math0105155/oct.tex
1: 
2:         %prepared in LaTeX   
3:    
4:         \input{epsf}   
5:    
6: 	\hfuzz = 9pt   
7:    
8:         \newcommand\Z{{\mathbb Z}}   
9: 	\newcommand\R{{\mathbb R}}   
10: 	\newcommand\C{{\mathbb C}}   
11:         \renewcommand\H{{\mathbb H}}   
12:         \newcommand\K{{\mathbb K}} 
13: 	\renewcommand\O{{\mathbb O}}   
14: 	\renewcommand\S{{\mathbb S}}      
15:            
16:         \newcommand\RP{{\mathbb {RP}}}   
17:         \newcommand\CP{{\mathbb {CP}}}   
18:         \newcommand\HP{{\mathbb {HP}}}   
19:         \newcommand\KP{{\mathbb {KP}}}   
20:         \newcommand\OP{{\mathbb {OP}}}   
21:         \renewcommand\P{{\mathbb P}}   
22:    
23:         \newcommand{\Cliff}{{\rm Cliff}}   
24:         \newcommand{\J}{{\rm J}} 
25:         \newcommand{\M}{{\rm M}}
26:   
27:         \newcommand{\OO}{{\rm O}}   
28: 	\newcommand{\SO}{{\rm SO}}   
29: 	\newcommand{\SL}{{\rm SL}}   
30:         \newcommand{\PSL}{{\rm PSL}}   
31: 	\newcommand{\SU}{{\rm SU}}   
32:         \newcommand{\Sp}{{\rm Sp}}
33:         \newcommand{\Spin}{{\rm Spin}}   
34:         \newcommand{\Pin}{{\rm Pin}}   
35: 	\newcommand{\U}{{\rm U}}   
36: 	\newcommand{\E}{{\rm E}}   
37: 	\newcommand{\F}{{\rm F}}   
38: 	\newcommand{\G}{{\rm G}}   
39:    
40: 	\newcommand{\so}{{\mathfrak {so}}}   
41: 	\newcommand{\Sl}{{\mathfrak {sl}}}   
42: 	\newcommand{\symp}{{\mathfrak {sp}}}   
43: 	\renewcommand{\u}{{\mathfrak {u}}}   
44: 	\newcommand{\e}{{\mathfrak {e}}}   
45: 	\newcommand{\f}{{\mathfrak {f}}}   
46: 	\newcommand{\g}{{\mathfrak {g}}}   
47: 	\newcommand{\su}{{\mathfrak {su}}}   
48:         \renewcommand{\a}{{\mathfrak {a}}}   
49:         \newcommand{\sa}{{\mathfrak {sa}}}   
50:         \newcommand{\h}{{\mathfrak {h}}}   
51:         \newcommand{\sh}{{\mathfrak {sh}}}   
52:         \newcommand{\coll}{{\mathfrak {coll}}}   
53:         \newcommand{\isom}{{\mathfrak {isom}}}   
54:         \newcommand{\Isom}{{\rm Isom}} 
55:         \newcommand{\Der}{{\mathfrak {der}}}   
56:         \newcommand{\Tri}{{\mathfrak {tri}}}  
57: 
58:    
59: 	\newcommand{\et}{\hspace{-0.08in}{\bf .}\hspace{0.1in}}   
60:         \newcommand{\DOT}{\hspace{-0.08in}{\bf .}\hspace{0.1in}}   
61:  	\newcommand{\BOX}{\hbox {$\sqcap$ \kern -1em $\sqcup$}}   
62: 	\newcommand{\qed}{\hskip 3em \hbox{\BOX} \vskip 2ex}   
63:    
64:         \renewcommand{\Re}{{\rm Re}}   
65:         \renewcommand{\Im}{{\rm Im}}   
66: 	\newcommand{\Hom}{{\rm hom}}   
67:         \newcommand{\Aut}{{\rm Aut}}   
68: 	\newcommand{\Inv}{{\rm Inv}}   
69: 	\newcommand{\tensor}{\otimes}   
70:         \newcommand{\iso}{\cong}   
71:         \newcommand{\implies}{\Longrightarrow}   
72:         \newcommand{\impliedby}{\Longleftarrow}   
73:    
74: 	\newcommand{\be}{\begin{equation}}   
75:         \newcommand{\ee}{\end{equation}}   
76:         \newcommand{\ba}{\begin{eqnarray}}   
77:         \newcommand{\ea}{\end{eqnarray}}   
78:         \newcommand{\ban}{\begin{eqnarray*}}   
79:         \newcommand{\ean}{\end{eqnarray*}}   
80:   
81:         \newcommand{\ad}{{\rm ad}}   
82: 	\newcommand{\Cob}{{\rm Cob}}   
83: 	\newcommand{\Vect}{{\rm Vect}}   
84: 	\newcommand{\Hilb}{{\rm Hilb}}   
85:         \newcommand{\maps}{\colon}   
86: 	\newcommand{\tr}{{\rm tr}}   
87: 	\newcommand{\rank}{{\rm rank}}   
88: 	\newcommand{\id}{{\rm id}}   
89: 	\newcommand{\Fun}{{\rm Fun}}   
90: 	\newcommand{\Ad}{{\rm Ad}}   
91: 	\newcommand{\om}{\omega}   
92: 	\newcommand{\eps}{\epsilon}  
93:  
94: 	\newtheorem{thm}{Theorem}   
95: 	\newtheorem{prop}{Proposition}   
96: 	\newtheorem{lemma}{Lemma}   
97: 	\newtheorem{corollary}{Corollary}   
98: 	\newtheorem{ex}{Example}   
99:  
100: \documentclass {article} 
101: \usepackage {amsfonts}
102: 
103: 	\textwidth 6in   
104: 	\textheight 8.5in	\evensidemargin .25in   
105: 	\oddsidemargin .25in   
106: 	\topmargin .25in   
107: 	\headsep 0in   
108: 	\headheight 0in   
109: 	\footskip .5in   
110: %If you want single spaced copy, delete the next two lines.   
111: %	\parskip 1.75\parskip plus 3pt minus 1pt   
112: %	\renewcommand{\baselinestretch}{1.5}   
113: 	\pagestyle{plain}   
114: 	\pagenumbering{arabic}   
115: 	\begin{document}   
116:  
117: 	\begin{center}   
118: 	{\bf  The Octonions \\}   
119: 	{\em John\ C.\ Baez\\}   
120: 	\vspace{0.3cm}   
121: 	{\small Department of Mathematics \\   
122: 	University of California\\   
123:         Riverside CA 92521\\}   
124:         {\small email:  baez@math.ucr.edu\\} 
125: 	\vspace{0.3cm}   
126: 	{May 16, 2001 \\}   
127: 	\vspace{0.3cm}   
128: 	\end{center}   
129: 
130: \begin{abstract} 
131: \noindent  
132: The octonions are the largest of the four normed division algebras.
133: While somewhat neglected due to their nonassociativity, they stand at
134: the crossroads of many interesting fields of mathematics.  Here we
135: describe them and their relation to Clifford algebras and spinors, Bott
136: periodicity, projective and Lorentzian geometry, Jordan algebras, and
137: the exceptional Lie groups.  We also touch upon their applications in
138: quantum logic, special relativity and supersymmetry. 
139: \end{abstract}
140: 
141: \section{Introduction}   
142:    
143: There are exactly four normed division algebras: the real numbers  
144: ($\R$), complex numbers ($\C$), quaternions ($\H$), and octonions  
145: ($\O$).  The real numbers are the dependable breadwinner of the family,  
146: the complete ordered field we all rely on.  The complex numbers are a  
147: slightly flashier but still respectable younger brother: not ordered,  
148: but algebraically complete.  The quaternions, being noncommutative, are  
149: the eccentric cousin who is shunned at important family gatherings.  But  
150: the octonions are the crazy old uncle nobody lets out of the attic: they  
151: are {\it nonassociative}.  
152:    
153: Most mathematicians have heard the story of how Hamilton invented the   
154: quaternions.  In 1835, at the age of 30, he had discovered how to treat   
155: complex numbers as pairs of real numbers.   Fascinated by the relation   
156: between $\C$ and 2-dimensional geometry, he tried for many years to   
157: invent a bigger algebra that would play a similar role in 3-dimensional   
158: geometry.  In modern language, it seems he was looking for a 3-dimensional   
159: normed division algebra.  His quest built to its climax in October 1843.   
160: He later wrote to his son, ``Every morning in the early part of the   
161: above-cited month, on my coming down to breakfast, your (then) little   
162: brother William Edwin, and yourself, used to ask me: `Well, Papa, can you   
163: {\it multiply} triplets?'  Whereto I was always obliged to reply, with a sad   
164: shake of the head: `No, I can only {\it add} and subtract them'.''     
165: The problem, of course, was that there exists no 3-dimensional normed    
166: division algebra.  He really needed a 4-dimensional algebra.   
167:    
168: Finally, on the 16th of October, 1843, while walking with his wife along   
169: the Royal Canal to a meeting of the Royal Irish Academy in Dublin, he made    
170: his momentous discovery.  ``That is to say, I then and there felt the    
171: galvanic circuit of thought {\it close}; and the sparks which fell from it    
172: were the {\it fundamental equations between $i,j,k$; exactly such} as I have    
173: used them ever since.''  And in a famous act of mathematical vandalism, he   
174: carved these equations into the stone of the Brougham Bridge:    
175: \[    i^2 = j^2 = k^2 = ijk = -1 .\]    
176:    
177: One reason this story is so well-known is that Hamilton spent the rest   
178: of his life obsessed with the quaternions and their applications to   
179: geometry \cite{Graves,Hankins}.  And for a while, quaternions  were   
180: fashionable.  They were made a mandatory examination topic in Dublin,   
181: and in some American universities they were the only advanced   
182: mathematics taught.  Much of what we now do with scalars and vectors in   
183: $\R^3$ was  then done using real and imaginary quaternions.   A school   
184: of `quaternionists' developed, which was led after Hamilton's death by   
185: Peter Tait of Edinburgh and Benjamin Peirce of Harvard.  Tait wrote 8   
186: books on the quaternions, emphasizing their applications to physics.    
187: When Gibbs invented the modern notation for the dot product and cross   
188: product, Tait condemned it as a ``hermaphrodite monstrosity''.  A war of   
189: polemics ensued, with such luminaries as Heaviside weighing   
190: in on the side of vectors.  Ultimately the quaternions lost, and   
191: acquired a slight taint of disgrace from which they have never fully   
192: recovered \cite{Crowe}.    
193:    
194: Less well-known is the discovery of the octonions by Hamilton's friend   
195: from college, John T.\ Graves.  It was Graves' interest in algebra that   
196: got Hamilton thinking about complex numbers and triplets in the first    
197: place.  The very day after his fateful walk, Hamilton sent an 8-page   
198: letter describing the quaternions to Graves.  Graves replied on October   
199: 26th, complimenting Hamilton on the boldness of the idea, but adding   
200: ``There is still something in the system which gravels me.  I have not   
201: yet any clear views as to the extent to which we are at liberty   
202: arbitrarily to create imaginaries, and to endow them with supernatural   
203: properties.''  And he asked: ``If with your alchemy you can make three   
204: pounds of gold, why should you stop there?''  
205: 
206: Graves then set to work on some gold of his own!  On December 26th,
207: he wrote to Hamilton describing a new 8-dimensional algebra, which he
208: called the `octaves'.   He showed that they were a normed division
209: algebra, and used this to express the product of two sums of eight
210: perfect squares as another sum of eight perfect squares: the `eight
211: squares theorem' \cite{Hamilton}. 
212: 
213: In January 1844, Graves sent three letters to Hamilton expanding on his   
214: discovery.  He considered the idea of a general theory of   
215: `$2^m$-ions', and tried to construct a 16-dimensional normed division   
216: algebra, but he ``met with an unexpected hitch'' and came to doubt that   
217: this was possible.  Hamilton offered to publicize Graves' discovery, but   
218: being busy with work on quaternions, he kept putting it off.  In July he   
219: wrote to Graves pointing out that the octonions were nonassociative:   
220: ``$A \cdot BC = AB \cdot C = ABC$, if $A,B,C$ be quaternions, but not   
221: so, generally, with your octaves.''  In fact, Hamilton first invented   
222: the term `associative' at about this time, so the octonions may have   
223: played a role in clarifying the importance of this concept.    
224:    
225: Meanwhile the young Arthur Cayley, fresh out of Cambridge, had been
226: thinking about the quaternions ever since Hamilton announced their
227: existence.  He seemed to be seeking relationships between the
228: quaternions and hyperelliptic functions.  In March of 1845, he published
229: a paper in the {\it Philosophical Magazine} entitled `On Jacobi's
230: Elliptic Functions, in Reply to the Rev.\ B.\ Bronwin; and on Quaternions'
231: \cite{Cayley}.  The bulk of this paper was an attempt to rebut an
232: article pointing out mistakes in Cayley's work on elliptic functions.
233: Apparently as an afterthought, he tacked on a brief description of the
234: octonions.  In fact, this paper was so full of errors that it was 
235: omitted from his collected works --- except for the part about octonions
236: \cite{Cayley2}.
237: 
238: Upset at being beaten to publication, Graves attached a postscript to a
239: paper of his own which was to appear in the following issue of the same
240: journal, saying that he had known of the octonions ever since Christmas,
241: 1843.  On June 14th, 1847, Hamilton contributed a short note to the
242: Transactions of the Royal Irish Academy, vouching for Graves' priority. 
243: But it was too late: the octonions became known as `Cayley numbers'. 
244: Still worse, Graves later found that his eight squares theorem had
245: already been discovered by C.\ F.\ Degen in 1818 \cite{Curtis,Dickson}.
246: 
247: Why have the octonions languished in such obscurity compared to the
248: quaternions?  Besides their rather inglorious birth, one reason is that
249: they lacked a tireless defender such as Hamilton.  But surely the reason
250: for {\it this} is that they lacked any clear application to geometry and
251: physics.  The unit quaternions form the group $\SU(2)$, which is the
252: double cover of the rotation group $\SO(3)$.  This makes them nicely
253: suited to the study of rotations and angular momentum, particularly in
254: the context of quantum mechanics.  These days we regard this phenomenon
255: as a special case of the theory of Clifford algebras.  Most of us no
256: longer attribute to the quaternions the cosmic significance that
257: Hamilton claimed for them, but they fit nicely into our understanding of
258: the scheme of things.
259:  
260: The octonions, on the other hand, do not.  Their relevance to geometry
261: was quite obscure until 1925, when \'Elie Cartan described `triality'
262: --- the symmetry between vectors and spinors in 8-dimensional Euclidean
263: space \cite{Cartan3}.  Their potential relevance to physics was noticed
264: in a 1934 paper by Jordan, von Neumann and Wigner on the foundations of
265: quantum mechanics \cite{JNW}.  However, attempts by Jordan and others to
266: apply octonionic quantum mechanics to nuclear and particle physics met
267: with little success.  Work along these lines continued quite slowly
268: until the 1980s, when it was realized that the octonions explain some
269: curious features of string theory \cite{KT}.  The Lagrangian for
270: the classical superstring involves a relationship between vectors and
271: spinors in Minkowski spacetime which holds only in 3, 4, 6, and 10
272: dimensions.  Note that these numbers are 2 more than the dimensions of
273: $\R,\C,\H$ and $\O$.  As we shall see, this is no coincidence: briefly, 
274: the isomorphisms 
275: \[ \begin{array}{lcl}
276:           \Sl(2,\R) &\iso& \so(2,1)   \\
277:           \Sl(2,\C) &\iso& \so(3,1)   \\   
278:           \Sl(2,\H) &\iso& \so(5,1)   \\
279:           \Sl(2,\O) &\iso& \so(9,1)   
280: \end{array}
281: \]
282: allow us to treat a spinor in one of these dimensions as a pair of   
283: elements of the corresponding division algebra.  It is fascinating   
284: that of these superstring Lagrangians, it is the 10-dimensional
285: octonionic one that gives the most promising candidate for a realistic
286: theory of fundamental physics!  However, there is still no {\it proof}
287: that the octonions are useful for understanding the real world.  We
288: can only hope that eventually this question will be settled one way or
289: another.
290: 
291: Besides their possible role in physics, the octonions are important   
292: because they tie together some algebraic structures that otherwise   
293: appear as isolated and inexplicable exceptions.  As we shall explain,   
294: the concept of an octonionic projective space $\OP^n$ only makes sense   
295: for $n \le 2$, due to the nonassociativity of $\O$.  This means that   
296: various structures associated to real, complex and quaternionic   
297: projective spaces have octonionic analogues only for $n \le 2$.    
298:    
299: Simple Lie algebras are a nice example of this phenomenon.   There are  
300: 3 infinite families of `classical' simple Lie algebras, which come from
301: the isometry groups of the projective spaces $\RP^n$, $\CP^n$ and   
302: $\HP^n$.  There are also 5 `exceptional' simple Lie algebras.   These 
303: were discovered by Killing and Cartan in the late 1800s.  At the time,
304: the significance of these exceptions was shrouded in mystery: they did
305: not arise as symmetry groups of known structures.  Only later did their
306: connection to the octonions become clear.  It turns out that 4 of them
307: come from the isometry groups of the projective planes over $\O$, $\O
308: \tensor \C$, $\O \tensor \H$ and $\O  \tensor \O$.  The remaining one is
309: the automorphism group of the octonions!   
310:    
311: Another good example is the classification of simple formally real
312: Jordan algebras.  Besides several infinite families of these, there
313: is the `exceptional' Jordan algebra, which consists of $3 \times 3$
314: hermitian octonionic matrices.   Minimal projections in this Jordan
315: algebra correspond to points of $\OP^2$, and the automorphism group of
316: this algebra is the same as the isometry group of $\OP^2$.   
317:    
318: The octonions also have fascinating connections to topology.  In 1957,  
319: Raoul Bott computed the homotopy groups of the topological group
320: $\OO(\infty)$, which is the inductive limit of the orthogonal groups
321: $\OO(n)$ as $n \to \infty$.  He proved that they repeat with period
322: 8:   
323: \[   \pi_{i+8}(\OO(\infty)) \iso \pi_i(\OO(\infty)).   \]   
324: This is known as `Bott periodicity'.  He also computed the first 8:   
325: \ban       
326:                \pi_0(\OO(\infty)) &\iso& \Z_2  \\    
327:                \pi_1(\OO(\infty)) &\iso& \Z_2  \\   
328:                \pi_2(\OO(\infty)) &\iso&  0    \\   
329:                \pi_3(\OO(\infty)) &\iso& \Z    \\   
330:                \pi_4(\OO(\infty)) &\iso&  0    \\   
331:                \pi_5(\OO(\infty)) &\iso&  0    \\   
332:                \pi_6(\OO(\infty)) &\iso&  0    \\   
333:                \pi_7(\OO(\infty)) &\iso& \Z       
334: \ean   
335: Note that the nonvanishing homotopy groups here occur in dimensions one   
336: less than the dimensions of $\R,\C,\H$, and $\O$.  This is no coincidence!   
337: In a normed division algebra, left multiplication by an element of norm   
338: one defines an orthogonal transformation of the algebra, and thus an   
339: element of $\OO(\infty)$.   This gives us maps from the spheres $S^0,    
340: S^1, S^3$ and $S^7$ to $\OO(\infty)$, and these maps generate the    
341: homotopy groups in those dimensions.     
342: 
343: Given this, one might naturally guess that the period-8 repetition in
344: the homotopy groups of $\OO(\infty)$ is in some sense `caused' by the
345: octonions.  As we shall see, this is true.  Conversely, Bott
346: periodicity is closely connected to the problem of how many pointwise
347: linearly independent smooth vector fields can be found on the
348: $n$-sphere \cite{Husemoller}.  There exist $n$ such vector fields only
349: when $n+1 = 1, 2, 4,$ or $8$, and this can be used to show that 
350: division algebras over the reals can only occur in these dimensions.
351: 
352: In what follows we shall try to explain the octonions and their role in 
353: algebra, geometry, and topology.  In Section \ref{constructing} we give 
354: four constructions of the octonions: first via their multiplication   
355: table, then using the Fano plane, then using the Cayley--Dickson   
356: construction and finally using Clifford algebras, spinors, and a  
357: generalized concept of `triality' advocated by Frank Adams \cite{Adams}.
358: Each approach has its own merits.  In Section \ref{proj} we discuss
359: the projective lines and planes over the normed division algebras --- 
360: especially $\O$ --- and describe their relation to Bott periodicity,  
361: the exceptional Jordan algebra, and the Lie algebra isomorphisms listed 
362: above.  Finally, in Section \ref{lie} we discuss octonionic  
363: constructions of the exceptional Lie groups, especially the `magic 
364: square'.   
365:    
366: \subsection{Preliminaries} \label{preliminaries}
367: 
368: Before our tour begins, let us settle on some definitions.  For us a
369: {\bf vector space} will always be a finite-dimensional module over the
370: field of real numbers.  An {\bf algebra} $A$ will be a vector space that
371: is equipped with a bilinear map $m \maps A \times A \to A$ called
372: `multiplication' and a nonzero element $1 \in A$ called the `unit' such
373: that $m(1,a) = m(a,1) = a$.  As usual, we abbreviate $m(a,b)$ as $ab$.
374: We do not assume our algebras are associative!  Given an algebra, we
375: will freely think of real numbers as elements of this algebra via the
376: map $\alpha \mapsto \alpha 1$.
377:    
378: An algebra $A$ is a {\bf division algebra} if given $a,b \in A$ with $ab
379: = 0$, then either $a = 0$ or $b = 0$.  Equivalently, $A$ is a division
380: algebra if the operations of left and right multiplication by any
381: nonzero element are invertible.  A {\bf normed division algebra} is an
382: algebra $A$ that is also a normed vector space with $\|ab\| = \|a\|
383: \|b\|$.  This implies that $A$ is a division algebra and that $\|1\| =
384: 1$.
385: 
386: We should warn the reader of some subtleties.  We say an algebra $A$ has
387: {\bf multiplicative inverses} if for any nonzero $a \in A$ there is an
388: element $a^{-1} \in A$ with $aa^{-1} = a^{-1}a = 1$.  An associative
389: algebra has multiplicative inverses iff it is a division
390: algebra.  However, this fails for nonassociative algebras!  In Section
391: \ref{cayley-dickson} we shall construct algebras that have
392: multiplicative inverses, but are not division algebras.  On the other
393: hand, we can construct a division algebra without multiplicative
394: inverses by taking the quaternions and modifying the product slightly,
395: setting $i^2 = -1 + \epsilon j$ for some small nonzero real number
396: $\epsilon$ while leaving the rest of the multiplication table unchanged.
397: The element $i$ then has both right and left inverses, but they are not
398: equal.  (We thank David Rusin for this example.)
399:    
400: There are three levels of associativity.  An algebra is {\bf   
401: power-associative} if the subalgebra generated by any one element is   
402: associative.  It is {\bf alternative} if the subalgebra generated by any   
403: two elements is associative.  Finally, if the subalgebra generated by any    
404: three elements is associative, the algebra is associative.     
405:    
406: As we shall see, the octonions are not associative, but they are alternative.   
407: How does one check a thing like this?  By a theorem of Emil Artin    
408: \cite{Schafer}, an algebra $A$ is alternative iff for all $a,b \in A$ we have   
409: \ba  (aa)b = a(ab), \qquad (ab)a = a(ba), \qquad (ba)a = b(aa)    
410: \label{alternative}   \ea   
411: In fact, any two of these equations implies the remaining one, so people   
412: usually take the first and last as the definition of `alternative'.   
413: To see this fact, note that any algebra has a trilinear map    
414: \[  [\cdot,\cdot,\cdot] \maps A^3 \to A  \]   
415: called the {\bf associator}, given by   
416: \[               [a,b,c] = (ab)c - a(bc)   .\]   
417: The associator measures the failure of associativity just as the 
418: commutator $[a,b] = ab - ba$ measures the failure of commutativity.   
419: Now, the commutator is an alternating bilinear map, meaning that it 
420: switches sign whenever the two arguments are exchanged:   
421: \[         [a,b] = -[b,a]   \]   
422: or equivalently, that it vanishes when they are equal:   
423: \[         [a,a] = 0 .\]   
424: This raises the question of whether the associator is alternating too.     
425: In fact, this holds precisely when $A$ is alternative!  The reason is   
426: that each equation in (\ref{alternative}) says that the associator   
427: vanishes when a certain pair of arguments are equal, or equivalently,   
428: that it changes sign when that pair of arguments is switched.  Note,   
429: however, that if the associator changes sign when we switch the $i$th   
430: and $j$th arguments, and also when we switch the $j$th and $k$th   
431: arguments, it must change sign when we switch the $i$th and $k$th.      
432: Thus any two of equations (\ref{alternative}) imply the third.     
433:    
434: Now we can say what is so great about $\R,\C,\H,$ and $\O$:   
435:  
436: \begin{thm}  \et \label{hurwitz} 
437: $\R,\C,\H$, and $\O$ are the only normed division algebras.  
438: \end{thm}  
439:  
440: \begin{thm} \et \label{zorn}   
441: $\R,\C,\H$, and $\O$ are the only alternative division algebras.   
442: \end{thm}   
443:    
444: The first theorem goes back to an 1898 paper by Hurwitz \cite{Hurwitz}. 
445: It was subsequently generalized in many directions, for example, to  
446: algebras over other fields.   A version of the second theorem appears in
447: an 1930 paper by Zorn \cite{Zorn} --- the guy with the lemma.  For
448: modern proofs of both these theorems, see Schafer's excellent book on
449: nonassociative algebras \cite{Schafer}.  We sketch a couple proofs of
450: Hurwitz's theorem in Section \ref{clifford}.  
451: 
452: Note that we did {\it not} state that $\R,\C,\H$ and $\O$ are the only   
453: division algebras.  This is not true.  For example, we have already  
454: described a way to get 4-dimensional division algebras that do not have  
455: multiplicative inverses.  However, we do have this fact:   
456: 
457: \begin{thm} \et \label{bott-milnor}  All division algebras have dimension    
458: $1, 2, 4,$ or $8$.    
459: \end{thm}  
460: 
461: \noindent
462: This was independently proved by Kervaire \cite{Kervaire} and
463: Bott--Milnor \cite{BM} in 1958.  We will say a bit about the proof in
464: Section \ref{OP1}.  However, in what follows our main focus will not be
465: on general results about division algebras.  Instead, we concentrate on
466: special features of the octonions.  Let us begin by constructing them.
467:   
468: \section{Constructing the Octonions} \label{constructing}   
469:    
470: The most elementary way to construct the octonions is to give their   
471: multiplication table.  The octonions are an 8-dimensional algebra   
472: with basis $1, e_1,e_2,e_3,e_4,e_5,e_6,e_7$,    
473: and their multiplication is given in this table, which describes   
474: the result of multiplying the element in the $i$th row by the   
475: element in the $j$th column:   
476:    
477: \vskip 2em   
478: {\vbox{   
479: \begin{center}   
480: {\small   
481: \begin{tabular}{|c|c|c|c|c|c|c|c|c|}                    \hline   
482:       & $e_1$ & $e_2$ & $e_3$ & $e_4$  & $e_5$ & $e_6$ & $e_7$ \\ \hline   
483:   $e_1$ & $-1$  & $e_4$ & $e_7$ & $-e_2$ & $e_6$ & $-e_5$ & $-e_3$ \\ \hline   
484: $e_2$ & $-e_4$ & $-1$ & $e_5$ & $e_1$ & $-e_3$ & $e_7$ & $-e_6$     \\ \hline   
485: $e_3$ & $-e_7$ & $-e_5$ & $-1$ & $e_6$ & $e_2$ & $-e_4$ & $e_1$   \\ \hline   
486: $e_4$ & $e_2$ & $-e_1$ & $-e_6$ & $-1$ & $e_7$ & $e_3$ & $-e_5$   \\ \hline   
487: $e_5$ & $-e_6$ & $e_3$ & $-e_2$ & $-e_7$ & $-1$ & $e_1$ & $e_4$    \\ \hline   
488: $e_6$ & $e_5$ & $-e_7$ & $e_4$ & $-e_3$ & $-e_1$ & $-1$ & $e_2$     \\ \hline   
489: $e_7$ & $e_3$ & $e_6$ & $-e_1$ & $e_5$ & $-e_4$ & $-e_2$ & $-1$    \\  \hline   
490: \end{tabular}} 
491: \end{center}   
492: \vskip 1em 
493: \centerline{Table 1 --- Octonion Multiplication Table}  
494: }}   
495: \vskip 1em   
496:    
497: \noindent
498: Unfortunately, this table is almost completely unenlightening!  About the only 
499: interesting things one can easily learn from it are:  
500: \begin{itemize}   
501: \item $e_1,\dots,e_7$ are square roots of -1,    
502: \item $e_i$ and $e_j$ anticommute when $i \ne j$:    
503: \[     e_i e_j = -e_j e_i   \]   
504: \item the {\bf index cycling} identity holds:   
505: \[       e_i e_j = e_k \; \implies\; e_{i+1} e_{j+1} = e_{k+1}  \]   
506: where we think of the indices as living in $\Z_7$, and   
507: \item the {\bf index doubling} identity holds:   
508: \[       e_i e_j = e_k \; \implies \; e_{2i} e_{2j} = e_{2k} . \]   
509: \end{itemize}   
510: Together with a single nontrivial product like $e_1 e_2 = e_4$, these   
511: facts are enough to recover the whole multiplication table.  However, we   
512: really want a better way to remember the octonion product.  We should   
513: become as comfortable with multiplying octonions as we are with   
514: multiplying matrices!  And ultimately, we want a more conceptual    
515: approach to the octonions, which explains their special properties and   
516: how they fit in with other mathematical ideas.  In what follows, we give   
517: some more descriptions of octonion multiplication, starting with a nice   
518: mnemonic, and working up to some deeper, more conceptual ones.   
519:    
520: \subsection{The Fano plane}    \label{fano}   
521:    
522: The quaternions, $\H$, are a 4-dimensional algebra with basis $1,i,j,k$.   
523: To describe the product we could give a multiplication   
524: table, but it is easier to remember that:   
525: \begin{itemize}   
526: \item $1$ is the multiplicative identity,   
527: \item $i,j,$ and $k$ are square roots of -1,   
528: \item we have $ij = k$, $ji = -k$, and all identities obtained    
529: from these by cyclic permutations of $(i,j,k)$.     
530: \end{itemize}   
531: We can summarize the last rule in a picture:   
532:  
533: \centerline{\epsfysize=1.5in\epsfbox{triangle.eps}}   
534: \label{triangle}   
535:  
536: \noindent   
537: When we multiply two elements going clockwise around the circle we get   
538: the next one: for example, $ij = k$.  But when we multiply two   
539: going around counterclockwise, we get {\it minus} the next one:    
540: for example, $ji = -k$.     
541:    
542: We can use the same sort of picture to remember how to multiply   
543: octonions:   
544: \medskip
545: 
546: \centerline{\epsfysize=1.5in\epsfbox{fano.eps}}   
547: \label{Fano}   
548: \medskip
549: 
550: \noindent   
551: This is the {\bf Fano plane}, a little gadget    
552: with 7 points and 7 lines.  The `lines' are the sides of the triangle,    
553: its altitudes, and the circle containing all the midpoints of the sides.   
554: Each pair of distinct points lies on a unique line.  Each line contains    
555: three points, and each of these triples has has a cyclic ordering    
556: shown by the arrows.  If $e_i, e_j,$ and $e_k$ are cyclically ordered    
557: in this way then    
558: \[            e_i e_j = e_k,  \qquad e_j e_i = -e_k  . \]   
559: Together with these rules:   
560: \begin{itemize}   
561: \item $1$ is the multiplicative identity,   
562: \item $e_1, \dots, e_7$ are square roots of -1,   
563: \end{itemize}   
564: the Fano plane completely describes the algebra structure of the
565: octonions.   Index-doubling corresponds to rotating the picture
566: a third of a turn.
567:    
568: This is certainly a neat mnemonic, but is there anything deeper lurking   
569: behind it?  Yes!  The Fano plane is the projective plane over the 2-element
570: field $\Z_2$.  In other words, it consists of lines through the origin 
571: in the vector space $\Z_2^3$.  Since every such line contains a single 
572: nonzero element, we can also think of the Fano plane as consisting of the 
573: seven nonzero elements of $\Z_2^3$.  If we think of the origin in $\Z_2^3$
574: as corresponding to $1 \in \O$, we get the following picture of the 
575: octonions:   
576: 
577: \medskip
578: \centerline{\epsfysize=1.5in\epsfbox{cube.eps}}   
579: \label{cube}   
580: \medskip
581: 
582: \noindent    
583: Note that planes through the origin of this 3-dimensional vector space    
584: give subalgebras of $\O$ isomorphic to the quaternions, lines through   
585: the origin give subalgebras isomorphic to the complex numbers, and   
586: the origin itself gives a subalgebra isomorphic to the real numbers.      
587:    
588: What we really have here is a description of the octonions as a
589: `twisted group algebra'.  Given any group $G$, the group algebra   
590: $\R[G]$ consists of all finite formal linear combinations of elements   
591: of $G$ with real coefficients.  This is an associative algebra with   
592: the product coming from that of $G$.  We can use any function    
593: \[       \alpha \maps G^2 \to \{ \pm 1 \}     \]   
594: to `twist' this product, defining a new product   
595: \[       \star \maps \R[G] \times \R[G] \to \R[G]  \]   
596: by:   
597: \[       g \star h = \alpha(g,h) \; gh,  \]   
598: where $g,h \in G \subset \R[G]$.  One can figure out an equation    
599: involving $\alpha$ that guarantees this new product will be associative.   
600: In this case we call $\alpha$ a `2-cocycle'.   If $\alpha$ satisfies  a   
601: certain extra equation, the product $\star$ will also be commutative,   
602: and we call $\alpha$ a `stable 2-cocycle'.  For example, the group   
603: algebra  $\R[\Z_2]$ is isomorphic to a product of 2 copies of $\R$,    
604: but we can twist it by a stable 2-cocyle to obtain the complex numbers.     
605: The group algebra $\R[\Z_2^2]$ is isomorphic to a product of 4 copies    
606: of $\R$, but we can twist it by a 2-cocycle to obtain the quaternions.      
607: Similarly, the group algebra $\R[\Z_2^3]$ is a product of 8 copies of $\R$,    
608: and what we have really done in this section is describe a function   
609: $\alpha$ that allows us to twist this group algebra to obtain the   
610: octonions.  Since the octonions are nonassociative, this function is   
611: not a 2-cocycle.  However, its coboundary is a `stable 3-cocycle', which  
612: allows one to define a new associator and braiding for the category of  
613: $\Z_2^3$-graded vector spaces, making it into a symmetric monoidal  
614: category \cite{AM}.  In this symmetric monoidal category, the octonions  
615: are a commutative monoid object.  In less technical terms: this category 
616: provides a context in which the octonions {\it are} commutative and 
617: associative!  So far this idea has just begun to be exploited. 
618:    
619: \subsection{The Cayley--Dickson construction}  \label{cayley-dickson}   
620:    
621: It would be nice to have a construction of the normed division algebras   
622: $\R,\C,\H,\O$ that explained why each one fits neatly inside the next.   
623: It would be nice if this construction made it clear why $\H$ is   
624: noncommutative and $\O$ is nonassociative.  It would be even better if   
625: this construction gave an infinite sequence of algebras, doubling in   
626: dimension each time, with the normed division algebras as the first   
627: four.  In fact, there is such a construction: it's called the   
628: Cayley--Dickson construction.   
629:    
630: As Hamilton noted, the complex number $a+bi$ can be thought of as a pair
631: $(a,b)$ of real numbers.  Addition is done component-wise, and
632: multiplication goes like this:
633: \[            (a,b)(c,d) = (ac - db,ad + cb)   \]   
634: We can also define the conjugate of a complex number by   
635: \[                (a,b)^* = (a,-b).   \]   
636:    
637: Now that we have the complex numbers, we can define the   
638: quaternions in a similar way.  A quaternion can be thought of    
639: as a pair of complex numbers.  Addition is done component-wise,   
640: and multiplication goes like this:   
641: \be          (a,b)(c,d) = (ac - db^*, a^* d + cb)      \label{cd1} \ee   
642: This is just like our formula for multiplication of complex numbers, but
643: with a couple of conjugates thrown in.  If we included them in 
644: the previous formula nothing would change, since the conjugate of a
645: real number is just itself.  We can also define the conjugate of a
646: quaternion by
647: \be          (a,b)^* = (a^*,-b).        \label{cd2} \ee   
648:    
649: The game continues!  Now we can define an octonion to be a pair of
650: quaternions.  As before, we add and multiply them using formulas
651: (\ref{cd1}) and (\ref{cd2}).  This trick for getting new algebras from
652: old is called the {\bf Cayley--Dickson construction}.
653:    
654: Why do the real numbers, complex numbers, quaternions   
655: and octonions have multiplicative inverses?  I take it as   
656: obvious for the real numbers.  For the complex numbers,    
657: one can check that   
658: \[    (a,b) (a,b)^* = (a,b)^* (a,b) = k (1,0)  \]   
659: where $k$ is a real number, the square of the norm of $(a,b)$.     
660: This means that whenever $(a,b)$ is nonzero, its multiplicative    
661: inverse is $(a,b)^*/k$.  One can check that the same holds for the    
662: quaternions and octonions.   
663:    
664: But this, of course, raises the question: why isn't there an {\it
665: infinite} sequence of division algebras, each one obtained from the
666: preceding one by the Cayley--Dickson construction?  The answer is that
667: each time we apply the construction, our algebra gets a bit worse.
668: First we lose the fact that every element is its own conjugate, then we
669: lose commutativity, then we lose associativity, and finally we lose the
670: division algebra property.
671:    
672: To see this clearly, it helps to be a bit more formal.  Define a {\bf   
673: $\ast$-algebra} to be an algebra $A$ equipped with a {\bf conjugation},   
674: that is, a real-linear map $\ast \maps A \to A$ with   
675: \[            a^{**} = a, \quad \quad (ab)^* = b^* a^*  \]   
676: for all $a,b \in A$.  We say a $\ast$-algebra is {\bf real} if  $a =   
677: a^*$ for every element $a$ of the algebra.  We say the $\ast$-algebra   
678: $A$ is {\bf nicely normed} if $a + a^* \in \R$ and $aa^* = a^* a > 0$ 
679: for all nonzero $a \in A$.  If $A$ is nicely normed we set   
680: \[             \Re(a) = (a + a^\ast)/2  \in \R, \qquad   
681:                \Im(a) = (a - a^\ast)/2 ,  \]   
682: and define a norm on $A$ by    
683: \[                 \|a\|^2 = aa^\ast .      \]   
684: If $A$ is nicely normed, it has multiplicative inverses given by   
685: \[               a^{-1} = a^\ast / \|a\|^2   .\]   
686: If $A$ is nicely normed and alternative, $A$ is a normed division   
687: algebra.  To see this, note that for any $a,b \in A$, all 4 elements   
688: $a,b,a^\ast,b^\ast$ lie in the associative algebra generated by $\Im(a)$   
689: and $\Im(b)$, so that    
690: \[     \|ab\|^2 = (ab)(ab)^\ast = ab(b^\ast a^\ast) =    
691: a(bb^\ast)a^\ast = \|a\|^2 \|b\|^2 . \]   
692:    
693: Starting from any $\ast$-algebra $A$, the Cayley--Dickson construction   
694: gives a new $\ast$-algebra $A'$.   Elements of $A'$ are pairs $(a,b) \in   
695: A^2$, and multiplication and conjugation are defined using equations   
696: (\ref{cd1}) and (\ref{cd2}).   The following propositions show the   
697: effect of repeatedly applying the Cayley--Dickson construction:   
698:    
699: \begin{prop} \et \label{CD1}   
700: $A'$ is never real.     
701: \end{prop}   
702:    
703: \begin{prop} \et \label{CD2}   
704: $A$ is real (and thus commutative) $\iff$ $A'$ is commutative.   
705: \end{prop}     
706:    
707: \begin{prop} \et \label{CD3}   
708: $A$ is commutative and associative $\iff$ $A'$ is associative.    
709: \end{prop}   
710:    
711: \begin{prop} \et \label{CD4}   
712: $A$ is associative and nicely normed $\iff$   
713: $A'$ is alternative and nicely normed.   
714: \end{prop}      
715:    
716: \begin{prop} \et\label{CD5}   
717: $A$ is nicely normed $\iff$ $A'$ is nicely normed.    
718: \end{prop}      
719:    
720: \noindent All of these follow from straightforward calculations; to   
721: prove them here would merely deprive the reader of the pleasure of doing   
722: so.  It follows from these propositions that:   
723: \begin{center}   
724: $\R$ is a real commutative associative nicely normed 
725: $\ast$-algebra $\implies$   
726:     
727: $\C$ is a commutative associative nicely normed $\ast$-algebra $\implies$   
728:    
729: $\H$ is an associative nicely normed $\ast$-algebra $\implies$   
730:    
731: $\O$ is an alternative nicely normed $\ast$-algebra    
732: \end{center}   
733: and therefore that $\R,\C,\H,$ and $\O$ are normed division algebras.     
734: It also follows that the octonions are neither real, nor commutative, nor   
735: associative.       
736:    
737: If we keep applying the Cayley--Dickson process to the octonions we get a   
738: sequence of $\ast$-algebras of dimension 16, 32, 64, and so on.   The   
739: first of these is called the {\bf sedenions}, presumably alluding to the   
740: fact that it is 16-dimensional \cite{LPS}.   It follows from the above   
741: results that all the $\ast$-algebras in this sequence are nicely normed   
742: but neither real, nor commutative, nor alternative.  They all have   
743: multiplicative inverses, since they are nicely normed.  But they are not   
744: division algebras, since an explicit calculation demonstrates that the   
745: sedenions, and thus all the rest, have zero divisors.   In fact    
746: \cite{Cohen,Moreno}, the zero divisors of norm one in the sedenions    
747: form a subspace that is homeomorphic to the exceptional Lie group $\G_2$.   
748:    
749: The Cayley--Dickson construction provides a nice way to obtain the
750: sequence $\R,\H,\C,\O$ and the basic properties of these algebras.  But
751: what is the meaning of this construction?  To answer this, it is better
752: to define $A'$ as the algebra formed by adjoining to $A$ an element $i$
753: satisfying $i^2 = -1$ together with the following relations:
754: \be  a(ib) = i(a^* b) ,   \qquad   
755:     (ai)b = (ab^*)i,      \qquad   
756:     (ia)(bi^{-1}) = (ab)^*  \label{cd3} \ee   
757: for all $a,b \in A$.  We make $A'$ into a $\ast$-algebra using the   
758: original conjugation on elements of $A$ and setting $i^* = -i$.  It is   
759: easy to check that every element of $A'$ can be uniquely written as $a +   
760: ib$ for some $a,b \in A$, and that this description of the   
761: Cayley--Dickson construction becomes equivalent to our previous one    
762: if we set $(a,b) = a + ib$.     
763:    
764: What is the significance of the relations in (\ref{cd3})?   Simply   
765: this: {\sl they express conjugation in terms of conjugation!}  This is a pun   
766: on the double meaning of the word `conjugation'.  What I really mean is   
767: that they express the $\ast$ operation in $A$ as conjugation by $i$.  In   
768: particular, we have   
769: \[       a^\ast = (ia)i^{-1} = i(ai^{-1})   \]   
770: for all $a \in A$.  Note that when $A'$ is associative, any one of the   
771: relations in (\ref{cd3}) implies the other two.  It is when $A'$ is   
772: nonassociative that we really need all three relations.   
773: 
774: This interpretation of the Cayley--Dickson construction makes it easier   
775: to see what happens as we repeatedly apply the construction starting with   
776: $\R$.  In $\R$ the $\ast$ operation does nothing, so when we do the   
777: Cayley--Dickson construction, conjugation by $i$ must have no effect on   
778: elements of $\R$.  Since $\R$ is commutative, this means that $\C = \R'$   
779: is commutative.  But $\C$ is no longer real, since $i^* = -i$.   
780:    
781: Next let us apply the Cayley--Dickson construction to $\C$.  Since $\C$   
782: is commutative, the $\ast$ operation in $\C$ is an automorphism.  Whenever   
783: we have an associative algebra $A$ equipped with an automorphism $\alpha$,    
784: we can always extend $A$ to a larger associative algebra by adjoining an    
785: invertible element $x$ with    
786: \[                \alpha(a) = xax^{-1}  \]    
787: for all $a \in A$.  Since $\C$ is associative, this means that $\C' =   
788: \H$ is associative.  But since $\C$ is not real, $\H$ cannot be   
789: commutative, since conjugation by the newly adjoined element $i$ must   
790: have a nontrivial effect.   
791:    
792: Finally, let us apply the Cayley--Dickson construction to $\H$.   Since   
793: $\H$ is noncommutative, the $\ast$ operation in $\H$ is not an   
794: automorphism; it is merely an antiautomorphism.   This means we cannot   
795: express it as conjugation by some element of a larger associative   
796: algebra.  Thus $\H' = \O$ must be nonassociative.     
797: 
798: \subsection{Clifford Algebras} \label{clifford}      
799: 
800: William Clifford invented his algebras in 1876 as an attempt to
801: generalize the quaternions to higher dimensions, and he published a
802: paper about them two years later \cite{Clifford}.  Given a real inner
803: product space $V$, the {\bf Clifford algebra} $\Cliff(V)$ is the
804: associative algebra freely generated by $V$ modulo the relations
805: \[  v^2 = -\|v\|^2 \] 
806: for all $v \in V$.  Equivalently, it is the associative algebra 
807: freely generated by $V$ modulo the relations 
808: \[              vw + wv = -2\langle v,w\rangle    \]  
809: for all $v, w \in V$.  If $V = \R^n$ with its usual inner product, we 
810: call this Clifford algebra $\Cliff(n)$.  Concretely, this is the 
811: associative algebra freely generated by $n$ anticommuting square roots 
812: of $-1$.  From this we easily see that  
813: \[  \Cliff(0) = \R, \qquad\qquad \Cliff(1) = \C, \qquad\qquad    
814: \Cliff(2) = \H .\]   
815: So far this sequence resembles the iterated Cayley-Dickson construction 
816: --- but the octonions are {\it not} a Clifford algebra, since they are 
817: nonassociative.   Nonetheless, there is a profound relation between  
818: Clifford algebras and normed division algebras.  This relationship gives
819: a nice way to prove that $\R, \C, \H$ and $\O$ are the only normed  
820: dvivision algebras.  It is also crucial for understanding the  
821: geometrical meaning of the octonions.    
822:   
823: To see this relation, first suppose $\K$ is a normed division algebra.    
824: Left multiplication by any element $a \in \K$ gives an operator  
825: \[   \begin{array}{ccccc}  
826:        L_a &\maps& \K & \to   &  \K  \\  
827:            &     & x &\mapsto&   ax .   
828: \end{array} \]  
829: If $\|a\| = 1$, the operator $L_a$ is norm-preserving, so it maps the  
830: unit sphere of $\K$ to itself.  Since $\K$ is a division algebra, we can  
831: find an operator of this form mapping any point on the unit sphere to  
832: any other point.  The only way the unit sphere in $\K$ can have this much  
833: symmetry is if the norm on $\K$ comes from an inner product.  Even better,  
834: this inner product is unique, since we can use the polarization identity  
835: \[   \langle x, y\rangle = {1\over 2}(\|x+y\|^2 - \|x\|^2 - \|y\|^2) \]   
836: to recover it from the norm.     
837:   
838: Using this inner product, we say an element $a \in \K$ is {\bf imaginary}  
839: if it is orthogonal to the element $1$, and we let $\Im(\K)$ be the space  
840: of imaginary elements of $\K$.  We can also think of $\Im(\K)$ as the  
841: tangent space of the unit sphere in $\K$ at the point $1$.  This has a  
842: nice consequence: since $a \mapsto L_a$ maps the unit sphere in $\K$ to  
843: the Lie group of orthogonal transformations of $\K$, it must send  
844: $\Im(\K)$ to the Lie algebra of skew-adjoint transformations of $\K$.  
845: In short, $L_a$ is skew-adjoint whenever $a$ is imaginary.  
846:   
847: The relation to Clifford algebras shows up when we compute the square of  
848: $L_a$ for $a \in \Im(\K)$.  We can do this most easily when $a$ has norm  
849: $1$.  Then $L_a$ is both orthogonal and skew-adjoint.  For any  
850: orthogonal transformation, we can find some orthonormal basis in which  
851: its matrix is block diagonal, built from $2 \times 2$ blocks that look  
852: like this:  
853: \[   \left( \begin{array}{cc} \cos \theta & \sin \theta \\  
854:                             -\sin \theta & \cos \theta   
855:             \end{array} \right)  \]  
856: and possibly a $1 \times 1$ block like this: $\left( 1 \right)$.  
857: Such a transformation can only be skew-adjoint if it consists solely of  
858: $2 \times 2$ blocks of this form:  
859: \[  \pm \left( \begin{array}{cc} 0 & 1 \\  
860:                              -1 & 0   
861:             \end{array} \right).  \]  
862: In this case, its square is $-1$.  We thus have $L_a^2 = -1$ when  
863: $a \in \Im(\K)$ has norm 1.  It follows that 
864: \[             L_a^2 = -\| a \|^2 \]  
865: for all $a \in \Im(\K)$.  We thus obtain a representation of the Clifford  
866: algebra $\Cliff(\Im(\K))$ on $\K$.   Any $n$-dimensional normed division  
867: algebra thus gives an $n$-dimensional representation of $\Cliff(n-1)$.    
868: As we shall see, this is very constraining.    
869:   
870: We have already described the Clifford algebras up to $\Cliff(2)$.  
871: Further calculations \cite{Harvey,Porteous} give the following table,  
872: where we use $A[n]$ to stand for $n\times n$ matrices with entries in  
873: the algebra $A$:  
874:    
875: \medskip  
876: {\vbox{   
877: \begin{center}   
878: {\small   
879: \begin{tabular}{|c|c|c|c|c|c|c|c|c|}                    \hline   
880: $n$ &    $\Cliff(n)$          \\ \hline   
881: $0$ &    $\R$                 \\ \hline   
882: $1$ &    $\C$                 \\ \hline   
883: $2$ &    $\H$                 \\ \hline   
884: $3$ &    $\H \oplus \H$       \\ \hline   
885: $4$ &    $\H[2]$              \\ \hline   
886: $5$ &    $\C[4]$              \\ \hline   
887: $6$ &    $\R[8]$              \\ \hline   
888: $7$ &    $\R[8] \oplus \R[8]$ \\ \hline   
889: \end{tabular}}  
890: \vskip 1em 
891: \centerline{Table 2 --- Clifford Algebras}  
892: \end{center}   
893: }}   
894: \medskip  
895:    
896: \noindent  
897: Starting at dimension 8, something marvelous happens: the table continues  
898: in the following fashion:   
899: \[    \Cliff(n+8) \iso \Cliff(n) \tensor \R[16]  .\]   
900: In other words, $\Cliff(n+8)$ consists of $16 \times 16$ matrices  
901: with entries in $\Cliff(n)$.  This `period-8' behavior was discovered  
902: by Cartan in 1908 \cite{Cartan2}, but we will take the liberty of 
903: calling it {\bf Bott periodicity}, since it has a far-ranging set of 
904: applications to topology, some of which were discovered by Bott.
905: 
906: Since Clifford algebras are built from matrix algebras over $\R,\C$ and
907: $\H$, it is easy to determine their representations.  Every
908: representation is a direct sum of irreducible ones.  The only
909: irreducible representation of $\R[n]$ is its obvious one via matrix
910: multiplication on $\R^n$.  Similarly, the only irreducible
911: representation of $\C[n]$ is the obvious one on $\C^n$, and the only
912: irreducible representation of $\H[n]$ is the obvious one on $\H^n$.
913:   
914: Glancing at the above table, we see that unless $n$ equals $3$ or $7$
915: modulo $8$, $\Cliff(n)$ is a real, complex or quaternionic matrix
916: algebra, so it has a unique irreducible representation.  For reasons to
917: be explained later, this irreducible representation is known as the
918: space of {\bf pinors} and denoted $P_n$.  When $n$ is $3$ or $7$ modulo
919: $8$, the algebra $\Cliff(n)$ is a direct sum of two real or quaternionic
920: matrix algebras, so it has two irreducible representations, which we
921: call the {\bf positive pinors} $P_n^+$ and {\bf negative pinors}
922: $P_n^-$.  We summarize these results in the following table:
923:   
924: \medskip  
925: {\vbox{   
926: \begin{center}   
927: {\small   
928: \begin{tabular}{|c|c|l|}                            \hline   
929: $n$ &    $\Cliff(n)$         & irreducible representations       \\ \hline
930: $0$ &    $\R$                & $P_0 = \R$                        \\ \hline   
931: $1$ &    $\C$                & $P_1 = \C$                        \\ \hline   
932: $2$ &    $\H$                & $P_2 = \H$                        \\ \hline   
933: $3$ &    $\H \oplus \H$      & $P^+_3 = \H,\, P^-_3 =\H$         \\ \hline   
934: $4$ &    $\H[2]$             & $P_4 = \H^2$                      \\ \hline   
935: $5$ &    $\C[4]$             & $P_5 = \C^4$                      \\ \hline   
936: $6$ &    $\R[8]$             & $P_6 =\R^8$                       \\ \hline   
937: $7$ &    $\R[8] \oplus \R[8]$& $P_7^+ = \R^8,\, P_7^- =\R^8$     \\ \hline   
938: \end{tabular}}  
939: \vskip 1em 
940: \centerline{Table 3 --- Pinor Representations}  
941: \end{center}   
942: }}   
943: \medskip  
944:   
945: \noindent 
946: Examining this table, we see that in the range of dimensions listed
947: there is an $n$-dimensional representation of $\Cliff(n-1)$ only for $n
948: = 1,2,4,$ and $8$.  What about higher dimensions?  By Bott periodicity,
949: the irreducible representations of $\Cliff(n+8)$ are obtained by
950: tensoring those of $\Cliff(n)$ by $\R^{16}$.  This multiplies the
951: dimension by 16, so one can easily check that for $n > 8$, the
952: irreducible representations of $\Cliff(n-1)$ always have dimension
953: greater than $n$.  
954: 
955: It follows that normed division algebras are only possible in dimensions  
956: $1,2,4,$ and $8$.  Having constructed $\R,\C,\H$ and $\O$, we also know  
957: that normed division algebras {\it exist} in these dimensions.  The only  
958: remaining question is whether they are {\it unique}.  For this it helps  
959: to investigate more deeply the relation between normed division algebras  
960: and the Cayley-Dickson construction.   In what follows, we outline an  
961: approach based on ideas in the book by Springer and Veldkamp \cite{SV}.   
962:   
963: First, suppose $\K$ is a normed division algebra.  Then there is a unique  
964: linear operator $\ast \maps \K \to \K$ such that $1^\ast = 1$ and $a^\ast  
965: = -a$ for $a \in \Im(\K)$.  With some calculation one can prove this  
966: makes $\K$ into a nicely normed $\ast$-algebra.    
967:   
968: Next, suppose that $\K_0$ is any subalgebra of the normed division algebra  
969: $\K$.  It is easy to check that $\K_0$ is a nicely normed $\ast$-algebra in  
970: its own right.  If $\K_0$ is not all of $\K$, we can find an element $i \in 
971: \K$ that is orthogonal to every element of $\K_0$.   Without loss of 
972: generality we shall assume this element has norm 1.  Since this element 
973: $i$ is orthogonal to $1 \in \K_0$, it is imaginary.  From the  definition 
974: of the $\ast$ operator it follows that $i^\ast = -i$, and from results 
975: earlier in this section we have $i^2 = -1$.  With  further calculation 
976: one can show that for all $a,a' \in \K_0$ we have   
977: \[  a(ia') = i(a^* a') ,  \qquad   
978:     (ai)a' = (aa'^*)i,      \qquad   
979:     (ia)(a'i^{-1}) = (aa')^*  \]  
980: A glance at equation (\ref{cd3}) reveals that these are exactly the  
981: relations defining the Cayley-Dickson construction!  With a little  
982: thought, it follows that the subalgebra of $\K$ generated by $\K_0$ and $i$  
983: is isomorphic as a $\ast$-algebra to $\K'_0$, the $\ast$-algebra obtained  
984: from $\K_0$ by the Cayley-Dickson construction.  
985:   
986: Thus, whenever we have a normed division algebra $\K$ we can find a 
987: chain of subalgebras $\R = \K_0 \subset \K_1 \subset \cdots \subset 
988: \K_n = \K$ such that $\K_{i+1} \iso \K_i'$.  To construct $\K_{i+1}$, we 
989: simply need to choose a norm-one element of $\K$ that is orthogonal to 
990: every element of $\K_i$.    It follows that the only normed division 
991: algebras of dimension 1, 2, 4 and 8 are $\R,\C,\H$ and $\O$.   This also 
992: gives an alternate proof that there are no normed division algebras of 
993: other dimensions: if there were any, there would have to be a 
994: 16-dimensional one, namely $\O'$ --- the sedenions.  But as mentioned
995: in Section \ref{cayley-dickson}, one can check explicitly that the 
996: sedenions are not a division algebra.    
997: 
998: \subsection{Spinors and Trialities}  \label{triality}   
999:    
1000: A nonassociative division algebra may seem like a strange thing to   
1001: bother with, but the notion of triality makes it seem a bit more   
1002: natural.  The concept of duality is important throughout linear algebra.   
1003: The concept of triality is similar, but considerably subtler.  Given   
1004: vector spaces $V_1$ and $V_2$, we may define a {\bf duality} to be a   
1005: bilinear map    
1006: \[    f \maps V_1 \times V_2 \to \R   \]   
1007: that is nondegenerate, meaning that if we fix either argument   
1008: to any nonzero value, the linear functional induced on the other vector   
1009: space is nonzero.  Similarly, given vector spaces $V_1,V_2,$ and $V_3$,   
1010: a {\bf triality} is a trilinear map    
1011: \[    t \maps V_1 \times V_2 \times V_3 \to \R   \]   
1012: that is nondegenerate in the sense that if we fix any two arguments to   
1013: any nonzero values, the linear functional induced on the third vector   
1014: space is nonzero.   
1015:    
1016: Dualities are easy to come by.  Trialities are much rarer.  For suppose   
1017: we have a triality  
1018: \[    t \maps V_1 \times V_2 \times V_3 \to \R  . \]   
1019: By dualizing, we can turn this into a bilinear map   
1020: \[     m \maps V_1 \times V_2 \to V_3^\ast     \]   
1021: which we call `multiplication'.  By the nondegeneracy of our triality,   
1022: left multiplication by any nonzero element of $V_1$ defines an   
1023: isomorphism from $V_2$ to $V_3^\ast$.  Similarly, right multiplication   
1024: by any nonzero element of $V_2$ defines an isomorphism from $V_1$ to   
1025: $V_3^\ast$.  If we choose nonzero elements $e_1 \in V_1$ and $e_2 \in   
1026: V_2$, we can thereby identify the spaces $V_1$, $V_2$ and $V_3^\ast$   
1027: with a single vector space, say $V$.   Note that this identifies   
1028: all three vectors $e_1 \in V_1$, $e_2 \in V_2$, and $e_1e_2 \in V_3^\ast$   
1029: with the same vector $e \in V$.   We thus obtain a product   
1030: \[      m \maps V \times V \to V        \]   
1031: for which $e$ is the left and right unit.  Since left or right   
1032: multiplication by any nonzero element is an isomorphism, $V$ is   
1033: actually a division algebra!   Conversely, any division algebra   
1034: gives a triality.     
1035: 
1036: It follows from Theorem \ref{bott-milnor}  that trialities only occur in
1037: dimensions 1, 2, 4, or 8.  This theorem is quite deep.  By comparison,
1038: Hurwitz's classification of {\it normed} division algebras is easy to
1039: prove.  Not surprisingly, these correspond to a special sort of
1040: triality, which we call a `normed' triality.  
1041:   
1042: To be precise, a {\bf normed triality} consists of inner product  
1043: spaces $V_1, V_2, V_3$ equipped with a trilinear map   
1044: $t \maps V_1 \times V_2 \times V_3 \to \R$ with  
1045: \[      |t(v_1, v_2, v_3)| \le \|v_1\| \, \|v_2\| \, \|v_3 \|,   \]    
1046: and such that for all $v_1, v_2$ there exists $v_3 \ne 0$ for which this   
1047: bound is attained --- and similarly for cyclic permutations of $1,2,3$.    
1048: Given a normed triality, picking unit vectors in any two of the spaces 
1049: $V_i$ allows us to identify all three spaces and get a normed division 
1050: algebra.  Conversely, any normed division algebra gives a normed triality.    
1051:   
1052: But where do normed trialities come from?  They come from the theory of  
1053: spinors!  From Section \ref{clifford}, we already know that any  
1054: $n$-dimensional normed division algebra is a representation of  
1055: $\Cliff(n-1)$, so it makes sense to look for normed trialities here. 
1056: In fact, representations of $\Cliff(n-1)$ give certain representations  
1057: of $\Spin(n)$, the double cover of the rotation group in $n$ dimensions.  
1058: These are called `spinors'.  As we shall see, the relation between   
1059: spinors and vectors gives a nice way to construct normed trialities in  
1060: dimensions 1, 2, 4 and 8.  
1061:   
1062: To see how this works, first let $\Pin(n)$ be the group sitting inside  
1063: $\Cliff(n)$ that consists of all products of unit vectors in $\R^n$.  
1064: This group is a double cover of the orthogonal group $\OO(n)$, where  
1065: given any unit vector $v \in \R^n$, we map both $\pm v \in \Pin(n)$ to  
1066: the element of $\OO(n)$ that reflects across the hyperplane perpendicular to  
1067: $v$.  Since every element of $\OO(n)$ is a product of reflections, this  
1068: homomorphism is indeed onto.      
1069:   
1070: Next, let $\Spin(n) \subset \Pin(n)$ be the subgroup consisting of all  
1071: elements that are a product of an even number of unit vectors in  
1072: $\R^n$.  An element of $\OO(n)$ has determinant 1 iff it is the product  
1073: of an even number of reflections, so just as $\Pin(n)$ is a double cover  
1074: of $\OO(n)$, $\Spin(n)$ is a double cover of $\SO(n)$.  Together with a  
1075: French dirty joke which we shall not explain, this analogy is the origin  
1076: of the terms `$\Pin$' and `pinor'.    
1077:   
1078: Since $\Pin(n)$ sits inside $\Cliff(n)$, the irreducible representations
1079: of $\Cliff(n)$ restrict to representations of $\Pin(n)$, which turn out
1080: to be still irreducible.  These are again called {\bf pinors}, and we
1081: know what they are from Table 3.  Similarly, $\Spin(n)$ sits inside the
1082: subalgebra
1083: \[            \Cliff_0(n) \subseteq \Cliff(n) \]  
1084: consisting of all linear combinations of products of an even number of
1085: vectors in $\R^n$.  Thus the irreducible representations of
1086: $\Cliff_0(n)$ restrict to representations of $\Spin(n)$, which turn out
1087: to be still irreducible.  These are called {\bf spinors} --- but we warn
1088: the reader that this term is also used for many slight variations on
1089: this concept.
1090:   
1091: In fact, there is an isomorphism   
1092: \[      \phi \maps \Cliff(n-1) \to \Cliff_0(n) \]  
1093: given as follows:  
1094: \[       \phi(e_i) = e_i e_n  ,   \qquad \qquad 1 \le i \le n-1 ,\]  
1095: where $\{e_i\}$ is an orthonormal basis for $\R^n$.  Thus spinors in $n$  
1096: dimensions are the same as pinors in $n-1$ dimensions!   Table 3  
1097: therefore yields the following table, where we use similar notation but  
1098: with `$S$' instead of `$P$':  
1099:   
1100: \medskip  
1101: {\vbox{   
1102: \begin{center}   
1103: {\small   
1104: \begin{tabular}{|c|c|l|}                           \hline   
1105: $n$ &  $\Cliff_0(n)$    & irreducible representations            \\ \hline  
1106: $1$ &  $\R$             & $S_1 = \R$                             \\ \hline   
1107: $2$ &  $\C$             & $S_2 = \C$                             \\ \hline   
1108: $3$ &  $\H$             & $S_3 = \H$                             \\ \hline   
1109: $4$ &  $\H \oplus \H$   & $S_4^+ = \H, \, S_4^- = \H$            \\ \hline   
1110: $5$ &  $\H[2]$          & $S_5 = \H^2$                           \\ \hline   
1111: $6$ &  $\C[4]$          & $S_6 = \C^4$                           \\ \hline   
1112: $7$ &  $\R[8]$          & $S_7 = \R^8$                           \\ \hline   
1113: $8$ &  $\R[8] \oplus \R[8]$ & $S_8^+ = \R^8,\, S_8^- = \R^8$     \\ \hline   
1114: \end{tabular}}  
1115: \vskip 1em 
1116: \centerline{Table 4 --- Spinor Representations}  
1117: \end{center}   
1118: }}   
1119: \medskip   
1120:   
1121: \noindent   
1122: We call $S_n^+$ and $S_n^-$ the {\bf right-handed} and {\bf left-handed}  
1123: spinor representations.  For $n > 8$ we can work out the spinor  
1124: representations using Bott periodicity:  
1125: \[      S_{n+8} \iso S_n \tensor \R^{16}   \]  
1126: and similarly for right-handed and left-handed spinors.  
1127:   
1128: 
1129: Now, besides its pinor representation(s), the group $\Pin(n)$ also has
1130: an irreducible representation where we first apply the 2--1 homomorphism
1131: $\Pin(n) \to \OO(n)$ and then use the obvious representation of $\OO(n)$
1132: on $\R^n$.  We call this the {\bf vector} representation, $V_n$.  As a
1133: vector space $V_n$ is just $\R^n$, and $\Cliff(n)$ is generated by
1134: $\R^n$, so we have an inclusion
1135: \[                  V_n \hookrightarrow \Cliff(n)  .\]  
1136: Using this, we can restrict the action of the Clifford algebra on pinors  
1137: to a map  
1138: \[    
1139: \begin{array}{lcc}       
1140: m_n \maps& V_n \times P_n^\pm \to P_n^\pm &n \equiv 3,7 \bmod 8  \\  
1141: m_n \maps& V_n \times P_n \to P_n         & {\rm otherwise.}   
1142: \end{array}   
1143: \]  
1144: This map is actually an intertwining operator between 
1145: representations of $\Pin(n)$.  If we restrict the vector representation 
1146: to the subgroup $\Spin(n)$, it remains irreducible.  This is not always 
1147: true for the pinor representations, but we can always decompose them as 
1148: a direct sum of spinor representations.  Applying this decomposition to 
1149: the map $m_n$, we get a map 
1150: \[    
1151: \begin{array}{lc}        
1152: m_n \maps V_n \times S_n^\pm \to S_n^\mp   &n \equiv 0,4 \bmod 8  \\   
1153: m_n \maps V_n \times S_n \to S_n         & {\rm otherwise.}    
1154: \end{array}    
1155: \]   
1156: All the spinor representations appearing here are self-dual, so we can  
1157: dualize the above maps and reinterpret them as trilinear maps   
1158: \[  
1159: \begin{array}{lc}        
1160: t_n \maps V_n \times S_n^+ \times S_n^- \to \R  & n \equiv 0,4 \bmod 8  \\   
1161: t_n \maps V_n \times S_n \times S_n \to \R      & {\rm otherwise.}    
1162: \end{array}   
1163: \]   
1164:  
1165: These trilinear maps are candidates for trialities!  However, they can  
1166: only be trialities when the dimension of the vector representation  
1167: matches that of the relevant spinor representations.    In the range of  
1168: the above table this happens only for $n = 1,2,4,8$.  In these cases we 
1169: actually do get normed trialities, which in turn give normed division algebras: 
1170: \[ 
1171: \begin{array}{ll} 
1172:   t_1 \maps V_1 \times S_1 \times S_1 \to \R  & 
1173: {\rm \; gives \;} \R.   \\ 
1174:    t_2 \maps V_2 \times S_2 \times S_2 \to \R & 
1175: {\rm \; gives \;} \C.    \\ 
1176:    t_4 \maps V_4 \times S_4^+ \times S_4^- \to \R & 
1177: {\rm \; gives \;} \H.    \\ 
1178:    t_8 \maps V_8 \times S_8^+ \times S_8^- \to \R & 
1179: {\rm \; gives \;} \O .  
1180: \end{array} 
1181: \] 
1182: In higher dimensions, the spinor representations become bigger than the 
1183: vector representation, so we get no more trialities this way --- and of 
1184: course, none exist.  
1185:  
1186: Of the four normed trialities, the one that gives the octonions 
1187: has an interesting property that the rest lack.  To see this property, 
1188: one must pay careful attention to the difference between a normed triality 
1189: and a normed division algebra.  To construct a normed division $\K$ 
1190: algebra from the normed triality $t \maps V_1 \times V_2 \times V_3 \to \R$, 
1191: we must arbitrarily choose unit vectors in two of the three spaces, so 
1192: the symmetry group of $\K$ is smaller than that of $t$.  More precisely,  
1193: let us define a {\bf automorphism} of the normed triality $t \maps V_1 \times 
1194: V_2 \times V_3 \to \R$ to be a triple of norm--preserving maps  
1195: $f_i \maps V_i \to V_i$ such that  
1196: \[        t(f_1(v_1), f_2(v_2), f_3(v_3)) = t(v_1,v_2,v_3)  \] 
1197: for all $v_i \in V_i$.   These automorphisms form a group we call  
1198: $\Aut(t)$.  If we construct a normed division algebra $\K$ from $t$ 
1199: by choosing unit vectors $e_1 \in V_1, e_2 \in V_2$, we have  
1200: \[   
1201: \Aut(\K) \iso \{(f_1,f_2,f_3) \in \Aut(t)\; \colon \; f_1(e_1) = e_1,  
1202: \; f_2(e_2) = e_2 \} . 
1203: \] 
1204:  
1205: In particular, it turns out that: 
1206: \be
1207: \begin{array}{lclclcl} 
1208:     1 &\iso& \Aut(\R) & \subseteq& \Aut(t_1) & \iso &
1209: \{ (g_1,g_2,g_3) \in \OO(1)^3 \colon \; g_1g_2g_3 = 1 \}  \\        
1210:     \Z_2 &\iso& \Aut(\C) & \subseteq& \Aut(t_2)& \iso &
1211: \{ (g_1,g_2,g_3) \in \U(1)^3 \colon \; g_1g_2g_3 = 1 \} \times \Z_2  \\        
1212:     \SO(3) &\iso& \Aut(\H) &\subseteq &\Aut(t_4)& \iso  &
1213: \Sp(1)^3 / \{ \pm(1,1,1) \} \\ 
1214:      \G_2 &\iso& \Aut(\O) &\subseteq& \Aut(t_8)& \iso& \Spin(8)  
1215: \end{array} 
1216: \label{Aut(t)}
1217: \ee
1218: where 
1219: \[ \OO(1) \iso \Z_2, \qquad \U(1) \iso \SO(2) , \qquad \Sp(1) \iso \SU(2) \] 
1220: are the unit spheres in $\R$, $\C$ and $\H$, respectively ---
1221: the only spheres that are Lie groups.
1222: $\G_2$ is just another name for the automorphism group of 
1223: the octonions; we shall study this group in Section \ref{G2}.   
1224: The bigger group $\Spin(8)$ acts as automorphisms of the triality  
1225: that gives the octonions, and it does so in an interesting way.
1226: Given any element $g \in \Spin(8)$, there exist unique elements
1227: $g_\pm \in \Spin(8)$ such that 
1228: \[      t(g(v_1), g_+(v_2), g_-(v_3)) = t(v_1,v_2,v_3)  \] 
1229: for all $v_1 \in V_8, v_2 \in S^+_8,$ and $v_3 \in S^-_8$.
1230: Moreover, the maps 
1231: \[   \alpha_\pm \maps g \to g_\pm  \] 
1232: are outer automorphisms of $\Spin(8)$.  In fact ${\rm Out(\Spin(8))}$
1233: is the permutation group on 3 letters, and there exist outer
1234: automorphisms that have the effect of permuting the vector, left-handed
1235: spinor, and right-handed spinor representations any way one likes;
1236: $\alpha_+$ and $\alpha_-$ are among these.   
1237: 
1238: In general, outer automorphisms of simple Lie groups come from
1239: symmetries of their Dynkin diagrams.  Of all the simple Lie groups,
1240: $\Spin(8)$ has the most symmetrical Dynkin diagram!  It looks like this:
1241: 
1242: \medskip 
1243: \centerline{\epsfysize=1.0in\epsfbox{triality.eps}}   
1244: \label{triality.figure}   
1245: \medskip
1246:  
1247: \noindent   
1248: Here the three outer nodes correspond to the vector, left-handed spinor
1249: and right-handed spinor representations of $\Spin(8)$, while the central
1250: node corresponds to the adjoint representation --- that is, the
1251: representation of $\Spin(8)$ on its own Lie algebra, better known as
1252: $\so(8)$.  The outer automorphisms corresponding to the symmetries of
1253: this diagram were discovered in 1925 by Cartan \cite{Cartan3}, who
1254: called these symmetries {\bf triality}.  The more general notion of
1255: `triality' we have been discussing here came later, and is apparently
1256: due to Adams \cite{Adams}.
1257:  
1258: The construction of division algebras from trialities has tantalizing 
1259: links to physics.   In the Standard Model of particle physics, all 
1260: particles other than the Higgs boson transform either as vectors or 
1261: spinors.  The vector particles are also called `gauge bosons', and they
1262: serve to carry the {\it forces} in the Standard Model.  The spinor
1263: particles  are also called `fermions', and they correspond to the basic
1264: forms of {\it matter}: quarks and leptons.   The interaction between
1265: matter and the forces is described by a trilinear map involving two
1266: spinors and one vector.  This map is often drawn as a Feynman diagram:  
1267: 
1268: \medskip
1269: \centerline{\epsfysize=1.0in\epsfbox{feynman.eps}}   
1270: \label{feynman}   
1271: \medskip 
1272: \noindent  
1273: where the straight lines denote spinors and the wiggly one denotes a 
1274: vector.  The most familiar example is the process whereby an electron 
1275: emits or absorbs a photon.   
1276:  
1277: It is fascinating that the same sort of mathematics can be used both to
1278: construct the normed division algebras and to describe the interaction
1279: between matter and forces.  Could this be important for physics?  One
1280: {\it prima facie} problem with this speculation is that physics uses
1281: spinors associated to Lorentz groups rather than rotation groups, due to
1282: the fact that spacetime has a Lorentzian rather than Euclidean metric.
1283: However, in Section \ref{lorentz} we describe a way around this problem.
1284: Just as octonions give the spinor representations of $\Spin(8)$, pairs
1285: of octonions give the spinor representations of $\Spin(9,1)$.  This is
1286: one reason so many theories of physics work best when spacetime is
1287: 10-dimensional!  Examples include superstring theory \cite{Deligne,GSW},
1288: supersymmetric gauge theories \cite{Evans,KT,Schray}, and Geoffrey
1289: Dixon's extension of the Standard Model based on the algebra $\C \tensor
1290: \H \tensor \O$, in which the 3 forces arise naturally from the three
1291: factors in this tensor product \cite{Dixon}.
1292:  
1293: \section{Octonionic Projective Geometry}   \label{proj} 
1294:  
1295: Projective geometry is a venerable subject that has its origins in the 
1296: study of perspective by Renaissance painters.  As seen by the eye, 
1297: parallel lines --- e.g., train tracks --- appear to meet at a `point at 
1298: infinity'.  When one changes ones viewpoint, distances and angles appear 
1299: to change, but points remain points and lines remain lines.  These facts 
1300: suggest a modification of Euclidean plane geometry, based on a set of 
1301: points, a set of lines, and relation whereby a point `lies on' a line, 
1302: satisfying the following axioms: 
1303: \begin{itemize} 
1304: \item For any two distinct points, there is a unique line on which they 
1305: both lie. 
1306: \item For any two distinct lines, there is a unique point which lies on 
1307: both of them. 
1308: \item There exist four points, no three of which lie on the same line. 
1309: \item There exist four lines, no three of which have the same point lying 
1310: on them.   
1311: \end{itemize} 
1312: A structure satisfying these axioms is called a {\bf projective plane}. 
1313: Part of the charm of this definition is that it is `self-dual': if we 
1314: switch the words `point' and `line' and switch who lies on whom, it 
1315: stays the same.   
1316:  
1317: We have already met one example of a projective plane in Section  
1318: \ref{fano}: the smallest one of all, the Fano plane.  The example 
1319: relevant to perspective is the real projective plane, $\RP^2$.  Here the 
1320: points are lines through the origin in $\R^3$, the lines are planes 
1321: through the origin in $\R^3$, and the relation of `lying on' is taken to 
1322: be inclusion.  Each point $(x,y) \in \R^2$ determines a point in 
1323: $\RP^2$, namely the line in $\R^3$ containing the origin and the point 
1324: $(x,y,-1)$:
1325: 
1326: \centerline{\epsfysize=2in\epsfbox{plane.eps}}   
1327: \label{plane}   
1328: 
1329: \noindent
1330: There are also other points in $\RP^2$, the `points at infinity',
1331: corresponding to lines through the origin in $\R^3$ that do not
1332: intersect the plane $\{z = -1\}$.   For example, any point on the 
1333: horizon in the above picture determines a point at infinity. 
1334:  
1335: Projective geometry is also interesting in higher dimensions. 
1336: One can define a {\bf projective space} by the following axioms: 
1337: \begin{itemize} 
1338: \item For any two distinct points $p,q$, there is a unique line 
1339: $pq$ on which they both lie. 
1340: \item For any line, there are at least three points lying on this line. 
1341: \item If $a,b,c,d$ are distinct points and there is a point lying 
1342: on both $ab$ and $cd$, then there is a point lying on both $ac$ and $bd$. 
1343: \end{itemize} 
1344: Given a projective space and a set $S$ of points in this space, we 
1345: define the {\bf span} of $S$ to be the smallest set $T$ of points
1346: containing $S$ such that if $a$ and $b$ lie in $T$, so do all points
1347: on the line $ab$.  The {\bf dimension}  
1348: of a projective space is defined to be one less than the minimal 
1349: cardinality of a set that spans the whole space.  The reader may enjoy 
1350: showing that a 2-dimensional projective space is the same thing as a 
1351: projective plane \cite{Garner}. 
1352:  
1353: If $\K$ is any field, there is an $n$-dimensional projective space  
1354: called $\KP^n$ where the points are lines through the origin in 
1355: $\K^{n+1}$, the lines are planes through the origin in $\K^{n+1}$, and 
1356: the relation of `lying on' is inclusion.  In fact, this construction 
1357: works even when $\K$ is a mere {\bf skew field}: a ring such that every 
1358: nonzero element has a left and right multiplicative inverse.   We just 
1359: need to be a bit careful about defining lines and planes through the 
1360: origin in $\K^{n+1}$.  To do this, we use the fact that $\K^{n+1}$ is a 
1361: $\K$-bimodule in an obvious way.  We take a line through the origin to 
1362: be any set  
1363: \[   L = \{ \alpha x \; \colon\; \alpha \in \K \}  \] 
1364: where $x \in \K^{n+1}$ is nonzero, and take a plane through the 
1365: origin to be any set 
1366: \[   P = \{ \alpha x + \beta y \; \colon \; \alpha,\beta \in \K \} \] 
1367: where $x,y \in \K^{n+1}$ are elements such that  
1368: $\alpha x + \beta y = 0$ implies $\alpha,\beta = 0$.  
1369:  
1370: Given this example, the question naturally arises whether {\it every} 
1371: projective $n$-space is of the form $\KP^n$ for some skew field $\K$. 
1372: The answer is quite surprising: yes, but only if $n > 2$.  Projective 
1373: planes are more subtle \cite{Stevenson}.  A projective plane comes 
1374: from a skew field if and only if it satisfies an extra axiom, the 
1375: `axiom of Desargues', which goes as follows.  Define a {\bf triangle}  
1376: to be a triple of points that don't all lie on the same line.  Now,  
1377: suppose we have two triangles $xyz$ and $x'y'z'$.  The sides of each  
1378: triangle determine three lines, say $LMN$ and $L'M'N'$.  Sometimes  
1379: the line through $x$ and $x'$, the line through $y$ and $y'$, and  
1380: the line through $z$ and $z'$ will all intersect at the same point: 
1381: 
1382: \vskip 1em
1383: {\hskip 15em}{\epsfysize=2in\epsfbox{desargues1.eps}}   
1384: \label{desargues1}   
1385: \vskip -3em 
1386: 
1387: \noindent 
1388: The {\bf axiom of Desargues} says that whenever this happens, something 
1389: else happens: the intersection of $L$ and $L'$, the intersection of $M$ 
1390: and $M'$, and the intersection of $N$ and $N'$ all lie on the same line: 
1391:  
1392: {\hskip 11em}{\epsfysize=2in\epsfbox{desargues2.eps}}   
1393: \label{desargues2}   
1394:  
1395:  
1396: \noindent  
1397: This axiom holds automatically for projective spaces of dimension 3 
1398: or more, but not for projective planes.  A projective plane satisfying  
1399: this axiom is called {\bf Desarguesian}. 
1400:  
1401: The axiom of Desargues is pretty, but what is its connection to skew 
1402: fields?  Suppose we start with a projective plane $P$ and try to 
1403: reconstruct a skew field from it.  We can choose any line $L$, choose 
1404: three distinct points on this line, call them $0, 1$, and $\infty$, and 
1405: set $\K = L - \{\infty\}$.   Copying geometric constructions that work 
1406: when $P = \RP^2$, we can define addition and multiplication of points in 
1407: $\K$.   In general the resulting structure $(\K,+,0,\cdot,1)$ will not 
1408: be a skew field.  Even worse, it will depend in a nontrivial way on the 
1409: choices made.  However, if we assume the axiom of Desargues, these 
1410: problems go away.  We thus obtain a one-to-one correspondence between 
1411: isomorphism classes of skew fields and isomorphism classes of 
1412: Desarguesian projective planes. 
1413:  
1414: Projective geometry was very fashionable in the 1800s, with such  
1415: worthies as Poncelet, Brianchon, Steiner and von Staudt making important 
1416: contributions.  Later it was overshadowed by other forms of geometry.   
1417: However, work on the subject continued, and in 1933 Ruth Moufang 
1418: constructed a remarkable example of a non-Desarguesian projective plane 
1419: using the octonions \cite{Moufang}.  As we shall see, this projective 
1420: plane deserves the name $\OP^2$.   
1421:  
1422: The 1930s also saw the rise of another reason for interest in projective 
1423: geometry: quantum mechanics!  Quantum theory is distressingly different 
1424: from the classical Newtonian physics we have learnt to love.  In 
1425: classical mechanics, observables are described by real-valued functions. 
1426: In quantum mechanics, they are often described by hermitian $n \times 
1427: n$ complex matrices.   In both cases, observables are closed under 
1428: addition and multiplication by real scalars.  However, in quantum 
1429: mechanics, observables do not form an associative algebra.   Still,  
1430: one can raise an observable to a power, and from squaring one 
1431: can construct a commutative but nonassociative product: 
1432: \[     a \circ b = {1\over 2}((a+b)^2 - a^2 - b^2) 
1433:                   = {1\over 2}(ab + ba) . \] 
1434: In 1932, Pascual Jordan attempted to understand this situation better by 
1435: isolating  the bare minimum axioms that an `algebra of observables' 
1436: should satisfy \cite{Jordan}.  He invented the definition of what is now 
1437: called a {\bf formally real Jordan algebra}: a commutative and 
1438: power-associative algebra satisfying 
1439: \[  a_1^2 + \cdots + a_n^2 = 0 \quad \implies \quad a_1 = \cdots = a_n = 0  \] 
1440: for all $n$.   The last condition gives the algebra a partial 
1441: ordering: if we write $a \le b$ when the element $b - a$ is a sum of 
1442: squares, it says that $a \le b$ and $b \le a$ imply $a = b$.  Though it 
1443: is not obvious, any formally real Jordan algebra satisfies the identity  
1444: \[         a \circ (b \circ a^2) = (a \circ b) \circ a^2 \] 
1445: for all elements $a$ and $b$.   Any commutative algebra satisfying 
1446: this identity is called a {\bf Jordan algebra}.  Jordan algebras are 
1447: automatically power-associative.   
1448:  
1449: In 1934, Jordan published a paper with von Neumann and Wigner  
1450: classifying all formally real Jordan algebras \cite{JNW}.  The 
1451: classification is nice and succinct.  An {\bf ideal} in the Jordan algebra 
1452: $A$ is a subspace $B \subseteq A$ such that $b \in B$ implies $a \circ b 
1453: \in B$ for all $a \in A$.  A Jordan algebra $A$ is {\bf simple} if its 
1454: only ideals are $\{0\}$ and $A$ itself.  Every formally real Jordan 
1455: algebra is a direct sum of simple ones.  The simple formally real Jordan 
1456: algebras consist of 4 infinite families and one exception.   
1457: \begin{enumerate} 
1458: \item The algebra $\h_n(\R)$  
1459: with the product $a \circ b = {1\over 2}(ab + ba)$.   
1460: \item The algebra $\h_n(\C)$  
1461: with the product $a \circ b = {1\over 2}(ab + ba)$.   
1462: \item The algebra $\h_n(\H)$  
1463: with the product $a \circ b = {1\over 2}(ab + ba)$.   
1464: \item The algebra $\R^n \oplus \R$ with the product  
1465: \[  (v,\alpha) \circ (w, \beta) =  
1466: (\alpha w + \beta v, \langle v,w\rangle + \alpha \beta).  \] 
1467: \item The algebra $\h_3(\O)$  
1468: with the product $a \circ b = {1\over 2}(ab + ba)$.   
1469: \end{enumerate} 
1470: Here we say a square matrix with entries in the $\ast$-algebra $A$ is 
1471: {\bf hermitian} if it equals its conjugate transpose, and we let 
1472: $\h_n(A)$ stand for the hermitian $n \times n$ matrices with entries in 
1473: $A$.  Jordan algebras in the fourth family are called {\bf spin 
1474: factors}, while $\h_3(\O)$ is called the {\bf exceptional Jordan 
1475: algebra}.  This classification raises some obvious questions.  Why does 
1476: nature prefer the Jordan algebras $\h_n(\C)$ over all the rest?  Or does 
1477: it?  Could the other Jordan algebras --- even the exceptional one --- 
1478: have some role to play in quantum physics?  Despite much research, these 
1479: questions remain unanswered to this day. 
1480:  
1481: The paper by Jordan, von Neumann and Wigner appears to have been 
1482: uninfluenced by Moufang's discovery of $\OP^2$, but in fact they are 
1483: related.  A {\bf projection} in a formally real Jordan algebra is 
1484: defined to be an element $p$ with $p^2 = p$.  In the familiar case of 
1485: $\h_n(\C)$, these correspond to hermitian matrices with eigenvalues $0$ 
1486: and $1$, so they are used to describe observables that assume only two 
1487: values --- e.g., `true' and `false'.  This suggests treating projections 
1488: in a formally real Jordan algebra as propositions in a kind of `quantum 
1489: logic'.  The partial order helps us do this: given projections $p$ and 
1490: $q$, we say that $p$ `implies' $q$ if $p \le q$.    
1491:  
1492: The relation between Jordan algebras and quantum logic is already 
1493: interesting \cite{Emch}, but the real fun starts when we note 
1494: that projections in $\h_n(\C)$ correspond to subspaces of $\C^n$.   This 
1495: sets up a relationship to projective geometry \cite{Varadarajan}, since 
1496: the projections onto 1-dimensional subspaces correspond to points in 
1497: $\CP^n$, while the projections onto 2-dimensional subspaces correspond 
1498: to lines.  Even better, we can work out the dimension of a subspace $V 
1499: \subseteq \C^n$ from the corresponding projection $p \maps \C^n \to V$ 
1500: using only the partial order on projections: $V$ has dimension $d$ iff 
1501: the longest chain of distinct projections  
1502: \[   0 = p_0 < \cdots < p_i = p  \] 
1503: has length $i = d$.  In fact, we can use this to define the {\bf rank} 
1504: of a projection in any formally real Jordan algebra.  We can then try to 
1505: construct a projective space whose points are the rank-1 projections and 
1506: whose lines are the rank-2 projections, with the relation of `lying on' 
1507: given by the partial order $\le$.   
1508:  
1509: If we try this starting with $\h_n(\R)$, $\h_n(\C)$ or $\h_n(\H)$, we 
1510: succeed when $n \ge 2$, and we obtain the projective spaces $\RP^n$, 
1511: $\CP^n$ and $\HP^n$, respectively.  If we try this starting with the 
1512: spin factor $\R^n \oplus \R$ we succeed when $n \ge 2$, and obtain a 
1513: series of 1-dimensional projective spaces related to Lorentzian 
1514: geometry.  Finally, in 1949 Jordan \cite{Jordan2} discovered that if we 
1515: try this construction starting with the exceptional Jordan algebra, we 
1516: get the projective plane discovered by Moufang: $\OP^2$.  
1517:  
1518: In what follows we describe the octonionic projective plane 
1519: and exceptional Jordan algebra in more detail.  But first let us
1520: consider the octonionic projective line, and the Jordan algebra
1521: $\h_2(\O)$.
1522:  
1523: \subsection{Projective Lines}  \label{OP1} 
1524:  
1525: A one-dimensional projective space is called a {\bf projective line}. 
1526: Projective lines are not very interesting from the viewpoint of 
1527: axiomatic projective geometry, since they have only one line on which 
1528: all the points lie.  Nonetheless, they can be geometrically and 
1529: topologically interesting.  This is especially true of the octonionic 
1530: projective line.  As we shall see, this space has a deep connection to 
1531: Bott periodicity, and also to the Lorentzian geometry of 10-dimensional 
1532: spacetime. 
1533:  
1534: Suppose $\K$ is a normed division algebra.  We have already defined 
1535: $\KP^1$ when $\K$ is associative, but this definition does not work well 
1536: for the octonions: it is wiser to take a detour through Jordan 
1537: algebras.  Let $\h_2(\K)$ be the space of $2 \times 2$ hermitian 
1538: matrices with entries in $\K$.  It is easy to check that this becomes a 
1539: Jordan algebra with the product $a \circ b = {1\over 2}(ab + ba)$.  We 
1540: can try to build a projective space from this Jordan algebra using the 
1541: construction in the previous section.  To see if this 
1542: succeeds, we need to ponder the projections in $\h_2(\K)$.  A little 
1543: calculation shows that besides the trivial projections 0 and 1, they 
1544: are all of the form  
1545: \[   
1546: \left( \begin{array}{c}  x^* \\ y^* \end{array} \right)  
1547: \left( \begin{array}{cc} \! x  &  y \! \end{array} \right)  
1548: =  
1549: \left( \begin{array}{cc}  
1550:                          x^* x  & x^* y    \\  
1551:                          y^* x  & y^* y 
1552: \end{array} \right)  
1553: \] 
1554: where $(x,y) \in \K^2$ has 
1555: \[         \|x\|^2 + \|y\|^2 = 1.   \] 
1556: These nontrivial projections all have rank 1, so they are the points of 
1557: our would--be projective space.  Our would--be projective space has just 
1558: one line, corresponding to the projection 1, and all the points lie on 
1559: this line.  It is easy to check that the axioms for a projective space 
1560: hold.  Since this projective space is 1-dimensional, we have succeeded 
1561: in creating the {\bf projective line over} $\K$.  We call the set of 
1562: points of this projective line $\KP^1$. 
1563:  
1564: Given any nonzero element $(x,y) \in \K^2$, we can normalize it and then 
1565: use the above formula to get a point in $\KP^1$, which we call 
1566: $[(x,y)]$.   This allows us to describe $\KP^1$ in terms 
1567: of lines through the origin, as follows.  Define an equivalence relation 
1568: on nonzero elements of $\K^2$ by 
1569: \[        (x,y) \sim (x',y') \; \iff \; [(x,y)] = [(x',y')]  .\] 
1570: We call an equivalence class for this relation a {\bf line through the 
1571: origin} in $\K^2$.  We can then identify  points in $\KP^1$ with lines 
1572: through the origin in $\K^2$.  
1573:  
1574: Be careful: when $\K$ is the octonions, the line through the 
1575: origin containing $(x,y)$ is not always equal to 
1576: \[    \{(\alpha x, \alpha y)\; \colon \; \alpha \in \K\}.   \] 
1577: This is only true when $\K$ is associative, or when $x$ or $y$ is 
1578: $1$.   Luckily, we have $(x,y) \sim (y^{-1}x,1)$ when $y \ne 0$ and 
1579: $(x,y) \sim (1,x^{-1}y)$ when $x \ne 0$.  Thus in either case we get a 
1580: concrete description of the line through the origin containing $(x,y)$: 
1581: when $x \ne 0$ it equals 
1582: \[    \{(\alpha(y^{-1}x), \alpha)\; \colon \; \alpha \in \K\} ,  \] 
1583: and when $y \ne 0$ it equals 
1584: \[    \{(\alpha, \alpha(x^{-1}y)\; \colon \; \alpha \in \K\} .  \] 
1585: In particular, the line through the origin containing $(x,y)$ is  
1586: always a real vector space isomorphic to $\K$. 
1587:  
1588: We can make $\KP^1$ into a manifold as follows.  By the above 
1589: observations, we can cover it with two coordinate charts: one containing 
1590: all points of the form $[(x,1)]$, the other containing all points of the 
1591: form $[(1,y)]$.   It is easy to check that $[(x,1)] = [(1,y)]$ iff $y = 
1592: x^{-1}$, so the transition function from the first chart to the second 
1593: is the map $x \mapsto x^{-1}$.  Since this transition function and its 
1594: inverse are smooth on the intersection of the two charts, $\KP^1$ 
1595: becomes a smooth manifold.  
1596: 
1597: When pondering the geometry of projective lines it is handy to
1598: visualize the complex case, since $\CP^1$ is just the familiar
1599: `Riemann sphere'.  In this case, the map   
1600: \[    x \mapsto [(x,1)]    \]  
1601: is given by stereographic projection: 
1602:   
1603: \begin{figure}[h]   
1604: \centerline{\epsfysize=1.5in\epsfbox{stereo.eps}}   
1605: \label{stereo}   
1606: \end{figure}   
1607:   
1608: \noindent  
1609: where we choose the sphere to have diameter 1.   This map from $\C$ to 
1610: $\CP^1$ is one-to-one and almost onto, missing only the point at 
1611: infinity, or `north pole'.  Similarly, the map  
1612: \[    y \mapsto [(1,y)]    \]  
1613: misses only the south pole. Composing the first map with the inverse of 
1614: the second, we get the map  $x \mapsto x^{-1}$, which goes by the name 
1615: of `conformal inversion'.  The southern hemisphere of the Riemann 
1616: sphere consists of all points $[(x,1)]$ with $\|x\| \le 1$, while the 
1617: northern hemisphere consists of all $[(1,y)]$ with $\|y\| \le 1$.  Unit 
1618: complex numbers $x$ give points $[(x,1)] = [(1,x^{-1})]$ on the equator.    
1619:   
1620: All these ideas painlessly generalize to $\KP^1$ for any normed division  
1621: algebra $\K$.  First of all, as a smooth manifold $\KP^1$ is just a   
1622: sphere with dimension equal to that of $\K$:  
1623: \[  
1624: \begin{array}{ccl}       
1625:           \RP^1 &\iso& S^1   \\  
1626:           \CP^1 &\iso& S^2   \\  
1627:           \HP^1 &\iso& S^4   \\  
1628:           \OP^1 &\iso& S^8.  
1629: \end{array}   
1630: \]  
1631: We can think of it as the one-point compactification of $\K$.   The  
1632: `southern hemisphere', `northern hemisphere', and `equator' of $\K$ have 
1633: descriptions exactly like those given above for the complex case.  Also, 
1634: as in the complex case, the maps $x \mapsto [(x,1)]$ and $y \mapsto [(1,y)]$ 
1635: are angle-preserving with respect to the usual Euclidean metric on $\K$ 
1636: and the round metric on the sphere.   
1637: 
1638: One of the nice things about $\KP^1$ is that it comes equipped with a 
1639: vector bundle whose fiber over the point $[(x,y)]$ is the line 
1640: through the origin corresponding to this point.  This bundle is called 
1641: the {\bf canonical line bundle}, $L_\K$.  Of course, when we are working 
1642: with a particular division algebra, `line' means a copy of this division 
1643: algebra, so if we think of them as real vector bundles, $L_\R, L_\C, 
1644: L_\H$ and $L_\O$ have dimensions 1,2,4, and 8, respectively.   
1645: 
1646: These bundles play an important role in topology, so it is good to
1647: understand them in a number of ways.  In general, any $k$-dimensional
1648: real vector bundle over $S^n$ can be formed by taking trivial bundles
1649: over the northern and southern hemispheres and gluing  them together
1650: along the equator via a map $f \maps S^{n-1} \to \OO(k)$.  We must
1651: therefore be able to build the canonical line bundles $L_\R, L_\C,
1652: L_\H$ and $L_\O$ using maps  
1653: \[  
1654: \begin{array}{cccl}       
1655:           f_\R \maps &S^0& \to & \OO(1)   \\  
1656:           f_\C \maps &S^1& \to & \OO(2)   \\  
1657:           f_\H \maps &S^3& \to & \OO(4)   \\  
1658:           f_\O \maps &S^7& \to & \OO(8).        
1659: \end{array}   
1660: \]  
1661: What are these maps?  We can describe them all simultaneously.  Suppose  
1662: $\K$ is a normed division algebra of dimension $n$.  In the southern  
1663: hemisphere of $\KP^1$, we can identify any fiber of $L_\K$  with $\K$ by 
1664: mapping the point $(\alpha x, \alpha)$ in the line $[(x,1)]$ to the 
1665: element $\alpha \in \K$.  This trivializes the canonical line bundle 
1666: over the southern hemisphere.  Similarly, we can trivialize this bundle 
1667: over the northern hemisphere by mapping the point $(\beta,\beta y)$ in 
1668: the line $[(1,y)]$ to the element $\beta \in \K$.  If $x \in \K$ has norm 
1669: one, $[(x,1)] = [(1,x^{-1})]$ is a point on the equator, so we get two 
1670: trivializations of the fiber over this point.  These are related as 
1671: follows: if $(\alpha x, \alpha) =  (\beta, \beta x^{-1})$ then $\beta = 
1672: \alpha x$.  The map $\alpha \mapsto \beta$ is thus right multiplication 
1673: by $x$.  In short,   
1674: \[       f_\K \maps S^{n-1} \to \OO(n)  \]  
1675: is just the map sending any norm-one element $x \in \K$ to the operation 
1676: of right multiplication by $x$.   
1677:  
1678: The importance of the map $f_\K$ becomes clearest if we form the 
1679: inductive limit of the groups $\OO(n)$ using the obvious inclusions 
1680: $\OO(n) \hookrightarrow \OO(n+1)$, obtaining a topological group 
1681: called $\OO(\infty)$.  Since $\OO(n)$ is included in $\OO(\infty)$, 
1682: we can think of $f_\K$ as a map from $S^{n-1}$ to $\OO(\infty)$.
1683: Its homotopy class $[f_\K]$ has the following marvelous property,
1684: mentioned in the Introduction:
1685: 
1686: \vbox{
1687: \begin{itemize}   
1688: \item $[f_\R]$ generates $\pi_0(\OO(\infty)) \iso \Z_2$.     
1689: \item $[f_\C]$ generates $\pi_1(\OO(\infty)) \iso \Z$.   
1690: \item $[f_\H]$ generates $\pi_3(\OO(\infty)) \iso \Z$.   
1691: \item $[f_\O]$ generates $\pi_7(\OO(\infty)) \iso \Z$. 
1692: \end{itemize}  
1693: }
1694: 
1695: Another nice perspective on the canonical line bundles $L_\K$ comes from 
1696: looking at their unit sphere bundles.  Any fiber of $L_\K$ is naturally 
1697: an inner product space, since it is a line through the origin in $\K^2$. 
1698: If we take the unit sphere in each fiber, we get a bundle of 
1699: $(n-1)$-spheres over $\KP^1$ called the {\bf Hopf bundle}: 
1700: \[               p_\K \maps E_\K \to \KP^1   \] 
1701: The projection $p_\K$ is called the {\bf Hopf map}.  The total space 
1702: $E_\K$ consists of all the unit vectors in $\K^2$, so it is a sphere  
1703: of dimension $2n-1$.  In short, the Hopf bundles look like this: 
1704: \[     
1705: \begin{array}{crccl}       
1706: \K = \R: \qquad &S^0 \hookrightarrow &S^1&    \to &S^1   \\ 
1707: \K = \C: \qquad &S^1 \hookrightarrow &S^3&    \to &S^2   \\ 
1708: \K = \H: \qquad &S^3 \hookrightarrow &S^7&    \to &S^4   \\ 
1709: \K = \O: \qquad &S^7 \hookrightarrow &S^{15}& \to &S^8    
1710: \end{array} 
1711: \] 
1712: 
1713: We can understand the Hopf maps better by thinking about inverse images 
1714: of points.   The inverse image $p_\K^{-1}(x)$ of any point $x \in S^n$ 
1715: is a $(n-1)$-sphere in $S^{2n-1}$, and the inverse image of any pair of 
1716: distinct points is a pair of linked spheres of this sort.  When $\K = 
1717: \C$  we get linked circles in $S^3$, which form the famous {\bf Hopf link}: 
1718:  
1719: \medskip 
1720: \centerline{\epsfysize=1.0in\epsfbox{hopf.eps}}   
1721: \label{hopf}   
1722: \medskip 
1723:  
1724: \noindent  
1725: When $\K = \O$, we get a pair of linked 7-spheres in $S^{15}$.   
1726:  
1727: To quantify this notion of linking, we can use the `Hopf invariant'.   
1728: Suppose for a moment that $n$ is any natural number greater than one,  
1729: and let $f \maps S^{2n-1} \to S^n$ be any smooth map.  If $\omega$ is 
1730: the normalized volume form on $S^n$, then $f^* \omega$ is a closed 
1731: $n$-form on $S^{2n-1}$.  Since the $n$th cohomology of $S^{2n-1}$ 
1732: vanishes, $f^\ast \omega = d \alpha$ for some $(n-1)$-form $\alpha$. 
1733: We define the {\bf Hopf invariant} of $f$ to be the number 
1734: \[       H(f) = \int_{S^{2n-1}} \alpha \wedge d\alpha  .\] 
1735: This is easily seen to be invariant under smooth homotopies of the map $f$.   
1736: 
1737: To see how the Hopf invariant is related to linking, we can compute it 
1738: using homology rather than cohomology.  If we take any two regular 
1739: values of $f$, say $x$ and $y$, the inverse images of these points are 
1740: compact oriented $(n-1)$-dimensional submanifolds of $S^{2n-1}$.   We 
1741: can always find an oriented $n$-dimensional submanifold $X \subset 
1742: S^{2n-1}$ that has boundary equal to $f^{-1}(x)$ and that intersects 
1743: $f^{-1}(y)$ transversely.  The dimensions of $X$ and $f^{-1}(y)$ add up 
1744: to $2n-1$, so their intersection number is well-defined.  By the  
1745: duality between homology and cohomology, this number equals the Hopf 
1746: invariant $H(f)$.   This shows that the Hopf invariant is an integer.  
1747: Moreover, it shows that when the Hopf invariant is nonzero, the inverse 
1748: images of $x$ and $y$ are linked. 
1749: 
1750: Using either of these approaches we can compute the Hopf invariant of 
1751: $p_\C$, $p_\H$ and $p_\O$.  They all turn out to have Hopf invariant 1.  
1752: This implies, for example, that the inverse images of distinct points 
1753: under $p_\O$ are nontrivially linked 7-spheres in $S^{15}$.  It 
1754: also implies that $p_\C$, $p_\H$ and $p_\O$ give nontrivial elements of 
1755: $\pi_{2n-1}(S^n)$ for $n = 2, 4$, and $8$.  In fact, these elements
1756: generate the torsion-free part of $\pi_{2n-1}(S^n)$.
1757: 
1758: A deep study of the Hopf invariant is one way to prove that any division
1759: algebra must have dimension 1, 2, 4 or 8.  One can show that if there
1760: exists an $n$-dimensional division algebra, then $S^{n-1}$ must be {\bf
1761: parallelizable}: it must admit $n - 1$ pointwise linearly independent
1762: smooth vector fields.  One can also show that for $n > 1$, $S^{n-1}$ is
1763: parallelizable iff there exists a map $f \maps S^{2n-1} \to S^n$ with
1764: $H(f) = 1$ \cite{AH,BM,Kervaire}.  The hard part is showing that a map from
1765: $S^{2n-1}$ to $S^n$ can have Hopf invariant 1 only if $n = 2, 4$, or
1766: $8$.  This was proved by Adams sometime about 1958 \cite{Adams0}.
1767: 
1768: \subsection{$\OP^1$ and Bott Periodicity} \label{bott}  
1769:  
1770: We already touched upon Bott periodicity when we mentioned that the 
1771: Clifford algebra $\Cliff_{n+8}$ is isomorphic to the algebra of $16  
1772: \times 16$ matrices with entries lying in $\Cliff_n$.  This is but one 
1773: of many related `period-8' phenomena that go by the name of Bott  
1774: periodicity.  The appearance of the number 8 here is no coincidence: all 
1775: these phenomena are related to the octonions!  Since this marvelous   
1776: fact is somewhat under-appreciated, it seems worthwhile to say a bit   
1777: about it.   Here we shall focus on those aspects that are related to   
1778: $\OP^1$ and the canonical octonionic line bundle over this space.    
1779:   
1780: Let us start with K-theory.   This is a way of gaining information about 
1781: a topological space by studying the vector bundles over it.  If the  
1782: space has holes in it, there will be nontrivial vector bundles that  
1783: have `twists' as we go around these holes.  The simplest example is   
1784: the `M\"obius strip' bundle over $S^1$, a 1-dimensional real vector 
1785: bundle which has a $180^\circ$ twist as we go around the circle.  In  
1786: fact, this is just the canonical line bundle $L_\R$.   The canonical 
1787: line bundles $L_\C, L_\H$ and $L_\O$ provide higher-dimensional 
1788: analogues of this example. 
1789:  
1790: K-theory tells us to study the vector bundles over a topological space
1791: $X$ by constructing an abelian group as follows.  First, take the set
1792: consisting of all isomorphism classes of real vector bundles over $X$.
1793: Our ability to take direct sums of vector bundles gives this set an
1794: `addition' operation making it into a commutative monoid.  Next, adjoin
1795: formal `additive inverses' for all the elements of this set, obtaining
1796: an abelian group.  This group is called $KO(X)$, the {\bf real K-theory}
1797: of $X$.  Alternatively we could start with complex vector bundles and 
1798: get a group called $K(X)$, but here we will be interested in real vector
1799: bundles.
1800:  
1801: Any real vector bundle $E$ over $X$ gives an element $[E] \in KO(X)$, and 
1802: these elements generate this group.  If we pick a point in $X$, there is an
1803: obvious homomorphism $\dim \maps KO(X) \to \Z$ sending $[E]$ to the
1804: dimension of the fiber of $E$ at this point.  Since the dimension is a
1805: rather obvious and boring invariant of vector bundles, it is nice to
1806: work with the kernel of this homomorphism, which is called the {\bf
1807: reduced} real K-theory of $X$ and denoted $\widetilde{KO}(X)$.  This is
1808: an invariant of pointed spaces, i.e.\ spaces equipped with a designated
1809: point or {\bf basepoint}.
1810:  
1811: Any sphere becomes a pointed space if we take the north pole as
1812: basepoint.  The reduced real K-theory of the first eight spheres 
1813: looks like this:
1814: \ban       
1815:                \widetilde{KO}(S^1) &\iso& \Z_2  \\    
1816:                \widetilde{KO}(S^2) &\iso& \Z_2  \\    
1817:                \widetilde{KO}(S^3) &\iso& 0  \\    
1818:                \widetilde{KO}(S^4) &\iso& \Z  \\    
1819:                \widetilde{KO}(S^5) &\iso& 0  \\    
1820:                \widetilde{KO}(S^6) &\iso& 0  \\    
1821:                \widetilde{KO}(S^7) &\iso& 0  \\    
1822:                \widetilde{KO}(S^8) &\iso& \Z  
1823: \ean  
1824: where, as one might guess, 
1825: \begin{itemize}    
1826: \item $[L_\R]$ generates $\widetilde{KO}(S^1)$.
1827: \item $[L_\C]$ generates $\widetilde{KO}(S^2)$.
1828: \item $[L_\H]$ generates $\widetilde{KO}(S^4)$.
1829: \item $[L_\O]$ generates $\widetilde{KO}(S^8)$.
1830: \end{itemize}
1831: As mentioned in the previous section, one can build any $k$-dimensional
1832: real vector bundle over $S^n$ using a map $f \maps S^{n-1} \to \OO(k)$.
1833: In fact, isomorphism classes of such bundles are in one-to-one correspondence
1834: with homotopy classes of such maps.  Moreover, two such bundles determine 
1835: the same element of $\widetilde{KO}(X)$ if and only if the corresponding
1836: maps become homotopy equivalent after we compose them with the 
1837: inclusion $\OO(k) \hookrightarrow \OO(\infty)$, where 
1838: $\OO(\infty)$ is the direct limit of the groups $\OO(k)$.   It follows that
1839: \[        \widetilde{KO}(S^n) \iso \pi_{n-1}(\OO(\infty)) .\]
1840: This fact gives us the list of homotopy groups of $\OO(\infty)$ which
1841: appears in the Introduction.   It also means that to prove
1842: Bott periodicity for these homotopy groups:
1843: \[   \pi_{i+8}(\OO(\infty)) \iso \pi_i(\OO(\infty)),   \]   
1844: it suffices to prove Bott periodicity for real K-theory:
1845: \[     \widetilde{KO}(S^{n+8}) \iso \widetilde{KO}(S^n) . \]
1846: 
1847: Why do we have Bott periodicity in real K-theory?  It turns out
1848: that there is a graded ring $KO$ with 
1849: \[              KO_n = \widetilde{KO}(S^n)  .\]
1850: The product in this ring comes from our ability to take
1851: `smash products' of spheres and also of real vector bundles over these
1852: spheres.  Multiplying by $[L_\O]$ gives an isomorphism
1853: \[
1854: \begin{array}{ccc}
1855:        \widetilde{KO}(S^n) &\to& \widetilde{KO}(S^{n+8})  \\
1856:                        x   &\mapsto& [L_\O] \,x 
1857: \end{array}
1858: \]
1859: In other words, the canonical octonionic line bundle over $\OP^1$
1860: generates Bott periodicity!
1861: 
1862: There is much more to say about this fact and how it relates to Bott
1863: periodicity for Clifford algebras, but alas, this would take us too far
1864: afield.  We recommend that the interested reader turn to some
1865: introductory texts on K-theory, for example the one by Dale Husemoller
1866: \cite{Husemoller}.  Unfortunately, all the books I know downplay the
1867: role of the octonions.  To spot it, one must bear in mind the relation 
1868: between the octonions and Clifford algebras, discussed in Section
1869: \ref{clifford} above.
1870:  
1871: \subsection{$\OP^1$ and Lorentzian Geometry}  \label{lorentz} 
1872:  
1873: In Section \ref{OP1} we sketched a systematic approach to projective 
1874: lines over the normed division algebras.  The most famous example is the 
1875: Riemann sphere, $\CP^1$.  As emphasized by Penrose \cite{PR}, this space 
1876: has a fascinating connection to Lorentzian geometry --- or in other words, 
1877: special relativity.  All conformal transformations of the Riemann 
1878: sphere come from fractional linear transformations   
1879: \[        z \mapsto {az + b\over cz + d}, \qquad \qquad a,b,c,d \in \C.  \] 
1880: It is easy to see that the group of such transformations is isomorphic 
1881: to $\PSL(2,\C)$: $2 \times 2$ complex matrices with determinant 1,  
1882: modulo scalar multiples of the identity.  Less obviously, it is also 
1883: isomorphic to the Lorentz group $\SO_0(3,1)$: the identity component of 
1884: the group of linear transformations of $\R^4$ that preserve the 
1885: Minkowski metric 
1886: \[     x \cdot y = x_1 y_1 + x_2 y_2 + x_3 y_3 - x_4 y_4 .\] 
1887: This fact has a nice explanation in terms of the `heavenly sphere'.  
1888: Mathematically, this is the 2-sphere consisting of all lines of the form 
1889: $\{\alpha x\}$ where $x \in \R^4$ has $x \cdot x = 0$.  In special 
1890: relativity such lines represent light rays, so the heavenly sphere is 
1891: the sphere on which the stars appear to lie when you look at the night 
1892: sky.  This sphere inherits a conformal structure from the Minkowski 
1893: metric on $\R^4$.  This allows us to identify the heavenly sphere with 
1894: $\CP^1$, and it implies that the Lorentz group acts as conformal 
1895: transformations of $\CP^1$.  In concrete terms, what this means is that 
1896: if you shoot past the earth at nearly the speed of light, the 
1897: constellations in the sky will appear distorted, but all {\it angles} 
1898: will be preserved. 
1899:  
1900: In fact, these results are not special to the complex case: the same
1901: ideas work for the other normed division algebras as well!  The algebras
1902: $\R, \C, \H$ and $\O$ are related to Lorentzian geometry in 3, 4, 6, and
1903: 10 dimensions, respectively \cite{MD,MS,MS2,Schray,Sudbery}.  Even
1904: better, a full explanation of this fact brings out new relationships
1905: between the normed division algebras and spinors.  In what follows we
1906: explain how this works for all 4 normed division algebras, with special
1907: attention to the peculiarities of the octonionic case.
1908:  
1909: To set the stage, we first recall the most mysterious of the four 
1910: infinite series of Jordan algebras listed at the beginning of Section 
1911: \ref{proj}: the spin factors.  We described these quite concretely, but 
1912: a more abstract approach brings out their kinship to Clifford algebras.  
1913: Given an $n$-dimensional real inner product space $V$, let the {\bf spin 
1914: factor} $\J(V)$ be the Jordan algebra freely generated by $V$ modulo
1915: relations  
1916: \[     v^2 = \|v\|^2  .\]  
1917: Polarizing and applying the commutative law, we obtain   
1918: \[   v\circ w = \langle v, w \rangle, \]  
1919: so $\J(V)$ is isomorphic to $V \oplus \R$ with the product  
1920: \[  (v,\alpha) \circ (w, \beta) =   
1921: (\alpha w + \beta v, \langle v,w\rangle + \alpha \beta). \] 
1922:  
1923: Though Jordan algebras were invented to study quantum mechanics, the 
1924: spin factors are also deeply related to special relativity.  We can 
1925: think of $\J(V) \iso V \oplus \R$ as {\bf Minkowksi spacetime}, with 
1926: $V$ as space and $\R$ as time.  The reason is that $\J(V)$ is naturally
1927: equipped with a symmetric bilinear form of signature $(n,1)$, the {\bf
1928: Minkowski metric}: 
1929: \[    (v,\alpha)\cdot (w,\beta) = \langle v,w\rangle - \alpha \beta. \] 
1930: The group of linear transformations preserving the Minkowski metric is 
1931: called $\OO(n,1)$, and the identity component of this is called the {\bf 
1932: Lorentz group}, $\SO_0(n,1)$.   We define the {\bf lightcone} ${\rm 
1933: C}(V)$ to consist of all nonzero $x \in \J(V)$ with $x \cdot x = 0$.  A 
1934: 1-dimensional subspace of $\J(V)$ spanned by an element of the 
1935: lightcone is called a {\bf light ray}, and the space of all light rays is 
1936: called the {\bf heavenly sphere} ${\rm S}(V)$.  We can identify the 
1937: heavenly sphere with the unit sphere in $V$, since every light ray is 
1938: spanned by an element of the form $(v,1)$ where $v \in V$ has norm one. 
1939: Here is a picture of the lightcone and the heavenly sphere when $V$ is 
1940: 2-dimensional: 
1941:  
1942: \medskip 
1943: \centerline{\epsfysize=1.5in\epsfbox{heavenly.eps}}   
1944: \label{heavenly}   
1945: \medskip 
1946:  
1947: When $V$ is at least 2-dimensional, we can build a projective space from 
1948: the Jordan algebra $\J(V)$.  The result is none other than the heavenly 
1949: sphere!   To see this, note that aside from the elements 0 and 1, all 
1950: projections in $\J(V)$ are of the form $p = \textstyle{1\over 2}(v, 1)$ 
1951: where $v \in V$ has norm one.  These are the points of our projective 
1952: space, but as we have seen, they also correspond to points of the 
1953: heavenly sphere.  Our projective space has just one line, corresponding 
1954: to the projection $1 \in \J(V)$.  We can visualize this line as the 
1955: heavenly sphere itself.  
1956:  
1957: What does all this have to do with normed division algebras?   To answer 
1958: this, let $\K$ be a normed division algebra of dimension $n$.  Then  
1959: the Jordan algebra $h_2(\K)$ is secretly a spin factor!  There is an  
1960: isomorphism  
1961: \[   \phi \maps \h_2(\K) \to J(\K \oplus \R) \iso \K \oplus \R \oplus \R \] 
1962: given by 
1963: \be  \phi \left( \begin{array}{cc}     \alpha + \beta & x     \\  
1964:                                    x^\ast & \alpha - \beta \\  
1965: \end{array} \right) = (x, \beta, \alpha) , 
1966: \qquad \qquad x \in \K, \; \alpha, \beta \in \R .  \label{h2} \ee
1967: Furthermore, the determinant of matrices in $\h_2(\K)$ is well-defined  
1968: even when $\K$ is noncommutative or nonassociative: 
1969: \[  \det \left( \begin{array}{cc}  \alpha + \beta & x     \\  
1970:                                    x^\ast & \alpha - \beta \\  
1971: \end{array} \right) = \alpha^2 - \beta^2 - \|x\|^2 ,  \] 
1972: and clearly we have  
1973: \[      \det(a) = -\phi(a) \cdot \phi(a) \] 
1974: for all $a \in \h_2(\K)$.    
1975:  
1976: These facts have a number of nice consequences.  First of all, since the 
1977: Jordan algebras $\J(\K \oplus \R)$ and $\h_2(\K)$ are isomorphic, so are 
1978: their associated projective spaces.   We have seen that the former space 
1979: is the heavenly sphere ${\rm S}(\K \oplus \R)$, and that the latter is 
1980: $\KP^1$.  It follows that 
1981: \[    \KP^1 \iso {\rm S}(\K \oplus \R) . \] 
1982: This gives another proof of something we already saw in Section 
1983: \ref{OP1}: $\KP^1$ is an $n$-sphere.  But it shows more.   The Lorentz 
1984: group $\SO_0(n+1,1)$ has an obvious action on the heavenly sphere, and 
1985: the usual conformal structure on the sphere is invariant under this 
1986: action.  Using the above isomorphism we can transfer this group action 
1987: and invariant conformal structure to $\KP^1$ in a natural way. 
1988:  
1989: Secondly, it follows that the determinant-preserving linear 
1990: transformations of $\h_2(\K)$ form a group isomorphic to $\OO(n+1,1)$.  
1991: How can we find some transformations of this sort?  If $\K = \R$, this 
1992: is easy: when $g \in \SL(2,\R)$ and $x \in \h_2(\R)$, we again have 
1993: $gxg^* \in \h_2(\R)$, and   
1994: \[     \det(gxg^*) = \det(x). \]  
1995: This gives a homomorphism from $\SL(2,\R)$ to ${\rm O}(2,1)$.  This 
1996: homomorphism is two--to--one, since both $g = 1$ and $g = -1$ act 
1997: trivially, and it maps $\SL(2,\R)$ onto the identity component of ${\rm 
1998: O}(2,1)$.  It follows that $\SL(2,\R)$ is a double cover of  
1999: $\SO_0(2,1)$.  The exact same construction works for $\K = \C$, so 
2000: $\SL(2,\C)$ is a double cover of $\SO_0(3,1)$.    
2001:  
2002: For the other two normed division algebras the above calculation
2003: involving determinants breaks down, and it even becomes tricky to define
2004: the group $\SL(2,\K)$, so we start by working at the Lie algebra level.
2005: We say a $m \times m$ matrix with entries in the normed division algebra
2006: $\K$ is {\bf traceless} if the sum of its diagonal entries is zero.  Any
2007: such traceless matrix acts as a real--linear operator on $\K^m$.  When
2008: $\K$ is commutative and associative, the space of operators coming from
2009: $m \times m$ traceless matrices with entries in $\K$ is closed under
2010: commutators, but otherwise it is not, so we define $\Sl(m,\K)$ to be the
2011: Lie algebra of operators on $\K^m$ {\it generated} by operators of this
2012: form.  This Lie algebra in turn generates a Lie group of real-linear
2013: operators on $\K^m$, which we call $\SL(m,\K)$.  Note that
2014: multiplication in this group is given by composition of real-linear
2015: operators, which is associative even for $\K = \O$.
2016:  
2017: The Lie algebra $\Sl(m,\K)$ comes born with a representation:  
2018: its {\bf fundamental representation} as real-linear operators on $\K^m$,  
2019: given by 
2020: \[   a \maps x \mapsto ax ,           \qquad \qquad x \in \K^m \] 
2021: whenever $a \in \Sl(m,\K)$ actually corresponds to a traceless $m \times 
2022: m$ matrix with entries in $\K$.  Tensoring the fundamental representation  
2023: with its dual, we get a representation of $\Sl(m,\K)$ on the space 
2024: of matrices $\K[m]$, given by 
2025: \[   a \maps x \mapsto ax + xa^*,       \qquad \qquad x \in \K[m]   \] 
2026: whenever $a$ is a traceless matrix with entries in $\K$.  Since $ax + 
2027: xa^*$ is hermitian whenever $x$ is, this representation restricts to a 
2028: representation of $\Sl(m,\K)$ on $\h_m(\K)$.   This in turn can be 
2029: exponentiated to obtain a representation of the group $\SL(m,\K)$ on 
2030: $\h_m(\K)$. 
2031:  
2032: Now let us return to the case $m = 2$.  One can prove that the 
2033: representation of $\SL(2,\K)$ on $\h_2(\K)$ is determinant-preserving 
2034: simply by checking that 
2035: \[    {d \over dt} \det(x + t(ax + xa^*))\, \Bigr|_{t = 0} = 0 \] 
2036: when $x$ lies in $\h_2(\K)$ and $a \in \K[2]$ is traceless.   Here the 
2037: crucial thing is to make sure that the calculation is not spoiled by 
2038: noncommutativity or nonassociativity.  It follows that we have a  
2039: homomorphism 
2040: \[     \alpha_\K \maps \SL(2,\K) \to \SO_0(n+1,1)  \] 
2041: One can check that this is onto, and that its kernel consists of the 
2042: matrices $\pm 1$.  Thus if we define 
2043: \[   \PSL(2,\K) = \SL(2,\K) / \{\pm 1\} ,  \] 
2044: we get isomorphisms 
2045: \[     
2046: \begin{array}{ccl}       
2047:           \PSL(2,\R)& \iso &\SO_0(2,1)   \\ 
2048:           \PSL(2,\C)& \iso &\SO_0(3,1)   \\   
2049:           \PSL(2,\H)& \iso &\SO_0(6,1)   \\    
2050:           \PSL(2,\O)& \iso &\SO_0(9,1) .   
2051: \end{array} 
2052: \] 
2053: Putting this together with our earlier observations, it follows that 
2054: $\PSL(2,\K)$ acts as conformal transformations of $\KP^1$. 
2055:  
2056: We conclude with some words about how all this relates to spinors.  The 
2057: machinery of Clifford algebras and spinors extends effortlessly from the 
2058: case of inner product spaces to vector spaces equipped with an 
2059: indefinite metric.  In particular, the Lorentz group $\SO_0(n+1,1)$ has 
2060: a double cover called $\Spin(n+1,1)$, and this group has certain 
2061: representations called spinor representations.  When $n = 1,2,4$ or $8$, 
2062: we actually have 
2063: \[         \Spin(n+1,1) \iso \SL(2,\K)  \] 
2064: where $\K$ is the normed division algebra of dimension $n$.   The
2065: fundamental representation of $\SL(2,\K)$ on $\K^2$ is the left-handed
2066: spinor representation of $\Spin(n+1,1)$.  Its dual is the right-handed
2067: spinor representation.  Moreover, the interaction between vectors and
2068: spinors that serves as the basis of supersymmetric theories of physics
2069: in spacetimes of dimension 3, 4, 6 and 10 is just the action of
2070: $\h_2(\K)$ on $\K^2$ by matrix multiplication.  In a 
2071: Feynman diagram, this is represented as follows:
2072: 
2073: \medskip
2074: \hskip 25em \raise  1ex \hbox{$\K^2$}
2075: 
2076: \centerline{\raise7.5ex \hbox{$\h_2(\K)$} \kern -.5em
2077: \epsfysize=1.0in\epsfbox{feynman.eps}}   
2078: 
2079: \hskip 25em $\K^2$
2080: \label{feynman2}   
2081: 
2082: In the case $\K = \C$, Penrose \cite{PR} has described a nice trick for
2083: getting points on the heavenly sphere from spinors.
2084: In fact, it also works for other normed division algebras: 
2085: if $(x,y) \in \K^2$ is nonzero, the hermitian matrix 
2086: \[   
2087: \left( \begin{array}{c}  x \\ y \end{array} \right)  
2088: \left( \begin{array}{cc} \! x^\ast  &  y^\ast \! \end{array} \right)  
2089: =  
2090: \left( \begin{array}{cc}  
2091:                          x x^\ast  & x y^\ast    \\  
2092:                          y x^\ast  & y y^\ast    
2093: \end{array} \right)   
2094: \] 
2095: is nonzero but has determinant zero, so it defines a point on the
2096: heavenly sphere.  If we restrict to spinors of norm one, this trick
2097: reduces to the Hopf map.  Moreover, it clarifies the curious double role
2098: of $\KP^1$ as both the heavenly sphere in special relativity and a space
2099: of propositions in the quantum logic associated to the Jordan algebra
2100: $\h_2(\K)$: any point on the heavenly sphere corresponds to a
2101: proposition specifying the state of a spinor!
2102: 
2103: \subsection{$\OP^2$ and the Exceptional Jordan Algebra}  \label{OP2} 
2104:  
2105: The octonions are fascinating in themselves, but the magic really starts
2106: when we use them to construct the exceptional Jordan algebra $\h_3(\O)$
2107: and its associated projective space, the octonionic projective plane. 
2108: The symmetry groups of these structures turn out to be exceptional Lie 
2109: groups, and triality gains an eerie pervasive influence over the
2110: proceedings, since an element of $\h_3(\O)$ consists of 3 octonions and
2111: 3 real numbers.  Using the relation between normed division algebras and
2112: trialities, we get an isomorphism 
2113: \be 
2114: \begin{array} {ccc} 
2115:   \h_3(\O) & \iso & \R^3 \oplus V_8 \oplus S_8^+ \oplus S_8^-   \\ 
2116: {} & {} & {}  \\
2117: \left( \begin{array}{ccc}  
2118:                          \alpha  &  z^*  & y^*    \\  
2119:                          z       & \beta & x      \\ 
2120:                          y       & x^*   & \gamma   
2121: \end{array} \right) & \mapsto & ((\alpha,\beta,\gamma),x,y,z)   
2122: \end{array} 
2123: \label{jordan.triality} \ee  
2124: where $\alpha,\beta,\gamma \in \R$ and $x,y,z \in \O$.   Examining the
2125: Jordan product in $\h_3(\O)$ then reveals a wonderful fact: while
2126: superficially this product is defined using the $\ast$-algebra structure
2127: of $\O$, it can actually be defined using only the natural maps 
2128: \[      V_8 \times S_8^+ \to S_8^- , \qquad
2129:         V_8 \times S_8^- \to S_8^+ , \qquad
2130:         S_8^+ \times S_8^- \to V_8  
2131: \]  
2132: together with the inner products on these 3 spaces.   All this
2133: information is contained in the normed triality 
2134: \[     t_8 \maps V_8 \times S_8^+ \times S_8^- \to \R ,\]
2135: so any automorphism of this triality gives a automorphism of $\h_3(\O)$.
2136: In Section \ref{triality} we saw that $\Aut(t_8) \iso \Spin(8)$.  With
2137: a little thought, it follows that 
2138: \[    \Spin(8) \subseteq \Aut(\h_3(\O))   .\] 
2139: 
2140: However, this picture of $\h_3(\O)$ in terms of 8-dimensional Euclidean
2141: geometry is just part of a bigger picture --- a picture set in
2142: 10-dimensional Minkowski spacetime!  If we regard $\h_2(\O)$ as sitting
2143: in the lower right-hand corner of $\h_3(\O)$, we get an isomorphism
2144: \be 
2145: \begin{array} {ccc} 
2146:   \h_3(\O) & \iso & \R \oplus \h_2(\O) \oplus \O^2   \\ 
2147: {} & {} & {}  \\
2148: \left( \begin{array}{cc}  
2149:                          \alpha &  \psi^*    \\  
2150:                          \psi   &  a      \\ 
2151: \end{array} \right) & \mapsto & (\alpha,a,\psi)   
2152: \end{array} 
2153: \label{jordan.10d} \ee  
2154: We saw in Section \ref{lorentz} that $a \in \h_2(\O)$ and $\psi \in
2155: \O^2$ can be identified with a vector and a spinor in 10-dimensional
2156: Minkowski spacetime, respectively.   Similarly, $\alpha$ is a scalar.  
2157: 
2158: This picture gives a representation of $\Spin(9,1)$ as linear
2159: transformations of $\h_3(\O)$.  Unfortunately, most of these
2160: transformations do not preserve the Jordan product on $\h_3(\O)$.  As we
2161: shall see, they only preserve a lesser structure on $\h_3(\O)$: the {\it
2162: determinant}.  However, the transformations coming from the subgroup
2163: $\Spin(9) \subset \Spin(9,1)$ do preserve the Jordan product.  We can
2164: see this as follows.  As a representation of $\Spin(9)$, $\h_2(\O)$
2165: splits into `space' and `time':
2166: \[       \h_2(\O) \iso V_9 \oplus \R  \]
2167: with the two pieces corresponding to the traceless elements of 
2168: $\h_2(\O)$ and the real multiples of the identity, respectively. 
2169: On the other hand, the spinor representation of $\so(9)$ splits
2170: as $S_8^+ \oplus S_8^-$ when we restrict it to $\so(8)$, so we 
2171: have 
2172: \[       \O^2 \iso S_9  .\]
2173: We thus obtain an isomorphism
2174: \be 
2175: \begin{array} {ccc} 
2176:   \h_3(\O) & \iso & \R^2 \oplus V_9 \oplus S_9   \\ 
2177: {} & {} & {}  \\
2178: \left( \begin{array}{cc}  
2179:                          \alpha &  \psi^*    \\  
2180:                          \psi   &  a + \beta     \\ 
2181: \end{array} \right) & \mapsto & ((\alpha,\beta),a,\psi)   
2182: \end{array} 
2183: \label{jordan.9d} \ee  
2184: where $a \in \h_2(\O)$ has vanishing trace and $\beta$ is a real
2185: multiple of the identity.  In these terms, one can easily check that the
2186: Jordan product in $\h_3(\O)$ is built from invariant operations on
2187: scalars, vectors and spinors in 9 dimensions.  It follows that
2188: \[    \Spin(9) \subseteq \Aut(\h_3(\O))   .\] 
2189: For more details on this, see Harvey's book \cite{Harvey}.
2190: 
2191: This does not exhaust all the symmetries of $\h_3(\O)$, since there are 
2192: other automorphisms coming from the permutation group on 3 letters, 
2193: which acts on $(\alpha,\beta,\gamma) \in \R^3$ and $(x,y,z) \in \O^3$ in 
2194: an obvious way.  Also, any matrix $g \in \OO(3)$ acts by conjugation as 
2195: an automorphism of $\h_3(\O)$; since the entries of $g$ are real, there 
2196: is no problem with nonassociativity here.   The group $\Spin(9)$ is 
2197: 36-dimensional, but the full automorphism group $\h_3(\O)$ is much  
2198: bigger: it is 52-dimensional.  As we explain in Section \ref{F4}, it 
2199: goes by the name of $\F_4$. 
2200: 
2201: However, we can already do something interesting with the automorphisms 
2202: we have: we can use them to diagonalize any element of $\h_3(\O)$.   To 
2203: see this, first note that the rotation group, and thus $\Spin(9)$, acts 
2204: transitively on the unit sphere in $V_9$.  This means we can use an  
2205: automorphism in our $\Spin(9)$ subgroup to bring any element of $\h_3(\O)$ 
2206: to the form 
2207: \[ 
2208: \left( \begin{array}{ccc}  
2209:                          \alpha  &  z^*  & y^*         \\  
2210:                          z       & \beta & x           \\ 
2211:                          y       & x^* & \gamma   \end{array} \right)  
2212: \]  
2213: where $x$ is real.   The next step is to apply an automorphism 
2214: that makes $y$ and $z$ real while leaving $x$ alone.  To do this, note 
2215: that the subgroup of $\Spin(9)$ fixing any nonzero vector in $V_9$ is 
2216: isomorphic to $\Spin(8)$.  When we restrict the representation $S_9$ to 
2217: this subgroup, it splits as $S_8^+ \oplus S_8^-$, and with some work 
2218: \cite{Harvey} one can show that $\Spin(8)$ acts on $S_8^+ \oplus S_8^- 
2219: \iso \O^2$ in such a way that any element $(y,z) \in \O^2$ can be 
2220: carried to an element with both components real.  The final step is to 
2221: take our element of $\h_3(\O)$ with all real entries and use an 
2222: automorphism to diagonalize it.  We can do this by conjugating it with a 
2223: suitable matrix in $\OO(3)$.   
2224:  
2225: To understand $\OP^2$, we need to understand projections in $\h_3(\O)$.  
2226: Here is where our ability to diagonalize matrices in $\h_3(\O)$ via 
2227: automorphisms comes in handy.  Up to automorphism, every projection in 
2228: $\h_3(\O)$ looks like one of these four: 
2229: \[  
2230: p_0 = 
2231: \left( \begin{array}{ccc}  
2232:                          0  &  0  & 0         \\  
2233:                          0  &  0  & 0           \\ 
2234:                          0  &  0  & 0   \end{array} \right) , \; \;
2235: p_1 =  
2236: \left( \begin{array}{ccc}  
2237:                          1  &  0  & 0         \\  
2238:                          0  &  0  & 0           \\ 
2239:                          0  &  0  & 0   \end{array} \right) , \; \;
2240: p_2 =  
2241: \left( \begin{array}{ccc}  
2242:                          1  &  0  & 0         \\  
2243:                          0  &  1  & 0           \\ 
2244:                          0  &  0  & 0   \end{array} \right) , \; \;
2245: p_3 =  
2246: \left( \begin{array}{ccc}  
2247:                          1  &  0  & 0         \\  
2248:                          0  &  1  & 0           \\ 
2249:                          0  &  0  & 1   \end{array} \right) .  \] 
2250: Now, the trace of a matrix in $\h_3(\O)$ is invariant under 
2251: automorphisms, because we can define it using only the Jordan algebra 
2252: structure: 
2253: \[        \tr(a) = {1\over 9} \tr(L_a)    , \qquad \qquad a \in \h_3(\O) \] 
2254: where $L_a$ is left multiplication by $a$.  It follows that the trace of
2255: any projection in $\h_3(\O)$ is 0,1,2, or 3.  Furthermore, the rank of
2256: any projection $p \in \h_3(\O)$ equals its trace.  To see this, first
2257: note that $\tr(p) \ge \rank(p)$, since $p < q$ implies $\tr(p) <
2258: \tr(q)$, and the trace goes up by integer steps.  Thus we only need
2259: show $\tr(p) \le \rank(p)$.  For this it suffices to consider the four
2260: projections shown above, as both trace and rank are invariant under
2261: automorphisms.  Since $p_0 < p_1 < p_2 < p_3$, it is clear that for
2262: these projections we indeed have $\tr(p) \le \rank(p)$.
2263: 
2264: It follows that the points of the octonionic projective plane are 
2265: projections with trace 1 in $\h_3(\O)$, while the lines are projections 
2266: with trace 2.  A calculation \cite{Harvey} shows that any projection 
2267: with trace 1 has the form 
2268: \be   p =  
2269: \left( \begin{array}{c}  x \\ y \\ z \end{array} \right)  
2270: \left( \begin{array}{ccc} \! x^\ast  &  y^\ast & z^\ast \! \end{array} \right)  
2271: =  
2272: \left( \begin{array}{ccc}  
2273:                          x x^*  & x y^*   & x z^* \\ 
2274:                          y x^*  & y y^*   & y z^* \\ 
2275:                          z x^*  & z y^*   & z z^*  
2276: \end{array} \right)   
2277: \label{projection} 
2278: \ee 
2279: where $(x,y,z) \in \O^3$ has 
2280: \[     (xy)z = x(yz), \qquad \qquad  \|x\|^2 + \|y\|^2 + \|z\|^2 = 1 .\] 
2281: On the other hand, any projection with trace 2 is of the form $1 - p$ 
2282: where $p$ has trace 1.  This sets up a one-to-one correspondence between 
2283: points and lines in the octonionic projective plane.  If we use this 
2284: correspondence to think of both as trace-1 projections, the point $p$ 
2285: lies on the line $p'$ if and only if $p < 1 - p'$.  Of course, $p < 1 - p'$ 
2286: iff $p' < 1 - p$.   The symmetry of this relation means the octonionic 
2287: projective plane is self-dual!  This is also true of the real, complex 
2288: and quaternionic projective planes.  In all cases, the operation that 
2289: switches points and lines corresponds in quantum logic to the `negation'
2290: of propositions \cite{Varadarajan}.
2291: 
2292: We use $\OP^2$ to stand for the set of points in the octonionic 
2293: projective plane.   Given any nonzero element $(x,y,z) \in \O^3$ with 
2294: $(xy)z = x(yz)$, we can normalize it and then use equation 
2295: (\ref{projection}) to obtain a point $[(x,y,z)] \in \OP^2$.  Copying the 
2296: strategy that worked for $\OP^1$, we can make $\OP^2$ into a smooth manifold  
2297: by covering it with three coordinate charts:
2298: \begin{itemize}
2299: \item one chart containing all points of the form $[(x,y,1)]$, 
2300: \item one chart containing all points of the form $[(x,1,z)]$, 
2301: \item one chart containing all points of the form $[(1,y,z)]$. 
2302: \end{itemize}
2303: Checking that this works is a simple calculation.  The only interesting
2304: part is to make sure that whenever the associative law might appear
2305: necessary, we can either use the alternativity of the octonions or the
2306: fact that only triples with $(xy)z = x(yz)$ give points $[(x,y,z)] \in
2307: \OP^2$.
2308:  
2309: We thus obtain the following picture of the octonionic projective plane.
2310: As a manifold, $\OP^2$ is 16-dimensional.  The lines in $\OP^2$ are
2311: copies of $\OP^1$, and thus 8-spheres.  For any two distinct points in
2312: $\OP^2$, there is a unique line on which they both lie.  For any two
2313: distinct lines, there is a unique point lying on both of them.  There is
2314: a `duality' transformation that maps points to lines and vice versa
2315: while preserving this incidence relation.  In particular, since the
2316: space of all points lying on any given line is a copy of $\OP^1$, so 
2317: is the space of all lines containing a given point!
2318: 
2319: To dig more deeply into the geometry of $\OP^2$ one needs another
2320: important structure on the exceptional Jordan algebra: the determinant. 
2321: We saw in Section \ref{lorentz} that despite noncommutativity and
2322: nonassociativity, the determinant of a matrix in $\h_2(\O)$ is a
2323: well-defined and useful concept.  The same holds for $\h_3(\O)$!   We
2324: can define the {\bf determinant} of a matrix in $\h_3(\O)$ by
2325: \[ 
2326: \det \left( \begin{array}{ccc}  
2327:                          \alpha  &  z^*  & y^*         \\  
2328:                          z       & \beta & x           \\ 
2329:                          y       & x^* & \gamma   \end{array} \right) = 
2330: \alpha \beta \gamma - (\alpha \|x\|^2 + \beta \|y\|^2 + \gamma \|z\|^2)
2331: + 2 \Re(xyz) .
2332: \]  
2333: We can express this in terms of the trace and product via
2334: \[   \det(a) = {1\over 3} \tr(a^3) - {1\over 2} \tr(a^2) \tr(a) + 
2335:                {1\over 6} {\tr(a)}^3  .\]
2336: This shows that the determinant is invariant under all automorphisms of
2337: $\h_3(\O)$.  However, the determinant is invariant under an even bigger
2338: group of linear transformations.  As we shall see in Section \ref{E6},
2339: this group is 78-dimensional: it is a noncompact real form of the
2340: exceptional Lie group $\E_6$.  This extra symmetry makes it worth
2341: seeing how much geometry we can do starting with just the determinant
2342: and the vector space structure of $\h_3(\O)$.  
2343: 
2344: The determinant is a cubic form on $\h_3(\O)$, so there is a unique
2345: symmetric trilinear form 
2346: \[  (\cdot,\cdot,\cdot) \maps \h_3(\O) \times \h_3(\O) \times \h_3(\O)
2347: \to \R \]
2348: such that
2349: \[      (a,a,a) = \det(a) .\]
2350: By dualizing this, we obtain the so-called {\bf cross product}
2351: \[  \times \maps \h_3(\O) \times \h_3(\O) \to \h_3(\O)^*. \]
2352: Explicitly, this is given by
2353: \[     (a \times b)(c) = (a,b,c) .\]
2354: Despite its name, this product is commutative.  
2355: 
2356: We have already seen that points of $\OP^2$ correspond to trace-1
2357: projections in $\h_3(\O)$.  Freudenthal \cite{Freudenthal} noticed that
2358: these are the same as elements $p \in \h_3(\O)$ with $\tr(p) = 1$ and $p
2359: \times p = 0$.    Even better, we can drop the equation $\tr(p) = 1$ as
2360: long as we promise to work with {\it equivalence classes} of nonzero 
2361: elements satisfying $p \times p = 0$, where two such elements are
2362: equivalent when one is a nonzero real multiple of the other.  Each
2363: such equivalence class $[p]$ corresponds to a unique point of $\OP^2$,
2364: and we get all the points this way.
2365: 
2366: Given two points $[p]$ and $[q]$, their cross product $p \times q$ is
2367: well-defined up to a nonzero real multiple.  This suggests that we define
2368: a `line' to be an equivalence class of elements $p \times q \in
2369: \h_3(\O)^*$, where again two such elements are deemed equivalent if one
2370: is a nonzero real multiple of the other.   Freudenthal showed that we
2371: get a projective plane isomorphic to $\OP^2$ if we take these as our
2372: definitions of points and lines and decree that the point $[p]$ lies on
2373: the line $[L]$ if and only if $L(p) = 0$.   Note that this equation
2374: makes sense even though $L$ and $p$ are only well-defined up to 
2375: nonzero real multiples.
2376: 
2377: One consequence of all this is that one can recover the structure of
2378: $\OP^2$ as a projective plane starting from just the determinant on
2379: $\h_3(\O)$: we did not need the Jordan algebra structure!  However, to
2380: get a `duality' map switching points and lines while preserving the
2381: incidence relation, we need a bit more: we need the nondegenerate
2382: pairing 
2383: \[        \langle a,b \rangle =  \tr(ab)    \]
2384: on $\h_3(\O)$.  This sets up an isomorphism 
2385: \[        \h_3(\O) \iso \h_3(\O)^* . \]
2386: This isomorphism turns out to map points to lines, and in fact, it sets
2387: up a one-to-one correspondence between points and lines.  We can use
2388: this correspondence to think of both points and lines in $\OP^2$ as
2389: equivalence classes of elements of $\h_3(\O)$.  In these terms, the
2390: point $p$ lies on the line $\ell$ iff $\langle \ell,p \rangle = 0$.  This
2391: relationship is symmetrical!  It follows that if we switch points and
2392: lines using this correspondence, the incidence relation is preserved.
2393: 
2394: We thus obtain a very pretty setup for working with $\OP^2$.  If we
2395: use the isomorphism between $\h_3(\O)$ and its dual to reinterpret
2396: the cross product as a map 
2397: \[  \times \maps \h_3(\O) \times \h_3(\O) \to \h_3(\O) ,\]
2398: then not only is the line through distinct points $[p]$ and $[q]$ given
2399: by $[p \times q]$, but also the point in which two distinct lines
2400: $[\ell]$ and $[m]$ meet is given by $[\ell \times m]$.  A 
2401: triple of points $[p], [q]$ and $[r]$ is collinear iff $(p,q,r) = 0$,
2402: and a triple of lines $[\ell]$, $[m]$, $[n]$ meets at a point iff
2403: $(\ell,m,n) = 0$.  In addition, there is a delightful bunch of
2404: identities relating the Jordan product, the determinant, the cross
2405: product and the inner product in $\h_3(\O)$.  
2406: 
2407: For more on octonionic geometry, the reader is urged to consult the
2408: original papers by Freudenthal
2409: \cite{Freudenthal4,Freudenthal,Freudenthal2,Freudenthal3}, Jacques Tits
2410: \cite{Tits,Tits2} and Tonny Springer
2411: \cite{Springer,Springer2,Springer3}.  The book by Helmut Salzmann {\it
2412: et al} is also good \cite{Salzmann}.  Unfortunately, we must now bid
2413: goodbye to this subject and begin our trip through the exceptional
2414: groups.  However, we shall return to study the symmetries of $\OP^2$ and
2415: the exceptional Jordan algebra in Sections \ref{F4} and \ref{E6}.
2416: 
2417: \section{Exceptional Lie Algebras}         \label{lie}   
2418:  
2419: On October 18th, 1887, Wilhelm Killing wrote a letter to Friedrich Engel
2420: saying that he had classified the simple Lie algebras.  In the next
2421: three years, this revolutionary work was published in a series of papers
2422: \cite{Killing}.  Besides what we now call the `classical' simple Lie
2423: algebras, he claimed to have found 6 `exceptional' ones --- new
2424: mathematical objects whose existence had never before been suspected. In
2425: fact he gave a rigorous construction of only the smallest of these. In
2426: his 1894 thesis, Cartan \cite{Cartan0} constructed all of them and
2427: noticed that the two 52-dimensional exceptional Lie algebras discovered
2428: by Killing were isomorphic, so that there are really only 5.
2429: 
2430: The Killing-Cartan classification of simple Lie algebras introduced much
2431: of the technology that is now covered in any introductory course on the
2432: subject, such as roots and weights.  In what follows we shall avoid this
2433: technology, since we wish instead to see the exceptional Lie algebras as
2434: octonionic relatives of the classical ones --- slightly eccentric relatives,
2435: but still having a close connection to {\it geometry}, in particular the
2436: Riemannian geometry of projective planes.  It is also for this reason
2437: that we shall focus on the compact real forms of the simple Lie algebras.
2438: 
2439: The classical simple Lie algebras can be organized in three infinite
2440: families:
2441: \[
2442: \begin{array}{lcl}
2443:  \so(n) &=&  \{ x \in \R[n] \colon \; x^* = -x, \; \tr(x) = 0\},   \\
2444:  \su(n) &=&  \{ x \in \C[n] \colon \; x^* = -x, \; \tr(x) = 0\},   \\
2445:  \symp(n) &=&  \{ x \in \H[n] \colon \; x^* = -x \}.   
2446: \end{array}
2447: \]
2448: The corresponding Lie groups are
2449: \[
2450: \begin{array}{lcl}
2451:  \SO(n) &=&  \{ x \in \R[n] \colon \; xx^* = 1, \; \det(x) = 1\},   \\
2452:  \SU(n) &=&  \{ x \in \C[n] \colon \; xx^* = 1, \; \det(x) = 1\},   \\
2453:  \Sp(n) &=&  \{ x \in \H[n] \colon \; xx^* = 1 \}.   
2454: \end{array}
2455: \]
2456: These arise naturally as symmetry groups of projective spaces over
2457: $\R$, $\C$, and $\H$, respectively.   More precisely, they arise as
2458: groups of {\bf isometries}: transformations that preserve a specified
2459: Riemannian metric.  Let us sketch how this works, as a warmup for the 
2460: exceptional groups.  
2461: 
2462: First consider the projective space $\RP^n$.  We can think of this as
2463: the unit sphere in $\R^{n+1}$ with antipodal points $x$ and $-x$
2464: identified.  It thus inherits a Riemannian metric from the sphere,
2465: and the obvious action of the rotation group $\OO(n+1)$ as isometries
2466: of the sphere yields an action of this group as isometries of $\RP^n$
2467: with this metric.  In fact, with this metric, the group of all isometries 
2468: of $\RP^n$ is just 
2469: \[       \Isom(\RP^n) \iso \OO(n+1)/\OO(1)  \]
2470: where $\OO(1) = \{\pm 1\}$ is the subgroup of $\OO(n+1)$ that acts trivially
2471: on $\RP^n$.  The Lie algebra of this isometry group is 
2472: \[       \isom(\RP^n) \iso \so(n+1) .\]
2473: 
2474: The case of $\CP^n$ is very similar.  We can think of this as the unit
2475: sphere in  $\C^{n+1}$ with points $x$ and $\alpha x$ identified whenever
2476: $\alpha$ is a unit complex number.  It thus inherits a Riemannian metric
2477: from this sphere, and the unitary group $\U(n+1)$ acts as isometries.  
2478: If we consider only the connected component of the isometry group and 
2479: ignore the orientation-reversing isometries coming from complex 
2480: conjugation, we have
2481: \[       \Isom_0(\CP^n) \iso \U(n+1)/\U(1)  \]
2482: where $\U(1)$ is the subgroup that acts trivially on $\CP^n$.  
2483: The Lie algebra of this isometry group is
2484: \[       \isom(\CP^n) \iso \su(n+1)  .\]
2485: 
2486: The case of $\HP^n$ is subtler, since we must take the noncommutativity
2487: of the quaternions into account.   We can think of $\HP^n$ as the unit
2488: sphere in $\H^{n+1}$ with points $x$ and $\alpha x$ identified whenever
2489: $\alpha$ is a unit quaternion, and as before, $\HP^n$ inherits a
2490: Riemannian metric.  The group $\Sp(n+1)$ acts as isometries of $\HP^n$,
2491: but this action comes from {\it right} multiplication, so 
2492: \[       \Isom(\HP^n) \iso \Sp(n+1)/\{ \pm 1 \},  \]
2493: since not $\Sp(1)$ but only its center $\{\pm 1\}$ acts trivially
2494: on $\HP^n$ by right multiplication.  At the Lie algebra level, this 
2495: gives
2496: \[       \isom(\HP^n) \iso \symp(n+1) . \]
2497: 
2498: For lovers of the octonions, it is tempting to try a similar
2499: construction starting with $\OP^2$.  While nonassociativity makes things
2500: a bit tricky, we show in Section \ref{F4} that it can in fact be done. 
2501: It turns out that $\Isom(\OP^2)$ is one of the exceptional Lie groups,
2502: namely $\F_4$.  Similarly, the exceptional Lie groups $\E_6$, $\E_7$
2503: and $\E_8$ are in a certain subtle sense the isometry groups of
2504: projective planes over the algebras $\C \tensor \O$, $\H \tensor \O$ and
2505: $\O \tensor \O$.  Together with $\F_4$, these groups can all be 
2506: defined by the so-called `magic square' construction, which makes use of
2507: much of the algebra we have described so far.  We explain three versions
2508: of this construction in Section \ref{magic}.  We then treat the groups
2509: $\E_6, \E_7$ and $\E_8$ individually in the following sections.  But
2510: first, we must introduce $\G_2$: the smallest of the exceptional Lie
2511: groups, and none other than the automorphism group of the octonions.
2512: 
2513: \subsection{$\G_2$}       \label{G2}   
2514:  
2515: In 1914, \'Elie Cartan noted that the smallest of the exceptional Lie 
2516: groups, $\G_2$, is the automorphism group of the octonions 
2517: \cite{Cartan}.  Its Lie algebra $\g_2$ is therefore $\Der(\O)$, the 
2518: derivations of the octonions.  Let us take these facts as definitions of 
2519: $\G_2$ and its Lie algebra, and work out some of the consequences.   
2520:    
2521: What are automorphisms of the octonions like?  One way to analyze this
2522: involves subalgebras of the octonions.  Any octonion $e_1$ whose square
2523: is $-1$ generates a subalgebra of $\O$ isomorphic to $\C$. If we then
2524: pick any octonion $e_2$ with square equal to $-1$ that anticommutes with
2525: $e_1$, the elements $e_1,e_2$ generate a subalgebra isomorphic to $\H$.
2526: Finally, if we pick any octonion $e_3$ with square equal to $-1$ that
2527: anticommutes with $e_1,e_2,$ and $e_1e_2$, the elements $e_1,e_2,e_3$
2528: generate all of $\O$.  We call such a triple of octonions a {\bf basic
2529: triple}.  Given any basic triple, there exists a unique way to define
2530: $e_4, \dots , e_7$ so that the whole multiplication table in Section
2531: \ref{constructing} holds.  In fact, this follows from the remarks on the
2532: Cayley--Dickson construction at the end of Section \ref{clifford}.
2533:    
2534: It follows that given any two basic triples, there exists a unique   
2535: automorphism of $\O$ mapping the first to the second.  Conversely, it is
2536: obvious that any automorphism maps basic triples to basic triples. 
2537: This gives a nice description of the group $\G_2$, as follows.   
2538:     
2539: Fix a basic triple $e_1,e_2,e_3$.   There is a unique automorphism   
2540: of the octonions mapping this to any other basic triple, say   
2541: $e'_1,e'_2,e'_3$.  Now our description of basic triples so far has   
2542: been purely algebraic, but we can also view them more geometrically   
2543: as follows: a basic triple is any triple of unit imaginary   
2544: octonions (i.e.\ imaginary octonions of norm one) such that each is   
2545: orthogonal to the algebra generated by the other two.  This means that   
2546: our automorphism can map $e_1$ to any point $e'_1$ on the 6-sphere of   
2547: unit imaginary octonions, then map $e_2$ to any point $e'_2$ on the   
2548: 5-sphere of unit imaginary octonions that are orthogonal to $e'_1$, and   
2549: then map $e_3$ to any point $e'_3$ on the 3-sphere of unit imaginary   
2550: octonions that are orthogonal to $e'_1,e'_2$ and $e'_1 e'_2$.  It follows   
2551: that      
2552: \[    \dim \G_2 = \dim S^6 + \dim S^5 + \dim S^3 = 14  .\]    
2553:    
2554: The triality description of the octonions in Section \ref{triality}   
2555: gives another picture of $\G_2$.  First, recall that $\Spin(8)$ is the   
2556: automorphism group of the triality $t_8 \maps V_8 \times S^+_8 \times  
2557: S^-_8 \to \R$.  To construct the octonions from this triality we need to  
2558: pick unit vectors in any two of these spaces, so we can think of $\G_2$  
2559: as the subgroup of $\Spin(8)$ fixing unit vectors in $V_8$ and $S^+_8$.  
2560: The subgroup of $\Spin(8)$ fixing a unit vector in $V_8$ is just   
2561: $\Spin(7)$, and when we restrict the representation $S^+_8$ to   
2562: $\Spin(7)$, we get the spinor representation $S_7$.  Thus $\G_2$ is the   
2563: subgroup of $\Spin(7)$ fixing a unit vector in $S_7$.  Since $\Spin(7)$   
2564: acts transitively on the unit sphere $S^7$ in this spinor representation
2565: \cite{Adams}, we have 
2566: \[          \Spin(7)/\G_2 = S^7 .  \]   
2567: It follows that   
2568: \[   \dim \G_2 = \dim (\Spin(7)) - \dim S^7 = 21 - 7 = 14  .\]   
2569:    
2570: This picture becomes a bit more vivid if we remember that after choosing
2571: unit vectors in $V_8$ and $S^+_8$, we can identify both these
2572: representations with the octonions, with both unit vectors corresponding
2573: to $1 \in \O$.  Thus what we are really saying is this: the subgroup of
2574: $\Spin(8)$ that fixes $1$ in the vector representation on $\O$ is
2575: $\Spin(7)$; the subgroup that fixes $1$ in both the vector and
2576: right-handed spinor representations is $\G_2$.  This subgroup also fixes
2577: the element $1$ in the left-handed spinor representation of $\Spin(8)$
2578: on $\O$.
2579:    
2580: Now, using the vector representation of $\Spin(8)$ on $\O$, we    
2581: get homomorphisms   
2582: \[       \G_2 \hookrightarrow \Spin(8) \to \SO(\O)  \]   
2583: where $\SO(\O) \iso \SO(8)$ is the rotation group of the octonions,   
2584: viewed as a real vector space with the inner product $\langle x,y\rangle   
2585: = \Re(x^\ast y)$.   The map from $\Spin(8)$ to $\SO(\O)$ is two-to-one,   
2586: but when we restrict it to $\G_2$ we get a one-to-one map   
2587: \[       \G_2 \hookrightarrow \SO(\O) . \]   
2588:    
2589: At the Lie algebra level, this construction gives an inclusion   
2590: \[      \g_2 \hookrightarrow \so(\O)  \]   
2591: where $\so(\O) \iso \so(8)$ is the Lie algebra of skew-adjoint   
2592: real-linear transformations of the octonions.   Since $\g_2$ is   
2593: 14-dimensional and $\so(\O)$ is 28-dimensional, it is nice to see   
2594: exactly where the extra 14 dimensions come from.  In fact, they come   
2595: from two copies of $\Im(\O)$, the 7-dimensional space consisting of all   
2596: imaginary octonions.     
2597:    
2598: More precisely, we have:   
2599: \be  \so(\O) = \g_2 \oplus L_{\Im(\O)} \oplus R_{\Im(\O)} \label{so(O)} \ee   
2600: (a direct sum of vector spaces, not Lie algebras), where $L_{\Im(\O)}$    
2601: is the space of linear transformations of $\O$ given by left multiplication   
2602: by imaginary octonions and $R_{\Im(\O)}$ is the space of linear   
2603: transformations of $\O$ given by right multiplication by imaginary   
2604: octonions \cite{Schafer}.   To see this, we first check that left   
2605: multiplication by an imaginary octonion is skew-adjoint.  Using   
2606: polarization, it suffices to note that   
2607: \[           \langle x,ax \rangle = \Re(x^*(ax)) = \Re((x^*a)x) =   
2608: \Re((a^*x)^*x) = -\Re((ax)^* x) = -\langle ax,x \rangle \]   
2609: for all $a \in \Im(\O)$ and $x \in \O$.  Note that this calculation only   
2610: uses the alternative law, not the associative law, since $x, x^\ast$   
2611: and $a$ all lie in the algebra generated by the two elements $a$ and    
2612: $\Im(x)$.  A similar argument shows that right multiplication by   
2613: an imaginary octonion is skew-adjoint.  It follows that    
2614: $\g_2$, $L_{\Im(\O)}$ and $R_{\Im(\O)}$ all naturally lie in $\so(8)$.  
2615: Next, with some easy calculations we can check that 
2616: \[     L_{\Im(\O)} \cap R_{\Im(\O)} = \{0\} \]
2617: and 
2618: \[    \g_2 \cap (L_{\Im(\O)} + R_{\Im(\O)}) = \{0\} .\]
2619: Using the fact that the dimensions of the 3 pieces adds to 28,
2620: equation (\ref{so(O)}) follows.
2621:    
2622: We have seen that $\G_2$ sits inside $\SO(8)$, but we can do better: it   
2623: actually sits inside $\SO(7)$.  After all, every automorphism of the   
2624: octonions fixes the identity, and thus preserves the space of octonions   
2625: orthogonal to the identity.  This space is just $\Im(\O)$, so   
2626: we have an inclusion   
2627: \[       \G_2 \hookrightarrow \SO(\Im(\O))  \]   
2628: where $\SO(\Im(\O)) \iso \SO(7)$ is the rotation group of the imaginary   
2629: octonions.  At the Lie algebra level this gives an inclusion   
2630: \[        \g_2 \hookrightarrow \so(\Im(\O))  . \]   
2631:    
2632: Since $\g_2$ is 14-dimensional and $\so(\Im(\O))$ is 21-dimensional, it   
2633: is nice to see where the 7 extra dimensions come from.  Examining   
2634: equation (\ref{so(O)}), it is clear that these extra dimensions must   
2635: come from the transformations in $ L_{\Im(\O)} \oplus R_{\Im(\O)}$ that   
2636: annihilate the identity $1 \in \O$.  The transformations that do this are   
2637: precisely those of the form   
2638: \[           \ad_a = L_a - R_a \]   
2639: for $a \in \Im(\O)$.   We thus have    
2640: \be  \so(\Im(\O)) \iso \Der(\O) \oplus \ad_{\Im(\O)}  \label{so(Im(O))}  \ee   
2641: where $\ad_{\Im(\O)}$ is the 7-dimensional space of such transformations.   
2642:    
2643: We may summarize some of the above results as follows:   
2644:    
2645: \begin{thm} \et \label{g2-description} The compact real form   
2646: of the Lie algebra $\g_2$ is given by   
2647: \[     \g_2 = \Der(\O) \subset \so(\Im(\O)) \subset \so(\O)  \]   
2648: and we have    
2649: \ban      \so(\Im(\O)) &=& \Der(\O) \oplus \ad_{\Im(\O)}    \\   
2650:           \so(\O) &=& \Der(\O) \oplus L_{\Im(\O)} \oplus R_{\Im(\O)}    
2651: \ean   
2652: where the Lie brackets in $\so(\Im(\O))$ and $\so(\O)$ are built   
2653: from natural bilinear operations on the summands.     
2654: \end{thm}   
2655:    
2656: As we have seen, $\G_2$ has a 7-dimensional representation $\Im(\O)$.   
2657: In fact, this is the smallest nontrivial representation of $\G_2$,
2658: so it is worth understanding in as many ways as possible.  The space
2659: $\Im(\O)$ has at least three natural structures that are preserved by
2660: the transformations in $\G_2$.  These give more descriptions of  
2661: $\G_2$ as a symmetry group, and they also shed some new light on the   
2662: octonions.  The first two of the structures we describe are analogous to
2663: more familiar ones that exist on the 3-dimensional space of imaginary 
2664: quaternions, $\Im(\H)$.  The third makes explicit use of the   
2665: nonassociativity of the octonions.   
2666:    
2667: First, both $\Im(\H)$ and $\Im(\O)$ are closed under the commutator. In    
2668: the case of $\Im(\H)$, the commutator divided by 2 is the familiar {\bf   
2669: cross product} in 3 dimensions:   
2670: \[            a \times b = {1\over 2}[a,b]  .\]   
2671: We can make the same definition for $\Im(\O)$, obtaining a 7-dimensional   
2672: analog of the cross product.   For both $\Im(\H)$ and $\Im(\O)$ the   
2673: cross product is bilinear and anticommutative.   The cross product   
2674: makes $\Im(\H)$ into a Lie algebra, but not $\Im(\O)$.  For both   
2675: $\Im(\H)$ and $\Im(\O)$, the cross product has two nice    
2676: geometrical properties.  On the one hand, its norm is determined by the    
2677: formula   
2678: \[    \|a \times b\|^2 + \langle a,b\rangle^2 = \|a\|^2 \, \|b\|^2 , \]   
2679: or equivalently,    
2680: \[   \|a \times b\| = | {\sin \theta}| \, \|a\| \, \|b\| ,  \]   
2681: where $\theta$ is the angle between $a$ and $b$.  On the other hand, $a   
2682: \times b$ is orthogonal to $a$ and $b$.  Both these properties follow   
2683: from easy calculations.  For $\Im(\H)$, these two properties are enough   
2684: to determine $x \times y$ up to a sign.  For $\Im(\O)$ they are not ---   
2685: but they become so if we also use the fact that $x \times y$ lies inside   
2686: a copy of $\Im(\H)$ that contains $x$ and $y$.    
2687:    
2688: It is clear that the group of all real-linear transformations of   
2689: $\Im(\H)$ preserving the cross product is just $\SO(3)$, which is also   
2690: the automorphism group of the quaternions.  One can similarly show that   
2691: the group of real-linear transformations of $\Im(\O)$ preserving the   
2692: cross product is exactly $\G_2$.  To see this, start by noting that any   
2693: element of $\G_2$ preserves the cross product on $\Im(\O)$, since the   
2694: cross product is defined using octonion multiplication.  To show that   
2695: conversely any transformation preserving the cross product lies in   
2696: $\G_2$, it suffices to express the multiplication of imaginary octonions    
2697: in terms of their cross product.  Using this identity:   
2698: \[  a \times b = ab + \langle a, b\rangle   , \]   
2699: it actually suffices to express the inner product on $\Im(\O)$ in terms   
2700: of the cross product.  Here the following identity does the job:    
2701: \be  \langle a,b\rangle =   
2702: -\textstyle{1\over 6}\tr(a \times (b \times \cdot\,))    
2703: \label{inner} \ee   
2704: where the right-hand side refers to the trace of the map   
2705: \[       a \times (b \times \cdot\,) \maps \Im(\O) \to \Im(\O)  .\]   
2706:    
2707: Second, both $\Im(\H)$ and $\Im(\O)$ are equipped with a natural 3-form,   
2708: or in other words, an alternating trilinear functional.  This is given   
2709: by   
2710: \[        \phi(x,y,z) = \langle x,yz \rangle  .\]   
2711: In the case of $\Im(\H)$ this is just the usual volume form, and the   
2712: group of real-linear transformations preserving it is $\SL(3, \R)$.   In   
2713: the case of $\Im(\O)$, the real-linear transformations preserving $\phi$   
2714: are exactly those in the group $\G_2$.  A proof of this by Robert  
2715: Bryant can be found in Reese Harvey's book \cite{Harvey}.   The 3-form   
2716: $\phi$ is important in the theory of `Joyce manifolds' \cite{Joyce},  
2717: which are 7-dimensional Riemannian manifolds with holonomy group equal  
2718: to $\G_2$.  
2719:    
2720: Third, both $\Im(\H)$ and $\Im(\O)$ are closed under the associator.    
2721: For $\Im(\H)$ this is boring, since the associator vanishes.  On  the  
2722: other hand, for $\Im(\O)$ the associator is interesting.  In fact, it  
2723: follows from results of Harvey \cite{Harvey} that a real-linear  
2724: transformation $T \maps \Im(\O) \to \Im(\O)$ preserves the associator  
2725: if and only if $\pm T$ lies in $\G_2$.  Thus the symmetry group of the
2726: associator is slightly bigger than $\G_2$: it is $\G_2 \times \Z_2$.  
2727:    
2728: Now we must make an embarrassing admission: these three structures on  
2729: $\Im(\O)$ are all almost the same thing!   Starting with the cross
2730: product   
2731: \[    \times \maps \Im(\O) \times \Im(\O) \to \Im(\O)  \]   
2732: we can recover the usual inner product on $\Im(\O)$ by equation   
2733: (\ref{inner}).  This inner product allows us to dualize the cross
2734: product and obtain a trilinear functional, which is, up to a constant,   
2735: just the 3-form    
2736: \[     \phi \maps \Im(\O) \times \Im(\O) \times \Im(\O) \to \R .\]   
2737: The cross product also determines an orientation on $\Im(\O)$ (we leave   
2738: this as an exercise for the reader).  This allows us to take the Hodge   
2739: dual of $\phi$, obtaining a 4-form $\psi$, i.e.\ an alternating   
2740: tetralinear functional    
2741: \[     \psi \maps \Im(\O) \times \Im(\O) \times \Im(\O) \times \Im(\O)   
2742: \to \R .\]   
2743: Dualizing yet again, this gives a ternary operation which, up   
2744: to a constant multiple, is the associator:   
2745: \[  [\cdot, \cdot, \cdot] \maps \Im(\O) \times \Im(\O) \times \Im(\O)    
2746: \to \Im(\O) .\]   
2747:    
2748: We conclude this section with a handy explicit formula for all the   
2749: derivations of the octonions.    In an associative algebra $A$, any   
2750: element $x$ defines an {\bf inner derivation} $\ad_x \maps A \to A$ by   
2751: \[      \ad_x (a) = [x,a]  \]   
2752: where the bracket denotes the commutator $xa - ax$.   
2753: In a nonassociative algebra, this formula usually does not define   
2754: a derivation.  However, if $A$ is alternative, any pair of elements   
2755: $x,y \in A$ define a derivation $D_{x,y} \maps A \to A$ by   
2756: \be       D_{x,y} a = [[x,y],a] - 3[x,y,a]          \label{D}  \ee   
2757: where $[a,b,x]$ denotes the associator $(ab)x - a(bx)$.   Moreover, 
2758: when $A$ is a normed division algebra, every derivation is a linear 
2759: combination of derivations of this form.  Unfortunately, proving these 
2760: facts seems to require some brutal calculations \cite{Schafer}.  
2761:    
2762: \subsection{$\F_4$}   \label{F4}   
2763: 
2764: The second smallest of the exceptional Lie groups is the 52-dimensional
2765: group $\F_4$.  The geometric meaning of this group became clear in a
2766: number of nearly simultaneous papers by various mathematicians.  In
2767: 1949, Jordan constructed the octonionic projective plane using
2768: projections in $\h_3(\O)$.   One year later, Armand Borel \cite{Borel}
2769: noted that $\F_4$ is the isometry group of a 16-dimensional projective
2770: plane.  In fact, this plane is none other than than $\OP^2$.   Also
2771: in 1950, Claude Chevalley and Richard Schafer \cite{CS} showed that
2772: $\F_4$ is the automorphism group of $\h_3(\O)$.   In 1951, Freudenthal
2773: \cite{Freudenthal4} embarked upon a long series of papers in which he
2774: described not only $\F_4$ but also the other exceptional Lie groups
2775: using octonionic projective geometry.  To survey these developments, one
2776: still cannot do better than to read his classic 1964 paper on Lie groups
2777: and the foundations of geometry \cite{Freudenthal3}.
2778: 
2779: Let us take Chevalley and Schafer's result as the definition of $\F_4$:
2780: \[     \F_4 = \Aut(\h_3(\O))  .\]
2781: Its Lie algebra is thus
2782: \[   \f_4 = \Der(\h_3(\O)).     \label{f4.1}  \]
2783: As we saw in Section \ref{OP2}, points of $\OP^2$ correspond to trace-1
2784: projections in the exceptional Jordan algebra.  It follows that $\F_4$
2785: acts as transformations of $\OP^2$.  In fact, we can equip $\OP^2$ with
2786: a Riemannian metric for which $\F_4$ is the isometry group. To get a
2787: sense of how this works, let us describe $\OP^2$ as a quotient space of
2788: $\F_4$.   
2789: 
2790: In Section \ref{OP2} we saw that the exceptional Jordan algebra can
2791: be built using natural operations on the scalar, vector and spinor
2792: representations of $\Spin(9)$.  This implies that $\Spin(9)$ is a
2793: subgroup of $\F_4$.  Equation (\ref{jordan.9d}) makes it clear that
2794: $\Spin(9)$ is precisely the subgroup fixing the element
2795:  \[   \left( \begin{array}{ccc} 1 & 0 & 0 \\  
2796:                                 0 & 0 & 0 \\  
2797:                                 0 & 0 & 0 \\  
2798: \end{array} \right).  \]  
2799: Since this element is a trace-one projection, it corresponds to a point
2800: of $\OP^2$.  We have already seen that $\F_4$ acts transitively on 
2801: $\OP^2$.  It follows that
2802: \be     \OP^2 \iso \F_4 /\Spin(9)  . \label{F4/Spin(9)} \ee
2803:   
2804: This fact has various nice spinoffs.  First, it gives an easy way to  
2805: compute the dimension of $\F_4$:  
2806: \[    \dim(\F_4) = \dim(\Spin(9)) + \dim(\OP^2) =  
2807: 36 + 16 = 52.\]  
2808: Second, since $\F_4$ is compact, we can take any Riemannian metric on $\OP^2$ 
2809: and average it with respect to the action of this group.  The isometry
2810: group of the resulting metric will automatically include $\F_4$ as a
2811: subgroup.  With more work \cite{Besse}, one can show that actually
2812: \[     \F_4 = \Isom(\OP^2) \]
2813: and thus
2814: \[   \f_4 = \isom(\OP^2).     \label{f4.2}  \]
2815: 
2816: Equation (\ref{F4/Spin(9)}) also implies that the tangent space of our
2817: chosen point in $\OP^2$ is isomorphic to $\f_4/\so(9)$.  But we already
2818: know that this tangent space is just $\O^2$, or in other words, the
2819: spinor representation of $\so(9)$.  We thus have
2820: \be   \f_4 \iso \so(9) \oplus S_9    \label{f4.3}   \ee   
2821: as vector spaces, where $\so(9)$ is a Lie subalgebra.  The bracket in  
2822: $\f_4$ is built from the bracket in $\so(9)$, the action $\so(9) \tensor  
2823: S_9 \to S_9$, and the map $S_9 \tensor S_9 \to \so(9)$  obtained by  
2824: dualizing this action.  We can also rewrite this description
2825: of $\f_4$ in terms of the octonions, as follows:   
2826: \[   \f_4 \iso \so(\O \oplus \R) \oplus \O^2    \label{f4.4}   \]
2827:    
2828: This last formula suggests that we decompose $\f_4$ further using the   
2829: splitting of $\O \oplus \R$ into $\O$ and $\R$.   
2830: It is easily seen by looking at matrices that for all $n,m$ we have    
2831: \be     
2832: \so(n+m) \iso \so(n) \oplus \so(m) \, \oplus\, V_n \tensor V_m.    
2833: \label{so(n+m)} \ee   
2834: Moreover, when we restrict the representation  
2835: $S_9$ to $\so(8)$, it splits as a direct sum $S_8^+ \oplus S_8^-$.  
2836: Using these facts and equation (\ref{f4.3}), we see  
2837: \be      
2838: \f_4 \iso \so(8) \oplus V_8 \oplus S_8^+ \oplus S_8^-     
2839: \label{f4.5}   
2840: \ee   
2841: This formula emphasizes the close relation between $\f_4$ and triality:   
2842: the Lie bracket in $\f_4$ is completely built out of maps involving   
2843: $\so(8)$ and its three 8-dimensional irreducible representations!     
2844: We can rewrite this in a way that brings out the role of the octonions:   
2845: \[
2846: \f_4 \iso \so(\O) \oplus \O^3    
2847: \label{f4.6}    
2848: \]
2849:   
2850: While elegant, none of these descriptions of $\f_4$ gives a  convenient
2851: picture of all the derivations of the exceptional Jordan algebra.   In
2852: fact, there is a nice picture of this sort for $\h_3(\K)$ whenever $\K$
2853: is a normed division algebra.    One way to get a derivation of the
2854: Jordan algebra $\h_3(\K)$ is to take a derivation of $\K$  and let it
2855: act on each entry of the matrices in $\h_3(\K)$.  Another way uses
2856: elements of 
2857: \[    \sa_3(\K) = \{  x \in \K[3] \colon \; x^* = -x,\; \tr(x) = 0 \}  .\]
2858: Given $x \in \sa_3(\K)$, there is a derivation $\ad_x$ of $\h_3(\K)$ given 
2859: by
2860: \[          \ad_x (a) = [x,a]   .\]   
2861: In fact \cite{BS}, every derivation of $\h_3(\K)$ can be uniquely 
2862: expressed as a linear combination of derivations of these two sorts,
2863: so we have 
2864: \be   \Der(\h_3(\K)) \iso \Der(\K) \oplus \sa_3(\K)   \label{der(h3K)} \ee
2865: as vector spaces.   In the case of the octonions, this decomposition 
2866: says that
2867: \[        \f_4 \iso \g_2 \oplus \sa_3(\O) .     \label{f4.7} \]
2868: 
2869: In equation (\ref{der(h3K)}), the subspace $\Der(\K)$ is always a Lie
2870: subalgebra, but $\sa_3(\K)$ is not unless $\K$ is commutative and
2871: associative --- in which case $\Der(\K)$ vanishes.  Nonetheless, there
2872: is a formula for the brackets in $\Der(\h_3(\K))$ which applies in every
2873: case \cite{OV}.  Given $D,D' \in \Der(\K)$ and $x,y \in \sa_3(\K)$, we
2874: have
2875: \be
2876: \begin{array}{lcl}
2877:             [D,D'] &=& DD' - D'D   \cr
2878:          [D,\ad_x] &=& \ad_{Dx}    \cr
2879:      [\ad_x,\ad_y] &=& \ad_{[x,y]_0} + 
2880: \displaystyle{{1\over 3} \sum_{i,j = 1}^3   D_{x_{ij},y_{ij}}  }
2881: \end{array} 
2882: \label{der(h3K)-bracket}
2883: \ee
2884: where $D$ acts on $x$ componentwise, $[x,y]_0$ is the trace-free   
2885: part of the commutator $[x,y]$, and $D_{x_{ij},y_{ij}}$ is the
2886: derivation of $\K$ defined using equation (\ref{D}).  
2887:    
2888: Summarizing these different descriptions of $\f_4$, we have:   
2889: \begin{thm} \et \label{f4-description}  The compact real form of    
2890: $\f_4$ is given by    
2891: \ban   \f_4 &\iso& \isom(\OP^2)  \\   
2892:             &\iso& \Der(\h_3(\O)) \\   
2893:             &\iso& \Der(\O) \oplus \sa_3(\O) \\   
2894:             &\iso& \so(\O \oplus \R) \oplus \O^2 \\   
2895:             &\iso& \so(\O) \oplus \O^3   
2896: \ean   
2897: where in each case the Lie bracket is built from    
2898: natural bilinear operations on the summands.     
2899: \end{thm}   
2900: 
2901: \subsection{The Magic Square}  \label{magic}   
2902:    
2903: Around 1956, Boris Rosenfeld \cite{Rosenfeld1} had the remarkable idea
2904: that just as $\F_4$ is the isometry group of the projective plane over
2905: the octonions, the exceptional Lie groups $\E_6$, $\E_7$ and $\E_8$ are
2906: the isometry groups of projective planes over the following three
2907: algebras, respectively:
2908: \begin{itemize}
2909: \item the {\bf bioctonions}, $\C \tensor \O$,
2910: \item the {\bf quateroctonions}, $\H \tensor \O$,
2911: \item the {\bf octooctonions}, $\O \tensor \O$.
2912: \end{itemize}
2913: There is definitely something right about this idea, because one would
2914: expect these projective planes to have dimensions 32, 64, and 128, and
2915: there indeed do exist compact Riemannian manifolds with these
2916: dimensions having $\E_6$, $\E_7$ and $\E_8$ as their isometry groups. 
2917: The problem is that the bioctonions, quateroctonions and and
2918: octooctonions are not division algebras, so it is a nontrivial matter to
2919: define projective planes over them!
2920: 
2921: The situation is not so bad for the bioctonions: $\h_3(\C \tensor \O)$
2922: is a simple Jordan algebra, though not a formally real one, and one can
2923: use this to define $(\C \tensor \O)\P^2$ in a manner modeled after one
2924: of the constructions of $\OP^2$.  Rosenfeld claimed that a similar
2925: construction worked for the quateroctonions and octooctonions, but this
2926: appears to be false.  Among other problems, $\h_3(\H \tensor \O)$ and
2927: $\h_3(\O \tensor \O)$  do not become Jordan algebras under the product
2928: $a \circ b =  {1\over 2}(ab + ba)$.  Scattered throughout the literature
2929: \cite{Besse,Freudenthal3,Freudenthal5} one can find frustrated comments
2930: about the lack of a really nice construction of $(\H \tensor \O)\P^2$
2931: and $(\O \tensor \O)\P^2$.  One problem is that these spaces do {\it
2932: not} satisfy the usual axioms for a projective plane.   Tits addressed
2933: this problem in his theory of `buildings', which allows one to construct
2934: a geometry having any desired algebraic group as symmetries
2935: \cite{Tits4}.  But alas, it still seems that the quickest way to get our
2936: hands on the quateroctonionic and octooctonionic `projective planes' is
2937: by {\it starting} with the Lie groups $\E_7$ and $\E_8$ and then taking
2938: quotients by suitable subgroups.
2939: 
2940: In short, more work must be done before we can claim to fully understand
2941: the geometrical meaning of the Lie groups $\E_6, \E_7$ and $\E_8$. 
2942: Luckily, Rosenfeld's ideas can be used to motivate a nice construction
2943: of their Lie algebras.  This goes by the name of the `magic square'.  
2944: Tits \cite{Tits3} and Freudenthal \cite{Freudenthal2} found two very
2945: different versions of this construction in about 1958, but we shall
2946: start by presenting a simplified version published by E.\ B.\ Vinberg
2947: \cite{Vinberg} in 1966.
2948: 
2949: First consider the projective plane $\KP^2$ when $\K$ is a normed
2950: division algebra $\K$.  The points of this plane are the rank-1
2951: projections in the Jordan algebra $\h_3(\K)$, and this plane admits a
2952: Riemannian metric such that
2953: \[   \isom(\KP^2) \iso \Der(\h_3(\K)). \] 
2954: Moreover, we have seen in equation (\ref{der(h3K)}) that 
2955: \[   \Der(\h_3(\K)) \iso \Der(\K) \oplus \sa_3(\K) . \]
2956: Combined with Rosenfeld's observations, these facts might lead one to 
2957: hope that whenever we have a pair of normed division algebras $\K$ and 
2958: $\K'$, there is a Riemannian manifold $(\K \tensor \K')\P^2$ with
2959: \[   \isom((\K\tensor \K')\P^2) \iso 
2960: \Der(\K) \oplus \Der(\K') \oplus \sa_3(\K \tensor \K')  \]
2961: where for any $\ast$-algebra $A$ we define
2962: \[
2963: \begin{array}{lcl}
2964:     \sh_n(A) & =& \{ x \in A[n] \colon \; x^* = x, \; \tr(x) = 0\}   \\
2965:     \sa_n(A) & =& \{ x \in A[n] \colon \; x^* = -x, \; \tr(x) = 0\}.  
2966: \end{array}
2967: \]
2968: 
2969: This motivated Vinberg's definition of the {\bf magic square} Lie
2970: algebras:
2971: \[
2972:     \M(\K,\K') = \Der(\K) \oplus \Der(\K') \oplus \sa_3(\K \tensor \K').
2973: \]
2974: Now, when $\K \tensor \K'$ is commutative and associative, $\sa_3(\K
2975: \tensor \K')$ is a Lie algebra with the commutator as its Lie bracket,
2976: but in the really interesting cases it is not.  Thus to make 
2977: $\M(\K,\K')$ into a Lie algebra we must give it a rather subtle bracket.
2978: We have already seen the special case $\K' = \R$ in equation
2979: (\ref{der(h3K)-bracket}).  In general, the Lie bracket in $\M(\K,\K')$
2980: is given as follows:
2981: \begin{enumerate}
2982: \item $\Der(\K)$ and $\Der(\K')$ are commuting Lie subalgebras of $\M(\K,\K')$.
2983: \item The bracket of $D \in \Der(K) \oplus \Der(\K')$ 
2984: with $x \in \sa_3(\K \tensor \K')$ is given by applying 
2985: $D$ to every entry of the matrix $x$, using the natural action of
2986: $\Der(K) \oplus \Der(\K')$ as derivations of $\K \tensor \K'$.
2987: \item Given $X,Y \in \sa_3(\K \tensor \K')$, 
2988: \[      [X,Y] = [X,Y]_0 + 
2989: {1\over 3}\displaystyle{\sum_{i,j = 1}^3   D_{X_{ij},Y_{ij}}  } .\]
2990: Here $[X,Y]_0$ is the traceless part of the $3 \times 3$ matrix $[X,Y]$,
2991: and given $x,y \in \K \tensor \K'$ we define $D_{x,y} \in 
2992: \Der(\K) \oplus \Der(\K')$ in the following way: $D_{x,y}$ is
2993: real-bilinear in $x$ and $y$, and 
2994: \[     D_{a \tensor a',b \tensor b'} = \langle a',b' \rangle D_{a,b} +
2995:                                        \langle a,b \rangle D_{a',b'} \]
2996: where $a,b \in \K$, $a',b' \in \K'$, and $D_{a,b},D_{a',b'}$ are defined
2997: as in equation (\ref{D}).
2998: \end{enumerate}
2999: With this construction we magically obtain the following square of Lie 
3000: algebras:
3001: \vskip 1em   
3002: {\vbox{   
3003: \begin{center}   
3004: \renewcommand{\baselinestretch}{1.3}  
3005: {\small
3006: \begin{tabular}{|c|c|c|c|c|}                    \hline   
3007:            & $\K' = \R$ & $\K' = \C$ & $\K' = \H$ & $\K' = \O$     \\  \hline
3008: $\K = \R$ &  $\so(3)$ & $\su(3)$  & $\symp(3)$  & $\f_4$           \\  \hline
3009: $\K = \C$ &  $\su(3)$ & $\su(3) \oplus \su(3)$ & $\su(6)$ & $\e_6$ \\  \hline
3010: $\K = \H$ &  $\symp(3)$ & $\su(6)$  & $\so(12)$  & $\e_7$          \\  \hline
3011: $\K = \O$ &  $\f_4$   & $\e_6$    & $\e_7$     & $\e_8$            \\  \hline
3012: \end{tabular}} \vskip 1em   
3013: Table 5 --- Magic Square Lie Algebras $\M(\K,\K')$ 
3014: \end{center}   
3015: }   }
3016: \vskip 0.5em   
3017: \noindent
3018: We will mainly be interested in the last row (or column), which is the
3019: one involving the octonions.  In this case we can take the magic square
3020: construction as {\it defining} the Lie algebras $\f_4$, $\e_6$, $\e_7$
3021: and $\e_8$.  This definition turns out to be consistent with our earlier
3022: definition of $\f_4$.
3023: 
3024: Starting from Vinberg's definition of the magic square Lie algebras, we
3025: can easily recover Tits' original definition.  To do so, we need two
3026: facts.  First, 
3027: \[  
3028: \sa_3(\K \otimes \K') 
3029: \iso \sa_3(\K') \, \oplus \, (\Im(\K) \! \tensor \! \sh_3(\K')).
3030: \]
3031: This is easily seen by direct examination of the relevant matrices.
3032: Second, 
3033: \[    \Der(\h_3(\K)) \iso \Der(\K) \oplus \sa_3(\K)  \]
3034: as vector spaces.  This is just equation (\ref{der(h3K)}).  Starting with
3035: Vinberg's definition and applying these two facts, we obtain
3036: \[
3037: \begin{array}{lcl}
3038: \M(\K,\K') &=& \Der(\K) \oplus \Der(\K') \oplus \sa_3(\K \tensor \K') \\
3039: &\iso& \Der(\K) \oplus \Der(\K') 
3040: \oplus \sa_3(\K') \, \oplus \, (\Im(\K) \tensor \sh_3(\K')) \\
3041: &\iso& 
3042: \Der(\K) \oplus \Der(\h_3(\K')) \, \oplus \, (\Im(\K) \tensor \sh_3(\K')) .
3043: \end{array}
3044: \]
3045: The last line is Tits' definition of the magic square Lie algebras.
3046: Unlike Vinberg's, it is not manifestly symmetrical in $\K$ and $\K'$.
3047: This unhappy feature is somewhat made up for by the fact that $\Der(\K)
3048: \oplus \Der(\h_3(\K'))$ is a nice big Lie subalgebra.  This subalgebra
3049: acts on $\Im(\K) \tensor \sh_3(\K')$ in an obvious way, using the fact
3050: that any derivation of $\K$ maps $\Im(\K)$ to itself, and any derivation
3051: of $\h_3(\K')$ maps $\sh_3(\K')$ to itself.  However, the bracket of two
3052: elements of $(\Im(\K) \tensor \sh_3(\K'))$ is a bit of a mess.
3053: 
3054: Yet another description of the magic square was recently given by
3055: Barton and Sudbery \cite{BS}.  This one emphasizes the role of 
3056: trialities.  Let $\Tri(\K)$ be the Lie algebra of the group $\Aut(t)$,
3057: where $t$ is the normed triality giving the normed division algebra
3058: $\K$.  From equation (\ref{Aut(t)}) we have
3059: \be
3060: \begin{array}{lcl} 
3061:     \Tri(\R) &\iso& \{0\}    \\
3062:     \Tri(\C) &\iso& \u(1)^2  \\
3063:     \Tri(\H) &\iso& \symp(1)^3 \\
3064:     \Tri(\O) &\iso& \so(8)  .
3065: \end{array} 
3066: \label{tri}
3067: \ee
3068: To express the magic square in terms of these Lie algebras, we need
3069: three facts.  First, it is easy to see that 
3070: \[     \sh_3(\K) \iso \K^3 \oplus \R^2 .\]
3071: Second, Barton and Sudbery show that as vector spaces,
3072: \[    \Der(\h_3(\K)) \iso \Tri(\K) \oplus \K^3 . \]
3073: This follows in a case--by--case way from equation
3074: (\ref{tri}), but they give a unified proof that covers all cases.
3075: Third, they show that as vector spaces,
3076: \[     \Tri(\K) \iso \Der(\K) \oplus \Im(\K)^2 . \]
3077: Now starting with Tits' definition of the magic square,
3078: applying the first two facts, regrouping terms, and applying 
3079: the third fact, we obtain Barton and Sudbery's version of the 
3080: magic square:
3081: \[
3082: \begin{array}{lcl}
3083: \M(\K,\K') 
3084: &\iso& \Der(\K) \oplus \Der(\h_3(\K')) \, \oplus \, 
3085: (\Im(\K) \tensor \sh_3(\K'))   \\
3086: &\iso & \Der(\K) \oplus \Tri(\K') \oplus \K'^3 \oplus 
3087: \Im(\K) \! \tensor \! (\K'^3 \oplus \R^2)  \\
3088: &\iso & \Der(\K) \oplus \Im(\K)^2 \oplus \Tri(\K') \oplus (\K \tensor \K')^3 \\
3089: &\iso &  \Tri(\K) \oplus \Tri(\K') \oplus (\K \tensor \K')^3  .
3090: \end{array}
3091: \]
3092: 
3093: In the next three sections we use all these different versions of the
3094: magic square to give lots of octonionic descriptions of $\e_6$, $\e_7$
3095: and $\e_8$.  To save space, we usually omit the formulas for the Lie
3096: bracket in these descriptions. However, the patient reader can
3097: reconstruct these with the help of Barton and Sudbery's paper, which 
3098: is packed with useful formulas.
3099: 
3100: As we continue our tour through the exceptional Lie algebras, we shall
3101: make contact with Adams' work \cite{Adams} constructing
3102: $\f_4,\e_6,\e_7,$ and $\e_8$ with the help of spinors and rotation group
3103: Lie algebras:
3104: \[
3105: \begin{array}{lcl} 
3106: \f_4 &\iso& \so(9) \oplus S_9    \\
3107: \e_6 &\iso& \so(10) \oplus \u(1) \oplus S_{10} \\
3108: \e_7 &\iso& \so(12) \oplus \symp(1) \oplus S_{12}^+ \\
3109: \e_8 &\iso& \so(16) \oplus S_{16}^+ 
3110: \end{array}
3111: \]
3112: as vector spaces.  Note that the numbers 9, 10, 12 and 16 are 8 more
3113: than the dimensions of $\R,\C,\H$ and $\O$.  As usual, this is no
3114: coincidence!  In terms of the octonions, Bott periodicity implies that
3115: \[              S_{n+8} \iso S_n \tensor \O^2  .\]
3116: This gives the following description of spinors in dimensions $\le 16$:
3117:   
3118: \medskip  
3119: {\vbox{   
3120: \begin{center}   
3121: \renewcommand{\baselinestretch}{1.3}  
3122: {\small   
3123: \begin{tabular}{|l|l|}                                  \hline   
3124: $S_1 = \R$           &  $S_9 = \O^2$                       \\ \hline   
3125: $S_2 = \C$           &  $S_{10} = (\C \tensor \O)^2$       \\ \hline  
3126: $S_3 = \H$           &  $S_{11} = (\H \tensor \O)^2$       \\ \hline     
3127: $S_4^\pm = \H$       &  $S_{12}^\pm = (\H \tensor \O)^2$   \\ \hline     
3128: $S_5 = \H^2$         &  $S_{13} = (\H^2 \tensor \O)^2$      \\ \hline     
3129: $S_6 = \C^4$         &  $S_{14} = (\C^4 \tensor \O)^2$      \\ \hline     
3130: $S_7 = \O$           &  $S_{15} = (\O \tensor \O)^2$       \\ \hline     
3131: $S_8^\pm = \O$       &  $S_{16}^\pm = (\O \tensor \O)^2$   \\ \hline     
3132: \end{tabular}}  
3133: \vskip 1em 
3134: \centerline{Table 6 --- Spinor Representations Revisited}  
3135: \end{center}   
3136: }}   
3137: \medskip   
3138: 
3139: \noindent
3140: Since spinors in dimensions 1,2,4 and 8 are isomorphic to the division
3141: algebras $\R,\C,\H$ and $\O$, spinors in dimensions 8 higher are 
3142: isomorphic to the `planes' $\O^2, (\C \tensor \O)^2, (\H \tensor \O)^2$ 
3143: and $(\O \tensor \O)^2$ --- and are thus closely linked to $\f_4$, $\e_6$, 
3144: $\e_7$ and $\e_8$, thanks to the magic square.
3145: 
3146: \subsection{$\E_6$}   \label{E6}   
3147: 
3148: We begin with the 78-dimensional exceptional Lie group $\E_6$.     
3149: As we mentioned in Section \ref{OP2}, there is a nice description of a 
3150: certain noncompact real form of $\E_6$ as the group of collineations
3151: of $\OP^2$, or equivalently, the group of determinant-preserving 
3152: linear transformations of $\h_3(\O)$.  But before going into these,
3153: we consider the magic square constructions of the Lie algebra $\e_6$.
3154: Vinberg's construction gives 
3155: \[
3156: \e_6 = \Der(\O) \oplus \sa_3(\C \tensor \O)  .
3157: \label{e6.1}
3158: \]
3159: Tits' construction, which is asymmetrical, gives
3160: \[
3161: \e_6 \iso \Der(\h_3(\O)) \oplus \sh_3(\O)  
3162: \label{e6.2}
3163: \]
3164: and also
3165: \[
3166: \e_6 \iso \Der(\O) \oplus \Der(\h_3(\C)) \oplus    
3167: (\Im(\O) \!\tensor \! \sh_3(\C))  .
3168: \label{e6.3}
3169: \]
3170: The Barton-Sudbery construction gives
3171: \[
3172: \e_6 \iso \Tri(\O) \oplus \Tri(\C) \oplus (\C \tensor \O)^3 .
3173: \label{e6.4}
3174: \]
3175: We can use any of these to determine the dimension of $\e_6$.   For
3176: example, we have 
3177: \[ \dim(\e_6) = \dim(\Der(\h_3(\O))) + \dim(\sh_3(\O)) = 52 + 26 = 78. \]
3178: 
3179: Starting from the Barton-Sudbery construction and using the concrete
3180: descriptions of $\Tri(\O)$ and $\Tri(\C)$ from equation
3181: (\ref{tri}), we obtain
3182: \[
3183: \e_6 \iso \so(\O) \oplus \so(\C) \oplus \Im(\C) \oplus (\C \tensor \O)^3 
3184: \label{e6.5}
3185: \]
3186: Using equation (\ref{so(n+m)}), we may rewrite this as
3187: \[
3188: \e_6 \iso \so(\O \oplus \C) \oplus \Im(\C) \oplus (\C \tensor \O)^2
3189: \label{e6.6}
3190: \]
3191: and it turns out that the summand $\so(\O \oplus \C) \oplus \Im(\C)$ is 
3192: actually a Lie subalgebra of $\e_6$.  This result can also be found in 
3193: Adams' book \cite{Adams}, phrased as follows: 
3194: \[
3195: \e_6 \iso \so(10) \oplus \u(1) \oplus S_{10}
3196: \label{e6.7}
3197: \]
3198: In fact, he describes the bracket in $\e_6$ in terms of natural
3199: operations involving $\so(10)$ and its spinor representation $S_{10}$. 
3200: The funny-looking factor of $\u(1)$ comes from the fact that this spinor
3201: representation is complex.  The bracket of an element of $\u(1)$ and
3202: an element of $S_{10}$ is another element of $S_{10}$, defined using
3203: the obvious action of $\u(1)$ on this complex vector space.  
3204: 
3205: If we define $\E_6$ to be the simply connected group with Lie algebra
3206: $\e_6$, it follows from results of Adams that the subgroup generated by
3207: the Lie subalgebra $\so(10) \oplus \u(1)$ is isomorphic to  $(\Spin(10)
3208: \times \U(1))/\Z_4$.  This lets us define the {\bf bioctonionic
3209: projective plane} by
3210: \[     (\C \tensor \O)\P^2 = \E_6\, / \, ((\Spin(10) \times \U(1))/\Z_4) \]
3211: and conclude that the tangent space at any point of this manifold is
3212: isomorphic to $S_{10} \iso (\C \tensor \O)^2$.  
3213: 
3214: Since $\E_6$ is compact, we can put an $\E_6$-invariant Riemannian metric on the bioctonionic
3215: projective plane by averaging any metric with respect to the action
3216: of this group.  It turns out \cite{Besse} that the isometry group of this 
3217: metric is exactly $\E_6$, so we have
3218: \[        \E_6 \iso \Isom((\C \tensor \O)\P^2).
3219: \]
3220: It follows that
3221: \[
3222: \e_6 \iso \isom((\C \tensor \O)\P^2)  .
3223: \label{e6.8}
3224: \]
3225: 
3226: Summarizing, we have 6 octonionic descriptions of $\e_6$:   
3227: \begin{thm} \et \label{e6-description}  The compact real form of    
3228: $\e_6$ is given by    
3229: \ban   
3230:  \e_6 &\iso& \isom((\C \tensor \O)\P^2) \\   
3231:  &\iso& \Der(\O) \oplus \Der(\h_3(\C)) \oplus    
3232:         (\Im(\O) \! \tensor\!  \sh_3(\C)) \\   
3233:  &\iso& \Der(\h_3(\O)) \oplus \sh_3(\O)  \\   
3234:  &\iso& \Der(\O) \oplus \sa_3(\C \tensor \O) \\   
3235:  &\iso& \so(\O \oplus \C) \oplus \Im(\C) \oplus (\C \tensor \O)^2  \\   
3236:  &\iso& \so(\O) \oplus \so(\C) \oplus \Im(\C) \oplus (\C \tensor \O)^3   
3237: \ean   
3238: where in each case the Lie bracket of $\e_6$ is built from    
3239: natural bilinear operations on the summands.     
3240: \end{thm}   
3241: 
3242: The smallest nontrivial representations of $\E_6$ are 27-dimensional:
3243: in fact it has two inequivalent representations of this dimension, which
3244: are dual to one another.  Now, the exceptional Jordan algebra is also
3245: 27-dimensional, and in 1950 this clue led Chevalley and Schafer \cite{CS}
3246: to give a nice description of $\E_6$ as symmetries of this algebra.
3247: These symmetries do not preserve the product, but only the determinant.
3248: 
3249: More precisely, the group of determinant-preserving linear
3250: transformations of $\h_3(\O)$ turns out to be a noncompact real form of
3251: $\E_6$.  This real form is sometimes called $\E_{6(-26)}$, because its
3252: Killing form has signature $-26$.  To
3253: see this, note that any automorphism of $\h_3(\O)$ preserves the
3254: determinant, so we get an inclusion
3255: \[            \F_4 \hookrightarrow \E_{6(-26)}  .\] 
3256: This means that $\F_4$ is a compact subgroup of $\E_{6(-26)}$.  In fact
3257: it is a maximal compact subgroup, since if there were a larger one, we
3258: could average a Riemannian metric group on $\OP^2$ with respect to this
3259: group and get a metric with an isometry group larger than $\F_4$, but no
3260: such metric exists.  It follows that the Killing form on the Lie algebra
3261: $\e_{6(-26)}$ is negative definite on its 52-dimensional maximal compact
3262: Lie algebra, $\f_4$, and positive definite on the complementary
3263: 26-dimensional subspace, giving a signature of $26 - 52 = -26$.
3264: 
3265: We saw in Section \ref{OP2} that the projective plane structure of
3266: $\OP^2$ can be constructed starting only with the determinant function
3267: on the vector space $\h_3(\O)$.   It follows that $\E_{6(-26)}$
3268: acts as {\bf collineations} on $\OP^2$, that is, line-preserving
3269: transformations.  In fact, the group of collineations of $\OP^2$ is
3270: precisely $\E_{6(-26)}$:
3271: \[   \E_{6(-26)}  \iso     {\rm Coll}(\OP^2). \]
3272: Moreover, just as the group of isometries of $\OP^2$ fixing a specific
3273: point is a copy of $\Spin(9)$, the group of collineations fixing a
3274: specific point is $\Spin(9,1)$.  This fact follows with some work
3275: starting from equation (\ref{jordan.10d}), and it gives us a commutative
3276: square of inclusions:
3277: \[
3278: \begin{array}{ccl} 
3279:    \Spin(9) &\longrightarrow& \Isom(\OP^2) \iso \F_4 \\
3280:    \downarrow & &     \;\;\;\;\;  \downarrow  \\
3281:    \Spin(9,1)    & \longrightarrow & {\rm Coll}(\OP^2) \iso \E_{6(-26)} 
3282: \end{array}
3283: \]
3284: where the groups on the top are maximal compact subgroups of those on
3285: the bottom.  Thus in a very real sense, $\F_4$ is to 9-dimensional
3286: Euclidean geometry as $\E_{6(-26)}$ is to 10-dimensional Lorentzian
3287: geometry.
3288: 
3289: \subsection{$\E_7$}  \label{E7}   
3290: 
3291: Next we turn to the 133-dimensional exceptional Lie group $\E_7$.
3292: In 1954, Freudenthal \cite{Freudenthal2} described this group as
3293: the automorphism group of a 56-dimensional octonionic structure now
3294: called a `Freudenthal triple system'.  We sketch this idea
3295: below, but first we give some magic square constructions.
3296: Vinberg's version of the magic square gives
3297: \[
3298: \e_7 = \Der(\H) \oplus \Der(\O) \oplus \sa_3(\H \tensor \O)  .
3299: \label{e7.1}
3300: \]
3301: Tits' version gives
3302: \be
3303: \e_7 \iso \Der(\H) \oplus \Der(\h_3(\O)) \oplus    
3304: (\Im(\H) \!\tensor \! \sh_3(\O))  
3305: \label{e7.2}
3306: \ee
3307: and also
3308: \[
3309: \e_7 \iso \Der(\O) \oplus \Der(\h_3(\H)) \oplus    
3310: (\Im(\O) \!\tensor \! \sh_3(\H))  
3311: \label{e7.3}
3312: \]
3313: The Barton-Sudbery version gives
3314: \be
3315: \e_7 \iso \Tri(\O) \oplus \Tri(\H) \oplus (\H \tensor \O)^3 
3316: \label{e7.4}
3317: \ee
3318:    
3319: Starting from equation (\ref{e7.2}) and using the fact that
3320: $\Der(\H) \iso \Im(\H)$ is 3-dimensional, we obtain the elegant
3321: formula
3322: \[ 
3323: \e_7 \iso \Der(\h_3(\O)) \, \oplus \, \h_3(\O)^3  .
3324: \label{e7.5}
3325: \]
3326: This gives an illuminating way to compute the dimension of $\e_7$:
3327: \[ \dim(\e_7) = \dim(\Der(\h_3(\O))) + 3 \dim(\h_3(\O)) = 52 + 3 \cdot 27 = 
3328: 133 .\]
3329: Starting from equation (\ref{e7.4}) and using the concrete 
3330: descriptions of $\Tri(\H)$ and $\Tri(\O)$ from equation
3331: (\ref{tri}), we obtain
3332: \[
3333: \e_7 \iso \so(\O) \oplus \so(\H) \oplus \Im(\H) \oplus (\H \tensor \O)^3 
3334: \label{e7.6}
3335: \]
3336: Using equation (\ref{so(n+m)}), we may rewrite this as
3337: \[
3338: \e_7 \iso \so(\O \oplus \H) \oplus \Im(\H) \oplus (\H \tensor \O)^2.
3339: \label{e7.7}
3340: \]
3341: Though not obvious from what we have done, the direct summand $\so(\O
3342: \oplus \H) \oplus \Im(\H)$ here is really a Lie subalgebra of $\e_7$. In
3343: less octonionic language, this result can also be found in Adams' book
3344: \cite{Adams}:
3345: \[
3346: \e_7 \iso \so(12) \oplus \symp(1) \oplus S_{12}^+ 
3347: \label{e7.8}
3348: \]
3349: He describes the bracket in $\e_7$ in terms of natural operations
3350: involving $\so(12)$ and its spinor representation $S_{12}^+$.   The
3351: funny-looking factor of $\symp(1)$ comes from the fact that this
3352: representation is quaternionic.  The bracket of an element of $\symp(1)$ and
3353: an element of $S_{12}^+$ is the element of $S_{12}^+$ defined using
3354: the natural action of $\symp(1)$ on this space.
3355: 
3356: If we let $\E_7$ be the simply connected group with Lie 
3357: algebra $\e_7$, it follows from results of Adams \cite{Adams} that the
3358: subgroup generated by the Lie subalgebra $\so(12) \oplus \symp(1)$ is
3359: isomorphic to $(\Spin(12) \times \Sp(1))/\Z_2$.  
3360: This lets us define the {\bf quateroctonionic projective plane} by
3361: \[    (\H \tensor \O)\P^2 = \E_7\, / \,((\Spin(12) \times \Sp(1))/\Z_2) \]
3362: and conclude that the tangent space at any point of this manifold is
3363: isomorphic to $S_{12}^+ \iso (\H \tensor \O)^2$.   We can put 
3364: an $\E_7$-invariant Riemannian metric on this manifold by the technique
3365: of averaging over the group action.  It then turns out \cite{Besse} that
3366: \[        \E_7 \iso \Isom((\H \tensor \O)\P^2)
3367: \]
3368: and thus
3369: \[
3370: \e_7 \iso \isom((\H \tensor \O)\P^2)  .
3371: \label{e7.9}
3372: \]
3373: 
3374: Summarizing, we have the following 7 octonionic descriptions of $\e_7$:   
3375:    
3376: \begin{thm} \et \label{e7-description}  The compact real form of    
3377: $\e_7$ is given by    
3378: \ban   
3379: \e_7  &\iso& \isom((\H \tensor \O)\P^2)    \\   
3380: &\iso& \Der(\h_3(\O)) \oplus \h_3(\O)^3   \\   
3381: &\iso& \Der(\O) \oplus \Der(\h_3(\H)) \oplus    
3382:        (\Im(\O) \!\tensor \! \sh_3(\H)) \\   
3383: &\iso& \Der(\H) \oplus \Der(\h_3(\O)) \oplus    
3384:         (\Im(\H) \! \tensor\! \sh_3(\O)) \\   
3385: &\iso& \Der(\O) \oplus \Der(\H) \oplus \sa_3(\H \tensor \O) \\   
3386: &\iso& \so(\O \oplus \H) \oplus \Im(\H) \oplus (\H \tensor \O)^2 \\   
3387: &\iso& \so(\O) \oplus \so(\H) \oplus \Im(\H) \oplus (\H \tensor \O)^3  
3388: \ean   
3389: where in each case the Lie bracket of $\e_7$ is built from    
3390: natural bilinear operations on the summands.     
3391: \end{thm}   
3392: 
3393: Before the magic square was developed, Freudenthal \cite{Freudenthal2} 
3394: used another octonionic construction to study $\E_7$.   The smallest
3395: nontrivial representation of this group is 56-dimensional.  Freudenthal
3396: realized we can define a 56-dimensional space 
3397: \[ F = \{ \left( \begin{array}{cc}  
3398:                          \alpha  & x     \\  
3399:                           y      & \beta \\ 
3400: \end{array} \right) : \;
3401: x,y \in \h_3(\O) , \; \alpha , \beta \in \R \} 
3402: \] 
3403: and equip this space with a symplectic structure 
3404: \[     \omega \maps F \times F \to \R  \]
3405: and trilinear product 
3406: \[     \tau \maps F \times F \times F \to F \]
3407: such that the group of linear transformations preserving both these
3408: structures is a certain noncompact real form of $\E_7$, namely
3409: $\E_{7(-25)}$.  The symplectic structure and trilinear product on
3410: $F$ satisfy some relations, and algebraists have made these into the
3411: definition of a `Freudenthal triple system' \cite{Brown,Faulkner,Meyberg}.
3412: The geometrical significance of this rather complicated sort of structure 
3413: has recently been clarified by some physicists working on string theory.
3414: At the end of the previous section, we
3415: mentioned a relation between 9-dimensional Euclidean geometry and
3416: $\F_4$, and a corresponding relation between 10-dimensional Lorentzian
3417: geometry and $\E_{6(-26)}$.  Murat G\"unaydin \cite{Gunaydin} has 
3418: extended this to a relation between 10-dimensional {\sl conformal} 
3419: geometry and $\E_{7(-25)}$, and in work with Kilian Koepsell 
3420: and Hermann Nikolai \cite{GKN} has explicated how this is connected 
3421: to Freudenthal triple systems.  
3422: 
3423: \subsection{$\E_8$}   \label{E8}   
3424:    
3425: With 248 dimensions, $\E_8$ is the biggest of the exceptional Lie
3426: groups, and in some ways the most mysterious.  The easiest way to
3427: understand a group is to realize it as as symmetries of a structure one
3428: already understands.  Of all the simple Lie groups, $\E_8$ is the only
3429: one whose smallest nontrivial representation is the adjoint
3430: representation.  This means that in the context of linear algebra,
3431: $\E_8$ is most simply described as the group of symmetries of its own
3432: Lie algebra!  One way out of this vicious circle would be to describe
3433: $\E_8$ as isometries of a Riemannian manifold.  As already mentioned,
3434: $\E_8$ is the isometry group of a 128-dimensional manifold called $(\O
3435: \tensor \O)\P^2$.  But alas, nobody seems to know how to define $(\O
3436: \tensor \O)\P^2$ without first defining $\E_8$.  Thus this group remains
3437: a bit enigmatic.
3438: 
3439: At present, to get our hands on $\E_8$ we must start with its Lie
3440: algebra.   We can define this using any of the three equivalent magic
3441: square constructions explained in Section \ref{magic}.  Vinberg's
3442: construction gives
3443: \[
3444: \e_8 = \Der(\O) \oplus \Der(\O) \oplus \sa_3(\O \tensor \O)  .
3445: \label{e8.1}
3446: \]
3447: Tits' construction gives
3448: \[
3449: \e_8 \iso \Der(\O) \oplus \Der(\h_3(\O)) \oplus    
3450: (\Im(\O) \!\tensor \! \sh_3(\O)) .
3451: \label{e8.2}
3452: \]
3453: The Barton-Sudbery construction gives
3454: \be
3455: \begin{array}{lcl}
3456:  \e_8 &\iso& \Tri(\O) \oplus \Tri(\O) \oplus (\O \tensor \O)^3 \\
3457:       &\iso& \so(\O) \oplus \so(\O) \oplus (\O \tensor \O)^3 
3458: \end{array}
3459: \label{e8.3}
3460: \ee
3461: We can use any of these to count the dimension of $\e_8$; for example,
3462: the last one gives
3463: \[   \dim \e_8 = 28 + 28 + 3 \cdot 8^2 = 248.\]   
3464: 
3465: To emphasize the importance of triality, we can rewrite equation 
3466: (\ref{e8.3}) as:
3467: \be
3468: \e_8  \iso \so(8) \oplus \so(8) \oplus (V_8 \tensor V_8) \oplus
3469:              (S_8^+ \tensor S_8^+) \oplus (S_8^- \tensor S_8^-).
3470: \label{e8.4}
3471: \ee
3472: Here the Lie bracket is built from natural maps relating $\so(8)$
3473: and its three 8-dimensional irreducible representations.  In particular,
3474: $\so(8) \oplus \so(8)$ is a Lie subalgebra, and the first copy of
3475: $\so(8)$ acts on the first factor in $V_8 \tensor V_8$, $S_8^+ \tensor
3476: S_8^+$, and $S_8^- \tensor S_8^-$, while the second copy acts on the
3477: second factor in each of these.   The reader should compare this to
3478: the description of $\f_4$ in equation (\ref{f4.5}).
3479: 
3480: Now, equation (\ref{so(n+m)}) implies that
3481: \[     \so(16) \iso \so(8) \oplus \so(8) \oplus (V_8 \tensor V_8) .\]
3482: Together with equation (\ref{e8.4}), this suggests that $\e_8$ contains
3483: $\so(16)$ as a Lie subalgebra. In fact this is true!  Even better, if we
3484: restrict the right-handed spinor representation of $\so(16)$ to $\so(8)
3485: \oplus \so(8)$, it decomposes as
3486: \[    S^+_{16} \iso (S_8^+ \tensor S_8^+) \oplus (S_8^- \tensor S_8^-),\]
3487: so we obtain
3488: \be   \e_8 \iso \so(16) \oplus S^+_{16}     \label{e8.5} \ee   
3489: or in more octonionic language,
3490: \[   
3491: \e_8 \iso \so(\O \oplus \O) \oplus (\O \otimes \O)^2 
3492: \label{e8.6} \]   
3493: where we use $\so(V)$ to mean the Lie algebra of skew-adjoint real-linear
3494: transformations of the real inner product space $V$.  
3495: 
3496: The really remarkable thing about equation (\ref{e8.5}) is that the Lie
3497: bracket in $\e_8$ is entirely built from natural maps involving
3498: $\so(16)$ and $S^+_{16}$:  
3499: \[         \so(16) \tensor \so(16) \to \so(16) , \qquad   
3500:            \so(16) \tensor S^+_{16} \to S^+_{16}  , \qquad   
3501:            S^+_{16} \tensor S^+_{16} \to \so(16) .\]   
3502: The first of these is the Lie bracket in $\so(16)$, the second is the
3503: action of $\so(16)$ on its right-handed spinor representation, and the
3504: third is obtained from the second by duality, using the natural inner
3505: product on $\so(16)$ and $S^+_{16}$ to identify these spaces with their
3506: duals.   In fact, this is a very efficient way to {\it define} $\e_8$. 
3507: If we take this approach, we must verify the Jacobi identity:   
3508: \[       [[a,b],c] = [a,[b,c]] - [b,[a,c]]  .\]   
3509: When all three of $a,b,c$ lie in $\so(16)$ this is just the Jacobi   
3510: identity for $\so(16)$.   When two of them lie in $\so(16)$, it boils   
3511: down to fact that spinors indeed form a representation of $\so(16)$.   
3512: Thanks to duality, the same is true when just one lies in $\so(16)$.  It
3513: thus suffices to consider the case when $a,b,c$ all lie in $S_{16}^+$. 
3514: This is the only case that uses anything special about the number 16. 
3515: Unfortunately, at this point a brute-force calculation seems to be
3516: required.  For two approaches that minimize the pain, see the books by
3517: Adams \cite{Adams} and by Green, Schwarz and Witten \cite{GSW}.   It
3518: would be nice to find a more conceptual approach.
3519:    
3520: Starting from $\e_8$, we can define $\E_8$ to be the simply-connected
3521: Lie group with this Lie algebra.  As shown by Adams \cite{Adams}, the 
3522: subgroup of $\E_8$ generated by the Lie subalgebra $\so(16) \subset
3523: \e_8$ is $\Spin(16)/\Z_2$.  This lets us define the {\bf octooctonionic
3524: projective plane} by  
3525: \[            (\O \tensor \O)\P^2 = \E_8\,/\,(\Spin(16)/\Z_2) . \]  
3526: By equation (\ref{e8.5}), the tangent space at any point 
3527: of this manifold is isomorphic to $S_{16}^+ \iso (\O \tensor \O)^2$.
3528: This partially justifies calling it `octooctonionic projective plane',
3529: though it seems not to satisfy the usual axioms for a projective plane.
3530: 
3531: We can put an $\E_8$-invariant Riemannian metric on the octooctonionic
3532: projective plane by the technique of averaging over the group action.
3533: It then turns out \cite{Besse} that 
3534: \[        \E_8 \iso \Isom((\O \tensor \O)\P^2)   \]
3535: and thus 
3536: \[       \e_8 \iso \isom((\O \tensor \O)\P^2)  . \label{e8.7} \]
3537:    
3538: Summarizing, we have the following octonionic descriptions of 
3539: $\E_8$:
3540: \begin{thm} \et \label{e8-description}  The compact real form of    
3541: $\e_8$ is given by    
3542: \ban   
3543: \e_8 &\iso& \isom((\O \tensor \O)\P^2)    \\  
3544:      &\iso& \Der(\O) \oplus \Der(\h_3(\O)) \oplus    
3545:             (\Im(\O) \!\tensor\! \sh_3(\O)) \\   
3546:      &\iso&  \Der(\O) \oplus \Der(\O) \oplus \sa_3(\O \tensor \O) \\   
3547:      &\iso& \so(\O \oplus \O) \oplus (\O \tensor \O)^2 \\   
3548:      &\iso& \so(\O) \oplus \so(\O) \oplus (\O \tensor \O)^3     
3549: \ean   
3550: where in each case the Lie bracket on $\e_8$ is built from    
3551: natural bilinear operations on the summands.     
3552: \end{thm}   
3553: 
3554: \section{Conclusions} 
3555: 
3556: It should be clear by now that besides being a fascinating mathematical
3557: object in their own right, the octonions link together many important
3558: phenomena whose connections would otherwise be completely mysterious.  
3559: Indeed, the full story of these connections is deeper and more elaborate
3560: than I have been able to explain here!   It also includes:
3561: \begin{itemize} 
3562: \item Attempts to set up an octonionic analogue of the theory of 
3563: analytic functions (see \cite{GT} and the references therein). 
3564: \item The role of Jordan pairs, Jordan triple systems and 
3565: Freudenthal triple systems in the construction of exceptional Lie groups 
3566: \cite{Brown,Faulkner,FF,GKN,GT,McCrimmon,Meyberg}.
3567: \item Constructions of the $\E_8$ lattice and Leech lattice using
3568: integral octonions \cite{Coxeter,EG}.
3569: \item Tensor-categorical approaches to normed division algebras
3570: and the invariant of framed trivalent graphs coming from the
3571: quantum group associated to $\G_2$ \cite{Boos,Bremner,Kuperberg,Rost}.
3572: \item Octonionic constructions of vertex operator algebras \cite{FFrenkel}. 
3573: \item Octonionic constructions of the exceptional simple Lie superalgebras 
3574: \cite{Sudbery2}. 
3575: \item Octonionic constructions of symmetric spaces \cite{Besse}.
3576: \item Octonions and the geometry of the `squashed 7-spheres', that is, 
3577: the homogeneous spaces $\Spin(7)/\G_2$, $\Spin(6)/\SU(3)$, and 
3578: $\Spin(5)/\SU(2)$, all of which are diffeomorphic to $S^7$ with its 
3579: usual smooth structure \cite{CD}. 
3580: \item The theory of `Joyce manifolds', that is, 7-dimensional Riemannian 
3581: manifolds with holonomy group $\G_2$ \cite{Joyce}. 
3582: \item The octonionic Hopf map and instanton solutions 
3583: of the Yang-Mills equations in 8 dimensions \cite{GKS}. 
3584: \item Octonionic aspects of 10-dimensional superstring theory and 
3585: 10-dimensional super-Yang-Mills theory 
3586: \cite{CH,Deligne,Evans,KT,Schray,Sierra}.
3587: \item Octonionic aspects of 11-dimensional supergravity and supermembrane 
3588: theories, and the role of Joyce manifolds in compactifying 11-dimensional 
3589: supergravity to obtain theories of physics in 4 dimensions \cite{Duff}. 
3590: \item Geoffrey Dixon's extension of the Standard Model based on the
3591: algebra $\C \tensor \H \tensor \O$ \cite{Dixon}.  
3592: \item Other attempts to use the octonions in physics 
3593: \cite{CMT,GT,LPS,Okubo}.
3594: \end{itemize}  
3595: 
3596: \noindent I urge the reader to explore these with the help of the references.
3597: 
3598: \subsection*{Acknowledgements}  
3599: 
3600: I thank John Barrett, Toby Bartels, Robert Bryant, Geoffrey Dixon, James 
3601: Dolan, Tevian Dray, Bertram Kostant, Linus Kramer, Pertti Lounesto,
3602: Corinne Manogue, John McKay, David Rusin, Tony Smith, Anthony Sudbery, and 
3603: Matthew Wiener for useful discussions.    
3604:    
3605: \begin{thebibliography}{10}   
3606: 
3607: \bibitem{Adams0} John F.\ Adams,
3608: On the non-existence of elements of Hopf invariant
3609: one, {\sl Ann.\ Math.\ }{\bf 72} (1960), 20--104.
3610: 
3611: \bibitem{Adams} John F.\ Adams, {\sl Lectures on Exceptional Lie   
3612: Groups}, eds.\ Zafer Mahmud and Mamoru Mimira, University of    
3613: Chicago Press, Chicago, 1996.   
3614: 
3615: \bibitem{AH} Michael Atiyah and Friedrich Hirzebruch, 
3616: Bott periodicity and the parallelizability of the spheres. 
3617: {\sl Proc.\ Cambridge Philos.\ Soc.\ }{\bf 57} (1961), 223--226. 
3618: 
3619: \bibitem{AM} Helena Albuquerque and Shahn Majid, Quasialgebra structure  
3620: of the octonions, preprint available as math.QA/9802116.   
3621:  
3622: \bibitem{BS} Chris H.\ Barton and Anthony Sudbery, Magic squares of Lie    
3623: algebras, preprint available as math.RA/0001083.   
3624:   
3625: \bibitem{Besse} Arthur L.\ Besse, {\sl Einstein Manifolds}, Springer,   
3626: Berlin, 1987, pp.\ 313--316.   
3627:  
3628: \bibitem{vdB} F.\ van der Blij, History of the octaves, {\sl Simon 
3629: Stevin} {\bf 34} (1961), 106---125. 
3630:  
3631: \bibitem{vdBS} F.\ van der Blij and Tonny A.\ Springer, Octaves and 
3632: triality, {\sl Nieuw Arch.\ v.\ Wiskunde} {\bf 8} (1960), 158--169. 
3633: 
3634: \bibitem{Boos} Dominik Boos, {\sl Ein tensorkategorieller Zugang zum Satz 
3635: von Hurwitz}, Diplomarbeit, ETH Zurich, March 1998.
3636: 
3637: \bibitem{Borel} Armand Borel, Le plan projectif des octaves et les sph\'eres
3638: commes espaces homog\`enes, {\sl Compt.\ Rend.\ Acad.\ Sci.\ }{\bf 230}
3639: (1950), 1378--1380.
3640: 
3641: \bibitem{BM} Raoul Bott and John Milnor, On the parallelizability   
3642: of the spheres, {\sl Bull.\ Amer.\ Math.\ Soc.\ } {\bf 64} (1958)   
3643: 87--89.   
3644:  
3645: \bibitem{Bremner} Murray Bremner, Quantum octonions, 
3646: {\sl Comm.\ Alg.} {\bf 27} (1999), 2809--2831.
3647: 
3648: \bibitem{Brown} Robert B.\ Brown, Groups of type $\E_7$, {\sl
3649: Jour.\ Reine Angew.\ Math.\ }{\bf 236} (1969), 79--102.
3650: 
3651: \bibitem{Cartan0} \'Elie Cartan, {\sl Sur la structure des groupes
3652: de tranformations finis et continus}, Th\`ese, Paris, Nony, 1894.
3653: 
3654: \bibitem{Cartan} \'Elie Cartan, Les groupes r\'eels simples finis et 
3655: continus, {\sl Ann.\ Sci.\ \'Ecole Norm.\ Sup.\ }{\bf 31} (1914),  
3656: 255--262. 
3657: 
3658: \bibitem{Cartan2} \'Elie Cartan, Nombres complexes, in {\sl 
3659: Encyclop\'edie des sciences math\'ematiques}, {\bf 1}, ed.\ J.\ Molk, 
3660: 1908, 329--468.  
3661: 
3662: \bibitem{Cartan3} \'Elie Cartan, Le principe de dualit\'e et la
3663: th\'eorie des groupes simple et semi-simples, {\sl Bull.\ Sci.\ 
3664: Math.\ }{\bf 49} (1925), 361--374.
3665: 
3666: \bibitem{Cayley} Arthur Cayley, On Jacobi's elliptic functions,
3667: in reply to the Rev.\ B.\ Bronwin; and on quaternions, {\sl Philos.\
3668: Mag.\ }{\bf 26} (1845), 208--211.  
3669: 
3670: \bibitem{Cayley2} Arthur Cayley, On Jacobi's elliptic functions, in
3671: reply to the Rev.\ B.\ Bronwin; and on quaternions (appendix only), in
3672: {\sl The Collected Mathematical Papers}, Johnson Reprint Co., New York,
3673: 1963, p.\ 127.
3674: 
3675: \bibitem{CMT} Sultan Catto, Carlos J.\ Moreno and Chia-Hsiung Tze,   
3676: {\sl Octonionic Structures in Physics}, to appear.     
3677:    
3678: \bibitem{CS} Claude Chevalley and Richard D.\ Schafer, The exceptional    
3679: simple Lie algebras $\F_4$ and $\E_6$, {\sl Proc.\ Nat.\ Acad.\ Sci.\ USA}   
3680: {\bf 36} (1950), 137--141.    
3681:  
3682: \bibitem{CD} Yvonne Choquet-Bruhat and C\'ecile DeWitt-Morette,  
3683: {\sl Analysis, Manifolds and Physics}, part II, Elsevier, Amsterdam, 
3684: 2000, pp.\ 263--274. 
3685:  
3686: \bibitem{Clifford} William K.\ Clifford, Applications of Grassmann's  
3687: extensive algebra, {\sl Amer.\ Jour.\ Math.\ }{\bf 1} (1878), 350--358.  
3688:    
3689: \bibitem{Cohen} Frederick R.\ Cohen, On Whitehead squares, Cayley--Dickson    
3690: algebras and rational functions, {\sl Bol.\ Soc.\ Mat.\ Mexicana}   
3691: {\bf 37} (1992), 55--62.    
3692: 
3693: \bibitem{CH} E.\ Corrigan and T.\ J.\ Hollowood, The exceptional Jordan
3694: algebra and the superstring, {\sl Comm.\ Math.\ Phys.\ }{\bf 122} (1989),
3695: 393--410.
3696: 
3697: \bibitem{Coxeter} Harold Scott MacDonald Coxeter, Integral Cayley    
3698: numbers, {\sl Duke Math.\ Jour.\ }{\bf 13} (1946), 561--578.   
3699:    
3700: \bibitem{Crowe} Michael J.\ Crowe, {\sl A History of Vector Analysis},   
3701: University of Notre Dame Press, Notre Dame, 1967.   
3702: 
3703: \bibitem{Curtis} C.\ W.\ Curtis, The four and eight square problem    
3704: and division algebras, in {\sl Studies in Modern Algebra}, ed.\    
3705: A.\ Albert, Prentice--Hall, Englewood Cliffs, New Jersey, 1963, pp.\ 100--125.
3706:  
3707: \bibitem{Deligne} Pierre Deligne {\it et al}, eds., {\sl Quantum
3708: Fields and Strings: A Course for Mathematicians}, 2 volumes, Amer.\ 
3709: Math.\ Soc., Providence, Rhode Island, 1999. 
3710:  
3711: \bibitem{Dickson} Leonard E.\ Dickson, On quaternions and their  
3712: generalization and the history of the eight square theorem, {\sl
3713: Ann.\ Math.\ }{\bf 20} (1919), 155--171.   
3714:    
3715: \bibitem{Dixon} Geoffrey M.\ Dixon, {\sl Division Algebras: Octonions,   
3716: Quaternions, Complex Numbers and the Algebraic Design of Physics},   
3717: Kluwer, Dordrecht, 1994.   
3718:  
3719: \bibitem{Duff} M.\ J.\ Duff, ed., {\sl The World in Eleven Dimensions:  
3720: Supergravity, Supermembranes and M-Theory}, Institute of Physics 
3721: Publishing, Bristol, 1999. 
3722:  
3723: \bibitem{EG} Noam Elkies and Benedict H.\ Gross, The exceptional
3724: cone and the Leech lattice, {\sl Internat.\ Math.\ Res.\ Notices}
3725: {\bf 14} (1996), 665--698.
3726: 
3727: \bibitem{Emch} Gerard G.\ Emch, {\sl Algebraic Methods in Statistical 
3728: Mechanics and Quantum Field Theory}, Wiley-Interscience, New York, 1972.
3729: 
3730: \bibitem{Evans} J.\ M.\ Evans, Supersymmetric Yang--Mills theories and   
3731: division algebras, {\sl Nucl.\ Phys.\ }{\bf B298} (1988), 92--108.   
3732:    
3733: \bibitem{Faulkner} John R.\ Faulkner, A construction of Lie algebras from
3734: a class of ternary algebras, {\sl Trans.\ Amer.\ Math.\ Soc.\ }{\bf 155}
3735: (1971), 397--408.
3736: 
3737: \bibitem{FF} John R.\ Faulkner and Joseph C.\ Ferrar,    
3738: Exceptional Lie algebras and related algebraic and geometric structures,   
3739: {\sl Bull.\ London Math.\ Soc.\ }{\bf 9} (1977), 1--35.   
3740: 
3741: \bibitem{FFrenkel} Alex J.\ Feingold, Igor B.\ Frenkel, and John F.\ X.\
3742: Ries, {\sl Spinor Construction of Vertex Operator Algebras, Triality, and
3743: $\E_8^{(1)}$}, Contemp.\ Math.\ 121, Amer.\ Math.\ Soc., 
3744: Providence, Rhode Island, 1991.
3745: 
3746: \bibitem{Freudenthal4}  Hans Freudenthal, Oktaven, Ausnahmegruppen und    
3747: Oktavengeometrie, mimeographed notes, 1951.  Also available in
3748: {\sl Geom.\ Dedicata} {\bf 19} (1985), 7--63.    
3749:    
3750: \bibitem{Freudenthal} Hans Freudenthal, Zur ebenen Oktavengeometrie,    
3751: {\sl Indag.\ Math.\ }{\bf 15} (1953), 195--200.   
3752:    
3753: \bibitem{Freudenthal2} Hans Freudenthal, Beziehungen der $\e_7$ und   
3754: $\e_8$ zur Oktavenebene, I, II, {\sl Indag.\ Math.\ }{\bf 16} (1954),   
3755: 218--230, 363--368.  III, IV, {\sl Indag.\ Math.\ }{\bf 17} (1955),   
3756: 151--157, 277--285.  V --- IX, {\sl Indag.\ Math.\ }{\bf 21} (1959),  
3757: 165--201, 447--474.  X, XI, {\sl Indag.\ Math.\ }{\bf 25} (1963) 457--487. 
3758:    
3759: \bibitem{Freudenthal3} Hans Freudenthal, Lie groups in the foundations of   
3760: geometry, {\sl Adv.\ Math.\ }{\bf 1} (1964), 145--190.   
3761: 
3762: \bibitem{Freudenthal5} Hans Freudenthal, Bericht \"uber die Theorie   
3763: der Rosenfeldschen elliptischen Ebenen, in {\sl Raumtheorie}, Wege  
3764: Der Forschung, CCLXX, Wissenschaftliche Buchgesellschaft, Darmstadt,  
3765: 1978, pp. 283--286.  
3766: 
3767: \bibitem{Garner} Lynn E.\ Garner, {\sl An Outline of Projective Geometry}, 
3768: North Holland, New York, 1981. 
3769:    
3770: \bibitem{Graves} Robert Perceval Graves, {\sl Life of Sir William Rowan   
3771: Hamilton}, 3 volumes, Arno Press, New York, 1975.   
3772:    
3773: \bibitem{GSW} Michael B.\ Green, John H.\ Schwarz and Edward Witten,   
3774: {\sl Superstring Theory}, volume 1, Cambridge University Press,    
3775: Cambridge, 1987, pp.\ 344--349.   
3776:  
3777: \bibitem{GKS} B.\ Grossman, T.\ E.\ Kephart, and James D.\ Stasheff, 
3778: Solutions to Yang-Mills field equations in eight dimensions and the  
3779: last Hopf map, {\sl Comm.\ Math.\ Phys.\ }{\bf 96} (1984), 431--437. 
3780: 
3781: \bibitem{Gunaydin} Murat G\"unaydin, Generalized conformal and
3782: superconformal group actions and Jordan algebras, 
3783: {\sl Mod.\ Phys.\ Lett.\ }{\bf A8} (1993), 1407--1416.
3784: 
3785: \bibitem{GKN} Murat G\"unaydin, Kilian Koepsell, and Hermann Nicolai, 
3786: Conformal and quasiconformal realizations of exceptional Lie groups,
3787: preprint available as hep-th/0008063.
3788: 
3789: \bibitem{GPR} Murat G\"unaydin, C.\ Piron and H.\ Ruegg, Moufang plane
3790: and octonionic quantum mechanics, {\sl Comm.\ Math.\ Phys.\ }{\bf 61}
3791: (1978), 69--85.
3792:   
3793: \bibitem{GT} Feza G\"ursey and Chia-Hsiung Tze, {\sl On the Role of Division, 
3794: Jordan, and Related Algebras in Particle Physics}, World Scientific,   
3795: Singapore, 1996.   
3796:    
3797: \bibitem{Hamilton} William Rowan Hamilton, Four and eight square   
3798: theorems, in Appendix 3 of {\sl The Mathematical Papers of William   
3799: Rowan Hamilton}, vol.\ 3, eds.\ H. Halberstam and R. E. Ingram,   
3800: Cambridge University Press, Cambridge, 1967, pp.\ 648--656.   
3801:    
3802: \bibitem{Hankins} Thomas L.\ Hankins, {\sl Sir William Rowan Hamilton},   
3803: John Hopkins University Press, Baltimore, 1980.   
3804:   
3805: \bibitem{Harvey} F.\ Reese Harvey, {\sl Spinors and Calibrations},    
3806: Academic Press, San Diego, 1990.   
3807:   
3808: \bibitem{Hurwitz} Adolf Hurwitz, \"Uber die Composition der quadratischen   
3809: Formen von beliebig vielen Variabeln, {\sl Nachr.\ Ges.\ Wiss.\ G\"ottingen} 
3810: (1898) 309--316.   
3811: 
3812: \bibitem{Husemoller} Dale Husemoller, {\sl Fibre Bundles}, Springer, Berlin,
3813: 1994.
3814: 
3815: \bibitem{Jordan} Pascual Jordan, \"Uber eine Klasse nichtassociativer   
3816: hyperkomplexer Algebren, {\sl Nachr.\ Ges.\ Wiss.\ G\"ottingen} (1932),  
3817: 569--575.   
3818:   
3819: \bibitem{Jordan2} Pascual Jordan, \"Uber eine nicht-desarguessche 
3820: ebene projektive Geometrie, {\sl Abh.\ Math.\ Sem.\ Hamburg} 
3821: {\bf 16} (1949), 74--76. 
3822:  
3823: \bibitem{JNW} Pascual Jordan, John von Neumann, Eugene Wigner,    
3824: On an algebraic generalization of the quantum mechanical formalism,   
3825: {\sl Ann.\ Math.\ }{\bf 35} (1934), 29--64.   
3826:  
3827: \bibitem{Joyce} Dominic Joyce, {\sl Compact Manifolds with Special  
3828: Holonomy}, Oxford U.\ Press, Oxford, 2000.  
3829:  
3830: \bibitem{KS} I.\ L.\ Kantor and A.\ S.\ Solodovnikov, {\sl Hypercomplex   
3831: Numbers --- an Elementary Introduction to Algebras,} Springer, Berlin, 1989.  
3832: 
3833: \bibitem{Kervaire} Michel Kervaire, Non-parallelizability of the $n$ 
3834: sphere for $n > 7$, {\sl Proc.\ Nat.\ Acad.\ Sci.\ USA} {\bf 44} (1958),
3835: 280--283.
3836: 
3837: \bibitem{Killing} Wilhelm Killing, Die Zusammensetzung der stetigen
3838: endlichen Transformationsgruppen I, {\sl Math.\ Ann.\ } {\bf 31} (1888),
3839: 252--290.  II, {\bf 33} (1889) 1--48.  III, {\bf 34} (1889), 57--122.
3840: IV {\bf 36} (1890), 161--189.
3841: 
3842: \bibitem{Kuperberg} Greg Kuperberg, Spiders for rank 2 Lie algebras,
3843: {\sl Comm.\ Math.\ Phys.\ }{\bf 180} (1996), 109--151.
3844: 
3845: \bibitem{KT} T.\ Kugo and P.--K.\ Townsend, Supersymmetry and the 
3846: division algebras, {\sl Nucl.\ Phys.\ }{\bf B221} (1983), 357--380. 
3847:  
3848: \bibitem{LM} J.\ M.\ Landsberg and L.\ Manivel: The projective geometry of   
3849: Freudenthal's magic square, preprint available as math.AG/9908039.   
3850:  
3851: \bibitem{LPS} Jaak Lohmus, Eugene Paal, and Leo Sorgsepp,    
3852: {\sl Nonassociative Algebras in Physics}, Hadronic Press, Palm   
3853: Harbor, Florida, 1994.   
3854: 
3855: \bibitem{MD} Corinne A.\ Manogue and Tevian Dray, Octonionic    
3856: M\"obius transformations, {\sl Mod.\ Phys.\ Lett.\ }{\bf A14} (1999), 
3857: 1243--1256. 
3858: 
3859: \bibitem{MS} Corinne A.\ Manogue and J\"org Schray, Finite Lorentz   
3860: transformations, automorphisms, and division algebras, {\sl Jour.\   
3861: Math.\ Phys.\ }{\bf 34} (1993), 3746--3767.   
3862:  
3863: \bibitem{MS2} Corinne A.\ Manogue and J\"org Schray, Octonionic    
3864: representations of Clifford algebras and triality, {\sl Found.\    
3865: Phys.\ }{\bf 26} (1996), 17--70.    
3866:  
3867: \bibitem{McCrimmon} Kevin McCrimmon, Jordan algebras and their applications,   
3868: {\sl Bull.\ Amer.\ Math.\ Soc.\ } {\bf 84} (1978), 612--627.     
3869: 
3870: \bibitem{Meyberg} K.\ Meyberg, Eine Theorie der Freudenthalschen 
3871: Tripelsysteme, I, II, {\sl Indag.\ Math.\ }{\bf 30} (1968),
3872: 162--190.
3873:  
3874: \bibitem{Moreno} R.\ Guillermo Moreno, The zero divisors of the    
3875: Cayley--Dickson algebras over the real numbers, preprint available    
3876: at q-alg/9710013.   
3877:   
3878: \bibitem{Moufang} Ruth Moufang, Alternativk\"orper und der Satz vom  
3879: vollst\"andigen Vierseit, {\sl Abh.\ Math.\ Sem.\ Hamburg} {\bf 9} 
3880: (1933), 207--222. 
3881:   
3882: \bibitem{OV} A.\ L. Onishchik and E.\ B.\ Vinberg, eds., {\sl Lie Groups 
3883: and Lie Algebras III}, Springer, Berlin, 1991, pp.\ 167--178.   
3884:    
3885: \bibitem{Okubo} Susumu Okubo, {\sl Introduction to Octonion and Other    
3886: Non-Associative Algebras in Physics}, Cambridge University Press,   
3887: Cambridge, 1995.   
3888:  
3889: \bibitem{PR} Roger Penrose and Wolfgang Rindler, {\sl Spinors and  
3890: Space-Time}, 2 volumes, Cambridge U.\ Press, Cambridge, 1985-86. 
3891:   
3892: \bibitem{Porteous} Ian R.\ Porteous, {\sl Topological Geometry},    
3893: Cambridge U.\ Press, 1981.   
3894: 
3895: \bibitem{Rosenfeld1} Boris A.\ Rosenfeld, Geometrical interpretation of
3896: the compact simple Lie groups of the class $\E$ (Russian), {\sl
3897: Dokl.\ Akad.\ Nauk.\ SSSR} (1956) {\bf 106}, 600-603.
3898: 
3899: \bibitem{Rosenfeld} Boris A.\ Rosenfeld, {\sl Geometry of Lie Groups},    
3900: Kluwer, Dordrecht, 1997.   
3901: 
3902: \bibitem{Rost} Markus Rost, On the dimension of a composition algebra, 
3903: {\sl Doc.\ Math.\ } {\bf 1} (1996), 209--214.  
3904: 
3905: \bibitem{Salzmann} Helmut Salzmann {\it et al}, {\sl Compact Projective
3906: Planes: With an Introduction to Octonion Geometry}, de Gruyter, Berlin,
3907: 1995.
3908: 
3909: \bibitem{Schafer1} Richard D.\ Schafer, On algebras formed by the    
3910: Cayley--Dickson process, {\sl Amer.\ Jour.\ of Math.\ }{\bf 76} (1954)   
3911: 435--446.   
3912:   
3913: \bibitem{Schafer} Richard D.\ Schafer, {\sl Introduction to Non-Associative 
3914: Algebras}, Dover, New York, 1995.   
3915: 
3916: \bibitem{Schray} J\"org Schray, {\sl Octonions and Supersymmetry},
3917: Ph.D.\ thesis, Department of Physics, Oregon State University, 1994.
3918:  
3919: \bibitem{Sierra} G.\ Sierra, An application of the theories of Jordan 
3920: algebras and Freudenthal triple systems to particles and strings,  
3921: {\sl Class.\ Quant.\ Grav.\ }{\bf 4} (1987), 227--236.
3922: 
3923: \bibitem{Springer} Tonny A.\ Springer, The projective octave plane, I--II, 
3924: {\sl Indag.\ Math.\ }{\bf 22} (1960), 74--101. 
3925:   
3926: \bibitem{Springer2} Tonny A.\ Springer, Characterization of a class of
3927: cubic forms, {\sl Indag.\ Math.\ }{\bf 24} (1962), 259--265.
3928: 
3929: \bibitem{Springer3} Tonny A.\ Springer, On the geometric algebra of 
3930: the octave planes, {\sl Indag.\ Math.\ }{\bf 24} (1962), 451--468.
3931:  
3932: \bibitem{SV} Tonny A.\ Springer and Ferdinand D.\ Veldkamp,    
3933: {\sl Octonions, Jordan Algebras and Exceptional Groups}, Springer,   
3934: Berlin, 2000.   
3935:  
3936: \bibitem{Stevenson} Frederick W. Stevenson, {\sl Projective Planes}, W.\ H.\ 
3937: Freeman and Company, San Francisco, 1972. 
3938:  
3939: \bibitem{Sudbery2} Anthony Sudbery, Octonionic description of exceptional 
3940: Lie superalgebras, {\sl Jour.\ Math.\ Phys.\ }{\bf 24} (1983), 1986--1988. 
3941:  
3942: \bibitem{Sudbery} Anthony Sudbery, Division algebras, (pseudo)orthogonal   
3943: groups and spinors, {\sl Jour.\ Phys.\ }{\bf A17} (1984), 939--955.   
3944:   
3945: \bibitem{Tits} Jacques Tits, Le plan projectif des octaves et les groupes 
3946: de Lie exceptionnels, {\sl Bull.\ Acad.\ Roy.\ Belg.\ Sci.\ }{\bf 39} (1953), 
3947: 309--329. 
3948:  
3949: \bibitem{Tits2} Jacques Tits, Le plan projectif des octaves et les groupes 
3950: exceptionnels $\E_6$ et $\E_7$, {\sl Bull.\ Acad.\ Roy.\ Belg.\ 
3951: Sci.\ }{\bf 40} (1954), 29--40. 
3952:  
3953: \bibitem{Tits3} Jacques Tits, Alg\`ebres alternatives, alg\`ebres de    
3954: Jordan et alg\`ebres de Lie exceptionnelles, {\sl Indag.\ Math.\ }{\bf 28} 
3955: (1966), 223--237.   
3956:  
3957: \bibitem{Tits4} Jacques Tits, {\sl Buildings of Spherical Type 
3958: and Finite BN-Pairs, } Lecture Notes in Mathematics, Vol.\ 386,
3959: Springer, Berlin, 1974.  
3960: 
3961: \bibitem{Varadarajan} V.\ S.\ Varadarajan, {\sl Geometry of Quantum 
3962: Theory}, Springer-Verlag, Berlin, 1985.   
3963: 
3964: \bibitem{Vinberg} E.\ B.\ Vinberg, A construction of exceptional
3965: simple Lie groups (Russian), {\sl Tr.\ Semin.\ Vektorn.\ Tensorn.\ 
3966: Anal.\ }{\bf 13} (1966), 7--9.
3967: 
3968: \bibitem{Zorn} Max Zorn, Theorie der alternativen Ringe,    
3969: {\sl Abh.\ Math.\ Sem.\ Univ.\ Hamburg} {\bf 8} (1930), 123--147.   
3970:   
3971: \bibitem{Zorn2} Max Zorn, Alternativk\"orper und quadratische Systeme,   
3972: {\sl Abh.\ Math.\ Sem.\ Univ.\ Hamburg} {\bf 9} (1933), 395--402.   
3973:   
3974: \end{thebibliography}   
3975: \end{document}   
3976:    
3977:    
3978:    
3979:    
3980: