J. Molk: Recherches sur la transformation, par des substitutions rdelles, d'une
somme de deux ou troix carrds en elle-m6me. J. Math. Pures Appl. (4) 2 (1886),
373-439. French rdsumd of all three chapters in Bull. Sci. Math. (2) 1 0 (1886),
163-183.
R. Lipschitz (signed): Correspondence. Ann. of Math. 69 (1959), 247-251. Reprinted
on pages 557-561 of A. Weil: (Euvres Scientijiques, Collected Papers, Volume 11.
P. Lounesto: Scalar products of spinors and an extension of Brauer-Wall groups.
Found. Phys. 11 (1981), 721-740.
P. Lounesto: Report of Conference, NATO and SERC Workshop on 'Clifford Algebras
and their Applications in Mathematical Physics' (Canterbury, 1985). Found. Phys.
1 6 (1986), 967-971.
P. Lounesto, P. Bergh: Axially symmetric vector fields and their complex potentials.
Complex Variables: Theory and Application 2 (1983), 139-150.
P. Lounesto, A. Springer: Mobius transformations and Clifford algebras of Euclidean
and anti-Euclidean spaces, pp. 79-90 in J. Lawrynowicz (ed.): Deformations of
Mathematical Structures. Kluwer, Dordrecht, The Netherlands,l989.
P. Lounesto, G.P. Wene: Idempotent structure of Clifford algebras. Acta Applic.
Math. 9 (1987), 165-173.
A. Micali, R. Boudet, J. Helmstetter (eds.): Proceedings of the Pd Workshop on 'Clif-
ford Algebras and their Applications i n Mathematical Physics' (Montpellier, 1989).
A. Micali, Ph. Revoy: Modules quadratiques. Cahiers Math. 10, Montpellier, 1977.
Bull. Soc. Math. France 63, suppl. (1979), 5-144.
M. Riesz: Sur certaines notions fondamentales en thdorie quantique relativiste; pp.
123-148 in C.R. 10e C0ngrZ.s Math. Scandinaves, Copenhagen, 1946. Jul. Gjellerups
Forlag, Copenhagen, 1947. Reprinted in L. GArding, L. Hormander (eds.): Marcel
Riesz, Collected Papers, Springer, Berlin, 1988, pp. 545-570.
Selected Reading
M. Riesz: L'kquation de Dirac en relativitC gknkrale; pp. 241-259 in Tolfte[=l2.] Skan-
dinaviska Matematikerkongressen i Lund, 1953. H h n Ohlssons boktryckeri, Lund,
1954. Reprinted in L. GBrding, L. Hormander (eds.): Marcel Riesz, Collected Pa-
pers, Springer, Berlin, 1988, pp. 814-832.
lands, 1993.
W.A. Rodrigues, Jr., E. Capelas de Oliveira: Dirac and Maxwell equations in the
Clifford and spin-Clifford bundles. Internat. J. Theoret. Phys. 29 (1990), 397-412.
A. Ronveaux, D. Lambert: Le probleme de factorisation de Hurwitz. Approche his-
torique, solutions, applications e n physique. Faculte Universitajres N.D. de la Paix,
Namur, 1991.
N.A. Salingaros, G.P. Wene: The Clifford algebra of differential forms. Acta Applic.
Math. 4 (1985), 271-292.
L. Silberstein: The Theory of Relativity. Macmillan, London, 1914.
J. Snygg: Clifford Algebra, a Computational Tool for Physicists. Oxford University
Press, Oxford, 1997.
K. Th. Vahlen: ~ b e hohere
r komplexe Zahlen. Schriften der phys.-6kon. Gesellschaft
zu Kiinigsberg 3 8 (1897), 72-78.
r und complexe Zahlen. Math. A n n . 55 (1902),
585-593.
B.L. van der Waerden: A History of Algebra. Springer, Berlin, 1985.
Math. 176 (1937), 31-44.
Index
algebra 21 Cayley algebra 200, 303

,automorphism of 22 Cayley transform 222
,Clifford 9, 23, 40, 56, 190 center 55
,division 200, 301 charge conjugation 162, 168
,exterior 40 Clifford
,Lie 95 algebra 9, 26, 40,53, 190
,opposite 175, 202 -conjugate 29, 56, 86
,simple 202 dual 39
alternation 194 product 9, 41, 189
alternative 200, 301 combination
automorphism ,linear 6
,similar 31 companion 307, 317
complex
basis 3 conjugate 18
,orthonormal 7 number 18
,standard 4 conjugate
bilinear 21 ,charge 162, 168
covariants 137, 142 ,Clifford- 29, 56, 86
bivector 8 ,complex 18
,opposite 34 ,quaternion 69
,simple 87 contraction 44, 46
,unit 25, 34 coordinate 3, 6
Brauer group 202 cover (ing)
Brauer-Wall group 239 ,two-fold 30
Brauer-Wall-Porteous group 240 Crawford 152, 156
cross product 37, 93
Cartan map 63
Cartan's principle of triality 308 dimension 6
Cayley-Dickson process 285, 302 grading 42
Index
Dirac
adjoint 136, 139 grade involution 29, 86
current 136 graded tensor product 202
equation 135, 136 grading
-Hestenes equation 144 ,dimension 42
matrices 135 group
directed line segment 1 ,Brauer 202
division ,Brauer-Wall-Porteous 240
algebra 200, 301 ,Lorent2 124
ring 61, 232 ,spin 220
dual
,Clifford 39
helicity 164
,Hodge 38
Hestenes 149
,Dirac-Hestenes equation 144
endomorphism 6 Hodge dual 38
energy hyperbolic plane 196
,kinetic 50
projection operator 138
ideal
,total 50
,left 52, 60
equation
,minimal 60
,Dirac 136 idempotent 52, 60
,Maxwell 101
,lattice of 227
,Schrodinger 50
,primitive 52, 61, 138, 164, 226
Euclidean
imaginary part 18
plane 7
inverse quaternion 70
space 93 involution 56
exterior
,grade 29, 56, 86
algebra 40
,parity 169
product 10, 34
irreducible
even
components 53
part 28
fields 260
subalgebra
left ideal 213
representation 228, 232
field 21 tensor 113
,automorphism of 22 isoclinic 89, 310
,ordering of 31 isotropic 196
function
,linear 5
Ii'S- transformation 148
,right linear 73
Kustaanheimo-Stiefel = K S 63
number 20
left ideal 52, 60 ,complex 18
,graded 213 ,negative 31
,irreducible 213 ,positive 31
,minimal 52, 60, 226
Lie algebra 95
octonion 97, 303
1'inear
odd part 28
combination 6
opposite
function 5
algebra 175, 202
isomorphism 6
bivector 34
,right 73
metric 174
space 5
product 3 18
structure 5
vector 2
linearly independent 6
ordering 31
Lorentz
force 100
group 124 parity involution 169
invariants 123 part
transformation 120 ,even 28
Lorenz ,odd 28
condition 106 Pauli
gauge 106 -Dirac representation 136
spin matrices 51, 73
spinor 52, 60
map, see function
polar form 19
mapping, see function
positive 31
matrix
product
,Pauli spin 5 1, 73.
,Clifford 9
Maxwell equations 101
,cross 37
metric 8
,exterior 10, 34
minimal left ideal 52, 60, 226
,graded tensor 202
Minkowski space-time 102, 121
,scalar 7, 92
multivector structure 43
of spinors 233
negative 3 1
quadratic
neutral
form 195
axis 315
,neutral 196
quadratic space 195
space 93, 195
norm 8, 19, 36, 37, 70
quaternion 68
null = isotropic 196
conjugate 69
Index
,Pauli 52, 60
real ,recovery of 155
part 10 regularization 63
structure 139 representation 53
representation ,semi- 228
,faithful 228 space 61
,irreducible 228 ,Weyl 164
,Pauli-Dirac 136 spinoriality 169
reversion 28, 56, 86 standard basis 4
rotation structure
,isoclinic 89, 310 ,complex 139
,simple 89 ,linear 5
Rodrigues formula 58, 71 ,multivector 43
,real 139
scalar 1
product 7, 92 tensor product 197, 201
of spinors 223 time reversal 169
Schrodinger equation 50 ,Wigner 169
-Pauli equation 64 triality 306, 309
similar 31 ,Cartan's principle of 308
simple triplet 308
Clifford algebra 228 two-fold 30
bivector 87
rotation 89
unit
space
bivector 25, 34
,Euclidean 93
circle 67
,linear 5
vector 2
,quadratic 93, 195
universal 192
span 6
spin 50
group 30, 59, 220 vector 1, 5
projection operator 138 space 4
spinor ,unit 2
,column 138
,Dirac 164, 167 Witt
,even 228 index 196
,ideal 138 ring 198
,Majorana 163
operator 63, 143, 145

Clifford Algebra and Spinors

Uploaded by

Copyright:

Available Formats

Clifford Algebra and Spinors

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Clifford Algebra and Spinors

Uploaded by

Copyright:

Available Formats

London Mathematical Society

Lecture Note Series 286

Clifford Algebras and Spinors

This book is in copyright. Subject to statutory exception

ISBN 0 521 0055 1 5 paperback

Helsinki, November 2000 PERTTI

Vectors provide a mathematical formulation for the notion of direction, thus

1.1 Scalars a n d vectors

1.2 Vector a d d i t i o n a n d subtraction

Vector addition is commutative, a + b = b +a, as can be seen by inspection of

1.3 Multiplication b y n u m b e r s (scalars)

1.4 Bases and coordinates

Conversely, we may start from the set R x R = {(z, y) I t ,y E R ) , and

of the 2-dimensional linear space R2.

1.6 Linear independence; dimension

which equals la1 = d m , and the angle given by

1.8 The Clifford product of vectors; the bivector

1.9 The Clifford algebra Ce2

The four elements

is a linear combination of a scalar uo, a vector ulel + u2e2 and a bivector

1.10 Exterior product = bivector part of the Clifford product

Area = lalb2- a2bl1

The parallelogram can be regarded as a kind of geometrical product of its

The commutative rule a . b = b . a together with the anticommutative rule

Adding and subtracting equations (a) and (b), we find

ab=-ba 9 a l b a.b=O 9 ab=aAb.

1.11 Components of a vector in given directions

11 The bivector valued exterior product a /\ b = (alb2 - a 2 b l ) e 1 2 ,which represents a plane

1.12 Perpendicular projections a n d reflections

In other words, the parallel component all is the scalar product a . b =

The area of the parallelogram with sides a, b is seen to be

1.13 M a t r i x representation of Ce2

In this representation the transpose of a matrix,

corresponds to the reverse

( ,b) [=(ad-bc)( ) for a d b c # 0 11

The reversion and Clifford-conjugation are anti-involutions, that is, involutory

for which ii = ii- = ii-

Define two anti-automorphisms a,p to be similar, if there is an

with squares I, I, and -I.

as can be seen by choosing the intertwining automorphism

for which a y = yp.

1 Electrical engineers denote the square root of -1 by j = G.

Addition of complex numbers can be illustrated by the parallelogram law of

although this result is also a consequence of distributivity, associativity and

Division is carried out as multiplication by the inverse: zl/z2 = z l z ~ l .

The real number cp is called the phase-angle or argument of z [sometimes all

3 Electrical engineers denote the polar form by r k .

= rl (cos p l + i sin pl) and z2 = r2(cos cp2 + i sin p2).

We write ez = exp(z). The series expansions of trigonometric functions result

The exponential form of multiplication seems natural:

Examples. (1 + i)-' = i ( 1 - i), & = &$(l+ i), eirl2 = i. I

2.1 The field @ versus the real algebra @

Definition. An algebra over a field IF is a linear space A over IF together with

To distinguish the field C from a real algebra C let us construct C as the

then we actually introduce a real linear structure on the set R 2 = R x R . This

as well as the unity, a(1) = 1. A function a : C + C is an autoniorphism of

as well as the unity, a(1) = 1.

2.2 The double-ring 2R of R

results in the double-ring 2R of R. The only automorphisms of the real algebra

Study numbers have Study conjugate (a + jb)- = a - j b , Lorentz squared