Clifford algebras and Dirac operators in harmonic analysis JOHN E. GILBERT University of Texas MARGARET A.M. MURRAY Virginia Polytechnic Institute and State University The rth ofthe University of Combridge "Print ond sell all manne of books wor pronted by Henry VI in 1336 The University hs printed ond, ‘continuo since 1586 a CAMBRIDGE UNIVERSITY PRESS Cambridge New York Port Chester Melbourne SydneyPublished by the Press Syndicate of the University of Cambridge The Pitt Building, Trumpington Street, Cambridge CB2 1RP 40 West 20th Street, New York, NY 10011,-4211 USA 10 Stamford Road, Oakleigh, Melbourne 3166, Australia © Cambridge University Press 1991 First published 1991 Printed in Great Britain at the University Press, Cambridge British Library cataloguing in publication data Gilbert, John E. Clifford algebras and Dirac operators in harmonic analysis. 1. Harmonie oscillations. Analysis. I. Title II. Murray, Margaret A.M 515, 2433 Library of Congress cataloguing in publication data available ISBN 0 521 34654 | hardbackweet Contents Introduction Chapter 1 Clifford algebras Quadratic spaces Clifford algebras Structure of Clifford algebras Orthogonal transformations Transformers, Clifford groups Spin groups The Euclidean case Spin(V, Q) as a Lie group Spin groups as classical Lie groups Notes and remarks for chapter 1 Chapter 2 Dirac operators and Clifford analyticity Cauchy-Riemann operators Dirac operators past and present Clifford analyticity Spaces of analytic functions Spaces of Clifford analytic functions I: the upper half-space Cauchy integrals and Hilbert transforms on Lipschitz domains Spaces of Clifford analytic functions II: Lipschitz domains Notes and remarks for chapter 2 108 119 125 135 140anene akon oakwene Contents Chapter 3 Representations of Spin(\',Q) Elements of representation theory Signature, fundamental representations, Class 1 representations Polynomials of matrix argument. Harmonic polynomials of matrix argument Notes and remarks for chapter 3 Chapter 4 Constant coefficient operators of Dirac type First-order systems: some general results Operators of Dirac type Rotation-invariant systems The operators 6, Critical indices of subharmonicity Notes and remarks for chapter 4 Chapter 5 Dirac operators and manifolds Local theory Global theory Dirac operators on hyperbolic and spherical space Representation theory for Sping(n, 1) Asymptotics for heat kernels The index theorem for Dirac operators Notes and remarks for chapter 5 References Index 143 144 147 164 173 193 201 203 203 208 213 220 232 244 247 248 263 272 284 296 309 317 321 328To Vicki Gilbert, for her love, patience, and understanding, and to Magdalene K. Murray, for her courageous and independent spirit.Introduction In this book we present a comprehensive introduction to the use of Clifford algebras and Dirac operators in harmonic analysis and analy- sis more generally. In the past 30 years, Clifford algebras and Dirac operators have played a key role in three of the most important ar- eas of mathematical research during that time: the boundedness of the Cauchy integral on Lipschitz surfaces, the realization of discrete series representations of semi-simple Lie groups, and the celebrated Atiyah— Singer index theorem. Much as an analyst would like to understand and appreciate these developments, however, there are formidable technical barriers to doing so, particularly for more classically trained analysts, as we have found to our cost over the years. Thus our aim from the outset has been to meld into a coherent and reasonably self-contained whole a body of ideas from classical singular integral theory, representation theory and analysis on manifolds, with a view to making this material accessible to more classically trained analysts. Now the starting point for much of classical harmonic analysis is the study of the boundary regularity of harmonic functions in domains in Euclidean space. Classical Hardy space theory explores the consequences of the improved boundary regularity obtained when consideration is restricted to analytic functions in the plane. On the other hand, for SL(2,IR), the starting-point for representation theory of semi-simple Lie groups, some important unitary representations become irreducible only on restriction to analytic functions. It may be a dramatic overstatement to characterize analytic functions as those in the kernel of a first-orderelliptic differential operator — the Cauchy-Riemann 9 operator — which factors the Laplacian and has rotation-invariant symbol; but it is pre- cisely such properties that one looks for in differential operators on more general manifolds. For one can then develop a Hardy H? theory on Eu- clidean space including an analysis of elliptic boundary value problems, as well as explicit realizations of semi-simple Lie groups on associated symmetric spaces. Index theorems arise in both cases, of course. Dirac operators and their generalization, the so-called operators of Dirac type, have such properties. Much earlier, quite independently of all these analytic ideas, Clifford introduced his algebras as a common generalization of Grassmann’s ex- terior algebra and Hamilton’s quaternions, both of which sought to cap- ture the geometric and algebraic properties of Euclidean space. Indeed, Clifford used the name ‘geometric algebras’ for his algebras quite appro- priately, because the universal Clifford algebra for IR” is the minimal enlargement of IR” to an associative algebra capturing precisely the al- gebraic, geometric and metric properties of Euclidean space. It is not surprising, therefore, that the bundle formed by the Clifford algebra of the tangent space at each point of a manifold should be so important in the geometric analysis of that manifold. In chapter 1 we present, the general theory of Clifford algebras in an elementary and thoroughgoing fashion, which should be accessible to the algebra ‘neophyte’; it is our aim to give a coherent account of mate- rial which is presently scattered throughout the literature with no one account being readily accessible. In chapter 2 we quickly review the classical Hardy space theory and its extension to minimally smooth do- mains, and then develop a higher-dimensional analogue for this theory based upon functions in the kernel of the Dirac operator. In chapter 3 we explore further the connections between Clifford algebras and represen- tations of the spin group and of the rotation group. Then, in chapter 4, we define a more general notion of operators of Dirac type, and show that. all of the important rotation-invariant geometric differential opera- tors of Euclidean analysis are in fact of Dirac type. Finally, in chapter 5 we introduce and then study Clifford algebras and Dirac operators on more general manifolds, concluding with a recent simplified proof of the local Atiyah-Singer index theorem. This book had its beginnings in the fall of 1985, when one of us (M.M.) was a visiting faculty member at the University of Texas at Austin. To whatever extent we have succeeded in our goal, we owe a debt of thanks to many of our friends and colleagues who have made this successpossible. In particular, we wish to thank René Beerends, Klaus Bichteler, Chris Meaney and John Ryan for numerous helpful discussions, but most of all we wish to thank Kathy Davis, Gene Fabes and Ray Kunze for immeasurable help in the formulation of ideas going into the book, as well as in the writing of the book. One of us (M.M.) would like to acknowledge the particular help and support of her good friends and colleagues Daniel Farkas, and Carol and Frank Burch-Brown, without whom this work might never have come to fruition. Partial support from the National Science Foundation is acknowledged by both of us, too. Finally, we wish to express our tremendous gratitude to Margaret Combs, whose patience, skill, and craftsmanship produced such a mar- velous typescript.1 Clifford algebras Associated with any Euclidean space IR" or Minkowski space R”’ is a universal Clifford algebra, denoted by %, and Up,9, respectively. Rough- ly speaking, a Clifford algebra is an associative algebra with unit into which a given Euclidean or Minkowski space may be embedded, in which the corresponding quadratic form may be expressed as the negative of a square. The real numbers R, the complex numbers C, and the quater- nions HH are the simplest examples. Our intent in this chapter is to give an elementary, coherent, and largely self-contained account of the theory of Clifford algebras. In sec- tions 1 and 2 we present the definitions basic to all of our work. The bal- ance of section 2 is devoted to three constructive proofs of the existence of universal Clifford algebras: two basis-free constructions using tensor algebras and exterior algebras, and a basis-dependent construction. The reader who is willing to accept the existence of Clifford algebras may wish to proceed directly to the statement of the major structural results in section 3. Sections 4, 5, and 6 explore the interconnections between Clif- ford algebras and orthogonal groups; the spin representation and spin groups will be studied in detail, with Spin(p,q) and Spin(p, q+ 1) both being realized in Wp, using the notion of transformers. The reader who is primarily interested in the analytic applications of Clifford algebras may wish to proceed directly to the discussion of the Euclidean case in section 7. Section 8 is a discussion of spin groups as Lie groups. In sec- tion 9 we construct various realizations of Spin(p, q), p + q < 6, whereby these groups are explicitly identified with classical Lie groups.6 Clifford algebras 1 Quadratic spaces Let V be a finite-dimensional vector space over the scalar field IF, where IF=RorC. A quadratic form on V is a mapping Q : V > F such that (L1G) QQv) =7Qv), AEF, veV, (1.1)(ii) the associated form Biv,w) = 3{ Q(v) + Aw) - Qw-w) }, vwevVv, is bilinear. When such a Q exists, the pair (V,Q) is said to be a quadratic space; every vector space over IF becomes a quadratic space with respect to the trivial quadratic form Q = 0, for instance. The significance of condition (1.1)(ii) is that the form B defines an inner product on V x V, and so all the usual geometric properties of inner product spaces can be exploited. Typically, a quadratic space arises from an inner product space, defining Q on V by, say, Q(v) = (v | v) where (- | -) is the inner product on V xV. For example, if (| -) is the usual Euclidean inner product on IR” and |v|? = (v | v), then both (IR”,| - |?) and (IR”, —| - |?) are real quadratic spaces with associated bilinear forms (- | :) and —(- | -) respectively. More gencrally, let p,q be non-negative integers with p+q > 0 and define a pseudo-Euclidean or Minkowski quadratic form on R?*? by (1.2) Qpalte) = (upto + up) + (urge Fug gy), w= (urs, Uptg) 3 the corresponding real quadratic space we shall call Minkowski space and denote it by (IR? Q),q)- Clearly (R™°, Qn,o) reduces to (IR®, —| - |?), while (R°", Qo,n) is just (IR",|-|?). By convention, R°° = {0}. In the complex case, (€”, Qn) becomes a complex quadratic space on setting (1.3) Qu(z) = apte teh, r= (zy. en) 5 note that in (1.3) the associated form B,(z,w) = zw, +-+' + 2nWn is complex linear in w, not conjugate-linear as in the usual inner product. on €”. Now let (V,Q) be an arbitrary quadratic space and {e;} a basis for V. Then Qe) = D7 Blej,ex)ujre, v= Dye; , 7 ik and if there is a basis which is B-orthogonal in the sense that B(e;,ex) = O when j # k, the expression for Q(v) reduces to diagonal form Av) = Ac)7, v= ye; a aQuadratic spaces 7 Such a basis is easily constructed. Let Rad(V,@Q) = { weV: Biv,w) =0, allveV }=vt be the radical of (V,Q). We say that (V,Q) is non-degenerate if Rad(V,Q) = {0}; otherwise (V,Q) is degenerate, in which case V can be written as the B-orthogonal direct sum V = Rad(V, Q) © Rad(V,Q)+ of Rad(V, Q) and its B-orthogonal complement. Clearly (Rad(V,Q)+, Q) is non-degenerate, and B-orthogonal bases {e;}, Q(e;) 4 0, can be con- structed in the usual way for Rad(V,Q)+, or for V if £V,Q) is already non-degenerate. Using (1.1)(i) to normalize the e,, we can also assume that Q(e;) = +1 when F = R, while Q(e;) = 1 when F = €. Now augment this basis by any basis of Rad(V,Q) if (V,Q) is degenerate. Since Q is trivial on Rad(V,Q), we thus obtain a basis {e;} for V such that (1.4) (i) Bej,en)=0, JAK, (ii) {e; : Q(e;) = 0} is a basis for Rad(V,Q) , (iii) {e; : Q(e;) £0} is a basis for Rad(V,Q)+ such that Q(e;) = +1 when F=R, while Q(e;) =1 when F=C. With some abuse of customary terminology, a basis for V satisfying (1.4)(i), (ii), (iii) will be said to be a normalized basis; many of the algebraic constructions to be discussed are conveniently given using such a basis. For instance, from such a basis it follows that every quadratic space (V,Q) is the sum of the particular examples given already. More precisely, we have the following. (1.5) Theorem. Let (V,Q) be a quadratic space with B-orthogonal decomposition V =Rad(V, Q) © Rad(V,@)*. Then (a) Q=0 on Rad(V,Q), (b) when IF = R, (Rad(V,Q)+,Q) is isomorphic to R?4 where p,q depend only on Q, (c) when F = C, (Rad(V,Q)+,Q) is isomorphic to (€",Qn) where n depends only on Q. Proof. Part (a) is an immediate consequence of the definition of8 Clifford algebras Rad(V,Q). If Rad(V,Q) # V, choose a basis {e;} satisfying (1.4)(i), (ii), (iii). In the case F = R this basis can be indexed so that Qu) = O(Dwses) =u a) + ga tot he)» 7 and dim(V,Q)+ = p+ q > 0; by Sylvester’s theorem, the values of p,q do not vary with the choice of basis. Part (b) is now clear, and part (c) is proved in the same way. a 2 Clifford algebras As Clifford’s paper introducing ‘geometric algebras’ shows, Clifford based his ideas on the common features he saw in the construction of Grassmann’s algebra and Hamilton’s quaternions. In the framework of modern algebra we shall derive both constructions simultaneously be- ginning with an arbitrary quadratic space (V,Q), V a finite-dimensional vector space over IF. Let A be an associative algebra over F with identity land v: V — A an F-linear embedding of V into A. (2.1) Definition. The pair (A,v) is said to be a Clifford algebra for (V,Q) when (i) A is generated as an algebra by {v(v) : v € V} and {AL: AE F}, (i) (Ve)? =-Q@)L, — allveV. Roughly speaking, therefore, condition (ii) ensures that A is an al- gebra in which there exists a ‘square root’ of the quadratic form —Q; condition (i) is a minimality restriction on the ‘size’ of A. Some simple examples illustrate how this definition contains the al- gebras whose structure prompted Clifford to introduce ‘geometric alge- bras’. (2.2) Examples. (i) When Q = 0 on V, let A be the exterior algebra A"\(V) = Seo A*(V) with n = dimV, A°(V) © B, and Al(V) & V, and let v: V — A(V). Since every element of A*(V), k > 2, is of the form v1 A--- A ug, clearly A*(V) is generated by {v(v) : v € V} and {Al : A € F}; in addition, (Vv)? = vAv=0= -Qv)l . Hence the Grassmann algebra A*(V) is a Clifford algebra for (V, Q) when Q=0.Clifford algebras 9 ii ine the Pauli matrices in (ii) Define the Paul c?*? by (2.3) o=[} 9]. o=[b SJ. o=[ oJ. a=[2 and associated Pauli matrices by (2.4) ee A ed ed As elements of the associative algebra €2%?, o=oi=o}=03=1, =I, € while (2.5) O50 = —i9e , een = ee when {j,k, €} is a cyclic permutation of {1,2,3}. These matrices will occur throughout the theory of Clifford algebras. For instance, set Woo = (Avo: AER}, wun{ly Hssen}. wwe([5 toveah, and 9 ian to+iz, a2+ir3| . LE 2 22 7 woo = {[ ey ay — ie |) 77 ER —m 4) '4SCf- Then each of these is an associative subalgebra (over R) of C2*? having an identity, and (2.6) WolR, MoZRER, WiC, We2=H where HH is Hamilton’s algebra of quaternions. As the notation suggests, Wp,q also is a Clifford algebra for IR’ with respective embeddings v given by 070, yyos, yr yer, (1,22) > r1e1 + r2e2 - In each case the proof amounts to a simple computation using properties of the oj and e;. For instance, in Wo, 2 (v(a1,2))” = (wie1 + x22)? = —(a} + 23)e9 since the cyclic permutation property of {e1, ¢2,e3} ensures that e,e2 + ¢e, =0; alternatively, by direct calculation we see that 7 2 iz, © ate 2a,|l1 0 [‘2 23] a t+28)() alee (iii) As an associative algebra over C, the matrix algebra €?*? is a10 Clifford algebras Clifford algebra for (C?, Q2) with 0 2 — 129 2 + tz 0 Again the proof is just a simple computation. Some elementary properties of a Clifford algebra follow readily from its definition. Let (V,Q) be a quadratic space of dimension n and (A, v) a Clifford algebra for (V,Q). It is very convenient not to make distinction between A € IF and Al € A, or between v € V and y(v) € A. With this understanding, A is generated by F and V; in addition v? = —Q(v). Similarly, if W is a subspace of V, then the subalgebra of A generated by F and W is a Clifford algebra for (W,Q). On the other hand, for arbitrary u,v in V (u +0) = —Q(u + ») = -2B(u,v) — Q(u) — Qe) = -2B(u,v) tu? +0; vi@ G0? vs (24,2) but, by direct expansion, (utvP=wtuvtevute?. Thus the algebra structure on A enables us to express the inner product. BonVxVas Blu, v) = —}(uv + vu) ; in particular, (2.7) een + exe; = —2Q(€;) bj 5 L where m; = 0 or 1 and e9e$ -- -e9 is interpreted as theClifford algebras ll identity in A. The linear span of all such reduced products must then be A, since A is generated as an algebra by V and F. These reduced products are very convenient to use both in devel- oping the general theory of Clifford algebras and in studying particu- lar examples, so it is worth having a simple notation for them. Let Ny = {1,2,...,dim V} and for each non-empty subset a of Ny set (2.10) Ca = Ca," Cay > a={a,...,an}, Loay<-++ B such that p = Gov and B(a) = In. The fundamental result of this section is that for each quadratic space (V,Q) there is a universal Clifford algebra. Since V and F generate évery Clifford algebra, it is virtually obvious that this universal Clif ford algebra is unique up to isomorphism; hence we shall denote it by Y(V, Q), or, more briefly, W when (V,Q) is understood, and speak of the universal Clifford algebra for (V,Q). In particular, we shall use the nota- tion W,, Wn, and C,, for the universal Clifford algebras UR", Qp,q), AR", Qon), and W(C", Q,) respectively. The next result relates the dimension of a Clifford algebra to its universality. (2.14) Theorem. A Clifford algebra (A, v) for (V,Q) is universal when dim A = 2%™V, Proof. Let {aj} be a normalized basis for V and (B,) an arbitrary Clifford algebra for (V,@). Defining ej = v(a;) in A and fj = y(xs) in B, we deduce from (2.7) that (2.15) even tenes =0, G#k, ef =-Q(z)la, while (2.16) fifet+ frfi=0, J#k, FF = —Q(a;)1p - Now the reduced products {e,} form a basis for A since A has maxi- mum dimension, and the reduced products {f.} span B. Thus the linear extension of the map f : €. — fa is a well-defined linear mapping from A onto B such that p = Sov on V and f(1q) = 1g. In addition, (2.15) and (2.16) ensure that BlCa + €6) = BET +--em «ey +e) = Bley) = fy = fa fo» establishing the required algebra homomorphism property of 8. Hence (A, v) is universal when dim A = 24m, a (2.17) Theorem. For every quadratic space (V,Q) there exists a universal Clifford al- gebra U(V,Q). Furthermore, every universal Clifford algebra has mazi- mum dimension. To prove this theorem we give three explicit constructions of a uni- versal Clifford algebra for (V,Q) having maximum dimension. SinceClifford algebras 13 universal Clifford algebras are unique up to isomorphism, it will then follow that a Clifford algebra is universal if and only if it has maximum dimension. First CONSTRUCTION. In view of theorem (1.5), every (real) quadratic space (V,Q) can be written as an orthogonal finite direct sum (V,Q) = (Vi, Q1) @ ++: ® (Ve, Qr) of quadratic spaces (V;,Qj;) where Q; = 0 on Vj or (Vj;,Q;) is isomor- phic to one of R™° or R°. In the complex case, R'° and R®! have to be replaced by (€',Qi). As we have seen, each of these quadratic spaces has a corresponding two-dimensional Clifford algebra, so theorem (2.14) ensures that for each (Vj, Qj) there is a universal Clifford algebra U(V;,Q;) with dim U(V;,Q;) = 2%". To complete the proof we shall combine these individual algebras into a universal Clifford algebra for (V,Q) using the following result which is of independent interest. (2.18) Theorem. Let (Ai,1), (Az,~2) be Clifford algebras for quadratic spaces (Vi,Qi), (V2,Q2). Then there is a linear isomorphism v from Vi & Vo into Ay @ Ag and an associative multiplicative structure on A; @ Az so that (A; @ Ao, v) is a Clifford algebra for (Vi ® V2, Qi © Qa). Applying theorem (2.18), we obtain a Clifford algebra (A, v) for (V, Q) with A=M(Vi,Q1)B--@A(V,,Qr), dim A = 24%... asim Ye — gdm Thus A has maximum dimension, and so (A,v) is a universal Clifford algebra for (V, Q). To prove theorem (2.18) a Z(2)-grading on a Clifford algebra is needed; this grading will be an integral part of its algebraic structure. (2.19) Definition. When (A,v) is a Clifford algebra for (V,Q) denote by A* and A~ the subspaces of A generated by the products of even and odd numbers of elements of V, respectively. Thus A = A+ @A~ and F C A+, V C A~. Suppose now that {e,} is a normalized basis for V; and if a C Ny, let |a| denote the cardinality of a. Then the set of reduced products eq with a C Ny and lal even spans A*, while the corresponding set of eg with a C Ny and |a| odd spans14 Clifford algebras A~. Since the product of any two elements in {eg : a € Ny, |a| even} again belongs to this set, we see that A* is a subalgebra of A; similarly, A~ is an At-module and products of elements from A~ belong to A*. In simple terms, therefore, (2.20) AtAtT=A*, A-AtT=A™>, ATA =At, which expresses the Z(2)-grading of A by At, A>. Proof of theorem (2.18). Define a linear mapping v from V; © V2 into A; ® Ap by V2 (01,02) 9 Uy Ble+1, Bw, vp EV, where 1, is the identity in Aj. This clearly is a linear isomorphism. To introduce a multiplicative structure on Ai @ Ag, first define products of elements a; ® a2, a, @ a2 where aj,a’, € A¥ by (a1 ® a2) (a4, ® a3) = (-1)** (aia, @ aap) with 0, meAt, ,_f% meat, e= f= 1, a2€ AZ, 1, aye Ar, and then extend by linearity to all of Ai @ A» using the fact that A; = Af © Aj and that every element of A: @ Ay is a linear combina- tion of elements of the form a; ® a2. Straightforward calculations show that A, @ Ag is an associative algebra with identity 1, ® 1, under this multiplication; in addition, A @ Aa is generated by {v(v1, v2) : vj € Vj} together with {A(1; @ lz) : \ € IF}. To complete the proof we have to show that. 2 (2.21) (v(wr,v2))" = —(@i(v1) + Qe(v2))1i @12, Yj EV; It is precisely at this point that the form of the ‘twisted’ multiplicative structure on A; @ A: is needed. Indeed, (v(x, v2)" = (v1 @ la +h @ v2)? = (of @ ly +11 @ 03) + (v1 @ la)(1r @ v2) + (11 @ v2)(v1 @ Ly) = -(Qi(r) + Q2(v2))h @ le +01 Bv2-— 11 Bra, since 1; € A} and v; € Aj. This establishes (2.21). ' SECOND CONSTRUCTION. We shall obtain the universal Clifford algebra %1(V,@) in very specific terms as a subalgebra of the algebra £(A*(V)) of linear transformations of A*(V). Recall that A"(V) = D7.) A“(V) where n =dimV, A°(V) =Clifford algebras 15 FF, V = A1(V) and each element of A‘(V) is a sum of wedge products vy A+++ Ave. For each v in V we define M, and 6, in £(A*(V)) as the linear extensions of MV =v, Mv Av Ave) vA A Ave and ‘ So(1) =0, — By(vpAs=-Ave) = D>(-1)“1 Bw, ve)Or A: AoA: “Ave 5 k=l , meaning as usual that v, is omitted from the product. The operator M, is just exterior multiplication by v, while 6, is interior multiplication with respect to the inner product induced on V x V by B; clearly 6, is the adjoint of M,. In the physics literature the operators M,,6y are referred to as creation and annihilation operators, respectively. Now define v : V — L(A*(V)) by v(v) = My — 6y; by definition v(v)1 = v, so v clearly is a linear isomorphism. We shall show that the subalgebra A of £(A*(V)) generated by {v(v) : v € V} and {Al : \ € IF} is a Clifford algebra for (V,Q) of maximum dimension. This will then be a universal Clifford algebra for (V,Q). The proof rests on three simple properties of the operators M,, 6,: (2.22) M2=8=0, 6, (wAz)=Blv,w)z —wAd,(z) for any w € V and z € A*(V). The first of these is an immediate consequence of the property v Av = 0. To establish the second, observe that 82(u, A-++ Ave) é S2(H1)FB(0, ve) (01 A= A Be Nv A 08) m i SO (HVT B(y, vj) Bw, va)Ur Ao AG] Av NGA Ave ISjcksé = YS dt *B(v, 05) Bw, ver A AGA AG Av NUE. lsk Av Ave + DE =0 for some & € A*(V). Since @ maps {AI : \ € FF} onto A°(V) and {v(v) : v € V} onto Al(V), an obvious induction argument shows that @ maps A onto A*(V). But then dim A > dimA*(V) = 2". Hence A must have maximum dimension, and so be a universal Clifford algebra for (V,Q). When Q = 0 on V, the realization of M(V, Q) in the second construc- tion reduces to the left regular representation of A*(V) on itself because 6, = 0 in this case. Even when Q # 0, however, the mapping @ defined by (2.23) identifies 2(V,Q) linearly with A*(V) in a completely explicit manner. THIRD CONSTRUCTION. This construction will establish directly the existence of a Clifford algebra for (V,Q) having the universal property rather than one having maximum dimension, but it does use the second construction in the process. Let T,(V) be the k-fold tensor product V ®---®V with Ty(V) = F and T;(V) = V, and let T(V) = >y2.9 Tk(V) be the tensor algebra over V. Let Ig be the two-sided ideal generated in T(V) by {v@v+ Q(v)1:Clifford algebras 17 v € V}, so that Ig is spanned by { u® (v@vt+ Qr)l) @wiuweET(V), vEV } > let Y be the quotient algebra T(V)/Ig, and let 7 : T(V) + W be the associated homomorphism. Define v : V — & by v(v) = mv). We shall prove that (2,v) is the required algebra. Now certainly 2% is an associative algebra with identity ly = 7(1) and is generated by {v(v) : v € V} and {An(I) : A € FF}, since T(V) is generated by V and IF. Also by construction, v(v)? = —Q(v)la. Thus (Q,v) will be a Clifford algebra for (V,Q) provided v : V > & is 1-1. To establish this and the universal property, first let (A, a) be an arbitrary Clifford algebra for (V,Q). By the universal property of T(V) with respect to associative algebras, there is an algebra homomorphism a’! : T(V) > A such that a/(1) = 14 and a/(v) = a(v), v eV. But then a (u(v @ v + Qvr)L)w) = a'(u) (aw)? + Q(v)1A)o'(w) = al (u)(a(v)? + Q(v)1a)a’(w) =0 for all u,w in T(V) and v in V. Thus Ig C ker(a’); hence a’: T(V) + A factors through Q%, so there is an algebra homomorphism 6: % > A for which a! = Go 7. Clearly then @(1lqy) = 14 while ov =a. This establishes the universal property, leaving only the 1-1 property of v to be established. Choose for (A, @) the subalgebra of £(A*(V)) generated by {M, — 6, : vu © V} and {AI : \ € IF}. By what we have just proved, there is an algebra homomorphism f from YW into £(A*(V)) such that B(V(v)) = My — 6, v € V. Since then @(v(v))1 = (My — 6y)1 = v, clearly 30 7 is the identity on V. Hence » must be 1-1, completing the construction. ' Three operations on Y(V,Q) will be basic to all that follows. To illustrate the various ideas developed in this section, we shall give both basis-dependent and basis-free definitions. The principal automorphism on U(V, Q), denoted by (’), is the algebra automorphism defined on basis elements €, by (2.24) é=(-1llen, anv, where |a| is the cardinality of @. The principal anti-automorphism, denoted by (*), is defined on basis elements €g by 1 (2.25) ef = (-1)2lalllal-De, while conjugation, denoted by (~), is the composition (2.26) Bu = (@5) = (ee) = (192MM,18 Clifford algebras of the principal automorphism and anti-automorphism. The term rever- sion is sometimes used instead of ‘principal anti-automorphism’ because (2.27) €4 = (Car €a2***Cax)” = Cay *** Cara - To prove this observe first that e;¢, = —exe; when j # k; thus ax 1° Caza, = ~Cax*** Cas €o Car = (“1 Peas Cage Calas = *** = (DEHN e605 °° bay = (-DEIMAID eg , establishing (2.27). That. (eaes)’ = eye, — (eaes)* = eet, hold as the names suggest can be checked by easy computation. The alternative basis-free approach to defining (’) and (*) makes these prop- erties clearer, however. Indeed, define a linear mapping ': V — %(V,Q) by v! = —v. Then (v")? = —Q(v), so, by the universal property, this lin- ear mapping extends uniquely to an algebra automorphism on (V,Q) which clearly is given by (2.24) on e,. Now define the anti-automorphism. (*) on the tensor algebra T(V) = p29 Tk(V) by (v1 @ +++ B ve) =U WB ; this leaves invariant the two-sided ideal Ig generated in T(V) by {v ® v+Q(v)I : uv € V}. Passing to the quotient T(V)/Ig, we obtain an anti-automorphism on (V,Q) given by (2.27) on elements eq. If U(V,Q) has a matrix realization, these three operations are often familiar ones from linear algebra. (2.28) Examples. (i) When C is a realization of %, under the embedding v : 2 — iz, then » a =Zz, zec, where Z is the usual conjugation on C; in particular, conjugation on C as a Clifford algebra coincides with the usual notion of conjugation. (ii) When C?*? is a realization of %3,9 under the embedding Zo ot = v: (0,21, 22) —> ered taeClifford algebras 19 |. of R*° as the traceless hermitian matrices in €?*?, then a b)’_[d -~@ a b]"_ fa ce ad} |-b a]? [ec a] [6 ab) _[d - e dj |-c al Consequently, in this interpretation of €?*? as a universal Clifford al- gebra, A* is the usual adjoint (= conjugate-transpose) of A, while A is the adjugate of A; in particular, AA = AA = (det A). (iii) When C2? is a realization of %,,2 under the embedding Zo a +izg —(x1 — ix2) —Zo : [ea -[ a] Ey -[% a]. fe eee terete So again A is the adjugate matrix of A, and AA = AA = (det A). These expressions are easily established using only routine calculations. They also generalize, replacing © by any universal Clifford algebra U(Vo, Qo) and €?%? by M(2,9(Vo, Qo) = {[2 ‘| :a,b,e,d€ MV, a0)} ; where (Vo, Qo) is a real, non-degenerate quadratic space. aor vy: (£9, 01,22) —> [ then (2.29) Theorem. Under the quadratic form Q0,2,y)=-(? +¥°+Qo(v)) (EW), on V = Vo ® R*”, there is a realization of U(V,Q) as M(2, (Vo, Qo)) such that [Eal-[% 2] > (aye Bee ee with respect to the corresponding operations on W(Vo, Qo). oe ear (2.30)20 Clifford algebras Just as theorem (2.29) reduces to (2.28)(ii) when M(Vo,Qo) = C, so the next result reduces to (2.28)(iii) when W(Vo, Qo) = C. (2.31) Theorem. Under the quadratic form Q,2,y) =Qolv) +y?-2? (VE Ve), on V = Vo @R", there is a realization of U(V,Q) as M(2, W(Vo, Qo)) such that [EJ- e] ey -[5 a, [- a] -[% F]. with respect to the corresponding operations on U(Vo, Qo). (2.32) Proof of theorem (2.29). Let {e1,...,€n} be a normalized basis for (Vo, Qo), n = dim Vo, and define 7,...,7%m-+1 in M(2, {(Vo, Qo)) by 1 0 0 e : ol w= | ro [ 2 j| ssn), m=[} 0| : Then ye + WIG = 2jeQo(e;) (LS G,k face adh ] = opt yy ty Is (v= Lue) j=l j= defines a linear embedding of V = Vo @R?°, in M(2,U(Vo, Qo)) satis- fying the Clifford condition with respect to the quadratic form (2.34) Q(v, 2,9) = —(0? + ¥? + Qolv)) on V. Now the universal Clifford algebra 2(Vo,—Qo) is realized in M(2, U(Vo, Qo)) by wv-a0)={[5 2 7 | :=€M*(V,0)} ; (Vo, —Qo) = {[é 5] £6 eM (Vo Qo)} ;Clifford algebras 21 these are generated by {7; Ya On the other hand, the universal Clifford algebra M29 of R®° is realized in M(2, M(Vo, Qo)) by M(2,R) = {[2 | :a,tedeR} : and this algebra is generated by {7o, ¥n+1}- Jointly, therefore, these two generating sets generate all of M(2, (Vo, Qo)). Since this last algebra has maximum dimension 4 - 24 Vo = 24imV (2, (Vo, Qo)) is a re- alization of U(V,Q). Straightforward but lengthy calculations establish (2.30). ' Proof of theorem (2.31). Define the basic elements 70, ...,7m41 now by 1 0 0 « . 0 1 w=[9 fai w-|2 | asim). ma=[% a]: and let. . Eo z ytou (2:35) von —[ ty UA] be the corresponding embedding of V = Vo @ R* in M(2, (Vo, Qo))- The Clifford condition is satisfied with respect to the quadratic form (2.36) Q(v,2,y) = y? — 2? + Qo(v) on V. Only trivial modifications to the proof of the previous theorem are required to complete the present. proof. ' Restricting theorems (2.29) and (2.31) to subspaces of Vp 9 R®° and Vo @ R™" respectively, we deduce the following. (2.37) Theorem. Under the ar form Q(v, 2) = -(2? + Qo(w)) (v € Vo) on V =Vo @R"™, there is a realization of U(V,Q) as the subalgebra {fz S]:scem.an} of M(2, U(Vo, Qo))- (2.38) Theorem. Under the quadratic form Q(v,2) = Qo(v) +2? (vE Vo)22 Clifford algebras on V=Vo@R"", there is a realization of U(V,Q) as the subalgebra {[2, §]:2cea,00)} of M(2, U(Vo, Qo)). Repeated use of these last two theorems actually provides yet another proof of theorem (2.17), but the main use of all four theorems will be in recognizing various matrix algebras as Clifford algebras and in con- structing the spin groups which will play such an important role later. 3 Structure of Clifford algebras The fundamental use of Clifford algebras in studying Dirac operators and Lie groups arises from basic structural identifications of 2I(V,Q) to be made in this section. It is not surprising that these identifications depend on whether F is R or C, but it is surprising that they depend on the parity of the dimension of V. For instance, example (2.2)(iii) shows that (3.1) UC ,Q)=CoC, AC’,Q)2C0%, so that the universal Clifford algebra ©, is isomorphic to the full matrix algebra €?*?, whereas ©, is isomorphic to the direct. sum of two full matrix algebras, regarding € as C'*'. These results are typical of the structure of Q(V,Q) for any non-degenerate complex quadratic space (V,Q). (3.2) Theorem. Let U(V, Q) be the universal Clifford algebra for a non-degenerate com- plex quadratic space (V,Q). Then (i) UV, Q) ts isomorphic to the full matric algebra C2" *?” when dim V = 2m, (ii) U(V,Q) is isomorphic to the direct sum €2"*?" @ €2"*?" when dim V = 2m +1. From theorem (3.2) the structural identification of W(V,Q) for any complex quadratic space (V, Q) follows easily. For (V,Q) can be written as a B-orthogonal direct sum (V,Q) = Rad(V, Q) ® Rad(V, Q)* of quadratic spaces where in one case Q = 0, while in the other Q is non-degenerate. But as we have seen (Rad(V, Q)) = A*(Rad(V,Q)) .Structure of Clifford algebras 23 Together with theorems (2.18) and (3.2), this identifies 2(V,Q) with one of the tensor products AMC) @ C2"*2",, At(C*) 8 (C2"2" 9 @2"2") according as dimV = k+2m or dimV = k+2m+1. It must be remembered, of course, that the multiplicative structure on these tensor products is the twisted multiplicative structure defined in the proof of theorem (2.18), thus eliminating any simple-minded proof based on that theorem. Nonetheless, this failure brings out the essential element of a correct proof. Indeed, suppose dim V = 2m and identify (V,Q) with the m-fold B-orthogonal direct sum (V,Q) = (€?,Q2) ®--- 8 (C*,Q2) - Then, by (3.1) and theorem (2.18), 2(V,Q) is isomorphic to the m-fold tensor product. UV.Q)=C?@..e0”? which in turn is isomorphic to €?"*"_ Because of the twisted multi- plicative structure on the tensor product, however, this last isomorphism is only a linear isomorphism when €?"*?” has its usual multiplicative structure. Thus, the linear isomorphism from V = €? @ --- @C? into c?"*2" (~ ©?*? g...@C2%?) defined by (3.8) Hiv = (V1, .--,Um) 4 UB12@W++-@lmt-+-+11B-+-@1m-19Um as in the proof of theorem (2.18) does not have the property p(v)? = —Q(v) with respect to the usual multiplication on €?”*2". The two proofs of theorem (3.2) that we shall give both amount to defining a linear isomorphism j: which does have the property y(v)? = —Q(v) in (?"*2" | Hence, by the universal property, extends to an algebra homomorphism 8 from Q(V, Q) into ©?" *?". Had we been able te apply theorem (2.18), we would have been able to deduce at once that the range of this homomorphism is all of €?"*?”, establishing the required isomorphism because dimW(V,Q) = 22" = dim€?"*?". Thus the second step in both proofs is to show that the dimension of 3(2I(V,Q)) = 2” In the first proof this is done directly by dimension and structural arguments which will be useful in both the real and the complex case, while the second uses a Wedderburn-type argument which brings out interesting algebraic properties of 2(V,Q). Similar comments apply also when V has odd dimension. As preparation for the proofs of theorem (3.2) we prove several useful results. For the moment no restriction is placed on the scalar field, i-e., F can be R or C, and (A,v) will be an arbitrary Clifford algebra for a24 Clifford algebras quadratic space (V,Q) over IF. Fix a normalized basis {e;} in V so that the set {eq : a C Ny} of reduced products spans A. (3.4) Lemma. For each a © Ny and basis element e; @) ejea =(-1)!*leae;, J ¢as (ii) ejea = (-1)*!"teae;, G Ea, where |a| is the cardinality of a. In particular, ej¢a = eae; for all j=1,...,dimV, éf and only if a =0 or Ny; the latter case can occur only when V is odd-dimensional. Proof. All the results follow from successive use of the anti-commutation property ejex + ee; = 0, j # k. For, if a = {a1,...,ax} and j ¢ a, Cjla = Cjlay Cay = (~l)earljlar *** Car = (-1)?ea,a2€y** Cay = 21° = (“1)!lenes 5 whereas, if j € a, one fewer (—1) factor will arise from the need to permute e; with all but one of the basis elements in eg. Thus ¢j€a = €a€; if and only if |a| is even and j ¢ a, or |a| is odd and j € a. Hence €j€a = Cae; for all j = 1,...,dimV, if and only if a = 0 or a= Ny and V is odd-dimensional. 1. The particular reduced product e€,€2--+€2m41, dim V = 2m + 1, will have a special role to play in the structural theory, so we shall denote it. by ey. The name pseudo-scalar is often used for ey. (3.5) Lemma. Suppose (V,Q) is non-degenerate and Dycy, Ages = 0 for a choice of \g € IF. Then either dg = 0 or Ay + Avev = 0; the latter case can only occur when V is odd-dimensional. Proof. ‘The non-degeneracy ensures that each e; is invertible in A since ej = £1. But then, if Dacy, Aves = 0, certainly =f fo(E wast +(E woh BONY 1 as = Yoda 5 (eresey* +e0) = D0 Aveo BON Bec where C={ BON: eres — epey }-Structure of Clifford algebras 25 Repeating this successively with €2,...,€dimv in place of e;, we deduce that gep Asa = 0, the sum now being taken just over the set D={ BCNy:e;e9 =ege; forall je Ny}. By lemma (3.4) the only possible choices for 9 are 3 = 9 or, when V is odd-dimensional, 6 = Ny. Hence »g = 0 or V is odd-dimensional and Ag +Avev = 0. ' (3.6) Corollary. Let (V,Q) be a non-degenerate quadratic space and suppose that either dim V is even, or dimV is odd and ey ¢ IF. Then every Clifford algebra for (V,Q) has maximum dimension. Proof. It is enough to show that the spanning set {eg : a C Ny} is linearly independent. Suppose then that }>, A€a = 0 for some choice of {Aa} © FF. Now each reduced product is invertible because (V,Q) is non-degenerate. Thus, for fixed 6 C Ny, Xo + > Aetaes) = (© Data)es! =0 ax5 o Hence As = 0 or V is odd-dimensional and As + Ayey = 0 for some y C Ny. In the latter case, however, ey would have to belong to F contrary to hypothesis. This establishes linear independence. a We can now give a direct constructive proof of theorem (3.2). First proof of (3.2)(i). When dim V = 2m, define Ey,...,E2m in the m-fold tensor product €2*? @ ---@ €?*? = €?"%?" py E,=01 8-80, 802818: @l, (3.7) l ALF, +++ + ems Famsi + Indeed, the first of the equalities in FP=([@--@NDO(U@--@D, FjyFet+FeFj=0 (#8), is clear, while the second follows from (3.10) since 0; @ --- @ 0 anti- commutes with each E;. Consequently, when A is the algebra generated by {i\j Fj : Aj € F} and {A @---@1) 6 (1 @--- @1): XE C}, the pair (A, ¥) is a Clifford algebra for (V,Q) which by corollary (3.6) willStructure of Clifford algebras 27 be of maximum dimension provided that (-1)"i FF: +++ Fm+1 is not a complex scalar. To evaluate Fi Fp +++ Fom41 first note that Ey E+ ++ Em = 02(01)"~! @ o2(01)"-? @--- ® 0201 @02 , while Emmet Em+2 +++ Eom = 03(01)"7* ® 03(01)"? © +++ @ a301 Bas - Thus, using (2.5) together with the fact that o? = J, we obtain E, Ez: ++ E2m = 02(01)™—103(01)"~! @ o2(o1)™? o3(01)""? @--- @ az03 = (-1)™"1a203 ® (-1)""?a203 @ --- @ 0203 =OmI1 BO @-+ Bor, where Om = (—1)2™"+Di". In this case Fi Fa+++ Fong = Om(I @ +++ @1)@(-(1@-:- @D) does not belong to IF. Corollary (3.6) now ensures that the Clifford algebra (A,v) has maximum dimension 2?”+!, so A is the direct sum of two copies of €?*? @---@€?*?. This completes the first proof of theorem (3.2). . The second proof rests on results describing the simplicity or semi- simplicity of a Clifford algebra. It will be instructive to present. this second proof as much as possible parallel to the first one, so once again for the moment. (V, Q) is an arbitrary quadratic space over F = R or C, and (A, v) a Clifford algebra for (V,Q). Let {e;} be a normalized basis of V. (3.11) Lemma. The center 3, of (A,v) consists of all x in A such that ejx = xe; for allj € Ny. Proof. Clearly ejz = xe; for all e; when « belongs to the center 3a. Conversely, if ej2 = xe; for all e;, x will commute with every reduced product €, = €a,--*€a,; since these reduced products span A, every such z must lie in 34. ' Lemma (3.4) now gives 3a explicitly at least for non-degenerate (V, Q). (3.12) Corollary. Suppose (V,Q) is a non-degenerate quadratic space and (A,v) a Clif-28 Clifford algebras ford algebra for (V,Q). Then 3,4 = F when V is even-dimensional, whereas 34 =F @ Fey when V is odd-dimensional. The non-degeneracy assumption in the corollary above is essential, since lemma (3.11) ensures that every wedge product €; A ex lies in the center of A*(V). Proof of (3.12). Non-degeneracy ensures that each ¢; is invertible in A. Let ¢ = Ygcm, Aveo be an element in 34. Then ej2 = ze; for all e;, and so in particular z= ff{erey' +z} = > grslereser’ +ea)= D Avea BONY pec where C = {8 C Ny : e1€g = egei}. Repeating this successively with €2,.--,€dimv in place of e;, we deduce that « = Deep Ages where D={BCNy:ejeg =ege;, forall jE Ny}. By lemma (3.4) 8 must be empty, or else # = Ny and V is odd- dimensional. Hence z = XJ or, if V is odd-dimensional, x = AJ + Avev. This proves the corollary. a Next we consider the ideal structure of (A,v), beginning with the following result. (3.13) Lemma. Suppose (V,Q) is non-degenerate and I is a non-trivial ideal in A. Then I contains the identity 1, or V is odd-dimensional and I contains an element of the form 1+ Avey for some dy in F. Proof. Let « = Dacny Ha€a be a non-zero element in I with, say, Hs # 0, for some 6 C Ny. But then, since jseg is invertible, the element ao = (uses) 12 = > (ug "Hades lea aCNy =1+ > Ape BEB also belongs to J where, crucially, B does not contain the empty set 0. The ideal I thus also contains the element 21 = 3{ermoey' + co} =1+ D> Ag(erepey + ep) bes =1+ > Ages BEB, where now By = {BE Bi eyes — eger}Structure of Clifford algebras 29 Repeating this successively with e2,...,€dimv in place of e;, we finally obtain an element y in J such that y = 1+ Dyce Asea where C= {8 € B: e;eg =ege; , forall 7 ¢ Ny}. Since 0 ¢ B, lemma (3.4) ensures that either C is empty, or V is odd- dimensional and C consists solely of the set Ny. The lemma follows immediately. ' Thus if (V, Q) is an even-dimensional non-degenerate quadratic space, any Clifford algebra (A, v) for (V, Q) is central and simple, in the sense that 34 = IF and J = A when J is a non-trivial ideal in A. In view of Wedderburn’s theorem for central simple algebras therefore, it is now not so surprising that A must be isomorphic to €?”"*?" when V is a complex space with dimension 2m. A proof using just Clifford algebra ideas can be given as follows. Second proof of (3.2)(i). By corollary (3.6), A has maximum dimension; consequently, we can assume A = 2(V,Q). Let {e; : 7 = 1,...,2m} be a normalized basis for V as usual and let (V’,Q’) be the complex quadratic space obtained by restricting Q to the subspace V’ of V having basis {e; : j = 1,...,m}. Since the universal Clifford algebra & = &(V’,Q’) has dimension 2™, we shall exhibit the isomorphism &(V,Q) ~ €?"*?” by constructing an alge- bra isomorphism + from %(V,Q) onto the algebra £(Q’) of all linear transformations on Y’. When {eq : a © Nyy} is a basis for 2’, define 6:V — £(2’) as the linear extension of (i) d(ejJea = eae; , (ii) S(em+Jea =(—1)liejea (1S j 2?". Simi- larly, defining 6. : V’ + I_ in an analogous way, we obtain dimJ_ > 2. Tt follows that dim I, = dimI_ = 2" , and that each is isomorphic to C(V’,Q’) & €?"*?"_ Hence A=hel =a" 9a" | completing the proof of theorem (3.2)(ii) once again. . We conclude this section with a characterization of the structure of real Clifford algebras. Suppose that (V, Q) is a non-degenerate quadratic space over IR, which is isomorphic to R”? with p-+q =n. Now %(V,Q) is isomorphic to a subalgebra of €,, (considered as an algebra over IR). To see this, let {e1,€2,...,€n} be a normalized basis for (C",Q,), so that {e, : a € {1,2,...,n}} spans €,, as a complex vector space. Let {fi,---, fn} be a normalized basis for (V,Q), ordered so that QA) =" = Qo) =-1, — Qfp41) = ++ = WSpta) = 1- Then the map »: V > €,,, given by w(f;)= te), LS5SP, wfors)=eptd, 1SISG and extended by linearity, satisfies 4(v)? = —Q(v) for allu € V. ByStructure of Clifford algebras 31 universality, 2 extends to an injection 9 of W(V,Q) into €,, which is a homomorphism of real algebras. Thus %(V,Q) is isomorphic to a 2”-dimensional real subalgebra of €,,. The following result is now im- mediate from theorem (3.2). (3.16) Corollary. Let 4(V,Q) be the universal Clifford algebra for a non-degenerate real quadratic space (V,Q). Then (i) UW(V,Q) is isomorphic to a real subalgebra of the matriz algebra Cx" when dimV = 2m, (ii) UV, Q) is isormorphic to a real subalgebra of the direct sum C?” *?” et?"*?” when dimV = 2m +1. As an easy consequence of our work in this section, we can characterize all possible Clifford algebras for (V,Q) in terms of dimension: (3.17) Theorem. Let A be a Clifford algebra for a non-degenerate real quadratic space (V,Q). Suppose further that (V,Q) is isomorphic to R?4, and let {€1,€2,-..,€p+q} be @ normalized basis for (V,Q) such that e? = +1 forl (Ai,---)Ap+q) is an orthogonal isomorphism from (V,Q) onto (IR’'",Q,,q). Similarly, when (V,Q) is a complex quadratic space of dimension n, there is a corresponding orthogonal isomorphism from (V,Q) onto (€", Qn). (ii) Fix v in (V,Q) with Q(v) # 0 and define S, : V V by 2B(u,v) Sou) =u Bay 8 uevVv. Since B(S.(u), Sy(w)) = Bu,w), uwev, each such S,, belongs to O(V, Q). Geometrically, S,(u) is a reflection of u in the subspace (IFv)+ of V which is B-orthogonal to Fv. Indeed, Bu, v) B(u,v) wm ape Gee) vey, is the decomposition of u with respect to the B-orthogonal direct sum V = (Fv) 6 (Fv)*, and Bu,v) Blu, v) ss) = a + (um GP) Clearly, (Sy 0 Sy)(u) =u, ie, 5? = Ion V. Under composition O(V, Q) is a group which we shall call the orthog- onal group associated with (V,Q). Example (4.3)(ii) shows that O(V, Q) contains all reflections {S, : Q(v) # 0}, hence all finite products of such reflections. In fact, more is true.34 Clifford algebras (4.4) Theorem. Each S in O(V,Q), S# I, can be written as the product of at most 2n reflections, n = dimV. The proof of theorem (4.4) is by induction on the dimension of V. When n = 1 the group O(V, Q) reduces to {+I} and —I can be written as the reflection S, for any non-zero v in V. So suppose that theorem (4.4) has been established for all quadratic spaces of dimension less than n. To establish the result for an n-dimensional space the following lemma is needed. (4.5) Lemma. Let u,v be elements of (V,Q) with Q(u) = Q(v) 40. Then u can be mapped to v by the product of at most two reflections. Proof. Since Biu-v, w+v) = Q(u) - Q(v) =0, the element u ~ v is B-orthogonal to u + v. In this case Q(u tv) + Q(u—v) = 2{Q(u) + Qv)} and so at least one of Q(u+v), Q(u—v) is non-zero. Now, if Q(u—v) # 0, Su-v(u) = Su-v($(u—v) + 3(u +) =—f(u—v) + Z(utv) =v. On the other hand, if Q(u+v) #0, Susv(u) = Susv(3(u—v) + $(utv)) = $(u—v) — (ut) = so that So S,4,:u—v. This proves the lemma. a v, To complete the proof of theorem (4.4) let {e;}_1 be a B-orthogonal basis for V and (V‘, Q’) the quadratic space obtained by restricting Q to the subspace of V spanned by {e; : 1 < j < n}. Now, by the preceding lemma, for each T in O(V,Q) there exists an S in O(V,Q) which is the product of at most two reflections and such that (SoT)en = en, because Q(T en) = Q(en) # 0. Consequently, SoT belongs to O(V’, Q’). If SoT # I, the induction hypothesis ensures that T = S-! oT, where T, is the product of at most 2(n —1) reflections, while if SoT = I already T = S-!. Since any reflection is self-inverse, ie., Sy = S51, it follows that T’ must be the product of at most 2n reflections, completing the proof. a The group O(V,Q) and various groups derived from it are funda- mental to all that follows because they will arise both geometrically as transformation groups on the particular differentiable manifolds to beOrthogonal transformations 35 considered and analytically as ‘symmetry groups’ for important differ- ential operators defined on such manifolds. For the remainder of the section (V,Q) will be assumed to be a real quadratic space. The group O(V, Q) often arises then as a subgroup of two important groups of transformations, the rigid motions or isometries and the conformal or Mébius transformations. (4.6) Definition. A mapping from V onto V is said to be a rigid motion (or isometry) when it preserves the distance between pairs of points, i.e., when Q(Tu-Tr)=Q(u-v), uvev. The set of all rigid motions will be denoted by E(V,Q); clearly it is a semigroup under composition, containing the composition (4.7) TiAw) it — Ar+w, cev, of rotation by A in O(V,Q) and translation by w € V. Consequently, E(V,Q) contains the reflection T : x + Syx+w of V in the hyper-plane (Rv)! +w, Q(v) £0, hence all finite products of such reflections. Just as each S in O(V, Q) is the finite product of the reflections S,, Q(v) 4 0, so E(V,Q) can be characterized by the following. (4.8) Theorem. Each T in E(V,Q) is a composition of finitely many reflections + — Syx+w in hyper-planes. Since the composition of finitely many such reflections is of the form T.a.w) for some A in O(V,Q) and w in V, we deduce the following. (4.9) Corollary. A mapping T from V onto V is a rigid motion if and only if it is @ composition T.a,w) of rotation by A in O(V,Q) and translation by w in V. In particular, E(V,Q) is a group isomorphic to the semi-direct product OV, Q)OV = {(A,w) : AE O(V,Q), we V} in which group multiplication is (Ai, w1)(A2, we) = (Ar Ae, Arwe + wi) - Proof of theorem (4.8). Each T in E(V,Q) can be written as Tv = Su +T(0), v € V, where S is an isometry such that S(0) = 0, Q(Sv) = Q(v). Consequently, if S is known in advance to be linear, then S$ will be36 Clifford algebras orthogonal and (4.4) can be applied, completing the proof; since linearity is not a part of (4.7), however, a separate proof is needed. Now, for any uvinV, B(Su, Sv) = }{Q(Su) + Q(Sv) — Q(Su — Sv)} = 3{Q(u) + Q(v) - Qu—v)} = Blu»), so S also preserves inner products. This ensures that the image {Se;} of any normalized basis {e;} of V is again a normalized basis. Conse- quently, for every v = 0; vje; in V, there exist 2; € IR such that Sv= Yo 2;(Se;) > &j = B(Sv,Se;) = Biv, ej) =v; - j Hence Su= (X ¥6) = Soy (Se;) 7 j establishing linearity. Proof of corollary (4.9). The identification of each T in E(V,Q) with an isometry T(4,..) was part of the previous proof. But T(4,y) is invertible within E(V,Q); in fact, (Taw) =Tez), B=A?, z= -Alw. Thus E(V,Q) is a group. On the other hand, Tay wr) © T(Ag,wa) 1 ¥ — (ArAa)v + (Arw2 +0) , so Tiaw) — (A,w) defines an isomorphism from E(V,Q) onto O(V,Q)®V, completing the proof. a The group of rigid motions E(n) associated with Euclidean space R™ is known as the Euclidean motion group, not surprisingly; its counterpart E(3, 1) associated with Minkowski space R®” is often called the Poincaré group, although this term is sometimes reserved for the two-fold covering group of E(3,1). More generally, E(p,q) will denote the group of rigid motions on R?*. Finally, we come to the conformal group C(V,Q) for (V,Q), which contains the motion group E(V, Q) and the orthogonal group O(V, Q) as subgroups. For simplicity here, we shall assume that (V, Q) is isomorphic to Euclidean space IR". Now a smooth mapping ¢ : Up — U; between open sets in V is said to be conformal on Up when for each x in Uo there exist A > 0 and A € O(V,Q) so that ¢/(x) = AA; thus ¢ : Up + Ui is conformal on Up when there is a real-valued function 7 on Uo such that (4.10) jo'(z)v? =e" |v? vEV, cE.Orthogonal transformations 37 By the chain rule, the composition of conformal mappings is again con- formal whenever the composition is defined. For example, each rigid motion T(4,w) : 2 —+ Ar + w obviously has derivative T/4 4) = A, so E(V,Q) is a group of conformal transformations on V. Every dilation a, : 2 — dx, X > 0, of V also will be conformal on V, since a = AI. Thus along with the group of rigid motions, the group {a, : \ > 0} generates the important group P(V, Q) of similarities (4.11) Tawa) iP > AAT Hw, TEV, of V, each of which is conformal on all of V. On the other hand, inversion z (4.12) Jit is well-defined for non-zero x in V, and has derivative "q) = L(y — 2282) 1 We) = Bal Spee) = Tap Se Consequently, J is conformal on V \ {0}. With all this in mind, we make the following definition. (4.13) Definition. The conformal group C(V,Q) of (V,Q) is the group of transformations of V generated by the group P(V,Q) together with the inversion J, i.e., by translations, rotations, dilations and inversion. A conformal sphere in V is any set (4.14) Sw = {ee V:alz|? —22-v4+8 = 0} with (4.15) w=(a,7,8)EROVOR, wF0. The notion of conformal sphere includes spheres in the usual sense (the case a # 0) as well as hyper-planes (the case a = 0). By reflection (or inversion) in a conformal sphere ©,,, we mean the function ¢ defined by jv? — 08 ge) = 0+ joa when © is a sphere, and B ae) = 8.(2) + (soa) » when © is a hyper-plane (i.e., w = (0,v,8)). These are not defined on all of V, but they are conformal where they are defined. We then define a Mobius transformation on (V,Q) to be any finite composition of reflections in conformal spheres. Since any such reflection is self-inverse, the set of all Mébins transformations is a subgroup, called the Mébius38 Clifford algebras group, of C(V,Q). In view of theorems (4.4) and (4.8), the Mébius group contains O(V,Q) and E(V,Q). In fact, there is a characterization of the connected component of the Mébius group which provides the appropriate analogue to these results. (4.16) Theorem. The connected component of the Mébius group coincides with the set of sense-preserving conformal transformations, i.e., with the subgroup of C(V,Q) generated by translations, proper rotations, dilations, and inversion. The proof of this result is somewhat beyond the scope of the present work. We do remark, however, that there are analogous results in the case where (V,Q) is an arbitrary non-degenerate quadratic space. 5 Transformers, Clifford groups Using the multiplicative structure on a Clifford algebra we shall con- struct covering groups for various of the orthogonal groups introduced earlier. Throughout this section (V,Q) will be an arbitrary non- degenerate quadratic space and % = Y%(V,Q) its universal Clifford alge- bra, no restrictions being placed on F unless specifically indicated. Now, because of the Clifford condition v? = —Q(v) on the elements of V, v will be invertible in % if and only if Q(v) # 0. Consequently, the set (5.1) { v1 ---0n 10; EV, Q(v;) #0 } of all finite products of such elements is a subgroup of GL(Q). It is a subgroup of the Clifford group T'(V,Q) in Y to be associated with (V,Q). On the other hand, the orthogonal group O(V, Q) was identified in theorem (4.4) as the subgroup (5.2) OV, Q) = { Soi So_ 209 EV , Qvj) #0} of GL(V) of all finite products of reflections. One of the principal goals of this section will be to extend the mapping v — S, to a covering homomorphism from (5.1) onto O(V,Q), but the route will be far from direct. The simple case of IR*° illustrates the basic ideas involved. Its uni- versal Clifford algebra Mz, can be identified with IR?** under the em- bedding (6.3) vin= (21,02) — moi +2203 =X,Transformers, Clifford groups 39 where X is the traceless symmetric matrix Ty 2 in R?*?, Thus (5.5) Qo0(z) = det(v(x)) =det(X), 2eR*°. This embedding extends to an embedding + (56) v:z=(¢1,22,y) Za yl + X= [’ Tis l t2 y-Z of IR" as the set of all symmetric matrices in R?*?. Again (5.7) Q2,1(2) =det(v(e)) =detZ, ze R*", although the Clifford condition is not now satisfied. In both cases, how- ever, the condition Q(v) # 0 is equivalent to v(v) € GL(2,R). (5.8) Theorem. When Uo,9 is identified with IR°*? by (5.3), Wo coincides with the set A= {Zy---Zq 2 Z; = (zy), 2 € R*} of all finite products of elements from the image of R™! in Woo, while GL(2,R) coincides with the set P= {Zp---Ze2 Zi = (2%) , Qaalz;) #0} of invertible such products. In the terminology to be introduced shortly, GL(2,IR) is the Clifford group of Ur,o, and Wp. itself is the Clifford semigroup of Wo,9. Proof of theorem (5.8). It is more convenient to deal with the invertible case first. Obviously, GL(2,IR) > I’. Conversely, by the polar decom- position of matrices, each A in GL(2,JR) can be written uniquely as A= BZ with B in O(2,R) and Z a positive-definite symmetric matrix in GL(2,IR). But all such Z are already in T, so it is enough to show that T > O(2,R). It is at this point that the subgroup (5.1) of determined by R®® enters. For if EB @ —sin "| = 0<6<27, sin? cos@ then (5.9) SO(2,IR) = {kp 10 <8 < 2n}40 Clifford algebras and (5.10) 0(2,R) = $0(2,R) U [? a S0(2,R) . On the other hand, ol sin@ cos@ (5.11) to=| | [ ; |. 10 cos@ —sin@ Consequently, SO(2,IR) consists of all even finite products in W2,o of elements X, det X = —1, from the image of IR?°, while O(2,IR) consists of all finite products of such elements. Hence T = GL(2, IR). To complete the proof we have to show that every A in IR?*? is a finite product of symmetric matrices even when det A = 0. So suppose a~[2 é| . 0b = 78. Then, if a #0, a-[5 30 IL ol fote °C] On the other hand, if a = 0, at least one of y, 3 must also be zero. But then 0 0] _fo oj[6 ¥ 0 B)_[é s]jo o 7 6} [o 1Jly 6} ’ |o 6]~ |B 6}lo 1] - This completes the proof. Now let o(A) be the modified similarity transformation (5.12) o(A):Z—+ (1/detA)AZA', AEGL(2,R), on R?*?, where At is the transpose of A. The restriction of o(A) to the space of symmetric matrices in IR?*? will determine a linear trans- formation of IR"; in fact, this will be an orthogonal transformation of R?" because det(a(A)Z) = det Z. Thus A — o(A) is a homomorphism from the Clifford group I into the orthogonal group O(2,1) associated with R?!. What is more, by restricting o(A) to traceless, symmetric matrices, we see that o(A) is an orthogonal transformation of R®° when Ais in the subgroup (5.13) {Xi-+ Xe: Xj = (23) , Q20(2s) # OF of GL(2,IR) of all finite products in Wz, of elements from the image of R®° (see (5.10), (5.11)). Thus A — o(A) is a homomorphism from (6.18) into O(2,IR). It can (and will later) be shown that o actually is a covering homomorphism from I onto $O(2,1) and from the subgroup (5.13) of P onto O(2,R). In order to proceed with the general construction, we introduce aTransformers, Clifford groups 41 general substitute for the determinant function on IR?*?. This is the norm function (5.14) A:U— aU, A(a) =Zr. Since (Ar)~ = Az for all \ € F and x € W, clearly A(Ar) = \?A(z); but, in general, A is not a quadratic form on W since A is not F-valued on all of W when dimV > 2. For, if dimV > 3 and c = 1+ e,€2¢3, then a= and A(x) = 2? = 1+ 2ere2€3 + (e,e2€3)? = A+ 2ereE2e3 for some d € IF; however, e,e2€3 cannot be in IF because both 1 and €1€2€3 belong to the basis of all reduced products for W(V,Q). Hence A(z) ¢ F. Nonetheless, on the subset (5.15) N= NR(V,Q) = {x E W(V,Q): A(z) EF } on which A is F-valued the basic properties of the norm function mirror those of the determinant. (5.16) Theorem. (i) The set N is a multiplicative semigroup in UW which is closed under scalar multiplication as well as under the principal automorphism and conjugation; furthermore, A(zy) = A(z)A(y), A(w’) = A(z) = A(a*) = A(z) , ryEN. (ii) Anz in N is invertible if and only if A(x) # 0, and then (5.17) a t=(1/A(e))e , Ale) =1/A(z) - Clearly, R(V, Q) always contains F © V because (5.18) A(A +) = (A= v)(A +0) =? = 0? = 7 + QW). Consequently, the Clifford semigroup (5.19) A(V,Q) = {w1--- we sw; EF OV} of all finite products in % of elements from F © V is a sub-semigroup of R(V,Q). This Clifford semigroup contains every reduced product €a:***€a, in U(V,Q), and in the case of R*° coincides with 2(V,Q); but in general, A(V,Q) # M(V,Q) since A is not always F-valued. Proof of theorem (5.16). If x,y € ®, then Alay) = Ty ry = y(tr)y = A(z) A(y) since A(z), A(y) € IF. If A(x) #0, then 2 is easily seen to be invertible and (5.17) is immediate; moreover, in this case, = A(x)z~! so that A(#) = A(z)2A(a27!) = A(x) «42 Clifford algebras If, on the other hand, A(x) = 0, then z is a zero divisor in 9%, hence not invertible; consequently 2 is not invertible, which forces A(#) = A(z) = 0. Finally, A(e’) = ra! = (#2) = Ala)’ =A) , Ale") =A@')= Ale), whenever r € 2. a From (5.16) and (5.19) we obtain immediately a multiplicative group, the so-called Clifford group T'(V,Q) of &(V,Q), which is the general analogue of the group GL(2, IR) in Uo. (5.20) Corollary. The Clifford group T(V,Q) = {w+ we: wy EF OV , A(wy) £0} of all invertible finite products in U(V,Q) of elements from F@ V is a multiplicative group in U which is closed under the principal automor- phism and anti-automorphism as well as under conjugation. To establish the connection with various orthogonal groups associated with (V,Q), we use a modified similarity transformation analogous to (5.12), regarding the quadratic space (IF@V, A) as an extension of (V,Q), just as IR?" was earlier regarded as an extension of R™° (see (5.18)). (5.21) Definition. An element A in U(V,Q) is said to be a transformer if to each w in FOV there corresponds a z in F@V so that Aw = 2A’. Obviously every element, of FF is a transformer since (5.22) Aw=wr=w', weFeV, AEF. In studying further this notion it is worth singling out the properties of (F@V,A) that are needed: (5.23)(i) VCFOV C R(V,Q), (5.23)(ii) GF @V,A) is a non-degenerate quadratic space, which is closed under the principal automorphism, (5.23) (iii) on (F@ V, A), the principal anti-automorphism is the — identity, i.e.,w* =w for w€FO V. The first result is the principal step along the path to the characteriza- tion of the Clifford group.Transformers, Clifford groups 43, (5.24) Theorem. The set of all transformers is a sub-semigroup of 9(V,Q) which is closed under the principal automorphism. Proof. By definition, to each transformer A and w in F © V there corresponds z in F @V so that Aw’ = zA’. But then Alw = (Aw')! = (2A = 2A. Since 2’ € F @ V (see (5.23)(iii)), it follows that A’ is a transformer. Moreover, if A, B are transformers, then for each w in F@ V there exist n,¢ in FV such that A(Bw) = A(nB’) = (CA')B' = ((AB)’ . Hence, the set of all transformers is a multiplicative semigroup which is closed under the principal automorphism. Finally, to show that A is F-valued on this set, we shall use the following lemma, which is of some independent interest. (5.25) Lemma. Ang in satisfies rv = va’ for all v in V if and only if x € F. Proof of lemma (5.25). As noted already in (5.22), the equality zu = vr’, v €V, holds for each z in F. Conversely, let {e;} be a B-orthogonal basis for V and {e,} the corresponding basis of reduced products for 2. Then, by (3.4), eien =(-D)Mlejen = {00 ES consequently, for any = 37, Aq€a in UY, eja! = (x Aveo) - (= ata) > ‘iga ‘jE while rey = (= dota) + (= ota) 7 ide ja Thus ej2’ = ze; can hold only if \q = 0 for those a containing j; hence va! = xv, v € V, can hold only when x = g, i-c., 2 € F. t To complete the proof of theorem (5.24), fix a transformer A and w in F@V. Then there is a z in F ® V so that A(A)w = A(zA’) = (Az) A! = (zA’)* A’ = (Aw)" A’ = (wA*)A' = w(A(A))’44 Clifford algebras using (5.23)(iii). Now, since V CF @ V (see (5.23)(i)), we have A(Ay = v(A(A))’, -vEV; thus lemma (5.25) ensures that A(A) €1F, completing the proof. The significance of the idea of transformer can now be seen. For if A is an invertible transformer, then by (5.24) A’ also will be invertible, and the modified similarity transformation (5.26) o(A):w> Av(A)"', weFovVv, will define a linear transformation on F @ V. This defines a homomor- phism which by (5.25) has kernel (5.27) kero = {A €F:\ 0}. On the other hand, since A* A’ = (A A)! = A(A), o can also be written as (5.28 a(A) : :w— —— 5.28) (A) i'w — ay which is exactly how (5.12) appears because of the form the principal anti-automorphism takes when 2, is realized as IR?*? (see (2.30)). Furthermore, each such 0(A) is orthogonal on F © V since (5.23)(iii) ensures that w= (w*)! = w’ and hence that A(o(A)w) = (0(A)w)'o(A)w = (A'w! A7) (Aw(A')“?) = Alu'w(A') = A’A(w)(4’) | = A(w) (we have used the fact that A(w) € IF, ie., (5.23)(i), in the last step). Consequently, ¢ : A > o(A) actually is a homomorphism from the group of invertible transformers into O(IF @ V, A). This homomorphism leads directly to the most important result of this section. AwA*, welFovV, (5.29) Theorem. The Clifford group T(V,Q) coincides with the group of invertible trans- formers in %(V,Q); in addition, o : A+ o(A) defines a covering ho- momorphism o:T(V,Q) — SO(IF@ V, A) whose kernel is {\ € IF: \ #0}. Proof. The principal step in the proof is to relate each w € F@V, A(w) # 0, as a transformer o(w) on F@ V with the reflection Sy, of IF®V determined by w. Now the bilinear form on F @ V associated to A is given by (5.30) Biw,2) = }(wz+20), w,zE FOV.Transformers, Clifford groups 45 Consequently, 2B(w, z) Foy jw, zeRav, whenever A(w) # 0. But, by (5.23)(iii), daw , ww’ =w'w=Aw), Syiz—sz- (et and so wz = —2'w' + 2B(w', 2) = —2'w! + 2B(w, 2’) = (+ Toa ; Dy)! = 84(—2!yu! i.e., o(w)z = Sy(—2"), whenever A(w) # 0. Hence the Clifford group is a subgroup of the group of all invertible transformers on F @V. We shall prove that these two groups coincide by showing that they have the same range SO(IF © V, A) with respect to the homomorphism o, since in both cases the kernel of o is just {\ € F : \ £ 0} (see (5.27)). First we must recognize the linear transformation z + —2', z € W. But, by (5.30), Biz,l)=8E+2)=2(2' +2), 2eW; consequently, ~z! =2-2B(z, 11 = $1(z), zew. Hence, as the composition 5, 0.5; of two reflections, ¢(w) is in SO(F © V,A), not just in O(F@V,A). Clearly o now extends to a homomor- phism from ['(V,Q) into SO(IF @ V, A) such that o(w') = o(w)~! = (Sw 0 1)? = $1 0 Sw since w'w (= A(w)) is in the kernel of 9; in particular, o(w2’) = SyoSz. On [(V, Q), therefore, o is a homomorphism having range all of SO(F® V,A) and kernel {\ € F : A 4 0}, since SO(FOV, A) consists of all even finite products of reflections. In particular, every element of SO(F ® V, A) arises as o( A) for some invertible transformer A. Consequently, if there exists an invertible transformer B such that (B) is in OUF@V, A) but not in SO(IF@ V, A), then ¢(B) = $;00(A) for some such A. Hence we can assume from the outset that S; = 0(B), ie., Si(z) =—-2' = B2(B')"'*" zew. For z in V, this reduces to Bz = zB’; and so B is in the kernel of o and o(B) =1 ¢ Sj, contradicting the original choice of B. Thus o also maps the group of all invertible transformers onto SO(F @ V, A), completing the proof. a46 Clifford algebras 6 Spin groups One of the most fundamental aspects of the theory of Clifford algebras is the construction of various ‘spin subgroups’ of the Clifford group of W. By restricting the homomorphism ¢ to such a subgroup, we thus obtain a covering of a subgroup of SO(IF@V, A), and the idea is to achieve this as economically as possible by ensuring that the kernel of the mapping is {+1} rather than {\ € F: \ £0}. Covering homomorphisms with this property will be said to be two-fold coverings. These constructions are made possible because all non-zero multiples Av of any vin V, Q(v) 4 0, determine the same reflection 2B(a,v) Qe) of V as v does. Consequently, every A in O(V,Q) is the finite product of reflections S, with v in Up(V) where Syy i 2 a (6.1)(i) Er(V) = {ve V: Qu) =+1} when IF = R, and (6.1)(ii) Le(V) = {v eV: Qv) =1} when IF = C; every A in SO(V, Q) is an even product of such reflections. Similarly, every A in SO(IF@V, A) is an even finite product of reflections Sw with w in the corresponding Dp(IF @ V). Throughout this section (V, Q) will be any real or complex, non-degenerate quadratic space unless specifically restricted. The prototypical spin group is the set (6.2) Spin(V,Q) = {v1 +--v2% vj € De(V)} of all even finite products in YW of elements from Ug(V). Since the inverse —v/Q(v) of each v in EF(V) is again in De(V), Spin(V,Q) is a subgroup of P(V,@); it is contained also in the even subalgebra Wy. of U. Thus on V each o(g), g € Spin(V, Q), actually is a genuine similarity transformation o(g) : v > gvg~! because the principal automorphism reduces to the identity on %,. (6.3) Theorem. The restriction of to Spin(V,Q) defines a two-fold covering homo- morphism (6.4) a: Spin(V,Q) —> SO(V,Q) , o(g):u—+gug™. Proof. As in the proof of the corresponding result for the Clifford group,Spin groups 47 the basic idea is to relate the transformer o(v) on V determined by each v in Ug(V) with the reflection S, of V that v determines. Now B(a,v) =-}(xv + vz) , zeV, is the bilinear form on V associated with Q. Thus 2B(a, vie) ; Qe) Consequently, o(v) = —S,, and so (6.4) is a covering homomorphism. Its kernel consists of all A in F which can be written as \ = v1 ---v2% with vj € Dp(V). But then d= AA) = A(ur ++ va) = [] Q(vs) = 1 j va = -av - 2B(a,v) =— (« whether IF = R or €. Hence o has kernel {1} in Spin(V,Q). ' Had we taken all finite products of elements from Up(V) instead of even finite products, the range of o would have been O(V,Q), not SO(V,Q). Thus in a jocular vein we are led to the group (6.5) Pin(V,Q) = {vr ---vn 0) € DE(V)} of all finite products in 2 of elements from Up(V). The same proof as for (6.3) gives the following. (6.6) Theorem. The restriction of o to Pin(V,Q) defines a two-fold covering homo- morphism o:Pin(V,Q) + OV,Q) , a(g):u—> gry’)? - Replacing De(V) by Ee (IF @ V) we obtain (6.7) Spoin(V,Q) = {wi-++- we: wj € De(F@V)} ; it too is a subgroup of [(V,Q). (6.8) Theorem. When F =R Spoin(V,Q) = {9 €T(V,Q): A(g) = +1} , whereas when F = C (6.9) Spoin(V,Q) = {9 ET(V,Q) : AG) = 1} - Proof. Suppose F = R. Then, since A : T(V,Q) > R. is a multiplica- tive real character, (6.10) {9 €T(V,Q) : A(g) = £1}48 Clifford algebras clearly is a subgroup of P(V,Q) containing Spoin(V,@). On the other hand, by (5.29), every g in (V,Q) can be written as (6.11) gare = |S? TT oar = Ala aw; with go in Spoin(V,Q). The subgroup (6.10) thus coincides with Spoin(V, Q). The proof for the case IF = C is the same, replacing (6.10) by the right-hand group in (6.9). . (6.12) Theorem. The restriction of o to Spoin(V,Q) defines a two-fold covering homo- Fabetaot a: Spoin(V,Q) —+ SO(F@®V,A) , o(g):w— gu(g')?. Proof. The proof of theorem (5.29) shows that, whenever w € De(IFOV), o(w) is equal to S, 0S, = Sy oS. Using the fact that S? = J, it is easy to see that o maps Spoin(V,Q) onto SO(IF @ V, A). The kernel of ¢ consists of all ) € IF which may be written as \ = wiw2-++ we, with w, € Of(IF @ V). A simple application of theorem (6.8) shows that the kernel of ¢ in Spoin(V,Q) is just {+1}. a The motivation for introducing Pin(V,Q) is clear from (6.6). What is not so clear perhaps is the need for the group Spoin(V, Q), since it is isomorphic to Spin(IF @ V, A) under the algebra isomorphism (6.13) WV, Q) =U (F@V,A) . In fact, in any theoretical discussion Spin(V,Q) is more convenient to study than Spoin(V,Q) because of the awkward dual nature of the role that IF is being forced to play in the latter case. But in practice, Spoin(-) realizes a particular spin group such as Spin(p,q) inside the Clifford algebra Wp,¢—1, whereas Spin(-) realizes it inside the larger algebra Wp,. On the grounds of economy of size, therefore, Spoin(-) is often preferable to Spin(-) in specific constructions, as we shall see. The connectedness of SO(V, Q) depends both on the scalar field IF and on Q. In turn this affects its two-fold covering Spin(V,Q), forcing the introduction of yet another Spin group, Sping(V, Q), in the disconnected case. When F = C, for instance, SO(V,Q) has a matrix realization as SO(n, C), n = dime V, with respect to any choice of normalized basis, and so will always be connected. By contrast, when FF = R, SO(V,Q) has a matrix realization as SO(p,q), p-+q = dim V. Such a group is connected if and only if p or q is zero, i.e., if and only if Q is negative.The Euclidean case 49 or positive-definite; otherwise, it has two components. When IF = R and Q is not definite, we shall denote by $0o(V,@) (resp. Spino(p, q)) the connected component of SO(V,Q) (resp. Spin(p, q)) containing the identity. (6.14) Definition. Let (V,Q) be an indefinite real quadratic space. Then the subgroup {9 € Spin(V, Q) : A(g) = 1} of Spin(V,Q) will be denoted by Sping(V, Q). Since A : Spin(V,Q) + {+1} is always a real multiplicative charac- ter, Sping(V, Q) is just the kernel of A, hence a subgroup of Spin(V, Q) whether IF = R or € and @ is definite or indefinite. But it is only when F = R and Q is indefinite, i.c., Q does not have constant sign on V, that Sping(V, Q) is a proper subgroup. When dim V > 2, we obtain the following result, extending (6.3). (6.15) Theorem. When (V,Q) is a real indefinite quadratic space, the restriction of o to Sping(V,Q) defines a two-fold covering homomorphism @: Sping(V,Q) — SOo(V,Q) , (9): —> grg7! As the proof of this theorem would take us too far afield, we omit it. The curious reader is referred to the excellent monograph of Porteous ([88). 7 The Euclidean case Because so much of the analysis to be developed on Euclidean space and on more general Riemannian manifolds depends ultimately on the theory of Clifford algebras associated with positive-definite quadratic spaces, it will be very convenient to formulate separately various aspects of the theory for this special class of spaces. Thus, for each n = 0,1,..., let %, (= Won) be the universal Clifford algebra associated with R”, adopting the convention Mo = IR. Formally, each %,, is a 2"-dimensional real associative algebra with identity containing linearly a copy of IR” so that for any orthonormal basis {€1,..., én} of IR”50 Clifford algebras (7.1)(i) jk + ene; = — Wye (LS I,R 0. ‘Three operators on Wl, are conveniently defined in terms of the a‘*); for each u in U1"), & > 0, set 1 1 (77) w= (Iku, ut =(-1)2*®F My, a= (-1) 2k Dy | and then extend linearly to all of {,. These are the principal auto- morphism, principal anti-automorphism and conjugation respectively, defined earlier, in section 2. Note that u + u* and u —> @ are involu- tions on Wy in the sense of being linear mappings such that (uw) =bu , (a) =u, and similarly for u— u*. As an element of 2%, each e; in an orthonormal basis for IR” will be said to be an imaginary unit because e = —1 in Y,,; furthermore, just. as C and H can be written as C=ReER , H=CoHjc,The Euclidean case 51 with respect to imaginary units i, 7, so %p41 can be written as (7.8) Ang = Un Beng An where e+, is the imaginary unit in Q,,4) which extends an orthonormal basis for IR” C ,, to one for R"*? C M41. The usual basis for IR” will be specified by calling it the standard basis. The next result exhibits an explicit and very useful linear isomorphism between the exterior algebra (7.9) A*(R") = A°(R") & A(R") ®---@ AT (IR") for R” and the Clifford algebra %, associated with R”. (7.10) Theorem. The linear extension of the mapping Aiea, No Alan — > Car"*Ca, — (k 2 0) to all of A*(IR") defines a linear isomorphism from A*(IR") onto Un, independently of the choice of orthonormal basis {e;} for R”. One consequence of theorem (7.10) is that all the usual Grassmannian geometry associated with IR” can be found naturally within @,. There are many other important consequences as we shall see. Proof of theorem (7.10). Since the sets of reduced products €q, A---A€a, and €g,-++€a, (1 < a) < ++: < ay ajranser Aes (F #K), res analogous to (7.6). Hence \ : A?(R") + U@) is well-defined indepen- dently of orthonormal basis. A corresponding proof is valid for every k > 0, completing the proof. a. Now the inner product u-v on JR” extends naturally to an inner product on A*(IR"). On A‘(IR”) it is defined explicitly in a basis-free way by (7.12) (Er As Abe m A+++ Ang) = det(é, ©) with an extension to A*(IR”) so that A‘(IR") is orthogonal to A™(IR”) when €#m. The basis of reduced products €a, A-+- A €a, for A*(IR") determined by an orthonormal basis {e;} for IR” clearly is orthonormal