Sei sulla pagina 1di 14

Clifford Algebra, Symmetries, and

Vectors

S. L. ALTMANN”
Brasenose College, University of Oxford, Oxford OX1 4AJ, United Kingdom

Received April 27, 1995; accepted for publication October 23, 1995

ABSTRACT rn
The realization of a Clifford algebra in laboratory space is considered and it is
demonstrated that the elements of the algebra cannot, as often assumed, be directly
identified with vectors in this space, but, rather, that they form the parametric space of
the symmetry operations of the Euclidean group as performed in the laboratory space.
Details of this parametrization are established and expressions are given that determine
the action of the Euclidean-group operations (screws included) on laboratory-space
vectors in terms of the elements of the Clifford algebra. A discussion of Clifford vectors,
bivectors, and pseudoscalars and their relation to the Gibbs vectors is provided. The
correct definition of axial and polar vectors within the Clifford algebra is carefully
discussed. It is shown how simple it is to generate finite point groups in 4-dimensional
space by means of the Clifford algebra. 0 1996 John Wiley & Sons, Inc.

usually been defined in relation to some abstract


Introduction linear vector spaces and it has often been the
practice in the past to identify in an axiomatic way
the elements of such spaces with physical objects

I n early work about molecular and solid-state


studies, the necessary symmetries were treated
by geometrical or matrix methods, the latter being
such as physical vectors. These realizations of the
abstract vector spaces are frequently regarded as
mappings which require no further justification
based on the clumsy Euler-angle parametrization than the verification of some formal rules. The
of rotations, subject to various singularities in its inventor of the first algebra that had direct rele-
parameter space. It is desirable in modern work to vance to the study of the rotation group S0(3),
move over to algebraic methods, which are far Hamilton [ 11, thus identified the algebra spanned
more precise and are easier to program. Progress by the three quaternion units with ordinary vec-
with algebraic methods, however, have been ham- tors in 3-space. Riesz [2] pointed out that such
pered by problems of interpretation. Algebras have identification was wrong since the vectors so de-
*It is my pleasure to dedicate this article to Roy McWeeny,
fined are axial and not polar vectors. Altmann [3]
in warm recognition for his extensive contributions to the study went further and showed that Hamilton’s “vec-
of symmetry. tors’’ were nothing else than binary rotations (rota-

InternationalJournal of Quantum Chemistry, Vol. 60, 359-372 (1996)


0 1996 John Wiley & Sons, Inc. CCC 0020-7608 I 9 6 I 0 10359-14
ALTMANN

tions by T).He showed that binary rotations could application of (I):


masquerade as vectors for the following reason: To
each rotation of SO(3), a unique point of the unit e2ez= 1, (2)
sphere can be associated with it, which is the point e~e e e 1 = -ee,e,e,el
~ 1 = -1, (3)
of the unit sphere left invariant by the rotation and
e,e,e,e,e,e, = -e,e,e,e,e,e,
such that from outside that point the Dtation is
seen as counterclockwise. Such poiltLaare called = e,e,e,eke,ek= - el el ek ek = -1. (4)
poles of the rotations. (Of course, the pole of a
rotation can be identified with the vector corre- The applications of Clifford algebra in physics
sponding to it, and in this article, we shall do so, have been much studied. After Marcel Riesz’s pio-
thus regarding the poles of rotations as vectors.) neering work, Hestenes [51 constructed a very ele-
Position vectors of points of the unit sphere can, if gant and powerful multivector algebra, further
necessary, be uniquely identified by poles of rota- developed by Hestenes and Sobczyk [61. This work
tions that coincide with them. That a position has been continued and extended by Greider [7]
vector and the pole of a rotation coincide does not and his school at Davis. (See also [8,9].) A very
alter the fact that the position vector and the sym- important contribution of Hestenes to the subject
metry operation used to map the vector via its is his insistence in giving paramount importance
pole are entirely different geometrical objects (be- to the geometrical meaning of any vector algebra
cause their transformation properties are different), used, a principle which will be followed very
a point that appears to have been glossed over in strictly in the present article, albeit from a different
the past. Handling a position vector via a pole point of view. The elements of Hestenes’ multivec-
must be done very carefully, as this article will tor algebra, are identified with vectors in 3-space,
demonstrate. but this identification has not always been done
The study of the symmetry operations in SO(3) without some uncomfortable results. Thus,
and O(3) can be done by means of the quaternion Hestenes [5] concluded that a vector product is a
algebra, as fully demonstrated in [41, where 75 polar and not an axial vector, which, of course,
point groups and their double groups have consis- disagrees with current practice in physics. (See
tently been tabulated by this method. Although also [lo].) Such interpretation was also followed by
the quaternion parametrization allows for a very Greider and his school (see, e.g., 1111, p. 11-24) but
compact presentation of the results, the inversion not necessarily by all proponents of the multivec-
has to be treated in an ad hoc manner since it tor algebra, [121. It should be stressed that what-
cannot be mapped within the algebra. When more ever the status of the above interpretation might
detail is required and especially if symmetry oper- be the value and importance of the multivector
ations in spaces of symmetry higher than three are algebra cannot be ignored. The geometrical inter-
to be studied, for which the quaternion method pretation to be discussed in this article, however,
fails, a more general algebra is required. It is well starts from a different point of view, namely, that
known that the Clifford algebra, C,, is most ade- the elements of the Clifford algebra must be recog-
quate for this purpose, although its practical appli- nized as mappings of symmetry operations of O(3)
cation to the study of symmetry has not been forming the parametric space of this group and
extensive. For simplicity, we shall discuss only C, that therein lies their primary geometrical mean-
in most of this article, which will give us all ing. As in the case of quaternions, if one wishes to
symmetries in three-dimensional space, although deal with laboratory-space vectors by means of the
point groups in higher-dimensional spaces will be Clifford algebra, position vectors must first be
discussed in a final section. In C,, we have three mapped by poles of symmetry operations and it is
units el ( i = 1,2,3) such that only then that carefully defined rules permit the
recovery of the ordinary vector algebra in labora-
tory space. Numerous examples of this important
e,e, + e,e, = 26,,. (1) technique will be given in this article.
The connection between the Clifford algebra
Given e,, e2, and e, terms of second and third and symmetry reflections is, of course, well known.
rank, e,e, and e,e,e,, respectively, can be formed Riesz 121 devoted a section of his book to this
and all these terms are units in the sense that their subject. In pure mathematics, since Cartan [13],
square is of unit modulus, as follows by repeated general results along this line valid for spaces of

360 VOL. 60,NO. 1


CLIFFORD ALGEBRA, SYMMETRIES, AND VECTORS

all dimensions have been obtained [14,15]. It The main purpose of this article was to present
should be stressed, nevertheless, that however well an alternative point of view to that of the multi-
known theoretically these results were they had vector algebras of the Hestenes type. Like Hestenes,
never been systematically exploited to parametrize we shall strongly emphasize at each stage the
the operations of O(3) in terms of Clifford-algebra geometrical meaning in laboratory space of the
elements. [This is not surprising, since the param- objects defined. We shall not define, however, any
etrization of SO(3) by the quaternion algebra, which new type of vector in the laboratory space: Our
is a great deal simpler, and which is the most vectors there will be the standard Gibbs vectors
natural way to deal with rotations, was practically with all their standard properties. On the other
ignored in the study of point groups until some 10 hand, we shall construct a Clifford space out of the
years ago or so.] There is a wide gap in practice standard Clifford algebra and we shall make the
between the existential results obtained in pure important connection between the two spaces by
mathematics as regards the mapping of Clifford proposing that the Clifford space be used as the
vectors onto reflections and the formulation of a parametric space of the symmetry operations in
workable parametrization of these symmetry oper- the laboratory space. Thus, the Clifford units will
ations, which is one of the main problems ad- be identified not with vectors but with orthogonal
dressed in this article. It cannot be stressed too reflections in O(3). It will be shown that this pro-
strongly that up to the present day the posal is entirely in agreement with the geometry of
parametrization of the operations of O(3)is carried O(3) and that the Clifford algebra spanned by the
out in the literature by ad hoc heuristic procedures units so-defined maps all the operations of O(3)
based on S 0 ( 3 ) , which, of course, affects pro- and, indeed, of the full Euclidean group. One of
foundly the study of the point groups so much the main results of this article, therefore, will be
used in crystallography and in the study of molec- the obtention of a parametrization of the Euclidean
ular structure. group, as defined in the laboratory space, by means
One other important preoccupation in this arti- of the Clifford algebra, in such a way as to ensure
cle is in relation to another aspect of the multivec- that the correct multiplication rules are obtained
tor algebras. It often appears in these algebras that and, most importantly, that one can operate on the
vectors are defined in terms of the Clifford units (Gibbs) vectors of the laboratory space correctly, a
and then, similarly, bivectors and pseudoscalars. point that requires a little care. It is self-evident
Fundamental distinctions are often made between that, in molecular applications for instance, the
the vectors and bivectors on the grounds that symmetry operations that are used must transform
when the units e, change sign then the vectors correctly all the physical Gibbs vectors required, a
change sign, whereas the bivectors (which are point that will be guaranteed by the present
written in terms of products e,e,) do not, and these method.
distinctions appear to be correlated with those As can be seen from the above, the general
between polar and axial vectors and pseu- approach of this article is firmly based on the
doscalars. It does not seem to be widely appreci- premise that geometry in the laboratory space is
ated that the distinction between polar and axial paramount and that no assertion about any map-
vectors which, whatever the terminology used in ping between the Clifford-space and the labora-
mathematics, is fundamental in much of physics, tory-space vectors should be made until rigorous
must be made not in terms of the inversion of an tests are made to ensure that all the properties of
arbitrarily defined vector space but rather in terms the laboratory-space geometry are correctly given.
of the inversion operation in the physical configu- For this reason, the language that we shall adopt
ration space (laboratory space). It is for this reason will be as nearly as possible that used in labora-
that in the main body of the article we shall stick tory-space geometry.
to the now rather old-fashioned terminology of The provision of an algebra along the lines
polar and axial vectors, leaving it to one of the last described, which as we shall see will allow us to
sections to make contact with the properties of transform vectors in the laboratory space under all
vectors and bivectors. It must be emphasized that the operations of the Euclidean group, is of some
the relation between polar and axial vectors is still importance in many practical applications, not only
most important in physical applications and that it in molecular and solid-state structure, but also in
cannot be ignored or tampered with without sig- such varied subjects as robotics and ray-tracing in
nificant consequences. optical systems. Before embarking into our main

INTERNATIONAL JOURNAL OF QUANTUM CHEMISTRY 361


ALTMANN

discussion, it will be useful to review the simpler


work that is done in SO(3) by means of the quater-
nion algebra.

with
Rotations and Quaternions
cos++, = 4cos++,cos++, - n1 * n 2 sin$+, sin++,,
A rotation R(+,n) is defined by an angle of
(9)
+
rotation 0 2 2 T and an axial vector n from the
fixed origin of the unit sphere. This vector is in- n3 sin+(b3= n 2 cos++, sin$+, + n1cos++,sin+(b,
variant under the rotation and it is called its pole:
+ n1 x n 2 sin:+, sin$+,. (10)

R(+,n)n = n. (5) Equations (9) and (10) are the well-known


quaternion-multiplication rules. If a vector such as
All rotation angles are defined as positive n is written in terms of the usual orthogonal unit
(counterclockwise) as seen from outside the sphere vectors i, j, and k (axial, of course), then
when looking at the head of the pole. The distinc-
tion between positive and negative rotations is i = LO, ill, j = [IO, jl, k = [O, kl (11)
done through the pole: All positive rotations and are the usual quaternion units which satisfy
all binary rotations are given poles in a conven- Hamilton’s well-known rules:
tionally defined positive hemisphere H and all
negative rotations are given poles in the negative i2 = -1, ij = k ,
hemisphere. The positive hemisphere need not be ijk 1, and cyclic permutations.
= - (12)
defined as a connected area on the unit sphere and
its choice must be carefully done if one is dealing It must be noted, however, that Hamilton, and
with finite subgroups of SO(3) (see [3]). most of the literature that followed on quater-
Normalized quaternions [a, A] (a, real scalar; A, nions, made no distinction between i and i, say.
axial vector; a’ + A’, equal to unity) can be Also, on comparing (11) with (6), it is evident that
mapped onto the rotations R(+,n) of SO(3) as the quaterinion units are not unit vectors at all, but,
follows: rather, binary rotations (rotations by T ) . (See [3]
for further discussion of the results above.)

(Notice that the vector n is axial.) For the purpose


Conventions
of multiplying rotations, which is the problem that
Before we go any further, we shall clarify a few
we have in mind, and purely for this purpose, it is
conventions that we shall use in this article and
possible to parametrize each rotation R(+,n) by
which are absolutely necessary to avoid such con-
the unique positive quaternion in ( 6 ) by means of
cepts as those of an axial vector to be ridden by
the following device: All poles n are chosen as
fallacies thLat easily result from mixing conventions
stated above, and all rotation angles are chosen in
or from the use of loose terminology. The right-
the range given in (6). When two rotations are
handed orthogonal axes i, j, and k are fixed in the
multiplied, the rotation angle in the product rota-
laboratory and never transformed. A (Gibbs) vec-
tion is always brought into the range stated, if
tor such as r is given by components rl, r2, and r3,
necessary, by addition of a multiple of 2 ~ and
r any
respectiveliy, in those axes. Symmetry operations
negative factor thus appearing is disregarded (al-
will always be understood in the active picture:
though it is, of course, essential if one wishes to
They transform all position vectors in the labora-
construct projective or spinor representations of
tory space but they do not change the axes. All
the rotation group). Under these restrictive condi-
symmetry operations are fixed in i, j, k, which
tions, the mapping is isomorphic: If
means that the symmetry elements, and therefore
also their poles, are never changed when symmetry
operations are performed in the laboratory space.

362 VOL. 60.NO. 1


CLIFFORD ALGEBRA, SYMMETRIES, AND VECTORS

The inversion operation changes the sign of the axial. Notice that endless confusion can be created
components of all position vectors in laboratory by loosely referring to ”vectors changing sign,”
space and of all polar vectors, but does not change meaning either that the vector reverses its direc-
the sign of the components of axial vectors. This is tion or that its components change sign, state-
identically the same, in the active picture but not ments which may or may not be equivalent. No-
necessarily otherwise, as saying that the inversion tice also that the positive sense of rotation should
operation reverses the direction of polar but not of be defined as that which takes i into j when
axial vectors. looking from the head of k, for any basis i, j, k of
It is sometimes stated that when inverting a X b arbitrary chirality.
not only the vectors must be inverted but also that It will be useful for later reference to summarize
the cross-product rule must be changed, thus re- the distinctions between polar and axial vectors,
sulting in the cross-product changing sign (and, within the conventions defined (always using ac-
thus, allegedly behaving like a polar vector). Al- tive transformations). Polar vectors reverse direc-
though there is an element of truth in this state- tion under inversion (and their components change
ment, it has to be very carefully formulated to sign), but axial vectors remain invariant. A polar
avoid the fallacious result stated. There is no doubt vector reverses direction when reflected through a
whatever that at least in low-energy physics it is perpendicular reflection plane but remains invari-
not possible to establish a difference between the ant when this plane is parallel. An axial vector
two possible chiralities (handedness). It is thus a reverses direction when reflected through a paral-
major principle that all statements involving chi- lel plane but remains invariant when this plane is
rality must be neutral, in the sense that one must perpendicular. (A simple geometrical treatment of
be able to use such statements without knowing
these results may be found in [16].) Also, within
which is the chirality that one is using. Thus, the
our conventions, ”reverse direction” of a vector is
instructions to construct a vector product must
equivalent to ”change the sign of its components.”
read as follows: Choose a basis of three orthogonal
unit vectors i, j, k. Then, given two polar vectors a
and b, a, b, and a x b must have the same chiral-
ity as that of i, j, k. It is thus obvious that in the Reflections and the Clifford Units
active picture, since there is no change of basis and
thus no change of chirality, the reversal of direc- Quaternion algebra is an excellent tool for deal-
tion of a and b under inversion does not entail a ing with all the operations of SO(3) but the inver-
reversal of direction of a X b, which is therefore sion and reflections that appear in O(3) are not
axial. tractable within this algebra, although it is possible
Although we do not use the passive picture in to deal with them by ad hoc techniques. The virtue
this article, we shall restate the last result in that of the Clifford algebra is not only that it deals with
picture. In the passive picture, the inversion re- all the operations of O(3) equally well, but that it
verses the direction of the i, j, k, vectors. Polar can also be extended to cover all the operations of
vectors do not change direction but their compo- the Euclidean group. The relation between reflec-
nents change sign. Axial vectors, instead, reverse tions and the Clifford units was already sketched
their direction and thus their components are in- in [ 3 ] ,but it will now be done in a more conve-
variant. (These are statements that can cause con- nient and general way. A reflection u is given as
fusion: For them to be absolutely clear in the the product iC, of the inversion i with a binary
passive picture, three unit vectors I, J, K must be rotation C, with the axis of rotation perpendicular
defined that coincide with i, j, k for the identity to the plane of u and it can be uniquely mapped
but, as a difference from i, j, k, are never trans- to an axial vector P,, which we shall call the pole
formed. When we say that a polar vector does not of the reflection. This is a unit axial vector normal
change direction, we mean that it is invariant with to the reflection plane and chosen so that it coin-
respect to the fixed axes I, J, K and contrarywise cides with the pole of the binary rotation C,, as
for an axial vector.) When i, j, k are inverted, a
chosen by the rules stated. The main property of P,
and b remain invariant in the sense just explained,
is that it is invariant under u.
but because the set a, b, a x b must change chiral-
ity (owing to the change in chirality of the basis),
a x b must reverse its direction and it is thus u P, = P, , (13)

INTERNATIONAL JOURNAL OF QUANTUM CHEMISTRY 363


ALTMANN

because an axial vector is invariant under a reflec- Hamilton's turns, much developed recently by
tion perpendicular to it. Biedenharn (see [ 181). The geometrical relation (15)
It will be important in what follows to keep in will be important for us as a check of the validity
mind some properties of the inversion and reflec- of the identification that we shall now provide of
tion operations. If we call the identity E and the reflections within the Clifford algebra.
anti-identity 2,the operations that multiply all If el, e,, e, are the units of C,, we propose the

-
operands by +1 and -I, respectively, it has been

-
following mapping:
shown [3,17] that there are two multiplication
rules for the inversion which are both fully com- el ax,e2 c-) my,e3 u2, (19)
patible with the conventions used in physics. They whereupon
are i2 = 1 or i2 = -1, which have been called by
Altmann the Pauli and Cartan gauges, respec-
tively. Whereas most of the literature on the sym-
metry operations of O(3) is based on the Pauli
gauge, it has been demonstrated [171 that the Car-
eIe3
e2e1
--
e3e2 c.) uzuY= C,,
u,u2= CZy= [O,j]
uYux
= C,,
=

=
[O,i]

[O, k]
=

=
i,

j,
k.
(20)
(21)
(22)
tan gauge cannot be ignored: its use, as we shall
see, is, in fact, unavoidable in the present work We have thus established a mapping between
[see Eq. (2611 and will henceforth be adopted. the bivectors e,el and the quaternion units. (Al-
As regards reflections, the orthogonal reflec- though the details of this mapping differ from
tions a,, uy, uz,respectively perpendicular to the those given in [3], the present choice is more
orthogonal directions x , y, z , will be useful. Given adequate for the purposes of this article.) We can
two reflections u2 and u, with poles P, and P,,
separated by the angle i4 from P, to P,, then,
from the quaternion multiplication rules (9) and
(101,
ii -
verify that this mapping is isomorphic:

e3e2e3e2= -e3e3e2e2 = -1,


ij * e,e2e,e, = e2e1 = c-) k ,
(23)
(24)
u2u, = ~~O~~,IliCO,P11 (14) in agreement with (12). It should be noticed that
- P,, P2 x PI]
-
= i 2 [ - P, within this mapping the inversion is a trivector:
= [P, - P,,P, x P2il (15) 1 = eIe2e3 = i,
uxuyu, (25)
= [cosi4,nsin$c,h] = R(+,n), (16)
and that we are unavoidably led to the Cartan
with gauge:
n = (PI x P,)/IP, x P21. (17) ii * z2 = e,e2e3ele2e3 = -1. (26)
It follows that for i4 5 i n , the vector n is in
(Notice that an upper-case letter has to be used for
the direction of P, X P,, but if i4 > i n , then n is
the image of the inversion in the Clifford algebra,
in the opposite direction, since, on using the rule
to avoid confusion with the quaternion unit i.
stated for the rotation products, the sign of the
Notice also that, because I behaves like the imagi-
quaternion requires change. (There is an element
nary unit i, it is often identified with it, which is
of convention here for the case i4 = in-,which
not entirely satisfactory: There are many objects in
has been chosen so as to agree without change
this work that share this form of behavior.)
with the products in the Cartan gauge as given in
It will presently be useful to consider the form
[ 3 ] . )From (16) and (17), we find, in particular, that
of a general binary rotation C, with pole n which,
uY uX = C2*,and cyclic permutations, (18) from (6), is

where C,, is a binary rotation about the z axis. C2 = LO, nl = [O, n,i + n2 j + n,kJ (27)
Relation (16) was obtained by Hamilton [l], al- = n,i + n, j + n,k. (28)
though, since he never parametrized rotations by
half-angles, he did not give it explicitly. Rather, We shall now identify a general reflection v
the construction of a rotation by 4 as the product with pole P as an element of the Clifford algebra
of two reflections separated by i+was the basis of C,. We write u for this purpose, as before, as iC2,

364 VOL. 60, NO. 1


CLIFFORD ALGEBRA, SYMMETRIES, AND VECTORS

with P as the pole of C2. On using (25) and (281, h f h= Zh, (37)
we have
and which is called the pole of the corresponding
u= e,e2e3(P,i + P, j + P 3 k ) (29) operation. (Strictly speaking, as we have discussed
= P,e, + P,e, + P3e3, (30) in the section on conventions, in the active picture
of symmetry operations, their poles are never af-
as can readily be derived on using (20) to (22) and fected by any of them. When we discuss the
(1). "transformation" property or, as here, the "invari-
We shall verify that the multiplication rule (15) ance" of a pole under a symmetry operation, we
is recovered with this identification, thus support- must assume that we superimpose on each pole a
ing the consistency of our parametrization. Con- position vector which is a copy of the pole and that
sider the reflections u, and ul, with poles P, and it is this copy which is then transformed or other-
PI, respectively, and write, from (301, wise left invariant.)
Consider now an arbitrary operation g of O(3)
= '2,lel + '2,Ze2 + P2,3e3f (31) that transforms the vector f k into another one, 8 i . h
(see Fig. 1). We assert that g z h coincides with the
Ul = P,,,e, + P,,ze2 + '1,3e3. (32)
pole of the operation ghg-l of O(3). In fact,
Therefore,

Let us call g the element of the Clifford algebra


that parametrizes the operation g. For a reflection,
it will be given by the right-hand side of (301, and
for a rotation, it will be given in terms of its
corresponding positive quaternion in (6). Similarly,
h will be the element of the Clifford algebra which
parametrizes the operation h. As is clear from (6)
and (30), each g and h will entail a vector part
which is the pole of the corresponding operation.
On the other hand, it follows from (38) that if we
want to transform a vector f h under g it is suffi-
This equation coincides with (15) and thus rela- cient to find the pole of ghg-', which is the vector
tion (16) is established, verifying (so far) the part of ghg-', so that we write
parametrization of reflections given by (30). Notice
that from (36) it follows that u 2 = + E , which is gzh = Ve(ghg-');
correct in the Cartan gauge. Also, to regain (151, no
arbitrary factors are allowed in (31) and (321, ex- (axial vector transformation). (39)
cept for a factor of -1 in both equations, which
merely entails a trivial redefinition of the positive The operator "Ve" here means: "find the com-
hemisphere H. ponents 1, 2, 3 of the Clifford vector (or bivector)

Transformation of a Vector
It will first of all be necessary to make a nota-
tional distinction between a polar vector r and an
axial vector with the same components, to be de-
noted with i.. (Two such vectors will be called
associated.) We have seen in (5) and (13) that for
any rotation or reflection h of O(3) a unique axial
vector f h can be given, which is invariant under
that operation FIGURE 1. Conjugate poles.

INTERNATIONAL JOURNAL OF QUANTUM CHEMISTRY 365


ALTMANN

on its right and use them as components, in the age of a syinmetry operation, in which case ghg-'
right order, of an axial vector spanned by the unit is the pararnetric image of another operation in the
vectors i, j, k in laboratory space." (If h is a same class of conjugation; hence, the norms of the
rotation and h is not a pure quaternion, it must be parameters, as well as those of any vector parts
remembered that the vector part of the quaternion that they might contain, must be left invariant.)
contains the vector in the form n sin$+.) Thus, if h is a rotation, in which case h must be a
In practice, however, we shall most often want quaternion, ghg-' is a rotation of the same angle
to transform a polar vector r. We do this, first, by so that the quaternions h and ghg-' have both the
finding a pole i,, formally equal to r and, second, same scalar part, which allows us to interpret the
by taking the vector part of ghg-', on changing at transforma tion entailed in these equations as one
the same time axial into polar vectors. This, how- that transforms only the underlying vectors, as we
ever, is not all that is required, for the following have assumed.
reason: In ghg-', we are transforming the axial We have two choices as regards the operation h
vector I,, under the operation g and this axial whose pole is going to map the desired vector r.
vector will transform identically with r if and only First, we can take h to be a rotation, in which case
if g is a proper rotation: Axial vectors transform
identically with polar vectors under proper opera- h = l a , ill. (42)
tions but change sign with respect to the transform
of a polar vector under an improper operation The scalar part a of the quaternion in this equa-
(reflection or inversion). Thus, if we now under- tion can be quite arbitrary and we have now
stand V e + to mean "apply V e on the expression dropped thle subscript of the pole. Second, we can
on its right and replace axial vectors by their take h to be a reflection, in which case, from (301,
associated polar vectors," we have
h = ?,el + ?,e, + F,e,. (43)
gr = ve+(ghg-'), g proper; We shall illustrate the first choice for mapping
( polar vector transformation). (40) the vector r in the case when it is desired to
gr = - V e + Cghg-' 1, g improper; transform this vector under g , a proper rotation
with pole P and rotation angle +. Equation (40)
( polar vector transformation). (41) then gives
For simplicity, we have now dropped the sub- R ( + . P )=
~ ve+{Ccost+,Psin~+1Ca,i~
script h from the vector r on the left of these
equations. Therefore, when using them, it must be X [ ; C O S -Psin++]}
+~, (44)
remembered that h is the Clifford-algebra element = Ve+[a, i'], (45)
that maps the operation of which r is the pole. One
important observation must be made about these where the quaternion on the right-hand side is the
equations: The transformed vector g? in the source product of the three quaternions in (44). Hamilton
Eq. (39) is the pole of the operation ghg-' conju- wrote this equation differently. First, he never
'
gate to h. Because h and ghg- belong to the same made a distinction between i and r. Second, he
class of conjugation, both operations must be of took a to vanish (a point about which far too much
the same point-group type (i.e., both rotations by has been made in the literature to this day). Third,
the same angle, or both reflections, etc.) and the he identified the so-called pure quaternion 60, r],
norms of their parametric elements must be pre- now appealring on the right-hand side of (441, with
served. This is the reason why the inner automor- the vector r. Thus, Eq. (44) is written in the Hamil-
phism ghg-' of the Clifford space is one that must ton sense as
preserve the norm of h, a well-known result that
can be directly tested and that it is often, since the
days of the original conical transformation of Cay-
ley and Hamilton, taken to "prove" that the "vec-
tor" h must be obtained from ghg-' by a symme- This is the Cayley-Hamilton conical trunsforma-
try operation, because symmetry operations must tion. Although (46) can be used heuristically and
be norm-preserving. (It should have been recog- although it is undoubtedly convenient to take a in
nized instead that h could be the parametric im- (42) to vanish, the identification of the pure quater-

366 VOL. 60, NO.1


CLIFFORD ALGEBRA, SYMMETRIES, AND VECTORS

nion [O,r] with the vector r is imprudent, since with that of Pgl, and all that we require in order
from comparison with (6) this quaternion maps a to apply the symbol is to remember to replace Y by
rotation by r r (binary rotation) and does not have its corresponding polar vector r. We thus write
the transformation properties of a vector. Thus,
although the Ve operator in (39) to (41) may ap- crr = -PuyPu, (51)
pear as somewhat clumsy, it does precisely the
operations that are left dangerously implicit in the but it must be understood that this is mere short-
more standard notation. Obviously, once one is hand: On the right-hand side of this expression,
absolutely sure of what one is doing, a simplified one is simply multiplying symmetry operations
notation in the style of (46) may be adopted, as rather than operating in any strict sense with vec-
will later on be done in this article. tors and one must also remember that the Clifford
We shall now use the transformation method components must be properly assembled into a
given by (41) in order to find the transform v r of a laboratory-axes vector.
(polar) position vector r under a reflection v on a We shall first of all use the transformation en-
plane through the origin 0 with pole P. This is a tailed in (491, for which purpose we write
rather trivial geometrical result:
[O,ZJ = + f , j + F,k
F,i
vr = r - 2(r P)P, (47)
= Fle,e2 + + F3e2el,
*
F2eie3 (52)
but its undoubted validity will provide a good
check on our method. [Notice that in this section and Pu as in (48). Therefore,
we identify explicitly axial vectors with a tilde and
that in (47) P must be replaced, as we have done, (
P , [ Y o , ~ ~=~ PFlel
, + F2e2 + P 3 e 3 )
by its associated polar vector.]
To do this work, we shall first of all rewrite (41) x(F,e,e, + f2e,e, + f,e2e,)
for the two choices for h given by (42) and (43). x ( F l e l + F2e2+ P3e,) (53)
We must take g in (41) to be a reflection cr with
pole P, whose parametric Clifford-algebra element = (I; - P)P {P x (I; x P)}li
-
will be written, from (30), as follows:
-{P x (i x P))2 j
P , = P,e,
- - -
+ P,e, + P,e,. -(P X (? X P)},k,
(48) (54)

It is very easy to prove, bearing in mind that P as it can easily be proved after some simple alge-
is a unit vector, that PwP, equals unity; hence, P, bra. On introducing (54) into (49), we obtain
is its own inverse. Thus, we have (first choice for
h) crr = -(r - P)P + {P x (r x PI}
= r - Z(r * P)P, (55)
crr = -Ve+{P,[O,?]q,} (49)
which, from (51), we can write as
or (second choice for h )
vr = -PurP, = r - Z(r - P)P. (56)
vr = -Ve+(P,(?,e, + f 2 e 2 + F,e,)P,}. (50)
Equation (56) agrees with the geometrical result
This last expression will be much simplified so (47). I should like to stress that the object of this
as to adopt a notation more familiar in the litera- test is the following: Equation (47) is correct be-
ture. First, we notice that as a difference with (49) yond any reasonable doubt; hence, its recovery in
there is no scalar part in the Clifford algebra prod- (56) validates our operating rules and, in particu-
uct and therefore that the vector to be extracted lar, our contention that the elements of the Clifford
from the right-hand side of (50) is obtained from algebra cannot be identified with vectors but that
the whole of the Clifford algebra element. This they are the parametric images of reflections and
property allows the latter, and, thus, the curly that here, as in the quaternion algebra, vectors can
bracket in (501, to be regarded as "vectors." Thus, only be manipulated by means of products of their
the symbol " V e + " can now be dispensed with, image symmetry operations. Moreover, the axial
because the axial nature of the vector in P , cancels property of the vectors used is established beyond

INTERNATIONAL JOURNAL OF QUANTUM CHEMISTRY 367


ALTMANN

reasonable doubt, since, otherwise, the vital sign in It must be remembered that in the active pic-
(49) would not be obtained. ture the translation, like all symmetry operations,
The geometrical result (55) can be identically does not clhange the symmetry elements at all.
reproduced on imaging the vector r by (43) instead Thus, the poles at 0 and 0’, e.g., are not trans-
of (42), i.e., by using (50) instead of (49). Although formed by the translation in (57). To relate Pxt and
(50) is more compact than is (49), the work is very Px under the translation u, dummy copies of these
much the same and it will not be reproduced here. vectors must be superimposed on them with a
The fact that the two different approaches agree view to transforming these dummies. They, how-
entirely is, however, not without significance, as ever, cannot be free vectors, since their points of
will be discussed in a later section on vectors and application are vital. Neither can they be trans-
bivectors. lated as the position vectors of the space are, in
(571, since they would then cease to be of unit
length. We must therefore define the translation by
The Euclidean Group: Translations u of the pole Pg (strictly speaking, that of its
dummy copy) as merely translating its point of
We want to consider now the Euclidean group. application. This is quite legitimate because all
Let us call ( g } the set of all point-group operations that we need is a relation between Pg. and Pg.
g E 0(3), such that their fixed points coincide with Under the irule just given, it is clear that the pole
the origin 0 of the space. We assume that these Px8 of an operation g’ must be given by upg,
operations have all been parametrized as de- where Pg is the pole of some operation g at 0. It is
scribed above and, in particular, that their poles very easy to prove, in the same manner as in (38),
have all been determined and that they form a set that upg is the pole of ugu-l so that g‘ must
(Pg}. We have seen in (40) and (41) how such coincide with this product. We can thus obtain the
point-group operations transform a position (polar) action on a vector r of any point-group operation
vector r. In the Euclidean group, however, at any of the Euclidean group, g’, the fixed-point of which
point 0’ of the space, we have another copy of has position vector u with respect to 0:
O(3) with point-group operation g ’ with fixed
point 0’ and another set of poles (Pgr](see Fig. 2). g’r = ugu-lr = u(r - u) = g(r - u) + u. (58)
We propose to express all the point-group opera-
tions of the Euclidean group in terms of the point- It is useful to recognize that, if the vector u is
group operations at the origin 0 of the laboratory taken to be a continuous function of a real parame-
space. We must therefore be able to express an ter T (which might be the time), Eq. (58) describes
operation g ‘ in terms of the point-group opera- a general screw in laboratory space. The point-
tions { g } at 0. Call for this purpose u the position group operation g will in this case be a rotation,
vector of 0’ (with respect, of course, to the origin which may be taken, in general, to be also a
0 of the laboratory space) and consider a transla- continuous function of 7 and, if convenient, to be
tion u by the vector u, which transforms all posi- an infinitesimal rotation.
tion vectors of the space as follows: We shall illustrate the general procedure given
by (58) for a reflection u ’ (see Fig. 3). On introduc-
ur = r + u. (57) ing (51) in that equation, we obtain

v’r = -P,,(Y - u ) P ,+u


= -PurP,, + P,,uP, + u. (59)

From (56),

PUuPu= -U +2 ( ~ PIP
. = -U + 2v, (60)

where v is the perpendicular vector from 0 to the


plane u ’. On introducing this result into (59), we
get
FIGURE 2. Parametrization of symmetry operations
centered at different points. u ’ r = -P,rP, + 2v. (61)

368 VOL. 60,NO. 1


ZLIFFORD ALGEBRA, SYMMETRIES, AND VECTORS

It is asserted that the first of these elements, r, can


be called a vector because on changing the sign of
the Clifford units (this, presumably, is thought to
be the passive picture of the inversion) its compc-
nents do change signs as vectors should do. The
bivector s, instead, does not change sign under
such ”operation” and it is thus said to behave like
an axial vector (or a pseudovector, although a
distinction between these two objects should be
In’ - made [12]). I , on the other hand, is in the same
o V manner identified with a pseudoscalar, since it
changes sign under the so-called inversion. This
FIGURE 3. A reflection CT’ at a plane that does not appears to work very well, because the pseu-
contain the origin. doscalar multiplied by the pseudovector should
give a vector, and it is, in fact, obvious that Is
We shall now show that translations may also does indeed give r. However convincing this argu-
be expressed within this algebra. Consider two ment appears, it has nothing whatever to do, un-
successive parallel reflections u ’u where the per- fortunately, with the distinction between Gibbs
pendicular vector from 0 (which is on the plane polar and axial vectors which must be done (in the
a ) to the plane u ’ is v. From (611, passive picture) by applying the inversion on the
unit vectors i, j, k, of the laboratory space, not on
a‘ar = -P,(ar)P, + 2v. (62) those of an arbitrarily defined vector space. Before
this important point is discussed in detail, the
On using (56) here, it follows that following remark is useful: In accordance with the
argument just given, the bivector in (52) should
u’ur = -Pu(-PgrPu)~+
, 2v (63) not behave in the same manner as does the vector
in (431, whereas we have seen after Eq. (56) that
= P:rP: + 2v = r + 2v, (64) both of them behave beyond any reasonable doubt
in identically the same manner as axial vectors. It
since, as already remarked, P , equals unity. This is useful to note that the careful consideration of
result shows that, as is well known, two successive the inversion of the laboratory-space coordinates is
reflections on two parallel planes such that the very important in the Lorentz group, where only
perpendicular distance from the first to the second the space coordinates can be inverted [19].
plane is given by the vector v is a translation t by It must be strongly stressed that axiality or
2v, thus completing the description of all the oper- polarity is a property that can only be discussed
ations of the Euclidean group within the Clifford for vectors defined in the laboratory space. Thus,
algebra. whereas this property is significant for the vectors
r and s in laboratory space that can be extracted
from r and s in (65) and (66), it makes little sense
Vectors and Bivectors for the vectors in the Clifford space. To hammer
this last point home, let us transform under the
I shall now deal with a troublesome fallacy. inversion the vectors r and s associated with r and
When using Clifford multivector algebras, it is s in (65) and (66). As always, we perform this
sometimes asserted that the old-fashioned con- operation within the active picture which requires,
cepts of polar and axial vectors are superseded in in principle, the use of the rule given in (41) for
these algebras. The reasoning behind this assertion applying an improper rotation on a vector. (Notice
appears to go roughly as follows: Define the fol- that this operation has nothing whatever to do
lowing elements of the algebra: with a transformation, whether active or passive,
in the space spanned by the Clifford units.) Since
r = + r2e2 + r3e3,
r,el (65) we assert, however, that the components rl, y 2 , rg
s = rle3e2 + r2ele3 + Y3e2el, (66) in (65) behave like the components of the axial
+
vector r = r,i + r2 j r3k, i.e., that they do not
I = e1e2e3. (67) change sign under the (active) inversion in the

INTERNATIONAL JOURNAL OF QUANTUM CHEMISTRY 369


ALTMANN

space spanned by i, j, and k, we shall revert to the are


use of the transformation rule for axial vectors
(39). We must, of course, change the notation a
little. First, f, will be replaced by r and, accord-
ingly, h must be replaced by Y from (65). Also, the
operation g will be chosen to be the inversion,
whereby g must be taken to be I from (67). It is
easy to prove that I-' equals e3e2e1. Finally, it
must be remembered that "Ve" requires that we In the first column here, we list the eight Clif-
extract the components 1, 2, and 3 of the expres- ford algebra elements and, on their right, in the
sion on its right and that we apply them, respec- correct order, the operations of the full rotation
tively, on the unit vectors i, j, k of the laboratory group that they map, for (72) from (19) and for (73)
space on which r has been defined. We therefore from (20) to (22). It should be noted that the
write Clifford bivectors are the three orthogonal binary
rotations that are the generators of the spinor rep-
ir = = Ve{e,e2e,(rle,
IYI-' + r2e2 + r3e3)e3e2el} resentations of S0(3), as shown in [3], p. 117.
(68) Likewise, we have seen in (26) that the mapping of
the trivector onto the inversion entails the Cartan
= Ve{r,el + r2e2 + r,e,}, (69) gauge. All this means that the elements of C, will
as can easily be verified. Clearly, the right-hand be isomorphic to spinor representations of O(3)
side of (69) gives and are capable of generating all of them. Because
the spinor representations are projective ones, if it
ir = r,i + r, j + r,k = r, (70) is desired to generate the corresponding groups,
thus verifying that, as asserted, r is an axial vector. projective factors of tl must be ignored when
This result, of course, merely demonstrates the obtaining cllosure. It is thus easy to verify that (71)
consistency of our work. The crucially important and (73) give the dihedral group D, under this
result, instead, is that we obtain precisely the same rule.
transformation rule for a bivector, thus giving the We can now deal with the Clifford algebra in
lie to the assumption that Clifford vectors and four dimensions, C,, the elements of which, with
bivectors behave differently under the inversion. It their identification (in the same order as given in
is, in fact, extremely easy to prove that (70) is still the first column), are listed below:
valid when s is substituted for Y in (68). [Both this
result and (70) were, of course, to be expected,
since the inversion commutes with all symmetry
operations: This is another clear example of how
transparent the meaning of products in the Clif-
ford space is, once their nature in terms of symme-
try operations is understood.] This proof not only
corroborates once more that the work on the trans-
formation of vectors that we have given is correct,
but it reinforces our statement that the property of
axiality has nothing whatever to do with the dis-
tinction between vectors and bivectors in the Clif- The identification of the symmetry operations is
ford space. very simply done by plugging them into an equa-
tion like (68). When this is done for e3e2, e.g., it is
found that the x and w components are invariant
Point Groups in Three and Four and the other two change sign. Since this opera-
Dimensions tion, like e3e2 before, must be a binary rotation
under which all vectors operate in the same man-
I shall first review briefly how point groups in ner, it means that it must be a binary rotation
three dimensions are generated by the Clifford about the hyperline xw.In the same manner, e4e,
algebra C,, which are listed below. The elements is a binary rotation about the hyperline yz. The

370 VOL. 60, NO. 1


CLIFFORD ALGEBRA, SYMMETRIES, AND VECTORS

product of e3e2 with e4e, is easily proved to be The latter group is a dihedral group with a
e1e2e3e4, which is a binary rotation of the whole binary rotation C, of the whole space, orthogonal
space (it changes all four components and behaves to the six binary rotations in (86). These examples
like a binary rotation around the center of a strictly will suffice to show how easily it is to generate
two-dimensional plane). It is thus sensible to iden- point groups in four dimensions by means of the
tify e4e, with a rotation Cix,orthogonal to CZx, in Clifford algebra, but it must be remembered that
analogy to the fact that, in three dimensions, the closure is only satisfied on disregarding the projec-
product of two perpendicular binaries is a binary tive factors since, strictly speaking, the Clifford
rotation perpendicular to the plane of the two algebra will always generate the covering or dou-
factors. ble group.
Going back to (76), the same test, remembering
now that under reflection axial components paral-
lel to the reflection plane change sign, shows that
e, is a reflection on the hyperplane yzw, normal to Conclusions
the x direction. The operation e,e,e, in (781, in-
stead, is a reflection on a plane normal to the The main object of this article was that of pro-
hyperplane xyz which is parallel to zu-hence, the viding a number of simple geometrical tests in the
symbol adopted. laboratory space in order to understand the prop-
It is useful to notice in (77) that all the un- erties of Clifford-space vectors and to construct a
primed operations commute, including the projec- workable parametrization of O(3). These tests al-
tive factor, and they likewise commute with the low us to obtain practical rules of transformation
primed operations with the same label. The primed of laboratory-space vectors under the most general
operations commute only if projective factors are transformations of the Euclidean group (such as
disregarded. Also, the operation in (79) when mul- screws). This formalism will thus be useful in any
tiplied by the binary rotations in (77) transforms work that requires the tracking of motions in Eu-
them in other binary rotations of the same list, in clidean space. The very simple geometrical tests
such a way that the unprimed rotations transform here provided show clearly that no element of a
into their orthogonal primed ones and vice versa. Clifford algebra can be used to directly map a
Just like in 0(3), the binary rotations C 2 x , CZy, vector. The Clifford space, in fact, is the parametric
C,, are the group generators and the six opera- space of O(3).Vectors in the laboratory space must
tions in (77) are the generators of 0(4), which
first be mapped by poles of rotations or reflections,
agrees with Wulfman [20], although he lists them
these operations being the entities directly mapped
in a different way (see also [21].)
by elements of the Clifford algebra. When this is
It is very easy from (75) to (79) to deduce finite
point subgroups of 0(4), which will be designated done with some care, the standard rules of trans-
with a notation similar to that used for those in formation of the Gibbs vectors are preserved with
0(3), but with a distinguishing superscript denot- no trouble.
ing the dimension of the space. The first few, e.g., The superiority of the Clifford over the quater-
are nion algebra must be noticed since, whereas in the
quaternion algebra the inversion, and thus all im-
proper operations, must be handled on an ad hoc
basis, the inversion, reflections, and translations
are all correctly mapped by the elements of the
Clifford algebra. Most importantly, it follows from
the present work that both Clifford vectors and
bivectors may be associated with laboratory-space
axial vectors and that much care must be exer-
cised in defining axiality or polarity when using
Clifford-algebra methods. Finally, the utility of
Clifford algebras in forming point groups in 4-
dimensional space was demonstrated and further
applications along this line are in preparation.

INTERNATIONAL JOURNAL OF QUANTUM CHEMISTRY 371


ALTMANN

ACKNOWLEDGMENTS 5. D. Hestenes, Space-Time Algebra (Gordon and Breach, New


York, 1966).
I should like to acknowledge stimulating corre- 6. D. Hestenes and G. Sobczyk, Clifford Algebra to Geometric
spondence with the late Professor Ken Greider. I Calculus. A Unified Language for Mathematics and Physics
am also grateful to Dr. M. F. Ross and Professor W. (Reidel, Dordrecht, 1984).
M. Pezzaglia for providing me with copies of their 7. K. Greider, Founds. Phys. 14,467 (1984).
theses and to Professor Pezzaglia for very useful 8. P. Lounesto, Founds. Phys. 11, 721 (1981).
correspondence. Professor H. K. Urbantke from 9. J. S. R. Chisholm and A. K. Common, Eds., Clzfford Algebras
and Their Applications in Mathematical Physics (Reidel, Dor-
Vienna very kindly read a draft of this article and drecht, 1986).
contributed useful comments, for which I am 10. D. Hestenes, Am. J. Phys. 39, 1013 (1971).
grateful, as I am for a discussion with Professor J. 11. W. M. Pezzaglia, Doctoral Thesis (University of California,
Morzymas from Wroclaw. A preliminary version Davis, 1975).
of this article was presented at the Second Interna- 12. W. E. Bayliss, J. Huschilt, and J. Wei, Am. J. Phys. 60, 788
tional School of Theoretical Physics at Poznafi, in (1992).
August 1992. This work was supported by a travel 13. E. Cartan, Legons sur la Th&oorie des Spineurs (Hermann,
grant of the Royal Society and by the Austrian Paris, 1938).
Ministry of Science under Project GZ 49.731/2- 14. M. Atiyah, R. Bott, and A. Shapiro, Topology 3 (Suppl. 11, 3
(1964).
24/91, for which help I am most grateful.
15. F. R. Harvey, Spinors and Calibrations (Academic Press,
London, 1990).
16. S. L. Altmann, Icons and Symmetries (Clarendon Press, Ox-
References ford, 1992).
17. S. L. Altmann, J. Phys. A: Math. Gen. 20, 4587 (1987).
1. W. R. Hamilton, Lectures on Quaternions (Hodges & Smith, 18. L. C. Biedcnharn and J. D. Louck, Angular Momentum in
Dublin, 1853). Quantum Physics. Theory and Application (Addison Wesley,
2. M. Riesz, Clifford Numbers and Spinors, Lecture Series No. 38 Reading, MA, 1981).
(Institute for Fluid Dynamics and Applied Mathematics, 19. R. U. Sex1 and H. K. Urbantke, Relativitut, Gruppen, Teilchen,
Baltimore, MD, 1958). 3rd ed. (Springer, Vienna, 1992).
3. S. L. Altmann, Rotations, Quaternions, and Double Groups 20. C. E. Wulfman, in Group Theory and Applications E. M.
(Clarendon Press, Oxford, 1986). Loebl, Ed. (Academic Press, New York, 1971), Vol. 2, p. 151.
4. S. L. Altmann and P. Herzig, Point-Group Theory Tables 21. H. S. M. Coxeter and W. 0. J. Moser, Generators and Rela-
(Clarendon Press, Oxford, 1994). tions for Discrete Groups, 4th ed. (Springer, Berlin, 1984).

372 VOL. 60.NO. 1

Potrebbero piacerti anche